safe/jmp/linux-2.6
15 years agooprofile: simplify add_sample() in cpu_buffer.c
Robert Richter [Tue, 30 Dec 2008 04:10:58 +0000 (05:10 +0100)]
oprofile: simplify add_sample() in cpu_buffer.c

Signed-off-by: Robert Richter <robert.richter@amd.com>
15 years agooprofile: simplify oprofile_begin_trace()
Robert Richter [Thu, 18 Dec 2008 21:09:13 +0000 (22:09 +0100)]
oprofile: simplify oprofile_begin_trace()

This patch removes the unused return parameter in
oprofile_begin_trace(). Also, oprofile_begin_trace() and
oprofile_end_trace() are inline now.

Signed-off-by: Robert Richter <robert.richter@amd.com>
15 years agooprofile: simplify sync_buffer()
Robert Richter [Thu, 18 Dec 2008 18:44:20 +0000 (19:44 +0100)]
oprofile: simplify sync_buffer()

Make code more readable. No functional changes.

Signed-off-by: Robert Richter <robert.richter@amd.com>
15 years agooprofile: simplify add_sample()
Robert Richter [Thu, 18 Dec 2008 18:44:20 +0000 (19:44 +0100)]
oprofile: simplify add_sample()

This patch removes add_us_sample() and simplifies add_sample(). Code
is much more readable now.

Signed-off-by: Robert Richter <robert.richter@amd.com>
15 years agooprofile: add inline function __oprofile_add_ext_sample()
Robert Richter [Tue, 16 Dec 2008 11:00:10 +0000 (12:00 +0100)]
oprofile: add inline function __oprofile_add_ext_sample()

This patch adds the inline function __oprofile_add_ext_sample() to
cpu_buffer.c and thus reduces overhead when calling
oprofile_add_sample().

Signed-off-by: Robert Richter <robert.richter@amd.com>
15 years agooprofile: reordering some code in cpu_buffer.c
Robert Richter [Tue, 23 Dec 2008 00:35:12 +0000 (01:35 +0100)]
oprofile: reordering some code in cpu_buffer.c

Reordering code to keep alloc/free functions together.

Signed-off-by: Robert Richter <robert.richter@amd.com>
15 years agox86/oprofile: fix pci_dev use count for AMD northbridge devices
Robert Richter [Mon, 15 Dec 2008 14:09:50 +0000 (15:09 +0100)]
x86/oprofile: fix pci_dev use count for AMD northbridge devices

This patch fixes the PCI device use count for AMD northbridge
devices. In case of an IBS LVT initialization failure, the PCI device
is released now by calling pci_dev_put().

If there are no initialization errors, the devices are released in
pci_get_device() while iterating.

Signed-off-by: Robert Richter <robert.richter@amd.com>
15 years agooprofile: remove ring buffer inline functions in cpu_buffer.h
Robert Richter [Tue, 16 Dec 2008 15:19:54 +0000 (16:19 +0100)]
oprofile: remove ring buffer inline functions in cpu_buffer.h

This patch moves ring buffer inline functions to cpu_buffer.c.

Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Robert Richter <robert.richter@amd.com>
15 years agooprofile: rename cpu buffer functions
Robert Richter [Wed, 24 Dec 2008 15:53:53 +0000 (16:53 +0100)]
oprofile: rename cpu buffer functions

This patch renames cpu buffer functions to something more oprofile
specific names. Functions will be moved to the global name space.

Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Robert Richter <robert.richter@amd.com>
15 years agooprofile: rename kernel-wide identifiers
Robert Richter [Tue, 16 Dec 2008 15:19:54 +0000 (16:19 +0100)]
oprofile: rename kernel-wide identifiers

This patch renames kernel-wide identifiers to something more oprofile
specific names.

Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Robert Richter <robert.richter@amd.com>
15 years agooprofile: select RING_BUFFER
Ingo Molnar [Fri, 12 Dec 2008 08:38:57 +0000 (09:38 +0100)]
oprofile: select RING_BUFFER

Impact: build fix

OProfile now depends on the ring buffer infrastructure:

 arch/x86/oprofile/built-in.o: In function `oprofile_add_ibs_sample':
 : undefined reference to `ring_buffer_unlock_commit'

Select TRACING and RING_BUFFER when oprofile is enabled.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agoring_buffer: adding EXPORT_SYMBOLs
Robert Richter [Thu, 11 Dec 2008 15:49:22 +0000 (16:49 +0100)]
ring_buffer: adding EXPORT_SYMBOLs

I added EXPORT_SYMBOL_GPLs for all functions part of the API
(ring_buffer.h). This is required since oprofile is using the ring
buffer and the compilation as modules would fail otherwise.

Signed-off-by: Robert Richter <robert.richter@amd.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
15 years agooprofile: fix lost sample counter
Robert Richter [Tue, 9 Dec 2008 01:13:25 +0000 (02:13 +0100)]
oprofile: fix lost sample counter

The number of lost samples could be greater than the number of
received samples. This patches fixes this. The implementation
introduces return values for add_sample() and add_code().

Signed-off-by: Robert Richter <robert.richter@amd.com>
15 years agooprofile: remove nr_available_slots()
Robert Richter [Mon, 8 Dec 2008 10:59:52 +0000 (11:59 +0100)]
oprofile: remove nr_available_slots()

This function is no longer available after the port to the new ring
buffer. Its removal can lead to incomplete sampling sequences since
IBS samples and backtraces are transfered in multiple samples. Due to
a full buffer, samples could be lost any time. The userspace daemon
has to live with such incomplete sampling sequences as long as the
data within one sample is consistent.

This will be fixed by changing the internal buffer data there all data
of one IBS sample or a backtrace is packed in a single ring buffer
entry. This is possible since the new ring buffer supports variable
data size.

Signed-off-by: Robert Richter <robert.richter@amd.com>
15 years agooprofile: port to the new ring_buffer
Robert Richter [Tue, 9 Dec 2008 00:21:32 +0000 (01:21 +0100)]
oprofile: port to the new ring_buffer

This patch replaces the current oprofile cpu buffer implementation
with the ring buffer provided by the tracing framework. The motivation
here is to leave the pain of implementing ring buffers to others. Oh,
no, there are more advantages. Main reason is the support of different
sample sizes that could be stored in the buffer. Use cases for this
are IBS and Cell spu profiling. Using the new ring buffer ensures
valid and complete samples and allows copying the cpu buffer stateless
without knowing its content. Second it will use generic kernel API and
also reduce code size. And hopefully, there are less bugs.

Since the new tracing ring buffer implementation uses spin locks to
protect the buffer during read/write access, it is difficult to use
the buffer in an NMI handler. In this case, writing to the buffer by
the NMI handler (x86) could occur also during critical sections when
reading the buffer. To avoid this, there are 2 buffers for independent
read and write access. Read access is in process context only, write
access only in the NMI handler. If the read buffer runs empty, both
buffers are swapped atomically. There is potentially a small window
during swapping where the buffers are disabled and samples could be
lost.

Using 2 buffers is a little bit overhead, but the solution is clear
and does not require changes in the ring buffer implementation. It can
be changed to a single buffer solution when the ring buffer access is
implemented as non-locking atomic code.

The new buffer requires more size to store the same amount of samples
because each sample includes an u32 header. Also, there is more code
to execute for buffer access. Nonetheless, the buffer implementation
is proven in the ftrace environment and worth to use also in oprofile.

Patches that changes the internal IBS buffer usage will follow.

Cc: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Robert Richter <robert.richter@amd.com>
15 years agoring_buffer: add remaining cpu functions to ring_buffer.h
Robert Richter [Wed, 26 Nov 2008 13:04:19 +0000 (14:04 +0100)]
ring_buffer: add remaining cpu functions to ring_buffer.h

These functions are not yet in ring_buffer.h though they seems to be
part of the API.

Cc: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Robert Richter <robert.richter@amd.com>
15 years agooprofile: moving cpu_buffer_reset() to cpu_buffer.h
Robert Richter [Thu, 4 Dec 2008 15:27:00 +0000 (16:27 +0100)]
oprofile: moving cpu_buffer_reset() to cpu_buffer.h

This is in preparation for changes in the cpu buffer implementation.

Signed-off-by: Robert Richter <robert.richter@amd.com>
15 years agooprofile: adding cpu_buffer_entries()
Robert Richter [Thu, 27 Nov 2008 21:33:37 +0000 (22:33 +0100)]
oprofile: adding cpu_buffer_entries()

This is in preparation for changes in the cpu buffer implementation.

Signed-off-by: Robert Richter <robert.richter@amd.com>
15 years agooprofile: adding cpu_buffer_write_commit()
Robert Richter [Thu, 27 Nov 2008 17:36:08 +0000 (18:36 +0100)]
oprofile: adding cpu_buffer_write_commit()

This is in preparation for changes in the cpu buffer implementation.

Signed-off-by: Robert Richter <robert.richter@amd.com>
15 years agooprofile: adding cpu buffer r/w access functions
Robert Richter [Thu, 27 Nov 2008 09:57:09 +0000 (10:57 +0100)]
oprofile: adding cpu buffer r/w access functions

This is in preparation for changes in the cpu buffer implementation.

Signed-off-by: Robert Richter <robert.richter@amd.com>
15 years agoftrace: remove unused function arg in trace_iterator_increment()
Robert Richter [Wed, 12 Nov 2008 11:59:32 +0000 (12:59 +0100)]
ftrace: remove unused function arg in trace_iterator_increment()

This removes the unused cpu function parameter.

Cc: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Robert Richter <robert.richter@amd.com>
15 years agoring_buffer: update description for ring_buffer_alloc()
Robert Richter [Mon, 24 Nov 2008 11:24:12 +0000 (12:24 +0100)]
ring_buffer: update description for ring_buffer_alloc()

Trivial patch.

Cc: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Robert Richter <robert.richter@amd.com>
15 years agooprofile: set values to default when creating oprofilefs
Robert Richter [Tue, 9 Dec 2008 15:56:01 +0000 (16:56 +0100)]
oprofile: set values to default when creating oprofilefs

This patch restores default values for:

/dev/oprofile/cpu_buffer_size
/dev/oprofile/buffer_watershed
/dev/oprofile/buffer_size

when creating the oprofilefs:

 # opcontrol --deinit
 # opcontrol --init
 # cat /dev/oprofile/cpu_buffer_size
 8192
 # echo 5123 > /dev/oprofile/cpu_buffer_size
 # cat /dev/oprofile/cpu_buffer_size
 5123
 # opcontrol --deinit
 # opcontrol --init
 # cat /dev/oprofile/cpu_buffer_size
 8192
 # opcontrol --deinit

This sets the values in a defined state. Before, there was no way to
restore the defaults without rebooting the system or reloading the
module.

Signed-off-by: Robert Richter <robert.richter@amd.com>
15 years agooprofile: implement switch/case in buffer_sync.c
Robert Richter [Fri, 26 Sep 2008 21:50:31 +0000 (17:50 -0400)]
oprofile: implement switch/case in buffer_sync.c

Signed-off-by: Robert Richter <robert.richter@amd.com>
15 years agox86/oprofile: cleanup IBS init/exit functions in op_model_amd.c
Robert Richter [Mon, 24 Nov 2008 13:58:03 +0000 (14:58 +0100)]
x86/oprofile: cleanup IBS init/exit functions in op_model_amd.c

Implementation of pairwise init/exit funcions for IBS and IBS NMI
setup. There are also some function renames and the removal of forward
function declarations.

Signed-off-by: Robert Richter <robert.richter@amd.com>
15 years agox86/oprofile: reordering IBS code in op_model_amd.c
Robert Richter [Mon, 24 Nov 2008 13:21:03 +0000 (14:21 +0100)]
x86/oprofile: reordering IBS code in op_model_amd.c

This is part of the cpu buffer rework.

Signed-off-by: Robert Richter <robert.richter@amd.com>
15 years agooprofile: fix typo
Robert Richter [Wed, 26 Nov 2008 14:00:52 +0000 (15:00 +0100)]
oprofile: fix typo

Signed-off-by: Robert Richter <robert.richter@amd.com>
15 years agooprofile: whitspace changes only
Robert Richter [Sat, 27 Sep 2008 02:18:44 +0000 (22:18 -0400)]
oprofile: whitspace changes only

Signed-off-by: Robert Richter <robert.richter@amd.com>
15 years agooprofile: update comment for oprofile_add_sample()
Robert Richter [Wed, 26 Nov 2008 11:02:53 +0000 (12:02 +0100)]
oprofile: update comment for oprofile_add_sample()

The cpu argument is no longer part of the parameter list.

Signed-off-by: Robert Richter <robert.richter@amd.com>
15 years agooprofile: comment cleanup
Robert Richter [Sun, 19 Oct 2008 19:00:09 +0000 (21:00 +0200)]
oprofile: comment cleanup

This fixes the coding style of some comments.

Signed-off-by: Robert Richter <robert.richter@amd.com>
15 years agoLinux 2.6.28-rc7 v2.6.28-rc7
Linus Torvalds [Tue, 2 Dec 2008 03:59:23 +0000 (19:59 -0800)]
Linux 2.6.28-rc7

15 years agoMerge branch 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mchehab...
Linus Torvalds [Tue, 2 Dec 2008 03:56:34 +0000 (19:56 -0800)]
Merge branch 'for_linus' of git://git./linux/kernel/git/mchehab/linux-2.6

* 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mchehab/linux-2.6: (25 commits)
  em28xx: remove backward compat macro added on a previous fix
  V4L/DVB (9748): em28xx: fix compile warning
  V4L/DVB (9743): em28xx: fix oops audio
  V4L/DVB (9742): em28xx-alsa: implement another locking schema
  V4L/DVB (9732): sms1xxx: use new firmware for Hauppauge WinTV MiniStick
  V4L/DVB (9691): gspca: Move the video device to a separate area.
  V4L/DVB (9690): gspca: Lock the subdrivers via module_get/put.
  V4L/DVB (9689): gspca: Memory leak when disconnect while streaming.
  V4L/DVB (9668): em28xx: fix a race condition with hald
  V4L/DVB (9664): af9015: don't reconnect device in USB-bus
  V4L/DVB (9647): em28xx: void having two concurrent control URB's
  V4L/DVB (9646): em28xx: avoid allocating/dealocating memory on every control urb
  V4L/DVB (9645): em28xx: Avoid memory leaks if registration fails
  V4L/DVB (9639): Make dib0700 remote control support work with firmware v1.20
  V4L/DVB (9635): v4l: s2255drv fix firmware test on big-endian
  V4L/DVB (9634): Make sure the i2c gate is open before powering down tuner
  V4L/DVB (9632): make em28xx aux audio input work
  V4L/DVB (9631): Make s2api work for ATSC support
  V4L/DVB (9627): em28xx: Avoid i2c register error for boards without eeprom
  V4L/DVB (9608): Fix section mismatch warning for dm1105 during make
  ...

15 years agodrivers/gpu/drm/i915/i915_irq.c: fix warning
Andrew Morton [Mon, 1 Dec 2008 21:14:08 +0000 (13:14 -0800)]
drivers/gpu/drm/i915/i915_irq.c: fix warning

drivers/gpu/drm/i915/i915_irq.c: In function 'i915_disable_pipestat':
drivers/gpu/drm/i915/i915_irq.c:101: warning: control may reach end of non-void function 'i915_pipestat' being inlined

Cc: Dave Airlie <airlied@linux.ie>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
15 years agoi82875p_edac: fix module remove
Jarkko Lavinen [Mon, 1 Dec 2008 21:14:08 +0000 (13:14 -0800)]
i82875p_edac: fix module remove

Fix module removal bugs of i82875p_edac.  Also i82975x_edac code seems to
have the same module removal bugs as in i82875p_edac.

The problems were:

1. In module removal i82875p_remove_one() is never called.

   Variable i82875p_registered is newer changed from 1, which
   guarantees i82875p_remove_one() is not called (and even if it were
   called, it would be called in wrong order).

   As a result, the edac_mc workque is not stopped and keeps probing.
   If kernel debugging options are not enabled, user may not notice
   anything going wrong.

   if debugging options are enabled and I do "rmmod i82875p_edac", I
   get:

      edac debug: edac_pci_workq_function() checking
      BUG: unable to handle kernel paging request at f882d16f
      ...
      call trace:
       [<f8834df3>] ? edac_mc_workq_function+0x55/0x7e [edac_core]
       [<c0233974>] ? run_workqueue+0xd7/0x1a5
       [<c023392f>] ? run_workqueue+0x92/0x1a5
       [<f8834d9e>] ? edac_mc_workq_function+0x0/0x7e [edac_core]
       [<c0233af9>] ? worker_thread+0xb7/0xc3
       [<c0236a7b>] ? autoremove_wake_function+0x0/0x33
       [<c0233a42>] ? worker_thread+0x0/0xc3
       [<c0236809>] ? kthread+0x3b/0x61
       [<c02367ce>] ? kthread+0x0/0x61
       [<c0204587>] ? kernel_thread_helper+0x7/0x10

   Fix for this is to get rid of needles variable i82875p_registered
   altogether and run i82875p_remove_one() *before*
   pci_unregister_driver().

2. edac_mc_del_mc() uses mci after freeing mci

   edac_mc_del_mc() calls calls edac_remove_sysfs_mci_device().  The
   kobject refcount of mci drops to 0 and mci is freed.  After this
   mci is accessed via debug print and i82875p_remove_one() still
   uses mci->pvt and tries to free mci again with edac_mc_free().

   The fix for this is add kobject_get(&mci->edac_mci_kobj) after
   edac_mc_alloc(). Then the mci is still available after returning
   from edac_mc_del_mc() with refcount 1, and mci->pvt is still
   available. When i82875p_remove_one() finally calls edac_mc_free(),
   this will cause kobject_put() and mci is released properly.

Signed-off-by: Jarkko Lavinen <jlavi@iki.fi>
Cc: Doug Thompson <norsk5@yahoo.com>
Cc: Alan Cox <alan@lxorguk.ukuu.org.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
15 years agoi82875p_edac: fix overflow device resource setup
Jarkko Lavinen [Mon, 1 Dec 2008 21:14:06 +0000 (13:14 -0800)]
i82875p_edac: fix overflow device resource setup

When I do "modprobe i82875p_edac" on my Asus P4C800 MB on kernels 2.6.26
or later, the module load fails due to BAR 0 collision.  On 2.6.25 the
module loads just fine.

The overflow device on the MB seems to be hidden and its resources are not
allocated at normal PCI bus init.  Log shows the missing resource problem:

  EDAC DEBUG: i82875p_probe1()
  PCI: 0000:00:06.0 reg 10 32bit mmio: [fecf0000fecf0fff]
  pci 0000:00:06.0: device not available because of BAR 0
[0xfecf0000-0xfecf0fff] collisions
  EDAC i82875p: i82875p_setup_overfl_dev(): Failed to enable overflow
device

The patch below fixes this by calling pci_bus_assign_resources() after
the overflow device is revealed and added to the bus. With this patch
I am again able to load and use the module.

Signed-off-by: Jarkko Lavinen <jlavi@iki.fi>
Cc: Doug Thompson <norsk5@yahoo.com>
Cc: Jesse Barnes <jbarnes@virtuousgeek.org>
Cc: Alan Cox <alan@lxorguk.ukuu.org.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
15 years agofbdev: fix FB console blanking
Dmitry Baryshkov [Mon, 1 Dec 2008 21:14:05 +0000 (13:14 -0800)]
fbdev: fix FB console blanking

The commit aef7db4bd5a3b6068dfa05919a3d685199eed116 fixed the problem with
recursive locking in fb blanking code if blank is caused by user setting
the /sys/class/graphics/fb*/blank.  However this broke the fbcon timeout
blanking.

If you use a driver that defines ->fb_blank operation and at the same time
that driver relies on other driver (e.g.  backlight or lcd class) to blank
the screen, when the fbcon times out and tries to blank the fb, it will
call only fb driver blanker and won't notify the other driver.  Thus FB
output is disabled, but the screen isn't blanked.

Restore fbcon blanking and at the same time apply the proper fix for the
above problem: if fbcon_blank is called with FBINFO_FLAG_USEREVENT, we are
already called through notification from fb_blank, thus we don't have to
blank the fb again.

Signed-off-by: Dmitry Baryshkov <dbaryshkov@gmail.com>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
15 years agontfs: don't fool kernel-doc
Randy Dunlap [Mon, 1 Dec 2008 21:14:04 +0000 (13:14 -0800)]
ntfs: don't fool kernel-doc

kernel-doc handles macros now (it has for quite some time), so change the
ntfs_debug() macro's kernel-doc to be just before the macro instead of
before a phony function prototype.

[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com>
Cc: Anton Altaparmakov <aia21@cantab.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
15 years agokernel-doc: handle varargs cleanly
Randy Dunlap [Mon, 1 Dec 2008 21:14:03 +0000 (13:14 -0800)]
kernel-doc: handle varargs cleanly

The method for listing varargs in kernel-doc notation is:
 * @...: these arguments are printed by the @fmt argument

but scripts/kernel-doc is confused:  it always lists varargs as:
... variable arguments
and ignores the @...: line's description, but then prints that
line after the list of function parameters as though it's
not part of the function parameters.

This patch makes kernel-doc print the supplied @...  description if it is
present; otherwise a boilerplate "variable arguments" is printed.

Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
15 years agolib/idr.c: fix rcu related race with idr_find
Manfred Spraul [Mon, 1 Dec 2008 21:14:02 +0000 (13:14 -0800)]
lib/idr.c: fix rcu related race with idr_find

2nd part of the fixes needed for
http://bugzilla.kernel.org/show_bug.cgi?id=11796.

When the idr tree is either grown or shrunk, then the update to the number
of layers and the top pointer were not atomic.  This race caused crashes.

The attached patch fixes that by replicating the layers counter in each
layer, thus idr_find doesn't need idp->layers anymore.

Signed-off-by: Manfred Spraul <manfred@colorfullife.com>
Cc: Clement Calmels <cboulte@gmail.com>
Cc: Nadia Derbey <Nadia.Derbey@bull.net>
Cc: Pierre Peiffer <peifferp@gmail.com>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
15 years agoDMA-API.txt: fix description of pci_map_sg/dma_map_sg scatterlists handling
FUJITA Tomonori [Mon, 1 Dec 2008 21:14:01 +0000 (13:14 -0800)]
DMA-API.txt: fix description of pci_map_sg/dma_map_sg scatterlists handling

- pci_map_sg/dma_map_sg are used with a scatter gather list that doesn't
  come from the block layer (e.g.  some network drivers do).

- how IOMMUs merge adjacent elements of the scatter/gather list is
  independent of how the block layer determines sees elements.

Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Cc: James Bottomley <James.Bottomley@HansenPartnership.com>
Cc: Tejun Heo <htejun@gmail.com>
Cc: Jens Axboe <jens.axboe@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
15 years agofrv: fix mmap2 error handling
David Howells [Mon, 1 Dec 2008 21:14:00 +0000 (13:14 -0800)]
frv: fix mmap2 error handling

Fix the error handling in sys_mmap2().  Currently, if the pgoff check
fails, fput() might have to be called (which it isn't), so do the pgoff
check first, before fget() is called.

Signed-off-by: David Howells <dhowells@redhat.com>
Reported-by: Julia Lawall <julia@diku.dk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
15 years agotaint: add missing comment
Arjan van de Ven [Mon, 1 Dec 2008 21:14:00 +0000 (13:14 -0800)]
taint: add missing comment

The description for 'D' was missing in the comment...  (causing me a
minute of WTF followed by looking at more of the code)

Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
15 years agoradeonfb: fix problem with color expansion & alignment
Benjamin Herrenschmidt [Mon, 1 Dec 2008 21:13:58 +0000 (13:13 -0800)]
radeonfb: fix problem with color expansion & alignment

The engine on some radeon variants locks up if color expansion is called
for non aligned source data.  This patch enables a feature of the core
fbdev to request aligned input pixmaps and uses the HW clipping engine to
clip the output to the requested size

Addresses http://bugzilla.kernel.org/show_bug.cgi?id=11875

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Tested-by: James Cloos <cloos@jhcloos.com>
Cc: "Rafael J. Wysocki" <rjw@sisk.pl>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Krzysztof Helt <krzysztof.h1@poczta.fm>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
15 years agospi: fix spi_s3c24xx_gpio num_chipselect
Ben Dooks [Mon, 1 Dec 2008 21:13:57 +0000 (13:13 -0800)]
spi: fix spi_s3c24xx_gpio num_chipselect

The spi master driver must have num_chipselect set to allow the bus to
initialise.  Pass this through the platform data.

Signed-off-by: Ben Dooks <ben-linux@fluff.org>
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
15 years agospi: fix spi_s3c24xx_gpio device handle lookup
Ben Dooks [Mon, 1 Dec 2008 21:13:56 +0000 (13:13 -0800)]
spi: fix spi_s3c24xx_gpio device handle lookup

The spidev_to_sg() call in spi_s3c24xx_gpio.c was using the wrong method
to convert the spi device into the private data for the driver.  Fix this
by using spi_master_get_devdata.

Signed-off-by: Ben Dooks <ben-linux@fluff.org>
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
15 years agospi: au1550_spi full duplex dma fix
Jan Nikitenko [Mon, 1 Dec 2008 21:13:56 +0000 (13:13 -0800)]
spi: au1550_spi full duplex dma fix

Fix unsafe order in dma mapping operation: always flush data from the
cache *BEFORE* invalidating it, to allow full duplex transfers where the
same buffer may be used for both writes and reads.  Tested with mmc-spi.

Signed-off-by: Jan Nikitenko <jan.nikitenko@gmail.com>
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
15 years agospi: fix spi_imx probe oopsing
Julien Boibessot [Mon, 1 Dec 2008 21:13:55 +0000 (13:13 -0800)]
spi: fix spi_imx probe oopsing

Corrects spi_imx driver oops during initialization/probing: can't use
drv_data before it's allocated.

Signed-off-by: Julien Boibessot <julien.boibessot@armadeus.com>
Acked-by: Sascha Hauer <s.hauer@pengutronix.de>
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
15 years agoepoll: introduce resource usage limits
Davide Libenzi [Mon, 1 Dec 2008 21:13:55 +0000 (13:13 -0800)]
epoll: introduce resource usage limits

It has been thought that the per-user file descriptors limit would also
limit the resources that a normal user can request via the epoll
interface.  Vegard Nossum reported a very simple program (a modified
version attached) that can make a normal user to request a pretty large
amount of kernel memory, well within the its maximum number of fds.  To
solve such problem, default limits are now imposed, and /proc based
configuration has been introduced.  A new directory has been created,
named /proc/sys/fs/epoll/ and inside there, there are two configuration
points:

  max_user_instances = Maximum number of devices - per user

  max_user_watches   = Maximum number of "watched" fds - per user

The current default for "max_user_watches" limits the memory used by epoll
to store "watches", to 1/32 of the amount of the low RAM.  As example, a
256MB 32bit machine, will have "max_user_watches" set to roughly 90000.
That should be enough to not break existing heavy epoll users.  The
default value for "max_user_instances" is set to 128, that should be
enough too.

This also changes the userspace, because a new error code can now come out
from EPOLL_CTL_ADD (-ENOSPC).  The EMFILE from epoll_create() was already
listed, so that should be ok.

[akpm@linux-foundation.org: use get_current_user()]
Signed-off-by: Davide Libenzi <davidel@xmailserver.org>
Cc: Michael Kerrisk <mtk.manpages@gmail.com>
Cc: <stable@kernel.org>
Cc: Cyrill Gorcunov <gorcunov@gmail.com>
Reported-by: Vegard Nossum <vegardno@ifi.uio.no>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
15 years agospi: mpc52xx_psc_spi chipselect bugfix
Stefano Babic [Mon, 1 Dec 2008 21:13:53 +0000 (13:13 -0800)]
spi: mpc52xx_psc_spi chipselect bugfix

According to the manual the "tdfOnExit" flag must be set on the last byte
we want to send.  The PSC controller holds SS low until the flag is set.

However, the flag was set always on the last byte of the FIFO,
independently if it is the last byte of the transfer.  This generates
spurious toggling of the SS signals that breaks the protocol of some
peripherals.  Fix.

Signed-off-by: Stefano Babic <sbabic@denx.de>
Acked-by: Grant Likely <grant.likely@secretlab.ca>
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
15 years agospi: avoid spidev crash when device is removed
Wolfgang Ocker [Mon, 1 Dec 2008 21:13:52 +0000 (13:13 -0800)]
spi: avoid spidev crash when device is removed

I saw a kernel oops in spidev_remove() when a spidev device was registered
and I unloaded the SPI master driver:

Unable to handle kernel paging request for data at address 0x00000004
Faulting instruction address: 0xc01c0c50
Oops: Kernel access of bad area, sig: 11 [#1]
CDSPR
Modules linked in: spi_ppc4xx(-)
NIP: c01c0c50 LR: c01bf9e4 CTR: c01c0c34
REGS: cec89c30 TRAP: 0300   Not tainted  (2.6.27.3izt)
MSR: 00021000 <ME>  CR: 24000228  XER: 20000007
DEAR: 00000004, ESR: 00800000
TASK = cf889040[2070] 'rmmod' THREAD: cec88000
GPR00: 00000000 cec89ce0 cf889040 cec8e000 00000004 cec8e000 ffffffff 00000000
GPR08: 0000001c c0336380 00000000 c01c0c34 00000001 1001a338 100e0000 100df49c
GPR16: 100b54c0 100df49c 100ddd20 100f05a8 100b5340 100efd68 00000000 00000000
GPR24: 100ec008 100f0428 c0327788 c0327794 cec8e0ac cec8e000 c0336380 00000000
NIP [c01c0c50] spidev_remove+0x1c/0xe4
LR [c01bf9e4] spi_drv_remove+0x2c/0x3c
Call Trace:
[cec89d00] [c01bf9e4] spi_drv_remove+0x2c/0x3c
[cec89d10] [c01859a0] __device_release_driver+0x78/0xb4
[cec89d20] [c0185ab0] device_release_driver+0x28/0x44
[cec89d40] [c0184be8] bus_remove_device+0xac/0xd8
[cec89d60] [c0183094] device_del+0x100/0x194
[cec89d80] [c0183140] device_unregister+0x18/0x30
[cec89da0] [c01bf30c] __unregister+0x20/0x34
[cec89db0] [c0182778] device_for_each_child+0x38/0x74
[cec89de0] [c01bf2d0] spi_unregister_master+0x28/0x44
[cec89e00] [c01bfeac] spi_bitbang_stop+0x1c/0x58
[cec89e20] [d908a5e0] spi_ppc4xx_of_remove+0x24/0x7c [spi_ppc4xx]
[...]

IMHO a call to spi_set_drvdata() is missing in spidev_probe(). The patch
below helped.

Signed-off-by: Wolfgang Ocker <weo@reccoware.de>
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
15 years agospi documentation: use __initdata on struct
roel kluin [Mon, 1 Dec 2008 21:13:51 +0000 (13:13 -0800)]
spi documentation: use __initdata on struct

Use __initdata for data, not __init.

Signed-off-by: Roel Kluin <roel.kluin@gmail.com>
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
15 years agohwmon: applesmc: make applesmc load automatically on startup
Henrik Rydberg [Mon, 1 Dec 2008 21:13:49 +0000 (13:13 -0800)]
hwmon: applesmc: make applesmc load automatically on startup

make use of the new dmi device loading support to automatically load the
applesmc driver based on the dmi_match table.

Signed-off-by: Henrik Rydberg <rydberg@euromail.se>
Cc: Nicolas Boichat <nicolas@boichat.ch>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
15 years agoparport_serial: fix array overflow
Takashi Iwai [Mon, 1 Dec 2008 21:13:49 +0000 (13:13 -0800)]
parport_serial: fix array overflow

The netmos_9xx5_combo type assumes that PCI SSID provides always the
correct value for the number of parallel and serial ports, but there are
indeed broken devices with wrong numbers, which may result in Oops.

This patch simply adds the check of the array range.

Reference: Novell bnc#447067
https://bugzilla.novell.com/show_bug.cgi?id=447067

Signed-off-by: Takashi Iwai <tiwai@suse.de>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
15 years agomemcg: memory hotplug fix for notifier callback
KAMEZAWA Hiroyuki [Mon, 1 Dec 2008 21:13:48 +0000 (13:13 -0800)]
memcg: memory hotplug fix for notifier callback

Fixes for memcg/memory hotplug.

While memory hotplug allocate/free memmap, page_cgroup doesn't free
page_cgroup at OFFLINE when page_cgroup is allocated via bootomem.
(Because freeing bootmem requires special care.)

Then, if page_cgroup is allocated by bootmem and memmap is freed/allocated
by memory hotplug, page_cgroup->page == page is no longer true.

But current MEM_ONLINE handler doesn't check it and update
page_cgroup->page if it's not necessary to allocate page_cgroup.  (This
was not found because memmap is not freed if SPARSEMEM_VMEMMAP is y.)

And I noticed that MEM_ONLINE can be called against "part of section".
So, freeing page_cgroup at CANCEL_ONLINE will cause trouble.  (freeing
used page_cgroup) Don't rollback at CANCEL.

One more, current memory hotplug notifier is stopped by slub because it
sets NOTIFY_STOP_MASK to return vaule.  So, page_cgroup's callback never
be called.  (low priority than slub now.)

I think this slub's behavior is not intentional(BUG). and fixes it.

Another way to be considered about page_cgroup allocation:
  - free page_cgroup at OFFLINE even if it's from bootmem
    and remove specieal handler. But it requires more changes.

Addresses http://bugzilla.kernel.org/show_bug.cgi?id=12041

Signed-off-by: KAMEZAWA Hiruyoki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Tested-by: Badari Pulavarty <pbadari@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
15 years agomm: vmalloc fix lazy unmapping cache aliasing
Nick Piggin [Mon, 1 Dec 2008 21:13:47 +0000 (13:13 -0800)]
mm: vmalloc fix lazy unmapping cache aliasing

Jim Radford has reported that the vmap subsystem rewrite was sometimes
causing his VIVT ARM system to behave strangely (seemed like going into
infinite loops trying to fault in pages to userspace).

We determined that the problem was most likely due to a cache aliasing
issue.  flush_cache_vunmap was only being called at the moment the page
tables were to be taken down, however with lazy unmapping, this can happen
after the page has subsequently been freed and allocated for something
else.  The dangling alias may still have dirty data attached to it.

The fix for this problem is to do the cache flushing when the caller has
called vunmap -- it would be a bug for them to write anything else to the
mapping at that point.

That appeared to solve Jim's problems.

Reported-by: Jim Radford <radford@blackbean.org>
Signed-off-by: Nick Piggin <npiggin@suse.de>
Cc: Russell King <rmk@arm.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
15 years agoMerge branch 'upstream-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mfashe...
Linus Torvalds [Tue, 2 Dec 2008 02:56:55 +0000 (18:56 -0800)]
Merge branch 'upstream-linus' of git://git./linux/kernel/git/mfasheh/ocfs2

* 'upstream-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mfasheh/ocfs2:
  ocfs2: fix regression in ocfs2_read_blocks_sync()
  ocfs2: fix return value set in init_dlmfs_fs()
  ocfs2: Small documentation update
  ocfs2: fix wake_up in unlock_ast
  ocfs2: initialize stack_user lvbptr
  ocfs2: comments typo fix

15 years agoocfs2: fix regression in ocfs2_read_blocks_sync()
Mark Fasheh [Fri, 21 Nov 2008 22:06:55 +0000 (14:06 -0800)]
ocfs2: fix regression in ocfs2_read_blocks_sync()

We're panicing in ocfs2_read_blocks_sync() if a jbd-managed buffer is seen.
At first glance, this seems ok but in reality it can happen. My test case
was to just run 'exorcist'. A struct inode is being pushed out of memory but
is then re-read at a later time, before the buffer has been checkpointed by
jbd. This causes a BUG to be hit in ocfs2_read_blocks_sync().

Reviewed-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: fix return value set in init_dlmfs_fs()
Coly Li [Mon, 17 Nov 2008 04:38:22 +0000 (12:38 +0800)]
ocfs2: fix return value set in init_dlmfs_fs()

In init_dlmfs_fs(), if calling kmem_cache_create() failed, the code will use return value from
calling bdi_init(). The correct behavior should be set status as -ENOMEM before going to "bail:".

Signed-off-by: Coly Li <coyli@suse.de>
Acked-by: Sunil Mushran <sunil.mushran@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: Small documentation update
Mark Fasheh [Tue, 18 Nov 2008 23:08:42 +0000 (15:08 -0800)]
ocfs2: Small documentation update

Remove some features from the "not-supported" list that are actually
supported now.

Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: fix wake_up in unlock_ast
David Teigland [Mon, 17 Nov 2008 18:28:48 +0000 (12:28 -0600)]
ocfs2: fix wake_up in unlock_ast

In ocfs2_unlock_ast(), call wake_up() on lockres before releasing
the spin lock on it.  As soon as the spin lock is released, the
lockres can be freed.

Signed-off-by: David Teigland <teigland@redhat.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: initialize stack_user lvbptr
David Teigland [Mon, 10 Nov 2008 22:24:57 +0000 (16:24 -0600)]
ocfs2: initialize stack_user lvbptr

The locking_state dump, ocfs2_dlm_seq_show, reads the lvb on locks where it
has not yet been initialized by a lock call.

Signed-off-by: David Teigland <teigland@redhat.com>
Acked-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: comments typo fix
Coly Li [Wed, 5 Nov 2008 07:16:24 +0000 (15:16 +0800)]
ocfs2: comments typo fix

This patch fixes two typos in comments of ocfs2.

Signed-off-by: Coly Li <coyli@suse.de>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoem28xx: remove backward compat macro added on a previous fix
Mauro Carvalho Chehab [Mon, 1 Dec 2008 20:04:14 +0000 (18:04 -0200)]
em28xx: remove backward compat macro added on a previous fix

commit 50f3beb50abe0cc0228363af804e50e710b3e5b0 fixed em28xx-alsa
locking schema. However, a backport macro was kept.

This patch removes the macro, since it is not needed for the module
compilation against upstream.

Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
15 years agoMerge branch 'upstream-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jgarzi...
Linus Torvalds [Mon, 1 Dec 2008 19:23:33 +0000 (11:23 -0800)]
Merge branch 'upstream-linus' of git://git./linux/kernel/git/jgarzik/libata-dev

* 'upstream-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jgarzik/libata-dev:
  libata: blacklist Seagate drives which time out FLUSH_CACHE when used with NCQ
  [libata] pata_rb532_cf: fix signature of the xfer function
  [libata] pata_rb532_cf: fix and rename register definitions
  ata_piix: add borked Tecra M4 to broken suspend list

15 years agoV4L/DVB (9748): em28xx: fix compile warning
Hans Verkuil [Mon, 24 Nov 2008 12:53:22 +0000 (09:53 -0300)]
V4L/DVB (9748): em28xx: fix compile warning

Label fail_unreg is no longer used.

Signed-off-by: Hans Verkuil <hverkuil@xs4all.nl>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
15 years agoV4L/DVB (9743): em28xx: fix oops audio
Douglas Schilling Landgraf [Mon, 24 Nov 2008 12:51:20 +0000 (09:51 -0300)]
V4L/DVB (9743): em28xx: fix oops audio

Replaced usb_kill_usb for usb_unlink_usb
(wait until urb to fully stop require USB core to put the calling process to sleep).

Oops:
http://www.kerneloops.org/raw.php?rawid=71799&msgid=

Signed-off-by: Douglas Schilling Landgraf <dougsland@linuxtv.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
15 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/roland...
Linus Torvalds [Mon, 1 Dec 2008 19:01:54 +0000 (11:01 -0800)]
Merge branch 'for-linus' of git://git./linux/kernel/git/roland/infiniband

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/roland/infiniband:
  IB/mlx4: Fix MTT leakage in resize CQ
  IB/ehca: Fix problem with generated flush work completions
  IB/ehca: Change misleading error message on memory hotplug
  mlx4_core: Save/restore default port IB capability mask

15 years agolibata: blacklist Seagate drives which time out FLUSH_CACHE when used with NCQ
Tejun Heo [Thu, 27 Nov 2008 04:36:48 +0000 (13:36 +0900)]
libata: blacklist Seagate drives which time out FLUSH_CACHE when used with NCQ

Some recent Seagate harddrives have firmware bug which causes FLUSH
CACHE to timeout under certain circumstances if NCQ is being used.
This can be worked around by disabling NCQ and fixed by updating the
firmware.  Implement ATA_HORKAGE_FIRMWARE_UPDATE and blacklist these
devices.

The wiki page has been updated to contain information on this issue.

  http://ata.wiki.kernel.org/index.php/Known_issues

Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
15 years ago[libata] pata_rb532_cf: fix signature of the xfer function
Phil Sutter [Fri, 28 Nov 2008 19:48:35 +0000 (20:48 +0100)]
[libata] pata_rb532_cf: fix signature of the xfer function

Per definition, this function should return the number of bytes
consumed. As the original parameter "buflen" is being decremented inside
the read/write loop, save it in "retlen" at the beginning.

Signed-off-by: Phil Sutter <n0-1@freewrt.org>
Acked-by: Sergei Shtyltov <sshtylyov@ru.mvista.com>
Acked-by: Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>
Acked-by: Florian Fainelli <florian@openwrt.org>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
15 years ago[libata] pata_rb532_cf: fix and rename register definitions
Phil Sutter [Fri, 28 Nov 2008 19:48:26 +0000 (20:48 +0100)]
[libata] pata_rb532_cf: fix and rename register definitions

The original standalone driver uses a custom address for the error
register. Use it in pata_rb532_cf, too.

Rename two register definitions:
- The address offset 0x0800 in fact is the ATA base, not ATA command
  address.
- The offset 0x0C00 is not a regular ATA data address, but a buffered one
  allowing 4-byte IO.

Signed-off-by: Phil Sutter <n0-1@freewrt.org>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
15 years agoata_piix: add borked Tecra M4 to broken suspend list
Tejun Heo [Sat, 29 Nov 2008 13:37:21 +0000 (22:37 +0900)]
ata_piix: add borked Tecra M4 to broken suspend list

Tecra M4 sometimes forget what it is and reports bogus data via DMI
which makes the machine evade broken suspend matching and thus fail
suspend/resume.  This patch updates piix_broken_suspend() such that it
can match such case.  As the borked DMI data is a bit generic,
matching many entries to make the match more specific is necessary.
As the usual DMI matching is limited to four entries, this patch uses
hard coded manual matching.

This is reported by Alexandru Romanescu.

Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Alexandru Romanescu <a_romanescu@yahoo.co.uk>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
15 years agoMerge branches 'ehca' and 'mlx4' into for-linus
Roland Dreier [Mon, 1 Dec 2008 18:11:50 +0000 (10:11 -0800)]
Merge branches 'ehca' and 'mlx4' into for-linus

15 years agoIB/mlx4: Fix MTT leakage in resize CQ
Jack Morgenstein [Mon, 1 Dec 2008 18:09:37 +0000 (10:09 -0800)]
IB/mlx4: Fix MTT leakage in resize CQ

When resizing a CQ, MTTs associated with the old CQE buffer were not
freed.  As a result, if any app used resize CQ repeatedly, all MTTs
were eventually exhausted, which led to all memory registration
operations failing until the driver is reloaded.

Once the RESIZE_CQ command returns successfully from FW, FW no longer
accesses the old CQ buffer, so it is safe to deallocate the MTT
entries used by the old CQ buffer.

Finally, if the RESIZE_CQ command fails, the MTTs allocated for the
new CQEs buffer also need to be de-allocated.

This fixes <https://bugs.openfabrics.org/show_bug.cgi?id=1416>.

Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
15 years agoIB/ehca: Fix problem with generated flush work completions
Stefan Roscher [Mon, 1 Dec 2008 18:05:50 +0000 (10:05 -0800)]
IB/ehca: Fix problem with generated flush work completions

This fix enables ehca device driver to generate flush work completions
even if the application doesn't request completions for all work
requests. The current implementation of ehca will generate flush work
completions for the wrong work requests if an application uses non
signaled work completions.

Signed-off-by: Stefan Roscher <stefan.roscher@de.ibm.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
15 years agoIB/ehca: Change misleading error message on memory hotplug
Joachim Fenkes [Mon, 1 Dec 2008 18:05:44 +0000 (10:05 -0800)]
IB/ehca: Change misleading error message on memory hotplug

The error message printed when the eHCA driver prevents memory hotplug
is misleading -- the user might think that hot-removing the lhca,
hotplugging memory, then hot-adding the lhca again will work, but it
actually doesn't.

Signed-off-by: Joachim Fenkes <fenkes@de.ibm.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
15 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ieee1394...
Linus Torvalds [Mon, 1 Dec 2008 17:34:23 +0000 (09:34 -0800)]
Merge branch 'for-linus' of git://git./linux/kernel/git/ieee1394/linux1394-2.6

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ieee1394/linux1394-2.6:
  ieee1394: sbp2: fix race condition in state change
  ieee1394: fix list corruption (reported at module removal)
  firewire: fw-sbp2: another iPod mini quirk entry
  ieee1394: sbp2: another iPod mini quirk entry

15 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jikos/hid
Linus Torvalds [Mon, 1 Dec 2008 16:33:59 +0000 (08:33 -0800)]
Merge branch 'for-linus' of git://git./linux/kernel/git/jikos/hid

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jikos/hid:
  HID: Apple ALU wireless keyboards are bluetooth devices
  HID: remove setup mutex, fix possible deadlock
  HID: add USB ID for another dual gameron adapter
  HID: unignore mouse on unibody macbooks
  HID: fix blacklist entries for greenasia/pantherlord

15 years agoAdd kref to fake tty used by USB console
Kevin Hao [Mon, 1 Dec 2008 11:36:16 +0000 (11:36 +0000)]
Add kref to fake tty used by USB console

We alloc a fake tty in usb serial console setup function. we should
init the tty's kref otherwise we will face WARN_ON after following
invoke of tty_port_tty_set --> tty_kref_get.

Signed-off-by: Kevin Hao <kexin.hao@windriver.com>
Signed-off-by: Alan Cox <alan@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
15 years agodrivers/char/tty_io.c: Avoid panic when no console is configured.
Will Newton [Mon, 1 Dec 2008 11:36:06 +0000 (11:36 +0000)]
drivers/char/tty_io.c: Avoid panic when no console is configured.

When no console is configured tty_open tries to call kref_get on a NULL
pointer, return ENODEV instead.

Signed-off-by: Will Newton <will.newton@gmail.com>
Signed-off-by: Alan Cox <alan@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
15 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/usb-2.6
Linus Torvalds [Mon, 1 Dec 2008 15:58:49 +0000 (07:58 -0800)]
Merge git://git./linux/kernel/git/gregkh/usb-2.6

* git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/usb-2.6:
  USB: serial: add more Onda device ids to option driver
  USB: usb-storage: unusual_devs entry for Nikon D2H
  USB: storage: unusual_devs entry for Mio C520-GPS
  USB: fsl_usb2_udc: Report disconnect before unbinding
  USB: fsl_qe_udc: Report disconnect before unbinding
  USB: fix SB600 USB subsystem hang bug
  Revert "USB: improve ehci_watchdog's side effect in CPU power management"

15 years agoMerge branch 'merge' of git://git.kernel.org/pub/scm/linux/kernel/git/paulus/powerpc
Linus Torvalds [Mon, 1 Dec 2008 15:58:23 +0000 (07:58 -0800)]
Merge branch 'merge' of git://git./linux/kernel/git/paulus/powerpc

* 'merge' of git://git.kernel.org/pub/scm/linux/kernel/git/paulus/powerpc:
  powerpc: Fix build for 32-bit SMP configs

15 years agovmscan: protect zone rotation stats by lru lock
Johannes Weiner [Mon, 1 Dec 2008 02:00:35 +0000 (03:00 +0100)]
vmscan: protect zone rotation stats by lru lock

The zone's rotation statistics must not be accessed without the
corresponding LRU lock held.  Fix an unprotected write in
shrink_active_list().

Acked-by: Rik van Riel <riel@redhat.com>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Johannes Weiner <hannes@saeurebad.de>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
15 years agoRevert "of_platform_driver noise on sparce"
Linus Torvalds [Mon, 1 Dec 2008 15:55:14 +0000 (07:55 -0800)]
Revert "of_platform_driver noise on sparce"

This reverts commit e669dae6141ff97d3c7566207f5de3b487dcf837, since it
is incomplete, and clashes with fuller patches and the sparc 32/64
unification effort.

Requested-by: David Miller <davem@davemloft.net>
Acked-by: Al Viro <viro@ZenIV.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
15 years agoUSB: serial: add more Onda device ids to option driver
Greg Kroah-Hartman [Sat, 29 Nov 2008 19:46:21 +0000 (11:46 -0800)]
USB: serial: add more Onda device ids to option driver

Thanks to Domenico Riccio for pointing these out.

Cc: Domenico Riccio <domenico.riccio@gmail.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
15 years agoUSB: usb-storage: unusual_devs entry for Nikon D2H
Tobias Kunze Briseño [Mon, 24 Nov 2008 16:28:31 +0000 (11:28 -0500)]
USB: usb-storage: unusual_devs entry for Nikon D2H

This patch adds an unusual_devs entry for the Nikon D2H camera.

From: Tobias Kunze Briseño <t@fictive.com>,
Signed-off-by: Alan Stern <stern@rowland.harvard.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
15 years agoUSB: storage: unusual_devs entry for Mio C520-GPS
Alan Stern [Fri, 21 Nov 2008 21:15:12 +0000 (16:15 -0500)]
USB: storage: unusual_devs entry for Mio C520-GPS

This patch (as1176) adds an unusual_devs entry for the Mio C520 GPS
unit.  Other devices also based on the Mitac hardware use the same USB
interface firmware, so the Vendor and Product names are generalized.

This fixes Bugzilla #11583.

Signed-off-by: Alan Stern <stern@rowland.harvard.edu>
Tested-by: Tamas Kerecsen <kerecsen@bigfoot.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
15 years agoUSB: fsl_usb2_udc: Report disconnect before unbinding
Anton Vorontsov [Thu, 13 Nov 2008 12:00:46 +0000 (15:00 +0300)]
USB: fsl_usb2_udc: Report disconnect before unbinding

Gadgets disable endpoints in their disconnect callbacks, so
we must call disconnect before unbinding.

The patch fixes following badness:

root@b1:~# insmod fsl_usb2_udc.ko
Freescale High-Speed USB SOC Device Controller driver (Apr 20, 2007)
root@b1:~# insmod g_ether.ko
g_ether gadget: using random self ethernet address
g_ether gadget: using random host ethernet address
usb0: MAC 26:07:ba:c0:44:33
usb0: HOST MAC 96:81:0c:05:4d:e3
g_ether gadget: Ethernet Gadget, version: Memorial Day 2008
g_ether gadget: g_ether ready
fsl-usb2-udc: bind to driver g_ether
g_ether gadget: high speed config #1: CDC Ethernet (ECM)
root@b1:~# rmmod g_ether.ko
------------[ cut here ]------------
Badness at drivers/usb/gadget/composite.c:871
[...]
NIP [e10c3454] composite_unbind+0x24/0x15c [g_ether]
LR [e10aa454] usb_gadget_unregister_driver+0x13c/0x164 [fsl_usb2_udc]
Call Trace:
[df145e80] [ffffff94] 0xffffff94 (unreliable)
[df145eb0] [e10aa454] usb_gadget_unregister_driver+0x13c/0x164 [fsl_usb2_udc]
[df145ed0] [e10c4c40] usb_composite_unregister+0x3c/0x4c [g_ether]
[df145ee0] [c006bcc0] sys_delete_module+0x130/0x19c
[df145f40] [c00142d8] ret_from_syscall+0x0/0x38
[...]
unregistered gadget driver 'g_ether'

Signed-off-by: Anton Vorontsov <avorontsov@ru.mvista.com>
Acked-by: David Brownell <dbrownell@users.sourceforge.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
15 years agoUSB: fsl_qe_udc: Report disconnect before unbinding
Anton Vorontsov [Thu, 13 Nov 2008 11:57:20 +0000 (14:57 +0300)]
USB: fsl_qe_udc: Report disconnect before unbinding

Gadgets disable endpoints in their disconnect callbacks, so
we must call disconnect before unbinding. This also fixes
muram memory leak, since we free muram in the qe_ep_disable().

But mainly the patch fixes following badness:

root@b1:~# insmod fsl_qe_udc.ko
fsl_qe_udc: Freescale QE/CPM USB Device Controller driver, 1.0
fsl_qe_udc e01006c0.usb: QE USB controller initialized as device
root@b1:~# insmod g_ether.ko
g_ether gadget: using random self ethernet address
g_ether gadget: using random host ethernet address
usb0: MAC be:2d:3c:fa:be:f0
usb0: HOST MAC 62:b8:6a:df:38:66
g_ether gadget: Ethernet Gadget, version: Memorial Day 2008
g_ether gadget: g_ether ready
fsl_qe_udc e01006c0.usb: fsl_qe_udc bind to driver g_ether
g_ether gadget: high speed config #1: CDC Ethernet (ECM)
root@b1:~# rmmod g_ether.ko
------------[ cut here ]------------
Badness at drivers/usb/gadget/composite.c:871
[...]
NIP [d10c1374] composite_unbind+0x24/0x15c [g_ether]
LR [d10a82f4] usb_gadget_unregister_driver+0x128/0x168 [fsl_qe_udc]
Call Trace:
[cfb93e80] [cfb1f3a0] 0xcfb1f3a0 (unreliable)
[cfb93eb0] [d10a82f4] usb_gadget_unregister_driver+0x128/0x168 [fsl_qe_udc]
[cfb93ed0] [d10c2a3c] usb_composite_unregister+0x3c/0x4c [g_ether]
[cfb93ee0] [c006bde0] sys_delete_module+0x130/0x19c
[cfb93f40] [c00142d8] ret_from_syscall+0x0/0x38
[...]
fsl_qe_udc e01006c0.usb: unregistered gadget driver 'g_ether'

Signed-off-by: Anton Vorontsov <avorontsov@ru.mvista.com>
Acked-by: David Brownell <dbrownell@users.sourceforge.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
15 years agoUSB: fix SB600 USB subsystem hang bug
Shane Huang [Tue, 25 Nov 2008 07:12:33 +0000 (15:12 +0800)]
USB: fix SB600 USB subsystem hang bug

This patch is required for all AMD SB600 revisions to avoid USB subsystem hang
symptom. The USB subsystem hang symptom is observed when the system has
multiple USB devices connected to it. In some cases a USB hub may be required
to observe this symptom.

Reported in bugzilla as #11599, the similar patch for SB700 old revision is:
commit b09bc6cbae4dd3a2d35722668ef2c502a7b8b093

Reported-by: raffaele <ralfconn@tele2.it>
Tested-by: Roman Mamedov <roman@rm.pp.ru>
Signed-off-by: Shane Huang <shane.huang@amd.com>
Cc: stable <stable@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
15 years agoRevert "USB: improve ehci_watchdog's side effect in CPU power management"
Greg Kroah-Hartman [Tue, 25 Nov 2008 21:34:45 +0000 (13:34 -0800)]
Revert "USB: improve ehci_watchdog's side effect in CPU power management"

This reverts commit f0d781d59cb621e1795d510039df973d0f8b23fc.

It was the wrong thing to do, and does not really do what it said
it did.

Cc: Yi Yang <yi.y.yang@intel.com>
Cc: David Brownell <dbrownell@users.sourceforge.net>
Cc: Alan Stern <stern@rowland.harvard.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
15 years agopowerpc: Fix build for 32-bit SMP configs
Milton Miller [Sun, 16 Nov 2008 11:44:42 +0000 (11:44 +0000)]
powerpc: Fix build for 32-bit SMP configs

attr_smt_snooze_delay is only defined for CONFIG_PPC64, so protect the
attribute removal with the same condition.  This fixes this build error
on 32-bit SMP configurations:

/data/home/miltonm/next.git/arch/powerpc/kernel/sysfs.c: In function ‘unregister_cpu_online’:
/data/home/miltonm/next.git/arch/powerpc/kernel/sysfs.c:722: error: ‘attr_smt_snooze_delay’ undeclared (first use in this function)
/data/home/miltonm/next.git/arch/powerpc/kernel/sysfs.c:722: error: (Each undeclared identifier is reported only once
/data/home/miltonm/next.git/arch/powerpc/kernel/sysfs.c:722: error: for each function it appears in.)

Signed-off-by: Paul Mackerras <paulus@samba.org>
15 years agoMerge branch 'drm-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/airlied...
Linus Torvalds [Mon, 1 Dec 2008 00:45:13 +0000 (16:45 -0800)]
Merge branch 'drm-fixes' of git://git./linux/kernel/git/airlied/drm-2.6

* 'drm-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/airlied/drm-2.6:
  drm/i915: Fix copy'n'pasteo that broke VT switch if flushing was non-empty.

15 years agoMerge branch 'merge' of git://git.kernel.org/pub/scm/linux/kernel/git/paulus/powerpc
Linus Torvalds [Mon, 1 Dec 2008 00:44:18 +0000 (16:44 -0800)]
Merge branch 'merge' of git://git./linux/kernel/git/paulus/powerpc

* 'merge' of git://git.kernel.org/pub/scm/linux/kernel/git/paulus/powerpc:
  powerpc: Fix system calls on Cell entered with XER.SO=1
  powerpc/cell: Fix GDB watchpoints, again
  powerpc/mpic: Don't reset affinity for secondary MPIC on boot
  powerpc/cell/axon-msi: Retry on missing interrupt
  powerpc: Fix boot freeze on machine with empty memory node
  powerpc: Fix IRQ assignment for some PCIe devices
  powerpc/spufs: Fix spinning in spufs_ps_fault on signal
  powerpc/mpc832x_rdb: fix swapped ethernet ids
  powerpc: Use generic PHY driver for Marvell 88E1111 PHY on GE Fanuc SBC610
  powerpc/85xx: L2 cache size wrong in 8572DS dts
  powerpc/virtex: Update defconfigs
  powerpc/52xx: update defconfigs
  xsysace: Fix driver to use resource_size_t instead of unsigned long
  powerpc/virtex: fix various format/casting printk mismatches
  powerpc/mpc5200: fix bestcomm Kconfig dependencies
  powerpc/44x: Fix 460EX/460GT machine check handling
  powerpc/40x: Limit allocable DRAM during early mapping

15 years agoMerge master.kernel.org:/home/rmk/linux-2.6-arm
Linus Torvalds [Mon, 1 Dec 2008 00:39:06 +0000 (16:39 -0800)]
Merge master.kernel.org:/home/rmk/linux-2.6-arm

* master.kernel.org:/home/rmk/linux-2.6-arm:
  Allow architectures to override copy_user_highpage()
  [ARM] pxa/palmtx: misc fixes to use generic GPIO API
  ARM: OMAP: Fixes for suspend / resume GPIO wake-up handling
  [ARM] pxa/corgi: update default config to exclude tosa from being built
  [ARM] pxa/pcm990: use negative number for an invalid GPIO in camera data
  ARM: OMAP: Typo fix for clock_allow_idle
  ARM: OMAP: Remove broken LCD driver for SX1
  [ARM] 5335/1: pxa25x_udc: Fix is_vbus_present to return 1 or 0
  [ARM] pxa/MioA701: bluetooth resume fix
  [ARM] pxa/MioA701: fix memory corruption.

15 years agodrm/i915: Fix copy'n'pasteo that broke VT switch if flushing was non-empty.
Eric Anholt [Mon, 1 Dec 2008 00:23:21 +0000 (10:23 +1000)]
drm/i915: Fix copy'n'pasteo that broke VT switch if flushing was non-empty.

Introduced in the "Avoid BUG_ONs on VT switch" commit.

Signed-off-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Dave Airlie <airlied@redhat.com>
15 years agopowerpc: Fix system calls on Cell entered with XER.SO=1
Paul Mackerras [Sun, 30 Nov 2008 11:49:45 +0000 (11:49 +0000)]
powerpc: Fix system calls on Cell entered with XER.SO=1

It turns out that on Cell, on a kernel with CONFIG_VIRT_CPU_ACCOUNTING
= y, if a program sets the SO (summary overflow) bit in the XER and
then does a system call, the SO bit in CR0 will be set on return
regardless of whether the system call detected an error.  Since CR0.SO
is used as the error indication from the system call, this means that
all system calls appear to fail.

The reason is that the workaround for the timebase bug on Cell uses a
compare instruction.  With CONFIG_VIRT_CPU_ACCOUNTING = y, the
ACCOUNT_CPU_USER_ENTRY macro reads the timebase, so we end up doing a
compare instruction, which copies XER.SO to CR0.SO.  Since we were
doing this in the system call entry patch after clearing CR0.SO but
before saving the CR, this meant that the saved CR image had CR0.SO
set if XER.SO was set on entry.

This fixes it by moving the clearing of CR0.SO to after the
ACCOUNT_CPU_USER_ENTRY call in the system call entry path.

Signed-off-by: Paul Mackerras <paulus@samba.org>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Acked-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
15 years agopowerpc/cell: Fix GDB watchpoints, again
Arnd Bergmann [Fri, 28 Nov 2008 09:51:24 +0000 (09:51 +0000)]
powerpc/cell: Fix GDB watchpoints, again

An earlier patch from Jens Osterkamp attempted to fix GDB
watchpoints by enabling the DABRX register at boot time.
Unfortunately, this did not work on SMP setups, where
secondary CPUs were still using the power-on DABRX value.

This introduces the same change for secondary CPUs on cell
as well.

Reported-by: Ulrich Weigand <Ulrich.Weigand@de.ibm.com>
Tested-by: Ulrich Weigand <Ulrich.Weigand@de.ibm.com>
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Paul Mackerras <paulus@samba.org>
15 years agopowerpc/mpic: Don't reset affinity for secondary MPIC on boot
Arnd Bergmann [Fri, 28 Nov 2008 09:51:23 +0000 (09:51 +0000)]
powerpc/mpic: Don't reset affinity for secondary MPIC on boot

Kexec/kdump currently fails on the IBM QS2x blades when the kexec happens
on a CPU other than the initial boot CPU.  It turns out that this is the
result of mpic_init trying to set affinity of each interrupt vector to the
current boot CPU.

As far as I can tell,  the same problem is likely to exist on any
secondary MPIC, because they have to deliver interrupts to the first
output all the time. There are two potential solutions for this: either
not set up affinity at all for secondary MPICs, or assume that a single
CPU output is connected to the upstream interrupt controller and hardcode
affinity to that per architecture.

This patch implements the second approach, defaulting to the first output.
Currently, all known secondary MPICs are routed to their upstream port
using the first destination, so we hardcode that.

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Paul Mackerras <paulus@samba.org>
15 years agopowerpc/cell/axon-msi: Retry on missing interrupt
Arnd Bergmann [Fri, 28 Nov 2008 09:51:22 +0000 (09:51 +0000)]
powerpc/cell/axon-msi: Retry on missing interrupt

The MSI capture logic for the axon bridge can sometimes
lose interrupts in case of high DMA and interrupt load,
when it signals an MSI interrupt to the MPIC interrupt
controller while we are already handling another MSI.

Each MSI vector gets written into a FIFO buffer in main
memory using DMA, and that DMA access is normally flushed
by the actual interrupt packet on the IOIF.  An MMIO
register in the MSIC holds the position of the last
entry in the FIFO buffer that was written.  However,
reading that position does not flush the DMA, so that
we can observe stale data in the buffer.

In a stress test, we have observed the DMA to arrive
up to 14 microseconds after reading the register.

This patch works around this problem by retrying the
access to the FIFO buffer.

We can reliably detect the conditioning by writing
an invalid MSI vector into the FIFO buffer after
reading from it, assuming that all MSIs we get
are valid.  After detecting an invalid MSI vector,
we udelay(1) in the interrupt cascade for up to
100 times before giving up.

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Paul Mackerras <paulus@samba.org>
15 years agopowerpc: Fix boot freeze on machine with empty memory node
Dave Hansen [Mon, 24 Nov 2008 12:02:35 +0000 (12:02 +0000)]
powerpc: Fix boot freeze on machine with empty memory node

I got a bug report about a distro kernel not booting on a particular
machine.  It would freeze during boot:

> ...
> Could not find start_pfn for node 1
> [boot]0015 Setup Done
> Built 2 zonelists in Node order, mobility grouping on.  Total pages: 123783
> Policy zone: DMA
> Kernel command line:
> [boot]0020 XICS Init
> [boot]0021 XICS Done
> PID hash table entries: 4096 (order: 12, 32768 bytes)
> clocksource: timebase mult[7d0000] shift[22] registered
> Console: colour dummy device 80x25
> console handover: boot [udbg0] -> real [hvc0]
> Dentry cache hash table entries: 1048576 (order: 7, 8388608 bytes)
> Inode-cache hash table entries: 524288 (order: 6, 4194304 bytes)
> freeing bootmem node 0

I've reproduced this on 2.6.27.7.  It is caused by commit
8f64e1f2d1e09267ac926e15090fd505c1c0cbcb ("powerpc: Reserve in bootmem
lmb reserved regions that cross NUMA nodes").

The problem is that Jon took a loop which was (in pseudocode):

for_each_node(nid)
NODE_DATA(nid) = careful_alloc(nid);
setup_bootmem(nid);
reserve_node_bootmem(nid);

and broke it up into:

for_each_node(nid)
NODE_DATA(nid) = careful_alloc(nid);
setup_bootmem(nid);
for_each_node(nid)
reserve_node_bootmem(nid);

The issue comes in when the 'careful_alloc()' is called on a node with
no memory.  It falls back to using bootmem from a previously-initialized
node.  But, bootmem has not yet been reserved when Jon's patch is
applied.  It gives back bogus memory (0xc000000000000000) and pukes
later in boot.

The following patch collapses the loop back together.  It also breaks
the mark_reserved_regions_for_nid() code out into a function and adds
some comments.  I think a huge part of introducing this bug is because
for loop was too long and hard to read.

The actual bug fix here is the:

+ if (end_pfn <= node->node_start_pfn ||
+     start_pfn >= node_end_pfn)
+ continue;

Signed-off-by: Dave Hansen <dave@linux.vnet.ibm.com>
Signed-off-by: Paul Mackerras <paulus@samba.org>