Unlike the other tty comment patch this one has code changes.  Specifically
it limits the queue size for a tty to 64K characters (128Kbytes) worst case
even if the tty is ignoring tty->throttle.  This is because certain drivers
don't honour the throttle value correctly, although it is a useful
safeguard anyway.
Signed-off-by: NAlan Cox <alan@redhat.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Doesn't fix them but does show up some interesting areas that need review
and fixing.
Signed-off-by: NAlan Cox <alan@redhat.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Fix printk format warning:
drivers/cdrom/gscd.c:269: warning: format ‘%lu’ expects type ‘long unsigned int’, but argument 2 has type ‘unsigned int’
Signed-off-by: NRandy Dunlap <rdunlap@xenotime.net>
Cc: Jens Axboe <axboe@suse.de>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

When we select NUMA with i386, the system is only X86_NUMAQ or using ACPI.
Signed-off-by: NKAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

None of the other /proc/meminfo lines have a space in the identifier.  This
post-2.6.17 addition has the potential to break existing parsers, so use an
underscore instead (like Committed_AS).

Cc: Christoph Lameter <clameter@engr.sgi.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

This fixes the locking error noticed by lockdep:

  =============================================
  [ INFO: possible recursive locking detected ]
  ---------------------------------------------
  init/1 is trying to acquire lock:
   (&sighand->siglock){....}, at: [<c047a78a>] flush_old_exec+0x3ae/0x859

  but task is already holding lock:
   (&sighand->siglock){....}, at: [<c047a77a>] flush_old_exec+0x39e/0x859

  other info that might help us debug this:
  2 locks held by init/1:
   #0:  (tasklist_lock){..--}, at: [<c047a76a>] flush_old_exec+0x38e/0x859
   #1:  (&sighand->siglock){....}, at: [<c047a77a>] flush_old_exec+0x39e/0x859

  stack backtrace:
   [<c04051e1>] show_trace_log_lvl+0x54/0xfd
   [<c040579d>] show_trace+0xd/0x10
   [<c04058b6>] dump_stack+0x19/0x1b
   [<c043b33a>] __lock_acquire+0x773/0x997
   [<c043bacf>] lock_acquire+0x4b/0x6c
   [<c060630b>] _spin_lock+0x19/0x28
   [<c047a78a>] flush_old_exec+0x3ae/0x859
   [<c0498053>] load_elf_binary+0x4aa/0x1628
   [<c0479cab>] search_binary_handler+0xa7/0x24e
   [<c047b577>] do_execve+0x15b/0x1f9
   [<c04022b4>] sys_execve+0x29/0x4d
   [<c0403faf>] syscall_call+0x7/0xb
Signed-off-by: NArjan van de Ven <arjan@infradead.org>
Signed-off-by: NDave Jones <davej@redhat.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

reiserfs seems to have another locking level layer for the i_mutex due to the
xattrs-are-a-directory thing.
Signed-off-by: NIngo Molnar <mingo@elte.hu>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

idescsi_pc_intr() uses local_irq_enable() in IRQ context: annotate it.

(this has no effect on kernels with lockdep disabled.  On kernels with lockdep
enabled this means that we wont actually disable interrupts, and the warning
message will go away as well.)
Signed-off-by: NIngo Molnar <mingo@elte.hu>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

  In file included from include/asm/mmzone.h:18,
                   from include/linux/mmzone.h:439,
  <snip>
  include/asm/srat.h:31:2: error: #error CONFIG_ACPI_SRAT not defined, and srat.h header has been included
  make[1]: *** [arch/i386/kernel/asm-offsets.s] Error 1

This can happen with CONFIG_NUMA && !CONFIG_ACPI && !CONFIG_X86_NUMAQ
Signed-off-by: NKAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

cpuset_excl_nodes_overlap always returns 0 if current is exiting.  This caused
customer's systems to panic in the OOM killer when processes were having
trouble getting memory for the final put_user in mm_release.  Even though
there were lots of processes to kill.

Change to returning 1 in this case.  This achieves parity with !CONFIG_CPUSETS
case, and was observed to fix the problem.
Signed-off-by: NNick Piggin <npiggin@suse.de>
Acked-by: NPaul Jackson <pj@sgi.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

register_one_node()'s should be defined under CONFIG_NUMA=n.
fixes following bug.

  CC	  init/version.o
  LD	  init/built-in.o
  LD	  .tmp_vmlinux1
  mm/built-in.o: In function `add_memory': undefined reference to `register_one_node'
Signed-off-by: NKAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: NYasunori Goto <y-goto@jp.fujitsu.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

JBD currently allocates commit and frozen buffers from slabs.  With
CONFIG_SLAB_DEBUG, its possible for an allocation to cross the page
boundary causing IO problems.

https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=200127

So, instead of allocating these from regular slabs - manage allocation from
its own slabs and disable slab debug for these slabs.

[akpm@osdl.org: cleanups]
Signed-off-by: NBadari Pulavarty <pbadari@us.ibm.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Change the list of cpus allowed to tasks in the top (root) cpuset to
dynamically track what cpus are online, using a CPU hotplug notifier.  Make
this top cpus file read-only.

On systems that have cpusets configured in their kernel, but that aren't
actively using cpusets (for some distros, this covers the majority of
systems) all tasks end up in the top cpuset.

If that system does support CPU hotplug, then these tasks cannot make use
of CPUs that are added after system boot, because the CPUs are not allowed
in the top cpuset.  This is a surprising regression over earlier kernels
that didn't have cpusets enabled.

In order to keep the behaviour of cpusets consistent between systems
actively making use of them and systems not using them, this patch changes
the behaviour of the 'cpus' file in the top (root) cpuset, making it read
only, and making it automatically track the value of cpu_online_map.  Thus
tasks in the top cpuset will have automatic use of hot plugged CPUs allowed
by their cpuset.

Thanks to Anton Blanchard and Nathan Lynch for reporting this problem,
driving the fix, and earlier versions of this patch.
Signed-off-by: NPaul Jackson <pj@sgi.com>
Cc: Nathan Lynch <ntl@pobox.com>
Cc: Anton Blanchard <anton@samba.org>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

A recent patch broke the ability to do a user-request check of a raid1.
This patch fixes the breakage and also moves a comment that was dislocated
by the same patch.
Signed-off-by: NNeil Brown <neilb@suse.de>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

If we
  - shut down a clean array,
  - restart with one (or more) drive(s) missing
  - make some changes
  - pause, so that they array gets marked 'clean',
the event count on the superblock of included drives
will be the same as that of the removed drives.
So adding the removed drive back in will cause it
to be included with no resync.

To avoid this, we only update the eventcount backwards when the array
is not degraded.  In this case there can (should) be no non-connected
drives that we can get confused with, and this is the particular case
where updating-backwards is valuable.
Signed-off-by: NNeil Brown <neilb@suse.de>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Fix two compile failures in eventpoll.c code which would happen if
DEBUG_EPOLL is bigger than zero.
Signed-off-by: NMasoud Sharbiani <masouds@google.com>
Cc: Davide Libenzi <davidel@xmailserver.org>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Here's updated documentation for the relay interface, rewritten to match
the relayfs->relay changes.  It also moves relayfs.txt to relay.txt in the
process.

It includes the changes to relayfs.txt previously posted by Randy Dunlap,
thanks for those.

The relay-apps examples have also been updated to match, and can be found
on the sourceforge relayfs website.
Signed-off-by: NTom Zanussi <zanussi@us.ibm.com>
Cc: "Randy.Dunlap" <rdunlap@xenotime.net>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

An up() is called in kernel/stop_machine.c on failure, and also in the
caller (unconditionally).
Signed-off-by: NZhou Yingchao <yingchao.zhou@gmail.com>
Cc: <stable@kernel.org>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

1) When we allocated last fragment in ufs_truncate, we read page, check
   if block mapped to address, and if not trying to allocate it.  This is
   wrong behaviour, fragment may be NOT allocated, but mapped, this
   happened because of "block map" function not checked allocated fragment
   or not, it just take address of the first fragment in the block, add
   offset of fragment and return result, this is correct behaviour in
   almost all situation except call from ufs_truncate.

2) Almost all implementation of UFS, which I can investigate have such
   "defect": if you have full disk, and try truncate file, for example 3GB
   to 2MB, and have hole in this region, truncate return -ENOSPC.  I tried
   evade from this problem, but "block allocation" algorithm is tied to
   right value of i_lastfrag, and fix of this corner case may slow down of
   ordinaries scenarios, so this patch makes behavior of "truncate"
   operations similar to what other UFS implementations do.
Signed-off-by: NEvgeniy Dushistov <dushistov@mail.ru>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

On UFS, this scenario:
	open(O_TRUNC)
	lseek(1024 * 1024 * 80)
	write("A")
	lseek(1024 * 2)
	write("A")

may cause access to invalid address.

This happened because of "goal" is calculated in wrong way in block
allocation path, as I see this problem exists also in 2.4.

We use construction like this i_data[lastfrag], i_data array of pointers to
direct blocks, indirect and so on, it has ceratain size ~20 elements, and
lastfrag may have value for example 40000.

Also this patch fixes related to handling such scenario issues, wrong
zeroing metadata, in case of block(not fragment) allocation, and wrong goal
calculation, when we allocate block
Signed-off-by: NEvgeniy Dushistov <dushistov@mail.ru>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

To handle the earlier bogus ENOSPC error caused by filesystem full of block
reservation, current code falls back to non block reservation, starts to
allocate block(s) from the goal allocation block group as if there is no
block reservation.

Current code needs to re-load the corresponding block group descriptor for
the initial goal block group in this case.  The patch fixes this.
Signed-off-by: NMingming Cao <cmm@us.ibm.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Mounting an ext2 filesystem with zero s_inodes_per_group will cause a
divide error.
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Mounting a (corrupt) minix filesystem with zero s_zmap_blocks
gives a spectacular crash on my 2.6.17.8 system, no doubt
because minix/inode.c does an unconditional
	minix_set_bit(0,sbi->s_zmap[0]->b_data);

[akpm@osdl.org: make labels conistent while we're there]
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

The recent hwctrl core conversion for MTD NAND devices broke the Amstrad
Delta driver.  This fixes it up and uses the existing control line defines
rather than unclear magic numbers.
Signed-off-by: NJonathan McDowell <noodles@earth.li>
Acked-by: NDavid Woodhouse <dwmw2@infradead.org>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

futex_find_get_task:

	if (p->state == EXIT_ZOMBIE || p->exit_state == EXIT_ZOMBIE)
		return NULL;

I can't understand this.  First, p->state can't be EXIT_ZOMBIE.  The
->exit_state check looks strange too.  Sub-threads or tasks whose ->parent
ignores SIGCHLD go directly to EXIT_DEAD state (I am ignoring a ptrace
case).  Why EXIT_DEAD tasks should be ok?  Yes, EXIT_ZOMBIE is more
important (a task may stay zombie for a long time), but this doesn't mean
we should explicitely ignore other EXIT_XXX states.
Signed-off-by: NOleg Nesterov <oleg@tv-sign.ru>
Acked-by: NIngo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

When reading /dev/vcsa while a font with more than 256 characters is
loaded, one of the attribute bits records the 9th bit of the character.
But depending on the console driver (vgacon or fbcon for instance), that's
bit 3 or bit 0.  And there is no way for userland to know that, thus no way
for userland to safely grab the screen content.  So here is a (tested)
patch:

Add a VT_GETHIFONTMASK ioctl for knowing which bit is the 9th bit for VC
text (vc_hi_font_mask field of the vc_data structure).
Signed-off-by: NSamuel Thibault <samuel.thibault@ens-lyon.org>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

I wish I was happier about this patch.  It'll serve as a placeholder for
the moment.  I'm still trying to get a G550 working in order to even
reproduce the problem this patch introduces.  I find that the G450 has
jitter even without this patch, so it won't show me what the patch changed.
 At this point, I'll continue trying to get the G550 to work, and in
parallel work with the G450 to work out the kinks.

The patch is below.

Set XDVICLKCTRL only on PPC, as doing this apparently introduces jitter on
the G550, at least on x86 architectures.
Signed-off-by: NPaul A. Clarke <pc@us.ibm.com>
Signed-off-by: NPetr Vandrovec <petr@vandrovec.name>
Cc: "Antonino A. Daplas" <adaplas@pol.net>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

While testing Moxa C218T/PCI on PowerPC 405EP I found that loading firmware
using the linux kernel driver fails because calculation of the checksum is
not endianess independent in the original code.

After I fixed this I found that uploading firmware in a system with
multiple cards causes a kernel oops.  I had a look in the recent moxa
sources and found that they do some kind of locking there.  Applying this
lock fixed the problem.

Alan sayeth:

  Checksum changes are clearly correct.  Other changes is an improvement but
  not I think enough to handle malicious firmware attacks.  That said such an
  attacker has CAP_SYS_RAWIO anyway so that part is irrelevant except for
  neatness.

[akpm@osdl.org: cleanups]
Signed-off-by: NDirk Eibach <eibach@gdsys.de>
Acked-by: NAlan Cox <alan@redhat.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Ignore the return value of early_init_acpi(), as it can give false error
messages.  If there is something really wrong, then register_driver will
fail cleanly with EINVAL later.

[ background: modprobe acpi-cpufreq on systems not capable of speed-scaling
  started failing with 'invalid argument', where previously it would only
  ever -ENODEV

  I'm not 100% happy with the solution. It'd be better to handle
  failure properly, but this is a low-impact change for 2.6.18
  We can always revisit doing this better in .19   --davej.]
Signed-off-by: NVenkatesh Pallipadi <venkatesh.pallipadi@intel.com>
Signed-off-by: NDave Jones <davej@redhat.com>
Cc: Greg KH <greg@kroah.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

sched_setscheduler() looks at ->signal->rlim[].  It is unsafe do
dereference ->signal unless tasklist_lock or ->siglock is held (or p ==
current).  We pin the task structure, but this can't prevent from
release_task()->__exit_signal() which sets ->signal = NULL.

Restore tasklist_lock across the setscheduler call.
Signed-off-by: NOleg Nesterov <oleg@tv-sign.ru>
Cc: Greg KH <greg@kroah.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Commit b64ef8af ("[PATCH] add imacfb
documentation and detection") contained a wrong DMI_MATCH.
Signed-off-by: NThomas Meyer <thomas@m3y3r.de>
Cc: Greg KH <greg@kroah.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Read the return value before we release the nand device otherwise the
value can become corrupted by another user of chip->ops, ultimately
resulting in filesystem corruption.
Signed-off-by: NRichard Purdie <rpurdie@rpsys.net>
Cc: David Woodhouse <dwmw2@infradead.org>
Acked-by: NJosh Boyer <jwboyer@gmail.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

On Wed, 2006-08-09 at 07:57 +0200, Rolf Eike Beer wrote:
> =============================================
> [ INFO: possible recursive locking detected ]
> ---------------------------------------------
> parted/7929 is trying to acquire lock:
>  (&bdev->bd_mutex){--..}, at: [<c105eb8d>] __blkdev_put+0x1e/0x13c
>
> but task is already holding lock:
>  (&bdev->bd_mutex){--..}, at: [<c105eec6>] do_open+0x72/0x3a8
>
> other info that might help us debug this:
> 1 lock held by parted/7929:
>  #0:  (&bdev->bd_mutex){--..}, at: [<c105eec6>] do_open+0x72/0x3a8
> stack backtrace:
>  [<c1003aad>] show_trace_log_lvl+0x58/0x15b
>  [<c100495f>] show_trace+0xd/0x10
>  [<c1004979>] dump_stack+0x17/0x1a
>  [<c102dee5>] __lock_acquire+0x753/0x99c
>  [<c102e3b0>] lock_acquire+0x4a/0x6a
>  [<c1204501>] mutex_lock_nested+0xc8/0x20c
>  [<c105eb8d>] __blkdev_put+0x1e/0x13c
>  [<c105ecc4>] blkdev_put+0xa/0xc
>  [<c105f18a>] do_open+0x336/0x3a8
>  [<c105f21b>] blkdev_open+0x1f/0x4c
>  [<c1057b40>] __dentry_open+0xc7/0x1aa
>  [<c1057c91>] nameidata_to_filp+0x1c/0x2e
>  [<c1057cd1>] do_filp_open+0x2e/0x35
>  [<c1057dd7>] do_sys_open+0x38/0x68
>  [<c1057e33>] sys_open+0x16/0x18
>  [<c1002845>] sysenter_past_esp+0x56/0x8d

OK, I'm having a look here; its all new to me so bear with me.

blkdev_open() calls
  do_open(bdev, ...,BD_MUTEX_NORMAL) and takes
    mutex_lock_nested(&bdev->bd_mutex, BD_MUTEX_NORMAL)

then something fails, and we're thrown to:

out_first: where
    if (bdev != bdev->bd_contains)
      blkdev_put(bdev->bd_contains) which is
        __blkdev_put(bdev->bd_contains, BD_MUTEX_NORMAL) which does
          mutex_lock_nested(&bdev->bd_contains->bd_mutex, BD_MUTEX_NORMAL) <--- lockdep trigger

When going to out_first, dbev->bd_contains is either bdev or whole, and
since we take the branch it must be whole. So it seems to me the
following patch would be the right one:

[akpm@osdl.org: compile fix]
Signed-off-by: NPeter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arjan van de Ven <arjan@linux.intel.com>
Acked-by: NNeilBrown <neilb@suse.de>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Recently a patch was added for preliminary suspend/resume handling on
!PPC_PMAC.  However, this broke both suspend and firewire on powerpc
because it saves the pci state after the device has already been disabled.

This moves the save state to before the pmac specific code.
Signed-off-by: NDanny Tholen <obiwan@mailmij.org>
Cc: Stefan Richter <stefanr@s5r6.in-berlin.de>
Acked-by: NBenjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Ben Collins <bcollins@ubuntu.com>
Cc: Jody McIntyre <scjody@modernduck.com>
Cc: <stable@kernel.org>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Sergey Vlasov noticed that there is not kernel.suid_dumpable, but
fs.suid_dumpable.

How KERN_SETUID_DUMPABLE ended up in fs_table[]? Hell knows...
Signed-off-by: NAlexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

When cdev_add() failed there is no reason to call cdev_del().
Signed-off-by: NRolf Eike Beer <eike-kernel@sf-tec.de>
Cc: Alan Cox <alan@lxorguk.ukuu.org.uk>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Fix the year check on setting the time with the S3C24XX RTC driver.  Also
move the debug to before the set to see what is going on if it does fail.
Signed-off-by: NBen Dooks <ben-linux@fluff.org>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

There is a bug in mm/swapfile.c#swap_type_of() that makes swsusp only be
able to use the first active swap partition as the resume device.  Fix it.
Signed-off-by: NRafael J. Wysocki <rjw@sisk.pl>
Cc: Hugh Dickins <hugh@veritas.com>
Acked-by: NPavel Machek <pavel@suse.cz>
Cc: <stable@kernel.org>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

On an nForce4-equipped machine with two SATA disk in raid1 setup using dmraid,
we experienced frequent deadlock of the system under high i/o load.  'cat
/dev/zero > ~/zero' was the most reliable way to reproduce them: Randomly
after a few GB, 'cp' would be left in 'D' state along with kjournald and
kmirrord.  The functions cp and kjournald were blocked in did vary, but
kmirrord's wchan always pointed to 'mempool_alloc()'.  We've seen this pattern
on 2.6.15 and 2.6.17 kernels.  http://lkml.org/lkml/2005/4/20/142 indicates
that this problem has been around even before.

So much for the facts, here's my interpretation: mempool_alloc() first tries
to atomically allocate the requested memory, or falls back to hand out
preallocated chunks from the mempool.  If both fail, it puts the calling
process (kmirrord in this case) on a private waitqueue until somebody refills
the pool.  Where the only 'somebody' is kmirrord itself, so we have a
deadlock.

I worked around this problem by falling back to a (blocking) kmalloc when
before kmirrord would have ended up on the waitqueue.  This defeats part of
the benefits of using the mempool, but at least keeps the system running.  And
it could be done with a two-line change.  Note that mempool_alloc() clears the
GFP_NOIO flag internally, and only uses it to decide whether to wait or return
an error if immediate allocation fails, so the attached patch doesn't change
behaviour in the non-deadlocking case.  Path is against current git
(2.6.18-rc4), but should apply to earlier versions as well.  I've tested on
2.6.15, where this patch makes the difference between random lockup and a
stable system.
Signed-off-by: NDaniel Kobras <kobras@linux.de>
Acked-by: NAlasdair G Kergon <agk@redhat.com>
Cc: <stable@kernel.org>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>