* [PATCH 1/2] cpumask: Convert cpumask_of_cpu to static array
2007-11-21 8:02 [PATCH 0/2] x86: Reduce pressure on stack from cpumask usage travis
@ 2007-11-21 8:02 ` travis
2007-11-21 8:02 ` [PATCH 2/2] cpumask: Convert set_cpus_allowed to use ptr for cpumask arg travis
1 sibling, 0 replies; 3+ messages in thread
From: travis @ 2007-11-21 8:02 UTC (permalink / raw)
To: Andrew Morton, Andi Kleen, Christoph Lameter
Cc: mingo, apw, Jack Steiner, linux-mm, linux-kernel
[-- Attachment #1: cpumask-to-percpu --]
[-- Type: text/plain, Size: 2866 bytes --]
Here is a simple patch to use a per cpu cpumask instead of constructing
one on the stack. I have been running awhile with this one:
Do not use stack to allocate cpumask for cpumask_of_cpu
Signed-off-by: Christoph Lameter <clameter@sgi.com>
Modified to be used only if NR_CPUS is greater than the BITS_PER_LONG
as well as fix cases where !SMP and both NR_CPUS > and < BITS_PER_LONG.
Signed-off-by: Mike Travis <travis@sgi.com>
---
arch/x86/kernel/process_64.c | 1 +
arch/x86/mm/numa_64.c | 1 +
include/linux/cpumask.h | 4 ++++
include/linux/sched.h | 4 ++++
kernel/sched.c | 9 +++++++++
5 files changed, 19 insertions(+)
--- a/include/linux/cpumask.h
+++ b/include/linux/cpumask.h
@@ -222,6 +222,9 @@ int __next_cpu(int n, const cpumask_t *s
#define next_cpu(n, src) 1
#endif
+#if defined(CONFIG_SMP) && (NR_CPUS > BITS_PER_LONG)
+#define cpumask_of_cpu(cpu) per_cpu(cpu_mask, cpu)
+#else
#define cpumask_of_cpu(cpu) \
({ \
typeof(_unused_cpumask_arg_) m; \
@@ -233,6 +236,7 @@ int __next_cpu(int n, const cpumask_t *s
} \
m; \
})
+#endif
#define CPU_MASK_LAST_WORD BITMAP_LAST_WORD_MASK(NR_CPUS)
--- a/kernel/sched.c
+++ b/kernel/sched.c
@@ -6732,6 +6732,11 @@ static void init_cfs_rq(struct cfs_rq *c
cfs_rq->min_vruntime = (u64)(-(1LL << 20));
}
+#if NR_CPUS > BITS_PER_LONG
+DEFINE_PER_CPU(cpumask_t, cpu_mask);
+EXPORT_PER_CPU_SYMBOL(cpu_mask);
+#endif
+
void __init sched_init(void)
{
int highest_cpu = 0;
@@ -6741,6 +6746,10 @@ void __init sched_init(void)
struct rt_prio_array *array;
struct rq *rq;
+#if NR_CPUS > BITS_PER_LONG
+ /* This makes cpumask_of_cpu work */
+ cpu_set(i, per_cpu(cpu_mask, i));
+#endif
rq = cpu_rq(i);
spin_lock_init(&rq->lock);
lockdep_set_class(&rq->lock, &rq->rq_lock_key);
--- a/arch/x86/kernel/process_64.c
+++ b/arch/x86/kernel/process_64.c
@@ -36,6 +36,7 @@
#include <linux/kprobes.h>
#include <linux/kdebug.h>
#include <linux/tick.h>
+#include <linux/percpu.h>
#include <asm/uaccess.h>
#include <asm/pgtable.h>
--- a/arch/x86/mm/numa_64.c
+++ b/arch/x86/mm/numa_64.c
@@ -11,6 +11,7 @@
#include <linux/ctype.h>
#include <linux/module.h>
#include <linux/nodemask.h>
+#include <linux/sched.h>
#include <asm/e820.h>
#include <asm/proto.h>
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -2024,6 +2024,10 @@ static inline void migration_init(void)
#define TASK_SIZE_OF(tsk) TASK_SIZE
#endif
+#if defined(CONFIG_SMP) && (NR_CPUS > BITS_PER_LONG)
+DECLARE_PER_CPU(cpumask_t, cpu_mask);
+#endif
+
#endif /* __KERNEL__ */
#endif
--
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>
^ permalink raw reply [flat|nested] 3+ messages in thread* [PATCH 2/2] cpumask: Convert set_cpus_allowed to use ptr for cpumask arg
2007-11-21 8:02 [PATCH 0/2] x86: Reduce pressure on stack from cpumask usage travis
2007-11-21 8:02 ` [PATCH 1/2] cpumask: Convert cpumask_of_cpu to static array travis
@ 2007-11-21 8:02 ` travis
1 sibling, 0 replies; 3+ messages in thread
From: travis @ 2007-11-21 8:02 UTC (permalink / raw)
To: Andrew Morton, Andi Kleen, Christoph Lameter
Cc: mingo, apw, Jack Steiner, linux-mm, linux-kernel
[-- Attachment #1: set_cpus_allowed-use-ptr-arg --]
[-- Type: text/plain, Size: 3131 bytes --]
Avoid pushing cpumask variables onto stack when calling set_cpus_allowed
when NR_CPUS > BITS_PER_LONG by passing cpumast_t arg as a pointer.
Signed-off-by: Mike Travis <travis@sgi.com>
---
include/linux/sched.h | 11 +++++++++--
kernel/sched.c | 22 ++++++++++++++++------
2 files changed, 25 insertions(+), 8 deletions(-)
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -1435,8 +1435,15 @@ static inline void put_task_struct(struc
#define tsk_used_math(p) ((p)->flags & PF_USED_MATH)
#define used_math() tsk_used_math(current)
-#ifdef CONFIG_SMP
-extern int set_cpus_allowed(struct task_struct *p, cpumask_t new_mask);
+#if defined(CONFIG_SMP)
+# if NR_CPUS > BITS_PER_LONG
+ /* avoid pushing cpumask variable onto the stack */
+# define set_cpus_allowed(p, new_mask) __set_cpus_allowed((p), &(new_mask))
+ extern int __set_cpus_allowed(struct task_struct *p, cpumask_t *new_mask);
+# else
+# define set_cpus_allowed(p, new_mask) __set_cpus_allowed((p), (new_mask))
+ extern int __set_cpus_allowed(struct task_struct *p, cpumask_t new_mask);
+# endif
#else
static inline int set_cpus_allowed(struct task_struct *p, cpumask_t new_mask)
{
--- a/kernel/sched.c
+++ b/kernel/sched.c
@@ -5036,7 +5036,16 @@ static inline void sched_init_granularit
* task must not exit() & deallocate itself prematurely. The
* call is not atomic; no spinlocks may be held.
*/
-int set_cpus_allowed(struct task_struct *p, cpumask_t new_mask)
+#if NR_CPUS > BITS_PER_LONG
+/*
+ * avoid pushing a large CPU count cpumask_t variable onto stack
+ * (relies on the fact that new_mask is a const arg to subfunctions)
+ */
+#define CPU_MASK_VAR(v) *v
+#else
+#define CPU_MASK_VAR(v) v
+#endif
+int __set_cpus_allowed(struct task_struct *p, cpumask_t CPU_MASK_VAR(new_mask))
{
struct migration_req req;
unsigned long flags;
@@ -5044,17 +5053,17 @@ int set_cpus_allowed(struct task_struct
int ret = 0;
rq = task_rq_lock(p, &flags);
- if (!cpus_intersects(new_mask, cpu_online_map)) {
+ if (!cpus_intersects(CPU_MASK_VAR(new_mask), cpu_online_map)) {
ret = -EINVAL;
goto out;
}
- p->cpus_allowed = new_mask;
+ p->cpus_allowed = CPU_MASK_VAR(new_mask);
/* Can the task run on the task's current CPU? If so, we're done */
- if (cpu_isset(task_cpu(p), new_mask))
+ if (cpu_isset(task_cpu(p), CPU_MASK_VAR(new_mask)))
goto out;
- if (migrate_task(p, any_online_cpu(new_mask), &req)) {
+ if (migrate_task(p, any_online_cpu(CPU_MASK_VAR(new_mask)), &req)) {
/* Need help from migration thread: drop lock and wait. */
task_rq_unlock(rq, &flags);
wake_up_process(rq->migration_thread);
@@ -5067,7 +5076,8 @@ out:
return ret;
}
-EXPORT_SYMBOL_GPL(set_cpus_allowed);
+/* keep CPU_MASK_VAR local to __set_cpus_allowed */
+EXPORT_SYMBOL_GPL(__set_cpus_allowed);
/*
* Move (not current) task off this cpu, onto dest cpu. We're doing
--
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>
^ permalink raw reply [flat|nested] 3+ messages in thread