sphinx.addnodesdocument)}( rawsourcechildren]( translations LanguagesNode)}(hhh](h pending_xref)}(hhh]docutils.nodesTextChinese (Simplified)}parenthsba attributes}(ids]classes]names]dupnames]backrefs] refdomainstdreftypedoc reftarget8/translations/zh_CN/RCU/Design/Requirements/RequirementsmodnameN classnameN refexplicitutagnamehhh ubh)}(hhh]hChinese (Traditional)}hh2sbah}(h]h ]h"]h$]h&] refdomainh)reftypeh+ reftarget8/translations/zh_TW/RCU/Design/Requirements/RequirementsmodnameN classnameN refexplicituh1hhh ubh)}(hhh]hItalian}hhFsbah}(h]h ]h"]h$]h&] refdomainh)reftypeh+ reftarget8/translations/it_IT/RCU/Design/Requirements/RequirementsmodnameN classnameN refexplicituh1hhh ubh)}(hhh]hJapanese}hhZsbah}(h]h ]h"]h$]h&] refdomainh)reftypeh+ reftarget8/translations/ja_JP/RCU/Design/Requirements/RequirementsmodnameN classnameN refexplicituh1hhh ubh)}(hhh]hKorean}hhnsbah}(h]h ]h"]h$]h&] refdomainh)reftypeh+ reftarget8/translations/ko_KR/RCU/Design/Requirements/RequirementsmodnameN classnameN refexplicituh1hhh ubh)}(hhh]hPortuguese (Brazilian)}hhsbah}(h]h ]h"]h$]h&] refdomainh)reftypeh+ reftarget8/translations/pt_BR/RCU/Design/Requirements/RequirementsmodnameN classnameN refexplicituh1hhh ubh)}(hhh]hSpanish}hhsbah}(h]h ]h"]h$]h&] refdomainh)reftypeh+ reftarget8/translations/sp_SP/RCU/Design/Requirements/RequirementsmodnameN classnameN refexplicituh1hhh ubeh}(h]h ]h"]h$]h&]current_languageEnglishuh1h hh _documenthsourceNlineNubhsection)}(hhh](htitle)}(h!A Tour Through RCU's Requirementsh]h#A Tour Through RCU’s Requirements}(hhhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhhhR/var/lib/git/docbuild/linux/Documentation/RCU/Design/Requirements/Requirements.rsthKubh paragraph)}(hCopyright IBM Corporation, 2015h]hCopyright IBM Corporation, 2015}(hhhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhKhhhhubh)}(hAuthor: Paul E. McKenneyh]hAuthor: Paul E. McKenney}(hhhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhKhhhhubh)}(hThe initial version of this document appeared in the `LWN `_ on those articles: `part 1 `_, `part 2 `_, and `part 3 `_.h](h5The initial version of this document appeared in the }(hhhhhNhNubh reference)}(h`LWN `_h]hLWN}(hhhhhNhNubah}(h]h ]h"]h$]h&]nameLWNrefurihttps://lwn.net/uh1hhhubhtarget)}(h h]h}(h]lwnah ]h"]lwnah$]h&]refurijuh1j referencedKhhubh on those articles: }(hhhhhNhNubh)}(h,`part 1 `_h]hpart 1}(hjhhhNhNubah}(h]h ]h"]h$]h&]namepart 1j https://lwn.net/Articles/652156/uh1hhhubj)}(h# h]h}(h]part-1ah ]h"]part 1ah$]h&]refurij)uh1jjKhhubh, }(hhhhhNhNubh)}(h,`part 2 `_h]hpart 2}(hj;hhhNhNubah}(h]h ]h"]h$]h&]namepart 2j https://lwn.net/Articles/652677/uh1hhhubj)}(h# h]h}(h]part-2ah ]h"]part 2ah$]h&]refurijKuh1jjKhhubh, and }(hhhhhNhNubh)}(h,`part 3 `_h]hpart 3}(hj]hhhNhNubah}(h]h ]h"]h$]h&]namepart 3j https://lwn.net/Articles/653326/uh1hhhubj)}(h# h]h}(h]part-3ah ]h"]part 3ah$]h&]refurijmuh1jjKhhubh.}(hhhhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhK hhhhubh)}(hhh](h)}(h Introductionh]h Introduction}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhjhhhhhKubh)}(hXRead-copy update (RCU) is a synchronization mechanism that is often used as a replacement for reader-writer locking. RCU is unusual in that updaters do not block readers, which means that RCU's read-side primitives can be exceedingly fast and scalable. In addition, updaters can make useful forward progress concurrently with readers. However, all this concurrency between RCU readers and updaters does raise the question of exactly what RCU readers are doing, which in turn raises the question of exactly what RCU's requirements are.h]hXRead-copy update (RCU) is a synchronization mechanism that is often used as a replacement for reader-writer locking. RCU is unusual in that updaters do not block readers, which means that RCU’s read-side primitives can be exceedingly fast and scalable. In addition, updaters can make useful forward progress concurrently with readers. However, all this concurrency between RCU readers and updaters does raise the question of exactly what RCU readers are doing, which in turn raises the question of exactly what RCU’s requirements are.}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhKhjhhubh)}(hXThis document therefore summarizes RCU's requirements, and can be thought of as an informal, high-level specification for RCU. It is important to understand that RCU's specification is primarily empirical in nature; in fact, I learned about many of these requirements the hard way. This situation might cause some consternation, however, not only has this learning process been a lot of fun, but it has also been a great privilege to work with so many people willing to apply technologies in interesting new ways.h]hXThis document therefore summarizes RCU’s requirements, and can be thought of as an informal, high-level specification for RCU. It is important to understand that RCU’s specification is primarily empirical in nature; in fact, I learned about many of these requirements the hard way. This situation might cause some consternation, however, not only has this learning process been a lot of fun, but it has also been a great privilege to work with so many people willing to apply technologies in interesting new ways.}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhKhjhhubh)}(hLAll that aside, here are the categories of currently known RCU requirements:h]hLAll that aside, here are the categories of currently known RCU requirements:}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhK$hjhhubhenumerated_list)}(hhh](h list_item)}(h`Fundamental Requirements`_h]h)}(hjh]h)}(hjh]hFundamental Requirements}(hjhhhNhNubah}(h]h ]h"]h$]h&]nameFundamental Requirementsrefidfundamental-requirementsuh1hhjresolvedKubah}(h]h ]h"]h$]h&]uh1hhhhK'hjubah}(h]h ]h"]h$]h&]uh1jhjhhhhhNubj)}(h`Fundamental Non-Requirements`_h]h)}(hjh]h)}(hjh]hFundamental Non-Requirements}(hjhhhNhNubah}(h]h ]h"]h$]h&]nameFundamental Non-Requirementsjfundamental-non-requirementsuh1hhjjKubah}(h]h ]h"]h$]h&]uh1hhhhK(hjubah}(h]h ]h"]h$]h&]uh1jhjhhhhhNubj)}(h`Parallelism Facts of Life`_h]h)}(hjh]h)}(hjh]hParallelism Facts of Life}(hjhhhNhNubah}(h]h ]h"]h$]h&]nameParallelism Facts of Lifejparallelism-facts-of-lifeuh1hhjjKubah}(h]h ]h"]h$]h&]uh1hhhhK)hjubah}(h]h ]h"]h$]h&]uh1jhjhhhhhNubj)}(h)`Quality-of-Implementation Requirements`_h]h)}(hj4h]h)}(hj4h]h&Quality-of-Implementation Requirements}(hj9hhhNhNubah}(h]h ]h"]h$]h&]name&Quality-of-Implementation Requirementsj&quality-of-implementation-requirementsuh1hhj6jKubah}(h]h ]h"]h$]h&]uh1hhhhK*hj2ubah}(h]h ]h"]h$]h&]uh1jhjhhhhhNubj)}(h`Linux Kernel Complications`_h]h)}(hjWh]h)}(hjWh]hLinux Kernel Complications}(hj\hhhNhNubah}(h]h ]h"]h$]h&]nameLinux Kernel Complicationsjlinux-kernel-complicationsuh1hhjYjKubah}(h]h ]h"]h$]h&]uh1hhhhK+hjUubah}(h]h ]h"]h$]h&]uh1jhjhhhhhNubj)}(h$`Software-Engineering Requirements`_h]h)}(hjzh]h)}(hjzh]h!Software-Engineering Requirements}(hjhhhNhNubah}(h]h ]h"]h$]h&]name!Software-Engineering Requirementsj!software-engineering-requirementsuh1hhj|jKubah}(h]h ]h"]h$]h&]uh1hhhhK,hjxubah}(h]h ]h"]h$]h&]uh1jhjhhhhhNubj)}(h`Other RCU Flavors`_h]h)}(hjh]h)}(hjh]hOther RCU Flavors}(hjhhhNhNubah}(h]h ]h"]h$]h&]nameOther RCU Flavorsjother-rcu-flavorsuh1hhjjKubah}(h]h ]h"]h$]h&]uh1hhhhK-hjubah}(h]h ]h"]h$]h&]uh1jhjhhhhhNubj)}(h`Possible Future Changes`_ h]h)}(h`Possible Future Changes`_h]h)}(hjh]hPossible Future Changes}(hjhhhNhNubah}(h]h ]h"]h$]h&]namePossible Future Changesjpossible-future-changesuh1hhjjKubah}(h]h ]h"]h$]h&]uh1hhhhK.hjubah}(h]h ]h"]h$]h&]uh1jhjhhhhhNubeh}(h]h ]h"]h$]h&]enumtypearabicprefixhsuffix.uh1jhjhhhhhK'ubh)}(hThis is followed by a summary_, however, the answers to each quick quiz immediately follows the quiz. Select the big white space with your mouse to see the answer.h](hThis is followed by a }(hjhhhNhNubh)}(hsummary_h]hsummary}(hjhhhNhNubah}(h]h ]h"]h$]h&]namesummaryjsummaryuh1hhjjKubh, however, the answers to each quick quiz immediately follows the quiz. Select the big white space with your mouse to see the answer.}(hjhhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhK0hjhhubeh}(h] introductionah ]h"] introductionah$]h&]uh1hhhhhhhhKubh)}(hhh](h)}(hFundamental Requirementsh]hFundamental Requirements}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhjhhhhhK5ubh)}(hjRCU's fundamental requirements are the closest thing RCU has to hard mathematical requirements. These are:h]hlRCU’s fundamental requirements are the closest thing RCU has to hard mathematical requirements. These are:}(hj)hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhK7hjhhubj)}(hhh](j)}(h`Grace-Period Guarantee`_h]h)}(hj<h]h)}(hj<h]hGrace-Period Guarantee}(hjAhhhNhNubah}(h]h ]h"]h$]h&]nameGrace-Period Guaranteejgrace-period-guaranteeuh1hhj>jKubah}(h]h ]h"]h$]h&]uh1hhhhK:hj:ubah}(h]h ]h"]h$]h&]uh1jhj7hhhhhNubj)}(h`Publish/Subscribe Guarantee`_h]h)}(hj_h]h)}(hj_h]hPublish/Subscribe Guarantee}(hjdhhhNhNubah}(h]h ]h"]h$]h&]namePublish/Subscribe Guaranteejpublish-subscribe-guaranteeuh1hhjajKubah}(h]h ]h"]h$]h&]uh1hhhhK;hj]ubah}(h]h ]h"]h$]h&]uh1jhj7hhhhhNubj)}(h`Memory-Barrier Guarantees`_h]h)}(hjh]h)}(hjh]hMemory-Barrier Guarantees}(hjhhhNhNubah}(h]h ]h"]h$]h&]nameMemory-Barrier Guaranteesjmemory-barrier-guaranteesuh1hhjjKubah}(h]h ]h"]h$]h&]uh1hhhhKhjubah}(h]h ]h"]h$]h&]uh1jhj7hhhhhNubeh}(h]h ]h"]h$]h&]jjjhjjuh1jhjhhhhhK:ubh)}(hhh](h)}(hGrace-Period Guaranteeh]hGrace-Period Guarantee}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhjhhhhhKAubh)}(hX;RCU's grace-period guarantee is unusual in being premeditated: Jack Slingwine and I had this guarantee firmly in mind when we started work on RCU (then called “rclock”) in the early 1990s. That said, the past two decades of experience with RCU have produced a much more detailed understanding of this guarantee.h]hX=RCU’s grace-period guarantee is unusual in being premeditated: Jack Slingwine and I had this guarantee firmly in mind when we started work on RCU (then called “rclock”) in the early 1990s. That said, the past two decades of experience with RCU have produced a much more detailed understanding of this guarantee.}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhKChjhhubh)}(hX+RCU's grace-period guarantee allows updaters to wait for the completion of all pre-existing RCU read-side critical sections. An RCU read-side critical section begins with the marker rcu_read_lock() and ends with the marker rcu_read_unlock(). These markers may be nested, and RCU treats a nested set as one big RCU read-side critical section. Production-quality implementations of rcu_read_lock() and rcu_read_unlock() are extremely lightweight, and in fact have exactly zero overhead in Linux kernels built for production use with ``CONFIG_PREEMPTION=n``.h](hXRCU’s grace-period guarantee allows updaters to wait for the completion of all pre-existing RCU read-side critical sections. An RCU read-side critical section begins with the marker rcu_read_lock() and ends with the marker rcu_read_unlock(). These markers may be nested, and RCU treats a nested set as one big RCU read-side critical section. Production-quality implementations of rcu_read_lock() and rcu_read_unlock() are extremely lightweight, and in fact have exactly zero overhead in Linux kernels built for production use with }(hjhhhNhNubhliteral)}(h``CONFIG_PREEMPTION=n``h]hCONFIG_PREEMPTION=n}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1jhjubh.}(hjhhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhKIhjhhubh)}(hbThis guarantee allows ordering to be enforced with extremely low overhead to readers, for example:h]hbThis guarantee allows ordering to be enforced with extremely low overhead to readers, for example:}(hj1hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhKShjhhubh block_quote)}(hX+:: 1 int x, y; 2 3 void thread0(void) 4 { 5 rcu_read_lock(); 6 r1 = READ_ONCE(x); 7 r2 = READ_ONCE(y); 8 rcu_read_unlock(); 9 } 10 11 void thread1(void) 12 { 13 WRITE_ONCE(x, 1); 14 synchronize_rcu(); 15 WRITE_ONCE(y, 1); 16 } h]h literal_block)}(h 1 int x, y; 2 3 void thread0(void) 4 { 5 rcu_read_lock(); 6 r1 = READ_ONCE(x); 7 r2 = READ_ONCE(y); 8 rcu_read_unlock(); 9 } 10 11 void thread1(void) 12 { 13 WRITE_ONCE(x, 1); 14 synchronize_rcu(); 15 WRITE_ONCE(y, 1); 16 }h]h 1 int x, y; 2 3 void thread0(void) 4 { 5 rcu_read_lock(); 6 r1 = READ_ONCE(x); 7 r2 = READ_ONCE(y); 8 rcu_read_unlock(); 9 } 10 11 void thread1(void) 12 { 13 WRITE_ONCE(x, 1); 14 synchronize_rcu(); 15 WRITE_ONCE(y, 1); 16 }}hjGsbah}(h]h ]h"]h$]h&] xml:spacepreserveuh1jEhhhKXhjAubah}(h]h ]h"]h$]h&]uh1j?hhhKVhjhhubh)}(hXBecause the synchronize_rcu() on line 14 waits for all pre-existing readers, any instance of thread0() that loads a value of zero from ``x`` must complete before thread1() stores to ``y``, so that instance must also load a value of zero from ``y``. Similarly, any instance of thread0() that loads a value of one from ``y`` must have started after the synchronize_rcu() started, and must therefore also load a value of one from ``x``. Therefore, the outcome:h](hBecause the synchronize_rcu() on line 14 waits for all pre-existing readers, any instance of thread0() that loads a value of zero from }(hj]hhhNhNubj)}(h``x``h]hx}(hjehhhNhNubah}(h]h ]h"]h$]h&]uh1jhj]ubh* must complete before thread1() stores to }(hj]hhhNhNubj)}(h``y``h]hy}(hjwhhhNhNubah}(h]h ]h"]h$]h&]uh1jhj]ubh7, so that instance must also load a value of zero from }(hj]hhhNhNubj)}(h``y``h]hy}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1jhj]ubhF. Similarly, any instance of thread0() that loads a value of one from }(hj]hhhNhNubj)}(h``y``h]hy}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1jhj]ubhi must have started after the synchronize_rcu() started, and must therefore also load a value of one from }(hj]hhhNhNubj)}(h``x``h]hx}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1jhj]ubh. Therefore, the outcome:}(hj]hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhKihjhhubj@)}(h:: (r1 == 0 && r2 == 1) h]jF)}(h(r1 == 0 && r2 == 1)h]h(r1 == 0 && r2 == 1)}hjsbah}(h]h ]h"]h$]h&]jUjVuh1jEhhhKshjubah}(h]h ]h"]h$]h&]uh1j?hhhKqhjhhubh)}(hcannot happen.h]hcannot happen.}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhKuhjhhubhtable)}(hhh]htgroup)}(hhh](hcolspec)}(hhh]h}(h]h ]h"]h$]h&]colwidthKGuh1jhjubhtbody)}(hhh](hrow)}(hhh]hentry)}(hhh]h)}(h**Quick Quiz**:h](hstrong)}(h**Quick Quiz**h]h Quick Quiz}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1jhjubh:}(hjhhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhKxhj ubah}(h]h ]h"]h$]h&]uh1j hjubah}(h]h ]h"]h$]h&]uh1jhjubj)}(hhh]j )}(hhh]h)}(hWait a minute! You said that updaters can make useful forward progress concurrently with readers, but pre-existing readers will block synchronize_rcu()!!! Just who are you trying to fool???h]hWait a minute! You said that updaters can make useful forward progress concurrently with readers, but pre-existing readers will block synchronize_rcu()!!! Just who are you trying to fool???}(hj@hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhKzhj=ubah}(h]h ]h"]h$]h&]uh1j hj:ubah}(h]h ]h"]h$]h&]uh1jhjubj)}(hhh]j )}(hhh]h)}(h **Answer**:h](j)}(h **Answer**h]hAnswer}(hjdhhhNhNubah}(h]h ]h"]h$]h&]uh1jhj`ubh:}(hj`hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhKhj]ubah}(h]h ]h"]h$]h&]uh1j hjZubah}(h]h ]h"]h$]h&]uh1jhjubj)}(hhh]j )}(hhh]h)}(hXFirst, if updaters do not wish to be blocked by readers, they can use call_rcu() or kfree_rcu(), which will be discussed later. Second, even when using synchronize_rcu(), the other update-side code does run concurrently with readers, whether pre-existing or not.h]hXFirst, if updaters do not wish to be blocked by readers, they can use call_rcu() or kfree_rcu(), which will be discussed later. Second, even when using synchronize_rcu(), the other update-side code does run concurrently with readers, whether pre-existing or not.}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhKhjubah}(h]h ]h"]h$]h&]uh1j hjubah}(h]h ]h"]h$]h&]uh1jhjubeh}(h]h ]h"]h$]h&]uh1jhjubeh}(h]h ]h"]h$]h&]colsKuh1jhjubah}(h]h ]h"]h$]h&]uh1jhjhhhhhNubh)}(hThis scenario resembles one of the first uses of RCU in `DYNIX/ptx `__, which managed a distributed lock manager's transition into a state suitable for handling recovery from node failure, more or less as follows:h](h8This scenario resembles one of the first uses of RCU in }(hjhhhNhNubh)}(h3`DYNIX/ptx `__h]h DYNIX/ptx}(hjhhhNhNubah}(h]h ]h"]h$]h&]name DYNIX/ptxj#https://en.wikipedia.org/wiki/DYNIXuh1hhjubh, which managed a distributed lock manager’s transition into a state suitable for handling recovery from node failure, more or less as follows:}(hjhhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhKhjhhubj@)}(hX:: 1 #define STATE_NORMAL 0 2 #define STATE_WANT_RECOVERY 1 3 #define STATE_RECOVERING 2 4 #define STATE_WANT_NORMAL 3 5 6 int state = STATE_NORMAL; 7 8 void do_something_dlm(void) 9 { 10 int state_snap; 11 12 rcu_read_lock(); 13 state_snap = READ_ONCE(state); 14 if (state_snap == STATE_NORMAL) 15 do_something(); 16 else 17 do_something_carefully(); 18 rcu_read_unlock(); 19 } 20 21 void start_recovery(void) 22 { 23 WRITE_ONCE(state, STATE_WANT_RECOVERY); 24 synchronize_rcu(); 25 WRITE_ONCE(state, STATE_RECOVERING); 26 recovery(); 27 WRITE_ONCE(state, STATE_WANT_NORMAL); 28 synchronize_rcu(); 29 WRITE_ONCE(state, STATE_NORMAL); 30 } h]jF)}(hX 1 #define STATE_NORMAL 0 2 #define STATE_WANT_RECOVERY 1 3 #define STATE_RECOVERING 2 4 #define STATE_WANT_NORMAL 3 5 6 int state = STATE_NORMAL; 7 8 void do_something_dlm(void) 9 { 10 int state_snap; 11 12 rcu_read_lock(); 13 state_snap = READ_ONCE(state); 14 if (state_snap == STATE_NORMAL) 15 do_something(); 16 else 17 do_something_carefully(); 18 rcu_read_unlock(); 19 } 20 21 void start_recovery(void) 22 { 23 WRITE_ONCE(state, STATE_WANT_RECOVERY); 24 synchronize_rcu(); 25 WRITE_ONCE(state, STATE_RECOVERING); 26 recovery(); 27 WRITE_ONCE(state, STATE_WANT_NORMAL); 28 synchronize_rcu(); 29 WRITE_ONCE(state, STATE_NORMAL); 30 }h]hX 1 #define STATE_NORMAL 0 2 #define STATE_WANT_RECOVERY 1 3 #define STATE_RECOVERING 2 4 #define STATE_WANT_NORMAL 3 5 6 int state = STATE_NORMAL; 7 8 void do_something_dlm(void) 9 { 10 int state_snap; 11 12 rcu_read_lock(); 13 state_snap = READ_ONCE(state); 14 if (state_snap == STATE_NORMAL) 15 do_something(); 16 else 17 do_something_carefully(); 18 rcu_read_unlock(); 19 } 20 21 void start_recovery(void) 22 { 23 WRITE_ONCE(state, STATE_WANT_RECOVERY); 24 synchronize_rcu(); 25 WRITE_ONCE(state, STATE_RECOVERING); 26 recovery(); 27 WRITE_ONCE(state, STATE_WANT_NORMAL); 28 synchronize_rcu(); 29 WRITE_ONCE(state, STATE_NORMAL); 30 }}hjsbah}(h]h ]h"]h$]h&]jUjVuh1jEhhhKhjubah}(h]h ]h"]h$]h&]uh1j?hhhKhjhhubh)}(hThe RCU read-side critical section in do_something_dlm() works with the synchronize_rcu() in start_recovery() to guarantee that do_something() never runs concurrently with recovery(), but with little or no synchronization overhead in do_something_dlm().h]hThe RCU read-side critical section in do_something_dlm() works with the synchronize_rcu() in start_recovery() to guarantee that do_something() never runs concurrently with recovery(), but with little or no synchronization overhead in do_something_dlm().}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhKhjhhubj)}(hhh]j)}(hhh](j)}(hhh]h}(h]h ]h"]h$]h&]colwidthKGuh1jhjubj)}(hhh](j)}(hhh]j )}(hhh]h)}(h**Quick Quiz**:h](j)}(h**Quick Quiz**h]h Quick Quiz}(hj!hhhNhNubah}(h]h ]h"]h$]h&]uh1jhjubh:}(hjhhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhKhjubah}(h]h ]h"]h$]h&]uh1j hjubah}(h]h ]h"]h$]h&]uh1jhjubj)}(hhh]j )}(hhh]h)}(h0Why is the synchronize_rcu() on line 28 needed?h]h0Why is the synchronize_rcu() on line 28 needed?}(hjKhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhKhjHubah}(h]h ]h"]h$]h&]uh1j hjEubah}(h]h ]h"]h$]h&]uh1jhjubj)}(hhh]j )}(hhh]h)}(h **Answer**:h](j)}(h **Answer**h]hAnswer}(hjohhhNhNubah}(h]h ]h"]h$]h&]uh1jhjkubh:}(hjkhhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhKhjhubah}(h]h ]h"]h$]h&]uh1j hjeubah}(h]h ]h"]h$]h&]uh1jhjubj)}(hhh]j )}(hhh]h)}(hWithout that extra grace period, memory reordering could result in do_something_dlm() executing do_something() concurrently with the last bits of recovery().h]hWithout that extra grace period, memory reordering could result in do_something_dlm() executing do_something() concurrently with the last bits of recovery().}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhKhjubah}(h]h ]h"]h$]h&]uh1j hjubah}(h]h ]h"]h$]h&]uh1jhjubeh}(h]h ]h"]h$]h&]uh1jhjubeh}(h]h ]h"]h$]h&]colsKuh1jhjubah}(h]h ]h"]h$]h&]uh1jhjhhhhhNubh)}(hX"In order to avoid fatal problems such as deadlocks, an RCU read-side critical section must not contain calls to synchronize_rcu(). Similarly, an RCU read-side critical section must not contain anything that waits, directly or indirectly, on completion of an invocation of synchronize_rcu().h]hX"In order to avoid fatal problems such as deadlocks, an RCU read-side critical section must not contain calls to synchronize_rcu(). Similarly, an RCU read-side critical section must not contain anything that waits, directly or indirectly, on completion of an invocation of synchronize_rcu().}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhKhjhhubh)}(hX3Although RCU's grace-period guarantee is useful in and of itself, with `quite a few use cases `__, it would be good to be able to use RCU to coordinate read-side access to linked data structures. For this, the grace-period guarantee is not sufficient, as can be seen in function add_gp_buggy() below. We will look at the reader's code later, but in the meantime, just think of the reader as locklessly picking up the ``gp`` pointer, and, if the value loaded is non-\ ``NULL``, locklessly accessing the ``->a`` and ``->b`` fields.h](hIAlthough RCU’s grace-period guarantee is useful in and of itself, with }(hjhhhNhNubh)}(h<`quite a few use cases `__h]hquite a few use cases}(hjhhhNhNubah}(h]h ]h"]h$]h&]namequite a few use casesj https://lwn.net/Articles/573497/uh1hhjubhXB, it would be good to be able to use RCU to coordinate read-side access to linked data structures. For this, the grace-period guarantee is not sufficient, as can be seen in function add_gp_buggy() below. We will look at the reader’s code later, but in the meantime, just think of the reader as locklessly picking up the }(hjhhhNhNubj)}(h``gp``h]hgp}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1jhjubh, pointer, and, if the value loaded is non- }(hjhhhNhNubj)}(h``NULL``h]hNULL}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1jhjubh, locklessly accessing the }(hjhhhNhNubj)}(h``->a``h]h->a}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1jhjubh and }(hjhhhNhNubj)}(h``->b``h]h->b}(hj'hhhNhNubah}(h]h ]h"]h$]h&]uh1jhjubh fields.}(hjhhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhKhjhhubj@)}(hX:: 1 bool add_gp_buggy(int a, int b) 2 { 3 p = kmalloc_obj(*p); 4 if (!p) 5 return -ENOMEM; 6 spin_lock(&gp_lock); 7 if (rcu_access_pointer(gp)) { 8 spin_unlock(&gp_lock); 9 return false; 10 } 11 p->a = a; 12 p->b = a; 13 gp = p; /* ORDERING BUG */ 14 spin_unlock(&gp_lock); 15 return true; 16 } h]jF)}(hXM 1 bool add_gp_buggy(int a, int b) 2 { 3 p = kmalloc_obj(*p); 4 if (!p) 5 return -ENOMEM; 6 spin_lock(&gp_lock); 7 if (rcu_access_pointer(gp)) { 8 spin_unlock(&gp_lock); 9 return false; 10 } 11 p->a = a; 12 p->b = a; 13 gp = p; /* ORDERING BUG */ 14 spin_unlock(&gp_lock); 15 return true; 16 }h]hXM 1 bool add_gp_buggy(int a, int b) 2 { 3 p = kmalloc_obj(*p); 4 if (!p) 5 return -ENOMEM; 6 spin_lock(&gp_lock); 7 if (rcu_access_pointer(gp)) { 8 spin_unlock(&gp_lock); 9 return false; 10 } 11 p->a = a; 12 p->b = a; 13 gp = p; /* ORDERING BUG */ 14 spin_unlock(&gp_lock); 15 return true; 16 }}hjCsbah}(h]h ]h"]h$]h&]jUjVuh1jEhhhKhj?ubah}(h]h ]h"]h$]h&]uh1j?hhhKhjhhubh)}(hvThe problem is that both the compiler and weakly ordered CPUs are within their rights to reorder this code as follows:h]hvThe problem is that both the compiler and weakly ordered CPUs are within their rights to reorder this code as follows:}(hjWhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhKhjhhubj@)}(hX:: 1 bool add_gp_buggy_optimized(int a, int b) 2 { 3 p = kmalloc_obj(*p); 4 if (!p) 5 return -ENOMEM; 6 spin_lock(&gp_lock); 7 if (rcu_access_pointer(gp)) { 8 spin_unlock(&gp_lock); 9 return false; 10 } 11 gp = p; /* ORDERING BUG */ 12 p->a = a; 13 p->b = a; 14 spin_unlock(&gp_lock); 15 return true; 16 } h]jF)}(hXW 1 bool add_gp_buggy_optimized(int a, int b) 2 { 3 p = kmalloc_obj(*p); 4 if (!p) 5 return -ENOMEM; 6 spin_lock(&gp_lock); 7 if (rcu_access_pointer(gp)) { 8 spin_unlock(&gp_lock); 9 return false; 10 } 11 gp = p; /* ORDERING BUG */ 12 p->a = a; 13 p->b = a; 14 spin_unlock(&gp_lock); 15 return true; 16 }h]hXW 1 bool add_gp_buggy_optimized(int a, int b) 2 { 3 p = kmalloc_obj(*p); 4 if (!p) 5 return -ENOMEM; 6 spin_lock(&gp_lock); 7 if (rcu_access_pointer(gp)) { 8 spin_unlock(&gp_lock); 9 return false; 10 } 11 gp = p; /* ORDERING BUG */ 12 p->a = a; 13 p->b = a; 14 spin_unlock(&gp_lock); 15 return true; 16 }}hjisbah}(h]h ]h"]h$]h&]jUjVuh1jEhhhKhjeubah}(h]h ]h"]h$]h&]uh1j?hhhKhjhhubh)}(hXIf an RCU reader fetches ``gp`` just after ``add_gp_buggy_optimized`` executes line 11, it will see garbage in the ``->a`` and ``->b`` fields. And this is but one of many ways in which compiler and hardware optimizations could cause trouble. Therefore, we clearly need some way to prevent the compiler and the CPU from reordering in this manner, which brings us to the publish-subscribe guarantee discussed in the next section.h](hIf an RCU reader fetches }(hj}hhhNhNubj)}(h``gp``h]hgp}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1jhj}ubh just after }(hj}hhhNhNubj)}(h``add_gp_buggy_optimized``h]hadd_gp_buggy_optimized}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1jhj}ubh/ executes line 11, it will see garbage in the }(hj}hhhNhNubj)}(h``->a``h]h->a}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1jhj}ubh and }(hj}hhhNhNubj)}(h``->b``h]h->b}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1jhj}ubhX% fields. And this is but one of many ways in which compiler and hardware optimizations could cause trouble. Therefore, we clearly need some way to prevent the compiler and the CPU from reordering in this manner, which brings us to the publish-subscribe guarantee discussed in the next section.}(hj}hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhKhjhhubeh}(h]jPah ]h"]grace-period guaranteeah$]h&]uh1hhjhhhhhKAjKubh)}(hhh](h)}(hPublish/Subscribe Guaranteeh]hPublish/Subscribe Guarantee}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhjhhhhhKubh)}(hX*RCU's publish-subscribe guarantee allows data to be inserted into a linked data structure without disrupting RCU readers. The updater uses rcu_assign_pointer() to insert the new data, and readers use rcu_dereference() to access data, whether new or old. The following shows an example of insertion:h]hX,RCU’s publish-subscribe guarantee allows data to be inserted into a linked data structure without disrupting RCU readers. The updater uses rcu_assign_pointer() to insert the new data, and readers use rcu_dereference() to access data, whether new or old. The following shows an example of insertion:}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhjhhubj@)}(hX|:: 1 bool add_gp(int a, int b) 2 { 3 p = kmalloc_obj(*p); 4 if (!p) 5 return -ENOMEM; 6 spin_lock(&gp_lock); 7 if (rcu_access_pointer(gp)) { 8 spin_unlock(&gp_lock); 9 return false; 10 } 11 p->a = a; 12 p->b = a; 13 rcu_assign_pointer(gp, p); 14 spin_unlock(&gp_lock); 15 return true; 16 } h]jF)}(hXG 1 bool add_gp(int a, int b) 2 { 3 p = kmalloc_obj(*p); 4 if (!p) 5 return -ENOMEM; 6 spin_lock(&gp_lock); 7 if (rcu_access_pointer(gp)) { 8 spin_unlock(&gp_lock); 9 return false; 10 } 11 p->a = a; 12 p->b = a; 13 rcu_assign_pointer(gp, p); 14 spin_unlock(&gp_lock); 15 return true; 16 }h]hXG 1 bool add_gp(int a, int b) 2 { 3 p = kmalloc_obj(*p); 4 if (!p) 5 return -ENOMEM; 6 spin_lock(&gp_lock); 7 if (rcu_access_pointer(gp)) { 8 spin_unlock(&gp_lock); 9 return false; 10 } 11 p->a = a; 12 p->b = a; 13 rcu_assign_pointer(gp, p); 14 spin_unlock(&gp_lock); 15 return true; 16 }}hjsbah}(h]h ]h"]h$]h&]jUjVuh1jEhhhM hjubah}(h]h ]h"]h$]h&]uh1j?hhhMhjhhubh)}(hXThe rcu_assign_pointer() on line 13 is conceptually equivalent to a simple assignment statement, but also guarantees that its assignment will happen after the two assignments in lines 11 and 12, similar to the C11 ``memory_order_release`` store operation. It also prevents any number of “interesting” compiler optimizations, for example, the use of ``gp`` as a scratch location immediately preceding the assignment.h](hThe rcu_assign_pointer() on line 13 is conceptually equivalent to a simple assignment statement, but also guarantees that its assignment will happen after the two assignments in lines 11 and 12, similar to the C11 }(hjhhhNhNubj)}(h``memory_order_release``h]hmemory_order_release}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1jhjubhs store operation. It also prevents any number of “interesting” compiler optimizations, for example, the use of }(hjhhhNhNubj)}(h``gp``h]hgp}(hj+hhhNhNubah}(h]h ]h"]h$]h&]uh1jhjubh< as a scratch location immediately preceding the assignment.}(hjhhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhjhhubj)}(hhh]j)}(hhh](j)}(hhh]h}(h]h ]h"]h$]h&]colwidthKGuh1jhjFubj)}(hhh](j)}(hhh]j )}(hhh]h)}(h**Quick Quiz**:h](j)}(h**Quick Quiz**h]h Quick Quiz}(hj`hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj\ubh:}(hj\hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhM"hjYubah}(h]h ]h"]h$]h&]uh1j hjVubah}(h]h ]h"]h$]h&]uh1jhjSubj)}(hhh]j )}(hhh]h)}(hBut rcu_assign_pointer() does nothing to prevent the two assignments to ``p->a`` and ``p->b`` from being reordered. Can't that also cause problems?h](hHBut rcu_assign_pointer() does nothing to prevent the two assignments to }(hjhhhNhNubj)}(h``p->a``h]hp->a}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1jhjubh and }(hjhhhNhNubj)}(h``p->b``h]hp->b}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1jhjubh8 from being reordered. Can’t that also cause problems?}(hjhhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhM$hjubah}(h]h ]h"]h$]h&]uh1j hjubah}(h]h ]h"]h$]h&]uh1jhjSubj)}(hhh]j )}(hhh]h)}(h **Answer**:h](j)}(h **Answer**h]hAnswer}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1jhjubh:}(hjhhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhM(hjubah}(h]h ]h"]h$]h&]uh1j hjubah}(h]h ]h"]h$]h&]uh1jhjSubj)}(hhh]j )}(hhh]h)}(hNo, it cannot. The readers cannot see either of these two fields until the assignment to ``gp``, by which time both fields are fully initialized. So reordering the assignments to ``p->a`` and ``p->b`` cannot possibly cause any problems.h](hYNo, it cannot. The readers cannot see either of these two fields until the assignment to }(hjhhhNhNubj)}(h``gp``h]hgp}(hj hhhNhNubah}(h]h ]h"]h$]h&]uh1jhjubhT, by which time both fields are fully initialized. So reordering the assignments to }(hjhhhNhNubj)}(h``p->a``h]hp->a}(hj hhhNhNubah}(h]h ]h"]h$]h&]uh1jhjubh and }(hjhhhNhNubj)}(h``p->b``h]hp->b}(hj( hhhNhNubah}(h]h ]h"]h$]h&]uh1jhjubh$ cannot possibly cause any problems.}(hjhhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhM*hjubah}(h]h ]h"]h$]h&]uh1j hjubah}(h]h ]h"]h$]h&]uh1jhjSubeh}(h]h ]h"]h$]h&]uh1jhjFubeh}(h]h ]h"]h$]h&]colsKuh1jhjCubah}(h]h ]h"]h$]h&]uh1jhjhhhhhNubh)}(hIt is tempting to assume that the reader need not do anything special to control its accesses to the RCU-protected data, as shown in do_something_gp_buggy() below:h]hIt is tempting to assume that the reader need not do anything special to control its accesses to the RCU-protected data, as shown in do_something_gp_buggy() below:}(hj_ hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM0hjhhubj@)}(hX':: 1 bool do_something_gp_buggy(void) 2 { 3 rcu_read_lock(); 4 p = gp; /* OPTIMIZATIONS GALORE!!! */ 5 if (p) { 6 do_something(p->a, p->b); 7 rcu_read_unlock(); 8 return true; 9 } 10 rcu_read_unlock(); 11 return false; 12 } h]jF)}(h 1 bool do_something_gp_buggy(void) 2 { 3 rcu_read_lock(); 4 p = gp; /* OPTIMIZATIONS GALORE!!! */ 5 if (p) { 6 do_something(p->a, p->b); 7 rcu_read_unlock(); 8 return true; 9 } 10 rcu_read_unlock(); 11 return false; 12 }h]h 1 bool do_something_gp_buggy(void) 2 { 3 rcu_read_lock(); 4 p = gp; /* OPTIMIZATIONS GALORE!!! */ 5 if (p) { 6 do_something(p->a, p->b); 7 rcu_read_unlock(); 8 return true; 9 } 10 rcu_read_unlock(); 11 return false; 12 }}hjq sbah}(h]h ]h"]h$]h&]jUjVuh1jEhhhM6hjm ubah}(h]h ]h"]h$]h&]uh1j?hhhM4hjhhubh)}(hXQHowever, this temptation must be resisted because there are a surprisingly large number of ways that the compiler (or weak ordering CPUs like the DEC Alpha) can trip this code up. For but one example, if the compiler were short of registers, it might choose to refetch from ``gp`` rather than keeping a separate copy in ``p`` as follows:h](hXHowever, this temptation must be resisted because there are a surprisingly large number of ways that the compiler (or weak ordering CPUs like the DEC Alpha) can trip this code up. For but one example, if the compiler were short of registers, it might choose to refetch from }(hj hhhNhNubj)}(h``gp``h]hgp}(hj hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj ubh( rather than keeping a separate copy in }(hj hhhNhNubj)}(h``p``h]hp}(hj hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj ubh as follows:}(hj hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMChjhhubj@)}(hX#:: 1 bool do_something_gp_buggy_optimized(void) 2 { 3 rcu_read_lock(); 4 if (gp) { /* OPTIMIZATIONS GALORE!!! */ 5 do_something(gp->a, gp->b); 6 rcu_read_unlock(); 7 return true; 8 } 9 rcu_read_unlock(); 10 return false; 11 } h]jF)}(h 1 bool do_something_gp_buggy_optimized(void) 2 { 3 rcu_read_lock(); 4 if (gp) { /* OPTIMIZATIONS GALORE!!! */ 5 do_something(gp->a, gp->b); 6 rcu_read_unlock(); 7 return true; 8 } 9 rcu_read_unlock(); 10 return false; 11 }h]h 1 bool do_something_gp_buggy_optimized(void) 2 { 3 rcu_read_lock(); 4 if (gp) { /* OPTIMIZATIONS GALORE!!! */ 5 do_something(gp->a, gp->b); 6 rcu_read_unlock(); 7 return true; 8 } 9 rcu_read_unlock(); 10 return false; 11 }}hj sbah}(h]h ]h"]h$]h&]jUjVuh1jEhhhMKhj ubah}(h]h ]h"]h$]h&]uh1j?hhhMIhjhhubh)}(hXQIf this function ran concurrently with a series of updates that replaced the current structure with a new one, the fetches of ``gp->a`` and ``gp->b`` might well come from two different structures, which could cause serious confusion. To prevent this (and much else besides), do_something_gp() uses rcu_dereference() to fetch from ``gp``:h](h~If this function ran concurrently with a series of updates that replaced the current structure with a new one, the fetches of }(hj hhhNhNubj)}(h ``gp->a``h]hgp->a}(hj hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj ubh and }(hj hhhNhNubj)}(h ``gp->b``h]hgp->b}(hj hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj ubh might well come from two different structures, which could cause serious confusion. To prevent this (and much else besides), do_something_gp() uses rcu_dereference() to fetch from }(hj hhhNhNubj)}(h``gp``h]hgp}(hj hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj ubh:}(hj hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMWhjhhubj@)}(hX:: 1 bool do_something_gp(void) 2 { 3 rcu_read_lock(); 4 p = rcu_dereference(gp); 5 if (p) { 6 do_something(p->a, p->b); 7 rcu_read_unlock(); 8 return true; 9 } 10 rcu_read_unlock(); 11 return false; 12 } h]jF)}(h 1 bool do_something_gp(void) 2 { 3 rcu_read_lock(); 4 p = rcu_dereference(gp); 5 if (p) { 6 do_something(p->a, p->b); 7 rcu_read_unlock(); 8 return true; 9 } 10 rcu_read_unlock(); 11 return false; 12 }h]h 1 bool do_something_gp(void) 2 { 3 rcu_read_lock(); 4 p = rcu_dereference(gp); 5 if (p) { 6 do_something(p->a, p->b); 7 rcu_read_unlock(); 8 return true; 9 } 10 rcu_read_unlock(); 11 return false; 12 }}hj sbah}(h]h ]h"]h$]h&]jUjVuh1jEhhhM_hj ubah}(h]h ]h"]h$]h&]uh1j?hhhM]hjhhubh)}(hX}The rcu_dereference() uses volatile casts and (for DEC Alpha) memory barriers in the Linux kernel. Should a |high-quality implementation of C11 memory_order_consume [PDF]|_ ever appear, then rcu_dereference() could be implemented as a ``memory_order_consume`` load. Regardless of the exact implementation, a pointer fetched by rcu_dereference() may not be used outside of the outermost RCU read-side critical section containing that rcu_dereference(), unless protection of the corresponding data element has been passed from RCU to some other synchronization mechanism, most commonly locking or reference counting (see ../../rcuref.rst).h](hlThe rcu_dereference() uses volatile casts and (for DEC Alpha) memory barriers in the Linux kernel. Should a }(hj+ hhhNhNubh)}(h@|high-quality implementation of C11 memory_order_consume [PDF]|_h](h#high-quality implementation of C11 }(hj3 hhhNhNubj)}(h``memory_order_consume``h]hmemory_order_consume}(hj; hhhNhNubah}(h]h ]h"]h$]h&]uh1jhNhNhj3 hhubh [PDF]}(hj3 hhhNhNubeh}(h]h ]h"]h$]h&]j>http://www.rdrop.com/users/paulmck/RCU/consume.2015.07.13a.pdfuh1hhj+ jKubh? ever appear, then rcu_dereference() could be implemented as a }(hj+ hhhNhNubj)}(h``memory_order_consume``h]hmemory_order_consume}(hjX hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj+ ubhXz load. Regardless of the exact implementation, a pointer fetched by rcu_dereference() may not be used outside of the outermost RCU read-side critical section containing that rcu_dereference(), unless protection of the corresponding data element has been passed from RCU to some other synchronization mechanism, most commonly locking or reference counting (see ../../rcuref.rst).}(hj+ hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMlhjhhubhsubstitution_definition)}(h.. |high-quality implementation of C11 memory_order_consume [PDF]| replace:: high-quality implementation of C11 ``memory_order_consume`` [PDF]h](h#high-quality implementation of C11 }hjr sbj)}(hj= h]hmemory_order_consume}hjz sbah}(h]h ]h"]h$]h&]uh1jhjr ubh [PDF]}hjr sbeh}(h]h ]h"]=high-quality implementation of C11 memory_order_consume [PDF]ah$]h&]uh1jp hhhMxhjhhubj)}(h.. _high-quality implementation of C11 memory_order_consume [PDF]: http://www.rdrop.com/users/paulmck/RCU/consume.2015.07.13a.pdfh]h}(h];high-quality-implementation-of-c11-memory-order-consume-pdfah ]h"]=high-quality implementation of c11 memory_order_consume [pdf]ah$]h&]jjS uh1jhMyhjhhhhjKubh)}(hIn short, updaters use rcu_assign_pointer() and readers use rcu_dereference(), and these two RCU API elements work together to ensure that readers have a consistent view of newly added data elements.h]hIn short, updaters use rcu_assign_pointer() and readers use rcu_dereference(), and these two RCU API elements work together to ensure that readers have a consistent view of newly added data elements.}(hj hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM{hjhhubh)}(hOf course, it is also necessary to remove elements from RCU-protected data structures, for example, using the following process:h]hOf course, it is also necessary to remove elements from RCU-protected data structures, for example, using the following process:}(hj hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhjhhubj)}(hhh](j)}(h5Remove the data element from the enclosing structure.h]h)}(hj h]h5Remove the data element from the enclosing structure.}(hj hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj ubah}(h]h ]h"]h$]h&]uh1jhj hhhhhNubj)}(hWait for all pre-existing RCU read-side critical sections to complete (because only pre-existing readers can possibly have a reference to the newly removed data element).h]h)}(hWait for all pre-existing RCU read-side critical sections to complete (because only pre-existing readers can possibly have a reference to the newly removed data element).h]hWait for all pre-existing RCU read-side critical sections to complete (because only pre-existing readers can possibly have a reference to the newly removed data element).}(hj hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj ubah}(h]h ]h"]h$]h&]uh1jhj hhhhhNubj)}(hAt this point, only the updater has a reference to the newly removed data element, so it can safely reclaim the data element, for example, by passing it to kfree(). h]h)}(hAt this point, only the updater has a reference to the newly removed data element, so it can safely reclaim the data element, for example, by passing it to kfree().h]hAt this point, only the updater has a reference to the newly removed data element, so it can safely reclaim the data element, for example, by passing it to kfree().}(hj hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj ubah}(h]h ]h"]h$]h&]uh1jhj hhhhhNubeh}(h]h ]h"]h$]h&]jjjhjjuh1jhjhhhhhMubh)}(h7This process is implemented by remove_gp_synchronous():h]h7This process is implemented by remove_gp_synchronous():}(hj hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhjhhubj@)}(hXu:: 1 bool remove_gp_synchronous(void) 2 { 3 struct foo *p; 4 5 spin_lock(&gp_lock); 6 p = rcu_access_pointer(gp); 7 if (!p) { 8 spin_unlock(&gp_lock); 9 return false; 10 } 11 rcu_assign_pointer(gp, NULL); 12 spin_unlock(&gp_lock); 13 synchronize_rcu(); 14 kfree(p); 15 return true; 16 } h]jF)}(hX@ 1 bool remove_gp_synchronous(void) 2 { 3 struct foo *p; 4 5 spin_lock(&gp_lock); 6 p = rcu_access_pointer(gp); 7 if (!p) { 8 spin_unlock(&gp_lock); 9 return false; 10 } 11 rcu_assign_pointer(gp, NULL); 12 spin_unlock(&gp_lock); 13 synchronize_rcu(); 14 kfree(p); 15 return true; 16 }h]hX@ 1 bool remove_gp_synchronous(void) 2 { 3 struct foo *p; 4 5 spin_lock(&gp_lock); 6 p = rcu_access_pointer(gp); 7 if (!p) { 8 spin_unlock(&gp_lock); 9 return false; 10 } 11 rcu_assign_pointer(gp, NULL); 12 spin_unlock(&gp_lock); 13 synchronize_rcu(); 14 kfree(p); 15 return true; 16 }}hj sbah}(h]h ]h"]h$]h&]jUjVuh1jEhhhMhj ubah}(h]h ]h"]h$]h&]uh1j?hhhMhjhhubh)}(hXHThis function is straightforward, with line 13 waiting for a grace period before line 14 frees the old data element. This waiting ensures that readers will reach line 7 of do_something_gp() before the data element referenced by ``p`` is freed. The rcu_access_pointer() on line 6 is similar to rcu_dereference(), except that:h](hThis function is straightforward, with line 13 waiting for a grace period before line 14 frees the old data element. This waiting ensures that readers will reach line 7 of do_something_gp() before the data element referenced by }(hj0 hhhNhNubj)}(h``p``h]hp}(hj8 hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj0 ubh\ is freed. The rcu_access_pointer() on line 6 is similar to rcu_dereference(), except that:}(hj0 hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhjhhubj)}(hhh](j)}(hThe value returned by rcu_access_pointer() cannot be dereferenced. If you want to access the value pointed to as well as the pointer itself, use rcu_dereference() instead of rcu_access_pointer().h]h)}(hThe value returned by rcu_access_pointer() cannot be dereferenced. If you want to access the value pointed to as well as the pointer itself, use rcu_dereference() instead of rcu_access_pointer().h]hThe value returned by rcu_access_pointer() cannot be dereferenced. If you want to access the value pointed to as well as the pointer itself, use rcu_dereference() instead of rcu_access_pointer().}(hjW hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhjS ubah}(h]h ]h"]h$]h&]uh1jhjP hhhhhNubj)}(hX The call to rcu_access_pointer() need not be protected. In contrast, rcu_dereference() must either be within an RCU read-side critical section or in a code segment where the pointer cannot change, for example, in code protected by the corresponding update-side lock. h]h)}(hX The call to rcu_access_pointer() need not be protected. In contrast, rcu_dereference() must either be within an RCU read-side critical section or in a code segment where the pointer cannot change, for example, in code protected by the corresponding update-side lock.h]hX The call to rcu_access_pointer() need not be protected. In contrast, rcu_dereference() must either be within an RCU read-side critical section or in a code segment where the pointer cannot change, for example, in code protected by the corresponding update-side lock.}(hjo hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhjk ubah}(h]h ]h"]h$]h&]uh1jhjP hhhhhNubeh}(h]h ]h"]h$]h&]jjjhjjuh1jhjhhhhhMubj)}(hhh]j)}(hhh](j)}(hhh]h}(h]h ]h"]h$]h&]colwidthKGuh1jhj ubj)}(hhh](j)}(hhh]j )}(hhh]h)}(h**Quick Quiz**:h](j)}(h**Quick Quiz**h]h Quick Quiz}(hj hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj ubh:}(hj hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhj ubah}(h]h ]h"]h$]h&]uh1j hj ubah}(h]h ]h"]h$]h&]uh1jhj ubj)}(hhh]j )}(hhh]h)}(hyWithout the rcu_dereference() or the rcu_access_pointer(), what destructive optimizations might the compiler make use of?h]hyWithout the rcu_dereference() or the rcu_access_pointer(), what destructive optimizations might the compiler make use of?}(hj hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj ubah}(h]h ]h"]h$]h&]uh1j hj ubah}(h]h ]h"]h$]h&]uh1jhj ubj)}(hhh]j )}(hhh]h)}(h **Answer**:h](j)}(h **Answer**h]hAnswer}(hj hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj ubh:}(hj hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhj ubah}(h]h ]h"]h$]h&]uh1j hj ubah}(h]h ]h"]h$]h&]uh1jhj ubj)}(hhh]j )}(hhh]h)}(hXLet's start with what happens to do_something_gp() if it fails to use rcu_dereference(). It could reuse a value formerly fetched from this same pointer. It could also fetch the pointer from ``gp`` in a byte-at-a-time manner, resulting in *load tearing*, in turn resulting a bytewise mash-up of two distinct pointer values. It might even use value-speculation optimizations, where it makes a wrong guess, but by the time it gets around to checking the value, an update has changed the pointer to match the wrong guess. Too bad about any dereferences that returned pre-initialization garbage in the meantime! For remove_gp_synchronous(), as long as all modifications to ``gp`` are carried out while holding ``gp_lock``, the above optimizations are harmless. However, ``sparse`` will complain if you define ``gp`` with ``__rcu`` and then access it without using either rcu_access_pointer() or rcu_dereference().h](hLet’s start with what happens to do_something_gp() if it fails to use rcu_dereference(). It could reuse a value formerly fetched from this same pointer. It could also fetch the pointer from }(hj hhhNhNubj)}(h``gp``h]hgp}(hj& hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj ubh* in a byte-at-a-time manner, resulting in }(hj hhhNhNubhemphasis)}(h*load tearing*h]h load tearing}(hj: hhhNhNubah}(h]h ]h"]h$]h&]uh1j8 hj ubhX, in turn resulting a bytewise mash-up of two distinct pointer values. It might even use value-speculation optimizations, where it makes a wrong guess, but by the time it gets around to checking the value, an update has changed the pointer to match the wrong guess. Too bad about any dereferences that returned pre-initialization garbage in the meantime! For remove_gp_synchronous(), as long as all modifications to }(hj hhhNhNubj)}(h``gp``h]hgp}(hjL hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj ubh are carried out while holding }(hj hhhNhNubj)}(h ``gp_lock``h]hgp_lock}(hj^ hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj ubh1, the above optimizations are harmless. However, }(hj hhhNhNubj)}(h ``sparse``h]hsparse}(hjp hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj ubh will complain if you define }(hj hhhNhNubj)}(h``gp``h]hgp}(hj hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj ubh with }(hj hhhNhNubj)}(h ``__rcu``h]h__rcu}(hj hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj ubhS and then access it without using either rcu_access_pointer() or rcu_dereference().}(hj hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhj ubah}(h]h ]h"]h$]h&]uh1j hj ubah}(h]h ]h"]h$]h&]uh1jhj ubeh}(h]h ]h"]h$]h&]uh1jhj ubeh}(h]h ]h"]h$]h&]colsKuh1jhj ubah}(h]h ]h"]h$]h&]uh1jhjhhhhhNubh)}(hXIn short, RCU's publish-subscribe guarantee is provided by the combination of rcu_assign_pointer() and rcu_dereference(). This guarantee allows data elements to be safely added to RCU-protected linked data structures without disrupting RCU readers. This guarantee can be used in combination with the grace-period guarantee to also allow data elements to be removed from RCU-protected linked data structures, again without disrupting RCU readers.h]hXIn short, RCU’s publish-subscribe guarantee is provided by the combination of rcu_assign_pointer() and rcu_dereference(). This guarantee allows data elements to be safely added to RCU-protected linked data structures without disrupting RCU readers. This guarantee can be used in combination with the grace-period guarantee to also allow data elements to be removed from RCU-protected linked data structures, again without disrupting RCU readers.}(hj hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhjhhubh)}(hXThis guarantee was only partially premeditated. DYNIX/ptx used an explicit memory barrier for publication, but had nothing resembling rcu_dereference() for subscription, nor did it have anything resembling the dependency-ordering barrier that was later subsumed into rcu_dereference() and later still into READ_ONCE(). The need for these operations made itself known quite suddenly at a late-1990s meeting with the DEC Alpha architects, back in the days when DEC was still a free-standing company. It took the Alpha architects a good hour to convince me that any sort of barrier would ever be needed, and it then took me a good *two* hours to convince them that their documentation did not make this point clear. More recent work with the C and C++ standards committees have provided much education on tricks and traps from the compiler. In short, compilers were much less tricky in the early 1990s, but in 2015, don't even think about omitting rcu_dereference()!h](hXtThis guarantee was only partially premeditated. DYNIX/ptx used an explicit memory barrier for publication, but had nothing resembling rcu_dereference() for subscription, nor did it have anything resembling the dependency-ordering barrier that was later subsumed into rcu_dereference() and later still into READ_ONCE(). The need for these operations made itself known quite suddenly at a late-1990s meeting with the DEC Alpha architects, back in the days when DEC was still a free-standing company. It took the Alpha architects a good hour to convince me that any sort of barrier would ever be needed, and it then took me a good }(hj hhhNhNubj9 )}(h*two*h]htwo}(hj hhhNhNubah}(h]h ]h"]h$]h&]uh1j8 hj ubhXL hours to convince them that their documentation did not make this point clear. More recent work with the C and C++ standards committees have provided much education on tricks and traps from the compiler. In short, compilers were much less tricky in the early 1990s, but in 2015, don’t even think about omitting rcu_dereference()!}(hj hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhjhhubeh}(h]jsah ]h"]publish/subscribe guaranteeah$]h&]uh1hhjhhhhhKjKubh)}(hhh](h)}(hMemory-Barrier Guaranteesh]hMemory-Barrier Guarantees}(hj hhhNhNubah}(h]h ]h"]h$]h&]uh1hhj hhhhhMubh)}(hThe previous section's simple linked-data-structure scenario clearly demonstrates the need for RCU's stringent memory-ordering guarantees on systems with more than one CPU:h]hThe previous section’s simple linked-data-structure scenario clearly demonstrates the need for RCU’s stringent memory-ordering guarantees on systems with more than one CPU:}(hj hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj hhubj)}(hhh](j)}(hXEach CPU that has an RCU read-side critical section that begins before synchronize_rcu() starts is guaranteed to execute a full memory barrier between the time that the RCU read-side critical section ends and the time that synchronize_rcu() returns. Without this guarantee, a pre-existing RCU read-side critical section might hold a reference to the newly removed ``struct foo`` after the kfree() on line 14 of remove_gp_synchronous().h]h)}(hXEach CPU that has an RCU read-side critical section that begins before synchronize_rcu() starts is guaranteed to execute a full memory barrier between the time that the RCU read-side critical section ends and the time that synchronize_rcu() returns. Without this guarantee, a pre-existing RCU read-side critical section might hold a reference to the newly removed ``struct foo`` after the kfree() on line 14 of remove_gp_synchronous().h](hXlEach CPU that has an RCU read-side critical section that begins before synchronize_rcu() starts is guaranteed to execute a full memory barrier between the time that the RCU read-side critical section ends and the time that synchronize_rcu() returns. Without this guarantee, a pre-existing RCU read-side critical section might hold a reference to the newly removed }(hj& hhhNhNubj)}(h``struct foo``h]h struct foo}(hj. hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj& ubh: after the kfree() on line 14 of remove_gp_synchronous().}(hj& hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhj" ubah}(h]h ]h"]h$]h&]uh1jhj hhhhhNubj)}(hXEach CPU that has an RCU read-side critical section that ends after synchronize_rcu() returns is guaranteed to execute a full memory barrier between the time that synchronize_rcu() begins and the time that the RCU read-side critical section begins. Without this guarantee, a later RCU read-side critical section running after the kfree() on line 14 of remove_gp_synchronous() might later run do_something_gp() and find the newly deleted ``struct foo``.h]h)}(hXEach CPU that has an RCU read-side critical section that ends after synchronize_rcu() returns is guaranteed to execute a full memory barrier between the time that synchronize_rcu() begins and the time that the RCU read-side critical section begins. Without this guarantee, a later RCU read-side critical section running after the kfree() on line 14 of remove_gp_synchronous() might later run do_something_gp() and find the newly deleted ``struct foo``.h](hXEach CPU that has an RCU read-side critical section that ends after synchronize_rcu() returns is guaranteed to execute a full memory barrier between the time that synchronize_rcu() begins and the time that the RCU read-side critical section begins. Without this guarantee, a later RCU read-side critical section running after the kfree() on line 14 of remove_gp_synchronous() might later run do_something_gp() and find the newly deleted }(hjP hhhNhNubj)}(h``struct foo``h]h struct foo}(hjX hhhNhNubah}(h]h ]h"]h$]h&]uh1jhjP ubh.}(hjP hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhjL ubah}(h]h ]h"]h$]h&]uh1jhj hhhhhNubj)}(hX1If the task invoking synchronize_rcu() remains on a given CPU, then that CPU is guaranteed to execute a full memory barrier sometime during the execution of synchronize_rcu(). This guarantee ensures that the kfree() on line 14 of remove_gp_synchronous() really does execute after the removal on line 11.h]h)}(hX1If the task invoking synchronize_rcu() remains on a given CPU, then that CPU is guaranteed to execute a full memory barrier sometime during the execution of synchronize_rcu(). This guarantee ensures that the kfree() on line 14 of remove_gp_synchronous() really does execute after the removal on line 11.h]hX1If the task invoking synchronize_rcu() remains on a given CPU, then that CPU is guaranteed to execute a full memory barrier sometime during the execution of synchronize_rcu(). This guarantee ensures that the kfree() on line 14 of remove_gp_synchronous() really does execute after the removal on line 11.}(hjz hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhjv ubah}(h]h ]h"]h$]h&]uh1jhj hhhhhNubj)}(hXIf the task invoking synchronize_rcu() migrates among a group of CPUs during that invocation, then each of the CPUs in that group is guaranteed to execute a full memory barrier sometime during the execution of synchronize_rcu(). This guarantee also ensures that the kfree() on line 14 of remove_gp_synchronous() really does execute after the removal on line 11, but also in the case where the thread executing the synchronize_rcu() migrates in the meantime. h]h)}(hXIf the task invoking synchronize_rcu() migrates among a group of CPUs during that invocation, then each of the CPUs in that group is guaranteed to execute a full memory barrier sometime during the execution of synchronize_rcu(). This guarantee also ensures that the kfree() on line 14 of remove_gp_synchronous() really does execute after the removal on line 11, but also in the case where the thread executing the synchronize_rcu() migrates in the meantime.h]hXIf the task invoking synchronize_rcu() migrates among a group of CPUs during that invocation, then each of the CPUs in that group is guaranteed to execute a full memory barrier sometime during the execution of synchronize_rcu(). This guarantee also ensures that the kfree() on line 14 of remove_gp_synchronous() really does execute after the removal on line 11, but also in the case where the thread executing the synchronize_rcu() migrates in the meantime.}(hj hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj ubah}(h]h ]h"]h$]h&]uh1jhj hhhhhNubeh}(h]h ]h"]h$]h&]jjjhjjuh1jhj hhhhhMubj)}(hhh]j)}(hhh](j)}(hhh]h}(h]h ]h"]h$]h&]colwidthKGuh1jhj ubj)}(hhh](j)}(hhh]j )}(hhh]h)}(h**Quick Quiz**:h](j)}(h**Quick Quiz**h]h Quick Quiz}(hj hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj ubh:}(hj hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhj ubah}(h]h ]h"]h$]h&]uh1j hj ubah}(h]h ]h"]h$]h&]uh1jhj ubj)}(hhh]j )}(hhh]h)}(hGiven that multiple CPUs can start RCU read-side critical sections at any time without any ordering whatsoever, how can RCU possibly tell whether or not a given RCU read-side critical section starts before a given instance of synchronize_rcu()?h]hGiven that multiple CPUs can start RCU read-side critical sections at any time without any ordering whatsoever, how can RCU possibly tell whether or not a given RCU read-side critical section starts before a given instance of synchronize_rcu()?}(hj hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj ubah}(h]h ]h"]h$]h&]uh1j hj ubah}(h]h ]h"]h$]h&]uh1jhj ubj)}(hhh]j )}(hhh]h)}(h **Answer**:h](j)}(h **Answer**h]hAnswer}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1jhjubh:}(hjhhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhM hjubah}(h]h ]h"]h$]h&]uh1j hj ubah}(h]h ]h"]h$]h&]uh1jhj ubj)}(hhh]j )}(hhh](h)}(hX9If RCU cannot tell whether or not a given RCU read-side critical section starts before a given instance of synchronize_rcu(), then it must assume that the RCU read-side critical section started first. In other words, a given instance of synchronize_rcu() can avoid waiting on a given RCU read-side critical section only if it can prove that synchronize_rcu() started first. A related question is “When rcu_read_lock() doesn't generate any code, why does it matter how it relates to a grace period?” The answer is that it is not the relationship of rcu_read_lock() itself that is important, but rather the relationship of the code within the enclosed RCU read-side critical section to the code preceding and following the grace period. If we take this viewpoint, then a given RCU read-side critical section begins before a given grace period when some access preceding the grace period observes the effect of some access within the critical section, in which case none of the accesses within the critical section may observe the effects of any access following the grace period.h]hX;If RCU cannot tell whether or not a given RCU read-side critical section starts before a given instance of synchronize_rcu(), then it must assume that the RCU read-side critical section started first. In other words, a given instance of synchronize_rcu() can avoid waiting on a given RCU read-side critical section only if it can prove that synchronize_rcu() started first. A related question is “When rcu_read_lock() doesn’t generate any code, why does it matter how it relates to a grace period?” The answer is that it is not the relationship of rcu_read_lock() itself that is important, but rather the relationship of the code within the enclosed RCU read-side critical section to the code preceding and following the grace period. If we take this viewpoint, then a given RCU read-side critical section begins before a given grace period when some access preceding the grace period observes the effect of some access within the critical section, in which case none of the accesses within the critical section may observe the effects of any access following the grace period.}(hjAhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM hj>ubh)}(hAs of late 2016, mathematical models of RCU take this viewpoint, for example, see slides 62 and 63 of the `2016 LinuxCon EU `__ presentation.h](hlAs of late 2016, mathematical models of RCU take this viewpoint, for example, see slides 62 and 63 of the }(hjOhhhNhNubh)}(hi`2016 LinuxCon EU `__h]h2016 LinuxCon EU}(hjWhhhNhNubah}(h]h ]h"]h$]h&]name2016 LinuxCon EUjQhttp://www2.rdrop.com/users/paulmck/scalability/paper/LinuxMM.2016.10.04c.LCE.pdfuh1hhjOubh presentation.}(hjOhhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhj>ubeh}(h]h ]h"]h$]h&]uh1j hj;ubah}(h]h ]h"]h$]h&]uh1jhj ubeh}(h]h ]h"]h$]h&]uh1jhj ubeh}(h]h ]h"]h$]h&]colsKuh1jhj ubah}(h]h ]h"]h$]h&]uh1jhj hhhhhNubj)}(hhh]j)}(hhh](j)}(hhh]h}(h]h ]h"]h$]h&]colwidthKGuh1jhjubj)}(hhh](j)}(hhh]j )}(hhh]h)}(h**Quick Quiz**:h](j)}(h**Quick Quiz**h]h Quick Quiz}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1jhjubh:}(hjhhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhM&hjubah}(h]h ]h"]h$]h&]uh1j hjubah}(h]h ]h"]h$]h&]uh1jhjubj)}(hhh]j )}(hhh]h)}(hvThe first and second guarantees require unbelievably strict ordering! Are all these memory barriers *really* required?h](hdThe first and second guarantees require unbelievably strict ordering! Are all these memory barriers }(hjhhhNhNubj9 )}(h*really*h]hreally}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1j8 hjubh required?}(hjhhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhM(hjubah}(h]h ]h"]h$]h&]uh1j hjubah}(h]h ]h"]h$]h&]uh1jhjubj)}(hhh]j )}(hhh]h)}(h **Answer**:h](j)}(h **Answer**h]hAnswer}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1jhj ubh:}(hj hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhM+hjubah}(h]h ]h"]h$]h&]uh1j hjubah}(h]h ]h"]h$]h&]uh1jhjubj)}(hhh]j )}(hhh](h)}(huYes, they really are required. To see why the first guarantee is required, consider the following sequence of events:h]huYes, they really are required. To see why the first guarantee is required, consider the following sequence of events:}(hj8hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM-hj5ubj)}(hhh](j)}(hCPU 1: rcu_read_lock()h]h)}(hjKh]hCPU 1: rcu_read_lock()}(hjMhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM0hjIubah}(h]h ]h"]h$]h&]uh1jhjFubj)}(hBCPU 1: ``q = rcu_dereference(gp); /* Very likely to return p. */``h]h)}(hjbh](hCPU 1: }(hjdhhhNhNubj)}(h;``q = rcu_dereference(gp); /* Very likely to return p. */``h]h7q = rcu_dereference(gp); /* Very likely to return p. */}(hjkhhhNhNubah}(h]h ]h"]h$]h&]uh1jhjdubeh}(h]h ]h"]h$]h&]uh1hhhhM1hj`ubah}(h]h ]h"]h$]h&]uh1jhjFubj)}(hCPU 0: ``list_del_rcu(p);``h]h)}(hjh](hCPU 0: }(hjhhhNhNubj)}(h``list_del_rcu(p);``h]hlist_del_rcu(p);}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1jhjubeh}(h]h ]h"]h$]h&]uh1hhhhM2hjubah}(h]h ]h"]h$]h&]uh1jhjFubj)}(h CPU 0: synchronize_rcu() starts.h]h)}(hjh]h CPU 0: synchronize_rcu() starts.}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM3hjubah}(h]h ]h"]h$]h&]uh1jhjFubj)}(hYCPU 1: ``do_something_with(q->a);`` ``/* No smp_mb(), so might happen after kfree(). */``h]h)}(hYCPU 1: ``do_something_with(q->a);`` ``/* No smp_mb(), so might happen after kfree(). */``h](hCPU 1: }(hjhhhNhNubj)}(h``do_something_with(q->a);``h]hdo_something_with(q->a);}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1jhjubh }(hjhhhNhNubj)}(h5``/* No smp_mb(), so might happen after kfree(). */``h]h1/* No smp_mb(), so might happen after kfree(). */}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1jhjubeh}(h]h ]h"]h$]h&]uh1hhhhM4hjubah}(h]h ]h"]h$]h&]uh1jhjFubj)}(hCPU 1: rcu_read_unlock()h]h)}(hjh]hCPU 1: rcu_read_unlock()}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM6hjubah}(h]h ]h"]h$]h&]uh1jhjFubj)}(h!CPU 0: synchronize_rcu() returns.h]h)}(hjh]h!CPU 0: synchronize_rcu() returns.}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM7hjubah}(h]h ]h"]h$]h&]uh1jhjFubj)}(hCPU 0: ``kfree(p);`` h]h)}(hCPU 0: ``kfree(p);``h](hCPU 0: }(hj+hhhNhNubj)}(h ``kfree(p);``h]h kfree(p);}(hj3hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj+ubeh}(h]h ]h"]h$]h&]uh1hhhhM8hj'ubah}(h]h ]h"]h$]h&]uh1jhjFubeh}(h]h ]h"]h$]h&]jjjhjjuh1jhj5ubh)}(hTherefore, there absolutely must be a full memory barrier between the end of the RCU read-side critical section and the end of the grace period.h]hTherefore, there absolutely must be a full memory barrier between the end of the RCU read-side critical section and the end of the grace period.}(hjShhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM:hj5ubh)}(hYThe sequence of events demonstrating the necessity of the second rule is roughly similar:h]hYThe sequence of events demonstrating the necessity of the second rule is roughly similar:}(hjahhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM>hj5ubj)}(hhh](j)}(hCPU 0: ``list_del_rcu(p);``h]h)}(hjth](hCPU 0: }(hjvhhhNhNubj)}(h``list_del_rcu(p);``h]hlist_del_rcu(p);}(hj}hhhNhNubah}(h]h ]h"]h$]h&]uh1jhjvubeh}(h]h ]h"]h$]h&]uh1hhhhMAhjrubah}(h]h ]h"]h$]h&]uh1jhjoubj)}(h CPU 0: synchronize_rcu() starts.h]h)}(hjh]h CPU 0: synchronize_rcu() starts.}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMBhjubah}(h]h ]h"]h$]h&]uh1jhjoubj)}(hCPU 1: rcu_read_lock()h]h)}(hjh]hCPU 1: rcu_read_lock()}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMChjubah}(h]h ]h"]h$]h&]uh1jhjoubj)}(hRCPU 1: ``q = rcu_dereference(gp);`` ``/* Might return p if no memory barrier. */``h]h)}(hRCPU 1: ``q = rcu_dereference(gp);`` ``/* Might return p if no memory barrier. */``h](hCPU 1: }(hjhhhNhNubj)}(h``q = rcu_dereference(gp);``h]hq = rcu_dereference(gp);}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1jhjubh }(hjhhhNhNubj)}(h.``/* Might return p if no memory barrier. */``h]h*/* Might return p if no memory barrier. */}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1jhjubeh}(h]h ]h"]h$]h&]uh1hhhhMDhjubah}(h]h ]h"]h$]h&]uh1jhjoubj)}(h!CPU 0: synchronize_rcu() returns.h]h)}(hjh]h!CPU 0: synchronize_rcu() returns.}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMFhjubah}(h]h ]h"]h$]h&]uh1jhjoubj)}(hCPU 0: ``kfree(p);``h]h)}(hjh](hCPU 0: }(hjhhhNhNubj)}(h ``kfree(p);``h]h kfree(p);}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1jhjubeh}(h]h ]h"]h$]h&]uh1hhhhMGhjubah}(h]h ]h"]h$]h&]uh1jhjoubj)}(h1CPU 1: ``do_something_with(q->a); /* Boom!!! */``h]h)}(hj;h](hCPU 1: }(hj=hhhNhNubj)}(h*``do_something_with(q->a); /* Boom!!! */``h]h&do_something_with(q->a); /* Boom!!! */}(hjDhhhNhNubah}(h]h ]h"]h$]h&]uh1jhj=ubeh}(h]h ]h"]h$]h&]uh1hhhhMHhj9ubah}(h]h ]h"]h$]h&]uh1jhjoubj)}(hCPU 1: rcu_read_unlock() h]h)}(hCPU 1: rcu_read_unlock()h]hCPU 1: rcu_read_unlock()}(hjbhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMIhj^ubah}(h]h ]h"]h$]h&]uh1jhjoubeh}(h]h ]h"]h$]h&]jjjhjjuh1jhj5ubh)}(hAnd similarly, without a memory barrier between the beginning of the grace period and the beginning of the RCU read-side critical section, CPU 1 might end up accessing the freelist.h]hAnd similarly, without a memory barrier between the beginning of the grace period and the beginning of the RCU read-side critical section, CPU 1 might end up accessing the freelist.}(hj|hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMKhj5ubh)}(hX'The “as if” rule of course applies, so that any implementation that acts as if the appropriate memory barriers were in place is a correct implementation. That said, it is much easier to fool yourself into believing that you have adhered to the as-if rule than it is to actually adhere to it!h]hX'The “as if” rule of course applies, so that any implementation that acts as if the appropriate memory barriers were in place is a correct implementation. That said, it is much easier to fool yourself into believing that you have adhered to the as-if rule than it is to actually adhere to it!}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMOhj5ubeh}(h]h ]h"]h$]h&]uh1j hj2ubah}(h]h ]h"]h$]h&]uh1jhjubeh}(h]h ]h"]h$]h&]uh1jhjubeh}(h]h ]h"]h$]h&]colsKuh1jhjubah}(h]h ]h"]h$]h&]uh1jhj hhhhhNubj)}(hhh]j)}(hhh](j)}(hhh]h}(h]h ]h"]h$]h&]colwidthKGuh1jhjubj)}(hhh](j)}(hhh]j )}(hhh]h)}(h**Quick Quiz**:h](j)}(h**Quick Quiz**h]h Quick Quiz}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1jhjubh:}(hjhhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMWhjubah}(h]h ]h"]h$]h&]uh1j hjubah}(h]h ]h"]h$]h&]uh1jhjubj)}(hhh]j )}(hhh]h)}(hXYou claim that rcu_read_lock() and rcu_read_unlock() generate absolutely no code in some kernel builds. This means that the compiler might arbitrarily rearrange consecutive RCU read-side critical sections. Given such rearrangement, if a given RCU read-side critical section is done, how can you be sure that all prior RCU read-side critical sections are done? Won't the compiler rearrangements make that impossible to determine?h]hXYou claim that rcu_read_lock() and rcu_read_unlock() generate absolutely no code in some kernel builds. This means that the compiler might arbitrarily rearrange consecutive RCU read-side critical sections. Given such rearrangement, if a given RCU read-side critical section is done, how can you be sure that all prior RCU read-side critical sections are done? Won’t the compiler rearrangements make that impossible to determine?}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMYhjubah}(h]h ]h"]h$]h&]uh1j hjubah}(h]h ]h"]h$]h&]uh1jhjubj)}(hhh]j )}(hhh]h)}(h **Answer**:h](j)}(h **Answer**h]hAnswer}(hj"hhhNhNubah}(h]h ]h"]h$]h&]uh1jhjubh:}(hjhhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMahjubah}(h]h ]h"]h$]h&]uh1j hjubah}(h]h ]h"]h$]h&]uh1jhjubj)}(hhh]j )}(hhh]h)}(hXIn cases where rcu_read_lock() and rcu_read_unlock() generate absolutely no code, RCU infers quiescent states only at special locations, for example, within the scheduler. Because calls to schedule() had better prevent calling-code accesses to shared variables from being rearranged across the call to schedule(), if RCU detects the end of a given RCU read-side critical section, it will necessarily detect the end of all prior RCU read-side critical sections, no matter how aggressively the compiler scrambles the code. Again, this all assumes that the compiler cannot scramble code across calls to the scheduler, out of interrupt handlers, into the idle loop, into user-mode code, and so on. But if your kernel build allows that sort of scrambling, you have broken far more than just RCU!h]hXIn cases where rcu_read_lock() and rcu_read_unlock() generate absolutely no code, RCU infers quiescent states only at special locations, for example, within the scheduler. Because calls to schedule() had better prevent calling-code accesses to shared variables from being rearranged across the call to schedule(), if RCU detects the end of a given RCU read-side critical section, it will necessarily detect the end of all prior RCU read-side critical sections, no matter how aggressively the compiler scrambles the code. Again, this all assumes that the compiler cannot scramble code across calls to the scheduler, out of interrupt handlers, into the idle loop, into user-mode code, and so on. But if your kernel build allows that sort of scrambling, you have broken far more than just RCU!}(hjLhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMchjIubah}(h]h ]h"]h$]h&]uh1j hjFubah}(h]h ]h"]h$]h&]uh1jhjubeh}(h]h ]h"]h$]h&]uh1jhjubeh}(h]h ]h"]h$]h&]colsKuh1jhjubah}(h]h ]h"]h$]h&]uh1jhj hhhhhNubh)}(hXNote that these memory-barrier requirements do not replace the fundamental RCU requirement that a grace period wait for all pre-existing readers. On the contrary, the memory barriers called out in this section must operate in such a way as to *enforce* this fundamental requirement. Of course, different implementations enforce this requirement in different ways, but enforce it they must.h](hNote that these memory-barrier requirements do not replace the fundamental RCU requirement that a grace period wait for all pre-existing readers. On the contrary, the memory barriers called out in this section must operate in such a way as to }(hjyhhhNhNubj9 )}(h *enforce*h]henforce}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1j8 hjyubh this fundamental requirement. Of course, different implementations enforce this requirement in different ways, but enforce it they must.}(hjyhhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMqhj hhubeh}(h]jah ]h"]memory-barrier guaranteesah$]h&]uh1hhjhhhhhMjKubh)}(hhh](h)}(h4RCU Primitives Guaranteed to Execute Unconditionallyh]h4RCU Primitives Guaranteed to Execute Unconditionally}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhjhhhhhMyubh)}(hThe common-case RCU primitives are unconditional. They are invoked, they do their job, and they return, with no possibility of error, and no need to retry. This is a key RCU design philosophy.h]hThe common-case RCU primitives are unconditional. They are invoked, they do their job, and they return, with no possibility of error, and no need to retry. This is a key RCU design philosophy.}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM{hjhhubh)}(hXfHowever, this philosophy is pragmatic rather than pigheaded. If someone comes up with a good justification for a particular conditional RCU primitive, it might well be implemented and added. After all, this guarantee was reverse-engineered, not premeditated. The unconditional nature of the RCU primitives was initially an accident of implementation, and later experience with synchronization primitives with conditional primitives caused me to elevate this accident to a guarantee. Therefore, the justification for adding a conditional primitive to RCU would need to be based on detailed and compelling use cases.h]hXfHowever, this philosophy is pragmatic rather than pigheaded. If someone comes up with a good justification for a particular conditional RCU primitive, it might well be implemented and added. After all, this guarantee was reverse-engineered, not premeditated. The unconditional nature of the RCU primitives was initially an accident of implementation, and later experience with synchronization primitives with conditional primitives caused me to elevate this accident to a guarantee. Therefore, the justification for adding a conditional primitive to RCU would need to be based on detailed and compelling use cases.}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhjhhubeh}(h]jah ]h"]4rcu primitives guaranteed to execute unconditionallyah$]h&]uh1hhjhhhhhMyjKubh)}(hhh](h)}(h Guaranteed Read-to-Write Upgradeh]h Guaranteed Read-to-Write Upgrade}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhjhhhhhMubh)}(hXGAs far as RCU is concerned, it is always possible to carry out an update within an RCU read-side critical section. For example, that RCU read-side critical section might search for a given data element, and then might acquire the update-side spinlock in order to update that element, all while remaining in that RCU read-side critical section. Of course, it is necessary to exit the RCU read-side critical section before invoking synchronize_rcu(), however, this inconvenience can be avoided through use of the call_rcu() and kfree_rcu() API members described later in this document.h]hXGAs far as RCU is concerned, it is always possible to carry out an update within an RCU read-side critical section. For example, that RCU read-side critical section might search for a given data element, and then might acquire the update-side spinlock in order to update that element, all while remaining in that RCU read-side critical section. Of course, it is necessary to exit the RCU read-side critical section before invoking synchronize_rcu(), however, this inconvenience can be avoided through use of the call_rcu() and kfree_rcu() API members described later in this document.}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhjhhubj)}(hhh]j)}(hhh](j)}(hhh]h}(h]h ]h"]h$]h&]colwidthKGuh1jhjubj)}(hhh](j)}(hhh]j )}(hhh]h)}(h**Quick Quiz**:h](j)}(h**Quick Quiz**h]h Quick Quiz}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1jhj ubh:}(hj hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhj ubah}(h]h ]h"]h$]h&]uh1j hjubah}(h]h ]h"]h$]h&]uh1jhjubj)}(hhh]j )}(hhh]h)}(hBBut how does the upgrade-to-write operation exclude other readers?h]hBBut how does the upgrade-to-write operation exclude other readers?}(hj:hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj7ubah}(h]h ]h"]h$]h&]uh1j hj4ubah}(h]h ]h"]h$]h&]uh1jhjubj)}(hhh]j )}(hhh]h)}(h **Answer**:h](j)}(h **Answer**h]hAnswer}(hj^hhhNhNubah}(h]h ]h"]h$]h&]uh1jhjZubh:}(hjZhhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhjWubah}(h]h ]h"]h$]h&]uh1j hjTubah}(h]h ]h"]h$]h&]uh1jhjubj)}(hhh]j )}(hhh]h)}(hPIt doesn't, just like normal RCU updates, which also do not exclude RCU readers.h]hRIt doesn’t, just like normal RCU updates, which also do not exclude RCU readers.}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhjubah}(h]h ]h"]h$]h&]uh1j hjubah}(h]h ]h"]h$]h&]uh1jhjubeh}(h]h ]h"]h$]h&]uh1jhjubeh}(h]h ]h"]h$]h&]colsKuh1jhjubah}(h]h ]h"]h$]h&]uh1jhjhhhhhNubh)}(hThis guarantee allows lookup code to be shared between read-side and update-side code, and was premeditated, appearing in the earliest DYNIX/ptx RCU documentation.h]hThis guarantee allows lookup code to be shared between read-side and update-side code, and was premeditated, appearing in the earliest DYNIX/ptx RCU documentation.}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhjhhubeh}(h]jah ]h"] guaranteed read-to-write upgradeah$]h&]uh1hhjhhhhhMjKubeh}(h]jah ]h"]fundamental requirementsah$]h&]uh1hhhhhhhhK5jKubh)}(hhh](h)}(hFundamental Non-Requirementsh]hFundamental Non-Requirements}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhjhhhhhMubh)}(hXRCU provides extremely lightweight readers, and its read-side guarantees, though quite useful, are correspondingly lightweight. It is therefore all too easy to assume that RCU is guaranteeing more than it really is. Of course, the list of things that RCU does not guarantee is infinitely long, however, the following sections list a few non-guarantees that have caused confusion. Except where otherwise noted, these non-guarantees were premeditated.h]hXRCU provides extremely lightweight readers, and its read-side guarantees, though quite useful, are correspondingly lightweight. It is therefore all too easy to assume that RCU is guaranteeing more than it really is. Of course, the list of things that RCU does not guarantee is infinitely long, however, the following sections list a few non-guarantees that have caused confusion. Except where otherwise noted, these non-guarantees were premeditated.}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhjhhubj)}(hhh](j)}(h"`Readers Impose Minimal Ordering`_h]h)}(hjh]h)}(hjh]hReaders Impose Minimal Ordering}(hjhhhNhNubah}(h]h ]h"]h$]h&]nameReaders Impose Minimal Orderingjreaders-impose-minimal-orderinguh1hhjjKubah}(h]h ]h"]h$]h&]uh1hhhhMhjubah}(h]h ]h"]h$]h&]uh1jhjhhhhhNubj)}(h"`Readers Do Not Exclude Updaters`_h]h)}(hjh]h)}(hjh]hReaders Do Not Exclude Updaters}(hjhhhNhNubah}(h]h ]h"]h$]h&]nameReaders Do Not Exclude Updatersjreaders-do-not-exclude-updatersuh1hhjjKubah}(h]h ]h"]h$]h&]uh1hhhhMhjubah}(h]h ]h"]h$]h&]uh1jhjhhhhhNubj)}(h%`Updaters Only Wait For Old Readers`_h]h)}(hj;h]h)}(hj;h]h"Updaters Only Wait For Old Readers}(hj@hhhNhNubah}(h]h ]h"]h$]h&]name"Updaters Only Wait For Old Readersj"updaters-only-wait-for-old-readersuh1hhj=jKubah}(h]h ]h"]h$]h&]uh1hhhhMhj9ubah}(h]h ]h"]h$]h&]uh1jhjhhhhhNubj)}(h<`Grace Periods Don't Partition Read-Side Critical Sections`_h]h)}(hj^h]h)}(hj^h]h;Grace Periods Don’t Partition Read-Side Critical Sections}(hjchhhNhNubah}(h]h ]h"]h$]h&]name9Grace Periods Don't Partition Read-Side Critical Sectionsj9grace-periods-don-t-partition-read-side-critical-sectionsuh1hhj`jKubah}(h]h ]h"]h$]h&]uh1hhhhMhj\ubah}(h]h ]h"]h$]h&]uh1jhjhhhhhNubj)}(h=`Read-Side Critical Sections Don't Partition Grace Periods`_ h]h)}(h<`Read-Side Critical Sections Don't Partition Grace Periods`_h]h)}(hjh]h;Read-Side Critical Sections Don’t Partition Grace Periods}(hjhhhNhNubah}(h]h ]h"]h$]h&]name9Read-Side Critical Sections Don't Partition Grace Periodsj9read-side-critical-sections-don-t-partition-grace-periodsuh1hhjjKubah}(h]h ]h"]h$]h&]uh1hhhhMhjubah}(h]h ]h"]h$]h&]uh1jhjhhhhhNubeh}(h]h ]h"]h$]h&]jjjhjjuh1jhjhhhhhMubh)}(hhh](h)}(hReaders Impose Minimal Orderingh]hReaders Impose Minimal Ordering}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhjhhhhhMubh)}(hReader-side markers such as rcu_read_lock() and rcu_read_unlock() provide absolutely no ordering guarantees except through their interaction with the grace-period APIs such as synchronize_rcu(). To see this, consider the following pair of threads:h]hReader-side markers such as rcu_read_lock() and rcu_read_unlock() provide absolutely no ordering guarantees except through their interaction with the grace-period APIs such as synchronize_rcu(). To see this, consider the following pair of threads:}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhjhhubj@)}(hX:: 1 void thread0(void) 2 { 3 rcu_read_lock(); 4 WRITE_ONCE(x, 1); 5 rcu_read_unlock(); 6 rcu_read_lock(); 7 WRITE_ONCE(y, 1); 8 rcu_read_unlock(); 9 } 10 11 void thread1(void) 12 { 13 rcu_read_lock(); 14 r1 = READ_ONCE(y); 15 rcu_read_unlock(); 16 rcu_read_lock(); 17 r2 = READ_ONCE(x); 18 rcu_read_unlock(); 19 } h]jF)}(hXX 1 void thread0(void) 2 { 3 rcu_read_lock(); 4 WRITE_ONCE(x, 1); 5 rcu_read_unlock(); 6 rcu_read_lock(); 7 WRITE_ONCE(y, 1); 8 rcu_read_unlock(); 9 } 10 11 void thread1(void) 12 { 13 rcu_read_lock(); 14 r1 = READ_ONCE(y); 15 rcu_read_unlock(); 16 rcu_read_lock(); 17 r2 = READ_ONCE(x); 18 rcu_read_unlock(); 19 }h]hXX 1 void thread0(void) 2 { 3 rcu_read_lock(); 4 WRITE_ONCE(x, 1); 5 rcu_read_unlock(); 6 rcu_read_lock(); 7 WRITE_ONCE(y, 1); 8 rcu_read_unlock(); 9 } 10 11 void thread1(void) 12 { 13 rcu_read_lock(); 14 r1 = READ_ONCE(y); 15 rcu_read_unlock(); 16 rcu_read_lock(); 17 r2 = READ_ONCE(x); 18 rcu_read_unlock(); 19 }}hjsbah}(h]h ]h"]h$]h&]jUjVuh1jEhhhMhjubah}(h]h ]h"]h$]h&]uh1j?hhhMhjhhubh)}(hPAfter thread0() and thread1() execute concurrently, it is quite possible to haveh]hPAfter thread0() and thread1() execute concurrently, it is quite possible to have}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhjhhubj@)}(h:: (r1 == 1 && r2 == 0) h]jF)}(h(r1 == 1 && r2 == 0)h]h(r1 == 1 && r2 == 0)}hjsbah}(h]h ]h"]h$]h&]jUjVuh1jEhhhMhjubah}(h]h ]h"]h$]h&]uh1j?hhhMhjhhubh)}(hXZ(that is, ``y`` appears to have been assigned before ``x``), which would not be possible if rcu_read_lock() and rcu_read_unlock() had much in the way of ordering properties. But they do not, so the CPU is within its rights to do significant reordering. This is by design: Any significant ordering constraints would slow down these fast-path APIs.h](h (that is, }(hjhhhNhNubj)}(h``y``h]hy}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1jhjubh& appears to have been assigned before }(hjhhhNhNubj)}(h``x``h]hx}(hj hhhNhNubah}(h]h ]h"]h$]h&]uh1jhjubhX ), which would not be possible if rcu_read_lock() and rcu_read_unlock() had much in the way of ordering properties. But they do not, so the CPU is within its rights to do significant reordering. This is by design: Any significant ordering constraints would slow down these fast-path APIs.}(hjhhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhjhhubj)}(hhh]j)}(hhh](j)}(hhh]h}(h]h ]h"]h$]h&]colwidthKGuh1jhj;ubj)}(hhh](j)}(hhh]j )}(hhh]h)}(h**Quick Quiz**:h](j)}(h**Quick Quiz**h]h Quick Quiz}(hjUhhhNhNubah}(h]h ]h"]h$]h&]uh1jhjQubh:}(hjQhhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhjNubah}(h]h ]h"]h$]h&]uh1j hjKubah}(h]h ]h"]h$]h&]uh1jhjHubj)}(hhh]j )}(hhh]h)}(h*Can't the compiler also reorder this code?h]h,Can’t the compiler also reorder this code?}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj|ubah}(h]h ]h"]h$]h&]uh1j hjyubah}(h]h ]h"]h$]h&]uh1jhjHubj)}(hhh]j )}(hhh]h)}(h **Answer**:h](j)}(h **Answer**h]hAnswer}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1jhjubh:}(hjhhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhjubah}(h]h ]h"]h$]h&]uh1j hjubah}(h]h ]h"]h$]h&]uh1jhjHubj)}(hhh]j )}(hhh]h)}(htNo, the volatile casts in READ_ONCE() and WRITE_ONCE() prevent the compiler from reordering in this particular case.h]htNo, the volatile casts in READ_ONCE() and WRITE_ONCE() prevent the compiler from reordering in this particular case.}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhjubah}(h]h ]h"]h$]h&]uh1j hjubah}(h]h ]h"]h$]h&]uh1jhjHubeh}(h]h ]h"]h$]h&]uh1jhj;ubeh}(h]h ]h"]h$]h&]colsKuh1jhj8ubah}(h]h ]h"]h$]h&]uh1jhjhhhhhNubeh}(h]j ah ]h"]readers impose minimal orderingah$]h&]uh1hhjhhhhhMjKubh)}(hhh](h)}(hReaders Do Not Exclude Updatersh]hReaders Do Not Exclude Updaters}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhjhhhhhMubh)}(hNeither rcu_read_lock() nor rcu_read_unlock() exclude updates. All they do is to prevent grace periods from ending. The following example illustrates this:h]hNeither rcu_read_lock() nor rcu_read_unlock() exclude updates. All they do is to prevent grace periods from ending. The following example illustrates this:}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhjhhubj@)}(hX:: 1 void thread0(void) 2 { 3 rcu_read_lock(); 4 r1 = READ_ONCE(y); 5 if (r1) { 6 do_something_with_nonzero_x(); 7 r2 = READ_ONCE(x); 8 WARN_ON(!r2); /* BUG!!! */ 9 } 10 rcu_read_unlock(); 11 } 12 13 void thread1(void) 14 { 15 spin_lock(&my_lock); 16 WRITE_ONCE(x, 1); 17 WRITE_ONCE(y, 1); 18 spin_unlock(&my_lock); 19 } h]jF)}(hXd 1 void thread0(void) 2 { 3 rcu_read_lock(); 4 r1 = READ_ONCE(y); 5 if (r1) { 6 do_something_with_nonzero_x(); 7 r2 = READ_ONCE(x); 8 WARN_ON(!r2); /* BUG!!! */ 9 } 10 rcu_read_unlock(); 11 } 12 13 void thread1(void) 14 { 15 spin_lock(&my_lock); 16 WRITE_ONCE(x, 1); 17 WRITE_ONCE(y, 1); 18 spin_unlock(&my_lock); 19 }h]hXd 1 void thread0(void) 2 { 3 rcu_read_lock(); 4 r1 = READ_ONCE(y); 5 if (r1) { 6 do_something_with_nonzero_x(); 7 r2 = READ_ONCE(x); 8 WARN_ON(!r2); /* BUG!!! */ 9 } 10 rcu_read_unlock(); 11 } 12 13 void thread1(void) 14 { 15 spin_lock(&my_lock); 16 WRITE_ONCE(x, 1); 17 WRITE_ONCE(y, 1); 18 spin_unlock(&my_lock); 19 }}hj$sbah}(h]h ]h"]h$]h&]jUjVuh1jEhhhMhj ubah}(h]h ]h"]h$]h&]uh1j?hhhMhjhhubh)}(hX"If the thread0() function's rcu_read_lock() excluded the thread1() function's update, the WARN_ON() could never fire. But the fact is that rcu_read_lock() does not exclude much of anything aside from subsequent grace periods, of which thread1() has none, so the WARN_ON() can and does fire.h]hX&If the thread0() function’s rcu_read_lock() excluded the thread1() function’s update, the WARN_ON() could never fire. But the fact is that rcu_read_lock() does not exclude much of anything aside from subsequent grace periods, of which thread1() has none, so the WARN_ON() can and does fire.}(hj8hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM hjhhubeh}(h]j,ah ]h"]readers do not exclude updatersah$]h&]uh1hhjhhhhhMjKubh)}(hhh](h)}(h"Updaters Only Wait For Old Readersh]h"Updaters Only Wait For Old Readers}(hjPhhhNhNubah}(h]h ]h"]h$]h&]uh1hhjMhhhhhMubh)}(hXIt might be tempting to assume that after synchronize_rcu() completes, there are no readers executing. This temptation must be avoided because new readers can start immediately after synchronize_rcu() starts, and synchronize_rcu() is under no obligation to wait for these new readers.h]hXIt might be tempting to assume that after synchronize_rcu() completes, there are no readers executing. This temptation must be avoided because new readers can start immediately after synchronize_rcu() starts, and synchronize_rcu() is under no obligation to wait for these new readers.}(hj^hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhjMhhubj)}(hhh]j)}(hhh](j)}(hhh]h}(h]h ]h"]h$]h&]colwidthKGuh1jhjoubj)}(hhh](j)}(hhh]j )}(hhh]h)}(h**Quick Quiz**:h](j)}(h**Quick Quiz**h]h Quick Quiz}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1jhjubh:}(hjhhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhjubah}(h]h ]h"]h$]h&]uh1j hjubah}(h]h ]h"]h$]h&]uh1jhj|ubj)}(hhh]j )}(hhh]h)}(hSuppose that synchronize_rcu() did wait until *all* readers had completed instead of waiting only on pre-existing readers. For how long would the updater be able to rely on there being no readers?h](h.Suppose that synchronize_rcu() did wait until }(hjhhhNhNubj9 )}(h*all*h]hall}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1j8 hjubh readers had completed instead of waiting only on pre-existing readers. For how long would the updater be able to rely on there being no readers?}(hjhhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhjubah}(h]h ]h"]h$]h&]uh1j hjubah}(h]h ]h"]h$]h&]uh1jhj|ubj)}(hhh]j )}(hhh]h)}(h **Answer**:h](j)}(h **Answer**h]hAnswer}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1jhjubh:}(hjhhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhM!hjubah}(h]h ]h"]h$]h&]uh1j hjubah}(h]h ]h"]h$]h&]uh1jhj|ubj)}(hhh]j )}(hhh]h)}(hFor no time at all. Even if synchronize_rcu() were to wait until all readers had completed, a new reader might start immediately after synchronize_rcu() completed. Therefore, the code following synchronize_rcu() can *never* rely on there being no readers.h](hFor no time at all. Even if synchronize_rcu() were to wait until all readers had completed, a new reader might start immediately after synchronize_rcu() completed. Therefore, the code following synchronize_rcu() can }(hjhhhNhNubj9 )}(h*never*h]hnever}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1j8 hjubh rely on there being no readers.}(hjhhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhM#hjubah}(h]h ]h"]h$]h&]uh1j hj ubah}(h]h ]h"]h$]h&]uh1jhj|ubeh}(h]h ]h"]h$]h&]uh1jhjoubeh}(h]h ]h"]h$]h&]colsKuh1jhjlubah}(h]h ]h"]h$]h&]uh1jhjMhhhhhNubeh}(h]jOah ]h"]"updaters only wait for old readersah$]h&]uh1hhjhhhhhMjKubh)}(hhh](h)}(h9Grace Periods Don't Partition Read-Side Critical Sectionsh]h;Grace Periods Don’t Partition Read-Side Critical Sections}(hj\hhhNhNubah}(h]h ]h"]h$]h&]uh1hhjYhhhhhM*ubh)}(hXIt is tempting to assume that if any part of one RCU read-side critical section precedes a given grace period, and if any part of another RCU read-side critical section follows that same grace period, then all of the first RCU read-side critical section must precede all of the second. However, this just isn't the case: A single grace period does not partition the set of RCU read-side critical sections. An example of this situation can be illustrated as follows, where ``x``, ``y``, and ``z`` are initially all zero:h](hXIt is tempting to assume that if any part of one RCU read-side critical section precedes a given grace period, and if any part of another RCU read-side critical section follows that same grace period, then all of the first RCU read-side critical section must precede all of the second. However, this just isn’t the case: A single grace period does not partition the set of RCU read-side critical sections. An example of this situation can be illustrated as follows, where }(hjjhhhNhNubj)}(h``x``h]hx}(hjrhhhNhNubah}(h]h ]h"]h$]h&]uh1jhjjubh, }(hjjhhhNhNubj)}(h``y``h]hy}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1jhjjubh, and }(hjjhhhNhNubj)}(h``z``h]hz}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1jhjjubh are initially all zero:}(hjjhhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhM,hjYhhubj@)}(hX:: 1 void thread0(void) 2 { 3 rcu_read_lock(); 4 WRITE_ONCE(a, 1); 5 WRITE_ONCE(b, 1); 6 rcu_read_unlock(); 7 } 8 9 void thread1(void) 10 { 11 r1 = READ_ONCE(a); 12 synchronize_rcu(); 13 WRITE_ONCE(c, 1); 14 } 15 16 void thread2(void) 17 { 18 rcu_read_lock(); 19 r2 = READ_ONCE(b); 20 r3 = READ_ONCE(c); 21 rcu_read_unlock(); 22 } h]jF)}(hXf 1 void thread0(void) 2 { 3 rcu_read_lock(); 4 WRITE_ONCE(a, 1); 5 WRITE_ONCE(b, 1); 6 rcu_read_unlock(); 7 } 8 9 void thread1(void) 10 { 11 r1 = READ_ONCE(a); 12 synchronize_rcu(); 13 WRITE_ONCE(c, 1); 14 } 15 16 void thread2(void) 17 { 18 rcu_read_lock(); 19 r2 = READ_ONCE(b); 20 r3 = READ_ONCE(c); 21 rcu_read_unlock(); 22 }h]hXf 1 void thread0(void) 2 { 3 rcu_read_lock(); 4 WRITE_ONCE(a, 1); 5 WRITE_ONCE(b, 1); 6 rcu_read_unlock(); 7 } 8 9 void thread1(void) 10 { 11 r1 = READ_ONCE(a); 12 synchronize_rcu(); 13 WRITE_ONCE(c, 1); 14 } 15 16 void thread2(void) 17 { 18 rcu_read_lock(); 19 r2 = READ_ONCE(b); 20 r3 = READ_ONCE(c); 21 rcu_read_unlock(); 22 }}hjsbah}(h]h ]h"]h$]h&]jUjVuh1jEhhhM7hjubah}(h]h ]h"]h$]h&]uh1j?hhhM5hjYhhubh)}(hIt turns out that the outcome:h]hIt turns out that the outcome:}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMNhjYhhubj@)}(h':: (r1 == 1 && r2 == 0 && r3 == 1) h]jF)}(h(r1 == 1 && r2 == 0 && r3 == 1)h]h(r1 == 1 && r2 == 0 && r3 == 1)}hjsbah}(h]h ]h"]h$]h&]jUjVuh1jEhhhMRhjubah}(h]h ]h"]h$]h&]uh1j?hhhMPhjYhhubh)}(hXKis entirely possible. The following figure show how this can happen, with each circled ``QS`` indicating the point at which RCU recorded a *quiescent state* for each thread, that is, a state in which RCU knows that the thread cannot be in the midst of an RCU read-side critical section that started before the current grace period:h](hWis entirely possible. The following figure show how this can happen, with each circled }(hjhhhNhNubj)}(h``QS``h]hQS}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1jhjubh. indicating the point at which RCU recorded a }(hjhhhNhNubj9 )}(h*quiescent state*h]hquiescent state}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1j8 hjubh for each thread, that is, a state in which RCU knows that the thread cannot be in the midst of an RCU read-side critical section that started before the current grace period:}(hjhhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMThjYhhubkfigure kernel_figure)}(hhh]hfigure)}(hhh]himage)}(h+.. kernel-figure:: GPpartitionReaders1.svg h]h}(h]h ]h"]h$]h&]uri/RCU/Design/Requirements/GPpartitionReaders1.svg candidates}*j6suh1j)hj&hhhKubah}(h]h ]h"]h$]h&]uh1j$hj!ubah}(h]h ]h"]h$]h&]uh1jhjYhhhhhM[ubh)}(hIf it is necessary to partition RCU read-side critical sections in this manner, it is necessary to use two grace periods, where the first grace period is known to end before the second grace period starts:h]hIf it is necessary to partition RCU read-side critical sections in this manner, it is necessary to use two grace periods, where the first grace period is known to end before the second grace period starts:}(hjFhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM\hjYhhubj@)}(hX,:: 1 void thread0(void) 2 { 3 rcu_read_lock(); 4 WRITE_ONCE(a, 1); 5 WRITE_ONCE(b, 1); 6 rcu_read_unlock(); 7 } 8 9 void thread1(void) 10 { 11 r1 = READ_ONCE(a); 12 synchronize_rcu(); 13 WRITE_ONCE(c, 1); 14 } 15 16 void thread2(void) 17 { 18 r2 = READ_ONCE(c); 19 synchronize_rcu(); 20 WRITE_ONCE(d, 1); 21 } 22 23 void thread3(void) 24 { 25 rcu_read_lock(); 26 r3 = READ_ONCE(b); 27 r4 = READ_ONCE(d); 28 rcu_read_unlock(); 29 } h]jF)}(hX 1 void thread0(void) 2 { 3 rcu_read_lock(); 4 WRITE_ONCE(a, 1); 5 WRITE_ONCE(b, 1); 6 rcu_read_unlock(); 7 } 8 9 void thread1(void) 10 { 11 r1 = READ_ONCE(a); 12 synchronize_rcu(); 13 WRITE_ONCE(c, 1); 14 } 15 16 void thread2(void) 17 { 18 r2 = READ_ONCE(c); 19 synchronize_rcu(); 20 WRITE_ONCE(d, 1); 21 } 22 23 void thread3(void) 24 { 25 rcu_read_lock(); 26 r3 = READ_ONCE(b); 27 r4 = READ_ONCE(d); 28 rcu_read_unlock(); 29 }h]hX 1 void thread0(void) 2 { 3 rcu_read_lock(); 4 WRITE_ONCE(a, 1); 5 WRITE_ONCE(b, 1); 6 rcu_read_unlock(); 7 } 8 9 void thread1(void) 10 { 11 r1 = READ_ONCE(a); 12 synchronize_rcu(); 13 WRITE_ONCE(c, 1); 14 } 15 16 void thread2(void) 17 { 18 r2 = READ_ONCE(c); 19 synchronize_rcu(); 20 WRITE_ONCE(d, 1); 21 } 22 23 void thread3(void) 24 { 25 rcu_read_lock(); 26 r3 = READ_ONCE(b); 27 r4 = READ_ONCE(d); 28 rcu_read_unlock(); 29 }}hjXsbah}(h]h ]h"]h$]h&]jUjVuh1jEhhhMbhjTubah}(h]h ]h"]h$]h&]uh1j?hhhM`hjYhhubh)}(hXHere, if ``(r1 == 1)``, then thread0()'s write to ``b`` must happen before the end of thread1()'s grace period. If in addition ``(r4 == 1)``, then thread3()'s read from ``b`` must happen after the beginning of thread2()'s grace period. If it is also the case that ``(r2 == 1)``, then the end of thread1()'s grace period must precede the beginning of thread2()'s grace period. This mean that the two RCU read-side critical sections cannot overlap, guaranteeing that ``(r3 == 1)``. As a result, the outcome:h](h Here, if }(hjlhhhNhNubj)}(h ``(r1 == 1)``h]h (r1 == 1)}(hjthhhNhNubah}(h]h ]h"]h$]h&]uh1jhjlubh, then thread0()’s write to }(hjlhhhNhNubj)}(h``b``h]hb}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1jhjlubhJ must happen before the end of thread1()’s grace period. If in addition }(hjlhhhNhNubj)}(h ``(r4 == 1)``h]h (r4 == 1)}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1jhjlubh, then thread3()’s read from }(hjlhhhNhNubj)}(h``b``h]hb}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1jhjlubh\ must happen after the beginning of thread2()’s grace period. If it is also the case that }(hjlhhhNhNubj)}(h ``(r2 == 1)``h]h (r2 == 1)}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1jhjlubh, then the end of thread1()’s grace period must precede the beginning of thread2()’s grace period. This mean that the two RCU read-side critical sections cannot overlap, guaranteeing that }(hjlhhhNhNubj)}(h ``(r3 == 1)``h]h (r3 == 1)}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1jhjlubh. As a result, the outcome:}(hjlhhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhjYhhubj@)}(h2:: (r1 == 1 && r2 == 1 && r3 == 0 && r4 == 1) h]jF)}(h*(r1 == 1 && r2 == 1 && r3 == 0 && r4 == 1)h]h*(r1 == 1 && r2 == 1 && r3 == 0 && r4 == 1)}hjsbah}(h]h ]h"]h$]h&]jUjVuh1jEhhhMhjubah}(h]h ]h"]h$]h&]uh1j?hhhMhjYhhubh)}(hcannot happen.h]hcannot happen.}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhjYhhubh)}(hyThis non-requirement was also non-premeditated, but became apparent when studying RCU's interaction with memory ordering.h]h{This non-requirement was also non-premeditated, but became apparent when studying RCU’s interaction with memory ordering.}(hj hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhjYhhubeh}(h]jrah ]h"]9grace periods don't partition read-side critical sectionsah$]h&]uh1hhjhhhhhM*jKubh)}(hhh](h)}(h9Read-Side Critical Sections Don't Partition Grace Periodsh]h;Read-Side Critical Sections Don’t Partition Grace Periods}(hj$hhhNhNubah}(h]h ]h"]h$]h&]uh1hhj!hhhhhMubh)}(hXIt is also tempting to assume that if an RCU read-side critical section happens between a pair of grace periods, then those grace periods cannot overlap. However, this temptation leads nowhere good, as can be illustrated by the following, with all variables initially zero:h]hXIt is also tempting to assume that if an RCU read-side critical section happens between a pair of grace periods, then those grace periods cannot overlap. However, this temptation leads nowhere good, as can be illustrated by the following, with all variables initially zero:}(hj2hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj!hhubj@)}(hX:: 1 void thread0(void) 2 { 3 rcu_read_lock(); 4 WRITE_ONCE(a, 1); 5 WRITE_ONCE(b, 1); 6 rcu_read_unlock(); 7 } 8 9 void thread1(void) 10 { 11 r1 = READ_ONCE(a); 12 synchronize_rcu(); 13 WRITE_ONCE(c, 1); 14 } 15 16 void thread2(void) 17 { 18 rcu_read_lock(); 19 WRITE_ONCE(d, 1); 20 r2 = READ_ONCE(c); 21 rcu_read_unlock(); 22 } 23 24 void thread3(void) 25 { 26 r3 = READ_ONCE(d); 27 synchronize_rcu(); 28 WRITE_ONCE(e, 1); 29 } 30 31 void thread4(void) 32 { 33 rcu_read_lock(); 34 r4 = READ_ONCE(b); 35 r5 = READ_ONCE(e); 36 rcu_read_unlock(); 37 } h]jF)}(hXP 1 void thread0(void) 2 { 3 rcu_read_lock(); 4 WRITE_ONCE(a, 1); 5 WRITE_ONCE(b, 1); 6 rcu_read_unlock(); 7 } 8 9 void thread1(void) 10 { 11 r1 = READ_ONCE(a); 12 synchronize_rcu(); 13 WRITE_ONCE(c, 1); 14 } 15 16 void thread2(void) 17 { 18 rcu_read_lock(); 19 WRITE_ONCE(d, 1); 20 r2 = READ_ONCE(c); 21 rcu_read_unlock(); 22 } 23 24 void thread3(void) 25 { 26 r3 = READ_ONCE(d); 27 synchronize_rcu(); 28 WRITE_ONCE(e, 1); 29 } 30 31 void thread4(void) 32 { 33 rcu_read_lock(); 34 r4 = READ_ONCE(b); 35 r5 = READ_ONCE(e); 36 rcu_read_unlock(); 37 }h]hXP 1 void thread0(void) 2 { 3 rcu_read_lock(); 4 WRITE_ONCE(a, 1); 5 WRITE_ONCE(b, 1); 6 rcu_read_unlock(); 7 } 8 9 void thread1(void) 10 { 11 r1 = READ_ONCE(a); 12 synchronize_rcu(); 13 WRITE_ONCE(c, 1); 14 } 15 16 void thread2(void) 17 { 18 rcu_read_lock(); 19 WRITE_ONCE(d, 1); 20 r2 = READ_ONCE(c); 21 rcu_read_unlock(); 22 } 23 24 void thread3(void) 25 { 26 r3 = READ_ONCE(d); 27 synchronize_rcu(); 28 WRITE_ONCE(e, 1); 29 } 30 31 void thread4(void) 32 { 33 rcu_read_lock(); 34 r4 = READ_ONCE(b); 35 r5 = READ_ONCE(e); 36 rcu_read_unlock(); 37 }}hjDsbah}(h]h ]h"]h$]h&]jUjVuh1jEhhhMhj@ubah}(h]h ]h"]h$]h&]uh1j?hhhMhj!hhubh)}(hIn this case, the outcome:h]hIn this case, the outcome:}(hjXhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj!hhubj@)}(h=:: (r1 == 1 && r2 == 1 && r3 == 1 && r4 == 0 && r5 == 1) h]jF)}(h5(r1 == 1 && r2 == 1 && r3 == 1 && r4 == 0 && r5 == 1)h]h5(r1 == 1 && r2 == 1 && r3 == 1 && r4 == 0 && r5 == 1)}hjjsbah}(h]h ]h"]h$]h&]jUjVuh1jEhhhMhjfubah}(h]h ]h"]h$]h&]uh1j?hhhMhj!hhubh)}(h+is entirely possible, as illustrated below:h]h+is entirely possible, as illustrated below:}(hj~hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj!hhubj )}(hhh]j%)}(hhh]j*)}(h+.. kernel-figure:: ReadersPartitionGP1.svg h]h}(h]h ]h"]h$]h&]uri/RCU/Design/Requirements/ReadersPartitionGP1.svgj7}j9jsuh1j)hjhhhKubah}(h]h ]h"]h$]h&]uh1j$hjubah}(h]h ]h"]h$]h&]uh1jhj!hhhhhMubh)}(hAgain, an RCU read-side critical section can overlap almost all of a given grace period, just so long as it does not overlap the entire grace period. As a result, an RCU read-side critical section cannot partition a pair of RCU grace periods.h]hAgain, an RCU read-side critical section can overlap almost all of a given grace period, just so long as it does not overlap the entire grace period. As a result, an RCU read-side critical section cannot partition a pair of RCU grace periods.}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj!hhubj)}(hhh]j)}(hhh](j)}(hhh]h}(h]h ]h"]h$]h&]colwidthKGuh1jhjubj)}(hhh](j)}(hhh]j )}(hhh]h)}(h**Quick Quiz**:h](j)}(h**Quick Quiz**h]h Quick Quiz}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1jhjubh:}(hjhhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhjubah}(h]h ]h"]h$]h&]uh1j hjubah}(h]h ]h"]h$]h&]uh1jhjubj)}(hhh]j )}(hhh]h)}(hHow long a sequence of grace periods, each separated by an RCU read-side critical section, would be required to partition the RCU read-side critical sections at the beginning and end of the chain?h]hHow long a sequence of grace periods, each separated by an RCU read-side critical section, would be required to partition the RCU read-side critical sections at the beginning and end of the chain?}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhjubah}(h]h ]h"]h$]h&]uh1j hjubah}(h]h ]h"]h$]h&]uh1jhjubj)}(hhh]j )}(hhh]h)}(h **Answer**:h](j)}(h **Answer**h]hAnswer}(hj$hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj ubh:}(hj hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhjubah}(h]h ]h"]h$]h&]uh1j hjubah}(h]h ]h"]h$]h&]uh1jhjubj)}(hhh]j )}(hhh]h)}(hIn theory, an infinite number. In practice, an unknown number that is sensitive to both implementation details and timing considerations. Therefore, even in practice, RCU users must abide by the theoretical rather than the practical answer.h]hIn theory, an infinite number. In practice, an unknown number that is sensitive to both implementation details and timing considerations. Therefore, even in practice, RCU users must abide by the theoretical rather than the practical answer.}(hjNhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhjKubah}(h]h ]h"]h$]h&]uh1j hjHubah}(h]h ]h"]h$]h&]uh1jhjubeh}(h]h ]h"]h$]h&]uh1jhjubeh}(h]h ]h"]h$]h&]colsKuh1jhjubah}(h]h ]h"]h$]h&]uh1jhj!hhhhhNubeh}(h]jah ]h"]9read-side critical sections don't partition grace periodsah$]h&]uh1hhjhhhhhMjKubeh}(h]jah ]h"]fundamental non-requirementsah$]h&]uh1hhhhhhhhMjKubh)}(hhh](h)}(hParallelism Facts of Lifeh]hParallelism Facts of Life}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhjhhhhhMubh)}(hThese parallelism facts of life are by no means specific to RCU, but the RCU implementation must abide by them. They therefore bear repeating:h]hThese parallelism facts of life are by no means specific to RCU, but the RCU implementation must abide by them. They therefore bear repeating:}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhjhhubj)}(hhh](j)}(hXAny CPU or task may be delayed at any time, and any attempts to avoid these delays by disabling preemption, interrupts, or whatever are completely futile. This is most obvious in preemptible user-level environments and in virtualized environments (where a given guest OS's VCPUs can be preempted at any time by the underlying hypervisor), but can also happen in bare-metal environments due to ECC errors, NMIs, and other hardware events. Although a delay of more than about 20 seconds can result in splats, the RCU implementation is obligated to use algorithms that can tolerate extremely long delays, but where “extremely long” is not long enough to allow wrap-around when incrementing a 64-bit counter.h]h)}(hXAny CPU or task may be delayed at any time, and any attempts to avoid these delays by disabling preemption, interrupts, or whatever are completely futile. This is most obvious in preemptible user-level environments and in virtualized environments (where a given guest OS's VCPUs can be preempted at any time by the underlying hypervisor), but can also happen in bare-metal environments due to ECC errors, NMIs, and other hardware events. Although a delay of more than about 20 seconds can result in splats, the RCU implementation is obligated to use algorithms that can tolerate extremely long delays, but where “extremely long” is not long enough to allow wrap-around when incrementing a 64-bit counter.h]hXAny CPU or task may be delayed at any time, and any attempts to avoid these delays by disabling preemption, interrupts, or whatever are completely futile. This is most obvious in preemptible user-level environments and in virtualized environments (where a given guest OS’s VCPUs can be preempted at any time by the underlying hypervisor), but can also happen in bare-metal environments due to ECC errors, NMIs, and other hardware events. Although a delay of more than about 20 seconds can result in splats, the RCU implementation is obligated to use algorithms that can tolerate extremely long delays, but where “extremely long” is not long enough to allow wrap-around when incrementing a 64-bit counter.}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhjubah}(h]h ]h"]h$]h&]uh1jhjhhhhhNubj)}(hBoth the compiler and the CPU can reorder memory accesses. Where it matters, RCU must use compiler directives and memory-barrier instructions to preserve ordering.h]h)}(hBoth the compiler and the CPU can reorder memory accesses. Where it matters, RCU must use compiler directives and memory-barrier instructions to preserve ordering.h]hBoth the compiler and the CPU can reorder memory accesses. Where it matters, RCU must use compiler directives and memory-barrier instructions to preserve ordering.}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhjubah}(h]h ]h"]h$]h&]uh1jhjhhhhhNubj)}(hX\Conflicting writes to memory locations in any given cache line will result in expensive cache misses. Greater numbers of concurrent writes and more-frequent concurrent writes will result in more dramatic slowdowns. RCU is therefore obligated to use algorithms that have sufficient locality to avoid significant performance and scalability problems.h]h)}(hX\Conflicting writes to memory locations in any given cache line will result in expensive cache misses. Greater numbers of concurrent writes and more-frequent concurrent writes will result in more dramatic slowdowns. RCU is therefore obligated to use algorithms that have sufficient locality to avoid significant performance and scalability problems.h]hX\Conflicting writes to memory locations in any given cache line will result in expensive cache misses. Greater numbers of concurrent writes and more-frequent concurrent writes will result in more dramatic slowdowns. RCU is therefore obligated to use algorithms that have sufficient locality to avoid significant performance and scalability problems.}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhjubah}(h]h ]h"]h$]h&]uh1jhjhhhhhNubj)}(hAs a rough rule of thumb, only one CPU's worth of processing may be carried out under the protection of any given exclusive lock. RCU must therefore use scalable locking designs.h]h)}(hAs a rough rule of thumb, only one CPU's worth of processing may be carried out under the protection of any given exclusive lock. RCU must therefore use scalable locking designs.h]hAs a rough rule of thumb, only one CPU’s worth of processing may be carried out under the protection of any given exclusive lock. RCU must therefore use scalable locking designs.}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhjubah}(h]h ]h"]h$]h&]uh1jhjhhhhhNubj)}(hXCounters are finite, especially on 32-bit systems. RCU's use of counters must therefore tolerate counter wrap, or be designed such that counter wrap would take way more time than a single system is likely to run. An uptime of ten years is quite possible, a runtime of a century much less so. As an example of the latter, RCU's dyntick-idle nesting counter allows 54 bits for interrupt nesting level (this counter is 64 bits even on a 32-bit system). Overflowing this counter requires 2\ :sup:`54` half-interrupts on a given CPU without that CPU ever going idle. If a half-interrupt happened every microsecond, it would take 570 years of runtime to overflow this counter, which is currently believed to be an acceptably long time.h]h)}(hXCounters are finite, especially on 32-bit systems. RCU's use of counters must therefore tolerate counter wrap, or be designed such that counter wrap would take way more time than a single system is likely to run. An uptime of ten years is quite possible, a runtime of a century much less so. As an example of the latter, RCU's dyntick-idle nesting counter allows 54 bits for interrupt nesting level (this counter is 64 bits even on a 32-bit system). Overflowing this counter requires 2\ :sup:`54` half-interrupts on a given CPU without that CPU ever going idle. If a half-interrupt happened every microsecond, it would take 570 years of runtime to overflow this counter, which is currently believed to be an acceptably long time.h](hXCounters are finite, especially on 32-bit systems. RCU’s use of counters must therefore tolerate counter wrap, or be designed such that counter wrap would take way more time than a single system is likely to run. An uptime of ten years is quite possible, a runtime of a century much less so. As an example of the latter, RCU’s dyntick-idle nesting counter allows 54 bits for interrupt nesting level (this counter is 64 bits even on a 32-bit system). Overflowing this counter requires 2 }(hjhhhNhNubh superscript)}(h :sup:`54`h]h54}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1jhjubh half-interrupts on a given CPU without that CPU ever going idle. If a half-interrupt happened every microsecond, it would take 570 years of runtime to overflow this counter, which is currently believed to be an acceptably long time.}(hjhhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhj ubah}(h]h ]h"]h$]h&]uh1jhjhhhhhNubj)}(hLinux systems can have thousands of CPUs running a single Linux kernel in a single shared-memory environment. RCU must therefore pay close attention to high-end scalability. h]h)}(hLinux systems can have thousands of CPUs running a single Linux kernel in a single shared-memory environment. RCU must therefore pay close attention to high-end scalability.h]hLinux systems can have thousands of CPUs running a single Linux kernel in a single shared-memory environment. RCU must therefore pay close attention to high-end scalability.}(hj;hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj7ubah}(h]h ]h"]h$]h&]uh1jhjhhhhhNubeh}(h]h ]h"]h$]h&]jjjhjjuh1jhjhhhhhMubh)}(hXBThis last parallelism fact of life means that RCU must pay special attention to the preceding facts of life. The idea that Linux might scale to systems with thousands of CPUs would have been met with some skepticism in the 1990s, but these requirements would have otherwise have been unsurprising, even in the early 1990s.h]hXBThis last parallelism fact of life means that RCU must pay special attention to the preceding facts of life. The idea that Linux might scale to systems with thousands of CPUs would have been met with some skepticism in the 1990s, but these requirements would have otherwise have been unsurprising, even in the early 1990s.}(hjUhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM hjhhubeh}(h]j%ah ]h"]parallelism facts of lifeah$]h&]uh1hhhhhhhhMjKubh)}(hhh](h)}(h&Quality-of-Implementation Requirementsh]h&Quality-of-Implementation Requirements}(hjmhhhNhNubah}(h]h ]h"]h$]h&]uh1hhjjhhhhhMubh)}(hXDThese sections list quality-of-implementation requirements. Although an RCU implementation that ignores these requirements could still be used, it would likely be subject to limitations that would make it inappropriate for industrial-strength production use. Classes of quality-of-implementation requirements are as follows:h]hXDThese sections list quality-of-implementation requirements. Although an RCU implementation that ignores these requirements could still be used, it would likely be subject to limitations that would make it inappropriate for industrial-strength production use. Classes of quality-of-implementation requirements are as follows:}(hj{hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhjjhhubj)}(hhh](j)}(h`Specialization`_h]h)}(hjh]h)}(hjh]hSpecialization}(hjhhhNhNubah}(h]h ]h"]h$]h&]nameSpecializationjspecializationuh1hhjjKubah}(h]h ]h"]h$]h&]uh1hhhhMhjubah}(h]h ]h"]h$]h&]uh1jhjhhhhhNubj)}(h`Performance and Scalability`_h]h)}(hjh]h)}(hjh]hPerformance and Scalability}(hjhhhNhNubah}(h]h ]h"]h$]h&]namePerformance and Scalabilityjperformance-and-scalabilityuh1hhjjKubah}(h]h ]h"]h$]h&]uh1hhhhMhjubah}(h]h ]h"]h$]h&]uh1jhjhhhhhNubj)}(h`Forward Progress`_h]h)}(hjh]h)}(hjh]hForward Progress}(hjhhhNhNubah}(h]h ]h"]h$]h&]nameForward Progressjforward-progressuh1hhjjKubah}(h]h ]h"]h$]h&]uh1hhhhMhjubah}(h]h ]h"]h$]h&]uh1jhjhhhhhNubj)}(h`Composability`_h]h)}(hjh]h)}(hjh]h Composability}(hjhhhNhNubah}(h]h ]h"]h$]h&]name Composabilityj composabilityuh1hhjjKubah}(h]h ]h"]h$]h&]uh1hhhhMhjubah}(h]h ]h"]h$]h&]uh1jhjhhhhhNubj)}(h`Corner Cases`_ h]h)}(h`Corner Cases`_h]h)}(hjh]h Corner Cases}(hj hhhNhNubah}(h]h ]h"]h$]h&]name Corner Casesj corner-casesuh1hhjjKubah}(h]h ]h"]h$]h&]uh1hhhhMhjubah}(h]h ]h"]h$]h&]uh1jhjhhhhhNubeh}(h]h ]h"]h$]h&]jjjhjjuh1jhjjhhhhhMubh)}(h3These classes is covered in the following sections.h]h3These classes is covered in the following sections.}(hjBhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM!hjjhhubh)}(hhh](h)}(hSpecializationh]hSpecialization}(hjShhhNhNubah}(h]h ]h"]h$]h&]uh1hhjPhhhhhM$ubh)}(hRCU is and always has been intended primarily for read-mostly situations, which means that RCU's read-side primitives are optimized, often at the expense of its update-side primitives. Experience thus far is captured by the following list of situations:h]hRCU is and always has been intended primarily for read-mostly situations, which means that RCU’s read-side primitives are optimized, often at the expense of its update-side primitives. Experience thus far is captured by the following list of situations:}(hjahhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM&hjPhhubj)}(hhh](j)}(hVRead-mostly data, where stale and inconsistent data is not a problem: RCU works great!h]h)}(hVRead-mostly data, where stale and inconsistent data is not a problem: RCU works great!h]hVRead-mostly data, where stale and inconsistent data is not a problem: RCU works great!}(hjvhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM+hjrubah}(h]h ]h"]h$]h&]uh1jhjohhhhhNubj)}(h@Read-mostly data, where data must be consistent: RCU works well.h]h)}(hjh]h@Read-mostly data, where data must be consistent: RCU works well.}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM-hjubah}(h]h ]h"]h$]h&]uh1jhjohhhhhNubj)}(hLRead-write data, where data must be consistent: RCU *might* work OK. Or not.h]h)}(hLRead-write data, where data must be consistent: RCU *might* work OK. Or not.h](h4Read-write data, where data must be consistent: RCU }(hjhhhNhNubj9 )}(h*might*h]hmight}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1j8 hjubh work OK. Or not.}(hjhhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhM.hjubah}(h]h ]h"]h$]h&]uh1jhjohhhhhNubj)}(hX Write-mostly data, where data must be consistent: RCU is very unlikely to be the right tool for the job, with the following exceptions, where RCU can provide: a. Existence guarantees for update-friendly mechanisms. b. Wait-free read-side primitives for real-time use. h](h)}(hWrite-mostly data, where data must be consistent: RCU is very unlikely to be the right tool for the job, with the following exceptions, where RCU can provide:h]hWrite-mostly data, where data must be consistent: RCU is very unlikely to be the right tool for the job, with the following exceptions, where RCU can provide:}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM0hjubj)}(hhh](j)}(h4Existence guarantees for update-friendly mechanisms.h]h)}(hjh]h4Existence guarantees for update-friendly mechanisms.}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM4hjubah}(h]h ]h"]h$]h&]uh1jhjubj)}(h2Wait-free read-side primitives for real-time use. h]h)}(h1Wait-free read-side primitives for real-time use.h]h1Wait-free read-side primitives for real-time use.}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM5hjubah}(h]h ]h"]h$]h&]uh1jhjubeh}(h]h ]h"]h$]h&]j loweralphajhjjuh1jhjubeh}(h]h ]h"]h$]h&]uh1jhjohhhNhNubeh}(h]h ]h"]h$]h&]jjjhjjuh1jhjPhhhhhM+ubh)}(hXThis focus on read-mostly situations means that RCU must interoperate with other synchronization primitives. For example, the add_gp() and remove_gp_synchronous() examples discussed earlier use RCU to protect readers and locking to coordinate updaters. However, the need extends much farther, requiring that a variety of synchronization primitives be legal within RCU read-side critical sections, including spinlocks, sequence locks, atomic operations, reference counters, and memory barriers.h]hXThis focus on read-mostly situations means that RCU must interoperate with other synchronization primitives. For example, the add_gp() and remove_gp_synchronous() examples discussed earlier use RCU to protect readers and locking to coordinate updaters. However, the need extends much farther, requiring that a variety of synchronization primitives be legal within RCU read-side critical sections, including spinlocks, sequence locks, atomic operations, reference counters, and memory barriers.}(hj"hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM7hjPhhubj)}(hhh]j)}(hhh](j)}(hhh]h}(h]h ]h"]h$]h&]colwidthKGuh1jhj3ubj)}(hhh](j)}(hhh]j )}(hhh]h)}(h**Quick Quiz**:h](j)}(h**Quick Quiz**h]h Quick Quiz}(hjMhhhNhNubah}(h]h ]h"]h$]h&]uh1jhjIubh:}(hjIhhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMAhjFubah}(h]h ]h"]h$]h&]uh1j hjCubah}(h]h ]h"]h$]h&]uh1jhj@ubj)}(hhh]j )}(hhh]h)}(hWhat about sleeping locks?h]hWhat about sleeping locks?}(hjwhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMChjtubah}(h]h ]h"]h$]h&]uh1j hjqubah}(h]h ]h"]h$]h&]uh1jhj@ubj)}(hhh]j )}(hhh]h)}(h **Answer**:h](j)}(h **Answer**h]hAnswer}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1jhjubh:}(hjhhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMEhjubah}(h]h ]h"]h$]h&]uh1j hjubah}(h]h ]h"]h$]h&]uh1jhj@ubj)}(hhh]j )}(hhh]h)}(hXyThese are forbidden within Linux-kernel RCU read-side critical sections because it is not legal to place a quiescent state (in this case, voluntary context switch) within an RCU read-side critical section. However, sleeping locks may be used within userspace RCU read-side critical sections, and also within Linux-kernel sleepable RCU `(SRCU) `__ read-side critical sections. In addition, the -rt patchset turns spinlocks into a sleeping locks so that the corresponding critical sections can be preempted, which also means that these sleeplockified spinlocks (but not other sleeping locks!) may be acquire within -rt-Linux-kernel RCU read-side critical sections. Note that it *is* legal for a normal RCU read-side critical section to conditionally acquire a sleeping locks (as in mutex_trylock()), but only as long as it does not loop indefinitely attempting to conditionally acquire that sleeping locks. The key point is that things like mutex_trylock() either return with the mutex held, or return an error indication if the mutex was not immediately available. Either way, mutex_trylock() returns immediately without sleeping.h](hXOThese are forbidden within Linux-kernel RCU read-side critical sections because it is not legal to place a quiescent state (in this case, voluntary context switch) within an RCU read-side critical section. However, sleeping locks may be used within userspace RCU read-side critical sections, and also within Linux-kernel sleepable RCU }(hjhhhNhNubh)}(h`(SRCU) `__h]h(SRCU)}(hjhhhNhNubah}(h]h ]h"]h$]h&]name(SRCU)j sleepable-rcuuh1hhjjKubhXJ read-side critical sections. In addition, the -rt patchset turns spinlocks into a sleeping locks so that the corresponding critical sections can be preempted, which also means that these sleeplockified spinlocks (but not other sleeping locks!) may be acquire within -rt-Linux-kernel RCU read-side critical sections. Note that it }(hjhhhNhNubj9 )}(h*is*h]his}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1j8 hjubhX legal for a normal RCU read-side critical section to conditionally acquire a sleeping locks (as in mutex_trylock()), but only as long as it does not loop indefinitely attempting to conditionally acquire that sleeping locks. The key point is that things like mutex_trylock() either return with the mutex held, or return an error indication if the mutex was not immediately available. Either way, mutex_trylock() returns immediately without sleeping.}(hjhhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMGhjubah}(h]h ]h"]h$]h&]uh1j hjubah}(h]h ]h"]h$]h&]uh1jhj@ubeh}(h]h ]h"]h$]h&]uh1jhj3ubeh}(h]h ]h"]h$]h&]colsKuh1jhj0ubah}(h]h ]h"]h$]h&]uh1jhjPhhhhhNubh)}(hXIt often comes as a surprise that many algorithms do not require a consistent view of data, but many can function in that mode, with network routing being the poster child. Internet routing algorithms take significant time to propagate updates, so that by the time an update arrives at a given system, that system has been sending network traffic the wrong way for a considerable length of time. Having a few threads continue to send traffic the wrong way for a few more milliseconds is clearly not a problem: In the worst case, TCP retransmissions will eventually get the data where it needs to go. In general, when tracking the state of the universe outside of the computer, some level of inconsistency must be tolerated due to speed-of-light delays if nothing else.h]hXIt often comes as a surprise that many algorithms do not require a consistent view of data, but many can function in that mode, with network routing being the poster child. Internet routing algorithms take significant time to propagate updates, so that by the time an update arrives at a given system, that system has been sending network traffic the wrong way for a considerable length of time. Having a few threads continue to send traffic the wrong way for a few more milliseconds is clearly not a problem: In the worst case, TCP retransmissions will eventually get the data where it needs to go. In general, when tracking the state of the universe outside of the computer, some level of inconsistency must be tolerated due to speed-of-light delays if nothing else.}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM\hjPhhubh)}(hXaFurthermore, uncertainty about external state is inherent in many cases. For example, a pair of veterinarians might use heartbeat to determine whether or not a given cat was alive. But how long should they wait after the last heartbeat to decide that the cat is in fact dead? Waiting less than 400 milliseconds makes no sense because this would mean that a relaxed cat would be considered to cycle between death and life more than 100 times per minute. Moreover, just as with human beings, a cat's heart might stop for some period of time, so the exact wait period is a judgment call. One of our pair of veterinarians might wait 30 seconds before pronouncing the cat dead, while the other might insist on waiting a full minute. The two veterinarians would then disagree on the state of the cat during the final 30 seconds of the minute following the last heartbeat.h]hXcFurthermore, uncertainty about external state is inherent in many cases. For example, a pair of veterinarians might use heartbeat to determine whether or not a given cat was alive. But how long should they wait after the last heartbeat to decide that the cat is in fact dead? Waiting less than 400 milliseconds makes no sense because this would mean that a relaxed cat would be considered to cycle between death and life more than 100 times per minute. Moreover, just as with human beings, a cat’s heart might stop for some period of time, so the exact wait period is a judgment call. One of our pair of veterinarians might wait 30 seconds before pronouncing the cat dead, while the other might insist on waiting a full minute. The two veterinarians would then disagree on the state of the cat during the final 30 seconds of the minute following the last heartbeat.}(hj'hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMihjPhhubh)}(hXInterestingly enough, this same situation applies to hardware. When push comes to shove, how do we tell whether or not some external server has failed? We send messages to it periodically, and declare it failed if we don't receive a response within a given period of time. Policy decisions can usually tolerate short periods of inconsistency. The policy was decided some time ago, and is only now being put into effect, so a few milliseconds of delay is normally inconsequential.h]hXInterestingly enough, this same situation applies to hardware. When push comes to shove, how do we tell whether or not some external server has failed? We send messages to it periodically, and declare it failed if we don’t receive a response within a given period of time. Policy decisions can usually tolerate short periods of inconsistency. The policy was decided some time ago, and is only now being put into effect, so a few milliseconds of delay is normally inconsequential.}(hj5hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMwhjPhhubh)}(hX]However, there are algorithms that absolutely must see consistent data. For example, the translation between a user-level SystemV semaphore ID to the corresponding in-kernel data structure is protected by RCU, but it is absolutely forbidden to update a semaphore that has just been removed. In the Linux kernel, this need for consistency is accommodated by acquiring spinlocks located in the in-kernel data structure from within the RCU read-side critical section, and this is indicated by the green box in the figure above. Many other techniques may be used, and are in fact used within the Linux kernel.h]hX]However, there are algorithms that absolutely must see consistent data. For example, the translation between a user-level SystemV semaphore ID to the corresponding in-kernel data structure is protected by RCU, but it is absolutely forbidden to update a semaphore that has just been removed. In the Linux kernel, this need for consistency is accommodated by acquiring spinlocks located in the in-kernel data structure from within the RCU read-side critical section, and this is indicated by the green box in the figure above. Many other techniques may be used, and are in fact used within the Linux kernel.}(hjChhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhjPhhubh)}(hXVIn short, RCU is not required to maintain consistency, and other mechanisms may be used in concert with RCU when consistency is required. RCU's specialization allows it to do its job extremely well, and its ability to interoperate with other synchronization mechanisms allows the right mix of synchronization tools to be used for a given job.h]hXXIn short, RCU is not required to maintain consistency, and other mechanisms may be used in concert with RCU when consistency is required. RCU’s specialization allows it to do its job extremely well, and its ability to interoperate with other synchronization mechanisms allows the right mix of synchronization tools to be used for a given job.}(hjQhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhjPhhubeh}(h]jah ]h"]specializationah$]h&]uh1hhjjhhhhhM$jKubh)}(hhh](h)}(hPerformance and Scalabilityh]hPerformance and Scalability}(hjihhhNhNubah}(h]h ]h"]h$]h&]uh1hhjfhhhhhMubh)}(hXEnergy efficiency is a critical component of performance today, and Linux-kernel RCU implementations must therefore avoid unnecessarily awakening idle CPUs. I cannot claim that this requirement was premeditated. In fact, I learned of it during a telephone conversation in which I was given “frank and open” feedback on the importance of energy efficiency in battery-powered systems and on specific energy-efficiency shortcomings of the Linux-kernel RCU implementation. In my experience, the battery-powered embedded community will consider any unnecessary wakeups to be extremely unfriendly acts. So much so that mere Linux-kernel-mailing-list posts are insufficient to vent their ire.h]hXEnergy efficiency is a critical component of performance today, and Linux-kernel RCU implementations must therefore avoid unnecessarily awakening idle CPUs. I cannot claim that this requirement was premeditated. In fact, I learned of it during a telephone conversation in which I was given “frank and open” feedback on the importance of energy efficiency in battery-powered systems and on specific energy-efficiency shortcomings of the Linux-kernel RCU implementation. In my experience, the battery-powered embedded community will consider any unnecessary wakeups to be extremely unfriendly acts. So much so that mere Linux-kernel-mailing-list posts are insufficient to vent their ire.}(hjwhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhjfhhubh)}(hXMemory consumption is not particularly important for in most situations, and has become decreasingly so as memory sizes have expanded and memory costs have plummeted. However, as I learned from Matt Mackall's `bloatwatch `__ efforts, memory footprint is critically important on single-CPU systems with non-preemptible (``CONFIG_PREEMPTION=n``) kernels, and thus `tiny RCU `__ was born. Josh Triplett has since taken over the small-memory banner with his `Linux kernel tinification `__ project, which resulted in `SRCU `__ becoming optional for those kernels not needing it.h](hMemory consumption is not particularly important for in most situations, and has become decreasingly so as memory sizes have expanded and memory costs have plummeted. However, as I learned from Matt Mackall’s }(hjhhhNhNubh)}(h1`bloatwatch `__h]h bloatwatch}(hjhhhNhNubah}(h]h ]h"]h$]h&]name bloatwatchj http://elinux.org/Linux_Tiny-FAQuh1hhjubh_ efforts, memory footprint is critically important on single-CPU systems with non-preemptible (}(hjhhhNhNubj)}(h``CONFIG_PREEMPTION=n``h]hCONFIG_PREEMPTION=n}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1jhjubh) kernels, and thus }(hjhhhNhNubh)}(hR`tiny RCU `__h]htiny RCU}(hjhhhNhNubah}(h]h ]h"]h$]h&]nametiny RCUjChttps://lore.kernel.org/r/20090113221724.GA15307@linux.vnet.ibm.comuh1hhjubhO was born. Josh Triplett has since taken over the small-memory banner with his }(hjhhhNhNubh)}(h=`Linux kernel tinification `__h]hLinux kernel tinification}(hjhhhNhNubah}(h]h ]h"]h$]h&]nameLinux kernel tinificationjhttps://tiny.wiki.kernel.org/uh1hhjubh project, which resulted in }(hjhhhNhNubh)}(h`SRCU `__h]hSRCU}(hjhhhNhNubah}(h]h ]h"]h$]h&]nameSRCUjjuh1hhjjKubh4 becoming optional for those kernels not needing it.}(hjhhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhjfhhubh)}(hXrThe remaining performance requirements are, for the most part, unsurprising. For example, in keeping with RCU's read-side specialization, rcu_dereference() should have negligible overhead (for example, suppression of a few minor compiler optimizations). Similarly, in non-preemptible environments, rcu_read_lock() and rcu_read_unlock() should have exactly zero overhead.h]hXtThe remaining performance requirements are, for the most part, unsurprising. For example, in keeping with RCU’s read-side specialization, rcu_dereference() should have negligible overhead (for example, suppression of a few minor compiler optimizations). Similarly, in non-preemptible environments, rcu_read_lock() and rcu_read_unlock() should have exactly zero overhead.}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhjfhhubh)}(hXIn preemptible environments, in the case where the RCU read-side critical section was not preempted (as will be the case for the highest-priority real-time process), rcu_read_lock() and rcu_read_unlock() should have minimal overhead. In particular, they should not contain atomic read-modify-write operations, memory-barrier instructions, preemption disabling, interrupt disabling, or backwards branches. However, in the case where the RCU read-side critical section was preempted, rcu_read_unlock() may acquire spinlocks and disable interrupts. This is why it is better to nest an RCU read-side critical section within a preempt-disable region than vice versa, at least in cases where that critical section is short enough to avoid unduly degrading real-time latencies.h]hXIn preemptible environments, in the case where the RCU read-side critical section was not preempted (as will be the case for the highest-priority real-time process), rcu_read_lock() and rcu_read_unlock() should have minimal overhead. In particular, they should not contain atomic read-modify-write operations, memory-barrier instructions, preemption disabling, interrupt disabling, or backwards branches. However, in the case where the RCU read-side critical section was preempted, rcu_read_unlock() may acquire spinlocks and disable interrupts. This is why it is better to nest an RCU read-side critical section within a preempt-disable region than vice versa, at least in cases where that critical section is short enough to avoid unduly degrading real-time latencies.}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhjfhhubh)}(hX~The synchronize_rcu() grace-period-wait primitive is optimized for throughput. It may therefore incur several milliseconds of latency in addition to the duration of the longest RCU read-side critical section. On the other hand, multiple concurrent invocations of synchronize_rcu() are required to use batching optimizations so that they can be satisfied by a single underlying grace-period-wait operation. For example, in the Linux kernel, it is not unusual for a single grace-period-wait operation to serve more than `1,000 separate invocations `__ of synchronize_rcu(), thus amortizing the per-invocation overhead down to nearly zero. However, the grace-period optimization is also required to avoid measurable degradation of real-time scheduling and interrupt latencies.h](hXThe synchronize_rcu() grace-period-wait primitive is optimized for throughput. It may therefore incur several milliseconds of latency in addition to the duration of the longest RCU read-side critical section. On the other hand, multiple concurrent invocations of synchronize_rcu() are required to use batching optimizations so that they can be satisfied by a single underlying grace-period-wait operation. For example, in the Linux kernel, it is not unusual for a single grace-period-wait operation to serve more than }(hjhhhNhNubh)}(h`1,000 separate invocations `__h]h1,000 separate invocations}(hjhhhNhNubah}(h]h ]h"]h$]h&]name1,000 separate invocationsjwhttps://www.usenix.org/conference/2004-usenix-annual-technical-conference/making-rcu-safe-deep-sub-millisecond-responseuh1hhjubh of synchronize_rcu(), thus amortizing the per-invocation overhead down to nearly zero. However, the grace-period optimization is also required to avoid measurable degradation of real-time scheduling and interrupt latencies.}(hjhhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhjfhhubh)}(hXIn some cases, the multi-millisecond synchronize_rcu() latencies are unacceptable. In these cases, synchronize_rcu_expedited() may be used instead, reducing the grace-period latency down to a few tens of microseconds on small systems, at least in cases where the RCU read-side critical sections are short. There are currently no special latency requirements for synchronize_rcu_expedited() on large systems, but, consistent with the empirical nature of the RCU specification, that is subject to change. However, there most definitely are scalability requirements: A storm of synchronize_rcu_expedited() invocations on 4096 CPUs should at least make reasonable forward progress. In return for its shorter latencies, synchronize_rcu_expedited() is permitted to impose modest degradation of real-time latency on non-idle online CPUs. Here, “modest” means roughly the same latency degradation as a scheduling-clock interrupt.h]hXIn some cases, the multi-millisecond synchronize_rcu() latencies are unacceptable. In these cases, synchronize_rcu_expedited() may be used instead, reducing the grace-period latency down to a few tens of microseconds on small systems, at least in cases where the RCU read-side critical sections are short. There are currently no special latency requirements for synchronize_rcu_expedited() on large systems, but, consistent with the empirical nature of the RCU specification, that is subject to change. However, there most definitely are scalability requirements: A storm of synchronize_rcu_expedited() invocations on 4096 CPUs should at least make reasonable forward progress. In return for its shorter latencies, synchronize_rcu_expedited() is permitted to impose modest degradation of real-time latency on non-idle online CPUs. Here, “modest” means roughly the same latency degradation as a scheduling-clock interrupt.}(hj7hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhjfhhubh)}(hThere are a number of situations where even synchronize_rcu_expedited()'s reduced grace-period latency is unacceptable. In these situations, the asynchronous call_rcu() can be used in place of synchronize_rcu() as follows:h]hThere are a number of situations where even synchronize_rcu_expedited()’s reduced grace-period latency is unacceptable. In these situations, the asynchronous call_rcu() can be used in place of synchronize_rcu() as follows:}(hjEhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhjfhhubj@)}(hXk:: 1 struct foo { 2 int a; 3 int b; 4 struct rcu_head rh; 5 }; 6 7 static void remove_gp_cb(struct rcu_head *rhp) 8 { 9 struct foo *p = container_of(rhp, struct foo, rh); 10 11 kfree(p); 12 } 13 14 bool remove_gp_asynchronous(void) 15 { 16 struct foo *p; 17 18 spin_lock(&gp_lock); 19 p = rcu_access_pointer(gp); 20 if (!p) { 21 spin_unlock(&gp_lock); 22 return false; 23 } 24 rcu_assign_pointer(gp, NULL); 25 call_rcu(&p->rh, remove_gp_cb); 26 spin_unlock(&gp_lock); 27 return true; 28 } h]jF)}(hX 1 struct foo { 2 int a; 3 int b; 4 struct rcu_head rh; 5 }; 6 7 static void remove_gp_cb(struct rcu_head *rhp) 8 { 9 struct foo *p = container_of(rhp, struct foo, rh); 10 11 kfree(p); 12 } 13 14 bool remove_gp_asynchronous(void) 15 { 16 struct foo *p; 17 18 spin_lock(&gp_lock); 19 p = rcu_access_pointer(gp); 20 if (!p) { 21 spin_unlock(&gp_lock); 22 return false; 23 } 24 rcu_assign_pointer(gp, NULL); 25 call_rcu(&p->rh, remove_gp_cb); 26 spin_unlock(&gp_lock); 27 return true; 28 }h]hX 1 struct foo { 2 int a; 3 int b; 4 struct rcu_head rh; 5 }; 6 7 static void remove_gp_cb(struct rcu_head *rhp) 8 { 9 struct foo *p = container_of(rhp, struct foo, rh); 10 11 kfree(p); 12 } 13 14 bool remove_gp_asynchronous(void) 15 { 16 struct foo *p; 17 18 spin_lock(&gp_lock); 19 p = rcu_access_pointer(gp); 20 if (!p) { 21 spin_unlock(&gp_lock); 22 return false; 23 } 24 rcu_assign_pointer(gp, NULL); 25 call_rcu(&p->rh, remove_gp_cb); 26 spin_unlock(&gp_lock); 27 return true; 28 }}hjWsbah}(h]h ]h"]h$]h&]jUjVuh1jEhhhMhjSubah}(h]h ]h"]h$]h&]uh1j?hhhMhjfhhubh)}(hX[A definition of ``struct foo`` is finally needed, and appears on lines 1-5. The function remove_gp_cb() is passed to call_rcu() on line 25, and will be invoked after the end of a subsequent grace period. This gets the same effect as remove_gp_synchronous(), but without forcing the updater to wait for a grace period to elapse. The call_rcu() function may be used in a number of situations where neither synchronize_rcu() nor synchronize_rcu_expedited() would be legal, including within preempt-disable code, local_bh_disable() code, interrupt-disable code, and interrupt handlers. However, even call_rcu() is illegal within NMI handlers and from idle and offline CPUs. The callback function (remove_gp_cb() in this case) will be executed within softirq (software interrupt) environment within the Linux kernel, either within a real softirq handler or under the protection of local_bh_disable(). In both the Linux kernel and in userspace, it is bad practice to write an RCU callback function that takes too long. Long-running operations should be relegated to separate threads or (in the Linux kernel) workqueues.h](hA definition of }(hjkhhhNhNubj)}(h``struct foo``h]h struct foo}(hjshhhNhNubah}(h]h ]h"]h$]h&]uh1jhjkubhX= is finally needed, and appears on lines 1-5. The function remove_gp_cb() is passed to call_rcu() on line 25, and will be invoked after the end of a subsequent grace period. This gets the same effect as remove_gp_synchronous(), but without forcing the updater to wait for a grace period to elapse. The call_rcu() function may be used in a number of situations where neither synchronize_rcu() nor synchronize_rcu_expedited() would be legal, including within preempt-disable code, local_bh_disable() code, interrupt-disable code, and interrupt handlers. However, even call_rcu() is illegal within NMI handlers and from idle and offline CPUs. The callback function (remove_gp_cb() in this case) will be executed within softirq (software interrupt) environment within the Linux kernel, either within a real softirq handler or under the protection of local_bh_disable(). In both the Linux kernel and in userspace, it is bad practice to write an RCU callback function that takes too long. Long-running operations should be relegated to separate threads or (in the Linux kernel) workqueues.}(hjkhhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhjfhhubj)}(hhh]j)}(hhh](j)}(hhh]h}(h]h ]h"]h$]h&]colwidthKGuh1jhjubj)}(hhh](j)}(hhh]j )}(hhh]h)}(h**Quick Quiz**:h](j)}(h**Quick Quiz**h]h Quick Quiz}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1jhjubh:}(hjhhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhjubah}(h]h ]h"]h$]h&]uh1j hjubah}(h]h ]h"]h$]h&]uh1jhjubj)}(hhh]j )}(hhh]h)}(hWhy does line 19 use rcu_access_pointer()? After all, call_rcu() on line 25 stores into the structure, which would interact badly with concurrent insertions. Doesn't this mean that rcu_dereference() is required?h]hWhy does line 19 use rcu_access_pointer()? After all, call_rcu() on line 25 stores into the structure, which would interact badly with concurrent insertions. Doesn’t this mean that rcu_dereference() is required?}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhjubah}(h]h ]h"]h$]h&]uh1j hjubah}(h]h ]h"]h$]h&]uh1jhjubj)}(hhh]j )}(hhh]h)}(h **Answer**:h](j)}(h **Answer**h]hAnswer}(hjhhhNhNubah}(h]h ]h"]h$]h&]uh1jhjubh:}(hjhhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhjubah}(h]h ]h"]h$]h&]uh1j hjubah}(h]h ]h"]h$]h&]uh1jhjubj)}(hhh]j )}(hhh]h)}(hX#Presumably the ``->gp_lock`` acquired on line 18 excludes any changes, including any insertions that rcu_dereference() would protect against. Therefore, any insertions will be delayed until after ``->gp_lock`` is released on line 25, which in turn means that rcu_access_pointer() suffices.h](hPresumably the }(hj hhhNhNubj)}(h ``->gp_lock``h]h ->gp_lock}(hj( hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj ubh acquired on line 18 excludes any changes, including any insertions that rcu_dereference() would protect against. Therefore, any insertions will be delayed until after }(hj hhhNhNubj)}(h ``->gp_lock``h]h ->gp_lock}(hj: hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj ubhQ is released on line 25, which in turn means that rcu_access_pointer() suffices.}(hj hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhj ubah}(h]h ]h"]h$]h&]uh1j hj ubah}(h]h ]h"]h$]h&]uh1jhjubeh}(h]h ]h"]h$]h&]uh1jhjubeh}(h]h ]h"]h$]h&]colsKuh1jhjubah}(h]h ]h"]h$]h&]uh1jhjfhhhhhNubh)}(hHowever, all that remove_gp_cb() is doing is invoking kfree() on the data element. This is a common idiom, and is supported by kfree_rcu(), which allows “fire and forget” operation as shown below:h]hHowever, all that remove_gp_cb() is doing is invoking kfree() on the data element. This is a common idiom, and is supported by kfree_rcu(), which allows “fire and forget” operation as shown below:}(hjq hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM!hjfhhubj@)}(hX:: 1 struct foo { 2 int a; 3 int b; 4 struct rcu_head rh; 5 }; 6 7 bool remove_gp_faf(void) 8 { 9 struct foo *p; 10 11 spin_lock(&gp_lock); 12 p = rcu_dereference(gp); 13 if (!p) { 14 spin_unlock(&gp_lock); 15 return false; 16 } 17 rcu_assign_pointer(gp, NULL); 18 kfree_rcu(p, rh); 19 spin_unlock(&gp_lock); 20 return true; 21 } h]jF)}(hXo 1 struct foo { 2 int a; 3 int b; 4 struct rcu_head rh; 5 }; 6 7 bool remove_gp_faf(void) 8 { 9 struct foo *p; 10 11 spin_lock(&gp_lock); 12 p = rcu_dereference(gp); 13 if (!p) { 14 spin_unlock(&gp_lock); 15 return false; 16 } 17 rcu_assign_pointer(gp, NULL); 18 kfree_rcu(p, rh); 19 spin_unlock(&gp_lock); 20 return true; 21 }h]hXo 1 struct foo { 2 int a; 3 int b; 4 struct rcu_head rh; 5 }; 6 7 bool remove_gp_faf(void) 8 { 9 struct foo *p; 10 11 spin_lock(&gp_lock); 12 p = rcu_dereference(gp); 13 if (!p) { 14 spin_unlock(&gp_lock); 15 return false; 16 } 17 rcu_assign_pointer(gp, NULL); 18 kfree_rcu(p, rh); 19 spin_unlock(&gp_lock); 20 return true; 21 }}hj sbah}(h]h ]h"]h$]h&]jUjVuh1jEhhhM(hj ubah}(h]h ]h"]h$]h&]uh1j?hhhM&hjfhhubh)}(hXNote that remove_gp_faf() simply invokes kfree_rcu() and proceeds, without any need to pay any further attention to the subsequent grace period and kfree(). It is permissible to invoke kfree_rcu() from the same environments as for call_rcu(). Interestingly enough, DYNIX/ptx had the equivalents of call_rcu() and kfree_rcu(), but not synchronize_rcu(). This was due to the fact that RCU was not heavily used within DYNIX/ptx, so the very few places that needed something like synchronize_rcu() simply open-coded it.h]hXNote that remove_gp_faf() simply invokes kfree_rcu() and proceeds, without any need to pay any further attention to the subsequent grace period and kfree(). It is permissible to invoke kfree_rcu() from the same environments as for call_rcu(). Interestingly enough, DYNIX/ptx had the equivalents of call_rcu() and kfree_rcu(), but not synchronize_rcu(). This was due to the fact that RCU was not heavily used within DYNIX/ptx, so the very few places that needed something like synchronize_rcu() simply open-coded it.}(hj hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM>hjfhhubj)}(hhh]j)}(hhh](j)}(hhh]h}(h]h ]h"]h$]h&]colwidthKGuh1jhj ubj)}(hhh](j)}(hhh]j )}(hhh]h)}(h**Quick Quiz**:h](j)}(h**Quick Quiz**h]h Quick Quiz}(hj hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj ubh:}(hj hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMIhj ubah}(h]h ]h"]h$]h&]uh1j hj ubah}(h]h ]h"]h$]h&]uh1jhj ubj)}(hhh]j )}(hhh]h)}(hX Earlier it was claimed that call_rcu() and kfree_rcu() allowed updaters to avoid being blocked by readers. But how can that be correct, given that the invocation of the callback and the freeing of the memory (respectively) must still wait for a grace period to elapse?h]hX Earlier it was claimed that call_rcu() and kfree_rcu() allowed updaters to avoid being blocked by readers. But how can that be correct, given that the invocation of the callback and the freeing of the memory (respectively) must still wait for a grace period to elapse?}(hj hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMKhj ubah}(h]h ]h"]h$]h&]uh1j hj ubah}(h]h ]h"]h$]h&]uh1jhj ubj)}(hhh]j )}(hhh]h)}(h **Answer**:h](j)}(h **Answer**h]hAnswer}(hj!hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj !ubh:}(hj !hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMQhj !ubah}(h]h ]h"]h$]h&]uh1j hj!ubah}(h]h ]h"]h$]h&]uh1jhj ubj)}(hhh]j )}(hhh]h)}(hXWe could define things this way, but keep in mind that this sort of definition would say that updates in garbage-collected languages cannot complete until the next time the garbage collector runs, which does not seem at all reasonable. The key point is that in most cases, an updater using either call_rcu() or kfree_rcu() can proceed to the next update as soon as it has invoked call_rcu() or kfree_rcu(), without having to wait for a subsequent grace period.h]hXWe could define things this way, but keep in mind that this sort of definition would say that updates in garbage-collected languages cannot complete until the next time the garbage collector runs, which does not seem at all reasonable. The key point is that in most cases, an updater using either call_rcu() or kfree_rcu() can proceed to the next update as soon as it has invoked call_rcu() or kfree_rcu(), without having to wait for a subsequent grace period.}(hj:!hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMShj7!ubah}(h]h ]h"]h$]h&]uh1j hj4!ubah}(h]h ]h"]h$]h&]uh1jhj ubeh}(h]h ]h"]h$]h&]uh1jhj ubeh}(h]h ]h"]h$]h&]colsKuh1jhj ubah}(h]h ]h"]h$]h&]uh1jhjfhhhhhNubh)}(hX,But what if the updater must wait for the completion of code to be executed after the end of the grace period, but has other tasks that can be carried out in the meantime? The polling-style get_state_synchronize_rcu() and cond_synchronize_rcu() functions may be used for this purpose, as shown below:h]hX,But what if the updater must wait for the completion of code to be executed after the end of the grace period, but has other tasks that can be carried out in the meantime? The polling-style get_state_synchronize_rcu() and cond_synchronize_rcu() functions may be used for this purpose, as shown below:}(hjg!hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM]hjfhhubj@)}(hX:: 1 bool remove_gp_poll(void) 2 { 3 struct foo *p; 4 unsigned long s; 5 6 spin_lock(&gp_lock); 7 p = rcu_access_pointer(gp); 8 if (!p) { 9 spin_unlock(&gp_lock); 10 return false; 11 } 12 rcu_assign_pointer(gp, NULL); 13 spin_unlock(&gp_lock); 14 s = get_state_synchronize_rcu(); 15 do_something_while_waiting(); 16 cond_synchronize_rcu(s); 17 kfree(p); 18 return true; 19 } h]jF)}(hX 1 bool remove_gp_poll(void) 2 { 3 struct foo *p; 4 unsigned long s; 5 6 spin_lock(&gp_lock); 7 p = rcu_access_pointer(gp); 8 if (!p) { 9 spin_unlock(&gp_lock); 10 return false; 11 } 12 rcu_assign_pointer(gp, NULL); 13 spin_unlock(&gp_lock); 14 s = get_state_synchronize_rcu(); 15 do_something_while_waiting(); 16 cond_synchronize_rcu(s); 17 kfree(p); 18 return true; 19 }h]hX 1 bool remove_gp_poll(void) 2 { 3 struct foo *p; 4 unsigned long s; 5 6 spin_lock(&gp_lock); 7 p = rcu_access_pointer(gp); 8 if (!p) { 9 spin_unlock(&gp_lock); 10 return false; 11 } 12 rcu_assign_pointer(gp, NULL); 13 spin_unlock(&gp_lock); 14 s = get_state_synchronize_rcu(); 15 do_something_while_waiting(); 16 cond_synchronize_rcu(s); 17 kfree(p); 18 return true; 19 }}hjy!sbah}(h]h ]h"]h$]h&]jUjVuh1jEhhhMehju!ubah}(h]h ]h"]h$]h&]uh1j?hhhMchjfhhubh)}(hXOn line 14, get_state_synchronize_rcu() obtains a “cookie” from RCU, then line 15 carries out other tasks, and finally, line 16 returns immediately if a grace period has elapsed in the meantime, but otherwise waits as required. The need for ``get_state_synchronize_rcu`` and cond_synchronize_rcu() has appeared quite recently, so it is too early to tell whether they will stand the test of time.h](hOn line 14, get_state_synchronize_rcu() obtains a “cookie” from RCU, then line 15 carries out other tasks, and finally, line 16 returns immediately if a grace period has elapsed in the meantime, but otherwise waits as required. The need for }(hj!hhhNhNubj)}(h``get_state_synchronize_rcu``h]hget_state_synchronize_rcu}(hj!hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj!ubh} and cond_synchronize_rcu() has appeared quite recently, so it is too early to tell whether they will stand the test of time.}(hj!hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMyhjfhhubh)}(hRCU thus provides a range of tools to allow updaters to strike the required tradeoff between latency, flexibility and CPU overhead.h]hRCU thus provides a range of tools to allow updaters to strike the required tradeoff between latency, flexibility and CPU overhead.}(hj!hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhjfhhubeh}(h]jah ]h"]performance and scalabilityah$]h&]uh1hhjjhhhhhMjKubh)}(hhh](h)}(hForward Progressh]hForward Progress}(hj!hhhNhNubah}(h]h ]h"]h$]h&]uh1hhj!hhhhhMubh)}(hXLIn theory, delaying grace-period completion and callback invocation is harmless. In practice, not only are memory sizes finite but also callbacks sometimes do wakeups, and sufficiently deferred wakeups can be difficult to distinguish from system hangs. Therefore, RCU must provide a number of mechanisms to promote forward progress.h]hXLIn theory, delaying grace-period completion and callback invocation is harmless. In practice, not only are memory sizes finite but also callbacks sometimes do wakeups, and sufficiently deferred wakeups can be difficult to distinguish from system hangs. Therefore, RCU must provide a number of mechanisms to promote forward progress.}(hj!hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj!hhubh)}(hX|These mechanisms are not foolproof, nor can they be. For one simple example, an infinite loop in an RCU read-side critical section must by definition prevent later grace periods from ever completing. For a more involved example, consider a 64-CPU system built with ``CONFIG_RCU_NOCB_CPU=y`` and booted with ``rcu_nocbs=1-63``, where CPUs 1 through 63 spin in tight loops that invoke call_rcu(). Even if these tight loops also contain calls to cond_resched() (thus allowing grace periods to complete), CPU 0 simply will not be able to invoke callbacks as fast as the other 63 CPUs can register them, at least not until the system runs out of memory. In both of these examples, the Spiderman principle applies: With great power comes great responsibility. However, short of this level of abuse, RCU is required to ensure timely completion of grace periods and timely invocation of callbacks.h](hX These mechanisms are not foolproof, nor can they be. For one simple example, an infinite loop in an RCU read-side critical section must by definition prevent later grace periods from ever completing. For a more involved example, consider a 64-CPU system built with }(hj!hhhNhNubj)}(h``CONFIG_RCU_NOCB_CPU=y``h]hCONFIG_RCU_NOCB_CPU=y}(hj!hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj!ubh and booted with }(hj!hhhNhNubj)}(h``rcu_nocbs=1-63``h]hrcu_nocbs=1-63}(hj!hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj!ubhX7, where CPUs 1 through 63 spin in tight loops that invoke call_rcu(). Even if these tight loops also contain calls to cond_resched() (thus allowing grace periods to complete), CPU 0 simply will not be able to invoke callbacks as fast as the other 63 CPUs can register them, at least not until the system runs out of memory. In both of these examples, the Spiderman principle applies: With great power comes great responsibility. However, short of this level of abuse, RCU is required to ensure timely completion of grace periods and timely invocation of callbacks.}(hj!hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhj!hhubh)}(hNRCU takes the following steps to encourage timely completion of grace periods:h]hNRCU takes the following steps to encourage timely completion of grace periods:}(hj"hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj!hhubj)}(hhh](j)}(hX-If a grace period fails to complete within 100 milliseconds, RCU causes future invocations of cond_resched() on the holdout CPUs to provide an RCU quiescent state. RCU also causes those CPUs' need_resched() invocations to return ``true``, but only after the corresponding CPU's next scheduling-clock.h]h)}(hX-If a grace period fails to complete within 100 milliseconds, RCU causes future invocations of cond_resched() on the holdout CPUs to provide an RCU quiescent state. RCU also causes those CPUs' need_resched() invocations to return ``true``, but only after the corresponding CPU's next scheduling-clock.h](hIf a grace period fails to complete within 100 milliseconds, RCU causes future invocations of cond_resched() on the holdout CPUs to provide an RCU quiescent state. RCU also causes those CPUs’ need_resched() invocations to return }(hj("hhhNhNubj)}(h``true``h]htrue}(hj0"hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj("ubhA, but only after the corresponding CPU’s next scheduling-clock.}(hj("hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhj$"ubah}(h]h ]h"]h$]h&]uh1jhj!"hhhhhNubj)}(hX CPUs mentioned in the ``nohz_full`` kernel boot parameter can run indefinitely in the kernel without scheduling-clock interrupts, which defeats the above need_resched() strategem. RCU will therefore invoke resched_cpu() on any ``nohz_full`` CPUs still holding out after 109 milliseconds.h]h)}(hX CPUs mentioned in the ``nohz_full`` kernel boot parameter can run indefinitely in the kernel without scheduling-clock interrupts, which defeats the above need_resched() strategem. RCU will therefore invoke resched_cpu() on any ``nohz_full`` CPUs still holding out after 109 milliseconds.h](hCPUs mentioned in the }(hjR"hhhNhNubj)}(h ``nohz_full``h]h nohz_full}(hjZ"hhhNhNubah}(h]h ]h"]h$]h&]uh1jhjR"ubh kernel boot parameter can run indefinitely in the kernel without scheduling-clock interrupts, which defeats the above need_resched() strategem. RCU will therefore invoke resched_cpu() on any }(hjR"hhhNhNubj)}(h ``nohz_full``h]h nohz_full}(hjl"hhhNhNubah}(h]h ]h"]h$]h&]uh1jhjR"ubh0 CPUs still holding out after 109 milliseconds.}(hjR"hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhjN"ubah}(h]h ]h"]h$]h&]uh1jhj!"hhhhhNubj)}(hIn kernels built with ``CONFIG_RCU_BOOST=y``, if a given task that has been preempted within an RCU read-side critical section is holding out for more than 500 milliseconds, RCU will resort to priority boosting.h]h)}(hIn kernels built with ``CONFIG_RCU_BOOST=y``, if a given task that has been preempted within an RCU read-side critical section is holding out for more than 500 milliseconds, RCU will resort to priority boosting.h](hIn kernels built with }(hj"hhhNhNubj)}(h``CONFIG_RCU_BOOST=y``h]hCONFIG_RCU_BOOST=y}(hj"hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj"ubh, if a given task that has been preempted within an RCU read-side critical section is holding out for more than 500 milliseconds, RCU will resort to priority boosting.}(hj"hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhj"ubah}(h]h ]h"]h$]h&]uh1jhj!"hhhhhNubj)}(hIf a CPU is still holding out 10 seconds into the grace period, RCU will invoke resched_cpu() on it regardless of its ``nohz_full`` state. h]h)}(hIf a CPU is still holding out 10 seconds into the grace period, RCU will invoke resched_cpu() on it regardless of its ``nohz_full`` state.h](hwIf a CPU is still holding out 10 seconds into the grace period, RCU will invoke resched_cpu() on it regardless of its }(hj"hhhNhNubj)}(h ``nohz_full``h]h nohz_full}(hj"hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj"ubh state.}(hj"hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhj"ubah}(h]h ]h"]h$]h&]uh1jhj!"hhhhhNubeh}(h]h ]h"]h$]h&]jjjhjjuh1jhj!hhhhhMubh)}(hXThe above values are defaults for systems running with ``HZ=1000``. They will vary as the value of ``HZ`` varies, and can also be changed using the relevant Kconfig options and kernel boot parameters. RCU currently does not do much sanity checking of these parameters, so please use caution when changing them. Note that these forward-progress measures are provided only for RCU, not for `SRCU `__ or `Tasks RCU`_.h](h7The above values are defaults for systems running with }(hj"hhhNhNubj)}(h ``HZ=1000``h]hHZ=1000}(hj"hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj"ubh!. They will vary as the value of }(hj"hhhNhNubj)}(h``HZ``h]hHZ}(hj"hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj"ubhX varies, and can also be changed using the relevant Kconfig options and kernel boot parameters. RCU currently does not do much sanity checking of these parameters, so please use caution when changing them. Note that these forward-progress measures are provided only for RCU, not for }(hj"hhhNhNubh)}(h`SRCU `__h]hSRCU}(hj#hhhNhNubah}(h]h ]h"]h$]h&]nameSRCUjjuh1hhj"jKubh or }(hj"hhhNhNubh)}(h `Tasks RCU`_h]h Tasks RCU}(hj$#hhhNhNubah}(h]h ]h"]h$]h&]name Tasks RCUj tasks-rcuuh1hhj"jKubh.}(hj"hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhj!hhubh)}(hRCU takes the following steps in call_rcu() to encourage timely invocation of callbacks when any given non-\ ``rcu_nocbs`` CPU has 10,000 callbacks, or has 10,000 more callbacks than it had the last time encouragement was provided:h](hmRCU takes the following steps in call_rcu() to encourage timely invocation of callbacks when any given non- }(hj?#hhhNhNubj)}(h ``rcu_nocbs``h]h rcu_nocbs}(hjG#hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj?#ubhm CPU has 10,000 callbacks, or has 10,000 more callbacks than it had the last time encouragement was provided:}(hj?#hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhj!hhubj)}(hhh](j)}(h9Starts a grace period, if one is not already in progress.h]h)}(hjd#h]h9Starts a grace period, if one is not already in progress.}(hjf#hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhjb#ubah}(h]h ]h"]h$]h&]uh1jhj_#hhhhhNubj)}(hForces immediate checking for quiescent states, rather than waiting for three milliseconds to have elapsed since the beginning of the grace period.h]h)}(hForces immediate checking for quiescent states, rather than waiting for three milliseconds to have elapsed since the beginning of the grace period.h]hForces immediate checking for quiescent states, rather than waiting for three milliseconds to have elapsed since the beginning of the grace period.}(hj}#hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhjy#ubah}(h]h ]h"]h$]h&]uh1jhj_#hhhhhNubj)}(hImmediately tags the CPU's callbacks with their grace period completion numbers, rather than waiting for the ``RCU_SOFTIRQ`` handler to get around to it.h]h)}(hImmediately tags the CPU's callbacks with their grace period completion numbers, rather than waiting for the ``RCU_SOFTIRQ`` handler to get around to it.h](hoImmediately tags the CPU’s callbacks with their grace period completion numbers, rather than waiting for the }(hj#hhhNhNubj)}(h``RCU_SOFTIRQ``h]h RCU_SOFTIRQ}(hj#hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj#ubh handler to get around to it.}(hj#hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhj#ubah}(h]h ]h"]h$]h&]uh1jhj_#hhhhhNubj)}(hzLifts callback-execution batch limits, which speeds up callback invocation at the expense of degrading realtime response. h]h)}(hyLifts callback-execution batch limits, which speeds up callback invocation at the expense of degrading realtime response.h]hyLifts callback-execution batch limits, which speeds up callback invocation at the expense of degrading realtime response.}(hj#hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj#ubah}(h]h ]h"]h$]h&]uh1jhj_#hhhhhNubeh}(h]h ]h"]h$]h&]jjjhjjuh1jhj!hhhhhMubh)}(hX3Again, these are default values when running at ``HZ=1000``, and can be overridden. Again, these forward-progress measures are provided only for RCU, not for `SRCU `__ or `Tasks RCU`_. Even for RCU, callback-invocation forward progress for ``rcu_nocbs`` CPUs is much less well-developed, in part because workloads benefiting from ``rcu_nocbs`` CPUs tend to invoke call_rcu() relatively infrequently. If workloads emerge that need both ``rcu_nocbs`` CPUs and high call_rcu() invocation rates, then additional forward-progress work will be required.h](h0Again, these are default values when running at }(hj#hhhNhNubj)}(h ``HZ=1000``h]hHZ=1000}(hj#hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj#ubhc, and can be overridden. Again, these forward-progress measures are provided only for RCU, not for }(hj#hhhNhNubh)}(h`SRCU `__h]hSRCU}(hj#hhhNhNubah}(h]h ]h"]h$]h&]nameSRCUjjuh1hhj#jKubh or }(hj#hhhNhNubh)}(h `Tasks RCU`_h]h Tasks RCU}(hj$hhhNhNubah}(h]h ]h"]h$]h&]name Tasks RCUjj4#uh1hhj#jKubh9. Even for RCU, callback-invocation forward progress for }(hj#hhhNhNubj)}(h ``rcu_nocbs``h]h rcu_nocbs}(hj$hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj#ubhM CPUs is much less well-developed, in part because workloads benefiting from }(hj#hhhNhNubj)}(h ``rcu_nocbs``h]h rcu_nocbs}(hj-$hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj#ubh\ CPUs tend to invoke call_rcu() relatively infrequently. If workloads emerge that need both }(hj#hhhNhNubj)}(h ``rcu_nocbs``h]h rcu_nocbs}(hj?$hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj#ubhc CPUs and high call_rcu() invocation rates, then additional forward-progress work will be required.}(hj#hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhj!hhubeh}(h]jah ]h"]forward progressah$]h&]uh1hhjjhhhhhMjKubh)}(hhh](h)}(h Composabilityh]h Composability}(hja$hhhNhNubah}(h]h ]h"]h$]h&]uh1hhj^$hhhhhMubh)}(hXComposability has received much attention in recent years, perhaps in part due to the collision of multicore hardware with object-oriented techniques designed in single-threaded environments for single-threaded use. And in theory, RCU read-side critical sections may be composed, and in fact may be nested arbitrarily deeply. In practice, as with all real-world implementations of composable constructs, there are limitations.h]hXComposability has received much attention in recent years, perhaps in part due to the collision of multicore hardware with object-oriented techniques designed in single-threaded environments for single-threaded use. And in theory, RCU read-side critical sections may be composed, and in fact may be nested arbitrarily deeply. In practice, as with all real-world implementations of composable constructs, there are limitations.}(hjo$hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj^$hhubh)}(hXlImplementations of RCU for which rcu_read_lock() and rcu_read_unlock() generate no code, such as Linux-kernel RCU when ``CONFIG_PREEMPTION=n``, can be nested arbitrarily deeply. After all, there is no overhead. Except that if all these instances of rcu_read_lock() and rcu_read_unlock() are visible to the compiler, compilation will eventually fail due to exhausting memory, mass storage, or user patience, whichever comes first. If the nesting is not visible to the compiler, as is the case with mutually recursive functions each in its own translation unit, stack overflow will result. If the nesting takes the form of loops, perhaps in the guise of tail recursion, either the control variable will overflow or (in the Linux kernel) you will get an RCU CPU stall warning. Nevertheless, this class of RCU implementations is one of the most composable constructs in existence.h](hwImplementations of RCU for which rcu_read_lock() and rcu_read_unlock() generate no code, such as Linux-kernel RCU when }(hj}$hhhNhNubj)}(h``CONFIG_PREEMPTION=n``h]hCONFIG_PREEMPTION=n}(hj$hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj}$ubhX, can be nested arbitrarily deeply. After all, there is no overhead. Except that if all these instances of rcu_read_lock() and rcu_read_unlock() are visible to the compiler, compilation will eventually fail due to exhausting memory, mass storage, or user patience, whichever comes first. If the nesting is not visible to the compiler, as is the case with mutually recursive functions each in its own translation unit, stack overflow will result. If the nesting takes the form of loops, perhaps in the guise of tail recursion, either the control variable will overflow or (in the Linux kernel) you will get an RCU CPU stall warning. Nevertheless, this class of RCU implementations is one of the most composable constructs in existence.}(hj}$hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhj^$hhubh)}(hXRCU implementations that explicitly track nesting depth are limited by the nesting-depth counter. For example, the Linux kernel's preemptible RCU limits nesting to ``INT_MAX``. This should suffice for almost all practical purposes. That said, a consecutive pair of RCU read-side critical sections between which there is an operation that waits for a grace period cannot be enclosed in another RCU read-side critical section. This is because it is not legal to wait for a grace period within an RCU read-side critical section: To do so would result either in deadlock or in RCU implicitly splitting the enclosing RCU read-side critical section, neither of which is conducive to a long-lived and prosperous kernel.h](hRCU implementations that explicitly track nesting depth are limited by the nesting-depth counter. For example, the Linux kernel’s preemptible RCU limits nesting to }(hj$hhhNhNubj)}(h ``INT_MAX``h]hINT_MAX}(hj$hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj$ubhX. This should suffice for almost all practical purposes. That said, a consecutive pair of RCU read-side critical sections between which there is an operation that waits for a grace period cannot be enclosed in another RCU read-side critical section. This is because it is not legal to wait for a grace period within an RCU read-side critical section: To do so would result either in deadlock or in RCU implicitly splitting the enclosing RCU read-side critical section, neither of which is conducive to a long-lived and prosperous kernel.}(hj$hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhj^$hhubh)}(hXnIt is worth noting that RCU is not alone in limiting composability. For example, many transactional-memory implementations prohibit composing a pair of transactions separated by an irrevocable operation (for example, a network receive operation). For another example, lock-based critical sections can be composed surprisingly freely, but only if deadlock is avoided.h]hXnIt is worth noting that RCU is not alone in limiting composability. For example, many transactional-memory implementations prohibit composing a pair of transactions separated by an irrevocable operation (for example, a network receive operation). For another example, lock-based critical sections can be composed surprisingly freely, but only if deadlock is avoided.}(hj$hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj^$hhubh)}(hIn short, although RCU read-side critical sections are highly composable, care is required in some situations, just as is the case for any other composable synchronization mechanism.h]hIn short, although RCU read-side critical sections are highly composable, care is required in some situations, just as is the case for any other composable synchronization mechanism.}(hj$hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj^$hhubeh}(h]j ah ]h"] composabilityah$]h&]uh1hhjjhhhhhMjKubh)}(hhh](h)}(h Corner Casesh]h Corner Cases}(hj$hhhNhNubah}(h]h ]h"]h$]h&]uh1hhj$hhhhhMubh)}(hXA given RCU workload might have an endless and intense stream of RCU read-side critical sections, perhaps even so intense that there was never a point in time during which there was not at least one RCU read-side critical section in flight. RCU cannot allow this situation to block grace periods: As long as all the RCU read-side critical sections are finite, grace periods must also be finite.h]hXA given RCU workload might have an endless and intense stream of RCU read-side critical sections, perhaps even so intense that there was never a point in time during which there was not at least one RCU read-side critical section in flight. RCU cannot allow this situation to block grace periods: As long as all the RCU read-side critical sections are finite, grace periods must also be finite.}(hj$hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj$hhubh)}(hXThat said, preemptible RCU implementations could potentially result in RCU read-side critical sections being preempted for long durations, which has the effect of creating a long-duration RCU read-side critical section. This situation can arise only in heavily loaded systems, but systems using real-time priorities are of course more vulnerable. Therefore, RCU priority boosting is provided to help deal with this case. That said, the exact requirements on RCU priority boosting will likely evolve as more experience accumulates.h]hXThat said, preemptible RCU implementations could potentially result in RCU read-side critical sections being preempted for long durations, which has the effect of creating a long-duration RCU read-side critical section. This situation can arise only in heavily loaded systems, but systems using real-time priorities are of course more vulnerable. Therefore, RCU priority boosting is provided to help deal with this case. That said, the exact requirements on RCU priority boosting will likely evolve as more experience accumulates.}(hj$hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM hj$hhubh)}(hXnOther workloads might have very high update rates. Although one can argue that such workloads should instead use something other than RCU, the fact remains that RCU must handle such workloads gracefully. This requirement is another factor driving batching of grace periods, but it is also the driving force behind the checks for large numbers of queued RCU callbacks in the call_rcu() code path. Finally, high update rates should not delay RCU read-side critical sections, although some small read-side delays can occur when using synchronize_rcu_expedited(), courtesy of this function's use of smp_call_function_single().h]hXpOther workloads might have very high update rates. Although one can argue that such workloads should instead use something other than RCU, the fact remains that RCU must handle such workloads gracefully. This requirement is another factor driving batching of grace periods, but it is also the driving force behind the checks for large numbers of queued RCU callbacks in the call_rcu() code path. Finally, high update rates should not delay RCU read-side critical sections, although some small read-side delays can occur when using synchronize_rcu_expedited(), courtesy of this function’s use of smp_call_function_single().}(hj %hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj$hhubh)}(hXAlthough all three of these corner cases were understood in the early 1990s, a simple user-level test consisting of ``close(open(path))`` in a tight loop in the early 2000s suddenly provided a much deeper appreciation of the high-update-rate corner case. This test also motivated addition of some RCU code to react to high update rates, for example, if a given CPU finds itself with more than 10,000 RCU callbacks queued, it will cause RCU to take evasive action by more aggressively starting grace periods and more aggressively forcing completion of grace-period processing. This evasive action causes the grace period to complete more quickly, but at the cost of restricting RCU's batching optimizations, thus increasing the CPU overhead incurred by that grace period.h](htAlthough all three of these corner cases were understood in the early 1990s, a simple user-level test consisting of }(hj%hhhNhNubj)}(h``close(open(path))``h]hclose(open(path))}(hj#%hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj%ubhX{ in a tight loop in the early 2000s suddenly provided a much deeper appreciation of the high-update-rate corner case. This test also motivated addition of some RCU code to react to high update rates, for example, if a given CPU finds itself with more than 10,000 RCU callbacks queued, it will cause RCU to take evasive action by more aggressively starting grace periods and more aggressively forcing completion of grace-period processing. This evasive action causes the grace period to complete more quickly, but at the cost of restricting RCU’s batching optimizations, thus increasing the CPU overhead incurred by that grace period.}(hj%hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhM hj$hhubeh}(h]j/ah ]h"] corner casesah$]h&]uh1hhjjhhhhhMjKubeh}(h]jHah ]h"]&quality-of-implementation requirementsah$]h&]uh1hhhhhhhhMjKubh)}(hhh](h)}(h!Software-Engineering Requirementsh]h!Software-Engineering Requirements}(hjL%hhhNhNubah}(h]h ]h"]h$]h&]uh1hhjI%hhhhhM.ubh)}(hdBetween Murphy's Law and “To err is human”, it is necessary to guard against mishaps and misuse:h]hfBetween Murphy’s Law and “To err is human”, it is necessary to guard against mishaps and misuse:}(hjZ%hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM0hjI%hhubj)}(hhh](j)}(hXIt is all too easy to forget to use rcu_read_lock() everywhere that it is needed, so kernels built with ``CONFIG_PROVE_RCU=y`` will splat if rcu_dereference() is used outside of an RCU read-side critical section. Update-side code can use rcu_dereference_protected(), which takes a `lockdep expression `__ to indicate what is providing the protection. If the indicated protection is not provided, a lockdep splat is emitted. Code shared between readers and updaters can use rcu_dereference_check(), which also takes a lockdep expression, and emits a lockdep splat if neither rcu_read_lock() nor the indicated protection is in place. In addition, rcu_dereference_raw() is used in those (hopefully rare) cases where the required protection cannot be easily described. Finally, rcu_read_lock_held() is provided to allow a function to verify that it has been invoked within an RCU read-side critical section. I was made aware of this set of requirements shortly after Thomas Gleixner audited a number of RCU uses.h]h)}(hXIt is all too easy to forget to use rcu_read_lock() everywhere that it is needed, so kernels built with ``CONFIG_PROVE_RCU=y`` will splat if rcu_dereference() is used outside of an RCU read-side critical section. Update-side code can use rcu_dereference_protected(), which takes a `lockdep expression `__ to indicate what is providing the protection. If the indicated protection is not provided, a lockdep splat is emitted. Code shared between readers and updaters can use rcu_dereference_check(), which also takes a lockdep expression, and emits a lockdep splat if neither rcu_read_lock() nor the indicated protection is in place. In addition, rcu_dereference_raw() is used in those (hopefully rare) cases where the required protection cannot be easily described. Finally, rcu_read_lock_held() is provided to allow a function to verify that it has been invoked within an RCU read-side critical section. I was made aware of this set of requirements shortly after Thomas Gleixner audited a number of RCU uses.h](hhIt is all too easy to forget to use rcu_read_lock() everywhere that it is needed, so kernels built with }(hjo%hhhNhNubj)}(h``CONFIG_PROVE_RCU=y``h]hCONFIG_PROVE_RCU=y}(hjw%hhhNhNubah}(h]h ]h"]h$]h&]uh1jhjo%ubh will splat if rcu_dereference() is used outside of an RCU read-side critical section. Update-side code can use rcu_dereference_protected(), which takes a }(hjo%hhhNhNubh)}(h9`lockdep expression `__h]hlockdep expression}(hj%hhhNhNubah}(h]h ]h"]h$]h&]namelockdep expressionj https://lwn.net/Articles/371986/uh1hhjo%ubhX to indicate what is providing the protection. If the indicated protection is not provided, a lockdep splat is emitted. Code shared between readers and updaters can use rcu_dereference_check(), which also takes a lockdep expression, and emits a lockdep splat if neither rcu_read_lock() nor the indicated protection is in place. In addition, rcu_dereference_raw() is used in those (hopefully rare) cases where the required protection cannot be easily described. Finally, rcu_read_lock_held() is provided to allow a function to verify that it has been invoked within an RCU read-side critical section. I was made aware of this set of requirements shortly after Thomas Gleixner audited a number of RCU uses.}(hjo%hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhM3hjk%ubah}(h]h ]h"]h$]h&]uh1jhjh%hhhhhNubj)}(hA given function might wish to check for RCU-related preconditions upon entry, before using any other RCU API. The rcu_lockdep_assert() does this job, asserting the expression in kernels having lockdep enabled and doing nothing otherwise.h]h)}(hA given function might wish to check for RCU-related preconditions upon entry, before using any other RCU API. The rcu_lockdep_assert() does this job, asserting the expression in kernels having lockdep enabled and doing nothing otherwise.h]hA given function might wish to check for RCU-related preconditions upon entry, before using any other RCU API. The rcu_lockdep_assert() does this job, asserting the expression in kernels having lockdep enabled and doing nothing otherwise.}(hj%hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMEhj%ubah}(h]h ]h"]h$]h&]uh1jhjh%hhhhhNubj)}(hXIt is also easy to forget to use rcu_assign_pointer() and rcu_dereference(), perhaps (incorrectly) substituting a simple assignment. To catch this sort of error, a given RCU-protected pointer may be tagged with ``__rcu``, after which sparse will complain about simple-assignment accesses to that pointer. Arnd Bergmann made me aware of this requirement, and also supplied the needed `patch series `__.h]h)}(hXIt is also easy to forget to use rcu_assign_pointer() and rcu_dereference(), perhaps (incorrectly) substituting a simple assignment. To catch this sort of error, a given RCU-protected pointer may be tagged with ``__rcu``, after which sparse will complain about simple-assignment accesses to that pointer. Arnd Bergmann made me aware of this requirement, and also supplied the needed `patch series `__.h](hIt is also easy to forget to use rcu_assign_pointer() and rcu_dereference(), perhaps (incorrectly) substituting a simple assignment. To catch this sort of error, a given RCU-protected pointer may be tagged with }(hj%hhhNhNubj)}(h ``__rcu``h]h__rcu}(hj%hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj%ubh, after which sparse will complain about simple-assignment accesses to that pointer. Arnd Bergmann made me aware of this requirement, and also supplied the needed }(hj%hhhNhNubh)}(h3`patch series `__h]h patch series}(hj%hhhNhNubah}(h]h ]h"]h$]h&]name patch seriesj https://lwn.net/Articles/376011/uh1hhj%ubh.}(hj%hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMIhj%ubah}(h]h ]h"]h$]h&]uh1jhjh%hhhhhNubj)}(hXKernels built with ``CONFIG_DEBUG_OBJECTS_RCU_HEAD=y`` will splat if a data element is passed to call_rcu() twice in a row, without a grace period in between. (This error is similar to a double free.) The corresponding ``rcu_head`` structures that are dynamically allocated are automatically tracked, but ``rcu_head`` structures allocated on the stack must be initialized with init_rcu_head_on_stack() and cleaned up with destroy_rcu_head_on_stack(). Similarly, statically allocated non-stack ``rcu_head`` structures must be initialized with init_rcu_head() and cleaned up with destroy_rcu_head(). Mathieu Desnoyers made me aware of this requirement, and also supplied the needed `patch `__.h]h)}(hXKernels built with ``CONFIG_DEBUG_OBJECTS_RCU_HEAD=y`` will splat if a data element is passed to call_rcu() twice in a row, without a grace period in between. (This error is similar to a double free.) The corresponding ``rcu_head`` structures that are dynamically allocated are automatically tracked, but ``rcu_head`` structures allocated on the stack must be initialized with init_rcu_head_on_stack() and cleaned up with destroy_rcu_head_on_stack(). Similarly, statically allocated non-stack ``rcu_head`` structures must be initialized with init_rcu_head() and cleaned up with destroy_rcu_head(). Mathieu Desnoyers made me aware of this requirement, and also supplied the needed `patch `__.h](hKernels built with }(hj&hhhNhNubj)}(h#``CONFIG_DEBUG_OBJECTS_RCU_HEAD=y``h]hCONFIG_DEBUG_OBJECTS_RCU_HEAD=y}(hj &hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj&ubh will splat if a data element is passed to call_rcu() twice in a row, without a grace period in between. (This error is similar to a double free.) The corresponding }(hj&hhhNhNubj)}(h ``rcu_head``h]hrcu_head}(hj&hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj&ubhJ structures that are dynamically allocated are automatically tracked, but }(hj&hhhNhNubj)}(h ``rcu_head``h]hrcu_head}(hj1&hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj&ubh structures allocated on the stack must be initialized with init_rcu_head_on_stack() and cleaned up with destroy_rcu_head_on_stack(). Similarly, statically allocated non-stack }(hj&hhhNhNubj)}(h ``rcu_head``h]hrcu_head}(hjC&hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj&ubh structures must be initialized with init_rcu_head() and cleaned up with destroy_rcu_head(). Mathieu Desnoyers made me aware of this requirement, and also supplied the needed }(hj&hhhNhNubh)}(hD`patch `__h]hpatch}(hjU&hhhNhNubah}(h]h ]h"]h$]h&]namepatchj8https://lore.kernel.org/r/20100319013024.GA28456@Krystaluh1hhj&ubh.}(hj&hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMPhj&ubah}(h]h ]h"]h$]h&]uh1jhjh%hhhhhNubj)}(hXAn infinite loop in an RCU read-side critical section will eventually trigger an RCU CPU stall warning splat, with the duration of “eventually” being controlled by the ``RCU_CPU_STALL_TIMEOUT`` ``Kconfig`` option, or, alternatively, by the ``rcupdate.rcu_cpu_stall_timeout`` boot/sysfs parameter. However, RCU is not obligated to produce this splat unless there is a grace period waiting on that particular RCU read-side critical section. Some extreme workloads might intentionally delay RCU grace periods, and systems running those workloads can be booted with ``rcupdate.rcu_cpu_stall_suppress`` to suppress the splats. This kernel parameter may also be set via ``sysfs``. Furthermore, RCU CPU stall warnings are counter-productive during sysrq dumps and during panics. RCU therefore supplies the rcu_sysrq_start() and rcu_sysrq_end() API members to be called before and after long sysrq dumps. RCU also supplies the rcu_panic() notifier that is automatically invoked at the beginning of a panic to suppress further RCU CPU stall warnings. This requirement made itself known in the early 1990s, pretty much the first time that it was necessary to debug a CPU stall. That said, the initial implementation in DYNIX/ptx was quite generic in comparison with that of Linux. h](h)}(hXAn infinite loop in an RCU read-side critical section will eventually trigger an RCU CPU stall warning splat, with the duration of “eventually” being controlled by the ``RCU_CPU_STALL_TIMEOUT`` ``Kconfig`` option, or, alternatively, by the ``rcupdate.rcu_cpu_stall_timeout`` boot/sysfs parameter. However, RCU is not obligated to produce this splat unless there is a grace period waiting on that particular RCU read-side critical section.h](hAn infinite loop in an RCU read-side critical section will eventually trigger an RCU CPU stall warning splat, with the duration of “eventually” being controlled by the }(hjz&hhhNhNubj)}(h``RCU_CPU_STALL_TIMEOUT``h]hRCU_CPU_STALL_TIMEOUT}(hj&hhhNhNubah}(h]h ]h"]h$]h&]uh1jhjz&ubh }(hjz&hhhNhNubj)}(h ``Kconfig``h]hKconfig}(hj&hhhNhNubah}(h]h ]h"]h$]h&]uh1jhjz&ubh# option, or, alternatively, by the }(hjz&hhhNhNubj)}(h"``rcupdate.rcu_cpu_stall_timeout``h]hrcupdate.rcu_cpu_stall_timeout}(hj&hhhNhNubah}(h]h ]h"]h$]h&]uh1jhjz&ubh boot/sysfs parameter. However, RCU is not obligated to produce this splat unless there is a grace period waiting on that particular RCU read-side critical section.}(hjz&hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhM]hjv&ubh)}(hXZSome extreme workloads might intentionally delay RCU grace periods, and systems running those workloads can be booted with ``rcupdate.rcu_cpu_stall_suppress`` to suppress the splats. This kernel parameter may also be set via ``sysfs``. Furthermore, RCU CPU stall warnings are counter-productive during sysrq dumps and during panics. RCU therefore supplies the rcu_sysrq_start() and rcu_sysrq_end() API members to be called before and after long sysrq dumps. RCU also supplies the rcu_panic() notifier that is automatically invoked at the beginning of a panic to suppress further RCU CPU stall warnings.h](h{Some extreme workloads might intentionally delay RCU grace periods, and systems running those workloads can be booted with }(hj&hhhNhNubj)}(h#``rcupdate.rcu_cpu_stall_suppress``h]hrcupdate.rcu_cpu_stall_suppress}(hj&hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj&ubhC to suppress the splats. This kernel parameter may also be set via }(hj&hhhNhNubj)}(h ``sysfs``h]hsysfs}(hj&hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj&ubhXp. Furthermore, RCU CPU stall warnings are counter-productive during sysrq dumps and during panics. RCU therefore supplies the rcu_sysrq_start() and rcu_sysrq_end() API members to be called before and after long sysrq dumps. RCU also supplies the rcu_panic() notifier that is automatically invoked at the beginning of a panic to suppress further RCU CPU stall warnings.}(hj&hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMehjv&ubh)}(hThis requirement made itself known in the early 1990s, pretty much the first time that it was necessary to debug a CPU stall. That said, the initial implementation in DYNIX/ptx was quite generic in comparison with that of Linux.h]hThis requirement made itself known in the early 1990s, pretty much the first time that it was necessary to debug a CPU stall. That said, the initial implementation in DYNIX/ptx was quite generic in comparison with that of Linux.}(hj&hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMphjv&ubeh}(h]h ]h"]h$]h&]uh1jhjh%hhhhhNubj)}(hXNAlthough it would be very good to detect pointers leaking out of RCU read-side critical sections, there is currently no good way of doing this. One complication is the need to distinguish between pointers leaking and pointers that have been handed off from RCU to some other synchronization mechanism, for example, reference counting.h]h)}(hXNAlthough it would be very good to detect pointers leaking out of RCU read-side critical sections, there is currently no good way of doing this. One complication is the need to distinguish between pointers leaking and pointers that have been handed off from RCU to some other synchronization mechanism, for example, reference counting.h]hXNAlthough it would be very good to detect pointers leaking out of RCU read-side critical sections, there is currently no good way of doing this. One complication is the need to distinguish between pointers leaking and pointers that have been handed off from RCU to some other synchronization mechanism, for example, reference counting.}(hj'hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMuhj'ubah}(h]h ]h"]h$]h&]uh1jhjh%hhhhhNubj)}(hdIn kernels built with ``CONFIG_RCU_TRACE=y``, RCU-related information is provided via event tracing.h]h)}(hdIn kernels built with ``CONFIG_RCU_TRACE=y``, RCU-related information is provided via event tracing.h](hIn kernels built with }(hj 'hhhNhNubj)}(h``CONFIG_RCU_TRACE=y``h]hCONFIG_RCU_TRACE=y}(hj('hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj 'ubh8, RCU-related information is provided via event tracing.}(hj 'hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMzhj'ubah}(h]h ]h"]h$]h&]uh1jhjh%hhhhhNubj)}(hXOpen-coded use of rcu_assign_pointer() and rcu_dereference() to create typical linked data structures can be surprisingly error-prone. Therefore, RCU-protected `linked lists `__ and, more recently, RCU-protected `hash tables `__ are available. Many other special-purpose RCU-protected data structures are available in the Linux kernel and the userspace RCU library.h]h)}(hXOpen-coded use of rcu_assign_pointer() and rcu_dereference() to create typical linked data structures can be surprisingly error-prone. Therefore, RCU-protected `linked lists `__ and, more recently, RCU-protected `hash tables `__ are available. Many other special-purpose RCU-protected data structures are available in the Linux kernel and the userspace RCU library.h](hOpen-coded use of rcu_assign_pointer() and rcu_dereference() to create typical linked data structures can be surprisingly error-prone. Therefore, RCU-protected }(hjJ'hhhNhNubh)}(hE`linked lists `__h]h linked lists}(hjR'hhhNhNubah}(h]h ]h"]h$]h&]name linked listsj2https://lwn.net/Articles/609973/#RCU%20List%20APIsuh1hhjJ'ubh# and, more recently, RCU-protected }(hjJ'hhhNhNubh)}(h2`hash tables `__h]h hash tables}(hjg'hhhNhNubah}(h]h ]h"]h$]h&]name hash tablesj https://lwn.net/Articles/612100/uh1hhjJ'ubh are available. Many other special-purpose RCU-protected data structures are available in the Linux kernel and the userspace RCU library.}(hjJ'hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhM|hjF'ubah}(h]h ]h"]h$]h&]uh1jhjh%hhhhhNubj)}(hSome linked structures are created at compile time, but still require ``__rcu`` checking. The RCU_POINTER_INITIALIZER() macro serves this purpose.h]h)}(hSome linked structures are created at compile time, but still require ``__rcu`` checking. The RCU_POINTER_INITIALIZER() macro serves this purpose.h](hFSome linked structures are created at compile time, but still require }(hj'hhhNhNubj)}(h ``__rcu``h]h__rcu}(hj'hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj'ubhC checking. The RCU_POINTER_INITIALIZER() macro serves this purpose.}(hj'hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhj'ubah}(h]h ]h"]h$]h&]uh1jhjh%hhhhhNubj)}(hIt is not necessary to use rcu_assign_pointer() when creating linked structures that are to be published via a single external pointer. The RCU_INIT_POINTER() macro is provided for this task. h]h)}(hIt is not necessary to use rcu_assign_pointer() when creating linked structures that are to be published via a single external pointer. The RCU_INIT_POINTER() macro is provided for this task.h]hIt is not necessary to use rcu_assign_pointer() when creating linked structures that are to be published via a single external pointer. The RCU_INIT_POINTER() macro is provided for this task.}(hj'hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj'ubah}(h]h ]h"]h$]h&]uh1jhjh%hhhhhNubeh}(h]h ]h"]h$]h&]jjjhjjuh1jhjI%hhhhhM3ubh)}(hThis not a hard-and-fast list: RCU's diagnostic capabilities will continue to be guided by the number and type of usage bugs found in real-world RCU usage.h]hThis not a hard-and-fast list: RCU’s diagnostic capabilities will continue to be guided by the number and type of usage bugs found in real-world RCU usage.}(hj'hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhjI%hhubeh}(h]jah ]h"]!software-engineering requirementsah$]h&]uh1hhhhhhhhM.jKubh)}(hhh](h)}(hLinux Kernel Complicationsh]hLinux Kernel Complications}(hj'hhhNhNubah}(h]h ]h"]h$]h&]uh1hhj'hhhhhMubh)}(hThe Linux kernel provides an interesting environment for all kinds of software, including RCU. Some of the relevant points of interest are as follows:h]hThe Linux kernel provides an interesting environment for all kinds of software, including RCU. Some of the relevant points of interest are as follows:}(hj'hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj'hhubj)}(hhh](j)}(h`Configuration`_h]h)}(hj (h]h)}(hj (h]h Configuration}(hj(hhhNhNubah}(h]h ]h"]h$]h&]name Configurationj configurationuh1hhj (jKubah}(h]h ]h"]h$]h&]uh1hhhhMhj(ubah}(h]h ]h"]h$]h&]uh1jhj(hhhhhNubj)}(h`Firmware Interface`_h]h)}(hj,(h]h)}(hj,(h]hFirmware Interface}(hj1(hhhNhNubah}(h]h ]h"]h$]h&]nameFirmware Interfacejfirmware-interfaceuh1hhj.(jKubah}(h]h ]h"]h$]h&]uh1hhhhMhj*(ubah}(h]h ]h"]h$]h&]uh1jhj(hhhhhNubj)}(h `Early Boot`_h]h)}(hjO(h]h)}(hjO(h]h Early Boot}(hjT(hhhNhNubah}(h]h ]h"]h$]h&]name Early Bootj early-bootuh1hhjQ(jKubah}(h]h ]h"]h$]h&]uh1hhhhMhjM(ubah}(h]h ]h"]h$]h&]uh1jhj(hhhhhNubj)}(h`Interrupts and NMIs`_h]h)}(hjr(h]h)}(hjr(h]hInterrupts and NMIs}(hjw(hhhNhNubah}(h]h ]h"]h$]h&]nameInterrupts and NMIsjinterrupts-and-nmisuh1hhjt(jKubah}(h]h ]h"]h$]h&]uh1hhhhMhjp(ubah}(h]h ]h"]h$]h&]uh1jhj(hhhhhNubj)}(h`Loadable Modules`_h]h)}(hj(h]h)}(hj(h]hLoadable Modules}(hj(hhhNhNubah}(h]h ]h"]h$]h&]nameLoadable Modulesjloadable-modulesuh1hhj(jKubah}(h]h ]h"]h$]h&]uh1hhhhMhj(ubah}(h]h ]h"]h$]h&]uh1jhj(hhhhhNubj)}(h`Hotplug CPU`_h]h)}(hj(h]h)}(hj(h]h Hotplug CPU}(hj(hhhNhNubah}(h]h ]h"]h$]h&]name Hotplug CPUj hotplug-cpuuh1hhj(jKubah}(h]h ]h"]h$]h&]uh1hhhhMhj(ubah}(h]h ]h"]h$]h&]uh1jhj(hhhhhNubj)}(h`Scheduler and RCU`_h]h)}(hj(h]h)}(hj(h]hScheduler and RCU}(hj(hhhNhNubah}(h]h ]h"]h$]h&]nameScheduler and RCUjscheduler-and-rcuuh1hhj(jKubah}(h]h ]h"]h$]h&]uh1hhhhMhj(ubah}(h]h ]h"]h$]h&]uh1jhj(hhhhhNubj)}(h`Tracing and RCU`_h]h)}(hj(h]h)}(hj(h]hTracing and RCU}(hj)hhhNhNubah}(h]h ]h"]h$]h&]nameTracing and RCUjtracing-and-rcuuh1hhj)jKubah}(h]h ]h"]h$]h&]uh1hhhhMhj(ubah}(h]h ]h"]h$]h&]uh1jhj(hhhhhNubj)}(h"`Accesses to User Memory and RCU`_h]h)}(hj!)h]h)}(hj!)h]hAccesses to User Memory and RCU}(hj&)hhhNhNubah}(h]h ]h"]h$]h&]nameAccesses to User Memory and RCUjaccesses-to-user-memory-and-rcuuh1hhj#)jKubah}(h]h ]h"]h$]h&]uh1hhhhMhj)ubah}(h]h ]h"]h$]h&]uh1jhj(hhhhhNubj)}(h`Energy Efficiency`_h]h)}(hjD)h]h)}(hjD)h]hEnergy Efficiency}(hjI)hhhNhNubah}(h]h ]h"]h$]h&]nameEnergy Efficiencyjenergy-efficiencyuh1hhjF)jKubah}(h]h ]h"]h$]h&]uh1hhhhMhjB)ubah}(h]h ]h"]h$]h&]uh1jhj(hhhhhNubj)}(h&`Scheduling-Clock Interrupts and RCU`_h]h)}(hjg)h]h)}(hjg)h]h#Scheduling-Clock Interrupts and RCU}(hjl)hhhNhNubah}(h]h ]h"]h$]h&]name#Scheduling-Clock Interrupts and RCUj#scheduling-clock-interrupts-and-rcuuh1hhji)jKubah}(h]h ]h"]h$]h&]uh1hhhhMhje)ubah}(h]h ]h"]h$]h&]uh1jhj(hhhhhNubj)}(h`Memory Efficiency`_h]h)}(hj)h]h)}(hj)h]hMemory Efficiency}(hj)hhhNhNubah}(h]h ]h"]h$]h&]nameMemory Efficiencyjmemory-efficiencyuh1hhj)jKubah}(h]h ]h"]h$]h&]uh1hhhhMhj)ubah}(h]h ]h"]h$]h&]uh1jhj(hhhhhNubj)}(h<`Performance, Scalability, Response Time, and Reliability`_ h]h)}(h;`Performance, Scalability, Response Time, and Reliability`_h]h)}(hj)h]h8Performance, Scalability, Response Time, and Reliability}(hj)hhhNhNubah}(h]h ]h"]h$]h&]name8Performance, Scalability, Response Time, and Reliabilityj5performance-scalability-response-time-and-reliabilityuh1hhj)jKubah}(h]h ]h"]h$]h&]uh1hhhhMhj)ubah}(h]h ]h"]h$]h&]uh1jhj(hhhhhNubeh}(h]h ]h"]h$]h&]jjjhjjuh1jhj'hhhhhMubh)}(hThis list is probably incomplete, but it does give a feel for the most notable Linux-kernel complications. Each of the following sections covers one of the above topics.h]hThis list is probably incomplete, but it does give a feel for the most notable Linux-kernel complications. Each of the following sections covers one of the above topics.}(hj)hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj'hhubh)}(hhh](h)}(h Configurationh]h Configuration}(hj)hhhNhNubah}(h]h ]h"]h$]h&]uh1hhj)hhhhhMubh)}(hRCU's goal is automatic configuration, so that almost nobody needs to worry about RCU's ``Kconfig`` options. And for almost all users, RCU does in fact work well “out of the box.”h](h\RCU’s goal is automatic configuration, so that almost nobody needs to worry about RCU’s }(hj)hhhNhNubj)}(h ``Kconfig``h]hKconfig}(hj)hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj)ubhT options. And for almost all users, RCU does in fact work well “out of the box.”}(hj)hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhj)hhubh)}(hX*However, there are specialized use cases that are handled by kernel boot parameters and ``Kconfig`` options. Unfortunately, the ``Kconfig`` system will explicitly ask users about new ``Kconfig`` options, which requires almost all of them be hidden behind a ``CONFIG_RCU_EXPERT`` ``Kconfig`` option.h](hXHowever, there are specialized use cases that are handled by kernel boot parameters and }(hj*hhhNhNubj)}(h ``Kconfig``h]hKconfig}(hj*hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj*ubh options. Unfortunately, the }(hj*hhhNhNubj)}(h ``Kconfig``h]hKconfig}(hj.*hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj*ubh, system will explicitly ask users about new }(hj*hhhNhNubj)}(h ``Kconfig``h]hKconfig}(hj@*hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj*ubh? options, which requires almost all of them be hidden behind a }(hj*hhhNhNubj)}(h``CONFIG_RCU_EXPERT``h]hCONFIG_RCU_EXPERT}(hjR*hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj*ubh }(hj*hhhNhNubj)}(h ``Kconfig``h]hKconfig}(hjd*hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj*ubh option.}(hj*hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhj)hhubh)}(hThis all should be quite obvious, but the fact remains that Linus Torvalds recently had to `remind `__ me of this requirement.h](h[This all should be quite obvious, but the fact remains that Linus Torvalds recently had to }(hj|*hhhNhNubh)}(hi`remind `__h]hremind}(hj*hhhNhNubah}(h]h ]h"]h$]h&]nameremindj\https://lore.kernel.org/r/CA+55aFy4wcCwaL4okTs8wXhGZ5h-ibecy_Meg9C4MNQrUnwMcg@mail.gmail.comuh1hhj|*ubh me of this requirement.}(hj|*hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhj)hhubeh}(h]j(ah ]h"] configurationah$]h&]uh1hhj'hhhhhMjKubh)}(hhh](h)}(hFirmware Interfaceh]hFirmware Interface}(hj*hhhNhNubah}(h]h ]h"]h$]h&]uh1hhj*hhhhhMubh)}(hIn many cases, kernel obtains information about the system from the firmware, and sometimes things are lost in translation. Or the translation is accurate, but the original message is bogus.h]hIn many cases, kernel obtains information about the system from the firmware, and sometimes things are lost in translation. Or the translation is accurate, but the original message is bogus.}(hj*hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj*hhubh)}(hXfFor example, some systems' firmware overreports the number of CPUs, sometimes by a large factor. If RCU naively believed the firmware, as it used to do, it would create too many per-CPU kthreads. Although the resulting system will still run correctly, the extra kthreads needlessly consume memory and can cause confusion when they show up in ``ps`` listings.h](hXXFor example, some systems’ firmware overreports the number of CPUs, sometimes by a large factor. If RCU naively believed the firmware, as it used to do, it would create too many per-CPU kthreads. Although the resulting system will still run correctly, the extra kthreads needlessly consume memory and can cause confusion when they show up in }(hj*hhhNhNubj)}(h``ps``h]hps}(hj*hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj*ubh listings.}(hj*hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhj*hhubh)}(hX'RCU must therefore wait for a given CPU to actually come online before it can allow itself to believe that the CPU actually exists. The resulting “ghost CPUs” (which are never going to come online) cause a number of `interesting complications `__.h](hRCU must therefore wait for a given CPU to actually come online before it can allow itself to believe that the CPU actually exists. The resulting “ghost CPUs” (which are never going to come online) cause a number of }(hj*hhhNhNubh)}(hJ`interesting complications `__h]hinteresting complications}(hj*hhhNhNubah}(h]h ]h"]h$]h&]nameinteresting complicationsj*https://paulmck.livejournal.com/37494.htmluh1hhj*ubh.}(hj*hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhj*hhubeh}(h]j@(ah ]h"]firmware interfaceah$]h&]uh1hhj'hhhhhMjKubh)}(hhh](h)}(h Early Booth]h Early Boot}(hj+hhhNhNubah}(h]h ]h"]h$]h&]uh1hhj+hhhhhMubh)}(hXThe Linux kernel's boot sequence is an interesting process, and RCU is used early, even before rcu_init() is invoked. In fact, a number of RCU's primitives can be used as soon as the initial task's ``task_struct`` is available and the boot CPU's per-CPU variables are set up. The read-side primitives (rcu_read_lock(), rcu_read_unlock(), rcu_dereference(), and rcu_access_pointer()) will operate normally very early on, as will rcu_assign_pointer().h](hThe Linux kernel’s boot sequence is an interesting process, and RCU is used early, even before rcu_init() is invoked. In fact, a number of RCU’s primitives can be used as soon as the initial task’s }(hj +hhhNhNubj)}(h``task_struct``h]h task_struct}(hj(+hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj +ubh is available and the boot CPU’s per-CPU variables are set up. The read-side primitives (rcu_read_lock(), rcu_read_unlock(), rcu_dereference(), and rcu_access_pointer()) will operate normally very early on, as will rcu_assign_pointer().}(hj +hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhj+hhubh)}(hXAlthough call_rcu() may be invoked at any time during boot, callbacks are not guaranteed to be invoked until after all of RCU's kthreads have been spawned, which occurs at early_initcall() time. This delay in callback invocation is due to the fact that RCU does not invoke callbacks until it is fully initialized, and this full initialization cannot occur until after the scheduler has initialized itself to the point where RCU can spawn and run its kthreads. In theory, it would be possible to invoke callbacks earlier, however, this is not a panacea because there would be severe restrictions on what operations those callbacks could invoke.h]hXAlthough call_rcu() may be invoked at any time during boot, callbacks are not guaranteed to be invoked until after all of RCU’s kthreads have been spawned, which occurs at early_initcall() time. This delay in callback invocation is due to the fact that RCU does not invoke callbacks until it is fully initialized, and this full initialization cannot occur until after the scheduler has initialized itself to the point where RCU can spawn and run its kthreads. In theory, it would be possible to invoke callbacks earlier, however, this is not a panacea because there would be severe restrictions on what operations those callbacks could invoke.}(hj@+hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj+hhubh)}(hX\Perhaps surprisingly, synchronize_rcu() and synchronize_rcu_expedited(), will operate normally during very early boot, the reason being that there is only one CPU and preemption is disabled. This means that the call synchronize_rcu() (or friends) itself is a quiescent state and thus a grace period, so the early-boot implementation can be a no-op.h]hX\Perhaps surprisingly, synchronize_rcu() and synchronize_rcu_expedited(), will operate normally during very early boot, the reason being that there is only one CPU and preemption is disabled. This means that the call synchronize_rcu() (or friends) itself is a quiescent state and thus a grace period, so the early-boot implementation can be a no-op.}(hjN+hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj+hhubh)}(hXHowever, once the scheduler has spawned its first kthread, this early boot trick fails for synchronize_rcu() (as well as for synchronize_rcu_expedited()) in ``CONFIG_PREEMPTION=y`` kernels. The reason is that an RCU read-side critical section might be preempted, which means that a subsequent synchronize_rcu() really does have to wait for something, as opposed to simply returning immediately. Unfortunately, synchronize_rcu() can't do this until all of its kthreads are spawned, which doesn't happen until some time during early_initcalls() time. But this is no excuse: RCU is nevertheless required to correctly handle synchronous grace periods during this time period. Once all of its kthreads are up and running, RCU starts running normally.h](hHowever, once the scheduler has spawned its first kthread, this early boot trick fails for synchronize_rcu() (as well as for synchronize_rcu_expedited()) in }(hj\+hhhNhNubj)}(h``CONFIG_PREEMPTION=y``h]hCONFIG_PREEMPTION=y}(hjd+hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj\+ubhX9 kernels. The reason is that an RCU read-side critical section might be preempted, which means that a subsequent synchronize_rcu() really does have to wait for something, as opposed to simply returning immediately. Unfortunately, synchronize_rcu() can’t do this until all of its kthreads are spawned, which doesn’t happen until some time during early_initcalls() time. But this is no excuse: RCU is nevertheless required to correctly handle synchronous grace periods during this time period. Once all of its kthreads are up and running, RCU starts running normally.}(hj\+hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhj+hhubj)}(hhh]j)}(hhh](j)}(hhh]h}(h]h ]h"]h$]h&]colwidthKGuh1jhj+ubj)}(hhh](j)}(hhh]j )}(hhh]h)}(h**Quick Quiz**:h](j)}(h**Quick Quiz**h]h Quick Quiz}(hj+hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj+ubh:}(hj+hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhj+ubah}(h]h ]h"]h$]h&]uh1j hj+ubah}(h]h ]h"]h$]h&]uh1jhj+ubj)}(hhh]j )}(hhh]h)}(hYHow can RCU possibly handle grace periods before all of its kthreads have been spawned???h]hYHow can RCU possibly handle grace periods before all of its kthreads have been spawned???}(hj+hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj+ubah}(h]h ]h"]h$]h&]uh1j hj+ubah}(h]h ]h"]h$]h&]uh1jhj+ubj)}(hhh]j )}(hhh]h)}(h **Answer**:h](j)}(h **Answer**h]hAnswer}(hj+hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj+ubh:}(hj+hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhj+ubah}(h]h ]h"]h$]h&]uh1j hj+ubah}(h]h ]h"]h$]h&]uh1jhj+ubj)}(hhh]j )}(hhh](h)}(hXVery carefully! During the “dead zone” between the time that the scheduler spawns the first task and the time that all of RCU's kthreads have been spawned, all synchronous grace periods are handled by the expedited grace-period mechanism. At runtime, this expedited mechanism relies on workqueues, but during the dead zone the requesting task itself drives the desired expedited grace period. Because dead-zone execution takes place within task context, everything works. Once the dead zone ends, expedited grace periods go back to using workqueues, as is required to avoid problems that would otherwise occur when a user task received a POSIX signal while driving an expedited grace period.h]hXVery carefully! During the “dead zone” between the time that the scheduler spawns the first task and the time that all of RCU’s kthreads have been spawned, all synchronous grace periods are handled by the expedited grace-period mechanism. At runtime, this expedited mechanism relies on workqueues, but during the dead zone the requesting task itself drives the desired expedited grace period. Because dead-zone execution takes place within task context, everything works. Once the dead zone ends, expedited grace periods go back to using workqueues, as is required to avoid problems that would otherwise occur when a user task received a POSIX signal while driving an expedited grace period.}(hj,hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj,ubh)}(hXAnd yes, this does mean that it is unhelpful to send POSIX signals to random tasks between the time that the scheduler spawns its first kthread and the time that RCU's kthreads have all been spawned. If there ever turns out to be a good reason for sending POSIX signals during that time, appropriate adjustments will be made. (If it turns out that POSIX signals are sent during this time for no good reason, other adjustments will be made, appropriate or otherwise.)h]hXAnd yes, this does mean that it is unhelpful to send POSIX signals to random tasks between the time that the scheduler spawns its first kthread and the time that RCU’s kthreads have all been spawned. If there ever turns out to be a good reason for sending POSIX signals during that time, appropriate adjustments will be made. (If it turns out that POSIX signals are sent during this time for no good reason, other adjustments will be made, appropriate or otherwise.)}(hj,hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj,ubeh}(h]h ]h"]h$]h&]uh1j hj ,ubah}(h]h ]h"]h$]h&]uh1jhj+ubeh}(h]h ]h"]h$]h&]uh1jhj+ubeh}(h]h ]h"]h$]h&]colsKuh1jhj|+ubah}(h]h ]h"]h$]h&]uh1jhj+hhhhhNubh)}(hRI learned of these boot-time requirements as a result of a series of system hangs.h]hRI learned of these boot-time requirements as a result of a series of system hangs.}(hjL,hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj+hhubeh}(h]jc(ah ]h"] early bootah$]h&]uh1hhj'hhhhhMjKubh)}(hhh](h)}(hInterrupts and NMIsh]hInterrupts and NMIs}(hjd,hhhNhNubah}(h]h ]h"]h$]h&]uh1hhja,hhhhhMubh)}(hThe Linux kernel has interrupts, and RCU read-side critical sections are legal within interrupt handlers and within interrupt-disabled regions of code, as are invocations of call_rcu().h]hThe Linux kernel has interrupts, and RCU read-side critical sections are legal within interrupt handlers and within interrupt-disabled regions of code, as are invocations of call_rcu().}(hjr,hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhja,hhubh)}(hXSome Linux-kernel architectures can enter an interrupt handler from non-idle process context, and then just never leave it, instead stealthily transitioning back to process context. This trick is sometimes used to invoke system calls from inside the kernel. These “half-interrupts” mean that RCU has to be very careful about how it counts interrupt nesting levels. I learned of this requirement the hard way during a rewrite of RCU's dyntick-idle code.h]hXSome Linux-kernel architectures can enter an interrupt handler from non-idle process context, and then just never leave it, instead stealthily transitioning back to process context. This trick is sometimes used to invoke system calls from inside the kernel. These “half-interrupts” mean that RCU has to be very careful about how it counts interrupt nesting levels. I learned of this requirement the hard way during a rewrite of RCU’s dyntick-idle code.}(hj,hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM!hja,hhubh)}(hThe Linux kernel has non-maskable interrupts (NMIs), and RCU read-side critical sections are legal within NMI handlers. Thankfully, RCU update-side primitives, including call_rcu(), are prohibited within NMI handlers.h]hThe Linux kernel has non-maskable interrupts (NMIs), and RCU read-side critical sections are legal within NMI handlers. Thankfully, RCU update-side primitives, including call_rcu(), are prohibited within NMI handlers.}(hj,hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM)hja,hhubh)}(hXThe name notwithstanding, some Linux-kernel architectures can have nested NMIs, which RCU must handle correctly. Andy Lutomirski `surprised me `__ with this requirement; he also kindly surprised me with `an algorithm `__ that meets this requirement.h](hThe name notwithstanding, some Linux-kernel architectures can have nested NMIs, which RCU must handle correctly. Andy Lutomirski }(hj,hhhNhNubh)}(ho`surprised me `__h]h surprised me}(hj,hhhNhNubah}(h]h ]h"]h$]h&]name surprised mej\https://lore.kernel.org/r/CALCETrXLq1y7e_dKFPgou-FKHB6Pu-r8+t-6Ds+8=va7anBWDA@mail.gmail.comuh1hhj,ubh9 with this requirement; he also kindly surprised me with }(hj,hhhNhNubh)}(ho`an algorithm `__h]h an algorithm}(hj,hhhNhNubah}(h]h ]h"]h$]h&]name an algorithmj\https://lore.kernel.org/r/CALCETrXSY9JpW3uE6H8WYk81sg56qasA2aqmjMPsq5dOtzso=g@mail.gmail.comuh1hhj,ubh that meets this requirement.}(hj,hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhM.hja,hhubh)}(hXFurthermore, NMI handlers can be interrupted by what appear to RCU to be normal interrupts. One way that this can happen is for code that directly invokes ct_irq_enter() and ct_irq_exit() to be called from an NMI handler. This astonishing fact of life prompted the current code structure, which has ct_irq_enter() invoking ct_nmi_enter() and ct_irq_exit() invoking ct_nmi_exit(). And yes, I also learned of this requirement the hard way.h]hXFurthermore, NMI handlers can be interrupted by what appear to RCU to be normal interrupts. One way that this can happen is for code that directly invokes ct_irq_enter() and ct_irq_exit() to be called from an NMI handler. This astonishing fact of life prompted the current code structure, which has ct_irq_enter() invoking ct_nmi_enter() and ct_irq_exit() invoking ct_nmi_exit(). And yes, I also learned of this requirement the hard way.}(hj,hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM5hja,hhubeh}(h]j(ah ]h"]interrupts and nmisah$]h&]uh1hhj'hhhhhMjKubh)}(hhh](h)}(hLoadable Modulesh]hLoadable Modules}(hj,hhhNhNubah}(h]h ]h"]h$]h&]uh1hhj,hhhhhM>ubh)}(hXThe Linux kernel has loadable modules, and these modules can also be unloaded. After a given module has been unloaded, any attempt to call one of its functions results in a segmentation fault. The module-unload functions must therefore cancel any delayed calls to loadable-module functions, for example, any outstanding mod_timer() must be dealt with via timer_shutdown_sync() or similar.h]hXThe Linux kernel has loadable modules, and these modules can also be unloaded. After a given module has been unloaded, any attempt to call one of its functions results in a segmentation fault. The module-unload functions must therefore cancel any delayed calls to loadable-module functions, for example, any outstanding mod_timer() must be dealt with via timer_shutdown_sync() or similar.}(hj,hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM@hj,hhubh)}(hXcUnfortunately, there is no way to cancel an RCU callback; once you invoke call_rcu(), the callback function is eventually going to be invoked, unless the system goes down first. Because it is normally considered socially irresponsible to crash the system in response to a module unload request, we need some other way to deal with in-flight RCU callbacks.h]hXcUnfortunately, there is no way to cancel an RCU callback; once you invoke call_rcu(), the callback function is eventually going to be invoked, unless the system goes down first. Because it is normally considered socially irresponsible to crash the system in response to a module unload request, we need some other way to deal with in-flight RCU callbacks.}(hj-hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMGhj,hhubh)}(hXRCU therefore provides rcu_barrier(), which waits until all in-flight RCU callbacks have been invoked. If a module uses call_rcu(), its exit function should therefore prevent any future invocation of call_rcu(), then invoke rcu_barrier(). In theory, the underlying module-unload code could invoke rcu_barrier() unconditionally, but in practice this would incur unacceptable latencies.h]hXRCU therefore provides rcu_barrier(), which waits until all in-flight RCU callbacks have been invoked. If a module uses call_rcu(), its exit function should therefore prevent any future invocation of call_rcu(), then invoke rcu_barrier(). In theory, the underlying module-unload code could invoke rcu_barrier() unconditionally, but in practice this would incur unacceptable latencies.}(hj-hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMNhj,hhubh)}(hNikita Danilov noted this requirement for an analogous filesystem-unmount situation, and Dipankar Sarma incorporated rcu_barrier() into RCU. The need for rcu_barrier() for module unloading became apparent later.h]hNikita Danilov noted this requirement for an analogous filesystem-unmount situation, and Dipankar Sarma incorporated rcu_barrier() into RCU. The need for rcu_barrier() for module unloading became apparent later.}(hj$-hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMVhj,hhubh important)}(hXThe rcu_barrier() function is not, repeat, *not*, obligated to wait for a grace period. It is instead only required to wait for RCU callbacks that have already been posted. Therefore, if there are no RCU callbacks posted anywhere in the system, rcu_barrier() is within its rights to return immediately. Even if there are callbacks posted, rcu_barrier() does not necessarily need to wait for a grace period.h]h)}(hXThe rcu_barrier() function is not, repeat, *not*, obligated to wait for a grace period. It is instead only required to wait for RCU callbacks that have already been posted. Therefore, if there are no RCU callbacks posted anywhere in the system, rcu_barrier() is within its rights to return immediately. Even if there are callbacks posted, rcu_barrier() does not necessarily need to wait for a grace period.h](h+The rcu_barrier() function is not, repeat, }(hj8-hhhNhNubj9 )}(h*not*h]hnot}(hj@-hhhNhNubah}(h]h ]h"]h$]h&]uh1j8 hj8-ubhXf, obligated to wait for a grace period. It is instead only required to wait for RCU callbacks that have already been posted. Therefore, if there are no RCU callbacks posted anywhere in the system, rcu_barrier() is within its rights to return immediately. Even if there are callbacks posted, rcu_barrier() does not necessarily need to wait for a grace period.}(hj8-hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhM]hj4-ubah}(h]h ]h"]h$]h&]uh1j2-hj,hhhhhNubj)}(hhh]j)}(hhh](j)}(hhh]h}(h]h ]h"]h$]h&]colwidthKGuh1jhja-ubj)}(hhh](j)}(hhh]j )}(hhh]h)}(h**Quick Quiz**:h](j)}(h**Quick Quiz**h]h Quick Quiz}(hj{-hhhNhNubah}(h]h ]h"]h$]h&]uh1jhjw-ubh:}(hjw-hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMfhjt-ubah}(h]h ]h"]h$]h&]uh1j hjq-ubah}(h]h ]h"]h$]h&]uh1jhjn-ubj)}(hhh]j )}(hhh]h)}(hXWait a minute! Each RCU callbacks must wait for a grace period to complete, and rcu_barrier() must wait for each pre-existing callback to be invoked. Doesn't rcu_barrier() therefore need to wait for a full grace period if there is even one callback posted anywhere in the system?h]hXWait a minute! Each RCU callbacks must wait for a grace period to complete, and rcu_barrier() must wait for each pre-existing callback to be invoked. Doesn’t rcu_barrier() therefore need to wait for a full grace period if there is even one callback posted anywhere in the system?}(hj-hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhhj-ubah}(h]h ]h"]h$]h&]uh1j hj-ubah}(h]h ]h"]h$]h&]uh1jhjn-ubj)}(hhh]j )}(hhh]h)}(h **Answer**:h](j)}(h **Answer**h]hAnswer}(hj-hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj-ubh:}(hj-hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMnhj-ubah}(h]h ]h"]h$]h&]uh1j hj-ubah}(h]h ]h"]h$]h&]uh1jhjn-ubj)}(hhh]j )}(hhh](h)}(hXAbsolutely not!!! Yes, each RCU callbacks must wait for a grace period to complete, but it might well be partly (or even completely) finished waiting by the time rcu_barrier() is invoked. In that case, rcu_barrier() need only wait for the remaining portion of the grace period to elapse. So even if there are quite a few callbacks posted, rcu_barrier() might well return quite quickly.h]hXAbsolutely not!!! Yes, each RCU callbacks must wait for a grace period to complete, but it might well be partly (or even completely) finished waiting by the time rcu_barrier() is invoked. In that case, rcu_barrier() need only wait for the remaining portion of the grace period to elapse. So even if there are quite a few callbacks posted, rcu_barrier() might well return quite quickly.}(hj-hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMphj-ubh)}(hSo if you need to wait for a grace period as well as for all pre-existing callbacks, you will need to invoke both synchronize_rcu() and rcu_barrier(). If latency is a concern, you can always use workqueues to invoke them concurrently. h]hSo if you need to wait for a grace period as well as for all pre-existing callbacks, you will need to invoke both synchronize_rcu() and rcu_barrier(). If latency is a concern, you can always use workqueues to invoke them concurrently.}(hj.hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMxhj-ubeh}(h]h ]h"]h$]h&]uh1j hj-ubah}(h]h ]h"]h$]h&]uh1jhjn-ubeh}(h]h ]h"]h$]h&]uh1jhja-ubeh}(h]h ]h"]h$]h&]colsKuh1jhj^-ubah}(h]h ]h"]h$]h&]uh1jhj,hhhhhNubeh}(h]j(ah ]h"]loadable modulesah$]h&]uh1hhj'hhhhhM>jKubh)}(hhh](h)}(h Hotplug CPUh]h Hotplug CPU}(hj8.hhhNhNubah}(h]h ]h"]h$]h&]uh1hhj5.hhhhhMubh)}(hXtThe Linux kernel supports CPU hotplug, which means that CPUs can come and go. It is of course illegal to use any RCU API member from an offline CPU, with the exception of `SRCU `__ read-side critical sections. This requirement was present from day one in DYNIX/ptx, but on the other hand, the Linux kernel's CPU-hotplug implementation is “interesting.”h](hThe Linux kernel supports CPU hotplug, which means that CPUs can come and go. It is of course illegal to use any RCU API member from an offline CPU, with the exception of }(hjF.hhhNhNubh)}(h`SRCU `__h]hSRCU}(hjN.hhhNhNubah}(h]h ]h"]h$]h&]nameSRCUjjuh1hhjF.jKubh read-side critical sections. This requirement was present from day one in DYNIX/ptx, but on the other hand, the Linux kernel’s CPU-hotplug implementation is “interesting.”}(hjF.hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhj5.hhubh)}(hX<The Linux-kernel CPU-hotplug implementation has notifiers that are used to allow the various kernel subsystems (including RCU) to respond appropriately to a given CPU-hotplug operation. Most RCU operations may be invoked from CPU-hotplug notifiers, including even synchronous grace-period operations such as (synchronize_rcu() and synchronize_rcu_expedited()). However, these synchronous operations do block and therefore cannot be invoked from notifiers that execute via stop_machine(), specifically those between the ``CPUHP_AP_OFFLINE`` and ``CPUHP_AP_ONLINE`` states.h](hXThe Linux-kernel CPU-hotplug implementation has notifiers that are used to allow the various kernel subsystems (including RCU) to respond appropriately to a given CPU-hotplug operation. Most RCU operations may be invoked from CPU-hotplug notifiers, including even synchronous grace-period operations such as (synchronize_rcu() and synchronize_rcu_expedited()). However, these synchronous operations do block and therefore cannot be invoked from notifiers that execute via stop_machine(), specifically those between the }(hjh.hhhNhNubj)}(h``CPUHP_AP_OFFLINE``h]hCPUHP_AP_OFFLINE}(hjp.hhhNhNubah}(h]h ]h"]h$]h&]uh1jhjh.ubh and }(hjh.hhhNhNubj)}(h``CPUHP_AP_ONLINE``h]hCPUHP_AP_ONLINE}(hj.hhhNhNubah}(h]h ]h"]h$]h&]uh1jhjh.ubh states.}(hjh.hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhj5.hhubh)}(hXIn addition, all-callback-wait operations such as rcu_barrier() may not be invoked from any CPU-hotplug notifier. This restriction is due to the fact that there are phases of CPU-hotplug operations where the outgoing CPU's callbacks will not be invoked until after the CPU-hotplug operation ends, which could also result in deadlock. Furthermore, rcu_barrier() blocks CPU-hotplug operations during its execution, which results in another type of deadlock when invoked from a CPU-hotplug notifier.h]hXIn addition, all-callback-wait operations such as rcu_barrier() may not be invoked from any CPU-hotplug notifier. This restriction is due to the fact that there are phases of CPU-hotplug operations where the outgoing CPU’s callbacks will not be invoked until after the CPU-hotplug operation ends, which could also result in deadlock. Furthermore, rcu_barrier() blocks CPU-hotplug operations during its execution, which results in another type of deadlock when invoked from a CPU-hotplug notifier.}(hj.hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj5.hhubh)}(hX%Finally, RCU must avoid deadlocks due to interaction between hotplug, timers and grace period processing. It does so by maintaining its own set of books that duplicate the centrally maintained ``cpu_online_mask``, and also by reporting quiescent states explicitly when a CPU goes offline. This explicit reporting of quiescent states avoids any need for the force-quiescent-state loop (FQS) to report quiescent states for offline CPUs. However, as a debugging measure, the FQS loop does splat if offline CPUs block an RCU grace period for too long.h](hFinally, RCU must avoid deadlocks due to interaction between hotplug, timers and grace period processing. It does so by maintaining its own set of books that duplicate the centrally maintained }(hj.hhhNhNubj)}(h``cpu_online_mask``h]hcpu_online_mask}(hj.hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj.ubhXQ, and also by reporting quiescent states explicitly when a CPU goes offline. This explicit reporting of quiescent states avoids any need for the force-quiescent-state loop (FQS) to report quiescent states for offline CPUs. However, as a debugging measure, the FQS loop does splat if offline CPUs block an RCU grace period for too long.}(hj.hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhj5.hhubh)}(h9An offline CPU's quiescent state will be reported either:h]h;An offline CPU’s quiescent state will be reported either:}(hj.hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj5.hhubj)}(hhh](j)}(hQAs the CPU goes offline using RCU's hotplug notifier (rcutree_report_cpu_dead()).h]h)}(hj.h]hSAs the CPU goes offline using RCU’s hotplug notifier (rcutree_report_cpu_dead()).}(hj.hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj.ubah}(h]h ]h"]h$]h&]uh1jhj.hhhhhNubj)}(hWhen grace period initialization (rcu_gp_init()) detects a race either with CPU offlining or with a task unblocking on a leaf ``rcu_node`` structure whose CPUs are all offline. h]h)}(hWhen grace period initialization (rcu_gp_init()) detects a race either with CPU offlining or with a task unblocking on a leaf ``rcu_node`` structure whose CPUs are all offline.h](h~When grace period initialization (rcu_gp_init()) detects a race either with CPU offlining or with a task unblocking on a leaf }(hj.hhhNhNubj)}(h ``rcu_node``h]hrcu_node}(hj.hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj.ubh& structure whose CPUs are all offline.}(hj.hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhj.ubah}(h]h ]h"]h$]h&]uh1jhj.hhhhhNubeh}(h]h ]h"]h$]h&]jjjhjjuh1jhj5.hhhhhMubh)}(hThe CPU-online path (rcutree_report_cpu_starting()) should never need to report a quiescent state for an offline CPU. However, as a debugging measure, it does emit a warning if a quiescent state was not already reported for that CPU.h]hThe CPU-online path (rcutree_report_cpu_starting()) should never need to report a quiescent state for an offline CPU. However, as a debugging measure, it does emit a warning if a quiescent state was not already reported for that CPU.}(hj /hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj5.hhubh)}(hXDuring the checking/modification of RCU's hotplug bookkeeping, the corresponding CPU's leaf node lock is held. This avoids race conditions between RCU's hotplug notifier hooks, the grace period initialization code, and the FQS loop, all of which refer to or modify this bookkeeping.h]hX During the checking/modification of RCU’s hotplug bookkeeping, the corresponding CPU’s leaf node lock is held. This avoids race conditions between RCU’s hotplug notifier hooks, the grace period initialization code, and the FQS loop, all of which refer to or modify this bookkeeping.}(hj./hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj5.hhubh)}(hXNote that grace period initialization (rcu_gp_init()) must carefully sequence CPU hotplug scanning with grace period state changes. For example, the following race could occur in rcu_gp_init() if rcu_seq_start() were to happen after the CPU hotplug scanning::h]hXNote that grace period initialization (rcu_gp_init()) must carefully sequence CPU hotplug scanning with grace period state changes. For example, the following race could occur in rcu_gp_init() if rcu_seq_start() were to happen after the CPU hotplug scanning:}(hjqsmaskinit = rnp->qsmaskinitnext; } rcutree_report_cpu_starting() rnp->qsmaskinitnext |= mask; rcu_read_lock() r0 = *X; r1 = *X; X = NULL; cookie = get_state_synchronize_rcu(); // cookie = 8 (future GP) rcu_seq_start(&rcu_state.gp_seq); // gp_seq = 5 // CPU1 now invisible to this GP! rcu_for_each_node_breadth_first() { rnp->qsmask = rnp->qsmaskinit; // CPU1 not included! } // GP completes without CPU1 rcu_seq_end(&rcu_state.gp_seq); // gp_seq = 8 poll_state_synchronize_rcu(cookie); // Returns true! kfree(r1); r2 = *r0; // USE-AFTER-FREE!h]hXCPU0 (rcu_gp_init) CPU1 CPU2 --------------------- ---- ---- // Hotplug scan first (WRONG ORDER) rcu_for_each_leaf_node(rnp) { rnp->qsmaskinit = rnp->qsmaskinitnext; } rcutree_report_cpu_starting() rnp->qsmaskinitnext |= mask; rcu_read_lock() r0 = *X; r1 = *X; X = NULL; cookie = get_state_synchronize_rcu(); // cookie = 8 (future GP) rcu_seq_start(&rcu_state.gp_seq); // gp_seq = 5 // CPU1 now invisible to this GP! rcu_for_each_node_breadth_first() { rnp->qsmask = rnp->qsmaskinit; // CPU1 not included! } // GP completes without CPU1 rcu_seq_end(&rcu_state.gp_seq); // gp_seq = 8 poll_state_synchronize_rcu(cookie); // Returns true! kfree(r1); r2 = *r0; // USE-AFTER-FREE!}hjJ/sbah}(h]h ]h"]h$]h&]jUjVuh1jEhhhMhj5.hhubh)}(hoBy incrementing ``gp_seq`` first, CPU1's RCU read-side critical section is guaranteed to not be missed by CPU2.h](hBy incrementing }(hjX/hhhNhNubj)}(h ``gp_seq``h]hgp_seq}(hj`/hhhNhNubah}(h]h ]h"]h$]h&]uh1jhjX/ubhW first, CPU1’s RCU read-side critical section is guaranteed to not be missed by CPU2.}(hjX/hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhj5.hhubh)}(hhh](h)}(h5Concurrent Quiescent State Reporting for Offline CPUsh]h5Concurrent Quiescent State Reporting for Offline CPUs}(hj{/hhhNhNubah}(h]h ]h"]h$]h&]uh1hhjx/hhhhhMubh)}(hRCU must ensure that CPUs going offline report quiescent states to avoid blocking grace periods. This requires careful synchronization to handle race conditionsh]hRCU must ensure that CPUs going offline report quiescent states to avoid blocking grace periods. This requires careful synchronization to handle race conditions}(hj/hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhjx/hhubeh}(h]5concurrent-quiescent-state-reporting-for-offline-cpusah ]h"]5concurrent quiescent state reporting for offline cpusah$]h&]uh1hhj5.hhhhhMubh)}(hhh](h)}(h-Race condition causing Offline CPU to hang GPh]h-Race condition causing Offline CPU to hang GP}(hj/hhhNhNubah}(h]h ]h"]h$]h&]uh1hhj/hhhhhMubh)}(hA race between CPU offlining and new GP initialization (gp_init()) may occur because rcu_report_qs_rnp() in rcutree_report_cpu_dead() must temporarily release the ``rcu_node`` lock to wake the RCU grace-period kthread::h](hA race between CPU offlining and new GP initialization (gp_init()) may occur because rcu_report_qs_rnp() in rcutree_report_cpu_dead() must temporarily release the }(hj/hhhNhNubj)}(h ``rcu_node``h]hrcu_node}(hj/hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj/ubh+ lock to wake the RCU grace-period kthread:}(hj/hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhj/hhubjF)}(hXCPU1 (going offline) CPU0 (GP kthread) -------------------- ----------------- rcutree_report_cpu_dead() rcu_report_qs_rnp() // Must release rnp->lock to wake GP kthread raw_spin_unlock_irqrestore_rcu_node() // Wakes up and starts new GP rcu_gp_init() // First loop: copies qsmaskinitnext->qsmaskinit // CPU1 still in qsmaskinitnext! // Second loop: rnp->qsmask = rnp->qsmaskinit mask = rnp->qsmask & ~rnp->qsmaskinitnext // mask is 0! CPU1 still in both masks // Reacquire lock (but too late) rnp->qsmaskinitnext &= ~mask // Finally clears bith]hXCPU1 (going offline) CPU0 (GP kthread) -------------------- ----------------- rcutree_report_cpu_dead() rcu_report_qs_rnp() // Must release rnp->lock to wake GP kthread raw_spin_unlock_irqrestore_rcu_node() // Wakes up and starts new GP rcu_gp_init() // First loop: copies qsmaskinitnext->qsmaskinit // CPU1 still in qsmaskinitnext! // Second loop: rnp->qsmask = rnp->qsmaskinit mask = rnp->qsmask & ~rnp->qsmaskinitnext // mask is 0! CPU1 still in both masks // Reacquire lock (but too late) rnp->qsmaskinitnext &= ~mask // Finally clears bit}hj/sbah}(h]h ]h"]h$]h&]jUjVuh1jEhhhMhj/hhubh)}(hWithout ``ofl_lock``, the new grace period includes the offline CPU and waits forever for its quiescent state causing a GP hang.h](hWithout }(hj/hhhNhNubj)}(h ``ofl_lock``h]hofl_lock}(hj/hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj/ubhl, the new grace period includes the offline CPU and waits forever for its quiescent state causing a GP hang.}(hj/hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhj/hhubeh}(h]-race-condition-causing-offline-cpu-to-hang-gpah ]h"]-race condition causing offline cpu to hang gpah$]h&]uh1hhj5.hhhhhMubh)}(hhh](h)}(hA solution with ofl_lockh]hA solution with ofl_lock}(hj 0hhhNhNubah}(h]h ]h"]h$]h&]uh1hhj0hhhhhMubh)}(hThe ``ofl_lock`` (offline lock) prevents rcu_gp_init() from running during the vulnerable window when rcu_report_qs_rnp() has released ``rnp->lock``::h](hThe }(hj0hhhNhNubj)}(h ``ofl_lock``h]hofl_lock}(hj0hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj0ubhw (offline lock) prevents rcu_gp_init() from running during the vulnerable window when rcu_report_qs_rnp() has released }(hj0hhhNhNubj)}(h ``rnp->lock``h]h rnp->lock}(hj10hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj0ubh:}(hj0hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhj0hhubjF)}(hXCPU0 (rcu_gp_init) CPU1 (rcutree_report_cpu_dead) ------------------ ------------------------------ rcu_for_each_leaf_node(rnp) { arch_spin_lock(&ofl_lock) -----> arch_spin_lock(&ofl_lock) [BLOCKED] // Safe: CPU1 can't interfere rnp->qsmaskinit = rnp->qsmaskinitnext arch_spin_unlock(&ofl_lock) ---> // Now CPU1 can proceed } // But snapshot already takenh]hXCPU0 (rcu_gp_init) CPU1 (rcutree_report_cpu_dead) ------------------ ------------------------------ rcu_for_each_leaf_node(rnp) { arch_spin_lock(&ofl_lock) -----> arch_spin_lock(&ofl_lock) [BLOCKED] // Safe: CPU1 can't interfere rnp->qsmaskinit = rnp->qsmaskinitnext arch_spin_unlock(&ofl_lock) ---> // Now CPU1 can proceed } // But snapshot already taken}hjI0sbah}(h]h ]h"]h$]h&]jUjVuh1jEhhhMhj0hhubeh}(h]a-solution-with-ofl-lockah ]h"]a solution with ofl_lockah$]h&]uh1hhj5.hhhhhMubh)}(hhh](h)}(hRAnother race causing GP hangs in rcu_gpu_init(): Reporting QS for Now-offline CPUsh]hRAnother race causing GP hangs in rcu_gpu_init(): Reporting QS for Now-offline CPUs}(hjb0hhhNhNubah}(h]h ]h"]h$]h&]uh1hhj_0hhhhhMubh)}(hAfter the first loop takes an atomic snapshot of online CPUs, as shown above, the second loop in rcu_gp_init() detects CPUs that went offline between releasing ``ofl_lock`` and acquiring the per-node ``rnp->lock``. This detection is crucial because:h](hAfter the first loop takes an atomic snapshot of online CPUs, as shown above, the second loop in rcu_gp_init() detects CPUs that went offline between releasing }(hjp0hhhNhNubj)}(h ``ofl_lock``h]hofl_lock}(hjx0hhhNhNubah}(h]h ]h"]h$]h&]uh1jhjp0ubh and acquiring the per-node }(hjp0hhhNhNubj)}(h ``rnp->lock``h]h rnp->lock}(hj0hhhNhNubah}(h]h ]h"]h$]h&]uh1jhjp0ubh$. This detection is crucial because:}(hjp0hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhj_0hhubj)}(hhh](j)}(hMThe CPU might have gone offline after the snapshot but before the second looph]h)}(hj0h]hMThe CPU might have gone offline after the snapshot but before the second loop}(hj0hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj0ubah}(h]h ]h"]h$]h&]uh1jhj0hhhhhNubj)}(h=The offline CPU cannot report its own QS if it's already deadh]h)}(hj0h]h?The offline CPU cannot report its own QS if it’s already dead}(hj0hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj0ubah}(h]h ]h"]h$]h&]uh1jhj0hhhhhNubj)}(h[Without this detection, the grace period would wait forever for CPUs that are now offline. h]h)}(hZWithout this detection, the grace period would wait forever for CPUs that are now offline.h]hZWithout this detection, the grace period would wait forever for CPUs that are now offline.}(hj0hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj0ubah}(h]h ]h"]h$]h&]uh1jhj0hhhhhNubeh}(h]h ]h"]h$]h&]jjjhjjuh1jhj_0hhhhhMubh)}(h0The second loop performs this detection safely::h]h/The second loop performs this detection safely:}(hj0hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj_0hhubjF)}(hXXrcu_for_each_node_breadth_first(rnp) { raw_spin_lock_irqsave_rcu_node(rnp, flags); rnp->qsmask = rnp->qsmaskinit; // Apply the snapshot // Detect CPUs offline after snapshot mask = rnp->qsmask & ~rnp->qsmaskinitnext; if (mask && rcu_is_leaf_node(rnp)) rcu_report_qs_rnp(mask, ...) // Report QS for offline CPUs }h]hXXrcu_for_each_node_breadth_first(rnp) { raw_spin_lock_irqsave_rcu_node(rnp, flags); rnp->qsmask = rnp->qsmaskinit; // Apply the snapshot // Detect CPUs offline after snapshot mask = rnp->qsmask & ~rnp->qsmaskinitnext; if (mask && rcu_is_leaf_node(rnp)) rcu_report_qs_rnp(mask, ...) // Report QS for offline CPUs }}hj0sbah}(h]h ]h"]h$]h&]jUjVuh1jEhhhM hj_0hhubh)}(hXgThis approach ensures atomicity: quiescent state reporting for offline CPUs happens either in rcu_gp_init() (second loop) or in rcutree_report_cpu_dead(), never both and never neither. The ``rnp->lock`` held throughout the sequence prevents races - rcutree_report_cpu_dead() also acquires this lock when clearing ``qsmaskinitnext``, ensuring mutual exclusion.h](hThis approach ensures atomicity: quiescent state reporting for offline CPUs happens either in rcu_gp_init() (second loop) or in rcutree_report_cpu_dead(), never both and never neither. The }(hj 1hhhNhNubj)}(h ``rnp->lock``h]h rnp->lock}(hj1hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj 1ubho held throughout the sequence prevents races - rcutree_report_cpu_dead() also acquires this lock when clearing }(hj 1hhhNhNubj)}(h``qsmaskinitnext``h]hqsmaskinitnext}(hj'1hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj 1ubh, ensuring mutual exclusion.}(hj 1hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhM+hj_0hhubeh}(h]Oanother-race-causing-gp-hangs-in-rcu-gpu-init-reporting-qs-for-now-offline-cpusah ]h"]Ranother race causing gp hangs in rcu_gpu_init(): reporting qs for now-offline cpusah$]h&]uh1hhj5.hhhhhMubeh}(h]j(ah ]h"] hotplug cpuah$]h&]uh1hhj'hhhhhMjKubh)}(hhh](h)}(hScheduler and RCUh]hScheduler and RCU}(hjQ1hhhNhNubah}(h]h ]h"]h$]h&]uh1hhjN1hhhhhM2ubh)}(hXRCU makes use of kthreads, and it is necessary to avoid excessive CPU-time accumulation by these kthreads. This requirement was no surprise, but RCU's violation of it when running context-switch-heavy workloads when built with ``CONFIG_NO_HZ_FULL=y`` `did come as a surprise [PDF] `__. RCU has made good progress towards meeting this requirement, even for context-switch-heavy ``CONFIG_NO_HZ_FULL=y`` workloads, but there is room for further improvement.h](hRCU makes use of kthreads, and it is necessary to avoid excessive CPU-time accumulation by these kthreads. This requirement was no surprise, but RCU’s violation of it when running context-switch-heavy workloads when built with }(hj_1hhhNhNubj)}(h``CONFIG_NO_HZ_FULL=y``h]hCONFIG_NO_HZ_FULL=y}(hjg1hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj_1ubh }(hj_1hhhNhNubh)}(hq`did come as a surprise [PDF] `__h]hdid come as a surprise [PDF]}(hjy1hhhNhNubah}(h]h ]h"]h$]h&]namedid come as a surprise [PDF]jNhttp://www.rdrop.com/users/paulmck/scalability/paper/BareMetal.2015.01.15b.pdfuh1hhj_1ubh]. RCU has made good progress towards meeting this requirement, even for context-switch-heavy }(hj_1hhhNhNubj)}(h``CONFIG_NO_HZ_FULL=y``h]hCONFIG_NO_HZ_FULL=y}(hj1hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj_1ubh6 workloads, but there is room for further improvement.}(hj_1hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhM4hjN1hhubh)}(hXThere is no longer any prohibition against holding any of scheduler's runqueue or priority-inheritance spinlocks across an rcu_read_unlock(), even if interrupts and preemption were enabled somewhere within the corresponding RCU read-side critical section. Therefore, it is now perfectly legal to execute rcu_read_lock() with preemption enabled, acquire one of the scheduler locks, and hold that lock across the matching rcu_read_unlock().h]hXThere is no longer any prohibition against holding any of scheduler’s runqueue or priority-inheritance spinlocks across an rcu_read_unlock(), even if interrupts and preemption were enabled somewhere within the corresponding RCU read-side critical section. Therefore, it is now perfectly legal to execute rcu_read_lock() with preemption enabled, acquire one of the scheduler locks, and hold that lock across the matching rcu_read_unlock().}(hj1hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM=hjN1hhubh)}(hXSimilarly, the RCU flavor consolidation has removed the need for negative nesting. The fact that interrupt-disabled regions of code act as RCU read-side critical sections implicitly avoids earlier issues that used to result in destructive recursion via interrupt handler's use of RCU.h]hXSimilarly, the RCU flavor consolidation has removed the need for negative nesting. The fact that interrupt-disabled regions of code act as RCU read-side critical sections implicitly avoids earlier issues that used to result in destructive recursion via interrupt handler’s use of RCU.}(hj1hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMEhjN1hhubeh}(h]j(ah ]h"]scheduler and rcuah$]h&]uh1hhj'hhhhhM2jKubh)}(hhh](h)}(hTracing and RCUh]hTracing and RCU}(hj1hhhNhNubah}(h]h ]h"]h$]h&]uh1hhj1hhhhhMKubh)}(hXIt is possible to use tracing on RCU code, but tracing itself uses RCU. For this reason, rcu_dereference_raw_check() is provided for use by tracing, which avoids the destructive recursion that could otherwise ensue. This API is also used by virtualization in some architectures, where RCU readers execute in environments in which tracing cannot be used. The tracing folks both located the requirement and provided the needed fix, so this surprise requirement was relatively painless.h]hXIt is possible to use tracing on RCU code, but tracing itself uses RCU. For this reason, rcu_dereference_raw_check() is provided for use by tracing, which avoids the destructive recursion that could otherwise ensue. This API is also used by virtualization in some architectures, where RCU readers execute in environments in which tracing cannot be used. The tracing folks both located the requirement and provided the needed fix, so this surprise requirement was relatively painless.}(hj1hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMMhj1hhubeh}(h]j)ah ]h"]tracing and rcuah$]h&]uh1hhj'hhhhhMKjKubh)}(hhh](h)}(hAccesses to User Memory and RCUh]hAccesses to User Memory and RCU}(hj1hhhNhNubah}(h]h ]h"]h$]h&]uh1hhj1hhhhhMVubh)}(hThe kernel needs to access user-space memory, for example, to access data referenced by system-call parameters. The get_user() macro does this job.h]hThe kernel needs to access user-space memory, for example, to access data referenced by system-call parameters. The get_user() macro does this job.}(hj2hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMXhj1hhubh)}(hXHowever, user-space memory might well be paged out, which means that get_user() might well page-fault and thus block while waiting for the resulting I/O to complete. It would be a very bad thing for the compiler to reorder a get_user() invocation into an RCU read-side critical section.h]hXHowever, user-space memory might well be paged out, which means that get_user() might well page-fault and thus block while waiting for the resulting I/O to complete. It would be a very bad thing for the compiler to reorder a get_user() invocation into an RCU read-side critical section.}(hj2hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM[hj1hhubh)}(h;For example, suppose that the source code looked like this:h]h;For example, suppose that the source code looked like this:}(hj2hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM`hj1hhubj@)}(h:: 1 rcu_read_lock(); 2 p = rcu_dereference(gp); 3 v = p->value; 4 rcu_read_unlock(); 5 get_user(user_v, user_p); 6 do_something_with(v, user_v); h]jF)}(h1 rcu_read_lock(); 2 p = rcu_dereference(gp); 3 v = p->value; 4 rcu_read_unlock(); 5 get_user(user_v, user_p); 6 do_something_with(v, user_v);h]h1 rcu_read_lock(); 2 p = rcu_dereference(gp); 3 v = p->value; 4 rcu_read_unlock(); 5 get_user(user_v, user_p); 6 do_something_with(v, user_v);}hj.2sbah}(h]h ]h"]h$]h&]jUjVuh1jEhhhMdhj*2ubah}(h]h ]h"]h$]h&]uh1j?hhhMbhj1hhubh)}(hTThe compiler must not be permitted to transform this source code into the following:h]hTThe compiler must not be permitted to transform this source code into the following:}(hjB2hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMkhj1hhubj@)}(h:: 1 rcu_read_lock(); 2 p = rcu_dereference(gp); 3 get_user(user_v, user_p); // BUG: POSSIBLE PAGE FAULT!!! 4 v = p->value; 5 rcu_read_unlock(); 6 do_something_with(v, user_v); h]jF)}(h1 rcu_read_lock(); 2 p = rcu_dereference(gp); 3 get_user(user_v, user_p); // BUG: POSSIBLE PAGE FAULT!!! 4 v = p->value; 5 rcu_read_unlock(); 6 do_something_with(v, user_v);h]h1 rcu_read_lock(); 2 p = rcu_dereference(gp); 3 get_user(user_v, user_p); // BUG: POSSIBLE PAGE FAULT!!! 4 v = p->value; 5 rcu_read_unlock(); 6 do_something_with(v, user_v);}hjT2sbah}(h]h ]h"]h$]h&]jUjVuh1jEhhhMphjP2ubah}(h]h ]h"]h$]h&]uh1j?hhhMnhj1hhubh)}(hXIf the compiler did make this transformation in a ``CONFIG_PREEMPTION=n`` kernel build, and if get_user() did page fault, the result would be a quiescent state in the middle of an RCU read-side critical section. This misplaced quiescent state could result in line 4 being a use-after-free access, which could be bad for your kernel's actuarial statistics. Similar examples can be constructed with the call to get_user() preceding the rcu_read_lock().h](h2If the compiler did make this transformation in a }(hjh2hhhNhNubj)}(h``CONFIG_PREEMPTION=n``h]hCONFIG_PREEMPTION=n}(hjp2hhhNhNubah}(h]h ]h"]h$]h&]uh1jhjh2ubhX} kernel build, and if get_user() did page fault, the result would be a quiescent state in the middle of an RCU read-side critical section. This misplaced quiescent state could result in line 4 being a use-after-free access, which could be bad for your kernel’s actuarial statistics. Similar examples can be constructed with the call to get_user() preceding the rcu_read_lock().}(hjh2hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMwhj1hhubh)}(hXEUnfortunately, get_user() doesn't have any particular ordering properties, and in some architectures the underlying ``asm`` isn't even marked ``volatile``. And even if it was marked ``volatile``, the above access to ``p->value`` is not volatile, so the compiler would not have any reason to keep those two accesses in order.h](hvUnfortunately, get_user() doesn’t have any particular ordering properties, and in some architectures the underlying }(hj2hhhNhNubj)}(h``asm``h]hasm}(hj2hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj2ubh isn’t even marked }(hj2hhhNhNubj)}(h ``volatile``h]hvolatile}(hj2hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj2ubh. And even if it was marked }(hj2hhhNhNubj)}(h ``volatile``h]hvolatile}(hj2hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj2ubh, the above access to }(hj2hhhNhNubj)}(h ``p->value``h]hp->value}(hj2hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj2ubh` is not volatile, so the compiler would not have any reason to keep those two accesses in order.}(hj2hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhj1hhubh)}(hTherefore, the Linux-kernel definitions of rcu_read_lock() and rcu_read_unlock() must act as compiler barriers, at least for outermost instances of rcu_read_lock() and rcu_read_unlock() within a nested set of RCU read-side critical sections.h]hTherefore, the Linux-kernel definitions of rcu_read_lock() and rcu_read_unlock() must act as compiler barriers, at least for outermost instances of rcu_read_lock() and rcu_read_unlock() within a nested set of RCU read-side critical sections.}(hj2hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj1hhubeh}(h]j5)ah ]h"]accesses to user memory and rcuah$]h&]uh1hhj'hhhhhMVjKubh)}(hhh](h)}(hEnergy Efficiencyh]hEnergy Efficiency}(hj2hhhNhNubah}(h]h ]h"]h$]h&]uh1hhj2hhhhhMubh)}(hX_Interrupting idle CPUs is considered socially unacceptable, especially by people with battery-powered embedded systems. RCU therefore conserves energy by detecting which CPUs are idle, including tracking CPUs that have been interrupted from idle. This is a large part of the energy-efficiency requirement, so I learned of this via an irate phone call.h]hX_Interrupting idle CPUs is considered socially unacceptable, especially by people with battery-powered embedded systems. RCU therefore conserves energy by detecting which CPUs are idle, including tracking CPUs that have been interrupted from idle. This is a large part of the energy-efficiency requirement, so I learned of this via an irate phone call.}(hj3hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj2hhubh)}(hBecause RCU avoids interrupting idle CPUs, it is illegal to execute an RCU read-side critical section on an idle CPU. (Kernels built with ``CONFIG_PROVE_RCU=y`` will splat if you try it.)h](hBecause RCU avoids interrupting idle CPUs, it is illegal to execute an RCU read-side critical section on an idle CPU. (Kernels built with }(hj3hhhNhNubj)}(h``CONFIG_PROVE_RCU=y``h]hCONFIG_PROVE_RCU=y}(hj3hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj3ubh will splat if you try it.)}(hj3hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhj2hhubh)}(hXEIt is similarly socially unacceptable to interrupt an ``nohz_full`` CPU running in userspace. RCU must therefore track ``nohz_full`` userspace execution. RCU must therefore be able to sample state at two points in time, and be able to determine whether or not some other CPU spent any time idle and/or executing in userspace.h](h6It is similarly socially unacceptable to interrupt an }(hj23hhhNhNubj)}(h ``nohz_full``h]h nohz_full}(hj:3hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj23ubh4 CPU running in userspace. RCU must therefore track }(hj23hhhNhNubj)}(h ``nohz_full``h]h nohz_full}(hjL3hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj23ubh userspace execution. RCU must therefore be able to sample state at two points in time, and be able to determine whether or not some other CPU spent any time idle and/or executing in userspace.}(hj23hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhj2hhubh)}(hXIThese energy-efficiency requirements have proven quite difficult to understand and to meet, for example, there have been more than five clean-sheet rewrites of RCU's energy-efficiency code, the last of which was finally able to demonstrate `real energy savings running on real hardware [PDF] `__. As noted earlier, I learned of many of these requirements via angry phone calls: Flaming me on the Linux-kernel mailing list was apparently not sufficient to fully vent their ire at RCU's energy-efficiency bugs!h](hThese energy-efficiency requirements have proven quite difficult to understand and to meet, for example, there have been more than five clean-sheet rewrites of RCU’s energy-efficiency code, the last of which was finally able to demonstrate }(hjd3hhhNhNubh)}(h`real energy savings running on real hardware [PDF] `__h]h2real energy savings running on real hardware [PDF]}(hjl3hhhNhNubah}(h]h ]h"]h$]h&]name2real energy savings running on real hardware [PDF]jKhttp://www.rdrop.com/users/paulmck/realtime/paper/AMPenergy.2013.04.19a.pdfuh1hhjd3ubh. As noted earlier, I learned of many of these requirements via angry phone calls: Flaming me on the Linux-kernel mailing list was apparently not sufficient to fully vent their ire at RCU’s energy-efficiency bugs!}(hjd3hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhj2hhubeh}(h]jX)ah ]h"]energy efficiencyah$]h&]uh1hhj'hhhhhMjKubh)}(hhh](h)}(h#Scheduling-Clock Interrupts and RCUh]h#Scheduling-Clock Interrupts and RCU}(hj3hhhNhNubah}(h]h ]h"]h$]h&]uh1hhj3hhhhhMubh)}(hThe kernel transitions between in-kernel non-idle execution, userspace execution, and the idle loop. Depending on kernel configuration, RCU handles these states differently:h]hThe kernel transitions between in-kernel non-idle execution, userspace execution, and the idle loop. Depending on kernel configuration, RCU handles these states differently:}(hj3hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj3hhubj)}(hhh]j)}(hhh](j)}(hhh]h}(h]h ]h"]h$]h&]colwidthKuh1jhj3ubj)}(hhh]h}(h]h ]h"]h$]h&]colwidthKuh1jhj3ubj)}(hhh]h}(h]h ]h"]h$]h&]colwidthKuh1jhj3ubj)}(hhh]h}(h]h ]h"]h$]h&]colwidthKuh1jhj3ubhthead)}(hhh]j)}(hhh](j )}(hhh]h)}(h``HZ`` Kconfigh](j)}(h``HZ``h]hHZ}(hj3hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj3ubh Kconfig}(hj3hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhj3ubah}(h]h ]h"]h$]h&]uh1j hj3ubj )}(hhh]h)}(h In-Kernelh]h In-Kernel}(hj 4hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj4ubah}(h]h ]h"]h$]h&]uh1j hj3ubj )}(hhh]h)}(hUsermodeh]hUsermode}(hj"4hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj4ubah}(h]h ]h"]h$]h&]uh1j hj3ubj )}(hhh]h)}(hIdleh]hIdle}(hj94hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj64ubah}(h]h ]h"]h$]h&]uh1j hj3ubeh}(h]h ]h"]h$]h&]uh1jhj3ubah}(h]h ]h"]h$]h&]uh1j3hj3ubj)}(hhh](j)}(hhh](j )}(hhh]h)}(h``HZ_PERIODIC``h]j)}(hjd4h]h HZ_PERIODIC}(hjf4hhhNhNubah}(h]h ]h"]h$]h&]uh1jhjb4ubah}(h]h ]h"]h$]h&]uh1hhhhMhj_4ubah}(h]h ]h"]h$]h&]uh1j hj\4ubj )}(hhh]h)}(h'Can rely on scheduling-clock interrupt.h]h'Can rely on scheduling-clock interrupt.}(hj4hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj4ubah}(h]h ]h"]h$]h&]uh1j hj\4ubj )}(hhh]h)}(hTCan rely on scheduling-clock interrupt and its detection of interrupt from usermode.h]hTCan rely on scheduling-clock interrupt and its detection of interrupt from usermode.}(hj4hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj4ubah}(h]h ]h"]h$]h&]uh1j hj\4ubj )}(hhh]h)}(h)Can rely on RCU's dyntick-idle detection.h]h+Can rely on RCU’s dyntick-idle detection.}(hj4hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj4ubah}(h]h ]h"]h$]h&]uh1j hj\4ubeh}(h]h ]h"]h$]h&]uh1jhjY4ubj)}(hhh](j )}(hhh]h)}(h``NO_HZ_IDLE``h]j)}(hj4h]h NO_HZ_IDLE}(hj4hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj4ubah}(h]h ]h"]h$]h&]uh1hhhhMhj4ubah}(h]h ]h"]h$]h&]uh1j hj4ubj )}(hhh]h)}(h'Can rely on scheduling-clock interrupt.h]h'Can rely on scheduling-clock interrupt.}(hj4hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj4ubah}(h]h ]h"]h$]h&]uh1j hj4ubj )}(hhh]h)}(hTCan rely on scheduling-clock interrupt and its detection of interrupt from usermode.h]hTCan rely on scheduling-clock interrupt and its detection of interrupt from usermode.}(hj5hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj5ubah}(h]h ]h"]h$]h&]uh1j hj4ubj )}(hhh]h)}(h)Can rely on RCU's dyntick-idle detection.h]h+Can rely on RCU’s dyntick-idle detection.}(hj5hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj5ubah}(h]h ]h"]h$]h&]uh1j hj4ubeh}(h]h ]h"]h$]h&]uh1jhjY4ubj)}(hhh](j )}(hhh]h)}(h``NO_HZ_FULL``h]j)}(hj@5h]h NO_HZ_FULL}(hjB5hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj>5ubah}(h]h ]h"]h$]h&]uh1hhhhMhj;5ubah}(h]h ]h"]h$]h&]uh1j hj85ubj )}(hhh]h)}(hCan only sometimes rely on scheduling-clock interrupt. In other cases, it is necessary to bound kernel execution times and/or use IPIs.h]hCan only sometimes rely on scheduling-clock interrupt. In other cases, it is necessary to bound kernel execution times and/or use IPIs.}(hj^5hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj[5ubah}(h]h ]h"]h$]h&]uh1j hj85ubj )}(hhh]h)}(h)Can rely on RCU's dyntick-idle detection.h]h+Can rely on RCU’s dyntick-idle detection.}(hju5hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhjr5ubah}(h]h ]h"]h$]h&]uh1j hj85ubj )}(hhh]h)}(h)Can rely on RCU's dyntick-idle detection.h]h+Can rely on RCU’s dyntick-idle detection.}(hj5hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj5ubah}(h]h ]h"]h$]h&]uh1j hj85ubeh}(h]h ]h"]h$]h&]uh1jhjY4ubeh}(h]h ]h"]h$]h&]uh1jhj3ubeh}(h]h ]h"]h$]h&]colsKuh1jhj3ubah}(h]h ]h"]h$]h&]uh1jhj3hhhhhNubj)}(hhh]j)}(hhh](j)}(hhh]h}(h]h ]h"]h$]h&]colwidthKGuh1jhj5ubj)}(hhh](j)}(hhh]j )}(hhh]h)}(h**Quick Quiz**:h](j)}(h**Quick Quiz**h]h Quick Quiz}(hj5hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj5ubh:}(hj5hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhj5ubah}(h]h ]h"]h$]h&]uh1j hj5ubah}(h]h ]h"]h$]h&]uh1jhj5ubj)}(hhh]j )}(hhh]h)}(hWhy can't ``NO_HZ_FULL`` in-kernel execution rely on the scheduling-clock interrupt, just like ``HZ_PERIODIC`` and ``NO_HZ_IDLE`` do?h](h Why can’t }(hj6hhhNhNubj)}(h``NO_HZ_FULL``h]h NO_HZ_FULL}(hj6hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj6ubhG in-kernel execution rely on the scheduling-clock interrupt, just like }(hj6hhhNhNubj)}(h``HZ_PERIODIC``h]h HZ_PERIODIC}(hj6hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj6ubh and }(hj6hhhNhNubj)}(h``NO_HZ_IDLE``h]h NO_HZ_IDLE}(hj,6hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj6ubh do?}(hj6hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhj5ubah}(h]h ]h"]h$]h&]uh1j hj5ubah}(h]h ]h"]h$]h&]uh1jhj5ubj)}(hhh]j )}(hhh]h)}(h **Answer**:h](j)}(h **Answer**h]hAnswer}(hjZ6hhhNhNubah}(h]h ]h"]h$]h&]uh1jhjV6ubh:}(hjV6hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhjS6ubah}(h]h ]h"]h$]h&]uh1j hjP6ubah}(h]h ]h"]h$]h&]uh1jhj5ubj)}(hhh]j )}(hhh]h)}(hBecause, as a performance optimization, ``NO_HZ_FULL`` does not necessarily re-enable the scheduling-clock interrupt on entry to each and every system call.h](h(Because, as a performance optimization, }(hj6hhhNhNubj)}(h``NO_HZ_FULL``h]h NO_HZ_FULL}(hj6hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj6ubhf does not necessarily re-enable the scheduling-clock interrupt on entry to each and every system call.}(hj6hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhj6ubah}(h]h ]h"]h$]h&]uh1j hj~6ubah}(h]h ]h"]h$]h&]uh1jhj5ubeh}(h]h ]h"]h$]h&]uh1jhj5ubeh}(h]h ]h"]h$]h&]colsKuh1jhj5ubah}(h]h ]h"]h$]h&]uh1jhj3hhhhhNubh)}(hX2However, RCU must be reliably informed as to whether any given CPU is currently in the idle loop, and, for ``NO_HZ_FULL``, also whether that CPU is executing in usermode, as discussed `earlier `__. It also requires that the scheduling-clock interrupt be enabled when RCU needs it to be:h](hkHowever, RCU must be reliably informed as to whether any given CPU is currently in the idle loop, and, for }(hj6hhhNhNubj)}(h``NO_HZ_FULL``h]h NO_HZ_FULL}(hj6hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj6ubh?, also whether that CPU is executing in usermode, as discussed }(hj6hhhNhNubh)}(h `earlier `__h]hearlier}(hj6hhhNhNubah}(h]h ]h"]h$]h&]nameearlierjjX)uh1hhj6jKubhZ. It also requires that the scheduling-clock interrupt be enabled when RCU needs it to be:}(hj6hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhj3hhubj)}(hhh](j)}(hXIf a CPU is either idle or executing in usermode, and RCU believes it is non-idle, the scheduling-clock tick had better be running. Otherwise, you will get RCU CPU stall warnings. Or at best, very long (11-second) grace periods, with a pointless IPI waking the CPU from time to time.h]h)}(hXIf a CPU is either idle or executing in usermode, and RCU believes it is non-idle, the scheduling-clock tick had better be running. Otherwise, you will get RCU CPU stall warnings. Or at best, very long (11-second) grace periods, with a pointless IPI waking the CPU from time to time.h]hXIf a CPU is either idle or executing in usermode, and RCU believes it is non-idle, the scheduling-clock tick had better be running. Otherwise, you will get RCU CPU stall warnings. Or at best, very long (11-second) grace periods, with a pointless IPI waking the CPU from time to time.}(hj6hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj6ubah}(h]h ]h"]h$]h&]uh1jhj6hhhhhNubj)}(hXIf a CPU is in a portion of the kernel that executes RCU read-side critical sections, and RCU believes this CPU to be idle, you will get random memory corruption. **DON'T DO THIS!!!** This is one reason to test with lockdep, which will complain about this sort of thing.h]h)}(hXIf a CPU is in a portion of the kernel that executes RCU read-side critical sections, and RCU believes this CPU to be idle, you will get random memory corruption. **DON'T DO THIS!!!** This is one reason to test with lockdep, which will complain about this sort of thing.h](hIf a CPU is in a portion of the kernel that executes RCU read-side critical sections, and RCU believes this CPU to be idle, you will get random memory corruption. }(hj7hhhNhNubj)}(h**DON'T DO THIS!!!**h]hDON’T DO THIS!!!}(hj7hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj7ubhW This is one reason to test with lockdep, which will complain about this sort of thing.}(hj7hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhj7ubah}(h]h ]h"]h$]h&]uh1jhj6hhhhhNubj)}(hXIf a CPU is in a portion of the kernel that is absolutely positively no-joking guaranteed to never execute any RCU read-side critical sections, and RCU believes this CPU to be idle, no problem. This sort of thing is used by some architectures for light-weight exception handlers, which can then avoid the overhead of ct_irq_enter() and ct_irq_exit() at exception entry and exit, respectively. Some go further and avoid the entireties of irq_enter() and irq_exit(). Just make very sure you are running some of your tests with ``CONFIG_PROVE_RCU=y``, just in case one of your code paths was in fact joking about not doing RCU read-side critical sections.h]h)}(hXIf a CPU is in a portion of the kernel that is absolutely positively no-joking guaranteed to never execute any RCU read-side critical sections, and RCU believes this CPU to be idle, no problem. This sort of thing is used by some architectures for light-weight exception handlers, which can then avoid the overhead of ct_irq_enter() and ct_irq_exit() at exception entry and exit, respectively. Some go further and avoid the entireties of irq_enter() and irq_exit(). Just make very sure you are running some of your tests with ``CONFIG_PROVE_RCU=y``, just in case one of your code paths was in fact joking about not doing RCU read-side critical sections.h](hX If a CPU is in a portion of the kernel that is absolutely positively no-joking guaranteed to never execute any RCU read-side critical sections, and RCU believes this CPU to be idle, no problem. This sort of thing is used by some architectures for light-weight exception handlers, which can then avoid the overhead of ct_irq_enter() and ct_irq_exit() at exception entry and exit, respectively. Some go further and avoid the entireties of irq_enter() and irq_exit(). Just make very sure you are running some of your tests with }(hj@7hhhNhNubj)}(h``CONFIG_PROVE_RCU=y``h]hCONFIG_PROVE_RCU=y}(hjH7hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj@7ubhi, just in case one of your code paths was in fact joking about not doing RCU read-side critical sections.}(hj@7hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMhj<7ubah}(h]h ]h"]h$]h&]uh1jhj6hhhhhNubj)}(hXiIf a CPU is executing in the kernel with the scheduling-clock interrupt disabled and RCU believes this CPU to be non-idle, and if the CPU goes idle (from an RCU perspective) every few jiffies, no problem. It is usually OK for there to be the occasional gap between idle periods of up to a second or so. If the gap grows too long, you get RCU CPU stall warnings.h]h)}(hXiIf a CPU is executing in the kernel with the scheduling-clock interrupt disabled and RCU believes this CPU to be non-idle, and if the CPU goes idle (from an RCU perspective) every few jiffies, no problem. It is usually OK for there to be the occasional gap between idle periods of up to a second or so. If the gap grows too long, you get RCU CPU stall warnings.h]hXiIf a CPU is executing in the kernel with the scheduling-clock interrupt disabled and RCU believes this CPU to be non-idle, and if the CPU goes idle (from an RCU perspective) every few jiffies, no problem. It is usually OK for there to be the occasional gap between idle periods of up to a second or so. If the gap grows too long, you get RCU CPU stall warnings.}(hjj7hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhjf7ubah}(h]h ]h"]h$]h&]uh1jhj6hhhhhNubj)}(hgIf a CPU is either idle or executing in usermode, and RCU believes it to be idle, of course no problem.h]h)}(hgIf a CPU is either idle or executing in usermode, and RCU believes it to be idle, of course no problem.h]hgIf a CPU is either idle or executing in usermode, and RCU believes it to be idle, of course no problem.}(hj7hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj~7ubah}(h]h ]h"]h$]h&]uh1jhj6hhhhhNubj)}(hXIf a CPU is executing in the kernel, the kernel code path is passing through quiescent states at a reasonable frequency (preferably about once per few jiffies, but the occasional excursion to a second or so is usually OK) and the scheduling-clock interrupt is enabled, of course no problem. If the gap between a successive pair of quiescent states grows too long, you get RCU CPU stall warnings. h]h)}(hXIf a CPU is executing in the kernel, the kernel code path is passing through quiescent states at a reasonable frequency (preferably about once per few jiffies, but the occasional excursion to a second or so is usually OK) and the scheduling-clock interrupt is enabled, of course no problem. If the gap between a successive pair of quiescent states grows too long, you get RCU CPU stall warnings.h]hXIf a CPU is executing in the kernel, the kernel code path is passing through quiescent states at a reasonable frequency (preferably about once per few jiffies, but the occasional excursion to a second or so is usually OK) and the scheduling-clock interrupt is enabled, of course no problem. If the gap between a successive pair of quiescent states grows too long, you get RCU CPU stall warnings.}(hj7hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMhj7ubah}(h]h ]h"]h$]h&]uh1jhj6hhhhhNubeh}(h]h ]h"]h$]h&]jjjhjjuh1jhj3hhhhhMubj)}(hhh]j)}(hhh](j)}(hhh]h}(h]h ]h"]h$]h&]colwidthKGuh1jhj7ubj)}(hhh](j)}(hhh]j )}(hhh]h)}(h**Quick Quiz**:h](j)}(h**Quick Quiz**h]h Quick Quiz}(hj7hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj7ubh:}(hj7hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhM hj7ubah}(h]h ]h"]h$]h&]uh1j hj7ubah}(h]h ]h"]h$]h&]uh1jhj7ubj)}(hhh]j )}(hhh]h)}(hBut what if my driver has a hardware interrupt handler that can run for many seconds? I cannot invoke schedule() from an hardware interrupt handler, after all!h]hBut what if my driver has a hardware interrupt handler that can run for many seconds? I cannot invoke schedule() from an hardware interrupt handler, after all!}(hj7hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM hj7ubah}(h]h ]h"]h$]h&]uh1j hj7ubah}(h]h ]h"]h$]h&]uh1jhj7ubj)}(hhh]j )}(hhh]h)}(h **Answer**:h](j)}(h **Answer**h]hAnswer}(hj8hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj8ubh:}(hj8hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhM hj8ubah}(h]h ]h"]h$]h&]uh1j hj8ubah}(h]h ]h"]h$]h&]uh1jhj7ubj)}(hhh]j )}(hhh]h)}(hXOne approach is to do ``ct_irq_exit();ct_irq_enter();`` every so often. But given that long-running interrupt handlers can cause other problems, not least for response time, shouldn't you work to keep your interrupt handler's runtime within reasonable bounds?h](hOne approach is to do }(hjI8hhhNhNubj)}(h!``ct_irq_exit();ct_irq_enter();``h]hct_irq_exit();ct_irq_enter();}(hjQ8hhhNhNubah}(h]h ]h"]h$]h&]uh1jhjI8ubh every so often. But given that long-running interrupt handlers can cause other problems, not least for response time, shouldn’t you work to keep your interrupt handler’s runtime within reasonable bounds?}(hjI8hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhM hjF8ubah}(h]h ]h"]h$]h&]uh1j hjC8ubah}(h]h ]h"]h$]h&]uh1jhj7ubeh}(h]h ]h"]h$]h&]uh1jhj7ubeh}(h]h ]h"]h$]h&]colsKuh1jhj7ubah}(h]h ]h"]h$]h&]uh1jhj3hhhhhNubh)}(hXEBut as long as RCU is properly informed of kernel state transitions between in-kernel execution, usermode execution, and idle, and as long as the scheduling-clock interrupt is enabled when RCU needs it to be, you can rest assured that the bugs you encounter will be in some other part of RCU or some other part of the kernel!h]hXEBut as long as RCU is properly informed of kernel state transitions between in-kernel execution, usermode execution, and idle, and as long as the scheduling-clock interrupt is enabled when RCU needs it to be, you can rest assured that the bugs you encounter will be in some other part of RCU or some other part of the kernel!}(hj8hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM hj3hhubeh}(h]j{)ah ]h"]#scheduling-clock interrupts and rcuah$]h&]uh1hhj'hhhhhMjKubh)}(hhh](h)}(hMemory Efficiencyh]hMemory Efficiency}(hj8hhhNhNubah}(h]h ]h"]h$]h&]uh1hhj8hhhhhM ubh)}(hXAlthough small-memory non-realtime systems can simply use Tiny RCU, code size is only one aspect of memory efficiency. Another aspect is the size of the ``rcu_head`` structure used by call_rcu() and kfree_rcu(). Although this structure contains nothing more than a pair of pointers, it does appear in many RCU-protected data structures, including some that are size critical. The ``page`` structure is a case in point, as evidenced by the many occurrences of the ``union`` keyword within that structure.h](hAlthough small-memory non-realtime systems can simply use Tiny RCU, code size is only one aspect of memory efficiency. Another aspect is the size of the }(hj8hhhNhNubj)}(h ``rcu_head``h]hrcu_head}(hj8hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj8ubh structure used by call_rcu() and kfree_rcu(). Although this structure contains nothing more than a pair of pointers, it does appear in many RCU-protected data structures, including some that are size critical. The }(hj8hhhNhNubj)}(h``page``h]hpage}(hj8hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj8ubhK structure is a case in point, as evidenced by the many occurrences of the }(hj8hhhNhNubj)}(h ``union``h]hunion}(hj8hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj8ubh keyword within that structure.}(hj8hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhM hj8hhubh)}(hX This need for memory efficiency is one reason that RCU uses hand-crafted singly linked lists to track the ``rcu_head`` structures that are waiting for a grace period to elapse. It is also the reason why ``rcu_head`` structures do not contain debug information, such as fields tracking the file and line of the call_rcu() or kfree_rcu() that posted them. Although this information might appear in debug-only kernel builds at some point, in the meantime, the ``->func`` field will often provide the needed debug information.h](hjThis need for memory efficiency is one reason that RCU uses hand-crafted singly linked lists to track the }(hj8hhhNhNubj)}(h ``rcu_head``h]hrcu_head}(hj8hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj8ubhU structures that are waiting for a grace period to elapse. It is also the reason why }(hj8hhhNhNubj)}(h ``rcu_head``h]hrcu_head}(hj 9hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj8ubh structures do not contain debug information, such as fields tracking the file and line of the call_rcu() or kfree_rcu() that posted them. Although this information might appear in debug-only kernel builds at some point, in the meantime, the }(hj8hhhNhNubj)}(h ``->func``h]h->func}(hj9hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj8ubh7 field will often provide the needed debug information.}(hj8hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhM' hj8hhubh)}(hXaHowever, in some cases, the need for memory efficiency leads to even more extreme measures. Returning to the ``page`` structure, the ``rcu_head`` field shares storage with a great many other structures that are used at various points in the corresponding page's lifetime. In order to correctly resolve certain `race conditions `__, the Linux kernel's memory-management subsystem needs a particular bit to remain zero during all phases of grace-period processing, and that bit happens to map to the bottom bit of the ``rcu_head`` structure's ``->next`` field. RCU makes this guarantee as long as call_rcu() is used to post the callback, as opposed to kfree_rcu() or some future “lazy” variant of call_rcu() that might one day be created for energy-efficiency purposes.h](hmHowever, in some cases, the need for memory efficiency leads to even more extreme measures. Returning to the }(hj69hhhNhNubj)}(h``page``h]hpage}(hj>9hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj69ubh structure, the }(hj69hhhNhNubj)}(h ``rcu_head``h]hrcu_head}(hjP9hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj69ubh field shares storage with a great many other structures that are used at various points in the corresponding page’s lifetime. In order to correctly resolve certain }(hj69hhhNhNubh)}(hr`race conditions `__h]hrace conditions}(hjb9hhhNhNubah}(h]h ]h"]h$]h&]namerace conditionsj\https://lore.kernel.org/r/1439976106-137226-1-git-send-email-kirill.shutemov@linux.intel.comuh1hhj69ubh, the Linux kernel’s memory-management subsystem needs a particular bit to remain zero during all phases of grace-period processing, and that bit happens to map to the bottom bit of the }(hj69hhhNhNubj)}(h ``rcu_head``h]hrcu_head}(hjw9hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj69ubh structure’s }(hj69hhhNhNubj)}(h ``->next``h]h->next}(hj9hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj69ubh field. RCU makes this guarantee as long as call_rcu() is used to post the callback, as opposed to kfree_rcu() or some future “lazy” variant of call_rcu() that might one day be created for energy-efficiency purposes.}(hj69hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhM0 hj8hhubh)}(hXThat said, there are limits. RCU requires that the ``rcu_head`` structure be aligned to a two-byte boundary, and passing a misaligned ``rcu_head`` structure to one of the call_rcu() family of functions will result in a splat. It is therefore necessary to exercise caution when packing structures containing fields of type ``rcu_head``. Why not a four-byte or even eight-byte alignment requirement? Because the m68k architecture provides only two-byte alignment, and thus acts as alignment's least common denominator.h](h3That said, there are limits. RCU requires that the }(hj9hhhNhNubj)}(h ``rcu_head``h]hrcu_headG}(hj9hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj9ubhG structure be aligned to a two-byte boundary, and passing a misaligned }(hj9hhhNhNubj)}(h ``rcu_head``h]hrcu_head}(hj9hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj9ubh structure to one of the call_rcu() family of functions will result in a splat. It is therefore necessary to exercise caution when packing structures containing fields of type }(hj9hhhNhNubj)}(h ``rcu_head``h]hrcu_head}(hj9hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj9ubh. Why not a four-byte or even eight-byte alignment requirement? Because the m68k architecture provides only two-byte alignment, and thus acts as alignment’s least common denominator.}(hj9hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhM> hj8hhubh)}(hXThe reason for reserving the bottom bit of pointers to ``rcu_head`` structures is to leave the door open to “lazy” callbacks whose invocations can safely be deferred. Deferring invocation could potentially have energy-efficiency benefits, but only if the rate of non-lazy callbacks decreases significantly for some important workload. In the meantime, reserving the bottom bit keeps this option open in case it one day becomes useful.h](h7The reason for reserving the bottom bit of pointers to }(hj9hhhNhNubj)}(h ``rcu_head``h]hrcu_head}(hj9hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj9ubhXs structures is to leave the door open to “lazy” callbacks whose invocations can safely be deferred. Deferring invocation could potentially have energy-efficiency benefits, but only if the rate of non-lazy callbacks decreases significantly for some important workload. In the meantime, reserving the bottom bit keeps this option open in case it one day becomes useful.}(hj9hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMG hj8hhubeh}(h]j)ah ]h"]memory efficiencyah$]h&]uh1hhj'hhhhhM jKubh)}(hhh](h)}(h8Performance, Scalability, Response Time, and Reliabilityh]h8Performance, Scalability, Response Time, and Reliability}(hj:hhhNhNubah}(h]h ]h"]h$]h&]uh1hhj :hhhhhMP ubh)}(hXExpanding on the `earlier discussion `__, RCU is used heavily by hot code paths in performance-critical portions of the Linux kernel's networking, security, virtualization, and scheduling code paths. RCU must therefore use efficient implementations, especially in its read-side primitives. To that end, it would be good if preemptible RCU's implementation of rcu_read_lock() could be inlined, however, doing this requires resolving ``#include`` issues with the ``task_struct`` structure.h](hExpanding on the }(hj:hhhNhNubh)}(h5`earlier discussion `__h]hearlier discussion}(hj%:hhhNhNubah}(h]h ]h"]h$]h&]nameearlier discussionjjuh1hhj:jKubhX, RCU is used heavily by hot code paths in performance-critical portions of the Linux kernel’s networking, security, virtualization, and scheduling code paths. RCU must therefore use efficient implementations, especially in its read-side primitives. To that end, it would be good if preemptible RCU’s implementation of rcu_read_lock() could be inlined, however, doing this requires resolving }(hj:hhhNhNubj)}(h ``#include``h]h#include}(hj9:hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj:ubh issues with the }(hj:hhhNhNubj)}(h``task_struct``h]h task_struct}(hjK:hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj:ubh structure.}(hj:hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMR hj :hhubh)}(hX;The Linux kernel supports hardware configurations with up to 4096 CPUs, which means that RCU must be extremely scalable. Algorithms that involve frequent acquisitions of global locks or frequent atomic operations on global variables simply cannot be tolerated within the RCU implementation. RCU therefore makes heavy use of a combining tree based on the ``rcu_node`` structure. RCU is required to tolerate all CPUs continuously invoking any combination of RCU's runtime primitives with minimal per-operation overhead. In fact, in many cases, increasing load must *decrease* the per-operation overhead, witness the batching optimizations for synchronize_rcu(), call_rcu(), synchronize_rcu_expedited(), and rcu_barrier(). As a general rule, RCU must cheerfully accept whatever the rest of the Linux kernel decides to throw at it.h](hXbThe Linux kernel supports hardware configurations with up to 4096 CPUs, which means that RCU must be extremely scalable. Algorithms that involve frequent acquisitions of global locks or frequent atomic operations on global variables simply cannot be tolerated within the RCU implementation. RCU therefore makes heavy use of a combining tree based on the }(hjc:hhhNhNubj)}(h ``rcu_node``h]hrcu_node}(hjk:hhhNhNubah}(h]h ]h"]h$]h&]uh1jhjc:ubh structure. RCU is required to tolerate all CPUs continuously invoking any combination of RCU’s runtime primitives with minimal per-operation overhead. In fact, in many cases, increasing load must }(hjc:hhhNhNubj9 )}(h *decrease*h]hdecrease}(hj}:hhhNhNubah}(h]h ]h"]h$]h&]uh1j8 hjc:ubh the per-operation overhead, witness the batching optimizations for synchronize_rcu(), call_rcu(), synchronize_rcu_expedited(), and rcu_barrier(). As a general rule, RCU must cheerfully accept whatever the rest of the Linux kernel decides to throw at it.}(hjc:hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhM\ hj :hhubh)}(hXThe Linux kernel is used for real-time workloads, especially in conjunction with the `-rt patchset `__. The real-time-latency response requirements are such that the traditional approach of disabling preemption across RCU read-side critical sections is inappropriate. Kernels built with ``CONFIG_PREEMPTION=y`` therefore use an RCU implementation that allows RCU read-side critical sections to be preempted. This requirement made its presence known after users made it clear that an earlier `real-time patch `__ did not meet their needs, in conjunction with some `RCU issues `__ encountered by a very early version of the -rt patchset.h](hUThe Linux kernel is used for real-time workloads, especially in conjunction with the }(hj:hhhNhNubh)}(h=`-rt patchset `__h]h -rt patchset}(hj:hhhNhNubah}(h]h ]h"]h$]h&]name -rt patchsetj*https://wiki.linuxfoundation.org/realtime/uh1hhj:ubh. The real-time-latency response requirements are such that the traditional approach of disabling preemption across RCU read-side critical sections is inappropriate. Kernels built with }(hj:hhhNhNubj)}(h``CONFIG_PREEMPTION=y``h]hCONFIG_PREEMPTION=y}(hj:hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj:ubh therefore use an RCU implementation that allows RCU read-side critical sections to be preempted. This requirement made its presence known after users made it clear that an earlier }(hj:hhhNhNubh)}(h6`real-time patch `__h]hreal-time patch}(hj:hhhNhNubah}(h]h ]h"]h$]h&]namereal-time patchj https://lwn.net/Articles/107930/uh1hhj:ubh4 did not meet their needs, in conjunction with some }(hj:hhhNhNubh)}(hK`RCU issues `__h]h RCU issues}(hj:hhhNhNubah}(h]h ]h"]h$]h&]name RCU issuesj:https://lore.kernel.org/r/20050318002026.GA2693@us.ibm.comuh1hhj:ubh9 encountered by a very early version of the -rt patchset.}(hj:hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMj hj :hhubh)}(hXIn addition, RCU must make do with a sub-100-microsecond real-time latency budget. In fact, on smaller systems with the -rt patchset, the Linux kernel provides sub-20-microsecond real-time latencies for the whole kernel, including RCU. RCU's scalability and latency must therefore be sufficient for these sorts of configurations. To my surprise, the sub-100-microsecond real-time latency budget `applies to even the largest systems [PDF] `__, up to and including systems with 4096 CPUs. This real-time requirement motivated the grace-period kthread, which also simplified handling of a number of race conditions.h](hXIn addition, RCU must make do with a sub-100-microsecond real-time latency budget. In fact, on smaller systems with the -rt patchset, the Linux kernel provides sub-20-microsecond real-time latencies for the whole kernel, including RCU. RCU’s scalability and latency must therefore be sufficient for these sorts of configurations. To my surprise, the sub-100-microsecond real-time latency budget }(hj:hhhNhNubh)}(h{`applies to even the largest systems [PDF] `__h]h)applies to even the largest systems [PDF]}(hj:hhhNhNubah}(h]h ]h"]h$]h&]name)applies to even the largest systems [PDF]jKhttp://www.rdrop.com/users/paulmck/realtime/paper/bigrt.2013.01.31a.LCA.pdfuh1hhj:ubh, up to and including systems with 4096 CPUs. This real-time requirement motivated the grace-period kthread, which also simplified handling of a number of race conditions.}(hj:hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMx hj :hhubh)}(hXCRCU must avoid degrading real-time response for CPU-bound threads, whether executing in usermode (which is one use case for ``CONFIG_NO_HZ_FULL=y``) or in the kernel. That said, CPU-bound loops in the kernel must execute cond_resched() at least once per few tens of milliseconds in order to avoid receiving an IPI from RCU.h](h|RCU must avoid degrading real-time response for CPU-bound threads, whether executing in usermode (which is one use case for }(hj;hhhNhNubj)}(h``CONFIG_NO_HZ_FULL=y``h]hCONFIG_NO_HZ_FULL=y}(hj;hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj;ubh) or in the kernel. That said, CPU-bound loops in the kernel must execute cond_resched() at least once per few tens of milliseconds in order to avoid receiving an IPI from RCU.}(hj;hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhM hj :hhubh)}(hXUFinally, RCU's status as a synchronization primitive means that any RCU failure can result in arbitrary memory corruption that can be extremely difficult to debug. This means that RCU must be extremely reliable, which in practice also means that RCU must have an aggressive stress-test suite. This stress-test suite is called ``rcutorture``.h](hXHFinally, RCU’s status as a synchronization primitive means that any RCU failure can result in arbitrary memory corruption that can be extremely difficult to debug. This means that RCU must be extremely reliable, which in practice also means that RCU must have an aggressive stress-test suite. This stress-test suite is called }(hj7;hhhNhNubj)}(h``rcutorture``h]h rcutorture}(hj?;hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj7;ubh.}(hj7;hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhM hj :hhubh)}(hXAlthough the need for ``rcutorture`` was no surprise, the current immense popularity of the Linux kernel is posing interesting—and perhaps unprecedented—validation challenges. To see this, keep in mind that there are well over one billion instances of the Linux kernel running today, given Android smartphones, Linux-powered televisions, and servers. This number can be expected to increase sharply with the advent of the celebrated Internet of Things.h](hAlthough the need for }(hjW;hhhNhNubj)}(h``rcutorture``h]h rcutorture}(hj_;hhhNhNubah}(h]h ]h"]h$]h&]uh1jhjW;ubhX was no surprise, the current immense popularity of the Linux kernel is posing interesting—and perhaps unprecedented—validation challenges. To see this, keep in mind that there are well over one billion instances of the Linux kernel running today, given Android smartphones, Linux-powered televisions, and servers. This number can be expected to increase sharply with the advent of the celebrated Internet of Things.}(hjW;hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhM hj :hhubh)}(hXRSuppose that RCU contains a race condition that manifests on average once per million years of runtime. This bug will be occurring about three times per *day* across the installed base. RCU could simply hide behind hardware error rates, given that no one should really expect their smartphone to last for a million years. However, anyone taking too much comfort from this thought should consider the fact that in most jurisdictions, a successful multi-year test of a given mechanism, which might include a Linux kernel, suffices for a number of types of safety-critical certifications. In fact, rumor has it that the Linux kernel is already being used in production for safety-critical applications. I don't know about you, but I would feel quite bad if a bug in RCU killed someone. Which might explain my recent focus on validation and verification.h](hSuppose that RCU contains a race condition that manifests on average once per million years of runtime. This bug will be occurring about three times per }(hjw;hhhNhNubj9 )}(h*day*h]hday}(hj;hhhNhNubah}(h]h ]h"]h$]h&]uh1j8 hjw;ubhX across the installed base. RCU could simply hide behind hardware error rates, given that no one should really expect their smartphone to last for a million years. However, anyone taking too much comfort from this thought should consider the fact that in most jurisdictions, a successful multi-year test of a given mechanism, which might include a Linux kernel, suffices for a number of types of safety-critical certifications. In fact, rumor has it that the Linux kernel is already being used in production for safety-critical applications. I don’t know about you, but I would feel quite bad if a bug in RCU killed someone. Which might explain my recent focus on validation and verification.}(hjw;hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhM hj :hhubeh}(h]j)ah ]h"]8performance, scalability, response time, and reliabilityah$]h&]uh1hhj'hhhhhMP jKubeh}(h]jkah ]h"]linux kernel complicationsah$]h&]uh1hhhhhhhhMjKubh)}(hhh](h)}(hOther RCU Flavorsh]hOther RCU Flavors}(hj;hhhNhNubah}(h]h ]h"]h$]h&]uh1hhj;hhhhhM ubh)}(hXgOne of the more surprising things about RCU is that there are now no fewer than five *flavors*, or API families. In addition, the primary flavor that has been the sole focus up to this point has two different implementations, non-preemptible and preemptible. The other four flavors are listed below, with requirements for each described in a separate section.h](hUOne of the more surprising things about RCU is that there are now no fewer than five }(hj;hhhNhNubj9 )}(h *flavors*h]hflavors}(hj;hhhNhNubah}(h]h ]h"]h$]h&]uh1j8 hj;ubhX , or API families. In addition, the primary flavor that has been the sole focus up to this point has two different implementations, non-preemptible and preemptible. The other four flavors are listed below, with requirements for each described in a separate section.}(hj;hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhM hj;hhubj)}(hhh](j)}(h"`Bottom-Half Flavor (Historical)`_h]h)}(hj;h]h)}(hj;h]hBottom-Half Flavor (Historical)}(hj;hhhNhNubah}(h]h ]h"]h$]h&]nameBottom-Half Flavor (Historical)jbottom-half-flavor-historicaluh1hhj;jKubah}(h]h ]h"]h$]h&]uh1hhhhM hj;ubah}(h]h ]h"]h$]h&]uh1jhj;hhhhhNubj)}(h`Sched Flavor (Historical)`_h]h)}(hj;h]h)}(hj;h]hSched Flavor (Historical)}(hj<hhhNhNubah}(h]h ]h"]h$]h&]nameSched Flavor (Historical)jsched-flavor-historicaluh1hhj<jKubah}(h]h ]h"]h$]h&]uh1hhhhM hj;ubah}(h]h ]h"]h$]h&]uh1jhj;hhhhhNubj)}(h`Sleepable RCU`_h]h)}(hj!<h]h)}(hj!<h]h Sleepable RCU}(hj&<hhhNhNubah}(h]h ]h"]h$]h&]name Sleepable RCUjjuh1hhj#<jKubah}(h]h ]h"]h$]h&]uh1hhhhM hj<ubah}(h]h ]h"]h$]h&]uh1jhj;hhhhhNubj)}(h `Tasks RCU`_h]h)}(hjC<h]h)}(hjC<h]h Tasks RCU}(hjH<hhhNhNubah}(h]h ]h"]h$]h&]name Tasks RCUjj4#uh1hhjE<jKubah}(h]h ]h"]h$]h&]uh1hhhhM hjA<ubah}(h]h ]h"]h$]h&]uh1jhj;hhhhhNubj)}(h`Tasks Trace RCU`_ h]h)}(h`Tasks Trace RCU`_h]h)}(hji<h]hTasks Trace RCU}(hjk<hhhNhNubah}(h]h ]h"]h$]h&]nameTasks Trace RCUjtasks-trace-rcuuh1hhjg<jKubah}(h]h ]h"]h$]h&]uh1hhhhM hjc<ubah}(h]h ]h"]h$]h&]uh1jhj;hhhhhNubeh}(h]h ]h"]h$]h&]jjjhjjuh1jhj;hhhhhM ubh)}(hhh](h)}(hBottom-Half Flavor (Historical)h]hBottom-Half Flavor (Historical)}(hj<hhhNhNubah}(h]h ]h"]h$]h&]uh1hhj<hhhhhM ubh)}(hXJThe RCU-bh flavor of RCU has since been expressed in terms of the other RCU flavors as part of a consolidation of the three flavors into a single flavor. The read-side API remains, and continues to disable softirq and to be accounted for by lockdep. Much of the material in this section is therefore strictly historical in nature.h]hXJThe RCU-bh flavor of RCU has since been expressed in terms of the other RCU flavors as part of a consolidation of the three flavors into a single flavor. The read-side API remains, and continues to disable softirq and to be accounted for by lockdep. Much of the material in this section is therefore strictly historical in nature.}(hj<hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM hj<hhubh)}(hXSThe softirq-disable (AKA “bottom-half”, hence the “_bh” abbreviations) flavor of RCU, or *RCU-bh*, was developed by Dipankar Sarma to provide a flavor of RCU that could withstand the network-based denial-of-service attacks researched by Robert Olsson. These attacks placed so much networking load on the system that some of the CPUs never exited softirq execution, which in turn prevented those CPUs from ever executing a context switch, which, in the RCU implementation of that time, prevented grace periods from ever ending. The result was an out-of-memory condition and a system hang.h](haThe softirq-disable (AKA “bottom-half”, hence the “_bh” abbreviations) flavor of RCU, or }(hj<hhhNhNubj9 )}(h*RCU-bh*h]hRCU-bh}(hj<hhhNhNubah}(h]h ]h"]h$]h&]uh1j8 hj<ubhX, was developed by Dipankar Sarma to provide a flavor of RCU that could withstand the network-based denial-of-service attacks researched by Robert Olsson. These attacks placed so much networking load on the system that some of the CPUs never exited softirq execution, which in turn prevented those CPUs from ever executing a context switch, which, in the RCU implementation of that time, prevented grace periods from ever ending. The result was an out-of-memory condition and a system hang.}(hj<hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhM hj<hhubh)}(hXThe solution was the creation of RCU-bh, which does local_bh_disable() across its read-side critical sections, and which uses the transition from one type of softirq processing to another as a quiescent state in addition to context switch, idle, user mode, and offline. This means that RCU-bh grace periods can complete even when some of the CPUs execute in softirq indefinitely, thus allowing algorithms based on RCU-bh to withstand network-based denial-of-service attacks.h]hXThe solution was the creation of RCU-bh, which does local_bh_disable() across its read-side critical sections, and which uses the transition from one type of softirq processing to another as a quiescent state in addition to context switch, idle, user mode, and offline. This means that RCU-bh grace periods can complete even when some of the CPUs execute in softirq indefinitely, thus allowing algorithms based on RCU-bh to withstand network-based denial-of-service attacks.}(hj<hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM hj<hhubh)}(hXBecause rcu_read_lock_bh() and rcu_read_unlock_bh() disable and re-enable softirq handlers, any attempt to start a softirq handlers during the RCU-bh read-side critical section will be deferred. In this case, rcu_read_unlock_bh() will invoke softirq processing, which can take considerable time. One can of course argue that this softirq overhead should be associated with the code following the RCU-bh read-side critical section rather than rcu_read_unlock_bh(), but the fact is that most profiling tools cannot be expected to make this sort of fine distinction. For example, suppose that a three-millisecond-long RCU-bh read-side critical section executes during a time of heavy networking load. There will very likely be an attempt to invoke at least one softirq handler during that three milliseconds, but any such invocation will be delayed until the time of the rcu_read_unlock_bh(). This can of course make it appear at first glance as if rcu_read_unlock_bh() was executing very slowly.h]hXBecause rcu_read_lock_bh() and rcu_read_unlock_bh() disable and re-enable softirq handlers, any attempt to start a softirq handlers during the RCU-bh read-side critical section will be deferred. In this case, rcu_read_unlock_bh() will invoke softirq processing, which can take considerable time. One can of course argue that this softirq overhead should be associated with the code following the RCU-bh read-side critical section rather than rcu_read_unlock_bh(), but the fact is that most profiling tools cannot be expected to make this sort of fine distinction. For example, suppose that a three-millisecond-long RCU-bh read-side critical section executes during a time of heavy networking load. There will very likely be an attempt to invoke at least one softirq handler during that three milliseconds, but any such invocation will be delayed until the time of the rcu_read_unlock_bh(). This can of course make it appear at first glance as if rcu_read_unlock_bh() was executing very slowly.}(hj<hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM hj<hhubh)}(hX5The `RCU-bh API `__ includes rcu_read_lock_bh(), rcu_read_unlock_bh(), rcu_dereference_bh(), rcu_dereference_bh_check(), and rcu_read_lock_bh_held(). However, the old RCU-bh update-side APIs are now gone, replaced by synchronize_rcu(), synchronize_rcu_expedited(), call_rcu(), and rcu_barrier(). In addition, anything that disables bottom halves also marks an RCU-bh read-side critical section, including local_bh_disable() and local_bh_enable(), local_irq_save() and local_irq_restore(), and so on.h](hThe }(hj<hhhNhNubh)}(hP`RCU-bh API `__h]h RCU-bh API}(hj<hhhNhNubah}(h]h ]h"]h$]h&]name RCU-bh APIj?https://lwn.net/Articles/609973/#RCU%20Per-Flavor%20API%20Tableuh1hhj<ubhX includes rcu_read_lock_bh(), rcu_read_unlock_bh(), rcu_dereference_bh(), rcu_dereference_bh_check(), and rcu_read_lock_bh_held(). However, the old RCU-bh update-side APIs are now gone, replaced by synchronize_rcu(), synchronize_rcu_expedited(), call_rcu(), and rcu_barrier(). In addition, anything that disables bottom halves also marks an RCU-bh read-side critical section, including local_bh_disable() and local_bh_enable(), local_irq_save() and local_irq_restore(), and so on.}(hj<hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhM hj<hhubeh}(h]j;ah ]h"]bottom-half flavor (historical)ah$]h&]uh1hhj;hhhhhM jKubh)}(hhh](h)}(hSched Flavor (Historical)h]hSched Flavor (Historical)}(hj=hhhNhNubah}(h]h ]h"]h$]h&]uh1hhj=hhhhhM ubh)}(hXPThe RCU-sched flavor of RCU has since been expressed in terms of the other RCU flavors as part of a consolidation of the three flavors into a single flavor. The read-side API remains, and continues to disable preemption and to be accounted for by lockdep. Much of the material in this section is therefore strictly historical in nature.h]hXPThe RCU-sched flavor of RCU has since been expressed in terms of the other RCU flavors as part of a consolidation of the three flavors into a single flavor. The read-side API remains, and continues to disable preemption and to be accounted for by lockdep. Much of the material in this section is therefore strictly historical in nature.}(hj#=hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM hj=hhubh)}(hXBefore preemptible RCU, waiting for an RCU grace period had the side effect of also waiting for all pre-existing interrupt and NMI handlers. However, there are legitimate preemptible-RCU implementations that do not have this property, given that any point in the code outside of an RCU read-side critical section can be a quiescent state. Therefore, *RCU-sched* was created, which follows “classic” RCU in that an RCU-sched grace period waits for pre-existing interrupt and NMI handlers. In kernels built with ``CONFIG_PREEMPTION=n``, the RCU and RCU-sched APIs have identical implementations, while kernels built with ``CONFIG_PREEMPTION=y`` provide a separate implementation for each.h](hX^Before preemptible RCU, waiting for an RCU grace period had the side effect of also waiting for all pre-existing interrupt and NMI handlers. However, there are legitimate preemptible-RCU implementations that do not have this property, given that any point in the code outside of an RCU read-side critical section can be a quiescent state. Therefore, }(hj1=hhhNhNubj9 )}(h *RCU-sched*h]h RCU-sched}(hj9=hhhNhNubah}(h]h ]h"]h$]h&]uh1j8 hj1=ubh was created, which follows “classic” RCU in that an RCU-sched grace period waits for pre-existing interrupt and NMI handlers. In kernels built with }(hj1=hhhNhNubj)}(h``CONFIG_PREEMPTION=n``h]hCONFIG_PREEMPTION=n}(hjK=hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj1=ubhV, the RCU and RCU-sched APIs have identical implementations, while kernels built with }(hj1=hhhNhNubj)}(h``CONFIG_PREEMPTION=y``h]hCONFIG_PREEMPTION=y}(hj]=hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj1=ubh, provide a separate implementation for each.}(hj1=hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhM hj=hhubh)}(hXHNote well that in ``CONFIG_PREEMPTION=y`` kernels, rcu_read_lock_sched() and rcu_read_unlock_sched() disable and re-enable preemption, respectively. This means that if there was a preemption attempt during the RCU-sched read-side critical section, rcu_read_unlock_sched() will enter the scheduler, with all the latency and overhead entailed. Just as with rcu_read_unlock_bh(), this can make it look as if rcu_read_unlock_sched() was executing very slowly. However, the highest-priority task won't be preempted, so that task will enjoy low-overhead rcu_read_unlock_sched() invocations.h](hNote well that in }(hju=hhhNhNubj)}(h``CONFIG_PREEMPTION=y``h]hCONFIG_PREEMPTION=y}(hj}=hhhNhNubah}(h]h ]h"]h$]h&]uh1jhju=ubhX! kernels, rcu_read_lock_sched() and rcu_read_unlock_sched() disable and re-enable preemption, respectively. This means that if there was a preemption attempt during the RCU-sched read-side critical section, rcu_read_unlock_sched() will enter the scheduler, with all the latency and overhead entailed. Just as with rcu_read_unlock_bh(), this can make it look as if rcu_read_unlock_sched() was executing very slowly. However, the highest-priority task won’t be preempted, so that task will enjoy low-overhead rcu_read_unlock_sched() invocations.}(hju=hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhM hj=hhubh)}(hXThe `RCU-sched API `__ includes rcu_read_lock_sched(), rcu_read_unlock_sched(), rcu_read_lock_sched_notrace(), rcu_read_unlock_sched_notrace(), rcu_dereference_sched(), rcu_dereference_sched_check(), and rcu_read_lock_sched_held(). However, the old RCU-sched update-side APIs are now gone, replaced by synchronize_rcu(), synchronize_rcu_expedited(), call_rcu(), and rcu_barrier(). In addition, anything that disables preemption also marks an RCU-sched read-side critical section, including preempt_disable() and preempt_enable(), local_irq_save() and local_irq_restore(), and so on.h](hThe }(hj=hhhNhNubh)}(hS`RCU-sched API `__h]h RCU-sched API}(hj=hhhNhNubah}(h]h ]h"]h$]h&]name RCU-sched APIj?https://lwn.net/Articles/609973/#RCU%20Per-Flavor%20API%20Tableuh1hhj=ubhX2 includes rcu_read_lock_sched(), rcu_read_unlock_sched(), rcu_read_lock_sched_notrace(), rcu_read_unlock_sched_notrace(), rcu_dereference_sched(), rcu_dereference_sched_check(), and rcu_read_lock_sched_held(). However, the old RCU-sched update-side APIs are now gone, replaced by synchronize_rcu(), synchronize_rcu_expedited(), call_rcu(), and rcu_barrier(). In addition, anything that disables preemption also marks an RCU-sched read-side critical section, including preempt_disable() and preempt_enable(), local_irq_save() and local_irq_restore(), and so on.}(hj=hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhM hj=hhubeh}(h]j<ah ]h"]sched flavor (historical)ah$]h&]uh1hhj;hhhhhM jKubh)}(hhh](h)}(h Sleepable RCUh]h Sleepable RCU}(hj=hhhNhNubah}(h]h ]h"]h$]h&]uh1hhj=hhhhhM ubh)}(hX?For well over a decade, someone saying “I need to block within an RCU read-side critical section” was a reliable indication that this someone did not understand RCU. After all, if you are always blocking in an RCU read-side critical section, you can probably afford to use a higher-overhead synchronization mechanism. However, that changed with the advent of the Linux kernel's notifiers, whose RCU read-side critical sections almost never sleep, but sometimes need to. This resulted in the introduction of `sleepable RCU `__, or *SRCU*.h](hXFor well over a decade, someone saying “I need to block within an RCU read-side critical section” was a reliable indication that this someone did not understand RCU. After all, if you are always blocking in an RCU read-side critical section, you can probably afford to use a higher-overhead synchronization mechanism. However, that changed with the advent of the Linux kernel’s notifiers, whose RCU read-side critical sections almost never sleep, but sometimes need to. This resulted in the introduction of }(hj=hhhNhNubh)}(h4`sleepable RCU `__h]h sleepable RCU}(hj=hhhNhNubah}(h]h ]h"]h$]h&]name sleepable RCUj https://lwn.net/Articles/202847/uh1hhj=ubh, or }(hj=hhhNhNubj9 )}(h*SRCU*h]hSRCU}(hj=hhhNhNubah}(h]h ]h"]h$]h&]uh1j8 hj=ubh.}(hj=hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhM hj=hhubh)}(hX<SRCU allows different domains to be defined, with each such domain defined by an instance of an ``srcu_struct`` structure. A pointer to this structure must be passed in to each SRCU function, for example, ``synchronize_srcu(&ss)``, where ``ss`` is the ``srcu_struct`` structure. The key benefit of these domains is that a slow SRCU reader in one domain does not delay an SRCU grace period in some other domain. That said, one consequence of these domains is that read-side code must pass a “cookie” from srcu_read_lock() to srcu_read_unlock(), for example, as follows:h](h`SRCU allows different domains to be defined, with each such domain defined by an instance of an }(hj>hhhNhNubj)}(h``srcu_struct``h]h srcu_struct}(hj >hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj>ubh^ structure. A pointer to this structure must be passed in to each SRCU function, for example, }(hj>hhhNhNubj)}(h``synchronize_srcu(&ss)``h]hsynchronize_srcu(&ss)}(hj>hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj>ubh, where }(hj>hhhNhNubj)}(h``ss``h]hss}(hj1>hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj>ubh is the }(hj>hhhNhNubj)}(h``srcu_struct``h]h srcu_struct}(hjC>hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj>ubhX1 structure. The key benefit of these domains is that a slow SRCU reader in one domain does not delay an SRCU grace period in some other domain. That said, one consequence of these domains is that read-side code must pass a “cookie” from srcu_read_lock() to srcu_read_unlock(), for example, as follows:}(hj>hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhM$ hj=hhubj@)}(hr:: 1 int idx; 2 3 idx = srcu_read_lock(&ss); 4 do_something(); 5 srcu_read_unlock(&ss, idx); h]jF)}(hY1 int idx; 2 3 idx = srcu_read_lock(&ss); 4 do_something(); 5 srcu_read_unlock(&ss, idx);h]hY1 int idx; 2 3 idx = srcu_read_lock(&ss); 4 do_something(); 5 srcu_read_unlock(&ss, idx);}hj_>sbah}(h]h ]h"]h$]h&]jUjVuh1jEhhhM0 hj[>ubah}(h]h ]h"]h$]h&]uh1j?hhhM. hj=hhubh)}(hX$As noted above, it is legal to block within SRCU read-side critical sections, however, with great power comes great responsibility. If you block forever in one of a given domain's SRCU read-side critical sections, then that domain's grace periods will also be blocked forever. Of course, one good way to block forever is to deadlock, which can happen if any operation in a given domain's SRCU read-side critical section can wait, either directly or indirectly, for that domain's grace period to elapse. For example, this results in a self-deadlock:h]hX,As noted above, it is legal to block within SRCU read-side critical sections, however, with great power comes great responsibility. If you block forever in one of a given domain’s SRCU read-side critical sections, then that domain’s grace periods will also be blocked forever. Of course, one good way to block forever is to deadlock, which can happen if any operation in a given domain’s SRCU read-side critical section can wait, either directly or indirectly, for that domain’s grace period to elapse. For example, this results in a self-deadlock:}(hjs>hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM6 hj=hhubj@)}(h:: 1 int idx; 2 3 idx = srcu_read_lock(&ss); 4 do_something(); 5 synchronize_srcu(&ss); 6 srcu_read_unlock(&ss, idx); h]jF)}(hr1 int idx; 2 3 idx = srcu_read_lock(&ss); 4 do_something(); 5 synchronize_srcu(&ss); 6 srcu_read_unlock(&ss, idx);h]hr1 int idx; 2 3 idx = srcu_read_lock(&ss); 4 do_something(); 5 synchronize_srcu(&ss); 6 srcu_read_unlock(&ss, idx);}hj>sbah}(h]h ]h"]h$]h&]jUjVuh1jEhhhMA hj>ubah}(h]h ]h"]h$]h&]uh1j?hhhM? hj=hhubh)}(hXrHowever, if line 5 acquired a mutex that was held across a synchronize_srcu() for domain ``ss``, deadlock would still be possible. Furthermore, if line 5 acquired a mutex that was held across a synchronize_srcu() for some other domain ``ss1``, and if an ``ss1``-domain SRCU read-side critical section acquired another mutex that was held across as ``ss``-domain synchronize_srcu(), deadlock would again be possible. Such a deadlock cycle could extend across an arbitrarily large number of different SRCU domains. Again, with great power comes great responsibility, though lockdep is now able to detect this sort of deadlock.h](hZHowever, if line 5 acquired a mutex that was held across a synchronize_srcu() for domain }(hj>hhhNhNubj)}(h``ss``h]hss}(hj>hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj>ubh, deadlock would still be possible. Furthermore, if line 5 acquired a mutex that was held across a synchronize_srcu() for some other domain }(hj>hhhNhNubj)}(h``ss1``h]hss1}(hj>hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj>ubh , and if an }(hj>hhhNhNubj)}(h``ss1``h]hss1}(hj>hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj>ubhW-domain SRCU read-side critical section acquired another mutex that was held across as }(hj>hhhNhNubj)}(h``ss``h]hss}(hj>hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj>ubhX-domain synchronize_srcu(), deadlock would again be possible. Such a deadlock cycle could extend across an arbitrarily large number of different SRCU domains. Again, with great power comes great responsibility, though lockdep is now able to detect this sort of deadlock.}(hj>hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMH hj=hhubh)}(hXUnlike the other RCU flavors, SRCU read-side critical sections can run on idle and even offline CPUs, with the exception of srcu_read_lock_fast() and friends. This ability requires that srcu_read_lock() and srcu_read_unlock() contain memory barriers, which means that SRCU readers will run a bit slower than would RCU readers. It also motivates the smp_mb__after_srcu_read_unlock() API, which, in combination with srcu_read_unlock(), guarantees a full memory barrier.h]hXUnlike the other RCU flavors, SRCU read-side critical sections can run on idle and even offline CPUs, with the exception of srcu_read_lock_fast() and friends. This ability requires that srcu_read_lock() and srcu_read_unlock() contain memory barriers, which means that SRCU readers will run a bit slower than would RCU readers. It also motivates the smp_mb__after_srcu_read_unlock() API, which, in combination with srcu_read_unlock(), guarantees a full memory barrier.}(hj>hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMS hj=hhubh)}(hXAlso unlike other RCU flavors, synchronize_srcu() may **not** be invoked from CPU-hotplug notifiers, due to the fact that SRCU grace periods make use of timers and the possibility of timers being temporarily “stranded” on the outgoing CPU. This stranding of timers means that timers posted to the outgoing CPU will not fire until late in the CPU-hotplug process. The problem is that if a notifier is waiting on an SRCU grace period, that grace period is waiting on a timer, and that timer is stranded on the outgoing CPU, then the notifier will never be awakened, in other words, deadlock has occurred. This same situation of course also prohibits srcu_barrier() from being invoked from CPU-hotplug notifiers.h](h6Also unlike other RCU flavors, synchronize_srcu() may }(hj>hhhNhNubj)}(h**not**h]hnot}(hj?hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj>ubhX be invoked from CPU-hotplug notifiers, due to the fact that SRCU grace periods make use of timers and the possibility of timers being temporarily “stranded” on the outgoing CPU. This stranding of timers means that timers posted to the outgoing CPU will not fire until late in the CPU-hotplug process. The problem is that if a notifier is waiting on an SRCU grace period, that grace period is waiting on a timer, and that timer is stranded on the outgoing CPU, then the notifier will never be awakened, in other words, deadlock has occurred. This same situation of course also prohibits srcu_barrier() from being invoked from CPU-hotplug notifiers.}(hj>hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhM[ hj=hhubh)}(hXSRCU also differs from other RCU flavors in that SRCU's expedited and non-expedited grace periods are implemented by the same mechanism. This means that in the current SRCU implementation, expediting a future grace period has the side effect of expediting all prior grace periods that have not yet completed. (But please note that this is a property of the current implementation, not necessarily of future implementations.) In addition, if SRCU has been idle for longer than the interval specified by the ``srcutree.exp_holdoff`` kernel boot parameter (25 microseconds by default), and if a synchronize_srcu() invocation ends this idle period, that invocation will be automatically expedited.h](hXSRCU also differs from other RCU flavors in that SRCU’s expedited and non-expedited grace periods are implemented by the same mechanism. This means that in the current SRCU implementation, expediting a future grace period has the side effect of expediting all prior grace periods that have not yet completed. (But please note that this is a property of the current implementation, not necessarily of future implementations.) In addition, if SRCU has been idle for longer than the interval specified by the }(hj?hhhNhNubj)}(h``srcutree.exp_holdoff``h]hsrcutree.exp_holdoff}(hj%?hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj?ubh kernel boot parameter (25 microseconds by default), and if a synchronize_srcu() invocation ends this idle period, that invocation will be automatically expedited.}(hj?hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhMg hj=hhubh)}(hXAs of v4.12, SRCU's callbacks are maintained per-CPU, eliminating a locking bottleneck present in prior kernel versions. Although this will allow users to put much heavier stress on call_srcu(), it is important to note that SRCU does not yet take any special steps to deal with callback flooding. So if you are posting (say) 10,000 SRCU callbacks per second per CPU, you are probably totally OK, but if you intend to post (say) 1,000,000 SRCU callbacks per second per CPU, please run some tests first. SRCU just might need a few adjustment to deal with that sort of load. Of course, your mileage may vary based on the speed of your CPUs and the size of your memory.h]hXAs of v4.12, SRCU’s callbacks are maintained per-CPU, eliminating a locking bottleneck present in prior kernel versions. Although this will allow users to put much heavier stress on call_srcu(), it is important to note that SRCU does not yet take any special steps to deal with callback flooding. So if you are posting (say) 10,000 SRCU callbacks per second per CPU, you are probably totally OK, but if you intend to post (say) 1,000,000 SRCU callbacks per second per CPU, please run some tests first. SRCU just might need a few adjustment to deal with that sort of load. Of course, your mileage may vary based on the speed of your CPUs and the size of your memory.}(hj=?hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhMr hj=hhubh)}(hXThe `SRCU API `__ includes srcu_read_lock(), srcu_read_unlock(), srcu_dereference(), srcu_dereference_check(), synchronize_srcu(), synchronize_srcu_expedited(), call_srcu(), srcu_barrier(), and srcu_read_lock_held(). It also includes DEFINE_SRCU(), DEFINE_STATIC_SRCU(), DEFINE_SRCU_FAST(), DEFINE_STATIC_SRCU_FAST(), init_srcu_struct(), and init_srcu_struct_fast() APIs for defining and initializing ``srcu_struct`` structures.h](hThe }(hjK?hhhNhNubh)}(hN`SRCU API `__h]hSRCU API}(hjS?hhhNhNubah}(h]h ]h"]h$]h&]nameSRCU APIj?https://lwn.net/Articles/609973/#RCU%20Per-Flavor%20API%20Tableuh1hhjK?ubhX includes srcu_read_lock(), srcu_read_unlock(), srcu_dereference(), srcu_dereference_check(), synchronize_srcu(), synchronize_srcu_expedited(), call_srcu(), srcu_barrier(), and srcu_read_lock_held(). It also includes DEFINE_SRCU(), DEFINE_STATIC_SRCU(), DEFINE_SRCU_FAST(), DEFINE_STATIC_SRCU_FAST(), init_srcu_struct(), and init_srcu_struct_fast() APIs for defining and initializing }(hjK?hhhNhNubj)}(h``srcu_struct``h]h srcu_struct}(hjh?hhhNhNubah}(h]h ]h"]h$]h&]uh1jhjK?ubh structures.}(hjK?hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhM} hj=hhubh)}(h9More recently, the SRCU API has added polling interfaces:h]h9More recently, the SRCU API has added polling interfaces:}(hj?hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM hj=hhubj)}(hhh](j)}(hstart_poll_synchronize_srcu() returns a cookie identifying the completion of a future SRCU grace period and ensures that this grace period will be started.h]h)}(hstart_poll_synchronize_srcu() returns a cookie identifying the completion of a future SRCU grace period and ensures that this grace period will be started.h]hstart_poll_synchronize_srcu() returns a cookie identifying the completion of a future SRCU grace period and ensures that this grace period will be started.}(hj?hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM hj?ubah}(h]h ]h"]h$]h&]uh1jhj?hhhhhNubj)}(h~poll_state_synchronize_srcu() returns ``true`` iff the specified cookie corresponds to an already-completed SRCU grace period.h]h)}(h~poll_state_synchronize_srcu() returns ``true`` iff the specified cookie corresponds to an already-completed SRCU grace period.h](h&poll_state_synchronize_srcu() returns }(hj?hhhNhNubj)}(h``true``h]htrue}(hj?hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj?ubhP iff the specified cookie corresponds to an already-completed SRCU grace period.}(hj?hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhM hj?ubah}(h]h ]h"]h$]h&]uh1jhj?hhhhhNubj)}(hget_state_synchronize_srcu() returns a cookie just like start_poll_synchronize_srcu() does, but differs in that it does nothing to ensure that any future SRCU grace period will be started. h]h)}(hget_state_synchronize_srcu() returns a cookie just like start_poll_synchronize_srcu() does, but differs in that it does nothing to ensure that any future SRCU grace period will be started.h]hget_state_synchronize_srcu() returns a cookie just like start_poll_synchronize_srcu() does, but differs in that it does nothing to ensure that any future SRCU grace period will be started.}(hj?hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM hj?ubah}(h]h ]h"]h$]h&]uh1jhj?hhhhhNubeh}(h]h ]h"]h$]h&]jjjhjjuh1jhj=hhhhhM ubh)}(hXThese functions are used to avoid unnecessary SRCU grace periods in certain types of buffer-cache algorithms having multi-stage age-out mechanisms. The idea is that by the time the block has aged completely from the cache, an SRCU grace period will be very likely to have elapsed.h]hXThese functions are used to avoid unnecessary SRCU grace periods in certain types of buffer-cache algorithms having multi-stage age-out mechanisms. The idea is that by the time the block has aged completely from the cache, an SRCU grace period will be very likely to have elapsed.}(hj?hhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM hj=hhubeh}(h]jah ]h"] sleepable rcuah$]h&]uh1hhj;hhhhhM jKubh)}(hhh](h)}(h Tasks RCUh]h Tasks RCU}(hj @hhhNhNubah}(h]h ]h"]h$]h&]uh1hhj@hhhhhM ubh)}(hX4Some forms of tracing use “trampolines” to handle the binary rewriting required to install different types of probes. It would be good to be able to free old trampolines, which sounds like a job for some form of RCU. However, because it is necessary to be able to install a trace anywhere in the code, it is not possible to use read-side markers such as rcu_read_lock() and rcu_read_unlock(). In addition, it does not work to have these markers in the trampoline itself, because there would need to be instructions following rcu_read_unlock(). Although synchronize_rcu() would guarantee that execution reached the rcu_read_unlock(), it would not be able to guarantee that execution had completely left the trampoline. Worse yet, in some situations the trampoline's protection must extend a few instructions *prior* to execution reaching the trampoline. For example, these few instructions might calculate the address of the trampoline, so that entering the trampoline would be pre-ordained a surprisingly long time before execution actually reached the trampoline itself.h](hX-Some forms of tracing use “trampolines” to handle the binary rewriting required to install different types of probes. It would be good to be able to free old trampolines, which sounds like a job for some form of RCU. However, because it is necessary to be able to install a trace anywhere in the code, it is not possible to use read-side markers such as rcu_read_lock() and rcu_read_unlock(). In addition, it does not work to have these markers in the trampoline itself, because there would need to be instructions following rcu_read_unlock(). Although synchronize_rcu() would guarantee that execution reached the rcu_read_unlock(), it would not be able to guarantee that execution had completely left the trampoline. Worse yet, in some situations the trampoline’s protection must extend a few instructions }(hj@hhhNhNubj9 )}(h*prior*h]hprior}(hj@hhhNhNubah}(h]h ]h"]h$]h&]uh1j8 hj@ubhX to execution reaching the trampoline. For example, these few instructions might calculate the address of the trampoline, so that entering the trampoline would be pre-ordained a surprisingly long time before execution actually reached the trampoline itself.}(hj@hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhM hj@hhubh)}(hXThe solution, in the form of `Tasks RCU `__, is to have implicit read-side critical sections that are delimited by voluntary context switches, that is, calls to schedule(), cond_resched(), and synchronize_rcu_tasks(). In addition, transitions to and from userspace execution also delimit tasks-RCU read-side critical sections. Idle tasks are ignored by Tasks RCU, and Tasks Rude RCU may be used to interact with them.h](hThe solution, in the form of }(hj7@hhhNhNubh)}(h0`Tasks RCU `__h]h Tasks RCU}(hj?@hhhNhNubah}(h]h ]h"]h$]h&]name Tasks RCUj https://lwn.net/Articles/607117/uh1hhj7@ubhXv, is to have implicit read-side critical sections that are delimited by voluntary context switches, that is, calls to schedule(), cond_resched(), and synchronize_rcu_tasks(). In addition, transitions to and from userspace execution also delimit tasks-RCU read-side critical sections. Idle tasks are ignored by Tasks RCU, and Tasks Rude RCU may be used to interact with them.}(hj7@hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhM hj@hhubh)}(hXNote well that involuntary context switches are *not* Tasks-RCU quiescent states. After all, in preemptible kernels, a task executing code in a trampoline might be preempted. In this case, the Tasks-RCU grace period clearly cannot end until that task resumes and its execution leaves that trampoline. This means, among other things, that cond_resched() does not provide a Tasks RCU quiescent state. (Instead, use rcu_softirq_qs() from softirq or rcu_tasks_classic_qs() otherwise.)h](h0Note well that involuntary context switches are }(hjZ@hhhNhNubj9 )}(h*not*h]hnot}(hjb@hhhNhNubah}(h]h ]h"]h$]h&]uh1j8 hjZ@ubhX Tasks-RCU quiescent states. After all, in preemptible kernels, a task executing code in a trampoline might be preempted. In this case, the Tasks-RCU grace period clearly cannot end until that task resumes and its execution leaves that trampoline. This means, among other things, that cond_resched() does not provide a Tasks RCU quiescent state. (Instead, use rcu_softirq_qs() from softirq or rcu_tasks_classic_qs() otherwise.)}(hjZ@hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhM hj@hhubh)}(hXThe tasks-RCU API is quite compact, consisting only of call_rcu_tasks(), synchronize_rcu_tasks(), and rcu_barrier_tasks(). In ``CONFIG_PREEMPTION=n`` kernels, trampolines cannot be preempted, so these APIs map to call_rcu(), synchronize_rcu(), and rcu_barrier(), respectively. In ``CONFIG_PREEMPTION=y`` kernels, trampolines can be preempted, and these three APIs are therefore implemented by separate functions that check for voluntary context switches.h](h~The tasks-RCU API is quite compact, consisting only of call_rcu_tasks(), synchronize_rcu_tasks(), and rcu_barrier_tasks(). In }(hjz@hhhNhNubj)}(h``CONFIG_PREEMPTION=n``h]hCONFIG_PREEMPTION=n}(hj@hhhNhNubah}(h]h ]h"]h$]h&]uh1jhjz@ubh kernels, trampolines cannot be preempted, so these APIs map to call_rcu(), synchronize_rcu(), and rcu_barrier(), respectively. In }(hjz@hhhNhNubj)}(h``CONFIG_PREEMPTION=y``h]hCONFIG_PREEMPTION=y}(hj@hhhNhNubah}(h]h ]h"]h$]h&]uh1jhjz@ubh kernels, trampolines can be preempted, and these three APIs are therefore implemented by separate functions that check for voluntary context switches.}(hjz@hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhM hj@hhubeh}(h]j4#ah ]h"] tasks rcuah$]h&]uh1hhj;hhhhhM jKubh)}(hhh](h)}(hTasks Rude RCUh]hTasks Rude RCU}(hj@hhhNhNubah}(h]h ]h"]h$]h&]uh1hhj@hhhhhM ubh)}(hXASome forms of tracing need to wait for all preemption-disabled regions of code running on any online CPU, including those executed when RCU is not watching. This means that synchronize_rcu() is insufficient, and Tasks Rude RCU must be used instead. This flavor of RCU does its work by forcing a workqueue to be scheduled on each online CPU, hence the "Rude" moniker. And this operation is considered to be quite rude by real-time workloads that don't want their ``nohz_full`` CPUs receiving IPIs and by battery-powered systems that don't want their idle CPUs to be awakened.h](hXSome forms of tracing need to wait for all preemption-disabled regions of code running on any online CPU, including those executed when RCU is not watching. This means that synchronize_rcu() is insufficient, and Tasks Rude RCU must be used instead. This flavor of RCU does its work by forcing a workqueue to be scheduled on each online CPU, hence the “Rude” moniker. And this operation is considered to be quite rude by real-time workloads that don’t want their }(hj@hhhNhNubj)}(h ``nohz_full``h]h nohz_full}(hj@hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj@ubhe CPUs receiving IPIs and by battery-powered systems that don’t want their idle CPUs to be awakened.}(hj@hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhM hj@hhubh)}(hOnce kernel entry/exit and deep-idle functions have been properly tagged ``noinstr``, Tasks RCU can start paying attention to idle tasks (except those that are idle from RCU's perspective) and then Tasks Rude RCU can be removed from the kernel.h](hIOnce kernel entry/exit and deep-idle functions have been properly tagged }(hj@hhhNhNubj)}(h ``noinstr``h]hnoinstr}(hj@hhhNhNubah}(h]h ]h"]h$]h&]uh1jhj@ubh, Tasks RCU can start paying attention to idle tasks (except those that are idle from RCU’s perspective) and then Tasks Rude RCU can be removed from the kernel.}(hj@hhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhM hj@hhubh)}(h}The tasks-rude-RCU API is also reader-marking-free and thus quite compact, consisting solely of synchronize_rcu_tasks_rude().h]h}The tasks-rude-RCU API is also reader-marking-free and thus quite compact, consisting solely of synchronize_rcu_tasks_rude().}(hjAhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM hj@hhubeh}(h]tasks-rude-rcuah ]h"]tasks rude rcuah$]h&]uh1hhj;hhhhhM ubh)}(hhh](h)}(hTasks Trace RCUh]hTasks Trace RCU}(hjAhhhNhNubah}(h]h ]h"]h$]h&]uh1hhjAhhhhhM ubh)}(hXSome forms of tracing need to sleep in readers, but cannot tolerate SRCU's read-side overhead, which includes a full memory barrier in both srcu_read_lock() and srcu_read_unlock(). This need is handled by a Tasks Trace RCU API implemented as thin wrappers around SRCU-fast, which avoids the read-side memory barriers, at least for architectures that apply noinstr to kernel entry/exit code (or that build with ``CONFIG_TASKS_TRACE_RCU_NO_MB=y``).h](hXSome forms of tracing need to sleep in readers, but cannot tolerate SRCU’s read-side overhead, which includes a full memory barrier in both srcu_read_lock() and srcu_read_unlock(). This need is handled by a Tasks Trace RCU API implemented as thin wrappers around SRCU-fast, which avoids the read-side memory barriers, at least for architectures that apply noinstr to kernel entry/exit code (or that build with }(hj+AhhhNhNubj)}(h"``CONFIG_TASKS_TRACE_RCU_NO_MB=y``h]hCONFIG_TASKS_TRACE_RCU_NO_MB=y}(hj3AhhhNhNubah}(h]h ]h"]h$]h&]uh1jhj+Aubh).}(hj+AhhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhM hjAhhubh)}(hX{Now that the implementation is based on SRCU-fast, a call to synchronize_rcu_tasks_trace() implies at least one call to synchronize_rcu(), that is, every Tasks Trace RCU grace period contains at least one plain vanilla RCU grace period. Should there ever be a synchronize_rcu_tasks_trace_expedited(), this guarantee would *not* necessarily apply to this hypothetical API member.h](hXCNow that the implementation is based on SRCU-fast, a call to synchronize_rcu_tasks_trace() implies at least one call to synchronize_rcu(), that is, every Tasks Trace RCU grace period contains at least one plain vanilla RCU grace period. Should there ever be a synchronize_rcu_tasks_trace_expedited(), this guarantee would }(hjKAhhhNhNubj9 )}(h*not*h]hnot}(hjSAhhhNhNubah}(h]h ]h"]h$]h&]uh1j8 hjKAubh3 necessarily apply to this hypothetical API member.}(hjKAhhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhM hjAhhubh)}(hThe tasks-trace-RCU API is also reasonably compact, consisting of rcu_read_lock_trace(), rcu_read_unlock_trace(), rcu_read_lock_trace_held(), call_rcu_tasks_trace(), synchronize_rcu_tasks_trace(), and rcu_barrier_tasks_trace().h]hThe tasks-trace-RCU API is also reasonably compact, consisting of rcu_read_lock_trace(), rcu_read_unlock_trace(), rcu_read_lock_trace_held(), call_rcu_tasks_trace(), synchronize_rcu_tasks_trace(), and rcu_barrier_tasks_trace().}(hjkAhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM hjAhhubeh}(h]jz<ah ]h"]tasks trace rcuah$]h&]uh1hhj;hhhhhM jKubeh}(h]jah ]h"]other rcu flavorsah$]h&]uh1hhhhhhhhM jKubh)}(hhh](h)}(hPossible Future Changesh]hPossible Future Changes}(hjAhhhNhNubah}(h]h ]h"]h$]h&]uh1hhjAhhhhhM ubh)}(hXOne of the tricks that RCU uses to attain update-side scalability is to increase grace-period latency with increasing numbers of CPUs. If this becomes a serious problem, it will be necessary to rework the grace-period state machine so as to avoid the need for the additional latency.h]hXOne of the tricks that RCU uses to attain update-side scalability is to increase grace-period latency with increasing numbers of CPUs. If this becomes a serious problem, it will be necessary to rework the grace-period state machine so as to avoid the need for the additional latency.}(hjAhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM hjAhhubh)}(hX2RCU disables CPU hotplug in a few places, perhaps most notably in the rcu_barrier() operations. If there is a strong reason to use rcu_barrier() in CPU-hotplug notifiers, it will be necessary to avoid disabling CPU hotplug. This would introduce some complexity, so there had better be a *very* good reason.h](hXRCU disables CPU hotplug in a few places, perhaps most notably in the rcu_barrier() operations. If there is a strong reason to use rcu_barrier() in CPU-hotplug notifiers, it will be necessary to avoid disabling CPU hotplug. This would introduce some complexity, so there had better be a }(hjAhhhNhNubj9 )}(h*very*h]hvery}(hjAhhhNhNubah}(h]h ]h"]h$]h&]uh1j8 hjAubh good reason.}(hjAhhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhM hjAhhubh)}(hXThe tradeoff between grace-period latency on the one hand and interruptions of other CPUs on the other hand may need to be re-examined. The desire is of course for zero grace-period latency as well as zero interprocessor interrupts undertaken during an expedited grace period operation. While this ideal is unlikely to be achievable, it is quite possible that further improvements can be made.h]hXThe tradeoff between grace-period latency on the one hand and interruptions of other CPUs on the other hand may need to be re-examined. The desire is of course for zero grace-period latency as well as zero interprocessor interrupts undertaken during an expedited grace period operation. While this ideal is unlikely to be achievable, it is quite possible that further improvements can be made.}(hjAhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM hjAhhubh)}(hXjThe multiprocessor implementations of RCU use a combining tree that groups CPUs so as to reduce lock contention and increase cache locality. However, this combining tree does not spread its memory across NUMA nodes nor does it align the CPU groups with hardware features such as sockets or cores. Such spreading and alignment is currently believed to be unnecessary because the hotpath read-side primitives do not access the combining tree, nor does call_rcu() in the common case. If you believe that your architecture needs such spreading and alignment, then your architecture should also benefit from the ``rcutree.rcu_fanout_leaf`` boot parameter, which can be set to the number of CPUs in a socket, NUMA node, or whatever. If the number of CPUs is too large, use a fraction of the number of CPUs. If the number of CPUs is a large prime number, well, that certainly is an “interesting” architectural choice! More flexible arrangements might be considered, but only if ``rcutree.rcu_fanout_leaf`` has proven inadequate, and only if the inadequacy has been demonstrated by a carefully run and realistic system-level workload.h](hX_The multiprocessor implementations of RCU use a combining tree that groups CPUs so as to reduce lock contention and increase cache locality. However, this combining tree does not spread its memory across NUMA nodes nor does it align the CPU groups with hardware features such as sockets or cores. Such spreading and alignment is currently believed to be unnecessary because the hotpath read-side primitives do not access the combining tree, nor does call_rcu() in the common case. If you believe that your architecture needs such spreading and alignment, then your architecture should also benefit from the v&}(hjAhhhNhNubj)}(h``rcutree.rcu_fanout_leaf``h]hrcutree.rcu_fanout_leaf}(hjAhhhNhNubah}(h]h ]h"]h$]h&]uh1jhjAubhXU boot parameter, which can be set to the number of CPUs in a socket, NUMA node, or whatever. If the number of CPUs is too large, use a fraction of the number of CPUs. If the number of CPUs is a large prime number, well, that certainly is an “interesting” architectural choice! More flexible arrangements might be considered, but only if }(hjAhhhNhNubj)}(h``rcutree.rcu_fanout_leaf``h]hrcutree.rcu_fanout_leaf}(hjAhhhNhNubah}(h]h ]h"]h$]h&]uh1jhjAubh has proven inadequate, and only if the inadequacy has been demonstrated by a carefully run and realistic system-level workload.}(hjAhhhNhNubeh}(h]h ]h"]h$]h&]uh1hhhhM hjAhhubh)}(hPlease note that arrangements that require RCU to remap CPU numbers will require extremely good demonstration of need and full exploration of alternatives.h]hPlease note that arrangements that require RCU to remap CPU numbers will require extremely good demonstration of need and full exploration of alternatives.}(hjBhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM hjAhhubh)}(hXRCU's various kthreads are reasonably recent additions. It is quite likely that adjustments will be required to more gracefully handle extreme loads. It might also be necessary to be able to relate CPU utilization by RCU's kthreads and softirq handlers to the code that instigated this CPU utilization. For example, RCU callback overhead might be charged back to the originating call_rcu() instance, though probably not in production kernels.h]hXRCU’s various kthreads are reasonably recent additions. It is quite likely that adjustments will be required to more gracefully handle extreme loads. It might also be necessary to be able to relate CPU utilization by RCU’s kthreads and softirq handlers to the code that instigated this CPU utilization. For example, RCU callback overhead might be charged back to the originating call_rcu() instance, though probably not in production kernels.}(hjBhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM hjAhhubh)}(hAdditional work may be required to provide reasonable forward-progress guarantees under heavy load for grace periods and for callback invocation.h]hAdditional work may be required to provide reasonable forward-progress guarantees under heavy load for grace periods and for callback invocation.}(hj"BhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM& hjAhhubeh}(h]jah ]h"]possible future changesah$]h&]uh1hhhhhhhhM jKubh)}(hhh](h)}(hSummaryh]hSummary}(hj:BhhhNhNubah}(h]h ]h"]h$]h&]uh1hhj7BhhhhhM+ ubh)}(hThis document has presented more than two decade's worth of RCU requirements. Given that the requirements keep changing, this will not be the last word on this subject, but at least it serves to get an important subset of the requirements set forth.h]hThis document has presented more than two decade’s worth of RCU requirements. Given that the requirements keep changing, this will not be the last word on this subject, but at least it serves to get an important subset of the requirements set forth.}(hjHBhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM- hj7Bhhubeh}(h]jah ]h"]summaryah$]h&]uh1hhhhhhhhM+ jKubh)}(hhh](h)}(hAcknowledgmentsh]hAcknowledgments}(hj`BhhhNhNubah}(h]h ]h"]h$]h&]uh1hhj]BhhhhhM3 ubh)}(hXDI am grateful to Steven Rostedt, Lai Jiangshan, Ingo Molnar, Oleg Nesterov, Borislav Petkov, Peter Zijlstra, Boqun Feng, and Andy Lutomirski for their help in rendering this article human readable, and to Michelle Rankin for her support of this effort. Other contributions are acknowledged in the Linux kernel's git archive.h]hXFI am grateful to Steven Rostedt, Lai Jiangshan, Ingo Molnar, Oleg Nesterov, Borislav Petkov, Peter Zijlstra, Boqun Feng, and Andy Lutomirski for their help in rendering this article human readable, and to Michelle Rankin for her support of this effort. Other contributions are acknowledged in the Linux kernel’s git archive.}(hjnBhhhNhNubah}(h]h ]h"]h$]h&]uh1hhhhM5 hj]Bhhubeh}(h]acknowledgmentsah ]h"]acknowledgmentsah$]h&]uh1hhhhhhhhM3 ubeh}(h]!a-tour-through-rcu-s-requirementsah ]h"]!a tour through rcu's requirementsah$]h&]uh1hhhhhhhhKubeh}(h]h ]h"]h$]h&]sourcehuh1hcurrent_sourceN current_lineNsettingsdocutils.frontendValues)}(hN generatorN datestampN source_linkN source_urlN toc_backlinksj footnote_backlinksK sectnum_xformKstrip_commentsNstrip_elements_with_classesN strip_classesN report_levelK halt_levelKexit_status_levelKdebugNwarning_streamN tracebackinput_encoding utf-8-siginput_encoding_error_handlerstrictoutput_encodingutf-8output_encoding_error_handlerjBerror_encodingutf-8error_encoding_error_handlerbackslashreplace language_codeenrecord_dependenciesNconfigN id_prefixhauto_id_prefixid dump_settingsNdump_internalsNdump_transformsNdump_pseudo_xmlNexpose_internalsNstrict_visitorN_disable_configN_sourcehʌ _destinationN _config_files]7/var/lib/git/docbuild/linux/Documentation/docutils.confafile_insertion_enabled raw_enabledKline_length_limitM'pep_referencesN pep_base_urlhttps://peps.python.org/pep_file_url_templatepep-%04drfc_referencesN rfc_base_url&https://datatracker.ietf.org/doc/html/ tab_widthKtrim_footnote_reference_spacesyntax_highlightlong smart_quotessmartquotes_locales]character_level_inline_markupdoctitle_xform docinfo_xformKsectsubtitle_xform image_loadinglinkembed_stylesheetcloak_email_addressessection_self_linkenvNubreporterNindirect_targets]substitution_defs}=high-quality implementation of C11 memory_order_consume [PDF]jr ssubstitution_names}=high-quality implementation of c11 memory_order_consume [pdf]jBsrefnames}(fundamental requirements]jafundamental non-requirements]japarallelism facts of life]ja&quality-of-implementation requirements]j9alinux kernel complications]j\a!software-engineering requirements]jaother rcu flavors]japossible future changes]jasummary]jagrace-period guarantee]jAapublish/subscribe guarantee]jdamemory-barrier guarantees]ja4rcu primitives guaranteed to execute unconditionally]ja guaranteed read-to-write upgrade]ja=high-quality implementation of c11 memory_order_consume [pdf]]j3 areaders impose minimal ordering]jareaders do not exclude updaters]ja"updaters only wait for old readers]j@a9grace periods don't partition read-side critical sections]jca9read-side critical sections don't partition grace periods]jaspecialization]japerformance and scalability](jj%:eforward progress]ja composability]ja corner cases]j a sleepable rcu](jjj#j#jN.j&<e tasks rcu](j$#j$jH<e configuration]j(afirmware interface]j1(a early boot]jT(ainterrupts and nmis]jw(aloadable modules]j(a hotplug cpu]j(ascheduler and rcu]j(atracing and rcu]j)aaccesses to user memory and rcu]j&)aenergy efficiency](jI)j6e#scheduling-clock interrupts and rcu]jl)amemory efficiency]j)a8performance, scalability, response time, and reliability]j)abottom-half flavor (historical)]j;asched flavor (historical)]j<atasks trace rcu]jk<aurefids}nameids}(jBjBjj j3j0jUjRjwjtjjjjjjPj jsj j jjjjjjjjjj jJj,jVjOjjrjjjgj%jF%jHjcjj!jj[$jj$j j?%j/j'jj;jkj*j(j +j@(j^,jc(j,j(j2.j(jK1j(j/j/j0j0j\0jY0jD1jA1j1j(j1j)j2j5)j3jX)j8j{)j :j)j;j)jAjj=j;j=j<j@jj@j4#jAjAj}Ajz<j4BjjZBjjBj~Bu nametypes}(jBjj3jUjwjjjj j jjjjjjJjVjjjgjF%jcj!j[$j$j?%j'j;j*j +j^,j,j2.jK1j/j0j\0jD1j1j1j2j3j8j :j;jAj=j=j@j@jAj}Aj4BjZBjBuh}(jBhj jj0j*jRjLjtjnjjjjjPjjsjj j jj jjjjjjj jj,jjOjMjrjYjj!j%jjHjjjjPjjfjj!j j^$j/j$jjI%jkj'j(j)j@(j*jc(j+j(ja,j(j,j(j5.j/jx/j0j/jY0j0jA1j_0j(jN1j)j1j5)j1jX)j2j{)j3j)j8j)j :jj;j;j<j<j=jj=j4#j@jAj@jz<jAjjAjj7Bj~Bj]Bu footnote_refs} citation_refs} autofootnotes]autofootnote_refs]symbol_footnotes]symbol_footnote_refs] footnotes] citations]autofootnote_startKsymbol_footnote_startK id_counter collectionsCounter}Rparse_messages]transform_messages] transformerN include_log] decorationNhhub.