printk: enable interrupts before calling console_trylock_for_printk()

We need interrupts disabled when calling console_trylock_for_printk()
only so that cpu id we pass to can_use_console() remains valid (for
other things console_sem provides all the exclusion we need and
deadlocks on console_sem due to interrupts are impossible because we use
down_trylock()).  However if we are rescheduled, we are guaranteed to
run on an online cpu so we can easily just get the cpu id in
can_use_console().

We can lose a bit of performance when we enable interrupts in
vprintk_emit() and then disable them again in console_unlock() but OTOH
it can somewhat reduce interrupt latency caused by console_unlock().

We differ from (reverted) commit 939f04bec1a4 in that we avoid calling
console_unlock() from vprintk_emit() with lockdep enabled as that has
unveiled quite some bugs leading to system freezes during boot (e.g.
  https://lkml.org/lkml/2014/5/30/242,
  https://lkml.org/lkml/2014/6/28/521).

Signed-off-by: Jan Kara <jack@suse.cz>
Tested-by: Andreas Bombe <aeb@debian.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

printk: miscellaneous cleanups

Some small cleanups to kernel/printk/printk.c.  None of them should
cause any change in behavior.

  - When CONFIG_PRINTK is defined, parenthesize the value of LOG_LINE_MAX.
  - When CONFIG_PRINTK is *not* defined, there is an extra LOG_LINE_MAX
    definition; delete it.
  - Pull an assignment out of a conditional expression in console_setup().
  - Use isdigit() in console_setup() rather than open coding it.
  - In update_console_cmdline(), drop a NUL-termination assignment;
    the strlcpy() call that precedes it guarantees it's not needed.
  - Simplify some logic in printk_timed_ratelimit().

Signed-off-by: Alex Elder <elder@linaro.org>
Reviewed-by: Petr Mladek <pmladek@suse.cz>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Borislav Petkov <bp@suse.de>
Cc: Jan Kara <jack@suse.cz>
Cc: John Stultz <john.stultz@linaro.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

printk: use a clever macro

Use the IS_ENABLED() macro rather than #ifdef blocks to set certain
global values.

Signed-off-by: Alex Elder <elder@linaro.org>
Acked-by: Borislav Petkov <bp@suse.de>
Reviewed-by: Petr Mladek <pmladek@suse.cz>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Jan Kara <jack@suse.cz>
Cc: John Stultz <john.stultz@linaro.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

printk: fix some comments

Fix a few comments that don't accurately describe their corresponding
code.  It also fixes some minor typographical errors.

Signed-off-by: Alex Elder <elder@linaro.org>
Reviewed-by: Petr Mladek <pmladek@suse.cz>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Borislav Petkov <bp@suse.de>
Cc: Jan Kara <jack@suse.cz>
Cc: John Stultz <john.stultz@linaro.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

printk: rename DEFAULT_MESSAGE_LOGLEVEL

Commit a8fe19ebfbfd ("kernel/printk: use symbolic defines for console
loglevels") makes consistent use of symbolic values for printk() log
levels.

The naming scheme used is different from the one used for
DEFAULT_MESSAGE_LOGLEVEL though.  Change that symbol name to be
MESSAGE_LOGLEVEL_DEFAULT for consistency.  And because the value of that
symbol comes from a similarly-named config option, rename
CONFIG_DEFAULT_MESSAGE_LOGLEVEL as well.

Signed-off-by: Alex Elder <elder@linaro.org>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Borislav Petkov <bp@suse.de>
Cc: Jan Kara <jack@suse.cz>
Cc: John Stultz <john.stultz@linaro.org>
Cc: Petr Mladek <pmladek@suse.cz>
Cc: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

printk: tweak do_syslog() to match comments

In do_syslog() there's a path used by kmsg_poll() and kmsg_read() that
only needs to know whether there's any data available to read (and not
its size).  These callers only check for non-zero return.  As a
shortcut, do_syslog() returns the difference between what has been
logged and what has been "seen."

The comments say that the "count of records" should be returned but it's
not.  Instead it returns (log_next_idx - syslog_idx), which is a
difference between buffer offsets--and the result could be negative.

The behavior is the same (it'll be zero or not in the same cases), but
the count of records is more meaningful and it matches what the comments
say.  So change the code to return that.

Signed-off-by: Alex Elder <elder@linaro.org>
Cc: Petr Mladek <pmladek@suse.cz>
Cc: Jan Kara <jack@suse.cz>
Cc: Joe Perches <joe@perches.com>
Cc: John Stultz <john.stultz@linaro.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

printk: allow increasing the ring buffer depending on the number of CPUs

The default size of the ring buffer is too small for machines with a
large amount of CPUs under heavy load.  What ends up happening when
debugging is the ring buffer overlaps and chews up old messages making
debugging impossible unless the size is passed as a kernel parameter.
An idle system upon boot up will on average spew out only about one or
two extra lines but where this really matters is on heavy load and that
will vary widely depending on the system and environment.

There are mechanisms to help increase the kernel ring buffer for tracing
through debugfs, and those interfaces even allow growing the kernel ring
buffer per CPU.  We also have a static value which can be passed upon
boot.  Relying on debugfs however is not ideal for production, and
relying on the value passed upon bootup is can only used *after* an
issue has creeped up.  Instead of being reactive this adds a proactive
measure which lets you scale the amount of contributions you'd expect to
the kernel ring buffer under load by each CPU in the worst case
scenario.

We use num_possible_cpus() to avoid complexities which could be
introduced by dynamically changing the ring buffer size at run time,
num_possible_cpus() lets us use the upper limit on possible number of
CPUs therefore avoiding having to deal with hotplugging CPUs on and off.
This introduces the kernel configuration option LOG_CPU_MAX_BUF_SHIFT
which is used to specify the maximum amount of contributions to the
kernel ring buffer in the worst case before the kernel ring buffer flips
over, the size is specified as a power of 2.  The total amount of
contributions made by each CPU must be greater than half of the default
kernel ring buffer size (1 << LOG_BUF_SHIFT bytes) in order to trigger
an increase upon bootup.  The kernel ring buffer is increased to the
next power of two that would fit the required minimum kernel ring buffer
size plus the additional CPU contribution.  For example if LOG_BUF_SHIFT
is 18 (256 KB) you'd require at least 128 KB contributions by other CPUs
in order to trigger an increase of the kernel ring buffer.  With a
LOG_CPU_BUF_SHIFT of 12 (4 KB) you'd require at least anything over > 64
possible CPUs to trigger an increase.  If you had 128 possible CPUs the
amount of minimum required kernel ring buffer bumps to:

   ((1 << 18) + ((128 - 1) * (1 << 12))) / 1024 = 764 KB

Since we require the ring buffer to be a power of two the new required
size would be 1024 KB.

This CPU contributions are ignored when the "log_buf_len" kernel
parameter is used as it forces the exact size of the ring buffer to an
expected power of two value.

[pmladek@suse.cz: fix build]
Signed-off-by: Luis R. Rodriguez <mcgrof@suse.com>
Signed-off-by: Petr Mladek <pmladek@suse.cz>
Tested-by: Davidlohr Bueso <davidlohr@hp.com>
Tested-by: Petr Mladek <pmladek@suse.cz>
Reviewed-by: Davidlohr Bueso <davidlohr@hp.com>
Cc: Andrew Lunn <andrew@lunn.ch>
Cc: Stephen Warren <swarren@wwwdotorg.org>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Petr Mladek <pmladek@suse.cz>
Cc: Joe Perches <joe@perches.com>
Cc: Arun KS <arunks.linux@gmail.com>
Cc: Kees Cook <keescook@chromium.org>
Cc: Davidlohr Bueso <davidlohr@hp.com>
Cc: Chris Metcalf <cmetcalf@tilera.com>
Cc: Jan Kara <jack@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

printk: make dynamic units clear for the kernel ring buffer

Signed-off-by: Luis R. Rodriguez <mcgrof@suse.com>
Suggested-by: Davidlohr Bueso <davidlohr@hp.com>
Cc: Andrew Lunn <andrew@lunn.ch>
Cc: Stephen Warren <swarren@wwwdotorg.org>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Petr Mladek <pmladek@suse.cz>
Cc: Joe Perches <joe@perches.com>
Cc: Arun KS <arunks.linux@gmail.com>
Cc: Kees Cook <keescook@chromium.org>
Cc: Davidlohr Bueso <davidlohr@hp.com>
Cc: Chris Metcalf <cmetcalf@tilera.com>
Cc: Jan Kara <jack@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

printk: move power of 2 practice of ring buffer size to a helper

In practice the power of 2 practice of the size of the kernel ring
buffer remains purely historical but not a requirement, specially now
that we have LOG_ALIGN and use it for both static and dynamic
allocations.  It could have helped with implicit alignment back in the
days given the even the dynamically sized ring buffer was guaranteed to
be aligned so long as CONFIG_LOG_BUF_SHIFT was set to produce a
__LOG_BUF_LEN which is architecture aligned, since log_buf_len=n would
be allowed only if it was > __LOG_BUF_LEN and we always ended up
rounding the log_buf_len=n to the next power of 2 with
roundup_pow_of_two(), any multiple of 2 then should be also architecture
aligned.  These assumptions of course relied heavily on
CONFIG_LOG_BUF_SHIFT producing an aligned value but users can always
change this.

We now have precise alignment requirements set for the log buffer size
for both static and dynamic allocations, but lets upkeep the old
practice of using powers of 2 for its size to help with easy expected
scalable values and the allocators for dynamic allocations.  We'll reuse
this later so move this into a helper.

Signed-off-by: Luis R. Rodriguez <mcgrof@suse.com>
Cc: Andrew Lunn <andrew@lunn.ch>
Cc: Stephen Warren <swarren@wwwdotorg.org>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Petr Mladek <pmladek@suse.cz>
Cc: Joe Perches <joe@perches.com>
Cc: Arun KS <arunks.linux@gmail.com>
Cc: Kees Cook <keescook@chromium.org>
Cc: Davidlohr Bueso <davidlohr@hp.com>
Cc: Chris Metcalf <cmetcalf@tilera.com>
Cc: Jan Kara <jack@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

printk: make dynamic kernel ring buffer alignment explicit

We have to consider alignment for the ring buffer both for the default
static size, and then also for when an dynamic allocation is made when
the log_buf_len=n kernel parameter is passed to set the size
specifically to a size larger than the default size set by the
architecture through CONFIG_LOG_BUF_SHIFT.

The default static kernel ring buffer can be aligned properly if
architectures set CONFIG_LOG_BUF_SHIFT properly, we provide ranges for
the size though so even if CONFIG_LOG_BUF_SHIFT has a sensible aligned
value it can be reduced to a non aligned value.  Commit 6ebb017de9
("printk: Fix alignment of buf causing crash on ARM EABI") by Andrew
Lunn ensures the static buffer is always aligned and the decision of
alignment is done by the compiler by using __alignof__(struct log).

When log_buf_len=n is used we allocate the ring buffer dynamically.
Dynamic allocation varies, for the early allocation called before
setup_arch() memblock_virt_alloc() requests a page aligment and for the
default kernel allocation memblock_virt_alloc_nopanic() requests no
special alignment, which in turn ends up aligning the allocation to
SMP_CACHE_BYTES, which is L1 cache aligned.

Since we already have the required alignment for the kernel ring buffer
though we can do better and request explicit alignment for LOG_ALIGN.
This does that to be safe and make dynamic allocation alignment
explicit.

Signed-off-by: Luis R. Rodriguez <mcgrof@suse.com>
Tested-by: Petr Mladek <pmladek@suse.cz>
Acked-by: Petr Mladek <pmladek@suse.cz>
Cc: Andrew Lunn <andrew@lunn.ch>
Cc: Stephen Warren <swarren@wwwdotorg.org>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Petr Mladek <pmladek@suse.cz>
Cc: Joe Perches <joe@perches.com>
Cc: Arun KS <arunks.linux@gmail.com>
Cc: Kees Cook <keescook@chromium.org>
Cc: Davidlohr Bueso <davidlohr@hp.com>
Cc: Chris Metcalf <cmetcalf@tilera.com>
Cc: Jan Kara <jack@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

include/linux/byteorder/generic.h: minor comment fix

Signed-off-by: Geoff Levand <geoff@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fs.h, drivers/hwmon/asus_atk0110.c: fix DEFINE_SIMPLE_ATTRIBUTE semicolon definition and use

The DEFINE_SIMPLE_ATTRIBUTE macro should not end in a ; Fix the one use
in the kernel tree that did not have a semicolon.

Signed-off-by: Joe Perches <joe@perches.com>
Acked-by: Guenter Roeck <linux@roeck-us.net>
Acked-by: Luca Tettamanti <kronos.it@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

./Makefile: tell gcc optimizer to never introduce new data races

We have been chasing a memory corruption bug, which turned out to be
caused by very old gcc (4.3.4), which happily turned conditional load
into a non-conditional one, and that broke correctness (the condition
was met only if lock was held) and corrupted memory.

This particular problem with that particular code did not happen when
never gccs were used.  I've brought this up with our gcc folks, as I
wanted to make sure that this can't really happen again, and it turns
out it actually can.

Quoting Martin Jambor <mjambor@suse.cz>:
 "More current GCCs are more careful when it comes to replacing a
  conditional load with a non-conditional one, most notably they check
  that a store happens in each iteration of _a_ loop but they assume
  loops are executed.  They also perform a simple check whether the
  store cannot trap which currently passes only for non-const
  variables.  A simple testcase demonstrating it on an x86_64 is for
  example the following:

  $ cat cond_store.c

  int g_1 = 1;

  int g_2[1024] __attribute__((section ("safe_section"), aligned (4096)));

  int c = 4;

  int __attribute__ ((noinline))
  foo (void)
  {
    int l;
    for (l = 0; (l != 4); l++) {
      if (g_1)
        return l;
      for (g_2[0] = 0; (g_2[0] >= 26); ++g_2[0])
        ;
    }
    return 2;
  }

  int main (int argc, char* argv[])
  {
    if (mprotect (g_2, sizeof(g_2), PROT_READ) == -1)
      {
        int e = errno;
        error (e, e, "mprotect error %i", e);
      }
    foo ();
    __builtin_printf("OK\n");
    return 0;
  }
  /* EOF */
  $ ~/gcc/trunk/inst/bin/gcc cond_store.c -O2 --param allow-store-data-races=0
  $ ./a.out
  OK
  $ ~/gcc/trunk/inst/bin/gcc cond_store.c -O2 --param allow-store-data-races=1
  $ ./a.out
  Segmentation fault

  The testcase fails the same at least with 4.9, 4.8 and 4.7.  Therefore
  I would suggest building kernels with this parameter set to zero. I
  also agree with Jikos that the default should be changed for -O2.  I
  have run most of the SPEC 2k6 CPU benchmarks (gamess and dealII
  failed, at -O2, not sure why) compiled with and without this option
  and did not see any real difference between respective run-times"

Hopefully the default will be changed in newer gccs, but let's force it
for kernel builds so that we are on a safe side even when older gcc are
used.

The code in question was out-of-tree printk-in-NMI (yeah, surprise
suprise, once again) patch written by Petr Mladek, let me quote his
comment from our internal bugzilla:

 "I have spent few days investigating inconsistent state of kernel ring buffer.
  It went out that it was caused by speculative store generated by
  gcc-4.3.4.

  The problem is in assembly generated for make_free_space(). The functions is
  called the following way:

  + vprintk_emit();
      + log = MAIN_LOG; // with logbuf_lock
         or
         log = NMI_LOG; // with nmi_logbuf_lock
         cont_add(log, ...);
          + cont_flush(log, ...);
              + log_store(log, ...);
                    + log_make_free_space(log, ...);

  If called with log = NMI_LOG then only nmi_log_* global variables are safe to
  modify but the generated code does store also into (main_)log_* global
  variables:

  <log_make_free_space>:
         55                      push   %rbp
         89 f6                   mov    %esi,%esi

         48 8b 05 03 99 51 01    mov    0x1519903(%rip),%rax       # ffffffff82620868 <nmi_log_next_id>
         44 8b 1d ec 98 51 01    mov    0x15198ec(%rip),%r11d      # ffffffff82620858 <log_next_idx>
         8b 35 36 60 14 01       mov    0x1146036(%rip),%esi       # ffffffff8224cfa8 <log_buf_len>
         44 8b 35 33 60 14 01    mov    0x1146033(%rip),%r14d      # ffffffff8224cfac <nmi_log_buf_len>
         4c 8b 2d d0 98 51 01    mov    0x15198d0(%rip),%r13       # ffffffff82620850 <log_next_seq>
         4c 8b 25 11 61 14 01    mov    0x1146111(%rip),%r12       # ffffffff8224d098 <log_buf>
         49 89 c2                mov    %rax,%r10
         48 21 c2                and    %rax,%rdx
         48 8b 1d 0c 99 55 01    mov    0x155990c(%rip),%rbx       # ffffffff826608a0 <nmi_log_buf>
         49 c1 ea 20             shr    $0x20,%r10
         48 89 55 d0             mov    %rdx,-0x30(%rbp)
         44 29 de                sub    %r11d,%esi
         45 29 d6                sub    %r10d,%r14d
         4c 8b 0d 97 98 51 01    mov    0x1519897(%rip),%r9 # ffffffff82620840 <log_first_seq>
         eb 7e                   jmp    ffffffff81107029 <log_make_free_space+0xe9>
  [...]
         85 ff                   test   %edi,%edi                  # edi = 1 for NMI_LOG
         4c 89 e8                mov    %r13,%rax
         4c 89 ca                mov    %r9,%rdx
         74 0a                   je     ffffffff8110703d <log_make_free_space+0xfd>
         8b 15 27 98 51 01       mov    0x1519827(%rip),%edx       # ffffffff82620860 <nmi_log_first_id>
         48 8b 45 d0             mov    -0x30(%rbp),%rax
         48 39 c2                cmp    %rax,%rdx                  # end of loop
         0f 84 da 00 00 00       je     ffffffff81107120 <log_make_free_space+0x1e0>
  [...]
         85 ff                   test   %edi,%edi                  # edi = 1 for NMI_LOG
         4c 89 0d 17 97 51 01    mov    %r9,0x1519717(%rip)        # ffffffff82620840 <log_first_seq>
                                 ^^^^^^^^^^^^^^^^^^^^^^^^^^
                                 KABOOOM
         74 35                   je     ffffffff81107160  <log_make_free_space+0x220>

  It stores log_first_seq when edi == NMI_LOG. This instructions are used also
  when edi == MAIN_LOG but the store is done speculatively before the condition
  is decided.  It is unsafe because we do not have "logbuf_lock" in NMI context
  and some other process migh modify "log_first_seq" in parallel"

I believe that the best course of action is both

 - building kernel (and anything multi-threaded, I guess) with that
   optimization turned off
 - persuade gcc folks to change the default for future releases

Signed-off-by: Jiri Kosina <jkosina@suse.cz>
Cc: Martin Jambor <mjambor@suse.cz>
Cc: Petr Mladek <pmladek@suse.cz>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Marek Polacek <polacek@redhat.com>
Cc: Jakub Jelinek <jakub@redhat.com>
Cc: Steven Noonan <steven@uplinklabs.net>
Cc: Richard Biener <richard.guenther@gmail.com>
Cc: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/zpool: update zswap to use zpool

Change zswap to use the zpool api instead of directly using zbud.  Add a
boot-time param to allow selecting which zpool implementation to use,
with zbud as the default.

Signed-off-by: Dan Streetman <ddstreet@ieee.org>
Tested-by: Seth Jennings <sjennings@variantweb.net>
Cc: Weijie Yang <weijie.yang@samsung.com>
Cc: 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>

mm/zpool: zbud/zsmalloc implement zpool

Update zbud and zsmalloc to implement the zpool api.

[fengguang.wu@intel.com: make functions static]
Signed-off-by: Dan Streetman <ddstreet@ieee.org>
Tested-by: Seth Jennings <sjennings@variantweb.net>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Nitin Gupta <ngupta@vflare.org>
Cc: Weijie Yang <weijie.yang@samsung.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/zpool: implement common zpool api to zbud/zsmalloc

Add zpool api.

zpool provides an interface for memory storage, typically of compressed
memory.  Users can select what backend to use; currently the only
implementations are zbud, a low density implementation with up to two
compressed pages per storage page, and zsmalloc, a higher density
implementation with multiple compressed pages per storage page.

Signed-off-by: Dan Streetman <ddstreet@ieee.org>
Tested-by: Seth Jennings <sjennings@variantweb.net>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Nitin Gupta <ngupta@vflare.org>
Cc: Weijie Yang <weijie.yang@samsung.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/zbud: change zbud_alloc size type to size_t

Change the type of the zbud_alloc() size param from unsigned int to
size_t.

Technically, this should not make any difference, as the zbud
implementation already restricts the size to well within either type's
limits; but as zsmalloc (and kmalloc) use size_t, and zpool will use
size_t, this brings the size parameter type in line with zsmalloc/zpool.

Signed-off-by: Dan Streetman <ddstreet@ieee.org>
Acked-by: Seth Jennings <sjennings@variantweb.net>
Tested-by: Seth Jennings <sjennings@variantweb.net>
Cc: Weijie Yang <weijie.yang@samsung.com>
Cc: 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>

zram: replace global tb_lock with fine grain lock

Currently, we use a rwlock tb_lock to protect concurrent access to the
whole zram meta table.  However, according to the actual access model,
there is only a small chance for upper user to access the same
table[index], so the current lock granularity is too big.

The idea of optimization is to change the lock granularity from whole
meta table to per table entry (table -> table[index]), so that we can
protect concurrent access to the same table[index], meanwhile allow the
maximum concurrency.

With this in mind, several kinds of locks which could be used as a
per-entry lock were tested and compared:

Test environment:
x86-64 Intel Core2 Q8400, system memory 4GB, Ubuntu 12.04,
kernel v3.15.0-rc3 as base, zram with 4 max_comp_streams LZO.

iozone test:
iozone -t 4 -R -r 16K -s 200M -I +Z
(1GB zram with ext4 filesystem, take the average of 10 tests, KB/s)

      Test       base      CAS    spinlock    rwlock   bit_spinlock
-------------------------------------------------------------------
 Initial write  1381094   1425435   1422860   1423075   1421521
       Rewrite  1529479   1641199   1668762   1672855   1654910
          Read  8468009  11324979  11305569  11117273  10997202
       Re-read  8467476  11260914  11248059  11145336  10906486
  Reverse Read  6821393   8106334   8282174   8279195   8109186
   Stride read  7191093   8994306   9153982   8961224   9004434
   Random read  7156353   8957932   9167098   8980465   8940476
Mixed workload  4172747   5680814   5927825   5489578   5972253
  Random write  1483044   1605588   1594329   1600453   1596010
        Pwrite  1276644   1303108   1311612   1314228   1300960
         Pread  4324337   4632869   4618386   4457870   4500166

To enhance the possibility of access the same table[index] concurrently,
set zram a small disksize(10MB) and let threads run with large loop
count.

fio test:
fio --bs=32k --randrepeat=1 --randseed=100 --refill_buffers
--scramble_buffers=1 --direct=1 --loops=3000 --numjobs=4
--filename=/dev/zram0 --name=seq-write --rw=write --stonewall
--name=seq-read --rw=read --stonewall --name=seq-readwrite
--rw=rw --stonewall --name=rand-readwrite --rw=randrw --stonewall
(10MB zram raw block device, take the average of 10 tests, KB/s)

    Test     base     CAS    spinlock    rwlock  bit_spinlock
-------------------------------------------------------------
seq-write   933789   999357   1003298    995961   1001958
 seq-read  5634130  6577930   6380861   6243912   6230006
   seq-rw  1405687  1638117   1640256   1633903   1634459
  rand-rw  1386119  1614664   1617211   1609267   1612471

All the optimization methods show a higher performance than the base,
however, it is hard to say which method is the most appropriate.

On the other hand, zram is mostly used on small embedded system, so we
don't want to increase any memory footprint.

This patch pick the bit_spinlock method, pack object size and page_flag
into an unsigned long table.value, so as to not increase any memory
overhead on both 32-bit and 64-bit system.

On the third hand, even though different kinds of locks have different
performances, we can ignore this difference, because: if zram is used as
zram swapfile, the swap subsystem can prevent concurrent access to the
same swapslot; if zram is used as zram-blk for set up filesystem on it,
the upper filesystem and the page cache also prevent concurrent access
of the same block mostly.  So we can ignore the different performances
among locks.

Acked-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Reviewed-by: Davidlohr Bueso <davidlohr@hp.com>
Signed-off-by: Weijie Yang <weijie.yang@samsung.com>
Signed-off-by: Minchan Kim <minchan@kernel.org>
Cc: Jerome Marchand <jmarchan@redhat.com>
Cc: Nitin Gupta <ngupta@vflare.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

zram: use size_t instead of u16

Some architectures (eg, hexagon and PowerPC) could use PAGE_SHIFT of 16
or more.  In these cases u16 is not sufficiently large to represent a
compressed page's size so use size_t.

Signed-off-by: Minchan Kim <minchan@kernel.org>
Reported-by: Weijie Yang <weijie.yang@samsung.com>
Acked-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Cc: Jerome Marchand <jmarchan@redhat.com>
Cc: Nitin Gupta <ngupta@vflare.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

zram: remove unused SECTOR_SIZE define

Drop SECTOR_SIZE define, because it's not used.

Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Nitin Gupta <ngupta@vflare.org>
Cc: Weijie Yang <weijie.yang@samsung.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

zram: rename struct `table' to `zram_table_entry'

Andrew Morton has recently noted that `struct table' actually represents
table entry and, thus, should be renamed.  Rename to `zram_table_entry'.

Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Nitin Gupta <ngupta@vflare.org>
Cc: Weijie Yang <weijie.yang@samsung.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/highmem: make kmap cache coloring aware

User-visible effect:
 Architectures that choose this method of maintaining cache coherency
(MIPS and xtensa currently) are able to use high memory on cores with
aliasing data cache.  Without this fix such architectures can not use
high memory (in case of xtensa it means that at most 128 MBytes of
physical memory is available).

The problem:
 VIPT cache with way size larger than MMU page size may suffer from
aliasing problem: a single physical address accessed via different
virtual addresses may end up in multiple locations in the cache.
Virtual mappings of a physical address that always get cached in
different cache locations are said to have different colors.  L1 caching
hardware usually doesn't handle this situation leaving it up to
software.  Software must avoid this situation as it leads to data
corruption.

What can be done:
 One way to handle this is to flush and invalidate data cache every time
page mapping changes color.  The other way is to always map physical
page at a virtual address with the same color.  Low memory pages already
have this property.  Giving architecture a way to control color of high
memory page mapping allows reusing of existing low memory cache alias
handling code.

How this is done with this patch:
 Provide hooks that allow architectures with aliasing cache to align
mapping address of high pages according to their color.  Such
architectures may enforce similar coloring of low- and high-memory page
mappings and reuse existing cache management functions to support
highmem.

This code is based on the implementation of similar feature for MIPS by
Leonid Yegoshin.

Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
Cc: Leonid Yegoshin <Leonid.Yegoshin@imgtec.com>
Cc: Chris Zankel <chris@zankel.net>
Cc: Marc Gauthier <marc@cadence.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Steven Hill <Steven.Hill@imgtec.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mmu_notifier: add call_srcu and sync function for listener to delay call and sync

When kernel device drivers or subsystems want to bind their lifespan to
t= he lifespan of the mm_struct, they usually use one of the following
methods:

1. Manually calling a function in the interested kernel module.  The
   funct= ion call needs to be placed in mmput.  This method was rejected
   by several ker= nel maintainers.

2. Registering to the mmu notifier release mechanism.

The problem with the latter approach is that the mmu_notifier_release
cal= lback is called from__mmu_notifier_release (called from exit_mmap).
That functi= on iterates over the list of mmu notifiers and don't expect
the release call= back function to remove itself from the list.
Therefore, the callback function= in the kernel module can't release the
mmu_notifier_object, which is actuall= y the kernel module's object
itself.  As a result, the destruction of the kernel module's object must
to be done in a delayed fashion.

This patch adds support for this delayed callback, by adding a new
mmu_notifier_call_srcu function that receives a function ptr and calls
th= at function with call_srcu.  In that function, the kernel module
releases its object.  To use mmu_notifier_call_srcu, the calling module
needs to call b= efore that a new function called
mmu_notifier_unregister_no_release that as its= name implies,
unregisters a notifier without calling its notifier release call= back.

This patch also adds a function that will call barrier_srcu so those
kern= el modules can sync with mmu_notifier.

Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Jérôme Glisse <jglisse@redhat.com>
Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: BUG when __kmap_atomic_idx equals KM_TYPE_NR

__kmap_atomic_idx is per_cpu variable.  Each CPU can use KM_TYPE_NR
entries from FIXMAP i.e.  from 0 to KM_TYPE_NR - 1.  Allowing
__kmap_atomic_idx to over- shoot to KM_TYPE_NR can mess up with next
CPU's 0th entry which is a bug.  Hence BUG_ON if __kmap_atomic_idx >=
KM_TYPE_NR.

Fix the off-by-on in this test.

Signed-off-by: Chintan Pandya <cpandya@codeaurora.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>

mm: memcontrol: clean up reclaim size variable use in try_charge()

Charge reclaim and OOM currently use the charge batch variable, but
batching is already disabled at that point.  To simplify the charge
logic, the batch variable is reset to the original request size when
reclaim is entered, so it's functionally equal, but it's misleading.

Switch reclaim/OOM to nr_pages, which is the original request size.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

kernel/smp.c:on_each_cpu_cond(): fix warning in fallback path

The rarely-executed memry-allocation-failed callback path generates a
WARN_ON_ONCE() when smp_call_function_single() succeeds.  Presumably
it's supposed to warn on failures.

Signed-off-by: Sasha Levin <sasha.levin@oracle.com>
Cc: Christoph Lameter <cl@gentwo.org>
Cc: Gilad Ben-Yossef <gilad@benyossef.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Tejun Heo <htejun@gmail.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: change confusing #ifdef use in __access_remote_vm

This patch changes confusing #ifdef use in __access_remote_vm into
merely ugly #ifdef use.

Addresses bug https://bugzilla.kernel.org/show_bug.cgi?id=81651

Signed-off-by: Rik van Riel <riel@redhat.com>
Reported-by: David Binderman <dcb314@hotmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: softdirty: respect VM_SOFTDIRTY in PTE holes

After a VMA is created with the VM_SOFTDIRTY flag set, /proc/pid/pagemap
should report that the VMA's virtual pages are soft-dirty until
VM_SOFTDIRTY is cleared (i.e., by the next write of "4" to
/proc/pid/clear_refs).  However, pagemap ignores the VM_SOFTDIRTY flag
for virtual addresses that fall in PTE holes (i.e., virtual addresses
that don't have a PMD, PUD, or PGD allocated yet).

To observe this bug, use mmap to create a VMA large enough such that
there's a good chance that the VMA will occupy an unused PMD, then test
the soft-dirty bit on its pages.  In practice, I found that a VMA that
covered a PMD's worth of address space was big enough.

This patch adds the necessary VMA lookup to the PTE hole callback in
/proc/pid/pagemap's page walk and sets soft-dirty according to the VMAs'
VM_SOFTDIRTY flag.

Signed-off-by: Peter Feiner <pfeiner@google.com>
Acked-by: Cyrill Gorcunov <gorcunov@openvz.org>
Cc: Pavel Emelyanov <xemul@parallels.com>
Cc: Hugh Dickins <hughd@google.com>
Acked-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: mark fault_around_bytes __read_mostly

fault_around_bytes can only be changed via debugfs.  Let's mark it
read-mostly.

Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Suggested-by: David Rientjes <rientjes@google.com>
Acked-by: David Rientjes <rientjes@google.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Andrey Ryabinin <a.ryabinin@samsung.com>
Cc: Sasha Levin <sasha.levin@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: close race between do_fault_around() and fault_around_bytes_set()

Things can go wrong if fault_around_bytes will be changed under
do_fault_around(): between fault_around_mask() and fault_around_pages().

Let's read fault_around_bytes only once during do_fault_around() and
calculate mask based on the reading.

Note: fault_around_bytes can only be updated via debug interface.  Also
I've tried but was not able to trigger a bad behaviour without the
patch.  So I would not consider this patch as urgent.

Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Andrey Ryabinin <a.ryabinin@samsung.com>
Cc: Sasha Levin <sasha.levin@oracle.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>

memcg, vmscan: Fix forced scan of anonymous pages

When memory cgoups are enabled, the code that decides to force to scan
anonymous pages in get_scan_count() compares global values (free,
high_watermark) to a value that is restricted to a memory cgroup (file).
It make the code over-eager to force anon scan.

For instance, it will force anon scan when scanning a memcg that is
mainly populated by anonymous page, even when there is plenty of file
pages to get rid of in others memcgs, even when swappiness == 0.  It
breaks user's expectation about swappiness and hurts performance.

This patch makes sure that forced anon scan only happens when there not
enough file pages for the all zone, not just in one random memcg.

[hannes@cmpxchg.org: cleanups]
Signed-off-by: Jerome Marchand <jmarchan@redhat.com>
Acked-by: Michal Hocko <mhocko@suse.cz>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Reviewed-by: Rik van Riel <riel@redhat.com>
Cc: Mel Gorman <mgorman@suse.de>
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, vmscan: fix an outdated comment still mentioning get_scan_ratio

Quite a while ago, get_scan_ratio() has been renamed get_scan_count(),
however a comment in shrink_active_list() still mention it.  This patch
fixes the outdated comment.

Signed-off-by: Jerome Marchand <jmarchan@redhat.com>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Mel Gorman <mgorman@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, oom: remove unnecessary exit_state check

The oom killer scans each process and determines whether it is eligible
for oom kill or whether the oom killer should abort because of
concurrent memory freeing.  It will abort when an eligible process is
found to have TIF_MEMDIE set, meaning it has already been oom killed and
we're waiting for it to exit.

Processes with task->mm == NULL should not be considered because they
are either kthreads or have already detached their memory and killing
them would not lead to memory freeing.  That memory is only freed after
exit_mm() has returned, however, and not when task->mm is first set to
NULL.

Clear TIF_MEMDIE after exit_mm()'s mmput() so that an oom killed process
is no longer considered for oom kill, but only until exit_mm() has
returned.  This was fragile in the past because it relied on
exit_notify() to be reached before no longer considering TIF_MEMDIE
processes.

Signed-off-by: David Rientjes <rientjes@google.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: fix potential infinite loop in dissolve_free_huge_pages()

It is possible for some platforms, such as powerpc to set HPAGE_SHIFT to
0 to indicate huge pages not supported.

When this is the case, hugetlbfs could be disabled during boot time:
hugetlbfs: disabling because there are no supported hugepage sizes

Then in dissolve_free_huge_pages(), order is kept maximum (64 for
64bits), and the for loop below won't end: for (pfn = start_pfn; pfn <
end_pfn; pfn += 1 << order)

As suggested by Naoya, below fix checks hugepages_supported() before
calling dissolve_free_huge_pages().

[rientjes@google.com: no legitimate reason to call dissolve_free_huge_pages() when !hugepages_supported()]
Signed-off-by: Li Zhong <zhong@linux.vnet.ibm.com>
Acked-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Acked-by: David Rientjes <rientjes@google.com>
Signed-off-by: David Rientjes <rientjes@google.com>
Cc: <stable@vger.kernel.org> [3.12+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, thp: restructure thp avoidance of light synchronous migration

__GFP_NO_KSWAPD, once the way to determine if an allocation was for thp
or not, has gained more users.  Their use is not necessarily wrong, they
are trying to do a memory allocation that can easily fail without
disturbing kswapd, so the bit has gained additional usecases.

This restructures the check to determine whether MIGRATE_SYNC_LIGHT
should be used for memory compaction in the page allocator.  Rather than
testing solely for __GFP_NO_KSWAPD, test for all bits that must be set
for thp allocations.

This also moves the check to be done only after the page allocator is
aborted for deferred or contended memory compaction since setting
migration_mode for this case is pointless.

Signed-off-by: David Rientjes <rientjes@google.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Rik van Riel <riel@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, oom: rename zonelist locking functions

try_set_zonelist_oom() and clear_zonelist_oom() are not named properly
to imply that they require locking semantics to avoid out_of_memory()
being reordered.

zone_scan_lock is required for both functions to ensure that there is
proper locking synchronization.

Rename try_set_zonelist_oom() to oom_zonelist_trylock() and rename
clear_zonelist_oom() to oom_zonelist_unlock() to imply there is proper
locking semantics.

At the same time, convert oom_zonelist_trylock() to return bool instead
of int since only success and failure are tested.

Signed-off-by: David Rientjes <rientjes@google.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, oom: ensure memoryless node zonelist always includes zones

With memoryless node support being worked on, it's possible that for
optimizations that a node may not have a non-NULL zonelist.  When
CONFIG_NUMA is enabled and node 0 is memoryless, this means the zonelist
for first_online_node may become NULL.

The oom killer requires a zonelist that includes all memory zones for
the sysrq trigger and pagefault out of memory handler.

Ensure that a non-NULL zonelist is always passed to the oom killer.

[akpm@linux-foundation.org: fix non-numa build]
Signed-off-by: David Rientjes <rientjes@google.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memory-hotplug: sh: suitable memory should go to ZONE_MOVABLE

This patch introduces zone_for_memory() to arch_add_memory() on sh to
ensure new, higher memory added into ZONE_MOVABLE if movable zone has
already setup.

Signed-off-by: Wang Nan <wangnan0@huawei.com>
Cc: Zhang Yanfei <zhangyanfei@cn.fujitsu.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Yinghai Lu <yinghai@kernel.org>
Cc: "Mel Gorman" <mgorman@suse.de>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Chris Metcalf <cmetcalf@tilera.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memory-hotplug: ppc: suitable memory should go to ZONE_MOVABLE

This patch introduces zone_for_memory() to arch_add_memory() on powerpc
to ensure new, higher memory added into ZONE_MOVABLE if movable zone has
already setup.

Signed-off-by: Wang Nan <wangnan0@huawei.com>
Cc: Zhang Yanfei <zhangyanfei@cn.fujitsu.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Yinghai Lu <yinghai@kernel.org>
Cc: "Mel Gorman" <mgorman@suse.de>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Chris Metcalf <cmetcalf@tilera.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memory-hotplug: ia64: suitable memory should go to ZONE_MOVABLE

This patch introduces zone_for_memory() to arch_add_memory() on ia64 to
ensure new, higher memory added into ZONE_MOVABLE if movable zone has
already setup.

Signed-off-by: Wang Nan <wangnan0@huawei.com>
Cc: Zhang Yanfei <zhangyanfei@cn.fujitsu.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Yinghai Lu <yinghai@kernel.org>
Cc: "Mel Gorman" <mgorman@suse.de>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Chris Metcalf <cmetcalf@tilera.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memory-hotplug: x86_32: suitable memory should go to ZONE_MOVABLE

This patch introduces zone_for_memory() to arch_add_memory() on x86_32
to ensure new, higher memory added into ZONE_MOVABLE if movable zone has
already setup.

Signed-off-by: Wang Nan <wangnan0@huawei.com>
Cc: Zhang Yanfei <zhangyanfei@cn.fujitsu.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Yinghai Lu <yinghai@kernel.org>
Cc: "Mel Gorman" <mgorman@suse.de>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Chris Metcalf <cmetcalf@tilera.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memory-hotplug: x86_64: suitable memory should go to ZONE_MOVABLE

This patch introduces zone_for_memory() to arch_add_memory() on x86_64
to ensure new, higher memory added into ZONE_MOVABLE if movable zone has
already setup.

Signed-off-by: Wang Nan <wangnan0@huawei.com>
Cc: Zhang Yanfei <zhangyanfei@cn.fujitsu.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Yinghai Lu <yinghai@kernel.org>
Cc: "Mel Gorman" <mgorman@suse.de>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Chris Metcalf <cmetcalf@tilera.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memory-hotplug: add zone_for_memory() for selecting zone for new memory

This series of patches fixes a problem when adding memory in bad manner.
For example: for a x86_64 machine booted with "mem=400M" and with 2GiB
memory installed, following commands cause problem:

  # echo 0x40000000 > /sys/devices/system/memory/probe
 [   28.613895] init_memory_mapping: [mem 0x40000000-0x47ffffff]
  # echo 0x48000000 > /sys/devices/system/memory/probe
 [   28.693675] init_memory_mapping: [mem 0x48000000-0x4fffffff]
  # echo online_movable > /sys/devices/system/memory/memory9/state
  # echo 0x50000000 > /sys/devices/system/memory/probe
 [   29.084090] init_memory_mapping: [mem 0x50000000-0x57ffffff]
  # echo 0x58000000 > /sys/devices/system/memory/probe
 [   29.151880] init_memory_mapping: [mem 0x58000000-0x5fffffff]
  # echo online_movable > /sys/devices/system/memory/memory11/state
  # echo online> /sys/devices/system/memory/memory8/state
  # echo online> /sys/devices/system/memory/memory10/state
  # echo offline> /sys/devices/system/memory/memory9/state
 [   30.558819] Offlined Pages 32768
  # free
              total       used       free     shared    buffers     cached
 Mem:        780588 18014398509432020     830552          0          0      51180
 -/+ buffers/cache: 18014398509380840     881732
 Swap:            0          0          0

This is because the above commands probe higher memory after online a
section with online_movable, which causes ZONE_HIGHMEM (or ZONE_NORMAL
for systems without ZONE_HIGHMEM) overlaps ZONE_MOVABLE.

After the second online_movable, the problem can be observed from
zoneinfo:

  # cat /proc/zoneinfo
  ...
  Node 0, zone  Movable
    pages free     65491
          min      250
          low      312
          high     375
          scanned  0
          spanned  18446744073709518848
          present  65536
          managed  65536
  ...

This series of patches solve the problem by checking ZONE_MOVABLE when
choosing zone for new memory.  If new memory is inside or higher than
ZONE_MOVABLE, makes it go there instead.

After applying this series of patches, following are free and zoneinfo
result (after offlining memory9):

  bash-4.2# free
                total       used       free     shared    buffers     cached
   Mem:        780956      80112     700844          0          0      51180
   -/+ buffers/cache:      28932     752024
   Swap:            0          0          0

  bash-4.2# cat /proc/zoneinfo

  Node 0, zone      DMA
    pages free     3389
          min      14
          low      17
          high     21
          scanned  0
          spanned  4095
          present  3998
          managed  3977
      nr_free_pages 3389
  ...
    start_pfn:         1
    inactive_ratio:    1
  Node 0, zone    DMA32
    pages free     73724
          min      341
          low      426
          high     511
          scanned  0
          spanned  98304
          present  98304
          managed  92958
      nr_free_pages 73724
    ...
    start_pfn:         4096
    inactive_ratio:    1
  Node 0, zone   Normal
    pages free     32630
          min      120
          low      150
          high     180
          scanned  0
          spanned  32768
          present  32768
          managed  32768
      nr_free_pages 32630
  ...
    start_pfn:         262144
    inactive_ratio:    1
  Node 0, zone  Movable
    pages free     65476
          min      241
          low      301
          high     361
          scanned  0
          spanned  98304
          present  65536
          managed  65536
      nr_free_pages 65476
  ...
    start_pfn:         294912
    inactive_ratio:    1

This patch (of 7):

Introduce zone_for_memory() in arch independent code for
arch_add_memory() use.

Many arch_add_memory() function simply selects ZONE_HIGHMEM or
ZONE_NORMAL and add new memory into it.  However, with the existance of
ZONE_MOVABLE, the selection method should be carefully considered: if
new, higher memory is added after ZONE_MOVABLE is setup, the default
zone and ZONE_MOVABLE may overlap each other.

should_add_memory_movable() checks the status of ZONE_MOVABLE.  If it
has already contain memory, compare the address of new memory and
movable memory.  If new memory is higher than movable, it should be
added into ZONE_MOVABLE instead of default zone.

Signed-off-by: Wang Nan <wangnan0@huawei.com>
Cc: Zhang Yanfei <zhangyanfei@cn.fujitsu.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Yinghai Lu <yinghai@kernel.org>
Cc: "Mel Gorman" <mgorman@suse.de>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Chris Metcalf <cmetcalf@tilera.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

slub: remove kmemcg id from create_unique_id

This function is never called for memcg caches, because they are
unmergeable, so remove the dead code.

Signed-off-by: Vladimir Davydov <vdavydov@parallels.com>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Christoph Lameter <cl@linux.com>
Reviewed-by: Pekka Enberg <penberg@kernel.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, writeback: prevent race when calculating dirty limits

Setting vm_dirty_bytes and dirty_background_bytes is not protected by
any serialization.

Therefore, it's possible for either variable to change value after the
test in global_dirty_limits() to determine whether available_memory
needs to be initialized or not.

Always ensure that available_memory is properly initialized.

Signed-off-by: David Rientjes <rientjes@google.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Michal Hocko <mhocko@suse.cz>
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>

mm, thp: only collapse hugepages to nodes with affinity for zone_reclaim_mode

Commit 9f1b868a13ac ("mm: thp: khugepaged: add policy for finding target
node") improved the previous khugepaged logic which allocated a
transparent hugepages from the node of the first page being collapsed.

However, it is still possible to collapse pages to remote memory which
may suffer from additional access latency.  With the current policy, it
is possible that 255 pages (with PAGE_SHIFT == 12) will be collapsed
remotely if the majority are allocated from that node.

When zone_reclaim_mode is enabled, it means the VM should make every
attempt to allocate locally to prevent NUMA performance degradation.  In
this case, we do not want to collapse hugepages to remote nodes that
would suffer from increased access latency.  Thus, when
zone_reclaim_mode is enabled, only allow collapsing to nodes with
RECLAIM_DISTANCE or less.

There is no functional change for systems that disable
zone_reclaim_mode.

Signed-off-by: David Rientjes <rientjes@google.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Mel Gorman <mgorman@suse.de>
Cc: Rik van Riel <riel@redhat.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Bob Liu <bob.liu@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/shmem.c: remove the unused gfp arg to shmem_add_to_page_cache()

The gfp arg is not used in shmem_add_to_page_cache.  Remove this unused
arg.

Signed-off-by: Wang Sheng-Hui <shhuiw@gmail.com>
Signed-off-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: describe mmap_sem rules for __lock_page_or_retry() and callers

Add a comment describing the circumstances in which
__lock_page_or_retry() will or will not release the mmap_sem when
returning 0.

Add comments to lock_page_or_retry()'s callers (filemap_fault(),
do_swap_page()) noting the impact on VM_FAULT_RETRY returns.

Add comments on up the call tree, particularly replacing the false "We
return with mmap_sem still held" comments.

Signed-off-by: Paul Cassella <cassella@cray.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: page_alloc: reduce cost of the fair zone allocation policy

The fair zone allocation policy round-robins allocations between zones
within a node to avoid age inversion problems during reclaim.  If the
first allocation fails, the batch counts are reset and a second attempt
made before entering the slow path.

One assumption made with this scheme is that batches expire at roughly
the same time and the resets each time are justified.  This assumption
does not hold when zones reach their low watermark as the batches will
be consumed at uneven rates.  Allocation failure due to watermark
depletion result in additional zonelist scans for the reset and another
watermark check before hitting the slowpath.

On UMA, the benefit is negligible -- around 0.25%.  On 4-socket NUMA
machine it's variable due to the variability of measuring overhead with
the vmstat changes.  The system CPU overhead comparison looks like

          3.16.0-rc3  3.16.0-rc3  3.16.0-rc3
             vanilla   vmstat-v5 lowercost-v5
User          746.94      774.56      802.00
System      65336.22    32847.27    40852.33
Elapsed     27553.52    27415.04    27368.46

However it is worth noting that the overall benchmark still completed
faster and intuitively it makes sense to take as few passes as possible
through the zonelists.

Signed-off-by: Mel Gorman <mgorman@suse.de>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: page_alloc: abort fair zone allocation policy when remotes nodes are encountered

The purpose of numa_zonelist_order=zone is to preserve lower zones for
use with 32-bit devices.  If locality is preferred then the
numa_zonelist_order=node policy should be used.

Unfortunately, the fair zone allocation policy overrides this by
skipping zones on remote nodes until the lower one is found.  While this
makes sense from a page aging and performance perspective, it breaks the
expected zonelist policy.  This patch restores the expected behaviour
for zone-list ordering.

Signed-off-by: Mel Gorman <mgorman@suse.de>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: vmscan: only update per-cpu thresholds for online CPU

When kswapd is awake reclaiming, the per-cpu stat thresholds are lowered
to get more accurate counts to avoid breaching watermarks.  This
threshold update iterates over all possible CPUs which is unnecessary.
Only online CPUs need to be updated.  If a new CPU is onlined,
refresh_zone_stat_thresholds() will set the thresholds correctly.

Signed-off-by: Mel Gorman <mgorman@suse.de>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: move zone->pages_scanned into a vmstat counter

zone->pages_scanned is a write-intensive cache line during page reclaim
and it's also updated during page free.  Move the counter into vmstat to
take advantage of the per-cpu updates and do not update it in the free
paths unless necessary.

On a small UMA machine running tiobench the difference is marginal.  On
a 4-node machine the overhead is more noticable.  Note that automatic
NUMA balancing was disabled for this test as otherwise the system CPU
overhead is unpredictable.

          3.16.0-rc3  3.16.0-rc3  3.16.0-rc3
             vanillarearrange-v5   vmstat-v5
User          746.94      759.78      774.56
System      65336.22    58350.98    32847.27
Elapsed     27553.52    27282.02    27415.04

Note that the overhead reduction will vary depending on where exactly
pages are allocated and freed.

Signed-off-by: Mel Gorman <mgorman@suse.de>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: rearrange zone fields into read-only, page alloc, statistics and page reclaim lines

The arrangement of struct zone has changed over time and now it has
reached the point where there is some inappropriate sharing going on.
On x86-64 for example

o The zone->node field is shared with the zone lock and zone->node is
  accessed frequently from the page allocator due to the fair zone
  allocation policy.

o span_seqlock is almost never used by shares a line with free_area

o Some zone statistics share a cache line with the LRU lock so
  reclaim-intensive and allocator-intensive workloads can bounce the cache
  line on a stat update

This patch rearranges struct zone to put read-only and read-mostly
fields together and then splits the page allocator intensive fields, the
zone statistics and the page reclaim intensive fields into their own
cache lines.  Note that the type of lowmem_reserve changes due to the
watermark calculations being signed and avoiding a signed/unsigned
conversion there.

On the test configuration I used the overall size of struct zone shrunk
by one cache line.  On smaller machines, this is not likely to be
noticable.  However, on a 4-node NUMA machine running tiobench the
system CPU overhead is reduced by this patch.

          3.16.0-rc3  3.16.0-rc3
             vanillarearrange-v5r9
User          746.94      759.78
System      65336.22    58350.98
Elapsed     27553.52    27282.02

Signed-off-by: Mel Gorman <mgorman@suse.de>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: pagemap: avoid unnecessary overhead when tracepoints are deactivated

This was formerly the series "Improve sequential read throughput" which
noted some major differences in performance of tiobench since 3.0.
While there are a number of factors, two that dominated were the
introduction of the fair zone allocation policy and changes to CFQ.

The behaviour of fair zone allocation policy makes more sense than
tiobench as a benchmark and CFQ defaults were not changed due to
insufficient benchmarking.

This series is what's left.  It's one functional fix to the fair zone
allocation policy when used on NUMA machines and a reduction of overhead
in general.  tiobench was used for the comparison despite its flaws as
an IO benchmark as in this case we are primarily interested in the
overhead of page allocator and page reclaim activity.

On UMA, it makes little difference to overhead

          3.16.0-rc3   3.16.0-rc3
             vanilla lowercost-v5
User          383.61      386.77
System        403.83      401.74
Elapsed      5411.50     5413.11

On a 4-socket NUMA machine it's a bit more noticable

          3.16.0-rc3   3.16.0-rc3
             vanilla lowercost-v5
User          746.94      802.00
System      65336.22    40852.33
Elapsed     27553.52    27368.46

This patch (of 6):

The LRU insertion and activate tracepoints take PFN as a parameter
forcing the overhead to the caller.  Move the overhead to the tracepoint
fast-assign method to ensure the cost is only incurred when the
tracepoint is active.

Signed-off-by: Mel Gorman <mgorman@suse.de>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: trace-vmscan-postprocess.pl: report the number of file/anon pages respectively

Until now, the reporting from trace-vmscan-postprocess.pl is not very
useful because we cannot directly use this script for checking the
file/anon ratio of scanning.  This patch aims to report respectively the
number of file/anon pages which were scanned/reclaimed by kswapd or
direct-reclaim.  Sample output is usually something like the following.

Summary
Direct reclaims:                          8823
Direct reclaim pages scanned:             2438797
Direct reclaim file pages scanned:        1315200
Direct reclaim anon pages scanned:        1123597
Direct reclaim pages reclaimed:           446139
Direct reclaim file pages reclaimed:      378668
Direct reclaim anon pages reclaimed:      67471
Direct reclaim write file sync I/O:       0
Direct reclaim write anon sync I/O:       0
Direct reclaim write file async I/O:      0
Direct reclaim write anon async I/O:      4240
Wake kswapd requests:                     122310
Time stalled direct reclaim:              13.78 seconds

Kswapd wakeups:                           25817
Kswapd pages scanned:                     170779115
Kswapd file pages scanned:                162725123
Kswapd anon pages scanned:                8053992
Kswapd pages reclaimed:                   129065738
Kswapd file pages reclaimed:              128500930
Kswapd anon pages reclaimed:              564808
Kswapd reclaim write file sync I/O:       0
Kswapd reclaim write anon sync I/O:       0
Kswapd reclaim write file async I/O:      36
Kswapd reclaim write anon async I/O:      730730
Time kswapd awake:                        1015.50 seconds

Signed-off-by: Chen Yucong <slaoub@gmail.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: update the description for vm_total_pages

vm_total_pages is calculated by nr_free_pagecache_pages(), which counts
the number of pages which are beyond the high watermark within all
zones.  So vm_total_pages is not equal to total number of pages which
the VM controls.

Signed-off-by: Wang Sheng-Hui <shhuiw@gmail.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Rik van Riel <riel@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/memory.c: don't forget to set softdirty on file mapped fault

Otherwise we may not notice that pte was softdirty because
pte_mksoft_dirty helper _returns_ new pte but doesn't modify the
argument.

In case if page fault happend on dirty filemapping the newly created pte
may loose softdirty bit thus if a userspace program is tracking memory
changes with help of a memory tracker (CONFIG_MEM_SOFT_DIRTY) it might
miss modification of a memory page (which in worts case may lead to data
inconsistency).

Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Acked-by: Pavel Emelyanov <xemul@parallels.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

drivers/firmware/memmap.c: don't allocate firmware_map_entry of same memory range

When limiting memory by mem= and ACPI DSDT table has PNP0C80,
firmware_map_entrys of same memory range are allocated and memmap X
sysfses which have same memory range are created as follows:

  # cat /sys/firmware/memmap/0/*
  0x407ffffffff
  0x40000000000
  System RAM
  # cat /sys/firmware/memmap/33/*
  0x407ffffffff
  0x40000000000
  System RAM
  # cat /sys/firmware/memmap/35/*
  0x407ffffffff
  0x40000000000
  System RAM

In this case, when hot-removing memory, kernel panic occurs, showing
following call trace:

  BUG: unable to handle kernel paging request at 00000001003e000b
  IP: sysfs_open_file+0x46/0x2b0
  PGD 203a89fe067 PUD 0
  Oops: 0000 [#1] SMP
  ...
  Call Trace:
    do_dentry_open+0x1ef/0x2a0
    finish_open+0x31/0x40
    do_last+0x57c/0x1220
    path_openat+0xc2/0x4c0
    do_filp_open+0x4b/0xb0
    do_sys_open+0xf3/0x1f0
    SyS_open+0x1e/0x20
    system_call_fastpath+0x16/0x1b

The problem occurs as follows:

When calling e820_reserve_resources(), firmware_map_entrys of all e820
memory map are allocated.  And all firmware_map_entrys is added
map_entries list as follows:

map_entries
 -> +--- entry A --------+ -> ...
    | start 0x407ffffffff|
    | end   0x40000000000|
    | type  System RAM   |
    +--------------------+

After that, if ACPI DSDT table has PNP0C80 and the memory range is
limited by mem=, the PNP0C80 is hot-added.  Then firmware_map_entry of
PNP0C80 is allocated and added map_entries list as follows:

map_entries
 -> +--- entry A --------+ -> ... -> +--- entry B --------+
    | start 0x407ffffffff|           | start 0x407ffffffff|
    | end   0x40000000000|           | end   0x40000000000|
    | type  System RAM   |           | type  System RAM   |
    +--------------------+           +--------------------+

Then memmap 0 sysfs for entry B is created.

After that, firmware_memmap_init() creates memmap sysfses of all
firmware_map_entrys in map_entries list.  As a result, memmap 33 sysfs
for entry A and memmap 35 sysfs for entry B are created.  But kobject of
entry B has been used by memmap 0 sysfs.  So when creating memmap 35
sysfs, the kobject is broken.

If hot-removing memory, memmap 0 sysfs is destroyed and kobject of
memmap 0 sysfs is freed.  But the kobject can be accessed via memmap 35
sysfs.  So when open memmap 35 sysfs, kernel panic occurs.

This patch checks whether there is firmware_map_entry of same memory
range in map_entries list and don't allocate firmware_map_entry of same
memroy range.

Signed-off-by: Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com>
Cc: Santosh Shilimkar <santosh.shilimkar@ti.com>
Cc: Toshi Kani <toshi.kani@hp.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

drivers/firmware/memmap.c: pass the correct argument to firmware_map_find_entry_bootmem()

firmware_map_add_hotplug() calls firmware_map_find_entry_bootmem() to
get free firmware_map_entry.  But end arguments is not correct.  So
firmware_map_find_entry_bootmem() cannot not find firmware_map_entry.

The patch passes the correct end argument to firmware_map_find_entry_bootmem().

Signed-off-by: Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com>
Cc: Santosh Shilimkar <santosh.shilimkar@ti.com>
Cc: Toshi Kani <toshi.kani@hp.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/vmalloc.c: clean up map_vm_area third argument

Currently map_vm_area() takes (struct page *** pages) as third argument,
and after mapping, it moves (*pages) to point to (*pages +
nr_mappped_pages).

It looks like this kind of increment is useless to its caller these
days.  The callers don't care about the increments and actually they're
trying to avoid this by passing another copy to map_vm_area().

The caller can always guarantee all the pages can be mapped into vm_area
as specified in first argument and the caller only cares about whether
map_vm_area() fails or not.

This patch cleans up the pointer movement in map_vm_area() and updates
its callers accordingly.

Signed-off-by: WANG Chao <chaowang@redhat.com>
Cc: Zhang Yanfei <zhangyanfei@cn.fujitsu.com>
Acked-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Nitin Gupta <ngupta@vflare.org>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: make copy_pte_range static again

Commit 71e3aac0724f ("thp: transparent hugepage core") adds
copy_pte_range prototype to huge_mm.h.  I'm not sure why (or if) this
function have been used outside of memory.c, but it currently isn't.
This patch makes copy_pte_range() static again.

Signed-off-by: Jerome Marchand <jmarchan@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>

mm, hugetlb: remove hugetlb_zero and hugetlb_infinity

They are unnecessary: "zero" can be used in place of "hugetlb_zero" and
passing extra2 == NULL is equivalent to infinity.

Signed-off-by: David Rientjes <rientjes@google.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Reviewed-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Reviewed-by: Luiz Capitulino <lcapitulino@redhat.com>
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>

mm, hugetlb: generalize writes to nr_hugepages

Three different interfaces alter the maximum number of hugepages for an
hstate:

 - /proc/sys/vm/nr_hugepages for global number of hugepages of the default
   hstate,

 - /sys/kernel/mm/hugepages/hugepages-X/nr_hugepages for global number of
   hugepages for a specific hstate, and

 - /sys/kernel/mm/hugepages/hugepages-X/nr_hugepages/mempolicy for number of
   hugepages for a specific hstate over the set of allowed nodes.

Generalize the code so that a single function handles all of these
writes instead of duplicating the code in two different functions.

This decreases the number of lines of code, but also reduces the size of
.text by about half a percent since set_max_huge_pages() can be inlined.

Signed-off-by: David Rientjes <rientjes@google.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Reviewed-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Reviewed-by: Luiz Capitulino <lcapitulino@redhat.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Acked-by: Davidlohr Bueso <davidlohr@hp.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hwpoison: fix race with changing page during offlining

When a hwpoison page is locked it could change state due to parallel
modifications.  The original compound page can be torn down and then
this 4k page becomes part of a differently-size compound page is is a
standalone regular page.

Check after the lock if the page is still the same compound page.

We could go back, grab the new head page and try again but it should be
quite rare, so I thought this was safest.  A retry loop would be more
difficult to test and may have more side effects.

The hwpoison code by design only tries to handle cases that are
reasonably common in workloads, as visible in page-flags.

I'm not really that concerned about handling this (likely rare case),
just not crashing on it.

Signed-off-by: Andi Kleen <ak@linux.intel.com>
Acked-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm,hugetlb: simplify error handling in hugetlb_cow()

When returning from hugetlb_cow(), we always (1) put back the refcount
for each referenced page -- always 'old', and 'new' if allocation was
successful.  And (2) retake the page table lock right before returning,
as the callers expects.  This logic can be simplified and encapsulated,
as proposed in this patch.  In addition to cleaner code, we also shave a
few bytes off the instruction text:

   text    data     bss     dec     hex filename
  28399     462   41328   70189   1122d mm/hugetlb.o-baseline
  28367     462   41328   70157   1120d mm/hugetlb.o-patched

Passes libhugetlbfs testcases.

Signed-off-by: Davidlohr Bueso <davidlohr@hp.com>
Cc: Aswin Chandramouleeswaran <aswin@hp.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>

mm,hugetlb: make unmap_ref_private() return void

This function always returns 1, thus no need to check return value in
hugetlb_cow().  By doing so, we can get rid of the unnecessary WARN_ON
call.  While this logic perhaps existed as a way of identifying future
unmap_ref_private() mishandling, reality is it serves no apparent
purpose.

Signed-off-by: Davidlohr Bueso <davidlohr@hp.com>
Cc: Aswin Chandramouleeswaran <aswin@hp.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>

mm: replace init_page_accessed by __SetPageReferenced

Do we really need an exported alias for __SetPageReferenced()? Its
callers better know what they're doing, in which case the page would not
be already marked referenced.  Kill init_page_accessed(), just
__SetPageReferenced() inline.

Signed-off-by: Hugh Dickins <hughd@google.com>
Acked-by: Mel Gorman <mgorman@suse.de>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Prabhakar Lad <prabhakar.csengg@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/hwpoison-inject.c: remove unnecessary null test before debugfs_remove_recursive

Fix checkpatch warning:
  "WARNING: debugfs_remove_recursive(NULL) is safe this check is probably not required"

Signed-off-by: Fabian Frederick <fabf@skynet.be>
Acked-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: export NR_SHMEM via sysinfo(2) / si_meminfo() interfaces

Historically, we exported shared pages to userspace via sysinfo(2)
sharedram and /proc/meminfo's "MemShared" fields.  With the advent of
tmpfs, from kernel v2.4 onward, that old way for accounting shared mem
was deemed inaccurate and we started to export a hard-coded 0 for
sysinfo.sharedram.  Later on, during the 2.6 timeframe, "MemShared" got
re-introduced to /proc/meminfo re-branded as "Shmem", but we're still
reporting sysinfo.sharedmem as that old hard-coded zero, which makes the
"shared memory" report inconsistent across interfaces.

This patch leverages the addition of explicit accounting for pages used
by shmem/tmpfs -- "4b02108 mm: oom analysis: add shmem vmstat" -- in
order to make the users of sysinfo(2) and si_meminfo*() friends aware of
that vmstat entry and make them report it consistently across the
interfaces, as well to make sysinfo(2) returned data consistent with our
current API documentation states.

Signed-off-by: Rafael Aquini <aquini@redhat.com>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Hugh Dickins <hughd@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: catch memory commitment underflow

Print a warning (if CONFIG_DEBUG_VM=y) when memory commitment becomes
too negative.

This shouldn't happen any more - the previous two patches fixed the
committed_as underflow issues.

[akpm@linux-foundation.org: use VM_WARN_ONCE, per Dave]
Signed-off-by: Konstantin Khlebnikov <koct9i@gmail.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

shmem: update memory reservation on truncate

A shared anonymous mapping created without MAP_NORESERVE holds memory
reservation for whole range of shmem segment.  Usually there is no way
to change its size, but /proc/<pid>/map_files/...  (available if
CONFIG_CHECKPOINT_RESTORE=y) allows that.

This patch adjusts the memory reservation in shmem_setattr().

Signed-off-by: Konstantin Khlebnikov <koct9i@gmail.com>
Acked-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

shmem: fix double uncharge in __shmem_file_setup()

If __shmem_file_setup() fails on struct file allocation it uncharges
memory commitment twice: first by shmem_unacct_size() and second time
implicitly in shmem_evict_inode() when it kills the newly created inode.

This patch removes shmem_unacct_size() from error path if the inode was
already there.

Signed-off-by: Konstantin Khlebnikov <koct9i@gmail.com>
Acked-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

include/linux/mmdebug.h: add VM_WARN_ONCE()

It was missing...

Cc: Konstantin Khlebnikov <koct9i@gmail.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Hugh Dickins <hughd@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, vmalloc: constify allocation mask

tmp_mask in the __vmalloc_area_node() iteration never changes so it can
be moved into function scope and marked with const.  This causes the
movl and orl to only be done once per call rather than area->nr_pages
times.

nested_gfp can also be marked const.

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>

mm/vmalloc.c: add a schedule point to vmalloc()

It is not uncommon on busy servers to get stuck hundred of ms in
vmalloc() calls (like file descriptor expansions).

Add a cond_resched() to __vmalloc_area_node() to be gentle to
other tasks.

[akpm@linux-foundation.org: only do it for __GFP_WAIT, per David]
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Hugh Dickins <hughd@google.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>

mm: update the description for madvise_remove

Currently, we have more filesystems supporting fallocate, e.g
ext4/btrfs.  Remove the outdated comment for madvise_remove.

Signed-off-by: Wang Sheng-Hui <shhuiw@gmail.com>
Reviewed-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm tracing: tell mm_migrate_pages event about numa_misplaced

The mm_migrate_pages trace event reports a reason for the migration,
typically as a symbolic string.  The exception is the reason
MR_NUMA_MISPLACED for which it just displays the numeric value:
mm_migrate_pages: nr_succeeded=1 nr_failed=0 mode=MIGRATE_ASYNC
reason=0x5

This patch makes the output consistent by introducing a string value for
MR_NUMA_MISPLACED.  The event is then reported as: mm_migrate_pages:
nr_succeeded=1 nr_failed=0 mode=MIGRATE_ASYNC reason=numa_misplaced

Signed-off-by: Max Asbock <masbock@linux.vnet.ibm.com>
Acked-by: Steven Rostedt <rostedt@goodmis.org>
Cc: Ingo Molnar <mingo@redhat.com>
Acked-by: Mel Gorman <mgorman@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: vmscan: clean up struct scan_control

Reorder the members by input and output, then turn the individual
integers for may_writepage, may_unmap, may_swap, compaction_ready,
hibernation_mode into bit fields to save stack space:

  +72/-296 -224
  kswapd                                       104     176     +72
  try_to_free_pages                             80      56     -24
  try_to_free_mem_cgroup_pages                  80      56     -24
  shrink_all_memory                             88      64     -24
  reclaim_clean_pages_from_list                168     144     -24
  mem_cgroup_shrink_node_zone                  104      80     -24
  __zone_reclaim                               176     152     -24
  balance_pgdat                                152       -    -152

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Suggested-by: Mel Gorman <mgorman@suse.de>
Acked-by: Mel Gorman <mgorman@suse.de>
Acked-by: Michal Hocko <mhocko@suse.cz>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Acked-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: vmscan: move swappiness out of scan_control

Swappiness is determined for each scanned memcg individually in
shrink_zone() and is not a parameter that applies throughout the reclaim
scan.  Move it out of struct scan_control to prevent accidental use of a
stale value.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Rik van Riel <riel@redhat.com>
Acked-by: Michal Hocko <mhocko@suse.cz>
Cc: Vlastimil Babka <vbabka@suse.cz>
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>

mm: vmscan: remove all_unreclaimable()

Direct reclaim currently calls shrink_zones() to reclaim all members of
a zonelist, and if that wasn't successful it does another pass through
the same zonelist to check overall reclaimability.

Just check reclaimability in shrink_zones() directly and propagate the
result through the return value.  Then remove all_unreclaimable().

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Rik van Riel <riel@redhat.com>
Acked-by: Michal Hocko <mhocko@suse.cz>
Cc: Vlastimil Babka <vbabka@suse.cz>
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>

mm: vmscan: rework compaction-ready signaling in direct reclaim

Page reclaim for a higher-order page runs until compaction is ready,
then aborts and signals this situation through the return value of
shrink_zones().  This is an oddly specific signal to encode in the
return value of shrink_zones(), though, and can be quite confusing.

Introduce sc->compaction_ready and signal the compactability of the
zones out-of-band to free up the return value of shrink_zones() for
actual zone reclaimability.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Rik van Riel <riel@redhat.com>
Cc: Michal Hocko <mhocko@suse.cz>
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>

mm: vmscan: remove remains of kswapd-managed zone->all_unreclaimable

shrink_zones() has a special branch to skip the all_unreclaimable()
check during hibernation, because a frozen kswapd can't mark a zone
unreclaimable.

But ever since commit 6e543d5780e3 ("mm: vmscan: fix
do_try_to_free_pages() livelock"), determining a zone to be
unreclaimable is done by directly looking at its scan history and no
longer relies on kswapd setting the per-zone flag.

Remove this branch and let shrink_zones() check the reclaimability of
the target zones regardless of hibernation state.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Mel Gorman <mgorman@suse.de>
Cc: Rik van Riel <riel@redhat.com>
Acked-by: Michal Hocko <mhocko@suse.cz>
Cc: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Minchan Kim <minchan@kernel.org>
Cc: KOSAKI Motohiro <Kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mem-hotplug: improve zone_movable_is_highmem logic

In original code, zone_movable_is_highmem() assumes ZONE_MOVABLE not
highmem if CONFIG_HAVE_MEMBLOCK_NODE_MAP is not set.  In online_pages,
it extracts pages from the previous zone before ZONE_MOVABLE.  Which is
logically inconsistent:

If HAVE_MEMBLOCK_NODE_MAP is turned off but HIGHMEM is on,
zone_movable_is_highmem() makes movable zone not highmem, but
online_pages() extracts pages from ZONE_HIGHMEM.

This inconsistency doesn't cause real problem currently, because all
architectures support online_pages also have HAVE_MEMBLOCK_NODE_MAP.
However, fixing it makes code clear, and also helps futher coding.

Signed-off-by: Wang Nan <wangnan0@huawei.com>
Cc: Zhang Zhen <zhangzhen@huawei.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Jiang Liu <liuj97@gmail.com>
Cc: Li Zefan <lizefan@huawei.com>
Cc: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/mem-hotplug: replace simple_strtoull() with kstrtoull()

Use the newer and more pleasant kstrtoull() to replace
simple_strtoull(), because simple_strtoull() is marked for obsoletion.

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>

mm: memcontrol: do not acquire page_cgroup lock for kmem pages

Kmem page charging and uncharging is serialized by means of exclusive
access to the page.  Do not take the page_cgroup lock and don't set
pc->flags atomically.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Michal Hocko <mhocko@suse.cz>
Acked-by: Vladimir Davydov <vdavydov@parallels.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Tejun Heo <tj@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcontrol: remove ordering between pc->mem_cgroup and PageCgroupUsed

There is a write barrier between setting pc->mem_cgroup and
PageCgroupUsed, which was added to allow LRU operations to lookup the
memcg LRU list of a page without acquiring the page_cgroup lock.

But ever since commit 38c5d72f3ebe ("memcg: simplify LRU handling by new
rule"), pages are ensured to be off-LRU while charging, so nobody else
is changing LRU state while pc->mem_cgroup is being written, and there
are no read barriers anymore.

Remove the unnecessary write barrier.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Michal Hocko <mhocko@suse.cz>
Cc: Hugh Dickins <hughd@google.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Vladimir Davydov <vdavydov@parallels.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcontrol: use root_mem_cgroup res_counter

Due to an old optimization to keep expensive res_counter changes at a
minimum, the root_mem_cgroup res_counter is never charged; there is no
limit at that level anyway, and any statistics can be generated on
demand by summing up the counters of all other cgroups.

However, with per-cpu charge caches, res_counter operations do not even
show up in profiles anymore, so this optimization is no longer
necessary.

Remove it to simplify the code.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Vladimir Davydov <vdavydov@parallels.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcontrol: catch root bypass in move precharge

When mem_cgroup_try_charge() returns -EINTR, it bypassed the charge to
the root memcg.  But move precharging does not catch this and treats
this case as if no charge had happened, thus leaking a charge against
root.  Because of an old optimization, the root memcg's res_counter is
not actually charged right now, but it's still an imbalance and
subsequent patches will charge the root memcg again.

Catch those bypasses to the root memcg and properly cancel them before
giving up the move.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Michal Hocko <mhocko@suse.cz>
Cc: Hugh Dickins <hughd@google.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Vladimir Davydov <vdavydov@parallels.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcontrol: simplify move precharge function

The move precharge function does some baroque things: it tries raw
res_counter charging of the entire amount first, and then falls back to
a loop of one-by-one charges, with checks for pending signals and
cond_resched() batching.

Just use mem_cgroup_try_charge() without __GFP_WAIT for the first bulk
charge attempt.  In the one-by-one loop, remove the signal check (this
is already checked in try_charge), and simply call cond_resched() after
every charge - it's not that expensive.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Michal Hocko <mhocko@suse.cz>
Cc: Hugh Dickins <hughd@google.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Vladimir Davydov <vdavydov@parallels.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcontrol: remove explicit OOM parameter in charge path

For the page allocator, __GFP_NORETRY implies that no OOM should be
triggered, whereas memcg has an explicit parameter to disable OOM.

The only callsites that want OOM disabled are THP charges and charge
moving.  THP already uses __GFP_NORETRY and charge moving can use it as
well - one full reclaim cycle should be plenty.  Switch it over, then
remove the OOM parameter.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Michal Hocko <mhocko@suse.cz>
Cc: Hugh Dickins <hughd@google.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Vladimir Davydov <vdavydov@parallels.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcontrol: retry reclaim for oom-disabled and __GFP_NOFAIL charges

There is no reason why oom-disabled and __GFP_NOFAIL charges should try
to reclaim only once when every other charge tries several times before
giving up.  Make them all retry the same number of times.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Michal Hocko <mhocko@suse.cz>
Cc: Hugh Dickins <hughd@google.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Vladimir Davydov <vdavydov@parallels.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: huge_memory: use GFP_TRANSHUGE when charging huge pages

Transparent huge page charges prefer falling back to regular pages
rather than spending a lot of time in direct reclaim.

Desired reclaim behavior is usually declared in the gfp mask, but THP
charges use GFP_KERNEL and then rely on the fact that OOM is disabled
for THP charges, and that OOM-disabled charges don't retry reclaim.
Needless to say, this is anything but obvious and quite error prone.

Convert THP charges to use GFP_TRANSHUGE instead, which implies
__GFP_NORETRY, to indicate the low-latency requirement.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Michal Hocko <mhocko@suse.cz>
Cc: Hugh Dickins <hughd@google.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Vladimir Davydov <vdavydov@parallels.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcontrol: reclaim at least once for __GFP_NORETRY

Currently, __GFP_NORETRY tries charging once and gives up before even
trying to reclaim.  Bring the behavior on par with the page allocator
and reclaim at least once before giving up.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Michal Hocko <mhocko@suse.cz>
Cc: Hugh Dickins <hughd@google.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Vladimir Davydov <vdavydov@parallels.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcontrol: rearrange charging fast path

The charging path currently starts out with OOM condition checks when
OOM is the rarest possible case.

Rearrange this code to run OOM/task dying checks only after trying the
percpu charge and the res_counter charge and bail out before entering
reclaim.  Attempting a charge does not hurt an (oom-)killed task as much
as every charge attempt having to check OOM conditions.  Also, only
check __GFP_NOFAIL when the charge would actually fail.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Michal Hocko <mhocko@suse.cz>
Cc: Hugh Dickins <hughd@google.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Vladimir Davydov <vdavydov@parallels.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: memcontrol: fold mem_cgroup_do_charge()

These patches rework memcg charge lifetime to integrate more naturally
with the lifetime of user pages.  This drastically simplifies the code
and reduces charging and uncharging overhead.  The most expensive part
of charging and uncharging is the page_cgroup bit spinlock, which is
removed entirely after this series.

Here are the top-10 profile entries of a stress test that reads a 128G
sparse file on a freshly booted box, without even a dedicated cgroup
(i.e. executing in the root memcg).  Before:

    15.36%              cat  [kernel.kallsyms]   [k] copy_user_generic_string
    13.31%              cat  [kernel.kallsyms]   [k] memset
    11.48%              cat  [kernel.kallsyms]   [k] do_mpage_readpage
     4.23%              cat  [kernel.kallsyms]   [k] get_page_from_freelist
     2.38%              cat  [kernel.kallsyms]   [k] put_page
     2.32%              cat  [kernel.kallsyms]   [k] __mem_cgroup_commit_charge
     2.18%          kswapd0  [kernel.kallsyms]   [k] __mem_cgroup_uncharge_common
     1.92%          kswapd0  [kernel.kallsyms]   [k] shrink_page_list
     1.86%              cat  [kernel.kallsyms]   [k] __radix_tree_lookup
     1.62%              cat  [kernel.kallsyms]   [k] __pagevec_lru_add_fn

After:

    15.67%           cat  [kernel.kallsyms]   [k] copy_user_generic_string
    13.48%           cat  [kernel.kallsyms]   [k] memset
    11.42%           cat  [kernel.kallsyms]   [k] do_mpage_readpage
     3.98%           cat  [kernel.kallsyms]   [k] get_page_from_freelist
     2.46%           cat  [kernel.kallsyms]   [k] put_page
     2.13%       kswapd0  [kernel.kallsyms]   [k] shrink_page_list
     1.88%           cat  [kernel.kallsyms]   [k] __radix_tree_lookup
     1.67%           cat  [kernel.kallsyms]   [k] __pagevec_lru_add_fn
     1.39%       kswapd0  [kernel.kallsyms]   [k] free_pcppages_bulk
     1.30%           cat  [kernel.kallsyms]   [k] kfree

As you can see, the memcg footprint has shrunk quite a bit.

   text    data     bss     dec     hex filename
  37970    9892     400   48262    bc86 mm/memcontrol.o.old
  35239    9892     400   45531    b1db mm/memcontrol.o

This patch (of 13):

This function was split out because mem_cgroup_try_charge() got too big.
But having essentially one sequence of operations arbitrarily split in
half is not good for reworking the code.  Fold it back in.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Michal Hocko <mhocko@suse.cz>
Cc: Hugh Dickins <hughd@google.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Vladimir Davydov <vdavydov@parallels.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: page-flags: clean up the page flag test, set, clear macros

- PAGEFLAG_FALSE only defines TEST, make it define SET and CLEAR as
  well, analogous to PAGEFLAG.

- Define TESTSETFLAG_FALSE, analogous to TESTSETFLAG.

- Define TESTSCFLAG_FALSE, analogous to TESTSCFLAG

- Make PG_mlocked accessors the same on both MMU and !MMU setups

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, thp: replace smp_mb after atomic_add by smp_mb__after_atomic

In some architectures like x86, atomic_add() is a full memory barrier.
In that case, an additional smp_mb() is just a waste of time.  This
patch replaces that smp_mb() by smp_mb__after_atomic() which will avoid
the redundant memory barrier in some architectures.

With a 3.16-rc1 based kernel, this patch reduced the execution time of
breaking 1000 transparent huge pages from 38,245us to 30,964us.  A
reduction of 19% which is quite sizeable.  It also reduces the %cpu time
of the __split_huge_page_refcount function in the perf profile from
2.18% to 1.15%.

Signed-off-by: Waiman Long <Waiman.Long@hp.com>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Rik van Riel <riel@redhat.com>
Cc: Scott J Norton <scott.norton@hp.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, thp: move invariant bug check out of loop in __split_huge_page_map

In __split_huge_page_map(), the check for page_mapcount(page) is
invariant within the for loop.  Because of the fact that the macro is
implemented using atomic_read(), the redundant check cannot be optimized
away by the compiler leading to unnecessary read to the page structure.

This patch moves the invariant bug check out of the loop so that it will
be done only once.  On a 3.16-rc1 based kernel, the execution time of a
microbenchmark that broke up 1000 transparent huge pages using munmap()
had an execution time of 38,245us and 38,548us with and without the
patch respectively.  The performance gain is about 1%.

Signed-off-by: Waiman Long <Waiman.Long@hp.com>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Rik van Riel <riel@redhat.com>
Cc: Scott J Norton <scott.norton@hp.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, CMA: clean-up log message

We don't need explicit 'CMA:' prefix, since we already define prefix
'cma:' in pr_fmt.  So remove it.

Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Acked-by: Michal Nazarewicz <mina86@mina86.com>
Reviewed-by: Zhang Yanfei <zhangyanfei@cn.fujitsu.com>
Cc: "Aneesh Kumar K.V" <aneesh.kumar@linux.vnet.ibm.com>
Cc: Alexander Graf <agraf@suse.de>
Cc: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Cc: Gleb Natapov <gleb@kernel.org>
Acked-by: Marek Szyprowski <m.szyprowski@samsung.com>
Tested-by: Marek Szyprowski <m.szyprowski@samsung.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, CMA: change cma_declare_contiguous() to obey coding convention

Conventionally, we put output param to the end of param list and put the
'base' ahead of 'size', but cma_declare_contiguous() doesn't look like
that, so change it.

Additionally, move down cma_areas reference code to the position where
it is really needed.

Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Acked-by: Michal Nazarewicz <mina86@mina86.com>
Reviewed-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Cc: Alexander Graf <agraf@suse.de>
Cc: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Cc: Gleb Natapov <gleb@kernel.org>
Acked-by: Marek Szyprowski <m.szyprowski@samsung.com>
Tested-by: Marek Szyprowski <m.szyprowski@samsung.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Zhang Yanfei <zhangyanfei@cn.fujitsu.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>