10 years agoadd pandora defconfigs pandora-37
Grazvydas Ignotas [Sun, 16 Jan 2011 22:50:30 +0000 (00:50 +0200)]
add pandora defconfigs

There are 2 defconfigs now:
- arch/arm/configs/omap3_pandora_defconfig
  proposed kernel config to be used for kernel in firmware
- arch/arm/configs/omap3_pandora_debug_defconfig
  defconfig with debugging options enabled and most stuff built in
  for quick debug sessions.

10 years agopower_supply: twl4030: allow USB charging by default
Grazvydas Ignotas [Sat, 26 Feb 2011 18:00:08 +0000 (20:00 +0200)]
power_supply: twl4030: allow USB charging by default

This is disabled in mainline because we don't know amount of current
host device can provide, but enabling it here regardless like we did
with 2.6.27.

10 years agopandora: setup fn keys for hacked twl4030_keypad
Grazvydas Ignotas [Sat, 26 Jun 2010 20:10:18 +0000 (23:10 +0300)]
pandora: setup fn keys for hacked twl4030_keypad

10 years agoinput: add hackish fn handling to twl4030_keypad
Grazvydas Ignotas [Sat, 26 Jun 2010 20:03:59 +0000 (23:03 +0300)]
input: add hackish fn handling to twl4030_keypad

10 years agopandora: add leds_pwm platform data
Grazvydas Ignotas [Sun, 27 Jun 2010 18:36:11 +0000 (21:36 +0300)]
pandora: add leds_pwm platform data

10 years agoleds: add TWL4030 PWM LED driver
Grazvydas Ignotas [Sun, 27 Jun 2010 16:59:03 +0000 (19:59 +0300)]
leds: add TWL4030 PWM LED driver

Some code contributed by Urja Rannikko <>

10 years agopandora: add backlight platform data
Grazvydas Ignotas [Mon, 21 Jun 2010 21:43:25 +0000 (00:43 +0300)]
pandora: add backlight platform data

10 years agobacklight: add backlight driver for pandora
Grazvydas Ignotas [Mon, 21 Jun 2010 21:29:10 +0000 (00:29 +0300)]
backlight: add backlight driver for pandora

This is fine-tuned driver for pandora's backlight connected to
TWL4030 PWM0 and LED driver chip.

10 years agopandora: add vsense platform data
Grazvydas Ignotas [Sun, 20 Jun 2010 19:30:21 +0000 (22:30 +0300)]
pandora: add vsense platform data

10 years agoinput: add VSense (analog controller) driver
Grazvydas Ignotas [Sun, 20 Jun 2010 18:34:00 +0000 (21:34 +0300)]
input: add VSense (analog controller) driver

This drives analog nubs on pandora.

10 years agopandora: create proc entry early
Grazvydas Ignotas [Tue, 22 Jun 2010 20:40:52 +0000 (23:40 +0300)]
pandora: create proc entry early

this will avoid needing to handle it in drivers.

10 years agoadd Pandora Linux logo :)
Grazvydas Ignotas [Sun, 20 Jun 2010 19:57:19 +0000 (22:57 +0300)]
add Pandora Linux logo :)

10 years agomusb VBUS workaround
Grazvydas Ignotas [Wed, 6 Oct 2010 22:12:09 +0000 (01:12 +0300)]
musb VBUS workaround

10 years agoaufs: aufs2.1 kbuild patch for linux-2.6.37
aufs project [Sat, 26 Feb 2011 16:26:34 +0000 (18:26 +0200)]
aufs: aufs2.1 kbuild patch for linux-2.6.37

10 years agoaufs: aufs2.1 standalone patch for linux-2.6.37
aufs project [Sat, 26 Feb 2011 16:23:05 +0000 (18:23 +0200)]
aufs: aufs2.1 standalone patch for linux-2.6.37

10 years agoaufs: aufs2.1 base patch for linux-2.6.37
aufs project [Sat, 26 Feb 2011 16:22:09 +0000 (18:22 +0200)]
aufs: aufs2.1 base patch for linux-2.6.37

10 years agoaufs: add files
aufs project [Sat, 26 Feb 2011 16:20:07 +0000 (18:20 +0200)]
aufs: add files

10 years agoAdd power-off support for the TWL4030 companion
Bernhard Wörndl-Aichriedler [Sat, 15 May 2010 14:34:05 +0000 (16:34 +0200)]
Add power-off support for the TWL4030 companion

This patch adds support for the power-off on shutdown feature of the TWL4030

10 years agoomap: mmc: Adjust dto to eliminate timeout errors
Steve Sakoman [Wed, 12 Jan 2011 13:54:55 +0000 (05:54 -0800)]
omap: mmc: Adjust dto to eliminate timeout errors

A number of SD card types were experiencing timeout errors.  This
could also lead to data corruption in some cases.

This fix proposed by Sukumar Ghoral of TI.

10 years agomtd: omap: fix subpage ecc issue with prefetch
Kishore Kadiyala [Wed, 11 May 2011 15:47:27 +0000 (21:17 +0530)]
mtd: omap: fix subpage ecc issue with prefetch

When reading/writing a subpage (When HW ECC is not available/enabled)
for number of bytes not aligned to 4, the mis-aligned bytes are handled
first (by cpu copy method) before enabling the Prefetch engine to/from
'p'(start of buffer 'buf'). Then it reads/writes rest of the bytes with
the help of Prefetch engine, if available, or again using cpu copy method.
Currently, reading/writing of rest of bytes, is not done correctly since
its trying to read/write again to/from begining of buffer 'buf',
overwriting the mis-aligned bytes.

Read & write using prefetch engine got broken in commit '2c01946c'.
We never hit a scenario of not getting 'gpmc_prefetch_enable' call
success. So, problem did not get caught up.

Signed-off-by: Kishore Kadiyala <>
Signed-off-by: Vimal Singh <>
Reported-by: Bryan DE FARIA <>
Cc: [2.6.35+]
Signed-off-by: Artem Bityutskiy <>
Signed-off-by: David Woodhouse <>
10 years agomtd: omap3: nand: report corrected ecc errors
John Ogness [Mon, 28 Feb 2011 12:12:46 +0000 (13:12 +0100)]
mtd: omap3: nand: report corrected ecc errors

The number of corrected ECC errors should be reported since other MTD
systems make use of this information (such as UBI data scrubbing).

Signed-off-by: John Ogness <>
Signed-off-by: Artem Bityutskiy <>
Signed-off-by: David Woodhouse <>
10 years agomtd: return badblockbits back
Artem Bityutskiy [Thu, 28 Apr 2011 17:26:59 +0000 (20:26 +0300)]
mtd: return badblockbits back

In commit c7b28e25cb9beb943aead770ff14551b55fa8c79 the initialization of
the backblockbits was accidentally removed. This patch returns it back,
because otherwise some NAND drivers are broken.

This problem was reported by "Saxena, Parth <>" here:

Reported-by: Saxena, Parth <>
Signed-off-by: Artem Bityutskiy <>
Tested-by: Saxena, Parth <>
Acked-by: Saxena, Parth <>
Acked-by: Brian Norris <>
Cc: [2.6.36+]
Signed-off-by: David Woodhouse <>
10 years agowl1251: Add support for idle mode
Jarkko Nikula [Mon, 4 Apr 2011 08:04:58 +0000 (11:04 +0300)]
wl1251: Add support for idle mode

On Nokia N900 the wl1251 consumes the most power when the interface is up
but not associated to access point (that supports PSM). In terms of battery
current consumption, the consumption is ~180 mA higher when the interface is
up but not associated and only ~5 mA higher when associated compared to
interface down and driver not loaded cases.

This patch adds support for the mac80211 idle notifications. Chip is put into
idle very much the same way when entering into PSM by utilizing the Extreme
Low Power (ELP) mode. I.e. idle is entered by setting necessary conditions
in wl1251_ps_set_mode followed by a call to wl1251_ps_elp_sleep.

It seems it is just enough the authorize ELP mode followed by
CMD_DISCONNECT (thanks to Kalle Valo about the idea to use it).
Without disconnect command the chip remains somewhat active and stays
consuming ~20 mA. Idle mode is left by same way than PSM. The wl1251_join
call is used to revert the CMD_DISCONNECT. Without it association to AP
doesn't work when trying second time.

With this patch the interface up but not associated case the battery current
consumption is less than 1 mA higher compared to interface down case.

Signed-off-by: Jarkko Nikula <>
Acked-by: Kalle Valo <>
Signed-off-by: John W. Linville <>
10 years agowl1251: Prepare for idle mode support
Jarkko Nikula [Mon, 4 Apr 2011 08:04:57 +0000 (11:04 +0300)]
wl1251: Prepare for idle mode support

RFC for WL1251 idle mode support brought a few issues that are worth to
update before adding the idle mode support.

Since the idle mode can reuse the code that is now used in Power Save Mode
(PSM), the flag psm in struct wl1251 is changed to variable station_mode
to be able to distinguish between PSM and idle modes.

As the station mode is different than the power power save mode command
that is sent to chip, the enum wl1251_cmd_ps_mod values are used only when
communicating with the chip and new enum wl1251_station_mode values are used
inside the driver.

Confusing comment about psm and elp relation is removed since the PSM is
actually activated by putting the chip into Entreme Low Power (ELP) mode.

Signed-off-by: Jarkko Nikula <>
Acked-by: Kalle Valo <>
Signed-off-by: John W. Linville <>
10 years agowl1251: fix elp_work race condition
Grazvydas Ignotas [Sun, 6 Mar 2011 17:23:37 +0000 (19:23 +0200)]
wl1251: fix elp_work race condition

While working on PS I've noticed elp_work is kicking rather often, and
sometimes the chip is put to sleep before 5ms delay expires. This
seems to happen because by the time wl1251_ps_elp_wakeup is called
elp_work might still be pending. After wakeup is done, the processing
may take some time, during which 5ms might expire and elp_work might
get scheduled. In this case, ss soon as 1st thread finishes work and
releases the mutex, elp_work will then put the device to sleep without
5ms delay. In addition 1st thread will queue additional elp_work

Fix this by cancelling work in wl1251_ps_elp_wakeup instead.

Signed-off-by: Grazvydas Ignotas <>
Acked-by: Kalle Valo <>
Signed-off-by: John W. Linville <>
10 years agowl1251: remove wl1251_ps_set_elp function
Grazvydas Ignotas [Sun, 6 Mar 2011 17:23:36 +0000 (19:23 +0200)]
wl1251: remove wl1251_ps_set_elp function

wl1251_ps_set_elp() only does acx_sleep_auth call and takes the chip
from/to ELP, however all callers of wl1251_ps_set_mode() have already
taken the chip out of ELP and puts it back to ELP when they finish.
This makes ELP calls (and register writes they result in) superfluous.

So remove wl1251_ps_set_elp function and call acx_sleep_auth directly.

Signed-off-by: Grazvydas Ignotas <>
Acked-by: Kalle Valo <>
Signed-off-by: John W. Linville <>
10 years agowl1251: set rate index and preamble flag on received packets
David Gnedt [Sun, 30 Jan 2011 19:11:10 +0000 (20:11 +0100)]
wl1251: set rate index and preamble flag on received packets

Set the rate index rate_idx and preamble flag RX_FLAG_SHORTPRE on received

Signed-off-by: David Gnedt <>
Acked-by: Kalle Valo <>
Signed-off-by: John W. Linville <>
10 years agowl1251: enable adhoc mode
David Gnedt [Sun, 30 Jan 2011 19:11:04 +0000 (20:11 +0100)]
wl1251: enable adhoc mode

Enable adhoc support in wl1251 driver.

Signed-off-by: David Gnedt <>
Acked-by: Kalle Valo <>
Signed-off-by: John W. Linville <>
10 years agowl1251: implement connection quality monitoring
David Gnedt [Sun, 30 Jan 2011 19:11:00 +0000 (20:11 +0100)]
wl1251: implement connection quality monitoring

Implement connection quality monitoring similar to the wl1271 driver.
It triggers ieee80211_cqm_rssi_notify with the corresponding event when
RSSI drops blow RSSI threshold or rises again above the RSSI threshold.
It should be noted that wl1251 doesn't support RSSI hysteresis, instead it
uses RSSI averageing and delays events until a certain count of frames
proved RSSI change.

Signed-off-by: David Gnedt <>
Acked-by: Kalle Valo <>
Signed-off-by: John W. Linville <>
10 years agowl1251: enable beacon early termination while in power-saving mode
David Gnedt [Sun, 30 Jan 2011 19:10:57 +0000 (20:10 +0100)]
wl1251: enable beacon early termination while in power-saving mode

Port the beacon early termination feature from wl1251 driver version
included in the Maemo Fremantle kernel.
It is enabled when going to power-saving mode and disabled when leaving
power-saving mode.

Signed-off-by: David Gnedt <>
Acked-by: Kalle Valo <>
Signed-off-by: John W. Linville <>
10 years agowl1251: fix 4-byte TX buffer alignment
David Gnedt [Sun, 30 Jan 2011 19:10:48 +0000 (20:10 +0100)]
wl1251: fix 4-byte TX buffer alignment

This implements TX buffer alignment for cloned or too small skb by
copying and replacing the original skb.
Recent changes in wireless-testing seems to make this really necessary.

Signed-off-by: David Gnedt <>
Acked-by: Kalle Valo <>
Signed-off-by: John W. Linville <>
10 years agowl1251: fix queue stopping/waking for TX path
David Gnedt [Sun, 30 Jan 2011 19:10:46 +0000 (20:10 +0100)]
wl1251: fix queue stopping/waking for TX path

The queue stopping/waking functionality was broken in a way that could
cause the TX to stall if the right circumstances are met.

The problem was caused by tx_work, which is scheduled on each TX operation.
If the firmware buffer is full, tx_work does nothing. In combinition with
stopped queues or non-continues transfers, tx_work is never scheduled again.
Moreover the low watermark introduced by
9df86e2e702c6d5547aced7f241addd2d698bb11 never takes effect because of some
old code.

Solve this by scheduling tx_work every time tx_queue is non-empty and
firmware buffer is freed on tx_complete.

This also solves a possible but unlikely case: If less frames than the high
watermark are queued, but more than firmware buffer can hold. This results
in queues staying awake but the only scheduled tx_work doesn't transfer all
frames, so the remaining frames are stuck in the queue until more frames
get queued and tx_work is scheduled again.

Signed-off-by: David Gnedt <>
Acked-by: Kalle Valo <>
Signed-off-by: John W. Linville <>
10 years agowireless, wl1251: Fix potential NULL pointer dereference in wl1251_op_bss_info_changed()
Jesper Juhl [Thu, 3 Feb 2011 20:14:01 +0000 (21:14 +0100)]
wireless, wl1251: Fix potential NULL pointer dereference in wl1251_op_bss_info_changed()

In drivers/net/wireless/wl1251/main.c:wl1251_op_bss_info_changed() we make
a call to ieee80211_beacon_get() which may return NULL, but we do not
check the return value before dereferencing the pointer.

Signed-off-by: Jesper Juhl <>
Signed-off-by: John W. Linville <>
10 years agowl1251: wl12xx_get_platform_data() returns an ERR_PTR
Dan Carpenter [Sat, 18 Dec 2010 08:52:10 +0000 (11:52 +0300)]
wl1251: wl12xx_get_platform_data() returns an ERR_PTR

wl12xx_get_platform_data() returns an ERR_PTR on failure and it never
returns a NULL.

Signed-off-by: Dan Carpenter <>
Acked-by: Kalle Valo <>
Signed-off-by: John W. Linville <>
10 years agoomap: pandora: fix wifi support
Grazvydas Ignotas [Sun, 19 Dec 2010 22:33:36 +0000 (22:33 +0000)]
omap: pandora: fix wifi support

After commit ed919b0 "mmc: sdio: fix runtime PM anomalies by introducing
MMC_CAP_POWER_OFF_CARD" it is required to specify MMC_CAP_POWER_OFF_CARD
to have runtime PM support. As the wl1251 driver expects card to be
powered down when it's not used, wifi will no longer work after interface
is brought down at least once without functioning runtime PM.

Fix this by declaring MMC_CAP_POWER_OFF_CARD for MMC3.

Signed-off-by: Grazvydas Ignotas <>
Signed-off-by: Tony Lindgren <>
10 years agowl1251: use wl12xx_platform_data to pass data
Grazvydas Ignotas [Wed, 3 Nov 2010 22:13:49 +0000 (00:13 +0200)]
wl1251: use wl12xx_platform_data to pass data

Make use the newly added method to pass platform data for wl1251 too.
This allows to eliminate some redundant code.

Cc: Ohad Ben-Cohen <>
Signed-off-by: Grazvydas Ignotas <>
Acked-by: Kalle Valo <>
Acked-by: Luciano Coelho <>
Acked-by: Tony Lindgren <>
Signed-off-by: John W. Linville <>
10 years agowl1251: add runtime PM support for SDIO
Grazvydas Ignotas [Mon, 8 Nov 2010 13:29:36 +0000 (15:29 +0200)]
wl1251: add runtime PM support for SDIO

Add runtime PM support, similar to how it's done for wl1271.
This allows to power down the card when the driver is loaded but
network is not in use.

Cc: Ohad Ben-Cohen <>
Signed-off-by: Grazvydas Ignotas <>
Acked-by: Kalle Valo <>
Signed-off-by: John W. Linville <>
10 years agowl1251: add power callback to wl1251_if_operations
Grazvydas Ignotas [Wed, 3 Nov 2010 22:13:47 +0000 (00:13 +0200)]
wl1251: add power callback to wl1251_if_operations

Call interface specific power callback before calling board specific
one. Also allow that callback to fail. This is how it's done for
wl1271 and will be used for runtime_pm support.

Signed-off-by: Grazvydas Ignotas <>
Acked-by: Kalle Valo <>
Signed-off-by: John W. Linville <>
10 years agoLinux v2.6.37.6
Greg Kroah-Hartman [Sun, 27 Mar 2011 19:01:41 +0000 (12:01 -0700)]

10 years agoext4: skip orphan cleanup if fs has unknown ROCOMPAT features
Amir Goldstein [Mon, 28 Feb 2011 05:53:45 +0000 (00:53 -0500)]
ext4: skip orphan cleanup if fs has unknown ROCOMPAT features

commit d39195c33bb1b5fdcb0f416e8a0b34bfdb07a027 upstream.

Orphan cleanup is currently executed even if the file system has some
number of unknown ROCOMPAT features, which deletes inodes and frees
blocks, which could be very bad for some RO_COMPAT features,
especially the SNAPSHOT feature.

This patch skips the orphan cleanup if it contains readonly compatible
features not known by this ext4 implementation, which would prevent
the fs from being mounted (or remounted) readwrite.

Signed-off-by: Amir Goldstein <>
Signed-off-by: "Theodore Ts'o" <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agodcdbas: force SMI to happen when expected
Stuart Hayes [Wed, 2 Mar 2011 12:42:05 +0000 (13:42 +0100)]
dcdbas: force SMI to happen when expected

commit dd65c736d1b5312c80c88a64bf521db4959eded5 upstream.

The dcdbas driver can do an I/O write to cause a SMI to occur.  The SMI handler
looks at certain registers and memory locations, so the SMI needs to happen
immediately.  On some systems I/O writes are posted, though, causing the SMI to
happen well after the "outb" occurred, which causes random failures.  Following
the "outb" with an "inb" forces the write to go through even if it is posted.

Signed-off-by: Stuart Hayes <>
Acked-by: Doug Warzecha <>
Cc: Chuck Ebbert <>
Signed-off-by: Jiri Kosina <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agofs: call security_d_instantiate in d_obtain_alias V2
Josef Bacik [Fri, 19 Nov 2010 01:52:55 +0000 (20:52 -0500)]
fs: call security_d_instantiate in d_obtain_alias V2

commit 24ff6663ccfdaf088dfa7acae489cb11ed4f43c4 upstream.

While trying to track down some NFS problems with BTRFS, I kept noticing I was
getting -EACCESS for no apparent reason.  Eric Paris and printk() helped me
figure out that it was SELinux that was giving me grief, with the following

type=AVC msg=audit(1290013638.413:95): avc:  denied  { 0x800000 } for  pid=1772
comm="nfsd" name="" dev=sda1 ino=256 scontext=system_u:system_r:kernel_t:s0
tcontext=system_u:object_r:unlabeled_t:s0 tclass=file

Turns out this is because in d_obtain_alias if we can't find an alias we create
one and do all the normal instantiation stuff, but we don't do the

Usually we are protected from getting a hashed dentry that hasn't yet run
security_d_instantiate() by the parent's i_mutex, but obviously this isn't an
option there, so in order to deal with the case that a second thread comes in
and finds our new dentry before we get to run security_d_instantiate(), we go
ahead and call it if we find a dentry already.  Eric assures me that this is ok
as the code checks to see if the dentry has been initialized already so calling
security_d_instantiate() against the same dentry multiple times is ok.  With
this patch I'm no longer getting errant -EACCESS values.

Signed-off-by: Josef Bacik <>
Signed-off-by: Al Viro <>
Cc: Chuck Ebbert <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agoSUNRPC: Never reuse the socket port after an xs_close()
Trond Myklebust [Tue, 22 Mar 2011 22:40:10 +0000 (18:40 -0400)]
SUNRPC: Never reuse the socket port after an xs_close()

commit 246408dcd5dfeef2df437ccb0ef4d6ee87805f58 upstream.

If we call xs_close(), we're in one of two situations:
 - Autoclose, which means we don't expect to resend a request
 - bind+connect failed, which probably means the port is in use

Signed-off-by: Trond Myklebust <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agoNFS: Fix a hang/infinite loop in nfs_wb_page()
Trond Myklebust [Mon, 21 Mar 2011 19:37:01 +0000 (15:37 -0400)]
NFS: Fix a hang/infinite loop in nfs_wb_page()

commit b8413f98f997bb3ed7327e6d7117e7e91ce010c3 upstream.

When one of the two waits in nfs_commit_inode() is interrupted, it
returns a non-negative value, which causes nfs_wb_page() to think
that the operation was successful causing it to busy-loop rather
than exiting.
It also causes nfs_file_fsync() to incorrectly report the file as
being successfully committed to disk.

This patch fixes both problems by ensuring that we return an error
if the attempts to wait fail.

Signed-off-by: Trond Myklebust <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agoperf: Fix tear-down of inherited group events
Peter Zijlstra [Tue, 15 Mar 2011 13:37:10 +0000 (14:37 +0100)]
perf: Fix tear-down of inherited group events

commit 38b435b16c36b0d863efcf3f07b34a6fac9873fd upstream.

When destroying inherited events, we need to destroy groups too,
otherwise the event iteration in perf_event_exit_task_context() will
miss group siblings and we leak events with all the consequences.

Reported-and-tested-by: Vince Weaver <>
Signed-off-by: Peter Zijlstra <>
LKML-Reference: <1300196470.2203.61.camel@twins>
Signed-off-by: Ingo Molnar <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agodrm/radeon/kms: fix hardcoded EDID handling
Alex Deucher [Wed, 23 Mar 2011 08:10:10 +0000 (08:10 +0000)]
drm/radeon/kms: fix hardcoded EDID handling

commit fafcf94e2b5732d1e13b440291c53115d2b172e9 upstream.

On some servers there is a hardcoded EDID provided
in the vbios so that the driver will always see a
display connected even if something like a KVM
prevents traditional means like DDC or load
detection from working properly.  Also most
server boards with DVI are not actually DVI, but
DVO connected to a virtual KVM service processor.
If we fail to detect a monitor via DDC or load
detection and a hardcoded EDID is available, use

Additionally, when using the hardcoded EDID, use
a copy of it rather than the actual one stored
in the driver as the detect() and get_modes()
functions may free it if DDC is successful.

This fixes the virtual KVM on several internal

Signed-off-by: Alex Deucher <>
Signed-off-by: Dave Airlie <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agodrm/radeon/kms: prefer legacy pll algo for tv-out
Alex Deucher [Tue, 22 Mar 2011 05:46:12 +0000 (01:46 -0400)]
drm/radeon/kms: prefer legacy pll algo for tv-out

commit 64146f8b2af1ba77fe3c21d9d6d7213b9bb72b40 upstream.

ntsc seems to work fine with either algo, some
pal TVs seem pickier.


Signed-off-by: Alex Deucher <>
Signed-off-by: Dave Airlie <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agodrm: Fix use-after-free in drm_gem_vm_close()
Chris Wilson [Thu, 17 Mar 2011 22:33:33 +0000 (22:33 +0000)]
drm: Fix use-after-free in drm_gem_vm_close()

commit b74ad5ae14def5e81ad0be3dddb96e485b861b1b upstream.

As we may release the last reference, we need to store the device in a
local variable in order to unlock afterwards.

[   60.140768] BUG: unable to handle kernel paging request at 6b6b6b9f
[   60.140973] IP: [<c1536d11>] __mutex_unlock_slowpath+0x5a/0x111
[   60.141014] *pdpt = 0000000024a54001 *pde = 0000000000000000
[   60.141014] Oops: 0002 [#1] PREEMPT SMP
[   60.141014] last sysfs file: /sys/devices/LNXSYSTM:00/device:00/PNP0A08:00/PNP0C0A:00/power_supply/BAT0/voltage_now
[   60.141014] Modules linked in: uvcvideo ath9k pegasus ath9k_common ath9k_hw hid_egalax ath3k joydev asus_laptop sparse_keymap battery input_polldev
[   60.141014]
[   60.141014] Pid: 771, comm: meego-ux-daemon Not tainted #1 EXOPC EXOPG06411/EXOPG06411
[   60.141014] EIP: 0060:[<c1536d11>] EFLAGS: 00010046 CPU: 0
[   60.141014] EIP is at __mutex_unlock_slowpath+0x5a/0x111
[   60.141014] EAX: 00000100 EBX: 6b6b6b9b ECX: e9b4a1b0 EDX: e4a4e580
[   60.141014] ESI: db162558 EDI: 00000246 EBP: e480be50 ESP: e480be44
[   60.141014]  DS: 007b ES: 007b FS: 00d8 GS: 0000 SS: 0068
[   60.141014] Process meego-ux-daemon (pid: 771, ti=e480a000 task=e9b4a1b0 task.ti=e480a000)
[   60.141014] Stack:
[   60.141014]  e4a4e580 db162558 f5a2f838 e480be58 c1536dd0 e480be68 c125ab1b db162558
[   60.141014]  db1624e0 e480be78 c10ba071 db162558 f760241c e480be94 c10bb0bc 000155fe
[   60.141014]  f760241c f5a2f838 f5a2f8c8 00000000 e480bea4 c1037c24 00000000 f5a2f838
[   60.141014] Call Trace:
[   60.141014]  [<c1536dd0>] ? mutex_unlock+0x8/0xa
[   60.141014]  [<c125ab1b>] ? drm_gem_vm_close+0x39/0x3d
[   60.141014]  [<c10ba071>] ? remove_vma+0x2d/0x58
[   60.141014]  [<c10bb0bc>] ? exit_mmap+0x126/0x13f
[   60.141014]  [<c1037c24>] ? mmput+0x37/0x9a
[   60.141014]  [<c10d450d>] ? exec_mmap+0x178/0x19c
[   60.141014]  [<c1537f85>] ? _raw_spin_unlock+0x1d/0x36
[   60.141014]  [<c10d4eb0>] ? flush_old_exec+0x42/0x75
[   60.141014]  [<c1104442>] ? load_elf_binary+0x32a/0x922
[   60.141014]  [<c10d3f76>] ? search_binary_handler+0x200/0x2ea
[   60.141014]  [<c10d3ecf>] ? search_binary_handler+0x159/0x2ea
[   60.141014]  [<c1104118>] ? load_elf_binary+0x0/0x922
[   60.141014]  [<c10d56b2>] ? do_execve+0x1ff/0x2e6
[   60.141014]  [<c100970e>] ? sys_execve+0x2d/0x55
[   60.141014]  [<c1002a5a>] ? ptregs_execve+0x12/0x18
[   60.141014]  [<c10029dc>] ? sysenter_do_call+0x12/0x3c
[   60.141014]  [<c1530000>] ? init_centaur+0x9c/0x1ba
[   60.141014] Code: c1 00 75 0f ba 38 01 00 00 b8 8c 3a 6c c1 e8 cc 2e b0 ff 9c 58 8d 74 26 00 89 c7 fa 90 8d 74 26 00 e8 d2 b4 b2 ff b8 00 01 00 00 <f0> 66 0f c1 43 04 38 e0 74 07 f3 90 8a 43 04 eb f5 83 3d 64 ef
[   60.141014] EIP: [<c1536d11>] __mutex_unlock_slowpath+0x5a/0x111 SS:ESP 0068:e480be44
[   60.141014] CR2: 000000006b6b6b9f

Reported-by: Rusty Lynch <>
Signed-off-by: Chris Wilson <>
Signed-off-by: Dave Airlie <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agodrm/i915: Prevent racy removal of request from client list
Herton Ronaldo Krzesinski [Thu, 17 Mar 2011 13:45:12 +0000 (13:45 +0000)]
drm/i915: Prevent racy removal of request from client list

commit 09bfa51773c1e90f13000dc2fc0c4b84047009bc upstream.

When i915_gem_retire_requests_ring calls i915_gem_request_remove_from_client,
the client_list for that request may already be removed in i915_gem_release.
So we may call twice list_del(&request->client_list), resulting in an
oops like this report:

[126167.230394] BUG: unable to handle kernel paging request at 00100104
[126167.230699] IP: [<f8c2ce44>] i915_gem_retire_requests_ring+0xd4/0x240 [i915]
[126167.231042] *pdpt = 00000000314c1001 *pde = 0000000000000000
[126167.231314] Oops: 0002 [#1] SMP
[126167.231471] last sysfs file: /sys/devices/LNXSYSTM:00/device:00/PNP0C0A:00/power_supply/BAT1/current_now
[126167.231901] Modules linked in: snd_seq_dummy nls_utf8 isofs btrfs zlib_deflate libcrc32c ufs qnx4 hfsplus hfs minix ntfs vfat msdos fat jfs xfs exportfs reiserfs cryptd aes_i586 aes_generic binfmt_misc vboxnetadp vboxnetflt vboxdrv parport_pc ppdev snd_hda_codec_hdmi snd_hda_codec_conexant snd_hda_intel snd_hda_codec snd_hwdep arc4 snd_pcm snd_seq_midi snd_rawmidi snd_seq_midi_event snd_seq uvcvideo videodev snd_timer snd_seq_device joydev iwlagn iwlcore mac80211 snd cfg80211 soundcore i915 drm_kms_helper snd_page_alloc psmouse drm serio_raw i2c_algo_bit video lp parport usbhid hid sky2 sdhci_pci ahci sdhci libahci
[126167.232018] Pid: 1101, comm: Xorg Not tainted 2.6.38-6-generic-pae #34-Ubuntu Gateway                          MC7833U /
[126167.232018] EIP: 0060:[<f8c2ce44>] EFLAGS: 00213246 CPU: 0
[126167.232018] EIP is at i915_gem_retire_requests_ring+0xd4/0x240 [i915]
[126167.232018] EAX: 00200200 EBX: f1ac25b0 ECX: 00000040 EDX: 00100100
[126167.232018] ESI: f1a2801c EDI: e87fc060 EBP: ef4d7dd8 ESP: ef4d7db0
[126167.232018]  DS: 007b ES: 007b FS: 00d8 GS: 00e0 SS: 0068
[126167.232018] Process Xorg (pid: 1101, ti=ef4d6000 task=f1ba6500 task.ti=ef4d6000)
[126167.232018] Stack:
[126167.232018]  f1a28000 f1a2809c f1a28094 0058bd97 f1aa2400 f1a2801c 0058bd7b 0058bd85
[126167.232018]  f1a2801c f1a28000 ef4d7e38 f8c2e995 ef4d7e30 ef4d7e60 c14d1ebc f6b3a040
[126167.232018]  f1522cc0 000000db 00000000 f1ba6500 ffffffa1 00000000 00000001 f1a29214
[126167.232018] Call Trace:

Unfortunately the call trace reported was cut, but looking at debug
symbols the crash is at __list_del, when probably list_del is called
twice on the same request->client_list, as the dereferenced value is
LIST_POISON1 + 4, and by looking more at the debug symbols before
list_del call it should have being called by

And as I can see in the code, it seems we indeed have the possibility
to remove a request->client_list twice, which would cause the above,
because we do list_del(&request->client_list) on both
i915_gem_request_remove_from_client and i915_gem_release

As Chris Wilson pointed out, it's indeed the case:
"(...) I had thought that the actual insertion/deletion was serialised
under the struct mutex and the intention of the spinlock was to protect
the unlocked list traversal during throttling. However, I missed that
i915_gem_release() is also called without struct mutex and so we do need
the double check for i915_gem_request_remove_from_client()."

This change does the required check to avoid the duplicate remove of

Signed-off-by: Herton Ronaldo Krzesinski <>
Signed-off-by: Chris Wilson <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agodrm: check for modesetting on modeset ioctls
Dave Airlie [Tue, 8 Feb 2011 03:55:21 +0000 (13:55 +1000)]
drm: check for modesetting on modeset ioctls

commit fb3b06c8a1fd1a80298f13b738ab38ef8c73baff upstream.

Noticed this while working on some other things, helps if we check for modeset
enabled on modesetting ioctls.

Signed-off-by: Dave Airlie <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agox86: Cleanup highmap after brk is concluded
Yinghai Lu [Fri, 18 Feb 2011 11:30:30 +0000 (11:30 +0000)]
x86: Cleanup highmap after brk is concluded

commit e5f15b45ddf3afa2bbbb10c7ea34fb32b6de0a0e upstream.

Now cleanup_highmap actually is in two steps: one is early in head64.c
and only clears above _end; a second one is in init_memory_mapping() and
tries to clean from _brk_end to _end.
It should check if those boundaries are PMD_SIZE aligned but currently
does not.
Also init_memory_mapping() is called several times for numa or memory
hotplug, so we really should not handle initial kernel mappings there.

This patch moves cleanup_highmap() down after _brk_end is settled so
we can do everything in one step.
Also we honor max_pfn_mapped in the implementation of cleanup_highmap.

Signed-off-by: Yinghai Lu <>
Signed-off-by: Stefano Stabellini <>
LKML-Reference: <alpine.DEB.2.00.1103171739050.3382@kaball-desktop>
Signed-off-by: H. Peter Anvin <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agoInput: xen-kbdfront - advertise either absolute or relative coordinates
Olaf Hering [Thu, 17 Mar 2011 05:11:46 +0000 (22:11 -0700)]
Input: xen-kbdfront - advertise either absolute or relative coordinates

commit 8c3c283e6bf463ab498d6e7823aff6c4762314b6 upstream.

A virtualized display device is usually viewed with the vncviewer
application, either by 'xm vnc domU' or with vncviewer localhost:port.
vncviewer and the RFB protocol provides absolute coordinates to the
virtual display. These coordinates are either passed through to a PV
guest or converted to relative coordinates for a HVM guest.

A PV guest receives these coordinates and passes them to the kernels
evdev driver. There it can be picked up by applications such as the
xorg-input drivers. Using absolute coordinates avoids issues such as
guest mouse pointer not tracking host mouse pointer due to wrong mouse
acceleration settings in the guests X display.

Advertise either absolute or relative coordinates to the input system
and the evdev driver, depending on what dom0 provides. The xorg-input
driver prefers relative coordinates even if a devices provides both.

Signed-off-by: Olaf Hering <>
Signed-off-by: Stefano Stabellini <>
Signed-off-by: Dmitry Torokhov <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agofs: assign sb->s_bdi to default_backing_dev_info if the bdi is going away
Jens Axboe [Thu, 17 Mar 2011 10:13:12 +0000 (11:13 +0100)]
fs: assign sb->s_bdi to default_backing_dev_info if the bdi is going away

commit 95f28604a65b1c40b6c6cd95e58439cd7ded3add upstream.

We don't have proper reference counting for this yet, so we run into
cases where the device is pulled and we OOPS on flushing the fs data.
This happens even though the dirty inodes have already been
migrated to the default_backing_dev_info.

Reported-by: Torsten Hilbrich <>
Tested-by: Torsten Hilbrich <>
Signed-off-by: Jens Axboe <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agoUSB: cdc-acm: fix potential null-pointer dereference on disconnect
Johan Hovold [Tue, 22 Mar 2011 10:12:11 +0000 (11:12 +0100)]
USB: cdc-acm: fix potential null-pointer dereference on disconnect

commit 7e7797e7f6f7bfab73fca02c65e40eaa5bb9000c upstream.

Fix potential null-pointer exception on disconnect introduced by commit
11ea859d64b69a747d6b060b9ed1520eab1161fe (USB: additional power savings
for cdc-acm devices that support remote wakeup).

Only access acm->dev after making sure it is non-null in control urb
completion handler.

Signed-off-by: Johan Hovold <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agoUSB: cdc-acm: fix potential null-pointer dereference
Johan Hovold [Tue, 22 Mar 2011 10:12:10 +0000 (11:12 +0100)]
USB: cdc-acm: fix potential null-pointer dereference

commit 15e5bee33ffc11d0e5c6f819a65e7881c5c407be upstream.

Must check return value of tty_port_tty_get.

Signed-off-by: Johan Hovold <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agoUSB: cdc-acm: fix memory corruption / panic
Johan Hovold [Tue, 22 Mar 2011 10:12:09 +0000 (11:12 +0100)]
USB: cdc-acm: fix memory corruption / panic

commit 23b80550e2aa61d0ba3af98b831b9195be0db9ee upstream.

Prevent read urbs from being resubmitted from tasklet after port close.

The receive tasklet was not disabled on port close, which could lead to
corruption of receive lists on consecutive port open. In particular,
read urbs could be re-submitted before port open, added to free list in
open, and then added a second time to the free list in the completion

cdc-acm.c: Entering acm_tty_open.
cdc-acm.c: acm_control_msg: rq: 0x22 val: 0x3 len: 0x0 result: 0
cdc-acm.c: Entering acm_rx_tasklet
cdc-acm.c: acm_rx_tasklet: sending urb 0xf50da280, rcv 0xf57fbc24, buf 0xf57fbd64
cdc-acm.c: set line: 115200 0 0 8
cdc-acm.c: acm_control_msg: rq: 0x20 val: 0x0 len: 0x7 result: 7
cdc-acm.c: acm_tty_close
cdc-acm.c: acm_port_down
cdc-acm.c: acm_control_msg: rq: 0x22 val: 0x0 len: 0x0 result: 0
cdc-acm.c: acm_ctrl_irq - urb shutting down with status: -2
cdc-acm.c: acm_rx_tasklet: sending urb 0xf50da300, rcv 0xf57fbc10, buf 0xf57fbd50
cdc-acm.c: Entering acm_read_bulk with status -2
cdc_acm 4-1:1.1: Aborting, acm not ready
cdc-acm.c: Entering acm_read_bulk with status -2
cdc_acm 4-1:1.1: Aborting, acm not ready
cdc-acm.c: acm_rx_tasklet: sending urb 0xf50da380, rcv 0xf57fbbfc, buf 0xf57fbd3c
cdc-acm.c: acm_rx_tasklet: sending urb 0xf50da400, rcv 0xf57fbbe8, buf 0xf57fbd28
cdc-acm.c: acm_rx_tasklet: sending urb 0xf50da480, rcv 0xf57fbbd4, buf 0xf57fbd14
cdc-acm.c: acm_rx_tasklet: sending urb 0xf50da900, rcv 0xf57fbbc0, buf 0xf57fbd00
cdc-acm.c: acm_rx_tasklet: sending urb 0xf50da980, rcv 0xf57fbbac, buf 0xf57fbcec
cdc-acm.c: acm_rx_tasklet: sending urb 0xf50daa00, rcv 0xf57fbb98, buf 0xf57fbcd8
cdc-acm.c: acm_rx_tasklet: sending urb 0xf50daa80, rcv 0xf57fbb84, buf 0xf57fbcc4
cdc-acm.c: acm_rx_tasklet: sending urb 0xf50dab00, rcv 0xf57fbb70, buf 0xf57fbcb0
cdc-acm.c: acm_rx_tasklet: sending urb 0xf50dab80, rcv 0xf57fbb5c, buf 0xf57fbc9c
cdc-acm.c: acm_rx_tasklet: sending urb 0xf50dac00, rcv 0xf57fbb48, buf 0xf57fbc88
cdc-acm.c: acm_rx_tasklet: sending urb 0xf50dac80, rcv 0xf57fbb34, buf 0xf57fbc74
cdc-acm.c: acm_rx_tasklet: sending urb 0xf50dad00, rcv 0xf57fbb20, buf 0xf57fbc60
cdc-acm.c: acm_rx_tasklet: sending urb 0xf50dad80, rcv 0xf57fbb0c, buf 0xf57fbc4c
cdc-acm.c: acm_rx_tasklet: sending urb 0xf50da880, rcv 0xf57fbaf8, buf 0xf57fbc38
cdc-acm.c: Entering acm_tty_open.
cdc-acm.c: acm_control_msg: rq: 0x22 val: 0x3 len: 0x0 result: 0
cdc-acm.c: Entering acm_rx_tasklet
cdc-acm.c: acm_rx_tasklet: sending urb 0xf50da280, rcv 0xf57fbc24, buf 0xf57fbd64
cdc-acm.c: Entering acm_tty_write to write 3 bytes,
cdc-acm.c: Get 3 bytes...
cdc-acm.c: acm_write_start susp_count: 0
cdc-acm.c: Entering acm_read_bulk with status 0
------------[ cut here ]------------
WARNING: at /home/johan/src/linux/linux-2.6/lib/list_debug.c:57 list_del+0x10c/0x120()
Hardware name: Vostro 1520
list_del corruption. next->prev should be f57fbc10, but was f57fbaf8
Modules linked in: cdc_acm
Pid: 3, comm: ksoftirqd/0 Not tainted 2.6.37+ #39
Call Trace:
 [<c103c7e2>] warn_slowpath_common+0x72/0xa0
 [<c11dd8ac>] ? list_del+0x10c/0x120
 [<c11dd8ac>] ? list_del+0x10c/0x120
 [<c103c8b3>] warn_slowpath_fmt+0x33/0x40
 [<c11dd8ac>] list_del+0x10c/0x120
 [<f8051dbf>] acm_rx_tasklet+0xef/0x3e0 [cdc_acm]
 [<c135465d>] ? net_rps_action_and_irq_enable+0x6d/0x80
 [<c1042bb6>] tasklet_action+0xe6/0x140
 [<c104342f>] __do_softirq+0xaf/0x210
 [<c1043380>] ? __do_softirq+0x0/0x210
 <IRQ>  [<c1042c9a>] ? run_ksoftirqd+0x8a/0x1c0
 [<c1042c10>] ? run_ksoftirqd+0x0/0x1c0
 [<c105ac24>] ? kthread+0x74/0x80
 [<c105abb0>] ? kthread+0x0/0x80
 [<c100337a>] ? kernel_thread_helper+0x6/0x10
---[ end trace efd9a11434f0082e ]---
------------[ cut here ]------------
WARNING: at /home/johan/src/linux/linux-2.6/lib/list_debug.c:57 list_del+0x10c/0x120()
Hardware name: Vostro 1520
list_del corruption. next->prev should be f57fbd50, but was f57fbdb0
Modules linked in: cdc_acm
Pid: 3, comm: ksoftirqd/0 Tainted: G        W   2.6.37+ #39
Call Trace:
 [<c103c7e2>] warn_slowpath_common+0x72/0xa0
 [<c11dd8ac>] ? list_del+0x10c/0x120
 [<c11dd8ac>] ? list_del+0x10c/0x120
 [<c103c8b3>] warn_slowpath_fmt+0x33/0x40
 [<c11dd8ac>] list_del+0x10c/0x120
 [<f8051dd6>] acm_rx_tasklet+0x106/0x3e0 [cdc_acm]
 [<c135465d>] ? net_rps_action_and_irq_enable+0x6d/0x80
 [<c1042bb6>] tasklet_action+0xe6/0x140
 [<c104342f>] __do_softirq+0xaf/0x210
 [<c1043380>] ? __do_softirq+0x0/0x210
 <IRQ>  [<c1042c9a>] ? run_ksoftirqd+0x8a/0x1c0
 [<c1042c10>] ? run_ksoftirqd+0x0/0x1c0
 [<c105ac24>] ? kthread+0x74/0x80
 [<c105abb0>] ? kthread+0x0/0x80
 [<c100337a>] ? kernel_thread_helper+0x6/0x10
---[ end trace efd9a11434f0082f ]---
cdc-acm.c: acm_rx_tasklet: sending urb 0xf50da300, rcv 0xf57fbc10, buf 0xf57fbd50
cdc-acm.c: disconnected from network
cdc-acm.c: acm_rx_tasklet: sending urb 0xf50da380, rcv 0xf57fbbfc, buf 0xf57fbd3c
cdc-acm.c: Entering acm_rx_tasklet
------------[ cut here ]------------
WARNING: at /home/johan/src/linux/linux-2.6/lib/list_debug.c:48 list_del+0xd5/0x120()
Hardware name: Vostro 1520
list_del corruption, next is LIST_POISON1 (00100100)
Modules linked in: cdc_acm
Pid: 3, comm: ksoftirqd/0 Tainted: G        W   2.6.37+ #39
Call Trace:
 [<c103c7e2>] warn_slowpath_common+0x72/0xa0
 [<c11dd875>] ? list_del+0xd5/0x120
 [<c11dd875>] ? list_del+0xd5/0x120
 [<c103c8b3>] warn_slowpath_fmt+0x33/0x40
 [<c11dd875>] list_del+0xd5/0x120
 [<f8051fac>] acm_rx_tasklet+0x2dc/0x3e0 [cdc_acm]
 [<c106dbab>] ? trace_hardirqs_on+0xb/0x10
 [<c1042b30>] ? tasklet_action+0x60/0x140
 [<c1042bb6>] tasklet_action+0xe6/0x140
 [<c104342f>] __do_softirq+0xaf/0x210
 [<c1043380>] ? __do_softirq+0x0/0x210
 <IRQ>  [<c1042c9a>] ? run_ksoftirqd+0x8a/0x1c0
 [<c1042c10>] ? run_ksoftirqd+0x0/0x1c0
 [<c105ac24>] ? kthread+0x74/0x80
 [<c105abb0>] ? kthread+0x0/0x80
 [<c100337a>] ? kernel_thread_helper+0x6/0x10
---[ end trace efd9a11434f00830 ]---
BUG: unable to handle kernel paging request at 00200200
IP: [<c11dd7bd>] list_del+0x1d/0x120
*pde = 00000000
Oops: 0000 [#1] PREEMPT SMP
last sysfs file: /sys/devices/pci0000:00/0000:00:1a.1/usb4/4-1/4-1:1.0/tty/ttyACM0/uevent
Modules linked in: cdc_acm
Pid: 3, comm: ksoftirqd/0 Tainted: G        W   2.6.37+ #39 0T816J/Vostro 1520
EIP: 0060:[<c11dd7bd>] EFLAGS: 00010046 CPU: 0
EIP is at list_del+0x1d/0x120
EAX: f57fbd3c EBX: f57fb800 ECX: ffff8000 EDX: 00200200
ESI: f57fbe90 EDI: f57fbd3c EBP: f600bf54 ESP: f600bf3c
 DS: 007b ES: 007b FS: 00d8 GS: 0000 SS: 0068
Process ksoftirqd/0 (pid: 3, ti=f600a000 task=f60791c0 task.ti=f6082000)
 c1527e84 00000030 c1527e54 00100100 f57fb800 f57fbd3c f600bf98 f8051fac
 f8053104 f8052b94 f600bf6c c106dbab f600bf80 00000286 f60791c0 c1042b30
 f57fbda8 f57f5800 f57fbdb0 f57fbd80 f57fbe7c c1656b04 00000000 f600bfb0
Call Trace:
 [<f8051fac>] ? acm_rx_tasklet+0x2dc/0x3e0 [cdc_acm]
 [<c106dbab>] ? trace_hardirqs_on+0xb/0x10
 [<c1042b30>] ? tasklet_action+0x60/0x140
 [<c1042bb6>] ? tasklet_action+0xe6/0x140
 [<c104342f>] ? __do_softirq+0xaf/0x210
 [<c1043380>] ? __do_softirq+0x0/0x210
 [<c1042c9a>] ? run_ksoftirqd+0x8a/0x1c0
 [<c1042c10>] ? run_ksoftirqd+0x0/0x1c0
 [<c105ac24>] ? kthread+0x74/0x80
 [<c105abb0>] ? kthread+0x0/0x80
 [<c100337a>] ? kernel_thread_helper+0x6/0x10
Code: ff 48 14 e9 57 ff ff ff 90 90 90 90 90 90 55 89 e5 83 ec 18 81 38 00 01 10 00 0f 84 9c 00 00 00 8b 50 04 81 fa 00 02 20 00 74 33 <8b> 12 39 d0 75 5c 8b 10 8b 4a 04 39 c8 0f 85 b5 00 00 00 8b 48
EIP: [<c11dd7bd>] list_del+0x1d/0x120 SS:ESP 0068:f600bf3c
CR2: 0000000000200200
---[ end trace efd9a11434f00831 ]---
Kernel panic - not syncing: Fatal exception in interrupt
Pid: 3, comm: ksoftirqd/0 Tainted: G      D W   2.6.37+ #39
Call Trace:
 [<c13fede1>] ? printk+0x1d/0x24
 [<c13fecce>] panic+0x66/0x15c
 [<c10067df>] oops_end+0x8f/0x90
 [<c1025476>] no_context+0xc6/0x160
 [<c10255a8>] __bad_area_nosemaphore+0x98/0x140
 [<c103cf68>] ? release_console_sem+0x1d8/0x210
 [<c1025667>] bad_area_nosemaphore+0x17/0x20
 [<c1025a49>] do_page_fault+0x279/0x420
 [<c1006a8f>] ? show_trace+0x1f/0x30
 [<c13fede1>] ? printk+0x1d/0x24
 [<c10257d0>] ? do_page_fault+0x0/0x420
 [<c140333b>] error_code+0x5f/0x64
 [<c103007b>] ? select_task_rq_fair+0x37b/0x6a0
 [<c10257d0>] ? do_page_fault+0x0/0x420
 [<c11dd7bd>] ? list_del+0x1d/0x120
 [<f8051fac>] acm_rx_tasklet+0x2dc/0x3e0 [cdc_acm]
 [<c106dbab>] ? trace_hardirqs_on+0xb/0x10
 [<c1042b30>] ? tasklet_action+0x60/0x140
 [<c1042bb6>] tasklet_action+0xe6/0x140
 [<c104342f>] __do_softirq+0xaf/0x210
 [<c1043380>] ? __do_softirq+0x0/0x210
 <IRQ>  [<c1042c9a>] ? run_ksoftirqd+0x8a/0x1c0
 [<c1042c10>] ? run_ksoftirqd+0x0/0x1c0
 [<c105ac24>] ? kthread+0x74/0x80
 [<c105abb0>] ? kthread+0x0/0x80
 [<c100337a>] ? kernel_thread_helper+0x6/0x10
panic occurred, switching back to text console
------------[ cut here ]------------

Signed-off-by: Johan Hovold <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agoUSB: Fix 'bad dma' problem on WDM device disconnect
Robert Lukassen [Wed, 16 Mar 2011 11:13:34 +0000 (12:13 +0100)]
USB: Fix 'bad dma' problem on WDM device disconnect

commit 878b753e32ca765cd346a5d3038d630178ec78ff upstream.

In the WDM class driver a disconnect event leads to calls to
usb_free_coherent to put back two USB DMA buffers allocated earlier.
The call to usb_free_coherent uses a different size parameter
(desc->wMaxCommand) than the corresponding call to usb_alloc_coherent

When a disconnect event occurs, this leads to 'bad dma' complaints
from usb core because the USB DMA buffer is being pushed back to the
'buffer-2048' pool from which it has not been allocated.

This patch against the most recent linux-2.6 kernel ensures that the
parameters used by usb_alloc_coherent & usb_free_coherent calls in
cdc-wdm.c match.

Signed-off-by: Robert Lukassen <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agoUSB: uss720 fixup refcount position
Peter Holik [Fri, 18 Mar 2011 17:47:44 +0000 (18:47 +0100)]
USB: uss720 fixup refcount position

commit adaa3c6342b249548ea830fe8e02aa5b45be8688 upstream.

My testprog do a lot of bitbang - after hours i got following warning and my machine lockups:
WARNING: at /build/buildd/linux-2.6.38/lib/kref.c:34
After debugging uss720 driver i discovered that the completion callback was called before
usb_submit_urb returns. The callback frees the request structure that is krefed on return by

Signed-off-by: Peter Holik <>
Acked-by: Thomas Sailer <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agoehci-hcd: Bug fix: don't set a QH's Halt bit
Alan Stern [Wed, 16 Mar 2011 14:57:15 +0000 (10:57 -0400)]
ehci-hcd: Bug fix: don't set a QH's Halt bit

commit b5a3b3d985493c173925907adfebf3edab236fe7 upstream.

This patch (as1453) fixes a long-standing bug in the ehci-hcd driver.

There is no need to set the Halt bit in the overlay region for an
unlinked or blocked QH.  Contrary to what the comment says, setting
the Halt bit does not cause the QH to be patched later; that decision
(made in qh_refresh()) depends only on whether the QH is currently
pointing to a valid qTD.  Likewise, setting the Halt bit does not
prevent completions from activating the QH while it is "stopped"; they
are prevented by the fact that qh_completions() temporarily changes
qh->qh_state to QH_STATE_COMPLETING.

On the other hand, there are circumstances in which the QH will be
reactivated _without_ being patched; this happens after an URB beyond
the head of the queue is unlinked.  Setting the Halt bit will then
cause the hardware to see the QH with both the Active and Halt bits
set, an invalid combination that will prevent the queue from
advancing and may even crash some controllers.

Apparently the only reason this hasn't been reported before is that
unlinking URBs from the middle of a running queue is quite uncommon.
However Test 17, recently added to the usbtest driver, does exactly
this, and it confirms the presence of the bug.

In short, there is no reason to set the Halt bit for an unlinked or
blocked QH, and there is a very good reason not to set it.  Therefore
the code that sets it is removed.

Signed-off-by: Alan Stern <>
Tested-by: Andiry Xu <>
CC: David Brownell <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agoUSB: Do not pass negative length to snoop_urb()
Michal Sojka [Tue, 15 Mar 2011 15:41:47 +0000 (16:41 +0100)]
USB: Do not pass negative length to snoop_urb()

commit 9d02b42614149ebccf12c9c580601ed01bd83070 upstream.

When `echo Y > /sys/module/usbcore/parameters/usbfs_snoop` and
usb_control_msg() returns error, a lot of kernel memory is dumped to dmesg
until unhandled kernel paging request occurs.

Signed-off-by: Michal Sojka <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agosh: Fix ptrace hw_breakpoint handling
David Engraf [Wed, 23 Mar 2011 11:35:42 +0000 (11:35 +0000)]
sh: Fix ptrace hw_breakpoint handling

commit fb7f045ace0624f1e59a7db8497e460bd54b1cbc upstream.

Since commit 34d0b5af50a063cded842716633501b38ff815fb it is no longer
possible to debug an application using singlestep. The old commit
converted singlestep handling via ptrace to hw_breakpoints. The
hw_breakpoint is disabled when an event is triggered and not re-enabled
again. This patch re-enables the existing hw_breakpoint before the
existing breakpoint is reused.

Signed-off-by: David Engraf <>
Signed-off-by: Paul Mundt <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agosh: Fix ptrace fpu state initialisation
Phil Edworthy [Fri, 18 Mar 2011 14:16:31 +0000 (14:16 +0000)]
sh: Fix ptrace fpu state initialisation

commit c49b6ecf0870e78fa40497cd8b142915c1d5c7c9 upstream.

Commit 0ea820cf introduced the PTRACE_GETFPREGS/SETFPREGS cmds,
but gdb-server still accesses the FPU state using the
PTRACE_PEEKUSR/POKEUSR commands. In this case, xstate was not

Signed-off-by: Phil Edworthy <>
Signed-off-by: Paul Mundt <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agouvcvideo: Fix descriptor parsing for video output devices
Laurent Pinchart [Wed, 23 Feb 2011 14:19:17 +0000 (11:19 -0300)]
uvcvideo: Fix descriptor parsing for video output devices

commit 4093a5c4a3f59cba1a085bbf87b6ffdddc5a443d upstream.

Commit 4057ac6ca9a77c4275b34b5925ab5c99557913b1

    V4L/DVB (13505): uvcvideo: Refactor chain scan

broke output terminals parsing. Fix it.

Signed-off-by: Laurent Pinchart <>
Signed-off-by: Mauro Carvalho Chehab <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agouvcvideo: Fix uvc_fixup_video_ctrl() format search
Stephan Lachowsky [Fri, 28 Jan 2011 02:04:33 +0000 (23:04 -0300)]
uvcvideo: Fix uvc_fixup_video_ctrl() format search

commit 38a66824d96de8aeeb915e6f46f0d3fe55828eb1 upstream.

The scheme used to index format in uvc_fixup_video_ctrl() is not robust:
format index is based on descriptor ordering, which does not necessarily
match bFormatIndex ordering.  Searching for first matching format will
prevent uvc_fixup_video_ctrl() from using the wrong format/frame to make

Signed-off-by: Stephan Lachowsky <>
Signed-off-by: Mauro Carvalho Chehab <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agonfsd: wrong index used in inner loop
Mi Jinlong [Fri, 11 Mar 2011 04:13:55 +0000 (12:13 +0800)]
nfsd: wrong index used in inner loop

commit 5a02ab7c3c4580f94d13c683721039855b67cda6 upstream.

We must not use dummy for index.
After the first index, READ32(dummy) will change dummy!!!!

Signed-off-by: Mi Jinlong <>
[ Trond points out READ_BUF alone is sufficient.]
Signed-off-by: J. Bruce Fields <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agonfsd4: fix struct file leak
J. Bruce Fields [Wed, 2 Mar 2011 23:01:35 +0000 (18:01 -0500)]
nfsd4: fix struct file leak

commit 0997b173609b9229ece28941c118a2a9b278796e upstream.

Make sure we properly reference count the struct files that a lock
depends on, and release them when the lock stateid is released.

This fixes a major leak of struct files when using locking over nfsv4.

Reported-by: Rick Koshi <>
Tested-by: Ivo Přikryl <>
Signed-off-by: J. Bruce Fields <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agonfsd4: minor nfs4state.c reshuffling
J. Bruce Fields [Thu, 3 Mar 2011 04:48:33 +0000 (23:48 -0500)]
nfsd4: minor nfs4state.c reshuffling

commit 529d7b2a7fa31e9f7d08bc790d232c3cbe64fa24 upstream.

Minor cleanup in preparation for a bugfix--moving some code to avoid
forward references, etc.  No change in functionality.

Signed-off-by: J. Bruce Fields <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agonfsd41: modify the members value of nfsd4_op_flags
Mi Jinlong [Fri, 18 Feb 2011 01:08:31 +0000 (09:08 +0800)]
nfsd41: modify the members value of nfsd4_op_flags

commit 5ece3cafbd88d4da5c734e1810c4a2e6474b57b2 upstream.

The members of nfsd4_op_flags, (ALLOWED_WITHOUT_FH | ALLOWED_ON_ABSENT_FS)
equals to  ALLOWED_AS_FIRST_OP, maybe that's not what we want.

can't appears as the first operation with out SEQUENCE ops.

This patch modify the wrong value of ALLOWED_WITHOUT_FH etc which
was introduced by f9bb94c4.

Reviewed-by: Benny Halevy <>
Signed-off-by: Mi Jinlong <>
Signed-off-by: J. Bruce Fields <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agofbcon: Bugfix soft cursor detection in Tile Blitting
Henry Nestler [Sun, 20 Feb 2011 20:50:56 +0000 (20:50 +0000)]
fbcon: Bugfix soft cursor detection in Tile Blitting

commit d6244bc0ed0c52a795e6f4dcab3886daf3e74fac upstream.

Use mask 0x10 for "soft cursor" detection on in function tile_cursor.
(Tile Blitting Operation in framebuffer console).

The old mask 0x01 for vc_cursor_type detects CUR_NONE, CUR_LOWER_THIRD
and every second mode value as "software cursor". This hides the cursor
for these modes (cursor.mode = 0). But, only CUR_NONE or "software cursor"
should hide the cursor.
See also 0x10 in functions add_softcursor, bit_cursor and cw_cursor.

Signed-off-by: Henry Nestler <>
Signed-off-by: Paul Mundt <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agoproc: protect mm start_code/end_code in /proc/pid/stat
Kees Cook [Wed, 23 Mar 2011 23:42:53 +0000 (16:42 -0700)]
proc: protect mm start_code/end_code in /proc/pid/stat

commit 5883f57ca0008ffc93e09cbb9847a1928e50c6f3 upstream.

While mm->start_stack was protected from cross-uid viewing (commit
f83ce3e6b02d5 ("proc: avoid information leaks to non-privileged
processes")), the start_code and end_code values were not.  This would
allow the text location of a PIE binary to leak, defeating ASLR.

Note that the value "1" is used instead of "0" for a protected value since
"ps", "killall", and likely other readers of /proc/pid/stat, take
start_code of "0" to mean a kernel thread and will misbehave.  Thanks to
Brad Spengler for pointing this out.

Addresses CVE-2011-0726

Signed-off-by: Kees Cook <>
Cc: Alexey Dobriyan <>
Cc: David Howells <>
Cc: Eugene Teo <>
Cc: Martin Schwidefsky <>
Cc: Brad Spengler <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agoprocfs: fix /proc/<pid>/maps heap check
Aaro Koskinen [Wed, 23 Mar 2011 23:42:50 +0000 (16:42 -0700)]
procfs: fix /proc/<pid>/maps heap check

commit 0db0c01b53a1a421513f91573241aabafb87802a upstream.

The current code fails to print the "[heap]" marking if the heap is split
into multiple mappings.

Fix the check so that the marking is displayed in all possible cases:
1. vma matches exactly the heap
2. the heap vma is merged e.g. with bss
3. the heap vma is splitted e.g. due to locked pages

Test cases. In all cases, the process should have mapping(s) with
[heap] marking:

(1) vma matches exactly the heap

#include <stdio.h>
#include <unistd.h>
#include <sys/types.h>

int main (void)
if (sbrk(4096) != (void *)-1) {
printf("check /proc/%d/maps\n", (int)getpid());
while (1)
return 0;

# ./test1
check /proc/553/maps
[1] + Stopped                    ./test1
# cat /proc/553/maps | head -4
00008000-00009000 r-xp 00000000 01:00 3113640    /test1
00010000-00011000 rw-p 00000000 01:00 3113640    /test1
00011000-00012000 rw-p 00000000 00:00 0          [heap]
4006f000-40070000 rw-p 00000000 00:00 0

(2) the heap vma is merged

#include <stdio.h>
#include <unistd.h>
#include <sys/types.h>

char foo[4096] = "foo";
char bar[4096];

int main (void)
if (sbrk(4096) != (void *)-1) {
printf("check /proc/%d/maps\n", (int)getpid());
while (1)
return 0;

# ./test2
check /proc/556/maps
[2] + Stopped                    ./test2
# cat /proc/556/maps | head -4
00008000-00009000 r-xp 00000000 01:00 3116312    /test2
00010000-00012000 rw-p 00000000 01:00 3116312    /test2
00012000-00014000 rw-p 00000000 00:00 0          [heap]
4004a000-4004b000 rw-p 00000000 00:00 0

(3) the heap vma is splitted (this fails without the patch)

#include <stdio.h>
#include <unistd.h>
#include <sys/mman.h>
#include <sys/types.h>

int main (void)
if ((sbrk(4096) != (void *)-1) && !mlockall(MCL_FUTURE) &&
    (sbrk(4096) != (void *)-1)) {
printf("check /proc/%d/maps\n", (int)getpid());
while (1)
return 0;

# ./test3
check /proc/559/maps
[1] + Stopped                    ./test3
# cat /proc/559/maps|head -4
00008000-00009000 r-xp 00000000 01:00 3119108    /test3
00010000-00011000 rw-p 00000000 01:00 3119108    /test3
00011000-00012000 rw-p 00000000 00:00 0          [heap]
00012000-00013000 rw-p 00000000 00:00 0          [heap]

It looks like the bug has been there forever, and since it only results in
some information missing from a procfile, it does not fulfil the -stable
"critical issue" criteria.

Signed-off-by: Aaro Koskinen <>
Reviewed-by: KOSAKI Motohiro <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agosysctl: restrict write access to dmesg_restrict
Richard Weinberger [Wed, 23 Mar 2011 23:43:11 +0000 (16:43 -0700)]
sysctl: restrict write access to dmesg_restrict

commit bfdc0b497faa82a0ba2f9dddcf109231dd519fcc upstream.

When dmesg_restrict is set to 1 CAP_SYS_ADMIN is needed to read the kernel
ring buffer.  But a root user without CAP_SYS_ADMIN is able to reset
dmesg_restrict to 0.

This is an issue when e.g.  LXC (Linux Containers) are used and complete
user space is running without CAP_SYS_ADMIN.  A unprivileged and jailed
root user can bypass the dmesg_restrict protection.

With this patch writing to dmesg_restrict is only allowed when root has

Signed-off-by: Richard Weinberger <>
Acked-by: Dan Rosenberg <>
Acked-by: Serge E. Hallyn <>
Cc: Eric Paris <>
Cc: Kees Cook <>
Cc: James Morris <>
Cc: Eugene Teo <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agoext3: skip orphan cleanup on rocompat fs
Amir Goldstein [Sat, 26 Feb 2011 20:40:19 +0000 (22:40 +0200)]
ext3: skip orphan cleanup on rocompat fs

commit ce654b37f87980d95f339080e4c3bdb2370bdf22 upstream.

Orphan cleanup is currently executed even if the file system has some
number of unknown ROCOMPAT features, which deletes inodes and frees
blocks, which could be very bad for some RO_COMPAT features.

This patch skips the orphan cleanup if it contains readonly compatible
features not known by this ext3 implementation, which would prevent
the fs from being mounted (or remounted) readwrite.

Signed-off-by: Amir Goldstein <>
Signed-off-by: Jan Kara <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agoALSA: HDA: Fix internal mic on Dell E5420/E5520
David Henningsson [Tue, 22 Mar 2011 09:23:28 +0000 (10:23 +0100)]
ALSA: HDA: Fix internal mic on Dell E5420/E5520

This is a fixup for the 2.6.38 kernel, as the issue is being resolved
by upstream commits 699d899560cd7e72da39231e584412e7ac8114a4 and
094a42452abd5564429045e210281c6d22e67fca - which are too invasive
to reach 2.6.38. Instead make pin fixes as a workaround.

Tested-by: Kent Baxley <>
Signed-off-by: David Henningsson <>
Acked-by: Takashi Iwai <>
10 years agoPrevent rt_sigqueueinfo and rt_tgsigqueueinfo from spoofing the signal code
Julien Tinnes [Fri, 18 Mar 2011 22:05:21 +0000 (15:05 -0700)]
Prevent rt_sigqueueinfo and rt_tgsigqueueinfo from spoofing the signal code

commit da48524eb20662618854bb3df2db01fc65f3070c upstream.

Userland should be able to trust the pid and uid of the sender of a
signal if the si_code is SI_TKILL.

Unfortunately, the kernel has historically allowed sigqueueinfo() to
send any si_code at all (as long as it was negative - to distinguish it
from kernel-generated signals like SIGILL etc), so it could spoof a
SI_TKILL with incorrect siginfo values.

Happily, it looks like glibc has always set si_code to the appropriate
SI_QUEUE, so there are probably no actual user code that ever uses
anything but the appropriate SI_QUEUE flag.

So just tighten the check for si_code (we used to allow any negative
value), and add a (one-time) warning in case there are binaries out
there that might depend on using other si_code values.

Signed-off-by: Julien Tinnes <>
Acked-by: Oleg Nesterov <>
Signed-off-by: Linus Torvalds <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agointel_idle: disable Atom/Lincroft HW C-state auto-demotion
Len Brown [Wed, 16 Feb 2011 06:32:48 +0000 (01:32 -0500)]
intel_idle: disable Atom/Lincroft HW C-state auto-demotion

commit bfb53ccf1c734b1907df7189eef4c08489827951 upstream.

Just as we had to disable auto-demotion for NHM/WSM,
we need to do the same for Atom (Lincroft version).

In particular, auto-demotion will prevent Lincroft
from entering the S0i3 idle power saving state.

Signed-off-by: Len Brown <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agointel_idle: disable NHM/WSM HW C-state auto-demotion
Len Brown [Wed, 19 Jan 2011 01:48:27 +0000 (20:48 -0500)]
intel_idle: disable NHM/WSM HW C-state auto-demotion

commit 14796fca2bd22acc73dd0887248d003b0f441d08 upstream.

Hardware C-state auto-demotion is a mechanism where the HW overrides
the OS C-state request, instead demoting to a shallower state,
which is less expensive, but saves less power.

Modern Linux should generally get exactly the states it requests.
In particular, when a CPU is taken off-line, it must not be demoted, else
it can prevent the entire package from reaching deep C-states.

Signed-off-by: Len Brown <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agoxen: set max_pfn_mapped to the last pfn mapped
Stefano Stabellini [Fri, 18 Feb 2011 11:32:40 +0000 (11:32 +0000)]
xen: set max_pfn_mapped to the last pfn mapped

commit 14988a4d350ce3b41ecad4f63c4f44c56f5ae34d upstream.

Do not set max_pfn_mapped to the end of the initial memory mappings,
that also contain pages that don't belong in pfn space (like the mfn

Set max_pfn_mapped to the last real pfn mapped in the initial memory
mappings that is the pfn backing _end.

Signed-off-by: Stefano Stabellini <>
Acked-by: Konrad Rzeszutek Wilk <>
LKML-Reference: <alpine.DEB.2.00.1103171739050.3382@kaball-desktop>
Signed-off-by: H. Peter Anvin <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agoPCI hotplug: acpiphp: set current_state to D0 in register_slot
Stefano Stabellini [Mon, 28 Feb 2011 16:20:11 +0000 (16:20 +0000)]
PCI hotplug: acpiphp: set current_state to D0 in register_slot

commit 47e9037ac16637cd7f12b8790ea7ce6680e42168 upstream.

If a device doesn't support power management (pm_cap == 0) but it is
acpi_pci_power_manageable() because there is a _PS0 method declared for
it and _EJ0 is also declared for the slot then nobody is going to set
current_state = PCI_D0 for this device.  This is what I think it is

/* here we do not set current_state because !pm_cap */
/* platform_pci_power_manageable() calls acpi_pci_power_manageable that
 * returns true */
/* acpi_pci_set_power_state gets called and does nothing because the
 * acpi device has _EJ0, see the comment "If the ACPI device has _EJ0,
 * ignore the device" */

at this point if we refer to the commit message that introduced the
comment above (10b3dcae0f275e2546e55303d64ddbb58cec7599), it is up to
the hotplug driver to set the state to D0.
However AFAICT the pci hotplug driver never does, in fact
drivers/pci/hotplug/acpiphp_glue.c:register_slot sets the slot flags to
(SLOT_ENABLED | SLOT_POWEREDON) but it does not set the pci device
current state to PCI_D0.

So my proposed fix is also to set current_state = PCI_D0 in
Comments are very welcome.

Signed-off-by: Stefano Stabellini <>
Signed-off-by: Jesse Barnes <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agooom: avoid deferring oom killer if exiting task is being traced
David Rientjes [Tue, 22 Mar 2011 23:30:12 +0000 (16:30 -0700)]
oom: avoid deferring oom killer if exiting task is being traced

commit edd45544c6f09550df0a5491aa8a07af24767e73 upstream.

The oom killer naturally defers killing anything if it finds an eligible
task that is already exiting and has yet to detach its ->mm.  This avoids
unnecessarily killing tasks when one is already in the exit path and may
free enough memory that the oom killer is no longer needed.  This is
detected by PF_EXITING since threads that have already detached its ->mm
are no longer considered at all.

The problem with always deferring when a thread is PF_EXITING, however, is
that it may never actually exit when being traced, specifically if another
task is tracing it with PTRACE_O_TRACEEXIT.  The oom killer does not want
to defer in this case since there is no guarantee that thread will ever
exit without intervention.

This patch will now only defer the oom killer when a thread is PF_EXITING
and no ptracer has stopped its progress in the exit path.  It also ensures
that a child is sacrificed for the chosen parent only if it has a
different ->mm as the comment implies: this ensures that the thread group
leader is always targeted appropriately.

Signed-off-by: David Rientjes <>
Reported-by: Oleg Nesterov <>
Cc: KOSAKI Motohiro <>
Cc: KAMEZAWA Hiroyuki <>
Cc: Hugh Dickins <>
Cc: Andrey Vagin <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agooom: skip zombies when iterating tasklist
Andrey Vagin [Tue, 22 Mar 2011 23:30:11 +0000 (16:30 -0700)]
oom: skip zombies when iterating tasklist

commit 30e2b41f20b6238f51e7cffb879c7a0f0073f5fe upstream.

We shouldn't defer oom killing if a thread has already detached its ->mm
and still has TIF_MEMDIE set.  Memory needs to be freed, so find kill
other threads that pin the same ->mm or find another task to kill.

Signed-off-by: Andrey Vagin <>
Signed-off-by: David Rientjes <>
Cc: KOSAKI Motohiro <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agooom: prevent unnecessary oom kills or kernel panics
David Rientjes [Tue, 22 Mar 2011 23:30:09 +0000 (16:30 -0700)]
oom: prevent unnecessary oom kills or kernel panics

commit 3a5dda7a17cf3706f79b86293f29db02d61e0d48 upstream.

This patch prevents unnecessary oom kills or kernel panics by reverting
two commits:

495789a5 (oom: make oom_score to per-process value)
cef1d352 (oom: multi threaded process coredump don't make deadlock)

First, 495789a5 (oom: make oom_score to per-process value) ignores the
fact that all threads in a thread group do not necessarily exit at the
same time.

It is imperative that select_bad_process() detect threads that are in the
exit path, specifically those with PF_EXITING set, to prevent needlessly
killing additional tasks.  If a process is oom killed and the thread group
leader exits, select_bad_process() cannot detect the other threads that
are PF_EXITING by iterating over only processes.  Thus, it currently
chooses another task unnecessarily for oom kill or panics the machine when
nothing else is eligible.

By iterating over threads instead, it is possible to detect threads that
are exiting and nominate them for oom kill so they get access to memory

Second, cef1d352 (oom: multi threaded process coredump don't make
deadlock) erroneously avoids making the oom killer a no-op when an
eligible thread other than current isfound to be exiting.  We want to
detect this situation so that we may allow that exiting thread time to
exit and free its memory; if it is able to exit on its own, that should
free memory so current is no loner oom.  If it is not able to exit on its
own, the oom killer will nominate it for oom kill which, in this case,
only means it will get access to memory reserves.

Without this change, it is easy for the oom killer to unnecessarily target
tasks when all threads of a victim don't exit before the thread group
leader or, in the worst case, panic the machine.

Signed-off-by: David Rientjes <>
Cc: KOSAKI Motohiro <>
Cc: KAMEZAWA Hiroyuki <>
Cc: Oleg Nesterov <>
Cc: Hugh Dickins <>
Cc: Andrey Vagin <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agomm: swap: unlock swapfile inode mutex before closing file on bad swapfiles
Mel Gorman [Tue, 22 Mar 2011 23:30:08 +0000 (16:30 -0700)]
mm: swap: unlock swapfile inode mutex before closing file on bad swapfiles

commit 52c50567d8ab0a0a87f12cceaa4194967854f0bd upstream.

If an administrator tries to swapon a file backed by NFS, the inode mutex is
taken (as it is for any swapfile) but later identified to be a bad swapfile
due to the lack of bmap and tries to cleanup. During cleanup, an attempt is
made to close the file but with inode->i_mutex still held. Closing an NFS
file syncs it which tries to acquire the inode mutex leading to deadlock. If
lockdep is enabled the following appears on the console;

    [ INFO: possible recursive locking detected ]
    2.6.38-rc8-autobuild #1
    swapon/2192 is trying to acquire lock:
     (&sb->s_type->i_mutex_key#13){+.+.+.}, at: vfs_fsync_range+0x47/0x7c

    but task is already holding lock:
     (&sb->s_type->i_mutex_key#13){+.+.+.}, at: sys_swapon+0x28d/0xae7

    other info that might help us debug this:
    1 lock held by swapon/2192:
     #0:  (&sb->s_type->i_mutex_key#13){+.+.+.}, at: sys_swapon+0x28d/0xae7

    stack backtrace:
    Pid: 2192, comm: swapon Not tainted 2.6.38-rc8-autobuild #1
    Call Trace:
        nfs_flush_one+0x0/0xdf [nfs]
        nfs_file_flush+0x64/0x69 [nfs]

This patch releases the mutex if its held before calling filep_close()
so swapon fails as expected without deadlock when the swapfile is backed
by NFS.  If accepted for 2.6.39, it should also be considered a -stable
candidate for 2.6.38 and 2.6.37.

Signed-off-by: Mel Gorman <>
Acked-by: Hugh Dickins <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agoshmem: let shared anonymous be nonlinear again
Hugh Dickins [Tue, 22 Mar 2011 23:33:43 +0000 (16:33 -0700)]
shmem: let shared anonymous be nonlinear again

commit bee4c36a5cf5c9f63ce1d7372aa62045fbd16d47 upstream.

Up to 2.6.22, you could use remap_file_pages(2) on a tmpfs file or a
shared mapping of /dev/zero or a shared anonymous mapping.  In 2.6.23 we
disabled it by default, but set VM_CAN_NONLINEAR to enable it on safe
mappings.  We made sure to set it in shmem_mmap() for tmpfs files, but
missed it in shmem_zero_setup() for the others.  Fix that at last.

Reported-by: Kenny Simpson <>
Signed-off-by: Hugh Dickins <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agocgroups: if you list_empty() a head then don't list_del() it
Phil Carmody [Tue, 22 Mar 2011 23:30:13 +0000 (16:30 -0700)]
cgroups: if you list_empty() a head then don't list_del() it

commit 8d2587970b8bdf7c8d9208e3f4bb93182aef1a0f upstream.

list_del() leaves poison in the prev and next pointers.  The next
list_empty() will compare those poisons, and say the list isn't empty.
Any list operations that assume the node is on a list because of such a
check will be fooled into dereferencing poison.  One needs to INIT the
node after the del, and fortunately there's already a wrapper for that -

Some of the dels are followed by deallocations, so can be ignored, and one
can be merged with an add to make a move.  Apart from that, I erred on the
side of caution in making nodes list_empty()-queriable.

Signed-off-by: Phil Carmody <>
Reviewed-by: Paul Menage <>
Cc: Li Zefan <>
Acked-by: Kirill A. Shutemov <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agoaio: wake all waiters when destroying ctx
Roland Dreier [Tue, 22 Mar 2011 23:35:10 +0000 (16:35 -0700)]
aio: wake all waiters when destroying ctx

commit e91f90bb0bb10be9cc8efd09a3cf4ecffcad0db1 upstream.

The test program below will hang because io_getevents() uses
add_wait_queue_exclusive(), which means the wake_up() in io_destroy() only
wakes up one of the threads.  Fix this by using wake_up_all() in the aio
code paths where we want to make sure no one gets stuck.

// t.c -- compile with gcc -lpthread -laio t.c

#include <libaio.h>
#include <pthread.h>
#include <stdio.h>
#include <unistd.h>

static const int nthr = 2;

void *getev(void *ctx)
struct io_event ev;
io_getevents(ctx, 1, 1, &ev, NULL);
printf("io_getevents returned\n");
return NULL;

int main(int argc, char *argv[])
io_context_t ctx = 0;
pthread_t thread[nthr];
int i;

io_setup(1024, &ctx);

for (i = 0; i < nthr; ++i)
pthread_create(&thread[i], NULL, getev, ctx);



for (i = 0; i < nthr; ++i)
pthread_join(thread[i], NULL);

return 0;

Signed-off-by: Roland Dreier <>
Reviewed-by: Jeff Moyer <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agoRevert "slab: Fix missing DEBUG_SLAB last user"
Pekka Enberg [Mon, 14 Feb 2011 15:46:21 +0000 (17:46 +0200)]
Revert "slab: Fix missing DEBUG_SLAB last user"

commit 3ff84a7f36554b257cd57325b1a7c1fa4b49fbe3 upstream.

This reverts commit 5c5e3b33b7cb959a401f823707bee006caadd76e.

The commit breaks ARM thusly:

| Mount-cache hash table entries: 512
| slab error in verify_redzone_free(): cache `idr_layer_cache': memory outside object was overwritten
| Backtrace:
| [<c0227088>] (dump_backtrace+0x0/0x110) from [<c0431afc>] (dump_stack+0x18/0x1c)
| [<c0431ae4>] (dump_stack+0x0/0x1c) from [<c0293304>] (__slab_error+0x28/0x30)
| [<c02932dc>] (__slab_error+0x0/0x30) from [<c0293a74>] (cache_free_debugcheck+0x1c0/0x2b8)
| [<c02938b4>] (cache_free_debugcheck+0x0/0x2b8) from [<c0293f78>] (kmem_cache_free+0x3c/0xc0)
| [<c0293f3c>] (kmem_cache_free+0x0/0xc0) from [<c032b1c8>] (ida_get_new_above+0x19c/0x1c0)
| [<c032b02c>] (ida_get_new_above+0x0/0x1c0) from [<c02af7ec>] (alloc_vfsmnt+0x54/0x144)
| [<c02af798>] (alloc_vfsmnt+0x0/0x144) from [<c0299830>] (vfs_kern_mount+0x30/0xec)
| [<c0299800>] (vfs_kern_mount+0x0/0xec) from [<c0299908>] (kern_mount_data+0x1c/0x20)
| [<c02998ec>] (kern_mount_data+0x0/0x20) from [<c02146c4>] (sysfs_init+0x68/0xc8)
| [<c021465c>] (sysfs_init+0x0/0xc8) from [<c02137d4>] (mnt_init+0x90/0x1b0)
| [<c0213744>] (mnt_init+0x0/0x1b0) from [<c0213388>] (vfs_caches_init+0x100/0x140)
| [<c0213288>] (vfs_caches_init+0x0/0x140) from [<c0208c0c>] (start_kernel+0x2e8/0x368)
| [<c0208924>] (start_kernel+0x0/0x368) from [<c0208034>] (__enable_mmu+0x0/0x2c)
c0113268: redzone 1:0xd84156c5c032b3ac, redzone 2:0xd84156c5635688c0.
| slab error in cache_alloc_debugcheck_after(): cache `idr_layer_cache': double free, or memory outside object was overwritten
| ...
c011307c: redzone 1:0x9f91102ffffffff, redzone 2:0x9f911029d74e35b
| slab: Internal list corruption detected in cache 'idr_layer_cache'(24), slabp c0113000(16). Hexdump:
| 000: 20 4f 10 c0 20 4f 10 c0 7c 00 00 00 7c 30 11 c0
| 010: 10 00 00 00 10 00 00 00 00 00 c9 17 fe ff ff ff
| 020: fe ff ff ff fe ff ff ff fe ff ff ff fe ff ff ff
| 030: fe ff ff ff fe ff ff ff fe ff ff ff fe ff ff ff
| 040: fe ff ff ff fe ff ff ff fe ff ff ff fe ff ff ff
| 050: fe ff ff ff fe ff ff ff fe ff ff ff 11 00 00 00
| 060: 12 00 00 00 13 00 00 00 14 00 00 00 15 00 00 00
| 070: 16 00 00 00 17 00 00 00 c0 88 56 63
| kernel BUG at /home/rmk/git/linux-2.6-rmk/mm/slab.c:2928!

Reported-and-analyzed-by: Russell King <>
Signed-off-by: Pekka Enberg <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agoethtool: Compat handling for struct ethtool_rxnfc
Ben Hutchings [Thu, 17 Mar 2011 07:34:32 +0000 (07:34 +0000)]
ethtool: Compat handling for struct ethtool_rxnfc

commit 3a7da39d165e0c363c294feec119db1427032afd upstream.

This structure was accidentally defined such that its layout can
differ between 32-bit and 64-bit processes.  Add compat structure
definitions and an ioctl wrapper function.

Signed-off-by: Ben Hutchings <>
Acked-by: Alexander Duyck <>
Signed-off-by: David S. Miller <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agoALSA: hda - VIA: Fix VT1708 can't build up Headphone control issue
Lydia Wang [Tue, 22 Mar 2011 08:26:36 +0000 (16:26 +0800)]
ALSA: hda - VIA: Fix VT1708 can't build up Headphone control issue

commit ee3c35c0827de02de414d08b2ddcbb910c2263ab upstream.

Since VT1708 didn't support the control of getting connection number,
building of headphone control will fail in via_hp_build() function.

Signed-off-by: Lydia Wang <>
Signed-off-by: Takashi Iwai <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agoALSA: hda - VIA: Fix codec type for VT1708BCE at the right timing
Lydia Wang [Tue, 22 Mar 2011 08:25:03 +0000 (16:25 +0800)]
ALSA: hda - VIA: Fix codec type for VT1708BCE at the right timing

commit 0341ccd7557fecafe6a79c55158670cf629d269e upstream.

Add get_codec_type() in via_new_spec() function to make sure getting
correct codec type before building mixer controls.

Signed-off-by: Lydia Wang <>
Signed-off-by: Takashi Iwai <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agoALSA: hda - VIA: Fix invalid A-A path volume adjust issue
Lydia Wang [Tue, 22 Mar 2011 08:24:10 +0000 (16:24 +0800)]
ALSA: hda - VIA: Fix invalid A-A path volume adjust issue

commit 169222813eec8403c76394fb7b35ecab98e3c607 upstream.

Modify vt_auto_create_analog_input_ctls() function to fix invalid a-a path
volume adjust issue for VT1708S, VT1702 and VT1716S codecs.

Signed-off-by: Lydia Wang <>
Signed-off-by: Takashi Iwai <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agoALSA: hda - VIA: Add missing support for VT1718S in A-A path
Lydia Wang [Tue, 22 Mar 2011 08:23:23 +0000 (16:23 +0800)]
ALSA: hda - VIA: Add missing support for VT1718S in A-A path

commit ab657e0cacc39d88145871c6a3c844597c02d406 upstream.

Modify mute_aa_path() function to support VT1718S codec.

Signed-off-by: Lydia Wang <>
Signed-off-by: Takashi Iwai <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agoALSA: hda - VIA: Fix independent headphone no sound issue
Lydia Wang [Tue, 22 Mar 2011 08:22:37 +0000 (16:22 +0800)]
ALSA: hda - VIA: Fix independent headphone no sound issue

commit ce0e5a9e81fbb153ee15ca60246c6722f07fc546 upstream.

Modify via_independent_hp_put() function to support VT1718S and VT1812
codecs, and fix independent headphone no sound issue.

Signed-off-by: Lydia Wang <>
Signed-off-by: Takashi Iwai <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agoALSA: hda - VIA: Fix stereo mixer recording no sound issue
Lydia Wang [Tue, 22 Mar 2011 08:21:38 +0000 (16:21 +0800)]
ALSA: hda - VIA: Fix stereo mixer recording no sound issue

commit bff5fbf50bd498c217994bd2d41a53ac3141185a upstream.

Modify function via_mux_enum_put() to fix stereo mixer recording
no sound issue.

Signed-off-by: Lydia Wang <>
Signed-off-by: Takashi Iwai <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agoLinux v2.6.37.5
Greg Kroah-Hartman [Wed, 23 Mar 2011 19:52:59 +0000 (12:52 -0700)]

10 years agohwmon: (sht15) Fix integer overflow in humidity calculation
Vivien Didelot [Mon, 21 Mar 2011 16:59:35 +0000 (17:59 +0100)]
hwmon: (sht15) Fix integer overflow in humidity calculation

commit ccd32e735de7a941906e093f8dca924bb05c5794 upstream.

An integer overflow occurs in the calculation of RHlinear when the
relative humidity is greater than around 30%. The consequence is a subtle
(but noticeable) error in the resulting humidity measurement.

Signed-off-by: Vivien Didelot <>
Signed-off-by: Jean Delvare <>
Cc: Jonathan Cameron <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agox86, binutils, xen: Fix another wrong size directive
Alexander van Heukelum [Fri, 11 Mar 2011 20:59:38 +0000 (21:59 +0100)]
x86, binutils, xen: Fix another wrong size directive

commit 371c394af27ab7d1e58a66bc19d9f1f3ac1f67b4 upstream.

The latest binutils ( breaks the build
yet another time, under CONFIG_XEN=y due to a .size directive that
refers to a slightly differently named (hence, to the now very
strict and unforgiving assembler, non-existent) symbol.

[ mingo:

   This unnecessary build breakage caused by new binutils
   version 2.21 gets escallated back several kernel releases spanning
   several years of Linux history, affecting over 130,000 upstream
   kernel commits (!), on CONFIG_XEN=y 64-bit kernels (i.e. essentially
   affecting all major Linux distro kernel configs).

   Git annotate tells us that this slight debug symbol code mismatch
   bug has been introduced in 2008 in commit 3d75e1b8:

     3d75e1b8        (Jeremy Fitzhardinge    2008-07-08 15:06:49 -0700 1231) ENTRY(xen_do_hypervisor_callback)   # do_hypervisor_callback(struct *pt_regs)

   The 'bug' is just a slight assymetry in ENTRY()/END()
   debug-symbols sequences, with lots of assembly code between the
   ENTRY() and the END():

     ENTRY(xen_do_hypervisor_callback)   # do_hypervisor_callback(struct *pt_regs)

   Human reviewers almost never catch such small mismatches, and binutils
   never even warned about it either.

   This new binutils version thus breaks the Xen build on all upstream kernels
   since v2.6.27, out of the blue.

   This makes a straightforward Git bisection of all 64-bit Xen-enabled kernels
   impossible on such binutils, for a bisection window of over hundred
   thousand historic commits. (!)

   This is a major fail on the side of binutils and binutils needs to turn
   this show-stopper build failure into a warning ASAP. ]

Signed-off-by: Alexander van Heukelum <>
Cc: Jeremy Fitzhardinge <>
Cc: Jan Beulich <>
Cc: H.J. Lu <>
Cc: Linus Torvalds <>
Cc: Andrew Morton <>
Cc: "H. Peter Anvin" <>
Cc: Kees Cook <>
LKML-Reference: <>
Signed-off-by: Ingo Molnar <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agommc: sdio: remember new card RCA when redetecting card
Stefan Nilsson XK [Tue, 1 Mar 2011 13:41:04 +0000 (14:41 +0100)]
mmc: sdio: remember new card RCA when redetecting card

commit 0aab3995485b8a994bf29a995a008c9ea4a28054 upstream.

During redetection of a SDIO card, a request for a new card RCA
was submitted to the card, but was then overwritten by the old RCA.
This caused the card to be deselected instead of selected when using
the incorrect RCA.  This bug's been present since the "oldcard"
handling was introduced in 2.6.32.

Signed-off-by: Stefan Nilsson XK <>
Reviewed-by: Ulf Hansson <>
Reviewed-by: Pawel Wieczorkiewicz <>
Signed-off-by: Linus Walleij <>
Signed-off-by: Chris Ball <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agommc: sdhci: Add Ricoh e823 PCI ID
Manoj Iyer [Fri, 11 Feb 2011 22:25:31 +0000 (16:25 -0600)]
mmc: sdhci: Add Ricoh e823 PCI ID

commit 5fd11c0754fa069b6aba64b65734aa2fb193552d upstream.

Signed-off-by: Manoj Iyer <>
Signed-off-by: Chris Ball <>
Signed-off-by: Greg Kroah-Hartman <>
10 years agoi2c: Fix typo in instantiating-devices document
Roman Fietze [Sun, 20 Mar 2011 13:50:52 +0000 (14:50 +0100)]
i2c: Fix typo in instantiating-devices document

commit 6ced9e6b3901af4ab6ac0a11231402c888286ea6 upstream.

The struct i2c_board_info member holding the name is "type", not

Signed-off-by: Roman Fietze <>
Signed-off-by: Jean Delvare <>
Signed-off-by: Greg Kroah-Hartman <>