Message ID | 20241017142504.1170208-2-wangkefeng.wang@huawei.com (mailing list archive) |
---|---|
State | New |
Headers | show |
Series | mm: shmem: convert to use folio_zero_range() | expand |
On Thu, Oct 17, 2024 at 10:25:04PM +0800, Kefeng Wang wrote: > Directly use folio_zero_range() to cleanup code. Are you sure there's no performance regression introduced by this? clear_highpage() is often optimised in ways that we can't optimise for a plain memset(). On the other hand, if the folio is large, maybe a modern CPU will be able to do better than clear-one-page-at-a-time. IOW, what performance testing have you done with this patch? > if (sgp != SGP_WRITE && !folio_test_uptodate(folio)) { > - long i, n = folio_nr_pages(folio); > - > - for (i = 0; i < n; i++) > - clear_highpage(folio_page(folio, i)); > - flush_dcache_folio(folio); > + folio_zero_range(folio, 0, folio_size(folio)); > folio_mark_uptodate(folio); > }
On 2024/10/17 23:09, Matthew Wilcox wrote: > On Thu, Oct 17, 2024 at 10:25:04PM +0800, Kefeng Wang wrote: >> Directly use folio_zero_range() to cleanup code. > > Are you sure there's no performance regression introduced by this? > clear_highpage() is often optimised in ways that we can't optimise for > a plain memset(). On the other hand, if the folio is large, maybe a > modern CPU will be able to do better than clear-one-page-at-a-time. > Right, I missing this, clear_page might be better than memset, I change this one when look at the shmem_writepage(), which already convert to use folio_zero_range() from clear_highpage(), also I grep folio_zero_range(), there are some other to use folio_zero_range(). fs/bcachefs/fs-io-buffered.c: folio_zero_range(folio, 0, folio_size(folio)); fs/bcachefs/fs-io-buffered.c: folio_zero_range(f, 0, folio_size(f)); fs/bcachefs/fs-io-buffered.c: folio_zero_range(f, 0, folio_size(f)); fs/libfs.c: folio_zero_range(folio, 0, folio_size(folio)); fs/ntfs3/frecord.c: folio_zero_range(folio, 0, folio_size(folio)); mm/page_io.c: folio_zero_range(folio, 0, folio_size(folio)); mm/shmem.c: folio_zero_range(folio, 0, folio_size(folio)); > IOW, what performance testing have you done with this patch? No performance test before, but I write a testcase, 1) allocate N large folios (folio_alloc(PMD_ORDER)) 2) then calculate the diff(us) when clear all N folios clear_highpage/folio_zero_range/folio_zero_user 3) release N folios the result(run 5 times) shown below on my machine, N=1, clear_highpage folio_zero_range folio_zero_user 1 69 74 177 2 57 62 168 3 54 58 234 4 54 58 157 5 56 62 148 avg 58 62.8 176.8 N=100 clear_highpage folio_zero_range folio_zero_user 1 11015 11309 32833 2 10385 11110 49751 3 10369 11056 33095 4 10332 11017 33106 5 10483 11000 49032 avg 10516.8 11098.4 39563.4 N=512 clear_highpage folio_zero_range folio_zero_user 1 55560 60055 156876 2 55485 60024 157132 3 55474 60129 156658 4 55555 59867 157259 5 55528 59932 157108 avg 55520.4 60001.4 157006.6 folio_zero_user with many cond_resched(), so time fluctuates a lot, clear_highpage is better folio_zero_range as you said. Maybe add a new helper to convert all folio_zero_range(folio, 0, folio_size(folio)) to use clear_highpage + flush_dcache_folio? > >> if (sgp != SGP_WRITE && !folio_test_uptodate(folio)) { >> - long i, n = folio_nr_pages(folio); >> - >> - for (i = 0; i < n; i++) >> - clear_highpage(folio_page(folio, i)); >> - flush_dcache_folio(folio); >> + folio_zero_range(folio, 0, folio_size(folio)); >> folio_mark_uptodate(folio); >> } >
On Fri, Oct 18, 2024 at 6:20 PM Kefeng Wang <wangkefeng.wang@huawei.com> wrote: > > > > On 2024/10/17 23:09, Matthew Wilcox wrote: > > On Thu, Oct 17, 2024 at 10:25:04PM +0800, Kefeng Wang wrote: > >> Directly use folio_zero_range() to cleanup code. > > > > Are you sure there's no performance regression introduced by this? > > clear_highpage() is often optimised in ways that we can't optimise for > > a plain memset(). On the other hand, if the folio is large, maybe a > > modern CPU will be able to do better than clear-one-page-at-a-time. > > > > Right, I missing this, clear_page might be better than memset, I change > this one when look at the shmem_writepage(), which already convert to > use folio_zero_range() from clear_highpage(), also I grep > folio_zero_range(), there are some other to use folio_zero_range(). > > fs/bcachefs/fs-io-buffered.c: folio_zero_range(folio, 0, > folio_size(folio)); > fs/bcachefs/fs-io-buffered.c: folio_zero_range(f, 0, folio_size(f)); > fs/bcachefs/fs-io-buffered.c: folio_zero_range(f, 0, folio_size(f)); > fs/libfs.c: folio_zero_range(folio, 0, folio_size(folio)); > fs/ntfs3/frecord.c: folio_zero_range(folio, 0, folio_size(folio)); > mm/page_io.c: folio_zero_range(folio, 0, folio_size(folio)); > mm/shmem.c: folio_zero_range(folio, 0, folio_size(folio)); > > > > IOW, what performance testing have you done with this patch? > > No performance test before, but I write a testcase, > > 1) allocate N large folios (folio_alloc(PMD_ORDER)) > 2) then calculate the diff(us) when clear all N folios > clear_highpage/folio_zero_range/folio_zero_user > 3) release N folios > > the result(run 5 times) shown below on my machine, > > N=1, > clear_highpage folio_zero_range folio_zero_user > 1 69 74 177 > 2 57 62 168 > 3 54 58 234 > 4 54 58 157 > 5 56 62 148 > avg 58 62.8 176.8 > > > N=100 > clear_highpage folio_zero_range folio_zero_user > 1 11015 11309 32833 > 2 10385 11110 49751 > 3 10369 11056 33095 > 4 10332 11017 33106 > 5 10483 11000 49032 > avg 10516.8 11098.4 39563.4 > > N=512 > clear_highpage folio_zero_range folio_zero_user > 1 55560 60055 156876 > 2 55485 60024 157132 > 3 55474 60129 156658 > 4 55555 59867 157259 > 5 55528 59932 157108 > avg 55520.4 60001.4 157006.6 > > > > folio_zero_user with many cond_resched(), so time fluctuates a lot, > clear_highpage is better folio_zero_range as you said. > > Maybe add a new helper to convert all folio_zero_range(folio, 0, > folio_size(folio)) > to use clear_highpage + flush_dcache_folio? If this also improves performance for other existing callers of folio_zero_range(), then that's a positive outcome. > > > > > >> if (sgp != SGP_WRITE && !folio_test_uptodate(folio)) { > >> - long i, n = folio_nr_pages(folio); > >> - > >> - for (i = 0; i < n; i++) > >> - clear_highpage(folio_page(folio, i)); > >> - flush_dcache_folio(folio); > >> + folio_zero_range(folio, 0, folio_size(folio)); > >> folio_mark_uptodate(folio); > >> } > > > > Thanks Barry
On 2024/10/18 13:23, Barry Song wrote: > On Fri, Oct 18, 2024 at 6:20 PM Kefeng Wang <wangkefeng.wang@huawei.com> wrote: >> >> >> >> On 2024/10/17 23:09, Matthew Wilcox wrote: >>> On Thu, Oct 17, 2024 at 10:25:04PM +0800, Kefeng Wang wrote: >>>> Directly use folio_zero_range() to cleanup code. >>> >>> Are you sure there's no performance regression introduced by this? >>> clear_highpage() is often optimised in ways that we can't optimise for >>> a plain memset(). On the other hand, if the folio is large, maybe a >>> modern CPU will be able to do better than clear-one-page-at-a-time. >>> >> >> Right, I missing this, clear_page might be better than memset, I change >> this one when look at the shmem_writepage(), which already convert to >> use folio_zero_range() from clear_highpage(), also I grep >> folio_zero_range(), there are some other to use folio_zero_range(). >> >> fs/bcachefs/fs-io-buffered.c: folio_zero_range(folio, 0, >> folio_size(folio)); >> fs/bcachefs/fs-io-buffered.c: folio_zero_range(f, 0, folio_size(f)); >> fs/bcachefs/fs-io-buffered.c: folio_zero_range(f, 0, folio_size(f)); >> fs/libfs.c: folio_zero_range(folio, 0, folio_size(folio)); >> fs/ntfs3/frecord.c: folio_zero_range(folio, 0, folio_size(folio)); >> mm/page_io.c: folio_zero_range(folio, 0, folio_size(folio)); >> mm/shmem.c: folio_zero_range(folio, 0, folio_size(folio)); >> >> >>> IOW, what performance testing have you done with this patch? >> >> No performance test before, but I write a testcase, >> >> 1) allocate N large folios (folio_alloc(PMD_ORDER)) >> 2) then calculate the diff(us) when clear all N folios >> clear_highpage/folio_zero_range/folio_zero_user >> 3) release N folios >> >> the result(run 5 times) shown below on my machine, >> >> N=1, >> clear_highpage folio_zero_range folio_zero_user >> 1 69 74 177 >> 2 57 62 168 >> 3 54 58 234 >> 4 54 58 157 >> 5 56 62 148 >> avg 58 62.8 176.8 >> >> >> N=100 >> clear_highpage folio_zero_range folio_zero_user >> 1 11015 11309 32833 >> 2 10385 11110 49751 >> 3 10369 11056 33095 >> 4 10332 11017 33106 >> 5 10483 11000 49032 >> avg 10516.8 11098.4 39563.4 >> >> N=512 >> clear_highpage folio_zero_range folio_zero_user >> 1 55560 60055 156876 >> 2 55485 60024 157132 >> 3 55474 60129 156658 >> 4 55555 59867 157259 >> 5 55528 59932 157108 >> avg 55520.4 60001.4 157006.6 >> >> >> >> folio_zero_user with many cond_resched(), so time fluctuates a lot, >> clear_highpage is better folio_zero_range as you said. >> >> Maybe add a new helper to convert all folio_zero_range(folio, 0, >> folio_size(folio)) >> to use clear_highpage + flush_dcache_folio? > > If this also improves performance for other existing callers of > folio_zero_range(), then that's a positive outcome. rm -f /tmp/test && fallocate -l 20G /tmp/test && fallocate -d -l 20G /tmp/test && time fallocate -l 20G /tmp/test 1)mount always(2M folio) with patch without patch real 0m1.214s 0m1.111s user 0m0.000s 0m0.000s sys 0m1.210s 0m1.109s With this patch, the performance does have regression, folio_zero_range() is bad than clear_highpage + flush_dcache_folio with patch 99.95% 0.00% fallocate [kernel.vmlinux] [k] vfs_fallocate vfs_fallocate - shmem_fallocate 98.54% __pi_clear_page - 1.38% shmem_get_folio_gfp filemap_get_entry without patch 99.89% 0.00% fallocate [kernel.vmlinux] [k] shmem_fallocate shmem_fallocate - shmem_get_folio_gfp 90.12% __memset - 9.42% zero_user_segments.constprop.0 8.16% flush_dcache_page 1.03% flush_dcache_folio 2)mount never (4K folio) real 0m3.159s 0m3.176s user 0m0.000s 0m0.000s sys 0m3.150s 0m3.169s But with this patch, the performance is improved a little, folio_zero_range() is better than clear_highpage + flush_dcache_folio with patch 97.77% 3.37% fallocate [kernel.vmlinux] [k] shmem_fallocate - 94.40% shmem_fallocate - 93.70% shmem_get_folio_gfp 66.60% __memset - 7.43% filemap_get_entry 3.49% xas_load 1.32% zero_user_segments.constprop.0 without patch 97.82% 3.22% fallocate [kernel.vmlinux] [k] shmem_fallocate - 94.61% shmem_fallocate 68.18% __pi_clear_page - 25.60% shmem_get_folio_gfp - 7.64% filemap_get_entry 3.51% xas_load > >> >> >>> >>>> if (sgp != SGP_WRITE && !folio_test_uptodate(folio)) { >>>> - long i, n = folio_nr_pages(folio); >>>> - >>>> - for (i = 0; i < n; i++) >>>> - clear_highpage(folio_page(folio, i)); >>>> - flush_dcache_folio(folio); >>>> + folio_zero_range(folio, 0, folio_size(folio)); >>>> folio_mark_uptodate(folio); >>>> } >>> >> >> > > Thanks > Barry
On 2024/10/18 15:32, Kefeng Wang wrote: > > > On 2024/10/18 13:23, Barry Song wrote: >> On Fri, Oct 18, 2024 at 6:20 PM Kefeng Wang >> <wangkefeng.wang@huawei.com> wrote: >>> >>> >>> >>> On 2024/10/17 23:09, Matthew Wilcox wrote: >>>> On Thu, Oct 17, 2024 at 10:25:04PM +0800, Kefeng Wang wrote: >>>>> Directly use folio_zero_range() to cleanup code. >>>> >>>> Are you sure there's no performance regression introduced by this? >>>> clear_highpage() is often optimised in ways that we can't optimise for >>>> a plain memset(). On the other hand, if the folio is large, maybe a >>>> modern CPU will be able to do better than clear-one-page-at-a-time. >>>> >>> >>> Right, I missing this, clear_page might be better than memset, I change >>> this one when look at the shmem_writepage(), which already convert to >>> use folio_zero_range() from clear_highpage(), also I grep >>> folio_zero_range(), there are some other to use folio_zero_range(). >>> >>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(folio, 0, >>> folio_size(folio)); >>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(f, >>> 0, folio_size(f)); >>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(f, >>> 0, folio_size(f)); >>> fs/libfs.c: folio_zero_range(folio, 0, folio_size(folio)); >>> fs/ntfs3/frecord.c: folio_zero_range(folio, 0, >>> folio_size(folio)); >>> mm/page_io.c: folio_zero_range(folio, 0, folio_size(folio)); >>> mm/shmem.c: folio_zero_range(folio, 0, folio_size(folio)); >>> >>> >>>> IOW, what performance testing have you done with this patch? >>> >>> No performance test before, but I write a testcase, >>> >>> 1) allocate N large folios (folio_alloc(PMD_ORDER)) >>> 2) then calculate the diff(us) when clear all N folios >>> clear_highpage/folio_zero_range/folio_zero_user >>> 3) release N folios >>> >>> the result(run 5 times) shown below on my machine, >>> >>> N=1, >>> clear_highpage folio_zero_range folio_zero_user >>> 1 69 74 177 >>> 2 57 62 168 >>> 3 54 58 234 >>> 4 54 58 157 >>> 5 56 62 148 >>> avg 58 62.8 176.8 >>> >>> >>> N=100 >>> clear_highpage folio_zero_range folio_zero_user >>> 1 11015 11309 32833 >>> 2 10385 11110 49751 >>> 3 10369 11056 33095 >>> 4 10332 11017 33106 >>> 5 10483 11000 49032 >>> avg 10516.8 11098.4 39563.4 >>> >>> N=512 >>> clear_highpage folio_zero_range folio_zero_user >>> 1 55560 60055 156876 >>> 2 55485 60024 157132 >>> 3 55474 60129 156658 >>> 4 55555 59867 157259 >>> 5 55528 59932 157108 >>> avg 55520.4 60001.4 157006.6 >>> >>> >>> >>> folio_zero_user with many cond_resched(), so time fluctuates a lot, >>> clear_highpage is better folio_zero_range as you said. >>> >>> Maybe add a new helper to convert all folio_zero_range(folio, 0, >>> folio_size(folio)) >>> to use clear_highpage + flush_dcache_folio? >> >> If this also improves performance for other existing callers of >> folio_zero_range(), then that's a positive outcome. > > > rm -f /tmp/test && fallocate -l 20G /tmp/test && fallocate -d -l 20G / > tmp/test && time fallocate -l 20G /tmp/test > > 1)mount always(2M folio) > with patch without patch > real 0m1.214s 0m1.111s > user 0m0.000s 0m0.000s > sys 0m1.210s 0m1.109s > > With this patch, the performance does have regression, > folio_zero_range() is bad than clear_highpage + flush_dcache_folio > > with patch Oh, this should without patch since it uses clear_highpage, > > 99.95% 0.00% fallocate [kernel.vmlinux] [k] vfs_fallocate > vfs_fallocate > - shmem_fallocate > 98.54% __pi_clear_page > - 1.38% shmem_get_folio_gfp > filemap_get_entry > and this one is with patch > without patch > 99.89% 0.00% fallocate [kernel.vmlinux] [k] shmem_fallocate > shmem_fallocate > - shmem_get_folio_gfp > 90.12% __memset > - 9.42% zero_user_segments.constprop.0 > 8.16% flush_dcache_page > 1.03% flush_dcache_folio > > > > > 2)mount never (4K folio) > real 0m3.159s 0m3.176s > user 0m0.000s 0m0.000s > sys 0m3.150s 0m3.169s > > But with this patch, the performance is improved a little, > folio_zero_range() is better than clear_highpage + flush_dcache_folio > For 4K, the result is fluctuating, so maybe no different. > with patch > 97.77% 3.37% fallocate [kernel.vmlinux] [k] shmem_fallocate > - 94.40% shmem_fallocate > - 93.70% shmem_get_folio_gfp > 66.60% __memset > - 7.43% filemap_get_entry > 3.49% xas_load > 1.32% zero_user_segments.constprop.0 > > without patch > 97.82% 3.22% fallocate [kernel.vmlinux] [k] shmem_fallocate > - 94.61% shmem_fallocate > 68.18% __pi_clear_page > - 25.60% shmem_get_folio_gfp > - 7.64% filemap_get_entry > 3.51% xas_load >> >>> >>> >>>> >>>>> if (sgp != SGP_WRITE && !folio_test_uptodate(folio)) { >>>>> - long i, n = folio_nr_pages(folio); >>>>> - >>>>> - for (i = 0; i < n; i++) >>>>> - clear_highpage(folio_page(folio, i)); >>>>> - flush_dcache_folio(folio); >>>>> + folio_zero_range(folio, 0, folio_size(folio)); >>>>> folio_mark_uptodate(folio); >>>>> } >>>> >>> >>> >> >> Thanks >> Barry > >
On Fri, Oct 18, 2024 at 8:48 PM Kefeng Wang <wangkefeng.wang@huawei.com> wrote: > > > > On 2024/10/18 15:32, Kefeng Wang wrote: > > > > > > On 2024/10/18 13:23, Barry Song wrote: > >> On Fri, Oct 18, 2024 at 6:20 PM Kefeng Wang > >> <wangkefeng.wang@huawei.com> wrote: > >>> > >>> > >>> > >>> On 2024/10/17 23:09, Matthew Wilcox wrote: > >>>> On Thu, Oct 17, 2024 at 10:25:04PM +0800, Kefeng Wang wrote: > >>>>> Directly use folio_zero_range() to cleanup code. > >>>> > >>>> Are you sure there's no performance regression introduced by this? > >>>> clear_highpage() is often optimised in ways that we can't optimise for > >>>> a plain memset(). On the other hand, if the folio is large, maybe a > >>>> modern CPU will be able to do better than clear-one-page-at-a-time. > >>>> > >>> > >>> Right, I missing this, clear_page might be better than memset, I change > >>> this one when look at the shmem_writepage(), which already convert to > >>> use folio_zero_range() from clear_highpage(), also I grep > >>> folio_zero_range(), there are some other to use folio_zero_range(). > >>> > >>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(folio, 0, > >>> folio_size(folio)); > >>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(f, > >>> 0, folio_size(f)); > >>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(f, > >>> 0, folio_size(f)); > >>> fs/libfs.c: folio_zero_range(folio, 0, folio_size(folio)); > >>> fs/ntfs3/frecord.c: folio_zero_range(folio, 0, > >>> folio_size(folio)); > >>> mm/page_io.c: folio_zero_range(folio, 0, folio_size(folio)); > >>> mm/shmem.c: folio_zero_range(folio, 0, folio_size(folio)); > >>> > >>> > >>>> IOW, what performance testing have you done with this patch? > >>> > >>> No performance test before, but I write a testcase, > >>> > >>> 1) allocate N large folios (folio_alloc(PMD_ORDER)) > >>> 2) then calculate the diff(us) when clear all N folios > >>> clear_highpage/folio_zero_range/folio_zero_user > >>> 3) release N folios > >>> > >>> the result(run 5 times) shown below on my machine, > >>> > >>> N=1, > >>> clear_highpage folio_zero_range folio_zero_user > >>> 1 69 74 177 > >>> 2 57 62 168 > >>> 3 54 58 234 > >>> 4 54 58 157 > >>> 5 56 62 148 > >>> avg 58 62.8 176.8 > >>> > >>> > >>> N=100 > >>> clear_highpage folio_zero_range folio_zero_user > >>> 1 11015 11309 32833 > >>> 2 10385 11110 49751 > >>> 3 10369 11056 33095 > >>> 4 10332 11017 33106 > >>> 5 10483 11000 49032 > >>> avg 10516.8 11098.4 39563.4 > >>> > >>> N=512 > >>> clear_highpage folio_zero_range folio_zero_user > >>> 1 55560 60055 156876 > >>> 2 55485 60024 157132 > >>> 3 55474 60129 156658 > >>> 4 55555 59867 157259 > >>> 5 55528 59932 157108 > >>> avg 55520.4 60001.4 157006.6 > >>> > >>> > >>> > >>> folio_zero_user with many cond_resched(), so time fluctuates a lot, > >>> clear_highpage is better folio_zero_range as you said. > >>> > >>> Maybe add a new helper to convert all folio_zero_range(folio, 0, > >>> folio_size(folio)) > >>> to use clear_highpage + flush_dcache_folio? > >> > >> If this also improves performance for other existing callers of > >> folio_zero_range(), then that's a positive outcome. > > > > > > rm -f /tmp/test && fallocate -l 20G /tmp/test && fallocate -d -l 20G / > > tmp/test && time fallocate -l 20G /tmp/test > > > > 1)mount always(2M folio) > > with patch without patch > > real 0m1.214s 0m1.111s > > user 0m0.000s 0m0.000s > > sys 0m1.210s 0m1.109s > > > > With this patch, the performance does have regression, > > folio_zero_range() is bad than clear_highpage + flush_dcache_folio > > > > with patch > > Oh, this should without patch since it uses clear_highpage, > > > > > 99.95% 0.00% fallocate [kernel.vmlinux] [k] vfs_fallocate > > vfs_fallocate > > - shmem_fallocate > > 98.54% __pi_clear_page > > - 1.38% shmem_get_folio_gfp > > filemap_get_entry > > > and this one is with patch > > without patch > > 99.89% 0.00% fallocate [kernel.vmlinux] [k] shmem_fallocate > > shmem_fallocate > > - shmem_get_folio_gfp > > 90.12% __memset > > - 9.42% zero_user_segments.constprop.0 > > 8.16% flush_dcache_page > > 1.03% flush_dcache_folio > > > > > > > > > > 2)mount never (4K folio) > > real 0m3.159s 0m3.176s > > user 0m0.000s 0m0.000s > > sys 0m3.150s 0m3.169s > > > > But with this patch, the performance is improved a little, > > folio_zero_range() is better than clear_highpage + flush_dcache_folio > > > > For 4K, the result is fluctuating, so maybe no different. hi Kefeng, what's your point? providing a helper like clear_highfolio() or similar? > > > with patch > > 97.77% 3.37% fallocate [kernel.vmlinux] [k] shmem_fallocate > > - 94.40% shmem_fallocate > > - 93.70% shmem_get_folio_gfp > > 66.60% __memset > > - 7.43% filemap_get_entry > > 3.49% xas_load > > 1.32% zero_user_segments.constprop.0 > > > > without patch > > 97.82% 3.22% fallocate [kernel.vmlinux] [k] shmem_fallocate > > - 94.61% shmem_fallocate > > 68.18% __pi_clear_page > > - 25.60% shmem_get_folio_gfp > > - 7.64% filemap_get_entry > > 3.51% xas_load > >> > >>> > >>> > >>>> > >>>>> if (sgp != SGP_WRITE && !folio_test_uptodate(folio)) { > >>>>> - long i, n = folio_nr_pages(folio); > >>>>> - > >>>>> - for (i = 0; i < n; i++) > >>>>> - clear_highpage(folio_page(folio, i)); > >>>>> - flush_dcache_folio(folio); > >>>>> + folio_zero_range(folio, 0, folio_size(folio)); > >>>>> folio_mark_uptodate(folio); > >>>>> } > >>>> > >>> > >>> > >> > >> Thanks > >> Barry > > > > >
On 2024/10/21 12:15, Barry Song wrote: > On Fri, Oct 18, 2024 at 8:48 PM Kefeng Wang <wangkefeng.wang@huawei.com> wrote: >> >> >> >> On 2024/10/18 15:32, Kefeng Wang wrote: >>> >>> >>> On 2024/10/18 13:23, Barry Song wrote: >>>> On Fri, Oct 18, 2024 at 6:20 PM Kefeng Wang >>>> <wangkefeng.wang@huawei.com> wrote: >>>>> >>>>> >>>>> >>>>> On 2024/10/17 23:09, Matthew Wilcox wrote: >>>>>> On Thu, Oct 17, 2024 at 10:25:04PM +0800, Kefeng Wang wrote: >>>>>>> Directly use folio_zero_range() to cleanup code. >>>>>> >>>>>> Are you sure there's no performance regression introduced by this? >>>>>> clear_highpage() is often optimised in ways that we can't optimise for >>>>>> a plain memset(). On the other hand, if the folio is large, maybe a >>>>>> modern CPU will be able to do better than clear-one-page-at-a-time. >>>>>> >>>>> >>>>> Right, I missing this, clear_page might be better than memset, I change >>>>> this one when look at the shmem_writepage(), which already convert to >>>>> use folio_zero_range() from clear_highpage(), also I grep >>>>> folio_zero_range(), there are some other to use folio_zero_range(). >>>>> >>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(folio, 0, >>>>> folio_size(folio)); >>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(f, >>>>> 0, folio_size(f)); >>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(f, >>>>> 0, folio_size(f)); >>>>> fs/libfs.c: folio_zero_range(folio, 0, folio_size(folio)); >>>>> fs/ntfs3/frecord.c: folio_zero_range(folio, 0, >>>>> folio_size(folio)); >>>>> mm/page_io.c: folio_zero_range(folio, 0, folio_size(folio)); >>>>> mm/shmem.c: folio_zero_range(folio, 0, folio_size(folio)); >>>>> >>>>> >>>>>> IOW, what performance testing have you done with this patch? >>>>> >>>>> No performance test before, but I write a testcase, >>>>> >>>>> 1) allocate N large folios (folio_alloc(PMD_ORDER)) >>>>> 2) then calculate the diff(us) when clear all N folios >>>>> clear_highpage/folio_zero_range/folio_zero_user >>>>> 3) release N folios >>>>> >>>>> the result(run 5 times) shown below on my machine, >>>>> >>>>> N=1, >>>>> clear_highpage folio_zero_range folio_zero_user >>>>> 1 69 74 177 >>>>> 2 57 62 168 >>>>> 3 54 58 234 >>>>> 4 54 58 157 >>>>> 5 56 62 148 >>>>> avg 58 62.8 176.8 >>>>> >>>>> >>>>> N=100 >>>>> clear_highpage folio_zero_range folio_zero_user >>>>> 1 11015 11309 32833 >>>>> 2 10385 11110 49751 >>>>> 3 10369 11056 33095 >>>>> 4 10332 11017 33106 >>>>> 5 10483 11000 49032 >>>>> avg 10516.8 11098.4 39563.4 >>>>> >>>>> N=512 >>>>> clear_highpage folio_zero_range folio_zero_user >>>>> 1 55560 60055 156876 >>>>> 2 55485 60024 157132 >>>>> 3 55474 60129 156658 >>>>> 4 55555 59867 157259 >>>>> 5 55528 59932 157108 >>>>> avg 55520.4 60001.4 157006.6 >>>>> >>>>> >>>>> >>>>> folio_zero_user with many cond_resched(), so time fluctuates a lot, >>>>> clear_highpage is better folio_zero_range as you said. >>>>> >>>>> Maybe add a new helper to convert all folio_zero_range(folio, 0, >>>>> folio_size(folio)) >>>>> to use clear_highpage + flush_dcache_folio? >>>> >>>> If this also improves performance for other existing callers of >>>> folio_zero_range(), then that's a positive outcome. >>> >>> >>> rm -f /tmp/test && fallocate -l 20G /tmp/test && fallocate -d -l 20G / >>> tmp/test && time fallocate -l 20G /tmp/test >>> >>> 1)mount always(2M folio) >>> with patch without patch >>> real 0m1.214s 0m1.111s >>> user 0m0.000s 0m0.000s >>> sys 0m1.210s 0m1.109s >>> >>> With this patch, the performance does have regression, >>> folio_zero_range() is bad than clear_highpage + flush_dcache_folio >>> >>> with patch >> >> Oh, this should without patch since it uses clear_highpage, >> >>> >>> 99.95% 0.00% fallocate [kernel.vmlinux] [k] vfs_fallocate >>> vfs_fallocate >>> - shmem_fallocate >>> 98.54% __pi_clear_page >>> - 1.38% shmem_get_folio_gfp >>> filemap_get_entry >>> >> and this one is with patch >>> without patch >>> 99.89% 0.00% fallocate [kernel.vmlinux] [k] shmem_fallocate >>> shmem_fallocate >>> - shmem_get_folio_gfp >>> 90.12% __memset >>> - 9.42% zero_user_segments.constprop.0 >>> 8.16% flush_dcache_page >>> 1.03% flush_dcache_folio >>> >>> >>> >>> >>> 2)mount never (4K folio) >>> real 0m3.159s 0m3.176s >>> user 0m0.000s 0m0.000s >>> sys 0m3.150s 0m3.169s >>> >>> But with this patch, the performance is improved a little, >>> folio_zero_range() is better than clear_highpage + flush_dcache_folio >>> >> >> For 4K, the result is fluctuating, so maybe no different. > > hi Kefeng, > what's your point? providing a helper like clear_highfolio() or similar? Yes, from above test, using clear_highpage/flush_dcache_folio is better than using folio_zero_range() for folio zero(especially for large folio), so I'd like to add a new helper, maybe name it folio_zero() since it zero the whole folio. > >> >>> with patch >>> 97.77% 3.37% fallocate [kernel.vmlinux] [k] shmem_fallocate >>> - 94.40% shmem_fallocate >>> - 93.70% shmem_get_folio_gfp >>> 66.60% __memset >>> - 7.43% filemap_get_entry >>> 3.49% xas_load >>> 1.32% zero_user_segments.constprop.0 >>> >>> without patch >>> 97.82% 3.22% fallocate [kernel.vmlinux] [k] shmem_fallocate >>> - 94.61% shmem_fallocate >>> 68.18% __pi_clear_page >>> - 25.60% shmem_get_folio_gfp >>> - 7.64% filemap_get_entry >>> 3.51% xas_load >>>> >>>>> >>>>> >>>>>> >>>>>>> if (sgp != SGP_WRITE && !folio_test_uptodate(folio)) { >>>>>>> - long i, n = folio_nr_pages(folio); >>>>>>> - >>>>>>> - for (i = 0; i < n; i++) >>>>>>> - clear_highpage(folio_page(folio, i)); >>>>>>> - flush_dcache_folio(folio); >>>>>>> + folio_zero_range(folio, 0, folio_size(folio)); >>>>>>> folio_mark_uptodate(folio); >>>>>>> } >>>>>> >>>>> >>>>> >>>> >>>> Thanks >>>> Barry >>> >>> >>
On Mon, Oct 21, 2024 at 6:16 PM Kefeng Wang <wangkefeng.wang@huawei.com> wrote: > > > > On 2024/10/21 12:15, Barry Song wrote: > > On Fri, Oct 18, 2024 at 8:48 PM Kefeng Wang <wangkefeng.wang@huawei.com> wrote: > >> > >> > >> > >> On 2024/10/18 15:32, Kefeng Wang wrote: > >>> > >>> > >>> On 2024/10/18 13:23, Barry Song wrote: > >>>> On Fri, Oct 18, 2024 at 6:20 PM Kefeng Wang > >>>> <wangkefeng.wang@huawei.com> wrote: > >>>>> > >>>>> > >>>>> > >>>>> On 2024/10/17 23:09, Matthew Wilcox wrote: > >>>>>> On Thu, Oct 17, 2024 at 10:25:04PM +0800, Kefeng Wang wrote: > >>>>>>> Directly use folio_zero_range() to cleanup code. > >>>>>> > >>>>>> Are you sure there's no performance regression introduced by this? > >>>>>> clear_highpage() is often optimised in ways that we can't optimise for > >>>>>> a plain memset(). On the other hand, if the folio is large, maybe a > >>>>>> modern CPU will be able to do better than clear-one-page-at-a-time. > >>>>>> > >>>>> > >>>>> Right, I missing this, clear_page might be better than memset, I change > >>>>> this one when look at the shmem_writepage(), which already convert to > >>>>> use folio_zero_range() from clear_highpage(), also I grep > >>>>> folio_zero_range(), there are some other to use folio_zero_range(). > >>>>> > >>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(folio, 0, > >>>>> folio_size(folio)); > >>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(f, > >>>>> 0, folio_size(f)); > >>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(f, > >>>>> 0, folio_size(f)); > >>>>> fs/libfs.c: folio_zero_range(folio, 0, folio_size(folio)); > >>>>> fs/ntfs3/frecord.c: folio_zero_range(folio, 0, > >>>>> folio_size(folio)); > >>>>> mm/page_io.c: folio_zero_range(folio, 0, folio_size(folio)); > >>>>> mm/shmem.c: folio_zero_range(folio, 0, folio_size(folio)); > >>>>> > >>>>> > >>>>>> IOW, what performance testing have you done with this patch? > >>>>> > >>>>> No performance test before, but I write a testcase, > >>>>> > >>>>> 1) allocate N large folios (folio_alloc(PMD_ORDER)) > >>>>> 2) then calculate the diff(us) when clear all N folios > >>>>> clear_highpage/folio_zero_range/folio_zero_user > >>>>> 3) release N folios > >>>>> > >>>>> the result(run 5 times) shown below on my machine, > >>>>> > >>>>> N=1, > >>>>> clear_highpage folio_zero_range folio_zero_user > >>>>> 1 69 74 177 > >>>>> 2 57 62 168 > >>>>> 3 54 58 234 > >>>>> 4 54 58 157 > >>>>> 5 56 62 148 > >>>>> avg 58 62.8 176.8 > >>>>> > >>>>> > >>>>> N=100 > >>>>> clear_highpage folio_zero_range folio_zero_user > >>>>> 1 11015 11309 32833 > >>>>> 2 10385 11110 49751 > >>>>> 3 10369 11056 33095 > >>>>> 4 10332 11017 33106 > >>>>> 5 10483 11000 49032 > >>>>> avg 10516.8 11098.4 39563.4 > >>>>> > >>>>> N=512 > >>>>> clear_highpage folio_zero_range folio_zero_user > >>>>> 1 55560 60055 156876 > >>>>> 2 55485 60024 157132 > >>>>> 3 55474 60129 156658 > >>>>> 4 55555 59867 157259 > >>>>> 5 55528 59932 157108 > >>>>> avg 55520.4 60001.4 157006.6 > >>>>> > >>>>> > >>>>> > >>>>> folio_zero_user with many cond_resched(), so time fluctuates a lot, > >>>>> clear_highpage is better folio_zero_range as you said. > >>>>> > >>>>> Maybe add a new helper to convert all folio_zero_range(folio, 0, > >>>>> folio_size(folio)) > >>>>> to use clear_highpage + flush_dcache_folio? > >>>> > >>>> If this also improves performance for other existing callers of > >>>> folio_zero_range(), then that's a positive outcome. > >>> > >>> > >>> rm -f /tmp/test && fallocate -l 20G /tmp/test && fallocate -d -l 20G / > >>> tmp/test && time fallocate -l 20G /tmp/test > >>> > >>> 1)mount always(2M folio) > >>> with patch without patch > >>> real 0m1.214s 0m1.111s > >>> user 0m0.000s 0m0.000s > >>> sys 0m1.210s 0m1.109s > >>> > >>> With this patch, the performance does have regression, > >>> folio_zero_range() is bad than clear_highpage + flush_dcache_folio > >>> > >>> with patch > >> > >> Oh, this should without patch since it uses clear_highpage, > >> > >>> > >>> 99.95% 0.00% fallocate [kernel.vmlinux] [k] vfs_fallocate > >>> vfs_fallocate > >>> - shmem_fallocate > >>> 98.54% __pi_clear_page > >>> - 1.38% shmem_get_folio_gfp > >>> filemap_get_entry > >>> > >> and this one is with patch > >>> without patch > >>> 99.89% 0.00% fallocate [kernel.vmlinux] [k] shmem_fallocate > >>> shmem_fallocate > >>> - shmem_get_folio_gfp > >>> 90.12% __memset > >>> - 9.42% zero_user_segments.constprop.0 > >>> 8.16% flush_dcache_page > >>> 1.03% flush_dcache_folio > >>> > >>> > >>> > >>> > >>> 2)mount never (4K folio) > >>> real 0m3.159s 0m3.176s > >>> user 0m0.000s 0m0.000s > >>> sys 0m3.150s 0m3.169s > >>> > >>> But with this patch, the performance is improved a little, > >>> folio_zero_range() is better than clear_highpage + flush_dcache_folio > >>> > >> > >> For 4K, the result is fluctuating, so maybe no different. > > > > hi Kefeng, > > what's your point? providing a helper like clear_highfolio() or similar? > > Yes, from above test, using clear_highpage/flush_dcache_folio is better > than using folio_zero_range() for folio zero(especially for large > folio), so I'd like to add a new helper, maybe name it folio_zero() > since it zero the whole folio. we already have a helper like folio_zero_user()? it is not good enough? > > > > >> > >>> with patch > >>> 97.77% 3.37% fallocate [kernel.vmlinux] [k] shmem_fallocate > >>> - 94.40% shmem_fallocate > >>> - 93.70% shmem_get_folio_gfp > >>> 66.60% __memset > >>> - 7.43% filemap_get_entry > >>> 3.49% xas_load > >>> 1.32% zero_user_segments.constprop.0 > >>> > >>> without patch > >>> 97.82% 3.22% fallocate [kernel.vmlinux] [k] shmem_fallocate > >>> - 94.61% shmem_fallocate > >>> 68.18% __pi_clear_page > >>> - 25.60% shmem_get_folio_gfp > >>> - 7.64% filemap_get_entry > >>> 3.51% xas_load > >>>> > >>>>> > >>>>> > >>>>>> > >>>>>>> if (sgp != SGP_WRITE && !folio_test_uptodate(folio)) { > >>>>>>> - long i, n = folio_nr_pages(folio); > >>>>>>> - > >>>>>>> - for (i = 0; i < n; i++) > >>>>>>> - clear_highpage(folio_page(folio, i)); > >>>>>>> - flush_dcache_folio(folio); > >>>>>>> + folio_zero_range(folio, 0, folio_size(folio)); > >>>>>>> folio_mark_uptodate(folio); > >>>>>>> } > >>>>>> > >>>>> > >>>>> > >>>> > >>>> Thanks > >>>> Barry > >>> > >>> > >> >
On 2024/10/21 13:38, Barry Song wrote: > On Mon, Oct 21, 2024 at 6:16 PM Kefeng Wang <wangkefeng.wang@huawei.com> wrote: >> >> >> >> On 2024/10/21 12:15, Barry Song wrote: >>> On Fri, Oct 18, 2024 at 8:48 PM Kefeng Wang <wangkefeng.wang@huawei.com> wrote: >>>> >>>> >>>> >>>> On 2024/10/18 15:32, Kefeng Wang wrote: >>>>> >>>>> >>>>> On 2024/10/18 13:23, Barry Song wrote: >>>>>> On Fri, Oct 18, 2024 at 6:20 PM Kefeng Wang >>>>>> <wangkefeng.wang@huawei.com> wrote: >>>>>>> >>>>>>> >>>>>>> >>>>>>> On 2024/10/17 23:09, Matthew Wilcox wrote: >>>>>>>> On Thu, Oct 17, 2024 at 10:25:04PM +0800, Kefeng Wang wrote: >>>>>>>>> Directly use folio_zero_range() to cleanup code. >>>>>>>> >>>>>>>> Are you sure there's no performance regression introduced by this? >>>>>>>> clear_highpage() is often optimised in ways that we can't optimise for >>>>>>>> a plain memset(). On the other hand, if the folio is large, maybe a >>>>>>>> modern CPU will be able to do better than clear-one-page-at-a-time. >>>>>>>> >>>>>>> >>>>>>> Right, I missing this, clear_page might be better than memset, I change >>>>>>> this one when look at the shmem_writepage(), which already convert to >>>>>>> use folio_zero_range() from clear_highpage(), also I grep >>>>>>> folio_zero_range(), there are some other to use folio_zero_range(). >>>>>>> >>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(folio, 0, >>>>>>> folio_size(folio)); >>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(f, >>>>>>> 0, folio_size(f)); >>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(f, >>>>>>> 0, folio_size(f)); >>>>>>> fs/libfs.c: folio_zero_range(folio, 0, folio_size(folio)); >>>>>>> fs/ntfs3/frecord.c: folio_zero_range(folio, 0, >>>>>>> folio_size(folio)); >>>>>>> mm/page_io.c: folio_zero_range(folio, 0, folio_size(folio)); >>>>>>> mm/shmem.c: folio_zero_range(folio, 0, folio_size(folio)); >>>>>>> >>>>>>> >>>>>>>> IOW, what performance testing have you done with this patch? >>>>>>> >>>>>>> No performance test before, but I write a testcase, >>>>>>> >>>>>>> 1) allocate N large folios (folio_alloc(PMD_ORDER)) >>>>>>> 2) then calculate the diff(us) when clear all N folios >>>>>>> clear_highpage/folio_zero_range/folio_zero_user >>>>>>> 3) release N folios >>>>>>> >>>>>>> the result(run 5 times) shown below on my machine, >>>>>>> >>>>>>> N=1, >>>>>>> clear_highpage folio_zero_range folio_zero_user >>>>>>> 1 69 74 177 >>>>>>> 2 57 62 168 >>>>>>> 3 54 58 234 >>>>>>> 4 54 58 157 >>>>>>> 5 56 62 148 >>>>>>> avg 58 62.8 176.8 >>>>>>> >>>>>>> >>>>>>> N=100 >>>>>>> clear_highpage folio_zero_range folio_zero_user >>>>>>> 1 11015 11309 32833 >>>>>>> 2 10385 11110 49751 >>>>>>> 3 10369 11056 33095 >>>>>>> 4 10332 11017 33106 >>>>>>> 5 10483 11000 49032 >>>>>>> avg 10516.8 11098.4 39563.4 >>>>>>> >>>>>>> N=512 >>>>>>> clear_highpage folio_zero_range folio_zero_user >>>>>>> 1 55560 60055 156876 >>>>>>> 2 55485 60024 157132 >>>>>>> 3 55474 60129 156658 >>>>>>> 4 55555 59867 157259 >>>>>>> 5 55528 59932 157108 >>>>>>> avg 55520.4 60001.4 157006.6 >>>>>>> >>>>>>> >>>>>>> >>>>>>> folio_zero_user with many cond_resched(), so time fluctuates a lot, >>>>>>> clear_highpage is better folio_zero_range as you said. >>>>>>> >>>>>>> Maybe add a new helper to convert all folio_zero_range(folio, 0, >>>>>>> folio_size(folio)) >>>>>>> to use clear_highpage + flush_dcache_folio? >>>>>> >>>>>> If this also improves performance for other existing callers of >>>>>> folio_zero_range(), then that's a positive outcome. >>>>> ... >>> hi Kefeng, >>> what's your point? providing a helper like clear_highfolio() or similar? >> >> Yes, from above test, using clear_highpage/flush_dcache_folio is better >> than using folio_zero_range() for folio zero(especially for large >> folio), so I'd like to add a new helper, maybe name it folio_zero() >> since it zero the whole folio. > > we already have a helper like folio_zero_user()? > it is not good enough? Since it is with many cond_resched(), the performance is worst...
On Mon, Oct 21, 2024 at 7:09 PM Kefeng Wang <wangkefeng.wang@huawei.com> wrote: > > > > On 2024/10/21 13:38, Barry Song wrote: > > On Mon, Oct 21, 2024 at 6:16 PM Kefeng Wang <wangkefeng.wang@huawei.com> wrote: > >> > >> > >> > >> On 2024/10/21 12:15, Barry Song wrote: > >>> On Fri, Oct 18, 2024 at 8:48 PM Kefeng Wang <wangkefeng.wang@huawei.com> wrote: > >>>> > >>>> > >>>> > >>>> On 2024/10/18 15:32, Kefeng Wang wrote: > >>>>> > >>>>> > >>>>> On 2024/10/18 13:23, Barry Song wrote: > >>>>>> On Fri, Oct 18, 2024 at 6:20 PM Kefeng Wang > >>>>>> <wangkefeng.wang@huawei.com> wrote: > >>>>>>> > >>>>>>> > >>>>>>> > >>>>>>> On 2024/10/17 23:09, Matthew Wilcox wrote: > >>>>>>>> On Thu, Oct 17, 2024 at 10:25:04PM +0800, Kefeng Wang wrote: > >>>>>>>>> Directly use folio_zero_range() to cleanup code. > >>>>>>>> > >>>>>>>> Are you sure there's no performance regression introduced by this? > >>>>>>>> clear_highpage() is often optimised in ways that we can't optimise for > >>>>>>>> a plain memset(). On the other hand, if the folio is large, maybe a > >>>>>>>> modern CPU will be able to do better than clear-one-page-at-a-time. > >>>>>>>> > >>>>>>> > >>>>>>> Right, I missing this, clear_page might be better than memset, I change > >>>>>>> this one when look at the shmem_writepage(), which already convert to > >>>>>>> use folio_zero_range() from clear_highpage(), also I grep > >>>>>>> folio_zero_range(), there are some other to use folio_zero_range(). > >>>>>>> > >>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(folio, 0, > >>>>>>> folio_size(folio)); > >>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(f, > >>>>>>> 0, folio_size(f)); > >>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(f, > >>>>>>> 0, folio_size(f)); > >>>>>>> fs/libfs.c: folio_zero_range(folio, 0, folio_size(folio)); > >>>>>>> fs/ntfs3/frecord.c: folio_zero_range(folio, 0, > >>>>>>> folio_size(folio)); > >>>>>>> mm/page_io.c: folio_zero_range(folio, 0, folio_size(folio)); > >>>>>>> mm/shmem.c: folio_zero_range(folio, 0, folio_size(folio)); > >>>>>>> > >>>>>>> > >>>>>>>> IOW, what performance testing have you done with this patch? > >>>>>>> > >>>>>>> No performance test before, but I write a testcase, > >>>>>>> > >>>>>>> 1) allocate N large folios (folio_alloc(PMD_ORDER)) > >>>>>>> 2) then calculate the diff(us) when clear all N folios > >>>>>>> clear_highpage/folio_zero_range/folio_zero_user > >>>>>>> 3) release N folios > >>>>>>> > >>>>>>> the result(run 5 times) shown below on my machine, > >>>>>>> > >>>>>>> N=1, > >>>>>>> clear_highpage folio_zero_range folio_zero_user > >>>>>>> 1 69 74 177 > >>>>>>> 2 57 62 168 > >>>>>>> 3 54 58 234 > >>>>>>> 4 54 58 157 > >>>>>>> 5 56 62 148 > >>>>>>> avg 58 62.8 176.8 > >>>>>>> > >>>>>>> > >>>>>>> N=100 > >>>>>>> clear_highpage folio_zero_range folio_zero_user > >>>>>>> 1 11015 11309 32833 > >>>>>>> 2 10385 11110 49751 > >>>>>>> 3 10369 11056 33095 > >>>>>>> 4 10332 11017 33106 > >>>>>>> 5 10483 11000 49032 > >>>>>>> avg 10516.8 11098.4 39563.4 > >>>>>>> > >>>>>>> N=512 > >>>>>>> clear_highpage folio_zero_range folio_zero_user > >>>>>>> 1 55560 60055 156876 > >>>>>>> 2 55485 60024 157132 > >>>>>>> 3 55474 60129 156658 > >>>>>>> 4 55555 59867 157259 > >>>>>>> 5 55528 59932 157108 > >>>>>>> avg 55520.4 60001.4 157006.6 > >>>>>>> > >>>>>>> > >>>>>>> > >>>>>>> folio_zero_user with many cond_resched(), so time fluctuates a lot, > >>>>>>> clear_highpage is better folio_zero_range as you said. > >>>>>>> > >>>>>>> Maybe add a new helper to convert all folio_zero_range(folio, 0, > >>>>>>> folio_size(folio)) > >>>>>>> to use clear_highpage + flush_dcache_folio? > >>>>>> > >>>>>> If this also improves performance for other existing callers of > >>>>>> folio_zero_range(), then that's a positive outcome. > >>>>> > ... > > >>> hi Kefeng, > >>> what's your point? providing a helper like clear_highfolio() or similar? > >> > >> Yes, from above test, using clear_highpage/flush_dcache_folio is better > >> than using folio_zero_range() for folio zero(especially for large > >> folio), so I'd like to add a new helper, maybe name it folio_zero() > >> since it zero the whole folio. > > > > we already have a helper like folio_zero_user()? > > it is not good enough? > > Since it is with many cond_resched(), the performance is worst... Not exactly? It should have zero cost for a preemptible kernel. For a non-preemptible kernel, it helps avoid clearing the folio from occupying the CPU and starving other processes, right?
On Mon, Oct 21, 2024 at 8:47 PM Barry Song <21cnbao@gmail.com> wrote: > > On Mon, Oct 21, 2024 at 7:09 PM Kefeng Wang <wangkefeng.wang@huawei.com> wrote: > > > > > > > > On 2024/10/21 13:38, Barry Song wrote: > > > On Mon, Oct 21, 2024 at 6:16 PM Kefeng Wang <wangkefeng.wang@huawei.com> wrote: > > >> > > >> > > >> > > >> On 2024/10/21 12:15, Barry Song wrote: > > >>> On Fri, Oct 18, 2024 at 8:48 PM Kefeng Wang <wangkefeng.wang@huawei.com> wrote: > > >>>> > > >>>> > > >>>> > > >>>> On 2024/10/18 15:32, Kefeng Wang wrote: > > >>>>> > > >>>>> > > >>>>> On 2024/10/18 13:23, Barry Song wrote: > > >>>>>> On Fri, Oct 18, 2024 at 6:20 PM Kefeng Wang > > >>>>>> <wangkefeng.wang@huawei.com> wrote: > > >>>>>>> > > >>>>>>> > > >>>>>>> > > >>>>>>> On 2024/10/17 23:09, Matthew Wilcox wrote: > > >>>>>>>> On Thu, Oct 17, 2024 at 10:25:04PM +0800, Kefeng Wang wrote: > > >>>>>>>>> Directly use folio_zero_range() to cleanup code. > > >>>>>>>> > > >>>>>>>> Are you sure there's no performance regression introduced by this? > > >>>>>>>> clear_highpage() is often optimised in ways that we can't optimise for > > >>>>>>>> a plain memset(). On the other hand, if the folio is large, maybe a > > >>>>>>>> modern CPU will be able to do better than clear-one-page-at-a-time. > > >>>>>>>> > > >>>>>>> > > >>>>>>> Right, I missing this, clear_page might be better than memset, I change > > >>>>>>> this one when look at the shmem_writepage(), which already convert to > > >>>>>>> use folio_zero_range() from clear_highpage(), also I grep > > >>>>>>> folio_zero_range(), there are some other to use folio_zero_range(). > > >>>>>>> > > >>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(folio, 0, > > >>>>>>> folio_size(folio)); > > >>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(f, > > >>>>>>> 0, folio_size(f)); > > >>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(f, > > >>>>>>> 0, folio_size(f)); > > >>>>>>> fs/libfs.c: folio_zero_range(folio, 0, folio_size(folio)); > > >>>>>>> fs/ntfs3/frecord.c: folio_zero_range(folio, 0, > > >>>>>>> folio_size(folio)); > > >>>>>>> mm/page_io.c: folio_zero_range(folio, 0, folio_size(folio)); > > >>>>>>> mm/shmem.c: folio_zero_range(folio, 0, folio_size(folio)); > > >>>>>>> > > >>>>>>> > > >>>>>>>> IOW, what performance testing have you done with this patch? > > >>>>>>> > > >>>>>>> No performance test before, but I write a testcase, > > >>>>>>> > > >>>>>>> 1) allocate N large folios (folio_alloc(PMD_ORDER)) > > >>>>>>> 2) then calculate the diff(us) when clear all N folios > > >>>>>>> clear_highpage/folio_zero_range/folio_zero_user > > >>>>>>> 3) release N folios > > >>>>>>> > > >>>>>>> the result(run 5 times) shown below on my machine, > > >>>>>>> > > >>>>>>> N=1, > > >>>>>>> clear_highpage folio_zero_range folio_zero_user > > >>>>>>> 1 69 74 177 > > >>>>>>> 2 57 62 168 > > >>>>>>> 3 54 58 234 > > >>>>>>> 4 54 58 157 > > >>>>>>> 5 56 62 148 > > >>>>>>> avg 58 62.8 176.8 > > >>>>>>> > > >>>>>>> > > >>>>>>> N=100 > > >>>>>>> clear_highpage folio_zero_range folio_zero_user > > >>>>>>> 1 11015 11309 32833 > > >>>>>>> 2 10385 11110 49751 > > >>>>>>> 3 10369 11056 33095 > > >>>>>>> 4 10332 11017 33106 > > >>>>>>> 5 10483 11000 49032 > > >>>>>>> avg 10516.8 11098.4 39563.4 > > >>>>>>> > > >>>>>>> N=512 > > >>>>>>> clear_highpage folio_zero_range folio_zero_user > > >>>>>>> 1 55560 60055 156876 > > >>>>>>> 2 55485 60024 157132 > > >>>>>>> 3 55474 60129 156658 > > >>>>>>> 4 55555 59867 157259 > > >>>>>>> 5 55528 59932 157108 > > >>>>>>> avg 55520.4 60001.4 157006.6 > > >>>>>>> > > >>>>>>> > > >>>>>>> > > >>>>>>> folio_zero_user with many cond_resched(), so time fluctuates a lot, > > >>>>>>> clear_highpage is better folio_zero_range as you said. > > >>>>>>> > > >>>>>>> Maybe add a new helper to convert all folio_zero_range(folio, 0, > > >>>>>>> folio_size(folio)) > > >>>>>>> to use clear_highpage + flush_dcache_folio? > > >>>>>> > > >>>>>> If this also improves performance for other existing callers of > > >>>>>> folio_zero_range(), then that's a positive outcome. > > >>>>> > > ... > > > > >>> hi Kefeng, > > >>> what's your point? providing a helper like clear_highfolio() or similar? > > >> > > >> Yes, from above test, using clear_highpage/flush_dcache_folio is better > > >> than using folio_zero_range() for folio zero(especially for large > > >> folio), so I'd like to add a new helper, maybe name it folio_zero() > > >> since it zero the whole folio. > > > > > > we already have a helper like folio_zero_user()? > > > it is not good enough? > > > > Since it is with many cond_resched(), the performance is worst... > > Not exactly? It should have zero cost for a preemptible kernel. > For a non-preemptible kernel, it helps avoid clearing the folio > from occupying the CPU and starving other processes, right? --- a/mm/shmem.c +++ b/mm/shmem.c @@ -2393,10 +2393,7 @@ static int shmem_get_folio_gfp(struct inode *inode, pgoff_t index, * it now, lest undo on failure cancel our earlier guarantee. */ if (sgp != SGP_WRITE && !folio_test_uptodate(folio)) { - long i, n = folio_nr_pages(folio); - - for (i = 0; i < n; i++) - clear_highpage(folio_page(folio, i)); + folio_zero_user(folio, vmf->address); flush_dcache_folio(folio); folio_mark_uptodate(folio); } Do we perform better or worse with the following?
On 2024/10/21 15:55, Barry Song wrote: > On Mon, Oct 21, 2024 at 8:47 PM Barry Song <21cnbao@gmail.com> wrote: >> >> On Mon, Oct 21, 2024 at 7:09 PM Kefeng Wang <wangkefeng.wang@huawei.com> wrote: >>> >>> >>> >>> On 2024/10/21 13:38, Barry Song wrote: >>>> On Mon, Oct 21, 2024 at 6:16 PM Kefeng Wang <wangkefeng.wang@huawei.com> wrote: >>>>> >>>>> >>>>> >>>>> On 2024/10/21 12:15, Barry Song wrote: >>>>>> On Fri, Oct 18, 2024 at 8:48 PM Kefeng Wang <wangkefeng.wang@huawei.com> wrote: >>>>>>> >>>>>>> >>>>>>> >>>>>>> On 2024/10/18 15:32, Kefeng Wang wrote: >>>>>>>> >>>>>>>> >>>>>>>> On 2024/10/18 13:23, Barry Song wrote: >>>>>>>>> On Fri, Oct 18, 2024 at 6:20 PM Kefeng Wang >>>>>>>>> <wangkefeng.wang@huawei.com> wrote: >>>>>>>>>> >>>>>>>>>> >>>>>>>>>> >>>>>>>>>> On 2024/10/17 23:09, Matthew Wilcox wrote: >>>>>>>>>>> On Thu, Oct 17, 2024 at 10:25:04PM +0800, Kefeng Wang wrote: >>>>>>>>>>>> Directly use folio_zero_range() to cleanup code. >>>>>>>>>>> >>>>>>>>>>> Are you sure there's no performance regression introduced by this? >>>>>>>>>>> clear_highpage() is often optimised in ways that we can't optimise for >>>>>>>>>>> a plain memset(). On the other hand, if the folio is large, maybe a >>>>>>>>>>> modern CPU will be able to do better than clear-one-page-at-a-time. >>>>>>>>>>> >>>>>>>>>> >>>>>>>>>> Right, I missing this, clear_page might be better than memset, I change >>>>>>>>>> this one when look at the shmem_writepage(), which already convert to >>>>>>>>>> use folio_zero_range() from clear_highpage(), also I grep >>>>>>>>>> folio_zero_range(), there are some other to use folio_zero_range(). >>>>>>>>>> >>>>>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(folio, 0, >>>>>>>>>> folio_size(folio)); >>>>>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(f, >>>>>>>>>> 0, folio_size(f)); >>>>>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(f, >>>>>>>>>> 0, folio_size(f)); >>>>>>>>>> fs/libfs.c: folio_zero_range(folio, 0, folio_size(folio)); >>>>>>>>>> fs/ntfs3/frecord.c: folio_zero_range(folio, 0, >>>>>>>>>> folio_size(folio)); >>>>>>>>>> mm/page_io.c: folio_zero_range(folio, 0, folio_size(folio)); >>>>>>>>>> mm/shmem.c: folio_zero_range(folio, 0, folio_size(folio)); >>>>>>>>>> >>>>>>>>>> >>>>>>>>>>> IOW, what performance testing have you done with this patch? >>>>>>>>>> >>>>>>>>>> No performance test before, but I write a testcase, >>>>>>>>>> >>>>>>>>>> 1) allocate N large folios (folio_alloc(PMD_ORDER)) >>>>>>>>>> 2) then calculate the diff(us) when clear all N folios >>>>>>>>>> clear_highpage/folio_zero_range/folio_zero_user >>>>>>>>>> 3) release N folios >>>>>>>>>> >>>>>>>>>> the result(run 5 times) shown below on my machine, >>>>>>>>>> >>>>>>>>>> N=1, >>>>>>>>>> clear_highpage folio_zero_range folio_zero_user >>>>>>>>>> 1 69 74 177 >>>>>>>>>> 2 57 62 168 >>>>>>>>>> 3 54 58 234 >>>>>>>>>> 4 54 58 157 >>>>>>>>>> 5 56 62 148 >>>>>>>>>> avg 58 62.8 176.8 >>>>>>>>>> >>>>>>>>>> >>>>>>>>>> N=100 >>>>>>>>>> clear_highpage folio_zero_range folio_zero_user >>>>>>>>>> 1 11015 11309 32833 >>>>>>>>>> 2 10385 11110 49751 >>>>>>>>>> 3 10369 11056 33095 >>>>>>>>>> 4 10332 11017 33106 >>>>>>>>>> 5 10483 11000 49032 >>>>>>>>>> avg 10516.8 11098.4 39563.4 >>>>>>>>>> >>>>>>>>>> N=512 >>>>>>>>>> clear_highpage folio_zero_range folio_zero_user >>>>>>>>>> 1 55560 60055 156876 >>>>>>>>>> 2 55485 60024 157132 >>>>>>>>>> 3 55474 60129 156658 >>>>>>>>>> 4 55555 59867 157259 >>>>>>>>>> 5 55528 59932 157108 >>>>>>>>>> avg 55520.4 60001.4 157006.6 >>>>>>>>>> >>>>>>>>>> >>>>>>>>>> >>>>>>>>>> folio_zero_user with many cond_resched(), so time fluctuates a lot, >>>>>>>>>> clear_highpage is better folio_zero_range as you said. >>>>>>>>>> >>>>>>>>>> Maybe add a new helper to convert all folio_zero_range(folio, 0, >>>>>>>>>> folio_size(folio)) >>>>>>>>>> to use clear_highpage + flush_dcache_folio? >>>>>>>>> >>>>>>>>> If this also improves performance for other existing callers of >>>>>>>>> folio_zero_range(), then that's a positive outcome. >>>>>>>> >>> ... >>> >>>>>> hi Kefeng, >>>>>> what's your point? providing a helper like clear_highfolio() or similar? >>>>> >>>>> Yes, from above test, using clear_highpage/flush_dcache_folio is better >>>>> than using folio_zero_range() for folio zero(especially for large >>>>> folio), so I'd like to add a new helper, maybe name it folio_zero() >>>>> since it zero the whole folio. >>>> >>>> we already have a helper like folio_zero_user()? >>>> it is not good enough? >>> >>> Since it is with many cond_resched(), the performance is worst... >> >> Not exactly? It should have zero cost for a preemptible kernel. >> For a non-preemptible kernel, it helps avoid clearing the folio >> from occupying the CPU and starving other processes, right? > > --- a/mm/shmem.c > +++ b/mm/shmem.c > > @@ -2393,10 +2393,7 @@ static int shmem_get_folio_gfp(struct inode > *inode, pgoff_t index, > * it now, lest undo on failure cancel our earlier guarantee. > */ > > if (sgp != SGP_WRITE && !folio_test_uptodate(folio)) { > - long i, n = folio_nr_pages(folio); > - > - for (i = 0; i < n; i++) > - clear_highpage(folio_page(folio, i)); > + folio_zero_user(folio, vmf->address); > flush_dcache_folio(folio); > folio_mark_uptodate(folio); > } > > Do we perform better or worse with the following? Here is for SGP_FALLOC, vmf = NULL, we could use folio_zero_user(folio, 0), I think the performance is worse, will retest once I can access hardware.
On Mon, Oct 21, 2024 at 9:14 PM Kefeng Wang <wangkefeng.wang@huawei.com> wrote: > > > > On 2024/10/21 15:55, Barry Song wrote: > > On Mon, Oct 21, 2024 at 8:47 PM Barry Song <21cnbao@gmail.com> wrote: > >> > >> On Mon, Oct 21, 2024 at 7:09 PM Kefeng Wang <wangkefeng.wang@huawei.com> wrote: > >>> > >>> > >>> > >>> On 2024/10/21 13:38, Barry Song wrote: > >>>> On Mon, Oct 21, 2024 at 6:16 PM Kefeng Wang <wangkefeng.wang@huawei.com> wrote: > >>>>> > >>>>> > >>>>> > >>>>> On 2024/10/21 12:15, Barry Song wrote: > >>>>>> On Fri, Oct 18, 2024 at 8:48 PM Kefeng Wang <wangkefeng.wang@huawei.com> wrote: > >>>>>>> > >>>>>>> > >>>>>>> > >>>>>>> On 2024/10/18 15:32, Kefeng Wang wrote: > >>>>>>>> > >>>>>>>> > >>>>>>>> On 2024/10/18 13:23, Barry Song wrote: > >>>>>>>>> On Fri, Oct 18, 2024 at 6:20 PM Kefeng Wang > >>>>>>>>> <wangkefeng.wang@huawei.com> wrote: > >>>>>>>>>> > >>>>>>>>>> > >>>>>>>>>> > >>>>>>>>>> On 2024/10/17 23:09, Matthew Wilcox wrote: > >>>>>>>>>>> On Thu, Oct 17, 2024 at 10:25:04PM +0800, Kefeng Wang wrote: > >>>>>>>>>>>> Directly use folio_zero_range() to cleanup code. > >>>>>>>>>>> > >>>>>>>>>>> Are you sure there's no performance regression introduced by this? > >>>>>>>>>>> clear_highpage() is often optimised in ways that we can't optimise for > >>>>>>>>>>> a plain memset(). On the other hand, if the folio is large, maybe a > >>>>>>>>>>> modern CPU will be able to do better than clear-one-page-at-a-time. > >>>>>>>>>>> > >>>>>>>>>> > >>>>>>>>>> Right, I missing this, clear_page might be better than memset, I change > >>>>>>>>>> this one when look at the shmem_writepage(), which already convert to > >>>>>>>>>> use folio_zero_range() from clear_highpage(), also I grep > >>>>>>>>>> folio_zero_range(), there are some other to use folio_zero_range(). > >>>>>>>>>> > >>>>>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(folio, 0, > >>>>>>>>>> folio_size(folio)); > >>>>>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(f, > >>>>>>>>>> 0, folio_size(f)); > >>>>>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(f, > >>>>>>>>>> 0, folio_size(f)); > >>>>>>>>>> fs/libfs.c: folio_zero_range(folio, 0, folio_size(folio)); > >>>>>>>>>> fs/ntfs3/frecord.c: folio_zero_range(folio, 0, > >>>>>>>>>> folio_size(folio)); > >>>>>>>>>> mm/page_io.c: folio_zero_range(folio, 0, folio_size(folio)); > >>>>>>>>>> mm/shmem.c: folio_zero_range(folio, 0, folio_size(folio)); > >>>>>>>>>> > >>>>>>>>>> > >>>>>>>>>>> IOW, what performance testing have you done with this patch? > >>>>>>>>>> > >>>>>>>>>> No performance test before, but I write a testcase, > >>>>>>>>>> > >>>>>>>>>> 1) allocate N large folios (folio_alloc(PMD_ORDER)) > >>>>>>>>>> 2) then calculate the diff(us) when clear all N folios > >>>>>>>>>> clear_highpage/folio_zero_range/folio_zero_user > >>>>>>>>>> 3) release N folios > >>>>>>>>>> > >>>>>>>>>> the result(run 5 times) shown below on my machine, > >>>>>>>>>> > >>>>>>>>>> N=1, > >>>>>>>>>> clear_highpage folio_zero_range folio_zero_user > >>>>>>>>>> 1 69 74 177 > >>>>>>>>>> 2 57 62 168 > >>>>>>>>>> 3 54 58 234 > >>>>>>>>>> 4 54 58 157 > >>>>>>>>>> 5 56 62 148 > >>>>>>>>>> avg 58 62.8 176.8 > >>>>>>>>>> > >>>>>>>>>> > >>>>>>>>>> N=100 > >>>>>>>>>> clear_highpage folio_zero_range folio_zero_user > >>>>>>>>>> 1 11015 11309 32833 > >>>>>>>>>> 2 10385 11110 49751 > >>>>>>>>>> 3 10369 11056 33095 > >>>>>>>>>> 4 10332 11017 33106 > >>>>>>>>>> 5 10483 11000 49032 > >>>>>>>>>> avg 10516.8 11098.4 39563.4 > >>>>>>>>>> > >>>>>>>>>> N=512 > >>>>>>>>>> clear_highpage folio_zero_range folio_zero_user > >>>>>>>>>> 1 55560 60055 156876 > >>>>>>>>>> 2 55485 60024 157132 > >>>>>>>>>> 3 55474 60129 156658 > >>>>>>>>>> 4 55555 59867 157259 > >>>>>>>>>> 5 55528 59932 157108 > >>>>>>>>>> avg 55520.4 60001.4 157006.6 > >>>>>>>>>> > >>>>>>>>>> > >>>>>>>>>> > >>>>>>>>>> folio_zero_user with many cond_resched(), so time fluctuates a lot, > >>>>>>>>>> clear_highpage is better folio_zero_range as you said. > >>>>>>>>>> > >>>>>>>>>> Maybe add a new helper to convert all folio_zero_range(folio, 0, > >>>>>>>>>> folio_size(folio)) > >>>>>>>>>> to use clear_highpage + flush_dcache_folio? > >>>>>>>>> > >>>>>>>>> If this also improves performance for other existing callers of > >>>>>>>>> folio_zero_range(), then that's a positive outcome. > >>>>>>>> > >>> ... > >>> > >>>>>> hi Kefeng, > >>>>>> what's your point? providing a helper like clear_highfolio() or similar? > >>>>> > >>>>> Yes, from above test, using clear_highpage/flush_dcache_folio is better > >>>>> than using folio_zero_range() for folio zero(especially for large > >>>>> folio), so I'd like to add a new helper, maybe name it folio_zero() > >>>>> since it zero the whole folio. > >>>> > >>>> we already have a helper like folio_zero_user()? > >>>> it is not good enough? > >>> > >>> Since it is with many cond_resched(), the performance is worst... > >> > >> Not exactly? It should have zero cost for a preemptible kernel. > >> For a non-preemptible kernel, it helps avoid clearing the folio > >> from occupying the CPU and starving other processes, right? > > > > --- a/mm/shmem.c > > +++ b/mm/shmem.c > > > > @@ -2393,10 +2393,7 @@ static int shmem_get_folio_gfp(struct inode > > *inode, pgoff_t index, > > * it now, lest undo on failure cancel our earlier guarantee. > > */ > > > > if (sgp != SGP_WRITE && !folio_test_uptodate(folio)) { > > - long i, n = folio_nr_pages(folio); > > - > > - for (i = 0; i < n; i++) > > - clear_highpage(folio_page(folio, i)); > > + folio_zero_user(folio, vmf->address); > > flush_dcache_folio(folio); > > folio_mark_uptodate(folio); > > } > > > > Do we perform better or worse with the following? > > Here is for SGP_FALLOC, vmf = NULL, we could use folio_zero_user(folio, > 0), I think the performance is worse, will retest once I can access > hardware. Perhaps, since the current code uses clear_hugepage(). Does using index << PAGE_SHIFT as the addr_hint offer any benefit? > >
On 2024/10/21 17:17, Barry Song wrote: > On Mon, Oct 21, 2024 at 9:14 PM Kefeng Wang <wangkefeng.wang@huawei.com> wrote: >> >> >> >> On 2024/10/21 15:55, Barry Song wrote: >>> On Mon, Oct 21, 2024 at 8:47 PM Barry Song <21cnbao@gmail.com> wrote: >>>> >>>> On Mon, Oct 21, 2024 at 7:09 PM Kefeng Wang <wangkefeng.wang@huawei.com> wrote: >>>>> >>>>> >>>>> >>>>> On 2024/10/21 13:38, Barry Song wrote: >>>>>> On Mon, Oct 21, 2024 at 6:16 PM Kefeng Wang <wangkefeng.wang@huawei.com> wrote: >>>>>>> >>>>>>> >>>>>>> >>>>>>> On 2024/10/21 12:15, Barry Song wrote: >>>>>>>> On Fri, Oct 18, 2024 at 8:48 PM Kefeng Wang <wangkefeng.wang@huawei.com> wrote: >>>>>>>>> >>>>>>>>> >>>>>>>>> >>>>>>>>> On 2024/10/18 15:32, Kefeng Wang wrote: >>>>>>>>>> >>>>>>>>>> >>>>>>>>>> On 2024/10/18 13:23, Barry Song wrote: >>>>>>>>>>> On Fri, Oct 18, 2024 at 6:20 PM Kefeng Wang >>>>>>>>>>> <wangkefeng.wang@huawei.com> wrote: >>>>>>>>>>>> >>>>>>>>>>>> >>>>>>>>>>>> >>>>>>>>>>>> On 2024/10/17 23:09, Matthew Wilcox wrote: >>>>>>>>>>>>> On Thu, Oct 17, 2024 at 10:25:04PM +0800, Kefeng Wang wrote: >>>>>>>>>>>>>> Directly use folio_zero_range() to cleanup code. >>>>>>>>>>>>> >>>>>>>>>>>>> Are you sure there's no performance regression introduced by this? >>>>>>>>>>>>> clear_highpage() is often optimised in ways that we can't optimise for >>>>>>>>>>>>> a plain memset(). On the other hand, if the folio is large, maybe a >>>>>>>>>>>>> modern CPU will be able to do better than clear-one-page-at-a-time. >>>>>>>>>>>>> >>>>>>>>>>>> >>>>>>>>>>>> Right, I missing this, clear_page might be better than memset, I change >>>>>>>>>>>> this one when look at the shmem_writepage(), which already convert to >>>>>>>>>>>> use folio_zero_range() from clear_highpage(), also I grep >>>>>>>>>>>> folio_zero_range(), there are some other to use folio_zero_range(). >>>>>>>>>>>> >>>>>>>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(folio, 0, >>>>>>>>>>>> folio_size(folio)); >>>>>>>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(f, >>>>>>>>>>>> 0, folio_size(f)); >>>>>>>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(f, >>>>>>>>>>>> 0, folio_size(f)); >>>>>>>>>>>> fs/libfs.c: folio_zero_range(folio, 0, folio_size(folio)); >>>>>>>>>>>> fs/ntfs3/frecord.c: folio_zero_range(folio, 0, >>>>>>>>>>>> folio_size(folio)); >>>>>>>>>>>> mm/page_io.c: folio_zero_range(folio, 0, folio_size(folio)); >>>>>>>>>>>> mm/shmem.c: folio_zero_range(folio, 0, folio_size(folio)); >>>>>>>>>>>> >>>>>>>>>>>> >>>>>>>>>>>>> IOW, what performance testing have you done with this patch? >>>>>>>>>>>> >>>>>>>>>>>> No performance test before, but I write a testcase, >>>>>>>>>>>> >>>>>>>>>>>> 1) allocate N large folios (folio_alloc(PMD_ORDER)) >>>>>>>>>>>> 2) then calculate the diff(us) when clear all N folios >>>>>>>>>>>> clear_highpage/folio_zero_range/folio_zero_user >>>>>>>>>>>> 3) release N folios >>>>>>>>>>>> >>>>>>>>>>>> the result(run 5 times) shown below on my machine, >>>>>>>>>>>> >>>>>>>>>>>> N=1, >>>>>>>>>>>> clear_highpage folio_zero_range folio_zero_user >>>>>>>>>>>> 1 69 74 177 >>>>>>>>>>>> 2 57 62 168 >>>>>>>>>>>> 3 54 58 234 >>>>>>>>>>>> 4 54 58 157 >>>>>>>>>>>> 5 56 62 148 >>>>>>>>>>>> avg 58 62.8 176.8 >>>>>>>>>>>> >>>>>>>>>>>> >>>>>>>>>>>> N=100 >>>>>>>>>>>> clear_highpage folio_zero_range folio_zero_user >>>>>>>>>>>> 1 11015 11309 32833 >>>>>>>>>>>> 2 10385 11110 49751 >>>>>>>>>>>> 3 10369 11056 33095 >>>>>>>>>>>> 4 10332 11017 33106 >>>>>>>>>>>> 5 10483 11000 49032 >>>>>>>>>>>> avg 10516.8 11098.4 39563.4 >>>>>>>>>>>> >>>>>>>>>>>> N=512 >>>>>>>>>>>> clear_highpage folio_zero_range folio_zero_user >>>>>>>>>>>> 1 55560 60055 156876 >>>>>>>>>>>> 2 55485 60024 157132 >>>>>>>>>>>> 3 55474 60129 156658 >>>>>>>>>>>> 4 55555 59867 157259 >>>>>>>>>>>> 5 55528 59932 157108 >>>>>>>>>>>> avg 55520.4 60001.4 157006.6 >>>>>>>>>>>> >>>>>>>>>>>> >>>>>>>>>>>> >>>>>>>>>>>> folio_zero_user with many cond_resched(), so time fluctuates a lot, >>>>>>>>>>>> clear_highpage is better folio_zero_range as you said. >>>>>>>>>>>> >>>>>>>>>>>> Maybe add a new helper to convert all folio_zero_range(folio, 0, >>>>>>>>>>>> folio_size(folio)) >>>>>>>>>>>> to use clear_highpage + flush_dcache_folio? >>>>>>>>>>> >>>>>>>>>>> If this also improves performance for other existing callers of >>>>>>>>>>> folio_zero_range(), then that's a positive outcome. >>>>>>>>>> >>>>> ... >>>>> >>>>>>>> hi Kefeng, >>>>>>>> what's your point? providing a helper like clear_highfolio() or similar? >>>>>>> >>>>>>> Yes, from above test, using clear_highpage/flush_dcache_folio is better >>>>>>> than using folio_zero_range() for folio zero(especially for large >>>>>>> folio), so I'd like to add a new helper, maybe name it folio_zero() >>>>>>> since it zero the whole folio. >>>>>> >>>>>> we already have a helper like folio_zero_user()? >>>>>> it is not good enough? >>>>> >>>>> Since it is with many cond_resched(), the performance is worst... >>>> >>>> Not exactly? It should have zero cost for a preemptible kernel. >>>> For a non-preemptible kernel, it helps avoid clearing the folio >>>> from occupying the CPU and starving other processes, right? >>> >>> --- a/mm/shmem.c >>> +++ b/mm/shmem.c >>> >>> @@ -2393,10 +2393,7 @@ static int shmem_get_folio_gfp(struct inode >>> *inode, pgoff_t index, >>> * it now, lest undo on failure cancel our earlier guarantee. >>> */ >>> >>> if (sgp != SGP_WRITE && !folio_test_uptodate(folio)) { >>> - long i, n = folio_nr_pages(folio); >>> - >>> - for (i = 0; i < n; i++) >>> - clear_highpage(folio_page(folio, i)); >>> + folio_zero_user(folio, vmf->address); >>> flush_dcache_folio(folio); >>> folio_mark_uptodate(folio); >>> } >>> >>> Do we perform better or worse with the following? >> >> Here is for SGP_FALLOC, vmf = NULL, we could use folio_zero_user(folio, >> 0), I think the performance is worse, will retest once I can access >> hardware. > > Perhaps, since the current code uses clear_hugepage(). Does using > index << PAGE_SHIFT as the addr_hint offer any benefit? > when use folio_zero_user(), the performance is vary bad with above fallocate test(mount huge=always), folio_zero_range clear_highpage folio_zero_user real 0m1.214s 0m1.111s 0m3.159s user 0m0.000s 0m0.000s 0m0.000s sys 0m1.210s 0m1.109s 0m3.152s I tried with addr_hint = 0/index << PAGE_SHIFT, no obvious different.
On Tue, Oct 22, 2024 at 4:33 AM Kefeng Wang <wangkefeng.wang@huawei.com> wrote: > > > > On 2024/10/21 17:17, Barry Song wrote: > > On Mon, Oct 21, 2024 at 9:14 PM Kefeng Wang <wangkefeng.wang@huawei.com> wrote: > >> > >> > >> > >> On 2024/10/21 15:55, Barry Song wrote: > >>> On Mon, Oct 21, 2024 at 8:47 PM Barry Song <21cnbao@gmail.com> wrote: > >>>> > >>>> On Mon, Oct 21, 2024 at 7:09 PM Kefeng Wang <wangkefeng.wang@huawei.com> wrote: > >>>>> > >>>>> > >>>>> > >>>>> On 2024/10/21 13:38, Barry Song wrote: > >>>>>> On Mon, Oct 21, 2024 at 6:16 PM Kefeng Wang <wangkefeng.wang@huawei.com> wrote: > >>>>>>> > >>>>>>> > >>>>>>> > >>>>>>> On 2024/10/21 12:15, Barry Song wrote: > >>>>>>>> On Fri, Oct 18, 2024 at 8:48 PM Kefeng Wang <wangkefeng.wang@huawei.com> wrote: > >>>>>>>>> > >>>>>>>>> > >>>>>>>>> > >>>>>>>>> On 2024/10/18 15:32, Kefeng Wang wrote: > >>>>>>>>>> > >>>>>>>>>> > >>>>>>>>>> On 2024/10/18 13:23, Barry Song wrote: > >>>>>>>>>>> On Fri, Oct 18, 2024 at 6:20 PM Kefeng Wang > >>>>>>>>>>> <wangkefeng.wang@huawei.com> wrote: > >>>>>>>>>>>> > >>>>>>>>>>>> > >>>>>>>>>>>> > >>>>>>>>>>>> On 2024/10/17 23:09, Matthew Wilcox wrote: > >>>>>>>>>>>>> On Thu, Oct 17, 2024 at 10:25:04PM +0800, Kefeng Wang wrote: > >>>>>>>>>>>>>> Directly use folio_zero_range() to cleanup code. > >>>>>>>>>>>>> > >>>>>>>>>>>>> Are you sure there's no performance regression introduced by this? > >>>>>>>>>>>>> clear_highpage() is often optimised in ways that we can't optimise for > >>>>>>>>>>>>> a plain memset(). On the other hand, if the folio is large, maybe a > >>>>>>>>>>>>> modern CPU will be able to do better than clear-one-page-at-a-time. > >>>>>>>>>>>>> > >>>>>>>>>>>> > >>>>>>>>>>>> Right, I missing this, clear_page might be better than memset, I change > >>>>>>>>>>>> this one when look at the shmem_writepage(), which already convert to > >>>>>>>>>>>> use folio_zero_range() from clear_highpage(), also I grep > >>>>>>>>>>>> folio_zero_range(), there are some other to use folio_zero_range(). > >>>>>>>>>>>> > >>>>>>>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(folio, 0, > >>>>>>>>>>>> folio_size(folio)); > >>>>>>>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(f, > >>>>>>>>>>>> 0, folio_size(f)); > >>>>>>>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(f, > >>>>>>>>>>>> 0, folio_size(f)); > >>>>>>>>>>>> fs/libfs.c: folio_zero_range(folio, 0, folio_size(folio)); > >>>>>>>>>>>> fs/ntfs3/frecord.c: folio_zero_range(folio, 0, > >>>>>>>>>>>> folio_size(folio)); > >>>>>>>>>>>> mm/page_io.c: folio_zero_range(folio, 0, folio_size(folio)); > >>>>>>>>>>>> mm/shmem.c: folio_zero_range(folio, 0, folio_size(folio)); > >>>>>>>>>>>> > >>>>>>>>>>>> > >>>>>>>>>>>>> IOW, what performance testing have you done with this patch? > >>>>>>>>>>>> > >>>>>>>>>>>> No performance test before, but I write a testcase, > >>>>>>>>>>>> > >>>>>>>>>>>> 1) allocate N large folios (folio_alloc(PMD_ORDER)) > >>>>>>>>>>>> 2) then calculate the diff(us) when clear all N folios > >>>>>>>>>>>> clear_highpage/folio_zero_range/folio_zero_user > >>>>>>>>>>>> 3) release N folios > >>>>>>>>>>>> > >>>>>>>>>>>> the result(run 5 times) shown below on my machine, > >>>>>>>>>>>> > >>>>>>>>>>>> N=1, > >>>>>>>>>>>> clear_highpage folio_zero_range folio_zero_user > >>>>>>>>>>>> 1 69 74 177 > >>>>>>>>>>>> 2 57 62 168 > >>>>>>>>>>>> 3 54 58 234 > >>>>>>>>>>>> 4 54 58 157 > >>>>>>>>>>>> 5 56 62 148 > >>>>>>>>>>>> avg 58 62.8 176.8 > >>>>>>>>>>>> > >>>>>>>>>>>> > >>>>>>>>>>>> N=100 > >>>>>>>>>>>> clear_highpage folio_zero_range folio_zero_user > >>>>>>>>>>>> 1 11015 11309 32833 > >>>>>>>>>>>> 2 10385 11110 49751 > >>>>>>>>>>>> 3 10369 11056 33095 > >>>>>>>>>>>> 4 10332 11017 33106 > >>>>>>>>>>>> 5 10483 11000 49032 > >>>>>>>>>>>> avg 10516.8 11098.4 39563.4 > >>>>>>>>>>>> > >>>>>>>>>>>> N=512 > >>>>>>>>>>>> clear_highpage folio_zero_range folio_zero_user > >>>>>>>>>>>> 1 55560 60055 156876 > >>>>>>>>>>>> 2 55485 60024 157132 > >>>>>>>>>>>> 3 55474 60129 156658 > >>>>>>>>>>>> 4 55555 59867 157259 > >>>>>>>>>>>> 5 55528 59932 157108 > >>>>>>>>>>>> avg 55520.4 60001.4 157006.6 > >>>>>>>>>>>> > >>>>>>>>>>>> > >>>>>>>>>>>> > >>>>>>>>>>>> folio_zero_user with many cond_resched(), so time fluctuates a lot, > >>>>>>>>>>>> clear_highpage is better folio_zero_range as you said. > >>>>>>>>>>>> > >>>>>>>>>>>> Maybe add a new helper to convert all folio_zero_range(folio, 0, > >>>>>>>>>>>> folio_size(folio)) > >>>>>>>>>>>> to use clear_highpage + flush_dcache_folio? > >>>>>>>>>>> > >>>>>>>>>>> If this also improves performance for other existing callers of > >>>>>>>>>>> folio_zero_range(), then that's a positive outcome. > >>>>>>>>>> > >>>>> ... > >>>>> > >>>>>>>> hi Kefeng, > >>>>>>>> what's your point? providing a helper like clear_highfolio() or similar? > >>>>>>> > >>>>>>> Yes, from above test, using clear_highpage/flush_dcache_folio is better > >>>>>>> than using folio_zero_range() for folio zero(especially for large > >>>>>>> folio), so I'd like to add a new helper, maybe name it folio_zero() > >>>>>>> since it zero the whole folio. > >>>>>> > >>>>>> we already have a helper like folio_zero_user()? > >>>>>> it is not good enough? > >>>>> > >>>>> Since it is with many cond_resched(), the performance is worst... > >>>> > >>>> Not exactly? It should have zero cost for a preemptible kernel. > >>>> For a non-preemptible kernel, it helps avoid clearing the folio > >>>> from occupying the CPU and starving other processes, right? > >>> > >>> --- a/mm/shmem.c > >>> +++ b/mm/shmem.c > >>> > >>> @@ -2393,10 +2393,7 @@ static int shmem_get_folio_gfp(struct inode > >>> *inode, pgoff_t index, > >>> * it now, lest undo on failure cancel our earlier guarantee. > >>> */ > >>> > >>> if (sgp != SGP_WRITE && !folio_test_uptodate(folio)) { > >>> - long i, n = folio_nr_pages(folio); > >>> - > >>> - for (i = 0; i < n; i++) > >>> - clear_highpage(folio_page(folio, i)); > >>> + folio_zero_user(folio, vmf->address); > >>> flush_dcache_folio(folio); > >>> folio_mark_uptodate(folio); > >>> } > >>> > >>> Do we perform better or worse with the following? > >> > >> Here is for SGP_FALLOC, vmf = NULL, we could use folio_zero_user(folio, > >> 0), I think the performance is worse, will retest once I can access > >> hardware. > > > > Perhaps, since the current code uses clear_hugepage(). Does using > > index << PAGE_SHIFT as the addr_hint offer any benefit? > > > > when use folio_zero_user(), the performance is vary bad with above > fallocate test(mount huge=always), > > folio_zero_range clear_highpage folio_zero_user > real 0m1.214s 0m1.111s 0m3.159s > user 0m0.000s 0m0.000s 0m0.000s > sys 0m1.210s 0m1.109s 0m3.152s > > I tried with addr_hint = 0/index << PAGE_SHIFT, no obvious different. Interesting. Does your kernel have preemption disabled or preemption_debug enabled? If not, it makes me wonder whether folio_zero_user() in alloc_anon_folio() is actually improving performance as expected, compared to the simpler folio_zero() you plan to implement. :-)
On 2024/10/22 4:32, Barry Song wrote: > On Tue, Oct 22, 2024 at 4:33 AM Kefeng Wang <wangkefeng.wang@huawei.com> wrote: >> >> >> >> On 2024/10/21 17:17, Barry Song wrote: >>> On Mon, Oct 21, 2024 at 9:14 PM Kefeng Wang <wangkefeng.wang@huawei.com> wrote: >>>> >>>> >>>> >>>> On 2024/10/21 15:55, Barry Song wrote: >>>>> On Mon, Oct 21, 2024 at 8:47 PM Barry Song <21cnbao@gmail.com> wrote: >>>>>> >>>>>> On Mon, Oct 21, 2024 at 7:09 PM Kefeng Wang <wangkefeng.wang@huawei.com> wrote: >>>>>>> >>>>>>> >>>>>>> >>>>>>> On 2024/10/21 13:38, Barry Song wrote: >>>>>>>> On Mon, Oct 21, 2024 at 6:16 PM Kefeng Wang <wangkefeng.wang@huawei.com> wrote: >>>>>>>>> >>>>>>>>> >>>>>>>>> >>>>>>>>> On 2024/10/21 12:15, Barry Song wrote: >>>>>>>>>> On Fri, Oct 18, 2024 at 8:48 PM Kefeng Wang <wangkefeng.wang@huawei.com> wrote: >>>>>>>>>>> >>>>>>>>>>> >>>>>>>>>>> >>>>>>>>>>> On 2024/10/18 15:32, Kefeng Wang wrote: >>>>>>>>>>>> >>>>>>>>>>>> >>>>>>>>>>>> On 2024/10/18 13:23, Barry Song wrote: >>>>>>>>>>>>> On Fri, Oct 18, 2024 at 6:20 PM Kefeng Wang >>>>>>>>>>>>> <wangkefeng.wang@huawei.com> wrote: >>>>>>>>>>>>>> >>>>>>>>>>>>>> >>>>>>>>>>>>>> >>>>>>>>>>>>>> On 2024/10/17 23:09, Matthew Wilcox wrote: >>>>>>>>>>>>>>> On Thu, Oct 17, 2024 at 10:25:04PM +0800, Kefeng Wang wrote: >>>>>>>>>>>>>>>> Directly use folio_zero_range() to cleanup code. >>>>>>>>>>>>>>> >>>>>>>>>>>>>>> Are you sure there's no performance regression introduced by this? >>>>>>>>>>>>>>> clear_highpage() is often optimised in ways that we can't optimise for >>>>>>>>>>>>>>> a plain memset(). On the other hand, if the folio is large, maybe a >>>>>>>>>>>>>>> modern CPU will be able to do better than clear-one-page-at-a-time. >>>>>>>>>>>>>>> >>>>>>>>>>>>>> >>>>>>>>>>>>>> Right, I missing this, clear_page might be better than memset, I change >>>>>>>>>>>>>> this one when look at the shmem_writepage(), which already convert to >>>>>>>>>>>>>> use folio_zero_range() from clear_highpage(), also I grep >>>>>>>>>>>>>> folio_zero_range(), there are some other to use folio_zero_range(). >>>>>>>>>>>>>> >>>>>>>>>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(folio, 0, >>>>>>>>>>>>>> folio_size(folio)); >>>>>>>>>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(f, >>>>>>>>>>>>>> 0, folio_size(f)); >>>>>>>>>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(f, >>>>>>>>>>>>>> 0, folio_size(f)); >>>>>>>>>>>>>> fs/libfs.c: folio_zero_range(folio, 0, folio_size(folio)); >>>>>>>>>>>>>> fs/ntfs3/frecord.c: folio_zero_range(folio, 0, >>>>>>>>>>>>>> folio_size(folio)); >>>>>>>>>>>>>> mm/page_io.c: folio_zero_range(folio, 0, folio_size(folio)); >>>>>>>>>>>>>> mm/shmem.c: folio_zero_range(folio, 0, folio_size(folio)); >>>>>>>>>>>>>> >>>>>>>>>>>>>> >>>>>>>>>>>>>>> IOW, what performance testing have you done with this patch? >>>>>>>>>>>>>> >>>>>>>>>>>>>> No performance test before, but I write a testcase, >>>>>>>>>>>>>> >>>>>>>>>>>>>> 1) allocate N large folios (folio_alloc(PMD_ORDER)) >>>>>>>>>>>>>> 2) then calculate the diff(us) when clear all N folios >>>>>>>>>>>>>> clear_highpage/folio_zero_range/folio_zero_user >>>>>>>>>>>>>> 3) release N folios >>>>>>>>>>>>>> >>>>>>>>>>>>>> the result(run 5 times) shown below on my machine, >>>>>>>>>>>>>> >>>>>>>>>>>>>> N=1, >>>>>>>>>>>>>> clear_highpage folio_zero_range folio_zero_user >>>>>>>>>>>>>> 1 69 74 177 >>>>>>>>>>>>>> 2 57 62 168 >>>>>>>>>>>>>> 3 54 58 234 >>>>>>>>>>>>>> 4 54 58 157 >>>>>>>>>>>>>> 5 56 62 148 >>>>>>>>>>>>>> avg 58 62.8 176.8 >>>>>>>>>>>>>> >>>>>>>>>>>>>> >>>>>>>>>>>>>> N=100 >>>>>>>>>>>>>> clear_highpage folio_zero_range folio_zero_user >>>>>>>>>>>>>> 1 11015 11309 32833 >>>>>>>>>>>>>> 2 10385 11110 49751 >>>>>>>>>>>>>> 3 10369 11056 33095 >>>>>>>>>>>>>> 4 10332 11017 33106 >>>>>>>>>>>>>> 5 10483 11000 49032 >>>>>>>>>>>>>> avg 10516.8 11098.4 39563.4 >>>>>>>>>>>>>> >>>>>>>>>>>>>> N=512 >>>>>>>>>>>>>> clear_highpage folio_zero_range folio_zero_user >>>>>>>>>>>>>> 1 55560 60055 156876 >>>>>>>>>>>>>> 2 55485 60024 157132 >>>>>>>>>>>>>> 3 55474 60129 156658 >>>>>>>>>>>>>> 4 55555 59867 157259 >>>>>>>>>>>>>> 5 55528 59932 157108 >>>>>>>>>>>>>> avg 55520.4 60001.4 157006.6 >>>>>>>>>>>>>> >>>>>>>>>>>>>> >>>>>>>>>>>>>> >>>>>>>>>>>>>> folio_zero_user with many cond_resched(), so time fluctuates a lot, >>>>>>>>>>>>>> clear_highpage is better folio_zero_range as you said. >>>>>>>>>>>>>> >>>>>>>>>>>>>> Maybe add a new helper to convert all folio_zero_range(folio, 0, >>>>>>>>>>>>>> folio_size(folio)) >>>>>>>>>>>>>> to use clear_highpage + flush_dcache_folio? >>>>>>>>>>>>> >>>>>>>>>>>>> If this also improves performance for other existing callers of >>>>>>>>>>>>> folio_zero_range(), then that's a positive outcome. >>>>>>>>>>>> >>>>>>> ... >>>>>>> >>>>>>>>>> hi Kefeng, >>>>>>>>>> what's your point? providing a helper like clear_highfolio() or similar? >>>>>>>>> >>>>>>>>> Yes, from above test, using clear_highpage/flush_dcache_folio is better >>>>>>>>> than using folio_zero_range() for folio zero(especially for large >>>>>>>>> folio), so I'd like to add a new helper, maybe name it folio_zero() >>>>>>>>> since it zero the whole folio. >>>>>>>> >>>>>>>> we already have a helper like folio_zero_user()? >>>>>>>> it is not good enough? >>>>>>> >>>>>>> Since it is with many cond_resched(), the performance is worst... >>>>>> >>>>>> Not exactly? It should have zero cost for a preemptible kernel. >>>>>> For a non-preemptible kernel, it helps avoid clearing the folio >>>>>> from occupying the CPU and starving other processes, right? >>>>> >>>>> --- a/mm/shmem.c >>>>> +++ b/mm/shmem.c >>>>> >>>>> @@ -2393,10 +2393,7 @@ static int shmem_get_folio_gfp(struct inode >>>>> *inode, pgoff_t index, >>>>> * it now, lest undo on failure cancel our earlier guarantee. >>>>> */ >>>>> >>>>> if (sgp != SGP_WRITE && !folio_test_uptodate(folio)) { >>>>> - long i, n = folio_nr_pages(folio); >>>>> - >>>>> - for (i = 0; i < n; i++) >>>>> - clear_highpage(folio_page(folio, i)); >>>>> + folio_zero_user(folio, vmf->address); >>>>> flush_dcache_folio(folio); >>>>> folio_mark_uptodate(folio); >>>>> } >>>>> >>>>> Do we perform better or worse with the following? >>>> >>>> Here is for SGP_FALLOC, vmf = NULL, we could use folio_zero_user(folio, >>>> 0), I think the performance is worse, will retest once I can access >>>> hardware. >>> >>> Perhaps, since the current code uses clear_hugepage(). Does using >>> index << PAGE_SHIFT as the addr_hint offer any benefit? >>> >> >> when use folio_zero_user(), the performance is vary bad with above >> fallocate test(mount huge=always), >> >> folio_zero_range clear_highpage folio_zero_user >> real 0m1.214s 0m1.111s 0m3.159s >> user 0m0.000s 0m0.000s 0m0.000s >> sys 0m1.210s 0m1.109s 0m3.152s >> >> I tried with addr_hint = 0/index << PAGE_SHIFT, no obvious different. > > Interesting. Does your kernel have preemption disabled or > preemption_debug enabled? ARM64 server, CONFIG_PREEMPT_NONE=y > > If not, it makes me wonder whether folio_zero_user() in > alloc_anon_folio() is actually improving performance as expected, > compared to the simpler folio_zero() you plan to implement. :-) Yes, maybe, the folio_zero_user(was clear_huge_page) is from 47ad8475c000 ("thp: clear_copy_huge_page"), so original clear_huge_page is used in HugeTLB, clear PUD size maybe spend many time, but for PMD or other size of large folio, cond_resched is not necessary since we already have some folio_zero_range() to clear large folio, and no issue was reported.
On Wed, Oct 23, 2024 at 4:10 AM Kefeng Wang <wangkefeng.wang@huawei.com> wrote: > > > > On 2024/10/22 4:32, Barry Song wrote: > > On Tue, Oct 22, 2024 at 4:33 AM Kefeng Wang <wangkefeng.wang@huawei.com> wrote: > >> > >> > >> > >> On 2024/10/21 17:17, Barry Song wrote: > >>> On Mon, Oct 21, 2024 at 9:14 PM Kefeng Wang <wangkefeng.wang@huawei.com> wrote: > >>>> > >>>> > >>>> > >>>> On 2024/10/21 15:55, Barry Song wrote: > >>>>> On Mon, Oct 21, 2024 at 8:47 PM Barry Song <21cnbao@gmail.com> wrote: > >>>>>> > >>>>>> On Mon, Oct 21, 2024 at 7:09 PM Kefeng Wang <wangkefeng.wang@huawei.com> wrote: > >>>>>>> > >>>>>>> > >>>>>>> > >>>>>>> On 2024/10/21 13:38, Barry Song wrote: > >>>>>>>> On Mon, Oct 21, 2024 at 6:16 PM Kefeng Wang <wangkefeng.wang@huawei.com> wrote: > >>>>>>>>> > >>>>>>>>> > >>>>>>>>> > >>>>>>>>> On 2024/10/21 12:15, Barry Song wrote: > >>>>>>>>>> On Fri, Oct 18, 2024 at 8:48 PM Kefeng Wang <wangkefeng.wang@huawei.com> wrote: > >>>>>>>>>>> > >>>>>>>>>>> > >>>>>>>>>>> > >>>>>>>>>>> On 2024/10/18 15:32, Kefeng Wang wrote: > >>>>>>>>>>>> > >>>>>>>>>>>> > >>>>>>>>>>>> On 2024/10/18 13:23, Barry Song wrote: > >>>>>>>>>>>>> On Fri, Oct 18, 2024 at 6:20 PM Kefeng Wang > >>>>>>>>>>>>> <wangkefeng.wang@huawei.com> wrote: > >>>>>>>>>>>>>> > >>>>>>>>>>>>>> > >>>>>>>>>>>>>> > >>>>>>>>>>>>>> On 2024/10/17 23:09, Matthew Wilcox wrote: > >>>>>>>>>>>>>>> On Thu, Oct 17, 2024 at 10:25:04PM +0800, Kefeng Wang wrote: > >>>>>>>>>>>>>>>> Directly use folio_zero_range() to cleanup code. > >>>>>>>>>>>>>>> > >>>>>>>>>>>>>>> Are you sure there's no performance regression introduced by this? > >>>>>>>>>>>>>>> clear_highpage() is often optimised in ways that we can't optimise for > >>>>>>>>>>>>>>> a plain memset(). On the other hand, if the folio is large, maybe a > >>>>>>>>>>>>>>> modern CPU will be able to do better than clear-one-page-at-a-time. > >>>>>>>>>>>>>>> > >>>>>>>>>>>>>> > >>>>>>>>>>>>>> Right, I missing this, clear_page might be better than memset, I change > >>>>>>>>>>>>>> this one when look at the shmem_writepage(), which already convert to > >>>>>>>>>>>>>> use folio_zero_range() from clear_highpage(), also I grep > >>>>>>>>>>>>>> folio_zero_range(), there are some other to use folio_zero_range(). > >>>>>>>>>>>>>> > >>>>>>>>>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(folio, 0, > >>>>>>>>>>>>>> folio_size(folio)); > >>>>>>>>>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(f, > >>>>>>>>>>>>>> 0, folio_size(f)); > >>>>>>>>>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(f, > >>>>>>>>>>>>>> 0, folio_size(f)); > >>>>>>>>>>>>>> fs/libfs.c: folio_zero_range(folio, 0, folio_size(folio)); > >>>>>>>>>>>>>> fs/ntfs3/frecord.c: folio_zero_range(folio, 0, > >>>>>>>>>>>>>> folio_size(folio)); > >>>>>>>>>>>>>> mm/page_io.c: folio_zero_range(folio, 0, folio_size(folio)); > >>>>>>>>>>>>>> mm/shmem.c: folio_zero_range(folio, 0, folio_size(folio)); > >>>>>>>>>>>>>> > >>>>>>>>>>>>>> > >>>>>>>>>>>>>>> IOW, what performance testing have you done with this patch? > >>>>>>>>>>>>>> > >>>>>>>>>>>>>> No performance test before, but I write a testcase, > >>>>>>>>>>>>>> > >>>>>>>>>>>>>> 1) allocate N large folios (folio_alloc(PMD_ORDER)) > >>>>>>>>>>>>>> 2) then calculate the diff(us) when clear all N folios > >>>>>>>>>>>>>> clear_highpage/folio_zero_range/folio_zero_user > >>>>>>>>>>>>>> 3) release N folios > >>>>>>>>>>>>>> > >>>>>>>>>>>>>> the result(run 5 times) shown below on my machine, > >>>>>>>>>>>>>> > >>>>>>>>>>>>>> N=1, > >>>>>>>>>>>>>> clear_highpage folio_zero_range folio_zero_user > >>>>>>>>>>>>>> 1 69 74 177 > >>>>>>>>>>>>>> 2 57 62 168 > >>>>>>>>>>>>>> 3 54 58 234 > >>>>>>>>>>>>>> 4 54 58 157 > >>>>>>>>>>>>>> 5 56 62 148 > >>>>>>>>>>>>>> avg 58 62.8 176.8 > >>>>>>>>>>>>>> > >>>>>>>>>>>>>> > >>>>>>>>>>>>>> N=100 > >>>>>>>>>>>>>> clear_highpage folio_zero_range folio_zero_user > >>>>>>>>>>>>>> 1 11015 11309 32833 > >>>>>>>>>>>>>> 2 10385 11110 49751 > >>>>>>>>>>>>>> 3 10369 11056 33095 > >>>>>>>>>>>>>> 4 10332 11017 33106 > >>>>>>>>>>>>>> 5 10483 11000 49032 > >>>>>>>>>>>>>> avg 10516.8 11098.4 39563.4 > >>>>>>>>>>>>>> > >>>>>>>>>>>>>> N=512 > >>>>>>>>>>>>>> clear_highpage folio_zero_range folio_zero_user > >>>>>>>>>>>>>> 1 55560 60055 156876 > >>>>>>>>>>>>>> 2 55485 60024 157132 > >>>>>>>>>>>>>> 3 55474 60129 156658 > >>>>>>>>>>>>>> 4 55555 59867 157259 > >>>>>>>>>>>>>> 5 55528 59932 157108 > >>>>>>>>>>>>>> avg 55520.4 60001.4 157006.6 > >>>>>>>>>>>>>> > >>>>>>>>>>>>>> > >>>>>>>>>>>>>> > >>>>>>>>>>>>>> folio_zero_user with many cond_resched(), so time fluctuates a lot, > >>>>>>>>>>>>>> clear_highpage is better folio_zero_range as you said. > >>>>>>>>>>>>>> > >>>>>>>>>>>>>> Maybe add a new helper to convert all folio_zero_range(folio, 0, > >>>>>>>>>>>>>> folio_size(folio)) > >>>>>>>>>>>>>> to use clear_highpage + flush_dcache_folio? > >>>>>>>>>>>>> > >>>>>>>>>>>>> If this also improves performance for other existing callers of > >>>>>>>>>>>>> folio_zero_range(), then that's a positive outcome. > >>>>>>>>>>>> > >>>>>>> ... > >>>>>>> > >>>>>>>>>> hi Kefeng, > >>>>>>>>>> what's your point? providing a helper like clear_highfolio() or similar? > >>>>>>>>> > >>>>>>>>> Yes, from above test, using clear_highpage/flush_dcache_folio is better > >>>>>>>>> than using folio_zero_range() for folio zero(especially for large > >>>>>>>>> folio), so I'd like to add a new helper, maybe name it folio_zero() > >>>>>>>>> since it zero the whole folio. > >>>>>>>> > >>>>>>>> we already have a helper like folio_zero_user()? > >>>>>>>> it is not good enough? > >>>>>>> > >>>>>>> Since it is with many cond_resched(), the performance is worst... > >>>>>> > >>>>>> Not exactly? It should have zero cost for a preemptible kernel. > >>>>>> For a non-preemptible kernel, it helps avoid clearing the folio > >>>>>> from occupying the CPU and starving other processes, right? > >>>>> > >>>>> --- a/mm/shmem.c > >>>>> +++ b/mm/shmem.c > >>>>> > >>>>> @@ -2393,10 +2393,7 @@ static int shmem_get_folio_gfp(struct inode > >>>>> *inode, pgoff_t index, > >>>>> * it now, lest undo on failure cancel our earlier guarantee. > >>>>> */ > >>>>> > >>>>> if (sgp != SGP_WRITE && !folio_test_uptodate(folio)) { > >>>>> - long i, n = folio_nr_pages(folio); > >>>>> - > >>>>> - for (i = 0; i < n; i++) > >>>>> - clear_highpage(folio_page(folio, i)); > >>>>> + folio_zero_user(folio, vmf->address); > >>>>> flush_dcache_folio(folio); > >>>>> folio_mark_uptodate(folio); > >>>>> } > >>>>> > >>>>> Do we perform better or worse with the following? > >>>> > >>>> Here is for SGP_FALLOC, vmf = NULL, we could use folio_zero_user(folio, > >>>> 0), I think the performance is worse, will retest once I can access > >>>> hardware. > >>> > >>> Perhaps, since the current code uses clear_hugepage(). Does using > >>> index << PAGE_SHIFT as the addr_hint offer any benefit? > >>> > >> > >> when use folio_zero_user(), the performance is vary bad with above > >> fallocate test(mount huge=always), > >> > >> folio_zero_range clear_highpage folio_zero_user > >> real 0m1.214s 0m1.111s 0m3.159s > >> user 0m0.000s 0m0.000s 0m0.000s > >> sys 0m1.210s 0m1.109s 0m3.152s > >> > >> I tried with addr_hint = 0/index << PAGE_SHIFT, no obvious different. > > > > Interesting. Does your kernel have preemption disabled or > > preemption_debug enabled? > > ARM64 server, CONFIG_PREEMPT_NONE=y this explains why the performance is much worse. > > > > > If not, it makes me wonder whether folio_zero_user() in > > alloc_anon_folio() is actually improving performance as expected, > > compared to the simpler folio_zero() you plan to implement. :-) > > Yes, maybe, the folio_zero_user(was clear_huge_page) is from > 47ad8475c000 ("thp: clear_copy_huge_page"), so original clear_huge_page > is used in HugeTLB, clear PUD size maybe spend many time, but for PMD or > other size of large folio, cond_resched is not necessary since we > already have some folio_zero_range() to clear large folio, and no issue > was reported. probably worth an optimization. calling cond_resched() for each page seems too aggressive and useless. diff --git a/mm/memory.c b/mm/memory.c index 0f614523b9f4..5fc38347d782 100644 --- a/mm/memory.c +++ b/mm/memory.c @@ -6738,6 +6738,19 @@ EXPORT_SYMBOL(__might_fault); #endif #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS) +/* + * To prevent process_huge_page() from starving other processes, + * we allow other processes a chance with each batch. + */ +static inline void batched_cond_resched(int *nr) +{ +#define BATCHED_PROCESS_NR 64 + if (*nr++ < BATCHED_PROCESS_NR) + return; + cond_resched(); + *nr = 0; +} + /* * Process all subpages of the specified huge page with the specified * operation. The target subpage will be processed last to keep its @@ -6748,7 +6761,7 @@ static inline int process_huge_page( int (*process_subpage)(unsigned long addr, int idx, void *arg), void *arg) { - int i, n, base, l, ret; + int i, n, base, l, ret, processed_nr = 0; unsigned long addr = addr_hint & ~(((unsigned long)nr_pages << PAGE_SHIFT) - 1); @@ -6761,7 +6774,7 @@ static inline int process_huge_page( l = n; /* Process subpages at the end of huge page */ for (i = nr_pages - 1; i >= 2 * n; i--) { - cond_resched(); + batched_cond_resched(&processed_nr); ret = process_subpage(addr + i * PAGE_SIZE, i, arg); if (ret) return ret; @@ -6772,7 +6785,7 @@ static inline int process_huge_page( l = nr_pages - n; /* Process subpages at the begin of huge page */ for (i = 0; i < base; i++) { - cond_resched(); + batched_cond_resched(&processed_nr); ret = process_subpage(addr + i * PAGE_SIZE, i, arg); if (ret) return ret; @@ -6786,11 +6799,11 @@ static inline int process_huge_page( int left_idx = base + i; int right_idx = base + 2 * l - 1 - i; - cond_resched(); + batched_cond_resched(&processed_nr); ret = process_subpage(addr + left_idx * PAGE_SIZE, left_idx, arg); if (ret) return ret; - cond_resched(); + batched_cond_resched(&processed_nr); ret = process_subpage(addr + right_idx * PAGE_SIZE, right_idx, arg); if (ret) return ret;
+CC Huang Ying, On 2024/10/23 6:56, Barry Song wrote: > On Wed, Oct 23, 2024 at 4:10 AM Kefeng Wang <wangkefeng.wang@huawei.com> wrote: >> >> ... >>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>> On 2024/10/17 23:09, Matthew Wilcox wrote: >>>>>>>>>>>>>>>>> On Thu, Oct 17, 2024 at 10:25:04PM +0800, Kefeng Wang wrote: >>>>>>>>>>>>>>>>>> Directly use folio_zero_range() to cleanup code. >>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>> Are you sure there's no performance regression introduced by this? >>>>>>>>>>>>>>>>> clear_highpage() is often optimised in ways that we can't optimise for >>>>>>>>>>>>>>>>> a plain memset(). On the other hand, if the folio is large, maybe a >>>>>>>>>>>>>>>>> modern CPU will be able to do better than clear-one-page-at-a-time. >>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>> Right, I missing this, clear_page might be better than memset, I change >>>>>>>>>>>>>>>> this one when look at the shmem_writepage(), which already convert to >>>>>>>>>>>>>>>> use folio_zero_range() from clear_highpage(), also I grep >>>>>>>>>>>>>>>> folio_zero_range(), there are some other to use folio_zero_range(). >>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(folio, 0, >>>>>>>>>>>>>>>> folio_size(folio)); >>>>>>>>>>>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(f, >>>>>>>>>>>>>>>> 0, folio_size(f)); >>>>>>>>>>>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(f, >>>>>>>>>>>>>>>> 0, folio_size(f)); >>>>>>>>>>>>>>>> fs/libfs.c: folio_zero_range(folio, 0, folio_size(folio)); >>>>>>>>>>>>>>>> fs/ntfs3/frecord.c: folio_zero_range(folio, 0, >>>>>>>>>>>>>>>> folio_size(folio)); >>>>>>>>>>>>>>>> mm/page_io.c: folio_zero_range(folio, 0, folio_size(folio)); >>>>>>>>>>>>>>>> mm/shmem.c: folio_zero_range(folio, 0, folio_size(folio)); >>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>> IOW, what performance testing have you done with this patch? >>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>> No performance test before, but I write a testcase, >>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>> 1) allocate N large folios (folio_alloc(PMD_ORDER)) >>>>>>>>>>>>>>>> 2) then calculate the diff(us) when clear all N folios >>>>>>>>>>>>>>>> clear_highpage/folio_zero_range/folio_zero_user >>>>>>>>>>>>>>>> 3) release N folios >>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>> the result(run 5 times) shown below on my machine, >>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>> N=1, >>>>>>>>>>>>>>>> clear_highpage folio_zero_range folio_zero_user >>>>>>>>>>>>>>>> 1 69 74 177 >>>>>>>>>>>>>>>> 2 57 62 168 >>>>>>>>>>>>>>>> 3 54 58 234 >>>>>>>>>>>>>>>> 4 54 58 157 >>>>>>>>>>>>>>>> 5 56 62 148 >>>>>>>>>>>>>>>> avg 58 62.8 176.8 >>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>> N=100 >>>>>>>>>>>>>>>> clear_highpage folio_zero_range folio_zero_user >>>>>>>>>>>>>>>> 1 11015 11309 32833 >>>>>>>>>>>>>>>> 2 10385 11110 49751 >>>>>>>>>>>>>>>> 3 10369 11056 33095 >>>>>>>>>>>>>>>> 4 10332 11017 33106 >>>>>>>>>>>>>>>> 5 10483 11000 49032 >>>>>>>>>>>>>>>> avg 10516.8 11098.4 39563.4 >>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>> N=512 >>>>>>>>>>>>>>>> clear_highpage folio_zero_range folio_zero_user >>>>>>>>>>>>>>>> 1 55560 60055 156876 >>>>>>>>>>>>>>>> 2 55485 60024 157132 >>>>>>>>>>>>>>>> 3 55474 60129 156658 >>>>>>>>>>>>>>>> 4 55555 59867 157259 >>>>>>>>>>>>>>>> 5 55528 59932 157108 >>>>>>>>>>>>>>>> avg 55520.4 60001.4 157006.6 >>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>> folio_zero_user with many cond_resched(), so time fluctuates a lot, >>>>>>>>>>>>>>>> clear_highpage is better folio_zero_range as you said. >>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>> Maybe add a new helper to convert all folio_zero_range(folio, 0, >>>>>>>>>>>>>>>> folio_size(folio)) >>>>>>>>>>>>>>>> to use clear_highpage + flush_dcache_folio? >>>>>>>>>>>>>>> >>>>>>>>>>>>>>> If this also improves performance for other existing callers of >>>>>>>>>>>>>>> folio_zero_range(), then that's a positive outcome. >>>>>>>>>>>>>> >>>>>>>>> ... >>>>>>>>> >>>>>>>>>>>> hi Kefeng, >>>>>>>>>>>> what's your point? providing a helper like clear_highfolio() or similar? >>>>>>>>>>> >>>>>>>>>>> Yes, from above test, using clear_highpage/flush_dcache_folio is better >>>>>>>>>>> than using folio_zero_range() for folio zero(especially for large >>>>>>>>>>> folio), so I'd like to add a new helper, maybe name it folio_zero() >>>>>>>>>>> since it zero the whole folio. >>>>>>>>>> >>>>>>>>>> we already have a helper like folio_zero_user()? >>>>>>>>>> it is not good enough? >>>>>>>>> >>>>>>>>> Since it is with many cond_resched(), the performance is worst... >>>>>>>> >>>>>>>> Not exactly? It should have zero cost for a preemptible kernel. >>>>>>>> For a non-preemptible kernel, it helps avoid clearing the folio >>>>>>>> from occupying the CPU and starving other processes, right? >>>>>>> >>>>>>> --- a/mm/shmem.c >>>>>>> +++ b/mm/shmem.c >>>>>>> >>>>>>> @@ -2393,10 +2393,7 @@ static int shmem_get_folio_gfp(struct inode >>>>>>> *inode, pgoff_t index, >>>>>>> * it now, lest undo on failure cancel our earlier guarantee. >>>>>>> */ >>>>>>> >>>>>>> if (sgp != SGP_WRITE && !folio_test_uptodate(folio)) { >>>>>>> - long i, n = folio_nr_pages(folio); >>>>>>> - >>>>>>> - for (i = 0; i < n; i++) >>>>>>> - clear_highpage(folio_page(folio, i)); >>>>>>> + folio_zero_user(folio, vmf->address); >>>>>>> flush_dcache_folio(folio); >>>>>>> folio_mark_uptodate(folio); >>>>>>> } >>>>>>> >>>>>>> Do we perform better or worse with the following? >>>>>> >>>>>> Here is for SGP_FALLOC, vmf = NULL, we could use folio_zero_user(folio, >>>>>> 0), I think the performance is worse, will retest once I can access >>>>>> hardware. >>>>> >>>>> Perhaps, since the current code uses clear_hugepage(). Does using >>>>> index << PAGE_SHIFT as the addr_hint offer any benefit? >>>>> >>>> >>>> when use folio_zero_user(), the performance is vary bad with above >>>> fallocate test(mount huge=always), >>>> >>>> folio_zero_range clear_highpage folio_zero_user >>>> real 0m1.214s 0m1.111s 0m3.159s >>>> user 0m0.000s 0m0.000s 0m0.000s >>>> sys 0m1.210s 0m1.109s 0m3.152s >>>> >>>> I tried with addr_hint = 0/index << PAGE_SHIFT, no obvious different. >>> >>> Interesting. Does your kernel have preemption disabled or >>> preemption_debug enabled? >> >> ARM64 server, CONFIG_PREEMPT_NONE=y > > this explains why the performance is much worse. > >> >>> >>> If not, it makes me wonder whether folio_zero_user() in >>> alloc_anon_folio() is actually improving performance as expected, >>> compared to the simpler folio_zero() you plan to implement. :-) >> >> Yes, maybe, the folio_zero_user(was clear_huge_page) is from >> 47ad8475c000 ("thp: clear_copy_huge_page"), so original clear_huge_page >> is used in HugeTLB, clear PUD size maybe spend many time, but for PMD or >> other size of large folio, cond_resched is not necessary since we >> already have some folio_zero_range() to clear large folio, and no issue >> was reported. > > probably worth an optimization. calling cond_resched() for each page > seems too aggressive and useless. After some test, I think the cond_resched() is not the root cause, no performance gained with batched cond_resched(), even I kill cond_resched() from process_huge_page, no improvement. But when I unconditionally use clear_gigantic_page() in folio_zero_user(patched), there is big improvement with above fallocate on tmpfs(mount huge=always), also I test some other testcase, 1) case-anon-w-seq-mt: (2M PMD THP) base: real 0m2.490s 0m2.254s 0m2.272s user 1m59.980s 2m23.431s 2m18.739s sys 1m3.675s 1m15.462s 1m15.030s patched: real 0m2.234s 0m2.225s 0m2.159s user 2m56.105s 2m57.117s 3m0.489s sys 0m17.064s 0m17.564s 0m16.150s Patched kernel win on sys and bad in user, but real is almost same, maybe a little better than base. 2) case-anon-w-seq-hugetlb:(2M PMD HugeTLB) base: real 0m5.175s 0m5.117s 0m4.856s user 5m15.943s 5m7.567s 4m29.273s sys 2m38.503s 2m21.949s 2m21.252s patched: real 0m4.966s 0m4.841s 0m4.561s user 6m30.123s 6m9.516s 5m49.733s sys 0m58.503s 0m47.847s 0m46.785s This case is similar to the case1. 3) fallocate hugetlb 20G (2M PMD HugeTLB) base: real 0m3.016s 0m3.019s 0m3.018s user 0m0.000s 0m0.000s 0m0.000s sys 0m3.009s 0m3.012s 0m3.010s patched: real 0m1.136s 0m1.136s 0m1.136s user 0m0.000s 0m0.000s 0m0.004s sys 0m1.133s 0m1.133s 0m1.129s There is big win on patched kernel, and it is similar to above tmpfs test, so maybe we could revert the commit c79b57e462b5 ("mm: hugetlb: clear target sub-page last when clearing huge page").
Hi, Kefeng, Kefeng Wang <wangkefeng.wang@huawei.com> writes: > +CC Huang Ying, > > On 2024/10/23 6:56, Barry Song wrote: >> On Wed, Oct 23, 2024 at 4:10 AM Kefeng Wang <wangkefeng.wang@huawei.com> wrote: >>> >>> > ... >>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>> On 2024/10/17 23:09, Matthew Wilcox wrote: >>>>>>>>>>>>>>>>>> On Thu, Oct 17, 2024 at 10:25:04PM +0800, Kefeng Wang wrote: >>>>>>>>>>>>>>>>>>> Directly use folio_zero_range() to cleanup code. >>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>> Are you sure there's no performance regression introduced by this? >>>>>>>>>>>>>>>>>> clear_highpage() is often optimised in ways that we can't optimise for >>>>>>>>>>>>>>>>>> a plain memset(). On the other hand, if the folio is large, maybe a >>>>>>>>>>>>>>>>>> modern CPU will be able to do better than clear-one-page-at-a-time. >>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>> Right, I missing this, clear_page might be better than memset, I change >>>>>>>>>>>>>>>>> this one when look at the shmem_writepage(), which already convert to >>>>>>>>>>>>>>>>> use folio_zero_range() from clear_highpage(), also I grep >>>>>>>>>>>>>>>>> folio_zero_range(), there are some other to use folio_zero_range(). >>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(folio, 0, >>>>>>>>>>>>>>>>> folio_size(folio)); >>>>>>>>>>>>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(f, >>>>>>>>>>>>>>>>> 0, folio_size(f)); >>>>>>>>>>>>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(f, >>>>>>>>>>>>>>>>> 0, folio_size(f)); >>>>>>>>>>>>>>>>> fs/libfs.c: folio_zero_range(folio, 0, folio_size(folio)); >>>>>>>>>>>>>>>>> fs/ntfs3/frecord.c: folio_zero_range(folio, 0, >>>>>>>>>>>>>>>>> folio_size(folio)); >>>>>>>>>>>>>>>>> mm/page_io.c: folio_zero_range(folio, 0, folio_size(folio)); >>>>>>>>>>>>>>>>> mm/shmem.c: folio_zero_range(folio, 0, folio_size(folio)); >>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>> IOW, what performance testing have you done with this patch? >>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>> No performance test before, but I write a testcase, >>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>> 1) allocate N large folios (folio_alloc(PMD_ORDER)) >>>>>>>>>>>>>>>>> 2) then calculate the diff(us) when clear all N folios >>>>>>>>>>>>>>>>> clear_highpage/folio_zero_range/folio_zero_user >>>>>>>>>>>>>>>>> 3) release N folios >>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>> the result(run 5 times) shown below on my machine, >>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>> N=1, >>>>>>>>>>>>>>>>> clear_highpage folio_zero_range folio_zero_user >>>>>>>>>>>>>>>>> 1 69 74 177 >>>>>>>>>>>>>>>>> 2 57 62 168 >>>>>>>>>>>>>>>>> 3 54 58 234 >>>>>>>>>>>>>>>>> 4 54 58 157 >>>>>>>>>>>>>>>>> 5 56 62 148 >>>>>>>>>>>>>>>>> avg 58 62.8 176.8 >>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>> N=100 >>>>>>>>>>>>>>>>> clear_highpage folio_zero_range folio_zero_user >>>>>>>>>>>>>>>>> 1 11015 11309 32833 >>>>>>>>>>>>>>>>> 2 10385 11110 49751 >>>>>>>>>>>>>>>>> 3 10369 11056 33095 >>>>>>>>>>>>>>>>> 4 10332 11017 33106 >>>>>>>>>>>>>>>>> 5 10483 11000 49032 >>>>>>>>>>>>>>>>> avg 10516.8 11098.4 39563.4 >>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>> N=512 >>>>>>>>>>>>>>>>> clear_highpage folio_zero_range folio_zero_user >>>>>>>>>>>>>>>>> 1 55560 60055 156876 >>>>>>>>>>>>>>>>> 2 55485 60024 157132 >>>>>>>>>>>>>>>>> 3 55474 60129 156658 >>>>>>>>>>>>>>>>> 4 55555 59867 157259 >>>>>>>>>>>>>>>>> 5 55528 59932 157108 >>>>>>>>>>>>>>>>> avg 55520.4 60001.4 157006.6 >>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>> folio_zero_user with many cond_resched(), so time fluctuates a lot, >>>>>>>>>>>>>>>>> clear_highpage is better folio_zero_range as you said. >>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>> Maybe add a new helper to convert all folio_zero_range(folio, 0, >>>>>>>>>>>>>>>>> folio_size(folio)) >>>>>>>>>>>>>>>>> to use clear_highpage + flush_dcache_folio? >>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>> If this also improves performance for other existing callers of >>>>>>>>>>>>>>>> folio_zero_range(), then that's a positive outcome. >>>>>>>>>>>>>>> >>>>>>>>>> ... >>>>>>>>>> >>>>>>>>>>>>> hi Kefeng, >>>>>>>>>>>>> what's your point? providing a helper like clear_highfolio() or similar? >>>>>>>>>>>> >>>>>>>>>>>> Yes, from above test, using clear_highpage/flush_dcache_folio is better >>>>>>>>>>>> than using folio_zero_range() for folio zero(especially for large >>>>>>>>>>>> folio), so I'd like to add a new helper, maybe name it folio_zero() >>>>>>>>>>>> since it zero the whole folio. >>>>>>>>>>> >>>>>>>>>>> we already have a helper like folio_zero_user()? >>>>>>>>>>> it is not good enough? >>>>>>>>>> >>>>>>>>>> Since it is with many cond_resched(), the performance is worst... >>>>>>>>> >>>>>>>>> Not exactly? It should have zero cost for a preemptible kernel. >>>>>>>>> For a non-preemptible kernel, it helps avoid clearing the folio >>>>>>>>> from occupying the CPU and starving other processes, right? >>>>>>>> >>>>>>>> --- a/mm/shmem.c >>>>>>>> +++ b/mm/shmem.c >>>>>>>> >>>>>>>> @@ -2393,10 +2393,7 @@ static int shmem_get_folio_gfp(struct inode >>>>>>>> *inode, pgoff_t index, >>>>>>>> * it now, lest undo on failure cancel our earlier guarantee. >>>>>>>> */ >>>>>>>> >>>>>>>> if (sgp != SGP_WRITE && !folio_test_uptodate(folio)) { >>>>>>>> - long i, n = folio_nr_pages(folio); >>>>>>>> - >>>>>>>> - for (i = 0; i < n; i++) >>>>>>>> - clear_highpage(folio_page(folio, i)); >>>>>>>> + folio_zero_user(folio, vmf->address); >>>>>>>> flush_dcache_folio(folio); >>>>>>>> folio_mark_uptodate(folio); >>>>>>>> } >>>>>>>> >>>>>>>> Do we perform better or worse with the following? >>>>>>> >>>>>>> Here is for SGP_FALLOC, vmf = NULL, we could use folio_zero_user(folio, >>>>>>> 0), I think the performance is worse, will retest once I can access >>>>>>> hardware. >>>>>> >>>>>> Perhaps, since the current code uses clear_hugepage(). Does using >>>>>> index << PAGE_SHIFT as the addr_hint offer any benefit? >>>>>> >>>>> >>>>> when use folio_zero_user(), the performance is vary bad with above >>>>> fallocate test(mount huge=always), >>>>> >>>>> folio_zero_range clear_highpage folio_zero_user >>>>> real 0m1.214s 0m1.111s 0m3.159s >>>>> user 0m0.000s 0m0.000s 0m0.000s >>>>> sys 0m1.210s 0m1.109s 0m3.152s >>>>> >>>>> I tried with addr_hint = 0/index << PAGE_SHIFT, no obvious different. >>>> >>>> Interesting. Does your kernel have preemption disabled or >>>> preemption_debug enabled? >>> >>> ARM64 server, CONFIG_PREEMPT_NONE=y >> this explains why the performance is much worse. >> >>> >>>> >>>> If not, it makes me wonder whether folio_zero_user() in >>>> alloc_anon_folio() is actually improving performance as expected, >>>> compared to the simpler folio_zero() you plan to implement. :-) >>> >>> Yes, maybe, the folio_zero_user(was clear_huge_page) is from >>> 47ad8475c000 ("thp: clear_copy_huge_page"), so original clear_huge_page >>> is used in HugeTLB, clear PUD size maybe spend many time, but for PMD or >>> other size of large folio, cond_resched is not necessary since we >>> already have some folio_zero_range() to clear large folio, and no issue >>> was reported. >> probably worth an optimization. calling cond_resched() for each page >> seems too aggressive and useless. > > After some test, I think the cond_resched() is not the root cause, > no performance gained with batched cond_resched(), even I kill > cond_resched() from process_huge_page, no improvement. > > But when I unconditionally use clear_gigantic_page() in > folio_zero_user(patched), there is big improvement with above > fallocate on tmpfs(mount huge=always), also I test some other testcase, > > > 1) case-anon-w-seq-mt: (2M PMD THP) > > base: > real 0m2.490s 0m2.254s 0m2.272s > user 1m59.980s 2m23.431s 2m18.739s > sys 1m3.675s 1m15.462s 1m15.030s > > patched: > real 0m2.234s 0m2.225s 0m2.159s > user 2m56.105s 2m57.117s 3m0.489s > sys 0m17.064s 0m17.564s 0m16.150s > > Patched kernel win on sys and bad in user, but real is almost same, > maybe a little better than base. We can find user time difference. That means the original cache hot behavior still applies on your system. However, it appears that the performance to clear page from end to begin is really bad on your system. So, I suggest to revise the current implementation to use sequential clearing as much as possible. > 2) case-anon-w-seq-hugetlb:(2M PMD HugeTLB) > > base: > real 0m5.175s 0m5.117s 0m4.856s > user 5m15.943s 5m7.567s 4m29.273s > sys 2m38.503s 2m21.949s 2m21.252s > > patched: > real 0m4.966s 0m4.841s 0m4.561s > user 6m30.123s 6m9.516s 5m49.733s > sys 0m58.503s 0m47.847s 0m46.785s > > > This case is similar to the case1. > > 3) fallocate hugetlb 20G (2M PMD HugeTLB) > > base: > real 0m3.016s 0m3.019s 0m3.018s > user 0m0.000s 0m0.000s 0m0.000s > sys 0m3.009s 0m3.012s 0m3.010s > > patched: > > real 0m1.136s 0m1.136s 0m1.136s > user 0m0.000s 0m0.000s 0m0.004s > sys 0m1.133s 0m1.133s 0m1.129s > > > There is big win on patched kernel, and it is similar to above tmpfs > test, so maybe we could revert the commit c79b57e462b5 ("mm: hugetlb: > clear target sub-page last when clearing huge page"). -- Best Regards, Huang, Ying
On 2024/10/25 10:59, Huang, Ying wrote: > Hi, Kefeng, > > Kefeng Wang <wangkefeng.wang@huawei.com> writes: > >> +CC Huang Ying, >> >> On 2024/10/23 6:56, Barry Song wrote: >>> On Wed, Oct 23, 2024 at 4:10 AM Kefeng Wang <wangkefeng.wang@huawei.com> wrote: >>>> >>>> >> ... >>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>> On 2024/10/17 23:09, Matthew Wilcox wrote: >>>>>>>>>>>>>>>>>>> On Thu, Oct 17, 2024 at 10:25:04PM +0800, Kefeng Wang wrote: >>>>>>>>>>>>>>>>>>>> Directly use folio_zero_range() to cleanup code. >>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>> Are you sure there's no performance regression introduced by this? >>>>>>>>>>>>>>>>>>> clear_highpage() is often optimised in ways that we can't optimise for >>>>>>>>>>>>>>>>>>> a plain memset(). On the other hand, if the folio is large, maybe a >>>>>>>>>>>>>>>>>>> modern CPU will be able to do better than clear-one-page-at-a-time. >>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>> Right, I missing this, clear_page might be better than memset, I change >>>>>>>>>>>>>>>>>> this one when look at the shmem_writepage(), which already convert to >>>>>>>>>>>>>>>>>> use folio_zero_range() from clear_highpage(), also I grep >>>>>>>>>>>>>>>>>> folio_zero_range(), there are some other to use folio_zero_range(). >>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(folio, 0, >>>>>>>>>>>>>>>>>> folio_size(folio)); >>>>>>>>>>>>>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(f, >>>>>>>>>>>>>>>>>> 0, folio_size(f)); >>>>>>>>>>>>>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(f, >>>>>>>>>>>>>>>>>> 0, folio_size(f)); >>>>>>>>>>>>>>>>>> fs/libfs.c: folio_zero_range(folio, 0, folio_size(folio)); >>>>>>>>>>>>>>>>>> fs/ntfs3/frecord.c: folio_zero_range(folio, 0, >>>>>>>>>>>>>>>>>> folio_size(folio)); >>>>>>>>>>>>>>>>>> mm/page_io.c: folio_zero_range(folio, 0, folio_size(folio)); >>>>>>>>>>>>>>>>>> mm/shmem.c: folio_zero_range(folio, 0, folio_size(folio)); >>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>> IOW, what performance testing have you done with this patch? >>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>> No performance test before, but I write a testcase, >>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>> 1) allocate N large folios (folio_alloc(PMD_ORDER)) >>>>>>>>>>>>>>>>>> 2) then calculate the diff(us) when clear all N folios >>>>>>>>>>>>>>>>>> clear_highpage/folio_zero_range/folio_zero_user >>>>>>>>>>>>>>>>>> 3) release N folios >>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>> the result(run 5 times) shown below on my machine, >>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>> N=1, >>>>>>>>>>>>>>>>>> clear_highpage folio_zero_range folio_zero_user >>>>>>>>>>>>>>>>>> 1 69 74 177 >>>>>>>>>>>>>>>>>> 2 57 62 168 >>>>>>>>>>>>>>>>>> 3 54 58 234 >>>>>>>>>>>>>>>>>> 4 54 58 157 >>>>>>>>>>>>>>>>>> 5 56 62 148 >>>>>>>>>>>>>>>>>> avg 58 62.8 176.8 >>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>> N=100 >>>>>>>>>>>>>>>>>> clear_highpage folio_zero_range folio_zero_user >>>>>>>>>>>>>>>>>> 1 11015 11309 32833 >>>>>>>>>>>>>>>>>> 2 10385 11110 49751 >>>>>>>>>>>>>>>>>> 3 10369 11056 33095 >>>>>>>>>>>>>>>>>> 4 10332 11017 33106 >>>>>>>>>>>>>>>>>> 5 10483 11000 49032 >>>>>>>>>>>>>>>>>> avg 10516.8 11098.4 39563.4 >>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>> N=512 >>>>>>>>>>>>>>>>>> clear_highpage folio_zero_range folio_zero_user >>>>>>>>>>>>>>>>>> 1 55560 60055 156876 >>>>>>>>>>>>>>>>>> 2 55485 60024 157132 >>>>>>>>>>>>>>>>>> 3 55474 60129 156658 >>>>>>>>>>>>>>>>>> 4 55555 59867 157259 >>>>>>>>>>>>>>>>>> 5 55528 59932 157108 >>>>>>>>>>>>>>>>>> avg 55520.4 60001.4 157006.6 >>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>> folio_zero_user with many cond_resched(), so time fluctuates a lot, >>>>>>>>>>>>>>>>>> clear_highpage is better folio_zero_range as you said. >>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>> Maybe add a new helper to convert all folio_zero_range(folio, 0, >>>>>>>>>>>>>>>>>> folio_size(folio)) >>>>>>>>>>>>>>>>>> to use clear_highpage + flush_dcache_folio? >>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>> If this also improves performance for other existing callers of >>>>>>>>>>>>>>>>> folio_zero_range(), then that's a positive outcome. >>>>>>>>>>>>>>>> >>>>>>>>>>> ... >>>>>>>>>>> >>>>>>>>>>>>>> hi Kefeng, >>>>>>>>>>>>>> what's your point? providing a helper like clear_highfolio() or similar? >>>>>>>>>>>>> >>>>>>>>>>>>> Yes, from above test, using clear_highpage/flush_dcache_folio is better >>>>>>>>>>>>> than using folio_zero_range() for folio zero(especially for large >>>>>>>>>>>>> folio), so I'd like to add a new helper, maybe name it folio_zero() >>>>>>>>>>>>> since it zero the whole folio. >>>>>>>>>>>> >>>>>>>>>>>> we already have a helper like folio_zero_user()? >>>>>>>>>>>> it is not good enough? >>>>>>>>>>> >>>>>>>>>>> Since it is with many cond_resched(), the performance is worst... >>>>>>>>>> >>>>>>>>>> Not exactly? It should have zero cost for a preemptible kernel. >>>>>>>>>> For a non-preemptible kernel, it helps avoid clearing the folio >>>>>>>>>> from occupying the CPU and starving other processes, right? >>>>>>>>> >>>>>>>>> --- a/mm/shmem.c >>>>>>>>> +++ b/mm/shmem.c >>>>>>>>> >>>>>>>>> @@ -2393,10 +2393,7 @@ static int shmem_get_folio_gfp(struct inode >>>>>>>>> *inode, pgoff_t index, >>>>>>>>> * it now, lest undo on failure cancel our earlier guarantee. >>>>>>>>> */ >>>>>>>>> >>>>>>>>> if (sgp != SGP_WRITE && !folio_test_uptodate(folio)) { >>>>>>>>> - long i, n = folio_nr_pages(folio); >>>>>>>>> - >>>>>>>>> - for (i = 0; i < n; i++) >>>>>>>>> - clear_highpage(folio_page(folio, i)); >>>>>>>>> + folio_zero_user(folio, vmf->address); >>>>>>>>> flush_dcache_folio(folio); >>>>>>>>> folio_mark_uptodate(folio); >>>>>>>>> } >>>>>>>>> >>>>>>>>> Do we perform better or worse with the following? >>>>>>>> >>>>>>>> Here is for SGP_FALLOC, vmf = NULL, we could use folio_zero_user(folio, >>>>>>>> 0), I think the performance is worse, will retest once I can access >>>>>>>> hardware. >>>>>>> >>>>>>> Perhaps, since the current code uses clear_hugepage(). Does using >>>>>>> index << PAGE_SHIFT as the addr_hint offer any benefit? >>>>>>> >>>>>> >>>>>> when use folio_zero_user(), the performance is vary bad with above >>>>>> fallocate test(mount huge=always), >>>>>> >>>>>> folio_zero_range clear_highpage folio_zero_user >>>>>> real 0m1.214s 0m1.111s 0m3.159s >>>>>> user 0m0.000s 0m0.000s 0m0.000s >>>>>> sys 0m1.210s 0m1.109s 0m3.152s >>>>>> >>>>>> I tried with addr_hint = 0/index << PAGE_SHIFT, no obvious different. >>>>> >>>>> Interesting. Does your kernel have preemption disabled or >>>>> preemption_debug enabled? >>>> >>>> ARM64 server, CONFIG_PREEMPT_NONE=y >>> this explains why the performance is much worse. >>> >>>> >>>>> >>>>> If not, it makes me wonder whether folio_zero_user() in >>>>> alloc_anon_folio() is actually improving performance as expected, >>>>> compared to the simpler folio_zero() you plan to implement. :-) >>>> >>>> Yes, maybe, the folio_zero_user(was clear_huge_page) is from >>>> 47ad8475c000 ("thp: clear_copy_huge_page"), so original clear_huge_page >>>> is used in HugeTLB, clear PUD size maybe spend many time, but for PMD or >>>> other size of large folio, cond_resched is not necessary since we >>>> already have some folio_zero_range() to clear large folio, and no issue >>>> was reported. >>> probably worth an optimization. calling cond_resched() for each page >>> seems too aggressive and useless. >> >> After some test, I think the cond_resched() is not the root cause, >> no performance gained with batched cond_resched(), even I kill >> cond_resched() from process_huge_page, no improvement. >> >> But when I unconditionally use clear_gigantic_page() in >> folio_zero_user(patched), there is big improvement with above >> fallocate on tmpfs(mount huge=always), also I test some other testcase, >> >> >> 1) case-anon-w-seq-mt: (2M PMD THP) >> >> base: >> real 0m2.490s 0m2.254s 0m2.272s >> user 1m59.980s 2m23.431s 2m18.739s >> sys 1m3.675s 1m15.462s 1m15.030s >> >> patched: >> real 0m2.234s 0m2.225s 0m2.159s >> user 2m56.105s 2m57.117s 3m0.489s >> sys 0m17.064s 0m17.564s 0m16.150s >> >> Patched kernel win on sys and bad in user, but real is almost same, >> maybe a little better than base. > > We can find user time difference. That means the original cache hot > behavior still applies on your system. > > However, it appears that the performance to clear page from end to begin > is really bad on your system. > > So, I suggest to revise the current implementation to use sequential > clearing as much as possible. > I test case-anon-cow-seq-hugetlb for copy_user_large_folio() base: real 0m6.259s 0m6.197s 0m6.316s user 1m31.176s 1m27.195s 1m29.594s sys 7m44.199s 7m51.490s 8m21.149s patched(use copy_user_gigantic_page for 2M hugetlb too) real 0m3.182s 0m3.002s 0m2.963s user 1m19.456s 1m3.107s 1m6.447s sys 2m59.222s 3m10.899s 3m1.027s and sequential copy is better than the current implementation, so I will use sequential clear and copy. >> 2) case-anon-w-seq-hugetlb:(2M PMD HugeTLB) >> >> base: >> real 0m5.175s 0m5.117s 0m4.856s >> user 5m15.943s 5m7.567s 4m29.273s >> sys 2m38.503s 2m21.949s 2m21.252s >> >> patched: >> real 0m4.966s 0m4.841s 0m4.561s >> user 6m30.123s 6m9.516s 5m49.733s >> sys 0m58.503s 0m47.847s 0m46.785s >> >> >> This case is similar to the case1. >> >> 3) fallocate hugetlb 20G (2M PMD HugeTLB) >> >> base: >> real 0m3.016s 0m3.019s 0m3.018s >> user 0m0.000s 0m0.000s 0m0.000s >> sys 0m3.009s 0m3.012s 0m3.010s >> >> patched: >> >> real 0m1.136s 0m1.136s 0m1.136s >> user 0m0.000s 0m0.000s 0m0.004s >> sys 0m1.133s 0m1.133s 0m1.129s >> >> >> There is big win on patched kernel, and it is similar to above tmpfs >> test, so maybe we could revert the commit c79b57e462b5 ("mm: hugetlb: >> clear target sub-page last when clearing huge page"). > > -- > Best Regards, > Huang, Ying >
Kefeng Wang <wangkefeng.wang@huawei.com> writes: > On 2024/10/25 10:59, Huang, Ying wrote: >> Hi, Kefeng, >> Kefeng Wang <wangkefeng.wang@huawei.com> writes: >> >>> +CC Huang Ying, >>> >>> On 2024/10/23 6:56, Barry Song wrote: >>>> On Wed, Oct 23, 2024 at 4:10 AM Kefeng Wang <wangkefeng.wang@huawei.com> wrote: >>>>> >>>>> >>> ... >>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>> On 2024/10/17 23:09, Matthew Wilcox wrote: >>>>>>>>>>>>>>>>>>>> On Thu, Oct 17, 2024 at 10:25:04PM +0800, Kefeng Wang wrote: >>>>>>>>>>>>>>>>>>>>> Directly use folio_zero_range() to cleanup code. >>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>> Are you sure there's no performance regression introduced by this? >>>>>>>>>>>>>>>>>>>> clear_highpage() is often optimised in ways that we can't optimise for >>>>>>>>>>>>>>>>>>>> a plain memset(). On the other hand, if the folio is large, maybe a >>>>>>>>>>>>>>>>>>>> modern CPU will be able to do better than clear-one-page-at-a-time. >>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>> Right, I missing this, clear_page might be better than memset, I change >>>>>>>>>>>>>>>>>>> this one when look at the shmem_writepage(), which already convert to >>>>>>>>>>>>>>>>>>> use folio_zero_range() from clear_highpage(), also I grep >>>>>>>>>>>>>>>>>>> folio_zero_range(), there are some other to use folio_zero_range(). >>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(folio, 0, >>>>>>>>>>>>>>>>>>> folio_size(folio)); >>>>>>>>>>>>>>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(f, >>>>>>>>>>>>>>>>>>> 0, folio_size(f)); >>>>>>>>>>>>>>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(f, >>>>>>>>>>>>>>>>>>> 0, folio_size(f)); >>>>>>>>>>>>>>>>>>> fs/libfs.c: folio_zero_range(folio, 0, folio_size(folio)); >>>>>>>>>>>>>>>>>>> fs/ntfs3/frecord.c: folio_zero_range(folio, 0, >>>>>>>>>>>>>>>>>>> folio_size(folio)); >>>>>>>>>>>>>>>>>>> mm/page_io.c: folio_zero_range(folio, 0, folio_size(folio)); >>>>>>>>>>>>>>>>>>> mm/shmem.c: folio_zero_range(folio, 0, folio_size(folio)); >>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>> IOW, what performance testing have you done with this patch? >>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>> No performance test before, but I write a testcase, >>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>> 1) allocate N large folios (folio_alloc(PMD_ORDER)) >>>>>>>>>>>>>>>>>>> 2) then calculate the diff(us) when clear all N folios >>>>>>>>>>>>>>>>>>> clear_highpage/folio_zero_range/folio_zero_user >>>>>>>>>>>>>>>>>>> 3) release N folios >>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>> the result(run 5 times) shown below on my machine, >>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>> N=1, >>>>>>>>>>>>>>>>>>> clear_highpage folio_zero_range folio_zero_user >>>>>>>>>>>>>>>>>>> 1 69 74 177 >>>>>>>>>>>>>>>>>>> 2 57 62 168 >>>>>>>>>>>>>>>>>>> 3 54 58 234 >>>>>>>>>>>>>>>>>>> 4 54 58 157 >>>>>>>>>>>>>>>>>>> 5 56 62 148 >>>>>>>>>>>>>>>>>>> avg 58 62.8 176.8 >>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>> N=100 >>>>>>>>>>>>>>>>>>> clear_highpage folio_zero_range folio_zero_user >>>>>>>>>>>>>>>>>>> 1 11015 11309 32833 >>>>>>>>>>>>>>>>>>> 2 10385 11110 49751 >>>>>>>>>>>>>>>>>>> 3 10369 11056 33095 >>>>>>>>>>>>>>>>>>> 4 10332 11017 33106 >>>>>>>>>>>>>>>>>>> 5 10483 11000 49032 >>>>>>>>>>>>>>>>>>> avg 10516.8 11098.4 39563.4 >>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>> N=512 >>>>>>>>>>>>>>>>>>> clear_highpage folio_zero_range folio_zero_user >>>>>>>>>>>>>>>>>>> 1 55560 60055 156876 >>>>>>>>>>>>>>>>>>> 2 55485 60024 157132 >>>>>>>>>>>>>>>>>>> 3 55474 60129 156658 >>>>>>>>>>>>>>>>>>> 4 55555 59867 157259 >>>>>>>>>>>>>>>>>>> 5 55528 59932 157108 >>>>>>>>>>>>>>>>>>> avg 55520.4 60001.4 157006.6 >>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>> folio_zero_user with many cond_resched(), so time fluctuates a lot, >>>>>>>>>>>>>>>>>>> clear_highpage is better folio_zero_range as you said. >>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>> Maybe add a new helper to convert all folio_zero_range(folio, 0, >>>>>>>>>>>>>>>>>>> folio_size(folio)) >>>>>>>>>>>>>>>>>>> to use clear_highpage + flush_dcache_folio? >>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>> If this also improves performance for other existing callers of >>>>>>>>>>>>>>>>>> folio_zero_range(), then that's a positive outcome. >>>>>>>>>>>>>>>>> >>>>>>>>>>>> ... >>>>>>>>>>>> >>>>>>>>>>>>>>> hi Kefeng, >>>>>>>>>>>>>>> what's your point? providing a helper like clear_highfolio() or similar? >>>>>>>>>>>>>> >>>>>>>>>>>>>> Yes, from above test, using clear_highpage/flush_dcache_folio is better >>>>>>>>>>>>>> than using folio_zero_range() for folio zero(especially for large >>>>>>>>>>>>>> folio), so I'd like to add a new helper, maybe name it folio_zero() >>>>>>>>>>>>>> since it zero the whole folio. >>>>>>>>>>>>> >>>>>>>>>>>>> we already have a helper like folio_zero_user()? >>>>>>>>>>>>> it is not good enough? >>>>>>>>>>>> >>>>>>>>>>>> Since it is with many cond_resched(), the performance is worst... >>>>>>>>>>> >>>>>>>>>>> Not exactly? It should have zero cost for a preemptible kernel. >>>>>>>>>>> For a non-preemptible kernel, it helps avoid clearing the folio >>>>>>>>>>> from occupying the CPU and starving other processes, right? >>>>>>>>>> >>>>>>>>>> --- a/mm/shmem.c >>>>>>>>>> +++ b/mm/shmem.c >>>>>>>>>> >>>>>>>>>> @@ -2393,10 +2393,7 @@ static int shmem_get_folio_gfp(struct inode >>>>>>>>>> *inode, pgoff_t index, >>>>>>>>>> * it now, lest undo on failure cancel our earlier guarantee. >>>>>>>>>> */ >>>>>>>>>> >>>>>>>>>> if (sgp != SGP_WRITE && !folio_test_uptodate(folio)) { >>>>>>>>>> - long i, n = folio_nr_pages(folio); >>>>>>>>>> - >>>>>>>>>> - for (i = 0; i < n; i++) >>>>>>>>>> - clear_highpage(folio_page(folio, i)); >>>>>>>>>> + folio_zero_user(folio, vmf->address); >>>>>>>>>> flush_dcache_folio(folio); >>>>>>>>>> folio_mark_uptodate(folio); >>>>>>>>>> } >>>>>>>>>> >>>>>>>>>> Do we perform better or worse with the following? >>>>>>>>> >>>>>>>>> Here is for SGP_FALLOC, vmf = NULL, we could use folio_zero_user(folio, >>>>>>>>> 0), I think the performance is worse, will retest once I can access >>>>>>>>> hardware. >>>>>>>> >>>>>>>> Perhaps, since the current code uses clear_hugepage(). Does using >>>>>>>> index << PAGE_SHIFT as the addr_hint offer any benefit? >>>>>>>> >>>>>>> >>>>>>> when use folio_zero_user(), the performance is vary bad with above >>>>>>> fallocate test(mount huge=always), >>>>>>> >>>>>>> folio_zero_range clear_highpage folio_zero_user >>>>>>> real 0m1.214s 0m1.111s 0m3.159s >>>>>>> user 0m0.000s 0m0.000s 0m0.000s >>>>>>> sys 0m1.210s 0m1.109s 0m3.152s >>>>>>> >>>>>>> I tried with addr_hint = 0/index << PAGE_SHIFT, no obvious different. >>>>>> >>>>>> Interesting. Does your kernel have preemption disabled or >>>>>> preemption_debug enabled? >>>>> >>>>> ARM64 server, CONFIG_PREEMPT_NONE=y >>>> this explains why the performance is much worse. >>>> >>>>> >>>>>> >>>>>> If not, it makes me wonder whether folio_zero_user() in >>>>>> alloc_anon_folio() is actually improving performance as expected, >>>>>> compared to the simpler folio_zero() you plan to implement. :-) >>>>> >>>>> Yes, maybe, the folio_zero_user(was clear_huge_page) is from >>>>> 47ad8475c000 ("thp: clear_copy_huge_page"), so original clear_huge_page >>>>> is used in HugeTLB, clear PUD size maybe spend many time, but for PMD or >>>>> other size of large folio, cond_resched is not necessary since we >>>>> already have some folio_zero_range() to clear large folio, and no issue >>>>> was reported. >>>> probably worth an optimization. calling cond_resched() for each page >>>> seems too aggressive and useless. >>> >>> After some test, I think the cond_resched() is not the root cause, >>> no performance gained with batched cond_resched(), even I kill >>> cond_resched() from process_huge_page, no improvement. >>> >>> But when I unconditionally use clear_gigantic_page() in >>> folio_zero_user(patched), there is big improvement with above >>> fallocate on tmpfs(mount huge=always), also I test some other testcase, >>> >>> >>> 1) case-anon-w-seq-mt: (2M PMD THP) >>> >>> base: >>> real 0m2.490s 0m2.254s 0m2.272s >>> user 1m59.980s 2m23.431s 2m18.739s >>> sys 1m3.675s 1m15.462s 1m15.030s >>> >>> patched: >>> real 0m2.234s 0m2.225s 0m2.159s >>> user 2m56.105s 2m57.117s 3m0.489s >>> sys 0m17.064s 0m17.564s 0m16.150s >>> >>> Patched kernel win on sys and bad in user, but real is almost same, >>> maybe a little better than base. >> We can find user time difference. That means the original cache hot >> behavior still applies on your system. >> However, it appears that the performance to clear page from end to >> begin >> is really bad on your system. >> So, I suggest to revise the current implementation to use sequential >> clearing as much as possible. >> > > I test case-anon-cow-seq-hugetlb for copy_user_large_folio() > > base: > real 0m6.259s 0m6.197s 0m6.316s > user 1m31.176s 1m27.195s 1m29.594s > sys 7m44.199s 7m51.490s 8m21.149s > > patched(use copy_user_gigantic_page for 2M hugetlb too) > real 0m3.182s 0m3.002s 0m2.963s > user 1m19.456s 1m3.107s 1m6.447s > sys 2m59.222s 3m10.899s 3m1.027s > > and sequential copy is better than the current implementation, > so I will use sequential clear and copy. Sorry, it appears that you misunderstanding my suggestion. I suggest to revise process_huge_page() to use more sequential memory clearing and copying to improve its performance on your platform. -- Best Regards, Huang, Ying >>> 2) case-anon-w-seq-hugetlb:(2M PMD HugeTLB) >>> >>> base: >>> real 0m5.175s 0m5.117s 0m4.856s >>> user 5m15.943s 5m7.567s 4m29.273s >>> sys 2m38.503s 2m21.949s 2m21.252s >>> >>> patched: >>> real 0m4.966s 0m4.841s 0m4.561s >>> user 6m30.123s 6m9.516s 5m49.733s >>> sys 0m58.503s 0m47.847s 0m46.785s >>> >>> >>> This case is similar to the case1. >>> >>> 3) fallocate hugetlb 20G (2M PMD HugeTLB) >>> >>> base: >>> real 0m3.016s 0m3.019s 0m3.018s >>> user 0m0.000s 0m0.000s 0m0.000s >>> sys 0m3.009s 0m3.012s 0m3.010s >>> >>> patched: >>> >>> real 0m1.136s 0m1.136s 0m1.136s >>> user 0m0.000s 0m0.000s 0m0.004s >>> sys 0m1.133s 0m1.133s 0m1.129s >>> >>> >>> There is big win on patched kernel, and it is similar to above tmpfs >>> test, so maybe we could revert the commit c79b57e462b5 ("mm: hugetlb: >>> clear target sub-page last when clearing huge page"). -- Best Regards, Huang, Ying
On 2024/10/25 15:47, Huang, Ying wrote: > Kefeng Wang <wangkefeng.wang@huawei.com> writes: > >> On 2024/10/25 10:59, Huang, Ying wrote: >>> Hi, Kefeng, >>> Kefeng Wang <wangkefeng.wang@huawei.com> writes: >>> >>>> +CC Huang Ying, >>>> >>>> On 2024/10/23 6:56, Barry Song wrote: >>>>> On Wed, Oct 23, 2024 at 4:10 AM Kefeng Wang <wangkefeng.wang@huawei.com> wrote: >>>>>> >>>>>> >>>> ... >>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>> On 2024/10/17 23:09, Matthew Wilcox wrote: >>>>>>>>>>>>>>>>>>>>> On Thu, Oct 17, 2024 at 10:25:04PM +0800, Kefeng Wang wrote: >>>>>>>>>>>>>>>>>>>>>> Directly use folio_zero_range() to cleanup code. >>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>> Are you sure there's no performance regression introduced by this? >>>>>>>>>>>>>>>>>>>>> clear_highpage() is often optimised in ways that we can't optimise for >>>>>>>>>>>>>>>>>>>>> a plain memset(). On the other hand, if the folio is large, maybe a >>>>>>>>>>>>>>>>>>>>> modern CPU will be able to do better than clear-one-page-at-a-time. >>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>> Right, I missing this, clear_page might be better than memset, I change >>>>>>>>>>>>>>>>>>>> this one when look at the shmem_writepage(), which already convert to >>>>>>>>>>>>>>>>>>>> use folio_zero_range() from clear_highpage(), also I grep >>>>>>>>>>>>>>>>>>>> folio_zero_range(), there are some other to use folio_zero_range(). >>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(folio, 0, >>>>>>>>>>>>>>>>>>>> folio_size(folio)); >>>>>>>>>>>>>>>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(f, >>>>>>>>>>>>>>>>>>>> 0, folio_size(f)); >>>>>>>>>>>>>>>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(f, >>>>>>>>>>>>>>>>>>>> 0, folio_size(f)); >>>>>>>>>>>>>>>>>>>> fs/libfs.c: folio_zero_range(folio, 0, folio_size(folio)); >>>>>>>>>>>>>>>>>>>> fs/ntfs3/frecord.c: folio_zero_range(folio, 0, >>>>>>>>>>>>>>>>>>>> folio_size(folio)); >>>>>>>>>>>>>>>>>>>> mm/page_io.c: folio_zero_range(folio, 0, folio_size(folio)); >>>>>>>>>>>>>>>>>>>> mm/shmem.c: folio_zero_range(folio, 0, folio_size(folio)); >>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>> IOW, what performance testing have you done with this patch? >>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>> No performance test before, but I write a testcase, >>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>> 1) allocate N large folios (folio_alloc(PMD_ORDER)) >>>>>>>>>>>>>>>>>>>> 2) then calculate the diff(us) when clear all N folios >>>>>>>>>>>>>>>>>>>> clear_highpage/folio_zero_range/folio_zero_user >>>>>>>>>>>>>>>>>>>> 3) release N folios >>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>> the result(run 5 times) shown below on my machine, >>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>> N=1, >>>>>>>>>>>>>>>>>>>> clear_highpage folio_zero_range folio_zero_user >>>>>>>>>>>>>>>>>>>> 1 69 74 177 >>>>>>>>>>>>>>>>>>>> 2 57 62 168 >>>>>>>>>>>>>>>>>>>> 3 54 58 234 >>>>>>>>>>>>>>>>>>>> 4 54 58 157 >>>>>>>>>>>>>>>>>>>> 5 56 62 148 >>>>>>>>>>>>>>>>>>>> avg 58 62.8 176.8 >>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>> N=100 >>>>>>>>>>>>>>>>>>>> clear_highpage folio_zero_range folio_zero_user >>>>>>>>>>>>>>>>>>>> 1 11015 11309 32833 >>>>>>>>>>>>>>>>>>>> 2 10385 11110 49751 >>>>>>>>>>>>>>>>>>>> 3 10369 11056 33095 >>>>>>>>>>>>>>>>>>>> 4 10332 11017 33106 >>>>>>>>>>>>>>>>>>>> 5 10483 11000 49032 >>>>>>>>>>>>>>>>>>>> avg 10516.8 11098.4 39563.4 >>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>> N=512 >>>>>>>>>>>>>>>>>>>> clear_highpage folio_zero_range folio_zero_user >>>>>>>>>>>>>>>>>>>> 1 55560 60055 156876 >>>>>>>>>>>>>>>>>>>> 2 55485 60024 157132 >>>>>>>>>>>>>>>>>>>> 3 55474 60129 156658 >>>>>>>>>>>>>>>>>>>> 4 55555 59867 157259 >>>>>>>>>>>>>>>>>>>> 5 55528 59932 157108 >>>>>>>>>>>>>>>>>>>> avg 55520.4 60001.4 157006.6 >>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>> folio_zero_user with many cond_resched(), so time fluctuates a lot, >>>>>>>>>>>>>>>>>>>> clear_highpage is better folio_zero_range as you said. >>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>> Maybe add a new helper to convert all folio_zero_range(folio, 0, >>>>>>>>>>>>>>>>>>>> folio_size(folio)) >>>>>>>>>>>>>>>>>>>> to use clear_highpage + flush_dcache_folio? >>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>> If this also improves performance for other existing callers of >>>>>>>>>>>>>>>>>>> folio_zero_range(), then that's a positive outcome. >>>>>>>>>>>>>>>>>> >>>>>>>>>>>>> ... >>>>>>>>>>>>> >>>>>>>>>>>>>>>> hi Kefeng, >>>>>>>>>>>>>>>> what's your point? providing a helper like clear_highfolio() or similar? >>>>>>>>>>>>>>> >>>>>>>>>>>>>>> Yes, from above test, using clear_highpage/flush_dcache_folio is better >>>>>>>>>>>>>>> than using folio_zero_range() for folio zero(especially for large >>>>>>>>>>>>>>> folio), so I'd like to add a new helper, maybe name it folio_zero() >>>>>>>>>>>>>>> since it zero the whole folio. >>>>>>>>>>>>>> >>>>>>>>>>>>>> we already have a helper like folio_zero_user()? >>>>>>>>>>>>>> it is not good enough? >>>>>>>>>>>>> >>>>>>>>>>>>> Since it is with many cond_resched(), the performance is worst... >>>>>>>>>>>> >>>>>>>>>>>> Not exactly? It should have zero cost for a preemptible kernel. >>>>>>>>>>>> For a non-preemptible kernel, it helps avoid clearing the folio >>>>>>>>>>>> from occupying the CPU and starving other processes, right? >>>>>>>>>>> >>>>>>>>>>> --- a/mm/shmem.c >>>>>>>>>>> +++ b/mm/shmem.c >>>>>>>>>>> >>>>>>>>>>> @@ -2393,10 +2393,7 @@ static int shmem_get_folio_gfp(struct inode >>>>>>>>>>> *inode, pgoff_t index, >>>>>>>>>>> * it now, lest undo on failure cancel our earlier guarantee. >>>>>>>>>>> */ >>>>>>>>>>> >>>>>>>>>>> if (sgp != SGP_WRITE && !folio_test_uptodate(folio)) { >>>>>>>>>>> - long i, n = folio_nr_pages(folio); >>>>>>>>>>> - >>>>>>>>>>> - for (i = 0; i < n; i++) >>>>>>>>>>> - clear_highpage(folio_page(folio, i)); >>>>>>>>>>> + folio_zero_user(folio, vmf->address); >>>>>>>>>>> flush_dcache_folio(folio); >>>>>>>>>>> folio_mark_uptodate(folio); >>>>>>>>>>> } >>>>>>>>>>> >>>>>>>>>>> Do we perform better or worse with the following? >>>>>>>>>> >>>>>>>>>> Here is for SGP_FALLOC, vmf = NULL, we could use folio_zero_user(folio, >>>>>>>>>> 0), I think the performance is worse, will retest once I can access >>>>>>>>>> hardware. >>>>>>>>> >>>>>>>>> Perhaps, since the current code uses clear_hugepage(). Does using >>>>>>>>> index << PAGE_SHIFT as the addr_hint offer any benefit? >>>>>>>>> >>>>>>>> >>>>>>>> when use folio_zero_user(), the performance is vary bad with above >>>>>>>> fallocate test(mount huge=always), >>>>>>>> >>>>>>>> folio_zero_range clear_highpage folio_zero_user >>>>>>>> real 0m1.214s 0m1.111s 0m3.159s >>>>>>>> user 0m0.000s 0m0.000s 0m0.000s >>>>>>>> sys 0m1.210s 0m1.109s 0m3.152s >>>>>>>> >>>>>>>> I tried with addr_hint = 0/index << PAGE_SHIFT, no obvious different. >>>>>>> >>>>>>> Interesting. Does your kernel have preemption disabled or >>>>>>> preemption_debug enabled? >>>>>> >>>>>> ARM64 server, CONFIG_PREEMPT_NONE=y >>>>> this explains why the performance is much worse. >>>>> >>>>>> >>>>>>> >>>>>>> If not, it makes me wonder whether folio_zero_user() in >>>>>>> alloc_anon_folio() is actually improving performance as expected, >>>>>>> compared to the simpler folio_zero() you plan to implement. :-) >>>>>> >>>>>> Yes, maybe, the folio_zero_user(was clear_huge_page) is from >>>>>> 47ad8475c000 ("thp: clear_copy_huge_page"), so original clear_huge_page >>>>>> is used in HugeTLB, clear PUD size maybe spend many time, but for PMD or >>>>>> other size of large folio, cond_resched is not necessary since we >>>>>> already have some folio_zero_range() to clear large folio, and no issue >>>>>> was reported. >>>>> probably worth an optimization. calling cond_resched() for each page >>>>> seems too aggressive and useless. >>>> >>>> After some test, I think the cond_resched() is not the root cause, >>>> no performance gained with batched cond_resched(), even I kill >>>> cond_resched() from process_huge_page, no improvement. >>>> >>>> But when I unconditionally use clear_gigantic_page() in >>>> folio_zero_user(patched), there is big improvement with above >>>> fallocate on tmpfs(mount huge=always), also I test some other testcase, >>>> >>>> >>>> 1) case-anon-w-seq-mt: (2M PMD THP) >>>> >>>> base: >>>> real 0m2.490s 0m2.254s 0m2.272s >>>> user 1m59.980s 2m23.431s 2m18.739s >>>> sys 1m3.675s 1m15.462s 1m15.030s >>>> >>>> patched: >>>> real 0m2.234s 0m2.225s 0m2.159s >>>> user 2m56.105s 2m57.117s 3m0.489s >>>> sys 0m17.064s 0m17.564s 0m16.150s >>>> >>>> Patched kernel win on sys and bad in user, but real is almost same, >>>> maybe a little better than base. >>> We can find user time difference. That means the original cache hot >>> behavior still applies on your system. >>> However, it appears that the performance to clear page from end to >>> begin >>> is really bad on your system. >>> So, I suggest to revise the current implementation to use sequential >>> clearing as much as possible. >>> >> >> I test case-anon-cow-seq-hugetlb for copy_user_large_folio() >> >> base: >> real 0m6.259s 0m6.197s 0m6.316s >> user 1m31.176s 1m27.195s 1m29.594s >> sys 7m44.199s 7m51.490s 8m21.149s >> >> patched(use copy_user_gigantic_page for 2M hugetlb too) >> real 0m3.182s 0m3.002s 0m2.963s >> user 1m19.456s 1m3.107s 1m6.447s >> sys 2m59.222s 3m10.899s 3m1.027s >> >> and sequential copy is better than the current implementation, >> so I will use sequential clear and copy. > > Sorry, it appears that you misunderstanding my suggestion. I suggest to > revise process_huge_page() to use more sequential memory clearing and > copying to improve its performance on your platform. > > -- > Best Regards, > Huang, Ying > >>>> 2) case-anon-w-seq-hugetlb:(2M PMD HugeTLB) >>>> >>>> base: >>>> real 0m5.175s 0m5.117s 0m4.856s >>>> user 5m15.943s 5m7.567s 4m29.273s >>>> sys 2m38.503s 2m21.949s 2m21.252s >>>> >>>> patched: >>>> real 0m4.966s 0m4.841s 0m4.561s >>>> user 6m30.123s 6m9.516s 5m49.733s >>>> sys 0m58.503s 0m47.847s 0m46.785s >>>> >>>> >>>> This case is similar to the case1. >>>> >>>> 3) fallocate hugetlb 20G (2M PMD HugeTLB) >>>> >>>> base: >>>> real 0m3.016s 0m3.019s 0m3.018s >>>> user 0m0.000s 0m0.000s 0m0.000s >>>> sys 0m3.009s 0m3.012s 0m3.010s >>>> >>>> patched: >>>> >>>> real 0m1.136s 0m1.136s 0m1.136s >>>> user 0m0.000s 0m0.000s 0m0.004s >>>> sys 0m1.133s 0m1.133s 0m1.129s >>>> >>>> >>>> There is big win on patched kernel, and it is similar to above tmpfs >>>> test, so maybe we could revert the commit c79b57e462b5 ("mm: hugetlb: >>>> clear target sub-page last when clearing huge page"). I tried the following changes, diff --git a/mm/memory.c b/mm/memory.c index 66cf855dee3f..e5cc75adfa10 100644 --- a/mm/memory.c +++ b/mm/memory.c @@ -6777,7 +6777,7 @@ static inline int process_huge_page( base = 0; l = n; /* Process subpages at the end of huge page */ - for (i = nr_pages - 1; i >= 2 * n; i--) { + for (i = 2 * n; i < nr_pages; i++) { cond_resched(); ret = process_subpage(addr + i * PAGE_SIZE, i, arg); if (ret) Since n = 0, so the copying is from start to end now, but not improvement for case-anon-cow-seq-hugetlb, and if use copy_user_gigantic_pager, the time reduced from 6s to 3s diff --git a/mm/memory.c b/mm/memory.c index fe21bd3beff5..2c6532d21d84 100644 --- a/mm/memory.c +++ b/mm/memory.c @@ -6876,10 +6876,7 @@ int copy_user_large_folio(struct folio *dst, struct folio *src, .vma = vma, }; - if (unlikely(nr_pages > MAX_ORDER_NR_PAGES)) - return copy_user_gigantic_page(dst, src, addr_hint, vma, nr_pages); - - return process_huge_page(addr_hint, nr_pages, copy_subpage, &arg); + return copy_user_gigantic_page(dst, src, addr_hint, vma, nr_pages); } > > -- > Best Regards, > Huang, Ying >
Kefeng Wang <wangkefeng.wang@huawei.com> writes: > On 2024/10/25 15:47, Huang, Ying wrote: >> Kefeng Wang <wangkefeng.wang@huawei.com> writes: >> >>> On 2024/10/25 10:59, Huang, Ying wrote: >>>> Hi, Kefeng, >>>> Kefeng Wang <wangkefeng.wang@huawei.com> writes: >>>> >>>>> +CC Huang Ying, >>>>> >>>>> On 2024/10/23 6:56, Barry Song wrote: >>>>>> On Wed, Oct 23, 2024 at 4:10 AM Kefeng Wang <wangkefeng.wang@huawei.com> wrote: >>>>>>> >>>>>>> >>>>> ... >>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>> On 2024/10/17 23:09, Matthew Wilcox wrote: >>>>>>>>>>>>>>>>>>>>>> On Thu, Oct 17, 2024 at 10:25:04PM +0800, Kefeng Wang wrote: >>>>>>>>>>>>>>>>>>>>>>> Directly use folio_zero_range() to cleanup code. >>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>> Are you sure there's no performance regression introduced by this? >>>>>>>>>>>>>>>>>>>>>> clear_highpage() is often optimised in ways that we can't optimise for >>>>>>>>>>>>>>>>>>>>>> a plain memset(). On the other hand, if the folio is large, maybe a >>>>>>>>>>>>>>>>>>>>>> modern CPU will be able to do better than clear-one-page-at-a-time. >>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>> Right, I missing this, clear_page might be better than memset, I change >>>>>>>>>>>>>>>>>>>>> this one when look at the shmem_writepage(), which already convert to >>>>>>>>>>>>>>>>>>>>> use folio_zero_range() from clear_highpage(), also I grep >>>>>>>>>>>>>>>>>>>>> folio_zero_range(), there are some other to use folio_zero_range(). >>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(folio, 0, >>>>>>>>>>>>>>>>>>>>> folio_size(folio)); >>>>>>>>>>>>>>>>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(f, >>>>>>>>>>>>>>>>>>>>> 0, folio_size(f)); >>>>>>>>>>>>>>>>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(f, >>>>>>>>>>>>>>>>>>>>> 0, folio_size(f)); >>>>>>>>>>>>>>>>>>>>> fs/libfs.c: folio_zero_range(folio, 0, folio_size(folio)); >>>>>>>>>>>>>>>>>>>>> fs/ntfs3/frecord.c: folio_zero_range(folio, 0, >>>>>>>>>>>>>>>>>>>>> folio_size(folio)); >>>>>>>>>>>>>>>>>>>>> mm/page_io.c: folio_zero_range(folio, 0, folio_size(folio)); >>>>>>>>>>>>>>>>>>>>> mm/shmem.c: folio_zero_range(folio, 0, folio_size(folio)); >>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>> IOW, what performance testing have you done with this patch? >>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>> No performance test before, but I write a testcase, >>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>> 1) allocate N large folios (folio_alloc(PMD_ORDER)) >>>>>>>>>>>>>>>>>>>>> 2) then calculate the diff(us) when clear all N folios >>>>>>>>>>>>>>>>>>>>> clear_highpage/folio_zero_range/folio_zero_user >>>>>>>>>>>>>>>>>>>>> 3) release N folios >>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>> the result(run 5 times) shown below on my machine, >>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>> N=1, >>>>>>>>>>>>>>>>>>>>> clear_highpage folio_zero_range folio_zero_user >>>>>>>>>>>>>>>>>>>>> 1 69 74 177 >>>>>>>>>>>>>>>>>>>>> 2 57 62 168 >>>>>>>>>>>>>>>>>>>>> 3 54 58 234 >>>>>>>>>>>>>>>>>>>>> 4 54 58 157 >>>>>>>>>>>>>>>>>>>>> 5 56 62 148 >>>>>>>>>>>>>>>>>>>>> avg 58 62.8 176.8 >>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>> N=100 >>>>>>>>>>>>>>>>>>>>> clear_highpage folio_zero_range folio_zero_user >>>>>>>>>>>>>>>>>>>>> 1 11015 11309 32833 >>>>>>>>>>>>>>>>>>>>> 2 10385 11110 49751 >>>>>>>>>>>>>>>>>>>>> 3 10369 11056 33095 >>>>>>>>>>>>>>>>>>>>> 4 10332 11017 33106 >>>>>>>>>>>>>>>>>>>>> 5 10483 11000 49032 >>>>>>>>>>>>>>>>>>>>> avg 10516.8 11098.4 39563.4 >>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>> N=512 >>>>>>>>>>>>>>>>>>>>> clear_highpage folio_zero_range folio_zero_user >>>>>>>>>>>>>>>>>>>>> 1 55560 60055 156876 >>>>>>>>>>>>>>>>>>>>> 2 55485 60024 157132 >>>>>>>>>>>>>>>>>>>>> 3 55474 60129 156658 >>>>>>>>>>>>>>>>>>>>> 4 55555 59867 157259 >>>>>>>>>>>>>>>>>>>>> 5 55528 59932 157108 >>>>>>>>>>>>>>>>>>>>> avg 55520.4 60001.4 157006.6 >>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>> folio_zero_user with many cond_resched(), so time fluctuates a lot, >>>>>>>>>>>>>>>>>>>>> clear_highpage is better folio_zero_range as you said. >>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>> Maybe add a new helper to convert all folio_zero_range(folio, 0, >>>>>>>>>>>>>>>>>>>>> folio_size(folio)) >>>>>>>>>>>>>>>>>>>>> to use clear_highpage + flush_dcache_folio? >>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>> If this also improves performance for other existing callers of >>>>>>>>>>>>>>>>>>>> folio_zero_range(), then that's a positive outcome. >>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>> ... >>>>>>>>>>>>>> >>>>>>>>>>>>>>>>> hi Kefeng, >>>>>>>>>>>>>>>>> what's your point? providing a helper like clear_highfolio() or similar? >>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>> Yes, from above test, using clear_highpage/flush_dcache_folio is better >>>>>>>>>>>>>>>> than using folio_zero_range() for folio zero(especially for large >>>>>>>>>>>>>>>> folio), so I'd like to add a new helper, maybe name it folio_zero() >>>>>>>>>>>>>>>> since it zero the whole folio. >>>>>>>>>>>>>>> >>>>>>>>>>>>>>> we already have a helper like folio_zero_user()? >>>>>>>>>>>>>>> it is not good enough? >>>>>>>>>>>>>> >>>>>>>>>>>>>> Since it is with many cond_resched(), the performance is worst... >>>>>>>>>>>>> >>>>>>>>>>>>> Not exactly? It should have zero cost for a preemptible kernel. >>>>>>>>>>>>> For a non-preemptible kernel, it helps avoid clearing the folio >>>>>>>>>>>>> from occupying the CPU and starving other processes, right? >>>>>>>>>>>> >>>>>>>>>>>> --- a/mm/shmem.c >>>>>>>>>>>> +++ b/mm/shmem.c >>>>>>>>>>>> >>>>>>>>>>>> @@ -2393,10 +2393,7 @@ static int shmem_get_folio_gfp(struct inode >>>>>>>>>>>> *inode, pgoff_t index, >>>>>>>>>>>> * it now, lest undo on failure cancel our earlier guarantee. >>>>>>>>>>>> */ >>>>>>>>>>>> >>>>>>>>>>>> if (sgp != SGP_WRITE && !folio_test_uptodate(folio)) { >>>>>>>>>>>> - long i, n = folio_nr_pages(folio); >>>>>>>>>>>> - >>>>>>>>>>>> - for (i = 0; i < n; i++) >>>>>>>>>>>> - clear_highpage(folio_page(folio, i)); >>>>>>>>>>>> + folio_zero_user(folio, vmf->address); >>>>>>>>>>>> flush_dcache_folio(folio); >>>>>>>>>>>> folio_mark_uptodate(folio); >>>>>>>>>>>> } >>>>>>>>>>>> >>>>>>>>>>>> Do we perform better or worse with the following? >>>>>>>>>>> >>>>>>>>>>> Here is for SGP_FALLOC, vmf = NULL, we could use folio_zero_user(folio, >>>>>>>>>>> 0), I think the performance is worse, will retest once I can access >>>>>>>>>>> hardware. >>>>>>>>>> >>>>>>>>>> Perhaps, since the current code uses clear_hugepage(). Does using >>>>>>>>>> index << PAGE_SHIFT as the addr_hint offer any benefit? >>>>>>>>>> >>>>>>>>> >>>>>>>>> when use folio_zero_user(), the performance is vary bad with above >>>>>>>>> fallocate test(mount huge=always), >>>>>>>>> >>>>>>>>> folio_zero_range clear_highpage folio_zero_user >>>>>>>>> real 0m1.214s 0m1.111s 0m3.159s >>>>>>>>> user 0m0.000s 0m0.000s 0m0.000s >>>>>>>>> sys 0m1.210s 0m1.109s 0m3.152s >>>>>>>>> >>>>>>>>> I tried with addr_hint = 0/index << PAGE_SHIFT, no obvious different. >>>>>>>> >>>>>>>> Interesting. Does your kernel have preemption disabled or >>>>>>>> preemption_debug enabled? >>>>>>> >>>>>>> ARM64 server, CONFIG_PREEMPT_NONE=y >>>>>> this explains why the performance is much worse. >>>>>> >>>>>>> >>>>>>>> >>>>>>>> If not, it makes me wonder whether folio_zero_user() in >>>>>>>> alloc_anon_folio() is actually improving performance as expected, >>>>>>>> compared to the simpler folio_zero() you plan to implement. :-) >>>>>>> >>>>>>> Yes, maybe, the folio_zero_user(was clear_huge_page) is from >>>>>>> 47ad8475c000 ("thp: clear_copy_huge_page"), so original clear_huge_page >>>>>>> is used in HugeTLB, clear PUD size maybe spend many time, but for PMD or >>>>>>> other size of large folio, cond_resched is not necessary since we >>>>>>> already have some folio_zero_range() to clear large folio, and no issue >>>>>>> was reported. >>>>>> probably worth an optimization. calling cond_resched() for each page >>>>>> seems too aggressive and useless. >>>>> >>>>> After some test, I think the cond_resched() is not the root cause, >>>>> no performance gained with batched cond_resched(), even I kill >>>>> cond_resched() from process_huge_page, no improvement. >>>>> >>>>> But when I unconditionally use clear_gigantic_page() in >>>>> folio_zero_user(patched), there is big improvement with above >>>>> fallocate on tmpfs(mount huge=always), also I test some other testcase, >>>>> >>>>> >>>>> 1) case-anon-w-seq-mt: (2M PMD THP) >>>>> >>>>> base: >>>>> real 0m2.490s 0m2.254s 0m2.272s >>>>> user 1m59.980s 2m23.431s 2m18.739s >>>>> sys 1m3.675s 1m15.462s 1m15.030s >>>>> >>>>> patched: >>>>> real 0m2.234s 0m2.225s 0m2.159s >>>>> user 2m56.105s 2m57.117s 3m0.489s >>>>> sys 0m17.064s 0m17.564s 0m16.150s >>>>> >>>>> Patched kernel win on sys and bad in user, but real is almost same, >>>>> maybe a little better than base. >>>> We can find user time difference. That means the original cache hot >>>> behavior still applies on your system. >>>> However, it appears that the performance to clear page from end to >>>> begin >>>> is really bad on your system. >>>> So, I suggest to revise the current implementation to use sequential >>>> clearing as much as possible. >>>> >>> >>> I test case-anon-cow-seq-hugetlb for copy_user_large_folio() >>> >>> base: >>> real 0m6.259s 0m6.197s 0m6.316s >>> user 1m31.176s 1m27.195s 1m29.594s >>> sys 7m44.199s 7m51.490s 8m21.149s >>> >>> patched(use copy_user_gigantic_page for 2M hugetlb too) >>> real 0m3.182s 0m3.002s 0m2.963s >>> user 1m19.456s 1m3.107s 1m6.447s >>> sys 2m59.222s 3m10.899s 3m1.027s >>> >>> and sequential copy is better than the current implementation, >>> so I will use sequential clear and copy. >> Sorry, it appears that you misunderstanding my suggestion. I >> suggest to >> revise process_huge_page() to use more sequential memory clearing and >> copying to improve its performance on your platform. >> -- >> Best Regards, >> Huang, Ying >> >>>>> 2) case-anon-w-seq-hugetlb:(2M PMD HugeTLB) >>>>> >>>>> base: >>>>> real 0m5.175s 0m5.117s 0m4.856s >>>>> user 5m15.943s 5m7.567s 4m29.273s >>>>> sys 2m38.503s 2m21.949s 2m21.252s >>>>> >>>>> patched: >>>>> real 0m4.966s 0m4.841s 0m4.561s >>>>> user 6m30.123s 6m9.516s 5m49.733s >>>>> sys 0m58.503s 0m47.847s 0m46.785s >>>>> >>>>> >>>>> This case is similar to the case1. >>>>> >>>>> 3) fallocate hugetlb 20G (2M PMD HugeTLB) >>>>> >>>>> base: >>>>> real 0m3.016s 0m3.019s 0m3.018s >>>>> user 0m0.000s 0m0.000s 0m0.000s >>>>> sys 0m3.009s 0m3.012s 0m3.010s >>>>> >>>>> patched: >>>>> >>>>> real 0m1.136s 0m1.136s 0m1.136s >>>>> user 0m0.000s 0m0.000s 0m0.004s >>>>> sys 0m1.133s 0m1.133s 0m1.129s >>>>> >>>>> >>>>> There is big win on patched kernel, and it is similar to above tmpfs >>>>> test, so maybe we could revert the commit c79b57e462b5 ("mm: hugetlb: >>>>> clear target sub-page last when clearing huge page"). > > I tried the following changes, > diff --git a/mm/memory.c b/mm/memory.c > index 66cf855dee3f..e5cc75adfa10 100644 > --- a/mm/memory.c > +++ b/mm/memory.c > @@ -6777,7 +6777,7 @@ static inline int process_huge_page( > base = 0; > l = n; > /* Process subpages at the end of huge page */ > - for (i = nr_pages - 1; i >= 2 * n; i--) { > + for (i = 2 * n; i < nr_pages; i++) { > cond_resched(); > ret = process_subpage(addr + i * PAGE_SIZE, i, > arg); > if (ret) > > Since n = 0, so the copying is from start to end now, but not > improvement for case-anon-cow-seq-hugetlb, > > and if use copy_user_gigantic_pager, the time reduced from 6s to 3s > > diff --git a/mm/memory.c b/mm/memory.c > index fe21bd3beff5..2c6532d21d84 100644 > --- a/mm/memory.c > +++ b/mm/memory.c > @@ -6876,10 +6876,7 @@ int copy_user_large_folio(struct folio *dst, > struct folio *src, > .vma = vma, > }; > > - if (unlikely(nr_pages > MAX_ORDER_NR_PAGES)) > - return copy_user_gigantic_page(dst, src, addr_hint, > vma, nr_pages); > - > - return process_huge_page(addr_hint, nr_pages, copy_subpage, &arg); > + return copy_user_gigantic_page(dst, src, addr_hint, vma, nr_pages); > } It appears that we have code generation issue here. Can you check it? Whether code is inlined in the same way? Maybe we can start with modified mm/memory.c @@ -6714,7 +6714,7 @@ EXPORT_SYMBOL(__might_fault); * operation. The target subpage will be processed last to keep its * cache lines hot. */ -static inline int process_huge_page( +static __always_inline int process_huge_page( unsigned long addr_hint, unsigned int nr_pages, int (*process_subpage)(unsigned long addr, int idx, void *arg), void *arg) -- Best Regards, Huang, Ying
On 2024/10/25 20:21, Huang, Ying wrote: > Kefeng Wang <wangkefeng.wang@huawei.com> writes: > >> On 2024/10/25 15:47, Huang, Ying wrote: >>> Kefeng Wang <wangkefeng.wang@huawei.com> writes: >>> >>>> On 2024/10/25 10:59, Huang, Ying wrote: >>>>> Hi, Kefeng, >>>>> Kefeng Wang <wangkefeng.wang@huawei.com> writes: >>>>> >>>>>> +CC Huang Ying, >>>>>> >>>>>> On 2024/10/23 6:56, Barry Song wrote: >>>>>>> On Wed, Oct 23, 2024 at 4:10 AM Kefeng Wang <wangkefeng.wang@huawei.com> wrote: >>>>>>>> >>>>>>>> >>>>>> ... >>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>> On 2024/10/17 23:09, Matthew Wilcox wrote: >>>>>>>>>>>>>>>>>>>>>>> On Thu, Oct 17, 2024 at 10:25:04PM +0800, Kefeng Wang wrote: >>>>>>>>>>>>>>>>>>>>>>>> Directly use folio_zero_range() to cleanup code. >>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>> Are you sure there's no performance regression introduced by this? >>>>>>>>>>>>>>>>>>>>>>> clear_highpage() is often optimised in ways that we can't optimise for >>>>>>>>>>>>>>>>>>>>>>> a plain memset(). On the other hand, if the folio is large, maybe a >>>>>>>>>>>>>>>>>>>>>>> modern CPU will be able to do better than clear-one-page-at-a-time. >>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>> Right, I missing this, clear_page might be better than memset, I change >>>>>>>>>>>>>>>>>>>>>> this one when look at the shmem_writepage(), which already convert to >>>>>>>>>>>>>>>>>>>>>> use folio_zero_range() from clear_highpage(), also I grep >>>>>>>>>>>>>>>>>>>>>> folio_zero_range(), there are some other to use folio_zero_range(). >>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(folio, 0, >>>>>>>>>>>>>>>>>>>>>> folio_size(folio)); >>>>>>>>>>>>>>>>>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(f, >>>>>>>>>>>>>>>>>>>>>> 0, folio_size(f)); >>>>>>>>>>>>>>>>>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(f, >>>>>>>>>>>>>>>>>>>>>> 0, folio_size(f)); >>>>>>>>>>>>>>>>>>>>>> fs/libfs.c: folio_zero_range(folio, 0, folio_size(folio)); >>>>>>>>>>>>>>>>>>>>>> fs/ntfs3/frecord.c: folio_zero_range(folio, 0, >>>>>>>>>>>>>>>>>>>>>> folio_size(folio)); >>>>>>>>>>>>>>>>>>>>>> mm/page_io.c: folio_zero_range(folio, 0, folio_size(folio)); >>>>>>>>>>>>>>>>>>>>>> mm/shmem.c: folio_zero_range(folio, 0, folio_size(folio)); >>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>> IOW, what performance testing have you done with this patch? >>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>> No performance test before, but I write a testcase, >>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>> 1) allocate N large folios (folio_alloc(PMD_ORDER)) >>>>>>>>>>>>>>>>>>>>>> 2) then calculate the diff(us) when clear all N folios >>>>>>>>>>>>>>>>>>>>>> clear_highpage/folio_zero_range/folio_zero_user >>>>>>>>>>>>>>>>>>>>>> 3) release N folios >>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>> the result(run 5 times) shown below on my machine, >>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>> N=1, >>>>>>>>>>>>>>>>>>>>>> clear_highpage folio_zero_range folio_zero_user >>>>>>>>>>>>>>>>>>>>>> 1 69 74 177 >>>>>>>>>>>>>>>>>>>>>> 2 57 62 168 >>>>>>>>>>>>>>>>>>>>>> 3 54 58 234 >>>>>>>>>>>>>>>>>>>>>> 4 54 58 157 >>>>>>>>>>>>>>>>>>>>>> 5 56 62 148 >>>>>>>>>>>>>>>>>>>>>> avg 58 62.8 176.8 >>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>> N=100 >>>>>>>>>>>>>>>>>>>>>> clear_highpage folio_zero_range folio_zero_user >>>>>>>>>>>>>>>>>>>>>> 1 11015 11309 32833 >>>>>>>>>>>>>>>>>>>>>> 2 10385 11110 49751 >>>>>>>>>>>>>>>>>>>>>> 3 10369 11056 33095 >>>>>>>>>>>>>>>>>>>>>> 4 10332 11017 33106 >>>>>>>>>>>>>>>>>>>>>> 5 10483 11000 49032 >>>>>>>>>>>>>>>>>>>>>> avg 10516.8 11098.4 39563.4 >>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>> N=512 >>>>>>>>>>>>>>>>>>>>>> clear_highpage folio_zero_range folio_zero_user >>>>>>>>>>>>>>>>>>>>>> 1 55560 60055 156876 >>>>>>>>>>>>>>>>>>>>>> 2 55485 60024 157132 >>>>>>>>>>>>>>>>>>>>>> 3 55474 60129 156658 >>>>>>>>>>>>>>>>>>>>>> 4 55555 59867 157259 >>>>>>>>>>>>>>>>>>>>>> 5 55528 59932 157108 >>>>>>>>>>>>>>>>>>>>>> avg 55520.4 60001.4 157006.6 >>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>> folio_zero_user with many cond_resched(), so time fluctuates a lot, >>>>>>>>>>>>>>>>>>>>>> clear_highpage is better folio_zero_range as you said. >>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>> Maybe add a new helper to convert all folio_zero_range(folio, 0, >>>>>>>>>>>>>>>>>>>>>> folio_size(folio)) >>>>>>>>>>>>>>>>>>>>>> to use clear_highpage + flush_dcache_folio? >>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>> If this also improves performance for other existing callers of >>>>>>>>>>>>>>>>>>>>> folio_zero_range(), then that's a positive outcome. >>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>> ... >>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>> hi Kefeng, >>>>>>>>>>>>>>>>>> what's your point? providing a helper like clear_highfolio() or similar? >>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>> Yes, from above test, using clear_highpage/flush_dcache_folio is better >>>>>>>>>>>>>>>>> than using folio_zero_range() for folio zero(especially for large >>>>>>>>>>>>>>>>> folio), so I'd like to add a new helper, maybe name it folio_zero() >>>>>>>>>>>>>>>>> since it zero the whole folio. >>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>> we already have a helper like folio_zero_user()? >>>>>>>>>>>>>>>> it is not good enough? >>>>>>>>>>>>>>> >>>>>>>>>>>>>>> Since it is with many cond_resched(), the performance is worst... >>>>>>>>>>>>>> >>>>>>>>>>>>>> Not exactly? It should have zero cost for a preemptible kernel. >>>>>>>>>>>>>> For a non-preemptible kernel, it helps avoid clearing the folio >>>>>>>>>>>>>> from occupying the CPU and starving other processes, right? >>>>>>>>>>>>> >>>>>>>>>>>>> --- a/mm/shmem.c >>>>>>>>>>>>> +++ b/mm/shmem.c >>>>>>>>>>>>> >>>>>>>>>>>>> @@ -2393,10 +2393,7 @@ static int shmem_get_folio_gfp(struct inode >>>>>>>>>>>>> *inode, pgoff_t index, >>>>>>>>>>>>> * it now, lest undo on failure cancel our earlier guarantee. >>>>>>>>>>>>> */ >>>>>>>>>>>>> >>>>>>>>>>>>> if (sgp != SGP_WRITE && !folio_test_uptodate(folio)) { >>>>>>>>>>>>> - long i, n = folio_nr_pages(folio); >>>>>>>>>>>>> - >>>>>>>>>>>>> - for (i = 0; i < n; i++) >>>>>>>>>>>>> - clear_highpage(folio_page(folio, i)); >>>>>>>>>>>>> + folio_zero_user(folio, vmf->address); >>>>>>>>>>>>> flush_dcache_folio(folio); >>>>>>>>>>>>> folio_mark_uptodate(folio); >>>>>>>>>>>>> } >>>>>>>>>>>>> >>>>>>>>>>>>> Do we perform better or worse with the following? >>>>>>>>>>>> >>>>>>>>>>>> Here is for SGP_FALLOC, vmf = NULL, we could use folio_zero_user(folio, >>>>>>>>>>>> 0), I think the performance is worse, will retest once I can access >>>>>>>>>>>> hardware. >>>>>>>>>>> >>>>>>>>>>> Perhaps, since the current code uses clear_hugepage(). Does using >>>>>>>>>>> index << PAGE_SHIFT as the addr_hint offer any benefit? >>>>>>>>>>> >>>>>>>>>> >>>>>>>>>> when use folio_zero_user(), the performance is vary bad with above >>>>>>>>>> fallocate test(mount huge=always), >>>>>>>>>> >>>>>>>>>> folio_zero_range clear_highpage folio_zero_user >>>>>>>>>> real 0m1.214s 0m1.111s 0m3.159s >>>>>>>>>> user 0m0.000s 0m0.000s 0m0.000s >>>>>>>>>> sys 0m1.210s 0m1.109s 0m3.152s >>>>>>>>>> >>>>>>>>>> I tried with addr_hint = 0/index << PAGE_SHIFT, no obvious different. >>>>>>>>> >>>>>>>>> Interesting. Does your kernel have preemption disabled or >>>>>>>>> preemption_debug enabled? >>>>>>>> >>>>>>>> ARM64 server, CONFIG_PREEMPT_NONE=y >>>>>>> this explains why the performance is much worse. >>>>>>> >>>>>>>> >>>>>>>>> >>>>>>>>> If not, it makes me wonder whether folio_zero_user() in >>>>>>>>> alloc_anon_folio() is actually improving performance as expected, >>>>>>>>> compared to the simpler folio_zero() you plan to implement. :-) >>>>>>>> >>>>>>>> Yes, maybe, the folio_zero_user(was clear_huge_page) is from >>>>>>>> 47ad8475c000 ("thp: clear_copy_huge_page"), so original clear_huge_page >>>>>>>> is used in HugeTLB, clear PUD size maybe spend many time, but for PMD or >>>>>>>> other size of large folio, cond_resched is not necessary since we >>>>>>>> already have some folio_zero_range() to clear large folio, and no issue >>>>>>>> was reported. >>>>>>> probably worth an optimization. calling cond_resched() for each page >>>>>>> seems too aggressive and useless. >>>>>> >>>>>> After some test, I think the cond_resched() is not the root cause, >>>>>> no performance gained with batched cond_resched(), even I kill >>>>>> cond_resched() from process_huge_page, no improvement. >>>>>> >>>>>> But when I unconditionally use clear_gigantic_page() in >>>>>> folio_zero_user(patched), there is big improvement with above >>>>>> fallocate on tmpfs(mount huge=always), also I test some other testcase, >>>>>> >>>>>> >>>>>> 1) case-anon-w-seq-mt: (2M PMD THP) >>>>>> >>>>>> base: >>>>>> real 0m2.490s 0m2.254s 0m2.272s >>>>>> user 1m59.980s 2m23.431s 2m18.739s >>>>>> sys 1m3.675s 1m15.462s 1m15.030s >>>>>> >>>>>> patched: >>>>>> real 0m2.234s 0m2.225s 0m2.159s >>>>>> user 2m56.105s 2m57.117s 3m0.489s >>>>>> sys 0m17.064s 0m17.564s 0m16.150s >>>>>> >>>>>> Patched kernel win on sys and bad in user, but real is almost same, >>>>>> maybe a little better than base. >>>>> We can find user time difference. That means the original cache hot >>>>> behavior still applies on your system. >>>>> However, it appears that the performance to clear page from end to >>>>> begin >>>>> is really bad on your system. >>>>> So, I suggest to revise the current implementation to use sequential >>>>> clearing as much as possible. >>>>> >>>> >>>> I test case-anon-cow-seq-hugetlb for copy_user_large_folio() >>>> >>>> base: >>>> real 0m6.259s 0m6.197s 0m6.316s >>>> user 1m31.176s 1m27.195s 1m29.594s >>>> sys 7m44.199s 7m51.490s 8m21.149s >>>> >>>> patched(use copy_user_gigantic_page for 2M hugetlb too) >>>> real 0m3.182s 0m3.002s 0m2.963s >>>> user 1m19.456s 1m3.107s 1m6.447s >>>> sys 2m59.222s 3m10.899s 3m1.027s >>>> >>>> and sequential copy is better than the current implementation, >>>> so I will use sequential clear and copy. >>> Sorry, it appears that you misunderstanding my suggestion. I >>> suggest to >>> revise process_huge_page() to use more sequential memory clearing and >>> copying to improve its performance on your platform. >>> -- >>> Best Regards, >>> Huang, Ying >>> >>>>>> 2) case-anon-w-seq-hugetlb:(2M PMD HugeTLB) >>>>>> >>>>>> base: >>>>>> real 0m5.175s 0m5.117s 0m4.856s >>>>>> user 5m15.943s 5m7.567s 4m29.273s >>>>>> sys 2m38.503s 2m21.949s 2m21.252s >>>>>> >>>>>> patched: >>>>>> real 0m4.966s 0m4.841s 0m4.561s >>>>>> user 6m30.123s 6m9.516s 5m49.733s >>>>>> sys 0m58.503s 0m47.847s 0m46.785s >>>>>> >>>>>> >>>>>> This case is similar to the case1. >>>>>> >>>>>> 3) fallocate hugetlb 20G (2M PMD HugeTLB) >>>>>> >>>>>> base: >>>>>> real 0m3.016s 0m3.019s 0m3.018s >>>>>> user 0m0.000s 0m0.000s 0m0.000s >>>>>> sys 0m3.009s 0m3.012s 0m3.010s >>>>>> >>>>>> patched: >>>>>> >>>>>> real 0m1.136s 0m1.136s 0m1.136s >>>>>> user 0m0.000s 0m0.000s 0m0.004s >>>>>> sys 0m1.133s 0m1.133s 0m1.129s >>>>>> >>>>>> >>>>>> There is big win on patched kernel, and it is similar to above tmpfs >>>>>> test, so maybe we could revert the commit c79b57e462b5 ("mm: hugetlb: >>>>>> clear target sub-page last when clearing huge page"). >> >> I tried the following changes, >> diff --git a/mm/memory.c b/mm/memory.c >> index 66cf855dee3f..e5cc75adfa10 100644 >> --- a/mm/memory.c >> +++ b/mm/memory.c >> @@ -6777,7 +6777,7 @@ static inline int process_huge_page( >> base = 0; >> l = n; >> /* Process subpages at the end of huge page */ >> - for (i = nr_pages - 1; i >= 2 * n; i--) { >> + for (i = 2 * n; i < nr_pages; i++) { >> cond_resched(); >> ret = process_subpage(addr + i * PAGE_SIZE, i, >> arg); >> if (ret) >> >> Since n = 0, so the copying is from start to end now, but not >> improvement for case-anon-cow-seq-hugetlb, >> >> and if use copy_user_gigantic_pager, the time reduced from 6s to 3s >> >> diff --git a/mm/memory.c b/mm/memory.c >> index fe21bd3beff5..2c6532d21d84 100644 >> --- a/mm/memory.c >> +++ b/mm/memory.c >> @@ -6876,10 +6876,7 @@ int copy_user_large_folio(struct folio *dst, >> struct folio *src, >> .vma = vma, >> }; >> >> - if (unlikely(nr_pages > MAX_ORDER_NR_PAGES)) >> - return copy_user_gigantic_page(dst, src, addr_hint, >> vma, nr_pages); >> - >> - return process_huge_page(addr_hint, nr_pages, copy_subpage, &arg); >> + return copy_user_gigantic_page(dst, src, addr_hint, vma, nr_pages); >> } > > It appears that we have code generation issue here. Can you check it? > Whether code is inlined in the same way? > No different, and I checked the asm, both process_huge_page and copy_user_gigantic_page are inlined, it is strange... > Maybe we can start with > > modified mm/memory.c > @@ -6714,7 +6714,7 @@ EXPORT_SYMBOL(__might_fault); > * operation. The target subpage will be processed last to keep its > * cache lines hot. > */ > -static inline int process_huge_page( > +static __always_inline int process_huge_page( > unsigned long addr_hint, unsigned int nr_pages, > int (*process_subpage)(unsigned long addr, int idx, void *arg), > void *arg) > > -- > Best Regards, > Huang, Ying >
Kefeng Wang <wangkefeng.wang@huawei.com> writes: > On 2024/10/25 20:21, Huang, Ying wrote: >> Kefeng Wang <wangkefeng.wang@huawei.com> writes: >> >>> On 2024/10/25 15:47, Huang, Ying wrote: >>>> Kefeng Wang <wangkefeng.wang@huawei.com> writes: >>>> >>>>> On 2024/10/25 10:59, Huang, Ying wrote: >>>>>> Hi, Kefeng, >>>>>> Kefeng Wang <wangkefeng.wang@huawei.com> writes: >>>>>> >>>>>>> +CC Huang Ying, >>>>>>> >>>>>>> On 2024/10/23 6:56, Barry Song wrote: >>>>>>>> On Wed, Oct 23, 2024 at 4:10 AM Kefeng Wang <wangkefeng.wang@huawei.com> wrote: >>>>>>>>> >>>>>>>>> >>>>>>> ... >>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>> On 2024/10/17 23:09, Matthew Wilcox wrote: >>>>>>>>>>>>>>>>>>>>>>>> On Thu, Oct 17, 2024 at 10:25:04PM +0800, Kefeng Wang wrote: >>>>>>>>>>>>>>>>>>>>>>>>> Directly use folio_zero_range() to cleanup code. >>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>> Are you sure there's no performance regression introduced by this? >>>>>>>>>>>>>>>>>>>>>>>> clear_highpage() is often optimised in ways that we can't optimise for >>>>>>>>>>>>>>>>>>>>>>>> a plain memset(). On the other hand, if the folio is large, maybe a >>>>>>>>>>>>>>>>>>>>>>>> modern CPU will be able to do better than clear-one-page-at-a-time. >>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>> Right, I missing this, clear_page might be better than memset, I change >>>>>>>>>>>>>>>>>>>>>>> this one when look at the shmem_writepage(), which already convert to >>>>>>>>>>>>>>>>>>>>>>> use folio_zero_range() from clear_highpage(), also I grep >>>>>>>>>>>>>>>>>>>>>>> folio_zero_range(), there are some other to use folio_zero_range(). >>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(folio, 0, >>>>>>>>>>>>>>>>>>>>>>> folio_size(folio)); >>>>>>>>>>>>>>>>>>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(f, >>>>>>>>>>>>>>>>>>>>>>> 0, folio_size(f)); >>>>>>>>>>>>>>>>>>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(f, >>>>>>>>>>>>>>>>>>>>>>> 0, folio_size(f)); >>>>>>>>>>>>>>>>>>>>>>> fs/libfs.c: folio_zero_range(folio, 0, folio_size(folio)); >>>>>>>>>>>>>>>>>>>>>>> fs/ntfs3/frecord.c: folio_zero_range(folio, 0, >>>>>>>>>>>>>>>>>>>>>>> folio_size(folio)); >>>>>>>>>>>>>>>>>>>>>>> mm/page_io.c: folio_zero_range(folio, 0, folio_size(folio)); >>>>>>>>>>>>>>>>>>>>>>> mm/shmem.c: folio_zero_range(folio, 0, folio_size(folio)); >>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>> IOW, what performance testing have you done with this patch? >>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>> No performance test before, but I write a testcase, >>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>> 1) allocate N large folios (folio_alloc(PMD_ORDER)) >>>>>>>>>>>>>>>>>>>>>>> 2) then calculate the diff(us) when clear all N folios >>>>>>>>>>>>>>>>>>>>>>> clear_highpage/folio_zero_range/folio_zero_user >>>>>>>>>>>>>>>>>>>>>>> 3) release N folios >>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>> the result(run 5 times) shown below on my machine, >>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>> N=1, >>>>>>>>>>>>>>>>>>>>>>> clear_highpage folio_zero_range folio_zero_user >>>>>>>>>>>>>>>>>>>>>>> 1 69 74 177 >>>>>>>>>>>>>>>>>>>>>>> 2 57 62 168 >>>>>>>>>>>>>>>>>>>>>>> 3 54 58 234 >>>>>>>>>>>>>>>>>>>>>>> 4 54 58 157 >>>>>>>>>>>>>>>>>>>>>>> 5 56 62 148 >>>>>>>>>>>>>>>>>>>>>>> avg 58 62.8 176.8 >>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>> N=100 >>>>>>>>>>>>>>>>>>>>>>> clear_highpage folio_zero_range folio_zero_user >>>>>>>>>>>>>>>>>>>>>>> 1 11015 11309 32833 >>>>>>>>>>>>>>>>>>>>>>> 2 10385 11110 49751 >>>>>>>>>>>>>>>>>>>>>>> 3 10369 11056 33095 >>>>>>>>>>>>>>>>>>>>>>> 4 10332 11017 33106 >>>>>>>>>>>>>>>>>>>>>>> 5 10483 11000 49032 >>>>>>>>>>>>>>>>>>>>>>> avg 10516.8 11098.4 39563.4 >>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>> N=512 >>>>>>>>>>>>>>>>>>>>>>> clear_highpage folio_zero_range folio_zero_user >>>>>>>>>>>>>>>>>>>>>>> 1 55560 60055 156876 >>>>>>>>>>>>>>>>>>>>>>> 2 55485 60024 157132 >>>>>>>>>>>>>>>>>>>>>>> 3 55474 60129 156658 >>>>>>>>>>>>>>>>>>>>>>> 4 55555 59867 157259 >>>>>>>>>>>>>>>>>>>>>>> 5 55528 59932 157108 >>>>>>>>>>>>>>>>>>>>>>> avg 55520.4 60001.4 157006.6 >>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>> folio_zero_user with many cond_resched(), so time fluctuates a lot, >>>>>>>>>>>>>>>>>>>>>>> clear_highpage is better folio_zero_range as you said. >>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>> Maybe add a new helper to convert all folio_zero_range(folio, 0, >>>>>>>>>>>>>>>>>>>>>>> folio_size(folio)) >>>>>>>>>>>>>>>>>>>>>>> to use clear_highpage + flush_dcache_folio? >>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>> If this also improves performance for other existing callers of >>>>>>>>>>>>>>>>>>>>>> folio_zero_range(), then that's a positive outcome. >>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>> ... >>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>> hi Kefeng, >>>>>>>>>>>>>>>>>>> what's your point? providing a helper like clear_highfolio() or similar? >>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>> Yes, from above test, using clear_highpage/flush_dcache_folio is better >>>>>>>>>>>>>>>>>> than using folio_zero_range() for folio zero(especially for large >>>>>>>>>>>>>>>>>> folio), so I'd like to add a new helper, maybe name it folio_zero() >>>>>>>>>>>>>>>>>> since it zero the whole folio. >>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>> we already have a helper like folio_zero_user()? >>>>>>>>>>>>>>>>> it is not good enough? >>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>> Since it is with many cond_resched(), the performance is worst... >>>>>>>>>>>>>>> >>>>>>>>>>>>>>> Not exactly? It should have zero cost for a preemptible kernel. >>>>>>>>>>>>>>> For a non-preemptible kernel, it helps avoid clearing the folio >>>>>>>>>>>>>>> from occupying the CPU and starving other processes, right? >>>>>>>>>>>>>> >>>>>>>>>>>>>> --- a/mm/shmem.c >>>>>>>>>>>>>> +++ b/mm/shmem.c >>>>>>>>>>>>>> >>>>>>>>>>>>>> @@ -2393,10 +2393,7 @@ static int shmem_get_folio_gfp(struct inode >>>>>>>>>>>>>> *inode, pgoff_t index, >>>>>>>>>>>>>> * it now, lest undo on failure cancel our earlier guarantee. >>>>>>>>>>>>>> */ >>>>>>>>>>>>>> >>>>>>>>>>>>>> if (sgp != SGP_WRITE && !folio_test_uptodate(folio)) { >>>>>>>>>>>>>> - long i, n = folio_nr_pages(folio); >>>>>>>>>>>>>> - >>>>>>>>>>>>>> - for (i = 0; i < n; i++) >>>>>>>>>>>>>> - clear_highpage(folio_page(folio, i)); >>>>>>>>>>>>>> + folio_zero_user(folio, vmf->address); >>>>>>>>>>>>>> flush_dcache_folio(folio); >>>>>>>>>>>>>> folio_mark_uptodate(folio); >>>>>>>>>>>>>> } >>>>>>>>>>>>>> >>>>>>>>>>>>>> Do we perform better or worse with the following? >>>>>>>>>>>>> >>>>>>>>>>>>> Here is for SGP_FALLOC, vmf = NULL, we could use folio_zero_user(folio, >>>>>>>>>>>>> 0), I think the performance is worse, will retest once I can access >>>>>>>>>>>>> hardware. >>>>>>>>>>>> >>>>>>>>>>>> Perhaps, since the current code uses clear_hugepage(). Does using >>>>>>>>>>>> index << PAGE_SHIFT as the addr_hint offer any benefit? >>>>>>>>>>>> >>>>>>>>>>> >>>>>>>>>>> when use folio_zero_user(), the performance is vary bad with above >>>>>>>>>>> fallocate test(mount huge=always), >>>>>>>>>>> >>>>>>>>>>> folio_zero_range clear_highpage folio_zero_user >>>>>>>>>>> real 0m1.214s 0m1.111s 0m3.159s >>>>>>>>>>> user 0m0.000s 0m0.000s 0m0.000s >>>>>>>>>>> sys 0m1.210s 0m1.109s 0m3.152s >>>>>>>>>>> >>>>>>>>>>> I tried with addr_hint = 0/index << PAGE_SHIFT, no obvious different. >>>>>>>>>> >>>>>>>>>> Interesting. Does your kernel have preemption disabled or >>>>>>>>>> preemption_debug enabled? >>>>>>>>> >>>>>>>>> ARM64 server, CONFIG_PREEMPT_NONE=y >>>>>>>> this explains why the performance is much worse. >>>>>>>> >>>>>>>>> >>>>>>>>>> >>>>>>>>>> If not, it makes me wonder whether folio_zero_user() in >>>>>>>>>> alloc_anon_folio() is actually improving performance as expected, >>>>>>>>>> compared to the simpler folio_zero() you plan to implement. :-) >>>>>>>>> >>>>>>>>> Yes, maybe, the folio_zero_user(was clear_huge_page) is from >>>>>>>>> 47ad8475c000 ("thp: clear_copy_huge_page"), so original clear_huge_page >>>>>>>>> is used in HugeTLB, clear PUD size maybe spend many time, but for PMD or >>>>>>>>> other size of large folio, cond_resched is not necessary since we >>>>>>>>> already have some folio_zero_range() to clear large folio, and no issue >>>>>>>>> was reported. >>>>>>>> probably worth an optimization. calling cond_resched() for each page >>>>>>>> seems too aggressive and useless. >>>>>>> >>>>>>> After some test, I think the cond_resched() is not the root cause, >>>>>>> no performance gained with batched cond_resched(), even I kill >>>>>>> cond_resched() from process_huge_page, no improvement. >>>>>>> >>>>>>> But when I unconditionally use clear_gigantic_page() in >>>>>>> folio_zero_user(patched), there is big improvement with above >>>>>>> fallocate on tmpfs(mount huge=always), also I test some other testcase, >>>>>>> >>>>>>> >>>>>>> 1) case-anon-w-seq-mt: (2M PMD THP) >>>>>>> >>>>>>> base: >>>>>>> real 0m2.490s 0m2.254s 0m2.272s >>>>>>> user 1m59.980s 2m23.431s 2m18.739s >>>>>>> sys 1m3.675s 1m15.462s 1m15.030s >>>>>>> >>>>>>> patched: >>>>>>> real 0m2.234s 0m2.225s 0m2.159s >>>>>>> user 2m56.105s 2m57.117s 3m0.489s >>>>>>> sys 0m17.064s 0m17.564s 0m16.150s >>>>>>> >>>>>>> Patched kernel win on sys and bad in user, but real is almost same, >>>>>>> maybe a little better than base. >>>>>> We can find user time difference. That means the original cache hot >>>>>> behavior still applies on your system. >>>>>> However, it appears that the performance to clear page from end to >>>>>> begin >>>>>> is really bad on your system. >>>>>> So, I suggest to revise the current implementation to use sequential >>>>>> clearing as much as possible. >>>>>> >>>>> >>>>> I test case-anon-cow-seq-hugetlb for copy_user_large_folio() >>>>> >>>>> base: >>>>> real 0m6.259s 0m6.197s 0m6.316s >>>>> user 1m31.176s 1m27.195s 1m29.594s >>>>> sys 7m44.199s 7m51.490s 8m21.149s >>>>> >>>>> patched(use copy_user_gigantic_page for 2M hugetlb too) >>>>> real 0m3.182s 0m3.002s 0m2.963s >>>>> user 1m19.456s 1m3.107s 1m6.447s >>>>> sys 2m59.222s 3m10.899s 3m1.027s >>>>> >>>>> and sequential copy is better than the current implementation, >>>>> so I will use sequential clear and copy. >>>> Sorry, it appears that you misunderstanding my suggestion. I >>>> suggest to >>>> revise process_huge_page() to use more sequential memory clearing and >>>> copying to improve its performance on your platform. >>>> -- >>>> Best Regards, >>>> Huang, Ying >>>> >>>>>>> 2) case-anon-w-seq-hugetlb:(2M PMD HugeTLB) >>>>>>> >>>>>>> base: >>>>>>> real 0m5.175s 0m5.117s 0m4.856s >>>>>>> user 5m15.943s 5m7.567s 4m29.273s >>>>>>> sys 2m38.503s 2m21.949s 2m21.252s >>>>>>> >>>>>>> patched: >>>>>>> real 0m4.966s 0m4.841s 0m4.561s >>>>>>> user 6m30.123s 6m9.516s 5m49.733s >>>>>>> sys 0m58.503s 0m47.847s 0m46.785s >>>>>>> >>>>>>> >>>>>>> This case is similar to the case1. >>>>>>> >>>>>>> 3) fallocate hugetlb 20G (2M PMD HugeTLB) >>>>>>> >>>>>>> base: >>>>>>> real 0m3.016s 0m3.019s 0m3.018s >>>>>>> user 0m0.000s 0m0.000s 0m0.000s >>>>>>> sys 0m3.009s 0m3.012s 0m3.010s >>>>>>> >>>>>>> patched: >>>>>>> >>>>>>> real 0m1.136s 0m1.136s 0m1.136s >>>>>>> user 0m0.000s 0m0.000s 0m0.004s >>>>>>> sys 0m1.133s 0m1.133s 0m1.129s >>>>>>> >>>>>>> >>>>>>> There is big win on patched kernel, and it is similar to above tmpfs >>>>>>> test, so maybe we could revert the commit c79b57e462b5 ("mm: hugetlb: >>>>>>> clear target sub-page last when clearing huge page"). >>> >>> I tried the following changes, >>> diff --git a/mm/memory.c b/mm/memory.c >>> index 66cf855dee3f..e5cc75adfa10 100644 >>> --- a/mm/memory.c >>> +++ b/mm/memory.c >>> @@ -6777,7 +6777,7 @@ static inline int process_huge_page( >>> base = 0; >>> l = n; >>> /* Process subpages at the end of huge page */ >>> - for (i = nr_pages - 1; i >= 2 * n; i--) { >>> + for (i = 2 * n; i < nr_pages; i++) { >>> cond_resched(); >>> ret = process_subpage(addr + i * PAGE_SIZE, i, >>> arg); >>> if (ret) >>> >>> Since n = 0, so the copying is from start to end now, but not >>> improvement for case-anon-cow-seq-hugetlb, >>> >>> and if use copy_user_gigantic_pager, the time reduced from 6s to 3s >>> >>> diff --git a/mm/memory.c b/mm/memory.c >>> index fe21bd3beff5..2c6532d21d84 100644 >>> --- a/mm/memory.c >>> +++ b/mm/memory.c >>> @@ -6876,10 +6876,7 @@ int copy_user_large_folio(struct folio *dst, >>> struct folio *src, >>> .vma = vma, >>> }; >>> >>> - if (unlikely(nr_pages > MAX_ORDER_NR_PAGES)) >>> - return copy_user_gigantic_page(dst, src, addr_hint, >>> vma, nr_pages); >>> - >>> - return process_huge_page(addr_hint, nr_pages, copy_subpage, &arg); >>> + return copy_user_gigantic_page(dst, src, addr_hint, vma, nr_pages); >>> } >> It appears that we have code generation issue here. Can you check >> it? >> Whether code is inlined in the same way? >> > > No different, and I checked the asm, both process_huge_page and > copy_user_gigantic_page are inlined, it is strange... It's not inlined in my configuration. And __always_inline below changes it for me. If it's already inlined and the code is actually almost same, why there's difference? Is it possible for you to do some profile or further analysis? >> Maybe we can start with >> modified mm/memory.c >> @@ -6714,7 +6714,7 @@ EXPORT_SYMBOL(__might_fault); >> * operation. The target subpage will be processed last to keep its >> * cache lines hot. >> */ >> -static inline int process_huge_page( >> +static __always_inline int process_huge_page( >> unsigned long addr_hint, unsigned int nr_pages, >> int (*process_subpage)(unsigned long addr, int idx, void *arg), >> void *arg) -- Best Regards, Huang, Ying
On 2024/10/28 10:39, Huang, Ying wrote: > Kefeng Wang <wangkefeng.wang@huawei.com> writes: > >> On 2024/10/25 20:21, Huang, Ying wrote: >>> Kefeng Wang <wangkefeng.wang@huawei.com> writes: >>> >>>> On 2024/10/25 15:47, Huang, Ying wrote: >>>>> Kefeng Wang <wangkefeng.wang@huawei.com> writes: >>>>> >>>>>> On 2024/10/25 10:59, Huang, Ying wrote: >>>>>>> Hi, Kefeng, >>>>>>> Kefeng Wang <wangkefeng.wang@huawei.com> writes: >>>>>>> >>>>>>>> +CC Huang Ying, >>>>>>>> >>>>>>>> On 2024/10/23 6:56, Barry Song wrote: >>>>>>>>> On Wed, Oct 23, 2024 at 4:10 AM Kefeng Wang <wangkefeng.wang@huawei.com> wrote: >>>>>>>>>> >>>>>>>>>> >>>>>>>> ... >>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>> On 2024/10/17 23:09, Matthew Wilcox wrote: >>>>>>>>>>>>>>>>>>>>>>>>> On Thu, Oct 17, 2024 at 10:25:04PM +0800, Kefeng Wang wrote: >>>>>>>>>>>>>>>>>>>>>>>>>> Directly use folio_zero_range() to cleanup code. >>>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>>> Are you sure there's no performance regression introduced by this? >>>>>>>>>>>>>>>>>>>>>>>>> clear_highpage() is often optimised in ways that we can't optimise for >>>>>>>>>>>>>>>>>>>>>>>>> a plain memset(). On the other hand, if the folio is large, maybe a >>>>>>>>>>>>>>>>>>>>>>>>> modern CPU will be able to do better than clear-one-page-at-a-time. >>>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>> Right, I missing this, clear_page might be better than memset, I change >>>>>>>>>>>>>>>>>>>>>>>> this one when look at the shmem_writepage(), which already convert to >>>>>>>>>>>>>>>>>>>>>>>> use folio_zero_range() from clear_highpage(), also I grep >>>>>>>>>>>>>>>>>>>>>>>> folio_zero_range(), there are some other to use folio_zero_range(). >>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(folio, 0, >>>>>>>>>>>>>>>>>>>>>>>> folio_size(folio)); >>>>>>>>>>>>>>>>>>>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(f, >>>>>>>>>>>>>>>>>>>>>>>> 0, folio_size(f)); >>>>>>>>>>>>>>>>>>>>>>>> fs/bcachefs/fs-io-buffered.c: folio_zero_range(f, >>>>>>>>>>>>>>>>>>>>>>>> 0, folio_size(f)); >>>>>>>>>>>>>>>>>>>>>>>> fs/libfs.c: folio_zero_range(folio, 0, folio_size(folio)); >>>>>>>>>>>>>>>>>>>>>>>> fs/ntfs3/frecord.c: folio_zero_range(folio, 0, >>>>>>>>>>>>>>>>>>>>>>>> folio_size(folio)); >>>>>>>>>>>>>>>>>>>>>>>> mm/page_io.c: folio_zero_range(folio, 0, folio_size(folio)); >>>>>>>>>>>>>>>>>>>>>>>> mm/shmem.c: folio_zero_range(folio, 0, folio_size(folio)); >>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>>> IOW, what performance testing have you done with this patch? >>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>> No performance test before, but I write a testcase, >>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>> 1) allocate N large folios (folio_alloc(PMD_ORDER)) >>>>>>>>>>>>>>>>>>>>>>>> 2) then calculate the diff(us) when clear all N folios >>>>>>>>>>>>>>>>>>>>>>>> clear_highpage/folio_zero_range/folio_zero_user >>>>>>>>>>>>>>>>>>>>>>>> 3) release N folios >>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>> the result(run 5 times) shown below on my machine, >>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>> N=1, >>>>>>>>>>>>>>>>>>>>>>>> clear_highpage folio_zero_range folio_zero_user >>>>>>>>>>>>>>>>>>>>>>>> 1 69 74 177 >>>>>>>>>>>>>>>>>>>>>>>> 2 57 62 168 >>>>>>>>>>>>>>>>>>>>>>>> 3 54 58 234 >>>>>>>>>>>>>>>>>>>>>>>> 4 54 58 157 >>>>>>>>>>>>>>>>>>>>>>>> 5 56 62 148 >>>>>>>>>>>>>>>>>>>>>>>> avg 58 62.8 176.8 >>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>> N=100 >>>>>>>>>>>>>>>>>>>>>>>> clear_highpage folio_zero_range folio_zero_user >>>>>>>>>>>>>>>>>>>>>>>> 1 11015 11309 32833 >>>>>>>>>>>>>>>>>>>>>>>> 2 10385 11110 49751 >>>>>>>>>>>>>>>>>>>>>>>> 3 10369 11056 33095 >>>>>>>>>>>>>>>>>>>>>>>> 4 10332 11017 33106 >>>>>>>>>>>>>>>>>>>>>>>> 5 10483 11000 49032 >>>>>>>>>>>>>>>>>>>>>>>> avg 10516.8 11098.4 39563.4 >>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>> N=512 >>>>>>>>>>>>>>>>>>>>>>>> clear_highpage folio_zero_range folio_zero_user >>>>>>>>>>>>>>>>>>>>>>>> 1 55560 60055 156876 >>>>>>>>>>>>>>>>>>>>>>>> 2 55485 60024 157132 >>>>>>>>>>>>>>>>>>>>>>>> 3 55474 60129 156658 >>>>>>>>>>>>>>>>>>>>>>>> 4 55555 59867 157259 >>>>>>>>>>>>>>>>>>>>>>>> 5 55528 59932 157108 >>>>>>>>>>>>>>>>>>>>>>>> avg 55520.4 60001.4 157006.6 >>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>> folio_zero_user with many cond_resched(), so time fluctuates a lot, >>>>>>>>>>>>>>>>>>>>>>>> clear_highpage is better folio_zero_range as you said. >>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>> Maybe add a new helper to convert all folio_zero_range(folio, 0, >>>>>>>>>>>>>>>>>>>>>>>> folio_size(folio)) >>>>>>>>>>>>>>>>>>>>>>>> to use clear_highpage + flush_dcache_folio? >>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>> If this also improves performance for other existing callers of >>>>>>>>>>>>>>>>>>>>>>> folio_zero_range(), then that's a positive outcome. >>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>> ... >>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>> hi Kefeng, >>>>>>>>>>>>>>>>>>>> what's your point? providing a helper like clear_highfolio() or similar? >>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>> Yes, from above test, using clear_highpage/flush_dcache_folio is better >>>>>>>>>>>>>>>>>>> than using folio_zero_range() for folio zero(especially for large >>>>>>>>>>>>>>>>>>> folio), so I'd like to add a new helper, maybe name it folio_zero() >>>>>>>>>>>>>>>>>>> since it zero the whole folio. >>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>> we already have a helper like folio_zero_user()? >>>>>>>>>>>>>>>>>> it is not good enough? >>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>> Since it is with many cond_resched(), the performance is worst... >>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>> Not exactly? It should have zero cost for a preemptible kernel. >>>>>>>>>>>>>>>> For a non-preemptible kernel, it helps avoid clearing the folio >>>>>>>>>>>>>>>> from occupying the CPU and starving other processes, right? >>>>>>>>>>>>>>> >>>>>>>>>>>>>>> --- a/mm/shmem.c >>>>>>>>>>>>>>> +++ b/mm/shmem.c >>>>>>>>>>>>>>> >>>>>>>>>>>>>>> @@ -2393,10 +2393,7 @@ static int shmem_get_folio_gfp(struct inode >>>>>>>>>>>>>>> *inode, pgoff_t index, >>>>>>>>>>>>>>> * it now, lest undo on failure cancel our earlier guarantee. >>>>>>>>>>>>>>> */ >>>>>>>>>>>>>>> >>>>>>>>>>>>>>> if (sgp != SGP_WRITE && !folio_test_uptodate(folio)) { >>>>>>>>>>>>>>> - long i, n = folio_nr_pages(folio); >>>>>>>>>>>>>>> - >>>>>>>>>>>>>>> - for (i = 0; i < n; i++) >>>>>>>>>>>>>>> - clear_highpage(folio_page(folio, i)); >>>>>>>>>>>>>>> + folio_zero_user(folio, vmf->address); >>>>>>>>>>>>>>> flush_dcache_folio(folio); >>>>>>>>>>>>>>> folio_mark_uptodate(folio); >>>>>>>>>>>>>>> } >>>>>>>>>>>>>>> >>>>>>>>>>>>>>> Do we perform better or worse with the following? >>>>>>>>>>>>>> >>>>>>>>>>>>>> Here is for SGP_FALLOC, vmf = NULL, we could use folio_zero_user(folio, >>>>>>>>>>>>>> 0), I think the performance is worse, will retest once I can access >>>>>>>>>>>>>> hardware. >>>>>>>>>>>>> >>>>>>>>>>>>> Perhaps, since the current code uses clear_hugepage(). Does using >>>>>>>>>>>>> index << PAGE_SHIFT as the addr_hint offer any benefit? >>>>>>>>>>>>> >>>>>>>>>>>> >>>>>>>>>>>> when use folio_zero_user(), the performance is vary bad with above >>>>>>>>>>>> fallocate test(mount huge=always), >>>>>>>>>>>> >>>>>>>>>>>> folio_zero_range clear_highpage folio_zero_user >>>>>>>>>>>> real 0m1.214s 0m1.111s 0m3.159s >>>>>>>>>>>> user 0m0.000s 0m0.000s 0m0.000s >>>>>>>>>>>> sys 0m1.210s 0m1.109s 0m3.152s >>>>>>>>>>>> >>>>>>>>>>>> I tried with addr_hint = 0/index << PAGE_SHIFT, no obvious different. >>>>>>>>>>> >>>>>>>>>>> Interesting. Does your kernel have preemption disabled or >>>>>>>>>>> preemption_debug enabled? >>>>>>>>>> >>>>>>>>>> ARM64 server, CONFIG_PREEMPT_NONE=y >>>>>>>>> this explains why the performance is much worse. >>>>>>>>> >>>>>>>>>> >>>>>>>>>>> >>>>>>>>>>> If not, it makes me wonder whether folio_zero_user() in >>>>>>>>>>> alloc_anon_folio() is actually improving performance as expected, >>>>>>>>>>> compared to the simpler folio_zero() you plan to implement. :-) >>>>>>>>>> >>>>>>>>>> Yes, maybe, the folio_zero_user(was clear_huge_page) is from >>>>>>>>>> 47ad8475c000 ("thp: clear_copy_huge_page"), so original clear_huge_page >>>>>>>>>> is used in HugeTLB, clear PUD size maybe spend many time, but for PMD or >>>>>>>>>> other size of large folio, cond_resched is not necessary since we >>>>>>>>>> already have some folio_zero_range() to clear large folio, and no issue >>>>>>>>>> was reported. >>>>>>>>> probably worth an optimization. calling cond_resched() for each page >>>>>>>>> seems too aggressive and useless. >>>>>>>> >>>>>>>> After some test, I think the cond_resched() is not the root cause, >>>>>>>> no performance gained with batched cond_resched(), even I kill >>>>>>>> cond_resched() from process_huge_page, no improvement. >>>>>>>> >>>>>>>> But when I unconditionally use clear_gigantic_page() in >>>>>>>> folio_zero_user(patched), there is big improvement with above >>>>>>>> fallocate on tmpfs(mount huge=always), also I test some other testcase, >>>>>>>> >>>>>>>> >>>>>>>> 1) case-anon-w-seq-mt: (2M PMD THP) >>>>>>>> >>>>>>>> base: >>>>>>>> real 0m2.490s 0m2.254s 0m2.272s >>>>>>>> user 1m59.980s 2m23.431s 2m18.739s >>>>>>>> sys 1m3.675s 1m15.462s 1m15.030s >>>>>>>> >>>>>>>> patched: >>>>>>>> real 0m2.234s 0m2.225s 0m2.159s >>>>>>>> user 2m56.105s 2m57.117s 3m0.489s >>>>>>>> sys 0m17.064s 0m17.564s 0m16.150s >>>>>>>> >>>>>>>> Patched kernel win on sys and bad in user, but real is almost same, >>>>>>>> maybe a little better than base. >>>>>>> We can find user time difference. That means the original cache hot >>>>>>> behavior still applies on your system. >>>>>>> However, it appears that the performance to clear page from end to >>>>>>> begin >>>>>>> is really bad on your system. >>>>>>> So, I suggest to revise the current implementation to use sequential >>>>>>> clearing as much as possible. >>>>>>> >>>>>> >>>>>> I test case-anon-cow-seq-hugetlb for copy_user_large_folio() >>>>>> >>>>>> base: >>>>>> real 0m6.259s 0m6.197s 0m6.316s >>>>>> user 1m31.176s 1m27.195s 1m29.594s >>>>>> sys 7m44.199s 7m51.490s 8m21.149s >>>>>> >>>>>> patched(use copy_user_gigantic_page for 2M hugetlb too) >>>>>> real 0m3.182s 0m3.002s 0m2.963s >>>>>> user 1m19.456s 1m3.107s 1m6.447s >>>>>> sys 2m59.222s 3m10.899s 3m1.027s >>>>>> >>>>>> and sequential copy is better than the current implementation, >>>>>> so I will use sequential clear and copy. >>>>> Sorry, it appears that you misunderstanding my suggestion. I >>>>> suggest to >>>>> revise process_huge_page() to use more sequential memory clearing and >>>>> copying to improve its performance on your platform. >>>>> -- >>>>> Best Regards, >>>>> Huang, Ying >>>>> >>>>>>>> 2) case-anon-w-seq-hugetlb:(2M PMD HugeTLB) >>>>>>>> >>>>>>>> base: >>>>>>>> real 0m5.175s 0m5.117s 0m4.856s >>>>>>>> user 5m15.943s 5m7.567s 4m29.273s >>>>>>>> sys 2m38.503s 2m21.949s 2m21.252s >>>>>>>> >>>>>>>> patched: >>>>>>>> real 0m4.966s 0m4.841s 0m4.561s >>>>>>>> user 6m30.123s 6m9.516s 5m49.733s >>>>>>>> sys 0m58.503s 0m47.847s 0m46.785s >>>>>>>> >>>>>>>> >>>>>>>> This case is similar to the case1. >>>>>>>> >>>>>>>> 3) fallocate hugetlb 20G (2M PMD HugeTLB) >>>>>>>> >>>>>>>> base: >>>>>>>> real 0m3.016s 0m3.019s 0m3.018s >>>>>>>> user 0m0.000s 0m0.000s 0m0.000s >>>>>>>> sys 0m3.009s 0m3.012s 0m3.010s >>>>>>>> >>>>>>>> patched: >>>>>>>> >>>>>>>> real 0m1.136s 0m1.136s 0m1.136s >>>>>>>> user 0m0.000s 0m0.000s 0m0.004s >>>>>>>> sys 0m1.133s 0m1.133s 0m1.129s >>>>>>>> >>>>>>>> >>>>>>>> There is big win on patched kernel, and it is similar to above tmpfs >>>>>>>> test, so maybe we could revert the commit c79b57e462b5 ("mm: hugetlb: >>>>>>>> clear target sub-page last when clearing huge page"). >>>> >>>> I tried the following changes, >>>> diff --git a/mm/memory.c b/mm/memory.c >>>> index 66cf855dee3f..e5cc75adfa10 100644 >>>> --- a/mm/memory.c >>>> +++ b/mm/memory.c >>>> @@ -6777,7 +6777,7 @@ static inline int process_huge_page( >>>> base = 0; >>>> l = n; >>>> /* Process subpages at the end of huge page */ >>>> - for (i = nr_pages - 1; i >= 2 * n; i--) { >>>> + for (i = 2 * n; i < nr_pages; i++) { >>>> cond_resched(); >>>> ret = process_subpage(addr + i * PAGE_SIZE, i, >>>> arg); >>>> if (ret) >>>> >>>> Since n = 0, so the copying is from start to end now, but not >>>> improvement for case-anon-cow-seq-hugetlb, >>>> >>>> and if use copy_user_gigantic_pager, the time reduced from 6s to 3s >>>> >>>> diff --git a/mm/memory.c b/mm/memory.c >>>> index fe21bd3beff5..2c6532d21d84 100644 >>>> --- a/mm/memory.c >>>> +++ b/mm/memory.c >>>> @@ -6876,10 +6876,7 @@ int copy_user_large_folio(struct folio *dst, >>>> struct folio *src, >>>> .vma = vma, >>>> }; >>>> >>>> - if (unlikely(nr_pages > MAX_ORDER_NR_PAGES)) >>>> - return copy_user_gigantic_page(dst, src, addr_hint, >>>> vma, nr_pages); >>>> - >>>> - return process_huge_page(addr_hint, nr_pages, copy_subpage, &arg); >>>> + return copy_user_gigantic_page(dst, src, addr_hint, vma, nr_pages); >>>> } >>> It appears that we have code generation issue here. Can you check >>> it? >>> Whether code is inlined in the same way? >>> >> >> No different, and I checked the asm, both process_huge_page and >> copy_user_gigantic_page are inlined, it is strange... > > It's not inlined in my configuration. And __always_inline below changes > it for me. > > If it's already inlined and the code is actually almost same, why > there's difference? Is it possible for you to do some profile or > further analysis? Yes, will continue to debug this. > >>> Maybe we can start with >>> modified mm/memory.c >>> @@ -6714,7 +6714,7 @@ EXPORT_SYMBOL(__might_fault); >>> * operation. The target subpage will be processed last to keep its >>> * cache lines hot. >>> */ >>> -static inline int process_huge_page( >>> +static __always_inline int process_huge_page( >>> unsigned long addr_hint, unsigned int nr_pages, >>> int (*process_subpage)(unsigned long addr, int idx, void *arg), >>> void *arg) > > -- > Best Regards, > Huang, Ying >
On 2024/10/28 14:37, Kefeng Wang wrote: > > > On 2024/10/28 10:39, Huang, Ying wrote: >> Kefeng Wang <wangkefeng.wang@huawei.com> writes: >> >>> On 2024/10/25 20:21, Huang, Ying wrote: >>>> Kefeng Wang <wangkefeng.wang@huawei.com> writes: >>>> >>>>> On 2024/10/25 15:47, Huang, Ying wrote: >>>>>> Kefeng Wang <wangkefeng.wang@huawei.com> writes: >>>>>> >>>>>>> On 2024/10/25 10:59, Huang, Ying wrote: >>>>>>>> Hi, Kefeng, >>>>>>>> Kefeng Wang <wangkefeng.wang@huawei.com> writes: >>>>>>>> >>>>>>>>> +CC Huang Ying, >>>>>>>>> >>>>>>>>> On 2024/10/23 6:56, Barry Song wrote: >>>>>>>>>> On Wed, Oct 23, 2024 at 4:10 AM Kefeng Wang >>>>>>>>>> <wangkefeng.wang@huawei.com> wrote: >>>>>>>>>>> >>>>>>>>>>> >>>>>>>>> ... >>>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>>> On 2024/10/17 23:09, Matthew Wilcox wrote: >>>>>>>>>>>>>>>>>>>>>>>>>> On Thu, Oct 17, 2024 at 10:25:04PM +0800, >>>>>>>>>>>>>>>>>>>>>>>>>> Kefeng Wang wrote: >>>>>>>>>>>>>>>>>>>>>>>>>>> Directly use folio_zero_range() to cleanup code. >>>>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>>>> Are you sure there's no performance regression >>>>>>>>>>>>>>>>>>>>>>>>>> introduced by this? >>>>>>>>>>>>>>>>>>>>>>>>>> clear_highpage() is often optimised in ways >>>>>>>>>>>>>>>>>>>>>>>>>> that we can't optimise for >>>>>>>>>>>>>>>>>>>>>>>>>> a plain memset(). On the other hand, if the >>>>>>>>>>>>>>>>>>>>>>>>>> folio is large, maybe a >>>>>>>>>>>>>>>>>>>>>>>>>> modern CPU will be able to do better than >>>>>>>>>>>>>>>>>>>>>>>>>> clear-one-page-at-a-time. >>>>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>>> Right, I missing this, clear_page might be >>>>>>>>>>>>>>>>>>>>>>>>> better than memset, I change >>>>>>>>>>>>>>>>>>>>>>>>> this one when look at the shmem_writepage(), >>>>>>>>>>>>>>>>>>>>>>>>> which already convert to >>>>>>>>>>>>>>>>>>>>>>>>> use folio_zero_range() from clear_highpage(), >>>>>>>>>>>>>>>>>>>>>>>>> also I grep >>>>>>>>>>>>>>>>>>>>>>>>> folio_zero_range(), there are some other to use >>>>>>>>>>>>>>>>>>>>>>>>> folio_zero_range(). >>>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>>> fs/bcachefs/fs-io-buffered.c: >>>>>>>>>>>>>>>>>>>>>>>>> folio_zero_range(folio, 0, >>>>>>>>>>>>>>>>>>>>>>>>> folio_size(folio)); >>>>>>>>>>>>>>>>>>>>>>>>> fs/bcachefs/fs-io-buffered.c: >>>>>>>>>>>>>>>>>>>>>>>>> folio_zero_range(f, >>>>>>>>>>>>>>>>>>>>>>>>> 0, folio_size(f)); >>>>>>>>>>>>>>>>>>>>>>>>> fs/bcachefs/fs-io-buffered.c: >>>>>>>>>>>>>>>>>>>>>>>>> folio_zero_range(f, >>>>>>>>>>>>>>>>>>>>>>>>> 0, folio_size(f)); >>>>>>>>>>>>>>>>>>>>>>>>> fs/libfs.c: folio_zero_range(folio, 0, >>>>>>>>>>>>>>>>>>>>>>>>> folio_size(folio)); >>>>>>>>>>>>>>>>>>>>>>>>> fs/ntfs3/frecord.c: >>>>>>>>>>>>>>>>>>>>>>>>> folio_zero_range(folio, 0, >>>>>>>>>>>>>>>>>>>>>>>>> folio_size(folio)); >>>>>>>>>>>>>>>>>>>>>>>>> mm/page_io.c: folio_zero_range(folio, 0, >>>>>>>>>>>>>>>>>>>>>>>>> folio_size(folio)); >>>>>>>>>>>>>>>>>>>>>>>>> mm/shmem.c: folio_zero_range(folio, >>>>>>>>>>>>>>>>>>>>>>>>> 0, folio_size(folio)); >>>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>>>> IOW, what performance testing have you done >>>>>>>>>>>>>>>>>>>>>>>>>> with this patch? >>>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>>> No performance test before, but I write a >>>>>>>>>>>>>>>>>>>>>>>>> testcase, >>>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>>> 1) allocate N large folios >>>>>>>>>>>>>>>>>>>>>>>>> (folio_alloc(PMD_ORDER)) >>>>>>>>>>>>>>>>>>>>>>>>> 2) then calculate the diff(us) when clear all N >>>>>>>>>>>>>>>>>>>>>>>>> folios >>>>>>>>>>>>>>>>>>>>>>>>> clear_highpage/folio_zero_range/ >>>>>>>>>>>>>>>>>>>>>>>>> folio_zero_user >>>>>>>>>>>>>>>>>>>>>>>>> 3) release N folios >>>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>>> the result(run 5 times) shown below on my machine, >>>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>>> N=1, >>>>>>>>>>>>>>>>>>>>>>>>> clear_highpage >>>>>>>>>>>>>>>>>>>>>>>>> folio_zero_range folio_zero_user >>>>>>>>>>>>>>>>>>>>>>>>> 1 69 74 >>>>>>>>>>>>>>>>>>>>>>>>> 177 >>>>>>>>>>>>>>>>>>>>>>>>> 2 57 62 >>>>>>>>>>>>>>>>>>>>>>>>> 168 >>>>>>>>>>>>>>>>>>>>>>>>> 3 54 58 >>>>>>>>>>>>>>>>>>>>>>>>> 234 >>>>>>>>>>>>>>>>>>>>>>>>> 4 54 58 >>>>>>>>>>>>>>>>>>>>>>>>> 157 >>>>>>>>>>>>>>>>>>>>>>>>> 5 56 62 >>>>>>>>>>>>>>>>>>>>>>>>> 148 >>>>>>>>>>>>>>>>>>>>>>>>> avg 58 62.8 >>>>>>>>>>>>>>>>>>>>>>>>> 176.8 >>>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>>> N=100 >>>>>>>>>>>>>>>>>>>>>>>>> clear_highpage >>>>>>>>>>>>>>>>>>>>>>>>> folio_zero_range folio_zero_user >>>>>>>>>>>>>>>>>>>>>>>>> 1 11015 11309 >>>>>>>>>>>>>>>>>>>>>>>>> 32833 >>>>>>>>>>>>>>>>>>>>>>>>> 2 10385 11110 >>>>>>>>>>>>>>>>>>>>>>>>> 49751 >>>>>>>>>>>>>>>>>>>>>>>>> 3 10369 11056 >>>>>>>>>>>>>>>>>>>>>>>>> 33095 >>>>>>>>>>>>>>>>>>>>>>>>> 4 10332 11017 >>>>>>>>>>>>>>>>>>>>>>>>> 33106 >>>>>>>>>>>>>>>>>>>>>>>>> 5 10483 11000 >>>>>>>>>>>>>>>>>>>>>>>>> 49032 >>>>>>>>>>>>>>>>>>>>>>>>> avg 10516.8 11098.4 >>>>>>>>>>>>>>>>>>>>>>>>> 39563.4 >>>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>>> N=512 >>>>>>>>>>>>>>>>>>>>>>>>> clear_highpage >>>>>>>>>>>>>>>>>>>>>>>>> folio_zero_range folio_zero_user >>>>>>>>>>>>>>>>>>>>>>>>> 1 55560 60055 >>>>>>>>>>>>>>>>>>>>>>>>> 156876 >>>>>>>>>>>>>>>>>>>>>>>>> 2 55485 60024 >>>>>>>>>>>>>>>>>>>>>>>>> 157132 >>>>>>>>>>>>>>>>>>>>>>>>> 3 55474 60129 >>>>>>>>>>>>>>>>>>>>>>>>> 156658 >>>>>>>>>>>>>>>>>>>>>>>>> 4 55555 59867 >>>>>>>>>>>>>>>>>>>>>>>>> 157259 >>>>>>>>>>>>>>>>>>>>>>>>> 5 55528 59932 >>>>>>>>>>>>>>>>>>>>>>>>> 157108 >>>>>>>>>>>>>>>>>>>>>>>>> avg 55520.4 60001.4 >>>>>>>>>>>>>>>>>>>>>>>>> 157006.6 >>>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>>> folio_zero_user with many cond_resched(), so >>>>>>>>>>>>>>>>>>>>>>>>> time fluctuates a lot, >>>>>>>>>>>>>>>>>>>>>>>>> clear_highpage is better folio_zero_range as >>>>>>>>>>>>>>>>>>>>>>>>> you said. >>>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>>> Maybe add a new helper to convert all >>>>>>>>>>>>>>>>>>>>>>>>> folio_zero_range(folio, 0, >>>>>>>>>>>>>>>>>>>>>>>>> folio_size(folio)) >>>>>>>>>>>>>>>>>>>>>>>>> to use clear_highpage + flush_dcache_folio? >>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>> If this also improves performance for other >>>>>>>>>>>>>>>>>>>>>>>> existing callers of >>>>>>>>>>>>>>>>>>>>>>>> folio_zero_range(), then that's a positive outcome. >>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>> ... >>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>> hi Kefeng, >>>>>>>>>>>>>>>>>>>>> what's your point? providing a helper like >>>>>>>>>>>>>>>>>>>>> clear_highfolio() or similar? >>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>> Yes, from above test, using clear_highpage/ >>>>>>>>>>>>>>>>>>>> flush_dcache_folio is better >>>>>>>>>>>>>>>>>>>> than using folio_zero_range() for folio >>>>>>>>>>>>>>>>>>>> zero(especially for large >>>>>>>>>>>>>>>>>>>> folio), so I'd like to add a new helper, maybe name >>>>>>>>>>>>>>>>>>>> it folio_zero() >>>>>>>>>>>>>>>>>>>> since it zero the whole folio. >>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>> we already have a helper like folio_zero_user()? >>>>>>>>>>>>>>>>>>> it is not good enough? >>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>> Since it is with many cond_resched(), the performance >>>>>>>>>>>>>>>>>> is worst... >>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>> Not exactly? It should have zero cost for a preemptible >>>>>>>>>>>>>>>>> kernel. >>>>>>>>>>>>>>>>> For a non-preemptible kernel, it helps avoid clearing >>>>>>>>>>>>>>>>> the folio >>>>>>>>>>>>>>>>> from occupying the CPU and starving other processes, >>>>>>>>>>>>>>>>> right? >>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>> --- a/mm/shmem.c >>>>>>>>>>>>>>>> +++ b/mm/shmem.c >>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>> @@ -2393,10 +2393,7 @@ static int >>>>>>>>>>>>>>>> shmem_get_folio_gfp(struct inode >>>>>>>>>>>>>>>> *inode, pgoff_t index, >>>>>>>>>>>>>>>> * it now, lest undo on failure cancel >>>>>>>>>>>>>>>> our earlier guarantee. >>>>>>>>>>>>>>>> */ >>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>> if (sgp != SGP_WRITE && ! >>>>>>>>>>>>>>>> folio_test_uptodate(folio)) { >>>>>>>>>>>>>>>> - long i, n = folio_nr_pages(folio); >>>>>>>>>>>>>>>> - >>>>>>>>>>>>>>>> - for (i = 0; i < n; i++) >>>>>>>>>>>>>>>> - clear_highpage(folio_page(folio, >>>>>>>>>>>>>>>> i)); >>>>>>>>>>>>>>>> + folio_zero_user(folio, vmf->address); >>>>>>>>>>>>>>>> flush_dcache_folio(folio); >>>>>>>>>>>>>>>> folio_mark_uptodate(folio); >>>>>>>>>>>>>>>> } >>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>> Do we perform better or worse with the following? >>>>>>>>>>>>>>> >>>>>>>>>>>>>>> Here is for SGP_FALLOC, vmf = NULL, we could use >>>>>>>>>>>>>>> folio_zero_user(folio, >>>>>>>>>>>>>>> 0), I think the performance is worse, will retest once I >>>>>>>>>>>>>>> can access >>>>>>>>>>>>>>> hardware. >>>>>>>>>>>>>> >>>>>>>>>>>>>> Perhaps, since the current code uses clear_hugepage(). >>>>>>>>>>>>>> Does using >>>>>>>>>>>>>> index << PAGE_SHIFT as the addr_hint offer any benefit? >>>>>>>>>>>>>> >>>>>>>>>>>>> >>>>>>>>>>>>> when use folio_zero_user(), the performance is vary bad >>>>>>>>>>>>> with above >>>>>>>>>>>>> fallocate test(mount huge=always), >>>>>>>>>>>>> >>>>>>>>>>>>> folio_zero_range clear_highpage >>>>>>>>>>>>> folio_zero_user >>>>>>>>>>>>> real 0m1.214s 0m1.111s 0m3.159s >>>>>>>>>>>>> user 0m0.000s 0m0.000s 0m0.000s >>>>>>>>>>>>> sys 0m1.210s 0m1.109s 0m3.152s >>>>>>>>>>>>> >>>>>>>>>>>>> I tried with addr_hint = 0/index << PAGE_SHIFT, no obvious >>>>>>>>>>>>> different. >>>>>>>>>>>> >>>>>>>>>>>> Interesting. Does your kernel have preemption disabled or >>>>>>>>>>>> preemption_debug enabled? >>>>>>>>>>> >>>>>>>>>>> ARM64 server, CONFIG_PREEMPT_NONE=y >>>>>>>>>> this explains why the performance is much worse. >>>>>>>>>> >>>>>>>>>>> >>>>>>>>>>>> >>>>>>>>>>>> If not, it makes me wonder whether folio_zero_user() in >>>>>>>>>>>> alloc_anon_folio() is actually improving performance as >>>>>>>>>>>> expected, >>>>>>>>>>>> compared to the simpler folio_zero() you plan to implement. :-) >>>>>>>>>>> >>>>>>>>>>> Yes, maybe, the folio_zero_user(was clear_huge_page) is from >>>>>>>>>>> 47ad8475c000 ("thp: clear_copy_huge_page"), so original >>>>>>>>>>> clear_huge_page >>>>>>>>>>> is used in HugeTLB, clear PUD size maybe spend many time, but >>>>>>>>>>> for PMD or >>>>>>>>>>> other size of large folio, cond_resched is not necessary >>>>>>>>>>> since we >>>>>>>>>>> already have some folio_zero_range() to clear large folio, >>>>>>>>>>> and no issue >>>>>>>>>>> was reported. >>>>>>>>>> probably worth an optimization. calling cond_resched() for >>>>>>>>>> each page >>>>>>>>>> seems too aggressive and useless. >>>>>>>>> >>>>>>>>> After some test, I think the cond_resched() is not the root cause, >>>>>>>>> no performance gained with batched cond_resched(), even I kill >>>>>>>>> cond_resched() from process_huge_page, no improvement. >>>>>>>>> >>>>>>>>> But when I unconditionally use clear_gigantic_page() in >>>>>>>>> folio_zero_user(patched), there is big improvement with above >>>>>>>>> fallocate on tmpfs(mount huge=always), also I test some other >>>>>>>>> testcase, >>>>>>>>> >>>>>>>>> >>>>>>>>> 1) case-anon-w-seq-mt: (2M PMD THP) >>>>>>>>> >>>>>>>>> base: >>>>>>>>> real 0m2.490s 0m2.254s 0m2.272s >>>>>>>>> user 1m59.980s 2m23.431s 2m18.739s >>>>>>>>> sys 1m3.675s 1m15.462s 1m15.030s >>>>>>>>> >>>>>>>>> patched: >>>>>>>>> real 0m2.234s 0m2.225s 0m2.159s >>>>>>>>> user 2m56.105s 2m57.117s 3m0.489s >>>>>>>>> sys 0m17.064s 0m17.564s 0m16.150s >>>>>>>>> >>>>>>>>> Patched kernel win on sys and bad in user, but real is almost >>>>>>>>> same, >>>>>>>>> maybe a little better than base. >>>>>>>> We can find user time difference. That means the original cache >>>>>>>> hot >>>>>>>> behavior still applies on your system. >>>>>>>> However, it appears that the performance to clear page from end to >>>>>>>> begin >>>>>>>> is really bad on your system. >>>>>>>> So, I suggest to revise the current implementation to use >>>>>>>> sequential >>>>>>>> clearing as much as possible. >>>>>>>> >>>>>>> >>>>>>> I test case-anon-cow-seq-hugetlb for copy_user_large_folio() >>>>>>> >>>>>>> base: >>>>>>> real 0m6.259s 0m6.197s 0m6.316s >>>>>>> user 1m31.176s 1m27.195s 1m29.594s >>>>>>> sys 7m44.199s 7m51.490s 8m21.149s >>>>>>> >>>>>>> patched(use copy_user_gigantic_page for 2M hugetlb too) >>>>>>> real 0m3.182s 0m3.002s 0m2.963s >>>>>>> user 1m19.456s 1m3.107s 1m6.447s >>>>>>> sys 2m59.222s 3m10.899s 3m1.027s >>>>>>> >>>>>>> and sequential copy is better than the current implementation, >>>>>>> so I will use sequential clear and copy. >>>>>> Sorry, it appears that you misunderstanding my suggestion. I >>>>>> suggest to >>>>>> revise process_huge_page() to use more sequential memory clearing and >>>>>> copying to improve its performance on your platform. >>>>>> -- >>>>>> Best Regards, >>>>>> Huang, Ying >>>>>> >>>>>>>>> 2) case-anon-w-seq-hugetlb:(2M PMD HugeTLB) >>>>>>>>> >>>>>>>>> base: >>>>>>>>> real 0m5.175s 0m5.117s 0m4.856s >>>>>>>>> user 5m15.943s 5m7.567s 4m29.273s >>>>>>>>> sys 2m38.503s 2m21.949s 2m21.252s >>>>>>>>> >>>>>>>>> patched: >>>>>>>>> real 0m4.966s 0m4.841s 0m4.561s >>>>>>>>> user 6m30.123s 6m9.516s 5m49.733s >>>>>>>>> sys 0m58.503s 0m47.847s 0m46.785s >>>>>>>>> >>>>>>>>> >>>>>>>>> This case is similar to the case1. >>>>>>>>> >>>>>>>>> 3) fallocate hugetlb 20G (2M PMD HugeTLB) >>>>>>>>> >>>>>>>>> base: >>>>>>>>> real 0m3.016s 0m3.019s 0m3.018s >>>>>>>>> user 0m0.000s 0m0.000s 0m0.000s >>>>>>>>> sys 0m3.009s 0m3.012s 0m3.010s >>>>>>>>> >>>>>>>>> patched: >>>>>>>>> >>>>>>>>> real 0m1.136s 0m1.136s 0m1.136s >>>>>>>>> user 0m0.000s 0m0.000s 0m0.004s >>>>>>>>> sys 0m1.133s 0m1.133s 0m1.129s >>>>>>>>> >>>>>>>>> >>>>>>>>> There is big win on patched kernel, and it is similar to above >>>>>>>>> tmpfs >>>>>>>>> test, so maybe we could revert the commit c79b57e462b5 ("mm: >>>>>>>>> hugetlb: >>>>>>>>> clear target sub-page last when clearing huge page"). >>>>> >>>>> I tried the following changes, >>>>> diff --git a/mm/memory.c b/mm/memory.c >>>>> index 66cf855dee3f..e5cc75adfa10 100644 >>>>> --- a/mm/memory.c >>>>> +++ b/mm/memory.c >>>>> @@ -6777,7 +6777,7 @@ static inline int process_huge_page( >>>>> base = 0; >>>>> l = n; >>>>> /* Process subpages at the end of huge page */ >>>>> - for (i = nr_pages - 1; i >= 2 * n; i--) { >>>>> + for (i = 2 * n; i < nr_pages; i++) { >>>>> cond_resched(); >>>>> ret = process_subpage(addr + i * >>>>> PAGE_SIZE, i, >>>>> arg); >>>>> if (ret) >>>>> >>>>> Since n = 0, so the copying is from start to end now, but not >>>>> improvement for case-anon-cow-seq-hugetlb, >>>>> >>>>> and if use copy_user_gigantic_pager, the time reduced from 6s to 3s >>>>> >>>>> diff --git a/mm/memory.c b/mm/memory.c >>>>> index fe21bd3beff5..2c6532d21d84 100644 >>>>> --- a/mm/memory.c >>>>> +++ b/mm/memory.c >>>>> @@ -6876,10 +6876,7 @@ int copy_user_large_folio(struct folio *dst, >>>>> struct folio *src, >>>>> .vma = vma, >>>>> }; >>>>> >>>>> - if (unlikely(nr_pages > MAX_ORDER_NR_PAGES)) >>>>> - return copy_user_gigantic_page(dst, src, addr_hint, >>>>> vma, nr_pages); >>>>> - >>>>> - return process_huge_page(addr_hint, nr_pages, copy_subpage, >>>>> &arg); >>>>> + return copy_user_gigantic_page(dst, src, addr_hint, vma, >>>>> nr_pages); >>>>> } >>>> It appears that we have code generation issue here. Can you check >>>> it? >>>> Whether code is inlined in the same way? >>>> >>> >>> No different, and I checked the asm, both process_huge_page and >>> copy_user_gigantic_page are inlined, it is strange... >> >> It's not inlined in my configuration. And __always_inline below changes >> it for me. >> >> If it's already inlined and the code is actually almost same, why >> there's difference? Is it possible for you to do some profile or >> further analysis? > > Yes, will continue to debug this. My bad, I has some refactor patch before using copy_user_large_folio(), ba3fda2a7b08 mm: use copy_user_large_page // good performance a88666ae8f4d mm: call might_sleep() in folio_zero/copy_user() 3ab7d4d405e9 mm: calculate the base address in the folio_zero/copy_user() 7b240664c07d mm: convert to folio_copy_user() // I made a mistake which use dst instead of src in copy_user_gigantic_page() 1a951e310aa9 mm: use aligned address in copy_user_gigantic_page() e095ce052607 mm: use aligned address in clear_gigantic_page() so please ignore the copy test result (case-anon-cow-seq-hugetlb) In summary: 1) for copying, no obvious different between copy_user_large_folio/process_huge_page(copying from last to start or copying from start to last) 2) for clearing, clear_gigantic_page is better than process_huge_page from my machine, and after clearing page from start to last(current, it process page from last to first), the performance is same to the clear_gigantic_page. > >> >>>> Maybe we can start with >>>> modified mm/memory.c >>>> @@ -6714,7 +6714,7 @@ EXPORT_SYMBOL(__might_fault); >>>> * operation. The target subpage will be processed last to keep its >>>> * cache lines hot. >>>> */ >>>> -static inline int process_huge_page( >>>> +static __always_inline int process_huge_page( >>>> unsigned long addr_hint, unsigned int nr_pages, >>>> int (*process_subpage)(unsigned long addr, int idx, void *arg), >>>> void *arg) >> >> -- >> Best Regards, >> Huang, Ying >> > >
Hi, Kefeng, Sorry for late reply, the email is buried in my inbox, just dig it out. Kefeng Wang <wangkefeng.wang@huawei.com> writes: > On 2024/10/28 14:37, Kefeng Wang wrote: >> On 2024/10/28 10:39, Huang, Ying wrote: >>> Kefeng Wang <wangkefeng.wang@huawei.com> writes: >>> >>>> On 2024/10/25 20:21, Huang, Ying wrote: >>>>> Kefeng Wang <wangkefeng.wang@huawei.com> writes: >>>>> >>>>>> On 2024/10/25 15:47, Huang, Ying wrote: >>>>>>> Kefeng Wang <wangkefeng.wang@huawei.com> writes: >>>>>>> >>>>>>>> On 2024/10/25 10:59, Huang, Ying wrote: >>>>>>>>> Hi, Kefeng, >>>>>>>>> Kefeng Wang <wangkefeng.wang@huawei.com> writes: >>>>>>>>> >>>>>>>>>> +CC Huang Ying, >>>>>>>>>> >>>>>>>>>> On 2024/10/23 6:56, Barry Song wrote: >>>>>>>>>>> On Wed, Oct 23, 2024 at 4:10 AM Kefeng Wang >>>>>>>>>>> <wangkefeng.wang@huawei.com> wrote: >>>>>>>>>>>> >>>>>>>>>>>> >>>>>>>>>> ... >>>>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>>>> On 2024/10/17 23:09, Matthew Wilcox wrote: >>>>>>>>>>>>>>>>>>>>>>>>>>> On Thu, Oct 17, 2024 at 10:25:04PM +0800, >>>>>>>>>>>>>>>>>>>>>>>>>>> Kefeng Wang wrote: >>>>>>>>>>>>>>>>>>>>>>>>>>>> Directly use folio_zero_range() to cleanup code. >>>>>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>>>>> Are you sure there's no performance >>>>>>>>>>>>>>>>>>>>>>>>>>> regression introduced by this? >>>>>>>>>>>>>>>>>>>>>>>>>>> clear_highpage() is often optimised in ways >>>>>>>>>>>>>>>>>>>>>>>>>>> that we can't optimise for >>>>>>>>>>>>>>>>>>>>>>>>>>> a plain memset(). On the other hand, if >>>>>>>>>>>>>>>>>>>>>>>>>>> the folio is large, maybe a >>>>>>>>>>>>>>>>>>>>>>>>>>> modern CPU will be able to do better than >>>>>>>>>>>>>>>>>>>>>>>>>>> clear-one-page-at-a-time. >>>>>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>>>> Right, I missing this, clear_page might be >>>>>>>>>>>>>>>>>>>>>>>>>> better than memset, I change >>>>>>>>>>>>>>>>>>>>>>>>>> this one when look at the shmem_writepage(), >>>>>>>>>>>>>>>>>>>>>>>>>> which already convert to >>>>>>>>>>>>>>>>>>>>>>>>>> use folio_zero_range() from >>>>>>>>>>>>>>>>>>>>>>>>>> clear_highpage(), also I grep >>>>>>>>>>>>>>>>>>>>>>>>>> folio_zero_range(), there are some other to >>>>>>>>>>>>>>>>>>>>>>>>>> use folio_zero_range(). >>>>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>>>> fs/bcachefs/fs-io-buffered.c: >>>>>>>>>>>>>>>>>>>>>>>>>> folio_zero_range(folio, 0, >>>>>>>>>>>>>>>>>>>>>>>>>> folio_size(folio)); >>>>>>>>>>>>>>>>>>>>>>>>>> fs/bcachefs/fs-io-buffered.c: >>>>>>>>>>>>>>>>>>>>>>>>>> folio_zero_range(f, >>>>>>>>>>>>>>>>>>>>>>>>>> 0, folio_size(f)); >>>>>>>>>>>>>>>>>>>>>>>>>> fs/bcachefs/fs-io-buffered.c: >>>>>>>>>>>>>>>>>>>>>>>>>> folio_zero_range(f, >>>>>>>>>>>>>>>>>>>>>>>>>> 0, folio_size(f)); >>>>>>>>>>>>>>>>>>>>>>>>>> fs/libfs.c: folio_zero_range(folio, 0, >>>>>>>>>>>>>>>>>>>>>>>>>> folio_size(folio)); >>>>>>>>>>>>>>>>>>>>>>>>>> fs/ntfs3/frecord.c: >>>>>>>>>>>>>>>>>>>>>>>>>> folio_zero_range(folio, 0, >>>>>>>>>>>>>>>>>>>>>>>>>> folio_size(folio)); >>>>>>>>>>>>>>>>>>>>>>>>>> mm/page_io.c: folio_zero_range(folio, 0, >>>>>>>>>>>>>>>>>>>>>>>>>> folio_size(folio)); >>>>>>>>>>>>>>>>>>>>>>>>>> mm/shmem.c: >>>>>>>>>>>>>>>>>>>>>>>>>> folio_zero_range(folio, 0, >>>>>>>>>>>>>>>>>>>>>>>>>> folio_size(folio)); >>>>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>>>>> IOW, what performance testing have you done >>>>>>>>>>>>>>>>>>>>>>>>>>> with this patch? >>>>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>>>> No performance test before, but I write a >>>>>>>>>>>>>>>>>>>>>>>>>> testcase, >>>>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>>>> 1) allocate N large folios >>>>>>>>>>>>>>>>>>>>>>>>>> (folio_alloc(PMD_ORDER)) >>>>>>>>>>>>>>>>>>>>>>>>>> 2) then calculate the diff(us) when clear >>>>>>>>>>>>>>>>>>>>>>>>>> all N folios >>>>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>>>> clear_highpage/folio_zero_range/ >>>>>>>>>>>>>>>>>>>>>>>>>> folio_zero_user >>>>>>>>>>>>>>>>>>>>>>>>>> 3) release N folios >>>>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>>>> the result(run 5 times) shown below on my machine, >>>>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>>>> N=1, >>>>>>>>>>>>>>>>>>>>>>>>>> clear_highpage >>>>>>>>>>>>>>>>>>>>>>>>>> folio_zero_range folio_zero_user >>>>>>>>>>>>>>>>>>>>>>>>>> 1 69 74 >>>>>>>>>>>>>>>>>>>>>>>>>> 177 >>>>>>>>>>>>>>>>>>>>>>>>>> 2 57 62 >>>>>>>>>>>>>>>>>>>>>>>>>> 168 >>>>>>>>>>>>>>>>>>>>>>>>>> 3 54 58 >>>>>>>>>>>>>>>>>>>>>>>>>> 234 >>>>>>>>>>>>>>>>>>>>>>>>>> 4 54 58 >>>>>>>>>>>>>>>>>>>>>>>>>> 157 >>>>>>>>>>>>>>>>>>>>>>>>>> 5 56 62 >>>>>>>>>>>>>>>>>>>>>>>>>> 148 >>>>>>>>>>>>>>>>>>>>>>>>>> avg 58 62.8 >>>>>>>>>>>>>>>>>>>>>>>>>> 176.8 >>>>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>>>> N=100 >>>>>>>>>>>>>>>>>>>>>>>>>> clear_highpage >>>>>>>>>>>>>>>>>>>>>>>>>> folio_zero_range folio_zero_user >>>>>>>>>>>>>>>>>>>>>>>>>> 1 11015 >>>>>>>>>>>>>>>>>>>>>>>>>> 11309 32833 >>>>>>>>>>>>>>>>>>>>>>>>>> 2 10385 >>>>>>>>>>>>>>>>>>>>>>>>>> 11110 49751 >>>>>>>>>>>>>>>>>>>>>>>>>> 3 10369 >>>>>>>>>>>>>>>>>>>>>>>>>> 11056 33095 >>>>>>>>>>>>>>>>>>>>>>>>>> 4 10332 >>>>>>>>>>>>>>>>>>>>>>>>>> 11017 33106 >>>>>>>>>>>>>>>>>>>>>>>>>> 5 10483 >>>>>>>>>>>>>>>>>>>>>>>>>> 11000 49032 >>>>>>>>>>>>>>>>>>>>>>>>>> avg 10516.8 11098.4 >>>>>>>>>>>>>>>>>>>>>>>>>> 39563.4 >>>>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>>>> N=512 >>>>>>>>>>>>>>>>>>>>>>>>>> clear_highpage >>>>>>>>>>>>>>>>>>>>>>>>>> folio_zero_range folio_zero_user >>>>>>>>>>>>>>>>>>>>>>>>>> 1 55560 >>>>>>>>>>>>>>>>>>>>>>>>>> 60055 156876 >>>>>>>>>>>>>>>>>>>>>>>>>> 2 55485 >>>>>>>>>>>>>>>>>>>>>>>>>> 60024 157132 >>>>>>>>>>>>>>>>>>>>>>>>>> 3 55474 >>>>>>>>>>>>>>>>>>>>>>>>>> 60129 156658 >>>>>>>>>>>>>>>>>>>>>>>>>> 4 55555 >>>>>>>>>>>>>>>>>>>>>>>>>> 59867 157259 >>>>>>>>>>>>>>>>>>>>>>>>>> 5 55528 >>>>>>>>>>>>>>>>>>>>>>>>>> 59932 157108 >>>>>>>>>>>>>>>>>>>>>>>>>> avg 55520.4 60001.4 >>>>>>>>>>>>>>>>>>>>>>>>>> 157006.6 >>>>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>>>> folio_zero_user with many cond_resched(), so >>>>>>>>>>>>>>>>>>>>>>>>>> time fluctuates a lot, >>>>>>>>>>>>>>>>>>>>>>>>>> clear_highpage is better folio_zero_range as >>>>>>>>>>>>>>>>>>>>>>>>>> you said. >>>>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>>>> Maybe add a new helper to convert all >>>>>>>>>>>>>>>>>>>>>>>>>> folio_zero_range(folio, 0, >>>>>>>>>>>>>>>>>>>>>>>>>> folio_size(folio)) >>>>>>>>>>>>>>>>>>>>>>>>>> to use clear_highpage + flush_dcache_folio? >>>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>>>>> If this also improves performance for other >>>>>>>>>>>>>>>>>>>>>>>>> existing callers of >>>>>>>>>>>>>>>>>>>>>>>>> folio_zero_range(), then that's a positive outcome. >>>>>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>> ... >>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>>> hi Kefeng, >>>>>>>>>>>>>>>>>>>>>> what's your point? providing a helper like >>>>>>>>>>>>>>>>>>>>>> clear_highfolio() or similar? >>>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>>> Yes, from above test, using clear_highpage/ >>>>>>>>>>>>>>>>>>>>> flush_dcache_folio is better >>>>>>>>>>>>>>>>>>>>> than using folio_zero_range() for folio >>>>>>>>>>>>>>>>>>>>> zero(especially for large >>>>>>>>>>>>>>>>>>>>> folio), so I'd like to add a new helper, maybe >>>>>>>>>>>>>>>>>>>>> name it folio_zero() >>>>>>>>>>>>>>>>>>>>> since it zero the whole folio. >>>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>>> we already have a helper like folio_zero_user()? >>>>>>>>>>>>>>>>>>>> it is not good enough? >>>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>>> Since it is with many cond_resched(), the >>>>>>>>>>>>>>>>>>> performance is worst... >>>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>>> Not exactly? It should have zero cost for a >>>>>>>>>>>>>>>>>> preemptible kernel. >>>>>>>>>>>>>>>>>> For a non-preemptible kernel, it helps avoid >>>>>>>>>>>>>>>>>> clearing the folio >>>>>>>>>>>>>>>>>> from occupying the CPU and starving other processes, >>>>>>>>>>>>>>>>>> right? >>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>> --- a/mm/shmem.c >>>>>>>>>>>>>>>>> +++ b/mm/shmem.c >>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>> @@ -2393,10 +2393,7 @@ static int >>>>>>>>>>>>>>>>> shmem_get_folio_gfp(struct inode >>>>>>>>>>>>>>>>> *inode, pgoff_t index, >>>>>>>>>>>>>>>>> * it now, lest undo on failure >>>>>>>>>>>>>>>>> cancel our earlier guarantee. >>>>>>>>>>>>>>>>> */ >>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>> if (sgp != SGP_WRITE && ! >>>>>>>>>>>>>>>>> folio_test_uptodate(folio)) { >>>>>>>>>>>>>>>>> - long i, n = folio_nr_pages(folio); >>>>>>>>>>>>>>>>> - >>>>>>>>>>>>>>>>> - for (i = 0; i < n; i++) >>>>>>>>>>>>>>>>> - >>>>>>>>>>>>>>>>> clear_highpage(folio_page(folio, i)); >>>>>>>>>>>>>>>>> + folio_zero_user(folio, vmf->address); >>>>>>>>>>>>>>>>> flush_dcache_folio(folio); >>>>>>>>>>>>>>>>> folio_mark_uptodate(folio); >>>>>>>>>>>>>>>>> } >>>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>>> Do we perform better or worse with the following? >>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>> Here is for SGP_FALLOC, vmf = NULL, we could use >>>>>>>>>>>>>>>> folio_zero_user(folio, >>>>>>>>>>>>>>>> 0), I think the performance is worse, will retest once >>>>>>>>>>>>>>>> I can access >>>>>>>>>>>>>>>> hardware. >>>>>>>>>>>>>>> >>>>>>>>>>>>>>> Perhaps, since the current code uses >>>>>>>>>>>>>>> clear_hugepage(). Does using >>>>>>>>>>>>>>> index << PAGE_SHIFT as the addr_hint offer any benefit? >>>>>>>>>>>>>>> >>>>>>>>>>>>>> >>>>>>>>>>>>>> when use folio_zero_user(), the performance is vary bad >>>>>>>>>>>>>> with above >>>>>>>>>>>>>> fallocate test(mount huge=always), >>>>>>>>>>>>>> >>>>>>>>>>>>>> folio_zero_range clear_highpage >>>>>>>>>>>>>> folio_zero_user >>>>>>>>>>>>>> real 0m1.214s 0m1.111s 0m3.159s >>>>>>>>>>>>>> user 0m0.000s 0m0.000s 0m0.000s >>>>>>>>>>>>>> sys 0m1.210s 0m1.109s 0m3.152s >>>>>>>>>>>>>> >>>>>>>>>>>>>> I tried with addr_hint = 0/index << PAGE_SHIFT, no >>>>>>>>>>>>>> obvious different. >>>>>>>>>>>>> >>>>>>>>>>>>> Interesting. Does your kernel have preemption disabled or >>>>>>>>>>>>> preemption_debug enabled? >>>>>>>>>>>> >>>>>>>>>>>> ARM64 server, CONFIG_PREEMPT_NONE=y >>>>>>>>>>> this explains why the performance is much worse. >>>>>>>>>>> >>>>>>>>>>>> >>>>>>>>>>>>> >>>>>>>>>>>>> If not, it makes me wonder whether folio_zero_user() in >>>>>>>>>>>>> alloc_anon_folio() is actually improving performance as >>>>>>>>>>>>> expected, >>>>>>>>>>>>> compared to the simpler folio_zero() you plan to implement. :-) >>>>>>>>>>>> >>>>>>>>>>>> Yes, maybe, the folio_zero_user(was clear_huge_page) is from >>>>>>>>>>>> 47ad8475c000 ("thp: clear_copy_huge_page"), so original >>>>>>>>>>>> clear_huge_page >>>>>>>>>>>> is used in HugeTLB, clear PUD size maybe spend many time, >>>>>>>>>>>> but for PMD or >>>>>>>>>>>> other size of large folio, cond_resched is not necessary >>>>>>>>>>>> since we >>>>>>>>>>>> already have some folio_zero_range() to clear large folio, >>>>>>>>>>>> and no issue >>>>>>>>>>>> was reported. >>>>>>>>>>> probably worth an optimization. calling cond_resched() for >>>>>>>>>>> each page >>>>>>>>>>> seems too aggressive and useless. >>>>>>>>>> >>>>>>>>>> After some test, I think the cond_resched() is not the root cause, >>>>>>>>>> no performance gained with batched cond_resched(), even I kill >>>>>>>>>> cond_resched() from process_huge_page, no improvement. >>>>>>>>>> >>>>>>>>>> But when I unconditionally use clear_gigantic_page() in >>>>>>>>>> folio_zero_user(patched), there is big improvement with above >>>>>>>>>> fallocate on tmpfs(mount huge=always), also I test some >>>>>>>>>> other testcase, >>>>>>>>>> >>>>>>>>>> >>>>>>>>>> 1) case-anon-w-seq-mt: (2M PMD THP) >>>>>>>>>> >>>>>>>>>> base: >>>>>>>>>> real 0m2.490s 0m2.254s 0m2.272s >>>>>>>>>> user 1m59.980s 2m23.431s 2m18.739s >>>>>>>>>> sys 1m3.675s 1m15.462s 1m15.030s >>>>>>>>>> >>>>>>>>>> patched: >>>>>>>>>> real 0m2.234s 0m2.225s 0m2.159s >>>>>>>>>> user 2m56.105s 2m57.117s 3m0.489s >>>>>>>>>> sys 0m17.064s 0m17.564s 0m16.150s >>>>>>>>>> >>>>>>>>>> Patched kernel win on sys and bad in user, but real is >>>>>>>>>> almost same, >>>>>>>>>> maybe a little better than base. >>>>>>>>> We can find user time difference. That means the original >>>>>>>>> cache hot >>>>>>>>> behavior still applies on your system. >>>>>>>>> However, it appears that the performance to clear page from end to >>>>>>>>> begin >>>>>>>>> is really bad on your system. >>>>>>>>> So, I suggest to revise the current implementation to use >>>>>>>>> sequential >>>>>>>>> clearing as much as possible. >>>>>>>>> >>>>>>>> >>>>>>>> I test case-anon-cow-seq-hugetlb for copy_user_large_folio() >>>>>>>> >>>>>>>> base: >>>>>>>> real 0m6.259s 0m6.197s 0m6.316s >>>>>>>> user 1m31.176s 1m27.195s 1m29.594s >>>>>>>> sys 7m44.199s 7m51.490s 8m21.149s >>>>>>>> >>>>>>>> patched(use copy_user_gigantic_page for 2M hugetlb too) >>>>>>>> real 0m3.182s 0m3.002s 0m2.963s >>>>>>>> user 1m19.456s 1m3.107s 1m6.447s >>>>>>>> sys 2m59.222s 3m10.899s 3m1.027s >>>>>>>> >>>>>>>> and sequential copy is better than the current implementation, >>>>>>>> so I will use sequential clear and copy. >>>>>>> Sorry, it appears that you misunderstanding my suggestion. I >>>>>>> suggest to >>>>>>> revise process_huge_page() to use more sequential memory clearing and >>>>>>> copying to improve its performance on your platform. >>>>>>> -- Best Regards, >>>>>>> Huang, Ying >>>>>>> >>>>>>>>>> 2) case-anon-w-seq-hugetlb:(2M PMD HugeTLB) >>>>>>>>>> >>>>>>>>>> base: >>>>>>>>>> real 0m5.175s 0m5.117s 0m4.856s >>>>>>>>>> user 5m15.943s 5m7.567s 4m29.273s >>>>>>>>>> sys 2m38.503s 2m21.949s 2m21.252s >>>>>>>>>> >>>>>>>>>> patched: >>>>>>>>>> real 0m4.966s 0m4.841s 0m4.561s >>>>>>>>>> user 6m30.123s 6m9.516s 5m49.733s >>>>>>>>>> sys 0m58.503s 0m47.847s 0m46.785s >>>>>>>>>> >>>>>>>>>> >>>>>>>>>> This case is similar to the case1. >>>>>>>>>> >>>>>>>>>> 3) fallocate hugetlb 20G (2M PMD HugeTLB) >>>>>>>>>> >>>>>>>>>> base: >>>>>>>>>> real 0m3.016s 0m3.019s 0m3.018s >>>>>>>>>> user 0m0.000s 0m0.000s 0m0.000s >>>>>>>>>> sys 0m3.009s 0m3.012s 0m3.010s >>>>>>>>>> >>>>>>>>>> patched: >>>>>>>>>> >>>>>>>>>> real 0m1.136s 0m1.136s 0m1.136s >>>>>>>>>> user 0m0.000s 0m0.000s 0m0.004s >>>>>>>>>> sys 0m1.133s 0m1.133s 0m1.129s >>>>>>>>>> >>>>>>>>>> >>>>>>>>>> There is big win on patched kernel, and it is similar to >>>>>>>>>> above tmpfs >>>>>>>>>> test, so maybe we could revert the commit c79b57e462b5 ("mm: >>>>>>>>>> hugetlb: >>>>>>>>>> clear target sub-page last when clearing huge page"). >>>>>> >>>>>> I tried the following changes, >>>>>> diff --git a/mm/memory.c b/mm/memory.c >>>>>> index 66cf855dee3f..e5cc75adfa10 100644 >>>>>> --- a/mm/memory.c >>>>>> +++ b/mm/memory.c >>>>>> @@ -6777,7 +6777,7 @@ static inline int process_huge_page( >>>>>> base = 0; >>>>>> l = n; >>>>>> /* Process subpages at the end of huge page */ >>>>>> - for (i = nr_pages - 1; i >= 2 * n; i--) { >>>>>> + for (i = 2 * n; i < nr_pages; i++) { >>>>>> cond_resched(); >>>>>> ret = process_subpage(addr + i * >>>>>> PAGE_SIZE, i, >>>>>> arg); >>>>>> if (ret) >>>>>> >>>>>> Since n = 0, so the copying is from start to end now, but not >>>>>> improvement for case-anon-cow-seq-hugetlb, >>>>>> >>>>>> and if use copy_user_gigantic_pager, the time reduced from 6s to 3s >>>>>> >>>>>> diff --git a/mm/memory.c b/mm/memory.c >>>>>> index fe21bd3beff5..2c6532d21d84 100644 >>>>>> --- a/mm/memory.c >>>>>> +++ b/mm/memory.c >>>>>> @@ -6876,10 +6876,7 @@ int copy_user_large_folio(struct folio *dst, >>>>>> struct folio *src, >>>>>> .vma = vma, >>>>>> }; >>>>>> >>>>>> - if (unlikely(nr_pages > MAX_ORDER_NR_PAGES)) >>>>>> - return copy_user_gigantic_page(dst, src, addr_hint, >>>>>> vma, nr_pages); >>>>>> - >>>>>> - return process_huge_page(addr_hint, nr_pages, >>>>>> copy_subpage, &arg); >>>>>> + return copy_user_gigantic_page(dst, src, addr_hint, vma, >>>>>> nr_pages); >>>>>> } >>>>> It appears that we have code generation issue here. Can you check >>>>> it? >>>>> Whether code is inlined in the same way? >>>>> >>>> >>>> No different, and I checked the asm, both process_huge_page and >>>> copy_user_gigantic_page are inlined, it is strange... >>> >>> It's not inlined in my configuration. And __always_inline below changes >>> it for me. >>> >>> If it's already inlined and the code is actually almost same, why >>> there's difference? Is it possible for you to do some profile or >>> further analysis? >> Yes, will continue to debug this. > > My bad, I has some refactor patch before using copy_user_large_folio(), > > ba3fda2a7b08 mm: use copy_user_large_page // good performance > a88666ae8f4d mm: call might_sleep() in folio_zero/copy_user() > 3ab7d4d405e9 mm: calculate the base address in the folio_zero/copy_user() > 7b240664c07d mm: convert to folio_copy_user() // I made a mistake > which use dst instead of src in copy_user_gigantic_page() > 1a951e310aa9 mm: use aligned address in copy_user_gigantic_page() > e095ce052607 mm: use aligned address in clear_gigantic_page() > > so please ignore the copy test result (case-anon-cow-seq-hugetlb) > > In summary: > 1) for copying, no obvious different between > copy_user_large_folio/process_huge_page(copying from last to start or > copying from start to last) > > 2) for clearing, clear_gigantic_page is better than process_huge_page > from my machine, and after clearing page from start to last(current, > it process page from last to first), the performance is same to the > clear_gigantic_page. Can you show detailed data, at least user/sys/real time? Previously, we can find user time reduction and sys time increment. -- Best Regards, Huang, Ying >> >>> >>>>> Maybe we can start with >>>>> modified mm/memory.c >>>>> @@ -6714,7 +6714,7 @@ EXPORT_SYMBOL(__might_fault); >>>>> * operation. The target subpage will be processed last to keep its >>>>> * cache lines hot. >>>>> */ >>>>> -static inline int process_huge_page( >>>>> +static __always_inline int process_huge_page( >>>>> unsigned long addr_hint, unsigned int nr_pages, >>>>> int (*process_subpage)(unsigned long addr, int idx, void *arg), >>>>> void *arg) >>> >>> -- Best Regards, >>> Huang, Ying >>> >>
diff --git a/mm/shmem.c b/mm/shmem.c index bd5ba016567d..247c0403af83 100644 --- a/mm/shmem.c +++ b/mm/shmem.c @@ -2392,11 +2392,7 @@ static int shmem_get_folio_gfp(struct inode *inode, pgoff_t index, * it now, lest undo on failure cancel our earlier guarantee. */ if (sgp != SGP_WRITE && !folio_test_uptodate(folio)) { - long i, n = folio_nr_pages(folio); - - for (i = 0; i < n; i++) - clear_highpage(folio_page(folio, i)); - flush_dcache_folio(folio); + folio_zero_range(folio, 0, folio_size(folio)); folio_mark_uptodate(folio); }
Directly use folio_zero_range() to cleanup code. Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com> --- mm/shmem.c | 6 +----- 1 file changed, 1 insertion(+), 5 deletions(-)