openwrt/staging/blogic.git
7 years agopowerpc/mm/autonuma: switch ppc64 to its own implementation of saved write
Aneesh Kumar K.V [Fri, 24 Feb 2017 22:59:21 +0000 (14:59 -0800)]
powerpc/mm/autonuma: switch ppc64 to its own implementation of saved write

With this our protnone becomes a present pte with READ/WRITE/EXEC bit
cleared.  By default we also set _PAGE_PRIVILEGED on such pte.  This is
now used to help us identify a protnone pte that as saved write bit.
For such pte, we will clear the _PAGE_PRIVILEGED bit.  The pte still
remain non-accessible from both user and kernel.

[aneesh.kumar@linux.vnet.ibm.com: v3]
Link: http://lkml.kernel.org/r/1487498625-10891-4-git-send-email-aneesh.kumar@linux.vnet.ibm.com
Link: http://lkml.kernel.org/r/1487050314-3892-3-git-send-email-aneesh.kumar@linux.vnet.ibm.com
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Acked-by: Michael Neuling <mikey@neuling.org>
Cc: Rik van Riel <riel@surriel.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Michael Ellerman <michaele@au1.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm/ksm: handle protnone saved writes when making page write protect
Aneesh Kumar K.V [Fri, 24 Feb 2017 22:59:19 +0000 (14:59 -0800)]
mm/ksm: handle protnone saved writes when making page write protect

Without this KSM will consider the page write protected, but a numa
fault can later mark the page writable.  This can result in memory
corruption.

Link: http://lkml.kernel.org/r/1487498625-10891-3-git-send-email-aneesh.kumar@linux.vnet.ibm.com
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm/autonuma: let architecture override how the write bit should be stashed in a protn...
Aneesh Kumar K.V [Fri, 24 Feb 2017 22:59:16 +0000 (14:59 -0800)]
mm/autonuma: let architecture override how the write bit should be stashed in a protnone pte.

Patch series "Numabalancing preserve write fix", v2.

This patch series address an issue w.r.t THP migration and autonuma
preserve write feature.  migrate_misplaced_transhuge_page() cannot deal
with concurrent modification of the page.  It does a page copy without
following the migration pte sequence.  IIUC, this was done to keep the
migration simpler and at the time of implemenation we didn't had THP
page cache which would have required a more elaborate migration scheme.
That means thp autonuma migration expect the protnone with saved write
to be done such that both kernel and user cannot update the page
content.  This patch series enables archs like ppc64 to do that.  We are
good with the hash translation mode with the current code, because we
never create a hardware page table entry for a protnone pte.

This patch (of 2):

Autonuma preserves the write permission across numa fault to avoid
taking a writefault after a numa fault (Commit: b191f9b106ea " mm: numa:
preserve PTE write permissions across a NUMA hinting fault").
Architecture can implement protnone in different ways and some may
choose to implement that by clearing Read/ Write/Exec bit of pte.
Setting the write bit on such pte can result in wrong behaviour.  Fix
this up by allowing arch to override how to save the write bit on a
protnone pte.

[aneesh.kumar@linux.vnet.ibm.com: don't mark pte saved write in case of dirty_accountable]
Link: http://lkml.kernel.org/r/1487942884-16517-1-git-send-email-aneesh.kumar@linux.vnet.ibm.com
[aneesh.kumar@linux.vnet.ibm.com: v3]
Link: http://lkml.kernel.org/r/1487498625-10891-2-git-send-email-aneesh.kumar@linux.vnet.ibm.com
Link: http://lkml.kernel.org/r/1487050314-3892-2-git-send-email-aneesh.kumar@linux.vnet.ibm.com
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Acked-by: Michael Neuling <mikey@neuling.org>
Cc: Rik van Riel <riel@surriel.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Michael Ellerman <michaele@au1.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm/autonuma: don't use set_pte_at when updating protnone ptes
Aneesh Kumar K.V [Fri, 24 Feb 2017 22:59:13 +0000 (14:59 -0800)]
mm/autonuma: don't use set_pte_at when updating protnone ptes

Architectures like ppc64, use privilege access bit to mark pte non
accessible.  This implies that kernel can do a copy_to_user to an
address marked for numa fault.  This also implies that there can be a
parallel hardware update for the pte.  set_pte_at cannot be used in such
scenarios.  Hence switch the pte update to use ptep_get_and_clear and
set_pte_at combination.

[akpm@linux-foundation.org: remove unwanted ppc change, per Aneesh]
Link: http://lkml.kernel.org/r/1486400776-28114-1-git-send-email-aneesh.kumar@linux.vnet.ibm.com
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Acked-by: Rik van Riel <riel@redhat.com>
Acked-by: Mel Gorman <mgorman@techsingularity.net>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm/shmem.c: fix unlikely() test of info->seals to test only for WRITE and GROW
Steven Rostedt (VMware) [Fri, 24 Feb 2017 22:59:10 +0000 (14:59 -0800)]
mm/shmem.c: fix unlikely() test of info->seals to test only for WRITE and GROW

Running my likely/unlikely profiler, I discovered that the test in
shmem_write_begin() that tests for info->seals as unlikely, is always
incorrect.  This is because shmem_get_inode() sets info->seals to have
F_SEAL_SEAL set by default, and it is unlikely to be cleared when
shmem_write_begin() is called.  Thus, the if statement is very likely.

But as the if statement block only cares about F_SEAL_WRITE and
F_SEAL_GROW, change the test to only test those two bits.

Link: http://lkml.kernel.org/r/20170203105656.7aec6237@gandalf.local.home
Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>
Acked-by: Hugh Dickins <hughd@google.com>
Cc: David Herrmann <dh.herrmann@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm, vmscan: clear PGDAT_WRITEBACK when zone is balanced
Mel Gorman [Fri, 24 Feb 2017 22:59:07 +0000 (14:59 -0800)]
mm, vmscan: clear PGDAT_WRITEBACK when zone is balanced

Hillf Danton pointed out that since commit 1d82de618dd ("mm, vmscan:
make kswapd reclaim in terms of nodes") that PGDAT_WRITEBACK is no
longer cleared.

It was not noticed as triggering it requires pages under writeback to
cycle twice through the LRU and before kswapd gets stalled.
Historically, such issues tended to occur on small machines writing
heavily to slow storage such as a USB stick.

Once kswapd stalls, direct reclaim stalls may be higher but due to the
fact that memory pressure is required, it would not be very noticable.

Michal Hocko suggested removing the flag entirely but the conservative
fix is to restore the intended PGDAT_WRITEBACK behaviour and clear the
flag when a suitable zone is balanced.

Fixes: 1d82de618ddd ("mm, vmscan: make kswapd reclaim in terms of nodes")
Link: http://lkml.kernel.org/r/20170203203222.gq7hk66yc36lpgtb@suse.de
Signed-off-by: Mel Gorman <mgorman@suse.de>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Hillf Danton <hillf.zj@alibaba-inc.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agodrm: remove unnecessary fault wrappers
Ross Zwisler [Fri, 24 Feb 2017 22:59:04 +0000 (14:59 -0800)]
drm: remove unnecessary fault wrappers

The fault wrappers drm_vm_fault(), drm_vm_shm_fault(),
drm_vm_dma_fault() and drm_vm_sg_fault() used to provide extra logic
beyond what was in the "drm_do_*" versions of these functions, but as of
commit ca0b07d9a969 ("drm: convert drm from nopage to fault") they are
just unnecessary wrappers that do nothing.

Remove them, and rename the the drm_do_* fault handlers to remove the
"do_" since they no longer have corresponding wrappers.

Link: http://lkml.kernel.org/r/1486155698-25717-1-git-send-email-ross.zwisler@linux.intel.com
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Sean Paul <seanpaul@chromium.org>
Acked-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: Dave Jiang <dave.jiang@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm: codgin-style fixes
Tobin C Harding [Fri, 24 Feb 2017 22:59:01 +0000 (14:59 -0800)]
mm: codgin-style fixes

Fix whitespace issues, extraneous braces.

Link: http://lkml.kernel.org/r/1485992240-10986-5-git-send-email-me@tobin.cc
Signed-off-by: Tobin C Harding <me@tobin.cc>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm/memory.c: use NULL instead of literal 0
Tobin C Harding [Fri, 24 Feb 2017 22:58:59 +0000 (14:58 -0800)]
mm/memory.c: use NULL instead of literal 0

Patch fixes sparse warning: Using plain integer as NULL pointer.
Replaces assignment of 0 to pointer with NULL assignment.

Link: http://lkml.kernel.org/r/1485992240-10986-2-git-send-email-me@tobin.cc
Signed-off-by: Tobin C Harding <me@tobin.cc>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm/page_alloc.c: remove duplicate inclusion of page_ext.h
Masanari Iida [Fri, 24 Feb 2017 22:58:56 +0000 (14:58 -0800)]
mm/page_alloc.c: remove duplicate inclusion of page_ext.h

Link: http://lkml.kernel.org/r/20170202011942.1609-1-standby24x7@gmail.com
Signed-off-by: Masanari Iida <standby24x7@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agovmalloc: back off when the current task is killed
Michal Hocko [Fri, 24 Feb 2017 22:58:53 +0000 (14:58 -0800)]
vmalloc: back off when the current task is killed

__vmalloc_area_node() allocates pages to cover the requested vmalloc
size.  This can be a lot of memory.  If the current task is killed by
the OOM killer, and thus has an unlimited access to memory reserves, it
can consume all the memory theoretically.  Fix this by checking for
fatal_signal_pending and back off early.

Link: http://lkml.kernel.org/r/20170201092706.9966-4-mhocko@kernel.org
Signed-off-by: Michal Hocko <mhocko@suse.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm: cma: print allocation failure reason and bitmap status
Jaewon Kim [Fri, 24 Feb 2017 22:58:50 +0000 (14:58 -0800)]
mm: cma: print allocation failure reason and bitmap status

There are many reasons of CMA allocation failure such as EBUSY, ENOMEM,
EINTR.  But we did not know error reason so far.  This patch prints the
error value.

Additionally if CONFIG_CMA_DEBUG is enabled, this patch shows bitmap
status to know available pages.  Actually CMA internally tries on all
available regions because some regions can be failed because of EBUSY.
Bitmap status is useful to know in detail on both ENONEM and EBUSY;

 ENOMEM: not tried at all because of no available region
         it could be too small total region or could be fragmentation issue
 EBUSY:  tried some region but all failed

This is an ENOMEM example with this patch.

    [2:   Binder:714_1:  744] cma: cma_alloc: alloc failed, req-size: 256 pages, ret: -12

If CONFIG_CMA_DEBUG is enabled, avabile pages also will be shown as
concatenated size@position format.  So 4@572 means that there are 4
available pages at 572 position starting from 0 position.

    [2:   Binder:714_1:  744] cma: number of available pages: 4@572+7@585+7@601+8@632+38@730+166@1114+127@1921=> 357 free of 2048 total pages

Link: http://lkml.kernel.org/r/1485909785-3952-1-git-send-email-jaewon31.kim@samsung.com
Signed-off-by: Jaewon Kim <jaewon31.kim@samsung.com>
Acked-by: Michal Nazarewicz <mina86@mina86.com>
Cc: Laura Abbott <labbott@redhat.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm, madvise: fail with ENOMEM when splitting vma will hit max_map_count
David Rientjes [Fri, 24 Feb 2017 22:58:47 +0000 (14:58 -0800)]
mm, madvise: fail with ENOMEM when splitting vma will hit max_map_count

If madvise(2) advice will result in the underlying vma being split and
the number of areas mapped by the process will exceed
/proc/sys/vm/max_map_count as a result, return ENOMEM instead of EAGAIN.

EAGAIN is returned by madvise(2) when a kernel resource, such as slab,
is temporarily unavailable.  It indicates that userspace should retry
the advice in the near future.  This is important for advice such as
MADV_DONTNEED which is often used by malloc implementations to free
memory back to the system: we really do want to free memory back when
madvise(2) returns EAGAIN because slab allocations (for vmas, anon_vmas,
or mempolicies) cannot be allocated.

Encountering /proc/sys/vm/max_map_count is not a temporary failure,
however, so return ENOMEM to indicate this is a more serious issue.  A
followup patch to the man page will specify this behavior.

Link: http://lkml.kernel.org/r/alpine.DEB.2.10.1701241431120.42507@chino.kir.corp.google.com
Signed-off-by: David Rientjes <rientjes@google.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Jerome Marchand <jmarchan@redhat.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Michael Kerrisk <mtk.manpages@googlemail.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>
7 years agomm: wire up GFP flag passing in dma_alloc_from_contiguous
Lucas Stach [Fri, 24 Feb 2017 22:58:44 +0000 (14:58 -0800)]
mm: wire up GFP flag passing in dma_alloc_from_contiguous

The callers of the DMA alloc functions already provide the proper
context GFP flags.  Make sure to pass them through to the CMA allocator,
to make the CMA compaction context aware.

Link: http://lkml.kernel.org/r/20170127172328.18574-3-l.stach@pengutronix.de
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Radim Krcmar <rkrcmar@redhat.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Cc: Chris Zankel <chris@zankel.net>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Alexander Graf <agraf@suse.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm: cma_alloc: allow to specify GFP mask
Lucas Stach [Fri, 24 Feb 2017 22:58:41 +0000 (14:58 -0800)]
mm: cma_alloc: allow to specify GFP mask

Most users of this interface just want to use it with the default
GFP_KERNEL flags, but for cases where DMA memory is allocated it may be
called from a different context.

No functional change yet, just passing through the flag to the
underlying alloc_contig_range function.

Link: http://lkml.kernel.org/r/20170127172328.18574-2-l.stach@pengutronix.de
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Radim Krcmar <rkrcmar@redhat.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Cc: Chris Zankel <chris@zankel.net>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Alexander Graf <agraf@suse.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm: alloc_contig_range: allow to specify GFP mask
Lucas Stach [Fri, 24 Feb 2017 22:58:37 +0000 (14:58 -0800)]
mm: alloc_contig_range: allow to specify GFP mask

Currently alloc_contig_range assumes that the compaction should be done
with the default GFP_KERNEL flags.  This is probably right for all
current uses of this interface, but may change as CMA is used in more
use-cases (including being the default DMA memory allocator on some
platforms).

Change the function prototype, to allow for passing through the GFP mask
set by upper layers.

Also respect global restrictions by applying memalloc_noio_flags to the
passed in flags.

Link: http://lkml.kernel.org/r/20170127172328.18574-1-l.stach@pengutronix.de
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Radim Krcmar <rkrcmar@redhat.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Cc: Chris Zankel <chris@zankel.net>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Alexander Graf <agraf@suse.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agouserfaultfd: documentation update
Mike Rapoport [Fri, 24 Feb 2017 22:58:34 +0000 (14:58 -0800)]
userfaultfd: documentation update

Add documentation about new userfaultfd features and events

Link: http://lkml.kernel.org/r/1487716431-5551-1-git-send-email-rppt@linux.vnet.ibm.com
Signed-off-by: Mike Rapoport <rppt@linux.vnet.ibm.com>
Reviewed-by: Andrea Arcangeli <aarcange@redhat.com>
Cc: "Dr. David Alan Gilbert" <dgilbert@redhat.com>
Cc: Hillf Danton <hillf.zj@alibaba-inc.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Pavel Emelyanov <xemul@virtuozzo.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agouserfaultfd_copy: return -ENOSPC in case mm has gone
Mike Rapoport [Fri, 24 Feb 2017 22:58:31 +0000 (14:58 -0800)]
userfaultfd_copy: return -ENOSPC in case mm has gone

In the non-cooperative userfaultfd case, the process exit may race with
outstanding mcopy_atomic called by the uffd monitor.  Returning -ENOSPC
instead of -EINVAL when mm is already gone will allow uffd monitor to
distinguish this case from other error conditions.

Link: http://lkml.kernel.org/r/1485542673-24387-6-git-send-email-rppt@linux.vnet.ibm.com
Signed-off-by: Mike Rapoport <rppt@linux.vnet.ibm.com>
Acked-by: Hillf Danton <hillf.zj@alibaba-inc.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: "Dr. David Alan Gilbert" <dgilbert@redhat.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Pavel Emelyanov <xemul@virtuozzo.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agouserfaultfd: mcopy_atomic: return -ENOENT when no compatible VMA found
Mike Rapoport [Fri, 24 Feb 2017 22:58:28 +0000 (14:58 -0800)]
userfaultfd: mcopy_atomic: return -ENOENT when no compatible VMA found

The memory mapping of a process may change between #PF event and the
call to mcopy_atomic that comes to resolve the page fault.  In such
case, there will be no VMA covering the range passed to mcopy_atomic or
the VMA will not have userfaultfd context.

To allow uffd monitor to distinguish those case from other errors, let's
return -ENOENT instead of -EINVAL.

Note, that despite availability of UFFD_EVENT_UNMAP there still might be
race between the processing of UFFD_EVENT_UNMAP and outstanding
mcopy_atomic in case of non-cooperative uffd usage.

[rppt@linux.vnet.ibm.com: update cases returning -ENOENT]
Link: http://lkml.kernel.org/r/20170207150249.GA6709@rapoport-lnx
[aarcange@redhat.com: merge fix]
[akpm@linux-foundation.org: fix the merge fix]
Link: http://lkml.kernel.org/r/1485542673-24387-5-git-send-email-rppt@linux.vnet.ibm.com
Signed-off-by: Mike Rapoport <rppt@linux.vnet.ibm.com>
Acked-by: Hillf Danton <hillf.zj@alibaba-inc.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: "Dr. David Alan Gilbert" <dgilbert@redhat.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Pavel Emelyanov <xemul@virtuozzo.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agouserfaultfd: non-cooperative: add event for exit() notification
Mike Rapoport [Fri, 24 Feb 2017 22:58:25 +0000 (14:58 -0800)]
userfaultfd: non-cooperative: add event for exit() notification

Allow userfaultfd monitor track termination of the processes that have
memory backed by the uffd.

[rppt@linux.vnet.ibm.com: add comment]
Link: http://lkml.kernel.org/r/20170202135448.GB19804@rapoport-lnxLink:
Signed-off-by: Mike Rapoport <rppt@linux.vnet.ibm.com>
Acked-by: Hillf Danton <hillf.zj@alibaba-inc.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: "Dr. David Alan Gilbert" <dgilbert@redhat.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Pavel Emelyanov <xemul@virtuozzo.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agouserfaultfd: non-cooperative: add event for memory unmaps
Mike Rapoport [Fri, 24 Feb 2017 22:58:22 +0000 (14:58 -0800)]
userfaultfd: non-cooperative: add event for memory unmaps

When a non-cooperative userfaultfd monitor copies pages in the
background, it may encounter regions that were already unmapped.
Addition of UFFD_EVENT_UNMAP allows the uffd monitor to track precisely
changes in the virtual memory layout.

Since there might be different uffd contexts for the affected VMAs, we
first should create a temporary representation for the unmap event for
each uffd context and then notify them one by one to the appropriate
userfault file descriptors.

The event notification occurs after the mmap_sem has been released.

[arnd@arndb.de: fix nommu build]
Link: http://lkml.kernel.org/r/20170203165141.3665284-1-arnd@arndb.de
[mhocko@suse.com: fix nommu build]
Link: http://lkml.kernel.org/r/20170202091503.GA22823@dhcp22.suse.cz
Link: http://lkml.kernel.org/r/1485542673-24387-3-git-send-email-rppt@linux.vnet.ibm.com
Signed-off-by: Mike Rapoport <rppt@linux.vnet.ibm.com>
Signed-off-by: Michal Hocko <mhocko@suse.com>
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Acked-by: Hillf Danton <hillf.zj@alibaba-inc.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: "Dr. David Alan Gilbert" <dgilbert@redhat.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Pavel Emelyanov <xemul@virtuozzo.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm: call vm_munmap in munmap syscall instead of using open coded version
Mike Rapoport [Fri, 24 Feb 2017 22:58:19 +0000 (14:58 -0800)]
mm: call vm_munmap in munmap syscall instead of using open coded version

Patch series "userfaultfd: non-cooperative: better tracking for mapping
changes", v2.

These patches try to address issues I've encountered during integration
of userfaultfd with CRIU.

Previously added userfaultfd events for fork(), madvise() and mremap()
unfortunately do not cover all possible changes to a process virtual
memory layout required for uffd monitor.

When one or more VMAs is removed from the process mm, the external uffd
monitor has no way to detect those changes and will attempt to fill the
removed regions with userfaultfd_copy.

Another problematic event is the exit() of the process.  Here again, the
external uffd monitor will try to use userfaultfd_copy, although mm
owning the memory has already gone.

The first patch in the series is a minor cleanup and it's not strictly
related to the rest of the series.

The patches 2 and 3 below add UFFD_EVENT_UNMAP and UFFD_EVENT_EXIT to
allow the uffd monitor track changes in the memory layout of a process.

The patches 4 and 5 amend error codes returned by userfaultfd_copy to
make the uffd monitor able to cope with races that might occur between
delivery of unmap and exit events and outstanding userfaultfd_copy's.

This patch (of 5):

Commit dc0ef0df7b6a ("mm: make mmap_sem for write waits killable for mm
syscalls") replaced call to vm_munmap in munmap syscall with open coded
version to allow different waits on mmap_sem in munmap syscall and
vm_munmap.

Now both functions use down_write_killable, so we can restore the call
to vm_munmap from the munmap system call.

Link: http://lkml.kernel.org/r/1485542673-24387-2-git-send-email-rppt@linux.vnet.ibm.com
Signed-off-by: Mike Rapoport <rppt@linux.vnet.ibm.com>
Acked-by: Hillf Danton <hillf.zj@alibaba-inc.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: "Dr. David Alan Gilbert" <dgilbert@redhat.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Pavel Emelyanov <xemul@virtuozzo.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm: convert remove_migration_pte() to use page_vma_mapped_walk()
Kirill A. Shutemov [Fri, 24 Feb 2017 22:58:16 +0000 (14:58 -0800)]
mm: convert remove_migration_pte() to use page_vma_mapped_walk()

remove_migration_pte() also can easily be converted to
page_vma_mapped_walk().

[akpm@linux-foundation.org: coding-style fixes]
Link: http://lkml.kernel.org/r/20170129173858.45174-13-kirill.shutemov@linux.intel.com
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hillf Danton <hillf.zj@alibaba-inc.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm: drop page_check_address{,_transhuge}
Kirill A. Shutemov [Fri, 24 Feb 2017 22:58:13 +0000 (14:58 -0800)]
mm: drop page_check_address{,_transhuge}

All users are gone. Let's drop them.

Link: http://lkml.kernel.org/r/20170129173858.45174-12-kirill.shutemov@linux.intel.com
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hillf Danton <hillf.zj@alibaba-inc.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm: convert page_mapped_in_vma() to use page_vma_mapped_walk()
Kirill A. Shutemov [Fri, 24 Feb 2017 22:58:10 +0000 (14:58 -0800)]
mm: convert page_mapped_in_vma() to use page_vma_mapped_walk()

For consistency, it worth converting all page_check_address() to
page_vma_mapped_walk(), so we could drop the former.

Link: http://lkml.kernel.org/r/20170129173858.45174-11-kirill.shutemov@linux.intel.com
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Acked-by: Hillf Danton <hillf.zj@alibaba-inc.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm, uprobes: convert __replace_page() to use page_vma_mapped_walk()
Kirill A. Shutemov [Fri, 24 Feb 2017 22:58:07 +0000 (14:58 -0800)]
mm, uprobes: convert __replace_page() to use page_vma_mapped_walk()

For consistency, it worth converting all page_check_address() to
page_vma_mapped_walk(), so we could drop the former.

Link: http://lkml.kernel.org/r/20170129173858.45174-10-kirill.shutemov@linux.intel.com
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Reviewed-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hillf Danton <hillf.zj@alibaba-inc.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm, ksm: convert write_protect_page() to use page_vma_mapped_walk()
Kirill A. Shutemov [Fri, 24 Feb 2017 22:58:04 +0000 (14:58 -0800)]
mm, ksm: convert write_protect_page() to use page_vma_mapped_walk()

For consistency, it worth converting all page_check_address() to
page_vma_mapped_walk(), so we could drop the former.

Link: http://lkml.kernel.org/r/20170129173858.45174-9-kirill.shutemov@linux.intel.com
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hillf Danton <hillf.zj@alibaba-inc.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm: convert try_to_unmap_one() to use page_vma_mapped_walk()
Kirill A. Shutemov [Fri, 24 Feb 2017 22:58:01 +0000 (14:58 -0800)]
mm: convert try_to_unmap_one() to use page_vma_mapped_walk()

For consistency, it worth converting all page_check_address() to
page_vma_mapped_walk(), so we could drop the former.

It also makes freeze_page() as we walk though rmap only once.

Link: http://lkml.kernel.org/r/20170129173858.45174-8-kirill.shutemov@linux.intel.com
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hillf Danton <hillf.zj@alibaba-inc.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm: convert page_mkclean_one() to use page_vma_mapped_walk()
Kirill A. Shutemov [Fri, 24 Feb 2017 22:57:57 +0000 (14:57 -0800)]
mm: convert page_mkclean_one() to use page_vma_mapped_walk()

For consistency, it worth converting all page_check_address() to
page_vma_mapped_walk(), so we could drop the former.

PMD handling here is future-proofing, we don't have users yet.  ext4
with huge pages will be the first.

Link: http://lkml.kernel.org/r/20170129173858.45174-7-kirill.shutemov@linux.intel.com
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hillf Danton <hillf.zj@alibaba-inc.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm, rmap: check all VMAs that PTE-mapped THP can be part of
Kirill A. Shutemov [Fri, 24 Feb 2017 22:57:54 +0000 (14:57 -0800)]
mm, rmap: check all VMAs that PTE-mapped THP can be part of

Current rmap code can miss a VMA that maps PTE-mapped THP if the first
suppage of the THP was unmapped from the VMA.

We need to walk rmap for the whole range of offsets that THP covers, not
only the first one.

vma_address() also need to be corrected to check the range instead of
the first subpage.

Link: http://lkml.kernel.org/r/20170129173858.45174-6-kirill.shutemov@linux.intel.com
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Acked-by: Hillf Danton <hillf.zj@alibaba-inc.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm: fix handling PTE-mapped THPs in page_idle_clear_pte_refs()
Kirill A. Shutemov [Fri, 24 Feb 2017 22:57:51 +0000 (14:57 -0800)]
mm: fix handling PTE-mapped THPs in page_idle_clear_pte_refs()

For PTE-mapped THP page_check_address_transhuge() is not adequate: it
cannot find all relevant PTEs, only the first one.i

Let's switch it to page_vma_mapped_walk().

I don't think it's subject for stable@: it's not fatal.

Link: http://lkml.kernel.org/r/20170129173858.45174-5-kirill.shutemov@linux.intel.com
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hillf Danton <hillf.zj@alibaba-inc.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm: fix handling PTE-mapped THPs in page_referenced()
Kirill A. Shutemov [Fri, 24 Feb 2017 22:57:48 +0000 (14:57 -0800)]
mm: fix handling PTE-mapped THPs in page_referenced()

For PTE-mapped THP page_check_address_transhuge() is not adequate: it
cannot find all relevant PTEs, only the first one.  It means we can miss
some references of the page and it can result in suboptimal decisions by
vmscan.

Let's switch it to page_vma_mapped_walk().

I don't think it's subject for stable@: it's not fatal.  The only side
effect is that THP can be swapped out when it shouldn't.

Link: http://lkml.kernel.org/r/20170129173858.45174-4-kirill.shutemov@linux.intel.com
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hillf Danton <hillf.zj@alibaba-inc.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm: introduce page_vma_mapped_walk()
Kirill A. Shutemov [Fri, 24 Feb 2017 22:57:45 +0000 (14:57 -0800)]
mm: introduce page_vma_mapped_walk()

Introduce a new interface to check if a page is mapped into a vma.  It
aims to address shortcomings of page_check_address{,_transhuge}.

Existing interface is not able to handle PTE-mapped THPs: it only finds
the first PTE.  The rest lefted unnoticed.

page_vma_mapped_walk() iterates over all possible mapping of the page in
the vma.

Link: http://lkml.kernel.org/r/20170129173858.45174-3-kirill.shutemov@linux.intel.com
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hillf Danton <hillf.zj@alibaba-inc.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agouprobes: split THPs before trying to replace them
Kirill A. Shutemov [Fri, 24 Feb 2017 22:57:42 +0000 (14:57 -0800)]
uprobes: split THPs before trying to replace them

Patch series "Fix few rmap-related THP bugs", v3.

The patchset fixes handing PTE-mapped THPs in page_referenced() and
page_idle_clear_pte_refs().

To achieve that I've intrdocued new helper -- page_vma_mapped_walk() --
which replaces all page_check_address{,_transhuge}() and covers all THP
cases.

Patchset overview:
  - First patch fixes one uprobe bug (unrelated to the rest of the
    patchset, just spotted it at the same time);

  - Patches 2-5 fix handling PTE-mapped THPs in page_referenced(),
    page_idle_clear_pte_refs() and rmap core;

  - Patches 6-12 convert all page_check_address{,_transhuge}() users
    (plus remove_migration_pte()) to page_vma_mapped_walk() and drop
    unused helpers.

I think the fixes are not critical enough for stable@ as they don't lead
to crashes or hangs, only suboptimal behaviour.

This patch (of 12):

For THPs page_check_address() always fails.  It leads to endless loop in
uprobe_write_opcode().

Testcase with huge-tmpfs (uprobes cannot probe anonymous memory).

mount -t debugfs none /sys/kernel/debug
mount -t tmpfs -o huge=always none /mnt
gcc -Wall -O2 -o /mnt/test -x c - <<EOF
int main(void)
{
return 0;
}
/* Padding to map the code segment with huge pmd */
asm (".zero 2097152");
EOF
echo 'p /mnt/test:0' > /sys/kernel/debug/tracing/uprobe_events
echo 1 > /sys/kernel/debug/tracing/events/uprobes/enable
/mnt/test

Let's split THPs before trying to replace.

Link: http://lkml.kernel.org/r/20170129173858.45174-2-kirill.shutemov@linux.intel.com
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Acked-by: Rik van Riel <riel@redhat.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Hillf Danton <hillf.zj@alibaba-inc.com>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm/hotplug: enable memory hotplug for non-lru movable pages
Yisheng Xie [Fri, 24 Feb 2017 22:57:39 +0000 (14:57 -0800)]
mm/hotplug: enable memory hotplug for non-lru movable pages

We had considered all of the non-lru pages as unmovable before commit
bda807d44454 ("mm: migrate: support non-lru movable page migration").
But now some of non-lru pages like zsmalloc, virtio-balloon pages also
become movable.  So we can offline such blocks by using non-lru page
migration.

This patch straightforwardly adds non-lru migration code, which means
adding non-lru related code to the functions which scan over pfn and
collect pages to be migrated and isolate them before migration.

Signed-off-by: Yisheng Xie <xieyisheng1@huawei.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Hanjun Guo <guohanjun@huawei.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Reza Arbab <arbab@linux.vnet.ibm.com>
Cc: Taku Izumi <izumi.taku@jp.fujitsu.com>
Cc: Vitaly Kuznetsov <vkuznets@redhat.com>
Cc: Xishi Qiu <qiuxishi@huawei.com>
Cc: Yisheng Xie <xieyisheng1@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agoHWPOISON: soft offlining for non-lru movable page
Yisheng Xie [Fri, 24 Feb 2017 22:57:35 +0000 (14:57 -0800)]
HWPOISON: soft offlining for non-lru movable page

Extend soft offlining framework to support non-lru page, which already
support migration after commit bda807d44454 ("mm: migrate: support
non-lru movable page migration")

When memory corrected errors occur on a non-lru movable page, we can
choose to stop using it by migrating data onto another page and disable
the original (maybe half-broken) one.

Link: http://lkml.kernel.org/r/1485867981-16037-4-git-send-email-ysxie@foxmail.com
Signed-off-by: Yisheng Xie <xieyisheng1@huawei.com>
Suggested-by: Michal Hocko <mhocko@kernel.org>
Suggested-by: Minchan Kim <minchan@kernel.org>
Reviewed-by: Minchan Kim <minchan@kernel.org>
Acked-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Hanjun Guo <guohanjun@huawei.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Reza Arbab <arbab@linux.vnet.ibm.com>
Cc: Taku Izumi <izumi.taku@jp.fujitsu.com>
Cc: Vitaly Kuznetsov <vkuznets@redhat.com>
Cc: Xishi Qiu <qiuxishi@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm/migration: make isolate_movable_page always defined
Yisheng Xie [Fri, 24 Feb 2017 22:57:32 +0000 (14:57 -0800)]
mm/migration: make isolate_movable_page always defined

Define isolate_movable_page as a static inline function when
CONFIG_MIGRATION is not enable.  It should return -EBUSY here which
means failed to isolate movable pages.

This patch do not have any functional change but prepare for later
patch.

Link: http://lkml.kernel.org/r/1485867981-16037-3-git-send-email-ysxie@foxmail.com
Signed-off-by: Yisheng Xie <xieyisheng1@huawei.com>
Acked-by: Minchan Kim <minchan@kernel.org>
Suggested-by: Michal Hocko <mhocko@kernel.org>
Cc: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Hanjun Guo <guohanjun@huawei.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Reza Arbab <arbab@linux.vnet.ibm.com>
Cc: Taku Izumi <izumi.taku@jp.fujitsu.com>
Cc: Vitaly Kuznetsov <vkuznets@redhat.com>
Cc: Xishi Qiu <qiuxishi@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm/migration: make isolate_movable_page() return int type
Yisheng Xie [Fri, 24 Feb 2017 22:57:29 +0000 (14:57 -0800)]
mm/migration: make isolate_movable_page() return int type

Patch series "HWPOISON: soft offlining for non-lru movable page", v6.

After Minchan's commit bda807d44454 ("mm: migrate: support non-lru
movable page migration"), some type of non-lru page like zsmalloc and
virtio-balloon page also support migration.

Therefore, we can:

1) soft offlining no-lru movable pages, which means when memory
   corrected errors occur on a non-lru movable page, we can stop to use
   it by migrating data onto another page and disable the original
   (maybe half-broken) one.

2) enable memory hotplug for non-lru movable pages, i.e. we may offline
   blocks, which include such pages, by using non-lru page migration.

This patchset is heavily dependent on non-lru movable page migration.

This patch (of 4):

Change the return type of isolate_movable_page() from bool to int.  It
will return 0 when isolate movable page successfully, and return -EBUSY
when it isolates failed.

There is no functional change within this patch but prepare for later
patch.

[xieyisheng1@huawei.com: v6]
Link: http://lkml.kernel.org/r/1486108770-630-2-git-send-email-xieyisheng1@huawei.com
Link: http://lkml.kernel.org/r/1485867981-16037-2-git-send-email-ysxie@foxmail.com
Signed-off-by: Yisheng Xie <xieyisheng1@huawei.com>
Suggested-by: Michal Hocko <mhocko@kernel.org>
Acked-by: Minchan Kim <minchan@kernel.org>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Hanjun Guo <guohanjun@huawei.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Cc: Reza Arbab <arbab@linux.vnet.ibm.com>
Cc: Taku Izumi <izumi.taku@jp.fujitsu.com>
Cc: Vitaly Kuznetsov <vkuznets@redhat.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Xishi Qiu <qiuxishi@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agoz3fold: add kref refcounting
Vitaly Wool [Fri, 24 Feb 2017 22:57:26 +0000 (14:57 -0800)]
z3fold: add kref refcounting

With both coming and already present locking optimizations, introducing
kref to reference-count z3fold objects is the right thing to do.
Moreover, it makes buddied list no longer necessary, and allows for a
simpler handling of headless pages.

[akpm@linux-foundation.org: coding-style fixes]
Link: http://lkml.kernel.org/r/20170131214650.8ea78033d91ded233f552bc0@gmail.com
Signed-off-by: Vitaly Wool <vitalywool@gmail.com>
Reviewed-by: Dan Streetman <ddstreet@ieee.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agoz3fold: use per-page spinlock
Vitaly Wool [Fri, 24 Feb 2017 22:57:23 +0000 (14:57 -0800)]
z3fold: use per-page spinlock

Most of z3fold operations are in-page, such as modifying z3fold page
header or moving z3fold objects within a page.  Taking per-pool spinlock
to protect per-page objects is therefore suboptimal, and the idea of
having a per-page spinlock (or rwlock) has been around for some time.

This patch implements spinlock-based per-page locking mechanism which is
lightweight enough to normally fit ok into the z3fold header.

Link: http://lkml.kernel.org/r/20170131214438.433e0a5fda908337b63206d3@gmail.com
Signed-off-by: Vitaly Wool <vitalywool@gmail.com>
Reviewed-by: Dan Streetman <ddstreet@ieee.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agoz3fold: extend compaction function
Vitaly Wool [Fri, 24 Feb 2017 22:57:20 +0000 (14:57 -0800)]
z3fold: extend compaction function

z3fold_compact_page() currently only handles the situation when there's
a single middle chunk within the z3fold page.  However it may be worth
it to move middle chunk closer to either first or last chunk, whichever
is there, if the gap between them is big enough.

This patch adds the relevant code, using BIG_CHUNK_GAP define as a
threshold for middle chunk to be worth moving.

Link: http://lkml.kernel.org/r/20170131214334.c4f3eac9a477af0fa9a22c46@gmail.com
Signed-off-by: Vitaly Wool <vitalywool@gmail.com>
Reviewed-by: Dan Streetman <ddstreet@ieee.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agoz3fold: fix header size related issues
Vitaly Wool [Fri, 24 Feb 2017 22:57:17 +0000 (14:57 -0800)]
z3fold: fix header size related issues

Currently the whole kernel build will be stopped if the size of struct
z3fold_header is greater than the size of one chunk, which is 64 bytes
by default.  This patch instead defines the offset for z3fold objects as
the size of the z3fold header in chunks.

Fixed also are the calculation of num_free_chunks() and the address to
move the middle chunk to in case of in-page compaction in
z3fold_compact_page().

Link: http://lkml.kernel.org/r/20170131214057.d98677032bc7b1c6c59a80c9@gmail.com
Signed-off-by: Vitaly Wool <vitalywool@gmail.com>
Reviewed-by: Dan Streetman <ddstreet@ieee.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agoz3fold: make pages_nr atomic
Vitaly Wool [Fri, 24 Feb 2017 22:57:15 +0000 (14:57 -0800)]
z3fold: make pages_nr atomic

Convert pages_nr per-pool counter to atomic64_t.

Link: http://lkml.kernel.org/r/20170131213946.b828676ab17bbea42022c213@gmail.com
Signed-off-by: Vitaly Wool <vitalywool@gmail.com>
Reviewed-by: Dan Streetman <ddstreet@ieee.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm: fix get_user_pages() vs device-dax pud mappings
Dan Williams [Fri, 24 Feb 2017 22:57:12 +0000 (14:57 -0800)]
mm: fix get_user_pages() vs device-dax pud mappings

A new unit test for the device-dax 1GB enabling currently fails with
this warning before hanging the test thread:

 WARNING: CPU: 0 PID: 21 at lib/percpu-refcount.c:155 percpu_ref_switch_to_atomic_rcu+0x1e3/0x1f0
 percpu ref (dax_pmem_percpu_release [dax_pmem]) <= 0 (0) after switching to atomic
 [..]
 CPU: 0 PID: 21 Comm: rcuos/1 Tainted: G           O    4.10.0-rc7-next-20170207+ #944
 [..]
 Call Trace:
  dump_stack+0x86/0xc3
  __warn+0xcb/0xf0
  warn_slowpath_fmt+0x5f/0x80
  ? rcu_nocb_kthread+0x27a/0x510
  ? dax_pmem_percpu_exit+0x50/0x50 [dax_pmem]
  percpu_ref_switch_to_atomic_rcu+0x1e3/0x1f0
  ? percpu_ref_exit+0x60/0x60
  rcu_nocb_kthread+0x339/0x510
  ? rcu_nocb_kthread+0x27a/0x510
  kthread+0x101/0x140

The get_user_pages() path needs to arrange for references to be taken
against the dev_pagemap instance backing the pud mapping.  Refactor the
existing __gup_device_huge_pmd() to also account for the pud case.

Link: http://lkml.kernel.org/r/148653181153.38226.9605457830505509385.stgit@dwillia2-desk3.amr.corp.intel.com
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Jiang <dave.jiang@intel.com>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Nilesh Choudhury <nilesh.choudhury@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm: replace FAULT_FLAG_SIZE with parameter to huge_fault
Dave Jiang [Fri, 24 Feb 2017 22:57:08 +0000 (14:57 -0800)]
mm: replace FAULT_FLAG_SIZE with parameter to huge_fault

Since the introduction of FAULT_FLAG_SIZE to the vm_fault flag, it has
been somewhat painful with getting the flags set and removed at the
correct locations.  More than one kernel oops was introduced due to
difficulties of getting the placement correctly.

Remove the flag values and introduce an input parameter to huge_fault
that indicates the size of the page entry.  This makes the code easier
to trace and should avoid the issues we see with the fault flags where
removal of the flag was necessary in the fallback paths.

Link: http://lkml.kernel.org/r/148615748258.43180.1690152053774975329.stgit@djiang5-desk3.ch.intel.com
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Tested-by: Dan Williams <dan.j.williams@intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Nilesh Choudhury <nilesh.choudhury@oracle.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agodax: support for transparent PUD pages for device DAX
Dave Jiang [Fri, 24 Feb 2017 22:57:05 +0000 (14:57 -0800)]
dax: support for transparent PUD pages for device DAX

Add transparent huge PUD pages support for device DAX by adding a
pud_fault handler.

Link: http://lkml.kernel.org/r/148545060002.17912.6765687780007547551.stgit@djiang5-desk3.ch.intel.com
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Jan Kara <jack@suse.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Nilesh Choudhury <nilesh.choudhury@oracle.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm, x86: add support for PUD-sized transparent hugepages
Matthew Wilcox [Fri, 24 Feb 2017 22:57:02 +0000 (14:57 -0800)]
mm, x86: add support for PUD-sized transparent hugepages

The current transparent hugepage code only supports PMDs.  This patch
adds support for transparent use of PUDs with DAX.  It does not include
support for anonymous pages.  x86 support code also added.

Most of this patch simply parallels the work that was done for huge
PMDs.  The only major difference is how the new ->pud_entry method in
mm_walk works.  The ->pmd_entry method replaces the ->pte_entry method,
whereas the ->pud_entry method works along with either ->pmd_entry or
->pte_entry.  The pagewalk code takes care of locking the PUD before
calling ->pud_walk, so handlers do not need to worry whether the PUD is
stable.

[dave.jiang@intel.com: fix SMP x86 32bit build for native_pud_clear()]
Link: http://lkml.kernel.org/r/148719066814.31111.3239231168815337012.stgit@djiang5-desk3.ch.intel.com
[dave.jiang@intel.com: native_pud_clear missing on i386 build]
Link: http://lkml.kernel.org/r/148640375195.69754.3315433724330910314.stgit@djiang5-desk3.ch.intel.com
Link: http://lkml.kernel.org/r/148545059381.17912.8602162635537598445.stgit@djiang5-desk3.ch.intel.com
Signed-off-by: Matthew Wilcox <mawilcox@microsoft.com>
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Tested-by: Alexander Kapshuk <alexander.kapshuk@gmail.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Jan Kara <jack@suse.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Nilesh Choudhury <nilesh.choudhury@oracle.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm,fs,dax: change ->pmd_fault to ->huge_fault
Dave Jiang [Fri, 24 Feb 2017 22:56:59 +0000 (14:56 -0800)]
mm,fs,dax: change ->pmd_fault to ->huge_fault

Patch series "1G transparent hugepage support for device dax", v2.

The following series implements support for 1G trasparent hugepage on
x86 for device dax.  The bulk of the code was written by Mathew Wilcox a
while back supporting transparent 1G hugepage for fs DAX.  I have
forward ported the relevant bits to 4.10-rc.  The current submission has
only the necessary code to support device DAX.

Comments from Dan Williams: So the motivation and intended user of this
functionality mirrors the motivation and users of 1GB page support in
hugetlbfs.  Given expected capacities of persistent memory devices an
in-memory database may want to reduce tlb pressure beyond what they can
already achieve with 2MB mappings of a device-dax file.  We have
customer feedback to that effect as Willy mentioned in his previous
version of these patches [1].

[1]: https://lkml.org/lkml/2016/1/31/52

Comments from Nilesh @ Oracle:

There are applications which have a process model; and if you assume
10,000 processes attempting to mmap all the 6TB memory available on a
server; we are looking at the following:

processes         : 10,000
memory            :    6TB
pte @ 4k page size: 8 bytes / 4K of memory * #processes = 6TB / 4k * 8 * 10000 = 1.5GB * 80000 = 120,000GB
pmd @ 2M page size: 120,000 / 512 = ~240GB
pud @ 1G page size: 240GB / 512 = ~480MB

As you can see with 2M pages, this system will use up an exorbitant
amount of DRAM to hold the page tables; but the 1G pages finally brings
it down to a reasonable level.  Memory sizes will keep increasing; so
this number will keep increasing.

An argument can be made to convert the applications from process model
to thread model, but in the real world that may not be always practical.
Hopefully this helps explain the use case where this is valuable.

This patch (of 3):

In preparation for adding the ability to handle PUD pages, convert
vm_operations_struct.pmd_fault to vm_operations_struct.huge_fault.  The
vm_fault structure is extended to include a union of the different page
table pointers that may be needed, and three flag bits are reserved to
indicate which type of pointer is in the union.

[ross.zwisler@linux.intel.com: remove unused function ext4_dax_huge_fault()]
Link: http://lkml.kernel.org/r/1485813172-7284-1-git-send-email-ross.zwisler@linux.intel.com
[dave.jiang@intel.com: clear PMD or PUD size flags when in fall through path]
Link: http://lkml.kernel.org/r/148589842696.5820.16078080610311444794.stgit@djiang5-desk3.ch.intel.com
Link: http://lkml.kernel.org/r/148545058784.17912.6353162518188733642.stgit@djiang5-desk3.ch.intel.com
Signed-off-by: Matthew Wilcox <mawilcox@microsoft.com>
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Jan Kara <jack@suse.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Nilesh Choudhury <nilesh.choudhury@oracle.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Dave Jiang <dave.jiang@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm, page_alloc: use static global work_struct for draining per-cpu pages
Mel Gorman [Fri, 24 Feb 2017 22:56:56 +0000 (14:56 -0800)]
mm, page_alloc: use static global work_struct for draining per-cpu pages

As suggested by Vlastimil Babka and Tejun Heo, this patch uses a static
work_struct to co-ordinate the draining of per-cpu pages on the
workqueue.  Only one task can drain at a time but this is better than
the previous scheme that allowed multiple tasks to send IPIs at a time.

One consideration is whether parallel requests should synchronise
against each other.  This patch does not synchronise for a global drain
as the common case for such callers is expected to be multiple parallel
direct reclaimers competing for pages when the watermark is close to
min.  Draining the per-cpu list is unlikely to make much progress and
serialising the drain is of dubious merit.  Drains are synchonrised for
callers such as memory hotplug and CMA that care about the drain being
complete when the function returns.

Link: http://lkml.kernel.org/r/20170125083038.rzb5f43nptmk7aed@techsingularity.net
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Suggested-by: Tejun Heo <tj@kernel.org>
Suggested-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Hillf Danton <hillf.zj@alibaba-inc.com>
Cc: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm, page_alloc: don't check cpuset allowed twice in fast-path
Vlastimil Babka [Fri, 24 Feb 2017 22:56:53 +0000 (14:56 -0800)]
mm, page_alloc: don't check cpuset allowed twice in fast-path

Since commit 682a3385e773 ("mm, page_alloc: inline the fast path of the
zonelist iterator") we replace a NULL nodemask with
cpuset_current_mems_allowed in the fast path, so that
get_page_from_freelist() filters nodes allowed by the cpuset via
for_next_zone_zonelist_nodemask().

In that case it's pointless to additionaly check __cpuset_zone_allowed()
in each iteration, which we can avoid by not adding ALLOC_CPUSET to
alloc_flags in that scenario.

This saves some cycles in the allocator fast path on systems with one or
more non-root cpuset configured.  In the slow path, ALLOC_CPUSET is
reset according to __alloc_pages_slowpath().  Without configured
cpusets, this code is disabled by a static key.

Link: http://lkml.kernel.org/r/20170124150511.5710-2-vbabka@suse.cz
Signed-off-by: Vlastimil Babka <vbabka@suse.cz>
Reviewed-by: Anshuman Khandual <khandual@linux.vnet.ibm.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Anshuman Khandual <khandual@linux.vnet.ibm.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm, page_alloc: remove redundant checks from alloc fastpath
Vlastimil Babka [Fri, 24 Feb 2017 22:56:50 +0000 (14:56 -0800)]
mm, page_alloc: remove redundant checks from alloc fastpath

The allocation fast path contains two similar checks for zoneref->zone
being NULL, where zoneref points either to the first zone in the
zonelist, or to the preferred zone.  These can be NULL either due to
empty zonelist, or no zone being compatible with given nodemask or
task's cpuset.

These checks are unnecessary, because the zonelist walks in
first_zones_zonelist() and get_page_from_freelist() handle a NULL
starting zoneref->zone or preferred_zoneref->zone safely.  It's safe to
fallback to __alloc_pages_slowpath() where we also have the check early
enough.

Link: http://lkml.kernel.org/r/20170124150511.5710-1-vbabka@suse.cz
Signed-off-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Anshuman Khandual <khandual@linux.vnet.ibm.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agozram: remove waitqueue for IO done
Minchan Kim [Fri, 24 Feb 2017 22:56:47 +0000 (14:56 -0800)]
zram: remove waitqueue for IO done

zram_reset_device() waits for ongoing writepage pages to be completed by
zram->refcount logic.  However, it's pointless because before the reset,
we prevent further opening of zram by zram->claim and flush all of
pending IO by fsync_bdev so there should be no pending IO at the
zram_reset_device().

So let's remove that code which is even broken due to the lack of
wake_up elsewhere.

Link: http://lkml.kernel.org/r/1485145031-11661-1-git-send-email-minchan@kernel.org
Signed-off-by: Minchan Kim <minchan@kernel.org>
Reviewed-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm: fix comments for mmap_init()
seokhoon.yoon [Fri, 24 Feb 2017 22:56:44 +0000 (14:56 -0800)]
mm: fix comments for mmap_init()

mmap_init() is no longer associated with VMA slab.  So fix it.

Link: http://lkml.kernel.org/r/1485182601-9294-1-git-send-email-iamyooon@gmail.com
Signed-off-by: seokhoon.yoon <iamyooon@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm, fs: reduce fault, page_mkwrite, and pfn_mkwrite to take only vmf
Dave Jiang [Fri, 24 Feb 2017 22:56:41 +0000 (14:56 -0800)]
mm, fs: reduce fault, page_mkwrite, and pfn_mkwrite to take only vmf

->fault(), ->page_mkwrite(), and ->pfn_mkwrite() calls do not need to
take a vma and vmf parameter when the vma already resides in vmf.

Remove the vma parameter to simplify things.

[arnd@arndb.de: fix ARM build]
Link: http://lkml.kernel.org/r/20170125223558.1451224-1-arnd@arndb.de
Link: http://lkml.kernel.org/r/148521301778.19116.10840599906674778980.stgit@djiang5-desk3.ch.intel.com
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Reviewed-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Theodore Ts'o <tytso@mit.edu>
Cc: Darrick J. Wong <darrick.wong@oracle.com>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Jan Kara <jack@suse.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>
7 years agomm, page_alloc: only use per-cpu allocator for irq-safe requests
Mel Gorman [Fri, 24 Feb 2017 22:56:38 +0000 (14:56 -0800)]
mm, page_alloc: only use per-cpu allocator for irq-safe requests

Many workloads that allocate pages are not handling an interrupt at a
time.  As allocation requests may be from IRQ context, it's necessary to
disable/enable IRQs for every page allocation.  This cost is the bulk of
the free path but also a significant percentage of the allocation path.

This patch alters the locking and checks such that only irq-safe
allocation requests use the per-cpu allocator.  All others acquire the
irq-safe zone->lock and allocate from the buddy allocator.  It relies on
disabling preemption to safely access the per-cpu structures.  It could
be slightly modified to avoid soft IRQs using it but it's not clear it's
worthwhile.

This modification may slow allocations from IRQ context slightly but the
main gain from the per-cpu allocator is that it scales better for
allocations from multiple contexts.  There is an implicit assumption
that intensive allocations from IRQ contexts on multiple CPUs from a
single NUMA node are rare and that the fast majority of scaling issues
are encountered in !IRQ contexts such as page faulting.  It's worth
noting that this patch is not required for a bulk page allocator but it
significantly reduces the overhead.

The following is results from a page allocator micro-benchmark.  Only
order-0 is interesting as higher orders do not use the per-cpu allocator

                                          4.10.0-rc2                 4.10.0-rc2
                                             vanilla               irqsafe-v1r5
Amean    alloc-odr0-1               287.15 (  0.00%)           219.00 ( 23.73%)
Amean    alloc-odr0-2               221.23 (  0.00%)           183.23 ( 17.18%)
Amean    alloc-odr0-4               187.00 (  0.00%)           151.38 ( 19.05%)
Amean    alloc-odr0-8               167.54 (  0.00%)           132.77 ( 20.75%)
Amean    alloc-odr0-16              156.00 (  0.00%)           123.00 ( 21.15%)
Amean    alloc-odr0-32              149.00 (  0.00%)           118.31 ( 20.60%)
Amean    alloc-odr0-64              138.77 (  0.00%)           116.00 ( 16.41%)
Amean    alloc-odr0-128             145.00 (  0.00%)           118.00 ( 18.62%)
Amean    alloc-odr0-256             136.15 (  0.00%)           125.00 (  8.19%)
Amean    alloc-odr0-512             147.92 (  0.00%)           121.77 ( 17.68%)
Amean    alloc-odr0-1024            147.23 (  0.00%)           126.15 ( 14.32%)
Amean    alloc-odr0-2048            155.15 (  0.00%)           129.92 ( 16.26%)
Amean    alloc-odr0-4096            164.00 (  0.00%)           136.77 ( 16.60%)
Amean    alloc-odr0-8192            166.92 (  0.00%)           138.08 ( 17.28%)
Amean    alloc-odr0-16384           159.00 (  0.00%)           138.00 ( 13.21%)
Amean    free-odr0-1                165.00 (  0.00%)            89.00 ( 46.06%)
Amean    free-odr0-2                113.00 (  0.00%)            63.00 ( 44.25%)
Amean    free-odr0-4                 99.00 (  0.00%)            54.00 ( 45.45%)
Amean    free-odr0-8                 88.00 (  0.00%)            47.38 ( 46.15%)
Amean    free-odr0-16                83.00 (  0.00%)            46.00 ( 44.58%)
Amean    free-odr0-32                80.00 (  0.00%)            44.38 ( 44.52%)
Amean    free-odr0-64                72.62 (  0.00%)            43.00 ( 40.78%)
Amean    free-odr0-128               78.00 (  0.00%)            42.00 ( 46.15%)
Amean    free-odr0-256               80.46 (  0.00%)            57.00 ( 29.16%)
Amean    free-odr0-512               96.38 (  0.00%)            64.69 ( 32.88%)
Amean    free-odr0-1024             107.31 (  0.00%)            72.54 ( 32.40%)
Amean    free-odr0-2048             108.92 (  0.00%)            78.08 ( 28.32%)
Amean    free-odr0-4096             113.38 (  0.00%)            82.23 ( 27.48%)
Amean    free-odr0-8192             112.08 (  0.00%)            82.85 ( 26.08%)
Amean    free-odr0-16384            110.38 (  0.00%)            81.92 ( 25.78%)
Amean    total-odr0-1               452.15 (  0.00%)           308.00 ( 31.88%)
Amean    total-odr0-2               334.23 (  0.00%)           246.23 ( 26.33%)
Amean    total-odr0-4               286.00 (  0.00%)           205.38 ( 28.19%)
Amean    total-odr0-8               255.54 (  0.00%)           180.15 ( 29.50%)
Amean    total-odr0-16              239.00 (  0.00%)           169.00 ( 29.29%)
Amean    total-odr0-32              229.00 (  0.00%)           162.69 ( 28.96%)
Amean    total-odr0-64              211.38 (  0.00%)           159.00 ( 24.78%)
Amean    total-odr0-128             223.00 (  0.00%)           160.00 ( 28.25%)
Amean    total-odr0-256             216.62 (  0.00%)           182.00 ( 15.98%)
Amean    total-odr0-512             244.31 (  0.00%)           186.46 ( 23.68%)
Amean    total-odr0-1024            254.54 (  0.00%)           198.69 ( 21.94%)
Amean    total-odr0-2048            264.08 (  0.00%)           208.00 ( 21.24%)
Amean    total-odr0-4096            277.38 (  0.00%)           219.00 ( 21.05%)
Amean    total-odr0-8192            279.00 (  0.00%)           220.92 ( 20.82%)
Amean    total-odr0-16384           269.38 (  0.00%)           219.92 ( 18.36%)

This is the alloc, free and total overhead of allocating order-0 pages
in batches of 1 page up to 16384 pages.  Avoiding disabling/enabling
overhead massively reduces overhead.  Alloc overhead is roughly reduced
by 14-20% in most cases.  The free path is reduced by 26-46% and the
total reduction is significant.

Many users require zeroing of pages from the page allocator which is the
vast cost of allocation.  Hence, the impact on a basic page faulting
benchmark is not that significant

                              4.10.0-rc2            4.10.0-rc2
                                 vanilla          irqsafe-v1r5
Hmean    page_test   656632.98 (  0.00%)   675536.13 (  2.88%)
Hmean    brk_test   3845502.67 (  0.00%)  3867186.94 (  0.56%)
Stddev   page_test    10543.29 (  0.00%)     4104.07 ( 61.07%)
Stddev   brk_test     33472.36 (  0.00%)    15538.39 ( 53.58%)
CoeffVar page_test        1.61 (  0.00%)        0.61 ( 62.15%)
CoeffVar brk_test         0.87 (  0.00%)        0.40 ( 53.84%)
Max      page_test   666513.33 (  0.00%)   678640.00 (  1.82%)
Max      brk_test   3882800.00 (  0.00%)  3887008.66 (  0.11%)

This is from aim9 and the most notable outcome is that fault variability
is reduced by the patch.  The headline improvement is small as the
overall fault cost, zeroing, page table insertion etc dominate relative
to disabling/enabling IRQs in the per-cpu allocator.

Similarly, little benefit was seen on networking benchmarks both
localhost and between physical server/clients where other costs
dominate.  It's possible that this will only be noticable on very high
speed networks.

Jesper Dangaard Brouer independently tested this with a separate
microbenchmark from
  https://github.com/netoptimizer/prototype-kernel/tree/master/kernel/mm/bench

Micro-benchmarked with [1] page_bench02:
 modprobe page_bench02 page_order=0 run_flags=$((2#010)) loops=$((10**8)); \
  rmmod page_bench02 ; dmesg --notime | tail -n 4

Compared to baseline: 213 cycles(tsc) 53.417 ns
 - against this     : 184 cycles(tsc) 46.056 ns
 - Saving           : -29 cycles
 - Very close to expected 27 cycles saving [see below [2]]

Micro benchmarking via time_bench_sample[3], we get the cost of these
operations:

 time_bench: Type:for_loop                 Per elem: 0 cycles(tsc) 0.232 ns (step:0)
 time_bench: Type:spin_lock_unlock         Per elem: 33 cycles(tsc) 8.334 ns (step:0)
 time_bench: Type:spin_lock_unlock_irqsave Per elem: 62 cycles(tsc) 15.607 ns (step:0)
 time_bench: Type:irqsave_before_lock      Per elem: 57 cycles(tsc) 14.344 ns (step:0)
 time_bench: Type:spin_lock_unlock_irq     Per elem: 34 cycles(tsc) 8.560 ns (step:0)
 time_bench: Type:simple_irq_disable_before_lock Per elem: 37 cycles(tsc) 9.289 ns (step:0)
 time_bench: Type:local_BH_disable_enable  Per elem: 19 cycles(tsc) 4.920 ns (step:0)
 time_bench: Type:local_IRQ_disable_enable Per elem: 7 cycles(tsc) 1.864 ns (step:0)
 time_bench: Type:local_irq_save_restore   Per elem: 38 cycles(tsc) 9.665 ns (step:0)
 [Mel's patch removes a ^^^^^^^^^^^^^^^^]            ^^^^^^^^^ expected saving - preempt cost
 time_bench: Type:preempt_disable_enable   Per elem: 11 cycles(tsc) 2.794 ns (step:0)
 [adds a preempt  ^^^^^^^^^^^^^^^^^^^^^^]            ^^^^^^^^^ adds this cost
 time_bench: Type:funcion_call_cost        Per elem: 6 cycles(tsc) 1.689 ns (step:0)
 time_bench: Type:func_ptr_call_cost       Per elem: 11 cycles(tsc) 2.767 ns (step:0)
 time_bench: Type:page_alloc_put           Per elem: 211 cycles(tsc) 52.803 ns (step:0)

Thus, expected improvement is: 38-11 = 27 cycles.

[mgorman@techsingularity.net: s/preempt_enable_no_resched/preempt_enable/]
Link: http://lkml.kernel.org/r/20170208143128.25ahymqlyspjcixu@techsingularity.net
Link: http://lkml.kernel.org/r/20170123153906.3122-5-mgorman@techsingularity.net
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Hillf Danton <hillf.zj@alibaba-inc.com>
Acked-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Hillf Danton <hillf.zj@alibaba-inc.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm, page_alloc: do not depend on cpu hotplug locks inside the allocator
Michal Hocko [Fri, 24 Feb 2017 22:56:35 +0000 (14:56 -0800)]
mm, page_alloc: do not depend on cpu hotplug locks inside the allocator

Dmitry has reported the following lockdep splat
  lock_acquire+0x2a1/0x630 kernel/locking/lockdep.c:3753
  __mutex_lock_common kernel/locking/mutex.c:521 [inline]
  mutex_lock_nested+0x24e/0xff0 kernel/locking/mutex.c:621
  pcpu_alloc+0xbda/0x1280 mm/percpu.c:896
  __alloc_percpu+0x24/0x30 mm/percpu.c:1075
  smpcfd_prepare_cpu+0x73/0xd0 kernel/smp.c:44
  cpuhp_invoke_callback+0x254/0x1480 kernel/cpu.c:136
  cpuhp_up_callbacks+0x81/0x2a0 kernel/cpu.c:493
  _cpu_up+0x1e3/0x2a0 kernel/cpu.c:1057
  do_cpu_up+0x73/0xa0 kernel/cpu.c:1087
  cpu_up+0x18/0x20 kernel/cpu.c:1095
  smp_init+0xe9/0xee kernel/smp.c:564
  kernel_init_freeable+0x439/0x690 init/main.c:1010
  kernel_init+0x13/0x180 init/main.c:941
  ret_from_fork+0x2a/0x40 arch/x86/entry/entry_64.S:433

cpu_hotplug_begin
  cpu_hotplug.lock
pcpu_alloc
  pcpu_alloc_mutex

  get_online_cpus+0x62/0x90 kernel/cpu.c:248
  drain_all_pages+0xf8/0x710 mm/page_alloc.c:2385
  __alloc_pages_direct_reclaim mm/page_alloc.c:3440 [inline]
  __alloc_pages_slowpath+0x8fd/0x2370 mm/page_alloc.c:3778
  __alloc_pages_nodemask+0x8f5/0xc60 mm/page_alloc.c:3980
  __alloc_pages include/linux/gfp.h:426 [inline]
  __alloc_pages_node include/linux/gfp.h:439 [inline]
  alloc_pages_node include/linux/gfp.h:453 [inline]
  pcpu_alloc_pages mm/percpu-vm.c:93 [inline]
  pcpu_populate_chunk+0x1e1/0x900 mm/percpu-vm.c:282
  pcpu_alloc+0xe01/0x1280 mm/percpu.c:998
  __alloc_percpu_gfp+0x27/0x30 mm/percpu.c:1062
  bpf_array_alloc_percpu kernel/bpf/arraymap.c:34 [inline]
  array_map_alloc+0x532/0x710 kernel/bpf/arraymap.c:99
  find_and_alloc_map kernel/bpf/syscall.c:34 [inline]
  map_create kernel/bpf/syscall.c:188 [inline]
  SYSC_bpf kernel/bpf/syscall.c:870 [inline]
  SyS_bpf+0xd64/0x2500 kernel/bpf/syscall.c:827
  entry_SYSCALL_64_fastpath+0x1f/0xc2

pcpu_alloc
  pcpu_alloc_mutex
drain_all_pages
  get_online_cpus
    cpu_hotplug.lock

  cpu_hotplug_begin+0x206/0x2e0 kernel/cpu.c:304
  _cpu_up+0xca/0x2a0 kernel/cpu.c:1011
  do_cpu_up+0x73/0xa0 kernel/cpu.c:1087
  cpu_up+0x18/0x20 kernel/cpu.c:1095
  smp_init+0xe9/0xee kernel/smp.c:564
  kernel_init_freeable+0x439/0x690 init/main.c:1010
  kernel_init+0x13/0x180 init/main.c:941
  ret_from_fork+0x2a/0x40 arch/x86/entry/entry_64.S:433

cpu_hotplug_begin
  cpu_hotplug.lock

Pulling cpu hotplug locks inside the page allocator is just too
dangerous.  Let's remove the dependency by dropping get_online_cpus()
from drain_all_pages.  This is not so simple though because now we do
not have a protection against cpu hotplug which means 2 things:

  - the work item might be executed on a different cpu in worker from
    unbound pool so it doesn't run on pinned on the cpu

  - we have to make sure that we do not race with page_alloc_cpu_dead
    calling drain_pages_zone

Disabling preemption in drain_local_pages_wq will solve the first
problem drain_local_pages will determine its local CPU from the WQ
context which will be stable after that point, page_alloc_cpu_dead is
pinned to the CPU already.  The later condition is achieved by disabling
IRQs in drain_pages_zone.

Fixes: mm, page_alloc: drain per-cpu pages from workqueue context
Link: http://lkml.kernel.org/r/20170207201950.20482-1-mhocko@kernel.org
Signed-off-by: Michal Hocko <mhocko@suse.com>
Reported-by: Dmitry Vyukov <dvyukov@google.com>
Acked-by: Tejun Heo <tj@kernel.org>
Acked-by: Mel Gorman <mgorman@techsingularity.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm, page_alloc: drain per-cpu pages from workqueue context
Mel Gorman [Fri, 24 Feb 2017 22:56:32 +0000 (14:56 -0800)]
mm, page_alloc: drain per-cpu pages from workqueue context

The per-cpu page allocator can be drained immediately via
drain_all_pages() which sends IPIs to every CPU.  In the next patch, the
per-cpu allocator will only be used for interrupt-safe allocations which
prevents draining it from IPI context.  This patch uses workqueues to
drain the per-cpu lists instead.

This is slower but no slowdown during intensive reclaim was measured and
the paths that use drain_all_pages() are not that sensitive to
performance.  This is particularly true as the path would only be
triggered when reclaim is failing.  It also makes a some sense to avoid
storming a machine with IPIs when it's under memory pressure.  Arguably,
it should be further adjusted so that only one caller at a time is
draining pages but it's beyond the scope of the current patch.

Link: http://lkml.kernel.org/r/20170123153906.3122-4-mgorman@techsingularity.net
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Hillf Danton <hillf.zj@alibaba-inc.com>
Cc: Jesper Dangaard Brouer <brouer@redhat.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>
7 years agomm, page_alloc: split alloc_pages_nodemask()
Mel Gorman [Fri, 24 Feb 2017 22:56:29 +0000 (14:56 -0800)]
mm, page_alloc: split alloc_pages_nodemask()

alloc_pages_nodemask does a number of preperation steps that determine
what zones can be used for the allocation depending on a variety of
factors.  This is fine but a hypothetical caller that wanted multiple
order-0 pages has to do the preparation steps multiple times.  This
patch structures __alloc_pages_nodemask such that it's relatively easy
to build a bulk order-0 page allocator.  There is no functional change.

Link: http://lkml.kernel.org/r/20170123153906.3122-3-mgorman@techsingularity.net
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Hillf Danton <hillf.zj@alibaba-inc.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm, page_alloc: split buffered_rmqueue()
Mel Gorman [Fri, 24 Feb 2017 22:56:26 +0000 (14:56 -0800)]
mm, page_alloc: split buffered_rmqueue()

Patch series "Use per-cpu allocator for !irq requests and prepare for a
bulk allocator", v5.

This series is motivated by a conversation led by Jesper Dangaard Brouer
at the last LSF/MM proposing a generic page pool for DMA-coherent pages.
Part of his motivation was due to the overhead of allocating multiple
order-0 that led some drivers to use high-order allocations and
splitting them.  This is very slow in some cases.

The first two patches in this series restructure the page allocator such
that it is relatively easy to introduce an order-0 bulk page allocator.
A patch exists to do that and has been handed over to Jesper until an
in-kernel users is created.  The third patch prevents the per-cpu
allocator being drained from IPI context as that can potentially corrupt
the list after patch four is merged.  The final patch alters the per-cpu
alloctor to make it exclusive to !irq requests.  This cuts
allocation/free overhead by roughly 30%.

Performance tests from both Jesper and me are included in the patch.

This patch (of 4):

buffered_rmqueue removes a page from a given zone and uses the per-cpu
list for order-0.  This is fine but a hypothetical caller that wanted
multiple order-0 pages has to disable/reenable interrupts multiple
times.  This patch structures buffere_rmqueue such that it's relatively
easy to build a bulk order-0 page allocator.  There is no functional
change.

[mgorman@techsingularity.net: failed per-cpu refill may blow up]
Link: http://lkml.kernel.org/r/20170124112723.mshmgwq2ihxku2um@techsingularity.net
Link: http://lkml.kernel.org/r/20170123153906.3122-2-mgorman@techsingularity.net
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Hillf Danton <hillf.zj@alibaba-inc.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm: vmscan: move dirty pages out of the way until they're flushed
Johannes Weiner [Fri, 24 Feb 2017 22:56:23 +0000 (14:56 -0800)]
mm: vmscan: move dirty pages out of the way until they're flushed

We noticed a performance regression when moving hadoop workloads from
3.10 kernels to 4.0 and 4.6.  This is accompanied by increased pageout
activity initiated by kswapd as well as frequent bursts of allocation
stalls and direct reclaim scans.  Even lowering the dirty ratios to the
equivalent of less than 1% of memory would not eliminate the issue,
suggesting that dirty pages concentrate where the scanner is looking.

This can be traced back to recent efforts of thrash avoidance.  Where
3.10 would not detect refaulting pages and continuously supply clean
cache to the inactive list, a thrashing workload on 4.0+ will detect and
activate refaulting pages right away, distilling used-once pages on the
inactive list much more effectively.  This is by design, and it makes
sense for clean cache.  But for the most part our workload's cache
faults are refaults and its use-once cache is from streaming writes.  We
end up with most of the inactive list dirty, and we don't go after the
active cache as long as we have use-once pages around.

But waiting for writes to avoid reclaiming clean cache that *might*
refault is a bad trade-off.  Even if the refaults happen, reads are
faster than writes.  Before getting bogged down on writeback, reclaim
should first look at *all* cache in the system, even active cache.

To accomplish this, activate pages that are dirty or under writeback
when they reach the end of the inactive LRU.  The pages are marked for
immediate reclaim, meaning they'll get moved back to the inactive LRU
tail as soon as they're written back and become reclaimable.  But in the
meantime, by reducing the inactive list to only immediately reclaimable
pages, we allow the scanner to deactivate and refill the inactive list
with clean cache from the active list tail to guarantee forward
progress.

[hannes@cmpxchg.org: update comment]
Link: http://lkml.kernel.org/r/20170202191957.22872-8-hannes@cmpxchg.org
Link: http://lkml.kernel.org/r/20170123181641.23938-6-hannes@cmpxchg.org
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Minchan Kim <minchan@kernel.org>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Hillf Danton <hillf.zj@alibaba-inc.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm: vmscan: only write dirty pages that the scanner has seen twice
Johannes Weiner [Fri, 24 Feb 2017 22:56:20 +0000 (14:56 -0800)]
mm: vmscan: only write dirty pages that the scanner has seen twice

Dirty pages can easily reach the end of the LRU while there are still
clean pages to reclaim around.  Don't let kswapd write them back just
because there are a lot of them.  It costs more CPU to find the clean
pages, but that's almost certainly better than to disrupt writeback from
the flushers with LRU-order single-page writes from reclaim.  And the
flushers have been woken up by that point, so we spend IO capacity on
flushing and CPU capacity on finding the clean cache.

Only start writing dirty pages if they have cycled around the LRU twice
now and STILL haven't been queued on the IO device.  It's possible that
the dirty pages are so sparsely distributed across different bdis,
inodes, memory cgroups, that the flushers take forever to get to the
ones we want reclaimed.  Once we see them twice on the LRU, we know
that's the quicker way to find them, so do LRU writeback.

Link: http://lkml.kernel.org/r/20170123181641.23938-5-hannes@cmpxchg.org
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Minchan Kim <minchan@kernel.org>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Mel Gorman <mgorman@suse.de>
Acked-by: Hillf Danton <hillf.zj@alibaba-inc.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm: vmscan: remove old flusher wakeup from direct reclaim path
Johannes Weiner [Fri, 24 Feb 2017 22:56:17 +0000 (14:56 -0800)]
mm: vmscan: remove old flusher wakeup from direct reclaim path

Direct reclaim has been replaced by kswapd reclaim in pretty much all
common memory pressure situations, so this code most likely doesn't
accomplish the described effect anymore.  The previous patch wakes up
flushers for all reclaimers when we encounter dirty pages at the tail
end of the LRU.  Remove the crufty old direct reclaim invocation.

Link: http://lkml.kernel.org/r/20170123181641.23938-4-hannes@cmpxchg.org
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Minchan Kim <minchan@kernel.org>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Hillf Danton <hillf.zj@alibaba-inc.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm: vmscan: kick flushers when we encounter dirty pages on the LRU
Johannes Weiner [Fri, 24 Feb 2017 22:56:14 +0000 (14:56 -0800)]
mm: vmscan: kick flushers when we encounter dirty pages on the LRU

Memory pressure can put dirty pages at the end of the LRU without
anybody running into dirty limits.  Don't start writing individual pages
from kswapd while the flushers might be asleep.

Unlike the old direct reclaim flusher wakeup (removed in the next patch)
that flushes the number of pages just scanned, this patch wakes the
flushers for all outstanding dirty pages.  That seemed to perform better
in a synthetic test that pushes dirty pages to the end of the LRU and
into reclaim, because we know LRU aging outstrips writeback already, and
this way we give younger dirty pages a headstart rather than wait until
reclaim runs into them as well.  It also means less plugging and risk of
exhausting the struct request pool from reclaim.

There is a concern that this will cause temporary files that used to get
dirtied and truncated before writeback to now get written to disk under
memory pressure.  If this turns out to be a real problem, we'll have to
revisit this and tame the reclaim flusher wakeups.

[hannes@cmpxchg.org: mention dirty expiration as a condition]
Link: http://lkml.kernel.org/r/20170126174739.GA30636@cmpxchg.org
Link: http://lkml.kernel.org/r/20170123181641.23938-3-hannes@cmpxchg.org
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Minchan Kim <minchan@kernel.org>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Mel Gorman <mgorman@suse.de>
Acked-by: Hillf Danton <hillf.zj@alibaba-inc.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm: vmscan: scan dirty pages even in laptop mode
Johannes Weiner [Fri, 24 Feb 2017 22:56:11 +0000 (14:56 -0800)]
mm: vmscan: scan dirty pages even in laptop mode

Patch series "mm: vmscan: fix kswapd writeback regression".

We noticed a regression on multiple hadoop workloads when moving from
3.10 to 4.0 and 4.6, which involves kswapd getting tangled up in page
writeout, causing direct reclaim herds that also don't make progress.

I tracked it down to the thrash avoidance efforts after 3.10 that make
the kernel better at keeping use-once cache and use-many cache sorted on
the inactive and active list, with more aggressive protection of the
active list as long as there is inactive cache.  Unfortunately, our
workload's use-once cache is mostly from streaming writes.  Waiting for
writes to avoid potential reloads in the future is not a good tradeoff.

These patches do the following:

1. Wake the flushers when kswapd sees a lump of dirty pages. It's
   possible to be below the dirty background limit and still have cache
   velocity push them through the LRU. So start a-flushin'.

2. Let kswapd only write pages that have been rotated twice. This makes
   sure we really tried to get all the clean pages on the inactive list
   before resorting to horrible LRU-order writeback.

3. Move rotating dirty pages off the inactive list. Instead of churning
   or waiting on page writeback, we'll go after clean active cache. This
   might lead to thrashing, but in this state memory demand outstrips IO
   speed anyway, and reads are faster than writes.

Mel backported the series to 4.10-rc5 with one minor conflict and ran a
couple of tests on it.  Mix of read/write random workload didn't show
anything interesting.  Write-only database didn't show much difference
in performance but there were slight reductions in IO -- probably in the
noise.

simoop did show big differences although not as big as Mel expected.
This is Chris Mason's workload that similate the VM activity of hadoop.
Mel won't go through the full details but over the samples measured
during an hour it reported

                                         4.10.0-rc5            4.10.0-rc5
                                            vanilla         johannes-v1r1
Amean    p50-Read             21346531.56 (  0.00%) 21697513.24 ( -1.64%)
Amean    p95-Read             24700518.40 (  0.00%) 25743268.98 ( -4.22%)
Amean    p99-Read             27959842.13 (  0.00%) 28963271.11 ( -3.59%)
Amean    p50-Write                1138.04 (  0.00%)      989.82 ( 13.02%)
Amean    p95-Write             1106643.48 (  0.00%)    12104.00 ( 98.91%)
Amean    p99-Write             1569213.22 (  0.00%)    36343.38 ( 97.68%)
Amean    p50-Allocation          85159.82 (  0.00%)    79120.70 (  7.09%)
Amean    p95-Allocation         204222.58 (  0.00%)   129018.43 ( 36.82%)
Amean    p99-Allocation         278070.04 (  0.00%)   183354.43 ( 34.06%)
Amean    final-p50-Read       21266432.00 (  0.00%) 21921792.00 ( -3.08%)
Amean    final-p95-Read       24870912.00 (  0.00%) 26116096.00 ( -5.01%)
Amean    final-p99-Read       28147712.00 (  0.00%) 29523968.00 ( -4.89%)
Amean    final-p50-Write          1130.00 (  0.00%)      977.00 ( 13.54%)
Amean    final-p95-Write       1033216.00 (  0.00%)     2980.00 ( 99.71%)
Amean    final-p99-Write       1517568.00 (  0.00%)    32672.00 ( 97.85%)
Amean    final-p50-Allocation    86656.00 (  0.00%)    78464.00 (  9.45%)
Amean    final-p95-Allocation   211712.00 (  0.00%)   116608.00 ( 44.92%)
Amean    final-p99-Allocation   287232.00 (  0.00%)   168704.00 ( 41.27%)

The latencies are actually completely horrific in comparison to 4.4 (and
4.10-rc5 is worse than 4.9 according to historical data for reasons Mel
hasn't analysed yet).

Still, 95% of write latency (p95-write) is halved by the series and
allocation latency is way down.  Direct reclaim activity is one fifth of
what it was according to vmstats.  Kswapd activity is higher but this is
not necessarily surprising.  Kswapd efficiency is unchanged at 99% (99%
of pages scanned were reclaimed) but direct reclaim efficiency went from
77% to 99%

In the vanilla kernel, 627MB of data was written back from reclaim
context.  With the series, no data was written back.  With or without
the patch, pages are being immediately reclaimed after writeback
completes.  However, with the patch, only 1/8th of the pages are
reclaimed like this.

This patch (of 5):

We have an elaborate dirty/writeback throttling mechanism inside the
reclaim scanner, but for that to work the pages have to go through
shrink_page_list() and get counted for what they are.  Otherwise, we
mess up the LRU order and don't match reclaim speed to writeback.

Especially during deactivation, there is never a reason to skip dirty
pages; nothing is even trying to write them out from there.  Don't mess
up the LRU order for nothing, shuffle these pages along.

Link: http://lkml.kernel.org/r/20170123181641.23938-2-hannes@cmpxchg.org
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Minchan Kim <minchan@kernel.org>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Mel Gorman <mgorman@suse.de>
Acked-by: Hillf Danton <hillf.zj@alibaba-inc.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agouserfaultfd: non-cooperative: selftest: enable REMOVE event test for shmem
Mike Rapoport [Fri, 24 Feb 2017 22:56:08 +0000 (14:56 -0800)]
userfaultfd: non-cooperative: selftest: enable REMOVE event test for shmem

Now when madvise(MADV_REMOVE) notifies uffd reader, we should verify
that appliciation actually sees zeros at the removed range.

Link: http://lkml.kernel.org/r/1484814154-1557-4-git-send-email-rppt@linux.vnet.ibm.com
Signed-off-by: Mike Rapoport <rppt@linux.vnet.ibm.com>
Reviewed-by: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hillf Danton <hillf.zj@alibaba-inc.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agouserfaultfd: non-cooperative: add madvise() event for MADV_REMOVE request
Mike Rapoport [Fri, 24 Feb 2017 22:56:05 +0000 (14:56 -0800)]
userfaultfd: non-cooperative: add madvise() event for MADV_REMOVE request

When a page is removed from a shared mapping, the uffd reader should be
notified, so that it won't attempt to handle #PF events for the removed
pages.

We can reuse the UFFD_EVENT_REMOVE because from the uffd monitor point
of view, the semantices of madvise(MADV_DONTNEED) and
madvise(MADV_REMOVE) is exactly the same.

Link: http://lkml.kernel.org/r/1484814154-1557-3-git-send-email-rppt@linux.vnet.ibm.com
Signed-off-by: Mike Rapoport <rppt@linux.vnet.ibm.com>
Reviewed-by: Andrea Arcangeli <aarcange@redhat.com>
Acked-by: Hillf Danton <hillf.zj@alibaba-inc.com>
Acked-by: Pavel Emelyanov <xemul@virtuozzo.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agouserfaultfd: non-cooperative: rename *EVENT_MADVDONTNEED to *EVENT_REMOVE
Mike Rapoport [Fri, 24 Feb 2017 22:56:02 +0000 (14:56 -0800)]
userfaultfd: non-cooperative: rename *EVENT_MADVDONTNEED to *EVENT_REMOVE

Patch series "userfaultfd: non-cooperative: add madvise() event for
MADV_REMOVE request".

These patches add notification of madvise(MADV_REMOVE) event to
non-cooperative userfaultfd monitor.

The first pacth renames EVENT_MADVDONTNEED to EVENT_REMOVE along with
relevant functions and structures.  Using _REMOVE instead of
_MADVDONTNEED describes the event semantics more clearly and I hope it's
not too late for such change in the ABI.

This patch (of 3):

The UFFD_EVENT_MADVDONTNEED purpose is to notify uffd monitor about
removal of certain range from address space tracked by userfaultfd.
Hence, UFFD_EVENT_REMOVE seems to better reflect the operation
semantics.  Respectively, 'madv_dn' field of uffd_msg is renamed to
'remove' and the madvise_userfault_dontneed callback is renamed to
userfaultfd_remove.

Link: http://lkml.kernel.org/r/1484814154-1557-2-git-send-email-rppt@linux.vnet.ibm.com
Signed-off-by: Mike Rapoport <rppt@linux.vnet.ibm.com>
Reviewed-by: Andrea Arcangeli <aarcange@redhat.com>
Acked-by: Hillf Danton <hillf.zj@alibaba-inc.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomemblock: embed memblock type name within struct memblock_type
Heiko Carstens [Fri, 24 Feb 2017 22:55:59 +0000 (14:55 -0800)]
memblock: embed memblock type name within struct memblock_type

Provide the name of each memblock type with struct memblock_type.  This
allows to get rid of the function memblock_type_name() and duplicating
the type names in __memblock_dump_all().

The only memblock_type usage out of mm/memblock.c seems to be
arch/s390/kernel/crash_dump.c.  While at it, give it a name.

Link: http://lkml.kernel.org/r/20170120123456.46508-4-heiko.carstens@de.ibm.com
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Philipp Hachtmann <phacht@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomemblock: also dump physmem list within __memblock_dump_all
Heiko Carstens [Fri, 24 Feb 2017 22:55:56 +0000 (14:55 -0800)]
memblock: also dump physmem list within __memblock_dump_all

Since commit 70210ed950b5 ("mm/memblock: add physical memory list") the
memblock structure knows about a physical memory list.

The physical memory list should also be dumped if memblock_dump_all() is
called in case memblock_debug is switched on.  This makes debugging a
bit easier.

Link: http://lkml.kernel.org/r/20170120123456.46508-3-heiko.carstens@de.ibm.com
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Philipp Hachtmann <phacht@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomemblock: let memblock_type_name know about physmem type
Heiko Carstens [Fri, 24 Feb 2017 22:55:54 +0000 (14:55 -0800)]
memblock: let memblock_type_name know about physmem type

Since commit 70210ed950b5 ("mm/memblock: add physical memory list") the
memblock structure knows about a physical memory list.

memblock_type_name() should return "physmem" instead of "unknown" if the
name of the physmem memblock_type is being asked for.

Link: http://lkml.kernel.org/r/20170120123456.46508-2-heiko.carstens@de.ibm.com
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Philipp Hachtmann <phacht@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm/memory_hotplug.c: unexport __remove_pages()
Andrew Morton [Fri, 24 Feb 2017 22:55:51 +0000 (14:55 -0800)]
mm/memory_hotplug.c: unexport __remove_pages()

It has no modular callers.

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>
7 years agomm: validate device_hotplug is held for memory hotplug
Dan Williams [Fri, 24 Feb 2017 22:55:48 +0000 (14:55 -0800)]
mm: validate device_hotplug is held for memory hotplug

mem_hotplug_begin() assumes that it can set mem_hotplug.active_writer
and run the hotplug process without racing another thread.  Validate
this assumption with a lockdep assertion.

Link: http://lkml.kernel.org/r/148693886229.16345.1770484669403334689.stgit@dwillia2-desk3.amr.corp.intel.com
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Reported-by: Ben Hutchings <ben@decadent.org.uk>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Toshi Kani <toshi.kani@hpe.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Logan Gunthorpe <logang@deltatee.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Masayoshi Mizuma <m.mizuma@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm, devm_memremap_pages: hold device_hotplug lock over mem_hotplug_{begin, done}
Dan Williams [Fri, 24 Feb 2017 22:55:45 +0000 (14:55 -0800)]
mm, devm_memremap_pages: hold device_hotplug lock over mem_hotplug_{begin, done}

The mem_hotplug_{begin,done} lock coordinates with {get,put}_online_mems()
to hold off "readers" of the current state of memory from new hotplug
actions.  mem_hotplug_begin() expects exclusive access, via the
device_hotplug lock, to set mem_hotplug.active_writer.  Calling
mem_hotplug_begin() without locking device_hotplug can lead to
corrupting mem_hotplug.refcount and missed wakeups / soft lockups.

[dan.j.williams@intel.com: v2]
Link: http://lkml.kernel.org/r/148728203365.38457.17804568297887708345.stgit@dwillia2-desk3.amr.corp.intel.com
Link: http://lkml.kernel.org/r/148693885680.16345.17802627926777862337.stgit@dwillia2-desk3.amr.corp.intel.com
Fixes: f931ab479dd2 ("mm: fix devm_memremap_pages crash, use mem_hotplug_{begin, done}")
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Reported-by: Ben Hutchings <ben@decadent.org.uk>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Toshi Kani <toshi.kani@hpe.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Logan Gunthorpe <logang@deltatee.com>
Cc: Masayoshi Mizuma <m.mizuma@jp.fujitsu.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm, oom: header nodemask is NULL when cpusets are disabled
David Rientjes [Fri, 24 Feb 2017 22:55:42 +0000 (14:55 -0800)]
mm, oom: header nodemask is NULL when cpusets are disabled

Commit 82e7d3abec86 ("oom: print nodemask in the oom report") implicitly
sets the allocation nodemask to cpuset_current_mems_allowed when there
is no effective mempolicy.  cpuset_current_mems_allowed is only
effective when cpusets are enabled, which is also printed by
dump_header(), so setting the nodemask to cpuset_current_mems_allowed is
redundant and prevents debugging issues where ac->nodemask is not set
properly in the page allocator.

This provides better debugging output since
cpuset_print_current_mems_allowed() is already provided.

[rientjes@google.com: newline per Hillf]
Link: http://lkml.kernel.org/r/alpine.DEB.2.10.1701200158300.88321@chino.kir.corp.google.com
Link: http://lkml.kernel.org/r/alpine.DEB.2.10.1701191454470.2381@chino.kir.corp.google.com
Signed-off-by: David Rientjes <rientjes@google.com>
Suggested-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Hillf Danton <hillf.zj@alibaba-inc.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Mel Gorman <mgorman@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agomm/ksm: improve deduplication of zero pages with colouring
Claudio Imbrenda [Fri, 24 Feb 2017 22:55:39 +0000 (14:55 -0800)]
mm/ksm: improve deduplication of zero pages with colouring

Some architectures have a set of zero pages (coloured zero pages)
instead of only one zero page, in order to improve the cache
performance.  In those cases, the kernel samepage merger (KSM) would
merge all the allocated pages that happen to be filled with zeroes to
the same deduplicated page, thus losing all the advantages of coloured
zero pages.

This behaviour is noticeable when a process accesses large arrays of
allocated pages containing zeroes.  A test I conducted on s390 shows
that there is a speed penalty when KSM merges such pages, compared to
not merging them or using actual zero pages from the start without
breaking the COW.

This patch fixes this behaviour.  When coloured zero pages are present,
the checksum of a zero page is calculated during initialisation, and
compared with the checksum of the current canditate during merging.  In
case of a match, the normal merging routine is used to merge the page
with the correct coloured zero page, which ensures the candidate page is
checked to be equal to the target zero page.

A sysfs entry is also added to toggle this behaviour, since it can
potentially introduce performance regressions, especially on
architectures without coloured zero pages.  The default value is
disabled, for backwards compatibility.

With this patch, the performance with KSM is the same as with non
COW-broken actual zero pages, which is also the same as without KSM.

[akpm@linux-foundation.org: make zero_checksum and ksm_use_zero_pages __read_mostly, per Andrea]
[imbrenda@linux.vnet.ibm.com: documentation for coloured zero pages deduplication]
Link: http://lkml.kernel.org/r/1484927522-1964-1-git-send-email-imbrenda@linux.vnet.ibm.com
Link: http://lkml.kernel.org/r/1484850953-23941-1-git-send-email-imbrenda@linux.vnet.ibm.com
Signed-off-by: Claudio Imbrenda <imbrenda@linux.vnet.ibm.com>
Cc: Christian Borntraeger <borntraeger@de.ibm.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agocris: use generic current.h
Davidlohr Bueso [Fri, 24 Feb 2017 22:55:36 +0000 (14:55 -0800)]
cris: use generic current.h

Given that the arch does not add its own implementations, simply use the
asm-generic/current.h (generic-y) header instead of duplicating code.

Link: http://lkml.kernel.org/r/1485992878-4780-3-git-send-email-dave@stgolabs.net
Signed-off-by: Davidlohr Bueso <dbueso@suse.de>
Cc: Mikael Starvik <starvik@axis.com>
Cc: Jesper Nilsson <jesper.nilsson@axis.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
7 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ebiederm...
Linus Torvalds [Fri, 24 Feb 2017 04:33:51 +0000 (20:33 -0800)]
Merge branch 'for-linus' of git://git./linux/kernel/git/ebiederm/user-namespace

Pull namespace updates from Eric Biederman:
 "There is a lot here. A lot of these changes result in subtle user
  visible differences in kernel behavior. I don't expect anything will
  care but I will revert/fix things immediately if any regressions show
  up.

  From Seth Forshee there is a continuation of the work to make the vfs
  ready for unpriviled mounts. We had thought the previous changes
  prevented the creation of files outside of s_user_ns of a filesystem,
  but it turns we missed the O_CREAT path. Ooops.

  Pavel Tikhomirov and Oleg Nesterov worked together to fix a long
  standing bug in the implemenation of PR_SET_CHILD_SUBREAPER where only
  children that are forked after the prctl are considered and not
  children forked before the prctl. The only known user of this prctl
  systemd forks all children after the prctl. So no userspace
  regressions will occur. Holding earlier forked children to the same
  rules as later forked children creates a semantic that is sane enough
  to allow checkpoing of processes that use this feature.

  There is a long delayed change by Nikolay Borisov to limit inotify
  instances inside a user namespace.

  Michael Kerrisk extends the API for files used to maniuplate
  namespaces with two new trivial ioctls to allow discovery of the
  hierachy and properties of namespaces.

  Konstantin Khlebnikov with the help of Al Viro adds code that when a
  network namespace exits purges it's sysctl entries from the dcache. As
  in some circumstances this could use a lot of memory.

  Vivek Goyal fixed a bug with stacked filesystems where the permissions
  on the wrong inode were being checked.

  I continue previous work on ptracing across exec. Allowing a file to
  be setuid across exec while being ptraced if the tracer has enough
  credentials in the user namespace, and if the process has CAP_SETUID
  in it's own namespace. Proc files for setuid or otherwise undumpable
  executables are now owned by the root in the user namespace of their
  mm. Allowing debugging of setuid applications in containers to work
  better.

  A bug I introduced with permission checking and automount is now
  fixed. The big change is to mark the mounts that the kernel initiates
  as a result of an automount. This allows the permission checks in sget
  to be safely suppressed for this kind of mount. As the permission
  check happened when the original filesystem was mounted.

  Finally a special case in the mount namespace is removed preventing
  unbounded chains in the mount hash table, and making the semantics
  simpler which benefits CRIU.

  The vfs fix along with related work in ima and evm I believe makes us
  ready to finish developing and merge fully unprivileged mounts of the
  fuse filesystem. The cleanups of the mount namespace makes discussing
  how to fix the worst case complexity of umount. The stacked filesystem
  fixes pave the way for adding multiple mappings for the filesystem
  uids so that efficient and safer containers can be implemented"

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ebiederm/user-namespace:
  proc/sysctl: Don't grab i_lock under sysctl_lock.
  vfs: Use upper filesystem inode in bprm_fill_uid()
  proc/sysctl: prune stale dentries during unregistering
  mnt: Tuck mounts under others instead of creating shadow/side mounts.
  prctl: propagate has_child_subreaper flag to every descendant
  introduce the walk_process_tree() helper
  nsfs: Add an ioctl() to return owner UID of a userns
  fs: Better permission checking for submounts
  exit: fix the setns() && PR_SET_CHILD_SUBREAPER interaction
  vfs: open() with O_CREAT should not create inodes with unknown ids
  nsfs: Add an ioctl() to return the namespace type
  proc: Better ownership of files for non-dumpable tasks in user namespaces
  exec: Remove LSM_UNSAFE_PTRACE_CAP
  exec: Test the ptracer's saved cred to see if the tracee can gain caps
  exec: Don't reset euid and egid when the tracee has CAP_SETUID
  inotify: Convert to using per-namespace limits

7 years agoMerge tag 'drm-for-v4.11-less-shouty' of git://people.freedesktop.org/~airlied/linux
Linus Torvalds [Fri, 24 Feb 2017 02:58:18 +0000 (18:58 -0800)]
Merge tag 'drm-for-v4.11-less-shouty' of git://people.freedesktop.org/~airlied/linux

Pull drm updates from Dave Airlie:
 "This is the main drm pull request for v4.11.

  Nothing too major, the tinydrm and mmu-less support should make
  writing smaller drivers easier for some of the simpler platforms, and
  there are a bunch of documentation updates.

  Intel grew displayport MST audio support which is hopefully useful to
  people, and FBC is on by default for GEN9+ (so people know where to
  look for regressions). AMDGPU has a lot of fixes that would like new
  firmware files installed for some GPUs.

  Other than that it's pretty scattered all over.

  I may have a follow up pull request as I know BenH has a bunch of AST
  rework and fixes and I'd like to get those in once they've been tested
  by AST, and I've got at least one pull request I'm just trying to get
  the author to fix up.

  Core:
   - drm_mm reworked
   - Connector list locking and iterators
   - Documentation updates
   - Format handling rework
   - MMU-less support for fbdev helpers
   - drm_crtc_from_index helper
   - Core CRC API
   - Remove drm_framebuffer_unregister_private
   - Debugfs cleanup
   - EDID/Infoframe fixes
   - Release callback
   - Tinydrm support (smaller drivers for simple hw)

  panel:
   - Add support for some new simple panels

  i915:
   - FBC by default for gen9+
   - Shared dpll cleanups and docs
   - GEN8 powerdomain cleanup
   - DMC support on GLK
   - DP MST audio support
   - HuC loading support
   - GVT init ordering fixes
   - GVT IOMMU workaround fix

  amdgpu/radeon:
   - Power/clockgating improvements
   - Preliminary SR-IOV support
   - TTM buffer priority and eviction fixes
   - SI DPM quirks removed due to firmware fixes
   - Powerplay improvements
   - VCE/UVD powergating fixes
   - Cleanup SI GFX code to match CI/VI
   - Support for > 2 displays on 3/5 crtc asics
   - SI headless fixes

  nouveau:
   - Rework securre boot code in prep for GP10x secure boot
   - Channel recovery improvements
   - Initial power budget code
   - MMU rework preperation

  vmwgfx:
   - Bunch of fixes and cleanups

  exynos:
   - Runtime PM support for MIC driver
   - Cleanups to use atomic helpers
   - UHD Support for TM2/TM2E boards
   - Trigger mode fix for Rinato board

  etnaviv:
   - Shader performance fix
   - Command stream validator fixes
   - Command buffer suballocator

  rockchip:
   - CDN DisplayPort support
   - IOMMU support for arm64 platform

  imx-drm:
   - Fix i.MX5 TV encoder probing
   - Remove lower fb size limits

  msm:
   - Support for HW cursor on MDP5 devices
   - DSI encoder cleanup
   - GPU DT bindings cleanup

  sti:
   - stih410 cleanups
   - Create fbdev at binding
   - HQVDP fixes
   - Remove stih416 chip functionality
   - DVI/HDMI mode selection fixes
   - FPS statistic reporting

  omapdrm:
   - IRQ code cleanup

  dwi-hdmi bridge:
   - Cleanups and fixes

  adv-bridge:
   - Updates for nexus

  sii8520 bridge:
   - Add interlace mode support
   - Rework HDMI and lots of fixes

  qxl:
   - probing/teardown cleanups

  ZTE drm:
   - HDMI audio via SPDIF interface
   - Video Layer overlay plane support
   - Add TV encoder output device

  atmel-hlcdc:
   - Rework fbdev creation logic

  tegra:
   - OF node fix

  fsl-dcu:
   - Minor fixes

  mali-dp:
   - Assorted fixes

  sunxi:
   - Minor fix"

[ This was the "fixed" pull, that still had build warnings due to people
  not even having build tested the result. I'm not a happy camper

  I've fixed the things I noticed up in this merge.      - Linus ]

* tag 'drm-for-v4.11-less-shouty' of git://people.freedesktop.org/~airlied/linux: (1177 commits)
  lib/Kconfig: make PRIME_NUMBERS not user selectable
  drm/tinydrm: helpers: Properly fix backlight dependency
  drm/tinydrm: mipi-dbi: Fix field width specifier warning
  drm/tinydrm: mipi-dbi: Silence: â€˜cmd’ may be used uninitialized
  drm/sti: fix build warnings in sti_drv.c and sti_vtg.c files
  drm/amd/powerplay: fix PSI feature on Polars12
  drm/amdgpu: refuse to reserve io mem for split VRAM buffers
  drm/ttm: fix use-after-free races in vm fault handling
  drm/tinydrm: Add support for Multi-Inno MI0283QT display
  dt-bindings: Add Multi-Inno MI0283QT binding
  dt-bindings: display/panel: Add common rotation property
  of: Add vendor prefix for Multi-Inno
  drm/tinydrm: Add MIPI DBI support
  drm/tinydrm: Add helper functions
  drm: Add DRM support for tiny LCD displays
  drm/amd/amdgpu: post card if there is real hw resetting performed
  drm/nouveau/tmr: provide backtrace when a timeout is hit
  drm/nouveau/pci/g92: Fix rearm
  drm/nouveau/drm/therm/fan: add a fallback if no fan control is specified in the vbios
  drm/nouveau/hwmon: expose power_max and power_crit
  ..

7 years agolib/Kconfig: make PRIME_NUMBERS not user selectable.
Dave Airlie [Fri, 24 Feb 2017 02:11:21 +0000 (12:11 +1000)]
lib/Kconfig: make PRIME_NUMBERS not user selectable.

Linus doesn't like it user selectable, so kill it until
someone needs it for something else.

Signed-off-by: Dave Airlie <airlied@redhat.com>
7 years agodrm/tinydrm: helpers: Properly fix backlight dependency
Noralf Trønnes [Thu, 23 Feb 2017 13:29:57 +0000 (14:29 +0100)]
drm/tinydrm: helpers: Properly fix backlight dependency

BACKLIGHT_CLASS_DEVICE was selected in the last version of the
tinydrm patchset to fix the backlight dependency, but the
ifdef CONFIG_BACKLIGHT_CLASS_DEVICE was forgotten. Fix that.

Signed-off-by: Noralf Trønnes <noralf@tronnes.org>
Signed-off-by: Dave Airlie <airlied@redhat.com>
7 years agodrm/tinydrm: mipi-dbi: Fix field width specifier warning
Noralf Trønnes [Thu, 23 Feb 2017 13:29:56 +0000 (14:29 +0100)]
drm/tinydrm: mipi-dbi: Fix field width specifier warning

This warning is seen on 64-bit builds in functions:
   'mipi_dbi_typec1_command':
   'mipi_dbi_typec3_command_read':
   'mipi_dbi_typec3_command':

>> drivers/gpu/drm/tinydrm/mipi-dbi.c:65:20: warning: field width specifier '*' expects argument of type 'int', but argument 5 has type 'size_t {aka long unsigned int}' [-Wformat=]
      DRM_DEBUG_DRIVER("cmd=%02x, par=%*ph\n", cmd, len, data); \
                       ^
   include/drm/drmP.h:228:40: note: in definition of macro 'DRM_DEBUG_DRIVER'
     drm_printk(KERN_DEBUG, DRM_UT_DRIVER, fmt, ##__VA_ARGS__)
                                           ^~~
>> drivers/gpu/drm/tinydrm/mipi-dbi.c:671:2: note: in expansion of macro 'MIPI_DBI_DEBUG_COMMAND'
     MIPI_DBI_DEBUG_COMMAND(cmd, parameters, num);
     ^~~~~~~~~~~~~~~~~~~~~~

Fix by casting 'len' to int in the macro MIPI_DBI_DEBUG_COMMAND().
There is no chance of overflow.

Signed-off-by: Noralf Trønnes <noralf@tronnes.org>
Signed-off-by: Dave Airlie <airlied@redhat.com>
7 years agodrm/tinydrm: mipi-dbi: Silence: ‘cmd’ may be used uninitialized
Noralf Trønnes [Thu, 23 Feb 2017 13:29:55 +0000 (14:29 +0100)]
drm/tinydrm: mipi-dbi: Silence: â€˜cmd’ may be used uninitialized

Fix this warning:
drivers/gpu/drm/tinydrm/mipi-dbi.c: In function â€˜mipi_dbi_debugfs_command_write’:
drivers/gpu/drm/tinydrm/mipi-dbi.c:905:8: warning: â€˜cmd’ may be used uninitialized in this function [-Wmaybe-uninitialized]
  ret = mipi_dbi_command_buf(mipi, cmd, parameters, i);
        ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

cmd can't be used uninitialized, but to satisfy the compiler,
initialize it to zero.

Signed-off-by: Noralf Trønnes <noralf@tronnes.org>
Signed-off-by: Dave Airlie <airlied@redhat.com>
7 years agoMerge tag 'usercopy-v4.11-rc1.fix' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Fri, 24 Feb 2017 00:55:48 +0000 (16:55 -0800)]
Merge tag 'usercopy-v4.11-rc1.fix' of git://git./linux/kernel/git/kees/linux

Pull usercopy test fix from Kees Cook:
 "Fix for non-MMU ARM testing, from Arnd Bergmann"

* tag 'usercopy-v4.11-rc1.fix' of git://git.kernel.org/pub/scm/linux/kernel/git/kees/linux:
  usercopy: ARM NOMMU has no 64-bit get_user

7 years agoMerge tag 'armsoc-drivers' of git://git.kernel.org/pub/scm/linux/kernel/git/arm/arm-soc
Linus Torvalds [Thu, 23 Feb 2017 23:57:04 +0000 (15:57 -0800)]
Merge tag 'armsoc-drivers' of git://git./linux/kernel/git/arm/arm-soc

Pull ARM SoC driver updates from Arnd Bergmann:
 "Driver updates for ARM SoCs.

  A handful of driver changes this time around. The larger changes are:

   - Reset drivers for hi3660 and zx2967

   - AHCI driver for Davinci, acked by Tejun and brought in here due to
     platform dependencies

   - Cleanups of atmel-ebi (External Bus Interface)

   - Tweaks for Rockchip GRF (General Register File) usage (kitchensink
     misc register range on the SoCs)

   - PM domains changes for support of two new ZTE SoCs (zx296718 and
     zx2967)"

* tag 'armsoc-drivers' of git://git.kernel.org/pub/scm/linux/kernel/git/arm/arm-soc: (53 commits)
  soc: samsung: pmu: Add register defines for pad retention control
  reset: make zx2967 explicitly non-modular
  reset: core: fix reset_control_put
  soc: samsung: pm_domains: Read domain name from the new label property
  soc: samsung: pm_domains: Remove message about failed memory allocation
  soc: samsung: pm_domains: Remove unused name field
  soc: samsung: pm_domains: Use full names in subdomains registration log
  sata: ahci-da850: un-hardcode the MPY bits
  sata: ahci-da850: add a workaround for controller instability
  sata: ahci: export ahci_do_hardreset() locally
  sata: ahci-da850: implement a workaround for the softreset quirk
  sata: ahci-da850: add device tree match table
  sata: ahci-da850: get the sata clock using a connection id
  soc: samsung: pmu: Remove duplicated define for ARM_L2_OPTION register
  memory: atmel-ebi: Enable the SMC clock if specified
  soc: samsung: pmu: Remove unused and duplicated defines
  memory: atmel-ebi: Properly handle multiple reference to the same CS
  memory: atmel-ebi: Fix the test to enable generic SMC logic
  soc: samsung: pm_domains: Add new Exynos5433 compatible
  soc: samsung: pmu: Add dummy support for Exynos5433 SoC
  ...

7 years agoMerge tag 'armsoc-dt64' of git://git.kernel.org/pub/scm/linux/kernel/git/arm/arm-soc
Linus Torvalds [Thu, 23 Feb 2017 23:52:14 +0000 (15:52 -0800)]
Merge tag 'armsoc-dt64' of git://git./linux/kernel/git/arm/arm-soc

Pull ARM 64-bit DT updates from Arnd Bergmann:
 "ARM64 DT updates are fairly small this time, only two new SoCs and a
  handful of new machines get added, all of them similar to other
  hardware we already support.

  New SoC:

   - HiSilicon Kirin960/Hi3660 and HiKey960 development board

   - NXP LS1012a with three reference boards:
        http://www.nxp.com/products/microcontrollers-and-processors/arm-processors/qoriq-layerscape-arm-processors/qoriq-layerscape-1012a-low-power-communication-processor:LS1012A

  New development board:

   - Banana Pi M64, based on Allwinner A64:
        http://www.banana-pi.org/m64.html

   - SolidRun MACCHIATOBin based on Marvell Armada 8K:
        https://www.solid-run.com/marvell-armada-family/armada-8040-community-board/

   - Broadcom BCM958712DxXMC NorthStar2 reference board (another one)

  A lot of platforms improve support for existing machines by adding
  extra devices for which a binding and driver is availabe:

  Allwinner:
   - MMC, USB

  ARM Juno:
   - Coresight, STM

  Broadcom:
   - NS2 GICv2m irqchip and PCIe

  Marvell:
   - Armada 3700 SPI, I2C, ethernet switch

  Mediatek:
   - MT8173 thermal

  NXP i.MX:
   - LS1046A thermal

  Qualcomm:
   - coresight on MSM8916, HDMI, WCNSS, SCM

  Renesas:
   - r8a779[56] thermal, powerdomain, ethernet, sound, pwm, can, can fd

  Rockchip:
   - thermal, eDP, pinctrl enhancements

  Samsung:
   - TM2 touchkey, Exynos5433 HDMI and power management improvements

  UniPhier:
   - SD reset, eMMC controller

  ZTE:
   - oppv2 cpufreq"

* tag 'armsoc-dt64' of git://git.kernel.org/pub/scm/linux/kernel/git/arm/arm-soc: (110 commits)
  arm64: dts: qcom: Add msm8916 CoreSight components
  arm64: dts: marvell: adjust name of sd-mmc-gop clock in syscon
  arm64: allwinner: add BananaPi-M64 support
  arm64: allwinner: a64: add UART1 pin nodes
  arm64: allwinner: pine64: add MMC support
  arm64: allwinner: a64: Increase the MMC max frequency
  arm64: allwinner: a64: Add MMC pinctrl nodes
  arm64: allwinner: a64: Add MMC nodes
  dt-bindings: clockgen: Add compatible string for LS1012A
  Documentation: DT: add LS1012A compatible for SCFG and DCFG
  Documentation: DT: Add entry for FSL LS1012A RDB, FRDM, QDS boards
  arm64: dts: marvell: add generic-ahci compatibles for CP110 ahci
  arm64: tegra: Use symbolic reset identifiers
  arm64: dts: r8a7796: Mark EthernetAVB device node disabled
  arm64: dts: r8a7795: Mark EthernetAVB device node disabled
  arm64: dts: r8a7795: tidyup audma definition order
  arm64: dts: r8a7796: Link ARM GIC to clock and clock domain
  arm64: dts: r8a7795: Link ARM GIC to clock and clock domain
  arm64: dts: r8a7796: Add R-Car Gen3 thermal support
  arm64: dts: r8a7795: Add R-Car Gen3 thermal support
  ...

7 years agoMerge tag 'armsoc-dt' of git://git.kernel.org/pub/scm/linux/kernel/git/arm/arm-soc
Linus Torvalds [Thu, 23 Feb 2017 23:46:25 +0000 (15:46 -0800)]
Merge tag 'armsoc-dt' of git://git./linux/kernel/git/arm/arm-soc

Pull ARM DT updates from Arnd Bergmann:
 "A total of 380 patches this time, mostly adding support for more
  hardware in the device tree descriptions. There is not much exciting
  here for 4.11, but I've tried my best to condense the information from
  the pull requests I got into a readable summary.

  Noteworthy changes to existing platforms include:

   - The GIC memory map was a bit wrong almost everywhere and now gets
     fixed up

   - The Allwinner platforms convert to the generic pinmux properties

   - The Marvell EBU platforms now use the new DSA binding

   - Samsung Exynos4212 was unused and gets removed

   - The Renesas power management got improved

  New production machines:

   - Lego Mindstorms EV3:
        https://www.lego.com/en-us/mindstorms/about-ev3

   - Beelink X2 Android media box:
        http://linux-sunxi.org/Beelink_X2

   - "Romulus" baseboard management controller for OpenPower

   - Axentia TSE-850 Data Radio Channel (DARC) encoder:
        http://www.axentia.se/db/equipment.html

   - Luxul XAP-1410 and XWR-1200 wireless access points:
        https://luxul.com/xap-1410

  New SoCs:

   - Allwinner H2+ and V3s, both minor variations of already supported
     chips:
        http://www.allwinnertech.com/index.php?c=product&a=index&id=38

   - Marvell Prestera DX packet processors based on Armada XP
     architecture:
        http://www.marvell.com/switching/prestera-dx/

   - Samsung Exynos4412 Prime gets added, a minor variation of
     Exynos4412

  New developer and reference boards:

   - Lichee Pi One, Lichee Pi Zero and Orange Pi Zero, all based on
     Allwinner SoCs:
        http://linux-sunxi.org/LicheePi_One
        http://www.orangepi.org/orangepizero/

   - SAMA5d36ek Reference platform:
        http://www.atmel.com/tools/sama5d36-ek.aspx

   - Beaglebone Green Wireless and Black Wireless:
        https://beagleboard.org/black-wireless
        https://beagleboard.org/green-wireless

   - phyCORE-AM335x System on Module:
        http://phytec.com/products/system-on-modules/phycore/am335x/

   - New revision of "vf610-zii" Zodiac Inflight Innovations board

   - Various i.MX System-on-Module: Is.IoT MX6UL, SavageBoard, Engicam
     i.Core:
        http://www.opossom.com/english/index.html
        http://www.savageboard.org/
        http://www.engicam.com/en/products/embedded/som/sodimm/is-iot-mx6ul
        http://www.engicam.com/en/products/embedded/som/sodimm/i-core-m6s-dl-d-q

   - Liebherr (LWN) monitor 6 based on i.MX6 Quad, no idea what this is

   - Cleanups and bugfixes on at91, bcm53xx, i.MX, mvebu, omap, oxnas,
     qcom, rockchip, sti, stm32 and tegra

  New device supports added to some boards and SoCs, briefly by platform:

   - Allwinner: SPDIF, A33 cpufreq, A33 Mali GPU

   - Aspeed: network, ipmi bt, gpio, pinmux

   - Broadcom: video encoder for raspberry pi, qspi, ethernet, sd/mmc

   - TI DaVinci: gpio, lcdc, usb, video-in, uart

   - TI Keystone 2: MSM RAM, power/reset, uart

   - Mediatek MT2701: clocks, iommu, spi, nand, adc, thermal

   - Marvell EBU: ethernet switch on Turris Omnia

   - NXP i.MX: otp ram, USB, wifi, bluetooth, spdif, spi, pmic, eeprom,
     mmc, nand

   - TI OMAP:

   - Qualcomm: coresight, gyro/accelerometer, hdmi

   - Renesas: pmic, soc-id

   - Rockchip: qos

   - Samsung: audio on Odroid-X

   - Socfpga: FPGA manager, i2c, led, can, watchdog, nand, power monitor

   - STi: video in/out

   - STM32: timer, pwm, i2c, rtc, add, i2s

   - NVIDIA Tegra: tpm

   - Uniphier: mmc/sd pinmux"

* tag 'armsoc-dt' of git://git.kernel.org/pub/scm/linux/kernel/git/arm/arm-soc: (380 commits)
  ARM: dts: armada-385-linksys: fix DSA compatible property
  ARM: dts: Fix typo in armada-xp-98dx4251
  ARM: DTS: Fix register map for virt-capable GIC
  dt-bindings: arm,gic: Fix binding example for a virt-capable GIC
  ARM: dts: sun8i: sinlinx: Enable audio nodes
  ARM: dts: sun8i: parrot: Enable audio nodes
  ARM: dts: sun8i: Add audio codec, dai and card for A33
  ARM: dts: Add EMAC AXI settings for Arria10
  ARM: dts: am335x-chiliboard: Support charger
  ARM: dts: am335x-chiliboard: Support power button
  ARM: sun8i: dt: Add mali node
  dt-bindings: gpu: Add Mali Utgard bindings
  ARM: dts: stm32: Add I2C1 support for STM32429 eval board
  ARM: dts: stm32: Add I2C1 support for STM32F429 SoC
  ARM: dts: stm32: Use clock DT binding definition on stm32f429 family
  dt-bindings: mfd: stm32f4: Add missing binding definition
  dt-bindings: mfd: stm32f4: Fix STM32F4_X_CLOCK() macro
  ARM: dts: stm32: Enable pwm1 and pwm3 for stm32f469-disco
  ARM: dts: stm32: add Timers driver for stm32f429 MCU
  ARM: dts: add the AB8500 sysclk to the device trees
  ...

7 years agoMerge tag 'armsoc-defconfig' of git://git.kernel.org/pub/scm/linux/kernel/git/arm...
Linus Torvalds [Thu, 23 Feb 2017 23:39:32 +0000 (15:39 -0800)]
Merge tag 'armsoc-defconfig' of git://git./linux/kernel/git/arm/arm-soc

Pull ARM SoC defconfig updates from Arnd Bergmann:
 "Defconfig additions, removals, etc. Almost all of them just turn on
  drivers that we want on some platform, usually after the driver has
  been merged into mainline.

  There is now a new defconfig file for tango4"

* tag 'armsoc-defconfig' of git://git.kernel.org/pub/scm/linux/kernel/git/arm/arm-soc: (44 commits)
  ARM: multi_v7_defconfig: enable pstore configs
  ARM: multi_v7_defconfig: enable some newly added crypto modules
  ARM: davinci_all_defconfig: enable SATA modules
  arm64: defconfig: enable CONFIG_MTD_NAND and CONFIG_MTD_NAND_DENALI_DT
  arm64: defconfig: enable CONFIG_MTD_BLOCK
  ARM: Import tango4_defconfig
  ARM: omap2plus_defconfig: Enable support for RTC M41T80
  ARM: omap2plus_defconfig: Enable support for micrell phys
  ARM: vf610m4: defconfig: enable EXT4 filesystem
  ARM: omap2plus_defconfig: Fix probe errors on UARTs 5 and 6
  arm64: defconfig: Enable NUMA and NUMA_BALANCING
  arm64: defconfig: enable SMMUv3 config
  ARM: davinci_all_defconfig: enable iio
  ARM: Keystone: Enable ARCH_HAS_RESET_CONTROLLER
  ARM: configs: stm32: Add RTC support in STM32 defconfig
  ARM: defconfig: qcom: add APQ8060 DragonBoard devices
  ARM: qcom_defconfig: enable thermal sensors
  ARM: qcom_defconfig: add ahci configs
  ARM: qcom_defconfig: add pcie and atl1c ethernet configs
  ARM: qcom_defconfig: add usb related configs
  ...

7 years agoMerge tag 'armsoc-arm64' of git://git.kernel.org/pub/scm/linux/kernel/git/arm/arm-soc
Linus Torvalds [Thu, 23 Feb 2017 23:35:57 +0000 (15:35 -0800)]
Merge tag 'armsoc-arm64' of git://git./linux/kernel/git/arm/arm-soc

Pull ARM SoC 64-bit updates from Arnd Bergmann:
 "Changes to platform code for 64-bit ARM platforms, only trivial stuff
  this time, a few defconfig changes to enable drivers, and a new entry
  for the Cavium ThunderX2 platform"

* tag 'armsoc-arm64' of git://git.kernel.org/pub/scm/linux/kernel/git/arm/arm-soc:
  MAINTAINERS: Add Cavium ThunderX2 entry
  arm64: add ARCH_THUNDER2 to defconfig
  arm64: add THUNDER2 processor family
  MAINTAINERS: Extend ARM/Mediatek SoC support section
  arm64: defconfig: enable CONFIG_MMC_SDHCI_CADENCE
  arm64: defconfig: enable XORv2 for Marvell Armada 7K/8K

7 years agoMerge tag 'armsoc-soc' of git://git.kernel.org/pub/scm/linux/kernel/git/arm/arm-soc
Linus Torvalds [Thu, 23 Feb 2017 23:33:54 +0000 (15:33 -0800)]
Merge tag 'armsoc-soc' of git://git./linux/kernel/git/arm/arm-soc

Pull ARM SoC platform updates from Arnd Bergmann:
 "In the SoC branch we normally collect classic arch/arm/mach-*
  contents, i.e. C code changes for SoC platforms. This release cycle
  the diffstat is quite nice, in that we're removing 3x the amount of
  code that's being added.

  The main reason for this is that there's a removal of camera drivers
  for Freescale i.MX chips (driver was removed so the device
  registration isn't needed any more). There's also removal of display
  initialization code for OMAP that is no longer needed.

  The rest are mostly minor tweaks and cleanups; constification on
  Samsung platforms, cleanup of ux500 platform data, purge of other
  unused platform data/device seutp on i.MX and other good stuff.

  New SoC support this cycle is for two Allwinner platforms, H2+ and
  V3s"

* tag 'armsoc-soc' of git://git.kernel.org/pub/scm/linux/kernel/git/arm/arm-soc: (55 commits)
  ARM: ux500: remove deleted file from Makefile
  ARM: ep93xx: Disable TS-72xx watchdog before uncompressing
  ARM: ux500: cut some platform data
  MAINTAINERS: Update for the current location of the bcm2835 tree.
  ARM: davinci: remove BUG_ON() from da850_register_sata()
  ARM: davinci: da850: model the SATA refclk
  ARM: davinci: da850: add con_id for the SATA clock
  ARM: davinci: da8xx-dt: add OF_DEV_AUXDATA entry for SATA
  arm: mvebu: support for SMP on 98DX3336 SoC
  dt-bindings: video: exynos7-decon: Remove obsolete samsung,power-domain property
  soc: dove: constify reset_control_ops structures
  ARM: mv78xx0: fix possible PCI buffer overflow
  MAINTAINERS: transfer maintainership for the EZX platform
  ARM: shmobile: rcar-gen2: Add more register documentation
  ARM: tegra: paz00: Fix __initdata placement
  ARM: OMAP: clock: Remove unused mpurate cmdline option
  ARM: davinci: add skeleton for pdata-quirks
  arm: sunxi: add support for V3s SoC
  ARM: OMAP2+: omap_hwmod: Add support for earlycon
  arm: hisi: drop extern hip01_cpu_die
  ...

7 years agoMerge tag 'armsoc-fixes-nc' of git://git.kernel.org/pub/scm/linux/kernel/git/arm...
Linus Torvalds [Thu, 23 Feb 2017 23:28:04 +0000 (15:28 -0800)]
Merge tag 'armsoc-fixes-nc' of git://git./linux/kernel/git/arm/arm-soc

Pull ARM SoC non-urgent fixes from Arnd Bergmann:
 "We sometimes collect non-critical fixes that come in during the later
  part of the merge window in a branch for the next release instead, and
  this is that contents for v4.11.

  Most of these are OMAP fixes, dealing with OMAP36/37 detection, quirks
  and setup. There's also some fixes for Davinci and a Kconfig fix for
  SCPI to only enable on ARM{,64}"

* tag 'armsoc-fixes-nc' of git://git.kernel.org/pub/scm/linux/kernel/git/arm/arm-soc:
  firmware: arm_scpi: Add hardware dependencies
  ARM: OMAP3: Fix SoC detection of OMAP36/37 Family
  ARM: OMAP5: Add HWMOD_SWSUP_SIDLE_ACT flag for UART
  ARM: dts: Fix compatible for ti81xx uarts for 8250
  ARM: dts: Fix am335x and dm814x scm syscon to probe children
  ARM: OMAP2+: Fix init for multiple quirks for the same SoC
  ARM: dts: Fix omap3 off mode pull defines
  bus: da850-mstpri: fix my e-mail address
  ARM: davinci: da850: fix da850_set_pll0rate()
  ARM: davinci: da850: coding style fix

7 years agoMerge branch 'drm-next-4.11' of git://people.freedesktop.org/~agd5f/linux into drm...
Dave Airlie [Thu, 23 Feb 2017 22:35:23 +0000 (08:35 +1000)]
Merge branch 'drm-next-4.11' of git://people.freedesktop.org/~agd5f/linux into drm-next

Some ttm/amd fixes.

* 'drm-next-4.11' of git://people.freedesktop.org/~agd5f/linux:
  drm/amd/powerplay: fix PSI feature on Polars12.
  drm/amdgpu: refuse to reserve io mem for split VRAM buffers
  drm/ttm: fix use-after-free races in vm fault handling
  drm/amd/amdgpu: post card if there is real hw resetting performed

7 years agoMerge tag 'drm/panel/for-4.11-rc1' of git://anongit.freedesktop.org/tegra/linux into...
Dave Airlie [Thu, 23 Feb 2017 22:29:03 +0000 (08:29 +1000)]
Merge tag 'drm/panel/for-4.11-rc1' of git://anongit.freedesktop.org/tegra/linux into drm-next

drm/panel: Changes for v4.11-rc1

This set contains a couple of cleanups as well as support for a few more
simple panels.

* tag 'drm/panel/for-4.11-rc1' of git://anongit.freedesktop.org/tegra/linux:
  drm/panel: simple: Specify bus width and flags for EDT displays
  drm/panel: simple: Add Netron DY E231732
  of: Add vendor prefix for Netron DY
  drm/panel: simple: Add support for Tianma TM070JDHG30
  of: Add vendor prefix for Tianma Micro-electronics
  drm/panel: simple: Add support BOE NV101WXMN51
  dt-bindings: display: Add BOE NV101WXMN51 panel binding
  drm/panel: Constify device node argument to of_drm_find_panel()

7 years agoMerge tag 'drm/tegra/for-4.11-rc1' of git://anongit.freedesktop.org/tegra/linux into...
Dave Airlie [Thu, 23 Feb 2017 22:26:48 +0000 (08:26 +1000)]
Merge tag 'drm/tegra/for-4.11-rc1' of git://anongit.freedesktop.org/tegra/linux into drm-next

drm/tegra: Changes for v4.11-rc1

Just a single change that hooks up the Tegra DRM parent device to the
correct device tree node.

* tag 'drm/tegra/for-4.11-rc1' of git://anongit.freedesktop.org/tegra/linux:
  gpu: host1x: Set OF node for new host1x devices

7 years agoMerge tag 'pci-v4.11-changes' of git://git.kernel.org/pub/scm/linux/kernel/git/helgaa...
Linus Torvalds [Thu, 23 Feb 2017 19:53:22 +0000 (11:53 -0800)]
Merge tag 'pci-v4.11-changes' of git://git./linux/kernel/git/helgaas/pci

Pull PCI updates from Bjorn Helgaas:

 - add ASPM L1 substate support

 - enable PCIe Extended Tags when supported

 - configure PCIe MPS settings on iProc, Versatile, X-Gene, and Xilinx

 - increase VPD access timeout

 - add ACS quirks for Intel Union Point, Qualcomm QDF2400 and QDF2432

 - use new pci_irq_alloc_vectors() in more drivers

 - fix MSI affinity memory leak

 - remove unused MSI interfaces and update documentation

 - remove unused AER .link_reset() callback

 - avoid pci_lock / p->pi_lock deadlock seen with perf

 - serialize sysfs enable/disable num_vfs operations

 - move DesignWare IP from drivers/pci/host/ to drivers/pci/dwc/ and
   refactor so we can support both hosts and endpoints

 - add DT ECAM-like support for HiSilicon Hip06/Hip07 controllers

 - add Rockchip system power management support

 - add Thunder-X cn81xx and cn83xx support

 - add Exynos 5440 PCIe PHY support

* tag 'pci-v4.11-changes' of git://git.kernel.org/pub/scm/linux/kernel/git/helgaas/pci: (93 commits)
  PCI: dwc: Remove dependency of designware on CONFIG_PCI
  PCI: dwc: Add CONFIG_PCIE_DW_HOST to enable PCI dwc host
  PCI: dwc: Split pcie-designware.c into host and core files
  PCI: dwc: designware: Fix style errors in pcie-designware.c
  PCI: dwc: designware: Parse "num-lanes" property in dw_pcie_setup_rc()
  PCI: dwc: all: Split struct pcie_port into host-only and core structures
  PCI: dwc: designware: Get device pointer at the start of dw_pcie_host_init()
  PCI: dwc: all: Rename cfg_read/cfg_write to read/write
  PCI: dwc: all: Use platform_set_drvdata() to save private data
  PCI: dwc: designware: Move register defines to designware header file
  PCI: dwc: Use PTR_ERR_OR_ZERO to simplify code
  PCI: dra7xx: Group PHY API invocations
  PCI: dra7xx: Enable MSI and legacy interrupts simultaneously
  PCI: dra7xx: Add support to force RC to work in GEN1 mode
  PCI: dra7xx: Simplify probe code with devm_gpiod_get_optional()
  PCI: Move DesignWare IP support to new drivers/pci/dwc/ directory
  PCI: exynos: Support the PHY generic framework
  Documentation: binding: Modify the exynos5440 PCIe binding
  phy: phy-exynos-pcie: Add support for Exynos PCIe PHY
  Documentation: samsung-phy: Add exynos-pcie-phy binding
  ...

7 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
Linus Torvalds [Thu, 23 Feb 2017 19:36:06 +0000 (11:36 -0800)]
Merge git://git./linux/kernel/git/davem/net

Pull networking fixes from David Miller:

 1) Some 'const'ing in qlogic networking drivers, from Bhumika Goyal.

 2) Fix scheduling while atomic in l2tp network namespace exit by
    deferring the work to the workqueue. From Ridge Kennedy.

 3) Fix use after free in dccp timewait handling, from Andrey Ryabinin.

 4) mlx5e CQE compression engine not initialized properly, from Tariq
    Toukan.

 5) Some UAPI header fixes from Dmitry V. Levin.

 6) Don't overwrite module parameter value in mlx4 driver, from Majd
    Dibbiny.

 7) Fix divide by zero in xt_hashlimit netfilter module, from Alban
    Browaeys.

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: (35 commits)
  bpf: Fix bpf_xdp_event_output
  net/mlx4_en: Use __skb_fill_page_desc()
  net/mlx4_core: Use cq quota in SRIOV when creating completion EQs
  net/mlx4_core: Fix VF overwrite of module param which disables DMFS on new probed PFs
  net/mlx4: Spoofcheck and zero MAC can't coexist
  net/mlx4: Change ENOTSUPP to EOPNOTSUPP
  uapi: fix linux/rds.h userspace compilation errors
  uapi: fix linux/seg6.h and linux/seg6_iptunnel.h userspace compilation errors
  lib: Remove string from parman config selection
  forcedeth: Remove return from a void function
  bpf: fix spelling mistake: "proccessed" -> "processed"
  uapi: fix linux/llc.h userspace compilation error
  uapi: fix linux/ip6_tunnel.h userspace compilation errors
  net/mlx5e: Fix wrong CQE decompression
  net/mlx5e: Update MPWQE stride size when modifying CQE compress state
  net/mlx5e: Fix broken CQE compression initialization
  net/mlx5e: Do not reduce LRO WQE size when not using build_skb
  net/mlx5e: Register/unregister vport representors on interface attach/detach
  net/mlx5e: s390 system compilation fix
  tcp: account for ts offset only if tsecr not zero
  ...

7 years agoMerge tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dledford/rdma
Linus Torvalds [Thu, 23 Feb 2017 19:27:49 +0000 (11:27 -0800)]
Merge tag 'for-linus' of git://git./linux/kernel/git/dledford/rdma

Pull Mellanox rdma updates from Doug Ledford:
 "Mellanox specific updates for 4.11 merge window

  Because the Mellanox code required being based on a net-next tree, I
  keept it separate from the remainder of the RDMA stack submission that
  is based on 4.10-rc3.

  This branch contains:

   - Various mlx4 and mlx5 fixes and minor changes

   - Support for adding a tag match rule to flow specs

   - Support for cvlan offload operation for raw ethernet QPs

   - A change to the core IB code to recognize raw eth capabilities and
     enumerate them (touches non-Mellanox code)

   - Implicit On-Demand Paging memory registration support"

* tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dledford/rdma: (40 commits)
  IB/mlx5: Fix configuration of port capabilities
  IB/mlx4: Take source GID by index from HW GID table
  IB/mlx5: Fix blue flame buffer size calculation
  IB/mlx4: Remove unused variable from function declaration
  IB: Query ports via the core instead of direct into the driver
  IB: Add protocol for USNIC
  IB/mlx4: Support raw packet protocol
  IB/mlx5: Support raw packet protocol
  IB/core: Add raw packet protocol
  IB/mlx5: Add implicit MR support
  IB/mlx5: Expose MR cache for mlx5_ib
  IB/mlx5: Add null_mkey access
  IB/umem: Indicate that process is being terminated
  IB/umem: Update on demand page (ODP) support
  IB/core: Add implicit MR flag
  IB/mlx5: Support creation of a WQ with scatter FCS offload
  IB/mlx5: Enable QP creation with cvlan offload
  IB/mlx5: Enable WQ creation and modification with cvlan offload
  IB/mlx5: Expose vlan offloads capabilities
  IB/uverbs: Enable QP creation with cvlan offload
  ...

7 years agoMerge tag 'vfio-v4.11-rc1' of git://github.com/awilliam/linux-vfio
Linus Torvalds [Thu, 23 Feb 2017 19:26:09 +0000 (11:26 -0800)]
Merge tag 'vfio-v4.11-rc1' of git://github.com/awilliam/linux-vfio

Pull VFIO updates from Alex Williamson:

 - Kconfig fixes for SPAPR_TCE_IOMMU=n (Michael Ellerman)

 - Module softdep rather than request_module to simplify usage from
   initrd (Alex Williamson)

 - Comment typo fix (Changbin Du)

* tag 'vfio-v4.11-rc1' of git://github.com/awilliam/linux-vfio:
  vfio: fix a typo in comment of function vfio_pin_pages
  vfio: Replace module request with softdep
  vfio/mdev: Use a module softdep for vfio_mdev
  vfio: Fix build break when SPAPR_TCE_IOMMU=n

7 years agobpf: Fix bpf_xdp_event_output
Martin KaFai Lau [Thu, 23 Feb 2017 18:40:34 +0000 (10:40 -0800)]
bpf: Fix bpf_xdp_event_output

Fix a typo. xdp->data instead of xdp should be copied to the perf-event's
dst_buff.

Fixes: 4de16969523c ("bpf: enable event output helper also for xdp types")
Reported-by: Huapeng Zhou <hzhou@fb.com>
Tested-by: Feixiong Zhang <feixiong@fb.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'linus' of git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6
Linus Torvalds [Thu, 23 Feb 2017 17:54:19 +0000 (09:54 -0800)]
Merge branch 'linus' of git://git./linux/kernel/git/herbert/crypto-2.6

Pull crypto update from Herbert Xu:
 "API:
   - Try to catch hash output overrun in testmgr
   - Introduce walksize attribute for batched walking
   - Make crypto_xor() and crypto_inc() alignment agnostic

  Algorithms:
   - Add time-invariant AES algorithm
   - Add standalone CBCMAC algorithm

  Drivers:
   - Add NEON acclerated chacha20 on ARM/ARM64
   - Expose AES-CTR as synchronous skcipher on ARM64
   - Add scalar AES implementation on ARM64
   - Improve scalar AES implementation on ARM
   - Improve NEON AES implementation on ARM/ARM64
   - Merge CRC32 and PMULL instruction based drivers on ARM64
   - Add NEON acclerated CBCMAC/CMAC/XCBC AES on ARM64
   - Add IPsec AUTHENC implementation in atmel
   - Add Support for Octeon-tx CPT Engine
   - Add Broadcom SPU driver
   - Add MediaTek driver"

* 'linus' of git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6: (142 commits)
  crypto: xts - Add ECB dependency
  crypto: cavium - switch to pci_alloc_irq_vectors
  crypto: cavium - switch to pci_alloc_irq_vectors
  crypto: cavium - remove dead MSI-X related define
  crypto: brcm - Avoid double free in ahash_finup()
  crypto: cavium - fix Kconfig dependencies
  crypto: cavium - cpt_bind_vq_to_grp could return an error code
  crypto: doc - fix typo
  hwrng: omap - update Kconfig help description
  crypto: ccm - drop unnecessary minimum 32-bit alignment
  crypto: ccm - honour alignmask of subordinate MAC cipher
  crypto: caam - fix state buffer DMA (un)mapping
  crypto: caam - abstract ahash request double buffering
  crypto: caam - fix error path for ctx_dma mapping failure
  crypto: caam - fix DMA API leaks for multiple setkey() calls
  crypto: caam - don't dma_map key for hash algorithms
  crypto: caam - use dma_map_sg() return code
  crypto: caam - replace sg_count() with sg_nents_for_len()
  crypto: caam - check sg_count() return value
  crypto: caam - fix HW S/G in ablkcipher_giv_edesc_alloc()
  ..

7 years agoMerge tag 'rpmsg-v4.11' of git://github.com/andersson/remoteproc
Linus Torvalds [Thu, 23 Feb 2017 17:41:03 +0000 (09:41 -0800)]
Merge tag 'rpmsg-v4.11' of git://github.com/andersson/remoteproc

Pull rpmsg updates from Bjorn Andersson:
 "This introduces an interface for user space to directly communicate on
  rpmsg endpoints without the implementation of specific kernel drivers,
  which is useful for e.g. debug channels:

* tag 'rpmsg-v4.11' of git://github.com/andersson/remoteproc:
  rpmsg: rpmsg_create_ept() returns NULL on error
  rpmsg: qcom: smd: Return positively when not enabled
  rpmsg: unlock on error in rpmsg_eptdev_read()
  rpmsg: char: add CONFIG_NET dependency
  rpmsg: smd: Register rpmsg user space interface for edges
  rpmsg: Driver for user space endpoint interface
  rpmsg: qcom_smd: Implement endpoint "poll"
  rpmsg: Introduce "poll" to endpoint ops
  rpmsg: qcom_smd: Add support for "label" property