When page allocation in direct reclaim path fails, the system will
make one attempt to shrink per-cpu page lists and free pages from
high alloc reserves. Draining per-cpu pages into buddy allocator can
be a very slow operation because it's done using workqueues and the
task in direct reclaim waits for all of them to finish before
proceeding. Currently this time is not accounted as psi memory stall.
While testing mobile devices under extreme memory pressure, when
allocations are failing during direct reclaim, we notices that psi
events which would be expected in such conditions were not triggered.
After profiling these cases it was determined that the reason for
missing psi events was that a big chunk of time spent in direct
reclaim is not accounted as memory stall, therefore psi would not
reach the levels at which an event is generated. Further investigation
revealed that the bulk of that unaccounted time was spent inside
drain_all_pages call.
A typical captured case when drain_all_pages path gets activated:
__alloc_pages_slowpath took 44.644.613ns
__perform_reclaim took 751.668ns (1.7%)
drain_all_pages took 43.887.167ns (98.3%)
PSI in this case records the time spent in __perform_reclaim but
ignores drain_all_pages, IOW it misses 98.3% of the time spent in
__alloc_pages_slowpath.
Annotate __alloc_pages_direct_reclaim in its entirety so that delays
from handling page allocation failure in the direct reclaim path are
accounted as memory stall.
Reported-by: Tim Murray <[email protected]>
Signed-off-by: Suren Baghdasaryan <[email protected]>
Acked-by: Johannes Weiner <[email protected]>
---
changes in v3:
- Moved psi_memstall_leave after the "out" label
mm/page_alloc.c | 10 ++++++----
1 file changed, 6 insertions(+), 4 deletions(-)
diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index 3589febc6d31..029bceb79861 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -4595,13 +4595,12 @@ __perform_reclaim(gfp_t gfp_mask, unsigned int order,
const struct alloc_context *ac)
{
unsigned int noreclaim_flag;
- unsigned long pflags, progress;
+ unsigned long progress;
cond_resched();
/* We now go into synchronous reclaim */
cpuset_memory_pressure_bump();
- psi_memstall_enter(&pflags);
fs_reclaim_acquire(gfp_mask);
noreclaim_flag = memalloc_noreclaim_save();
@@ -4610,7 +4609,6 @@ __perform_reclaim(gfp_t gfp_mask, unsigned int order,
memalloc_noreclaim_restore(noreclaim_flag);
fs_reclaim_release(gfp_mask);
- psi_memstall_leave(&pflags);
cond_resched();
@@ -4624,11 +4622,13 @@ __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
unsigned long *did_some_progress)
{
struct page *page = NULL;
+ unsigned long pflags;
bool drained = false;
+ psi_memstall_enter(&pflags);
*did_some_progress = __perform_reclaim(gfp_mask, order, ac);
if (unlikely(!(*did_some_progress)))
- return NULL;
+ goto out;
retry:
page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
@@ -4644,6 +4644,8 @@ __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
drained = true;
goto retry;
}
+out:
+ psi_memstall_leave(&pflags);
return page;
}
--
2.35.1.473.g83b2b277ed-goog
On Wed, Feb 23, 2022 at 11:48:12AM -0800, Suren Baghdasaryan wrote:
> When page allocation in direct reclaim path fails, the system will
> make one attempt to shrink per-cpu page lists and free pages from
> high alloc reserves. Draining per-cpu pages into buddy allocator can
> be a very slow operation because it's done using workqueues and the
> task in direct reclaim waits for all of them to finish before
> proceeding. Currently this time is not accounted as psi memory stall.
> While testing mobile devices under extreme memory pressure, when
> allocations are failing during direct reclaim, we notices that psi
> events which would be expected in such conditions were not triggered.
> After profiling these cases it was determined that the reason for
> missing psi events was that a big chunk of time spent in direct
> reclaim is not accounted as memory stall, therefore psi would not
> reach the levels at which an event is generated. Further investigation
> revealed that the bulk of that unaccounted time was spent inside
> drain_all_pages call.
> A typical captured case when drain_all_pages path gets activated:
> __alloc_pages_slowpath took 44.644.613ns
> __perform_reclaim took 751.668ns (1.7%)
> drain_all_pages took 43.887.167ns (98.3%)
> PSI in this case records the time spent in __perform_reclaim but
> ignores drain_all_pages, IOW it misses 98.3% of the time spent in
> __alloc_pages_slowpath.
> Annotate __alloc_pages_direct_reclaim in its entirety so that delays
> from handling page allocation failure in the direct reclaim path are
> accounted as memory stall.
> Reported-by: Tim Murray <[email protected]>
> Signed-off-by: Suren Baghdasaryan <[email protected]>
> Acked-by: Johannes Weiner <[email protected]>
Reviewed-by: Shakeel Butt <[email protected]>
On Wed 23-02-22 11:48:12, Suren Baghdasaryan wrote:
> When page allocation in direct reclaim path fails, the system will
> make one attempt to shrink per-cpu page lists and free pages from
> high alloc reserves. Draining per-cpu pages into buddy allocator can
> be a very slow operation because it's done using workqueues and the
> task in direct reclaim waits for all of them to finish before
> proceeding. Currently this time is not accounted as psi memory stall.
>
> While testing mobile devices under extreme memory pressure, when
> allocations are failing during direct reclaim, we notices that psi
> events which would be expected in such conditions were not triggered.
> After profiling these cases it was determined that the reason for
> missing psi events was that a big chunk of time spent in direct
> reclaim is not accounted as memory stall, therefore psi would not
> reach the levels at which an event is generated. Further investigation
> revealed that the bulk of that unaccounted time was spent inside
> drain_all_pages call.
>
> A typical captured case when drain_all_pages path gets activated:
>
> __alloc_pages_slowpath took 44.644.613ns
> __perform_reclaim took 751.668ns (1.7%)
> drain_all_pages took 43.887.167ns (98.3%)
Although the draining is done in the slow path these numbers suggest
that we should really reconsider the use of WQ both for draining and
other purposes (like vmstats).
> PSI in this case records the time spent in __perform_reclaim but
> ignores drain_all_pages, IOW it misses 98.3% of the time spent in
> __alloc_pages_slowpath.
>
> Annotate __alloc_pages_direct_reclaim in its entirety so that delays
> from handling page allocation failure in the direct reclaim path are
> accounted as memory stall.
>
> Reported-by: Tim Murray <[email protected]>
> Signed-off-by: Suren Baghdasaryan <[email protected]>
> Acked-by: Johannes Weiner <[email protected]>
Acked-by: Michal Hocko <[email protected]>
Thanks!
> ---
> changes in v3:
> - Moved psi_memstall_leave after the "out" label
>
> mm/page_alloc.c | 10 ++++++----
> 1 file changed, 6 insertions(+), 4 deletions(-)
>
> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> index 3589febc6d31..029bceb79861 100644
> --- a/mm/page_alloc.c
> +++ b/mm/page_alloc.c
> @@ -4595,13 +4595,12 @@ __perform_reclaim(gfp_t gfp_mask, unsigned int order,
> const struct alloc_context *ac)
> {
> unsigned int noreclaim_flag;
> - unsigned long pflags, progress;
> + unsigned long progress;
>
> cond_resched();
>
> /* We now go into synchronous reclaim */
> cpuset_memory_pressure_bump();
> - psi_memstall_enter(&pflags);
> fs_reclaim_acquire(gfp_mask);
> noreclaim_flag = memalloc_noreclaim_save();
>
> @@ -4610,7 +4609,6 @@ __perform_reclaim(gfp_t gfp_mask, unsigned int order,
>
> memalloc_noreclaim_restore(noreclaim_flag);
> fs_reclaim_release(gfp_mask);
> - psi_memstall_leave(&pflags);
>
> cond_resched();
>
> @@ -4624,11 +4622,13 @@ __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
> unsigned long *did_some_progress)
> {
> struct page *page = NULL;
> + unsigned long pflags;
> bool drained = false;
>
> + psi_memstall_enter(&pflags);
> *did_some_progress = __perform_reclaim(gfp_mask, order, ac);
> if (unlikely(!(*did_some_progress)))
> - return NULL;
> + goto out;
>
> retry:
> page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
> @@ -4644,6 +4644,8 @@ __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
> drained = true;
> goto retry;
> }
> +out:
> + psi_memstall_leave(&pflags);
>
> return page;
> }
> --
> 2.35.1.473.g83b2b277ed-goog
--
Michal Hocko
SUSE Labs
On Thu, Feb 24, 2022 at 12:53 AM 'Michal Hocko' via kernel-team
<[email protected]> wrote:
>
> On Wed 23-02-22 11:48:12, Suren Baghdasaryan wrote:
> > When page allocation in direct reclaim path fails, the system will
> > make one attempt to shrink per-cpu page lists and free pages from
> > high alloc reserves. Draining per-cpu pages into buddy allocator can
> > be a very slow operation because it's done using workqueues and the
> > task in direct reclaim waits for all of them to finish before
> > proceeding. Currently this time is not accounted as psi memory stall.
> >
> > While testing mobile devices under extreme memory pressure, when
> > allocations are failing during direct reclaim, we notices that psi
> > events which would be expected in such conditions were not triggered.
> > After profiling these cases it was determined that the reason for
> > missing psi events was that a big chunk of time spent in direct
> > reclaim is not accounted as memory stall, therefore psi would not
> > reach the levels at which an event is generated. Further investigation
> > revealed that the bulk of that unaccounted time was spent inside
> > drain_all_pages call.
> >
> > A typical captured case when drain_all_pages path gets activated:
> >
> > __alloc_pages_slowpath took 44.644.613ns
> > __perform_reclaim took 751.668ns (1.7%)
> > drain_all_pages took 43.887.167ns (98.3%)
>
> Although the draining is done in the slow path these numbers suggest
> that we should really reconsider the use of WQ both for draining and
> other purposes (like vmstats).
Yep, I'm testing the kthread_create_worker_on_cpu approach suggested
by Petr. Will post it later today if nothing regresses.
>
> > PSI in this case records the time spent in __perform_reclaim but
> > ignores drain_all_pages, IOW it misses 98.3% of the time spent in
> > __alloc_pages_slowpath.
> >
> > Annotate __alloc_pages_direct_reclaim in its entirety so that delays
> > from handling page allocation failure in the direct reclaim path are
> > accounted as memory stall.
> >
> > Reported-by: Tim Murray <[email protected]>
> > Signed-off-by: Suren Baghdasaryan <[email protected]>
> > Acked-by: Johannes Weiner <[email protected]>
>
> Acked-by: Michal Hocko <[email protected]>
>
> Thanks!
>
> > ---
> > changes in v3:
> > - Moved psi_memstall_leave after the "out" label
> >
> > mm/page_alloc.c | 10 ++++++----
> > 1 file changed, 6 insertions(+), 4 deletions(-)
> >
> > diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> > index 3589febc6d31..029bceb79861 100644
> > --- a/mm/page_alloc.c
> > +++ b/mm/page_alloc.c
> > @@ -4595,13 +4595,12 @@ __perform_reclaim(gfp_t gfp_mask, unsigned int order,
> > const struct alloc_context *ac)
> > {
> > unsigned int noreclaim_flag;
> > - unsigned long pflags, progress;
> > + unsigned long progress;
> >
> > cond_resched();
> >
> > /* We now go into synchronous reclaim */
> > cpuset_memory_pressure_bump();
> > - psi_memstall_enter(&pflags);
> > fs_reclaim_acquire(gfp_mask);
> > noreclaim_flag = memalloc_noreclaim_save();
> >
> > @@ -4610,7 +4609,6 @@ __perform_reclaim(gfp_t gfp_mask, unsigned int order,
> >
> > memalloc_noreclaim_restore(noreclaim_flag);
> > fs_reclaim_release(gfp_mask);
> > - psi_memstall_leave(&pflags);
> >
> > cond_resched();
> >
> > @@ -4624,11 +4622,13 @@ __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
> > unsigned long *did_some_progress)
> > {
> > struct page *page = NULL;
> > + unsigned long pflags;
> > bool drained = false;
> >
> > + psi_memstall_enter(&pflags);
> > *did_some_progress = __perform_reclaim(gfp_mask, order, ac);
> > if (unlikely(!(*did_some_progress)))
> > - return NULL;
> > + goto out;
> >
> > retry:
> > page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
> > @@ -4644,6 +4644,8 @@ __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
> > drained = true;
> > goto retry;
> > }
> > +out:
> > + psi_memstall_leave(&pflags);
> >
> > return page;
> > }
> > --
> > 2.35.1.473.g83b2b277ed-goog
>
> --
> Michal Hocko
> SUSE Labs
>
> --
> To unsubscribe from this group and stop receiving emails from it, send an email to [email protected].
>
On Thu, Feb 24, 2022 at 8:28 AM Suren Baghdasaryan <[email protected]> wrote:
>
> On Thu, Feb 24, 2022 at 12:53 AM 'Michal Hocko' via kernel-team
> <[email protected]> wrote:
> >
> > On Wed 23-02-22 11:48:12, Suren Baghdasaryan wrote:
> > > When page allocation in direct reclaim path fails, the system will
> > > make one attempt to shrink per-cpu page lists and free pages from
> > > high alloc reserves. Draining per-cpu pages into buddy allocator can
> > > be a very slow operation because it's done using workqueues and the
> > > task in direct reclaim waits for all of them to finish before
> > > proceeding. Currently this time is not accounted as psi memory stall.
> > >
> > > While testing mobile devices under extreme memory pressure, when
> > > allocations are failing during direct reclaim, we notices that psi
> > > events which would be expected in such conditions were not triggered.
> > > After profiling these cases it was determined that the reason for
> > > missing psi events was that a big chunk of time spent in direct
> > > reclaim is not accounted as memory stall, therefore psi would not
> > > reach the levels at which an event is generated. Further investigation
> > > revealed that the bulk of that unaccounted time was spent inside
> > > drain_all_pages call.
> > >
> > > A typical captured case when drain_all_pages path gets activated:
> > >
> > > __alloc_pages_slowpath took 44.644.613ns
> > > __perform_reclaim took 751.668ns (1.7%)
> > > drain_all_pages took 43.887.167ns (98.3%)
> >
> > Although the draining is done in the slow path these numbers suggest
> > that we should really reconsider the use of WQ both for draining and
> > other purposes (like vmstats).
>
> Yep, I'm testing the kthread_create_worker_on_cpu approach suggested
> by Petr. Will post it later today if nothing regresses.
An RFC for kthreads approach is posted at
https://lore.kernel.org/all/[email protected]/
>
> >
> > > PSI in this case records the time spent in __perform_reclaim but
> > > ignores drain_all_pages, IOW it misses 98.3% of the time spent in
> > > __alloc_pages_slowpath.
> > >
> > > Annotate __alloc_pages_direct_reclaim in its entirety so that delays
> > > from handling page allocation failure in the direct reclaim path are
> > > accounted as memory stall.
> > >
> > > Reported-by: Tim Murray <[email protected]>
> > > Signed-off-by: Suren Baghdasaryan <[email protected]>
> > > Acked-by: Johannes Weiner <[email protected]>
> >
> > Acked-by: Michal Hocko <[email protected]>
> >
> > Thanks!
> >
> > > ---
> > > changes in v3:
> > > - Moved psi_memstall_leave after the "out" label
> > >
> > > mm/page_alloc.c | 10 ++++++----
> > > 1 file changed, 6 insertions(+), 4 deletions(-)
> > >
> > > diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> > > index 3589febc6d31..029bceb79861 100644
> > > --- a/mm/page_alloc.c
> > > +++ b/mm/page_alloc.c
> > > @@ -4595,13 +4595,12 @@ __perform_reclaim(gfp_t gfp_mask, unsigned int order,
> > > const struct alloc_context *ac)
> > > {
> > > unsigned int noreclaim_flag;
> > > - unsigned long pflags, progress;
> > > + unsigned long progress;
> > >
> > > cond_resched();
> > >
> > > /* We now go into synchronous reclaim */
> > > cpuset_memory_pressure_bump();
> > > - psi_memstall_enter(&pflags);
> > > fs_reclaim_acquire(gfp_mask);
> > > noreclaim_flag = memalloc_noreclaim_save();
> > >
> > > @@ -4610,7 +4609,6 @@ __perform_reclaim(gfp_t gfp_mask, unsigned int order,
> > >
> > > memalloc_noreclaim_restore(noreclaim_flag);
> > > fs_reclaim_release(gfp_mask);
> > > - psi_memstall_leave(&pflags);
> > >
> > > cond_resched();
> > >
> > > @@ -4624,11 +4622,13 @@ __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
> > > unsigned long *did_some_progress)
> > > {
> > > struct page *page = NULL;
> > > + unsigned long pflags;
> > > bool drained = false;
> > >
> > > + psi_memstall_enter(&pflags);
> > > *did_some_progress = __perform_reclaim(gfp_mask, order, ac);
> > > if (unlikely(!(*did_some_progress)))
> > > - return NULL;
> > > + goto out;
> > >
> > > retry:
> > > page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
> > > @@ -4644,6 +4644,8 @@ __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
> > > drained = true;
> > > goto retry;
> > > }
> > > +out:
> > > + psi_memstall_leave(&pflags);
> > >
> > > return page;
> > > }
> > > --
> > > 2.35.1.473.g83b2b277ed-goog
> >
> > --
> > Michal Hocko
> > SUSE Labs
> >
> > --
> > To unsubscribe from this group and stop receiving emails from it, send an email to [email protected].
> >