platform/kernel/linux-exynos.git
9 years ago9p: don't leave a half-initialized inode sitting around
Al Viro [Sun, 12 Jul 2015 14:34:29 +0000 (10:34 -0400)]
9p: don't leave a half-initialized inode sitting around

commit 0a73d0a204a4a04a1e110539c5a524ae51f91d6d upstream.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years ago9p: forgetting to cancel request on interrupted zero-copy RPC
Al Viro [Sat, 4 Jul 2015 20:04:19 +0000 (16:04 -0400)]
9p: forgetting to cancel request on interrupted zero-copy RPC

commit a84b69cb6e0a41e86bc593904faa6def3b957343 upstream.

If we'd already sent a request and decide to abort it, we *must*
issue TFLUSH properly and not just blindly reuse the tag, or
we'll get seriously screwed when response eventually arrives
and we confuse it for response to later request that had reused
the same tag.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agoSUNRPC: Fix a memory leak in the backchannel code
Trond Myklebust [Mon, 1 Jun 2015 19:10:25 +0000 (15:10 -0400)]
SUNRPC: Fix a memory leak in the backchannel code

commit 88de6af24f2b48b06c514d3c3d0a8f22fafe30bd upstream.

req->rq_private_buf isn't initialised when xprt_setup_backchannel calls
xprt_free_allocation.

Fixes: fb7a0b9addbdb ("nfs41: New backchannel helper routines")
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agonfs: always update creds in mirror, even when we have an already connected ds
Jeff Layton [Wed, 24 Jun 2015 16:10:24 +0000 (12:10 -0400)]
nfs: always update creds in mirror, even when we have an already connected ds

commit 0c8315dd56577445dd1afe6b9cfa06b7efdf2f82 upstream.

A ds can be associated with more than one mirror, but we currently skip
setting a mirror's credentials if we find that it's already set up with
a connected client.

The upshot is that we can end up sending DS writes with MDS credentials
instead of properly setting them up. Fix nfs4_ff_layout_prepare_ds to
always verify that the mirror's credentials are set up, even when we
have a DS that's already connected.

Reported-by: Tom Haynes <thomas.haynes@primarydata.com>
Signed-off-by: Jeff Layton <jeff.layton@primarydata.com>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agonfs: fix potential credential leak in ff_layout_update_mirror_cred
Jeff Layton [Wed, 24 Jun 2015 16:10:23 +0000 (12:10 -0400)]
nfs: fix potential credential leak in ff_layout_update_mirror_cred

commit a24221dca1868101c9b4b5adde4a6a5b1a3a64a7 upstream.

If we have two tasks racing to update a mirror's credentials, then they
can end up leaking one (or more) sets of credentials. The first task
will set mirror->cred and then the second task will just overwrite it.

Use a cmpxchg to ensure that the creds are only set once. If we get to
the point where we would set mirror->cred and find that they're already
set, then we just release the creds that were just found.

Signed-off-by: Jeff Layton <jeff.layton@primarydata.com>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agoNFS: Ensure we set NFS_CONTEXT_RESEND_WRITES when requeuing writes
Trond Myklebust [Wed, 17 Jun 2015 23:56:22 +0000 (19:56 -0400)]
NFS: Ensure we set NFS_CONTEXT_RESEND_WRITES when requeuing writes

commit c70701131f7a8edea91fc49d11796d342cff7c62 upstream.

If a write attempt fails, and the write is queued up for resending to
the server, as opposed to being dropped, then we need to set the
appropriate flag so that nfs_file_fsync() does the right thing.

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agonfs: increase size of EXCHANGE_ID name string buffer
Jeff Layton [Tue, 9 Jun 2015 23:43:56 +0000 (19:43 -0400)]
nfs: increase size of EXCHANGE_ID name string buffer

commit 764ad8ba8cd4c6f836fca9378f8c5121aece0842 upstream.

The current buffer is much too small if you have a relatively long
hostname. Bring it up to the size of the one that SETCLIENTID has.

Reported-by: Michael Skralivetsky <michael.skralivetsky@primarydata.com>
Signed-off-by: Jeff Layton <jeff.layton@primarydata.com>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agofixing infinite OPEN loop in 4.0 stateid recovery
Olga Kornievskaia [Fri, 15 May 2015 15:45:31 +0000 (11:45 -0400)]
fixing infinite OPEN loop in 4.0 stateid recovery

commit e8d975e73e5fa05f983fbf2723120edcf68e0b38 upstream.

Problem: When an operation like WRITE receives a BAD_STATEID, even though
recovery code clears the RECLAIM_NOGRACE recovery flag before recovering
the open state, because of clearing delegation state for the associated
inode, nfs_inode_find_state_and_recover() gets called and it makes the
same state with RECLAIM_NOGRACE flag again. As a results, when we restart
looking over the open states, we end up in the infinite loop instead of
breaking out in the next test of state flags.

Solution: unset the RECLAIM_NOGRACE set because of
calling of nfs_inode_find_state_and_recover() after returning from calling
recover_open() function.

Signed-off-by: Olga Kornievskaia <kolga@netapp.com>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agoNFS: Fix size of NFSACL SETACL operations
Chuck Lever [Tue, 26 May 2015 15:53:52 +0000 (11:53 -0400)]
NFS: Fix size of NFSACL SETACL operations

commit d683cc49daf7c5afca8cd9654aaa1bf63cdf2ad9 upstream.

When encoding the NFSACL SETACL operation, reserve just the estimated
size of the ACL rather than a fixed maximum. This eliminates needless
zero padding on the wire that the server ignores.

Fixes: ee5dc7732bd5 ('NFS: Fix "kernel BUG at fs/nfs/nfs3xdr.c:1338!"')
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agopNFS/flexfiles: Fix the reset of struct pgio_header when resending
Trond Myklebust [Fri, 26 Jun 2015 19:37:58 +0000 (15:37 -0400)]
pNFS/flexfiles: Fix the reset of struct pgio_header when resending

commit d620876990f02788d5a663075df007ffb91bdfad upstream.

hdr->good_bytes needs to be set to the length of the request, not
zero.

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agopNFS: Fix a memory leak when attempted pnfs fails
Trond Myklebust [Wed, 17 Jun 2015 23:41:51 +0000 (19:41 -0400)]
pNFS: Fix a memory leak when attempted pnfs fails

commit 1ca018d28d96d07788474abf66a5f3e9594841f5 upstream.

pnfs_do_write() expects the call to pnfs_write_through_mds() to free the
pgio header and to release the layout segment before exiting. The problem
is that nfs_pgio_data_destroy() doesn't actually do this; it only frees
the memory allocated by nfs_generic_pgio().

Ditto for pnfs_do_read()...

Fix in both cases is to add a call to hdr->release(hdr).

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agoclk: qcom: Use parent rate when set rate to pixel RCG clock
Hai Li [Thu, 25 Jun 2015 22:35:33 +0000 (18:35 -0400)]
clk: qcom: Use parent rate when set rate to pixel RCG clock

commit 6d451367bfa16fc103604bacd258f534c65d1540 upstream.

Since the parent rate has been recalculated, pixel RCG clock
should rely on it to find the correct M/N values during set_rate,
instead of calling __clk_round_rate() to its parent again.

Signed-off-by: Hai Li <hali@codeaurora.org>
Tested-by: Archit Taneja <architt@codeaurora.org>
Fixes: 99cbd064b059 ("clk: qcom: Support display RCG clocks")
[sboyd@codeaurora.org: Silenced unused parent variable warning]
Signed-off-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agoclk: ti: dra7-atl-clock: Fix possible ERR_PTR dereference
Krzysztof Kozlowski [Wed, 13 May 2015 06:54:40 +0000 (15:54 +0900)]
clk: ti: dra7-atl-clock: Fix possible ERR_PTR dereference

commit e0cdcda508f110b7ec190dc7c5eb2869ba73a535 upstream.

of_clk_get_from_provider() returns ERR_PTR on failure. The
dra7-atl-clock driver was not checking its return value and
immediately used it in __clk_get_hw().  __clk_get_hw()
dereferences supplied clock, if it is not NULL, so in that case
it would dereference an ERR_PTR.

Fixes: 9ac33b0ce81f ("CLK: TI: Driver for DRA7 ATL (Audio Tracking Logic)")
Signed-off-by: Krzysztof Kozlowski <k.kozlowski@samsung.com>
Signed-off-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agoclk: Fix JSON output in debugfs
Stefan Wahren [Wed, 29 Apr 2015 16:36:43 +0000 (16:36 +0000)]
clk: Fix JSON output in debugfs

commit 7cb81136d2efe0f5ed9d965857f4756a15e6c338 upstream.

key/value pairs in a JSON object must be separated by a comma.
After adding the properties "accuracy" and "phase" the JSON output
of /sys/kernel/debug/clk/clk_dump is invalid.

So add the missing commas to fix it.

Fixes: 5279fc402ae5 ("clk: add clk accuracy retrieval support")
Signed-off-by: Stefan Wahren <stefan.wahren@i2se.com>
[sboyd@codeaurora.org: Added comment in function]
Signed-off-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agogpiolib: Add missing dummies for the unified device properties interface
Geert Uytterhoeven [Thu, 7 May 2015 08:08:08 +0000 (01:08 -0700)]
gpiolib: Add missing dummies for the unified device properties interface

commit 496e7ce2a46562938edcb74f65b26068ee8895f6 upstream.

If GPIOLIB=n:

    drivers/leds/leds-gpio.c: In function ‘gpio_leds_create’:
    drivers/leds/leds-gpio.c:187: error: implicit declaration of function ‘devm_get_gpiod_from_child’
    drivers/leds/leds-gpio.c:187: warning: assignment makes pointer from integer without a cast

Add dummies for fwnode_get_named_gpiod() and devm_get_gpiod_from_child()
for the !GPIOLIB case to fix this.

Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>
Fixes: 40b7318319281b1b ("gpio: Support for unified device properties interface")
Acked-by: Alexandre Courbot <acourbot@nvidia.com>
Acked-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: Bryan Wu <cooloney@gmail.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agowatchdog: omap: assert the counter being stopped before reprogramming
Uwe Kleine-König [Wed, 29 Apr 2015 18:38:46 +0000 (20:38 +0200)]
watchdog: omap: assert the counter being stopped before reprogramming

commit 530c11d432727c697629ad5f9d00ee8e2864d453 upstream.

The omap watchdog has the annoying behaviour that writes to most
registers don't have any effect when the watchdog is already running.
Quoting the AM335x reference manual:

To modify the timer counter value (the WDT_WCRR register),
prescaler ratio (the WDT_WCLR[4:2] PTV bit field), delay
configuration value (the WDT_WDLY[31:0] DLY_VALUE bit field), or
the load value (the WDT_WLDR[31:0] TIMER_LOAD bit field), the
watchdog timer must be disabled by using the start/stop sequence
(the WDT_WSPR register).

Currently the timer is stopped in the .probe callback but still there
are possibilities that yield to a situation where omap_wdt_start is
entered with the timer running (e.g. when /dev/watchdog is closed
without stopping and then reopened). In such a case programming the
timeout silently fails!

To circumvent this stop the timer before reprogramming.

Assuming one of the first things the watchdog user does is setting the
timeout explicitly nothing too bad should happen because this explicit
setting works fine.

Fixes: 7768a13c252a ("[PATCH] OMAP: Add Watchdog driver support")
Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Reviewed-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: Wim Van Sebroeck <wim@iguana.be>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agoof: return NUMA_NO_NODE from fallback of_node_to_nid()
Konstantin Khlebnikov [Wed, 8 Apr 2015 16:59:20 +0000 (19:59 +0300)]
of: return NUMA_NO_NODE from fallback of_node_to_nid()

commit c8fff7bc5bba6bd59cad40441c189c4efe7190f6 upstream.

Node 0 might be offline as well as any other numa node,
in this case kernel cannot handle memory allocation and crashes.

Signed-off-by: Konstantin Khlebnikov <khlebnikov@yandex-team.ru>
Fixes: 0c3f061c195c ("of: implement of_node_to_nid as a weak function")
Signed-off-by: Grant Likely <grant.likely@linaro.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agoovl: lookup whiteouts outside iterate_dir()
Miklos Szeredi [Mon, 22 Jun 2015 11:53:48 +0000 (13:53 +0200)]
ovl: lookup whiteouts outside iterate_dir()

commit cdb672795876d7bc1870aed9a2d7cb59f43d1d96 upstream.

If jffs2 can deadlock on overlayfs readdir because it takes the same lock
on ->iterate() as in ->lookup().

Fix by moving whiteout checking outside iterate_dir().  Optimized by
collecting potential whiteouts (DT_CHR) in a temporary list and if
non-empty iterating throug these and checking for a 0/0 chardev.

Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Fixes: 49c21e1cacd7 ("ovl: check whiteout while reading directory")
Reported-by: Roman Yeryomin <leroi.lists@gmail.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agodell-laptop: Fix allocating & freeing SMI buffer page
Pali Rohár [Tue, 23 Jun 2015 08:11:19 +0000 (10:11 +0200)]
dell-laptop: Fix allocating & freeing SMI buffer page

commit b8830a4e71b15d0364ac8e6c55301eea73f211da upstream.

This commit fix kernel crash when probing for rfkill devices in dell-laptop
driver failed. Function free_page() was incorrectly used on struct page *
instead of virtual address of SMI buffer.

This commit also simplify allocating page for SMI buffer by using
__get_free_page() function instead of sequential call of functions
alloc_page() and page_address().

Signed-off-by: Pali Rohár <pali.rohar@gmail.com>
Acked-by: Michal Hocko <mhocko@suse.cz>
Signed-off-by: Darren Hart <dvhart@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agoof/address: use atomic allocation in pci_register_io_range()
Jingoo Han [Wed, 17 Jun 2015 15:12:27 +0000 (00:12 +0900)]
of/address: use atomic allocation in pci_register_io_range()

commit 294240ffe784e951dc2ef070da04fa31ef6db3a0 upstream.

When kzalloc() is called under spin_lock(), GFP_ATOMIC should be
used to avoid sleeping allocation.
The call tree is:
  of_pci_range_to_resource()
    --> pci_register_io_range() <-- takes spin_lock(&io_range_lock);
       --> kzalloc()

Signed-off-by: Jingoo Han <jingoohan1@gmail.com>
Signed-off-by: Rob Herring <robh@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agoideapad: fix software rfkill setting
Arnd Bergmann [Sat, 13 Jun 2015 13:23:33 +0000 (15:23 +0200)]
ideapad: fix software rfkill setting

commit 4b200b4604bec3388426159f1656109d19fadf6e upstream.

This fixes a several year old regression that I found while trying
to get the Yoga 3 11 to work. The ideapad_rfk_set function is meant
to send a command to the embedded controller through ACPI, but
as of c1f73658ed, it sends the index of the rfkill device instead
of the command, and ignores the opcode field.

This changes it back to the original behavior, which indeed
flips the rfkill state as seen in the debugfs interface.

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Fixes: c1f73658ed ("ideapad: pass ideapad_priv as argument (part 2)")
Signed-off-by: Darren Hart <dvhart@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agoideapad_laptop: Lenovo G50-30 fix rfkill reports wireless blocked
Dmitry Tunin [Sun, 18 Jan 2015 12:44:40 +0000 (15:44 +0300)]
ideapad_laptop: Lenovo G50-30 fix rfkill reports wireless blocked

commit 4fa9dabcffc8e16601307d3d56b58c68d9716ba4 upstream.

Lenovo G30-50 does not have a hardware wireless switch and wireless
is always blocked.

BugLink: https://bugs.launchpad.net/bugs/1397021
Signed-off-by: Dmitry Tunin <hanipouspilot@gmail.com>
Signed-off-by: Philippe Coval <philippe.coval@open.eurogiciel.org>
[dvhart@linux.intel.com: Reordered dmi id per Phillippe's later version]
Signed-off-by: Darren Hart <dvhart@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agoclocksource: exynos_mct: Avoid blocking calls in the cpu hotplug notifier
Damian Eppel [Fri, 26 Jun 2015 13:23:04 +0000 (15:23 +0200)]
clocksource: exynos_mct: Avoid blocking calls in the cpu hotplug notifier

commit 56a94f13919c0db5958611b388e1581b4852f3c9 upstream.

Whilst testing cpu hotplug events on kernel configured with
DEBUG_PREEMPT and DEBUG_ATOMIC_SLEEP we get following BUG message,
caused by calling request_irq() and free_irq() in the context of
hotplug notification (which is in this case atomic context).

[   40.785859] CPU1: Software reset
[   40.786660] BUG: sleeping function called from invalid context at mm/slub.c:1241
[   40.786668] in_atomic(): 1, irqs_disabled(): 128, pid: 0, name: swapper/1
[   40.786678] Preemption disabled at:[<  (null)>]   (null)
[   40.786681]
[   40.786692] CPU: 1 PID: 0 Comm: swapper/1 Not tainted 3.19.0-rc4-00024-g7dca860 #36
[   40.786698] Hardware name: SAMSUNG EXYNOS (Flattened Device Tree)
[   40.786728] [<c0014a00>] (unwind_backtrace) from [<c0011980>] (show_stack+0x10/0x14)
[   40.786747] [<c0011980>] (show_stack) from [<c0449ba0>] (dump_stack+0x70/0xbc)
[   40.786767] [<c0449ba0>] (dump_stack) from [<c00c6124>] (kmem_cache_alloc+0xd8/0x170)
[   40.786785] [<c00c6124>] (kmem_cache_alloc) from [<c005d6f8>] (request_threaded_irq+0x64/0x128)
[   40.786804] [<c005d6f8>] (request_threaded_irq) from [<c0350b8c>] (exynos4_local_timer_setup+0xc0/0x13c)
[   40.786820] [<c0350b8c>] (exynos4_local_timer_setup) from [<c0350ca8>] (exynos4_mct_cpu_notify+0x30/0xa8)
[   40.786838] [<c0350ca8>] (exynos4_mct_cpu_notify) from [<c003b330>] (notifier_call_chain+0x44/0x84)
[   40.786857] [<c003b330>] (notifier_call_chain) from [<c0022fd4>] (__cpu_notify+0x28/0x44)
[   40.786873] [<c0022fd4>] (__cpu_notify) from [<c0013714>] (secondary_start_kernel+0xec/0x150)
[   40.786886] [<c0013714>] (secondary_start_kernel) from [<40008764>] (0x40008764)

Interrupts cannot be requested/freed in the CPU_STARTING/CPU_DYING
notifications which run on the hotplugged cpu with interrupts and
preemption disabled.

To avoid the issue, request the interrupts for all possible cpus in
the boot code. The interrupts are marked NO_AUTOENABLE to avoid a racy
request_irq/disable_irq() sequence. The flag prevents the
request_irq() code from enabling the interrupt immediately.

The interrupt is then enabled in the CPU_STARTING notifier of the
hotplugged cpu and again disabled with disable_irq_nosync() in the
CPU_DYING notifier.

[ tglx: Massaged changelog to match the patch ]

Fixes: 7114cd749a12 ("clocksource: exynos_mct: use (request/free)_irq calls for local timer registration")
Reported-by: Krzysztof Kozlowski <k.kozlowski@samsung.com>
Reviewed-by: Krzysztof Kozlowski <k.kozlowski@samsung.com>
Tested-by: Krzysztof Kozlowski <k.kozlowski@samsung.com>
Tested-by: Marcin Jabrzyk <m.jabrzyk@samsung.com>
Signed-off-by: Damian Eppel <d.eppel@samsung.com>
Cc: m.szyprowski@samsung.com
Cc: kyungmin.park@samsung.com
Cc: daniel.lezcano@linaro.org
Cc: kgene@kernel.org
Cc: linux-arm-kernel@lists.infradead.org
Link: http://lkml.kernel.org/r/1435324984-7328-1-git-send-email-d.eppel@samsung.com
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: <stable@vger.kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agoe1000e: Cleanup handling of VLAN_HLEN as a part of max frame size
Alexander Duyck [Sat, 2 May 2015 07:52:00 +0000 (00:52 -0700)]
e1000e: Cleanup handling of VLAN_HLEN as a part of max frame size

commit 8084b86dcfbc4b4822868c1dbdb429b5c08154e2 upstream.

When the VLAN_HLEN was added to the calculation for the maximum frame size
there seems to have been a number of issues added to the driver.

The first issue is that in some cases the maximum frame size for a device
never really reached the actual maximum frame size as the VLAN header
length was not included the calculation for that value.  As a result some
parts only supported a maximum frame size of either 1496 in the case of
parts that didn't support jumbo frames, and 8996 in the case of the parts
that do.

The second issue is the fact that there were several checks that weren't
updated so as a result setting an MTU of 1500 was treated as enabling jumbo
frames as the calculated value was 1522 instead of 1518.  I have addressed
those by replacing ETH_FRAME_LEN with VLAN_ETH_FRAME_LEN where appropriate.

The final issue was the fact that lowering the MTU below 1500 would cause
the driver to allocate 2K buffers for the rings.  This is an old issue that
was fixed several years ago in igb/ixgbe and I am addressing now by just
replacing == with a <= so that we always just round up to 1522 for anything
that isn't a jumbo frame.

Fixes: c751a3d58cf2d ("e1000e: Correctly include VLAN_HLEN when changing interface MTU")
Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agomac80211: prevent possible crypto tx tailroom corruption
Michal Kazior [Fri, 22 May 2015 08:22:40 +0000 (10:22 +0200)]
mac80211: prevent possible crypto tx tailroom corruption

commit ab499db80fcf07c18e4053f91a619500f663e90e upstream.

There was a possible race between
ieee80211_reconfig() and
ieee80211_delayed_tailroom_dec(). This could
result in inability to transmit data if driver
crashed during roaming or rekeying and subsequent
skbs with insufficient tailroom appeared.

This race was probably never seen in the wild
because a device driver would have to crash AND
recover within 0.5s which is very unlikely.

I was able to prove this race exists after
changing the delay to 10s locally and crashing
ath10k via debugfs immediately after GTK
rekeying. In case of ath10k the counter went below
0. This was harmless but other drivers which
actually require tailroom (e.g. for WEP ICV or
MMIC) could end up with the counter at 0 instead
of >0 and introduce insufficient skb tailroom
failures because mac80211 would not resize skbs
appropriately anymore.

Fixes: 8d1f7ecd2af5 ("mac80211: defer tailroom counter manipulation when roaming")
Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agocfg80211: ignore netif running state when changing iftype
Michal Kazior [Fri, 22 May 2015 08:57:22 +0000 (10:57 +0200)]
cfg80211: ignore netif running state when changing iftype

commit 6cbfb1bb66e4e85da5db78e8ff429a85bd84ce64 upstream.

It was possible for mac80211 to be coerced into an
unexpected flow causing sdata union to become
corrupted. Station pointer was put into
sdata->u.vlan.sta memory location while it was
really master AP's sdata->u.ap.next_beacon. This
led to station entry being later freed as
next_beacon before __sta_info_flush() in
ieee80211_stop_ap() and a subsequent invalid
pointer dereference crash.

The problem was that ieee80211_ptr->use_4addr
wasn't cleared on interface type changes.

This could be reproduced with the following steps:

 # host A and host B have just booted; no
 # wpa_s/hostapd running; all vifs are down
 host A> iw wlan0 set type station
 host A> iw wlan0 set 4addr on
 host A> printf 'interface=wlan0\nssid=4addrcrash\nchannel=1\nwds_sta=1' > /tmp/hconf
 host A> hostapd -B /tmp/conf
 host B> iw wlan0 set 4addr on
 host B> ifconfig wlan0 up
 host B> iw wlan0 connect -w hostAssid
 host A> pkill hostapd
 # host A crashed:

 [  127.928192] BUG: unable to handle kernel NULL pointer dereference at 00000000000006c8
 [  127.929014] IP: [<ffffffff816f4f32>] __sta_info_flush+0xac/0x158
 ...
 [  127.934578]  [<ffffffff8170789e>] ieee80211_stop_ap+0x139/0x26c
 [  127.934578]  [<ffffffff8100498f>] ? dump_trace+0x279/0x28a
 [  127.934578]  [<ffffffff816dc661>] __cfg80211_stop_ap+0x84/0x191
 [  127.934578]  [<ffffffff816dc7ad>] cfg80211_stop_ap+0x3f/0x58
 [  127.934578]  [<ffffffff816c5ad6>] nl80211_stop_ap+0x1b/0x1d
 [  127.934578]  [<ffffffff815e53f8>] genl_family_rcv_msg+0x259/0x2b5

Note: This isn't a revert of f8cdddb8d61d
("cfg80211: check iface combinations only when
iface is running") as far as functionality is
considered because b6a550156bc ("cfg80211/mac80211:
move more combination checks to mac80211") moved
the logic somewhere else already.

Fixes: f8cdddb8d61d ("cfg80211: check iface combinations only when iface is running")
Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agoiwlwifi: mvm: fix ROC reference accounting
Eliad Peller [Sun, 19 Apr 2015 08:41:04 +0000 (11:41 +0300)]
iwlwifi: mvm: fix ROC reference accounting

commit c779273b37bec14c33feeab11c4d457a24bc64e0 upstream.

commit b112889c5af8124 ("iwlwifi: mvm: add Aux ROC request/response flow")
added aux ROC flow in addition to the existing ROC flow. While doing
it, it moved the ROC reference release to a common work item, which
is being called for both the ROC and aux ROC flows.

This resulted in invalid reference accounting, as no reference was
taken in case of aux ROC, while a reference was released on completion.

Fix it by adding a reference for the aux ROC as well, and release
only the relevant references on completion (according to the set bits).

While at it, convert cancel_work_sync() to flush_work(), in order
to make sure the references are being cleaned properly.

Fixes: b112889c5af8 ("iwlwifi: mvm: add Aux ROC request/response flow")
Signed-off-by: Eliad Peller <eliadx.peller@intel.com>
Reviewed-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agomac80211: fix the beacon csa counter for mesh and ibss
Chun-Yeow Yeoh [Tue, 9 Jun 2015 05:35:33 +0000 (13:35 +0800)]
mac80211: fix the beacon csa counter for mesh and ibss

commit 8df734e865b74d9f273216482a45a38269dc767a upstream.

The csa counter has moved from sdata to beacon/presp but
it is not updated accordingly for mesh and ibss. Fix this.

Fixes: af296bdb8da4 ("mac80211: move csa counters from sdata to beacon/presp")
Signed-off-by: Chun-Yeow Yeoh <yeohchunyeow@gmail.com>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agosecurity_syslog() should be called once only
Vasily Averin [Thu, 25 Jun 2015 22:01:44 +0000 (15:01 -0700)]
security_syslog() should be called once only

commit d194e5d666225b04c7754471df0948f645b6ab3a upstream.

The final version of commit 637241a900cb ("kmsg: honor dmesg_restrict
sysctl on /dev/kmsg") lost few hooks, as result security_syslog() are
processed incorrectly:

- open of /dev/kmsg checks syslog access permissions by using
  check_syslog_permissions() where security_syslog() is not called if
  dmesg_restrict is set.

- syslog syscall and /proc/kmsg calls do_syslog() where security_syslog
  can be executed twice (inside check_syslog_permissions() and then
  directly in do_syslog())

With this patch security_syslog() is called once only in all
syslog-related operations regardless of dmesg_restrict value.

Fixes: 637241a900cb ("kmsg: honor dmesg_restrict sysctl on /dev/kmsg")
Signed-off-by: Vasily Averin <vvs@virtuozzo.com>
Cc: Kees Cook <keescook@chromium.org>
Cc: Josh Boyer <jwboyer@redhat.com>
Cc: Eric Paris <eparis@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years ago__bitmap_parselist: fix bug in empty string handling
Chris Metcalf [Thu, 25 Jun 2015 22:02:08 +0000 (15:02 -0700)]
__bitmap_parselist: fix bug in empty string handling

commit 2528a8b8f457d7432552d0e2b6f0f4046bb702f4 upstream.

bitmap_parselist("", &mask, nmaskbits) will erroneously set bit zero in
the mask.  The same bug is visible in cpumask_parselist() since it is
layered on top of the bitmask code, e.g.  if you boot with "isolcpus=",
you will actually end up with cpu zero isolated.

The bug was introduced in commit 4b060420a596 ("bitmap, irq: add
smp_affinity_list interface to /proc/irq") when bitmap_parselist() was
generalized to support userspace as well as kernelspace.

Fixes: 4b060420a596 ("bitmap, irq: add smp_affinity_list interface to /proc/irq")
Signed-off-by: Chris Metcalf <cmetcalf@ezchip.com>
Cc: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agocompiler-intel: fix wrong compiler barrier() macro
Daniel Borkmann [Thu, 25 Jun 2015 22:01:05 +0000 (15:01 -0700)]
compiler-intel: fix wrong compiler barrier() macro

commit b86a50c3b5414eafdbee7f34af4a201a4a7817c2 upstream.

Cleanup commit 73679e508201 ("compiler-intel.h: Remove duplicate
definition") removed the double definition of __memory_barrier()
intrinsics.

However, in doing so, it also removed the preceding #undef barrier by
accident, meaning, the actual barrier() macro from compiler-gcc.h with
inline asm is still in place as __GNUC__ is provided.

Subsequently, barrier() can never be defined as __memory_barrier() from
compiler.h since it already has a definition in place and if we trust
the comment in compiler-intel.h, ecc doesn't support gcc specific asm
statements.

I don't have an ecc at hand (unsure if that's still used in the field?)
and only found this by accident during code review, a revert of that
cleanup would be simplest option.

Fixes: 73679e508201 ("compiler-intel.h: Remove duplicate definition")
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Pranith Kumar <bobby.prani@gmail.com>
Cc: Pranith Kumar <bobby.prani@gmail.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: mancha security <mancha1@zoho.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agofirmware: dmi_scan: Only honor end-of-table for 64-bit tables
Jean Delvare [Thu, 25 Jun 2015 07:06:55 +0000 (09:06 +0200)]
firmware: dmi_scan: Only honor end-of-table for 64-bit tables

commit 17cd5bd5391e6e7b363d66335e1bc6760ae969b9 upstream.

A 32-bit entry point to a DMI table says how many structures the table
contains. The SMBIOS specification explicitly says that end-of-table
markers should be ignored if they are not actually at the end of the
DMI table. So only honor the end-of-table marker for tables accessed
through 64-bit entry points, as they do not specify a structure count.

Fixes: fc43026278 ("dmi: add support for SMBIOS 3.0 64-bit entry point")
Signed-off-by: Jean Delvare <jdelvare@suse.de>
Acked-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Cc: Leif Lindholm <leif.lindholm@linaro.org>
Cc: Matt Fleming <matt.fleming@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agoPM / sleep: Increase default DPM watchdog timeout to 60
Takashi Iwai [Wed, 24 Jun 2015 22:35:16 +0000 (00:35 +0200)]
PM / sleep: Increase default DPM watchdog timeout to 60

commit fff3b16d2754a061a3549c4307a186423a0128fd upstream.

Many harddisks (mostly WD ones) have firmware problems and take too
long, more than 10 seconds, to resume from suspend.  And this often
exceeds the default DPM watchdog timeout (12 seconds), resulting in a
kernel panic out of sudden.

Since most distros just take the default as is, we should give a bit
more safer value.  This patch increases the default value from 12
seconds to one minute, which has been confirmed to be long enough for
such problematic disks.

Link: https://bugzilla.kernel.org/show_bug.cgi?id=91921
Fixes: 70fea60d888d (PM / Sleep: Detect device suspend/resume lockup and log event)
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agomm/hugetlb: introduce minimum hugepage order
Naoya Horiguchi [Wed, 24 Jun 2015 23:56:59 +0000 (16:56 -0700)]
mm/hugetlb: introduce minimum hugepage order

commit 641844f5616d7c6597309f560838f996466d7aac upstream.

Currently the initial value of order in dissolve_free_huge_page is 64 or
32, which leads to the following warning in static checker:

  mm/hugetlb.c:1203 dissolve_free_huge_pages()
  warn: potential right shift more than type allows '9,18,64'

This is a potential risk of infinite loop, because 1 << order (== 0) is used
in for-loop like this:

  for (pfn =3D start_pfn; pfn < end_pfn; pfn +=3D 1 << order)
      ...

So this patch fixes it by using global minimum_order calculated at boot time.

    text    data     bss     dec     hex filename
   28313     469   84236  113018   1b97a mm/hugetlb.o
   28256     473   84236  112965   1b945 mm/hugetlb.o (patched)

Fixes: c8721bbbdd36 ("mm: memory-hotplug: enable memory hotplug to handle hugepage")
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agotty: remove platform_sysrq_reset_seq
Arnd Bergmann [Tue, 26 May 2015 21:45:29 +0000 (14:45 -0700)]
tty: remove platform_sysrq_reset_seq

commit ffb6e0c9a0572f8e5f8e9337a1b40ac2ec1493a1 upstream.

The platform_sysrq_reset_seq code was intended as a way for an embedded
platform to provide its own sysrq sequence at compile time. After over two
years, nobody has started using it in an upstream kernel, and the platforms
that were interested in it have moved on to devicetree, which can be used
to configure the sequence without requiring kernel changes. The method is
also incompatible with the way that most architectures build support for
multiple platforms into a single kernel.

Now the code is producing warnings when built with gcc-5.1:

drivers/tty/sysrq.c: In function 'sysrq_init':
drivers/tty/sysrq.c:959:33: warning: array subscript is above array bounds [-Warray-bounds]
   key = platform_sysrq_reset_seq[i];

We could fix this, but it seems unlikely that it will ever be used, so
let's just remove the code instead. We still have the option to pass the
sequence either in DT, using the kernel command line, or using the
/sys/module/sysrq/parameters/reset_seq file.

Fixes: 154b7a489a ("Input: sysrq - allow specifying alternate reset sequence")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Dmitry Torokhov <dmitry.torokhov@gmail.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agoRDMA/ocrdma: fix double free on pd
Colin Ian King [Fri, 5 Jun 2015 14:47:27 +0000 (15:47 +0100)]
RDMA/ocrdma: fix double free on pd

commit 4dc544427991e3cef38ce3ae124b7e6557063bd3 upstream.

A reorganisation of the PD allocation and deallocation in commit
9ba1377daa ("RDMA/ocrdma: Move PD resource management to driver.")
introduced a double free on pd, as detected by static analysis by
smatch:

drivers/infiniband/hw/ocrdma/ocrdma_verbs.c:682 ocrdma_alloc_pd()
  error: double free of 'pd'^

The original call to ocrdma_mbx_dealloc_pd() (which does not kfree
pd) was replaced with a call to _ocrdma_dealloc_pd() (which does
kfree pd).  The kfree following this call causes the double free,
so just remove it to fix the problem.

Fixes: 9ba1377daa ("RDMA/ocrdma: Move PD resource management to driver.")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Acked-By: Devesh Sharma <devesh.sharma@avagotech.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agoPM / clk: Fix clock error check in __pm_clk_add()
Geert Uytterhoeven [Fri, 8 May 2015 08:47:43 +0000 (10:47 +0200)]
PM / clk: Fix clock error check in __pm_clk_add()

commit 3fc3a0be0dab352e065d1dad7d3f81953ed0d4bc upstream.

In the final iteration of commit 245bd6f6af8a62a2 ("PM / clock_ops: Add
pm_clk_add_clk()"), a refcount increment was added by Grygorii Strashko.
However, the accompanying IS_ERR() check operates on the wrong clock
pointer, which is always zero at this point, i.e. not an error.
This may lead to a NULL pointer dereference later, when __clk_get()
tries to dereference an error pointer.

Check the passed clock pointer instead to fix this.

Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Fixes: 245bd6f6af8a62a2 ("PM / clock_ops: Add pm_clk_add_clk()")
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agommc: sdhci: Restore behavior while creating OCR mask
Ulf Hansson [Fri, 5 Jun 2015 09:40:08 +0000 (11:40 +0200)]
mmc: sdhci: Restore behavior while creating OCR mask

commit 5fd26c7ecb32082745b0bd33c8e35badd1cb5a91 upstream.

Commit 3a48edc4bd68 ("mmc: sdhci: Use mmc core regulator infrastucture")
changed the behavior for how to assign the ocr_avail mask for the mmc
host. More precisely it started to mask the bits instead of assigning
them.

Restore the behavior, but also make it clear that an OCR mask created
from an external regulator overrides the other ones. The OCR mask is
determined by one of the following with this priority:

1. Supported ranges of external regulator if one supplies VDD
2. Host OCR mask if set by the driver (based on DT properties)
3. The capabilities reported by the controller itself

Fixes: 3a48edc4bd68 ("mmc: sdhci: Use mmc core regulator infrastucture")
Cc: Tim Kryger <tim.kryger@gmail.com>
Reported-by: Yangbo Lu <yangbo.lu@freescale.com>
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
Reviewed-by: Tim Kryger <tim.kryger@gmail.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agommc: card: Fixup request missing in mmc_blk_issue_rw_rq
Ding Wang [Mon, 18 May 2015 12:14:15 +0000 (20:14 +0800)]
mmc: card: Fixup request missing in mmc_blk_issue_rw_rq

commit 29535f7b797df35cc9b6b3bca635591cdd3dd2a8 upstream.

The current handler of MMC_BLK_CMD_ERR in mmc_blk_issue_rw_rq function
may cause new coming request permanent missing when the ongoing
request (previoulsy started) complete end.

The problem scenario is as follows:
(1) Request A is ongoing;
(2) Request B arrived, and finally mmc_blk_issue_rw_rq() is called;
(3) Request A encounters the MMC_BLK_CMD_ERR error;
(4) In the error handling of MMC_BLK_CMD_ERR, suppose mmc_blk_cmd_err()
    end request A completed and return zero. Continue the error handling,
    suppose mmc_blk_reset() reset device success;
(5) Continue the execution, while loop completed because variable ret
    is zero now;
(6) Finally, mmc_blk_issue_rw_rq() return without processing request B.

The process related to the missing request may wait that IO request
complete forever, possibly crashing the application or hanging the system.

Fix this issue by starting new request when reset success.

Signed-off-by: Ding Wang <justin.wang@spreadtrum.com>
Fixes: 67716327eec7 ("mmc: block: add eMMC hardware reset support")
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agoserial: samsung: only use earlycon for console
Arnd Bergmann [Tue, 19 May 2015 20:26:04 +0000 (22:26 +0200)]
serial: samsung: only use earlycon for console

commit 357d56151976a78d90dc3dfac01777de0ef05212 upstream.

A configuration that enables earlycon but not the core console
code causes a link error:

  drivers/built-in.o: In function `setup_earlycon':
  drivers/tty/serial/earlycon.c:70: undefined reference to `uart_parse_earlycon'

That error can be triggered by the newly added samsung earlycon support,
which is missing a 'select' statement.

As suggested by Peter Hurley, solves the problem by moving the
'select SERIAL_EARLYCON' statement to the samsung console driver
option, as it is done by all other console drivers.

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Fixes: b94ba0328d3b3 ("serial: samsung: Add support for early console")
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agoACPI / PCI: Fix regressions caused by resource_size_t overflow with 32-bit kernel
Jiang Liu [Wed, 8 Jul 2015 07:26:39 +0000 (15:26 +0800)]
ACPI / PCI: Fix regressions caused by resource_size_t overflow with 32-bit kernel

commit 1fb01ca93a1348a1469b8777326cd7632483de77 upstream.

Zoltan Boszormenyi reported this regression:
  "There's a Realtek RTL8111/8168/8411 (PCI ID 10ec:8168, Subsystem ID
   1565:230e) network chip on the mainboard. After the r8169 driver loaded
   the IRQs in the machine went berserk. Keyboard keypressed arrived with
   considerable latency and duplicated, so no real work was possible.
   The machine responded to the power button but didn't actually power
   down. It just stuck at the powering down message. I had to press the
   power button for 4 seconds to power it down.

   The computer is a POS machine with a big battery inside. Because of this,
   either ACPI or the Realtek chip kept the bad state and after rebooting,
   the network chip didn't even show up in lspci. Not even the PXE ROM
   announced itself during boot. I had to disconnect the battery to beat
   some sense back to the computer.

   The regression happens with 4.0.5, 4.1.0-rc8 and 4.1.0-final. 3.18.16 was
   good."

The regression is caused by commit 593669c2ac0f (x86/PCI/ACPI: Use common
ACPI resource interfaces to simplify implementation). Since commit
593669c2ac0f, x86 PCI ACPI host bridge driver validates ACPI resources by
first converting an ACPI resource to a 'struct resource' structure and
then applying checks against the converted resource structure. The 'start'
and 'end' fields in 'struct resource' are defined to be type of
resource_size_t, which may be 32 bits or 64 bits depending on
CONFIG_PHYS_ADDR_T_64BIT.

This may cause incorrect resource validation results with 32-bit kernels
because 64-bit ACPI resource descriptors may get truncated when converting
to 32-bit 'start' and 'end' fields in 'struct resource'. It eventually
affects PCI resource allocation subsystem and makes some PCI devices and
the system behave abnormally due to incorrect resource assignment.

So enhance the ACPI resource parsing interfaces to ignore ACPI resource
descriptors with address/offset above 4G when running in 32-bit mode.

With the fix applied, the behavior of the machine was restored to how
3.18.16 worked, i.e. the memory range that is over 4GB is ignored again,
and lspci -vvxxx shows that everything is at the same memory window as
they were with 3.18.16.

Reported-and-tested-by: Boszormenyi Zoltan <zboszor@pr.hu>
Fixes: 593669c2ac0f (x86/PCI/ACPI: Use common ACPI resource interfaces to simplify implementation)
Signed-off-by: Jiang Liu <jiang.liu@linux.intel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agoACPICA: Tables: Enable default 64-bit FADT addresses favor
Lv Zheng [Wed, 1 Jul 2015 06:43:34 +0000 (14:43 +0800)]
ACPICA: Tables: Enable default 64-bit FADT addresses favor

commit 0ea61381788a37d864f9841b0fe97d40f7058f3b upstream.

ACPICA commit 4da56eeae0749dfe8491285c1e1fad48f6efafd8

The following commit temporarily disables correct 64-bit FADT addresses
favor during the period the root cause of the bug is not fixed:
 Commit: 85dbd5801f62b66e2aa7826aaefcaebead44c8a6
 ACPICA: Tables: Restore old behavor to favor 32-bit FADT addresses.

With enough protections, this patch re-enables 64-bit FADT addresses by
default. If regressions are reported against such change, this patch should
be bisected and reverted.
Note that 64-bit FACS favor and 64-bit firmware waking vector favor are
excluded by this commit in order not to break OSPMs. Lv Zheng.

Link: https://bugzilla.kernel.org/show_bug.cgi?id=74021
Link: https://github.com/acpica/acpica/commit/4da56eea
Reported-and-tested-by: Oswald Buddenhagen <ossi@kde.org>
Signed-off-by: Lv Zheng <lv.zheng@intel.com>
Signed-off-by: Bob Moore <robert.moore@intel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agoACPICA: Tables: Fix an issue that FACS initialization is performed twice
Lv Zheng [Wed, 1 Jul 2015 06:43:26 +0000 (14:43 +0800)]
ACPICA: Tables: Fix an issue that FACS initialization is performed twice

commit c04be18448355441a0c424362df65b6422e27bda upstream.

ACPICA commit 90f5332a15e9d9ba83831ca700b2b9f708274658

This patch adds a new FACS initialization flag for acpi_tb_initialize().
acpi_enable_subsystem() might be invoked several times in OS bootup process,
and we don't want FACS initialization to be invoked twice. Lv Zheng.

Link: https://github.com/acpica/acpica/commit/90f5332a
Signed-off-by: Lv Zheng <lv.zheng@intel.com>
Signed-off-by: Bob Moore <robert.moore@intel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agoACPICA: Tables: Enable both 32-bit and 64-bit FACS
Lv Zheng [Wed, 1 Jul 2015 06:43:11 +0000 (14:43 +0800)]
ACPICA: Tables: Enable both 32-bit and 64-bit FACS

commit c04e1fb4396d27f18296db0f914760fa7fe8223a upstream.

ACPICA commit f7b86f35416e3d1f71c3d816ff5075ddd33ed486

The following commit is reported to have broken s2ram on some platforms:
 Commit: 0249ed2444d65d65fc3f3f64f398f1ad0b7e54cd
 ACPICA: Add option to favor 32-bit FADT addresses.
The platform reports 2 FACS tables (which is not allowed by ACPI
specification) and the new 32-bit address favor rule forces OSPMs to use
the FACS table reported via FADT's X_FIRMWARE_CTRL field.

The root cause of the reported bug might be one of the followings:
1. BIOS may favor the 64-bit firmware waking vector address when the
   version of the FACS is greater than 0 and Linux currently only supports
   resuming from the real mode, so the 64-bit firmware waking vector has
   never been set and might be invalid to BIOS while the commit enables
   higher version FACS.
2. BIOS may favor the FACS reported via the "FIRMWARE_CTRL" field in the
   FADT while the commit doesn't set the firmware waking vector address of
   the FACS reported by "FIRMWARE_CTRL", it only sets the firware waking
   vector address of the FACS reported by "X_FIRMWARE_CTRL".

This patch excludes the cases that can trigger the bugs caused by the root
cause 2.

There is no handshaking mechanism can be used by OSPM to tell BIOS which
FACS is currently used. Thus the FACS reported by "FIRMWARE_CTRL" may still
be used by BIOS and the 0 value of the 32-bit firmware waking vector might
trigger such failure.

This patch tries to favor 32bit FACS address in another way where both the
FACS reported by "FIRMWARE_CTRL" and the FACS reported by "X_FIRMWARE_CTRL"
are loaded so that further commit can set firmware waking vector in the
both tables to ensure we can exclude the cases that trigger the bugs caused
by the root cause 2. The exclusion is split into 2 commits as this commit
is also useful for dumping more ACPI tables, it won't get reverted when
such exclusion is no longer necessary. Lv Zheng.

Link: https://bugzilla.kernel.org/show_bug.cgi?id=74021
Link: https://github.com/acpica/acpica/commit/f7b86f35
Reported-and-tested-by: Oswald Buddenhagen <ossi@kde.org>
Signed-off-by: Lv Zheng <lv.zheng@intel.com>
Signed-off-by: Bob Moore <robert.moore@intel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agoACPI / LPSS: Fix up acpi_lpss_create_device()
Rafael J. Wysocki [Mon, 6 Jul 2015 22:31:47 +0000 (00:31 +0200)]
ACPI / LPSS: Fix up acpi_lpss_create_device()

commit d3e13ff3c1aa2403d9a5f371baac088daeb8f56d upstream.

Fix a return value (which should be a negative error code) and a
memory leak (the list allocated by acpi_dev_get_resources() needs
to be freed on ioremap() errors too) in acpi_lpss_create_device()
introduced by commit 4483d59e29fe 'ACPI / LPSS: check the result
of ioremap()'.

Fixes: 4483d59e29fe 'ACPI / LPSS: check the result of ioremap()'
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agoACPI / PNP: Reserve ACPI resources at the fs_initcall_sync stage
Rafael J. Wysocki [Sat, 4 Jul 2015 01:09:03 +0000 (03:09 +0200)]
ACPI / PNP: Reserve ACPI resources at the fs_initcall_sync stage

commit 0294112ee3135fbd15eaa70015af8283642dd970 upstream.

This effectively reverts the following three commits:

 7bc10388ccdd ACPI / resources: free memory on error in add_region_before()
 0f1b414d1907 ACPI / PNP: Avoid conflicting resource reservations
 b9a5e5e18fbf ACPI / init: Fix the ordering of acpi_reserve_resources()

(commit b9a5e5e18fbf introduced regressions some of which, but not
all, were addressed by commit 0f1b414d1907 and commit 7bc10388ccdd
was a fixup on top of the latter) and causes ACPI fixed hardware
resources to be reserved at the fs_initcall_sync stage of system
initialization.

The story is as follows.  First, a boot regression was reported due
to an apparent resource reservation ordering change after a commit
that shouldn't lead to such changes.  Investigation led to the
conclusion that the problem happened because acpi_reserve_resources()
was executed at the device_initcall() stage of system initialization
which wasn't strictly ordered with respect to driver initialization
(and with respect to the initialization of the pcieport driver in
particular), so a random change causing the device initcalls to be
run in a different order might break things.

The response to that was to attempt to run acpi_reserve_resources()
as soon as we knew that ACPI would be in use (commit b9a5e5e18fbf).
However, that turned out to be too early, because it caused resource
reservations made by the PNP system driver to fail on at least one
system and that failure was addressed by commit 0f1b414d1907.

That fix still turned out to be insufficient, though, because
calling acpi_reserve_resources() before the fs_initcall stage of
system initialization caused a boot regression to happen on the
eCAFE EC-800-H20G/S netbook.  That meant that we only could call
acpi_reserve_resources() at the fs_initcall initialization stage
or later, but then we might just as well call it after the PNP
initalization in which case commit 0f1b414d1907 wouldn't be
necessary any more.

For this reason, the changes made by commit 0f1b414d1907 are reverted
(along with a memory leak fixup on top of that commit), the changes
made by commit b9a5e5e18fbf that went too far are reverted too and
acpi_reserve_resources() is changed into fs_initcall_sync, which
will cause it to be executed after the PNP subsystem initialization
(which is an fs_initcall) and before device initcalls (including
the pcieport driver initialization) which should avoid the initial
issue.

Link: https://bugzilla.kernel.org/show_bug.cgi?id=100581
Link: http://marc.info/?t=143092384600002&r=1&w=2
Link: https://bugzilla.kernel.org/show_bug.cgi?id=99831
Link: http://marc.info/?t=143389402600001&r=1&w=2
Fixes: b9a5e5e18fbf "ACPI / init: Fix the ordering of acpi_reserve_resources()"
Reported-by: Roland Dreier <roland@purestorage.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agoACPI / resources: free memory on error in add_region_before()
Dan Carpenter [Wed, 24 Jun 2015 14:30:15 +0000 (17:30 +0300)]
ACPI / resources: free memory on error in add_region_before()

commit 7bc10388ccdd79b3d20463151a1f8e7a590a775b upstream.

There is a small memory leak on error.

Fixes: 0f1b414d1907 (ACPI / PNP: Avoid conflicting resource reservations)
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agocrush: fix a bug in tree bucket decode
Ilya Dryomov [Mon, 29 Jun 2015 16:30:23 +0000 (19:30 +0300)]
crush: fix a bug in tree bucket decode

commit 82cd003a77173c91b9acad8033fb7931dac8d751 upstream.

struct crush_bucket_tree::num_nodes is u8, so ceph_decode_8_safe()
should be used.  -Wconversion catches this, but I guess it went
unnoticed in all the noise it spews.  The actual problem (at least for
common crushmaps) isn't the u32 -> u8 truncation though - it's the
advancement by 4 bytes instead of 1 in the crushmap buffer.

Fixes: http://tracker.ceph.com/issues/2759

Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Reviewed-by: Josh Durgin <jdurgin@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agofuse: initialize fc->release before calling it
Miklos Szeredi [Wed, 1 Jul 2015 14:25:55 +0000 (16:25 +0200)]
fuse: initialize fc->release before calling it

commit 0ad0b3255a08020eaf50e34ef0d6df5bdf5e09ed upstream.

fc->release is called from fuse_conn_put() which was used in the error
cleanup before fc->release was initialized.

[Jeremiah Mahler <jmmahler@gmail.com>: assign fc->release after calling
fuse_conn_init(fc) instead of before.]

Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Fixes: a325f9b92273 ("fuse: update fuse_conn_init() and separate out fuse_conn_kill()")
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agoselinux: fix mprotect PROT_EXEC regression caused by mm change
Stephen Smalley [Fri, 10 Jul 2015 13:40:59 +0000 (09:40 -0400)]
selinux: fix mprotect PROT_EXEC regression caused by mm change

commit 892e8cac99a71f6254f84fc662068d912e1943bf upstream.

commit 66fc13039422ba7df2d01a8ee0873e4ef965b50b ("mm: shmem_zero_setup
skip security check and lockdep conflict with XFS") caused a regression
for SELinux by disabling any SELinux checking of mprotect PROT_EXEC on
shared anonymous mappings.  However, even before that regression, the
checking on such mprotect PROT_EXEC calls was inconsistent with the
checking on a mmap PROT_EXEC call for a shared anonymous mapping.  On a
mmap, the security hook is passed a NULL file and knows it is dealing
with an anonymous mapping and therefore applies an execmem check and no
file checks.  On a mprotect, the security hook is passed a vma with a
non-NULL vm_file (as this was set from the internally-created shmem
file during mmap) and therefore applies the file-based execute check
and no execmem check.  Since the aforementioned commit now marks the
shmem zero inode with the S_PRIVATE flag, the file checks are disabled
and we have no checking at all on mprotect PROT_EXEC.  Add a test to
the mprotect hook logic for such private inodes, and apply an execmem
check in that case.  This makes the mmap and mprotect checking
consistent for shared anonymous mappings, as well as for /dev/zero and
ashmem.

Signed-off-by: Stephen Smalley <sds@tycho.nsa.gov>
Signed-off-by: Paul Moore <pmoore@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agoselinux: don't waste ebitmap space when importing NetLabel categories
Paul Moore [Thu, 9 Jul 2015 18:20:36 +0000 (14:20 -0400)]
selinux: don't waste ebitmap space when importing NetLabel categories

commit 3324603524925c7727207027d1c15e597412d15e upstream.

At present we don't create efficient ebitmaps when importing NetLabel
category bitmaps.  This can present a problem when comparing ebitmaps
since ebitmap_cmp() is very strict about these things and considers
these wasteful ebitmaps not equal when compared to their more
efficient counterparts, even if their values are the same.  This isn't
likely to cause problems on 64-bit systems due to a bit of luck on
how NetLabel/CIPSO works and the default ebitmap size, but it can be
a problem on 32-bit systems.

This patch fixes this problem by being a bit more intelligent when
importing NetLabel category bitmaps by skipping over empty sections
which should result in a nice, efficient ebitmap.

Signed-off-by: Paul Moore <pmoore@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agoBtrfs: fix file corruption after cloning inline extents
Filipe Manana [Tue, 14 Jul 2015 15:09:39 +0000 (16:09 +0100)]
Btrfs: fix file corruption after cloning inline extents

commit ed958762644b404654a6f5d23e869f496fe127c6 upstream.

Using the clone ioctl (or extent_same ioctl, which calls the same extent
cloning function as well) we end up allowing copy an inline extent from
the source file into a non-zero offset of the destination file. This is
something not expected and that the btrfs code is not prepared to deal
with - all inline extents must be at a file offset equals to 0.

For example, the following excerpt of a test case for fstests triggers
a crash/BUG_ON() on a write operation after an inline extent is cloned
into a non-zero offset:

  _scratch_mkfs >>$seqres.full 2>&1
  _scratch_mount

  # Create our test files. File foo has the same 2K of data at offset 4K
  # as file bar has at its offset 0.
  $XFS_IO_PROG -f -s -c "pwrite -S 0xaa 0 4K" \
      -c "pwrite -S 0xbb 4k 2K" \
      -c "pwrite -S 0xcc 8K 4K" \
      $SCRATCH_MNT/foo | _filter_xfs_io

  # File bar consists of a single inline extent (2K size).
  $XFS_IO_PROG -f -s -c "pwrite -S 0xbb 0 2K" \
     $SCRATCH_MNT/bar | _filter_xfs_io

  # Now call the clone ioctl to clone the extent of file bar into file
  # foo at its offset 4K. This made file foo have an inline extent at
  # offset 4K, something which the btrfs code can not deal with in future
  # IO operations because all inline extents are supposed to start at an
  # offset of 0, resulting in all sorts of chaos.
  # So here we validate that clone ioctl returns an EOPNOTSUPP, which is
  # what it returns for other cases dealing with inlined extents.
  $CLONER_PROG -s 0 -d $((4 * 1024)) -l $((2 * 1024)) \
      $SCRATCH_MNT/bar $SCRATCH_MNT/foo

  # Because of the inline extent at offset 4K, the following write made
  # the kernel crash with a BUG_ON().
  $XFS_IO_PROG -c "pwrite -S 0xdd 6K 2K" $SCRATCH_MNT/foo | _filter_xfs_io

  status=0
  exit

The stack trace of the BUG_ON() triggered by the last write is:

  [152154.035903] ------------[ cut here ]------------
  [152154.036424] kernel BUG at mm/page-writeback.c:2286!
  [152154.036424] invalid opcode: 0000 [#1] PREEMPT SMP DEBUG_PAGEALLOC
  [152154.036424] Modules linked in: btrfs dm_flakey dm_mod crc32c_generic xor raid6_pq nfsd auth_rpcgss oid_registry nfs_acl nfs lockd grace fscache sunrpc loop fuse parport_pc acpi_cpu$
  [152154.036424] CPU: 2 PID: 17873 Comm: xfs_io Tainted: G        W       4.1.0-rc6-btrfs-next-11+ #2
  [152154.036424] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.8.1-0-g4adadbd-20150316_085822-nilsson.home.kraxel.org 04/01/2014
  [152154.036424] task: ffff880429f70990 ti: ffff880429efc000 task.ti: ffff880429efc000
  [152154.036424] RIP: 0010:[<ffffffff8111a9d5>]  [<ffffffff8111a9d5>] clear_page_dirty_for_io+0x1e/0x90
  [152154.036424] RSP: 0018:ffff880429effc68  EFLAGS: 00010246
  [152154.036424] RAX: 0200000000000806 RBX: ffffea0006a6d8f0 RCX: 0000000000000001
  [152154.036424] RDX: 0000000000000000 RSI: ffffffff81155d1b RDI: ffffea0006a6d8f0
  [152154.036424] RBP: ffff880429effc78 R08: ffff8801ce389fe0 R09: 0000000000000001
  [152154.036424] R10: 0000000000002000 R11: ffffffffffffffff R12: ffff8800200dce68
  [152154.036424] R13: 0000000000000000 R14: ffff8800200dcc88 R15: ffff8803d5736d80
  [152154.036424] FS:  00007fbf119f6700(0000) GS:ffff88043d280000(0000) knlGS:0000000000000000
  [152154.036424] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
  [152154.036424] CR2: 0000000001bdc000 CR3: 00000003aa555000 CR4: 00000000000006e0
  [152154.036424] Stack:
  [152154.036424]  ffff8803d5736d80 0000000000000001 ffff880429effcd8 ffffffffa04e97c1
  [152154.036424]  ffff880429effd68 ffff880429effd60 0000000000000001 ffff8800200dc9c8
  [152154.036424]  0000000000000001 ffff8800200dcc88 0000000000000000 0000000000001000
  [152154.036424] Call Trace:
  [152154.036424]  [<ffffffffa04e97c1>] lock_and_cleanup_extent_if_need+0x147/0x18d [btrfs]
  [152154.036424]  [<ffffffffa04ea82c>] __btrfs_buffered_write+0x245/0x4c8 [btrfs]
  [152154.036424]  [<ffffffffa04ed14b>] ? btrfs_file_write_iter+0x150/0x3e0 [btrfs]
  [152154.036424]  [<ffffffffa04ed15a>] ? btrfs_file_write_iter+0x15f/0x3e0 [btrfs]
  [152154.036424]  [<ffffffffa04ed2c7>] btrfs_file_write_iter+0x2cc/0x3e0 [btrfs]
  [152154.036424]  [<ffffffff81165a4a>] __vfs_write+0x7c/0xa5
  [152154.036424]  [<ffffffff81165f89>] vfs_write+0xa0/0xe4
  [152154.036424]  [<ffffffff81166855>] SyS_pwrite64+0x64/0x82
  [152154.036424]  [<ffffffff81465197>] system_call_fastpath+0x12/0x6f
  [152154.036424] Code: 48 89 c7 e8 0f ff ff ff 5b 41 5c 5d c3 0f 1f 44 00 00 55 48 89 e5 41 54 53 48 89 fb e8 ae ef 00 00 49 89 c4 48 8b 03 a8 01 75 02 <0f> 0b 4d 85 e4 74 59 49 8b 3c 2$
  [152154.036424] RIP  [<ffffffff8111a9d5>] clear_page_dirty_for_io+0x1e/0x90
  [152154.036424]  RSP <ffff880429effc68>
  [152154.242621] ---[ end trace e3d3376b23a57041 ]---

Fix this by returning the error EOPNOTSUPP if an attempt to copy an
inline extent into a non-zero offset happens, just like what is done for
other scenarios that would require copying/splitting inline extents,
which were introduced by the following commits:

   00fdf13a2e9f ("Btrfs: fix a crash of clone with inline extents's split")
   3f9e3df8da3c ("btrfs: replace error code from btrfs_drop_extents")

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agoBtrfs: fix list transaction->pending_ordered corruption
Filipe Manana [Fri, 3 Jul 2015 19:30:34 +0000 (20:30 +0100)]
Btrfs: fix list transaction->pending_ordered corruption

commit d3efe08400317888f559bbedf0e42cd31575d0ef upstream.

When we call btrfs_commit_transaction(), we splice the list "ordered"
of our transaction handle into the transaction's "pending_ordered"
list, but we don't re-initialize the "ordered" list of our transaction
handle, this means it still points to the same elements it used to
before the splice. Then we check if the current transaction's state is
>= TRANS_STATE_COMMIT_START and if it is we end up calling
btrfs_end_transaction() which simply splices again the "ordered" list
of our handle into the transaction's "pending_ordered" list, leaving
multiple pointers to the same ordered extents which results in list
corruption when we are iterating, removing and freeing ordered extents
at btrfs_wait_pending_ordered(), resulting in access to dangling
pointers / use-after-free issues.
Similarly, btrfs_end_transaction() can end up in some cases calling
btrfs_commit_transaction(), and both did a list splice of the transaction
handle's "ordered" list into the transaction's "pending_ordered" without
re-initializing the handle's "ordered" list, resulting in exactly the
same problem.

This produces the following warning on a kernel with linked list
debugging enabled:

[109749.265416] ------------[ cut here ]------------
[109749.266410] WARNING: CPU: 7 PID: 324 at lib/list_debug.c:59 __list_del_entry+0x5a/0x98()
[109749.267969] list_del corruption. prev->next should be ffff8800ba087e20, but was fffffff8c1f7c35d
(...)
[109749.287505] Call Trace:
[109749.288135]  [<ffffffff8145f077>] dump_stack+0x4f/0x7b
[109749.298080]  [<ffffffff81095de5>] ? console_unlock+0x356/0x3a2
[109749.331605]  [<ffffffff8104b3b0>] warn_slowpath_common+0xa1/0xbb
[109749.334849]  [<ffffffff81260642>] ? __list_del_entry+0x5a/0x98
[109749.337093]  [<ffffffff8104b410>] warn_slowpath_fmt+0x46/0x48
[109749.337847]  [<ffffffff81260642>] __list_del_entry+0x5a/0x98
[109749.338678]  [<ffffffffa053e8bf>] btrfs_wait_pending_ordered+0x46/0xdb [btrfs]
[109749.340145]  [<ffffffffa058a65f>] ? __btrfs_run_delayed_items+0x149/0x163 [btrfs]
[109749.348313]  [<ffffffffa054077d>] btrfs_commit_transaction+0x36b/0xa10 [btrfs]
[109749.349745]  [<ffffffff81087310>] ? trace_hardirqs_on+0xd/0xf
[109749.350819]  [<ffffffffa055370d>] btrfs_sync_file+0x36f/0x3fc [btrfs]
[109749.351976]  [<ffffffff8118ec98>] vfs_fsync_range+0x8f/0x9e
[109749.360341]  [<ffffffff8118ecc3>] vfs_fsync+0x1c/0x1e
[109749.368828]  [<ffffffff8118ee1d>] do_fsync+0x34/0x4e
[109749.369790]  [<ffffffff8118f045>] SyS_fsync+0x10/0x14
[109749.370925]  [<ffffffff81465197>] system_call_fastpath+0x12/0x6f
[109749.382274] ---[ end trace 48e0d07f7c03d95a ]---

On a non-debug kernel this leads to invalid memory accesses, causing a
crash. Fix this by using list_splice_init() instead of list_splice() in
btrfs_commit_transaction() and btrfs_end_transaction().

Fixes: 50d9aa99bd35 ("Btrfs: make sure logged extents complete in the current transaction V3"
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agoBtrfs: fix memory leak in the extent_same ioctl
Filipe Manana [Fri, 3 Jul 2015 07:36:11 +0000 (08:36 +0100)]
Btrfs: fix memory leak in the extent_same ioctl

commit 497b4050e0eacd4c746dd396d14916b1e669849d upstream.

We were allocating memory with memdup_user() but we were never releasing
that memory. This affected pretty much every call to the ioctl, whether
it deduplicated extents or not.

This issue was reported on IRC by Julian Taylor and on the mailing list
by Marcel Ritter, credit goes to them for finding the issue.

Reported-by: Julian Taylor <jtaylor.debian@googlemail.com>
Reported-by: Marcel Ritter <ritter.marcel@gmail.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: Mark Fasheh <mfasheh@suse.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agoBtrfs: fix fsync data loss after append write
Filipe Manana [Wed, 17 Jun 2015 11:49:23 +0000 (12:49 +0100)]
Btrfs: fix fsync data loss after append write

commit e4545de5b035c7debb73d260c78377dbb69cbfb5 upstream.

If we do an append write to a file (which increases its inode's i_size)
that does not have the flag BTRFS_INODE_NEEDS_FULL_SYNC set in its inode,
and the previous transaction added a new hard link to the file, which sets
the flag BTRFS_INODE_COPY_EVERYTHING in the file's inode, and then fsync
the file, the inode's new i_size isn't logged. This has the consequence
that after the fsync log is replayed, the file size remains what it was
before the append write operation, which means users/applications will
not be able to read the data that was successsfully fsync'ed before.

This happens because neither the inode item nor the delayed inode get
their i_size updated when the append write is made - doing so would
require starting a transaction in the buffered write path, something that
we do not do intentionally for performance reasons.

Fix this by making sure that when the flag BTRFS_INODE_COPY_EVERYTHING is
set the inode is logged with its current i_size (log the in-memory inode
into the log tree).

This issue is not a recent regression and is easy to reproduce with the
following test case for fstests:

  seq=`basename $0`
  seqres=$RESULT_DIR/$seq
  echo "QA output created by $seq"

  here=`pwd`
  tmp=/tmp/$$
  status=1 # failure is the default!

  _cleanup()
  {
          _cleanup_flakey
          rm -f $tmp.*
  }
  trap "_cleanup; exit \$status" 0 1 2 3 15

  # get standard environment, filters and checks
  . ./common/rc
  . ./common/filter
  . ./common/dmflakey

  # real QA test starts here
  _supported_fs generic
  _supported_os Linux
  _need_to_be_root
  _require_scratch
  _require_dm_flakey
  _require_metadata_journaling $SCRATCH_DEV

  _crash_and_mount()
  {
          # Simulate a crash/power loss.
          _load_flakey_table $FLAKEY_DROP_WRITES
          _unmount_flakey
          # Allow writes again and mount. This makes the fs replay its fsync log.
          _load_flakey_table $FLAKEY_ALLOW_WRITES
          _mount_flakey
  }

  rm -f $seqres.full

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

  # Create the test file with some initial data and then fsync it.
  # The fsync here is only needed to trigger the issue in btrfs, as it causes the
  # the flag BTRFS_INODE_NEEDS_FULL_SYNC to be removed from the btrfs inode.
  $XFS_IO_PROG -f -c "pwrite -S 0xaa 0 32k" \
                  -c "fsync" \
                  $SCRATCH_MNT/foo | _filter_xfs_io
  sync

  # Add a hard link to our file.
  # On btrfs this sets the flag BTRFS_INODE_COPY_EVERYTHING on the btrfs inode,
  # which is a necessary condition to trigger the issue.
  ln $SCRATCH_MNT/foo $SCRATCH_MNT/bar

  # Sync the filesystem to force a commit of the current btrfs transaction, this
  # is a necessary condition to trigger the bug on btrfs.
  sync

  # Now append more data to our file, increasing its size, and fsync the file.
  # In btrfs because the inode flag BTRFS_INODE_COPY_EVERYTHING was set and the
  # write path did not update the inode item in the btree nor the delayed inode
  # item (in memory struture) in the current transaction (created by the fsync
  # handler), the fsync did not record the inode's new i_size in the fsync
  # log/journal. This made the data unavailable after the fsync log/journal is
  # replayed.
  $XFS_IO_PROG -c "pwrite -S 0xbb 32K 32K" \
               -c "fsync" \
               $SCRATCH_MNT/foo | _filter_xfs_io

  echo "File content after fsync and before crash:"
  od -t x1 $SCRATCH_MNT/foo

  _crash_and_mount

  echo "File content after crash and log replay:"
  od -t x1 $SCRATCH_MNT/foo

  status=0
  exit

The expected file output before and after the crash/power failure expects the
appended data to be available, which is:

  0000000 aa aa aa aa aa aa aa aa aa aa aa aa aa aa aa aa
  *
  0100000 bb bb bb bb bb bb bb bb bb bb bb bb bb bb bb bb
  *
  0200000

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Chris Mason <clm@fb.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agoBtrfs: fix race between caching kthread and returning inode to inode cache
Filipe Manana [Sat, 13 Jun 2015 05:52:57 +0000 (06:52 +0100)]
Btrfs: fix race between caching kthread and returning inode to inode cache

commit ae9d8f17118551bedd797406a6768b87c2146234 upstream.

While the inode cache caching kthread is calling btrfs_unpin_free_ino(),
we could have a concurrent call to btrfs_return_ino() that adds a new
entry to the root's free space cache of pinned inodes. This concurrent
call does not acquire the fs_info->commit_root_sem before adding a new
entry if the caching state is BTRFS_CACHE_FINISHED, which is a problem
because the caching kthread calls btrfs_unpin_free_ino() after setting
the caching state to BTRFS_CACHE_FINISHED and therefore races with
the task calling btrfs_return_ino(), which is adding a new entry, while
the former (caching kthread) is navigating the cache's rbtree, removing
and freeing nodes from the cache's rbtree without acquiring the spinlock
that protects the rbtree.

This race resulted in memory corruption due to double free of struct
btrfs_free_space objects because both tasks can end up doing freeing the
same objects. Note that adding a new entry can result in merging it with
other entries in the cache, in which case those entries are freed.
This is particularly important as btrfs_free_space structures are also
used for the block group free space caches.

This memory corruption can be detected by a debugging kernel, which
reports it with the following trace:

[132408.501148] slab error in verify_redzone_free(): cache `btrfs_free_space': double free detected
[132408.505075] CPU: 15 PID: 12248 Comm: btrfs-ino-cache Tainted: G        W       4.1.0-rc5-btrfs-next-10+ #1
[132408.505075] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.8.1-0-g4adadbd-20150316_085822-nilsson.home.kraxel.org 04/01/2014
[132408.505075]  ffff880023e7d320 ffff880163d73cd8 ffffffff8145eec7 ffffffff81095dce
[132408.505075]  ffff880009735d40 ffff880163d73ce8 ffffffff81154e1e ffff880163d73d68
[132408.505075]  ffffffff81155733 ffffffffa054a95a ffff8801b6099f00 ffffffffa0505b5f
[132408.505075] Call Trace:
[132408.505075]  [<ffffffff8145eec7>] dump_stack+0x4f/0x7b
[132408.505075]  [<ffffffff81095dce>] ? console_unlock+0x356/0x3a2
[132408.505075]  [<ffffffff81154e1e>] __slab_error.isra.28+0x25/0x36
[132408.505075]  [<ffffffff81155733>] __cache_free+0xe2/0x4b6
[132408.505075]  [<ffffffffa054a95a>] ? __btrfs_add_free_space+0x2f0/0x343 [btrfs]
[132408.505075]  [<ffffffffa0505b5f>] ? btrfs_unpin_free_ino+0x8e/0x99 [btrfs]
[132408.505075]  [<ffffffff810f3b30>] ? time_hardirqs_off+0x15/0x28
[132408.505075]  [<ffffffff81084d42>] ? trace_hardirqs_off+0xd/0xf
[132408.505075]  [<ffffffff811563a1>] ? kfree+0xb6/0x14e
[132408.505075]  [<ffffffff811563d0>] kfree+0xe5/0x14e
[132408.505075]  [<ffffffffa0505b5f>] btrfs_unpin_free_ino+0x8e/0x99 [btrfs]
[132408.505075]  [<ffffffffa0505e08>] caching_kthread+0x29e/0x2d9 [btrfs]
[132408.505075]  [<ffffffffa0505b6a>] ? btrfs_unpin_free_ino+0x99/0x99 [btrfs]
[132408.505075]  [<ffffffff8106698f>] kthread+0xef/0xf7
[132408.505075]  [<ffffffff810f3b08>] ? time_hardirqs_on+0x15/0x28
[132408.505075]  [<ffffffff810668a0>] ? __kthread_parkme+0xad/0xad
[132408.505075]  [<ffffffff814653d2>] ret_from_fork+0x42/0x70
[132408.505075]  [<ffffffff810668a0>] ? __kthread_parkme+0xad/0xad
[132408.505075] ffff880023e7d320: redzone 1:0x9f911029d74e35b, redzone 2:0x9f911029d74e35b.
[132409.501654] slab: double free detected in cache 'btrfs_free_space', objp ffff880023e7d320
[132409.503355] ------------[ cut here ]------------
[132409.504241] kernel BUG at mm/slab.c:2571!

Therefore fix this by having btrfs_unpin_free_ino() acquire the lock
that protects the rbtree while doing the searches and removing entries.

Fixes: 1c70d8fb4dfa ("Btrfs: fix inode caching vs tree log")
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agoBtrfs: use kmem_cache_free when freeing entry in inode cache
Filipe Manana [Sat, 13 Jun 2015 05:52:56 +0000 (06:52 +0100)]
Btrfs: use kmem_cache_free when freeing entry in inode cache

commit c3f4a1685bb87e59c886ee68f7967eae07d4dffa upstream.

The free space entries are allocated using kmem_cache_zalloc(),
through __btrfs_add_free_space(), therefore we should use
kmem_cache_free() and not kfree() to avoid any confusion and
any potential problem. Looking at the kfree() definition at
mm/slab.c it has the following comment:

  /*
   * (...)
   *
   * Don't free memory not originally allocated by kmalloc()
   * or you will run into trouble.
   */

So better be safe and use kmem_cache_free().

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.cz>
Signed-off-by: Chris Mason <clm@fb.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agomd: fix a build warning
Firo Yang [Thu, 11 Jun 2015 01:41:10 +0000 (09:41 +0800)]
md: fix a build warning

commit 4e023612325a9034a542bfab79f78b1fe5ebb841 upstream.

Warning like this:

drivers/md/md.c: In function "update_array_info":
drivers/md/md.c:6394:26: warning: logical not is only applied
to the left hand side of comparison [-Wlogical-not-parentheses]
      !mddev->persistent  != info->not_persistent||

Fix it as Neil Brown said:
mddev->persistent != !info->not_persistent ||

Signed-off-by: Firo Yang <firogm@gmail.com>
Signed-off-by: NeilBrown <neilb@suse.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agoBtrfs: don't invalidate root dentry when subvolume deletion fails
Omar Sandoval [Wed, 3 Jun 2015 00:31:00 +0000 (17:31 -0700)]
Btrfs: don't invalidate root dentry when subvolume deletion fails

commit 64ad6c488975d7516230cf7849190a991fd615ae upstream.

Since commit bafc9b754f75 ("vfs: More precise tests in d_invalidate"),
mounted subvolumes can be deleted because d_invalidate() won't fail.
However, we run into problems when we attempt to delete the default
subvolume while it is mounted as the root filesystem:

# btrfs subvol list /
ID 257 gen 306 top level 5 path rootvol
ID 267 gen 334 top level 5 path snap1
# btrfs subvol get-default /
ID 267 gen 334 top level 5 path snap1
# btrfs inspect-internal rootid /
267
# mount -o subvol=/ /dev/vda1 /mnt
# btrfs subvol del /mnt/snap1
Delete subvolume (no-commit): '/mnt/snap1'
ERROR: cannot delete '/mnt/snap1' - Operation not permitted
# findmnt /
findmnt: can't read /proc/mounts: No such file or directory
# ls /proc
#

Markus reported that this same scenario simply led to a kernel oops.

This happens because in btrfs_ioctl_snap_destroy(), we call
d_invalidate() before we check may_destroy_subvol(), which means that we
detach the submounts and drop the dentry before erroring out. Instead,
we should only invalidate the dentry once the deletion has succeeded.
Additionally, the shrink_dcache_sb() isn't necessary; d_invalidate()
will prune the dcache for the deleted subvolume.

Fixes: bafc9b754f75 ("vfs: More precise tests in d_invalidate")
Reported-by: Markus Schauler <mschauler@gmail.com>
Signed-off-by: Omar Sandoval <osandov@osandov.com>
Signed-off-by: Chris Mason <clm@fb.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agoARM: dts: mx23: fix iio-hwmon support
Stefan Wahren [Tue, 2 Jun 2015 22:03:28 +0000 (22:03 +0000)]
ARM: dts: mx23: fix iio-hwmon support

commit e8e94ed6285428ab780cd7b0df4622f71eceb39e upstream.

In order to get iio-hwmon support, the lradc must be declared as an
iio provider. So fix this issue by adding the #io-channel-cells property.

Signed-off-by: Stefan Wahren <stefan.wahren@i2se.com>
Fixes: bd798f9c7b30 ("ARM: dts: mxs: Add iio-hwmon to mx23 soc")
Reviewed-by: Marek Vasut <marex@denx.de>
Reviewed-by: Alexandre Belloni <alexandre.belloni@free-electrons.com>
Signed-off-by: Shawn Guo <shawn.guo@linaro.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agohwmon: (nct7802) fix visibility of temp3
Constantine Shulyupin [Fri, 26 Jun 2015 14:47:44 +0000 (17:47 +0300)]
hwmon: (nct7802) fix visibility of temp3

commit 56172d81a9bc37a69b95dd627b8d48135c9c7b31 upstream.

Excerpt from datasheet:
7.2.32 Mode Selection Register
RTD3_MD : 00=Closed , 01=Reserved , 10=Thermistor mode , 11=Voltage sense

Show temp3 only in Thermistor mode

Signed-off-by: Constantine Shulyupin <const@MakeLinux.com>
Signed-off-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agohwmon: (mcp3021) Fix broken output scaling
Stevens, Nick [Wed, 1 Jul 2015 16:07:41 +0000 (16:07 +0000)]
hwmon: (mcp3021) Fix broken output scaling

commit 347d7e45bd09ce09cbc30d5cea9de377eb22f55c upstream.

The mcp3021 scaling code is dividing the VDD (full-scale) value in
millivolts by the A2D resolution to obtain the scaling factor. When VDD
is 3300mV (the standard value) and the resolution is 12-bit (4096
divisions), the result is a scale factor of 3300/4096, which is always
one.  Effectively, the raw A2D reading is always being returned because
no scaling is applied.

This patch fixes the issue and simplifies the register-to-volts
calculation, removing the unneeded "output_scale" struct member.

Signed-off-by: Nick Stevens <Nick.Stevens@digi.com>
[Guenter Roeck: Dropped unnecessary value check]
Signed-off-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agomd: Skip cluster setup for dm-raid
Goldwyn Rodrigues [Wed, 22 Jul 2015 17:09:17 +0000 (12:09 -0500)]
md: Skip cluster setup for dm-raid

commit d3b178adb3a3adf54ecf77758138b654c3ee7f09 upstream.

There is a bug that the bitmap superblock isn't initialised properly for
dm-raid, so a new field can have garbage in new fields.
(dm-raid does initialisation in the kernel - md initialised the
 superblock in mdadm).

This means that for dm-raid we cannot currently trust the new ->nodes
field. So:
 - use __GFP_ZERO to initialise the superblock properly for all new
    arrays
 - initialise all fields in bitmap_info in bitmap_new_disk_sb
 - ignore ->nodes for dm arrays (yes, this is a hack)

This bug exposes dm-raid to bug in the (still experimental) md-cluster
code, so it is suitable for -stable.  It does cause crashes.

References: https://bugzilla.kernel.org/show_bug.cgi?id=100491
Signed-off-By: Goldwyn Rodrigues <rgoldwyn@suse.com>
Signed-off-by: NeilBrown <neilb@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agomd: unlock mddev_lock on an error path.
NeilBrown [Thu, 25 Jun 2015 07:06:40 +0000 (17:06 +1000)]
md: unlock mddev_lock on an error path.

commit 9a8c0fa861e4db60409b4dda254cef5e17e4d43c upstream.

This error path retuns while still holding the lock - bad.

Fixes: 6791875e2e53 ("md: make reconfig_mutex optional for writes to md sysfs files.")
Signed-off-by: NeilBrown <neilb@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agomd: clear mddev->private when it has been freed.
NeilBrown [Thu, 25 Jun 2015 07:01:40 +0000 (17:01 +1000)]
md: clear mddev->private when it has been freed.

commit bd6919228d7e1867ae9e24ab27e3e4a366c87d21 upstream.

If ->private is set when ->run is called, it is assumed to be
a 'config'  prepared as part of 'reshape'.

So it is important when we free that config, that we also clear ->private.
This is not often a problem as the mddev will normally be discarded
shortly after the config us freed.
However if an 'assemble' races with a final close, the assemble can use
the old mddev which has a stale ->private.  This leads to any of
various sorts of crashes.

So clear ->private after calling ->free().

Reported-by: Nate Clark <nate@neworld.us>
Fixes: afa0f557cb15 ("md: rename ->stop to ->free")
Signed-off-by: NeilBrown <neilb@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agodmaengine: mv_xor: bug fix for racing condition in descriptors cleanup
Lior Amsalem [Tue, 26 May 2015 13:07:32 +0000 (15:07 +0200)]
dmaengine: mv_xor: bug fix for racing condition in descriptors cleanup

commit 9136291f1dbc1d4d1cacd2840fb35f4f3ce16c46 upstream.

This patch fixes a bug in the XOR driver where the cleanup function can be
called and free descriptors that never been processed by the engine (which
result in data errors).

The cleanup function will free descriptors based on the ownership bit in
the descriptors.

Fixes: ff7b04796d98 ("dmaengine: DMA engine driver for Marvell XOR engine")
Signed-off-by: Lior Amsalem <alior@marvell.com>
Signed-off-by: Maxime Ripard <maxime.ripard@free-electrons.com>
Reviewed-by: Ofer Heifetz <oferh@marvell.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agotracing: Fix sample output of dynamic arrays
Steven Rostedt (Red Hat) [Fri, 17 Jul 2015 18:03:26 +0000 (14:03 -0400)]
tracing: Fix sample output of dynamic arrays

commit d6726c8145290bef950ae2538ea6ae1d96a1944b upstream.

He Kuang noticed that the trace event samples for arrays was broken:

"The output result of trace_foo_bar event in traceevent samples is
 wrong. This problem can be reproduced as following:

  (Build kernel with SAMPLE_TRACE_EVENTS=m)

  $ insmod trace-events-sample.ko

  $ echo 1 > /sys/kernel/debug/tracing/events/sample-trace/foo_bar/enable

  $ cat /sys/kernel/debug/tracing/trace

  event-sample-980 [000] ....  43.649559: foo_bar: foo hello 21 0x15
  BIT1|BIT3|0x10 {0x1,0x6f6f6e53,0xff007970,0xffffffff} Snoopy
                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                 The array length is not right, should be {0x1}.
  (ffffffff,ffffffff)

  event-sample-980 [000] ....  44.653827: foo_bar: foo hello 22 0x16
  BIT2|BIT3|0x10
  {0x1,0x2,0x646e6147,0x666c61,0xffffffff,0xffffffff,0x750aeffe,0x7}
  ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                 The array length is not right, should be {0x1,0x2}.
  Gandalf (ffffffff,ffffffff)"

This was caused by an update to have __print_array()'s second parameter
be the count of items in the array and not the size of the array.

As there is already users of __print_array(), it can not change. But
the sample code can and we can also improve on the documentation about
__print_array() and __get_dynamic_array_len().

Link: http://lkml.kernel.org/r/1436839171-31527-2-git-send-email-hekuang@huawei.com
Fixes: ac01ce1410fc2 ("tracing: Make ftrace_print_array_seq compute buf_len")
Reported-by: He Kuang <hekuang@huawei.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agotracing: Have branch tracer use recursive field of task struct
Steven Rostedt (Red Hat) [Tue, 7 Jul 2015 19:05:03 +0000 (15:05 -0400)]
tracing: Have branch tracer use recursive field of task struct

commit 6224beb12e190ff11f3c7d4bf50cb2922878f600 upstream.

Fengguang Wu's tests triggered a bug in the branch tracer's start up
test when CONFIG_DEBUG_PREEMPT set. This was because that config
adds some debug logic in the per cpu field, which calls back into
the branch tracer.

The branch tracer has its own recursive checks, but uses a per cpu
variable to implement it. If retrieving the per cpu variable calls
back into the branch tracer, you can see how things will break.

Instead of using a per cpu variable, use the trace_recursion field
of the current task struct. Simply set a bit when entering the
branch tracing and clear it when leaving. If the bit is set on
entry, just don't do the tracing.

There's also the case with lockdep, as the local_irq_save() called
before the recursion can also trigger code that can call back into
the function. Changing that to a raw_local_irq_save() will protect
that as well.

This prevents the recursion and the inevitable crash that follows.

Link: http://lkml.kernel.org/r/20150630141803.GA28071@wfg-t540p.sh.intel.com
Reported-by: Fengguang Wu <fengguang.wu@intel.com>
Tested-by: Fengguang Wu <fengguang.wu@intel.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agotracing: Fix typo from "static inlin" to "static inline"
Steven Rostedt (Red Hat) [Thu, 25 Jun 2015 22:19:37 +0000 (18:19 -0400)]
tracing: Fix typo from "static inlin" to "static inline"

commit cc9e4bde03f2b4cfba52406c021364cbd2a4a0f3 upstream.

The trace.h header when called without CONFIG_EVENT_TRACING enabled
(seldom done), will not compile because of a typo in the protocol
of trace_event_enum_update().

Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agotracing/filter: Do not allow infix to exceed end of string
Steven Rostedt (Red Hat) [Thu, 25 Jun 2015 22:10:09 +0000 (18:10 -0400)]
tracing/filter: Do not allow infix to exceed end of string

commit 6b88f44e161b9ee2a803e5b2b1fbcf4e20e8b980 upstream.

While debugging a WARN_ON() for filtering, I found that it is possible
for the filter string to be referenced after its end. With the filter:

 # echo '>' > /sys/kernel/debug/events/ext4/ext4_truncate_exit/filter

The filter_parse() function can call infix_get_op() which calls
infix_advance() that updates the infix filter pointers for the cnt
and tail without checking if the filter is already at the end, which
will put the cnt to zero and the tail beyond the end. The loop then calls
infix_next() that has

ps->infix.cnt--;
return ps->infix.string[ps->infix.tail++];

The cnt will now be below zero, and the tail that is returned is
already passed the end of the filter string. So far the allocation
of the filter string usually has some buffer that is zeroed out, but
if the filter string is of the exact size of the allocated buffer
there's no guarantee that the charater after the nul terminating
character will be zero.

Luckily, only root can write to the filter.

Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agotracing/filter: Do not WARN on operand count going below zero
Steven Rostedt (Red Hat) [Thu, 25 Jun 2015 22:02:29 +0000 (18:02 -0400)]
tracing/filter: Do not WARN on operand count going below zero

commit b4875bbe7e68f139bd3383828ae8e994a0df6d28 upstream.

When testing the fix for the trace filter, I could not come up with
a scenario where the operand count goes below zero, so I added a
WARN_ON_ONCE(cnt < 0) to the logic. But there is legitimate case
that it can happen (although the filter would be wrong).

 # echo '>' > /sys/kernel/debug/events/ext4/ext4_truncate_exit/filter

That is, a single operation without any operands will hit the path
where the WARN_ON_ONCE() can trigger. Although this is harmless,
and the filter is reported as a error. But instead of spitting out
a warning to the kernel dmesg, just fail nicely and report it via
the proper channels.

Link: http://lkml.kernel.org/r/558C6082.90608@oracle.com
Reported-by: Vince Weaver <vincent.weaver@maine.edu>
Reported-by: Sasha Levin <sasha.levin@oracle.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agoima: update builtin policies
Mimi Zohar [Fri, 12 Jun 2015 00:48:33 +0000 (20:48 -0400)]
ima: update builtin policies

commit 24fd03c87695a76f0517df42a37e51b1597d2c8a upstream.

This patch defines a builtin measurement policy "tcb", similar to the
existing "ima_tcb", but with additional rules to also measure files
based on the effective uid and to measure files opened with the "read"
mode bit set (eg. read, read-write).

Changing the builtin "ima_tcb" policy could potentially break existing
users.  Instead of defining a new separate boot command line option each
time the builtin measurement policy is modified, this patch defines a
single generic boot command line option "ima_policy=" to specify the
builtin policy and deprecates the use of the builtin ima_tcb policy.

[The "ima_policy=" boot command line option is based on Roberto Sassu's
"ima: added new policy type exec" patch.]

Signed-off-by: Mimi Zohar <zohar@linux.vnet.ibm.com>
Signed-off-by: Dr. Greg Wettstein <gw@idfusion.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agoima: extend "mask" policy matching support
Mimi Zohar [Wed, 5 Nov 2014 12:53:55 +0000 (07:53 -0500)]
ima: extend "mask" policy matching support

commit 4351c294b8c1028077280f761e158d167b592974 upstream.

The current "mask" policy option matches files opened as MAY_READ,
MAY_WRITE, MAY_APPEND or MAY_EXEC.  This patch extends the "mask"
option to match files opened containing one of these modes.  For
example, "mask=^MAY_READ" would match files opened read-write.

Signed-off-by: Mimi Zohar <zohar@linux.vnet.ibm.com>
Signed-off-by: Dr. Greg Wettstein <gw@idfusion.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agoima: add support for new "euid" policy condition
Mimi Zohar [Wed, 5 Nov 2014 12:48:36 +0000 (07:48 -0500)]
ima: add support for new "euid" policy condition

commit 139069eff7388407f19794384c42a534d618ccd7 upstream.

The new "euid" policy condition measures files with the specified
effective uid (euid).  In addition, for CAP_SETUID files it measures
files with the specified uid or suid.

Changelog:
- fixed checkpatch.pl warnings
- fixed avc denied {setuid} messages - based on Roberto's feedback

Signed-off-by: Mimi Zohar <zohar@linux.vnet.ibm.com>
Signed-off-by: Dr. Greg Wettstein <gw@idfusion.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agoima: fix ima_show_template_data_ascii()
Mimi Zohar [Thu, 11 Jun 2015 15:54:42 +0000 (11:54 -0400)]
ima: fix ima_show_template_data_ascii()

commit 45b26133b97871896b8c5241d59f4ff7839db7b2 upstream.

This patch fixes a bug introduced in "4d7aeee ima: define new template
ima-ng and template fields d-ng and n-ng".

Changelog:
- change int to uint32 (Roberto Sassu's suggestion)

Signed-off-by: Mimi Zohar <zohar@linux.vnet.ibm.com>
Signed-off-by: Roberto Sassu <rsassu@suse.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agoevm: labeling pseudo filesystems exception
Mimi Zohar [Tue, 21 Apr 2015 17:59:31 +0000 (13:59 -0400)]
evm: labeling pseudo filesystems exception

commit 5101a1850bb7ccbf107929dee9af0cd2f400940f upstream.

To prevent offline stripping of existing file xattrs and relabeling of
them at runtime, EVM allows only newly created files to be labeled.  As
pseudo filesystems are not persistent, stripping of xattrs is not a
concern.

Some LSMs defer file labeling on pseudo filesystems.  This patch
permits the labeling of existing files on pseudo files systems.

Signed-off-by: Mimi Zohar <zohar@linux.vnet.ibm.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agoima: do not measure or appraise the NSFS filesystem
Mimi Zohar [Tue, 21 Apr 2015 20:54:24 +0000 (16:54 -0400)]
ima: do not measure or appraise the NSFS filesystem

commit cd025f7f94108995383edddfb61fc8afea6c66a9 upstream.

Include don't appraise or measure rules for the NSFS filesystem
in the builtin ima_tcb and ima_appraise_tcb policies.

Changelog:
- Update documentation

Signed-off-by: Mimi Zohar <zohar@linux.vnet.ibm.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agoima: cleanup ima_init_policy() a little
Dan Carpenter [Tue, 7 Apr 2015 09:22:11 +0000 (12:22 +0300)]
ima: cleanup ima_init_policy() a little

commit 5577857f8e26e9027271f10daf96361640907300 upstream.

It's a bit easier to read this if we split it up into two for loops.

Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Mimi Zohar <zohar@linux.vnet.ibm.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agoima: skip measurement of cgroupfs files and update documentation
Roberto Sassu [Sat, 11 Apr 2015 15:13:06 +0000 (17:13 +0200)]
ima: skip measurement of cgroupfs files and update documentation

commit 6438de9f3fb5180d78a0422695d0b88c687757d3 upstream.

This patch adds a rule in the default measurement policy to skip inodes
in the cgroupfs filesystem. Measurements for this filesystem can be
avoided, as all the digests collected have the same value of the digest of
an empty file.

Furthermore, this patch updates the documentation of IMA policies in
Documentation/ABI/testing/ima_policy to make it consistent with
the policies set in security/integrity/ima/ima_policy.c.

Signed-off-by: Roberto Sassu <rsassu@suse.de>
Signed-off-by: Mimi Zohar <zohar@linux.vnet.ibm.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agoKEYS: ensure we free the assoc array edit if edit is valid
Colin Ian King [Mon, 27 Jul 2015 14:23:43 +0000 (15:23 +0100)]
KEYS: ensure we free the assoc array edit if edit is valid

commit ca4da5dd1f99fe9c59f1709fb43e818b18ad20e0 upstream.

__key_link_end is not freeing the associated array edit structure
and this leads to a 512 byte memory leak each time an identical
existing key is added with add_key().

The reason the add_key() system call returns okay is that
key_create_or_update() calls __key_link_begin() before checking to see
whether it can update a key directly rather than adding/replacing - which
it turns out it can.  Thus __key_link() is not called through
__key_instantiate_and_link() and __key_link_end() must cancel the edit.

CVE-2015-1333

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: James Morris <james.l.morris@oracle.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agoKEYS: fix "ca_keys=" partial key matching
Mimi Zohar [Wed, 11 Feb 2015 12:33:34 +0000 (07:33 -0500)]
KEYS: fix "ca_keys=" partial key matching

commit f2b3dee484f9cee967a54ef05a66866282337519 upstream.

The call to asymmetric_key_hex_to_key_id() from ca_keys_setup()
silently fails with -ENOMEM.  Instead of dynamically allocating
memory from a __setup function, this patch defines a variable
and calls __asymmetric_key_hex_to_key_id(), a new helper function,
directly.

This bug was introduced by 'commit 46963b774d44 ("KEYS: Overhaul
key identification when searching for asymmetric keys")'.

Changelog:
- for clarification, rename hexlen to asciihexlen in
  asymmetric_key_hex_to_key_id()
- add size argument to __asymmetric_key_hex_to_key_id() - David Howells
- inline __asymmetric_key_hex_to_key_id() - David Howells
- remove duplicate strlen() calls

Acked-by: David Howells <dhowells@redhat.com>
Signed-off-by: Mimi Zohar <zohar@linux.vnet.ibm.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agotpm, tpm_crb: fail when TPM2 ACPI table contents look corrupted
Jarkko Sakkinen [Wed, 24 Jun 2015 14:14:55 +0000 (17:14 +0300)]
tpm, tpm_crb: fail when TPM2 ACPI table contents look corrupted

commit b371616b8537d6450ebca0819defbf53452bebf3 upstream.

At least some versions of AMI BIOS have corrupted contents in the TPM2
ACPI table and namely the physical address of the control area is set to
zero.

This patch changes the driver to fail gracefully  when we observe a zero
address instead of continuing to ioremap.

Signed-off-by: Jarkko Sakkinen <jarkko.sakkinen@linux.intel.com>
Reviewed-by: Peter Huewe <peterhuewe@gmx.de>
Signed-off-by: Peter Huewe <peterhuewe@gmx.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agotpm: Fix initialization of the cdev
Jason Gunthorpe [Tue, 30 Jun 2015 19:15:31 +0000 (13:15 -0600)]
tpm: Fix initialization of the cdev

commit ba0ef85479c46a2ab354c2220bdb6152f7f4baf3 upstream.

When a cdev is contained in a dynamic structure the cdev parent kobj
should be set to the kobj that controls the lifetime of the enclosing
structure. In TPM's case this is the embedded struct device.

Also, cdev_init 0's the whole structure, so all sets must be after,
not before. This fixes module ref counting and cdev.

Fixes: 313d21eeab92 ("tpm: device class for tpm")
Signed-off-by: Jason Gunthorpe <jgunthorpe@obsidianresearch.com>
Reviewed-by: Dmitry Torokhov <dmitry.torokhov@gmail.com>
Reviewed-by: Jarkko Sakkinen <jarkko.sakkinen@linux.intel.com>
Tested-by: Jarkko Sakkinen <jarkko.sakkinen@linux.intel.com>
Signed-off-by: Peter Huewe <peterhuewe@gmx.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agovTPM: set virtual device before passing to ibmvtpm_reset_crq
Hon Ching \\(Vicky\\) Lo [Fri, 22 May 2015 17:23:02 +0000 (13:23 -0400)]
vTPM: set virtual device before passing to ibmvtpm_reset_crq

commit 9d75f08946e8485109458ccf16f714697c207f41 upstream.

tpm_ibmvtpm_probe() calls ibmvtpm_reset_crq(ibmvtpm) without having yet
set the virtual device in the ibmvtpm structure. So in ibmvtpm_reset_crq,
the phype call contains empty unit addresses, ibmvtpm->vdev->unit_address.

Signed-off-by: Hon Ching(Vicky) Lo <honclo@linux.vnet.ibm.com>
Signed-off-by: Joy Latten <jmlatten@linux.vnet.ibm.com>
Reviewed-by: Ashley Lai <ashley@ahsleylai.com>
Fixes: 132f76294744 ("drivers/char/tpm: Add new device driver to support IBM vTPM")
Signed-off-by: Peter Huewe <peterhuewe@gmx.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agotpm, tpm_crb: fix le64_to_cpu conversions in crb_acpi_add()
Jarkko Sakkinen [Tue, 9 Jun 2015 12:07:59 +0000 (15:07 +0300)]
tpm, tpm_crb: fix le64_to_cpu conversions in crb_acpi_add()

commit 49afd7289bd937401c5f7faa193054bc3c41dad6 upstream.

le64_to_cpu() was applied twice to the physical addresses read from the
control area. This hasn't shown any visible regressions because CRB
driver has been tested only on the little endian platofrms so far.

Reported-by: Matt Fleming <matt.fleming@intel.com>
Signed-off-by: Jarkko Sakkinen <jarkko.sakkinen@linux.intel.com>
Reviewed-By: Jason Gunthorpe <jgunthorpe@obsidianresearch.com>
Fixes: 30fc8d138e91 ("tpm: TPM 2.0 CRB Interface")
Signed-off-by: Peter Huewe <peterhuewe@gmx.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agow1_therm reference count family data
David Fries [Sat, 9 May 2015 00:51:50 +0000 (19:51 -0500)]
w1_therm reference count family data

commit f7134eea05b2fb4a2c0935f8a540539fff01f3eb upstream.

A temperature conversion can take 750 ms and when possible the
w1_therm slave driver drops the bus_mutex to allow other bus
operations, but that includes operations such as a periodic slave
search, which can remove this slave when it is no longer detected.
If that happens the sl->family_data will be freed and set to NULL
causing w1_slave_show to crash when it wakes up.

Signed-off-by: David Fries <David@Fries.net>
Reported-By: Thorsten Bschorr <thorsten@bschorr.de>
Tested-by: Thorsten Bschorr <thorsten@bschorr.de>
Acked-by: Evgeniy Polyakov <zbr@ioremap.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agoxfs: don't truncate attribute extents if no extents exist
Brian Foster [Mon, 22 Jun 2015 22:47:20 +0000 (08:47 +1000)]
xfs: don't truncate attribute extents if no extents exist

commit f66bf042693b620133d39af8d2f13615f03eadfc upstream.

The xfs_attr3_root_inactive() call from xfs_attr_inactive() assumes that
attribute blocks exist to invalidate. It is possible to have an
attribute fork without extents, however. Consider the case where the
attribute fork is created towards the beginning of xfs_attr_set() but
some part of the subsequent attribute set fails.

If an inode in such a state hits xfs_attr_inactive(), it eventually
calls xfs_dabuf_map() and possibly xfs_bmapi_read(). The former emits a
filesystem corruption warning, returns an error that bubbles back up to
xfs_attr_inactive(), and leads to destruction of the in-core attribute
fork without an on-disk reset. If the inode happens to make it back
through xfs_inactive() in this state (e.g., via a concurrent bulkstat
that cycles the inode from the reclaim state and releases it), i_afp
might not exist when xfs_bmapi_read() is called and causes a NULL
dereference panic.

A '-p 2' fsstress run to ENOSPC on a relatively small fs (1GB)
reproduces these problems. The behavior is a regression caused by:

6dfe5a0 xfs: xfs_attr_inactive leaves inconsistent attr fork state behind

... which removed logic that avoided the attribute extent truncate when
no extents exist. Restore this logic to ensure the attribute fork is
destroyed and reset correctly if it exists without any allocated
extents.

Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agoxfs: fix remote symlinks on V5/CRC filesystems
Eric Sandeen [Sun, 21 Jun 2015 23:42:48 +0000 (09:42 +1000)]
xfs: fix remote symlinks on V5/CRC filesystems

commit 2ac56d3d4bd625450a54d4c3f9292d58f6b88232 upstream.

If we create a CRC filesystem, mount it, and create a symlink with
a path long enough that it can't live in the inode, we get a very
strange result upon remount:

# ls -l mnt
total 4
lrwxrwxrwx. 1 root root 929 Jun 15 16:58 link -> XSLM

XSLM is the V5 symlink block header magic (which happens to be
followed by a NUL, so the string looks terminated).

xfs_readlink_bmap() advanced cur_chunk by the size of the header
for CRC filesystems, but never actually used that pointer; it
kept reading from bp->b_addr, which is the start of the block,
rather than the start of the symlink data after the header.

Looks like this problem goes back to v3.10.

Fixing this gets us reading the proper link target, again.

Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agolibata: Fix regression when the NCQ Send and Receive log page is absent
Martin K. Petersen [Tue, 19 May 2015 23:44:17 +0000 (19:44 -0400)]
libata: Fix regression when the NCQ Send and Receive log page is absent

commit eab6ee1ce3c4678224d70338134f7a02005768cb upstream.

Commit 5d3abf8ff67f ("libata: Fall back to unqueued READ LOG EXT if
the DMA variant fails") allowed us to fall back to the unqueued READ
LOG variant if the queued version failed. However, if the device did
not support the page at all we would end up looping due to a merge
snafu.

Ensure we only take the fallback path once.

Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Reported-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Tested-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agodrm: Stop resetting connector state to unknown
Daniel Vetter [Thu, 16 Jul 2015 14:47:50 +0000 (16:47 +0200)]
drm: Stop resetting connector state to unknown

commit 5677d67ae3949f09f57357241b88222d49b8c782 upstream.

It's causing piles of issues since we've stopped forcing full detect
cycles in the sysfs interfaces with

commit c484f02d0f02fbbfc6decc945a69aae011041a27
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date:   Fri Mar 6 12:36:42 2015 +0000

    drm: Lighten sysfs connector 'status'

The original justification for this was that the hpd handlers could
use the unknown state as a hint to force a full detection. But current
i915 code isn't doing that any more, and no one else really uses reset
on resume. So instead just keep the old state around.

References: http://article.gmane.org/gmane.comp.freedesktop.xorg.drivers.intel/62584
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=100641
Cc: Rui Matos <tiagomatos@gmail.com>
Cc: Julien Wajsberg <felash@gmail.com>
Cc: kuddel.mail@gmx.de
Cc: Lennart Poettering <mzxreary@0pointer.de>
Acked-by: Rob Clark <robdclark@gmail.com>
Tested-by: Rui Tiago Cação Matos <tiagomatos@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agodrm: Provide compat ioctl for addfb2.1
Tvrtko Ursulin [Tue, 14 Jul 2015 10:13:08 +0000 (11:13 +0100)]
drm: Provide compat ioctl for addfb2.1

commit c631d5f90e7ee246536c72f80ade86e9ef4d2f13 upstream.

Frame buffer modifiers extensions provided in;

  commit e3eb3250d84ef97b766312345774367b6a310db8
  Author: Rob Clark <robdclark@gmail.com>
  Date:   Thu Feb 5 14:41:52 2015 +0000

      drm: add support for tiled/compressed/etc modifier in addfb2

Missed the structure packing/alignment problem where 64-bit
members were added after the odd number of 32-bit ones. This
makes the compiler produce structures of different sizes under
32- and 64-bit x86 targets and makes the ioctl need explicit
compat handling.

v2: Removed the typedef. (Daniel Vetter)

Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: dri-devel@lists.freedesktop.org
Cc: Rob Clark <robdclark@gmail.com>
Cc: Daniel Stone <daniels@collabora.com>
Cc: Daniel Vetter <daniel.vetter@intel.com>
[danvet: Squash in compile fix from Mika.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agodrm: add a check for x/y in drm_mode_setcrtc
Zhao Junwang [Tue, 7 Jul 2015 09:08:35 +0000 (17:08 +0800)]
drm: add a check for x/y in drm_mode_setcrtc

commit 01447e9f04ba1c49a9534ae6a5a6f26c2bb05226 upstream.

legacy setcrtc ioctl does take a 32 bit value which might indeed
overflow

the checks of crtc_req->x > INT_MAX and crtc_req->y > INT_MAX aren't
needed any more with this

v2: -polish the annotation according to Daniel's comment

Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Zhao Junwang <zhjwpku@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agodrm/rockchip: use drm_gem_mmap helpers
Daniel Kurtz [Tue, 7 Jul 2015 09:03:36 +0000 (17:03 +0800)]
drm/rockchip: use drm_gem_mmap helpers

commit 41315b793e13f884cda79389f0d5d44d027e57d1 upstream.

Rather than (incompletely [0]) re-implementing drm_gem_mmap() and
drm_gem_mmap_obj() helpers, call them directly from the rockchip mmap
routines.

Once the core functions return successfully, the rockchip mmap routines
can still use dma_mmap_attrs() to simply mmap the entire buffer.

[0] Previously, we were performing the mmap() without first taking a
reference on the underlying gem buffer.  This could leak ptes if the gem
object is destroyed while userspace is still holding the mapping.

Signed-off-by: Daniel Kurtz <djkurtz@chromium.org>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agodrm/radeon/ci: silence a harmless PCC warning
Alex Deucher [Thu, 16 Jul 2015 14:17:09 +0000 (10:17 -0400)]
drm/radeon/ci: silence a harmless PCC warning

commit bda5e3e97ffe80c5a793383df5681d3581d46ac8 upstream.

This has been a source of confusion.  Make it debug only.

Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agodrm/radeon: fix user ptr race condition
Christian König [Tue, 14 Jul 2015 13:58:30 +0000 (15:58 +0200)]
drm/radeon: fix user ptr race condition

commit 12f1384da650bdb835fff63e66fe815ea882fc0e upstream.

Port of amdgpu patch 9298e52f8b51d1e4acd68f502832f3a97f8cf892.

Signed-off-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agodrm/radeon: add a dpm quirk for Sapphire Radeon R9 270X 2GB GDDR5
Alex Deucher [Fri, 10 Jul 2015 01:08:17 +0000 (21:08 -0400)]
drm/radeon: add a dpm quirk for Sapphire Radeon R9 270X 2GB GDDR5

commit 5dfc71bc44d91d1620505c064fa22b0b3db58a9d upstream.

bug:
https://bugs.freedesktop.org/show_bug.cgi?id=76490

Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agodrm/radeon: Don't flush the GART TLB if rdev->gart.ptr == NULL
Michel Dänzer [Fri, 3 Jul 2015 01:02:27 +0000 (10:02 +0900)]
drm/radeon: Don't flush the GART TLB if rdev->gart.ptr == NULL

commit 233709d2cd6bbaaeda0aeb8d11f6ca7f98563b39 upstream.

This can be the case when the GPU is powered off, e.g. via vgaswitcheroo
or runpm. When the GPU is powered up again, radeon_gart_table_vram_pin
flushes the TLB after setting rdev->gart.ptr to non-NULL.

Fixes panic on powering off R7xx GPUs.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=61529
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Michel Dänzer <michel.daenzer@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agodrm/radeon: unpin cursor BOs on suspend and pin them again on resume (v2)
Grigori Goronzy [Tue, 7 Jul 2015 07:27:29 +0000 (16:27 +0900)]
drm/radeon: unpin cursor BOs on suspend and pin them again on resume (v2)

commit f3cbb17bcf676a2fc6aedebe9fbebd59e550c51a upstream.

Everything is evicted from VRAM before suspend, so we need to make
sure all BOs are unpinned and re-pinned after resume. Fixes broken
mouse cursor after resume introduced by commit b9729b17.

[Michel Dänzer: Add pinning BOs on resume]

v2:
[Alex Deucher: merge cursor unpin into fb unpin loop]

Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=100541
Reviewed-by: Christian König <christian.koenig@amd.com> (v1)
Signed-off-by: Grigori Goronzy <greg@chown.ath.cx>
Signed-off-by: Michel Dänzer <michel.daenzer@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agodrm/radeon: Clean up reference counting and pinning of the cursor BOs
Michel Dänzer [Tue, 7 Jul 2015 07:27:28 +0000 (16:27 +0900)]
drm/radeon: Clean up reference counting and pinning of the cursor BOs

commit cd404af0c930104462aa91344f07d002cf8248ed upstream.

Take a GEM reference for and pin the new cursor BO, unpin and drop the
GEM reference for the old cursor BO in radeon_crtc_cursor_set2, and use
radeon_crtc->cursor_addr in radeon_set_cursor.

This fixes radeon_cursor_reset accidentally incrementing the cursor BO
pin count, and cleans up the code a little.

Reviewed-by: Grigori Goronzy <greg@chown.ath.cx>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Michel Dänzer <michel.daenzer@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
9 years agodrm/radeon: Handle irqs only based on irq ring, not irq status regs.
Mario Kleiner [Fri, 3 Jul 2015 04:03:06 +0000 (06:03 +0200)]
drm/radeon: Handle irqs only based on irq ring, not irq status regs.

commit 07f18f0bb8d8d65badd8b4988b40d329fc0cc6dc upstream.

Trying to resolve issues with missed vblanks and impossible
values inside delivered kms pageflip completion events showed
that radeon's irq handling sometimes doesn't handle valid irqs,
but silently skips them. This was observed for vblank interrupts.

Although those irqs have corresponding events queued in the gpu's
irq ring at time of interrupt, and therefore the corresponding
handling code gets triggered by these events, the handling code
sometimes silently skipped processing the irq. The reason for those
skips is that the handling code double-checks for each irq event if
the corresponding irq status bits in the irq status registers
are set. Sometimes those bits are not set at time of check
for valid irqs, maybe due to some hardware race on some setups?

The problem only seems to happen on some machine + card combos
sometimes, e.g., never happened during my testing of different PC
cards of the DCE-2/3/4 generation a year ago, but happens consistently
now on two different Apple Mac cards (RV730, DCE-3, Apple iMac and
Evergreen JUNIPER, DCE-4 in a Apple MacPro). It also doesn't happen
at each interrupt but only occassionally every couple of
hundred or thousand vblank interrupts.

This results in XOrg warning messages like

"[  7084.472] (WW) RADEON(0): radeon_dri2_flip_event_handler:
Pageflip completion event has impossible msc 420120 < target_msc 420121"

as well as skipped frames and problems for applications that
use kms pageflip events or vblank events, e.g., users of DRI2 and
DRI3/Present, Waylands Weston compositor, etc. See also

https://bugs.freedesktop.org/show_bug.cgi?id=85203

After some talking to Alex and Michel, we decided to fix this
by turning the double-check for asserted irq status bits into a
warning. Whenever a irq event is queued in the IH ring, always
execute the corresponding interrupt handler. Still check the irq
status bits, but only to log a DRM_DEBUG message on a mismatch.

This fixed the problems reliably on both previously failing
cards, RV-730 dual-head tested on both crtcs (pipes D1 and D2)
and a triple-output Juniper HD-5770 card tested on all three
available crtcs (D1/D2/D3). The r600 and evergreen irq handling
is therefore tested, but the cik an si handling is only compile
tested due to lack of hw.

Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Mario Kleiner <mario.kleiner.de@gmail.com>
CC: Michel Dänzer <michel.daenzer@amd.com>
CC: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>