firefly-linux-kernel-4.4.55.git
13 years agofs/9p: Fid is not valid after a failed clunk.
Aneesh Kumar K.V [Mon, 11 Jul 2011 16:40:58 +0000 (16:40 +0000)]
fs/9p: Fid is not valid after a failed clunk.

commit 5034990e28efb2d232ee82443a9edd62defd17ba upstream.

free the fid even in case of failed clunk.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Eric Van Hensbergen <ericvh@gmail.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoVirtIO can transfer VIRTQUEUE_NUM of pages.
jvrao [Thu, 30 Jun 2011 23:18:41 +0000 (23:18 +0000)]
VirtIO can transfer VIRTQUEUE_NUM of pages.

commit 7f781679dd596c8abde8336b4d0d166d6a4aad04 upstream.

Signed-off-by: Venkateswararao Jujjuri "<jvrao@linux.vnet.ibm.com>
Reviewed-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Eric Van Hensbergen <ericvh@gmail.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoFix the size of receive buffer packing onto VirtIO ring.
jvrao [Thu, 30 Jun 2011 23:18:39 +0000 (23:18 +0000)]
Fix the size of receive buffer packing onto VirtIO ring.

commit 114e6f3a5ede73d5b56e145f04680c61c3dd67c4 upstream.

Signed-off-by: Venkateswararao Jujjuri "<jvrao@linux.vnet.ibm.com>
Reviewed-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Eric Van Hensbergen <ericvh@gmail.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agonet/9p: fix client code to fail more gracefully on protocol error
Eric Van Hensbergen [Thu, 14 Jul 2011 00:12:18 +0000 (19:12 -0500)]
net/9p: fix client code to fail more gracefully on protocol error

commit b85f7d92d7bd7e3298159e8b1eed8cb8cbbb0348 upstream.

There was a BUG_ON to protect against a bad id which could be dealt with
more gracefully.

Reported-by: Natalie Orlin <norlin@us.ibm.com>
Signed-off-by: Eric Van Hensbergen <ericvh@gmail.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agovp7045: fix buffer setup
Florian Mickler [Wed, 10 Aug 2011 10:05:20 +0000 (07:05 -0300)]
vp7045: fix buffer setup

commit fc61ccd35fd59d5362d37c8bf9c0526c85086c84 upstream.

dvb_usb_device_init calls the frontend_attach method of this driver which
uses vp7045_usb_ob. In order to have a buffer ready in vp7045_usb_op, it has to
be allocated before that happens.

Luckily we can use the whole private data as the buffer as it gets separately
allocated on the heap via kzalloc in dvb_usb_device_init and is thus apt for
use via usb_control_msg.

This fixes a
BUG: unable to handle kernel paging request at 0000000000001e78

reported by Tino Keitel and diagnosed by Dan Carpenter.

Tested-by: Tino Keitel <tino.keitel@tikei.de>
Signed-off-by: Florian Mickler <florian@mickler.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agonuvoton-cir: simplify raw IR sample handling
Jarod Wilson [Mon, 8 Aug 2011 20:20:40 +0000 (17:20 -0300)]
nuvoton-cir: simplify raw IR sample handling

commit de4ed0c111ed078b8729a5cc49c23197740f5bad upstream.

The nuvoton-cir driver was storing up consecutive pulse-pulse and
space-space samples internally, for no good reason, since
ir_raw_event_store_with_filter() already merges back to back like
samples types for us. This should also fix a regression introduced late
in 3.0 that related to a timeout change, which actually becomes correct
when coupled with this change. Tested with RC6 and RC5 on my own
nuvoton-cir hardware atop vanilla 3.0.0, after verifying quirky
behavior in 3.0 due to the timeout change.

Reported-by: Stephan Raue <sraue@openelec.tv>
CC: Stephan Raue <sraue@openelec.tv>
Signed-off-by: Jarod Wilson <jarod@redhat.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agomd: Fix handling for devices from 2TB to 4TB in 0.90 metadata.
NeilBrown [Sat, 10 Sep 2011 07:21:28 +0000 (17:21 +1000)]
md: Fix handling for devices from 2TB to 4TB in 0.90 metadata.

commit 27a7b260f71439c40546b43588448faac01adb93 upstream.

0.90 metadata uses an unsigned 32bit number to count the number of
kilobytes used from each device.
This should allow up to 4TB per device.
However we multiply this by 2 (to get sectors) before casting to a
larger type, so sizes above 2TB get truncated.

Also we allow rdev->sectors to be larger than 4TB, so it is possible
for the array to be resized larger than the metadata can handle.
So make sure rdev->sectors never exceeds 4TB when 0.90 metadata is in
used.

Also the sanity check at the end of super_90_load should include level
1 as it used ->size too. (RAID0 and Linear don't use ->size at all).

Reported-by: Pim Zandbergen <P.Zandbergen@macroscoop.nl>
Signed-off-by: NeilBrown <neilb@suse.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoAvoid dereferencing a 'request_queue' after last close.
NeilBrown [Sat, 10 Sep 2011 07:20:21 +0000 (17:20 +1000)]
Avoid dereferencing a 'request_queue' after last close.

commit 94007751bb02797ba87bac7aacee2731ac2039a3 upstream.

On the last close of an 'md' device which as been stopped, the device
is destroyed and in particular the request_queue is freed.  The free
is done in a separate thread so it might happen a short time later.

__blkdev_put calls bdev_inode_switch_bdi *after* ->release has been
called.

Since commit f758eeabeb96f878c860e8f110f94ec8820822a9
bdev_inode_switch_bdi will dereference the 'old' bdi, which lives
inside a request_queue, to get a spin lock.  This causes the last
close on an md device to sometime take a spin_lock which lives in
freed memory - which results in an oops.

So move the called to bdev_inode_switch_bdi before the call to
->release.

Cc: Christoph Hellwig <hch@lst.de>
Cc: Hugh Dickins <hughd@google.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Acked-by: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: NeilBrown <neilb@suse.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agodrm/nouveau: properly handle allocation failure in nouveau_sgdma_populate
Marcin Slusarz [Mon, 22 Aug 2011 21:14:05 +0000 (23:14 +0200)]
drm/nouveau: properly handle allocation failure in nouveau_sgdma_populate

commit 17c8b960930da3599e47801a54ac0ea1070545d2 upstream.

Not cleaning after alloc failure would result in crash on destroy,
because nouveau_sgdma_clear assumes "ttm_alloced" to be not null when
"pages" is not null.

Signed-off-by: Marcin Slusarz <marcin.slusarz@gmail.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoARM: davinci: fix cache flush build error
Linus Walleij [Tue, 2 Aug 2011 15:48:38 +0000 (17:48 +0200)]
ARM: davinci: fix cache flush build error

commit 897a6a1a14837d6d582bfd1fd7aba00be44b6469 upstream.

The TNET variant of DaVinci compiles some code that it shares
with other DaVinci variants, however it has a V6 CPU rather than
an ARM926T, thus the hardcoded call to arm926_flush_kern_cache_all()
in sleep.S will obviously fail, and we need to build with the
v6_flush_kern_cache_all() call instead. This was triggered by
manually altering the DaVinci config to build the TNET version.

Cc: Dave Martin <dave.martin@linaro.org>
Cc: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: Sekhar Nori <nsekhar@ti.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoARM: davinci: da850 EVM: read mac address from SPI flash
Sudhakar Rajashekhara [Tue, 12 Jul 2011 10:28:53 +0000 (15:58 +0530)]
ARM: davinci: da850 EVM: read mac address from SPI flash

commit 810198bc9c109489dfadc57131c5183ce6ad2d7d upstream.

DA850/OMAP-L138 EMAC driver uses random mac address instead of
a fixed one because the mac address is not stuffed into EMAC
platform data.

This patch provides a function which reads the mac address
stored in SPI flash (registered as MTD device) and populates the
EMAC platform data. The function which reads the mac address is
registered as a callback which gets called upon addition of MTD
device.

NOTE: In case the MAC address stored in SPI flash is erased, follow
the instructions at [1] to restore it.

[1] http://processors.wiki.ti.com/index.php/GSG:_OMAP-L138_DVEVM_Additional_Procedures#Restoring_MAC_address_on_SPI_Flash

Modifications in v2:
Guarded registering the mtd_notifier only when MTD is enabled.
Earlier this was handled using mtd_has_partitions() call, but
this has been removed in Linux v3.0.

Modifications in v3:
a. Guarded da850_evm_m25p80_notify_add() function and
   da850evm_spi_notifier structure with CONFIG_MTD macros.
b. Renamed da850_evm_register_mtd_user() function to
   da850_evm_setup_mac_addr() and removed the struct mtd_notifier
   argument to this function.
c. Passed the da850evm_spi_notifier structure to register_mtd_user()
   function.

Modifications in v4:
Moved the da850_evm_setup_mac_addr() function within the first
CONFIG_MTD ifdef construct.

Signed-off-by: Sudhakar Rajashekhara <sudhakar.raj@ti.com>
Signed-off-by: Sekhar Nori <nsekhar@ti.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoARM: 7081/1: mach-integrator: fix the clocksource
Linus Walleij [Tue, 6 Sep 2011 07:08:13 +0000 (08:08 +0100)]
ARM: 7081/1: mach-integrator: fix the clocksource

commit bb9ea77846620ed2b37e74c852d72c7a476b248c upstream.

I was intrigued by the fact that the clock stood still on
the Integrator, but it wasn't strange at all, because the
timer was set up all wrong and probably has been for a
while. With this patch the clock starts ticking again:
make the timer periodic (reload), |= on the divisor bit
and load the timer before starting it.

Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agohwmon: (max16065) Fix current calculation
Guenter Roeck [Sun, 28 Aug 2011 20:01:49 +0000 (13:01 -0700)]
hwmon: (max16065) Fix current calculation

commit ff71c182f461da5ae9d2d65f8a63f5a9193b9be1 upstream.

Current calculation is completely wrong. Add missing brackets to fix it.

Signed-off-by: Guenter Roeck <guenter.roeck@ericsson.com>
Acked-by: Jean Delvare <khali@linux-fr.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoxen/smp: Warn user why they keel over - nosmp or noapic and what to use instead.
Konrad Rzeszutek Wilk [Thu, 1 Sep 2011 13:48:27 +0000 (09:48 -0400)]
xen/smp: Warn user why they keel over - nosmp or noapic and what to use instead.

commit ed467e69f16e6b480e2face7bc5963834d025f91 upstream.

We have hit a couple of customer bugs where they would like to
use those parameters to run an UP kernel - but both of those
options turn of important sources of interrupt information so
we end up not being able to boot. The correct way is to
pass in 'dom0_max_vcpus=1' on the Xen hypervisor line and
the kernel will patch itself to be a UP kernel.

Fixes bug: http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=637308

Acked-by: Ian Campbell <Ian.Campbell@eu.citrix.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoxen: x86_32: do not enable iterrupts when returning from exception in interrupt context
Igor Mammedov [Thu, 1 Sep 2011 11:46:55 +0000 (13:46 +0200)]
xen: x86_32: do not enable iterrupts when returning from exception in interrupt context

commit d198d499148a0c64a41b3aba9e7dd43772832b91 upstream.

If vmalloc page_fault happens inside of interrupt handler with interrupts
disabled then on exit path from exception handler when there is no pending
interrupts, the following code (arch/x86/xen/xen-asm_32.S:112):

cmpw $0x0001, XEN_vcpu_info_pending(%eax)
sete XEN_vcpu_info_mask(%eax)

will enable interrupts even if they has been previously disabled according to
eflags from the bounce frame (arch/x86/xen/xen-asm_32.S:99)

testb $X86_EFLAGS_IF>>8, 8+1+ESP_OFFSET(%esp)
setz XEN_vcpu_info_mask(%eax)

Solution is in setting XEN_vcpu_info_mask only when it should be set
according to
cmpw $0x0001, XEN_vcpu_info_pending(%eax)
but not clearing it if there isn't any pending events.

Reproducer for bug is attached to RHBZ 707552

Signed-off-by: Igor Mammedov <imammedo@redhat.com>
Acked-by: Jeremy Fitzhardinge <jeremy@goop.org>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agommc: sdhci-s3c: Fix mmc card I/O problem
Girish K S [Fri, 26 Aug 2011 09:28:18 +0000 (14:58 +0530)]
mmc: sdhci-s3c: Fix mmc card I/O problem

commit 49bb1e619568ec84785ceb366f07db2a6f0b64cc upstream.

This patch fixes the problem in sdhci-s3c host driver for Samsung Soc's.
During the card identification stage the mmc core driver enumerates for
the best bus width in combination with the highest available data rate.
It starts enumerating from the highest bus width (8) to lowest width (1).

In case of few MMC cards the 4-bit bus enumeration fails and tries
the 1-bit bus enumeration. When switched to 1-bit bus mode the host driver
has to clear the previous bus width setting and apply the new setting.

The current patch will clear the previous bus mode and apply the new
mode setting.

Signed-off-by: Girish K S <girish.shivananjappa@linaro.org>
Acked-by: Jaehoon Chung <jh80.chung@samsung.com>
Signed-off-by: Chris Ball <cjb@laptop.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agommc: core: use non-reentrant workqueue for clock gating
Mika Westerberg [Thu, 18 Aug 2011 12:23:49 +0000 (15:23 +0300)]
mmc: core: use non-reentrant workqueue for clock gating

commit 50a50f9248497484c678631a9c1a719f1aaeab79 upstream.

The default multithread workqueue can cause the same work to be executed
concurrently on a different CPUs. This isn't really suitable for clock
gating as it might already gated the clock and gating it twice results both
host->clk_old and host->ios.clock to be set to 0.

To prevent this from happening we use system_nrt_wq instead.

Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com>
Reviewed-by: Linus Walleij <linus.walleij@linaro.org>
Tested-by: Chris Ball <cjb@laptop.org>
Signed-off-by: Chris Ball <cjb@laptop.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agommc: core: prevent aggressive clock gating racing with ios updates
Mika Westerberg [Thu, 18 Aug 2011 12:23:48 +0000 (15:23 +0300)]
mmc: core: prevent aggressive clock gating racing with ios updates

commit 778e277cb82411c9002ca28ccbd216c4d9eb9158 upstream.

We have seen at least two different races when clock gating kicks in in a
middle of ios structure update.

First one happens when ios->clock is changed outside of aggressive clock
gating framework, for example via mmc_set_clock(). The race might happen
when we run following code:

mmc_set_ios():
...
if (ios->clock > 0)
mmc_set_ungated(host);

Now if gating kicks in right after the condition check we end up setting
host->clk_gated to false even though we have just gated the clock. Next
time a request is started we try to ungate and restore the clock in
mmc_host_clk_hold(). However since we have host->clk_gated set to false the
original clock is not restored.

This eventually will cause the host controller to hang since its clock is
disabled while we are trying to issue a request. For example on Intel
Medfield platform we see:

[   13.818610] mmc2: Timeout waiting for hardware interrupt.
[   13.818698] sdhci: =========== REGISTER DUMP (mmc2)===========
[   13.818753] sdhci: Sys addr: 0x00000000 | Version:  0x00008901
[   13.818804] sdhci: Blk size: 0x00000000 | Blk cnt:  0x00000000
[   13.818853] sdhci: Argument: 0x00000000 | Trn mode: 0x00000000
[   13.818903] sdhci: Present:  0x1fff0000 | Host ctl: 0x00000001
[   13.818951] sdhci: Power:    0x0000000d | Blk gap:  0x00000000
[   13.819000] sdhci: Wake-up:  0x00000000 | Clock:    0x00000000
[   13.819049] sdhci: Timeout:  0x00000000 | Int stat: 0x00000000
[   13.819098] sdhci: Int enab: 0x00ff00c3 | Sig enab: 0x00ff00c3
[   13.819147] sdhci: AC12 err: 0x00000000 | Slot int: 0x00000000
[   13.819196] sdhci: Caps:     0x6bee32b2 | Caps_1:   0x00000000
[   13.819245] sdhci: Cmd:      0x00000000 | Max curr: 0x00000000
[   13.819292] sdhci: Host ctl2: 0x00000000
[   13.819331] sdhci: ADMA Err: 0x00000000 | ADMA Ptr: 0x00000000
[   13.819377] sdhci: ===========================================
[   13.919605] mmc2: Reset 0x2 never completed.

and it never recovers.

Second race might happen while running mmc_power_off():

static void mmc_power_off(struct mmc_host *host)
{
host->ios.clock = 0;
host->ios.vdd = 0;

[ clock gating kicks in here ]

/*
 * Reset ocr mask to be the highest possible voltage supported for
 * this mmc host. This value will be used at next power up.
 */
host->ocr = 1 << (fls(host->ocr_avail) - 1);

if (!mmc_host_is_spi(host)) {
host->ios.bus_mode = MMC_BUSMODE_OPENDRAIN;
host->ios.chip_select = MMC_CS_DONTCARE;
}
host->ios.power_mode = MMC_POWER_OFF;
host->ios.bus_width = MMC_BUS_WIDTH_1;
host->ios.timing = MMC_TIMING_LEGACY;
mmc_set_ios(host);
}

If the clock gating worker kicks in while we are only partially updated the
ios structure the host controller gets incomplete ios and might not work as
supposed. Again on Intel Medfield platform we get:

[    4.185349] kernel BUG at drivers/mmc/host/sdhci.c:1155!
[    4.185422] invalid opcode: 0000 [#1] PREEMPT SMP
[    4.185509] Modules linked in:
[    4.185565]
[    4.185608] Pid: 4, comm: kworker/0:0 Not tainted 3.0.0+ #240 Intel Corporation Medfield/iCDKA
[    4.185742] EIP: 0060:[<c136364e>] EFLAGS: 00010083 CPU: 0
[    4.185827] EIP is at sdhci_set_power+0x3e/0xd0
[    4.185891] EAX: f5ff98e0 EBX: f5ff98e0 ECX: 00000000 EDX: 00000001
[    4.185970] ESI: f5ff977c EDI: f5ff9904 EBP: f644fe98 ESP: f644fe94
[    4.186049]  DS: 007b ES: 007b FS: 00d8 GS: 0000 SS: 0068
[    4.186125] Process kworker/0:0 (pid: 4, ti=f644e000 task=f644c0e0 task.ti=f644e000)
[    4.186219] Stack:
[    4.186257]  f5ff98e0 f644feb0 c1365173 00000282 f5ff9460 f5ff96e0 f5ff96e0 f644feec
[    4.186418]  c1355bd8 f644c0e0 c1499c3d f5ff96e0 f644fed4 00000006 f5ff96e0 00000286
[    4.186579]  f644fedc c107922b f644feec 00000286 f5ff9460 f5ff9700 f644ff10 c135839e
[    4.186739] Call Trace:
[    4.186802]  [<c1365173>] sdhci_set_ios+0x1c3/0x340
[    4.186883]  [<c1355bd8>] mmc_gate_clock+0x68/0x120
[    4.186963]  [<c1499c3d>] ? _raw_spin_unlock_irqrestore+0x4d/0x60
[    4.187052]  [<c107922b>] ? trace_hardirqs_on+0xb/0x10
[    4.187134]  [<c135839e>] mmc_host_clk_gate_delayed+0xbe/0x130
[    4.187219]  [<c105ec09>] ? process_one_work+0xf9/0x5b0
[    4.187300]  [<c135841d>] mmc_host_clk_gate_work+0xd/0x10
[    4.187379]  [<c105ec82>] process_one_work+0x172/0x5b0
[    4.187457]  [<c105ec09>] ? process_one_work+0xf9/0x5b0
[    4.187538]  [<c1358410>] ? mmc_host_clk_gate_delayed+0x130/0x130
[    4.187625]  [<c105f3c8>] worker_thread+0x118/0x330
[    4.187700]  [<c1496cee>] ? preempt_schedule+0x2e/0x50
[    4.187779]  [<c105f2b0>] ? rescuer_thread+0x1f0/0x1f0
[    4.187857]  [<c1062cf4>] kthread+0x74/0x80
[    4.187931]  [<c1062c80>] ? __init_kthread_worker+0x60/0x60
[    4.188015]  [<c149acfa>] kernel_thread_helper+0x6/0xd
[    4.188079] Code: 81 fa 00 00 04 00 0f 84 a7 00 00 00 7f 21 81 fa 80 00 00 00 0f 84 92 00 00 00 81 fa 00 00 0
[    4.188780] EIP: [<c136364e>] sdhci_set_power+0x3e/0xd0 SS:ESP 0068:f644fe94
[    4.188898] ---[ end trace a7b23eecc71777e4 ]---

This BUG() comes from the fact that ios.power_mode was still in previous
value (MMC_POWER_ON) and ios.vdd was set to zero.

We prevent these by inhibiting the clock gating while we update the ios
structure.

Both problems can be reproduced by simply running the device in a reboot
loop.

Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com>
Reviewed-by: Linus Walleij <linus.walleij@linaro.org>
Tested-by: Chris Ball <cjb@laptop.org>
Signed-off-by: Chris Ball <cjb@laptop.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agommc: rename mmc_host_clk_{ungate|gate} to mmc_host_clk_{hold|release}
Mika Westerberg [Thu, 18 Aug 2011 12:23:47 +0000 (15:23 +0300)]
mmc: rename mmc_host_clk_{ungate|gate} to mmc_host_clk_{hold|release}

commit 08c14071fda4e69abb9d5b1566651cd092b158d3 upstream.

As per suggestion by Linus Walleij:

  > If you think the names of the functions are confusing then
  > you may rename them, say like this:
  >
  > mmc_host_clk_ungate() -> mmc_host_clk_hold()
  > mmc_host_clk_gate() -> mmc_host_clk_release()
  >
  > Which would make the usecases more clear

(This is CC'd to stable@ because the next two patches, which fix
observable races, depend on it.)

Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com>
Reviewed-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: Chris Ball <cjb@laptop.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agox86, perf: Check that current->mm is alive before getting user callchain
Andrey Vagin [Tue, 30 Aug 2011 08:32:36 +0000 (12:32 +0400)]
x86, perf: Check that current->mm is alive before getting user callchain

commit 20afc60f892d285fde179ead4b24e6a7938c2f1b upstream.

An event may occur when an mm is already released.

I added an event in dequeue_entity() and caught a panic with
the following backtrace:

[  434.421110] BUG: unable to handle kernel NULL pointer dereference at 0000000000000050
[  434.421258] IP: [<ffffffff810464ac>] __get_user_pages_fast+0x9c/0x120
...
[  434.421258] Call Trace:
[  434.421258]  [<ffffffff8101ae81>] copy_from_user_nmi+0x51/0xf0
[  434.421258]  [<ffffffff8109a0d5>] ? sched_clock_local+0x25/0x90
[  434.421258]  [<ffffffff8101b048>] perf_callchain_user+0x128/0x170
[  434.421258]  [<ffffffff811154cd>] ? __perf_event_header__init_id+0xed/0x100
[  434.421258]  [<ffffffff81116690>] perf_prepare_sample+0x200/0x280
[  434.421258]  [<ffffffff81118da8>] __perf_event_overflow+0x1b8/0x290
[  434.421258]  [<ffffffff81065240>] ? tg_shares_up+0x0/0x670
[  434.421258]  [<ffffffff8104fe1a>] ? walk_tg_tree+0x6a/0xb0
[  434.421258]  [<ffffffff81118f44>] perf_swevent_overflow+0xc4/0xf0
[  434.421258]  [<ffffffff81119150>] do_perf_sw_event+0x1e0/0x250
[  434.421258]  [<ffffffff81119204>] perf_tp_event+0x44/0x70
[  434.421258]  [<ffffffff8105701f>] ftrace_profile_sched_block+0xdf/0x110
[  434.421258]  [<ffffffff8106121d>] dequeue_entity+0x2ad/0x2d0
[  434.421258]  [<ffffffff810614ec>] dequeue_task_fair+0x1c/0x60
[  434.421258]  [<ffffffff8105818a>] dequeue_task+0x9a/0xb0
[  434.421258]  [<ffffffff810581e2>] deactivate_task+0x42/0xe0
[  434.421258]  [<ffffffff814bc019>] thread_return+0x191/0x808
[  434.421258]  [<ffffffff81098a44>] ? switch_task_namespaces+0x24/0x60
[  434.421258]  [<ffffffff8106f4c4>] do_exit+0x464/0x910
[  434.421258]  [<ffffffff8106f9c8>] do_group_exit+0x58/0xd0
[  434.421258]  [<ffffffff8106fa57>] sys_exit_group+0x17/0x20
[  434.421258]  [<ffffffff8100b202>] system_call_fastpath+0x16/0x1b

Signed-off-by: Andrey Vagin <avagin@openvz.org>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1314693156-24131-1-git-send-email-avagin@openvz.org
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agosched: Fix a memory leak in __sdt_free()
WANG Cong [Thu, 18 Aug 2011 12:36:57 +0000 (20:36 +0800)]
sched: Fix a memory leak in __sdt_free()

commit feff8fa0075bdfd43c841e9d689ed81adda988d6 upstream.

This patch fixes the following memory leak:

unreferenced object 0xffff880107266800 (size 512):
  comm "sched-powersave", pid 3718, jiffies 4323097853 (age 27495.450s)
  hex dump (first 32 bytes):
    00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00  ................
    00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00  ................
  backtrace:
    [<ffffffff81133940>] create_object+0x187/0x28b
    [<ffffffff814ac103>] kmemleak_alloc+0x73/0x98
    [<ffffffff811232ba>] __kmalloc_node+0x104/0x159
    [<ffffffff81044b98>] kzalloc_node.clone.97+0x15/0x17
    [<ffffffff8104cb90>] build_sched_domains+0xb7/0x7f3
    [<ffffffff8104d4df>] partition_sched_domains+0x1db/0x24a
    [<ffffffff8109ee4a>] do_rebuild_sched_domains+0x3b/0x47
    [<ffffffff810a00c7>] rebuild_sched_domains+0x10/0x12
    [<ffffffff8104d5ba>] sched_power_savings_store+0x6c/0x7b
    [<ffffffff8104d5df>] sched_mc_power_savings_store+0x16/0x18
    [<ffffffff8131322c>] sysdev_class_store+0x20/0x22
    [<ffffffff81193876>] sysfs_write_file+0x108/0x144
    [<ffffffff81135b10>] vfs_write+0xaf/0x102
    [<ffffffff81135d23>] sys_write+0x4d/0x74
    [<ffffffff814c8a42>] system_call_fastpath+0x16/0x1b
    [<ffffffffffffffff>] 0xffffffffffffffff

Signed-off-by: WANG Cong <amwang@redhat.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1313671017-4112-1-git-send-email-amwang@redhat.com
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agosched: Move blk_schedule_flush_plug() out of __schedule()
Thomas Gleixner [Wed, 22 Jun 2011 17:47:01 +0000 (19:47 +0200)]
sched: Move blk_schedule_flush_plug() out of __schedule()

commit 9c40cef2b799f9b5e7fa5de4d2ad3a0168ba118c upstream.

There is no real reason to run blk_schedule_flush_plug() with
interrupts and preemption disabled.

Move it into schedule() and call it when the task is going voluntarily
to sleep. There might be false positives when the task is woken
between that call and actually scheduling, but that's not really
different from being woken immediately after switching away.

This fixes a deadlock in the scheduler where the
blk_schedule_flush_plug() callchain enables interrupts and thereby
allows a wakeup to happen of the task that's going to sleep.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Tejun Heo <tj@kernel.org>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Link: http://lkml.kernel.org/n/tip-dwfxtra7yg1b5r65m32ywtct@git.kernel.org
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agosched: Separate the scheduler entry for preemption
Thomas Gleixner [Wed, 22 Jun 2011 17:47:00 +0000 (19:47 +0200)]
sched: Separate the scheduler entry for preemption

commit c259e01a1ec90063042f758e409cd26b2a0963c8 upstream.

Block-IO and workqueues call into notifier functions from the
scheduler core code with interrupts and preemption disabled. These
calls should be made before entering the scheduler core.

To simplify this, separate the scheduler core code into
__schedule(). __schedule() is directly called from the places which
set PREEMPT_ACTIVE and from schedule(). This allows us to add the work
checks into schedule(), so they are only called when a task voluntary
goes to sleep.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Tejun Heo <tj@kernel.org>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Link: http://lkml.kernel.org/r/20110622174918.813258321@linutronix.de
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agortc: Fix RTC PIE frequency limit
John Stultz [Fri, 22 Jul 2011 09:12:51 +0000 (09:12 +0000)]
rtc: Fix RTC PIE frequency limit

commit 938f97bcf1bdd1b681d5d14d1d7117a2e22d4434 upstream.

Thomas earlier submitted a fix to limit the RTC PIE freq, but
picked 5000Hz out of the air. Willy noticed that we should
instead use the 8192Hz max from the rtc man documentation.

Cc: Willy Tarreau <w@1wt.eu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: John Stultz <john.stultz@linaro.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoalarmtimers: Avoid possible denial of service with high freq periodic timers
John Stultz [Wed, 10 Aug 2011 17:26:09 +0000 (10:26 -0700)]
alarmtimers: Avoid possible denial of service with high freq periodic timers

commit 6af7e471e5a7746b8024d70b4363d3dfe41d36b8 upstream.

Its possible to jam up the alarm timers by setting very small interval
timers, which will cause the alarmtimer subsystem to spend all of its time
firing and restarting timers. This can effectivly lock up a box.

A deeper fix is needed, closely mimicking the hrtimer code, but for now
just cap the interval to 100us to avoid userland hanging the system.

CC: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: John Stultz <john.stultz@linaro.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoalarmtimers: Memset itimerspec passed into alarm_timer_get
John Stultz [Thu, 4 Aug 2011 14:51:56 +0000 (07:51 -0700)]
alarmtimers: Memset itimerspec passed into alarm_timer_get

commit ea7802f630d356acaf66b3c0b28c00a945fc35dc upstream.

Following common_timer_get, zero out the itimerspec passed in.

CC: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: John Stultz <john.stultz@linaro.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoalarmtimers: Avoid possible null pointer traversal
John Stultz [Thu, 4 Aug 2011 14:25:35 +0000 (07:25 -0700)]
alarmtimers: Avoid possible null pointer traversal

commit 971c90bfa2f0b4fe52d6d9002178d547706f1343 upstream.

We don't check if old_setting is non null before assigning it, so
correct this.

CC: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: John Stultz <john.stultz@linaro.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoMXC: iomux-v3: correct NO_PAD_CTRL definition
Troy Kisky [Fri, 24 Jun 2011 17:52:56 +0000 (10:52 -0700)]
MXC: iomux-v3: correct NO_PAD_CTRL definition

commit 425933b30b0ccfac58065bca6c853ea627443cdf upstream.

iomux-v3.c uses NO_PAD_CTRL as a 32 bit value
so it should not be shifted left by MUX_PAD_CTRL_SHIFT(41)

Previously, anything requesting NO_PAD_CTRL would get
their pad control register set to 0.

Since it is a pad control mask, place it with the other mask values.

Signed-off-by: Troy Kisky <troy.kisky@boundarydevices.com>
Acked-by: Lothar Waßmann <LW@KARO-electronics.de>
Tested-by: Lothar Waßmann <LW@KARO-electronics.de>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
Cc: John Ogness <john.ogness@linutronix.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoigb: fix WOL on second port of i350 device
Carolyn Wyborny [Thu, 7 Jul 2011 00:24:56 +0000 (00:24 +0000)]
igb: fix WOL on second port of i350 device

commit 6d337dce664b6872ddf1655f6b1fcab76ce35b08 upstream.

This patch fixes a problem where WOL would fail on second port of i350
device.

Reported-by: Martin Wilck <martin.wilck@ts.fujitsu.com>
Reported-by: Stefan Assmann<sassmann@redhat.com>
Signed-off-by: Carolyn Wyborny <carolyn.wyborny@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agomm: page allocator: reconsider zones for allocation after direct reclaim
Mel Gorman [Tue, 26 Jul 2011 00:12:30 +0000 (17:12 -0700)]
mm: page allocator: reconsider zones for allocation after direct reclaim

commit 76d3fbf8fbf6cc78ceb63549e0e0c5bc8a88f838 upstream.

With zone_reclaim_mode enabled, it's possible for zones to be considered
full in the zonelist_cache so they are skipped in the future.  If the
process enters direct reclaim, the ZLC may still consider zones to be full
even after reclaiming pages.  Reconsider all zones for allocation if
direct reclaim returns successfully.

Signed-off-by: Mel Gorman <mgorman@suse.de>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Christoph Lameter <cl@linux.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Stefan Priebe <s.priebe@profihost.ag>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agomm: page allocator: initialise ZLC for first zone eligible for zone_reclaim
Mel Gorman [Tue, 26 Jul 2011 00:12:29 +0000 (17:12 -0700)]
mm: page allocator: initialise ZLC for first zone eligible for zone_reclaim

commit cd38b115d5ad79b0100ac6daa103c4fe2c50a913 upstream.

There have been a small number of complaints about significant stalls
while copying large amounts of data on NUMA machines reported on a
distribution bugzilla.  In these cases, zone_reclaim was enabled by
default due to large NUMA distances.  In general, the complaints have not
been about the workload itself unless it was a file server (in which case
the recommendation was disable zone_reclaim).

The stalls are mostly due to significant amounts of time spent scanning
the preferred zone for pages to free.  After a failure, it might fallback
to another node (as zonelists are often node-ordered rather than
zone-ordered) but stall quickly again when the next allocation attempt
occurs.  In bad cases, each page allocated results in a full scan of the
preferred zone.

Patch 1 checks the preferred zone for recent allocation failure
        which is particularly important if zone_reclaim has failed
        recently.  This avoids rescanning the zone in the near future and
        instead falling back to another node.  This may hurt node locality
        in some cases but a failure to zone_reclaim is more expensive than
        a remote access.

Patch 2 clears the zlc information after direct reclaim.
        Otherwise, zone_reclaim can mark zones full, direct reclaim can
        reclaim enough pages but the zone is still not considered for
        allocation.

This was tested on a 24-thread 2-node x86_64 machine.  The tests were
focused on large amounts of IO.  All tests were bound to the CPUs on
node-0 to avoid disturbances due to processes being scheduled on different
nodes.  The kernels tested are

3.0-rc6-vanilla Vanilla 3.0-rc6
zlcfirst Patch 1 applied
zlcreconsider Patches 1+2 applied

FS-Mark
./fs_mark  -d  /tmp/fsmark-10813  -D  100  -N  5000  -n  208  -L  35  -t  24  -S0  -s  524288
                fsmark-3.0-rc6       3.0-rc6        3.0-rc6
                   vanilla  zlcfirs  zlcreconsider
Files/s  min          54.90 ( 0.00%)       49.80 (-10.24%)       49.10 (-11.81%)
Files/s  mean        100.11 ( 0.00%)      135.17 (25.94%)      146.93 (31.87%)
Files/s  stddev       57.51 ( 0.00%)      138.97 (58.62%)      158.69 (63.76%)
Files/s  max         361.10 ( 0.00%)      834.40 (56.72%)      802.40 (55.00%)
Overhead min       76704.00 ( 0.00%)    76501.00 ( 0.27%)    77784.00 (-1.39%)
Overhead mean    1485356.51 ( 0.00%)  1035797.83 (43.40%)  1594680.26 (-6.86%)
Overhead stddev  1848122.53 ( 0.00%)   881489.88 (109.66%)  1772354.90 ( 4.27%)
Overhead max     7989060.00 ( 0.00%)  3369118.00 (137.13%) 10135324.00 (-21.18%)
MMTests Statistics: duration
User/Sys Time Running Test (seconds)        501.49    493.91    499.93
Total Elapsed Time (seconds)               2451.57   2257.48   2215.92

MMTests Statistics: vmstat
Page Ins                                       46268       63840       66008
Page Outs                                   90821596    90671128    88043732
Swap Ins                                           0           0           0
Swap Outs                                          0           0           0
Direct pages scanned                        13091697     8966863     8971790
Kswapd pages scanned                               0     1830011     1831116
Kswapd pages reclaimed                             0     1829068     1829930
Direct pages reclaimed                      13037777     8956828     8648314
Kswapd efficiency                               100%         99%         99%
Kswapd velocity                                0.000     810.643     826.346
Direct efficiency                                99%         99%         96%
Direct velocity                             5340.128    3972.068    4048.788
Percentage direct scans                         100%         83%         83%
Page writes by reclaim                             0           3           0
Slabs scanned                                 796672      720640      720256
Direct inode steals                          7422667     7160012     7088638
Kswapd inode steals                                0     1736840     2021238

Test completes far faster with a large increase in the number of files
created per second.  Standard deviation is high as a small number of
iterations were much higher than the mean.  The number of pages scanned by
zone_reclaim is reduced and kswapd is used for more work.

LARGE DD
                3.0-rc6       3.0-rc6       3.0-rc6
                    vanilla     zlcfirst     zlcreconsider
download tar           59 ( 0.00%)   59 ( 0.00%)   55 ( 7.27%)
dd source files       527 ( 0.00%)  296 (78.04%)  320 (64.69%)
delete source          36 ( 0.00%)   19 (89.47%)   20 (80.00%)
MMTests Statistics: duration
User/Sys Time Running Test (seconds)        125.03    118.98    122.01
Total Elapsed Time (seconds)                624.56    375.02    398.06

MMTests Statistics: vmstat
Page Ins                                     3594216      439368      407032
Page Outs                                   23380832    23380488    23377444
Swap Ins                                           0           0           0
Swap Outs                                          0         436         287
Direct pages scanned                        17482342    69315973    82864918
Kswapd pages scanned                               0      519123      575425
Kswapd pages reclaimed                             0      466501      522487
Direct pages reclaimed                       5858054     2732949     2712547
Kswapd efficiency                               100%         89%         90%
Kswapd velocity                                0.000    1384.254    1445.574
Direct efficiency                                33%          3%          3%
Direct velocity                            27991.453  184832.737  208171.929
Percentage direct scans                         100%         99%         99%
Page writes by reclaim                             0        5082       13917
Slabs scanned                                  17280       29952       35328
Direct inode steals                           115257     1431122      332201
Kswapd inode steals                                0           0      979532

This test downloads a large tarfile and copies it with dd a number of
times - similar to the most recent bug report I've dealt with.  Time to
completion is reduced.  The number of pages scanned directly is still
disturbingly high with a low efficiency but this is likely due to the
number of dirty pages encountered.  The figures could probably be improved
with more work around how kswapd is used and how dirty pages are handled
but that is separate work and this result is significant on its own.

Streaming Mapped Writer
MMTests Statistics: duration
User/Sys Time Running Test (seconds)        124.47    111.67    112.64
Total Elapsed Time (seconds)               2138.14   1816.30   1867.56

MMTests Statistics: vmstat
Page Ins                                       90760       89124       89516
Page Outs                                  121028340   120199524   120736696
Swap Ins                                           0          86          55
Swap Outs                                          0           0           0
Direct pages scanned                       114989363    96461439    96330619
Kswapd pages scanned                        56430948    56965763    57075875
Kswapd pages reclaimed                      27743219    27752044    27766606
Direct pages reclaimed                         49777       46884       36655
Kswapd efficiency                                49%         48%         48%
Kswapd velocity                            26392.541   31363.631   30561.736
Direct efficiency                                 0%          0%          0%
Direct velocity                            53780.091   53108.759   51581.004
Percentage direct scans                          67%         62%         62%
Page writes by reclaim                           385         122        1513
Slabs scanned                                  43008       39040       42112
Direct inode steals                                0          10           8
Kswapd inode steals                              733         534         477

This test just creates a large file mapping and writes to it linearly.
Time to completion is again reduced.

The gains are mostly down to two things.  In many cases, there is less
scanning as zone_reclaim simply gives up faster due to recent failures.
The second reason is that memory is used more efficiently.  Instead of
scanning the preferred zone every time, the allocator falls back to
another zone and uses it instead improving overall memory utilisation.

This patch: initialise ZLC for first zone eligible for zone_reclaim.

The zonelist cache (ZLC) is used among other things to record if
zone_reclaim() failed for a particular zone recently.  The intention is to
avoid a high cost scanning extremely long zonelists or scanning within the
zone uselessly.

Currently the zonelist cache is setup only after the first zone has been
considered and zone_reclaim() has been called.  The objective was to avoid
a costly setup but zone_reclaim is itself quite expensive.  If it is
failing regularly such as the first eligible zone having mostly mapped
pages, the cost in scanning and allocation stalls is far higher than the
ZLC initialisation step.

This patch initialises ZLC before the first eligible zone calls
zone_reclaim().  Once initialised, it is checked whether the zone failed
zone_reclaim recently.  If it has, the zone is skipped.  As the first zone
is now being checked, additional care has to be taken about zones marked
full.  A zone can be marked "full" because it should not have enough
unmapped pages for zone_reclaim but this is excessive as direct reclaim or
kswapd may succeed where zone_reclaim fails.  Only mark zones "full" after
zone_reclaim fails if it failed to reclaim enough pages after scanning.

Signed-off-by: Mel Gorman <mgorman@suse.de>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Christoph Lameter <cl@linux.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Stefan Priebe <s.priebe@profihost.ag>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agodrm/radeon/kms: make sure pci max read request size is valid on evergreen+ (v2)
Alex Deucher [Thu, 1 Sep 2011 17:46:15 +0000 (17:46 +0000)]
drm/radeon/kms: make sure pci max read request size is valid on evergreen+ (v2)

commit d054ac16eeb658bccadb06b12c39cee22243b10f upstream.

If the bios or OS sets the pci max read request size to 0 or an
invalid value (6,7), it can result in a hang or slowdown.  Check
and set it to something sane if it's invalid.

Fixes:
https://bugzilla.kernel.org/show_bug.cgi?id=42162

v2: use pci reg defines from include/linux/pci_regs.h

Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agodrm/radeon/kms: set a default max_pixel_clock
Dave Airlie [Tue, 30 Aug 2011 19:22:04 +0000 (20:22 +0100)]
drm/radeon/kms: set a default max_pixel_clock

commit 9adceaa5b3d2480e2252c4a7f9c4bd7d66b8c4a2 upstream.

On some Power rv100 cards, we have no ATY OF table, but we have
no combios table either, and hence we refuse all modes on VGA-0
since we end up with a 0 max pixel clock.

Signed-off-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Alex Deucher <alexdeucher@gmail.com>
Reviewed-by: Jerome Glisse <jglisse@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agomd/linear: avoid corrupting structure while waiting for rcu_free to complete.
NeilBrown [Thu, 25 Aug 2011 04:43:53 +0000 (14:43 +1000)]
md/linear: avoid corrupting structure while waiting for rcu_free to complete.

commit 1b6afa17581027218088a18a9ceda600e0ddba7a upstream.

I don't know what I was thinking putting 'rcu' after a dynamically
sized array!  The array could still be in use when we call rcu_free()
(That is the point) so we mustn't corrupt it.

Signed-off-by: NeilBrown <neilb@suse.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoARM: 7014/1: cache-l2x0: Fix L2 Cache size calculation.
Srinivas Kandagatla [Mon, 15 Aug 2011 09:43:44 +0000 (10:43 +0100)]
ARM: 7014/1: cache-l2x0: Fix L2 Cache size calculation.

commit 43c734be5571a4daad9f0a3e0b3229a1c0049917 upstream.

This patch fixes L2 Cache size calculations for L2C-210, L2C-310 and
PL310, by changing the L2X0_AUX_CTRL_WAY_SIZE_MASK from 2 bits to 3
bits.

The Auxiliary Control Register for L2C-210, L2C-310 and PL310 has 3bits
[19:17] for Way size, however the existing code only uses 2 bits to
get this value. This results in incorrect cachesize calculations.

It also results in performing operations on the whole cache when we
erroneously decide that the range is big enough (due to l2x0_size being
too small) and also prints incorrect cachesize.

Signed-off-by: Srinivas Kandagatla <srinivas.kandagatla@st.com>
Acked-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agodrm/radeon/kms: evergreen & ni reset SPI block on CP resume
Jerome Glisse [Wed, 24 Aug 2011 20:00:17 +0000 (20:00 +0000)]
drm/radeon/kms: evergreen & ni reset SPI block on CP resume

commit a49a50dad48586d42ebac1a6730c3a3cd5603421 upstream.

For some reason SPI block is in broken state after module
unloading. This lead to broken rendering after reloading
module. Fix this by reseting SPI block in CP resume function

Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agodrm/radeon/kms: add s/r quirk for Compaq Presario V5245EU
Alex Deucher [Mon, 29 Aug 2011 14:55:25 +0000 (14:55 +0000)]
drm/radeon/kms: add s/r quirk for Compaq Presario V5245EU

commit 302a8e8b06d312dcb3b718dfeb42aa912b5f426b upstream.

Fixes resume on Compaq Presario V5245EU.

Fixes:
https://bugzilla.kernel.org/show_bug.cgi?id=41642

Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agosparc64: Only Panther cheetah+ chips have POPC.
David S. Miller [Tue, 30 Aug 2011 04:14:29 +0000 (21:14 -0700)]
sparc64: Only Panther cheetah+ chips have POPC.

commit 1a8e0da5937a6c87807083baa318cf8f98dac9aa upstream.

Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoregulator: tps65910: Add missing breaks in switch/case
Axel Lin [Sun, 10 Jul 2011 13:44:09 +0000 (21:44 +0800)]
regulator: tps65910: Add missing breaks in switch/case

commit d04156bca629740a661fd0738cd69ba1f08b2b20 upstream.

Also add a default case in tps65910_list_voltage_dcdc to silence
'volt' may be used uninitialized in this function warning.

Signed-off-by: Axel Lin <axel.lin@gmail.com>
Acked-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
Signed-off-by: Liam Girdwood <lrg@slimlogic.co.uk>
Cc: Johan Hovold <jhovold@gmail.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agosparc32,sun4d: Change IPI IRQ level to prevent collision between IPI and timer interrupt
Kjetil Oftedal [Sun, 28 Aug 2011 22:16:28 +0000 (00:16 +0200)]
sparc32,sun4d: Change IPI IRQ level to prevent collision between IPI and timer interrupt

commit 38f7f8f05e8239e9871f7e1c4b0a842080e85315 upstream.

On Sun4d systems running in SMP mode, IRQ 14 is used for timer interrupts
and has a specialized interrupt handler. IPI is currently set to use IRQ 14
as well, which causes it to trigger the timer interrupt handler, and not the
IPI interrupt handler.

The IPI interrupt is therefore changed to IRQ 13, which is the highest
normally handled interrupt. This IRQ is also used for SBUS interrupts,
however there is nothing in the IPI/SBUS interrupt handlers that indicate
that they will not handle sharing the interrupt.
(IRQ 13 is indicated as audio interrupt, which is unlikely to be found in a
sun4d system)

Signed-off-by: Kjetil Oftedal <oftedal@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agosparc: fix array bounds error setting up PCIC NMI trap
Ian Campbell [Wed, 17 Aug 2011 22:14:57 +0000 (22:14 +0000)]
sparc: fix array bounds error setting up PCIC NMI trap

commit 4a0342ca8e8150bd47e7118a76e300692a1b6b7b upstream.

  CC      arch/sparc/kernel/pcic.o
arch/sparc/kernel/pcic.c: In function 'pcic_probe':
arch/sparc/kernel/pcic.c:359:33: error: array subscript is above array bounds [-Werror=array-bounds]
arch/sparc/kernel/pcic.c:359:8: error: array subscript is above array bounds [-Werror=array-bounds]
arch/sparc/kernel/pcic.c:360:33: error: array subscript is above array bounds [-Werror=array-bounds]
arch/sparc/kernel/pcic.c:360:8: error: array subscript is above array bounds [-Werror=array-bounds]
arch/sparc/kernel/pcic.c:361:33: error: array subscript is above array bounds [-Werror=array-bounds]
arch/sparc/kernel/pcic.c:361:8: error: array subscript is above array bounds [-Werror=array-bounds]
cc1: all warnings being treated as errors

I'm not particularly familiar with sparc but t_nmi (defined in head_32.S via
the TRAP_ENTRY macro) and pcic_nmi_trap_patch (defined in entry.S) both appear
to be 4 instructions long and I presume from the usage that instructions are
int sized.

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: sparclinux@vger.kernel.org
Reviewed-by: Sam Ravnborg <sam@ravnborg.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agosparc64: Set HAVE_C_RECORDMCOUNT
David S. Miller [Mon, 15 Aug 2011 21:45:17 +0000 (14:45 -0700)]
sparc64: Set HAVE_C_RECORDMCOUNT

[ Upstream commit 178a29600340bef5b13cd4157053679debe35351 ]

Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agosparc: Allow handling signals when stack is corrupted.
David S. Miller [Sun, 21 Aug 2011 00:14:54 +0000 (17:14 -0700)]
sparc: Allow handling signals when stack is corrupted.

commit 5598473a5b40c47a8c5349dd2c2630797169cf1a upstream.

If we can't push the pending register windows onto the user's stack,
we disallow signal delivery even if the signal would be delivered on a
valid seperate signal stack.

Add a register window save area in the signal frame, and store any
unsavable windows there.

On sigreturn, if any windows are still queued up in the signal frame,
try to push them back onto the stack and if that fails we kill the
process immediately.

This allows the debug/tst-longjmp_chk2 glibc test case to pass.

Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agosparc32: unbreak arch_write_unlock()
Mikael Pettersson [Mon, 15 Aug 2011 10:11:50 +0000 (10:11 +0000)]
sparc32: unbreak arch_write_unlock()

commit 3f6aa0b113846a8628baa649af422cfc6fb1d786 upstream.

The sparc32 version of arch_write_unlock() is just a plain assignment.
Unfortunately this allows the compiler to schedule side-effects in a
protected region to occur after the HW-level unlock, which is broken.
E.g., the following trivial test case gets miscompiled:

#include <linux/spinlock.h>
rwlock_t lock;
int counter;
void foo(void) { write_lock(&lock); ++counter; write_unlock(&lock); }

Fixed by adding a compiler memory barrier to arch_write_unlock().  The
sparc64 version combines the barrier and assignment into a single asm(),
and implements the operation as a static inline, so that's what I did too.

Compile-tested with sparc32_defconfig + CONFIG_SMP=y.

Signed-off-by: Mikael Pettersson <mikpe@it.uu.se>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agosparc64: remove unnecessary macros from spinlock_64.h
Mikael Pettersson [Mon, 15 Aug 2011 10:10:31 +0000 (10:10 +0000)]
sparc64: remove unnecessary macros from spinlock_64.h

commit a0fba3eb059e73fed2d376a901f8117734c12f1f upstream.

The sparc64 spinlock_64.h contains a number of operations defined
first as static inline functions, and then as macros with the same
names and parameters as the functions.  Maybe this was needed at
some point in the past, but now nothing seems to depend on these
macros (checked with a recursive grep looking for ifdefs on these
names).  Other archs don't define these identity-macros.

So this patch deletes these unnecessary macros.

Compile-tested with sparc64_defconfig.

Signed-off-by: Mikael Pettersson <mikpe@it.uu.se>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agort2x00: fix crash in rt2800usb_get_txwi
Stanislaw Gruszka [Thu, 25 Aug 2011 15:14:26 +0000 (17:14 +0200)]
rt2x00: fix crash in rt2800usb_get_txwi

commit 674db1344443204b6ce3293f2df8fd1b7665deea upstream.

Patch should fix this oops:

BUG: unable to handle kernel NULL pointer dereference at 000000a0
IP: [<f81b30c9>] rt2800usb_get_txwi+0x19/0x70 [rt2800usb]
*pdpt = 0000000000000000 *pde = f000ff53f000ff53
Oops: 0000 [#1] SMP
Pid: 198, comm: kworker/u:3 Tainted: G        W   3.0.0-wl+ #9 LENOVO 6369CTO/6369CTO
EIP: 0060:[<f81b30c9>] EFLAGS: 00010283 CPU: 1
EIP is at rt2800usb_get_txwi+0x19/0x70 [rt2800usb]
EAX: 00000000 EBX: f465e140 ECX: f4494960 EDX: ef24c5f8
ESI: 810f21f5 EDI: f1da9960 EBP: f4581e80 ESP: f4581e70
 DS: 007b ES: 007b FS: 00d8 GS: 00e0 SS: 0068
Process kworker/u:3 (pid: 198, ti=f4580000 task=f4494960 task.ti=f4580000)
Call Trace:
 [<f804790f>] rt2800_txdone_entry+0x2f/0xf0 [rt2800lib]
 [<c045110d>] ? warn_slowpath_common+0x7d/0xa0
 [<f81b3a38>] ? rt2800usb_work_txdone+0x288/0x360 [rt2800usb]
 [<f81b3a38>] ? rt2800usb_work_txdone+0x288/0x360 [rt2800usb]
 [<f81b3a13>] rt2800usb_work_txdone+0x263/0x360 [rt2800usb]
 [<c046a8d6>] process_one_work+0x186/0x440
 [<c046a85a>] ? process_one_work+0x10a/0x440
 [<f81b37b0>] ? rt2800usb_probe_hw+0x120/0x120 [rt2800usb]
 [<c046c283>] worker_thread+0x133/0x310
 [<c04885db>] ? trace_hardirqs_on+0xb/0x10
 [<c046c150>] ? manage_workers+0x1e0/0x1e0
 [<c047054c>] kthread+0x7c/0x90
 [<c04704d0>] ? __init_kthread_worker+0x60/0x60
 [<c0826b42>] kernel_thread_helper+0x6/0x1

Oops might happen because we check rt2x00queue_empty(queue) twice,
but this condition can change and we can process entry in
rt2800_txdone_entry(), which was already processed by
rt2800usb_txdone_entry_check() -> rt2x00lib_txdone_noinfo() and
has nullify entry->skb .

Reported-by: Justin Piszcz <jpiszcz@lucidpixels.com>
Signed-off-by: Stanislaw Gruszka <sgruszka@redhat.com>
Acked-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agort2x00: fix crash in rt2800usb_write_tx_desc
Stanislaw Gruszka [Thu, 25 Aug 2011 15:14:24 +0000 (17:14 +0200)]
rt2x00: fix crash in rt2800usb_write_tx_desc

commit 4b1bfb7d2d125af6653d6c2305356b2677f79dc6 upstream.

Patch should fix this oops:

BUG: unable to handle kernel NULL pointer dereference at 000000a0
IP: [<f8e06078>] rt2800usb_write_tx_desc+0x18/0xc0 [rt2800usb]
*pdpt = 000000002408c001 *pde = 0000000024079067 *pte = 0000000000000000
Oops: 0000 [#1] SMP
EIP: 0060:[<f8e06078>] EFLAGS: 00010282 CPU: 0
EIP is at rt2800usb_write_tx_desc+0x18/0xc0 [rt2800usb]
EAX: 00000035 EBX: ef2bef10 ECX: 00000000 EDX: d40958a0
ESI: ef1865f8 EDI: ef1865f8 EBP: d4095878 ESP: d409585c
 DS: 007b ES: 007b FS: 00d8 GS: 00e0 SS: 0068
Call Trace:
 [<f8da5e85>] rt2x00queue_write_tx_frame+0x155/0x300 [rt2x00lib]
 [<f8da424c>] rt2x00mac_tx+0x7c/0x370 [rt2x00lib]
 [<c04882b2>] ? mark_held_locks+0x62/0x90
 [<c081f645>] ? _raw_spin_unlock_irqrestore+0x35/0x60
 [<c04884ba>] ? trace_hardirqs_on_caller+0x5a/0x170
 [<c04885db>] ? trace_hardirqs_on+0xb/0x10
 [<f8d618ac>] __ieee80211_tx+0x5c/0x1e0 [mac80211]
 [<f8d631fc>] ieee80211_tx+0xbc/0xe0 [mac80211]
 [<f8d63163>] ? ieee80211_tx+0x23/0xe0 [mac80211]
 [<f8d632e1>] ieee80211_xmit+0xc1/0x200 [mac80211]
 [<f8d63220>] ? ieee80211_tx+0xe0/0xe0 [mac80211]
 [<c0487d45>] ? lock_release_holdtime+0x35/0x1b0
 [<f8d63986>] ? ieee80211_subif_start_xmit+0x446/0x5f0 [mac80211]
 [<f8d637dd>] ieee80211_subif_start_xmit+0x29d/0x5f0 [mac80211]
 [<f8d63924>] ? ieee80211_subif_start_xmit+0x3e4/0x5f0 [mac80211]
 [<c0760188>] ? sock_setsockopt+0x6a8/0x6f0
 [<c0760000>] ? sock_setsockopt+0x520/0x6f0
 [<c076daef>] dev_hard_start_xmit+0x2ef/0x650

Oops might happen because we perform parallel putting new entries in a
queue (rt2x00queue_write_tx_frame()) and removing entries after
finishing transmitting (rt2800usb_work_txdone()). There are cases when
_txdone may process an entry that was not fully send and nullify
entry->skb .

To fix check in _txdone if entry has flags that indicate pending
transmission and wait until flags get cleared.

Reported-by: Justin Piszcz <jpiszcz@lucidpixels.com>
Signed-off-by: Stanislaw Gruszka <sgruszka@redhat.com>
Acked-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoatm: br2684: Fix oops due to skb->dev being NULL
Daniel Schwierzeck [Fri, 19 Aug 2011 12:04:20 +0000 (12:04 +0000)]
atm: br2684: Fix oops due to skb->dev being NULL

commit fbe5e29ec1886967255e76946aaf537b8cc9b81e upstream.

This oops have been already fixed with commit

    27141666b69f535a4d63d7bc6d9e84ee5032f82a

    atm: [br2684] Fix oops due to skb->dev being NULL

    It happens that if a packet arrives in a VC between the call to open it on
    the hardware and the call to change the backend to br2684, br2684_regvcc
    processes the packet and oopses dereferencing skb->dev because it is
    NULL before the call to br2684_push().

but have been introduced again with commit

    b6211ae7f2e56837c6a4849316396d1535606e90

    atm: Use SKB queue and list helpers instead of doing it by-hand.

Signed-off-by: Daniel Schwierzeck <daniel.schwierzeck@googlemail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agopata_via: disable ATAPI DMA on AVERATEC 3200
Tejun Heo [Thu, 4 Aug 2011 09:15:07 +0000 (11:15 +0200)]
pata_via: disable ATAPI DMA on AVERATEC 3200

commit 6d0e194d2eefcaab6dbdca1f639748660144acb5 upstream.

On AVERATEC 3200, pata_via causes memory corruption with ATAPI DMA,
which often leads to random kernel oops.  The cause of the problem is
not well understood yet and only small subset of machines using the
controller seem affected.  Blacklist ATAPI DMA on the machine.

Signed-off-by: Tejun Heo <tj@kernel.org>
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=11426
Reported-and-tested-by: Jim Bray <jimsantelmo@gmail.com>
Cc: Alan Cox <alan@linux.intel.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agosavagedb: Fix typo causing regression in savage4 series video chip detection
John Stanley [Thu, 4 Aug 2011 00:41:00 +0000 (20:41 -0400)]
savagedb: Fix typo causing regression in savage4 series video chip detection

commit 4b00e4b3940eabb38adeec0823751820fe2d6fda upstream.

Two additional savage4 variants were added, but the S3_SAVAGE4_SERIES
macro was incompletely modified, resulting in a false positive detection
of a savage4 card regardless of which savage card is actually present.

For non-savage4 series cards, such as a Savage/IX-MV card, this results
in garbled video and/or a hard-hang at boot time.  Fix this by changing
an '||' to an '&&' in the S3_SAVAGE4_SERIES macro.

Signed-off-by: John P. Stanley <jpsinthemix@verizon.net>
Reviewed-by: Tormod Volden <debian.tormod@gmail.com>
[ The macros have incomplete parenthesis too, but whatever ..  -Linus ]
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agort2x00: do not drop usb dev reference counter on suspend
Stanislaw Gruszka [Fri, 12 Aug 2011 12:02:04 +0000 (14:02 +0200)]
rt2x00: do not drop usb dev reference counter on suspend

commit 543cc38c8fe86deba4169977c61eb88491036837 upstream.

When hibernating ->resume may not be called by usb core, but disconnect
and probe instead, so we do not increase the counter after decreasing
it in ->supend. As a result we free memory early, and get crash when
unplugging usb dongle.

BUG: unable to handle kernel paging request at 6b6b6b9f
IP: [<c06909b0>] driver_sysfs_remove+0x10/0x30
*pdpt = 0000000034f21001 *pde = 0000000000000000
Pid: 20, comm: khubd Not tainted 3.1.0-rc1-wl+ #20 LENOVO 6369CTO/6369CTO
EIP: 0060:[<c06909b0>] EFLAGS: 00010202 CPU: 1
EIP is at driver_sysfs_remove+0x10/0x30
EAX: 6b6b6b6b EBX: f52bba34 ECX: 00000000 EDX: 6b6b6b6b
ESI: 6b6b6b6b EDI: c0a0ea20 EBP: f61c9e68 ESP: f61c9e64
 DS: 007b ES: 007b FS: 00d8 GS: 00e0 SS: 0068
Process khubd (pid: 20, ti=f61c8000 task=f6138270 task.ti=f61c8000)
Call Trace:
 [<c06909ef>] __device_release_driver+0x1f/0xa0
 [<c0690b20>] device_release_driver+0x20/0x40
 [<c068fd64>] bus_remove_device+0x84/0xe0
 [<c068e12a>] ? device_remove_attrs+0x2a/0x80
 [<c068e267>] device_del+0xe7/0x170
 [<c06d93d4>] usb_disconnect+0xd4/0x180
 [<c06d9d61>] hub_thread+0x691/0x1600
 [<c0473260>] ? wake_up_bit+0x30/0x30
 [<c0442a39>] ? complete+0x49/0x60
 [<c06d96d0>] ? hub_disconnect+0xd0/0xd0
 [<c06d96d0>] ? hub_disconnect+0xd0/0xd0
 [<c0472eb4>] kthread+0x74/0x80
 [<c0472e40>] ? kthread_worker_fn+0x150/0x150
 [<c0809b3e>] kernel_thread_helper+0x6/0x10

Signed-off-by: Stanislaw Gruszka <sgruszka@redhat.com>
Acked-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoath9k_hw: Fix STA (AR9485) bringup issue due to incorrect MAC address
Senthil Balasubramanian [Fri, 19 Aug 2011 13:13:06 +0000 (18:43 +0530)]
ath9k_hw: Fix STA (AR9485) bringup issue due to incorrect MAC address

commit b503c7a273c0a3018ad11ea8c513c639120afbf4 upstream.

Due to some recent optimization done in the way the mac address
bytes are written into the OTP memory, some AR9485 chipsets were
forced to use the first byte from the eeprom template and the
remaining bytes are read from OTP.

AR9485 happens to use generic eeprom template which has 0x1 as
the first byte causes issues in bringing up the card.

So fixed the eeprom template accordingly to address the issue.

Cc: Paul Stewart <pstew@google.com>
Signed-off-by: Senthil Balasubramanian <senthilb@qca.qualcomm.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agocarl9170: Fix mismatch in carl9170_op_set_key mutex lock-unlock
Alexey Khoroshilov [Tue, 23 Aug 2011 20:44:32 +0000 (00:44 +0400)]
carl9170: Fix mismatch in carl9170_op_set_key mutex lock-unlock

commit 66cb54bd24086b2d871a03035de9b0e79b2b725e upstream.

If is_main_vif(ar, vif) reports that we have to fall back
to software encryption, we goto err_softw; before locking ar->mutex.
As a result, we have unprotected call to carl9170_set_operating_mode
and unmatched mutex_unlock.

The patch fix the issue by adding mutex_lock before goto.

Found by Linux Driver Verification project (linuxtesting.org).

Signed-off-by: Alexey Khoroshilov <khoroshilov@ispras.ru>
Acked-By: Christian Lamparter <chunkeey@googlemail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoibmveth: Fix leak when recycling skb and hypervisor returns error
Anton Blanchard [Thu, 25 Aug 2011 00:56:15 +0000 (17:56 -0700)]
ibmveth: Fix leak when recycling skb and hypervisor returns error

commit c6f59d13e24187ff95427a9f4a5a7e14fb8faf5a upstream.

If h_add_logical_lan_buffer returns an error we need to free
the skb.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoath9k: Fix PS wrappers in ath9k_set_coverage_class
Mohammed Shafi Shajakhan [Wed, 24 Aug 2011 16:08:07 +0000 (21:38 +0530)]
ath9k: Fix PS wrappers in ath9k_set_coverage_class

commit 8b2a3827bb12430d932cd479b22d906baf08c212 upstream.

this callback is called during suspend/resume and also via iw command.
it configures parameters like sifs, slottime, acktimeout in
ath9k_hw_init_global_settings where few REG_READ, REG_RMW are also done
and hence the need for PS wrappers

Signed-off-by: Mohammed Shafi Shajakhan <mohammed@qca.qualcomm.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agosendmmsg/sendmsg: fix unsafe user pointer access
Mathieu Desnoyers [Thu, 25 Aug 2011 02:45:03 +0000 (19:45 -0700)]
sendmmsg/sendmsg: fix unsafe user pointer access

commit bc909d9ddbf7778371e36a651d6e4194b1cc7d4c upstream.

Dereferencing a user pointer directly from kernel-space without going
through the copy_from_user family of functions is a bad idea. Two of
such usages can be found in the sendmsg code path called from sendmmsg,
added by

commit c71d8ebe7a4496fb7231151cb70a6baa0cb56f9a upstream.
commit 5b47b8038f183b44d2d8ff1c7d11a5c1be706b34 in the 3.0-stable tree.

Usages are performed through memcmp() and memcpy() directly. Fix those
by using the already copied msg_sys structure instead of the __user *msg
structure. Note that msg_sys can be set to NULL by verify_compat_iovec()
or verify_iovec(), which requires additional NULL pointer checks.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: David Goulet <dgoulet@ev0ke.net>
CC: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
CC: Anton Blanchard <anton@samba.org>
CC: David S. Miller <davem@davemloft.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoxhci: Handle zero-length isochronous packets.
Sarah Sharp [Fri, 12 Aug 2011 17:23:01 +0000 (10:23 -0700)]
xhci: Handle zero-length isochronous packets.

commit 48df4a6fd8c40c0bbcbca2044f5f2bc75dcf6db1 upstream.

For a long time, the xHCI driver has had this note:
/* FIXME: Ignoring zero-length packets, can those happen? */

It turns out that, yes, there are drivers that need to queue zero-length
transfers for isochronous OUT transfers.  Without this patch, users will
see kernel hang messages when a driver attempts to enqueue an isochronous
URB with a zero length transfer (because count_isoc_trbs_needed will return
zero for that TD, xhci_td->last_trb will never be set, and updating the
dequeue pointer will cause an infinite loop).

MatÄ›j ran into this issue when using an NI Audio4DJ USB soundcard
with the snd-usb-caiaq driver.  See
https://bugzilla.kernel.org/show_bug.cgi?id=40702

Fix count_isoc_trbs_needed() to return 1 for zero-length transfers (thanks
Alan on the math help).  Update the various TRB field calculations to deal
with zero-length transfers.  We're still transferring one packet with a
zero-length data payload, so the total_packet_count should be 1. The
Transfer Burst Count (TBC) and Transfer Last Burst Packet Count (TLBPC)
fields should be set to zero.

This patch should be backported to kernels as old as 2.6.36.

Signed-off-by: Sarah Sharp <sarah.a.sharp@linux.intel.com>
Tested-by: Matěj Laitl <matej@laitl.cz>
Cc: Daniel Mack <zonque@gmail.com>
Cc: Alan Stern <stern@rowland.harvard.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoxhci: Remove TDs from TD lists when URBs are canceled.
Sarah Sharp [Tue, 2 Aug 2011 22:43:40 +0000 (15:43 -0700)]
xhci: Remove TDs from TD lists when URBs are canceled.

commit 585df1d90cb07a02ca6c7a7d339e56e46d50dafb upstream.

When a driver tries to cancel an URB, and the host controller is dying,
xhci_urb_dequeue will giveback the URB without removing the xhci_tds
that comprise that URB from the td_list or the cancelled_td_list.  This
can cause a race condition between the driver calling URB dequeue and
the stop endpoint command watchdog timer.

If the timer fires on a dying host, and a driver attempts to resubmit
while the watchdog timer has dropped the xhci->lock to giveback a
cancelled URB, URBs may be given back by the xhci_urb_dequeue() function.
At that point, the URB's priv pointer will be freed and set to NULL, but
the TDs will remain on the td_list.  This will cause an oops in
xhci_giveback_urb_in_irq() when the watchdog timer attempts to loop
through the endpoints' td_lists, giving back killed URBs.

Make sure that xhci_urb_dequeue() removes TDs from the TD lists and
canceled TD lists before it gives back the URB.

This patch should be backported to kernels as old as 2.6.36.

Signed-off-by: Sarah Sharp <sarah.a.sharp@linux.intel.com>
Cc: Andiry Xu <andiry.xu@amd.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoxhci: Fix failed enqueue in the middle of isoch TD.
Sarah Sharp [Fri, 29 Jul 2011 19:44:32 +0000 (12:44 -0700)]
xhci: Fix failed enqueue in the middle of isoch TD.

commit 522989a27c7badb608155b1f1dea3487ed431f74 upstream.

When an isochronous transfer is enqueued, xhci_queue_isoc_tx_prepare()
will ensure that there is enough room on the transfer rings for all of the
isochronous TDs for that URB.  However, when xhci_queue_isoc_tx() is
enqueueing individual isoc TDs, the prepare_transfer() function can fail
if the endpoint state has changed to disabled, error, or some other
unknown state.

With the current code, if Nth TD (not the first TD) fails, the ring is
left in a sorry state.  The partially enqueued TDs are left on the ring,
and the first TRB of the TD is not given back to the hardware.  The
enqueue pointer is left on the TRB after the last successfully enqueued
TD.  This means the ring is basically useless.  Any new transfers will be
enqueued after the failed TDs, which the hardware will never read because
the cycle bit indicates it does not own them.  The ring will fill up with
untransferred TDs, and the endpoint will be basically unusable.

The untransferred TDs will also remain on the TD list.  Since the td_list
is a FIFO, this basically means the ring handler will be waiting on TDs
that will never be completed (or worse, dereference memory that doesn't
exist any more).

Change the code to clean up the isochronous ring after a failed transfer.
If the first TD failed, simply return and allow the xhci_urb_enqueue
function to free the urb_priv.  If the Nth TD failed, first remove the TDs
from the td_list.  Then convert the TRBs that were enqueued into No-op
TRBs.  Make sure to flip the cycle bit on all enqueued TRBs (including any
link TRBs in the middle or between TDs), but leave the cycle bit of the
first TRB (which will show software-owned) intact.  Then move the ring
enqueue pointer back to the first TRB and make sure to change the
xhci_ring's cycle state to what is appropriate for that ring segment.

This ensures that the No-op TRBs will be overwritten by subsequent TDs,
and the hardware will not start executing random TRBs because the cycle
bit was left as hardware-owned.

This bug is unlikely to be hit, but it was something I noticed while
tracking down the watchdog timer issue.  I verified that the fix works by
injecting some errors on the 250th isochronous URB queued, although I
could not verify that the ring is in the correct state because uvcvideo
refused to talk to the device after the first usb_submit_urb() failed.
Ring debugging shows that the ring looks correct, however.

This patch should be backported to kernels as old as 2.6.36.

Signed-off-by: Sarah Sharp <sarah.a.sharp@linux.intel.com>
Cc: Andiry Xu <andiry.xu@amd.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoxhci: Fix memory leak during failed enqueue.
Sarah Sharp [Fri, 22 Jul 2011 21:34:34 +0000 (14:34 -0700)]
xhci: Fix memory leak during failed enqueue.

commit d13565c12828ce0cd2a3862bf6260164a0653352 upstream.

When the isochronous transfer support was introduced, and the xHCI driver
switched to using urb->hcpriv to store an "urb_priv" pointer, a couple of
memory leaks were introduced into the URB enqueue function in its error
handling paths.

xhci_urb_enqueue allocates urb_priv, but it doesn't free it if changing
the control endpoint's max packet size fails or the bulk endpoint is in
the middle of allocating or deallocating streams.

xhci_urb_enqueue also doesn't free urb_priv if any of the four endpoint
types' enqueue functions fail.  Instead, it expects those functions to
free urb_priv if an error occurs.  However, the bulk, control, and
interrupt enqueue functions do not free urb_priv if the endpoint ring is
NULL.  It will, however, get freed if prepare_transfer() fails in those
enqueue functions.

Several of the error paths in the isochronous endpoint enqueue function
also fail to free it.  xhci_queue_isoc_tx_prepare() doesn't free urb_priv
if prepare_ring() indicates there is not enough room for all the
isochronous TDs in this URB.  If individual isochronous TDs fail to be
queued (perhaps due to an endpoint state change), urb_priv is also leaked.

This argues that the freeing of urb_priv should be done in the function
that allocated it, xhci_urb_enqueue.

This patch looks rather ugly, but refactoring the code will have to wait
because this patch needs to be backported to stable kernels.

This patch should be backported to kernels as old as 2.6.36.

Signed-off-by: Sarah Sharp <sarah.a.sharp@linux.intel.com>
Cc: Andiry Xu <andiry.xu@amd.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoxHCI: report USB2 port in resuming as suspend
Andiry Xu [Wed, 3 Aug 2011 08:46:49 +0000 (16:46 +0800)]
xHCI: report USB2 port in resuming as suspend

commit 8a8ff2f9399b23b968901f585ccb5a70a537c5ae upstream.

When a USB2 port initiate a remote wakeup, software shall ensure that
resume is signaled for at least 20ms, and then write '0' to the PLS field.
According to this, xhci driver do the following things:

1. When receive a remote wakeup event in irq_handler, set the resume_done
   value as jiffies + 20ms, and modify rh_timer to poll root hub status at
   that time;
2. When receive a GetPortStatus request, if the jiffies is after the
   resume_done value, clear the resume signal and resume_done.

However, if usb_port_resume() is called before the rh_timer triggered, it
will indicate the port as Suspend Cleared and skip the clear resume signal
part. The device will fail the usb_get_status request in finish_port_resume(),
and usbcore will try a reset-resume instead. Device will work OK after
reset-resume, but resume_done value is not cleared in this case, and
xhci_bus_suspend() will fail because when it finds a non-zero resume_done
value, it will regard the port as resuming and return -EBUSY.

This causes issue on some platforms that the system fail to suspend
after remote wakeup from suspend by USB2 devices connected to xHCI port.

To fix this issue, report the port status as suspend if the resume is
signaling less that 20ms, and usb_port_resume() will wait 25ms and check
port status again, so xHCI driver can clear the resume signaling and
resume_done value.

This should be backported to kernels as old as 2.6.37.

Signed-off-by: Andiry Xu <andiry.xu@amd.com>
Signed-off-by: Sarah Sharp <sarah.a.sharp@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoxHCI: fix port U3 status check condition
Andiry Xu [Wed, 3 Aug 2011 08:46:48 +0000 (16:46 +0800)]
xHCI: fix port U3 status check condition

commit 5ac04bf190e6f8b17238aef179ebd7f2bdfec919 upstream.

Fix the port U3 status check when Clear PORT_SUSPEND Feature.
The port status should be masked with PORT_PLS_MASK to check if it's in
U3 state.

This should be backported to kernels as old as 2.6.37.

Signed-off-by: Andiry Xu <andiry.xu@amd.com>
Signed-off-by: Sarah Sharp <sarah.a.sharp@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoUSB: EHCI: Do not rely on PORT_SUSPEND to stop USB resuming in ehci_bus_resume().
Wang Zhi [Wed, 17 Aug 2011 02:39:31 +0000 (10:39 +0800)]
USB: EHCI: Do not rely on PORT_SUSPEND to stop USB resuming in ehci_bus_resume().

commit d0f2fb2500b1c5fe4967eb45d8c9bc758d7aef80 upstream.

From EHCI Spec p.28 HC should clear PORT_SUSPEND when SW clears
PORT_RESUME. In Intel Oaktrail platform, MPH (Multi-Port Host
Controller) core clears PORT_SUSPEND directly when SW sets PORT_RESUME
bit. If we rely on PORT_SUSPEND bit to stop USB resume, we will miss
the action of clearing PORT_RESUME. This will cause unexpected long
resume signal on USB bus.

Signed-off-by: Wang Zhi <zhi.wang@windriver.com>
Signed-off-by: Alan Stern <stern@rowland.harvard.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agousb: musb: cppi: fix build errors due to DBG and missing musb variable
Per Forlin [Wed, 3 Aug 2011 13:39:15 +0000 (15:39 +0200)]
usb: musb: cppi: fix build errors due to DBG and missing musb variable

commit f847a79ab3c1faca3022061045cd22e4678c1b1c upstream.

Replace DBG with dev_dbg and fix invalid access of musb->controller.
With this patch cppi_dma builds successfully.

Signed-off-by: Per Forlin <per.forlin@linaro.org>
Signed-off-by: Felipe Balbi <balbi@ti.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoUSB option driver K3765/K4505 avoid CDC_DATA interface
Andrew Bird [Tue, 16 Aug 2011 23:20:03 +0000 (00:20 +0100)]
USB option driver K3765/K4505 avoid CDC_DATA interface

commit 6118514e8749105334f46ccec6faf9a439be6cf9 upstream.

Currently the Option driver avoids binding interface 1 on Huawei K3765
and K4505 broadband modems as it should be handled by the cdc_ether
driver instead. This patch ensures we don't bind the interface 2
on those devices as that is CDC_DATA.

Signed-off-by: Andrew Bird <ajb@spheresystems.co.uk>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoUSB: option: add YUGA device id to driver
Gavin.zhu [Mon, 22 Aug 2011 20:51:53 +0000 (13:51 -0700)]
USB: option: add YUGA device id to driver

commit c6eb2d75ffcdfafa37ff010bf467de20d468ef79 upstream.

Signed-off-by: Gavin.zhu <gavin.kx@qq.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoUSB option driver add PID of Huawei Vodafone K4605
Andrew Bird [Tue, 16 Aug 2011 19:58:21 +0000 (13:58 -0600)]
USB option driver add PID of Huawei Vodafone K4605

commit 7e1805844da18a37e6d251d286f93c94b52d791e upstream.

This patch adds the product ID of Huawei's Vodafone K4605 mobile broadband
modem to option.c. This is necessary so that the driver gets loaded on
demand without the intervention of usb_modeswitch. This has the benefit of
it becoming available faster and also ensures that the option driver is not
bound to a network interface that should be claimed by suitable network
driver.

Signed-off-by: Andrew Bird <ajb@spheresystems.co.uk>
Signed-off-by: Alex Chiang <achiang@canonical.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoUSB option driver add PID of Huawei Vodafone K3806
Andrew Bird [Tue, 16 Aug 2011 19:57:14 +0000 (13:57 -0600)]
USB option driver add PID of Huawei Vodafone K3806

commit 0e69d75ccb2f091757b38d4d6a2ed739e06b615e upstream.

This patch adds the product ID of Huawei's Vodafone K3806 mobile broadband
modem to option.c. This is necessary so that the driver gets loaded on
demand without the intervention of usb_modeswitch. This has the benefit of
it becoming available faster and also ensures that the option driver is not
bound to a network interface that should be claimed by cdc_ether.

Signed-off-by: Andrew Bird <ajb@spheresystems.co.uk>
Signed-off-by: Alex Chiang <achiang@canonical.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agousb: s5p-ehci: fix a NULL pointer deference
Yulgon Kim [Thu, 18 Aug 2011 05:02:45 +0000 (14:02 +0900)]
usb: s5p-ehci: fix a NULL pointer deference

commit e5d3d4463fb30998385f9e78ab3c7f63b5813000 upstream.

This patch fixes a NULL pointer deference. A NULL pointer
dereference happens since s5p_ehci->hcd field is not initialized
yet in probe function.

[jg1.han@samsung.com: edit commit message]
Signed-off-by: Yulgon Kim <yulgon.kim@samsung.com>
Signed-off-by: Jingoo Han <jg1.han@samsung.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoUSB: ftdi_sio: add Calao reference board support
Jean-Christophe PLAGNIOL-VILLARD [Thu, 25 Aug 2011 09:46:58 +0000 (11:46 +0200)]
USB: ftdi_sio: add Calao reference board support

commit c96fbdd0ab97235f930ebf24b38fa42a2e3458cf upstream.

Calao use on there dev kits a FT2232 where the port 0 is used for the JTAG and
port 1 for the UART

They use the same VID and PID as FTDI Chip but they program the manufacturer
name in the eeprom

So use this information to detect it

Signed-off-by: Jean-Christophe PLAGNIOL-VILLARD <plagnioj@jcrosoft.com>
Cc: Gregory Hermant <gregory.hermant@calao-systems.com>
Cc: Alan Cox <alan@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoTTY: pty, fix pty counting
Jiri Slaby [Wed, 10 Aug 2011 12:59:28 +0000 (14:59 +0200)]
TTY: pty, fix pty counting

commit 24d406a6bf736f7aebdc8fa0f0ec86e0890c6d24 upstream.

tty_operations->remove is normally called like:
queue_release_one_tty
 ->tty_shutdown
   ->tty_driver_remove_tty
     ->tty_operations->remove

However tty_shutdown() is called from queue_release_one_tty() only if
tty_operations->shutdown is NULL. But for pty, it is not.
pty_unix98_shutdown() is used there as ->shutdown.

So tty_operations->remove of pty (i.e. pty_unix98_remove()) is never
called. This results in invalid pty_count. I.e. what can be seen in
/proc/sys/kernel/pty/nr.

I see this was already reported at:
  https://lkml.org/lkml/2009/11/5/370
But it was not fixed since then.

This patch is kind of a hackish way. The problem lies in ->install. We
allocate there another tty (so-called tty->link). So ->install is
called once, but ->remove twice, for both tty and tty->link. The fix
here is to count both tty and tty->link and divide the count by 2 for
user.

And to have ->remove called, let's make tty_driver_remove_tty() global
and call that from pty_unix98_shutdown() (tty_operations->shutdown).

While at it, let's document that when ->shutdown is defined,
tty_shutdown() is not called.

Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Cc: Alan Cox <alan@linux.intel.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agotty: Add "spi:" prefix for spi modalias
Axel Lin [Mon, 1 Aug 2011 13:20:10 +0000 (21:20 +0800)]
tty: Add "spi:" prefix for spi modalias

commit 8c4074cd2254606aeb788d518ccc27c9f97129e1 upstream.

Since commit e0626e38 (spi: prefix modalias with "spi:"),
the spi modalias is prefixed with "spi:".

This patch adds "spi:" prefix and removes "-spi" suffix in the modalias.

Signed-off-by: Axel Lin <axel.lin@gmail.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years ago8250: Fix race condition in serial8250_backup_timeout().
Al Cooper [Mon, 25 Jul 2011 20:19:52 +0000 (16:19 -0400)]
8250: Fix race condition in serial8250_backup_timeout().

commit dbb3b1ca5609d1f3848cd387d06cc60aaacf7f98 upstream.

This is to fix an issue where output will suddenly become very slow.
The problem occurs on 8250 UARTS with the hardware bug UART_BUG_THRE.

BACKGROUND
For normal UARTs (without UART_BUG_THRE): When the serial core layer
gets new transmit data and the transmitter is idle, it buffers the
data and calls the 8250s' serial8250_start_tx() routine which will
simply enable the TX interrupt in the IER register and return. This
should immediately fire a THRE interrupt and begin transmitting the
data.
For buggy UARTs (with UART_BUG_THRE): merely enabling the TX interrupt
in IER does not necessarily generate a new THRE interrupt.
Therefore, a background timer periodically checks to see if there is
pending data, and starts transmission if that is the case.

The bug happens on SMP systems when the system has nothing to transmit,
the transmit interrupt is disabled and the following sequence occurs:
- CPU0: The background timer routine serial8250_backup_timeout()
  starts and saves the state of the interrupt enable register (IER)
  and then disables all interrupts in IER. NOTE: The transmit interrupt
  (TI) bit is saved as disabled.
- CPU1: The serial core gets data to transmit, grabs the port lock and
  calls serial8250_start_tx() which enables the TI in IER.
- CPU0: serial8250_backup_timeout() waits for the port lock.
- CPU1: finishes (with TI enabled) and releases the port lock.
- CPU0: serial8250_backup_timeout() calls the interrupt routine which
  will transmit the next fifo's worth of data and then restores the
  IER from the previously saved value (TI disabled).
At this point, as long as the serial core has more transmit data
buffered, it will not call serial8250_start_tx() again and the
background timer routine will slowly transmit the data.

The fix is to have serial8250_start_tx() get the port lock before
it saves the IER state and release it after restoring IER. This will
prevent serial8250_start_tx() from running in parallel.

Signed-off-by: Al Cooper <alcooperx@gmail.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years ago8250_pci: add support for Rosewill RC-305 4x serial port card
Eric Smith [Tue, 12 Jul 2011 04:53:13 +0000 (22:53 -0600)]
8250_pci: add support for Rosewill RC-305 4x serial port card

commit 44178176ecc55ad370b837dd2c4b4b8bed1e3823 upstream.

This patch adds support for the Rosewill RC-305 four-port PCI serial
card, and probably any other four-port serial cards based on the
Moschip MCS9865 chip, assuming that the EEPROM on the card was
programmed in accordance with Table 6 of the MCS9865 EEPROM
Application Note version 0.3 dated 16-May-2008, available from the
Moschip web site (registration required).

This patch is based on an earlier patch [1] for the SYBA 6x serial
port card by Ira W. Snyder.

[1]: http://www.gossamer-threads.com/lists/linux/kernel/1162435

Signed-off-by: Eric Smith <eric@brouhaha.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoserial: 8250_pnp: add Intermec CV60 touchscreen device
Bjorn Helgaas [Tue, 16 Aug 2011 18:02:28 +0000 (12:02 -0600)]
serial: 8250_pnp: add Intermec CV60 touchscreen device

commit ab8ba3a2d2cba6a658ef596cd5b2e0905b6c8a9f upstream.

It would have been nice if Intermec had supplied a PNP0501 _CID for the
COM3 device, but they didn't, so we have to recognize it explicitly.

Reference: https://bugzilla.kernel.org/show_bug.cgi?id=40612
CC: Jeff Chua <jeff.chua.linux@gmail.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Acked-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoomap-serial: Allow IXON and IXOFF to be disabled.
Nick Pelly [Fri, 15 Jul 2011 20:53:08 +0000 (13:53 -0700)]
omap-serial: Allow IXON and IXOFF to be disabled.

commit b280a97d1caf6fe1d38b51ebb31219391f5ad1a0 upstream.

Fixes logic bug that software flow control cannot be disabled, because
serial_omap_configure_xonxoff() is not called if both IXON and IXOFF bits
are cleared.

Signed-off-by: Nick Pelly <npelly@google.com>
Acked-by: Govindraj.R <govindraj.raja@ti.com>
Tested-by: Govindraj.R <govindraj.raja@ti.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoASoC: ad193x: fix dac word len setting
Scott Jiang [Fri, 12 Aug 2011 22:04:11 +0000 (18:04 -0400)]
ASoC: ad193x: fix dac word len setting

commit 95c93d8525ebce1024bda7316f602ae45c36cd6f upstream.

dac word len value should left shift before setting

Signed-off-by: Scott Jiang <scott.jiang.linux@gmail.com>
Acked-by: Barry Song <21cnbao@gmail.com>
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoASoC: ad193x: fix registers definition
Scott Jiang [Fri, 12 Aug 2011 22:04:10 +0000 (18:04 -0400)]
ASoC: ad193x: fix registers definition

commit bf545ed72f2eeac664695a8ea2199d9ddaef6020 upstream.

fix dac word len mask and adc tdm fmt shift value

Signed-off-by: Scott Jiang <scott.jiang.linux@gmail.com>
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoASoC: soc-jack: Fix checking return value of request_any_context_irq
Axel Lin [Sat, 13 Aug 2011 11:15:01 +0000 (19:15 +0800)]
ASoC: soc-jack: Fix checking return value of request_any_context_irq

commit d2b4c7bd7eabfaa2e3e5b8107d5eeb56ac879813 upstream.

request_any_context_irq() returns a negative value on failure.
On success, it returns either IRQC_IS_HARDIRQ or IRQC_IS_NESTED.

Signed-off-by: Axel Lin <axel.lin@gmail.com>
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoarch/powerpc/sysdev/fsl_rio.c: correct IECSR register clear value
Liu Gang-B34182 [Thu, 25 Aug 2011 22:59:25 +0000 (15:59 -0700)]
arch/powerpc/sysdev/fsl_rio.c: correct IECSR register clear value

commit 671ee7f0ce62e4b991b47fcf1c161c3f710dabbc upstream.

This bug causes the IECSR register clear failure.  In this case, the RETE
(retry error threshold exceeded) interrupt will be generated and cannot be
cleared.  So the related ISR may be called persistently.

The RETE bit in IECSR is cleared by writing a 1 to it.

Signed-off-by: Liu Gang <Gang.Liu@freescale.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Kumar Gala <galak@kernel.crashing.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agorapidio: fix use of non-compatible registers
Alexandre Bounine [Thu, 25 Aug 2011 22:59:13 +0000 (15:59 -0700)]
rapidio: fix use of non-compatible registers

commit 284fb68d00c56e971ed01e0b4bac5ddd4d1b74ab upstream.

Replace/remove use of RIO v.1.2 registers/bits that are not
forward-compatible with newer versions of RapidIO specification.

RapidIO specification v.1.3 removed Write Port CSR, Doorbell CSR,
Mailbox CSR and Mailbox and Doorbell bits of the PEF CAR.

Use of removed (since RIO v.1.3) register bits affects users of
currently available 1.3 and 2.x compliant devices who may use not so
recent kernel versions.

Removing checks for unsupported bits makes corresponding routines
compatible with all versions of RapidIO specification.  Therefore,
backporting makes stable kernel versions compliant with RIO v.1.3 and
later as well.

Signed-off-by: Alexandre Bounine <alexandre.bounine@idt.com>
Cc: Kumar Gala <galak@kernel.crashing.org>
Cc: Matt Porter <mporter@kernel.crashing.org>
Cc: Li Yang <leoli@freescale.com>
Cc: Thomas Moll <thomas.moll@sysgo.com>
Cc: Chul Kim <chul.kim@idt.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agokernel/printk: do not turn off bootconsole in printk_late_init() if keep_bootcon
Nishanth Aravamudan [Thu, 25 Aug 2011 22:59:11 +0000 (15:59 -0700)]
kernel/printk: do not turn off bootconsole in printk_late_init() if keep_bootcon

commit 4c30c6f566c0989ddaee3407da44751e340a63ed upstream.

It seems that 7bf693951a8e ("console: allow to retain boot console via
boot option keep_bootcon") doesn't always achieve what it aims, as when
printk_late_init() runs it unconditionally turns off all boot consoles.
With this patch, I am able to see more messages on the boot console in
KVM guests than I can without, when keep_bootcon is specified.

I think it is appropriate for the relevant -stable trees.  However, it's
more of an annoyance than a serious bug (ideally you don't need to keep
the boot console around as console handover should be working -- I was
encountering a situation where the console handover wasn't working and
not having the boot console available meant I couldn't see why).

Signed-off-by: Nishanth Aravamudan <nacc@us.ibm.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Alan Cox <alan@lxorguk.ukuu.org.uk>
Cc: Greg KH <gregkh@suse.de>
Acked-by: Fabio M. Di Nitto <fdinitto@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoLinux 3.0.4
Greg Kroah-Hartman [Mon, 29 Aug 2011 20:56:30 +0000 (13:56 -0700)]
Linux 3.0.4

13 years agoAdd a personality to report 2.6.x version numbers
Andi Kleen [Fri, 19 Aug 2011 23:15:10 +0000 (16:15 -0700)]
Add a personality to report 2.6.x version numbers

commit be27425dcc516fd08245b047ea57f83b8f6f0903 upstream.

I ran into a couple of programs which broke with the new Linux 3.0
version.  Some of those were binary only.  I tried to use LD_PRELOAD to
work around it, but it was quite difficult and in one case impossible
because of a mix of 32bit and 64bit executables.

For example, all kind of management software from HP doesnt work, unless
we pretend to run a 2.6 kernel.

  $ uname -a
  Linux svivoipvnx001 3.0.0-08107-g97cd98f #1062 SMP Fri Aug 12 18:11:45 CEST 2011 i686 i686 i386 GNU/Linux

  $ hpacucli ctrl all show

  Error: No controllers detected.

  $ rpm -qf /usr/sbin/hpacucli
  hpacucli-8.75-12.0

Another notable case is that Python now reports "linux3" from
sys.platform(); which in turn can break things that were checking
sys.platform() == "linux2":

  https://bugzilla.mozilla.org/show_bug.cgi?id=664564

It seems pretty clear to me though it's a bug in the apps that are using
'==' instead of .startswith(), but this allows us to unbreak broken
programs.

This patch adds a UNAME26 personality that makes the kernel report a
2.6.40+x version number instead.  The x is the x in 3.x.

I know this is somewhat ugly, but I didn't find a better workaround, and
compatibility to existing programs is important.

Some programs also read /proc/sys/kernel/osrelease.  This can be worked
around in user space with mount --bind (and a mount namespace)

To use:

  wget ftp://ftp.kernel.org/pub/linux/kernel/people/ak/uname26/uname26.c
  gcc -o uname26 uname26.c
  ./uname26 program

Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoigb: Fix lack of flush after register write and before delay
Carolyn Wyborny [Sat, 25 Jun 2011 13:18:12 +0000 (13:18 +0000)]
igb: Fix lack of flush after register write and before delay

commit 064b43304ed8ede8e13ff7b4338d09fd37bcffb1 upstream.

Register writes followed by a delay are required to have a flush
before the delay in order to commit the values to the register.  Without
the flush, the code following the delay may not function correctly.

Reported-by: Tong Ho <tong.ho@ericsson.com>
Reported-by: Guenter Roeck <guenter.roeck@ericsson.com>
Signed-off-by: Carolyn Wyborny <carolyn.wyborny@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agommc: sdhci: fix retuning timer wrongly deleted in sdhci_tasklet_finish
Aaron Lu [Mon, 11 Jul 2011 05:27:11 +0000 (13:27 +0800)]
mmc: sdhci: fix retuning timer wrongly deleted in sdhci_tasklet_finish

commit 78869618a886d33d8cdfcb78cf9b245b5250e465 upstream.

Currently, the retuning timer for retuning mode 1 will be deleted in
function sdhci_tasklet_finish after a mmc request done, which will make
retuning timing never trigger again. This patch fixed this problem.

Signed-off-by: Aaron Lu <Aaron.Lu@amd.com>
Reviewed-by: Philip Rakity <prakity@marvell.com>
Signed-off-by: Chris Ball <cjb@laptop.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agort2x00: fix order of entry flags modification
Stanislaw Gruszka [Wed, 10 Aug 2011 13:32:23 +0000 (15:32 +0200)]
rt2x00: fix order of entry flags modification

commit df71c9cfceea801e7e26e2c74241758ef9c042e5 upstream.

In rt2800usb_work_txdone we check flags in order:

- ENTRY_OWNER_DEVICE_DATA
- ENTRY_DATA_STATUS_PENDING
- ENTRY_DATA_IO_FAILED

Modify flags in separate order in rt2x00usb_interrupt_txdone, to avoid
processing entries in _txdone with wrong flags or skip processing
ready entries.

Reported-by: Justin Piszcz <jpiszcz@lucidpixels.com>
Signed-off-by: Stanislaw Gruszka <sgruszka@redhat.com>
Acked-by: Ivo van Doorn <IvDoorn@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agofuse: check size of FUSE_NOTIFY_INVAL_ENTRY message
Miklos Szeredi [Wed, 24 Aug 2011 08:20:17 +0000 (10:20 +0200)]
fuse: check size of FUSE_NOTIFY_INVAL_ENTRY message

commit c2183d1e9b3f313dd8ba2b1b0197c8d9fb86a7ae upstream.

FUSE_NOTIFY_INVAL_ENTRY didn't check the length of the write so the
message processing could overrun and result in a "kernel BUG at
fs/fuse/dev.c:629!"

Reported-by: Han-Wen Nienhuys <hanwenn@gmail.com>
Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agodrm/radeon: Extended DDC Probing for Toshiba L300D Radeon Mobility X1100 HDMI-A Connector
Thomas Reim [Wed, 17 Aug 2011 09:03:32 +0000 (09:03 +0000)]
drm/radeon: Extended DDC Probing for Toshiba L300D Radeon Mobility X1100 HDMI-A Connector

commit f2b60717e692550bf753a5d64a5b69ea430fc832 upstream.

Toshiba Satellite L300D with ATI Mobility Radeon X1100 sends data
   to i2c bus for a HDMI connector that is not implemented/existent
   on the notebook's board.

   Fix by applying extented DDC probing for this connector.

   Requires [PATCH] drm/radeon: Extended DDC Probing for Connectors
   with Improperly Wired DDC Lines

   Tested for kernel 2.6.38 on Toshiba Satellite L300D notebook

BugLink: http://bugs.launchpad.net/bugs/826677
Signed-off-by: Thomas Reim <reimth@gmail.com>
Acked-by: Chris Routh <routhy@gmail.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agodrm/ttm: fix ttm_bo_add_ttm(user) failure path
Marcin Slusarz [Mon, 22 Aug 2011 21:17:57 +0000 (21:17 +0000)]
drm/ttm: fix ttm_bo_add_ttm(user) failure path

commit 7c4c3960dff109bc5db4c35da481c212dadb5eb5 upstream.

ttm_tt_destroy kfrees passed object, so we need to nullify
a reference to it.

Signed-off-by: Marcin Slusarz <marcin.slusarz@gmail.com>
Reviewed-by: Thomas Hellstrom <thellstrom@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agox86, UV: Remove UV delay in starting slave cpus
Jack Steiner [Fri, 5 Aug 2011 14:09:00 +0000 (09:09 -0500)]
x86, UV: Remove UV delay in starting slave cpus

commit 05e33fc20ea5e493a2a1e7f1d04f43cdf89f83ed upstream.

Delete the 10 msec delay between the INIT and SIPI when starting
slave cpus. I can find no requirement for this delay. BIOS also
has similar code sequences without the delay.

Removing the delay reduces boot time by 40 sec. Every bit helps.

Signed-off-by: Jack Steiner <steiner@sgi.com>
Link: http://lkml.kernel.org/r/20110805140900.GA6774@sgi.com
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agox86-32, vdso: On system call restart after SYSENTER, use int $0x80
H. Peter Anvin [Mon, 22 Aug 2011 20:27:06 +0000 (13:27 -0700)]
x86-32, vdso: On system call restart after SYSENTER, use int $0x80

commit 7ca0758cdb7c241cb4e0490a8d95f0eb5b861daf upstream.

When we enter a 32-bit system call via SYSENTER or SYSCALL, we shuffle
the arguments to match the int $0x80 calling convention.  This was
probably a design mistake, but it's what it is now.  This causes
errors if the system call as to be restarted.

For SYSENTER, we have to invoke the instruction from the vdso as the
return address is hardcoded.  Accordingly, we can simply replace the
jump in the vdso with an int $0x80 instruction and use the slower
entry point for a post-restart.

Suggested-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
Link: http://lkml.kernel.org/r/CA%2B55aFztZ=r5wa0x26KJQxvZOaQq8s2v3u50wCyJcA-Sc4g8gQ@mail.gmail.com
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agox86, olpc: Wait for last byte of EC command to be accepted
Paul Fox [Tue, 26 Jul 2011 15:42:26 +0000 (16:42 +0100)]
x86, olpc: Wait for last byte of EC command to be accepted

commit a3ea14df0e383f44dcb2e61badb71180dbffe526 upstream.

When executing EC commands, only waiting when there are still
more bytes to write is usually fine. However, if the system
suspends very quickly after a call to olpc_ec_cmd(), the last
data byte may not yet be transferred to the EC, and the command
will not complete.

This solves a bug where the SCI wakeup mask was not correctly
written when going into suspend.

It means that sometimes, on XO-1.5 (but not XO-1), the
devices that were marked as wakeup sources can't wake up
the system. e.g. you ask for wifi wakeups, suspend, but then
incoming wifi frames don't wake up the system as they should.

Signed-off-by: Paul Fox <pgf@laptop.org>
Signed-off-by: Daniel Drake <dsd@laptop.org>
Acked-by: Andres Salomon <dilinger@queued.net>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoxen: Do not enable PV IPIs when vector callback not present
Stefano Stabellini [Wed, 17 Aug 2011 13:15:00 +0000 (15:15 +0200)]
xen: Do not enable PV IPIs when vector callback not present

commit 3c05c4bed4ccce3f22f6d7899b308faae24ad198 upstream.

Fix regression for HVM case on older (<4.1.1) hypervisors caused by

  commit 99bbb3a84a99cd04ab16b998b20f01a72cfa9f4f
  Author: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
  Date:   Thu Dec 2 17:55:10 2010 +0000

    xen: PV on HVM: support PV spinlocks and IPIs

This change replaced the SMP operations with event based handlers without
taking into account that this only works when the hypervisor supports
callback vectors. This causes unexplainable hangs early on boot for
HVM guests with more than one CPU.

BugLink: http://bugs.launchpad.net/bugs/791850
Signed-off-by: Stefan Bader <stefan.bader@canonical.com>
Signed-off-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
Tested-and-Reported-by: Stefan Bader <stefan.bader@canonical.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoxen/x86: replace order-based range checking of M2P table by linear one
Jan Beulich [Tue, 16 Aug 2011 14:07:41 +0000 (15:07 +0100)]
xen/x86: replace order-based range checking of M2P table by linear one

commit ccbcdf7cf1b5f6c6db30d84095b9c6c53043af55 upstream.

The order-based approach is not only less efficient (requiring a shift
and a compare, typical generated code looking like this

mov eax, [machine_to_phys_order]
mov ecx, eax
shr ebx, cl
test ebx, ebx
jnz ...

whereas a direct check requires just a compare, like in

cmp ebx, [machine_to_phys_nr]
jae ...

), but also slightly dangerous in the 32-on-64 case - the element
address calculation can wrap if the next power of two boundary is
sufficiently far away from the actual upper limit of the table, and
hence can result in user space addresses being accessed (with it being
unknown what may actually be mapped there).

Additionally, the elimination of the mistaken use of fls() here (should
have been __fls()) fixes a latent issue on x86-64 that would trigger
if the code was run on a system with memory extending beyond the 44-bit
boundary.

Signed-off-by: Jan Beulich <jbeulich@novell.com>
[v1: Based on Jeremy's feedback]
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoxen-blkfront: Fix one off warning about name clash
Stefan Bader [Thu, 14 Jul 2011 13:30:37 +0000 (15:30 +0200)]
xen-blkfront: Fix one off warning about name clash

commit 89153b5cae9f40c224a5d321665a97bf14220c2c upstream.

Avoid telling users to use xvde and onwards when using xvde.

Acked-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
Signed-off-by: Stefan Bader <stefan.bader@canonical.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoxen-blkfront: Drop name and minor adjustments for emulated scsi devices
Stefan Bader [Thu, 14 Jul 2011 13:30:22 +0000 (15:30 +0200)]
xen-blkfront: Drop name and minor adjustments for emulated scsi devices

commit 196cfe2ae8fcdc03b3c7d627e7dfe8c0ce7229f9 upstream.

These were intended to avoid the namespace clash when representing
emulated IDE and SCSI devices. However that seems to confuse users
more than expected (a disk defined as sda becomes xvde).
So for now go back to the scheme which does no adjustments. This
will break when mixing IDE and SCSI names in the configuration of
guests but should be by now expected.

Acked-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
Signed-off-by: Stefan Bader <stefan.bader@canonical.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoext4: fix nomblk_io_submit option so it correctly converts uninit blocks
Theodore Ts'o [Sat, 13 Aug 2011 16:58:21 +0000 (12:58 -0400)]
ext4: fix nomblk_io_submit option so it correctly converts uninit blocks

commit 9dd75f1f1a02d656a11a7b9b9e6c2759b9c1e946 upstream.

Bug discovered by Jan Kara:

Finally, commit 1449032be17abb69116dbc393f67ceb8bd034f92 returned back
the old IO submission code but apparently it forgot to return the old
handling of uninitialized buffers so we unconditionnaly call
block_write_full_page() without specifying end_io function. So AFAICS
we never convert unwritten extents to written in some cases. For
example when I mount the fs as: mount -t ext4 -o
nomblk_io_submit,dioread_nolock /dev/ubdb /mnt and do
        int fd = open(argv[1], O_RDWR | O_CREAT | O_TRUNC, 0600);
        char buf[1024];
        memset(buf, 'a', sizeof(buf));
        fallocate(fd, 0, 0, 16384);
        write(fd, buf, sizeof(buf));

I get a file full of zeros (after remounting the filesystem so that
pagecache is dropped) instead of seeing the first KB contain 'a's.

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoext4: Resolve the hang of direct i/o read in handling EXT4_IO_END_UNWRITTEN.
Tao Ma [Sat, 13 Aug 2011 16:30:59 +0000 (12:30 -0400)]
ext4: Resolve the hang of direct i/o read in handling EXT4_IO_END_UNWRITTEN.

commit 32c80b32c053dc52712dedac5e4d0aa7c93fc353 upstream.

EXT4_IO_END_UNWRITTEN flag set and the increase of i_aiodio_unwritten
should be done simultaneously since ext4_end_io_nolock always clear
the flag and decrease the counter in the same time.

We don't increase i_aiodio_unwritten when setting
EXT4_IO_END_UNWRITTEN so it will go nagative and causes some process
to wait forever.

Part of the patch came from Eric in his e-mail, but it doesn't fix the
problem met by Michael actually.

http://marc.info/?l=linux-ext4&m=131316851417460&w=2

Reported-and-Tested-by: Michael Tokarev<mjt@tls.msk.ru>
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Tao Ma <boyu.mt@taobao.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoext4: call ext4_ioend_wait and ext4_flush_completed_IO in ext4_evict_inode
Jiaying Zhang [Sat, 13 Aug 2011 16:17:13 +0000 (12:17 -0400)]
ext4: call ext4_ioend_wait and ext4_flush_completed_IO in ext4_evict_inode

commit 2581fdc810889fdea97689cb62481201d579c796 upstream.

Flush inode's i_completed_io_list before calling ext4_io_wait to
prevent the following deadlock scenario: A page fault happens while
some process is writing inode A. During page fault,
shrink_icache_memory is called that in turn evicts another inode
B. Inode B has some pending io_end work so it calls ext4_ioend_wait()
that waits for inode B's i_ioend_count to become zero. However, inode
B's ioend work was queued behind some of inode A's ioend work on the
same cpu's ext4-dio-unwritten workqueue. As the ext4-dio-unwritten
thread on that cpu is processing inode A's ioend work, it tries to
grab inode A's i_mutex lock. Since the i_mutex lock of inode A is
still hold before the page fault happened, we enter a deadlock.

Also moves ext4_flush_completed_IO and ext4_ioend_wait from
ext4_destroy_inode() to ext4_evict_inode(). During inode deleteion,
ext4_evict_inode() is called before ext4_destroy_inode() and in
ext4_evict_inode(), we may call ext4_truncate() without holding
i_mutex lock. As a result, there is a race between flush_completed_IO
that is called from ext4_ext_truncate() and ext4_end_io_work, which
may cause corruption on an io_end structure. This change moves
ext4_flush_completed_IO and ext4_ioend_wait from ext4_destroy_inode()
to ext4_evict_inode() to resolve the race between ext4_truncate() and
ext4_end_io_work during inode deletion.

Signed-off-by: Jiaying Zhang <jiayingz@google.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>