i386: cpu/transmeta.c: fix definition of USER686

The definition of USER686 is supposed to be a mask of feature bits,
not an OR of feature numbers!  It happened to work anyway on the only
processor affected, simply by pure coincidence.  Fix.

Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

md: improve partition detection in md array

md currently uses ->media_changed to make sure rescan_partitions
is call on md array after they are assembled.

However that doesn't happen until the array is opened, which is later
than some people would like.

So use blkdev_ioctl to do the rescan immediately that the
array has been assembled.

This means we can remove all the ->change infrastructure as it was only used
to trigger a partition rescan.

Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

md: allow reshape_position for md arrays to be set via sysfs

"reshape_position" records how much progress has been made on a "reshape"
(adding drives, changing layout or chunksize).

When it is set, the number of drives, layout and chunksize can have
two possible values, an old an a new.

So allow these different values to be visible, and allow both old and new to
be set: Set the old ones first, then the reshape_position, then the new
values.

Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

md: remove the slash from the name of a kmem_cache used by raid5

SLUB doesn't like slashes as it wants to use the cache name as the name of a
directory (or symlink) in sysfs.

Signed-off-by: Neil Brown <neilb@suse.de>
Acked-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

md: stop using csum_partial for checksum calculation in md

If CONFIG_NET is not selected, csum_partial is not exported, so md.ko cannot
use it.  We shouldn't really be using csum_partial anyway as it is an
internal-to-networking interface.

So replace it with C code to do the same thing.  Speed is not crucial here, so
something simple and correct is best.

Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

md: move test for whether level supports bitmap to correct place

We need to check for internal-consistency of superblock in load_super.
validate_super is for inter-device consistency.

With the test in the wrong place, a badly created array will confuse md rather
an produce sensible errors.

Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

md: cleanup: use seq_release_private() where appropriate

We can save some lines of code by using seq_release_private().

Signed-off-by: Martin Peschke <mp3@de.ibm.com>
Acked-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

drivers/md.c: Use ARRAY_SIZE macro when appropriate

Use ARRAY_SIZE macro already defined in kernel.h

Signed-off-by: Ahmed S. Darwish <darwish.07@gmail.com>
Acked-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

frame buffer: geforce 7300 gt

My geforce isn't supported by nvidia frame buffer.

/sbin/lspci
01:00.0 VGA compatible controller: nVidia Corporation Unknown device 02e2 (rev a2)

/usr/sbin/fbset -i

mode "1024x768-60"
    # D: 65.003 MHz, H: 48.365 kHz, V: 60.006 Hz
    geometry 1024 768 1024 32767 8
    timings 15384 160 24 29 3 136 6
    accel true
    rgba 8/0,8/0,8/0,0/0
endmode

Frame buffer device information:
    Name        : NV2e
    Address     : 0xe0000000
    Size        : 134217728
    Type        : PACKED PIXELS
    Visual      : PSEUDOCOLOR
    XPanStep    : 8
    YPanStep    : 1
    YWrapStep   : 0
    LineLength  : 1024
    MMIO Address: 0xf6000000
    MMIO Size   : 16777216
    Accelerator : Unknown (46)

Here is a patch for this problem.

Signed-off-by: Michal Piotrowski <michal.k.k.piotrowski@gmail.com>
Cc: "Antonino A. Daplas" <adaplas@pol.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fbdev: add support for AVR32

Provide framebuffer page protection flags and definitions of
fb_readl/fb_writel for AVR32.

Signed-off-by: Haavard Skinnemoen <hskinnemoen@atmel.com>
Cc: "Antonino A. Daplas" <adaplas@pol.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

svgalib: move fb_get_caps to svgalib

Move fb_get_caps() method to svgalib.c as svga_get_caps() so it can be used by
s3fb, arkfb and vt8623fb.

Signed-off-by: Antonino Daplas <adaplas@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

arkfb: new framebuffer driver for ARK Logic cards

This patch adds fbdev driver for graphics cards with ARK Logic 2000PV graphics
chip with ICS 5342 ramdac.

[adaplas@gmail.com: build fixes]
Signed-off-by: Ondrej Zajicek <santiago@crfreenet.org>
Signed-off-by: Antonino Daplas <adaplas@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

vt8623fb: new framebuffer driver for VIA VT8623

This patch adds fbdev driver for graphics core in VIA VT8623

[adaplas@gmail.com: build fixes]
Signed-off-by: Ondrej Zajicek <santiago@crfreenet.org>
Signed-off-by: Antonino Daplas <adaplas@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

i386 mmzone: use __maybe_unused

Replace automatic variable instances of __attribute__ ((unused)) with
__maybe_unused.

Cc: Andy Whitcroft <apw@shadowen.org>
Signed-off-by: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

i386: voyager: use __maybe_unused

Replace automatic variable instances of __attribute__((unused)) with
__maybe_unused in mca_nmi_hook().

Cc: James Bottomley <James.Bottomley@HansenPartnership.com>
Signed-off-by: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

sh: dma: use __maybe_unused

There is no such thing as labeling a variable as __attribute__((used)).  Since
ts_shift is not referenced in inline assembly, we assume that we're simply
suppressing a warning here if the variable is declared but unreferenced.

Cc: Paul Mundt <lethal@linux-sh.org>
Signed-off-by: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

i386 pci: use __maybe_unused

Use the new macro here

Cc: Andi Kleen <ak@suse.de>
Signed-off-by: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

compiler: introduce __used and __maybe_unused

__used is defined to be __attribute__((unused)) for all pre-3.3 gcc
compilers to suppress warnings for unused functions because perhaps they
are referenced only in inline assembly.  It is defined to be
__attribute__((used)) for gcc 3.3 and later so that the code is still
emitted for such functions.

__maybe_unused is defined to be __attribute__((unused)) for both function
and variable use if it could possibly be unreferenced due to the evaluation
of preprocessor macros.  Function prototypes shall be marked with
__maybe_unused if the actual definition of the function is dependant on
preprocessor macros.

No update to compiler-intel.h is necessary because ICC supports both
__attribute__((used)) and __attribute__((unused)) as specified by the gcc
manual.

__attribute_used__ is deprecated and will be removed once all current
code is converted to using __used.

Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Adrian Bunk <bunk@stusta.de>
Signed-off-by: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

rename thread_info to stack

This finally renames the thread_info field in task structure to stack, so that
the assumptions about this field are gone and archs have more freedom about
placing the thread_info structure.

Nonbroken archs which have a proper thread pointer can do the access to both
current thread and task structure via a single pointer.

It'll allow for a few more cleanups of the fork code, from which e.g.  ia64
could benefit.

Signed-off-by: Roman Zippel <zippel@linux-m68k.org>
[akpm@linux-foundation.org: build fix]
Cc: Richard Henderson <rth@twiddle.net>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Cc: Russell King <rmk@arm.linux.org.uk>
Cc: Ian Molton <spyro@f2s.com>
Cc: Haavard Skinnemoen <hskinnemoen@atmel.com>
Cc: Mikael Starvik <starvik@axis.com>
Cc: David Howells <dhowells@redhat.com>
Cc: Yoshinori Sato <ysato@users.sourceforge.jp>
Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: Hirokazu Takata <takata@linux-m32r.org>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Roman Zippel <zippel@linux-m68k.org>
Cc: Greg Ungerer <gerg@uclinux.org>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Paul Mundt <lethal@linux-sh.org>
Cc: Kazumoto Kojima <kkojima@rr.iij4u.or.jp>
Cc: Richard Curnow <rc@rc0.org.uk>
Cc: William Lee Irwin III <wli@holomorphy.com>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Jeff Dike <jdike@addtoit.com>
Cc: Paolo 'Blaisorblade' Giarrusso <blaisorblade@yahoo.it>
Cc: Miles Bader <uclinux-v850@lsi.nec.co.jp>
Cc: Andi Kleen <ak@muc.de>
Cc: Chris Zankel <chris@zankel.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

wrap access to thread_info

Recently a few direct accesses to the thread_info in the task structure snuck
back, so this wraps them with the appropriate wrapper.

Signed-off-by: Roman Zippel <zippel@linux-m68k.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Allow arch to initialize arch field of the module structure

This will later allow an arch to add module specific information via linker
generated tables instead of poking directly in the module object structure.

Signed-off-by: Roman Zippel <zippel@linux-m68k.org>
Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

clocksource: fix resume logic

We need to make sure that the clocksources are resumed, when timekeeping is
resumed.  The current resume logic does not guarantee this.

Add a resume function pointer to the clocksource struct, so clocksource
drivers which need to reinitialize the clocksource can provide a resume
function.

Add a resume function, which calls the maybe available clocksource resume
functions and resets the watchdog function, so a stable TSC can be used
accross suspend/resume.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: john stultz <johnstul@us.ibm.com>
Cc: Andi Kleen <ak@suse.de>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Move remote node draining out of slab allocators

Currently the slab allocators contain callbacks into the page allocator to
perform the draining of pagesets on remote nodes.  This requires SLUB to have
a whole subsystem in order to be compatible with SLAB.  Moving node draining
out of the slab allocators avoids a section of code in SLUB.

Move the node draining so that is is done when the vm statistics are updated.
At that point we are already touching all the cachelines with the pagesets of
a processor.

Add a expire counter there.  If we have to update per zone or global vm
statistics then assume that the pageset will require subsequent draining.

The expire counter will be decremented on each vm stats update pass until it
reaches zero.  Then we will drain one batch from the pageset.  The draining
will cause vm counter updates which will then cause another expiration until
the pcp is empty.  So we will drain a batch every 3 seconds.

Note that remote node draining is a somewhat esoteric feature that is required
on large NUMA systems because otherwise significant portions of system memory
can become trapped in pcp queues.  The number of pcp is determined by the
number of processors and nodes in a system.  A system with 4 processors and 2
nodes has 8 pcps which is okay.  But a system with 1024 processors and 512
nodes has 512k pcps with a high potential for large amount of memory being
caught in them.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Make vm statistics update interval configurable

Make it configurable.  Code in mm makes the vm statistics intervals
independent from the cache reaper use that opportunity to make it
configurable.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

vmstat: use our own timer events

vmstat is currently using the cache reaper to periodically bring the
statistics up to date.  The cache reaper does only exists in SLUB as a way to
provide compatibility with SLAB.  This patch removes the vmstat calls from the
slab allocators and provides its own handling.

The advantage is also that we can use a different frequency for the updates.
Refreshing vm stats is a pretty fast job so we can run this every second and
stagger this by only one tick.  This will lead to some overlap in large
systems.  F.e a system running at 250 HZ with 1024 processors will have 4 vm
updates occurring at once.

However, the vm stats update only accesses per node information.  It is only
necessary to stagger the vm statistics updates per processor in each node.  Vm
counter updates occurring on distant nodes will not cause cacheline
contention.

We could implement an alternate approach that runs the first processor on each
node at the second and then each of the other processor on a node on a
subsequent tick.  That may be useful to keep a large amount of the second free
of timer activity.  Maybe the timer folks will have some feedback on this one?

[jirislaby@gmail.com: add missing break]
Cc: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Jiri Slaby <jirislaby@gmail.com>
Cc: Oleg Nesterov <oleg@tv-sign.ru>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

microcode: use suspend-related CPU hotplug notifications

Make the microcode driver use the suspend-related CPU hotplug notifications
to handle the CPU hotplug events occuring during system-wide suspend and
resume transitions.  Remove the global variable suspend_cpu_hotplug
previously used for this purpose.

Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Cc: Gautham R Shenoy <ego@in.ibm.com>
Cc: Pavel Machek <pavel@ucw.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Add suspend-related notifications for CPU hotplug

Since nonboot CPUs are now disabled after tasks and devices have been
frozen and the CPU hotplug infrastructure is used for this purpose, we need
special CPU hotplug notifications that will help the CPU-hotplug-aware
subsystems distinguish normal CPU hotplug events from CPU hotplug events
related to a system-wide suspend or resume operation in progress.  This
patch introduces such notifications and causes them to be used during
suspend and resume transitions.  It also changes all of the
CPU-hotplug-aware subsystems to take these notifications into consideration
(for now they are handled in the same way as the corresponding "normal"
ones).

[oleg@tv-sign.ru: cleanups]
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Cc: Gautham R Shenoy <ego@in.ibm.com>
Cc: Pavel Machek <pavel@ucw.cz>
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fs: deprecate memclear_highpage_flush

Now that all the in-tree users are converted over to zero_user_page(),
deprecate the old memclear_highpage_flush() call.

Signed-off-by: Nate Diller <nate.diller@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

reiserfs: use zero_user_page

Use zero_user_page() instead of open-coding it.

Signed-off-by: Nate Diller <nate.diller@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ext3: use zero_user_page

Use zero_user_page() instead of open-coding it.

Signed-off-by: Nate Diller <nate.diller@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

affs: use zero_user_page

Use zero_user_page() instead of open-coding it.

Signed-off-by: Nate Diller <nate.diller@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fs: convert core functions to zero_user_page

It's very common for file systems to need to zero part or all of a page,
the simplist way is just to use kmap_atomic() and memset().  There's
actually a library function in include/linux/highmem.h that does exactly
that, but it's confusingly named memclear_highpage_flush(), which is
descriptive of *how* it does the work rather than what the *purpose* is.
So this patchset renames the function to zero_user_page(), and calls it
from the various places that currently open code it.

This first patch introduces the new function call, and converts all the
core kernel callsites, both the open-coded ones and the old
memclear_highpage_flush() ones.  Following this patch is a series of
conversions for each file system individually, per AKPM, and finally a
patch deprecating the old call.  The diffstat below shows the entire
patchset.

[akpm@linux-foundation.org: fix a few things]
Signed-off-by: Nate Diller <nate.diller@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

timer: parenthesis fix in tbase_get_deferrable() etc

Signed-off-by: Jarek Poplawski <jarkao2@o2.pl>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

FUTEX: new PRIVATE futexes

  Analysis of current linux futex code :
  --------------------------------------

A central hash table futex_queues[] holds all contexts (futex_q) of waiting
threads.

Each futex_wait()/futex_wait() has to obtain a spinlock on a hash slot to
perform lookups or insert/deletion of a futex_q.

When a futex_wait() is done, calling thread has to :

1) - Obtain a read lock on mmap_sem to be able to validate the user pointer
     (calling find_vma()). This validation tells us if the futex uses
     an inode based store (mapped file), or mm based store (anonymous mem)

2) - compute a hash key

3) - Atomic increment of reference counter on an inode or a mm_struct

4) - lock part of futex_queues[] hash table

5) - perform the test on value of futex.
(rollback is value != expected_value, returns EWOULDBLOCK)
(various loops if test triggers mm faults)

6) queue the context into hash table, release the lock got in 4)

7) - release the read_lock on mmap_sem

   <block>

8) Eventually unqueue the context (but rarely, as this part  may be done
   by the futex_wake())

Futexes were designed to improve scalability but current implementation has
various problems :

- Central hashtable :

  This means scalability problems if many processes/threads want to use
  futexes at the same time.
  This means NUMA unbalance because this hashtable is located on one node.

- Using mmap_sem on every futex() syscall :

  Even if mmap_sem is a rw_semaphore, up_read()/down_read() are doing atomic
  ops on mmap_sem, dirtying cache line :
    - lot of cache line ping pongs on SMP configurations.

  mmap_sem is also extensively used by mm code (page faults, mmap()/munmap())
  Highly threaded processes might suffer from mmap_sem contention.

  mmap_sem is also used by oprofile code. Enabling oprofile hurts threaded
  programs because of contention on the mmap_sem cache line.

- Using an atomic_inc()/atomic_dec() on inode ref counter or mm ref counter:
  It's also a cache line ping pong on SMP. It also increases mmap_sem hold time
  because of cache misses.

Most of these scalability problems come from the fact that futexes are in
one global namespace.  As we use a central hash table, we must make sure
they are all using the same reference (given by the mm subsystem).  We
chose to force all futexes be 'shared'.  This has a cost.

But fact is POSIX defined PRIVATE and SHARED, allowing clear separation,
and optimal performance if carefuly implemented.  Time has come for linux
to have better threading performance.

The goal is to permit new futex commands to avoid :
 - Taking the mmap_sem semaphore, conflicting with other subsystems.
 - Modifying a ref_count on mm or an inode, still conflicting with mm or fs.

This is possible because, for one process using PTHREAD_PROCESS_PRIVATE
futexes, we only need to distinguish futexes by their virtual address, no
matter the underlying mm storage is.

If glibc wants to exploit this new infrastructure, it should use new
_PRIVATE futex subcommands for PTHREAD_PROCESS_PRIVATE futexes.  And be
prepared to fallback on old subcommands for old kernels.  Using one global
variable with the FUTEX_PRIVATE_FLAG or 0 value should be OK.

PTHREAD_PROCESS_SHARED futexes should still use the old subcommands.

Compatibility with old applications is preserved, they still hit the
scalability problems, but new applications can fly :)

Note : the same SHARED futex (mapped on a file) can be used by old binaries
*and* new binaries, because both binaries will use the old subcommands.

Note : Vast majority of futexes should be using PROCESS_PRIVATE semantic,
as this is the default semantic. Almost all applications should benefit
of this changes (new kernel and updated libc)

Some bench results on a Pentium M 1.6 GHz (SMP kernel on a UP machine)

/* calling futex_wait(addr, value) with value != *addr */
433 cycles per futex(FUTEX_WAIT) call (mixing 2 futexes)
424 cycles per futex(FUTEX_WAIT) call (using one futex)
334 cycles per futex(FUTEX_WAIT_PRIVATE) call (mixing 2 futexes)
334 cycles per futex(FUTEX_WAIT_PRIVATE) call (using one futex)
For reference :
187 cycles per getppid() call
188 cycles per umask() call
181 cycles per ni_syscall() call

Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Pierre Peiffer <pierre.peiffer@bull.net>
Cc: "Ulrich Drepper" <drepper@gmail.com>
Cc: "Nick Piggin" <nickpiggin@yahoo.com.au>
Cc: "Ingo Molnar" <mingo@elte.hu>
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>

futex_requeue_pi optimization

This patch provides the futex_requeue_pi functionality, which allows some
threads waiting on a normal futex to be requeued on the wait-queue of a
PI-futex.

This provides an optimization, already used for (normal) futexes, to be used
with the PI-futexes.

This optimization is currently used by the glibc in pthread_broadcast, when
using "normal" mutexes.  With futex_requeue_pi, it can be used with
PRIO_INHERIT mutexes too.

Signed-off-by: Pierre Peiffer <pierre.peiffer@bull.net>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Ulrich Drepper <drepper@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Make futex_wait() use an hrtimer for timeout

This patch modifies futex_wait() to use an hrtimer + schedule() in place of
schedule_timeout().

schedule_timeout() is tick based, therefore the timeout granularity is the
tick (1 ms, 4 ms or 10 ms depending on HZ).  By using a high resolution timer
for timeout wakeup, we can attain a much finer timeout granularity (in the
microsecond range).  This parallels what is already done for futex_lock_pi().

The timeout passed to the syscall is no longer converted to jiffies and is
therefore passed to do_futex() and futex_wait() as an absolute ktime_t
therefore keeping nanosecond resolution.

Also this removes the need to pass the nanoseconds timeout part to
futex_lock_pi() in val2.

In futex_wait(), if there is no timeout then a regular schedule() is
performed.  Otherwise, an hrtimer is fired before schedule() is called.

[akpm@linux-foundation.org: fix `make headers_check']
Signed-off-by: Sebastien Dugue <sebastien.dugue@bull.net>
Signed-off-by: Pierre Peiffer <pierre.peiffer@bull.net>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Ulrich Drepper <drepper@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

futex priority based wakeup

Today, all threads waiting for a given futex are woken in FIFO order (first
waiter woken first) instead of priority order.

This patch makes use of plist (pirotity ordered lists) instead of simple list
in futex_hash_bucket.

All non-RT threads are stored with priority MAX_RT_PRIO, causing them to be
woken last, in FIFO order (RT-threads are woken first, in priority order).

Signed-off-by: Sebastien Dugue <sebastien.dugue@bull.net>
Signed-off-by: Pierre Peiffer <pierre.peiffer@bull.net>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Ulrich Drepper <drepper@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

declare struct ktime

Some smarty went and inflicted ktime_t as a typedef upon us, so we cannot
forward declare it.

Create a new `union ktime', map ktime_t onto that.  Now we need to kill off
this ktime_t thing.

Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: john stultz <johnstul@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

aio is unlikely

Stick an unlikely() around is_aio(): I assert that most IO is synchronous.

Cc: Suparna Bhattacharya <suparna@in.ibm.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Benjamin LaHaise <bcrl@kvack.org>
Cc: Zach Brown <zach.brown@oracle.com>
Cc: Ulrich Drepper <drepper@redhat.com>
Cc: Christoph Hellwig <hch@lst.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

knfsd: avoid Oops if buggy userspace performs confusing filehandle->dentry mapping

When a lookup request arrives, nfsd uses information provided by userspace
(mountd) to find the right filesystem.

It then assumes that the same filehandle type as the incoming filehandle can
be used to create an outgoing filehandle.

However if mountd is buggy, or maybe just being creative, the filesystem may
not support that filesystem type, and the kernel could oops, particularly if
'ex_uuid' is NULL but a FSID_UUID* filehandle type is used.

So add some proper checking that the fsid version/type from the incoming
filehandle is actually supportable, and ignore that information if it isn't
supportable.

Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

knfsd: various nfsd xdr cleanups

1/ decode_sattr and decode_sattr3 never return NULL, so remove
   several checks for that. ditto for xdr_decode_hyper.

2/ replace some open coded XDR_QUADLEN calls with calls to
   XDR_QUADLEN

3/ in decode_writeargs, simply an 'if' to use a single
   calculation.
   .page_len is the length of that part of the packet that did
   not fit in the first page (the head).
   So the length of the data part is the remainder of the
   head, plus page_len.

3/ other minor cleanups.

Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

knfsd: trivial makefile cleanup

kbuild directly interprets <modulename>-y as objects to build into a module,
no need to assign it to the old foo-objs variable.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

knfsd: simplify a 'while' condition in svcsock.c

This while loop has an overly complex condition, which performs a couple of
assignments.  This hurts readability.

We don't really need a loop at all.  We can just return -EAGAIN and (providing
we set SK_DATA), the function will be called again.

So discard the loop, make the complex conditional become a few clear function
calls, and hopefully improve readability.

Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

knfsd: rpcgss: RPC_GSS_PROC_ DESTROY request will get a bad rpc

If I send a RPC_GSS_PROC_DESTROY message to NFSv4 server, it will reply with a
bad rpc reply which lacks an authentication verifier.  Maybe this patch is
needed.

Send/recv packets as following:

send:

RemoteProcedureCall
    xid
    rpcvers = 2
    prog = 100003
    vers = 4
    proc = 0
    cred = AUTH_GSS
        version = 1
        gss_proc = 3 (RPCSEC_GSS_DESTROY)
        service  = 1 (RPC_GSS_SVC_NONE)
    verf = AUTH_GSS
        checksum

reply:

RemoteProcedureReply
    xid
    msg_type
    reply_stat
    accepted_reply

Signed-off-by: Wei Yongjun <yjwei@cn.fujitsu.com>
Signed-off-by: "J. Bruce Fields" <bfields@citi.umich.edu>
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

knfsd: fix resource leak resulting in module refcount leak for rpcsec_gss_krb5.ko

I have been investigating a module reference count leak on the server for
rpcsec_gss_krb5.ko.  It turns out the problem is a reference count leak for
the security context in net/sunrpc/auth_gss/svcauth_gss.c.

The problem is that gss_write_init_verf() calls gss_svc_searchbyctx() which
does a rsc_lookup() but never releases the reference to the context.  There is
another issue that rpc.svcgssd sets an "end of time" expiration for the
context

By adding a cache_put() call in gss_svc_searchbyctx(), and setting an
expiration timeout in the downcall, cache_clean() does clean up the context
and the module reference count now goes to zero after unmount.

I also verified that if the context expires and then the client makes a new
request, a new context is established.

Here is the patch to fix the kernel, I will start a separate thread to discuss
what expiration time should be set by rpc.svcgssd.

Acked-by: "J. Bruce Fields" <bfields@citi.umich.edu>
Signed-off-by: Frank Filz <ffilzlnx@us.ibm.com>
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

knfsd: rpc: fix server-side wrapping of krb5i replies

It's not necessarily correct to assume that the xdr_buf used to hold the
server's reply must have page data whenever it has tail data.

And there's no need for us to deal with that case separately anyway.

Acked-by: "J. Bruce Fields" <bfields@citi.umich.edu>
Signed-off-by: Neil Brown <neilb@suse.de>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

knfsd: avoid use of unitialised variables on error path when nfs exports

We need to zero various parts of 'exp' before any 'goto out', otherwise when
we go to free the contents...  we die.

Signed-off-by: Neil Brown <neilb@suse.de>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

sunrpc: fix error path in module_init

register_rpc_pipefs() needs to clean up rpc_inode_cache
by kmem_cache_destroy() on register_filesystem() failure.

init_sunrpc() needs to unregister rpc_pipe_fs by unregister_rpc_pipefs()
when rpc_init_mempool() returns error.

Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Cc: Neil Brown <neilb@suse.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

RPC: add wrapper for svc_reserve to account for checksum

When the kernel calls svc_reserve to downsize the expected size of an RPC
reply, it fails to account for the possibility of a checksum at the end of
the packet.  If a client mounts a NFSv2/3 with sec=krb5i/p, and does I/O
then you'll generally see messages similar to this in the server's ring
buffer:

RPC request reserved 164 but used 208

While I was never able to verify it, I suspect that this problem is also
the root cause of some oopses I've seen under these conditions:

https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=227726

This is probably also a problem for other sec= types and for NFSv4.  The
large reserved size for NFSv4 compound packets seems to generally paper
over the problem, however.

This patch adds a wrapper for svc_reserve that accounts for the possibility
of a checksum.  It also fixes up the appropriate callers of svc_reserve to
call the wrapper.  For now, it just uses a hardcoded value that I
determined via testing.  That value may need to be revised upward as things
change, or we may want to eventually add a new auth_op that attempts to
calculate this somehow.

Unfortunately, there doesn't seem to be a good way to reliably determine
the expected checksum length prior to actually calculating it, particularly
with schemes like spkm3.

Signed-off-by: Jeff Layton <jlayton@redhat.com>
Acked-by: Neil Brown <neilb@suse.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Acked-by: J. Bruce Fields <bfields@citi.umich.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

nfsd/nfs4state: remove unnecessary daemonize call

Acked-by: Neil Brown <neilb@suse.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

knfsd: rename sk_defer_lock to sk_lock

Now that sk_defer_lock protects two different things, make the name more
generic.

Also don't bother with disabling _bh as the lock is only ever taken from
process context.

Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

The NFSv2/NFSv3 server does not handle zero length WRITE requests correctly

The NFSv2 and NFSv3 servers do not handle WRITE requests for 0 bytes
correctly.  The specifications indicate that the server should accept the
request, but it should mostly turn into a no-op.  Currently, the server
will return an XDR decode error, which it should not.

Attached is a patch which addresses this issue.  It also adds some boundary
checking to ensure that the request contains as much data as was requested
to be written.  It also correctly handles an NFSv3 request which requests
to write more data than the server has stated that it is prepared to
handle.  Previously, there was some support which looked like it should
work, but wasn't quite right.

Signed-off-by: Peter Staubach <staubach@redhat.com>
Acked-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

remove nfs4_acl_add_ace()

nfs4_acl_add_ace() can now be removed.

Signed-off-by: Adrian Bunk <bunk@stusta.de>
Acked-by: Neil Brown <neilb@cse.unsw.edu.au>
Acked-by: J. Bruce Fields <bfields@citi.umich.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

make cancel_rearming_delayed_work() reliable

Thanks to Jarek Poplawski for the ideas and for spotting the bug in the
initial draft patch.

cancel_rearming_delayed_work() currently has many limitations, because it
requires that dwork always re-arms itself via queue_delayed_work().  So it
hangs forever if dwork doesn't do this, or cancel_rearming_delayed_work/
cancel_delayed_work was already called.  It uses flush_workqueue() in a
loop, so it can't be used if workqueue was freezed, and it is potentially
live- lockable on busy system if delay is small.

With this patch cancel_rearming_delayed_work() doesn't make any assumptions
about dwork, it can re-arm itself via queue_delayed_work(), or
queue_work(), or do nothing.

As a "side effect", cancel_work_sync() was changed to handle re-arming works
as well.

Disadvantages:

- this patch adds wmb() to insert_work().

- slowdowns the fast path (when del_timer() succeeds on entry) of
  cancel_rearming_delayed_work(), because wait_on_work() is called
  unconditionally. In that case, compared to the old version, we are
  doing "unneeded" lock/unlock for each online CPU.

  On the other hand, this means we don't need to use cancel_work_sync()
  after cancel_rearming_delayed_work().

- complicates the code (.text grows by 130 bytes).

[akpm@linux-foundation.org: fix speling]
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: David Chinner <dgc@sgi.com>
Cc: David Howells <dhowells@redhat.com>
Cc: Gautham Shenoy <ego@in.ibm.com>
Acked-by: Jarek Poplawski <jarkao2@o2.pl>
Cc: Srivatsa Vaddagiri <vatsa@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Remove kthread_bind() call from _cpu_down()

We are anyway kthread_stop()ping other per-cpu kernel threads after
move_task_off_dead_cpu(), so we can do it with the stop_machine_run thread
as well.

I just checked with Vatsa if there was any subtle reason why they
had put in the kthread_bind() in cpu.c. Vatsa cannot seem to recollect
any and I can't see any. So let us just remove the kthread_bind.

Signed-off-by: Gautham R Shenoy <ego@in.ibm.com>
Cc: Oleg Nesterov <oleg@tv-sign.ru>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: "Rafael J. Wysocki" <rjw@sisk.pl>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

change kernel threads to ignore signals instead of blocking them

Currently kernel threads use sigprocmask(SIG_BLOCK) to protect against
signals.  This doesn't prevent the signal delivery, this only blocks
signal_wake_up().  Every "killall -33 kthreadd" means a "struct siginfo"
leak.

Change kthreadd_setup() to set all handlers to SIG_IGN instead of blocking
them (make a new helper ignore_signals() for that).  If the kernel thread
needs some signal, it should use allow_signal() anyway, and in that case it
should not use CLONE_SIGHAND.

Note that we can't change daemonize() (should die!) in the same way,
because it can be used along with CLONE_SIGHAND.  This means that
allow_signal() still should unblock the signal to work correctly with
daemonize()ed threads.

However, disallow_signal() doesn't block the signal any longer but ignores
it.

NOTE: with or without this patch the kernel threads are not protected from
handle_stop_signal(), this seems harmless, but not good.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Acked-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

worker_thread: don't play with SIGCHLD and numa policy

worker_thread() inherits ignored SIGCHLD and numa_default_policy() from its
parent, kthreadd.  No need to setup this again.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Acked-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

wait_for_helper: remove unneeded do_sigaction()

allow_signal(SIGCHLD) does all necessary job, no need to call do_sigaction()
prior to.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Rusty Russell <rusty@rustcorp.com.au>
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>

usbatm_heavy_init: don't use CLONE_SIGHAND

usbatm_do_heavy_init() calls allow_signal() which plays with parent process's
->sighand.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Acked-by: Duncan Sands <duncan.sands@free.fr>
Cc: Greg KH <greg@kroah.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Change reparent_to_init to reparent_to_kthreadd

When a kernel thread calls daemonize, instead of reparenting the thread to
init reparent the thread to kthreadd next to the threads created by
kthread_create.

This is really just a stop gap until daemonize goes away, but it does
ensure no kernel threads are under init and they are all in one place that
is easy to find.

Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Cc: Oleg Nesterov <oleg@tv-sign.ru>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

kthread: don't depend on work queues

Currently there is a circular reference between work queue initialization
and kthread initialization.  This prevents the kthread infrastructure from
initializing until after work queues have been initialized.

We want the properties of tasks created with kthread_create to be as close
as possible to the init_task and to not be contaminated by user processes.
The later we start our kthreadd that creates these tasks the harder it is
to avoid contamination from user processes and the more of a mess we have
to clean up because the defaults have changed on us.

So this patch modifies the kthread support to not use work queues but to
instead use a simple list of structures, and to have kthreadd start from
init_task immediately after our kernel thread that execs /sbin/init.

By being a true child of init_task we only have to change those process
settings that we want to have different from init_task, such as our process
name, the cpus that are allowed, blocking all signals and setting SIGCHLD
to SIG_IGN so that all of our children are reaped automatically.

By being a true child of init_task we also naturally get our ppid set to 0
and do not wind up as a child of PID == 1.  Ensuring that tasks generated
by kthread_create will not slow down the functioning of the wait family of
functions.

[akpm@linux-foundation.org: use interruptible sleeps]
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Cc: Oleg Nesterov <oleg@tv-sign.ru>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

____call_usermodehelper: don't flush_signals()

____call_usermodehelper() has no reason for flush_signals().  It is a fresh
forked process which is going to exec a user-space application or exit on
failure.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
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>

unify flush_work/flush_work_keventd and rename it to cancel_work_sync

flush_work(wq, work) doesn't need the first parameter, we can use cwq->wq
(this was possible from the very beginnig, I missed this).  So we can unify
flush_work_keventd and flush_work.

Also, rename flush_work() to cancel_work_sync() and fix all callers.
Perhaps this is not the best name, but "flush_work" is really bad.

(akpm: this is why the earlier patches bypassed maintainers)

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Jeff Garzik <jeff@garzik.org>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Jens Axboe <jens.axboe@oracle.com>
Cc: Tejun Heo <htejun@gmail.com>
Cc: Auke Kok <auke-jan.h.kok@intel.com>,
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

slab: shut down cache_reaper when cpu goes down

Shutdown the cache_reaper if the cpu is brought down and set the
cache_reap.func to NULL.  Otherwise hotplug shuts down the reaper for good.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Cc: Oleg Nesterov <oleg@tv-sign.ru>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

zap_other_threads: remove unneeded ->exit_signal change

We already depend on fact that all sub-threads have ->exit_signal == -1, no
need to set it in zap_other_threads().

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

worker_thread: fix racy try_to_freeze() usage

worker_thread() can miss freeze_process()->signal_wake_up() if it happens
between try_to_freeze() and prepare_to_wait().  We should check freezing()
before entering schedule().

This race was introduced by me in

[PATCH 1/1] workqueue: don't migrate pending works from the dead CPU

Looks like mm/vmscan.c:kswapd() has the same race.

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

worker_thread: don't play with signals

worker_thread() doesn't need to "Block and flush all signals", this was
already done by its caller, kthread().

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

workqueue: kill NOAUTOREL works

We don't have any users, and it is not so trivial to use NOAUTOREL works
correctly.  It is better to simplify API.

Delete NOAUTOREL support and rename work_release to work_clear_pending to
avoid a confusion.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Acked-by: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ipvs: flush defense_work before module unload

net/ipv4/ipvs/ip_vs_core.c

module_exit
    ip_vs_cleanup
ip_vs_control_cleanup
    cancel_rearming_delayed_work
// done

This is unsafe.  The module may be unloaded and the memory may be freed
while defense_work's handler is still running/preempted.

Do flush_work(&defense_work.work) after cancel_rearming_delayed_work().

Alternatively, we could add flush_work() to cancel_rearming_delayed_work(),
but note that we can't change cancel_delayed_work() in the same manner
because it may be called from atomic context.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
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>

make cancel_rearming_delayed_work() work on any workqueue, not just keventd_wq

cancel_rearming_delayed_workqueue(wq, dwork) doesn't need the first
parameter.  We don't hang on un-queued dwork any longer, and work->data
doesn't change its type.  This means we can always figure out "wq" from
dwork when it is needed.

Remove this parameter, and rename the function to
cancel_rearming_delayed_work().  Re-create an inline "obsolete"
cancel_rearming_delayed_workqueue(wq) which just calls
cancel_rearming_delayed_work().

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

workqueue: introduce wq_per_cpu() helper

Cleanup.  A number of per_cpu_ptr(wq->cpu_wq, cpu) users have to check that
cpu is valid for this wq.  Make a simple helper.

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

unify queue_delayed_work() and queue_delayed_work_on()

Change queue_delayed_work() to use queue_delayed_work_on() to avoid the code
duplication (saves 133 bytes).

Q: queue_delayed_work() enqueues &dwork->work directly when delay == 0, why?

[jirislaby@gmail.com: oops fix]
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Signed-off-by: Jiri Slaby <jirislaby@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

make queue_delayed_work() friendly to flush_fork()

Currently typeof(delayed_work->work.data) is

"struct workqueue_struct" when the timer is pending

"struct cpu_workqueue_struct" whe the work is queued

This makes impossible to use flush_fork(delayed_work->work) in addition
to cancel_delayed_work/cancel_rearming_delayed_work, not good.

Change queue_delayed_work/delayed_work_timer_fn to use cwq, not wq. This
complicates (and uglifies) these functions a little bit, but alows us to
use flush_fork(dwork) and imho makes the whole code more consistent.

Also, document the fact that cancel_rearming_delayed_work() doesn't garantee
the completion of work->func() upon return.

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

workqueues: shift kthread_bind() from CPU_UP_PREPARE to CPU_ONLINE

CPU_UP_PREPARE binds cwq->thread to the new CPU.  So CPU_UP_CANCELED tries to
wake up the task which is bound to the failed CPU.

With this patch we don't bind cwq->thread until CPU becomes online.  The first
wake_up() after kthread_create() is a bit special, make a simple helper for
that.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Gautham R Shenoy <ego@in.ibm.com>
Cc: "Rafael J. Wysocki" <rjw@sisk.pl>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

workqueue: make init_workqueues() __init

The only caller of init_workqueues() is do_basic_setup().

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

workqueue: introduce workqueue_struct->singlethread

Add explicit workqueue_struct->singlethread flag.  This lessens .text a
little, but most importantly this allows us to manipulate wq->list without
changine the meaning of is_single_threaded().

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

workqueue: introduce cpu_singlethread_map

The code like

if (is_single_threaded(wq))
do_something(singlethread_cpu);
else {
for_each_cpu_mask(cpu, cpu_populated_map)
do_something(cpu);
}

looks very annoying. We can add "static cpumask_t cpu_singlethread_map" and
simplify the code. Lessens .text a bit, and imho makes the code more readable.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

workqueue: make cancel_rearming_delayed_workqueue() work on idle dwork

cancel_rearming_delayed_workqueue(dwork) will hang forever if dwork was not
scheduled, because in that case cancel_delayed_work()->del_timer_sync() never
returns true.

I don't know if there are any callers which may have problems, but this is not
so convenient, and the fix is very simple.

Q: looks like we don't need "struct workqueue_struct *wq" parameter.  If the
timer was aborted successfully, get_wq_data() == wq.  Is it worth to add the
new function?

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

workqueue: don't save interrupts in run_workqueue()

work->func() may sleep, it's a bug to call run_workqueue() with irqs disabled.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

workqueue: kill run_scheduled_work()

Because it has no callers.

Actually, I think the whole idea of run_scheduled_work() was not right, not
good to mix "unqueue this work and execute its ->func()" in one function.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

workqueue: don't migrate pending works from the dead CPU

Currently CPU_DEAD uses kthread_stop() to stop cwq->thread and then
transfers cwq->worklist to another CPU.  However, it is very unlikely that
worker_thread() will notice kthread_should_stop() before flushing
cwq->worklist.  It is only possible if worker_thread() was preempted after
run_workqueue(cwq), a new work_struct was added, and CPU_DEAD happened
before cwq->thread has a chance to run.

This means that take_over_work() mostly adds unneeded complications.  Note
also that kthread_stop() is not good per se, wake_up_process() may confuse
work->func() if it sleeps waiting for some event.

Remove take_over_work() and migrate_sequence complications.  CPU_DEAD sets
the cwq->should_stop flag (introduced by this patch) and waits for
cwq->thread to flush cwq->worklist and exit.  Because the dead CPU is not
on cpu_online_map, no more works can be added to that cwq.

cpu_populated_map was introduced to optimize for_each_possible_cpu(), it is
not strictly needed, and it is more a documentation in fact.

Saves 418 bytes.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Srivatsa Vaddagiri <vatsa@in.ibm.com>
Cc: "Pallipadi, Venkatesh" <venkatesh.pallipadi@intel.com>
Cc: Gautham shenoy <ego@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

workqueue: don't clear cwq->thread until it exits

Pointed out by Srivatsa Vaddagiri.

cleanup_workqueue_thread() sets cwq->thread = NULL and does kthread_stop().
This breaks the "if (cwq->thread == current)" logic in flush_cpu_workqueue()
and leads to deadlock.

Kill the thead first, then clear cwq->thread. workqueue_mutex protects us
from create_workqueue_thread() so we don't need cwq->lock.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Srivatsa Vaddagiri <vatsa@in.ibm.com>
Cc: "Pallipadi, Venkatesh" <venkatesh.pallipadi@intel.com>
Cc: Gautham shenoy <ego@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

workqueue: fix flush_workqueue() vs CPU_DEAD race

Many thanks to Srivatsa Vaddagiri for the helpful discussion and for spotting
the bug in my previous attempt.

work->func() (and thus flush_workqueue()) must not use workqueue_mutex,
this leads to deadlock when CPU_DEAD does kthread_stop(). However without
this mutex held we can't detect CPU_DEAD in progress, which can move pending
works to another CPU while the dead one is not on cpu_online_map.

Change flush_workqueue() to use for_each_possible_cpu(). This means that
flush_cpu_workqueue() may hit CPU which is already dead. However in that
case

!list_empty(&cwq->worklist) || cwq->current_work != NULL

means that CPU_DEAD in progress, it will do kthread_stop() + take_over_work()
so we can proceed and insert a barrier. We hold cwq->lock, so we are safe.

Also, add migrate_sequence incremented by take_over_work() under cwq->lock.
If take_over_work() happened before we checked this CPU, we should see the
new value after spin_unlock().

Further possible changes:

remove CPU_DEAD handling (along with take_over_work, migrate_sequence)
from workqueue.c. CPU_DEAD just sets cwq->please_exit_after_flush flag.

CPU_UP_PREPARE->create_workqueue_thread() clears this flag, and creates
the new thread if cwq->thread == NULL.

This way the workqueue/cpu-hotplug interaction is almost zero, workqueue_mutex
just protects "workqueues" list, CPU_LOCK_ACQUIRE/CPU_LOCK_RELEASE go away.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Srivatsa Vaddagiri <vatsa@in.ibm.com>
Cc: "Pallipadi, Venkatesh" <venkatesh.pallipadi@intel.com>
Cc: Gautham shenoy <ego@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

workqueue: fix freezeable workqueues implementation

Currently ->freezeable is per-cpu, this is wrong. CPU_UP_PREPARE creates
cwq->thread which is not freezeable. Move ->freezeable to workqueue_struct.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Srivatsa Vaddagiri <vatsa@in.ibm.com>
Cc: "Pallipadi, Venkatesh" <venkatesh.pallipadi@intel.com>
Cc: Gautham shenoy <ego@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

slab: use CPU_LOCK_[ACQUIRE|RELEASE]

Looks like this was forgotten when CPU_LOCK_[ACQUIRE|RELEASE] was
introduced.

Cc: Pekka Enberg <penberg@cs.helsinki.fi>
Cc: Srivatsa Vaddagiri <vatsa@in.ibm.com>
Cc: Gautham Shenoy <ego@in.ibm.com>
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Oleg Nesterov <oleg@tv-sign.ru>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

call cpu_chain with CPU_DOWN_FAILED if CPU_DOWN_PREPARE failed

This makes cpu hotplug symmetrical: if CPU_UP_PREPARE fails we get
CPU_UP_CANCELED, so we can undo what ever happened on PREPARE.  The same
should happen for CPU_DOWN_PREPARE.

[akpm@linux-foundation.org: fix for reduce-size-of-task_struct-on-64-bit-machines]
Cc: Srivatsa Vaddagiri <vatsa@in.ibm.com>
Cc: Gautham Shenoy <ego@in.ibm.com>
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Oleg Nesterov <oleg@tv-sign.ru>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Eliminate lock_cpu_hotplug in kernel/schedc

Eliminate lock_cpu_hotplug from kernel/sched.c and use sched_hotcpu_mutex
instead to postpone a hotplug event.

In the migration_call hotcpu callback function, take sched_hotcpu_mutex
while handling the event CPU_LOCK_ACQUIRE and release it while handling
CPU_LOCK_RELEASE event.

[akpm@linux-foundation.org: fix deadlock]
Signed-off-by: Gautham R Shenoy <ego@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Define and use new events,CPU_LOCK_ACQUIRE and CPU_LOCK_RELEASE

This is an attempt to provide an alternate mechanism for postponing
a hotplug event instead of using a global mechanism like lock_cpu_hotplug.

The proposal is to add two new events namely CPU_LOCK_ACQUIRE and
CPU_LOCK_RELEASE. The notification for these two events would be sent
out before and after a cpu_hotplug event respectively.

During the CPU_LOCK_ACQUIRE event, a cpu-hotplug-aware subsystem is
supposed to acquire any per-subsystem hotcpu mutex ( Eg. workqueue_mutex
in kernel/workqueue.c ).

During the CPU_LOCK_RELEASE release event the cpu-hotplug-aware subsystem
is supposed to release the per-subsystem hotcpu mutex.

The reasons for defining new events as opposed to reusing the existing events
like CPU_UP_PREPARE/CPU_UP_FAILED/CPU_ONLINE for locking/unlocking of
per-subsystem hotcpu mutexes are as follow:

- CPU_LOCK_ACQUIRE: All hotcpu mutexes are taken before subsystems
start handling pre-hotplug events like CPU_UP_PREPARE/CPU_DOWN_PREPARE
etc, thus ensuring a clean handling of these events.

- CPU_LOCK_RELEASE: The hotcpu mutexes will be released only after
all subsystems have handled post-hotplug events like CPU_DOWN_FAILED,
CPU_DEAD,CPU_ONLINE etc thereby ensuring that there are no subsequent
clashes amongst the interdependent subsystems after a cpu hotplugs.

This patch also uses __raw_notifier_call chain in _cpu_up to take care
of the dependency between the two consequetive calls to
raw_notifier_call_chain.

[akpm@linux-foundation.org: fix a bug]
Signed-off-by: Gautham R Shenoy <ego@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Extend notifier_call_chain to count nr_calls made

Since 2.6.18-something, the community has been bugged by the problem to
provide a clean and a stable mechanism to postpone a cpu-hotplug event as
lock_cpu_hotplug was badly broken.

This is another proposal towards solving that problem.  This one is along the
lines of the solution provided in kernel/workqueue.c

Instead of having a global mechanism like lock_cpu_hotplug, we allow the
subsytems to define their own per-subsystem hot cpu mutexes.  These would be
taken(released) where ever we are currently calling
lock_cpu_hotplug(unlock_cpu_hotplug).

Also, in the per-subsystem hotcpu callback function,we take this mutex before
we handle any pre-cpu-hotplug events and release it once we finish handling
the post-cpu-hotplug events.  A standard means for doing this has been
provided in [PATCH 2/4] and demonstrated in [PATCH 3/4].

The ordering of these per-subsystem mutexes might still prove to be a
problem, but hopefully lockdep should help us get out of that muddle.

The patch set to be applied against linux-2.6.19-rc5 is as follows:

[PATCH 1/4] : Extend notifier_call_chain with an option to specify the
number of notifications to be sent and also count the
number of notifications actually sent.

[PATCH 2/4] : Define events CPU_LOCK_ACQUIRE and CPU_LOCK_RELEASE
and send out notifications for these in _cpu_up and
_cpu_down. This would help us standardise the acquire and
release of the subsystem locks in the hotcpu
callback functions of these subsystems.

[PATCH 3/4] : Eliminate lock_cpu_hotplug from kernel/sched.c.

[PATCH 4/4] : In workqueue_cpu_callback function, acquire(release) the
workqueue_mutex while handling
CPU_LOCK_ACQUIRE(CPU_LOCK_RELEASE).

If the per-subsystem-locking approach survives the test of time, we can expect
a slow phasing out of lock_cpu_hotplug, which has not yet been eliminated in
these patches :)

This patch:

Provide notifier_call_chain with an option to call only a specified number of
notifiers and also record the number of call to notifiers made.

The need for this enhancement was identified in the post entitled
"Slab - Eliminate lock_cpu_hotplug from slab"
(http://lkml.org/lkml/2006/10/28/92) by Ravikiran G Thirumalai and
Andrew Morton.

This patch adds two additional parameters to notifier_call_chain API namely
 - int nr_to_calls : Number of notifier_functions to be called.
       The don't care value is -1.

 - unsigned int *nr_calls : Records the total number of notifier_funtions
    called by notifier_call_chain. The don't care
    value is NULL.

[michal.k.k.piotrowski@gmail.com: build fix]
Credit: Andrew Morton <akpm@osdl.org>
Signed-off-by: Gautham R Shenoy <ego@in.ibm.com>
Signed-off-by: Michal Piotrowski <michal.k.k.piotrowski@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

relay: use plain timer instead of delayed work

relay doesn't need to use schedule_delayed_work() for waking readers
when a simple timer will do.

Signed-off-by: Tom Zanussi <zanussi@comcast.net>
Cc: Satyam Sharma <satyam.sharma@gmail.com>
Cc: Oleg Nesterov <oleg@tv-sign.ru>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

phy: use flush_work_keventd()

(akpm: bypassed maintainers, sorry.  There are other patches which depend on
this)

Cc: "Maciej W. Rozycki" <macro@linux-mips.org>
Cc: David Howells <dhowells@redhat.com>
Cc: Jeff Garzik <jeff@garzik.org>
Cc: Oleg Nesterov <oleg@tv-sign.ru>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

libata: use flush_work()

(akpm: bypassed maintainers, sorry.  There are other patches which depend on
this)

Cc: "Maciej W. Rozycki" <macro@linux-mips.org>
Cc: David Howells <dhowells@redhat.com>
Cc: Jeff Garzik <jeff@garzik.org>
Cc: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Tejun Heo <htejun@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

e1000: use flush_work_keventd()

Switch e1000 over to flush_work_keventd().  This probably fixes a netdev-close
versus linkwatch rtnl_lock() deadlock which nobody knew about.

(akpm: bypassed maintainers, sorry.  There are other patches which depend on
this)

Cc: "Maciej W. Rozycki" <macro@linux-mips.org>
Cc: David Howells <dhowells@redhat.com>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Jeff Garzik <jeff@garzik.org>
Acked-by: Auke Kok <auke-jan.h.kok@intel.com>
Cc: Oleg Nesterov <oleg@tv-sign.ru>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

tg3: use flush_work_keventd()

Convert tg3 over to flush_work_keventd().  Remove nasty now-unneeded deadlock
avoidance logic.

(akpm: bypassed maintainers, sorry.  There are other patches which depend on
this)

Cc: "Maciej W. Rozycki" <macro@linux-mips.org>
Cc: David Howells <dhowells@redhat.com>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Michael Chan <mchan@broadcom.com>
Cc: Jeff Garzik <jeff@garzik.org>
Cc: Oleg Nesterov <oleg@tv-sign.ru>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

kblockd: use flush_work

Switch the kblockd flushing from a global flush to a more specific
flush_work().

(akpm: bypassed maintainers, sorry.  There are other patches which depend on
this)

Cc: "Maciej W. Rozycki" <macro@linux-mips.org>
Cc: David Howells <dhowells@redhat.com>
Cc: Jens Axboe <axboe@suse.de>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Oleg Nesterov <oleg@tv-sign.ru>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

aio: use flush_work()

Migrate AIO over to use flush_work().

Cc: "Maciej W. Rozycki" <macro@linux-mips.org>
Cc: David Howells <dhowells@redhat.com>
Cc: Zach Brown <zach.brown@oracle.com>
Cc: Benjamin LaHaise <bcrl@kvack.org>
Cc: Oleg Nesterov <oleg@tv-sign.ru>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

flush_cpu_workqueue: don't flush an empty ->worklist

Now when we have ->current_work we can avoid adding a barrier and waiting
for its completition when cwq's queue is empty.

Note: this change is also useful if we change flush_workqueue() to also
check the dead CPUs.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Srivatsa Vaddagiri <vatsa@in.ibm.com>
Cc: Gautham Shenoy <ego@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

flush_workqueue(): use preempt_disable to hold off cpu hotplug

Cc: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Srivatsa Vaddagiri <vatsa@in.ibm.com>
Cc: Gautham Shenoy <ego@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

implement flush_work()

A basic problem with flush_scheduled_work() is that it blocks behind _all_
presently-queued works, rather than just the work whcih the caller wants to
flush.  If the caller holds some lock, and if one of the queued work happens
to want that lock as well then accidental deadlocks can occur.

One example of this is the phy layer: it wants to flush work while holding
rtnl_lock().  But if a linkwatch event happens to be queued, the phy code will
deadlock because the linkwatch callback function takes rtnl_lock.

So we implement a new function which will flush a *single* work - just the one
which the caller wants to free up.  Thus we avoid the accidental deadlocks
which can arise from unrelated subsystems' callbacks taking shared locks.

flush_work() non-blockingly dequeues the work_struct which we want to kill,
then it waits for its handler to complete on all CPUs.

Add ->current_work to the "struct cpu_workqueue_struct", it points to
currently running "struct work_struct". When flush_work(work) detects
->current_work == work, it inserts a barrier at the _head_ of ->worklist
(and thus right _after_ that work) and waits for completition. This means
that the next work fired on that CPU will be this barrier, or another
barrier queued by concurrent flush_work(), so the caller of flush_work()
will be woken before any "regular" work has a chance to run.

When wait_on_work() unlocks workqueue_mutex (or whatever we choose to protect
against CPU hotplug), CPU may go away. But in that case take_over_work() will
move a barrier we queued to another CPU, it will be fired sometime, and
wait_on_work() will be woken.

Actually, we are doing cleanup_workqueue_thread()->kthread_stop() before
take_over_work(), so cwq->thread should complete its ->worklist (and thus
the barrier), because currently we don't check kthread_should_stop() in
run_workqueue(). But even if we did, everything should be ok.

[akpm@osdl.org: cleanup]
[akpm@osdl.org: add flush_work_keventd() wrapper]
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

reimplement flush_workqueue()

Remove ->remove_sequence, ->insert_sequence, and ->work_done from struct
cpu_workqueue_struct.  To implement flush_workqueue() we can queue a
barrier work on each CPU and wait for its completition.

The barrier is queued under workqueue_mutex to ensure that per cpu
wq->cpu_wq is alive, we drop this mutex before going to sleep.  If CPU goes
down while we are waiting for completition, take_over_work() will move the
barrier on another CPU, and the handler will wake up us eventually.

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