kvmalloc warned about incompatible gfp_mask to catch abusers (mostly
GFP_NOFS) with an intention that this will motivate authors of the code
to fix those. Linus argues that this just motivates people to do even
more hacks like
if (gfp == GFP_KERNEL)
kvmalloc
else
kmalloc
I haven't seen this happening much (Linus pointed to bucket_lock special
cases an atomic allocation but my git foo hasn't found much more) but it
is true that we can grow those in future. Therefore Linus suggested to
simply not fallback to vmalloc for incompatible gfp flags and rather
stick with the kmalloc path.
Link: http://lkml.kernel.org/r/20180601115329.27807-1-mhocko@kernel.org
Signed-off-by: Michal Hocko <mhocko@suse.com>
Suggested-by: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Tom Herbert <tom@quantonium.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
In __alloc_pages_slowpath() we reset zonelist and preferred_zoneref for
allocations that can ignore memory policies. The zonelist is obtained
from current CPU's node. This is a problem for __GFP_THISNODE
allocations that want to allocate on a different node, e.g. because the
allocating thread has been migrated to a different CPU.
This has been observed to break SLAB in our 4.4-based kernel, because
there it relies on __GFP_THISNODE working as intended. If a slab page
is put on wrong node's list, then further list manipulations may corrupt
the list because page_to_nid() is used to determine which node's
list_lock should be locked and thus we may take a wrong lock and race.
Current SLAB implementation seems to be immune by luck thanks to commit
511e3a0588 ("mm/slab: make cache_grow() handle the page allocated on
arbitrary node") but there may be others assuming that __GFP_THISNODE
works as promised.
We can fix it by simply removing the zonelist reset completely. There
is actually no reason to reset it, because memory policies and cpusets
don't affect the zonelist choice in the first place. This was different
when commit 183f6371aa ("mm: ignore mempolicies when using
ALLOC_NO_WATERMARK") introduced the code, as mempolicies provided their
own restricted zonelists.
We might consider this for 4.17 although I don't know if there's
anything currently broken.
SLAB is currently not affected, but in kernels older than 4.7 that don't
yet have 511e3a0588 ("mm/slab: make cache_grow() handle the page
allocated on arbitrary node") it is. That's at least 4.4 LTS. Older
ones I'll have to check.
So stable backports should be more important, but will have to be
reviewed carefully, as the code went through many changes. BTW I think
that also the ac->preferred_zoneref reset is currently useless if we
don't also reset ac->nodemask from a mempolicy to NULL first (which we
probably should for the OOM victims etc?), but I would leave that for a
separate patch.
Link: http://lkml.kernel.org/r/20180525130853.13915-1-vbabka@suse.cz
Signed-off-by: Vlastimil Babka <vbabka@suse.cz>
Fixes: 183f6371aa ("mm: ignore mempolicies when using ALLOC_NO_WATERMARK")
Acked-by: Mel Gorman <mgorman@techsingularity.net>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
If a process monitored with userfaultfd changes it's memory mappings or
forks() at the same time as uffd monitor fills the process memory with
UFFDIO_COPY, the actual creation of page table entries and copying of
the data in mcopy_atomic may happen either before of after the memory
mapping modifications and there is no way for the uffd monitor to
maintain consistent view of the process memory layout.
For instance, let's consider fork() running in parallel with
userfaultfd_copy():
process | uffd monitor
---------------------------------+------------------------------
fork() | userfaultfd_copy()
... | ...
dup_mmap() | down_read(mmap_sem)
down_write(mmap_sem) | /* create PTEs, copy data */
dup_uffd() | up_read(mmap_sem)
copy_page_range() |
up_write(mmap_sem) |
dup_uffd_complete() |
/* notify monitor */ |
If the userfaultfd_copy() takes the mmap_sem first, the new page(s) will
be present by the time copy_page_range() is called and they will appear
in the child's memory mappings. However, if the fork() is the first to
take the mmap_sem, the new pages won't be mapped in the child's address
space.
If the pages are not present and child tries to access them, the monitor
will get page fault notification and everything is fine. However, if
the pages *are present*, the child can access them without uffd
noticing. And if we copy them into child it'll see the wrong data.
Since we are talking about background copy, we'd need to decide whether
the pages should be copied or not regardless #PF notifications.
Since userfaultfd monitor has no way to determine what was the order,
let's disallow userfaultfd_copy in parallel with the non-cooperative
events. In such case we return -EAGAIN and the uffd monitor can
understand that userfaultfd_copy() clashed with a non-cooperative event
and take an appropriate action.
Link: http://lkml.kernel.org/r/1527061324-19949-1-git-send-email-rppt@linux.vnet.ibm.com
Signed-off-by: Mike Rapoport <rppt@linux.vnet.ibm.com>
Acked-by: Pavel Emelyanov <xemul@virtuozzo.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Andrei Vagin <avagin@virtuozzo.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Currently an attempt to set swap.max into a value lower than the actual
swap usage fails, which causes configuration problems as there's no way
of lowering the configuration below the current usage short of turning
off swap entirely. This makes swap.max difficult to use and allows
delegatees to lock the delegator out of reducing swap allocation.
This patch updates swap_max_write() so that the limit can be lowered
below the current usage. It doesn't implement active reclaiming of swap
entries for the following reasons.
* mem_cgroup_swap_full() already tells the swap machinary to
aggressively reclaim swap entries if the usage is above 50% of
limit, so simply lowering the limit automatically triggers gradual
reclaim.
* Forcing back swapped out pages is likely to heavily impact the
workload and mess up the working set. Given that swap usually is a
lot less valuable and less scarce, letting the existing usage
dissipate over time through the above gradual reclaim and as they're
falted back in is likely the better behavior.
Link: http://lkml.kernel.org/r/20180523185041.GR1718769@devbig577.frc2.facebook.com
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Roman Gushchin <guro@fb.com>
Acked-by: Rik van Riel <riel@surriel.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Shaohua Li <shli@fb.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Patch series "Rearrange struct page", v6.
As presented at LSFMM, this patch-set rearranges struct page to give
more contiguous usable space to users who have allocated a struct page
for their own purposes. For a graphical view of before-and-after, see
the first two tabs of
https://docs.google.com/spreadsheets/d/1tvCszs_7FXrjei9_mtFiKV6nW1FLnYyvPvW-qNZhdog/edit?usp=sharing
Highlights:
- deferred_list now really exists in struct page instead of just a comment.
- hmm_data also exists in struct page instead of being a nasty hack.
- x86's PGD pages have a real pointer to the mm_struct.
- VMalloc pages now have all sorts of extra information stored in them
to help with debugging and tuning.
- rcu_head is no longer tied to slab in case anyone else wants to
free pages by RCU.
- slub's counters no longer share space with _refcount.
- slub's freelist+counters are now naturally dword aligned.
- slub loses a parameter to a lot of functions and a sysfs file.
This patch (of 17):
s390 borrows the storage used for _mapcount in struct page in order to
account whether the bottom or top half is being used for 2kB page tables.
I want to use that for something else, so use the top byte of _refcount
instead of the bottom byte of _mapcount. _refcount may temporarily be
incremented by other CPUs that see a stale pointer to this page in the
page cache, but each CPU can only increment it by one, and there are no
systems with 2^24 CPUs today, so they will not change the upper byte of
_refcount. We do have to be a little careful not to lose any of their
writes (as they will subsequently decrement the counter).
Link: http://lkml.kernel.org/r/20180518194519.3820-2-willy@infradead.org
Signed-off-by: Matthew Wilcox <mawilcox@microsoft.com>
Acked-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: "Kirill A . Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Lai Jiangshan <jiangshanlai@gmail.com>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Jérôme Glisse <jglisse@redhat.com>
Cc: Randy Dunlap <rdunlap@infradead.org>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This is to take better advantage of general huge page clearing
optimization (commit c79b57e462: "mm: hugetlb: clear target sub-page
last when clearing huge page") for hugetlbfs.
In the general optimization patch, the sub-page to access will be
cleared last to avoid the cache lines of to access sub-page to be
evicted when clearing other sub-pages. This works better if we have the
address of the sub-page to access, that is, the fault address inside the
huge page. So the hugetlbfs no page fault handler is changed to pass
that information. This will benefit workloads which don't access the
begin of the hugetlbfs huge page after the page fault under heavy cache
contention for shared last level cache.
The patch is a generic optimization which should benefit quite some
workloads, not for a specific use case. To demonstrate the performance
benefit of the patch, we tested it with vm-scalability run on hugetlbfs.
With this patch, the throughput increases ~28.1% in vm-scalability
anon-w-seq test case with 88 processes on a 2 socket Xeon E5 2699 v4
system (44 cores, 88 threads). The test case creates 88 processes, each
process mmaps a big anonymous memory area with MAP_HUGETLB and writes to
it from the end to the begin. For each process, other processes could
be seen as other workload which generates heavy cache pressure. At the
same time, the cache miss rate reduced from ~36.3% to ~25.6%, the IPC
(instruction per cycle) increased from 0.3 to 0.37, and the time spent
in user space is reduced ~19.3%.
Link: http://lkml.kernel.org/r/20180517083539.9242-1-ying.huang@intel.com
Signed-off-by: "Huang, Ying" <ying.huang@intel.com>
Reviewed-by: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Andi Kleen <andi.kleen@intel.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Shaohua Li <shli@fb.com>
Cc: Christopher Lameter <cl@linux.com>
Cc: "Aneesh Kumar K.V" <aneesh.kumar@linux.vnet.ibm.com>
Cc: Punit Agrawal <punit.agrawal@arm.com>
Cc: Anshuman Khandual <khandual@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Memory controller implements the memory.low best-effort memory
protection mechanism, which works perfectly in many cases and allows
protecting working sets of important workloads from sudden reclaim.
But its semantics has a significant limitation: it works only as long as
there is a supply of reclaimable memory. This makes it pretty useless
against any sort of slow memory leaks or memory usage increases. This
is especially true for swapless systems. If swap is enabled, memory
soft protection effectively postpones problems, allowing a leaking
application to fill all swap area, which makes no sense. The only
effective way to guarantee the memory protection in this case is to
invoke the OOM killer.
It's possible to handle this case in userspace by reacting on MEMCG_LOW
events; but there is still a place for a fail-safe in-kernel mechanism
to provide stronger guarantees.
This patch introduces the memory.min interface for cgroup v2 memory
controller. It works very similarly to memory.low (sharing the same
hierarchical behavior), except that it's not disabled if there is no
more reclaimable memory in the system.
If cgroup is not populated, its memory.min is ignored, because otherwise
even the OOM killer wouldn't be able to reclaim the protected memory,
and the system can stall.
[guro@fb.com: s/low/min/ in docs]
Link: http://lkml.kernel.org/r/20180510130758.GA9129@castle.DHCP.thefacebook.com
Link: http://lkml.kernel.org/r/20180509180734.GA4856@castle.DHCP.thefacebook.com
Signed-off-by: Roman Gushchin <guro@fb.com>
Reviewed-by: Randy Dunlap <rdunlap@infradead.org>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Tejun Heo <tj@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
While revisiting my Btrfs swapfile series [1], I introduced a situation
in which reclaim would lock i_rwsem, and even though the swapon() path
clearly made GFP_KERNEL allocations while holding i_rwsem, I got no
complaints from lockdep. It turns out that the rework of the fs_reclaim
annotation was broken: if the current task has PF_MEMALLOC set, we don't
acquire the dummy fs_reclaim lock, but when reclaiming we always check
this _after_ we've just set the PF_MEMALLOC flag. In most cases, we can
fix this by moving the fs_reclaim_{acquire,release}() outside of the
memalloc_noreclaim_{save,restore}(), althought kswapd is slightly
different. After applying this, I got the expected lockdep splats.
1: https://lwn.net/Articles/625412/
Link: http://lkml.kernel.org/r/9f8aa70652a98e98d7c4de0fc96a4addcee13efe.1523778026.git.osandov@fb.com
Fixes: d92a8cfcb3 ("locking/lockdep: Rework FS_RECLAIM annotation")
Signed-off-by: Omar Sandoval <osandov@fb.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Cc: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Since tmpfs THP was supported in 4.8, hugetlbfs is not the only
filesystem with huge page support anymore. tmpfs can use huge page via
THP when mounting by "huge=" mount option.
When applications use huge page on hugetlbfs, it just need check the
filesystem magic number, but it is not enough for tmpfs. Make
stat.st_blksize return huge page size if it is mounted by appropriate
"huge=" option to give applications a hint to optimize the behavior with
THP.
Some applications may not do wisely with THP. For example, QEMU may
mmap file on non huge page aligned hint address with MAP_FIXED, which
results in no pages are PMD mapped even though THP is used. Some
applications may mmap file with non huge page aligned offset. Both
behaviors make THP pointless.
statfs.f_bsize still returns 4KB for tmpfs since THP could be split, and
it also may fallback to 4KB page silently if there is not enough huge
page. Furthermore, different f_bsize makes max_blocks and free_blocks
calculation harder but without too much benefit. Returning huge page
size via stat.st_blksize sounds good enough.
Since PUD size huge page for THP has not been supported, now it just
returns HPAGE_PMD_SIZE.
Hugh said:
: Sorry, I have no enthusiasm for this patch; but do I feel strongly
: enough to override you and everyone else to NAK it? No, I don't feel
: that strongly, maybe st_blksize isn't worth arguing over.
:
: We did look at struct stat when designing huge tmpfs, to see if there
: were any fields that should be adjusted for it; but concluded none.
: Yes, it would sometimes be nice to have a quickly accessible indicator
: for when tmpfs has been mounted huge (scanning /proc/mounts for options
: can be tiresome, agreed); but since tmpfs tries to supply huge (or not)
: pages transparently, no difference seemed right.
:
: So, because st_blksize is a not very useful field of struct stat, with
: "size" in the name, we're going to put HPAGE_PMD_SIZE in there instead
: of PAGE_SIZE, if the tmpfs was mounted with one of the huge "huge"
: options (force or always, okay; within_size or advise, not so much).
: Though HPAGE_PMD_SIZE is no more its "preferred I/O size" or "blocksize
: for file system I/O" than PAGE_SIZE was.
:
: Which we can expect to speed up some applications and disadvantage
: others, depending on how they interpret st_blksize: just like if we
: changed it in the same way on non-huge tmpfs. (Did I actually try
: changing st_blksize early on, and find it broke something? If so, I've
: now forgotten what, and a search through commit messages didn't find
: it; but I guess we'll find out soon enough.)
:
: If there were an mstat() syscall, returning a field "preferred
: alignment", then we could certainly agree to put HPAGE_PMD_SIZE in
: there; but in stat()'s st_blksize? And what happens when (in future)
: mm maps this or that hard-disk filesystem's blocks with a pmd mapping -
: should that filesystem then advertise a bigger st_blksize, despite the
: same disk layout as before? What happens with DAX?
:
: And this change is not going to help the QEMU suboptimality that
: brought you here (or does QEMU align mmaps according to st_blksize?).
: QEMU ought to work well with kernels without this change, and kernels
: with this change; and I hope it can easily deal with both by avoiding
: that use of MAP_FIXED which prevented the kernel's intended alignment.
[akpm@linux-foundation.org: remove unneeded `else']
Link: http://lkml.kernel.org/r/1524665633-83806-1-git-send-email-yang.shi@linux.alibaba.com
Signed-off-by: Yang Shi <yang.shi@linux.alibaba.com>
Suggested-by: Christoph Hellwig <hch@infradead.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mmap_sem will be read locked when calling follow_pmd_mask(). But this
cannot prevent PMD from being changed for all cases when PTL is
unlocked, for example, from pmd_trans_huge() to pmd_none() via
MADV_DONTNEED. So it is possible for the pmd_present() check in
follow_pmd_mask() to encounter an invalid PMD. This may cause an
incorrect VM_BUG_ON() or an infinite loop. Fix this by reading the PMD
entry into a local variable with READ_ONCE() and checking the local
variable and pmd_none() in the retry loop.
As Kirill pointed out, with PTL unlocked, the *pmd may be changed under
us, so reading it directly again and again may incur weird bugs. So
although using *pmd directly other than for pmd_present() checking may
be safe, it is still better to replace them to read *pmd once and check
the local variable multiple times.
When PTL unlocked, replace all *pmd with local variable was suggested by
Kirill.
Link: http://lkml.kernel.org/r/20180419083514.1365-1-ying.huang@intel.com
Signed-off-by: "Huang, Ying" <ying.huang@intel.com>
Reviewed-by: Zi Yan <zi.yan@cs.rutgers.edu>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: "Aneesh Kumar K.V" <aneesh.kumar@linux.vnet.ibm.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
