| Message ID | 1584124476-76534-1-git-send-email-yang.shi@linux.alibaba.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | [1/2] mm: swap: make page_evictable() inline | expand |
On Fri, Mar 13, 2020 at 11:34 AM Yang Shi <yang.shi@linux.alibaba.com> wrote: > > When backporting commit 9c4e6b1a7027 ("mm, mlock, vmscan: no more > skipping pagevecs") to our 4.9 kernel, our test bench noticed around 10% > down with a couple of vm-scalability's test cases (lru-file-readonce, > lru-file-readtwice and lru-file-mmap-read). I didn't see that much down > on my VM (32c-64g-2nodes). It might be caused by the test configuration, > which is 32c-256g with NUMA disabled and the tests were run in root memcg, > so the tests actually stress only one inactive and active lru. It > sounds not very usual in mordern production environment. > > That commit did two major changes: > 1. Call page_evictable() > 2. Use smp_mb to force the PG_lru set visible > > It looks they contribute the most overhead. The page_evictable() is a > function which does function prologue and epilogue, and that was used by > page reclaim path only. However, lru add is a very hot path, so it > sounds better to make it inline. However, it calls page_mapping() which > is not inlined either, but the disassemble shows it doesn't do push and > pop operations and it sounds not very straightforward to inline it. > > Other than this, it sounds smp_mb() is not necessary for x86 since > SetPageLRU is atomic which enforces memory barrier already, replace it > with smp_mb__after_atomic() in the following patch. > > With the two fixes applied, the tests can get back around 5% on that > test bench and get back normal on my VM. Since the test bench > configuration is not that usual and I also saw around 6% up on the > latest upstream, so it sounds good enough IMHO. > > The below is test data (lru-file-readtwice throughput) against the v5.6-rc4: > mainline w/ inline fix > 150MB 154MB > What is the test setup for the above experiment? I would like to get a repro. > With this patch the throughput gets 2.67% up. The data with using > smp_mb__after_atomic() is showed in the following patch. > > Fixes: 9c4e6b1a7027 ("mm, mlock, vmscan: no more skipping pagevecs") > Cc: Shakeel Butt <shakeelb@google.com> > Cc: Vlastimil Babka <vbabka@suse.cz> > Signed-off-by: Yang Shi <yang.shi@linux.alibaba.com> > --- > include/linux/swap.h | 24 +++++++++++++++++++++++- > mm/vmscan.c | 23 ----------------------- > 2 files changed, 23 insertions(+), 24 deletions(-) > > diff --git a/include/linux/swap.h b/include/linux/swap.h > index 1e99f7a..297eb66 100644 > --- a/include/linux/swap.h > +++ b/include/linux/swap.h > @@ -374,7 +374,29 @@ extern unsigned long mem_cgroup_shrink_node(struct mem_cgroup *mem, > #define node_reclaim_mode 0 > #endif > > -extern int page_evictable(struct page *page); > +/* > + * page_evictable - test whether a page is evictable > + * @page: the page to test > + * > + * Test whether page is evictable--i.e., should be placed on active/inactive > + * lists vs unevictable list. > + * > + * Reasons page might not be evictable: > + * (1) page's mapping marked unevictable > + * (2) page is part of an mlocked VMA > + * > + */ > +static inline int page_evictable(struct page *page) > +{ > + int ret; > + > + /* Prevent address_space of inode and swap cache from being freed */ > + rcu_read_lock(); > + ret = !mapping_unevictable(page_mapping(page)) && !PageMlocked(page); > + rcu_read_unlock(); > + return ret; > +} > + > extern void check_move_unevictable_pages(struct pagevec *pvec); > > extern int kswapd_run(int nid); > diff --git a/mm/vmscan.c b/mm/vmscan.c > index 8763705..855c395 100644 > --- a/mm/vmscan.c > +++ b/mm/vmscan.c > @@ -4277,29 +4277,6 @@ int node_reclaim(struct pglist_data *pgdat, gfp_t gfp_mask, unsigned int order) > } > #endif > > -/* > - * page_evictable - test whether a page is evictable > - * @page: the page to test > - * > - * Test whether page is evictable--i.e., should be placed on active/inactive > - * lists vs unevictable list. > - * > - * Reasons page might not be evictable: > - * (1) page's mapping marked unevictable > - * (2) page is part of an mlocked VMA > - * > - */ > -int page_evictable(struct page *page) > -{ > - int ret; > - > - /* Prevent address_space of inode and swap cache from being freed */ > - rcu_read_lock(); > - ret = !mapping_unevictable(page_mapping(page)) && !PageMlocked(page); > - rcu_read_unlock(); > - return ret; > -} > - > /** > * check_move_unevictable_pages - check pages for evictability and move to > * appropriate zone lru list > -- > 1.8.3.1 >
On 3/13/20 12:33 PM, Shakeel Butt wrote: > On Fri, Mar 13, 2020 at 11:34 AM Yang Shi <yang.shi@linux.alibaba.com> wrote: >> When backporting commit 9c4e6b1a7027 ("mm, mlock, vmscan: no more >> skipping pagevecs") to our 4.9 kernel, our test bench noticed around 10% >> down with a couple of vm-scalability's test cases (lru-file-readonce, >> lru-file-readtwice and lru-file-mmap-read). I didn't see that much down >> on my VM (32c-64g-2nodes). It might be caused by the test configuration, >> which is 32c-256g with NUMA disabled and the tests were run in root memcg, >> so the tests actually stress only one inactive and active lru. It >> sounds not very usual in mordern production environment. >> >> That commit did two major changes: >> 1. Call page_evictable() >> 2. Use smp_mb to force the PG_lru set visible >> >> It looks they contribute the most overhead. The page_evictable() is a >> function which does function prologue and epilogue, and that was used by >> page reclaim path only. However, lru add is a very hot path, so it >> sounds better to make it inline. However, it calls page_mapping() which >> is not inlined either, but the disassemble shows it doesn't do push and >> pop operations and it sounds not very straightforward to inline it. >> >> Other than this, it sounds smp_mb() is not necessary for x86 since >> SetPageLRU is atomic which enforces memory barrier already, replace it >> with smp_mb__after_atomic() in the following patch. >> >> With the two fixes applied, the tests can get back around 5% on that >> test bench and get back normal on my VM. Since the test bench >> configuration is not that usual and I also saw around 6% up on the >> latest upstream, so it sounds good enough IMHO. >> >> The below is test data (lru-file-readtwice throughput) against the v5.6-rc4: >> mainline w/ inline fix >> 150MB 154MB >> > What is the test setup for the above experiment? I would like to get a repro. Just startup a VM with two nodes, then run case-lru-file-readtwice or case-lru-file-readonce in vm-scalability in root memcg or with memcg disabled. Then get the average throughput (dd result) from the test. Our test bench uses the script from lkp, but I just ran it manually. Single node VM should be more obvious showed in my test. > >> With this patch the throughput gets 2.67% up. The data with using >> smp_mb__after_atomic() is showed in the following patch. >> >> Fixes: 9c4e6b1a7027 ("mm, mlock, vmscan: no more skipping pagevecs") >> Cc: Shakeel Butt <shakeelb@google.com> >> Cc: Vlastimil Babka <vbabka@suse.cz> >> Signed-off-by: Yang Shi <yang.shi@linux.alibaba.com> >> --- >> include/linux/swap.h | 24 +++++++++++++++++++++++- >> mm/vmscan.c | 23 ----------------------- >> 2 files changed, 23 insertions(+), 24 deletions(-) >> >> diff --git a/include/linux/swap.h b/include/linux/swap.h >> index 1e99f7a..297eb66 100644 >> --- a/include/linux/swap.h >> +++ b/include/linux/swap.h >> @@ -374,7 +374,29 @@ extern unsigned long mem_cgroup_shrink_node(struct mem_cgroup *mem, >> #define node_reclaim_mode 0 >> #endif >> >> -extern int page_evictable(struct page *page); >> +/* >> + * page_evictable - test whether a page is evictable >> + * @page: the page to test >> + * >> + * Test whether page is evictable--i.e., should be placed on active/inactive >> + * lists vs unevictable list. >> + * >> + * Reasons page might not be evictable: >> + * (1) page's mapping marked unevictable >> + * (2) page is part of an mlocked VMA >> + * >> + */ >> +static inline int page_evictable(struct page *page) >> +{ >> + int ret; >> + >> + /* Prevent address_space of inode and swap cache from being freed */ >> + rcu_read_lock(); >> + ret = !mapping_unevictable(page_mapping(page)) && !PageMlocked(page); >> + rcu_read_unlock(); >> + return ret; >> +} >> + >> extern void check_move_unevictable_pages(struct pagevec *pvec); >> >> extern int kswapd_run(int nid); >> diff --git a/mm/vmscan.c b/mm/vmscan.c >> index 8763705..855c395 100644 >> --- a/mm/vmscan.c >> +++ b/mm/vmscan.c >> @@ -4277,29 +4277,6 @@ int node_reclaim(struct pglist_data *pgdat, gfp_t gfp_mask, unsigned int order) >> } >> #endif >> >> -/* >> - * page_evictable - test whether a page is evictable >> - * @page: the page to test >> - * >> - * Test whether page is evictable--i.e., should be placed on active/inactive >> - * lists vs unevictable list. >> - * >> - * Reasons page might not be evictable: >> - * (1) page's mapping marked unevictable >> - * (2) page is part of an mlocked VMA >> - * >> - */ >> -int page_evictable(struct page *page) >> -{ >> - int ret; >> - >> - /* Prevent address_space of inode and swap cache from being freed */ >> - rcu_read_lock(); >> - ret = !mapping_unevictable(page_mapping(page)) && !PageMlocked(page); >> - rcu_read_unlock(); >> - return ret; >> -} >> - >> /** >> * check_move_unevictable_pages - check pages for evictability and move to >> * appropriate zone lru list >> -- >> 1.8.3.1 >>
On Fri, Mar 13, 2020 at 12:46 PM Yang Shi <yang.shi@linux.alibaba.com> wrote: > > > > On 3/13/20 12:33 PM, Shakeel Butt wrote: > > On Fri, Mar 13, 2020 at 11:34 AM Yang Shi <yang.shi@linux.alibaba.com> wrote: > >> When backporting commit 9c4e6b1a7027 ("mm, mlock, vmscan: no more > >> skipping pagevecs") to our 4.9 kernel, our test bench noticed around 10% > >> down with a couple of vm-scalability's test cases (lru-file-readonce, > >> lru-file-readtwice and lru-file-mmap-read). I didn't see that much down > >> on my VM (32c-64g-2nodes). It might be caused by the test configuration, > >> which is 32c-256g with NUMA disabled and the tests were run in root memcg, > >> so the tests actually stress only one inactive and active lru. It > >> sounds not very usual in mordern production environment. > >> > >> That commit did two major changes: > >> 1. Call page_evictable() > >> 2. Use smp_mb to force the PG_lru set visible > >> > >> It looks they contribute the most overhead. The page_evictable() is a > >> function which does function prologue and epilogue, and that was used by > >> page reclaim path only. However, lru add is a very hot path, so it > >> sounds better to make it inline. However, it calls page_mapping() which > >> is not inlined either, but the disassemble shows it doesn't do push and > >> pop operations and it sounds not very straightforward to inline it. > >> > >> Other than this, it sounds smp_mb() is not necessary for x86 since > >> SetPageLRU is atomic which enforces memory barrier already, replace it > >> with smp_mb__after_atomic() in the following patch. > >> > >> With the two fixes applied, the tests can get back around 5% on that > >> test bench and get back normal on my VM. Since the test bench > >> configuration is not that usual and I also saw around 6% up on the > >> latest upstream, so it sounds good enough IMHO. > >> > >> The below is test data (lru-file-readtwice throughput) against the v5.6-rc4: > >> mainline w/ inline fix > >> 150MB 154MB > >> > > What is the test setup for the above experiment? I would like to get a repro. > > Just startup a VM with two nodes, then run case-lru-file-readtwice or > case-lru-file-readonce in vm-scalability in root memcg or with memcg > disabled. Then get the average throughput (dd result) from the test. > Our test bench uses the script from lkp, but I just ran it manually. > Single node VM should be more obvious showed in my test. > Thanks, I will try this on a real machine.
On 3/13/20 12:50 PM, Shakeel Butt wrote: > On Fri, Mar 13, 2020 at 12:46 PM Yang Shi <yang.shi@linux.alibaba.com> wrote: >> >> >> On 3/13/20 12:33 PM, Shakeel Butt wrote: >>> On Fri, Mar 13, 2020 at 11:34 AM Yang Shi <yang.shi@linux.alibaba.com> wrote: >>>> When backporting commit 9c4e6b1a7027 ("mm, mlock, vmscan: no more >>>> skipping pagevecs") to our 4.9 kernel, our test bench noticed around 10% >>>> down with a couple of vm-scalability's test cases (lru-file-readonce, >>>> lru-file-readtwice and lru-file-mmap-read). I didn't see that much down >>>> on my VM (32c-64g-2nodes). It might be caused by the test configuration, >>>> which is 32c-256g with NUMA disabled and the tests were run in root memcg, >>>> so the tests actually stress only one inactive and active lru. It >>>> sounds not very usual in mordern production environment. >>>> >>>> That commit did two major changes: >>>> 1. Call page_evictable() >>>> 2. Use smp_mb to force the PG_lru set visible >>>> >>>> It looks they contribute the most overhead. The page_evictable() is a >>>> function which does function prologue and epilogue, and that was used by >>>> page reclaim path only. However, lru add is a very hot path, so it >>>> sounds better to make it inline. However, it calls page_mapping() which >>>> is not inlined either, but the disassemble shows it doesn't do push and >>>> pop operations and it sounds not very straightforward to inline it. >>>> >>>> Other than this, it sounds smp_mb() is not necessary for x86 since >>>> SetPageLRU is atomic which enforces memory barrier already, replace it >>>> with smp_mb__after_atomic() in the following patch. >>>> >>>> With the two fixes applied, the tests can get back around 5% on that >>>> test bench and get back normal on my VM. Since the test bench >>>> configuration is not that usual and I also saw around 6% up on the >>>> latest upstream, so it sounds good enough IMHO. >>>> >>>> The below is test data (lru-file-readtwice throughput) against the v5.6-rc4: >>>> mainline w/ inline fix >>>> 150MB 154MB >>>> >>> What is the test setup for the above experiment? I would like to get a repro. >> Just startup a VM with two nodes, then run case-lru-file-readtwice or >> case-lru-file-readonce in vm-scalability in root memcg or with memcg >> disabled. Then get the average throughput (dd result) from the test. >> Our test bench uses the script from lkp, but I just ran it manually. >> Single node VM should be more obvious showed in my test. >> > Thanks, I will try this on a real machine Real machine should be better. Our test bench is bare metal with NUMA disabled. On my test VM it is not that obvious.
On Sat, Mar 14, 2020 at 02:34:35AM +0800, Yang Shi wrote: > -extern int page_evictable(struct page *page); > +/* This seems to be in kernel-doc format already; could you add the extra '*' so it is added to the fine documentation? > + * page_evictable - test whether a page is evictable > + * @page: the page to test > + * > + * Test whether page is evictable--i.e., should be placed on active/inactive > + * lists vs unevictable list. > + * > + * Reasons page might not be evictable: > + * (1) page's mapping marked unevictable > + * (2) page is part of an mlocked VMA > + * > + */ > +static inline int page_evictable(struct page *page) > +{ > + int ret; > + > + /* Prevent address_space of inode and swap cache from being freed */ > + rcu_read_lock(); > + ret = !mapping_unevictable(page_mapping(page)) && !PageMlocked(page); > + rcu_read_unlock(); > + return ret; > +} This seems like it should return bool ... that might even lead to code generation improvement.
On 3/14/20 9:01 AM, Matthew Wilcox wrote: > On Sat, Mar 14, 2020 at 02:34:35AM +0800, Yang Shi wrote: >> -extern int page_evictable(struct page *page); >> +/* > This seems to be in kernel-doc format already; could you add the extra > '*' so it is added to the fine documentation? Yes, sure. > >> + * page_evictable - test whether a page is evictable >> + * @page: the page to test >> + * >> + * Test whether page is evictable--i.e., should be placed on active/inactive >> + * lists vs unevictable list. >> + * >> + * Reasons page might not be evictable: >> + * (1) page's mapping marked unevictable >> + * (2) page is part of an mlocked VMA >> + * >> + */ >> +static inline int page_evictable(struct page *page) >> +{ >> + int ret; >> + >> + /* Prevent address_space of inode and swap cache from being freed */ >> + rcu_read_lock(); >> + ret = !mapping_unevictable(page_mapping(page)) && !PageMlocked(page); >> + rcu_read_unlock(); >> + return ret; >> +} > This seems like it should return bool ... that might even lead to code > generation improvement. Thanks for catching this. It looks mapping_unevictable() needs to be converted to bool as well.
diff --git a/include/linux/swap.h b/include/linux/swap.h index 1e99f7a..297eb66 100644 --- a/include/linux/swap.h +++ b/include/linux/swap.h @@ -374,7 +374,29 @@ extern unsigned long mem_cgroup_shrink_node(struct mem_cgroup *mem, #define node_reclaim_mode 0 #endif -extern int page_evictable(struct page *page); +/* + * page_evictable - test whether a page is evictable + * @page: the page to test + * + * Test whether page is evictable--i.e., should be placed on active/inactive + * lists vs unevictable list. + * + * Reasons page might not be evictable: + * (1) page's mapping marked unevictable + * (2) page is part of an mlocked VMA + * + */ +static inline int page_evictable(struct page *page) +{ + int ret; + + /* Prevent address_space of inode and swap cache from being freed */ + rcu_read_lock(); + ret = !mapping_unevictable(page_mapping(page)) && !PageMlocked(page); + rcu_read_unlock(); + return ret; +} + extern void check_move_unevictable_pages(struct pagevec *pvec); extern int kswapd_run(int nid); diff --git a/mm/vmscan.c b/mm/vmscan.c index 8763705..855c395 100644 --- a/mm/vmscan.c +++ b/mm/vmscan.c @@ -4277,29 +4277,6 @@ int node_reclaim(struct pglist_data *pgdat, gfp_t gfp_mask, unsigned int order) } #endif -/* - * page_evictable - test whether a page is evictable - * @page: the page to test - * - * Test whether page is evictable--i.e., should be placed on active/inactive - * lists vs unevictable list. - * - * Reasons page might not be evictable: - * (1) page's mapping marked unevictable - * (2) page is part of an mlocked VMA - * - */ -int page_evictable(struct page *page) -{ - int ret; - - /* Prevent address_space of inode and swap cache from being freed */ - rcu_read_lock(); - ret = !mapping_unevictable(page_mapping(page)) && !PageMlocked(page); - rcu_read_unlock(); - return ret; -} - /** * check_move_unevictable_pages - check pages for evictability and move to * appropriate zone lru list
When backporting commit 9c4e6b1a7027 ("mm, mlock, vmscan: no more skipping pagevecs") to our 4.9 kernel, our test bench noticed around 10% down with a couple of vm-scalability's test cases (lru-file-readonce, lru-file-readtwice and lru-file-mmap-read). I didn't see that much down on my VM (32c-64g-2nodes). It might be caused by the test configuration, which is 32c-256g with NUMA disabled and the tests were run in root memcg, so the tests actually stress only one inactive and active lru. It sounds not very usual in mordern production environment. That commit did two major changes: 1. Call page_evictable() 2. Use smp_mb to force the PG_lru set visible It looks they contribute the most overhead. The page_evictable() is a function which does function prologue and epilogue, and that was used by page reclaim path only. However, lru add is a very hot path, so it sounds better to make it inline. However, it calls page_mapping() which is not inlined either, but the disassemble shows it doesn't do push and pop operations and it sounds not very straightforward to inline it. Other than this, it sounds smp_mb() is not necessary for x86 since SetPageLRU is atomic which enforces memory barrier already, replace it with smp_mb__after_atomic() in the following patch. With the two fixes applied, the tests can get back around 5% on that test bench and get back normal on my VM. Since the test bench configuration is not that usual and I also saw around 6% up on the latest upstream, so it sounds good enough IMHO. The below is test data (lru-file-readtwice throughput) against the v5.6-rc4: mainline w/ inline fix 150MB 154MB With this patch the throughput gets 2.67% up. The data with using smp_mb__after_atomic() is showed in the following patch. Fixes: 9c4e6b1a7027 ("mm, mlock, vmscan: no more skipping pagevecs") Cc: Shakeel Butt <shakeelb@google.com> Cc: Vlastimil Babka <vbabka@suse.cz> Signed-off-by: Yang Shi <yang.shi@linux.alibaba.com> --- include/linux/swap.h | 24 +++++++++++++++++++++++- mm/vmscan.c | 23 ----------------------- 2 files changed, 23 insertions(+), 24 deletions(-)