DMA mapped IO should be unmapped on the error path in probe() and
unconditionally on remove().

Fixes: 62936009 ([libata] Add 460EX on-chip SATA driver, sata_dwc_460ex)
Signed-off-by: NAndy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: NTejun Heo <tj@kernel.org>

The driver sata_dwc_460ex is using this symbol. To build it as a
module we have to have the symbol exported. This patch adds
EXPORT_SYMBOL_GPL() macro for that.

tj: Updated to use EXPORT_SYMBOL_GPL() instead of EXPORT_SYMBOL() as
    the only known user is an in-tree driver.  Suggested by Sergei.
Signed-off-by: NAndy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: NTejun Heo <tj@kernel.org>
Cc: Sergei Shtylyov <sergei.shtylyov@cogentembedded.com>

This patch addresses the issue with ATA_CMD_PACKET pio mode
command for enumeration and device detection with ATAPI devices.
The X-Gene AHCI controller has an errata in which it cannot clear
the BSY bit after the PIO setup FIS. The dma state machine enters
CMFatalErrorUpdate state and locks up.
Signed-off-by: NSuman Tripathi <stripathi@apm.com>
Signed-off-by: NTejun Heo <tj@kernel.org>

This patch fixes the big endian mode issue with function
xgene_ahci_read_id.
Signed-off-by: NSuman Tripathi <stripathi@apm.com>
Signed-off-by: NTejun Heo <tj@kernel.org>

Signed-off-by: NWolfram Sang <wsa+renesas@sang-engineering.com>
Signed-off-by: NWolfram Sang <wsa@the-dreams.de>

Having switched over all of the users of CONFIG_PM_RUNTIME to use
CONFIG_PM directly, turn the latter into a user-selectable option
and drop the former entirely from the tree.
Signed-off-by: NRafael J. Wysocki <rafael.j.wysocki@intel.com>
Reviewed-by: NUlf Hansson <ulf.hansson@linaro.org>
Acked-by: NKevin Hilman <khilman@linaro.org>

I'm leaving Red Hat at the end of December 2014, so remove all
references to my soon-to-be-dead address.

(There are some references left in the tree, that need additional
changes, I'll send those through the AGP maintainers).
Signed-off-by: NDave Jones <davej@codemonkey.org.uk>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

The new driver is around for more than 2 years now, so the old one can
go. Getting rid of it helps the removal of the legacy .attach_adapter
callback of the I2C subsystem.
Signed-off-by: NWolfram Sang <wsa@the-dreams.de>
Acked-by: NBenjamin Herrenschmidt <benh@kernel.crashing.org>

The 8250_omap serial driver is a new user of CONFIG_PM_RUNTIME.

However, after commit b2b49ccb (PM: Kconfig: Set PM_RUNTIME if
PM_SLEEP is selected) PM_RUNTIME is always set if PM is set, so
#ifdef blocks depending on CONFIG_PM_RUNTIME may now be changed to
depend on CONFIG_PM.

Replace CONFIG_PM_RUNTIME with CONFIG_PM in 8250_omap.c.
Signed-off-by: NRafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: NGreg Kroah-Hartman <gregkh@linuxfoundation.org>

A couple of new CONFIG_PM_RUNTIME users have been added recently
in the SPI subsystem.

However, after commit b2b49ccb (PM: Kconfig: Set PM_RUNTIME if
PM_SLEEP is selected) PM_RUNTIME is always set if PM is set, so
#ifdef blocks depending on CONFIG_PM_RUNTIME may now be changed to
depend on CONFIG_PM.

Replace CONFIG_PM_RUNTIME with CONFIG_PM everywhere under
drivers/spi/ (again).
Signed-off-by: NRafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: NMark Brown <broonie@kernel.org>

commit 4dbd2771 "tick: export nohz tick idle symbols for module
use" was merged via the thermal tree without an explicit ack from the
relevant maintainers.

The exports are abused by the intel powerclamp driver which implements
a fake idle state from a sched FIFO task. This causes all kinds of
wreckage in the NOHZ core code which rightfully assumes that
tick_nohz_idle_enter/exit() are only called from the idle task itself.

Recent changes in the NOHZ core lead to a failure of the powerclamp
driver and now people try to hack completely broken and backwards
workarounds into the NOHZ core code. This is completely unacceptable
and just papers over the real problem. There are way more subtle
issues lurking around the corner.

The real solution is to fix the powerclamp driver by rewriting it with
a sane concept, but that's beyond the scope of this.

So the only solution for now is to remove the calls into the core NOHZ
code from the powerclamp trainwreck along with the exports. 

Fixes: d6d71ee4 "PM: Introduce Intel PowerClamp Driver"
Signed-off-by: NThomas Gleixner <tglx@linutronix.de>
Cc: Preeti U Murthy <preeti@linux.vnet.ibm.com>
Cc: Viresh Kumar <viresh.kumar@linaro.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Fengguang Wu <fengguang.wu@intel.com>
Cc: Frederic Weisbecker <frederic@kernel.org>
Cc: Pan Jacob jun <jacob.jun.pan@intel.com>
Cc: LKP <lkp@01.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Zhang Rui <rui.zhang@intel.com>
Cc: stable@vger.kernel.org
Link: http://lkml.kernel.org/r/alpine.DEB.2.11.1412181110110.17382@nanosSigned-off-by: NThomas Gleixner <tglx@linutronix.de>

This patch changes core_scsi3_pro_release() logic to allow an
existing AllRegistrants type reservation to be re-reserved by
any registered I_T nexus.

This addresses a issue where AllRegistrants type RESERVE was
receiving RESERVATION_CONFLICT status if dev_pr_res_holder did
not match the same I_T nexus, instead of just returning GOOD
status following spc4r34 Section 5.9.9:

"If the device server receives a PERSISTENT RESERVE OUT command
 with RESERVE service action where the TYPE field and the SCOPE
 field contain the same values as the existing type and scope
 from a persistent reservation holder, it shall not make any
 change to the existing persistent reservation and shall complete
 the command with GOOD status."
Reported-by: NIlias Tsitsimpis <i.tsitsimpis@gmail.com>
Cc: Ilias Tsitsimpis <i.tsitsimpis@gmail.com>
Cc: Lee Duncan <lduncan@suse.com>
Cc: James Bottomley <James.Bottomley@HansenPartnership.com>
Signed-off-by: NNicholas Bellinger <nab@linux-iscsi.org>

An earlier commit to resolve an issue with encapsulation offloads missed
setting a bit in the outer netdev features flag. This results in loss of TSO
feature on a VxLAN interface.

Fixes: 630f4b70 ("Export tunnel offloads only when a VxLAN tunnel is created")
Signed-off-by: NSriharsha Basavapatna <sriharsha.basavapatna@emulex.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Noticed when debugging ptp.
Signed-off-by: NJiri Benc <jbenc@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Commit bc96f648 (xen-netback: make
feature-rx-notify mandatory) incorrectly assumed that there were no
frontends in use that did not support this feature.  But the frontend
driver in MiniOS does not and since this is used by (qemu) stubdoms,
these stopped working.

Netback sort of works as-is in this mode except:

- If there are no Rx requests and the internal Rx queue fills, only
  the drain timeout will wake the thread.  The default drain timeout
  of 10 s would give unacceptable pauses.

- If an Rx stall was detected and the internal Rx queue is drained,
  then the Rx thread would never wake.

Handle these two cases (when feature-rx-notify is disabled) by:

- Reducing the drain timeout to 30 ms.

- Disabling Rx stall detection.
Reported-by: NJohn <jw@nuclearfallout.net>
Tested-by: NJohn <jw@nuclearfallout.net>
Signed-off-by: NDavid Vrabel <david.vrabel@citrix.com>
Reviewed-by: NWei Liu <wei.liu2@citrix.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

QSA module was getting decoded as QSFP module in ethtool get settings, this
patch fixes it.
Signed-off-by: NHariprasad Shenai <hariprasad@chelsio.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Fix the following build warning by passing the expected argument type to
watchdog_active():

drivers/watchdog/imx2_wdt.c: In function 'imx2_wdt_suspend':
drivers/watchdog/imx2_wdt.c:340:2: warning: passing argument 1 of 'watchdog_active' from incompatible pointer type [enabled by default]
In file included from drivers/watchdog/imx2_wdt.c:38:0:
include/linux/watchdog.h:104:20: note: expected 'struct watchdog_device *' but argument is of type 'struct watchdog_device **'
Reported-by: NOlof's autobuilder <build@lixom.net>
Signed-off-by: NFabio Estevam <fabio.estevam@freescale.com>
Reviewed-by: NGuenter Roeck <linux@roeck-us.net>
Signed-off-by: NWim Van Sebroeck <wim@iguana.be>

Add power management operations(suspend and resume) as part of
dev_pm_ops for IMX2 watchdog driver.
Signed-off-by: NXiubo Li <Li.Xiubo@freescale.com>
Reviewed-by: NGuenter Roeck <linux@roeck-us.net>
Signed-off-by: NWim Van Sebroeck <wim@iguana.be>

If NO_DMA=y:

drivers/built-in.o: In function `sh_mobile_i2c_dma_unmap':
i2c-sh_mobile.c:(.text+0x60de42): undefined reference to `dma_unmap_single'
drivers/built-in.o: In function `sh_mobile_i2c_xfer_dma':
i2c-sh_mobile.c:(.text+0x60df22): undefined reference to `dma_map_single'
i2c-sh_mobile.c:(.text+0x60df2e): undefined reference to `dma_mapping_error'
Signed-off-by: NGeert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: NWolfram Sang <wsa@the-dreams.de>

DMA is opt-in for this driver. So, we can't use deferred probing for
requesting DMA channels in probe, because our driver would get endlessly
deferred if DMA support is compiled in AND the DMA driver is missing.
Because we can't know when the DMA driver might show up, we always try
again when a DMA transfer would be possible. The downside is that there
is more overhead for setting up PIO transfers under the above scenario.
But well, having DMA enabled and the proper DMA driver missing looks
like a broken or test config anyhow.
Reported-by: NGeert Uytterhoeven <geert+renesas@glider.be>
Signed-off-by: NWolfram Sang <wsa+renesas@sang-engineering.com>
Signed-off-by: NWolfram Sang <wsa@the-dreams.de>

Refactor DMA setup to keep the errno so we can implement better
deferred probe support in the next step.
Signed-off-by: NWolfram Sang <wsa+renesas@sang-engineering.com>
Signed-off-by: NWolfram Sang <wsa@the-dreams.de>

Originally, the I2C controller supported by the i2c-mv64xxx driver
requires a lot of software support: an interrupt is generated at each
step of an I2C transaction (after the start bit, after sending the
address, etc.) and the driver is in charge of re-programming the I2C
controller to do the next step of the I2C transaction. This explains
the fairly complex state machine that the driver has.

On Marvell Armada XP and later processors (Armada 375, 38x, etc.), the
I2C controller was extended with a part called the "I2C Bridge", which
allows to offload the I2C transaction completely to the
hardware. Initial support for this mechanism was added in commit
930ab3d4 ("i2c: mv64xxx: Add I2C Transaction Generator support").

However, the implementation done in this commit has two related
issues, which this commit fixes by completely changing how the offload
implementation is done:

 * SMBus read transfers, where there is one write to select the
   register immediately followed in the same transaction by one read,
   were making the processor hang. This was easier visible on the
   Marvell Armada XP WRT1900AC platform using a driver for an I2C LED
   controller, or on other Armada XP platforms by using a simple
   'i2cget' command to read an I2C EEPROM.

 * The implementation was based on the fact that the offload engine
   was re-programmed to transfer each message of an I2C xfer: this
   meant that each message sent with the offload engine was starting
   with a normal I2C start sequence. However, the I2C subsystem
   assumes that all messages belonging to the same xfer will use the
   so-called "repeated start" so that the entire I2C xfer is seen as
   one transfer by the I2C devices and cannot be interrupt by other
   I2C masters on the same bus.

In fact, the "I2C Bridge" allows to offload three types of xfer:

 - xfer of one write message
 - xfer of one read message
 - xfer of one write message followed by one read message

For all other situations, we have to fallback to not using the "I2C
Bridge" in order to get proper I2C semantics.

Therefore, this commit reworks the offload implementation to put it
not at the message level, but at the xfer level: in the
mv64xxx_i2c_xfer() function, we decide if the transaction can be
offloaded (in which case it is handled by the
mv64xxx_i2c_offload_xfer() function), or otherwise it is handled by
the slow path (implemented in the existing mv64xxx_i2c_execute_msg()).

This allows to simplify the state machine, which no longer needs to
have any state related to the offload implementation: the offload
implementation is now completely separated from the slow path (with
the exception of the interrupt handler, of course).

In summary:

 - mv64xxx_i2c_can_offload() will analyze an I2C xfer and decided of
   the "I2C Bridge" can be used to offload it or not.

 - mv64xxx_i2c_offload_xfer() will actually program the "I2C Bridge"
   to offload one xfer (of either one or two messages), and block
   using mv64xxx_i2c_wait_for_completion() until the xfer completes.

 - The interrupt handler mv64xxx_i2c_intr() is modified to push the
   offload related code to a separate function,
   mv64xxx_i2c_intr_offload(). It will take care of reading the
   received data if needed.

This commit was tested on:

 - Armada XP OpenBlocks AX3-4 (EEPROM on I2C and RTC on I2C)
 - Armada XP WRT1900AC (LED controller on I2C)
 - Armada XP GP (EEPROM on I2C)

Fixes: 930ab3d4 ("i2c: mv64xxx: Add I2C Transaction Generator support")
Cc: <stable@vger.kernel.org> # v3.12+
Signed-off-by: NThomas Petazzoni <thomas.petazzoni@free-electrons.com>
[wsa: fixed checkpatch warnings]
Signed-off-by: NWolfram Sang <wsa@the-dreams.de>

Signed-off-by: NThomas Petazzoni <thomas.petazzoni@free-electrons.com>
Signed-off-by: NWolfram Sang <wsa@the-dreams.de>

CEPH_OSD_OP_DELETE is not an extent op, stop treating it as such.  This
sneaked in with discard patches - it's one of the three osd ops (the
other two are CEPH_OSD_OP_TRUNCATE and CEPH_OSD_OP_ZERO) that discard
is implemented with.
Signed-off-by: NIlya Dryomov <idryomov@redhat.com>
Reviewed-by: NAlex Elder <elder@linaro.org>

The functions ceph_put_snap_context() and iput() test whether their
argument is NULL and then return immediately. Thus the test around the
call is not needed.

This issue was detected by using the Coccinelle software.
Signed-off-by: NMarkus Elfring <elfring@users.sourceforge.net>
[idryomov@redhat.com: squashed rbd.c hunk, changelog]
Signed-off-by: NIlya Dryomov <idryomov@redhat.com>

Audio subsystem clocks are located in separate block. On Exynos 5420 if
clock for this block (from main clock domain) 'mau_epll' is gated then
any read or write to audss registers will block.

This kind of boot hang was observed on Arndale Octa and Peach Pi/Pit
after introducing runtime PM to pl330 DMA driver. After that commit the
'mau_epll' was gated, because the "amba" clock was disabled and there
were no more users of mau_epll.

The system hang on one of steps:
1. Disabling unused clocks from audss block.
2. During audss GPIO setup (just before probing i2s0 because
   samsung_pinmux_setup() tried to access memory from audss block which was
   gated.

Add a workaround for this by enabling the 'mau_epll' clock in probe.
Signed-off-by: NKrzysztof Kozlowski <k.kozlowski@samsung.com>
Acked-by: NSylwester Nawrocki <s.nawrocki@samsung.com>
Tested-by: NJavier Martinez Canillas <javier.martinez@collabora.co.uk>
Tested-by: NKevin Hilman <khilman@linaro.org>
Signed-off-by: NMichael Turquette <mturquette@linaro.org>

* removed several pieces of dead code in lustre_compat25.h
* don't open-code current_umask() (and BTW, 0755 & (S_IRWXUGO | S_ISVTX)
is better spelled as 0755)
* fix broken attempt to get the pathname by dentry - abusing d_path() for
that is simply wrong.
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

These drivers haven't been tested in a long, long time. The hardware is
ancient and hopelessly obsolete. These drivers also need to be converted
to newer media frameworks but due to the lack of hardware that's going
to be impossible. In addition, cheaper and vastly better hardware is
available today.

So these drivers are a prime candidate for removal. If someone is
interested in working on these drivers to prevent their removal, then
please contact the linux-media mailinglist.

Let's be honest, the age of parallel port webcams and ISA video capture
boards is really gone.
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@osg.samsung.com>

This driver hasn't been tested in a long, long time. The company that made
this chip has gone bust many years ago and hardware using this chip is next
to impossible to find.

This driver needs to be converted to newer media frameworks but due to the
lack of hardware that's going to be impossible. Since cheap alternatives are
easily available, there is little point in keeping this driver alive.

In other words, this driver is a prime candidate for removal. If someone is
interested in working on this driver to prevent its removal, then please
contact the linux-media mailinglist.
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Acked-by: NHuang Shijie <shijie8@gmail.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@osg.samsung.com>

These drivers haven't been tested in a long, long time. The hardware is
ancient and hopelessly obsolete. These drivers also need to be converted
to newer media frameworks but due to the lack of hardware that's going
to be impossible.

So these drivers are a prime candidate for removal. If someone is
interested in working on these drivers to prevent their removal, then
please contact the linux-media mailinglist.
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@osg.samsung.com>

The start_streaming op is responsible for starting the video dma,
so it shouldn't be called anymore from the buf_queue op.

Unfortunately, this call to start_video_dma() was added to the
start_streaming op, but was forgotten to be removed from the
buf_queue op, which is where it used to be before the vb2 conversion.

Calling this function twice causes very hard to find errors: sometimes
it works, sometimes it doesn't. It took me a whole friggin' day
to track this down, and in the end it was just luck that my eye suddenly
triggered on that line.
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@osg.samsung.com>

The cx88 vb2 conversion and the vb2 dma_sg improvements were developed separately and
were merged separately. Unfortunately, the patch updating drivers to the dma_sg
improvements didn't take the updated cx88 driver into account. Basically two ships
passing in the night, unaware of one another even though both ships have the same
owner, i.e. me :-)
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Reported-by: NChris Lee <updatelee@gmail.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@osg.samsung.com>

This is easy to forget to do in drivers. While v4l2-compliance will check for it,
not everyone remembers to run it. So warn about it.
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@osg.samsung.com>

An error was returned if composing was not supported, instead of if
cropping was not supported.

A classic copy-and-paste bug. Found with v4l2-compliance.
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Cc: <stable@vger.kernel.org>      # for v3.18
Signed-off-by: NMauro Carvalho Chehab <mchehab@osg.samsung.com>

When the virtio_pci driver was moved into virtio_pci_legacy.c the module
licence and other attributes went AWOL.  This patch restores them.
Signed-off-by: NHerbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: NMichael S. Tsirkin <mst@redhat.com>

The current implementations all use dev_uc_add_excl() and such whose API
doesn't support vlans, so we can't make it with NICs HW for now.

Fixes: f6f6424b ('net: make vid as a parameter for ndo_fdb_add/ndo_fdb_del')
Signed-off-by: NOr Gerlitz <ogerlitz@mellanox.com>
Reviewed-by: NJiri Pirko <jiri@resnulli.us>
Acked-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The Exynos IOMMU driver uses the ARM specific dmac_flush_range() and
outer_flush_range() functions. This breaks the build on arm64 allmodconfig
in -next since support has been merged for some Exynos ARMv8 SoCs. Add a
dependency on ARM to keep things building until either the driver has the
ARM dependencies removed or the ARMv8 architecture code implements these
ARM specific APIs.
Signed-off-by: NMark Brown <broonie@kernel.org>
Signed-off-by: NArnd Bergmann <arnd@arndb.de>

This commit contains 2 fixes for the 128B CQE/EQE stride feaure.
Wei found that mlx4_QUERY_HCA function marked the wrong capability
in flags (64B CQE/EQE), when CQE/EQE stride feature was enabled.
Also added small fix in initial CQE ownership bit assignment, when CQE
is size is not default 32B.

Fixes: 77507aa2 (net/mlx4: Enable CQE/EQE stride support)
Signed-off-by: NWei Yang <weiyang@linux.vnet.ibm.com>
Signed-off-by: NIdo Shamay <idos@mellanox.com>
Signed-off-by: NAmir Vadai <amirv@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Do camera capture test on i.MX6q sabresd board, and save the capture data to
nfs rootfs. The command is:
gst-launch-1.0 -e imxv4l2src device=/dev/video1 num-buffers=2592000 ! tee name=t !
queue ! imxv4l2sink sync=false t. ! queue ! vpuenc ! queue ! mux. pulsesrc num-buffers=3720937
blocksize=4096 ! 'audio/x-raw, rate=44100, channels=2' ! queue ! imxmp3enc ! mpegaudioparse !
queue ! mux. qtmux name=mux ! filesink location=video_recording_long.mov

After about 10 hours running, there have net watchdog timeout kernel dump:
...
WARNING: CPU: 0 PID: 0 at net/sched/sch_generic.c:264 dev_watchdog+0x2b4/0x2d8()
NETDEV WATCHDOG: eth0 (fec): transmit queue 0 timed out
CPU: 0 PID: 0 Comm: swapper/0 Not tainted 3.14.24-01051-gdb840b7 #440
[<80014e6c>] (unwind_backtrace) from [<800118ac>] (show_stack+0x10/0x14)
[<800118ac>] (show_stack) from [<806ae3f0>] (dump_stack+0x78/0xc0)
[<806ae3f0>] (dump_stack) from [<8002b504>] (warn_slowpath_common+0x68/0x8c)
[<8002b504>] (warn_slowpath_common) from [<8002b558>] (warn_slowpath_fmt+0x30/0x40)
[<8002b558>] (warn_slowpath_fmt) from [<8055e0d4>] (dev_watchdog+0x2b4/0x2d8)
[<8055e0d4>] (dev_watchdog) from [<800352d8>] (call_timer_fn.isra.33+0x24/0x8c)
[<800352d8>] (call_timer_fn.isra.33) from [<800354c4>] (run_timer_softirq+0x184/0x220)
[<800354c4>] (run_timer_softirq) from [<8002f420>] (__do_softirq+0xc0/0x22c)
[<8002f420>] (__do_softirq) from [<8002f804>] (irq_exit+0xa8/0xf4)
[<8002f804>] (irq_exit) from [<8000ee5c>] (handle_IRQ+0x54/0xb4)
[<8000ee5c>] (handle_IRQ) from [<80008598>] (gic_handle_irq+0x28/0x5c)
[<80008598>] (gic_handle_irq) from [<800123c0>] (__irq_svc+0x40/0x74)
Exception stack(0x80d27f18 to 0x80d27f60)
7f00:                                                       80d27f60 0000014c
7f20: 8858c60e 0000004d 884e4540 0000004d ab7250d0 80d34348 00000000 00000000
7f40: 00000001 00000000 00000017 80d27f60 800702a4 80476e6c 600f0013 ffffffff
[<800123c0>] (__irq_svc) from [<80476e6c>] (cpuidle_enter_state+0x50/0xe0)
[<80476e6c>] (cpuidle_enter_state) from [<80476fa8>] (cpuidle_idle_call+0xac/0x154)
[<80476fa8>] (cpuidle_idle_call) from [<8000f174>] (arch_cpu_idle+0x8/0x44)
[<8000f174>] (arch_cpu_idle) from [<80064c54>] (cpu_startup_entry+0x100/0x158)
[<80064c54>] (cpu_startup_entry) from [<80cd8a9c>] (start_kernel+0x304/0x368)
---[ end trace 09ebd32fb032f86d ]---
...

There might have a race in napi_schedule(), leaving interrupts disabled forever.
After these patch, the case still work more than 40 hours running.
Signed-off-by: NFugang Duan <B38611@freescale.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

After d75b1ade (net: less interrupt
masking in NAPI) the napi instance is removed from the per-cpu list
prior to calling the n->poll(), and is only requeued if all of the
budget was used.  This inadvertently broke netfront because netfront
does not use NAPI correctly.

If netfront had not used all of its budget it would do a final check
for any Rx responses and avoid calling napi_complete() if there were
more responses.  It would still return under budget so it would never
be rescheduled.  The final check would also not re-enable the Rx
interrupt.

Additionally, xenvif_poll() would also call napi_complete() /after/
enabling the interrupt.  This resulted in a race between the
napi_complete() and the napi_schedule() in the interrupt handler.  The
use of local_irq_save/restore() avoided by race iff the handler is
running on the same CPU but not if it was running on a different CPU.

Fix both of these by always calling napi_compete() if the budget was
not all used, and then calling napi_schedule() if the final checks
says there's more work.
Signed-off-by: NDavid Vrabel <david.vrabel@citrix.com>
Cc: Eric Dumazet <edumazet@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>