memcg: dirty pages accounting and limiting infrastructure
in [Kernel]
|
Prev: [PATCH -tip 5/9] perf probe: Use elfutils-libdw for analyzing debuginfo
Next: [announce] gujin GPL bootloader version 2.8
From: Andrea Righi on 8 Mar 2010 19:20 On Mon, Mar 08, 2010 at 05:31:00PM +0900, KAMEZAWA Hiroyuki wrote: > On Mon, 8 Mar 2010 17:07:11 +0900 > Daisuke Nishimura <nishimura(a)mxp.nes.nec.co.jp> wrote: > > > On Mon, 8 Mar 2010 11:37:11 +0900, KAMEZAWA Hiroyuki <kamezawa.hiroyu(a)jp.fujitsu.com> wrote: > > > On Mon, 8 Mar 2010 11:17:24 +0900 > > > Daisuke Nishimura <nishimura(a)mxp.nes.nec.co.jp> wrote: > > > > > > > > But IIRC, clear_writeback is done under treelock.... No ? > > > > > > > > > The place where NR_WRITEBACK is updated is out of tree_lock. > > > > > > > > 1311 int test_clear_page_writeback(struct page *page) > > > > 1312 { > > > > 1313 struct address_space *mapping = page_mapping(page); > > > > 1314 int ret; > > > > 1315 > > > > 1316 if (mapping) { > > > > 1317 struct backing_dev_info *bdi = mapping->backing_dev_info; > > > > 1318 unsigned long flags; > > > > 1319 > > > > 1320 spin_lock_irqsave(&mapping->tree_lock, flags); > > > > 1321 ret = TestClearPageWriteback(page); > > > > 1322 if (ret) { > > > > 1323 radix_tree_tag_clear(&mapping->page_tree, > > > > 1324 page_index(page), > > > > 1325 PAGECACHE_TAG_WRITEBACK); > > > > 1326 if (bdi_cap_account_writeback(bdi)) { > > > > 1327 __dec_bdi_stat(bdi, BDI_WRITEBACK); > > > > 1328 __bdi_writeout_inc(bdi); > > > > 1329 } > > > > 1330 } > > > > 1331 spin_unlock_irqrestore(&mapping->tree_lock, flags); > > > > 1332 } else { > > > > 1333 ret = TestClearPageWriteback(page); > > > > 1334 } > > > > 1335 if (ret) > > > > 1336 dec_zone_page_state(page, NR_WRITEBACK); > > > > 1337 return ret; > > > > 1338 } > > > > > > We can move this up to under tree_lock. Considering memcg, all our target has "mapping". > > > > > > If we newly account bounce-buffers (for NILFS, FUSE, etc..), which has no ->mapping, > > > we need much more complex new charge/uncharge theory. > > > > > > But yes, adding new lock scheme seems complicated. (Sorry Andrea.) > > > My concerns is performance. We may need somehing new re-implementation of > > > locks/migrate/charge/uncharge. > > > > > I agree. Performance is my concern too. > > > > I made a patch below and measured the time(average of 10 times) of kernel build > > on tmpfs(make -j8 on 8 CPU machine with 2.6.33 defconfig). > > > > <before> > > - root cgroup: 190.47 sec > > - child cgroup: 192.81 sec > > > > <after> > > - root cgroup: 191.06 sec > > - child cgroup: 193.06 sec > > > > Hmm... about 0.3% slower for root, 0.1% slower for child. > > > > Hmm...accepatable ? (sounds it's in error-range) > > BTW, why local_irq_disable() ? > local_irq_save()/restore() isn't better ? Probably there's not the overhead of saving flags? Anyway, it would make the code much more readable... Thanks, -Andrea > > Thanks, > -Kame > > > === > > From: Daisuke Nishimura <nishimura(a)mxp.nes.nec.co.jp> > > > > In current implementation, we don't have to disable irq at lock_page_cgroup() > > because the lock is never acquired in interrupt context. > > But we are going to do it in later patch, so this patch encloses all of > > lock_page_cgroup()/unlock_page_cgroup() with irq_disabled()/irq_enabled(). > > > > Signed-off-by: Daisuke Nishimura <nishimura(a)mxp.nes.nec.co.jp> > > --- > > mm/memcontrol.c | 17 +++++++++++++++++ > > 1 files changed, 17 insertions(+), 0 deletions(-) > > > > diff --git a/mm/memcontrol.c b/mm/memcontrol.c > > index 02ea959..e5ae1a1 100644 > > --- a/mm/memcontrol.c > > +++ b/mm/memcontrol.c > > @@ -1359,6 +1359,7 @@ void mem_cgroup_update_file_mapped(struct page *page, int val) > > if (unlikely(!pc)) > > return; > > > > + local_irq_disable(); > > lock_page_cgroup(pc); > > mem = pc->mem_cgroup; > > if (!mem) > > @@ -1374,6 +1375,7 @@ void mem_cgroup_update_file_mapped(struct page *page, int val) > > > > done: > > unlock_page_cgroup(pc); > > + local_irq_enable(); > > } > > > > /* > > @@ -1711,6 +1713,7 @@ struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) > > VM_BUG_ON(!PageLocked(page)); > > > > pc = lookup_page_cgroup(page); > > + local_irq_disable(); > > lock_page_cgroup(pc); > > if (PageCgroupUsed(pc)) { > > mem = pc->mem_cgroup; > > @@ -1726,6 +1729,7 @@ struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) > > rcu_read_unlock(); > > } > > unlock_page_cgroup(pc); > > + local_irq_enable(); > > return mem; > > } > > > > @@ -1742,9 +1746,11 @@ static void __mem_cgroup_commit_charge(struct mem_cgroup *mem, > > if (!mem) > > return; > > > > + local_irq_disable(); > > lock_page_cgroup(pc); > > if (unlikely(PageCgroupUsed(pc))) { > > unlock_page_cgroup(pc); > > + local_irq_enable(); > > mem_cgroup_cancel_charge(mem); > > return; > > } > > @@ -1775,6 +1781,7 @@ static void __mem_cgroup_commit_charge(struct mem_cgroup *mem, > > mem_cgroup_charge_statistics(mem, pc, true); > > > > unlock_page_cgroup(pc); > > + local_irq_enable(); > > /* > > * "charge_statistics" updated event counter. Then, check it. > > * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree. > > @@ -1844,12 +1851,14 @@ static int mem_cgroup_move_account(struct page_cgroup *pc, > > struct mem_cgroup *from, struct mem_cgroup *to, bool uncharge) > > { > > int ret = -EINVAL; > > + local_irq_disable(); > > lock_page_cgroup(pc); > > if (PageCgroupUsed(pc) && pc->mem_cgroup == from) { > > __mem_cgroup_move_account(pc, from, to, uncharge); > > ret = 0; > > } > > unlock_page_cgroup(pc); > > + local_irq_enable(); > > /* > > * check events > > */ > > @@ -1981,12 +1990,15 @@ int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, > > pc = lookup_page_cgroup(page); > > if (!pc) > > return 0; > > + local_irq_disable(); > > lock_page_cgroup(pc); > > if (PageCgroupUsed(pc)) { > > unlock_page_cgroup(pc); > > + local_irq_enable(); > > return 0; > > } > > unlock_page_cgroup(pc); > > + local_irq_enable(); > > } > > > > if (unlikely(!mm && !mem)) > > @@ -2182,6 +2194,7 @@ __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype) > > if (unlikely(!pc || !PageCgroupUsed(pc))) > > return NULL; > > > > + local_irq_disable(); > > lock_page_cgroup(pc); > > > > mem = pc->mem_cgroup; > > @@ -2222,6 +2235,7 @@ __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype) > > > > mz = page_cgroup_zoneinfo(pc); > > unlock_page_cgroup(pc); > > + local_irq_enable(); > > > > memcg_check_events(mem, page); > > /* at swapout, this memcg will be accessed to record to swap */ > > @@ -2232,6 +2246,7 @@ __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype) > > > > unlock_out: > > unlock_page_cgroup(pc); > > + local_irq_enable(); > > return NULL; > > } > > > > @@ -2424,12 +2439,14 @@ int mem_cgroup_prepare_migration(struct page *page, struct mem_cgroup **ptr) > > return 0; > > > > pc = lookup_page_cgroup(page); > > + local_irq_disable(); > > lock_page_cgroup(pc); > > if (PageCgroupUsed(pc)) { > > mem = pc->mem_cgroup; > > css_get(&mem->css); > > } > > unlock_page_cgroup(pc); > > + local_irq_enable(); > > > > if (mem) { > > ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem, false); > > -- > > 1.6.4 > > > > -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo(a)vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
From: KAMEZAWA Hiroyuki on 8 Mar 2010 19:30 On Tue, 9 Mar 2010 01:12:52 +0100 Andrea Righi <arighi(a)develer.com> wrote: > On Mon, Mar 08, 2010 at 05:31:00PM +0900, KAMEZAWA Hiroyuki wrote: > > On Mon, 8 Mar 2010 17:07:11 +0900 > > Daisuke Nishimura <nishimura(a)mxp.nes.nec.co.jp> wrote: > > > > > On Mon, 8 Mar 2010 11:37:11 +0900, KAMEZAWA Hiroyuki <kamezawa.hiroyu(a)jp.fujitsu.com> wrote: > > > > On Mon, 8 Mar 2010 11:17:24 +0900 > > > > Daisuke Nishimura <nishimura(a)mxp.nes.nec.co.jp> wrote: > > > > > > > > > > But IIRC, clear_writeback is done under treelock.... No ? > > > > > > > > > > > The place where NR_WRITEBACK is updated is out of tree_lock. > > > > > > > > > > 1311 int test_clear_page_writeback(struct page *page) > > > > > 1312 { > > > > > 1313 struct address_space *mapping = page_mapping(page); > > > > > 1314 int ret; > > > > > 1315 > > > > > 1316 if (mapping) { > > > > > 1317 struct backing_dev_info *bdi = mapping->backing_dev_info; > > > > > 1318 unsigned long flags; > > > > > 1319 > > > > > 1320 spin_lock_irqsave(&mapping->tree_lock, flags); > > > > > 1321 ret = TestClearPageWriteback(page); > > > > > 1322 if (ret) { > > > > > 1323 radix_tree_tag_clear(&mapping->page_tree, > > > > > 1324 page_index(page), > > > > > 1325 PAGECACHE_TAG_WRITEBACK); > > > > > 1326 if (bdi_cap_account_writeback(bdi)) { > > > > > 1327 __dec_bdi_stat(bdi, BDI_WRITEBACK); > > > > > 1328 __bdi_writeout_inc(bdi); > > > > > 1329 } > > > > > 1330 } > > > > > 1331 spin_unlock_irqrestore(&mapping->tree_lock, flags); > > > > > 1332 } else { > > > > > 1333 ret = TestClearPageWriteback(page); > > > > > 1334 } > > > > > 1335 if (ret) > > > > > 1336 dec_zone_page_state(page, NR_WRITEBACK); > > > > > 1337 return ret; > > > > > 1338 } > > > > > > > > We can move this up to under tree_lock. Considering memcg, all our target has "mapping". > > > > > > > > If we newly account bounce-buffers (for NILFS, FUSE, etc..), which has no ->mapping, > > > > we need much more complex new charge/uncharge theory. > > > > > > > > But yes, adding new lock scheme seems complicated. (Sorry Andrea.) > > > > My concerns is performance. We may need somehing new re-implementation of > > > > locks/migrate/charge/uncharge. > > > > > > > I agree. Performance is my concern too. > > > > > > I made a patch below and measured the time(average of 10 times) of kernel build > > > on tmpfs(make -j8 on 8 CPU machine with 2.6.33 defconfig). > > > > > > <before> > > > - root cgroup: 190.47 sec > > > - child cgroup: 192.81 sec > > > > > > <after> > > > - root cgroup: 191.06 sec > > > - child cgroup: 193.06 sec > > > > > > Hmm... about 0.3% slower for root, 0.1% slower for child. > > > > > > > Hmm...accepatable ? (sounds it's in error-range) > > > > BTW, why local_irq_disable() ? > > local_irq_save()/restore() isn't better ? > > Probably there's not the overhead of saving flags? maybe. > Anyway, it would make the code much more readable... > ok. please go ahead in this direction. Nishimura-san, would you post an independent patch ? If no, Andrea-san, please. Thanks, -Kame -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo(a)vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
From: KAMEZAWA Hiroyuki on 8 Mar 2010 19:30 On Tue, 9 Mar 2010 09:18:45 +0900 Daisuke Nishimura <nishimura(a)mxp.nes.nec.co.jp> wrote: > On Mon, 8 Mar 2010 17:31:00 +0900, KAMEZAWA Hiroyuki <kamezawa.hiroyu(a)jp.fujitsu.com> wrote: > > On Mon, 8 Mar 2010 17:07:11 +0900 > > Daisuke Nishimura <nishimura(a)mxp.nes.nec.co.jp> wrote: > > Hmm...accepatable ? (sounds it's in error-range) > > > > BTW, why local_irq_disable() ? > > local_irq_save()/restore() isn't better ? > > > I don't have any strong reason. All of lock_page_cgroup() is *now* called w/o irq disabled, > so I used just disable()/enable() instead of save()/restore(). My point is, this will be used under treelock soon. Thanks, -Kame -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo(a)vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
From: Daisuke Nishimura on 8 Mar 2010 19:30 On Mon, 8 Mar 2010 17:31:00 +0900, KAMEZAWA Hiroyuki <kamezawa.hiroyu(a)jp.fujitsu.com> wrote: > On Mon, 8 Mar 2010 17:07:11 +0900 > Daisuke Nishimura <nishimura(a)mxp.nes.nec.co.jp> wrote: > > > On Mon, 8 Mar 2010 11:37:11 +0900, KAMEZAWA Hiroyuki <kamezawa.hiroyu(a)jp.fujitsu.com> wrote: > > > On Mon, 8 Mar 2010 11:17:24 +0900 > > > Daisuke Nishimura <nishimura(a)mxp.nes.nec.co.jp> wrote: > > > > > > > > But IIRC, clear_writeback is done under treelock.... No ? > > > > > > > > > The place where NR_WRITEBACK is updated is out of tree_lock. > > > > > > > > 1311 int test_clear_page_writeback(struct page *page) > > > > 1312 { > > > > 1313 struct address_space *mapping = page_mapping(page); > > > > 1314 int ret; > > > > 1315 > > > > 1316 if (mapping) { > > > > 1317 struct backing_dev_info *bdi = mapping->backing_dev_info; > > > > 1318 unsigned long flags; > > > > 1319 > > > > 1320 spin_lock_irqsave(&mapping->tree_lock, flags); > > > > 1321 ret = TestClearPageWriteback(page); > > > > 1322 if (ret) { > > > > 1323 radix_tree_tag_clear(&mapping->page_tree, > > > > 1324 page_index(page), > > > > 1325 PAGECACHE_TAG_WRITEBACK); > > > > 1326 if (bdi_cap_account_writeback(bdi)) { > > > > 1327 __dec_bdi_stat(bdi, BDI_WRITEBACK); > > > > 1328 __bdi_writeout_inc(bdi); > > > > 1329 } > > > > 1330 } > > > > 1331 spin_unlock_irqrestore(&mapping->tree_lock, flags); > > > > 1332 } else { > > > > 1333 ret = TestClearPageWriteback(page); > > > > 1334 } > > > > 1335 if (ret) > > > > 1336 dec_zone_page_state(page, NR_WRITEBACK); > > > > 1337 return ret; > > > > 1338 } > > > > > > We can move this up to under tree_lock. Considering memcg, all our target has "mapping". > > > > > > If we newly account bounce-buffers (for NILFS, FUSE, etc..), which has no ->mapping, > > > we need much more complex new charge/uncharge theory. > > > > > > But yes, adding new lock scheme seems complicated. (Sorry Andrea.) > > > My concerns is performance. We may need somehing new re-implementation of > > > locks/migrate/charge/uncharge. > > > > > I agree. Performance is my concern too. > > > > I made a patch below and measured the time(average of 10 times) of kernel build > > on tmpfs(make -j8 on 8 CPU machine with 2.6.33 defconfig). > > > > <before> > > - root cgroup: 190.47 sec > > - child cgroup: 192.81 sec > > > > <after> > > - root cgroup: 191.06 sec > > - child cgroup: 193.06 sec > > > > Hmm... about 0.3% slower for root, 0.1% slower for child. > > > > Hmm...accepatable ? (sounds it's in error-range) > > BTW, why local_irq_disable() ? > local_irq_save()/restore() isn't better ? > I don't have any strong reason. All of lock_page_cgroup() is *now* called w/o irq disabled, so I used just disable()/enable() instead of save()/restore(). I think disable()/enable() is better in those cases because we need not to save/restore eflags register by pushf/popf, but, I don't have any numbers though, there wouldn't be a big difference in performance. Thanks, Daisuke Nishimura. -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo(a)vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
From: Daisuke Nishimura on 8 Mar 2010 20:10
On Tue, 9 Mar 2010 09:20:54 +0900, KAMEZAWA Hiroyuki <kamezawa.hiroyu(a)jp.fujitsu.com> wrote: > On Tue, 9 Mar 2010 09:18:45 +0900 > Daisuke Nishimura <nishimura(a)mxp.nes.nec.co.jp> wrote: > > > On Mon, 8 Mar 2010 17:31:00 +0900, KAMEZAWA Hiroyuki <kamezawa.hiroyu(a)jp.fujitsu.com> wrote: > > > On Mon, 8 Mar 2010 17:07:11 +0900 > > > Daisuke Nishimura <nishimura(a)mxp.nes.nec.co.jp> wrote: > > > > Hmm...accepatable ? (sounds it's in error-range) > > > > > > BTW, why local_irq_disable() ? > > > local_irq_save()/restore() isn't better ? > > > > > I don't have any strong reason. All of lock_page_cgroup() is *now* called w/o irq disabled, > > so I used just disable()/enable() instead of save()/restore(). > > My point is, this will be used under treelock soon. > I agree. I'll update the patch using save()/restore(), and repost later. Thanks, Daisuke Nishimura. -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo(a)vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/ |