]> git.proxmox.com Git - mirror_ubuntu-eoan-kernel.git/log
mirror_ubuntu-eoan-kernel.git
9 years agohfs,hfsplus: cache pages correctly between bnode_create and bnode_free
Hin-Tak Leung [Wed, 9 Sep 2015 22:38:04 +0000 (15:38 -0700)]
hfs,hfsplus: cache pages correctly between bnode_create and bnode_free

Pages looked up by __hfs_bnode_create() (called by hfs_bnode_create() and
hfs_bnode_find() for finding or creating pages corresponding to an inode)
are immediately kmap()'ed and used (both read and write) and kunmap()'ed,
and should not be page_cache_release()'ed until hfs_bnode_free().

This patch fixes a problem I first saw in July 2012: merely running "du"
on a large hfsplus-mounted directory a few times on a reasonably loaded
system would get the hfsplus driver all confused and complaining about
B-tree inconsistencies, and generates a "BUG: Bad page state".  Most
recently, I can generate this problem on up-to-date Fedora 22 with shipped
kernel 4.0.5, by running "du /" (="/" + "/home" + "/mnt" + other smaller
mounts) and "du /mnt" simultaneously on two windows, where /mnt is a
lightly-used QEMU VM image of the full Mac OS X 10.9:

$ df -i / /home /mnt
Filesystem                  Inodes   IUsed      IFree IUse% Mounted on
/dev/mapper/fedora-root    3276800  551665    2725135   17% /
/dev/mapper/fedora-home   52879360  716221   52163139    2% /home
/dev/nbd0p2             4294967295 1387818 4293579477    1% /mnt

After applying the patch, I was able to run "du /" (60+ times) and "du
/mnt" (150+ times) continuously and simultaneously for 6+ hours.

There are many reports of the hfsplus driver getting confused under load
and generating "BUG: Bad page state" or other similar issues over the
years.  [1]

The unpatched code [2] has always been wrong since it entered the kernel
tree.  The only reason why it gets away with it is that the
kmap/memcpy/kunmap follow very quickly after the page_cache_release() so
the kernel has not had a chance to reuse the memory for something else,
most of the time.

The current RW driver appears to have followed the design and development
of the earlier read-only hfsplus driver [3], where-by version 0.1 (Dec
2001) had a B-tree node-centric approach to
read_cache_page()/page_cache_release() per bnode_get()/bnode_put(),
migrating towards version 0.2 (June 2002) of caching and releasing pages
per inode extents.  When the current RW code first entered the kernel [2]
in 2005, there was an REF_PAGES conditional (and "//" commented out code)
to switch between B-node centric paging to inode-centric paging.  There
was a mistake with the direction of one of the REF_PAGES conditionals in
__hfs_bnode_create().  In a subsequent "remove debug code" commit [4], the
read_cache_page()/page_cache_release() per bnode_get()/bnode_put() were
removed, but a page_cache_release() was mistakenly left in (propagating
the "REF_PAGES <-> !REF_PAGE" mistake), and the commented-out
page_cache_release() in bnode_release() (which should be spanned by
!REF_PAGES) was never enabled.

References:
[1]:
Michael Fox, Apr 2013
http://www.spinics.net/lists/linux-fsdevel/msg63807.html
("hfsplus volume suddenly inaccessable after 'hfs: recoff %d too large'")

Sasha Levin, Feb 2015
http://lkml.org/lkml/2015/2/20/85 ("use after free")

https://bugs.launchpad.net/ubuntu/+source/linux/+bug/740814
https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1027887
https://bugzilla.kernel.org/show_bug.cgi?id=42342
https://bugzilla.kernel.org/show_bug.cgi?id=63841
https://bugzilla.kernel.org/show_bug.cgi?id=78761

[2]:
http://git.kernel.org/cgit/linux/kernel/git/tglx/history.git/commit/\
fs/hfs/bnode.c?id=d1081202f1d0ee35ab0beb490da4b65d4bc763db
commit d1081202f1d0ee35ab0beb490da4b65d4bc763db
Author: Andrew Morton <akpm@osdl.org>
Date:   Wed Feb 25 16:17:36 2004 -0800

    [PATCH] HFS rewrite

http://git.kernel.org/cgit/linux/kernel/git/tglx/history.git/commit/\
fs/hfsplus/bnode.c?id=91556682e0bf004d98a529bf829d339abb98bbbd

commit 91556682e0bf004d98a529bf829d339abb98bbbd
Author: Andrew Morton <akpm@osdl.org>
Date:   Wed Feb 25 16:17:48 2004 -0800

    [PATCH] HFS+ support

[3]:
http://sourceforge.net/projects/linux-hfsplus/

http://sourceforge.net/projects/linux-hfsplus/files/Linux%202.4.x%20patch/hfsplus%200.1/
http://sourceforge.net/projects/linux-hfsplus/files/Linux%202.4.x%20patch/hfsplus%200.2/

http://linux-hfsplus.cvs.sourceforge.net/viewvc/linux-hfsplus/linux/\
fs/hfsplus/bnode.c?r1=1.4&r2=1.5

Date:   Thu Jun 6 09:45:14 2002 +0000
Use buffer cache instead of page cache in bnode.c. Cache inode extents.

[4]:
http://git.kernel.org/cgit/linux/kernel/git/\
stable/linux-stable.git/commit/?id=a5e3985fa014029eb6795664c704953720cc7f7d

commit a5e3985fa014029eb6795664c704953720cc7f7d
Author: Roman Zippel <zippel@linux-m68k.org>
Date:   Tue Sep 6 15:18:47 2005 -0700

[PATCH] hfs: remove debug code

Signed-off-by: Hin-Tak Leung <htl10@users.sourceforge.net>
Signed-off-by: Sergei Antonov <saproj@gmail.com>
Reviewed-by: Anton Altaparmakov <anton@tuxera.com>
Reported-by: Sasha Levin <sasha.levin@oracle.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Vyacheslav Dubeyko <slava@dubeyko.com>
Cc: Sougata Santra <sougata@tuxera.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agofs/coda: fix readlink buffer overflow
Jan Harkes [Wed, 9 Sep 2015 22:38:01 +0000 (15:38 -0700)]
fs/coda: fix readlink buffer overflow

Dan Carpenter discovered a buffer overflow in the Coda file system
readlink code.  A userspace file system daemon can return a 4096 byte
result which then triggers a one byte write past the allocated readlink
result buffer.

This does not trigger with an unmodified Coda implementation because Coda
has a 1024 byte limit for symbolic links, however other userspace file
systems using the Coda kernel module could be affected.

Although this is an obvious overflow, I don't think this has to be handled
as too sensitive from a security perspective because the overflow is on
the Coda userspace daemon side which already needs root to open Coda's
kernel device and to mount the file system before we get to the point that
links can be read.

[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Jan Harkes <jaharkes@cs.cmu.edu>
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agocheckpatch: add constant comparison on left side test
Joe Perches [Wed, 9 Sep 2015 22:37:58 +0000 (15:37 -0700)]
checkpatch: add constant comparison on left side test

"CONST <comparison> variable" checks like:

        if (NULL != foo)
and
        while (0 < bar(...))

where a constant (or what appears to be a constant like an upper case
identifier) is on the left of a comparison are generally preferred to be
written using the constant on the right side like:

        if (foo != NULL)
and
        while (bar(...) > 0)

Add a test for this.

Add a --fix option too, but only do it when the code is immediately
surrounded by parentheses to avoid misfixing things like "(0 < bar() +
constant)"

Signed-off-by: Joe Perches <joe@perches.com>
Cc: Nicolas Morey Chaisemartin <nmorey@kalray.eu>
Cc: Viresh Kumar <viresh.kumar@linaro.org>
Cc: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agocheckpatch: add __pmem to $Sparse annotations
Joe Perches [Wed, 9 Sep 2015 22:37:55 +0000 (15:37 -0700)]
checkpatch: add __pmem to $Sparse annotations

commit 61031952f4c8 ("arch, x86: pmem api for ensuring durability of
persistent memory updates") added a new __pmem annotation for sparse
verification.  Add __pmem to the $Sparse variable so checkpatch can
appropriately ignore uses of this attribute too.

Signed-off-by: Joe Perches <joe@perches.com>
Reviewed-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Acked-by: Andy Whitcroft <apw@canonical.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agocheckpatch: fix left brace warning
Eddie Kovsky [Wed, 9 Sep 2015 22:37:52 +0000 (15:37 -0700)]
checkpatch: fix left brace warning

Using checkpatch.pl with Perl 5.22.0 generates the following warning:

    Unescaped left brace in regex is deprecated, passed through in regex;

This patch fixes the warnings by escaping occurrences of the left brace
inside the regular expression.

Signed-off-by: Eddie Kovsky <ewk@edkovsky.org>
Cc: Joe Perches <joe@perches.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agocheckpatch: avoid some commit message long line warnings
Joe Perches [Wed, 9 Sep 2015 22:37:50 +0000 (15:37 -0700)]
checkpatch: avoid some commit message long line warnings

Fixes: and Link: lines may exceed 75 chars in the commit log.
So too can stack dump and dmesg lines and lines that seem
like filenames.

And Fixes: lines don't need to have a "commit" prefix before the
commit id.

Add exceptions for these types of lines.

Signed-off-by: Joe Perches <joe@perches.com>
Reported-by: Paul Bolle <pebolle@tiscali.nl>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agocheckpatch: emit an error on formats with 0x%<decimal>
Joe Perches [Wed, 9 Sep 2015 22:37:47 +0000 (15:37 -0700)]
checkpatch: emit an error on formats with 0x%<decimal>

Using 0x%d is wrong.  Emit a message when it happens.

Miscellanea:

Improve the %Lu warning to match formats like %16Lu.

Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agocheckpatch: make --strict the default for drivers/staging files and patches
Joe Perches [Wed, 9 Sep 2015 22:37:44 +0000 (15:37 -0700)]
checkpatch: make --strict the default for drivers/staging files and patches

Making --strict the default for staging may help some people submit
patches without obvious defects.

Signed-off-by: Joe Perches <joe@perches.com>
Cc: Dan Carpenter <dan.carpenter@oracle.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agocheckpatch: always check block comment styles
Joe Perches [Wed, 9 Sep 2015 22:37:41 +0000 (15:37 -0700)]
checkpatch: always check block comment styles

Some of the block comment tests that are used only for networking are
appropriate for all patches.

For example, these styles are not encouraged:

/*
 block comment without introductory *
*/
and
/*
 * block comment with line terminating */

Remove the networking specific test and add comments.

There are some infrequent false positives where code is lazily
commented out using /* and */ rather than using #if 0/#endif blocks
like:
/* case foo:
case bar: */
case baz:

Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agocheckpatch: report the right line # when using --emacs and --file
Joe Perches [Wed, 9 Sep 2015 22:37:39 +0000 (15:37 -0700)]
checkpatch: report the right line # when using --emacs and --file

commit 34d8815f9512 ("checkpatch: add --showfile to allow input via pipe
to show filenames") broke the --emacs with --file option.

Fix it.

Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agocheckpatch: add some <foo>_destroy functions to NEEDLESS_IF tests
Joe Perches [Wed, 9 Sep 2015 22:37:36 +0000 (15:37 -0700)]
checkpatch: add some <foo>_destroy functions to NEEDLESS_IF tests

Sergey Senozhatsky has modified several destroy functions that can
now be called with NULL values.

 - kmem_cache_destroy()
 - mempool_destroy()
 - dma_pool_destroy()

Update checkpatch to warn when those functions are preceded by an if.

Update checkpatch to --fix all the calls too only when the code style
form is using leading tabs.

from:
if (foo)
<func>(foo);
to:
<func>(foo);

Signed-off-by: Joe Perches <joe@perches.com>
Tested-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Julia Lawall <julia.lawall@lip6.fr>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agocheckpatch: Allow longer declaration macros
Joe Perches [Wed, 9 Sep 2015 22:37:33 +0000 (15:37 -0700)]
checkpatch: Allow longer declaration macros

Some really long declaration macros exist.

For instance;
   DEFINE_DMA_BUF_EXPORT_INFO(exp_info);
and
DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(name, description)

Increase the limit from 2 words to 6 after DECLARE/DEFINE uses.

Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agocheckpatch: improve SUSPECT_CODE_INDENT test
Joe Perches [Wed, 9 Sep 2015 22:37:30 +0000 (15:37 -0700)]
checkpatch: improve SUSPECT_CODE_INDENT test

Many lines exist like

if (foo)
bar;

where the tabbed indentation of the branch is not one more than the "if"
line above it.

checkpatch should emit a warning on those lines.

Miscellenea:

o Remove comments from branch blocks
o Skip blank lines in block

Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agocheckpatch: add warning on BUG/BUG_ON use
Joe Perches [Wed, 9 Sep 2015 22:37:27 +0000 (15:37 -0700)]
checkpatch: add warning on BUG/BUG_ON use

Using BUG/BUG_ON crashes the kernel and is just unfriendly.

Enable code that emits a warning on BUG/BUG_ON use.

Make the code emit the message at WARNING level when scanning a patch and
at CHECK level when scanning files so that script users don't feel an
obligation to fix code that might be above their pay grade.

Signed-off-by: Joe Perches <joe@perches.com>
Reported-by: Geert Uytterhoeven <geert@linux-m68k.org>
Tested-by: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agocheckpatch: warn on bare SHA-1 commit IDs in commit logs
Joe Perches [Wed, 9 Sep 2015 22:37:25 +0000 (15:37 -0700)]
checkpatch: warn on bare SHA-1 commit IDs in commit logs

Commit IDs should have commit descriptions too.  Warn when a 12 to 40 byte
SHA-1 is used in commit logs.

Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agolib/test_kasan.c: make kmalloc_oob_krealloc_less more correctly
Wang Long [Wed, 9 Sep 2015 22:37:22 +0000 (15:37 -0700)]
lib/test_kasan.c: make kmalloc_oob_krealloc_less more correctly

In kmalloc_oob_krealloc_less, I think it is better to test
the size2 boundary.

If we do not call krealloc, the access of position size1 will still cause
out-of-bounds and access of position size2 does not.  After call krealloc,
the access of position size2 cause out-of-bounds.  So using size2 is more
correct.

Signed-off-by: Wang Long <long.wanglong@huawei.com>
Cc: Andrey Ryabinin <ryabinin.a.a@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agolib/test_kasan.c: fix a typo
Wang Long [Wed, 9 Sep 2015 22:37:19 +0000 (15:37 -0700)]
lib/test_kasan.c: fix a typo

Signed-off-by: Wang Long <long.wanglong@huawei.com>
Cc: Andrey Ryabinin <ryabinin.a.a@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agolib/string_helpers: rename "esc" arg to "only"
Kees Cook [Wed, 9 Sep 2015 22:37:16 +0000 (15:37 -0700)]
lib/string_helpers: rename "esc" arg to "only"

To further clarify the purpose of the "esc" argument, rename it to "only"
to reflect that it is a limit, not a list of additional characters to
escape.

Signed-off-by: Kees Cook <keescook@chromium.org>
Suggested-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agolib/string_helpers: clarify esc arg in string_escape_mem
Kees Cook [Wed, 9 Sep 2015 22:37:14 +0000 (15:37 -0700)]
lib/string_helpers: clarify esc arg in string_escape_mem

The esc argument is used to reduce which characters will be escaped.  For
example, using " " with ESCAPE_SPACE will not produce any escaped spaces.

Signed-off-by: Kees Cook <keescook@chromium.org>
Cc: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Cc: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Cc: Mathias Krause <minipli@googlemail.com>
Cc: James Bottomley <James.Bottomley@HansenPartnership.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agohexdump: do not print debug dumps for !CONFIG_DEBUG
Linus Walleij [Wed, 9 Sep 2015 22:37:11 +0000 (15:37 -0700)]
hexdump: do not print debug dumps for !CONFIG_DEBUG

print_hex_dump_debug() is likely supposed to be analogous to pr_debug() or
dev_dbg() & friends.  Currently it will adhere to dynamic debug, but will
not stub out prints if CONFIG_DEBUG is not set.  Let's make it do the
right thing, because I am tired of having my dmesg buffer full of hex
dumps on production systems.

Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agolib/bitmap.c: bitmap_parselist can accept string with whitespaces on head or tail
Pan Xinhui [Wed, 9 Sep 2015 22:37:08 +0000 (15:37 -0700)]
lib/bitmap.c: bitmap_parselist can accept string with whitespaces on head or tail

In __bitmap_parselist we can accept whitespaces on head or tail during
every parsing procedure.  If input has valid ranges, there is no reason to
reject the user.

For example, bitmap_parselist(" 1-3, 5, ", &mask, nmaskbits).  After
separating the string, we get " 1-3", " 5", and " ".  It's possible and
reasonable to accept such string as long as the parsing result is correct.

Signed-off-by: Pan Xinhui <xinhuix.pan@intel.com>
Cc: Yury Norov <yury.norov@gmail.com>
Cc: Chris Metcalf <cmetcalf@ezchip.com>
Cc: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Cc: Sudeep Holla <sudeep.holla@arm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agolib/bitmap.c: fix a special string handling bug in __bitmap_parselist
Pan Xinhui [Wed, 9 Sep 2015 22:37:05 +0000 (15:37 -0700)]
lib/bitmap.c: fix a special string handling bug in __bitmap_parselist

If string end with '-', for exapmle, bitmap_parselist("1,0-",&mask,
nmaskbits), It is not in a valid pattern, so add a check after loop.
Return -EINVAL on such condition.

Signed-off-by: Pan Xinhui <xinhuix.pan@intel.com>
Cc: Yury Norov <yury.norov@gmail.com>
Cc: Chris Metcalf <cmetcalf@ezchip.com>
Cc: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Cc: Sudeep Holla <sudeep.holla@arm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agolib/bitmap.c: correct a code style and do some, optimization
Pan Xinhui [Wed, 9 Sep 2015 22:37:02 +0000 (15:37 -0700)]
lib/bitmap.c: correct a code style and do some, optimization

We can avoid in-loop incrementation of ndigits.  Save current totaldigits
to ndigits before loop, and check ndigits against totaldigits after the
loop.

Signed-off-by: Pan Xinhui <xinhuix.pan@intel.com>
Cc: Yury Norov <yury.norov@gmail.com>
Cc: Chris Metcalf <cmetcalf@ezchip.com>
Cc: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Cc: Sudeep Holla <sudeep.holla@arm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agoproc: convert to kstrto*()/kstrto*_from_user()
Alexey Dobriyan [Wed, 9 Sep 2015 22:36:59 +0000 (15:36 -0700)]
proc: convert to kstrto*()/kstrto*_from_user()

Convert from manual allocation/copy_from_user/...  to kstrto*() family
which were designed for exactly that.

One case can not be converted to kstrto*_from_user() to make code even
more simpler because of whitespace stripping, oh well...

Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agokstrto*: accept "-0" for signed conversion
Alexey Dobriyan [Wed, 9 Sep 2015 22:36:17 +0000 (15:36 -0700)]
kstrto*: accept "-0" for signed conversion

strtol(3) et al accept "-0", so should we.

Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Cc: David Howells <dhowells@redhat.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Mark Fasheh <mfasheh@suse.com>
Cc: Theodore Ts'o <tytso@mit.edu>
Cc: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agoMAINTAINERS/CREDITS: mark MaxRAID as Orphan, move Anil Ravindranath to CREDITS
Joe Perches [Wed, 9 Sep 2015 22:36:14 +0000 (15:36 -0700)]
MAINTAINERS/CREDITS: mark MaxRAID as Orphan, move Anil Ravindranath to CREDITS

Anil's email address bounces and he hasn't had a signoff
in over 5 years.

Signed-off-by: Joe Perches <joe@perches.com>
Cc: James Bottomley <James.Bottomley@HansenPartnership.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agoinclude/linux/printk.h: include pr_fmt in pr_debug_ratelimited
Jason A. Donenfeld [Wed, 9 Sep 2015 22:36:12 +0000 (15:36 -0700)]
include/linux/printk.h: include pr_fmt in pr_debug_ratelimited

The other two implementations of pr_debug_ratelimited include pr_fmt,
along with every other pr_* function.  But pr_debug_ratelimited forgot to
add it with the CONFIG_DYNAMIC_DEBUG implementation.

This patch unifies the behavior.

Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agokernel/cred.c: remove unnecessary kdebug atomic reads
Joe Perches [Wed, 9 Sep 2015 22:36:09 +0000 (15:36 -0700)]
kernel/cred.c: remove unnecessary kdebug atomic reads

Commit e0e817392b9a ("CRED: Add some configurable debugging [try #6]")
added the kdebug mechanism to this file back in 2009.

The kdebug macro calls no_printk which always evaluates arguments.

Most of the kdebug uses have an unnecessary call of
atomic_read(&cred->usage)

Make the kdebug macro do nothing by defining it with
do { if (0) no_printk(...); } while (0)
when not enabled.

$ size kernel/cred.o* (defconfig x86-64)
   text    data     bss     dec     hex filename
   2748     336       8    3092     c14 kernel/cred.o.new
   2788     336       8    3132     c3c kernel/cred.o.old

Miscellanea:
o Neaten the #define kdebug macros while there

Signed-off-by: Joe Perches <joe@perches.com>
Cc: David Howells <dhowells@redhat.com>
Cc: James Morris <jmorris@namei.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agokernel/extable.c: remove duplicated include
Wei Yongjun [Wed, 9 Sep 2015 22:36:06 +0000 (15:36 -0700)]
kernel/extable.c: remove duplicated include

Signed-off-by: Wei Yongjun <yongjun_wei@trendmicro.com.cn>
Acked-by: Steven Rostedt <rostedt@goodmis.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agoinclude/linux/poison.h: remove not-used poison pointer macros
Vasily Kulikov [Wed, 9 Sep 2015 22:36:03 +0000 (15:36 -0700)]
include/linux/poison.h: remove not-used poison pointer macros

Signed-off-by: Vasily Kulikov <segoon@openwall.com>
Cc: Solar Designer <solar@openwall.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agoinclude/linux/poison.h: fix LIST_POISON{1,2} offset
Vasily Kulikov [Wed, 9 Sep 2015 22:36:00 +0000 (15:36 -0700)]
include/linux/poison.h: fix LIST_POISON{1,2} offset

Poison pointer values should be small enough to find a room in
non-mmap'able/hardly-mmap'able space.  E.g.  on x86 "poison pointer space"
is located starting from 0x0.  Given unprivileged users cannot mmap
anything below mmap_min_addr, it should be safe to use poison pointers
lower than mmap_min_addr.

The current poison pointer values of LIST_POISON{1,2} might be too big for
mmap_min_addr values equal or less than 1 MB (common case, e.g.  Ubuntu
uses only 0x10000).  There is little point to use such a big value given
the "poison pointer space" below 1 MB is not yet exhausted.  Changing it
to a smaller value solves the problem for small mmap_min_addr setups.

The values are suggested by Solar Designer:
http://www.openwall.com/lists/oss-security/2015/05/02/6

Signed-off-by: Vasily Kulikov <segoon@openwall.com>
Cc: Solar Designer <solar@openwall.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agoproc: change proc_subdir_lock to a rwlock
Waiman Long [Wed, 9 Sep 2015 22:35:57 +0000 (15:35 -0700)]
proc: change proc_subdir_lock to a rwlock

The proc_subdir_lock spinlock is used to allow only one task to make
change to the proc directory structure as well as looking up information
in it.  However, the information lookup part can actually be entered by
more than one task as the pde_get() and pde_put() reference count update
calls in the critical sections are atomic increment and decrement
respectively and so are safe with concurrent updates.

The x86 architecture has already used qrwlock which is fair and other
architectures like ARM are in the process of switching to qrwlock.  So
unfairness shouldn't be a concern in that conversion.

This patch changed the proc_subdir_lock to a rwlock in order to enable
concurrent lookup. The following functions were modified to take a
write lock:
 - proc_register()
 - remove_proc_entry()
 - remove_proc_subtree()

The following functions were modified to take a read lock:
 - xlate_proc_name()
 - proc_lookup_de()
 - proc_readdir_de()

A parallel /proc filesystem search with the "find" command (1000 threads)
was run on a 4-socket Haswell-EX box (144 threads).  Before the patch, the
parallel search took about 39s.  After the patch, the parallel find took
only 25s, a saving of about 14s.

The micro-benchmark that I used was artificial, but it was used to
reproduce an exit hanging problem that I saw in real application.  In
fact, only allow one task to do a lookup seems too limiting to me.

Signed-off-by: Waiman Long <Waiman.Long@hp.com>
Acked-by: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Alexey Dobriyan <adobriyan@gmail.com>
Cc: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Scott J Norton <scott.norton@hp.com>
Cc: Douglas Hatch <doug.hatch@hp.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agoprocfs: always expose /proc/<pid>/map_files/ and make it readable
Calvin Owens [Wed, 9 Sep 2015 22:35:54 +0000 (15:35 -0700)]
procfs: always expose /proc/<pid>/map_files/ and make it readable

Currently, /proc/<pid>/map_files/ is restricted to CAP_SYS_ADMIN, and is
only exposed if CONFIG_CHECKPOINT_RESTORE is set.

Each mapped file region gets a symlink in /proc/<pid>/map_files/
corresponding to the virtual address range at which it is mapped.  The
symlinks work like the symlinks in /proc/<pid>/fd/, so you can follow them
to the backing file even if that backing file has been unlinked.

Currently, files which are mapped, unlinked, and closed are impossible to
stat() from userspace.  Exposing /proc/<pid>/map_files/ closes this
functionality "hole".

Not being able to stat() such files makes noticing and explicitly
accounting for the space they use on the filesystem impossible.  You can
work around this by summing up the space used by every file in the
filesystem and subtracting that total from what statfs() tells you, but
that obviously isn't great, and it becomes unworkable once your filesystem
becomes large enough.

This patch moves map_files/ out from behind CONFIG_CHECKPOINT_RESTORE, and
adjusts the permissions enforced on it as follows:

* proc_map_files_lookup()
* proc_map_files_readdir()
* map_files_d_revalidate()

Remove the CAP_SYS_ADMIN restriction, leaving only the current
restriction requiring PTRACE_MODE_READ. The information made
available to userspace by these three functions is already
available in /proc/PID/maps with MODE_READ, so I don't see any
reason to limit them any further (see below for more detail).

* proc_map_files_follow_link()

This stub has been added, and requires that the user have
CAP_SYS_ADMIN in order to follow the links in map_files/,
since there was concern on LKML both about the potential for
bypassing permissions on ancestor directories in the path to
files pointed to, and about what happens with more exotic
memory mappings created by some drivers (ie dma-buf).

In older versions of this patch, I changed every permission check in
the four functions above to enforce MODE_ATTACH instead of MODE_READ.
This was an oversight on my part, and after revisiting the discussion
it seems that nobody was concerned about anything outside of what is
made possible by ->follow_link(). So in this version, I've left the
checks for PTRACE_MODE_READ as-is.

[akpm@linux-foundation.org: catch up with concurrent proc_pid_follow_link() changes]
Signed-off-by: Calvin Owens <calvinowens@fb.com>
Reviewed-by: Kees Cook <keescook@chromium.org>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Cyrill Gorcunov <gorcunov@openvz.org>
Cc: Joe Perches <joe@perches.com>
Cc: Kirill A. Shutemov <kirill@shutemov.name>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agoproc: add cond_resched to /proc/kpage* read/write loop
Vladimir Davydov [Wed, 9 Sep 2015 22:35:51 +0000 (15:35 -0700)]
proc: add cond_resched to /proc/kpage* read/write loop

Reading/writing a /proc/kpage* file may take long on machines with a lot
of RAM installed.

Signed-off-by: Vladimir Davydov <vdavydov@parallels.com>
Suggested-by: Andres Lagar-Cavilla <andreslc@google.com>
Reviewed-by: Andres Lagar-Cavilla <andreslc@google.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Raghavendra K T <raghavendra.kt@linux.vnet.ibm.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Greg Thelen <gthelen@google.com>
Cc: Michel Lespinasse <walken@google.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Pavel Emelyanov <xemul@parallels.com>
Cc: Cyrill Gorcunov <gorcunov@openvz.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agoproc: export idle flag via kpageflags
Vladimir Davydov [Wed, 9 Sep 2015 22:35:48 +0000 (15:35 -0700)]
proc: export idle flag via kpageflags

As noted by Minchan, a benefit of reading idle flag from /proc/kpageflags
is that one can easily filter dirty and/or unevictable pages while
estimating the size of unused memory.

Note that idle flag read from /proc/kpageflags may be stale in case the
page was accessed via a PTE, because it would be too costly to iterate
over all page mappings on each /proc/kpageflags read to provide an
up-to-date value.  To make sure the flag is up-to-date one has to read
/sys/kernel/mm/page_idle/bitmap first.

Signed-off-by: Vladimir Davydov <vdavydov@parallels.com>
Reviewed-by: Andres Lagar-Cavilla <andreslc@google.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Raghavendra K T <raghavendra.kt@linux.vnet.ibm.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Greg Thelen <gthelen@google.com>
Cc: Michel Lespinasse <walken@google.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Pavel Emelyanov <xemul@parallels.com>
Cc: Cyrill Gorcunov <gorcunov@openvz.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm: introduce idle page tracking
Vladimir Davydov [Wed, 9 Sep 2015 22:35:45 +0000 (15:35 -0700)]
mm: introduce idle page tracking

Knowing the portion of memory that is not used by a certain application or
memory cgroup (idle memory) can be useful for partitioning the system
efficiently, e.g.  by setting memory cgroup limits appropriately.
Currently, the only means to estimate the amount of idle memory provided
by the kernel is /proc/PID/{clear_refs,smaps}: the user can clear the
access bit for all pages mapped to a particular process by writing 1 to
clear_refs, wait for some time, and then count smaps:Referenced.  However,
this method has two serious shortcomings:

 - it does not count unmapped file pages
 - it affects the reclaimer logic

To overcome these drawbacks, this patch introduces two new page flags,
Idle and Young, and a new sysfs file, /sys/kernel/mm/page_idle/bitmap.
A page's Idle flag can only be set from userspace by setting bit in
/sys/kernel/mm/page_idle/bitmap at the offset corresponding to the page,
and it is cleared whenever the page is accessed either through page tables
(it is cleared in page_referenced() in this case) or using the read(2)
system call (mark_page_accessed()). Thus by setting the Idle flag for
pages of a particular workload, which can be found e.g.  by reading
/proc/PID/pagemap, waiting for some time to let the workload access its
working set, and then reading the bitmap file, one can estimate the amount
of pages that are not used by the workload.

The Young page flag is used to avoid interference with the memory
reclaimer.  A page's Young flag is set whenever the Access bit of a page
table entry pointing to the page is cleared by writing to the bitmap file.
If page_referenced() is called on a Young page, it will add 1 to its
return value, therefore concealing the fact that the Access bit was
cleared.

Note, since there is no room for extra page flags on 32 bit, this feature
uses extended page flags when compiled on 32 bit.

[akpm@linux-foundation.org: fix build]
[akpm@linux-foundation.org: kpageidle requires an MMU]
[akpm@linux-foundation.org: decouple from page-flags rework]
Signed-off-by: Vladimir Davydov <vdavydov@parallels.com>
Reviewed-by: Andres Lagar-Cavilla <andreslc@google.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Raghavendra K T <raghavendra.kt@linux.vnet.ibm.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Greg Thelen <gthelen@google.com>
Cc: Michel Lespinasse <walken@google.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Pavel Emelyanov <xemul@parallels.com>
Cc: Cyrill Gorcunov <gorcunov@openvz.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agommu-notifier: add clear_young callback
Vladimir Davydov [Wed, 9 Sep 2015 22:35:41 +0000 (15:35 -0700)]
mmu-notifier: add clear_young callback

In the scope of the idle memory tracking feature, which is introduced by
the following patch, we need to clear the referenced/accessed bit not only
in primary, but also in secondary ptes.  The latter is required in order
to estimate wss of KVM VMs.  At the same time we want to avoid flushing
tlb, because it is quite expensive and it won't really affect the final
result.

Currently, there is no function for clearing pte young bit that would meet
our requirements, so this patch introduces one.  To achieve that we have
to add a new mmu-notifier callback, clear_young, since there is no method
for testing-and-clearing a secondary pte w/o flushing tlb.  The new method
is not mandatory and currently only implemented by KVM.

Signed-off-by: Vladimir Davydov <vdavydov@parallels.com>
Reviewed-by: Andres Lagar-Cavilla <andreslc@google.com>
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Raghavendra K T <raghavendra.kt@linux.vnet.ibm.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Greg Thelen <gthelen@google.com>
Cc: Michel Lespinasse <walken@google.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Pavel Emelyanov <xemul@parallels.com>
Cc: Cyrill Gorcunov <gorcunov@openvz.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agoproc: add kpagecgroup file
Vladimir Davydov [Wed, 9 Sep 2015 22:35:38 +0000 (15:35 -0700)]
proc: add kpagecgroup file

/proc/kpagecgroup contains a 64-bit inode number of the memory cgroup each
page is charged to, indexed by PFN.  Having this information is useful for
estimating a cgroup working set size.

The file is present if CONFIG_PROC_PAGE_MONITOR && CONFIG_MEMCG.

Signed-off-by: Vladimir Davydov <vdavydov@parallels.com>
Reviewed-by: Andres Lagar-Cavilla <andreslc@google.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Raghavendra K T <raghavendra.kt@linux.vnet.ibm.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Greg Thelen <gthelen@google.com>
Cc: Michel Lespinasse <walken@google.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Pavel Emelyanov <xemul@parallels.com>
Cc: Cyrill Gorcunov <gorcunov@openvz.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomemcg: zap try_get_mem_cgroup_from_page
Vladimir Davydov [Wed, 9 Sep 2015 22:35:35 +0000 (15:35 -0700)]
memcg: zap try_get_mem_cgroup_from_page

It is only used in mem_cgroup_try_charge, so fold it in and zap it.

Signed-off-by: Vladimir Davydov <vdavydov@parallels.com>
Reviewed-by: Andres Lagar-Cavilla <andreslc@google.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Raghavendra K T <raghavendra.kt@linux.vnet.ibm.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Greg Thelen <gthelen@google.com>
Cc: Michel Lespinasse <walken@google.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Pavel Emelyanov <xemul@parallels.com>
Cc: Cyrill Gorcunov <gorcunov@openvz.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agohwpoison: use page_cgroup_ino for filtering by memcg
Vladimir Davydov [Wed, 9 Sep 2015 22:35:31 +0000 (15:35 -0700)]
hwpoison: use page_cgroup_ino for filtering by memcg

Hwpoison allows to filter pages by memory cgroup ino.  Currently, it
calls try_get_mem_cgroup_from_page to obtain the cgroup from a page and
then its ino using cgroup_ino, but now we have a helper method for
that, page_cgroup_ino, so use it instead.

This patch also loosens the hwpoison memcg filter dependency rules - it
makes it depend on CONFIG_MEMCG instead of CONFIG_MEMCG_SWAP, because
hwpoison memcg filter does not require anything (nor it used to) from
CONFIG_MEMCG_SWAP side.

Signed-off-by: Vladimir Davydov <vdavydov@parallels.com>
Reviewed-by: Andres Lagar-Cavilla <andreslc@google.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Raghavendra K T <raghavendra.kt@linux.vnet.ibm.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Greg Thelen <gthelen@google.com>
Cc: Michel Lespinasse <walken@google.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Pavel Emelyanov <xemul@parallels.com>
Cc: Cyrill Gorcunov <gorcunov@openvz.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomemcg: add page_cgroup_ino helper
Vladimir Davydov [Wed, 9 Sep 2015 22:35:28 +0000 (15:35 -0700)]
memcg: add page_cgroup_ino helper

This patchset introduces a new user API for tracking user memory pages
that have not been used for a given period of time.  The purpose of this
is to provide the userspace with the means of tracking a workload's
working set, i.e.  the set of pages that are actively used by the
workload.  Knowing the working set size can be useful for partitioning the
system more efficiently, e.g.  by tuning memory cgroup limits
appropriately, or for job placement within a compute cluster.

==== USE CASES ====

The unified cgroup hierarchy has memory.low and memory.high knobs, which
are defined as the low and high boundaries for the workload working set
size.  However, the working set size of a workload may be unknown or
change in time.  With this patch set, one can periodically estimate the
amount of memory unused by each cgroup and tune their memory.low and
memory.high parameters accordingly, therefore optimizing the overall
memory utilization.

Another use case is balancing workloads within a compute cluster.  Knowing
how much memory is not really used by a workload unit may help take a more
optimal decision when considering migrating the unit to another node
within the cluster.

Also, as noted by Minchan, this would be useful for per-process reclaim
(https://lwn.net/Articles/545668/). With idle tracking, we could reclaim idle
pages only by smart user memory manager.

==== USER API ====

The user API consists of two new files:

 * /sys/kernel/mm/page_idle/bitmap.  This file implements a bitmap where each
   bit corresponds to a page, indexed by PFN. When the bit is set, the
   corresponding page is idle. A page is considered idle if it has not been
   accessed since it was marked idle. To mark a page idle one should set the
   bit corresponding to the page by writing to the file. A value written to the
   file is OR-ed with the current bitmap value. Only user memory pages can be
   marked idle, for other page types input is silently ignored. Writing to this
   file beyond max PFN results in the ENXIO error. Only available when
   CONFIG_IDLE_PAGE_TRACKING is set.

   This file can be used to estimate the amount of pages that are not
   used by a particular workload as follows:

   1. mark all pages of interest idle by setting corresponding bits in the
      /sys/kernel/mm/page_idle/bitmap
   2. wait until the workload accesses its working set
   3. read /sys/kernel/mm/page_idle/bitmap and count the number of bits set

 * /proc/kpagecgroup.  This file contains a 64-bit inode number of the
   memory cgroup each page is charged to, indexed by PFN. Only available when
   CONFIG_MEMCG is set.

   This file can be used to find all pages (including unmapped file pages)
   accounted to a particular cgroup. Using /sys/kernel/mm/page_idle/bitmap, one
   can then estimate the cgroup working set size.

For an example of using these files for estimating the amount of unused
memory pages per each memory cgroup, please see the script attached
below.

==== REASONING ====

The reason to introduce the new user API instead of using
/proc/PID/{clear_refs,smaps} is that the latter has two serious
drawbacks:

 - it does not count unmapped file pages
 - it affects the reclaimer logic

The new API attempts to overcome them both. For more details on how it
is achieved, please see the comment to patch 6.

==== PATCHSET STRUCTURE ====

The patch set is organized as follows:

 - patch 1 adds page_cgroup_ino() helper for the sake of
   /proc/kpagecgroup and patches 2-3 do related cleanup
 - patch 4 adds /proc/kpagecgroup, which reports cgroup ino each page is
   charged to
 - patch 5 introduces a new mmu notifier callback, clear_young, which is
   a lightweight version of clear_flush_young; it is used in patch 6
 - patch 6 implements the idle page tracking feature, including the
   userspace API, /sys/kernel/mm/page_idle/bitmap
 - patch 7 exports idle flag via /proc/kpageflags

==== SIMILAR WORKS ====

Originally, the patch for tracking idle memory was proposed back in 2011
by Michel Lespinasse (see http://lwn.net/Articles/459269/).  The main
difference between Michel's patch and this one is that Michel implemented
a kernel space daemon for estimating idle memory size per cgroup while
this patch only provides the userspace with the minimal API for doing the
job, leaving the rest up to the userspace.  However, they both share the
same idea of Idle/Young page flags to avoid affecting the reclaimer logic.

==== PERFORMANCE EVALUATION ====

SPECjvm2008 (https://www.spec.org/jvm2008/) was used to evaluate the
performance impact introduced by this patch set.  Three runs were carried
out:

 - base: kernel without the patch
 - patched: patched kernel, the feature is not used
 - patched-active: patched kernel, 1 minute-period daemon is used for
   tracking idle memory

For tracking idle memory, idlememstat utility was used:
https://github.com/locker/idlememstat

testcase            base            patched        patched-active

compiler       537.40 ( 0.00)%   532.26 (-0.96)%   538.31 ( 0.17)%
compress       305.47 ( 0.00)%   301.08 (-1.44)%   300.71 (-1.56)%
crypto         284.32 ( 0.00)%   282.21 (-0.74)%   284.87 ( 0.19)%
derby          411.05 ( 0.00)%   413.44 ( 0.58)%   412.07 ( 0.25)%
mpegaudio      189.96 ( 0.00)%   190.87 ( 0.48)%   189.42 (-0.28)%
scimark.large   46.85 ( 0.00)%    46.41 (-0.94)%    47.83 ( 2.09)%
scimark.small  412.91 ( 0.00)%   415.41 ( 0.61)%   421.17 ( 2.00)%
serial         204.23 ( 0.00)%   213.46 ( 4.52)%   203.17 (-0.52)%
startup         36.76 ( 0.00)%    35.49 (-3.45)%    35.64 (-3.05)%
sunflow        115.34 ( 0.00)%   115.08 (-0.23)%   117.37 ( 1.76)%
xml            620.55 ( 0.00)%   619.95 (-0.10)%   620.39 (-0.03)%

composite      211.50 ( 0.00)%   211.15 (-0.17)%   211.67 ( 0.08)%

time idlememstat:

17.20user 65.16system 2:15:23elapsed 1%CPU (0avgtext+0avgdata 8476maxresident)k
448inputs+40outputs (1major+36052minor)pagefaults 0swaps

==== SCRIPT FOR COUNTING IDLE PAGES PER CGROUP ====
#! /usr/bin/python
#

import os
import stat
import errno
import struct

CGROUP_MOUNT = "/sys/fs/cgroup/memory"
BUFSIZE = 8 * 1024  # must be multiple of 8

def get_hugepage_size():
    with open("/proc/meminfo", "r") as f:
        for s in f:
            k, v = s.split(":")
            if k == "Hugepagesize":
                return int(v.split()[0]) * 1024

PAGE_SIZE = os.sysconf("SC_PAGE_SIZE")
HUGEPAGE_SIZE = get_hugepage_size()

def set_idle():
    f = open("/sys/kernel/mm/page_idle/bitmap", "wb", BUFSIZE)
    while True:
        try:
            f.write(struct.pack("Q", pow(2, 64) - 1))
        except IOError as err:
            if err.errno == errno.ENXIO:
                break
            raise
    f.close()

def count_idle():
    f_flags = open("/proc/kpageflags", "rb", BUFSIZE)
    f_cgroup = open("/proc/kpagecgroup", "rb", BUFSIZE)

    with open("/sys/kernel/mm/page_idle/bitmap", "rb", BUFSIZE) as f:
        while f.read(BUFSIZE): pass  # update idle flag

    idlememsz = {}
    while True:
        s1, s2 = f_flags.read(8), f_cgroup.read(8)
        if not s1 or not s2:
            break

        flags, = struct.unpack('Q', s1)
        cgino, = struct.unpack('Q', s2)

        unevictable = (flags >> 18) & 1
        huge = (flags >> 22) & 1
        idle = (flags >> 25) & 1

        if idle and not unevictable:
            idlememsz[cgino] = idlememsz.get(cgino, 0) + \
                (HUGEPAGE_SIZE if huge else PAGE_SIZE)

    f_flags.close()
    f_cgroup.close()
    return idlememsz

if __name__ == "__main__":
    print "Setting the idle flag for each page..."
    set_idle()

    raw_input("Wait until the workload accesses its working set, "
              "then press Enter")

    print "Counting idle pages..."
    idlememsz = count_idle()

    for dir, subdirs, files in os.walk(CGROUP_MOUNT):
        ino = os.stat(dir)[stat.ST_INO]
        print dir + ": " + str(idlememsz.get(ino, 0) / 1024) + " kB"
==== END SCRIPT ====

This patch (of 8):

Add page_cgroup_ino() helper to memcg.

This function returns the inode number of the closest online ancestor of
the memory cgroup a page is charged to.  It is required for exporting
information about which page is charged to which cgroup to userspace,
which will be introduced by a following patch.

Signed-off-by: Vladimir Davydov <vdavydov@parallels.com>
Reviewed-by: Andres Lagar-Cavilla <andreslc@google.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Raghavendra K T <raghavendra.kt@linux.vnet.ibm.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Greg Thelen <gthelen@google.com>
Cc: Michel Lespinasse <walken@google.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Pavel Emelyanov <xemul@parallels.com>
Cc: Cyrill Gorcunov <gorcunov@openvz.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agozswap: update docs for runtime-changeable attributes
Dan Streetman [Wed, 9 Sep 2015 22:35:25 +0000 (15:35 -0700)]
zswap: update docs for runtime-changeable attributes

Change the Documentation/vm/zswap.txt doc to indicate that the "zpool" and
"compressor" params are now changeable at runtime.

Signed-off-by: Dan Streetman <ddstreet@ieee.org>
Cc: Seth Jennings <sjennings@variantweb.net>
Cc: Sergey Senozhatsky <sergey.senozhatsky.work@gmail.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>
9 years agozswap: change zpool/compressor at runtime
Dan Streetman [Wed, 9 Sep 2015 22:35:21 +0000 (15:35 -0700)]
zswap: change zpool/compressor at runtime

Update the zpool and compressor parameters to be changeable at runtime.
When changed, a new pool is created with the requested zpool/compressor,
and added as the current pool at the front of the pool list.  Previous
pools remain in the list only to remove existing compressed pages from.
The old pool(s) are removed once they become empty.

Signed-off-by: Dan Streetman <ddstreet@ieee.org>
Acked-by: Seth Jennings <sjennings@variantweb.net>
Cc: Sergey Senozhatsky <sergey.senozhatsky.work@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agozswap: dynamic pool creation
Dan Streetman [Wed, 9 Sep 2015 22:35:19 +0000 (15:35 -0700)]
zswap: dynamic pool creation

Add dynamic creation of pools.  Move the static crypto compression per-cpu
transforms into each pool.  Add a pointer to zswap_entry to the pool it's
in.

This is required by the following patch which enables changing the zswap
zpool and compressor params at runtime.

[akpm@linux-foundation.org: fix merge snafus]
Signed-off-by: Dan Streetman <ddstreet@ieee.org>
Acked-by: Seth Jennings <sjennings@variantweb.net>
Cc: Sergey Senozhatsky <sergey.senozhatsky.work@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agozpool: add zpool_has_pool()
Dan Streetman [Wed, 9 Sep 2015 22:35:16 +0000 (15:35 -0700)]
zpool: add zpool_has_pool()

This series makes creation of the zpool and compressor dynamic, so that
they can be changed at runtime.  This makes using/configuring zswap
easier, as before this zswap had to be configured at boot time, using boot
params.

This uses a single list to track both the zpool and compressor together,
although Seth had mentioned an alternative which is to track the zpools
and compressors using separate lists.  In the most common case, only a
single zpool and single compressor, using one list is slightly simpler
than using two lists, and for the uncommon case of multiple zpools and/or
compressors, using one list is slightly less simple (and uses slightly
more memory, probably) than using two lists.

This patch (of 4):

Add zpool_has_pool() function, indicating if the specified type of zpool
is available (i.e.  zsmalloc or zbud).  This allows checking if a pool is
available, without actually trying to allocate it, similar to
crypto_has_alg().

This is used by a following patch to zswap that enables the dynamic
runtime creation of zswap zpools.

Signed-off-by: Dan Streetman <ddstreet@ieee.org>
Acked-by: Seth Jennings <sjennings@variantweb.net>
Cc: Sergey Senozhatsky <sergey.senozhatsky.work@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agoMerge tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dledford/rdma
Linus Torvalds [Wed, 9 Sep 2015 15:33:31 +0000 (08:33 -0700)]
Merge tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dledford/rdma

Pull inifiniband/rdma updates from Doug Ledford:
 "This is a fairly sizeable set of changes.  I've put them through a
  decent amount of testing prior to sending the pull request due to
  that.

  There are still a few fixups that I know are coming, but I wanted to
  go ahead and get the big, sizable chunk into your hands sooner rather
  than waiting for those last few fixups.

  Of note is the fact that this creates what is intended to be a
  temporary area in the drivers/staging tree specifically for some
  cleanups and additions that are coming for the RDMA stack.  We
  deprecated two drivers (ipath and amso1100) and are waiting to hear
  back if we can deprecate another one (ehca).  We also put Intel's new
  hfi1 driver into this area because it needs to be refactored and a
  transfer library created out of the factored out code, and then it and
  the qib driver and the soft-roce driver should all be modified to use
  that library.

  I expect drivers/staging/rdma to be around for three or four kernel
  releases and then to go away as all of the work is completed and final
  deletions of deprecated drivers are done.

  Summary of changes for 4.3:

   - Create drivers/staging/rdma
   - Move amso1100 driver to staging/rdma and schedule for deletion
   - Move ipath driver to staging/rdma and schedule for deletion
   - Add hfi1 driver to staging/rdma and set TODO for move to regular
     tree
   - Initial support for namespaces to be used on RDMA devices
   - Add RoCE GID table handling to the RDMA core caching code
   - Infrastructure to support handling of devices with differing read
     and write scatter gather capabilities
   - Various iSER updates
   - Kill off unsafe usage of global mr registrations
   - Update SRP driver
   - Misc  mlx4 driver updates
   - Support for the mr_alloc verb
   - Support for a netlink interface between kernel and user space cache
     daemon to speed path record queries and route resolution
   - Ininitial support for safe hot removal of verbs devices"

* tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dledford/rdma: (136 commits)
  IB/ipoib: Suppress warning for send only join failures
  IB/ipoib: Clean up send-only multicast joins
  IB/srp: Fix possible protection fault
  IB/core: Move SM class defines from ib_mad.h to ib_smi.h
  IB/core: Remove unnecessary defines from ib_mad.h
  IB/hfi1: Add PSM2 user space header to header_install
  IB/hfi1: Add CSRs for CONFIG_SDMA_VERBOSITY
  mlx5: Fix incorrect wc pkey_index assignment for GSI messages
  IB/mlx5: avoid destroying a NULL mr in reg_user_mr error flow
  IB/uverbs: reject invalid or unknown opcodes
  IB/cxgb4: Fix if statement in pick_local_ip6adddrs
  IB/sa: Fix rdma netlink message flags
  IB/ucma: HW Device hot-removal support
  IB/mlx4_ib: Disassociate support
  IB/uverbs: Enable device removal when there are active user space applications
  IB/uverbs: Explicitly pass ib_dev to uverbs commands
  IB/uverbs: Fix race between ib_uverbs_open and remove_one
  IB/uverbs: Fix reference counting usage of event files
  IB/core: Make ib_dealloc_pd return void
  IB/srp: Create an insecure all physical rkey only if needed
  ...

9 years agoMerge tag 'for-linus-4.3' of git://git.code.sf.net/p/openipmi/linux-ipmi
Linus Torvalds [Wed, 9 Sep 2015 01:19:17 +0000 (18:19 -0700)]
Merge tag 'for-linus-4.3' of git://git.code.sf.net/p/openipmi/linux-ipmi

Pull IPMI updates from Corey Minyard:
 "Most of these have been sitting in linux-next for more than a release,
  particularly commit 0fbcf4af7c83 ("ipmi: Convert the IPMI SI ACPI
  handling to a platform device") which is probably the most complex
  patch.

  That is also the one that changes drivers/acpi/acpi_pnp.c.  The change
  in that file is only removing IPMI from a "special platform devices"
  list, since I convert it to the standard PNP interface.  I posted this
  one to the ACPI list twice and got no response, and it seems to work
  well in my testing, so I'm hoping it's good.

  Hidehiro Kawai posted a set of changes that improves the panic time
  handling in the IPMI driver.

  The rest of the changes are minor bug fixes or cleanups and some
  documentation"

* tag 'for-linus-4.3' of git://git.code.sf.net/p/openipmi/linux-ipmi:
  ipmi:ssif: Add a module parm to specify that SMBus alerts don't work
  ipmi: add of_device_id in MODULE_DEVICE_TABLE
  ipmi: Compensate for BMCs that wont set the irq enable bit
  ipmi: Don't call receive handler in the panic context
  ipmi: Avoid touching possible corrupted lists in the panic context
  ipmi: Don't flush messages in sender() in run-to-completion mode
  ipmi: Factor out message flushing procedure
  ipmi: Remove unneeded set_run_to_completion call
  ipmi: Make some data const that was only read
  ipmi: constify SSIF ACPI device ids
  ipmi: Delete an unnecessary check before the function call "cleanup_one_si"
  char:ipmi - Change 1 to true for bool type variables during initialization.
  impi:Remove unneeded setting of module owner to THIS_MODULE in the platform structure, powernv_ipmi_driver
  ipmi: Add a comment in how messages are delivered from the lower layer
  ipmi/powernv: Fix potential invalid pointer dereference
  ipmi: Convert the IPMI SI ACPI handling to a platform device
  ipmi: Add device tree bindings information

9 years agoMerge branch 'akpm' (patches from Andrew)
Linus Torvalds [Wed, 9 Sep 2015 00:52:23 +0000 (17:52 -0700)]
Merge branch 'akpm' (patches from Andrew)

Merge second patch-bomb from Andrew Morton:
 "Almost all of the rest of MM.  There was an unusually large amount of
  MM material this time"

* emailed patches from Andrew Morton <akpm@linux-foundation.org>: (141 commits)
  zpool: remove no-op module init/exit
  mm: zbud: constify the zbud_ops
  mm: zpool: constify the zpool_ops
  mm: swap: zswap: maybe_preload & refactoring
  zram: unify error reporting
  zsmalloc: remove null check from destroy_handle_cache()
  zsmalloc: do not take class lock in zs_shrinker_count()
  zsmalloc: use class->pages_per_zspage
  zsmalloc: consider ZS_ALMOST_FULL as migrate source
  zsmalloc: partial page ordering within a fullness_list
  zsmalloc: use shrinker to trigger auto-compaction
  zsmalloc: account the number of compacted pages
  zsmalloc/zram: introduce zs_pool_stats api
  zsmalloc: cosmetic compaction code adjustments
  zsmalloc: introduce zs_can_compact() function
  zsmalloc: always keep per-class stats
  zsmalloc: drop unused variable `nr_to_migrate'
  mm/memblock.c: fix comment in __next_mem_range()
  mm/page_alloc.c: fix type information of memoryless node
  memory-hotplug: fix comments in zone_spanned_pages_in_node() and zone_spanned_pages_in_node()
  ...

9 years agoMerge branch 'parisc-4.3-1' of git://git.kernel.org/pub/scm/linux/kernel/git/deller...
Linus Torvalds [Wed, 9 Sep 2015 00:45:05 +0000 (17:45 -0700)]
Merge branch 'parisc-4.3-1' of git://git.kernel.org/pub/scm/linux/kernel/git/deller/parisc-linux

Pull parisc updates from Helge Deller:
 "The most important changes in this patchset are:

   - re-enable 64bit PCI bus addresses which were temporarily disabled
     for PA-RISC in kernel 4.2

   - fix the 64bit CAS operation in the LWS path which now enables us to
     enable the 64bit gcc atomic builtins even on 32bit userspace with
     64bit kernel

   - fix a long-standing bug which sometimes crashed kernel at bootup
     while serial interrupt wasn't registered yet"

* 'parisc-4.3-1' of git://git.kernel.org/pub/scm/linux/kernel/git/deller/parisc-linux:
  parisc: Use platform_device_register_simple("rtc-generic")
  parisc: Drop CONFIG_SMP around update_cr16_clocksource()
  parisc: Use double word condition in 64bit CAS operation
  parisc: Filter out spurious interrupts in PA-RISC irq handler
  parisc: Additionally check for in_atomic() in page fault handler
  PCI,parisc: Enable 64-bit bus addresses on PA-RISC
  parisc: Define ioremap_uc and ioremap_wc

9 years agoMerge tag 'linux-kselftest-4.3-rc1' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Wed, 9 Sep 2015 00:39:10 +0000 (17:39 -0700)]
Merge tag 'linux-kselftest-4.3-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/shuah/linux-kselftest

Pull kselftest update from Shuah Khan:
 "This update adds new zram test and fixes to problems found during
  testing this new zram test.  In addition, there are a few bug fixes
  and ksefltest improvement patches from Linaro developers.

  I will send another update later on this week to fix kselftest
  breakage due to commit 2bf9e0ab08c6 ("locking/static_keys: Provide a
  selftest") after the fix soaks in next for a couple of days"

* tag 'linux-kselftest-4.3-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/shuah/linux-kselftest:
  selftests/zram: Makefile fix
  selftests/zram: must be run as root
  selftests: breakpoints: fix installing error on the architecture except x86
  selftests: check before install
  selftests/zram: Adding zram tests

9 years agoMerge tag 'iommu-updates-v4.3' of git://git.kernel.org/pub/scm/linux/kernel/git/joro...
Linus Torvalds [Wed, 9 Sep 2015 00:22:35 +0000 (17:22 -0700)]
Merge tag 'iommu-updates-v4.3' of git://git.kernel.org/pub/scm/linux/kernel/git/joro/iommu

Pull iommu updates for from Joerg Roedel:
 "This time the IOMMU updates are mostly cleanups or fixes.  No big new
  features or drivers this time.  In particular the changes include:

   - Bigger cleanup of the Domain<->IOMMU data structures and the code
     that manages them in the Intel VT-d driver.  This makes the code
     easier to understand and maintain, and also easier to keep the data
     structures in sync.  It is also a preparation step to make use of
     default domains from the IOMMU core in the Intel VT-d driver.

   - Fixes for a couple of DMA-API misuses in ARM IOMMU drivers, namely
     in the ARM and Tegra SMMU drivers.

   - Fix for a potential buffer overflow in the OMAP iommu driver's
     debug code

   - A couple of smaller fixes and cleanups in various drivers

   - One small new feature: Report domain-id usage in the Intel VT-d
     driver to easier detect bugs where these are leaked"

* tag 'iommu-updates-v4.3' of git://git.kernel.org/pub/scm/linux/kernel/git/joro/iommu: (83 commits)
  iommu/vt-d: Really use upper context table when necessary
  x86/vt-d: Fix documentation of DRHD
  iommu/fsl: Really fix init section(s) content
  iommu/io-pgtable-arm: Unmap and free table when overwriting with block
  iommu/io-pgtable-arm: Move init-fn declarations to io-pgtable.h
  iommu/msm: Use BUG_ON instead of if () BUG()
  iommu/vt-d: Access iomem correctly
  iommu/vt-d: Make two functions static
  iommu/vt-d: Use BUG_ON instead of if () BUG()
  iommu/vt-d: Return false instead of 0 in irq_remapping_cap()
  iommu/amd: Use BUG_ON instead of if () BUG()
  iommu/amd: Make a symbol static
  iommu/amd: Simplify allocation in irq_remapping_alloc()
  iommu/tegra-smmu: Parameterize number of TLB lines
  iommu/tegra-smmu: Factor out tegra_smmu_set_pde()
  iommu/tegra-smmu: Extract tegra_smmu_pte_get_use()
  iommu/tegra-smmu: Use __GFP_ZERO to allocate zeroed pages
  iommu/tegra-smmu: Remove PageReserved manipulation
  iommu/tegra-smmu: Convert to use DMA API
  iommu/tegra-smmu: smmu_flush_ptc() wants device addresses
  ...

9 years agoMerge tag 'regmap-v4.3' of git://git.kernel.org/pub/scm/linux/kernel/git/broonie...
Linus Torvalds [Tue, 8 Sep 2015 23:48:55 +0000 (16:48 -0700)]
Merge tag 'regmap-v4.3' of git://git.kernel.org/pub/scm/linux/kernel/git/broonie/regmap

Pull regmap updates from Mark Brown:
 "This has been a busy release for regmap.

  By far the biggest set of changes here are those from Markus Pargmann
  which implement support for block transfers in smbus devices.  This
  required quite a bit of refactoring but leaves us better able to
  handle odd restrictions that controllers may have and with better
  performance on smbus.

  Other new features include:

   - Fix interactions with lockdep for nested regmaps (eg, when a device
     using regmap is connected to a bus where the bus controller has a
     separate regmap).  Lockdep's default class identification is too
     crude to work without help.

   - Support for must write bitfield operations, useful for operations
     which require writing a bit to trigger them from Kuniori Morimoto.

   - Support for delaying during register patch application from Nariman
     Poushin.

   - Support for overriding cache state via the debugfs implementation
     from Richard Fitzgerald"

* tag 'regmap-v4.3' of git://git.kernel.org/pub/scm/linux/kernel/git/broonie/regmap: (25 commits)
  regmap: fix a NULL pointer dereference in __regmap_init
  regmap: Support bulk reads for devices without raw formatting
  regmap-i2c: Add smbus i2c block support
  regmap: Add raw_write/read checks for max_raw_write/read sizes
  regmap: regmap max_raw_read/write getter functions
  regmap: Introduce max_raw_read/write for regmap_bulk_read/write
  regmap: Add missing comments about struct regmap_bus
  regmap: No multi_write support if bus->write does not exist
  regmap: Split use_single_rw internally into use_single_read/write
  regmap: Fix regmap_bulk_write for bus writes
  regmap: regmap_raw_read return error on !bus->read
  regulator: core: Print at debug level on debugfs creation failure
  regmap: Fix regmap_can_raw_write check
  regmap: fix typos in regmap.c
  regmap: Fix integertypes for register address and value
  regmap: Move documentation to regmap.h
  regmap: Use different lockdep class for each regmap init call
  thermal: sti: Add parentheses around bridge->ops->regmap_init call
  mfd: vexpress: Add parentheses around bridge->ops->regmap_init call
  regmap: debugfs: Fix misuse of IS_ENABLED
  ...

9 years agoMerge tag 'fbdev-4.3' of git://git.kernel.org/pub/scm/linux/kernel/git/tomba/linux
Linus Torvalds [Tue, 8 Sep 2015 23:42:55 +0000 (16:42 -0700)]
Merge tag 'fbdev-4.3' of git://git.kernel.org/pub/scm/linux/kernel/git/tomba/linux

Pull fbdev updates from Tomi Valkeinen:
 "Minor fixes and cleanups"

* tag 'fbdev-4.3' of git://git.kernel.org/pub/scm/linux/kernel/git/tomba/linux:
  video: fbdev: atmel_lcdfb: remove useless include
  video: fbdev: pxa168fb: Use devm_clk_get
  fbdev: ssd1307fb: fix error return code
  fbdev: fix snprintf() limit in show_bl_curve()
  video: fbdev: s3c-fb: Constify platform_device_id
  video: fbdev: atmel: fix warning for const return value
  video: fbdev: Drop owner assignment from platform_driver
  video: fbdev: Drop owner assignment from i2c_driver
  fbdev: remove unnecessary memset in vfb
  framebuffer: disable vgacon on microblaze arch
  fbdev: udlfb: remove unneeded initialization in few places
  fbdev: Allow compile test of GPIO consumers if !GPIOLIB
  fbdev: fix cea_modes array size

9 years agoMerge tag 'mmc-v4.3' of git://git.linaro.org/people/ulf.hansson/mmc
Linus Torvalds [Tue, 8 Sep 2015 23:33:16 +0000 (16:33 -0700)]
Merge tag 'mmc-v4.3' of git://git.linaro.org/people/ulf.hansson/mmc

Pull MMC updates from Ulf Hansson:
 "MMC core:
   - Fix a race condition in the request handling
   - Skip trim commands for some buggy kingston eMMCs
   - An optimization and a correction for erase groups
   - Set CMD23 quirk for some Sandisk cards

  MMC host:
   - sdhci: Give GPIO CD higher precedence and don't poll when it's used
   - sdhci: Fix DMA memory leakage
   - sdhci: Some updates for clock management
   - sdhci-of-at91: introduce driver for the Atmel SDMMC
   - sdhci-of-arasan: Add support for sdhci-5.1
   - sdhci-esdhc-imx: Add support for imx7d which also supports HS400
   - sdhci: A collection of fixes and improvements for various sdhci hosts
   - omap_hsmmc: Modernization of the regulator code
   - dw_mmc: A couple of fixes for DMA and PIO mode
   - usdhi6rol0: A few fixes and support probe deferral for regulators
   - pxamci: Convert to use dmaengine
   - sh_mmcif: Fix the suspend process in a short term solution
   - tmio: Adjust timeout for commands
   - sunxi: Fix timeout while gating/ungating clock"

* tag 'mmc-v4.3' of git://git.linaro.org/people/ulf.hansson/mmc: (67 commits)
  mmc: android-goldfish: remove incorrect __iomem annotation
  mmc: core: fix race condition in mmc_wait_data_done
  mmc: host: omap_hsmmc: remove CONFIG_REGULATOR check
  mmc: host: omap_hsmmc: use ios->vdd for setting vmmc voltage
  mmc: host: omap_hsmmc: use regulator_is_enabled to find pbias status
  mmc: host: omap_hsmmc: enable/disable vmmc_aux regulator based on previous state
  mmc: host: omap_hsmmc: don't use ->set_power to set initial regulator state
  mmc: host: omap_hsmmc: avoid pbias regulator enable on power off
  mmc: host: omap_hsmmc: add separate function to set pbias
  mmc: host: omap_hsmmc: add separate functions for enable/disable supply
  mmc: host: omap_hsmmc: return error if any of the regulator APIs fail
  mmc: host: omap_hsmmc: remove unnecessary pbias set_voltage
  mmc: host: omap_hsmmc: use mmc_host's vmmc and vqmmc
  mmc: host: omap_hsmmc: use the ocrmask provided by the vmmc regulator
  mmc: host: omap_hsmmc: cleanup omap_hsmmc_reg_get()
  mmc: host: omap_hsmmc: return on fatal errors from omap_hsmmc_reg_get
  mmc: host: omap_hsmmc: use devm_regulator_get_optional() for vmmc
  mmc: sdhci-of-at91: fix platform_no_drv_owner.cocci warnings
  mmc: sh_mmcif: Fix suspend process
  mmc: usdhi6rol0: fix error return code
  ...

9 years agoMerge tag 'platform-drivers-x86-v4.3-1' of git://git.infradead.org/users/dvhart/linux...
Linus Torvalds [Tue, 8 Sep 2015 23:26:18 +0000 (16:26 -0700)]
Merge tag 'platform-drivers-x86-v4.3-1' of git://git.infradead.org/users/dvhart/linux-platform-drivers-x86

Pull x86 platform driver updates from Darren Hart:
 "Significant work on toshiba_acpi, including new hardware support,
  refactoring, and cleanups.  Extend device support for asus, ideapad,
  and acer systems.  New surface pro 3 buttons driver.  Misc minor
  cleanups for thinkpad and hp-wireless.

  acer-wmi:
   - No rfkill on HP Omen 15 wifi

  thinkpad_acpi:
   - Remove side effects from vdbg_printk -> no_printk macro

  surface pro 3:
   - Add support driver for Surface Pro 3 buttons

  hp-wireless:
   - remove unneeded goto/label in hpwl_init

  ideapad-laptop:
   - add alternative representation for Yoga 2 to DMI table
   - Add Lenovo Yoga 3 14 to no_hw_rfkill dmi list

  asus-laptop:
   - Add key found on Asus F3M

  MAINTAINERS:
   - Remove Toshiba Linux mailing list address

  toshiba_acpi:
   - Bump driver version to 0.23
   - Remove unnecessary checks and returns in HCI/SCI functions
   - Refactor *{get, set} functions return value
   - Remove "*not supported" feature prints
   - Change *available functions return type
   - Add set_fan_status function
   - Change some variables to avoid warnings from ninja-check
   - Reorder toshiba_acpi_alt_keymap entries
   - Remove unused wireless defines
   - Transflective backlight updates
   - Avoid registering input device on WMI event laptops
   - Add /dev/toshiba_acpi device
   - Adapt /proc/acpi/toshiba/keys to TOS1900 devices"

* tag 'platform-drivers-x86-v4.3-1' of git://git.infradead.org/users/dvhart/linux-platform-drivers-x86: (21 commits)
  acer-wmi: No rfkill on HP Omen 15 wifi
  thinkpad_acpi: Remove side effects from vdbg_printk -> no_printk macro
  surface pro 3: Add support driver for Surface Pro 3 buttons
  hp-wireless: remove unneeded goto/label in hpwl_init
  ideapad-laptop: add alternative representation for Yoga 2 to DMI table
  asus-laptop: Add key found on Asus F3M
  MAINTAINERS: Remove Toshiba Linux mailing list address
  ideapad-laptop: Add Lenovo Yoga 3 14 to no_hw_rfkill dmi list
  toshiba_acpi: Bump driver version to 0.23
  toshiba_acpi: Remove unnecessary checks and returns in HCI/SCI functions
  toshiba_acpi: Refactor *{get, set} functions return value
  toshiba_acpi: Remove "*not supported" feature prints
  toshiba_acpi: Change *available functions return type
  toshiba_acpi: Add set_fan_status function
  toshiba_acpi: Change some variables to avoid warnings from ninja-check
  toshiba_acpi: Reorder toshiba_acpi_alt_keymap entries
  toshiba_acpi: Remove unused wireless defines
  toshiba_acpi: Transflective backlight updates
  toshiba_acpi: Avoid registering input device on WMI event laptops
  toshiba_acpi: Add /dev/toshiba_acpi device
  ...

9 years agoMerge branch 'i2c/for-4.3' of git://git.kernel.org/pub/scm/linux/kernel/git/wsa/linux
Linus Torvalds [Tue, 8 Sep 2015 23:16:26 +0000 (16:16 -0700)]
Merge branch 'i2c/for-4.3' of git://git.kernel.org/pub/scm/linux/kernel/git/wsa/linux

Pull i2c updates from Wolfram Sang:
 "Features:

   - new drivers: Renesas EMEV2, register based MUX, NXP LPC2xxx
   - core: scans DT and assigns wakeup interrupts.  no driver changes needed.
   - core: some refcouting issues fixed and better API for that
   - core: new helper function for best effort block read emulation
   - slave framework: proper DT bindings and userspace instantiation
   - some bigger work for xiic, pxa, omap drivers

  .. and quite a number of smaller driver fixes, cleanups, improvements"

* 'i2c/for-4.3' of git://git.kernel.org/pub/scm/linux/kernel/git/wsa/linux: (65 commits)
  i2c: mux: reg Change ioread endianness for readback
  i2c: mux: reg: fix compilation warnings
  i2c: mux: reg: simplify register size checking
  i2c: muxes: fix leaked i2c adapter device node references
  i2c: allow specifying separate wakeup interrupt in device tree
  of/irq: export of_get_irq_byname()
  i2c: xgene-slimpro: dma_mapping_error() doesn't return an error code
  i2c: Replace I2C_CROS_EC_TUNNEL dependency
  eeprom: at24: use i2c_smbus_read_i2c_block_data_or_emulated
  i2c: core: Add support for best effort block read emulation
  i2c: lpc2k: add driver
  i2c: mux: Add register-based mux i2c-mux-reg
  i2c: dt: describe generic bindings
  i2c: slave: print warning if slave flag not set
  i2c: support 10 bit and slave addresses in sysfs 'new_device'
  i2c: take address space into account when checking for used addresses
  i2c: apply DT flags when probing
  i2c: make address check indpendent from client struct
  i2c: rename address check functions
  i2c: apply address offset for slaves, too
  ...

9 years agoMerge tag 'rtc-v4.3' of git://git.kernel.org/pub/scm/linux/kernel/git/abelloni/linux
Linus Torvalds [Tue, 8 Sep 2015 22:46:31 +0000 (15:46 -0700)]
Merge tag 'rtc-v4.3' of git://git.kernel.org/pub/scm/linux/kernel/git/abelloni/linux

Pull RTC updates from Alexandre Belloni:
 "Core:
   - use is_visible() to control sysfs attributes
   - switch wakealarm attribute to DEVICE_ATTR_RW
   - make rtc_does_wakealarm() return boolean
   - properly manage lifetime of dev and cdev in rtc device
   - remove unnecessary device_get() in rtc_device_unregister
   - fix double free in rtc_register_device() error path

  New drivers:
   - NXP LPC24xx
   - Xilinx Zynq MP
   - Dialog DA9062

  Subsystem wide cleanups:
   - fix drivers that consider 0 as a valid IRQ in client->irq
   - Drop (un)likely before IS_ERR(_OR_NULL)
   - drop the remaining owner assignment for i2c_driver and
     platform_driver
   - module autoload fixes

  Drivers:
   - 88pm80x: add device tree support
   - abx80x: fix RTC write bit
   - ab8500: Add a sentinel to ab85xx_rtc_ids[]
   - armada38x: Align RTC set time procedure with the official errata
   - as3722: correct month value
   - at91sam9: cleanups
   - at91rm9200: get and use slow clock and cleanups
   - bq32k: remove redundant check
   - cmos: century support, proper fix for the spurious wakeup
   - ds1307: cleanups and wakeup irq support
   - ds1374: Remove unused variable
   - ds1685: Use module_platform_driver
   - ds3232: fix WARNING trace in resume function
   - gemini: fix ptr_ret.cocci warnings
   - mt6397: implement suspend/resume
   - omap: support internal and external clock enabling
   - opal: Enable alarms only when opal supports tpo
   - pcf2127: use OFS flag to detect unreliable date and warn the user
   - pl031: fix typo for author email
   - rx8025: huge cleanup and fixes
   - sa1100/pxa: share common code
   - s5m: fix to update ctrl register
   - s3c: fix clocks and wakeup, cleanup
   - sirfsoc: use regmap
   - nvram_read()/nvram_write() functions for cmos, ds1305, ds1307,
     ds1343, ds1511, ds1553, ds1742, m48t59, rp5c01, stk17ta8, tx4939
   - use rtc_valid_tm() error code when reading date/time instead of 0
     for isl12022, pcf2123, pcf2127"

* tag 'rtc-v4.3' of git://git.kernel.org/pub/scm/linux/kernel/git/abelloni/linux: (90 commits)
  rtc: abx80x: fix RTC write bit
  rtc: ab8500: Add a sentinel to ab85xx_rtc_ids[]
  rtc: ds1374: Remove unused variable
  rtc: Fix module autoload for OF platform drivers
  rtc: Fix module autoload for rtc-{ab8500,max8997,s5m} drivers
  rtc: omap: Add external clock enabling support
  rtc: omap: Add internal clock enabling support
  ARM: dts: AM437x: Add the internal and external clock nodes for rtc
  rtc: s5m: fix to update ctrl register
  rtc: add xilinx zynqmp rtc driver
  devicetree: bindings: rtc: add bindings for xilinx zynqmp rtc
  rtc: as3722: correct month value
  ARM: config: Switch PXA27x platforms to use PXA RTC driver
  ARM: mmp: remove unused RTC register definitions
  ARM: sa1100: remove unused RTC register definitions
  rtc: sa1100/pxa: convert to run-time register mapping
  ARM: pxa: add memory resource to SA1100 RTC device
  rtc: pxa: convert to use shared sa1100 functions
  rtc: sa1100: prepare to share sa1100_rtc_ops
  rtc: ds3232: fix WARNING trace in resume function
  ...

9 years agozpool: remove no-op module init/exit
Dan Streetman [Tue, 8 Sep 2015 22:05:09 +0000 (15:05 -0700)]
zpool: remove no-op module init/exit

Remove zpool_init() and zpool_exit(); they do nothing other than print
"loaded" and "unloaded".

Signed-off-by: Dan Streetman <ddstreet@ieee.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm: zbud: constify the zbud_ops
Krzysztof Kozlowski [Tue, 8 Sep 2015 22:05:06 +0000 (15:05 -0700)]
mm: zbud: constify the zbud_ops

The structure zbud_ops is not modified so make the pointer to it a
pointer to const.

Signed-off-by: Krzysztof Kozlowski <k.kozlowski@samsung.com>
Acked-by: Dan Streetman <ddstreet@ieee.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm: zpool: constify the zpool_ops
Krzysztof Kozlowski [Tue, 8 Sep 2015 22:05:03 +0000 (15:05 -0700)]
mm: zpool: constify the zpool_ops

The structure zpool_ops is not modified so make the pointer to it a
pointer to const.

Signed-off-by: Krzysztof Kozlowski <k.kozlowski@samsung.com>
Acked-by: Dan Streetman <ddstreet@ieee.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm: swap: zswap: maybe_preload & refactoring
Dmitry Safonov [Tue, 8 Sep 2015 22:05:00 +0000 (15:05 -0700)]
mm: swap: zswap: maybe_preload & refactoring

zswap_get_swap_cache_page and read_swap_cache_async have pretty much the
same code with only significant difference in return value and usage of
swap_readpage.

I a helper __read_swap_cache_async() with the common code.  Behavior
change: now zswap_get_swap_cache_page will use radix_tree_maybe_preload
instead radix_tree_preload.  Looks like, this wasn't changed only by the
reason of code duplication.

Signed-off-by: Dmitry Safonov <0x7f454c46@gmail.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Vladimir Davydov <vdavydov@parallels.com>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Hugh Dickins <hughd@google.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Tejun Heo <tj@kernel.org>
Cc: Jens Axboe <axboe@fb.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: David Herrmann <dh.herrmann@gmail.com>
Cc: Seth Jennings <sjennings@variantweb.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agozram: unify error reporting
Sergey Senozhatsky [Tue, 8 Sep 2015 22:04:58 +0000 (15:04 -0700)]
zram: unify error reporting

Make zram syslog error reporting more consistent. We have random
error levels in some places. For example, critical errors like
  "Error allocating memory for compressed page"
and
  "Unable to allocate temp memory"
are reported as KERN_INFO messages.

a) Reassign error levels

Error messages that directly affect zram
functionality -- pr_err():

 Error allocating zram address table
 Error creating memory pool
 Decompression failed! err=%d, page=%u
 Unable to allocate temp memory
 Compression failed! err=%d
 Error allocating memory for compressed page: %u, size=%zu
 Cannot initialise %s compressing backend
 Error allocating disk queue for device %d
 Error allocating disk structure for device %d
 Error creating sysfs group for device %d
 Unable to register zram-control class
 Unable to get major number

Messages that do not affect functionality, but user
must be warned (because sysfs attrs will be removed in
this particular case) -- pr_warn():

 %d (%s) Attribute %s (and others) will be removed. %s

Messages that do not affect functionality and mostly are
informative -- pr_info():

 Cannot change max compression streams
 Can't change algorithm for initialized device
 Cannot change disksize for initialized device
 Added device: %s
 Removed device: %s

b) Update sysfs_create_group() error message

First, it lacks a trailing new line; add it.  Second, every error message
in zram_add() has a "for device %d" part, which makes errors more
informative.  Add missing part to "Error creating sysfs group" message.

Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.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>
9 years agozsmalloc: remove null check from destroy_handle_cache()
Sergey Senozhatsky [Tue, 8 Sep 2015 22:04:55 +0000 (15:04 -0700)]
zsmalloc: remove null check from destroy_handle_cache()

We can pass a NULL cache pointer to kmem_cache_destroy(), because it
NULL-checks its argument now.  Remove redundant test from
destroy_handle_cache().

Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Acked-by: Minchan Kim <minchan@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agozsmalloc: do not take class lock in zs_shrinker_count()
Sergey Senozhatsky [Tue, 8 Sep 2015 22:04:52 +0000 (15:04 -0700)]
zsmalloc: do not take class lock in zs_shrinker_count()

We can avoid taking class ->lock around zs_can_compact() in
zs_shrinker_count(), because the number that we return back is outdated
in general case, by design.  We have different sources that are able to
change class's state right after we return from zs_can_compact() --
ongoing I/O operations, manually triggered compaction, or two of them
happening simultaneously.

We re-do this calculations during compaction on a per class basis
anyway.

zs_unregister_shrinker() will not return until we have an active
shrinker, so classes won't unexpectedly disappear while
zs_shrinker_count() iterates them.

Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Acked-by: Minchan Kim <minchan@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agozsmalloc: use class->pages_per_zspage
Minchan Kim [Tue, 8 Sep 2015 22:04:49 +0000 (15:04 -0700)]
zsmalloc: use class->pages_per_zspage

There is no need to recalcurate pages_per_zspage in runtime.  Just use
class->pages_per_zspage to avoid unnecessary runtime overhead.

Signed-off-by: Minchan Kim <minchan@kernel.org>
Acked-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agozsmalloc: consider ZS_ALMOST_FULL as migrate source
Minchan Kim [Tue, 8 Sep 2015 22:04:47 +0000 (15:04 -0700)]
zsmalloc: consider ZS_ALMOST_FULL as migrate source

There is no reason to prevent select ZS_ALMOST_FULL as migration source
if we cannot find source from ZS_ALMOST_EMPTY.

With this patch, zs_can_compact will return more exact result.

Signed-off-by: Minchan Kim <minchan.kim@lge.com>
Acked-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agozsmalloc: partial page ordering within a fullness_list
Sergey Senozhatsky [Tue, 8 Sep 2015 22:04:44 +0000 (15:04 -0700)]
zsmalloc: partial page ordering within a fullness_list

We want to see more ZS_FULL pages and less ZS_ALMOST_{FULL, EMPTY}
pages.  Put a page with higher ->inuse count first within its
->fullness_list, which will give us better chances to fill up this page
with new objects (find_get_zspage() return ->fullness_list head for new
object allocation), so some zspages will become ZS_ALMOST_FULL/ZS_FULL
quicker.

It performs a trivial and cheap ->inuse compare which does not slow down
zsmalloc and in the worst case keeps the list pages in no particular
order.

A more expensive solution could sort fullness_list by ->inuse count.

[minchan@kernel.org: code adjustments]
Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.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>
9 years agozsmalloc: use shrinker to trigger auto-compaction
Sergey Senozhatsky [Tue, 8 Sep 2015 22:04:41 +0000 (15:04 -0700)]
zsmalloc: use shrinker to trigger auto-compaction

Perform automatic pool compaction by a shrinker when system is getting
tight on memory.

User-space has a very little knowledge regarding zsmalloc fragmentation
and basically has no mechanism to tell whether compaction will result in
any memory gain.  Another issue is that user space is not always aware
of the fact that system is getting tight on memory.  Which leads to very
uncomfortable scenarios when user space may start issuing compaction
'randomly' or from crontab (for example).  Fragmentation is not always
necessarily bad, allocated and unused objects, after all, may be filled
with the data later, w/o the need of allocating a new zspage.  On the
other hand, we obviously don't want to waste memory when the system
needs it.

Compaction now has a relatively quick pool scan so we are able to
estimate the number of pages that will be freed easily, which makes it
possible to call this function from a shrinker->count_objects()
callback.  We also abort compaction as soon as we detect that we can't
free any pages any more, preventing wasteful objects migrations.

Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Suggested-by: Minchan Kim <minchan@kernel.org>
Acked-by: Minchan Kim <minchan@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agozsmalloc: account the number of compacted pages
Sergey Senozhatsky [Tue, 8 Sep 2015 22:04:38 +0000 (15:04 -0700)]
zsmalloc: account the number of compacted pages

Compaction returns back to zram the number of migrated objects, which is
quite uninformative -- we have objects of different sizes so user space
cannot obtain any valuable data from that number.  Change compaction to
operate in terms of pages and return back to compaction issuer the
number of pages that were freed during compaction.  So from now on we
will export more meaningful value in zram<id>/mm_stat -- the number of
freed (compacted) pages.

This requires:
 (a) a rename of `num_migrated' to 'pages_compacted'
 (b) a internal API change -- return first_page's fullness_group from
     putback_zspage(), so we know when putback_zspage() did
     free_zspage().  It helps us to account compaction stats correctly.

Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Acked-by: Minchan Kim <minchan@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agozsmalloc/zram: introduce zs_pool_stats api
Sergey Senozhatsky [Tue, 8 Sep 2015 22:04:35 +0000 (15:04 -0700)]
zsmalloc/zram: introduce zs_pool_stats api

`zs_compact_control' accounts the number of migrated objects but it has
a limited lifespan -- we lose it as soon as zs_compaction() returns back
to zram.  It worked fine, because (a) zram had it's own counter of
migrated objects and (b) only zram could trigger compaction.  However,
this does not work for automatic pool compaction (not issued by zram).
To account objects migrated during auto-compaction (issued by the
shrinker) we need to store this number in zs_pool.

Define a new `struct zs_pool_stats' structure to keep zs_pool's stats
there.  It provides only `num_migrated', as of this writing, but it
surely can be extended.

A new zsmalloc zs_pool_stats() symbol exports zs_pool's stats back to
caller.

Use zs_pool_stats() in zram and remove `num_migrated' from zram_stats.

Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Suggested-by: Minchan Kim <minchan@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agozsmalloc: cosmetic compaction code adjustments
Sergey Senozhatsky [Tue, 8 Sep 2015 22:04:33 +0000 (15:04 -0700)]
zsmalloc: cosmetic compaction code adjustments

Change zs_object_copy() argument order to be (DST, SRC) rather than
(SRC, DST).  copy/move functions usually have (to, from) arguments
order.

Rename alloc_target_page() to isolate_target_page().  This function
doesn't allocate anything, it isolates target page, pretty much like
isolate_source_page().

Tweak __zs_compact() comment.

Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Acked-by: Minchan Kim <minchan@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agozsmalloc: introduce zs_can_compact() function
Sergey Senozhatsky [Tue, 8 Sep 2015 22:04:30 +0000 (15:04 -0700)]
zsmalloc: introduce zs_can_compact() function

This function checks if class compaction will free any pages.
Rephrasing -- do we have enough unused objects to form at least one
ZS_EMPTY page and free it.  It aborts compaction if class compaction
will not result in any (further) savings.

EXAMPLE (this debug output is not part of this patch set):

 - class size
 - number of allocated objects
 - number of used objects
 - max objects per zspage
 - pages per zspage
 - estimated number of pages that will be freed

[..]
class-512 objs:544 inuse:540 maxobj-per-zspage:8  pages-per-zspage:1 zspages-to-free:0
 ... class-512 compaction is useless. break
class-496 objs:660 inuse:570 maxobj-per-zspage:33 pages-per-zspage:4 zspages-to-free:2
class-496 objs:627 inuse:570 maxobj-per-zspage:33 pages-per-zspage:4 zspages-to-free:1
class-496 objs:594 inuse:570 maxobj-per-zspage:33 pages-per-zspage:4 zspages-to-free:0
 ... class-496 compaction is useless. break
class-448 objs:657 inuse:617 maxobj-per-zspage:9  pages-per-zspage:1 zspages-to-free:4
class-448 objs:648 inuse:617 maxobj-per-zspage:9  pages-per-zspage:1 zspages-to-free:3
class-448 objs:639 inuse:617 maxobj-per-zspage:9  pages-per-zspage:1 zspages-to-free:2
class-448 objs:630 inuse:617 maxobj-per-zspage:9  pages-per-zspage:1 zspages-to-free:1
class-448 objs:621 inuse:617 maxobj-per-zspage:9  pages-per-zspage:1 zspages-to-free:0
 ... class-448 compaction is useless. break
class-432 objs:728 inuse:685 maxobj-per-zspage:28 pages-per-zspage:3 zspages-to-free:1
class-432 objs:700 inuse:685 maxobj-per-zspage:28 pages-per-zspage:3 zspages-to-free:0
 ... class-432 compaction is useless. break
class-416 objs:819 inuse:705 maxobj-per-zspage:39 pages-per-zspage:4 zspages-to-free:2
class-416 objs:780 inuse:705 maxobj-per-zspage:39 pages-per-zspage:4 zspages-to-free:1
class-416 objs:741 inuse:705 maxobj-per-zspage:39 pages-per-zspage:4 zspages-to-free:0
 ... class-416 compaction is useless. break
class-400 objs:690 inuse:674 maxobj-per-zspage:10 pages-per-zspage:1 zspages-to-free:1
class-400 objs:680 inuse:674 maxobj-per-zspage:10 pages-per-zspage:1 zspages-to-free:0
 ... class-400 compaction is useless. break
class-384 objs:736 inuse:709 maxobj-per-zspage:32 pages-per-zspage:3 zspages-to-free:0
 ... class-384 compaction is useless. break
[..]

Every "compaction is useless" indicates that we saved CPU cycles.

class-512 has
544 object allocated
540 objects used
8 objects per-page

Even if we have a ALMOST_EMPTY zspage, we still don't have enough room to
migrate all of its objects and free this zspage; so compaction will not
make a lot of sense, it's better to just leave it as is.

Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Acked-by: Minchan Kim <minchan@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agozsmalloc: always keep per-class stats
Sergey Senozhatsky [Tue, 8 Sep 2015 22:04:27 +0000 (15:04 -0700)]
zsmalloc: always keep per-class stats

Always account per-class `zs_size_stat' stats.  This data will help us
make better decisions during compaction.  We are especially interested
in OBJ_ALLOCATED and OBJ_USED, which can tell us if class compaction
will result in any memory gain.

For instance, we know the number of allocated objects in the class, the
number of objects being used (so we also know how many objects are not
used) and the number of objects per-page.  So we can ensure if we have
enough unused objects to form at least one ZS_EMPTY zspage during
compaction.

We calculate this value on per-class basis so we can calculate a total
number of zspages that can be released.  Which is exactly what a
shrinker wants to know.

Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Acked-by: Minchan Kim <minchan@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agozsmalloc: drop unused variable `nr_to_migrate'
Sergey Senozhatsky [Tue, 8 Sep 2015 22:04:24 +0000 (15:04 -0700)]
zsmalloc: drop unused variable `nr_to_migrate'

This patchset tweaks compaction and makes it possible to trigger pool
compaction automatically when system is getting low on memory.

zsmalloc in some cases can suffer from a notable fragmentation and
compaction can release some considerable amount of memory.  The problem
here is that currently we fully rely on user space to perform compaction
when needed.  However, performing zsmalloc compaction is not always an
obvious thing to do.  For example, suppose we have a `idle' fragmented
(compaction was never performed) zram device and system is getting low
on memory due to some 3rd party user processes (gcc LTO, or firefox,
etc.).  It's quite unlikely that user space will issue zpool compaction
in this case.  Besides, user space cannot tell for sure how badly pool
is fragmented; however, this info is known to zsmalloc and, hence, to a
shrinker.

This patch (of 7):

__zs_compact() does not use `nr_to_migrate', drop it.

Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Acked-by: Minchan Kim <minchan@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm/memblock.c: fix comment in __next_mem_range()
Alexander Kuleshov [Tue, 8 Sep 2015 22:04:22 +0000 (15:04 -0700)]
mm/memblock.c: fix comment in __next_mem_range()

Signed-off-by: Alexander Kuleshov <kuleshovmail@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm/page_alloc.c: fix type information of memoryless node
Zhen Lei [Tue, 8 Sep 2015 22:04:19 +0000 (15:04 -0700)]
mm/page_alloc.c: fix type information of memoryless node

For a memoryless node, the output of get_pfn_range_for_nid are all zero.
It will display mem from 0 to -1.

Signed-off-by: Zhen Lei <thunder.leizhen@huawei.com>
Acked-by: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomemory-hotplug: fix comments in zone_spanned_pages_in_node() and zone_spanned_pages_i...
Xishi Qiu [Tue, 8 Sep 2015 22:04:16 +0000 (15:04 -0700)]
memory-hotplug: fix comments in zone_spanned_pages_in_node() and zone_spanned_pages_in_node()

When hot adding a node from add_memory(), we will add memblock first, so
the node is not empty.  But when called from cpu_up(), the node should
be empty.

Signed-off-by: Xishi Qiu <qiuxishi@huawei.com>
Cc: Tang Chen <tangchen@cn.fujitsu.com>
Cc: Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com>
Cc: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Taku Izumi <izumi.taku@jp.fujitsu.com>\
Acked-by: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm/page_alloc.c: change sysctl_lower_zone_reserve_ratio to sysctl_lowmem_reserve_rati...
Yaowei Bai [Tue, 8 Sep 2015 22:04:13 +0000 (15:04 -0700)]
mm/page_alloc.c: change sysctl_lower_zone_reserve_ratio to sysctl_lowmem_reserve_ratio in comments

We use sysctl_lowmem_reserve_ratio rather than
sysctl_lower_zone_reserve_ratio to determine how aggressive the kernel
is in defending lowmem from the possibility of being captured into
pinned user memory.  To avoid misleading, correct it in some comments.

Signed-off-by: Yaowei Bai <bywxiaobai@163.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm/page_alloc.c: fix a misleading comment
Yaowei Bai [Tue, 8 Sep 2015 22:04:10 +0000 (15:04 -0700)]
mm/page_alloc.c: fix a misleading comment

The comment says that the per-cpu batchsize and zone watermarks are
determined by present_pages which is definitely wrong, they are both
calculated from managed_pages.  Fix it.

Signed-off-by: Yaowei Bai <bywxiaobai@163.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm/mmap.c:insert_vm_struct(): check for failure before setting values
Chen Gang [Tue, 8 Sep 2015 22:04:08 +0000 (15:04 -0700)]
mm/mmap.c:insert_vm_struct(): check for failure before setting values

There's no point in initializing vma->vm_pgoff if the insertion attempt
will be failing anyway.  Run the checks before performing the
initialization.

Signed-off-by: Chen Gang <gang.chen.5i5j@gmail.com>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm/khugepaged: allow interruption of allocation sleep again
Petr Mladek [Tue, 8 Sep 2015 22:04:05 +0000 (15:04 -0700)]
mm/khugepaged: allow interruption of allocation sleep again

Commit 1dfb059b9438 ("thp: reduce khugepaged freezing latency") fixed
khugepaged to do not block a system suspend.  But the result is that it
could not get interrupted before the given timeout because the condition
for the wait event is "false".

This patch puts back the original approach but it uses
freezable_schedule_timeout_interruptible() instead of
schedule_timeout_interruptible().  It does the right thing.  I am pretty
sure that the freezable variant was not used in the original fix only
because it was not available at that time.

The regression has been there for ages.  It was not critical.  It just
did the allocation throttling a little bit more aggressively.

I found this problem when converting the kthread to kthread worker API
and trying to understand the code.

This bug is thought to have minimal userspace-visible impact.  Somebody
could set a high alloc_sleep value by mistake, and then try to fix it
back, but khugepaged would keep sleeping until the high value expires.

Signed-off-by: Petr Mladek <pmladek@suse.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: "Aneesh Kumar K.V" <aneesh.kumar@linux.vnet.ibm.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Ebru Akagunduz <ebru.akagunduz@gmail.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Jiri Kosina <jkosina@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm/memblock.c: fiy typos in comments
Alexander Kuleshov [Tue, 8 Sep 2015 22:04:02 +0000 (15:04 -0700)]
mm/memblock.c: fiy typos in comments

s/succees/success/

Signed-off-by: Alexander Kuleshov <kuleshovmail@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm/compaction: correct to flush migrated pages if pageblock skip happens
Joonsoo Kim [Tue, 8 Sep 2015 22:03:59 +0000 (15:03 -0700)]
mm/compaction: correct to flush migrated pages if pageblock skip happens

We cache isolate_start_pfn before entering isolate_migratepages().  If
pageblock is skipped in isolate_migratepages() due to whatever reason,
cc->migrate_pfn can be far from isolate_start_pfn hence we flush pages
that were freed.  For example, the following scenario can be possible:

- assume order-9 compaction, pageblock order is 9
- start_isolate_pfn is 0x200
- isolate_migratepages()
  - skip a number of pageblocks
  - start to isolate from pfn 0x600
  - cc->migrate_pfn = 0x620
  - return
- last_migrated_pfn is set to 0x200
- check flushing condition
  - current_block_start is set to 0x600
  - last_migrated_pfn < current_block_start then do useless flush

This wrong flush would not help the performance and success rate so this
patch tries to fix it.  One simple way to know the exact position where
we start to isolate migratable pages is that we cache it in
isolate_migratepages() before entering actual isolation.  This patch
implements that and fixes the problem.

Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Rik van Riel <riel@redhat.com>
Cc: David Rientjes <rientjes@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>
9 years agomm: use numa_mem_id() in alloc_pages_node()
Vlastimil Babka [Tue, 8 Sep 2015 22:03:56 +0000 (15:03 -0700)]
mm: use numa_mem_id() in alloc_pages_node()

alloc_pages_node() might fail when called with NUMA_NO_NODE and
__GFP_THISNODE on a CPU belonging to a memoryless node.  To make the
local-node fallback more robust and prevent such situations, use
numa_mem_id(), which was introduced for similar scenarios in the slab
context.

Suggested-by: Christoph Lameter <cl@linux.com>
Signed-off-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: David Rientjes <rientjes@google.com>
Acked-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Christoph Lameter <cl@linux.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>
9 years agomm: unify checks in alloc_pages_node() and __alloc_pages_node()
Vlastimil Babka [Tue, 8 Sep 2015 22:03:53 +0000 (15:03 -0700)]
mm: unify checks in alloc_pages_node() and __alloc_pages_node()

Perform the same debug checks in alloc_pages_node() as are done in
__alloc_pages_node(), by making the former function a wrapper of the
latter one.

In addition to better diagnostics in DEBUG_VM builds for situations
which have been already fatal (e.g.  out-of-bounds node id), there are
two visible changes for potential existing buggy callers of
alloc_pages_node():

- calling alloc_pages_node() with any negative nid (e.g. due to arithmetic
  overflow) was treated as passing NUMA_NO_NODE and fallback to local node was
  applied. This will now be fatal.
- calling alloc_pages_node() with an offline node will now be checked for
  DEBUG_VM builds. Since it's not fatal if the node has been previously online,
  and this patch may expose some existing buggy callers, change the VM_BUG_ON
  in __alloc_pages_node() to VM_WARN_ON.

Signed-off-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: David Rientjes <rientjes@google.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Christoph Lameter <cl@linux.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm: rename alloc_pages_exact_node() to __alloc_pages_node()
Vlastimil Babka [Tue, 8 Sep 2015 22:03:50 +0000 (15:03 -0700)]
mm: rename alloc_pages_exact_node() to __alloc_pages_node()

alloc_pages_exact_node() was introduced in commit 6484eb3e2a81 ("page
allocator: do not check NUMA node ID when the caller knows the node is
valid") as an optimized variant of alloc_pages_node(), that doesn't
fallback to current node for nid == NUMA_NO_NODE.  Unfortunately the
name of the function can easily suggest that the allocation is
restricted to the given node and fails otherwise.  In truth, the node is
only preferred, unless __GFP_THISNODE is passed among the gfp flags.

The misleading name has lead to mistakes in the past, see for example
commits 5265047ac301 ("mm, thp: really limit transparent hugepage
allocation to local node") and b360edb43f8e ("mm, mempolicy:
migrate_to_node should only migrate to node").

Another issue with the name is that there's a family of
alloc_pages_exact*() functions where 'exact' means exact size (instead
of page order), which leads to more confusion.

To prevent further mistakes, this patch effectively renames
alloc_pages_exact_node() to __alloc_pages_node() to better convey that
it's an optimized variant of alloc_pages_node() not intended for general
usage.  Both functions get described in comments.

It has been also considered to really provide a convenience function for
allocations restricted to a node, but the major opinion seems to be that
__GFP_THISNODE already provides that functionality and we shouldn't
duplicate the API needlessly.  The number of users would be small
anyway.

Existing callers of alloc_pages_exact_node() are simply converted to
call __alloc_pages_node(), with the exception of sba_alloc_coherent()
which open-codes the check for NUMA_NO_NODE, so it is converted to use
alloc_pages_node() instead.  This means it no longer performs some
VM_BUG_ON checks, and since the current check for nid in
alloc_pages_node() uses a 'nid < 0' comparison (which includes
NUMA_NO_NODE), it may hide wrong values which would be previously
exposed.

Both differences will be rectified by the next patch.

To sum up, this patch makes no functional changes, except temporarily
hiding potentially buggy callers.  Restricting the checks in
alloc_pages_node() is left for the next patch which can in turn expose
more existing buggy callers.

Signed-off-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Robin Holt <robinmholt@gmail.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Christoph Lameter <cl@linux.com>
Acked-by: Michael Ellerman <mpe@ellerman.id.au>
Cc: Mel Gorman <mgorman@suse.de>
Cc: David Rientjes <rientjes@google.com>
Cc: Greg Thelen <gthelen@google.com>
Cc: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Cc: Tony Luck <tony.luck@intel.com>
Cc: Fenghua Yu <fenghua.yu@intel.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Gleb Natapov <gleb@kernel.org>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Cliff Whickman <cpw@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm, vmscan: unlock page while waiting on writeback
Hugh Dickins [Tue, 8 Sep 2015 22:03:46 +0000 (15:03 -0700)]
mm, vmscan: unlock page while waiting on writeback

This is merely a politeness: I've not found that shrink_page_list()
leads to deadlock with the page it holds locked across
wait_on_page_writeback(); but nevertheless, why hold others off by
keeping the page locked there?

And while we're at it: remove the mistaken "not " from the commentary on
this Case 3 (and a distracting blank line from Case 2, if I may).

Signed-off-by: Hugh Dickins <hughd@google.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Mel Gorman <mgorman@techsingularity.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agolist_lru: don't call list_lru_from_kmem if the list_head is empty
Jeff Layton [Tue, 8 Sep 2015 22:03:44 +0000 (15:03 -0700)]
list_lru: don't call list_lru_from_kmem if the list_head is empty

If the list_head is empty then we'll have called list_lru_from_kmem for
nothing.  Move that call inside of the list_empty if block.

Signed-off-by: Jeff Layton <jeff.layton@primarydata.com>
Reviewed-by: Vladimir Davydov <vdavydov@parallels.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agokmemleak: record accurate early log buffer count and report when exceeded
Wang Kai [Tue, 8 Sep 2015 22:03:41 +0000 (15:03 -0700)]
kmemleak: record accurate early log buffer count and report when exceeded

In log_early function, crt_early_log should also count once when
'crt_early_log >= ARRAY_SIZE(early_log)'.  Otherwise the reported count
from kmemleak_init is one less than 'actual number'.

Then, in kmemleak_init, if early_log buffer size equal actual number,
kmemleak will init sucessful, so change warning condition to
'crt_early_log > ARRAY_SIZE(early_log)'.

Signed-off-by: Wang Kai <morgan.wang@huawei.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>
9 years agomm/mmap.c: simplify the failure return working flow
Chen Gang [Tue, 8 Sep 2015 22:03:38 +0000 (15:03 -0700)]
mm/mmap.c: simplify the failure return working flow

__split_vma() doesn't need out_err label, neither need initializing err.

copy_vma() can return NULL directly when kmem_cache_alloc() fails.

Signed-off-by: Chen Gang <gang.chen.5i5j@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agoshmem: recalculate file inode when fstat
Yu Zhao [Tue, 8 Sep 2015 22:03:33 +0000 (15:03 -0700)]
shmem: recalculate file inode when fstat

Shmem uses shmem_recalc_inode to update i_blocks when it allocates page,
undoes range or swaps.  But mm can drop clean page without notifying
shmem.  This makes fstat sometimes return out-of-date block size.

The problem can be partially solved when we add
inode_operations->getattr which calls shmem_recalc_inode to update
i_blocks for fstat.

shmem_recalc_inode also updates counter used by statfs and
vm_committed_as.  For them the situation is not changed.  They still
suffer from the discrepancy after dropping clean page and before the
function is called by aforementioned triggers.

Signed-off-by: Yu Zhao <yuzhao@google.com>
Signed-off-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm/memblock.c: rename local variable of memblock_type to 'type'
Alexander Kuleshov [Tue, 8 Sep 2015 22:03:33 +0000 (15:03 -0700)]
mm/memblock.c: rename local variable of memblock_type to 'type'

Since commit e3239ff92a17 ("memblock: Rename memblock_region to
memblock_type and memblock_property to memblock_region"), all local
variables of the membock_type type were renamed to 'type'.  This commit
renames all remaining local variables with the memblock_type type to the
same view.

Signed-off-by: Alexander Kuleshov <kuleshovmail@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm/hwpoison: don't try to unpoison containment-failed pages
Naoya Horiguchi [Tue, 8 Sep 2015 22:03:29 +0000 (15:03 -0700)]
mm/hwpoison: don't try to unpoison containment-failed pages

memory_failure() can be called at any page at any time, which means that
we can't eliminate the possibility of containment failure.  In such case
the best option is to leak the page intentionally (and never touch it
later.)

We have an unpoison function for testing, and it cannot handle such
containment-failed pages, which results in kernel panic (visible with
various calltraces.) So this patch suggests that we limit the
unpoisonable pages to properly contained pages and ignore any other
ones.

Testers are recommended to keep in mind that there're un-unpoisonable
pages when writing test programs.

Signed-off-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Tested-by: Wanpeng Li <wanpeng.li@hotmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm/hwpoison: fix race between soft_offline_page and unpoison_memory
Wanpeng Li [Tue, 8 Sep 2015 22:03:27 +0000 (15:03 -0700)]
mm/hwpoison: fix race between soft_offline_page and unpoison_memory

Wanpeng Li reported a race between soft_offline_page() and
unpoison_memory(), which causes the following kernel panic:

   BUG: Bad page state in process bash  pfn:97000
   page:ffffea00025c0000 count:0 mapcount:1 mapping:          (null) index:0x7f4fdbe00
   flags: 0x1fffff80080048(uptodate|active|swapbacked)
   page dumped because: PAGE_FLAGS_CHECK_AT_FREE flag(s) set
   bad because of flags:
   flags: 0x40(active)
   Modules linked in: snd_hda_codec_hdmi i915 rpcsec_gss_krb5 nfsv4 dns_resolver bnep rfcomm nfsd bluetooth auth_rpcgss nfs_acl nfs rfkill lockd grace sunrpc i2c_algo_bit drm_kms_helper snd_hda_codec_realtek snd_hda_codec_generic drm snd_hda_intel fscache snd_hda_codec x86_pkg_temp_thermal coretemp kvm_intel snd_hda_core snd_hwdep kvm snd_pcm snd_seq_dummy snd_seq_oss crct10dif_pclmul snd_seq_midi crc32_pclmul snd_seq_midi_event ghash_clmulni_intel snd_rawmidi aesni_intel lrw gf128mul snd_seq glue_helper ablk_helper snd_seq_device cryptd fuse snd_timer dcdbas serio_raw mei_me parport_pc snd mei ppdev i2c_core video lp soundcore parport lpc_ich shpchp mfd_core ext4 mbcache jbd2 sd_mod e1000e ahci ptp libahci crc32c_intel libata pps_core
   CPU: 3 PID: 2211 Comm: bash Not tainted 4.2.0-rc5-mm1+ #45
   Hardware name: Dell Inc. OptiPlex 7020/0F5C5X, BIOS A03 01/08/2015
   Call Trace:
     dump_stack+0x48/0x5c
     bad_page+0xe6/0x140
     free_pages_prepare+0x2f9/0x320
     ? uncharge_list+0xdd/0x100
     free_hot_cold_page+0x40/0x170
     __put_single_page+0x20/0x30
     put_page+0x25/0x40
     unmap_and_move+0x1a6/0x1f0
     migrate_pages+0x100/0x1d0
     ? kill_procs+0x100/0x100
     ? unlock_page+0x6f/0x90
     __soft_offline_page+0x127/0x2a0
     soft_offline_page+0xa6/0x200

This race is explained like below:

  CPU0                    CPU1

  soft_offline_page
  __soft_offline_page
  TestSetPageHWPoison
                        unpoison_memory
                        PageHWPoison check (true)
                        TestClearPageHWPoison
                        put_page    -> release refcount held by get_hwpoison_page in unpoison_memory
                        put_page    -> release refcount held by isolate_lru_page in __soft_offline_page
  migrate_pages

The second put_page() releases refcount held by isolate_lru_page() which
will lead to unmap_and_move() releases the last refcount of page and w/
mapcount still 1 since try_to_unmap() is not called if there is only one
user map the page.  Anyway, the page refcount and mapcount will still
mess if the page is mapped by multiple users.

This race was introduced by commit 4491f71260 ("mm/memory-failure: set
PageHWPoison before migrate_pages()"), which focuses on preventing the
reuse of successfully migrated page.  Before this commit we prevent the
reuse by changing the migratetype to MIGRATE_ISOLATE during soft
offlining, which has the following problems, so simply reverting the
commit is not a best option:

  1) it doesn't eliminate the reuse completely, because
     set_migratetype_isolate() can fail to set MIGRATE_ISOLATE to the
     target page if the pageblock of the page contains one or more
     unmovable pages (i.e.  has_unmovable_pages() returns true).

  2) the original code changes migratetype to MIGRATE_ISOLATE
     forcibly, and sets it to MIGRATE_MOVABLE forcibly after soft offline,
     regardless of the original migratetype state, which could impact
     other subsystems like memory hotplug or compaction.

This patch moves PageSetHWPoison just after put_page() in
unmap_and_move(), which closes up the reported race window and minimizes
another race window b/w SetPageHWPoison and reallocation (which causes
the reuse of soft-offlined page.) The latter race window still exists
but it's acceptable, because it's rare and effectively the same as
ordinary "containment failure" case even if it happens, so keep the
window open is acceptable.

Fixes: 4491f71260 ("mm/memory-failure: set PageHWPoison before migrate_pages()")
Signed-off-by: Wanpeng Li <wanpeng.li@hotmail.com>
Signed-off-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Reported-by: Wanpeng Li <wanpeng.li@hotmail.com>
Tested-by: Wanpeng Li <wanpeng.li@hotmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm/hwpoison: introduce num_poisoned_pages wrappers
Naoya Horiguchi [Tue, 8 Sep 2015 22:03:24 +0000 (15:03 -0700)]
mm/hwpoison: introduce num_poisoned_pages wrappers

num_poisoned_pages counter will be changed outside mm/memory-failure.c
by a subsequent patch, so this patch prepares wrappers to manipulate it.

Signed-off-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Tested-by: Wanpeng Li <wanpeng.li@hotmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm/hwpoison: replace most of put_page in memory error handling by put_hwpoison_page
Wanpeng Li [Tue, 8 Sep 2015 22:03:21 +0000 (15:03 -0700)]
mm/hwpoison: replace most of put_page in memory error handling by put_hwpoison_page

Replace most instances of put_page() in memory error handling with
put_hwpoison_page().

Signed-off-by: Wanpeng Li <wanpeng.li@hotmail.com>
Cc: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm/hwpoison: fix refcount of THP head page in no-injection case
Wanpeng Li [Tue, 8 Sep 2015 22:03:18 +0000 (15:03 -0700)]
mm/hwpoison: fix refcount of THP head page in no-injection case

Hwpoison injection takes a refcount of target page and another refcount
of head page of THP if the target page is the tail page of a THP.
However, current code doesn't release the refcount of head page if the
THP is not supported to be injected wrt hwpoison filter.

Fix it by reducing the refcount of head page if the target page is the
tail page of a THP and it is not supported to be injected.

Signed-off-by: Wanpeng Li <wanpeng.li@hotmail.com>
Acked-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm/hwpoison: introduce put_hwpoison_page to put refcount for memory error handling
Wanpeng Li [Tue, 8 Sep 2015 22:03:15 +0000 (15:03 -0700)]
mm/hwpoison: introduce put_hwpoison_page to put refcount for memory error handling

Introduce put_hwpoison_page to put refcount for memory error handling.

Signed-off-by: Wanpeng Li <wanpeng.li@hotmail.com>
Suggested-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Acked-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm/hwpoison: fix PageHWPoison test/set race
Wanpeng Li [Tue, 8 Sep 2015 22:03:13 +0000 (15:03 -0700)]
mm/hwpoison: fix PageHWPoison test/set race

There is a race between madvise_hwpoison path and memory_failure:

 CPU0 CPU1

madvise_hwpoison
get_user_pages_fast
PageHWPoison check (false)
memory_failure
TestSetPageHWPoison
soft_offline_page
PageHWPoison check (true)
return -EBUSY (without put_page)

Signed-off-by: Wanpeng Li <wanpeng.li@hotmail.com>
Suggested-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm/hwpoison: fix failure to split thp w/ refcount held
Wanpeng Li [Tue, 8 Sep 2015 22:03:10 +0000 (15:03 -0700)]
mm/hwpoison: fix failure to split thp w/ refcount held

THP pages will get a refcount in madvise_hwpoison() w/
MF_COUNT_INCREASED flag, however, the refcount is still held when fail
to split THP pages.

Fix it by reducing the refcount of THP pages when fail to split THP.

Signed-off-by: Wanpeng Li <wanpeng.li@hotmail.com>
Cc: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>