From: Mateusz Nosek <[email protected]>
Previously 0 was assigned to 'sc->skipped_deactivate'. It could happen only
if 'sc->skipped_deactivate' was 0 so the assignment is unnecessary and can
be removed.
Signed-off-by: Mateusz Nosek <[email protected]>
---
mm/vmscan.c | 1 -
1 file changed, 1 deletion(-)
diff --git a/mm/vmscan.c b/mm/vmscan.c
index dca623db51c8..453ff2abcb58 100644
--- a/mm/vmscan.c
+++ b/mm/vmscan.c
@@ -3093,7 +3093,6 @@ static unsigned long do_try_to_free_pages(struct zonelist *zonelist,
if (sc->memcg_low_skipped) {
sc->priority = initial_priority;
sc->force_deactivate = 0;
- sc->skipped_deactivate = 0;
sc->memcg_low_reclaim = 1;
sc->memcg_low_skipped = 0;
goto retry;
--
2.17.1
It is usually preferable to Cc author of the code (added Johannes)
On Thu 19-03-20 17:59:38, [email protected] wrote:
> From: Mateusz Nosek <[email protected]>
>
> Previously 0 was assigned to 'sc->skipped_deactivate'. It could happen only
> if 'sc->skipped_deactivate' was 0 so the assignment is unnecessary and can
> be removed.
The above wording was a bit hard to understdand for me. I would go with
"
sc->memcg_low_skipped resets skipped_deactivate to 0 but this is not
needed as this code path is never reachable with skipped_deactivate != 0
due to previous sc->skipped_deactivate branch.
"
> Signed-off-by: Mateusz Nosek <[email protected]>
The patch is correct. I am not sure it results in a better code though.
I will defer to Johannes here. I suspect he simply wanted to express
that skipped_deactivate should be always reset when retrying the direct
reclaim. After this patch this could be lost in future changes so the
code would be more subtle. But I am only guessing here.
> ---
> mm/vmscan.c | 1 -
> 1 file changed, 1 deletion(-)
>
> diff --git a/mm/vmscan.c b/mm/vmscan.c
> index dca623db51c8..453ff2abcb58 100644
> --- a/mm/vmscan.c
> +++ b/mm/vmscan.c
> @@ -3093,7 +3093,6 @@ static unsigned long do_try_to_free_pages(struct zonelist *zonelist,
> if (sc->memcg_low_skipped) {
> sc->priority = initial_priority;
> sc->force_deactivate = 0;
> - sc->skipped_deactivate = 0;
> sc->memcg_low_reclaim = 1;
> sc->memcg_low_skipped = 0;
> goto retry;
> --
> 2.17.1
--
Michal Hocko
SUSE Labs
On Thu, Mar 19, 2020 at 06:13:34PM +0100, Michal Hocko wrote:
> It is usually preferable to Cc author of the code (added Johannes)
>
> On Thu 19-03-20 17:59:38, [email protected] wrote:
> > From: Mateusz Nosek <[email protected]>
> >
> > Previously 0 was assigned to 'sc->skipped_deactivate'. It could happen only
> > if 'sc->skipped_deactivate' was 0 so the assignment is unnecessary and can
> > be removed.
>
> The above wording was a bit hard to understdand for me. I would go with
> "
> sc->memcg_low_skipped resets skipped_deactivate to 0 but this is not
> needed as this code path is never reachable with skipped_deactivate != 0
> due to previous sc->skipped_deactivate branch.
> "
Yeah that sounds good.
> > Signed-off-by: Mateusz Nosek <[email protected]>
>
> The patch is correct. I am not sure it results in a better code though.
> I will defer to Johannes here. I suspect he simply wanted to express
> that skipped_deactivate should be always reset when retrying the direct
> reclaim. After this patch this could be lost in future changes so the
> code would be more subtle. But I am only guessing here.
It's a valid concern, but I think in this case specifically we're very
unlikely to change the ordering here - violate memory.low before going
after active pages of unprotected cgroups.
I indeed just kept it stupid: reset everything, then retry. But it
appears that the unnecessary assignment trips people up and wastes
their time, so I'm in favor of removing it.
Acked-by: Johannes Weiner <[email protected]>