6 years agor8169: missing barriers.
Francois Romieu [Fri, 5 Oct 2012 21:29:16 +0000 (23:29 +0200)]
r8169: missing barriers.

commit 1e874e041fc7c222cbd85b20c4406070be1f687a upstream.

Signed-off-by: Francois Romieu <>
Cc: Hayes Wang <>
Acked-by: David S. Miller <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agor8169: fix Config2 MSIEnable bit setting.
françois romieu [Fri, 5 Oct 2012 21:29:15 +0000 (23:29 +0200)]
r8169: fix Config2 MSIEnable bit setting.

commit 2ca6cf06d988fea21e812a86be79353352677c9c upstream.

The MSIEnable bit is only available for the 8169.

Avoid Config2 writes for the post-8169 8168 and 810x.

Reported-by: Su Kang Yin <>
Signed-off-by: Francois Romieu <>
Cc: Hayes Wang <>
Signed-off-by: David S. Miller <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agor8169: Rx FIFO overflow fixes.
Francois Romieu [Fri, 5 Oct 2012 21:29:14 +0000 (23:29 +0200)]
r8169: Rx FIFO overflow fixes.

commit 811fd3010cf512f2e23e6c4c912aad54516dc706 upstream.

Realtek has specified that the post 8168c gigabit chips and the post
8105e fast ethernet chips recover automatically from a Rx FIFO overflow.
The driver does not need to clear the RxFIFOOver bit of IntrStatus and
it should rather avoid messing it.

The implementation deserves some explanation:
1. events outside of the intr_event bit mask are now ignored. It enforces
   a no-processing policy for the events that either should not be there
   or should be ignored.

2. RxFIFOOver was already ignored in rtl_cfg_infos[RTL_CFG_1] for the
   whole 8168 line of chips with two exceptions:
   - RTL_GIGA_MAC_VER_22 since b5ba6d12bdac21bc0620a5089e0f24e362645efd
     ("use RxFIFO overflow workaround for 8168c chipset.").
     This one should now be correctly handled.
   - RTL_GIGA_MAC_VER_11 (8168b) which requires a different Rx FIFO
     overflow processing.

   Though it does not conform to Realtek suggestion above, the updated
   driver includes no change for RTL_GIGA_MAC_VER_12 and RTL_GIGA_MAC_VER_17.
   Both are 8168b. RTL_GIGA_MAC_VER_12 is common and a bit old so I'd rather
   wait for experimental evidence that the change suggested by Realtek really
   helps or does not hurt in unexpected ways.

   Removed case statements in rtl8169_interrupt are only 8168 relevant.

3. RxFIFOOver is masked for post 8105e 810x chips, namely the sole 8105e
   (RTL_GIGA_MAC_VER_30) itself.

Signed-off-by: Francois Romieu <>
Cc: hayeswang <>
Signed-off-by: David S. Miller <>
Reviewed-by: Jonathan Nieder <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agor8169: increase the delay parameter of pm_schedule_suspend
hayeswang [Fri, 5 Oct 2012 21:29:13 +0000 (23:29 +0200)]
r8169: increase the delay parameter of pm_schedule_suspend

commit 10953db8e1a278742ef7e64a3d1491802bcfa98b upstream
The link down would occur when reseting PHY. And it would take about 2 ~ 5
seconds from link down to link up. If the delay of pm_schedule_suspend is
not long enough, the device would enter runtime_suspend before link up.
After link up, the device would wake up and reset PHY again. Then, you
would find the driver keep in a loop of runtime_suspend and rumtime_resume.

Signed-off-by: Hayes Wang <>
Acked-by: Francois Romieu <>
Signed-off-by: David S. Miller <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agor8169: expand received packet length indication.
Francois Romieu [Fri, 5 Oct 2012 21:29:12 +0000 (23:29 +0200)]
r8169: expand received packet length indication.

commit deb9d93c89d311714a60809b28160e538e1cbb43 upstream.

8168d and above allow jumbo frames beyond 8k. Bump the received
packet length check before enabling jumbo frames on these chipsets.

Frame length indication covers bits 0..13 of the first Rx descriptor
32 bits for the 8169 and 8168. I only have authoritative documentation
for the allowed use of the extra (13) bit with the 8169 and 8168c.
Realtek's drivers use the same mask for the 816x and the fast ethernet
only 810x.

Signed-off-by: Francois Romieu <>
Acked-by: David S. Miller <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agor8169: jumbo fixes.
Francois Romieu [Fri, 5 Oct 2012 21:29:11 +0000 (23:29 +0200)]
r8169: jumbo fixes.

commit d58d46b5d85139d18eb939aa7279c160bab70484 upstream.

- fix features : jumbo frames and checksumming can not be used at the
  same time.

- introduce hw_jumbo_{enable / disable} helpers. Their content has been
  creatively extracted from Realtek's own drivers. As an illustration,
  it would be nice to know how/if the MaxTxPacketSize register operates
  when the device can work with a 9k jumbo frame as its documentation
  (8168c) can not be applied beyond ~7k.

- rtl_tx_performance_tweak is moved forward. No change.

Signed-off-by: Francois Romieu <>
Acked-by: David S. Miller <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agor8169: remove erroneous processing of always set bit.
Francois Romieu [Fri, 5 Oct 2012 21:29:10 +0000 (23:29 +0200)]
r8169: remove erroneous processing of always set bit.

commit e03f33af79f0772156e1a1a1e36bdddf8012b2e4 upstream.

When set, RxFOVF (resp. RxBOVF) is always 1 (resp. 0).

Signed-off-by: Francois Romieu <>
Cc: Hayes <>
Acked-by: David S. Miller <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agor8169: don't enable rx when shutdown.
Hayes Wang [Fri, 5 Oct 2012 21:29:09 +0000 (23:29 +0200)]
r8169: don't enable rx when shutdown.

commit aaa89c08d9ffa3739c93d65d98b73ec2aa2e93a5 upstream.

Only 8111b needs to enable rx when shutdowning with WoL.

Signed-off-by: Hayes Wang <>
Acked-by: Francois Romieu <>
Acked-by: David S. Miller <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agor8169: fix wake on lan setting for non-8111E.
Hayes Wang [Fri, 5 Oct 2012 21:29:08 +0000 (23:29 +0200)]
r8169: fix wake on lan setting for non-8111E.

commit d4ed95d796e5126bba51466dc07e287cebc8bd19 upstream.

Only 8111E needs enable RxConfig bit 0 ~ 3 when suspending or
shutdowning for wake on lan.

Signed-off-by: Hayes Wang <>
Acked-by: Francois Romieu <>
Acked-by: David S. Miller <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agorcu: Fix day-one dyntick-idle stall-warning bug
Paul E. McKenney [Sat, 22 Sep 2012 20:55:30 +0000 (13:55 -0700)]
rcu: Fix day-one dyntick-idle stall-warning bug

commit a10d206ef1a83121ab7430cb196e0376a7145b22 upstream.

Each grace period is supposed to have at least one callback waiting
for that grace period to complete.  However, if CONFIG_NO_HZ=n, an
extra callback-free grace period is no big problem -- it will chew up
a tiny bit of CPU time, but it will complete normally.  In contrast,
CONFIG_NO_HZ=y kernels have the potential for all the CPUs to go to
sleep indefinitely, in turn indefinitely delaying completion of the
callback-free grace period.  Given that nothing is waiting on this grace
period, this is also not a problem.

That is, unless RCU CPU stall warnings are also enabled, as they are
in recent kernels.  In this case, if a CPU wakes up after at least one
minute of inactivity, an RCU CPU stall warning will result.  The reason
that no one noticed until quite recently is that most systems have enough
OS noise that they will never remain absolutely idle for a full minute.
But there are some embedded systems with cut-down userspace configurations
that consistently get into this situation.

All this begs the question of exactly how a callback-free grace period
gets started in the first place.  This can happen due to the fact that
CPUs do not necessarily agree on which grace period is in progress.
If a CPU still believes that the grace period that just completed is
still ongoing, it will believe that it has callbacks that need to wait for
another grace period, never mind the fact that the grace period that they
were waiting for just completed.  This CPU can therefore erroneously
decide to start a new grace period.  Note that this can happen in
TREE_RCU and TREE_PREEMPT_RCU even on a single-CPU system:  Deadlock
considerations mean that the CPU that detected the end of the grace
period is not necessarily officially informed of this fact for some time.

Once this CPU notices that the earlier grace period completed, it will
invoke its callbacks.  It then won't have any callbacks left.  If no
other CPU has any callbacks, we now have a callback-free grace period.

This commit therefore makes CPUs check more carefully before starting a
new grace period.  This new check relies on an array of tail pointers
into each CPU's list of callbacks.  If the CPU is up to date on which
grace periods have completed, it checks to see if any callbacks follow
the RCU_DONE_TAIL segment, otherwise it checks to see if any callbacks
follow the RCU_WAIT_TAIL segment.  The reason that this works is that
the RCU_WAIT_TAIL segment will be promoted to the RCU_DONE_TAIL segment
as soon as the CPU is officially notified that the old grace period
has ended.

This change is to cpu_needs_another_gp(), which is called in a number
of places.  The only one that really matters is in rcu_start_gp(), where
the root rcu_node structure's ->lock is held, which prevents any
other CPU from starting or completing a grace period, so that the
comparison that determines whether the CPU is missing the completion
of a grace period is stable.

Reported-by: Becky Bruce <>
Reported-by: Subodh Nijsure <>
Reported-by: Paul Walmsley <>
Signed-off-by: Paul E. McKenney <>
Signed-off-by: Paul E. McKenney <>
Tested-by: Paul Walmsley <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agodrm/radeon: force MSIs on RS690 asics
Alex Deucher [Wed, 26 Sep 2012 16:40:45 +0000 (12:40 -0400)]
drm/radeon: force MSIs on RS690 asics

commit fb6ca6d154cdcd53e7f27f8dbba513830372699b upstream.

There are so many quirks, lets just try and force
this for all RS690s.  See:

Signed-off-by: Alex Deucher <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agodrm/radeon: Add MSI quirk for gateway RS690
Alex Deucher [Wed, 26 Sep 2012 16:31:45 +0000 (12:31 -0400)]
drm/radeon: Add MSI quirk for gateway RS690

commit 3a6d59df80897cc87812b6826d70085905bed013 upstream.

Fixes another system on:

Signed-off-by: Alex Deucher <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agodrm/radeon: only adjust default clocks on NI GPUs
Alex Deucher [Fri, 14 Sep 2012 14:59:26 +0000 (10:59 -0400)]
drm/radeon: only adjust default clocks on NI GPUs

commit 2e3b3b105ab3bb5b6a37198da4f193cd13781d13 upstream.

SI asics store voltage information differently so we
don't have a way to deal with it properly yet.

Signed-off-by: Alex Deucher <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agoALSA: USB: Support for (original) Xbox Communicator
Marko Friedemann [Mon, 3 Sep 2012 08:12:40 +0000 (10:12 +0200)]
ALSA: USB: Support for (original) Xbox Communicator

commit c05fce586d4da2dfe0309bef3795a8586e967bc3 upstream.

Added support for Xbox Communicator to USB quirks.

Signed-off-by: Marko Friedemann <>
Signed-off-by: Takashi Iwai <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agoALSA: usb - disable broken hw volume for Tenx TP6911
David Henningsson [Thu, 20 Sep 2012 08:20:41 +0000 (10:20 +0200)]
ALSA: usb - disable broken hw volume for Tenx TP6911

commit c10514394ef9e8de93a4ad8c8904d71dcd82c122 upstream.

While going through Ubuntu bugs, I discovered this patch being
posted and a confirmation that the patch works as expected.

Finding out how the hw volume really works would be preferrable
to just disabling the broken one, but this would be better than

Credit: sndfnsdfin (qawsnews)
Signed-off-by: David Henningsson <>
Signed-off-by: Takashi Iwai <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agoALSA: aloop - add locking to timer access
Omair Mohammed Abdullah [Sat, 29 Sep 2012 06:54:05 +0000 (12:24 +0530)]
ALSA: aloop - add locking to timer access

commit d4f1e48bd11e3df6a26811f7a1f06c4225d92f7d upstream.

When the loopback timer handler is running, calling del_timer() (for STOP
trigger) will not wait for the handler to complete before deactivating the
timer. The timer gets rescheduled in the handler as usual. Then a subsequent
START trigger will try to start the timer using add_timer() with a timer pending
leading to a kernel panic.

Serialize the calls to add_timer() and del_timer() using a spin lock to avoid

Signed-off-by: Omair Mohammed Abdullah <>
Signed-off-by: Vinod Koul <>
Signed-off-by: Takashi Iwai <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agomm: thp: fix pmd_present for split_huge_page and PROT_NONE with THP
Andrea Arcangeli [Mon, 8 Oct 2012 23:33:27 +0000 (16:33 -0700)]
mm: thp: fix pmd_present for split_huge_page and PROT_NONE with THP

commit 027ef6c87853b0a9df53175063028edb4950d476 upstream.

In many places !pmd_present has been converted to pmd_none.  For pmds
that's equivalent and pmd_none is quicker so using pmd_none is better.

However (unless we delete pmd_present) we should provide an accurate
pmd_present too.  This will avoid the risk of code thinking the pmd is non
present because it's under __split_huge_page_map, see the pmd_mknotpresent
there and the comment above it.

If the page has been mprotected as PROT_NONE, it would also lead to a
pmd_present false negative in the same way as the race with

Because the PSE bit stays on at all times (both during split_huge_page and
when the _PAGE_PROTNONE bit get set), we could only check for the PSE bit,
but checking the PROTNONE bit too is still good to remember pmd_present
must always keep PROT_NONE into account.

This explains a not reproducible BUG_ON that was seldom reported on the

The same issue is in pmd_large, it would go wrong with both PROT_NONE and
if it races with split_huge_page.

Signed-off-by: Andrea Arcangeli <>
Acked-by: Rik van Riel <>
Cc: Johannes Weiner <>
Cc: Hugh Dickins <>
Cc: Mel Gorman <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agomm: fix invalidate_complete_page2() lock ordering
Hugh Dickins [Mon, 8 Oct 2012 23:33:14 +0000 (16:33 -0700)]
mm: fix invalidate_complete_page2() lock ordering

commit ec4d9f626d5908b6052c2973f37992f1db52e967 upstream.

In fuzzing with trinity, lockdep protested "possible irq lock inversion
dependency detected" when isolate_lru_page() reenabled interrupts while
still holding the supposedly irq-safe tree_lock:


isolate_lru_page() is correct to enable interrupts unconditionally:
invalidate_complete_page2() is incorrect to call clear_page_mlock() while
holding tree_lock, which is supposed to nest inside lru_lock.

Both truncate_complete_page() and invalidate_complete_page() call
clear_page_mlock() before taking tree_lock to remove page from radix_tree.
 I guess invalidate_complete_page2() preferred to test PageDirty (again)
under tree_lock before committing to the munlock; but since the page has
already been unmapped, its state is already somewhat inconsistent, and no
worse if clear_page_mlock() moved up.

Reported-by: Sasha Levin <>
Deciphered-by: Andrew Morton <>
Signed-off-by: Hugh Dickins <>
Acked-by: Mel Gorman <>
Cc: Rik van Riel <>
Cc: Johannes Weiner <>
Cc: Michel Lespinasse <>
Cc: Ying Han <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agoASoC: wm9712: Fix name of Capture Switch
Mark Brown [Tue, 31 Jul 2012 17:37:29 +0000 (18:37 +0100)]
ASoC: wm9712: Fix name of Capture Switch

commit 689185b78ba6fbe0042f662a468b5565909dff7a upstream.

Help UIs associate it with the matching gain control.

Signed-off-by: Mark Brown <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agoext4: fix fdatasync() for files with only i_size changes
Jan Kara [Thu, 27 Sep 2012 01:52:20 +0000 (21:52 -0400)]
ext4: fix fdatasync() for files with only i_size changes

commit b71fc079b5d8f42b2a52743c8d2f1d35d655b1c5 upstream.

Code tracking when transaction needs to be committed on fdatasync(2) forgets
to handle a situation when only inode's i_size is changed. Thus in such
situations fdatasync(2) doesn't force transaction with new i_size to disk
and that can result in wrong i_size after a crash.

Fix the issue by updating inode's i_datasync_tid whenever its size is

Reported-by: Kristian Nielsen <>
Signed-off-by: Jan Kara <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agoext4: always set i_op in ext4_mknod()
Bernd Schubert [Thu, 27 Sep 2012 01:24:57 +0000 (21:24 -0400)]
ext4: always set i_op in ext4_mknod()

commit 6a08f447facb4f9e29fcc30fb68060bb5a0d21c2 upstream.

ext4_special_inode_operations have their own ifdef CONFIG_EXT4_FS_XATTR
to mask those methods. And ext4_iget also always sets it, so there is
an inconsistency.

Signed-off-by: Bernd Schubert <>
Signed-off-by: "Theodore Ts'o" <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agoext4: online defrag is not supported for journaled files
Dmitry Monakhov [Wed, 26 Sep 2012 16:32:54 +0000 (12:32 -0400)]
ext4: online defrag is not supported for journaled files

commit f066055a3449f0e5b0ae4f3ceab4445bead47638 upstream.

Proper block swap for inodes with full journaling enabled is
truly non obvious task. In order to be on a safe side let's
explicitly disable it for now.

Signed-off-by: Dmitry Monakhov <>
Signed-off-by: "Theodore Ts'o" <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agoPCI: Check P2P bridge for invalid secondary/subordinate range
Yinghai Lu [Tue, 11 Sep 2012 00:19:33 +0000 (17:19 -0700)]
PCI: Check P2P bridge for invalid secondary/subordinate range

commit 1965f66e7db08d1ebccd24a59043eba826cc1ce8 upstream.

For bridges with "secondary > subordinate", i.e., invalid bus number
apertures, we don't enumerate anything behind the bridge unless the
user specified "pci=assign-busses".

This patch makes us automatically try to reassign the downstream bus
numbers in this case (just for that bridge, not for all bridges as
"pci=assign-busses" does).

We don't discover all the devices on the Intel DP43BF motherboard
without this change (or "pci=assign-busses") because its BIOS configures
a bridge as:

    pci 0000:00:1e.0: PCI bridge to [bus 20-08] (subtractive decode)

[bhelgaas: changelog, change message to dev_info]
Reported-by: Brian C. Huffman <>
Reported-by: VL <>
Tested-by: VL <>
Signed-off-by: Yinghai Lu <>
Signed-off-by: Bjorn Helgaas <>
6 years agoSCSI: zfcp: only access zfcp_scsi_dev for valid scsi_device
Martin Peschke [Tue, 4 Sep 2012 13:23:36 +0000 (15:23 +0200)]
SCSI: zfcp: only access zfcp_scsi_dev for valid scsi_device

commit d436de8ce25f53a8a880a931886821f632247943 upstream.

__scsi_remove_device (e.g. due to dev_loss_tmo) calls
zfcp_scsi_slave_destroy which in turn sends a close LUN FSF request to
the adapter. After 30 seconds without response,
zfcp_erp_timeout_handler kicks the ERP thread failing the close LUN
ERP action. zfcp_erp_wait in zfcp_erp_lun_shutdown_wait and thus
zfcp_scsi_slave_destroy returns and then scsi_device is no longer
valid. Sometime later the response to the close LUN FSF request may
finally come in. However, commit
"[SCSI] zfcp: Use SCSI device data zfcp_scsi_dev instead of zfcp_unit"
introduced a number of attempts to unconditionally access struct
zfcp_scsi_dev through struct scsi_device causing a use-after-free.
This leads to an Oops due to kernel page fault in one of:
zfcp_fsf_abort_fcp_command_handler, zfcp_fsf_open_lun_handler,
zfcp_fsf_close_lun_handler, zfcp_fsf_req_trace,
Move dereferencing of zfcp private data zfcp_scsi_dev allocated in
scsi_device via scsi_transport_reserve_device after the check for
potentially aborted FSF request and thus no longer valid scsi_device.
Only then assign sdev_to_zfcp(sdev) to the local auto variable struct
zfcp_scsi_dev *zfcp_sdev.

Signed-off-by: Martin Peschke <>
Signed-off-by: Steffen Maier <>
Signed-off-by: James Bottomley <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agoSCSI: zfcp: restore refcount check on port_remove
Steffen Maier [Tue, 4 Sep 2012 13:23:34 +0000 (15:23 +0200)]
SCSI: zfcp: restore refcount check on port_remove

commit d99b601b63386f3395dc26a699ae703a273d9982 upstream.

Upstream commit f3450c7b917201bb49d67032e9f60d5125675d6a
"[SCSI] zfcp: Replace local reference counting with common kref"
accidentally dropped a reference count check before tearing down
zfcp_ports that are potentially in use by zfcp_units.
Even remote ports in use can be removed causing
unreachable garbage objects zfcp_ports with zfcp_units.
Thus units won't come back even after a manual port_rescan.
The kref of zfcp_port->dev.kobj is already used by the driver core.
We cannot re-use it to track the number of zfcp_units.
Re-introduce our own counter for units per port
and check on port_remove.

Signed-off-by: Steffen Maier <>
Reviewed-by: Heiko Carstens <>
Signed-off-by: James Bottomley <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agoSCSI: zfcp: remove invalid reference to list iterator variable
Julia Lawall [Tue, 4 Sep 2012 13:23:33 +0000 (15:23 +0200)]
SCSI: zfcp: remove invalid reference to list iterator variable

commit ca579c9f136af4274ccfd1bcaee7f38a29a0e2e9 upstream.

If list_for_each_entry, etc complete a traversal of the list, the iterator
variable ends up pointing to an address at an offset from the list head,
and not a meaningful structure.  Thus this value should not be used after
the end of the iterator.  Replace port->adapter->scsi_host by

This problem was found using Coccinelle (

Oversight in upsteam commit of v2.6.37
"[SCSI] zfcp: Move ACL/CFDC code to zfcp_cfdc.c"
which merged the content of zfcp_erp_port_access_changed().

Signed-off-by: Julia Lawall <>
Signed-off-by: Steffen Maier <>
Reviewed-by: Martin Peschke <>
Signed-off-by: James Bottomley <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agoSCSI: zfcp: Do not wakeup while suspended
Steffen Maier [Tue, 4 Sep 2012 13:23:32 +0000 (15:23 +0200)]
SCSI: zfcp: Do not wakeup while suspended

commit cb45214960bc989af8b911ebd77da541c797717d upstream.

If the mapping of FCP device bus ID and corresponding subchannel
is modified while the Linux image is suspended, the resume of FCP
devices can fail. During resume, zfcp gets callbacks from cio regarding
the modified subchannels but they can be arbitrarily mixed with the
restore/resume callback. Since the cio callbacks would trigger
adapter recovery, zfcp could wakeup before the resume callback.
Therefore, ignore the cio callbacks regarding subchannels while
being suspended. We can safely do so, since zfcp does not deal itself
with subchannels. For problem determination purposes, we still trace the
ignored callback events.

The following kernel messages could be seen on resume:

kernel: <WWPN>: parent <FCP device bus ID> should not be sleeping

As part of adapter reopen recovery, zfcp performs auto port scanning
which can erroneously try to register new remote ports with
scsi_transport_fc and the device core code complains about the parent
(adapter) still sleeping.

kernel: zfcp.3dff9c: <FCP device bus ID>:\
 Setting up the QDIO connection to the FCP adapter failed
<last kernel message repeated 3 more times>
kernel: zfcp.574d43: <FCP device bus ID>:\
 ERP cannot recover an error on the FCP device

In such cases, the adapter gave up recovery and remained blocked along
with its child objects: remote ports and LUNs/scsi devices. Even the
adapter shutdown as part of giving up recovery failed because the ccw
device state remained disconnected. Later, the corresponding remote
ports ran into dev_loss_tmo. As a result, the LUNs were erroneously
not available again after resume.

Even a manually triggered adapter recovery (e.g. sysfs attribute
failed, or device offline/online via sysfs) could not recover the
adapter due to the remaining disconnected state of the corresponding
ccw device.

Signed-off-by: Steffen Maier <>
Signed-off-by: James Bottomley <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agoSCSI: zfcp: Make trace record tags unique
Steffen Maier [Tue, 4 Sep 2012 13:23:30 +0000 (15:23 +0200)]
SCSI: zfcp: Make trace record tags unique

commit 0100998dbfe6dfcd90a6e912ca7ed6f255d48f25 upstream.

Duplicate fssrh_2 from a54ca0f62f953898b05549391ac2a8a4dad6482b
"[SCSI] zfcp: Redesign of the debug tracing for HBA records."
complicates distinction of generic status read response from
local link up.
Duplicate fsscth1 from 2c55b750a884b86dea8b4cc5f15e1484cc47a25c
"[SCSI] zfcp: Redesign of the debug tracing for SAN records."
complicates distinction of good common transport response from
invalid port handle.

Signed-off-by: Steffen Maier <>
Reviewed-by: Martin Peschke <>
Signed-off-by: James Bottomley <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agotg3: Fix TSO CAP for 5704 devs w / ASF enabled
Matt Carlson [Mon, 28 Nov 2011 09:41:03 +0000 (09:41 +0000)]
tg3: Fix TSO CAP for 5704 devs w / ASF enabled

[ Upstream commit cf9ecf4b631f649a964fa611f1a5e8874f2a76db ]

On the earliest TSO capable devices, TSO was accomplished through
firmware.  The TSO cannot coexist with ASF management firmware though.
The tg3 driver determines whether or not ASF is enabled by calling
tg3_get_eeprom_hw_cfg(), which checks a particular bit of NIC memory.
Commit dabc5c670d3f86d15ee4f42ab38ec5bd2682487d, entitled "tg3: Move
TSO_CAPABLE assignment", accidentally moved the code that determines
TSO capabilities earlier than the call to tg3_get_eeprom_hw_cfg().  As a
consequence, the driver was attempting to determine TSO capabilities
before it had all the data it needed to make the decision.

This patch fixes the problem by revisiting and reevaluating the decision
after tg3_get_eeprom_hw_cfg() is called.

Signed-off-by: Matt Carlson <>
Signed-off-by: Michael Chan <>
Signed-off-by: David S. Miller <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agoaoe: assert AoE packets marked as requiring no checksum
Ed Cashin [Wed, 19 Sep 2012 15:46:39 +0000 (15:46 +0000)]
aoe: assert AoE packets marked as requiring no checksum

[ Upstream commit 8babe8cc6570ed896b7b596337eb8fe730c3ff45 ]

In order for the network layer to see that AoE requires
no checksumming in a generic way, the packets must be
marked as requiring no checksum, so we make this requirement
explicit with the assertion.

Signed-off-by: Ed Cashin <>
Signed-off-by: David S. Miller <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agonet: do not disable sg for packets requiring no checksum
Ed Cashin [Wed, 19 Sep 2012 15:49:00 +0000 (15:49 +0000)]
net: do not disable sg for packets requiring no checksum

[ Upstream commit c0d680e577ff171e7b37dbdb1b1bf5451e851f04 ]

A change in a series of VLAN-related changes appears to have
inadvertently disabled the use of the scatter gather feature of
network cards for transmission of non-IP ethernet protocols like ATA
over Ethernet (AoE).  Below is a reference to the commit that
introduces a "harmonize_features" function that turns off scatter
gather when the NIC does not support hardware checksumming for the
ethernet protocol of an sk buff.

  commit f01a5236bd4b140198fbcc550f085e8361fd73fa
  Author: Jesse Gross <>
  Date:   Sun Jan 9 06:23:31 2011 +0000

      net offloading: Generalize netif_get_vlan_features().

The can_checksum_protocol function is not equipped to consider a
protocol that does not require checksumming.  Calling it for a
protocol that requires no checksum is inappropriate.

The patch below has harmonize_features call can_checksum_protocol when
the protocol needs a checksum, so that the network layer is not forced
to perform unnecessary skb linearization on the transmission of AoE
packets.  Unnecessary linearization results in decreased performance
and increased memory pressure, as reported here:

The problem has probably not been widely experienced yet, because
only recently has the aoe driver acquired the
ability to use payloads of over a page in size, with the patchset
recently included in the mm tree:

The aoe driver already could use payloads of
greater than a page in size, but its users generally do not use the
newest kernels.

Signed-off-by: Ed Cashin <>
Signed-off-by: David S. Miller <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agonetrom: copy_datagram_iovec can fail
Alan Cox [Tue, 4 Sep 2012 04:13:18 +0000 (04:13 +0000)]
netrom: copy_datagram_iovec can fail

[ Upstream commit 6cf5c951175abcec4da470c50565cc0afe6cd11d ]

Check for an error from this and if so bail properly.

Signed-off-by: Alan Cox <>
Signed-off-by: David S. Miller <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agol2tp: fix a typo in l2tp_eth_dev_recv()
Eric Dumazet [Tue, 4 Sep 2012 19:54:55 +0000 (15:54 -0400)]
l2tp: fix a typo in l2tp_eth_dev_recv()

[ Upstream commit c0cc88a7627c333de50b07b7c60b1d49d9d2e6cc ]

While investigating l2tp bug, I hit a bug in eth_type_trans(),
because not enough bytes were pulled in skb head.

Signed-off-by: Eric Dumazet <>
Signed-off-by: David S. Miller <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agoipv6: mip6: fix mip6_mh_filter()
Eric Dumazet [Tue, 25 Sep 2012 20:01:28 +0000 (22:01 +0200)]
ipv6: mip6: fix mip6_mh_filter()

[ Upstream commit 96af69ea2a83d292238bdba20e4508ee967cf8cb ]

mip6_mh_filter() should not modify its input, or else its caller
would need to recompute ipv6_hdr() if skb->head is reallocated.

Use skb_header_pointer() instead of pskb_may_pull()

Signed-off-by: Eric Dumazet <>
Signed-off-by: David S. Miller <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agoipv6: raw: fix icmpv6_filter()
Eric Dumazet [Tue, 25 Sep 2012 07:03:40 +0000 (07:03 +0000)]
ipv6: raw: fix icmpv6_filter()

[ Upstream commit 1b05c4b50edbddbdde715c4a7350629819f6655e ]

icmpv6_filter() should not modify its input, or else its caller
would need to recompute ipv6_hdr() if skb->head is reallocated.

Use skb_header_pointer() instead of pskb_may_pull() and
change the prototype to make clear both sk and skb are const.

Also, if icmpv6 header cannot be found, do not deliver the packet,
as we do in IPv4.

Signed-off-by: Eric Dumazet <>
Signed-off-by: David S. Miller <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agoipv4: raw: fix icmp_filter()
Eric Dumazet [Sat, 22 Sep 2012 00:08:29 +0000 (00:08 +0000)]
ipv4: raw: fix icmp_filter()

[ Upstream commit ab43ed8b7490cb387782423ecf74aeee7237e591 ]

icmp_filter() should not modify its input, or else its caller
would need to recompute ip_hdr() if skb->head is reallocated.

Use skb_header_pointer() instead of pskb_may_pull() and
change the prototype to make clear both sk and skb are const.

Signed-off-by: Eric Dumazet <>
Signed-off-by: David S. Miller <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agonet: guard tcp_set_keepalive() to tcp sockets
Eric Dumazet [Mon, 24 Sep 2012 07:00:11 +0000 (07:00 +0000)]
net: guard tcp_set_keepalive() to tcp sockets

[ Upstream commit 3e10986d1d698140747fcfc2761ec9cb64c1d582 ]

Its possible to use RAW sockets to get a crash in
tcp_set_keepalive() / sk_reset_timer()

Fix is to make sure socket is a SOCK_STREAM one.

Reported-by: Dave Jones <>
Signed-off-by: Eric Dumazet <>
Signed-off-by: David S. Miller <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agonet: small bug on rxhash calculation
Chema Gonzalez [Fri, 7 Sep 2012 13:40:50 +0000 (13:40 +0000)]
net: small bug on rxhash calculation

[ Upstream commit 6862234238e84648c305526af2edd98badcad1e0 ]

In the current rxhash calculation function, while the
sorting of the ports/addrs is coherent (you get the
same rxhash for packets sharing the same 4-tuple, in
both directions), ports and addrs are sorted
independently. This implies packets from a connection
between the same addresses but crossed ports hash to
the same rxhash.

For example, traffic between A=S:l and B=L:s is hashed
(in both directions) from {L, S, {s, l}}. The same
rxhash is obtained for packets between C=S:s and D=L:l.

This patch ensures that you either swap both addrs and ports,
or you swap none. Traffic between A and B, and traffic
between C and D, get their rxhash from different sources
({L, S, {l, s}} for A<->B, and {L, S, {s, l}} for C<->D)

The patch is co-written with Eric Dumazet <>

Signed-off-by: Chema Gonzalez <>
Signed-off-by: Eric Dumazet <>
Signed-off-by: David S. Miller <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agopppoe: drop PPPOX_ZOMBIEs in pppoe_release
Xiaodong Xu [Sat, 22 Sep 2012 00:09:32 +0000 (00:09 +0000)]
pppoe: drop PPPOX_ZOMBIEs in pppoe_release

[ Upstream commit 2b018d57ff18e5405823e5cb59651a5b4d946d7b ]

When PPPOE is running over a virtual ethernet interface (e.g., a
bonding interface) and the user tries to delete the interface in case
the PPPOE state is ZOMBIE, the kernel will loop forever while
unregistering net_device for the reference count is not decreased to
zero which should have been done with dev_put().

Signed-off-by: Xiaodong Xu <>
Signed-off-by: David S. Miller <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agosctp: Don't charge for data in sndbuf again when transmitting packet
Thomas Graf [Mon, 3 Sep 2012 04:27:42 +0000 (04:27 +0000)]
sctp: Don't charge for data in sndbuf again when transmitting packet

[ Upstream commit 4c3a5bdae293f75cdf729c6c00124e8489af2276 ]

SCTP charges wmem_alloc via sctp_set_owner_w() in sctp_sendmsg() and via
skb_set_owner_w() in sctp_packet_transmit(). If a sender runs out of
sndbuf it will sleep in sctp_wait_for_sndbuf() and expects to be waken up
by __sctp_write_space().

Buffer space charged via sctp_set_owner_w() is released in sctp_wfree()
which calls __sctp_write_space() directly.

Buffer space charged via skb_set_owner_w() is released via sock_wfree()
which calls sk->sk_write_space() _if_ SOCK_USE_WRITE_QUEUE is not set.
sctp_endpoint_init() sets SOCK_USE_WRITE_QUEUE on all sockets.

Therefore if sctp_packet_transmit() manages to queue up more than sndbuf
bytes, sctp_wait_for_sndbuf() will never be woken up again unless it is
interrupted by a signal.

This could be fixed by clearing the SOCK_USE_WRITE_QUEUE flag but ...

Charging for the data twice does not make sense in the first place, it
leads to overcharging sndbuf by a factor 2. Therefore this patch only
charges a single byte in wmem_alloc when transmitting an SCTP packet to
ensure that the socket stays alive until the packet has been released.

This means that control chunks are no longer accounted for in wmem_alloc
which I believe is not a problem as skb->truesize will typically lead
to overcharging anyway and thus compensates for any control overhead.

Signed-off-by: Thomas Graf <>
CC: Vlad Yasevich <>
CC: Neil Horman <>
CC: David Miller <>
Acked-by: Vlad Yasevich <>
Signed-off-by: David S. Miller <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agotcp: flush DMA queue before sk_wait_data if rcv_wnd is zero
Michal Kubeček [Fri, 14 Sep 2012 04:59:52 +0000 (04:59 +0000)]
tcp: flush DMA queue before sk_wait_data if rcv_wnd is zero

[ Upstream commit 15c041759bfcd9ab0a4e43f1c16e2644977d0467 ]

If recv() syscall is called for a TCP socket so that
  - IOAT DMA is used
  - MSG_WAITALL flag is used
  - requested length is bigger than sk_rcvbuf
  - enough data has already arrived to bring rcv_wnd to zero
then when tcp_recvmsg() gets to calling sk_wait_data(), receive
window can be still zero while sk_async_wait_queue exhausts
enough space to keep it zero. As this queue isn't cleaned until
the tcp_service_net_dma() call, sk_wait_data() cannot receive
any data and blocks forever.

If zero receive window and non-empty sk_async_wait_queue is
detected before calling sk_wait_data(), process the queue first.

Signed-off-by: Michal Kubecek <>
Signed-off-by: David S. Miller <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agoipv6: release reference of ip6_null_entry's dst entry in __ip6_del_rt
Gao feng [Wed, 19 Sep 2012 19:25:34 +0000 (19:25 +0000)]
ipv6: release reference of ip6_null_entry's dst entry in __ip6_del_rt

[ Upstream commit 6825a26c2dc21eb4f8df9c06d3786ddec97cf53b ]

as we hold dst_entry before we call __ip6_del_rt,
so we should alse call dst_release not only return
-ENOENT when the rt6_info is ip6_null_entry.

and we already hold the dst entry, so I think it's
safe to call dst_release out of the write-read lock.

Signed-off-by: Gao feng <>
Signed-off-by: David S. Miller <>
Signed-off-by: Greg Kroah-Hartman <>
6 years ago8021q: fix mac_len recomputation in vlan_untag()
Antonio Quartulli [Tue, 2 Oct 2012 06:14:17 +0000 (06:14 +0000)]
8021q: fix mac_len recomputation in vlan_untag()

[ Upstream commit 5316cf9a5197eb80b2800e1acadde287924ca975 ]

skb_reset_mac_len() relies on the value of the skb->network_header pointer,
therefore we must wait for such pointer to be recalculated before computing
the new mac_len value.

Signed-off-by: Antonio Quartulli <>
Signed-off-by: David S. Miller <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agosierra_net: Endianess bug fix.
Lennart Sorensen [Fri, 7 Sep 2012 12:14:02 +0000 (12:14 +0000)]
sierra_net: Endianess bug fix.

[ Upstream commit 2120c52da6fe741454a60644018ad2a6abd957ac ]

I discovered I couldn't get sierra_net to work on a powerpc.  Turns out
the firmware attribute check assumes the system is little endian and
hence fails because the attributes is a 16 bit value.

Signed-off-by: Len Sorensen <>
Signed-off-by: David S. Miller <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agopkt_sched: fix virtual-start-time update in QFQ
Paolo Valente [Sat, 15 Sep 2012 00:41:35 +0000 (00:41 +0000)]
pkt_sched: fix virtual-start-time update in QFQ

[ Upstream commit 71261956973ba9e0637848a5adb4a5819b4bae83 ]

If the old timestamps of a class, say cl, are stale when the class
becomes active, then QFQ may assign to cl a much higher start time
than the maximum value allowed. This may happen when QFQ assigns to
the start time of cl the finish time of a group whose classes are
characterized by a higher value of the ratio
max_class_pkt/weight_of_the_class with respect to that of
cl. Inserting a class with a too high start time into the bucket list
corrupts the data structure and may eventually lead to crashes.
This patch limits the maximum start time assigned to a class.

Signed-off-by: Paolo Valente <>
Signed-off-by: David S. Miller <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agonet-sched: sch_cbq: avoid infinite loop
Eric Dumazet [Tue, 11 Sep 2012 13:11:12 +0000 (13:11 +0000)]
net-sched: sch_cbq: avoid infinite loop

[ Upstream commit bdfc87f7d1e253e0a61e2fc6a75ea9d76f7fc03a ]

Its possible to setup a bad cbq configuration leading to
an infinite loop in cbq_classify()

ICMP="match ip protocol 1 0xff"
U32="protocol ip u32"
DST="match ip dst"
tc qdisc add dev $DEV_OUT root handle 1: cbq avpkt 1000 \
bandwidth 100mbit
tc class add dev $DEV_OUT parent 1: classid 1:1 cbq \
rate 512kbit allot 1500 prio 5 bounded isolated
tc filter add dev $DEV_OUT parent 1: prio 3 $U32 \
$ICMP $DST flowid 1:

Reported-by: Denys Fedoryschenko <>
Tested-by: Denys Fedoryschenko <>
Signed-off-by: Eric Dumazet <>
Signed-off-by: David S. Miller <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agonetxen: check for root bus in netxen_mask_aer_correctable
Nikolay Aleksandrov [Fri, 14 Sep 2012 05:50:03 +0000 (05:50 +0000)]
netxen: check for root bus in netxen_mask_aer_correctable

[ Upstream commit e4d1aa40e363ed3e0486aeeeb0d173f7f822737e ]

Add a check if pdev->bus->self == NULL (root bus). When attaching
a netxen NIC to a VM it can be on the root bus and the guest would
crash in netxen_mask_aer_correctable() because of a NULL pointer
dereference if CONFIG_PCIEAER is present.

Signed-off-by: Nikolay Aleksandrov <>
Signed-off-by: David S. Miller <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agoixp4xx_hss: fix build failure due to missing linux/module.h inclusion
Florian Fainelli [Mon, 10 Sep 2012 12:06:58 +0000 (14:06 +0200)]
ixp4xx_hss: fix build failure due to missing linux/module.h inclusion

[ Upstream commit 0b836ddde177bdd5790ade83772860940bd481ea ]

Commit 36a1211970193ce215de50ed1e4e1272bc814df1 (netprio_cgroup.h:
dont include module.h from other includes) made the following build
error on ixp4xx_hss pop up:

  CC [M]  drivers/net/wan/ixp4xx_hss.o
 drivers/net/wan/ixp4xx_hss.c:1412:20: error: expected ';', ',' or ')'
 before string constant
 drivers/net/wan/ixp4xx_hss.c:1413:25: error: expected ';', ',' or ')'
 before string constant
 drivers/net/wan/ixp4xx_hss.c:1414:21: error: expected ';', ',' or ')'
 before string constant
 drivers/net/wan/ixp4xx_hss.c:1415:19: error: expected ';', ',' or ')'
 before string constant
 make[8]: *** [drivers/net/wan/ixp4xx_hss.o] Error 1

This was previously hidden because ixp4xx_hss includes linux/hdlc.h which
includes linux/netdevice.h which includes linux/netprio_cgroup.h which
used to include linux/module.h. The real issue was actually present since
the initial commit that added this driver since it uses macros from
linux/module.h without including this file.

Signed-off-by: Florian Fainelli <>
Signed-off-by: David S. Miller <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agonet: ethernet: davinci_cpdma: decrease the desc count when cleaning up the remaining...
htbegin [Mon, 1 Oct 2012 16:42:43 +0000 (16:42 +0000)]
net: ethernet: davinci_cpdma: decrease the desc count when cleaning up the remaining packets

[ Upstream commit ffb5ba90017505a19e238e986e6d33f09e4df765 ]

chan->count is used by rx channel. If the desc count is not updated by
the clean up loop in cpdma_chan_stop, the value written to the rxfree
register in cpdma_chan_start will be incorrect.

Signed-off-by: Tao Hou <>
Signed-off-by: David S. Miller <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agoxfrm_user: ensure user supplied esn replay window is valid
Mathias Krause [Thu, 20 Sep 2012 10:01:49 +0000 (10:01 +0000)]
xfrm_user: ensure user supplied esn replay window is valid

[ Upstream commit ecd7918745234e423dd87fcc0c077da557909720 ]

The current code fails to ensure that the netlink message actually
contains as many bytes as the header indicates. If a user creates a new
state or updates an existing one but does not supply the bytes for the
whole ESN replay window, the kernel copies random heap bytes into the
replay bitmap, the ones happen to follow the XFRMA_REPLAY_ESN_VAL
netlink attribute. This leads to following issues:

1. The replay window has random bits set confusing the replay handling
   code later on.

2. A malicious user could use this flaw to leak up to ~3.5kB of heap
   memory when she has access to the XFRM netlink interface (requires

Known users of the ESN replay window are strongSwan and Steffen's
iproute2 patch (<>). The latter
uses the interface with a bitmap supplied while the former does not.
strongSwan is therefore prone to run into issue 1.

To fix both issues without breaking existing userland allow using the
XFRMA_REPLAY_ESN_VAL netlink attribute with either an empty bitmap or a
fully specified one. For the former case we initialize the in-kernel
bitmap with zero, for the latter we copy the user supplied bitmap. For
state updates the full bitmap must be supplied.

To prevent overflows in the bitmap length calculation the maximum size
of bmp_len is limited to 128 by this patch -- resulting in a maximum
replay window of 4096 packets. This should be sufficient for all real
life scenarios (RFC 4303 recommends a default replay window size of 64).

Signed-off-by: Mathias Krause <>
Cc: Steffen Klassert <>
Cc: Martin Willi <>
Cc: Ben Hutchings <>
Signed-off-by: David S. Miller <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agoxfrm_user: don't copy esn replay window twice for new states
Mathias Krause [Wed, 19 Sep 2012 11:33:43 +0000 (11:33 +0000)]
xfrm_user: don't copy esn replay window twice for new states

[ Upstream commit e3ac104d41a97b42316915020ba228c505447d21 ]

The ESN replay window was already fully initialized in
xfrm_alloc_replay_state_esn(). No need to copy it again.

Signed-off-by: Mathias Krause <>
Cc: Steffen Klassert <>
Acked-by: Steffen Klassert <>
Signed-off-by: David S. Miller <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agoxfrm_user: fix info leak in copy_to_user_tmpl()
Mathias Krause [Wed, 19 Sep 2012 11:33:41 +0000 (11:33 +0000)]
xfrm_user: fix info leak in copy_to_user_tmpl()

[ Upstream commit 1f86840f897717f86d523a13e99a447e6a5d2fa5 ]

The memory used for the template copy is a local stack variable. As
struct xfrm_user_tmpl contains multiple holes added by the compiler for
alignment, not initializing the memory will lead to leaking stack bytes
to userland. Add an explicit memset(0) to avoid the info leak.

Initial version of the patch by Brad Spengler.

Signed-off-by: Mathias Krause <>
Cc: Brad Spengler <>
Acked-by: Steffen Klassert <>
Signed-off-by: David S. Miller <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agoxfrm_user: fix info leak in copy_to_user_policy()
Mathias Krause [Wed, 19 Sep 2012 11:33:40 +0000 (11:33 +0000)]
xfrm_user: fix info leak in copy_to_user_policy()

[ Upstream commit 7b789836f434c87168eab067cfbed1ec4783dffd ]

The memory reserved to dump the xfrm policy includes multiple padding
bytes added by the compiler for alignment (padding bytes in struct
xfrm_selector and struct xfrm_userpolicy_info). Add an explicit
memset(0) before filling the buffer to avoid the heap info leak.

Signed-off-by: Mathias Krause <>
Acked-by: Steffen Klassert <>
Signed-off-by: David S. Miller <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agoxfrm_user: fix info leak in copy_to_user_state()
Mathias Krause [Wed, 19 Sep 2012 11:33:39 +0000 (11:33 +0000)]
xfrm_user: fix info leak in copy_to_user_state()

[ Upstream commit f778a636713a435d3a922c60b1622a91136560c1 ]

The memory reserved to dump the xfrm state includes the padding bytes of
struct xfrm_usersa_info added by the compiler for alignment (7 for
amd64, 3 for i386). Add an explicit memset(0) before filling the buffer
to avoid the info leak.

Signed-off-by: Mathias Krause <>
Acked-by: Steffen Klassert <>
Signed-off-by: David S. Miller <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agoxfrm_user: fix info leak in copy_to_user_auth()
Mathias Krause [Wed, 19 Sep 2012 11:33:38 +0000 (11:33 +0000)]
xfrm_user: fix info leak in copy_to_user_auth()

[ Upstream commit 4c87308bdea31a7b4828a51f6156e6f721a1fcc9 ]

copy_to_user_auth() fails to initialize the remainder of alg_name and
therefore discloses up to 54 bytes of heap memory via netlink to

Use strncpy() instead of strcpy() to fill the trailing bytes of alg_name
with null bytes.

Signed-off-by: Mathias Krause <>
Acked-by: Steffen Klassert <>
Signed-off-by: David S. Miller <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agoxfrm: fix a read lock imbalance in make_blackhole
Li RongQing [Mon, 17 Sep 2012 22:40:10 +0000 (22:40 +0000)]
xfrm: fix a read lock imbalance in make_blackhole

[ Upstream commit 433a19548061bb5457b6ab77ed7ea58ca6e43ddb ]

if xfrm_policy_get_afinfo returns 0, it has already released the read
lock, xfrm_policy_put_afinfo should not be called again.

Signed-off-by: Li RongQing <>
Signed-off-by: David S. Miller <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agoxfrm_user: return error pointer instead of NULL #2
Mathias Krause [Fri, 14 Sep 2012 09:58:32 +0000 (09:58 +0000)]
xfrm_user: return error pointer instead of NULL #2

[ Upstream commit c25463722509fef0ed630b271576a8c9a70236f3 ]

When dump_one_policy() returns an error, e.g. because of a too small
buffer to dump the whole xfrm policy, xfrm_policy_netlink() returns
NULL instead of an error pointer. But its caller expects an error
pointer and therefore continues to operate on a NULL skbuff.

Signed-off-by: Mathias Krause <>
Acked-by: Steffen Klassert <>
Signed-off-by: David S. Miller <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agoxfrm_user: return error pointer instead of NULL
Mathias Krause [Thu, 13 Sep 2012 11:41:26 +0000 (11:41 +0000)]
xfrm_user: return error pointer instead of NULL

[ Upstream commit 864745d291b5ba80ea0bd0edcbe67273de368836 ]

When dump_one_state() returns an error, e.g. because of a too small
buffer to dump the whole xfrm state, xfrm_state_netlink() returns NULL
instead of an error pointer. But its callers expect an error pointer
and therefore continue to operate on a NULL skbuff.

This could lead to a privilege escalation (execution of user code in
kernel context) if the attacker has CAP_NET_ADMIN and is able to map
address 0.

Signed-off-by: Mathias Krause <>
Acked-by: Steffen Klassert <>
Signed-off-by: David S. Miller <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agoxfrm: Workaround incompatibility of ESN and async crypto
Steffen Klassert [Tue, 4 Sep 2012 00:03:29 +0000 (00:03 +0000)]
xfrm: Workaround incompatibility of ESN and async crypto

[ Upstream commit 3b59df46a449ec9975146d71318c4777ad086744 ]

ESN for esp is defined in RFC 4303. This RFC assumes that the
sequence number counters are always up to date. However,
this is not true if an async crypto algorithm is employed.

If the sequence number counters are not up to date on sequence
number check, we may incorrectly update the upper 32 bit of
the sequence number. This leads to a DOS.

We workaround this by comparing the upper sequence number,
(used for authentication) with the upper sequence number
computed after the async processing. We drop the packet
if these numbers are different.

To do this, we introduce a recheck function that does this
check in the ESN case.

Signed-off-by: Steffen Klassert <>
Acked-by: Herbert Xu <>
Signed-off-by: David S. Miller <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agoworkqueue: add missing smp_wmb() in process_one_work()
Tejun Heo [Fri, 3 Aug 2012 17:30:45 +0000 (10:30 -0700)]
workqueue: add missing smp_wmb() in process_one_work()

commit 959d1af8cffc8fd38ed53e8be1cf4ab8782f9c00 upstream.

WORK_STRUCT_PENDING is used to claim ownership of a work item and
process_one_work() releases it before starting execution.  When
someone else grabs PENDING, all pre-release updates to the work item
should be visible and all updates made by the new owner should happen

Grabbing PENDING uses test_and_set_bit() and thus has a full barrier;
however, clearing doesn't have a matching wmb.  Given the preceding
spin_unlock and use of clear_bit, I don't believe this can be a
problem on an actual machine and there hasn't been any related report
but it still is theretically possible for clear_pending to permeate
upwards and happen before work->entry update.

Add an explicit smp_wmb() before work_clear_pending().

Signed-off-by: Tejun Heo <>
Cc: Oleg Nesterov <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agodrivers/scsi/atp870u.c: fix bad use of udelay
Martin Michlmayr [Fri, 5 Oct 2012 00:11:25 +0000 (17:11 -0700)]
drivers/scsi/atp870u.c: fix bad use of udelay

commit 0f6d93aa9d96cc9022b51bd10d462b03296be146 upstream.

The ACARD driver calls udelay() with a value > 2000, which leads to to
the following compilation error on ARM:

  ERROR: "__bad_udelay" [drivers/scsi/atp870u.ko] undefined!
  make[1]: *** [__modpost] Error 1

This is because udelay is defined on ARM, roughly speaking, as

#define udelay(n) ((n) > 2000 ? __bad_udelay() : \
__const_udelay((n) * ((2199023U*HZ)>>11)))

The argument to __const_udelay is the number of jiffies to wait divided
by 4, but this does not work unless the multiplication does not
overflow, and that is what the build error is designed to prevent.  The
intended behavior can be achieved by using mdelay to call udelay
multiple times in a loop.

[ adding context]
Signed-off-by: Martin Michlmayr <>
Signed-off-by: Jonathan Nieder <>
Cc: James Bottomley <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agokernel/sys.c: call disable_nonboot_cpus() in kernel_restart()
Shawn Guo [Fri, 5 Oct 2012 00:12:23 +0000 (17:12 -0700)]
kernel/sys.c: call disable_nonboot_cpus() in kernel_restart()

commit f96972f2dc6365421cf2366ebd61ee4cf060c8d5 upstream.

As kernel_power_off() calls disable_nonboot_cpus(), we may also want to
have kernel_restart() call disable_nonboot_cpus().  Doing so can help
machines that require boot cpu be the last alive cpu during reboot to
survive with kernel restart.

This fixes one reboot issue seen on imx6q (Cortex-A9 Quad).  The machine
requires that the restart routine be run on the primary cpu rather than
secondary ones.  Otherwise, the secondary core running the restart
routine will fail to come to online after reboot.

Signed-off-by: Shawn Guo <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agolib/gcd.c: prevent possible div by 0
Davidlohr Bueso [Fri, 5 Oct 2012 00:13:18 +0000 (17:13 -0700)]
lib/gcd.c: prevent possible div by 0

commit e96875677fb2b7cb739c5d7769824dff7260d31d upstream.

Account for all properties when a and/or b are 0:
gcd(0, 0) = 0
gcd(a, 0) = a
gcd(0, b) = b

Fixes no known problems in current kernels.

Signed-off-by: Davidlohr Bueso <>
Cc: Eric Dumazet <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agoPCI: acpiphp: check whether _ADR evaluation succeeded
Bjorn Helgaas [Wed, 20 Jun 2012 22:18:29 +0000 (16:18 -0600)]
PCI: acpiphp: check whether _ADR evaluation succeeded

commit dfb117b3e50c52c7b3416db4a4569224b8db80bb upstream.

Check whether we evaluated _ADR successfully.  Previously we ignored
failure, so we would have used garbage data from the stack as the device
and function number.

We return AE_OK so that we ignore only this slot and continue looking
for other slots.

Found by Coverity (CID 113981).

Signed-off-by: Bjorn Helgaas <>
[bwh: Backported to 2.6.32/3.0: adjust context]
Signed-off-by: Ben Hutchings <>
Signed-off-by: Greg Kroah-Hartman <>
Lin Ming [Mon, 16 Jul 2012 08:30:21 +0000 (16:30 +0800)]

commit fc54ab72959edbf229b65ac74b2f122d799ca002 upstream.

The _OSC method may exist in module level code,
so it must be called after ACPI_FULL_INITIALIZATION

On some new platforms with Zero-Power-Optical-Disk-Drive (ZPODD)
support, this fix is necessary to save power.

Signed-off-by: Lin Ming <>
Tested-by: Aaron Lu <>
Signed-off-by: Len Brown <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agomedia: rc: ite-cir: Initialise ite_dev::rdev earlier
Ben Hutchings [Sun, 19 Aug 2012 22:32:27 +0000 (19:32 -0300)]
media: rc: ite-cir: Initialise ite_dev::rdev earlier

commit 4b961180ef275035b1538317ffd0e21e80e63e77 upstream.

ite_dev::rdev is currently initialised in ite_probe() after
rc_register_device() returns.  If a newly registered device is opened
quickly enough, we may enable interrupts and try to use ite_dev::rdev
before it has been initialised.  Move it up to the earliest point we
can, right after calling rc_allocate_device().

Reported-and-tested-by: YunQiang Su <>
Signed-off-by: Ben Hutchings <>
Signed-off-by: Mauro Carvalho Chehab <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agokbuild: make: fix if_changed when command contains backslashes
Sascha Hauer [Fri, 5 Oct 2012 00:11:17 +0000 (17:11 -0700)]
kbuild: make: fix if_changed when command contains backslashes

commit c353acba28fb3fa1fd05fd6b85a9fc7938330f9c upstream.

The call if_changed mechanism does not work when the command contains
backslashes.  This basically is an issue with lzo and bzip2 compressed
kernels.  The compressed binaries do not contain the uncompressed image
size, so these use size_append to append the size.  This results in
backslashes in the executed command.  With this if_changed always
detects a change in the command and rebuilds the compressed image even
if nothing has changed.

Fix this by escaping backslashes in make-cmd

Signed-off-by: Sascha Hauer <>
Signed-off-by: Jan Luebbe <>
Cc: Sam Ravnborg <>
Cc: Bernhard Walle <>
Cc: Michal Marek <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agomn10300: only add -mmem-funcs to KBUILD_CFLAGS if gcc supports it
Geert Uytterhoeven [Fri, 5 Oct 2012 00:11:13 +0000 (17:11 -0700)]
mn10300: only add -mmem-funcs to KBUILD_CFLAGS if gcc supports it

commit 9957423f035c2071f6d1c5d2f095cdafbeb25ad7 upstream.

It seems the current (gcc 4.6.3) no longer provides this so make it

As reported by Tony before, the mn10300 architecture cross-compiles with
gcc-4.6.3 if -mmem-funcs is not added to KBUILD_CFLAGS.

Reported-by: Tony Breeds <>
Signed-off-by: Geert Uytterhoeven <>
Cc: David Howells <>
Cc: Koichi Yasutake <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agoLinux 3.0.45
Greg Kroah-Hartman [Sun, 7 Oct 2012 15:28:29 +0000 (08:28 -0700)]
Linux 3.0.45

6 years agoSCSI: scsi_dh_alua: Enable STPG for unavailable ports
Bart Van Assche [Fri, 24 Aug 2012 09:08:41 +0000 (09:08 +0000)]
SCSI: scsi_dh_alua: Enable STPG for unavailable ports

commit e47f8976d8e573928824a06748f7bc82c58d747f upstream.

A quote from SPC-4: "While in the unavailable primary target port
asymmetric access state, the device server shall support those of
the following commands that it supports while in the active/optimized
state: [ ... ] d) SET TARGET PORT GROUPS; [ ... ]". Hence enable
sending STPG to a target port group that is in the unavailable state.

Signed-off-by: Bart Van Assche <>
Reviewed-by: Mike Christie <>
Acked-by: Hannes Reinecke <>
Signed-off-by: James Bottomley <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agoSCSI: scsi_remove_target: fix softlockup regression on hot remove
Dan Williams [Wed, 29 Aug 2012 05:12:10 +0000 (22:12 -0700)]
SCSI: scsi_remove_target: fix softlockup regression on hot remove

commit bc3f02a795d3b4faa99d37390174be2a75d091bd upstream.

John reports:
 BUG: soft lockup - CPU#2 stuck for 23s! [kworker/u:8:2202]
 Call Trace:
  [<ffffffff8141782a>] scsi_remove_target+0xda/0x1f0
  [<ffffffff81421de5>] sas_rphy_remove+0x55/0x60
  [<ffffffff81421e01>] sas_rphy_delete+0x11/0x20
  [<ffffffff81421e35>] sas_port_delete+0x25/0x160
  [<ffffffff814549a3>] mptsas_del_end_device+0x183/0x270

...introduced by commit 3b661a9 "[SCSI] fix hot unplug vs async scan race".

Don't restart lookup of more stargets in the multi-target case, just
arrange to traverse the list once, on the assumption that new targets
are always added at the end.  There is no guarantee that the target will
change state in scsi_target_reap() so we can end up spinning if we

Acked-by: Jack Wang <>
LKML-Reference: <>
Reported-by: John Drescher <>
Tested-by: John Drescher <>
Signed-off-by: Dan Williams <>
Signed-off-by: James Bottomley <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agoPCI: honor child buses add_size in hot plug configuration
Yinghai Lu [Mon, 25 Jul 2011 20:08:38 +0000 (13:08 -0700)]
PCI: honor child buses add_size in hot plug configuration

commit be768912a49b10b68e96fbd8fa3cab0adfbd3091 upstream.

git commit c8adf9a3e873eddaaec11ac410a99ef6b9656938
    "PCI: pre-allocate additional resources to devices only after
successful allocation of essential resources."

fails to take into consideration the optional-resources needed by children
devices while calculating the optional-resource needed by the bridge.

This can be a problem on some setup. For example, if a hotplug bridge has 8
children hotplug bridges, the bridge should have enough resources to accomodate
the hotplug requirements for each of its children hotplug bridges.  Currently
this is not the case.

This patch fixes the problem.

Signed-off-by: Yinghai Lu <>
Reviewed-by: Ram Pai <>
Signed-off-by: Jesse Barnes <>
Cc: Andrew Worsley <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agox86/alternatives: Fix p6 nops on non-modular kernels
Avi Kivity [Wed, 22 Aug 2012 10:03:48 +0000 (13:03 +0300)]
x86/alternatives: Fix p6 nops on non-modular kernels

commit cb09cad44f07044d9810f18f6f9a6a6f3771f979 upstream.

Probably a leftover from the early days of self-patching, p6nops
are marked __initconst_or_module, which causes them to be
discarded in a non-modular kernel.  If something later triggers
patching, it will overwrite kernel code with garbage.

Reported-by: Tomas Racek <>
Signed-off-by: Avi Kivity <>
Cc: Michael Tokarev <>
Cc: Borislav Petkov <>
Cc: Marcelo Tosatti <>
Cc: Anthony Liguori <>
Cc: H. Peter Anvin <>
Cc: Alan Cox <>
Cc: Alan Cox <>
Signed-off-by: Ingo Molnar <>
Cc: Ben Jencks <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agoisci: fix isci_pci_probe() generates warning on efi failure path
Dan Williams [Fri, 22 Jun 2012 18:31:14 +0000 (11:31 -0700)]
isci: fix isci_pci_probe() generates warning on efi failure path

commit 6d70a74ffd616073a68ae0974d98819bfa8e6da6 upstream.

The oem parameter image embedded in the efi variable is at an offset
from the start of the variable.  However, in the failure path we try to
free the 'orom' pointer which is only valid when the paramaters are
being read from the legacy option-rom space.

Since failure to load the oem parameters is unlikely and we keep the
memory around in the success case just defer all de-allocation to devm.

Reported-by: Don Morris <>
Signed-off-by: Dan Williams <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agoIB/srp: Avoid having aborted requests hang
Bart Van Assche [Fri, 24 Aug 2012 10:29:11 +0000 (10:29 +0000)]
IB/srp: Avoid having aborted requests hang

commit d8536670916a685df116b5c2cb256573fd25e4e3 upstream.

We need to call scsi_done() for commands after we abort them.

Signed-off-by: Bart Van Assche <>
Acked-by: David Dillow <>
Signed-off-by: Roland Dreier <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agoIB/srp: Fix use-after-free in srp_reset_req()
Bart Van Assche [Fri, 24 Aug 2012 10:27:54 +0000 (10:27 +0000)]
IB/srp: Fix use-after-free in srp_reset_req()

commit 9b796d06d5d1b1e85ae2316a283ea11dd739ef96 upstream.

srp_free_req() uses the scsi_cmnd structure contents to unmap
buffers, so we must invoke srp_free_req() before we release
ownership of that structure.

Signed-off-by: Bart Van Assche <>
Acked-by: David Dillow <>
Signed-off-by: Roland Dreier <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agoIPoIB: Fix use-after-free of multicast object
Patrick McHardy [Thu, 30 Aug 2012 07:01:30 +0000 (07:01 +0000)]
IPoIB: Fix use-after-free of multicast object

commit bea1e22df494a729978e7f2c54f7bda328f74bc3 upstream.

Fix a crash in ipoib_mcast_join_task().  (with help from Or Gerlitz)

Commit c8c2afe360b7 ("IPoIB: Use rtnl lock/unlock when changing device
flags") added a call to rtnl_lock() in ipoib_mcast_join_task(), which
is run from the ipoib_workqueue, and hence the workqueue can't be
flushed from the context of ipoib_stop().

In the current code, ipoib_stop() (which doesn't flush the workqueue)
calls ipoib_mcast_dev_flush(), which goes and deletes all the
multicast entries.  This takes place without any synchronization with
a possible running instance of ipoib_mcast_join_task() for the same
ipoib device, leading to a crash due to NULL pointer dereference.

Fix this by making sure that the workqueue is flushed before
ipoib_mcast_dev_flush() is called.  To make that possible, we move the
RTNL-lock wrapped code to ipoib_mcast_join_finish().

Signed-off-by: Patrick McHardy <>
Signed-off-by: Roland Dreier <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agocan: mscan-mpc5xxx: fix return value check in mpc512x_can_get_clock()
Wei Yongjun [Fri, 21 Sep 2012 07:09:47 +0000 (15:09 +0800)]
can: mscan-mpc5xxx: fix return value check in mpc512x_can_get_clock()

commit f61bd0585dfc7d99db4936d7467de4ca8e2f7ea0 upstream.

In case of error, the function clk_get() returns ERR_PTR()
and never returns NULL pointer. The NULL test in the error
handling should be replaced with IS_ERR().

dpatch engine is used to auto generated this patch.

Signed-off-by: Wei Yongjun <>
Acked-by: Wolfgang Grandegger <>
Signed-off-by: Marc Kleine-Budde <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agoSCSI: hpsa: Use LUN reset instead of target reset
Stephen M. Cameron [Thu, 26 Jul 2012 16:34:10 +0000 (11:34 -0500)]
SCSI: hpsa: Use LUN reset instead of target reset

commit 21e89afd325849eb38adccf382df16cc895911f9 upstream.

It turns out Smart Array logical drives do not support target
reset and when the target reset fails, the logical drive will
be taken off line.  Symptoms look like this:

hpsa 0000:03:00.0: Abort request on C1:B0:T0:L0
hpsa 0000:03:00.0: resetting device 1:0:0:0
hpsa 0000:03:00.0: cp ffff880037c56000 is reported invalid (probably means target device no longer present)
hpsa 0000:03:00.0: resetting device failed.
sd 1:0:0:0: Device offlined - not ready after error recovery
sd 1:0:0:0: rejecting I/O to offline device
EXT3-fs error (device sdb1): read_block_bitmap:

LUN reset is supported though, and is what we should be using.
Target reset is also disruptive in shared SAS situations,
for example, an external MSA1210m which does support target
reset attached to Smart Arrays in multiple hosts -- a target
reset from one host is disruptive to other hosts as all LUNs
on the target will be reset and will abort all outstanding i/os
back to all the attached hosts.  So we should use LUN reset,
not target reset.

Tested this with Smart Array logical drives and with tape drives.
Not sure how this bug survived since 2009, except it must be very
rare for a Smart Array to require more than 30s to complete a request.

Signed-off-by: Stephen M. Cameron <>
Signed-off-by: James Bottomley <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agoSCSI: ibmvscsi: Fix host config length field overflow
Benjamin Herrenschmidt [Mon, 30 Jul 2012 01:33:05 +0000 (11:33 +1000)]
SCSI: ibmvscsi: Fix host config length field overflow

commit 225c56960fcafeccc2b6304f96cd3f0dbf42a16a upstream.

The length field in the host config packet is only 16-bit long, so
passing it 0x10000 (64K which is our standard PAGE_SIZE) doesn't
work and result in an empty config from the server.

Signed-off-by: Benjamin Herrenschmidt <>
Acked-by: Robert Jennings <>
Signed-off-by: James Bottomley <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agoUBI: fix autoresize handling in R/O mode
Artem Bityutskiy [Sat, 18 Aug 2012 12:11:42 +0000 (14:11 +0200)]
UBI: fix autoresize handling in R/O mode

commit abb3e01103eb4e2ea5c15e6fedbc74e08bd4cc2b upstream.

Currently UBI fails in autoresize when it is in R/O mode (e.g., because the
underlying MTD device is R/O). This patch fixes the issue - we just skip
autoresize and print a warning.

Reported-by: Pali Rohár <>
Signed-off-by: Artem Bityutskiy <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agon_gsm: memory leak in uplink error path
Russ Gorby [Mon, 13 Aug 2012 12:45:30 +0000 (13:45 +0100)]
n_gsm: memory leak in uplink error path

commit 88ed2a60610974443335c924d7cb8e5dcf9dbdc1 upstream.

Uplink (TX) network data will go through gsm_dlci_data_output_framed
there is a bug where if memory allocation fails, the skb which
has already been pulled off the list will be lost.

In addition TX skbs were being processed in LIFO order

Fixed the memory leak, and changed to FIFO order processing

Signed-off-by: Russ Gorby <>
Tested-by: Kappel, LaurentX <>
Signed-off-by: Alan Cox <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agoIncrease XHCI suspend timeout to 16ms
Michael Spang [Fri, 14 Sep 2012 17:05:49 +0000 (13:05 -0400)]
Increase XHCI suspend timeout to 16ms

commit a6e097dfdfd189b6929af6efa1d289af61858386 upstream.

The Intel XHCI specification says that after clearing the run/stop bit
the controller may take up to 16ms to halt. We've seen a device take
14ms, which with the current timeout of 10ms causes the kernel to
abort the suspend. Increasing the timeout to the recommended value
fixes the problem.

This patch should be backported to kernels as old as 2.6.37, that
contain the commit 5535b1d5f8885695c6ded783c692e3c0d0eda8ca "USB: xHCI:
PCI power management implementation".

Signed-off-by: Michael Spang <>
Signed-off-by: Sarah Sharp <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agocoredump: prevent double-free on an error path in core dumper
Denys Vlasenko [Wed, 26 Sep 2012 01:34:50 +0000 (11:34 +1000)]
coredump: prevent double-free on an error path in core dumper

commit f34f9d186df35e5c39163444c43b4fc6255e39c5 upstream.

In !CORE_DUMP_USE_REGSET case, if elf_note_info_init fails to allocate
memory for info->fields, it frees already allocated stuff and returns
error to its caller, fill_note_info.  Which in turn returns error to its
caller, elf_core_dump.  Which jumps to cleanup label and calls
free_note_info, which will happily try to free all info->fields again.

This is the fix.

Signed-off-by: Oleg Nesterov <>
Signed-off-by: Denys Vlasenko <>
Cc: Venu Byravarasu <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agon_gsm: added interlocking for gsm_data_lock for certain code paths
Russ Gorby [Mon, 13 Aug 2012 12:44:40 +0000 (13:44 +0100)]
n_gsm: added interlocking for gsm_data_lock for certain code paths

commit 5e44708f75b0f8712da715d6babb0c21089b2317 upstream.

There were some locking holes in the management of the MUX's
message queue for 2 code paths:
1) gsmld_write_wakeup
2) receipt of CMD_FCON flow-control message
In both cases gsm_data_kick is called w/o locking so it can collide
with other other instances of gsm_data_kick (pulling messages tx_tail)
or potentially other instances of __gsm_data_queu (adding messages to tx_head)

Changed to take the tx_lock in these 2 cases

Signed-off-by: Russ Gorby <>
Tested-by: Yin, Fengwei <>
Signed-off-by: Alan Cox <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agoxhci: Intel Panther Point BEI quirk.
Sarah Sharp [Wed, 19 Sep 2012 23:27:26 +0000 (16:27 -0700)]
xhci: Intel Panther Point BEI quirk.

commit 80fab3b244a22e0ca539d2439bdda50e81e5666f upstream.

When a device with an isochronous endpoint is behind a hub plugged into
the Intel Panther Point xHCI host controller, and the driver submits
multiple frames per URB, the xHCI driver will set the Block Event
Interrupt (BEI) flag on all but the last TD for the URB.  This causes
the host controller to place an event on the event ring, but not send an
interrupt.  When the last TD for the URB completes, BEI is cleared, and
we get an interrupt for the whole URB.

However, under a Panther Point xHCI host controller, if the parent hub
is unplugged when one or more events from transfers with BEI set are on
the event ring, a port status change event is placed on the event ring,
but no interrupt is generated.  This means URBs stop completing, and the
USB device disconnect is not noticed.  Something like a USB headset will
cause mplayer to hang when the device is disconnected.

If another transfer is sent (such as running `sudo lsusb -v`), the next
transfer event seems to "unstick" the event ring, the xHCI driver gets
an interrupt, and the disconnect is reported to the USB core.

The fix is not to use the BEI flag under the Panther Point xHCI host.
This will impact power consumption and system responsiveness, because
the xHCI driver will receive an interrupt for every frame in all
isochronous URBs instead of once per URB.

Intel chipset developers confirm that this bug will be hit if the BEI
flag is used on any endpoint, not just ones that are behind a hub.

This patch should be backported to kernels as old as 3.0, that contain
the commit 69e848c2090aebba5698a1620604c7dccb448684 "Intel xhci: Support
EHCI/xHCI port switching."

Signed-off-by: Sarah Sharp <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agofirmware: Add missing attributes to EFI variable attribute print out from sysfs
Khalid Aziz [Mon, 10 Sep 2012 18:52:42 +0000 (12:52 -0600)]
firmware: Add missing attributes to EFI variable attribute print out from sysfs

commit 7083909023bbe29b3176e92d2d089def1aa7aa1e upstream.

Some of the EFI variable attributes are missing from print out from
/sys/firmware/efi/vars/*/attributes. This patch adds those in. It also
updates code to use pre-defined constants for masking current value
of attributes.

Signed-off-by: Khalid Aziz <>
Reviewed-by: Kees Cook <>
Acked-by: Matthew Garrett <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agob43legacy: Fix crash on unload when firmware not available
Larry Finger [Wed, 26 Sep 2012 17:32:02 +0000 (12:32 -0500)]
b43legacy: Fix crash on unload when firmware not available

commit 2d838bb608e2d1f6cb4280e76748cb812dc822e7 upstream.

When b43legacy is loaded without the firmware being available, a following
unload generates a kernel NULL pointer dereference BUG as follows:

[  214.330789] BUG: unable to handle kernel NULL pointer dereference at 0000004c
[  214.330997] IP: [<c104c395>] drain_workqueue+0x15/0x170
[  214.331179] *pde = 00000000
[  214.331311] Oops: 0000 [#1] SMP
[  214.331471] Modules linked in: b43legacy(-) ssb pcmcia mac80211 cfg80211 af_packet mperf arc4 ppdev sr_mod cdrom sg shpchp yenta_socket pcmcia_rsrc pci_hotplug pcmcia_core battery parport_pc parport floppy container ac button edd autofs4 ohci_hcd ehci_hcd usbcore usb_common thermal processor scsi_dh_rdac scsi_dh_hp_sw scsi_dh_emc scsi_dh_alua scsi_dh fan thermal_sys hwmon ata_generic pata_ali libata [last unloaded: cfg80211]
[  214.333421] Pid: 3639, comm: modprobe Not tainted 3.6.0-rc6-wl+ #163 Source Technology VIC 9921/ALI Based Notebook
[  214.333580] EIP: 0060:[<c104c395>] EFLAGS: 00010246 CPU: 0
[  214.333687] EIP is at drain_workqueue+0x15/0x170
[  214.333788] EAX: c162ac40 EBX: cdfb8360 ECX: 0000002a EDX: 00002a2a
[  214.333890] ESI: 00000000 EDI: 00000000 EBP: cd767e7c ESP: cd767e5c
[  214.333957]  DS: 007b ES: 007b FS: 00d8 GS: 0033 SS: 0068
[  214.333957] CR0: 8005003b CR2: 0000004c CR3: 0c96a000 CR4: 00000090
[  214.333957] DR0: 00000000 DR1: 00000000 DR2: 00000000 DR3: 00000000
[  214.333957] DR6: ffff0ff0 DR7: 00000400
[  214.333957] Process modprobe (pid: 3639, ti=cd766000 task=cf802e90 task.ti=cd766000)
[  214.333957] Stack:
[  214.333957]  00000292 cd767e74 c12c5e09 00000296 00000296 cdfb8360 cdfb9220 00000000
[  214.333957]  cd767e90 c104c4fd cdfb8360 cdfb9220 cd682800 cd767ea4 d0c10184 cd682800
[  214.333957]  cd767ea4 cba31064 cd767eb8 d0867908 cba31064 d087e09c cd96f034 cd767ec4
[  214.333957] Call Trace:
[  214.333957]  [<c12c5e09>] ? skb_dequeue+0x49/0x60
[  214.333957]  [<c104c4fd>] destroy_workqueue+0xd/0x150
[  214.333957]  [<d0c10184>] ieee80211_unregister_hw+0xc4/0x100 [mac80211]
[  214.333957]  [<d0867908>] b43legacy_remove+0x78/0x80 [b43legacy]
[  214.333957]  [<d083654d>] ssb_device_remove+0x1d/0x30 [ssb]
[  214.333957]  [<c126f15a>] __device_release_driver+0x5a/0xb0
[  214.333957]  [<c126fb07>] driver_detach+0x87/0x90
[  214.333957]  [<c126ef4c>] bus_remove_driver+0x6c/0xe0
[  214.333957]  [<c1270120>] driver_unregister+0x40/0x70
[  214.333957]  [<d083686b>] ssb_driver_unregister+0xb/0x10 [ssb]
[  214.333957]  [<d087c488>] b43legacy_exit+0xd/0xf [b43legacy]
[  214.333957]  [<c1089dde>] sys_delete_module+0x14e/0x2b0
[  214.333957]  [<c110a4a7>] ? vfs_write+0xf7/0x150
[  214.333957]  [<c1240050>] ? tty_write_lock+0x50/0x50
[  214.333957]  [<c110a6f8>] ? sys_write+0x38/0x70
[  214.333957]  [<c1397c55>] syscall_call+0x7/0xb
[  214.333957] Code: bc 27 00 00 00 00 a1 74 61 56 c1 55 89 e5 e8 a3 fc ff ff 5d c3 90 55 89 e5 57 56 89 c6 53 b8 40 ac 62 c1 83 ec 14 e8 bb b7 34 00 <8b> 46 4c 8d 50 01 85 c0 89 56 4c 75 03 83 0e 40 80 05 40 ac 62
[  214.333957] EIP: [<c104c395>] drain_workqueue+0x15/0x170 SS:ESP 0068:cd767e5c
[  214.333957] CR2: 000000000000004c
[  214.341110] ---[ end trace c7e90ec026d875a6 ]---Index: wireless-testing/drivers/net/wireless/b43legacy/main.c

The problem is fixed by making certain that the ucode pointer is not NULL
before deregistering the driver in mac80211.

Signed-off-by: Larry Finger <>
Signed-off-by: John W. Linville <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agoserial: set correct baud_base for EXSYS EX-41092 Dual 16950
Flavio Leitner [Sat, 22 Sep 2012 00:04:34 +0000 (21:04 -0300)]
serial: set correct baud_base for EXSYS EX-41092 Dual 16950

commit 26e8220adb0aec43b7acafa0f1431760eee28522 upstream.

Apparently the same card model has two IDs, so this patch
complements the commit 39aced68d664291db3324d0fcf0985ab5626aac2
adding the missing one.

Signed-off-by: Flavio Leitner <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agoserial: pl011: handle corruption at high clock speeds
Linus Walleij [Wed, 26 Sep 2012 15:21:36 +0000 (17:21 +0200)]
serial: pl011: handle corruption at high clock speeds

commit c5dd553b9fd069892c9e2de734f4f604e280fa7a upstream.

This works around a few glitches in the ST version of the PL011
serial driver when using very high baud rates, as we do in the
Ux500: 3, 3.25, 4 and 4.05 Mbps.

Problem Observed/rootcause:

When using high baud-rates, and the baudrate*8 is getting close to
the provided clock frequency (so a division factor close to 1), when
using bursts of characters (so they are abutted), then it seems as if
there is not enough time to detect the beginning of the start-bit which
is a timing reference for the entire character, and thus the sampling
moment of character bits is moving towards the end of each bit, instead
of the middle.

Increase slightly the RX baud rate of the UART above the theoretical
baudrate by 5%. This will definitely give more margin time to the
UART_RX to correctly sample the data at the middle of the bit period.

Also fix the ages old copy-paste error in the very stressed comment,
it's referencing the registers used in the PL010 driver rather than
the PL011 ones.

Signed-off-by: Guillaume Jaunet <>
Signed-off-by: Christophe Arnal <>
Signed-off-by: Matthias Locher <>
Signed-off-by: Rajanikanth HV <>
Cc: Bibek Basu <>
Cc: Par-Gunnar Hjalmdahl <>
Signed-off-by: Linus Walleij <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agoTTY: ttyprintk, don't touch behind tty->write_buf
Jiri Slaby [Tue, 7 Aug 2012 19:47:39 +0000 (21:47 +0200)]
TTY: ttyprintk, don't touch behind tty->write_buf

commit ee8b593affdf893012e57f4c54a21984d1b0d92e upstream.

If a user provides a buffer larger than a tty->write_buf chunk and
passes '\r' at the end of the buffer, we touch an out-of-bound memory.

Add a check there to prevent this.

Signed-off-by: Jiri Slaby <>
Cc: Samo Pogacnik <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agoRemove BUG_ON from n_tty_read()
Stanislav Kozina [Thu, 16 Aug 2012 11:01:47 +0000 (12:01 +0100)]
Remove BUG_ON from n_tty_read()

commit e9490e93c1978b6669f3e993caa3189be13ce459 upstream.

Change the BUG_ON to WARN_ON and return in case of tty->read_buf==NULL. We want to track a
couple of long standing reports of this but at the same time we can avoid killing the box.

Signed-off-by: Stanislav Kozina <>
Signed-off-by: Alan Cox <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agostaging: comedi: fix memory leak for saved channel list
Ian Abbott [Wed, 19 Sep 2012 18:37:39 +0000 (19:37 +0100)]
staging: comedi: fix memory leak for saved channel list

commit c8cad4c89ee3b15935c532210ae6ebb5c0a2734d upstream.

When `do_cmd_ioctl()` allocates memory for the kernel copy of a channel
list, it frees any previously allocated channel list in
`async->cmd.chanlist` and replaces it with the new one.  However, if the
device is ever removed (or "detached") the cleanup code in
`cleanup_device()` in "drivers.c" does not free this memory so it is

A sensible place to free the kernel copy of the channel list is in
`do_become_nonbusy()` as at that point the comedi asynchronous command
associated with the channel list is no longer valid.  Free the channel
list in `do_become_nonbusy()` instead of `do_cmd_ioctl()` and clear the
pointer to prevent it being freed more than once.

Note that `cleanup_device()` could be called at an inappropriate time
while the comedi device is open, but that's a separate bug not related
to this this patch.

Signed-off-by: Ian Abbott <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agostaging: comedi: don't dereference user memory for INSN_INTTRIG
Ian Abbott [Tue, 18 Sep 2012 18:46:58 +0000 (19:46 +0100)]
staging: comedi: don't dereference user memory for INSN_INTTRIG

commit 5d06e3df280bd230e2eadc16372e62818c63e894 upstream.

`parse_insn()` is dereferencing the user-space pointer `insn->data`
directly when handling the `INSN_INTTRIG` comedi instruction.  It
shouldn't be using `insn->data` at all; it should be using the separate
`data` pointer passed to the function.  Fix it.

Signed-off-by: Ian Abbott <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agostaging: comedi: jr3_pci: fix iomem dereference
Ian Abbott [Thu, 27 Sep 2012 16:45:27 +0000 (17:45 +0100)]
staging: comedi: jr3_pci: fix iomem dereference

commit e1878957b4676a17cf398f7f5723b365e9a2ca48 upstream.

Correct a direct dereference of I/O memory to use an appropriate I/O
memory access function.  Note that the pointer being dereferenced is not
currently tagged with `__iomem` but I plan to correct that for 3.7.

Signed-off-by: Ian Abbott <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agostaging: comedi: s626: don't dereference insn->data
Ian Abbott [Mon, 24 Sep 2012 16:20:52 +0000 (17:20 +0100)]
staging: comedi: s626: don't dereference insn->data

commit b655c2c4782ed3e2e71d2608154e295a3e860311 upstream.

`s626_enc_insn_config()` is incorrectly dereferencing `insn->data` which
is a pointer to user memory.  It should be dereferencing the separate
`data` parameter that points to a copy of the data in kernel memory.

Signed-off-by: Ian Abbott <>
Reviewed-by: H Hartley Sweeten <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agostaging: speakup_soft: Fix reading of init string
Ben Hutchings [Sun, 16 Sep 2012 03:18:50 +0000 (04:18 +0100)]
staging: speakup_soft: Fix reading of init string

commit 40fe4f89671fb3c7ded94190fb267402a38b0261 upstream.

softsynth_read() reads a character at a time from the init string;
when it finds the null terminator it sets the initialized flag but
then repeats the last character.

Additionally, if the read() buffer is not big enough for the init
string, the next read() will start reading from the beginning again.
So the caller may never progress to reading anything else.

Replace the simple initialized flag with the current position in
the init string, carried over between calls.  Switch to reading
real data once this reaches the null terminator.

(This assumes that the length of the init string can't change, which
seems to be the case.  Really, the string and position belong together
in a per-file private struct.)

Tested-by: Samuel Thibault <>
Signed-off-by: Ben Hutchings <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agoUSB: qcaux: add Pantech vendor class match
Bjørn Mork [Wed, 19 Sep 2012 20:02:03 +0000 (22:02 +0200)]
USB: qcaux: add Pantech vendor class match

commit c638eb2872b3af079501e7ee44cbb8a5cce9b4b5 upstream.

The three Pantech devices UML190 (106c:3716), UML290 (106c:3718) and
P4200 (106c:3721) all use the same subclasses to identify vendor
specific functions.  Replace the existing device specific entries
with generic vendor matching, adding support for the P4200.

Signed-off-by: Bjørn Mork <>
Cc: Thomas Schäfer <>
Acked-by: Dan Williams <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agoUSB: ftdi_sio: add TIAO USB Multi-Protocol Adapter (TUMPA) support
Antonio Ospite [Sun, 23 Sep 2012 07:57:25 +0000 (09:57 +0200)]
USB: ftdi_sio: add TIAO USB Multi-Protocol Adapter (TUMPA) support

commit 54575b05af36959dfb6a49a3e9ca0c2b456b7126 upstream.

TIAO/DIYGADGET USB Multi-Protocol Adapter (TUMPA) is an FTDI FT2232H
based device which provides an easily accessible JTAG, SPI, I2C, serial

FTDI FT2232H provides two serial channels (A and B), but on the TUMPA
channel A is dedicated to JTAG/SPI while channel B can be used for
UART/RS-232: use the ftdi_jtag_quirk to expose only channel B as
a usb-serial interface to userspace.

Signed-off-by: Antonio Ospite <>
Signed-off-by: Greg Kroah-Hartman <>
6 years agoUSB: option: blacklist QMI interface on ZTE MF683
Bjørn Mork [Wed, 19 Sep 2012 20:02:12 +0000 (22:02 +0200)]
USB: option: blacklist QMI interface on ZTE MF683

commit 160c9425ac52cb30502be2d9c5e848cec91bb115 upstream.

Interface #5 on ZTE MF683 is a QMI/wwan interface.

Signed-off-by: Bjørn Mork <>
Cc: Shawn J. Goff <>
Signed-off-by: Greg Kroah-Hartman <>