- 22 Mar, 2012 40 commits
-
-
Naoya Horiguchi authored
Currently we can't do task migration among memory cgroups without THP split, which means processes heavily using THP experience large overhead in task migration. This patch introduces the code for moving charge of THP and makes THP more valuable. Signed-off-by:
Naoya Horiguchi <n-horiguchi@ah.jp.nec.com> Acked-by:
Hillf Danton <dhillf@gmail.com> Cc: Andrea Arcangeli <aarcange@redhat.com> Acked-by:
KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: David Rientjes <rientjes@google.com> Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp> Signed-off-by:
Andrew Morton <akpm@linux-foundation.org> Signed-off-by:
Linus Torvalds <torvalds@linux-foundation.org>
-
Naoya Horiguchi authored
These macros will be used in a later patch, where all usages are expected to be optimized away without #ifdef CONFIG_TRANSPARENT_HUGEPAGE. But to detect unexpected usages, we convert the existing BUG() to BUILD_BUG(). [akpm@linux-foundation.org: fix build in mm/pgtable-generic.c] Signed-off-by:
Andrea Arcangeli <aarcange@redhat.com> Reviewed-by:
David Rientjes <rientjes@google.com> Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp> Signed-off-by:
-
Naoya Horiguchi authored
- Replace lengthy function name is_target_pte_for_mc() with a shorter one in order to avoid ugly line breaks. - explicitly use MC_TARGET_* instead of simply using integers. Signed-off-by:
-
Jeff Liu authored
Signed-off-by:
Jie Liu <jeff.liu@oracle.com> Signed-off-by:
-
Anton Vorontsov authored
In the following code: if (type == _MEM) thresholds = &memcg->thresholds; else if (type == _MEMSWAP) thresholds = &memcg->memsw_thresholds; else BUG(); BUG_ON(!thresholds); The BUG_ON() seems redundant. Signed-off-by:
Anton Vorontsov <anton.vorontsov@linaro.org> Signed-off-by:
-
Andrew Morton authored
A grammatical fix. Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Signed-off-by:
-
KAMEZAWA Hiroyuki authored
mem_cgroup_begin_update_page_stat() should be very fast because it's called very frequently. Now, it needs to look up page_cgroup and its memcg....this is slow. This patch adds a global variable to check "any memcg is moving or not". With this, the caller doesn't need to visit page_cgroup and memcg. Here is a test result. A test program makes page faults onto a file, MAP_SHARED and makes each page's page_mapcount(page) > 1, and free the range by madvise() and page fault again. This program causes 26214400 times of page fault onto a file(size was 1G.) and shows shows the cost of mem_cgroup_begin_update_page_stat(). Before this patch for mem_cgroup_begin_update_page_stat() [kamezawa@bluextal test]$ time ./mmap 1G real 0m21.765s user 0m5.999s sys 0m15.434s 27.46% mmap mmap [.] reader 21.15% mmap [kernel.kallsyms] [k] page_fault 9.17% mmap [kernel.kallsyms] [k] filemap_fault 2.96% mmap [kernel.kallsyms] [k] __do_fault 2.83% mmap [kernel.kallsyms] [k] __mem_cgroup_begin_update_page_stat After this patch [root@bluextal test]# time ./mmap 1G real 0m21.373s user 0m6.113s sys 0m15.016s In usual path, calls to __mem_cgroup_begin_update_page_stat() goes away. Note: we may be able to remove this optimization in future if we can get pointer to memcg directly from struct page. [akpm@linux-foundation.org: don't return a void] Signed-off-by:Greg Thelen <gthelen@google.com> Acked-by:
Johannes Weiner <hannes@cmpxchg.org> Cc: Michal Hocko <mhocko@suse.cz> Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Cc: Ying Han <yinghan@google.com> Signed-off-by:
-
KAMEZAWA Hiroyuki authored
With the new lock scheme for updating memcg's page stat, we don't need a flag PCG_FILE_MAPPED which was duplicated information of page_mapped(). [hughd@google.com: cosmetic fix] [hughd@google.com: add comment to MEM_CGROUP_CHARGE_TYPE_MAPPED case in __mem_cgroup_uncharge_common()] Signed-off-by:
-
KAMEZAWA Hiroyuki authored
Now, page-stat-per-memcg is recorded into per page_cgroup flag by duplicating page's status into the flag. The reason is that memcg has a feature to move a page from a group to another group and we have race between "move" and "page stat accounting", Under current logic, assume CPU-A and CPU-B. CPU-A does "move" and CPU-B does "page stat accounting". When CPU-A goes 1st, CPU-A CPU-B update "struct page" info. move_lock_mem_cgroup(memcg) see pc->flags copy page stat to new group overwrite pc->mem_cgroup. move_unlock_mem_cgroup(memcg) move_lock_mem_cgroup(mem) set pc->flags update page stat accounting move_unlock_mem_cgroup(mem) stat accounting is guarded by move_lock_mem_cgroup() and "move" logic (CPU-A) doesn't see changes in "struct page" information. But it's costly to have the same information both in 'struct page' and 'struct page_cgroup'. And, there is a potential problem. For example, assume we have PG_dirty accounting in memcg. PG_..is a flag for struct page. PCG_ is a flag for struct page_cgroup. (This is just an example. The same problem can be found in any kind of page stat accounting.) CPU-A CPU-B TestSet PG_dirty (delay) TestClear PG_dirty if (TestClear(PCG_dirty)) memcg->nr_dirty-- if (TestSet(PCG_dirty)) memcg->nr_dirty++ Here, memcg->nr_dirty = +1, this is wrong. This race was reported by Greg Thelen <gthelen@google.com>. Now, only FILE_MAPPED is supported but fortunately, it's serialized by page table lock and this is not real bug, _now_, If this potential problem is caused by having duplicated information in struct page and struct page_cgroup, we may be able to fix this by using original 'struct page' information. But we'll have a problem in "move account" Assume we use only PG_dirty. CPU-A CPU-B TestSet PG_dirty (delay) move_lock_mem_cgroup() if (PageDirty(page)) new_memcg->nr_dirty++ pc->mem_cgroup = new_memcg; move_unlock_mem_cgroup() move_lock_mem_cgroup() memcg = pc->mem_cgroup new_memcg->nr_dirty++ accounting information may be double-counted. This was original reason to have PCG_xxx flags but it seems PCG_xxx has another problem. I think we need a bigger lock as move_lock_mem_cgroup(page) TestSetPageDirty(page) update page stats (without any checks) move_unlock_mem_cgroup(page) This fixes both of problems and we don't have to duplicate page flag into page_cgroup. Please note: move_lock_mem_cgroup() is held only when there are possibility of "account move" under the system. So, in most path, status update will go without atomic locks. This patch introduces mem_cgroup_begin_update_page_stat() and mem_cgroup_end_update_page_stat() both should be called at modifying 'struct page' information if memcg takes care of it. as mem_cgroup_begin_update_page_stat() modify page information mem_cgroup_update_page_stat() => never check any 'struct page' info, just update counters. mem_cgroup_end_update_page_stat(). This patch is slow because we need to call begin_update_page_stat()/ end_update_page_stat() regardless of accounted will be changed or not. A following patch adds an easy optimization and reduces the cost. [akpm@linux-foundation.org: s/lock/locked/] [hughd@google.com: fix deadlock by avoiding stat lock when anon] Signed-off-by:Hugh Dickins <hughd@google.com> Signed-off-by:
-
KAMEZAWA Hiroyuki authored
PCG_MOVE_LOCK is used for bit spinlock to avoid race between overwriting pc->mem_cgroup and page statistics accounting per memcg. This lock helps to avoid the race but the race is very rare because moving tasks between cgroup is not a usual job. So, it seems using 1bit per page is too costly. This patch changes this lock as per-memcg spinlock and removes PCG_MOVE_LOCK. If smaller lock is required, we'll be able to add some hashes but I'd like to start from this. Signed-off-by:
-
KAMEZAWA Hiroyuki authored
In memcg, for avoiding take-lock-irq-off at accessing page_cgroup, a logic, flag + rcu_read_lock(), is used. This works as following CPU-A CPU-B rcu_read_lock() set flag if(flag is set) take heavy lock do job. synchronize_rcu() rcu_read_unlock() take heavy lock. In recent discussion, it's argued that using per-cpu value for this flag just complicates the code because 'set flag' is very rare. This patch changes 'flag' implementation from percpu to atomic_t. This will be much simpler. Signed-off-by: -
KAMEZAWA Hiroyuki authored
As described in the log, I guess EXPORT was for preparing dirty accounting. But _now_, we don't need to export this. Remove this for now. Signed-off-by:
-
Konstantin Khlebnikov authored
This code was removed in 25edde03 ("vmscan: kill prev_priority completely") Signed-off-by:
Konstantin Khlebnikov <khlebnikov@openvz.org> Acked-by:
KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Signed-off-by:
-
KAMEZAWA Hiroyuki authored
We record 'the page is cache' with the PCG_CACHE bit in page_cgroup. Here, "CACHE" means anonymous user pages (and SwapCache). This doesn't include shmem. Considering callers, at charge/uncharge, the caller should know what the page is and we don't need to record it by using one bit per page. This patch removes PCG_CACHE bit and make callers of mem_cgroup_charge_statistics() to specify what the page is. About page migration: Mapping of the used page is not touched during migra tion (see page_remove_rmap) so we can rely on it and push the correct charge type down to __mem_cgroup_uncharge_common from end_migration for unused page. The force flag was misleading was abused for skipping the needless page_mapped() / PageCgroupMigration() check, as we know the unused page is no longer mapped and cleared the migration flag just a few lines up. But doing the checks is no biggie and it's not worth adding another flag just to skip them. [akpm@linux-foundation.org: checkpatch fixes] [hughd@google.com: fix PageAnon uncharging] Signed-off-by:
Michal Hocko <mhocko@suse.cz> Acked-by:
-
Hugh Dickins authored
Remove lock and unlock around css_get_next()'s call to idr_get_next(). memcg iterators (only users of css_get_next) already did rcu_read_lock(), and its comment demands that; but add a WARN_ON_ONCE to make sure of it. Signed-off-by:
Li Zefan <lizf@cn.fujitsu.com> Cc: Eric Dumazet <eric.dumazet@gmail.com> Acked-by:
Tejun Heo <tj@kernel.org> Signed-off-by:
-
Hugh Dickins authored
Commit c1e2ee2d ("memcg: replace ss->id_lock with a rwlock") has now been seen to cause the unfair behavior we should have expected from converting a spinlock to an rwlock: softlockup in cgroup_mkdir(), whose get_new_cssid() is waiting for the wlock, while there are 19 tasks using the rlock in css_get_next() to get on with their memcg workload (in an artificial test, admittedly). Yet lib/idr.c was made suitable for RCU way back: revert that commit, restoring ss->id_lock to a spinlock. Signed-off-by:
-
Hugh Dickins authored
Make one small adjustment to idr_get_next(): take the height from the top layer (stable under RCU) instead of from the root (unprotected by RCU), as idr_find() does: so that it can be used with RCU locking. Copied comment on RCU locking from idr_find(). Signed-off-by:
-
KAMEZAWA Hiroyuki authored
Commit e94c8a9c ("memcg: make mem_cgroup_split_huge_fixup() more efficient") removed move_lock_page_cgroup(). So we do not have to check PageTransHuge in mem_cgroup_update_page_stat() and fallback into the locked accounting because both move_account() and thp split are done with compound_lock so they cannot race. The race between update vs. move is protected by mem_cgroup_stealed. PageTransHuge pages shouldn't appear in this code path currently because we are tracking only file pages at the moment but later we are planning to track also other pages (e.g. mlocked ones). Signed-off-by:
-
Hugh Dickins authored
Remove redundant returns from ends of functions, and one blank line. Signed-off-by:
Kirill A. Shutemov <kirill@shutemov.name> Acked-by:
-
Hugh Dickins authored
Mostly we use "enum lru_list lru": change those few "l"s to "lru"s. Signed-off-by:
-
Hugh Dickins authored
I never understood why we need a MEM_CGROUP_ZSTAT(mz, idx) macro to obscure the LRU counts. For easier searching? So call it lru_size rather than bare count (lru_length sounds better, but would be wrong, since each huge page raises lru_size hugely). Signed-off-by:
-
Hugh Dickins authored
Replace mem and mem_cont stragglers in memcontrol.c by memcg. Signed-off-by:
-
Hugh Dickins authored
Correct an #endif comment in memcontrol.h from MEM_CONT to MEM_RES_CTLR. Signed-off-by:
-
Shaohua Li authored
When swapon() was not passed the SWAP_FLAG_DISCARD option, sys_swapon() will still perform a discard operation. This can cause problems if discard is slow or buggy. Reverse the order of the check so that a discard operation is performed only if the sys_swapon() caller is attempting to enable discard. Signed-off-by:
Shaohua Li <shli@fusionio.com> Reported-by:
Holger Kiehl <Holger.Kiehl@dwd.de> Tested-by:
-
Konstantin Khlebnikov authored
Reset the reclaim mode in shrink_active_list() to RECLAIM_MODE_SINGLE | RECLAIM_MODE_ASYNC. (sync/async sign is used only in shrink_page_list and does not affect shrink_active_list) Currenly shrink_active_list() sometimes works in lumpy-reclaim mode, if RECLAIM_MODE_LUMPYRECLAIM is left over from an earlier shrink_inactive_list(). Meanwhile, in age_active_anon() sc->reclaim_mode is totally zero. So the current behavior is too complex and confusing, and this looks like bug. In general, shrink_active_list() populates the inactive list for the next shrink_inactive_list(). Lumpy shring_inactive_list() isolates pages around the chosen one from both the active and inactive lists. So, there is no reason for lumpy isolation in shrink_active_list(). See also: https://lkml.org/lkml/2012/3/15/583Signed-off-by:
-
Kautuk Consul authored
The comment above __insert_vm_struct seems to suggest that this function is also going to link the VMA with the anon_vma, but this is not true. This function only links the VMA to the mm->mm_rb tree and the mm->mmap linked list. [akpm@linux-foundation.org: improve comment layout and text] Signed-off-by:
Kautuk Consul <consul.kautuk@gmail.com> Acked-by:
-
Andrea Arcangeli authored
Without this fix the cpumask_of_node() for a fake=numa=2 is: cpumask 0 ff cpumask 1 ff with the fix it's correct and it's set to: cpumask 0 55 cpumask 1 aa Signed-off-by: -
Kautuk Consul authored
find_zone_movable_pfns_for_nodes() does not use its argument. Signed-off-by:
-
Kautuk Consul authored
add_from_early_node_map() is unused. Signed-off-by:
-
Hillf Danton authored
Return an errno upon failure to create inode kmem cache, and unregister the FS upon failure to mount. [akpm@linux-foundation.org: remove unneeded test of `error'] Signed-off-by:
-
Xiao Guangrong authored
After looking up the vma which covers or follows the cached search address, the following condition is always true: !prev_vma || (addr >= prev_vma->vm_end) so we can stop checking the previous VMA altogether. Signed-off-by:
Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com> Signed-off-by:
-
Steven Truelove authored
When calling shmget() with SHM_HUGETLB, shmget aligns the request size to PAGE_SIZE, but this is not sufficient. Modify hugetlb_file_setup() to align requests to the huge page size, and to accept an address argument so that all alignment checks can be performed in hugetlb_file_setup(), rather than in its callers. Change newseg() and mmap_pgoff() to match the new prototype and eliminate a now redundant alignment check. [akpm@linux-foundation.org: fix build] Signed-off-by:
Steven Truelove <steven.truelove@utoronto.ca> Cc: Hugh Dickins <hughd@google.com> Signed-off-by:
-
David Rientjes authored
Add the thread name and pid of the application that is allocating shm segments with MAP_HUGETLB without being a part of /proc/sys/vm/hugetlb_shm_group or having CAP_IPC_LOCK. This identifies the application so it may be fixed by avoiding using the deprecated exception (see Documentation/feature-removal-schedule.txt). Signed-off-by:
-
David Rientjes authored
There's no difference between sync_mm_rss() and __sync_task_rss_stat(), so fold the latter into the former. Signed-off-by:
-
David Rientjes authored
sync_mm_rss() can only be used for current to avoid race conditions in iterating and clearing its per-task counters. Remove the task argument for it and its helper function, __sync_task_rss_stat(), to avoid thinking it can be used safely for anything other than current. Signed-off-by:
-
David Gibson authored
hugetlbfs_{get,put}_quota() are badly named. They don't interact with the general quota handling code, and they don't much resemble its behaviour. Rather than being about maintaining limits on on-disk block usage by particular users, they are instead about maintaining limits on in-memory page usage (including anonymous MAP_PRIVATE copied-on-write pages) associated with a particular hugetlbfs filesystem instance. Worse, they work by having callbacks to the hugetlbfs filesystem code from the low-level page handling code, in particular from free_huge_page(). This is a layering violation of itself, but more importantly, if the kernel does a get_user_pages() on hugepages (which can happen from KVM amongst others), then the free_huge_page() can be delayed until after the associated inode has already been freed. If an unmount occurs at the wrong time, even the hugetlbfs superblock where the "quota" limits are stored may have been freed. Andrew Barry proposed a patch to fix this by having hugepages, instead of storing a pointer to their address_space and reaching the superblock from there, had the hugepages store pointers directly to the superblock, bumping the reference count as appropriate to avoid it being freed. Andrew Morton rejected that version, however, on the grounds that it made the existing layering violation worse. This is a reworked version of Andrew's patch, which removes the extra, and some of the existing, layering violation. It works by introducing the concept of a hugepage "subpool" at the lower hugepage mm layer - that is a finite logical pool of hugepages to allocate from. hugetlbfs now creates a subpool for each filesystem instance with a page limit set, and a pointer to the subpool gets added to each allocated hugepage, instead of the address_space pointer used now. The subpool has its own lifetime and is only freed once all pages in it _and_ all other references to it (i.e. superblocks) are gone. subpools are optional - a NULL subpool pointer is taken by the code to mean that no subpool limits are in effect. Previous discussion of this bug found in: "Fix refcounting in hugetlbfs quota handling.". See: https://lkml.org/lkml/2011/8/11/28 or http://marc.info/?l=linux-mm&m=126928970510627&w=1 v2: Fixed a bug spotted by Hillf Danton, and removed the extra parameter to alloc_huge_page() - since it already takes the vma, it is not necessary. Signed-off-by:Andrew Barry <abarry@cray.com> Signed-off-by:
David Gibson <david@gibson.dropbear.id.au> Cc: Hugh Dickins <hughd@google.com> Cc: Mel Gorman <mgorman@suse.de> Cc: Minchan Kim <minchan.kim@gmail.com> Cc: Hillf Danton <dhillf@gmail.com> Cc: Paul Mackerras <paulus@samba.org> Signed-off-by:
-
David Gibson authored
Make a couple of small cleanups to linux/include/hugetlb.h. The set_file_hugepages() function, which was not used anywhere is removed, and the hugetlbfs_config and hugetlbfs_inode_info structures with its HUGETLBFS_I helper function are moved into inode.c, the only place they were used. These structures are really linked to the hugetlbfs filesystem specifically not to hugepage mm handling in general, so they belong in the filesystem code not in a generally available header. It would be nice to move the hugetlbfs_sb_info (superblock) structure in there as well, but it's currently needed in a number of places via the hstate_vma() and hstate_inode(). Signed-off-by:
-
Bob Liu authored
There are multiple places which perform the same check. Add a new find_mergeable_vma() to handle this. Signed-off-by:
Bob Liu <lliubbo@gmail.com> Acked-by:
-
Mel Gorman authored
Commit c0ff7453 ("cpuset,mm: fix no node to alloc memory when changing cpuset's mems") wins a super prize for the largest number of memory barriers entered into fast paths for one commit. [get|put]_mems_allowed is incredibly heavy with pairs of full memory barriers inserted into a number of hot paths. This was detected while investigating at large page allocator slowdown introduced some time after 2.6.32. The largest portion of this overhead was shown by oprofile to be at an mfence introduced by this commit into the page allocator hot path. For extra style points, the commit introduced the use of yield() in an implementation of what looks like a spinning mutex. This patch replaces the full memory barriers on both read and write sides with a sequence counter with just read barriers on the fast path side. This is much cheaper on some architectures, including x86. The main bulk of the patch is the retry logic if the nodemask changes in a manner that can cause a false failure. While updating the nodemask, a check is made to see if a false failure is a risk. If it is, the sequence number gets bumped and parallel allocators will briefly stall while the nodemask update takes place. In a page fault test microbenchmark, oprofile samples from __alloc_pages_nodemask went from 4.53% of all samples to 1.15%. The actual results were 3.3.0-rc3 3.3.0-rc3 rc3-vanilla nobarrier-v2r1 Clients 1 UserTime 0.07 ( 0.00%) 0.08 (-14.19%) Clients 2 UserTime 0.07 ( 0.00%) 0.07 ( 2.72%) Clients 4 UserTime 0.08 ( 0.00%) 0.07 ( 3.29%) Clients 1 SysTime 0.70 ( 0.00%) 0.65 ( 6.65%) Clients 2 SysTime 0.85 ( 0.00%) 0.82 ( 3.65%) Clients 4 SysTime 1.41 ( 0.00%) 1.41 ( 0.32%) Clients 1 WallTime 0.77 ( 0.00%) 0.74 ( 4.19%) Clients 2 WallTime 0.47 ( 0.00%) 0.45 ( 3.73%) Clients 4 WallTime 0.38 ( 0.00%) 0.37 ( 1.58%) Clients 1 Flt/sec/cpu 497620.28 ( 0.00%) 520294.53 ( 4.56%) Clients 2 Flt/sec/cpu 414639.05 ( 0.00%) 429882.01 ( 3.68%) Clients 4 Flt/sec/cpu 257959.16 ( 0.00%) 258761.48 ( 0.31%) Clients 1 Flt/sec 495161.39 ( 0.00%) 517292.87 ( 4.47%) Clients 2 Flt/sec 820325.95 ( 0.00%) 850289.77 ( 3.65%) Clients 4 Flt/sec 1020068.93 ( 0.00%) 1022674.06 ( 0.26%) MMTests Statistics: duration Sys Time Running Test (seconds) 135.68 132.17 User+Sys Time Running Test (seconds) 164.2 160.13 Total Elapsed Time (seconds) 123.46 120.87 The overall improvement is small but the System CPU time is much improved and roughly in correlation to what oprofile reported (these performance figures are without profiling so skew is expected). The actual number of page faults is noticeably improved. For benchmarks like kernel builds, the overall benefit is marginal but the system CPU time is slightly reduced. To test the actual bug the commit fixed I opened two terminals. The first ran within a cpuset and continually ran a small program that faulted 100M of anonymous data. In a second window, the nodemask of the cpuset was continually randomised in a loop. Without the commit, the program would fail every so often (usually within 10 seconds) and obviously with the commit everything worked fine. With this patch applied, it also worked fine so the fix should be functionally equivalent. Signed-off-by:
Mel Gorman <mgorman@suse.de> Cc: Miao Xie <miaox@cn.fujitsu.com> Cc: David Rientjes <rientjes@google.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Christoph Lameter <cl@linux.com> Signed-off-by:
-
David Rientjes authored
The oom killer typically displays the allocation order at the time of oom as a part of its diangostic messages (for global, cpuset, and mempolicy ooms). The memory controller may also pass the charge order to the oom killer so it can emit the same information. This is useful in determining how large the memory allocation is that triggered the oom killer. Signed-off-by:
-
