08 Oct, 2016
2 commits
-
Several people have reported premature OOMs for order-2 allocations
(stack) due to OOM rework in 4.7. In the scenario (parallel kernel
build and dd writing to two drives) many pageblocks get marked as
Unmovable and compaction free scanner struggles to isolate free pages.
Joonsoo Kim pointed out that the free scanner skips pageblocks that are
not movable to prevent filling them and forcing non-movable allocations
to fallback to other pageblocks. Such heuristic makes sense to help
prevent long-term fragmentation, but premature OOMs are relatively more
urgent problem. As a compromise, this patch disables the heuristic only
for the ultimate compaction priority.Link: http://lkml.kernel.org/r/20160906135258.18335-5-vbabka@suse.cz
Reported-by: Ralf-Peter Rohbeck
Reported-by: Arkadiusz Miskiewicz
Reported-by: Olaf Hering
Suggested-by: Joonsoo Kim
Signed-off-by: Vlastimil Babka
Acked-by: Michal Hocko
Cc: Michal Hocko
Cc: Mel Gorman
Cc: Joonsoo Kim
Cc: David Rientjes
Cc: Rik van Riel
Cc: Tetsuo Handa
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds -
Patch series "make direct compaction more deterministic")
This is mostly a followup to Michal's oom detection rework, which
highlighted the need for direct compaction to provide better feedback in
reclaim/compaction loop, so that it can reliably recognize when
compaction cannot make further progress, and allocation should invoke
OOM killer or fail. We've discussed this at LSF/MM [1] where I proposed
expanding the async/sync migration mode used in compaction to more
general "priorities". This patchset adds one new priority that just
overrides all the heuristics and makes compaction fully scan all zones.
I don't currently think that we need more fine-grained priorities, but
we'll see. Other than that there's some smaller fixes and cleanups,
mainly related to the THP-specific hacks.I've tested this with stress-highalloc in GFP_KERNEL order-4 and
THP-like order-9 scenarios. There's some improvement for compaction
stats for the order-4, which is likely due to the better watermarks
handling. In the previous version I reported mostly noise wrt
compaction stats, and decreased direct reclaim - now the reclaim is
without difference. I believe this is due to the less aggressive
compaction priority increase in patch 6."before" is a mmotm tree prior to 4.7 release plus the first part of the
series that was sent and merged separatelybefore after
order-4:Compaction stalls 27216 30759
Compaction success 19598 25475
Compaction failures 7617 5283
Page migrate success 370510 464919
Page migrate failure 25712 27987
Compaction pages isolated 849601 1041581
Compaction migrate scanned 143146541 101084990
Compaction free scanned 208355124 144863510
Compaction cost 1403 1210order-9:
Compaction stalls 7311 7401
Compaction success 1634 1683
Compaction failures 5677 5718
Page migrate success 194657 183988
Page migrate failure 4753 4170
Compaction pages isolated 498790 456130
Compaction migrate scanned 565371 524174
Compaction free scanned 4230296 4250744
Compaction cost 215 203[1] https://lwn.net/Articles/684611/
This patch (of 11):
A recent patch has added whole_zone flag that compaction sets when
scanning starts from the zone boundary, in order to report that zone has
been fully scanned in one attempt. For allocations that want to try
really hard or cannot fail, we will want to introduce a mode where
scanning whole zone is guaranteed regardless of the cached positions.This patch reuses the whole_zone flag in a way that if it's already
passed true to compaction, the cached scanner positions are ignored.
Employing this flag during reclaim/compaction loop will be done in the
next patch. This patch however converts compaction invoked from
userspace via procfs to use this flag. Before this patch, the cached
positions were first reset to zone boundaries and then read back from
struct zone, so there was a window where a parallel compaction could
replace the reset values, making the manual compaction less effective.
Using the flag instead of performing reset is more robust.[akpm@linux-foundation.org: coding-style fixes]
Link: http://lkml.kernel.org/r/20160810091226.6709-2-vbabka@suse.cz
Signed-off-by: Vlastimil Babka
Tested-by: Lorenzo Stoakes
Acked-by: Michal Hocko
Cc: Mel Gorman
Cc: Joonsoo Kim
Cc: David Rientjes
Cc: Rik van Riel
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds
29 Jul, 2016
4 commits
-
Async compaction detects contention either due to failing trylock on
zone->lock or lru_lock, or by need_resched(). Since 1f9efdef4f3f ("mm,
compaction: khugepaged should not give up due to need_resched()") the
code got quite complicated to distinguish these two up to the
__alloc_pages_slowpath() level, so different decisions could be taken
for khugepaged allocations.After the recent changes, khugepaged allocations don't check for
contended compaction anymore, so we again don't need to distinguish lock
and sched contention, and simplify the current convoluted code a lot.However, I believe it's also possible to simplify even more and
completely remove the check for contended compaction after the initial
async compaction for costly orders, which was originally aimed at THP
page fault allocations. There are several reasons why this can be done
now:- with the new defaults, THP page faults no longer do reclaim/compaction at
all, unless the system admin has overridden the default, or application has
indicated via madvise that it can benefit from THP's. In both cases, it
means that the potential extra latency is expected and worth the benefits.
- even if reclaim/compaction proceeds after this patch where it previously
wouldn't, the second compaction attempt is still async and will detect the
contention and back off, if the contention persists
- there are still heuristics like deferred compaction and pageblock skip bits
in place that prevent excessive THP page fault latenciesLink: http://lkml.kernel.org/r/20160721073614.24395-9-vbabka@suse.cz
Signed-off-by: Vlastimil Babka
Acked-by: Michal Hocko
Acked-by: Mel Gorman
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds -
The fair zone allocation policy interleaves allocation requests between
zones to avoid an age inversion problem whereby new pages are reclaimed
to balance a zone. Reclaim is now node-based so this should no longer
be an issue and the fair zone allocation policy is not free. This patch
removes it.Link: http://lkml.kernel.org/r/1467970510-21195-30-git-send-email-mgorman@techsingularity.net
Signed-off-by: Mel Gorman
Acked-by: Vlastimil Babka
Cc: Hillf Danton
Acked-by: Johannes Weiner
Cc: Joonsoo Kim
Cc: Michal Hocko
Cc: Minchan Kim
Cc: Rik van Riel
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds -
As reclaim is now per-node based, convert zone_reclaim to be
node_reclaim. It is possible that a node will be reclaimed multiple
times if it has multiple zones but this is unavoidable without caching
all nodes traversed so far. The documentation and interface to
userspace is the same from a configuration perspective and will will be
similar in behaviour unless the node-local allocation requests were also
limited to lower zones.Link: http://lkml.kernel.org/r/1467970510-21195-24-git-send-email-mgorman@techsingularity.net
Signed-off-by: Mel Gorman
Acked-by: Vlastimil Babka
Cc: Hillf Danton
Acked-by: Johannes Weiner
Cc: Joonsoo Kim
Cc: Michal Hocko
Cc: Minchan Kim
Cc: Rik van Riel
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds -
This moves the LRU lists from the zone to the node and related data such
as counters, tracing, congestion tracking and writeback tracking.Unfortunately, due to reclaim and compaction retry logic, it is
necessary to account for the number of LRU pages on both zone and node
logic. Most reclaim logic is based on the node counters but the retry
logic uses the zone counters which do not distinguish inactive and
active sizes. It would be possible to leave the LRU counters on a
per-zone basis but it's a heavier calculation across multiple cache
lines that is much more frequent than the retry checks.Other than the LRU counters, this is mostly a mechanical patch but note
that it introduces a number of anomalies. For example, the scans are
per-zone but using per-node counters. We also mark a node as congested
when a zone is congested. This causes weird problems that are fixed
later but is easier to review.In the event that there is excessive overhead on 32-bit systems due to
the nodes being on LRU then there are two potential solutions1. Long-term isolation of highmem pages when reclaim is lowmem
When pages are skipped, they are immediately added back onto the LRU
list. If lowmem reclaim persisted for long periods of time, the same
highmem pages get continually scanned. The idea would be that lowmem
keeps those pages on a separate list until a reclaim for highmem pages
arrives that splices the highmem pages back onto the LRU. It potentially
could be implemented similar to the UNEVICTABLE list.That would reduce the skip rate with the potential corner case is that
highmem pages have to be scanned and reclaimed to free lowmem slab pages.2. Linear scan lowmem pages if the initial LRU shrink fails
This will break LRU ordering but may be preferable and faster during
memory pressure than skipping LRU pages.Link: http://lkml.kernel.org/r/1467970510-21195-4-git-send-email-mgorman@techsingularity.net
Signed-off-by: Mel Gorman
Acked-by: Johannes Weiner
Acked-by: Vlastimil Babka
Cc: Hillf Danton
Cc: Joonsoo Kim
Cc: Michal Hocko
Cc: Minchan Kim
Cc: Rik van Riel
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds
27 Jul, 2016
3 commits
-
The idea borrowed from Peter's patch from patchset on speculative page
faults[1]:Instead of passing around the endless list of function arguments,
replace the lot with a single structure so we can change context without
endless function signature changes.The changes are mostly mechanical with exception of faultaround code:
filemap_map_pages() got reworked a bit.This patch is preparation for the next one.
[1] http://lkml.kernel.org/r/20141020222841.302891540@infradead.org
Link: http://lkml.kernel.org/r/1466021202-61880-9-git-send-email-kirill.shutemov@linux.intel.com
Signed-off-by: Kirill A. Shutemov
Acked-by: Peter Zijlstra (Intel)
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds -
This patch makes swapin readahead to improve thp collapse rate. When
khugepaged scanned pages, there can be a few of the pages in swap area.With the patch THP can collapse 4kB pages into a THP when there are up
to max_ptes_swap swap ptes in a 2MB range.The patch was tested with a test program that allocates 400B of memory,
writes to it, and then sleeps. I force the system to swap out all.
Afterwards, the test program touches the area by writing, it skips a
page in each 20 pages of the area.Without the patch, system did not swap in readahead. THP rate was %65
of the program of the memory, it did not change over time.With this patch, after 10 minutes of waiting khugepaged had collapsed
%99 of the program's memory.[kirill.shutemov@linux.intel.com: trivial cleanup of exit path of the function]
[kirill.shutemov@linux.intel.com: __collapse_huge_page_swapin(): drop unused 'pte' parameter]
[kirill.shutemov@linux.intel.com: do not hold anon_vma lock during swap in]
Signed-off-by: Ebru Akagunduz
Acked-by: Rik van Riel
Cc: Naoya Horiguchi
Cc: Andrea Arcangeli
Cc: Joonsoo Kim
Cc: Xie XiuQi
Cc: Cyrill Gorcunov
Cc: Mel Gorman
Cc: David Rientjes
Cc: Vlastimil Babka
Cc: Aneesh Kumar K.V
Cc: Hugh Dickins
Cc: Johannes Weiner
Cc: Michal Hocko
Signed-off-by: Kirill A. Shutemov
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds -
This patch is motivated from Hugh and Vlastimil's concern [1].
There are two ways to get freepage from the allocator. One is using
normal memory allocation API and the other is __isolate_free_page()
which is internally used for compaction and pageblock isolation. Later
usage is rather tricky since it doesn't do whole post allocation
processing done by normal API.One problematic thing I already know is that poisoned page would not be
checked if it is allocated by __isolate_free_page(). Perhaps, there
would be more.We could add more debug logic for allocated page in the future and this
separation would cause more problem. I'd like to fix this situation at
this time. Solution is simple. This patch commonize some logic for
newly allocated page and uses it on all sites. This will solve the
problem.[1] http://marc.info/?i=alpine.LSU.2.11.1604270029350.7066%40eggly.anvils%3E
[iamjoonsoo.kim@lge.com: mm-page_alloc-introduce-post-allocation-processing-on-page-allocator-v3]
Link: http://lkml.kernel.org/r/1464230275-25791-7-git-send-email-iamjoonsoo.kim@lge.com
Link: http://lkml.kernel.org/r/1466150259-27727-9-git-send-email-iamjoonsoo.kim@lge.com
Link: http://lkml.kernel.org/r/1464230275-25791-7-git-send-email-iamjoonsoo.kim@lge.com
Signed-off-by: Joonsoo Kim
Acked-by: Vlastimil Babka
Cc: Mel Gorman
Cc: Minchan Kim
Cc: Alexander Potapenko
Cc: Hugh Dickins
Cc: Michal Hocko
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds
25 Jun, 2016
1 commit
-
Commit d0164adc89f6 ("mm, page_alloc: distinguish between being unable
to sleep, unwilling to sleep and avoiding waking kswapd") modified
__GFP_WAIT to explicitly identify the difference between atomic callers
and those that were unwilling to sleep. Later the definition was
removed entirely.The GFP_RECLAIM_MASK is the set of flags that affect watermark checking
and reclaim behaviour but __GFP_ATOMIC was never added. Without it,
atomic users of the slab allocator strip the __GFP_ATOMIC flag and
cannot access the page allocator atomic reserves. This patch addresses
the problem.The user-visible impact depends on the workload but potentially atomic
allocations unnecessarily fail without this path.Link: http://lkml.kernel.org/r/20160610093832.GK2527@techsingularity.net
Signed-off-by: Mel Gorman
Reported-by: Marcin Wojtas
Acked-by: Vlastimil Babka
Acked-by: Michal Hocko
Cc: [4.4+]
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds
24 May, 2016
2 commits
-
All the callers of vm_mmap seem to check for the failure already and
bail out in one way or another on the error which means that we can
change it to use killable version of vm_mmap_pgoff and return -EINTR if
the current task gets killed while waiting for mmap_sem. This also
means that vm_mmap_pgoff can be killable by default and drop the
additional parameter.This will help in the OOM conditions when the oom victim might be stuck
waiting for the mmap_sem for write which in turn can block oom_reaper
which relies on the mmap_sem for read to make a forward progress and
reclaim the address space of the victim.Please note that load_elf_binary is ignoring vm_mmap error for
current->personality & MMAP_PAGE_ZERO case but that shouldn't be a
problem because the address is not used anywhere and we never return to
the userspace if we got killed.Signed-off-by: Michal Hocko
Acked-by: Vlastimil Babka
Cc: "Kirill A. Shutemov"
Cc: Mel Gorman
Cc: Oleg Nesterov
Cc: Andrea Arcangeli
Cc: Al Viro
Cc: Johannes Weiner
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds -
This is a follow up work for oom_reaper [1]. As the async OOM killing
depends on oom_sem for read we would really appreciate if a holder for
write didn't stood in the way. This patchset is changing many of
down_write calls to be killable to help those cases when the writer is
blocked and waiting for readers to release the lock and so help
__oom_reap_task to process the oom victim.Most of the patches are really trivial because the lock is help from a
shallow syscall paths where we can return EINTR trivially and allow the
current task to die (note that EINTR will never get to the userspace as
the task has fatal signal pending). Others seem to be easy as well as
the callers are already handling fatal errors and bail and return to
userspace which should be sufficient to handle the failure gracefully.
I am not familiar with all those code paths so a deeper review is really
appreciated.As this work is touching more areas which are not directly connected I
have tried to keep the CC list as small as possible and people who I
believed would be familiar are CCed only to the specific patches (all
should have received the cover though).This patchset is based on linux-next and it depends on
down_write_killable for rw_semaphores which got merged into tip
locking/rwsem branch and it is merged into this next tree. I guess it
would be easiest to route these patches via mmotm because of the
dependency on the tip tree but if respective maintainers prefer other
way I have no objections.I haven't covered all the mmap_write(mm->mmap_sem) instances here
$ git grep "down_write(.*\)" next/master | wc -l
98
$ git grep "down_write(.*\)" | wc -l
62I have tried to cover those which should be relatively easy to review in
this series because this alone should be a nice improvement. Other
places can be changed on top.[0] http://lkml.kernel.org/r/1456752417-9626-1-git-send-email-mhocko@kernel.org
[1] http://lkml.kernel.org/r/1452094975-551-1-git-send-email-mhocko@kernel.org
[2] http://lkml.kernel.org/r/1456750705-7141-1-git-send-email-mhocko@kernel.orgThis patch (of 18):
This is the first step in making mmap_sem write waiters killable. It
focuses on the trivial ones which are taking the lock early after
entering the syscall and they are not changing state before.Therefore it is very easy to change them to use down_write_killable and
immediately return with -EINTR. This will allow the waiter to pass away
without blocking the mmap_sem which might be required to make a forward
progress. E.g. the oom reaper will need the lock for reading to
dismantle the OOM victim address space.The only tricky function in this patch is vm_mmap_pgoff which has many
call sites via vm_mmap. To reduce the risk keep vm_mmap with the
original non-killable semantic for now.vm_munmap callers do not bother checking the return value so open code
it into the munmap syscall path for now for simplicity.Signed-off-by: Michal Hocko
Acked-by: Vlastimil Babka
Cc: Mel Gorman
Cc: "Kirill A. Shutemov"
Cc: Konstantin Khlebnikov
Cc: Hugh Dickins
Cc: Andrea Arcangeli
Cc: David Rientjes
Cc: Dave Hansen
Cc: Johannes Weiner
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds
21 May, 2016
1 commit
-
COMPACT_COMPLETE now means that compaction and free scanner met. This
is not very useful information if somebody just wants to use this
feedback and make any decisions based on that. The current caller might
be a poor guy who just happened to scan tiny portion of the zone and
that could be the reason no suitable pages were compacted. Make sure we
distinguish the full and partial zone walks.Consumers should treat COMPACT_PARTIAL_SKIPPED as a potential success
and be optimistic in retrying.The existing users of COMPACT_COMPLETE are conservatively changed to use
COMPACT_PARTIAL_SKIPPED as well but some of them should be probably
reconsidered and only defer the compaction only for COMPACT_COMPLETE
with the new semantic.This patch shouldn't introduce any functional changes.
Signed-off-by: Michal Hocko
Acked-by: Vlastimil Babka
Acked-by: Hillf Danton
Cc: David Rientjes
Cc: Johannes Weiner
Cc: Joonsoo Kim
Cc: Mel Gorman
Cc: Tetsuo Handa
Cc: Vladimir Davydov
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds
20 May, 2016
4 commits
-
The classzone_idx can be inferred from preferred_zoneref so remove the
unnecessary field and save stack space.Signed-off-by: Mel Gorman
Cc: Vlastimil Babka
Cc: Jesper Dangaard Brouer
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds -
The allocator fast path looks up the first usable zone in a zonelist and
then get_page_from_freelist does the same job in the zonelist iterator.
This patch preserves the necessary information.4.6.0-rc2 4.6.0-rc2
fastmark-v1r20 initonce-v1r20
Min alloc-odr0-1 364.00 ( 0.00%) 359.00 ( 1.37%)
Min alloc-odr0-2 262.00 ( 0.00%) 260.00 ( 0.76%)
Min alloc-odr0-4 214.00 ( 0.00%) 214.00 ( 0.00%)
Min alloc-odr0-8 186.00 ( 0.00%) 186.00 ( 0.00%)
Min alloc-odr0-16 173.00 ( 0.00%) 173.00 ( 0.00%)
Min alloc-odr0-32 165.00 ( 0.00%) 165.00 ( 0.00%)
Min alloc-odr0-64 161.00 ( 0.00%) 162.00 ( -0.62%)
Min alloc-odr0-128 159.00 ( 0.00%) 161.00 ( -1.26%)
Min alloc-odr0-256 168.00 ( 0.00%) 170.00 ( -1.19%)
Min alloc-odr0-512 180.00 ( 0.00%) 181.00 ( -0.56%)
Min alloc-odr0-1024 190.00 ( 0.00%) 190.00 ( 0.00%)
Min alloc-odr0-2048 196.00 ( 0.00%) 196.00 ( 0.00%)
Min alloc-odr0-4096 202.00 ( 0.00%) 202.00 ( 0.00%)
Min alloc-odr0-8192 206.00 ( 0.00%) 205.00 ( 0.49%)
Min alloc-odr0-16384 206.00 ( 0.00%) 205.00 ( 0.49%)The benefit is negligible and the results are within the noise but each
cycle counts.Signed-off-by: Mel Gorman
Cc: Vlastimil Babka
Cc: Jesper Dangaard Brouer
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds -
alloc_flags is a bitmask of flags but it is signed which does not
necessarily generate the best code depending on the compiler. Even
without an impact, it makes more sense that this be unsigned.Signed-off-by: Mel Gorman
Acked-by: Vlastimil Babka
Cc: Jesper Dangaard Brouer
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds -
Many developers already know that field for reference count of the
struct page is _count and atomic type. They would try to handle it
directly and this could break the purpose of page reference count
tracepoint. To prevent direct _count modification, this patch rename it
to _refcount and add warning message on the code. After that, developer
who need to handle reference count will find that field should not be
accessed directly.[akpm@linux-foundation.org: fix comments, per Vlastimil]
[akpm@linux-foundation.org: Documentation/vm/transhuge.txt too]
[sfr@canb.auug.org.au: sync ethernet driver changes]
Signed-off-by: Joonsoo Kim
Signed-off-by: Stephen Rothwell
Cc: Vlastimil Babka
Cc: Hugh Dickins
Cc: Johannes Berg
Cc: "David S. Miller"
Cc: Sunil Goutham
Cc: Chris Metcalf
Cc: Manish Chopra
Cc: Yuval Mintz
Cc: Tariq Toukan
Cc: Saeed Mahameed
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds
26 Mar, 2016
1 commit
-
This patch (of 5):
This is based on the idea from Mel Gorman discussed during LSFMM 2015
and independently brought up by Oleg Nesterov.The OOM killer currently allows to kill only a single task in a good
hope that the task will terminate in a reasonable time and frees up its
memory. Such a task (oom victim) will get an access to memory reserves
via mark_oom_victim to allow a forward progress should there be a need
for additional memory during exit path.It has been shown (e.g. by Tetsuo Handa) that it is not that hard to
construct workloads which break the core assumption mentioned above and
the OOM victim might take unbounded amount of time to exit because it
might be blocked in the uninterruptible state waiting for an event (e.g.
lock) which is blocked by another task looping in the page allocator.This patch reduces the probability of such a lockup by introducing a
specialized kernel thread (oom_reaper) which tries to reclaim additional
memory by preemptively reaping the anonymous or swapped out memory owned
by the oom victim under an assumption that such a memory won't be needed
when its owner is killed and kicked from the userspace anyway. There is
one notable exception to this, though, if the OOM victim was in the
process of coredumping the result would be incomplete. This is
considered a reasonable constrain because the overall system health is
more important than debugability of a particular application.A kernel thread has been chosen because we need a reliable way of
invocation so workqueue context is not appropriate because all the
workers might be busy (e.g. allocating memory). Kswapd which sounds
like another good fit is not appropriate as well because it might get
blocked on locks during reclaim as well.oom_reaper has to take mmap_sem on the target task for reading so the
solution is not 100% because the semaphore might be held or blocked for
write but the probability is reduced considerably wrt. basically any
lock blocking forward progress as described above. In order to prevent
from blocking on the lock without any forward progress we are using only
a trylock and retry 10 times with a short sleep in between. Users of
mmap_sem which need it for write should be carefully reviewed to use
_killable waiting as much as possible and reduce allocations requests
done with the lock held to absolute minimum to reduce the risk even
further.The API between oom killer and oom reaper is quite trivial.
wake_oom_reaper updates mm_to_reap with cmpxchg to guarantee only
NULL->mm transition and oom_reaper clear this atomically once it is done
with the work. This means that only a single mm_struct can be reaped at
the time. As the operation is potentially disruptive we are trying to
limit it to the ncessary minimum and the reaper blocks any updates while
it operates on an mm. mm_struct is pinned by mm_count to allow parallel
exit_mmap and a race is detected by atomic_inc_not_zero(mm_users).Signed-off-by: Michal Hocko
Suggested-by: Oleg Nesterov
Suggested-by: Mel Gorman
Acked-by: Mel Gorman
Acked-by: David Rientjes
Cc: Mel Gorman
Cc: Tetsuo Handa
Cc: Oleg Nesterov
Cc: Hugh Dickins
Cc: Andrea Argangeli
Cc: Rik van Riel
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds
18 Mar, 2016
4 commits
-
Most of the mm subsystem uses pr_ so make it consistent.
Miscellanea:
- Realign arguments
- Add missing newline to format
- kmemleak-test.c has a "kmemleak: " prefix added to the
"Kmemleak testing" logging message via pr_fmtSigned-off-by: Joe Perches
Acked-by: Tejun Heo [percpu]
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds -
The success of CMA allocation largely depends on the success of
migration and key factor of it is page reference count. Until now, page
reference is manipulated by direct calling atomic functions so we cannot
follow up who and where manipulate it. Then, it is hard to find actual
reason of CMA allocation failure. CMA allocation should be guaranteed
to succeed so finding offending place is really important.In this patch, call sites where page reference is manipulated are
converted to introduced wrapper function. This is preparation step to
add tracepoint to each page reference manipulation function. With this
facility, we can easily find reason of CMA allocation failure. There is
no functional change in this patch.In addition, this patch also converts reference read sites. It will
help a second step that renames page._count to something else and
prevents later attempt to direct access to it (Suggested by Andrew).Signed-off-by: Joonsoo Kim
Acked-by: Michal Nazarewicz
Acked-by: Vlastimil Babka
Cc: Minchan Kim
Cc: Mel Gorman
Cc: "Kirill A. Shutemov"
Cc: Sergey Senozhatsky
Cc: Steven Rostedt
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds -
Similarly to direct reclaim/compaction, kswapd attempts to combine
reclaim and compaction to attempt making memory allocation of given
order available.The details differ from direct reclaim e.g. in having high watermark as
a goal. The code involved in kswapd's reclaim/compaction decisions has
evolved to be quite complex.Testing reveals that it doesn't actually work in at least one scenario,
and closer inspection suggests that it could be greatly simplified
without compromising on the goal (make high-order page available) or
efficiency (don't reclaim too much). The simplification relieas of
doing all compaction in kcompactd, which is simply woken up when high
watermarks are reached by kswapd's reclaim.The scenario where kswapd compaction doesn't work was found with mmtests
test stress-highalloc configured to attempt order-9 allocations without
direct reclaim, just waking up kswapd. There was no compaction attempt
from kswapd during the whole test. Some added instrumentation shows
what happens:- balance_pgdat() sets end_zone to Normal, as it's not balanced
- reclaim is attempted on DMA zone, which sets nr_attempted to 99, but
it cannot reclaim anything, so sc.nr_reclaimed is 0
- for zones DMA32 and Normal, kswapd_shrink_zone uses testorder=0, so
it merely checks if high watermarks were reached for base pages.
This is true, so no reclaim is attempted. For DMA, testorder=0
wasn't used, as compaction_suitable() returned COMPACT_SKIPPED
- even though the pgdat_needs_compaction flag wasn't set to false, no
compaction happens due to the condition sc.nr_reclaimed >
nr_attempted being false (as 0 < 99)
- priority-- due to nr_reclaimed being 0, repeat until priority reaches
0 pgdat_balanced() is false as only the small zone DMA appears
balanced (curiously in that check, watermark appears OK and
compaction_suitable() returns COMPACT_PARTIAL, because a lower
classzone_idx is used there)Now, even if it was decided that reclaim shouldn't be attempted on the
DMA zone, the scenario would be the same, as (sc.nr_reclaimed=0 >
nr_attempted=0) is also false. The condition really should use >= as
the comment suggests. Then there is a mismatch in the check for setting
pgdat_needs_compaction to false using low watermark, while the rest uses
high watermark, and who knows what other subtlety. Hopefully this
demonstrates that this is unsustainable.Luckily we can simplify this a lot. The reclaim/compaction decisions
make sense for direct reclaim scenario, but in kswapd, our primary goal
is to reach high watermark in order-0 pages. Afterwards we can attempt
compaction just once. Unlike direct reclaim, we don't reclaim extra
pages (over the high watermark), the current code already disallows it
for good reasons.After this patch, we simply wake up kcompactd to process the pgdat,
after we have either succeeded or failed to reach the high watermarks in
kswapd, which goes to sleep. We pass kswapd's order and classzone_idx,
so kcompactd can apply the same criteria to determine which zones are
worth compacting. Note that we use the classzone_idx from
wakeup_kswapd(), not balanced_classzone_idx which can include higher
zones that kswapd tried to balance too, but didn't consider them in
pgdat_balanced().Since kswapd now cannot create high-order pages itself, we need to
adjust how it determines the zones to be balanced. The key element here
is adding a "highorder" parameter to zone_balanced, which, when set to
false, makes it consider only order-0 watermark instead of the desired
higher order (this was done previously by kswapd_shrink_zone(), but not
elsewhere). This false is passed for example in pgdat_balanced().
Importantly, wakeup_kswapd() uses true to make sure kswapd and thus
kcompactd are woken up for a high-order allocation failure.The last thing is to decide what to do with pageblock_skip bitmap
handling. Compaction maintains a pageblock_skip bitmap to record
pageblocks where isolation recently failed. This bitmap can be reset by
three ways:1) direct compaction is restarting after going through the full deferred cycle
2) kswapd goes to sleep, and some other direct compaction has previously
finished scanning the whole zone and set zone->compact_blockskip_flush.
Note that a successful direct compaction clears this flag.3) compaction was invoked manually via trigger in /proc
The case 2) is somewhat fuzzy to begin with, but after introducing
kcompactd we should update it. The check for direct compaction in 1),
and to set the flush flag in 2) use current_is_kswapd(), which doesn't
work for kcompactd. Thus, this patch adds bool direct_compaction to
compact_control to use in 2). For the case 1) we remove the check
completely - unlike the former kswapd compaction, kcompactd does use the
deferred compaction functionality, so flushing tied to restarting from
deferred compaction makes sense here.Note that when kswapd goes to sleep, kcompactd is woken up, so it will
see the flushed pageblock_skip bits. This is different from when the
former kswapd compaction observed the bits and I believe it makes more
sense. Kcompactd can afford to be more thorough than a direct
compaction trying to limit allocation latency, or kswapd whose primary
goal is to reclaim.For testing, I used stress-highalloc configured to do order-9
allocations with GFP_NOWAIT|__GFP_HIGH|__GFP_COMP, so they relied just
on kswapd/kcompactd reclaim/compaction (the interfering kernel builds in
phases 1 and 2 work as usual):stress-highalloc
4.5-rc1+before 4.5-rc1+after
-nodirect -nodirect
Success 1 Min 1.00 ( 0.00%) 5.00 (-66.67%)
Success 1 Mean 1.40 ( 0.00%) 6.20 (-55.00%)
Success 1 Max 2.00 ( 0.00%) 7.00 (-16.67%)
Success 2 Min 1.00 ( 0.00%) 5.00 (-66.67%)
Success 2 Mean 1.80 ( 0.00%) 6.40 (-52.38%)
Success 2 Max 3.00 ( 0.00%) 7.00 (-16.67%)
Success 3 Min 34.00 ( 0.00%) 62.00 ( 1.59%)
Success 3 Mean 41.80 ( 0.00%) 63.80 ( 1.24%)
Success 3 Max 53.00 ( 0.00%) 65.00 ( 2.99%)User 3166.67 3181.09
System 1153.37 1158.25
Elapsed 1768.53 1799.374.5-rc1+before 4.5-rc1+after
-nodirect -nodirect
Direct pages scanned 32938 32797
Kswapd pages scanned 2183166 2202613
Kswapd pages reclaimed 2152359 2143524
Direct pages reclaimed 32735 32545
Percentage direct scans 1% 1%
THP fault alloc 579 612
THP collapse alloc 304 316
THP splits 0 0
THP fault fallback 793 778
THP collapse fail 11 16
Compaction stalls 1013 1007
Compaction success 92 67
Compaction failures 920 939
Page migrate success 238457 721374
Page migrate failure 23021 23469
Compaction pages isolated 504695 1479924
Compaction migrate scanned 661390 8812554
Compaction free scanned 13476658 84327916
Compaction cost 262 838After this patch we see improvements in allocation success rate
(especially for phase 3) along with increased compaction activity. The
compaction stalls (direct compaction) in the interfering kernel builds
(probably THP's) also decreased somewhat thanks to kcompactd activity,
yet THP alloc successes improved a bit.Note that elapsed and user time isn't so useful for this benchmark,
because of the background interference being unpredictable. It's just
to quickly spot some major unexpected differences. System time is
somewhat more useful and that didn't increase.Also (after adjusting mmtests' ftrace monitor):
Time kswapd awake 2547781 2269241
Time kcompactd awake 0 119253
Time direct compacting 939937 557649
Time kswapd compacting 0 0
Time kcompactd compacting 0 119099The decrease of overal time spent compacting appears to not match the
increased compaction stats. I suspect the tasks get rescheduled and
since the ftrace monitor doesn't see that, the reported time is wall
time, not CPU time. But arguably direct compactors care about overall
latency anyway, whether busy compacting or waiting for CPU doesn't
matter. And that latency seems to almost halved.It's also interesting how much time kswapd spent awake just going
through all the priorities and failing to even try compacting, over and
over.We can also configure stress-highalloc to perform both direct
reclaim/compaction and wakeup kswapd/kcompactd, by using
GFP_KERNEL|__GFP_HIGH|__GFP_COMP:stress-highalloc
4.5-rc1+before 4.5-rc1+after
-direct -direct
Success 1 Min 4.00 ( 0.00%) 9.00 (-50.00%)
Success 1 Mean 8.00 ( 0.00%) 10.00 (-19.05%)
Success 1 Max 12.00 ( 0.00%) 11.00 ( 15.38%)
Success 2 Min 4.00 ( 0.00%) 9.00 (-50.00%)
Success 2 Mean 8.20 ( 0.00%) 10.00 (-16.28%)
Success 2 Max 13.00 ( 0.00%) 11.00 ( 8.33%)
Success 3 Min 75.00 ( 0.00%) 74.00 ( 1.33%)
Success 3 Mean 75.60 ( 0.00%) 75.20 ( 0.53%)
Success 3 Max 77.00 ( 0.00%) 76.00 ( 0.00%)User 3344.73 3246.04
System 1194.24 1172.29
Elapsed 1838.04 1836.764.5-rc1+before 4.5-rc1+after
-direct -direct
Direct pages scanned 125146 120966
Kswapd pages scanned 2119757 2135012
Kswapd pages reclaimed 2073183 2108388
Direct pages reclaimed 124909 120577
Percentage direct scans 5% 5%
THP fault alloc 599 652
THP collapse alloc 323 354
THP splits 0 0
THP fault fallback 806 793
THP collapse fail 17 16
Compaction stalls 2457 2025
Compaction success 906 518
Compaction failures 1551 1507
Page migrate success 2031423 2360608
Page migrate failure 32845 40852
Compaction pages isolated 4129761 4802025
Compaction migrate scanned 11996712 21750613
Compaction free scanned 214970969 344372001
Compaction cost 2271 2694In this scenario, this patch doesn't change the overall success rate as
direct compaction already tries all it can. There's however significant
reduction in direct compaction stalls (that is, the number of
allocations that went into direct compaction). The number of successes
(i.e. direct compaction stalls that ended up with successful
allocation) is reduced by the same number. This means the offload to
kcompactd is working as expected, and direct compaction is reduced
either due to detecting contention, or compaction deferred by kcompactd.
In the previous version of this patchset there was some apparent
reduction of success rate, but the changes in this version (such as
using sync compaction only), new baseline kernel, and/or averaging
results from 5 executions (my bet), made this go away.Ftrace-based stats seem to roughly agree:
Time kswapd awake 2532984 2326824
Time kcompactd awake 0 257916
Time direct compacting 864839 735130
Time kswapd compacting 0 0
Time kcompactd compacting 0 257585Signed-off-by: Vlastimil Babka
Cc: Andrea Arcangeli
Cc: "Kirill A. Shutemov"
Cc: Rik van Riel
Cc: Joonsoo Kim
Cc: Mel Gorman
Cc: David Rientjes
Cc: Michal Hocko
Cc: Johannes Weiner
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds -
Currently /proc/kpageflags returns nothing for "tail" buddy pages, which
is inconvenient when grasping how free pages are distributed. This
patch sets KPF_BUDDY for such pages.With this patch:
$ grep MemFree /proc/meminfo ; tools/vm/page-types -b buddy
MemFree: 3134992 kB
flags page-count MB symbolic-flags long-symbolic-flags
0x0000000000000400 779272 3044 __________B_______________________________ buddy
0x0000000000000c00 4385 17 __________BM______________________________ buddy,mmap
total 783657 3061783657 pages is 3134628 kB (roughly consistent with the global counter,)
so it's OK.[akpm@linux-foundation.org: update comment, per Naoya]
Signed-off-by: Naoya Horiguchi
Reviewed-by: Vladimir Davydov >
Cc: Konstantin Khlebnikov
Cc: Naoya Horiguchi
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds
16 Mar, 2016
2 commits
-
There is a performance drop report due to hugepage allocation and in
there half of cpu time are spent on pageblock_pfn_to_page() in
compaction [1].In that workload, compaction is triggered to make hugepage but most of
pageblocks are un-available for compaction due to pageblock type and
skip bit so compaction usually fails. Most costly operations in this
case is to find valid pageblock while scanning whole zone range. To
check if pageblock is valid to compact, valid pfn within pageblock is
required and we can obtain it by calling pageblock_pfn_to_page(). This
function checks whether pageblock is in a single zone and return valid
pfn if possible. Problem is that we need to check it every time before
scanning pageblock even if we re-visit it and this turns out to be very
expensive in this workload.Although we have no way to skip this pageblock check in the system where
hole exists at arbitrary position, we can use cached value for zone
continuity and just do pfn_to_page() in the system where hole doesn't
exist. This optimization considerably speeds up in above workload.Before vs After
Max: 1096 MB/s vs 1325 MB/s
Min: 635 MB/s 1015 MB/s
Avg: 899 MB/s 1194 MB/sAvg is improved by roughly 30% [2].
[1]: http://www.spinics.net/lists/linux-mm/msg97378.html
[2]: https://lkml.org/lkml/2015/12/9/23[akpm@linux-foundation.org: don't forget to restore zone->contiguous on error path, per Vlastimil]
Signed-off-by: Joonsoo Kim
Reported-by: Aaron Lu
Acked-by: Vlastimil Babka
Tested-by: Aaron Lu
Cc: Mel Gorman
Cc: Rik van Riel
Cc: David Rientjes
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds -
In mm we use several kinds of flags bitfields that are sometimes printed
for debugging purposes, or exported to userspace via sysfs. To make
them easier to interpret independently on kernel version and config, we
want to dump also the symbolic flag names. So far this has been done
with repeated calls to pr_cont(), which is unreliable on SMP, and not
usable for e.g. sysfs export.To get a more reliable and universal solution, this patch extends
printk() format string for pointers to handle the page flags (%pGp),
gfp_flags (%pGg) and vma flags (%pGv). Existing users of
dump_flag_names() are converted and simplified.It would be possible to pass flags by value instead of pointer, but the
%p format string for pointers already has extensions for various kernel
structures, so it's a good fit, and the extra indirection in a
non-critical path is negligible.[linux@rasmusvillemoes.dk: lots of good implementation suggestions]
Signed-off-by: Vlastimil Babka
Acked-by: Michal Hocko
Cc: Steven Rostedt
Cc: Peter Zijlstra
Cc: Arnaldo Carvalho de Melo
Cc: Ingo Molnar
Cc: Rasmus Villemoes
Cc: Joonsoo Kim
Cc: Minchan Kim
Cc: Sasha Levin
Cc: "Kirill A. Shutemov"
Cc: Mel Gorman
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds
04 Feb, 2016
2 commits
-
* add VM_STACK as alias for VM_GROWSUP/DOWN depending on architecture
* always account VMAs with flag VM_STACK as stack (as it was before)
* cleanup classifying helpers
* update comments and documentationSigned-off-by: Konstantin Khlebnikov
Tested-by: Sudip Mukherjee
Cc: Cyrill Gorcunov
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds -
This patch provides a way of working around a slight regression
introduced by commit 84638335900f ("mm: rework virtual memory
accounting").Before that commit RLIMIT_DATA have control only over size of the brk
region. But that change have caused problems with all existing versions
of valgrind, because it set RLIMIT_DATA to zero.This patch fixes rlimit check (limit actually in bytes, not pages) and
by default turns it into warning which prints at first VmData misuse:"mmap: top (795): VmData 516096 exceed data ulimit 512000. Will be forbidden soon."
Behavior is controlled by boot param ignore_rlimit_data=y/n and by sysfs
/sys/module/kernel/parameters/ignore_rlimit_data. For now it set to "y".[akpm@linux-foundation.org: tweak kernel-parameters.txt text[
Signed-off-by: Konstantin Khlebnikov
Link: http://lkml.kernel.org/r/20151228211015.GL2194@uranus
Reported-by: Christian Borntraeger
Cc: Cyrill Gorcunov
Cc: Linus Torvalds
Cc: Vegard Nossum
Cc: Peter Zijlstra
Cc: Vladimir Davydov
Cc: Andy Lutomirski
Cc: Quentin Casasnovas
Cc: Kees Cook
Cc: Willy Tarreau
Cc: Pavel Emelyanov
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds
16 Jan, 2016
2 commits
-
This patch adds implementation of split_huge_page() for new
refcountings.Unlike previous implementation, new split_huge_page() can fail if
somebody holds GUP pin on the page. It also means that pin on page
would prevent it from bening split under you. It makes situation in
many places much cleaner.The basic scheme of split_huge_page():
- Check that sum of mapcounts of all subpage is equal to page_count()
plus one (caller pin). Foll off with -EBUSY. This way we can avoid
useless PMD-splits.- Freeze the page counters by splitting all PMD and setup migration
PTEs.- Re-check sum of mapcounts against page_count(). Page's counts are
stable now. -EBUSY if page is pinned.- Split compound page.
- Unfreeze the page by removing migration entries.
Signed-off-by: Kirill A. Shutemov
Tested-by: Sasha Levin
Tested-by: Aneesh Kumar K.V
Acked-by: Jerome Marchand
Cc: Vlastimil Babka
Cc: Andrea Arcangeli
Cc: Hugh Dickins
Cc: Dave Hansen
Cc: Mel Gorman
Cc: Rik van Riel
Cc: Naoya Horiguchi
Cc: Steve Capper
Cc: Johannes Weiner
Cc: Michal Hocko
Cc: Christoph Lameter
Cc: David RientjesSigned-off-by: Andrew Morton
Signed-off-by: Linus Torvalds -
Tail page refcounting is utterly complicated and painful to support.
It uses ->_mapcount on tail pages to store how many times this page is
pinned. get_page() bumps ->_mapcount on tail page in addition to
->_count on head. This information is required by split_huge_page() to
be able to distribute pins from head of compound page to tails during
the split.We will need ->_mapcount to account PTE mappings of subpages of the
compound page. We eliminate need in current meaning of ->_mapcount in
tail pages by forbidding split entirely if the page is pinned.The only user of tail page refcounting is THP which is marked BROKEN for
now.Let's drop all this mess. It makes get_page() and put_page() much
simpler.Signed-off-by: Kirill A. Shutemov
Tested-by: Sasha Levin
Tested-by: Aneesh Kumar K.V
Acked-by: Vlastimil Babka
Acked-by: Jerome Marchand
Cc: Andrea Arcangeli
Cc: Hugh Dickins
Cc: Dave Hansen
Cc: Mel Gorman
Cc: Rik van Riel
Cc: Naoya Horiguchi
Cc: Steve Capper
Cc: Johannes Weiner
Cc: Michal Hocko
Cc: Christoph Lameter
Cc: David Rientjes
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds
07 Nov, 2015
4 commits
-
Let's try to be consistent about data type of page order.
[sfr@canb.auug.org.au: fix build (type of pageblock_order)]
[hughd@google.com: some configs end up with MAX_ORDER and pageblock_order having different types]
Signed-off-by: Kirill A. Shutemov
Acked-by: Michal Hocko
Acked-by: Vlastimil Babka
Reviewed-by: Andrea Arcangeli
Cc: "Paul E. McKenney"
Cc: Andi Kleen
Cc: Aneesh Kumar K.V
Cc: Christoph Lameter
Cc: David Rientjes
Cc: Joonsoo Kim
Cc: Sergey Senozhatsky
Signed-off-by: Stephen Rothwell
Signed-off-by: Hugh Dickins
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds -
Hugh has pointed that compound_head() call can be unsafe in some
context. There's one example:CPU0 CPU1
isolate_migratepages_block()
page_count()
compound_head()
!!PageTail() == true
put_page()
tail->first_page = NULL
head = tail->first_page
alloc_pages(__GFP_COMP)
prep_compound_page()
tail->first_page = head
__SetPageTail(p);
!!PageTail() == true
The race is pure theoretical. I don't it's possible to trigger it in
practice. But who knows.We can fix the race by changing how encode PageTail() and compound_head()
within struct page to be able to update them in one shot.The patch introduces page->compound_head into third double word block in
front of compound_dtor and compound_order. Bit 0 encodes PageTail() and
the rest bits are pointer to head page if bit zero is set.The patch moves page->pmd_huge_pte out of word, just in case if an
architecture defines pgtable_t into something what can have the bit 0
set.hugetlb_cgroup uses page->lru.next in the second tail page to store
pointer struct hugetlb_cgroup. The patch switch it to use page->private
in the second tail page instead. The space is free since ->first_page is
removed from the union.The patch also opens possibility to remove HUGETLB_CGROUP_MIN_ORDER
limitation, since there's now space in first tail page to store struct
hugetlb_cgroup pointer. But that's out of scope of the patch.That means page->compound_head shares storage space with:
- page->lru.next;
- page->next;
- page->rcu_head.next;That's too long list to be absolutely sure, but looks like nobody uses
bit 0 of the word.page->rcu_head.next guaranteed[1] to have bit 0 clean as long as we use
call_rcu(), call_rcu_bh(), call_rcu_sched(), or call_srcu(). But future
call_rcu_lazy() is not allowed as it makes use of the bit and we can
get false positive PageTail().[1] http://lkml.kernel.org/g/20150827163634.GD4029@linux.vnet.ibm.com
Signed-off-by: Kirill A. Shutemov
Acked-by: Michal Hocko
Reviewed-by: Andrea Arcangeli
Cc: Hugh Dickins
Cc: David Rientjes
Cc: Vlastimil Babka
Acked-by: Paul E. McKenney
Cc: Aneesh Kumar K.V
Cc: Andi Kleen
Cc: Christoph Lameter
Cc: Joonsoo Kim
Cc: Sergey Senozhatsky
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds -
Andrew stated the following
We have quite a history of remote parts of the kernel using
weird/wrong/inexplicable combinations of __GFP_ flags. I tend
to think that this is because we didn't adequately explain the
interface.And I don't think that gfp.h really improved much in this area as
a result of this patchset. Could you go through it some time and
decide if we've adequately documented all this stuff?This patches first moves some GFP flag combinations that are part of the MM
internals to mm/internal.h. The rest of the patch documents the __GFP_FOO
bits under various headings and then documents the flag combinations. It
will not help callers that are brain damaged but the clarity might motivate
some fixes and avoid future mistakes.Signed-off-by: Mel Gorman
Cc: Johannes Weiner
Cc: Rik van Riel
Cc: Vlastimil Babka
Cc: David Rientjes
Cc: Joonsoo Kim
Cc: Michal Hocko
Cc: Vitaly Wool
Cc: Rik van Riel
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds -
File-backed pages that will be immediately written are balanced between
zones. This heuristic tries to avoid having a single zone filled with
recently dirtied pages but the checks are unnecessarily expensive. Move
consider_zone_balanced into the alloc_context instead of checking bitmaps
multiple times. The patch also gives the parameter a more meaningful
name.Signed-off-by: Mel Gorman
Acked-by: David Rientjes
Acked-by: Michal Hocko
Acked-by: Vlastimil Babka
Acked-by: Johannes Weiner
Cc: Christoph Lameter
Cc: Vitaly Wool
Cc: Rik van Riel
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds
06 Nov, 2015
1 commit
-
Commit e6c509f85455 ("mm: use clear_page_mlock() in page_remove_rmap()")
in v3.7 inadvertently made mlock_migrate_page() impotent: page migration
unmaps the page from userspace before migrating, and that commit clears
PageMlocked on the final unmap, leaving mlock_migrate_page() with
nothing to do. Not a serious bug, the next attempt at reclaiming the
page would fix it up; but a betrayal of page migration's intent - the
new page ought to emerge as PageMlocked.I don't see how to fix it for mlock_migrate_page() itself; but easily
fixed in remove_migration_pte(), by calling mlock_vma_page() when the vma
is VM_LOCKED - under pte lock as in try_to_unmap_one().Delete mlock_migrate_page()? Not quite, it does still serve a purpose for
migrate_misplaced_transhuge_page(): where we could replace it by a test,
clear_page_mlock(), mlock_vma_page() sequence; but would that be an
improvement? mlock_migrate_page() is fairly lean, and let's make it
leaner by skipping the irq save/restore now clearly not needed.Signed-off-by: Hugh Dickins
Cc: Christoph Lameter
Cc: "Kirill A. Shutemov"
Cc: Rik van Riel
Acked-by: Vlastimil Babka
Cc: Davidlohr Bueso
Cc: Oleg Nesterov
Cc: Sasha Levin
Cc: Dmitry Vyukov
Cc: KOSAKI Motohiro
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds
09 Sep, 2015
1 commit
-
We cache isolate_start_pfn before entering isolate_migratepages(). If
pageblock is skipped in isolate_migratepages() due to whatever reason,
cc->migrate_pfn can be far from isolate_start_pfn hence we flush pages
that were freed. For example, the following scenario can be possible:- assume order-9 compaction, pageblock order is 9
- start_isolate_pfn is 0x200
- isolate_migratepages()
- skip a number of pageblocks
- start to isolate from pfn 0x600
- cc->migrate_pfn = 0x620
- return
- last_migrated_pfn is set to 0x200
- check flushing condition
- current_block_start is set to 0x600
- last_migrated_pfn < current_block_start then do useless flushThis wrong flush would not help the performance and success rate so this
patch tries to fix it. One simple way to know the exact position where
we start to isolate migratable pages is that we cache it in
isolate_migratepages() before entering actual isolation. This patch
implements that and fixes the problem.Signed-off-by: Joonsoo Kim
Acked-by: Vlastimil Babka
Cc: Mel Gorman
Cc: Rik van Riel
Cc: David Rientjes
Cc: Minchan Kim
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds
05 Sep, 2015
2 commits
-
If a PTE is unmapped and it's dirty then it was writable recently. Due to
deferred TLB flushing, it's best to assume a writable TLB cache entry
exists. With that assumption, the TLB must be flushed before any IO can
start or the page is freed to avoid lost writes or data corruption. This
patch defers flushing of potentially writable TLBs as long as possible.Signed-off-by: Mel Gorman
Reviewed-by: Rik van Riel
Cc: Dave Hansen
Acked-by: Ingo Molnar
Cc: Linus Torvalds
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds -
An IPI is sent to flush remote TLBs when a page is unmapped that was
potentially accesssed by other CPUs. There are many circumstances where
this happens but the obvious one is kswapd reclaiming pages belonging to a
running process as kswapd and the task are likely running on separate
CPUs.On small machines, this is not a significant problem but as machine gets
larger with more cores and more memory, the cost of these IPIs can be
high. This patch uses a simple structure that tracks CPUs that
potentially have TLB entries for pages being unmapped. When the unmapping
is complete, the full TLB is flushed on the assumption that a refill cost
is lower than flushing individual entries.Architectures wishing to do this must give the following guarantee.
If a clean page is unmapped and not immediately flushed, the
architecture must guarantee that a write to that linear address
from a CPU with a cached TLB entry will trap a page fault.This is essentially what the kernel already depends on but the window is
much larger with this patch applied and is worth highlighting. The
architecture should consider whether the cost of the full TLB flush is
higher than sending an IPI to flush each individual entry. An additional
architecture helper called flush_tlb_local is required. It's a trivial
wrapper with some accounting in the x86 case.The impact of this patch depends on the workload as measuring any benefit
requires both mapped pages co-located on the LRU and memory pressure. The
case with the biggest impact is multiple processes reading mapped pages
taken from the vm-scalability test suite. The test case uses NR_CPU
readers of mapped files that consume 10*RAM.Linear mapped reader on a 4-node machine with 64G RAM and 48 CPUs
4.2.0-rc1 4.2.0-rc1
vanilla flushfull-v7
Ops lru-file-mmap-read-elapsed 159.62 ( 0.00%) 120.68 ( 24.40%)
Ops lru-file-mmap-read-time_range 30.59 ( 0.00%) 2.80 ( 90.85%)
Ops lru-file-mmap-read-time_stddv 6.70 ( 0.00%) 0.64 ( 90.38%)4.2.0-rc1 4.2.0-rc1
vanilla flushfull-v7
User 581.00 611.43
System 5804.93 4111.76
Elapsed 161.03 122.12This is showing that the readers completed 24.40% faster with 29% less
system CPU time. From vmstats, it is known that the vanilla kernel was
interrupted roughly 900K times per second during the steady phase of the
test and the patched kernel was interrupts 180K times per second.The impact is lower on a single socket machine.
4.2.0-rc1 4.2.0-rc1
vanilla flushfull-v7
Ops lru-file-mmap-read-elapsed 25.33 ( 0.00%) 20.38 ( 19.54%)
Ops lru-file-mmap-read-time_range 0.91 ( 0.00%) 1.44 (-58.24%)
Ops lru-file-mmap-read-time_stddv 0.28 ( 0.00%) 0.47 (-65.34%)4.2.0-rc1 4.2.0-rc1
vanilla flushfull-v7
User 58.09 57.64
System 111.82 76.56
Elapsed 27.29 22.55It's still a noticeable improvement with vmstat showing interrupts went
from roughly 500K per second to 45K per second.The patch will have no impact on workloads with no memory pressure or have
relatively few mapped pages. It will have an unpredictable impact on the
workload running on the CPU being flushed as it'll depend on how many TLB
entries need to be refilled and how long that takes. Worst case, the TLB
will be completely cleared of active entries when the target PFNs were not
resident at all.[sasha.levin@oracle.com: trace tlb flush after disabling preemption in try_to_unmap_flush]
Signed-off-by: Mel Gorman
Reviewed-by: Rik van Riel
Cc: Dave Hansen
Acked-by: Ingo Molnar
Cc: Linus Torvalds
Signed-off-by: Sasha Levin
Cc: Michal Hocko
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds
01 Jul, 2015
4 commits
-
Waiman Long reported that 24TB machines hit OOM during basic setup when
struct page initialisation was deferred. One approach is to initialise
memory on demand but it interferes with page allocator paths. This patch
creates dedicated threads to initialise memory before basic setup. It
then blocks on a rw_semaphore until completion as a wait_queue and counter
is overkill. This may be slower to boot but it's simplier overall and
also gets rid of a section mangling which existed so kswapd could do the
initialisation.[akpm@linux-foundation.org: include rwsem.h, use DECLARE_RWSEM, fix comment, remove unneeded cast]
Signed-off-by: Mel Gorman
Cc: Waiman Long
Cc: Dave Hansen
Cc: Scott Norton
Tested-by: Daniel J Blueman
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds -
mminit_verify_page_links() is an extremely paranoid check that was
introduced when memory initialisation was being heavily reworked.
Profiles indicated that up to 10% of parallel memory initialisation was
spent on checking this for every page. The cost could be reduced but in
practice this check only found problems very early during the
initialisation rewrite and has found nothing since. This patch removes an
expensive unnecessary check.Signed-off-by: Mel Gorman
Tested-by: Nate Zimmer
Tested-by: Waiman Long
Tested-by: Daniel J Blueman
Acked-by: Pekka Enberg
Cc: Robin Holt
Cc: Nate Zimmer
Cc: Dave Hansen
Cc: Waiman Long
Cc: Scott Norton
Cc: "Luck, Tony"
Cc: Ingo Molnar
Cc: "H. Peter Anvin"
Cc: Thomas Gleixner
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds -
Only a subset of struct pages are initialised at the moment. When this
patch is applied kswapd initialise the remaining struct pages in parallel.This should boot faster by spreading the work to multiple CPUs and
initialising data that is local to the CPU. The user-visible effect on
large machines is that free memory will appear to rapidly increase early
in the lifetime of the system until kswapd reports that all memory is
initialised in the kernel log. Once initialised there should be no other
user-visibile effects.Signed-off-by: Mel Gorman
Tested-by: Nate Zimmer
Tested-by: Waiman Long
Tested-by: Daniel J Blueman
Acked-by: Pekka Enberg
Cc: Robin Holt
Cc: Nate Zimmer
Cc: Dave Hansen
Cc: Waiman Long
Cc: Scott Norton
Cc: "Luck, Tony"
Cc: Ingo Molnar
Cc: "H. Peter Anvin"
Cc: Thomas Gleixner
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds -
This patch initalises all low memory struct pages and 2G of the highest
zone on each node during memory initialisation if
CONFIG_DEFERRED_STRUCT_PAGE_INIT is set. That config option cannot be set
but will be available in a later patch. Parallel initialisation of struct
page depends on some features from memory hotplug and it is necessary to
alter alter section annotations.Signed-off-by: Mel Gorman
Tested-by: Nate Zimmer
Tested-by: Waiman Long
Tested-by: Daniel J Blueman
Acked-by: Pekka Enberg
Cc: Robin Holt
Cc: Nate Zimmer
Cc: Dave Hansen
Cc: Waiman Long
Cc: Scott Norton
Cc: "Luck, Tony"
Cc: Ingo Molnar
Cc: "H. Peter Anvin"
Cc: Thomas Gleixner
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds