Skip to contentSkip to conversation

mm/page_alloc: remove repetitious local_irq_save() in __zone_pcp_update()

PATCH In review Cody P Schafer 3 participants 7 messages Jun 18, 2013 – Jun 20, 2013

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 mm/page_alloc: remove repetitious local_irq_save() in __zone_pcp_update()

__zone_pcp_update() is called via stop_machine(), which already disables local irq.

---
 mm/page_alloc.c | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)
mm/page_alloc.c +1 -3
index bac3107..b46b54a 100644
@@ -6179,7 +6179,7 @@ static int __meminit __zone_pcp_update(void *data)
 {
 	struct zone *zone = data;
 	int cpu;
-	unsigned long batch = zone_batchsize(zone), flags;
+	unsigned long batch = zone_batchsize(zone);
 
 	for_each_possible_cpu(cpu) {
 		struct per_cpu_pageset *pset;
@@ -6188,12 +6188,10 @@ static int __meminit __zone_pcp_update(void *data)
 		pset = per_cpu_ptr(zone->pageset, cpu);
 		pcp = &pset->pcp;
 
-		local_irq_save(flags);
 		if (pcp->count > 0)
 			free_pcppages_bulk(zone, pcp->count, pcp);
 		drain_zonestat(zone, pset);
 		setup_pageset(pset, batch);
-		local_irq_restore(flags);
 	}
 	return 0;
 }
Signature · 7 lines
1.8.3.1

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to [email protected].  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
Reviewed-by on v1 1/1
#1 lore raw
Quoted 4 lines · Cody P Schafer

__zone_pcp_update() is called via stop_machine(), which already disables local irq.

Regards,
Srivatsa S. Bhat

Quoted 21 lines
---
 mm/page_alloc.c | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)
mm/page_alloc.c +1 -1
index bac3107..b46b54a 100644
@@ -6179,7 +6179,7 @@ static int __meminit __zone_pcp_update(void *data)
 {
 	struct zone *zone = data;
 	int cpu;
-	unsigned long batch = zone_batchsize(zone), flags;
+	unsigned long batch = zone_batchsize(zone);
 
 	for_each_possible_cpu(cpu) {
 		struct per_cpu_pageset *pset;
@@ -6188,12 +6188,10 @@ static int __meminit __zone_pcp_update(void *data)
 		pset = per_cpu_ptr(zone->pageset, cpu);
 		pcp = &pset->pcp;

mm/page_alloc.c static int __meminit __zone_pcp_update(void *data) · line 6197

Quoted 10 lines
mm/page_alloc.c +0 -2
 
-		local_irq_save(flags);
 		if (pcp->count > 0)
 			free_pcppages_bulk(zone, pcp->count, pcp);
 		drain_zonestat(zone, pset);
 		setup_pageset(pset, batch);
-		local_irq_restore(flags);
 	}
 	return 0;
 }

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

#2 lore raw
Quoted 7 lines · Srivatsa S. Bhat
Quoted 4 lines

__zone_pcp_update() is called via stop_machine(), which already disables local irq.

What was reviewed?

Quoted 21 lines
Quoted 21 lines
---
 mm/page_alloc.c | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)
mm/page_alloc.c +1 -1
index bac3107..b46b54a 100644
@@ -6179,7 +6179,7 @@ static int __meminit __zone_pcp_update(void *data)
 {
 	struct zone *zone = data;
 	int cpu;
-	unsigned long batch = zone_batchsize(zone), flags;
+	unsigned long batch = zone_batchsize(zone);
 
 	for_each_possible_cpu(cpu) {
 		struct per_cpu_pageset *pset;
@@ -6188,12 +6188,10 @@ static int __meminit __zone_pcp_update(void *data)
 		pset = per_cpu_ptr(zone->pageset, cpu);
 		pcp = &pset->pcp;
 
Quoted 6 lines

mm/page_alloc.c static int __meminit __zone_pcp_update(void *data) · line 6196

Quoted 6 lines
mm/page_alloc.c +0 -2
-		local_irq_save(flags);
 		if (pcp->count > 0)
 			free_pcppages_bulk(zone, pcp->count, pcp);
 		drain_zonestat(zone, pset);
 		setup_pageset(pset, batch);
-		local_irq_restore(flags);

This seems like a fine cleanup because stop_machine() disable irqs, but it appears like there is two problems with this function already:

- it's doing for_each_possible_cpu() internally, why? local_irq_save() works on the local cpu and won't protect per_cpu_ptr(zone->pageset, cpu)->pcp of some random cpu, and

- setup_pageset() is what is ultimately responsible for doing pcp->count = 0 after free_pcppages_bulk(), but what happens if pcp->count is read in between the two on the cpu that has not disabled irqs?

You can't just do

	for_each_possible_cpu(cpu) {
		unsigned long flags;

		local_irq_save(flags);
		...
		local_irq_restore(flags);
	}

This is just disabling irqs locally over and over again, not on the cpu you're manipulating in its per-cpu critical section.

I don't think we hit this because onlining and offlining memory isn't a very common operation, but it doesn't change the fact that it's broken.

Quoted 4 lines

mm/page_alloc.c static int __meminit __zone_pcp_update(void *data) · line 6197

Quoted 4 lines
mm/page_alloc.c +0 -0
 	}
 	return 0;
 }

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

#3 lore raw
Quoted 11 lines · David Rientjes
Quoted 5 lines · Srivatsa S. Bhat
Quoted 5 lines

__zone_pcp_update() is called via stop_machine(), which already disables local irq.

  mm/page_alloc.c | 4 +---
  1 file changed, 1 insertion(+), 3 deletions(-)

This seems like a fine cleanup because stop_machine() disable irqs, but it appears like there is two problems with this function already:

Re-examining this, I've realized that my previous patchset containing
"mm/page_alloc: convert zone_pcp_update() to rely on memory barriers
instead of stop_machine()"

already went through and fixed this up (the right way). So ignore this patch.

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

#4 lore raw
Quoted 6 lines · Cody P Schafer

Re-examining this, I've realized that my previous patchset containing
"mm/page_alloc: convert zone_pcp_update() to rely on memory barriers
instead of stop_machine()"

already went through and fixed this up (the right way). So ignore this patch.

If you're referring to

mm-page_alloc-factor-out-setting-of-pcp-high-and-pcp-batch.patch mm-page_alloc-prevent-concurrent-updaters-of-pcp-batch-and-high.patch mm-page_alloc-insert-memory-barriers-to-allow-async-update-of-pcp-batch-and-high.patch mm-page_alloc-protect-pcp-batch-accesses-with-access_once.patch mm-page_alloc-convert-zone_pcp_update-to-rely-on-memory-barriers-instead-of-stop_machine.patch mm-page_alloc-when-handling-percpu_pagelist_fraction-dont-unneedly-recalulate-high.patch mm-page_alloc-factor-setup_pageset-into-pageset_init-and-pageset_set_batch.patch mm-page_alloc-relocate-comment-to-be-directly-above-code-it-refers-to.patch mm-page_alloc-factor-zone_pageset_init-out-of-setup_zone_pageset.patch mm-page_alloc-in-zone_pcp_update-uze-zone_pageset_init.patch mm-page_alloc-rename-setup_pagelist_highmark-to-match-naming-of-pageset_set_batch.patch

from -mm then I'll review them separately because they have their own issues. Thanks.

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

#5 lore raw
Quoted 23 lines · David Rientjes
Quoted 6 lines · Cody P Schafer

Re-examining this, I've realized that my previous patchset containing
"mm/page_alloc: convert zone_pcp_update() to rely on memory barriers
instead of stop_machine()"

already went through and fixed this up (the right way). So ignore this patch.

If you're referring to

mm-page_alloc-factor-out-setting-of-pcp-high-and-pcp-batch.patch mm-page_alloc-prevent-concurrent-updaters-of-pcp-batch-and-high.patch mm-page_alloc-insert-memory-barriers-to-allow-async-update-of-pcp-batch-and-high.patch mm-page_alloc-protect-pcp-batch-accesses-with-access_once.patch mm-page_alloc-convert-zone_pcp_update-to-rely-on-memory-barriers-instead-of-stop_machine.patch mm-page_alloc-when-handling-percpu_pagelist_fraction-dont-unneedly-recalulate-high.patch mm-page_alloc-factor-setup_pageset-into-pageset_init-and-pageset_set_batch.patch mm-page_alloc-relocate-comment-to-be-directly-above-code-it-refers-to.patch mm-page_alloc-factor-zone_pageset_init-out-of-setup_zone_pageset.patch mm-page_alloc-in-zone_pcp_update-uze-zone_pageset_init.patch mm-page_alloc-rename-setup_pagelist_highmark-to-match-naming-of-pageset_set_batch.patch

Quoted 3 lines

from -mm then I'll review them separately because they have their own issues. Thanks.

I am. You may also want to note mm-page_alloc-dont-re-init-pageset-in-zone_pcp_update.patch

which fixes a bug in that patchset.

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

#6 lore raw
In reply to David Rientjes
Quoted 5 lines · David Rientjes
Quoted 3 lines · Srivatsa S. Bhat
Quoted 3 lines

__zone_pcp_update() is called via stop_machine(), which already disables local irq.

Quoted 7 lines
Quoted 4 lines
Quoted 1 line

What was reviewed?

See below.

Quoted 21 lines
Quoted 21 lines
Quoted 21 lines
---
 mm/page_alloc.c | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)
mm/page_alloc.c +1 -1
index bac3107..b46b54a 100644
@@ -6179,7 +6179,7 @@ static int __meminit __zone_pcp_update(void *data)
 {
 	struct zone *zone = data;
 	int cpu;
-	unsigned long batch = zone_batchsize(zone), flags;
+	unsigned long batch = zone_batchsize(zone);
 
 	for_each_possible_cpu(cpu) {
 		struct per_cpu_pageset *pset;
@@ -6188,12 +6188,10 @@ static int __meminit __zone_pcp_update(void *data)
 		pset = per_cpu_ptr(zone->pageset, cpu);
 		pcp = &pset->pcp;
 
Quoted 8 lines
Quoted 6 lines

mm/page_alloc.c static int __meminit __zone_pcp_update(void *data) · line 6196

Quoted 6 lines
mm/page_alloc.c +0 -2
-		local_irq_save(flags);
 		if (pcp->count > 0)
 			free_pcppages_bulk(zone, pcp->count, pcp);
 		drain_zonestat(zone, pset);
 		setup_pageset(pset, batch);
-		local_irq_restore(flags);

This seems like a fine cleanup because stop_machine() disable irqs,

I hope you are not missing the fact that stop_machine() disables irqs on *all* CPUs.

Quoted 7 lines

but it
appears like there is two problems with this function already:

- it's doing for_each_possible_cpu() internally, why? local_irq_save() works on the local cpu and won't protect per_cpu_ptr(zone->pageset, cpu)->pcp of some random cpu, and

stop_machine() allows only _your function_ to run and nothing else, on the entire system. All other CPUs loop with interrupts disabled until the function is completed.

Quoted 5 lines

- setup_pageset() is what is ultimately responsible for doing pcp->count = 0 after free_pcppages_bulk(), but what happens if pcp->count is read in between the two on the cpu that has not disabled irqs?

Nobody can do anything else when this function runs. That's precisely why its named as stop-*machine*.

Quoted 10 lines

You can't just do

	for_each_possible_cpu(cpu) {
		unsigned long flags;

		local_irq_save(flags);
		...
		local_irq_restore(flags);
	}
Quoted 3 lines

This is just disabling irqs locally over and over again, not on the cpu you're manipulating in its per-cpu critical section.

stop-machine() takes care of disabling irqs on every online CPU.

Quoted 3 lines

I don't think we hit this because onlining and offlining memory isn't a very common operation, but it doesn't change the fact that it's broken.

If __zone_pcp_update() is called only from stop_machine() (and looking at the current code, that's true), then there is no problem, due to the reasons explained above.

Regards,
Srivatsa S. Bhat

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>