cris: cpuinfo_op should depend on CONFIG_PROC_FS

Fix breakage which will be exposed by the patch "kconfig: make allnoconfig
disable options behind EMBEDDED and EXPERT".

Now allnoconfig started disabling CONFIG_PROC_FS:

    arch/cris/kernel/built-in.o:(.rodata+0xc): undefined reference to `show_cpuinfo'
    make: *** [vmlinux] Error 1

Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Cc: Mikael Starvik <starvik@axis.com>
Cc: Jesper Nilsson <jesper.nilsson@axis.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

cris: make ETRAX_ARCH_V10 select TTY for use in debugport

Fix breakage which will be exposed by the patch "kconfig: make allnoconfig
disable options behind EMBEDDED and EXPERT".

arch/cris/arch-v10/kernel/debugport.c, compiled in unconditionally with
ETRAX_ARCH_V10, requires TTY, so select TTY to avoid a build failure.

Signed-off-by: Josh Triplett <josh@joshtriplett.org>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Cc: Mikael Starvik <starvik@axis.com>
Cc: Jesper Nilsson <jesper.nilsson@axis.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

drivers/misc/sgi-gru/grukdump.c: cleanup gru_dump_context() a little

"ret" is zero here so we can remove the "!ret" part of the condition.
"uhdr" is alread a __user pointer so we can remove the cast.

Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Acked-by: Dimitri Sivanich <sivanich@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

kernel/panic.c: display reason at end + pr_emerg

Currently, booting without initrd specified on 80x25 screen gives a call
trace followed by atkbd : Spurious ACK.  Original message ("VFS: Unable
to mount root fs") is not available.  Of course this could happen in
other situations...

This patch displays panic reason after call trace which could help lot
of people even if it's not the very last line on screen.

Also, convert all panic.c printk(KERN_EMERG to pr_emerg(

[akpm@linux-foundation.org: missed a couple of pr_ conversions]
Signed-off-by: Fabian Frederick <fabf@skynet.be>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fs/bfs/inode.c: add __init to init_inodecache()

init_inodecache is only called by __init init_bfs_fs

Signed-off-by: Fabian Frederick <fabf@skynet.be>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

affs: add mount option to avoid filename truncates

Normal behavior for filenames exceeding specific filesystem limits is to
refuse operation.

AFFS standard name length being only 30 characters against 255 for usual
Linux filesystems, original implementation does filename truncate by
default with a define value AFFS_NO_TRUNCATE which can be enabled but
needs module compilation.

This patch adds 'nofilenametruncate' mount option so that user can
easily activate that feature and avoid a lot of problems (eg overwrite
files ...)

Signed-off-by: Fabian Frederick <fabf@skynet.be>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fs/affs/dir.c: unlock/brelse dir on failure + code clean-up

Commit 0edf977d2ae3 ("[readdir] convert affs") returns directly -EIO
without unlocking dir inode and releasing dir bh when second affs_bread
sequence fails.  This patch restores initial behaviour.  It also fixes
pr_debug and affs_error to fit in 80 columns + removes reference to
filldir (replaced by dir_emit in the commit above).

Signed-off-by: Fabian Frederick <fabf@skynet.be>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

affs: add __init to init_inodecache ()

init_inodecache is only called by __init init_affs_fs

Signed-off-by: Fabian Frederick <fabf@skynet.be>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fs/adfs/super.c: add __init to init_inodecache()

init_inodecache is only called by __init init_adfs_fs.

Signed-off-by: Fabian Frederick <fabf@skynet.be>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hung_task: check the value of "sysctl_hung_task_timeout_sec"

As sysctl_hung_task_timeout_sec is unsigned long, when this value is
larger then LONG_MAX/HZ, the function schedule_timeout_interruptible in
watchdog will return immediately without sleep and with print :

  schedule_timeout: wrong timeout value ffffffffffffff83

and then the funtion watchdog will call schedule_timeout_interruptible
again and again.  The screen will be filled with

"schedule_timeout: wrong timeout value ffffffffffffff83"

This patch does some check and correction in sysctl, to let the function
schedule_timeout_interruptible allways get the valid parameter.

Signed-off-by: Liu Hua <sdu.liu@huawei.com>
Tested-by: Satoru Takeuchi <satoru.takeuchi@gmail.com>
Cc: <stable@vger.kernel.org> [3.4+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

rapidio: rework device hierarchy and introduce mport class of devices

This patch removes an artificial RapidIO bus root device and establishes
actual device hierarchy by providing reference to real parent devices.
It also introduces device class for RapidIO controller devices (on-chip
or an eternal bridge, known as "mport").

Existing implementation was sufficient for SoC-based platforms that have
a single RapidIO controller.  With introduction of devices using
multiple RapidIO controllers and PCIe-to-RapidIO bridges the old scheme
is very limiting or does not work at all.  The implemented changes allow
to properly reference platform's local RapidIO mport devices and provide
device details needed for upper layers.

This change to RapidIO device hierarchy does not break any known
existing kernel or user space interfaces.

Signed-off-by: Alexandre Bounine <alexandre.bounine@idt.com>
Cc: Matt Porter <mporter@kernel.crashing.org>
Cc: Li Yang <leoli@freescale.com>
Cc: Kumar Gala <galak@kernel.crashing.org>
Cc: Andre van Herk <andre.van.herk@prodrive-technologies.com>
Cc: Stef van Os <stef.van.os@prodrive-technologies.com>
Cc: Jerry Jacobs <jerry.jacobs@prodrive-technologies.com>
Cc: Arno Tiemersma <arno.tiemersma@prodrive-technologies.com>
Cc: Rob Landley <rob@landley.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

drivers/rapidio/devices/tsi721_dma.c: optimize use of BDMA descriptors

Combine SG entries describing single contiguous memory block into one
Tsi721 BDMA descriptor.  This reduces number of hardware descriptors
required for large data transfers and improves performance on the PCIe
side by reducing number of descriptor fetch requests.

Signed-off-by: Alexandre Bounine <alexandre.bounine@idt.com>
Cc: Matt Porter <mporter@kernel.crashing.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

lib/idr.c: use RCU_INIT_POINTER(x, NULL)

Replace rcu_assign_pointer(x, NULL) with RCU_INIT_POINTER(x, NULL)

The rcu_assign_pointer() ensures that the initialization of a structure
is carried out before storing a pointer to that structure.  And in the
case of the NULL pointer, there is no structure to initialize.

So, rcu_assign_pointer(p, NULL) can be safely converted to
RCU_INIT_POINTER(p, NULL)

Signed-off-by: Monam Agarwal <monamagarwal123@gmail.com>
Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

idr: remove dead code

Remove no longer used deprecated code, and make local functions
static.

Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Acked-by: Jean Delvare <jdelvare@suse.de>
Acked-by: Tejun Heo <tj@kernel.org>
Cc: Jeff Layton <jlayton@redhat.com>
Cc: Philipp Reisner <philipp.reisner@linbit.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: George Spelvin <linux@horizon.com>
Cc: Randy Dunlap <rdunlap@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

vmcore: continue vmcore initialization if PT_NOTE is found empty

Currently when an empty PT_NOTE is detected, vmcore initialization
fails.  It sounds too harsh.  Because PT_NOTE could be empty, for
example, one offlined a cpu but never restarted kdump service, and after
crash, PT_NOTE program header is there but no data contains.  It's
better to warn about the empty PT_NOTE and continue to initialise
vmcore.

And ultimately the multiple PT_NOTE are merged into a single one, all
empty PT_NOTE are discarded naturally during the merge.  So empty
PT_NOTE is not visible to user space and vmcore is as good as expected.

Signed-off-by: WANG Chao <chaowang@redhat.com>
Cc: Vivek Goyal <vgoyal@redhat.com>
Cc: HATAYAMA Daisuke <d.hatayama@jp.fujitsu.com>
Cc: Greg Pearson <greg.pearson@hp.com>
Cc: Baoquan He <bhe@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

include/linux/crash_dump.h: add vmcore_cleanup() prototype

Eliminate the following warning in proc/vmcore.c:

  fs/proc/vmcore.c:1088:6: warning: no previous prototype for `vmcore_cleanup' [-Wmissing-prototypes]

[akpm@linux-foundation.org: clean up powerpc, remove unneeded EXPORT_SYMBOL]
Signed-off-by: Rashika Kheria <rashika.kheria@gmail.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

wait: WSTOPPED|WCONTINUED doesn't work if a zombie leader is traced by another process

Even if the main thread is dead the process still can stop/continue.
However, if the leader is ptraced wait_consider_task(ptrace => false)
always skips wait_task_stopped/wait_task_continued, so WSTOPPED or
WCONTINUED can never work for the natural parent in this case.

Move the "A zombie ptracee is only visible to its ptracer" check into the
"if (!delay_group_leader(p))" block.  ->notask_error is cleared by the
"fall through" code below.

This depends on the previous change, wait_task_stopped/continued must be
avoided if !delay_group_leader() and the tracer is ->real_parent.
Otherwise WSTOPPED|WEXITED could wrongly report "stopped" when the child
is already dead (single-threaded or not).  If it is traced by another task
then the "stopped" state is fine until the debugger detaches and reveals a
zombie state.

Stupid test-case:

void *tfunc(void *arg)
{
sleep(1); // wait for zombie leader
raise(SIGSTOP);
exit(0x13);
return NULL;
}

int run_child(void)
{
pthread_t thread;

if (!fork()) {
int tracee = getppid();

assert(ptrace(PTRACE_ATTACH, tracee, 0,0) == 0);
do
ptrace(PTRACE_CONT, tracee, 0,0);
while (wait(NULL) > 0);

return 0;
}

sleep(1); // wait for PTRACE_ATTACH
assert(pthread_create(&thread, NULL, tfunc, NULL) == 0);
pthread_exit(NULL);
}

int main(void)
{
int child, stat;

child = fork();
if (!child)
return run_child();

assert(child == waitpid(-1, &stat, WSTOPPED));
assert(stat == 0x137f);

kill(child, SIGCONT);

assert(child == waitpid(-1, &stat, WCONTINUED));
assert(stat == 0xffff);

assert(child == waitpid(-1, &stat, 0));
assert(stat == 0x1300);

return 0;
}

Without this patch it hangs in waitpid(WSTOPPED), wait_task_stopped() is
never called.

Note: this doesn't fix all problems with a zombie delay_group_leader(),
WCONTINUED | WEXITED check is not exactly right.  debugger can't assume it
will be notified if another thread reaps the whole thread group.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Cc: Jan Kratochvil <jan.kratochvil@redhat.com>
Cc: Lennart Poettering <lpoetter@redhat.com>
Cc: Michal Schmidt <mschmidt@redhat.com>
Cc: Roland McGrath <roland@hack.frob.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>

wait: WSTOPPED|WCONTINUED hangs if a zombie child is traced by real_parent

"A zombie is only visible to its ptracer" logic in wait_consider_task()
is very wrong. Trivial test-case:

#include <unistd.h>
#include <sys/ptrace.h>
#include <sys/wait.h>
#include <assert.h>

int main(void)
{
int child = fork();

if (!child) {
assert(ptrace(PTRACE_TRACEME, 0,0,0) == 0);
return 0x23;
}

assert(waitid(P_ALL, child, NULL, WEXITED | WNOWAIT) == 0);
assert(waitid(P_ALL, 0, NULL, WSTOPPED) == -1);
return 0;
}

it hangs in waitpid(WSTOPPED) despite the fact it has a single zombie
child.  This is because wait_consider_task(ptrace => 0) sees p->ptrace and
cleares ->notask_error assuming that the debugger should detach and notify
us.

Change wait_consider_task(ptrace => 0) to pretend that ptrace == T if the
child is traced by us.  This really simplifies the logic and allows us to
do more fixes, see the next changes.  This also hides the unwanted group
stop state automatically, we can remove another ptrace_reparented() check.

Unfortunately, this adds the following behavioural changes:

1. Before this patch wait(WEXITED | __WNOTHREAD) does not reap
   a natural child if it is traced by the caller's sub-thread.

   Hopefully nobody will ever notice this change, and I think
   that nobody should rely on this behaviour anyway.

2. SIGNAL_STOP_CONTINUED is no longer hidden from debugger if
   it is real parent.

   While this change comes as a side effect, I think it is good
   by itself. The group continued state can not be consumed by
   another process in this case, it doesn't depend on ptrace,
   it doesn't make sense to hide it from real parent.

   Perhaps we should add the thread_group_leader() check before
   wait_task_continued()? May be, but this shouldn't depend on
   ptrace_reparented().

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Cc: Jan Kratochvil <jan.kratochvil@redhat.com>
Cc: Lennart Poettering <lpoetter@redhat.com>
Cc: Michal Schmidt <mschmidt@redhat.com>
Cc: Roland McGrath <roland@hack.frob.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>

wait: swap EXIT_ZOMBIE and EXIT_DEAD to hide EXIT_TRACE from user-space

get_task_state() uses the most significant bit to report the state to
user-space, this means that EXIT_ZOMBIE->EXIT_TRACE->EXIT_DEAD transition
can be noticed via /proc as Z -> X -> Z change.  Note that this was
possible even before EXIT_TRACE was introduced.

This is not really bad but imho it make sense to hide EXIT_TRACE from
user-space completely.  So the patch simply swaps EXIT_ZOMBIE and
EXIT_DEAD, this way EXIT_TRACE will be seen as EXIT_ZOMBIE by user-space.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: Jan Kratochvil <jan.kratochvil@redhat.com>
Cc: Michal Schmidt <mschmidt@redhat.com>
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Cc: Lennart Poettering <lpoetter@redhat.com>
Cc: Roland McGrath <roland@hack.frob.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>

wait: completely ignore the EXIT_DEAD tasks

Now that EXIT_DEAD is the terminal state it doesn't make sense to call
eligible_child() or security_task_wait() if the task is really dead.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Tested-by: Michal Schmidt <mschmidt@redhat.com>
Cc: Jan Kratochvil <jan.kratochvil@redhat.com>
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Cc: Lennart Poettering <lpoetter@redhat.com>
Cc: Roland McGrath <roland@hack.frob.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>

wait: use EXIT_TRACE only if thread_group_leader(zombie)

wait_task_zombie() always uses EXIT_TRACE/ptrace_unlink() if
ptrace_reparented().  This is suboptimal and a bit confusing: we do not
need do_notify_parent(p) if !thread_group_leader(p) and in this case we
also do not need ptrace_unlink(), we can rely on ptrace_release_task().

Change wait_task_zombie() to check thread_group_leader() along with
ptrace_reparented() and simplify the final p->exit_state transition.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Tested-by: Michal Schmidt <mschmidt@redhat.com>
Cc: Jan Kratochvil <jan.kratochvil@redhat.com>
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Cc: Lennart Poettering <lpoetter@redhat.com>
Cc: Roland McGrath <roland@hack.frob.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>

wait: introduce EXIT_TRACE to avoid the racy EXIT_DEAD->EXIT_ZOMBIE transition

wait_task_zombie() first does EXIT_ZOMBIE->EXIT_DEAD transition and
drops tasklist_lock.  If this task is not the natural child and it is
traced, we change its state back to EXIT_ZOMBIE for ->real_parent.

The last transition is racy, this is even documented in 50b8d257486a
"ptrace: partially fix the do_wait(WEXITED) vs EXIT_DEAD->EXIT_ZOMBIE
race".  wait_consider_task() tries to detect this transition and clear
->notask_error but we can't rely on ptrace_reparented(), debugger can
exit and do ptrace_unlink() before its sub-thread sets EXIT_ZOMBIE.

And there is another problem which were missed before: this transition
can also race with reparent_leader() which doesn't reset >exit_signal if
EXIT_DEAD, assuming that this task must be reaped by someone else.  So
the tracee can be re-parented with ->exit_signal != SIGCHLD, and if
/sbin/init doesn't use __WALL it becomes unreapable.  This was fixed by
the previous commit, but it was the temporary hack.

1. Add the new exit_state, EXIT_TRACE. It means that the task is the
   traced zombie, debugger is going to detach and notify its natural
   parent.

   This new state is actually EXIT_ZOMBIE | EXIT_DEAD. This way we
   can avoid the changes in proc/kgdb code, get_task_state() still
   reports "X (dead)" in this case.

   Note: with or without this change userspace can see Z -> X -> Z
   transition. Not really bad, but probably makes sense to fix.

2. Change wait_task_zombie() to use EXIT_TRACE instead of EXIT_DEAD
   if we need to notify the ->real_parent.

3. Revert the previous hack in reparent_leader(), now that EXIT_DEAD
   is always the final state we can safely ignore such a task.

4. Change wait_consider_task() to check EXIT_TRACE separately and kill
   the racy and no longer needed ptrace_reparented() case.

   If ptrace == T an EXIT_TRACE thread should be simply ignored, the
   owner of this state is going to ptrace_unlink() this task. We can
   pretend that it was already removed from ->ptraced list.

   Otherwise we should skip this thread too but clear ->notask_error,
   we must be the natural parent and debugger is going to untrace and
   notify us. IOW, this doesn't differ from "EXIT_ZOMBIE && p->ptrace"
   even if the task was already untraced.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Reported-by: Jan Kratochvil <jan.kratochvil@redhat.com>
Reported-by: Michal Schmidt <mschmidt@redhat.com>
Tested-by: Michal Schmidt <mschmidt@redhat.com>
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Cc: Lennart Poettering <lpoetter@redhat.com>
Cc: Roland McGrath <roland@hack.frob.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>

wait: fix reparent_leader() vs EXIT_DEAD->EXIT_ZOMBIE race

wait_task_zombie() first does EXIT_ZOMBIE->EXIT_DEAD transition and
drops tasklist_lock.  If this task is not the natural child and it is
traced, we change its state back to EXIT_ZOMBIE for ->real_parent.

The last transition is racy, this is even documented in 50b8d257486a
"ptrace: partially fix the do_wait(WEXITED) vs EXIT_DEAD->EXIT_ZOMBIE
race".  wait_consider_task() tries to detect this transition and clear
->notask_error but we can't rely on ptrace_reparented(), debugger can
exit and do ptrace_unlink() before its sub-thread sets EXIT_ZOMBIE.

And there is another problem which were missed before: this transition
can also race with reparent_leader() which doesn't reset >exit_signal if
EXIT_DEAD, assuming that this task must be reaped by someone else.  So
the tracee can be re-parented with ->exit_signal != SIGCHLD, and if
/sbin/init doesn't use __WALL it becomes unreapable.

Change reparent_leader() to update ->exit_signal even if EXIT_DEAD.
Note: this is the simple temporary hack for -stable, it doesn't try to
solve all problems, it will be reverted by the next changes.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Reported-by: Jan Kratochvil <jan.kratochvil@redhat.com>
Reported-by: Michal Schmidt <mschmidt@redhat.com>
Tested-by: Michal Schmidt <mschmidt@redhat.com>
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Cc: Lennart Poettering <lpoetter@redhat.com>
Cc: Roland McGrath <roland@hack.frob.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

exec: kill bprm->tcomm[], simplify the "basename" logic

Starting from commit c4ad8f98bef7 ("execve: use 'struct filename *' for
executable name passing") bprm->filename can not go away after
flush_old_exec(), so we do not need to save the binary name in
bprm->tcomm[] added by 96e02d158678 ("exec: fix use-after-free bug in
setup_new_exec()").

And there was never need for filename_to_taskname-like code, we can
simply do set_task_comm(kbasename(filename).

This patch has to change set_task_comm() and trace_task_rename() to
accept "const char *", but I think this change is also good.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

procfs: make /proc/*/pagemap 0400

The /proc/*/pagemap contain sensitive information and currently its mode
is 0444.  Change this to 0400, so the VFS will prevent unprivileged
processes from getting file descriptors on arbitrary privileged
/proc/*/pagemap files.

This reduces the scope of address space leaking and bypasses by protecting
already running processes.

Signed-off-by: Djalal Harouni <tixxdz@opendz.org>
Acked-by: Kees Cook <keescook@chromium.org>
Acked-by: Andy Lutomirski <luto@amacapital.net>
Cc: Eric W. Biederman <ebiederm@xmission.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

procfs: make /proc/*/{stack,syscall,personality} 0400

These procfs files contain sensitive information and currently their
mode is 0444.  Change this to 0400, so the VFS will be able to block
unprivileged processes from getting file descriptors on arbitrary
privileged /proc/*/{stack,syscall,personality} files.

This reduces the scope of ASLR leaking and bypasses by protecting already
running processes.

Signed-off-by: Djalal Harouni <tixxdz@opendz.org>
Acked-by: Kees Cook <keescook@chromium.org>
Acked-by: Andy Lutomirski <luto@amacapital.net>
Cc: Eric W. Biederman <ebiederm@xmission.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fs/proc/inode.c: use RCU_INIT_POINTER(x, NULL)

Replace rcu_assign_pointer(x, NULL) with RCU_INIT_POINTER(x, NULL)

The rcu_assign_pointer() ensures that the initialization of a structure
is carried out before storing a pointer to that structure.  And in the
case of the NULL pointer, there is no structure to initialize.  So,
rcu_assign_pointer(p, NULL) can be safely converted to
RCU_INIT_POINTER(p, NULL)

Signed-off-by: Monam Agarwal <monamagarwal123@gmail.com>
Cc: "Paul E. McKenney" <paulmck@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

proc: show mnt_id in /proc/pid/fdinfo

Currently we don't have a way how to determing from which mount point
file has been opened.  This information is required for proper dumping
and restoring file descriptos due to presence of mount namespaces.  It's
possible, that two file descriptors are opened using the same paths, but
one fd references mount point from one namespace while the other fd --
from other namespace.

$ ls -l /proc/1/fd/1
lrwx------ 1 root root 64 Mar 19 23:54 /proc/1/fd/1 -> /dev/null

$ cat /proc/1/fdinfo/1
pos: 0
flags: 0100002
mnt_id: 16

$ cat /proc/1/mountinfo | grep ^16
16 32 0:4 / /dev rw,nosuid shared:2 - devtmpfs devtmpfs rw,size=1013356k,nr_inodes=253339,mode=755

Signed-off-by: Andrey Vagin <avagin@openvz.org>
Acked-by: Pavel Emelyanov <xemul@parallels.com>
Acked-by: Cyrill Gorcunov <gorcunov@openvz.org>
Cc: Rob Landley <rob@landley.net>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fs/proc/meminfo: meminfo_proc_show(): fix typo in comment

It should read "reclaimable slab" and not "reclaimable swap".

Signed-off-by: Luiz Capitulino <lcapitulino@redhat.com>
Reviewed-by: Rik van Riel <riel@redhat.com>
Acked-by: Rafael Aquini <aquini@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

kernel/exit.c: call proc_exit_connector() after exit_state is set

The process events connector delivers a notification when a process
exits.  This is really convenient for a process that spawns and wants to
monitor its children through an epoll-able() interface.

Unfortunately, there is a small window between when the event is
delivered and the child become wait()-able.

This is creates a race if the parent wants to make sure that it knows
about the exit, e.g

pid_t pid = fork();
if (pid > 0) {
register_interest_for_pid(pid);
if (waitpid(pid, NULL, WNOHANG) > 0)
{
  /* We might have raced with exit() */
}
return;
}

/* Child */
execve(...)

register_interest_for_pid() would be telling the the connector socket
reader to pay attention to events related to pid.

Though this is not a bug, I think it would make the connector a bit more
usable if this race was closed by simply moving the call to
proc_exit_connector() from just before exit_notify() to right after.

Oleg said:

: Even with this patch the code above is still "racy" if the child is
: multi-threaded.  Plus it should obviously filter-out subthreads.  And
: afaics there is no way to make it reliable, even if you change the code
: above so that waitpid() is called only after the last thread exits WNOHANG
: still can fail.

Signed-off-by: Guillaume Morin <guillaume@morinfr.org>
Cc: Matt Helsley <matt.helsley@gmail.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: David S. Miller <davem@davemloft.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

exit: move check_stack_usage() to the end of do_exit()

It is not clear why check_stack_usage() is called so early and thus it
never checks the stack usage in, say, exit_notify() or
flush_ptrace_hw_breakpoint() or other functions which are only called by
do_exit().

Move the callsite down to the last preempt_disable/schedule.

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

exit: call disassociate_ctty() before exit_task_namespaces()

Commit 8aac62706ada ("move exit_task_namespaces() outside of
exit_notify()") breaks pppd and the exiting service crashes the kernel:

    BUG: unable to handle kernel NULL pointer dereference at 0000000000000028
    IP: ppp_register_channel+0x13/0x20 [ppp_generic]
    Call Trace:
      ppp_asynctty_open+0x12b/0x170 [ppp_async]
      tty_ldisc_open.isra.2+0x27/0x60
      tty_ldisc_hangup+0x1e3/0x220
      __tty_hangup+0x2c4/0x440
      disassociate_ctty+0x61/0x270
      do_exit+0x7f2/0xa50

ppp_register_channel() needs ->net_ns and current->nsproxy == NULL.

Move disassociate_ctty() before exit_task_namespaces(), it doesn't make
sense to delay it after perf_event_exit_task() or cgroup_exit().

This also allows to use task_work_add() inside the (nontrivial) code
paths in disassociate_ctty().

Investigated by Peter Hurley.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Reported-by: Sree Harsha Totakura <sreeharsha@totakura.in>
Cc: Peter Hurley <peter@hurleysoftware.com>
Cc: Sree Harsha Totakura <sreeharsha@totakura.in>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Jeff Dike <jdike@addtoit.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Andrey Vagin <avagin@openvz.org>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: <stable@vger.kernel.org> [v3.10+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/zswap.c: remove unnecessary parentheses

Fix following trivial checkpatch error:

  ERROR: return is not a function, parentheses are not required

Signed-off-by: SeongJae Park <sj38.park@gmail.com>
Acked-by: Seth Jennings <sjennings@variantweb.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/zswap: support multiple swap devices

Cai Liu reporeted that now zbud pool pages counting has a problem when
multiple swap is used because it just counts only one swap intead of all
of swap so zswap cannot control writeback properly.  The result is
unnecessary writeback or no writeback when we should really writeback.

IOW, it made zswap crazy.

Another problem in zswap is:

For example, let's assume we use two swap A and B with different
priority and A already has charged 19% long time ago and let's assume
that A swap is full now so VM start to use B so that B has charged 1%
recently.  It menas zswap charged (19% + 1%) is full by default.  Then,
if VM want to swap out more pages into B, zbud_reclaim_page would be
evict one of pages in B's pool and it would be repeated continuously.
It's totally LRU reverse problem and swap thrashing in B would happen.

This patch makes zswap consider mutliple swap by creating *a* zbud pool
which will be shared by multiple swap so all of zswap pages in multiple
swap keep order by LRU so it can prevent above two problems.

Signed-off-by: Minchan Kim <minchan@kernel.org>
Reported-by: Cai Liu <cai.liu@samsung.com>
Suggested-by: Weijie Yang <weijie.yang.kh@gmail.com>
Cc: Seth Jennings <sjennings@variantweb.net>
Reviewed-by: 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/zswap.c: update zsmalloc in comment to zbud

zswap used zsmalloc before and now using zbud.  But, some comments saying
it use zsmalloc yet.  Fix the trivial problems.

Signed-off-by: SeongJae Park <sj38.park@gmail.com>
Cc: Seth Jennings <sjenning@linux.vnet.ibm.com>
Cc: Minchan Kim <minchan@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/zswap.c: fix trivial typo and arrange indentation

Signed-off-by: SeongJae Park <sj38.park@gmail.com>
Cc: Seth Jennings <sjenning@linux.vnet.ibm.com>
Cc: Minchan Kim <minchan@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

zram: support REQ_DISCARD

zram is ram based block device and can be used by backend of filesystem.
When filesystem deletes a file, it normally doesn't do anything on data
block of that file.  It just marks on metadata of that file.  This
behavior has no problem on disk based block device, but has problems on
ram based block device, since we can't free memory used for data block.
To overcome this disadvantage, there is REQ_DISCARD functionality.  If
block device support REQ_DISCARD and filesystem is mounted with discard
option, filesystem sends REQ_DISCARD to block device whenever some data
blocks are discarded.  All we have to do is to handle this request.

This patch implements to flag up QUEUE_FLAG_DISCARD and handle this
REQ_DISCARD request.  With it, we can free memory used by zram if it isn't
used.

[akpm@linux-foundation.org: tweak comments]
Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Nitin Gupta <ngupta@vflare.org>
Cc: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Cc: Jerome Marchand <jmarchan@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

zram: use scnprintf() in attrs show() methods

sysfs.txt documentation lists the following requirements:

 - The buffer will always be PAGE_SIZE bytes in length. On i386, this
   is 4096.

 - show() methods should return the number of bytes printed into the
   buffer. This is the return value of scnprintf().

 - show() should always use scnprintf().

Use scnprintf() in show() functions.

Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Acked-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: propagate error to user

When we initialized zcomp with single, we couldn't change
max_comp_streams without zram reset but current interface doesn't show
any error to user and even it changes max_comp_streams's value without
any effect so it would make user very confusing.

This patch prevents max_comp_streams's change when zcomp was initialized
as single zcomp and emit the error to user(ex, echo).

[akpm@linux-foundation.org: don't return with the lock held, per Sergey]
[fengguang.wu@intel.com: fix coccinelle warnings]
Signed-off-by: Minchan Kim <minchan@kernel.org>
Cc: Nitin Gupta <ngupta@vflare.org>
Cc: Jerome Marchand <jmarchan@redhat.com>
Acked-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Signed-off-by: Fengguang Wu <fengguang.wu@intel.com>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

zram: return error-valued pointer from zcomp_create()

Instead of returning just NULL, return ERR_PTR from zcomp_create() if
compressing backend creation has failed.  ERR_PTR(-EINVAL) for unsupported
compression algorithm request, ERR_PTR(-ENOMEM) for allocation (zcomp or
compression stream) error.

Perform IS_ERR() check of returned from zcomp_create() value in
disksize_store() and set return code to PTR_ERR().

Change suggested by Jerome Marchand.

[akpm@linux-foundation.org: clean up error recovery flow]
Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Reported-by: Jerome Marchand <jmarchan@redhat.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Nitin Gupta <ngupta@vflare.org>
Cc: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

zram: move comp allocation out of init_lock

While fixing lockdep spew of ->init_lock reported by Sasha Levin [1],
Minchan Kim noted [2] that it's better to move compression backend
allocation (using GPF_KERNEL) out of the ->init_lock lock, same way as
with zram_meta_alloc(), in order to prevent the same lockdep spew.

[1] https://lkml.org/lkml/2014/2/27/337
[2] https://lkml.org/lkml/2014/3/3/32

Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Reported-by: Minchan Kim <minchan@kernel.org>
Acked-by: Minchan Kim <minchan@kernel.org>
Cc: Sasha Levin <sasha.levin@oracle.com>
Acked-by: Jerome Marchand <jmarchan@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

zram: add lz4 algorithm backend

Introduce LZ4 compression backend and make it available for selection.
LZ4 support is optional and requires user to set ZRAM_LZ4_COMPRESS config
option.  The default compression backend is LZO.

TEST

(x86_64, core i5, 2 cores + 2 hyperthreading, zram disk size 1G,
ext4 file system, 3 compression streams)

iozone -t 3 -R -r 16K -s 60M -I +Z

       Test           LZO           LZ4
----------------------------------------------
  Initial write   1642744.62    1317005.09
        Rewrite   2498980.88    1800645.16
           Read   3957026.38    5877043.75
        Re-read   3950997.38    5861847.00
   Reverse Read   2937114.56    5047384.00
    Stride read   2948163.19    4929587.38
    Random read   3292692.69    4880793.62
 Mixed workload   1545602.62    3502940.38
   Random write   2448039.75    1758786.25
         Pwrite   1670051.03    1338329.69
          Pread   2530682.00    5097177.62
         Fwrite   3232085.62    3275942.56
          Fread   6306880.25    6645271.12

So on my system LZ4 is slower in write-only tests, while it performs
better in read-only and mixed (reads + writes) tests.

Official LZ4 benchmarks available here http://code.google.com/p/lz4/
(linux kernel uses revision r90).

Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Acked-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: make compression algorithm selection possible

Add and document `comp_algorithm' device attribute.  This attribute allows
to show supported compression and currently selected compression
algorithms:

cat /sys/block/zram0/comp_algorithm
[lzo] lz4

and change selected compression algorithm:
echo lzo > /sys/block/zram0/comp_algorithm

Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Acked-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: add set_max_streams knob

This patch allows to change max_comp_streams on initialised zcomp.

Introduce zcomp set_max_streams() knob, zcomp_strm_multi_set_max_streams()
and zcomp_strm_single_set_max_streams() callbacks to change streams limit
for zcomp_strm_multi and zcomp_strm_single, accordingly.  set_max_streams
for single steam zcomp does nothing.

If user has lowered the limit, then zcomp_strm_multi_set_max_streams()
attempts to immediately free extra streams (as much as it can, depending
on idle streams availability).

Note, this patch does not allow to change stream 'policy' from single to
multi stream (or vice versa) on already initialised compression backend.

Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Acked-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: add multi stream functionality

Existing zram (zcomp) implementation has only one compression stream
(buffer and algorithm private part), so in order to prevent data
corruption only one write (compress operation) can use this compression
stream, forcing all concurrent write operations to wait for stream lock
to be released.  This patch changes zcomp to keep a compression streams
list of user-defined size (via sysfs device attr).  Each write operation
still exclusively holds compression stream, the difference is that we
can have N write operations (depending on size of streams list)
executing in parallel.  See TEST section later in commit message for
performance data.

Introduce struct zcomp_strm_multi and a set of functions to manage
zcomp_strm stream access.  zcomp_strm_multi has a list of idle
zcomp_strm structs, spinlock to protect idle list and wait queue, making
it possible to perform parallel compressions.

The following set of functions added:
- zcomp_strm_multi_find()/zcomp_strm_multi_release()
  find and release a compression stream, implement required locking
- zcomp_strm_multi_create()/zcomp_strm_multi_destroy()
  create and destroy zcomp_strm_multi

zcomp ->strm_find() and ->strm_release() callbacks are set during
initialisation to zcomp_strm_multi_find()/zcomp_strm_multi_release()
correspondingly.

Each time zcomp issues a zcomp_strm_multi_find() call, the following set
of operations performed:

- spin lock strm_lock
- if idle list is not empty, remove zcomp_strm from idle list, spin
  unlock and return zcomp stream pointer to caller
- if idle list is empty, current adds itself to wait queue. it will be
  awaken by zcomp_strm_multi_release() caller.

zcomp_strm_multi_release():
- spin lock strm_lock
- add zcomp stream to idle list
- spin unlock, wake up sleeper

Minchan Kim reported that spinlock-based locking scheme has demonstrated
a severe perfomance regression for single compression stream case,
comparing to mutex-based (see https://lkml.org/lkml/2014/2/18/16)

base                      spinlock                    mutex

==Initial write           ==Initial write             ==Initial  write
records:  5               records:  5                 records:   5
avg:      1642424.35      avg:      699610.40         avg:       1655583.71
std:      39890.95(2.43%) std:      232014.19(33.16%) std:       52293.96
max:      1690170.94      max:      1163473.45        max:       1697164.75
min:      1568669.52      min:      573429.88         min:       1553410.23
==Rewrite                 ==Rewrite                   ==Rewrite
records:  5               records:  5                 records:   5
avg:      1611775.39      avg:      501406.64         avg:       1684419.11
std:      17144.58(1.06%) std:      15354.41(3.06%)   std:       18367.42
max:      1641800.95      max:      531356.78         max:       1706445.84
min:      1593515.27      min:      488817.78         min:       1655335.73

When only one compression stream available, mutex with spin on owner
tends to perform much better than frequent wait_event()/wake_up().  This
is why single stream implemented as a special case with mutex locking.

Introduce and document zram device attribute max_comp_streams.  This
attr shows and stores current zcomp's max number of zcomp streams
(max_strm).  Extend zcomp's zcomp_create() with `max_strm' parameter.
`max_strm' limits the number of zcomp_strm structs in compression
backend's idle list (max_comp_streams).

max_comp_streams used during initialisation as follows:
-- passing to zcomp_create() max_strm equals to 1 will initialise zcomp
using single compression stream zcomp_strm_single (mutex-based locking).
-- passing to zcomp_create() max_strm greater than 1 will initialise zcomp
using multi compression stream zcomp_strm_multi (spinlock-based locking).

default max_comp_streams value is 1, meaning that zram with single stream
will be initialised.

Later patch will introduce configuration knob to change max_comp_streams
on already initialised and used zcomp.

TEST
iozone -t 3 -R -r 16K -s 60M -I +Z

       test           base       1 strm (mutex)     3 strm (spinlock)
-----------------------------------------------------------------------
 Initial write      589286.78       583518.39          718011.05
       Rewrite      604837.97       596776.38         1515125.72
  Random write      584120.11       595714.58         1388850.25
        Pwrite      535731.17       541117.38          739295.27
        Fwrite     1418083.88      1478612.72         1484927.06

Usage example:
set max_comp_streams to 4
        echo 4 > /sys/block/zram0/max_comp_streams

show current max_comp_streams (default value is 1).
        cat /sys/block/zram0/max_comp_streams

Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Acked-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: factor out single stream compression

This is preparation patch to add multi stream support to zcomp.

Introduce struct zcomp_strm_single and a set of functions to manage
zcomp_strm stream access.  zcomp_strm_single implements single compession
stream, same way as current zcomp implementation.  This moves zcomp_strm
stream control and locking from zcomp, so compressing backend zcomp is not
aware of required locking.

Single and multi streams require different locking schemes.  Minchan Kim
reported that spinlock-based locking scheme (which is used in multi stream
implementation) has demonstrated a severe perfomance regression for single
compression stream case, comparing to mutex-based.  see
https://lkml.org/lkml/2014/2/18/16

The following set of functions added:
- zcomp_strm_single_find()/zcomp_strm_single_release()
  find and release a compression stream, implement required locking
- zcomp_strm_single_create()/zcomp_strm_single_destroy()
  create and destroy zcomp_strm_single

New ->strm_find() and ->strm_release() callbacks added to zcomp, which are
set to zcomp_strm_single_find() and zcomp_strm_single_release() during
initialisation.  Instead of direct locking and zcomp_strm access from
zcomp_strm_find() and zcomp_strm_release(), zcomp now calls ->strm_find()
and ->strm_release() correspondingly.

Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Acked-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 zcomp compressing backends

Do not perform direct LZO compress/decompress calls, initialise
and use zcomp LZO backend (single compression stream) instead.

[akpm@linux-foundation.org: resolve conflicts with zram-delete-zram_init_device-fix.patch]
Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Acked-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: introduce compressing backend abstraction

ZRAM performs direct LZO compression algorithm calls, making it the one
and only option.  While LZO is generally performs well, LZ4 algorithm
tends to have a faster decompression (see http://code.google.com/p/lz4/
for full report)

Name            Ratio  C.speed D.speed
                        MB/s    MB/s
LZ4 (r101)      2.084    422    1820
LZO 2.06        2.106    414     600

Thus, users who have mostly read (decompress) usage scenarious or mixed
workflow (writes with relatively high read ops number) will benefit from
using LZ4 compression backend.

Introduce compressing backend abstraction zcomp in order to support
multiple compression algorithms with the following set of operations:

        .create
        .destroy
        .compress
        .decompress

Schematically zram write() usually contains the following steps:
0) preparation (decompression of partioal IO, etc.)
1) lock buffer_lock mutex (protects meta compress buffers)
2) compress (using meta compress buffers)
3) alloc and map zs_pool object
4) copy compressed data (from meta compress buffers) to object allocated by 3)
5) free previous pool page, assign a new one
6) unlock buffer_lock mutex

As we can see, compressing buffers must remain untouched from 1) to 4),
because, otherwise, concurrent write() can overwrite data.  At the same
time, zram_meta must be aware of a) specific compression algorithm memory
requirements and b) necessary locking to protect compression buffers.  To
remove requirement a) new struct zcomp_strm introduced, which contains a
compress/decompress `buffer' and compression algorithm `private' part.
While struct zcomp implements zcomp_strm stream handling and locking and
removes requirement b) from zram meta.  zcomp ->create() and ->destroy(),
respectively, allocate and deallocate algorithm specific zcomp_strm
`private' part.

Every zcomp has zcomp stream and mutex to protect its compression stream.
Stream usage semantics remains the same -- only one write can hold stream
lock and use its buffers.  zcomp_strm_find() turns caller into exclusive
user of a stream (holding stream mutex until zram release stream), and
zcomp_strm_release() makes zcomp stream available (unlock the stream
mutex).  Hence no concurrent write (compression) operations possible at
the moment.

iozone -t 3 -R -r 16K -s 60M -I +Z

       test            base           patched
--------------------------------------------------
  Initial write      597992.91       591660.58
        Rewrite      609674.34       616054.97
           Read     2404771.75      2452909.12
        Re-read     2459216.81      2470074.44
   Reverse Read     1652769.66      1589128.66
    Stride read     2202441.81      2202173.31
    Random read     2236311.47      2276565.31
 Mixed workload     1423760.41      1709760.06
   Random write      579584.08       615933.86
         Pwrite      597550.02       594933.70
          Pread     1703672.53      1718126.72
         Fwrite     1330497.06      1461054.00
          Fread     3922851.00      3957242.62

Usage examples:

comp = zcomp_create(NAME) /* NAME e.g. "lzo" */

which initialises compressing backend if requested algorithm is supported.

Compress:
zstrm = zcomp_strm_find(comp)
zcomp_compress(comp, zstrm, src, &dst_len)
[..] /* copy compressed data */
zcomp_strm_release(comp, zstrm)

Decompress:
zcomp_decompress(comp, src, src_len, dst);

Free compessing backend and its zcomp stream:
zcomp_destroy(comp)

Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Acked-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: delete zram_init_device()

allocate new `zram_meta' in disksize_store() only for uninitialised zram
device, saving a number of allocations and deallocations in case if
disksize_store() was called on currently used device.  at the same time
zram_meta stack variable is not necessary, because we can set ->meta
directly.  there is also no need in setting QUEUE_FLAG_NONROT queue on
every disksize_store(), set it once during device creation.

[minchan@kernel.org: handle zram->meta alloc fail case]
[minchan@kernel.org: prevent lockdep spew of init_lock]
Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Signed-off-by: Minchan Kim <minchan@kernel.org>
Acked-by: Jerome Marchand <jmarchan@redhat.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>

zram: document failed_reads, failed_writes stats

Document `failed_reads' and `failed_writes' device attributes.
Remove info about `discard' - there is no such zram attr.

Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Cc: 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: move zram size warning to documentation

Move zram warning about disksize and size of memory correlation to zram
documentation.

Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Acked-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: drop not used table `count' member

struct table `count' member is not used.

Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Cc: Minchan Kim <minchan@kernel.org>
Acked-by: 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: report failed read and write stats

zram accounted but did not report numbers of failed read and write
queries.  make these stats available as failed_reads and failed_writes
attrs.

Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Acked-by: Minchan Kim <minchan@kernel.org>
Acked-by: 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 zram stats code duplication

Introduce ZRAM_ATTR_RO macro that generates device_attribute and default
ATTR show() function for existing atomic64_t zram stats.

Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Acked-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 atomic64_t for all zram stats

This is a preparation patch for stats code duplication removal.

1) use atomic64_t for `pages_zero' and `pages_stored' zram stats.

2) `compr_size' and `pages_zero' struct zram_stats members did not
   follow the existing device attr naming scheme: zram_stats.ATTR has
   ATTR_show() function.  rename them:

   -- compr_size -> compr_data_size
   -- pages_zero -> zero_pages

Minchan Kim's note:
 If we really have trouble with atomic stat operation, we could
 change it with percpu_counter so that it could solve atomic overhead and
 unnecessary memory space by introducing unsigned long instead of 64bit
 atomic_t.

Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Acked-by: Minchan Kim <minchan@kernel.org>
Acked-by: 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 good and bad compress stats

Remove `good' and `bad' compressed sub-requests stats.  RW request may
cause a number of RW sub-requests.  zram used to account `good' compressed
sub-queries (with compressed size less than 50% of original size), `bad'
compressed sub-queries (with compressed size greater that 75% of original
size), leaving sub-requests with compression size between 50% and 75% of
original size not accounted and not reported.  zram already accounts each
sub-request's compression size so we can calculate real device compression
ratio.

Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Acked-by: Minchan Kim <minchan@kernel.org>
Acked-by: 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: do not pass rw argument to __zram_make_request()

Do not pass rw argument down the __zram_make_request() -> zram_bvec_rw()
chain, decode it in zram_bvec_rw() instead.  Besides, this is the place
where we distinguish READ and WRITE bio data directions, so account zram
RW stats here, instead of __zram_make_request().  This also allows to
account a real number of zram READ/WRITE operations, not just requests
(single RW request may cause a number of zram RW ops with separate
locking, compression/decompression, etc).

Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Acked-by: Minchan Kim <minchan@kernel.org>
Acked-by: 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: drop `init_done' struct zram member

Introduce init_done() helper function which allows us to drop `init_done'
struct zram member.  init_done() uses the fact that ->init_done == 1
equals to ->meta != NULL.

Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Acked-by: Minchan Kim <minchan@kernel.org>
Acked-by: 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>

mm/page_alloc.c: change mm debug routines back to EXPORT_SYMBOL

A new dump_page() routine was recently added, and marked
EXPORT_SYMBOL_GPL.  dump_page() was also added to the VM_BUG_ON_PAGE()
macro, and so the end result is that non-GPL code can no longer call
get_page() and a few other routines.

This only happens if the kernel was compiled with CONFIG_DEBUG_VM.

Change dump_page() to be EXPORT_SYMBOL.

Longer explanation:

Prior to commit 309381feaee5 ("mm: dump page when hitting a VM_BUG_ON
using VM_BUG_ON_PAGE") , it was possible to build MIT-licensed (non-GPL)
drivers on Fedora.  Fedora is semi-unique, in that it sets
CONFIG_VM_DEBUG.

Because Fedora sets CONFIG_VM_DEBUG, they end up pulling in dump_page(),
via VM_BUG_ON_PAGE, via get_page().  As one of the authors of NVIDIA's
new, open source, "UVM-Lite" kernel module, I originally choose to use
the kernel's get_page() routine from within nvidia_uvm_page_cache.c,
because get_page() has always seemed to be very clearly intended for use
by non-GPL, driver code.

So I'm hoping that making get_page() widely accessible again will not be
too controversial.  We did check with Fedora first, and they responded
(https://bugzilla.redhat.com/show_bug.cgi?id=1074710#c3) that we should
try to get upstream changed, before asking Fedora to change.  Their
reasoning seems beneficial to Linux: leaving CONFIG_DEBUG_VM set allows
Fedora to help catch mm bugs.

Signed-off-by: John Hubbard <jhubbard@nvidia.com>
Cc: Sasha Levin <sasha.levin@oracle.com>
Cc: Josh Boyer <jwboyer@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

numa: use LAST_CPUPID_SHIFT to calculate LAST_CPUPID_MASK

LAST_CPUPID_MASK is calculated using LAST_CPUPID_WIDTH.  However
LAST_CPUPID_WIDTH itself can be 0.  (when LAST_CPUPID_NOT_IN_PAGE_FLAGS is
set).  In such a case LAST_CPUPID_MASK turns out to be 0.

But with recent commit 1ae71d0319: (mm: numa: bugfix for
LAST_CPUPID_NOT_IN_PAGE_FLAGS) if LAST_CPUPID_MASK is 0,
page_cpupid_xchg_last() and page_cpupid_reset_last() causes
page->_last_cpupid to be set to 0.

This causes performance regression. Its almost as if numa_balancing is
off.

Fix LAST_CPUPID_MASK by using LAST_CPUPID_SHIFT instead of
LAST_CPUPID_WIDTH.

Some performance numbers and perf stats with and without the fix.

(3.14-rc6)
----------
numa01

 Performance counter stats for '/usr/bin/time -f %e %S %U %c %w -o start_bench.out -a ./numa01':

         12,27,462 cs                                                           [100.00%]
          2,41,957 migrations                                                   [100.00%]
       1,68,01,713 faults                                                       [100.00%]
    7,99,35,29,041 cache-misses
            98,808 migrate:mm_migrate_pages                                     [100.00%]

    1407.690148814 seconds time elapsed

numa02

 Performance counter stats for '/usr/bin/time -f %e %S %U %c %w -o start_bench.out -a ./numa02':

            63,065 cs                                                           [100.00%]
            14,364 migrations                                                   [100.00%]
          2,08,118 faults                                                       [100.00%]
      25,32,59,404 cache-misses
                12 migrate:mm_migrate_pages                                     [100.00%]

      63.840827219 seconds time elapsed

(3.14-rc6 with fix)
-------------------
numa01

 Performance counter stats for '/usr/bin/time -f %e %S %U %c %w -o start_bench.out -a ./numa01':

          9,68,911 cs                                                           [100.00%]
          1,01,414 migrations                                                   [100.00%]
         88,38,697 faults                                                       [100.00%]
    4,42,92,51,042 cache-misses
          4,25,060 migrate:mm_migrate_pages                                     [100.00%]

     685.965331189 seconds time elapsed

numa02

 Performance counter stats for '/usr/bin/time -f %e %S %U %c %w -o start_bench.out -a ./numa02':

            17,543 cs                                                           [100.00%]
             2,962 migrations                                                   [100.00%]
          1,17,843 faults                                                       [100.00%]
      11,80,61,644 cache-misses
            12,358 migrate:mm_migrate_pages                                     [100.00%]

      20.380132343 seconds time elapsed

Signed-off-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: Liu Ping Fan <pingfank@linux.vnet.ibm.com>
Reviewed-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

madvise: correct the comment of MADV_DODUMP flag

s/MADV_NODUMP/MADV_DONTDUMP/

Signed-off-by: Zhang Yanfei <zhangyanfei@cn.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/readahead.c: inline ra_submit

Commit f9acc8c7b35a ("readahead: sanify file_ra_state names") left
ra_submit with a single function call.

Move ra_submit to internal.h and inline it to save some stack.  Thanks
to Andrew Morton for commenting different versions.

Signed-off-by: Fabian Frederick <fabf@skynet.be>
Suggested-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: hugetlb: fix softlockup when a large number of hugepages are freed.

When I decrease the value of nr_hugepage in procfs a lot, softlockup
happens.  It is because there is no chance of context switch during this
process.

On the other hand, when I allocate a large number of hugepages, there is
some chance of context switch.  Hence softlockup doesn't happen during
this process.  So it's necessary to add the context switch in the
freeing process as same as allocating process to avoid softlockup.

When I freed 12 TB hugapages with kernel-2.6.32-358.el6, the freeing
process occupied a CPU over 150 seconds and following softlockup message
appeared twice or more.

$ echo 6000000 > /proc/sys/vm/nr_hugepages
$ cat /proc/sys/vm/nr_hugepages
6000000
$ grep ^Huge /proc/meminfo
HugePages_Total:   6000000
HugePages_Free:    6000000
HugePages_Rsvd:        0
HugePages_Surp:        0
Hugepagesize:       2048 kB
$ echo 0 > /proc/sys/vm/nr_hugepages

BUG: soft lockup - CPU#16 stuck for 67s! [sh:12883] ...
Pid: 12883, comm: sh Not tainted 2.6.32-358.el6.x86_64 #1
Call Trace:
  free_pool_huge_page+0xb8/0xd0
  set_max_huge_pages+0x128/0x190
  hugetlb_sysctl_handler_common+0x113/0x140
  hugetlb_sysctl_handler+0x1e/0x20
  proc_sys_call_handler+0x97/0xd0
  proc_sys_write+0x14/0x20
  vfs_write+0xb8/0x1a0
  sys_write+0x51/0x90
  __audit_syscall_exit+0x265/0x290
  system_call_fastpath+0x16/0x1b

I have not confirmed this problem with upstream kernels because I am not
able to prepare the machine equipped with 12TB memory now.  However I
confirmed that the amount of decreasing hugepages was directly
proportional to the amount of required time.

I measured required times on a smaller machine.  It showed 130-145
hugepages decreased in a millisecond.

  Amount of decreasing     Required time      Decreasing rate
  hugepages                     (msec)         (pages/msec)
  ------------------------------------------------------------
  10,000 pages == 20GB         70 -  74          135-142
  30,000 pages == 60GB        208 - 229          131-144

It means decrement of 6TB hugepages will trigger softlockup with the
default threshold 20sec, in this decreasing rate.

Signed-off-by: Masayoshi Mizuma <m.mizuma@jp.fujitsu.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Wanpeng Li <liwanp@linux.vnet.ibm.com>
Cc: Aneesh Kumar <aneesh.kumar@linux.vnet.ibm.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Naoya Horiguchi <n-horiguchi@ah.jp.nec.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/memblock.c: use PFN_PHYS()

Replace ((phys_addr_t)(x) << PAGE_SHIFT) by pfn macro.

Signed-off-by: Fabian Frederick <fabf@skynet.be>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memblock: use for_each_memblock()

This is a small cleanup.

Signed-off-by: Emil Medve <Emilian.Medve@Freescale.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: remove unused arg of set_page_dirty_balance()

There's only one caller of set_page_dirty_balance() and that will call it
with page_mkwrite == 0.

The page_mkwrite argument was unused since commit b827e496c893 "mm: close
page_mkwrite races".

Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: try_to_unmap_cluster() should lock_page() before mlocking

A BUG_ON(!PageLocked) was triggered in mlock_vma_page() by Sasha Levin
fuzzing with trinity.  The call site try_to_unmap_cluster() does not lock
the pages other than its check_page parameter (which is already locked).

The BUG_ON in mlock_vma_page() is not documented and its purpose is
somewhat unclear, but apparently it serializes against page migration,
which could otherwise fail to transfer the PG_mlocked flag.  This would
not be fatal, as the page would be eventually encountered again, but
NR_MLOCK accounting would become distorted nevertheless.  This patch adds
a comment to the BUG_ON in mlock_vma_page() and munlock_vma_page() to that
effect.

The call site try_to_unmap_cluster() is fixed so that for page !=
check_page, trylock_page() is attempted (to avoid possible deadlocks as we
already have check_page locked) and mlock_vma_page() is performed only
upon success.  If the page lock cannot be obtained, the page is left
without PG_mlocked, which is again not a problem in the whole unevictable
memory design.

Signed-off-by: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Bob Liu <bob.liu@oracle.com>
Reported-by: Sasha Levin <sasha.levin@oracle.com>
Cc: Wanpeng Li <liwanp@linux.vnet.ibm.com>
Cc: Michel Lespinasse <walken@google.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Hugh Dickins <hughd@google.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.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: page_alloc: spill to remote nodes before waking kswapd

On NUMA systems, a node may start thrashing cache or even swap anonymous
pages while there are still free pages on remote nodes.

This is a result of commits 81c0a2bb515f ("mm: page_alloc: fair zone
allocator policy") and fff4068cba48 ("mm: page_alloc: revert NUMA aspect
of fair allocation policy").

Before those changes, the allocator would first try all allowed zones,
including those on remote nodes, before waking any kswapds.  But now,
the allocator fastpath doubles as the fairness pass, which in turn can
only consider the local node to prevent remote spilling based on
exhausted fairness batches alone.  Remote nodes are only considered in
the slowpath, after the kswapds are woken up.  But if remote nodes still
have free memory, kswapd should not be woken to rebalance the local node
or it may thrash cash or swap prematurely.

Fix this by adding one more unfair pass over the zonelist that is
allowed to spill to remote nodes after the local fairness pass fails but
before entering the slowpath and waking the kswapds.

This also gets rid of the GFP_THISNODE exemption from the fairness
protocol because the unfair pass is no longer tied to kswapd, which
GFP_THISNODE is not allowed to wake up.

However, because remote spills can be more frequent now - we prefer them
over local kswapd reclaim - the allocation batches on remote nodes could
underflow more heavily.  When resetting the batches, use
atomic_long_read() directly instead of zone_page_state() to calculate the
delta as the latter filters negative counter values.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Rik van Riel <riel@redhat.com>
Acked-by: Mel Gorman <mgorman@suse.de>
Cc: <stable@kernel.org> [3.12+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: rename high level charging functions

mem_cgroup_newpage_charge is used only for charging anonymous memory so
it is better to rename it to mem_cgroup_charge_anon.

mem_cgroup_cache_charge is used for file backed memory so rename it to
mem_cgroup_charge_file.

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

memcg: sanitize __mem_cgroup_try_charge() call protocol

Some callsites pass a memcg directly, some callsites pass an mm that
then has to be translated to a memcg.  This makes for a terrible
function interface.

Just push the mm-to-memcg translation into the respective callsites and
always pass a memcg to mem_cgroup_try_charge().

[mhocko@suse.cz: add charge mm helper]
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Michal Hocko <mhocko@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: do not replicate get_mem_cgroup_from_mm in __mem_cgroup_try_charge

__mem_cgroup_try_charge duplicates get_mem_cgroup_from_mm for charges
which came without a memcg.  The only reason seems to be a tiny
optimization when css_tryget is not called if the charge can be consumed
from the stock.  Nevertheless css_tryget is very cheap since it has been
reworked to use per-cpu counting so this optimization doesn't give us
anything these days.

So let's drop the code duplication so that the code is more readable.

Signed-off-by: Michal Hocko <mhocko@suse.cz>
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>

memcg: get_mem_cgroup_from_mm()

Instead of returning NULL from try_get_mem_cgroup_from_mm() when the mm
owner is exiting, just return root_mem_cgroup.  This makes sense for all
callsites and gets rid of some of them having to fallback manually.

[fengguang.wu@intel.com: fix warnings]
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Fengguang Wu <fengguang.wu@intel.com>
Acked-by: Michal Hocko <mhocko@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: remove unnecessary !mm check from try_get_mem_cgroup_from_mm()

Users pass either a mm that has been established under task lock, or use
a verified current->mm, which means the task can't be exiting.

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

mm: memcg: push !mm handling out to page cache charge function

Only page cache charges can happen without an mm context, so push this
special case out of the inner core and into the cache charge function.

An ancient comment explains that the mm can also be NULL in case the
task is currently being migrated, but that is not actually true with the
current case, so just remove it.

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>

mm: memcg: inline mem_cgroup_charge_common()

mem_cgroup_charge_common() is used by both cache and anon pages, but
most of its body only applies to anon pages and the remainder is not
worth having in a separate function.

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

mm: memcg: remove mem_cgroup_move_account_page_stat()

It used to disable preemption and run sanity checks but now it's only
taking a number out of one percpu counter and putting it into another.
Do this directly in the callsite and save the indirection.

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

mm: memcg: remove unnecessary preemption disabling

lock_page_cgroup() disables preemption, remove explicit preemption
disabling for code paths holding this lock.

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

mm: use 'const char *' insted of 'char *' for reason in dump_page()

I tried to use 'dump_page(page, __func__)' for debugging, but it triggers
warning:

  warning: passing argument 2 of `dump_page' discards `const' qualifier from pointer target type [enabled by default]

Let's convert 'reason' to 'const char *' in dump_page() and friends: we
shouldn't modify it anyway.

Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/vmalloc.c: enhance vm_map_ram() comment

vm_map_ram() has a fragmentation problem when it cannot purge a
chunk(ie, 4M address space) if there is a pinning object in that
addresss space.  So it could consume all VMALLOC address space easily.

We can fix the fragmentation problem by using vmap instead of
vm_map_ram() but vmap() is known to be slow compared to vm_map_ram().
Minchan said vm_map_ram is 5 times faster than vmap in his tests.  So I
thought we should fix fragment problem of vm_map_ram because our
proprietary GPU driver has used it heavily.

On second thought, it's not an easy because we should reuse freed space
for solving the problem and it could make more IPI and bitmap operation
for searching hole.  It could mitigate API's goal which is very fast
mapping.  And even fragmentation problem wouldn't show in 64 bit
machine.

Another option is that the user should separate long-life and short-life
object and use vmap for long-life but vm_map_ram for short-life.  If we
inform the user about the characteristic of vm_map_ram the user can
choose one according to the page lifetime.

Let's add some notice messages to user.

[akpm@linux-foundation.org: tweak comment text]
Signed-off-by: Gioh Kim <gioh.kim@lge.com>
Reviewed-by: Zhang Yanfei <zhangyanfei@cn.fujitsu.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: fix 'ERROR: do not initialise globals to 0 or NULL' and coding style

Signed-off-by: Choi Gi-yong <yong@gnoy.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mempool: add unlikely and likely hints

Add unlikely and likely hints to the function mempool_free.  It lays out
the code in such a way that the common path is executed straighforward and
saves a cache line.

Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, compaction: determine isolation mode only once

The conditions that control the isolation mode in
isolate_migratepages_range() do not change during the iteration, so
extract them out and only define the value once.

This actually does have an effect, gcc doesn't optimize it itself because
of cc->sync.

Signed-off-by: David Rientjes <rientjes@google.com>
Cc: Mel Gorman <mgorman@suse.de>
Acked-by: Rik van Riel <riel@redhat.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
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>

res_counter: remove interface for locked charging and uncharging

The res_counter_{charge,uncharge}_locked() variants are not used in the
kernel outside of the resource counter code itself, so remove the
interface.

Signed-off-by: David Rientjes <rientjes@google.com>
Acked-by: Michal Hocko <mhocko@suse.cz>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Christoph Lameter <cl@linux-foundation.org>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: Tejun Heo <tj@kernel.org>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Jianguo Wu <wujianguo@huawei.com>
Cc: Tim Hockin <thockin@google.com>
Cc: Christoph Lameter <cl@linux.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, mempolicy: remove per-process flag

PF_MEMPOLICY is an unnecessary optimization for CONFIG_SLAB users.
There's no significant performance degradation to checking
current->mempolicy rather than current->flags & PF_MEMPOLICY in the
allocation path, especially since this is considered unlikely().

Running TCP_RR with netperf-2.4.5 through localhost on 16 cpu machine with
64GB of memory and without a mempolicy:

threads before after
16 1249409 1244487
32 1281786 1246783
48 1239175 1239138
64 1244642 1241841
80 1244346 1248918
96 1266436 1254316
112 1307398 1312135
128 1327607 1326502

Per-process flags are a scarce resource so we should free them up whenever
possible and make them available.  We'll be using it shortly for memcg oom
reserves.

Signed-off-by: David Rientjes <rientjes@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Christoph Lameter <cl@linux-foundation.org>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: Tejun Heo <tj@kernel.org>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Jianguo Wu <wujianguo@huawei.com>
Cc: Tim Hockin <thockin@google.com>
Cc: Christoph Lameter <cl@linux.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, mempolicy: rename slab_node for clarity

slab_node() is actually a mempolicy function, so rename it to
mempolicy_slab_node() to make it clearer that it used for processes with
mempolicies.

At the same time, cleanup its code by saving numa_mem_id() in a local
variable (since we require a node with memory, not just any node) and
remove an obsolete comment that assumes the mempolicy is actually passed
into the function.

Signed-off-by: David Rientjes <rientjes@google.com>
Acked-by: Christoph Lameter <cl@linux.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Christoph Lameter <cl@linux-foundation.org>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: Tejun Heo <tj@kernel.org>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Jianguo Wu <wujianguo@huawei.com>
Cc: Tim Hockin <thockin@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fork: collapse copy_flags into copy_process

copy_flags() does not use the clone_flags formal and can be collapsed
into copy_process() for cleaner code.

Signed-off-by: David Rientjes <rientjes@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Christoph Lameter <cl@linux-foundation.org>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: Tejun Heo <tj@kernel.org>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Jianguo Wu <wujianguo@huawei.com>
Cc: Tim Hockin <thockin@google.com>
Cc: Christoph Lameter <cl@linux.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: use macros from compiler.h instead of __attribute__((...))

To increase compiler portability there is <linux/compiler.h> which
provides convenience macros for various gcc constructs.  Eg: __weak for
__attribute__((weak)).  I've replaced all instances of gcc attributes with
the right macro in the memory management (/mm) subsystem.

[akpm@linux-foundation.org: while-we're-there consistency tweaks]
Signed-off-by: Gideon Israel Dsouza <gidisrael@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: per-thread vma caching

This patch is a continuation of efforts trying to optimize find_vma(),
avoiding potentially expensive rbtree walks to locate a vma upon faults.
The original approach (https://lkml.org/lkml/2013/11/1/410), where the
largest vma was also cached, ended up being too specific and random,
thus further comparison with other approaches were needed.  There are
two things to consider when dealing with this, the cache hit rate and
the latency of find_vma().  Improving the hit-rate does not necessarily
translate in finding the vma any faster, as the overhead of any fancy
caching schemes can be too high to consider.

We currently cache the last used vma for the whole address space, which
provides a nice optimization, reducing the total cycles in find_vma() by
up to 250%, for workloads with good locality.  On the other hand, this
simple scheme is pretty much useless for workloads with poor locality.
Analyzing ebizzy runs shows that, no matter how many threads are
running, the mmap_cache hit rate is less than 2%, and in many situations
below 1%.

The proposed approach is to replace this scheme with a small per-thread
cache, maximizing hit rates at a very low maintenance cost.
Invalidations are performed by simply bumping up a 32-bit sequence
number.  The only expensive operation is in the rare case of a seq
number overflow, where all caches that share the same address space are
flushed.  Upon a miss, the proposed replacement policy is based on the
page number that contains the virtual address in question.  Concretely,
the following results are seen on an 80 core, 8 socket x86-64 box:

1) System bootup: Most programs are single threaded, so the per-thread
   scheme does improve ~50% hit rate by just adding a few more slots to
   the cache.

+----------------+----------+------------------+
| caching scheme | hit-rate | cycles (billion) |
+----------------+----------+------------------+
| baseline       | 50.61%   | 19.90            |
| patched        | 73.45%   | 13.58            |
+----------------+----------+------------------+

2) Kernel build: This one is already pretty good with the current
   approach as we're dealing with good locality.

+----------------+----------+------------------+
| caching scheme | hit-rate | cycles (billion) |
+----------------+----------+------------------+
| baseline       | 75.28%   | 11.03            |
| patched        | 88.09%   | 9.31             |
+----------------+----------+------------------+

3) Oracle 11g Data Mining (4k pages): Similar to the kernel build workload.

+----------------+----------+------------------+
| caching scheme | hit-rate | cycles (billion) |
+----------------+----------+------------------+
| baseline       | 70.66%   | 17.14            |
| patched        | 91.15%   | 12.57            |
+----------------+----------+------------------+

4) Ebizzy: There's a fair amount of variation from run to run, but this
   approach always shows nearly perfect hit rates, while baseline is just
   about non-existent.  The amounts of cycles can fluctuate between
   anywhere from ~60 to ~116 for the baseline scheme, but this approach
   reduces it considerably.  For instance, with 80 threads:

+----------------+----------+------------------+
| caching scheme | hit-rate | cycles (billion) |
+----------------+----------+------------------+
| baseline       | 1.06%    | 91.54            |
| patched        | 99.97%   | 14.18            |
+----------------+----------+------------------+

[akpm@linux-foundation.org: fix nommu build, per Davidlohr]
[akpm@linux-foundation.org: document vmacache_valid() logic]
[akpm@linux-foundation.org: attempt to untangle header files]
[akpm@linux-foundation.org: add vmacache_find() BUG_ON]
[hughd@google.com: add vmacache_valid_mm() (from Oleg)]
[akpm@linux-foundation.org: coding-style fixes]
[akpm@linux-foundation.org: adjust and enhance comments]
Signed-off-by: Davidlohr Bueso <davidlohr@hp.com>
Reviewed-by: Rik van Riel <riel@redhat.com>
Acked-by: Linus Torvalds <torvalds@linux-foundation.org>
Reviewed-by: Michel Lespinasse <walken@google.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Tested-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: implement ->map_pages for shmem/tmpfs

In shmem/tmpfs, we also use the generic filemap_map_pages, seems the
additional checking is not worth a separate version of map_pages for it.

Signed-off-by: Ning Qu <quning@google.com>
Acked-by: Hugh Dickins <hughd@google.com>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Rik van Riel <riel@redhat.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Dave Chinner <david@fromorbit.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: add debugfs tunable for fault_around_order

Let's allow people to tweak faultaround at runtime.

[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Rik van Riel <riel@redhat.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Matthew Wilcox <matthew.r.wilcox@intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Ning Qu <quning@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: cleanup size checks in filemap_fault() and filemap_map_pages()

Minor cleanups:
 - 'size' variable is now in bytes, not pages;
 - use round_up(): it should be easier to read.

Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Rik van Riel <riel@redhat.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Matthew Wilcox <matthew.r.wilcox@intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Ning Qu <quning@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: implement ->map_pages for page cache

filemap_map_pages() is generic implementation of ->map_pages() for
filesystems who uses page cache.

It should be safe to use filemap_map_pages() for ->map_pages() if
filesystem use filemap_fault() for ->fault().

Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Acked-by: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Rik van Riel <riel@redhat.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Matthew Wilcox <matthew.r.wilcox@intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Ning Qu <quning@gmail.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: introduce vm_ops->map_pages()

Here's new version of faultaround patchset.  It took a while to tune it
and collect performance data.

First patch adds new callback ->map_pages to vm_operations_struct.

->map_pages() is called when VM asks to map easy accessible pages.
Filesystem should find and map pages associated with offsets from
"pgoff" till "max_pgoff".  ->map_pages() is called with page table
locked and must not block.  If it's not possible to reach a page without
blocking, filesystem should skip it.  Filesystem should use do_set_pte()
to setup page table entry.  Pointer to entry associated with offset
"pgoff" is passed in "pte" field in vm_fault structure.  Pointers to
entries for other offsets should be calculated relative to "pte".

Currently VM use ->map_pages only on read page fault path.  We try to
map FAULT_AROUND_PAGES a time.  FAULT_AROUND_PAGES is 16 for now.
Performance data for different FAULT_AROUND_ORDER is below.

TODO:
 - implement ->map_pages() for shmem/tmpfs;
 - modify get_user_pages() to be able to use ->map_pages() and implement
   mmap(MAP_POPULATE|MAP_NONBLOCK) on top.

=========================================================================
Tested on 4-socket machine (120 threads) with 128GiB of RAM.

Few real-world workloads. The sweet spot for FAULT_AROUND_ORDER here is
somewhere between 3 and 5. Let's say 4 :)

Linux build (make -j60)
FAULT_AROUND_ORDER Baseline 1 3 4 5 7 9
minor-faults 283,301,572 247,151,987 212,215,789 204,772,882 199,568,944 194,703,779 193,381,485
time, seconds 151.227629483 153.920996480 151.356125472 150.863792049 150.879207877 151.150764954 151.450962358
Linux rebuild (make -j60)
FAULT_AROUND_ORDER Baseline 1 3 4 5 7 9
minor-faults 5,396,854 4,148,444 2,855,286 2,577,282 2,361,957 2,169,573 2,112,643
time, seconds 27.404543757 27.559725591 27.030057426 26.855045126 26.678618635 26.974523490 26.761320095
Git test suite (make -j60 test)
FAULT_AROUND_ORDER Baseline 1 3 4 5 7 9
minor-faults 129,591,823 99,200,751 66,106,718 57,606,410 51,510,808 45,776,813 44,085,515
time, seconds 66.087215026 64.784546905 64.401156567 65.282708668 66.034016829 66.793780811 67.237810413

Two synthetic tests: access every word in file in sequential/random order.
It doesn't improve much after FAULT_AROUND_ORDER == 4.

Sequential access 16GiB file
FAULT_AROUND_ORDER Baseline 1 3 4 5 7 9
 1 thread
minor-faults 4,195,437 2,098,275 525,068 262,251 131,170 32,856 8,282
time, seconds 7.250461742 6.461711074 5.493859139 5.488488147 5.707213983 5.898510832 5.109232856
 8 threads
minor-faults 33,557,540 16,892,728 4,515,848 2,366,999 1,423,382 442,732 142,339
time, seconds 16.649304881 9.312555263 6.612490639 6.394316732 6.669827501 6.75078944 6.371900528
 32 threads
minor-faults 134,228,222 67,526,810 17,725,386 9,716,537 4,763,731 1,668,921 537,200
time, seconds 49.164430543 29.712060103 12.938649729 10.175151004 11.840094583 9.594081325 9.928461797
 60 threads
minor-faults 251,687,988 126,146,952 32,919,406 18,208,804 10,458,947 2,733,907 928,217
time, seconds 86.260656897 49.626551828 22.335007632 17.608243696 16.523119035 16.339489186 16.326390902
 120 threads
minor-faults 503,352,863 252,939,677 67,039,168 35,191,827 19,170,091 4,688,357 1,471,862
time, seconds 124.589206333 79.757867787 39.508707872 32.167281632 29.972989292 28.729834575 28.042251622
Random access 1GiB file
 1 thread
minor-faults 262,636 132,743 34,369 17,299 8,527 3,451 1,222
time, seconds 15.351890914 16.613802482 16.569227308 15.179220992 16.557356122 16.578247824 15.365266994
 8 threads
minor-faults 2,098,948 1,061,871 273,690 154,501 87,110 25,663 7,384
time, seconds 15.040026343 15.096933500 14.474757288 14.289129964 14.411537468 14.296316837 14.395635804
 32 threads
minor-faults 8,390,734 4,231,023 1,054,432 528,847 269,242 97,746 26,881
time, seconds 20.430433109 21.585235358 22.115062928 14.872878951 14.880856305 14.883370649 14.821261690
 60 threads
minor-faults 15,733,258 7,892,809 1,973,393 988,266 594,789 164,994 51,691
time, seconds 26.577302548 25.692397770 18.728863715 20.153026398 21.619101933 17.745086260 17.613215273
 120 threads
minor-faults 31,471,111 15,816,616 3,959,209 1,978,685 1,008,299 264,635 96,010
time, seconds 41.835322703 40.459786095 36.085306105 35.313894834 35.814445675 36.552633793 34.289210594

Touch only one page in page table in 16GiB file
FAULT_AROUND_ORDER Baseline 1 3 4 5 7 9
 1 thread
minor-faults 8,372 8,324 8,270 8,260 8,249 8,239 8,237
time, seconds 0.039892712 0.045369149 0.051846126 0.063681685 0.079095975 0.17652406 0.541213386
 8 threads
minor-faults 65,731 65,681 65,628 65,620 65,608 65,599 65,596
time, seconds 0.124159196 0.488600638 0.156854426 0.191901957 0.242631486 0.543569456 1.677303984
 32 threads
minor-faults 262,388 262,341 262,285 262,276 262,266 262,257 263,183
time, seconds 0.452421421 0.488600638 0.565020946 0.648229739 0.789850823 1.651584361 5.000361559
 60 threads
minor-faults 491,822 491,792 491,723 491,711 491,701 491,691 491,825
time, seconds 0.763288616 0.869620515 0.980727360 1.161732354 1.466915814 3.04041448 9.308612938
 120 threads
minor-faults 983,466 983,655 983,366 983,372 983,363 984,083 984,164
time, seconds 1.595846553 1.667902182 2.008959376 2.425380942 2.941368804 5.977807890 18.401846125

This patch (of 2):

Introduce new vm_ops callback ->map_pages() and uses it for mapping easy
accessible pages around fault address.

On read page fault, if filesystem provides ->map_pages(), we try to map up
to FAULT_AROUND_PAGES pages around page fault address in hope to reduce
number of minor page faults.

We call ->map_pages first and use ->fault() as fallback if page by the
offset is not ready to be mapped (cold page cache or something).

Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Acked-by: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Rik van Riel <riel@redhat.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Matthew Wilcox <matthew.r.wilcox@intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Ning Qu <quning@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

drivers/lguest/page_tables.c: rename do_set_pte()

"mm: introduce vm_ops->map_pages()" wants to export a do_set_pte() from core
kernel.  Rename lguest's do_set_pte() to something more lguest-specific.

Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
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>

tools/vm/page-types.c: page-cache sniffing feature

After this patch 'page-types' can walk over a file's mappings and
analyze populated page cache pages mostly without disturbing its state.

It maps chunk of file, marks VMA as MADV_RANDOM to turn off readahead,
pokes VMA via mincore() to determine cached pages, triggers page-fault
only for them, and finally gathers information via pagemap/kpageflags.
Before unmap it marks VMA as MADV_SEQUENTIAL for ignoring reference
bits.

usage: page-types -f <path>

If <path> is directory it will analyse all files in all subdirectories.

Symlinks are not followed as well as mount points.  Hardlinks aren't
handled, they'll be dumped as many times as they are found.  Recursive
walk brings all dentries into dcache and populates page cache of
block-devices aka 'Buffers'.

Probably it's worth to add ioctl for dumping file page cache as array of
PFNs as a replacement for this hackish juggling with
mmap/madvise/mincore/pagemap.  Also recursive walk could be replaced
with dumping cached inodes via some ioctl or debugfs interface followed
by openning them via open_by_handle_at, this would fix hardlinks
handling and unneeded population of dcache and buffers.  This interface
might be used as data source for constructing readahead plans and for
background optimizations of actively used files.

collateral changes:
+ fix 64-bit LFS: define _FILE_OFFSET_BITS instead of _LARGEFILE64_SOURCE
+ replace lseek + read with single pread
+ make show_page_range() reusable after flush

usage example:

  ~/src/linux/tools/vm$ sudo ./page-types -L -f page-types
  foffset offset    flags
  page-types       Inode: 2229277       Size: 89065 (22 pages)
  Modify: Tue Feb 25 12:00:59 2014 (162 seconds ago)
  Access: Tue Feb 25 12:01:00 2014 (161 seconds ago)
  0       3cbf3b     __RU_lA____M________________________
  1       38946a     __RU_lA____M________________________
  2       1a3cec     __RU_lA____M________________________
  3       1a8321     __RU_lA____M________________________
  4       3af7cc     __RU_lA____M________________________
  5       1ed532     __RU_lA_____________________________
  6       2e436a     __RU_lA_____________________________
  7       29a35e     ___U_lA_____________________________
  8       2de86e     ___U_lA_____________________________
  9       3bdfb4     ___U_lA_____________________________
  10      3cd8a3     ___U_lA_____________________________
  11      2afa50     ___U_lA_____________________________
  12      2534c2     ___U_lA_____________________________
  13      1b7a40     ___U_lA_____________________________
  14      17b0be     ___U_lA_____________________________
  15      392b0c     ___U_lA_____________________________
  16      3ba46a     __RU_lA_____________________________
  17      397dc8     ___U_lA_____________________________
  18      1f2a36     ___U_lA_____________________________
  19      21fd30     __RU_lA_____________________________
  20      2c35ba     __RU_l______________________________
  21      20f181     __RU_l______________________________

               flags page-count   MB  symbolic-flags                        long-symbolic-flags
  0x000000000000002c          2    0  __RU_l______________________________  referenced,uptodate,lru
  0x0000000000000068         11    0  ___U_lA_____________________________  uptodate,lru,active
  0x000000000000006c          4    0  __RU_lA_____________________________  referenced,uptodate,lru,active
  0x000000000000086c          5    0  __RU_lA____M________________________  referenced,uptodate,lru,active,mmap
               total         22    0

  ~/src/linux/tools/vm$ sudo ./page-types -f /
               flags page-count     MB  symbolic-flags                        long-symbolic-flags
  0x0000000000000028      21761     85  ___U_l______________________________  uptodate,lru
  0x000000000000002c     127279    497  __RU_l______________________________  referenced,uptodate,lru
  0x0000000000000068      74160    289  ___U_lA_____________________________  uptodate,lru,active
  0x000000000000006c      84469    329  __RU_lA_____________________________  referenced,uptodate,lru,active
  0x000000000000007c          1      0  __RUDlA_____________________________  referenced,uptodate,dirty,lru,active
  0x0000000000000228        370      1  ___U_l___I__________________________  uptodate,lru,reclaim
  0x0000000000000828         49      0  ___U_l_____M________________________  uptodate,lru,mmap
  0x000000000000082c        126      0  __RU_l_____M________________________  referenced,uptodate,lru,mmap
  0x0000000000000868        137      0  ___U_lA____M________________________  uptodate,lru,active,mmap
  0x000000000000086c      12890     50  __RU_lA____M________________________  referenced,uptodate,lru,active,mmap
               total     321242   1254

Signed-off-by: Konstantin Khlebnikov <koct9i@gmail.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Fengguang Wu <fengguang.wu@intel.com>
Cc: Borislav Petkov <bp@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: disable split page table lock for !MMU

There's no reason to enable split page table lock if don't have page
tables.

It also triggers build error at least on ARM since we don't define
pmd_page() for !MMU.

  In file included from arch/arm/kernel/asm-offsets.c:14:0:
  include/linux/mm.h: In function 'pte_lockptr':
  include/linux/mm.h:1392:2: error: implicit declaration of function 'pmd_page' [-Werror=implicit-function-declaration]
  include/linux/mm.h:1392:2: warning: passing argument 1 of 'ptlock_ptr' makes pointer from integer without a cast [enabled by default]
  include/linux/mm.h:1384:27: note: expected 'struct page *' but argument is of type 'int'

Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Reported-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

exec: kill the unnecessary mm->def_flags setting in load_elf_binary()

load_elf_binary() sets current->mm->def_flags = def_flags and def_flags
is always zero.  Not only this looks strange, this is unnecessary
because mm_init() has already set ->def_flags = 0.

Signed-off-by: Alex Thorlton <athorlton@sgi.com>
Suggested-by: Oleg Nesterov <oleg@redhat.com>
Cc: Gerald Schaefer <gerald.schaefer@de.ibm.com>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Christian Borntraeger <borntraeger@de.ibm.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Mel Gorman <mgorman@suse.de>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, thp: add VM_INIT_DEF_MASK and PRCTL_THP_DISABLE

Add VM_INIT_DEF_MASK, to allow us to set the default flags for VMs.  It
also adds a prctl control which allows us to set the THP disable bit in
mm->def_flags so that VMs will pick up the setting as they are created.

Signed-off-by: Alex Thorlton <athorlton@sgi.com>
Suggested-by: Oleg Nesterov <oleg@redhat.com>
Cc: Gerald Schaefer <gerald.schaefer@de.ibm.com>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Christian Borntraeger <borntraeger@de.ibm.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Mel Gorman <mgorman@suse.de>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: revert "thp: make MADV_HUGEPAGE check for mm->def_flags"

The main motivation behind this patch is to provide a way to disable THP
for jobs where the code cannot be modified, and using a malloc hook with
madvise is not an option (i.e.  statically allocated data).  This patch
allows us to do just that, without affecting other jobs running on the
system.

We need to do this sort of thing for jobs where THP hurts performance,
due to the possibility of increased remote memory accesses that can be
created by situations such as the following:

When you touch 1 byte of an untouched, contiguous 2MB chunk, a THP will
be handed out, and the THP will be stuck on whatever node the chunk was
originally referenced from.  If many remote nodes need to do work on
that same chunk, they'll be making remote accesses.

With THP disabled, 4K pages can be handed out to separate nodes as
they're needed, greatly reducing the amount of remote accesses to
memory.

This patch is based on some of my work combined with some
suggestions/patches given by Oleg Nesterov.  The main goal here is to
add a prctl switch to allow us to disable to THP on a per mm_struct
basis.

Here's a bit of test data with the new patch in place...

First with the flag unset:

  # perf stat -a ./prctl_wrapper_mmv3 0 ./thp_pthread -C 0 -m 0 -c 512 -b 256g
  Setting thp_disabled for this task...
  thp_disable: 0
  Set thp_disabled state to 0
  Process pid = 18027

                                                                                                                       PF/
                                  MAX        MIN                                  TOTCPU/      TOT_PF/   TOT_PF/     WSEC/
  TYPE:               CPUS       WALL       WALL        SYS     USER     TOTCPU       CPU     WALL_SEC   SYS_SEC       CPU   NODES
   512      1.120      0.060      0.000    0.110      0.110     0.000    28571428864 -9223372036854775808  55803572      23

   Performance counter stats for './prctl_wrapper_mmv3_hack 0 ./thp_pthread -C 0 -m 0 -c 512 -b 256g':

    273719072.841402 task-clock                #  641.026 CPUs utilized           [100.00%]
           1,008,986 context-switches          #    0.000 M/sec                   [100.00%]
               7,717 CPU-migrations            #    0.000 M/sec                   [100.00%]
           1,698,932 page-faults               #    0.000 M/sec
  355,222,544,890,379 cycles                   #    1.298 GHz                     [100.00%]
  536,445,412,234,588 stalled-cycles-frontend  #  151.02% frontend cycles idle    [100.00%]
  409,110,531,310,223 stalled-cycles-backend   #  115.17% backend  cycles idle    [100.00%]
  148,286,797,266,411 instructions             #    0.42  insns per cycle
                                               #    3.62  stalled cycles per insn [100.00%]
  27,061,793,159,503 branches                  #   98.867 M/sec                   [100.00%]
       1,188,655,196 branch-misses             #    0.00% of all branches

       427.001706337 seconds time elapsed

Now with the flag set:

  # perf stat -a ./prctl_wrapper_mmv3 1 ./thp_pthread -C 0 -m 0 -c 512 -b 256g
  Setting thp_disabled for this task...
  thp_disable: 1
  Set thp_disabled state to 1
  Process pid = 144957

                                                                                                                       PF/
                                  MAX        MIN                                  TOTCPU/      TOT_PF/   TOT_PF/     WSEC/
  TYPE:               CPUS       WALL       WALL        SYS     USER     TOTCPU       CPU     WALL_SEC   SYS_SEC       CPU   NODES
   512      0.620      0.260      0.250    0.320      0.570     0.001    51612901376 128000000000 100806448      23

   Performance counter stats for './prctl_wrapper_mmv3_hack 1 ./thp_pthread -C 0 -m 0 -c 512 -b 256g':

    138789390.540183 task-clock                #  641.959 CPUs utilized           [100.00%]
             534,205 context-switches          #    0.000 M/sec                   [100.00%]
               4,595 CPU-migrations            #    0.000 M/sec                   [100.00%]
          63,133,119 page-faults               #    0.000 M/sec
  147,977,747,269,768 cycles                   #    1.066 GHz                     [100.00%]
  200,524,196,493,108 stalled-cycles-frontend  #  135.51% frontend cycles idle    [100.00%]
  105,175,163,716,388 stalled-cycles-backend   #   71.07% backend  cycles idle    [100.00%]
  180,916,213,503,160 instructions             #    1.22  insns per cycle
                                               #    1.11  stalled cycles per insn [100.00%]
  26,999,511,005,868 branches                  #  194.536 M/sec                   [100.00%]
         714,066,351 branch-misses             #    0.00% of all branches

       216.196778807 seconds time elapsed

As with previous versions of the patch, We're getting about a 2x
performance increase here.  Here's a link to the test case I used, along
with the little wrapper to activate the flag:

  http://oss.sgi.com/projects/memtests/thp_pthread_mmprctlv3.tar.gz

This patch (of 3):

Revert commit 8e72033f2a48 and add in code to fix up any issues caused
by the revert.

The revert is necessary because hugepage_madvise would return -EINVAL
when VM_NOHUGEPAGE is set, which will break subsequent chunks of this
patch set.

Here's a snip of an e-mail from Gerald detailing the original purpose of
this code, and providing justification for the revert:

  "The intent of commit 8e72033f2a48 was to guard against any future
   programming errors that may result in an madvice(MADV_HUGEPAGE) on
   guest mappings, which would crash the kernel.

   Martin suggested adding the bit to arch/s390/mm/pgtable.c, if
   8e72033f2a48 was to be reverted, because that check will also prevent
   a kernel crash in the case described above, it will now send a
   SIGSEGV instead.

   This would now also allow to do the madvise on other parts, if
   needed, so it is a more flexible approach.  One could also say that
   it would have been better to do it this way right from the
   beginning..."

Signed-off-by: Alex Thorlton <athorlton@sgi.com>
Suggested-by: Oleg Nesterov <oleg@redhat.com>
Tested-by: Christian Borntraeger <borntraeger@de.ibm.com>
Cc: Gerald Schaefer <gerald.schaefer@de.ibm.com>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Rik van Riel <riel@redhat.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/compaction: clean-up code on success of ballon isolation

It is just for clean-up to reduce code size and improve readability.
There is no functional change.

Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>