lkml.org 
[lkml]   [2021]   [Oct]   [14]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
Date
From
SubjectRe: [PATCH 2/8] mm/vmscan: Throttle reclaim and compaction when too may pages are isolated
On Thu, Oct 14, 2021 at 10:06:25AM +0200, Vlastimil Babka wrote:
> On 10/8/21 15:53, Mel Gorman wrote:
> > Page reclaim throttles on congestion if too many parallel reclaim instances
> > have isolated too many pages. This makes no sense, excessive parallelisation
> > has nothing to do with writeback or congestion.
> >
> > This patch creates an additional workqueue to sleep on when too many
> > pages are isolated. The throttled tasks are woken when the number
> > of isolated pages is reduced or a timeout occurs. There may be
> > some false positive wakeups for GFP_NOIO/GFP_NOFS callers but
> > the tasks will throttle again if necessary.
> >
> > [shy828301@gmail.com: Wake up from compaction context]
> > Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
>
> ...
>
> > diff --git a/mm/internal.h b/mm/internal.h
> > index 90764d646e02..06d0c376efcd 100644
> > --- a/mm/internal.h
> > +++ b/mm/internal.h
> > @@ -45,6 +45,15 @@ static inline void acct_reclaim_writeback(struct page *page)
> > __acct_reclaim_writeback(pgdat, page, nr_throttled);
> > }
> >
> > +static inline void wake_throttle_isolated(pg_data_t *pgdat)
> > +{
> > + wait_queue_head_t *wqh;
> > +
> > + wqh = &pgdat->reclaim_wait[VMSCAN_THROTTLE_ISOLATED];
> > + if (waitqueue_active(wqh))
> > + wake_up_all(wqh);
>
> Again, would it be better to wake up just one task to prevent possible
> thundering herd? We can assume that that task will call too_many_isolated()
> eventually to wake up the next one?

Same problem as the writeback throttling, there is no prioritsation of
light vs heavy allocators.

> Although it seems strange that
> too_many_isolated() is the place where we detect the situation for wake up.
> Simpler than to hook into NR_ISOLATED decrementing I guess.
>

Simplier but more costly. Every decrement would have to check
too_many_isolated(). I think the cost of that is too high given that the
VMSCAN_THROTTLE_ISOLATED is relatively hard to trigger and the minority
of throttling events.

> > +}
> > +
> > vm_fault_t do_swap_page(struct vm_fault *vmf);
> >
> > void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *start_vma,
> ...
> > --- a/mm/vmscan.c
> > +++ b/mm/vmscan.c
> > @@ -1006,11 +1006,10 @@ static void handle_write_error(struct address_space *mapping,
> > unlock_page(page);
> > }
> >
> > -static void
> > -reclaim_throttle(pg_data_t *pgdat, enum vmscan_throttle_state reason,
> > +void reclaim_throttle(pg_data_t *pgdat, enum vmscan_throttle_state reason,
> > long timeout)
> > {
> > - wait_queue_head_t *wqh = &pgdat->reclaim_wait;
> > + wait_queue_head_t *wqh = &pgdat->reclaim_wait[reason];
>
> It seems weird that later in this function we increase nr_reclaim_throttled
> without distinguishing the reason, so effectively throttling for isolated
> pages will trigger acct_reclaim_writeback() doing the NR_THROTTLED_WRITTEN
> counting, although it's not related at all? Maybe either have separate
> nr_reclaim_throttled counters per vmscan_throttle_state (if counter of
> isolated is useful, I haven't seen the rest of series yet), or count only
> VMSCAN_THROTTLE_WRITEBACK tasks?
>

Very good point, it would be more appropriate to only count the
writeback reason.

Diff on top is below. It'll cause minor conflicts later in the series.

diff --git a/include/linux/mmzone.h b/include/linux/mmzone.h
index ca65d6a64bdd..58a25d42c31c 100644
--- a/include/linux/mmzone.h
+++ b/include/linux/mmzone.h
@@ -849,7 +849,7 @@ typedef struct pglist_data {
wait_queue_head_t kswapd_wait;
wait_queue_head_t pfmemalloc_wait;
wait_queue_head_t reclaim_wait[NR_VMSCAN_THROTTLE];
- atomic_t nr_reclaim_throttled; /* nr of throtted tasks */
+ atomic_t nr_writeback_throttled;/* nr of writeback-throttled tasks */
unsigned long nr_reclaim_start; /* nr pages written while throttled
* when throttling started. */
struct task_struct *kswapd; /* Protected by
diff --git a/mm/internal.h b/mm/internal.h
index 06d0c376efcd..3461a1055975 100644
--- a/mm/internal.h
+++ b/mm/internal.h
@@ -39,7 +39,7 @@ void __acct_reclaim_writeback(pg_data_t *pgdat, struct page *page,
static inline void acct_reclaim_writeback(struct page *page)
{
pg_data_t *pgdat = page_pgdat(page);
- int nr_throttled = atomic_read(&pgdat->nr_reclaim_throttled);
+ int nr_throttled = atomic_read(&pgdat->nr_writeback_throttled);

if (nr_throttled)
__acct_reclaim_writeback(pgdat, page, nr_throttled);
diff --git a/mm/vmscan.c b/mm/vmscan.c
index 6e198bbbd86a..29434d4fc1c7 100644
--- a/mm/vmscan.c
+++ b/mm/vmscan.c
@@ -1011,6 +1011,7 @@ void reclaim_throttle(pg_data_t *pgdat, enum vmscan_throttle_state reason,
{
wait_queue_head_t *wqh = &pgdat->reclaim_wait[reason];
long ret;
+ bool acct_writeback = (reason == VMSCAN_THROTTLE_WRITEBACK);
DEFINE_WAIT(wait);

/*
@@ -1022,7 +1023,8 @@ void reclaim_throttle(pg_data_t *pgdat, enum vmscan_throttle_state reason,
current->flags & (PF_IO_WORKER|PF_KTHREAD))
return;

- if (atomic_inc_return(&pgdat->nr_reclaim_throttled) == 1) {
+ if (acct_writeback &&
+ atomic_inc_return(&pgdat->nr_writeback_throttled) == 1) {
WRITE_ONCE(pgdat->nr_reclaim_start,
node_page_state(pgdat, NR_THROTTLED_WRITTEN));
}
@@ -1030,7 +1032,9 @@ void reclaim_throttle(pg_data_t *pgdat, enum vmscan_throttle_state reason,
prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
ret = schedule_timeout(timeout);
finish_wait(wqh, &wait);
- atomic_dec(&pgdat->nr_reclaim_throttled);
+
+ if (acct_writeback)
+ atomic_dec(&pgdat->nr_writeback_throttled);

trace_mm_vmscan_throttled(pgdat->node_id, jiffies_to_usecs(timeout),
jiffies_to_usecs(timeout - ret),
@@ -4349,7 +4353,7 @@ static int kswapd(void *p)

WRITE_ONCE(pgdat->kswapd_order, 0);
WRITE_ONCE(pgdat->kswapd_highest_zoneidx, MAX_NR_ZONES);
- atomic_set(&pgdat->nr_reclaim_throttled, 0);
+ atomic_set(&pgdat->nr_writeback_throttled, 0);
for ( ; ; ) {
bool ret;

\
 
 \ /
  Last update: 2021-10-14 13:57    [W:0.225 / U:0.628 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site