de-xchg fork.c
Subsystems (3)
PATCH 1/1 de-xchg fork.c
Don't know who put this in for 2.5.38.
I realize that using xchg() makes you 'leet. But doing an atomic op where none is required is suboptimal and confusing.
Hey, at least it wasn't:
if (likely(tsk = xchg(task_cache + cpu, tsk))) {Cheers,
Rusty.
kernel/fork.c +2 -1
@@ -64,11 +64,12 @@ void __put_task_struct(struct task_struc }else {int cpu =smp_processor_id (); - tsk =xchg (task_cache + cpu, tsk); + tsk = task_cache[cpu];if (tsk) {free_thread_info (tsk->thread_info);kmem_cache_free (task_struct_cachep,tsk); } + task_cache[cpu] = current; } }
--
Anyone who quotes me in their sig is an idiot. -- Rusty Russell.
Review
Quoted 2 lines · Rusty Russell
(quoted diff) +1
-1- tsk = xchg(task_cache + cpu, tsk); + tsk = task_cache[cpu];
this was me - i forgot that this are per-CPU fields. (hey and i wrote the original task_cache code so there's no excuse :-)
the attached patch (against BK-curr) fixes all xchg()'s and a preemption bug.
Ingo
kernel/fork.c renamed from kernel/fork.c.orig +6 -2
@@ -64,11 +64,12 @@ }else {int cpu =smp_processor_id (); - tsk =xchg (task_cache + cpu, tsk); + tsk = task_cache[cpu];if (tsk) {free_thread_info (tsk->thread_info);kmem_cache_free (task_struct_cachep,tsk); } + task_cache[cpu] = current; } }
@@ -126,8 +127,11 @@ {struct task_struct *tsk;struct thread_info *ti; +int cpu =get_cpu (); - tsk =xchg (task_cache +smp_processor_id (),NULL ); + tsk = task_cache[cpu]; + task_cache[cpu] =NULL ; +put_cpu ();if (!tsk) { ti =alloc_thread_info ();if (!ti)
Quoted 2 lines · Ingo Molnar
the attached patch (against BK-curr) fixes all xchg()'s and a preemption bug.
Thanks, Ingo.
Quoted 5 lines
kernel/fork.c renamed from kernel/fork.c.orig +0
-0@@ -64,11 +64,12 @@ }else {int cpu =smp_processor_id ();
kernel/fork.c line 74
Quoted 10 lines
kernel/fork.c renamed from kernel/fork.c.orig +2
-1- tsk =xchg (task_cache + cpu, tsk); + tsk = task_cache[cpu];if (tsk) {free_thread_info (tsk->thread_info);kmem_cache_free (task_struct_cachep,tsk); } + task_cache[cpu] = current; } }
I think you need get/put_cpu() here, too?
Robert Love
Quoted 5 lines · Robert Love
Review
Quoted 3 lines
(quoted diff) +1
-0+ task_cache[cpu] = current; } }I think you need get/put_cpu() here, too?
you are right - new patch attached.
Ingo
kernel/fork.c renamed from kernel/fork.c.orig +8 -3
@@ -62,13 +62,15 @@free_thread_info (tsk->thread_info);kmem_cache_free (task_struct_cachep,tsk); }else { -int cpu =smp_processor_id (); +int cpu =get_cpu (); - tsk =xchg (task_cache + cpu, tsk); + tsk = task_cache[cpu];if (tsk) {free_thread_info (tsk->thread_info);kmem_cache_free (task_struct_cachep,tsk); } + task_cache[cpu] = current; +put_cpu (); } }
@@ -126,8 +128,11 @@ {struct task_struct *tsk;struct thread_info *ti; +int cpu =get_cpu (); - tsk =xchg (task_cache +smp_processor_id (),NULL ); + tsk = task_cache[cpu]; + task_cache[cpu] =NULL ; +put_cpu ();if (!tsk) { ti =alloc_thread_info ();if (!ti)
Hi!
Quoted 4 lines
Don't know who put this in for 2.5.38.
I realize that using
xchg()makes you 'leet. But doing an atomic op where none is required is suboptimal and confusing.
Well, atomic op is also more expensive. i think ingo did this. Ingo, is patch below safe? It is faster *and* it works on 386.
Pavel
Quoted 16 lines
diff -urpN --exclude TAGS -X
/home/rusty/devel/kernel/kernel-patches/current-dontdiff--minimallinux-2.5.38/kernel/fork.cworking-2.5.38-unxchg/kernel/fork.ckernel/fork.c +2
-1@@ -64,11 +64,12 @@ void __put_task_struct(struct task_struc }else {int cpu =smp_processor_id (); - tsk =xchg (task_cache + cpu, tsk); + tsk = task_cache[cpu];if (tsk) {free_thread_info (tsk->thread_info);kmem_cache_free (task_struct_cachep,tsk); } + task_cache[cpu] = current; } }
Signature · 2 lines
Philips Velo 1: 1"x4"x8", 300gram, 60, 12MB, 40bogomips, linux, mutt, details at http://atrey.karlin.mff.cuni.cz/~pavel/velo/index.html.