mm: vmpressure: fix sending wrong events on underflow
Subsystems (4)
PATCH v2 1/2 mm: vmpressure: fix sending wrong events on underflow
At the end of a window period, if the reclaimed pages is greater than scanned, an unsigned underflow can result in a huge pressure value and thus a critical event. Reclaimed pages is found to go higher than scanned because of the addition of reclaimed slab pages to reclaimed in shrink_node without a corresponding increment to scanned pages. Minchan Kim mentioned that this can also happen in the case of a THP page where the scanned is 1 and reclaimed could be 512.
---
v2: Adding a comment and reordering the patches
as per Michal's suggestion
mm/vmpressure.c | 10 +++++++++- 1 file changed, 9 insertions(+), 1 deletion(-)
mm/vmpressure.c +9 -1
@@ -112,9 +112,16 @@ static enum vmpressure_levels vmpressure_calc_level(unsigned long scanned, unsigned long reclaimed) {unsigned long scale = scanned + reclaimed; -unsigned long pressure; +unsigned long pressure =0 ;/* +* reclaimed can be greater than scanned in cases +* like THP, where the scanned is 1 and reclaimed +* could be 512 +*/ +if (reclaimed >= scanned) + goto out; +/* * We calculate the ratio (in percents) of how many pages were * scanned vs. reclaimed in a given time frame (window). Note that * time is in VM reclaimer's "ticks", i.e. number of pages
@@ -124,6 +131,7 @@ static enum vmpressure_levels vmpressure_calc_level(unsigned long scanned, pressure = scale - (reclaimed * scale / scanned); pressure = pressure *100 / scale; +out:pr_debug ("%s: %3lu (s: %lu r: %lu)\n" , __func__, pressure, scanned, reclaimed);
Signature · 8 lines
QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum, hosted by The Linux Foundation -- 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>
PATCH v5 2/2 mm: vmscan: do not pass reclaimed slab to vmpressure
During global reclaim, the nr_reclaimed passed to vmpressure includes the pages reclaimed from slab. But the corresponding scanned slab pages is not passed. There is an impact to the vmpressure values because of this. While moving from kernel version 3.18 to 4.4, a difference is seen in the vmpressure values for the same workload resulting in a different behaviour of the vmpressure consumer. One such case is of a vmpressure based lowmemorykiller. It is observed that the vmpressure events are received late and less in number resulting in tasks not being killed at the right time. The following numbers show the impact on reclaim activity due to the change in behaviour of lowmemorykiller on a 4GB device. The test launches a number of apps in sequence and repeats it multiple times.
v4.4 v3.18 pgpgin 163016456 145617236 pgpgout 4366220 4188004 workingset_refault 29857868 26781854 workingset_activate 6293946 5634625 pswpin 1327601 1133912 pswpout 3593842 3229602 pgalloc_dma 99520618 94402970 pgalloc_normal 104046854 98124798 pgfree 203772640 192600737 pgmajfault 2126962 1851836 pgsteal_kswapd_dma 19732899 18039462 pgsteal_kswapd_normal 19945336 17977706 pgsteal_direct_dma 206757 131376 pgsteal_direct_normal 236783 138247 pageoutrun 116622 108370 allocstall 7220 4684 compact_stall 931 856
This is a regression introduced by commit 6b4f7799c6a5 ("mm: vmscan:
invoke slab shrinkers from shrink_zone()").
So do not consider reclaimed slab pages for vmpressure calculation. The reclaimed pages from slab can be excluded because the freeing of a page by slab shrinking depends on each slab's object population, making the cost model (i.e. scan:free) different from that of LRU. Also, not every shrinker accounts the pages it reclaims. But ideally the pages reclaimed from slab should be passed to vmpressure, otherwise higher vmpressure levels can be triggered even when there is a reclaim progress. But accounting only the reclaimed slab pages without the scanned, and adding something which does not fit into the cost model just adds noise to the vmpressure values.
---
v5: Modifying the changelog and reordering the patches
as per Michal's suggestion
mm/vmscan.c | 17 ++++++++++++----- 1 file changed, 12 insertions(+), 5 deletions(-)
mm/vmscan.c +12 -5
@@ -2594,16 +2594,23 @@ static bool shrink_node(pg_data_t *pgdat, struct scan_control *sc) sc->nr_scanned - nr_scanned, node_lru_pages); +/* +* Record the subtree's reclaim efficiency. The reclaimed +* pages from slab is excluded here because the corresponding +* scanned pages is not accounted. Moreover, freeing a page +* by slab shrinking depends on each slab's object population, +* making the cost model (i.e. scan:free) different from that +* of LRU. +*/ +vmpressure (sc->gfp_mask, sc->target_mem_cgroup, true, + sc->nr_scanned - nr_scanned, + sc->nr_reclaimed - nr_reclaimed); +if (reclaim_state) { sc->nr_reclaimed += reclaim_state->reclaimed_slab; reclaim_state->reclaimed_slab =0 ; } -/* Record the subtree's reclaim efficiency */ -vmpressure (sc->gfp_mask, sc->target_mem_cgroup, true, - sc->nr_scanned - nr_scanned, - sc->nr_reclaimed - nr_reclaimed); -if (sc->nr_reclaimed - nr_reclaimed) reclaimable = true;
Signature · 8 lines
QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum, hosted by The Linux Foundation -- 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>
Michal Hocko
reviewerRe: [PATCH 2/2 v5] mm: vmscan: do not pass reclaimed slab to vmpressure
Quoted 11 lines · Vinayak Menon
During global reclaim, the
nr_reclaimedpassed to vmpressure includes the pages reclaimed from slab. But the corresponding scanned slab pages is not passed. There is an impact to the vmpressure values because of this. While moving from kernel version 3.18 to 4.4, a difference is seen in the vmpressure values for the same workload resulting in a different behaviour of the vmpressure consumer. One such case is of a vmpressure based lowmemorykiller. It is observed that the vmpressure events are received late and less in number resulting in tasks not being killed at the right time. The following numbers show the impact on reclaim activity due to the change in behaviour of lowmemorykiller on a 4GB device. The test launches a number of apps in sequence and repeats it multiple times.
this is really vague description of your workload and doesn't really explain why getting critical events later is a bad thing.
Quoted 11 lines
v4.4 v3.18
pgpgin 163016456 145617236
pgpgout 4366220 4188004
workingset_refault 29857868 26781854
workingset_activate 6293946 5634625
pswpin 1327601 1133912
pswpout 3593842 3229602
pgalloc_dma 99520618 94402970
pgalloc_normal 104046854 98124798
pgfree 203772640 192600737Quoted 8 lines
pgmajfault 2126962 1851836 pgsteal_kswapd_dma 19732899 18039462 pgsteal_kswapd_normal 19945336 17977706 pgsteal_direct_dma 206757 131376 pgsteal_direct_normal 236783 138247 pageoutrun 116622 108370 allocstall 7220 4684 compact_stall 931 856
this is missing any vmpressure events data and so it is not very useful on its own
Quoted 14 lines
This is a regression introduced by commit
6b4f7799c6a5("mm: vmscan:
invoke slab shrinkers fromshrink_zone()").So do not consider reclaimed slab pages for vmpressure calculation. The reclaimed pages from slab can be excluded because the freeing of a page by slab shrinking depends on each slab's object population, making the cost model (i.e. scan:free) different from that of LRU. Also, not every shrinker accounts the pages it reclaims. But ideally the pages reclaimed from slab should be passed to vmpressure, otherwise higher vmpressure levels can be triggered even when there is a reclaim progress. But accounting only the reclaimed slab pages without the scanned, and adding something which does not fit into the cost model just adds noise to the vmpressure values.
Quoted 11 lines
Fixes 6b4f7799c6a5 ("mm: vmscan: invoke slab shrinkers from shrink_zone()")Acked-by Minchan Kim <[email protected]>Cc Johannes Weiner <[email protected]>Cc Mel Gorman <[email protected]>Cc Vlastimil Babka <[email protected]>Cc Michal Hocko <[email protected]>Cc Rik van Riel <[email protected]>Cc Vladimir Davydov <[email protected]>Cc Anton Vorontsov <[email protected]>Cc Shiraz Hashim <[email protected]>Signed-off-by Vinayak Menon <[email protected]>
I have already said I will _not_ NAK the patch but we need a much better description and justification why the older behavior was better to consider this a regression before this can be merged. It is hard to expect that the underlying implementation of the vmpressure will stay carved in stone and there might be changes in this area in the future. I want to hear why we believe that the tested workload is sufficiently universal and we won't see another report in few months because somebody else will see higher vmpressure levels even though we make reclaim progress. I have asked those questions already but it seems those were ignored.
Signature · 8 lines
Michal Hocko SUSE Labs -- 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>
Re: [PATCH 2/2 v5] mm: vmscan: do not pass reclaimed slab to vmpressure
Quoted 13 lines · Michal Hocko
Quoted 11 lines · Vinayak Menon
During global reclaim, the
nr_reclaimedpassed to vmpressure includes the pages reclaimed from slab. But the corresponding scanned slab pages is not passed. There is an impact to the vmpressure values because of this. While moving from kernel version 3.18 to 4.4, a difference is seen in the vmpressure values for the same workload resulting in a different behaviour of the vmpressure consumer. One such case is of a vmpressure based lowmemorykiller. It is observed that the vmpressure events are received late and less in number resulting in tasks not being killed at the right time. The following numbers show the impact on reclaim activity due to the change in behaviour of lowmemorykiller on a 4GB device. The test launches a number of apps in sequence and repeats it multiple times.
Quoted 2 lines
this is really vague description of your workload and doesn't really explain why getting critical events later is a bad thing.
Ok. I will add that.
Quoted 21 lines
Quoted 12 lines
v4.4 v3.18 pgpgin 163016456 145617236 pgpgout 4366220 4188004 workingset_refault 29857868 26781854 workingset_activate 6293946 5634625 pswpin 1327601 1133912 pswpout 3593842 3229602 pgalloc_dma 99520618 94402970 pgalloc_normal 104046854 98124798 pgfree 203772640 192600737Quoted 8 lines
pgmajfault 2126962 1851836 pgsteal_kswapd_dma 19732899 18039462 pgsteal_kswapd_normal 19945336 17977706 pgsteal_direct_dma 206757 131376 pgsteal_direct_normal 236783 138247 pageoutrun 116622 108370 allocstall 7220 4684 compact_stall 931 856
Quoted 2 lines
this is missing any vmpressure events data and so it is not very useful on its own
Done.
Quoted 27 lines
Quoted 15 lines
This is a regression introduced by commit
6b4f7799c6a5("mm: vmscan:
invoke slab shrinkers fromshrink_zone()").So do not consider reclaimed slab pages for vmpressure calculation. The reclaimed pages from slab can be excluded because the freeing of a page by slab shrinking depends on each slab's object population, making the cost model (i.e. scan:free) different from that of LRU. Also, not every shrinker accounts the pages it reclaims. But ideally the pages reclaimed from slab should be passed to vmpressure, otherwise higher vmpressure levels can be triggered even when there is a reclaim progress. But accounting only the reclaimed slab pages without the scanned, and adding something which does not fit into the cost model just adds noise to the vmpressure values.
Quoted 11 lines
Fixes 6b4f7799c6a5 ("mm: vmscan: invoke slab shrinkers from shrink_zone()")Acked-by Minchan Kim <[email protected]>Cc Johannes Weiner <[email protected]>Cc Mel Gorman <[email protected]>Cc Vlastimil Babka <[email protected]>Cc Michal Hocko <[email protected]>Cc Rik van Riel <[email protected]>Cc Vladimir Davydov <[email protected]>Cc Anton Vorontsov <[email protected]>Cc Shiraz Hashim <[email protected]>Signed-off-by Vinayak Menon <[email protected]>
Quoted 10 lines
I have already said I will _not_ NAK the patch but we need a much better description and justification why the older behavior was better to consider this a regression before this can be merged. It is hard to expect that the underlying implementation of the vmpressure will stay carved in stone and there might be changes in this area in the future. I want to hear why we believe that the tested workload is sufficiently universal and we won't see another report in few months because somebody else will see higher vmpressure levels even though we make reclaim progress. I have asked those questions already but it seems those were ignored.
The tested workload is not universal. The lowmemorykiller example was used just to mention the effect of vmpressure change on one of the workloads. I can drop the
reclaim stats and just keep the stats of change observed in vmpressure critical events.
I am not sure whether we would see another issue reported with this patch. We may because
someone would have written a code that works with this new vmpressure values. I am not
sure whether that matters because the core issue is whether the kernel is reporting the
right values. This could be termed as a regression because,
1) Accounting only reclaimed pages to a model which works on scanned and reclaimed
seems like a wrong thing. It is just adding noise to it. There could be issues with vmpressure
implementation, but it at least gives an estimate on what the pressure on LRU is. There are
many other shrinkers like zsmalloc which does not report reclaimed pages, and when add those
also in a similar fashion without considering the cost part, vmpressure values would always
remain low. So util we have a way to give correct information to vmpressure about non-LRU
reclaimers, I feel its better to keep it in its original form.
2) As Minchan mentioned, the cost model is different and thus adding slab reclaimed would
not be the right thing to do at this point.
But if you feel we don't have to fix this now and that it is better to fix the core problems with
vmpressure first, that's ok. Anyway I will sent a patch with a new changelog with vmpressure
event details. Thanks Michal for your comments.
Vinayak
--
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>
Michal Hocko
reviewerRe: [PATCH 2/2 v5] mm: vmscan: do not pass reclaimed slab to vmpressure
Quoted 1 line · vinayak menon
On Thu, Feb 9, 2017 at 5:50 PM, Michal Hocko <[email protected]> wrote:
[...]
Quoted 11 lines
Quoted 10 lines
I have already said I will _not_ NAK the patch but we need a much better description and justification why the older behavior was better to consider this a regression before this can be merged. It is hard to expect that the underlying implementation of the vmpressure will stay carved in stone and there might be changes in this area in the future. I want to hear why we believe that the tested workload is sufficiently universal and we won't see another report in few months because somebody else will see higher vmpressure levels even though we make reclaim progress. I have asked those questions already but it seems those were ignored.
Quoted 2 lines
The tested workload is not universal. The lowmemorykiller example was used just to mention the effect of vmpressure change on one of the workloads.
My point is whether this workload even matters. AFAIU the test benefits from killing as quickly as possible, right? So it directly benefits from seeing critical events as soon as possible even when the reclaim makes progress.
Quoted 6 lines
I can drop the reclaim stats and just keep the stats of change observed in vmpressure critical events. I am not sure whether we would see another issue reported with this patch. We may because someone would have written a code that works with this new vmpressure values. I am not sure whether that matters because the core issue is whether the kernel is reporting the right values.
Right. THe right values is a bit fuzzy, though.
Quoted 12 lines
This could be
termed as a regression because,1) Accounting only reclaimed pages to a model which works on scanned and reclaimed seems like a wrong thing. It is just adding noise to it. There could be issues with vmpressure implementation, but it at least gives an estimate on what the pressure on LRU is. There are many other shrinkers like zsmalloc which does not report reclaimed pages, and when add those also in a similar fashion without considering the cost part, vmpressure values would always remain low. So util we have a way to give correct information to vmpressure about non-LRU reclaimers, I feel its better to keep it in its original form.
Yeah, I understand that the current cost model is far from ideal and it needs fixing. My main question would be whether the model would be much better if we exclude pages freed from the slab shrinkers. I can only say it would be more pesimistic that way. Is this a good thing? If yes, why?
Quoted 5 lines
2) As Minchan mentioned, the cost model is different and thus adding slab reclaimed would not be the right thing to do at this point.
But if you feel we don't have to fix this now and that it is better to fix the core problems with vmpressure first, that's ok.
Yes, I believe we should reconsider how we calculate the pressure levels. This seems a larger project but definitely something we need. I do not have a good ideas how to do this properly
Signature · 8 lines
Michal Hocko SUSE Labs -- 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>
Quoted 10 lines · Vinayak Menon
At the end of a window period, if the reclaimed pages is greater than scanned, an unsigned underflow can result in a huge pressure value and thus a critical event. Reclaimed pages is found to go higher than scanned because of the addition of reclaimed slab pages to reclaimed in
shrink_nodewithout a corresponding increment to scanned pages. Minchan Kim mentioned that this can also happen in the case of a THP page where the scanned is 1 and reclaimed could be 512.
Quoted 2 lines
Acked-by Minchan Kim <[email protected]>Signed-off-by Vinayak Menon <[email protected]>
I would prefer the fixup in vmpressure() as already mentioned but this should work as well.
Quoted 41 lines
---
v2: Adding a comment and reordering the patches
as per Michal's suggestionmm/vmpressure.c | 10 +++++++++- 1 file changed, 9 insertions(+), 1 deletion(-)mm/vmpressure.c +9
-1@@ -112,9 +112,16 @@ static enum vmpressure_levels vmpressure_calc_level(unsigned long scanned, unsigned long reclaimed) {unsigned long scale = scanned + reclaimed; -unsigned long pressure; +unsigned long pressure =0 ;/* +* reclaimed can be greater than scanned in cases +* like THP, where the scanned is 1 and reclaimed +* could be 512 +*/ +if (reclaimed >= scanned) + goto out; +/* * We calculate the ratio (in percents) of how many pages were * scanned vs. reclaimed in a given time frame (window). Note that * time is in VM reclaimer's "ticks", i.e. number of pages @@ -124,6 +131,7 @@ static enum vmpressure_levels vmpressure_calc_level(unsigned long scanned, pressure = scale - (reclaimed * scale / scanned); pressure = pressure *100 / scale; +out:pr_debug ("%s: %3lu (s: %lu r: %lu)\n" , __func__, pressure, scanned, reclaimed);Signature · 2 lines
QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum, hosted by The Linux Foundation
Signature · 8 lines
Michal Hocko SUSE Labs -- 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>
Michal Hocko
reviewerQuoted 11 lines · Michal Hocko
Quoted 10 lines · Vinayak Menon
At the end of a window period, if the reclaimed pages is greater than scanned, an unsigned underflow can result in a huge pressure value and thus a critical event. Reclaimed pages is found to go higher than scanned because of the addition of reclaimed slab pages to reclaimed in
shrink_nodewithout a corresponding increment to scanned pages. Minchan Kim mentioned that this can also happen in the case of a THP page where the scanned is 1 and reclaimed could be 512.
Quoted 7 lines
Quoted 2 lines
Acked-by Minchan Kim <[email protected]>Signed-off-by Vinayak Menon <[email protected]>Acked-by Michal Hocko <[email protected]>I would prefer the fixup in
vmpressure()as already mentioned but this should work as well.
Btw. I guess this should be good to mark for stable. Reclaiming THP is not all that rare (even though we try to avoid anon reclaim as much as possible) and hitting critical events can lead to disruptive actions to early.
Signature · 8 lines
Michal Hocko SUSE Labs -- 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>