openwrt/staging/blogic.git
6 years agohugetlb: harmonize hugetlb.h arch specific defines with pgtable.h
Alexandre Ghiti [Fri, 26 Oct 2018 22:07:59 +0000 (15:07 -0700)]
hugetlb: harmonize hugetlb.h arch specific defines with pgtable.h

In order to reduce copy/paste of functions across architectures and then
make riscv hugetlb port (and future ports) simpler and smaller, this
patchset intends to factorize the numerous hugetlb primitives that are
defined across all the architectures.

Except for prepare_hugepage_range, this patchset moves the versions that
are just pass-through to standard pte primitives into
asm-generic/hugetlb.h by using the same #ifdef semantic that can be found
in asm-generic/pgtable.h, i.e.  __HAVE_ARCH_***.

s390 architecture has not been tackled in this serie since it does not use
asm-generic/hugetlb.h at all.

This patchset has been compiled on all addressed architectures with
success (except for parisc, but the problem does not come from this
series).

This patch (of 11):

asm-generic/hugetlb.h proposes generic implementations of hugetlb related
functions: use __HAVE_ARCH_HUGE* defines in order to make arch specific
implementations of hugetlb functions consistent with pgtable.h scheme.

Link: http://lkml.kernel.org/r/20180920060358.16606-2-alex@ghiti.fr
Signed-off-by: Alexandre Ghiti <alex@ghiti.fr>
Reviewed-by: Luiz Capitulino <lcapitulino@redhat.com>
Reviewed-by: Mike Kravetz <mike.kravetz@oracle.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com> [arm64]
Cc: Russell King <linux@armlinux.org.uk>
Cc: Will Deacon <will.deacon@arm.com>
Cc: Tony Luck <tony.luck@intel.com>
Cc: Fenghua Yu <fenghua.yu@intel.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: Paul Burton <paul.burton@mips.com>
Cc: James Hogan <jhogan@kernel.org>
Cc: James E.J. Bottomley <jejb@parisc-linux.org>
Cc: Helge Deller <deller@gmx.de>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Yoshinori Sato <ysato@users.sourceforge.jp>
Cc: Rich Felker <dalias@libc.org>
Cc: David S. Miller <davem@davemloft.net>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Ingo Molnar <mingo@kernel.org> [x86]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agomm: remove unnecessary local variable addr in __get_user_pages_fast()
Wei Yang [Fri, 26 Oct 2018 22:07:55 +0000 (15:07 -0700)]
mm: remove unnecessary local variable addr in __get_user_pages_fast()

The local variable `addr' in __get_user_pages_fast() is just a shadow of
`start'.  Since `start' never changes after assignment to `addr', it is
fine to replace `start' with it.

Also the meaning of [start, end] is more obvious than [addr, end] when
passed to gup_pgd_range().

Link: http://lkml.kernel.org/r/20180925021448.20265-1-richard.weiyang@gmail.com
Signed-off-by: Wei Yang <richard.weiyang@gmail.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agomm: defer ZONE_DEVICE page initialization to the point where we init pgmap
Alexander Duyck [Fri, 26 Oct 2018 22:07:52 +0000 (15:07 -0700)]
mm: defer ZONE_DEVICE page initialization to the point where we init pgmap

The ZONE_DEVICE pages were being initialized in two locations.  One was
with the memory_hotplug lock held and another was outside of that lock.
The problem with this is that it was nearly doubling the memory
initialization time.  Instead of doing this twice, once while holding a
global lock and once without, I am opting to defer the initialization to
the one outside of the lock.  This allows us to avoid serializing the
overhead for memory init and we can instead focus on per-node init times.

One issue I encountered is that devm_memremap_pages and
hmm_devmmem_pages_create were initializing only the pgmap field the same
way.  One wasn't initializing hmm_data, and the other was initializing it
to a poison value.  Since this is something that is exposed to the driver
in the case of hmm I am opting for a third option and just initializing
hmm_data to 0 since this is going to be exposed to unknown third party
drivers.

[alexander.h.duyck@linux.intel.com: fix reference count for pgmap in devm_memremap_pages]
Link: http://lkml.kernel.org/r/20181008233404.1909.37302.stgit@localhost.localdomain
Link: http://lkml.kernel.org/r/20180925202053.3576.66039.stgit@localhost.localdomain
Signed-off-by: Alexander Duyck <alexander.h.duyck@linux.intel.com>
Reviewed-by: Pavel Tatashin <pavel.tatashin@microsoft.com>
Tested-by: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Michal Hocko <mhocko@suse.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agomm: create non-atomic version of SetPageReserved for init use
Alexander Duyck [Fri, 26 Oct 2018 22:07:48 +0000 (15:07 -0700)]
mm: create non-atomic version of SetPageReserved for init use

It doesn't make much sense to use the atomic SetPageReserved at init time
when we are using memset to clear the memory and manipulating the page
flags via simple "&=" and "|=" operations in __init_single_page.

This patch adds a non-atomic version __SetPageReserved that can be used
during page init and shows about a 10% improvement in initialization times
on the systems I have available for testing.  On those systems I saw
initialization times drop from around 35 seconds to around 32 seconds to
initialize a 3TB block of persistent memory.  I believe the main advantage
of this is that it allows for more compiler optimization as the __set_bit
operation can be reordered whereas the atomic version cannot.

I tried adding a bit of documentation based on f1dd2cd13c4 ("mm,
memory_hotplug: do not associate hotadded memory to zones until online").

Ideally the reserved flag should be set earlier since there is a brief
window where the page is initialization via __init_single_page and we have
not set the PG_Reserved flag.  I'm leaving that for a future patch set as
that will require a more significant refactor.

Link: http://lkml.kernel.org/r/20180925202018.3576.11607.stgit@localhost.localdomain
Signed-off-by: Alexander Duyck <alexander.h.duyck@linux.intel.com>
Reviewed-by: Pavel Tatashin <pavel.tatashin@microsoft.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agomm: provide kernel parameter to allow disabling page init poisoning
Alexander Duyck [Fri, 26 Oct 2018 22:07:45 +0000 (15:07 -0700)]
mm: provide kernel parameter to allow disabling page init poisoning

Patch series "Address issues slowing persistent memory initialization", v5.

The main thing this patch set achieves is that it allows us to initialize
each node worth of persistent memory independently.  As a result we reduce
page init time by about 2 minutes because instead of taking 30 to 40
seconds per node and going through each node one at a time, we process all
4 nodes in parallel in the case of a 12TB persistent memory setup spread
evenly over 4 nodes.

This patch (of 3):

On systems with a large amount of memory it can take a significant amount
of time to initialize all of the page structs with the PAGE_POISON_PATTERN
value.  I have seen it take over 2 minutes to initialize a system with
over 12TB of RAM.

In order to work around the issue I had to disable CONFIG_DEBUG_VM and
then the boot time returned to something much more reasonable as the
arch_add_memory call completed in milliseconds versus seconds.  However in
doing that I had to disable all of the other VM debugging on the system.

In order to work around a kernel that might have CONFIG_DEBUG_VM enabled
on a system that has a large amount of memory I have added a new kernel
parameter named "vm_debug" that can be set to "-" in order to disable it.

Link: http://lkml.kernel.org/r/20180925201921.3576.84239.stgit@localhost.localdomain
Reviewed-by: Pavel Tatashin <pavel.tatashin@microsoft.com>
Signed-off-by: Alexander Duyck <alexander.h.duyck@linux.intel.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Michal Hocko <mhocko@suse.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agomemcg: remove memcg_kmem_skip_account
Shakeel Butt [Fri, 26 Oct 2018 22:07:41 +0000 (15:07 -0700)]
memcg: remove memcg_kmem_skip_account

The flag memcg_kmem_skip_account was added during the era of opt-out kmem
accounting.  There is no need for such flag in the opt-in world as there
aren't any __GFP_ACCOUNT allocations within memcg_create_cache_enqueue().

Link: http://lkml.kernel.org/r/20180919004501.178023-1-shakeelb@google.com
Signed-off-by: Shakeel Butt <shakeelb@google.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Greg Thelen <gthelen@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agomm/memory_hotplug.c: clean up node_states_check_changes_offline()
Oscar Salvador [Fri, 26 Oct 2018 22:07:38 +0000 (15:07 -0700)]
mm/memory_hotplug.c: clean up node_states_check_changes_offline()

This patch, as the previous one, gets rid of the wrong if statements.
While at it, I realized that the comments are sometimes very confusing,
to say the least, and wrong.
For example:

___
zone_last = ZONE_MOVABLE;

/*
 * check whether node_states[N_HIGH_MEMORY] will be changed
 * If we try to offline the last present @nr_pages from the node,
 * we can determind we will need to clear the node from
 * node_states[N_HIGH_MEMORY].
 */

for (; zt <= zone_last; zt++)
        present_pages += pgdat->node_zones[zt].present_pages;
if (nr_pages >= present_pages)
        arg->status_change_nid = zone_to_nid(zone);
else
        arg->status_change_nid = -1;
___

In case the node gets empry, it must be removed from N_MEMORY.  We already
check N_HIGH_MEMORY a bit above within the CONFIG_HIGHMEM ifdef code.  Not
to say that status_change_nid is for N_MEMORY, and not for N_HIGH_MEMORY.

So I re-wrote some of the comments to what I think is better.

[osalvador@suse.de: address feedback from Pavel]
Link: http://lkml.kernel.org/r/20180921132634.10103-5-osalvador@techadventures.net
Link: http://lkml.kernel.org/r/20180919100819.25518-6-osalvador@techadventures.net
Signed-off-by: Oscar Salvador <osalvador@suse.de>
Reviewed-by: Pavel Tatashin <pavel.tatashin@microsoft.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Jonathan Cameron <Jonathan.Cameron@huawei.com>
Cc: <yasu.isimatu@gmail.com>
Cc: Mathieu Malaterre <malat@debian.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agomm/memory_hotplug.c: simplify node_states_check_changes_online
Oscar Salvador [Fri, 26 Oct 2018 22:07:34 +0000 (15:07 -0700)]
mm/memory_hotplug.c: simplify node_states_check_changes_online

While looking at node_states_check_changes_online, I stumbled upon some
confusing things.

Right after entering the function, we find this:

if (N_MEMORY == N_NORMAL_MEMORY)
        zone_last = ZONE_MOVABLE;

This is wrong.
N_MEMORY cannot really be equal to N_NORMAL_MEMORY.
My guess is that this wanted to be something like:

if (N_NORMAL_MEMORY == N_HIGH_MEMORY)

to check if we have CONFIG_HIGHMEM.

Later on, in the CONFIG_HIGHMEM block, we have:

if (N_MEMORY == N_HIGH_MEMORY)
        zone_last = ZONE_MOVABLE;

Again, this is wrong, and will never be evaluated to true.

Besides removing these wrong if statements, I simplified the function a
bit.

[osalvador@suse.de: address feedback from Pavel]
Link: http://lkml.kernel.org/r/20180921132634.10103-4-osalvador@techadventures.net
Link: http://lkml.kernel.org/r/20180919100819.25518-5-osalvador@techadventures.net
Signed-off-by: Oscar Salvador <osalvador@suse.de>
Reviewed-by: Pavel Tatashin <pavel.tatashin@microsoft.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Jonathan Cameron <Jonathan.Cameron@huawei.com>
Cc: Mathieu Malaterre <malat@debian.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: <yasu.isimatu@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agomm/memory_hotplug.c: tidy up node_states_clear_node()
Oscar Salvador [Fri, 26 Oct 2018 22:07:28 +0000 (15:07 -0700)]
mm/memory_hotplug.c: tidy up node_states_clear_node()

node_states_clear has the following if statements:

if ((N_MEMORY != N_NORMAL_MEMORY) &&
    (arg->status_change_nid_high >= 0))
        ...

if ((N_MEMORY != N_HIGH_MEMORY) &&
    (arg->status_change_nid >= 0))
        ...

N_MEMORY can never be equal to neither N_NORMAL_MEMORY nor
N_HIGH_MEMORY.

Similar problem was found in [1].
Since this is wrong, let us get rid of it.

[1] https://patchwork.kernel.org/patch/10579155/

Link: http://lkml.kernel.org/r/20180919100819.25518-4-osalvador@techadventures.net
Signed-off-by: Oscar Salvador <osalvador@suse.de>
Reviewed-by: Pavel Tatashin <pavel.tatashin@microsoft.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Jonathan Cameron <Jonathan.Cameron@huawei.com>
Cc: Mathieu Malaterre <malat@debian.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: <yasu.isimatu@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agomm/memory_hotplug.c: spare unnecessary calls to node_set_state
Oscar Salvador [Fri, 26 Oct 2018 22:07:25 +0000 (15:07 -0700)]
mm/memory_hotplug.c: spare unnecessary calls to node_set_state

In node_states_check_changes_online, we check if the node will have to be
set for any of the N_*_MEMORY states after the pages have been onlined.

Later on, we perform the activation in node_states_set_node.  Currently,
in node_states_set_node we set the node to N_MEMORY unconditionally.

This means that we call node_set_state for N_MEMORY every time pages go
online, but we only need to do it if the node has not yet been set for
N_MEMORY.

Fix this by checking status_change_nid.

Link: http://lkml.kernel.org/r/20180919100819.25518-2-osalvador@techadventures.net
Signed-off-by: Oscar Salvador <osalvador@suse.de>
Reviewed-by: Pavel Tatashin <pavel.tatashin@microsoft.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Jonathan Cameron <Jonathan.Cameron@huawei.com>
Cc: <yasu.isimatu@gmail.com>
Cc: Mathieu Malaterre <malat@debian.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agomm/filemap.c: Use existing variable
haiqing.shq [Fri, 26 Oct 2018 22:07:22 +0000 (15:07 -0700)]
mm/filemap.c: Use existing variable

Use the variable write_len instead of ov_iter_count(from).

Link: http://lkml.kernel.org/r/1537375855-2088-1-git-send-email-leviathan0992@gmail.com
Signed-off-by: haiqing.shq <leviathan0992@gmail.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Jan Kara <jack@suse.cz>
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>
6 years agomm: unmap VM_PFNMAP mappings with optimized path
Yang Shi [Fri, 26 Oct 2018 22:07:18 +0000 (15:07 -0700)]
mm: unmap VM_PFNMAP mappings with optimized path

When unmapping VM_PFNMAP mappings, vm flags need to be updated.  Since the
vmas have been detached, so it sounds safe to update vm flags with read
mmap_sem.

Link: http://lkml.kernel.org/r/1537376621-51150-4-git-send-email-yang.shi@linux.alibaba.com
Signed-off-by: Yang Shi <yang.shi@linux.alibaba.com>
Reviewed-by: Matthew Wilcox <willy@infradead.org>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agomm: unmap VM_HUGETLB mappings with optimized path
Yang Shi [Fri, 26 Oct 2018 22:07:15 +0000 (15:07 -0700)]
mm: unmap VM_HUGETLB mappings with optimized path

When unmapping VM_HUGETLB mappings, vm flags need to be updated.  Since
the vmas have been detached, so it sounds safe to update vm flags with
read mmap_sem.

Link: http://lkml.kernel.org/r/1537376621-51150-3-git-send-email-yang.shi@linux.alibaba.com
Signed-off-by: Yang Shi <yang.shi@linux.alibaba.com>
Reviewed-by: Matthew Wilcox <willy@infradead.org>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agomm: mmap: zap pages with read mmap_sem in munmap
Yang Shi [Fri, 26 Oct 2018 22:07:11 +0000 (15:07 -0700)]
mm: mmap: zap pages with read mmap_sem in munmap

Patch series "mm: zap pages with read mmap_sem in munmap for large
mapping", v11.

Background:
Recently, when we ran some vm scalability tests on machines with large memory,
we ran into a couple of mmap_sem scalability issues when unmapping large memory
space, please refer to https://lkml.org/lkml/2017/12/14/733 and
https://lkml.org/lkml/2018/2/20/576.

History:
Then akpm suggested to unmap large mapping section by section and drop mmap_sem
at a time to mitigate it (see https://lkml.org/lkml/2018/3/6/784).

V1 patch series was submitted to the mailing list per Andrew's suggestion
(see https://lkml.org/lkml/2018/3/20/786).  Then I received a lot great
feedback and suggestions.

Then this topic was discussed on LSFMM summit 2018.  In the summit, Michal
Hocko suggested (also in the v1 patches review) to try "two phases"
approach.  Zapping pages with read mmap_sem, then doing via cleanup with
write mmap_sem (for discussion detail, see
https://lwn.net/Articles/753269/)

Approach:
Zapping pages is the most time consuming part, according to the suggestion from
Michal Hocko [1], zapping pages can be done with holding read mmap_sem, like
what MADV_DONTNEED does. Then re-acquire write mmap_sem to cleanup vmas.

But, we can't call MADV_DONTNEED directly, since there are two major drawbacks:
  * The unexpected state from PF if it wins the race in the middle of munmap.
    It may return zero page, instead of the content or SIGSEGV.
  * Can't handle VM_LOCKED | VM_HUGETLB | VM_PFNMAP and uprobe mappings, which
    is a showstopper from akpm

But, some part may need write mmap_sem, for example, vma splitting. So,
the design is as follows:
        acquire write mmap_sem
        lookup vmas (find and split vmas)
        deal with special mappings
        detach vmas
        downgrade_write

        zap pages
        free page tables
        release mmap_sem

The vm events with read mmap_sem may come in during page zapping, but
since vmas have been detached before, they, i.e.  page fault, gup, etc,
will not be able to find valid vma, then just return SIGSEGV or -EFAULT as
expected.

If the vma has VM_HUGETLB | VM_PFNMAP, they are considered as special
mappings.  They will be handled by falling back to regular do_munmap()
with exclusive mmap_sem held in this patch since they may update vm flags.

But, with the "detach vmas first" approach, the vmas have been detached
when vm flags are updated, so it sounds safe to update vm flags with read
mmap_sem for this specific case.  So, VM_HUGETLB and VM_PFNMAP will be
handled by using the optimized path in the following separate patches for
bisectable sake.

Unmapping uprobe areas may need update mm flags (MMF_RECALC_UPROBES).
However it is fine to have false-positive MMF_RECALC_UPROBES according to
uprobes developer.  So, uprobe unmap will not be handled by the regular
path.

With the "detach vmas first" approach we don't have to re-acquire mmap_sem
again to clean up vmas to avoid race window which might get the address
space changed since downgrade_write() doesn't release the lock to lead
regression, which simply downgrades to read lock.

And, since the lock acquire/release cost is managed to the minimum and
almost as same as before, the optimization could be extended to any size
of mapping without incurring significant penalty to small mappings.

For the time being, just do this in munmap syscall path.  Other
vm_munmap() or do_munmap() call sites (i.e mmap, mremap, etc) remain
intact due to some implementation difficulties since they acquire write
mmap_sem from very beginning and hold it until the end, do_munmap() might
be called in the middle.  But, the optimized do_munmap would like to be
called without mmap_sem held so that we can do the optimization.  So, if
we want to do the similar optimization for mmap/mremap path, I'm afraid we
would have to redesign them.  mremap might be called on very large area
depending on the usecases, the optimization to it will be considered in
the future.

This patch (of 3):

When running some mmap/munmap scalability tests with large memory (i.e.
> 300GB), the below hung task issue may happen occasionally.

INFO: task ps:14018 blocked for more than 120 seconds.
       Tainted: G            E 4.9.79-009.ali3000.alios7.x86_64 #1
 "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this
message.
 ps              D    0 14018      1 0x00000004
  ffff885582f84000 ffff885e8682f000 ffff880972943000 ffff885ebf499bc0
  ffff8828ee120000 ffffc900349bfca8 ffffffff817154d0 0000000000000040
  00ffffff812f872a ffff885ebf499bc0 024000d000948300 ffff880972943000
 Call Trace:
  [<ffffffff817154d0>] ? __schedule+0x250/0x730
  [<ffffffff817159e6>] schedule+0x36/0x80
  [<ffffffff81718560>] rwsem_down_read_failed+0xf0/0x150
  [<ffffffff81390a28>] call_rwsem_down_read_failed+0x18/0x30
  [<ffffffff81717db0>] down_read+0x20/0x40
  [<ffffffff812b9439>] proc_pid_cmdline_read+0xd9/0x4e0
  [<ffffffff81253c95>] ? do_filp_open+0xa5/0x100
  [<ffffffff81241d87>] __vfs_read+0x37/0x150
  [<ffffffff812f824b>] ? security_file_permission+0x9b/0xc0
  [<ffffffff81242266>] vfs_read+0x96/0x130
  [<ffffffff812437b5>] SyS_read+0x55/0xc0
  [<ffffffff8171a6da>] entry_SYSCALL_64_fastpath+0x1a/0xc5

It is because munmap holds mmap_sem exclusively from very beginning to all
the way down to the end, and doesn't release it in the middle.  When
unmapping large mapping, it may take long time (take ~18 seconds to unmap
320GB mapping with every single page mapped on an idle machine).

Zapping pages is the most time consuming part, according to the suggestion
from Michal Hocko [1], zapping pages can be done with holding read
mmap_sem, like what MADV_DONTNEED does.  Then re-acquire write mmap_sem to
cleanup vmas.

But, some part may need write mmap_sem, for example, vma splitting. So,
the design is as follows:
        acquire write mmap_sem
        lookup vmas (find and split vmas)
        deal with special mappings
        detach vmas
        downgrade_write

        zap pages
        free page tables
        release mmap_sem

The vm events with read mmap_sem may come in during page zapping, but
since vmas have been detached before, they, i.e.  page fault, gup, etc,
will not be able to find valid vma, then just return SIGSEGV or -EFAULT as
expected.

If the vma has VM_HUGETLB | VM_PFNMAP, they are considered as special
mappings.  They will be handled by without downgrading mmap_sem in this
patch since they may update vm flags.

But, with the "detach vmas first" approach, the vmas have been detached
when vm flags are updated, so it sounds safe to update vm flags with read
mmap_sem for this specific case.  So, VM_HUGETLB and VM_PFNMAP will be
handled by using the optimized path in the following separate patches for
bisectable sake.

Unmapping uprobe areas may need update mm flags (MMF_RECALC_UPROBES).
However it is fine to have false-positive MMF_RECALC_UPROBES according to
uprobes developer.

With the "detach vmas first" approach we don't have to re-acquire mmap_sem
again to clean up vmas to avoid race window which might get the address
space changed since downgrade_write() doesn't release the lock to lead
regression, which simply downgrades to read lock.

And, since the lock acquire/release cost is managed to the minimum and
almost as same as before, the optimization could be extended to any size
of mapping without incurring significant penalty to small mappings.

For the time being, just do this in munmap syscall path.  Other
vm_munmap() or do_munmap() call sites (i.e mmap, mremap, etc) remain
intact due to some implementation difficulties since they acquire write
mmap_sem from very beginning and hold it until the end, do_munmap() might
be called in the middle.  But, the optimized do_munmap would like to be
called without mmap_sem held so that we can do the optimization.  So, if
we want to do the similar optimization for mmap/mremap path, I'm afraid we
would have to redesign them.  mremap might be called on very large area
depending on the usecases, the optimization to it will be considered in
the future.

With the patches, exclusive mmap_sem hold time when munmap a 80GB address
space on a machine with 32 cores of E5-2680 @ 2.70GHz dropped to us level
from second.

munmap_test-15002 [008]   594.380138: funcgraph_entry: |
__vm_munmap() {
munmap_test-15002 [008]   594.380146: funcgraph_entry:      !2485684 us
|    unmap_region();
munmap_test-15002 [008]   596.865836: funcgraph_exit:       !2485692 us
|  }

Here the execution time of unmap_region() is used to evaluate the time of
holding read mmap_sem, then the remaining time is used with holding
exclusive lock.

[1] https://lwn.net/Articles/753269/

Link: http://lkml.kernel.org/r/1537376621-51150-2-git-send-email-yang.shi@linux.alibaba.com
Signed-off-by: Yang Shi <yang.shi@linux.alibaba.com>Suggested-by: Michal Hocko <mhocko@kernel.org>
Suggested-by: Kirill A. Shutemov <kirill@shutemov.name>
Suggested-by: Matthew Wilcox <willy@infradead.org>
Reviewed-by: Matthew Wilcox <willy@infradead.org>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Laurent Dufour <ldufour@linux.vnet.ibm.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agovfree: add debug might_sleep()
Andrey Ryabinin [Fri, 26 Oct 2018 22:07:07 +0000 (15:07 -0700)]
vfree: add debug might_sleep()

Add might_sleep() call to vfree() to catch potential sleep-in-atomic bugs
earlier.

[aryabinin@virtuozzo.com: drop might_sleep_if() from kvfree()]
Link: http://lkml.kernel.org/r/7e19e4df-b1a6-29bd-9ae7-0266d50bef1d@virtuozzo.com
Link: http://lkml.kernel.org/r/20180914130512.10394-3-aryabinin@virtuozzo.com
Signed-off-by: Andrey Ryabinin <aryabinin@virtuozzo.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agomm/vmalloc.c: improve vfree() kerneldoc
Andrey Ryabinin [Fri, 26 Oct 2018 22:07:03 +0000 (15:07 -0700)]
mm/vmalloc.c: improve vfree() kerneldoc

vfree() might sleep if called not in interrupt context.  Explain that in
the comment.

Link: http://lkml.kernel.org/r/20180914130512.10394-2-aryabinin@virtuozzo.com
Signed-off-by: Andrey Ryabinin <aryabinin@virtuozzo.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agokvfree(): fix misleading comment
Andrey Ryabinin [Fri, 26 Oct 2018 22:07:00 +0000 (15:07 -0700)]
kvfree(): fix misleading comment

vfree() might sleep if called not in interrupt context.  So does kvfree()
too.  Fix misleading kvfree()'s comment about allowed context.

Link: http://lkml.kernel.org/r/20180914130512.10394-1-aryabinin@virtuozzo.com
Fixes: 04b8e946075d ("mm/util.c: improve kvfree() kerneldoc")
Signed-off-by: Andrey Ryabinin <aryabinin@virtuozzo.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agomm/mempolicy.c: use match_string() helper to simplify the code
zhong jiang [Fri, 26 Oct 2018 22:06:57 +0000 (15:06 -0700)]
mm/mempolicy.c: use match_string() helper to simplify the code

match_string() returns the index of an array for a matching string, which
can be used intead of open coded implementation.

Link: http://lkml.kernel.org/r/1536988365-50310-1-git-send-email-zhongjiang@huawei.com
Signed-off-by: zhong jiang <zhongjiang@huawei.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.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>
6 years agomm/swap.c: remove duplicated include
YueHaibing [Fri, 26 Oct 2018 22:06:53 +0000 (15:06 -0700)]
mm/swap.c: remove duplicated include

Remove duplicated include linux/memremap.h

Link: http://lkml.kernel.org/r/20180917131308.16420-1-yuehaibing@huawei.com
Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agomm, page_alloc: drop should_suppress_show_mem
Michal Hocko [Fri, 26 Oct 2018 22:06:49 +0000 (15:06 -0700)]
mm, page_alloc: drop should_suppress_show_mem

should_suppress_show_mem() was introduced to reduce the overhead of
show_mem on large NUMA systems.  Things have changed since then though.
Namely c78e93630d15 ("mm: do not walk all of system memory during
show_mem") has reduced the overhead considerably.

Moreover warn_alloc_show_mem clears SHOW_MEM_FILTER_NODES when called from
the IRQ context already so we are not printing per node stats.

Remove should_suppress_show_mem because we are losing potentially
interesting information about allocation failures.  We have seen a bug
report where system gets unresponsive under memory pressure and there is
only

kernel: [2032243.696888] qlge 0000:8b:00.1 ql1: Could not get a page chunk, i=8, clean_idx =200 .
kernel: [2032243.710725] swapper/7: page allocation failure: order:1, mode:0x1084120(GFP_ATOMIC|__GFP_COLD|__GFP_COMP)

without an additional information for debugging.  It would be great to see
the state of the page allocator at the moment.

Link: http://lkml.kernel.org/r/20180907114334.7088-1-mhocko@kernel.org
Signed-off-by: Michal Hocko <mhocko@suse.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: David Rientjes <rientjes@google.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>
6 years agomm/memcontrol.c: fix memory.stat item ordering
Johannes Weiner [Fri, 26 Oct 2018 22:06:45 +0000 (15:06 -0700)]
mm/memcontrol.c: fix memory.stat item ordering

The refault stats go better with the page fault stats, and are of
higher interest than the stats on LRU operations. In fact they used to
be grouped together; when the LRU operation stats were added later on,
they were wedged in between.

Move them back together. Documentation/admin-guide/cgroup-v2.rst
already lists them in the right order.

Link: http://lkml.kernel.org/r/20181010140239.GA2527@cmpxchg.org
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agomm: zero-seek shrinkers
Johannes Weiner [Fri, 26 Oct 2018 22:06:42 +0000 (15:06 -0700)]
mm: zero-seek shrinkers

The page cache and most shrinkable slab caches hold data that has been
read from disk, but there are some caches that only cache CPU work, such
as the dentry and inode caches of procfs and sysfs, as well as the subset
of radix tree nodes that track non-resident page cache.

Currently, all these are shrunk at the same rate: using DEFAULT_SEEKS for
the shrinker's seeks setting tells the reclaim algorithm that for every
two page cache pages scanned it should scan one slab object.

This is a bogus setting.  A virtual inode that required no IO to create is
not twice as valuable as a page cache page; shadow cache entries with
eviction distances beyond the size of memory aren't either.

In most cases, the behavior in practice is still fine.  Such virtual
caches don't tend to grow and assert themselves aggressively, and usually
get picked up before they cause problems.  But there are scenarios where
that's not true.

Our database workloads suffer from two of those.  For one, their file
workingset is several times bigger than available memory, which has the
kernel aggressively create shadow page cache entries for the non-resident
parts of it.  The workingset code does tell the VM that most of these are
expendable, but the VM ends up balancing them 2:1 to cache pages as per
the seeks setting.  This is a huge waste of memory.

These workloads also deal with tens of thousands of open files and use
/proc for introspection, which ends up growing the proc_inode_cache to
absurdly large sizes - again at the cost of valuable cache space, which
isn't a reasonable trade-off, given that proc inodes can be re-created
without involving the disk.

This patch implements a "zero-seek" setting for shrinkers that results in
a target ratio of 0:1 between their objects and IO-backed caches.  This
allows such virtual caches to grow when memory is available (they do
cache/avoid CPU work after all), but effectively disables them as soon as
IO-backed objects are under pressure.

It then switches the shrinkers for procfs and sysfs metadata, as well as
excess page cache shadow nodes, to the new zero-seek setting.

Link: http://lkml.kernel.org/r/20181009184732.762-5-hannes@cmpxchg.org
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Reported-by: Domas Mituzas <dmituzas@fb.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Rik van Riel <riel@surriel.com>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agomm: workingset: add vmstat counter for shadow nodes
Johannes Weiner [Fri, 26 Oct 2018 22:06:39 +0000 (15:06 -0700)]
mm: workingset: add vmstat counter for shadow nodes

Make it easier to catch bugs in the shadow node shrinker by adding a
counter for the shadow nodes in circulation.

[akpm@linux-foundation.org: assert that irqs are disabled, for __inc_lruvec_page_state()]
[akpm@linux-foundation.org: s/WARN_ON_ONCE/VM_WARN_ON_ONCE/, per Johannes]
Link: http://lkml.kernel.org/r/20181009184732.762-4-hannes@cmpxchg.org
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
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>
6 years agomm: workingset: use cheaper __inc_lruvec_state in irqsafe node reclaim
Johannes Weiner [Fri, 26 Oct 2018 22:06:35 +0000 (15:06 -0700)]
mm: workingset: use cheaper __inc_lruvec_state in irqsafe node reclaim

No need to use the preemption-safe lruvec state function inside the
reclaim region that has irqs disabled.

Link: http://lkml.kernel.org/r/20181009184732.762-3-hannes@cmpxchg.org
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Rik van Riel <riel@surriel.com>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agopsi: cgroup support
Johannes Weiner [Fri, 26 Oct 2018 22:06:31 +0000 (15:06 -0700)]
psi: cgroup support

On a system that executes multiple cgrouped jobs and independent
workloads, we don't just care about the health of the overall system, but
also that of individual jobs, so that we can ensure individual job health,
fairness between jobs, or prioritize some jobs over others.

This patch implements pressure stall tracking for cgroups.  In kernels
with CONFIG_PSI=y, cgroup2 groups will have cpu.pressure, memory.pressure,
and io.pressure files that track aggregate pressure stall times for only
the tasks inside the cgroup.

Link: http://lkml.kernel.org/r/20180828172258.3185-10-hannes@cmpxchg.org
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Tejun Heo <tj@kernel.org>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Tested-by: Daniel Drake <drake@endlessm.com>
Tested-by: Suren Baghdasaryan <surenb@google.com>
Cc: Christopher Lameter <cl@linux.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Johannes Weiner <jweiner@fb.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Enderborg <peter.enderborg@sony.com>
Cc: Randy Dunlap <rdunlap@infradead.org>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Vinayak Menon <vinmenon@codeaurora.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agopsi: pressure stall information for CPU, memory, and IO
Johannes Weiner [Fri, 26 Oct 2018 22:06:27 +0000 (15:06 -0700)]
psi: pressure stall information for CPU, memory, and IO

When systems are overcommitted and resources become contended, it's hard
to tell exactly the impact this has on workload productivity, or how close
the system is to lockups and OOM kills.  In particular, when machines work
multiple jobs concurrently, the impact of overcommit in terms of latency
and throughput on the individual job can be enormous.

In order to maximize hardware utilization without sacrificing individual
job health or risk complete machine lockups, this patch implements a way
to quantify resource pressure in the system.

A kernel built with CONFIG_PSI=y creates files in /proc/pressure/ that
expose the percentage of time the system is stalled on CPU, memory, or IO,
respectively.  Stall states are aggregate versions of the per-task delay
accounting delays:

       cpu: some tasks are runnable but not executing on a CPU
       memory: tasks are reclaiming, or waiting for swapin or thrashing cache
       io: tasks are waiting for io completions

These percentages of walltime can be thought of as pressure percentages,
and they give a general sense of system health and productivity loss
incurred by resource overcommit.  They can also indicate when the system
is approaching lockup scenarios and OOMs.

To do this, psi keeps track of the task states associated with each CPU
and samples the time they spend in stall states.  Every 2 seconds, the
samples are averaged across CPUs - weighted by the CPUs' non-idle time to
eliminate artifacts from unused CPUs - and translated into percentages of
walltime.  A running average of those percentages is maintained over 10s,
1m, and 5m periods (similar to the loadaverage).

[hannes@cmpxchg.org: doc fixlet, per Randy]
Link: http://lkml.kernel.org/r/20180828205625.GA14030@cmpxchg.org
[hannes@cmpxchg.org: code optimization]
Link: http://lkml.kernel.org/r/20180907175015.GA8479@cmpxchg.org
[hannes@cmpxchg.org: rename psi_clock() to psi_update_work(), per Peter]
Link: http://lkml.kernel.org/r/20180907145404.GB11088@cmpxchg.org
[hannes@cmpxchg.org: fix build]
Link: http://lkml.kernel.org/r/20180913014222.GA2370@cmpxchg.org
Link: http://lkml.kernel.org/r/20180828172258.3185-9-hannes@cmpxchg.org
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Tested-by: Daniel Drake <drake@endlessm.com>
Tested-by: Suren Baghdasaryan <surenb@google.com>
Cc: Christopher Lameter <cl@linux.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Johannes Weiner <jweiner@fb.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Enderborg <peter.enderborg@sony.com>
Cc: Randy Dunlap <rdunlap@infradead.org>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Vinayak Menon <vinmenon@codeaurora.org>
Cc: Randy Dunlap <rdunlap@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agosched: introduce this_rq_lock_irq()
Johannes Weiner [Fri, 26 Oct 2018 22:06:23 +0000 (15:06 -0700)]
sched: introduce this_rq_lock_irq()

do_sched_yield() disables IRQs, looks up this_rq() and locks it.  The next
patch is adding another site with the same pattern, so provide a
convenience function for it.

Link: http://lkml.kernel.org/r/20180828172258.3185-8-hannes@cmpxchg.org
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Tested-by: Suren Baghdasaryan <surenb@google.com>
Tested-by: Daniel Drake <drake@endlessm.com>
Cc: Christopher Lameter <cl@linux.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Johannes Weiner <jweiner@fb.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Enderborg <peter.enderborg@sony.com>
Cc: Randy Dunlap <rdunlap@infradead.org>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Vinayak Menon <vinmenon@codeaurora.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agosched: sched.h: make rq locking and clock functions available in stats.h
Johannes Weiner [Fri, 26 Oct 2018 22:06:19 +0000 (15:06 -0700)]
sched: sched.h: make rq locking and clock functions available in stats.h

kernel/sched/sched.h includes "stats.h" half-way through the file.  The
next patch introduces users of sched.h's rq locking functions and
update_rq_clock() in kernel/sched/stats.h.  Move those definitions up in
the file so they are available in stats.h.

Link: http://lkml.kernel.org/r/20180828172258.3185-7-hannes@cmpxchg.org
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Tested-by: Suren Baghdasaryan <surenb@google.com>
Tested-by: Daniel Drake <drake@endlessm.com>
Cc: Christopher Lameter <cl@linux.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Johannes Weiner <jweiner@fb.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Enderborg <peter.enderborg@sony.com>
Cc: Randy Dunlap <rdunlap@infradead.org>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Vinayak Menon <vinmenon@codeaurora.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agosched: loadavg: make calc_load_n() public
Johannes Weiner [Fri, 26 Oct 2018 22:06:16 +0000 (15:06 -0700)]
sched: loadavg: make calc_load_n() public

It's going to be used in a later patch. Keep the churn separate.

Link: http://lkml.kernel.org/r/20180828172258.3185-6-hannes@cmpxchg.org
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Tested-by: Suren Baghdasaryan <surenb@google.com>
Tested-by: Daniel Drake <drake@endlessm.com>
Cc: Christopher Lameter <cl@linux.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Johannes Weiner <jweiner@fb.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Enderborg <peter.enderborg@sony.com>
Cc: Randy Dunlap <rdunlap@infradead.org>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Vinayak Menon <vinmenon@codeaurora.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agosched: loadavg: consolidate LOAD_INT, LOAD_FRAC, CALC_LOAD
Johannes Weiner [Fri, 26 Oct 2018 22:06:11 +0000 (15:06 -0700)]
sched: loadavg: consolidate LOAD_INT, LOAD_FRAC, CALC_LOAD

There are several definitions of those functions/macros in places that
mess with fixed-point load averages.  Provide an official version.

[akpm@linux-foundation.org: fix missed conversion in block/blk-iolatency.c]
Link: http://lkml.kernel.org/r/20180828172258.3185-5-hannes@cmpxchg.org
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Tested-by: Suren Baghdasaryan <surenb@google.com>
Tested-by: Daniel Drake <drake@endlessm.com>
Cc: Christopher Lameter <cl@linux.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Johannes Weiner <jweiner@fb.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Enderborg <peter.enderborg@sony.com>
Cc: Randy Dunlap <rdunlap@infradead.org>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Vinayak Menon <vinmenon@codeaurora.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agodelayacct: track delays from thrashing cache pages
Johannes Weiner [Fri, 26 Oct 2018 22:06:08 +0000 (15:06 -0700)]
delayacct: track delays from thrashing cache pages

Delay accounting already measures the time a task spends in direct reclaim
and waiting for swapin, but in low memory situations tasks spend can spend
a significant amount of their time waiting on thrashing page cache.  This
isn't tracked right now.

To know the full impact of memory contention on an individual task,
measure the delay when waiting for a recently evicted active cache page to
read back into memory.

Also update tools/accounting/getdelays.c:

     [hannes@computer accounting]$ sudo ./getdelays -d -p 1
     print delayacct stats ON
     PID     1

     CPU             count     real total  virtual total    delay total  delay average
                     50318      745000000      847346785      400533713          0.008ms
     IO              count    delay total  delay average
                       435      122601218              0ms
     SWAP            count    delay total  delay average
                         0              0              0ms
     RECLAIM         count    delay total  delay average
                         0              0              0ms
     THRASHING       count    delay total  delay average
                        19       12621439              0ms

Link: http://lkml.kernel.org/r/20180828172258.3185-4-hannes@cmpxchg.org
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Tested-by: Daniel Drake <drake@endlessm.com>
Tested-by: Suren Baghdasaryan <surenb@google.com>
Cc: Christopher Lameter <cl@linux.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Johannes Weiner <jweiner@fb.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Enderborg <peter.enderborg@sony.com>
Cc: Randy Dunlap <rdunlap@infradead.org>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Vinayak Menon <vinmenon@codeaurora.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agomm: workingset: tell cache transitions from workingset thrashing
Johannes Weiner [Fri, 26 Oct 2018 22:06:04 +0000 (15:06 -0700)]
mm: workingset: tell cache transitions from workingset thrashing

Refaults happen during transitions between workingsets as well as in-place
thrashing.  Knowing the difference between the two has a range of
applications, including measuring the impact of memory shortage on the
system performance, as well as the ability to smarter balance pressure
between the filesystem cache and the swap-backed workingset.

During workingset transitions, inactive cache refaults and pushes out
established active cache.  When that active cache isn't stale, however,
and also ends up refaulting, that's bonafide thrashing.

Introduce a new page flag that tells on eviction whether the page has been
active or not in its lifetime.  This bit is then stored in the shadow
entry, to classify refaults as transitioning or thrashing.

How many page->flags does this leave us with on 32-bit?

20 bits are always page flags

21 if you have an MMU

23 with the zone bits for DMA, Normal, HighMem, Movable

29 with the sparsemem section bits

30 if PAE is enabled

31 with this patch.

So on 32-bit PAE, that leaves 1 bit for distinguishing two NUMA nodes.  If
that's not enough, the system can switch to discontigmem and re-gain the 6
or 7 sparsemem section bits.

Link: http://lkml.kernel.org/r/20180828172258.3185-3-hannes@cmpxchg.org
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Tested-by: Daniel Drake <drake@endlessm.com>
Tested-by: Suren Baghdasaryan <surenb@google.com>
Cc: Christopher Lameter <cl@linux.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Johannes Weiner <jweiner@fb.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Enderborg <peter.enderborg@sony.com>
Cc: Randy Dunlap <rdunlap@infradead.org>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Vinayak Menon <vinmenon@codeaurora.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agomm: workingset: don't drop refault information prematurely
Johannes Weiner [Fri, 26 Oct 2018 22:05:59 +0000 (15:05 -0700)]
mm: workingset: don't drop refault information prematurely

Patch series "psi: pressure stall information for CPU, memory, and IO", v4.

Overview

PSI reports the overall wallclock time in which the tasks in a system (or
cgroup) wait for (contended) hardware resources.

This helps users understand the resource pressure their workloads are
under, which allows them to rootcause and fix throughput and latency
problems caused by overcommitting, underprovisioning, suboptimal job
placement in a grid; as well as anticipate major disruptions like OOM.

Real-world applications

We're using the data collected by PSI (and its previous incarnation,
memdelay) quite extensively at Facebook, and with several success stories.

One usecase is avoiding OOM hangs/livelocks.  The reason these happen is
because the OOM killer is triggered by reclaim not being able to free
pages, but with fast flash devices there is *always* some clean and
uptodate cache to reclaim; the OOM killer never kicks in, even as tasks
spend 90% of the time thrashing the cache pages of their own executables.
There is no situation where this ever makes sense in practice.  We wrote a
<100 line POC python script to monitor memory pressure and kill stuff way
before such pathological thrashing leads to full system losses that would
require forcible hard resets.

We've since extended and deployed this code into other places to guarantee
latency and throughput SLAs, since they're usually violated way before the
kernel OOM killer would ever kick in.

It is available here: https://github.com/facebookincubator/oomd

Eventually we probably want to trigger the in-kernel OOM killer based on
extreme sustained pressure as well, so that Linux can avoid memory
livelocks - which technically aren't deadlocks, but to the user
indistinguishable from them - out of the box.  We'd continue using OOMD as
the first line of defense to ensure workload health and implement complex
kill policies that are beyond the scope of the kernel.

We also use PSI memory pressure for loadshedding.  Our batch job
infrastructure used to use heuristics based on various VM stats to
anticipate OOM situations, with lackluster success.  We switched it to PSI
and managed to anticipate and avoid OOM kills and lockups fairly reliably.
The reduction of OOM outages in the worker pool raised the pool's
aggregate productivity, and we were able to switch that service to smaller
machines.

Lastly, we use cgroups to isolate a machine's main workload from
maintenance crap like package upgrades, logging, configuration, as well as
to prevent multiple workloads on a machine from stepping on each others'
toes.  We were not able to configure this properly without the pressure
metrics; we would see latency or bandwidth drops, but it would often be
hard to impossible to rootcause it post-mortem.

We now log and graph pressure for the containers in our fleet and can
trivially link latency spikes and throughput drops to shortages of
specific resources after the fact, and fix the job config/scheduling.

PSI has also received testing, feedback, and feature requests from Android
and EndlessOS for the purpose of low-latency OOM killing, to intervene in
pressure situations before the UI starts hanging.

How do you use this feature?

A kernel with CONFIG_PSI=y will create a /proc/pressure directory with 3
files: cpu, memory, and io.  If using cgroup2, cgroups will also have
cpu.pressure, memory.pressure and io.pressure files, which simply
aggregate task stalls at the cgroup level instead of system-wide.

The cpu file contains one line:

some avg10=2.04 avg60=0.75 avg300=0.40 total=157656722

The averages give the percentage of walltime in which one or more tasks
are delayed on the runqueue while another task has the CPU.  They're
recent averages over 10s, 1m, 5m windows, so you can tell short term
trends from long term ones, similarly to the load average.

The total= value gives the absolute stall time in microseconds.  This
allows detecting latency spikes that might be too short to sway the
running averages.  It also allows custom time averaging in case the
10s/1m/5m windows aren't adequate for the usecase (or are too coarse with
future hardware).

What to make of this "some" metric?  If CPU utilization is at 100% and CPU
pressure is 0, it means the system is perfectly utilized, with one
runnable thread per CPU and nobody waiting.  At two or more runnable tasks
per CPU, the system is 100% overcommitted and the pressure average will
indicate as much.  From a utilization perspective this is a great state of
course: no CPU cycles are being wasted, even when 50% of the threads were
to go idle (as most workloads do vary).  From the perspective of the
individual job it's not great, however, and they would do better with more
resources.  Depending on what your priority and options are, raised "some"
numbers may or may not require action.

The memory file contains two lines:

some avg10=70.24 avg60=68.52 avg300=69.91 total=3559632828
full avg10=57.59 avg60=58.06 avg300=60.38 total=3300487258

The some line is the same as for cpu, the time in which at least one task
is stalled on the resource.  In the case of memory, this includes waiting
on swap-in, page cache refaults and page reclaim.

The full line, however, indicates time in which *nobody* is using the CPU
productively due to pressure: all non-idle tasks are waiting for memory in
one form or another.  Significant time spent in there is a good trigger
for killing things, moving jobs to other machines, or dropping incoming
requests, since neither the jobs nor the machine overall are making too
much headway.

The io file is similar to memory.  Because the block layer doesn't have a
concept of hardware contention right now (how much longer is my IO request
taking due to other tasks?), it reports CPU potential lost on all IO
delays, not just the potential lost due to competition.

FAQ

Q: How is PSI's CPU component different from the load average?

A: There are several quirks in the load average that make it hard to
   impossible to tell how overcommitted the CPU really is.

   1. The load average is reported as a raw number of active tasks.
      You need to know how many CPUs there are in the system, how many
      CPUs the workload is allowed to use, then think about what the
      proportion between load and the number of CPUs mean for the
      tasks trying to run.

      PSI reports the percentage of wallclock time in which tasks are
      waiting for a CPU to run on. It doesn't matter how many CPUs are
      present or usable. The number always tells the quality of life
      of tasks in the system or in a particular cgroup.

   2. The shortest averaging window is 1m, which is extremely coarse,
      and it's sampled in 5s intervals. A *lot* can happen on a CPU in
      5 seconds. This *may* be able to identify persistent long-term
      trends and very clear and obvious overloads, but it's unusable
      for latency spikes and more subtle overutilization.

      PSI's shortest window is 10s. It also exports the cumulative
      stall times (in microseconds) of synchronously recorded events.

   3. On Linux, the load average for historical reasons includes all
      TASK_UNINTERRUPTIBLE tasks. This gives a broader sense of how
      busy the system is, but on the flipside it doesn't distinguish
      whether tasks are likely to contend over the CPU or IO - which
      obviously requires very different interventions from a sys admin
      or a job scheduler.

      PSI reports independent metrics for CPU and IO. You can tell
      which resource is making the tasks wait, but in conjunction
      still see how overloaded the system is overall.

Q: What's the cost / performance impact of this feature?

A: PSI's primary cost is in the scheduler, in particular task wakeups
   and sleeps.

   I benchmarked this code using Facebook's two most scheduling
   sensitive workloads: memcache and webserver. They handle a ton of
   small requests - lots of wakeups and sleeps with little actual work
   in between - so they tend to be canaries for scheduler regressions.

   In the tests, the boxes were handling live traffic over the course
   of several hours. Half the machines, the control, ran with
   CONFIG_PSI=n.

   For memcache I used eight machines total. They're 2-socket, 14
   core, 56 thread boxes. The test runs for half the test period,
   flips the test and control kernels on the hardware to rule out HW
   factors, DC location etc., then runs the other half of the test.

   For the webservers, I used 32 machines total. They're single
   socket, 16 core, 32 thread machines.

   During the memcache test, CPU load was nopsi=78.05% psi=78.98% in
   the first half and nopsi=77.52% psi=78.25%, so PSI added between
   0.7 and 0.9 percentage points to the CPU load, a difference of
   about 1%.

   UPDATE: I re-ran this test with the v3 version of this patch set
   and the CPU utilization was equivalent between test and control.

   UPDATE: v4 is on par with v3.

   As far as end-to-end request latency from the client perspective
   goes, we don't sample those finely enough to capture the requests
   going to those particular machines during the test, but we know the
   p50 turnaround time in this workload is 54us, and perf bench sched
   pipe on those machines show nopsi=5.232666 us/op and psi=5.587347
   us/op, so this doesn't add much here either.

   The profile for the pipe benchmark shows:

        0.87%  sched-pipe  [kernel.vmlinux]    [k] psi_group_change
        0.83%  perf.real   [kernel.vmlinux]    [k] psi_group_change
        0.82%  perf.real   [kernel.vmlinux]    [k] psi_task_change
        0.58%  sched-pipe  [kernel.vmlinux]    [k] psi_task_change

   The webserver load is running inside 4 nested cgroup levels. The
   CPU load with both nopsi and psi kernels was indistinguishable at
   81%.

   For comparison, we had to disable the cgroup cpu controller on the
   webservers because it added 4 percentage points to the CPU% during
   this same exact test.

   Versions of this accounting code now run on 80% of our fleet. None
   of our workloads have reported regressions during the rollout.

Daniel Drake said:

: I just retested the latest version at
: http://git.cmpxchg.org/cgit.cgi/linux-psi.git (Linux 4.18) and the results
: are great.
:
: Test setup:
: Endless OS
: GeminiLake N4200 low end laptop
: 2GB RAM
: swap (and zram swap) disabled
:
: Baseline test: open a handful of large-ish apps and several website
: tabs in Google Chrome.
:
: Results: after a couple of minutes, system is excessively thrashing, mouse
: cursor can barely be moved, UI is not responding to mouse clicks, so it's
: impractical to recover from this situation as an ordinary user
:
: Add my simple killer:
: https://gist.github.com/dsd/a8988bf0b81a6163475988120fe8d9cd
:
: Results: when the thrashing causes the UI to become sluggish, the killer
: steps in and kills something (usually a chrome tab), and the system
: remains usable.  I repeatedly opened more apps and more websites over a 15
: minute period but I wasn't able to get the system to a point of UI
: unresponsiveness.

Suren said:

: Backported to 4.9 and retested on ARMv8 8 code system running Android.
: Signals behave as expected reacting to memory pressure, no jumps in
: "total" counters that would indicate an overflow/underflow issues.  Nicely
: done!

This patch (of 9):

If we keep just enough refault information to match the *current* page
cache during reclaim time, we could lose a lot of events when there is
only a temporary spike in non-cache memory consumption that pushes out all
the cache.  Once cache comes back, we won't see those refaults.  They
might not be actionable for LRU aging, but we want to know about them for
measuring memory pressure.

[hannes@cmpxchg.org: switch to NUMA-aware lru and slab counters]
Link: http://lkml.kernel.org/r/20181009184732.762-2-hannes@cmpxchg.org
Link: http://lkml.kernel.org/r/20180828172258.3185-2-hannes@cmpxchg.org
Signed-off-by: Johannes Weiner <jweiner@fb.com>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Rik van Riel <riel@surriel.com>
Tested-by: Daniel Drake <drake@endlessm.com>
Tested-by: Suren Baghdasaryan <surenb@google.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Vinayak Menon <vinmenon@codeaurora.org>
Cc: Christopher Lameter <cl@linux.com>
Cc: Peter Enderborg <peter.enderborg@sony.com>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Randy Dunlap <rdunlap@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agomm, slab: shorten kmalloc cache names for large sizes
Vlastimil Babka [Fri, 26 Oct 2018 22:05:55 +0000 (15:05 -0700)]
mm, slab: shorten kmalloc cache names for large sizes

Kmalloc cache names can get quite long for large object sizes, when the
sizes are expressed in bytes.  Use 'k' and 'M' prefixes to make the names
as short as possible e.g.  in /proc/slabinfo.  This works, as we mostly
use power-of-two sizes, with exceptions only below 1k.

Example: 'kmalloc-4194304' becomes 'kmalloc-4M'

Link: http://lkml.kernel.org/r/20180731090649.16028-7-vbabka@suse.cz
Suggested-by: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Christoph Lameter <cl@linux.com>
Acked-by: Roman Gushchin <guro@fb.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Laura Abbott <labbott@redhat.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Sumit Semwal <sumit.semwal@linaro.org>
Cc: Vijayanand Jitta <vjitta@codeaurora.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agomm, proc: add KReclaimable to /proc/meminfo
Vlastimil Babka [Fri, 26 Oct 2018 22:05:50 +0000 (15:05 -0700)]
mm, proc: add KReclaimable to /proc/meminfo

The vmstat NR_KERNEL_MISC_RECLAIMABLE counter is for kernel non-slab
allocations that can be reclaimed via shrinker.  In /proc/meminfo, we can
show the sum of all reclaimable kernel allocations (including slab) as
"KReclaimable".  Add the same counter also to per-node meminfo under /sys

With this counter, users will have more complete information about kernel
memory usage.  Non-slab reclaimable pages (currently just the ION
allocator) will not be missing from /proc/meminfo, making users wonder
where part of their memory went.  More precisely, they already appear in
MemAvailable, but without the new counter, it's not obvious why the value
in MemAvailable doesn't fully correspond with the sum of other counters
participating in it.

Link: http://lkml.kernel.org/r/20180731090649.16028-6-vbabka@suse.cz
Signed-off-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Roman Gushchin <guro@fb.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Laura Abbott <labbott@redhat.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Sumit Semwal <sumit.semwal@linaro.org>
Cc: Vijayanand Jitta <vjitta@codeaurora.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agomm: rename and change semantics of nr_indirectly_reclaimable_bytes
Vlastimil Babka [Fri, 26 Oct 2018 22:05:46 +0000 (15:05 -0700)]
mm: rename and change semantics of nr_indirectly_reclaimable_bytes

The vmstat counter NR_INDIRECTLY_RECLAIMABLE_BYTES was introduced by
commit eb59254608bc ("mm: introduce NR_INDIRECTLY_RECLAIMABLE_BYTES") with
the goal of accounting objects that can be reclaimed, but cannot be
allocated via a SLAB_RECLAIM_ACCOUNT cache.  This is now possible via
kmalloc() with __GFP_RECLAIMABLE flag, and the dcache external names user
is converted.

The counter is however still useful for accounting direct page allocations
(i.e.  not slab) with a shrinker, such as the ION page pool.  So keep it,
and:

- change granularity to pages to be more like other counters; sub-page
  allocations should be able to use kmalloc
- rename the counter to NR_KERNEL_MISC_RECLAIMABLE
- expose the counter again in vmstat as "nr_kernel_misc_reclaimable"; we can
  again remove the check for not printing "hidden" counters

Link: http://lkml.kernel.org/r/20180731090649.16028-5-vbabka@suse.cz
Signed-off-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Christoph Lameter <cl@linux.com>
Acked-by: Roman Gushchin <guro@fb.com>
Cc: Vijayanand Jitta <vjitta@codeaurora.org>
Cc: Laura Abbott <labbott@redhat.com>
Cc: Sumit Semwal <sumit.semwal@linaro.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agodcache: allocate external names from reclaimable kmalloc caches
Vlastimil Babka [Fri, 26 Oct 2018 22:05:41 +0000 (15:05 -0700)]
dcache: allocate external names from reclaimable kmalloc caches

We can use the newly introduced kmalloc-reclaimable-X caches, to allocate
external names in dcache, which will take care of the proper accounting
automatically, and also improve anti-fragmentation page grouping.

This effectively reverts commit f1782c9bc547 ("dcache: account external
names as indirectly reclaimable memory") and instead passes
__GFP_RECLAIMABLE to kmalloc().  The accounting thus moves from
NR_INDIRECTLY_RECLAIMABLE_BYTES to NR_SLAB_RECLAIMABLE, which is also
considered in MemAvailable calculation and overcommit decisions.

Link: http://lkml.kernel.org/r/20180731090649.16028-4-vbabka@suse.cz
Signed-off-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Roman Gushchin <guro@fb.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Laura Abbott <labbott@redhat.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Sumit Semwal <sumit.semwal@linaro.org>
Cc: Vijayanand Jitta <vjitta@codeaurora.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agomm, slab/slub: introduce kmalloc-reclaimable caches
Vlastimil Babka [Fri, 26 Oct 2018 22:05:38 +0000 (15:05 -0700)]
mm, slab/slub: introduce kmalloc-reclaimable caches

Kmem caches can be created with a SLAB_RECLAIM_ACCOUNT flag, which
indicates they contain objects which can be reclaimed under memory
pressure (typically through a shrinker).  This makes the slab pages
accounted as NR_SLAB_RECLAIMABLE in vmstat, which is reflected also the
MemAvailable meminfo counter and in overcommit decisions.  The slab pages
are also allocated with __GFP_RECLAIMABLE, which is good for
anti-fragmentation through grouping pages by mobility.

The generic kmalloc-X caches are created without this flag, but sometimes
are used also for objects that can be reclaimed, which due to varying size
cannot have a dedicated kmem cache with SLAB_RECLAIM_ACCOUNT flag.  A
prominent example are dcache external names, which prompted the creation
of a new, manually managed vmstat counter NR_INDIRECTLY_RECLAIMABLE_BYTES
in commit f1782c9bc547 ("dcache: account external names as indirectly
reclaimable memory").

To better handle this and any other similar cases, this patch introduces
SLAB_RECLAIM_ACCOUNT variants of kmalloc caches, named kmalloc-rcl-X.
They are used whenever the kmalloc() call passes __GFP_RECLAIMABLE among
gfp flags.  They are added to the kmalloc_caches array as a new type.
Allocations with both __GFP_DMA and __GFP_RECLAIMABLE will use a dma type
cache.

This change only applies to SLAB and SLUB, not SLOB.  This is fine, since
SLOB's target are tiny system and this patch does add some overhead of
kmem management objects.

Link: http://lkml.kernel.org/r/20180731090649.16028-3-vbabka@suse.cz
Signed-off-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Christoph Lameter <cl@linux.com>
Acked-by: Roman Gushchin <guro@fb.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Laura Abbott <labbott@redhat.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Sumit Semwal <sumit.semwal@linaro.org>
Cc: Vijayanand Jitta <vjitta@codeaurora.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agomm, slab: combine kmalloc_caches and kmalloc_dma_caches
Vlastimil Babka [Fri, 26 Oct 2018 22:05:34 +0000 (15:05 -0700)]
mm, slab: combine kmalloc_caches and kmalloc_dma_caches

Patch series "kmalloc-reclaimable caches", v4.

As discussed at LSF/MM [1] here's a patchset that introduces
kmalloc-reclaimable caches (more details in the second patch) and uses
them for dcache external names.  That allows us to repurpose the
NR_INDIRECTLY_RECLAIMABLE_BYTES counter later in the series.

With patch 3/6, dcache external names are allocated from kmalloc-rcl-*
caches, eliminating the need for manual accounting.  More importantly, it
also ensures the reclaimable kmalloc allocations are grouped in pages
separate from the regular kmalloc allocations.  The need for proper
accounting of dcache external names has shown it's easy for misbehaving
process to allocate lots of them, causing premature OOMs.  Without the
added grouping, it's likely that a similar workload can interleave the
dcache external names allocations with regular kmalloc allocations (note:
I haven't searched myself for an example of such regular kmalloc
allocation, but I would be very surprised if there wasn't some).  A
pathological case would be e.g.  one 64byte regular allocations with 63
external dcache names in a page (64x64=4096), which means the page is not
freed even after reclaiming after all dcache names, and the process can
thus "steal" the whole page with single 64byte allocation.

If other kmalloc users similar to dcache external names become identified,
they can also benefit from the new functionality simply by adding
__GFP_RECLAIMABLE to the kmalloc calls.

Side benefits of the patchset (that could be also merged separately)
include removed branch for detecting __GFP_DMA kmalloc(), and shortening
kmalloc cache names in /proc/slabinfo output.  The latter is potentially
an ABI break in case there are tools parsing the names and expecting the
values to be in bytes.

This is how /proc/slabinfo looks like after booting in virtme:

...
kmalloc-rcl-4M         0      0 4194304    1 1024 : tunables    1    1    0 : slabdata      0      0      0
...
kmalloc-rcl-96         7     32    128   32    1 : tunables  120   60    8 : slabdata      1      1      0
kmalloc-rcl-64        25    128     64   64    1 : tunables  120   60    8 : slabdata      2      2      0
kmalloc-rcl-32         0      0     32  124    1 : tunables  120   60    8 : slabdata      0      0      0
kmalloc-4M             0      0 4194304    1 1024 : tunables    1    1    0 : slabdata      0      0      0
kmalloc-2M             0      0 2097152    1  512 : tunables    1    1    0 : slabdata      0      0      0
kmalloc-1M             0      0 1048576    1  256 : tunables    1    1    0 : slabdata      0      0      0
...

/proc/vmstat with renamed nr_indirectly_reclaimable_bytes counter:

...
nr_slab_reclaimable 2817
nr_slab_unreclaimable 1781
...
nr_kernel_misc_reclaimable 0
...

/proc/meminfo with new KReclaimable counter:

...
Shmem:               564 kB
KReclaimable:      11260 kB
Slab:              18368 kB
SReclaimable:      11260 kB
SUnreclaim:         7108 kB
KernelStack:        1248 kB
...

This patch (of 6):

The kmalloc caches currently mainain separate (optional) array
kmalloc_dma_caches for __GFP_DMA allocations.  There are tests for
__GFP_DMA in the allocation hotpaths.  We can avoid the branches by
combining kmalloc_caches and kmalloc_dma_caches into a single
two-dimensional array where the outer dimension is cache "type".  This
will also allow to add kmalloc-reclaimable caches as a third type.

Link: http://lkml.kernel.org/r/20180731090649.16028-2-vbabka@suse.cz
Signed-off-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Christoph Lameter <cl@linux.com>
Acked-by: Roman Gushchin <guro@fb.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Laura Abbott <labbott@redhat.com>
Cc: Sumit Semwal <sumit.semwal@linaro.org>
Cc: Vijayanand Jitta <vjitta@codeaurora.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agouserfaultfd: allow get_mempolicy(MPOL_F_NODE|MPOL_F_ADDR) to trigger userfaults
Andrea Arcangeli [Fri, 26 Oct 2018 22:05:16 +0000 (15:05 -0700)]
userfaultfd: allow get_mempolicy(MPOL_F_NODE|MPOL_F_ADDR) to trigger userfaults

get_mempolicy(MPOL_F_NODE|MPOL_F_ADDR) called a get_user_pages that would
not be waiting for userfaults before failing and it would hit on a SIGBUS
instead.  Using get_user_pages_locked/unlocked instead will allow
get_mempolicy to allow userfaults to resolve the fault and fill the hole,
before grabbing the node id of the page.

If the user calls get_mempolicy() with MPOL_F_ADDR | MPOL_F_NODE for an
address inside an area managed by uffd and there is no page at that
address, the page allocation from within get_mempolicy() will fail
because get_user_pages() does not allow for page fault retry required
for uffd; the user will get SIGBUS.

With this patch, the page fault will be resolved by the uffd and the
get_mempolicy() will continue normally.

Background:

Via code review, previously the syscall would have returned -EFAULT
(vm_fault_to_errno), now it will block and wait for an userfault (if
it's waken before the fault is resolved it'll still -EFAULT).

This way get_mempolicy will give a chance to an "unaware" app to be
compliant with userfaults.

The reason this visible change is that becoming "userfault compliant"
cannot regress anything: all other syscalls including read(2)/write(2)
had to become "userfault compliant" long time ago (that's one of the
things userfaultfd can do that PROT_NONE and trapping segfaults can't).

So this is just one more syscall that become "userfault compliant" like
all other major ones already were.

This has been happening on virtio-bridge dpdk process which just called
get_mempolicy on the guest space post live migration, but before the
memory had a chance to be migrated to destination.

I didn't run an strace to be able to show the -EFAULT going away, but
I've the confirmation of the below debug aid information (only visible
with CONFIG_DEBUG_VM=y) going away with the patch:

    [20116.371461] FAULT_FLAG_ALLOW_RETRY missing 0
    [20116.371464] CPU: 1 PID: 13381 Comm: vhost-events Not tainted 4.17.12-200.fc28.x86_64 #1
    [20116.371465] Hardware name: LENOVO 20FAS2BN0A/20FAS2BN0A, BIOS N1CET54W (1.22 ) 02/10/2017
    [20116.371466] Call Trace:
    [20116.371473]  dump_stack+0x5c/0x80
    [20116.371476]  handle_userfault.cold.37+0x1b/0x22
    [20116.371479]  ? remove_wait_queue+0x20/0x60
    [20116.371481]  ? poll_freewait+0x45/0xa0
    [20116.371483]  ? do_sys_poll+0x31c/0x520
    [20116.371485]  ? radix_tree_lookup_slot+0x1e/0x50
    [20116.371488]  shmem_getpage_gfp+0xce7/0xe50
    [20116.371491]  ? page_add_file_rmap+0x1a/0x2c0
    [20116.371493]  shmem_fault+0x78/0x1e0
    [20116.371495]  ? filemap_map_pages+0x3a1/0x450
    [20116.371498]  __do_fault+0x1f/0xc0
    [20116.371500]  __handle_mm_fault+0xe2e/0x12f0
    [20116.371502]  handle_mm_fault+0xda/0x200
    [20116.371504]  __get_user_pages+0x238/0x790
    [20116.371506]  get_user_pages+0x3e/0x50
    [20116.371510]  kernel_get_mempolicy+0x40b/0x700
    [20116.371512]  ? vfs_write+0x170/0x1a0
    [20116.371515]  __x64_sys_get_mempolicy+0x21/0x30
    [20116.371517]  do_syscall_64+0x5b/0x160
    [20116.371520]  entry_SYSCALL_64_after_hwframe+0x44/0xa9

The above harmless debug message (not a kernel crash, just a
dump_stack()) is shown with CONFIG_DEBUG_VM=y to more quickly identify
and improve kernel spots that may have to become "userfaultfd
compliant" like this one (without having to run an strace and search
for syscall misbehavior).  Spots like the above are more closer to a
kernel bug for the non-cooperative usages that Mike focuses on, than
for for dpdk qemu-cooperative usages that reproduced it, but it's still
nicer to get this fixed for dpdk too.

The part of the patch that caused me to think is only the
implementation issue of mpol_get, but it looks like it should work safe
no matter the kind of mempolicy structure that is (the default static
policy also starts at 1 so it'll go to 2 and back to 1 without crashing
everything at 0).

[rppt@linux.vnet.ibm.com: changelog addition]
  http://lkml.kernel.org/r/20180904073718.GA26916@rapoport-lnx
Link: http://lkml.kernel.org/r/20180831214848.23676-1-aarcange@redhat.com
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Reported-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Tested-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Mike Rapoport <rppt@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agoalpha: switch to NO_BOOTMEM
Mike Rapoport [Fri, 26 Oct 2018 22:05:10 +0000 (15:05 -0700)]
alpha: switch to NO_BOOTMEM

Replace bootmem allocator with memblock and enable use of NO_BOOTMEM like
on most other architectures.

Alpha gets the description of the physical memory from the firmware as an
array of memory clusters.  Each cluster that is not reserved by the
firmware is added to memblock.memory.

Once the memblock.memory is set up, we reserve the kernel and initrd pages
with memblock reserve.

Since we don't need the bootmem bitmap anymore, the code that finds an
appropriate place is removed.

The conversion does not take care of NUMA support which is marked broken
for more than 10 years now.

Link: http://lkml.kernel.org/r/1535952894-10967-1-git-send-email-rppt@linux.vnet.ibm.com
Signed-off-by: Mike Rapoport <rppt@linux.vnet.ibm.com>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agounicore32: switch to NO_BOOTMEM
Mike Rapoport [Fri, 26 Oct 2018 22:05:05 +0000 (15:05 -0700)]
unicore32: switch to NO_BOOTMEM

The unicore32 architecture already supports memblock and uses it for some
early memory reservations, e.g initrd and the page tables.

At some point unicore32 allocates the bootmem bitmap from the memblock and
then hands over the memory reservations from memblock to bootmem.

This patch removes the bootmem initialization and leaves memblock as the
only boot time memory manager for unicore32.

Link: http://lkml.kernel.org/r/1533326330-31677-8-git-send-email-rppt@linux.vnet.ibm.com
Signed-off-by: Mike Rapoport <rppt@linux.vnet.ibm.com>
Acked-by: Guan Xuetao <gxt@pku.edu.cn>
Cc: Ley Foon Tan <ley.foon.tan@intel.com>
Cc: Richard Kuo <rkuo@codeaurora.org>
Cc: Richard Weinberger <richard@nod.at>
Cc: Rob Herring <robh@kernel.org>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agoum: switch to NO_BOOTMEM
Mike Rapoport [Fri, 26 Oct 2018 22:05:02 +0000 (15:05 -0700)]
um: switch to NO_BOOTMEM

Replace bootmem initialization with memblock_add and memblock_reserve calls
and explicit initialization of {min,max}_low_pfn.

Link: http://lkml.kernel.org/r/1533326330-31677-7-git-send-email-rppt@linux.vnet.ibm.com
Signed-off-by: Mike Rapoport <rppt@linux.vnet.ibm.com>
Acked-by: Richard Weinberger <richard@nod.at>
Cc: Guan Xuetao <gxt@pku.edu.cn>
Cc: Ley Foon Tan <ley.foon.tan@intel.com>
Cc: Richard Kuo <rkuo@codeaurora.org>
Cc: Rob Herring <robh@kernel.org>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agoum: setup_physmem: stop using global variables
Mike Rapoport [Fri, 26 Oct 2018 22:04:58 +0000 (15:04 -0700)]
um: setup_physmem: stop using global variables

The setup_physmem() function receives uml_physmem and uml_reserved as
parameters and still used these global variables.  Replace such usage with
local variables.

Link: http://lkml.kernel.org/r/1533326330-31677-6-git-send-email-rppt@linux.vnet.ibm.com
Signed-off-by: Mike Rapoport <rppt@linux.vnet.ibm.com>
Acked-by: Richard Weinberger <richard@nod.at>
Cc: Guan Xuetao <gxt@pku.edu.cn>
Cc: Ley Foon Tan <ley.foon.tan@intel.com>
Cc: Richard Kuo <rkuo@codeaurora.org>
Cc: Rob Herring <robh@kernel.org>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agonios2: switch to NO_BOOTMEM
Mike Rapoport [Fri, 26 Oct 2018 22:04:55 +0000 (15:04 -0700)]
nios2: switch to NO_BOOTMEM

Remove bootmem bitmap initialization and replace reserve_bootmem() with
memblock_reserve().

Link: http://lkml.kernel.org/r/1533326330-31677-5-git-send-email-rppt@linux.vnet.ibm.com
Signed-off-by: Mike Rapoport <rppt@linux.vnet.ibm.com>
Acked-by: Ley Foon Tan <ley.foon.tan@intel.com>
Cc: Guan Xuetao <gxt@pku.edu.cn>
Cc: Richard Kuo <rkuo@codeaurora.org>
Cc: Richard Weinberger <richard@nod.at>
Cc: Rob Herring <robh@kernel.org>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agonios2: use generic early_init_dt_add_memory_arch
Mike Rapoport [Fri, 26 Oct 2018 22:04:51 +0000 (15:04 -0700)]
nios2: use generic early_init_dt_add_memory_arch

All we have to do is to enable memblock, the generic FDT code will take
care of the rest.

Link: http://lkml.kernel.org/r/1533326330-31677-4-git-send-email-rppt@linux.vnet.ibm.com
Signed-off-by: Mike Rapoport <rppt@linux.vnet.ibm.com>
Acked-by: Ley Foon Tan <ley.foon.tan@intel.com>
Cc: Guan Xuetao <gxt@pku.edu.cn>
Cc: Richard Kuo <rkuo@codeaurora.org>
Cc: Richard Weinberger <richard@nod.at>
Cc: Rob Herring <robh@kernel.org>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agoof: ignore sub-page memory regions
Mike Rapoport [Fri, 26 Oct 2018 22:04:48 +0000 (15:04 -0700)]
of: ignore sub-page memory regions

Memory region size is rounded down to page boundary and with sub-page
region it becomes 0 and there is no point to add an empty region.
Moreover, when the base is less than PAGE_SIZE we get a bogus size as
(base + size - 1) evaluates to -1.

8cccffc52694 ("of: check for size < 0 after rounding in
early_init_dt_add_memory_arch") introduced a test for wrap around for the
case when base is not page aligned, the same test can be used to ignore
sub-page region sizes.

Link: http://lkml.kernel.org/r/1533326330-31677-3-git-send-email-rppt@linux.vnet.ibm.com
Signed-off-by: Mike Rapoport <rppt@linux.vnet.ibm.com>
Reviewed-by: Rob Herring <robh@kernel.org>
Cc: Guan Xuetao <gxt@pku.edu.cn>
Cc: Ley Foon Tan <ley.foon.tan@intel.com>
Cc: Richard Kuo <rkuo@codeaurora.org>
Cc: Richard Weinberger <richard@nod.at>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agohexagon: switch to NO_BOOTMEM
Mike Rapoport [Fri, 26 Oct 2018 22:04:44 +0000 (15:04 -0700)]
hexagon: switch to NO_BOOTMEM

Patch series "switch several architectures NO_BOOTMEM".

These patches perform conversion to NO_BOOTMEM of hexagon, nios2, uml and
unicore32.

This patch (of 7):

Add registration of the system memory with memblock, eliminate bootmem
initialization and convert early memory reservations from bootmem to
memblock.

Link: http://lkml.kernel.org/r/1533326330-31677-2-git-send-email-rppt@linux.vnet.ibm.com
Signed-off-by: Mike Rapoport <rppt@linux.vnet.ibm.com>
Acked-by: Richard Kuo <rkuo@codeaurora.org>
Cc: Guan Xuetao <gxt@pku.edu.cn>
Cc: Ley Foon Tan <ley.foon.tan@intel.com>
Cc: Richard Weinberger <richard@nod.at>
Cc: Rob Herring <robh@kernel.org>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agomm: convert insert_pfn() to vm_fault_t
Matthew Wilcox [Fri, 26 Oct 2018 22:04:40 +0000 (15:04 -0700)]
mm: convert insert_pfn() to vm_fault_t

All callers convert its errno into a vm_fault_t, so convert it to return a
vm_fault_t directly.

Link: http://lkml.kernel.org/r/20180828145728.11873-11-willy@infradead.org
Signed-off-by: Matthew Wilcox <willy@infradead.org>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Nicolas Pitre <nicolas.pitre@linaro.org>
Cc: Souptick Joarder <jrdr.linux@gmail.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>
6 years agomm: convert __vm_insert_mixed() to vm_fault_t
Matthew Wilcox [Fri, 26 Oct 2018 22:04:37 +0000 (15:04 -0700)]
mm: convert __vm_insert_mixed() to vm_fault_t

Both of its callers currently convert its errno return into a vm_fault_t,
so move the conversion into __vm_insert_mixed().

Link: http://lkml.kernel.org/r/20180828145728.11873-10-willy@infradead.org
Signed-off-by: Matthew Wilcox <willy@infradead.org>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Nicolas Pitre <nicolas.pitre@linaro.org>
Cc: Souptick Joarder <jrdr.linux@gmail.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>
6 years agomm: inline vm_insert_pfn_prot() into caller
Matthew Wilcox [Fri, 26 Oct 2018 22:04:33 +0000 (15:04 -0700)]
mm: inline vm_insert_pfn_prot() into caller

vm_insert_pfn_prot() is only called from vmf_insert_pfn_prot(), so inline
it and convert some of the errnos into vm_fault codes earlier.

Link: http://lkml.kernel.org/r/20180828145728.11873-9-willy@infradead.org
Signed-off-by: Matthew Wilcox <willy@infradead.org>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Nicolas Pitre <nicolas.pitre@linaro.org>
Cc: Souptick Joarder <jrdr.linux@gmail.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>
6 years agomm: remove vm_insert_pfn()
Matthew Wilcox [Fri, 26 Oct 2018 22:04:29 +0000 (15:04 -0700)]
mm: remove vm_insert_pfn()

All callers are now converted to vmf_insert_pfn() so convert
vmf_insert_pfn() from being a compatibility wrapper around vm_insert_pfn()
to being a compatibility wrapper around vmf_insert_pfn_prot().

Link: http://lkml.kernel.org/r/20180828145728.11873-8-willy@infradead.org
Signed-off-by: Matthew Wilcox <willy@infradead.org>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Nicolas Pitre <nicolas.pitre@linaro.org>
Cc: Souptick Joarder <jrdr.linux@gmail.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>
6 years agomm: remove references to vm_insert_pfn()
Matthew Wilcox [Fri, 26 Oct 2018 22:04:26 +0000 (15:04 -0700)]
mm: remove references to vm_insert_pfn()

Documentation and comments.

Link: http://lkml.kernel.org/r/20180828145728.11873-7-willy@infradead.org
Signed-off-by: Matthew Wilcox <willy@infradead.org>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Nicolas Pitre <nicolas.pitre@linaro.org>
Cc: Souptick Joarder <jrdr.linux@gmail.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>
6 years agomm: make vm_insert_pfn_prot() static
Matthew Wilcox [Fri, 26 Oct 2018 22:04:21 +0000 (15:04 -0700)]
mm: make vm_insert_pfn_prot() static

Now this is no longer used outside mm/memory.c, make it static.

Link: http://lkml.kernel.org/r/20180828145728.11873-6-willy@infradead.org
Signed-off-by: Matthew Wilcox <willy@infradead.org>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Nicolas Pitre <nicolas.pitre@linaro.org>
Cc: Souptick Joarder <jrdr.linux@gmail.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>
6 years agox86: convert vdso to use vm_fault_t
Matthew Wilcox [Fri, 26 Oct 2018 22:04:16 +0000 (15:04 -0700)]
x86: convert vdso to use vm_fault_t

Return vm_fault_t codes directly from the appropriate mm routines instead
of converting from errnos ourselves.  Fixes a minor bug where we'd return
SIGBUS instead of the correct OOM code if we ran out of memory allocating
page tables.

Link: http://lkml.kernel.org/r/20180828145728.11873-5-willy@infradead.org
Signed-off-by: Matthew Wilcox <willy@infradead.org>
Reviewed-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Nicolas Pitre <nicolas.pitre@linaro.org>
Cc: Souptick Joarder <jrdr.linux@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agomm: introduce vmf_insert_pfn_prot()
Matthew Wilcox [Fri, 26 Oct 2018 22:04:13 +0000 (15:04 -0700)]
mm: introduce vmf_insert_pfn_prot()

Like vm_insert_pfn_prot(), but returns a vm_fault_t instead of an errno.
Also unexport vm_insert_pfn_prot as it has no modular users.

Link: http://lkml.kernel.org/r/20180828145728.11873-4-willy@infradead.org
Signed-off-by: Matthew Wilcox <willy@infradead.org>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Nicolas Pitre <nicolas.pitre@linaro.org>
Cc: Souptick Joarder <jrdr.linux@gmail.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>
6 years agomm: remove vm_insert_mixed()
Matthew Wilcox [Fri, 26 Oct 2018 22:04:10 +0000 (15:04 -0700)]
mm: remove vm_insert_mixed()

All callers are now converted to vmf_insert_mixed() so convert
vmf_insert_mixed() from being a compatibility wrapper into the real
function.

Link: http://lkml.kernel.org/r/20180828145728.11873-3-willy@infradead.org
Signed-off-by: Matthew Wilcox <willy@infradead.org>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Nicolas Pitre <nicolas.pitre@linaro.org>
Cc: Souptick Joarder <jrdr.linux@gmail.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>
6 years agocramfs: convert to use vmf_insert_mixed
Nicolas Pitre [Fri, 26 Oct 2018 22:04:06 +0000 (15:04 -0700)]
cramfs: convert to use vmf_insert_mixed

cramfs is the only remaining user of vm_insert_mixed() and should be
converted to vmf_insert_mixed().

Based on a previous patch from Matthew Wilcox.

Link: http://lkml.kernel.org/r/nycvar.YSQ.7.76.1808290945450.10215@knanqh.ubzr
Signed-off-by: Nicolas Pitre <nico@linaro.org>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Souptick Joarder <jrdr.linux@gmail.com>a
Cc: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agomm: convert to use vm_fault_t
Souptick Joarder [Fri, 26 Oct 2018 22:04:03 +0000 (15:04 -0700)]
mm: convert to use vm_fault_t

As part of vm_fault_t conversion filemap_page_mkwrite() for the NOMMU case
was missed.  Now converted.

Link: http://lkml.kernel.org/r/20180828174952.GA29229@jordon-HP-15-Notebook-PC
Signed-off-by: Souptick Joarder <jrdr.linux@gmail.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agomm/page_alloc.c: clean up check_for_memory()
Oscar Salvador [Fri, 26 Oct 2018 22:03:58 +0000 (15:03 -0700)]
mm/page_alloc.c: clean up check_for_memory()

check_for_memory() looks a bit confusing.  First of all, we have this:

if (N_MEMORY == N_NORMAL_MEMORY)
return;

Checking the ENUM declaration, looks like N_MEMORY canot be equal to
N_NORMAL_MEMORY.

I could not find where N_MEMORY is set to N_NORMAL_MEMORY, or the other
way around either, so unless I am missing something, this condition will
never evaluate to true.  It makes sense to get rid of it.

Moving forward, the operations within the loop look a bit confusing as
well.

We set N_HIGH_MEMORY unconditionally, and then we set N_NORMAL_MEMORY in
case we have CONFIG_HIGHMEM (N_NORMAL_MEMORY != N_HIGH_MEMORY) and zone <=
ZONE_NORMAL.  (N_HIGH_MEMORY falls back to N_NORMAL_MEMORY on
!CONFIG_HIGHMEM systems, and that is why we can just go ahead and set
N_HIGH_MEMORY unconditionally)

Although this works, it is a bit subtle.

I think that this could be easier to follow:

First, we should only set N_HIGH_MEMORY in case we have CONFIG_HIGHMEM.
And then we should set N_NORMAL_MEMORY in case zone <= ZONE_NORMAL,
without further checking whether we have CONFIG_HIGHMEM or not.

Link: http://lkml.kernel.org/r/20180828210158.4617-1-osalvador@techadventures.net
Signed-off-by: Oscar Salvador <osalvador@suse.de>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Michael Hocko <mhocko@suse.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Pavel Tatashin <pavel.tatashin@microsoft.com
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
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>
6 years agomm/swapfile.c: clear si->swap_map[] in swap_free_cluster()
Huang Ying [Fri, 26 Oct 2018 22:03:53 +0000 (15:03 -0700)]
mm/swapfile.c: clear si->swap_map[] in swap_free_cluster()

si->swap_map[] of the swap entries in cluster needs to be cleared during
freeing.  Previously, this is done in the caller of swap_free_cluster().
This may cause code duplication (one user now, will add more users later)
and lock/unlock cluster unnecessarily.  In this patch, the clearing code
is moved to swap_free_cluster() to avoid the downside.

Link: http://lkml.kernel.org/r/20180827075535.17406-4-ying.huang@intel.com
Signed-off-by: "Huang, Ying" <ying.huang@intel.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Shaohua Li <shli@kernel.org>
Cc: Hugh Dickins <hughd@google.com>
Cc: Minchan Kim <minchan@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agomm/swapfile.c: call free_swap_slot() in __swap_entry_free()
Huang Ying [Fri, 26 Oct 2018 22:03:49 +0000 (15:03 -0700)]
mm/swapfile.c: call free_swap_slot() in __swap_entry_free()

This is a code cleanup patch without functionality change.

Originally, when __swap_entry_free() is called, and its return value is 0,
free_swap_slot() will always be called to free the swap entry to the
per-CPU pool.  So move the call to free_swap_slot() to __swap_entry_free()
to simplify the code.

Link: http://lkml.kernel.org/r/20180827075535.17406-3-ying.huang@intel.com
Signed-off-by: "Huang, Ying" <ying.huang@intel.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Shaohua Li <shli@kernel.org>
Cc: Hugh Dickins <hughd@google.com>
Cc: Minchan Kim <minchan@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agomm/swapfile.c: use __try_to_reclaim_swap() in free_swap_and_cache()
Huang Ying [Fri, 26 Oct 2018 22:03:46 +0000 (15:03 -0700)]
mm/swapfile.c: use __try_to_reclaim_swap() in free_swap_and_cache()

The code path to reclaim the swap entry in free_swap_and_cache() is
almost same as that of __try_to_reclaim_swap().  The largest
difference is just coding style.  So the support to the additional
requirement of free_swap_and_cache() is added into
__try_to_reclaim_swap().  free_swap_and_cache() is changed to call
__try_to_reclaim_swap(), and delete the duplicated code.  This will
improve code readability and reduce the potential bugs.

There are 2 functionality differences between __try_to_reclaim_swap()
and swap entry reclaim code of free_swap_and_cache().

- free_swap_and_cache() only reclaims the swap entry if the page is
  unmapped or swap is getting full.  The support has been added into
  __try_to_reclaim_swap().

- try_to_free_swap() (called by __try_to_reclaim_swap()) checks
  pm_suspended_storage(), while free_swap_and_cache() not.  I think
  this is OK.  Because the page and the swap entry can be reclaimed
  later eventually.

Link: http://lkml.kernel.org/r/20180827075535.17406-2-ying.huang@intel.com
Signed-off-by: "Huang, Ying" <ying.huang@intel.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Shaohua Li <shli@kernel.org>
Cc: Hugh Dickins <hughd@google.com>
Cc: Minchan Kim <minchan@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agokmemleak: add module param to print warnings to dmesg
Vincent Whitchurch [Fri, 26 Oct 2018 22:03:42 +0000 (15:03 -0700)]
kmemleak: add module param to print warnings to dmesg

Currently, kmemleak only prints the number of suspected leaks to dmesg but
requires the user to read a debugfs file to get the actual stack traces of
the objects' allocation points.  Add a module option to print the full
object information to dmesg too.  It can be enabled with
kmemleak.verbose=1 on the kernel command line, or "echo 1 >
/sys/module/kmemleak/parameters/verbose":

This allows easier integration of kmemleak into test systems: We have
automated test infrastructure to test our Linux systems.  With this
option, running our tests with kmemleak is as simple as enabling kmemleak
and passing this command line option; the test infrastructure knows how to
save kernel logs, which will now include kmemleak reports.  Without this
option, the test infrastructure needs to be specifically taught to read
out the kmemleak debugfs file.  Removing this need for special handling
makes kmemleak more similar to other kernel debug options (slab debugging,
debug objects, etc).

Link: http://lkml.kernel.org/r/20180903144046.21023-1-vincent.whitchurch@axis.com
Signed-off-by: Vincent Whitchurch <vincent.whitchurch@axis.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agoRevert "mm, mmu_notifier: annotate mmu notifiers with blockable invalidate callbacks"
Michal Hocko [Fri, 26 Oct 2018 22:03:39 +0000 (15:03 -0700)]
Revert "mm, mmu_notifier: annotate mmu notifiers with blockable invalidate callbacks"

Revert 5ff7091f5a2ca ("mm, mmu_notifier: annotate mmu notifiers with
blockable invalidate callbacks").

MMU_INVALIDATE_DOES_NOT_BLOCK flags was the only one used and it is no
longer needed since 93065ac753e4 ("mm, oom: distinguish blockable mode for
mmu notifiers").  We now have a full support for per range !blocking
behavior so we can drop the stop gap workaround which the per notifier
flag was used for.

Link: http://lkml.kernel.org/r/20180827112623.8992-4-mhocko@kernel.org
Signed-off-by: Michal Hocko <mhocko@suse.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Cc: Jerome Glisse <jglisse@redhat.com>
Cc: Juergen Gross <jgross@suse.com>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agomm, mmu_notifier: be explicit about range invalition non-blocking mode
Michal Hocko [Fri, 26 Oct 2018 22:03:35 +0000 (15:03 -0700)]
mm, mmu_notifier: be explicit about range invalition non-blocking mode

If invalidate_range_start() is called for !blocking mode then all
callbacks have to guarantee they will no block/sleep.  The same obviously
applies to invalidate_range_end because this operation pairs with the
former and they are called from the same context.  Make sure this is
appropriately documented.

Link: http://lkml.kernel.org/r/20180827112623.8992-3-mhocko@kernel.org
Signed-off-by: Michal Hocko <mhocko@suse.com>
Reviewed-by: Jerome Glisse <jglisse@redhat.com>
Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Juergen Gross <jgross@suse.com>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agomm,page_alloc: PF_WQ_WORKER threads must sleep at should_reclaim_retry()
Michal Hocko [Fri, 26 Oct 2018 22:03:31 +0000 (15:03 -0700)]
mm,page_alloc: PF_WQ_WORKER threads must sleep at should_reclaim_retry()

Tetsuo Handa has reported that it is possible to bypass the short sleep
for PF_WQ_WORKER threads which was introduced by commit 373ccbe5927034b5
("mm, vmstat: allow WQ concurrency to discover memory reclaim doesn't make
any progress") and lock up the system if OOM.

The primary reason is that WQ_MEM_RECLAIM WQs are not guaranteed to run
even when they have a rescuer available.  Those workers might be essential
for reclaim to make a forward progress, however.  If we are too unlucky
all the allocations requests can get stuck waiting for a WQ_MEM_RECLAIM
work item and the system is essentially stuck in an OOM condition without
much hope to move on.  Tetsuo has seen the reclaim stuck on
drain_local_pages_wq or xlog_cil_push_work (xfs).  There might be others.

Since should_reclaim_retry() should be a natural reschedule point,
let's do the short sleep for PF_WQ_WORKER threads unconditionally in
order to guarantee that other pending work items are started.  This
will workaround this problem and it is less fragile than hunting down
when the sleep is missed.  Having a single sleeping point is more
robust.

[akpm@linux-foundation.org: reflow comment to 80 cols to save a couple of lines]
Link: http://lkml.kernel.org/r/20180827135101.15700-1-mhocko@kernel.org
Signed-off-by: Michal Hocko <mhocko@suse.com>
Debugged-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Reported-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Roman Gushchin <guro@fb.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: David Rientjes <rientjes@google.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>
6 years agomm: don't miss the last page because of round-off error
Roman Gushchin [Fri, 26 Oct 2018 22:03:27 +0000 (15:03 -0700)]
mm: don't miss the last page because of round-off error

I've noticed, that dying memory cgroups are often pinned in memory by a
single pagecache page.  Even under moderate memory pressure they sometimes
stayed in such state for a long time.  That looked strange.

My investigation showed that the problem is caused by applying the LRU
pressure balancing math:

  scan = div64_u64(scan * fraction[lru], denominator),

where

  denominator = fraction[anon] + fraction[file] + 1.

Because fraction[lru] is always less than denominator, if the initial scan
size is 1, the result is always 0.

This means the last page is not scanned and has
no chances to be reclaimed.

Fix this by rounding up the result of the division.

In practice this change significantly improves the speed of dying cgroups
reclaim.

[guro@fb.com: prevent double calculation of DIV64_U64_ROUND_UP() arguments]
Link: http://lkml.kernel.org/r/20180829213311.GA13501@castle
Link: http://lkml.kernel.org/r/20180827162621.30187-3-guro@fb.com
Signed-off-by: Roman Gushchin <guro@fb.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Tejun Heo <tj@kernel.org>
Cc: Rik van Riel <riel@surriel.com>
Cc: Konstantin Khlebnikov <koct9i@gmail.com>
Cc: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agomm: drain memcg stocks on css offlining
Roman Gushchin [Fri, 26 Oct 2018 22:03:23 +0000 (15:03 -0700)]
mm: drain memcg stocks on css offlining

Memcg charge is batched using per-cpu stocks, so an offline memcg can be
pinned by a cached charge up to a moment, when a process belonging to some
other cgroup will charge some memory on the same cpu.  In other words,
cached charges can prevent a memory cgroup from being reclaimed for some
time, without any clear need.

Let's optimize it by explicit draining of all stocks on css offlining.  As
draining is performed asynchronously, and is skipped if any parallel
draining is happening, it's cheap.

Link: http://lkml.kernel.org/r/20180827162621.30187-2-guro@fb.com
Signed-off-by: Roman Gushchin <guro@fb.com>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Acked-by: Michal Hocko <mhocko@kernel.org>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Konstantin Khlebnikov <koct9i@gmail.com>
Cc: Tejun Heo <tj@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agomm: rework memcg kernel stack accounting
Roman Gushchin [Fri, 26 Oct 2018 22:03:19 +0000 (15:03 -0700)]
mm: rework memcg kernel stack accounting

If CONFIG_VMAP_STACK is set, kernel stacks are allocated using
__vmalloc_node_range() with __GFP_ACCOUNT.  So kernel stack pages are
charged against corresponding memory cgroups on allocation and uncharged
on releasing them.

The problem is that we do cache kernel stacks in small per-cpu caches and
do reuse them for new tasks, which can belong to different memory cgroups.

Each stack page still holds a reference to the original cgroup, so the
cgroup can't be released until the vmap area is released.

To make this happen we need more than two subsequent exits without forks
in between on the current cpu, which makes it very unlikely to happen.  As
a result, I saw a significant number of dying cgroups (in theory, up to 2
* number_of_cpu + number_of_tasks), which can't be released even by
significant memory pressure.

As a cgroup structure can take a significant amount of memory (first of
all, per-cpu data like memcg statistics), it leads to a noticeable waste
of memory.

Link: http://lkml.kernel.org/r/20180827162621.30187-1-guro@fb.com
Fixes: ac496bf48d97 ("fork: Optimize task creation by caching two thread stacks per CPU if CONFIG_VMAP_STACK=y")
Signed-off-by: Roman Gushchin <guro@fb.com>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Acked-by: Michal Hocko <mhocko@kernel.org>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Konstantin Khlebnikov <koct9i@gmail.com>
Cc: Tejun Heo <tj@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agoslub: extend slub debug to handle multiple slabs
Aaron Tomlin [Fri, 26 Oct 2018 22:03:15 +0000 (15:03 -0700)]
slub: extend slub debug to handle multiple slabs

Extend the slub_debug syntax to "slub_debug=<flags>[,<slub>]*", where
<slub> may contain an asterisk at the end.  For example, the following
would poison all kmalloc slabs:

slub_debug=P,kmalloc*

and the following would apply the default flags to all kmalloc and all
block IO slabs:

slub_debug=,bio*,kmalloc*

Please note that a similar patch was posted by Iliyan Malchev some time
ago but was never merged:

https://marc.info/?l=linux-mm&m=131283905330474&w=2

Link: http://lkml.kernel.org/r/20180928111139.27962-1-atomlin@redhat.com
Signed-off-by: Aaron Tomlin <atomlin@redhat.com>
Acked-by: Christoph Lameter <cl@linux.com>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Iliyan Malchev <malchev@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agomm: don't warn about large allocations for slab
Dmitry Vyukov [Fri, 26 Oct 2018 22:03:12 +0000 (15:03 -0700)]
mm: don't warn about large allocations for slab

Slub does not call kmalloc_slab() for sizes > KMALLOC_MAX_CACHE_SIZE,
instead it falls back to kmalloc_large().

For slab KMALLOC_MAX_CACHE_SIZE == KMALLOC_MAX_SIZE and it calls
kmalloc_slab() for all allocations relying on NULL return value for
over-sized allocations.

This inconsistency leads to unwanted warnings from kmalloc_slab() for
over-sized allocations for slab.  Returning NULL for failed allocations is
the expected behavior.

Make slub and slab code consistent by checking size >
KMALLOC_MAX_CACHE_SIZE in slab before calling kmalloc_slab().

While we are here also fix the check in kmalloc_slab().  We should check
against KMALLOC_MAX_CACHE_SIZE rather than KMALLOC_MAX_SIZE.  It all kinda
worked because for slab the constants are the same, and slub always checks
the size against KMALLOC_MAX_CACHE_SIZE before kmalloc_slab().  But if we
get there with size > KMALLOC_MAX_CACHE_SIZE anyhow bad things will
happen.  For example, in case of a newly introduced bug in slub code.

Also move the check in kmalloc_slab() from function entry to the size >
192 case.  This partially compensates for the additional check in slab
code and makes slub code a bit faster (at least theoretically).

Also drop __GFP_NOWARN in the warning check.  This warning means a bug in
slab code itself, user-passed flags have nothing to do with it.

Nothing of this affects slob.

Link: http://lkml.kernel.org/r/20180927171502.226522-1-dvyukov@gmail.com
Signed-off-by: Dmitry Vyukov <dvyukov@google.com>
Reported-by: syzbot+87829a10073277282ad1@syzkaller.appspotmail.com
Reported-by: syzbot+ef4e8fc3a06e9019bb40@syzkaller.appspotmail.com
Reported-by: syzbot+6e438f4036df52cbb863@syzkaller.appspotmail.com
Reported-by: syzbot+8574471d8734457d98aa@syzkaller.appspotmail.com
Reported-by: syzbot+af1504df0807a083dbd9@syzkaller.appspotmail.com
Acked-by: Christoph Lameter <cl@linux.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: David Rientjes <rientjes@google.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>
6 years agomm/slub.c: switch to bitmap_zalloc()
Andy Shevchenko [Fri, 26 Oct 2018 22:03:06 +0000 (15:03 -0700)]
mm/slub.c: switch to bitmap_zalloc()

Switch to bitmap_zalloc() to show clearly what we are allocating.  Besides
that it returns pointer of bitmap type instead of opaque void *.

Link: http://lkml.kernel.org/r/20180830104301.61649-1-andriy.shevchenko@linux.intel.com
Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Acked-by: Christoph Lameter <cl@linux.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: David Rientjes <rientjes@google.com>
Cc: Pekka Enberg <penberg@kernel.org>
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>
6 years agoxtensa: use generic vga.h
Jiri Slaby [Fri, 26 Oct 2018 22:03:02 +0000 (15:03 -0700)]
xtensa: use generic vga.h

What xtensa has in asm/vga.h is the same as what can be found in
asm-generic/vga.h.  So use the latter header.

Link: http://lkml.kernel.org/r/20180907132219.12979-1-jslaby@suse.cz
Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Acked-by: Max Filippov <jcmvbkbc@gmail.com>
Cc: Chris Zankel <chris@zankel.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agofs/iomap.c: change return type to vm_fault_t
Souptick Joarder [Fri, 26 Oct 2018 22:02:59 +0000 (15:02 -0700)]
fs/iomap.c: change return type to vm_fault_t

Change iomap_page_mkwrite() return type to vm_fault_t.

see commit 1c8f422059ae ("mm: change return type to vm_fault_t") for
reference.

Link: http://lkml.kernel.org/r/20180827172050.GA18673@jordon-HP-15-Notebook-PC
Signed-off-by: Souptick Joarder <jrdr.linux@gmail.com>
Reviewed-by: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agoocfs2: remove set but not used variable 'rb'
YueHaibing [Fri, 26 Oct 2018 22:02:56 +0000 (15:02 -0700)]
ocfs2: remove set but not used variable 'rb'

Fixes gcc '-Wunused-but-set-variable' warning:

fs/ocfs2/refcounttree.c: In function 'ocfs2_create_reflink_node':
fs/ocfs2/refcounttree.c:4138:31: warning:
 variable 'rb' set but not used [-Wunused-but-set-variable]

Link: http://lkml.kernel.org/r/1536198443-113047-1-git-send-email-yuehaibing@huawei.com
Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Mark Fasheh <mark@fasheh.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Junxiao Bi <junxiao.bi@oracle.com>
Cc: Joseph Qi <jiangqi903@gmail.com>
Cc: Changwei Ge <ge.changwei@h3c.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agofs/ocfs2/dlm/dlmdebug.c: fix a sleep-in-atomic-context bug in dlm_print_one_mle()
Jia-Ju Bai [Fri, 26 Oct 2018 22:02:52 +0000 (15:02 -0700)]
fs/ocfs2/dlm/dlmdebug.c: fix a sleep-in-atomic-context bug in dlm_print_one_mle()

The kernel module may sleep with holding a spinlock.

The function call paths (from bottom to top) in Linux-4.16 are:

[FUNC] get_zeroed_page(GFP_NOFS)
fs/ocfs2/dlm/dlmdebug.c, 332: get_zeroed_page in dlm_print_one_mle
fs/ocfs2/dlm/dlmmaster.c, 240: dlm_print_one_mle in __dlm_put_mle
fs/ocfs2/dlm/dlmmaster.c, 255: __dlm_put_mle in dlm_put_mle
fs/ocfs2/dlm/dlmmaster.c, 254: spin_lock in dlm_put_ml

[FUNC] get_zeroed_page(GFP_NOFS)
fs/ocfs2/dlm/dlmdebug.c, 332: get_zeroed_page in dlm_print_one_mle
fs/ocfs2/dlm/dlmmaster.c, 240: dlm_print_one_mle in __dlm_put_mle
fs/ocfs2/dlm/dlmmaster.c, 222: __dlm_put_mle in dlm_put_mle_inuse
fs/ocfs2/dlm/dlmmaster.c, 219: spin_lock in dlm_put_mle_inuse

To fix this bug, GFP_NOFS is replaced with GFP_ATOMIC.

This bug is found by my static analysis tool DSAC.

Link: http://lkml.kernel.org/r/20180901112528.27025-1-baijiaju1990@gmail.com
Signed-off-by: Jia-Ju Bai <baijiaju1990@gmail.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Mark Fasheh <mark@fasheh.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Junxiao Bi <junxiao.bi@oracle.com>
Cc: Joseph Qi <jiangqi903@gmail.com>
Cc: Changwei Ge <ge.changwei@h3c.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agoocfs2: remove unneeded null check
Ding Xiang [Fri, 26 Oct 2018 22:02:48 +0000 (15:02 -0700)]
ocfs2: remove unneeded null check

Null check for kfree is unnecessary, so remove it.

Link: http://lkml.kernel.org/r/1535704514-26559-1-git-send-email-dingxiang@cmss.chinamobile.com
Signed-off-by: Ding Xiang <dingxiang@cmss.chinamobile.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Mark Fasheh <mark@fasheh.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Junxiao Bi <junxiao.bi@oracle.com>
Cc: Joseph Qi <jiangqi903@gmail.com>
Cc: Changwei Ge <ge.changwei@h3c.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agoocfs2: remove unused pointer 'eb'
Colin Ian King [Fri, 26 Oct 2018 22:02:45 +0000 (15:02 -0700)]
ocfs2: remove unused pointer 'eb'

Pointer 'eb' is being assigned but is never used hence it is
redundant and can be removed.

Cleans up clang warning:
warning: variable 'eb' set but not used [-Wunused-but-set-variable]

Link: http://lkml.kernel.org/r/20180828141907.10826-1-colin.king@canonical.com
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Mark Fasheh <mark@fasheh.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Junxiao Bi <junxiao.bi@oracle.com>
Cc: Joseph Qi <jiangqi903@gmail.com>
Cc: Changwei Ge <ge.changwei@h3c.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agoocfs2/dlm: remove unnecessary parentheses
Nathan Chancellor [Fri, 26 Oct 2018 22:02:41 +0000 (15:02 -0700)]
ocfs2/dlm: remove unnecessary parentheses

Clang warns when more than one set of parentheses is used for a
single conditional statement:

fs/ocfs2/dlm/dlmthread.c:534:18: warning: equality comparison with extraneous
      parentheses [-Wparentheses-equality]
        if ((res->owner == dlm->node_num)) {
             ~~~~~~~~~~~^~~~~~~~~~~~~~~~
fs/ocfs2/dlm/dlmthread.c:534:18: note: remove extraneous parentheses around the
      comparison to silence this warning
        if ((res->owner == dlm->node_num)) {
            ~           ^               ~

Link: http://lkml.kernel.org/r/20180924181929.6853-1-natechancellor@gmail.com
Signed-off-by: Nathan Chancellor <natechancellor@gmail.com>
Reported-by: Nick Desaulniers <ndesaulniers@google.com>
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agoscripts/tags.sh: add DECLARE_HASHTABLE()
Kirill Tkhai [Fri, 26 Oct 2018 22:02:38 +0000 (15:02 -0700)]
scripts/tags.sh: add DECLARE_HASHTABLE()

In addition to DEFINE_HASHTABLE() add DECLARE_ variant.

Link: http://lkml.kernel.org/r/153683203215.13678.11468076350083405643.stgit@localhost.localdomain
Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Masahiro Yamada <yamada.masahiro@socionext.com>
Cc: Constantine Shulyupin <const@MakeLinux.com>
Cc: Arend van Spriel <arend.vanspriel@broadcom.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Joey Pabalinas <joeypabalinas@gmail.com>
Cc: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agolib/test_kasan.c: add tests for several string/memory API functions
Andrey Ryabinin [Fri, 26 Oct 2018 22:02:34 +0000 (15:02 -0700)]
lib/test_kasan.c: add tests for several string/memory API functions

Arch code may have asm implementation of string/memory API functions
instead of using generic one from lib/string.c.  KASAN don't see memory
accesses in asm code, thus can miss many bugs.

E.g.  on ARM64 KASAN don't see bugs in memchr(), memcmp(), str[r]chr(),
str[n]cmp(), str[n]len().  Add tests for these functions to be sure that
we notice the problem on other architectures.

Link: http://lkml.kernel.org/r/20180920135631.23833-3-aryabinin@virtuozzo.com
Signed-off-by: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Alexander Potapenko <glider@google.com>
Cc: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: Kyeongdon Kim <kyeongdon.kim@lge.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agoarm64: lib: use C string functions with KASAN enabled
Andrey Ryabinin [Fri, 26 Oct 2018 22:02:30 +0000 (15:02 -0700)]
arm64: lib: use C string functions with KASAN enabled

ARM64 has asm implementation of memchr(), memcmp(), str[r]chr(),
str[n]cmp(), str[n]len().  KASAN don't see memory accesses in asm code,
thus it can potentially miss many bugs.

Ifdef out __HAVE_ARCH_* defines of these functions when KASAN is enabled,
so the generic implementations from lib/string.c will be used.

We can't just remove the asm functions because efistub uses them.  And we
can't have two non-weak functions either, so declare the asm functions as
weak.

Link: http://lkml.kernel.org/r/20180920135631.23833-2-aryabinin@virtuozzo.com
Signed-off-by: Andrey Ryabinin <aryabinin@virtuozzo.com>
Reported-by: Kyeongdon Kim <kyeongdon.kim@lge.com>
Cc: Alexander Potapenko <glider@google.com>
Cc: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agoinclude/linux/linkage.h: align weak symbols
Andrey Ryabinin [Fri, 26 Oct 2018 22:02:27 +0000 (15:02 -0700)]
include/linux/linkage.h: align weak symbols

Since WEAK() supposed to be used instead of ENTRY() to define weak
symbols, but unlike ENTRY() it doesn't have ALIGN directive.  It seems
there is no actual reason to not have, so let's add ALIGN to WEAK() too.

Link: http://lkml.kernel.org/r/20180920135631.23833-1-aryabinin@virtuozzo.com
Signed-off-by: Andrey Ryabinin <aryabinin@virtuozzo.com>
Will Deacon <will.deacon@arm.com>, Catalin Marinas <catalin.marinas@arm.com>
Cc: Kyeongdon Kim <kyeongdon.kim@lge.com>
Cc: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Cc: Alexander Potapenko <glider@google.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agoinclude/linux/pfn_t.h: force '~' to be parsed as an unary operator
Sebastien Boisvert [Fri, 26 Oct 2018 22:02:23 +0000 (15:02 -0700)]
include/linux/pfn_t.h: force '~' to be parsed as an unary operator

Tracing the event "fs_dax:dax_pmd_insert_mapping" with perf produces this
warning:

      [fs_dax:dax_pmd_insert_mapping] unknown op '~'

It is printed in process_op (tools/lib/traceevent/event-parse.c) because
'~' is parsed as a binary operator.

perf reads the format of fs_dax:dax_pmd_insert_mapping ("print fmt") from
/sys/kernel/debug/tracing/events/fs_dax/dax_pmd_insert_mapping/format .

The format contains:

~(((u64) ~(~(((1UL) << 12)-1)))
         ^
         \ interpreted as a binary operator by process_op().

This part is generated in the declaration of the event class
dax_pmd_insert_mapping_class in include/trace/events/fs_dax.h :

__print_flags_u64(__entry->pfn_val & PFN_FLAGS_MASK, "|",
PFN_FLAGS_TRACE),

This patch adds a pair of parentheses in the declaration of PFN_FLAGS_MASK
to make sure that '~' is parsed as a unary operator by perf.

The part of the format that was problematic is now:

~(((u64) (~(~(((1UL) << 12)-1))))

Now, all the '~' are parsed as unary operators.

Link: http://lkml.kernel.org/r/20181021145939.8760-1-sebhtml@videotron.qc.ca
Signed-off-by: Sebastien Boisvert <sebhtml@videotron.qc.ca>
Acked-by: Dan Williams <dan.j.williams@intel.com>
Cc: "Steven Rostedt (VMware)" <rostedt@goodmis.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: "Tzvetomir Stoyanov (VMware)" <tz.stoyanov@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Ross Zwisler <zwisler@kernel.org>
Cc: Elenie Godzaridis <arangradient@gmail.com>
Cc: <stable@vger.kerenl.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agouserfaultfd: disable irqs when taking the waitqueue lock
Christoph Hellwig [Fri, 26 Oct 2018 22:02:19 +0000 (15:02 -0700)]
userfaultfd: disable irqs when taking the waitqueue lock

userfaultfd contains howe-grown locking of the waitqueue lock, and does
not disable interrupts.  This relies on the fact that no one else takes it
from interrupt context and violates an invariat of the normal waitqueue
locking scheme.  With aio poll it is easy to trigger other locks that
disable interrupts (or are called from interrupt context).

Link: http://lkml.kernel.org/r/20181018154101.18750-1-hch@lst.de
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Andrea Arcangeli <aarcange@redhat.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: <stable@vger.kernel.org> [4.19.x]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agomm: /proc/pid/smaps_rollup: fix NULL pointer deref in smaps_pte_range()
Vlastimil Babka [Fri, 26 Oct 2018 22:02:16 +0000 (15:02 -0700)]
mm: /proc/pid/smaps_rollup: fix NULL pointer deref in smaps_pte_range()

Leonardo reports an apparent regression in 4.19-rc7:

 BUG: unable to handle kernel NULL pointer dereference at 00000000000000f0
 PGD 0 P4D 0
 Oops: 0000 [#1] PREEMPT SMP PTI
 CPU: 3 PID: 6032 Comm: python Not tainted 4.19.0-041900rc7-lowlatency #201810071631
 Hardware name: LENOVO 80UG/Toronto 4A2, BIOS 0XCN45WW 08/09/2018
 RIP: 0010:smaps_pte_range+0x32d/0x540
 Code: 80 00 00 00 00 74 a9 48 89 de 41 f6 40 52 40 0f 85 04 02 00 00 49 2b 30 48 c1 ee 0c 49 03 b0 98 00 00 00 49 8b 80 a0 00 00 00 <48> 8b b8 f0 00 00 00 e8 b7 ef ec ff 48 85 c0 0f 84 71 ff ff ff a8
 RSP: 0018:ffffb0cbc484fb88 EFLAGS: 00010202
 RAX: 0000000000000000 RBX: 0000560ddb9e9000 RCX: 0000000000000000
 RDX: 0000000000000000 RSI: 0000000560ddb9e9 RDI: 0000000000000001
 RBP: ffffb0cbc484fbc0 R08: ffff94a5a227a578 R09: ffff94a5a227a578
 R10: 0000000000000000 R11: 0000560ddbbe7000 R12: ffffe903098ba728
 R13: ffffb0cbc484fc78 R14: ffffb0cbc484fcf8 R15: ffff94a5a2e9cf48
 FS:  00007f6dfb683740(0000) GS:ffff94a5aaf80000(0000) knlGS:0000000000000000
 CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
 CR2: 00000000000000f0 CR3: 000000011c118001 CR4: 00000000003606e0
 DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
 DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
 Call Trace:
  __walk_page_range+0x3c2/0x6f0
  walk_page_vma+0x42/0x60
  smap_gather_stats+0x79/0xe0
  ? gather_pte_stats+0x320/0x320
  ? gather_hugetlb_stats+0x70/0x70
  show_smaps_rollup+0xcd/0x1c0
  seq_read+0x157/0x400
  __vfs_read+0x3a/0x180
  ? security_file_permission+0x93/0xc0
  ? security_file_permission+0x93/0xc0
  vfs_read+0x8f/0x140
  ksys_read+0x55/0xc0
  __x64_sys_read+0x1a/0x20
  do_syscall_64+0x5a/0x110
  entry_SYSCALL_64_after_hwframe+0x44/0xa9

Decoded code matched to local compilation+disassembly points to
smaps_pte_entry():

        } else if (unlikely(IS_ENABLED(CONFIG_SHMEM) && mss->check_shmem_swap
                                                        && pte_none(*pte))) {
                page = find_get_entry(vma->vm_file->f_mapping,
                                                linear_page_index(vma, addr));

Here, vma->vm_file is NULL.  mss->check_shmem_swap should be false in that
case, however for smaps_rollup, smap_gather_stats() can set the flag true
for one vma and leave it true for subsequent vma's where it should be
false.

To fix, reset the check_shmem_swap flag to false.  There's also related
bug which sets mss->swap to shmem_swapped, which in the context of
smaps_rollup overwrites any value accumulated from previous vma's.  Fix
that as well.

Note that the report suggests a regression between 4.17.19 and 4.19-rc7,
which makes the 4.19 series ending with commit 258f669e7e88 ("mm:
/proc/pid/smaps_rollup: convert to single value seq_file") suspicious.
But the mss was reused for rollup since 493b0e9d945f ("mm: add
/proc/pid/smaps_rollup") so let's play it safe with the stable backport.

Link: http://lkml.kernel.org/r/555fbd1f-4ac9-0b58-dcd4-5dc4380ff7ca@suse.cz
Link: https://bugzilla.kernel.org/show_bug.cgi?id=201377
Fixes: 493b0e9d945f ("mm: add /proc/pid/smaps_rollup")
Signed-off-by: Vlastimil Babka <vbabka@suse.cz>
Reported-by: Leonardo Soares Müller <leozinho29_eu@hotmail.com>
Tested-by: Leonardo Soares Müller <leozinho29_eu@hotmail.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Daniel Colascione <dancol@google.com>
Cc: Alexey Dobriyan <adobriyan@gmail.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
6 years agoMerge tag 'char-misc-4.20-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/gregk...
Linus Torvalds [Fri, 26 Oct 2018 16:11:43 +0000 (09:11 -0700)]
Merge tag 'char-misc-4.20-rc1' of git://git./linux/kernel/git/gregkh/char-misc

Pull char/misc driver updates from Greg KH:
 "Here is the big set of char/misc patches for 4.20-rc1.

  Loads of things here, we have new code in all of these driver
  subsystems:
   - fpga
   - stm
   - extcon
   - nvmem
   - eeprom
   - hyper-v
   - gsmi
   - coresight
   - thunderbolt
   - vmw_balloon
   - goldfish
   - soundwire
  along with lots of fixes and minor changes to other small drivers.

  All of these have been in linux-next for a while with no reported
  issues"

* tag 'char-misc-4.20-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/char-misc: (245 commits)
  Documentation/security-bugs: Clarify treatment of embargoed information
  lib: Fix ia64 bootloader linkage
  MAINTAINERS: Clarify UIO vs UIOVEC maintainer
  docs/uio: fix a grammar nitpick
  docs: fpga: document programming fpgas using regions
  fpga: add devm_fpga_region_create
  fpga: bridge: add devm_fpga_bridge_create
  fpga: mgr: add devm_fpga_mgr_create
  hv_balloon: Replace spin_is_locked() with lockdep
  sgi-xp: Replace spin_is_locked() with lockdep
  eeprom: New ee1004 driver for DDR4 memory
  eeprom: at25: remove unneeded 'at25_remove'
  w1: IAD Register is yet readable trough iad sys file. Fix snprintf (%u for unsigned, count for max size).
  misc: mic: scif: remove set but not used variables 'src_dma_addr, dst_dma_addr'
  misc: mic: fix a DMA pool free failure
  platform: goldfish: pipe: Add a blank line to separate varibles and code
  platform: goldfish: pipe: Remove redundant casting
  platform: goldfish: pipe: Call misc_deregister if init fails
  platform: goldfish: pipe: Move the file-scope goldfish_pipe_dev variable into the driver state
  platform: goldfish: pipe: Move the file-scope goldfish_pipe_miscdev variable into the driver state
  ...

6 years agoMerge tag 'driver-core-4.20-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Fri, 26 Oct 2018 15:42:25 +0000 (08:42 -0700)]
Merge tag 'driver-core-4.20-rc1' of git://git./linux/kernel/git/gregkh/driver-core

Pull driver core updates from Greg KH:
 "Here is a small number of driver core patches for 4.20-rc1.

  Not much happened here this merge window, only a very tiny number of
  patches that do:

   - add BUS_ATTR_WO() for use by drivers

   - component error path fixes

   - kernfs range check fix

   - other tiny error path fixes and const changes

  All of these have been in linux-next with no reported issues for a
  while"

* tag 'driver-core-4.20-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/driver-core:
  devres: provide devm_kstrdup_const()
  mm: move is_kernel_rodata() to asm-generic/sections.h
  devres: constify p in devm_kfree()
  driver core: add BUS_ATTR_WO() macro
  kernfs: Fix range checks in kernfs_get_target_path
  component: fix loop condition to call unbind() if bind() fails
  drivers/base/devtmpfs.c: don't pretend path is const in delete_path
  kernfs: update comment about kernfs_path() return value

6 years agoMerge tag 'usb-4.20-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/usb
Linus Torvalds [Fri, 26 Oct 2018 15:14:13 +0000 (08:14 -0700)]
Merge tag 'usb-4.20-rc1' of git://git./linux/kernel/git/gregkh/usb

Pull USB/PHY updates from Greg KH:
 "Here is the big USB/PHY driver patches for 4.20-rc1

  Lots of USB changes in here, primarily in these areas:

   - typec updates and new drivers

   - new PHY drivers

   - dwc2 driver updates and additions (this old core keeps getting
     added to new devices.)

   - usbtmc major update based on the industry group coming together and
     working to add new features and performance to the driver.

   - USB gadget additions for new features

   - USB gadget configfs updates

   - chipidea driver updates

   - other USB gadget updates

   - USB serial driver updates

   - renesas driver updates

   - xhci driver updates

   - other tiny USB driver updates

  All of these have been in linux-next for a while with no reported
  issues"

* tag 'usb-4.20-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/usb: (229 commits)
  usb: phy: ab8500: silence some uninitialized variable warnings
  usb: xhci: tegra: Add genpd support
  usb: xhci: tegra: Power-off power-domains on removal
  usbip:vudc: BUG kmalloc-2048 (Not tainted): Poison overwritten
  usbip: tools: fix atoi() on non-null terminated string
  USB: misc: appledisplay: fix backlight update_status return code
  phy: phy-pxa-usb: add a new driver
  usb: host: add DT bindings for faraday fotg2
  usb: host: ohci-at91: fix request of irq for optional gpio
  usb/early: remove set but not used variable 'remain_length'
  usb: typec: Fix copy/paste on typec_set_vconn_role() kerneldoc
  usb: typec: tcpm: Report back negotiated PPS voltage and current
  USB: core: remove set but not used variable 'udev'
  usb: core: fix memory leak on port_dev_path allocation
  USB: net2280: Remove ->disconnect() callback from net2280_pullup()
  usb: dwc2: disable power_down on rockchip devices
  usb: gadget: udc: renesas_usb3: add support for r8a77990
  dt-bindings: usb: renesas_usb3: add bindings for r8a77990
  usb: gadget: udc: renesas_usb3: Add r8a774a1 support
  USB: serial: cypress_m8: remove set but not used variable 'iflag'
  ...

6 years agoMerge tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma
Linus Torvalds [Fri, 26 Oct 2018 14:38:19 +0000 (07:38 -0700)]
Merge tag 'for-linus' of git://git./linux/kernel/git/rdma/rdma

Pull rdma updates from Jason Gunthorpe:
 "This has been a smaller cycle with many of the commits being smallish
  code fixes and improvements across the drivers.

   - Driver updates for bnxt_re, cxgb4, hfi1, hns, mlx5, nes, qedr, and
     rxe

   - Memory window support in hns

   - mlx5 user API 'flow mutate/steering' allows accessing the full
     packet mangling and matching machinery from user space

   - Support inter-working with verbs API calls in the 'devx' mlx5 user
     API, and provide options to use devx with less privilege

   - Modernize the use of syfs and the device interface to use attribute
     groups and cdev properly for uverbs, and clean up some of the core
     code's device list management

   - More progress on net namespaces for RDMA devices

   - Consolidate driver BAR mmapping support into core code helpers and
     rework how RDMA holds poitners to mm_struct for get_user_pages
     cases

   - First pass to use 'dev_name' instead of ib_device->name

   - Device renaming for RDMA devices"

* tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma: (242 commits)
  IB/mlx5: Add support for extended atomic operations
  RDMA/core: Fix comment for hw stats init for port == 0
  RDMA/core: Refactor ib_register_device() function
  RDMA/core: Fix unwinding flow in case of error to register device
  ib_srp: Remove WARN_ON in srp_terminate_io()
  IB/mlx5: Allow scatter to CQE without global signaled WRs
  IB/mlx5: Verify that driver supports user flags
  IB/mlx5: Support scatter to CQE for DC transport type
  RDMA/drivers: Use core provided API for registering device attributes
  RDMA/core: Allow existing drivers to set one sysfs group per device
  IB/rxe: Remove unnecessary enum values
  RDMA/umad: Use kernel API to allocate umad indexes
  RDMA/uverbs: Use kernel API to allocate uverbs indexes
  RDMA/core: Increase total number of RDMA ports across all devices
  IB/mlx4: Add port and TID to MAD debug print
  IB/mlx4: Enable debug print of SMPs
  RDMA/core: Rename ports_parent to ports_kobj
  RDMA/core: Do not expose unsupported counters
  IB/mlx4: Refer to the device kobject instead of ports_parent
  RDMA/nldev: Allow IB device rename through RDMA netlink
  ...

6 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/sparc
Linus Torvalds [Fri, 26 Oct 2018 01:14:31 +0000 (18:14 -0700)]
Merge git://git./linux/kernel/git/davem/sparc

Pull sparc fix from David Miller:
 "Build regression fix"

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/sparc:
  sparc: Fix VDSO build with older binutils.

6 years agoMerge tag 'riscv-for-linus-4.20-mw0' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Fri, 26 Oct 2018 01:01:29 +0000 (18:01 -0700)]
Merge tag 'riscv-for-linus-4.20-mw0' of git://git./linux/kernel/git/palmer/riscv-linux

Pull RISC-V updates from Palmer Dabbelt:
 "This patch set contains a lot (at least, for me) of improvements to
  the RISC-V kernel port:

   - The removal of some cacheinfo values that were bogus.

   - On systems with F but without D the kernel will not show the F
     extension to userspace, as it isn't actually supported.

   - Support for futexes.

   - Removal of some unused code.

   - Cleanup of some menuconfig entries.

   - Support for systems without a floating-point unit, and for building
     kernels that will never use the floating-point unit.

   - More fixes to the RV32I port, which regressed again. It's really
     time to get this into a regression test somewhere so I stop
     breaking it. Thanks to Zong for resurrecting it again!

   - Various fixes that resulted from a year old review of our original
     patch set that I finally got around to.

   - Various improvements to SMP support, largely based around having
     switched to logical hart numbering, as well as some interrupt
     improvements. This one is in the same patch set as above, thanks to
     Atish for sheparding everything though as my patch set was a bit of
     a mess.

  I'm pretty sure this is our largest patch set since the original
  kernel contribution, and it's certainly the one with the most
  contributors. While I don't have anything else I know I'm going to
  submit for the merge window, I would be somewhat surprised if I didn't
  screw anything up.

  Thanks for the help, everyone!"

* tag 'riscv-for-linus-4.20-mw0' of git://git.kernel.org/pub/scm/linux/kernel/git/palmer/riscv-linux: (31 commits)
  RISC-V: Cosmetic menuconfig changes
  riscv: move GCC version check for ARCH_SUPPORTS_INT128 to Kconfig
  RISC-V: remove the unused return_to_handler export
  RISC-V: Add futex support.
  RISC-V: Add FP register ptrace support for gdb.
  RISC-V: Mask out the F extension on systems without D
  RISC-V: Don't set cacheinfo.{physical_line_partition,attributes}
  RISC-V: Show IPI stats
  RISC-V: Show CPU ID and Hart ID separately in /proc/cpuinfo
  RISC-V: Use Linux logical CPU number instead of hartid
  RISC-V: Add logical CPU indexing for RISC-V
  RISC-V: Use WRITE_ONCE instead of direct access
  RISC-V: Use mmgrab()
  RISC-V: Rename im_okay_therefore_i_am to found_boot_cpu
  RISC-V: Rename riscv_of_processor_hart to riscv_of_processor_hartid
  RISC-V: Provide a cleaner raw_smp_processor_id()
  RISC-V: Disable preemption before enabling interrupts
  RISC-V: Comment on the TLB flush in smp_callin()
  RISC-V: Filter ISA and MMU values in cpuinfo
  RISC-V: Don't set cacheinfo.{physical_line_partition,attributes}
  ...

6 years agoMerge tag 'kvm-4.20-1' of git://git.kernel.org/pub/scm/virt/kvm/kvm
Linus Torvalds [Fri, 26 Oct 2018 00:57:35 +0000 (17:57 -0700)]
Merge tag 'kvm-4.20-1' of git://git./virt/kvm/kvm

Pull KVM updates from Radim Krčmář:
 "ARM:
   - Improved guest IPA space support (32 to 52 bits)

   - RAS event delivery for 32bit

   - PMU fixes

   - Guest entry hardening

   - Various cleanups

   - Port of dirty_log_test selftest

  PPC:
   - Nested HV KVM support for radix guests on POWER9. The performance
     is much better than with PR KVM. Migration and arbitrary level of
     nesting is supported.

   - Disable nested HV-KVM on early POWER9 chips that need a particular
     hardware bug workaround

   - One VM per core mode to prevent potential data leaks

   - PCI pass-through optimization

   - merge ppc-kvm topic branch and kvm-ppc-fixes to get a better base

  s390:
   - Initial version of AP crypto virtualization via vfio-mdev

   - Improvement for vfio-ap

   - Set the host program identifier

   - Optimize page table locking

  x86:
   - Enable nested virtualization by default

   - Implement Hyper-V IPI hypercalls

   - Improve #PF and #DB handling

   - Allow guests to use Enlightened VMCS

   - Add migration selftests for VMCS and Enlightened VMCS

   - Allow coalesced PIO accesses

   - Add an option to perform nested VMCS host state consistency check
     through hardware

   - Automatic tuning of lapic_timer_advance_ns

   - Many fixes, minor improvements, and cleanups"

* tag 'kvm-4.20-1' of git://git.kernel.org/pub/scm/virt/kvm/kvm: (204 commits)
  KVM/nVMX: Do not validate that posted_intr_desc_addr is page aligned
  Revert "kvm: x86: optimize dr6 restore"
  KVM: PPC: Optimize clearing TCEs for sparse tables
  x86/kvm/nVMX: tweak shadow fields
  selftests/kvm: add missing executables to .gitignore
  KVM: arm64: Safety check PSTATE when entering guest and handle IL
  KVM: PPC: Book3S HV: Don't use streamlined entry path on early POWER9 chips
  arm/arm64: KVM: Enable 32 bits kvm vcpu events support
  arm/arm64: KVM: Rename function kvm_arch_dev_ioctl_check_extension()
  KVM: arm64: Fix caching of host MDCR_EL2 value
  KVM: VMX: enable nested virtualization by default
  KVM/x86: Use 32bit xor to clear registers in svm.c
  kvm: x86: Introduce KVM_CAP_EXCEPTION_PAYLOAD
  kvm: vmx: Defer setting of DR6 until #DB delivery
  kvm: x86: Defer setting of CR2 until #PF delivery
  kvm: x86: Add payload operands to kvm_multiple_exception
  kvm: x86: Add exception payload fields to kvm_vcpu_events
  kvm: x86: Add has_payload and payload to kvm_queued_exception
  KVM: Documentation: Fix omission in struct kvm_vcpu_events
  KVM: selftests: add Enlightened VMCS test
  ...

6 years agoMerge branch 'for-4.20' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/cgroup
Linus Torvalds [Fri, 26 Oct 2018 00:15:46 +0000 (17:15 -0700)]
Merge branch 'for-4.20' of git://git./linux/kernel/git/tj/cgroup

Pull cgroup updates from Tejun Heo:
 "All trivial changes - simplification, typo fix and adding
  cond_resched() in a netclassid update loop"

* 'for-4.20' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/cgroup:
  cgroup, netclassid: add a preemption point to write_classid
  rdmacg: fix a typo in rdmacg documentation
  cgroup: Simplify cgroup_ancestor

6 years agoMerge tag 'printk-for-4.20' of git://git.kernel.org/pub/scm/linux/kernel/git/pmladek...
Linus Torvalds [Fri, 26 Oct 2018 00:11:52 +0000 (17:11 -0700)]
Merge tag 'printk-for-4.20' of git://git./linux/kernel/git/pmladek/printk

Pull printk updates from Petr Mladek:

 - Fix two more locations where printf formatting leaked pointers

 - Better log_buf_len parameter handling

 - Add prefix to messages from printk code

 - Do not miss messages on other consoles when the log is replayed on a
   new one

 - Reduce race between console registration and panic() when the log
   might get replayed on all consoles

 - Some cont buffer code clean up

 - Call console only when there is something to do (log vs cont buffer)

* tag 'printk-for-4.20' of git://git.kernel.org/pub/scm/linux/kernel/git/pmladek/printk:
  lib/vsprintf: Hash printed address for netdev bits fallback
  lib/vsprintf: Hash legacy clock addresses
  lib/vsprintf: Prepare for more general use of ptr_to_id()
  lib/vsprintf: Make ptr argument conts in ptr_to_id()
  printk: fix integer overflow in setup_log_buf()
  printk: do not preliminary split up cont buffer
  printk: lock/unlock console only for new logbuf entries
  printk: keep kernel cont support always enabled
  printk: Give error on attempt to set log buffer length to over 2G
  printk: Add KBUILD_MODNAME and remove a redundant print prefix
  printk: Correct wrong casting
  printk: Fix panic caused by passing log_buf_len to command line
  printk: CON_PRINTBUFFER console registration is a bit racy
  printk: Do not miss new messages when replaying the log

6 years agoMerge branch 'linus' of git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6
Linus Torvalds [Thu, 25 Oct 2018 23:43:35 +0000 (16:43 -0700)]
Merge branch 'linus' of git://git./linux/kernel/git/herbert/crypto-2.6

Pull crypto updates from Herbert Xu:
 "API:
   - Remove VLA usage
   - Add cryptostat user-space interface
   - Add notifier for new crypto algorithms

  Algorithms:
   - Add OFB mode
   - Remove speck

  Drivers:
   - Remove x86/sha*-mb as they are buggy
   - Remove pcbc(aes) from x86/aesni
   - Improve performance of arm/ghash-ce by up to 85%
   - Implement CTS-CBC in arm64/aes-blk, faster by up to 50%
   - Remove PMULL based arm64/crc32 driver
   - Use PMULL in arm64/crct10dif
   - Add aes-ctr support in s5p-sss
   - Add caam/qi2 driver

  Others:
   - Pick better transform if one becomes available in crc-t10dif"

* 'linus' of git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6: (124 commits)
  crypto: chelsio - Update ntx queue received from cxgb4
  crypto: ccree - avoid implicit enum conversion
  crypto: caam - add SPDX license identifier to all files
  crypto: caam/qi - simplify CGR allocation, freeing
  crypto: mxs-dcp - make symbols 'sha1_null_hash' and 'sha256_null_hash' static
  crypto: arm64/aes-blk - ensure XTS mask is always loaded
  crypto: testmgr - fix sizeof() on COMP_BUF_SIZE
  crypto: chtls - remove set but not used variable 'csk'
  crypto: axis - fix platform_no_drv_owner.cocci warnings
  crypto: x86/aes-ni - fix build error following fpu template removal
  crypto: arm64/aes - fix handling sub-block CTS-CBC inputs
  crypto: caam/qi2 - avoid double export
  crypto: mxs-dcp - Fix AES issues
  crypto: mxs-dcp - Fix SHA null hashes and output length
  crypto: mxs-dcp - Implement sha import/export
  crypto: aegis/generic - fix for big endian systems
  crypto: morus/generic - fix for big endian systems
  crypto: lrw - fix rebase error after out of bounds fix
  crypto: cavium/nitrox - use pci_alloc_irq_vectors() while enabling MSI-X.
  crypto: cavium/nitrox - NITROX command queue changes.
  ...

6 years agoMerge branch 'next-loadpin' of git://git.kernel.org/pub/scm/linux/kernel/git/jmorris...
Linus Torvalds [Thu, 25 Oct 2018 20:32:00 +0000 (13:32 -0700)]
Merge branch 'next-loadpin' of git://git./linux/kernel/git/jmorris/linux-security

Pull LoadPin updates from James Morris:
 "From Kees: This is a small reporting improvement and the param change
  needed for the ordering series (but since the loadpin change is
  desired and separable, I'm putting it here)"

* 'next-loadpin' of git://git.kernel.org/pub/scm/linux/kernel/git/jmorris/linux-security:
  LoadPin: Rename boot param "enabled" to "enforce"
  LoadPin: Report friendly block device name

6 years agoMerge branch 'next-smack' of git://git.kernel.org/pub/scm/linux/kernel/git/jmorris...
Linus Torvalds [Thu, 25 Oct 2018 20:29:51 +0000 (13:29 -0700)]
Merge branch 'next-smack' of git://git./linux/kernel/git/jmorris/linux-security

Pull smack updates from James Morris:
 "From Casey: three patches for Smack for 4.20. Two clean up warnings
  and one is a rarely encountered ptrace capability check"

* 'next-smack' of git://git.kernel.org/pub/scm/linux/kernel/git/jmorris/linux-security:
  Smack: Mark expected switch fall-through
  Smack: ptrace capability use fixes
  Smack: remove set but not used variable 'root_inode'

6 years agoMerge branch 'next-tpm' of git://git.kernel.org/pub/scm/linux/kernel/git/jmorris...
Linus Torvalds [Thu, 25 Oct 2018 20:25:18 +0000 (13:25 -0700)]
Merge branch 'next-tpm' of git://git./linux/kernel/git/jmorris/linux-security

Pull TPM updates from James Morris:
 "From Jarkko: The only new feature is non-blocking operation for
  /dev/tpm0"

* 'next-tpm' of git://git.kernel.org/pub/scm/linux/kernel/git/jmorris/linux-security:
  tpm: Restore functionality to xen vtpm driver.
  tpm: add support for nonblocking operation
  tpm: add ptr to the tpm_space struct to file_priv
  tpm: Make SECURITYFS a weak dependency
  tpm: suppress transmit cmd error logs when TPM 1.2 is disabled/deactivated
  tpm: fix response size validation in tpm_get_random()