cond_resched() optimization
Subsystems (18)
PATCH 2/2 sched: optimize cond_resched()
In reply to Peter ZijlstraOptimize cond_resched() by removing one conditional.
Currently cond_resched() checks system_state == SYSTEM_RUNNING in order to avoid scheduling before the scheduler is running.
We can however, as per suggestion of Matt, use PREEMPT_ACTIVE to accomplish that very same.
--- include/linux/sched.h | 5 ++++- kernel/sched.c | 14 +++++++++----- 2 files changed, 13 insertions(+), 6 deletions(-)
include/linux/sched.h +4 -1
@@ -503,8 +503,11 @@ struct task_cputime {/* * Disable preemption until the scheduler is running. * Reset by start_kernel()->sched_init()->init_idle(). +* +* We include PREEMPT_ACTIVE to avoid cond_resched() from working +* before the scheduler is active -- see should_resched(). */ -#define INIT_PREEMPT_COUNT (1) +#define INIT_PREEMPT_COUNT (1 + PREEMPT_ACTIVE)/** * struct thread_group_cputimer - thread group interval timer counts
kernel/sched.c +9 -5
@@ -6580,6 +6580,11 @@ SYSCALL_DEFINE0(sched_yield)return 0 ; } +static inline int should_resched (void ) +{ +return need_resched () && !(preempt_count () &PREEMPT_ACTIVE ); +} +static void __cond_resched (void ) {#ifdef CONFIG_DEBUG_SPINLOCK_SLEEP
@@ -6599,8 +6604,7 @@ static void __cond_resched(void) int __sched _cond_resched(void) { -if (need_resched () && !(preempt_count () &PREEMPT_ACTIVE ) && - system_state ==SYSTEM_RUNNING ) { +if (should_resched ()) {__cond_resched ();return 1 ; }
@@ -6618,12 +6622,12 @@ EXPORT_SYMBOL(_cond_resched);*/ int cond_resched_lock (spinlock_t *lock) { -int resched =need_resched () && system_state ==SYSTEM_RUNNING ; +int resched =should_resched ();int ret =0 ;if (spin_needbreak (lock) || resched) {spin_unlock (lock); -if (resched &&need_resched ()) +if (resched)__cond_resched ();else cpu_relax ();
@@ -6638,7 +6642,7 @@ int __sched cond_resched_softirq(void) {BUG_ON (!in_softirq ()); -if (need_resched () && system_state ==SYSTEM_RUNNING ) { +if (should_resched ()) {local_bh_enable ();__cond_resched ();local_bh_disable ();
cover cond_resched() optimization
Matt suggested, whereupon Linus sayeth:
"This looks like a good patch. Please make it so"
Compile and boot tested on x86_64 only.
PATCH 1/2 sched: INIT_PREEMPT_COUNT
Pull the initial preempt_count value into a single definition site.
Maintainers for: alpha, ia64 and m68k, please have a look, your arch code is funny.
The header magic is a bit odd, but similar to the KERNEL_DS one, CPP waits with expanding these macros until the INIT_THREAD_INFO macro itself is expanded, which is in arch/*/kernel/init_task.c where we've already included sched.h so we're good.
--- arch/alpha/include/asm/thread_info.h | 1 + arch/arm/include/asm/thread_info.h | 2 +- arch/avr32/include/asm/thread_info.h | 2 +- arch/blackfin/include/asm/thread_info.h | 2 +- arch/cris/include/asm/thread_info.h | 4 +--- arch/frv/include/asm/thread_info.h | 4 +--- arch/h8300/include/asm/thread_info.h | 2 +- arch/ia64/include/asm/thread_info.h | 2 +- arch/m32r/include/asm/thread_info.h | 4 +--- arch/m68k/include/asm/thread_info_mm.h | 1 + arch/m68k/include/asm/thread_info_no.h | 1 + arch/microblaze/include/asm/thread_info.h | 4 +--- arch/mips/include/asm/thread_info.h | 4 +--- arch/mn10300/include/asm/thread_info.h | 4 +--- arch/parisc/include/asm/thread_info.h | 2 +- arch/powerpc/include/asm/thread_info.h | 4 +--- arch/s390/include/asm/thread_info.h | 2 +- arch/sh/include/asm/thread_info.h | 2 +- arch/sparc/include/asm/thread_info_32.h | 4 +--- arch/sparc/include/asm/thread_info_64.h | 4 +--- arch/um/include/asm/thread_info.h | 2 +- arch/x86/include/asm/thread_info.h | 2 +- arch/xtensa/include/asm/thread_info.h | 4 +--- include/linux/sched.h | 6 ++++++ 24 files changed, 29 insertions(+), 40 deletions(-)
arch/alpha/include/asm/thread_info.h +1 -0
@@ -37,6 +37,7 @@ struct thread_info { .task = &tsk, \ .exec_domain = &default_exec_domain, \ .addr_limit = KERNEL_DS, \ + .preempt_count = INIT_PREEMPT_COUNT, \ .restart_block = { \ .fn = do_no_restart_syscall, \ }, \
arch/arm/include/asm/thread_info.h +1 -1
@@ -73,7 +73,7 @@ struct thread_info { .task = &tsk, \ .exec_domain = &default_exec_domain, \ .flags = 0, \ - .preempt_count = 1, \ + .preempt_count = INIT_PREEMPT_COUNT, \ .addr_limit = KERNEL_DS, \ .cpu_domain =domain_val (DOMAIN_USER ,DOMAIN_MANAGER ) | \ domain_val(DOMAIN_KERNEL, DOMAIN_MANAGER) | \
arch/avr32/include/asm/thread_info.h +1 -1
@@ -40,7 +40,7 @@ struct thread_info { .exec_domain = &default_exec_domain, \ .flags = 0, \ .cpu = 0, \ - .preempt_count = 1, \ + .preempt_count = INIT_PREEMPT_COUNT, \ .restart_block = { \ .fn = do_no_restart_syscall \ } \
arch/blackfin/include/asm/thread_info.h +1 -1
@@ -77,7 +77,7 @@ struct thread_info { .exec_domain = &default_exec_domain, \ .flags = 0, \ .cpu = 0, \ - .preempt_count = 1, \ + .preempt_count = INIT_PREEMPT_COUNT, \ .restart_block = { \ .fn = do_no_restart_syscall, \ }, \
arch/cris/include/asm/thread_info.h +1 -3
@@ -50,8 +50,6 @@ struct thread_info {/* * macros/functions for gaining access to the thread information structure -* -* preempt_count needs to be 1 initially, until the scheduler is functional. */ #ifndef __ASSEMBLY__#define INIT_THREAD_INFO (tsk) \
@@ -60,7 +58,7 @@ struct thread_info { .exec_domain = &default_exec_domain, \ .flags = 0, \ .cpu = 0, \ - .preempt_count = 1, \ + .preempt_count = INIT_PREEMPT_COUNT, \ .addr_limit = KERNEL_DS, \ .restart_block = { \ .fn = do_no_restart_syscall, \
arch/frv/include/asm/thread_info.h +1 -3
@@ -56,8 +56,6 @@ struct thread_info {/* * macros/functions for gaining access to the thread information structure -* -* preempt_count needs to be 1 initially, until the scheduler is functional. */ #ifndef __ASSEMBLY__
@@ -67,7 +65,7 @@ struct thread_info { .exec_domain = &default_exec_domain, \ .flags = 0, \ .cpu = 0, \ - .preempt_count = 1, \ + .preempt_count = INIT_PREEMPT_COUNT, \ .addr_limit = KERNEL_DS, \ .restart_block = { \ .fn = do_no_restart_syscall, \
arch/h8300/include/asm/thread_info.h +1 -1
@@ -36,7 +36,7 @@ struct thread_info { .exec_domain = &default_exec_domain, \ .flags = 0, \ .cpu = 0, \ - .preempt_count = 1, \ + .preempt_count = INIT_PREEMPT_COUNT, \ .restart_block = { \ .fn = do_no_restart_syscall, \ }, \
arch/ia64/include/asm/thread_info.h +1 -1
@@ -48,7 +48,7 @@ struct thread_info { .flags = 0, \ .cpu = 0, \ .addr_limit = KERNEL_DS, \ - .preempt_count = 0, \ + .preempt_count = INIT_PREEMPT_COUNT, \ .restart_block = { \ .fn = do_no_restart_syscall, \ }, \
arch/m32r/include/asm/thread_info.h +1 -3
@@ -57,8 +57,6 @@ struct thread_info {/* * macros/functions for gaining access to the thread information structure -* -* preempt_count needs to be 1 initially, until the scheduler is functional. */ #ifndef __ASSEMBLY__
@@ -68,7 +66,7 @@ struct thread_info { .exec_domain = &default_exec_domain, \ .flags = 0, \ .cpu = 0, \ - .preempt_count = 1, \ + .preempt_count = INIT_PREEMPT_COUNT, \ .addr_limit = KERNEL_DS, \ .restart_block = { \ .fn = do_no_restart_syscall, \
arch/m68k/include/asm/thread_info_mm.h +1 -0
@@ -19,6 +19,7 @@ struct thread_info { { \ .task = &tsk, \ .exec_domain = &default_exec_domain, \ + .preempt_count = INIT_PREEMPT_COUNT, \ .restart_block = { \ .fn = do_no_restart_syscall, \ }, \
arch/m68k/include/asm/thread_info_no.h +1 -0
@@ -49,6 +49,7 @@ struct thread_info { .exec_domain = &default_exec_domain, \ .flags = 0, \ .cpu = 0, \ + .preempt_count = INIT_PREEMPT_COUNT, \ .restart_block = { \ .fn = do_no_restart_syscall, \ }, \
arch/microblaze/include/asm/thread_info.h +1 -3
@@ -75,8 +75,6 @@ struct thread_info {/* * macros/functions for gaining access to the thread information structure -* -* preempt_count needs to be 1 initially, until the scheduler is functional. */ #define INIT_THREAD_INFO (tsk) \ { \
@@ -84,7 +82,7 @@ struct thread_info { .exec_domain = &default_exec_domain, \ .flags = 0, \ .cpu = 0, \ - .preempt_count = 1, \ + .preempt_count = INIT_PREEMPT_COUNT, \ .addr_limit = KERNEL_DS, \ .restart_block = { \ .fn = do_no_restart_syscall, \
arch/mips/include/asm/thread_info.h +1 -3
@@ -39,8 +39,6 @@ struct thread_info {/* * macros/functions for gaining access to the thread information structure -* -* preempt_count needs to be 1 initially, until the scheduler is functional. */ #define INIT_THREAD_INFO (tsk) \ { \
@@ -48,7 +46,7 @@ struct thread_info { .exec_domain = &default_exec_domain, \ .flags = _TIF_FIXADE, \ .cpu = 0, \ - .preempt_count = 1, \ + .preempt_count = INIT_PREEMPT_COUNT, \ .addr_limit = KERNEL_DS, \ .restart_block = { \ .fn = do_no_restart_syscall, \
arch/mn10300/include/asm/thread_info.h +1 -3
@@ -65,8 +65,6 @@ struct thread_info {/* * macros/functions for gaining access to the thread information structure -* -* preempt_count needs to be 1 initially, until the scheduler is functional. */ #ifndef __ASSEMBLY__
@@ -76,7 +74,7 @@ struct thread_info { .exec_domain = &default_exec_domain, \ .flags = 0, \ .cpu = 0, \ - .preempt_count = 1, \ + .preempt_count = INIT_PREEMPT_COUNT, \ .addr_limit = KERNEL_DS, \ .restart_block = { \ .fn = do_no_restart_syscall, \
arch/parisc/include/asm/thread_info.h +1 -1
@@ -23,7 +23,7 @@ struct thread_info { .flags = 0, \ .cpu = 0, \ .addr_limit = KERNEL_DS, \ - .preempt_count = 1, \ + .preempt_count = INIT_PREEMPT_COUNT, \ .restart_block = { \ .fn = do_no_restart_syscall \ } \
arch/powerpc/include/asm/thread_info.h +1 -3
@@ -46,15 +46,13 @@ struct thread_info {/* * macros/functions for gaining access to the thread information structure -* -* preempt_count needs to be 1 initially, until the scheduler is functional. */ #define INIT_THREAD_INFO (tsk) \ { \ .task = &tsk, \ .exec_domain = &default_exec_domain, \ .cpu = 0, \ - .preempt_count = 1, \ + .preempt_count = INIT_PREEMPT_COUNT, \ .restart_block = { \ .fn = do_no_restart_syscall, \ }, \
arch/s390/include/asm/thread_info.h +1 -1
@@ -61,7 +61,7 @@ struct thread_info { .exec_domain = &default_exec_domain, \ .flags = 0, \ .cpu = 0, \ - .preempt_count = 1, \ + .preempt_count = INIT_PREEMPT_COUNT, \ .restart_block = { \ .fn = do_no_restart_syscall, \ }, \
arch/sh/include/asm/thread_info.h +1 -1
@@ -51,7 +51,7 @@ struct thread_info { .exec_domain = &default_exec_domain, \ .flags = 0, \ .cpu = 0, \ - .preempt_count = 1, \ + .preempt_count = INIT_PREEMPT_COUNT, \ .addr_limit = KERNEL_DS, \ .restart_block = { \ .fn = do_no_restart_syscall, \
arch/sparc/include/asm/thread_info_32.h +1 -3
@@ -54,8 +54,6 @@ struct thread_info {/* * macros/functions for gaining access to the thread information structure -* -* preempt_count needs to be 1 initially, until the scheduler is functional. */ #define INIT_THREAD_INFO (tsk) \ { \
@@ -64,7 +62,7 @@ struct thread_info { .exec_domain = &default_exec_domain, \ .flags = 0, \ .cpu = 0, \ - .preempt_count = 1, \ + .preempt_count = INIT_PREEMPT_COUNT, \ .restart_block = { \ .fn = do_no_restart_syscall, \ }, \
arch/sparc/include/asm/thread_info_64.h +1 -3
@@ -125,8 +125,6 @@ struct thread_info {/* * macros/functions for gaining access to the thread information structure -* -* preempt_count needs to be 1 initially, until the scheduler is functional. */ #ifndef __ASSEMBLY__
@@ -135,7 +133,7 @@ struct thread_info { .task = &tsk, \ .flags = ((unsigned long)ASI_P) << TI_FLAG_CURRENT_DS_SHIFT, \ .exec_domain = &default_exec_domain, \ - .preempt_count = 1, \ + .preempt_count = INIT_PREEMPT_COUNT, \ .restart_block = { \ .fn = do_no_restart_syscall, \ }, \
arch/um/include/asm/thread_info.h +1 -1
@@ -32,7 +32,7 @@ struct thread_info { .exec_domain = &default_exec_domain, \ .flags = 0, \ .cpu = 0, \ - .preempt_count = 1, \ + .preempt_count = INIT_PREEMPT_COUNT, \ .addr_limit = KERNEL_DS, \ .restart_block = { \ .fn = do_no_restart_syscall, \
arch/x86/include/asm/thread_info.h +1 -1
@@ -49,7 +49,7 @@ struct thread_info { .exec_domain = &default_exec_domain, \ .flags = 0, \ .cpu = 0, \ - .preempt_count = 1, \ + .preempt_count = INIT_PREEMPT_COUNT, \ .addr_limit = KERNEL_DS, \ .restart_block = { \ .fn = do_no_restart_syscall, \
arch/xtensa/include/asm/thread_info.h +1 -3
@@ -80,8 +80,6 @@ struct thread_info {/* * macros/functions for gaining access to the thread information structure -* -* preempt_count needs to be 1 initially, until the scheduler is functional. */ #ifndef __ASSEMBLY__
@@ -92,7 +90,7 @@ struct thread_info { .exec_domain = &default_exec_domain, \ .flags = 0, \ .cpu = 0, \ - .preempt_count = 1, \ + .preempt_count = INIT_PREEMPT_COUNT, \ .addr_limit = KERNEL_DS, \ .restart_block = { \ .fn = do_no_restart_syscall, \
include/linux/sched.h +6 -0
@@ -500,6 +500,12 @@ struct task_cputime { .sum_exec_runtime = 0, \ } +/* +* Disable preemption until the scheduler is running. +* Reset by start_kernel()->sched_init()->init_idle(). +*/ +#define INIT_PREEMPT_COUNT (1) +/** * struct thread_group_cputimer - thread group interval timer counts * @cputime: thread group interval timers.
Re: [PATCH 1/2] sched: INIT_PREEMPT_COUNT
include/linux/sched.h struct task_cputime { · line 508
Quoted 6 lines · Peter Zijlstra
include/linux/sched.h +6
-0+/* +* Disable preemption until the scheduler is running. +* Reset by start_kernel()->sched_init()->init_idle(). +*/ +#define INIT_PREEMPT_COUNT (1) +
I had to look at this for quite some time before it sank in that it wasn't a reset of a #define, or a reset of (1) (anybody else remember changing the value of '5' in a Fortran program?). Especially when stuck in with a bunch of cputimer defines. Would have taken even longer if I was looking in sched.h for something and not looking at this patch at the same time.
Can we fix this comment to mention it's thread_info.preempt_count that needs the reset?
Re: [PATCH 1/2] sched: INIT_PREEMPT_COUNT
Quoted 15 lines · On Fri, 2009-07-10 at 11:42 -0400, [email protected] wrote:
Review
Quoted 6 lines · Peter Zijlstra
(quoted diff) +6
-0+/* + * Disable preemption until the scheduler is running. + * Reset by start_kernel()->sched_init()->init_idle(). + */ +#define INIT_PREEMPT_COUNT (1) +I had to look at this for quite some time before it sank in that it wasn't a reset of a #define, or a reset of (1) (anybody else remember changing the value of '5' in a Fortran program?). Especially when stuck in with a bunch of cputimer defines. Would have taken even longer if I was looking in
sched.hfor something and not looking at this patch at the same time.
Quoted 2 lines
Can we fix this comment to mention it's thread_info.preempt_count that needs the reset?
Something along the lines of the below?
---
include/linux/sched.h +3 -1
@@ -502,7 +502,9 @@ struct task_cputime {/* * Disable preemption until the scheduler is running. -* Reset by start_kernel()->sched_init()->init_idle(). +* +* We reset this initial offset of init_thread_info.preempt_count in: +* start_kernel()->sched_init()->init_idle(). */ #define INIT_PREEMPT_COUNT (1)
Re: [PATCH 2/2] sched: optimize cond_resched()
In reply to Peter ZijlstraQuoted 9 lines · Peter Zijlstra
plain text document attachment (sched_opt_cond_resched.patch) Optimize
cond_resched()by removing one conditional.Currently
cond_resched()checkssystem_state==SYSTEM_RUNNINGin order to avoid scheduling before the scheduler is running.We can however, as per suggestion of Matt, use
PREEMPT_ACTIVEto accomplish that very same.
Pedantically, introducing should_resched should be its own patch, but other than that, these two patches look good.
Quoted 75 lines
Suggested-by Matt Mackall <[email protected]>Signed-off-by Peter Zijlstra <[email protected]>--- include/linux/sched.h | 5 ++++- kernel/sched.c | 14 +++++++++----- 2 files changed, 13 insertions(+), 6 deletions(-)Index:
linux-2.6/include/linux/sched.h
===================================================================include/linux/sched.h +4
-1@@ -503,8 +503,11 @@ struct task_cputime {/* * Disable preemption until the scheduler is running. * Reset by start_kernel()->sched_init()->init_idle(). +* +* We include PREEMPT_ACTIVE to avoid cond_resched() from working +* before the scheduler is active -- see should_resched(). */ -#define INIT_PREEMPT_COUNT (1) +#define INIT_PREEMPT_COUNT (1 + PREEMPT_ACTIVE)/** * struct thread_group_cputimer - thread group interval timer counts Index:
linux-2.6/kernel/sched.c
===================================================================kernel/sched.c +9
-5@@ -6580,6 +6580,11 @@ SYSCALL_DEFINE0(sched_yield)return 0 ; } +static inline int should_resched (void ) +{ +return need_resched () && !(preempt_count () &PREEMPT_ACTIVE ); +} +static void __cond_resched (void ) {#ifdef CONFIG_DEBUG_SPINLOCK_SLEEP @@ -6599,8 +6604,7 @@ static void __cond_resched(void) int __sched _cond_resched(void) { -if (need_resched () && !(preempt_count () &PREEMPT_ACTIVE ) && - system_state ==SYSTEM_RUNNING ) { +if (should_resched ()) {__cond_resched ();return 1 ; }@@ -6618,12 +6622,12 @@ EXPORT_SYMBOL(_cond_resched);*/ int cond_resched_lock (spinlock_t *lock) { -int resched =need_resched () && system_state ==SYSTEM_RUNNING ; +int resched =should_resched ();int ret =0 ;if (spin_needbreak (lock) || resched) {spin_unlock (lock); -if (resched &&need_resched ()) +if (resched)__cond_resched ();else cpu_relax ();@@ -6638,7 +6642,7 @@ int __sched cond_resched_softirq(void) {BUG_ON (!in_softirq ()); -if (need_resched () && system_state ==SYSTEM_RUNNING ) { +if (should_resched ()) {local_bh_enable ();__cond_resched ();local_bh_disable ();
Signature · 1 line
http://selenic.com : development and support for Mercurial and Linux
Re: [PATCH 1/2] sched: INIT_PREEMPT_COUNT
In reply to Peter ZijlstraQuoted 8 lines · Peter Zijlstra
Something along the lines of the below?
---
Index:linux-2.6/include/linux/sched.h
===================================================================
include/linux/sched.h struct task_cputime { · line 509
Quoted 9 lines
include/linux/sched.h +3
-1@@ -502,7 +502,9 @@ struct task_cputime {/* * Disable preemption until the scheduler is running. -* Reset by start_kernel()->sched_init()->init_idle(). +* +* We reset this initial offset of init_thread_info.preempt_count in: +* start_kernel()->sched_init()->init_idle(). */ #define INIT_PREEMPT_COUNT (1)
Works for me, thanks...
Re: [PATCH 1/2] sched: INIT_PREEMPT_COUNT
In reply to Peter ZijlstraQuoted 1 line · Peter Zijlstra
+#define INIT_PREEMPT_COUNT (1)
If the parentheses are useful or necessary, I need to be educated about why. They look like superfluous paranoia to me.
Quoted 5 lines · Peter Zijlstra
Matt suggested, whereupon Linus sayeth:
"This looks like a good patch. Please make it so"
Compile and boot tested on
x86_64only.
Tested on ppc32, boots fine.
Thanks,
Signature · 3 lines
Anton Vorontsov email: [email protected] irc://irc.freenode.net/bd2
Re: [PATCH 1/2] sched: INIT_PREEMPT_COUNT
In reply to Bjorn HelgaasQuoted 5 lines · Bjorn Helgaas
Quoted 1 line · Peter Zijlstra
+#define
INIT_PREEMPT_COUNT(1)If the parentheses are useful or necessary, I need to be educated about why. They look like superfluous paranoia to me.
They are strictly speaking superfluous, but since I already knew I'd be adding a term I added them.
Re: [PATCH 1/2] sched: INIT_PREEMPT_COUNT
In reply to Peter ZijlstraQuoted 5 lines · Peter Zijlstra
Pull the initial
preempt_countvalue into a single definition site.Maintainers for: alpha, ia64 and m68k, please have a look, your arch code is funny.
Goodness. I remember sending patches to fix the initial value of preempt_count on these architectures about four-five years ago. Guess they were ignored. It just shows that we need to keep pulling stuff out of arch/.
Signature · 4 lines
Matthew Wilcox Intel Open Source Technology Centre "Bill, look, we understand that you're interested in selling us this operating system, but compare it to ours. We can't possibly take such a retrograde step."
-tip: ACPICA: Do not schedule during early init
In reply to Peter ZijlstraQuoted 5 lines · Peter Zijlstra
Matt suggested, whereupon Linus sayeth:
"This looks like a good patch. Please make it so"
Compile and boot tested on
x86_64only.
I wanted to wait with his for .32, but Linus applied it yesterday so lets hope it's all fine on all architectures ;-)
There's one new test failure on x86 caused by this commit: a scheduling-while-atomic assert during early ACPI init - fixed by the patch below. It triggers on two separate test-systems.
Btw., that ACPI_PREEMPTION_POINT() wrapper is both superfluous (we should not wrap something as obvious as a cond_resched()) and shows signs of problems:
#define ACPI_PREEMPTION_POINT() \
do { \
if (!irqs_disabled()) \
cond_resched(); \
} while (0)that should have been 1) an inline function 2) should not check for whether irqs are off. If we need to check for irqs-off like this then this is a sign that either the code flow is too unbalanced (mixing different things into the same function), or that the preemption point has been applied at a too low level.
So a followup cleanup would likely be in order, especially that this was the last (and only) call site of ACPI_PREEMPTION_POINT(). I'd suggest to remove it.
( I'm not sure what prompted the addition of this rescheduling point
- if there was a strong reason for it then we should probably add back the preemption point to the place that is causing the latency. )
Ingo
-------------------->
From 2b2b96115287177600c0158c95e87c5ab44f8379 Mon Sep 17 00:00:00 2001
From: Ingo Molnar <[email protected]>
Date: Sat, 11 Jul 2009 13:15:04 +0200
Subject: [PATCH] ACPICA: Do not schedule during early init
-tip testing found a test failure on x86, a scheduling-while-atomic bug during early ACPI init:
[ 0.048083] ACPI: Core revision 20090521 [ 0.051854] BUG: scheduling while atomic: swapper/0/0x10000002 [ 0.052076] no locks held by swapper/0. [ 0.053000] Pid: 0, comm: swapper Not tainted 2.6.31-rc2-tip-01240-gd51b6ef-dirty #69901 [ 0.053997] Call Trace: [ 0.055006] [<ffffffff8107afe2>] __schedule_bug+0x80/0x9c [ 0.057001] [<ffffffff81f1160a>] schedule+0xe6/0x5de [ 0.058000] [<ffffffff8151531f>] ? acpi_os_release_object+0x1c/0x34 [ 0.059002] [<ffffffff8154d16c>] ? acpi_ut_exit+0x40/0x5c [ 0.060020] [<ffffffff8107e1dc>] __cond_resched+0x37/0x69 [ 0.060998] [<ffffffff81f11d55>] _cond_resched+0x37/0x56 [ 0.061999] [<ffffffff81546748>] acpi_ps_complete_op+0x412/0x457 [ 0.062998] [<ffffffff8154588d>] ? acpi_ps_next_parse_state+0x14a/0x16b [ 0.064019] [<ffffffff81546e87>] acpi_ps_parse_loop+0x47e/0x513 [ 0.064998] [<ffffffff81545418>] acpi_ps_parse_aml+0x177/0x4a2 [ 0.065998] [<ffffffff8154a83b>] ? acpi_get_table_by_index+0x138/0x159 [ 0.066998] [<ffffffff81543c0d>] acpi_ns_one_complete_parse+0x1d1/0x220 [ 0.068019] [<ffffffff81543cd7>] acpi_ns_parse_table+0x7b/0x148 [ 0.068997] [<ffffffff8154b229>] ? acpi_tb_allocate_owner_id+0x95/0xb4 [ 0.069997] [<ffffffff8153e62d>] acpi_ns_load_table+0xc5/0x1b8 [ 0.070998] [<ffffffff8288a140>] ? early_idt_handler+0x0/0x71 [ 0.072018] [<ffffffff8154aa03>] acpi_tb_load_namespace+0x9d/0x1ba [ 0.072996] [<ffffffff8288a140>] ? early_idt_handler+0x0/0x71 [ 0.073996] [<ffffffff8154ab5a>] acpi_load_tables+0x3a/0x99 [ 0.074997] [<ffffffff828bef2e>] acpi_early_init+0x71/0x11a [ 0.076997] [<ffffffff8288a140>] ? early_idt_handler+0x0/0x71 [ 0.077996] [<ffffffff8288b0eb>] start_kernel+0x39f/0x3c1 [ 0.078995] [<ffffffff8288a140>] ? early_idt_handler+0x0/0x71 [ 0.080016] [<ffffffff8288a140>] ? early_idt_handler+0x0/0x71 [ 0.080995] [<ffffffff8288a2ce>] x86_64_start_reservations+0xb9/0xd4 [ 0.081995] [<ffffffff8288a000>] ? __init_begin+0x0/0x140 [ 0.082995] [<ffffffff8288a3ed>] x86_64_start_kernel+0x104/0x127 [ 0.088045] BUG: scheduling while atomic: swapper/0/0x10000002 [ 0.089999] no locks held by swapper/0. [ 0.090993] Pid: 0, comm: swapper Not tainted 2.6.31-rc2-tip-01240-gd51b6ef-dirty #69901
The problem is that drivers/acpi/acpica/psloop.c has this line:
ACPI_PREEMPTION_POINT();
Which does not mix well with this early init stage - we have preemption disabled and the init task has not started up yet, so we really should not schedule.
Remove this explicit preemption point.
--- drivers/acpi/acpica/psloop.c | 2 -- 1 files changed, 0 insertions(+), 2 deletions(-)
drivers/acpi/acpica/psloop.c +0 -2
@@ -720,8 +720,6 @@ acpi_ps_complete_op(struct acpi_walk_state *walk_state, *op =NULL ; } -ACPI_PREEMPTION_POINT (); -return_ACPI_STATUS (AE_OK ); }