pandora-kernel.git
11 years agoxen/x86: don't corrupt %eip when returning from a signal handler
David Vrabel [Fri, 19 Oct 2012 16:29:07 +0000 (17:29 +0100)]
xen/x86: don't corrupt %eip when returning from a signal handler

commit a349e23d1cf746f8bdc603dcc61fae9ee4a695f6 upstream.

In 32 bit guests, if a userspace process has %eax == -ERESTARTSYS
(-512) or -ERESTARTNOINTR (-513) when it is interrupted by an event
/and/ the process has a pending signal then %eip (and %eax) are
corrupted when returning to the main process after handling the
signal.  The application may then crash with SIGSEGV or a SIGILL or it
may have subtly incorrect behaviour (depending on what instruction it
returned to).

The occurs because handle_signal() is incorrectly thinking that there
is a system call that needs to restarted so it adjusts %eip and %eax
to re-execute the system call instruction (even though user space had
not done a system call).

If %eax == -514 (-ERESTARTNOHAND (-514) or -ERESTART_RESTARTBLOCK
(-516) then handle_signal() only corrupted %eax (by setting it to
-EINTR).  This may cause the application to crash or have incorrect
behaviour.

handle_signal() assumes that regs->orig_ax >= 0 means a system call so
any kernel entry point that is not for a system call must push a
negative value for orig_ax.  For example, for physical interrupts on
bare metal the inverse of the vector is pushed and page_fault() sets
regs->orig_ax to -1, overwriting the hardware provided error code.

xen_hypervisor_callback() was incorrectly pushing 0 for orig_ax
instead of -1.

Classic Xen kernels pushed %eax which works as %eax cannot be both
non-negative and -RESTARTSYS (etc.), but using -1 is consistent with
other non-system call entry points and avoids some of the tests in
handle_signal().

There were similar bugs in xen_failsafe_callback() of both 32 and
64-bit guests. If the fault was corrected and the normal return path
was used then 0 was incorrectly pushed as the value for orig_ax.

Signed-off-by: David Vrabel <david.vrabel@citrix.com>
Acked-by: Jan Beulich <JBeulich@suse.com>
Acked-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agos390: fix linker script for 31 bit builds
Heiko Carstens [Thu, 18 Oct 2012 09:11:01 +0000 (11:11 +0200)]
s390: fix linker script for 31 bit builds

commit c985cb37f1b39c2c8035af741a2a0b79f1fbaca7 upstream.

Because of a change in the s390 arch backend of binutils (commit 23ecd77
"Pick the default arch depending on the target size" in binutils repo)
31 bit builds will fail since the linker would now try to create 64 bit
binary output.
Fix this by setting OUTPUT_ARCH to s390:31-bit instead of s390.
Thanks to Andreas Krebbel for figuring out the issue.

Fixes this build error:

  LD      init/built-in.o
s390x-4.7.2-ld: s390:31-bit architecture of input file
 `arch/s390/kernel/head.o' is incompatible with s390:64-bit output

Cc: Andreas Krebbel <Andreas.Krebbel@de.ibm.com>
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agoUSB: option: add more ZTE devices
Bjørn Mork [Thu, 18 Oct 2012 15:14:17 +0000 (17:14 +0200)]
USB: option: add more ZTE devices

commit 4b35f1c52943851b310afb09047bfe991ac8f5ae upstream.

Signed-off-by: Bjørn Mork <bjorn@mork.no>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agoUSB: option: blacklist net interface on ZTE devices
Bjørn Mork [Thu, 18 Oct 2012 15:19:53 +0000 (17:19 +0200)]
USB: option: blacklist net interface on ZTE devices

commit 1452df6f1b7e396d89c2a1fdbdc0e0e839f97671 upstream.

Based on information from the ZTE Windows drivers.

Signed-off-by: Bjørn Mork <bjorn@mork.no>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agousb: host: xhci: New system added for Compliance Mode Patch on SN65LVPE502CP
Alexis R. Cortes [Wed, 17 Oct 2012 19:09:12 +0000 (14:09 -0500)]
usb: host: xhci: New system added for Compliance Mode Patch on SN65LVPE502CP

commit 470809741a28c3092279f4e1f3f432e534d46068 upstream.

This minor change adds a new system to which the "Fix Compliance Mode
on SN65LVPE502CP Hardware" patch has to be applied also.

System added:
Vendor: Hewlett-Packard. System Model: Z1

Signed-off-by: Alexis R. Cortes <alexis.cortes@ti.com>
Acked-by: Sarah Sharp <sarah.a.sharp@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agousb: acm: fix the computation of the number of data bits
Nicolas Boullis [Mon, 15 Oct 2012 22:06:23 +0000 (00:06 +0200)]
usb: acm: fix the computation of the number of data bits

commit 301a29da6e891e7eb95c843af0ecdbe86d01f723 upstream.

The current code assumes that CSIZE is 0000060, which appears to be
wrong on some arches (such as powerpc).

Signed-off-by: Nicolas Boullis <nboullis@debian.org>
Acked-by: Oliver Neukum <oneukum@suse.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agoUSB: cdc-acm: fix pipe type of write endpoint
Ming Lei [Tue, 16 Oct 2012 13:21:21 +0000 (21:21 +0800)]
USB: cdc-acm: fix pipe type of write endpoint

commit c5211187f7ff8e8dbff4ebf7c011ac4c0ffe319c upstream.

If the write endpoint is interrupt type, usb_sndintpipe() should
be passed to usb_fill_int_urb() instead of usb_sndbulkpipe().

Cc: Oliver Neukum <oneukum@suse.de>
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agoSUNRPC: Prevent kernel stack corruption on long values of flush
Sasha Levin [Mon, 16 Jul 2012 22:01:26 +0000 (00:01 +0200)]
SUNRPC: Prevent kernel stack corruption on long values of flush

commit 212ba90696ab4884e2025b0b13726d67aadc2cd4 upstream.

The buffer size in read_flush() is too small for the longest possible values
for it. This can lead to a kernel stack corruption:

[   43.047329] Kernel panic - not syncing: stack-protector: Kernel stack is corrupted in: ffffffff833e64b4
[   43.047329]
[   43.049030] Pid: 6015, comm: trinity-child18 Tainted: G        W    3.5.0-rc7-next-20120716-sasha #221
[   43.050038] Call Trace:
[   43.050435]  [<ffffffff836c60c2>] panic+0xcd/0x1f4
[   43.050931]  [<ffffffff833e64b4>] ? read_flush.isra.7+0xe4/0x100
[   43.051602]  [<ffffffff810e94e6>] __stack_chk_fail+0x16/0x20
[   43.052206]  [<ffffffff833e64b4>] read_flush.isra.7+0xe4/0x100
[   43.052951]  [<ffffffff833e6500>] ? read_flush_pipefs+0x30/0x30
[   43.053594]  [<ffffffff833e652c>] read_flush_procfs+0x2c/0x30
[   43.053596]  [<ffffffff812b9a8c>] proc_reg_read+0x9c/0xd0
[   43.053596]  [<ffffffff812b99f0>] ? proc_reg_write+0xd0/0xd0
[   43.053596]  [<ffffffff81250d5b>] do_loop_readv_writev+0x4b/0x90
[   43.053596]  [<ffffffff81250fd6>] do_readv_writev+0xf6/0x1d0
[   43.053596]  [<ffffffff812510ee>] vfs_readv+0x3e/0x60
[   43.053596]  [<ffffffff812511b8>] sys_readv+0x48/0xb0
[   43.053596]  [<ffffffff8378167d>] system_call_fastpath+0x1a/0x1f

Signed-off-by: Sasha Levin <levinsasha928@gmail.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agoNLM: nlm_lookup_file() may return NLMv4-specific error codes
Trond Myklebust [Sat, 13 Oct 2012 04:30:28 +0000 (00:30 -0400)]
NLM: nlm_lookup_file() may return NLMv4-specific error codes

commit cd0b16c1c3cda12dbed1f8de8f1a9b0591990724 upstream.

If the filehandle is stale, or open access is denied for some reason,
nlm_fopen() may return one of the NLMv4-specific error codes nlm4_stale_fh
or nlm4_failed. These get passed right through nlm_lookup_file(),
and so when nlmsvc_retrieve_args() calls the latter, it needs to filter
the result through the cast_status() machinery.

Failure to do so, will trigger the BUG_ON() in encode_nlm_stat...

Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Reported-by: Larry McVoy <lm@bitmover.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agoALSA: hda - Always check array bounds in alc_get_line_out_pfx
David Henningsson [Wed, 17 Oct 2012 10:43:44 +0000 (12:43 +0200)]
ALSA: hda - Always check array bounds in alc_get_line_out_pfx

commit 71aa5ebe36a4e936eff281b375a4707b6a8320f2 upstream.

Even when CONFIG_SND_DEBUG is not enabled, we don't want to
return an arbitrary memory location when the channel count is
larger than we expected.

Signed-off-by: David Henningsson <david.henningsson@canonical.com>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agocgroup: notify_on_release may not be triggered in some cases
Daisuke Nishimura [Thu, 4 Oct 2012 07:37:16 +0000 (16:37 +0900)]
cgroup: notify_on_release may not be triggered in some cases

commit 1f5320d5972aa50d3e8d2b227b636b370e608359 upstream.

notify_on_release must be triggered when the last process in a cgroup is
move to another. But if the first(and only) process in a cgroup is moved to
another, notify_on_release is not triggered.

# mkdir /cgroup/cpu/SRC
# mkdir /cgroup/cpu/DST
#
# echo 1 >/cgroup/cpu/SRC/notify_on_release
# echo 1 >/cgroup/cpu/DST/notify_on_release
#
# sleep 300 &
[1] 8629
#
# echo 8629 >/cgroup/cpu/SRC/tasks
# echo 8629 >/cgroup/cpu/DST/tasks
-> notify_on_release for /SRC must be triggered at this point,
   but it isn't.

This is because put_css_set() is called before setting CGRP_RELEASABLE
in cgroup_task_migrate(), and is a regression introduce by the
commit:74a1166d(cgroups: make procs file writable), which was merged
into v3.0.

Cc: Ben Blum <bblum@andrew.cmu.edu>
Acked-by: Li Zefan <lizefan@huawei.com>
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: Tejun Heo <tj@kernel.org>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agoiwlwifi: fix 6000 series channel switch command
Johannes Berg [Tue, 25 Sep 2012 14:40:12 +0000 (16:40 +0200)]
iwlwifi: fix 6000 series channel switch command

commit 8f7b8db6e0557c8437adf9371e020cd89a7e85dc upstream.

The channel switch command for 6000 series devices
is larger than the maximum inline command size of
320 bytes. The command is therefore refused with a
warning. Fix this by allocating the command and
using the NOCOPY mechanism.

Reviewed-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
[bwh: Backported to 3.2: adjust filename, context]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agonet/wireless: ipw2200: Fix panic occurring in ipw_handle_promiscuous_tx()
Stanislav Yakovlev [Mon, 15 Oct 2012 14:14:32 +0000 (14:14 +0000)]
net/wireless: ipw2200: Fix panic occurring in ipw_handle_promiscuous_tx()

commit bf11315eeda510ea4fc1a2bf972d8155d31d89b4 upstream.

The driver does not count space of radiotap fields when allocating skb for
radiotap packet. This leads to kernel panic with the following call trace:

...
[67607.676067] [<c152f90f>] error_code+0x67/0x6c
[67607.676067] [<c142f831>] ? skb_put+0x91/0xa0
[67607.676067] [<f8cf5e5b>] ? ipw_handle_promiscuous_tx+0x16b/0x2d0 [ipw2200]
[67607.676067] [<f8cf5e5b>] ipw_handle_promiscuous_tx+0x16b/0x2d0 [ipw2200]
[67607.676067] [<f8cf899b>] ipw_net_hard_start_xmit+0x8b/0x90 [ipw2200]
[67607.676067] [<f8741c5a>] libipw_xmit+0x55a/0x980 [libipw]
[67607.676067] [<c143d3e8>] dev_hard_start_xmit+0x218/0x4d0
...

This bug was found by VittGam.
https://bugzilla.kernel.org/show_bug.cgi?id=43255

Signed-off-by: Stanislav Yakovlev <stas.yakovlev@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agobcma: fix unregistration of cores
Piotr Haber [Thu, 11 Oct 2012 12:05:15 +0000 (14:05 +0200)]
bcma: fix unregistration of cores

commit 1fffa905adffbf0d3767fc978ef09afb830275eb upstream.

When cores are unregistered, entries
need to be removed from cores list in a safe manner.

Reported-by: Stanislaw Gruszka <sgruszka@redhat.com>
Reviewed-by: Arend Van Spriel <arend@broadcom.com>
Signed-off-by: Piotr Haber <phaber@broadcom.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agomac80211: check if key has TKIP type before updating IV
Stanislaw Gruszka [Tue, 2 Oct 2012 19:34:23 +0000 (21:34 +0200)]
mac80211: check if key has TKIP type before updating IV

commit 4045f72bcf3c293c7c5932ef001742d8bb5ded76 upstream.

This patch fix corruption which can manifest itself by following crash
when switching on rfkill switch with rt2x00 driver:
https://bugzilla.redhat.com/attachment.cgi?id=615362

Pointer key->u.ccmp.tfm of group key get corrupted in:

ieee80211_rx_h_michael_mic_verify():

        /* update IV in key information to be able to detect replays */
        rx->key->u.tkip.rx[rx->security_idx].iv32 = rx->tkip_iv32;
        rx->key->u.tkip.rx[rx->security_idx].iv16 = rx->tkip_iv16;

because rt2x00 always set RX_FLAG_MMIC_STRIPPED, even if key is not TKIP.

We already check type of the key in different path in
ieee80211_rx_h_michael_mic_verify() function, so adding additional
check here is reasonable.

Signed-off-by: Stanislaw Gruszka <sgruszka@redhat.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agodrm/radeon: Don't destroy I2C Bus Rec in radeon_ext_tmds_enc_destroy().
Egbert Eich [Mon, 15 Oct 2012 06:21:39 +0000 (08:21 +0200)]
drm/radeon: Don't destroy I2C Bus Rec in radeon_ext_tmds_enc_destroy().

commit 082918471139b07964967cfe5f70230909c82ae1 upstream.

radeon_i2c_fini() walks thru the list of I2C bus recs rdev->i2c_bus[]
to destroy each of them.
radeon_ext_tmds_enc_destroy() however also has code to destroy it's
associated I2C bus rec which has been obtained by radeon_i2c_lookup()
and is therefore also in the i2c_bus[] list.
This causes a double free resulting in a kernel panic when unloading
the radeon driver.
Removing destroy code from radeon_ext_tmds_enc_destroy() fixes this
problem.

agd5f: fix compiler warning

Signed-off-by: Egbert Eich <eich@suse.de>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agooprofile, x86: Fix wrapping bug in op_x86_get_ctrl()
Dan Carpenter [Wed, 10 Oct 2012 07:18:35 +0000 (10:18 +0300)]
oprofile, x86: Fix wrapping bug in op_x86_get_ctrl()

commit 44009105081b51417f311f4c3be0061870b6b8ed upstream.

The "event" variable is a u16 so the shift will always wrap to zero
making the line a no-op.

Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Robert Richter <robert.richter@amd.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agopcmcia: sharpsl: don't discard sharpsl_pcmcia_ops
Arnd Bergmann [Mon, 30 Apr 2012 13:50:56 +0000 (13:50 +0000)]
pcmcia: sharpsl: don't discard sharpsl_pcmcia_ops

commit fdc858a466b738d35d3492bc7cf77b1dac98bf7c upstream.

The sharpsl_pcmcia_ops structure gets passed into
sa11xx_drv_pcmcia_probe, where it gets accessed at run-time,
unlike all other pcmcia drivers that pass their structures
into platform_device_add_data, which makes a copy.

This means the gcc warning is valid and the structure
must not be marked as __initdata.

Without this patch, building collie_defconfig results in:

drivers/pcmcia/pxa2xx_sharpsl.c:22:31: fatal error: mach-pxa/hardware.h: No such file or directory
compilation terminated.
make[3]: *** [drivers/pcmcia/pxa2xx_sharpsl.o] Error 1
make[2]: *** [drivers/pcmcia] Error 2
make[1]: *** [drivers] Error 2
make: *** [sub-make] Error 2

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Cc: Dominik Brodowski <linux@dominikbrodowski.net>
Cc: Russell King <rmk+kernel@arm.linux.org.uk>
Cc: Pavel Machek <pavel@suse.cz>
Cc: linux-pcmcia@lists.infradead.org
Cc: Jochen Friedrich <jochen@scram.de>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agoxen/bootup: allow {read|write}_cr8 pvops call.
Konrad Rzeszutek Wilk [Wed, 10 Oct 2012 17:25:48 +0000 (13:25 -0400)]
xen/bootup: allow {read|write}_cr8 pvops call.

commit 1a7bbda5b1ab0e02622761305a32dc38735b90b2 upstream.

We actually do not do anything about it. Just return a default
value of zero and if the kernel tries to write anything but 0
we BUG_ON.

This fixes the case when an user tries to suspend the machine
and it blows up in save_processor_state b/c 'read_cr8' is set
to NULL and we get:

kernel BUG at /home/konrad/ssd/linux/arch/x86/include/asm/paravirt.h:100!
invalid opcode: 0000 [#1] SMP
Pid: 2687, comm: init.late Tainted: G           O 3.6.0upstream-00002-gac264ac-dirty #4 Bochs Bochs
RIP: e030:[<ffffffff814d5f42>]  [<ffffffff814d5f42>] save_processor_state+0x212/0x270

.. snip..
Call Trace:
 [<ffffffff810733bf>] do_suspend_lowlevel+0xf/0xac
 [<ffffffff8107330c>] ? x86_acpi_suspend_lowlevel+0x10c/0x150
 [<ffffffff81342ee2>] acpi_suspend_enter+0x57/0xd5

Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agoxen/bootup: allow read_tscp call for Xen PV guests.
Konrad Rzeszutek Wilk [Wed, 10 Oct 2012 17:30:47 +0000 (13:30 -0400)]
xen/bootup: allow read_tscp call for Xen PV guests.

commit cd0608e71e9757f4dae35bcfb4e88f4d1a03a8ab upstream.

The hypervisor will trap it. However without this patch,
we would crash as the .read_tscp is set to NULL. This patch
fixes it and sets it to the native_read_tscp call.

Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agomips,kgdb: fix recursive page fault with CONFIG_KPROBES
Jason Wessel [Fri, 10 Aug 2012 17:21:15 +0000 (12:21 -0500)]
mips,kgdb: fix recursive page fault with CONFIG_KPROBES

commit f0a996eeeda214f4293e234df33b29bec003b536 upstream.

This fault was detected using the kgdb test suite on boot and it
crashes recursively due to the fact that CONFIG_KPROBES on mips adds
an extra die notifier in the page fault handler.  The crash signature
looks like this:

kgdbts:RUN bad memory access test
KGDB: re-enter exception: ALL breakpoints killed
Call Trace:
[<807b7548>] dump_stack+0x20/0x54
[<807b7548>] dump_stack+0x20/0x54

The fix for now is to have kgdb return immediately if the fault type
is DIE_PAGE_FAULT and allow the kprobe code to decide what is supposed
to happen.

Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: David S. Miller <davem@davemloft.net>
Signed-off-by: Jason Wessel <jason.wessel@windriver.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agoBluetooth: SMP: Fix setting unknown auth_req bits
Johan Hedberg [Thu, 11 Oct 2012 14:26:06 +0000 (16:26 +0200)]
Bluetooth: SMP: Fix setting unknown auth_req bits

commit 065a13e2cc665f6547dc7e8a9d6b6565badf940a upstream.

When sending a pairing request or response we should not just blindly
copy the value that the remote device sent. Instead we should at least
make sure to mask out any unknown bits. This is particularly critical
from the upcoming LE Secure Connections feature perspective as
incorrectly indicating support for it (by copying the remote value)
would cause a failure to pair with devices that support it.

Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Acked-by: Marcel Holtmann <marcel@holtmann.org>
Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agomd/raid10: use correct limit variable
Dan Carpenter [Thu, 11 Oct 2012 03:20:58 +0000 (14:20 +1100)]
md/raid10: use correct limit variable

commit 91502f099dfc5a1e8812898e26ee280713e1d002 upstream.

Clang complains that we are assigning a variable to itself.  This should
be using bad_sectors like the similar earlier check does.

Bug has been present since 3.1-rc1.  It is minor but could
conceivably cause corruption or other bad behaviour.

Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: NeilBrown <neilb@suse.de>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agopktgen: fix crash when generating IPv6 packets
Amerigo Wang [Tue, 9 Oct 2012 17:48:16 +0000 (17:48 +0000)]
pktgen: fix crash when generating IPv6 packets

commit 5aa8b572007c4bca1e6d3dd4c4820f1ae49d6bb2 upstream.

For IPv6, sizeof(struct ipv6hdr) = 40, thus the following
expression will result negative:

        datalen = pkt_dev->cur_pkt_size - 14 -
                  sizeof(struct ipv6hdr) - sizeof(struct udphdr) -
                  pkt_dev->pkt_overhead;

And,  the check "if (datalen < sizeof(struct pktgen_hdr))" will be
passed as "datalen" is promoted to unsigned, therefore will cause
a crash later.

This is a quick fix by checking if "datalen" is negative. The following
patch will increase the default value of 'min_pkt_size' for IPv6.

This bug should exist for a long time, so Cc -stable too.

Cc: David S. Miller <davem@davemloft.net>
Signed-off-by: Cong Wang <amwang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agoALSA: hda - do not detect jack on internal speakers for Realtek
David Henningsson [Wed, 10 Oct 2012 14:32:09 +0000 (16:32 +0200)]
ALSA: hda - do not detect jack on internal speakers for Realtek

commit f7f4b2322bf7b8c5929b7eb5a667091f32592580 upstream.

This caused the internal speaker to mute itself because it was
present, which happened after powersave.
It was found on Dell XPS 15 (L502x), ALC665.

Reported-by: Da Fox <da.fox.mail@gmail.com>
Signed-off-by: David Henningsson <david.henningsson@canonical.com>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agonohz: Fix idle ticks in cpu summary line of /proc/stat
Michal Hocko [Wed, 10 Oct 2012 06:21:09 +0000 (11:51 +0530)]
nohz: Fix idle ticks in cpu summary line of /proc/stat

commit 7386cdbf2f57ea8cff3c9fde93f206e58b9fe13f upstream.

Git commit 09a1d34f8535ecf9 "nohz: Make idle/iowait counter update
conditional" introduced a bug in regard to cpu hotplug. The effect is
that the number of idle ticks in the cpu summary line in /proc/stat is
still counting ticks for offline cpus.

Reproduction is easy, just start a workload that keeps all cpus busy,
switch off one or more cpus and then watch the idle field in top.
On a dual-core with one cpu 100% busy and one offline cpu you will get
something like this:

%Cpu(s): 48.7 us,  1.3 sy,  0.0 ni, 50.0 id,  0.0 wa,  0.0 hi,  0.0 si,
%0.0 st

The problem is that an offline cpu still has ts->idle_active == 1.
To fix this we should make sure that the cpu is online when calling
get_cpu_idle_time_us and get_cpu_iowait_time_us.

[Srivatsa: Rebased to current mainline]

Reported-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Michal Hocko <mhocko@suse.cz>
Reviewed-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Link: http://lkml.kernel.org/r/20121010061820.8999.57245.stgit@srivatsabhat.in.ibm.com
Cc: deepthi@linux.vnet.ibm.com
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agoALSA: hda - Fix memory leaks at error path in patch_cirrus.c
Takashi Iwai [Wed, 10 Oct 2012 06:50:35 +0000 (08:50 +0200)]
ALSA: hda - Fix memory leaks at error path in patch_cirrus.c

commit c5e0b6dbad9b4d18c561af90b384d02373f1c994 upstream.

The proper destructor should be called at the error path.

Signed-off-by: Takashi Iwai <tiwai@suse.de>
[bwh: Backported to 3.2: drop the change to nonexistent patch_cs4213()]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agoext4: race-condition protection for ext4_convert_unwritten_extents_endio
Dmitry Monakhov [Wed, 10 Oct 2012 05:04:58 +0000 (01:04 -0400)]
ext4: race-condition protection for ext4_convert_unwritten_extents_endio

commit dee1f973ca341c266229faa5a1a5bb268bed3531 upstream.

We assumed that at the time we call ext4_convert_unwritten_extents_endio()
extent in question is fully inside [map.m_lblk, map->m_len] because
it was already split during submission.  But this may not be true due to
a race between writeback vs fallocate.

If extent in question is larger than requested we will split it again.
Special precautions should being done if zeroout required because
[map.m_lblk, map->m_len] already contains valid data.

Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agotmpfs,ceph,gfs2,isofs,reiserfs,xfs: fix fh_len checking
Hugh Dickins [Mon, 8 Oct 2012 03:32:51 +0000 (20:32 -0700)]
tmpfs,ceph,gfs2,isofs,reiserfs,xfs: fix fh_len checking

commit 35c2a7f4908d404c9124c2efc6ada4640ca4d5d5 upstream.

Fuzzing with trinity oopsed on the 1st instruction of shmem_fh_to_dentry(),
u64 inum = fid->raw[2];
which is unhelpfully reported as at the end of shmem_alloc_inode():

BUG: unable to handle kernel paging request at ffff880061cd3000
IP: [<ffffffff812190d0>] shmem_alloc_inode+0x40/0x40
Oops: 0000 [#1] PREEMPT SMP DEBUG_PAGEALLOC
Call Trace:
 [<ffffffff81488649>] ? exportfs_decode_fh+0x79/0x2d0
 [<ffffffff812d77c3>] do_handle_open+0x163/0x2c0
 [<ffffffff812d792c>] sys_open_by_handle_at+0xc/0x10
 [<ffffffff83a5f3f8>] tracesys+0xe1/0xe6

Right, tmpfs is being stupid to access fid->raw[2] before validating that
fh_len includes it: the buffer kmalloc'ed by do_sys_name_to_handle() may
fall at the end of a page, and the next page not be present.

But some other filesystems (ceph, gfs2, isofs, reiserfs, xfs) are being
careless about fh_len too, in fh_to_dentry() and/or fh_to_parent(), and
could oops in the same way: add the missing fh_len checks to those.

Reported-by: Sasha Levin <levinsasha928@gmail.com>
Signed-off-by: Hugh Dickins <hughd@google.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Sage Weil <sage@inktank.com>
Cc: Steven Whitehouse <swhiteho@redhat.com>
Cc: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agovideo/udlfb: fix line counting in fb_write
Alexander Holler [Tue, 14 Aug 2012 07:11:09 +0000 (09:11 +0200)]
video/udlfb: fix line counting in fb_write

commit b8c4321f3d194469007f5f5f2b34ec278c264a04 upstream.

Line 0 and 1 were both written to line 0 (on the display) and all subsequent
lines had an offset of -1. The result was that the last line on the display
was never overwritten by writes to /dev/fbN.

Signed-off-by: Alexander Holler <holler@ahsoftware.de>
Acked-by: Bernie Thompson <bernie@plugable.com>
Signed-off-by: Florian Tobias Schandinat <FlorianSchandinat@gmx.de>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agotimekeeping: Cast raw_interval to u64 to avoid shift overflow
Dan Carpenter [Tue, 9 Oct 2012 07:18:23 +0000 (10:18 +0300)]
timekeeping: Cast raw_interval to u64 to avoid shift overflow

commit 5b3900cd409466c0070b234d941650685ad0c791 upstream.

We fixed a bunch of integer overflows in timekeeping code during the 3.6
cycle.  I did an audit based on that and found this potential overflow.

Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Acked-by: John Stultz <johnstul@us.ibm.com>
Link: http://lkml.kernel.org/r/20121009071823.GA19159@elgon.mountain
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
[bwh: Backported to 3.2: adjust context; use timekeeper.raw_interval]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agotimers: Fix endless looping between cascade() and internal_add_timer()
Hildner, Christian [Mon, 8 Oct 2012 13:49:03 +0000 (15:49 +0200)]
timers: Fix endless looping between cascade() and internal_add_timer()

commit 26cff4e2aa4d666dc6a120ea34336b5057e3e187 upstream.

Adding two (or more) timers with large values for "expires" (they have
to reside within tv5 in the same list) leads to endless looping
between cascade() and internal_add_timer() in case CONFIG_BASE_SMALL
is one and jiffies are crossing the value 1 << 18. The bug was
introduced between 2.6.11 and 2.6.12 (and survived for quite some
time).

This patch ensures that when cascade() is called timers within tv5 are
not added endlessly to their own list again, instead they are added to
the next lower tv level tv4 (as expected).

Signed-off-by: Christian Hildner <christian.hildner@siemens.com>
Reviewed-by: Jan Kiszka <jan.kiszka@siemens.com>
Link: http://lkml.kernel.org/r/98673C87CB31274881CFFE0B65ECC87B0F5FC1963E@DEFTHW99EA4MSX.ww902.siemens.net
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agostorvsc: Account for in-transit packets in the RESET path
K. Y. Srinivasan [Tue, 2 Oct 2012 18:03:31 +0000 (11:03 -0700)]
storvsc: Account for in-transit packets in the RESET path

commit 5c1b10ab7f93d24f29b5630286e323d1c5802d5c upstream.

Properly account for I/O in transit before returning from the RESET call.
In the absense of this patch, we could have a situation where the host may
respond to a command that was issued prior to the issuance of the RESET
command at some arbitrary time after responding to the RESET command.
Currently, the host does not do anything with the RESET command.

Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: James Bottomley <JBottomley@Parallels.com>
[bwh: Backported to 3.2: adjust filename, context]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agoscsi_debug: Fix off-by-one bug when unmapping region
Lukas Czerner [Thu, 16 Aug 2012 14:38:45 +0000 (16:38 +0200)]
scsi_debug: Fix off-by-one bug when unmapping region

commit bc977749e967daa56de1922cf4cb38525631c51c upstream.

Currently it is possible to unmap one more block than user requested to
due to the off-by-one error in unmap_region(). This is probably due to
the fact that the end variable despite its name actually points to the
last block to unmap + 1. However in the condition it is handled as the
last block of the region to unmap.

The bug was not previously spotted probably due to the fact that the
region was not zeroed, which has changed with commit
be1dd78de5686c062bb3103f9e86d444a10ed783. With that commit we were able
to corrupt the ext4 file system on 256M scsi_debug device with LBPRZ
enabled using fstrim.

Since the 'end' semantic is the same in several functions there this
commit just fixes the condition to use the 'end' variable correctly in
that context.

Reported-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Lukas Czerner <lczerner@redhat.com>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Acked-by: Douglas Gilbert <dgilbert@interlog.com>
Signed-off-by: James Bottomley <JBottomley@Parallels.com>
[bwh: Backported to 3.2: adjust context; unwrap the if-statement]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agoARM: vfp: fix saving d16-d31 vfp registers on v6+ kernels
Russell King [Tue, 9 Oct 2012 10:13:26 +0000 (11:13 +0100)]
ARM: vfp: fix saving d16-d31 vfp registers on v6+ kernels

commit 846a136881b8f73c1f74250bf6acfaa309cab1f2 upstream.

Michael Olbrich reported that his test program fails when built with
-O2 -mcpu=cortex-a8 -mfpu=neon, and a kernel which supports v6 and v7
CPUs:

volatile int x = 2;
volatile int64_t y = 2;

int main() {
volatile int a = 0;
volatile int64_t b = 0;
while (1) {
a = (a + x) % (1 << 30);
b = (b + y) % (1 << 30);
assert(a == b);
}
}

and two instances are run.  When built for just v7 CPUs, this program
works fine.  It uses the "vadd.i64 d19, d18, d16" VFP instruction.

It appears that we do not save the high-16 double VFP registers across
context switches when the kernel is built for v6 CPUs.  Fix that.

Tested-By: Michael Olbrich <m.olbrich@pengutronix.de>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agoACPI: EC: Add a quirk for CLEVO M720T/M730T laptop
Feng Tang [Fri, 28 Sep 2012 07:22:01 +0000 (15:22 +0800)]
ACPI: EC: Add a quirk for CLEVO M720T/M730T laptop

commit 67bfa9b60bd689601554526d144b21d529f78a09 upstream.

By enlarging the GPE storm threshold back to 20, that laptop's
EC works fine with interrupt mode instead of polling mode.

https://bugzilla.kernel.org/show_bug.cgi?id=45151

Reported-and-Tested-by: Francesco <trentini@dei.unipd.it>
Signed-off-by: Feng Tang <feng.tang@intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agoACPI: EC: Make the GPE storm threshold a module parameter
Feng Tang [Fri, 28 Sep 2012 07:22:00 +0000 (15:22 +0800)]
ACPI: EC: Make the GPE storm threshold a module parameter

commit a520d52e99b14ba7db135e916348f12f2a6e09be upstream.

The Linux EC driver includes a mechanism to detect GPE storms,
and switch from interrupt-mode to polling mode.  However, polling
mode sometimes doesn't work, so the workaround is problematic.
Also, different systems seem to need the threshold for detecting
the GPE storm at different levels.

ACPI_EC_STORM_THRESHOLD was initially 20 when it's created, and
was changed to 8 in 2.6.28 commit 06cf7d3c7 "ACPI: EC: lower interrupt storm
threshold" to fix kernel bug 11892 by forcing the laptop in that bug to
work in polling mode. However in bug 45151, it works fine in interrupt
mode if we lift the threshold back to 20.

This patch makes the threshold a module parameter so that user has a
flexible option to debug/workaround this issue.

The default is unchanged.

This is also a preparation patch to fix specific systems:
https://bugzilla.kernel.org/show_bug.cgi?id=45151

Signed-off-by: Feng Tang <feng.tang@intel.com>
Signed-off-by: Len Brown <len.brown@intel.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agomodule: taint kernel when lve module is loaded
Matthew Garrett [Fri, 22 Jun 2012 17:49:31 +0000 (13:49 -0400)]
module: taint kernel when lve module is loaded

commit c99af3752bb52ba3aece5315279a57a477edfaf1 upstream.

Cloudlinux have a product called lve that includes a kernel module. This
was previously GPLed but is now under a proprietary license, but the
module continues to declare MODULE_LICENSE("GPL") and makes use of some
EXPORT_SYMBOL_GPL symbols. Forcibly taint it in order to avoid this.

Signed-off-by: Matthew Garrett <mjg59@srcf.ucam.org>
Cc: Alex Lyashkov <umka@cloudlinux.com>
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agoviafb: don't touch clock state on OLPC XO-1.5
Daniel Drake [Tue, 4 Sep 2012 15:45:32 +0000 (11:45 -0400)]
viafb: don't touch clock state on OLPC XO-1.5

commit 012a1211845eab69a5488d59eb87d24cc518c627 upstream.

As detailed in the thread titled "viafb PLL/clock tweaking causes XO-1.5
instability," enabling or disabling the IGA1/IGA2 clocks causes occasional
stability problems during suspend/resume cycles on this platform.

This is rather odd, as the documentation suggests that clocks have two
states (on/off) and the default (stable) configuration is configured to
enable the clock only when it is needed. However, explicitly enabling *or*
disabling the clock triggers this system instability, suggesting that there
is a 3rd state at play here.

Leaving the clock enable/disable registers alone solves this problem.
This fixes spurious reboots during suspend/resume behaviour introduced by
commit b692a63a.

Signed-off-by: Daniel Drake <dsd@laptop.org>
Signed-off-by: Florian Tobias Schandinat <FlorianSchandinat@gmx.de>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agonfsd4: fix nfs4 stateid leak
J. Bruce Fields [Wed, 29 Aug 2012 22:21:58 +0000 (15:21 -0700)]
nfsd4: fix nfs4 stateid leak

commit cf9182e90b2af04245ac4fae497fe73fc71285b4 upstream.

Processes that open and close multiple files may end up setting this
oo_last_closed_stid without freeing what was previously pointed to.
This can result in a major leak, visible for example by watching the
nfsd4_stateids line of /proc/slabinfo.

Reported-by: Cyril B. <cbay@excellency.fr>
Tested-by: Cyril B. <cbay@excellency.fr>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agostaging: comedi: amplc_pc236: fix invalid register access during detach
Ian Abbott [Wed, 3 Oct 2012 15:25:17 +0000 (16:25 +0100)]
staging: comedi: amplc_pc236: fix invalid register access during detach

commit aaeb61a97b7159ebe30b18a422d04eeabfa8790b upstream.

`pc236_detach()` is called by the comedi core if it attempted to attach
a device and failed.  `pc236_detach()` calls `pc236_intr_disable()` if
the comedi device private data pointer (`devpriv`) is non-null.  This
test is insufficient as `pc236_intr_disable()` accesses hardware
registers and the attach routine may have failed before it has saved
their I/O base addresses.

Fix it by checking `dev->iobase` is non-zero before calling
`pc236_intr_disable()` as that means the I/O base addresses have been
saved and the hardware registers can be accessed.  It also implies the
comedi device private data pointer is valid, so there is no need to
check it.

Signed-off-by: Ian Abbott <abbotti@mev.co.uk>
[Ian Abbott: This patch is for the stable 3.0 kernel.]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agotcp: resets are misrouted
Alexey Kuznetsov [Fri, 12 Oct 2012 04:34:17 +0000 (04:34 +0000)]
tcp: resets are misrouted

[ Upstream commit 4c67525849e0b7f4bd4fab2487ec9e43ea52ef29 ]

After commit e2446eaa ("tcp_v4_send_reset: binding oif to iif in no
sock case").. tcp resets are always lost, when routing is asymmetric.
Yes, backing out that patch will result in misrouting of resets for
dead connections which used interface binding when were alive, but we
actually cannot do anything here.  What's died that's died and correct
handling normal unbound connections is obviously a priority.

Comment to comment:
> This has few benefits:
>   1. tcp_v6_send_reset already did that.

It was done to route resets for IPv6 link local addresses. It was a
mistake to do so for global addresses. The patch fixes this as well.

Actually, the problem appears to be even more serious than guaranteed
loss of resets.  As reported by Sergey Soloviev <sol@eqv.ru>, those
misrouted resets create a lot of arp traffic and huge amount of
unresolved arp entires putting down to knees NAT firewalls which use
asymmetric routing.

Signed-off-by: Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agoRDS: fix rds-ping spinlock recursion
jeff.liu [Mon, 8 Oct 2012 18:57:27 +0000 (18:57 +0000)]
RDS: fix rds-ping spinlock recursion

[ Upstream commit 5175a5e76bbdf20a614fb47ce7a38f0f39e70226 ]

This is the revised patch for fixing rds-ping spinlock recursion
according to Venkat's suggestions.

RDS ping/pong over TCP feature has been broken for years(2.6.39 to
3.6.0) since we have to set TCP cork and call kernel_sendmsg() between
ping/pong which both need to lock "struct sock *sk". However, this
lock has already been hold before rds_tcp_data_ready() callback is
triggerred. As a result, we always facing spinlock resursion which
would resulting in system panic.

Given that RDS ping is only used to test the connectivity and not for
serious performance measurements, we can queue the pong transmit to
rds_wq as a delayed response.

Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
CC: Venkat Venkatsubra <venkat.x.venkatsubra@oracle.com>
CC: David S. Miller <davem@davemloft.net>
CC: James Morris <james.l.morris@oracle.com>
Signed-off-by: Jie Liu <jeff.liu@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agovlan: don't deliver frames for unknown vlans to protocols
Florian Zumbiehl [Sun, 7 Oct 2012 15:51:58 +0000 (15:51 +0000)]
vlan: don't deliver frames for unknown vlans to protocols

[ Upstream commit 48cc32d38a52d0b68f91a171a8d00531edc6a46e ]

6a32e4f9dd9219261f8856f817e6655114cfec2f made the vlan code skip marking
vlan-tagged frames for not locally configured vlans as PACKET_OTHERHOST if
there was an rx_handler, as the rx_handler could cause the frame to be received
on a different (virtual) vlan-capable interface where that vlan might be
configured.

As rx_handlers do not necessarily return RX_HANDLER_ANOTHER, this could cause
frames for unknown vlans to be delivered to the protocol stack as if they had
been received untagged.

For example, if an ipv6 router advertisement that's tagged for a locally not
configured vlan is received on an interface with macvlan interfaces attached,
macvlan's rx_handler returns RX_HANDLER_PASS after delivering the frame to the
macvlan interfaces, which caused it to be passed to the protocol stack, leading
to ipv6 addresses for the announced prefix being configured even though those
are completely unusable on the underlying interface.

The fix moves marking as PACKET_OTHERHOST after the rx_handler so the
rx_handler, if there is one, sees the frame unchanged, but afterwards,
before the frame is delivered to the protocol stack, it gets marked whether
there is an rx_handler or not.

Signed-off-by: Florian Zumbiehl <florz@florz.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agoskge: Add DMA mask quirk for Marvell 88E8001 on ASUS P5NSLI motherboard
Graham Gower [Mon, 8 Oct 2012 08:34:50 +0000 (08:34 +0000)]
skge: Add DMA mask quirk for Marvell 88E8001 on ASUS P5NSLI motherboard

[ Upstream commit a2af139ff1cd85df586690ff626619ab1ee88b0a ]

Marvell 88E8001 on an ASUS P5NSLI motherboard is unable to send/receive
packets on a system with >4gb ram unless a 32bit DMA mask is used.

This issue has been around for years and a fix was sent 3.5 years ago, but
there was some debate as to whether it should instead be fixed as a PCI quirk.
http://www.spinics.net/lists/netdev/msg88670.html

However, 18 months later a similar workaround was introduced for another
chipset exhibiting the same problem.
http://www.spinics.net/lists/netdev/msg142287.html

Signed-off-by: Graham Gower <graham.gower@gmail.com>
Signed-off-by: Jan Ceuleers <jan.ceuleers@computer.org>
Acked-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agonet: Fix skb_under_panic oops in neigh_resolve_output
ramesh.nagappa@gmail.com [Fri, 5 Oct 2012 19:10:15 +0000 (19:10 +0000)]
net: Fix skb_under_panic oops in neigh_resolve_output

[ Upstream commit e1f165032c8bade3a6bdf546f8faf61fda4dd01c ]

The retry loop in neigh_resolve_output() and neigh_connected_output()
call dev_hard_header() with out reseting the skb to network_header.
This causes the retry to fail with skb_under_panic. The fix is to
reset the network_header within the retry loop.

Signed-off-by: Ramesh Nagappa <ramesh.nagappa@ericsson.com>
Reviewed-by: Shawn Lu <shawn.lu@ericsson.com>
Reviewed-by: Robert Coulson <robert.coulson@ericsson.com>
Reviewed-by: Billie Alsup <billie.alsup@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agoau0828: fix case where STREAMOFF being called on stopped stream causes BUG()
Devin Heitmueller [Tue, 7 Aug 2012 01:47:03 +0000 (22:47 -0300)]
au0828: fix case where STREAMOFF being called on stopped stream causes BUG()

commit a595c1ce4c9d572cf53513570b9f1a263d7867f2 upstream.

We weren't checking whether the resource was in use before calling
res_free(), so applications which called STREAMOFF on a v4l2 device that
wasn't already streaming would cause a BUG() to be hit (MythTV).

Reported-by: Larry Finger <larry.finger@lwfinger.net>
Reported-by: Jay Harbeston <jharbestonus@gmail.com>
Signed-off-by: Devin Heitmueller <dheitmueller@kernellabs.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agosparc64: Be less verbose during vmemmap population.
David S. Miller [Wed, 15 Aug 2012 07:37:29 +0000 (00:37 -0700)]
sparc64: Be less verbose during vmemmap population.

[ Upstream commit 2856cc2e4d0852c3ddaae9dcb19cb9396512eb08 ]

On a 2-node machine with 256GB of ram we get 512 lines of
console output, which is just too much.

This mimicks Yinghai Lu's x86 commit c2b91e2eec9678dbda274e906cc32ea8f711da3b
(x86_64/mm: check and print vmemmap allocation continuous) except that
we aren't ever going to get contiguous block pointers in between calls
so just print when the virtual address or node changes.

This decreases the output by an order of 16.

Also demote this to KERN_DEBUG.

Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agosparc64: do not clobber personality flags in sys_sparc64_personality()
Jiri Kosina [Wed, 1 Aug 2012 19:10:51 +0000 (21:10 +0200)]
sparc64: do not clobber personality flags in sys_sparc64_personality()

[ Upstream commit a27032eee8cb6e16516f13c8a9752e9d5d4cc430 ]

There are multiple errors in how sys_sparc64_personality() handles
personality flags stored in top three bytes.

- directly comparing current->personality against PER_LINUX32 doesn't work
  in cases when any of the personality flags stored in the top three bytes
  are used.
- directly forcefully setting personality to PER_LINUX32 or PER_LINUX
  discards any flags stored in the top three bytes

Fix the first one by properly using personality() macro to compare only
PER_MASK bytes.
Fix the second one by setting only the bits that should be set, instead of
overwriting the whole value.

Signed-off-by: Jiri Kosina <jkosina@suse.cz>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agosparc64: Fix bit twiddling in sparc_pmu_enable_event().
David S. Miller [Tue, 16 Oct 2012 20:05:25 +0000 (13:05 -0700)]
sparc64: Fix bit twiddling in sparc_pmu_enable_event().

[ Upstream commit e793d8c6740f8fe704fa216e95685f4d92c4c4b9 ]

There was a serious disconnect in the logic happening in
sparc_pmu_disable_event() vs. sparc_pmu_enable_event().

Event disable is implemented by programming a NOP event into the PCR.

However, event enable was not reversing this operation.  Instead, it
was setting the User/Priv/Hypervisor trace enable bits.

That's not sparc_pmu_enable_event()'s job, that's what
sparc_pmu_enable() and sparc_pmu_disable() do .

The intent of sparc_pmu_enable_event() is clear, since it first clear
out the event type encoding field.  So fix this by OR'ing in the event
encoding rather than the trace enable bits.

Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agosparc64: Like x86 we should check current->mm during perf backtrace generation.
David S. Miller [Mon, 15 Oct 2012 00:59:40 +0000 (17:59 -0700)]
sparc64: Like x86 we should check current->mm during perf backtrace generation.

[ Upstream commit 08280e6c4c2e8049ac61d9e8e3536ec1df629c0d ]

If the MM is not active, only report the top-level PC.  Do not try to
access the address space.

Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agosparc64: fix ptrace interaction with force_successful_syscall_return()
Al Viro [Thu, 11 Oct 2012 00:25:00 +0000 (17:25 -0700)]
sparc64: fix ptrace interaction with force_successful_syscall_return()

[ Upstream commit 55c2770e413e96871147b9406a9c41fe9bc5209c ]

we want syscall_trace_leave() called on exit from any syscall;
skipping its call in case we'd done force_successful_syscall_return()
is broken...

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agodrm/i915: no lvds quirk for Zotac ZDBOX SD ID12/ID13
Sjoerd Simons [Fri, 22 Jun 2012 07:43:07 +0000 (09:43 +0200)]
drm/i915: no lvds quirk for Zotac ZDBOX SD ID12/ID13

commit 9756fe38d10b2bf90c81dc4d2f17d5632e135364 upstream.

This box claims to have an LVDS interface but doesn't
actually have one.

Signed-off-by: Sjoerd Simons <sjoerd.simons@collabora.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agodrm/i915: don't pwrite tiled objects through the gtt
Daniel Vetter [Fri, 13 Apr 2012 13:51:51 +0000 (15:51 +0200)]
drm/i915: don't pwrite tiled objects through the gtt

commit c07496fa61f4c5cb2addd1c57f6b22fcaeea2eeb upstream.

... we will botch up the bit17 swizzling. Furthermore tiled pwrite is
a (now) unused slowpath, so no one really cares.

This fixes the last swizzling issues I have with i-g-t on my bit17
swizzling i915G. No regression, it's been broken since the dawn of
gem, but it's nice for regression tracking when really _all_ i-g-t
tests work.

Actually this is not true, Chris Wilson noticed while reviewing this
patch that the commit

commit d9e86c0ee60f323e890484628f351bf50fa9a15d
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date:   Wed Nov 10 16:40:20 2010 +0000

    drm/i915: Pipelined fencing [infrastructure]

contained a functional change that broke things.

Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
[Luís Picciochi Oliveira: I picked the commit Daniel suggested and
edited it to affect the code that matched the same logical location on
the same function.]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agoSUNRPC: Set alloc_slot for backchannel tcp ops
Bryan Schumaker [Mon, 24 Sep 2012 17:39:01 +0000 (13:39 -0400)]
SUNRPC: Set alloc_slot for backchannel tcp ops

commit 84e28a307e376f271505af65a7b7e212dd6f61f4 upstream.

f39c1bfb5a03e2d255451bff05be0d7255298fa4 (SUNRPC: Fix a UDP transport
regression) introduced the "alloc_slot" function for xprt operations,
but never created one for the backchannel operations.  This patch fixes
a null pointer dereference when mounting NFS over v4.1.

Call Trace:
 [<ffffffffa0207957>] ? xprt_reserve+0x47/0x50 [sunrpc]
 [<ffffffffa02023a4>] call_reserve+0x34/0x60 [sunrpc]
 [<ffffffffa020e280>] __rpc_execute+0x90/0x400 [sunrpc]
 [<ffffffffa020e61a>] rpc_async_schedule+0x2a/0x40 [sunrpc]
 [<ffffffff81073589>] process_one_work+0x139/0x500
 [<ffffffff81070e70>] ? alloc_worker+0x70/0x70
 [<ffffffffa020e5f0>] ? __rpc_execute+0x400/0x400 [sunrpc]
 [<ffffffff81073d1e>] worker_thread+0x15e/0x460
 [<ffffffff8145c839>] ? preempt_schedule+0x49/0x70
 [<ffffffff81073bc0>] ? rescuer_thread+0x230/0x230
 [<ffffffff81079603>] kthread+0x93/0xa0
 [<ffffffff81465d04>] kernel_thread_helper+0x4/0x10
 [<ffffffff81079570>] ? kthread_freezable_should_stop+0x70/0x70
 [<ffffffff81465d00>] ? gs_change+0x13/0x13

Signed-off-by: Bryan Schumaker <bjschuma@netapp.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agoe1000: fix vlan processing regression
Jiri Pirko [Tue, 20 Mar 2012 18:10:01 +0000 (18:10 +0000)]
e1000: fix vlan processing regression

commit 52f5509fe8ccb607ff9b84ad618f244262336475 upstream.

This patch fixes a regression introduced by commit "e1000: do vlan
cleanup (799d531)".

Apparently some e1000 chips (not mine) are sensitive about the order of
setting vlan filter and vlan stripping/inserting functionality. So this
patch changes the order so it's the same as before vlan cleanup.

Reported-by: Ben Greear <greearb@candelatech.com>
Signed-off-by: Jiri Pirko <jpirko@redhat.com>
Tested-by: Ben Greear <greearb@candelatech.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
[Jonathan Nieder: It doesn't apply cleanly to kernels before
v3.3-rc1~182^2~581 (net: introduce and use netdev_features_t for
device features sets) but a backport is straightforward.]
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Tested-by: Andrey Jr. Melnikov <temnota@kmv.ru>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agoALSA: hda - Fix oops caused by recent commit "Fix internal mic for Lenovo Ideapad...
David Henningsson [Tue, 10 Apr 2012 11:05:29 +0000 (13:05 +0200)]
ALSA: hda - Fix oops caused by recent commit "Fix internal mic for Lenovo Ideapad U300s"

commit 83b0c6ba999643ee8ad6329f26e1cdc870e1a920 upstream.

Make sure we don't dereference the "quirk" pointer when it is null.

Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: David Henningsson <david.henningsson@canonical.com>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agomtd: nand: allow NAND_NO_SUBPAGE_WRITE to be set from driver
Brian Norris [Fri, 13 Jul 2012 16:28:24 +0000 (09:28 -0700)]
mtd: nand: allow NAND_NO_SUBPAGE_WRITE to be set from driver

commit bf7a01bf7987b63b121d572b240c132ec44129c4 upstream.

The NAND_CHIPOPTIONS_MSK has limited utility and is causing real bugs. It
silently masks off at least one flag that might be set by the driver
(NAND_NO_SUBPAGE_WRITE). This breaks the GPMI NAND driver and possibly
others.

Really, as long as driver writers exercise a small amount of care with
NAND_* options, this mask is not necessary at all; it was only here to
prevent certain options from accidentally being set by the driver. But the
original thought turns out to be a bad idea occasionally. Thus, kill it.

Note, this patch fixes some major gpmi-nand breakage.

Signed-off-by: Brian Norris <computersforpeace@gmail.com>
Tested-by: Huang Shijie <shijie8@gmail.com>
Cc: stable@vger.kernel.org
Signed-off-by: Artem Bityutskiy <artem.bityutskiy@linux.intel.com>
Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>
[Brian Norris: This is a backport for v3.2 stable.]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agojbd: Fix assertion failure in commit code due to lacking transaction credits
Jan Kara [Wed, 11 Jul 2012 21:16:25 +0000 (23:16 +0200)]
jbd: Fix assertion failure in commit code due to lacking transaction credits

commit 09e05d4805e6c524c1af74e524e5d0528bb3fef3 upstream.

ext3 users of data=journal mode with blocksize < pagesize were occasionally
hitting assertion failure in journal_commit_transaction() checking whether the
transaction has at least as many credits reserved as buffers attached.  The
core of the problem is that when a file gets truncated, buffers that still need
checkpointing or that are attached to the committing transaction are left with
buffer_mapped set. When this happens to buffers beyond i_size attached to a
page stradding i_size, subsequent write extending the file will see these
buffers and as they are mapped (but underlying blocks were freed) things go
awry from here.

The assertion failure just coincidentally (and in this case luckily as we would
start corrupting filesystem) triggers due to journal_head not being properly
cleaned up as well.

Under some rare circumstances this bug could even hit data=ordered mode users.
There the assertion won't trigger and we would end up corrupting the
filesystem.

We fix the problem by unmapping buffers if possible (in lots of cases we just
need a buffer attached to a transaction as a place holder but it must not be
written out anyway). And in one case, we just have to bite the bullet and wait
for transaction commit to finish.

Reviewed-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agosamsung-laptop: make the dmi check less strict (part 2)
Ben Hutchings [Sun, 9 Sep 2012 17:49:14 +0000 (18:49 +0100)]
samsung-laptop: make the dmi check less strict (part 2)

commit 3be324a94df0c3f032178d04549dbfbf6cccb09a upstream.

These are the hunks that I dropped when backporting for 3.2.24,
which are applicable now that we also have commit f34cd9ca
('samsung-laptop: don't handle backlight if handled by acpi/video').

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agosamsung-laptop: don't handle backlight if handled by acpi/video
Corentin Chary [Sat, 26 Nov 2011 10:00:00 +0000 (11:00 +0100)]
samsung-laptop: don't handle backlight if handled by acpi/video

commit f34cd9ca9320876e9c12764f052004628a03ba2d upstream.

samsung-laptop is not at all related to ACPI, but since this interface
is not documented at all, and the driver has to use it at load to
understand how it works on the laptop, I think it's a good idea to
disable it if a better solution is available.

Signed-off-by: Corentin Chary <corentincj@iksaif.net>
Signed-off-by: Matthew Garrett <mjg@redhat.com>
[bwh: Backported to 3.2:
 - Adjust context, and change return to goto, since we do not have commit
   5dea7a2 ('samsung-laptop: move code into init/exit functions')
 - Use static variable since we do not have commit a6df489
   ('samsung-laptop: put all local variables in a single structure')]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agonetfilter: ipset: avoid use of kernel-only types
Jan Engelhardt [Sat, 14 Jan 2012 16:38:55 +0000 (17:38 +0100)]
netfilter: ipset: avoid use of kernel-only types

commit 5276e16bb6f35412583518d6f04651dd9dc114be upstream.

When using the xt_set.h header in userspace, one will get these gcc
reports:

ipset/ip_set.h:184:1: error: unknown type name "u16"
In file included from libxt_SET.c:21:0:
netfilter/xt_set.h:61:2: error: unknown type name "u32"
netfilter/xt_set.h:62:2: error: unknown type name "u32"

Signed-off-by: Jan Engelhardt <jengelh@medozas.de>
Signed-off-by: Jozsef Kadlecsik <kadlec@blackhole.kfki.hu>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agoLinux 3.2.32 v3.2.32
Ben Hutchings [Wed, 17 Oct 2012 02:50:15 +0000 (03:50 +0100)]
Linux 3.2.32

11 years agodrm/i915: clear fencing tracking state when retiring requests
Daniel Vetter [Wed, 11 Apr 2012 23:27:57 +0000 (01:27 +0200)]
drm/i915: clear fencing tracking state when retiring requests

commit 15a13bbdffb0d6288a5dd04aee9736267da1335f upstream.

This fixes a resume regression introduced in

commit 7dd4906586274f3945f2aeaaa5a33b451c3b4bba
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date:   Wed Mar 21 10:48:18 2012 +0000

    drm/i915: Mark untiled BLT commands as fenced on gen2/3

which fixed fencing tracking for untiled blt commands.

A side effect of that patch was that now also untiled objects have a
non-zero obj->last_fenced_seqno to track when a fence can be set up
after a pipelined tiling change. Unfortunately this was only cleared
by the fence setup and teardown code, resulting in tons of untiled but
inactive objects with non-zero last_fenced_seqno.

Now after resume we completely reset the seqno tracking, both on the
driver side (by setting dev_priv->next_seqno = 1) and on the hw side
(by allocating a new hws page, which contains the seqnos). Hilarity
and indefinite waits ensued from the stale seqnos in
obj->last_fenced_seqno from before the suspend.

The fix is to properly clear the fencing tracking state like we
already do for the normal gpu rendering while moving objects off the
active list.

Reported-and-tested-by: "Rafael J. Wysocki" <rjw@sisk.pl>
Cc: Jiri Slaby <jslaby@suse.cz>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agodrm/i915: Mark untiled BLT commands as fenced on gen2/3
Chris Wilson [Wed, 21 Mar 2012 10:48:18 +0000 (10:48 +0000)]
drm/i915: Mark untiled BLT commands as fenced on gen2/3

commit 7dd4906586274f3945f2aeaaa5a33b451c3b4bba upstream.

The BLT commands on gen2/3 utilize the fence registers and so we cannot
modify any fences for the object whilst those commands are in flight.
Currently we marked tiled commands as occupying a fence, but forgot to
restrict the untiled commands from preventing a fence being assigned
before they were completed.

One side-effect is that we ten have to double check that a fence was
allocated for a fenced buffer during move-to-active.

Reported-by: Jiri Slaby <jirislaby@gmail.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=43427
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=47990
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Testcase: i-g-t/tests/gem_tiled_after_untiled_blt
Tested-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
[bwh: Backported to 3.2: The nesting of if-statements in the old
 i915_gem_execbuffer_reserve() differs from pin_and_fence_object(),
 so don't move the assignment of obj->pending_fenced_gpu_access but
 adjust the boolean expression as recommended by Daniel Vetter.]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agodrm/i915: fix swizzle detection for gen3
Daniel Vetter [Wed, 14 Dec 2011 12:57:15 +0000 (13:57 +0100)]
drm/i915: fix swizzle detection for gen3

commit c9c4b6f6c28354f1df9bd288dc33ba7ae0e66aaa upstream.

It looks like the desktop variants of i915 and i945 also have the DCC
register to control dram channel interleave and cpu side bit6
swizzling.

Unfortunately internal Cspec/ConfigDB documentation for these ancient chips
have already been dropped and there seem to be no archives. Also
somebody thought the swizzling behaviour is surely a worthy secret to
keep and redacted any mention of these fields from the published Intel
datasheets.

I suspect the hw engineers were really proud of the page coloring
they've achieved in their first dual channel dram controller with
bit17 - after all Bspec explains in great length the optimal layout of
page frame numbers modulo 4 for the color and depth buffers, too.
Later on when they've started to work on VT-d they shamefully
discoverd their stupidity and tried to cover the tracks ...

Tested-by: Daniel Vetter <daniel.vetter@ffwll.ch> (i915g)
Tested-by: Pavel Ondračka <pavel.ondracka@email.cz> (i945g)
Tested-by: Chris Wilson <chris@chris-wilson.co.uk>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=42625
Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agoxHCI: handle command after aborting the command ring
Elric Fu [Wed, 27 Jun 2012 08:55:43 +0000 (16:55 +0800)]
xHCI: handle command after aborting the command ring

commit b63f4053cc8aa22a98e3f9a97845afe6c15d0a0d upstream.

According to xHCI spec section 4.6.1.1 and section 4.6.1.2,
after aborting a command on the command ring, xHC will
generate a command completion event with its completion
code set to Command Ring Stopped at least. If a command is
currently executing at the time of aborting a command, xHC
also generate a command completion event with its completion
code set to Command Abort. When the command ring is stopped,
software may remove, add, or rearrage Command Descriptors.

To cancel a command, software will initialize a command
descriptor for the cancel command, and add it into a
cancel_cmd_list of xhci. When the command ring is stopped,
software will find the command trbs described by command
descriptors in cancel_cmd_list and modify it to No Op
command. If software can't find the matched trbs, we can
think it had been finished.

This patch should be backported to kernels as old as 3.0, that contain
the commit 7ed603ecf8b68ab81f4c83097d3063d43ec73bb8 "xhci: Add an
assertion to check for virt_dev=0 bug." That commit papers over a NULL
pointer dereference, and this patch fixes the underlying issue that
caused the NULL pointer dereference.

Signed-off-by: Elric Fu <elricfu1@gmail.com>
Signed-off-by: Sarah Sharp <sarah.a.sharp@linux.intel.com>
Tested-by: Miroslav Sabljic <miroslav.sabljic@avl.com>
[bwh: Backported to 3.2: inc_deq() needs an additional 'consumer' argument;
 Jonathan Nieder worked out that this should be false]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agoe1000: fix lockdep splat in shutdown handler
Jesse Brandeburg [Wed, 4 Jan 2012 20:23:33 +0000 (20:23 +0000)]
e1000: fix lockdep splat in shutdown handler

commit 3a3847e007aae732d64d8fd1374126393e9879a3 upstream.

As reported by Steven Rostedt, e1000 has a lockdep splat added
during the recent merge window.  The issue is that
cancel_delayed_work is called while holding our private mutex.

There is no reason that I can see to hold the mutex during pci
shutdown, it was more just paranoia that I put the mutex_lock
around the call to e1000_down.

In a quick survey lots of drivers handle locking differently when
being called by the pci layer.  The assumption here is that we
don't need the mutexes' protection in this function because
the driver could not be unloaded while in the shutdown handler
which is only called at reboot or poweroff.

Reported-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Steven Rostedt <rostedt@goodmis.org>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agonetfilter: xt_limit: have r->cost != 0 case work
Jan Engelhardt [Fri, 21 Sep 2012 22:26:52 +0000 (22:26 +0000)]
netfilter: xt_limit: have r->cost != 0 case work

commit 82e6bfe2fbc4d48852114c4f979137cd5bf1d1a8 upstream.

Commit v2.6.19-rc1~1272^2~41 tells us that r->cost != 0 can happen when
a running state is saved to userspace and then reinstated from there.

Make sure that private xt_limit area is initialized with correct values.
Otherwise, random matchings due to use of uninitialized memory.

Signed-off-by: Jan Engelhardt <jengelh@inai.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agonetfilter: limit, hashlimit: avoid duplicated inline
Florian Westphal [Mon, 7 May 2012 10:51:43 +0000 (10:51 +0000)]
netfilter: limit, hashlimit: avoid duplicated inline

commit 7a909ac70f6b0823d9f23a43f19598d4b57ac901 upstream.

credit_cap can be set to credit, which avoids inlining user2credits
twice. Also, remove inline keyword and let compiler decide.

old:
    684     192       0     876     36c net/netfilter/xt_limit.o
   4927     344      32    5303    14b7 net/netfilter/xt_hashlimit.o
now:
    668     192       0     860     35c net/netfilter/xt_limit.o
   4793     344      32    5169    1431 net/netfilter/xt_hashlimit.o

Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agoipvs: fix oops on NAT reply in br_nf context
Lin Ming [Sat, 7 Jul 2012 10:26:10 +0000 (18:26 +0800)]
ipvs: fix oops on NAT reply in br_nf context

commit 9e33ce453f8ac8452649802bee1f410319408f4b upstream.

IPVS should not reset skb->nf_bridge in FORWARD hook
by calling nf_reset for NAT replies. It triggers oops in
br_nf_forward_finish.

[  579.781508] BUG: unable to handle kernel NULL pointer dereference at 0000000000000004
[  579.781669] IP: [<ffffffff817b1ca5>] br_nf_forward_finish+0x58/0x112
[  579.781792] PGD 218f9067 PUD 0
[  579.781865] Oops: 0000 [#1] SMP
[  579.781945] CPU 0
[  579.781983] Modules linked in:
[  579.782047]
[  579.782080]
[  579.782114] Pid: 4644, comm: qemu Tainted: G        W    3.5.0-rc5-00006-g95e69f9 #282 Hewlett-Packard  /30E8
[  579.782300] RIP: 0010:[<ffffffff817b1ca5>]  [<ffffffff817b1ca5>] br_nf_forward_finish+0x58/0x112
[  579.782455] RSP: 0018:ffff88007b003a98  EFLAGS: 00010287
[  579.782541] RAX: 0000000000000008 RBX: ffff8800762ead00 RCX: 000000000001670a
[  579.782653] RDX: 0000000000000000 RSI: 000000000000000a RDI: ffff8800762ead00
[  579.782845] RBP: ffff88007b003ac8 R08: 0000000000016630 R09: ffff88007b003a90
[  579.782957] R10: ffff88007b0038e8 R11: ffff88002da37540 R12: ffff88002da01a02
[  579.783066] R13: ffff88002da01a80 R14: ffff88002d83c000 R15: ffff88002d82a000
[  579.783177] FS:  0000000000000000(0000) GS:ffff88007b000000(0063) knlGS:00000000f62d1b70
[  579.783306] CS:  0010 DS: 002b ES: 002b CR0: 000000008005003b
[  579.783395] CR2: 0000000000000004 CR3: 00000000218fe000 CR4: 00000000000027f0
[  579.783505] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[  579.783684] DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400
[  579.783795] Process qemu (pid: 4644, threadinfo ffff880021b20000, task ffff880021aba760)
[  579.783919] Stack:
[  579.783959]  ffff88007693cedc ffff8800762ead00 ffff88002da01a02 ffff8800762ead00
[  579.784110]  ffff88002da01a02 ffff88002da01a80 ffff88007b003b18 ffffffff817b26c7
[  579.784260]  ffff880080000000 ffffffff81ef59f0 ffff8800762ead00 ffffffff81ef58b0
[  579.784477] Call Trace:
[  579.784523]  <IRQ>
[  579.784562]
[  579.784603]  [<ffffffff817b26c7>] br_nf_forward_ip+0x275/0x2c8
[  579.784707]  [<ffffffff81704b58>] nf_iterate+0x47/0x7d
[  579.784797]  [<ffffffff817ac32e>] ? br_dev_queue_push_xmit+0xae/0xae
[  579.784906]  [<ffffffff81704bfb>] nf_hook_slow+0x6d/0x102
[  579.784995]  [<ffffffff817ac32e>] ? br_dev_queue_push_xmit+0xae/0xae
[  579.785175]  [<ffffffff8187fa95>] ? _raw_write_unlock_bh+0x19/0x1b
[  579.785179]  [<ffffffff817ac417>] __br_forward+0x97/0xa2
[  579.785179]  [<ffffffff817ad366>] br_handle_frame_finish+0x1a6/0x257
[  579.785179]  [<ffffffff817b2386>] br_nf_pre_routing_finish+0x26d/0x2cb
[  579.785179]  [<ffffffff817b2cf0>] br_nf_pre_routing+0x55d/0x5c1
[  579.785179]  [<ffffffff81704b58>] nf_iterate+0x47/0x7d
[  579.785179]  [<ffffffff817ad1c0>] ? br_handle_local_finish+0x44/0x44
[  579.785179]  [<ffffffff81704bfb>] nf_hook_slow+0x6d/0x102
[  579.785179]  [<ffffffff817ad1c0>] ? br_handle_local_finish+0x44/0x44
[  579.785179]  [<ffffffff81551525>] ? sky2_poll+0xb35/0xb54
[  579.785179]  [<ffffffff817ad62a>] br_handle_frame+0x213/0x229
[  579.785179]  [<ffffffff817ad417>] ? br_handle_frame_finish+0x257/0x257
[  579.785179]  [<ffffffff816e3b47>] __netif_receive_skb+0x2b4/0x3f1
[  579.785179]  [<ffffffff816e69fc>] process_backlog+0x99/0x1e2
[  579.785179]  [<ffffffff816e6800>] net_rx_action+0xdf/0x242
[  579.785179]  [<ffffffff8107e8a8>] __do_softirq+0xc1/0x1e0
[  579.785179]  [<ffffffff8135a5ba>] ? trace_hardirqs_off_thunk+0x3a/0x6c
[  579.785179]  [<ffffffff8188812c>] call_softirq+0x1c/0x30

The steps to reproduce as follow,

1. On Host1, setup brige br0(192.168.1.106)
2. Boot a kvm guest(192.168.1.105) on Host1 and start httpd
3. Start IPVS service on Host1
   ipvsadm -A -t 192.168.1.106:80 -s rr
   ipvsadm -a -t 192.168.1.106:80 -r 192.168.1.105:80 -m
4. Run apache benchmark on Host2(192.168.1.101)
   ab -n 1000 http://192.168.1.106/

ip_vs_reply4
  ip_vs_out
    handle_response
      ip_vs_notrack
        nf_reset()
        {
          skb->nf_bridge = NULL;
        }

Actually, IPVS wants in this case just to replace nfct
with untracked version. So replace the nf_reset(skb) call
in ip_vs_notrack() with a nf_conntrack_put(skb->nfct) call.

Signed-off-by: Lin Ming <mlin@ss.pku.edu.cn>
Signed-off-by: Julian Anastasov <ja@ssi.bg>
Signed-off-by: Simon Horman <horms@verge.net.au>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agonetfilter: nf_ct_expect: fix possible access to uninitialized timer
Pablo Neira Ayuso [Thu, 16 Aug 2012 00:25:24 +0000 (02:25 +0200)]
netfilter: nf_ct_expect: fix possible access to uninitialized timer

commit 2614f86490122bf51eb7c12ec73927f1900f4e7d upstream.

In __nf_ct_expect_check, the function refresh_timer returns 1
if a matching expectation is found and its timer is successfully
refreshed. This results in nf_ct_expect_related returning 0.
Note that at this point:

- the passed expectation is not inserted in the expectation table
  and its timer was not initialized, since we have refreshed one
  matching/existing expectation.

- nf_ct_expect_alloc uses kmem_cache_alloc, so the expectation
  timer is in some undefined state just after the allocation,
  until it is appropriately initialized.

This can be a problem for the SIP helper during the expectation
addition:

 ...
 if (nf_ct_expect_related(rtp_exp) == 0) {
         if (nf_ct_expect_related(rtcp_exp) != 0)
                 nf_ct_unexpect_related(rtp_exp);
 ...

Note that nf_ct_expect_related(rtp_exp) may return 0 for the timer refresh
case that is detailed above. Then, if nf_ct_unexpect_related(rtcp_exp)
returns != 0, nf_ct_unexpect_related(rtp_exp) is called, which does:

 spin_lock_bh(&nf_conntrack_lock);
 if (del_timer(&exp->timeout)) {
         nf_ct_unlink_expect(exp);
         nf_ct_expect_put(exp);
 }
 spin_unlock_bh(&nf_conntrack_lock);

Note that del_timer always returns false if the timer has been
initialized.  However, the timer was not initialized since setup_timer
was not called, therefore, the expectation timer remains in some
undefined state. If I'm not missing anything, this may lead to the
removal an unexistent expectation.

To fix this, the optimization that allows refreshing an expectation
is removed. Now nf_conntrack_expect_related looks more consistent
to me since it always add the expectation in case that it returns
success.

Thanks to Patrick McHardy for participating in the discussion of
this patch.

I think this may be the source of the problem described by:
http://marc.info/?l=netfilter-devel&m=134073514719421&w=2

Reported-by: Rafal Fitt <rafalf@aplusc.com.pl>
Acked-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agonetfilter: nf_nat_sip: fix via header translation with multiple parameters
Patrick McHardy [Thu, 9 Aug 2012 10:08:47 +0000 (10:08 +0000)]
netfilter: nf_nat_sip: fix via header translation with multiple parameters

commit f22eb25cf5b1157b29ef88c793b71972efc47143 upstream.

Via-headers are parsed beginning at the first character after the Via-address.
When the address is translated first and its length decreases, the offset to
start parsing at is incorrect and header parameters might be missed.

Update the offset after translating the Via-address to fix this.

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agonetfilter: nf_nat_sip: fix incorrect handling of EBUSY for RTCP expectation
Pablo Neira Ayuso [Wed, 29 Aug 2012 15:24:09 +0000 (15:24 +0000)]
netfilter: nf_nat_sip: fix incorrect handling of EBUSY for RTCP expectation

commit 3f509c689a07a4aa989b426893d8491a7ffcc410 upstream.

We're hitting bug while trying to reinsert an already existing
expectation:

kernel BUG at kernel/timer.c:895!
invalid opcode: 0000 [#1] SMP
[...]
Call Trace:
 <IRQ>
 [<ffffffffa0069563>] nf_ct_expect_related_report+0x4a0/0x57a [nf_conntrack]
 [<ffffffff812d423a>] ? in4_pton+0x72/0x131
 [<ffffffffa00ca69e>] ip_nat_sdp_media+0xeb/0x185 [nf_nat_sip]
 [<ffffffffa00b5b9b>] set_expected_rtp_rtcp+0x32d/0x39b [nf_conntrack_sip]
 [<ffffffffa00b5f15>] process_sdp+0x30c/0x3ec [nf_conntrack_sip]
 [<ffffffff8103f1eb>] ? irq_exit+0x9a/0x9c
 [<ffffffffa00ca738>] ? ip_nat_sdp_media+0x185/0x185 [nf_nat_sip]

We have to remove the RTP expectation if the RTCP expectation hits EBUSY
since we keep trying with other ports until we succeed.

Reported-by: Rafal Fitt <rafalf@aplusc.com.pl>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agonetfilter: nf_ct_ipv4: packets with wrong ihl are invalid
Jozsef Kadlecsik [Tue, 3 Apr 2012 20:02:01 +0000 (22:02 +0200)]
netfilter: nf_ct_ipv4: packets with wrong ihl are invalid

commit 07153c6ec074257ade76a461429b567cff2b3a1e upstream.

It was reported that the Linux kernel sometimes logs:

klogd: [2629147.402413] kernel BUG at net / netfilter /
nf_conntrack_proto_tcp.c: 447!
klogd: [1072212.887368] kernel BUG at net / netfilter /
nf_conntrack_proto_tcp.c: 392

ipv4_get_l4proto() in nf_conntrack_l3proto_ipv4.c and tcp_error() in
nf_conntrack_proto_tcp.c should catch malformed packets, so the errors
at the indicated lines - TCP options parsing - should not happen.
However, tcp_error() relies on the "dataoff" offset to the TCP header,
calculated by ipv4_get_l4proto().  But ipv4_get_l4proto() does not check
bogus ihl values in IPv4 packets, which then can slip through tcp_error()
and get caught at the TCP options parsing routines.

The patch fixes ipv4_get_l4proto() by invalidating packets with bogus
ihl value.

The patch closes netfilter bugzilla id 771.

Signed-off-by: Jozsef Kadlecsik <kadlec@blackhole.kfki.hu>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agosched: Fix migration thread runtime bogosity
Mike Galbraith [Sat, 4 Aug 2012 03:44:14 +0000 (05:44 +0200)]
sched: Fix migration thread runtime bogosity

commit 8f6189684eb4e85e6c593cd710693f09c944450a upstream.

Make stop scheduler class do the same accounting as other classes,

Migration threads can be caught in the act while doing exec balancing,
leading to the below due to use of unmaintained ->se.exec_start.  The
load that triggered this particular instance was an apparently out of
control heavily threaded application that does system monitoring in
what equated to an exec bomb, with one of the VERY frequently migrated
tasks being ps.

%CPU   PID USER     CMD
99.3    45 root     [migration/10]
97.7    53 root     [migration/12]
97.0    57 root     [migration/13]
90.1    49 root     [migration/11]
89.6    65 root     [migration/15]
88.7    17 root     [migration/3]
80.4    37 root     [migration/8]
78.1    41 root     [migration/9]
44.2    13 root     [migration/2]

Signed-off-by: Mike Galbraith <mgalbraith@suse.de>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1344051854.6739.19.camel@marge.simpson.net
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
[Steven Rostedt: backport for 3.2.]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agohpsa: dial down lockup detection during firmware flash
Stephen M. Cameron [Tue, 1 May 2012 16:43:42 +0000 (11:43 -0500)]
hpsa: dial down lockup detection during firmware flash

commit e85c59746957fd6e3595d02cf614370056b5816e upstream.

Dial back the aggressiveness of the controller lockup detection thread.
Currently it will declare the controller to be locked up if it goes
for 10 seconds with no interrupts and no change in the heartbeat
register.  Dial back this to 30 seconds with no heartbeat change, and
also snoop the ioctl path and if a firmware flash command is detected,
dial it back further to 4 minutes until the firmware flash command
completes.  The reason for this is that during the firmware flash
operation, the controller apparently doesn't update the heartbeat
register as frequently as it is supposed to, and we can get a false
positive.

Signed-off-by: Stephen M. Cameron <scameron@beardog.cce.hp.com>
Signed-off-by: James Bottomley <JBottomley@Parallels.com>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agor8169: 8168c and later require bit 0x20 to be set in Config2 for PME signaling.
Francois Romieu [Tue, 17 Apr 2012 09:12:01 +0000 (11:12 +0200)]
r8169: 8168c and later require bit 0x20 to be set in Config2 for PME signaling.

commit d387b427c973974dd619a33549c070ac5d0e089f upstream.

The new 84xx stopped flying below the radars.

Signed-off-by: Francois Romieu <romieu@fr.zoreil.com>
Cc: Hayes Wang <hayeswang@realtek.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agor8169: Config1 is read-only on 8168c and later.
Francois Romieu [Tue, 17 Apr 2012 09:10:11 +0000 (11:10 +0200)]
r8169: Config1 is read-only on 8168c and later.

commit 851e60221926a53344b4227879858bef841b0477 upstream.

Suggested by Hayes.

Signed-off-by: Francois Romieu <romieu@fr.zoreil.com>
Cc: Hayes Wang <hayeswang@realtek.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agomempolicy: fix a memory corruption by refcount imbalance in alloc_pages_vma()
Mel Gorman [Tue, 9 Oct 2012 16:58:41 +0000 (17:58 +0100)]
mempolicy: fix a memory corruption by refcount imbalance in alloc_pages_vma()

commit 00442ad04a5eac08a98255697c510e708f6082e2 upstream.

Commit cc9a6c877661 ("cpuset: mm: reduce large amounts of memory barrier
related damage v3") introduced a potential memory corruption.
shmem_alloc_page() uses a pseudo vma and it has one significant unique
combination, vma->vm_ops=NULL and vma->policy->flags & MPOL_F_SHARED.

get_vma_policy() does NOT increase a policy ref when vma->vm_ops=NULL
and mpol_cond_put() DOES decrease a policy ref when a policy has
MPOL_F_SHARED.  Therefore, when a cpuset update race occurs,
alloc_pages_vma() falls in 'goto retry_cpuset' path, decrements the
reference count and frees the policy prematurely.

Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Mel Gorman <mgorman@suse.de>
Reviewed-by: Christoph Lameter <cl@linux.com>
Cc: Josh Boyer <jwboyer@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Mel Gorman <mgorman@suse.de>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agomempolicy: fix refcount leak in mpol_set_shared_policy()
KOSAKI Motohiro [Mon, 8 Oct 2012 23:29:19 +0000 (16:29 -0700)]
mempolicy: fix refcount leak in mpol_set_shared_policy()

commit 63f74ca21f1fad36d075e063f06dcc6d39fe86b2 upstream.

When shared_policy_replace() fails to allocate new->policy is not freed
correctly by mpol_set_shared_policy().  The problem is that shared
mempolicy code directly call kmem_cache_free() in multiple places where
it is easy to make a mistake.

This patch creates an sp_free wrapper function and uses it. The bug was
introduced pre-git age (IOW, before 2.6.12-rc2).

[mgorman@suse.de: Editted changelog]
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Mel Gorman <mgorman@suse.de>
Reviewed-by: Christoph Lameter <cl@linux.com>
Cc: Josh Boyer <jwboyer@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Mel Gorman <mgorman@suse.de>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agomempolicy: fix a race in shared_policy_replace()
Mel Gorman [Tue, 9 Oct 2012 16:58:39 +0000 (17:58 +0100)]
mempolicy: fix a race in shared_policy_replace()

commit b22d127a39ddd10d93deee3d96e643657ad53a49 upstream.

shared_policy_replace() use of sp_alloc() is unsafe.  1) sp_node cannot
be dereferenced if sp->lock is not held and 2) another thread can modify
sp_node between spin_unlock for allocating a new sp node and next
spin_lock.  The bug was introduced before 2.6.12-rc2.

Kosaki's original patch for this problem was to allocate an sp node and
policy within shared_policy_replace and initialise it when the lock is
reacquired.  I was not keen on this approach because it partially
duplicates sp_alloc().  As the paths were sp->lock is taken are not that
performance critical this patch converts sp->lock to sp->mutex so it can
sleep when calling sp_alloc().

[kosaki.motohiro@jp.fujitsu.com: Original patch]
Signed-off-by: Mel Gorman <mgorman@suse.de>
Acked-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Reviewed-by: Christoph Lameter <cl@linux.com>
Cc: Josh Boyer <jwboyer@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Mel Gorman <mgorman@suse.de>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agomempolicy: remove mempolicy sharing
KOSAKI Motohiro [Mon, 8 Oct 2012 23:29:16 +0000 (16:29 -0700)]
mempolicy: remove mempolicy sharing

commit 869833f2c5c6e4dd09a5378cfc665ffb4615e5d2 upstream.

Dave Jones' system call fuzz testing tool "trinity" triggered the
following bug error with slab debugging enabled

    =============================================================================
    BUG numa_policy (Not tainted): Poison overwritten
    -----------------------------------------------------------------------------

    INFO: 0xffff880146498250-0xffff880146498250. First byte 0x6a instead of 0x6b
    INFO: Allocated in mpol_new+0xa3/0x140 age=46310 cpu=6 pid=32154
     __slab_alloc+0x3d3/0x445
     kmem_cache_alloc+0x29d/0x2b0
     mpol_new+0xa3/0x140
     sys_mbind+0x142/0x620
     system_call_fastpath+0x16/0x1b

    INFO: Freed in __mpol_put+0x27/0x30 age=46268 cpu=6 pid=32154
     __slab_free+0x2e/0x1de
     kmem_cache_free+0x25a/0x260
     __mpol_put+0x27/0x30
     remove_vma+0x68/0x90
     exit_mmap+0x118/0x140
     mmput+0x73/0x110
     exit_mm+0x108/0x130
     do_exit+0x162/0xb90
     do_group_exit+0x4f/0xc0
     sys_exit_group+0x17/0x20
     system_call_fastpath+0x16/0x1b

    INFO: Slab 0xffffea0005192600 objects=27 used=27 fp=0x          (null) flags=0x20000000004080
    INFO: Object 0xffff880146498250 @offset=592 fp=0xffff88014649b9d0

The problem is that the structure is being prematurely freed due to a
reference count imbalance. In the following case mbind(addr, len) should
replace the memory policies of both vma1 and vma2 and thus they will
become to share the same mempolicy and the new mempolicy will have the
MPOL_F_SHARED flag.

  +-------------------+-------------------+
  |     vma1          |     vma2(shmem)   |
  +-------------------+-------------------+
  |                                       |
 addr                                 addr+len

alloc_pages_vma() uses get_vma_policy() and mpol_cond_put() pair for
maintaining the mempolicy reference count.  The current rule is that
get_vma_policy() only increments refcount for shmem VMA and
mpol_conf_put() only decrements refcount if the policy has
MPOL_F_SHARED.

In above case, vma1 is not shmem vma and vma->policy has MPOL_F_SHARED!
The reference count will be decreased even though was not increased
whenever alloc_page_vma() is called.  This has been broken since commit
[52cd3b07: mempolicy: rework mempolicy Reference Counting] in 2008.

There is another serious bug with the sharing of memory policies.
Currently, mempolicy rebind logic (it is called from cpuset rebinding)
ignores a refcount of mempolicy and override it forcibly.  Thus, any
mempolicy sharing may cause mempolicy corruption.  The bug was
introduced by commit [68860ec1: cpusets: automatic numa mempolicy
rebinding].

Ideally, the shared policy handling would be rewritten to either
properly handle COW of the policy structures or at least reference count
MPOL_F_SHARED based exclusively on information within the policy.
However, this patch takes the easier approach of disabling any policy
sharing between VMAs.  Each new range allocated with sp_alloc will
allocate a new policy, set the reference count to 1 and drop the
reference count of the old policy.  This increases the memory footprint
but is not expected to be a major problem as mbind() is unlikely to be
used for fine-grained ranges.  It is also inefficient because it means
we allocate a new policy even in cases where mbind_range() could use the
new_policy passed to it.  However, it is more straight-forward and the
change should be invisible to the user.

[mgorman@suse.de: Edited changelog]
Reported-by: Dave Jones <davej@redhat.com>,
Cc: Christoph Lameter <cl@linux.com>,
Reviewed-by: Christoph Lameter <cl@linux.com>
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Mel Gorman <mgorman@suse.de>
Cc: Josh Boyer <jwboyer@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Mel Gorman <mgorman@suse.de>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agoefi: initialize efi.runtime_version to make query_variable_info/update_capsule workable
Seiji Aguchi [Tue, 24 Jul 2012 13:27:23 +0000 (13:27 +0000)]
efi: initialize efi.runtime_version to make query_variable_info/update_capsule workable

commit d6cf86d8f23253225fe2a763d627ecf7dfee9dae upstream.

A value of efi.runtime_version is checked before calling
update_capsule()/query_variable_info() as follows.
But it isn't initialized anywhere.

<snip>
static efi_status_t virt_efi_query_variable_info(u32 attr,
                                                 u64 *storage_space,
                                                 u64 *remaining_space,
                                                 u64 *max_variable_size)
{
        if (efi.runtime_version < EFI_2_00_SYSTEM_TABLE_REVISION)
                return EFI_UNSUPPORTED;
<snip>

This patch initializes a value of efi.runtime_version at boot time.

Signed-off-by: Seiji Aguchi <seiji.aguchi@hds.com>
Acked-by: Matthew Garrett <mjg@redhat.com>
Signed-off-by: Matt Fleming <matt.fleming@intel.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agodrm/radeon: properly handle mc_stop/mc_resume on evergreen+ (v2)
Alex Deucher [Wed, 15 Aug 2012 21:18:42 +0000 (17:18 -0400)]
drm/radeon: properly handle mc_stop/mc_resume on evergreen+ (v2)

commit 62444b7462a2b98bc78d68736c03a7c4e66ba7e2 upstream.

- Stop the displays from accessing the FB
- Block CPU access
- Turn off MC client access

This should fix issues some users have seen, especially
with UEFI, when changing the MC FB location that result
in hangs or display corruption.

v2: fix crtc enabled check noticed by Luca Tettamanti

Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[bwh: Backported to 3.2:
 - Drop DCE6 cases
 - Call evergreen_mc_wait_for_idle() directly
 - Add dce4_wait_for_vblank() (commits 3ae19b750bdc09ce233e1504348320141593ffda
   and 4a15903db02026728d0cf2755c6fabae16b8db6a) and call it directly
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agoeCryptfs: Call lower ->flush() from ecryptfs_flush()
Tyler Hicks [Thu, 13 Sep 2012 01:38:00 +0000 (18:38 -0700)]
eCryptfs: Call lower ->flush() from ecryptfs_flush()

commit 64e6651dcc10e9d2cc6230208a8e6c2cfd19ae18 upstream.

Since eCryptfs only calls fput() on the lower file in
ecryptfs_release(), eCryptfs should call the lower filesystem's
->flush() from ecryptfs_flush().

If the lower filesystem implements ->flush(), then eCryptfs should try
to flush out any dirty pages prior to calling the lower ->flush(). If
the lower filesystem does not implement ->flush(), then eCryptfs has no
need to do anything in ecryptfs_flush() since dirty pages are now
written out to the lower filesystem in ecryptfs_release().

Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agoeCryptfs: Write out all dirty pages just before releasing the lower file
Tyler Hicks [Thu, 13 Sep 2012 01:02:46 +0000 (18:02 -0700)]
eCryptfs: Write out all dirty pages just before releasing the lower file

commit 7149f2558d5b5b988726662fe58b1c388337805b upstream.

Fixes a regression caused by:

821f749 eCryptfs: Revert to a writethrough cache model

That patch reverted some code (specifically, 32001d6f) that was
necessary to properly handle open() -> mmap() -> close() -> dirty pages
-> munmap(), because the lower file could be closed before the dirty
pages are written out.

Rather than reapplying 32001d6f, this approach is a better way of
ensuring that the lower file is still open in order to handle writing
out the dirty pages. It is called from ecryptfs_release(), while we have
a lock on the lower file pointer, just before the lower file gets the
final fput() and we overwrite the pointer.

https://launchpad.net/bugs/1047261

Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Reported-by: Artemy Tregubenko <me@arty.name>
Tested-by: Artemy Tregubenko <me@arty.name>
Tested-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agoeCryptfs: Revert to a writethrough cache model
Tyler Hicks [Tue, 3 Jul 2012 23:50:57 +0000 (16:50 -0700)]
eCryptfs: Revert to a writethrough cache model

commit 821f7494a77627fb1ab539591c57b22cdca702d6 upstream.

A change was made about a year ago to get eCryptfs to better utilize its
page cache during writes. The idea was to do the page encryption
operations during page writeback, rather than doing them when initially
writing into the page cache, to reduce the number of page encryption
operations during sequential writes. This meant that the encrypted page
would only be written to the lower filesystem during page writeback,
which was a change from how eCryptfs had previously wrote to the lower
filesystem in ecryptfs_write_end().

The change caused a few eCryptfs-internal bugs that were shook out.
Unfortunately, more grave side effects have been identified that will
force changes outside of eCryptfs. Because the lower filesystem isn't
consulted until page writeback, eCryptfs has no way to pass lower write
errors (ENOSPC, mainly) back to userspace. Additionaly, it was reported
that quotas could be bypassed because of the way eCryptfs may sometimes
open the lower filesystem using a privileged kthread.

It would be nice to resolve the latest issues, but it is best if the
eCryptfs commits be reverted to the old behavior in the meantime.

This reverts:
32001d6f "eCryptfs: Flush file in vma close"
5be79de2 "eCryptfs: Flush dirty pages in setattr"
57db4e8d "ecryptfs: modify write path to encrypt page in writepage"

Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Tested-by: Colin King <colin.king@canonical.com>
Cc: Colin King <colin.king@canonical.com>
Cc: Thieu Le <thieule@google.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agoeCryptfs: Initialize empty lower files when opening them
Tyler Hicks [Thu, 21 Jun 2012 06:50:59 +0000 (23:50 -0700)]
eCryptfs: Initialize empty lower files when opening them

commit e3ccaa9761200952cc269b1f4b7d7bb77a5e071b upstream.

Historically, eCryptfs has only initialized lower files in the
ecryptfs_create() path. Lower file initialization is the act of writing
the cryptographic metadata from the inode's crypt_stat to the header of
the file. The ecryptfs_open() path already expects that metadata to be
in the header of the file.

A number of users have reported empty lower files in beneath their
eCryptfs mounts. Most of the causes for those empty files being left
around have been addressed, but the presence of empty files causes
problems due to the lack of proper cryptographic metadata.

To transparently solve this problem, this patch initializes empty lower
files in the ecryptfs_open() error path. If the metadata is unreadable
due to the lower inode size being 0, plaintext passthrough support is
not in use, and the metadata is stored in the header of the file (as
opposed to the user.ecryptfs extended attribute), the lower file will be
initialized.

The number of nested conditionals in ecryptfs_open() was getting out of
hand, so a helper function was created. To avoid the same nested
conditional problem, the conditional logic was reversed inside of the
helper function.

https://launchpad.net/bugs/911507

Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Cc: John Johansen <john.johansen@canonical.com>
Cc: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agoeCryptfs: Unlink lower inode when ecryptfs_create() fails
Tyler Hicks [Tue, 22 May 2012 20:09:50 +0000 (15:09 -0500)]
eCryptfs: Unlink lower inode when ecryptfs_create() fails

commit 8bc2d3cf612994a960c2e8eaea37f6676f67082a upstream.

ecryptfs_create() creates a lower inode, allocates an eCryptfs inode,
initializes the eCryptfs inode and cryptographic metadata attached to
the inode, and then writes the metadata to the header of the file.

If an error was to occur after the lower inode was created, an empty
lower file would be left in the lower filesystem. This is a problem
because ecryptfs_open() refuses to open any lower files which do not
have the appropriate metadata in the file header.

This patch properly unlinks the lower inode when an error occurs in the
later stages of ecryptfs_create(), reducing the chance that an empty
lower file will be left in the lower filesystem.

https://launchpad.net/bugs/872905

Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Cc: John Johansen <john.johansen@canonical.com>
Cc: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agoudf: fix retun value on error path in udf_load_logicalvol
Nikola Pajkovsky [Tue, 14 Aug 2012 22:38:08 +0000 (00:38 +0200)]
udf: fix retun value on error path in udf_load_logicalvol

commit 68766a2edcd5cd744262a70a2f67a320ac944760 upstream.

In case we detect a problem and bail out, we fail to set "ret" to a
nonzero value, and udf_load_logicalvol will mistakenly report success.

Signed-off-by: Nikola Pajkovsky <npajkovs@redhat.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agoautofs4 - fix reset pending flag on mount fail
Ian Kent [Thu, 11 Oct 2012 00:00:33 +0000 (08:00 +0800)]
autofs4 - fix reset pending flag on mount fail

commit 49999ab27eab6289a8e4f450e148bdab521361b2 upstream.

In autofs4_d_automount(), if a mount fail occurs the AUTOFS_INF_PENDING
mount pending flag is not cleared.

One effect of this is when using the "browse" option, directory entry
attributes show up with all "?"s due to the incorrect callback and
subsequent failure return (when in fact no callback should be made).

Signed-off-by: Ian Kent <ikent@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agofirewire: cdev: fix user memory corruption (i386 userland on amd64 kernel)
Stefan Richter [Sat, 6 Oct 2012 12:12:56 +0000 (14:12 +0200)]
firewire: cdev: fix user memory corruption (i386 userland on amd64 kernel)

commit 790198f74c9d1b46b6a89504361b1a844670d050 upstream.

Fix two bugs of the /dev/fw* character device concerning the
FW_CDEV_IOC_GET_INFO ioctl with nonzero fw_cdev_get_info.bus_reset.
(Practically all /dev/fw* clients issue this ioctl right after opening
the device.)

Both bugs are caused by sizeof(struct fw_cdev_event_bus_reset) being 36
without natural alignment and 40 with natural alignment.

 1) Memory corruption, affecting i386 userland on amd64 kernel:
    Userland reserves a 36 bytes large buffer, kernel writes 40 bytes.
    This has been first found and reported against libraw1394 if
    compiled with gcc 4.7 which happens to order libraw1394's stack such
    that the bug became visible as data corruption.

 2) Information leak, affecting all kernel architectures except i386:
    4 bytes of random kernel stack data were leaked to userspace.

Hence limit the respective copy_to_user() to the 32-bit aligned size of
struct fw_cdev_event_bus_reset.

Reported-by: Simon Kirby <sim@hostway.ca>
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agohugetlb: do not use vma_hugecache_offset() for vma_prio_tree_foreach
Michal Hocko [Mon, 8 Oct 2012 23:33:31 +0000 (16:33 -0700)]
hugetlb: do not use vma_hugecache_offset() for vma_prio_tree_foreach

commit 36e4f20af833d1ce196e6a4ade05dc26c44652d1 upstream.

Commit 0c176d52b0b2 ("mm: hugetlb: fix pgoff computation when unmapping
page from vma") fixed pgoff calculation but it has replaced it by
vma_hugecache_offset() which is not approapriate for offsets used for
vma_prio_tree_foreach() because that one expects index in page units
rather than in huge_page_shift.

Johannes said:

: The resulting index may not be too big, but it can be too small: assume
: hpage size of 2M and the address to unmap to be 0x200000.  This is regular
: page index 512 and hpage index 1.  If you have a VMA that maps the file
: only starting at the second huge page, that VMAs vm_pgoff will be 512 but
: you ask for offset 1 and miss it even though it does map the page of
: interest.  hugetlb_cow() will try to unmap, miss the vma, and retry the
: cow until the allocation succeeds or the skipped vma(s) go away.

Signed-off-by: Michal Hocko <mhocko@suse.cz>
Acked-by: Hillf Danton <dhillf@gmail.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: David Rientjes <rientjes@google.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agomm: hugetlb: fix pgoff computation when unmapping page from vma
Hillf Danton [Tue, 10 Jan 2012 23:08:19 +0000 (15:08 -0800)]
mm: hugetlb: fix pgoff computation when unmapping page from vma

commit 0c176d52b0b2619f231b2bbf329b90c028134f58 upstream.

The computation for pgoff is incorrect, at least with

(vma->vm_pgoff >> PAGE_SHIFT)

involved.  It is fixed with the available method if HPAGE_SIZE is
concerned in page cache lookup.

[akpm@linux-foundation.org: use vma_hugecache_offset() directly, per Michal]
Signed-off-by: Hillf Danton <dhillf@gmail.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Michal Hocko <mhocko@suse.cz>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: David Rientjes <rientjes@google.com>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agomm: thp: fix pmd_present for split_huge_page and PROT_NONE with THP
Andrea Arcangeli [Mon, 8 Oct 2012 23:33:27 +0000 (16:33 -0700)]
mm: thp: fix pmd_present for split_huge_page and PROT_NONE with THP

commit 027ef6c87853b0a9df53175063028edb4950d476 upstream.

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

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

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

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

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

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

Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Johannes Weiner <jweiner@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Mel Gorman <mgorman@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agomm: fix invalidate_complete_page2() lock ordering
Hugh Dickins [Mon, 8 Oct 2012 23:33:14 +0000 (16:33 -0700)]
mm: fix invalidate_complete_page2() lock ordering

commit ec4d9f626d5908b6052c2973f37992f1db52e967 upstream.

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

invalidate_inode_pages2
  invalidate_complete_page2
    spin_lock_irq(&mapping->tree_lock)
    clear_page_mlock
      isolate_lru_page
        spin_unlock_irq(&zone->lru_lock)

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

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

Reported-by: Sasha Levin <levinsasha928@gmail.com>
Deciphered-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Hugh Dickins <hughd@google.com>
Acked-by: Mel Gorman <mel@csn.ul.ie>
Cc: Rik van Riel <riel@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michel Lespinasse <walken@google.com>
Cc: Ying Han <yinghan@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agoARM: OMAP: counter: add locking to read_persistent_clock
Colin Cross [Mon, 8 Oct 2012 21:01:12 +0000 (14:01 -0700)]
ARM: OMAP: counter: add locking to read_persistent_clock

commit 9d7d6e363b06934221b81a859d509844c97380df upstream.

read_persistent_clock uses a global variable, use a spinlock to
ensure non-atomic updates to the variable don't overlap and cause
time to move backwards.

Signed-off-by: Colin Cross <ccross@android.com>
Signed-off-by: R Sricharan <r.sricharan@ti.com>
Signed-off-by: Tony Lindgren <tony@atomide.com>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agommc: sdhci-s3c: fix the wrong number of max bus clocks
Jaehoon Chung [Wed, 19 Sep 2012 06:43:33 +0000 (14:43 +0800)]
mmc: sdhci-s3c: fix the wrong number of max bus clocks

commit 5feb54a1ab91a237e247c013b8c4fb100ea347b1 upstream.

We can use up to four bus-clocks; but on module remove, we didn't
disable the fourth bus clock.

Signed-off-by: Jaehoon Chung <jh80.chung@samsung.com>
Signed-off-by: Kyungmin Park <kyungmin.park@samsung.com>
Signed-off-by: Chris Ball <cjb@laptop.org>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
11 years agoALSA: aloop - add locking to timer access
Omair Mohammed Abdullah [Sat, 29 Sep 2012 06:54:05 +0000 (12:24 +0530)]
ALSA: aloop - add locking to timer access

commit d4f1e48bd11e3df6a26811f7a1f06c4225d92f7d upstream.

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

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

Signed-off-by: Omair Mohammed Abdullah <omair.m.abdullah@linux.intel.com>
Signed-off-by: Vinod Koul <vinod.koul@linux.intel.com>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>