gru: add debug option for cache flushing

Add a debug option to the SGI GRU driver for flushing GRU cache lines from
memory.  In theory this is not needed but it is useful for debugging.
This has no use by end users.

Signed-off-by: Jack Steiner <steiner@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

gru: handle failures to mmu_notifier_register

Under some conditions, mmu_notifier_register() will fail to register a
mmu_notifier.  Fix the GRU driver to correctly handle these failures.

Signed-off-by: Jack Steiner <steiner@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

gru: support 64-bit GRU addresses

Increase the maximum address supported by the SGI GRU driver to a full 64
bits.  Note that GRU addresses are not always the same as socket virtual
addresses.  Sockets may not necessarily support the full 64 bits.

Signed-off-by: Jack Steiner <steiner@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

gru: improve messages for malfunctioning GRUs

Improve error messages for malfunctioning GRUs.  Identify the type of
instruction that is failing.

Signed-off-by: Jack Steiner <steiner@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

gru: fix bug in module unload

Fix bug in module unload.  Previous code was not correctly deleting the
files in /proc.

Signed-off-by: Jack Steiner <steiner@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

gru: allow users to specify gru chiplet 3

This patch builds on the infrastructure introduced in the patches that
allow user specification of GRU blades & chiplets for context allocation.

This patch simplifies the algorithms for migrating GRU contexts between
blades.

No new functionality is introduced.

Signed-off-by: Jack Steiner <steiner@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

gru: allow users to specify gru chiplet 2

Add support to the GRU driver to allow users to specify the blade &
chiplet for allocation of GRU contexts.  Add new statistics for context
loading/unloading/retargeting.  Also deleted a few GRU stats that were no
longer being unused.

Signed-off-by: Jack Steiner <steiner@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

gru: allow users to specify gru chiplet 1

Add table & user request infrastructure that is needed to allow users to
specify the blade and chiplet for allocation of GRU contexts.  Use of this
information is in a subsequent patch.

Signed-off-by: Jack Steiner <steiner@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

gru: handle blades without memory

Do not use alloc_pages_exact_node() to allocate GRU tables.  If a blade
has no local memory, nid will be -1.  Use alloc_pages_node() instead.

Signed-off-by: Jack Steiner <steiner@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

gru: fix istatus race in GRU tlb dropin

TLB dropins require updates to the CBR instruction istatus field.  This is
needed to resolve race conditions in the chip.

The code currently uses the user address of the CBR.  This works but opens
up additional endcases related to stealing of contexts and accessing the
CBR from tasks that do not have access to the user address space.  (Some
of this non-user task access is debug code that is not currently being
pushed to the community).

User CBRs are also directly accessible using the kernel mapping of the
CBR.  Change the TLB dropin code to use the the kernel mapping of the CBR.

Signed-off-by: Jack Steiner <steiner@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

gru: add comments raised in previous code reviews

Add comments from previous code reviews.  The comments help explain some
of the more esoteric aspects of the driver.

Move a free() to the other side of an unlock.

Signed-off-by: Jack Steiner <steiner@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

gru: initial GRU based on blade topology

Change the GRU initialization code to initialize based on blade topology
instead of node topology.  The result is the same but blade-based
initialization is cleaner.

Signed-off-by: Jack Steiner <steiner@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

UV - XPC: pass nasid instead of nid to gru_create_message_queue

Currently, the UV xpc code is passing nid to the gru_create_message_queue
instead of nasid as it expects.

Signed-off-by: Robin Holt <holt@sgi.com>
Signed-off-by: Jack Steiner <steiner@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

x86: uv: XPC receive message reuse triggers invalid BUG_ON()

This was a difficult bug to trip.  XPC was in the middle of sending an
acknowledgement for a received message.

In xpc_received_payload_uv():
.
        ret = xpc_send_gru_msg(ch->sn.uv.cached_notify_gru_mq_desc, msg,
                               sizeof(struct xpc_notify_mq_msghdr_uv));
        if (ret != xpSuccess)
                XPC_DEACTIVATE_PARTITION(&xpc_partitions[ch->partid], ret);

        msg->hdr.msg_slot_number += ch->remote_nentries;

at the point in xpc_send_gru_msg() where the hardware has dispatched the
acknowledgement, the remote side is able to reuse the message structure
and send a message with a different slot number.  This problem is made
worse by interrupts.

The adjustment of msg_slot_number and the BUG_ON in
xpc_handle_notify_mq_msg_uv() which verifies the msg_slot_number is
consistent are only used for debug purposes.  Since a fix for this that
preserves the debug functionality would either have to infringe upon the
payload or allocate another structure just for debug, I decided to remove
it entirely.

Signed-off-by: Robin Holt <holt@sgi.com>
Cc: Jack Steiner <steiner@sgi.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>

X86: uv: xpc_make_first_contact hang due to not accepting ACTIVE state

Many times while the initial connection is being made, the contacted
partition will send back both the ACTIVATING and the ACTIVE
remote_act_state changes in very close succescion.  The 1/4 second delay
in the make first contact loop is large enough to nearly always miss the
ACTIVATING state change.

Since either state indicates the remote partition has acknowledged our
state change, accept either.

Signed-off-by: Robin Holt <holt@sgi.com>
Cc: Jack Steiner <steiner@sgi.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>

x86: uv: xpc NULL deref when mesq becomes empty

Under heavy load conditions, our set of xpc messages may become exhausted.
 The code handles this correctly with the exception of the management code
which hits a NULL pointer dereference.

Signed-off-by: Robin Holt <holt@sgi.com>
Cc: Jack Steiner <steiner@sgi.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>

x86: uv: update XPC to handle updated BIOS interface

The UV BIOS has moved the location of some of their pointers to the
"partition reserved page" from memory into a uv hub MMR.  The GRU does not
support bcopy operations from MMR space so we need to special case the MMR
addresses using VLOAD operations.

Additionally, the BIOS call for registering a message queue watchlist has
removed the 'blade' value and eliminated the structure that was being
passed in.  This is also reflected in this patch.

Signed-off-by: Robin Holt <holt@sgi.com>
Cc: Jack Steiner <steiner@sgi.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>

X86: uv: implement a gru_read_gpa kernel function

The BIOS has decided to store a pointer to the partition reserved page in
a scratch MMR.  The GRU is only able to read an MMR using a vload
instruction.  The gru_read_gpa() function will implemented.

Signed-off-by: Robin Holt <holt@sgi.com>
Signed-off-by: Jack Steiner <steiner@sgi.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>

x86: uv: introduce uv_gpa_is_mmr

Provide a mechanism for determining if a global physical address is
pointing to a UV hub MMR.

Signed-off-by: Robin Holt <holt@sgi.com>
Cc: Jack Steiner <steiner@sgi.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>

x86: uv: xpc needs to provide an abstraction for uv_gpa

Provide an SGI SN2/UV agnositic method for converting a global physical
address into a socket physical address.

Signed-off-by: Robin Holt <holt@sgi.com>
Cc: Jack Steiner <steiner@sgi.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>

x86: uv: introduce a means to translate from gpa -> socket_paddr

The UV BIOS has been updated to implement some of our interface
functionality differently than originally expected.  These patches update
the kernel to the bios implementation and include a few minor bug fixes
which prevent us from doing significant testing on real hardware.

This patch:

For SGI UV systems, translate from a global physical address back to a
socket physical address.  This does nothing to ensure the socket physical
address is actually addressable by the kernel.  That is the responsibility
of the user of the function.

Signed-off-by: Robin Holt <holt@sgi.com>
Cc: Jack Steiner <steiner@sgi.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>

direct-io: cleanup blockdev_direct_IO locking

Currently the locking in blockdev_direct_IO is a mess, we have three
different locking types and very confusing checks for some of them.  The
most complicated one is DIO_OWN_LOCKING for reads, which happens to not
actually be used.

This patch gets rid of the DIO_OWN_LOCKING - as mentioned above the read
case is unused anyway, and the write side is almost identical to
DIO_NO_LOCKING.  The difference is that DIO_NO_LOCKING always sets the
create argument for the get_blocks callback to zero, but we can easily
move that to the actual get_blocks callbacks.  There are four users of the
DIO_NO_LOCKING mode: gfs already ignores the create argument and thus is
fine with the new version, ocfs2 only errors out if create were ever set,
and we can remove this dead code now, the block device code only ever uses
create for an error message if we are fully beyond the device which can
never happen, and last but not least XFS will need the new behavour for
writes.

Now we can replace the lock_type variable with a flags one, where no flag
means the DIO_NO_LOCKING behaviour and DIO_LOCKING is kept as the first
flag.  Separate out the check for not allowing to fill holes into a
separate flag, although for now both flags always get set at the same
time.

Also revamp the documentation of the locking scheme to actually make
sense.

[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Christoph Hellwig <hch@lst.de>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Badari Pulavarty <pbadari@us.ibm.com>
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Jens Axboe <jens.axboe@oracle.com>
Cc: Zach Brown <zach.brown@oracle.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Alex Elder <aelder@sgi.com>
Cc: Mark Fasheh <mfasheh@suse.com>
Cc: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

dio: don't zero out the pages array inside struct dio

Intel reported a performance regression caused by the following commit:

commit 848c4dd5153c7a0de55470ce99a8e13a63b4703f
Author: Zach Brown <zach.brown@oracle.com>
Date:   Mon Aug 20 17:12:01 2007 -0700

    dio: zero struct dio with kzalloc instead of manually

    This patch uses kzalloc to zero all of struct dio rather than
    manually trying to track which fields we rely on being zero.  It
    passed aio+dio stress testing and some bug regression testing on
    ext3.

    This patch was introduced by Linus in the conversation that lead up
    to Badari's minimal fix to manually zero .map_bh.b_state in commit:

      6a648fa72161d1f6468dabd96c5d3c0db04f598a

    It makes the code a bit smaller.  Maybe a couple fewer cachelines to
    load, if we're lucky:

       text    data     bss     dec     hex filename
    3285925  568506 1304616 5159047  4eb887 vmlinux
    3285797  568506 1304616 5158919  4eb807 vmlinux.patched

    I was unable to measure a stable difference in the number of cpu
    cycles spent in blockdev_direct_IO() when pushing aio+dio 256K reads
    at ~340MB/s.

    So the resulting intent of the patch isn't a performance gain but to
    avoid exposing ourselves to the risk of finding another field like
    .map_bh.b_state where we rely on zeroing but don't enforce it in the
    code.

Zach surmised that zeroing out the page array was what caused most of
the problem, and suggested the approach taken in the attached patch for
resolving the issue.  Intel re-tested with this patch and saw a 0.6%
performance gain (the original regression was 0.5%).

[akpm@linux-foundation.org: add comment]
Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
Acked-by: Zach Brown <zach.brown@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

aio: remove unused field

Don't know the reason, but it appears ki_wait field of iocb never gets used.

Signed-off-by: Shaohua Li <shaohua.li@intel.com>
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Benjamin LaHaise <bcrl@kvack.org>
Cc: Zach Brown <zach.brown@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

FS-Cache: Avoid maybe-used-uninitialised warning on variable

Andrew Morton's compiler sees the following warning in FS-Cache:

fs/fscache/object-list.c: In function 'fscache_objlist_lookup':
fs/fscache/object-list.c:94: warning: 'obj' may be used uninitialized in this function

which my compiler doesn't.  This is a false positive as obj can only be
used in the comparison against minobj if minobj has been set to something
other than NULL, but for that to happen, obj has to be first set to
something.

Deal with this by preclearing obj too.

Reported-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

kexec: premit reduction of the reserved memory size

Implement shrinking the reserved memory for crash kernel, if it is more
than enough.

For example, if you have already reserved 128M, now you just want 100M,
you can do:

# echo $((100*1024*1024)) > /sys/kernel/kexec_crash_size

Note, you can only do this before loading the crash kernel.

Signed-off-by: WANG Cong <amwang@redhat.com>
Cc: Neil Horman <nhorman@redhat.com>
Acked-by: Eric W. Biederman <ebiederm@xmission.com>
Cc: Andi Kleen <andi@firstfloor.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

parport_pc.c: use correct length in strncmp

Signed-off-by: Joe Perches <joe@perches.com>
Acked-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

dma-mapping: fix off-by-one error in dma_capable()

dma_mask is, when interpreted as address, the last valid byte, and hence
comparison msut also be done using the last valid of the buffer in
question.

Also fix the open-coded instances in lib/swiotlb.c.

Signed-off-by: Jan Beulich <jbeulich@novell.com>
Cc: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Cc: Becky Bruce <beckyb@kernel.crashing.org>
Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

edac: i5100 add 6 ranks per channel

Add support for 6 ranks per channel to the i5100 chipset.  I have tested
the patch as far as possible with correctible errors and things appear
good.  The DIMM mapping is correct for our board, but boards may differ.

Signed-off-by: Nils Carlson <nils.carlson@ludd.ltu.se>
Acked-by: Arthur Jones <ajones@riverbed.com>
Signed-off-by: Doug Thompson <dougthompson@xmission.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

edac: i5100 add scrubbing

Addscrubbing to the i5100 chipset.  The i5100 chipset only supports one
scrubbing rate, which is not constant but dependent on memory load.  The
rate returned by this driver is an estimate based on some experimentation,
but is substantially closer to the truth than the speed supplied in the
documentation.

Also, scrubbing is done once, and then a done-bit is set.  This means that
to accomplish continuous scrubbing a re-enabling mechanism must be used.
I have created the simplest possible such mechanism in the form of a
work-queue which will check every five minutes.  This interval is quite
arbitrary but should be sufficient for all sizes of system memory.

Signed-off-by: Nils Carlson <nils.carlson@ludd.ltu.se>
Signed-off-by: Doug Thompson <dougthompson@xmission.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

edac: i5100 clean controller to channel terms

The i5100 driver uses the word controller instead of channel in a lot of
places, this is simply a cleanup of the patch.

Signed-off-by: Nils Carlson <nils.carlson@ludd.ltu.se>
Signed-off-by: Doug Thompson <dougthompson@xmission.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

pid: reduce code size by using a pointer to iterate over array

It decreases code size by 16 bytes on my gcc 4.4.1 on Core 2:
  text    data     bss     dec     hex filename
  4314    2216       8    6538    198a kernel/pid.o-BEFORE
  4298    2216       8    6522    197a kernel/pid.o-AFTER

Signed-off-by: André Goddard Rosa <andre.goddard@gmail.com>
Cc: Pekka Enberg <penberg@cs.helsinki.fi>
Cc: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

pid: tighten pidmap spinlock critical section by removing kfree()

Avoid calling kfree() under pidmap spinlock, calling it afterwards.

Normally kfree() is fast, but sometimes it can be slow, so avoid
calling it under the spinlock if we can do it.

Signed-off-by: André Goddard Rosa <andre.goddard@gmail.com>
Cc: Pekka Enberg <penberg@cs.helsinki.fi>
Cc: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

elf: kill USE_ELF_CORE_DUMP

Currently all architectures but microblaze unconditionally define
USE_ELF_CORE_DUMP.  The microblaze omission seems like an error to me, so
let's kill this ifdef and make sure we are the same everywhere.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Acked-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Cc: <linux-arch@vger.kernel.org>
Cc: Michal Simek <michal.simek@petalogix.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

drivers/char/ipmi: Use KCS_IDLE_STATE

KCS_IDLE and KCS_IDLE state have the same value, but in this function the
constants ending in _STATE are compared to the state variable.

Signed-off-by: Julia Lawall <julia@diku.dk>
Acked-by: Core Minyard <cminyard@mvista.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ipc: HARD_MSGMAX should be higher not lower on 64bit

We have HARD_MSGMAX lower on 64bit than on 32bit, since usually 64bit
machines have more memory than 32bit machines.

Making it higher on 64bit seems reasonable, and keep the original number
on 32bit.

Acked-by: Serge E. Hallyn <serue@us.ibm.com>
Cc: Cedric Le Goater <clg@fr.ibm.com>
Signed-off-by: WANG Cong <amwang@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ipc: remove unreachable code in sem.c

This line is unreachable, remove it.

[akpm@linux-foundation.org: remove unneeded initialisation of `err']
Signed-off-by: WANG Cong <amwang@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ipc/sem.c: optimize single sops when semval is zero

If multiple simple decrements on the same semaphore are pending, then the
current code scans all decrement operations, even if the semaphore value
is already 0.

The patch optimizes that: if the semaphore value is 0, then there is no
need to scan the q->alter entries.

Note that this is a common case: It happens if 100 decrements by one are
pending and now an increment by one increases the semaphore value from 0
to 1.  Without this patch, all 100 entries are scanned.  With the patch,
only one entry is scanned, then woken up.  Then the new rule triggers and
the scanning is aborted, without looking at the remaining 99 tasks.

With this patch, single sop increment/decrement by 1 are now O(1).
(same as with Nick's patch)

Signed-off-by: Manfred Spraul <manfred@colorfullife.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Pierre Peiffer <peifferp@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ipc/sem.c: optimize single semop operations

sysv sem has the concept of semaphore arrays that consist out of multiple
semaphores.  Atomic operations that affect multiple semaphores are
supported.

The patch optimizes single semaphore operation calls that affect only one
semaphore: It's not necessary to scan all pending operations, it is
sufficient to scan the per-semaphore list.

The idea is from Nick Piggin version of an ipc sem improvement, the
implementation is different: The code tries to keep as much common code as
possible.

As the result, the patch is simpler, but optimizes fewer cases.

Signed-off-by: Manfred Spraul <manfred@colorfullife.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Pierre Peiffer <peifferp@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ipc/sem.c: add a per-semaphore pending list

Based on Nick's findings:

sysv sem has the concept of semaphore arrays that consist out of multiple
semaphores.  Atomic operations that affect multiple semaphores are
supported.

The patch is the first step for optimizing simple, single semaphore
operations: In addition to the global list of all pending operations, a
2nd, per-semaphore list with the simple operations is added.

Note: this patch does not make sense by itself, the new list is used
nowhere.

Signed-off-by: Manfred Spraul <manfred@colorfullife.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Pierre Peiffer <peifferp@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ipc/sem.c: optimize if semops fail

Reduce the amount of scanning of the list of pending semaphore operations:
If try_atomic_semop failed, then no changes were applied.  Thus no need to
restart.

Additionally, this patch correct an incorrect comment: It's possible to
wait for arbitrary semaphore values (do a dec by <x>, wait-for-zero, inc
by <x> in one atomic operation)

Both changes are from Nick Piggin, the patch is the result of a different
split of the individual changes.

Signed-off-by: Manfred Spraul <manfred@colorfullife.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Pierre Peiffer <peifferp@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ipc/sem.c: sem preempt improve

The strange sysv semaphore wakeup scheme has a kind of busy-wait lock
involved, which could deadlock if preemption is enabled during the "lock".

It is an implementation detail (due to a spinlock being held) that this is
actually the case.  However if "spinlocks" are made preemptible, or if the
sem lock is changed to a sleeping lock for example, then the wakeup would
become buggy.  So this might be a bugfix for -rt kernels.

Imagine waker being preempted by wakee and never clearing IN_WAKEUP -- if
wakee has higher RT priority then there is a priority inversion deadlock.
Even if there is not a priority inversion to cause a deadlock, then there
is still time wasted spinning.

Signed-off-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Manfred Spraul <manfred@colorfullife.com>
Cc: Pierre Peiffer <peifferp@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ipc/sem.c: sem use list operations

Replace the handcoded list operations in update_queue() with the standard
list_for_each_entry macros.

list_for_each_entry_safe() must be used, because list entries can
disappear immediately uppon the wakeup event.

Signed-off-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Manfred Spraul <manfred@colorfullife.com>
Cc: Pierre Peiffer <peifferp@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ipc/sem.c: sem optimise undo list search

Around a month ago, there was some discussion about an improvement of the
sysv sem algorithm: Most (at least: some important) users only use simple
semaphore operations, therefore it's worthwile to optimize this use case.

This patch:

Move last looked up sem_undo struct to the head of the task's undo list.
Attempt to move common entries to the front of the list so search time is
reduced.  This reduces lookup_undo on oprofile of problematic SAP workload
by 30% (see patch 4 for a description of SAP workload).

Signed-off-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Manfred Spraul <manfred@colorfullife.com>
Cc: Pierre Peiffer <peifferp@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ipc ns: fix memory leak (idr)

We have apparently had a memory leak since
7ca7e564e049d8b350ec9d958ff25eaa24226352 "ipc: store ipcs into IDRs" in
2007.  The idr of which 3 exist for each ipc namespace is never freed.

This patch simply frees them when the ipcns is freed.  I don't believe any
idr_remove() are done from rcu (and could therefore be delayed until after
this idr_destroy()), so the patch should be safe.  Some quick testing
showed no harm, and the memory leak fixed.

Caught by kmemleak.

Signed-off-by: Serge E. Hallyn <serue@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

signals: check ->group_stop_count after tracehook_get_signal()

Move the call to do_signal_stop() down, after tracehook call.  This makes
->group_stop_count condition visible to tracers before do_signal_stop()
will participate in this group-stop.

Currently the patch has no effect, tracehook_get_signal() always returns 0.

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

signals: kill force_sig_specific()

Kill force_sig_specific(), this trivial wrapper has no callers.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: Roland McGrath <roland@redhat.com>
Cc: Sukadev Bhattiprolu <sukadev@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

signals: cosmetic, collect_signal: use SI_USER

Trivial, s/0/SI_USER/ in collect_signal() for grep.

This is a bit confusing, we don't know the source of this signal.
But we don't care, and "info->si_code = 0" is imho worse.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: Roland McGrath <roland@redhat.com>
Cc: Sukadev Bhattiprolu <sukadev@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

signals: send_signal: use si_fromuser() to detect from_ancestor_ns

Change send_signal() to use si_fromuser().  From now SEND_SIG_NOINFO
triggers the "from_ancestor_ns" check.

This fixes reparent_thread()->group_send_sig_info(pdeath_signal)
behaviour, before this patch send_signal() does not detect the
cross-namespace case when the child of the dying parent belongs to the
sub-namespace.

This patch can affect the behaviour of send_sig(), kill_pgrp() and
kill_pid() when the caller sends the signal to the sub-namespace with
"priv == 0" but surprisingly all callers seem to use them correctly,
including disassociate_ctty(on_exit).

Except: drivers/staging/comedi/drivers/addi-data/*.c incorrectly use
send_sig(priv => 0).  But his is minor and should be fixed anyway.

Reported-by: Daniel Lezcano <dlezcano@fr.ibm.com>
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: Roland McGrath <roland@redhat.com>
Reviewed-by: Sukadev Bhattiprolu <sukadev@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

signals: SEND_SIG_NOINFO should be considered as SI_FROMUSER()

No changes in compiled code. The patch adds the new helper, si_fromuser()
and changes check_kill_permission() to use this helper.

The real effect of this patch is that from now we "officially" consider
SEND_SIG_NOINFO signal as "from user-space" signals. This is already true
if we look at the code which uses SEND_SIG_NOINFO, except __send_signal()
has another opinion - see the next patch.

The naming of these special SEND_SIG_XXX siginfo's is really bad
imho.  From __send_signal()'s pov they mean

SEND_SIG_NOINFO from user
SEND_SIG_PRIV from kernel
SEND_SIG_FORCED no info

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: Roland McGrath <roland@redhat.com>
Reviewed-by: Sukadev Bhattiprolu <sukadev@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ptrace: x86: change syscall_trace_leave() to rely on tracehook when stepping

Suggested by Roland.

Unlike powepc, x86 always calls tracehook_report_syscall_exit(step) with
step = 0, and sends the trap by hand.

This results in unnecessary SIGTRAP when PTRACE_SINGLESTEP follows the
syscall-exit stop.

Change syscall_trace_leave() to pass the correct "step" argument to
tracehook and remove the send_sigtrap() logic.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Roland McGrath <roland@redhat.com>
Cc: <linux-arch@vger.kernel.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ptrace: x86: implement user_single_step_siginfo()

Suggested by Roland.

Implement user_single_step_siginfo() for x86.  Extract this code from
send_sigtrap().

Since x86 calls tracehook_report_syscall_exit(step => 0) the new helper is
not used yet.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Roland McGrath <roland@redhat.com>
Cc: <linux-arch@vger.kernel.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ptrace: change tracehook_report_syscall_exit() to handle stepping

Suggested by Roland.

Change tracehook_report_syscall_exit() to look at step flag and send the
trap signal if needed.

This change affects ia64, microblaze, parisc, powerpc, sh.  They pass
nonzero "step" argument to tracehook but since it was ignored the tracee
reports via ptrace_notify(), this is not right and not consistent.

- PTRACE_SETSIGINFO doesn't work

- if the tracer resumes the tracee with signr != 0 the new signal
  is generated rather than delivering it

- If PT_TRACESYSGOOD is set the tracee reports the wrong exit_code

I don't have a powerpc machine, but I think this test-case should see the
difference:

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

int main(void)
{
int pid, status;

if (!(pid = fork())) {
assert(ptrace(PTRACE_TRACEME) == 0);
kill(getpid(), SIGSTOP);

getppid();

return 0;
}

assert(pid == wait(&status));
assert(ptrace(PTRACE_SETOPTIONS, pid, 0, PTRACE_O_TRACESYSGOOD) == 0);

assert(ptrace(PTRACE_SYSCALL, pid, 0,0) == 0);
assert(pid == wait(&status));

assert(ptrace(PTRACE_SINGLESTEP, pid, 0,0) == 0);
assert(pid == wait(&status));

if (status == 0x57F)
return 0;

printf("kernel bug: status=%X shouldn't have 0x80\n", status);
return 1;
}

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Roland McGrath <roland@redhat.com>
Cc: <linux-arch@vger.kernel.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ptrace: powerpc: implement user_single_step_siginfo()

Suggested by Roland.

Implement user_single_step_siginfo() for powerpc.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Roland McGrath <roland@redhat.com>
Cc: <linux-arch@vger.kernel.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ptrace: introduce user_single_step_siginfo() helper

Suggested by Roland.

Currently there is no way to synthesize a single-stepping trap in the
arch-independent manner.  This patch adds the default helper which fills
siginfo_t, arch/ can can override it.

Architetures which implement user_enable_single_step() should add
user_single_step_siginfo() also.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Roland McGrath <roland@redhat.com>
Cc: <linux-arch@vger.kernel.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ptrace: copy_process() should disable stepping

If the tracee calls fork() after PTRACE_SINGLESTEP, the forked child
starts with TIF_SINGLESTEP/X86_EFLAGS_TF bits copied from ptraced parent.
This is not right, especially when the new child is not auto-attaced: in
this case it is killed by SIGTRAP.

Change copy_process() to call user_disable_single_step(). Tested on x86.

Test-case:

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

int main(void)
{
int pid, status;

if (!(pid = fork())) {
assert(ptrace(PTRACE_TRACEME) == 0);
kill(getpid(), SIGSTOP);

if (!fork()) {
/* kernel bug: this child will be killed by SIGTRAP */
printf("Hello world\n");
return 43;
}

wait(&status);
return WEXITSTATUS(status);
}

for (;;) {
assert(pid == wait(&status));
if (WIFEXITED(status))
break;
assert(ptrace(PTRACE_SINGLESTEP, pid, 0,0) == 0);
}

assert(WEXITSTATUS(status) == 43);
return 0;
}

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

ptrace: cleanup ptrace_init_task()->ptrace_link() path

No functional changes.

ptrace_init_task() looks confusing, as if we always auto-attach when "bool
ptrace" argument is true, while in fact we attach only if current is
traced.

Make the code more explicit and kill now unused ptrace_link().

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

memcg: code clean, remove unused variable in mem_cgroup_resize_limit()

Variable `progress' isn't used in mem_cgroup_resize_limit() any more.
Remove it.

[akpm@linux-foundation.org: cleanup]
Signed-off-by: Bob Liu <lliubbo@gmail.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Reviewed-by: Balbir Singh <balbir@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: remove memcg_tasklist

memcg_tasklist was introduced at commit 7f4d454d(memcg: avoid deadlock
caused by race between oom and cpuset_attach) instead of cgroup_mutex to
fix a deadlock problem.  The cgroup_mutex, which was removed by the
commit, in mem_cgroup_out_of_memory() was originally introduced at commit
c7ba5c9e (Memory controller: OOM handling).

IIUC, the intention of this cgroup_mutex was to prevent task move during
select_bad_process() so that situations like below can be avoided.

  Assume cgroup "foo" has exceeded its limit and is about to trigger oom.
  1. Process A, which has been in cgroup "baa" and uses large memory, is just
     moved to cgroup "foo". Process A can be the candidates for being killed.
  2. Process B, which has been in cgroup "foo" and uses large memory, is just
     moved from cgroup "foo". Process B can be excluded from the candidates for
     being killed.

But these race window exists anyway even if we hold a lock, because
__mem_cgroup_try_charge() decides wether it should trigger oom or not
outside of the lock.  So the original cgroup_mutex in
mem_cgroup_out_of_memory and thus current memcg_tasklist has no use.  And
IMHO, those races are not so critical for users.

This patch removes it and make codes simpler.

Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@linux.vnet.ibm.com>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: avoid oom-killing innocent task in case of use_hierarchy

task_in_mem_cgroup(), which is called by select_bad_process() to check
whether a task can be a candidate for being oom-killed from memcg's limit,
checks "curr->use_hierarchy"("curr" is the mem_cgroup the task belongs
to).

But this check return true(it's false positive) when:

<some path>/aa use_hierarchy == 0 <- hitting limit
  <some path>/aa/00 use_hierarchy == 1 <- the task belongs to

This leads to killing an innocent task in aa/00.  This patch is a fix for
this bug.  And this patch also fixes the arg for
mem_cgroup_print_oom_info().  We should print information of mem_cgroup
which the task being killed, not current, belongs to.

Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Acked-by: Balbir Singh <balbir@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: cleanup mem_cgroup_move_parent()

mem_cgroup_move_parent() calls try_charge first and cancel_charge on
failure.  IMHO, charge/uncharge(especially charge) is high cost operation,
so we should avoid it as far as possible.

This patch tries to delay try_charge in mem_cgroup_move_parent() by
re-ordering checks it does.

And this patch renames mem_cgroup_move_account() to
__mem_cgroup_move_account(), changes the return value of
__mem_cgroup_move_account() from int to void, and adds a new
wrapper(mem_cgroup_move_account()), which checks whether a @pc is valid
for moving account and calls __mem_cgroup_move_account().

This patch removes the last caller of trylock_page_cgroup(), so removes
its definition too.

Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: add mem_cgroup_cancel_charge()

There are some places calling both res_counter_uncharge() and css_put() to
cancel the charge and the refcnt we have got by mem_cgroup_tyr_charge().

This patch introduces mem_cgroup_cancel_charge() and call it in those
places.

Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Reviewed-by: Balbir Singh <balbir@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: make memcg's file mapped consistent with global VM

In global VM, FILE_MAPPED is used but memcg uses MAPPED_FILE.  This makes
grep difficult.  Replace memcg's MAPPED_FILE with FILE_MAPPED

And in global VM, mapped shared memory is accounted into FILE_MAPPED.
But memcg doesn't. fix it.
Note:
  page_is_file_cache() just checks SwapBacked or not.
  So, we need to check PageAnon.

Cc: Balbir Singh <balbir@in.ibm.com>
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: coalesce charging via percpu storage

This is a patch for coalescing access to res_counter at charging by percpu
caching.  At charge, memcg charges 64pages and remember it in percpu
cache.  Because it's cache, drain/flush if necessary.

This version uses public percpu area.
 2 benefits for using public percpu area.
 1. Sum of stocked charge in the system is limited to # of cpus
    not to the number of memcg. This shows better synchonization.
 2. drain code for flush/cpuhotplug is very easy (and quick)

The most important point of this patch is that we never touch res_counter
in fast path. The res_counter is system-wide shared counter which is modified
very frequently. We shouldn't touch it as far as we can for avoiding
false sharing.

On x86-64 8cpu server, I tested overheads of memcg at page fault by
running a program which does map/fault/unmap in a loop. Running
a task per a cpu by taskset and see sum of the number of page faults
in 60secs.

[without memcg config]
  40156968  page-faults              #      0.085 M/sec   ( +-   0.046% )
  27.67 cache-miss/faults

[root cgroup]
  36659599  page-faults              #      0.077 M/sec   ( +-   0.247% )
  31.58 cache miss/faults

[in a child cgroup]
  18444157  page-faults              #      0.039 M/sec   ( +-   0.133% )
  69.96 cache miss/faults

[ + coalescing uncharge patch]
  27133719  page-faults              #      0.057 M/sec   ( +-   0.155% )
  47.16 cache miss/faults

[ + coalescing uncharge patch + this patch ]
  34224709  page-faults              #      0.072 M/sec   ( +-   0.173% )
  34.69 cache miss/faults

Changelog (since Oct/2):
  - updated comments
  - replaced get_cpu_var() with __get_cpu_var() if possible.
  - removed mutex for system-wide drain. adds a counter instead of it.
  - removed CONFIG_HOTPLUG_CPU

Changelog (old):
  - rebased onto the latest mmotm
  - moved charge size check before __GFP_WAIT check for avoiding unnecesary
  - added asynchronous flush routine.
  - fixed bugs pointed out by Nishimura-san.

[akpm@linux-foundation.org: tweak comments]
[nishimura@mxp.nes.nec.co.jp: don't do INIT_WORK() repeatedly against the same work_struct]
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: coalesce uncharge during unmap/truncate

In massive parallel enviroment, res_counter can be a performance
bottleneck.  One strong techinque to reduce lock contention is reducing
calls by coalescing some amount of calls into one.

Considering charge/uncharge chatacteristic,
- charge is done one by one via demand-paging.
- uncharge is done by
- in chunk at munmap, truncate, exit, execve...
- one by one via vmscan/paging.

It seems we have a chance to coalesce uncharges for improving scalability
at unmap/truncation.

This patch is a for coalescing uncharge.  For avoiding scattering memcg's
structure to functions under /mm, this patch adds memcg batch uncharge
information to the task.  A reason for per-task batching is for making use
of caller's context information.  We do batched uncharge (deleyed
uncharge) when truncation/unmap occurs but do direct uncharge when
uncharge is called by memory reclaim (vmscan.c).

The degree of coalescing depends on callers
  - at invalidate/trucate... pagevec size
  - at unmap ....ZAP_BLOCK_SIZE
(memory itself will be freed in this degree.)
Then, we'll not coalescing too much.

On x86-64 8cpu server, I tested overheads of memcg at page fault by
running a program which does map/fault/unmap in a loop. Running
a task per a cpu by taskset and see sum of the number of page faults
in 60secs.

[without memcg config]
  40156968  page-faults              #      0.085 M/sec   ( +-   0.046% )
  27.67 cache-miss/faults
[root cgroup]
  36659599  page-faults              #      0.077 M/sec   ( +-   0.247% )
  31.58 miss/faults
[in a child cgroup]
  18444157  page-faults              #      0.039 M/sec   ( +-   0.133% )
  69.96 miss/faults
[child with this patch]
  27133719  page-faults              #      0.057 M/sec   ( +-   0.155% )
  47.16 miss/faults

We can see some amounts of improvement.
(root cgroup doesn't affected by this patch)
Another patch for "charge" will follow this and above will be improved more.

Changelog(since 2009/10/02):
 - renamed filed of memcg_batch (as pages to bytes, memsw to memsw_bytes)
 - some clean up and commentary/description updates.
 - added initialize code to copy_process(). (possible bug fix)

Changelog(old):
 - fixed !CONFIG_MEM_CGROUP case.
 - rebased onto the latest mmotm + softlimit fix patches.
 - unified patch for callers
 - added commetns.
 - make ->do_batch as bool.
 - removed css_get() at el. We don't need it.

Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: fix memory.memsw.usage_in_bytes for root cgroup

A memory cgroup has a memory.memsw.usage_in_bytes file.  It shows the sum
of the usage of pages and swapents in the cgroup.  Presently the root
cgroup's memsw.usage_in_bytes shows the wrong value - the number of
swapents are not added.

So take MEM_CGROUP_STAT_SWAPOUT into account.

Signed-off-by: Kirill A. Shutemov <kirill@shutemov.name>
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

seq_file: use proc_create() in documentation

Using create_proc_entry() + ->proc_fops assignment is racy because
->proc_fops will be NULL for some time, use proc_create() to avoid race.

Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Cc: Randy Dunlap <randy.dunlap@oracle.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

proc: remove docbook and example

Example is outdated, it still uses old ->read_proc interfaces and "fb"
example is plain racy.  There are better examples all over the tree.

Docbook itself says almost nothing about /proc and contain quite a number
of simply wrong facts, e.g.  device nodes support.  What it does is
describing at great length interface which are going to be removed.

There are Documentation/filesystems/seq_file.txt in exchange.

Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Acked-by: Erik Mouw <mouw@nl.linux.org>
Cc: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

doc: SubmitChecklist, add ioctls, remove OSDL reference

If a patch adds ioctls, then Documentation/ioctl/ioctl-number.txt
should also be updated.

Remove reference to the OSDL PLM build farm.

Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fatfs: use common time_to_tm in fat_time_unix2fat()

It is not necessary to write custom code for convert calendar time to
broken-down time.  time_to_tm() is more generic to do that.

Signed-off-by: Zhao Lei <zhaolei@cn.fujitsu.com>
Cc: OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Pavel Machek <pavel@ucw.cz>
Cc: Andi Kleen <andi@firstfloor.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hpfs: use bitmap_weight()

Use bitmap_weight instead of doing hweight32 for each 32bit in bitmap.

Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Cc: Mikulas Patocka <mikulas@artax.karlin.mff.cuni.cz>
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>

hpfs: use hweight32

Use hweight32 instead of counting for each bit

Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Cc: Mikulas Patocka <mikulas@artax.karlin.mff.cuni.cz>
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>

reiserfs: don't compile procfs.o at all if no support

* small define cleanup in header
* fix #ifdeffery in procfs.c via Kconfig

Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Cc: Jeff Mahoney <jeffm@suse.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

reiserfs: remove /proc/fs/reiserfs/version

/proc/fs/reiserfs/version is on the way of removing ->read_proc interface.
 It's empty however, so simply remove it instead of doing dummy
conversion.  It's hard to see what information userspace can extract from
empty file.

Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Cc: Jeff Mahoney <jeffm@suse.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ufs: NFS support

Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Acked-by: Evgeniy Dushistov <dushistov@mail.ru>
Cc: Christoph Hellwig <hch@lst.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ufs: pass qstr instead of dentry where necessary for NFS

Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Cc: Evgeniy Dushistov <dushistov@mail.ru>
Cc: Christoph Hellwig <hch@lst.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ext2: report metadata errors during fsync

When an IO error happens while writing metadata buffers, we should better
report it and call ext2_error since the filesystem is probably no longer
consistent.  Sometimes such IO errors happen while flushing thread does
background writeback, the buffer gets later evicted from memory, and thus
the only trace of the error remains as AS_EIO bit set in blockdevice's
mapping.  So we check this bit in ext2_fsync and report the error although
we cannot be really sure which buffer we failed to write.

Signed-off-by: Jan Kara <jack@suse.cz>
Cc: Chris Mason <chris.mason@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ext2: avoid WARN() messages when failing to write to the superblock

This fixes a common warning reported by kerneloops.org

[Kernel summit hacking hour]
[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

const: constify remaining pipe_buf_operations

Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

pnpbios: convert to seq_file

Convert code away from ->read_proc/->write_proc interfaces.  Switch to
proc_create()/proc_create_data() which make addition of proc entries
reliable wrt NULL ->proc_fops, NULL ->data and so on.

Problem with ->read_proc et al is described here commit
786d7e1612f0b0adb6046f19b906609e4fe8b1ba "Fix rmmod/read/write races in
/proc entries"

Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Cc: Adam Belay <abelay@mit.edu>
Cc: Bjorn Helgaas <bjorn.helgaas@hp.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

da850/omap-l138: add callback to control LCD panel power

Add the platform specific callback to control LCD panel and backlight
power.

Signed-off-by: Chaithrika U S <chaithrika@ti.com>
Acked-by: Kevin Hilman <khilman@deeprootsystems.com>
Cc: Krzysztof Helt <krzysztof.h1@poczta.fm>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

intelfb: fix setting of active pipe with LVDS displays

The intelfb driver sets color map depending on currently active pipe.
However, if an LVDS display is attached (like in laptop) the active pipe
variable is never set.  The default value is PIPE_A and can be wrong.  Set
up the pipe variable during driver initialization after hardware state was
read.

Also, the detection of the active display (and hence the pipe) is wrong.
The pipes are assigned to so called planes.  Both pipes are always enabled
on my laptop but only one plane is enabled (the plane A for the CRT or the
plane B for the LVDS).  Change active pipe detection code to take into
account a status of the plane assigned to each pipe.

The problem is visible in the 8 bpp mode if colors above 15 are used.  The
first 16 color entries are displayed correctly.

The graphics chip description is here (G45 vol. 3):
http://intellinuxgraphics.org/documentation.html

Addresses http://bugzilla.kernel.org/show_bug.cgi?id=13285

Signed-off-by: Krzysztof Helt <krzysztof.h1@wp.pl>
Cc: Michal Suchanek <hramrach@centrum.cz>
Cc: Dean Menezes <samanddeanus@yahoo.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

viafb: cosmetic cleanup of function integrated_lvds_enable()

A humble attempt to simplify the coding style to improve readability

Signed-off-by: Harald Welte <HaraldWelte@viatech.com>
Signed-off-by: Florian Tobias Schandinat <FlorianSchandinat@gmx.de>
Cc: Joseph Chan <JosephChan@via.com.tw>
Cc: Scott Fang <ScottFang@viatech.com.cn>
Cc: Krzysztof Helt <krzysztof.h1@poczta.fm>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

viafb: documentation update

We now support the VX855, and the VX800 is no longer unaccellerated.
viafb_video_dev was removed as it was useless.

Signed-off-by: Harald Welte <HaraldWelte@viatech.com>
Signed-off-by: Florian Tobias Schandinat <FlorianSchandinat@gmx.de>
Cc: Joseph Chan <JosephChan@via.com.tw>
Cc: Scott Fang <ScottFang@viatech.com.cn>
Cc: Krzysztof Helt <krzysztof.h1@poczta.fm>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

matroxfb: fix problems with display stability

Regression caused in 2.6.23 and then despite repeated requests never fixed
or dealt with (Petr promised to sort it in 2008 but seems to have
forgotten).

Enough is enough - remove the problem line that was added.  If it upsets
someone they've had two years to deal with it and at the very least it'll
rattle their cage and wake them up.

Addresses http://bugzilla.kernel.org/show_bug.cgi?id=9709

Signed-off-by: Alan Cox <alan@linux.intel.com>
Reported-by: Damon <account@bugzilla.kernel.org.juxtaposition.net>
Tested-by: Ruud van Melick <rvm1974@raketnet.nl>
Cc: Petr Vandrovec <VANDROVE@vc.cvut.cz>
Cc: Pekka Enberg <penberg@cs.helsinki.fi>
Cc: Paul A. Clarke <pc@us.ibm.com>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

davinci: fb: add framebuffer blank operation

Implement frame buffer blank operation feature for DA8xx/OMAP-L1xx driver.

Signed-off-by: Chaithrika U S <chaithrika@ti.com>
Cc: Kevin Hilman <khilman@deeprootsystems.com>
Cc: Krzysztof Helt <krzysztof.h1@poczta.fm>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

davinci: fb : add suspend/resume suuport for DA8xx/OMAP-L1xx fb driver

Suspend/resume support DA8xx/OMAP-L1xx frame buffer driver.  This feature
has been tested on DA850/OMAP-L138 EVM.  For the purpose of testing, the
patch series[1] which adds suspend support for DA850/OMAP-L138 SoC was
applied.

[1] http://patchwork.kernel.org/patch/60260/

Signed-off-by: Chaithrika U S <chaithrika@ti.com>
Cc: Kevin Hilman <khilman@deeprootsystems.com>
Cc: Krzysztof Helt <krzysztof.h1@poczta.fm>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

davinci: fb: update the driver in preparation for addition of power management features

Add a helper function to enable raster.  Also add one member in the
private data structure to track the current blank status, another function
pointer which takes in the platform specific callback function to control
panel power.

These updates will help in adding suspend/resume and frame buffer blank
operation features.

Signed-off-by: Chaithrika U S <chaithrika@ti.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

clps711xfb: convert to proc_fops

Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

drivers/video/via/viafbdev.c: fix oops with no /proc

Fixed a typo: missing *.  This would lead to a kernel oops if the kernel
was compiled without support for the /proc file system.

Found with a static checker.  Compile tested.

Signed-off-by: Dan Carpenter <error27@gmail.com>
Acked-by: Florian Tobias Schandinat <FlorianSchandinat@gmx.de>
Cc: Joseph Chan <JosephChan@via.com.tw>
Cc: Scott Fang <ScottFang@viatech.com.cn>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

sm501: implement acceleration features

This patch provides the acceleration entry points for the SM501
framebuffer driver.

This patch provides the sync, copyarea and fillrect entry points, using
the SM501's 2D acceleration engine to perform the operations in-chip
rather than across the bus.

Signed-off-by: Ben Dooks <ben@simtec.co.uk>
Signed-off-by: Simtec Linux Team <linux@simtec.co.uk>
Signed-off-by: Vincent Sanders <vince@simtec.co.uk>
Cc: Krzysztof Helt <krzysztof.h1@poczta.fm>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

sm501: fix use of old <asm/io.h> instead of <linux/io.h>

Fix the old style use of <asm/io.h> by replacing it with <linux/io.h>.

Signed-off-by: Ben Dooks <ben@simtec.co.uk>
Signed-off-by: Simtec Linux Team <linux@simtec.co.uk>
Cc: Vincent Sanders <vince@simtec.co.uk>
Cc: Krzysztof Helt <krzysztof.h1@poczta.fm>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

sm501: fix missing uses of resource_size()

There are several places in the SM501 fb driver that could do with using
resource_size() to calculate the size of a resource.

Also fix a bug where request_mem_region() is being passed one too few
bytes when requesting the register memory region, which was causing the
following in /proc/iomem:

 13e80000-13e8ffff : sm501-fb.0
   13e80000-13e8fffe : sm501-fb

fixed, this reads:

 13e80000-13e8ffff : sm501-fb.0
   13e80000-13e8ffff : sm501-fb

Signed-off-by: Ben Dooks <ben@simtec.co.uk>
Signed-off-by: Simtec Linux Team <linux@simtec.co.uk>
Cc: Vincent Sanders <vince@simtec.co.uk>
Cc: Krzysztof Helt <krzysztof.h1@poczta.fm>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

i810fb: fix stack exploding

Alan Cox has found that the i810fb function "uses a whopping 2.5K of stack".

Signed-off-by: Krzysztof Helt <krzysztof.h1@wp.pl>
Reported-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

lxfb: add copyright/license info

The GRUB folks requested copyright/license info for lxfb.h, so here it is.

Signed-off-by: Andres Salomon <dilinger@collabora.co.uk>
Cc: Jordan Crouse <jordan@cosmicpenguin.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

davinci: fb: add cpufreq support

Add cpufreq support for DA8xx/OMAP-L1xx frame buffer driver

Signed-off-by: Chaithrika U S <chaithrika@ti.com>
Cc: Krzysztof Helt <krzysztof.h1@poczta.fm>
Cc: Kevin Hilman <khilman@deeprootsystems.com>
Cc: Dave Jones <davej@codemonkey.org.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

davinci: fb: calculate the clock divider from pixel clock info

The clock divider value can be calculated from the pixel clock value for
the panel.  This gives more flexiblity to the driver to change the divider
value on the fly as in the case of cpufreq feature- support for which will
be added shortly.

Signed-off-by: Chaithrika U S <chaithrika@ti.com>
Cc: Sudhakar Rajashekhara <sudhakar.raj@ti.com>
Cc: Steve Chen <schen@mvista.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mb862xxfb: add acceleration support for Coral-P/Coral-PA. * imageblt * copyarea * fillrect

[akpm@linux-foundation.org: clean up comments, remove do-nothing tests, use min()]
[akpm@linux-foundation.org: fix Alexandr's email address]
[akpm@linux-foundation.org: export mb862xxfb_init_accel]
Signed-off-by: Valentin Sitdikov <valentin.sitdikov@siemens.com>
Cc: Krzysztof Helt <krzysztof.h1@poczta.fm>
Cc: Anatolij Gustschin <agust@denx.de>
Signed-off-by: Alexander Shishkin <virtuoso@slind.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

drivers/video: Move dereference after NULL test

If the NULL test on fbi is needed, then the dereference should be after the
NULL test.

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

// <smpl>
@match exists@
expression x, E;
identifier fld;
@@

* x->fld
  ... when != \(x = E\|&x\)
* x == NULL
// </smpl>

Signed-off-by: Julia Lawall <julia@diku.dk>
Cc: Krzysztof Helt <krzysztof.h1@poczta.fm>
Cc: Eric Miao <eric.y.miao@gmail.com>
Cc: Daniel Mack <daniel@caiaq.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fbdev: add palette register check to several drivers

Add check if palette register number is in correct range for few drivers
which miss it.  The regno value comes indirectly from user space.

Two drivers has converted check from BUG_ON() macro to just return an
error (non-zero value).

Signed-off-by: Krzysztof Helt <krzysztof.h1@wp.pl>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>