28 Feb, 2013
1 commit
-
munlock_vma_pages_range() was always incrementing addresses by PAGE_SIZE
at a time. When munlocking THP pages (or the huge zero page), this
resulted in taking the mm->page_table_lock 512 times in a row.We can do better by making use of the page_mask returned by
follow_page_mask (for the huge zero page case), or the size of the page
munlock_vma_page() operated on (for the true THP page case).Signed-off-by: Michel Lespinasse
Cc: Andrea Arcangeli
Cc: Rik van Riel
Cc: Mel Gorman
Cc: Hugh Dickins
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds
24 Feb, 2013
1 commit
-
In find_extend_vma(), we don't need mlock_vma_pages_range() to verify
the vma type - we know we're working with a stack. So, we can call
directly into __mlock_vma_pages_range(), and remove the last
make_pages_present() call site.Note that we don't use mm_populate() here, so we can't release the
mmap_sem while allocating new stack pages. This is deemed acceptable,
because the stack vmas grow by a bounded number of pages at a time, and
these are anon pages so we don't have to read from disk to populate
them.Signed-off-by: Michel Lespinasse
Acked-by: Rik van Riel
Tested-by: Andy Lutomirski
Cc: Greg Ungerer
Cc: David Howells
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds
12 Jan, 2013
1 commit
-
Eric Wong reported on 3.7 and 3.8-rc2 that ppoll() got stuck when
waiting for POLLIN on a local TCP socket. It was easier to trigger if
there was disk IO and dirty pages at the same time and he bisected it to
commit 1fb3f8ca0e92 ("mm: compaction: capture a suitable high-order page
immediately when it is made available").The intention of that patch was to improve high-order allocations under
memory pressure after changes made to reclaim in 3.6 drastically hurt
THP allocations but the approach was flawed. For Eric, the problem was
that page->pfmemalloc was not being cleared for captured pages leading
to a poor interaction with swap-over-NFS support causing the packets to
be dropped. However, I identified a few more problems with the patch
including the fact that it can increase contention on zone->lock in some
cases which could result in async direct compaction being aborted early.In retrospect the capture patch took the wrong approach. What it should
have done is mark the pageblock being migrated as MIGRATE_ISOLATE if it
was allocating for THP and avoided races that way. While the patch was
showing to improve allocation success rates at the time, the benefit is
marginal given the relative complexity and it should be revisited from
scratch in the context of the other reclaim-related changes that have
taken place since the patch was first written and tested. This patch
partially reverts commit 1fb3f8ca "mm: compaction: capture a suitable
high-order page immediately when it is made available".Reported-and-tested-by: Eric Wong
Tested-by: Eric Dumazet
Cc: stable@vger.kernel.org
Signed-off-by: Mel Gorman
Signed-off-by: Linus Torvalds
17 Dec, 2012
1 commit
-
Pull Automatic NUMA Balancing bare-bones from Mel Gorman:
"There are three implementations for NUMA balancing, this tree
(balancenuma), numacore which has been developed in tip/master and
autonuma which is in aa.git.In almost all respects balancenuma is the dumbest of the three because
its main impact is on the VM side with no attempt to be smart about
scheduling. In the interest of getting the ball rolling, it would be
desirable to see this much merged for 3.8 with the view to building
scheduler smarts on top and adapting the VM where required for 3.9.The most recent set of comparisons available from different people are
mel: https://lkml.org/lkml/2012/12/9/108
mingo: https://lkml.org/lkml/2012/12/7/331
tglx: https://lkml.org/lkml/2012/12/10/437
srikar: https://lkml.org/lkml/2012/12/10/397The results are a mixed bag. In my own tests, balancenuma does
reasonably well. It's dumb as rocks and does not regress against
mainline. On the other hand, Ingo's tests shows that balancenuma is
incapable of converging for this workloads driven by perf which is bad
but is potentially explained by the lack of scheduler smarts. Thomas'
results show balancenuma improves on mainline but falls far short of
numacore or autonuma. Srikar's results indicate we all suffer on a
large machine with imbalanced node sizes.My own testing showed that recent numacore results have improved
dramatically, particularly in the last week but not universally.
We've butted heads heavily on system CPU usage and high levels of
migration even when it shows that overall performance is better.
There are also cases where it regresses. Of interest is that for
specjbb in some configurations it will regress for lower numbers of
warehouses and show gains for higher numbers which is not reported by
the tool by default and sometimes missed in treports. Recently I
reported for numacore that the JVM was crashing with
NullPointerExceptions but currently it's unclear what the source of
this problem is. Initially I thought it was in how numacore batch
handles PTEs but I'm no longer think this is the case. It's possible
numacore is just able to trigger it due to higher rates of migration.These reports were quite late in the cycle so I/we would like to start
with this tree as it contains much of the code we can agree on and has
not changed significantly over the last 2-3 weeks."* tag 'balancenuma-v11' of git://git.kernel.org/pub/scm/linux/kernel/git/mel/linux-balancenuma: (50 commits)
mm/rmap, migration: Make rmap_walk_anon() and try_to_unmap_anon() more scalable
mm/rmap: Convert the struct anon_vma::mutex to an rwsem
mm: migrate: Account a transhuge page properly when rate limiting
mm: numa: Account for failed allocations and isolations as migration failures
mm: numa: Add THP migration for the NUMA working set scanning fault case build fix
mm: numa: Add THP migration for the NUMA working set scanning fault case.
mm: sched: numa: Delay PTE scanning until a task is scheduled on a new node
mm: sched: numa: Control enabling and disabling of NUMA balancing if !SCHED_DEBUG
mm: sched: numa: Control enabling and disabling of NUMA balancing
mm: sched: Adapt the scanning rate if a NUMA hinting fault does not migrate
mm: numa: Use a two-stage filter to restrict pages being migrated for unlikely tasknode relationships
mm: numa: migrate: Set last_nid on newly allocated page
mm: numa: split_huge_page: Transfer last_nid on tail page
mm: numa: Introduce last_nid to the page frame
sched: numa: Slowly increase the scanning period as NUMA faults are handled
mm: numa: Rate limit setting of pte_numa if node is saturated
mm: numa: Rate limit the amount of memory that is migrated between nodes
mm: numa: Structures for Migrate On Fault per NUMA migration rate limiting
mm: numa: Migrate pages handled during a pmd_numa hinting fault
mm: numa: Migrate on reference policy
...
12 Dec, 2012
1 commit
-
Several place need to find the pmd by(mm_struct, address), so introduce a
function to simplify it.[akpm@linux-foundation.org: fix warning]
Signed-off-by: Bob Liu
Cc: Andrea Arcangeli
Cc: Michal Hocko
Cc: Minchan Kim
Cc: Ni zhan Chen
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds
11 Dec, 2012
1 commit
-
Note: This is very heavily based on a patch from Peter Zijlstra with
fixes from Ingo Molnar, Hugh Dickins and Johannes Weiner. That patch
put a lot of migration logic into mm/huge_memory.c where it does
not belong. This version puts tries to share some of the migration
logic with migrate_misplaced_page. However, it should be noted
that now migrate.c is doing more with the pagetable manipulation
than is preferred. The end result is barely recognisable so as
before, the signed-offs had to be removed but will be re-added if
the original authors are ok with it.Add THP migration for the NUMA working set scanning fault case.
It uses the page lock to serialize. No migration pte dance is
necessary because the pte is already unmapped when we decide
to migrate.[dhillf@gmail.com: Fix memory leak on isolation failure]
[dhillf@gmail.com: Fix transfer of last_nid information]
Signed-off-by: Mel Gorman
09 Oct, 2012
12 commits
-
NR_MLOCK is only accounted in single page units: there's no logic to
handle transparent hugepages. This patch checks the appropriate number of
pages to adjust the statistics by so that the correct amount of memory is
reflected.Currently:
$ grep Mlocked /proc/meminfo
Mlocked: 19636 kB#define MAP_SIZE (4 << 30) /* 4GB */
void *ptr = mmap(NULL, MAP_SIZE, PROT_READ | PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
mlock(ptr, MAP_SIZE);$ grep Mlocked /proc/meminfo
Mlocked: 29844 kBmunlock(ptr, MAP_SIZE);
$ grep Mlocked /proc/meminfo
Mlocked: 19636 kBAnd with this patch:
$ grep Mlock /proc/meminfo
Mlocked: 19636 kBmlock(ptr, MAP_SIZE);
$ grep Mlock /proc/meminfo
Mlocked: 4213664 kBmunlock(ptr, MAP_SIZE);
$ grep Mlock /proc/meminfo
Mlocked: 19636 kBSigned-off-by: David Rientjes
Reported-by: Hugh Dickens
Acked-by: Hugh Dickins
Reviewed-by: Andrea Arcangeli
Cc: Naoya Horiguchi
Cc: KAMEZAWA Hiroyuki
Cc: Johannes Weiner
Reviewed-by: Michel Lespinasse
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds -
Presently CMA cannot migrate mlocked pages so it ends up failing to allocate
contiguous memory space.This patch makes mlocked pages be migrated out. Of course, it can affect
realtime processes but in CMA usecase, contiguous memory allocation failing
is far worse than access latency to an mlocked page being variable while
CMA is running. If someone wants to make the system realtime, he shouldn't
enable CMA because stalls can still happen at random times.[akpm@linux-foundation.org: tweak comment text, per Mel]
Signed-off-by: Minchan Kim
Acked-by: Mel Gorman
Cc: Michal Nazarewicz
Cc: Bartlomiej Zolnierkiewicz
Cc: Marek Szyprowski
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds -
We had thought that pages could no longer get freed while still marked as
mlocked; but Johannes Weiner posted this program to demonstrate that
truncating an mlocked private file mapping containing COWed pages is still
mishandled:#include
#include
#include
#include
#include
#include
#includeint main(void)
{
char *map;
int fd;system("grep mlockfreed /proc/vmstat");
fd = open("chigurh", O_CREAT|O_EXCL|O_RDWR);
unlink("chigurh");
ftruncate(fd, 4096);
map = mmap(NULL, 4096, PROT_WRITE, MAP_PRIVATE, fd, 0);
map[0] = 11;
mlock(map, sizeof(fd));
ftruncate(fd, 0);
close(fd);
munlock(map, sizeof(fd));
munmap(map, 4096);
system("grep mlockfreed /proc/vmstat");
return 0;
}The anon COWed pages are not caught by truncation's clear_page_mlock() of
the pagecache pages; but unmap_mapping_range() unmaps them, so we ought to
look out for them there in page_remove_rmap(). Indeed, why should
truncation or invalidation be doing the clear_page_mlock() when removing
from pagecache? mlock is a property of mapping in userspace, not a
property of pagecache: an mlocked unmapped page is nonsensical.Reported-by: Johannes Weiner
Signed-off-by: Hugh Dickins
Cc: Mel Gorman
Cc: Rik van Riel
Cc: Michel Lespinasse
Cc: Ying Han
Acked-by: Johannes Weiner
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds -
page_evictable(page, vma) is an irritant: almost all its callers pass
NULL for vma. Remove the vma arg and use mlocked_vma_newpage(vma, page)
explicitly in the couple of places it's needed. But in those places we
don't even need page_evictable() itself! They're dealing with a freshly
allocated anonymous page, which has no "mapping" and cannot be mlocked yet.Signed-off-by: Hugh Dickins
Acked-by: Mel Gorman
Cc: Rik van Riel
Acked-by: Johannes Weiner
Cc: Michel Lespinasse
Cc: Ying Han
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds -
This is almost entirely based on Rik's previous patches and discussions
with him about how this might be implemented.Order > 0 compaction stops when enough free pages of the correct page
order have been coalesced. When doing subsequent higher order
allocations, it is possible for compaction to be invoked many times.However, the compaction code always starts out looking for things to
compact at the start of the zone, and for free pages to compact things to
at the end of the zone.This can cause quadratic behaviour, with isolate_freepages starting at the
end of the zone each time, even though previous invocations of the
compaction code already filled up all free memory on that end of the zone.
This can cause isolate_freepages to take enormous amounts of CPU with
certain workloads on larger memory systems.This patch caches where the migration and free scanner should start from
on subsequent compaction invocations using the pageblock-skip information.
When compaction starts it begins from the cached restart points and will
update the cached restart points until a page is isolated or a pageblock
is skipped that would have been scanned by synchronous compaction.Signed-off-by: Mel Gorman
Acked-by: Rik van Riel
Cc: Richard Davies
Cc: Shaohua Li
Cc: Avi Kivity
Acked-by: Rafael Aquini
Cc: Fengguang Wu
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds -
When compaction was implemented it was known that scanning could
potentially be excessive. The ideal was that a counter be maintained for
each pageblock but maintaining this information would incur a severe
penalty due to a shared writable cache line. It has reached the point
where the scanning costs are a serious problem, particularly on
long-lived systems where a large process starts and allocates a large
number of THPs at the same time.Instead of using a shared counter, this patch adds another bit to the
pageblock flags called PG_migrate_skip. If a pageblock is scanned by
either migrate or free scanner and 0 pages were isolated, the pageblock is
marked to be skipped in the future. When scanning, this bit is checked
before any scanning takes place and the block skipped if set.The main difficulty with a patch like this is "when to ignore the cached
information?" If it's ignored too often, the scanning rates will still be
excessive. If the information is too stale then allocations will fail
that might have otherwise succeeded. In this patcho CMA always ignores the information
o If the migrate and free scanner meet then the cached information will
be discarded if it's at least 5 seconds since the last time the cache
was discarded
o If there are a large number of allocation failures, discard the cache.The time-based heuristic is very clumsy but there are few choices for a
better event. Depending solely on multiple allocation failures still
allows excessive scanning when THP allocations are failing in quick
succession due to memory pressure. Waiting until memory pressure is
relieved would cause compaction to continually fail instead of using
reclaim/compaction to try allocate the page. The time-based mechanism is
clumsy but a better option is not obvious.Signed-off-by: Mel Gorman
Acked-by: Rik van Riel
Cc: Richard Davies
Cc: Shaohua Li
Cc: Avi Kivity
Acked-by: Rafael Aquini
Cc: Fengguang Wu
Cc: Michal Nazarewicz
Cc: Bartlomiej Zolnierkiewicz
Cc: Kyungmin Park
Cc: Mark Brown
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds -
This reverts commit 7db8889ab05b ("mm: have order > 0 compaction start
off where it left") and commit de74f1cc ("mm: have order > 0 compaction
start near a pageblock with free pages"). These patches were a good
idea and tests confirmed that they massively reduced the amount of
scanning but the implementation is complex and tricky to understand. A
later patch will cache what pageblocks should be skipped and
reimplements the concept of compact_cached_free_pfn on top for both
migration and free scanners.Signed-off-by: Mel Gorman
Acked-by: Rik van Riel
Cc: Richard Davies
Cc: Shaohua Li
Cc: Avi Kivity
Acked-by: Rafael Aquini
Acked-by: Minchan Kim
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds -
isolate_migratepages_range() might isolate no pages if for example when
zone->lru_lock is contended and running asynchronous compaction. In this
case, we should abort compaction, otherwise, compact_zone will run a
useless loop and make zone->lru_lock is even contended.An additional check is added to ensure that cc.migratepages and
cc.freepages get properly drained whan compaction is aborted.[minchan@kernel.org: Putback pages isolated for migration if aborting]
[akpm@linux-foundation.org: compact_zone_order requires non-NULL arg contended]
[akpm@linux-foundation.org: make compact_zone_order() require non-NULL arg `contended']
[minchan@kernel.org: Putback pages isolated for migration if aborting]
Signed-off-by: Andrea Arcangeli
Signed-off-by: Shaohua Li
Signed-off-by: Mel Gorman
Acked-by: Minchan Kim
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds -
* Add ALLOC_CMA alloc flag and pass it to [__]zone_watermark_ok()
(from Minchan Kim).* During watermark check decrease available free pages number by
free CMA pages number if necessary (unmovable allocations cannot
use pages from CMA areas).Signed-off-by: Bartlomiej Zolnierkiewicz
Signed-off-by: Kyungmin Park
Cc: Marek Szyprowski
Cc: Michal Nazarewicz
Cc: Minchan Kim
Cc: Mel Gorman
Cc: Hugh Dickins
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds -
Drop clean cache pages instead of migration during alloc_contig_range() to
minimise allocation latency by reducing the amount of migration that is
necessary. It's useful for CMA because latency of migration is more
important than evicting the background process's working set. In
addition, as pages are reclaimed then fewer free pages for migration
targets are required so it avoids memory reclaiming to get free pages,
which is a contributory factor to increased latency.I measured elapsed time of __alloc_contig_migrate_range() which migrates
10M in 40M movable zone in QEMU machine.Before - 146ms, After - 7ms
[akpm@linux-foundation.org: fix nommu build]
Signed-off-by: Mel Gorman
Signed-off-by: Minchan Kim
Reviewed-by: Mel Gorman
Cc: Marek Szyprowski
Acked-by: Michal Nazarewicz
Cc: Rik van Riel
Tested-by: Kyungmin Park
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds -
Make sure the #endif that terminates the standard #ifndef / #define /
#endif construct gets labeled, and gets positioned at the end of the file
as is normally the case.Signed-off-by: Michel Lespinasse
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds -
While compaction is migrating pages to free up large contiguous blocks
for allocation it races with other allocation requests that may steal
these blocks or break them up. This patch alters direct compaction to
capture a suitable free page as soon as it becomes available to reduce
this race. It uses similar logic to split_free_page() to ensure that
watermarks are still obeyed.Signed-off-by: Mel Gorman
Reviewed-by: Rik van Riel
Reviewed-by: Minchan Kim
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds
22 Aug, 2012
1 commit
-
Jim Schutt reported a problem that pointed at compaction contending
heavily on locks. The workload is straight-forward and in his own words;The systems in question have 24 SAS drives spread across 3 HBAs,
running 24 Ceph OSD instances, one per drive. FWIW these servers
are dual-socket Intel 5675 Xeons w/48 GB memory. I've got ~160
Ceph Linux clients doing dd simultaneously to a Ceph file system
backed by 12 of these servers.Early in the test everything looks fine
procs -------------------memory------------------ ---swap-- -----io---- --system-- -----cpu-------
r b swpd free buff cache si so bi bo in cs us sy id wa st
31 15 0 287216 576 38606628 0 0 2 1158 2 14 1 3 95 0 0
27 15 0 225288 576 38583384 0 0 18 2222016 203357 134876 11 56 17 15 0
28 17 0 219256 576 38544736 0 0 11 2305932 203141 146296 11 49 23 17 0
6 18 0 215596 576 38552872 0 0 7 2363207 215264 166502 12 45 22 20 0
22 18 0 226984 576 38596404 0 0 3 2445741 223114 179527 12 43 23 22 0and then it goes to pot
procs -------------------memory------------------ ---swap-- -----io---- --system-- -----cpu-------
r b swpd free buff cache si so bi bo in cs us sy id wa st
163 8 0 464308 576 36791368 0 0 11 22210 866 536 3 13 79 4 0
207 14 0 917752 576 36181928 0 0 712 1345376 134598 47367 7 90 1 2 0
123 12 0 685516 576 36296148 0 0 429 1386615 158494 60077 8 84 5 3 0
123 12 0 598572 576 36333728 0 0 1107 1233281 147542 62351 7 84 5 4 0
622 7 0 660768 576 36118264 0 0 557 1345548 151394 59353 7 85 4 3 0
223 11 0 283960 576 36463868 0 0 46 1107160 121846 33006 6 93 1 1 0Note that system CPU usage is very high blocks being written out has
dropped by 42%. He analysed this with perf and foundperf record -g -a sleep 10
perf report --sort symbol --call-graph fractal,5
34.63% [k] _raw_spin_lock_irqsave
|
|--97.30%-- isolate_freepages
| compaction_alloc
| unmap_and_move
| migrate_pages
| compact_zone
| compact_zone_order
| try_to_compact_pages
| __alloc_pages_direct_compact
| __alloc_pages_slowpath
| __alloc_pages_nodemask
| alloc_pages_vma
| do_huge_pmd_anonymous_page
| handle_mm_fault
| do_page_fault
| page_fault
| |
| |--87.39%-- skb_copy_datagram_iovec
| | tcp_recvmsg
| | inet_recvmsg
| | sock_recvmsg
| | sys_recvfrom
| | system_call
| | __recv
| | |
| | --100.00%-- (nil)
| |
| --12.61%-- memcpy
--2.70%-- [...]There was other data but primarily it is all showing that compaction is
contended heavily on the zone->lock and zone->lru_lock.commit [b2eef8c0: mm: compaction: minimise the time IRQs are disabled
while isolating pages for migration] noted that it was possible for
migration to hold the lru_lock for an excessive amount of time. Very
broadly speaking this patch expands the concept.This patch introduces compact_checklock_irqsave() to check if a lock
is contended or the process needs to be scheduled. If either condition
is true then async compaction is aborted and the caller is informed.
The page allocator will fail a THP allocation if compaction failed due
to contention. This patch also introduces compact_trylock_irqsave()
which will acquire the lock only if it is not contended and the process
does not need to schedule.Reported-by: Jim Schutt
Tested-by: Jim Schutt
Signed-off-by: Mel Gorman
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds
01 Aug, 2012
4 commits
-
Change the skb allocation API to indicate RX usage and use this to fall
back to the PFMEMALLOC reserve when needed. SKBs allocated from the
reserve are tagged in skb->pfmemalloc. If an SKB is allocated from the
reserve and the socket is later found to be unrelated to page reclaim, the
packet is dropped so that the memory remains available for page reclaim.
Network protocols are expected to recover from this packet loss.[a.p.zijlstra@chello.nl: Ideas taken from various patches]
[davem@davemloft.net: Use static branches, coding style corrections]
[sebastian@breakpoint.cc: Avoid unnecessary cast, fix !CONFIG_NET build]
Signed-off-by: Mel Gorman
Acked-by: David S. Miller
Cc: Neil Brown
Cc: Peter Zijlstra
Cc: Mike Christie
Cc: Eric B Munson
Cc: Eric Dumazet
Cc: Sebastian Andrzej Siewior
Cc: Mel Gorman
Cc: Christoph Lameter
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds -
When a user or administrator requires swap for their application, they
create a swap partition and file, format it with mkswap and activate it
with swapon. Swap over the network is considered as an option in diskless
systems. The two likely scenarios are when blade servers are used as part
of a cluster where the form factor or maintenance costs do not allow the
use of disks and thin clients.The Linux Terminal Server Project recommends the use of the Network Block
Device (NBD) for swap according to the manual at
https://sourceforge.net/projects/ltsp/files/Docs-Admin-Guide/LTSPManual.pdf/download
There is also documentation and tutorials on how to setup swap over NBD at
places like https://help.ubuntu.com/community/UbuntuLTSP/EnableNBDSWAP The
nbd-client also documents the use of NBD as swap. Despite this, the fact
is that a machine using NBD for swap can deadlock within minutes if swap
is used intensively. This patch series addresses the problem.The core issue is that network block devices do not use mempools like
normal block devices do. As the host cannot control where they receive
packets from, they cannot reliably work out in advance how much memory
they might need. Some years ago, Peter Zijlstra developed a series of
patches that supported swap over an NFS that at least one distribution is
carrying within their kernels. This patch series borrows very heavily
from Peter's work to support swapping over NBD as a pre-requisite to
supporting swap-over-NFS. The bulk of the complexity is concerned with
preserving memory that is allocated from the PFMEMALLOC reserves for use
by the network layer which is needed for both NBD and NFS.Patch 1 adds knowledge of the PFMEMALLOC reserves to SLAB and SLUB to
preserve access to pages allocated under low memory situations
to callers that are freeing memory.Patch 2 optimises the SLUB fast path to avoid pfmemalloc checks
Patch 3 introduces __GFP_MEMALLOC to allow access to the PFMEMALLOC
reserves without setting PFMEMALLOC.Patch 4 opens the possibility for softirqs to use PFMEMALLOC reserves
for later use by network packet processing.Patch 5 only sets page->pfmemalloc when ALLOC_NO_WATERMARKS was required
Patch 6 ignores memory policies when ALLOC_NO_WATERMARKS is set.
Patches 7-12 allows network processing to use PFMEMALLOC reserves when
the socket has been marked as being used by the VM to clean pages. If
packets are received and stored in pages that were allocated under
low-memory situations and are unrelated to the VM, the packets
are dropped.Patch 11 reintroduces __skb_alloc_page which the networking
folk may object to but is needed in some cases to propogate
pfmemalloc from a newly allocated page to an skb. If there is a
strong objection, this patch can be dropped with the impact being
that swap-over-network will be slower in some cases but it should
not fail.Patch 13 is a micro-optimisation to avoid a function call in the
common case.Patch 14 tags NBD sockets as being SOCK_MEMALLOC so they can use
PFMEMALLOC if necessary.Patch 15 notes that it is still possible for the PFMEMALLOC reserve
to be depleted. To prevent this, direct reclaimers get throttled on
a waitqueue if 50% of the PFMEMALLOC reserves are depleted. It is
expected that kswapd and the direct reclaimers already running
will clean enough pages for the low watermark to be reached and
the throttled processes are woken up.Patch 16 adds a statistic to track how often processes get throttled
Some basic performance testing was run using kernel builds, netperf on
loopback for UDP and TCP, hackbench (pipes and sockets), iozone and
sysbench. Each of them were expected to use the sl*b allocators
reasonably heavily but there did not appear to be significant performance
variances.For testing swap-over-NBD, a machine was booted with 2G of RAM with a
swapfile backed by NBD. 8*NUM_CPU processes were started that create
anonymous memory mappings and read them linearly in a loop. The total
size of the mappings were 4*PHYSICAL_MEMORY to use swap heavily under
memory pressure.Without the patches and using SLUB, the machine locks up within minutes
and runs to completion with them applied. With SLAB, the story is
different as an unpatched kernel run to completion. However, the patched
kernel completed the test 45% faster.MICRO
3.5.0-rc2 3.5.0-rc2
vanilla swapnbd
Unrecognised test vmscan-anon-mmap-write
MMTests Statistics: duration
Sys Time Running Test (seconds) 197.80 173.07
User+Sys Time Running Test (seconds) 206.96 182.03
Total Elapsed Time (seconds) 3240.70 1762.09This patch: mm: sl[au]b: add knowledge of PFMEMALLOC reserve pages
Allocations of pages below the min watermark run a risk of the machine
hanging due to a lack of memory. To prevent this, only callers who have
PF_MEMALLOC or TIF_MEMDIE set and are not processing an interrupt are
allowed to allocate with ALLOC_NO_WATERMARKS. Once they are allocated to
a slab though, nothing prevents other callers consuming free objects
within those slabs. This patch limits access to slab pages that were
alloced from the PFMEMALLOC reserves.When this patch is applied, pages allocated from below the low watermark
are returned with page->pfmemalloc set and it is up to the caller to
determine how the page should be protected. SLAB restricts access to any
page with page->pfmemalloc set to callers which are known to able to
access the PFMEMALLOC reserve. If one is not available, an attempt is
made to allocate a new page rather than use a reserve. SLUB is a bit more
relaxed in that it only records if the current per-CPU page was allocated
from PFMEMALLOC reserve and uses another partial slab if the caller does
not have the necessary GFP or process flags. This was found to be
sufficient in tests to avoid hangs due to SLUB generally maintaining
smaller lists than SLAB.In low-memory conditions it does mean that !PFMEMALLOC allocators can fail
a slab allocation even though free objects are available because they are
being preserved for callers that are freeing pages.[a.p.zijlstra@chello.nl: Original implementation]
[sebastian@breakpoint.cc: Correct order of page flag clearing]
Signed-off-by: Mel Gorman
Cc: David Miller
Cc: Neil Brown
Cc: Peter Zijlstra
Cc: Mike Christie
Cc: Eric B Munson
Cc: Eric Dumazet
Cc: Sebastian Andrzej Siewior
Cc: Mel Gorman
Cc: Christoph Lameter
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds -
On architectures with CONFIG_HUGETLB_PAGE_SIZE_VARIABLE set, such as
Itanium, pageblock_order is a variable with default value of 0. It's set
to the right value by set_pageblock_order() in function
free_area_init_core().But pageblock_order may be used by sparse_init() before free_area_init_core()
is called along path:
sparse_init()
->sparse_early_usemaps_alloc_node()
->usemap_size()
->SECTION_BLOCKFLAGS_BITS
->((1UL << (PFN_SECTION_SHIFT - pageblock_order)) *
NR_PAGEBLOCK_BITS)The uninitialized pageblock_size will cause memory wasting because
usemap_size() returns a much bigger value then it's really needed.For example, on an Itanium platform,
sparse_init() pageblock_order=0 usemap_size=24576
free_area_init_core() before pageblock_order=0, usemap_size=24576
free_area_init_core() after pageblock_order=12, usemap_size=8That means 24K memory has been wasted for each section, so fix it by calling
set_pageblock_order() from sparse_init().Signed-off-by: Xishi Qiu
Signed-off-by: Jiang Liu
Cc: Tony Luck
Cc: Yinghai Lu
Cc: KAMEZAWA Hiroyuki
Cc: Benjamin Herrenschmidt
Cc: KOSAKI Motohiro
Cc: David Rientjes
Cc: Keping Chen
Cc:
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds -
Order > 0 compaction stops when enough free pages of the correct page
order have been coalesced. When doing subsequent higher order
allocations, it is possible for compaction to be invoked many times.However, the compaction code always starts out looking for things to
compact at the start of the zone, and for free pages to compact things to
at the end of the zone.This can cause quadratic behaviour, with isolate_freepages starting at the
end of the zone each time, even though previous invocations of the
compaction code already filled up all free memory on that end of the zone.This can cause isolate_freepages to take enormous amounts of CPU with
certain workloads on larger memory systems.The obvious solution is to have isolate_freepages remember where it left
off last time, and continue at that point the next time it gets invoked
for an order > 0 compaction. This could cause compaction to fail if
cc->free_pfn and cc->migrate_pfn are close together initially, in that
case we restart from the end of the zone and try once more.Forced full (order == -1) compactions are left alone.
[akpm@linux-foundation.org: checkpatch fixes]
[akpm@linux-foundation.org: s/laste/last/, use 80 cols]
Signed-off-by: Rik van Riel
Reported-by: Jim Schutt
Tested-by: Jim Schutt
Cc: Minchan Kim
Reviewed-by: KAMEZAWA Hiroyuki
Acked-by: Mel Gorman
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds
04 Jun, 2012
1 commit
-
This reverts commit 5ceb9ce6fe9462a298bb2cd5c9f1ca6cb80a0199.
That commit seems to be the cause of the mm compation list corruption
issues that Dave Jones reported. The locking (or rather, absense
there-of) is dubious, as is the use of the 'page' variable once it has
been found to be outside the pageblock range.So revert it for now, we can re-visit this for 3.6. If we even need to:
as Minchan Kim says, "The patch wasn't a bug fix and even test workload
was very theoretical".Reported-and-tested-by: Dave Jones
Acked-by: Hugh Dickins
Acked-by: KOSAKI Motohiro
Acked-by: Minchan Kim
Cc: Bartlomiej Zolnierkiewicz
Cc: Kyungmin Park
Cc: Andrew Morton
Signed-off-by: Linus Torvalds
02 Jun, 2012
1 commit
-
Pull vfs changes from Al Viro.
"A lot of misc stuff. The obvious groups:
* Miklos' atomic_open series; kills the damn abuse of
->d_revalidate() by NFS, which was the major stumbling block for
all work in that area.
* ripping security_file_mmap() and dealing with deadlocks in the
area; sanitizing the neighborhood of vm_mmap()/vm_munmap() in
general.
* ->encode_fh() switched to saner API; insane fake dentry in
mm/cleancache.c gone.
* assorted annotations in fs (endianness, __user)
* parts of Artem's ->s_dirty work (jff2 and reiserfs parts)
* ->update_time() work from Josef.
* other bits and pieces all over the place.Normally it would've been in two or three pull requests, but
signal.git stuff had eaten a lot of time during this cycle ;-/"Fix up trivial conflicts in Documentation/filesystems/vfs.txt (the
'truncate_range' inode method was removed by the VM changes, the VFS
update adds an 'update_time()' method), and in fs/btrfs/ulist.[ch] (due
to sparse fix added twice, with other changes nearby).* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs: (95 commits)
nfs: don't open in ->d_revalidate
vfs: retry last component if opening stale dentry
vfs: nameidata_to_filp(): don't throw away file on error
vfs: nameidata_to_filp(): inline __dentry_open()
vfs: do_dentry_open(): don't put filp
vfs: split __dentry_open()
vfs: do_last() common post lookup
vfs: do_last(): add audit_inode before open
vfs: do_last(): only return EISDIR for O_CREAT
vfs: do_last(): check LOOKUP_DIRECTORY
vfs: do_last(): make ENOENT exit RCU safe
vfs: make follow_link check RCU safe
vfs: do_last(): use inode variable
vfs: do_last(): inline walk_component()
vfs: do_last(): make exit RCU safe
vfs: split do_lookup()
Btrfs: move over to use ->update_time
fs: introduce inode operation ->update_time
reiserfs: get rid of resierfs_sync_super
reiserfs: mark the superblock as dirty a bit later
...
01 Jun, 2012
1 commit
-
take it to mm/util.c, convert vm_mmap() to use of that one and
take it to mm/util.c as well, convert both sys_mmap_pgoff() to
use of vm_mmap_pgoff()Signed-off-by: Al Viro
30 May, 2012
2 commits
-
When MIGRATE_UNMOVABLE pages are freed from MIGRATE_UNMOVABLE type
pageblock (and some MIGRATE_MOVABLE pages are left in it) waiting until an
allocation takes ownership of the block may take too long. The type of
the pageblock remains unchanged so the pageblock cannot be used as a
migration target during compaction.Fix it by:
* Adding enum compact_mode (COMPACT_ASYNC_[MOVABLE,UNMOVABLE], and
COMPACT_SYNC) and then converting sync field in struct compact_control
to use it.* Adding nr_pageblocks_skipped field to struct compact_control and
tracking how many destination pageblocks were of MIGRATE_UNMOVABLE type.
If COMPACT_ASYNC_MOVABLE mode compaction ran fully in
try_to_compact_pages() (COMPACT_COMPLETE) it implies that there is not a
suitable page for allocation. In this case then check how if there were
enough MIGRATE_UNMOVABLE pageblocks to try a second pass in
COMPACT_ASYNC_UNMOVABLE mode.* Scanning the MIGRATE_UNMOVABLE pageblocks (during COMPACT_SYNC and
COMPACT_ASYNC_UNMOVABLE compaction modes) and building a count based on
finding PageBuddy pages, page_count(page) == 0 or PageLRU pages. If all
pages within the MIGRATE_UNMOVABLE pageblock are in one of those three
sets change the whole pageblock type to MIGRATE_MOVABLE.My particular test case (on a ARM EXYNOS4 device with 512 MiB, which means
131072 standard 4KiB pages in 'Normal' zone) is to:- allocate 120000 pages for kernel's usage
- free every second page (60000 pages) of memory just allocated
- allocate and use 60000 pages from user space
- free remaining 60000 pages of kernel memory
(now we have fragmented memory occupied mostly by user space pages)
- try to allocate 100 order-9 (2048 KiB) pages for kernel's usageThe results:
- with compaction disabled I get 11 successful allocations
- with compaction enabled - 14 successful allocations
- with this patch I'm able to get all 100 successful allocationsNOTE: If we can make kswapd aware of order-0 request during compaction, we
can enhance kswapd with changing mode to COMPACT_ASYNC_FULL
(COMPACT_ASYNC_MOVABLE + COMPACT_ASYNC_UNMOVABLE). Please see the
following thread:http://marc.info/?l=linux-mm&m=133552069417068&w=2
[minchan@kernel.org: minor cleanups]
Cc: Mel Gorman
Cc: Minchan Kim
Cc: Rik van Riel
Cc: Marek Szyprowski
Signed-off-by: Bartlomiej Zolnierkiewicz
Signed-off-by: Kyungmin Park
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds -
Andrew pointed out that the is_mlocked_vma() is misnamed. A function
with name like that would expect bool return and no side-effects.Since it is called on the fault path for new page, rename it in this
patch.Signed-off-by: Ying Han
Reviewed-by: Rik van Riel
Acked-by: KOSAKI Motohiro
Acked-by: KAMEZAWA Hiroyuki
Reviewed-by: Minchan Kim
[akpm@linux-foundation.org: s/mlock_vma_newpage/mlock_vma_newpage/, per Minchan]
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds
21 May, 2012
1 commit
-
This commit exports some of the functions from compaction.c file
outside of it adding their declaration into internal.h header
file so that other mm related code can use them.This forced compaction.c to always be compiled (as opposed to being
compiled only if CONFIG_COMPACTION is defined) but as to avoid
introducing code that user did not ask for, part of the compaction.c
is now wrapped in on #ifdef.Signed-off-by: Michal Nazarewicz
Signed-off-by: Marek Szyprowski
Acked-by: Mel Gorman
Reviewed-by: KAMEZAWA Hiroyuki
Tested-by: Rob Clark
Tested-by: Ohad Ben-Cohen
Tested-by: Benjamin Gaignard
Tested-by: Robert Nelson
Tested-by: Barry Song
03 Nov, 2011
1 commit
-
Michel while working on the working set estimation code, noticed that
calling get_page_unless_zero() on a random pfn_to_page(random_pfn)
wasn't safe, if the pfn ended up being a tail page of a transparent
hugepage under splitting by __split_huge_page_refcount().He then found the problem could also theoretically materialize with
page_cache_get_speculative() during the speculative radix tree lookups
that uses get_page_unless_zero() in SMP if the radix tree page is freed
and reallocated and get_user_pages is called on it before
page_cache_get_speculative has a chance to call get_page_unless_zero().So the best way to fix the problem is to keep page_tail->_count zero at
all times. This will guarantee that get_page_unless_zero() can never
succeed on any tail page. page_tail->_mapcount is guaranteed zero and
is unused for all tail pages of a compound page, so we can simply
account the tail page references there and transfer them to
tail_page->_count in __split_huge_page_refcount() (in addition to the
head_page->_mapcount).While debugging this s/_count/_mapcount/ change I also noticed get_page is
called by direct-io.c on pages returned by get_user_pages. That wasn't
entirely safe because the two atomic_inc in get_page weren't atomic. As
opposed to other get_user_page users like secondary-MMU page fault to
establish the shadow pagetables would never call any superflous get_page
after get_user_page returns. It's safer to make get_page universally safe
for tail pages and to use get_page_foll() within follow_page (inside
get_user_pages()). get_page_foll() is safe to do the refcounting for tail
pages without taking any locks because it is run within PT lock protected
critical sections (PT lock for pte and page_table_lock for
pmd_trans_huge).The standard get_page() as invoked by direct-io instead will now take
the compound_lock but still only for tail pages. The direct-io paths
are usually I/O bound and the compound_lock is per THP so very
finegrined, so there's no risk of scalability issues with it. A simple
direct-io benchmarks with all lockdep prove locking and spinlock
debugging infrastructure enabled shows identical performance and no
overhead. So it's worth it. Ideally direct-io should stop calling
get_page() on pages returned by get_user_pages(). The spinlock in
get_page() is already optimized away for no-THP builds but doing
get_page() on tail pages returned by GUP is generally a rare operation
and usually only run in I/O paths.This new refcounting on page_tail->_mapcount in addition to avoiding new
RCU critical sections will also allow the working set estimation code to
work without any further complexity associated to the tail page
refcounting with THP.Signed-off-by: Andrea Arcangeli
Reported-by: Michel Lespinasse
Reviewed-by: Michel Lespinasse
Reviewed-by: Minchan Kim
Cc: Peter Zijlstra
Cc: Hugh Dickins
Cc: Johannes Weiner
Cc: Rik van Riel
Cc: Mel Gorman
Cc: KOSAKI Motohiro
Cc: Benjamin Herrenschmidt
Cc: David Gibson
Cc:
Cc:
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds
25 May, 2011
1 commit
-
When I was reading nommu code, I found that it handles the vma list/tree
in an unusual way. IIUC, because there can be more than one
identical/overrapped vmas in the list/tree, it sorts the tree more
strictly and does a linear search on the tree. But it doesn't applied to
the list (i.e. the list could be constructed in a different order than
the tree so that we can't use the list when finding the first vma in that
order).Since inserting/sorting a vma in the tree and link is done at the same
time, we can easily construct both of them in the same order. And linear
searching on the tree could be more costly than doing it on the list, it
can be converted to use the list.Also, after the commit 297c5eee3724 ("mm: make the vma list be doubly
linked") made the list be doubly linked, there were a couple of code need
to be fixed to construct the list properly.Patch 1/6 is a preparation. It maintains the list sorted same as the tree
and construct doubly-linked list properly. Patch 2/6 is a simple
optimization for the vma deletion. Patch 3/6 and 4/6 convert tree
traversal to list traversal and the rest are simple fixes and cleanups.This patch:
@vma added into @mm should be sorted by start addr, end addr and VMA
struct addr in that order because we may get identical VMAs in the @mm.
However this was true only for the rbtree, not for the list.This patch fixes this by remembering 'rb_prev' during the tree traversal
like find_vma_prepare() does and linking the @vma via __vma_link_list().
After this patch, we can iterate the whole VMAs in correct order simply by
using @mm->mmap list.[akpm@linux-foundation.org: avoid duplicating __vma_link_list()]
Signed-off-by: Namhyung Kim
Acked-by: Greg Ungerer
Cc: David Howells
Cc: Paul Mundt
Cc: Geert Uytterhoeven
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds
31 Mar, 2011
1 commit
-
Fixes generated by 'codespell' and manually reviewed.
Signed-off-by: Lucas De Marchi
18 Mar, 2011
1 commit
-
In most cases, get_user_pages and get_user_pages_fast should be used
to pin user pages in memory. But sometimes, some special flags except
FOLL_GET, FOLL_WRITE and FOLL_FORCE are needed, for example in
following patch, KVM needs FOLL_HWPOISON. To support these users,
__get_user_pages is exported directly.There are some symbol name conflicts in infiniband driver, fixed them too.
Signed-off-by: Huang Ying
CC: Andrew Morton
CC: Michel Lespinasse
CC: Roland Dreier
CC: Ralph Campbell
Signed-off-by: Marcelo Tosatti
18 Jan, 2011
1 commit
-
This reverts commit 744ed1442757767ffede5008bb13e0805085902e.
Chris Mason ended up chasing down some page allocation errors and pages
stuck waiting on the IO scheduler, and was able to narrow it down to two
commits: commit 744ed1442757 ("mm: batch activate_page() to reduce lock
contention") and d8505dee1a87 ("mm: simplify code of swap.c").This reverts the first of them.
Reported-and-debugged-by: Chris Mason
Cc: Mel Gorman
Cc: Andrew Morton
Cc: Jens Axboe
Cc: linux-mm
Cc: KAMEZAWA Hiroyuki
Cc: Andrea Arcangeli
Cc: Shaohua Li
Signed-off-by: Linus Torvalds
14 Jan, 2011
3 commits
-
The zone->lru_lock is heavily contented in workload where activate_page()
is frequently used. We could do batch activate_page() to reduce the lock
contention. The batched pages will be added into zone list when the pool
is full or page reclaim is trying to drain them.For example, in a 4 socket 64 CPU system, create a sparse file and 64
processes, processes shared map to the file. Each process read access the
whole file and then exit. The process exit will do unmap_vmas() and cause
a lot of activate_page() call. In such workload, we saw about 58% total
time reduction with below patch. Other workloads with a lot of
activate_page also benefits a lot too.I tested some microbenchmarks:
case-anon-cow-rand-mt 0.58%
case-anon-cow-rand -3.30%
case-anon-cow-seq-mt -0.51%
case-anon-cow-seq -5.68%
case-anon-r-rand-mt 0.23%
case-anon-r-rand 0.81%
case-anon-r-seq-mt -0.71%
case-anon-r-seq -1.99%
case-anon-rx-rand-mt 2.11%
case-anon-rx-seq-mt 3.46%
case-anon-w-rand-mt -0.03%
case-anon-w-rand -0.50%
case-anon-w-seq-mt -1.08%
case-anon-w-seq -0.12%
case-anon-wx-rand-mt -5.02%
case-anon-wx-seq-mt -1.43%
case-fork 1.65%
case-fork-sleep -0.07%
case-fork-withmem 1.39%
case-hugetlb -0.59%
case-lru-file-mmap-read-mt -0.54%
case-lru-file-mmap-read 0.61%
case-lru-file-mmap-read-rand -2.24%
case-lru-file-readonce -0.64%
case-lru-file-readtwice -11.69%
case-lru-memcg -1.35%
case-mmap-pread-rand-mt 1.88%
case-mmap-pread-rand -15.26%
case-mmap-pread-seq-mt 0.89%
case-mmap-pread-seq -69.72%
case-mmap-xread-rand-mt 0.71%
case-mmap-xread-seq-mt 0.38%The most significent are:
case-lru-file-readtwice -11.69%
case-mmap-pread-rand -15.26%
case-mmap-pread-seq -69.72%which use activate_page a lot. others are basically variations because
each run has slightly difference.[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Shaohua Li
Cc: Andi Kleen
Cc: Minchan Kim
Cc: KOSAKI Motohiro
Cc: Rik van Riel
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds -
Lately I've been working to make KVM use hugepages transparently without
the usual restrictions of hugetlbfs. Some of the restrictions I'd like to
see removed:1) hugepages have to be swappable or the guest physical memory remains
locked in RAM and can't be paged out to swap2) if a hugepage allocation fails, regular pages should be allocated
instead and mixed in the same vma without any failure and without
userland noticing3) if some task quits and more hugepages become available in the
buddy, guest physical memory backed by regular pages should be
relocated on hugepages automatically in regions under
madvise(MADV_HUGEPAGE) (ideally event driven by waking up the
kernel deamon if the order=HPAGE_PMD_SHIFT-PAGE_SHIFT list becomes
not null)4) avoidance of reservation and maximization of use of hugepages whenever
possible. Reservation (needed to avoid runtime fatal faliures) may be ok for
1 machine with 1 database with 1 database cache with 1 database cache size
known at boot time. It's definitely not feasible with a virtualization
hypervisor usage like RHEV-H that runs an unknown number of virtual machines
with an unknown size of each virtual machine with an unknown amount of
pagecache that could be potentially useful in the host for guest not using
O_DIRECT (aka cache=off).hugepages in the virtualization hypervisor (and also in the guest!) are
much more important than in a regular host not using virtualization,
becasue with NPT/EPT they decrease the tlb-miss cacheline accesses from 24
to 19 in case only the hypervisor uses transparent hugepages, and they
decrease the tlb-miss cacheline accesses from 19 to 15 in case both the
linux hypervisor and the linux guest both uses this patch (though the
guest will limit the addition speedup to anonymous regions only for
now...). Even more important is that the tlb miss handler is much slower
on a NPT/EPT guest than for a regular shadow paging or no-virtualization
scenario. So maximizing the amount of virtual memory cached by the TLB
pays off significantly more with NPT/EPT than without (even if there would
be no significant speedup in the tlb-miss runtime).The first (and more tedious) part of this work requires allowing the VM to
handle anonymous hugepages mixed with regular pages transparently on
regular anonymous vmas. This is what this patch tries to achieve in the
least intrusive possible way. We want hugepages and hugetlb to be used in
a way so that all applications can benefit without changes (as usual we
leverage the KVM virtualization design: by improving the Linux VM at
large, KVM gets the performance boost too).The most important design choice is: always fallback to 4k allocation if
the hugepage allocation fails! This is the _very_ opposite of some large
pagecache patches that failed with -EIO back then if a 64k (or similar)
allocation failed...Second important decision (to reduce the impact of the feature on the
existing pagetable handling code) is that at any time we can split an
hugepage into 512 regular pages and it has to be done with an operation
that can't fail. This way the reliability of the swapping isn't decreased
(no need to allocate memory when we are short on memory to swap) and it's
trivial to plug a split_huge_page* one-liner where needed without
polluting the VM. Over time we can teach mprotect, mremap and friends to
handle pmd_trans_huge natively without calling split_huge_page*. The fact
it can't fail isn't just for swap: if split_huge_page would return -ENOMEM
(instead of the current void) we'd need to rollback the mprotect from the
middle of it (ideally including undoing the split_vma) which would be a
big change and in the very wrong direction (it'd likely be simpler not to
call split_huge_page at all and to teach mprotect and friends to handle
hugepages instead of rolling them back from the middle). In short the
very value of split_huge_page is that it can't fail.The collapsing and madvise(MADV_HUGEPAGE) part will remain separated and
incremental and it'll just be an "harmless" addition later if this initial
part is agreed upon. It also should be noted that locking-wise replacing
regular pages with hugepages is going to be very easy if compared to what
I'm doing below in split_huge_page, as it will only happen when
page_count(page) matches page_mapcount(page) if we can take the PG_lock
and mmap_sem in write mode. collapse_huge_page will be a "best effort"
that (unlike split_huge_page) can fail at the minimal sign of trouble and
we can try again later. collapse_huge_page will be similar to how KSM
works and the madvise(MADV_HUGEPAGE) will work similar to
madvise(MADV_MERGEABLE).The default I like is that transparent hugepages are used at page fault
time. This can be changed with
/sys/kernel/mm/transparent_hugepage/enabled. The control knob can be set
to three values "always", "madvise", "never" which mean respectively that
hugepages are always used, or only inside madvise(MADV_HUGEPAGE) regions,
or never used. /sys/kernel/mm/transparent_hugepage/defrag instead
controls if the hugepage allocation should defrag memory aggressively
"always", only inside "madvise" regions, or "never".The pmd_trans_splitting/pmd_trans_huge locking is very solid. The
put_page (from get_user_page users that can't use mmu notifier like
O_DIRECT) that runs against a __split_huge_page_refcount instead was a
pain to serialize in a way that would result always in a coherent page
count for both tail and head. I think my locking solution with a
compound_lock taken only after the page_first is valid and is still a
PageHead should be safe but it surely needs review from SMP race point of
view. In short there is no current existing way to serialize the O_DIRECT
final put_page against split_huge_page_refcount so I had to invent a new
one (O_DIRECT loses knowledge on the mapping status by the time gup_fast
returns so...). And I didn't want to impact all gup/gup_fast users for
now, maybe if we change the gup interface substantially we can avoid this
locking, I admit I didn't think too much about it because changing the gup
unpinning interface would be invasive.If we ignored O_DIRECT we could stick to the existing compound refcounting
code, by simply adding a get_user_pages_fast_flags(foll_flags) where KVM
(and any other mmu notifier user) would call it without FOLL_GET (and if
FOLL_GET isn't set we'd just BUG_ON if nobody registered itself in the
current task mmu notifier list yet). But O_DIRECT is fundamental for
decent performance of virtualized I/O on fast storage so we can't avoid it
to solve the race of put_page against split_huge_page_refcount to achieve
a complete hugepage feature for KVM.Swap and oom works fine (well just like with regular pages ;). MMU
notifier is handled transparently too, with the exception of the young bit
on the pmd, that didn't have a range check but I think KVM will be fine
because the whole point of hugepages is that EPT/NPT will also use a huge
pmd when they notice gup returns pages with PageCompound set, so they
won't care of a range and there's just the pmd young bit to check in that
case.NOTE: in some cases if the L2 cache is small, this may slowdown and waste
memory during COWs because 4M of memory are accessed in a single fault
instead of 8k (the payoff is that after COW the program can run faster).
So we might want to switch the copy_huge_page (and clear_huge_page too) to
not temporal stores. I also extensively researched ways to avoid this
cache trashing with a full prefault logic that would cow in 8k/16k/32k/64k
up to 1M (I can send those patches that fully implemented prefault) but I
concluded they're not worth it and they add an huge additional complexity
and they remove all tlb benefits until the full hugepage has been faulted
in, to save a little bit of memory and some cache during app startup, but
they still don't improve substantially the cache-trashing during startup
if the prefault happens in >4k chunks. One reason is that those 4k pte
entries copied are still mapped on a perfectly cache-colored hugepage, so
the trashing is the worst one can generate in those copies (cow of 4k page
copies aren't so well colored so they trashes less, but again this results
in software running faster after the page fault). Those prefault patches
allowed things like a pte where post-cow pages were local 4k regular anon
pages and the not-yet-cowed pte entries were pointing in the middle of
some hugepage mapped read-only. If it doesn't payoff substantially with
todays hardware it will payoff even less in the future with larger l2
caches, and the prefault logic would blot the VM a lot. If one is
emebdded transparent_hugepage can be disabled during boot with sysfs or
with the boot commandline parameter transparent_hugepage=0 (or
transparent_hugepage=2 to restrict hugepages inside madvise regions) that
will ensure not a single hugepage is allocated at boot time. It is simple
enough to just disable transparent hugepage globally and let transparent
hugepages be allocated selectively by applications in the MADV_HUGEPAGE
region (both at page fault time, and if enabled with the
collapse_huge_page too through the kernel daemon).This patch supports only hugepages mapped in the pmd, archs that have
smaller hugepages will not fit in this patch alone. Also some archs like
power have certain tlb limits that prevents mixing different page size in
the same regions so they will not fit in this framework that requires
"graceful fallback" to basic PAGE_SIZE in case of physical memory
fragmentation. hugetlbfs remains a perfect fit for those because its
software limits happen to match the hardware limits. hugetlbfs also
remains a perfect fit for hugepage sizes like 1GByte that cannot be hoped
to be found not fragmented after a certain system uptime and that would be
very expensive to defragment with relocation, so requiring reservation.
hugetlbfs is the "reservation way", the point of transparent hugepages is
not to have any reservation at all and maximizing the use of cache and
hugepages at all times automatically.Some performance result:
vmx andrea # LD_PRELOAD=/usr/lib64/libhugetlbfs.so HUGETLB_MORECORE=yes HUGETLB_PATH=/mnt/huge/ ./largep
ages3
memset page fault 1566023
memset tlb miss 453854
memset second tlb miss 453321
random access tlb miss 41635
random access second tlb miss 41658
vmx andrea # LD_PRELOAD=/usr/lib64/libhugetlbfs.so HUGETLB_MORECORE=yes HUGETLB_PATH=/mnt/huge/ ./largepages3
memset page fault 1566471
memset tlb miss 453375
memset second tlb miss 453320
random access tlb miss 41636
random access second tlb miss 41637
vmx andrea # ./largepages3
memset page fault 1566642
memset tlb miss 453417
memset second tlb miss 453313
random access tlb miss 41630
random access second tlb miss 41647
vmx andrea # ./largepages3
memset page fault 1566872
memset tlb miss 453418
memset second tlb miss 453315
random access tlb miss 41618
random access second tlb miss 41659
vmx andrea # echo 0 > /proc/sys/vm/transparent_hugepage
vmx andrea # ./largepages3
memset page fault 2182476
memset tlb miss 460305
memset second tlb miss 460179
random access tlb miss 44483
random access second tlb miss 44186
vmx andrea # ./largepages3
memset page fault 2182791
memset tlb miss 460742
memset second tlb miss 459962
random access tlb miss 43981
random access second tlb miss 43988============
#include
#include
#include
#include#define SIZE (3UL*1024*1024*1024)
int main()
{
char *p = malloc(SIZE), *p2;
struct timeval before, after;gettimeofday(&before, NULL);
memset(p, 0, SIZE);
gettimeofday(&after, NULL);
printf("memset page fault %Lu\n",
(after.tv_sec-before.tv_sec)*1000000UL +
after.tv_usec-before.tv_usec);gettimeofday(&before, NULL);
memset(p, 0, SIZE);
gettimeofday(&after, NULL);
printf("memset tlb miss %Lu\n",
(after.tv_sec-before.tv_sec)*1000000UL +
after.tv_usec-before.tv_usec);gettimeofday(&before, NULL);
memset(p, 0, SIZE);
gettimeofday(&after, NULL);
printf("memset second tlb miss %Lu\n",
(after.tv_sec-before.tv_sec)*1000000UL +
after.tv_usec-before.tv_usec);gettimeofday(&before, NULL);
for (p2 = p; p2 < p+SIZE; p2 += 4096)
*p2 = 0;
gettimeofday(&after, NULL);
printf("random access tlb miss %Lu\n",
(after.tv_sec-before.tv_sec)*1000000UL +
after.tv_usec-before.tv_usec);gettimeofday(&before, NULL);
for (p2 = p; p2 < p+SIZE; p2 += 4096)
*p2 = 0;
gettimeofday(&after, NULL);
printf("random access second tlb miss %Lu\n",
(after.tv_sec-before.tv_sec)*1000000UL +
after.tv_usec-before.tv_usec);return 0;
}
============Signed-off-by: Andrea Arcangeli
Acked-by: Rik van Riel
Signed-off-by: Johannes Weiner
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds -
__get_user_pages gets a new 'nonblocking' parameter to signal that the
caller is prepared to re-acquire mmap_sem and retry the operation if
needed. This is used to split off long operations if they are going to
block on a disk transfer, or when we detect contention on the mmap_sem.[akpm@linux-foundation.org: remove ref to rwsem_is_contended()]
Signed-off-by: Michel Lespinasse
Cc: Hugh Dickins
Cc: Rik van Riel
Cc: Peter Zijlstra
Cc: Nick Piggin
Cc: KOSAKI Motohiro
Cc: Ingo Molnar
Cc: "H. Peter Anvin"
Cc: Thomas Gleixner
Cc: David Howells
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds
27 Oct, 2010
1 commit
-
page_order() is called by memory hotplug's user interface to check the
section is removable or not. (is_mem_section_removable())It calls page_order() withoug holding zone->lock.
So, even if the caller doesif (PageBuddy(page))
ret = page_order(page) ...
The caller may hit BUG_ON().For fixing this, there are 2 choices.
1. add zone->lock.
2. remove BUG_ON().is_mem_section_removable() is used for some "advice" and doesn't need to
be 100% accurate. This is_removable() can be called via user program..
We don't want to take this important lock for long by user's request. So,
this patch removes BUG_ON().Signed-off-by: KAMEZAWA Hiroyuki
Acked-by: Wu Fengguang
Acked-by: Michal Hocko
Acked-by: Mel Gorman
Cc:
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds
16 Dec, 2009
2 commits
-
In some use cases, user doesn't need extra filtering. E.g. user program
can inject errors through madvise syscall to its own pages, however it
might not know what the page state exactly is or which inode the page
belongs to.So introduce an one-off interface "corrupt-filter-enable".
Echo 0 to switch off page filters, and echo 1 to switch on the filters.
[AK: changed default to 0]Signed-off-by: Haicheng Li
Signed-off-by: Wu Fengguang
Signed-off-by: Andi Kleen -
The hwpoison test suite need to inject hwpoison to a collection of
selected task pages, and must not touch pages not owned by them and
thus kill important system processes such as init. (But it's OK to
mis-hwpoison free/unowned pages as well as shared clean pages.
Mis-hwpoison of shared dirty pages will kill all tasks, so the test
suite will target all or non of such tasks in the first place.)The memory cgroup serves this purpose well. We can put the target
processes under the control of a memory cgroup, and tell the hwpoison
injection code to only kill pages associated with some active memory
cgroup.The prerequisite for doing hwpoison stress tests with mem_cgroup is,
the mem_cgroup code tracks task pages _accurately_ (unless page is
locked). Which we believe is/should be true.The benefits are simplification of hwpoison injector code. Also the
mem_cgroup code will automatically be tested by hwpoison test cases.The alternative interfaces pin-pfn/unpin-pfn can also delegate the
(process and page flags) filtering functions reliably to user space.
However prototype implementation shows that this scheme adds more
complexity than we wanted.Example test case:
mkdir /cgroup/hwpoison
usemem -m 100 -s 1000 &
echo `jobs -p` > /cgroup/hwpoison/tasksmemcg_ino=$(ls -id /cgroup/hwpoison | cut -f1 -d' ')
echo $memcg_ino > /debug/hwpoison/corrupt-filter-memcgpage-types -p `pidof init` --hwpoison # shall do nothing
page-types -p `pidof usemem` --hwpoison # poison its pages[AK: Fix documentation]
[Add fix for problem noticed by Li Zefan ;
dentry in the css could be NULL]CC: KOSAKI Motohiro
CC: Hugh Dickins
CC: Daisuke Nishimura
CC: Balbir Singh
CC: KAMEZAWA Hiroyuki
CC: Li Zefan
CC: Paul Menage
CC: Nick Piggin
CC: Andi Kleen
Signed-off-by: Wu Fengguang
Signed-off-by: Andi Kleen
