There are three callers of task_numa_fault():
- do_huge_pmd_numa_page():
Accounts against the current node, not the node where the
page resides, unless we migrated, in which case it accounts
against the node we migrated to.
- do_numa_page():
Accounts against the current node, not the node where the
page resides, unless we migrated, in which case it accounts
against the node we migrated to.
- do_pmd_numa_page():
Accounts not at all when the page isn't migrated, otherwise
accounts against the node we migrated towards.
This seems wrong to me; all three sites should have the same
sementaics, furthermore we should accounts against where the page
really is, we already know where the task is.
So modify all three sites to always account; we did after all receive
the fault; and always account to where the page is after migration,
regardless of success.
They all still differ on when they clear the PTE/PMD; ideally that
would get sorted too.
Signed-off-by: Mel Gorman <mgorman@suse.de>
Reviewed-by: Rik van Riel <riel@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: <stable@kernel.org>
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1381141781-10992-8-git-send-email-mgorman@suse.de
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Commit 3812c8c8f3 ("mm: memcg: do not trap chargers with full
callstack on OOM") assumed that only a few places that can trigger a
memcg OOM situation do not return VM_FAULT_OOM, like optional page cache
readahead. But there are many more and it's impractical to annotate
them all.
First of all, we don't want to invoke the OOM killer when the failed
allocation is gracefully handled, so defer the actual kill to the end of
the fault handling as well. This simplifies the code quite a bit for
added bonus.
Second, since a failed allocation might not be the abrupt end of the
fault, the memcg OOM handler needs to be re-entrant until the fault
finishes for subsequent allocation attempts. If an allocation is
attempted after the task already OOMed, allow it to bypass the limit so
that it can quickly finish the fault and invoke the OOM killer.
Reported-by: azurIt <azurit@pobox.sk>
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The memcg OOM handling is incredibly fragile and can deadlock. When a
task fails to charge memory, it invokes the OOM killer and loops right
there in the charge code until it succeeds. Comparably, any other task
that enters the charge path at this point will go to a waitqueue right
then and there and sleep until the OOM situation is resolved. The problem
is that these tasks may hold filesystem locks and the mmap_sem; locks that
the selected OOM victim may need to exit.
For example, in one reported case, the task invoking the OOM killer was
about to charge a page cache page during a write(), which holds the
i_mutex. The OOM killer selected a task that was just entering truncate()
and trying to acquire the i_mutex:
OOM invoking task:
mem_cgroup_handle_oom+0x241/0x3b0
mem_cgroup_cache_charge+0xbe/0xe0
add_to_page_cache_locked+0x4c/0x140
add_to_page_cache_lru+0x22/0x50
grab_cache_page_write_begin+0x8b/0xe0
ext3_write_begin+0x88/0x270
generic_file_buffered_write+0x116/0x290
__generic_file_aio_write+0x27c/0x480
generic_file_aio_write+0x76/0xf0 # takes ->i_mutex
do_sync_write+0xea/0x130
vfs_write+0xf3/0x1f0
sys_write+0x51/0x90
system_call_fastpath+0x18/0x1d
OOM kill victim:
do_truncate+0x58/0xa0 # takes i_mutex
do_last+0x250/0xa30
path_openat+0xd7/0x440
do_filp_open+0x49/0xa0
do_sys_open+0x106/0x240
sys_open+0x20/0x30
system_call_fastpath+0x18/0x1d
The OOM handling task will retry the charge indefinitely while the OOM
killed task is not releasing any resources.
A similar scenario can happen when the kernel OOM killer for a memcg is
disabled and a userspace task is in charge of resolving OOM situations.
In this case, ALL tasks that enter the OOM path will be made to sleep on
the OOM waitqueue and wait for userspace to free resources or increase
the group's limit. But a userspace OOM handler is prone to deadlock
itself on the locks held by the waiting tasks. For example one of the
sleeping tasks may be stuck in a brk() call with the mmap_sem held for
writing but the userspace handler, in order to pick an optimal victim,
may need to read files from /proc/<pid>, which tries to acquire the same
mmap_sem for reading and deadlocks.
This patch changes the way tasks behave after detecting a memcg OOM and
makes sure nobody loops or sleeps with locks held:
1. When OOMing in a user fault, invoke the OOM killer and restart the
fault instead of looping on the charge attempt. This way, the OOM
victim can not get stuck on locks the looping task may hold.
2. When OOMing in a user fault but somebody else is handling it
(either the kernel OOM killer or a userspace handler), don't go to
sleep in the charge context. Instead, remember the OOMing memcg in
the task struct and then fully unwind the page fault stack with
-ENOMEM. pagefault_out_of_memory() will then call back into the
memcg code to check if the -ENOMEM came from the memcg, and then
either put the task to sleep on the memcg's OOM waitqueue or just
restart the fault. The OOM victim can no longer get stuck on any
lock a sleeping task may hold.
Debugged by Michal Hocko.
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Reported-by: azurIt <azurit@pobox.sk>
Acked-by: Michal Hocko <mhocko@suse.cz>
Cc: David Rientjes <rientjes@google.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Ben Tebulin reported:
"Since v3.7.2 on two independent machines a very specific Git
repository fails in 9/10 cases on git-fsck due to an SHA1/memory
failures. This only occurs on a very specific repository and can be
reproduced stably on two independent laptops. Git mailing list ran
out of ideas and for me this looks like some very exotic kernel issue"
and bisected the failure to the backport of commit 53a59fc67f ("mm:
limit mmu_gather batching to fix soft lockups on !CONFIG_PREEMPT").
That commit itself is not actually buggy, but what it does is to make it
much more likely to hit the partial TLB invalidation case, since it
introduces a new case in tlb_next_batch() that previously only ever
happened when running out of memory.
The real bug is that the TLB gather virtual memory range setup is subtly
buggered. It was introduced in commit 597e1c3580 ("mm/mmu_gather:
enable tlb flush range in generic mmu_gather"), and the range handling
was already fixed at least once in commit e6c495a96c ("mm: fix the TLB
range flushed when __tlb_remove_page() runs out of slots"), but that fix
was not complete.
The problem with the TLB gather virtual address range is that it isn't
set up by the initial tlb_gather_mmu() initialization (which didn't get
the TLB range information), but it is set up ad-hoc later by the
functions that actually flush the TLB. And so any such case that forgot
to update the TLB range entries would potentially miss TLB invalidates.
Rather than try to figure out exactly which particular ad-hoc range
setup was missing (I personally suspect it's the hugetlb case in
zap_huge_pmd(), which didn't have the same logic as zap_pte_range()
did), this patch just gets rid of the problem at the source: make the
TLB range information available to tlb_gather_mmu(), and initialize it
when initializing all the other tlb gather fields.
This makes the patch larger, but conceptually much simpler. And the end
result is much more understandable; even if you want to play games with
partial ranges when invalidating the TLB contents in chunks, now the
range information is always there, and anybody who doesn't want to
bother with it won't introduce subtle bugs.
Ben verified that this fixes his problem.
Reported-bisected-and-tested-by: Ben Tebulin <tebulin@googlemail.com>
Build-testing-by: Stephen Rothwell <sfr@canb.auug.org.au>
Build-testing-by: Richard Weinberger <richard.weinberger@gmail.com>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Cc: stable@vger.kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Andy Lutomirski reported that if a page with _PAGE_SOFT_DIRTY bit set
get swapped out, the bit is getting lost and no longer available when
pte read back.
To resolve this we introduce _PTE_SWP_SOFT_DIRTY bit which is saved in
pte entry for the page being swapped out. When such page is to be read
back from a swap cache we check for bit presence and if it's there we
clear it and restore the former _PAGE_SOFT_DIRTY bit back.
One of the problem was to find a place in pte entry where we can save
the _PTE_SWP_SOFT_DIRTY bit while page is in swap. The _PAGE_PSE was
chosen for that, it doesn't intersect with swap entry format stored in
pte.
Reported-by: Andy Lutomirski <luto@amacapital.net>
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Acked-by: Pavel Emelyanov <xemul@parallels.com>
Cc: Matt Mackall <mpm@selenic.com>
Cc: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@gmail.com>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: "Aneesh Kumar K.V" <aneesh.kumar@linux.vnet.ibm.com>
Reviewed-by: Minchan Kim <minchan@kernel.org>
Reviewed-by: Wanpeng Li <liwanp@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
These VM_<READfoo> macros aren't used very often and three of them
aren't used at all.
Expand the ones that are used in-place, and remove all the now unused
#define VM_<foo> macros.
VM_READHINTMASK, VM_NormalReadHint and VM_ClearReadHint were added just
before 2.4 and appears have never been used.
Signed-off-by: Joe Perches <joe@perches.com>
Acked-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Pull voluntary preemption fixes from Ingo Molnar:
"This tree contains a speedup which is achieved through better
might_sleep()/might_fault() preemption point annotations for uaccess
functions, by Michael S Tsirkin:
1. The only reason uaccess routines might sleep is if they fault.
Make this explicit for all architectures.
2. A voluntary preemption point in uaccess functions means compiler
can't inline them efficiently, this breaks assumptions that they
are very fast and small that e.g. net code seems to make. Remove
this preemption point so behaviour matches with what callers
assume.
3. Accesses (e.g through socket ops) to kernel memory with KERNEL_DS
like net/sunrpc does will never sleep. Remove an unconditinal
might_sleep() in the might_fault() inline in kernel.h (used when
PROVE_LOCKING is not set).
4. Accesses with pagefault_disable() return EFAULT but won't cause
caller to sleep. Check for that and thus avoid might_sleep() when
PROVE_LOCKING is set.
These changes offer a nice speedup for CONFIG_PREEMPT_VOLUNTARY=y
kernels, here's a network bandwidth measurement between a virtual
machine and the host:
before:
incoming: 7122.77 Mb/s
outgoing: 8480.37 Mb/s
after:
incoming: 8619.24 Mb/s [ +21.0% ]
outgoing: 9455.42 Mb/s [ +11.5% ]
I kept these changes in a separate tree, separate from scheduler
changes, because it's a mixed MM and scheduler topic"
* 'sched-mm-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
mm, sched: Allow uaccess in atomic with pagefault_disable()
mm, sched: Drop voluntary schedule from might_fault()
x86: uaccess s/might_sleep/might_fault/
tile: uaccess s/might_sleep/might_fault/
powerpc: uaccess s/might_sleep/might_fault/
mn10300: uaccess s/might_sleep/might_fault/
microblaze: uaccess s/might_sleep/might_fault/
m32r: uaccess s/might_sleep/might_fault/
frv: uaccess s/might_sleep/might_fault/
arm64: uaccess s/might_sleep/might_fault/
asm-generic: uaccess s/might_sleep/might_fault/
Since the introduction of preemptible mmu_gather TLB fast mode has been
broken. TLB fast mode relies on there being absolutely no concurrency;
it frees pages first and invalidates TLBs later.
However now we can get concurrency and stuff goes *bang*.
This patch removes all tlb_fast_mode() code; it was found the better
option vs trying to patch the hole by entangling tlb invalidation with
the scheduler.
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Russell King <linux@arm.linux.org.uk>
Cc: Tony Luck <tony.luck@intel.com>
Reported-by: Max Filippov <jcmvbkbc@gmail.com>
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This changes might_fault() so that it does not
trigger a false positive diagnostic for e.g. the following
sequence:
spin_lock_irqsave()
pagefault_disable()
copy_to_user()
pagefault_enable()
spin_unlock_irqrestore()
In particular vhost wants to do this, to call
socket ops from under a lock.
There are 3 cases to consider:
- CONFIG_PROVE_LOCKING - might_fault is non-inline
so it's easy to move the in_atomic test to fix
up the false positive warning.
- CONFIG_DEBUG_ATOMIC_SLEEP - might_fault
is currently inline, but we are calling a
non-inline __might_sleep anyway,
so let's use the non-line version of might_fault
that does the right thing.
- !CONFIG_DEBUG_ATOMIC_SLEEP && !CONFIG_PROVE_LOCKING
__might_sleep is a nop so might_fault is a nop.
Make this explicit.
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1369577426-26721-11-git-send-email-mst@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
might_fault() is called from functions like copy_to_user()
which most callers expect to be very fast, like a couple of
instructions.
So functions like memcpy_toiovec() call them many times in a loop.
But might_fault() calls might_sleep() and with CONFIG_PREEMPT_VOLUNTARY
this results in a function call.
Let's not do this - just call __might_sleep() that produces
a diagnostic for sleep within atomic, but drop
might_preempt().
Here's a test sending traffic between the VM and the host,
host is built with CONFIG_PREEMPT_VOLUNTARY:
before:
incoming: 7122.77 Mb/s
outgoing: 8480.37 Mb/s
after:
incoming: 8619.24 Mb/s
outgoing: 9455.42 Mb/s
As a side effect, this fixes an issue pointed
out by Ingo: might_fault might schedule differently
depending on PROVE_LOCKING. Now there's no
preemption point in both cases, so it's consistent.
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1369577426-26721-10-git-send-email-mst@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Currently the memory barrier in __do_huge_pmd_anonymous_page doesn't
work. Because lru_cache_add_lru uses pagevec so it could miss spinlock
easily so above rule was broken so user might see inconsistent data.
I was not first person who pointed out the problem. Mel and Peter
pointed out a few months ago and Peter pointed out further that even
spin_lock/unlock can't make sure of it:
http://marc.info/?t=134333512700004
In particular:
*A = a;
LOCK
UNLOCK
*B = b;
may occur as:
LOCK, STORE *B, STORE *A, UNLOCK
At last, Hugh pointed out that even we don't need memory barrier in
there because __SetPageUpdate already have done it from Nick's commit
0ed361dec3 ("mm: fix PageUptodate data race") explicitly.
So this patch fixes comment on THP and adds same comment for
do_anonymous_page, too because everybody except Hugh was missing that.
It means we need a comment about that.
Signed-off-by: Minchan Kim <minchan@kernel.org>
Acked-by: Andrea Arcangeli <aarcange@redhat.com>
Acked-by: David Rientjes <rientjes@google.com>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Hugh Dickins <hughd@google.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Use the new vsprintf extension to avoid any possible
message interleaving.
Signed-off-by: Joe Perches <joe@perches.com>
Acked-by: Christoph Lameter <cl@linux.com>
Signed-off-by: Jiri Kosina <jkosina@suse.cz>
Various drivers end up replicating the code to mmap() their memory
buffers into user space, and our core memory remapping function may be
very flexible but it is unnecessarily complicated for the common cases
to use.
Our internal VM uses pfn's ("page frame numbers") which simplifies
things for the VM, and allows us to pass physical addresses around in a
denser and more efficient format than passing a "phys_addr_t" around,
and having to shift it up and down by the page size. But it just means
that drivers end up doing that shifting instead at the interface level.
It also means that drivers end up mucking around with internal VM things
like the vma details (vm_pgoff, vm_start/end) way more than they really
need to.
So this just exports a function to map a certain physical memory range
into user space (using a phys_addr_t based interface that is much more
natural for a driver) and hides all the complexity from the driver.
Some drivers will still end up tweaking the vm_page_prot details for
things like prefetching or cacheability etc, but that's actually
relevant to the driver, rather than caring about what the page offset of
the mapping is into the particular IO memory region.
Acked-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
