From: "Chen, Kenneth W" For architecture like ia64, the switch stack structure is fairly large (currently 528 bytes). For context switch intensive application, we found that significant amount of cache misses occurs in switch_to() function. The following patch adds a hook in the schedule() function to prefetch switch stack structure as soon as 'next' task is determined. This allows maximum overlap in prefetch cache lines for that structure. Signed-off-by: Ken Chen Cc: Ingo Molnar Cc: "Luck, Tony" Signed-off-by: Andrew Morton --- arch/ia64/kernel/entry.S | 23 +++++++++++++++++++++++ include/asm-ia64/system.h | 1 + include/linux/sched.h | 5 +++++ kernel/sched.c | 1 + 4 files changed, 30 insertions(+) diff -puN arch/ia64/kernel/entry.S~prefetch-kernel-stacks-to-speed-up-context-switch arch/ia64/kernel/entry.S --- devel/arch/ia64/kernel/entry.S~prefetch-kernel-stacks-to-speed-up-context-switch 2005-09-07 20:10:38.000000000 -0700 +++ devel-akpm/arch/ia64/kernel/entry.S 2005-09-07 20:10:38.000000000 -0700 @@ -470,6 +470,29 @@ ENTRY(load_switch_stack) br.cond.sptk.many b7 END(load_switch_stack) +GLOBAL_ENTRY(prefetch_stack) + add r14 = -IA64_SWITCH_STACK_SIZE, sp + add r15 = IA64_TASK_THREAD_KSP_OFFSET, in0 + ;; + ld8 r16 = [r15] // load next's stack pointer + lfetch.fault.excl [r14], 128 + ;; + lfetch.fault.excl [r14], 128 + lfetch.fault [r16], 128 + ;; + lfetch.fault.excl [r14], 128 + lfetch.fault [r16], 128 + ;; + lfetch.fault.excl [r14], 128 + lfetch.fault [r16], 128 + ;; + lfetch.fault.excl [r14], 128 + lfetch.fault [r16], 128 + ;; + lfetch.fault [r16], 128 + br.ret.sptk.many rp +END(prefetch_switch_stack) + GLOBAL_ENTRY(execve) mov r15=__NR_execve // put syscall number in place break __BREAK_SYSCALL diff -puN include/asm-ia64/system.h~prefetch-kernel-stacks-to-speed-up-context-switch include/asm-ia64/system.h --- devel/include/asm-ia64/system.h~prefetch-kernel-stacks-to-speed-up-context-switch 2005-09-07 20:10:38.000000000 -0700 +++ devel-akpm/include/asm-ia64/system.h 2005-09-07 20:10:38.000000000 -0700 @@ -275,6 +275,7 @@ extern void ia64_load_extra (struct task */ #define __ARCH_WANT_UNLOCKED_CTXSW +#define ARCH_HAS_PREFETCH_SWITCH_STACK #define ia64_platform_is(x) (strcmp(x, platform_name) == 0) void cpu_idle_wait(void); diff -puN include/linux/sched.h~prefetch-kernel-stacks-to-speed-up-context-switch include/linux/sched.h --- devel/include/linux/sched.h~prefetch-kernel-stacks-to-speed-up-context-switch 2005-09-07 20:10:38.000000000 -0700 +++ devel-akpm/include/linux/sched.h 2005-09-07 20:10:38.000000000 -0700 @@ -604,6 +604,11 @@ extern int groups_search(struct group_in #define GROUP_AT(gi, i) \ ((gi)->blocks[(i)/NGROUPS_PER_BLOCK][(i)%NGROUPS_PER_BLOCK]) +#ifdef ARCH_HAS_PREFETCH_SWITCH_STACK +extern void prefetch_stack(struct task_struct*); +#else +static inline void prefetch_stack(struct task_struct *t) { } +#endif struct audit_context; /* See audit.c */ struct mempolicy; diff -puN kernel/sched.c~prefetch-kernel-stacks-to-speed-up-context-switch kernel/sched.c --- devel/kernel/sched.c~prefetch-kernel-stacks-to-speed-up-context-switch 2005-09-07 20:10:38.000000000 -0700 +++ devel-akpm/kernel/sched.c 2005-09-07 20:10:38.000000000 -0700 @@ -2888,6 +2888,7 @@ switch_tasks: if (next == rq->idle) schedstat_inc(rq, sched_goidle); prefetch(next); + prefetch_stack(next); clear_tsk_need_resched(prev); rcu_qsctr_inc(task_cpu(prev)); _