cachestat: a new syscall for page cache state of files
Subsystems (22)
Nhat Pham
maintainerv6 cover cachestat: a new syscall for page cache state of files
Changelog:
v6:
* Add a missing fdput() (suggested by Brian Foster) (patch 2)
* Replace cstat_size with cstat_version (suggested by Brian Foster)
(patch 2)
* Add conditional resched to the xas walk. (suggested by Hillf Danton)
(patch 2)v5:
* Separate first patch into its own series. (suggested by Andrew Morton)
* Expose filemap_cachestat() to non-syscall usage (patch 2) (suggested by Brian Foster).
* Fix some build errors from last version.
(patch 2)
* Explain eviction and recent eviction in the draft man page and documentation (suggested by Andrew Morton).
(patch 2)
v4:
* Refactor cachestat and move it to mm/filemap.c (patch 3) (suggested by Brian Foster)
* Remove redundant checks (!folio, access_ok)
(patch 3) (suggested by Matthew Wilcox and Al Viro)
* Fix a bug in handling multipages folio.
(patch 3) (suggested by Matthew Wilcox)
* Add a selftest for shmem files, which can be used to test huge pages (patch 4) (suggested by Johannes Weiner)
v3:
* Fix some minor formatting issues and build errors.
* Add the new syscall entry to missing architecture syscall tables.
(patch 3).
* Add flags argument for the syscall. (patch 3).
* Clean up the recency refactoring (patch 2) (suggested by Yu Zhao)
* Add the new Kconfig (CONFIG_CACHESTAT) to disable the syscall. (patch 3) (suggested by Josh Triplett)
v2:
* len == 0 means query to EOF. len < 0 is invalid.
(patch 3) (suggested by Brian Foster)
* Make cachestat extensible by adding the cstat_size argument in the syscall (patch 3)
There is currently no good way to query the page cache state of large file sets and directory trees. There is mincore(), but it scales poorly:
the kernel writes out a lot of bitmap data that userspace has to aggregate, when the user really doesn not care about per-page information in that case. The user also needs to mmap and unmap each file as it goes along, which can be quite slow as well.
This series of patches introduces a new system call, cachestat, that summarizes the page cache statistics (number of cached pages, dirty pages, pages marked for writeback, evicted pages etc.) of a file, in a specified range of bytes. It also include a selftest suite that tests some typical usage
This interface is inspired by past discussion and concerns with fincore, which has a similar design (and as a result, issues) as mincore. Relevant links:
https://lkml.indiana.edu/hypermail/linux/kernel/1302.1/04207.html
https://lkml.indiana.edu/hypermail/linux/kernel/1302.1/04209.html
For comparison with mincore, I ran both syscalls on a 2TB sparse file:
Using mincore:
real 0m37.510s user 0m2.934s sys 0m34.558s
Using cachestat:
real 0m0.009s user 0m0.000s sys 0m0.009s
This series should be applied on top of:
workingset: fix confusion around eviction vs refault container
https://lkml.org/lkml/2023/1/4/1066
This series consist of 3 patches:
Nhat Pham (3):
workingset: refactor LRU refault to expose refault recency check
cachestat: implement cachestat syscall
selftests: Add selftests for cachestat
MAINTAINERS | 7 + arch/alpha/kernel/syscalls/syscall.tbl | 1 + arch/arm/tools/syscall.tbl | 1 + arch/ia64/kernel/syscalls/syscall.tbl | 1 + arch/m68k/kernel/syscalls/syscall.tbl | 1 + arch/microblaze/kernel/syscalls/syscall.tbl | 1 + arch/parisc/kernel/syscalls/syscall.tbl | 1 + arch/powerpc/kernel/syscalls/syscall.tbl | 1 + arch/s390/kernel/syscalls/syscall.tbl | 1 + arch/sh/kernel/syscalls/syscall.tbl | 1 + arch/sparc/kernel/syscalls/syscall.tbl | 1 + arch/x86/entry/syscalls/syscall_32.tbl | 1 + arch/x86/entry/syscalls/syscall_64.tbl | 1 + arch/xtensa/kernel/syscalls/syscall.tbl | 1 + include/linux/fs.h | 3 + include/linux/swap.h | 1 + include/linux/syscalls.h | 4 + include/uapi/asm-generic/unistd.h | 5 +- include/uapi/linux/mman.h | 9 + init/Kconfig | 10 + kernel/sys_ni.c | 1 + mm/filemap.c | 154 +++++++++++ mm/workingset.c | 129 ++++++--- tools/testing/selftests/Makefile | 1 + tools/testing/selftests/cachestat/.gitignore | 2 + tools/testing/selftests/cachestat/Makefile | 8 + .../selftests/cachestat/test_cachestat.c | 260 ++++++++++++++++++ 27 files changed, 568 insertions(+), 39 deletions(-) create mode 100644 tools/testing/selftests/cachestat/.gitignore create mode 100644 tools/testing/selftests/cachestat/Makefile create mode 100644 tools/testing/selftests/cachestat/test_cachestat.c
base-commit: 1440f576022887004f719883acb094e7e0dd4944
prerequisite-patch-id: 171a43d333e1b267ce14188a5beaea2f313787fb
Signature · 1 line
2.30.2
Nhat Pham
maintainerPATCH v6 1/3 workingset: refactor LRU refault to expose refault recency check
In preparation for computing recently evicted pages in cachestat, refactor workingset_refault and lru_gen_refault to expose a helper function that would test if an evicted page is recently evicted.
--- include/linux/swap.h | 1 + mm/workingset.c | 129 ++++++++++++++++++++++++++++++------------- 2 files changed, 92 insertions(+), 38 deletions(-)
include/linux/swap.h +1 -0
@@ -361,6 +361,7 @@ static inline void folio_set_swap_entry(struct folio *folio, swp_entry_t entry) }/* linux/mm/workingset.c */ +bool workingset_test_recent (void *shadow,bool file,bool *workingset);void workingset_age_nonresident (struct lruvec *lruvec,unsigned long nr_pages);void *workingset_eviction (struct folio *folio,struct mem_cgroup *target_memcg);void workingset_refault (struct folio *folio,void *shadow);
mm/workingset.c +91 -38
185 lines not on this page · raw message
Signature · 1 line
2.30.2
Re: [PATCH v6 1/3] workingset: refactor LRU refault to expose refault recency check
Quoted 6 lines · Nhat Pham
In preparation for computing recently evicted pages in cachestat, refactor
workingset_refaultandlru_gen_refaultto expose a helper function that would test if an evicted page is recently evicted.Signed-off-by Nhat Pham <[email protected]>---
Hi Nhat,
I'm not terribly familiar with the workingset management code, but a few thoughts now that I've stared at it a bit...
Quoted 27 lines
include/linux/swap.h | 1 + mm/workingset.c | 129 ++++++++++++++++++++++++++++++------------- 2 files changed, 92 insertions(+), 38 deletions(-)include/linux/swap.h +1
-0@@ -361,6 +361,7 @@ static inline void folio_set_swap_entry(struct folio *folio, swp_entry_t entry) }/* linux/mm/workingset.c */ +bool workingset_test_recent (void *shadow,bool file,bool *workingset);void workingset_age_nonresident (struct lruvec *lruvec,unsigned long nr_pages);void *workingset_eviction (struct folio *folio,struct mem_cgroup *target_memcg);void workingset_refault (struct folio *folio,void *shadow);mm/workingset.c +3
-0@@ -244,6 +244,33 @@ static void *lru_gen_eviction(struct folio *folio)return pack_shadow (mem_cgroup_id (memcg), pgdat, token, refs); } +/* +* Test if the folio is recently evicted. +*
mm/workingset.c static void *lru_gen_eviction(struct folio *folio) · line 259
Quoted 10 lines
mm/workingset.c +10
-0+* As a side effect, also populates the references with +* values unpacked from the shadow of the evicted folio. +*/ +static bool lru_gen_test_recent (void *shadow,bool file,bool *workingset) +{ +struct mem_cgroup *eviction_memcg; +struct lruvec *lruvec; +struct lru_gen_struct *lrugen; +unsigned long min_seq; +
Extra whitespace looks a bit funny here.
mm/workingset.c static void *lru_gen_eviction(struct folio *folio) · line 271
Quoted 12 lines
mm/workingset.c +12
-0+int memcgid; +struct pglist_data *pgdat; +unsigned long token; + +unpack_shadow (shadow, &memcgid, &pgdat, &token, workingset); + eviction_memcg =mem_cgroup_from_id (memcgid); + + lruvec =mem_cgroup_lruvec (eviction_memcg, pgdat); + lrugen = &lruvec->lrugen; + + min_seq =READ_ONCE (lrugen->min_seq[file]); +return !((token >>LRU_REFS_WIDTH ) != (min_seq & (EVICTION_MASK >>LRU_REFS_WIDTH )));
I think this might be more readable without the double negative.
Also it looks like this logic is pulled from lru_gen_refault(). Any reason the caller isn't refactored to use this helper, similar to how workingset_refault() is modified? It seems like a potential landmine to duplicate the logic here for cachestat purposes and somewhere else for actual workingset management.
mm/workingset.c static void *lru_gen_eviction(struct folio *folio) · line 339
Quoted 13 lines
mm/workingset.c +6
-0+} +static void lru_gen_refault (struct folio *folio,void *shadow) {int hist, tier, refs;@@ -306,6 +333,11 @@ static void *lru_gen_eviction(struct folio *folio)return NULL ; } +static bool lru_gen_test_recent (void *shadow,bool file,bool *workingset) +{ +return true; +}
I guess this is a no-op for !MGLRU but given the context (i.e. special treatment for "recent" refaults), perhaps false is a more sane default?
Quoted 38 lines
mm/workingset.c +8
-14+static void lru_gen_refault (struct folio *folio,void *shadow) { }@@ -373,40 +405,31 @@ void *workingset_eviction(struct folio *folio, struct mem_cgroup *target_memcg) folio_test_workingset(folio)); } -/** -* workingset_refault - Evaluate the refault of a previously evicted folio. -* @folio: The freshly allocated replacement folio. -* @shadow: Shadow entry of the evicted folio. +/* +* Test if the folio is recently evicted by checking if +* refault distance of shadow exceeds workingset size. * -* Calculates and evaluates the refault distance of the previously -* evicted folio in the context of the node and the memcg whose memory -* pressure caused the eviction. +* As a side effect, populate workingset with the value +* unpacked from shadow. */ -void workingset_refault (struct folio *folio,void *shadow) +bool workingset_test_recent (void *shadow,bool file,bool *workingset) { -bool file =folio_is_file_lru (folio);struct mem_cgroup *eviction_memcg;struct lruvec *eviction_lruvec;unsigned long refault_distance;unsigned long workingset_size; -struct pglist_data *pgdat; -struct mem_cgroup *memcg; -unsigned long eviction; -struct lruvec *lruvec;unsigned long refault; -bool workingset; +int memcgid;
mm/workingset.c void *workingset_eviction(struct folio *folio, struct mem_cgroup *target_memcg) · line 428
Quoted 10 lines
mm/workingset.c +4
-5-long nr; +struct pglist_data *pgdat; +unsigned long eviction; -if (lru_gen_enabled ()) { -lru_gen_refault (folio, shadow); -return ; - } +if (lru_gen_enabled ()) +return lru_gen_test_recent (shadow, file, workingset);
Hmm.. so this function is only called by workingset_refault() when lru_gen_enabled() == false, otherwise it calls into lru_gen_refault(), which as noted above duplicates some of the recency logic.
I'm assuming this lru_gen_test_recent() call is so filemap_cachestat() can just call workingset_test_recent(). That seems reasonable, but makes me wonder...
Quoted 62 lines
mm/workingset.c +19
-19-unpack_shadow (shadow, &memcgid, &pgdat, &eviction, &workingset); +unpack_shadow (shadow, &memcgid, &pgdat, &eviction, workingset); eviction <<= bucket_order; -rcu_read_lock ();/* * Look up the memcg associated with the stored ID. It might * have been deleted since the folio's eviction. @@ -425,7 +448,8 @@ void workingset_refault(struct folio *folio, void *shadow)*/ eviction_memcg =mem_cgroup_from_id (memcgid);if (!mem_cgroup_disabled () && !eviction_memcg) - goto out; +return false; + eviction_lruvec =mem_cgroup_lruvec (eviction_memcg, pgdat); refault =atomic_long_read (&eviction_lruvec->nonresident_age);@@ -447,21 +471,6 @@ void workingset_refault(struct folio *folio, void *shadow)*/ refault_distance = (refault - eviction) &EVICTION_MASK ; -/* -* The activation decision for this folio is made at the level -* where the eviction occurred, as that is where the LRU order -* during folio reclaim is being determined. -* -* However, the cgroup that will own the folio is the one that -* is actually experiencing the refault event. -*/ - nr =folio_nr_pages (folio); - memcg =folio_memcg (folio); - pgdat =folio_pgdat (folio); - lruvec =mem_cgroup_lruvec (memcg, pgdat); - -mod_lruvec_state (lruvec,WORKINGSET_REFAULT_BASE + file, nr); -mem_cgroup_flush_stats_delayed ();/* * Compare the distance to the existing workingset size. We @@ -483,8 +492,51 @@ void workingset_refault(struct folio *folio, void *shadow) NR_INACTIVE_ANON); } } -if (refault_distance > workingset_size) + +return refault_distance <= workingset_size; +} + +/** +* workingset_refault - Evaluate the refault of a previously evicted folio. +* @folio: The freshly allocated replacement folio. +* @shadow: Shadow entry of the evicted folio. +* +* Calculates and evaluates the refault distance of the previously +* evicted folio in the context of the node and the memcg whose memory +* pressure caused the eviction. +*/ +void workingset_refault (struct folio *folio,void *shadow) +{ +bool file =folio_is_file_lru (folio);
mm/workingset.c void workingset_refault(struct folio *folio, void *shadow) · line 520
Quoted 10 lines
mm/workingset.c +10
-0+struct pglist_data *pgdat; +struct mem_cgroup *memcg; +struct lruvec *lruvec; +bool workingset; +long nr; + +if (lru_gen_enabled ()) { +lru_gen_refault (folio, shadow); +return ; + }
... if perhaps this should call workingset_test_recent() a bit earlier, since it also covers the lru_gen_*() case..? That may or may not be cleaner. It _seems like_ it might produce a bit more consistent logic, but just a thought and I could easily be missing details.
Quoted 7 lines
mm/workingset.c +7
-0+ +rcu_read_lock (); + + nr =folio_nr_pages (folio); + memcg =folio_memcg (folio); + pgdat =folio_pgdat (folio); + lruvec =mem_cgroup_lruvec (memcg, pgdat);
mm/workingset.c void workingset_refault(struct folio *folio, void *shadow) · line 537
Quoted 10 lines
mm/workingset.c +10
-0+ +if (!workingset_test_recent (shadow, file, &workingset)) { +/* +* The activation decision for this folio is made at the level +* where the eviction occurred, as that is where the LRU order +* during folio reclaim is being determined. +* +* However, the cgroup that will own the folio is the one that +* is actually experiencing the refault event. +*/
IIUC, this comment is explaining the difference between using the eviction lru (based on the shadow entry) to calculate recency vs. the lru for the current folio to process the refault. If so, perhaps it should go right above the workingset_test_recent() call? (Then the if braces could go away as well..).
mm/workingset.c void workingset_refault(struct folio *folio, void *shadow) · line 553
Quoted 10 lines
mm/workingset.c +2
-0goto out; + }folio_set_active (folio);workingset_age_nonresident (lruvec, nr);@@ -498,6 +550,7 @@ void workingset_refault(struct folio *folio, void *shadow)mod_lruvec_state (lruvec,WORKINGSET_RESTORE_BASE + file, nr); } out: +mod_lruvec_state (lruvec,WORKINGSET_REFAULT_BASE + file, nr);
Why not just leave this up earlier in the function (i.e. before the recency check) as it was originally?
Brian
Quoted 6 lines
mm/workingset.c +0
-0rcu_read_unlock (); }Signature · 1 line
2.30.2
Johannes Weiner
maintainerRe: [PATCH v6 1/3] workingset: refactor LRU refault to expose refault recency check
Quoted 16 lines · Brian Foster
mm/workingset.c static void *lru_gen_eviction(struct folio *folio) · line 271
Quoted 12 lines · Nhat Pham
mm/workingset.c +12
-0+int memcgid; +struct pglist_data *pgdat; +unsigned long token; + +unpack_shadow (shadow, &memcgid, &pgdat, &token, workingset); + eviction_memcg =mem_cgroup_from_id (memcgid); + + lruvec =mem_cgroup_lruvec (eviction_memcg, pgdat); + lrugen = &lruvec->lrugen; + + min_seq =READ_ONCE (lrugen->min_seq[file]); +return !((token >>LRU_REFS_WIDTH ) != (min_seq & (EVICTION_MASK >>LRU_REFS_WIDTH )));I think this might be more readable without the double negative.
Quoted 5 lines
Also it looks like this logic is pulled from
lru_gen_refault(). Any reason the caller isn't refactored to use this helper, similar to howworkingset_refault()is modified? It seems like a potential landmine to duplicate the logic here for cachestat purposes and somewhere else for actual workingset management.
The initial version was refactored. Yu explicitly requested it be duplicated [1] to cut down on some boiler plate.
I have to agree with Brian on this one, though. The factored version is better for maintenance than duplicating the core logic here. Even if it ends up a bit more boiler plate - it's harder to screw that up, and easier to catch at compile time, than the duplicates diverging.
[1] https://lore.kernel.org/lkml/CAOUHufZKTqoD2rFwrX9-eCknBmeWqP88rZ7X7A_5KHHbGBUP=A@mail.gmail.com/
Re: [PATCH v6 1/3] workingset: refactor LRU refault to expose refault recency check
Quoted 17 lines · Johannes Weiner
Quoted 16 lines · Brian Foster
mm/workingset.c static void *lru_gen_eviction(struct folio *folio) · line 271
Quoted 12 lines · Nhat Pham
mm/workingset.c +12
-0+int memcgid; +struct pglist_data *pgdat; +unsigned long token; + +unpack_shadow (shadow, &memcgid, &pgdat, &token, workingset); + eviction_memcg =mem_cgroup_from_id (memcgid); + + lruvec =mem_cgroup_lruvec (eviction_memcg, pgdat); + lrugen = &lruvec->lrugen; + + min_seq =READ_ONCE (lrugen->min_seq[file]); +return !((token >>LRU_REFS_WIDTH ) != (min_seq & (EVICTION_MASK >>LRU_REFS_WIDTH )));I think this might be more readable without the double negative.
Quoted 9 lines
Quoted 5 lines
Also it looks like this logic is pulled from
lru_gen_refault(). Any reason the caller isn't refactored to use this helper, similar to howworkingset_refault()is modified? It seems like a potential landmine to duplicate the logic here for cachestat purposes and somewhere else for actual workingset management.The initial version was refactored. Yu explicitly requested it be duplicated [1] to cut down on some boiler plate.
Ah, sorry for missing the previous discussion. TBH I wasn't terribly comfortable reviewing this one until I had made enough passes at the second patch..
Quoted 5 lines
I have to agree with Brian on this one, though. The factored version is better for maintenance than duplicating the core logic here. Even if it ends up a bit more boiler plate - it's harder to screw that up, and easier to catch at compile time, than the duplicates diverging.
It seems more elegant to me, FWIW. Glad I'm not totally off the rails at least. ;) But I'll defer to those who know the code better and the author, so that's just my .02. I don't want to cause this to go around in circles..
Brian
Re: [PATCH v6 1/3] workingset: refactor LRU refault to expose refault recency check
In reply to Johannes WeinerQuoted 27 lines · Johannes Weiner
Quoted 16 lines · Brian Foster
mm/workingset.c static void *lru_gen_eviction(struct folio *folio) · line 271
Quoted 12 lines · Nhat Pham
mm/workingset.c +12
-0+int memcgid; +struct pglist_data *pgdat; +unsigned long token; + +unpack_shadow (shadow, &memcgid, &pgdat, &token, workingset); + eviction_memcg =mem_cgroup_from_id (memcgid); + + lruvec =mem_cgroup_lruvec (eviction_memcg, pgdat); + lrugen = &lruvec->lrugen; + + min_seq =READ_ONCE (lrugen->min_seq[file]); +return !((token >>LRU_REFS_WIDTH ) != (min_seq & (EVICTION_MASK >>LRU_REFS_WIDTH )));I think this might be more readable without the double negative.
Quoted 5 lines
Also it looks like this logic is pulled from
lru_gen_refault(). Any reason the caller isn't refactored to use this helper, similar to howworkingset_refault()is modified? It seems like a potential landmine to duplicate the logic here for cachestat purposes and somewhere else for actual workingset management.The initial version was refactored. Yu explicitly requested it be duplicated [1] to cut down on some boiler plate.
Quoted 6 lines
I have to agree with Brian on this one, though. The factored version is better for maintenance than duplicating the core logic here. Even if it ends up a bit more boiler plate - it's harder to screw that up, and easier to catch at compile time, than the duplicates diverging.
[1] https://lore.kernel.org/lkml/CAOUHufZKTqoD2rFwrX9-eCknBmeWqP88rZ7X7A_5KHHbGBUP=A@mail.gmail.com/
No objections to either way. I'll take a look at the final version and we are good as long as it works as intended.
Nhat Pham
maintainerRe: [PATCH v6 1/3] workingset: refactor LRU refault to expose refault recency check
In reply to Brian FosterQuoted 54 lines · Brian Foster
Quoted 6 lines · Nhat Pham
In preparation for computing recently evicted pages in cachestat, refactor
workingset_refaultandlru_gen_refaultto expose a helper function that would test if an evicted page is recently evicted.Signed-off-by Nhat Pham <[email protected]>---
Hi Nhat,
I'm not terribly familiar with the workingset management code, but a few thoughts now that I've stared at it a bit...
Quoted 27 lines
include/linux/swap.h | 1 + mm/workingset.c | 129 ++++++++++++++++++++++++++++++------------- 2 files changed, 92 insertions(+), 38 deletions(-)include/linux/swap.h +1
-0@@ -361,6 +361,7 @@ static inline void folio_set_swap_entry(struct folio *folio, swp_entry_t entry) }/* linux/mm/workingset.c */ +bool workingset_test_recent (void *shadow,bool file,bool *workingset);void workingset_age_nonresident (struct lruvec *lruvec,unsigned long nr_pages);void *workingset_eviction (struct folio *folio,struct mem_cgroup *target_memcg);void workingset_refault (struct folio *folio,void *shadow);mm/workingset.c +3
-0@@ -244,6 +244,33 @@ static void *lru_gen_eviction(struct folio *folio)return pack_shadow (mem_cgroup_id (memcg), pgdat, token, refs); } +/* +* Test if the folio is recently evicted. +* mm/workingset.c static void *lru_gen_eviction(struct folio *folio) · line 259
Quoted 10 lines
mm/workingset.c +10
-0+* As a side effect, also populates the references with +* values unpacked from the shadow of the evicted folio. +*/ +static bool lru_gen_test_recent (void *shadow,bool file,bool *workingset) +{ +struct mem_cgroup *eviction_memcg; +struct lruvec *lruvec; +struct lru_gen_struct *lrugen; +unsigned long min_seq; +Extra whitespace looks a bit funny here.
Quoted 14 lines
mm/workingset.c static void *lru_gen_eviction(struct folio *folio) · line 271
Quoted 12 lines
mm/workingset.c +12
-0+int memcgid; +struct pglist_data *pgdat; +unsigned long token; + +unpack_shadow (shadow, &memcgid, &pgdat, &token, workingset); + eviction_memcg =mem_cgroup_from_id (memcgid); + + lruvec =mem_cgroup_lruvec (eviction_memcg, pgdat); + lrugen = &lruvec->lrugen; + + min_seq =READ_ONCE (lrugen->min_seq[file]); +return !((token >>LRU_REFS_WIDTH ) != (min_seq & (EVICTION_MASK >>LRU_REFS_WIDTH )));I think this might be more readable without the double negative.
Hmm indeed. I was just making sure that I did not mess up Yu's original logic here (by just wrapping it in a parentheses and negate the whole thing), but if I understand it correctly it's just an equality check. I'll fix it in the next version to make it cleaner.
Quoted 6 lines
Also it looks like this logic is pulled from
lru_gen_refault(). Any reason the caller isn't refactored to use this helper, similar to howworkingset_refault()is modified? It seems like a potential landmine to duplicate the logic here for cachestat purposes and somewhere else for actual workingset management.
In V2, it is actually refactored analogously as well - but we had a discussion about it here:
https://lkml.org/lkml/2022/12/5/1321
Quoted 10 lines
Quoted 10 lines
mm/workingset.c +2
-0+} +static void lru_gen_refault (struct folio *folio,void *shadow) {int hist, tier, refs;@@ -306,6 +333,11 @@ static void *lru_gen_eviction(struct folio *folio)return NULL ; }
Quoted 7 lines
mm/workingset.c static void *lru_gen_eviction(struct folio *folio) · line 339
Quoted 4 lines
mm/workingset.c +4
-0+static bool lru_gen_test_recent (void *shadow,bool file,bool *workingset) +{ +return true; +}I guess this is a no-op for !MGLRU but given the context (i.e. special treatment for "recent" refaults), perhaps false is a more sane default?
Hmm, fair point. Let me fix that in the next version.
Quoted 54 lines
Quoted 39 lines
mm/workingset.c +8
-14+static void lru_gen_refault (struct folio *folio,void *shadow) { }@@ -373,40 +405,31 @@ void *workingset_eviction(struct folio *folio, struct mem_cgroup *target_memcg) folio_test_workingset(folio)); } -/** -* workingset_refault - Evaluate the refault of a previously evicted folio. -* @folio: The freshly allocated replacement folio. -* @shadow: Shadow entry of the evicted folio. +/* +* Test if the folio is recently evicted by checking if +* refault distance of shadow exceeds workingset size. * -* Calculates and evaluates the refault distance of the previously -* evicted folio in the context of the node and the memcg whose memory -* pressure caused the eviction. +* As a side effect, populate workingset with the value +* unpacked from shadow. */ -void workingset_refault (struct folio *folio,void *shadow) +bool workingset_test_recent (void *shadow,bool file,bool *workingset) { -bool file =folio_is_file_lru (folio);struct mem_cgroup *eviction_memcg;struct lruvec *eviction_lruvec;unsigned long refault_distance;unsigned long workingset_size; -struct pglist_data *pgdat; -struct mem_cgroup *memcg; -unsigned long eviction; -struct lruvec *lruvec;unsigned long refault; -bool workingset; +int memcgid;mm/workingset.c void *workingset_eviction(struct folio *folio, struct mem_cgroup *target_memcg) · line 428
Quoted 10 lines
mm/workingset.c +4
-5-long nr; +struct pglist_data *pgdat; +unsigned long eviction; -if (lru_gen_enabled ()) { -lru_gen_refault (folio, shadow); -return ; - } +if (lru_gen_enabled ()) +return lru_gen_test_recent (shadow, file, workingset);Hmm.. so this function is only called by
workingset_refault()whenlru_gen_enabled()== false, otherwise it calls intolru_gen_refault(), which as noted above duplicates some of the recency logic.
Quoted 3 lines
I'm assuming this
lru_gen_test_recent()call is sofilemap_cachestat()can just callworkingset_test_recent(). That seems reasonable, but makes me wonder...
You're right. It's a bit clunky...
Quoted 70 lines
Quoted 70 lines
mm/workingset.c +26
-19-unpack_shadow (shadow, &memcgid, &pgdat, &eviction, &workingset); +unpack_shadow (shadow, &memcgid, &pgdat, &eviction, workingset); eviction <<= bucket_order; -rcu_read_lock ();/* * Look up the memcg associated with the stored ID. It might * have been deleted since the folio's eviction. @@ -425,7 +448,8 @@ void workingset_refault(struct folio *folio, void *shadow)*/ eviction_memcg =mem_cgroup_from_id (memcgid);if (!mem_cgroup_disabled () && !eviction_memcg) - goto out; +return false; + eviction_lruvec =mem_cgroup_lruvec (eviction_memcg, pgdat); refault =atomic_long_read (&eviction_lruvec->nonresident_age);@@ -447,21 +471,6 @@ void workingset_refault(struct folio *folio, void *shadow)*/ refault_distance = (refault - eviction) &EVICTION_MASK ; -/* -* The activation decision for this folio is made at the level -* where the eviction occurred, as that is where the LRU order -* during folio reclaim is being determined. -* -* However, the cgroup that will own the folio is the one that -* is actually experiencing the refault event. -*/ - nr =folio_nr_pages (folio); - memcg =folio_memcg (folio); - pgdat =folio_pgdat (folio); - lruvec =mem_cgroup_lruvec (memcg, pgdat); - -mod_lruvec_state (lruvec,WORKINGSET_REFAULT_BASE + file, nr); -mem_cgroup_flush_stats_delayed ();/* * Compare the distance to the existing workingset size. We @@ -483,8 +492,51 @@ void workingset_refault(struct folio *folio, void *shadow) NR_INACTIVE_ANON); } } -if (refault_distance > workingset_size) + +return refault_distance <= workingset_size; +} + +/** +* workingset_refault - Evaluate the refault of a previously evicted folio. +* @folio: The freshly allocated replacement folio. +* @shadow: Shadow entry of the evicted folio. +* +* Calculates and evaluates the refault distance of the previously +* evicted folio in the context of the node and the memcg whose memory +* pressure caused the eviction. +*/ +void workingset_refault (struct folio *folio,void *shadow) +{ +bool file =folio_is_file_lru (folio); +struct pglist_data *pgdat; +struct mem_cgroup *memcg; +struct lruvec *lruvec; +bool workingset; +long nr; + +if (lru_gen_enabled ()) {
Quoted 8 lines
mm/workingset.c void workingset_refault(struct folio *folio, void *shadow) · line 520
Quoted 3 lines
mm/workingset.c +3
-0+lru_gen_refault (folio, shadow); +return ; + }... if perhaps this should call
workingset_test_recent()a bit earlier, since it also covers the lru_gen_*() case..? That may or may not be cleaner. It_seemslike_it might produce a bit more consistent logic, but just a thought and I could easily be missing details.
Hmm you mean before/in place of the lru_gen_refault call? workingset_test_recent only covers lru_gen_test_recent, not the rest of the logic of lru_gen_refault I believe.
Quoted 16 lines
Quoted 16 lines
mm/workingset.c +15
-0+ +rcu_read_lock (); + + nr =folio_nr_pages (folio); + memcg =folio_memcg (folio); + pgdat =folio_pgdat (folio); + lruvec =mem_cgroup_lruvec (memcg, pgdat); + +if (!workingset_test_recent (shadow, file, &workingset)) { +/* +* The activation decision for this folio is made at the level +* where the eviction occurred, as that is where the LRU order +* during folio reclaim is being determined. +* +* However, the cgroup that will own the folio is the one that
Quoted 8 lines
mm/workingset.c void workingset_refault(struct folio *folio, void *shadow) · line 537
Quoted 2 lines
mm/workingset.c +2
-0+* is actually experiencing the refault event. +*/ IIUC, this comment is explaining the difference between using the eviction lru (based on the shadow entry) to calculate recency vs. the lru for the current folio to process the refault. If so, perhaps it should go right above the
workingset_test_recent()call? (Then the if braces could go away as well..).
You're right! I think it should go above nr = folio_nr_pages(folio); call.
Quoted 4 lines
Quoted 4 lines
mm/workingset.c +1
-0goto out; + }
Quoted 10 lines
mm/workingset.c void workingset_refault(struct folio *folio, void *shadow) · line 553
Quoted 7 lines
mm/workingset.c +1
-0folio_set_active (folio);workingset_age_nonresident (lruvec, nr);@@ -498,6 +550,7 @@ void workingset_refault(struct folio *folio, void *shadow)mod_lruvec_state (lruvec,WORKINGSET_RESTORE_BASE + file, nr); } out: +mod_lruvec_state (lruvec,WORKINGSET_REFAULT_BASE + file, nr);Why not just leave this up earlier in the function (i.e. before the recency check) as it was originally?
Let me double check, but I think this is a relic from the old (and incorrect) version of workingset code.
Originally, mod_lruvec_state uses the lruvec computed from a variable (pgdat) that was unpacked from the shadow. So this mod_lruvec_state has to go after the unpack_shadow call (which has been moved inside of workingset_test_recent).
This is actually wrong - we actually want the pgdat from the folio. It has been fixed in a separate patch:
https://lore.kernel.org/all/[email protected]/T/#u
But I didn't update it here. Let me stare at it a bit more to make sure, and then fix it in the next version. It should not change the behavior, but it should be cleaner.
Quoted 10 lines
Brian
Quoted 6 lines
mm/workingset.c +0
-1rcu_read_unlock (); } --2.30.2
Re: [PATCH v6 1/3] workingset: refactor LRU refault to expose refault recency check
Quoted 82 lines · Nhat Pham
Quoted 54 lines · Brian Foster
Quoted 6 lines · Nhat Pham
In preparation for computing recently evicted pages in cachestat, refactor
workingset_refaultandlru_gen_refaultto expose a helper function that would test if an evicted page is recently evicted.Signed-off-by Nhat Pham <[email protected]>---
Hi Nhat,
I'm not terribly familiar with the workingset management code, but a few thoughts now that I've stared at it a bit...
Quoted 27 lines
include/linux/swap.h | 1 + mm/workingset.c | 129 ++++++++++++++++++++++++++++++------------- 2 files changed, 92 insertions(+), 38 deletions(-)include/linux/swap.h +1
-0@@ -361,6 +361,7 @@ static inline void folio_set_swap_entry(struct folio *folio, swp_entry_t entry) }/* linux/mm/workingset.c */ +bool workingset_test_recent (void *shadow,bool file,bool *workingset);void workingset_age_nonresident (struct lruvec *lruvec,unsigned long nr_pages);void *workingset_eviction (struct folio *folio,struct mem_cgroup *target_memcg);void workingset_refault (struct folio *folio,void *shadow);mm/workingset.c +3
-0@@ -244,6 +244,33 @@ static void *lru_gen_eviction(struct folio *folio)return pack_shadow (mem_cgroup_id (memcg), pgdat, token, refs); } +/* +* Test if the folio is recently evicted. +* mm/workingset.c static void *lru_gen_eviction(struct folio *folio) · line 259
Quoted 10 lines
mm/workingset.c +10
-0+* As a side effect, also populates the references with +* values unpacked from the shadow of the evicted folio. +*/ +static bool lru_gen_test_recent (void *shadow,bool file,bool *workingset) +{ +struct mem_cgroup *eviction_memcg; +struct lruvec *lruvec; +struct lru_gen_struct *lrugen; +unsigned long min_seq; +Extra whitespace looks a bit funny here.
Quoted 14 lines
mm/workingset.c static void *lru_gen_eviction(struct folio *folio) · line 271
Quoted 12 lines
mm/workingset.c +12
-0+int memcgid; +struct pglist_data *pgdat; +unsigned long token; + +unpack_shadow (shadow, &memcgid, &pgdat, &token, workingset); + eviction_memcg =mem_cgroup_from_id (memcgid); + + lruvec =mem_cgroup_lruvec (eviction_memcg, pgdat); + lrugen = &lruvec->lrugen; + + min_seq =READ_ONCE (lrugen->min_seq[file]); +return !((token >>LRU_REFS_WIDTH ) != (min_seq & (EVICTION_MASK >>LRU_REFS_WIDTH )));I think this might be more readable without the double negative.
Hmm indeed. I was just making sure that I did not mess up Yu's original logic here (by just wrapping it in a parentheses and negate the whole thing), but if I understand it correctly it's just an equality check. I'll fix it in the next version to make it cleaner.
Quoted 6 lines
Also it looks like this logic is pulled from
lru_gen_refault(). Any reason the caller isn't refactored to use this helper, similar to howworkingset_refault()is modified? It seems like a potential landmine to duplicate the logic here for cachestat purposes and somewhere else for actual workingset management.
Quoted 5 lines
In V2, it is actually refactored analogously as well - but we had a discussion about it here:
Yeah, sorry.. replied to Johannes.
Quoted 159 lines
Quoted 10 lines
Quoted 10 lines
mm/workingset.c +2
-0+} +static void lru_gen_refault (struct folio *folio,void *shadow) {int hist, tier, refs;@@ -306,6 +333,11 @@ static void *lru_gen_eviction(struct folio *folio)return NULL ; }Quoted 7 lines
mm/workingset.c static void *lru_gen_eviction(struct folio *folio) · line 339
Quoted 4 lines
mm/workingset.c +4
-0+static bool lru_gen_test_recent (void *shadow,bool file,bool *workingset) +{ +return true; +}I guess this is a no-op for !MGLRU but given the context (i.e. special treatment for "recent" refaults), perhaps false is a more sane default?
Hmm, fair point. Let me fix that in the next version.
Quoted 54 lines
Quoted 39 lines
mm/workingset.c +8
-14+static void lru_gen_refault (struct folio *folio,void *shadow) { }@@ -373,40 +405,31 @@ void *workingset_eviction(struct folio *folio, struct mem_cgroup *target_memcg) folio_test_workingset(folio)); } -/** -* workingset_refault - Evaluate the refault of a previously evicted folio. -* @folio: The freshly allocated replacement folio. -* @shadow: Shadow entry of the evicted folio. +/* +* Test if the folio is recently evicted by checking if +* refault distance of shadow exceeds workingset size. * -* Calculates and evaluates the refault distance of the previously -* evicted folio in the context of the node and the memcg whose memory -* pressure caused the eviction. +* As a side effect, populate workingset with the value +* unpacked from shadow. */ -void workingset_refault (struct folio *folio,void *shadow) +bool workingset_test_recent (void *shadow,bool file,bool *workingset) { -bool file =folio_is_file_lru (folio);struct mem_cgroup *eviction_memcg;struct lruvec *eviction_lruvec;unsigned long refault_distance;unsigned long workingset_size; -struct pglist_data *pgdat; -struct mem_cgroup *memcg; -unsigned long eviction; -struct lruvec *lruvec;unsigned long refault; -bool workingset; +int memcgid;mm/workingset.c void *workingset_eviction(struct folio *folio, struct mem_cgroup *target_memcg) · line 428
Quoted 10 lines
mm/workingset.c +4
-5-long nr; +struct pglist_data *pgdat; +unsigned long eviction; -if (lru_gen_enabled ()) { -lru_gen_refault (folio, shadow); -return ; - } +if (lru_gen_enabled ()) +return lru_gen_test_recent (shadow, file, workingset);Hmm.. so this function is only called by
workingset_refault()whenlru_gen_enabled()== false, otherwise it calls intolru_gen_refault(), which as noted above duplicates some of the recency logic.Quoted 3 lines
I'm assuming this
lru_gen_test_recent()call is sofilemap_cachestat()can just callworkingset_test_recent(). That seems reasonable, but makes me wonder...You're right. It's a bit clunky...
Quoted 70 lines
Quoted 70 lines
mm/workingset.c +26
-19-unpack_shadow (shadow, &memcgid, &pgdat, &eviction, &workingset); +unpack_shadow (shadow, &memcgid, &pgdat, &eviction, workingset); eviction <<= bucket_order; -rcu_read_lock ();/* * Look up the memcg associated with the stored ID. It might * have been deleted since the folio's eviction. @@ -425,7 +448,8 @@ void workingset_refault(struct folio *folio, void *shadow)*/ eviction_memcg =mem_cgroup_from_id (memcgid);if (!mem_cgroup_disabled () && !eviction_memcg) - goto out; +return false; + eviction_lruvec =mem_cgroup_lruvec (eviction_memcg, pgdat); refault =atomic_long_read (&eviction_lruvec->nonresident_age);@@ -447,21 +471,6 @@ void workingset_refault(struct folio *folio, void *shadow)*/ refault_distance = (refault - eviction) &EVICTION_MASK ; -/* -* The activation decision for this folio is made at the level -* where the eviction occurred, as that is where the LRU order -* during folio reclaim is being determined. -* -* However, the cgroup that will own the folio is the one that -* is actually experiencing the refault event. -*/ - nr =folio_nr_pages (folio); - memcg =folio_memcg (folio); - pgdat =folio_pgdat (folio); - lruvec =mem_cgroup_lruvec (memcg, pgdat); - -mod_lruvec_state (lruvec,WORKINGSET_REFAULT_BASE + file, nr); -mem_cgroup_flush_stats_delayed ();/* * Compare the distance to the existing workingset size. We @@ -483,8 +492,51 @@ void workingset_refault(struct folio *folio, void *shadow) NR_INACTIVE_ANON); } } -if (refault_distance > workingset_size) + +return refault_distance <= workingset_size; +} + +/** +* workingset_refault - Evaluate the refault of a previously evicted folio. +* @folio: The freshly allocated replacement folio. +* @shadow: Shadow entry of the evicted folio. +* +* Calculates and evaluates the refault distance of the previously +* evicted folio in the context of the node and the memcg whose memory +* pressure caused the eviction. +*/ +void workingset_refault (struct folio *folio,void *shadow) +{ +bool file =folio_is_file_lru (folio); +struct pglist_data *pgdat; +struct mem_cgroup *memcg; +struct lruvec *lruvec; +bool workingset; +long nr; + +if (lru_gen_enabled ()) {Quoted 8 lines
mm/workingset.c void workingset_refault(struct folio *folio, void *shadow) · line 520
Quoted 3 lines
mm/workingset.c +3
-0+lru_gen_refault (folio, shadow); +return ; + }... if perhaps this should call
workingset_test_recent()a bit earlier, since it also covers the lru_gen_*() case..? That may or may not be cleaner. It_seemslike_it might produce a bit more consistent logic, but just a thought and I could easily be missing details.
Quoted 4 lines
Hmm you mean before/in place of the
lru_gen_refaultcall?workingset_test_recentonly coverslru_gen_test_recent, not the rest of the logic oflru_gen_refaultI believe.
When reading through the code I got the impression that if workingset_test_recent() -> lru_gen_test_recent() returned false, then lru_gen_refault() didn't really do anything. I.e., the token/min_seq check fails and it skips out to the end of the function. That had me wondering whether workingset_refault() could just skip out of either mode if workingset_test_recent() returns false (since it calls lru_gen_test_recent()), but that may not work. Specifically I'm not sure about that mod_lruvec_state()) call (discussed below) for the MGLRU case. If that call is correct as is for !MGLRU, maybe it could be made conditional based on lru_gen_enabled()..?
I guess what I was trying to get at here is that you've created this nice workingset_test_recent() helper to cover both modes for filemap_cachestat(), so it would be nice if it could be used in that way in the core workingset code as well. :)
I suppose yet another option could be to skip the lru_gen_test_recent() check in workingset_test_recent() and instead call it from lru_gen_refault(), but then I guess you'd have to open code both checks in the filemap_cachestat() call, which sounds kind of ugly..
Quoted 26 lines
Quoted 17 lines
Quoted 17 lines
mm/workingset.c +15
-0+ +rcu_read_lock (); + + nr =folio_nr_pages (folio); + memcg =folio_memcg (folio); + pgdat =folio_pgdat (folio); + lruvec =mem_cgroup_lruvec (memcg, pgdat); + +if (!workingset_test_recent (shadow, file, &workingset)) { +/* +* The activation decision for this folio is made at the level +* where the eviction occurred, as that is where the LRU order +* during folio reclaim is being determined. +* +* However, the cgroup that will own the folio is the one that Quoted 8 lines
mm/workingset.c void workingset_refault(struct folio *folio, void *shadow) · line 537
Quoted 2 lines
mm/workingset.c +2
-0+* is actually experiencing the refault event. +*/ IIUC, this comment is explaining the difference between using the eviction lru (based on the shadow entry) to calculate recency vs. the lru for the current folio to process the refault. If so, perhaps it should go right above the
workingset_test_recent()call? (Then the if braces could go away as well..).
Quoted 2 lines
You're right! I think it should go above
nr = folio_nr_pages(folio);call.
Sounds good.
Quoted 23 lines
Quoted 4 lines
Quoted 4 lines
mm/workingset.c +1
-0goto out; + }Quoted 10 lines
mm/workingset.c void workingset_refault(struct folio *folio, void *shadow) · line 553
Quoted 7 lines
mm/workingset.c +1
-0folio_set_active (folio);workingset_age_nonresident (lruvec, nr);@@ -498,6 +550,7 @@ void workingset_refault(struct folio *folio, void *shadow)mod_lruvec_state (lruvec,WORKINGSET_RESTORE_BASE + file, nr); } out: +mod_lruvec_state (lruvec,WORKINGSET_REFAULT_BASE + file, nr);Why not just leave this up earlier in the function (i.e. before the recency check) as it was originally?
Let me double check, but I think this is a relic from the old (and incorrect) version of workingset code.
Originally,
mod_lruvec_stateuses the lruvec computed from a variable (pgdat) that was unpacked from the shadow. So thismod_lruvec_statehas to go after theunpack_shadowcall (which has been moved inside ofworkingset_test_recent).
Quoted 5 lines
This is actually wrong - we actually want the pgdat from the folio. It has been fixed in a separate patch:
Yep, I had applied this series on top of that one..
Quoted 4 lines
But I didn't update it here. Let me stare at it a bit more to make sure, and then fix it in the next version. It should not change the behavior, but it should be cleaner.
Sounds good. FWIW it looked like the logic hadn't changed with this series so I just assumed it was correct, just possibly moved around unnecessarily. I'll try to make more sense of it in the next version. Thanks.
Brian
Quoted 11 lines
Quoted 10 lines
Brian
Quoted 6 lines
mm/workingset.c +0
-1rcu_read_unlock (); } --2.30.2
Nhat Pham
maintainerPATCH v6 2/3 cachestat: implement cachestat syscall
In reply to Nhat PhamThere is currently no good way to query the page cache state of large file sets and directory trees. There is mincore(), but it scales poorly:
the kernel writes out a lot of bitmap data that userspace has to aggregate, when the user really doesn not care about per-page information in that case. The user also needs to mmap and unmap each file as it goes along, which can be quite slow as well.
This patch implements a new syscall that queries cache state of a file and summarizes the number of cached pages, number of dirty pages, number of pages marked for writeback, number of (recently) evicted pages, etc. in a given range.
NAME
cachestat - query the page cache statistics of a file.
SYNOPSIS
#include <sys/mman.h>
struct cachestat {
__u64 nr_cache;
__u64 nr_dirty;
__u64 nr_writeback;
__u64 nr_evicted;
__u64 nr_recently_evicted;
};
int cachestat(unsigned int fd, off_t off, size_t len,
unsigned int cstat_version, struct cachestat *cstat,
unsigned int flags);DESCRIPTION
cachestat() queries the number of cached pages, number of dirty pages, number of pages marked for writeback, number of evicted pages, number of recently evicted pages, in the bytes range given by off and len.
An evicted page is a page that is previously in the page cache but has been evicted since. A page is recently evicted if its last eviction was recent enough that its reentry to the cache would indicate that it is actively being used by the system, and that there is memory pressure on the system.
These values are returned in a cachestat struct, whose address is given by the cstat argument.
The off and len arguments must be non-negative integers. If len > 0, the queried range is [off, off + len]. If len == 0, we will query in the range from off to the end of the file.
cstat_version is an unsigned integer indicating the specific version of the cachestat struct. It must be at least 1, and does not exceed the latest version number (which is currently 1). For now, user should just pass 1.
The flags argument is unused for now, but is included for future extensibility. User should pass 0 (i.e no flag specified).
RETURN VALUE
On success, cachestat returns 0. On error, -1 is returned, and errno is set to indicate the error.
ERRORS
EFAULT cstat points to an invalid address.
EINVAL invalid cstat_version or flags
EBADF invalid file descriptor.
--- arch/alpha/kernel/syscalls/syscall.tbl | 1 + arch/arm/tools/syscall.tbl | 1 + arch/ia64/kernel/syscalls/syscall.tbl | 1 + arch/m68k/kernel/syscalls/syscall.tbl | 1 + arch/microblaze/kernel/syscalls/syscall.tbl | 1 + arch/parisc/kernel/syscalls/syscall.tbl | 1 + arch/powerpc/kernel/syscalls/syscall.tbl | 1 + arch/s390/kernel/syscalls/syscall.tbl | 1 + arch/sh/kernel/syscalls/syscall.tbl | 1 + arch/sparc/kernel/syscalls/syscall.tbl | 1 + arch/x86/entry/syscalls/syscall_32.tbl | 1 + arch/x86/entry/syscalls/syscall_64.tbl | 1 + arch/xtensa/kernel/syscalls/syscall.tbl | 1 + include/linux/fs.h | 3 + include/linux/syscalls.h | 4 + include/uapi/asm-generic/unistd.h | 5 +- include/uapi/linux/mman.h | 9 ++ init/Kconfig | 10 ++ kernel/sys_ni.c | 1 + mm/filemap.c | 154 ++++++++++++++++++++ 20 files changed, 198 insertions(+), 1 deletion(-)
arch/alpha/kernel/syscalls/syscall.tbl +1 -0
@@ -490,3 +490,4 @@ 558 common process_mrelease sys_process_mrelease 559 common futex_waitv sys_futex_waitv 560 common set_mempolicy_home_node sys_ni_syscall +561 common cachestat sys_cachestat
arch/arm/tools/syscall.tbl +1 -0
@@ -464,3 +464,4 @@ 448 common process_mrelease sys_process_mrelease 449 common futex_waitv sys_futex_waitv 450 common set_mempolicy_home_node sys_set_mempolicy_home_node +451 common cachestat sys_cachestat
arch/ia64/kernel/syscalls/syscall.tbl +1 -0
@@ -371,3 +371,4 @@ 448 common process_mrelease sys_process_mrelease 449 common futex_waitv sys_futex_waitv 450 common set_mempolicy_home_node sys_set_mempolicy_home_node +451 common cachestat sys_cachestat
arch/m68k/kernel/syscalls/syscall.tbl +1 -0
@@ -450,3 +450,4 @@ 448 common process_mrelease sys_process_mrelease 449 common futex_waitv sys_futex_waitv 450 common set_mempolicy_home_node sys_set_mempolicy_home_node +451 common cachestat sys_cachestat
arch/microblaze/kernel/syscalls/syscall.tbl +1 -0
@@ -456,3 +456,4 @@ 448 common process_mrelease sys_process_mrelease 449 common futex_waitv sys_futex_waitv 450 common set_mempolicy_home_node sys_set_mempolicy_home_node +451 common cachestat sys_cachestat
arch/parisc/kernel/syscalls/syscall.tbl +1 -0
@@ -448,3 +448,4 @@ 448 common process_mrelease sys_process_mrelease 449 common futex_waitv sys_futex_waitv 450 common set_mempolicy_home_node sys_set_mempolicy_home_node +451 common cachestat sys_cachestat
arch/powerpc/kernel/syscalls/syscall.tbl +1 -0
@@ -530,3 +530,4 @@ 448 common process_mrelease sys_process_mrelease 449 common futex_waitv sys_futex_waitv 450 nospu set_mempolicy_home_node sys_set_mempolicy_home_node +451 common cachestat sys_cachestat
arch/s390/kernel/syscalls/syscall.tbl +1 -0
@@ -453,3 +453,4 @@ 448 common process_mrelease sys_process_mrelease sys_process_mrelease 449 common futex_waitv sys_futex_waitv sys_futex_waitv 450 common set_mempolicy_home_node sys_set_mempolicy_home_node sys_set_mempolicy_home_node +451 common cachestat sys_cachestat sys_cachestat
arch/sh/kernel/syscalls/syscall.tbl +1 -0
@@ -453,3 +453,4 @@ 448 common process_mrelease sys_process_mrelease 449 common futex_waitv sys_futex_waitv 450 common set_mempolicy_home_node sys_set_mempolicy_home_node +451 common cachestat sys_cachestat
arch/sparc/kernel/syscalls/syscall.tbl +1 -0
@@ -496,3 +496,4 @@ 448 common process_mrelease sys_process_mrelease 449 common futex_waitv sys_futex_waitv 450 common set_mempolicy_home_node sys_set_mempolicy_home_node +451 common cachestat sys_cachestat
arch/x86/entry/syscalls/syscall_32.tbl +1 -0
@@ -455,3 +455,4 @@ 448 i386 process_mrelease sys_process_mrelease 449 i386 futex_waitv sys_futex_waitv 450 i386 set_mempolicy_home_node sys_set_mempolicy_home_node +451 i386 cachestat sys_cachestat
arch/x86/entry/syscalls/syscall_64.tbl +1 -0
@@ -372,6 +372,7 @@ 448 common process_mrelease sys_process_mrelease 449 common futex_waitv sys_futex_waitv 450 common set_mempolicy_home_node sys_set_mempolicy_home_node +451 common cachestat sys_cachestat # # Due to a historical design error, certain syscalls are numbered differently
arch/xtensa/kernel/syscalls/syscall.tbl +1 -0
@@ -421,3 +421,4 @@ 448 common process_mrelease sys_process_mrelease 449 common futex_waitv sys_futex_waitv 450 common set_mempolicy_home_node sys_set_mempolicy_home_node +451 common cachestat sys_cachestat
include/linux/fs.h +3 -0
@@ -75,6 +75,7 @@ struct fs_context;struct fs_parameter_spec ;struct fileattr ;struct iomap_ops ; +struct cachestat ;extern void __init inode_init (void );extern void __init inode_init_early (void );
@@ -830,6 +831,8 @@ void filemap_invalidate_lock_two(struct address_space *mapping1, struct address_space *mapping2);void filemap_invalidate_unlock_two (struct address_space *mapping1,struct address_space *mapping2); +void filemap_cachestat (struct address_space *mapping,pgoff_t first_index, +pgoff_t last_index,struct cachestat *cs);/*
include/linux/syscalls.h +4 -0
@@ -72,6 +72,7 @@ struct open_how;struct mount_attr ;struct landlock_ruleset_attr ;enum landlock_rule_type ; +struct cachestat ;#include <linux/types.h> #include <linux/aio_abi.h>
@@ -1056,6 +1057,9 @@ asmlinkage long sys_memfd_secret(unsigned int flags);asmlinkage long sys_set_mempolicy_home_node (unsigned long start,unsigned long len,unsigned long home_node,unsigned long flags); +asmlinkage long sys_cachestat (unsigned int fd,off_t off,size_t len, +unsigned int cstat_version,struct cachestat __user *cstat, +unsigned int flags);/* * Architecture-specific system calls
include/uapi/asm-generic/unistd.h +4 -1
@@ -886,8 +886,11 @@ __SYSCALL(__NR_futex_waitv, sys_futex_waitv)#define __NR_set_mempolicy_home_node 450 __SYSCALL(__NR_set_mempolicy_home_node, sys_set_mempolicy_home_node) +#define __NR_cachestat 451 +__SYSCALL(__NR_cachestat, sys_cachestat) +#undef __NR_syscalls -#define __NR_syscalls 451 +#define __NR_syscalls 452/* * 32 bit systems traditionally used different
include/uapi/linux/mman.h +9 -0
@@ -4,6 +4,7 @@#include <asm/mman.h> #include <asm-generic/hugetlb_encode.h> +#include <linux/types.h> #define MREMAP_MAYMOVE 1#define MREMAP_FIXED 2
@@ -41,4 +42,12 @@#define MAP_HUGE_2GB HUGETLB_FLAG_ENCODE_2GB#define MAP_HUGE_16GB HUGETLB_FLAG_ENCODE_16GB +struct cachestat { +__u64 nr_cache; +__u64 nr_dirty; +__u64 nr_writeback; +__u64 nr_evicted; +__u64 nr_recently_evicted; +}; +#endif /* _UAPI_LINUX_MMAN_H */
init/Kconfig +10 -0
@@ -1798,6 +1798,16 @@ config RSEQ If unsure, say Y. +config CACHESTAT_SYSCALL + bool "Enable cachestat() system call" if EXPERT + default y + help + Enable the cachestat system call, which queries the page cache + statistics of a file (number of cached pages, dirty pages, + pages marked for writeback, (recently) evicted pages). + + If unsure say Y here. + config DEBUG_RSEQ default n bool "Enabled debugging of rseq() system call" if EXPERT
kernel/sys_ni.c +1 -0
@@ -299,6 +299,7 @@ COND_SYSCALL(set_mempolicy);COND_SYSCALL (migrate_pages);COND_SYSCALL (move_pages);COND_SYSCALL (set_mempolicy_home_node); +COND_SYSCALL (cachestat);COND_SYSCALL (perf_event_open);COND_SYSCALL (accept4);
mm/filemap.c +154 -0
169 lines not on this page · raw message
Signature · 1 line
2.30.2
Re: [PATCH v6 2/3] cachestat: implement cachestat syscall
Quoted 45 lines · Nhat Pham
There is currently no good way to query the page cache state of large file sets and directory trees. There is
mincore(), but it scales poorly:
the kernel writes out a lot of bitmap data that userspace has to aggregate, when the user really doesn not care about per-page information in that case. The user also needs to mmap and unmap each file as it goes along, which can be quite slow as well.This patch implements a new syscall that queries cache state of a file and summarizes the number of cached pages, number of dirty pages, number of pages marked for writeback, number of (recently) evicted pages, etc. in a given range.
NAME
cachestat - query the page cache statistics of a file.SYNOPSIS
#include <sys/mman.h> struct cachestat { __u64 nr_cache; __u64 nr_dirty; __u64 nr_writeback; __u64 nr_evicted; __u64 nr_recently_evicted; }; int cachestat(unsigned int fd, off_t off, size_t len, unsigned int cstat_version, struct cachestat *cstat, unsigned int flags);DESCRIPTION
cachestat()queries the number of cached pages, number of dirty pages, number of pages marked for writeback, number of evicted pages, number of recently evicted pages, in the bytes range given byoffandlen.An evicted page is a page that is previously in the page cache but has been evicted since. A page is recently evicted if its last eviction was recent enough that its reentry to the cache would indicate that it is actively being used by the system, and that there is memory pressure on the system.
These values are returned in a cachestat struct, whose address is given by the
cstatargument.
Quoted 9 lines
The
offandlenarguments must be non-negative integers. Iflen> 0, the queried range is [off,off+len]. Iflen== 0, we will query in the range fromoffto the end of the file.
cstat_versionis an unsigned integer indicating the specific version of the cachestat struct. It must be at least 1, and does not exceed the latest version number (which is currently 1). For now, user should just pass 1.
Still not sure about this vs. just requiring the structure size to match (and maybe padding it), but perhaps linux-api can comment on the best way to future proof.
I'm not the person to ack a syscall regardless, but since I've had various previous comments on this patch and no further issues stand out to me:
Quoted 455 lines
The
flagsargument is unused for now, but is included for future extensibility. User should pass 0 (i.e no flag specified).RETURN VALUE
On success, cachestat returns 0. On error, -1 is returned, and errno is set to indicate the error.ERRORS
EFAULT cstat points to an invalid address.EINVAL invalid
cstat_versionorflagsEBADF invalid file descriptor.
Signed-off-by Nhat Pham <[email protected]>--- arch/alpha/kernel/syscalls/syscall.tbl | 1 + arch/arm/tools/syscall.tbl | 1 + arch/ia64/kernel/syscalls/syscall.tbl | 1 + arch/m68k/kernel/syscalls/syscall.tbl | 1 + arch/microblaze/kernel/syscalls/syscall.tbl | 1 + arch/parisc/kernel/syscalls/syscall.tbl | 1 + arch/powerpc/kernel/syscalls/syscall.tbl | 1 + arch/s390/kernel/syscalls/syscall.tbl | 1 + arch/sh/kernel/syscalls/syscall.tbl | 1 + arch/sparc/kernel/syscalls/syscall.tbl | 1 + arch/x86/entry/syscalls/syscall_32.tbl | 1 + arch/x86/entry/syscalls/syscall_64.tbl | 1 + arch/xtensa/kernel/syscalls/syscall.tbl | 1 + include/linux/fs.h | 3 + include/linux/syscalls.h | 4 + include/uapi/asm-generic/unistd.h | 5 +- include/uapi/linux/mman.h | 9 ++ init/Kconfig | 10 ++ kernel/sys_ni.c | 1 + mm/filemap.c | 154 ++++++++++++++++++++ 20 files changed, 198 insertions(+), 1 deletion(-)arch/alpha/kernel/syscalls/syscall.tbl +1
-0@@ -490,3 +490,4 @@ 558 common process_mrelease sys_process_mrelease 559 common futex_waitv sys_futex_waitv 560 common set_mempolicy_home_node sys_ni_syscall +561 common cachestat sys_cachestatarch/arm/tools/syscall.tbl +1
-0@@ -464,3 +464,4 @@ 448 common process_mrelease sys_process_mrelease 449 common futex_waitv sys_futex_waitv 450 common set_mempolicy_home_node sys_set_mempolicy_home_node +451 common cachestat sys_cachestatarch/ia64/kernel/syscalls/syscall.tbl +1
-0@@ -371,3 +371,4 @@ 448 common process_mrelease sys_process_mrelease 449 common futex_waitv sys_futex_waitv 450 common set_mempolicy_home_node sys_set_mempolicy_home_node +451 common cachestat sys_cachestatarch/m68k/kernel/syscalls/syscall.tbl +1
-0@@ -450,3 +450,4 @@ 448 common process_mrelease sys_process_mrelease 449 common futex_waitv sys_futex_waitv 450 common set_mempolicy_home_node sys_set_mempolicy_home_node +451 common cachestat sys_cachestatarch/microblaze/kernel/syscalls/syscall.tbl +1
-0@@ -456,3 +456,4 @@ 448 common process_mrelease sys_process_mrelease 449 common futex_waitv sys_futex_waitv 450 common set_mempolicy_home_node sys_set_mempolicy_home_node +451 common cachestat sys_cachestatarch/parisc/kernel/syscalls/syscall.tbl +1
-0@@ -448,3 +448,4 @@ 448 common process_mrelease sys_process_mrelease 449 common futex_waitv sys_futex_waitv 450 common set_mempolicy_home_node sys_set_mempolicy_home_node +451 common cachestat sys_cachestatarch/powerpc/kernel/syscalls/syscall.tbl +1
-0@@ -530,3 +530,4 @@ 448 common process_mrelease sys_process_mrelease 449 common futex_waitv sys_futex_waitv 450 nospu set_mempolicy_home_node sys_set_mempolicy_home_node +451 common cachestat sys_cachestatarch/s390/kernel/syscalls/syscall.tbl +1
-0@@ -453,3 +453,4 @@ 448 common process_mrelease sys_process_mrelease sys_process_mrelease 449 common futex_waitv sys_futex_waitv sys_futex_waitv 450 common set_mempolicy_home_node sys_set_mempolicy_home_node sys_set_mempolicy_home_node +451 common cachestat sys_cachestat sys_cachestatarch/sh/kernel/syscalls/syscall.tbl +1
-0@@ -453,3 +453,4 @@ 448 common process_mrelease sys_process_mrelease 449 common futex_waitv sys_futex_waitv 450 common set_mempolicy_home_node sys_set_mempolicy_home_node +451 common cachestat sys_cachestatarch/sparc/kernel/syscalls/syscall.tbl +1
-0@@ -496,3 +496,4 @@ 448 common process_mrelease sys_process_mrelease 449 common futex_waitv sys_futex_waitv 450 common set_mempolicy_home_node sys_set_mempolicy_home_node +451 common cachestat sys_cachestatarch/x86/entry/syscalls/syscall_32.tbl +1
-0@@ -455,3 +455,4 @@ 448 i386 process_mrelease sys_process_mrelease 449 i386 futex_waitv sys_futex_waitv 450 i386 set_mempolicy_home_node sys_set_mempolicy_home_node +451 i386 cachestat sys_cachestatarch/x86/entry/syscalls/syscall_64.tbl +1
-0@@ -372,6 +372,7 @@ 448 common process_mrelease sys_process_mrelease 449 common futex_waitv sys_futex_waitv 450 common set_mempolicy_home_node sys_set_mempolicy_home_node +451 common cachestat sys_cachestat # # Due to a historical design error, certain syscalls are numbered differentlyarch/xtensa/kernel/syscalls/syscall.tbl +1
-0@@ -421,3 +421,4 @@ 448 common process_mrelease sys_process_mrelease 449 common futex_waitv sys_futex_waitv 450 common set_mempolicy_home_node sys_set_mempolicy_home_node +451 common cachestat sys_cachestatinclude/linux/fs.h +3
-0@@ -75,6 +75,7 @@ struct fs_context;struct fs_parameter_spec ;struct fileattr ;struct iomap_ops ; +struct cachestat ;extern void __init inode_init (void );extern void __init inode_init_early (void );@@ -830,6 +831,8 @@ void filemap_invalidate_lock_two(struct address_space *mapping1, struct address_space *mapping2);void filemap_invalidate_unlock_two (struct address_space *mapping1,struct address_space *mapping2); +void filemap_cachestat (struct address_space *mapping,pgoff_t first_index, +pgoff_t last_index,struct cachestat *cs);/* include/linux/syscalls.h +4
-0@@ -72,6 +72,7 @@ struct open_how;struct mount_attr ;struct landlock_ruleset_attr ;enum landlock_rule_type ; +struct cachestat ;#include <linux/types.h> #include <linux/aio_abi.h> @@ -1056,6 +1057,9 @@ asmlinkage long sys_memfd_secret(unsigned int flags);asmlinkage long sys_set_mempolicy_home_node (unsigned long start,unsigned long len,unsigned long home_node,unsigned long flags); +asmlinkage long sys_cachestat (unsigned int fd,off_t off,size_t len, +unsigned int cstat_version,struct cachestat __user *cstat, +unsigned int flags);/* * Architecture-specific system calls include/uapi/asm-generic/unistd.h +4
-1@@ -886,8 +886,11 @@ __SYSCALL(__NR_futex_waitv, sys_futex_waitv)#define __NR_set_mempolicy_home_node 450 __SYSCALL(__NR_set_mempolicy_home_node, sys_set_mempolicy_home_node) +#define __NR_cachestat 451 +__SYSCALL(__NR_cachestat, sys_cachestat) +#undef __NR_syscalls -#define __NR_syscalls 451 +#define __NR_syscalls 452/* * 32 bit systems traditionally used different include/uapi/linux/mman.h +9
-0@@ -4,6 +4,7 @@#include <asm/mman.h> #include <asm-generic/hugetlb_encode.h> +#include <linux/types.h> #define MREMAP_MAYMOVE 1#define MREMAP_FIXED 2@@ -41,4 +42,12 @@#define MAP_HUGE_2GB HUGETLB_FLAG_ENCODE_2GB#define MAP_HUGE_16GB HUGETLB_FLAG_ENCODE_16GB +struct cachestat { +__u64 nr_cache; +__u64 nr_dirty; +__u64 nr_writeback; +__u64 nr_evicted; +__u64 nr_recently_evicted; +}; +#endif /* _UAPI_LINUX_MMAN_H */ init/Kconfig +10
-0@@ -1798,6 +1798,16 @@ config RSEQ If unsure, say Y. +config CACHESTAT_SYSCALL + bool "Enable cachestat() system call" if EXPERT + default y + help + Enable the cachestat system call, which queries the page cache + statistics of a file (number of cached pages, dirty pages, + pages marked for writeback, (recently) evicted pages). + + If unsure say Y here. + config DEBUG_RSEQ default n bool "Enabled debugging of rseq() system call" if EXPERTkernel/sys_ni.c +1
-0@@ -299,6 +299,7 @@ COND_SYSCALL(set_mempolicy);COND_SYSCALL (migrate_pages);COND_SYSCALL (move_pages);COND_SYSCALL (set_mempolicy_home_node); +COND_SYSCALL (cachestat);COND_SYSCALL (perf_event_open);COND_SYSCALL (accept4);mm/filemap.c +154
-0@@ -22,6 +22,7 @@#include <linux/mm.h> #include <linux/swap.h> #include <linux/swapops.h> +#include <linux/syscalls.h> #include <linux/mman.h> #include <linux/pagemap.h> #include <linux/file.h> @@ -55,6 +56,13 @@#include <linux/buffer_head.h> /* for try_to_free_buffers */ #include <asm/mman.h> +#include <uapi/linux/mman.h> + +#include "swap.h" + +#ifdef CONFIG_CACHESTAT_SYSCALL +#define LATEST_CACHESTAT_VERSION 1 +#endif /* * Shared mappings implemented 30.11.1994. It's not fully working yet, @@ -3949,3 +3957,149 @@ bool filemap_release_folio(struct folio *folio, gfp_t gfp)return try_to_free_buffers (folio); }EXPORT_SYMBOL (filemap_release_folio); + +/** +* filemap_cachestat() - compute the page cache statistics of a mapping +* @mapping: The mapping to compute the statistics for. +* @first_index: The starting page cache index. +* @last_index: The final page index (inclusive). +* @cs: the cachestat struct to write the result to. +* +* This will query the page cache statistics of a mapping in the +* page range of [first_index, last_index] (inclusive). The statistics +* queried include: number of dirty pages, number of pages marked for +* writeback, and the number of (recently) evicted pages. +*/ +void filemap_cachestat (struct address_space *mapping,pgoff_t first_index, +pgoff_t last_index,struct cachestat *cs) +{ +XA_STATE (xas, &mapping->i_pages, first_index); +struct folio *folio; + +rcu_read_lock (); +xas_for_each (&xas, folio, last_index) { +unsigned long nr_pages; +pgoff_t folio_first_index, folio_last_index; + +if (xas_retry (&xas, folio)) +continue ; + + nr_pages =folio_nr_pages (folio); + folio_first_index =folio_pgoff (folio); + folio_last_index = folio_first_index + nr_pages -1 ; + +/* Folios might straddle the range boundaries, only count covered subpages */ +if (folio_first_index < first_index) + nr_pages -= first_index - folio_first_index; + +if (folio_last_index > last_index) + nr_pages -= folio_last_index - last_index; + +if (xa_is_value (folio)) { +/* page is evicted */ +void *shadow = (void *)folio; +bool workingset;/* not used */ + + cs->nr_evicted += nr_pages; + +#ifdef CONFIG_SWAP /* implies CONFIG_MMU */ +if (shmem_mapping (mapping)) { +/* shmem file - in swap cache */ +swp_entry_t swp =radix_to_swp_entry (folio); + + shadow =get_shadow_from_swap_cache (swp); + } +#endif +if (workingset_test_recent (shadow, true, &workingset)) + cs->nr_recently_evicted += nr_pages; + + goto resched; + } + +/* page is in cache */ + cs->nr_cache += nr_pages; + +if (folio_test_dirty (folio)) + cs->nr_dirty += nr_pages; + +if (folio_test_writeback (folio)) + cs->nr_writeback += nr_pages; + +resched: +if (need_resched ()) { +xas_pause (&xas); +cond_resched_rcu (); + } + } +rcu_read_unlock (); +} +EXPORT_SYMBOL (filemap_cachestat); + +#ifdef CONFIG_CACHESTAT_SYSCALL +/* +* The cachestat(5) system call. +* +* cachestat() returns the page cache statistics of a file in the +* bytes range specified by `off` and `len`: number of cached pages, +* number of dirty pages, number of pages marked for writeback, +* number of evicted pages, and number of recently evicted pages. +* +* An evicted page is a page that is previously in the page cache +* but has been evicted since. A page is recently evicted if its last +* eviction was recent enough that its reentry to the cache would +* indicate that it is actively being used by the system, and that +* there is memory pressure on the system. +* +* `off` and `len` must be non-negative integers. If `len` > 0, +* the queried range is [`off`, `off` + `len`]. If `len` == 0, +* we will query in the range from `off` to the end of the file. +* +* `cstat_version` is an unsigned integer indicating the specific version +* of the cachestat struct. It must be at least 1, and does not exceed the +* latest version number (which is currently 1). For now, user should +* just pass 1. +* +* The `flags` argument is unused for now, but is included for future +* extensibility. User should pass 0 (i.e no flag specified). +* +* Because the status of a page can change after cachestat() checks it +* but before it returns to the application, the returned values may +* contain stale information. +* +* return values: +* zero - success +* -EFAULT - cstat points to an illegal address +* -EINVAL - invalid arguments +* -EBADF - invalid file descriptor +*/ +SYSCALL_DEFINE6(cachestat, unsigned int, fd, off_t, off, size_t, len, + unsigned int, cstat_version, struct cachestat __user *, cstat, + unsigned int, flags) +{ +struct fd f =fdget (fd); +struct address_space *mapping; +struct cachestat cs; +pgoff_t first_index = off >>PAGE_SHIFT ; +pgoff_t last_index = + len ==0 ?ULONG_MAX : (off + len -1 ) >>PAGE_SHIFT ; + +if (!f.file) +return -EBADF ; + +if (off <0 || flags !=0 || cstat_version <1 || + cstat_version >LATEST_CACHESTAT_VERSION ) { +fdput (f); +return -EINVAL ; + } + +memset (&cs,0 ,sizeof (struct cachestat )); + mapping = f.file->f_mapping; +filemap_cachestat (mapping, first_index, last_index, &cs); +fdput (f); + +if (copy_to_user (cstat, &cs,sizeof (struct cachestat ))) +return -EFAULT ; + +return 0 ; +} +#endif /* CONFIG_CACHESTAT_SYSCALL */ Signature · 1 line
2.30.2
Nhat Pham
maintainerRe: [PATCH v6 2/3] cachestat: implement cachestat syscall
Quoted 61 lines · Brian Foster
Quoted 45 lines · Nhat Pham
There is currently no good way to query the page cache state of large file sets and directory trees. There is
mincore(), but it scales poorly:
the kernel writes out a lot of bitmap data that userspace has to aggregate, when the user really doesn not care about per-page information in that case. The user also needs to mmap and unmap each file as it goes along, which can be quite slow as well.This patch implements a new syscall that queries cache state of a file and summarizes the number of cached pages, number of dirty pages, number of pages marked for writeback, number of (recently) evicted pages, etc. in a given range.
NAME
cachestat - query the page cache statistics of a file.SYNOPSIS
#include <sys/mman.h> struct cachestat { __u64 nr_cache; __u64 nr_dirty; __u64 nr_writeback; __u64 nr_evicted; __u64 nr_recently_evicted; }; int cachestat(unsigned int fd, off_t off, size_t len, unsigned int cstat_version, struct cachestat *cstat, unsigned int flags);DESCRIPTION
cachestat()queries the number of cached pages, number of dirty pages, number of pages marked for writeback, number of evicted pages, number of recently evicted pages, in the bytes range given byoffandlen.An evicted page is a page that is previously in the page cache but has been evicted since. A page is recently evicted if its last eviction was recent enough that its reentry to the cache would indicate that it is actively being used by the system, and that there is memory pressure on the system.
These values are returned in a cachestat struct, whose address is given by the
cstatargument.Quoted 9 lines
The
offandlenarguments must be non-negative integers. Iflen> 0, the queried range is [off,off+len]. Iflen== 0, we will query in the range fromoffto the end of the file.
cstat_versionis an unsigned integer indicating the specific version of the cachestat struct. It must be at least 1, and does not exceed the latest version number (which is currently 1). For now, user should just pass 1.Still not sure about this vs. just requiring the structure size to match (and maybe padding it), but perhaps linux-api can comment on the best way to future proof.
Quoted 5 lines
I'm not the person to ack a syscall regardless, but since I've had various previous comments on this patch and no further issues stand out to me:
Reviewed-by Brian Foster <[email protected]>
Thanks for the review and feedback, Brian!
Quoted 457 lines
Quoted 456 lines
The
flagsargument is unused for now, but is included for future extensibility. User should pass 0 (i.e no flag specified).RETURN VALUE
On success, cachestat returns 0. On error, -1 is returned, and errno is set to indicate the error.ERRORS
EFAULT cstat points to an invalid address.EINVAL invalid
cstat_versionorflagsEBADF invalid file descriptor.
Signed-off-by Nhat Pham <[email protected]>--- arch/alpha/kernel/syscalls/syscall.tbl | 1 + arch/arm/tools/syscall.tbl | 1 + arch/ia64/kernel/syscalls/syscall.tbl | 1 + arch/m68k/kernel/syscalls/syscall.tbl | 1 + arch/microblaze/kernel/syscalls/syscall.tbl | 1 + arch/parisc/kernel/syscalls/syscall.tbl | 1 + arch/powerpc/kernel/syscalls/syscall.tbl | 1 + arch/s390/kernel/syscalls/syscall.tbl | 1 + arch/sh/kernel/syscalls/syscall.tbl | 1 + arch/sparc/kernel/syscalls/syscall.tbl | 1 + arch/x86/entry/syscalls/syscall_32.tbl | 1 + arch/x86/entry/syscalls/syscall_64.tbl | 1 + arch/xtensa/kernel/syscalls/syscall.tbl | 1 + include/linux/fs.h | 3 + include/linux/syscalls.h | 4 + include/uapi/asm-generic/unistd.h | 5 +- include/uapi/linux/mman.h | 9 ++ init/Kconfig | 10 ++ kernel/sys_ni.c | 1 + mm/filemap.c | 154 ++++++++++++++++++++ 20 files changed, 198 insertions(+), 1 deletion(-)arch/alpha/kernel/syscalls/syscall.tbl +1
-0@@ -490,3 +490,4 @@ 558 common process_mrelease sys_process_mrelease 559 common futex_waitv sys_futex_waitv 560 common set_mempolicy_home_node sys_ni_syscall +561 common cachestat sys_cachestatarch/arm/tools/syscall.tbl +1
-0@@ -464,3 +464,4 @@ 448 common process_mrelease sys_process_mrelease 449 common futex_waitv sys_futex_waitv 450 common set_mempolicy_home_node sys_set_mempolicy_home_node +451 common cachestat sys_cachestatarch/ia64/kernel/syscalls/syscall.tbl +1
-0@@ -371,3 +371,4 @@ 448 common process_mrelease sys_process_mrelease 449 common futex_waitv sys_futex_waitv 450 common set_mempolicy_home_node sys_set_mempolicy_home_node +451 common cachestat sys_cachestatarch/m68k/kernel/syscalls/syscall.tbl +1
-0@@ -450,3 +450,4 @@ 448 common process_mrelease sys_process_mrelease 449 common futex_waitv sys_futex_waitv 450 common set_mempolicy_home_node sys_set_mempolicy_home_node +451 common cachestat sys_cachestatarch/microblaze/kernel/syscalls/syscall.tbl +1
-0@@ -456,3 +456,4 @@ 448 common process_mrelease sys_process_mrelease 449 common futex_waitv sys_futex_waitv 450 common set_mempolicy_home_node sys_set_mempolicy_home_node +451 common cachestat sys_cachestatarch/parisc/kernel/syscalls/syscall.tbl +1
-0@@ -448,3 +448,4 @@ 448 common process_mrelease sys_process_mrelease 449 common futex_waitv sys_futex_waitv 450 common set_mempolicy_home_node sys_set_mempolicy_home_node +451 common cachestat sys_cachestatarch/powerpc/kernel/syscalls/syscall.tbl +1
-0@@ -530,3 +530,4 @@ 448 common process_mrelease sys_process_mrelease 449 common futex_waitv sys_futex_waitv 450 nospu set_mempolicy_home_node sys_set_mempolicy_home_node +451 common cachestat sys_cachestatarch/s390/kernel/syscalls/syscall.tbl +1
-0@@ -453,3 +453,4 @@ 448 common process_mrelease sys_process_mrelease sys_process_mrelease 449 common futex_waitv sys_futex_waitv sys_futex_waitv 450 common set_mempolicy_home_node sys_set_mempolicy_home_node sys_set_mempolicy_home_node +451 common cachestat sys_cachestat sys_cachestatarch/sh/kernel/syscalls/syscall.tbl +1
-0@@ -453,3 +453,4 @@ 448 common process_mrelease sys_process_mrelease 449 common futex_waitv sys_futex_waitv 450 common set_mempolicy_home_node sys_set_mempolicy_home_node +451 common cachestat sys_cachestatarch/sparc/kernel/syscalls/syscall.tbl +1
-0@@ -496,3 +496,4 @@ 448 common process_mrelease sys_process_mrelease 449 common futex_waitv sys_futex_waitv 450 common set_mempolicy_home_node sys_set_mempolicy_home_node +451 common cachestat sys_cachestatarch/x86/entry/syscalls/syscall_32.tbl +1
-0@@ -455,3 +455,4 @@ 448 i386 process_mrelease sys_process_mrelease 449 i386 futex_waitv sys_futex_waitv 450 i386 set_mempolicy_home_node sys_set_mempolicy_home_node +451 i386 cachestat sys_cachestatarch/x86/entry/syscalls/syscall_64.tbl +1
-0@@ -372,6 +372,7 @@ 448 common process_mrelease sys_process_mrelease 449 common futex_waitv sys_futex_waitv 450 common set_mempolicy_home_node sys_set_mempolicy_home_node +451 common cachestat sys_cachestat # # Due to a historical design error, certain syscalls are numbered differentlyarch/xtensa/kernel/syscalls/syscall.tbl +1
-0@@ -421,3 +421,4 @@ 448 common process_mrelease sys_process_mrelease 449 common futex_waitv sys_futex_waitv 450 common set_mempolicy_home_node sys_set_mempolicy_home_node +451 common cachestat sys_cachestatinclude/linux/fs.h +3
-0@@ -75,6 +75,7 @@ struct fs_context;struct fs_parameter_spec ;struct fileattr ;struct iomap_ops ; +struct cachestat ;extern void __init inode_init (void );extern void __init inode_init_early (void );@@ -830,6 +831,8 @@ void filemap_invalidate_lock_two(struct address_space *mapping1, struct address_space *mapping2);void filemap_invalidate_unlock_two (struct address_space *mapping1,struct address_space *mapping2); +void filemap_cachestat (struct address_space *mapping,pgoff_t first_index, +pgoff_t last_index,struct cachestat *cs);/* include/linux/syscalls.h +4
-0@@ -72,6 +72,7 @@ struct open_how;struct mount_attr ;struct landlock_ruleset_attr ;enum landlock_rule_type ; +struct cachestat ;#include <linux/types.h> #include <linux/aio_abi.h> @@ -1056,6 +1057,9 @@ asmlinkage long sys_memfd_secret(unsigned int flags);asmlinkage long sys_set_mempolicy_home_node (unsigned long start,unsigned long len,unsigned long home_node,unsigned long flags); +asmlinkage long sys_cachestat (unsigned int fd,off_t off,size_t len, +unsigned int cstat_version,struct cachestat __user *cstat, +unsigned int flags);/* * Architecture-specific system calls include/uapi/asm-generic/unistd.h +4
-1@@ -886,8 +886,11 @@ __SYSCALL(__NR_futex_waitv, sys_futex_waitv)#define __NR_set_mempolicy_home_node 450 __SYSCALL(__NR_set_mempolicy_home_node, sys_set_mempolicy_home_node) +#define __NR_cachestat 451 +__SYSCALL(__NR_cachestat, sys_cachestat) +#undef __NR_syscalls -#define __NR_syscalls 451 +#define __NR_syscalls 452/* * 32 bit systems traditionally used different include/uapi/linux/mman.h +9
-0@@ -4,6 +4,7 @@#include <asm/mman.h> #include <asm-generic/hugetlb_encode.h> +#include <linux/types.h> #define MREMAP_MAYMOVE 1#define MREMAP_FIXED 2@@ -41,4 +42,12 @@#define MAP_HUGE_2GB HUGETLB_FLAG_ENCODE_2GB#define MAP_HUGE_16GB HUGETLB_FLAG_ENCODE_16GB +struct cachestat { +__u64 nr_cache; +__u64 nr_dirty; +__u64 nr_writeback; +__u64 nr_evicted; +__u64 nr_recently_evicted; +}; +#endif /* _UAPI_LINUX_MMAN_H */ init/Kconfig +10
-0@@ -1798,6 +1798,16 @@ config RSEQ If unsure, say Y. +config CACHESTAT_SYSCALL + bool "Enable cachestat() system call" if EXPERT + default y + help + Enable the cachestat system call, which queries the page cache + statistics of a file (number of cached pages, dirty pages, + pages marked for writeback, (recently) evicted pages). + + If unsure say Y here. + config DEBUG_RSEQ default n bool "Enabled debugging of rseq() system call" if EXPERTkernel/sys_ni.c +1
-0@@ -299,6 +299,7 @@ COND_SYSCALL(set_mempolicy);COND_SYSCALL (migrate_pages);COND_SYSCALL (move_pages);COND_SYSCALL (set_mempolicy_home_node); +COND_SYSCALL (cachestat);COND_SYSCALL (perf_event_open);COND_SYSCALL (accept4);mm/filemap.c +154
-1@@ -22,6 +22,7 @@#include <linux/mm.h> #include <linux/swap.h> #include <linux/swapops.h> +#include <linux/syscalls.h> #include <linux/mman.h> #include <linux/pagemap.h> #include <linux/file.h> @@ -55,6 +56,13 @@#include <linux/buffer_head.h> /* for try_to_free_buffers */ #include <asm/mman.h> +#include <uapi/linux/mman.h> + +#include "swap.h" + +#ifdef CONFIG_CACHESTAT_SYSCALL +#define LATEST_CACHESTAT_VERSION 1 +#endif /* * Shared mappings implemented 30.11.1994. It's not fully working yet, @@ -3949,3 +3957,149 @@ bool filemap_release_folio(struct folio *folio, gfp_t gfp)return try_to_free_buffers (folio); }EXPORT_SYMBOL (filemap_release_folio); + +/** +* filemap_cachestat() - compute the page cache statistics of a mapping +* @mapping: The mapping to compute the statistics for. +* @first_index: The starting page cache index. +* @last_index: The final page index (inclusive). +* @cs: the cachestat struct to write the result to. +* +* This will query the page cache statistics of a mapping in the +* page range of [first_index, last_index] (inclusive). The statistics +* queried include: number of dirty pages, number of pages marked for +* writeback, and the number of (recently) evicted pages. +*/ +void filemap_cachestat (struct address_space *mapping,pgoff_t first_index, +pgoff_t last_index,struct cachestat *cs) +{ +XA_STATE (xas, &mapping->i_pages, first_index); +struct folio *folio; + +rcu_read_lock (); +xas_for_each (&xas, folio, last_index) { +unsigned long nr_pages; +pgoff_t folio_first_index, folio_last_index; + +if (xas_retry (&xas, folio)) +continue ; + + nr_pages =folio_nr_pages (folio); + folio_first_index =folio_pgoff (folio); + folio_last_index = folio_first_index + nr_pages -1 ; + +/* Folios might straddle the range boundaries, only count covered subpages */ +if (folio_first_index < first_index) + nr_pages -= first_index - folio_first_index; + +if (folio_last_index > last_index) + nr_pages -= folio_last_index - last_index; + +if (xa_is_value (folio)) { +/* page is evicted */ +void *shadow = (void *)folio; +bool workingset;/* not used */ + + cs->nr_evicted += nr_pages; + +#ifdef CONFIG_SWAP /* implies CONFIG_MMU */ +if (shmem_mapping (mapping)) { +/* shmem file - in swap cache */ +swp_entry_t swp =radix_to_swp_entry (folio); + + shadow =get_shadow_from_swap_cache (swp); + } +#endif +if (workingset_test_recent (shadow, true, &workingset)) + cs->nr_recently_evicted += nr_pages; + + goto resched; + } + +/* page is in cache */ + cs->nr_cache += nr_pages; + +if (folio_test_dirty (folio)) + cs->nr_dirty += nr_pages; + +if (folio_test_writeback (folio)) + cs->nr_writeback += nr_pages; + +resched: +if (need_resched ()) { +xas_pause (&xas); +cond_resched_rcu (); + } + } +rcu_read_unlock (); +} +EXPORT_SYMBOL (filemap_cachestat); + +#ifdef CONFIG_CACHESTAT_SYSCALL +/* +* The cachestat(5) system call. +* +* cachestat() returns the page cache statistics of a file in the +* bytes range specified by `off` and `len`: number of cached pages, +* number of dirty pages, number of pages marked for writeback, +* number of evicted pages, and number of recently evicted pages. +* +* An evicted page is a page that is previously in the page cache +* but has been evicted since. A page is recently evicted if its last +* eviction was recent enough that its reentry to the cache would +* indicate that it is actively being used by the system, and that +* there is memory pressure on the system. +* +* `off` and `len` must be non-negative integers. If `len` > 0, +* the queried range is [`off`, `off` + `len`]. If `len` == 0, +* we will query in the range from `off` to the end of the file. +* +* `cstat_version` is an unsigned integer indicating the specific version +* of the cachestat struct. It must be at least 1, and does not exceed the +* latest version number (which is currently 1). For now, user should +* just pass 1. +* +* The `flags` argument is unused for now, but is included for future +* extensibility. User should pass 0 (i.e no flag specified). +* +* Because the status of a page can change after cachestat() checks it +* but before it returns to the application, the returned values may +* contain stale information. +* +* return values: +* zero - success +* -EFAULT - cstat points to an illegal address +* -EINVAL - invalid arguments +* -EBADF - invalid file descriptor +*/ +SYSCALL_DEFINE6(cachestat, unsigned int, fd, off_t, off, size_t, len, + unsigned int, cstat_version, struct cachestat __user *, cstat, + unsigned int, flags) +{ +struct fd f =fdget (fd); +struct address_space *mapping; +struct cachestat cs; +pgoff_t first_index = off >>PAGE_SHIFT ; +pgoff_t last_index = + len ==0 ?ULONG_MAX : (off + len -1 ) >>PAGE_SHIFT ; + +if (!f.file) +return -EBADF ; + +if (off <0 || flags !=0 || cstat_version <1 || + cstat_version >LATEST_CACHESTAT_VERSION ) { +fdput (f); +return -EINVAL ; + } + +memset (&cs,0 ,sizeof (struct cachestat )); + mapping = f.file->f_mapping; +filemap_cachestat (mapping, first_index, last_index, &cs); +fdput (f); + +if (copy_to_user (cstat, &cs,sizeof (struct cachestat ))) +return -EFAULT ; + +return 0 ; +} +#endif /* CONFIG_CACHESTAT_SYSCALL */ --2.30.2
Nhat Pham
maintainerPATCH v6 3/3 selftests: Add selftests for cachestat
In reply to Nhat PhamTest cachestat on a newly created file, /dev/ files, and /proc/ files. Also test on a shmem file (which can also be tested with huge pages since tmpfs supports huge pages).
--- MAINTAINERS | 7 + tools/testing/selftests/Makefile | 1 + tools/testing/selftests/cachestat/.gitignore | 2 + tools/testing/selftests/cachestat/Makefile | 8 + .../selftests/cachestat/test_cachestat.c | 260 ++++++++++++++++++ 5 files changed, 278 insertions(+) create mode 100644 tools/testing/selftests/cachestat/.gitignore create mode 100644 tools/testing/selftests/cachestat/Makefile create mode 100644 tools/testing/selftests/cachestat/test_cachestat.c
MAINTAINERS +7 -0
@@ -4552,6 +4552,13 @@ S: Supported F: Documentation/filesystems/caching/cachefiles.rst F: fs/cachefiles/ +CACHESTAT: PAGE CACHE STATS FOR A FILE +M: Nhat Pham <[email protected]> +M: Johannes Weiner <[email protected]> +L: [email protected] +S: Maintained +F: tools/testing/selftests/cachestat/test_cachestat.c + CADENCE MIPI-CSI2 BRIDGES M: Maxime Ripard <[email protected]> L: [email protected]
tools/testing/selftests/Makefile +1 -0
@@ -4,6 +4,7 @@ TARGETS += amd-pstate TARGETS += arm64 TARGETS += bpf TARGETS += breakpoints +TARGETS += cachestat TARGETS += capabilities TARGETS += cgroup TARGETS += clone3
tools/testing/selftests/cachestat/.gitignore new file +2 -0
@@ -0,0 +1,2 @@ +# SPDX-License-Identifier: GPL-2.0-only +test_cachestat
tools/testing/selftests/cachestat/Makefile new file +8 -0
@@ -0,0 +1,8 @@ +# SPDX-License-Identifier: GPL-2.0 +TEST_GEN_PROGS := test_cachestat + +CFLAGS += $(KHDR_INCLUDES) +CFLAGS += -Wall +CFLAGS += -lrt + +include ../lib.mk
tools/testing/selftests/cachestat/test_cachestat.c new file +260 -0
260 lines not on this page · raw message
Signature · 1 line
2.30.2