platform/kernel/linux-rpi.git
9 years agoMAINTAINERS: Use tabs consistently
Joe Perches [Wed, 15 Apr 2015 23:17:31 +0000 (16:17 -0700)]
MAINTAINERS: Use tabs consistently

Consistently use a single tab after the "specifier:" type.

Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agolib/string_helpers.c: change semantics of string_escape_mem
Rasmus Villemoes [Wed, 15 Apr 2015 23:17:28 +0000 (16:17 -0700)]
lib/string_helpers.c: change semantics of string_escape_mem

The current semantics of string_escape_mem are inadequate for one of its
current users, vsnprintf().  If that is to honour its contract, it must
know how much space would be needed for the entire escaped buffer, and
string_escape_mem provides no way of obtaining that (short of allocating a
large enough buffer (~4 times input string) to let it play with, and
that's definitely a big no-no inside vsnprintf).

So change the semantics for string_escape_mem to be more snprintf-like:
Return the size of the output that would be generated if the destination
buffer was big enough, but of course still only write to the part of dst
it is allowed to, and (contrary to snprintf) don't do '\0'-termination.
It is then up to the caller to detect whether output was truncated and to
append a '\0' if desired.  Also, we must output partial escape sequences,
otherwise a call such as snprintf(buf, 3, "%1pE", "\123") would cause
printf to write a \0 to buf[2] but leaving buf[0] and buf[1] with whatever
they previously contained.

This also fixes a bug in the escaped_string() helper function, which used
to unconditionally pass a length of "end-buf" to string_escape_mem();
since the latter doesn't check osz for being insanely large, it would
happily write to dst.  For example, kasprintf(GFP_KERNEL, "something and
then %pE", ...); is an easy way to trigger an oops.

In test-string_helpers.c, the -ENOMEM test is replaced with testing for
getting the expected return value even if the buffer is too small.  We
also ensure that nothing is written (by relying on a NULL pointer deref)
if the output size is 0 by passing NULL - this has to work for
kasprintf("%pE") to work.

In net/sunrpc/cache.c, I think qword_add still has the same semantics.
Someone should definitely double-check this.

In fs/proc/array.c, I made the minimum possible change, but longer-term it
should stop poking around in seq_file internals.

[andriy.shevchenko@linux.intel.com: simplify qword_add]
[andriy.shevchenko@linux.intel.com: add missed curly braces]
Signed-off-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Acked-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: 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>
9 years agolib/string_helpers.c: refactor string_escape_mem
Rasmus Villemoes [Wed, 15 Apr 2015 23:17:25 +0000 (16:17 -0700)]
lib/string_helpers.c: refactor string_escape_mem

When printf is given the format specifier %pE, it needs a way of obtaining
the total output size that would be generated if the buffer was large
enough, and string_escape_mem doesn't easily provide that.  This is a
refactorization of string_escape_mem in preparation of changing its
external API to provide that information.

The somewhat ugly early returns and subsequent seemingly redundant
conditionals are to make the following patch touch as little as possible
in string_helpers.c while still preserving the current behaviour of never
outputting partial escape sequences.  That behaviour must also change for
%pE to work as one expects from every other printf specifier.

Signed-off-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Acked-by: 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>
9 years agolib/vsprintf.c: fix potential NULL deref in hex_string
Rasmus Villemoes [Wed, 15 Apr 2015 23:17:23 +0000 (16:17 -0700)]
lib/vsprintf.c: fix potential NULL deref in hex_string

The helper hex_string() is broken in two ways.  First, it doesn't
increment buf regardless of whether there is room to print, so callers
such as kasprintf() that try to probe the correct storage to allocate will
get a too small return value.  But even worse, kasprintf() (and likely
anyone else trying to find the size of the result) pass NULL for buf and 0
for size, so we also have end == NULL.  But this means that the end-1 in
hex_string() is (char*)-1, so buf < end-1 is true and we get a NULL
pointer deref.  I double-checked this with a trivial kernel module that
just did a kasprintf(GFP_KERNEL, "%14ph", "CrashBoomBang").

Nobody seems to be using %ph with kasprintf, but we might as well fix it
before it hits someone.

Signed-off-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Acked-by: 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>
9 years agolib/vsprintf: add %pC{,n,r} format specifiers for clocks
Geert Uytterhoeven [Wed, 15 Apr 2015 23:17:20 +0000 (16:17 -0700)]
lib/vsprintf: add %pC{,n,r} format specifiers for clocks

Add format specifiers for printing struct clk:
  - '%pC' or '%pCn': name (Common Clock Framework) or address (legacy
    clock framework) of the clock,
  - '%pCr': rate of the clock.

[akpm@linux-foundation.org: omit code if !CONFIG_HAVE_CLK]
Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Mike Turquette <mturquette@linaro.org>
Cc: Stephen Boyd <sboyd@codeaurora.org>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agolib/vsprintf: Move integer format types to the top
Geert Uytterhoeven [Wed, 15 Apr 2015 23:17:17 +0000 (16:17 -0700)]
lib/vsprintf: Move integer format types to the top

Move the format types for 64-bit integers and configurable size integers
to the top, so they're next to the other integer format types.  While at
it, add the missing format types for s32 and u32.

Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Mike Turquette <mturquette@linaro.org>
Cc: Stephen Boyd <sboyd@codeaurora.org>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agolib/vsprintf: document %p parameters passed by reference
Geert Uytterhoeven [Wed, 15 Apr 2015 23:17:14 +0000 (16:17 -0700)]
lib/vsprintf: document %p parameters passed by reference

This patch series improves the documentation for printk() formats, and
adds support for printing clocks.  The latter has always been a hassle if
you wanted to support both the common and legacy clock frameworks.

  - '%pC' and '%pCn' print the name (Common Clock Framework) or address
    (legacy clock framework) of a clock,
  - '%pCr' prints the current clock rate.

This patch (of 3):

Make sure all %p extensions that take parameters by references are
documented to do so.

Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Mike Turquette <mturquette@linaro.org>
Cc: Stephen Boyd <sboyd@codeaurora.org>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agolib/vsprintf.c: another small hack
Rasmus Villemoes [Wed, 15 Apr 2015 23:17:11 +0000 (16:17 -0700)]
lib/vsprintf.c: another small hack

Making ZEROPAD == '0'-' ', we can eliminate a few more instructions.

Signed-off-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Tejun Heo <tj@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agolib/vsprintf.c: eliminate duplicate hex string array
Rasmus Villemoes [Wed, 15 Apr 2015 23:17:08 +0000 (16:17 -0700)]
lib/vsprintf.c: eliminate duplicate hex string array

gcc doesn't merge or overlap const char[] objects with identical contents
(probably language lawyers would also insist that these things have
different addresses), but there's no reason to have the string
"0123456789ABCDEF" occur in multiple places.  hex_asc_upper is declared in
kernel.h and defined in lib/hexdump.c, which is unconditionally compiled
in.

Signed-off-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Tejun Heo <tj@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agolib/vsprintf.c: reduce stack use in number()
Rasmus Villemoes [Wed, 15 Apr 2015 23:17:05 +0000 (16:17 -0700)]
lib/vsprintf.c: reduce stack use in number()

At least since the initial git commit, when base was passed as a separate
parameter, number() has only been called with bases 8, 10 and 16.  I'm
guessing that 66 was to accommodate 64 0/1, a sign and a '\0', but the
buffer is only used for the actual digits.  Octal digits carry 3 bits of
information, so 24 is enough.  Spell that 3*sizeof(num) so one less place
needs to be changed should long long ever be 128 bits.  Also remove the
commented-out code that would handle an arbitrary base.

Signed-off-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Tejun Heo <tj@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agolib/vsprintf.c: eliminate some branches
Rasmus Villemoes [Wed, 15 Apr 2015 23:17:02 +0000 (16:17 -0700)]
lib/vsprintf.c: eliminate some branches

Since FORMAT_TYPE_INT is simply 1 more than FORMAT_TYPE_UINT, and
similarly for BYTE/UBYTE, SHORT/USHORT, LONG/ULONG, we can eliminate a few
instructions by making SIGN have the value 1 instead of 2, and then use
arithmetic instead of branches for computing the right spec->type.  It's a
little hacky, but certainly in the same spirit as SMALL needing to have
the value 0x20.  For example for the spec->qualifier == 'l' case, gcc now
generates

     75e:       0f b6 53 01             movzbl 0x1(%rbx),%edx
     762:       83 e2 01                and    $0x1,%edx
     765:       83 c2 09                add    $0x9,%edx
     768:       88 13                   mov    %dl,(%rbx)

instead of

     763:       0f b6 53 01             movzbl 0x1(%rbx),%edx
     767:       83 e2 02                and    $0x2,%edx
     76a:       80 fa 01                cmp    $0x1,%dl
     76d:       19 d2                   sbb    %edx,%edx
     76f:       83 c2 0a                add    $0xa,%edx
     772:       88 13                   mov    %dl,(%rbx)

Signed-off-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Tejun Heo <tj@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agoprintk: comment pr_cont() stating it is only to continue a line
Steven Rostedt [Wed, 15 Apr 2015 23:16:59 +0000 (16:16 -0700)]
printk: comment pr_cont() stating it is only to continue a line

KERN_CONT is nicely commented in kern_levels.h, but pr_cont() is now used
more often, and it lacks the comment stating what it is used for.  It can
be confused as continuing the log level, but that is not its purpose.  Its
purpose is to continue a line that had no newline enclosed.  This should
be documented by pr_cont() as well.

Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Acked-by: Borislav Petkov <bp@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agopowerpc/powernv: reboot when requested by firmware
Joel Stanley [Wed, 15 Apr 2015 23:16:56 +0000 (16:16 -0700)]
powerpc/powernv: reboot when requested by firmware

Use orderly_reboot so userspace will to shut itself down via the reboot
path.  This is required for graceful reboot initiated by the BMC, such as
when a user uses ipmitool to issue a 'chassis power cycle' command.

Signed-off-by: Joel Stanley <joel@jms.id.au>
Acked-by: Michael Ellerman <mpe@ellerman.id.au>
Cc: Fabian Frederick <fabf@skynet.be>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Jeremy Kerr <jk@ozlabs.org>
Cc: David S. Miller <davem@davemloft.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agokernel/reboot.c: add orderly_reboot for graceful reboot
Joel Stanley [Wed, 15 Apr 2015 23:16:53 +0000 (16:16 -0700)]
kernel/reboot.c: add orderly_reboot for graceful reboot

The kernel has orderly_poweroff which allows the kernel to initiate a
graceful shutdown of userspace, by running /sbin/poweroff.  This adds
orderly_reboot that will cause userspace to shut itself down by calling
/sbin/reboot.

This will be used for shutdown initiated by a system controller on
platforms that do not use ACPI.

orderly_reboot() should be used when the system wants to allow userspace
to gracefully shut itself down.  For cases where the system may imminently
catch on fire, the existing emergency_restart() provides an immediate
reboot without involving userspace.

Signed-off-by: Joel Stanley <joel@jms.id.au>
Cc: Fabian Frederick <fabf@skynet.be>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Jeremy Kerr <jk@ozlabs.org>
Cc: David S. Miller <davem@davemloft.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agodrivers/sbus/char/envctrl.c: ignore orderly_poweroff return value
Joel Stanley [Wed, 15 Apr 2015 23:16:50 +0000 (16:16 -0700)]
drivers/sbus/char/envctrl.c: ignore orderly_poweroff return value

orderly_poweroff() unconditionally returns 0, so remove the dead code that
checks the return value.

A future patch will change the return type to void.

Signed-off-by: Joel Stanley <joel@jms.id.au>
Acked-by: David S. Miller <davem@davemloft.net>
Cc: Fabian Frederick <fabf@skynet.be>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Jeremy Kerr <jk@ozlabs.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agokernel/hung_task.c: change hung_task.c to use for_each_process_thread()
Aaron Tomlin [Wed, 15 Apr 2015 23:16:47 +0000 (16:16 -0700)]
kernel/hung_task.c: change hung_task.c to use for_each_process_thread()

In check_hung_uninterruptible_tasks() avoid the use of deprecated
while_each_thread().

The "max_count" logic will prevent a livelock - see commit 0c740d0a
("introduce for_each_thread() to replace the buggy while_each_thread()").
Having said this let's use for_each_process_thread().

Signed-off-by: Aaron Tomlin <atomlin@redhat.com>
Acked-by: Oleg Nesterov <oleg@redhat.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Dave Wysochanski <dwysocha@redhat.com>
Cc: Aaron Tomlin <atomlin@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agokernel/resource.c: remove deprecated __check_region() and friends
Jakub Sitnicki [Wed, 15 Apr 2015 23:16:44 +0000 (16:16 -0700)]
kernel/resource.c: remove deprecated __check_region() and friends

All users of __check_region(), check_region(), and check_mem_region() are
gone.  We got rid of the last user in v4.0-rc1.  Remove them.

bloat-o-meter on x86_64 shows:

add/remove: 0/3 grow/shrink: 0/0 up/down: 0/-102 (-102)
function                                     old     new   delta
__kstrtab___check_region                      15       -     -15
__ksymtab___check_region                      16       -     -16
__check_region                                71       -     -71

Signed-off-by: Jakub Sitnicki <jsitnicki@gmail.com>
Cc: Bjorn Helgaas <bhelgaas@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agokernel: conditionally support non-root users, groups and capabilities
Iulia Manda [Wed, 15 Apr 2015 23:16:41 +0000 (16:16 -0700)]
kernel: conditionally support non-root users, groups and capabilities

There are a lot of embedded systems that run most or all of their
functionality in init, running as root:root.  For these systems,
supporting multiple users is not necessary.

This patch adds a new symbol, CONFIG_MULTIUSER, that makes support for
non-root users, non-root groups, and capabilities optional.  It is enabled
under CONFIG_EXPERT menu.

When this symbol is not defined, UID and GID are zero in any possible case
and processes always have all capabilities.

The following syscalls are compiled out: setuid, setregid, setgid,
setreuid, setresuid, getresuid, setresgid, getresgid, setgroups,
getgroups, setfsuid, setfsgid, capget, capset.

Also, groups.c is compiled out completely.

In kernel/capability.c, capable function was moved in order to avoid
adding two ifdef blocks.

This change saves about 25 KB on a defconfig build.  The most minimal
kernels have total text sizes in the high hundreds of kB rather than
low MB.  (The 25k goes down a bit with allnoconfig, but not that much.

The kernel was booted in Qemu.  All the common functionalities work.
Adding users/groups is not possible, failing with -ENOSYS.

Bloat-o-meter output:
add/remove: 7/87 grow/shrink: 19/397 up/down: 1675/-26325 (-24650)

[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Iulia Manda <iulia.manda21@gmail.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
Acked-by: Geert Uytterhoeven <geert@linux-m68k.org>
Tested-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agolib/test-hexdump.c: fix initconst confusion
Andi Kleen [Wed, 15 Apr 2015 23:16:39 +0000 (16:16 -0700)]
lib/test-hexdump.c: fix initconst confusion

const char *...[] is not const, but an array of pointer to const.  So
these arrays cannot be __initconst, but must be __initdata

This fixes section conflicts with LTO.

Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agoparide: fix the "verbose" module param
Dan Carpenter [Wed, 15 Apr 2015 23:16:36 +0000 (16:16 -0700)]
paride: fix the "verbose" module param

The verbose module parameter can be set to 2 for extremely verbose
messages so the type should be int instead of bool.

Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Cc: Tim Waugh <tim@cyberelk.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agoinclude/linux: remove empty conditionals
Rasmus Villemoes [Wed, 15 Apr 2015 23:16:33 +0000 (16:16 -0700)]
include/linux: remove empty conditionals

Commit 607ca46e97a1 ("UAPI: (Scripted) Disintegrate include/linux") left
behind some empty conditional blocks.  Since they are useless and may
cause a reader to wonder whether something is missing, remove them.

Signed-off-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years ago/proc/PID/status: show all sets of pid according to ns
Chen Hanxiao [Wed, 15 Apr 2015 23:16:30 +0000 (16:16 -0700)]
/proc/PID/status: show all sets of pid according to ns

If some issues occurred inside a container guest, host user could not know
which process is in trouble just by guest pid: the users of container
guest only knew the pid inside containers.  This will bring obstacle for
trouble shooting.

This patch adds four fields: NStgid, NSpid, NSpgid and NSsid:

a) In init_pid_ns, nothing changed;

b) In one pidns, will tell the pid inside containers:
  NStgid: 21776   5       1
  NSpid:  21776   5       1
  NSpgid: 21776   5       1
  NSsid:  21729   1       0
  ** Process id is 21776 in level 0, 5 in level 1, 1 in level 2.

c) If pidns is nested, it depends on which pidns are you in.
  NStgid: 5       1
  NSpid:  5       1
  NSpgid: 5       1
  NSsid:  1       0
  ** Views from level 1

[akpm@linux-foundation.org: add CONFIG_PID_NS ifdef]
Signed-off-by: Chen Hanxiao <chenhanxiao@cn.fujitsu.com>
Acked-by: Serge Hallyn <serge.hallyn@canonical.com>
Acked-by: "Eric W. Biederman" <ebiederm@xmission.com>
Tested-by: Serge Hallyn <serge.hallyn@canonical.com>
Tested-by: Nathan Scott <nathans@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agozram: fix error return code
Julia Lawall [Wed, 15 Apr 2015 23:16:27 +0000 (16:16 -0700)]
zram: fix error return code

Return a negative error code on failure.

A simplified version of the semantic match that finds this problem is as
follows: (http://coccinelle.lip6.fr/)

// <smpl>
@@
identifier ret; expression e1,e2;
@@
(
if (\(ret < 0\|ret != 0\))
 { ... return ret; }
|
ret = 0
)
... when != ret = e1
    when != &ret
*if(...)
{
  ... when != ret = e2
      when forall
 return ret;
}
// </smpl>

Signed-off-by: Julia Lawall <Julia.Lawall@lip6.fr>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Nitin Gupta <ngupta@vflare.org>
Acked-by: Sergey Senozhatsky <sergey.senozhatsky.work@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agozsmalloc: remove extra cond_resched() in __zs_compact
Sergey Senozhatsky [Wed, 15 Apr 2015 23:16:24 +0000 (16:16 -0700)]
zsmalloc: remove extra cond_resched() in __zs_compact

Do not perform cond_resched() before the busy compaction loop in
__zs_compact(), because this loop does it when needed.

Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Acked-by: Minchan Kim <minchan@kernel.org>
Cc: Nitin Gupta <ngupta@vflare.org>
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>
9 years agozsmalloc: fix fatal corruption due to wrong size class selection
Heesub Shin [Wed, 15 Apr 2015 23:16:21 +0000 (16:16 -0700)]
zsmalloc: fix fatal corruption due to wrong size class selection

There is no point in overriding the size class below.  It causes fatal
corruption on the next chunk on the 3264-bytes size class, which is the
last size class that is not huge.

For example, if the requested size was exactly 3264 bytes, current
zsmalloc allocates and returns a chunk from the size class of 3264 bytes,
not 4096.  User access to this chunk may overwrite head of the next
adjacent chunk.

Here is the panic log captured when freelist was corrupted due to this:

    Kernel BUG at ffffffc00030659c [verbose debug info unavailable]
    Internal error: Oops - BUG: 96000006 [#1] PREEMPT SMP
    Modules linked in:
    exynos-snapshot: core register saved(CPU:5)
    CPUMERRSR: 0000000000000000, L2MERRSR: 0000000000000000
    exynos-snapshot: context saved(CPU:5)
    exynos-snapshot: item - log_kevents is disabled
    CPU: 5 PID: 898 Comm: kswapd0 Not tainted 3.10.61-4497415-eng #1
    task: ffffffc0b8783d80 ti: ffffffc0b71e8000 task.ti: ffffffc0b71e8000
    PC is at obj_idx_to_offset+0x0/0x1c
    LR is at obj_malloc+0x44/0xe8
    pc : [<ffffffc00030659c>] lr : [<ffffffc000306604>] pstate: a0000045
    sp : ffffffc0b71eb790
    x29: ffffffc0b71eb790 x28: ffffffc00204c000
    x27: 000000000001d96f x26: 0000000000000000
    x25: ffffffc098cc3500 x24: ffffffc0a13f2810
    x23: ffffffc098cc3501 x22: ffffffc0a13f2800
    x21: 000011e1a02006e3 x20: ffffffc0a13f2800
    x19: ffffffbc02a7e000 x18: 0000000000000000
    x17: 0000000000000000 x16: 0000000000000feb
    x15: 0000000000000000 x14: 00000000a01003e3
    x13: 0000000000000020 x12: fffffffffffffff0
    x11: ffffffc08b264000 x10: 00000000e3a01004
    x9 : ffffffc08b263fea x8 : ffffffc0b1e611c0
    x7 : ffffffc000307d24 x6 : 0000000000000000
    x5 : 0000000000000038 x4 : 000000000000011e
    x3 : ffffffbc00003e90 x2 : 0000000000000cc0
    x1 : 00000000d0100371 x0 : ffffffbc00003e90

Reported-by: Sooyong Suk <s.suk@samsung.com>
Signed-off-by: Heesub Shin <heesub.shin@samsung.com>
Tested-by: Sooyong Suk <s.suk@samsung.com>
Acked-by: Minchan Kim <minchan@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agozsmalloc: remove unnecessary insertion/removal of zspage in compaction
Minchan Kim [Wed, 15 Apr 2015 23:16:18 +0000 (16:16 -0700)]
zsmalloc: remove unnecessary insertion/removal of zspage in compaction

In putback_zspage, we don't need to insert a zspage into list of zspage
in size_class again to just fix fullness group. We could do directly
without reinsertion so we could save some instuctions.

Reported-by: Heesub Shin <heesub.shin@samsung.com>
Signed-off-by: Minchan Kim <minchan@kernel.org>
Cc: Nitin Gupta <ngupta@vflare.org>
Cc: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Cc: Dan Streetman <ddstreet@ieee.org>
Cc: Seth Jennings <sjennings@variantweb.net>
Cc: Ganesh Mahendran <opensource.ganesh@gmail.com>
Cc: Luigi Semenzato <semenzato@google.com>
Cc: Gunho Lee <gunho.lee@lge.com>
Cc: Juneho Choi <juno.choi@lge.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agozsmalloc: micro-optimize zs_object_copy()
Sergey Senozhatsky [Wed, 15 Apr 2015 23:16:15 +0000 (16:16 -0700)]
zsmalloc: micro-optimize zs_object_copy()

A micro-optimization.  Avoid additional branching and reduce (a bit)
registry pressure (f.e.  s_off += size; d_off += size; may be calculated
twise: first for >= PAGE_SIZE check and later for offset update in "else"
clause).

scripts/bloat-o-meter shows some improvement

add/remove: 0/0 grow/shrink: 0/1 up/down: 0/-10 (-10)
function                          old     new   delta
zs_object_copy                    550     540     -10

Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Acked-by: Minchan Kim <minchan@kernel.org>
Cc: Nitin Gupta <ngupta@vflare.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agozsmalloc: remove synchronize_rcu from zs_compact()
Sergey Senozhatsky [Wed, 15 Apr 2015 23:16:12 +0000 (16:16 -0700)]
zsmalloc: remove synchronize_rcu from zs_compact()

Do not synchronize rcu in zs_compact(). Neither zsmalloc not
zram use rcu.

Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Acked-by: Minchan Kim <minchan@kernel.org>
Cc: Nitin Gupta <ngupta@vflare.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agozram: deprecate zram attrs sysfs nodes
Sergey Senozhatsky [Wed, 15 Apr 2015 23:16:09 +0000 (16:16 -0700)]
zram: deprecate zram attrs sysfs nodes

Add Documentation/ABI/obsolete/sysfs-block-zram file and list obsolete and
deprecated attributes there.  The patch also adds additional information
to zram documentation and describes the basic strategy:

- the existing RW nodes will be downgraded to WO nodes (in 4.11)
- deprecated RO sysfs nodes will eventually be removed (in 4.11)

Users will be additionally notified about deprecated attr usage by
pr_warn_once() (added to every deprecated attr _show()), as suggested by
Minchan Kim.

User space is advised to use zram<id>/stat, zram<id>/io_stat and
zram<id>/mm_stat files.

Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Reported-by: Minchan Kim <minchan@kernel.org>
Cc: Nitin Gupta <ngupta@vflare.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agozram: export new 'mm_stat' sysfs attrs
Sergey Senozhatsky [Wed, 15 Apr 2015 23:16:06 +0000 (16:16 -0700)]
zram: export new 'mm_stat' sysfs attrs

Per-device `zram<id>/mm_stat' file provides mm statistics of a particular
zram device in a format similar to block layer statistics.  The file
consists of a single line and represents the following stats (separated by
whitespace):

        orig_data_size
        compr_data_size
        mem_used_total
        mem_limit
        mem_used_max
        zero_pages
        num_migrated

Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Acked-by: Minchan Kim <minchan@kernel.org>
Cc: Nitin Gupta <ngupta@vflare.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agozram: export new 'io_stat' sysfs attrs
Sergey Senozhatsky [Wed, 15 Apr 2015 23:16:03 +0000 (16:16 -0700)]
zram: export new 'io_stat' sysfs attrs

Per-device `zram<id>/io_stat' file provides accumulated I/O statistics of
particular zram device in a format similar to block layer statistics.  The
file consists of a single line and represents the following stats
(separated by whitespace):

        failed_reads
        failed_writes
        invalid_io
        notify_free

Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Acked-by: Minchan Kim <minchan@kernel.org>
Cc: Nitin Gupta <ngupta@vflare.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agozram: describe device attrs in documentation
Sergey Senozhatsky [Wed, 15 Apr 2015 23:16:00 +0000 (16:16 -0700)]
zram: describe device attrs in documentation

Briefly describe exported device stat attrs in zram documentation.  We
will eventually get rid of per-stat sysfs nodes and, thus, clean up
Documentation/ABI/testing/sysfs-block-zram file, which is the only source
of information about device sysfs nodes.

Add `num_migrated' description, since there is no independent
`num_migrated' sysfs node (and no corresponding sysfs-block-zram entry),
it will be exported via zram<id>/mm_stat file.

At this point we can provide minimal description, because sysfs-block-zram
still contains detailed information.

Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Acked-by: Minchan Kim <minchan@kernel.org>
Cc: Nitin Gupta <ngupta@vflare.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agozram: use generic start/end io accounting
Sergey Senozhatsky [Wed, 15 Apr 2015 23:15:57 +0000 (16:15 -0700)]
zram: use generic start/end io accounting

Use bio generic_start_io_acct() and generic_end_io_acct() to account
device's block layer statistics.  This will let users to monitor zram
activities using sysstat and similar packages/tools.

Apart from the usual per-stat sysfs attr, zram IO stats are now also
available in '/sys/block/zram<id>/stat' and '/proc/diskstats' files.

We will slowly get rid of per-stat sysfs files.

Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Acked-by: Minchan Kim <minchan@kernel.org>
Cc: Nitin Gupta <ngupta@vflare.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agozram: move compact_store() to sysfs functions area
Sergey Senozhatsky [Wed, 15 Apr 2015 23:15:55 +0000 (16:15 -0700)]
zram: move compact_store() to sysfs functions area

A cosmetic change.  We have a new code layout and keep zram per-device
sysfs store and show functions in one place.  Move compact_store() to that
handlers block to conform to current layout.

Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Acked-by: Minchan Kim <minchan@kernel.org>
Cc: Nitin Gupta <ngupta@vflare.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agozram: remove `num_migrated' device attr
Sergey Senozhatsky [Wed, 15 Apr 2015 23:15:52 +0000 (16:15 -0700)]
zram: remove `num_migrated' device attr

This patch introduces rework to zram stats.  We have per-stat sysfs nodes,
and it makes things a bit hard to use in user space: it doesn't give an
immediate stats 'snapshot', it requires user space to use more syscalls -
open, read, close for every stat file, with appropriate error checks on
every step, etc.

First, zram now accounts block layer statistics, available in
/sys/block/zram<id>/stat and /proc/diskstats files.  So some new stats are
available (see Documentation/block/stat.txt), besides, zram's activities
now can be monitored by sysstat's iostat or similar tools.

Example:
cat /sys/block/zram0/stat
248     0    1984    0   251029     0  2008232   5120   0   5116   5116

Second, group currently exported on per-stat basis nodes into two
categories (files):

-- zram<id>/io_stat
accumulates device's IO stats, that are not accounted by block layer,
and contains:
        failed_reads
        failed_writes
        invalid_io
        notify_free

Example:
cat /sys/block/zram0/io_stat
0        0        0   652572

-- zram<id>/mm_stat
accumulates zram mm stats and contains:
        orig_data_size
        compr_data_size
        mem_used_total
        mem_limit
        mem_used_max
        zero_pages
        num_migrated

Example:
cat /sys/block/zram0/mm_stat
434634752 270288572 279158784        0 579895296    15060        0

per-stat sysfs nodes are now considered to be deprecated and we plan to
remove them (and clean up some of the existing stat code) in two years (as
of now, there is no warning printed to syslog about deprecated stats being
used).  User space is advised to use the above mentioned 3 files.

This patch (of 7):

Remove sysfs `num_migrated' attribute.  We are moving away from per-stat
device attrs towards 3 stat files that will accumulate io and mm stats in
a format similar to block layer statistics in /sys/block/<dev>/stat.  That
will be easier to use in user space, and reduce the number of syscalls
needed to read zram device statistics.

`num_migrated' will return back in zram<id>/mm_stat file.

Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Acked-by: Minchan Kim <minchan@kernel.org>
Cc: Nitin Gupta <ngupta@vflare.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm/zsmalloc.c: fix comment for get_pages_per_zspage
Yinghao Xie [Wed, 15 Apr 2015 23:15:49 +0000 (16:15 -0700)]
mm/zsmalloc.c: fix comment for get_pages_per_zspage

Signed-off-by: Yinghao Xie <yinghao.xie@sumsung.com>
Suggested-by: Minchan Kim <minchan@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agozsmalloc: zsmalloc documentation
Minchan Kim [Wed, 15 Apr 2015 23:15:46 +0000 (16:15 -0700)]
zsmalloc: zsmalloc documentation

Create zsmalloc doc which explains design concept and stat information.

Signed-off-by: Minchan Kim <minchan@kernel.org>
Cc: Juneho Choi <juno.choi@lge.com>
Cc: Gunho Lee <gunho.lee@lge.com>
Cc: Luigi Semenzato <semenzato@google.com>
Cc: Dan Streetman <ddstreet@ieee.org>
Cc: Seth Jennings <sjennings@variantweb.net>
Cc: Nitin Gupta <ngupta@vflare.org>
Cc: Jerome Marchand <jmarchan@redhat.com>
Cc: Sergey Senozhatsky <sergey.senozhatsky@gmail.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>
9 years agozsmalloc: add fullness into stat
Minchan Kim [Wed, 15 Apr 2015 23:15:42 +0000 (16:15 -0700)]
zsmalloc: add fullness into stat

During investigating compaction, fullness information of each class is
helpful for investigating how the compaction works well.  With that, we
could know how compaction works well more clear on each size class.

Signed-off-by: Minchan Kim <minchan@kernel.org>
Cc: Juneho Choi <juno.choi@lge.com>
Cc: Gunho Lee <gunho.lee@lge.com>
Cc: Luigi Semenzato <semenzato@google.com>
Cc: Dan Streetman <ddstreet@ieee.org>
Cc: Seth Jennings <sjennings@variantweb.net>
Cc: Nitin Gupta <ngupta@vflare.org>
Cc: Jerome Marchand <jmarchan@redhat.com>
Cc: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agozsmalloc: record handle in page->private for huge object
Minchan Kim [Wed, 15 Apr 2015 23:15:39 +0000 (16:15 -0700)]
zsmalloc: record handle in page->private for huge object

We store handle on header of each allocated object so it increases the
size of each object by sizeof(unsigned long).

If zram stores 4096 bytes to zsmalloc(ie, bad compression), zsmalloc needs
4104B-class to add handle.

However, 4104B-class has 1-pages_per_zspage so wasted size by internal
fragment is 8192 - 4104, which is terrible.

So this patch records the handle in page->private on such huge object(ie,
pages_per_zspage == 1 && maxobj_per_zspage == 1) instead of header of each
object so we could use 4096B-class, not 4104B-class.

Signed-off-by: Minchan Kim <minchan@kernel.org>
Cc: Juneho Choi <juno.choi@lge.com>
Cc: Gunho Lee <gunho.lee@lge.com>
Cc: Luigi Semenzato <semenzato@google.com>
Cc: Dan Streetman <ddstreet@ieee.org>
Cc: Seth Jennings <sjennings@variantweb.net>
Cc: Nitin Gupta <ngupta@vflare.org>
Cc: Jerome Marchand <jmarchan@redhat.com>
Cc: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agozram: support compaction
Minchan Kim [Wed, 15 Apr 2015 23:15:36 +0000 (16:15 -0700)]
zram: support compaction

Now that zsmalloc supports compaction, zram can use it.  For the first
step, this patch exports compact knob via sysfs so user can do compaction
via "echo 1 > /sys/block/zram0/compact".

Signed-off-by: Minchan Kim <minchan@kernel.org>
Cc: Juneho Choi <juno.choi@lge.com>
Cc: Gunho Lee <gunho.lee@lge.com>
Cc: Luigi Semenzato <semenzato@google.com>
Cc: Dan Streetman <ddstreet@ieee.org>
Cc: Seth Jennings <sjennings@variantweb.net>
Cc: Nitin Gupta <ngupta@vflare.org>
Cc: Jerome Marchand <jmarchan@redhat.com>
Cc: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agozsmalloc: adjust ZS_ALMOST_FULL
Minchan Kim [Wed, 15 Apr 2015 23:15:33 +0000 (16:15 -0700)]
zsmalloc: adjust ZS_ALMOST_FULL

Curretly, zsmalloc regards a zspage as ZS_ALMOST_EMPTY if the zspage has
under 1/4 used objects(ie, fullness_threshold_frac).  It could make result
in loose packing since zsmalloc migrates only ZS_ALMOST_EMPTY zspage out.

This patch changes the rule so that zsmalloc makes zspage which has above
3/4 used object ZS_ALMOST_FULL so it could make tight packing.

Signed-off-by: Minchan Kim <minchan@kernel.org>
Cc: Juneho Choi <juno.choi@lge.com>
Cc: Gunho Lee <gunho.lee@lge.com>
Cc: Luigi Semenzato <semenzato@google.com>
Cc: Dan Streetman <ddstreet@ieee.org>
Cc: Seth Jennings <sjennings@variantweb.net>
Cc: Nitin Gupta <ngupta@vflare.org>
Cc: Jerome Marchand <jmarchan@redhat.com>
Cc: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agozsmalloc: support compaction
Minchan Kim [Wed, 15 Apr 2015 23:15:30 +0000 (16:15 -0700)]
zsmalloc: support compaction

This patch provides core functions for migration of zsmalloc.  Migraion
policy is simple as follows.

for each size class {
        while {
                src_page = get zs_page from ZS_ALMOST_EMPTY
                if (!src_page)
                        break;
                dst_page = get zs_page from ZS_ALMOST_FULL
                if (!dst_page)
                        dst_page = get zs_page from ZS_ALMOST_EMPTY
                if (!dst_page)
                        break;
                migrate(from src_page, to dst_page);
        }
}

For migration, we need to identify which objects in zspage are allocated
to migrate them out.  We could know it by iterating of freed objects in a
zspage because first_page of zspage keeps free objects singly-linked list
but it's not efficient.  Instead, this patch adds a tag(ie,
OBJ_ALLOCATED_TAG) in header of each object(ie, handle) so we could check
whether the object is allocated easily.

This patch adds another status bit in handle to synchronize between user
access through zs_map_object and migration.  During migration, we cannot
move objects user are using due to data coherency between old object and
new object.

[akpm@linux-foundation.org: zsmalloc.c needs sched.h for cond_resched()]
Signed-off-by: Minchan Kim <minchan@kernel.org>
Cc: Juneho Choi <juno.choi@lge.com>
Cc: Gunho Lee <gunho.lee@lge.com>
Cc: Luigi Semenzato <semenzato@google.com>
Cc: Dan Streetman <ddstreet@ieee.org>
Cc: Seth Jennings <sjennings@variantweb.net>
Cc: Nitin Gupta <ngupta@vflare.org>
Cc: Jerome Marchand <jmarchan@redhat.com>
Cc: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agozsmalloc: factor out obj_[malloc|free]
Minchan Kim [Wed, 15 Apr 2015 23:15:26 +0000 (16:15 -0700)]
zsmalloc: factor out obj_[malloc|free]

In later patch, migration needs some part of functions in zs_malloc and
zs_free so this patch factor out them.

Signed-off-by: Minchan Kim <minchan@kernel.org>
Cc: Juneho Choi <juno.choi@lge.com>
Cc: Gunho Lee <gunho.lee@lge.com>
Cc: Luigi Semenzato <semenzato@google.com>
Cc: Dan Streetman <ddstreet@ieee.org>
Cc: Seth Jennings <sjennings@variantweb.net>
Cc: Nitin Gupta <ngupta@vflare.org>
Cc: Jerome Marchand <jmarchan@redhat.com>
Cc: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agozsmalloc: decouple handle and object
Minchan Kim [Wed, 15 Apr 2015 23:15:23 +0000 (16:15 -0700)]
zsmalloc: decouple handle and object

Recently, we started to use zram heavily and some of issues
popped.

1) external fragmentation

I got a report from Juneho Choi that fork failed although there are plenty
of free pages in the system.  His investigation revealed zram is one of
the culprit to make heavy fragmentation so there was no more contiguous
16K page for pgd to fork in the ARM.

2) non-movable pages

Other problem of zram now is that inherently, user want to use zram as
swap in small memory system so they use zRAM with CMA to use memory
efficiently.  However, unfortunately, it doesn't work well because zRAM
cannot use CMA's movable pages unless it doesn't support compaction.  I
got several reports about that OOM happened with zram although there are
lots of swap space and free space in CMA area.

3) internal fragmentation

zRAM has started support memory limitation feature to limit memory usage
and I sent a patchset(https://lkml.org/lkml/2014/9/21/148) for VM to be
harmonized with zram-swap to stop anonymous page reclaim if zram consumed
memory up to the limit although there are free space on the swap.  One
problem for that direction is zram has no way to know any hole in memory
space zsmalloc allocated by internal fragmentation so zram would regard
swap is full although there are free space in zsmalloc.  For solving the
issue, zram want to trigger compaction of zsmalloc before it decides full
or not.

This patchset is first step to support above issues.  For that, it adds
indirect layer between handle and object location and supports manual
compaction to solve 3th problem first of all.

After this patchset got merged, next step is to make VM aware of zsmalloc
compaction so that generic compaction will move zsmalloced-pages
automatically in runtime.

In my imaginary experiment(ie, high compress ratio data with heavy swap
in/out on 8G zram-swap), data is as follows,

Before =
zram allocated object :      60212066 bytes
zram total used:     140103680 bytes
ratio:         42.98 percent
MemFree:          840192 kB

Compaction

After =
frag ratio after compaction
zram allocated object :      60212066 bytes
zram total used:      76185600 bytes
ratio:         79.03 percent
MemFree:          901932 kB

Juneho reported below in his real platform with small aging.
So, I think the benefit would be bigger in real aging system
for a long time.

- frag_ratio increased 3% (ie, higher is better)
- memfree increased about 6MB
- In buddy info, Normal 2^3: 4, 2^2: 1: 2^1 increased, Highmem: 2^1 21 increased

frag ratio after swap fragment
used :        156677 kbytes
total:        166092 kbytes
frag_ratio :  94
meminfo before compaction
MemFree:           83724 kB
Node 0, zone   Normal  13642   1364     57     10     61     17      9      5      4      0      0
Node 0, zone  HighMem    425     29      1      0      0      0      0      0      0      0      0

num_migrated :  23630
compaction done

frag ratio after compaction
used :        156673 kbytes
total:        160564 kbytes
frag_ratio :  97
meminfo after compaction
MemFree:           89060 kB
Node 0, zone   Normal  14076   1544     67     14     61     17      9      5      4      0      0
Node 0, zone  HighMem    863     50      1      0      0      0      0      0      0      0      0

This patchset adds more logics(about 480 lines) in zsmalloc but when I
tested heavy swapin/out program, the regression for swapin/out speed is
marginal because most of overheads were caused by compress/decompress and
other MM reclaim stuff.

This patch (of 7):

Currently, handle of zsmalloc encodes object's location directly so it
makes support of migration hard.

This patch decouples handle and object via adding indirect layer.  For
that, it allocates handle dynamically and returns it to user.  The handle
is the address allocated by slab allocation so it's unique and we could
keep object's location in the memory space allocated for handle.

With it, we can change object's position without changing handle itself.

Signed-off-by: Minchan Kim <minchan@kernel.org>
Cc: Juneho Choi <juno.choi@lge.com>
Cc: Gunho Lee <gunho.lee@lge.com>
Cc: Luigi Semenzato <semenzato@google.com>
Cc: Dan Streetman <ddstreet@ieee.org>
Cc: Seth Jennings <sjennings@variantweb.net>
Cc: Nitin Gupta <ngupta@vflare.org>
Cc: Jerome Marchand <jmarchan@redhat.com>
Cc: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm/compaction.c: fix "suitable_migration_target() unused" warning
Andrew Morton [Wed, 15 Apr 2015 23:15:20 +0000 (16:15 -0700)]
mm/compaction.c: fix "suitable_migration_target() unused" warning

mm/compaction.c:250:13: warning: 'suitable_migration_target' defined but not used [-Wunused-function]

Reported-by: Fengguang Wu <fengguang.wu@gmail.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agodax: unify ext2/4_{dax,}_file_operations
Boaz Harrosh [Wed, 15 Apr 2015 23:15:17 +0000 (16:15 -0700)]
dax: unify ext2/4_{dax,}_file_operations

The original dax patchset split the ext2/4_file_operations because of the
two NULL splice_read/splice_write in the dax case.

In the vfs if splice_read/splice_write are NULL we then call
default_splice_read/write.

What we do here is make generic_file_splice_read aware of IS_DAX() so the
original ext2/4_file_operations can be used as is.

For write it appears that iter_file_splice_write is just fine.  It uses
the regular f_op->write(file,..) or new_sync_write(file, ...).

Signed-off-by: Boaz Harrosh <boaz@plexistor.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: Dave Chinner <dchinner@redhat.com>
Cc: Matthew Wilcox <willy@linux.intel.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agodax: use pfn_mkwrite to update c/mtime + freeze protection
Boaz Harrosh [Wed, 15 Apr 2015 23:15:14 +0000 (16:15 -0700)]
dax: use pfn_mkwrite to update c/mtime + freeze protection

From: Yigal Korman <yigal@plexistor.com>

[v1]
Without this patch, c/mtime is not updated correctly when mmap'ed page is
first read from and then written to.

A new xfstest is submitted for testing this (generic/080)

[v2]
Jan Kara has pointed out that if we add the
sb_start/end_pagefault pair in the new pfn_mkwrite we
are then fixing another bug where: A user could start
writing to the page while filesystem is frozen.

Signed-off-by: Yigal Korman <yigal@plexistor.com>
Signed-off-by: Boaz Harrosh <boaz@plexistor.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: Matthew Wilcox <matthew.r.wilcox@intel.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm: new pfn_mkwrite same as page_mkwrite for VM_PFNMAP
Boaz Harrosh [Wed, 15 Apr 2015 23:15:11 +0000 (16:15 -0700)]
mm: new pfn_mkwrite same as page_mkwrite for VM_PFNMAP

This will allow FS that uses VM_PFNMAP | VM_MIXEDMAP (no page structs) to
get notified when access is a write to a read-only PFN.

This can happen if we mmap() a file then first mmap-read from it to
page-in a read-only PFN, than we mmap-write to the same page.

We need this functionality to fix a DAX bug, where in the scenario above
we fail to set ctime/mtime though we modified the file.  An xfstest is
attached to this patchset that shows the failure and the fix.  (A DAX
patch will follow)

This functionality is extra important for us, because upon dirtying of a
pmem page we also want to RDMA the page to a remote cluster node.

We define a new pfn_mkwrite and do not reuse page_mkwrite because
  1 - The name ;-)
  2 - But mainly because it would take a very long and tedious
      audit of all page_mkwrite functions of VM_MIXEDMAP/VM_PFNMAP
      users. To make sure they do not now CRASH. For example current
      DAX code (which this is for) would crash.
      If we would want to reuse page_mkwrite, We will need to first
      patch all users, so to not-crash-on-no-page. Then enable this
      patch. But even if I did that I would not sleep so well at night.
      Adding a new vector is the safest thing to do, and is not that
      expensive. an extra pointer at a static function vector per driver.
      Also the new vector is better for performance, because else we
      Will call all current Kernel vectors, so to:
        check-ha-no-page-do-nothing and return.

No need to call it from do_shared_fault because do_wp_page is called to
change pte permissions anyway.

Signed-off-by: Yigal Korman <yigal@plexistor.com>
Signed-off-by: Boaz Harrosh <boaz@plexistor.com>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Matthew Wilcox <matthew.r.wilcox@intel.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Hugh Dickins <hughd@google.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Dave Chinner <david@fromorbit.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm/memory: also print a_ops->readpage in print_bad_pte()
Konstantin Khlebnikov [Wed, 15 Apr 2015 23:15:08 +0000 (16:15 -0700)]
mm/memory: also print a_ops->readpage in print_bad_pte()

A lot of filesystems use generic_file_mmap() and filemap_fault(),
f_op->mmap and vm_ops->fault aren't enough to identify filesystem.

This prints file name, vm_ops->fault, f_op->mmap and a_ops->readpage
(which is almost always implemented and filesystem-specific).

Example:

[   23.676410] BUG: Bad page map in process sh  pte:1b7e6025 pmd:19bbd067
[   23.676887] page:ffffea00006df980 count:4 mapcount:1 mapping:ffff8800196426c0 index:0x97
[   23.677481] flags: 0x10000000000000c(referenced|uptodate)
[   23.677896] page dumped because: bad pte
[   23.678205] addr:00007f52fcb17000 vm_flags:00000075 anon_vma:          (null) mapping:ffff8800196426c0 index:97
[   23.678922] file:libc-2.19.so fault:filemap_fault mmap:generic_file_readonly_mmap readpage:v9fs_vfs_readpage

[akpm@linux-foundation.org: use pr_alert, per Kirill]
Signed-off-by: Konstantin Khlebnikov <khlebnikov@yandex-team.ru>
Cc: Sasha Levin <sasha.levin@oracle.com>
Acked-by: Kirill A. Shutemov <kirill@shutemov.name>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm/mempool.c: kasan: poison mempool elements
Andrey Ryabinin [Wed, 15 Apr 2015 23:15:05 +0000 (16:15 -0700)]
mm/mempool.c: kasan: poison mempool elements

Mempools keep allocated objects in reserved for situations when ordinary
allocation may not be possible to satisfy.  These objects shouldn't be
accessed before they leave the pool.

This patch poison elements when get into the pool and unpoison when they
leave it.  This will let KASan to detect use-after-free of mempool's
elements.

Signed-off-by: Andrey Ryabinin <a.ryabinin@samsung.com>
Tested-by: David Rientjes <rientjes@google.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Dmitry Chernenkov <drcheren@gmail.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: Alexander Potapenko <glider@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm/cma_debug.c: remove blank lines before DEFINE_SIMPLE_ATTRIBUTE()
Andrew Morton [Wed, 15 Apr 2015 23:15:02 +0000 (16:15 -0700)]
mm/cma_debug.c: remove blank lines before DEFINE_SIMPLE_ATTRIBUTE()

Like EXPORT_SYMBOL(): the positioning communicates that the macro pertains
to the immediately preceding function.

Cc: Dmitry Safonov <d.safonov@partner.samsung.com>
Cc: Michal Nazarewicz <mina86@mina86.com>
Cc: Stefan Strogin <stefan.strogin@gmail.com>
Cc: Marek Szyprowski <m.szyprowski@samsung.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Pintu Kumar <pintu.k@samsung.com>
Cc: Weijie Yang <weijie.yang@samsung.com>
Cc: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>
Cc: Vyacheslav Tyrtov <v.tyrtov@samsung.com>
Cc: Aleksei Mateosian <a.mateosian@samsung.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm: cma: add functions to get region pages counters
Dmitry Safonov [Wed, 15 Apr 2015 23:14:59 +0000 (16:14 -0700)]
mm: cma: add functions to get region pages counters

Here are two functions that provide interface to compute/get used size and
size of biggest free chunk in cma region.  Add that information to
debugfs.

[akpm@linux-foundation.org: move debug code from cma.c into cma_debug.c]
[stefan.strogin@gmail.com: move code from cma_get_used() and cma_get_maxchunk() to cma_used_get() and cma_maxchunk_get()]
Signed-off-by: Dmitry Safonov <d.safonov@partner.samsung.com>
Signed-off-by: Stefan Strogin <stefan.strogin@gmail.com>
Acked-by: Michal Nazarewicz <mina86@mina86.com>
Cc: Marek Szyprowski <m.szyprowski@samsung.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Pintu Kumar <pintu.k@samsung.com>
Cc: Weijie Yang <weijie.yang@samsung.com>
Cc: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>
Cc: Vyacheslav Tyrtov <v.tyrtov@samsung.com>
Cc: Aleksei Mateosian <a.mateosian@samsung.com>
Signed-off-by: Stefan Strogin <stefan.strogin@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agothp: cleanup khugepaged startup
Kirill A. Shutemov [Wed, 15 Apr 2015 23:14:56 +0000 (16:14 -0700)]
thp: cleanup khugepaged startup

Few trivial cleanups:

 - no need to call set_recommended_min_free_kbytes() from
   late_initcall() -- start_khugepaged() calls it;

 - no need to call set_recommended_min_free_kbytes() from
   start_khugepaged() if khugepaged is not started;

 - there isn't much point in running start_khugepaged() if we've just
   set transparent_hugepage_flags to zero;

 - start_khugepaged() is misnamed -- it also used to stop the thread;

Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm: uninline and cleanup page-mapping related helpers
Kirill A. Shutemov [Wed, 15 Apr 2015 23:14:53 +0000 (16:14 -0700)]
mm: uninline and cleanup page-mapping related helpers

Most-used page->mapping helper -- page_mapping() -- has already uninlined.
 Let's uninline also page_rmapping() and page_anon_vma().  It saves us
depending on configuration around 400 bytes in text:

   text    data     bss     dec     hex filename
 660318   99254  410000 1169572  11d8a4 mm/built-in.o-before
 659854   99254  410000 1169108  11d6d4 mm/built-in.o

I also tried to make code a bit more clean.

[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Konstantin Khlebnikov <koct9i@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>
9 years agomm: cma: add trace events for CMA allocations and freeings
Stefan Strogin [Wed, 15 Apr 2015 23:14:50 +0000 (16:14 -0700)]
mm: cma: add trace events for CMA allocations and freeings

Add trace events for cma_alloc() and cma_release().

The cma_alloc tracepoint is used both for successful and failed allocations,
in case of allocation failure pfn=-1UL is stored and printed.

Signed-off-by: Stefan Strogin <stefan.strogin@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.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>
9 years agoinclude/linux/mm.h: simplify flag check
Borislav Petkov [Wed, 15 Apr 2015 23:14:47 +0000 (16:14 -0700)]
include/linux/mm.h: simplify flag check

Flip the flag test so that it is the simplest.  No functional change, just
a small readability improvement:

No code changed:

  # arch/x86/kernel/sys_x86_64.o:

   text    data     bss     dec     hex filename
   1551      24       0    1575     627 sys_x86_64.o.before
   1551      24       0    1575     627 sys_x86_64.o.after

md5:
   70708d1b1ad35cc891118a69dc1a63f9  sys_x86_64.o.before.asm
   70708d1b1ad35cc891118a69dc1a63f9  sys_x86_64.o.after.asm

Signed-off-by: Borislav Petkov <bp@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm/memblock.c: add debug output for memblock_add()
Alexander Kuleshov [Wed, 15 Apr 2015 23:14:44 +0000 (16:14 -0700)]
mm/memblock.c: add debug output for memblock_add()

memblock_reserve() calls memblock_reserve_region() which prints debugging
information if 'memblock=debug' was passed on the command line.  This
patch adds the same behaviour, but for memblock_add function().

[akpm@linux-foundation.org: s/memblock_memory/memblock_add/ in message]
Signed-off-by: Alexander Kuleshov <kuleshovmail@gmail.com>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Philipp Hachtmann <phacht@linux.vnet.ibm.com>
Cc: Fabian Frederick <fabf@skynet.be>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Emil Medve <Emilian.Medve@freescale.com>
Cc: Akinobu Mita <akinobu.mita@gmail.com>
Cc: Tang Chen <tangchen@cn.fujitsu.com>
Cc: Tony Luck <tony.luck@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm: hugetlb: cleanup using paeg_huge_active()
Naoya Horiguchi [Wed, 15 Apr 2015 23:14:41 +0000 (16:14 -0700)]
mm: hugetlb: cleanup using paeg_huge_active()

Now we have an easy access to hugepages' activeness, so existing helpers to
get the information can be cleaned up.

[akpm@linux-foundation.org: s/PageHugeActive/page_huge_active/]
Signed-off-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Cc: Hugh Dickins <hughd@google.com>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm: hugetlb: introduce page_huge_active
Naoya Horiguchi [Wed, 15 Apr 2015 23:14:38 +0000 (16:14 -0700)]
mm: hugetlb: introduce page_huge_active

We are not safe from calling isolate_huge_page() on a hugepage
concurrently, which can make the victim hugepage in invalid state and
results in BUG_ON().

The root problem of this is that we don't have any information on struct
page (so easily accessible) about hugepages' activeness.  Note that
hugepages' activeness means just being linked to
hstate->hugepage_activelist, which is not the same as normal pages'
activeness represented by PageActive flag.

Normal pages are isolated by isolate_lru_page() which prechecks PageLRU
before isolation, so let's do similarly for hugetlb with a new
paeg_huge_active().

set/clear_page_huge_active() should be called within hugetlb_lock.  But
hugetlb_cow() and hugetlb_no_page() don't do this, being justified because
in these functions set_page_huge_active() is called right after the
hugepage is allocated and no other thread tries to isolate it.

[akpm@linux-foundation.org: s/PageHugeActive/page_huge_active/, make it return bool]
[fengguang.wu@intel.com: set_page_huge_active() can be static]
Signed-off-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Cc: Hugh Dickins <hughd@google.com>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Fengguang Wu <fengguang.wu@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm: don't call __page_cache_release for hugetlb
Naoya Horiguchi [Wed, 15 Apr 2015 23:14:35 +0000 (16:14 -0700)]
mm: don't call __page_cache_release for hugetlb

__put_compound_page() calls __page_cache_release() to do some freeing
work, but it's obviously for thps, not for hugetlb.  We don't care because
PageLRU is always cleared and page->mem_cgroup is always NULL for hugetlb.
But it's not correct and has potential risks, so let's make it
conditional.

Signed-off-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Cc: Hugh Dickins <hughd@google.com>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm/mmap.c: use while instead of if+goto
Rasmus Villemoes [Wed, 15 Apr 2015 23:14:32 +0000 (16:14 -0700)]
mm/mmap.c: use while instead of if+goto

The creators of the C language gave us the while keyword. Let's use
that instead of synthesizing it from if+goto.

Made possible by 6597d783397a ("mm/mmap.c: replace find_vma_prepare()
with clearer find_vma_links()").

[akpm@linux-foundation.org: fix 80-col overflows]
Signed-off-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Sasha Levin <sasha.levin@oracle.com>
Cc: Cyrill Gorcunov <gorcunov@openvz.org>
Cc: Roman Gushchin <klamm@yandex-team.ru>
Cc: Hugh Dickins <hughd@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm, selftests: test return value of munmap for MAP_HUGETLB memory
David Rientjes [Wed, 15 Apr 2015 23:14:29 +0000 (16:14 -0700)]
mm, selftests: test return value of munmap for MAP_HUGETLB memory

When MAP_HUGETLB memory is unmapped, the length must be hugepage aligned,
otherwise it fails with -EINVAL.

All tests currently behave correctly, but it's better to explcitly test
the return value for completeness and document the requirement, especially
if users copy map_hugetlb.c as a sample implementation.

Signed-off-by: David Rientjes <rientjes@google.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Davide Libenzi <davidel@xmailserver.org>
Cc: Luiz Capitulino <lcapitulino@redhat.com>
Cc: Shuah Khan <shuahkh@osg.samsung.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Joern Engel <joern@logfs.org>
Cc: Jianguo Wu <wujianguo@huawei.com>
Cc: Eric B Munson <emunson@akamai.com>
Acked-by: Michael Ellerman <mpe@ellerman.id.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm, doc: cleanup and clarify munmap behavior for hugetlb memory
David Rientjes [Wed, 15 Apr 2015 23:14:26 +0000 (16:14 -0700)]
mm, doc: cleanup and clarify munmap behavior for hugetlb memory

munmap(2) of hugetlb memory requires a length that is hugepage aligned,
otherwise it may fail.  Add this to the documentation.

This also cleans up the documentation and separates it into logical units:
one part refers to MAP_HUGETLB and another part refers to requirements for
shared memory segments.

Signed-off-by: David Rientjes <rientjes@google.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Davide Libenzi <davidel@xmailserver.org>
Cc: Luiz Capitulino <lcapitulino@redhat.com>
Cc: Shuah Khan <shuahkh@osg.samsung.com>
Acked-by: Hugh Dickins <hughd@google.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Joern Engel <joern@logfs.org>
Cc: Jianguo Wu <wujianguo@huawei.com>
Cc: Eric B Munson <emunson@akamai.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agothp: do not adjust zone water marks if khugepaged is not started
Kirill A. Shutemov [Wed, 15 Apr 2015 23:14:23 +0000 (16:14 -0700)]
thp: do not adjust zone water marks if khugepaged is not started

set_recommended_min_free_kbytes() adjusts zone water marks to be suitable
for khugepaged. We avoid doing this if khugepaged is disabled, but don't
catch the case when khugepaged is failed to start.

Let's address this by checking khugepaged_thread instead of
khugepaged_enabled() in set_recommended_min_free_kbytes().
It's NULL if the kernel thread is stopped or failed to start.

Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agothp: handle errors in hugepage_init() properly
Kirill A. Shutemov [Wed, 15 Apr 2015 23:14:20 +0000 (16:14 -0700)]
thp: handle errors in hugepage_init() properly

We miss error-handling in few cases hugepage_init(). Let's fix that.

Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Acked-by: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm, mempool: poison elements backed by slab allocator
David Rientjes [Wed, 15 Apr 2015 23:14:17 +0000 (16:14 -0700)]
mm, mempool: poison elements backed by slab allocator

Mempools keep elements in a reserved pool for contexts in which allocation
may not be possible.  When an element is allocated from the reserved pool,
its memory contents is the same as when it was added to the reserved pool.

Because of this, elements lack any free poisoning to detect use-after-free
errors.

This patch adds free poisoning for elements backed by the slab allocator.
This is possible because the mempool layer knows the object size of each
element.

When an element is added to the reserved pool, it is poisoned with
POISON_FREE.  When it is removed from the reserved pool, the contents are
checked for POISON_FREE.  If there is a mismatch, a warning is emitted to
the kernel log.

This is only effective for configs with CONFIG_DEBUG_SLAB or
CONFIG_SLUB_DEBUG_ON.

[fabio.estevam@freescale.com: use '%zu' for printing 'size_t' variable]
[arnd@arndb.de: add missing include]
Signed-off-by: David Rientjes <rientjes@google.com>
Cc: Dave Kleikamp <shaggy@kernel.org>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Sebastian Ott <sebott@linux.vnet.ibm.com>
Cc: Mikulas Patocka <mpatocka@redhat.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Fabio Estevam <fabio.estevam@freescale.com>
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm, mempool: disallow mempools based on slab caches with constructors
David Rientjes [Wed, 15 Apr 2015 23:14:14 +0000 (16:14 -0700)]
mm, mempool: disallow mempools based on slab caches with constructors

All occurrences of mempools based on slab caches with object constructors
have been removed from the tree, so disallow creating them.

We can only dereference mem->ctor in mm/mempool.c without including
mm/slab.h in include/linux/mempool.h.  So simply note the restriction,
just like the comment restricting usage of __GFP_ZERO, and warn on kernels
with CONFIG_DEBUG_VM() if such a mempool is allocated from.

We don't want to incur this check on every element allocation, so use
VM_BUG_ON().

Signed-off-by: David Rientjes <rientjes@google.com>
Cc: Dave Kleikamp <shaggy@kernel.org>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Sebastian Ott <sebott@linux.vnet.ibm.com>
Cc: Mikulas Patocka <mpatocka@redhat.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agofs, jfs: remove slab object constructor
David Rientjes [Wed, 15 Apr 2015 23:14:11 +0000 (16:14 -0700)]
fs, jfs: remove slab object constructor

Mempools based on slab caches with object constructors are risky because
element allocation can happen either from the slab cache itself, meaning
the constructor is properly called before returning, or from the mempool
reserve pool, meaning the constructor is not called before returning,
depending on the allocation context.

For this reason, we should disallow creating mempools based on slab caches
that have object constructors.  Callers of mempool_alloc() will be
responsible for properly initializing the returned element.

Then, it doesn't matter if the element came from the slab cache or the
mempool reserved pool.

The only occurrence of a mempool being based on a slab cache with an
object constructor in the tree is in fs/jfs/jfs_metapage.c.  Remove it and
properly initialize the element in alloc_metapage().

At the same time, META_free is never used, so remove it as well.

Signed-off-by: David Rientjes <rientjes@google.com>
Acked-by: Dave Kleikamp <dave.kleikamp@oracle.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Sebastian Ott <sebott@linux.vnet.ibm.com>
Cc: Mikulas Patocka <mpatocka@redhat.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm: remove rest of ACCESS_ONCE() usages
Jason Low [Wed, 15 Apr 2015 23:14:08 +0000 (16:14 -0700)]
mm: remove rest of ACCESS_ONCE() usages

We converted some of the usages of ACCESS_ONCE to READ_ONCE in the mm/
tree since it doesn't work reliably on non-scalar types.

This patch removes the rest of the usages of ACCESS_ONCE, and use the new
READ_ONCE API for the read accesses.  This makes things cleaner, instead
of using separate/multiple sets of APIs.

Signed-off-by: Jason Low <jason.low2@hp.com>
Acked-by: Michal Hocko <mhocko@suse.cz>
Acked-by: Davidlohr Bueso <dave@stgolabs.net>
Acked-by: Rik van Riel <riel@redhat.com>
Reviewed-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm: use READ_ONCE() for non-scalar types
Jason Low [Wed, 15 Apr 2015 23:14:05 +0000 (16:14 -0700)]
mm: use READ_ONCE() for non-scalar types

Commit 38c5ce936a08 ("mm/gup: Replace ACCESS_ONCE with READ_ONCE")
converted ACCESS_ONCE usage in gup_pmd_range() to READ_ONCE, since
ACCESS_ONCE doesn't work reliably on non-scalar types.

This patch also fixes the other ACCESS_ONCE usages in gup_pte_range()
and __get_user_pages_fast() in mm/gup.c

Signed-off-by: Jason Low <jason.low2@hp.com>
Acked-by: Michal Hocko <mhocko@suse.cz>
Acked-by: Davidlohr Bueso <dave@stgolabs.net>
Acked-by: Rik van Riel <riel@redhat.com>
Reviewed-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm/mremap.c: clean up goto just return ERR_PTR
Derek [Wed, 15 Apr 2015 23:14:02 +0000 (16:14 -0700)]
mm/mremap.c: clean up goto just return ERR_PTR

As suggested by Kirill the "goto"s in vma_to_resize aren't necessary, just
change them to explicit return.

Signed-off-by: Derek Che <crquan@ymail.com>
Suggested-by: "Kirill A. Shutemov" <kirill@shutemov.name>
Acked-by: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomremap should return -ENOMEM when __vm_enough_memory fail
Derek [Wed, 15 Apr 2015 23:13:58 +0000 (16:13 -0700)]
mremap should return -ENOMEM when __vm_enough_memory fail

Recently I straced bash behavior in this dd zero pipe to read test, in
part of testing under vm.overcommit_memory=2 (OVERCOMMIT_NEVER mode):

    # dd if=/dev/zero | read x

The bash sub shell is calling mremap to reallocate more and more memory
untill it finally failed -ENOMEM (I expect), or to be killed by system OOM
killer (which should not happen under OVERCOMMIT_NEVER mode); But the
mremap system call actually failed of -EFAULT, which is a surprise to me,
I think it's supposed to be -ENOMEM?  then I wrote this piece of C code
testing confirmed it: https://gist.github.com/crquan/326bde37e1ddda8effe5

    $ ./remap
    allocated one page @0x7f686bf71000, (PAGE_SIZE: 4096)
    grabbed 7680512000 bytes of memory (1875125 pages) @ 00007f6690993000.
    mremap failed Bad address (14).

The -EFAULT comes from the branch of security_vm_enough_memory_mm failure,
underlyingly it calls __vm_enough_memory which returns only 0 for success
or -ENOMEM; So why vma_to_resize needs to return -EFAULT in this case?
this sounds like a mistake to me.

Some more digging into git history:

1) Before commit 119f657c7 ("RLIMIT_AS checking fix") in May 1 2005
   (pre 2.6.12 days) it was returning -ENOMEM for this failure;

2) but commit 119f657c7 ("untangling do_mremap(), part 1") changed it
   accidentally, to what ever is preserved in local ret, which happened to
   be -EFAULT, in a previous assignment;

3) then in commit 54f5de709 code refactoring, it's explicitly returning
   -EFAULT, should be wrong.

Signed-off-by: Derek Che <crquan@ymail.com>
Acked-by: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm/vmalloc: get rid of dirty bitmap inside vmap_block structure
Roman Pen [Wed, 15 Apr 2015 23:13:55 +0000 (16:13 -0700)]
mm/vmalloc: get rid of dirty bitmap inside vmap_block structure

In original implementation of vm_map_ram made by Nick Piggin there were
two bitmaps: alloc_map and dirty_map.  None of them were used as supposed
to be: finding a suitable free hole for next allocation in block.
vm_map_ram allocates space sequentially in block and on free call marks
pages as dirty, so freed space can't be reused anymore.

Actually it would be very interesting to know the real meaning of those
bitmaps, maybe implementation was incomplete, etc.

But long time ago Zhang Yanfei removed alloc_map by these two commits:

  mm/vmalloc.c: remove dead code in vb_alloc
     3fcd76e8028e0be37b02a2002b4f56755daeda06
  mm/vmalloc.c: remove alloc_map from vmap_block
     b8e748b6c32999f221ea4786557b8e7e6c4e4e7a

In this patch I replaced dirty_map with two range variables: dirty min and
max.  These variables store minimum and maximum position of dirty space in
a block, since we need only to know the dirty range, not exact position of
dirty pages.

Why it was made?  Several reasons: at first glance it seems that
vm_map_ram allocator concerns about fragmentation thus it uses bitmaps for
finding free hole, but it is not true.  To avoid complexity seems it is
better to use something simple, like min or max range values.  Secondly,
code also becomes simpler, without iteration over bitmap, just comparing
values in min and max macros.  Thirdly, bitmap occupies up to 1024 bits
(4MB is a max size of a block).  Here I replaced the whole bitmap with two
longs.

Finally vm_unmap_aliases should be slightly faster and the whole
vmap_block structure occupies less memory.

Signed-off-by: Roman Pen <r.peniaev@gmail.com>
Cc: Zhang Yanfei <zhangyanfei@cn.fujitsu.com>
Cc: Eric Dumazet <edumazet@google.com>
Acked-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: David Rientjes <rientjes@google.com>
Cc: WANG Chao <chaowang@redhat.com>
Cc: Fabian Frederick <fabf@skynet.be>
Cc: Christoph Lameter <cl@linux.com>
Cc: Gioh Kim <gioh.kim@lge.com>
Cc: Rob Jones <rob.jones@codethink.co.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm/vmalloc: occupy newly allocated vmap block just after allocation
Roman Pen [Wed, 15 Apr 2015 23:13:52 +0000 (16:13 -0700)]
mm/vmalloc: occupy newly allocated vmap block just after allocation

Previous implementation allocates new vmap block and repeats search of a
free block from the very beginning, iterating over the CPU free list.

Why it can be better??

1. Allocation can happen on one CPU, but search can be done on another CPU.
   In worst case we preallocate amount of vmap blocks which is equal to
   CPU number on the system.

2. In previous patch I added newly allocated block to the tail of free list
   to avoid soon exhaustion of virtual space and give a chance to occupy
   blocks which were allocated long time ago.  Thus to find newly allocated
   block all the search sequence should be repeated, seems it is not efficient.

In this patch newly allocated block is occupied right away, address of
virtual space is returned to the caller, so there is no any need to repeat
the search sequence, allocation job is done.

Signed-off-by: Roman Pen <r.peniaev@gmail.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Eric Dumazet <edumazet@google.com>
Acked-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: David Rientjes <rientjes@google.com>
Cc: WANG Chao <chaowang@redhat.com>
Cc: Fabian Frederick <fabf@skynet.be>
Cc: Christoph Lameter <cl@linux.com>
Cc: Gioh Kim <gioh.kim@lge.com>
Cc: Rob Jones <rob.jones@codethink.co.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm/vmalloc: fix possible exhaustion of vmalloc space caused by vm_map_ram allocator
Roman Pen [Wed, 15 Apr 2015 23:13:48 +0000 (16:13 -0700)]
mm/vmalloc: fix possible exhaustion of vmalloc space caused by vm_map_ram allocator

Recently I came across high fragmentation of vm_map_ram allocator:
vmap_block has free space, but still new blocks continue to appear.
Further investigation showed that certain mapping/unmapping sequences
can exhaust vmalloc space.  On small 32bit systems that's not a big
problem, cause purging will be called soon on a first allocation failure
(alloc_vmap_area), but on 64bit machines, e.g.  x86_64 has 45 bits of
vmalloc space, that can be a disaster.

1) I came up with a simple allocation sequence, which exhausts virtual
   space very quickly:

  while (iters) {

                /* Map/unmap big chunk */
                vaddr = vm_map_ram(pages, 16, -1, PAGE_KERNEL);
                vm_unmap_ram(vaddr, 16);

                /* Map/unmap small chunks.
                 *
                 * -1 for hole, which should be left at the end of each block
                 * to keep it partially used, with some free space available */
                for (i = 0; i < (VMAP_BBMAP_BITS - 16) / 8 - 1; i++) {
                        vaddr = vm_map_ram(pages, 8, -1, PAGE_KERNEL);
                        vm_unmap_ram(vaddr, 8);
                }
  }

The idea behind is simple:

 1. We have to map a big chunk, e.g. 16 pages.

 2. Then we have to occupy the remaining space with smaller chunks, i.e.
    8 pages. At the end small hole should remain to keep block in free list,
    but do not let big chunk to occupy remaining space.

 3. Goto 1 - allocation request of 16 pages can't be completed (only 8 slots
    are left free in the block in the #2 step), new block will be allocated,
    all further requests will lay into newly allocated block.

To have some measurement numbers for all further tests I setup ftrace and
enabled 4 basic calls in a function profile:

        echo vm_map_ram              > /sys/kernel/debug/tracing/set_ftrace_filter;
        echo alloc_vmap_area        >> /sys/kernel/debug/tracing/set_ftrace_filter;
        echo vm_unmap_ram           >> /sys/kernel/debug/tracing/set_ftrace_filter;
        echo free_vmap_block        >> /sys/kernel/debug/tracing/set_ftrace_filter;

So for this scenario I got these results:

BEFORE (all new blocks are put to the head of a free list)
# cat /sys/kernel/debug/tracing/trace_stat/function0
  Function                               Hit    Time            Avg             s^2
  --------                               ---    ----            ---             ---
  vm_map_ram                          126000    30683.30 us     0.243 us        30819.36 us
  vm_unmap_ram                        126000    22003.24 us     0.174 us        340.886 us
  alloc_vmap_area                       1000    4132.065 us     4.132 us        0.903 us

AFTER (all new blocks are put to the tail of a free list)
# cat /sys/kernel/debug/tracing/trace_stat/function0
  Function                               Hit    Time            Avg             s^2
  --------                               ---    ----            ---             ---
  vm_map_ram                          126000    28713.13 us     0.227 us        24944.70 us
  vm_unmap_ram                        126000    20403.96 us     0.161 us        1429.872 us
  alloc_vmap_area                        993    3916.795 us     3.944 us        29.370 us
  free_vmap_block                        992    654.157 us      0.659 us        1.273 us

SUMMARY:

The most interesting numbers in those tables are numbers of block
allocations and deallocations: alloc_vmap_area and free_vmap_block
calls, which show that before the change blocks were not freed, and
virtual space and physical memory (vmap_block structure allocations,
etc) were consumed.

Average time which were spent in vm_map_ram/vm_unmap_ram became slightly
better.  That can be explained with a reasonable amount of blocks in a
free list, which we need to iterate to find a suitable free block.

2) Another scenario is a random allocation:

  while (iters) {

                /* Randomly take number from a range [1..32/64] */
                nr = rand(1, VMAP_MAX_ALLOC);
                vaddr = vm_map_ram(pages, nr, -1, PAGE_KERNEL);
                vm_unmap_ram(vaddr, nr);
  }

I chose mersenne twister PRNG to generate persistent random state to
guarantee that both runs have the same random sequence.  For each
vm_map_ram call random number from [1..32/64] was taken to represent
amount of pages which I do map.

I did 10'000 vm_map_ram calls and got these two tables:

BEFORE (all new blocks are put to the head of a free list)

# cat /sys/kernel/debug/tracing/trace_stat/function0
  Function                               Hit    Time            Avg             s^2
  --------                               ---    ----            ---             ---
  vm_map_ram                           10000    10170.01 us     1.017 us        993.609 us
  vm_unmap_ram                         10000    5321.823 us     0.532 us        59.789 us
  alloc_vmap_area                        420    2150.239 us     5.119 us        3.307 us
  free_vmap_block                         37    159.587 us      4.313 us        134.344 us

AFTER (all new blocks are put to the tail of a free list)

# cat /sys/kernel/debug/tracing/trace_stat/function0
  Function                               Hit    Time            Avg             s^2
  --------                               ---    ----            ---             ---
  vm_map_ram                           10000    7745.637 us     0.774 us        395.229 us
  vm_unmap_ram                         10000    5460.573 us     0.546 us        67.187 us
  alloc_vmap_area                        414    2201.650 us     5.317 us        5.591 us
  free_vmap_block                        412    574.421 us      1.394 us        15.138 us

SUMMARY:

'BEFORE' table shows, that 420 blocks were allocated and only 37 were
freed.  Remained 383 blocks are still in a free list, consuming virtual
space and physical memory.

'AFTER' table shows, that 414 blocks were allocated and 412 were really
freed.  2 blocks remained in a free list.

So fragmentation was dramatically reduced.  Why? Because when we put
newly allocated block to the head, all further requests will occupy new
block, regardless remained space in other blocks.  In this scenario all
requests come randomly.  Eventually remained free space will be less
than requested size, free list will be iterated and it is possible that
nothing will be found there - finally new block will be created.  So
exhaustion in random scenario happens for the maximum possible
allocation size: 32 pages for 32-bit system and 64 pages for 64-bit
system.

Also average cost of vm_map_ram was reduced from 1.017 us to 0.774 us.
Again this can be explained by iteration through smaller list of free
blocks.

3) Next simple scenario is a sequential allocation, when the allocation
   order is increased for each block.  This scenario forces allocator to
   reach maximum amount of partially free blocks in a free list:

  while (iters) {

                /* Populate free list with blocks with remaining space */
                for (order = 0; order <= ilog2(VMAP_MAX_ALLOC); order++) {
                        nr = VMAP_BBMAP_BITS / (1 << order);

                        /* Leave a hole */
                        nr -= 1;

                        for (i = 0; i < nr; i++) {
                                vaddr = vm_map_ram(pages, (1 << order), -1, PAGE_KERNEL);
                                vm_unmap_ram(vaddr, (1 << order));
                }

                /* Completely occupy blocks from a free list */
                for (order = 0; order <= ilog2(VMAP_MAX_ALLOC); order++) {
                        vaddr = vm_map_ram(pages, (1 << order), -1, PAGE_KERNEL);
                        vm_unmap_ram(vaddr, (1 << order));
                }
  }

Results which I got:

BEFORE (all new blocks are put to the head of a free list)

# cat /sys/kernel/debug/tracing/trace_stat/function0
  Function                               Hit    Time            Avg             s^2
  --------                               ---    ----            ---             ---
  vm_map_ram                         2032000    399545.2 us     0.196 us        467123.7 us
  vm_unmap_ram                       2032000    363225.7 us     0.178 us        111405.9 us
  alloc_vmap_area                       7001    30627.76 us     4.374 us        495.755 us
  free_vmap_block                       6993    7011.685 us     1.002 us        159.090 us

AFTER (all new blocks are put to the tail of a free list)

# cat /sys/kernel/debug/tracing/trace_stat/function0
  Function                               Hit    Time            Avg             s^2
  --------                               ---    ----            ---             ---
  vm_map_ram                         2032000    394259.7 us     0.194 us        589395.9 us
  vm_unmap_ram                       2032000    292500.7 us     0.143 us        94181.08 us
  alloc_vmap_area                       7000    31103.11 us     4.443 us        703.225 us
  free_vmap_block                       7000    6750.844 us     0.964 us        119.112 us

SUMMARY:

No surprises here, almost all numbers are the same.

Fixing this fragmentation problem I also did some improvements in a
allocation logic of a new vmap block: occupy block immediately and get
rid of extra search in a free list.

Also I replaced dirty bitmap with min/max dirty range values to make the
logic simpler and slightly faster, since two longs comparison costs
less, than loop thru bitmap.

This patchset raises several questions:

 Q: Think the problem you comments is already known so that I wrote comments
    about it as "it could consume lots of address space through fragmentation".
    Could you tell me about your situation and reason why it should be avoided?
                                                                     Gioh Kim

 A: Indeed, there was a commit 364376383 which adds explicit comment about
    fragmentation.  But fragmentation which is described in this comment caused
    by mixing of long-lived and short-lived objects, when a whole block is pinned
    in memory because some page slots are still in use.  But here I am talking
    about blocks which are free, nobody uses them, and allocator keeps them alive
    forever, continuously allocating new blocks.

 Q: I think that if you put newly allocated block to the tail of a free
    list, below example would results in enormous performance degradation.

    new block: 1MB (256 pages)

    while (iters--) {
      vm_map_ram(3 or something else not dividable for 256) * 85
      vm_unmap_ram(3) * 85
    }

    On every iteration, it needs newly allocated block and it is put to the
    tail of a free list so finding it consumes large amount of time.
                                                                    Joonsoo Kim

 A: Second patch in current patchset gets rid of extra search in a free list,
    so new block will be immediately occupied..

    Also, the scenario above is impossible, cause vm_map_ram allocates virtual
    range in orders, i.e. 2^n.  I.e. passing 3 to vm_map_ram you will allocate
    4 slots in a block and 256 slots (capacity of a block) of course dividable
    on 4, so block will be completely occupied.

    But there is a worst case which we can achieve: each free block has a hole
    equal to order size.

    The maximum size of allocation is 64 pages for 64-bit system
    (if you try to map more, original alloc_vmap_area will be called).

    So the maximum order is 6.  That means that worst case, before allocator
    makes a decision to allocate a new block, is to iterate 7 blocks:

    HEAD
    1st block - has 1  page slot  free (order 0)
    2nd block - has 2  page slots free (order 1)
    3rd block - has 4  page slots free (order 2)
    4th block - has 8  page slots free (order 3)
    5th block - has 16 page slots free (order 4)
    6th block - has 32 page slots free (order 5)
    7th block - has 64 page slots free (order 6)
    TAIL

    So the worst scenario on 64-bit system is that each CPU queue can have 7
    blocks in a free list.

    This can happen only and only if you allocate blocks increasing the order.
    (as I did in the function written in the comment of the first patch)
    This is weird and rare case, but still it is possible.  Afterwards you will
    get 7 blocks in a list.

    All further requests should be placed in a newly allocated block or some
    free slots should be found in a free list.
    Seems it does not look dramatically awful.

This patch (of 3):

If suitable block can't be found, new block is allocated and put into a
head of a free list, so on next iteration this new block will be found
first.

That's bad, because old blocks in a free list will not get a chance to be
fully used, thus fragmentation will grow.

Let's consider this simple example:

 #1 We have one block in a free list which is partially used, and where only
    one page is free:

    HEAD |xxxxxxxxx-| TAIL
                   ^
                   free space for 1 page, order 0

 #2 New allocation request of order 1 (2 pages) comes, new block is allocated
    since we do not have free space to complete this request. New block is put
    into a head of a free list:

    HEAD |----------|xxxxxxxxx-| TAIL

 #3 Two pages were occupied in a new found block:

    HEAD |xx--------|xxxxxxxxx-| TAIL
          ^
          two pages mapped here

 #4 New allocation request of order 0 (1 page) comes.  Block, which was created
    on #2 step, is located at the beginning of a free list, so it will be found
    first:

  HEAD |xxX-------|xxxxxxxxx-| TAIL
          ^                 ^
          page mapped here, but better to use this hole

It is obvious, that it is better to complete request of #4 step using the
old block, where free space is left, because in other case fragmentation
will be highly increased.

But fragmentation is not only the case.  The worst thing is that I can
easily create scenario, when the whole vmalloc space is exhausted by
blocks, which are not used, but already dirty and have several free pages.

Let's consider this function which execution should be pinned to one CPU:

static void exhaust_virtual_space(struct page *pages[16], int iters)
{
        /* Firstly we have to map a big chunk, e.g. 16 pages.
         * Then we have to occupy the remaining space with smaller
         * chunks, i.e. 8 pages. At the end small hole should remain.
         * So at the end of our allocation sequence block looks like
         * this:
         *                XX  big chunk
         * |XXxxxxxxx-|    x  small chunk
         *                 -  hole, which is enough for a small chunk,
         *                    but is not enough for a big chunk
         */
        while (iters--) {
                int i;
                void *vaddr;

                /* Map/unmap big chunk */
                vaddr = vm_map_ram(pages, 16, -1, PAGE_KERNEL);
                vm_unmap_ram(vaddr, 16);

                /* Map/unmap small chunks.
                 *
                 * -1 for hole, which should be left at the end of each block
                 * to keep it partially used, with some free space available */
                for (i = 0; i < (VMAP_BBMAP_BITS - 16) / 8 - 1; i++) {
                        vaddr = vm_map_ram(pages, 8, -1, PAGE_KERNEL);
                        vm_unmap_ram(vaddr, 8);
                }
        }
}

On every iteration new block (1MB of vm area in my case) will be
allocated and then will be occupied, without attempt to resolve small
allocation request using previously allocated blocks in a free list.

In case of random allocation (size should be randomly taken from the
range [1..64] in 64-bit case or [1..32] in 32-bit case) situation is the
same: new blocks continue to appear if maximum possible allocation size
(32 or 64) passed to the allocator, because all remaining blocks in a
free list do not have enough free space to complete this allocation
request.

In summary if new blocks are put into the head of a free list eventually
virtual space will be exhausted.

In current patch I simply put newly allocated block to the tail of a
free list, thus reduce fragmentation, giving a chance to resolve
allocation request using older blocks with possible holes left.

Signed-off-by: Roman Pen <r.peniaev@gmail.com>
Cc: Eric Dumazet <edumazet@google.com>
Acked-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: David Rientjes <rientjes@google.com>
Cc: WANG Chao <chaowang@redhat.com>
Cc: Fabian Frederick <fabf@skynet.be>
Cc: Christoph Lameter <cl@linux.com>
Cc: Gioh Kim <gioh.kim@lge.com>
Cc: Rob Jones <rob.jones@codethink.co.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agohugetlbfs: document min_size mount option and cleanup
Mike Kravetz [Wed, 15 Apr 2015 23:13:45 +0000 (16:13 -0700)]
hugetlbfs: document min_size mount option and cleanup

Add min_size mount option to the hugetlbfs documentation.  Also, add the
missing pagesize option and mention that size can be specified as bytes or
a percentage of huge page pool.

Signed-off-by: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Davidlohr Bueso <dave@stgolabs.net>
Cc: Aneesh Kumar <aneesh.kumar@linux.vnet.ibm.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agohugetlbfs: accept subpool min_size mount option and setup accordingly
Mike Kravetz [Wed, 15 Apr 2015 23:13:42 +0000 (16:13 -0700)]
hugetlbfs: accept subpool min_size mount option and setup accordingly

Make 'min_size=<value>' be an option when mounting a hugetlbfs.  This
option takes the same value as the 'size' option.  min_size can be
specified without specifying size.  If both are specified, min_size must
be less that or equal to size else the mount will fail.  If min_size is
specified, then at mount time an attempt is made to reserve min_size
pages.  If the reservation fails, the mount fails.  At umount time, the
reserved pages are released.

Signed-off-by: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Davidlohr Bueso <dave@stgolabs.net>
Cc: Aneesh Kumar <aneesh.kumar@linux.vnet.ibm.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agohugetlbfs: add minimum size accounting to subpools
Mike Kravetz [Wed, 15 Apr 2015 23:13:39 +0000 (16:13 -0700)]
hugetlbfs: add minimum size accounting to subpools

The same routines that perform subpool maximum size accounting
hugepage_subpool_get/put_pages() are modified to also perform minimum size
accounting.  When a delta value is passed to these routines, calculate how
global reservations must be adjusted to maintain the subpool minimum size.
 The routines now return this global reserve count adjustment.  This
global reserve count adjustment is then passed to the global accounting
routine hugetlb_acct_memory().

Signed-off-by: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Davidlohr Bueso <dave@stgolabs.net>
Cc: Aneesh Kumar <aneesh.kumar@linux.vnet.ibm.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agohugetlbfs: add minimum size tracking fields to subpool structure
Mike Kravetz [Wed, 15 Apr 2015 23:13:36 +0000 (16:13 -0700)]
hugetlbfs: add minimum size tracking fields to subpool structure

hugetlbfs allocates huge pages from the global pool as needed.  Even if
the global pool contains a sufficient number pages for the filesystem size
at mount time, those global pages could be grabbed for some other use.  As
a result, filesystem huge page allocations may fail due to lack of pages.

Applications such as a database want to use huge pages for performance
reasons.  hugetlbfs filesystem semantics with ownership and modes work
well to manage access to a pool of huge pages.  However, the application
would like some reasonable assurance that allocations will not fail due to
a lack of huge pages.  At application startup time, the application would
like to configure itself to use a specific number of huge pages.  Before
starting, the application can check to make sure that enough huge pages
exist in the system global pools.  However, there are no guarantees that
those pages will be available when needed by the application.  What the
application wants is exclusive use of a subset of huge pages.

Add a new hugetlbfs mount option 'min_size=<value>' to indicate that the
specified number of pages will be available for use by the filesystem.  At
mount time, this number of huge pages will be reserved for exclusive use
of the filesystem.  If there is not a sufficient number of free pages, the
mount will fail.  As pages are allocated to and freeed from the
filesystem, the number of reserved pages is adjusted so that the specified
minimum is maintained.

This patch (of 4):

Add a field to the subpool structure to indicate the minimimum number of
huge pages to always be used by this subpool.  This minimum count includes
allocated pages as well as reserved pages.  If the minimum number of pages
for the subpool have not been allocated, pages are reserved up to this
minimum.  An additional field (rsv_hpages) is used to track the number of
pages reserved to meet this minimum size.  The hstate pointer in the
subpool is convenient to have when reserving and unreserving the pages.

Signed-off-by: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Davidlohr Bueso <dave@stgolabs.net>
Cc: Aneesh Kumar <aneesh.kumar@linux.vnet.ibm.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm/compaction: reset compaction scanner positions
Gioh Kim [Wed, 15 Apr 2015 23:13:33 +0000 (16:13 -0700)]
mm/compaction: reset compaction scanner positions

When the compaction is activated via /proc/sys/vm/compact_memory it would
better scan the whole zone.  And some platforms, for instance ARM, have
the start_pfn of a zone at zero.  Therefore the first try to compact via
/proc doesn't work.  It needs to reset the compaction scanner position
first.

Signed-off-by: Gioh Kim <gioh.kim@lge.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: David Rientjes <rientjes@google.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm, memcg: sync allocation and memcg charge gfp flags for THP
Michal Hocko [Wed, 15 Apr 2015 23:13:29 +0000 (16:13 -0700)]
mm, memcg: sync allocation and memcg charge gfp flags for THP

memcg currently uses hardcoded GFP_TRANSHUGE gfp flags for all THP
charges.  THP allocations, however, might be using different flags
depending on /sys/kernel/mm/transparent_hugepage/{,khugepaged/}defrag and
the current allocation context.

The primary difference is that defrag configured to "madvise" value will
clear __GFP_WAIT flag from the core gfp mask to make the allocation
lighter for all mappings which are not backed by VM_HUGEPAGE vmas.  If
memcg charge path ignores this fact we will get light allocation but the a
potential memcg reclaim would kill the whole point of the configuration.

Fix the mismatch by providing the same gfp mask used for the allocation to
the charge functions.  This is quite easy for all paths except for
hugepaged kernel thread with !CONFIG_NUMA which is doing a pre-allocation
long before the allocated page is used in collapse_huge_page via
khugepaged_alloc_page.  To prevent from cluttering the whole code path
from khugepaged_do_scan we simply return the current flags as per
khugepaged_defrag() value which might have changed since the
preallocation.  If somebody changed the value of the knob we would charge
differently but this shouldn't happen often and it is definitely not
critical because it would only lead to a reduced success rate of one-off
THP promotion.

[akpm@linux-foundation.org: fix weird code layout while we're there]
[rientjes@google.com: clean up around alloc_hugepage_gfpmask()]
Signed-off-by: Michal Hocko <mhocko@suse.cz>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: David Rientjes <rientjes@google.com>
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>
9 years agomm: rename deactivate_page to deactivate_file_page
Minchan Kim [Wed, 15 Apr 2015 23:13:26 +0000 (16:13 -0700)]
mm: rename deactivate_page to deactivate_file_page

"deactivate_page" was created for file invalidation so it has too
specific logic for file-backed pages.  So, let's change the name of the
function and date to a file-specific one and yield the generic name.

Signed-off-by: Minchan Kim <minchan@kernel.org>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Rik van Riel <riel@redhat.com>
Cc: Shaohua Li <shli@kernel.org>
Cc: Wang, Yalin <Yalin.Wang@sonymobile.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agoDocumentation/vm/unevictable-lru.txt: document interaction between compaction and...
Eric B Munson [Wed, 15 Apr 2015 23:13:23 +0000 (16:13 -0700)]
Documentation/vm/unevictable-lru.txt: document interaction between compaction and the unevictable LRU

The memory compaction code uses the migration code to do most of the
work in compaction.  However, the compaction code interacts with the
unevictable LRU differently than migration code and this difference
should be noted in the documentation.

[akpm@linux-foundation.org: identify /proc/sys/vm/compact_unevictable directly]
Signed-off-by: Eric B Munson <emunson@akamai.com>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Christoph Lameter <cl@linux.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Mel Gorman <mgorman@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm: allow compaction of unevictable pages
Eric B Munson [Wed, 15 Apr 2015 23:13:20 +0000 (16:13 -0700)]
mm: allow compaction of unevictable pages

Currently, pages which are marked as unevictable are protected from
compaction, but not from other types of migration.  The POSIX real time
extension explicitly states that mlock() will prevent a major page
fault, but the spirit of this is that mlock() should give a process the
ability to control sources of latency, including minor page faults.
However, the mlock manpage only explicitly says that a locked page will
not be written to swap and this can cause some confusion.  The
compaction code today does not give a developer who wants to avoid swap
but wants to have large contiguous areas available any method to achieve
this state.  This patch introduces a sysctl for controlling compaction
behavior with respect to the unevictable lru.  Users who demand no page
faults after a page is present can set compact_unevictable_allowed to 0
and users who need the large contiguous areas can enable compaction on
locked memory by leaving the default value of 1.

To illustrate this problem I wrote a quick test program that mmaps a
large number of 1MB files filled with random data.  These maps are
created locked and read only.  Then every other mmap is unmapped and I
attempt to allocate huge pages to the static huge page pool.  When the
compact_unevictable_allowed sysctl is 0, I cannot allocate hugepages
after fragmenting memory.  When the value is set to 1, allocations
succeed.

Signed-off-by: Eric B Munson <emunson@akamai.com>
Acked-by: Michal Hocko <mhocko@suse.cz>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Christoph Lameter <cl@linux.com>
Acked-by: David Rientjes <rientjes@google.com>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Christoph Lameter <cl@linux.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Mel Gorman <mgorman@suse.de>
Cc: David Rientjes <rientjes@google.com>
Cc: Michal Hocko <mhocko@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm/page-writeback: check-before-clear PageReclaim
Naoya Horiguchi [Wed, 15 Apr 2015 23:13:17 +0000 (16:13 -0700)]
mm/page-writeback: check-before-clear PageReclaim

With the page flag sanitization patchset, an invalid usage of
ClearPageReclaim() is detected in set_page_dirty().  This can be called
from __unmap_hugepage_range(), so let's check PageReclaim() before trying
to clear it to avoid the misuse.

Signed-off-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm/migrate: check-before-clear PageSwapCache
Naoya Horiguchi [Wed, 15 Apr 2015 23:13:15 +0000 (16:13 -0700)]
mm/migrate: check-before-clear PageSwapCache

With the page flag sanitization patchset, an invalid usage of
ClearPageSwapCache() is detected in migration_page_copy().
migrate_page_copy() is shared by both normal and hugepage (both thp and
hugetlb) code path, so let's check PageSwapCache() and clear it if it's
set to avoid misuse of the invalid clear operation.

Signed-off-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm: avoid tail page refcounting on non-THP compound pages
Kirill A. Shutemov [Wed, 15 Apr 2015 23:13:12 +0000 (16:13 -0700)]
mm: avoid tail page refcounting on non-THP compound pages

THP uses tail page refcounting to be able to split huge pages at any time.
 Tail page refcounting is not needed for other users of compound pages and
it's harmful because of overhead.

We try to exclude non-THP pages from tail page refcounting using
__compound_tail_refcounted() check.  It excludes most common non-THP
compound pages: SL*B and hugetlb, but it doesn't catch rest of __GFP_COMP
users -- drivers.

And it's not only about overhead.

Drivers might want to use compound pages to get refcounting semantics
suitable for mapping high-order pages to userspace.  But tail page
refcounting breaks it.

Tail page refcounting uses ->_mapcount in tail pages to store GUP pins on
them.  It means GUP pins would affect page_mapcount() for tail pages.
It's not a problem for THP, because it never maps tail pages.  But unlike
THP, drivers map parts of compound pages with PTEs and it makes
page_mapcount() be called for tail pages.

In particular, GUP pins would shift PSS up and affect /proc/kpagecount for
such pages.  But, I'm not aware about anything which can lead to crash or
other serious misbehaviour.

Since currently all THP pages are anonymous and all drivers pages are not,
we can fix the __compound_tail_refcounted() check by requiring PageAnon()
to enable tail page refcounting.

Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Acked-by: Hugh Dickins <hughd@google.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm: consolidate all page-flags helpers in <linux/page-flags.h>
Kirill A. Shutemov [Wed, 15 Apr 2015 23:13:08 +0000 (16:13 -0700)]
mm: consolidate all page-flags helpers in <linux/page-flags.h>

Currently we take a naive approach to page flags on compound pages - we
set the flag on the page without consideration if the flag makes sense
for tail page or for compound page in general.  This patchset try to
sort this out by defining per-flag policy on what need to be done if
page-flag helper operate on compound page.

The last patch in the patchset also sanitizes usege of page->mapping for
tail pages.  We don't define the meaning of page->mapping for tail
pages.  Currently it's always NULL, which can be inconsistent with head
page and potentially lead to problems.

For now I caught one case of illegal usage of page flags or ->mapping:
sound subsystem allocates pages with __GFP_COMP and maps them with PTEs.
It leads to setting dirty bit on tail pages and access to tail_page's
->mapping.  I don't see any bad behaviour caused by this, but worth
fixing anyway.

This patchset makes more sense if you take my THP refcounting into
account: we will see more compound pages mapped with PTEs and we need to
define behaviour of flags on compound pages to avoid bugs.

This patch (of 16):

We have page-flags helper function declarations/definitions spread over
several header files.  Let's consolidate them in <linux/page-flags.h>.

Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Acked-by: Hugh Dickins <hughd@google.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Rik van Riel <riel@redhat.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Christoph Lameter <cl@linux.com>
Cc: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Cc: Steve Capper <steve.capper@linaro.org>
Cc: "Aneesh Kumar K.V" <aneesh.kumar@linux.vnet.ibm.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Jerome Marchand <jmarchan@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm/memory-failure.c: define page types for action_result() in one place
Naoya Horiguchi [Wed, 15 Apr 2015 23:13:05 +0000 (16:13 -0700)]
mm/memory-failure.c: define page types for action_result() in one place

This cleanup patch moves all strings passed to action_result() into a
singl= e array action_page_type so that a reader can easily find which
kind of actio= n results are possible.  And this patch also fixes the
odd lines to be printed out, like "unknown page state page" or "free
buddy, 2nd try page".

[akpm@linux-foundation.org: rename messages, per David]
[akpm@linux-foundation.org: s/DIRTY_UNEVICTABLE_LRU/CLEAN_UNEVICTABLE_LRU', per Andi]
Signed-off-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Reviewed-by: Andi Kleen <ak@linux.intel.com>
Cc: Tony Luck <tony.luck@intel.com>
Cc: "Xie XiuQi" <xiexiuqi@huawei.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Chen Gong <gong.chen@linux.intel.com>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomemcg: remove obsolete comment
Vladimir Davydov [Wed, 15 Apr 2015 23:13:03 +0000 (16:13 -0700)]
memcg: remove obsolete comment

Low and high watermarks, as they defined in the TODO to the mem_cgroup
struct, have already been implemented by Johannes, so remove the stale
comment.

Signed-off-by: Vladimir Davydov <vdavydov@parallels.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Michal Hocko <mhocko@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomemcg: zap mem_cgroup_lookup()
Vladimir Davydov [Wed, 15 Apr 2015 23:13:00 +0000 (16:13 -0700)]
memcg: zap mem_cgroup_lookup()

mem_cgroup_lookup() is a wrapper around mem_cgroup_from_id(), which
checks that id != 0 before issuing the function call.  Today, there is
no point in this additional check apart from optimization, because there
is no css with id <= 0, so that css_from_id, called by
mem_cgroup_from_id, will return NULL for any id <= 0.

Since mem_cgroup_from_id is only called from mem_cgroup_lookup, let us
zap mem_cgroup_lookup, substituting calls to it with mem_cgroup_from_id
and moving the check if id > 0 to css_from_id.

Signed-off-by: Vladimir Davydov <vdavydov@parallels.com>
Acked-by: Michal Hocko <mhocko@suse.cz>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm: refactor zone_movable_is_highmem()
Zhang Zhen [Wed, 15 Apr 2015 23:12:57 +0000 (16:12 -0700)]
mm: refactor zone_movable_is_highmem()

All callers of zone_movable_is_highmem are under #ifdef CONFIG_HIGHMEM,
so the else branch return 0 is not needed.

Signed-off-by: Zhang Zhen <zhenzhang.zhang@huawei.com>
Acked-by: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm/oom_kill.c: fix typo in comment
Yaowei Bai [Wed, 15 Apr 2015 23:12:54 +0000 (16:12 -0700)]
mm/oom_kill.c: fix typo in comment

Alter 'taks' -> 'task'

Signed-off-by: Yaowei Bai <bywxiaobai@163.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agovfs: delete vfs_readdir function declaration
Zhang Zhen [Wed, 15 Apr 2015 23:12:51 +0000 (16:12 -0700)]
vfs: delete vfs_readdir function declaration

vfs_readdir() was replaced by iterate_dir() in commit 5c0ba4e0762e
("[readdir] introduce iterate_dir() and dir_context").

Signed-off-by: Zhang Zhen <zhenzhang.zhang@huawei.com>
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>
9 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-next
Linus Torvalds [Wed, 15 Apr 2015 16:00:47 +0000 (09:00 -0700)]
Merge git://git./linux/kernel/git/davem/net-next

Pull networking updates from David Miller:

 1) Add BQL support to via-rhine, from Tino Reichardt.

 2) Integrate SWITCHDEV layer support into the DSA layer, so DSA drivers
    can support hw switch offloading.  From Floria Fainelli.

 3) Allow 'ip address' commands to initiate multicast group join/leave,
    from Madhu Challa.

 4) Many ipv4 FIB lookup optimizations from Alexander Duyck.

 5) Support EBPF in cls_bpf classifier and act_bpf action, from Daniel
    Borkmann.

 6) Remove the ugly compat support in ARP for ugly layers like ax25,
    rose, etc.  And use this to clean up the neigh layer, then use it to
    implement MPLS support.  All from Eric Biederman.

 7) Support L3 forwarding offloading in switches, from Scott Feldman.

 8) Collapse the LOCAL and MAIN ipv4 FIB tables when possible, to speed
    up route lookups even further.  From Alexander Duyck.

 9) Many improvements and bug fixes to the rhashtable implementation,
    from Herbert Xu and Thomas Graf.  In particular, in the case where
    an rhashtable user bulk adds a large number of items into an empty
    table, we expand the table much more sanely.

10) Don't make the tcp_metrics hash table per-namespace, from Eric
    Biederman.

11) Extend EBPF to access SKB fields, from Alexei Starovoitov.

12) Split out new connection request sockets so that they can be
    established in the main hash table.  Much less false sharing since
    hash lookups go direct to the request sockets instead of having to
    go first to the listener then to the request socks hashed
    underneath.  From Eric Dumazet.

13) Add async I/O support for crytpo AF_ALG sockets, from Tadeusz Struk.

14) Support stable privacy address generation for RFC7217 in IPV6.  From
    Hannes Frederic Sowa.

15) Hash network namespace into IP frag IDs, also from Hannes Frederic
    Sowa.

16) Convert PTP get/set methods to use 64-bit time, from Richard
    Cochran.

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-next: (1816 commits)
  fm10k: Bump driver version to 0.15.2
  fm10k: corrected VF multicast update
  fm10k: mbx_update_max_size does not drop all oversized messages
  fm10k: reset head instead of calling update_max_size
  fm10k: renamed mbx_tx_dropped to mbx_tx_oversized
  fm10k: update xcast mode before synchronizing multicast addresses
  fm10k: start service timer on probe
  fm10k: fix function header comment
  fm10k: comment next_vf_mbx flow
  fm10k: don't handle mailbox events in iov_event path and always process mailbox
  fm10k: use separate workqueue for fm10k driver
  fm10k: Set PF queues to unlimited bandwidth during virtualization
  fm10k: expose tx_timeout_count as an ethtool stat
  fm10k: only increment tx_timeout_count in Tx hang path
  fm10k: remove extraneous "Reset interface" message
  fm10k: separate PF only stats so that VF does not display them
  fm10k: use hw->mac.max_queues for stats
  fm10k: only show actual queues, not the maximum in hardware
  fm10k: allow creation of VLAN on default vid
  fm10k: fix unused warnings
  ...

9 years agoMerge branch 'for-linus' of git://ftp.arm.linux.org.uk/~rmk/linux-arm
Linus Torvalds [Wed, 15 Apr 2015 04:03:26 +0000 (21:03 -0700)]
Merge branch 'for-linus' of git://ftp.arm.linux.org.uk/~rmk/linux-arm

Pull ARM updates from Russell King:
 "Included in this update are both some long term fixes and some new
  features.

  Fixes:

   - An integer overflow in the calculation of ELF_ET_DYN_BASE.

   - Avoiding OOMs for high-order IOMMU allocations

   - SMP requires the data cache to be enabled for synchronisation
     primitives to work, so prevent the CPU_DCACHE_DISABLE option being
     visible on SMP builds.

   - A bug going back 10+ years in the noMMU ARM94* CPU support code,
     where it corrupts registers.  Found by folk getting Linux running
     on their cameras.

   - Versatile Express needs an errata workaround enabled for CPU
     hot-unplug to work.

  Features:

   - Clean up module linker by handling out of range relocations
     separately from relocation cases we don't handle.

   - Fix a long term bug in the pci_mmap_page_range() code, which we
     hope won't impact userspace (we hope there's no users of the
     existing broken interface.)

   - Don't map DMA coherent allocations when we don't have a MMU.

   - Drop experimental status for SMP_ON_UP.

   - Warn when DT doesn't specify ePAPR mandatory cache properties.

   - Add documentation concerning how we find the start of physical
     memory for AUTO_ZRELADDR kernels, detailing why we have chosen the
     mask and the implications of changing it.

   - Updates from Ard Biesheuvel to address some issues with large
     kernels (such as allyesconfig) failing to link.

   - Allow hibernation to work on modern (ARMv7) CPUs - this appears to
     have never worked in the past on these CPUs.

   - Enable IRQ_SHOW_LEVEL, which changes the /proc/interrupts output
     format (hopefully without userspace breaking...  let's hope that if
     it causes someone a problem, they tell us.)

   - Fix tegra-ahb DT offsets.

   - Rework ARM errata 643719 code (and ARMv7 flush_cache_louis()/
     flush_dcache_all()) code to be more efficient, and enable this
     errata workaround by default for ARMv7+SMP CPUs.  This complements
     the Versatile Express fix above.

   - Rework ARMv7 context code for errata 430973, so that only Cortex A8
     CPUs are impacted by the branch target buffer flush when this
     errata is enabled.  Also update the help text to indicate that all
     r1p* A8 CPUs are impacted.

   - Switch ARM to the generic show_mem() implementation, it conveys all
     the information which we were already reporting.

   - Prevent slow timer sources being used for udelay() - timers running
     at less than 1MHz are not useful for this, and can cause udelay()
     to return immediately, without any wait.  Using such a slow timer
     is silly.

   - VDSO support for 32-bit ARM, mainly for gettimeofday() using the
     ARM architected timer.

   - Perf support for Scorpion performance monitoring units"

vdso semantic conflict fixed up as per linux-next.

* 'for-linus' of git://ftp.arm.linux.org.uk/~rmk/linux-arm: (52 commits)
  ARM: update errata 430973 documentation to cover Cortex A8 r1p*
  ARM: ensure delay timer has sufficient accuracy for delays
  ARM: switch to use the generic show_mem() implementation
  ARM: proc-v7: avoid errata 430973 workaround for non-Cortex A8 CPUs
  ARM: enable ARM errata 643719 workaround by default
  ARM: cache-v7: optimise test for Cortex A9 r0pX devices
  ARM: cache-v7: optimise branches in v7_flush_cache_louis
  ARM: cache-v7: consolidate initialisation of cache level index
  ARM: cache-v7: shift CLIDR to extract appropriate field before masking
  ARM: cache-v7: use movw/movt instructions
  ARM: allow 16-bit instructions in ALT_UP()
  ARM: proc-arm94*.S: fix setup function
  ARM: vexpress: fix CPU hotplug with CT9x4 tile.
  ARM: 8276/1: Make CPU_DCACHE_DISABLE depend on !SMP
  ARM: 8335/1: Documentation: DT bindings: Tegra AHB: document the legacy base address
  ARM: 8334/1: amba: tegra-ahb: detect and correct bogus base address
  ARM: 8333/1: amba: tegra-ahb: fix register offsets in the macros
  ARM: 8339/1: Enable CONFIG_GENERIC_IRQ_SHOW_LEVEL
  ARM: 8338/1: kexec: Relax SMP validation to improve DT compatibility
  ARM: 8337/1: mm: Do not invoke OOM for higher order IOMMU DMA allocations
  ...

9 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/s390/linux
Linus Torvalds [Wed, 15 Apr 2015 03:51:44 +0000 (20:51 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/s390/linux

Pull s390 updates from Martin Schwidefsky:
 "The major change in this merge is the removal of the support for
  31-bit kernels.  Naturally 31-bit user space will continue to work via
  the compat layer.

  And then some cleanup, some improvements and bug fixes"

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/s390/linux: (23 commits)
  s390/smp: wait until secondaries are active & online
  s390/hibernate: fix save and restore of kernel text section
  s390/cacheinfo: add missing facility check
  s390/syscalls: simplify syscall_get_arch()
  s390/irq: enforce correct irqclass_sub_desc array size
  s390: remove "64" suffix from mem64.S and swsusp_asm64.S
  s390/ipl: cleanup macro usage
  s390/ipl: cleanup shutdown_action attributes
  s390/ipl: cleanup bin attr usage
  s390/uprobes: fix address space annotation
  s390: add missing arch_release_task_struct() declaration
  s390: make couple of functions and variables static
  s390/maccess: improve s390_kernel_write()
  s390/maccess: remove potentially broken probe_kernel_write()
  s390/watchdog: support for KVM hypervisors and delete pr_info messages
  s390/watchdog: enable KEEPALIVE for /dev/watchdog
  s390/dasd: remove setting of scheduler from driver
  s390/traps: panic() instead of die() on translation exception
  s390: remove test_facility(2) (== z/Architecture mode active) checks
  s390/cmpxchg: simplify cmpxchg_double
  ...

9 years agoMerge tag 'pm+acpi-4.1-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael...
Linus Torvalds [Wed, 15 Apr 2015 03:21:54 +0000 (20:21 -0700)]
Merge tag 'pm+acpi-4.1-rc1' of git://git./linux/kernel/git/rafael/linux-pm

Pull power management and ACPI updates from Rafael Wysocki:
 "These are mostly fixes and cleanups all over, although there are a few
  items that sort of fall into the new feature category.

  First off, we have new callbacks for PM domains that should help us to
  handle some issues related to device initialization in a better way.

  There also is some consolidation in the unified device properties API
  area allowing us to use that inferface for accessing data coming from
  platform initialization code in addition to firmware-provided data.

  We have some new device/CPU IDs in a few drivers, support for new
  chips and a new cpufreq driver too.

  Specifics:

   - Generic PM domains support update including new PM domain callbacks
     to handle device initialization better (Russell King, Rafael J
     Wysocki, Kevin Hilman)

   - Unified device properties API update including a new mechanism for
     accessing data provided by platform initialization code (Rafael J
     Wysocki, Adrian Hunter)

   - ARM cpuidle update including ARM32/ARM64 handling consolidation
     (Daniel Lezcano)

   - intel_idle update including support for the Silvermont Core in the
     Baytrail SOC and for the Airmont Core in the Cherrytrail and
     Braswell SOCs (Len Brown, Mathias Krause)

   - New cpufreq driver for Hisilicon ACPU (Leo Yan)

   - intel_pstate update including support for the Knights Landing chip
     (Dasaratharaman Chandramouli, Kristen Carlson Accardi)

   - QorIQ cpufreq driver update (Tang Yuantian, Arnd Bergmann)

   - powernv cpufreq driver update (Shilpasri G Bhat)

   - devfreq update including Tegra support changes (Tomeu Vizoso,
     MyungJoo Ham, Chanwoo Choi)

   - powercap RAPL (Running-Average Power Limit) driver update including
     support for Intel Broadwell server chips (Jacob Pan, Mathias Krause)

   - ACPI device enumeration update related to the handling of the
     special PRP0001 device ID allowing DT-style 'compatible' property
     to be used for ACPI device identification (Rafael J Wysocki)

   - ACPI EC driver update including limited _DEP support (Lan Tianyu,
     Lv Zheng)

   - ACPI backlight driver update including a new mechanism to allow
     native backlight handling to be forced on non-Windows 8 systems and
     a new quirk for Lenovo Ideapad Z570 (Aaron Lu, Hans de Goede)

   - New Windows Vista compatibility quirk for Sony VGN-SR19XN (Chen Yu)

   - Assorted ACPI fixes and cleanups (Aaron Lu, Martin Kepplinger,
     Masanari Iida, Mika Westerberg, Nan Li, Rafael J Wysocki)

   - Fixes related to suspend-to-idle for the iTCO watchdog driver and
     the ACPI core system suspend/resume code (Rafael J Wysocki, Chen Yu)

   - PM tracing support for the suspend phase of system suspend/resume
     transitions (Zhonghui Fu)

   - Configurable delay for the system suspend/resume testing facility
     (Brian Norris)

   - PNP subsystem cleanups (Peter Huewe, Rafael J Wysocki)"

* tag 'pm+acpi-4.1-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm: (74 commits)
  ACPI / scan: Fix NULL pointer dereference in acpi_companion_match()
  ACPI / scan: Rework modalias creation when "compatible" is present
  intel_idle: mark cpu id array as __initconst
  powercap / RAPL: mark rapl_ids array as __initconst
  powercap / RAPL: add ID for Broadwell server
  intel_pstate: Knights Landing support
  intel_pstate: remove MSR test
  cpufreq: fix qoriq uniprocessor build
  ACPI / scan: Take the PRP0001 position in the list of IDs into account
  ACPI / scan: Simplify acpi_match_device()
  ACPI / scan: Generalize of_compatible matching
  device property: Introduce firmware node type for platform data
  device property: Make it possible to use secondary firmware nodes
  PM / watchdog: iTCO: stop watchdog during system suspend
  cpufreq: hisilicon: add acpu driver
  ACPI / EC: Call acpi_walk_dep_device_list() after installing EC opregion handler
  cpufreq: powernv: Report cpu frequency throttling
  intel_idle: Add support for the Airmont Core in the Cherrytrail and Braswell SOCs
  intel_idle: Update support for Silvermont Core in Baytrail SOC
  PM / devfreq: tegra: Register governor on module init
  ...

9 years agoMerge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/next...
David S. Miller [Wed, 15 Apr 2015 02:29:27 +0000 (22:29 -0400)]
Merge branch 'master' of git://git./linux/kernel/git/jkirsher/next-queue

Jeff Kirsher says:

====================
Intel Wired LAN Driver Updates 2015-04-14

This series contains updates to fm10k only.

Fixed transmit statistics which was actually using values from the
receive ring, instead of the transmit ring.  Fixed up spelling mistakes
in code comments and resolved unused argument warnings.  Added support
for netconsole.  Fixed up statistic reporting so that we are only
reporting from actual queues as well as display PF only stats for
just the PF and not the VF.  Also fixed an issue that when returning
virtualization queues from the VF back to the PF, we were retaining
the VF rate limiter.

Fixed up the driver to use a separate workqueue, which helps reduce
and stabilize latency between scheduling the work in our interrupt and
actually performing the work.

Fixed a bug where the VF tried to set a multicast address before
requesting the required xcast mode.

Fix VF multicast update since VFs were being improperly added to the
switch's mutlicast group.  The error stems from the fact that incorrect
arguments were passed to the update_mc_addr().

Thanks to Alex Duyck for the extensive review.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
9 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input
Linus Torvalds [Wed, 15 Apr 2015 01:25:15 +0000 (18:25 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/dtor/input

Pull input subsystem updates from Dmitry Torokhov:
 "You will get the following new drivers:

   - Qualcomm PM8941 power key drver
   - ChipOne icn8318 touchscreen controller driver
   - Broadcom iProc touchscreen and keypad drivers
   - Semtech SX8654 I2C touchscreen controller driver

  ALPS driver now supports newer SS4 devices; Elantech got a fix that
  should make it work on some ASUS laptops; and a slew of other
  enhancements and random fixes"

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input: (51 commits)
  Input: alps - non interleaved V2 dualpoint has separate stick button bits
  Input: alps - fix touchpad buttons getting stuck when used with trackpoint
  Input: atkbd - document "no new force-release quirks" policy
  Input: ALPS - make alps_get_pkt_id_ss4_v2() and others static
  Input: ALPS - V7 devices can report 5-finger taps
  Input: ALPS - add support for SS4 touchpad devices
  Input: ALPS - refactor alps_set_abs_params_mt()
  Input: elantech - fix absolute mode setting on some ASUS laptops
  Input: atmel_mxt_ts - split out touchpad initialisation logic
  Input: atmel_mxt_ts - implement support for T100 touch object
  Input: cros_ec_keyb - fix clearing keyboard state on wakeup
  Input: gscps2 - drop pci_ids dependency
  Input: synaptics - allocate 3 slots to keep stability in image sensors
  Input: Revert "Revert "synaptics - use dmax in input_mt_assign_slots""
  Input: MT - make slot assignment work for overcovered solutions
  mfd: tc3589x: enforce device-tree only mode
  Input: tc3589x - localize platform data
  Input: tsc2007 - Convert msecs to jiffies only once
  Input: edt-ft5x06 - remove EV_SYN event report
  Input: edt-ft5x06 - allow to setting the maximum axes value through the DT
  ...