| Age | Commit message (Collapse) | Author | Files | Lines |
|---|
|
The initial multifs support for EFI was saving all device handles which
are logical partitions regardless which disk they belong to, so indexing
disks other than 0 (or even partitions outside disk 0) would not work.
This patch fixes enumeration of all logical partitions per disk thus
making possible to index both disk and partition correctly.
Signed-off-by: Paulo Alcantara <pcacjr@zytor.com>
|
|
Signed-off-by: Raphael S. Carvalho <raphaelsc@cloudius-systems.com>
Signed-off-by: Paulo Alcantara <pcacjr@zytor.com>
|
|
The get_dev_info_priv() function cannot return a reference of a static
variable since there might be multiple device handles and each one per
logical partition. A typical example of this failure is when a
configuration file contains multifs-path-syntax like, so multifs will
initialised to the first partition found (including its device handle)
and the other ones will be using the same partition to do I/O.
Cc: Gene Cumm <gene.cumm@gmail.com>
Signed-off-by: Paulo Alcantara <pcacjr@zytor.com>
|
|
The previous multifs implementation only supported BIOS, so this patch
basically keeps common code in a single place and separate
firmware-specific functions that'll be exposed later.
A generic interface has been added to be able to switch among different
FSes regardless which firmware is being used. Although, each
implementation has to implement init() and get_is_info() functions, pass
through multifs_ops structure and they'll get called in starting of
ldlinux.
Cc: Gene Cumm <gene.cumm@gmail.com>
Signed-off-by: Paulo Alcantara <pcacjr@zytor.com>
|
|
By looking for PATH entries and trying to find multifs-like paths does
not make any sense. Only try to open it once.
Cc: Gene Cumm <gene.cumm@gmail.com>
Signed-off-by: Paulo Alcantara <pcacjr@zytor.com>
|
|
The system firmware will expose device handles of all MBR/GPT partitions
which support BlockIo and/or DiskIo protocols, so find them and pass
through multifs driver.
Even in EFI environment ldlinux will be loaded up, so to avoid
initialise multifs twice call init_multifs() earlier.
Cc: Gene Cumm <gene.cumm@gmail.com>
Signed-off-by: Paulo Alcantara <pcacjr@zytor.com>
|
|
The initramfs path in the configuration file might come with multifs
path syntax (e.g. (hdX,Y)/initramfs...), so we must include it so
multifs will know how fs ops to switch to.
Cc: Gene Cumm <gene.cumm@gmail.com>
Signed-off-by: Paulo Alcantara <pcacjr@zytor.com>
|
|
This patch finishes the multifs support.
init_multifs gets called in the main (startup) function of ldlinux.c32,
so multifs will be initialized automatically.
init_multifs calls enable_multifs (lives in the core) to hook get_fs_info.
Subsequent accesses will callback the get_fs_info living in ldlinux.c32.
Cc: Gene Cumm <gene.cumm@gmail.com>
Signed-off-by: Raphael S. Carvalho <raphael.scarv@gmail.com>
Signed-off-by: Paulo Alcantara <pcacjr@zytor.com>
|
|
multifs syntax:
(hd[disk number],[partition number])/path/to/file
The meaning of this_fs was changed to improve the flexibility of the support.
Now, this_fs means the file system being currently used.
root_fs was created to save the context of the main file system (where ldlinux.sys lives in).
get_fs_info is a function pointer that will be later hooked to a function from ldlinux.c32.
get_fs_info is expected to return a fs_info structure given the multifs path.
Cc: Gene Cumm <gene.cumm@gmail.com>
Signed-off-by: Raphael S. Carvalho <raphael.scarv@gmail.com>
Signed-off-by: Paulo Alcantara <pcacjr@zytor.com>
|
|
multifs depends on the availability of partiter to find a partition.
Cc: Gene Cumm <gene.cumm@gmail.com>
Signed-off-by: Raphael S. Carvalho <raphael.scarv@gmail.com>
Signed-off-by: Paulo Alcantara <pcacjr@zytor.com>
|
|
'git://github.com/Celelibi/syslinux.git/fix/bios/tail-call-stack-overflows'
FIXME: this circumvents some of the issue but is not the cleanest and may
reoccur.
|
|
"may be used uninitialized in this function [-Wuninitialized]" even though logic dictates
it must be OK.
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
|
|
The current btrfs support did not handled multiple stripes stored in
chunk items, hence skipping the physical addresses that were needed to
do the mapping.
Besides, the chunk tree may contain DEV_ITEM keys which store
information on all of the underlying block devices, so we must skip them
instead of finishing lookup.
The bug was reproduced with btrfs-progs v4.2.2.
Cc: Gene Cumm <gene.cumm@gmail.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Signed-off-by: Paulo Alcantara <pcacjr@zytor.com>
---
v1 -> v2:
* Do not set ignore_key multiple times. Set it before parsing chunk
tree.
v2 -> v3:
* Replace an unnecessary goto with a continue statement.
|
|
Besides supporting newer version of xfs file system, this patch also
does some code refactoring and fix completely broken listing and
searching on v2-3 node directories.
Cc: Gene Cumm <gene.cumm@gmail.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Raphael S. Carvalho <raphael.scarv@gmail.com>
Cc: Ady <ady-sf@hotmail.com>
Signed-off-by: Paulo Alcantara <pcacjr@zytor.com>
|
|
Previous commit changed the dependency filenames to not have a leading '.'
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
We can let compiler driver pass the right options to preprocessor after
processing -Mxy options, right now its bypassing the gcc driver and
handing them straight to cpp
This also helps in other compilers processing these options correctly
for their preprocessors consumption
Signed-off-by: Khem Raj <raj.khem@gmail.com>
|
|
This patch is coming after some bad experience with gfxboot.
GFXboot was padding unexpected options on the chain.c32 commande line.
As a result, chain was confused and didn't provided the expected
behavior.
This patch is a workaround to enforce chain.c32 not considering anything
on the command line after this keyword.
Example:
chain.c32 hd0 exit ...garbage...
|
|
When parsing the modules.alias, we shall report it as broken if we
cannot find any valid line in it.
This patch simply count the number of valid lines and report a missing
modules.alias if no valid lines are found.
|
|
This option is mandatory when compiling for EFI as some event handlers
may interupt the running code and use the space just above the reserved
stack space.
Signed-off-by: Sylvain Gault <sylvain.gault@gmail.com>
|
|
This variable indicates whether or nor the files are compiled for EFI.
The lack of it lead the Makefiles to forget to add some compilation
options specific to EFI.
Signed-off-by: Sylvain Gault <sylvain.gault@gmail.com>
|
|
Merge btrfs_read_adv and xfs_read_adv into a single generic function
ext_read_adv and split ext_write_adv_offset out of ext_write_adv.
Use those new functions in rewrite_boot_image and btrfs_install_file
where it is actually hardcoded.
Signed-off-by: Nicolas Cornu <nicolac76@yahoo.fr>
Signed-off-by: Paulo Alcantara <pcacjr@zytor.com>
|
|
Spaces, ASCII 0x20, have been
* moved
* inserted
* deleted
Signed-off-by: Geert Stappers <stappers@stappers.nl>
|
|
As of commit ID 37eef640 (before 6.03-pre12, before 6.03), it is now in
Boot Area B.
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
Some typo fixes in extlinux/main.c.
None of them affects code execution.
Signed-off-by: Geert Stappers <stappers@stappers.nl>
|
|
Due to size constraints on DOS systems, do not include whole syslxcom
into DOS-based installer for using xpread() and xpwrite() functions,
instead make them part of another separate library and include it only.
Signed-off-by: Paulo Alcantara <pcacjr@zytor.com>
|
|
Signed-off-by: Nicolas Cornu <ncornu@aldebaran.com>
Signed-off-by: Paulo Alcantara <pcacjr@zytor.com>
|
|
xfs_is_valid_magicnum is not actually a generic function that checks for
magic numbers, instead it checks only for superblock's one.
Signed-off-by: Paulo Alcantara <pcacjr@zytor.com>
Reviewed-by: Gene Cumm <gene.cumm@gmail.com>
|
|
For the compression / decompression to succeed, the sections layout must
be the same between the virtual memory and load memory. The section
alignment was kept in sync by introducing aligment that should be
greater or equal to the actual section alignment.
This patch compute the load memory addresses of the sections so that
the layout is the same as the virtual memory addresses.
Signed-off-by: Sylvain Gault <sylvain.gault@gmail.com>
Tested-by: poma <pomidorabelisima@gmail.com>
Signed-off-by: Paulo Alcantara <pcacjr@zytor.com>
|
|
The section aligment specified in the ld scripts have to be greater or
equal to those in the .o files generated by gcc.
Signed-off-by: Sylvain Gault <sylvain.gault@gmail.com>
Tested-by: poma <pomidorabelisima@gmail.com>
Signed-off-by: Paulo Alcantara <pcacjr@zytor.com>
|
|
The command-line parsing used to return a pointer to a local array. The
code used to work by chance, but now, gcc 5 is able to detect it and
return a NULL pointer instead.
The buffer is now marked static. This shouldn't be a problem as only one
command line can be read at a time.
Signed-off-by: Sylvain Gault <sylvain.gault@gmail.com>
Tested-by: poma <pomidorabelisima at gmail.com>
Signed-off-by: Paulo Alcantara <pcacjr@zytor.com>
|
|
Some machines don't like UseDefaultAddress; fallback to manually
configuring. HP EFI servers like DL160 Gen9 are affected.
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
The behavior when running a "CONFIG" command line is to reload
ldlinux.c32 with the new file as argument. This call never return.
In order to avoid stacking up the calls to start_ldlinux, this patch
introduce a setjmp/longjmp to return to the first call to start_ldlinux,
thus freeing all the stack space.
Signed-off-by: Sylvain Gault <sylvain.gault@gmail.com>
|
|
When a COM32 module exits, the functions never return and a new call to
ldlinux_enter_command is made. This could fill the stack and overflow on
some data present in memory.
This patch use setjmp/longjmp to return to the main function and restart
from there when a COM32 module exits.
Signed-off-by: Sylvain Gault <sylvain.gault@gmail.com>
|
|
Wrong translation type used.
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
mboot.c32 lacks support for loading ELF64 binaries (supported by GRUB2,
for example). This is a trivial patch to add such feature.
[reflow some changes to fit ~80 columns; remove trailing whitespace
- gene.cumm@gmail.com]
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
HTTP/1.1 header Host must contain the port number if not default for the
protocol. Host isn't a part of HTTP/1.0 but let's implement it right.
Reported-By: Michael DeCandia <michael.decandia@gmail.com>
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
Apparently some servers don't bother setting siaddr when pointing to
itself for TFTP. Re-allow 54 but always set IPInfo.serverip from
siaddr/dhcp->sip in packet #3 (PXEReply/proxyDHCP).
Always set from siaddr if good in case parsing after-DHCP options.
Reported-by: Celelibi <celelibi@gmail.com>
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
Check if the first character matches the character to replace, rather
than skipping it and starting with the second.
Signed-off-by: Josh Triplett <josh@joshtriplett.org>
|
|
devname is put on heap for all cases to avoid memory
leak, and ease of use in future as well
Signed-off-by: Imran Zaman <imran.zaman@intel.com>
Reviewed-by: Paulo Alcantara <pcacjr@zytor.com>
|
|
gpxe/ has a lot of binary blobs after a spotless. Start the cleanup
measures. do-spotless-gpxe for bios target might be better in another
spot.
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
When adding lwIP functionality, the gPXE/iPXE callback was broken. This
prevented pxelinux.0 from calling gPXE/iPXE for HTTP and FTP URLs. Re-add
for pxelinux.0 and add code to find file size.
Move to core/legacynet/core.c to access packet_buf and leave a dummy
function for lpxelinux.0.
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
When multiple architectures are on a single media, all will try to use the
same config, preventing PATH from pointing to an architecture-specific
directory.
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
Certain modules directly make BIOS calls or call library functions that
will not be ported from BIOS (like syslinux_shuffle_boot_rm()).
It appears dir.c32 was long-forgotten for the Makefile
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
OBJ directory creation changed from a .PHONY target to a real target
used through an order-only dependency.
A target depending on another target marked .PHONY is always rebuilt,
thus forcing all the .o files to be rebuilt everytime.
Signed-off-by: Sylvain Gault <sylvain.gault@gmail.com>
Reviewed-by: Paulo Alcantara <pcacjr@zytor.com>
|
|
Originally-By: Patrick Masotta <masottaus@yahoo.com>
[Use (void(0)) instead of a dummy function - gene.cumm@gmail.com]
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
Missing */ at the end of a comment.
Signed-off-by: Sylvain Gault <sylvain.gault@gmail.com>
Reviewed-by: Paulo Alcantara <pcacjr@zytor.com>
|
|
Server Identifier is NEVER the next server where Syslinux was loaded from.
BOOTP field siaddr, BOOTP field sname and DHCP option 66 TFTP server name
may contain this information. For now, just use siaddr.
|
|
Merge git://github.com/Celelibi/syslinux.git branch 'fix/efi/cleanup'
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
As per bug #4, we should report a clear failure if the target path is
not writable.
The current code was catching the fact the file was not writable
but it didn't wrote an explicit message and even more confusing, was
trying to process the file descriptor leading to a creepy "Bad file
descriptor" error message.
This patch does return EACCES to avoid process the file description.
It also print an explicit message saying a particular file isn't
writable which generates a fatal error.
A typical output looks like :
[root@host]: extlinux --once=plop /boot
/boot is device /dev/sda
Cannot open file '/boot/ldlinux.sys' in read/write mode !
Fatal error, exiting.
|
|
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
Correct the order of the parameters for MENU RESOLUTION.
Update references.
Rewording.
Typos.
[certain phrases seem clearer in original wording - gene.cumm@gmail.com]
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
Syslinux 6.03 (efi64) fails to boot a 32-bit kernel. The way Syslinux
leaves long mode in kernel_jump assembly routine does not follow AMD64
specifications. More precisely:
1. After setting a new GADT, `cs` has to be refresh by doing a long
jump, but it is not
2. Other segments have to be updated, but they are not
3. Disabling paging has to be done before disabling long mode, but the
implementation does the opposite
In most cases, a computer that tries to execute the kernel_jump routine
reboot (it can also hangs). This patch fixes the kernel_jump routine.
Signed-off-by: Thomas Letan <thomas.letan@ssi.gouv.fr>
Tested-by: Patrick Masotta <masottaus@yahoo.com>
Tested-by: Celelibi <celelibi@gmail.com>
|
|
The libupload was using the pxe_call() directly for doing the tftp
uploading stuff. This was only working with pxelinux.
Since we do have lpxelinux, the libupload should use the core functions
to get rid of thoses direct PXE calls.
This patch does
- add a tftp_put() function which supports core functions.
- implement the call from libupload making the code much more simplier
As a result {l}pxelinux can upload data to a tftp server is a similar
way. HDT is getting the benefit of such code.
|
|
When debugging, it is very useful to get the ip adress of the server
that reponsded to that packet.
|
|
The current code was making a mistake by searching the -err element in
the array as it was resulting into a negative value.
The current patch does print the error number which could be useful but
also retrieve the proper string to indicate the real error we
encountered.
|
|
When uploading the dump file to a tftp server, removing the plus & comma
sign from the filename could make the file easier to manipulate later.
|
|
When debugging some syslinux code with debugging enabled, the threading
code is so verbose that it completly slow down syslinux but also is so
verbose that any other trace is invisible in the flood.
This commit aims at requesting people to explicitly enable the
threading logs by using -DDEBUG_THREAD.
|
|
This is a simple rebase of the current code against current dmidecode
version. This is pretty lame copy/paste but as we don't have a
libification of dmidecode, that does the job.
This patch does improve mostly the cpu & ram reporting.
|
|
Since commit 1697594b61f9a8f9d092996afc0e2c80bbb2a20a, some commands are
said to be "nomodule" like "say".
This patch was adding a check if the nomodule flag was set but didn't
checked that the associated structure did exist leading to false
positive detection.
As a result, the commands were not executed meaning the CLI was unsuable
since ... 3 years.... *shame*
This commit simply avoid considering the nomodule flag if the structure
is not allocated
|
|
argv shall be passed as a pointer to avoid a crash when running
command like "dmi".
Thanks genec for pointing this out.
|
|
When dynamic debug is engaged, the output is pretty flooded by
malloc/free messages while you are looking at other traces.
As devel.mk have a DEBUG_MALLOC option, it seems pretty logical to
enable the malloc/free dprintf() only if this option is engaged.
That patch make the dynamic debug output less floody while letting the
choice to get the malloc/free dprintf() messages.
|
|
PCs without keyboards may hang as they might return a status other than
EFI_NOT_READY.
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
Memory allocation and copy length should be SECTOR * size for sectors
operation.
Signed-off-by: Robert <luyao-c@360.cn>
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
Signed-off-by: Sylvain Gault <sylvain.gault@gmail.com>
|
|
The assignment looks suspicious but is actually legit since it is
protected by the type check.
Signed-off-by: Sylvain Gault <sylvain.gault@gmail.com>
|
|
The function efi_get_MAC was given a pointer to array instead of a
simple pointer, generating a warning with gcc.
Signed-off-by: Sylvain Gault <sylvain.gault@gmail.com>
|
|
If LibLocateHandle() returns success and either 0 handles or we find no
matching handles, treat it the same.
Originally-By: Patrick Masotta <masottaus@yahoo.com>
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
Extracts a MAC address from a device path
Originally-By: Patrick Masotta <masottaus@yahoo.com>
[gene.cumm@gmail.com: Respace]
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
More consistent. Also ensures it's properly recast.
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
Add 0x7f as a space character
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
The intention is to nullify not copy the IRQ pointers
Reported-By: Thomas Schmitt <scdbackup@gmx.net>
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
Unnecessary. On some systems, opens NIC in promiscuous mode or
breaks UseDefaultAddress.
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
There are more unsuccessful returns that are ignored. Restructure to
allow for trapping other values.
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
This clause closes the protocol before creating the child
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
Store and reuse handle found with EFI_LOADED_IMAGE_PROTOCOL for
EFI_PXE_BASE_CODE_PROTOCOL and EFI_UDP4_SERVICE_BINDING_PROTOCOL
This caused machines with multiple NICs to not reliably attach to the
correct NIC handle.
gnu-efi LoadedImageProtocol PxeBaseCodeProtocol Udp4ServiceBindingProtocol
Reported-By: Holger Baust <holger.baust@freenet.ag>
Reported-By: Michael Glasgow <glasgow@beer.net>
Reported-By: Da Shi Cao <dscao999@gmail.com>
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
Filter (by setting a minimum option number) the options based on pkt_type.
DHCPDiscover/PXEReply should only contain info about client ID, boot
server and PXELINUX options.
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
This prevented a proxyDHCP/PXEReply from overriding the IPInfo.serverip,
most often stored in BOOTP field siaddr or DHCP option 54.
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
Add packet type so we can eventually only grab certain data elements from
the DHCP packets appropriately
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
parse_dhcp() does this also, seeming to set a bad value for some clients.
Fixes some deaf clients.
Signed-off-by: Patrick Masotta <masottaus@yahoo.com>
[gene.cumm@gmail.com: Just move existing code]
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
EFI allows us to have access to more DHCP-related packets. Choose the
packet with the highest priority for the third to parse.
Signed-off-by: Patrick Masotta <masottaus@yahoo.com>
[gene.cumm@gmail.com: Only change the third parsing iteration;
reflow to better fit code standards; don't reparse DhcpAck]
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
For the SERIAL directive, allow octal (0###) and hexidecimal (0x###)
notation on the port and flow control arguments.
Two of the old ASM getint/parseint callers
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
A lot more items have identical dependencies; regroup. Certain items make
more sense to build first; resequence. Group library dependencies before
COM32 directories
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
The code was passing sizeof(const char *) - not the length of the string
or the size of the buffer - as the length to strncmp(). These two cases
don't need the length, so switch to strcmp.
Fixes the warning: argument to 'sizeof' in 'strncmp' call is the same
expression as the second source; did you mean to provide an explicit
length?
Signed-off-by: Jonathan Boeing <jonathan.n.boeing@gmail.com>
|
|
The code was passing sizeof(char *) - not the length of the buffer - to
memset. Change the function to take the length of the buffer as a
parameter.
Fixes the warning: argument to 'sizeof' in 'memset' call is the same
expression as the destination; did you mean to provide an explicit
length?
Signed-off-by: Jonathan Boeing <jonathan.n.boeing@gmail.com>
|
|
The code was passing sizeof(char *) - not the length of the buffer - to
memset. Change the function to take the length of the buffer as a
parameter.
Fixes the warning: argument to 'sizeof' in 'memset' call is the same
expression as the destination; did you mean to provide an explicit
length?
Signed-off-by: Jonathan Boeing <jonathan.n.boeing@gmail.com>
|
|
The code was passing sizeof(char *) - not the length of the buffer - to
strlcpy and snprintf. Change the function to take the length of the
buffer as a parameter.
Fixes the warning: argument to 'sizeof' in 'snprintf' call is the same
expression as the destination; did you mean to provide an explicit
length?
Signed-off-by: Jonathan Boeing <jonathan.n.boeing@gmail.com>
|
|
When building efi64, there's a size mismatch between the uint32_t
variable (32 bits) and the -1UL (64 bits). This fixes the warning:
large integer implicitly truncated to unsigned type
Signed-off-by: Jonathan Boeing <jonathan.n.boeing@gmail.com>
|
|
Fixes the warning: format '%x' expects argument of type 'unsigned int', but
argument 2 has type 'size_t'
Signed-off-by: Jonathan Boeing <jonathan.n.boeing@gmail.com>
|
|
Signed-off-by: Jonathan Boeing <jonathan.n.boeing@gmail.com>
|
|
Signed-off-by: Jonathan Boeing <jonathan.n.boeing@gmail.com>
|
|
If the kernel is relocatable and the protected mode code will not fit
in the initially determined location, that code will be moved to the
next available location. However, beginning with commit 8f470e7b, the
code is moved to the initially determined location instead of the next
available location because prot_mode_base is no longer updated to the
correct location. Since whdr->code32_start is updated, it is pointing
to the wrong execution start location, random code is executed and
the machine is rebooted.
Restore the old behavior by assigning prot_mode_base the value of
base. Tested on a machine that exposed this behavior.
Signed-off-by: Scot Doyle <lkml14@scotdoyle.com>
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
Signed-off-by: Michal Soltys <soltys@ziu.info>
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
This can be useful to force boot even if checksums of GPT header and/or
partition list are invalid. This works independently from 'strict'
option.
Signed-off-by: Michal Soltys <soltys@ziu.info>
|
|
Instead of calling it just once after the examination of both headers
(the 2nd one only if necessary) - call it per each header during
examination.
Thus in unlikely situation of passing crc check on primary gpt header
and at the same time failing sanity checks, it will also try backup
header.
Signed-off-by: Michal Soltys <soltys@ziu.info>
|
|
It's relatively simple to just apply the geodsp1s_f.bin force drive 80h
after applying the plain geodsp1s.img.xz
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
Remove 1 spelling error
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
Move library dependendcies to a variable; Add another _f (Force DL == 80h)
variant.
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
Print drive number at start. Print '!' before data when an error is
encountered (but continue). Several macros to change codesize. Code
refactoring to reduce size.
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
mk-lba-img.pl misdirects a status message about len/tail such that it hits
stdout (which may be our file) instead of stderr.
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
Correct base's type to match its initialization from prot_mode_base and
passage to syslinux_memmap_find(). Tested with extlinux.
Signed-off-by: Scot Doyle <lkml14@scotdoyle.com>
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
remove mention of the deprecated "-o" option
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
Correct typos.
Improve format for asciidoc-to-manpage conversion.
Wrap text.
Correct info.
Help phase-out deprecated and/or conflicting options.
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
The build would be failed and exit if build-gnu-efi.sh is failed, so
we'd better not suppress the output of build-gnu-efi.sh, which is good
for debugging.
Signed-off-by: Robert Yang <liezhi.yang@windriver.com>
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
efi/clean-gnu-efi.sh presumes it's in a git repo. Test first.
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
efi/build-gnu-efi.sh presumes it's in a git repo. Test first.
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
Reflow for proper AsciiDoc formatting.
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
Use the z size specifier to printf-ing size_t variables to get rid of gcc
warning format ‘%08x’ expects type ‘unsigned int’, but argument 2 has
type ‘long unsigned int’
Author: Martin Str|mberg <ams@ludd.ltu.se>
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
The description of SYSAPPEND for the DMI information states that the spaces
are replaced by underscores, but this replacement does not occur in 6.03.
Signed-off-by: Dany St-Amant <dany.ephemeral.2014@icloud.com>
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
It has been observed that binaries contents
are depending on the order of linked objects.
This order is caused by GNU make's wildcard function
and the position of sources on filesystem.
This change tries to prevent this kind of randomness.
Also consider building using -j1 flag
to make it even more reproductible.
Change-Id: Ie8eee7f336e6f1fa2863c4150d967afd15519f1d
Bug: http://bugzilla.syslinux.org/show_bug.cgi?id=57#related
Signed-off-by: Philippe Coval <philippe.coval at open.eurogiciel.org>
|
|
Get rid of printf format warning format ‘%llx’ expects type
‘long long unsigned int’, but argument 2 has type ‘uint64_t’
Use <inttypes.h> PRIx64 instead of llx to get rid of gcc warning
format ‘%llx’ expects type ‘long long unsigned int’, but argument 2
has type ‘uint64_t’
Author: Martin Str|mberg <ams@ludd.ltu.se>
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
Resort and collapse the list; add libutil to dep lib
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
From: Chong Lu <Chong.Lu at windriver.com>
There might be an error when parallel build:
[snip]
cp: cannot create directory `tmp/sysroots/x86_64-linux/usr/share/
syslinux/com32/include/gplinclude': No such file or directory
make[4]: *** [install] Error 1
make[3]: *** [gpllib] Error 2
[snip]
This is a potential issue. In ${S}/com32/gpllib/Makefile file,
install target wants to copy $(SRC)/../gplinclude to
$(INSTALLROOT)$(COM32DIR)/include/ directory, but in ${S}/com32/lib/Makefile
file, the install target will remove $(INSTALLROOT)$(COM32DIR)/include
directory. We need to do com32/lib first.
The patch make com32/gpllib depends on com32/lib to fix this issue.
Upstream-Status: Pending
Signed-off-by: Chong Lu <Chong.Lu at windriver.com>
Signed-off-by: Robert Yang <liezhi.yang at windriver.com>
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
Fixed:
cp -r syslinux-6.01/com32/libupload/*.h image/usr/share/syslinux/com32/include/
[snip]
rm -rf image/usr/share/syslinux/com32/include
[snip]
cp: cannot create regular file `image/usr/share/syslinux/com32/include/serial.h': No such file or directory
The cp is happened in the "libupload" dir, while "rm -fr" is happend in
"lib" dir, let "libupload" depend "lib" will fix the problem.
Signed-off-by: Robert Yang <liezhi.yang at windriver.com>
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
Accidentally misdocumented.
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
Document the change introduced in 386b59e1 and extended in 7c5efd0
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
gcc complains about the size of the constant of BTRFS_MAGIC_N. It is a
64bit value, so it needs the ULL suffix.
Signed-off-by: MartinS <ams@luminous.ludd.ltu.se>
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
Begin 6.04 in 2015.
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
Signed-off-by: Ferenc Wágner <wferi@niif.hu>
|
|
Signed-off-by: Ferenc Wágner <wferi@niif.hu>
|
|
Signed-off-by: Ferenc Wágner <wferi@niif.hu>
|
|
Signed-off-by: Ferenc Wágner <wferi@niif.hu>
|
|
Signed-off-by: Ferenc Wágner <wferi@niif.hu>
|
|
Signed-off-by: Ferenc Wágner <wferi@niif.hu>
|
|
Signed-off-by: Ferenc Wágner <wferi@niif.hu>
|
|
Signed-off-by: Ferenc Wágner <wferi@niif.hu>
|
|
Not expecting them does not cause any problems, they are ignored by Lua anyway.
Signed-off-by: Ferenc Wágner <wferi@niif.hu>
|
|
This also enables manual unloading of files by callig f:__gc().
Signed-off-by: Ferenc Wágner <wferi@niif.hu>
|
|
Signed-off-by: Ferenc Wágner <wferi@niif.hu>
|
|
Light userdata don't have individual metatables, thus they aren't
suitable for type checking. Also, a200ad6d replaced relying on reused
memory (from luaL_checkstring) with writing to uninitialized pointers;
copy the filename instead into freshly allocated memory.
Signed-off-by: Ferenc Wágner <wferi@niif.hu>
|
|
Signed-off-by: Ferenc Wágner <wferi@niif.hu>
|
|
Signed-off-by: Ferenc Wágner <wferi@niif.hu>
|
|
Signed-off-by: Ferenc Wágner <wferi@niif.hu>
|
|
Signed-off-by: Ferenc Wágner <wferi@niif.hu>
|
|
Signed-off-by: Ferenc Wágner <wferi@niif.hu>
|
|
Signed-off-by: Ferenc Wágner <wferi@niif.hu>
|
|
Signed-off-by: Ferenc Wágner <wferi@niif.hu>
|
|
Signed-off-by: Ferenc Wágner <wferi@niif.hu>
|
|
Signed-off-by: Ferenc Wágner <wferi@niif.hu>
|
|
Signed-off-by: Ferenc Wágner <wferi@niif.hu>
|
|
Signed-off-by: Ferenc Wágner <wferi@niif.hu>
|
|
Signed-off-by: Ferenc Wágner <wferi@niif.hu>
|
|
Signed-off-by: Ferenc Wágner <wferi@niif.hu>
|
|
Thanks to Ady <ady-sf@hotmail.com> for most of this writeup.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
|
|
_BSD_SOURCE was deprecated in favour of _DEFAULT_SOURCE since glibc
2.20[1]. To avoid build warning on glibc2.20, _DEFAULT_SOURCE should
also be defined.
[1]: https://sourceware.org/glibc/wiki/Release/2.20
Signed-off-by: Chanho Park <chanho61.park@samsung.com>
_BSD_SOURCE was fist used at 1769d57c94d7a965168b72b6fd8d48251710b452
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
In gpt header, lba_cur and lba_alt alternate depending on whether we
read primary or backup copy.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
|
|
This provides more fine grained control than single relax flag. to cover
case with wrong disk sizes.
relax and nostrict are equivalent to strict=0
norelax and strict are equivalent to strict=2
strict=1 does the same as strict=2, but ignores checks against disk size
The current default is strict=1. Options: 'fixchs', '[un]hide[all]' and
'save' will forcibly enable strict=2 (can be overridden by the user).
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
|
|
As partiter doesn't deallocate itself after finish (anymore), it should
be deleted after each loop iteration.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
|
|
Use <0 for errors, 0 for normal state, and >0 for clean completion.
In future this would be necessary if it's decided to make partiter
a generic lib (similar to disklib) - though it has to be quieted first
and provide strerr()-like functionality in place of its verbosity.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
|
|
This patch fixes non-NULL value returned from pi_begin() in case of an
error (which resulted in further hang instead of proper exit). Probably
a leftover from pi_dealloc() times.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
|
|
Previous version had some issues in case of error situations, among
those:
- backup gpt header was not read if reading of primary failed
- alternating nature of lba_cur and lba_alt was ignored
This patch fixes this and extends the gpt handling.
The current behavior is:
- try primary header; if unreadable or checksum fails (or sanity checks
fail unless relax flag is set) - try secondary header
- try main partition table referenced in the header that was read; if
it's unreadable or if its crc check fails - try alternative (so if we
managed to read primary header, table at the end would be considered
as alternative - if we managed to read secondary header, table at the
beginning would be treated as such)
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
|
|
Like 2fe3a7bd, certain Dell machines (Latitude E6510) state interrupts
should work but effectively don't. Force polling.
Reported-by:
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
When saving/restoring the flags around the stack switch test, we
cannot leave live data on the stack *across* the stack switch that we
are going to use on the other side (unlike the "big" stack frame which
we only use once we are back on the original stack.)
Use register BP, which is not live at either point, as a temporary
holding place for the flags from the stack.
Reported-by: Frank Mehnert <frank.mehnert@googlemail.com>
Link: http://bugzilla.syslinux.org/show_bug.cgi?id=54
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
|
|
Like 2fe3a7bd, certain Dell machines (OptiPlex 990) state interrupts
should work but effectively don't. Force polling.
Reported-by: Alexander Perlis <aperlis@math.lsu.edu>
Signed-off-by: Gene Cumm <gene.cumm@gmail.com>
|
|
Use version 2.07 of the LZO compression library.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
|
|
This adds checks for Watcom C DOS cross-compiler in a manner analogous to mingw
compiler for Windows. The compiler is not commonly present in Linux distributions
due to a non-free license.
Signed-off-by: Lubomir Rintel <lkundrak@v3.sk>
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
|
|
This change enables the single-letter options -u, -m, -b as promised
by the help text.
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
|
|
commit 415d571 ("adv: Remove double defintion") introduced a regression
for the EFI boot loader. The commit is correct, but should have tagged
the __syslinux_adv_ptr and __syslinux_adv_size data objects in the EFI
core as __export.
This change is required because symbols in the EFI code are not global
by default, and so, when ldlinux.{e32,e64} tries to bind to the adv
symbols it fails.
The following demonstrates the ELF symbol changes,
Before:
659: 0000000000034f40 8 OBJECT LOCAL DEFAULT 16 __syslinux_adv_ptr
After:
957: 0000000000036fe0 8 OBJECT GLOBAL DEFAULT 16 __syslinux_adv_ptr
Signed-off-by: Matt Fleming <matt.fleming@intel.com>
|
|
Rather than open-coding the content of UTF-16LE strings, just add a
simple function to convert ASCII (only) strings to UTF-16LE.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
|
|
standalone
Although isohybrid.c is supposed to be a companion of the local SYSLINUX
installation, there may be situations where the file isolinux.bin and the
matching MBR template do not stem directly from such an installation.
This change adds an option --mbr, which allows to load an MBR template
file. This may be an isohdp[fp]x*.bin MBR template from the local SYSLINUX
installation, or the first 512 bytes of the isohybrid-capable ISO image
from which isolinux.bin and the other ISOLINUX files are taken.
If macro ISOHYBRID_C_STANDALONE is defined, then the hardcoded MBR templates
are not accessible and isohdpfx.o is not needed at compile time. In this
case, option --mbr becomes mandatory.
I used this for testing my changes with Fedora-Live-Desktop-x86_64-20-1.iso.
isohybrid.c is then compilable without further components of ISOLINUX by:
cc -DISOHYBRID_C_STANDALONE -Wall -o isohybrid isohybrid.c -luuid
Test run:
cp Fedora-Live-Desktop-x86_64-20-1.iso \
Fedora-Live-Desktop-x86_64-20-1-rehybrid.iso
valgrind ./isohybrid --uefi --mac \
--mbr Fedora-Live-Desktop-x86_64-20-1.mbr \
Fedora-Live-Desktop-x86_64-20-1-rehybrid.iso
yields:
==13828== ERROR SUMMARY: 0 errors from 0 contexts (suppressed: 2 from 1)
...
==13828== LEAK SUMMARY:
==13828== definitely lost: 2,048 bytes in 1 blocks.
(Not that valgrind would have detected the memcpy() abuse of patch 001.
But at least i seem to have not introduced more obvious sins.)
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
|
|
It seems unwise to offer future programmers fseek(3) calls for copy+paste.
They are simply insufficient for large image files.
This change switches all calls of fseek(3) to fseeko(3) and takes care
that the offset value if of type off_t.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
|
|
The GPT backup header block should start 512 bytes before the end of the
image file (resp. end of the disk device). This block and the backup GPT
array were wrongly written 512 bytes too early.
This change brings the backup GPT at its correct position.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
|
|
The GPT partition 1 covers the whole ISO filesystem size. GPT specs
demand that the partition end block number shall be the last valid
block in the partition. isohybrid.c rather wrote the number of the
first block after the partition end.
This change reduces the number by 1.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
|
|
The block counts in the APM partitions assumed 512 bytes per block,
whereas the start block numbers assume 2048 as announced in the APM header.
This change divides the affected block counts by 4 to correct the assumption.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
|
|
The worst sin of isohybrid.c was to compose GPT partition names by
8-bit characters and to memcpy() them as if they were 16 bit wide.
GPT names are encoded as UTF-16LE. It is trivial to create this encoding
from 7-bit ASCII.
This change introduces two byte arrays with the desired UTF-16LE names
which replace the string constants "ISOHybrid ISO" and "ISOHybrid".
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
|
|
On general principles, define _FILE_OFFSET_BITS in the file itself
(even though it is also in the Makefile.)
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
|
|
When call isohybrid with option '-u', it overflows on a 32 bits host. It
seeks to 512 bytes before the end of the image to install gpt header. If
the size of image is larger than LONG_MAX, it overflows fseek() and
cause error:
isohybrid: image-x86-64-20140505110100.iso: seek error - 8: Invalid argument
Replace fseek with fseeko to fix this issue.
Signed-off-by: Kai Kang <kai.kang@windriver.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
|
|
Fix bug where the data would invariably be incorrectly written.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
|
|
For EFI, rather than mucking with the PECOFF or ELF headers (we have
both!) just use a fixed-size buffer embedded in the image with a large
magic number that can be scanned for safely.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
|
|
Fix the support for embedded DHCP options. Although we were
diligently saving them away, we never actually parsed them.
This fixes embedded options for BIOS only -- for EFI we need to modify
the encoding scheme so that it can fit inside an EFI PECOFF image.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
|
|
This reverts commit 7e7139788c8ac6ffbf474976ad4d1727e0677b9f.
Requested by Ady <ady-sf@hotmail.com> who also requested the original
patch:
After some tests with MENU SHIFTKEY with different versions of
Syslinux, I came to the conclusion that, at least for now, it would
be better to undo commit 7e7139788c8ac6ffbf474976ad4d1727e0677b9f:
"ldlinux: Allow a held shift key to override UI as well as DEFAULT"
committed on Thu, 22 May 2014.
Although the aforementioned commit makes the current 6.03 branch
behave as version 4.xx, my tests show that the behavior is
effectively inadequate and inconsistent.
Although I had been previously nagging you in irc about maintaining
backward compatibility of the "force-prompt" keys, I have to ask you
to undo the aforementioned commit before the final release of 6.03
stable. The behavior of Syslinux 6.03 without this particular commit
will slightly "break" with the prior "force-prompt" behavior as seen
in 4.xx when using [vesa]menu.c32, but it will be more consistent and
clean.
Without this commit, the user can still avoid the automatic launch of
the default label by the same means (the "escape" "shift" keys),
which is the most important characteristic of MENU SHIFTKEY.
Please consider this request before the release of 6.03 stable.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
|
|
Lua shall be one of the latest module to build as it requires some other
components to be built before.
Prior this patch, it was impossible to make a parallel build when "j"
was greater than 26.
This patch fix the ordering and allow a "make -j" of com32 directory.
|
|
memdump was a com16 binary. If still useful, it should be replaced
with a com32 library using libupload, but it has largely been replaced
by sysdump anyway.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
|
|
Sysdump was forked from memdump, but the README was never fixed. It
should be updated, but until then, delete it.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
|
|
This makes symbols defined in com32.ld and syslinux.ld HIDDEN, to
avoid collisions. Additional linker scripts should have this done, too.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
|
|
__syslinux_adv_* symbols are now part of the core, so don't export
them again elsewhere. This broke all uses of the ADV, including
--once.
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
|
|
It's already loading modules successfully, booting Linux, and both
UFS version 1 and 2 seem to be working correctly.
Signed-off-by: Raphael S. Carvalho <raphael.scarv@gmail.com>
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
|
|
It's needed to enumerate both UFS1 and UFS2 as they have different magic numbers
and super block offsets.
Neither UFS1 nor UFS2 move files around, therefore, ldlinux.sys can be
installed as a regular file.
Signed-off-by: Raphael S. Carvalho <raphael.scarv@gmail.com>
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
|
|
file->offset is used to store position in index root between
ntfs_readdir calls. Previously, pointer to buffer was stored in this
field. However this buffer is reallocated and read each ntfs_readdir
call so the pointer may become incorrect. Now offset in index root
rather than pointer is stored in this field.
[ hpa: applied patch manually as it arrived whitespace-corrupted ]
Signed-off-by: Andy Alex <andy at r-tt.com>
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
|
|
btrfs_read() was really just a function to read data linearly while
using the metadata cache. Move it to cache.c and rename it
cache_read() so other filesystems can make use of it as well.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
|
|
Remove the use of static variables and replace them with per-instance
allocations that can be found from the fs_info structure.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
|
|
Newer btrfs has a node size of more than 4K. Make sure we have a
buffer big enough to hold a node -- instead of allocating it on the
stack, allocate it at startup time.
While changing this code, remove a completely unnecessary arbitrary
64-bit divide.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
|
|
For compatibility with 4.xx, make a held shift escape to override UI
as well as DEFAULT.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
|
|
Correct the macros used in BTRFS_ADV_OFFSET.
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
|
|
ldlinux.sys is frequently too big to fit into Boot Area A (64K). Per
a recent ruling by Chris Mason, we have a 768K Boot Area B at offset
256K, so use that to store ldlinux.sys. In theory we could split it
up and make it slightly larger, however, ldlinux.sys is loaded below
the 640K mark so it has to be smaller than that size anyway.
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
|
|
Holding down shift is supposed to force a command line unless noescape
is provided; make it behave that way.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
|
|
Mixing extended and non-extended keyboard functions can be a bad idea.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
|
|
Getting the keyboard shift state is a firmware method. It is
unfortunately unclear if it is even possible on EFI.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
|
|
At least one USB keyboard BIOS has been reported to not properly
handle the keyboard flags in memory. See if actually using INT 16h
works better.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
|
|
NTFS $MFT file may be fragmented by itself (and actually is in most
cases). However, such a situation was not handled. This patch adds
support for a fragmented $MFT file.
Signed-off-by: Andy Alex <andy at r-tt.com>
Fixed-by: Ady <ady-sf@hotmail.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
|
|
Most of file system drivers initialize the cache themselves.
The problem is that the same cache could be again initialized later,
then invalidating the previous one. This patch fixes this.
Problem found while auditing the code.
Signed-off-by: Raphael S. Carvalho <raphael.scarv@gmail.com>
|
|
More obsolete cruft...
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
|
