- 11 Dec, 2014 40 commits
-
-
Johannes Weiner authored
Now that the external page_cgroup data structure and its lookup is gone, the only code remaining in there is swap slot accounting. Rename it and move the conditional compilation into mm/Makefile. Signed-off-by:
Johannes Weiner <hannes@cmpxchg.org> Acked-by:
Michal Hocko <mhocko@suse.cz> Acked-by:
Vladimir Davydov <vdavydov@parallels.com> Acked-by:
David S. Miller <davem@davemloft.net> Acked-by:
KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: "Kirill A. Shutemov" <kirill@shutemov.name> Cc: Tejun Heo <tj@kernel.org> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Signed-off-by:
Andrew Morton <akpm@linux-foundation.org> Signed-off-by:
Linus Torvalds <torvalds@linux-foundation.org>
-
Johannes Weiner authored
Memory cgroups used to have 5 per-page pointers. To allow users to disable that amount of overhead during runtime, those pointers were allocated in a separate array, with a translation layer between them and struct page. There is now only one page pointer remaining: the memcg pointer, that indicates which cgroup the page is associated with when charged. The complexity of runtime allocation and the runtime translation overhead is no longer justified to save that *potential* 0.19% of memory. With CONFIG_SLUB, page->mem_cgroup actually sits in the doubleword padding after the page->private member and doesn't even increase struct page, and then this patch actually saves space. Remaining users that care can still compile their kernels without CONFIG_MEMCG. text data bss dec hex filename 8828345 1725264 983040 11536649 b00909 vmlinux.old 8827425 1725264 966656 11519345 afc571 vmlinux.new [mhocko@suse.cz: update Documentation/cgroups/memory.txt] Signed-off-by:Konstantin Khlebnikov <koct9i@gmail.com> Signed-off-by:
-
Johannes Weiner authored
There is no cgroup-specific page lock anymore. Signed-off-by:
-
Li Haifeng authored
The largest index of swap device is MAX_SWAPFILES-1. So the type should be less than MAX_SWAPFILES. Signed-off-by:
Haifeng Li <omycle@gmail.com> Acked-by:
Konrad Rzeszutek Wilk <konrad.wilk@oracle.com> Signed-off-by:
-
Wei Yuan authored
Signed-off-by Wei Yuan <weiyuan.wei@huawei.com> Acked-by:
Rik van Riel <riel@redhat.com> Signed-off-by:
-
Hillf Danton authored
First, after flushing TLB, we have no need to scan pte from start again. Second, before bail out loop, the address is forwarded one step. Signed-off-by:
Hillf Danton <hillf.zj@alibaba-inc.com> Reviewed-by:
David Rientjes <rientjes@google.com> Signed-off-by:
-
Michal Hocko authored
Since commit d7365e78 ("mm: memcontrol: fix missed end-writeback page accounting") mem_cgroup_end_page_stat consumes locked and flags variables directly rather than via pointers which might trigger C undefined behavior as those variables are initialized only in the slow path of mem_cgroup_begin_page_stat. Although mem_cgroup_end_page_stat handles parameters correctly and touches them only when they hold a sensible value it is caller which loads a potentially uninitialized value which then might allow compiler to do crazy things. I haven't seen any warning from gcc and it seems that the current version (4.9) doesn't exploit this type undefined behavior but Sasha has reported the following: UBSan: Undefined behaviour in mm/rmap.c:1084:2 load of value 255 is not a valid value for type '_Bool' CPU: 4 PID: 8304 Comm: rngd Not tainted 3.18.0-rc2-next-20141029-sasha-00039-g77ed13d-dirty #1427 Call Trace: dump_stack (lib/dump_stack.c:52) ubsan_epilogue (lib/ubsan.c:159) __ubsan_handle_load_invalid_value (lib/ubsan.c:482) page_remove_rmap (mm/rmap.c:1084 mm/rmap.c:1096) unmap_page_range (./arch/x86/include/asm/atomic.h:27 include/linux/mm.h:463 mm/memory.c:1146 mm/memory.c:1258 mm/memory.c:1279 mm/memory.c:1303) unmap_single_vma (mm/memory.c:1348) unmap_vmas (mm/memory.c:1377 (discriminator 3)) exit_mmap (mm/mmap.c:2837) mmput (kernel/fork.c:659) do_exit (./arch/x86/include/asm/thread_info.h:168 kernel/exit.c:462 kernel/exit.c:747) do_group_exit (include/linux/sched.h:775 kernel/exit.c:873) SyS_exit_group (kernel/exit.c:901) tracesys_phase2 (arch/x86/kernel/entry_64.S:529) Fix this by using pointer parameters for both locked and flags and be more robust for future compiler changes even though the current code is implemented correctly. Signed-off-by:
Sasha Levin <sasha.levin@oracle.com> Acked-by:
-
Kirill A. Shutemov authored
As a small zero page, huge zero page should not be accounted in smaps report as normal page. For small pages we rely on vm_normal_page() to filter out zero page, but vm_normal_page() is not designed to handle pmds. We only get here due hackish cast pmd to pte in smaps_pte_range() -- pte and pmd format is not necessary compatible on each and every architecture. Let's add separate codepath to handle pmds. follow_trans_huge_pmd() will detect huge zero page for us. We would need pmd_dirty() helper to do this properly. The patch adds it to THP-enabled architectures which don't yet have one. [akpm@linux-foundation.org: use do_div to fix 32-bit build] Signed-off-by:
"Kirill A. Shutemov" <kirill@shutemov.name> Reported-by:
Fengguang Wu <fengguang.wu@intel.com> Tested-by:
Fengwei Yin <yfw.kernel@gmail.com> Signed-off-by:
-
Johannes Weiner authored
None of the mem_cgroup_same_or_subtree() callers actually require it to take the RCU lock, either because they hold it themselves or they have css references. Remove it. To make the API change clear, rename the leftover helper to mem_cgroup_is_descendant() to match cgroup_is_descendant(). Signed-off-by:
-
Johannes Weiner authored
The NULL in mm_match_cgroup() comes from a possibly exiting mm->owner. It makes a lot more sense to check where it's looked up, rather than check for it in __mem_cgroup_same_or_subtree() where it's unexpected. No other callsite passes NULL to __mem_cgroup_same_or_subtree(). Signed-off-by:
-
Johannes Weiner authored
That function acts like a typecast - unless NULL is passed in, no NULL can come out. task_in_mem_cgroup() callers don't pass NULL tasks. Signed-off-by:
-
Johannes Weiner authored
While moving charges from one memcg to another, page stat updates must acquire the old memcg's move_lock to prevent double accounting. That situation is denoted by an increased memcg->move_accounting. However, the charge moving code declares this way too early for now, even before summing up the RSS and pre-allocating destination charges. Shorten this slowpath mode by increasing memcg->move_accounting only right before walking the task's address space with the intention of actually moving the pages. Signed-off-by:
-
Kirill A. Shutemov authored
Zero pages can be used only in anonymous mappings, which never have writable vma->vm_page_prot: see protection_map in mm/mmap.c and __PX1X definitions. Let's drop redundant pmd_wrprotect() in set_huge_zero_page(). Signed-off-by:
-
Vladimir Davydov authored
Let's use generic slab_start/next/stop for showing memcg caches info. In contrast to the current implementation, this will work even if all memcg caches' info doesn't fit into a seq buffer (a page), plus it simply looks neater. Actually, the main reason I do this isn't mere cleanup. I'm going to zap the memcg_slab_caches list, because I find it useless provided we have the slab_caches list, and this patch is a step in this direction. It should be noted that before this patch an attempt to read memory.kmem.slabinfo of a cgroup that doesn't have kmem limit set resulted in -EIO, while after this patch it will silently show nothing except the header, but I don't think it will frustrate anyone. Signed-off-by:
-
Vladimir Davydov authored
mem_cgroup_reclaimable() checks whether a cgroup has reclaimable pages on *any* NUMA node. However, the only place where it's called is mem_cgroup_soft_reclaim(), which tries to reclaim memory from a *specific* zone. So the way it is used is incorrect - it will return true even if the cgroup doesn't have pages on the zone we're scanning. I think we can get rid of this check completely, because mem_cgroup_shrink_node_zone(), which is called by mem_cgroup_soft_reclaim() if mem_cgroup_reclaimable() returns true, is equivalent to shrink_lruvec(), which exits almost immediately if the lruvec passed to it is empty. So there's no need to optimize anything here. Besides, we don't have such a check in the general scan path (shrink_zone) either. Signed-off-by:
-
Sasha Levin authored
hstate_sizelog() would shift left an int rather than long, triggering undefined behaviour and passing an incorrect value when the requested page size was more than 4GB, thus breaking >4GB pages. Signed-off-by:
-
Johannes Weiner authored
Having these functions and their documentation split out and somewhere makes it harder, not easier, to follow what's going on. Inline them directly where charge moving is prepared and finished, and put an explanation right next to it. Signed-off-by:
-
Johannes Weiner authored
mem_cgroup_end_move() checks if the passed memcg is NULL, along with a lengthy comment to explain why this seemingly non-sensical situation is even possible. Check in cancel_attach() itself whether can_attach() set up the move context or not, it's a lot more obvious from there. Then remove the check and comment in mem_cgroup_end_move(). Signed-off-by:
-
Johannes Weiner authored
The wrappers around taking and dropping the memcg->move_lock spinlock add nothing of value. Inline the spinlock calls into the callsites. Signed-off-by:
-
Johannes Weiner authored
pc->mem_cgroup had to be left intact after uncharge for the final LRU removal, and !PCG_USED indicated whether the page was uncharged. But since commit 0a31bc97 ("mm: memcontrol: rewrite uncharge API") pages are uncharged after the final LRU removal. Uncharge can simply clear the pointer and the PCG_USED/PageCgroupUsed sites can test that instead. Because this is the last page_cgroup flag, this patch reduces the memcg per-page overhead to a single pointer. [akpm@linux-foundation.org: remove unneeded initialization of `memcg', per Michal] Signed-off-by:
-
Johannes Weiner authored
PCG_MEM is a remnant from an earlier version of 0a31bc97 ("mm: memcontrol: rewrite uncharge API"), used to tell whether migration cleared a charge while leaving pc->mem_cgroup valid and PCG_USED set. But in the final version, mem_cgroup_migrate() directly uncharges the source page, rendering this distinction unnecessary. Remove it. Signed-off-by:
-
Johannes Weiner authored
Now that mem_cgroup_swapout() fully uncharges the page, every page that is still in use when reaching mem_cgroup_uncharge() is known to carry both the memory and the memory+swap charge. Simplify the uncharge path and remove the PCG_MEMSW page flag accordingly. Signed-off-by:
-
Johannes Weiner authored
This series gets rid of the remaining page_cgroup flags, thus cutting the memcg per-page overhead down to one pointer. This patch (of 4): mem_cgroup_swapout() is called with exclusive access to the page at the end of the page's lifetime. Instead of clearing the PCG_MEMSW flag and deferring the uncharge, just do it right away. This allows follow-up patches to simplify the uncharge code. Signed-off-by:
-
Michal Hocko authored
Don't call lookup_page_cgroup() when memcg is disabled. Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Vladimir Davydov <vdavydov@parallels.com> Signed-off-by:
-
Vladimir Davydov authored
The activate_kmem_mutex is used to serialize memcg.kmem.limit updates, but we already serialize them with memcg_limit_mutex so let's remove the former. Signed-off-by:
-
Johannes Weiner authored
Better explain re-entrant migration when compaction races with reclaim, and also mention swapcache readahead pages as possible uncharged migration sources. Signed-off-by:
-
Johannes Weiner authored
Commit 7512102c ("memcg: fix GPF when cgroup removal races with last exit") added a pc->mem_cgroup reset into mem_cgroup_page_lruvec() to prevent a crash where an anon page gets uncharged on unmap, the memcg is released, and then the final LRU isolation on free dereferences the stale pc->mem_cgroup pointer. But since commit 0a31bc97 ("mm: memcontrol: rewrite uncharge API"), pages are only uncharged AFTER that final LRU isolation, which guarantees the memcg's lifetime until then. pc->mem_cgroup now only needs to be reset for swapcache readahead pages. Update the comment and callsite requirements accordingly. Signed-off-by:
-
Vladimir Davydov authored
If we fail to reclaim anything from a cgroup during a soft reclaim pass we want to get the next largest cgroup exceeding its soft limit. To achieve this, we should obviously remove the current group from the tree and then pick the largest group. Currently we have a weird loop instead. Let's simplify it. Signed-off-by:
-
Aneesh Kumar K.V authored
Add the default enable config option after the NUMA_BALANCING option so that it appears related in the nconfig interface. Signed-off-by:
Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com> Acked-by:
-
Vlastimil Babka authored
The goal of memory compaction is to create high-order freepages through page migration. Page migration however puts pages on the per-cpu lru_add cache, which is later flushed to per-cpu pcplists, and only after pcplists are drained the pages can actually merge. This can happen due to the per-cpu caches becoming full through further freeing, or explicitly. During direct compaction, it is useful to do the draining explicitly so that pages merge as soon as possible and compaction can detect success immediately and keep the latency impact at minimum. However the current implementation is far from ideal. Draining is done only in __alloc_pages_direct_compact(), after all zones were already compacted, and the decisions to continue or stop compaction in individual zones was done without the last batch of migrations being merged. It is also missing the draining of lru_add cache before the pcplists. This patch moves the draining for direct compaction into compact_zone(). It adds the missing lru_cache draining and uses the newly introduced single zone pcplists draining to reduce overhead and avoid impact on unrelated zones. Draining is only performed when it can actually lead to merging of a page of desired order (passed by cc->order). This means it is only done when migration occurred in the previously scanned cc->order aligned block(s) and the migration scanner is now pointing to the next cc->order aligned block. The patch has been tested with stress-highalloc benchmark from mmtests. Although overal allocation success rates of the benchmark were not affected, the number of detected compaction successes has doubled. This suggests that allocations were previously successful due to implicit merging caused by background activity, making a later allocation attempt succeed immediately, but not attributing the success to compaction. Since stress-highalloc always tries to allocate almost the whole memory, it cannot show the improvement in its reported success rate metric. However after this patch, compaction should detect success and terminate earlier, reducing the direct compaction latencies in a real scenario. Signed-off-by:
Vlastimil Babka <vbabka@suse.cz> Cc: Minchan Kim <minchan@kernel.org> Cc: Mel Gorman <mgorman@suse.de> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Michal Nazarewicz <mina86@mina86.com> Cc: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com> Cc: Christoph Lameter <cl@linux.com> Acked-by:
-
Vlastimil Babka authored
Compaction caches the migration and free scanner positions between compaction invocations, so that the whole zone gets eventually scanned and there is no bias towards the initial scanner positions at the beginning/end of the zone. The cached positions are continuously updated as scanners progress and the updating stops as soon as a page is successfully isolated. The reasoning behind this is that a pageblock where isolation succeeded is likely to succeed again in near future and it should be worth revisiting it. However, the downside is that potentially many pages are rescanned without successful isolation. At worst, there might be a page where isolation from LRU succeeds but migration fails (potentially always). So upon encountering this page, cached position would always stop being updated for no good reason. It might have been useful to let such page be rescanned with sync compaction after async one failed, but this is now handled by caching scanner position for async and sync mode separately since commit 35979ef3 ("mm, compaction: add per-zone migration pfn cache for async compaction"). After this patch, cached positions are updated unconditionally. In stress-highalloc benchmark, this has decreased the numbers of scanned pages by few percent, without affecting allocation success rates. To prevent free scanner from leaving free pages behind after they are returned due to page migration failure, the cached scanner pfn is changed to point to the pageblock of the returned free page with the highest pfn, before leaving compact_zone(). [akpm@linux-foundation.org: coding-style fixes] Signed-off-by:
-
Vlastimil Babka authored
Deferred compaction is employed to avoid compacting zone where sync direct compaction has recently failed. As such, it makes sense to only defer when a full zone was scanned, which is when compact_zone returns with COMPACT_COMPLETE. It's less useful to defer when compact_zone returns with apparent success (COMPACT_PARTIAL), followed by a watermark check failure, which can happen due to parallel allocation activity. It also does not make much sense to defer compaction which was completely skipped (COMPACT_SKIP) for being unsuitable in the first place. This patch therefore makes deferred compaction trigger only when COMPACT_COMPLETE is returned from compact_zone(). Results of stress-highalloc becnmark show the difference is within measurement error, so the issue is rather cosmetic. Signed-off-by:
-
Vlastimil Babka authored
Since commit 53853e2d ("mm, compaction: defer each zone individually instead of preferred zone"), compaction is deferred for each zone where sync direct compaction fails, and reset where it succeeds. However, it was observed that for DMA zone compaction often appeared to succeed while subsequent allocation attempt would not, due to different outcome of watermark check. In order to properly defer compaction in this zone, the candidate zone has to be passed back to __alloc_pages_direct_compact() and compaction deferred in the zone after the allocation attempt fails. The large source of mismatch between watermark check in compaction and allocation was the lack of alloc_flags and classzone_idx values in compaction, which has been fixed in the previous patch. So with this problem fixed, we can simplify the code by removing the candidate_zone parameter and deferring in __alloc_pages_direct_compact(). After this patch, the compaction activity during stress-highalloc benchmark is still somewhat increased, but it's negligible compared to the increase that occurred without the better watermark checking. This suggests that it is still possible to apparently succeed in compaction but fail to allocate, possibly due to parallel allocation activity. [akpm@linux-foundation.org: fix build] Suggested-by:
Joonsoo Kim <iamjoonsoo.kim@lge.com> Signed-off-by:
-
Vlastimil Babka authored
Compaction relies on zone watermark checks for decisions such as if it's worth to start compacting in compaction_suitable() or whether compaction should stop in compact_finished(). The watermark checks take classzone_idx and alloc_flags parameters, which are related to the memory allocation request. But from the context of compaction they are currently passed as 0, including the direct compaction which is invoked to satisfy the allocation request, and could therefore know the proper values. The lack of proper values can lead to mismatch between decisions taken during compaction and decisions related to the allocation request. Lack of proper classzone_idx value means that lowmem_reserve is not taken into account. This has manifested (during recent changes to deferred compaction) when DMA zone was used as fallback for preferred Normal zone. compaction_suitable() without proper classzone_idx would think that the watermarks are already satisfied, but watermark check in get_page_from_freelist() would fail. Because of this problem, deferring compaction has extra complexity that can be removed in the following patch. The issue (not confirmed in practice) with missing alloc_flags is opposite in nature. For allocations that include ALLOC_HIGH, ALLOC_HIGHER or ALLOC_CMA in alloc_flags (the last includes all MOVABLE allocations on CMA-enabled systems) the watermark checking in compaction with 0 passed will be stricter than in get_page_from_freelist(). In these cases compaction might be running for a longer time than is really needed. Another issue compaction_suitable() is that the check for "does the zone need compaction at all?" comes only after the check "does the zone have enough free free pages to succeed compaction". The latter considers extra pages for migration and can therefore in some situations fail and return COMPACT_SKIPPED, although the high-order allocation would succeed and we should return COMPACT_PARTIAL. This patch fixes these problems by adding alloc_flags and classzone_idx to struct compact_control and related functions involved in direct compaction and watermark checking. Where possible, all other callers of compaction_suitable() pass proper values where those are known. This is currently limited to classzone_idx, which is sometimes known in kswapd context. However, the direct reclaim callers should_continue_reclaim() and compaction_ready() do not currently know the proper values, so the coordination between reclaim and compaction may still not be as accurate as it could. This can be fixed later, if it's shown to be an issue. Additionaly the checks in compact_suitable() are reordered to address the second issue described above. The effect of this patch should be slightly better high-order allocation success rates and/or less compaction overhead, depending on the type of allocations and presence of CMA. It allows simplifying deferred compaction code in a followup patch. When testing with stress-highalloc, there was some slight improvement (which might be just due to variance) in success rates of non-THP-like allocations. Signed-off-by:
-
Jamie Liu authored
shrink_page_list() counts all pages with a mapping, including clean pages, toward nr_congested if they're on a write-congested BDI. shrink_inactive_list() then sets ZONE_CONGESTED if nr_dirty == nr_congested. Fix this apples-to-oranges comparison by only counting pages for nr_congested if they count for nr_dirty. Signed-off-by:
Jamie Liu <jamieliu@google.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Mel Gorman <mgorman@suse.de> Cc: Greg Thelen <gthelen@google.com> Cc: Hugh Dickins <hughd@google.com> Signed-off-by:
-
Yu Zhao authored
This allows us to catch the bug fixed in the previous patch (mm: free compound page with correct order). Here we also verify whether a page is tail page or not -- tail pages are supposed to be freed along with their head, not by themselves. Signed-off-by:
Yu Zhao <yuzhao@google.com> Reviewed-by:
-
Akinobu Mita authored
This makes CMA memory area size zero for x86 in default configuration (doesn't change on the other architectures). If default CMA size is zero, DMA_CMA is disabled. It can be enabled by passing cma= to the kernel. This makes less impact on x86. Because there is no mainline driver that requires it for x86, and Peter Hurley reported the performance regression, as this is trying to drive _all_ dma mapping allocations through a _very_ small window. Signed-off-by:
Akinobu Mita <akinobu.mita@gmail.com> Reported-by:
Peter Hurley <peter@hurleysoftware.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Chuck Ebbert <cebbert.lkml@gmail.com> Cc: Jean Delvare <jdelvare@suse.de> Acked-by:
Marek Szyprowski <m.szyprowski@samsung.com> Cc: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com> Cc: David Woodhouse <dwmw2@infradead.org> Cc: Don Dutile <ddutile@redhat.com> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Ingo Molnar <mingo@redhat.com> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: Andi Kleen <andi@firstfloor.org> Cc: Yinghai Lu <yinghai@kernel.org> Signed-off-by:
-
Vlastimil Babka authored
Memory hotplug and failure mechanisms have several places where pcplists are drained so that pages are returned to the buddy allocator and can be e.g. prepared for offlining. This is always done in the context of a single zone, we can reduce the pcplists drain to the single zone, which is now possible. The change should make memory offlining due to hotremove or failure faster and not disturbing unrelated pcplists anymore. Signed-off-by:
-
Vlastimil Babka authored
CMA allocation drains pcplists so that pages can merge back to buddy allocator. Since it operates on a single zone, we can reduce the pcplists drain to the single zone, which is now possible. The change should make CMA allocations faster and not disturbing unrelated pcplists anymore. Signed-off-by:
-
Vlastimil Babka authored
When setting MIGRATETYPE_ISOLATE on a pageblock, pcplists are drained to have a better chance that all pages will be successfully isolated and not left in the per-cpu caches. Since isolation is always concerned with a single zone, we can reduce the pcplists drain to the single zone, which is now possible. The change should make memory isolation faster and not disturbing unrelated pcplists anymore. Signed-off-by:
-
