]> Git Repo - linux.git/log
linux.git
10 years agoNFS: Fix a regression in the read() syscall
Trond Myklebust [Tue, 3 Mar 2015 04:32:08 +0000 (23:32 -0500)]
NFS: Fix a regression in the read() syscall

When invalidating the page cache for a regular file, we want to first
sync all dirty data to disk and then call invalidate_inode_pages2().
The latter relies on nfs_launder_page() and nfs_release_page() to deal
respectively with dirty pages, and unstable written pages.

When commit 9590544694bec ("NFS: avoid deadlocks with loop-back mounted
NFS filesystems.") changed the behaviour of nfs_release_page(), then it
made it possible for invalidate_inode_pages2() to fail with an EBUSY.
Unfortunately, that error is then propagated back to read().

Let's therefore work around the problem for now by protecting the call
to sync the data and invalidate_inode_pages2() so that they are atomic
w.r.t. the addition of new writes.
Later on, we can revisit whether or not we still need nfs_launder_page()
and nfs_release_page().

Signed-off-by: Trond Myklebust <[email protected]>
10 years agoLinux 4.0-rc2 v4.0-rc2
Linus Torvalds [Tue, 3 Mar 2015 17:04:59 +0000 (09:04 -0800)]
Linux 4.0-rc2

10 years agodrm/i915: Fix modeset state confusion in the load detect code
Daniel Vetter [Tue, 3 Mar 2015 16:31:21 +0000 (17:31 +0100)]
drm/i915: Fix modeset state confusion in the load detect code

This is a tricky story of the new atomic state handling and the legacy
code fighting over each another. The bug at hand is an underrun of the
framebuffer reference with subsequent hilarity caused by the load
detect code. Which is peculiar since the the exact same code works
fine as the implementation of the legacy setcrtc ioctl.

Let's look at the ingredients:

- Currently our code is a crazy mix of legacy modeset interfaces to
  set the parameters and half-baked atomic state tracking underneath.
  While this transition is going we're using the transitional plane
  helpers to update the atomic side (drm_plane_helper_disable/update
  and friends), i.e. plane->state->fb. Since the state structure owns
  the fb those functions take care of that themselves.

  The legacy state (specifically crtc->primary->fb) is still managed
  by the old code (and mostly by the drm core), with the fb reference
  counting done by callers (core drm for the ioctl or the i915 load
  detect code). The relevant commit is

  commit ea2c67bb4affa84080c616920f3899f123786e56
  Author: Matt Roper <[email protected]>
  Date:   Tue Dec 23 10:41:52 2014 -0800

      drm/i915: Move to atomic plane helpers (v9)

- drm_plane_helper_disable has special code to handle multiple calls
  in a row - it checks plane->crtc == NULL and bails out. This is to
  match the proper atomic implementation which needs the crtc to get
  at the implied locking context atomic updates always need. See

  commit acf24a395c5a9290189b080383564437101d411c
  Author: Daniel Vetter <[email protected]>
  Date:   Tue Jul 29 15:33:05 2014 +0200

      drm/plane-helper: transitional atomic plane helpers

- The universal plane code split out the implicit primary plane from
  the CRTC into it's own full-blown drm_plane object. As part of that
  the setcrtc ioctl (which updated both the crtc mode and primary
  plane) learned to set crtc->primary->crtc on modeset to make sure
  the plane->crtc assignments statate up to date in

  commit e13161af80c185ecd8dc4641d0f5df58f9e3e0af
  Author: Matt Roper <[email protected]>
  Date:   Tue Apr 1 15:22:38 2014 -0700

      drm: Add drm_crtc_init_with_planes() (v2)

  Unfortunately we've forgotten to update the load detect code. Which
  wasn't a problem since the load detect modeset is temporary and
  always undone before we drop the locks.

- Finally there is a organically grown history (i.e. don't ask) around
  who sets the legacy plane->fb for the various driver entry points.
  Originally updating that was the drivers duty, but for almost all
  places we've moved that (plus updating the refcounts) into the core.
  Again the exception is the load detect code.

Taking all together the following happens:
- The load detect code doesn't set crtc->primary->crtc. This is only
  really an issue on crtcs never before used or when userspace
  explicitly disabled the primary plane.

- The plane helper glue code short-circuits because of that and leaves
  a non-NULL fb behind in plane->state->fb and plane->fb. The state
  fb isn't a real problem (it's properly refcounted on its own), it's
  just the canary.

- Load detect code drops the reference for that fb, but doesn't set
  plane->fb = NULL. This is ok since it's still living in that old
  world where drivers had to clear the pointer but the core/callers
  handled the refcounting.

- On the next modeset the drm core notices plane->fb and takes care of
  refcounting it properly by doing another unref. This drops the
  refcount to zero, leaving state->plane now pointing at freed memory.

- intel_plane_duplicate_state still assume it owns a reference to that
  very state->fb and bad things start to happen.

Fix this all by applying the same duct-tape as for the legacy setcrtc
ioctl code and set crtc->primary->crtc properly.

Cc: Matt Roper <[email protected]>
Cc: Paul Bolle <[email protected]>
Cc: Rob Clark <[email protected]>
Cc: Paulo Zanoni <[email protected]>
Cc: Sean Paul <[email protected]>
Cc: Matt Roper <[email protected]>
Reported-and-tested-by: Linus Torvalds <[email protected]>
Reported-by: Paul Bolle <[email protected]>
Signed-off-by: Daniel Vetter <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
10 years agospi: img-spfi: Verify max spfi transfer length
Sifan Naeem [Mon, 2 Mar 2015 16:06:46 +0000 (16:06 +0000)]
spi: img-spfi: Verify max spfi transfer length

Maximum transfer length supported by SPFI is 65535, this is limited
by the number of bits available in SPFI TSize register to represent
the transfer size.
For transfer requests larger than the maximum supported the driver
will return an invalid argument error.

Signed-off-by: Sifan Naeem <[email protected]>
Reviewed-by: Andrew Bresticker <[email protected]>
Signed-off-by: Mark Brown <[email protected]>
10 years agoASoC: simple-card: Add a NULL pointer check in asoc_simple_card_dai_link_of
Vishal Thanki [Tue, 3 Mar 2015 13:29:00 +0000 (18:59 +0530)]
ASoC: simple-card: Add a NULL pointer check in asoc_simple_card_dai_link_of

Make sure devm_kzalloc() succeeds.

Signed-off-by: Vishal Thanki <[email protected]>
Signed-off-by: Mark Brown <[email protected]>
10 years agospi: fix a typo in comment.
Marcin Bis [Sun, 1 Mar 2015 12:49:32 +0000 (13:49 +0100)]
spi: fix a typo in comment.

alway -> always

Signed-off-by: Marcin Bis <[email protected]>
Signed-off-by: Mark Brown <[email protected]>
10 years agoiommu/io-pgtable-arm: Add built time dependency
Jean Delvare [Mon, 23 Feb 2015 09:05:21 +0000 (10:05 +0100)]
iommu/io-pgtable-arm: Add built time dependency

If io-pgtable-arm is an ARM-specific driver then configuration option
IOMMU_IO_PGTABLE_LPAE should not be presented to the user by default
for non-ARM kernels.

Signed-off-by: Jean Delvare <[email protected]>
Cc: Will Deacon <[email protected]>
Cc: Joerg Roedel <[email protected]>
Acked-by: Will Deacon <[email protected]>
Signed-off-by: Joerg Roedel <[email protected]>
10 years agoRevert "drm/rockchip: Flip select/depends in Kconfig"
Dave Airlie [Tue, 3 Mar 2015 10:58:43 +0000 (20:58 +1000)]
Revert "drm/rockchip: Flip select/depends in Kconfig"

This reverts commit 9c58e8dbd3bfe7197323c88a784617afeffa9f87.

This doesn't seem to fully fix this, Kbuild who knows.

Signed-off-by: Dave Airlie <[email protected]>
10 years agoKVM: s390/cpacf: Enable key wrapping by default
Tony Krowiak [Tue, 24 Feb 2015 19:06:57 +0000 (14:06 -0500)]
KVM: s390/cpacf: Enable key wrapping by default

z/VM and LPAR enable key wrapping by default, lets do the same on KVM.

Signed-off-by: Tony Krowiak <[email protected]>
Signed-off-by: Christian Borntraeger <[email protected]>
10 years agodrm/rockchip: Flip select/depends in Kconfig
Daniel Vetter [Tue, 3 Mar 2015 09:58:08 +0000 (10:58 +0100)]
drm/rockchip: Flip select/depends in Kconfig

Otherwise Kconfig gets confused and somehow ends up creating a 2nd drm
submenu. I couldn't find i915 because of this any more at first.

Cc: Andy Yan <[email protected]>
Cc: Russell King <[email protected]>
Cc: Philipp Zabel <[email protected]>
Cc: "Yann E. MORIN" <[email protected]>
Cc: [email protected]
Signed-off-by: Daniel Vetter <[email protected]>
Signed-off-by: Dave Airlie <[email protected]>
10 years agoMerge tag 'imx-drm-fixes-2015-02-24' of git://git.pengutronix.de/git/pza/linux into...
Dave Airlie [Tue, 3 Mar 2015 09:57:41 +0000 (19:57 +1000)]
Merge tag 'imx-drm-fixes-2015-02-24' of git://git.pengutronix.de/git/pza/linux into drm-fixes

imx-drm fixes for mode fixup, dw_hdmi/imx, and parallel-display

- A clock fix for too large pixel clocks depending on the
  DI clock flag simplification patch
- Pruning of unsupported modes and a missing end of array element
  for dw_hdmi-imx
- LVDS modeset fix for mode fixup
- Fix parallel-display deferred probing if drm_panel is used

* tag 'imx-drm-fixes-2015-02-24' of git://git.pengutronix.de/git/pza/linux:
  DRM: i.MX: parallel display: Support probe deferral for finding DRM panel
  drm/imx: imx-ldb: enable DI clock in encoder_mode_set
  drm/imx: dw_hdmi-imx: add end of array element to current control array
  drm/imx: dw_hdmi-imx: add mode_valid callback prune unsupported modes
  gpu: ipu-v3: do not divide by zero if the pixel clock is too large

10 years agoeCryptfs: don't pass fs-specific ioctl commands through
Tyler Hicks [Wed, 25 Feb 2015 01:28:10 +0000 (19:28 -0600)]
eCryptfs: don't pass fs-specific ioctl commands through

eCryptfs can't be aware of what to expect when after passing an
arbitrary ioctl command through to the lower filesystem. The ioctl
command may trigger an action in the lower filesystem that is
incompatible with eCryptfs.

One specific example is when one attempts to use the Btrfs clone
ioctl command when the source file is in the Btrfs filesystem that
eCryptfs is mounted on top of and the destination fd is from a new file
created in the eCryptfs mount. The ioctl syscall incorrectly returns
success because the command is passed down to Btrfs which thinks that it
was able to do the clone operation. However, the result is an empty
eCryptfs file.

This patch allows the trim, {g,s}etflags, and {g,s}etversion ioctl
commands through and then copies up the inode metadata from the lower
inode to the eCryptfs inode to catch any changes made to the lower
inode's metadata. Those five ioctl commands are mostly common across all
filesystems but the whitelist may need to be further pruned in the
future.

https://bugzilla.kernel.org/show_bug.cgi?id=93691
https://launchpad.net/bugs/1305335

Signed-off-by: Tyler Hicks <[email protected]>
Cc: Rocko <[email protected]>
Cc: Colin Ian King <[email protected]>
Cc: [email protected] # v2.6.36+: c43f7b8 eCryptfs: Handle ioctl calls with unlocked and compat functions
10 years agomtd: nand: MTD_NAND_HISI504 should depend on HAS_DMA
Geert Uytterhoeven [Sun, 1 Mar 2015 09:35:52 +0000 (10:35 +0100)]
mtd: nand: MTD_NAND_HISI504 should depend on HAS_DMA

If NO_DMA=y:

    drivers/built-in.o: In function `hisi_nfc_probe':
    hisi504_nand.c:(.text+0x23e646): undefined reference to `dmam_alloc_coherent'

Signed-off-by: Geert Uytterhoeven <[email protected]>
Signed-off-by: Brian Norris <[email protected]>
10 years agousb: ftdi_sio: Add jtag quirk support for Cyber Cortex AV boards
Max Mansfield [Tue, 3 Mar 2015 01:38:02 +0000 (18:38 -0700)]
usb: ftdi_sio: Add jtag quirk support for Cyber Cortex AV boards

This patch integrates Cyber Cortex AV boards with the existing
ftdi_jtag_quirk in order to use serial port 0 with JTAG which is
required by the manufacturers' software.

Steps: 2

[ftdi_sio_ids.h]
1. Defined the device PID

[ftdi_sio.c]
2. Added a macro declaration to the ids array, in order to enable the
jtag quirk for the device.

Signed-off-by: Max Mansfield <[email protected]>
Cc: stable <[email protected]>
Signed-off-by: Johan Hovold <[email protected]>
10 years agoudp: only allow UFO for packets from SOCK_DGRAM sockets
Michal Kubeček [Mon, 2 Mar 2015 17:27:11 +0000 (18:27 +0100)]
udp: only allow UFO for packets from SOCK_DGRAM sockets

If an over-MTU UDP datagram is sent through a SOCK_RAW socket to a
UFO-capable device, ip_ufo_append_data() sets skb->ip_summed to
CHECKSUM_PARTIAL unconditionally as all GSO code assumes transport layer
checksum is to be computed on segmentation. However, in this case,
skb->csum_start and skb->csum_offset are never set as raw socket
transmit path bypasses udp_send_skb() where they are usually set. As a
result, driver may access invalid memory when trying to calculate the
checksum and store the result (as observed in virtio_net driver).

Moreover, the very idea of modifying the userspace provided UDP header
is IMHO against raw socket semantics (I wasn't able to find a document
clearly stating this or the opposite, though). And while allowing
CHECKSUM_NONE in the UFO case would be more efficient, it would be a bit
too intrusive change just to handle a corner case like this. Therefore
disallowing UFO for packets from SOCK_DGRAM seems to be the best option.

Signed-off-by: Michal Kubecek <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
10 years agoMerge branch 'sh_eth'
David S. Miller [Tue, 3 Mar 2015 02:31:00 +0000 (21:31 -0500)]
Merge branch 'sh_eth'

Ben Hutchings says:

====================
Fixes for sh_eth #4 v2

I'm continuing review and testing of Ethernet support on the R-Car H2
chip, with help from a colleague.  This series fixes a few more issues.

These are not tested on any of the other supported chips.

v2: Add note that the revert is not a pure revert.
====================

Signed-off-by: David S. Miller <[email protected]>
10 years agosh_eth: Really fix padding of short frames on TX
Ben Hutchings [Tue, 3 Mar 2015 00:53:08 +0000 (00:53 +0000)]
sh_eth: Really fix padding of short frames on TX

My previous fix to clear padding of short frames used skb->len as the
DMA length, assuming that skb_padto() extended skb->len to include the
padding.  That isn't the case; we need to use skb_put_padto() instead.

(This wasn't immediately obvious because software padding isn't
actually needed on the R-Car H2.  We could make it conditional on
which chip is being driven, but it's probably not worth the effort.)

Reported-by: "Violeta Menéndez González" <[email protected]>
Fixes: 612a17a54b50 ("sh_eth: Fix padding of short frames on TX")
Signed-off-by: Ben Hutchings <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
10 years agoRevert "sh_eth: Enable Rx descriptor word 0 shift for r8a7790"
Ben Hutchings [Tue, 3 Mar 2015 00:52:39 +0000 (00:52 +0000)]
Revert "sh_eth: Enable Rx descriptor word 0 shift for r8a7790"

This reverts commit fd9af07c3404ac9ecbd0d859563360f51ce1ffde.

The hardware manual states that the frame error and multicast bits are
copied to bits 9:0 of RD0, not bits 25:16.  I've tested that this is
true for RFS1 (CRC error), RFS3 (frame too short), RFS4 (frame too
long) and RFS8 (multicast).

Also adjust a comment to agree with this.

Signed-off-by: Ben Hutchings <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
10 years agosh_eth: Fix RX recovery on R-Car in case of RX ring underrun
Ben Hutchings [Tue, 3 Mar 2015 00:52:08 +0000 (00:52 +0000)]
sh_eth: Fix RX recovery on R-Car in case of RX ring underrun

In case of RX ring underrun (RDE), we attempt to reset the software
descriptor pointers (dirty_rx and cur_rx) to match where the hardware
will read the next descriptor from, as that might not be the first
dirty descriptor.  This relies on reading RDFAR, but that register
doesn't exist on all supported chips - specifically, not on the R-Car
chips.  This will result in unpredictable behaviour on those chips
after an RDE.

Make this pointer reset conditional and assume that it isn't needed on
the R-Car chips.  This fix also assumes that RDFAR is never exposed at
offset 0 in the memory map - this is currently true, and a subsequent
commit will fix the ambiguity between offset 0 and no-offset in the
register offset maps.

Fixes: 79fba9f51755 ("net: sh_eth: fix the rxdesc pointer when rx ...")
Signed-off-by: Ben Hutchings <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
10 years agosh_eth: Ensure proper ordering of descriptor active bit write/read
Ben Hutchings [Tue, 3 Mar 2015 00:52:00 +0000 (00:52 +0000)]
sh_eth: Ensure proper ordering of descriptor active bit write/read

When submitting a DMA descriptor, the active bit must be written last.
When reading a completed DMA descriptor, the active bit must be read
first.

Add memory barriers to ensure that this ordering is maintained.

Signed-off-by: Ben Hutchings <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
10 years agoMerge branch 'fixes' of github.com:lmajewski/linux-samsung-thermal into work-fixes
Eduardo Valentin [Tue, 3 Mar 2015 02:15:34 +0000 (22:15 -0400)]
Merge branch 'fixes' of github.com:lmajewski/linux-samsung-thermal into work-fixes

Pull samsung thermal fixes from Lukasz Majewski:
"Changes:
- Exynos7 power down detection mode fix
- Fix for cpufreq cooling device regression
- Updating MAINTAINER's entry for Samsung Exynos Thermal"

Signed-off-by: Eduardo Valentin <[email protected]>
10 years agolivepatch: fix RCU usage in klp_find_external_symbol()
Peter Zijlstra [Sat, 28 Feb 2015 21:24:48 +0000 (22:24 +0100)]
livepatch: fix RCU usage in klp_find_external_symbol()

While one must hold RCU-sched (aka. preempt_disable) for find_symbol()
one must equally hold it over the use of the object returned.

The moment you release the RCU-sched read lock, the object can be dead
and gone.

[[email protected]: change subject line to be aligned with other patches]
Cc: Seth Jennings <[email protected]>
Cc: Josh Poimboeuf <[email protected]>
Cc: Masami Hiramatsu <[email protected]>
Cc: Miroslav Benes <[email protected]>
Cc: Petr Mladek <[email protected]>
Cc: Jiri Kosina <[email protected]>
Cc: "Paul E. McKenney" <[email protected]>
Cc: Rusty Russell <[email protected]>
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Reviewed-by: Masami Hiramatsu <[email protected]>
Acked-by: Paul E. McKenney <[email protected]>
Acked-by: Josh Poimboeuf <[email protected]>
Signed-off-by: Jiri Kosina <[email protected]>
10 years agoInput: sun4i-ts - add thermal driver dependency
Arnd Bergmann [Thu, 26 Feb 2015 01:26:58 +0000 (17:26 -0800)]
Input: sun4i-ts - add thermal driver dependency

The sun4i-ts driver has had a dependency on the thermal code
with the addition of the thermal zone sensor support, but this
is not currently enforced in Kconfig, so with TOUCHSCREEN_SUN4I=y,
THERMAL=m and THERMAL_OF=y we get

drivers/built-in.o: In function `sun4i_ts_remove':
:(.text+0x2376f4): undefined reference to `thermal_zone_of_sensor_unregister'
drivers/built-in.o: In function `sun4i_ts_probe':
:(.text+0x237a94): undefined reference to `thermal_zone_of_sensor_register'
:(.text+0x237c00): undefined reference to `thermal_zone_of_sensor_unregister'

We need the dependency on THERMAL in order to ensure that this
driver becomes a loadable module if the thermal support itself
is modular, while the dependency on THERMAL_OF is a runtime
dependency and the driver will still build if it is missing.
It is entirely possible to build sun4i-ts without THERMAL_OF
just to use the hwmon sensors and/or touchscreen.

Fixes: 223697107949 ("Input: sun4i-ts - add thermal zone sensor support")
Signed-off-by: Arnd Bergmann <[email protected]>
[[email protected]: Fix description and Kconfig dependencies]
Signed-off-by: Chen-Yu Tsai <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>
10 years agoNFSv4: Ensure we skip delegations that are already being returned
Trond Myklebust [Thu, 26 Feb 2015 19:05:05 +0000 (14:05 -0500)]
NFSv4: Ensure we skip delegations that are already being returned

In nfs_client_return_marked_delegations() and nfs_delegation_reap_unclaimed()
we want to optimise the loop traversal by skipping delegations that are
already in the process of being returned.

Signed-off-by: Trond Myklebust <[email protected]>
10 years agoNFSv4: Pin the superblock while we're returning the delegation
Trond Myklebust [Thu, 26 Feb 2015 14:57:34 +0000 (09:57 -0500)]
NFSv4: Pin the superblock while we're returning the delegation

This patch ensures that the superblock doesn't go ahead and disappear
underneath us while the state manager thread is returning delegations.

Signed-off-by: Trond Myklebust <[email protected]>
10 years agoNFSv4: Ensure we honour NFS_DELEGATION_RETURNING in nfs_inode_set_delegation()
Trond Myklebust [Fri, 27 Feb 2015 19:25:50 +0000 (14:25 -0500)]
NFSv4: Ensure we honour NFS_DELEGATION_RETURNING in nfs_inode_set_delegation()

Ensure that nfs_inode_set_delegation() doesn't inadvertently detach a
delegation that is already in the process of being returned.

Signed-off-by: Trond Myklebust <[email protected]>
10 years agoNFSv4: Ensure that we don't reap a delegation that is being returned
Trond Myklebust [Thu, 26 Feb 2015 18:59:38 +0000 (13:59 -0500)]
NFSv4: Ensure that we don't reap a delegation that is being returned

Signed-off-by: Trond Myklebust <[email protected]>
10 years agoNFS: Fix stateid used for NFS v4 closes
Anna Schumaker [Mon, 2 Mar 2015 21:46:09 +0000 (16:46 -0500)]
NFS: Fix stateid used for NFS v4 closes

After 566fcec60 the client uses the "current stateid" from the
nfs4_state structure to close a file.  This could potentially contain a
delegation stateid, which is disallowed by the protocol and causes
servers to return NFS4ERR_BAD_STATEID.  This patch restores the
(correct) behavior of sending the open stateid to close a file.

Reported-by: Olga Kornievskaia <[email protected]>
Fixes: 566fcec60 (NFSv4: Fix an atomicity problem in CLOSE)
Signed-off-by: Anna Schumaker <[email protected]>
Signed-off-by: Trond Myklebust <[email protected]>
10 years agoKVM: MIPS: Enable after disabling interrupt
Tapasweni Pathak [Sun, 22 Feb 2015 16:18:21 +0000 (21:48 +0530)]
KVM: MIPS: Enable after disabling interrupt

Enable disabled interrupt, on unsuccessful operation.

Found by Coccinelle.

Signed-off-by: Tapasweni Pathak <[email protected]>
Acked-by: Julia Lawall <[email protected]>
Reviewed-by: James Hogan <[email protected]>
Signed-off-by: Marcelo Tosatti <[email protected]>
10 years agoKVM: MIPS: Fix trace event to save PC directly
James Hogan [Tue, 24 Feb 2015 11:46:20 +0000 (11:46 +0000)]
KVM: MIPS: Fix trace event to save PC directly

Currently the guest exit trace event saves the VCPU pointer to the
structure, and the guest PC is retrieved by dereferencing it when the
event is printed rather than directly from the trace record. This isn't
safe as the printing may occur long afterwards, after the PC has changed
and potentially after the VCPU has been freed. Usually this results in
the same (wrong) PC being printed for multiple trace events. It also
isn't portable as userland has no way to access the VCPU data structure
when interpreting the trace record itself.

Lets save the actual PC in the structure so that the correct value is
accessible later.

Fixes: 669e846e6c4e ("KVM/MIPS32: MIPS arch specific APIs for KVM")
Signed-off-by: James Hogan <[email protected]>
Cc: Paolo Bonzini <[email protected]>
Cc: Ralf Baechle <[email protected]>
Cc: Marcelo Tosatti <[email protected]>
Cc: Gleb Natapov <[email protected]>
Cc: Steven Rostedt <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: [email protected]
Cc: [email protected]
Cc: <[email protected]> # v3.10+
Acked-by: Steven Rostedt <[email protected]>
Signed-off-by: Marcelo Tosatti <[email protected]>
10 years agoMerge tag 'gpio-v4.0-2' of git://git.kernel.org/pub/scm/linux/kernel/git/linusw/linux...
Linus Torvalds [Mon, 2 Mar 2015 22:13:39 +0000 (14:13 -0800)]
Merge tag 'gpio-v4.0-2' of git://git.kernel.org/pub/scm/linux/kernel/git/linusw/linux-gpio

Pull GPIO fixes from Linus Walleij:
 "Two GPIO fixes:

   - Fix a translation problem in of_get_named_gpiod_flags()

   - Fix a long standing container_of() mistake in the TPS65912 driver"

* tag 'gpio-v4.0-2' of git://git.kernel.org/pub/scm/linux/kernel/git/linusw/linux-gpio:
  gpio: tps65912: fix wrong container_of arguments
  gpiolib: of: allow of_gpiochip_find_and_xlate to find more than one chip per node

10 years agoMerge branch 'fixes-for-4.0-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Mon, 2 Mar 2015 22:08:10 +0000 (14:08 -0800)]
Merge branch 'fixes-for-4.0-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/evalenti/linux-soc-thermal

Pull thermal management fixes from Eduardo Valentin:
 "Specifics:

   - Several fixes in tmon tool.

   - Fixes in intel int340x for _ART and _TRT tables.

   - Add id for Avoton SoC into powerclamp driver.

   - Fixes in RCAR thermal driver to remove race conditions and fix fail
     path

   - Fixes in TI thermal driver: removal of unnecessary code and build
     fix if !CONFIG_PM_SLEEP

   - Cleanups in exynos thermal driver

   - Add stubs for include/linux/thermal.h.  Now drivers using thermal
     calls but that also work without CONFIG_THERMAL will be able to
     compile for systems that don't care about thermal.

  Note: I am sending this pull on Rui's behalf while he fixes issues in
  his Linux box"

* 'fixes-for-4.0-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/evalenti/linux-soc-thermal:
  thermal: int340x_thermal: Ignore missing _ART, _TRT tables
  thermal/intel_powerclamp: add id for Avoton SoC
  tools/thermal: tmon: silence 'set but not used' warnings
  tools/thermal: tmon: use pkg-config to determine library dependencies
  tools/thermal: tmon: support cross-compiling
  tools/thermal: tmon: add .gitignore
  tools/thermal: tmon: fixup tui windowing calculations
  tools/thermal: tmon: tui: don't hard-code dialog window size assumptions
  tools/thermal: tmon: add min/max macros
  tools/thermal: tmon: add --target-temp parameter
  thermal: exynos: Clean-up code to use oneline entry for exynos compatible table
  thermal: rcar: Make error and remove paths symmetrical with init
  thermal: rcar: Fix race condition between init and interrupt
  thermal: Introduce dummy functions when thermal is not defined
  ti-soc-thermal: Delete an unnecessary check before the function call "cpufreq_cooling_unregister"
  thermal: ti-soc-thermal: bandgap: Fix build warning if !CONFIG_PM_SLEEP

10 years agoBtrfs: incremental send, don't rename a directory too soon
Filipe Manana [Sat, 28 Feb 2015 22:29:22 +0000 (22:29 +0000)]
Btrfs: incremental send, don't rename a directory too soon

There's one more case where we can't issue a rename operation for a
directory as soon as we process it. We used to delay directory renames
only if they have some ancestor directory with a higher inode number
that got renamed too, but there's another case where we need to delay
the rename too - when a directory A is renamed to the old name of a
directory B but that directory B has its rename delayed because it
has now (in the send root) an ancestor with a higher inode number that
was renamed. If we don't delay the directory rename in this case, the
receiving end of the send stream will attempt to rename A to the old
name of B before B got renamed to its new name, which results in a
"directory not empty" error. So fix this by delaying directory renames
for this case too.

Steps to reproduce:

  $ mkfs.btrfs -f /dev/sdb
  $ mount /dev/sdb /mnt

  $ mkdir /mnt/a
  $ mkdir /mnt/b
  $ mkdir /mnt/c
  $ touch /mnt/a/file

  $ btrfs subvolume snapshot -r /mnt /mnt/snap1

  $ mv /mnt/c /mnt/x
  $ mv /mnt/a /mnt/x/y
  $ mv /mnt/b /mnt/a

  $ btrfs subvolume snapshot -r /mnt /mnt/snap2

  $ btrfs send /mnt/snap1 -f /tmp/1.send
  $ btrfs send -p /mnt/snap1 /mnt/snap2 -f /tmp/2.send

  $ mkfs.btrfs -f /dev/sdc
  $ mount /dev/sdc /mnt2
  $ btrfs receive /mnt2 -f /tmp/1.send
  $ btrfs receive /mnt2 -f /tmp/2.send
  ERROR: rename b -> a failed. Directory not empty

A test case for xfstests follows soon.

Reported-by: Ames Cornish <[email protected]>
Signed-off-by: Filipe Manana <[email protected]>
Signed-off-by: Chris Mason <[email protected]>
10 years agobtrfs: fix lost return value due to variable shadowing
David Sterba [Tue, 24 Feb 2015 17:57:18 +0000 (18:57 +0100)]
btrfs: fix lost return value due to variable shadowing

A block-local variable stores error code but btrfs_get_blocks_direct may
not return it in the end as there's a ret defined in the function scope.

CC: <[email protected]> # 3.6+
Fixes: d187663ef24c ("Btrfs: lock extents as we map them in DIO")
Signed-off-by: David Sterba <[email protected]>
Signed-off-by: Chris Mason <[email protected]>
10 years agoBtrfs: do not ignore errors from btrfs_lookup_xattr in do_setxattr
Filipe Manana [Mon, 23 Feb 2015 19:50:49 +0000 (19:50 +0000)]
Btrfs: do not ignore errors from btrfs_lookup_xattr in do_setxattr

The return value from btrfs_lookup_xattr() can be a pointer encoding an
error, therefore deal with it. This fixes commit 5f5bc6b1e2d5
("Btrfs: make xattr replace operations atomic").

Signed-off-by: Filipe Manana <[email protected]>
Signed-off-by: Chris Mason <[email protected]>
10 years agoBtrfs: fix off-by-one logic error in btrfs_realloc_node
Filipe Manana [Mon, 23 Feb 2015 19:48:52 +0000 (19:48 +0000)]
Btrfs: fix off-by-one logic error in btrfs_realloc_node

The end_slot variable actually matches the number of pointers in the
node and not the last slot (which is 'nritems - 1'). Therefore in order
to check that the current slot in the for loop doesn't match the last
one, the correct logic is to check if 'i' is less than 'end_slot - 1'
and not 'end_slot - 2'.

Fix this and set end_slot to be 'nritems - 1', as it's less confusing
since the variable name implies it's inclusive rather then exclusive.

Signed-off-by: Filipe Manana <[email protected]>
Signed-off-by: Chris Mason <[email protected]>
10 years agoBtrfs: add missing inode update when punching hole
Filipe Manana [Sun, 15 Feb 2015 22:38:54 +0000 (22:38 +0000)]
Btrfs: add missing inode update when punching hole

When punching a file hole if we endup only zeroing parts of a page,
because the start offset isn't a multiple of the sector size or the
start offset and length fall within the same page, we were not updating
the inode item. This prevented an fsync from doing anything, if no other
file changes happened in the current transaction, because the fields
in btrfs_inode used to check if the inode needs to be fsync'ed weren't
updated.

This issue is easy to reproduce and the following excerpt from the
xfstest case I made shows how to trigger it:

  _scratch_mkfs >> $seqres.full 2>&1
  _init_flakey
  _mount_flakey

  # Create our test file.
  $XFS_IO_PROG -f -c "pwrite -S 0x22 -b 16K 0 16K" \
      $SCRATCH_MNT/foo | _filter_xfs_io

  # Fsync the file, this makes btrfs update some btrfs inode specific fields
  # that are used to track if the inode needs to be written/updated to the fsync
  # log or not. After this fsync, the new values for those fields indicate that
  # a subsequent fsync does not need to touch the fsync log.
  $XFS_IO_PROG -c "fsync" $SCRATCH_MNT/foo

  # Force a commit of the current transaction. After this point, any operation
  # that modifies the data or metadata of our file, should update those fields in
  # the btrfs inode with values that make the next fsync operation write to the
  # fsync log.
  sync

  # Punch a hole in our file. This small range affects only 1 page.
  # This made the btrfs hole punching implementation write only some zeroes in
  # one page, but it did not update the btrfs inode fields used to determine if
  # the next fsync needs to write to the fsync log.
  $XFS_IO_PROG -c "fpunch 8000 4K" $SCRATCH_MNT/foo

  # Another variation of the previously mentioned case.
  $XFS_IO_PROG -c "fpunch 15000 100" $SCRATCH_MNT/foo

  # Now fsync the file. This was a no-operation because the previous hole punch
  # operation didn't update the inode's fields mentioned before, so they remained
  # with the values they had after the first fsync - that is, they indicate that
  # it is not needed to write to fsync log.
  $XFS_IO_PROG -c "fsync" $SCRATCH_MNT/foo

  echo "File content before:"
  od -t x1 $SCRATCH_MNT/foo

  # Simulate a crash/power loss.
  _load_flakey_table $FLAKEY_DROP_WRITES
  _unmount_flakey

  # Enable writes and mount the fs. This makes the fsync log replay code run.
  _load_flakey_table $FLAKEY_ALLOW_WRITES
  _mount_flakey

  # Because the last fsync didn't do anything, here the file content matched what
  # it was after the first fsync, before the holes were punched, and not what it
  # was after the holes were punched.
  echo "File content after:"
  od -t x1 $SCRATCH_MNT/foo

This issue has been around since 2012, when the punch hole implementation
was added, commit 2aaa66558172 ("Btrfs: add hole punching").

A test case for xfstests follows soon.

Signed-off-by: Filipe Manana <[email protected]>
Reviewed-by: Liu Bo <[email protected]>
Signed-off-by: Chris Mason <[email protected]>
10 years agoBtrfs: abort the transaction if we fail to update the free space cache inode
Josef Bacik [Thu, 12 Feb 2015 14:43:51 +0000 (09:43 -0500)]
Btrfs: abort the transaction if we fail to update the free space cache inode

Our gluster boxes were hitting a problem where they'd run out of space when
updating the block group cache and therefore wouldn't be able to update the free
space inode.  This is a problem because this is how we invalidate the cache and
protect ourselves from errors further down the stack, so if this fails we have
to abort the transaction so we make sure we don't end up with stale free space
cache.  Thanks,

Signed-off-by: Josef Bacik <[email protected]>
Signed-off-by: Chris Mason <[email protected]>
10 years agoBtrfs: fix fsync race leading to ordered extent memory leaks
Filipe Manana [Mon, 9 Feb 2015 17:17:43 +0000 (17:17 +0000)]
Btrfs: fix fsync race leading to ordered extent memory leaks

We can have multiple fsync operations against the same file during the
same transaction and they can collect the same ordered extents while they
don't complete (still accessible from the inode's ordered tree). If this
happens, those ordered extents will never get their reference counts
decremented to 0, leading to memory leaks and inode leaks (an iput for an
ordered extent's inode is scheduled only when the ordered extent's refcount
drops to 0). The following sequence diagram explains this race:

         CPU 1                                         CPU 2

btrfs_sync_file()

                                                 btrfs_sync_file()

  mutex_lock(inode->i_mutex)
  btrfs_log_inode()
    btrfs_get_logged_extents()
      --> collects ordered extent X
      --> increments ordered
          extent X's refcount
    btrfs_submit_logged_extents()
  mutex_unlock(inode->i_mutex)

                                                   mutex_lock(inode->i_mutex)
  btrfs_sync_log()
     btrfs_wait_logged_extents()
       --> list_del_init(&ordered->log_list)
                                                     btrfs_log_inode()
                                                       btrfs_get_logged_extents()
                                                         --> Adds ordered extent X
                                                             to logged_list because
                                                             at this point:
                                                             list_empty(&ordered->log_list)
                                                             && test_bit(BTRFS_ORDERED_LOGGED,
                                                                         &ordered->flags) == 0
                                                         --> Increments ordered extent
                                                             X's refcount
       --> check if ordered extent's io is
           finished or not, start it if
           necessary and wait for it to finish
       --> sets bit BTRFS_ORDERED_LOGGED
           on ordered extent X's flags
           and adds it to trans->ordered
  btrfs_sync_log() finishes

                                                       btrfs_submit_logged_extents()
                                                     btrfs_log_inode() finishes
                                                   mutex_unlock(inode->i_mutex)

btrfs_sync_file() finishes

                                                   btrfs_sync_log()
                                                      btrfs_wait_logged_extents()
                                                        --> Sees ordered extent X has the
                                                            bit BTRFS_ORDERED_LOGGED set in
                                                            its flags
                                                        --> X's refcount is untouched
                                                   btrfs_sync_log() finishes

                                                 btrfs_sync_file() finishes

btrfs_commit_transaction()
  --> called by transaction kthread for e.g.
  btrfs_wait_pending_ordered()
    --> waits for ordered extent X to
        complete
    --> decrements ordered extent X's
        refcount by 1 only, corresponding
        to the increment done by the fsync
        task ran by CPU 1

In the scenario of the above diagram, after the transaction commit,
the ordered extent will remain with a refcount of 1 forever, leaking
the ordered extent structure and preventing the i_count of its inode
from ever decreasing to 0, since the delayed iput is scheduled only
when the ordered extent's refcount drops to 0, preventing the inode
from ever being evicted by the VFS.

Fix this by using the flag BTRFS_ORDERED_LOGGED differently. Use it to
mean that an ordered extent is already being processed by an fsync call,
which will attach it to the current transaction, preventing it from being
collected by subsequent fsync operations against the same inode.

This race was introduced with the following change (added in 3.19 and
backported to stable 3.18 and 3.17):

  Btrfs: make sure logged extents complete in the current transaction V3
  commit 50d9aa99bd35c77200e0e3dd7a72274f8304701f

I ran into this issue while running xfstests/generic/113 in a loop, which
failed about 1 out of 10 runs with the following warning in dmesg:

[ 2612.440038] WARNING: CPU: 4 PID: 22057 at fs/btrfs/disk-io.c:3558 free_fs_root+0x36/0x133 [btrfs]()
[ 2612.442810] Modules linked in: btrfs crc32c_generic xor raid6_pq nfsd auth_rpcgss oid_registry nfs_acl nfs lockd grace fscache sunrpc loop processor parport_pc parport psmouse therma
l_sys i2c_piix4 serio_raw pcspkr evdev microcode button i2c_core ext4 crc16 jbd2 mbcache sd_mod sg sr_mod cdrom virtio_scsi ata_generic virtio_pci ata_piix virtio_ring libata virtio flo
ppy e1000 scsi_mod [last unloaded: btrfs]
[ 2612.452711] CPU: 4 PID: 22057 Comm: umount Tainted: G        W      3.19.0-rc5-btrfs-next-4+ #1
[ 2612.454921] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.7.5-0-ge51488c-20140602_164612-nilsson.home.kraxel.org 04/01/2014
[ 2612.457709]  0000000000000009 ffff8801342c3c78 ffffffff8142425e ffff88023ec8f2d8
[ 2612.459829]  0000000000000000 ffff8801342c3cb8 ffffffff81045308 ffff880046460000
[ 2612.461564]  ffffffffa036da56 ffff88003d07b000 ffff880046460000 ffff880046460068
[ 2612.463163] Call Trace:
[ 2612.463719]  [<ffffffff8142425e>] dump_stack+0x4c/0x65
[ 2612.464789]  [<ffffffff81045308>] warn_slowpath_common+0xa1/0xbb
[ 2612.466026]  [<ffffffffa036da56>] ? free_fs_root+0x36/0x133 [btrfs]
[ 2612.467247]  [<ffffffff810453c5>] warn_slowpath_null+0x1a/0x1c
[ 2612.468416]  [<ffffffffa036da56>] free_fs_root+0x36/0x133 [btrfs]
[ 2612.469625]  [<ffffffffa036f2a7>] btrfs_drop_and_free_fs_root+0x93/0x9b [btrfs]
[ 2612.471251]  [<ffffffffa036f353>] btrfs_free_fs_roots+0xa4/0xd6 [btrfs]
[ 2612.472536]  [<ffffffff8142612e>] ? wait_for_completion+0x24/0x26
[ 2612.473742]  [<ffffffffa0370bbc>] close_ctree+0x1f3/0x33c [btrfs]
[ 2612.475477]  [<ffffffff81059d1d>] ? destroy_workqueue+0x148/0x1ba
[ 2612.476695]  [<ffffffffa034e3da>] btrfs_put_super+0x19/0x1b [btrfs]
[ 2612.477911]  [<ffffffff81153e53>] generic_shutdown_super+0x73/0xef
[ 2612.479106]  [<ffffffff811540e2>] kill_anon_super+0x13/0x1e
[ 2612.480226]  [<ffffffffa034e1e3>] btrfs_kill_super+0x17/0x23 [btrfs]
[ 2612.481471]  [<ffffffff81154307>] deactivate_locked_super+0x3b/0x50
[ 2612.482686]  [<ffffffff811547a7>] deactivate_super+0x3f/0x43
[ 2612.483791]  [<ffffffff8116b3ed>] cleanup_mnt+0x59/0x78
[ 2612.484842]  [<ffffffff8116b44c>] __cleanup_mnt+0x12/0x14
[ 2612.485900]  [<ffffffff8105d019>] task_work_run+0x8f/0xbc
[ 2612.486960]  [<ffffffff810028d8>] do_notify_resume+0x5a/0x6b
[ 2612.488083]  [<ffffffff81236e5b>] ? trace_hardirqs_on_thunk+0x3a/0x3f
[ 2612.489333]  [<ffffffff8142a17f>] int_signal+0x12/0x17
[ 2612.490353] ---[ end trace 54a960a6bdcb8d93 ]---
[ 2612.557253] VFS: Busy inodes after unmount of sdb. Self-destruct in 5 seconds.  Have a nice day...

Kmemleak confirmed the ordered extent leak (and btrfs inode specific
structures such as delayed nodes):

$ cat /sys/kernel/debug/kmemleak
unreferenced object 0xffff880154290db0 (size 576):
  comm "btrfsck", pid 21980, jiffies 4295542503 (age 1273.412s)
  hex dump (first 32 bytes):
    01 40 00 00 01 00 00 00 b0 1d f1 4e 01 88 ff ff  [email protected]....
    00 00 00 00 00 00 00 00 c8 0d 29 54 01 88 ff ff  ..........)T....
  backtrace:
    [<ffffffff8141d74d>] kmemleak_update_trace+0x4c/0x6a
    [<ffffffff8122f2c0>] radix_tree_node_alloc+0x6d/0x83
    [<ffffffff8122fb26>] __radix_tree_create+0x109/0x190
    [<ffffffff8122fbdd>] radix_tree_insert+0x30/0xac
    [<ffffffffa03b9bde>] btrfs_get_or_create_delayed_node+0x130/0x187 [btrfs]
    [<ffffffffa03bb82d>] btrfs_delayed_delete_inode_ref+0x32/0xac [btrfs]
    [<ffffffffa0379dae>] __btrfs_unlink_inode+0xee/0x288 [btrfs]
    [<ffffffffa037c715>] btrfs_unlink_inode+0x1e/0x40 [btrfs]
    [<ffffffffa037c797>] btrfs_unlink+0x60/0x9b [btrfs]
    [<ffffffff8115d7f0>] vfs_unlink+0x9c/0xed
    [<ffffffff8115f5de>] do_unlinkat+0x12c/0x1fa
    [<ffffffff811601a7>] SyS_unlinkat+0x29/0x2b
    [<ffffffff81429e92>] system_call_fastpath+0x12/0x17
    [<ffffffffffffffff>] 0xffffffffffffffff
unreferenced object 0xffff88014ef11db0 (size 576):
  comm "rm", pid 22009, jiffies 4295542593 (age 1273.052s)
  hex dump (first 32 bytes):
    02 00 00 00 01 00 00 00 00 00 00 00 00 00 00 00  ................
    00 00 00 00 00 00 00 00 c8 1d f1 4e 01 88 ff ff  ...........N....
  backtrace:
    [<ffffffff8141d74d>] kmemleak_update_trace+0x4c/0x6a
    [<ffffffff8122f2c0>] radix_tree_node_alloc+0x6d/0x83
    [<ffffffff8122fb26>] __radix_tree_create+0x109/0x190
    [<ffffffff8122fbdd>] radix_tree_insert+0x30/0xac
    [<ffffffffa03b9bde>] btrfs_get_or_create_delayed_node+0x130/0x187 [btrfs]
    [<ffffffffa03bb82d>] btrfs_delayed_delete_inode_ref+0x32/0xac [btrfs]
    [<ffffffffa0379dae>] __btrfs_unlink_inode+0xee/0x288 [btrfs]
    [<ffffffffa037c715>] btrfs_unlink_inode+0x1e/0x40 [btrfs]
    [<ffffffffa037c797>] btrfs_unlink+0x60/0x9b [btrfs]
    [<ffffffff8115d7f0>] vfs_unlink+0x9c/0xed
    [<ffffffff8115f5de>] do_unlinkat+0x12c/0x1fa
    [<ffffffff811601a7>] SyS_unlinkat+0x29/0x2b
    [<ffffffff81429e92>] system_call_fastpath+0x12/0x17
    [<ffffffffffffffff>] 0xffffffffffffffff
unreferenced object 0xffff8800336feda8 (size 584):
  comm "aio-stress", pid 22031, jiffies 4295543006 (age 1271.400s)
  hex dump (first 32 bytes):
    00 40 3e 00 00 00 00 00 00 00 8f 42 00 00 00 00  .@>........B....
    00 00 01 00 00 00 00 00 00 00 01 00 00 00 00 00  ................
  backtrace:
    [<ffffffff8114eb34>] create_object+0x172/0x29a
    [<ffffffff8141d790>] kmemleak_alloc+0x25/0x41
    [<ffffffff81141ae6>] kmemleak_alloc_recursive.constprop.52+0x16/0x18
    [<ffffffff81145288>] kmem_cache_alloc+0xf7/0x198
    [<ffffffffa0389243>] __btrfs_add_ordered_extent+0x43/0x309 [btrfs]
    [<ffffffffa038968b>] btrfs_add_ordered_extent_dio+0x12/0x14 [btrfs]
    [<ffffffffa03810e2>] btrfs_get_blocks_direct+0x3ef/0x571 [btrfs]
    [<ffffffff81181349>] do_blockdev_direct_IO+0x62a/0xb47
    [<ffffffff8118189a>] __blockdev_direct_IO+0x34/0x36
    [<ffffffffa03776e5>] btrfs_direct_IO+0x16a/0x1e8 [btrfs]
    [<ffffffff81100373>] generic_file_direct_write+0xb8/0x12d
    [<ffffffffa038615c>] btrfs_file_write_iter+0x24b/0x42f [btrfs]
    [<ffffffff8118bb0d>] aio_run_iocb+0x2b7/0x32e
    [<ffffffff8118c99a>] do_io_submit+0x26e/0x2ff
    [<ffffffff8118ca3b>] SyS_io_submit+0x10/0x12
    [<ffffffff81429e92>] system_call_fastpath+0x12/0x17

CC: <[email protected]> # 3.19, 3.18 and 3.17
Signed-off-by: Filipe Manana <[email protected]>
Signed-off-by: Chris Mason <[email protected]>
10 years agoKVM: SVM: fix interrupt injection (apic->isr_count always 0)
Radim Krčmář [Fri, 27 Feb 2015 15:32:38 +0000 (16:32 +0100)]
KVM: SVM: fix interrupt injection (apic->isr_count always 0)

In commit b4eef9b36db4, we started to use hwapic_isr_update() != NULL
instead of kvm_apic_vid_enabled(vcpu->kvm).  This didn't work because
SVM had it defined and "apicv" path in apic_{set,clear}_isr() does not
change apic->isr_count, because it should always be 1.  The initial
value of apic->isr_count was based on kvm_apic_vid_enabled(vcpu->kvm),
which is always 0 for SVM, so KVM could have injected interrupts when it
shouldn't.

Fix it by implicitly setting SVM's hwapic_isr_update to NULL and make the
initial isr_count depend on hwapic_isr_update() for good measure.

Fixes: b4eef9b36db4 ("kvm: x86: vmx: NULL out hwapic_isr_update() in case of !enable_apicv")
Reported-and-tested-by: Borislav Petkov <[email protected]>
Signed-off-by: Radim Krčmář <[email protected]>
Signed-off-by: Marcelo Tosatti <[email protected]>
10 years agoMerge tag 'md/4.0-fixes' of git://neil.brown.name/md
Linus Torvalds [Mon, 2 Mar 2015 22:03:27 +0000 (14:03 -0800)]
Merge tag 'md/4.0-fixes' of git://neil.brown.name/md

Pull md fixes from Neil Brown:
 "Three md fixes:

   - fix a read-balance problem that was reported 2 years ago, but that
     I never noticed the report :-(

   - fix for rare RAID6 problem causing incorrect bitmap updates when
     two devices fail.

   - add __ATTR_PREALLOC annotation now that it is possible"

* tag 'md/4.0-fixes' of git://neil.brown.name/md:
  md: mark some attributes as pre-alloc
  raid5: check faulty flag for array status during recovery.
  md/raid1: fix read balance when a drive is write-mostly.

10 years agoMerge tag 'metag-fixes-v4.0-1' of git://git.kernel.org/pub/scm/linux/kernel/git/jhoga...
Linus Torvalds [Mon, 2 Mar 2015 22:02:17 +0000 (14:02 -0800)]
Merge tag 'metag-fixes-v4.0-1' of git://git.kernel.org/pub/scm/linux/kernel/git/jhogan/metag

Pull arch/metag fix from James Hogan:
 "This is just a single patch to fix the KSTK_EIP() and KSTK_ESP()
  macros for metag which have always been erronously returning the PC
  and stack pointer of the task's kernel context rather than from its
  user context saved at entry from userland into the kernel, which
  affects the contents of /proc/<pid>/maps and /proc/<pid>/stat"

* tag 'metag-fixes-v4.0-1' of git://git.kernel.org/pub/scm/linux/kernel/git/jhogan/metag:
  metag: Fix KSTK_EIP() and KSTK_ESP() macros

10 years agoInput: cyapa - remove superfluous type check in cyapa_gen5_read_idac_data()
Geert Uytterhoeven [Mon, 2 Mar 2015 17:39:24 +0000 (09:39 -0800)]
Input: cyapa - remove superfluous type check in cyapa_gen5_read_idac_data()

drivers/input/mouse/cyapa_gen5.c: In function ‘cyapa_gen5_read_idac_data’:
drivers/input/mouse/cyapa_gen5.c:1876: warning: ‘max_element_cnt’ may be used uninitialized in this function
drivers/input/mouse/cyapa_gen5.c:1873: warning: ‘offset’ may be used uninitialized in this function

If *data_size is non-zero, and idac_data_type contains an unknown type,
max_element_cnt and offset will be uninitialized, and the loop will
process non-existing data.

However, this cannot happen (for now), as there's a test for unknown
types at the top of cyapa_gen5_read_idac_data().

As no "if ... else if ..." is used in other places, remove the
superfluous "if" to silence the compiler warning.

Signed-off-by: Geert Uytterhoeven <[email protected]>
Acked-by: Dudley Du <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>
10 years agoInput: cyapa - fix unaligned functions redefinition error
Dudley Du [Mon, 2 Mar 2015 17:37:59 +0000 (09:37 -0800)]
Input: cyapa - fix unaligned functions redefinition error

Use asm/unaligned.h instead of linux/unaligned/access_ok.h header file to
fix compiling issues such as following while doing cross platform
compiling:

"include/linux/unaligned/access_ok.h:7:19: error: redefinition of
 'get_unaligned_le16'
...
include/linux/unaligned/le_struct.h:6:19: note: previous definition of
 'get_unaligned_le16' was here".

Reported-by: kbuild test robot <[email protected]>
Signed-off-by: Dudley Du <[email protected]>
Signed-off-by: Dmitry Torokhov <[email protected]>
10 years agocpuidle: Clean up fallback handling in cpuidle_idle_call()
Rafael J. Wysocki [Mon, 2 Mar 2015 21:25:37 +0000 (22:25 +0100)]
cpuidle: Clean up fallback handling in cpuidle_idle_call()

Move the fallback code path in cpuidle_idle_call() to the end of the
function to avoid jumping to a label in an if () branch.

Signed-off-by: Rafael J. Wysocki <[email protected]>
10 years agoMerge branch 'mlx4'
David S. Miller [Mon, 2 Mar 2015 20:27:23 +0000 (15:27 -0500)]
Merge branch 'mlx4'

Or Gerlitz says:

====================
Mellanox driver fixes

Two small fixes, please apply to net.

Both patches should go to 3.19.y too.
====================

Signed-off-by: David S. Miller <[email protected]>
10 years agonet/mlx4_en: Disbale GRO for incoming loopback/selftest packets
Ido Shamay [Mon, 2 Mar 2015 16:22:16 +0000 (18:22 +0200)]
net/mlx4_en: Disbale GRO for incoming loopback/selftest packets

Packets which are sent from the selftest (ethtool) flow,
should not be passed to GRO stack but rather dropped by
the driver after validation. To achieve that, we disable
GRO for the duration of the selftest.

Fixes: dd65beac48a5 ("net/mlx4_en: Extend usage of napi_gro_frags")
Reported-by: Carol Soto <[email protected]>
Signed-off-by: Ido Shamay <[email protected]>
Signed-off-by: Or Gerlitz <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
10 years agonet/mlx4_core: Fix wrong mask and error flow for the update-qp command
Or Gerlitz [Mon, 2 Mar 2015 16:22:15 +0000 (18:22 +0200)]
net/mlx4_core: Fix wrong mask and error flow for the update-qp command

The bit mask for currently supported driver features (MLX4_UPDATE_QP_SUPPORTED_ATTRS)
of the update-qp command was defined twice (using enum value and pre-processor
define directive) and wrong.

The return value of the call to mlx4_update_qp() from within the SRIOV
resource-tracker was wrongly voided down.

Fix both issues.

issue: none
Fixes: 09e05c3f78e9 ('net/mlx4: Set vlan stripping policy by the right command')
Fixes: ce8d9e0d6746 ('net/mlx4_core: Add UPDATE_QP SRIOV wrapper support')
Signed-off-by: Matan Barak <[email protected]>
Signed-off-by: Or Gerlitz <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
10 years agos390/mm: fix incorrect ASCE after crst_table_downgrade
Martin Schwidefsky [Sun, 1 Mar 2015 14:56:45 +0000 (06:56 -0800)]
s390/mm: fix incorrect ASCE after crst_table_downgrade

The switch_mm function does nothing in case the prev and next mm
are the same. It can happen that a crst_table_downgrade has changed
the top-level pgd in the meantime on a different CPU. Always store
the new ASCE to be picked up in entry.S.

[[email protected]]: Bug was introduced with git commit
53e857f30867 ("s390/mm,tlb: race of lazy TLB flush vs. recreation
of TLB entries") and causes random crashes due to broken page tables
being used.

Reported-by: Dominik Vogt <[email protected]>
Signed-off-by: Martin Schwidefsky <[email protected]>
Signed-off-by: Heiko Carstens <[email protected]>
10 years agos390/ftrace: fix crashes when switching tracers / add notrace to cpu_relax()
Heiko Carstens [Sat, 28 Feb 2015 10:35:26 +0000 (11:35 +0100)]
s390/ftrace: fix crashes when switching tracers / add notrace to cpu_relax()

With git commit 4d92f50249eb ("s390: reintroduce diag 44 calls for
cpu_relax()") I reintroduced a non-trivial cpu_relax() variant on s390.

The difference to the previous variant however is that the new version is
an out-of-line function, which will be traced if function tracing is enabled.

Switching to different tracers includes instruction patching. Therefore this
is done within stop_machine() "context" to prevent that any function tracing
is going on while instructions are being patched.
With the new out-of-line variant of cpu_relax() this is not true anymore,
since cpu_relax() gets called in a busy loop by all waiting cpus within
stop_machine() until function patching is finished.
Therefore cpu_relax() must be marked notrace.

This fixes kernel crashes when frequently switching between "function" and
"function_graph" tracers.

Moving cpu_relax() to a header file again, doesn't work because of header
include order dependencies.

Signed-off-by: Heiko Carstens <[email protected]>
Signed-off-by: Martin Schwidefsky <[email protected]>
10 years agosata-fsl: Apply link speed limits
Martin Hicks [Thu, 19 Feb 2015 20:05:47 +0000 (15:05 -0500)]
sata-fsl: Apply link speed limits

The driver was ignoring limits requested by libata.force.  The output
would look like:

fsl-sata ffe18000.sata: Sata FSL Platform/CSB Driver init
ata1: FORCE: PHY spd limit set to 1.5Gbps
ata1: SATA max UDMA/133 irq 74
ata1: Signature Update detected @ 0 msecs
ata1: SATA link up 3.0 Gbps (SStatus 123 SControl 310)

Signed-off-by: Martin Hicks <[email protected]>
Signed-off-by: Tejun Heo <[email protected]>
10 years agocpuset: Fix cpuset sched_relax_domain_level
Jason Low [Fri, 13 Feb 2015 03:58:07 +0000 (11:58 +0800)]
cpuset: Fix cpuset sched_relax_domain_level

The cpuset.sched_relax_domain_level can control how far we do
immediate load balancing on a system. However, it was found on recent
kernels that echo'ing a value into cpuset.sched_relax_domain_level
did not reduce any immediate load balancing.

The reason this occurred was because the update_domain_attr_tree() traversal
did not update for the "top_cpuset". This resulted in nothing being changed
when modifying the sched_relax_domain_level parameter.

This patch is able to address that problem by having update_domain_attr_tree()
allow updates for the root in the cpuset traversal.

Fixes: fc560a26acce ("cpuset: replace cpuset->stack_list with cpuset_for_each_descendant_pre()")
Cc: <[email protected]> # 3.9+
Signed-off-by: Jason Low <[email protected]>
Signed-off-by: Zefan Li <[email protected]>
Signed-off-by: Tejun Heo <[email protected]>
Tested-by: Serge Hallyn <[email protected]>
10 years agocpuset: fix a warning when clearing configured masks in old hierarchy
Zefan Li [Fri, 13 Feb 2015 03:20:30 +0000 (11:20 +0800)]
cpuset: fix a warning when clearing configured masks in old hierarchy

When we clear cpuset.cpus, cpuset.effective_cpus won't be cleared:

  # mount -t cgroup -o cpuset xxx /mnt
  # mkdir /mnt/tmp
  # echo 0 > /mnt/tmp/cpuset.cpus
  # echo > /mnt/tmp/cpuset.cpus
  # cat cpuset.cpus

  # cat cpuset.effective_cpus
  0-15

And a kernel warning in update_cpumasks_hier() is triggered:

 ------------[ cut here ]------------
 WARNING: CPU: 0 PID: 4028 at kernel/cpuset.c:894 update_cpumasks_hier+0x471/0x650()

Cc: <[email protected]> # 3.17+
Signed-off-by: Zefan Li <[email protected]>
Signed-off-by: Tejun Heo <[email protected]>
Tested-by: Serge Hallyn <[email protected]>
10 years agocpuset: initialize effective masks when clone_children is enabled
Zefan Li [Fri, 13 Feb 2015 03:19:49 +0000 (11:19 +0800)]
cpuset: initialize effective masks when clone_children is enabled

If clone_children is enabled, effective masks won't be initialized
due to the bug:

  # mount -t cgroup -o cpuset xxx /mnt
  # echo 1 > cgroup.clone_children
  # mkdir /mnt/tmp
  # cat /mnt/tmp/
  # cat cpuset.effective_cpus

  # cat cpuset.cpus
  0-15

And then this cpuset won't constrain the tasks in it.

Either the bug or the fix has no effect on unified hierarchy, as
there's no clone_chidren flag there any more.

Reported-by: Christian Brauner <[email protected]>
Reported-by: Serge Hallyn <[email protected]>
Cc: <[email protected]> # 3.17+
Signed-off-by: Zefan Li <[email protected]>
Signed-off-by: Tejun Heo <[email protected]>
Tested-by: Serge Hallyn <[email protected]>
10 years agodmaengine: shdma: Move DMA stop to (runtime) suspend callbacks
Geert Uytterhoeven [Thu, 26 Feb 2015 10:26:34 +0000 (11:26 +0100)]
dmaengine: shdma: Move DMA stop to (runtime) suspend callbacks

During system reboot, the sh-dma-engine device may be runtime-suspended,
causing a crash:

    Unhandled fault: imprecise external abort (0x1406) at 0x0002c02c
    Internal error: : 1406 [#1] SMP ARM
    ...
    PC is at sh_dmae_ctl_stop+0x28/0x64
    LR is at sh_dmae_ctl_stop+0x24/0x64

If the sh-dma-engine is runtime-suspended, its module clock is turned
off, and its registers cannot be accessed.

To fix this, move the call to sh_dmae_ctl_stop(), which touches the
DMAOR register, to the sh_dmae_suspend() and sh_dmae_runtime_suspend()
callbacks.  This makes PM operations more symmetric, as both
sh_dmae_resume() and sh_dmae_runtime_resume() already call sh_dmae_rst()
to re-initialize the DMAOR register.

Remove sh_dmae_shutdown(), as it became empty.

Signed-off-by: Geert Uytterhoeven <[email protected]>
Reviewed-by: Ulf Hansson <[email protected]>
Signed-off-by: Vinod Koul <[email protected]>
10 years agoASoC: sam9g20_wm8731: drop machine_is_xxx
Alexandre Belloni [Tue, 6 Jan 2015 11:14:32 +0000 (12:14 +0100)]
ASoC: sam9g20_wm8731: drop machine_is_xxx

Atmel based boards can now only be used with device tree. Drop non DT
initialization.

Signed-off-by: Alexandre Belloni <[email protected]>
Signed-off-by: Mark Brown <[email protected]>
10 years agoserial: add device tree binding documentation for ETRAX FS UART
Niklas Cassel [Wed, 18 Feb 2015 22:29:38 +0000 (23:29 +0100)]
serial: add device tree binding documentation for ETRAX FS UART

Signed-off-by: Niklas Cassel <[email protected]>
Signed-off-by: Rob Herring <[email protected]>
10 years agoof/overlay: Directly include idr.h
Mark Brown [Tue, 17 Feb 2015 02:36:58 +0000 (11:36 +0900)]
of/overlay: Directly include idr.h

The overlay code uses IDRs but does not explicitly include the header
providing the interface, instead relying on an implicit inclusion. Make
the dependency explicit to avoid potential future build issues if the
implicit inclusion goes away.

Signed-off-by: Mark Brown <[email protected]>
Signed-off-by: Rob Herring <[email protected]>
10 years agoof: Drop superfluous dependance for OF_OVERLAY
Matwey V. Kornilov [Mon, 16 Feb 2015 06:27:24 +0000 (09:27 +0300)]
of: Drop superfluous dependance for OF_OVERLAY

The whole menu already depends on OF, so there is no need to additionaly specify it.

Suggested-by: Paul Bolle <[email protected]>
Signed-off-by: Matwey V. Kornilov <[email protected]>
Signed-off-by: Rob Herring <[email protected]>
10 years agoof: Add vendor prefix for Arasan
Michal Simek [Thu, 12 Feb 2015 08:38:40 +0000 (09:38 +0100)]
of: Add vendor prefix for Arasan

PATA(pata_arasan_cf.c) and SDHCI(sdhci-of-arasan.c) drivers
are already using this prefix.

Signed-off-by: Michal Simek <[email protected]>
Signed-off-by: Rob Herring <[email protected]>
10 years agoof: Add prompt for OF_OVERLAY config
Matwey V. Kornilov [Sun, 15 Feb 2015 15:22:47 +0000 (18:22 +0300)]
of: Add prompt for OF_OVERLAY config

Signed-off-by: Matwey V. Kornilov <[email protected]>
Signed-off-by: Rob Herring <[email protected]>
10 years agoMerge tag 'efi-urgent' of git://git.kernel.org/pub/scm/linux/kernel/git/mfleming...
Ingo Molnar [Mon, 2 Mar 2015 13:18:57 +0000 (14:18 +0100)]
Merge tag 'efi-urgent' of git://git.kernel.org/pub/scm/linux/kernel/git/mfleming/efi into x86/urgent

Pull EFI fixes from Matt Fleming:

" - Fix regression in DMI sysfs code for handling "End of Table" entry
    and a type bug that could lead to integer overflow. (Ivan Khoronzhuk)

  - Fix boundary checking in efi_high_alloc() which can lead to memory
    corruption in the EFI boot stubs. (Yinghai Lu)"

Signed-off-by: Ingo Molnar <[email protected]>
10 years agoxen: Remove trailing semicolon from xenbus_register_frontend() definition
Yuval Shaia [Sun, 1 Mar 2015 10:57:38 +0000 (02:57 -0800)]
xen: Remove trailing semicolon from xenbus_register_frontend() definition

Signed-off-by: Yuval Shaia <[email protected]>
Signed-off-by: David Vrabel <[email protected]>
10 years agoMAINTAINERS: Add entry for SAMSUNG THERMAL DRIVER
Lukasz Majewski [Tue, 24 Feb 2015 15:31:28 +0000 (16:31 +0100)]
MAINTAINERS: Add entry for SAMSUNG THERMAL DRIVER

This patch adds entry for SAMSUNG THERMAL DRIVER in the MAINTAINERS file.
It has been agreed, that pull request are going to be sent to Eduardo
Valentin.

Signed-off-by: Lukasz Majewski <[email protected]>
Acked-by: Eduardo Valentin <[email protected]>
10 years agocpufreq: exynos: Use simple approach to asses if cpu cooling can be used
Lukasz Majewski [Thu, 5 Feb 2015 15:45:06 +0000 (16:45 +0100)]
cpufreq: exynos: Use simple approach to asses if cpu cooling can be used

Commit: e725d26c4857e5e41975b5e74e64ce6ab09a7121 provided possibility to
use device tree to asses if cpu can be used as cooling device. Since the
code was somewhat awkward, simpler approach has been proposed.

Test HW: Exynos 4412 - Odroid U3.

Suggested-by: Viresh Kumar <[email protected]>
Signed-off-by: Lukasz Majewski <[email protected]>
Acked-by: Viresh Kumar <[email protected]>
10 years agothermal: exynos: Fix wrong control of power down detection mode for Exynos7
Chanwoo Choi [Tue, 24 Feb 2015 04:56:54 +0000 (13:56 +0900)]
thermal: exynos: Fix wrong control of power down detection mode for Exynos7

This patch fixes the wrong control of PD_DET_EN (power down detection mode)
for Exynos7 because exynos7_tmu_control() always enables the power down detection
mode regardless 'on' parameter.

Cc: Zhang Rui <[email protected]>
Cc: Eduardo Valentin <[email protected]>
Signed-off-by: Chanwoo Choi <[email protected]>
Tested-by: Abhilash Kesavan <[email protected]>
Signed-off-by: Lukasz Majewski <[email protected]>
10 years agoUSB: ch341: set tty baud speed according to tty struct
Nicolas PLANEL [Sun, 1 Mar 2015 18:47:22 +0000 (13:47 -0500)]
USB: ch341: set tty baud speed according to tty struct

The ch341_set_baudrate() function initialize the device baud speed
according to the value on priv->baud_rate. By default the ch341_open() set
it to a hardcoded value (DEFAULT_BAUD_RATE 9600). Unfortunately, the
tty_struct is not initialized with the same default value. (usually 56700)

This means that the tty_struct and the device baud rate generator are not
synchronized after opening the port.

Fixup is done by calling ch341_set_termios() if tty exist.
Remove unnecessary variable priv->baud_rate setup as it's already done by
ch341_port_probe().
Remove unnecessary call to ch341_set_{handshake,baudrate}() in
ch341_open() as there already called in ch341_configure() and
ch341_set_termios()

Signed-off-by: Nicolas PLANEL <[email protected]>
Signed-off-by: Johan Hovold <[email protected]>
10 years agoNFSv4: Don't call put_rpccred() under the rcu_read_lock()
Trond Myklebust [Thu, 26 Feb 2015 17:54:46 +0000 (12:54 -0500)]
NFSv4: Don't call put_rpccred() under the rcu_read_lock()

put_rpccred() can sleep.

Fixes: 8f649c3762547 ("NFSv4: Fix the locking in nfs_inode_reclaim_delegation()")
Cc: [email protected] # 2.6.35+
Signed-off-by: Trond Myklebust <[email protected]>
10 years agoNFS: Don't require a filehandle to refresh the inode in nfs_prime_dcache()
Trond Myklebust [Mon, 23 Feb 2015 23:51:32 +0000 (18:51 -0500)]
NFS: Don't require a filehandle to refresh the inode in nfs_prime_dcache()

If the server does not return a valid set of attributes that we can
use to either create a file or refresh the inode, then there is no
value in calling nfs_prime_dcache().

However if we're just refreshing the inode using the attributes that
the server returned, then it shouldn't matter whether or not we have
a filehandle, as long as we check the fsid+fileid combination.

Signed-off-by: Trond Myklebust <[email protected]>
10 years agoNFSv3: Use the readdir fileid as the mounted-on-fileid
Trond Myklebust [Mon, 23 Feb 2015 21:15:00 +0000 (16:15 -0500)]
NFSv3: Use the readdir fileid as the mounted-on-fileid

When we call readdirplus, set the fileid normally returned by readdir
as the mounted-on-fileid, since that is commonly the case if there is
a mountpoint. To ensure that we get it right, we only set the flag if
the readdir fileid differs from the one returned in the readdirplus
attributes.

This again means that we can avoid the issues described in commit
2ef47eb1aee17 ("NFS: Fix use of nfs_attr_use_mounted_on_fileid()"),
which only fixed NFSv4.

Signed-off-by: Trond Myklebust <[email protected]>
10 years agoNFS: Don't invalidate a submounted dentry in nfs_prime_dcache()
Trond Myklebust [Sun, 22 Feb 2015 21:35:36 +0000 (16:35 -0500)]
NFS: Don't invalidate a submounted dentry in nfs_prime_dcache()

If we're traversing a directory which contains a submounted filesystem,
or one that has a referral, the NFS server that is processing the READDIR
request will often return information for the underlying (mounted-on)
directory. It may, or may not, also return filehandle information.

If this happens, and the lookup in nfs_prime_dcache() returns the
dentry for the submounted directory, the filehandle comparison will
fail, and we call d_invalidate(). Post-commit 8ed936b5671bf
("vfs: Lazily remove mounts on unlinked files and directories."), this
means the entire subtree is unmounted.

The following minimal patch addresses this problem by punting on
the invalidation if there is a submount.

Kudos to Neil Brown <[email protected]> for having tracked down this
issue (see link).

Reported-by: Nix <[email protected]>
Link: http://lkml.kernel.org/r/[email protected]
Cc: [email protected] # 3.18+
Signed-off-by: Trond Myklebust <[email protected]>
10 years agoNFSv4: Set a barrier in the update_changeattr() helper
Trond Myklebust [Fri, 27 Feb 2015 00:52:06 +0000 (19:52 -0500)]
NFSv4: Set a barrier in the update_changeattr() helper

Ensure that we don't regress the changes that were made to the
directory.

Signed-off-by: Trond Myklebust <[email protected]>
Tested-by: Chuck Lever <[email protected]>
10 years agoNFS: Fix nfs_post_op_update_inode() to set an attribute barrier
Trond Myklebust [Fri, 27 Feb 2015 00:48:26 +0000 (19:48 -0500)]
NFS: Fix nfs_post_op_update_inode() to set an attribute barrier

nfs_post_op_update_inode() is called after a self-induced attribute
update. Ensure that it also sets the barrier.

Signed-off-by: Trond Myklebust <[email protected]>
Tested-by: Chuck Lever <[email protected]>
10 years agoNFS: Remove size hack in nfs_inode_attrs_need_update()
Trond Myklebust [Thu, 26 Feb 2015 22:57:14 +0000 (17:57 -0500)]
NFS: Remove size hack in nfs_inode_attrs_need_update()

Prior to this patch, we used to always OK attribute updates that extended
the file size on the assumption that we might be performing writeback.
Now that we have attribute barriers to protect the writeback related updates,
we should remove this hack, as it can cause truncate() operations to
apparently be reverted if/when a readahead or getattr RPC call races
with our on-the-wire SETATTR.

Signed-off-by: Trond Myklebust <[email protected]>
Tested-by: Chuck Lever <[email protected]>
10 years agoNFSv4: Add attribute update barriers to delegreturn and pNFS layoutcommit
Trond Myklebust [Thu, 26 Feb 2015 22:54:58 +0000 (17:54 -0500)]
NFSv4: Add attribute update barriers to delegreturn and pNFS layoutcommit

Ensure that other operations that race with delegreturn and layoutcommit
cannot revert the attribute updates that were made on the server.

Signed-off-by: Trond Myklebust <[email protected]>
Tested-by: Chuck Lever <[email protected]>
10 years agoNFS: Add attribute update barriers to NFS writebacks
Trond Myklebust [Thu, 26 Feb 2015 22:36:09 +0000 (17:36 -0500)]
NFS: Add attribute update barriers to NFS writebacks

Ensure that other operations that race with our write RPC calls
cannot revert the file size updates that were made on the server.

Signed-off-by: Trond Myklebust <[email protected]>
Tested-by: Chuck Lever <[email protected]>
10 years agoNFS: Set an attribute barrier on all updates
Trond Myklebust [Fri, 27 Feb 2015 00:34:32 +0000 (19:34 -0500)]
NFS: Set an attribute barrier on all updates

Ensure that we update the attribute barrier even if there were no
invalidations, provided that this value is newer than the old one.

Signed-off-by: Trond Myklebust <[email protected]>
Tested-by: Chuck Lever <[email protected]>
10 years agoNFS: Add attribute update barriers to nfs_setattr_update_inode()
Trond Myklebust [Thu, 26 Feb 2015 21:09:04 +0000 (16:09 -0500)]
NFS: Add attribute update barriers to nfs_setattr_update_inode()

Ensure that other operations which raced with our setattr RPC call
cannot revert the file attribute changes that were made on the server.
To do so, we artificially bump the attribute generation counter on
the inode so that all calls to nfs_fattr_init() that precede ours
will be dropped.

The motivation for the patch came from Chuck Lever's reports of readaheads
racing with truncate operations and causing the file size to be reverted.

Reported-by: Chuck Lever <[email protected]>
Signed-off-by: Trond Myklebust <[email protected]>
Tested-by: Chuck Lever <[email protected]>
10 years agoNFS: Add a helper to set attribute barriers
Trond Myklebust [Thu, 26 Feb 2015 22:42:42 +0000 (17:42 -0500)]
NFS: Add a helper to set attribute barriers

Signed-off-by: Trond Myklebust <[email protected]>
Tested-by: Chuck Lever <[email protected]>
10 years agoNFS: Ensure that buffered writes wait for O_DIRECT writes to complete
Trond Myklebust [Sat, 28 Feb 2015 03:54:19 +0000 (22:54 -0500)]
NFS: Ensure that buffered writes wait for O_DIRECT writes to complete

The O_DIRECT code will grab the inode->i_mutex and flush out buffered
writes, before scheduling a read or a write. However there is no
equivalent in the buffered write code to wait for O_DIRECT to complete.

Fixes a reported issue in xfstests generic/133, when first performing an
O_DIRECT write followed by a buffered write.

Signed-off-by: Trond Myklebust <[email protected]>
Tested-by: Chuck Lever <[email protected]>
10 years agomei: make device disabled on stop unconditionally
Alexander Usyskin [Tue, 10 Feb 2015 08:36:36 +0000 (10:36 +0200)]
mei: make device disabled on stop unconditionally

Set the internal device state to to disabled after hardware reset in stop flow.
This will cover cases when driver was not brought to disabled state because of
an error and in stop flow we wish not to retry the reset.

Cc: <[email protected]> #3.10+
Signed-off-by: Alexander Usyskin <[email protected]>
Signed-off-by: Tomas Winkler <[email protected]>
Signed-off-by: Greg Kroah-Hartman <[email protected]>
10 years agostaging: comedi: adv_pci1710: fix AI INSN_READ for non-zero channel
Ian Abbott [Fri, 27 Feb 2015 16:04:42 +0000 (16:04 +0000)]
staging: comedi: adv_pci1710: fix AI INSN_READ for non-zero channel

Reading of analog input channels by the `INSN_READ` comedi instruction
is broken for all except channel 0.  `pci171x_ai_insn_read()` calls
`pci171x_ai_read_sample()` with the wrong value for the third parameter.
It is supposed to be the current index in a channel list (which is
always of length 1 in this case, so the index should be 0), but instead
it is passing the actual channel number.  `pci171x_ai_read_sample()`
checks the channel number encoded in the raw sample value read from the
hardware matches the channel number stored in the specified index of the
previously set up channel list and returns `-ENODATA` if it doesn't
match.  Since the index should always be 0 in this case, the match will
fail unless the channel number is also 0.  Fix it by passing 0 as the
channel index.

Note that when the bug first appeared, it was `pci171x_ai_dropout()`
that was called with the wrong parameter value.  `pci171x_ai_dropout()`
got replaced with `pci171x_ai_read_sample()` in commit 7fd2dae2500d
("staging: comedi: adv_pci1710: introduce pci171x_ai_read_sample()").

Fixes: 16c7eb6047bb ("staging: comedi: adv_pci1710: always enable PCI171x_PARANOIDCHECK code")
Signed-off-by: Ian Abbott <[email protected]>
Cc: stable <[email protected]> # 3.16+
Signed-off-by: Greg Kroah-Hartman <[email protected]>
10 years agoandroid: binder: fix binder mmap failures
Andrey Ryabinin [Fri, 27 Feb 2015 17:44:21 +0000 (20:44 +0300)]
android: binder: fix binder mmap failures

binder_update_page_range() initializes only addr and size
fields in 'struct vm_struct tmp_area;' and passes it to
map_vm_area().

Before 71394fe50146 ("mm: vmalloc: add flag preventing guard hole allocation")
this was because map_vm_area() didn't use any other fields
in vm_struct except addr and size.

Now get_vm_area_size() (used in map_vm_area()) reads vm_struct's
flags to determine whether vm area has guard hole or not.

binder_update_page_range() don't initialize flags field, so
this causes following binder mmap failures:
-----------[ cut here ]------------
WARNING: CPU: 0 PID: 1971 at mm/vmalloc.c:130
vmap_page_range_noflush+0x119/0x144()
CPU: 0 PID: 1971 Comm: healthd Not tainted 4.0.0-rc1-00399-g7da3fdc-dirty #157
Hardware name: ARM-Versatile Express
[<c001246d>] (unwind_backtrace) from [<c000f7f9>] (show_stack+0x11/0x14)
[<c000f7f9>] (show_stack) from [<c049a221>] (dump_stack+0x59/0x7c)
[<c049a221>] (dump_stack) from [<c001cf21>] (warn_slowpath_common+0x55/0x84)
[<c001cf21>] (warn_slowpath_common) from [<c001cfe3>]
(warn_slowpath_null+0x17/0x1c)
[<c001cfe3>] (warn_slowpath_null) from [<c00c66c5>]
(vmap_page_range_noflush+0x119/0x144)
[<c00c66c5>] (vmap_page_range_noflush) from [<c00c716b>] (map_vm_area+0x27/0x48)
[<c00c716b>] (map_vm_area) from [<c038ddaf>]
(binder_update_page_range+0x12f/0x27c)
[<c038ddaf>] (binder_update_page_range) from [<c038e857>]
(binder_mmap+0xbf/0x1ac)
[<c038e857>] (binder_mmap) from [<c00c2dc7>] (mmap_region+0x2eb/0x4d4)
[<c00c2dc7>] (mmap_region) from [<c00c3197>] (do_mmap_pgoff+0x1e7/0x250)
[<c00c3197>] (do_mmap_pgoff) from [<c00b35b5>] (vm_mmap_pgoff+0x45/0x60)
[<c00b35b5>] (vm_mmap_pgoff) from [<c00c1f39>] (SyS_mmap_pgoff+0x5d/0x80)
[<c00c1f39>] (SyS_mmap_pgoff) from [<c000ce81>] (ret_fast_syscall+0x1/0x5c)
---[ end trace 48c2c4b9a1349e54 ]---
binder: 1982: binder_alloc_buf failed to map page at f0e00000 in kernel
binder: binder_mmap: 1982 b6bde000-b6cdc000 alloc small buf failed -12

Use map_kernel_range_noflush() instead of map_vm_area() as this is better
API for binder's purposes and it allows to get rid of 'vm_struct tmp_area' at all.

Fixes: 71394fe50146 ("mm: vmalloc: add flag preventing guard hole allocation")
Signed-off-by: Andrey Ryabinin <[email protected]>
Reported-by: Amit Pundir <[email protected]>
Tested-by: Amit Pundir <[email protected]>
Acked-by: David Rientjes <[email protected]>
Tested-by: John Stultz <[email protected]>
Signed-off-by: Greg Kroah-Hartman <[email protected]>
10 years agoARM: OMAP4+: PRM: fix omap4 version of prm_save_and_clear_irqen
Tero Kristo [Fri, 27 Feb 2015 13:59:26 +0000 (15:59 +0200)]
ARM: OMAP4+: PRM: fix omap4 version of prm_save_and_clear_irqen

This was incorrectly reading the irq status registers during the save
and clear, instead of the irq enable. This worked because there is only
one user for the prcm interrupts currently, namely the io-chain. Whenever
the function was called, an io-chain interrupt was both pending and
enabled.

Signed-off-by: Tero Kristo <[email protected]>
Cc: Paul Walmsley <[email protected]>
Cc: Tony Lindgren <[email protected]>
Signed-off-by: Paul Walmsley <[email protected]>
10 years agoARM: OMAP2+: hwmod: fix deassert hardreset clkdm usecounting
Tero Kristo [Thu, 26 Feb 2015 16:06:00 +0000 (18:06 +0200)]
ARM: OMAP2+: hwmod: fix deassert hardreset clkdm usecounting

Deasserting hardreset increases the usecount for the hwmod parent clockdomain
always, however usecount is only decreased at end in certain error cases.
This causes software supervised clockdomains to remain always on, preventing
idle. Fixed by always releasing the hwmods clockdomain parent when exiting
the function.

Signed-off-by: Tero Kristo <[email protected]>
Tested-by: Carlos Hernandez <[email protected]>
Cc: Paul Walmsley <[email protected]>
Cc: Tony Lindgren <[email protected]>
Signed-off-by: Paul Walmsley <[email protected]>
10 years agoMerge branch 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 1 Mar 2015 20:22:44 +0000 (12:22 -0800)]
Merge branch 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip

Pull x86 fixes from Ingo Molnar:
 "A CR4-shadow 32-bit init fix, plus two typo fixes"

* 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  x86: Init per-cpu shadow copy of CR4 on 32-bit CPUs too
  x86/platform/intel-mid: Fix trivial printk message typo in intel_mid_arch_setup()
  x86/cpu/intel: Fix trivial typo in intel_tlb_table[]

10 years agoMerge branch 'timers-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 1 Mar 2015 20:00:25 +0000 (12:00 -0800)]
Merge branch 'timers-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip

Pull timer fixes from Ingo Molnar:
 "Three clockevents/clocksource driver fixes"

* 'timers-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  clocksource: pxa: Fix section mismatch
  clocksource: mtk: Fix race conditions in probe code
  clockevents: asm9260: Fix compilation error with sparc/sparc64 allyesconfig

10 years agoMerge branch 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 1 Mar 2015 19:56:13 +0000 (11:56 -0800)]
Merge branch 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip

Pull perf fixes from Ingo Molnar:
 "Two kprobes fixes and a handful of tooling fixes"

* 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  perf tools: Make sparc64 arch point to sparc
  perf symbols: Define EM_AARCH64 for older OSes
  perf top: Fix SIGBUS on sparc64
  perf tools: Fix probing for PERF_FLAG_FD_CLOEXEC flag
  perf tools: Fix pthread_attr_setaffinity_np build error
  perf tools: Define _GNU_SOURCE on pthread_attr_setaffinity_np feature check
  perf bench: Fix order of arguments to memcpy_alloc_mem
  kprobes/x86: Check for invalid ftrace location in __recover_probed_insn()
  kprobes/x86: Use 5-byte NOP when the code might be modified by ftrace

10 years agoMerge branch 'locking-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 1 Mar 2015 19:27:04 +0000 (11:27 -0800)]
Merge branch 'locking-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip

Pull locking fix from Ingo Molnar:
 "An rtmutex deadlock path fixlet"

* 'locking-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  locking/rtmutex: Set state back to running on error

10 years agoMerge branch 'bcmgenet_systemport_stats'
David S. Miller [Sun, 1 Mar 2015 19:02:24 +0000 (14:02 -0500)]
Merge branch 'bcmgenet_systemport_stats'

Florian Fainelli says:

====================
net: bcmgenet and systemport statistics fixes

This two patches fix a similar problem in the GENET and SYSTEMPORT drivers
for software maintained statistics used to track DMA mapping and SKB
re-allocation failures.
====================

Signed-off-by: David S. Miller <[email protected]>
10 years agonet: systemport: fix software maintained statistics
Florian Fainelli [Sun, 1 Mar 2015 02:09:17 +0000 (18:09 -0800)]
net: systemport: fix software maintained statistics

Commit 60b4ea1781fd ("net: systemport: log RX buffer allocation and RX/TX DMA
failures") added a few software maintained statistics using
BCM_SYSPORT_STAT_MIB_RX and BCM_SYSPORT_STAT_MIB_TX. These statistics are read
from the hardware MIB counters, such that bcm_sysport_update_mib_counters() was
trying to read from a non-existing MIB offset for these counters.

Fix this by introducing a special type: BCM_SYSPORT_STAT_SOFT, similar to
BCM_SYSPORT_STAT_NETDEV, such that bcm_sysport_get_ethtool_stats will read from
the software mib.

Fixes: 60b4ea1781fd ("net: systemport: log RX buffer allocation and RX/TX DMA failures")
Signed-off-by: Florian Fainelli <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
10 years agonet: bcmgenet: fix software maintained statistics
Florian Fainelli [Sun, 1 Mar 2015 02:09:16 +0000 (18:09 -0800)]
net: bcmgenet: fix software maintained statistics

Commit 44c8bc3ce39f ("net: bcmgenet: log RX buffer allocation and RX/TX dma
failures") added a few software maintained statistics using
BCMGENET_STAT_MIB_RX and BCMGENET_STAT_MIB_TX. These statistics are read from
the hardware MIB counters, such that bcmgenet_update_mib_counters() was trying
to read from a non-existing MIB offset for these counters.

Fix this by introducing a special type: BCMGENET_STAT_SOFT, similar to
BCMGENET_STAT_NETDEV, such that bcmgenet_get_ethtool_stats will read from the
software mib.

Fixes: 44c8bc3ce39f ("net: bcmgenet: log RX buffer allocation and RX/TX dma failures")
Signed-off-by: Florian Fainelli <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
10 years agorxrpc: don't multiply with HZ twice
Florian Westphal [Sat, 28 Feb 2015 10:51:37 +0000 (11:51 +0100)]
rxrpc: don't multiply with HZ twice

rxrpc_resend_timeout has an initial value of 4 * HZ; use it as-is.

Signed-off-by: Florian Westphal <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
10 years agorxrpc: terminate retrans loop when sending of skb fails
Florian Westphal [Sat, 28 Feb 2015 10:51:36 +0000 (11:51 +0100)]
rxrpc: terminate retrans loop when sending of skb fails

Typo, 'stop' is never set to true.
Seems intent is to not attempt to retransmit more packets after sendmsg
returns an error.

This change is based on code inspection only.

Signed-off-by: Florian Westphal <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
10 years agonet/hsr: Fix NULL pointer dereference and refcnt bugs when deleting a HSR interface.
Arvid Brodin [Fri, 27 Feb 2015 20:26:03 +0000 (21:26 +0100)]
net/hsr: Fix NULL pointer dereference and refcnt bugs when deleting a HSR interface.

To repeat:

$ sudo ip link del hsr0
BUG: unable to handle kernel NULL pointer dereference at 0000000000000018
IP: [<ffffffff8187f495>] hsr_del_port+0x15/0xa0
etc...

Bug description:

As part of the hsr master device destruction, hsr_del_port() is called for each of
the hsr ports. At each such call, the master device is updated regarding features
and mtu. When the master device is freed before the slave interfaces, master will
be NULL in hsr_del_port(), which led to a NULL pointer dereference.

Additionally, dev_put() was called on the master device itself in hsr_del_port(),
causing a refcnt error.

A third bug in the same code path was that the rtnl lock was not taken before
hsr_del_port() was called as part of hsr_dev_destroy().

The reporter (Nicolas Dichtel) also said: "hsr_netdev_notify() supposes that the
port will always be available when the notification is for an hsr interface. It's
wrong. For example, netdev_wait_allrefs() may resend NETDEV_UNREGISTER.". As a
precaution against this, a check for port == NULL was added in hsr_dev_notify().

Reported-by: Nicolas Dichtel <[email protected]>
Fixes: 51f3c605318b056a ("net/hsr: Move slave init to hsr_slave.c.")
Signed-off-by: Arvid Brodin <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
10 years agonet: pasemi: Use setup_timer and mod_timer
Vaishali Thakkar [Fri, 27 Feb 2015 18:50:59 +0000 (00:20 +0530)]
net: pasemi: Use setup_timer and mod_timer

Use timer API functions setup_timer and mod_timer instead
of structure assignments as they are standard way to set
the timer and to update the expire field of an active timer
respectively.

This is done using Coccinelle and semantic patch used for
this is as follows:

// <smpl>
@@
expression x,y,z,a,b;
@@

-init_timer (&x);
+setup_timer (&x, y, z);
+mod_timer (&a, b);
-x.function = y;
-x.data = z;
-x.expires = b;
-add_timer(&a);
// </smpl>

Signed-off-by: Vaishali Thakkar <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
10 years agonet: stmmac: Use setup_timer and mod_timer
Vaishali Thakkar [Fri, 27 Feb 2015 18:42:34 +0000 (00:12 +0530)]
net: stmmac: Use setup_timer and mod_timer

Use timer API functions setup_timer and mod_timer instead
of structure assignments as they are standard way to set
the timer and to update the expire field of an active timer
respectively.

This is done using Coccinelle and semantic patch used for
this is as follows:

// <smpl>
@@
expression x,y,z,a,b;
@@

-init_timer (&x);
+setup_timer (&x, y, z);
+mod_timer (&a, b);
-x.function = y;
-x.data = z;
-x.expires = b;
-add_timer(&a);
// </smpl>

Signed-off-by: Vaishali Thakkar <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
10 years agonet: 8390: axnet_cs: Use setup_timer and mod_timer
Vaishali Thakkar [Fri, 27 Feb 2015 18:32:34 +0000 (00:02 +0530)]
net: 8390: axnet_cs: Use setup_timer and mod_timer

Use timer API functions setup_timer and mod_timer instead
of structure assignments as they are standard way to set
the timer and to update the expire field of an active timer
respectively.

This is done using Coccinelle and semantic patch used for
this is as follows:

// <smpl>
@@
expression x,y,z,a,b;
@@

-init_timer (&x);
+setup_timer (&x, y, z);
+mod_timer (&a, b);
-x.function = y;
-x.data = z;
-x.expires = b;
-add_timer(&a);
// </smpl>

Signed-off-by: Vaishali Thakkar <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
10 years agonet: 8390: pcnet_cs: Use setup_timer and mod_timer
Vaishali Thakkar [Fri, 27 Feb 2015 18:23:03 +0000 (23:53 +0530)]
net: 8390: pcnet_cs: Use setup_timer and mod_timer

Use timer API functions setup_timer and mod_timer instead
of structure assignments as they are standard way to set
the timer and to update the expire field of an active timer
respectively.

This is done using Coccinelle and semantic patch used for
this is as follows:

// <smpl>
@@
expression x,y,z,a,b;
@@

-init_timer (&x);
+setup_timer (&x, y, z);
+mod_timer (&a, b);
-x.function = y;
-x.data = z;
-x.expires = b;
-add_timer(&a);
// </smpl>

Signed-off-by: Vaishali Thakkar <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
10 years agonet: smc91c92_cs: Use setup_timer and mod_timer
Vaishali Thakkar [Fri, 27 Feb 2015 18:10:24 +0000 (23:40 +0530)]
net: smc91c92_cs: Use setup_timer and mod_timer

Use timer API functions setup_timer and mod_timer instead
of structure assignments as they are standard way to set
the timer and to update the expire field of an active timer
respectively.

This is done using Coccinelle and semantic patch used for
this is as follows:

// <smpl>
@@
expression x,y,z,a,b;
@@

-init_timer (&x);
+setup_timer (&x, y, z);
+mod_timer (&a, b);
-x.function = y;
-x.data = z;
-x.expires = b;
-add_timer(&a);
// </smpl>

Signed-off-by: Vaishali Thakkar <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
This page took 0.132903 seconds and 4 git commands to generate.