cascardo/linux.git
8 years agoget_maintainer: add subsystem to reviewer output
Joe Perches [Sat, 7 Nov 2015 00:30:52 +0000 (16:30 -0800)]
get_maintainer: add subsystem to reviewer output

Reviewer output currently does not include the subsystem
that matched.  Add it.

Miscellanea:

o Add a get_subsystem_name routine to centralize this

Signed-off-by: Joe Perches <joe@perches.com>
Tested-by: Krzysztof Kozlowski <k.kozlowski@samsung.com>
Cc: Lee Jones <lee.jones@linaro.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agoget_maintainer: --r (list reviewer) is on by default
Brian Norris [Sat, 7 Nov 2015 00:30:49 +0000 (16:30 -0800)]
get_maintainer: --r (list reviewer) is on by default

We don't consistenly document the default value next to the option
listing, but we do have a list of defaults here, so let's keep it up to
date.

Signed-off-by: Brian Norris <computersforpeace@gmail.com>
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>
8 years agoget_maintainer: add --no-foo options to --help
Brian Norris [Sat, 7 Nov 2015 00:30:46 +0000 (16:30 -0800)]
get_maintainer: add --no-foo options to --help

Many flag options are boolean and support both a positive and a negative
invocation from the command line.  Some of these are even mentioned by
example (e.g., --nogit is mentioned as a default option), but they aren't
explicitly mentioned in the list of options.  It happens that some of
these are pretty important, as they are default-on, and to turn them off,
you have to know about the --no-foo version.

Rather than clutter the whole help text with bracketed '--[no]foo', let's
just mention the general rule, a la 'man gcc'.

Signed-off-by: Brian Norris <computersforpeace@gmail.com>
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>
8 years agoget_maintainer: it's '--pattern-depth', not '-pattern-depth'
Brian Norris [Sat, 7 Nov 2015 00:30:43 +0000 (16:30 -0800)]
get_maintainer: it's '--pattern-depth', not '-pattern-depth'

Though it appears that Perl's GetOptions will take either, the latter is
not documented in the options listing.

Signed-off-by: Brian Norris <computersforpeace@gmail.com>
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>
8 years agoget_maintainer: add missing documentation for --git-blame-signatures
Brian Norris [Sat, 7 Nov 2015 00:30:41 +0000 (16:30 -0800)]
get_maintainer: add missing documentation for --git-blame-signatures

I really haven't used this option much myself, so feel free to improve on
the documentation for it.  I just noticed it while inspecting this script
for undocumented features.

Signed-off-by: Brian Norris <computersforpeace@gmail.com>
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>
8 years agoprintk: prevent userland from spoofing kernel messages
Mathias Krause [Sat, 7 Nov 2015 00:30:38 +0000 (16:30 -0800)]
printk: prevent userland from spoofing kernel messages

The following statement of ABI/testing/dev-kmsg is not quite right:

   It is not possible to inject messages from userspace with the
   facility number LOG_KERN (0), to make sure that the origin of the
   messages can always be reliably determined.

Userland actually can inject messages with a facility of 0 by abusing the
fact that the facility is stored in a u8 data type.  By using a facility
which is a multiple of 256 the assignment of msg->facility in log_store()
implicitly truncates it to 0, i.e.  LOG_KERN, allowing users of /dev/kmsg
to spoof kernel messages as shown below:

The following call...
   # printf '<%d>Kernel panic - not syncing: beer empty\n' 0 >/dev/kmsg
...leads to the following log entry (dmesg -x | tail -n 1):
   user  :emerg : [   66.137758] Kernel panic - not syncing: beer empty

However, this call...
   # printf '<%d>Kernel panic - not syncing: beer empty\n' 0x800 >/dev/kmsg
...leads to the slightly different log entry (note the kernel facility):
   kern  :emerg : [   74.177343] Kernel panic - not syncing: beer empty

Fix that by limiting the user provided facility to 8 bit right from the
beginning and catch the truncation early.

Fixes: 7ff9554bb578 ("printk: convert byte-buffer to variable-length...")
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Petr Mladek <pmladek@suse.cz>
Cc: Alex Elder <elder@linaro.org>
Cc: Joe Perches <joe@perches.com>
Cc: Kay Sievers <kay@vrfy.org>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agolib/vsprintf.c: update documentation
Rasmus Villemoes [Sat, 7 Nov 2015 00:30:35 +0000 (16:30 -0800)]
lib/vsprintf.c: update documentation

%n is no longer just ignored; it results in early return from vsnprintf.
Also add a request to add test cases for future %p extensions.

Signed-off-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Reviewed-by: Martin Kletzander <mkletzan@redhat.com>
Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agoselftests: run lib/test_printf module
Kees Cook [Sat, 7 Nov 2015 00:30:32 +0000 (16:30 -0800)]
selftests: run lib/test_printf module

This runs the lib/test_printf module to make sure printf is operating
sanely.

Signed-off-by: Kees Cook <keescook@chromium.org>
Cc: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Cc: Shuah Khan <shuahkh@osg.samsung.com>
Cc: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agotest_printf: test printf family at runtime
Rasmus Villemoes [Sat, 7 Nov 2015 00:30:29 +0000 (16:30 -0800)]
test_printf: test printf family at runtime

This adds a simple module for testing the kernel's printf facilities.
Previously, some %p extensions have caused a wrong return value in case
the entire output didn't fit and/or been unusable in kasprintf().  This
should help catch such issues.  Also, it should help ensure that changes
to the formatting algorithms don't break anything.

I'm not sure if we have a struct dentry or struct file lying around at
boot time or if we can fake one, but most %p extensions should be
testable, as should the ordinary number and string formatting.

The nature of vararg functions means we can't use a more conventional
table-driven approach.

For now, this is mostly a skeleton; contributions are very
welcome. Some tests are/will be slightly annoying to write, since the
expected output depends on stuff like CONFIG_*, sizeof(long), runtime
values etc.

Signed-off-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Reviewed-by: Kees Cook <keescook@chromium.org>
Cc: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Cc: Martin Kletzander <mkletzan@redhat.com>
Cc: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agolib/vsprintf.c: remove SPECIAL handling in pointer()
Rasmus Villemoes [Sat, 7 Nov 2015 00:30:26 +0000 (16:30 -0800)]
lib/vsprintf.c: remove SPECIAL handling in pointer()

As a quick

   git grep -E '%[ +0#-]*#[ +0#-]*(\*|[0-9]+)?(\.(\*|[0-9]+)?)?p'

shows, nobody uses the # flag with %p. Should one try to do so, one
will be met with

  warning: `#' flag used with `%p' gnu_printf format [-Wformat]

(POSIX and C99 both say "... For other conversion specifiers, the
behavior is undefined.". Obviously, the kernel can choose to define
the behaviour however it wants, but as long as gcc issues that
warning, users are unlikely to show up.)

Since default_width is effectively always 2*sizeof(void*), we can
simplify the prologue of pointer() and save a few instructions.

Signed-off-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Acked-by: Kees Cook <keescook@chromium.org>
Cc: Martin Kletzander <mkletzan@redhat.com>
Cc: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agolib/vsprintf.c: also improve sanity check in bstr_printf()
Rasmus Villemoes [Sat, 7 Nov 2015 00:30:23 +0000 (16:30 -0800)]
lib/vsprintf.c: also improve sanity check in bstr_printf()

Quoting from 2aa2f9e21e4e ("lib/vsprintf.c: improve sanity check in
vsnprintf()"):

    On 64 bit, size may very well be huge even if bit 31 happens to be 0.
    Somehow it doesn't feel right that one can pass a 5 GiB buffer but not a
    3 GiB one.  So cap at INT_MAX as was probably the intention all along.
    This is also the made-up value passed by sprintf and vsprintf.

I should have seen this copy-pasted instance back then, but let's just
do it now.

Signed-off-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Acked-by: Kees Cook <keescook@chromium.org>
Cc: Martin Kletzander <mkletzan@redhat.com>
Cc: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agolib/vsprintf.c: handle invalid format specifiers more robustly
Rasmus Villemoes [Sat, 7 Nov 2015 00:30:20 +0000 (16:30 -0800)]
lib/vsprintf.c: handle invalid format specifiers more robustly

If we meet any invalid or unsupported format specifier, 'handling' it by
just printing it as a literal string is not safe: Presumably the format
string and the arguments passed gcc's type checking, but that means
something like sprintf(buf, "%n %pd", &intvar, dentry) would end up
interpreting &intvar as a struct dentry*.

When the offending specifier was %n it used to be at the end of the format
string, but we can't rely on that always being the case.  Also, gcc
doesn't complain about some more or less exotic qualifiers (or 'length
modifiers' in posix-speak) such as 'j' or 'q', but being unrecognized by
the kernel's printf implementation, they'd be interpreted as unknown
specifiers, and the rest of arguments would be interpreted wrongly.

So let's complain about anything we don't understand, not just %n, and
stop pretending that we'd be able to make sense of the rest of the
format/arguments.  If the offending specifier is in a printk() call we
unfortunately only get a "BUG: recent printk recursion!", but at least
direct users of the sprintf family will be caught.

Signed-off-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Acked-by: Kees Cook <keescook@chromium.org>
Cc: Martin Kletzander <mkletzan@redhat.com>
Cc: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agoprintk: synchronize %p formatting documentation
Martin Kletzander [Sat, 7 Nov 2015 00:30:17 +0000 (16:30 -0800)]
printk: synchronize %p formatting documentation

Move all pointer-formatting documentation to one place in the code and one
place in the documentation instead of keeping it in three places with
different level of completeness.  Documentation/printk-formats.txt has
detailed information about each modifier, docstring above pointer() has
short descriptions of them (as that is the function dealing with %p) and
docstring above vsprintf() is removed as redundant.  Both docstrings in
the code that were modified are updated with a reminder of updating the
documentation upon any further change.

[akpm@linux-foundation.org: fix comment]
Signed-off-by: Martin Kletzander <mkletzan@redhat.com>
Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Cc: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Cc: Jonathan Corbet <corbet@lwn.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agolib/dynamic_debug.c: use kstrdup_const
Rasmus Villemoes [Sat, 7 Nov 2015 00:30:15 +0000 (16:30 -0800)]
lib/dynamic_debug.c: use kstrdup_const

Using kstrdup_const, thus reusing .rodata when possible, saves around 2 kB
of runtime memory on my laptop/.config combination.

Signed-off-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Cc: Jason Baron <jbaron@akamai.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agofs/jffs2/wbuf.c: remove stray semicolon
Andrew Morton [Sat, 7 Nov 2015 00:30:12 +0000 (16:30 -0800)]
fs/jffs2/wbuf.c: remove stray semicolon

Reported-by: Wu Fengguang <fengguang.wu@intel.com>
Cc: Sasha Levin <sasha.levin@oracle.com>
Cc: David Woodhouse <dwmw2@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agoinclude/linux/compiler-gcc.h: improve __visible documentation
Andrew Morton [Sat, 7 Nov 2015 00:30:09 +0000 (16:30 -0800)]
include/linux/compiler-gcc.h: improve __visible documentation

Cc: Andi Kleen <andi@firstfloor.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agoproc: actually make proc_fd_permission() thread-friendly
Oleg Nesterov [Sat, 7 Nov 2015 00:30:06 +0000 (16:30 -0800)]
proc: actually make proc_fd_permission() thread-friendly

The commit 96d0df79f264 ("proc: make proc_fd_permission() thread-friendly")
fixed the access to /proc/self/fd from sub-threads, but introduced another
problem: a sub-thread can't access /proc/<tid>/fd/ or /proc/thread-self/fd
if generic_permission() fails.

Change proc_fd_permission() to check same_thread_group(pid_task(), current).

Fixes: 96d0df79f264 ("proc: make proc_fd_permission() thread-friendly")
Reported-by: "Jin, Yihua" <yihua.jin@intel.com>
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agofs/proc/array.c: set overflow flag in case of error
Andy Shevchenko [Sat, 7 Nov 2015 00:30:03 +0000 (16:30 -0800)]
fs/proc/array.c: set overflow flag in case of error

For now in task_name() we ignore the return code of string_escape_str()
call.  This is not good if buffer suddenly becomes not big enough.  Do the
proper error handling there.

Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agomm: use 'unsigned int' for compound_dtor/compound_order on 64BIT
Kirill A. Shutemov [Sat, 7 Nov 2015 00:30:00 +0000 (16:30 -0800)]
mm: use 'unsigned int' for compound_dtor/compound_order on 64BIT

On 64 bit system we have enough space in struct page to encode
compound_dtor and compound_order with unsigned int.

On x86-64 it leads to slightly smaller code size due usesage of plain
MOV instead of MOVZX (zero-extended move) or similar effect.

allyesconfig:

   text    data     bss     dec     hex filename
159520446 48146736 72196096 279863278 10ae5fee vmlinux.pre
159520382 48146736 72196096 279863214 10ae5fae vmlinux.post

On other architectures without native support of 16-bit data types the

Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Reviewed-by: Andrea Arcangeli <aarcange@redhat.com>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Sergey Senozhatsky <sergey.senozhatsky@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>
8 years agomm: use 'unsigned int' for page order
Kirill A. Shutemov [Sat, 7 Nov 2015 00:29:57 +0000 (16:29 -0800)]
mm: use 'unsigned int' for page order

Let's try to be consistent about data type of page order.

[sfr@canb.auug.org.au: fix build (type of pageblock_order)]
[hughd@google.com: some configs end up with MAX_ORDER and pageblock_order having different types]
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Reviewed-by: Andrea Arcangeli <aarcange@redhat.com>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
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>
8 years agomm: make compound_head() robust
Kirill A. Shutemov [Sat, 7 Nov 2015 00:29:54 +0000 (16:29 -0800)]
mm: make compound_head() robust

Hugh has pointed that compound_head() call can be unsafe in some
context. There's one example:

CPU0 CPU1

isolate_migratepages_block()
  page_count()
    compound_head()
      !!PageTail() == true
put_page()
  tail->first_page = NULL
      head = tail->first_page
alloc_pages(__GFP_COMP)
   prep_compound_page()
     tail->first_page = head
     __SetPageTail(p);
      !!PageTail() == true
    <head == NULL dereferencing>

The race is pure theoretical. I don't it's possible to trigger it in
practice. But who knows.

We can fix the race by changing how encode PageTail() and compound_head()
within struct page to be able to update them in one shot.

The patch introduces page->compound_head into third double word block in
front of compound_dtor and compound_order. Bit 0 encodes PageTail() and
the rest bits are pointer to head page if bit zero is set.

The patch moves page->pmd_huge_pte out of word, just in case if an
architecture defines pgtable_t into something what can have the bit 0
set.

hugetlb_cgroup uses page->lru.next in the second tail page to store
pointer struct hugetlb_cgroup. The patch switch it to use page->private
in the second tail page instead. The space is free since ->first_page is
removed from the union.

The patch also opens possibility to remove HUGETLB_CGROUP_MIN_ORDER
limitation, since there's now space in first tail page to store struct
hugetlb_cgroup pointer. But that's out of scope of the patch.

That means page->compound_head shares storage space with:

 - page->lru.next;
 - page->next;
 - page->rcu_head.next;

That's too long list to be absolutely sure, but looks like nobody uses
bit 0 of the word.

page->rcu_head.next guaranteed[1] to have bit 0 clean as long as we use
call_rcu(), call_rcu_bh(), call_rcu_sched(), or call_srcu(). But future
call_rcu_lazy() is not allowed as it makes use of the bit and we can
get false positive PageTail().

[1] http://lkml.kernel.org/g/20150827163634.GD4029@linux.vnet.ibm.com

Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Reviewed-by: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agomm: pack compound_dtor and compound_order into one word in struct page
Kirill A. Shutemov [Sat, 7 Nov 2015 00:29:50 +0000 (16:29 -0800)]
mm: pack compound_dtor and compound_order into one word in struct page

The patch halves space occupied by compound_dtor and compound_order in
struct page.

For compound_order, it's trivial long -> short conversion.

For get_compound_page_dtor(), we now use hardcoded table for destructor
lookup and store its index in the struct page instead of direct pointer
to destructor. It shouldn't be a big trouble to maintain the table: we
have only two destructor and NULL currently.

This patch free up one word in tail pages for reuse. This is preparation
for the next patch.

Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Reviewed-by: Michal Hocko <mhocko@suse.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Reviewed-by: Andrea Arcangeli <aarcange@redhat.com>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agozsmalloc: use page->private instead of page->first_page
Kirill A. Shutemov [Sat, 7 Nov 2015 00:29:47 +0000 (16:29 -0800)]
zsmalloc: use page->private instead of page->first_page

We are going to rework how compound_head() work. It will not use
page->first_page as we have it now.

The only other user of page->first_page beyond compound pages is
zsmalloc.

Let's use page->private instead of page->first_page here. It occupies
the same storage space.

Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Reviewed-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Reviewed-by: Andrea Arcangeli <aarcange@redhat.com>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.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>
8 years agoslab, slub: use page->rcu_head instead of page->lru plus cast
Kirill A. Shutemov [Sat, 7 Nov 2015 00:29:44 +0000 (16:29 -0800)]
slab, slub: use page->rcu_head instead of page->lru plus cast

We have properly typed page->rcu_head, no need to cast page->lru.

Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Reviewed-by: Andrea Arcangeli <aarcange@redhat.com>
Acked-by: Christoph Lameter <cl@linux.com>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Sergey Senozhatsky <sergey.senozhatsky@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>
8 years agomm: drop page->slab_page
Kirill A. Shutemov [Sat, 7 Nov 2015 00:29:40 +0000 (16:29 -0800)]
mm: drop page->slab_page

Since 8456a648cf44 ("slab: use struct page for slab management") nobody
uses slab_page field in struct page.

Let's drop it.

Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Acked-by: Christoph Lameter <cl@linux.com>
Acked-by: David Rientjes <rientjes@google.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Reviewed-by: Andrea Arcangeli <aarcange@redhat.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Cc: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agozsmalloc: reduce size_class memory usage
Sergey Senozhatsky [Sat, 7 Nov 2015 00:29:38 +0000 (16:29 -0800)]
zsmalloc: reduce size_class memory usage

Each `struct size_class' contains `struct zs_size_stat': an array of
NR_ZS_STAT_TYPE `unsigned long'.  For zsmalloc built with no
CONFIG_ZSMALLOC_STAT this results in a waste of `2 * sizeof(unsigned
long)' per-class.

The patch removes unneeded `struct zs_size_stat' members by redefining
NR_ZS_STAT_TYPE (max stat idx in array).

Since both NR_ZS_STAT_TYPE and zs_stat_type are compile time constants,
GCC can eliminate zs_stat_inc()/zs_stat_dec() calls that use zs_stat_type
larger than NR_ZS_STAT_TYPE: CLASS_ALMOST_EMPTY and CLASS_ALMOST_FULL at
the moment.

./scripts/bloat-o-meter mm/zsmalloc.o.old mm/zsmalloc.o.new
add/remove: 0/0 grow/shrink: 0/3 up/down: 0/-39 (-39)
function                                     old     new   delta
fix_fullness_group                            97      94      -3
insert_zspage                                100      86     -14
remove_zspage                                141     119     -22

To summarize:
a) each class now uses less memory
b) we avoid a number of dec/inc stats (a minor optimization,
   but still).

The gain will increase once we introduce additional stats.

A simple IO test.

iozone -t 4 -R -r 32K -s 60M -I +Z
                        patched                 base
"  Initial write "       4145599.06              4127509.75
"        Rewrite "       4146225.94              4223618.50
"           Read "      17157606.00             17211329.50
"        Re-read "      17380428.00             17267650.50
"   Reverse Read "      16742768.00             16162732.75
"    Stride read "      16586245.75             16073934.25
"    Random read "      16349587.50             15799401.75
" Mixed workload "      10344230.62              9775551.50
"   Random write "       4277700.62              4260019.69
"         Pwrite "       4302049.12              4313703.88
"          Pread "       6164463.16              6126536.72
"         Fwrite "       7131195.00              6952586.00
"          Fread "      12682602.25             12619207.50

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>
8 years agomm/zsmalloc.c: remove useless line in obj_free()
Hui Zhu [Sat, 7 Nov 2015 00:29:35 +0000 (16:29 -0800)]
mm/zsmalloc.c: remove useless line in obj_free()

Signed-off-by: Hui Zhu <zhuhui@xiaomi.com>
Reviewed-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>
8 years agozsmalloc: don't test shrinker_enabled in zs_shrinker_count()
Sergey Senozhatsky [Sat, 7 Nov 2015 00:29:32 +0000 (16:29 -0800)]
zsmalloc: don't test shrinker_enabled in zs_shrinker_count()

We don't let user to disable shrinker in zsmalloc (once it's been
enabled), so no need to check ->shrinker_enabled in zs_shrinker_count(),
at the moment at least.

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>
8 years agozsmalloc: use preempt.h for in_interrupt()
Sergey Senozhatsky [Sat, 7 Nov 2015 00:29:29 +0000 (16:29 -0800)]
zsmalloc: use preempt.h for in_interrupt()

A cosmetic change.

Commit c60369f01125 ("staging: zsmalloc: prevent mappping in interrupt
context") added in_interrupt() check to zs_map_object() and 'hardirq.h'
include; but in_interrupt() macro is defined in 'preempt.h' not in
'hardirq.h', so include it instead.

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>
8 years agozsmalloc: fix obj_to_head use page_private(page) as value but not pointer
Hui Zhu [Sat, 7 Nov 2015 00:29:26 +0000 (16:29 -0800)]
zsmalloc: fix obj_to_head use page_private(page) as value but not pointer

In obj_malloc():

if (!class->huge)
/* record handle in the header of allocated chunk */
link->handle = handle;
else
/* record handle in first_page->private */
set_page_private(first_page, handle);

In the hugepage we save handle to private directly.

But in obj_to_head():

if (class->huge) {
VM_BUG_ON(!is_first_page(page));
return *(unsigned long *)page_private(page);
} else
return *(unsigned long *)obj;

It is used as a pointer.

The reason why there is no problem until now is huge-class page is born
with ZS_FULL so it can't be migrated.  However, we need this patch for
future work: "VM-aware zsmalloced page migration" to reduce external
fragmentation.

Signed-off-by: Hui Zhu <zhuhui@xiaomi.com>
Acked-by: Minchan Kim <minchan@kernel.org>
Cc: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agozsmalloc: add comments for ->inuse to zspage
Hui Zhu [Sat, 7 Nov 2015 00:29:23 +0000 (16:29 -0800)]
zsmalloc: add comments for ->inuse to zspage

[akpm@linux-foundation.org: fix grammar]
Signed-off-by: Hui Zhu <zhuhui@xiaomi.com>
Reviewed-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Cc: Dan Streetman <ddstreet@ieee.org>
Cc: Minchan Kim <minchan@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agomm: zsmalloc: constify struct zs_pool name
Sergey SENOZHATSKY [Sat, 7 Nov 2015 00:29:21 +0000 (16:29 -0800)]
mm: zsmalloc: constify struct zs_pool name

Constify `struct zs_pool' ->name.

[akpm@inux-foundation.org: constify zpool_create_pool()'s `type' arg also]
Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Acked-by: Dan Streetman <ddstreet@ieee.org>
Cc: Minchan Kim <minchan@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agozpool: remove redundant zpool->type string, const-ify zpool_get_type
Dan Streetman [Sat, 7 Nov 2015 00:29:18 +0000 (16:29 -0800)]
zpool: remove redundant zpool->type string, const-ify zpool_get_type

Make the return type of zpool_get_type const; the string belongs to the
zpool driver and should not be modified.  Remove the redundant type field
in the struct zpool; it is private to zpool.c and isn't needed since
->driver->type can be used directly.  Add comments indicating strings must
be null-terminated.

Signed-off-by: Dan Streetman <ddstreet@ieee.org>
Cc: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Cc: Seth Jennings <sjennings@variantweb.net>
Cc: Minchan Kim <minchan@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agozswap: use charp for zswap param strings
Dan Streetman [Sat, 7 Nov 2015 00:29:15 +0000 (16:29 -0800)]
zswap: use charp for zswap param strings

Instead of using a fixed-length string for the zswap params, use charp.
This simplifies the code and uses less memory, as most zswap param strings
will be less than the current maximum length.

Signed-off-by: Dan Streetman <ddstreet@ieee.org>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Seth Jennings <sjennings@variantweb.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agomodule: export param_free_charp()
Dan Streetman [Sat, 7 Nov 2015 00:29:12 +0000 (16:29 -0800)]
module: export param_free_charp()

Change the param_free_charp() function from static to exported.

It is used by zswap in the next patch ("zswap: use charp for zswap param
strings").

Signed-off-by: Dan Streetman <ddstreet@ieee.org>
Acked-by: Rusty Russell <rusty@rustcorp.com.au>
Cc: Seth Jennings <sjennings@variantweb.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agomm/zswap.c: remove unneeded initialization to NULL in zswap_entry_find_get()
Alexey Klimov [Sat, 7 Nov 2015 00:29:09 +0000 (16:29 -0800)]
mm/zswap.c: remove unneeded initialization to NULL in zswap_entry_find_get()

On the next line entry variable will be re-initialized so no need to init
it with NULL.

Signed-off-by: Alexey Klimov <alexey.klimov@linaro.org>
Cc: Seth Jennings <sjennings@variantweb.net>
Cc: Dan Streetman <ddstreet@ieee.org>
Cc: Minchan Kim <minchan@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agozram: make is_partial_io/valid_io_request/page_zero_filled return boolean
Geliang Tang [Sat, 7 Nov 2015 00:29:06 +0000 (16:29 -0800)]
zram: make is_partial_io/valid_io_request/page_zero_filled return boolean

Make is_partial_io()/valid_io_request()/page_zero_filled() return boolean,
since each function only uses either one or zero as its return value.

Signed-off-by: Geliang Tang <geliangtang@163.com>
Reviewed-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>
8 years agozram: keep the exact overcommited value in mem_used_max
Sergey SENOZHATSKY [Sat, 7 Nov 2015 00:29:04 +0000 (16:29 -0800)]
zram: keep the exact overcommited value in mem_used_max

`mem_used_max' is designed to store the max amount of memory zram consumed
to store the data.  However, it does not represent the actual
'overcommited' (max) value.  The existing code goes to -ENOMEM
overcommited case before it updates `->stats.max_used_pages', which hides
the reason we went to -ENOMEM in the first place -- we actually used more
memory than `->limit_pages':

        alloced_pages = zs_get_total_pages(meta->mem_pool);
        if (zram->limit_pages && alloced_pages > zram->limit_pages) {
                zs_free(meta->mem_pool, handle);
                ret = -ENOMEM;
                goto out;
        }

        update_used_max(zram, alloced_pages);

Which is misleading.  User will see -ENOMEM, check `->limit_pages', check
`->stats.max_used_pages', which will keep the value BEFORE zram passed
`->limit_pages', and see:
`->stats.max_used_pages' < `->limit_pages'

Move update_used_max() before we do `->limit_pages' check, so that
user will see:
`->stats.max_used_pages' > `->limit_pages'
should the overcommit and -ENOMEM happen.

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>
8 years agozram: introduce comp algorithm fallback functionality
Luis Henriques [Sat, 7 Nov 2015 00:29:01 +0000 (16:29 -0800)]
zram: introduce comp algorithm fallback functionality

When the user supplies an unsupported compression algorithm, keep the
previously selected one (knowingly supported) or the default one (if the
compression algorithm hasn't been changed yet).

Note that previously this operation (i.e. setting an invalid algorithm)
would result in no algorithm being selected, which means that this
represents a small change in the default behaviour.

Minchan said:

For initializing zram, we need to set up 3 optional parameters in advance.

1. the number of compression streams
2. memory limitation
3. compression algorithm

Although user pass completely wrong value to set up for 1 and 2
parameters, it's okay because they have default value so zram will be
initialized with the default value (of course, when user passes a wrong
value via *echo*, sysfs returns -EINVAL so the user can notice it).

But 3 is not consistent with other optional parameters.  IOW, if the
user passes a wrong value to set up 3 parameter, zram's initialization
would fail unlike other optional parameters.

So this patch makes them consistent.

Signed-off-by: Luis Henriques <luis.henriques@canonical.com>
Acked-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>
8 years agomm/memcontrol.c: uninline mem_cgroup_usage
Andrew Morton [Sat, 7 Nov 2015 00:28:58 +0000 (16:28 -0800)]
mm/memcontrol.c: uninline mem_cgroup_usage

gcc version 5.2.1 20151010 (Debian 5.2.1-22)
$ size mm/memcontrol.o mm/memcontrol.o.before
   text    data     bss     dec     hex filename
  35535    7908      64   43507    a9f3 mm/memcontrol.o
  35762    7908      64   43734    aad6 mm/memcontrol.o.before

Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agofs/sync.c: make sync_file_range(2) use WB_SYNC_NONE writeback
Jan Kara [Sat, 7 Nov 2015 00:28:55 +0000 (16:28 -0800)]
fs/sync.c: make sync_file_range(2) use WB_SYNC_NONE writeback

sync_file_range(2) is documented to issue writeback only for pages that
are not currently being written.  After all the system call has been
created for userspace to be able to issue background writeout and so
waiting for in-flight IO is undesirable there.  However commit
ee53a891f474 ("mm: do_sync_mapping_range integrity fix") switched
do_sync_mapping_range() and thus sync_file_range() to issue writeback in
WB_SYNC_ALL mode since do_sync_mapping_range() was used by other code
relying on WB_SYNC_ALL semantics.

These days do_sync_mapping_range() went away and we can switch
sync_file_range(2) back to issuing WB_SYNC_NONE writeback.  That should
help PostgreSQL avoid large latency spikes when flushing data in the
background.

Andres measured a 20% increase in transactions per second on an SSD disk.

Signed-off-by: Jan Kara <jack@suse.com>
Reported-by: Andres Freund <andres@anarazel.de>
Tested-By: Andres Freund <andres@anarazel.de>
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agothp: remove unused vma parameter from khugepaged_alloc_page
Aaron Tomlin [Sat, 7 Nov 2015 00:28:52 +0000 (16:28 -0800)]
thp: remove unused vma parameter from khugepaged_alloc_page

The "vma" parameter to khugepaged_alloc_page() is unused.  It has to
remain unused or the drop read lock 'map_sem' optimisation introduce by
commit 8b1645685acf ("mm, THP: don't hold mmap_sem in khugepaged when
allocating THP") wouldn't be safe.  So let's remove it.

Signed-off-by: Aaron Tomlin <atomlin@redhat.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agomm, fs: introduce mapping_gfp_constraint()
Michal Hocko [Sat, 7 Nov 2015 00:28:49 +0000 (16:28 -0800)]
mm, fs: introduce mapping_gfp_constraint()

There are many places which use mapping_gfp_mask to restrict a more
generic gfp mask which would be used for allocations which are not
directly related to the page cache but they are performed in the same
context.

Let's introduce a helper function which makes the restriction explicit and
easier to track.  This patch doesn't introduce any functional changes.

[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Michal Hocko <mhocko@suse.com>
Suggested-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agoinclude/linux/mmzone.h: reflow comment
Andrew Morton [Sat, 7 Nov 2015 00:28:46 +0000 (16:28 -0800)]
include/linux/mmzone.h: reflow comment

Someone has an 86 column display.

Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agomm: page_alloc: hide some GFP internals and document the bits and flag combinations
Mel Gorman [Sat, 7 Nov 2015 00:28:43 +0000 (16:28 -0800)]
mm: page_alloc: hide some GFP internals and document the bits and flag combinations

Andrew stated the following

We have quite a history of remote parts of the kernel using
weird/wrong/inexplicable combinations of __GFP_ flags. I tend
to think that this is because we didn't adequately explain the
interface.

And I don't think that gfp.h really improved much in this area as
a result of this patchset.  Could you go through it some time and
decide if we've adequately documented all this stuff?

This patches first moves some GFP flag combinations that are part of the MM
internals to mm/internal.h. The rest of the patch documents the __GFP_FOO
bits under various headings and then documents the flag combinations. It
will not help callers that are brain damaged but the clarity might motivate
some fixes and avoid future mistakes.

Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: David Rientjes <rientjes@google.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Vitaly Wool <vitalywool@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agomm, page_alloc: only enforce watermarks for order-0 allocations
Mel Gorman [Sat, 7 Nov 2015 00:28:40 +0000 (16:28 -0800)]
mm, page_alloc: only enforce watermarks for order-0 allocations

The primary purpose of watermarks is to ensure that reclaim can always
make forward progress in PF_MEMALLOC context (kswapd and direct reclaim).
These assume that order-0 allocations are all that is necessary for
forward progress.

High-order watermarks serve a different purpose.  Kswapd had no high-order
awareness before they were introduced
(https://lkml.kernel.org/r/413AA7B2.4000907@yahoo.com.au).  This was
particularly important when there were high-order atomic requests.  The
watermarks both gave kswapd awareness and made a reserve for those atomic
requests.

There are two important side-effects of this.  The most important is that
a non-atomic high-order request can fail even though free pages are
available and the order-0 watermarks are ok.  The second is that
high-order watermark checks are expensive as the free list counts up to
the requested order must be examined.

With the introduction of MIGRATE_HIGHATOMIC it is no longer necessary to
have high-order watermarks.  Kswapd and compaction still need high-order
awareness which is handled by checking that at least one suitable
high-order page is free.

With the patch applied, there was little difference in the allocation
failure rates as the atomic reserves are small relative to the number of
allocation attempts.  The expected impact is that there will never be an
allocation failure report that shows suitable pages on the free lists.

The one potential side-effect of this is that in a vanilla kernel, the
watermark checks may have kept a free page for an atomic allocation.  Now,
we are 100% relying on the HighAtomic reserves and an early allocation to
have allocated them.  If the first high-order atomic allocation is after
the system is already heavily fragmented then it'll fail.

[akpm@linux-foundation.org: simplify __zone_watermark_ok(), per Vlastimil]
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Christoph Lameter <cl@linux.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Vitaly Wool <vitalywool@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agomm, page_alloc: reserve pageblocks for high-order atomic allocations on demand
Mel Gorman [Sat, 7 Nov 2015 00:28:37 +0000 (16:28 -0800)]
mm, page_alloc: reserve pageblocks for high-order atomic allocations on demand

High-order watermark checking exists for two reasons -- kswapd high-order
awareness and protection for high-order atomic requests.  Historically the
kernel depended on MIGRATE_RESERVE to preserve min_free_kbytes as
high-order free pages for as long as possible.  This patch introduces
MIGRATE_HIGHATOMIC that reserves pageblocks for high-order atomic
allocations on demand and avoids using those blocks for order-0
allocations.  This is more flexible and reliable than MIGRATE_RESERVE was.

A MIGRATE_HIGHORDER pageblock is created when an atomic high-order
allocation request steals a pageblock but limits the total number to 1% of
the zone.  Callers that speculatively abuse atomic allocations for
long-lived high-order allocations to access the reserve will quickly fail.
 Note that SLUB is currently not such an abuser as it reclaims at least
once.  It is possible that the pageblock stolen has few suitable
high-order pages and will need to steal again in the near future but there
would need to be strong justification to search all pageblocks for an
ideal candidate.

The pageblocks are unreserved if an allocation fails after a direct
reclaim attempt.

The watermark checks account for the reserved pageblocks when the
allocation request is not a high-order atomic allocation.

The reserved pageblocks can not be used for order-0 allocations.  This may
allow temporary wastage until a failed reclaim reassigns the pageblock.
This is deliberate as the intent of the reservation is to satisfy a
limited number of atomic high-order short-lived requests if the system
requires them.

The stutter benchmark was used to evaluate this but while it was running
there was a systemtap script that randomly allocated between 1 high-order
page and 12.5% of memory's worth of order-3 pages using GFP_ATOMIC.  This
is much larger than the potential reserve and it does not attempt to be
realistic.  It is intended to stress random high-order allocations from an
unknown source, show that there is a reduction in failures without
introducing an anomaly where atomic allocations are more reliable than
regular allocations.  The amount of memory reserved varied throughout the
workload as reserves were created and reclaimed under memory pressure.
The allocation failures once the workload warmed up were as follows;

4.2-rc5-vanilla 70%
4.2-rc5-atomic-reserve 56%

The failure rate was also measured while building multiple kernels.  The
failure rate was 14% but is 6% with this patch applied.

Overall, this is a small reduction but the reserves are small relative to
the number of allocation requests.  In early versions of the patch, the
failure rate reduced by a much larger amount but that required much larger
reserves and perversely made atomic allocations seem more reliable than
regular allocations.

[yalin.wang2010@gmail.com: fix redundant check and a memory leak]
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Christoph Lameter <cl@linux.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Vitaly Wool <vitalywool@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: yalin wang <yalin.wang2010@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agomm, page_alloc: remove MIGRATE_RESERVE
Mel Gorman [Sat, 7 Nov 2015 00:28:34 +0000 (16:28 -0800)]
mm, page_alloc: remove MIGRATE_RESERVE

MIGRATE_RESERVE preserves an old property of the buddy allocator that
existed prior to fragmentation avoidance -- min_free_kbytes worth of pages
tended to remain contiguous until the only alternative was to fail the
allocation.  At the time it was discovered that high-order atomic
allocations relied on this property so MIGRATE_RESERVE was introduced.  A
later patch will introduce an alternative MIGRATE_HIGHATOMIC so this patch
deletes MIGRATE_RESERVE and supporting code so it'll be easier to review.
Note that this patch in isolation may look like a false regression if
someone was bisecting high-order atomic allocation failures.

Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Christoph Lameter <cl@linux.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Vitaly Wool <vitalywool@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agomm, page_alloc: delete the zonelist_cache
Mel Gorman [Sat, 7 Nov 2015 00:28:31 +0000 (16:28 -0800)]
mm, page_alloc: delete the zonelist_cache

The zonelist cache (zlc) was introduced to skip over zones that were
recently known to be full.  This avoided expensive operations such as the
cpuset checks, watermark calculations and zone_reclaim.  The situation
today is different and the complexity of zlc is harder to justify.

1) The cpuset checks are no-ops unless a cpuset is active and in general
   are a lot cheaper.

2) zone_reclaim is now disabled by default and I suspect that was a large
   source of the cost that zlc wanted to avoid. When it is enabled, it's
   known to be a major source of stalling when nodes fill up and it's
   unwise to hit every other user with the overhead.

3) Watermark checks are expensive to calculate for high-order
   allocation requests. Later patches in this series will reduce the cost
   of the watermark checking.

4) The most important issue is that in the current implementation it
   is possible for a failed THP allocation to mark a zone full for order-0
   allocations and cause a fallback to remote nodes.

The last issue could be addressed with additional complexity but as the
benefit of zlc is questionable, it is better to remove it.  If stalls due
to zone_reclaim are ever reported then an alternative would be to
introduce deferring logic based on a timeout inside zone_reclaim itself
and leave the page allocator fast paths alone.

The impact on page-allocator microbenchmarks is negligible as they don't
hit the paths where the zlc comes into play.  Most page-reclaim related
workloads showed no noticeable difference as a result of the removal.

The impact was noticeable in a workload called "stutter".  One part uses a
lot of anonymous memory, a second measures mmap latency and a third copies
a large file.  In an ideal world the latency application would not notice
the mmap latency.  On a 2-node machine the results of this patch are

stutter
                             4.3.0-rc1             4.3.0-rc1
                              baseline              nozlc-v4
Min         mmap     20.9243 (  0.00%)     20.7716 (  0.73%)
1st-qrtle   mmap     22.0612 (  0.00%)     22.0680 ( -0.03%)
2nd-qrtle   mmap     22.3291 (  0.00%)     22.3809 ( -0.23%)
3rd-qrtle   mmap     25.2244 (  0.00%)     25.2396 ( -0.06%)
Max-90%     mmap     48.0995 (  0.00%)     28.3713 ( 41.02%)
Max-93%     mmap     52.5557 (  0.00%)     36.0170 ( 31.47%)
Max-95%     mmap     55.8173 (  0.00%)     47.3163 ( 15.23%)
Max-99%     mmap     67.3781 (  0.00%)     70.1140 ( -4.06%)
Max         mmap  24447.6375 (  0.00%)  12915.1356 ( 47.17%)
Mean        mmap     33.7883 (  0.00%)     27.7944 ( 17.74%)
Best99%Mean mmap     27.7825 (  0.00%)     25.2767 (  9.02%)
Best95%Mean mmap     26.3912 (  0.00%)     23.7994 (  9.82%)
Best90%Mean mmap     24.9886 (  0.00%)     23.2251 (  7.06%)
Best50%Mean mmap     22.0157 (  0.00%)     22.0261 ( -0.05%)
Best10%Mean mmap     21.6705 (  0.00%)     21.6083 (  0.29%)
Best5%Mean  mmap     21.5581 (  0.00%)     21.4611 (  0.45%)
Best1%Mean  mmap     21.3079 (  0.00%)     21.1631 (  0.68%)

Note that the maximum stall latency went from 24 seconds to 12 which is
still bad but an improvement.  The milage varies considerably 2-node
machine on an earlier test went from 494 seconds to 47 seconds and a
4-node machine that tested an earlier version of this patch went from a
worst case stall time of 6 seconds to 67ms.  The nature of the benchmark
is inherently unpredictable as it is hammering the system and the milage
will vary between machines.

There is a secondary impact with potentially more direct reclaim because
zones are now being considered instead of being skipped by zlc.  In this
particular test run it did not occur so will not be described.  However,
in at least one test the following was observed

1. Direct reclaim rates were higher. This was likely due to direct reclaim
  being entered instead of the zlc disabling a zone and busy looping.
  Busy looping may have the effect of allowing kswapd to make more
  progress and in some cases may be better overall. If this is found then
  the correct action is to put direct reclaimers to sleep on a waitqueue
  and allow kswapd make forward progress. Busy looping on the zlc is even
  worse than when the allocator used to blindly call congestion_wait().

2. There was higher swap activity as direct reclaim was active.

3. Direct reclaim efficiency was lower. This is related to 1 as more
  scanning activity also encountered more pages that could not be
  immediately reclaimed

In that case, the direct page scan and reclaim rates are noticeable but
it is not considered a problem for a few reasons

1. The test is primarily concerned with latency. The mmap attempts are also
   faulted which means there are THP allocation requests. The ZLC could
   cause zones to be disabled causing the process to busy loop instead
   of reclaiming.  This looks like elevated direct reclaim activity but
   it's the correct action to take based on what processes requested.

2. The test hammers reclaim and compaction heavily. The number of successful
   THP faults is highly variable but affects the reclaim stats. It's not a
   realistic or reasonable measure of page reclaim activity.

3. No other page-reclaim intensive workload that was tested showed a problem.

4. If a workload is identified that benefitted from the busy looping then it
   should be fixed by having direct reclaimers sleep on a wait queue until
   woken by kswapd instead of busy looping. We had this class of problem before
   when congestion_waits() with a fixed timeout was a brain damaged decision
   but happened to benefit some workloads.

If a workload is identified that relied on the zlc to busy loop then it
should be fixed correctly and have a direct reclaimer sleep on a waitqueue
until woken by kswapd.

Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: David Rientjes <rientjes@google.com>
Acked-by: Christoph Lameter <cl@linux.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Vitaly Wool <vitalywool@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agomm, page_alloc: rename __GFP_WAIT to __GFP_RECLAIM
Mel Gorman [Sat, 7 Nov 2015 00:28:28 +0000 (16:28 -0800)]
mm, page_alloc: rename __GFP_WAIT to __GFP_RECLAIM

__GFP_WAIT was used to signal that the caller was in atomic context and
could not sleep.  Now it is possible to distinguish between true atomic
context and callers that are not willing to sleep.  The latter should
clear __GFP_DIRECT_RECLAIM so kswapd will still wake.  As clearing
__GFP_WAIT behaves differently, there is a risk that people will clear the
wrong flags.  This patch renames __GFP_WAIT to __GFP_RECLAIM to clearly
indicate what it does -- setting it allows all reclaim activity, clearing
them prevents it.

[akpm@linux-foundation.org: fix build]
[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Christoph Lameter <cl@linux.com>
Acked-by: David Rientjes <rientjes@google.com>
Cc: Vitaly Wool <vitalywool@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agomm: page_alloc: remove GFP_IOFS
Mel Gorman [Sat, 7 Nov 2015 00:28:25 +0000 (16:28 -0800)]
mm: page_alloc: remove GFP_IOFS

GFP_IOFS was intended to be shorthand for clearing two flags, not a set of
allocation flags.  There is only one user of this flag combination now and
there appears to be no reason why Lustre had to be protected from reclaim
stalls.  As none of the sites appear to be atomic, this patch simply
deletes GFP_IOFS and converts Lustre to using GFP_KERNEL, GFP_NOFS or
GFP_NOIO as appropriate.

Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Cc: Oleg Drokin <oleg.drokin@intel.com>
Cc: Andreas Dilger <andreas.dilger@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agomm, page_alloc: distinguish between being unable to sleep, unwilling to sleep and...
Mel Gorman [Sat, 7 Nov 2015 00:28:21 +0000 (16:28 -0800)]
mm, page_alloc: distinguish between being unable to sleep, unwilling to sleep and avoiding waking kswapd

__GFP_WAIT has been used to identify atomic context in callers that hold
spinlocks or are in interrupts.  They are expected to be high priority and
have access one of two watermarks lower than "min" which can be referred
to as the "atomic reserve".  __GFP_HIGH users get access to the first
lower watermark and can be called the "high priority reserve".

Over time, callers had a requirement to not block when fallback options
were available.  Some have abused __GFP_WAIT leading to a situation where
an optimisitic allocation with a fallback option can access atomic
reserves.

This patch uses __GFP_ATOMIC to identify callers that are truely atomic,
cannot sleep and have no alternative.  High priority users continue to use
__GFP_HIGH.  __GFP_DIRECT_RECLAIM identifies callers that can sleep and
are willing to enter direct reclaim.  __GFP_KSWAPD_RECLAIM to identify
callers that want to wake kswapd for background reclaim.  __GFP_WAIT is
redefined as a caller that is willing to enter direct reclaim and wake
kswapd for background reclaim.

This patch then converts a number of sites

o __GFP_ATOMIC is used by callers that are high priority and have memory
  pools for those requests. GFP_ATOMIC uses this flag.

o Callers that have a limited mempool to guarantee forward progress clear
  __GFP_DIRECT_RECLAIM but keep __GFP_KSWAPD_RECLAIM. bio allocations fall
  into this category where kswapd will still be woken but atomic reserves
  are not used as there is a one-entry mempool to guarantee progress.

o Callers that are checking if they are non-blocking should use the
  helper gfpflags_allow_blocking() where possible. This is because
  checking for __GFP_WAIT as was done historically now can trigger false
  positives. Some exceptions like dm-crypt.c exist where the code intent
  is clearer if __GFP_DIRECT_RECLAIM is used instead of the helper due to
  flag manipulations.

o Callers that built their own GFP flags instead of starting with GFP_KERNEL
  and friends now also need to specify __GFP_KSWAPD_RECLAIM.

The first key hazard to watch out for is callers that removed __GFP_WAIT
and was depending on access to atomic reserves for inconspicuous reasons.
In some cases it may be appropriate for them to use __GFP_HIGH.

The second key hazard is callers that assembled their own combination of
GFP flags instead of starting with something like GFP_KERNEL.  They may
now wish to specify __GFP_KSWAPD_RECLAIM.  It's almost certainly harmless
if it's missed in most cases as other activity will wake kswapd.

Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Christoph Lameter <cl@linux.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Vitaly Wool <vitalywool@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agomm, page_alloc: use masks and shifts when converting GFP flags to migrate types
Mel Gorman [Sat, 7 Nov 2015 00:28:18 +0000 (16:28 -0800)]
mm, page_alloc: use masks and shifts when converting GFP flags to migrate types

This patch redefines which GFP bits are used for specifying mobility and
the order of the migrate types.  Once redefined it's possible to convert
GFP flags to a migrate type with a simple mask and shift.  The only
downside is that readers of OOM kill messages and allocation failures may
have been used to the existing values but scripts/gfp-translate will help.

Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Christoph Lameter <cl@linux.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Vitaly Wool <vitalywool@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agomm, page_alloc: remove unnecessary taking of a seqlock when cpusets are disabled
Mel Gorman [Sat, 7 Nov 2015 00:28:15 +0000 (16:28 -0800)]
mm, page_alloc: remove unnecessary taking of a seqlock when cpusets are disabled

There is a seqcounter that protects against spurious allocation failures
when a task is changing the allowed nodes in a cpuset.  There is no need
to check the seqcounter until a cpuset exists.

Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Christoph Lameter <cl@linux.com>
Acked-by: David Rientjes <rientjes@google.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Vitaly Wool <vitalywool@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agomm, page_alloc: remove unnecessary recalculations for dirty zone balancing
Mel Gorman [Sat, 7 Nov 2015 00:28:12 +0000 (16:28 -0800)]
mm, page_alloc: remove unnecessary recalculations for dirty zone balancing

File-backed pages that will be immediately written are balanced between
zones.  This heuristic tries to avoid having a single zone filled with
recently dirtied pages but the checks are unnecessarily expensive.  Move
consider_zone_balanced into the alloc_context instead of checking bitmaps
multiple times.  The patch also gives the parameter a more meaningful
name.

Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: David Rientjes <rientjes@google.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Christoph Lameter <cl@linux.com>
Cc: Vitaly Wool <vitalywool@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agomm, page_alloc: remove unnecessary parameter from zone_watermark_ok_safe
Mel Gorman [Sat, 7 Nov 2015 00:28:09 +0000 (16:28 -0800)]
mm, page_alloc: remove unnecessary parameter from zone_watermark_ok_safe

Overall, the intent of this series is to remove the zonelist cache which
was introduced to avoid high overhead in the page allocator.  Once this is
done, it is necessary to reduce the cost of watermark checks.

The series starts with minor micro-optimisations.

Next it notes that GFP flags that affect watermark checks are abused.
__GFP_WAIT historically identified callers that could not sleep and could
access reserves.  This was later abused to identify callers that simply
prefer to avoid sleeping and have other options.  A patch distinguishes
between atomic callers, high-priority callers and those that simply wish
to avoid sleep.

The zonelist cache has been around for a long time but it is of dubious
merit with a lot of complexity and some issues that are explained.  The
most important issue is that a failed THP allocation can cause a zone to
be treated as "full".  This potentially causes unnecessary stalls, reclaim
activity or remote fallbacks.  The issues could be fixed but it's not
worth it.  The series places a small number of other micro-optimisations
on top before examining GFP flags watermarks.

High-order watermarks enforcement can cause high-order allocations to fail
even though pages are free.  The watermark checks both protect high-order
atomic allocations and make kswapd aware of high-order pages but there is
a much better way that can be handled using migrate types.  This series
uses page grouping by mobility to reserve pageblocks for high-order
allocations with the size of the reservation depending on demand.  kswapd
awareness is maintained by examining the free lists.  By patch 12 in this
series, there are no high-order watermark checks while preserving the
properties that motivated the introduction of the watermark checks.

This patch (of 10):

No user of zone_watermark_ok_safe() specifies alloc_flags.  This patch
removes the unnecessary parameter.

Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: David Rientjes <rientjes@google.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Michal Hocko <mhocko@suse.com>
Reviewed-by: Christoph Lameter <cl@linux.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agomm/oom_kill.c: introduce is_sysrq_oom helper
Yaowei Bai [Sat, 7 Nov 2015 00:28:06 +0000 (16:28 -0800)]
mm/oom_kill.c: introduce is_sysrq_oom helper

Introduce is_sysrq_oom helper function indicating oom kill triggered
by sysrq to improve readability.

No functional changes.

Signed-off-by: Yaowei Bai <bywxiaobai@163.com>
Acked-by: David Rientjes <rientjes@google.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>
8 years agoMerge tag 'backlight-for-linus-4.4' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Fri, 6 Nov 2015 18:53:48 +0000 (10:53 -0800)]
Merge tag 'backlight-for-linus-4.4' of git://git./linux/kernel/git/lee/backlight

Pull backlight updates from Lee Jones:
 "New Device Support
   - None

  New Functionality:
   - None

  Core Frameworks:
   - Reject legacy PWM request for device defined in DT

  Fix-ups:
   - Remove unnecessary MODULE_ALIAS(); adp8860_bl, adp8870_bl
   - Simplify code: pm8941-wled
   - Supply default-brightness logic; pm8941-wled

  Bug Fixes:
   - Clean up OF node; 88pm860x_bl
   - Ensure struct is zeroed; lp855x_bl"

* tag 'backlight-for-linus-4.4' of git://git.kernel.org/pub/scm/linux/kernel/git/lee/backlight:
  backlight: pm8941-wled: Add default-brightness property
  backlight: pm8941-wled: Fix ptr_ret.cocci warnings
  backlight: pwm: Reject legacy PWM request for device defined in DT
  backlight: 88pm860x_bl: Add missing of_node_put
  backlight: adp8870: Remove unnecessary MODULE_ALIAS()
  backlight: adp8860: Remove unnecessary MODULE_ALIAS()
  backlight: lp855x: Make sure props struct is zeroed

8 years agomfd: avoid newly introduced compiler warning
Linus Torvalds [Fri, 6 Nov 2015 18:47:12 +0000 (10:47 -0800)]
mfd: avoid newly introduced compiler warning

Commit b158b69a3765 ("mfd: rtsx: Simplify function return logic")
removed the use of the 'err' variable, but left the variable itself
around, resulting in gcc quite reasonably warning:

    drivers/mfd/rtsx_pcr.c: In function â€˜rtsx_pci_set_pull_ctl’:
    drivers/mfd/rtsx_pcr.c:565:6: warning: unused variable â€˜err’ [-Wunused-variable]
      int err;
          ^

Get rid of the unused variable, and avoid the new warning.

Cc: Javier Martinez Canillas <javier@osg.samsung.com>
Cc: Lee Jones <lee.jones@linaro.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agoMerge tag 'mfd-for-linus-4.4' of git://git.kernel.org/pub/scm/linux/kernel/git/lee/mfd
Linus Torvalds [Fri, 6 Nov 2015 18:23:50 +0000 (10:23 -0800)]
Merge tag 'mfd-for-linus-4.4' of git://git./linux/kernel/git/lee/mfd

Pull MFD updates from Lee Jones:
 "New Device Support:
   - Add support for 88pm860; 88pm80x
   - Add support for 24c08 EEPROM; at24
   - Add support for Broxton Whiskey Cove; intel*
   - Add support for RTS522A; rts5227
   - Add support for I2C devices; intel_quark_i2c_gpio

  New Functionality:
   - Add microphone support; arizona
   - Add general purpose switch support; arizona
   - Add fuel-gauge support; da9150-core
   - Add shutdown support; sec-core
   - Add charger support; tps65217
   - Add flexible serial communication unit support; atmel-flexcom
   - Add power button support; axp20x
   - Add led-flash support; rt5033

  Core Frameworks:
   - Supply a generic macro for defining Regmap IRQs
   - Rework ACPI child device matching

  Fix-ups:
   - Use Regmap to access registers; tps6105x
   - Use DEFINE_RES_IRQ_NAMED() macro; da9150
   - Re-arrange device registration order; intel_quark_i2c_gpio
   - Allow OF matching; cros_ec_i2c, atmel-hlcdc, hi6421-pmic, max8997, sm501
   - Handle deferred probe; twl6040
   - Improve accuracy of headphone detect; arizona
   - Unnecessary MODULE_ALIAS() removal; bcm590xx, rt5033
   - Remove unused code; htc-i2cpld, arizona, pcf50633-irq, sec-core
   - Simplify code; kempld, rts5209, da903x, lm3533, da9052, arizona
   - Remove #iffery; arizona
   - DT binding adaptions; many

  Bug Fixes:
   - Fix possible NULL pointer dereference; wm831x, tps6105x
   - Fix 64bit bug; intel_soc_pmic_bxtwc
   - Fix signedness issue; arizona"

* tag 'mfd-for-linus-4.4' of git://git.kernel.org/pub/scm/linux/kernel/git/lee/mfd: (73 commits)
  bindings: mfd: s2mps11: Add documentation for s2mps15 PMIC
  mfd: sec-core: Remove unused s2mpu02-rtc and s2mpu02-clk children
  extcon: arizona: Add extcon specific device tree binding document
  MAINTAINERS: Add binding docs for Cirrus Logic/Wolfson Arizona devices
  mfd: arizona: Remove bindings covered in new subsystem specific docs
  mfd: rt5033: Add RT5033 Flash led sub device
  mfd: lpss: Add Intel Broxton PCI IDs
  mfd: lpss: Add Broxton ACPI IDs
  mfd: arizona: Signedness bug in arizona_runtime_suspend()
  mfd: axp20x: Add a cell for the power button part of the, axp288 PMICs
  mfd: dt-bindings: Document pulled down WRSTBI pin on S2MPS1X
  mfd: sec-core: Disable buck voltage reset on watchdog falling edge
  mfd: sec-core: Dump PMIC revision to find out the HW
  mfd: arizona: Use correct type ID for device tree config
  mfd: arizona: Remove use of codec build config #ifdefs
  mfd: arizona: Simplify adding subdevices
  mfd: arizona: Downgrade type mismatch messages to dev_warn
  mfd: arizona: Factor out checking of jack detection state
  mfd: arizona: Factor out DCVDD isolation control
  mfd: Make TPS6105X select REGMAP_I2C
  ...

8 years agox86: don't make DEBUG_WX default to 'y' even with DEBUG_RODATA
Linus Torvalds [Fri, 6 Nov 2015 17:12:41 +0000 (09:12 -0800)]
x86: don't make DEBUG_WX default to 'y' even with DEBUG_RODATA

It turns out that we still have issues with the EFI memory map that ends
up polluting our kernel page tables with writable executable pages.

That will get sorted out, but in the meantime let's not make the scary
complaint about them be on by default.  The code is useful for
developers, but not ready for end user testing yet.

Acked-by: Borislav Petkov <bp@alien8.de>
Acked-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agoMerge tag 'platform-drivers-x86-v4.4-1' of git://git.infradead.org/users/dvhart/linux...
Linus Torvalds [Fri, 6 Nov 2015 07:45:53 +0000 (23:45 -0800)]
Merge tag 'platform-drivers-x86-v4.4-1' of git://git.infradead.org/users/dvhart/linux-platform-drivers-x86

Pull x86 platform driver update from Darren Hart:
 "Various toshiba hotkey and keyboard related fixes and a new WMI
  driver.  Several intel_scu_ipc cleanups and a locking fix.  A
  spattering of small single fixes across various platforms.

  I was asked to pick up an OLPC cleanup as the driver appeared
  unmaintained and it seemed similar to what is maintained in
  platform/drivers/x86.  I have included the patch and an update to the
  MAINTAINERS file.

  toshiba_acpi:
   - Initialize hotkey_event_type variable
   - Remove unneeded u32 variables from *setup_keyboard
   - Add 0x prefix to available_kbd_modes_show function
   - Change default Hotkey enabling value
   - Unify hotkey enabling functions

  toshiba-wmi:
   - Toshiba WMI Hotkey Driver

  intel_scu_ipc:
   - Protect dev member assignment on ->remove()
   - Switch to use module_pci_driver() macro
   - Convert to use struct device *
   - Propagate pointer to struct intel_scu_ipc_dev
   - Fix error path by turning to devm_* / pcim_*

  acer-wmi:
   - remove threeg and interface sysfs interfaces

  OLPC:
   - Use %*ph specifier instead of passing direct values

  MAINTAINERS:
   - Add drivers/platform/olpc to drivers/platform/x86

  sony-laptop:
   - Fix handling sony_nc_hotkeys_decode result

  intel_mid_powerbtn:
   - Remove misuse of IRQF_NO_SUSPEND flag

  compal-laptop:
   - Add charge control limit

  asus-wmi:
   - restore kbd led level after resume"

* tag 'platform-drivers-x86-v4.4-1' of git://git.infradead.org/users/dvhart/linux-platform-drivers-x86:
  toshiba_acpi: Initialize hotkey_event_type variable
  intel_scu_ipc: Protect dev member assignment on ->remove()
  intel_scu_ipc: Switch to use module_pci_driver() macro
  intel_scu_ipc: Convert to use struct device *
  intel_scu_ipc: Propagate pointer to struct intel_scu_ipc_dev
  intel_scu_ipc: Fix error path by turning to devm_* / pcim_*
  acer-wmi: remove threeg and interface sysfs interfaces
  OLPC: Use %*ph specifier instead of passing direct values
  MAINTAINERS: Add drivers/platform/olpc to drivers/platform/x86
  platform/x86: Toshiba WMI Hotkey Driver
  sony-laptop: Fix handling sony_nc_hotkeys_decode result
  intel_mid_powerbtn: Remove misuse of IRQF_NO_SUSPEND flag
  compal-laptop: Add charge control limit
  asus-wmi: restore kbd led level after resume
  toshiba_acpi: Remove unneeded u32 variables from *setup_keyboard
  toshiba_acpi: Add 0x prefix to available_kbd_modes_show function
  toshiba_acpi: Change default Hotkey enabling value
  toshiba_acpi: Unify hotkey enabling functions

8 years agoMerge tag 'powerpc-4.4-1' of git://git.kernel.org/pub/scm/linux/kernel/git/powerpc...
Linus Torvalds [Fri, 6 Nov 2015 07:38:43 +0000 (23:38 -0800)]
Merge tag 'powerpc-4.4-1' of git://git./linux/kernel/git/powerpc/linux

Pull powerpc updates from Michael Ellerman:

 - Kconfig: remove BE-only platforms from LE kernel build from Boqun
   Feng
 - Refresh ps3_defconfig from Geoff Levand
 - Emit GNU & SysV hashes for the vdso from Michael Ellerman
 - Define an enum for the bolted SLB indexes from Anshuman Khandual
 - Use a local to avoid multiple calls to get_slb_shadow() from Michael
   Ellerman
 - Add gettimeofday() benchmark from Michael Neuling
 - Avoid link stack corruption in __get_datapage() from Michael Neuling
 - Add virt_to_pfn and use this instead of opencoding from Aneesh Kumar
   K.V
 - Add ppc64le_defconfig from Michael Ellerman
 - pseries: extract of_helpers module from Andy Shevchenko
 - Correct string length in pseries_of_derive_parent() from Nathan
   Fontenot
 - Free the MSI bitmap if it was slab allocated from Denis Kirjanov
 - Shorten irq_chip name for the SIU from Christophe Leroy
 - Wait 1s for secondaries to enter OPAL during kexec from Samuel
   Mendoza-Jonas
 - Fix _ALIGN_* errors due to type difference, from Aneesh Kumar K.V
 - powerpc/pseries/hvcserver: don't memset pi_buff if it is null from
   Colin Ian King
 - Disable hugepd for 64K page size, from Aneesh Kumar K.V
 - Differentiate between hugetlb and THP during page walk from Aneesh
   Kumar K.V
 - Make PCI non-optional for pseries from Michael Ellerman
 - Individual System V IPC system calls from Sam bobroff
 - Add selftest of unmuxed IPC calls from Michael Ellerman
 - discard .exit.data at runtime from Stephen Rothwell
 - Delete old orphaned PrPMC 280/2800 DTS and boot file, from Paul
   Gortmaker
 - Use of_get_next_parent to simplify code from Christophe Jaillet
 - Paginate some xmon output from Sam bobroff
 - Add some more elements to the xmon PACA dump from Michael Ellerman
 - Allow the tm-syscall selftest to build with old headers from Michael
   Ellerman
 - Run EBB selftests only on POWER8 from Denis Kirjanov
 - Drop CONFIG_TUNE_CELL in favour of CONFIG_CELL_CPU from Michael
   Ellerman
 - Avoid reference to potentially freed memory in prom.c from Christophe
   Jaillet
 - Quieten boot wrapper output with run_cmd from Geoff Levand
 - EEH fixes and cleanups from Gavin Shan
 - Fix recursive fenced PHB on Broadcom shiner adapter from Gavin Shan
 - Use of_get_next_parent() in of_get_ibm_chip_id() from Michael
   Ellerman
 - Fix section mismatch warning in msi_bitmap_alloc() from Denis
   Kirjanov
 - Fix ps3-lpm white space from Rudhresh Kumar J
 - Fix ps3-vuart null dereference from Colin King
 - nvram: Add missing kfree in error path from Christophe Jaillet
 - nvram: Fix function name in some errors messages, from Christophe
   Jaillet
 - drivers/macintosh: adb: fix misleading Kconfig help text from Aaro
   Koskinen
 - agp/uninorth: fix a memleak in create_gatt_table from Denis Kirjanov
 - cxl: Free virtual PHB when removing from Andrew Donnellan
 - scripts/kconfig/Makefile: Allow KBUILD_DEFCONFIG to be a target from
   Michael Ellerman
 - scripts/kconfig/Makefile: Fix KBUILD_DEFCONFIG check when building
   with O= from Michael Ellerman
 - Freescale updates from Scott: Highlights include 64-bit book3e
   kexec/kdump support, a rework of the qoriq clock driver, device tree
   changes including qoriq fman nodes, support for a new 85xx board, and
   some fixes.
 - MPC5xxx updates from Anatolij: Highlights include a driver for
   MPC512x LocalPlus Bus FIFO with its device tree binding
   documentation, mpc512x device tree updates and some minor fixes.

* tag 'powerpc-4.4-1' of git://git.kernel.org/pub/scm/linux/kernel/git/powerpc/linux: (106 commits)
  powerpc/msi: Fix section mismatch warning in msi_bitmap_alloc()
  powerpc/prom: Use of_get_next_parent() in of_get_ibm_chip_id()
  powerpc/pseries: Correct string length in pseries_of_derive_parent()
  powerpc/e6500: hw tablewalk: make sure we invalidate and write to the same tlb entry
  powerpc/mpc85xx: Add FSL QorIQ DPAA FMan support to the SoC device tree(s)
  powerpc/mpc85xx: Create dts components for the FSL QorIQ DPAA FMan
  powerpc/fsl: Add #clock-cells and clockgen label to clockgen nodes
  powerpc: handle error case in cpm_muram_alloc()
  powerpc: mpic: use IRQCHIP_SKIP_SET_WAKE instead of redundant mpic_irq_set_wake
  powerpc/book3e-64: Enable kexec
  powerpc/book3e-64/kexec: Set "r4 = 0" when entering spinloop
  powerpc/booke: Only use VIRT_PHYS_OFFSET on booke32
  powerpc/book3e-64/kexec: Enable SMP release
  powerpc/book3e-64/kexec: create an identity TLB mapping
  powerpc/book3e-64: Don't limit paca to 256 MiB
  powerpc/book3e/kdump: Enable crash_kexec_wait_realmode
  powerpc/book3e: support CONFIG_RELOCATABLE
  powerpc/booke64: Fix args to copy_and_flush
  powerpc/book3e-64: rename interrupt_end_book3e with __end_interrupts
  powerpc/e6500: kexec: Handle hardware threads
  ...

8 years agoMerge branch 'akpm' (patches from Andrew)
Linus Torvalds [Fri, 6 Nov 2015 07:10:54 +0000 (23:10 -0800)]
Merge branch 'akpm' (patches from Andrew)

Merge patch-bomb from Andrew Morton:

 - inotify tweaks

 - some ocfs2 updates (many more are awaiting review)

 - various misc bits

 - kernel/watchdog.c updates

 - Some of mm.  I have a huge number of MM patches this time and quite a
   lot of it is quite difficult and much will be held over to next time.

* emailed patches from Andrew Morton <akpm@linux-foundation.org>: (162 commits)
  selftests: vm: add tests for lock on fault
  mm: mlock: add mlock flags to enable VM_LOCKONFAULT usage
  mm: introduce VM_LOCKONFAULT
  mm: mlock: add new mlock system call
  mm: mlock: refactor mlock, munlock, and munlockall code
  kasan: always taint kernel on report
  mm, slub, kasan: enable user tracking by default with KASAN=y
  kasan: use IS_ALIGNED in memory_is_poisoned_8()
  kasan: Fix a type conversion error
  lib: test_kasan: add some testcases
  kasan: update reference to kasan prototype repo
  kasan: move KASAN_SANITIZE in arch/x86/boot/Makefile
  kasan: various fixes in documentation
  kasan: update log messages
  kasan: accurately determine the type of the bad access
  kasan: update reported bug types for kernel memory accesses
  kasan: update reported bug types for not user nor kernel memory accesses
  mm/kasan: prevent deadlock in kasan reporting
  mm/kasan: don't use kasan shadow pointer in generic functions
  mm/kasan: MODULE_VADDR is not available on all archs
  ...

8 years agovfs: clear remainder of 'full_fds_bits' in dup_fd()
Eric Biggers [Fri, 6 Nov 2015 06:32:04 +0000 (00:32 -0600)]
vfs: clear remainder of 'full_fds_bits' in dup_fd()

This fixes a bug from commit f3f86e33dc3d ("vfs: Fix pathological
performance case for __alloc_fd()").

v2: refactor to share fd bitmap copying code
Signed-off-by: Eric Biggers <ebiggers3@gmail.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agoselftests: vm: add tests for lock on fault
Eric B Munson [Fri, 6 Nov 2015 02:51:43 +0000 (18:51 -0800)]
selftests: vm: add tests for lock on fault

Test the mmap() flag, and the mlockall() flag.  These tests ensure that
pages are not faulted in until they are accessed, that the pages are
unevictable once faulted in, and that VMA splitting and merging works with
the new VM flag.  The second test ensures that mlock limits are respected.
 Note that the limit test needs to be run a normal user.

Also add tests to use the new mlock2 family of system calls.

[treding@nvidia.com: : Fix mlock2-tests for 32-bit architectures]
[treding@nvidia.com: ensure the mlock2 syscall number can be found]
[treding@nvidia.com: use the right arguments for main()]
Signed-off-by: Eric B Munson <emunson@akamai.com>
Cc: Shuah Khan <shuahkh@osg.samsung.com>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Guenter Roeck <linux@roeck-us.net>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Michael Kerrisk <mtk.manpages@gmail.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agomm: mlock: add mlock flags to enable VM_LOCKONFAULT usage
Eric B Munson [Fri, 6 Nov 2015 02:51:39 +0000 (18:51 -0800)]
mm: mlock: add mlock flags to enable VM_LOCKONFAULT usage

The previous patch introduced a flag that specified pages in a VMA should
be placed on the unevictable LRU, but they should not be made present when
the area is created.  This patch adds the ability to set this state via
the new mlock system calls.

We add MLOCK_ONFAULT for mlock2 and MCL_ONFAULT for mlockall.
MLOCK_ONFAULT will set the VM_LOCKONFAULT modifier for VM_LOCKED.
MCL_ONFAULT should be used as a modifier to the two other mlockall flags.
When used with MCL_CURRENT, all current mappings will be marked with
VM_LOCKED | VM_LOCKONFAULT.  When used with MCL_FUTURE, the mm->def_flags
will be marked with VM_LOCKED | VM_LOCKONFAULT.  When used with both
MCL_CURRENT and MCL_FUTURE, all current mappings and mm->def_flags will be
marked with VM_LOCKED | VM_LOCKONFAULT.

Prior to this patch, mlockall() will unconditionally clear the
mm->def_flags any time it is called without MCL_FUTURE.  This behavior is
maintained after adding MCL_ONFAULT.  If a call to mlockall(MCL_FUTURE) is
followed by mlockall(MCL_CURRENT), the mm->def_flags will be cleared and
new VMAs will be unlocked.  This remains true with or without MCL_ONFAULT
in either mlockall() invocation.

munlock() will unconditionally clear both vma flags.  munlockall()
unconditionally clears for VMA flags on all VMAs and in the mm->def_flags
field.

Signed-off-by: Eric B Munson <emunson@akamai.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Guenter Roeck <linux@roeck-us.net>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Michael Kerrisk <mtk.manpages@gmail.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: Shuah Khan <shuahkh@osg.samsung.com>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agomm: introduce VM_LOCKONFAULT
Eric B Munson [Fri, 6 Nov 2015 02:51:36 +0000 (18:51 -0800)]
mm: introduce VM_LOCKONFAULT

The cost of faulting in all memory to be locked can be very high when
working with large mappings.  If only portions of the mapping will be used
this can incur a high penalty for locking.

For the example of a large file, this is the usage pattern for a large
statical language model (probably applies to other statical or graphical
models as well).  For the security example, any application transacting in
data that cannot be swapped out (credit card data, medical records, etc).

This patch introduces the ability to request that pages are not
pre-faulted, but are placed on the unevictable LRU when they are finally
faulted in.  The VM_LOCKONFAULT flag will be used together with VM_LOCKED
and has no effect when set without VM_LOCKED.  Setting the VM_LOCKONFAULT
flag for a VMA will cause pages faulted into that VMA to be added to the
unevictable LRU when they are faulted or if they are already present, but
will not cause any missing pages to be faulted in.

Exposing this new lock state means that we cannot overload the meaning of
the FOLL_POPULATE flag any longer.  Prior to this patch it was used to
mean that the VMA for a fault was locked.  This means we need the new
FOLL_MLOCK flag to communicate the locked state of a VMA.  FOLL_POPULATE
will now only control if the VMA should be populated and in the case of
VM_LOCKONFAULT, it will not be set.

Signed-off-by: Eric B Munson <emunson@akamai.com>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Guenter Roeck <linux@roeck-us.net>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Michael Kerrisk <mtk.manpages@gmail.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: Shuah Khan <shuahkh@osg.samsung.com>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agomm: mlock: add new mlock system call
Eric B Munson [Fri, 6 Nov 2015 02:51:33 +0000 (18:51 -0800)]
mm: mlock: add new mlock system call

With the refactored mlock code, introduce a new system call for mlock.
The new call will allow the user to specify what lock states are being
added.  mlock2 is trivial at the moment, but a follow on patch will add a
new mlock state making it useful.

Signed-off-by: Eric B Munson <emunson@akamai.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Cc: Guenter Roeck <linux@roeck-us.net>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Michael Kerrisk <mtk.manpages@gmail.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: Shuah Khan <shuahkh@osg.samsung.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agomm: mlock: refactor mlock, munlock, and munlockall code
Eric B Munson [Fri, 6 Nov 2015 02:51:29 +0000 (18:51 -0800)]
mm: mlock: refactor mlock, munlock, and munlockall code

mlock() allows a user to control page out of program memory, but this
comes at the cost of faulting in the entire mapping when it is allocated.
For large mappings where the entire area is not necessary this is not
ideal.  Instead of forcing all locked pages to be present when they are
allocated, this set creates a middle ground.  Pages are marked to be
placed on the unevictable LRU (locked) when they are first used, but they
are not faulted in by the mlock call.

This series introduces a new mlock() system call that takes a flags
argument along with the start address and size.  This flags argument gives
the caller the ability to request memory be locked in the traditional way,
or to be locked after the page is faulted in.  A new MCL flag is added to
mirror the lock on fault behavior from mlock() in mlockall().

There are two main use cases that this set covers.  The first is the
security focussed mlock case.  A buffer is needed that cannot be written
to swap.  The maximum size is known, but on average the memory used is
significantly less than this maximum.  With lock on fault, the buffer is
guaranteed to never be paged out without consuming the maximum size every
time such a buffer is created.

The second use case is focussed on performance.  Portions of a large file
are needed and we want to keep the used portions in memory once accessed.
This is the case for large graphical models where the path through the
graph is not known until run time.  The entire graph is unlikely to be
used in a given invocation, but once a node has been used it needs to stay
resident for further processing.  Given these constraints we have a number
of options.  We can potentially waste a large amount of memory by mlocking
the entire region (this can also cause a significant stall at startup as
the entire file is read in).  We can mlock every page as we access them
without tracking if the page is already resident but this introduces large
overhead for each access.  The third option is mapping the entire region
with PROT_NONE and using a signal handler for SIGSEGV to
mprotect(PROT_READ) and mlock() the needed page.  Doing this page at a
time adds a significant performance penalty.  Batching can be used to
mitigate this overhead, but in order to safely avoid trying to mprotect
pages outside of the mapping, the boundaries of each mapping to be used in
this way must be tracked and available to the signal handler.  This is
precisely what the mm system in the kernel should already be doing.

For mlock(MLOCK_ONFAULT) the user is charged against RLIMIT_MEMLOCK as if
mlock(MLOCK_LOCKED) or mmap(MAP_LOCKED) was used, so when the VMA is
created not when the pages are faulted in.  For mlockall(MCL_ONFAULT) the
user is charged as if MCL_FUTURE was used.  This decision was made to keep
the accounting checks out of the page fault path.

To illustrate the benefit of this set I wrote a test program that mmaps a
5 GB file filled with random data and then makes 15,000,000 accesses to
random addresses in that mapping.  The test program was run 20 times for
each setup.  Results are reported for two program portions, setup and
execution.  The setup phase is calling mmap and optionally mlock on the
entire region.  For most experiments this is trivial, but it highlights
the cost of faulting in the entire region.  Results are averages across
the 20 runs in milliseconds.

mmap with mlock(MLOCK_LOCKED) on entire range:
Setup avg:      8228.666
Processing avg: 8274.257

mmap with mlock(MLOCK_LOCKED) before each access:
Setup avg:      0.113
Processing avg: 90993.552

mmap with PROT_NONE and signal handler and batch size of 1 page:
With the default value in max_map_count, this gets ENOMEM as I attempt
to change the permissions, after upping the sysctl significantly I get:
Setup avg:      0.058
Processing avg: 69488.073
mmap with PROT_NONE and signal handler and batch size of 8 pages:
Setup avg:      0.068
Processing avg: 38204.116

mmap with PROT_NONE and signal handler and batch size of 16 pages:
Setup avg:      0.044
Processing avg: 29671.180

mmap with mlock(MLOCK_ONFAULT) on entire range:
Setup avg:      0.189
Processing avg: 17904.899

The signal handler in the batch cases faulted in memory in two steps to
avoid having to know the start and end of the faulting mapping.  The first
step covers the page that caused the fault as we know that it will be
possible to lock.  The second step speculatively tries to mlock and
mprotect the batch size - 1 pages that follow.  There may be a clever way
to avoid this without having the program track each mapping to be covered
by this handeler in a globally accessible structure, but I could not find
it.  It should be noted that with a large enough batch size this two step
fault handler can still cause the program to crash if it reaches far
beyond the end of the mapping.

These results show that if the developer knows that a majority of the
mapping will be used, it is better to try and fault it in at once,
otherwise mlock(MLOCK_ONFAULT) is significantly faster.

The performance cost of these patches are minimal on the two benchmarks I
have tested (stream and kernbench).  The following are the average values
across 20 runs of stream and 10 runs of kernbench after a warmup run whose
results were discarded.

Avg throughput in MB/s from stream using 1000000 element arrays
Test     4.2-rc1      4.2-rc1+lock-on-fault
Copy:    10,566.5     10,421
Scale:   10,685       10,503.5
Add:     12,044.1     11,814.2
Triad:   12,064.8     11,846.3

Kernbench optimal load
                 4.2-rc1  4.2-rc1+lock-on-fault
Elapsed Time     78.453   78.991
User Time        64.2395  65.2355
System Time      9.7335   9.7085
Context Switches 22211.5  22412.1
Sleeps           14965.3  14956.1

This patch (of 6):

Extending the mlock system call is very difficult because it currently
does not take a flags argument.  A later patch in this set will extend
mlock to support a middle ground between pages that are locked and faulted
in immediately and unlocked pages.  To pave the way for the new system
call, the code needs some reorganization so that all the actual entry
point handles is checking input and translating to VMA flags.

Signed-off-by: Eric B Munson <emunson@akamai.com>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Michael Kerrisk <mtk.manpages@gmail.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Guenter Roeck <linux@roeck-us.net>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: Shuah Khan <shuahkh@osg.samsung.com>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agokasan: always taint kernel on report
Andrey Ryabinin [Fri, 6 Nov 2015 02:51:26 +0000 (18:51 -0800)]
kasan: always taint kernel on report

Currently we already taint the kernel in some cases.  E.g.  if we hit some
bug in slub memory we call object_err() which will taint the kernel with
TAINT_BAD_PAGE flag.  But for other kind of bugs kernel left untainted.

Always taint with TAINT_BAD_PAGE if kasan found some bug.  This is useful
for automated testing.

Signed-off-by: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Alexander Potapenko <glider@google.com>
Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agomm, slub, kasan: enable user tracking by default with KASAN=y
Andrey Ryabinin [Fri, 6 Nov 2015 02:51:23 +0000 (18:51 -0800)]
mm, slub, kasan: enable user tracking by default with KASAN=y

It's recommended to have slub's user tracking enabled with CONFIG_KASAN,
because:

a) User tracking disables slab merging which improves
    detecting out-of-bounds accesses.
b) User tracking metadata acts as redzone which also improves
    detecting out-of-bounds accesses.
c) User tracking provides additional information about object.
    This information helps to understand bugs.

Currently it is not enabled by default.  Besides recompiling the kernel
with KASAN and reinstalling it, user also have to change the boot cmdline,
which is not very handy.

Enable slub user tracking by default with KASAN=y, since there is no good
reason to not do this.

[akpm@linux-foundation.org: little fixes, per David]
Signed-off-by: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agokasan: use IS_ALIGNED in memory_is_poisoned_8()
Xishi Qiu [Fri, 6 Nov 2015 02:51:21 +0000 (18:51 -0800)]
kasan: use IS_ALIGNED in memory_is_poisoned_8()

Use IS_ALIGNED() to determine whether the shadow span two bytes.  It
generates less code and more readable.  Also add some comments in shadow
check functions.

Signed-off-by: Xishi Qiu <qiuxishi@huawei.com>
Acked-by: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Andrey Konovalov <adech.fo@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agokasan: Fix a type conversion error
Wang Long [Fri, 6 Nov 2015 02:51:18 +0000 (18:51 -0800)]
kasan: Fix a type conversion error

The current KASAN code can not find the following out-of-bounds bugs:

        char *ptr;
        ptr = kmalloc(8, GFP_KERNEL);
        memset(ptr+7, 0, 2);

the cause of the problem is the type conversion error in
*memory_is_poisoned_n* function.  So this patch fix that.

Signed-off-by: Wang Long <long.wanglong@huawei.com>
Acked-by: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Vladimir Murzin <vladimir.murzin@arm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agolib: test_kasan: add some testcases
Wang Long [Fri, 6 Nov 2015 02:51:15 +0000 (18:51 -0800)]
lib: test_kasan: add some testcases

Add some out of bounds testcases to test_kasan module.

Signed-off-by: Wang Long <long.wanglong@huawei.com>
Acked-by: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Vladimir Murzin <vladimir.murzin@arm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agokasan: update reference to kasan prototype repo
Andrey Konovalov [Fri, 6 Nov 2015 02:51:12 +0000 (18:51 -0800)]
kasan: update reference to kasan prototype repo

Update the reference to the kasan prototype repository on github, since it
was renamed.

Signed-off-by: Andrey Konovalov <andreyknvl@google.com>
Cc: Andrey Ryabinin <ryabinin.a.a@gmail.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: Alexander Potapenko <glider@google.com>
Cc: Konstantin Serebryany <kcc@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agokasan: move KASAN_SANITIZE in arch/x86/boot/Makefile
Andrey Konovalov [Fri, 6 Nov 2015 02:51:09 +0000 (18:51 -0800)]
kasan: move KASAN_SANITIZE in arch/x86/boot/Makefile

Move KASAN_SANITIZE in arch/x86/boot/Makefile above the comment
related to SVGA_MODE, since the comment refers to 'the next line'.

Signed-off-by: Andrey Konovalov <andreyknvl@google.com>
Cc: Andrey Ryabinin <ryabinin.a.a@gmail.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: Alexander Potapenko <glider@google.com>
Cc: Konstantin Serebryany <kcc@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agokasan: various fixes in documentation
Andrey Konovalov [Fri, 6 Nov 2015 02:51:06 +0000 (18:51 -0800)]
kasan: various fixes in documentation

[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Andrey Konovalov <andreyknvl@google.com>
Cc: Andrey Ryabinin <ryabinin.a.a@gmail.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: Alexander Potapenko <glider@google.com>
Cc: Konstantin Serebryany <kcc@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agokasan: update log messages
Andrey Konovalov [Fri, 6 Nov 2015 02:51:03 +0000 (18:51 -0800)]
kasan: update log messages

We decided to use KASAN as the short name of the tool and
KernelAddressSanitizer as the full one.  Update log messages according to
that.

Signed-off-by: Andrey Konovalov <andreyknvl@google.com>
Cc: Andrey Ryabinin <ryabinin.a.a@gmail.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: Alexander Potapenko <glider@google.com>
Cc: Konstantin Serebryany <kcc@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agokasan: accurately determine the type of the bad access
Andrey Konovalov [Fri, 6 Nov 2015 02:51:01 +0000 (18:51 -0800)]
kasan: accurately determine the type of the bad access

Makes KASAN accurately determine the type of the bad access. If the shadow
byte value is in the [0, KASAN_SHADOW_SCALE_SIZE) range we can look at
the next shadow byte to determine the type of the access.

Signed-off-by: Andrey Konovalov <andreyknvl@google.com>
Cc: Andrey Ryabinin <ryabinin.a.a@gmail.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: Alexander Potapenko <glider@google.com>
Cc: Konstantin Serebryany <kcc@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agokasan: update reported bug types for kernel memory accesses
Andrey Konovalov [Fri, 6 Nov 2015 02:50:58 +0000 (18:50 -0800)]
kasan: update reported bug types for kernel memory accesses

Update the names of the bad access types to better reflect the type of
the access that happended and make these error types "literals" that can
be used for classification and deduplication in scripts.

Signed-off-by: Andrey Konovalov <andreyknvl@google.com>
Cc: Andrey Ryabinin <ryabinin.a.a@gmail.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: Alexander Potapenko <glider@google.com>
Cc: Konstantin Serebryany <kcc@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agokasan: update reported bug types for not user nor kernel memory accesses
Andrey Konovalov [Fri, 6 Nov 2015 02:50:55 +0000 (18:50 -0800)]
kasan: update reported bug types for not user nor kernel memory accesses

Each access with address lower than
kasan_shadow_to_mem(KASAN_SHADOW_START) is reported as user-memory-access.
This is not always true, the accessed address might not be in user space.
Fix this by reporting such accesses as null-ptr-derefs or
wild-memory-accesses.

There's another reason for this change.  For userspace ASan we have a
bunch of systems that analyze error types for the purpose of
classification and deduplication.  Sooner of later we will write them to
KASAN as well.  Then clearly and explicitly stated error types will bring
value.

Signed-off-by: Andrey Konovalov <andreyknvl@google.com>
Cc: Andrey Ryabinin <ryabinin.a.a@gmail.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: Alexander Potapenko <glider@google.com>
Cc: Konstantin Serebryany <kcc@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agomm/kasan: prevent deadlock in kasan reporting
Aneesh Kumar K.V [Fri, 6 Nov 2015 02:50:51 +0000 (18:50 -0800)]
mm/kasan: prevent deadlock in kasan reporting

When we end up calling kasan_report in real mode, our shadow mapping for
the spinlock variable will show poisoned.  This will result in us calling
kasan_report_error with lock_report spin lock held.  To prevent this
disable kasan reporting when we are priting error w.r.t kasan.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Reviewed-by: 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>
8 years agomm/kasan: don't use kasan shadow pointer in generic functions
Aneesh Kumar K.V [Fri, 6 Nov 2015 02:50:48 +0000 (18:50 -0800)]
mm/kasan: don't use kasan shadow pointer in generic functions

We can't use generic functions like print_hex_dump to access kasan shadow
region.  This require us to setup another kasan shadow region for the
address passed (kasan shadow address).  Some architectures won't be able
to do that.  Hence make a copy of the shadow region row and pass that to
generic functions.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Reviewed-by: 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>
8 years agomm/kasan: MODULE_VADDR is not available on all archs
Aneesh Kumar K.V [Fri, 6 Nov 2015 02:50:46 +0000 (18:50 -0800)]
mm/kasan: MODULE_VADDR is not available on all archs

Use is_module_address instead

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Reviewed-by: 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>
8 years agomm/kasan: rename kasan_enabled() to kasan_report_enabled()
Aneesh Kumar K.V [Fri, 6 Nov 2015 02:50:43 +0000 (18:50 -0800)]
mm/kasan: rename kasan_enabled() to kasan_report_enabled()

The function only disable/enable reporting.  In the later patch we will be
adding a kasan early enable/disable.  Rename kasan_enabled to properly
reflect its function.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Reviewed-by: 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>
8 years agomm: remove refresh_cpu_vm_stats() definition for !SMP kernel
Tetsuo Handa [Fri, 6 Nov 2015 02:50:40 +0000 (18:50 -0800)]
mm: remove refresh_cpu_vm_stats() definition for !SMP kernel

refresh_cpu_vm_stats(int cpu) is no longer referenced by !SMP kernel
since Linux 3.12.

Signed-off-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agoDocumentation/filesystems/proc.txt: a little tidying
Hugh Dickins [Fri, 6 Nov 2015 02:50:37 +0000 (18:50 -0800)]
Documentation/filesystems/proc.txt: a little tidying

There's an odd line about "Locked" at the head of the description of
/proc/meminfo: it seems to have strayed from /proc/PID/smaps, so lead it
back there.  Move "Swap" and "SwapPss" descriptions down above it, to
match the order in the file (though "PageSize"s still undescribed).

The example of "Locked: 374 kB" (the same as Pss, neither Rss nor Size) is
so unlikely as to be misleading: just make it 0, this is /bin/bash text;
which would be "dw" (disabled write) not "de" (do not expand).

Signed-off-by: Hugh Dickins <hughd@google.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agotmpfs: avoid a little creat and stat slowdown
Hugh Dickins [Fri, 6 Nov 2015 02:50:34 +0000 (18:50 -0800)]
tmpfs: avoid a little creat and stat slowdown

LKP reports that v4.2 commit afa2db2fb6f1 ("tmpfs: truncate prealloc
blocks past i_size") causes a 14.5% slowdown in the AIM9 creat-clo
benchmark.

creat-clo does just what you'd expect from the name, and creat's O_TRUNC
on 0-length file does indeed get into more overhead now shmem_setattr()
tests "0 <= 0" instead of "0 < 0".

I'm not sure how much we care, but I think it would not be too VW-like to
add in a check for whether any pages (or swap) are allocated: if none are
allocated, there's none to remove from the radix_tree.  At first I thought
that check would be good enough for the unmaps too, but no: we should not
skip the unlikely case of unmapping pages beyond the new EOF, which were
COWed from holes which have now been reclaimed, leaving none.

This gives me an 8.5% speedup: on Haswell instead of LKP's Westmere, and
running a debug config before and after: I hope those account for the
lesser speedup.

And probably someone has a benchmark where a thousand threads keep on
stat'ing the same file repeatedly: forestall that report by adjusting v4.3
commit 44a30220bc0a ("shmem: recalculate file inode when fstat") not to
take the spinlock in shmem_getattr() when there's no work to do.

Signed-off-by: Hugh Dickins <hughd@google.com>
Reported-by: Ying Huang <ying.huang@linux.intel.com>
Tested-by: Ying Huang <ying.huang@linux.intel.com>
Cc: Josef Bacik <jbacik@fb.com>
Cc: Yu Zhao <yuzhao@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agomm, oom: add comment for why oom_adj exists
David Rientjes [Fri, 6 Nov 2015 02:50:32 +0000 (18:50 -0800)]
mm, oom: add comment for why oom_adj exists

/proc/pid/oom_adj exists solely to avoid breaking existing userspace
binaries that write to the tunable.

Add a comment in the only possible location within the kernel tree to
describe the situation and motivation for keeping it around.

Signed-off-by: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agomemcg: fix thresholds for 32b architectures.
Michal Hocko [Fri, 6 Nov 2015 02:50:29 +0000 (18:50 -0800)]
memcg: fix thresholds for 32b architectures.

Commit 424cdc141380 ("memcg: convert threshold to bytes") has fixed a
regression introduced by 3e32cb2e0a12 ("mm: memcontrol: lockless page
counters") where thresholds were silently converted to use page units
rather than bytes when interpreting the user input.

The fix is not complete, though, as properly pointed out by Ben Hutchings
during stable backport review.  The page count is converted to bytes but
unsigned long is used to hold the value which would be obviously not
sufficient for 32b systems with more than 4G thresholds.  The same applies
to usage as taken from mem_cgroup_usage which might overflow.

Let's remove this bytes vs.  pages internal tracking differences and
handle thresholds in page units internally.  Chage mem_cgroup_usage() to
return the value in page units and revert 424cdc141380 because this should
be sufficient for the consistent handling.  mem_cgroup_read_u64 as the
only users of mem_cgroup_usage outside of the threshold handling code is
converted to give the proper in bytes result.  It is doing that already
for page_counter output so this is more consistent as well.

The value presented to the userspace is still in bytes units.

Fixes: 424cdc141380 ("memcg: convert threshold to bytes")
Fixes: 3e32cb2e0a12 ("mm: memcontrol: lockless page counters")
Signed-off-by: Michal Hocko <mhocko@suse.com>
Reported-by: Ben Hutchings <ben@decadent.org.uk>
Reviewed-by: Vladimir Davydov <vdavydov@virtuozzo.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: <stable@vger.kernel.org>
From: Michal Hocko <mhocko@kernel.org>
Subject: memcg-fix-thresholds-for-32b-architectures-fix

Cc: Ben Hutchings <ben@decadent.org.uk>
Cc: Vladimir Davydov <vdavydov@virtuozzo.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
From: Andrew Morton <akpm@linux-foundation.org>
Subject: memcg-fix-thresholds-for-32b-architectures-fix-fix

don't attempt to inline mem_cgroup_usage()

The compiler ignores the inline anwyay.  And __always_inlining it adds 600
bytes of goop to the .o file.

Cc: Ben Hutchings <ben@decadent.org.uk>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Vladimir Davydov <vdavydov@virtuozzo.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agomm: page_counter: let page_counter_try_charge() return bool
Johannes Weiner [Fri, 6 Nov 2015 02:50:26 +0000 (18:50 -0800)]
mm: page_counter: let page_counter_try_charge() return bool

page_counter_try_charge() currently returns 0 on success and -ENOMEM on
failure, which is surprising behavior given the function name.

Make it follow the expected pattern of try_stuff() functions that return a
boolean true to indicate success, or false for failure.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Vladimir Davydov <vdavydov@virtuozzo.com
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agomm: memcontrol: eliminate root memory.current
Johannes Weiner [Fri, 6 Nov 2015 02:50:23 +0000 (18:50 -0800)]
mm: memcontrol: eliminate root memory.current

memory.current on the root level doesn't add anything that wouldn't be
more accurate and detailed using system statistics.  It already doesn't
include slabs, and it'll be a pain to keep in sync when further memory
types are accounted in the memory controller.  Remove it.

Note that this applies to the new unified hierarchy interface only.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Tejun Heo <tj@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agomm, hugetlbfs: optimize when NUMA=n
Dave Hansen [Fri, 6 Nov 2015 02:50:20 +0000 (18:50 -0800)]
mm, hugetlbfs: optimize when NUMA=n

My recent patch "mm, hugetlb: use memory policy when available" added some
bloat to hugetlb.o.  This patch aims to get some of the bloat back,
especially when NUMA is not in play.

It does this with an implicit #ifdef and marking some things static that
should have been static in my first patch.  It also makes the warnings
only VM_WARN_ON()s.  They were responsible for a pretty big chunk of the
bloat.

Doing this gets our NUMA=n text size back to a wee bit _below_ where we
started before the original patch.

It also shaves a bit of space off the NUMA=y case, but not much.
Enforcing the mempolicy definitely takes some text and it's hard to avoid.

size(1) output:

   text    data     bss     dec     hex filename
  30745    3433    2492   36670    8f3e hugetlb.o.nonuma.baseline
  31305    3755    2492   37552    92b0 hugetlb.o.nonuma.patch1
  30713    3433    2492   36638    8f1e hugetlb.o.nonuma.patch2 (this patch)
  25235     473   41276   66984   105a8 hugetlb.o.numa.baseline
  25715     475   41276   67466   1078a hugetlb.o.numa.patch1
  25491     473   41276   67240   106a8 hugetlb.o.numa.patch2 (this patch)

Signed-off-by: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Hillf Danton <hillf.zj@alibaba-inc.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>
8 years agomm, hugetlb: use memory policy when available
Dave Hansen [Fri, 6 Nov 2015 02:50:17 +0000 (18:50 -0800)]
mm, hugetlb: use memory policy when available

I have a hugetlbfs user which is never explicitly allocating huge pages
with 'nr_hugepages'.  They only set 'nr_overcommit_hugepages' and then let
the pages be allocated from the buddy allocator at fault time.

This works, but they noticed that mbind() was not doing them any good and
the pages were being allocated without respect for the policy they
specified.

The code in question is this:

> struct page *alloc_huge_page(struct vm_area_struct *vma,
...
>         page = dequeue_huge_page_vma(h, vma, addr, avoid_reserve, gbl_chg);
>         if (!page) {
>                 page = alloc_buddy_huge_page(h, NUMA_NO_NODE);

dequeue_huge_page_vma() is smart and will respect the VMA's memory policy.
 But, it only grabs _existing_ huge pages from the huge page pool.  If the
pool is empty, we fall back to alloc_buddy_huge_page() which obviously
can't do anything with the VMA's policy because it isn't even passed the
VMA.

Almost everybody preallocates huge pages.  That's probably why nobody has
ever noticed this.  Looking back at the git history, I don't think this
_ever_ worked from when alloc_buddy_huge_page() was introduced in
7893d1d5, 8 years ago.

The fix is to pass vma/addr down in to the places where we actually call
in to the buddy allocator.  It's fairly straightforward plumbing.  This
has been lightly tested.

Signed-off-by: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Hillf Danton <hillf.zj@alibaba-inc.com>
Cc: David Rientjes <rientjes@google.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agomm/hugetlb: make node_hstates array static
Alexander Kuleshov [Fri, 6 Nov 2015 02:50:14 +0000 (18:50 -0800)]
mm/hugetlb: make node_hstates array static

There are no users of the node_hstates array outside of the
mm/hugetlb.c. So let's make it static.

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>
8 years agomm/maccess.c: actually return -EFAULT from strncpy_from_unsafe
Rasmus Villemoes [Fri, 6 Nov 2015 02:50:11 +0000 (18:50 -0800)]
mm/maccess.c: actually return -EFAULT from strncpy_from_unsafe

As far as I can tell, strncpy_from_unsafe never returns -EFAULT.  ret is
the result of a __copy_from_user_inatomic(), which is 0 for success and
positive (in this case necessarily 1) for access error - it is never
negative.  So we were always returning the length of the, possibly
truncated, destination string.

Signed-off-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agomm/cma.c: suppress warning
Andrew Morton [Fri, 6 Nov 2015 02:50:08 +0000 (18:50 -0800)]
mm/cma.c: suppress warning

mm/cma.c: In function 'cma_alloc':
mm/cma.c:366: warning: 'pfn' may be used uninitialized in this function

The patch actually improves the tracing a bit: if alloc_contig_range()
fails, tracing will display the offending pfn rather than -1.

Cc: Stefan Strogin <stefan.strogin@gmail.com>
Cc: Michal Nazarewicz <mpn@google.com>
Cc: Marek Szyprowski <m.szyprowski@samsung.com>
Cc: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>
Cc: Thierry Reding <treding@nvidia.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agomm: migrate dirty page without clear_page_dirty_for_io etc
Hugh Dickins [Fri, 6 Nov 2015 02:50:05 +0000 (18:50 -0800)]
mm: migrate dirty page without clear_page_dirty_for_io etc

clear_page_dirty_for_io() has accumulated writeback and memcg subtleties
since v2.6.16 first introduced page migration; and the set_page_dirty()
which completed its migration of PageDirty, later had to be moderated to
__set_page_dirty_nobuffers(); then PageSwapBacked had to skip that too.

No actual problems seen with this procedure recently, but if you look into
what the clear_page_dirty_for_io(page)+set_page_dirty(newpage) is actually
achieving, it turns out to be nothing more than moving the PageDirty flag,
and its NR_FILE_DIRTY stat from one zone to another.

It would be good to avoid a pile of irrelevant decrementations and
incrementations, and improper event counting, and unnecessary descent of
the radix_tree under tree_lock (to set the PAGECACHE_TAG_DIRTY which
radix_tree_replace_slot() left in place anyway).

Do the NR_FILE_DIRTY movement, like the other stats movements, while
interrupts still disabled in migrate_page_move_mapping(); and don't even
bother if the zone is the same.  Do the PageDirty movement there under
tree_lock too, where old page is frozen and newpage not yet visible:
bearing in mind that as soon as newpage becomes visible in radix_tree, an
un-page-locked set_page_dirty() might interfere (or perhaps that's just
not possible: anything doing so should already hold an additional
reference to the old page, preventing its migration; but play safe).

But we do still need to transfer PageDirty in migrate_page_copy(), for
those who don't go the mapping route through migrate_page_move_mapping().

Signed-off-by: Hugh Dickins <hughd@google.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Davidlohr Bueso <dave@stgolabs.net>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Sasha Levin <sasha.levin@oracle.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agomm: page migration avoid touching newpage until no going back
Hugh Dickins [Fri, 6 Nov 2015 02:50:02 +0000 (18:50 -0800)]
mm: page migration avoid touching newpage until no going back

We have had trouble in the past from the way in which page migration's
newpage is initialized in dribs and drabs - see commit 8bdd63809160 ("mm:
fix direct reclaim writeback regression") which proposed a cleanup.

We have no actual problem now, but I think the procedure would be clearer
(and alternative get_new_page pools safer to implement) if we assert that
newpage is not touched until we are sure that it's going to be used -
except for taking the trylock on it in __unmap_and_move().

So shift the early initializations from move_to_new_page() into
migrate_page_move_mapping(), mapping and NULL-mapping paths.  Similarly
migrate_huge_page_move_mapping(), but its NULL-mapping path can just be
deleted: you cannot reach hugetlbfs_migrate_page() with a NULL mapping.

Adjust stages 3 to 8 in the Documentation file accordingly.

Signed-off-by: Hugh Dickins <hughd@google.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Davidlohr Bueso <dave@stgolabs.net>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Sasha Levin <sasha.levin@oracle.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>