summaryrefslogtreecommitdiff
path: root/mm/damon
diff options
context:
space:
mode:
authorSeongJae Park <sj@kernel.org>2023-09-16 02:09:37 +0000
committerAndrew Morton <akpm@linux-foundation.org>2023-10-04 10:32:31 -0700
commitaffa87c708185cab194099ee51b946ef0297f063 (patch)
treead27de2cbe6f41b9769942e8821f39098f044e7a /mm/damon
parentd5b43e9683ec5c78524f9ae93c1824edcbc1f08d (diff)
mm/damon/core: make DAMOS uses nr_accesses_bp instead of nr_accesses
Patch series "mm/damon: implement DAMOS apply intervals". DAMON-based operation schemes are applied for every aggregation interval. That is mainly because schemes are using nr_accesses, which be complete to be used for every aggregation interval. This makes some DAMOS use cases be tricky. Quota setting under long aggregation interval is one such example. Suppose the aggregation interval is ten seconds, and there is a scheme having CPU quota 100ms per 1s. The scheme will actually uses 100ms per ten seconds, since it cannobe be applied before next aggregation interval. The feature is working as intended, but the results might not that intuitive for some users. This could be fixed by updating the quota to 1s per 10s. But, in the case, the CPU usage of DAMOS could look like spikes, and actually make a bad effect to other CPU-sensitive workloads. Also, with such huge aggregation interval, users may want schemes to be applied more frequently. DAMON provides nr_accesses_bp, which is updated for each sampling interval in a way that reasonable to be used. By using that instead of nr_accesses, DAMOS can have its own time interval and mitigate abovely mentioned issues. This patchset makes DAMOS schemes to use nr_accesses_bp instead of nr_accesses, and have their own timing intervals. Also update DAMOS tried regions sysfs files and DAMOS before_apply tracepoint to use the new data as their source. Note that the interval is zero by default, and it is interpreted to use the aggregation interval instead. This avoids making user-visible behavioral changes. Patches Seuqeunce ----------------- The first patch (patch 1/9) makes DAMOS uses nr_accesses_bp instead of nr_accesses, and following two patches (patches 2/9 and 3/9) updates DAMON sysfs interface for DAMOS tried regions and the DAMOS before_apply tracespoint to use nr_accesses_bp instead of nr_accesses, respectively. The following two patches (patches 4/9 and 5/9) implements the scheme-specific apply interval for DAMON kernel API users and update the design document for the new feature. Finally, the following four patches (patches 6/9, 7/9, 8/9 and 9/9) add support of the feature in DAMON sysfs interface, add a simple selftest test case, and document the new file on the usage and the ABI documents, repsectively. This patch (of 9): DAMON provides nr_accesses_bp, which becomes same to nr_accesses * 10000 for every aggregation interval, but updated every sampling interval with a reasonable accuracy. Since DAMON-based operation schemes are applied in every aggregation interval using nr_accesses, using nr_accesses_bp instead will make no difference to users. Meanwhile, it allows DAMOS to apply the schemes in a time interval that less than the aggregation interval. It could be useful and more flexible for some cases. Do it. Link: https://lkml.kernel.org/r/20230916020945.47296-1-sj@kernel.org Link: https://lkml.kernel.org/r/20230916020945.47296-2-sj@kernel.org Signed-off-by: SeongJae Park <sj@kernel.org> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Shuah Khan <shuah@kernel.org> Cc: Steven Rostedt (Google) <rostedt@goodmis.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Diffstat (limited to 'mm/damon')
-rw-r--r--mm/damon/core.c5
1 files changed, 3 insertions, 2 deletions
diff --git a/mm/damon/core.c b/mm/damon/core.c
index b15cf47d2d29..79fef5145a4b 100644
--- a/mm/damon/core.c
+++ b/mm/damon/core.c
@@ -770,12 +770,13 @@ static void damon_split_region_at(struct damon_target *t,
static bool __damos_valid_target(struct damon_region *r, struct damos *s)
{
unsigned long sz;
+ unsigned int nr_accesses = r->nr_accesses_bp / 10000;
sz = damon_sz_region(r);
return s->pattern.min_sz_region <= sz &&
sz <= s->pattern.max_sz_region &&
- s->pattern.min_nr_accesses <= r->nr_accesses &&
- r->nr_accesses <= s->pattern.max_nr_accesses &&
+ s->pattern.min_nr_accesses <= nr_accesses &&
+ nr_accesses <= s->pattern.max_nr_accesses &&
s->pattern.min_age_region <= r->age &&
r->age <= s->pattern.max_age_region;
}