* [PATCH] mm/page_alloc: call check_new_pages() while zone spinlock is not held
@ 2022-03-04 1:59 Eric Dumazet
2022-03-04 2:09 ` Eric Dumazet
2022-03-04 8:26 ` Vlastimil Babka
0 siblings, 2 replies; 5+ messages in thread
From: Eric Dumazet @ 2022-03-04 1:59 UTC (permalink / raw)
To: Andrew Morton
Cc: linux-kernel, linux-mm, Eric Dumazet, Eric Dumazet, Mel Gorman,
Vlastimil Babka, Michal Hocko, Shakeel Butt, Wei Xu, Greg Thelen,
Hugh Dickins, David Rientjes
From: Eric Dumazet <edumazet@google.com>
For high order pages not using pcp, rmqueue() is currently calling
the costly check_new_pages() while zone spinlock is held.
This is not needed, we can release the spinlock sooner to reduce
zone spinlock contention.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Wei Xu <weixugc@google.com>
Cc: Greg Thelen <gthelen@google.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: David Rientjes <rientjes@google.com>
---
mm/page_alloc.c | 17 +++++++----------
1 file changed, 7 insertions(+), 10 deletions(-)
diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index 3589febc6d31928f850ebe5a4015ddc40e0469f3..0890a65f8cc2259e82bc1f5ba95a592fb30f9fb8 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -3685,7 +3685,6 @@ struct page *rmqueue(struct zone *preferred_zone,
gfp_t gfp_flags, unsigned int alloc_flags,
int migratetype)
{
- unsigned long flags;
struct page *page;
if (likely(pcp_allowed_order(order))) {
@@ -3706,10 +3705,12 @@ struct page *rmqueue(struct zone *preferred_zone,
* allocate greater than order-1 page units with __GFP_NOFAIL.
*/
WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) && (order > 1));
- spin_lock_irqsave(&zone->lock, flags);
do {
+ unsigned long flags;
+
page = NULL;
+ spin_lock_irqsave(&zone->lock, flags);
/*
* order-0 request can reach here when the pcplist is skipped
* due to non-CMA allocation context. HIGHATOMIC area is
@@ -3723,13 +3724,13 @@ struct page *rmqueue(struct zone *preferred_zone,
}
if (!page)
page = __rmqueue(zone, order, migratetype, alloc_flags);
- } while (page && check_new_pages(page, order));
- if (!page)
- goto failed;
+ spin_unlock_irqrestore(&zone->lock, flags);
+ if (!page)
+ return NULL;
+ } while (check_new_pages(page, order));
__mod_zone_freepage_state(zone, -(1 << order),
get_pcppage_migratetype(page));
- spin_unlock_irqrestore(&zone->lock, flags);
__count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order);
zone_statistics(preferred_zone, zone, 1);
@@ -3743,10 +3744,6 @@ struct page *rmqueue(struct zone *preferred_zone,
VM_BUG_ON_PAGE(page && bad_range(zone, page), page);
return page;
-
-failed:
- spin_unlock_irqrestore(&zone->lock, flags);
- return NULL;
}
#ifdef CONFIG_FAIL_PAGE_ALLOC
--
2.35.1.616.g0bdcbb4464-goog
^ permalink raw reply related [flat|nested] 5+ messages in thread
* Re: [PATCH] mm/page_alloc: call check_new_pages() while zone spinlock is not held
2022-03-04 1:59 [PATCH] mm/page_alloc: call check_new_pages() while zone spinlock is not held Eric Dumazet
@ 2022-03-04 2:09 ` Eric Dumazet
2022-03-04 8:32 ` Vlastimil Babka
2022-03-04 8:26 ` Vlastimil Babka
1 sibling, 1 reply; 5+ messages in thread
From: Eric Dumazet @ 2022-03-04 2:09 UTC (permalink / raw)
To: Eric Dumazet
Cc: Andrew Morton, linux-kernel, linux-mm, Mel Gorman,
Vlastimil Babka, Michal Hocko, Shakeel Butt, Wei Xu, Greg Thelen,
Hugh Dickins, David Rientjes
On Thu, Mar 3, 2022 at 5:59 PM Eric Dumazet <eric.dumazet@gmail.com> wrote:
>
> From: Eric Dumazet <edumazet@google.com>
>
> For high order pages not using pcp, rmqueue() is currently calling
> the costly check_new_pages() while zone spinlock is held.
>
> This is not needed, we can release the spinlock sooner to reduce
> zone spinlock contention.
>
> Signed-off-by: Eric Dumazet <edumazet@google.com>
> Cc: Mel Gorman <mgorman@techsingularity.net>
> Cc: Vlastimil Babka <vbabka@suse.cz>
> Cc: Michal Hocko <mhocko@kernel.org>
> Cc: Shakeel Butt <shakeelb@google.com>
> Cc: Wei Xu <weixugc@google.com>
> Cc: Greg Thelen <gthelen@google.com>
> Cc: Hugh Dickins <hughd@google.com>
> Cc: David Rientjes <rientjes@google.com>
> ---
> mm/page_alloc.c | 17 +++++++----------
> 1 file changed, 7 insertions(+), 10 deletions(-)
>
> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> index 3589febc6d31928f850ebe5a4015ddc40e0469f3..0890a65f8cc2259e82bc1f5ba95a592fb30f9fb8 100644
> --- a/mm/page_alloc.c
> +++ b/mm/page_alloc.c
> @@ -3685,7 +3685,6 @@ struct page *rmqueue(struct zone *preferred_zone,
> gfp_t gfp_flags, unsigned int alloc_flags,
> int migratetype)
> {
> - unsigned long flags;
> struct page *page;
>
> if (likely(pcp_allowed_order(order))) {
> @@ -3706,10 +3705,12 @@ struct page *rmqueue(struct zone *preferred_zone,
> * allocate greater than order-1 page units with __GFP_NOFAIL.
> */
> WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) && (order > 1));
> - spin_lock_irqsave(&zone->lock, flags);
>
> do {
> + unsigned long flags;
> +
> page = NULL;
> + spin_lock_irqsave(&zone->lock, flags);
> /*
> * order-0 request can reach here when the pcplist is skipped
> * due to non-CMA allocation context. HIGHATOMIC area is
> @@ -3723,13 +3724,13 @@ struct page *rmqueue(struct zone *preferred_zone,
> }
> if (!page)
> page = __rmqueue(zone, order, migratetype, alloc_flags);
> - } while (page && check_new_pages(page, order));
> - if (!page)
> - goto failed;
> + spin_unlock_irqrestore(&zone->lock, flags);
> + if (!page)
> + return NULL;
> + } while (check_new_pages(page, order));
>
Oh well, it seems hard irqs have to be disabled when calling the
following function.
I will send a V2.
> __mod_zone_freepage_state(zone, -(1 << order),
> get_pcppage_migratetype(page));
> - spin_unlock_irqrestore(&zone->lock, flags);
>
> __count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order);
> zone_statistics(preferred_zone, zone, 1);
> @@ -3743,10 +3744,6 @@ struct page *rmqueue(struct zone *preferred_zone,
>
> VM_BUG_ON_PAGE(page && bad_range(zone, page), page);
> return page;
> -
> -failed:
> - spin_unlock_irqrestore(&zone->lock, flags);
> - return NULL;
> }
>
> #ifdef CONFIG_FAIL_PAGE_ALLOC
> --
> 2.35.1.616.g0bdcbb4464-goog
>
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH] mm/page_alloc: call check_new_pages() while zone spinlock is not held
2022-03-04 1:59 [PATCH] mm/page_alloc: call check_new_pages() while zone spinlock is not held Eric Dumazet
2022-03-04 2:09 ` Eric Dumazet
@ 2022-03-04 8:26 ` Vlastimil Babka
1 sibling, 0 replies; 5+ messages in thread
From: Vlastimil Babka @ 2022-03-04 8:26 UTC (permalink / raw)
To: Eric Dumazet, Andrew Morton
Cc: linux-kernel, linux-mm, Eric Dumazet, Mel Gorman, Michal Hocko,
Shakeel Butt, Wei Xu, Greg Thelen, Hugh Dickins, David Rientjes
On 3/4/22 02:59, Eric Dumazet wrote:
> From: Eric Dumazet <edumazet@google.com>
>
> For high order pages not using pcp, rmqueue() is currently calling
> the costly check_new_pages() while zone spinlock is held.
>
> This is not needed, we can release the spinlock sooner to reduce
> zone spinlock contention.
>
> Signed-off-by: Eric Dumazet <edumazet@google.com>
> Cc: Mel Gorman <mgorman@techsingularity.net>
> Cc: Vlastimil Babka <vbabka@suse.cz>
> Cc: Michal Hocko <mhocko@kernel.org>
> Cc: Shakeel Butt <shakeelb@google.com>
> Cc: Wei Xu <weixugc@google.com>
> Cc: Greg Thelen <gthelen@google.com>
> Cc: Hugh Dickins <hughd@google.com>
> Cc: David Rientjes <rientjes@google.com>
> ---
> mm/page_alloc.c | 17 +++++++----------
> 1 file changed, 7 insertions(+), 10 deletions(-)
Just a quick reply from what AFAIK immediately see as a issue.
>
> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> index 3589febc6d31928f850ebe5a4015ddc40e0469f3..0890a65f8cc2259e82bc1f5ba95a592fb30f9fb8 100644
> --- a/mm/page_alloc.c
> +++ b/mm/page_alloc.c
> @@ -3685,7 +3685,6 @@ struct page *rmqueue(struct zone *preferred_zone,
> gfp_t gfp_flags, unsigned int alloc_flags,
> int migratetype)
> {
> - unsigned long flags;
> struct page *page;
>
> if (likely(pcp_allowed_order(order))) {
> @@ -3706,10 +3705,12 @@ struct page *rmqueue(struct zone *preferred_zone,
> * allocate greater than order-1 page units with __GFP_NOFAIL.
> */
> WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) && (order > 1));
> - spin_lock_irqsave(&zone->lock, flags);
>
> do {
> + unsigned long flags;
> +
> page = NULL;
> + spin_lock_irqsave(&zone->lock, flags);
> /*
> * order-0 request can reach here when the pcplist is skipped
> * due to non-CMA allocation context. HIGHATOMIC area is
> @@ -3723,13 +3724,13 @@ struct page *rmqueue(struct zone *preferred_zone,
> }
> if (!page)
> page = __rmqueue(zone, order, migratetype, alloc_flags);
> - } while (page && check_new_pages(page, order));
> - if (!page)
> - goto failed;
> + spin_unlock_irqrestore(&zone->lock, flags);
> + if (!page)
> + return NULL;
> + } while (check_new_pages(page, order));
>
> __mod_zone_freepage_state(zone, -(1 << order),
> get_pcppage_migratetype(page));
> - spin_unlock_irqrestore(&zone->lock, flags);
__mod_zone_freepage_state() outside the lock is AFAIK unsafe.
>
> __count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order);
> zone_statistics(preferred_zone, zone, 1);
> @@ -3743,10 +3744,6 @@ struct page *rmqueue(struct zone *preferred_zone,
>
> VM_BUG_ON_PAGE(page && bad_range(zone, page), page);
> return page;
> -
> -failed:
> - spin_unlock_irqrestore(&zone->lock, flags);
> - return NULL;
> }
>
> #ifdef CONFIG_FAIL_PAGE_ALLOC
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH] mm/page_alloc: call check_new_pages() while zone spinlock is not held
2022-03-04 2:09 ` Eric Dumazet
@ 2022-03-04 8:32 ` Vlastimil Babka
2022-03-04 16:50 ` Eric Dumazet
0 siblings, 1 reply; 5+ messages in thread
From: Vlastimil Babka @ 2022-03-04 8:32 UTC (permalink / raw)
To: Eric Dumazet, Eric Dumazet
Cc: Andrew Morton, linux-kernel, linux-mm, Mel Gorman, Michal Hocko,
Shakeel Butt, Wei Xu, Greg Thelen, Hugh Dickins, David Rientjes
On 3/4/22 03:09, Eric Dumazet wrote:
> On Thu, Mar 3, 2022 at 5:59 PM Eric Dumazet <eric.dumazet@gmail.com> wrote:
>>
>> From: Eric Dumazet <edumazet@google.com>
>>
>> For high order pages not using pcp, rmqueue() is currently calling
>> the costly check_new_pages() while zone spinlock is held.
>>
>> This is not needed, we can release the spinlock sooner to reduce
>> zone spinlock contention.
>>
>> Signed-off-by: Eric Dumazet <edumazet@google.com>
>> Cc: Mel Gorman <mgorman@techsingularity.net>
>> Cc: Vlastimil Babka <vbabka@suse.cz>
>> Cc: Michal Hocko <mhocko@kernel.org>
>> Cc: Shakeel Butt <shakeelb@google.com>
>> Cc: Wei Xu <weixugc@google.com>
>> Cc: Greg Thelen <gthelen@google.com>
>> Cc: Hugh Dickins <hughd@google.com>
>> Cc: David Rientjes <rientjes@google.com>
>> ---
>> mm/page_alloc.c | 17 +++++++----------
>> 1 file changed, 7 insertions(+), 10 deletions(-)
>>
>> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
>> index 3589febc6d31928f850ebe5a4015ddc40e0469f3..0890a65f8cc2259e82bc1f5ba95a592fb30f9fb8 100644
>> --- a/mm/page_alloc.c
>> +++ b/mm/page_alloc.c
>> @@ -3685,7 +3685,6 @@ struct page *rmqueue(struct zone *preferred_zone,
>> gfp_t gfp_flags, unsigned int alloc_flags,
>> int migratetype)
>> {
>> - unsigned long flags;
>> struct page *page;
>>
>> if (likely(pcp_allowed_order(order))) {
>> @@ -3706,10 +3705,12 @@ struct page *rmqueue(struct zone *preferred_zone,
>> * allocate greater than order-1 page units with __GFP_NOFAIL.
>> */
>> WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) && (order > 1));
>> - spin_lock_irqsave(&zone->lock, flags);
>>
>> do {
>> + unsigned long flags;
>> +
>> page = NULL;
>> + spin_lock_irqsave(&zone->lock, flags);
>> /*
>> * order-0 request can reach here when the pcplist is skipped
>> * due to non-CMA allocation context. HIGHATOMIC area is
>> @@ -3723,13 +3724,13 @@ struct page *rmqueue(struct zone *preferred_zone,
>> }
>> if (!page)
>> page = __rmqueue(zone, order, migratetype, alloc_flags);
>> - } while (page && check_new_pages(page, order));
>> - if (!page)
>> - goto failed;
>> + spin_unlock_irqrestore(&zone->lock, flags);
>> + if (!page)
>> + return NULL;
>> + } while (check_new_pages(page, order));
>>
>
> Oh well, it seems hard irqs have to be disabled when calling the
> following function.
Ah, you found out already. Well maybe it could simply be moved inside the
loop under the locked section and always done when page != NULL? I mean if
check_new_pages() fails we just leak the problematic pages anyway so they
are no longer free to allocate anymore and we should not count them as such.
> I will send a V2.
>
>> __mod_zone_freepage_state(zone, -(1 << order),
>> get_pcppage_migratetype(page));
>> - spin_unlock_irqrestore(&zone->lock, flags);
>>
>> __count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order);
>> zone_statistics(preferred_zone, zone, 1);
>> @@ -3743,10 +3744,6 @@ struct page *rmqueue(struct zone *preferred_zone,
>>
>> VM_BUG_ON_PAGE(page && bad_range(zone, page), page);
>> return page;
>> -
>> -failed:
>> - spin_unlock_irqrestore(&zone->lock, flags);
>> - return NULL;
>> }
>>
>> #ifdef CONFIG_FAIL_PAGE_ALLOC
>> --
>> 2.35.1.616.g0bdcbb4464-goog
>>
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH] mm/page_alloc: call check_new_pages() while zone spinlock is not held
2022-03-04 8:32 ` Vlastimil Babka
@ 2022-03-04 16:50 ` Eric Dumazet
0 siblings, 0 replies; 5+ messages in thread
From: Eric Dumazet @ 2022-03-04 16:50 UTC (permalink / raw)
To: Vlastimil Babka
Cc: Eric Dumazet, Andrew Morton, linux-kernel, linux-mm, Mel Gorman,
Michal Hocko, Shakeel Butt, Wei Xu, Greg Thelen, Hugh Dickins,
David Rientjes
On Fri, Mar 4, 2022 at 12:32 AM Vlastimil Babka <vbabka@suse.cz> wrote:
>
>
> Ah, you found out already. Well maybe it could simply be moved inside the
> loop under the locked section and always done when page != NULL? I mean if
> check_new_pages() fails we just leak the problematic pages anyway so they
> are no longer free to allocate anymore and we should not count them as such.
>
Yes, and I guess that calling __mod_zone_freepage_state() for the page
about to be leaked
is just fine.
This means we can call check_newpages() without holding zone spinlock
or hard irqs being masked.
I will test something like:
diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index 3589febc6d31928f850ebe5a4015ddc40e0469f3..1804287c1b792b8aa0e964b17eb002b6b1115258
100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -3706,10 +3706,10 @@ struct page *rmqueue(struct zone *preferred_zone,
* allocate greater than order-1 page units with __GFP_NOFAIL.
*/
WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) && (order > 1));
- spin_lock_irqsave(&zone->lock, flags);
do {
page = NULL;
+ spin_lock_irqsave(&zone->lock, flags);
/*
* order-0 request can reach here when the pcplist is skipped
* due to non-CMA allocation context. HIGHATOMIC area is
@@ -3721,15 +3721,15 @@ struct page *rmqueue(struct zone *preferred_zone,
if (page)
trace_mm_page_alloc_zone_locked(page,
order, migratetype);
}
- if (!page)
+ if (!page) {
page = __rmqueue(zone, order, migratetype, alloc_flags);
- } while (page && check_new_pages(page, order));
- if (!page)
- goto failed;
-
- __mod_zone_freepage_state(zone, -(1 << order),
- get_pcppage_migratetype(page));
- spin_unlock_irqrestore(&zone->lock, flags);
+ if (!page)
+ goto failed;
+ }
+ __mod_zone_freepage_state(zone, -(1 << order),
+ get_pcppage_migratetype(page));
+ spin_unlock_irqrestore(&zone->lock, flags);
+ } while (check_new_pages(page, order));
__count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order);
zone_statistics(preferred_zone, zone, 1);
^ permalink raw reply [flat|nested] 5+ messages in thread
end of thread, other threads:[~2022-03-04 16:50 UTC | newest]
Thread overview: 5+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2022-03-04 1:59 [PATCH] mm/page_alloc: call check_new_pages() while zone spinlock is not held Eric Dumazet
2022-03-04 2:09 ` Eric Dumazet
2022-03-04 8:32 ` Vlastimil Babka
2022-03-04 16:50 ` Eric Dumazet
2022-03-04 8:26 ` Vlastimil Babka
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).