Skip to contentSkip to conversation

SMP: kill redundant call_function_data->cpumask_ipi field

PATCH In review Jiang Liu 3 participants 5 messages Sep 8, 2013 – Sep 10, 2013 CPU hotplug

lore
Share & tools
lore

Jiang Liu

author
#0 lore raw

PATCH 1/3 SMP: kill redundant call_function_data->cpumask_ipi field

From: Jiang Liu <[email protected]>

Commit f44310b98ddb7 "smp: Fix SMP function call empty cpu mask race" introduced field call_function_data->cpumask_ipi to resolve a race condition in smp_call_function_many().

Later commit 9a46ad6d6df3 "smp: make smp_call_function_many() use logic similar to smp_call_function_single()" fixed the same issue in another way when optimizing smp_call_function_many(), which then obsoletes changes introduced by commit f44310b98ddb7. So revert it.

We may also keep call_function_data->cpumask_ipi field and use it to optimize smp_call_function_many() as below:

kernel/smp.c +2 -0
index fe9f773..dd852fb 100644
@@ -428,6 +428,8 @@ void smp_call_function_many(const struct cpumask *mask,
                csd->info = info;
 
                raw_spin_lock_irqsave(&dst->lock, flags);
+               if (list_empty(&dst->list))
+                       cpumask_clear_cpu(cpu, cfd->cpumask_ipi);
                list_add_tail(&csd->list, &dst->list);
                raw_spin_unlock_irqrestore(&dst->lock, flags);
        }

But I have no platforms to verify the optimization effect, so simpify the code first.

Cc Jiang Liu <[email protected]>
Cc Andrew Morton <[email protected]>
Cc Wang YanQing <[email protected]>
Cc Shaohua Li <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Steven Rostedt <[email protected]>
Cc Paul Gortmaker <[email protected]>
Cc liguang <[email protected]>
---
 kernel/smp.c | 16 ++--------------
 1 file changed, 2 insertions(+), 14 deletions(-)
kernel/smp.c +2 -14
index fe9f773..a034712 100644
@@ -23,7 +23,6 @@ enum {
 struct call_function_data {
 	struct call_single_data	__percpu *csd;
 	cpumask_var_t		cpumask;
-	cpumask_var_t		cpumask_ipi;
 };
 
 static DEFINE_PER_CPU_SHARED_ALIGNED(struct call_function_data, cfd_data);
@@ -47,9 +46,6 @@ hotplug_cfd(struct notifier_block *nfb, unsigned long action, void *hcpu)
 		if (!zalloc_cpumask_var_node(&cfd->cpumask, GFP_KERNEL,
 				cpu_to_node(cpu)))
 			return notifier_from_errno(-ENOMEM);
-		if (!zalloc_cpumask_var_node(&cfd->cpumask_ipi, GFP_KERNEL,
-				cpu_to_node(cpu)))
-			return notifier_from_errno(-ENOMEM);
 		cfd->csd = alloc_percpu(struct call_single_data);
 		if (!cfd->csd) {
 			free_cpumask_var(cfd->cpumask);
@@ -64,7 +60,6 @@ hotplug_cfd(struct notifier_block *nfb, unsigned long action, void *hcpu)
 	case CPU_DEAD:
 	case CPU_DEAD_FROZEN:
 		free_cpumask_var(cfd->cpumask);
-		free_cpumask_var(cfd->cpumask_ipi);
 		free_percpu(cfd->csd);
 		break;
 #endif
@@ -255,9 +250,9 @@ int smp_call_function_single(int cpu, smp_call_func_t func, void *info,
 				csd = &__get_cpu_var(csd_data);
 
 			csd_lock(csd);
-
 			csd->func = func;
 			csd->info = info;
+
 			generic_exec_single(cpu, csd, wait);
 		} else {
 			err = -ENXIO;	/* CPU not online */
@@ -410,13 +405,6 @@ void smp_call_function_many(const struct cpumask *mask,
 	if (unlikely(!cpumask_weight(cfd->cpumask)))
 		return;
 
-	/*
-	 * After we put an entry into the list, cfd->cpumask may be cleared
-	 * again when another CPU sends another IPI for a SMP function call, so
-	 * cfd->cpumask will be zero.
-	 */
-	cpumask_copy(cfd->cpumask_ipi, cfd->cpumask);
-
 	for_each_cpu(cpu, cfd->cpumask) {
 		struct call_single_data *csd = per_cpu_ptr(cfd->csd, cpu);
 		struct call_single_queue *dst =
@@ -433,7 +421,7 @@ void smp_call_function_many(const struct cpumask *mask,
 	}
 
 	/* Send a message to all CPUs in the map */
-	arch_send_call_function_ipi_mask(cfd->cpumask_ipi);
+	arch_send_call_function_ipi_mask(cfd->cpumask);
 
 	if (wait) {
 		for_each_cpu(cpu, cfd->cpumask) {
Signature · 1 line
1.8.1.2
author
#1 lore raw

PATCH 2/3 SMP: simpilify function generic_smp_call_function_single_interrupt()

From: Jiang Liu <[email protected]>

Now the call_single_data data structure is always locked by invoking csd_lock() before inserting it into corresponding call_single_queue list, so no need to save and check csd->flags in function generic_smp_call_function_single_interrupt().

Cc Jiang Liu <[email protected]>
Cc Andrew Morton <[email protected]>
Cc Wang YanQing <[email protected]>
Cc Shaohua Li <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Steven Rostedt <[email protected]>
Cc Paul Gortmaker <[email protected]>
Cc liguang <[email protected]>
---
 kernel/smp.c | 16 +---------------
 1 file changed, 1 insertion(+), 15 deletions(-)
kernel/smp.c +1 -15
index a034712..baa2a65 100644
@@ -181,25 +181,11 @@ void generic_smp_call_function_single_interrupt(void)
 
 	while (!list_empty(&list)) {
 		struct call_single_data *csd;
-		unsigned int csd_flags;
 
 		csd = list_entry(list.next, struct call_single_data, list);
 		list_del(&csd->list);
-
-		/*
-		 * 'csd' can be invalid after this call if flags == 0
-		 * (when called through generic_exec_single()),
-		 * so save them away before making the call:
-		 */
-		csd_flags = csd->flags;
-
 		csd->func(csd->info);
-
-		/*
-		 * Unlocked CSDs are valid through generic_exec_single():
-		 */
-		if (csd_flags & CSD_FLAG_LOCK)
-			csd_unlock(csd);
+		csd_unlock(csd);
 	}
 }
 
Signature · 1 line
1.8.1.2
#2 lore raw

Re: [PATCH v1 2/3] SMP: simpilify function generic_smp_call_function_single_interrupt()

Quoted 7 lines · Jiang Liu

From: Jiang Liu <[email protected]>

Now the call_single_data data structure is always locked by invoking csd_lock() before inserting it into corresponding call_single_queue list, so no need to save and check csd->flags in function generic_smp_call_function_single_interrupt().

Quoted 11 lines
Cc Jiang Liu <[email protected]>
Cc Andrew Morton <[email protected]>
Cc Wang YanQing <[email protected]>
Cc Shaohua Li <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Steven Rostedt <[email protected]>
Cc Paul Gortmaker <[email protected]>
Cc liguang <[email protected]>

Hi, I've sent a same patch before and it has been applied by Ingo.
See http://git.kernel.org/cgit/linux/kernel/git/tip/tip.git/commit/?id=46591962cb5bfd2bfb0baf42497119c816503598

Quoted 36 lines
---
 kernel/smp.c | 16 +---------------
 1 file changed, 1 insertion(+), 15 deletions(-)
kernel/smp.c +1 -15
index a034712..baa2a65 100644
@@ -181,25 +181,11 @@ void generic_smp_call_function_single_interrupt(void)
 
 	while (!list_empty(&list)) {
 		struct call_single_data *csd;
-		unsigned int csd_flags;
 
 		csd = list_entry(list.next, struct call_single_data, list);
 		list_del(&csd->list);
-
-		/*
-		 * 'csd' can be invalid after this call if flags == 0
-		 * (when called through generic_exec_single()),
-		 * so save them away before making the call:
-		 */
-		csd_flags = csd->flags;
-
 		csd->func(csd->info);
-
-		/*
-		 * Unlocked CSDs are valid through generic_exec_single():
-		 */
-		if (csd_flags & CSD_FLAG_LOCK)
-			csd_unlock(csd);
+		csd_unlock(csd);
 	}
 }
 
author
#3 lore raw

PATCH 3/3 SMP, trivial: remove unused code from smp_boot.h

In reply to Jiang Liu

From: Jiang Liu <[email protected]>

Function smpboot_thread_schedule() is never defined or used, so remove it from smp_boot.h.

Cc Jiang Liu <[email protected]>
Cc Jiang Liu <[email protected]>
Cc Andrew Morton <[email protected]>
Cc Wang YanQing <[email protected]>
Cc Shaohua Li <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Steven Rostedt <[email protected]>
Cc Paul Gortmaker <[email protected]>
Cc liguang <[email protected]>
---
 include/linux/smpboot.h | 1 -
 1 file changed, 1 deletion(-)
include/linux/smpboot.h +0 -1
index 13e9296..d600afb 100644
@@ -47,6 +47,5 @@ struct smp_hotplug_thread {
 
 int smpboot_register_percpu_thread(struct smp_hotplug_thread *plug_thread);
 void smpboot_unregister_percpu_thread(struct smp_hotplug_thread *plug_thread);
-int smpboot_thread_schedule(void);
 
 #endif
Signature · 1 line
1.8.1.2
Acked-by on v1 1/3
#4 lore raw
In reply to Jiang Liu
Quoted 11 lines · Jiang Liu

From: Jiang Liu <[email protected]>

Commit f44310b98ddb7 "smp: Fix SMP function call empty cpu mask race" introduced field call_function_data->cpumask_ipi to resolve a race condition in smp_call_function_many().

Later commit 9a46ad6d6df3 "smp: make smp_call_function_many() use logic similar to smp_call_function_single()" fixed the same issue in another way when optimizing smp_call_function_many(), which then obsoletes changes introduced by commit f44310b98ddb7. So revert it.

Yes, you are right, after commit 9a46ad6d6df3, we can revert f44310b98ddb7. Maybe use "Revert smp: Fix SMP function call empty cpu mask race" is a better subject.

Quoted 8 lines

We may also keep call_function_data->cpumask_ipi field and use it to optimize smp_call_function_many() as below:

kernel/smp.c +0 -0
index fe9f773..dd852fb 100644

kernel/smp.c void smp_call_function_many(const struct cpumask *mask, · line 435

Quoted 8 lines
kernel/smp.c +2 -0
@@ -428,6 +428,8 @@ void smp_call_function_many(const struct cpumask *mask,
                csd->info = info;
 
                raw_spin_lock_irqsave(&dst->lock, flags);
+               if (list_empty(&dst->list))
+                       cpumask_clear_cpu(cpu, cfd->cpumask_ipi);
                list_add_tail(&csd->list, &dst->list);
                raw_spin_unlock_irqrestore(&dst->lock, flags);
        }

Your optimization don't need to keep cpumask_ipi, just clear cfd->cpumask, and test whether cfd->cpumask is empty after "for_each_cpu(cpu, cfd->cpumask)".

Thanks.