7 years agomm/hugetlb: add migration/hwpoisoned entry check in hugetlb_change_protection
Naoya Horiguchi [Wed, 11 Feb 2015 23:25:28 +0000 (15:25 -0800)]
mm/hugetlb: add migration/hwpoisoned entry check in hugetlb_change_protection

commit a8bda28d87c38c6aa93de28ba5d30cc18e865a11 upstream.

There is a race condition between hugepage migration and
change_protection(), where hugetlb_change_protection() doesn't care about
migration entries and wrongly overwrites them.  That causes unexpected
results like kernel crash.  HWPoison entries also can cause the same

This patch adds is_hugetlb_entry_(migration|hwpoisoned) check in this
function to do proper actions.

Fixes: 290408d4a2 ("hugetlb: hugepage migration core")
Signed-off-by: Naoya Horiguchi <>
Cc: Hugh Dickins <>
Cc: James Hogan <>
Cc: David Rientjes <>
Cc: Mel Gorman <>
Cc: Johannes Weiner <>
Cc: Michal Hocko <>
Cc: Rik van Riel <>
Cc: Andrea Arcangeli <>
Cc: Luiz Capitulino <>
Cc: Nishanth Aravamudan <>
Cc: Lee Schermerhorn <>
Cc: Steve Capper <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
[bwh: Backported to 3.2:
 - Adjust context
 - We don't have split page table locks, so don't unlock inside the loop
 - We don't count pages here]
Signed-off-by: Ben Hutchings <>
7 years agomm/hugetlb: fix getting refcount 0 page in hugetlb_fault()
Naoya Horiguchi [Wed, 11 Feb 2015 23:25:25 +0000 (15:25 -0800)]
mm/hugetlb: fix getting refcount 0 page in hugetlb_fault()

commit 0f792cf949a0be506c2aa8bfac0605746b146dda upstream.

When running the test which causes the race as shown in the previous patch,
we can hit the BUG "get_page() on refcount 0 page" in hugetlb_fault().

This race happens when pte turns into migration entry just after the first
check of is_hugetlb_entry_migration() in hugetlb_fault() passed with false.
To fix this, we need to check pte_present() again after huge_ptep_get().

This patch also reorders taking ptl and doing pte_page(), because
pte_page() should be done in ptl.  Due to this reordering, we need use
trylock_page() in page != pagecache_page case to respect locking order.

Fixes: 66aebce747ea ("hugetlb: fix race condition in hugetlb_fault()")
Signed-off-by: Naoya Horiguchi <>
Cc: Hugh Dickins <>
Cc: James Hogan <>
Cc: David Rientjes <>
Cc: Mel Gorman <>
Cc: Johannes Weiner <>
Cc: Michal Hocko <>
Cc: Rik van Riel <>
Cc: Andrea Arcangeli <>
Cc: Luiz Capitulino <>
Cc: Nishanth Aravamudan <>
Cc: Lee Schermerhorn <>
Cc: Steve Capper <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
[bwh: Backported to 3.2:
 - Adjust context
 - Error label is named 'out_page_table_lock' not 'out_ptl']
Signed-off-by: Ben Hutchings <>
7 years agocpufreq: speedstep-smi: enable interrupts when waiting
Mikulas Patocka [Mon, 9 Feb 2015 18:38:17 +0000 (13:38 -0500)]
cpufreq: speedstep-smi: enable interrupts when waiting

commit d4d4eda23794c701442e55129dd4f8f2fefd5e4d upstream.

On Dell Latitude C600 laptop with Pentium 3 850MHz processor, the
speedstep-smi driver sometimes loads and sometimes doesn't load with
"change to state X failed" message.

The hardware sometimes refuses to change frequency and in this case, we
need to retry later. I found out that we need to enable interrupts while
waiting. When we enable interrupts, the hardware blockage that prevents
frequency transition resolves and the transition is possible. With
disabled interrupts, the blockage doesn't resolve (no matter how long do
we wait). The exact reasons for this hardware behavior are unknown.

This patch enables interrupts in the function speedstep_set_state that can
be called with disabled interrupts. However, this function is called with
disabled interrupts only from speedstep_get_freqs, so it shouldn't cause
any problem.

Signed-off-by: Mikulas Patocka <
Acked-by: Viresh Kumar <>
Signed-off-by: Rafael J. Wysocki <>
Signed-off-by: Ben Hutchings <>
7 years agoNFSv4.1: Fix a kfree() of uninitialised pointers in decode_cb_sequence_args
Trond Myklebust [Wed, 11 Feb 2015 22:27:55 +0000 (17:27 -0500)]
NFSv4.1: Fix a kfree() of uninitialised pointers in decode_cb_sequence_args

commit d8ba1f971497c19cf80da1ea5391a46a5f9fbd41 upstream.

If the call to decode_rc_list() fails due to a memory allocation error,
then we need to truncate the array size to ensure that we only call
kfree() on those pointer that were allocated.

Reported-by: David Ramos <>
Fixes: 4aece6a19cf7f ("nfs41: cb_sequence xdr implementation")
Signed-off-by: Trond Myklebust <>
Signed-off-by: Ben Hutchings <>
7 years agofsnotify: fix handling of renames in audit
Jan Kara [Tue, 10 Feb 2015 22:08:32 +0000 (14:08 -0800)]
fsnotify: fix handling of renames in audit

commit 6ee8e25fc3e916193bce4ebb43d5439e1e2144ab upstream.

Commit e9fd702a58c4 ("audit: convert audit watches to use fsnotify
instead of inotify") broke handling of renames in audit.  Audit code
wants to update inode number of an inode corresponding to watched name
in a directory.  When something gets renamed into a directory to a
watched name, inotify previously passed moved inode to audit code
however new fsnotify code passes directory inode where the change
happened.  That confuses audit and it starts watching parent directory
instead of a file in a directory.

This can be observed for example by doing:

  cd /tmp
  touch foo bar
  auditctl -w /tmp/foo
  touch foo
  mv bar foo
  touch foo

In audit log we see events like:

  type=CONFIG_CHANGE msg=audit(1423563584.155:90): auid=1000 ses=2 op="updated rules" path="/tmp/foo" key=(null) list=4 res=1
  type=PATH msg=audit(1423563584.155:91): item=2 name="bar" inode=1046884 dev=08:0 2 mode=0100644 ouid=0 ogid=0 rdev=00:00 nametype=DELETE
  type=PATH msg=audit(1423563584.155:91): item=3 name="foo" inode=1046842 dev=08:0 2 mode=0100644 ouid=0 ogid=0 rdev=00:00 nametype=DELETE
  type=PATH msg=audit(1423563584.155:91): item=4 name="foo" inode=1046884 dev=08:0 2 mode=0100644 ouid=0 ogid=0 rdev=00:00 nametype=CREATE

and that's it - we see event for the first touch after creating the
audit rule, we see events for rename but we don't see any event for the
last touch.  However we start seeing events for unrelated stuff
happening in /tmp.

Fix the problem by passing moved inode as data in the FS_MOVED_FROM and
FS_MOVED_TO events instead of the directory where the change happens.
This doesn't introduce any new problems because noone besides
audit_watch.c cares about the passed value:

  fs/notify/fanotify/fanotify.c cares only about FSNOTIFY_EVENT_PATH events.
  fs/notify/dnotify/dnotify.c doesn't care about passed 'data' value at all.
  fs/notify/inotify/inotify_fsnotify.c uses 'data' only for FSNOTIFY_EVENT_PATH.
  kernel/audit_tree.c doesn't care about passed 'data' at all.
  kernel/audit_watch.c expects moved inode as 'data'.

Fixes: e9fd702a58c49db ("audit: convert audit watches to use fsnotify instead of inotify")
Signed-off-by: Jan Kara <>
Cc: Paul Moore <>
Cc: Eric Paris <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
Signed-off-by: Ben Hutchings <>
7 years agoALSA: off by one bug in snd_riptide_joystick_probe()
Dan Carpenter [Mon, 9 Feb 2015 13:51:40 +0000 (16:51 +0300)]
ALSA: off by one bug in snd_riptide_joystick_probe()

commit e4940626defdf6c92da1052ad3f12741c1a28c90 upstream.

The problem here is that we check:

if (dev >= SNDRV_CARDS)

Then we increment "dev".

       if (!joystick_port[dev++])

Then we use it as an offset into a array with SNDRV_CARDS elements.

if (!request_region(joystick_port[dev], 8, "Riptide gameport")) {

This has 3 effects:
1) If you use the module option to specify the joystick port then it has
   to be shifted one space over.
2) The wrong error message will be printed on failure if you have over
   32 cards.
3) Static checkers will correctly complain that are off by one.

Fixes: db1005ec6ff8 ('ALSA: riptide - Fix joystick resource handling')
Signed-off-by: Dan Carpenter <>
Signed-off-by: Takashi Iwai <>
Signed-off-by: Ben Hutchings <>
7 years agortnetlink: ifla_vf_policy: fix misuses of NLA_BINARY
Daniel Borkmann [Thu, 5 Feb 2015 17:44:04 +0000 (18:44 +0100)]
rtnetlink: ifla_vf_policy: fix misuses of NLA_BINARY

commit 364d5716a7adb91b731a35765d369602d68d2881 upstream.

ifla_vf_policy[] is wrong in advertising its individual member types as
NLA_BINARY since .type = NLA_BINARY in combination with .len declares the
len member as *max* attribute length [0, len].

The issue is that when do_setvfinfo() is being called to set up a VF
through ndo handler, we could set corrupted data if the attribute length
is less than the size of the related structure itself.

The intent is exactly the opposite, namely to make sure to pass at least
data of minimum size of len.

Fixes: ebc08a6f47ee ("rtnetlink: Add VF config code to rtnetlink")
Cc: Mitch Williams <>
Cc: Jeff Kirsher <>
Signed-off-by: Daniel Borkmann <>
Acked-by: Thomas Graf <>
Signed-off-by: David S. Miller <>
[bwh: Backported to 3.2: drop the unsupported attributes]
Signed-off-by: Ben Hutchings <>
7 years agoxen/manage: Fix USB interaction issues when resuming
Ross Lagerwall [Mon, 19 Jan 2015 13:19:38 +0000 (13:19 +0000)]
xen/manage: Fix USB interaction issues when resuming

commit 72978b2fe2f2cdf9f319c6c6dcdbe92b38de2be2 upstream.

Commit 61a734d305e1 ("xen/manage: Always freeze/thaw processes when
suspend/resuming") ensured that userspace processes were always frozen
before suspending to reduce interaction issues when resuming devices.
However, freeze_processes() does not freeze kernel threads.  Freeze
kernel threads as well to prevent deadlocks with the khubd thread when
resuming devices.

This is what native suspend and resume does.

Example deadlock:
[ 7279.648010]  [<ffffffff81446bde>] ? xen_poll_irq_timeout+0x3e/0x50
[ 7279.648010]  [<ffffffff81448d60>] xen_poll_irq+0x10/0x20
[ 7279.648010]  [<ffffffff81011723>] xen_lock_spinning+0xb3/0x120
[ 7279.648010]  [<ffffffff810115d1>] __raw_callee_save_xen_lock_spinning+0x11/0x20
[ 7279.648010]  [<ffffffff815620b6>] ? usb_control_msg+0xe6/0x120
[ 7279.648010]  [<ffffffff81747e50>] ? _raw_spin_lock_irq+0x50/0x60
[ 7279.648010]  [<ffffffff8174522c>] wait_for_completion+0xac/0x160
[ 7279.648010]  [<ffffffff8109c520>] ? try_to_wake_up+0x2c0/0x2c0
[ 7279.648010]  [<ffffffff814b60f2>] dpm_wait+0x32/0x40
[ 7279.648010]  [<ffffffff814b6eb0>] device_resume+0x90/0x210
[ 7279.648010]  [<ffffffff814b7d71>] dpm_resume+0x121/0x250
[ 7279.648010]  [<ffffffff8144c570>] ? xenbus_dev_request_and_reply+0xc0/0xc0
[ 7279.648010]  [<ffffffff814b80d5>] dpm_resume_end+0x15/0x30
[ 7279.648010]  [<ffffffff81449fba>] do_suspend+0x10a/0x200
[ 7279.648010]  [<ffffffff8144a2f0>] ? xen_pre_suspend+0x20/0x20
[ 7279.648010]  [<ffffffff8144a1d0>] shutdown_handler+0x120/0x150
[ 7279.648010]  [<ffffffff8144c60f>] xenwatch_thread+0x9f/0x160
[ 7279.648010]  [<ffffffff810ac510>] ? finish_wait+0x80/0x80
[ 7279.648010]  [<ffffffff8108d189>] kthread+0xc9/0xe0
[ 7279.648010]  [<ffffffff8108d0c0>] ? flush_kthread_worker+0x80/0x80
[ 7279.648010]  [<ffffffff8175087c>] ret_from_fork+0x7c/0xb0
[ 7279.648010]  [<ffffffff8108d0c0>] ? flush_kthread_worker+0x80/0x80

[ 7441.216287] INFO: task khubd:89 blocked for more than 120 seconds.
[ 7441.219457]       Tainted: G            X #1
[ 7441.222176] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
[ 7441.225827] khubd           D ffff88003f433440     0    89      2 0x00000000
[ 7441.229258]  ffff88003ceb9b98 0000000000000046 ffff88003ce83000 0000000000013440
[ 7441.232959]  ffff88003ceb9fd8 0000000000013440 ffff88003cd13000 ffff88003ce83000
[ 7441.236658]  0000000000000286 ffff88003d3e0000 ffff88003ceb9bd0 00000001001aa01e
[ 7441.240415] Call Trace:
[ 7441.241614]  [<ffffffff817442f9>] schedule+0x29/0x70
[ 7441.243930]  [<ffffffff81743406>] schedule_timeout+0x166/0x2c0
[ 7441.246681]  [<ffffffff81075b80>] ? call_timer_fn+0x110/0x110
[ 7441.249339]  [<ffffffff8174357e>] schedule_timeout_uninterruptible+0x1e/0x20
[ 7441.252644]  [<ffffffff81077710>] msleep+0x20/0x30
[ 7441.254812]  [<ffffffff81555f00>] hub_port_reset+0xf0/0x580
[ 7441.257400]  [<ffffffff81558465>] hub_port_init+0x75/0xb40
[ 7441.259981]  [<ffffffff814bb3c9>] ? update_autosuspend+0x39/0x60
[ 7441.262817]  [<ffffffff814bb4f0>] ? pm_runtime_set_autosuspend_delay+0x50/0xa0
[ 7441.266212]  [<ffffffff8155a64a>] hub_thread+0x71a/0x1750
[ 7441.268728]  [<ffffffff810ac510>] ? finish_wait+0x80/0x80
[ 7441.271272]  [<ffffffff81559f30>] ? usb_port_resume+0x670/0x670
[ 7441.274067]  [<ffffffff8108d189>] kthread+0xc9/0xe0
[ 7441.276305]  [<ffffffff8108d0c0>] ? flush_kthread_worker+0x80/0x80
[ 7441.279131]  [<ffffffff8175087c>] ret_from_fork+0x7c/0xb0
[ 7441.281659]  [<ffffffff8108d0c0>] ? flush_kthread_worker+0x80/0x80

Signed-off-by: Ross Lagerwall <>
Signed-off-by: David Vrabel <>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <>
7 years agolmedm04: Fix usb_submit_urb BOGUS urb xfer, pipe 1 != type 3 in interrupt urb
Malcolm Priestley [Fri, 2 Jan 2015 13:56:28 +0000 (10:56 -0300)]
lmedm04: Fix usb_submit_urb BOGUS urb xfer, pipe 1 != type 3 in interrupt urb

commit 15e1ce33182d1d5dbd8efe8d382b9352dc857527 upstream.

A quirk of some older firmwares that report endpoint pipe type as PIPE_BULK
but the endpoint otheriwse functions as interrupt.

Check if usb_endpoint_type is USB_ENDPOINT_XFER_BULK and set as usb_rcvbulkpipe.

Signed-off-by: Malcolm Priestley <>
Signed-off-by: Mauro Carvalho Chehab <>
[bwh: Backported to 3.2:
 - Adjust filename, context
 - Add definition of the local variable 'd']
Signed-off-by: Ben Hutchings <>
7 years agotty: Prevent untrappable signals from malicious program
Peter Hurley [Mon, 19 Jan 2015 18:05:03 +0000 (13:05 -0500)]
tty: Prevent untrappable signals from malicious program

commit 37480a05685ed5b8e1b9bf5e5c53b5810258b149 upstream.

Commit 26df6d13406d1a5 ("tty: Add EXTPROC support for LINEMODE")
allows a process which has opened a pty master to send _any_ signal
to the process group of the pty slave. Although potentially
exploitable by a malicious program running a setuid program on
a pty slave, it's unknown if this exploit currently exists.

Limit to signals actually used.

Cc: Theodore Ts'o <>
Cc: Howard Chu <>
Cc: One Thousand Gnomes <>
Cc: Jiri Slaby <>
Signed-off-by: Peter Hurley <>
Signed-off-by: Greg Kroah-Hartman <>
Signed-off-by: Ben Hutchings <>
7 years agovt: provide notifications on selection changes
Nicolas Pitre [Fri, 23 Jan 2015 22:07:21 +0000 (17:07 -0500)]
vt: provide notifications on selection changes

commit 19e3ae6b4f07a87822c1c9e7ed99d31860e701af upstream.

The vcs device's poll/fasync support relies on the vt notifier to signal
changes to the screen content.  Notifier invocations were missing for
changes that comes through the selection interface though.  Fix that.

Tested with BRLTTY 5.2.

Signed-off-by: Nicolas Pitre <>
Cc: Dave Mielke <>
Signed-off-by: Greg Kroah-Hartman <>
Signed-off-by: Ben Hutchings <>
7 years agoUSB: fix use-after-free bug in usb_hcd_unlink_urb()
Alan Stern [Fri, 30 Jan 2015 17:58:26 +0000 (12:58 -0500)]
USB: fix use-after-free bug in usb_hcd_unlink_urb()

commit c99197902da284b4b723451c1471c45b18537cde upstream.

The usb_hcd_unlink_urb() routine in hcd.c contains two possible
use-after-free errors.  The dev_dbg() statement at the end of the
routine dereferences urb and urb->dev even though both structures may
have been deallocated.

This patch fixes the problem by storing urb->dev in a local variable
(avoiding the dereference of urb) and moving the dev_dbg() up before
the usb_put_dev() call.

Signed-off-by: Alan Stern <>
Reported-by: Joe Lawrence <>
Tested-by: Joe Lawrence <>
Signed-off-by: Greg Kroah-Hartman <>
Signed-off-by: Ben Hutchings <>
7 years agoUSB: add flag for HCDs that can't receive wakeup requests (isp1760-hcd)
Alan Stern [Thu, 29 Jan 2015 20:05:04 +0000 (15:05 -0500)]
USB: add flag for HCDs that can't receive wakeup requests (isp1760-hcd)

commit 074f9dd55f9cab1b82690ed7e44bcf38b9616ce0 upstream.

Currently the USB stack assumes that all host controller drivers are
capable of receiving wakeup requests from downstream devices.
However, this isn't true for the isp1760-hcd driver, which means that
it isn't safe to do a runtime suspend of any device attached to a
root-hub port if the device requires wakeup.

This patch adds a "cant_recv_wakeups" flag to the usb_hcd structure
and sets the flag in isp1760-hcd.  The core is modified to prevent a
direct child of the root hub from being put into runtime suspend with
wakeup enabled if the flag is set.

Signed-off-by: Alan Stern <>
Tested-by: Nicolas Pitre <>
Signed-off-by: Greg Kroah-Hartman <>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <>
7 years agocdc-acm: add sanity checks
Oliver Neukum [Wed, 28 Jan 2015 10:14:55 +0000 (11:14 +0100)]
cdc-acm: add sanity checks

commit 7e860a6e7aa62b337a61110430cd633db5b0d2dd upstream.

Check the special CDC headers for a plausible minimum length.
Another big operating systems ignores such garbage.

Signed-off-by: Oliver Neukum <>
Reviewed-by: Adam Lee <>
Tested-by: Adam Lee <>
Signed-off-by: Greg Kroah-Hartman <>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <>
7 years agonfs: don't call blocking operations while !TASK_RUNNING
Jeff Layton [Wed, 14 Jan 2015 18:08:57 +0000 (13:08 -0500)]
nfs: don't call blocking operations while !TASK_RUNNING

commit 6ffa30d3f734d4f6b478081dfc09592021028f90 upstream.

Bruce reported seeing this warning pop when mounting using v4.1:

     ------------[ cut here ]------------
     WARNING: CPU: 1 PID: 1121 at kernel/sched/core.c:7300 __might_sleep+0xbd/0xd0()
    do not call blocking ops when !TASK_RUNNING; state=1 set at [<ffffffff810ff58f>] prepare_to_wait+0x2f/0x90
    Modules linked in: rpcsec_gss_krb5 auth_rpcgss nfsv4 dns_resolver nfs lockd grace sunrpc fscache ip6t_rpfilter ip6t_REJECT nf_reject_ipv6 xt_conntrack ebtable_nat ebtable_broute bridge stp llc ebtable_filter ebtables ip6table_nat nf_conntrack_ipv6 nf_defrag_ipv6 nf_nat_ipv6 ip6table_mangle ip6table_security ip6table_raw ip6table_filter ip6_tables iptable_nat nf_conntrack_ipv4 nf_defrag_ipv4 nf_nat_ipv4 nf_nat nf_conntrack iptable_mangle iptable_security iptable_raw snd_hda_codec_generic snd_hda_intel snd_hda_controller snd_hda_codec snd_hwdep snd_pcm snd_timer ppdev joydev snd virtio_console virtio_balloon pcspkr serio_raw parport_pc parport pvpanic floppy soundcore i2c_piix4 virtio_blk virtio_net qxl drm_kms_helper ttm drm virtio_pci virtio_ring ata_generic virtio pata_acpi
    CPU: 1 PID: 1121 Comm: nfsv4.1-svc Not tainted 3.19.0-rc4+ #25
    Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.7.5-20140709_153950- 04/01/2014
     0000000000000000 000000004e5e3f73 ffff8800b998fb48 ffffffff8186ac78
     0000000000000000 ffff8800b998fba0 ffff8800b998fb88 ffffffff810ac9da
     ffff8800b998fb68 ffffffff81c923e7 00000000000004d9 0000000000000000
    Call Trace:
     [<ffffffff8186ac78>] dump_stack+0x4c/0x65
     [<ffffffff810ac9da>] warn_slowpath_common+0x8a/0xc0
     [<ffffffff810aca65>] warn_slowpath_fmt+0x55/0x70
     [<ffffffff810ff58f>] ? prepare_to_wait+0x2f/0x90
     [<ffffffff810ff58f>] ? prepare_to_wait+0x2f/0x90
     [<ffffffff810dd2ad>] __might_sleep+0xbd/0xd0
     [<ffffffff8124c973>] kmem_cache_alloc_trace+0x243/0x430
     [<ffffffff810d941e>] ? groups_alloc+0x3e/0x130
     [<ffffffff810d941e>] groups_alloc+0x3e/0x130
     [<ffffffffa0301b1e>] svcauth_unix_accept+0x16e/0x290 [sunrpc]
     [<ffffffffa0300571>] svc_authenticate+0xe1/0xf0 [sunrpc]
     [<ffffffffa02fc564>] svc_process_common+0x244/0x6a0 [sunrpc]
     [<ffffffffa02fd044>] bc_svc_process+0x1c4/0x260 [sunrpc]
     [<ffffffffa03d5478>] nfs41_callback_svc+0x128/0x1f0 [nfsv4]
     [<ffffffff810ff970>] ? wait_woken+0xc0/0xc0
     [<ffffffffa03d5350>] ? nfs4_callback_svc+0x60/0x60 [nfsv4]
     [<ffffffff810d45bf>] kthread+0x11f/0x140
     [<ffffffff810ea815>] ? local_clock+0x15/0x30
     [<ffffffff810d44a0>] ? kthread_create_on_node+0x250/0x250
     [<ffffffff81874bfc>] ret_from_fork+0x7c/0xb0
     [<ffffffff810d44a0>] ? kthread_create_on_node+0x250/0x250
    ---[ end trace 675220a11e30f4f2 ]---

nfs41_callback_svc does most of its work while in TASK_INTERRUPTIBLE,
which is just wrong. Fix that by finishing the wait immediately if we've
found that the list has something on it.

Also, we don't expect this kthread to accept signals, so we should be
using a TASK_UNINTERRUPTIBLE sleep instead. That however, opens us up
hung task warnings from the watchdog, so have the schedule_timeout
wake up every 60s if there's no callback activity.

Reported-by: "J. Bruce Fields" <>
Signed-off-by: Jeff Layton <>
Signed-off-by: Trond Myklebust <>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <>
7 years agoxprtrdma: Free the pd if ib_query_qp() fails
Chuck Lever [Wed, 21 Jan 2015 16:03:19 +0000 (11:03 -0500)]
xprtrdma: Free the pd if ib_query_qp() fails

commit 5ae711a24601257f395c1f8746ac95be0cbd75e5 upstream.

If ib_query_qp() fails or the memory registration mode isn't
supported, don't leak the PD. An orphaned IB/core resource will
cause IB module removal to hang.

Fixes: bd7ed1d13304 ("RPC/RDMA: check selected memory registration ...")
Signed-off-by: Chuck Lever <>
Reviewed-by: Steve Wise <>
Signed-off-by: Anna Schumaker <>
[bwh: Backported to 3.2:
 - Adjust context
 - There are only 2 goto's to be changed, not 3]
Signed-off-by: Ben Hutchings <>
7 years agoARM: 8284/1: sa1100: clear RCSR_SMR on resume
Dmitry Eremin-Solenikov [Thu, 15 Jan 2015 02:06:22 +0000 (03:06 +0100)]
ARM: 8284/1: sa1100: clear RCSR_SMR on resume

commit e461894dc2ce7778ccde1c3483c9b15a85a7fc5f upstream.

StrongARM core uses RCSR SMR bit to tell to bootloader that it was reset
by entering the sleep mode. After we have resumed, there is little point
in having that bit enabled. Moreover, if this bit is set before reboot,
the bootloader can become confused. Thus clear the SMR bit on resume
just before clearing the scratchpad (resume address) register.

Signed-off-by: Dmitry Eremin-Solenikov <>
Signed-off-by: Russell King <>
Signed-off-by: Ben Hutchings <>
7 years agostaging: comedi: comedi_compat32.c: fix COMEDI_CMD copy back
Ian Abbott [Tue, 27 Jan 2015 18:16:51 +0000 (18:16 +0000)]
staging: comedi: comedi_compat32.c: fix COMEDI_CMD copy back

commit 42b8ce6f55facfa101462e694d33fc6bca471138 upstream.

`do_cmd_ioctl()` in "comedi_fops.c" handles the `COMEDI_CMD` ioctl.
This returns `-EAGAIN` if it has copied a modified `struct comedi_cmd`
back to user-space.  (This occurs when the low-level Comedi driver's
`do_cmdtest()` handler returns non-zero to indicate a problem with the
contents of the `struct comedi_cmd`, or when the `struct comedi_cmd` has
the `CMDF_BOGUS` flag set.)

`compat_cmd()` in "comedi_compat32.c" handles the 32-bit compatible
version of the `COMEDI_CMD` ioctl.  Currently, it never copies a 32-bit
compatible version of `struct comedi_cmd` back to user-space, which is
at odds with the way the regular `COMEDI_CMD` ioctl is handled.  To fix
it, change `compat_cmd()` to copy a 32-bit compatible version of the
`struct comedi_cmd` back to user-space when the main ioctl handler
returns `-EAGAIN`.

Signed-off-by: Ian Abbott <>
Reviewed-by: H Hartley Sweeten <>
Signed-off-by: Greg Kroah-Hartman <>
Signed-off-by: Ben Hutchings <>
7 years agoiio: imu: adis16400: Fix sign extension
Rasmus Villemoes [Thu, 22 Jan 2015 23:34:02 +0000 (00:34 +0100)]
iio: imu: adis16400: Fix sign extension

commit 19e353f2b344ad86cea6ebbc0002e5f903480a90 upstream.

The intention is obviously to sign-extend a 12 bit quantity. But
because of C's promotion rules, the assignment is equivalent to "val16
&= 0xfff;". Use the proper API for this.

Signed-off-by: Rasmus Villemoes <>
Acked-by: Lars-Peter Clausen <>
Signed-off-by: Jonathan Cameron <>
[bwh: Backported to 3.2: adjust filename, context]
Signed-off-by: Ben Hutchings <>
7 years agoUSB: cp210x: add ID for RUGGEDCOM USB Serial Console
Lennart Sorensen [Wed, 21 Jan 2015 20:24:27 +0000 (15:24 -0500)]
USB: cp210x: add ID for RUGGEDCOM USB Serial Console

commit a6f0331236fa75afba14bbcf6668d42cebb55c43 upstream.

Added the USB serial console device ID for Siemens Ruggedcom devices
which have a USB port for their serial console.

Signed-off-by: Len Sorensen <>
Signed-off-by: Johan Hovold <>
Signed-off-by: Ben Hutchings <>
7 years agoPCI: Fix infinite loop with ROM image of size 0
Michel Dänzer [Mon, 19 Jan 2015 08:53:20 +0000 (17:53 +0900)]
PCI: Fix infinite loop with ROM image of size 0

commit 16b036af31e1456cb69243a5a0c9ef801ecd1f17 upstream.

If the image size would ever read as 0, pci_get_rom_size() could keep
processing the same image over and over again.  Exit the loop if we ever
read a length of zero.

This fixes a soft lockup on boot when the radeon driver calls
pci_get_rom_size() on an AMD Radeon R7 250X PCIe discrete graphics card.

[bhelgaas: changelog, reference]
Reported-by: Federico <>
Signed-off-by: Michel Dänzer <>
Signed-off-by: Bjorn Helgaas <>
Reviewed-by: Alex Deucher <>
Signed-off-by: Ben Hutchings <>
7 years agoKVM: s390: base hrtimer on a monotonic clock
David Hildenbrand [Fri, 12 Dec 2014 14:17:31 +0000 (15:17 +0100)]
KVM: s390: base hrtimer on a monotonic clock

commit 0ac96caf0f9381088c673a16d910b1d329670edf upstream.

The hrtimer that handles the wait with enabled timer interrupts
should not be disturbed by changes of the host time.

This patch changes our hrtimer to be based on a monotonic clock.

Signed-off-by: David Hildenbrand <>
Acked-by: Cornelia Huck <>
Signed-off-by: Christian Borntraeger <>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <>
7 years agosmack: fix possible use after frees in task_security() callers
Andrey Ryabinin [Tue, 13 Jan 2015 15:52:40 +0000 (18:52 +0300)]
smack: fix possible use after frees in task_security() callers

commit 6d1cff2a885850b78b40c34777b46cf5da5d1050 upstream.

We hit use after free on dereferncing pointer to task_smack struct in
smk_of_task() called from smack_task_to_inode().

task_security() macro uses task_cred_xxx() to get pointer to the task_smack.
task_cred_xxx() could be used only for non-pointer members of task's
credentials. It cannot be used for pointer members since what they point
to may disapper after dropping RCU read lock.

Mainly task_security() used this way:

Intead of this introduce function smk_of_task_struct() which
takes task_struct as argument and returns pointer to smk_known struct
and do this under RCU read lock.
Bogus task_security() macro is not used anymore, so remove it.

KASan's report for this:

AddressSanitizer: use after free in smack_task_to_inode+0x50/0x70 at addr c4635600
BUG kmalloc-64 (Tainted: PO): kasan error

Disabling lock debugging due to kernel taint
INFO: Allocated in new_task_smack+0x44/0xd8 age=39 cpu=0 pid=1866
INFO: Freed in smack_cred_free+0xc4/0xd0 age=27 cpu=0 pid=1564
INFO: Slab 0xd3b23f50 objects=32 used=31 fp=0xc4635600 flags=0x4080
INFO: Object 0xc4635600 @offset=5632 fp=0x  (null)

Bytes b4 c46355f0: 5a 5a 5a 5a 5a 5a 5a 5a 5a 5a 5a 5a 5a 5a 5a 5a  ZZZZZZZZZZZZZZZZ
Object c4635600: 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b  kkkkkkkkkkkkkkkk
Object c4635610: 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b  kkkkkkkkkkkkkkkk
Object c4635620: 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b  kkkkkkkkkkkkkkkk
Object c4635630: 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b a5  kkkkkkkkkkkkkkk.
Redzone c4635640: bb bb bb bb                                      ....
Padding c46356e8: 5a 5a 5a 5a 5a 5a 5a 5a 5a 5a 5a 5a 5a 5a 5a 5a  ZZZZZZZZZZZZZZZZ
Padding c46356f8: 5a 5a 5a 5a 5a 5a 5a 5a                          ZZZZZZZZ
CPU: 5 PID: 834 Comm: launchpad_prelo Tainted: PBO 3.10.30 #1
[<c00233a4>] (dump_backtrace+0x0/0x158) from [<c0023dec>] (show_stack+0x20/0x24)
 r7:c4634010 r6:d3b23f50 r5:c4635600 r4:d1002140
[<c0023dcc>] (show_stack+0x0/0x24) from [<c06d6d7c>] (dump_stack+0x20/0x28)
[<c06d6d5c>] (dump_stack+0x0/0x28) from [<c01c1d50>] (print_trailer+0x124/0x144)
[<c01c1c2c>] (print_trailer+0x0/0x144) from [<c01c1e88>] (object_err+0x3c/0x44)
 r7:c4635600 r6:d1002140 r5:d3b23f50 r4:c4635600
[<c01c1e4c>] (object_err+0x0/0x44) from [<c01cac18>] (kasan_report_error+0x2b8/0x538)
 r6:d1002140 r5:d3b23f50 r4:c6429cf8 r3:c09e1aa7
[<c01ca960>] (kasan_report_error+0x0/0x538) from [<c01c9430>] (__asan_load4+0xd4/0xf8)
[<c01c935c>] (__asan_load4+0x0/0xf8) from [<c031e168>] (smack_task_to_inode+0x50/0x70)
 r5:c4635600 r4:ca9da000
[<c031e118>] (smack_task_to_inode+0x0/0x70) from [<c031af64>] (security_task_to_inode+0x3c/0x44)
 r5:cca25e80 r4:c0ba9780
[<c031af28>] (security_task_to_inode+0x0/0x44) from [<c023d614>] (pid_revalidate+0x124/0x178)
 r6:00000000 r5:cca25e80 r4:cbabe3c0 r3:00008124
[<c023d4f0>] (pid_revalidate+0x0/0x178) from [<c01db98c>] (lookup_fast+0x35c/0x43y4)
 r9:c6429efc r8:00000101 r7:c079d940 r6:c6429e90 r5:c6429ed8 r4:c83c4148
[<c01db630>] (lookup_fast+0x0/0x434) from [<c01deec8>] (do_last.isra.24+0x1c0/0x1108)
[<c01ded08>] (do_last.isra.24+0x0/0x1108) from [<c01dff04>] (path_openat.isra.25+0xf4/0x648)
[<c01dfe10>] (path_openat.isra.25+0x0/0x648) from [<c01e1458>] (do_filp_open+0x3c/0x88)
[<c01e141c>] (do_filp_open+0x0/0x88) from [<c01ccb28>] (do_sys_open+0xf0/0x198)
 r7:00000001 r6:c0ea2180 r5:0000000b r4:00000000
[<c01cca38>] (do_sys_open+0x0/0x198) from [<c01ccc00>] (SyS_open+0x30/0x34)
[<c01ccbd0>] (SyS_open+0x0/0x34) from [<c001db80>] (ret_fast_syscall+0x0/0x48)
Read of size 4 by thread T834:
Memory state around the buggy address:
 c4635380: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
 c4635400: 00 00 00 00 00 00 00 00 fc fc fc fc fc fc fc fc
 c4635480: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
 c4635500: 00 00 00 00 00 fc fc fc fc fc fc fc fc fc fc fc
 c4635580: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
>c4635600: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
 c4635680: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
 c4635700: 00 00 00 00 04 fc fc fc fc fc fc fc fc fc fc fc
 c4635780: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
 c4635800: 00 00 00 00 00 00 04 fc fc fc fc fc fc fc fc fc
 c4635880: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc

Signed-off-by: Andrey Ryabinin <>
[bwh: Backported to 3.2:
 - smk_of_task() and similar functions return char * not struct smack_known *
 - The callers of task_security() are quite different, but most can be changed
   to use smk_of_task_struct() just as in the upstream version
 - Use open-coded RCU locking in the one place using smk_of_forked() instead
   of smk_of_task()]
Signed-off-by: Ben Hutchings <>
7 years agoBluetooth: ath3k: Add support of AR3012 bluetooth 13d3:3423 device
Dmitry Tunin [Sat, 17 Jan 2015 21:16:51 +0000 (00:16 +0300)]
Bluetooth: ath3k: Add support of AR3012 bluetooth 13d3:3423 device

commit 033efa920a7f22a8caf7a38d851a2f451781bbf7 upstream.

Add support of 13d3:3423 device.

T: Bus=01 Lev=02 Prnt=03 Port=00 Cnt=01 Dev#= 5 Spd=12 MxCh= 0
D: Ver= 1.10 Cls=e0(wlcon) Sub=01 Prot=01 MxPS=64 #Cfgs= 1
P: Vendor=13d3 ProdID=3423 Rev= 0.01
C:* #Ifs= 2 Cfg#= 1 Atr=e0 MxPwr=100mA
A: FirstIf#= 0 IfCount= 2 Cls=e0(wlcon) Sub=01 Prot=01
I:* If#= 0 Alt= 0 #EPs= 3 Cls=e0(wlcon) Sub=01 Prot=01 Driver=btusb
E: Ad=81(I) Atr=03(Int.) MxPS= 16 Ivl=1ms
E: Ad=82(I) Atr=02(Bulk) MxPS= 64 Ivl=0ms
E: Ad=02(O) Atr=02(Bulk) MxPS= 64 Ivl=0ms
I:* If#= 1 Alt= 0 #EPs= 2 Cls=e0(wlcon) Sub=01 Prot=01 Driver=btusb
E: Ad=83(I) Atr=01(Isoc) MxPS= 0 Ivl=1ms
E: Ad=03(O) Atr=01(Isoc) MxPS= 0 Ivl=1ms
I: If#= 1 Alt= 1 #EPs= 2 Cls=e0(wlcon) Sub=01 Prot=01 Driver=btusb
E: Ad=83(I) Atr=01(Isoc) MxPS= 9 Ivl=1ms
E: Ad=03(O) Atr=01(Isoc) MxPS= 9 Ivl=1ms
I: If#= 1 Alt= 2 #EPs= 2 Cls=e0(wlcon) Sub=01 Prot=01 Driver=btusb
E: Ad=83(I) Atr=01(Isoc) MxPS= 17 Ivl=1ms
E: Ad=03(O) Atr=01(Isoc) MxPS= 17 Ivl=1ms
I: If#= 1 Alt= 3 #EPs= 2 Cls=e0(wlcon) Sub=01 Prot=01 Driver=btusb
E: Ad=83(I) Atr=01(Isoc) MxPS= 25 Ivl=1ms
E: Ad=03(O) Atr=01(Isoc) MxPS= 25 Ivl=1ms
I: If#= 1 Alt= 4 #EPs= 2 Cls=e0(wlcon) Sub=01 Prot=01 Driver=btusb
E: Ad=83(I) Atr=01(Isoc) MxPS= 33 Ivl=1ms
E: Ad=03(O) Atr=01(Isoc) MxPS= 33 Ivl=1ms
I: If#= 1 Alt= 5 #EPs= 2 Cls=e0(wlcon) Sub=01 Prot=01 Driver=btusb
E: Ad=83(I) Atr=01(Isoc) MxPS= 49 Ivl=1ms
E: Ad=03(O) Atr=01(Isoc) MxPS= 49 Ivl=1ms

Signed-off-by: Dmitry Tunin <>
Signed-off-by: Marcel Holtmann <>
Signed-off-by: Ben Hutchings <>
7 years agoTPM: Add new TPMs to the tail of the list to prevent inadvertent change of dev
David Howells [Fri, 29 Aug 2014 09:33:02 +0000 (10:33 +0100)]
TPM: Add new TPMs to the tail of the list to prevent inadvertent change of dev

commit 398a1e71dc827b994b7f2f56c7c2186fea7f8d75 upstream.

Add newly registered TPMs to the tail of the list, not the beginning, so that
things that are specifying TPM_ANY_NUM don't find that the device they're
using has inadvertently changed.  Adding a second device would break IMA, for

Signed-off-by: David Howells <>
Reviewed-by: Jason Gunthorpe <>
Signed-off-by: Peter Huewe <>
[bwh: Backported to 3.2: adjust filename]
Signed-off-by: Ben Hutchings <>
7 years agoaxonram: Fix bug in direct_access
Matthew Wilcox [Wed, 7 Jan 2015 16:04:18 +0000 (18:04 +0200)]
axonram: Fix bug in direct_access

commit 91117a20245b59f70b563523edbf998a62fc6383 upstream.

The 'pfn' returned by axonram was completely bogus, and has been since

Signed-off-by: Matthew Wilcox <>
Reviewed-by: Jan Kara <>
Reviewed-by: Mathieu Desnoyers <>
Signed-off-by: Jens Axboe <>
Signed-off-by: Ben Hutchings <>
7 years agousb: core: buffer: smallest buffer should start at ARCH_DMA_MINALIGN
Sebastian Andrzej Siewior [Fri, 5 Dec 2014 14:13:54 +0000 (15:13 +0100)]
usb: core: buffer: smallest buffer should start at ARCH_DMA_MINALIGN

commit 5efd2ea8c9f4f12916ffc8ba636792ce052f6911 upstream.

the following error pops up during "testusb -a -t 10"
| musb-hdrc dma_pool_free buffer-128, f134e000/be842000 (bad dma)
hcd_buffer_create() creates a few buffers, the smallest has 32 bytes of
size. ARCH_KMALLOC_MINALIGN is set to 64 bytes. This combo results in
hcd_buffer_alloc() returning memory which is 32 bytes aligned and it
might by identified by buffer_offset() as another buffer. This means the
buffer which is on a 32 byte boundary will not get freed, instead it
tries to free another buffer with the error message.

This patch fixes the issue by creating the smallest DMA buffer with the
smaller). This might be 32, 64 or even 128 bytes. The next three pools
will have the size 128, 512 and 2048.
In case the smallest pool is 128 bytes then we have only three pools
instead of four (and zero the first entry in the array).
The last pool size is always 2048 bytes which is the assumed PAGE_SIZE /
2 of 4096. I doubt it makes sense to continue using PAGE_SIZE / 2 where
we would end up with 8KiB buffer in case we have 16KiB pages.
Instead I think it makes sense to have a common size(s) and extend them
if there is need to.
There is a BUILD_BUG_ON() now in case someone has a minalign of more than
128 bytes.

Signed-off-by: Sebastian Andrzej Siewior <>
Acked-by: Alan Stern <>
Signed-off-by: Greg Kroah-Hartman <>
Signed-off-by: Ben Hutchings <>
7 years agokernel.h: add BUILD_BUG() macro
David Daney [Tue, 10 Jan 2012 23:07:25 +0000 (15:07 -0800)]
kernel.h: add BUILD_BUG() macro

commit 1399ff86f2a2bbacbbe68fa00c5f8c752b344723 upstream.

We can place this in definitions that we expect the compiler to remove by
dead code elimination.  If this assertion fails, we get a nice error
message at build time.

The GCC function attribute error("message") was added in version 4.3, so
we define a new macro __linktime_error(message) to expand to this for
GCC-4.3 and later.  This will give us an error diagnostic from the
compiler on the line that fails.  For other compilers
__linktime_error(message) expands to nothing, and we have to be content
with a link time error, but at least we will still get a build error.

BUILD_BUG() expands to the undefined function __build_bug_failed() and
will fail at link time if the compiler ever emits code for it.  On GCC-4.3
and later, attribute((error())) is used so that the failure will be noted
at compile time instead.

Signed-off-by: David Daney <>
Acked-by: David Rientjes <>
Cc: DM <>
Cc: Ralf Baechle <>
Acked-by: David Howells <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <>
7 years agoPCI: Generate uppercase hex for modalias var in uevent
Ricardo Ribalda Delgado [Tue, 2 Dec 2014 16:35:04 +0000 (17:35 +0100)]
PCI: Generate uppercase hex for modalias var in uevent

commit 145b3fe579db66fbe999a2bc3fd5b63dffe9636d upstream.

Some implementations of modprobe fail to load the driver for a PCI device
automatically because the "interface" part of the modalias from the kernel
is lowercase, and the modalias from file2alias is uppercase.

The "interface" is the low-order byte of the Class Code, defined in PCI
r3.0, Appendix D.  Most interface types defined in the spec do not use
alpha characters, so they won't be affected.  For example, 00h, 01h, 10h,
20h, etc. are unaffected.

Print the "interface" byte of the Class Code in uppercase hex, as we
already do for the Vendor ID, Device ID, Class, etc.

Commit 89ec3dcf17fd ("PCI: Generate uppercase hex for modalias interface
class") fixed only half of the problem.  Some udev implementations rely on
the uevent file and not the modalias file.

Fixes: d1ded203adf1 ("PCI: add MODALIAS to hotplug event for pci devices")
Fixes: 89ec3dcf17fd ("PCI: Generate uppercase hex for modalias interface class")
Signed-off-by: Ricardo Ribalda Delgado <>
Signed-off-by: Bjorn Helgaas <>
Acked-by: Greg Kroah-Hartman <>
[bwh: Backported to 3.2: adjust filename]
Signed-off-by: Ben Hutchings <>
7 years agoudf: Check length of extended attributes and allocation descriptors
Jan Kara [Wed, 7 Jan 2015 12:49:08 +0000 (13:49 +0100)]
udf: Check length of extended attributes and allocation descriptors

commit 23b133bdc452aa441fcb9b82cbf6dd05cfd342d0 upstream.

Check length of extended attributes and allocation descriptors when
loading inodes from disk. Otherwise corrupted filesystems could confuse
the code and make the kernel oops.

Reported-by: Carl Henrik Lunde <>
Signed-off-by: Jan Kara <>
[bwh: Backported to 3.16: use make_bad_inode() instead of returning error]
Signed-off-by: Ben Hutchings <>
7 years agoudf: Remove repeated loads blocksize
Jan Kara [Wed, 7 Jan 2015 12:46:16 +0000 (13:46 +0100)]
udf: Remove repeated loads blocksize

commit 79144954278d4bb5989f8b903adcac7a20ff2a5a upstream.

Store blocksize in a local variable in udf_fill_inode() since it is used
a lot of times.

Signed-off-by: Jan Kara <>
[bwh: Needed for the following fix. Backported to 3.16: adjust context.]
Signed-off-by: Ben Hutchings <>
7 years agohx4700: regulator: declare full constraints
Martin Vajnar [Tue, 23 Dec 2014 23:27:57 +0000 (00:27 +0100)]
hx4700: regulator: declare full constraints

commit a52d209336f8fc7483a8c7f4a8a7d2a8e1692a6c upstream.

Since the removal of CONFIG_REGULATOR_DUMMY option, the touchscreen stopped
working. This patch enables the "replacement" for REGULATOR_DUMMY and
allows the touchscreen to work even though there is no regulator for "vcc".

Signed-off-by: Martin Vajnar <>
Signed-off-by: Robert Jarzmik <>
Signed-off-by: Ben Hutchings <>
7 years agoARM: pxa: add regulator_has_full_constraints to spitz board file
Dmitry Eremin-Solenikov [Thu, 4 Dec 2014 11:10:02 +0000 (14:10 +0300)]
ARM: pxa: add regulator_has_full_constraints to spitz board file

commit baad2dc49c5d970ea881d92981a1b76c94a7b7a1 upstream.

Add regulator_has_full_constraints() call to spitz board file to let
regulator core know that we do not have any additional regulators left.
This lets it substitute unprovided regulators with dummy ones.

This fixes the following warnings that can be seen on spitz if
regulators are enabled:

ads7846 spi2.0: unable to get regulator: -517
spi spi2.0: Driver ads7846 requests probe deferral

Signed-off-by: Dmitry Eremin-Solenikov <>
Acked-by: Mark Brown <>
Signed-off-by: Robert Jarzmik <>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <>
7 years agoARM: pxa: add regulator_has_full_constraints to poodle board file
Dmitry Eremin-Solenikov [Thu, 4 Dec 2014 11:10:01 +0000 (14:10 +0300)]
ARM: pxa: add regulator_has_full_constraints to poodle board file

commit 9bc78f32c2e430aebf6def965b316aa95e37a20c upstream.

Add regulator_has_full_constraints() call to poodle board file to let
regulator core know that we do not have any additional regulators left.
This lets it substitute unprovided regulators with dummy ones.

This fixes the following warnings that can be seen on poodle if
regulators are enabled:

ads7846 spi1.0: unable to get regulator: -517
spi spi1.0: Driver ads7846 requests probe deferral
wm8731 0-001b: Failed to get supply 'AVDD': -517
wm8731 0-001b: Failed to request supplies: -517
wm8731 0-001b: ASoC: failed to probe component -517

Signed-off-by: Dmitry Eremin-Solenikov <>
Acked-by: Mark Brown <>
Signed-off-by: Robert Jarzmik <>
Signed-off-by: Ben Hutchings <>
7 years agoARM: pxa: add regulator_has_full_constraints to corgi board file
Dmitry Eremin-Solenikov [Thu, 4 Dec 2014 11:10:00 +0000 (14:10 +0300)]
ARM: pxa: add regulator_has_full_constraints to corgi board file

commit 271e80176aae4e5b481f4bb92df9768c6075bbca upstream.

Add regulator_has_full_constraints() call to corgi board file to let
regulator core know that we do not have any additional regulators left.
This lets it substitute unprovided regulators with dummy ones.

This fixes the following warnings that can be seen on corgi if
regulators are enabled:

ads7846 spi1.0: unable to get regulator: -517
spi spi1.0: Driver ads7846 requests probe deferral
wm8731 0-001b: Failed to get supply 'AVDD': -517
wm8731 0-001b: Failed to request supplies: -517
wm8731 0-001b: ASoC: failed to probe component -517
corgi-audio corgi-audio: ASoC: failed to instantiate card -517

Signed-off-by: Dmitry Eremin-Solenikov <>
Acked-by: Mark Brown <>
Signed-off-by: Robert Jarzmik <>
Signed-off-by: Ben Hutchings <>
8 years agoLinux 3.2.68 v3.2.68
Ben Hutchings [Fri, 6 Mar 2015 00:39:21 +0000 (00:39 +0000)]
Linux 3.2.68

8 years agoBluetooth: ath3k: workaround the compatibility issue with xHCI controller
Adam Lee [Wed, 28 Jan 2015 20:30:27 +0000 (15:30 -0500)]
Bluetooth: ath3k: workaround the compatibility issue with xHCI controller

commit c561a5753dd631920c4459a067d22679b3d110d6 upstream.

ath3k devices fail to load firmwares on xHCI buses, but work well on
EHCI, this might be a compatibility issue between xHCI and ath3k chips.
As my testing result, those chips will work on xHCI buses again with
this patch.

This workaround is from Qualcomm, they also did some workarounds in
Windows driver.

Signed-off-by: Adam Lee <>
Signed-off-by: Marcel Holtmann <>
Signed-off-by: Ben Hutchings <>
8 years agoipv6: fib: fix fib dump restart
Kumar Sundararajan [Thu, 24 Apr 2014 13:48:53 +0000 (09:48 -0400)]
ipv6: fib: fix fib dump restart

commit 1c2658545816088477e91860c3a645053719cb54 upstream.

When the ipv6 fib changes during a table dump, the walk is
restarted and the number of nodes dumped are skipped. But the existing
code doesn't advance to the next node after a node is skipped. This can
cause the dump to loop or produce lots of duplicates when the fib
is modified during the dump.

This change advances the walk to the next node if the current node is
skipped after a restart.

Signed-off-by: Kumar Sundararajan <>
Signed-off-by: Chris Mason <>
Signed-off-by: David S. Miller <>
Signed-off-by: Ben Hutchings <>
8 years agoipv6: fib: fix fib dump restart
Eric Dumazet [Mon, 25 Jun 2012 22:37:19 +0000 (15:37 -0700)]
ipv6: fib: fix fib dump restart

commit fa809e2fd6e317226c046202a88520962672eac0 upstream.

Commit 2bec5a369ee79576a3 (ipv6: fib: fix crash when changing large fib
while dumping it) introduced ability to restart the dump at tree root,
but failed to skip correctly a count of already dumped entries. Code
didn't match Patrick intent.

We must skip exactly the number of already dumped entries.

Note that like other /proc/net files or netlink producers, we could
still dump some duplicates entries.

Reported-by: Debabrata Banerjee <>
Reported-by: Josh Hunt <>
Signed-off-by: Eric Dumazet <>
Signed-off-by: David S. Miller <>
Signed-off-by: Ben Hutchings <>
8 years agontp: Fixup adjtimex freq validation on 32-bit systems
John Stultz [Tue, 10 Feb 2015 07:30:36 +0000 (23:30 -0800)]
ntp: Fixup adjtimex freq validation on 32-bit systems

commit 29183a70b0b828500816bd794b3fe192fce89f73 upstream.

Additional validation of adjtimex freq values to avoid
potential multiplication overflows were added in commit
5e5aeb4367b (time: adjtimex: Validate the ADJ_FREQUENCY values)

Unfortunately the patch used LONG_MAX/MIN instead of
LLONG_MAX/MIN, which was fine on 64-bit systems, but being
much smaller on 32-bit systems caused false positives
resulting in most direct frequency adjustments to fail w/

ntpd only does direct frequency adjustments at startup, so
the issue was not as easily observed there, but other time
sync applications like ptpd and chrony were more effected by
the bug.

See bugs:

This patch changes the checks to use LLONG_MAX for
clarity, and additionally the checks are disabled
on 32-bit systems since LLONG_MAX/PPM_SCALE is always
larger then the 32-bit long freq value, so multiplication
overflows aren't possible there.

Reported-by: Josh Boyer <>
Reported-by: George Joseph <>
Tested-by: George Joseph <>
Signed-off-by: John Stultz <>
Signed-off-by: Peter Zijlstra (Intel) <>
Cc: Linus Torvalds <>
Cc: Sasha Levin <>
[ Prettified the changelog and the comments a bit. ]
Signed-off-by: Ingo Molnar <>
Signed-off-by: Ben Hutchings <>
8 years agotime: adjtimex: Validate the ADJ_FREQUENCY values
Sasha Levin [Thu, 4 Dec 2014 00:25:05 +0000 (19:25 -0500)]
time: adjtimex: Validate the ADJ_FREQUENCY values

commit 5e5aeb4367b450a28f447f6d5ab57d8f2ab16a5f upstream.

Verify that the frequency value from userspace is valid and makes sense.

Unverified values can cause overflows later on.

Cc: Thomas Gleixner <>
Cc: Ingo Molnar <>
Signed-off-by: Sasha Levin <>
[jstultz: Fix up bug for negative values and drop redunent cap check]
Signed-off-by: John Stultz <>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <>
8 years agosched/rt: Reduce rq lock contention by eliminating locking of non-feasible target
Tim Chen [Fri, 12 Dec 2014 23:38:12 +0000 (15:38 -0800)]
sched/rt: Reduce rq lock contention by eliminating locking of non-feasible target

commit 80e3d87b2c5582db0ab5e39610ce3707d97ba409 upstream.

This patch adds checks that prevens futile attempts to move rt tasks
to a CPU with active tasks of equal or higher priority.

This reduces run queue lock contention and improves the performance of
a well known OLTP benchmark by 0.7%.

Signed-off-by: Tim Chen <>
Signed-off-by: Peter Zijlstra (Intel) <>
Cc: Shawn Bohrer <>
Cc: Suruchi Kadu <>
Cc: Doug Nelson<>
Cc: Steven Rostedt <>
Cc: Linus Torvalds <>
Signed-off-by: Ingo Molnar <>
[bwh: Backported to 3.2: adjust filename]
Signed-off-by: Ben Hutchings <>
8 years agomedia/rc: Send sync space information on the lirc device
Austin Lund [Thu, 24 Jul 2014 10:40:20 +0000 (07:40 -0300)]
media/rc: Send sync space information on the lirc device

commit a8f29e89f2b54fbf2c52be341f149bc195b63a8b upstream.

Userspace expects to see a long space before the first pulse is sent on
the lirc device.  Currently, if a long time has passed and a new packet
is started, the lirc codec just returns and doesn't send anything.  This
makes lircd ignore many perfectly valid signals unless they are sent in
quick sucession.  When a reset event is delivered, we cannot know
anything about the duration of the space.  But it should be safe to
assume it has been a long time and we just set the duration to maximum.

Signed-off-by: Austin Lund <>
Signed-off-by: Mauro Carvalho Chehab <>
Signed-off-by: Ben Hutchings <>
8 years agostaging: comedi: cb_pcidas64: fix incorrect AI range code handling
Ian Abbott [Fri, 27 Feb 2015 18:52:08 +0000 (18:52 +0000)]
staging: comedi: cb_pcidas64: fix incorrect AI range code handling

commit be8e89087ec2d2c8a1ad1e3db64bf4efdfc3c298 upstream.

The hardware range code values and list of valid ranges for the AI
subdevice is incorrect for several supported boards.  The hardware range
code values for all boards except PCI-DAS4020/12 is determined by
calling `ai_range_bits_6xxx()` based on the maximum voltage of the range
and whether it is bipolar or unipolar, however it only returns the
correct hardware range code for the PCI-DAS60xx boards.  For
PCI-DAS6402/16 (and /12) it returns the wrong code for the unipolar
ranges.  For PCI-DAS64/Mx/16 it returns the wrong code for all the
ranges and the comedi range table is incorrect.

Change `ai_range_bits_6xxx()` to use a look-up table pointed to by new
member `ai_range_codes` of `struct pcidas64_board` to map the comedi
range table indices to the hardware range codes.  Use a new comedi range
table for the PCI-DAS64/Mx/16 boards (and the commented out variants).

Signed-off-by: Ian Abbott <>
Signed-off-by: Greg Kroah-Hartman <>
Signed-off-by: Ben Hutchings <>
8 years agoDrivers: hv: vmbus: incorrect device name is printed when child device is unregistered
Fernando Soto [Fri, 14 Jun 2013 23:13:35 +0000 (23:13 +0000)]
Drivers: hv: vmbus: incorrect device name is printed when child device is unregistered

commit 84672369ffb98a51d4ddf74c20a23636da3ad615 upstream.

Whenever a device is unregistered in vmbus_device_unregister (drivers/hv/vmbus_drv.c), the device name in the log message may contain garbage as the memory has already been freed by the time pr_info is called. Log example:
 [ 3149.170475] hv_vmbus: child device àõsèè0_5 unregistered

By logging the message just before calling device_unregister, the correct device name is printed:
[ 3145.034652] hv_vmbus: child device vmbus_0_5 unregistered

Also changing register & unregister messages to debug to avoid unnecessarily cluttering the kernel log.

Signed-off-by: Fernando M Soto <>
Signed-off-by: K. Y. Srinivasan <>
Signed-off-by: Greg Kroah-Hartman <>
Signed-off-by: Ben Hutchings <>
8 years agonilfs2: fix deadlock of segment constructor over I_SYNC flag
Ryusuke Konishi [Thu, 5 Feb 2015 20:25:20 +0000 (12:25 -0800)]
nilfs2: fix deadlock of segment constructor over I_SYNC flag

commit 7ef3ff2fea8bf5e4a21cef47ad87710a3d0fdb52 upstream.

Nilfs2 eventually hangs in a stress test with fsstress program.  This
issue was caused by the following deadlock over I_SYNC flag between
nilfs_segctor_thread() and writeback_sb_inodes():

                inode_wait_for_writeback()  * wait for I_SYNC flag

     * set I_SYNC flag on inode->i_state
               * wait for completion of segment constructor
       * clear I_SYNC flag after __writeback_single_inode() completed

writeback_sb_inodes() calls do_writepages() for dirty inodes after
setting I_SYNC flag on inode->i_state.  do_writepages() in turn calls
nilfs_writepages(), which can run segment constructor and wait for its
completion.  On the other hand, segment constructor calls iput(), which
can call evict() and wait for the I_SYNC flag on

Since segment constructor doesn't know when I_SYNC will be set, it
cannot know whether iput() will block or not unless inode->i_nlink has a
non-zero count.  We can prevent evict() from being called in iput() by
implementing sop->drop_inode(), but it's not preferable to leave inodes
with i_nlink == 0 for long periods because it even defers file
truncation and inode deallocation.  So, this instead resolves the
deadlock by calling iput() asynchronously with a workqueue for inodes
with i_nlink == 0.

Signed-off-by: Ryusuke Konishi <>
Cc: Al Viro <>
Tested-by: Ryusuke Konishi <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
Signed-off-by: Ben Hutchings <>
8 years agomm: pagewalk: call pte_hole() for VM_PFNMAP during walk_page_range
Shiraz Hashim [Thu, 5 Feb 2015 20:25:06 +0000 (12:25 -0800)]
mm: pagewalk: call pte_hole() for VM_PFNMAP during walk_page_range

commit 23aaed6659df9adfabe9c583e67a36b54e21df46 upstream.

walk_page_range() silently skips vma having VM_PFNMAP set, which leads
to undesirable behaviour at client end (who called walk_page_range).
Userspace applications get the wrong data, so the effect is like just
confusing users (if the applications just display the data) or sometimes
killing the processes (if the applications do something with
misunderstanding virtual addresses due to the wrong data.)

For example for pagemap_read, when no callbacks are called against
VM_PFNMAP vma, pagemap_read may prepare pagemap data for next virtual
address range at wrong index.

Eventually userspace may get wrong pagemap data for a task.
Corresponding to a VM_PFNMAP marked vma region, kernel may report
mappings from subsequent vma regions.  User space in turn may account
more pages (than really are) to the task.

In my case I was using procmem, procrack (Android utility) which uses
pagemap interface to account RSS pages of a task.  Due to this bug it
was giving a wrong picture for vmas (with VM_PFNMAP set).

Fixes: a9ff785e4437 ("mm/pagewalk.c: walk_page_range should avoid VM_PFNMAP areas")
Signed-off-by: Shiraz Hashim <>
Acked-by: Naoya Horiguchi <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
Signed-off-by: Ben Hutchings <>
8 years agonet: sctp: fix passing wrong parameter header to param_type2af in sctp_process_param
Saran Maruti Ramanara [Thu, 29 Jan 2015 10:05:58 +0000 (11:05 +0100)]
net: sctp: fix passing wrong parameter header to param_type2af in sctp_process_param

commit cfbf654efc6d78dc9812e030673b86f235bf677d upstream.

When making use of RFC5061, section 4.2.4. for setting the primary IP
address, we're passing a wrong parameter header to param_type2af(),
resulting always in NULL being returned.

At this point, param.p points to a sctp_addip_param struct, containing
a sctp_paramhdr (type = 0xc004, length = var), and crr_id as a correlation
id. Followed by that, as also presented in RFC5061 section 4.2.4., comes
the actual sctp_addr_param, which also contains a sctp_paramhdr, but
this time with the correct type SCTP_PARAM_IPV{4,6}_ADDRESS that
param_type2af() can make use of. Since we already hold a pointer to
addr_param from previous line, just reuse it for param_type2af().

Fixes: d6de3097592b ("[SCTP]: Add the handling of "Set Primary IP Address" parameter to INIT")
Signed-off-by: Saran Maruti Ramanara <>
Signed-off-by: Daniel Borkmann <>
Acked-by: Vlad Yasevich <>
Acked-by: Neil Horman <>
Signed-off-by: David S. Miller <>
Signed-off-by: Ben Hutchings <>
8 years agogpio: sysfs: fix memory leak in gpiod_sysfs_set_active_low
Johan Hovold [Mon, 26 Jan 2015 11:02:46 +0000 (12:02 +0100)]
gpio: sysfs: fix memory leak in gpiod_sysfs_set_active_low

commit 49d2ca84e433dab854c7a866bc6add09cfab682d upstream.

Fix memory leak in the gpio sysfs interface due to failure to drop
reference to device returned by class_find_device when setting the
gpio-line polarity.

Fixes: 0769746183ca ("gpiolib: add support for changing value polarity
in sysfs")
Signed-off-by: Johan Hovold <>
Signed-off-by: Linus Walleij <>
[bwh: Backported to 3.2: adjust filename]
Signed-off-by: Ben Hutchings <>
8 years agogpio: sysfs: fix memory leak in gpiod_export_link
Johan Hovold [Mon, 26 Jan 2015 11:02:45 +0000 (12:02 +0100)]
gpio: sysfs: fix memory leak in gpiod_export_link

commit 0f303db08df0df9bd0966443ad6001e63960af16 upstream.

Fix memory leak in the gpio sysfs interface due to failure to drop
reference to device returned by class_find_device when creating a link.

Fixes: a4177ee7f1a8 ("gpiolib: allow exported GPIO nodes to be named
using sysfs links")
Signed-off-by: Johan Hovold <>
Signed-off-by: Linus Walleij <>
[bwh: Backported to 3.2: adjust filename]
Signed-off-by: Ben Hutchings <>
8 years agoMIPS: Fix kernel lockup or crash after CPU offline/online
Hemmo Nieminen [Thu, 15 Jan 2015 21:01:59 +0000 (23:01 +0200)]
MIPS: Fix kernel lockup or crash after CPU offline/online

commit c7754e75100ed5e3068ac5085747f2bfc386c8d6 upstream.

As printk() invocation can cause e.g. a TLB miss, printk() cannot be
called before the exception handlers have been properly initialized.
This can happen e.g. when netconsole has been loaded as a kernel module
and the TLB table has been cleared when a CPU was offline.

Call cpu_report() in start_secondary() only after the exception handlers
have been initialized to fix this.

Without the patch the kernel will randomly either lockup or crash
after a CPU is onlined and the console driver is a module.

Signed-off-by: Hemmo Nieminen <>
Signed-off-by: Aaro Koskinen <>
Cc: David Daney <>
Signed-off-by: Ralf Baechle <>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <>
8 years agocaif: remove wrong dev_net_set() call
Nicolas Dichtel [Mon, 26 Jan 2015 21:28:13 +0000 (22:28 +0100)]
caif: remove wrong dev_net_set() call

commit 8997c27ec41127bf57421cc0205413d525421ddc upstream.

src_net points to the netns where the netlink message has been received. This
netns may be different from the netns where the interface is created (because
the user may add IFLA_NET_NS_[PID|FD]). In this case, src_net is the link netns.

It seems wrong to override the netns in the newlink() handler because if it
was not already src_net, it means that the user explicitly asks to create the
netdevice in another netns.

CC: Sjur Brændeland <>
CC: Dmitry Tarnyagin <>
Fixes: 8391c4aab1aa ("caif: Bugfixes in CAIF netdevice for close and flow control")
Fixes: c41254006377 ("caif-hsi: Add rtnl support")
Signed-off-by: Nicolas Dichtel <>
Signed-off-by: David S. Miller <>
[bwh: Backported to 3.2: drop the change to caif_hsi change]
Signed-off-by: Ben Hutchings <>
8 years agolib/checksum.c: fix build for generic csum_tcpudp_nofold
karl beldan [Thu, 29 Jan 2015 10:10:22 +0000 (11:10 +0100)]
lib/checksum.c: fix build for generic csum_tcpudp_nofold

commit 9ce357795ef208faa0d59894d9d119a7434e37f3 upstream.

Fixed commit added from64to32 under _#ifndef do_csum_ but used it
under _#ifndef csum_tcpudp_nofold_, breaking some builds (Fengguang's
robot reported TILEGX's). Move from64to32 under the latter.

Fixes: 150ae0e94634 ("lib/checksum.c: fix carry in csum_tcpudp_nofold")
Reported-by: kbuild test robot <>
Signed-off-by: Karl Beldan <>
Cc: Eric Dumazet <>
Cc: David S. Miller <>
Signed-off-by: David S. Miller <>
Signed-off-by: Ben Hutchings <>
8 years agolib/checksum.c: fix carry in csum_tcpudp_nofold
karl beldan [Wed, 28 Jan 2015 09:58:11 +0000 (10:58 +0100)]
lib/checksum.c: fix carry in csum_tcpudp_nofold

commit 150ae0e94634714b23919f0c333fee28a5b199d5 upstream.

The carry from the 64->32bits folding was dropped, e.g with:
saddr=0xFFFFFFFF daddr=0xFF0000FF len=0xFFFF proto=0 sum=1,
csum_tcpudp_nofold returned 0 instead of 1.

Signed-off-by: Karl Beldan <>
Cc: Al Viro <>
Cc: Eric Dumazet <>
Cc: Arnd Bergmann <>
Cc: Mike Frysinger <>
Signed-off-by: Eric Dumazet <>
Signed-off-by: David S. Miller <>
Signed-off-by: Ben Hutchings <>
8 years agoALSA: ak411x: Fix stall in work callback
Takashi Iwai [Tue, 13 Jan 2015 09:53:20 +0000 (10:53 +0100)]
ALSA: ak411x: Fix stall in work callback

commit 4161b4505f1690358ac0a9ee59845a7887336b21 upstream.

When ak4114 work calls its callback and the callback invokes
ak4114_reinit(), it stalls due to flush_delayed_work().  For avoiding
this, control the reentrance by introducing a refcount.  Also
flush_delayed_work() is replaced with cancel_delayed_work_sync().

The exactly same bug is present in ak4113.c and fixed as well.

Reported-by: Pavel Hofman <>
Acked-by: Jaroslav Kysela <>
Tested-by: Pavel Hofman <>
Signed-off-by: Takashi Iwai <>
[bwh: Backported to 3.2: snd_ak411{3,4}_reinit were previously using
 flush_delayed_work_sync() rather than flush_delayed_work()]
Signed-off-by: Ben Hutchings <>
8 years agoASoC: atmel_ssc_dai: fix start event for I2S mode
Bo Shen [Tue, 20 Jan 2015 07:43:16 +0000 (15:43 +0800)]
ASoC: atmel_ssc_dai: fix start event for I2S mode

commit a43bd7e125143b875caae6d4f9938855b440faaf upstream.

According to the I2S specification information as following:
  - WS = 0, channel 1 (left)
  - WS = 1, channel 2 (right)
So, the start event should be TF/RF falling edge.

Reported-by: Songjun Wu <>
Signed-off-by: Bo Shen <>
Signed-off-by: Mark Brown <>
Signed-off-by: Ben Hutchings <>
8 years agoMIPS: IRQ: Fix disable_irq on CPU IRQs
Felix Fietkau [Thu, 15 Jan 2015 18:05:28 +0000 (19:05 +0100)]
MIPS: IRQ: Fix disable_irq on CPU IRQs

commit a3e6c1eff54878506b2dddcc202df9cc8180facb upstream.

If the irq_chip does not define .irq_disable, any call to disable_irq
will defer disabling the IRQ until it fires while marked as disabled.
This assumes that the handler function checks for this condition, which
handle_percpu_irq does not. In this case, calling disable_irq leads to
an IRQ storm, if the interrupt fires while disabled.

This optimization is only useful when disabling the IRQ is slow, which
is not true for the MIPS CPU IRQ.

Disable this optimization by implementing .irq_disable and .irq_enable

Signed-off-by: Felix Fietkau <>
Signed-off-by: Ralf Baechle <>
Signed-off-by: Ben Hutchings <>
8 years agox86: mm/fault: Fix semaphore imbalance
Ben Hutchings [Wed, 25 Feb 2015 00:02:28 +0000 (00:02 +0000)]
x86: mm/fault: Fix semaphore imbalance

When backporting commit 33692f27597f ('vm: add VM_FAULT_SIGSEGV
handling support') I didn't notice that it depended on a recent change
to the locking context of mm_fault_error() (commit 7fb08eca4527,
'x86: mm: move mmap_sem unlock from mm_fault_error() to caller').
That isn't easily applicable to 3.2, so instead make sure we drop
mm->mmap_sem on the new branch of mm_fault_error().

Signed-off-by: Ben Hutchings <>
8 years agoPCI: quirks: Fix backport of quirk_io()
Ben Hutchings [Mon, 23 Feb 2015 21:37:52 +0000 (21:37 +0000)]
PCI: quirks: Fix backport of quirk_io()

Commit 06cf35f903aa ('PCI: Handle read-only BARs on AMD CS553x
devices') added the function quirk_io() which calls

Prior to Linux 3.14, pcibios_bus_to_resource() takes a pointer to
struct pci_dev and looks up the device's bus itself, so we need
to pass dev not dev->bus.

Signed-off-by: Ben Hutchings <>
8 years agoLinux 3.2.67 v3.2.67
Ben Hutchings [Fri, 20 Feb 2015 00:49:43 +0000 (00:49 +0000)]
Linux 3.2.67

8 years agoPCI: Handle read-only BARs on AMD CS553x devices
Myron Stowe [Tue, 3 Feb 2015 23:01:24 +0000 (16:01 -0700)]
PCI: Handle read-only BARs on AMD CS553x devices

commit 06cf35f903aa6da0cc8d9f81e9bcd1f7e1b534bb upstream.

Some AMD CS553x devices have read-only BARs because of a firmware or
hardware defect.  There's a workaround in quirk_cs5536_vsa(), but it no
longer works after 36e8164882ca ("PCI: Restore detection of read-only
BARs").  Prior to 36e8164882ca, we filled in res->start; afterwards we
leave it zeroed out.  The quirk only updated the size, so the driver tried
to use a region starting at zero, which didn't work.

Expand quirk_cs5536_vsa() to read the base addresses from the BARs and
hard-code the sizes.

On Nix's system BAR 2's read-only value is 0x6200.  Prior to 36e8164882ca,
we interpret that as a 512-byte BAR based on the lowest-order bit set.  Per
datasheet sec 5.6.1, that BAR (MFGPT) requires only 64 bytes; use that to
avoid clearing any address bits if a platform uses only 64-byte alignment.

[bhelgaas: changelog, reduce BAR 2 size to 64]
Fixes: 36e8164882ca ("PCI: Restore detection of read-only BARs")
Reported-and-tested-by: Nix <>
Signed-off-by: Myron Stowe <>
Signed-off-by: Bjorn Helgaas <>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <>
8 years agoKVM: x86: SYSENTER emulation is broken
Nadav Amit [Thu, 1 Jan 2015 21:11:11 +0000 (23:11 +0200)]
KVM: x86: SYSENTER emulation is broken

commit f3747379accba8e95d70cec0eae0582c8c182050 upstream.

SYSENTER emulation is broken in several ways:
1. It misses the case of 16-bit code segments completely (CVE-2015-0239).
2. MSR_IA32_SYSENTER_CS is checked in 64-bit mode incorrectly (bits 0 and 1 can
   still be set without causing #GP).
3. MSR_IA32_SYSENTER_EIP and MSR_IA32_SYSENTER_ESP are not masked in
4. There is some unneeded code.

Fix it.

Signed-off-by: Nadav Amit <>
Signed-off-by: Paolo Bonzini <>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <>
8 years agoKVM: x86 emulator: reject SYSENTER in compatibility mode on AMD guests
Avi Kivity [Wed, 1 Feb 2012 10:23:21 +0000 (12:23 +0200)]
KVM: x86 emulator: reject SYSENTER in compatibility mode on AMD guests

commit 1a18a69b762374c423305772500f36eb8984ca52 upstream.

If the guest thinks it's an AMD, it will not have prepared the SYSENTER MSRs,
and if the guest executes SYSENTER in compatibility mode, it will fails.

Detect this condition and #UD instead, like the spec says.

Signed-off-by: Avi Kivity <>
Signed-off-by: Ben Hutchings <>
8 years agonetfilter: conntrack: disable generic tracking for known protocols
Florian Westphal [Fri, 26 Sep 2014 09:35:42 +0000 (11:35 +0200)]
netfilter: conntrack: disable generic tracking for known protocols

commit db29a9508a9246e77087c5531e45b2c88ec6988b upstream.

Given following iptables ruleset:

-A FORWARD -m sctp --dport 9 -j ACCEPT
-A FORWARD -p tcp --dport 80 -j ACCEPT
-A FORWARD -p tcp -m conntrack -m state ESTABLISHED,RELATED -j ACCEPT

One would assume that this allows SCTP on port 9 and TCP on port 80.
Unfortunately, if the SCTP conntrack module is not loaded, this allows
*all* SCTP communication, to pass though, i.e. -p sctp -j ACCEPT,
which we think is a security issue.

This is because on the first SCTP packet on port 9, we create a dummy
"generic l4" conntrack entry without any port information (since
conntrack doesn't know how to extract this information).

All subsequent packets that are unknown will then be in established
state since they will fallback to proto_generic and will match the
'generic' entry.

Our originally proposed version [1] completely disabled generic protocol
tracking, but Jozsef suggests to not track protocols for which a more
suitable helper is available, hence we now mitigate the issue for in
tree known ct protocol helpers only, so that at least NAT and direction
information will still be preserved for others.


Joint work with Daniel Borkmann.

Signed-off-by: Florian Westphal <>
Signed-off-by: Daniel Borkmann <>
Acked-by: Jozsef Kadlecsik <>
Signed-off-by: Pablo Neira Ayuso <>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <>
8 years agosplice: Apply generic position and size checks to each write
Ben Hutchings [Thu, 29 Jan 2015 02:50:33 +0000 (02:50 +0000)]
splice: Apply generic position and size checks to each write

We need to check the position and size of file writes against various
limits, using generic_write_check().  This was not being done for
the splice write path.  It was fixed upstream by commit 8d0207652cbe
("->splice_write() via ->write_iter()") but we can't apply that.


Signed-off-by: Ben Hutchings <>
8 years agovfs: Fix vfsmount_lock imbalance in path_init()
Ben Hutchings [Mon, 16 Feb 2015 03:21:17 +0000 (03:21 +0000)]
vfs: Fix vfsmount_lock imbalance in path_init()

When backporting commit 4023bfc9f351 ("be careful with nd->inode in
path_init() and follow_dotdot_rcu()"), I failed to account for the
vfsmount_lock that is used in 3.2 but not upstream.  path_init() takes
the lock if performing RCU lookup, but must drop it if (and only if)
it subsequently fails.

Signed-off-by: Ben Hutchings <>
8 years agonet/core: Handle csum for CHECKSUM_COMPLETE VXLAN forwarding
Jay Vosburgh [Fri, 19 Dec 2014 23:32:00 +0000 (15:32 -0800)]
net/core: Handle csum for CHECKSUM_COMPLETE VXLAN forwarding

[ Upstream commit 2c26d34bbcc0b3f30385d5587aa232289e2eed8e ]

When using VXLAN tunnels and a sky2 device, I have experienced
checksum failures of the following type:

[ 4297.761899] eth0: hw csum failure
[ 4297.765223] Call Trace:
[ 4297.765224]  <IRQ>  [<ffffffff8172f026>] dump_stack+0x46/0x58
[ 4297.765235]  [<ffffffff8162ba52>] netdev_rx_csum_fault+0x42/0x50
[ 4297.765238]  [<ffffffff8161c1a0>] ? skb_push+0x40/0x40
[ 4297.765240]  [<ffffffff8162325c>] __skb_checksum_complete+0xbc/0xd0
[ 4297.765243]  [<ffffffff8168c602>] tcp_v4_rcv+0x2e2/0x950
[ 4297.765246]  [<ffffffff81666ca0>] ? ip_rcv_finish+0x360/0x360

These are reliably reproduced in a network topology of:

container:eth0 == host(OVS VXLAN on VLAN) == bond0 == eth0 (sky2) -> switch

When VXLAN encapsulated traffic is received from a similarly
configured peer, the above warning is generated in the receive
processing of the encapsulated packet.  Note that the warning is
associated with the container eth0.

        The skbs from sky2 have ip_summed set to CHECKSUM_COMPLETE, and
because the packet is an encapsulated Ethernet frame, the checksum
generated by the hardware includes the inner protocol and Ethernet

The receive code is careful to update the skb->csum, except in
__dev_forward_skb, as called by dev_forward_skb.  __dev_forward_skb
calls eth_type_trans, which in turn calls skb_pull_inline(skb, ETH_HLEN)
to skip over the Ethernet header, but does not update skb->csum when
doing so.

This patch resolves the problem by adding a call to
skb_postpull_rcsum to update the skb->csum after the call to

Signed-off-by: Jay Vosburgh <>
Signed-off-by: David S. Miller <>
Signed-off-by: Ben Hutchings <>
8 years agoenic: fix rx skb checksum
Govindarajulu Varadarajan [Thu, 18 Dec 2014 10:28:42 +0000 (15:58 +0530)]
enic: fix rx skb checksum

[ Upstream commit 17e96834fd35997ca7cdfbf15413bcd5a36ad448 ]

Hardware always provides compliment of IP pseudo checksum. Stack expects
whole packet checksum without pseudo checksum if CHECKSUM_COMPLETE is set.

This causes checksum error in nf & ovs.

kernel: qg-19546f09-f2: hw csum failure
kernel: CPU: 9 PID: 0 Comm: swapper/9 Tainted: GF          O--------------   3.10.0-123.8.1.el7.x86_64 #1
kernel: Hardware name: Cisco Systems Inc UCSB-B200-M3/UCSB-B200-M3, BIOS B200M3. 08/08/2014
kernel: ffff881218f40000 df68243feb35e3a8 ffff881237a43ab8 ffffffff815e237b
kernel: ffff881237a43ad0 ffffffff814cd4ca ffff8829ec71eb00 ffff881237a43af0
kernel: ffffffff814c6232 0000000000000286 ffff8829ec71eb00 ffff881237a43b00
kernel: Call Trace:
kernel: <IRQ>  [<ffffffff815e237b>] dump_stack+0x19/0x1b
kernel: [<ffffffff814cd4ca>] netdev_rx_csum_fault+0x3a/0x40
kernel: [<ffffffff814c6232>] __skb_checksum_complete_head+0x62/0x70
kernel: [<ffffffff814c6251>] __skb_checksum_complete+0x11/0x20
kernel: [<ffffffff8155a20c>] nf_ip_checksum+0xcc/0x100
kernel: [<ffffffffa049edc7>] icmp_error+0x1f7/0x35c [nf_conntrack_ipv4]
kernel: [<ffffffff814cf419>] ? netif_rx+0xb9/0x1d0
kernel: [<ffffffffa040eb7b>] ? internal_dev_recv+0xdb/0x130 [openvswitch]
kernel: [<ffffffffa04c8330>] nf_conntrack_in+0xf0/0xa80 [nf_conntrack]
kernel: [<ffffffff81509380>] ? inet_del_offload+0x40/0x40
kernel: [<ffffffffa049e302>] ipv4_conntrack_in+0x22/0x30 [nf_conntrack_ipv4]
kernel: [<ffffffff815005ca>] nf_iterate+0xaa/0xc0
kernel: [<ffffffff81509380>] ? inet_del_offload+0x40/0x40
kernel: [<ffffffff81500664>] nf_hook_slow+0x84/0x140
kernel: [<ffffffff81509380>] ? inet_del_offload+0x40/0x40
kernel: [<ffffffff81509dd4>] ip_rcv+0x344/0x380

Hardware verifies IP & tcp/udp header checksum but does not provide payload
checksum, use CHECKSUM_UNNECESSARY. Set it only if its valid IP tcp/udp packet.

Cc: Jiri Benc <>
Cc: Stefan Assmann <>
Reported-by: Sunil Choudhary <>
Signed-off-by: Govindarajulu Varadarajan <>
Reviewed-by: Jiri Benc <>
Signed-off-by: David S. Miller <>
Signed-off-by: Ben Hutchings <>
8 years agotg3: tg3_disable_ints using uninitialized mailbox value to disable interrupts
Prashant Sreedharan [Sat, 20 Dec 2014 20:16:17 +0000 (12:16 -0800)]
tg3: tg3_disable_ints using uninitialized mailbox value to disable interrupts

[ Upstream commit 05b0aa579397b734f127af58e401a30784a1e315 ]

During driver load in tg3_init_one, if the driver detects DMA activity before
intializing the chip tg3_halt is called. As part of tg3_halt interrupts are
disabled using routine tg3_disable_ints. This routine was using mailbox value
which was not initialized (default value is 0). As a result driver was writing
0x00000001 to pci config space register 0, which is the vendor id / device id.

This driver bug was exposed because of the commit a7877b17a667 (PCI: Check only
the Vendor ID to identify Configuration Request Retry). Also this issue is only
seen in older generation chipsets like 5722 because config space write to offset
0 from driver is possible. The newer generation chips ignore writes to offset 0.
Also without commit a7877b17a667, for these older chips when a GRC reset is
issued the Bootcode would reprogram the vendor id/device id, which is the reason
this bug was masked earlier.

Fixed by initializing the interrupt mailbox registers before calling tg3_halt.

Please queue for -stable.

Reported-by: Nils Holland <>
Reported-by: Marcelo Ricardo Leitner <>
Signed-off-by: Prashant Sreedharan <>
Signed-off-by: Michael Chan <>
Signed-off-by: David S. Miller <>
Signed-off-by: Ben Hutchings <>
8 years agodcache: Fix locking bugs in backported "deal with deadlock in d_walk()"
Ben Hutchings [Wed, 11 Feb 2015 03:16:35 +0000 (03:16 +0000)]
dcache: Fix locking bugs in backported "deal with deadlock in d_walk()"

Steven Rostedt reported:
> Porting -rt to the latest 3.2 stable tree I triggered this bug:
> =====================================
> [ BUG: bad unlock balance detected! ]
> -------------------------------------
> rm/1638 is trying to release lock (rcu_read_lock) at:
> [<c04fde6c>] rcu_read_unlock+0x0/0x23
> but there are no more locks to release!
> other info that might help us debug this:
> 2 locks held by rm/1638:
>  #0:  (&sb->s_type->i_mutex_key#9/1){+.+.+.}, at: [<c04f93eb>] do_rmdir+0x5f/0xd2
>  #1:  (&sb->s_type->i_mutex_key#9){+.+.+.}, at: [<c04f9329>] vfs_rmdir+0x49/0xac
> stack backtrace:
> Pid: 1638, comm: rm Not tainted 3.2.66-test-rt96+ #2
> Call Trace:
>  [<c083f390>] ? printk+0x1d/0x1f
>  [<c0463cdf>] print_unlock_inbalance_bug+0xc3/0xcd
>  [<c04653a8>] lock_release_non_nested+0x98/0x1ec
>  [<c046228d>] ? trace_hardirqs_off_caller+0x18/0x90
>  [<c0456f1c>] ? local_clock+0x2d/0x50
>  [<c04fde6c>] ? d_hash+0x2f/0x2f
>  [<c04fde6c>] ? d_hash+0x2f/0x2f
>  [<c046568e>] lock_release+0x192/0x1ad
>  [<c04fde83>] rcu_read_unlock+0x17/0x23
>  [<c04ff344>] shrink_dcache_parent+0x227/0x270
>  [<c04f9348>] vfs_rmdir+0x68/0xac
>  [<c04f9424>] do_rmdir+0x98/0xd2
>  [<c04f03ad>] ? fput+0x1a3/0x1ab
>  [<c084dd42>] ? sysenter_exit+0xf/0x1a
>  [<c0465b58>] ? trace_hardirqs_on_caller+0x118/0x149
>  [<c04fa3e0>] sys_unlinkat+0x2b/0x35
>  [<c084dd13>] sysenter_do_call+0x12/0x12
> There's a path to calling rcu_read_unlock() without calling
> rcu_read_lock() in have_submounts().
>  goto positive;
> positive:
>  if (!locked && read_seqretry(&rename_lock, seq))
>  goto rename_retry;
> rename_retry:
>  rcu_read_unlock();
> in the above path, rcu_read_lock() is never done before calling
> rcu_read_unlock();

I reviewed locking contexts in all three functions that I changed when
backporting "deal with deadlock in d_walk()".  It's actually worse
than this:

- We don't hold this_parent->d_lock at the 'positive' label in
  have_submounts(), but it is unlocked after 'rename_retry'.
- There is an rcu_read_unlock() after the 'out' label in
  select_parent(), but it's not held at the 'goto out'.

Fix all three lock imbalances.

Reported-by: Steven Rostedt <>
Signed-off-by: Ben Hutchings <>
Tested-by: Steven Rostedt <>
8 years agonetfilter: ipset: small potential read beyond the end of buffer
Dan Carpenter [Mon, 10 Nov 2014 16:11:21 +0000 (17:11 +0100)]
netfilter: ipset: small potential read beyond the end of buffer

commit 2196937e12b1b4ba139806d132647e1651d655df upstream.

We could be reading 8 bytes into a 4 byte buffer here.  It seems
harmless but adding a check is the right thing to do and it silences a
static checker warning.

Signed-off-by: Dan Carpenter <>
Acked-by: Jozsef Kadlecsik <>
Signed-off-by: Pablo Neira Ayuso <>
Signed-off-by: Ben Hutchings <>
8 years agoKEYS: close race between key lookup and freeing
Sasha Levin [Mon, 29 Dec 2014 14:39:01 +0000 (09:39 -0500)]
KEYS: close race between key lookup and freeing

commit a3a8784454692dd72e5d5d34dcdab17b4420e74c upstream.

When a key is being garbage collected, it's key->user would get put before
the ->destroy() callback is called, where the key is removed from it's
respective tracking structures.

This leaves a key hanging in a semi-invalid state which leaves a window open
for a different task to try an access key->user. An example is
find_keyring_by_name() which would dereference key->user for a key that is
in the process of being garbage collected (where key->user was freed but
->destroy() wasn't called yet - so it's still present in the linked list).

This would cause either a panic, or corrupt memory.

Fixes CVE-2014-9529.

Signed-off-by: Sasha Levin <>
Signed-off-by: David Howells <>
[bwh: Backported to 3.2: adjust indentation]
Signed-off-by: Ben Hutchings <>
8 years agofsnotify: next_i is freed during fsnotify_unmount_inodes.
Jerry Hoemann [Wed, 29 Oct 2014 21:50:22 +0000 (14:50 -0700)]
fsnotify: next_i is freed during fsnotify_unmount_inodes.

commit 6424babfd68dd8a83d9c60a5242d27038856599f upstream.

During file system stress testing on 3.10 and 3.12 based kernels, the
umount command occasionally hung in fsnotify_unmount_inodes in the
section of code:

                if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) {

As this section of code holds the global inode_sb_list_lock, eventually
the system hangs trying to acquire the lock.

Multiple crash dumps showed:

The inode->i_state == 0x60 and i_count == 0 and i_sb_list would point
back at itself.  As this is not the value of list upon entry to the
function, the kernel never exits the loop.

To help narrow down problem, the call to list_del_init in
inode_sb_list_del was changed to list_del.  This poisons the pointers in
the i_sb_list and causes a kernel to panic if it transverse a freed

Subsequent stress testing paniced in fsnotify_unmount_inodes at the
bottom of the list_for_each_entry_safe loop showing next_i had become

We believe the root cause of the problem is that next_i is being freed
during the window of time that the list_for_each_entry_safe loop
temporarily releases inode_sb_list_lock to call fsnotify and

The code in fsnotify_unmount_inodes attempts to prevent the freeing of
inode and next_i by calling __iget.  However, the code doesn't do the
__iget call on next_i

if i_count == 0 or
if i_state & (I_FREEING | I_WILL_FREE)

The patch addresses this issue by advancing next_i in the above two cases
until we either find a next_i which we can __iget or we reach the end of
the list.  This makes the handling of next_i more closely match the
handling of the variable "inode."

The time to reproduce the hang is highly variable (from hours to days.) We
ran the stress test on a 3.10 kernel with the proposed patch for a week
without failure.

During list_for_each_entry_safe, next_i is becoming free causing
the loop to never terminate.  Advance next_i in those cases where
__iget is not done.

Signed-off-by: Jerry Hoemann <>
Cc: Jeff Kirsher <>
Cc: Ken Helias <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
Signed-off-by: Ben Hutchings <>
Cc: Jan Kara <>
8 years agox86, cpu, amd: Add workaround for family 16h, erratum 793
Borislav Petkov [Tue, 14 Jan 2014 23:07:11 +0000 (00:07 +0100)]
x86, cpu, amd: Add workaround for family 16h, erratum 793

commit 3b56496865f9f7d9bcb2f93b44c63f274f08e3b6 upstream.

This adds the workaround for erratum 793 as a precaution in case not
every BIOS implements it.  This addresses CVE-2013-6885.

Erratum text:

[Revision Guide for AMD Family 16h Models 00h-0Fh Processors,
document 51810 Rev. 3.04 November 2013]

793 Specific Combination of Writes to Write Combined Memory Types and
Locked Instructions May Cause Core Hang


Under a highly specific and detailed set of internal timing
conditions, a locked instruction may trigger a timing sequence whereby
the write to a write combined memory type is not flushed, causing the
locked instruction to stall indefinitely.

Potential Effect on System

Processor core hang.

Suggested Workaround

BIOS should set MSR
C001_1020[15] = 1b.

Fix Planned

No fix planned

[ hpa: updated description, fixed typo in MSR name ]

Signed-off-by: Borislav Petkov <>
Tested-by: Aravind Gopalakrishnan <>
Signed-off-by: H. Peter Anvin <>
[bwh: Backported to 3.2:
 - Adjust filename
 - Venkatesh Srinivas pointed out we should use {rd,wr}msrl_safe() to
   avoid crashing on KVM.  This was fixed upstream by commit 8f86a7373a1c
   ("x86, AMD: Convert to the new bit access MSR accessors") but that's too
   much trouble to backport.  Here we must use {rd,wr}msrl_amd_safe().]
Signed-off-by: Ben Hutchings <>
Cc: Moritz Muehlenhoff <>
Cc: Venkatesh Srinivas <>
8 years agos390/3215: fix tty output containing tabs
Martin Schwidefsky [Wed, 13 Aug 2014 10:01:30 +0000 (12:01 +0200)]
s390/3215: fix tty output containing tabs

commit e512d56c799517f33b301d81e9a5e0ebf30c2d1e upstream.

git commit 37f81fa1f63ad38e16125526bb2769ae0ea8d332
"n_tty: do O_ONLCR translation as a single write"
surfaced a bug in the 3215 device driver. In combination this
broke tab expansion for tty ouput.

The cause is an asymmetry in the behaviour of tty3215_ops->write
vs tty3215_ops->put_char. The put_char function scans for '\t'
but the write function does not.

As the driver has logic for the '\t' expansion remove XTABS
from c_oflag of the initial termios as well.

Reported-by: Stephen Powell <>
Signed-off-by: Martin Schwidefsky <>
Signed-off-by: Ben Hutchings <>
8 years agoACPI / EC: Fix regression due to conflicting firmware behavior between Samsung and...
Lv Zheng [Wed, 29 Oct 2014 03:33:49 +0000 (11:33 +0800)]
ACPI / EC: Fix regression due to conflicting firmware behavior between Samsung and Acer.

commit 79149001105f18bd2285ada109f9229ea24a7571 upstream.

It is reported that Samsung laptops that need to poll events are broken by
the following commit:
 Commit 3afcf2ece453e1a8c2c6de19cdf06da3772a1b08
 Subject: ACPI / EC: Add support to disallow QR_EC to be issued when SCI_EVT isn't set

The behaviors of the 2 vendor firmwares are conflict:
 1. Acer: OSPM shouldn't issue QR_EC unless SCI_EVT is set, firmware
         automatically sets SCI_EVT as long as there is event queued up.
 2. Samsung: OSPM should issue QR_EC whatever SCI_EVT is set, firmware
            returns 0 when there is no event queued up.

This patch is a quick fix to distinguish the behaviors to make Acer
behavior only effective for Acer EC firmware so that the breakages on
Samsung EC firmware can be avoided.

Fixes: 3afcf2ece453 (ACPI / EC: Add support to disallow QR_EC to be issued ...)
Reported-and-tested-by: Ortwin Glück <>
Signed-off-by: Lv Zheng <>
[ rjw : Subject ]
Signed-off-by: Rafael J. Wysocki <>
Signed-off-by: Ben Hutchings <>
Cc: Kamal Mostafa <>
8 years agoRevert "x86, 64bit, mm: Mark data/bss/brk to nx"
Ben Hutchings [Sun, 11 Jan 2015 20:32:41 +0000 (20:32 +0000)]
Revert "x86, 64bit, mm: Mark data/bss/brk to nx"

This reverts commit e105c8187b7101e8a8a54ac0218c9d9c9463c636 which
was commit 72212675d1c96f5db8ec6fb35701879911193158 upstream.

This caused suspend/resume to stop working on at least some systems -
specifically, the system would reboot when woken.

Signed-off-by: Ben Hutchings <>
Cc: Steven Rostedt <>
8 years agoRevert "x86, mm: Set NX across entire PMD at boot"
Ben Hutchings [Sun, 11 Jan 2015 00:30:55 +0000 (00:30 +0000)]
Revert "x86, mm: Set NX across entire PMD at boot"

This reverts commit a5c187d92d2ce30315f333b9dff33af832e8b443 which
was commit 45e2a9d4701d8c624d4a4bcdd1084eae31e92f58 upstream.

The previous commit caused suspend/resume to stop working on at least
some systems - specifically, the system would reboot when woken.

Signed-off-by: Ben Hutchings <>
Cc: Steven Rostedt <>
8 years agovm: make stack guard page errors return VM_FAULT_SIGSEGV rather than SIGBUS
Linus Torvalds [Thu, 29 Jan 2015 19:15:17 +0000 (11:15 -0800)]
vm: make stack guard page errors return VM_FAULT_SIGSEGV rather than SIGBUS

commit 9c145c56d0c8a0b62e48c8d71e055ad0fb2012ba upstream.

The stack guard page error case has long incorrectly caused a SIGBUS
rather than a SIGSEGV, but nobody actually noticed until commit
fee7e49d4514 ("mm: propagate error from stack expansion even for guard
page") because that error case was never actually triggered in any
normal situations.

Now that we actually report the error, people noticed the wrong signal
that resulted.  So far, only the test suite of libsigsegv seems to have
actually cared, but there are real applications that use libsigsegv, so
let's not wait for any of those to break.

Reported-and-tested-by: Takashi Iwai <>
Tested-by: Jan Engelhardt <>
Acked-by: Heiko Carstens <> # "s390 still compiles and boots"
Signed-off-by: Linus Torvalds <>
Signed-off-by: Ben Hutchings <>
8 years agovm: add VM_FAULT_SIGSEGV handling support
Linus Torvalds [Thu, 29 Jan 2015 18:51:32 +0000 (10:51 -0800)]
vm: add VM_FAULT_SIGSEGV handling support

commit 33692f27597fcab536d7cbbcc8f52905133e4aa7 upstream.

The core VM already knows about VM_FAULT_SIGBUS, but cannot return a
"you should SIGSEGV" error, because the SIGSEGV case was generally
handled by the caller - usually the architecture fault handler.

That results in lots of duplication - all the architecture fault
handlers end up doing very similar "look up vma, check permissions, do
retries etc" - but it generally works.  However, there are cases where
the VM actually wants to SIGSEGV, and applications _expect_ SIGSEGV.

In particular, when accessing the stack guard page, libsigsegv expects a
SIGSEGV.  And it usually got one, because the stack growth is handled by
that duplicated architecture fault handler.

However, when the generic VM layer started propagating the error return
from the stack expansion in commit fee7e49d4514 ("mm: propagate error
from stack expansion even for guard page"), that now exposed the
existing VM_FAULT_SIGBUS result to user space.  And user space really
expected SIGSEGV, not SIGBUS.

To fix that case, we need to add a VM_FAULT_SIGSEGV, and teach all those
duplicate architecture fault handlers about it.  They all already have
the code to handle SIGSEGV, so it's about just tying that new return
value to the existing code, but it's all a bit annoying.

This is the mindless minimal patch to do this.  A more extensive patch
would be to try to gather up the mostly shared fault handling logic into
one generic helper routine, and long-term we really should do that

Just from this patch, you can generally see that most architectures just
copied (directly or indirectly) the old x86 way of doing things, but in
the meantime that original x86 model has been improved to hold the VM
semaphore for shorter times etc and to handle VM_FAULT_RETRY and other
"newer" things, so it would be a good idea to bring all those
improvements to the generic case and teach other architectures about
them too.

Reported-and-tested-by: Takashi Iwai <>
Tested-by: Jan Engelhardt <>
Acked-by: Heiko Carstens <> # "s390 still compiles and boots"
Signed-off-by: Linus Torvalds <>
[bwh: Backported to 3.2:
 - Adjust filenames, context
 - Drop arc, metag, nios2 and lustre changes
 - For sh, patch both 32-bit and 64-bit implementations to use goto bad_area
 - For s390, pass int_code and trans_exc_code as arguments to do_no_context()
   and do_sigsegv()]
Signed-off-by: Ben Hutchings <>
8 years agonet: sctp: fix slab corruption from use after free on INIT collisions
Daniel Borkmann [Thu, 22 Jan 2015 17:26:54 +0000 (18:26 +0100)]
net: sctp: fix slab corruption from use after free on INIT collisions

commit 600ddd6825543962fb807884169e57b580dba208 upstream.

When hitting an INIT collision case during the 4WHS with AUTH enabled, as
already described in detail in commit 1be9a950c646 ("net: sctp: inherit
auth_capable on INIT collisions"), it can happen that we occasionally
still remotely trigger the following panic on server side which seems to
have been uncovered after the fix from commit 1be9a950c646 ...

[  533.876389] BUG: unable to handle kernel paging request at 00000000ffffffff
[  533.913657] IP: [<ffffffff811ac385>] __kmalloc+0x95/0x230
[  533.940559] PGD 5030f2067 PUD 0
[  533.957104] Oops: 0000 [#1] SMP
[  533.974283] Modules linked in: sctp mlx4_en [...]
[  534.939704] Call Trace:
[  534.951833]  [<ffffffff81294e30>] ? crypto_init_shash_ops+0x60/0xf0
[  534.984213]  [<ffffffff81294e30>] crypto_init_shash_ops+0x60/0xf0
[  535.015025]  [<ffffffff8128c8ed>] __crypto_alloc_tfm+0x6d/0x170
[  535.045661]  [<ffffffff8128d12c>] crypto_alloc_base+0x4c/0xb0
[  535.074593]  [<ffffffff8160bd42>] ? _raw_spin_lock_bh+0x12/0x50
[  535.105239]  [<ffffffffa0418c11>] sctp_inet_listen+0x161/0x1e0 [sctp]
[  535.138606]  [<ffffffff814e43bd>] SyS_listen+0x9d/0xb0
[  535.166848]  [<ffffffff816149a9>] system_call_fastpath+0x16/0x1b

... or depending on the the application, for example this one:

[ 1370.026490] BUG: unable to handle kernel paging request at 00000000ffffffff
[ 1370.026506] IP: [<ffffffff811ab455>] kmem_cache_alloc+0x75/0x1d0
[ 1370.054568] PGD 633c94067 PUD 0
[ 1370.070446] Oops: 0000 [#1] SMP
[ 1370.085010] Modules linked in: sctp kvm_amd kvm [...]
[ 1370.963431] Call Trace:
[ 1370.974632]  [<ffffffff8120f7cf>] ? SyS_epoll_ctl+0x53f/0x960
[ 1371.000863]  [<ffffffff8120f7cf>] SyS_epoll_ctl+0x53f/0x960
[ 1371.027154]  [<ffffffff812100d3>] ? anon_inode_getfile+0xd3/0x170
[ 1371.054679]  [<ffffffff811e3d67>] ? __alloc_fd+0xa7/0x130
[ 1371.080183]  [<ffffffff816149a9>] system_call_fastpath+0x16/0x1b

With slab debugging enabled, we can see that the poison has been overwritten:

[  669.826368] BUG kmalloc-128 (Tainted: G        W     ): Poison overwritten
[  669.826385] INFO: 0xffff880228b32e50-0xffff880228b32e50. First byte 0x6a instead of 0x6b
[  669.826414] INFO: Allocated in sctp_auth_create_key+0x23/0x50 [sctp] age=3 cpu=0 pid=18494
[  669.826424]  __slab_alloc+0x4bf/0x566
[  669.826433]  __kmalloc+0x280/0x310
[  669.826453]  sctp_auth_create_key+0x23/0x50 [sctp]
[  669.826471]  sctp_auth_asoc_create_secret+0xcb/0x1e0 [sctp]
[  669.826488]  sctp_auth_asoc_init_active_key+0x68/0xa0 [sctp]
[  669.826505]  sctp_do_sm+0x29d/0x17c0 [sctp] [...]
[  669.826629] INFO: Freed in kzfree+0x31/0x40 age=1 cpu=0 pid=18494
[  669.826635]  __slab_free+0x39/0x2a8
[  669.826643]  kfree+0x1d6/0x230
[  669.826650]  kzfree+0x31/0x40
[  669.826666]  sctp_auth_key_put+0x19/0x20 [sctp]
[  669.826681]  sctp_assoc_update+0x1ee/0x2d0 [sctp]
[  669.826695]  sctp_do_sm+0x674/0x17c0 [sctp]

Since this only triggers in some collision-cases with AUTH, the problem at
heart is that sctp_auth_key_put() on asoc->asoc_shared_key is called twice
when having refcnt 1, once directly in sctp_assoc_update() and yet again
from within sctp_auth_asoc_init_active_key() via sctp_assoc_update() on
the already kzfree'd memory, which is also consistent with the observation
of the poison decrease from 0x6b to 0x6a (note: the overwrite is detected
at a later point in time when poison is checked on new allocation).

Reference counting of auth keys revisited:

Shared keys for AUTH chunks are being stored in endpoints and associations
in endpoint_shared_keys list. On endpoint creation, a null key is being
added; on association creation, all endpoint shared keys are being cached
and thus cloned over to the association. struct sctp_shared_key only holds
a pointer to the actual key bytes, that is, struct sctp_auth_bytes which
keeps track of users internally through refcounting. Naturally, on assoc
or enpoint destruction, sctp_shared_key are being destroyed directly and
the reference on sctp_auth_bytes dropped.

User space can add keys to either list via setsockopt(2) through struct
sctp_authkey and by passing that to sctp_auth_set_key() which replaces or
adds a new auth key. There, sctp_auth_create_key() creates a new sctp_auth_bytes
with refcount 1 and in case of replacement drops the reference on the old
sctp_auth_bytes. A key can be set active from user space through setsockopt()
on the id via sctp_auth_set_active_key(), which iterates through either
endpoint_shared_keys and in case of an assoc, invokes (one of various places)

sctp_auth_asoc_init_active_key() computes the actual secret from local's
and peer's random, hmac and shared key parameters and returns a new key
directly as sctp_auth_bytes, that is asoc->asoc_shared_key, plus drops
the reference if there was a previous one. The secret, which where we
eventually double drop the ref comes from sctp_auth_asoc_set_secret() with
intitial refcount of 1, which also stays unchanged eventually in
sctp_assoc_update(). This key is later being used for crypto layer to
set the key for the hash in crypto_hash_setkey() from sctp_auth_calculate_hmac().

To close the loop: asoc->asoc_shared_key is freshly allocated secret
material and independant of the sctp_shared_key management keeping track
of only shared keys in endpoints and assocs. Hence, also commit 4184b2a79a76
("net: sctp: fix memory leak in auth key management") is independant of
this bug here since it concerns a different layer (though same structures
being used eventually). asoc->asoc_shared_key is reference dropped correctly
on assoc destruction in sctp_association_free() and when active keys are
being replaced in sctp_auth_asoc_init_active_key(), it always has a refcount
of 1. Hence, it's freed prematurely in sctp_assoc_update(). Simple fix is
to remove that sctp_auth_key_put() from there which fixes these panics.

Fixes: 730fc3d05cd4 ("[SCTP]: Implete SCTP-AUTH parameter processing")
Signed-off-by: Daniel Borkmann <>
Acked-by: Vlad Yasevich <>
Acked-by: Neil Horman <>
Signed-off-by: David S. Miller <>
Signed-off-by: Ben Hutchings <>
8 years agoALSA: seq-dummy: remove deadlock-causing events on close
Clemens Ladisch [Sun, 25 Jan 2015 13:34:29 +0000 (14:34 +0100)]
ALSA: seq-dummy: remove deadlock-causing events on close

commit 0767e95bb96d7fdddcd590fb809e6975d93aebc5 upstream.

When the last subscriber to a "Through" port has been removed, the
subscribed destination ports might still be active, so it would be
wrong to send "all sounds off" and "reset controller" events to them.
The proper place for such a shutdown would be the closing of the actual
MIDI port (and close_substream() in rawmidi.c already can do this).

This also fixes a deadlock when dummy_unuse() tries to send events to
its own port that is already locked because it is being freed.

Reported-by: Peter Billam <>
Signed-off-by: Clemens Ladisch <>
Signed-off-by: Takashi Iwai <>
Signed-off-by: Ben Hutchings <>
8 years agodrm/i915: Only fence tiled region of object.
Bob Paauwe [Thu, 18 Dec 2014 17:51:26 +0000 (09:51 -0800)]
drm/i915: Only fence tiled region of object.

commit af1a7301c7cf8912dca03065d448c4437c5c239f upstream.

When creating a fence for a tiled object, only fence the area that
makes up the actual tiles.  The object may be larger than the tiled
area and if we allow those extra addresses to be fenced, they'll
get converted to addresses beyond where the object is mapped. This
opens up the possiblity of writes beyond the end of object.

To prevent this, we adjust the size of the fence to only encompass
the area that makes up the actual tiles.  The extra space is considered
un-tiled and now behaves as if it was a linear object.

Testcase: igt/gem_tiled_fence_overflow
Reported-by: Dan Hettena <>
Signed-off-by: Bob Paauwe <>
Reviewed-by: Daniel Vetter <>
Signed-off-by: Jani Nikula <>
[bwh: Backported to 3.2:
 - Adjust context, indentation
 - Apply to both i965_write_fence_reg() and sandybridge_write_fence_reg(),
   which have been combined into one function upstream]
Signed-off-by: Ben Hutchings <>
8 years agoUSB: Add OTG PET device to TPL
Macpaul Lin [Fri, 23 Jan 2015 06:39:02 +0000 (14:39 +0800)]
USB: Add OTG PET device to TPL

commit e5dff0e80463cc3fa236e898ef1491b40be70b19 upstream.

OTG device shall support this device for allowing compliance automated testing.
The modification is derived from Pavankumar and Vijayavardhans' previous work.

Signed-off-by: Macpaul Lin <>
Cc: Pavankumar Kondeti <>
Cc: Vijayavardhan Vennapusa <>
Signed-off-by: Greg Kroah-Hartman <>
Signed-off-by: Ben Hutchings <>
8 years agousb-core bInterval quirk
James P Michels III [Sun, 27 Jul 2014 17:28:04 +0000 (13:28 -0400)]
usb-core bInterval quirk

commit cd83ce9e6195aa3ea15ab4db92892802c20df5d0 upstream.

This patch adds a usb quirk to support devices with interupt endpoints
and bInterval values expressed as microframes. The quirk causes the
parse endpoint function to modify the reported bInterval to a standards
conforming value.

There is currently code in the endpoint parser that checks for
bIntervals that are outside of the valid range (1-16 for USB 2+ high
speed and super speed interupt endpoints). In this case, the code assumes
the bInterval is being reported in 1ms frames. As well, the correction
is only applied if the original bInterval value is out of the 1-16 range.

With this quirk applied to the device, the bInterval will be
accurately adjusted from microframes to an exponent.

Signed-off-by: James P Michels III <>
Signed-off-by: Greg Kroah-Hartman <>
Signed-off-by: Ben Hutchings <>
8 years agousb-storage/SCSI: blacklist FUA on JMicron 152d:2566 USB-SATA controller
Dmitry Nezhevenko [Mon, 12 Jan 2015 17:13:01 +0000 (19:13 +0200)]
usb-storage/SCSI: blacklist FUA on JMicron 152d:2566 USB-SATA controller

commit bf5c4136fa5ce471bdbf4cf59a813e32755fd014 upstream.

It looks like FUA support is broken on JMicron 152d:2566 bridge:

[223159.885704] sd 7:0:0:0: [sdc] Write Protect is off
[223159.885706] sd 7:0:0:0: [sdc] Mode Sense: 47 00 10 08
[223159.885942] sd 7:0:0:0: [sdc] Write cache: enabled, read cache: enabled, supports DPO and FUA

[223283.691677] sd 7:0:0:0: [sdc]
[223283.691680] Result: hostbyte=DID_OK driverbyte=DRIVER_SENSE
[223283.691681] sd 7:0:0:0: [sdc]
[223283.691682] Sense Key : Illegal Request [current]
[223283.691684] sd 7:0:0:0: [sdc]
[223283.691685] Add. Sense: Invalid field in cdb
[223283.691686] sd 7:0:0:0: [sdc] CDB:
[223283.691687] Write(10): 2a 08 15 d0 83 0d 00 00 01 00
[223283.691690] blk_update_request: critical target error, dev sdc, sector 2927892584

This patch adds blacklist flag so that sd will not use FUA

Signed-off-by: Dmitry Nezhevenko <>
Cc: Phil Dibowitz <>
Cc: Alan Stern <>
Signed-off-by: Greg Kroah-Hartman <>
Signed-off-by: Ben Hutchings <>
8 years agonl80211: fix per-station group key get/del and memory leak
Johannes Berg [Fri, 23 Jan 2015 10:10:12 +0000 (11:10 +0100)]
nl80211: fix per-station group key get/del and memory leak

commit 0fa7b39131576dd1baa6ca17fca53c65d7f62249 upstream.

In case userspace attempts to obtain key information for or delete a
unicast key, this is currently erroneously rejected unless the driver
sets the WIPHY_FLAG_IBSS_RSN flag. Apparently enough drivers do so it
was never noticed.

Fix that, and while at it fix a potential memory leak: the error path
in the get_key() function was placed after allocating a message but
didn't free it - move it to a better place. Luckily admin permissions
are needed to call this operation.

Fixes: e31b82136d1ad ("cfg80211/mac80211: allow per-station GTKs")
Signed-off-by: Johannes Berg <>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <>
8 years agox86, tls: Interpret an all-zero struct user_desc as "no segment"
Andy Lutomirski [Thu, 22 Jan 2015 19:27:59 +0000 (11:27 -0800)]
x86, tls: Interpret an all-zero struct user_desc as "no segment"

commit 3669ef9fa7d35f573ec9c0e0341b29251c2734a7 upstream.

The Witcher 2 did something like this to allocate a TLS segment index:

        struct user_desc u_info;
        bzero(&u_info, sizeof(u_info));
        u_info.entry_number = (uint32_t)-1;

        syscall(SYS_set_thread_area, &u_info);

Strictly speaking, this code was never correct.  It should have set
read_exec_only and seg_not_present to 1 to indicate that it wanted
to find a free slot without putting anything there, or it should
have put something sensible in the TLS slot if it wanted to allocate
a TLS entry for real.  The actual effect of this code was to
allocate a bogus segment that could be used to exploit espfix.

The set_thread_area hardening patches changed the behavior, causing
set_thread_area to return -EINVAL and crashing the game.

This changes set_thread_area to interpret this as a request to find
a free slot and to leave it empty, which isn't *quite* what the game
expects but should be close enough to keep it working.  In
particular, using the code above to allocate two segments will
allocate the same segment both times.

According to FrostbittenKing on Github, this fixes The Witcher 2.

If this somehow still causes problems, we could instead allocate
a limit==0 32-bit data segment, but that seems rather ugly to me.

Fixes: 41bdc78544b8 x86/tls: Validate TLS entries to protect espfix
Signed-off-by: Andy Lutomirski <>
Signed-off-by: Thomas Gleixner <>
Signed-off-by: Ben Hutchings <>
8 years agoInput: i8042 - add noloop quirk for Medion Akoya E7225 (MD98857)
Jochen Hein [Thu, 22 Jan 2015 20:03:15 +0000 (12:03 -0800)]
Input: i8042 - add noloop quirk for Medion Akoya E7225 (MD98857)

commit 1d90d6d5522befa8efa1a7ea406be65cf865ded4 upstream.

Without this the aux port does not get detected, and consequently the touchpad
will not work.

With this patch the touchpad is detected:

$ dmesg | grep -E "(SYN|i8042|serio)"
pnp 00:03: Plug and Play ACPI device, IDs SYN1d22 PNP0f13 (active)
i8042: PNP: PS/2 Controller [PNP0303:PS2K,PNP0f13:PS2M] at 0x60,0x64 irq 1,12
serio: i8042 KBD port at 0x60,0x64 irq 1
serio: i8042 AUX port at 0x60,0x64 irq 12
input: AT Translated Set 2 keyboard as /devices/platform/i8042/serio0/input/input4
psmouse serio1: synaptics: Touchpad model: 1, fw: 8.1, id: 0x1e2b1, caps: 0xd00123/0x840300/0x126800, board id: 2863, fw id: 1473085
input: SynPS/2 Synaptics TouchPad as /devices/platform/i8042/serio1/input/input6

dmidecode excerpt for this laptop is:

Handle 0x0001, DMI type 1, 27 bytes
System Information
        Manufacturer: Medion
        Product Name: Akoya E7225
        Version: 1.0

Signed-off-by: Jochen Hein <>
Signed-off-by: Dmitry Torokhov <>
Signed-off-by: Ben Hutchings <>
8 years agox86, tls, ldt: Stop checking lm in LDT_empty
Andy Lutomirski [Thu, 22 Jan 2015 19:27:58 +0000 (11:27 -0800)]
x86, tls, ldt: Stop checking lm in LDT_empty

commit e30ab185c490e9a9381385529e0fd32f0a399495 upstream.

32-bit programs don't have an lm bit in their ABI, so they can't
reliably cause LDT_empty to return true without resorting to memset.
They shouldn't need to do this.

This should fix a longstanding, if minor, issue in all 64-bit kernels
as well as a potential regression in the TLS hardening code.

Fixes: 41bdc78544b8 x86/tls: Validate TLS entries to protect espfix
Signed-off-by: Andy Lutomirski <>
Signed-off-by: Thomas Gleixner <>
Signed-off-by: Ben Hutchings <>
8 years agox86, hyperv: Mark the Hyper-V clocksource as being continuous
K. Y. Srinivasan [Tue, 13 Jan 2015 00:26:02 +0000 (16:26 -0800)]
x86, hyperv: Mark the Hyper-V clocksource as being continuous

commit 32c6590d126836a062b3140ed52d898507987017 upstream.

The Hyper-V clocksource is continuous; mark it accordingly.

Signed-off-by: K. Y. Srinivasan <>
Signed-off-by: Thomas Gleixner <>
Signed-off-by: Ben Hutchings <>
8 years agolibata: prevent HSM state change race between ISR and PIO
David Jeffery [Mon, 19 Jan 2015 19:03:25 +0000 (13:03 -0600)]
libata: prevent HSM state change race between ISR and PIO

commit ce7514526742c0898b837d4395f515b79dfb5a12 upstream.

It is possible for ata_sff_flush_pio_task() to set ap->hsm_task_state to
HSM_ST_IDLE in between the time __ata_sff_port_intr() checks for HSM_ST_IDLE
and before it calls ata_sff_hsm_move() causing ata_sff_hsm_move() to BUG().

This problem is hard to reproduce making this patch hard to verify, but this
fix will prevent the race.

I have not been able to reproduce the problem, but here is a crash dump from
a 2.6.32 kernel.

On examining the ata port's state, its hsm_task_state field has a value of HSM_ST_IDLE:

crash> struct ata_port.hsm_task_state ffff881c1121c000
  hsm_task_state = 0

Normally, this should not be possible as ata_sff_hsm_move() was called from ata_sff_host_intr(),
which checks hsm_task_state and won't call ata_sff_hsm_move() if it has a HSM_ST_IDLE value.

PID: 11053  TASK: ffff8816e846cae0  CPU: 0   COMMAND: "sshd"
 #0 [ffff88008ba03960] machine_kexec at ffffffff81038f3b
 #1 [ffff88008ba039c0] crash_kexec at ffffffff810c5d92
 #2 [ffff88008ba03a90] oops_end at ffffffff8152b510
 #3 [ffff88008ba03ac0] die at ffffffff81010e0b
 #4 [ffff88008ba03af0] do_trap at ffffffff8152ad74
 #5 [ffff88008ba03b50] do_invalid_op at ffffffff8100cf95
 #6 [ffff88008ba03bf0] invalid_op at ffffffff8100bf9b
    [exception RIP: ata_sff_hsm_move+317]
    RIP: ffffffff813a77ad  RSP: ffff88008ba03ca0  RFLAGS: 00010097
    RAX: 0000000000000000  RBX: ffff881c1121dc60  RCX: 0000000000000000
    RDX: ffff881c1121dd10  RSI: ffff881c1121dc60  RDI: ffff881c1121c000
    RBP: ffff88008ba03d00   R8: 0000000000000000   R9: 000000000000002e
    R10: 000000000001003f  R11: 000000000000009b  R12: ffff881c1121c000
    R13: 0000000000000000  R14: 0000000000000050  R15: ffff881c1121dd78
    ORIG_RAX: ffffffffffffffff  CS: 0010  SS: 0018
 #7 [ffff88008ba03d08] ata_sff_host_intr at ffffffff813a7fbd
 #8 [ffff88008ba03d38] ata_sff_interrupt at ffffffff813a821e
 #9 [ffff88008ba03d78] handle_IRQ_event at ffffffff810e6ec0
>--- <IRQ stack> ---
    [exception RIP: pipe_poll+48]
    RIP: ffffffff81192780  RSP: ffff880f26d459b8  RFLAGS: 00000246
    RAX: 0000000000000000  RBX: ffff880f26d459c8  RCX: 0000000000000000
    RDX: 0000000000000001  RSI: 0000000000000000  RDI: ffff881a0539fa80
    RBP: ffffffff8100bb8e   R8: ffff8803b23324a0   R9: 0000000000000000
    R10: ffff880f26d45dd0  R11: 0000000000000008  R12: ffffffff8109b646
    R13: ffff880f26d45948  R14: 0000000000000246  R15: 0000000000000246
    ORIG_RAX: ffffffffffffff10  CS: 0010  SS: 0018
    RIP: 00007f26017435c3  RSP: 00007fffe020c420  RFLAGS: 00000206
    RAX: 0000000000000017  RBX: ffffffff8100b072  RCX: 00007fffe020c45c
    RDX: 00007f2604a3f120  RSI: 00007f2604a3f140  RDI: 000000000000000d
    RBP: 0000000000000000   R8: 00007fffe020e570   R9: 0101010101010101
    R10: 0000000000000000  R11: 0000000000000246  R12: 00007fffe020e5f0
    R13: 00007fffe020e5f4  R14: 00007f26045f373c  R15: 00007fffe020e5e0
    ORIG_RAX: 0000000000000017  CS: 0033  SS: 002b

Somewhere between the ata_sff_hsm_move() check and the ata_sff_host_intr() check, the value changed.
On examining the other cpus to see what else was running, another cpu was running the error handler

PID: 326    TASK: ffff881c11014aa0  CPU: 1   COMMAND: "scsi_eh_1"
 #0 [ffff88008ba27e90] crash_nmi_callback at ffffffff8102fee6
 #1 [ffff88008ba27ea0] notifier_call_chain at ffffffff8152d515
 #2 [ffff88008ba27ee0] atomic_notifier_call_chain at ffffffff8152d57a
 #3 [ffff88008ba27ef0] notify_die at ffffffff810a154e
 #4 [ffff88008ba27f20] do_nmi at ffffffff8152b1db
 #5 [ffff88008ba27f50] nmi at ffffffff8152aaa0
    [exception RIP: _spin_lock_irqsave+47]
    RIP: ffffffff8152a1ff  RSP: ffff881c11a73aa0  RFLAGS: 00000006
    RAX: 0000000000000001  RBX: ffff881c1121deb8  RCX: 0000000000000000
    RDX: 0000000000000246  RSI: 0000000000000020  RDI: ffff881c122612d8
    RBP: ffff881c11a73aa0   R8: ffff881c17083800   R9: 0000000000000000
    R10: 0000000000000000  R11: 0000000000000000  R12: ffff881c1121c000
    R13: 000000000000001f  R14: ffff881c1121dd50  R15: ffff881c1121dc60
    ORIG_RAX: ffffffffffffffff  CS: 0010  SS: 0000
>--- <NMI exception stack> ---
 #6 [ffff881c11a73aa0] _spin_lock_irqsave at ffffffff8152a1ff
 #7 [ffff881c11a73aa8] ata_exec_internal_sg at ffffffff81396fb5
 #8 [ffff881c11a73b58] ata_exec_internal at ffffffff81397109
 #9 [ffff881c11a73bd8] atapi_eh_request_sense at ffffffff813a34eb

Before it tried to acquire a spinlock, ata_exec_internal_sg() called ata_sff_flush_pio_task().
This function will set ap->hsm_task_state to HSM_ST_IDLE, and has no locking around setting this
value. ata_sff_flush_pio_task() can then race with the interrupt handler and potentially set
HSM_ST_IDLE at a fatal moment, which will trigger a kernel BUG.

v2: Fixup comment in ata_sff_flush_pio_task()

tj: Further updated comment.  Use ap->lock instead of shost lock and
    use the [un]lock_irq variant instead of the irqsave/restore one.

Signed-off-by: David Milburn <>
Signed-off-by: Tejun Heo <>
Signed-off-by: Ben Hutchings <>
8 years agoscripts/ There is no -m32 gcc option on Super-H anymore
Michael Karcher [Sat, 17 Jan 2015 23:36:15 +0000 (00:36 +0100)]
scripts/ There is no -m32 gcc option on Super-H anymore

commit 1caf6aaaa47471831d77c75f094d4e00ad1ec808 upstream.

Compiling SH with gcc-4.8 fails due to the -m32 option not being


      CC      init/main.o
    gcc-4.8: error: unrecognized command line option '-m32'
    ld: cannot find init/.tmp_mc_main.o: No such file or directory
    objcopy: 'init/.tmp_mx_main.o': No such file
    rm: cannot remove 'init/.tmp_mx_main.o': No such file or directory
    rm: cannot remove 'init/.tmp_mc_main.o': No such file or directory

Cc: Matt Fleming <>
Reported-by: John Paul Adrian Glaubitz <>
Signed-off-by: Michael Karcher <>
Signed-off-by: Steven Rostedt <>
Signed-off-by: Ben Hutchings <>
8 years agolibata: allow sata_sil24 to opt-out of tag ordered submission
Dan Williams [Fri, 16 Jan 2015 23:13:02 +0000 (15:13 -0800)]
libata: allow sata_sil24 to opt-out of tag ordered submission

commit 72dd299d5039a336493993dcc63413cf31d0e662 upstream.

Ronny reports:
    "Since commit 8a4aeec8d "libata/ahci: accommodate tag ordered
    controllers" the access to the harddisk on the first SATA-port is
    failing on its first access. The access to the harddisk on the
    second port is working normal.

    When reverting the above commit, access to both harddisks is working
    fine again."

Maintain tag ordered submission as the default, but allow sata_sil24 to
continue with the old behavior.

Cc: Tejun Heo <>
Reported-by: Ronny Hegewald <>
Signed-off-by: Dan Williams <>
Signed-off-by: Tejun Heo <>
Signed-off-by: Ben Hutchings <>
8 years agoALSA: usb-audio: Add mic volume fix quirk for Logitech Webcam C210
Jason Lee Cragg [Sat, 17 Jan 2015 17:28:29 +0000 (12:28 -0500)]
ALSA: usb-audio: Add mic volume fix quirk for Logitech Webcam C210

commit 6455931186bff407493135e74c5f32efd30860e2 upstream.

Signed-off-by: Jason Lee Cragg <>
Signed-off-by: Takashi Iwai <>
Signed-off-by: Ben Hutchings <>
8 years agonet: sctp: fix race for one-to-many sockets in sendmsg's auto associate
Daniel Borkmann [Thu, 15 Jan 2015 15:34:35 +0000 (16:34 +0100)]
net: sctp: fix race for one-to-many sockets in sendmsg's auto associate

commit 2061dcd6bff8b774b4fac8b0739b6be3f87bc9f2 upstream.

I.e. one-to-many sockets in SCTP are not required to explicitly
call into connect(2) or sctp_connectx(2) prior to data exchange.
Instead, they can directly invoke sendmsg(2) and the SCTP stack
will automatically trigger connection establishment through 4WHS
via sctp_primitive_ASSOCIATE(). However, this in its current
implementation is racy: INIT is being sent out immediately (as
it cannot be bundled anyway) and the rest of the DATA chunks are
queued up for later xmit when connection is established, meaning
sendmsg(2) will return successfully. This behaviour can result
in an undesired side-effect that the kernel made the application
think the data has already been transmitted, although none of it
has actually left the machine, worst case even after close(2)'ing
the socket.

Instead, when the association from client side has been shut down
e.g. first gracefully through SCTP_EOF and then close(2), the
client could afterwards still receive the server's INIT_ACK due
to a connection with higher latency. This INIT_ACK is then considered
out of the blue and hence responded with ABORT as there was no
alive assoc found anymore. This can be easily reproduced f.e.
with sctp_test application from lksctp. One way to fix this race
is to wait for the handshake to actually complete.

The fix defers waiting after sctp_primitive_ASSOCIATE() and
sctp_primitive_SEND() succeeded, so that DATA chunks cooked up
from sctp_sendmsg() have already been placed into the output
queue through the side-effect interpreter, and therefore can then
be bundeled together with COOKIE_ECHO control chunks.

strace from example application (shortened):

sendmsg(3, {msg_name(28)={sa_family=AF_INET, sin_port=htons(8888), sin_addr=inet_addr("")},
           msg_iov(1)=[{"hello", 5}], msg_controllen=0, msg_flags=0}, 0) = 5
sendmsg(3, {msg_name(28)={sa_family=AF_INET, sin_port=htons(8888), sin_addr=inet_addr("")},
           msg_iov(1)=[{"hello", 5}], msg_controllen=0, msg_flags=0}, 0) = 5
sendmsg(3, {msg_name(28)={sa_family=AF_INET, sin_port=htons(8888), sin_addr=inet_addr("")},
           msg_iov(1)=[{"hello", 5}], msg_controllen=0, msg_flags=0}, 0) = 5
sendmsg(3, {msg_name(28)={sa_family=AF_INET, sin_port=htons(8888), sin_addr=inet_addr("")},
           msg_iov(1)=[{"hello", 5}], msg_controllen=0, msg_flags=0}, 0) = 5
sendmsg(3, {msg_name(28)={sa_family=AF_INET, sin_port=htons(8888), sin_addr=inet_addr("")},
           msg_iov(0)=[], msg_controllen=48, {cmsg_len=48, cmsg_level=0x84 /* SOL_??? */, cmsg_type=, ...},
           msg_flags=0}, 0) = 0 // graceful shutdown for SOCK_SEQPACKET via SCTP_EOF
close(3) = 0

tcpdump before patch (fooling the application):

22:33:36.306142 IP > sctp (1) [INIT] [init tag: 3879023686] [rwnd: 106496] [OS: 10] [MIS: 65535] [init TSN: 3139201684]
22:33:36.316619 IP > sctp (1) [INIT ACK] [init tag: 3345394793] [rwnd: 106496] [OS: 10] [MIS: 10] [init TSN: 3380109591]
22:33:36.317600 IP > sctp (1) [ABORT]

tcpdump after patch:

14:28:58.884116 IP > sctp (1) [INIT] [init tag: 438593213] [rwnd: 106496] [OS: 10] [MIS: 65535] [init TSN: 3092969729]
14:28:58.888414 IP > sctp (1) [INIT ACK] [init tag: 381429855] [rwnd: 106496] [OS: 10] [MIS: 10] [init TSN: 2141904492]
14:28:58.888638 IP > sctp (1) [COOKIE ECHO] , (2) [DATA] (B)(E) [TSN: 3092969729] [...]
14:28:58.893278 IP > sctp (1) [COOKIE ACK] , (2) [SACK] [cum ack 3092969729] [a_rwnd 106491] [#gap acks 0] [#dup tsns 0]
14:28:58.893591 IP > sctp (1) [DATA] (B)(E) [TSN: 3092969730] [...]
14:28:59.096963 IP > sctp (1) [SACK] [cum ack 3092969730] [a_rwnd 106496] [#gap acks 0] [#dup tsns 0]
14:28:59.097086 IP > sctp (1) [DATA] (B)(E) [TSN: 3092969731] [...] , (2) [DATA] (B)(E) [TSN: 3092969732] [...]
14:28:59.103218 IP > sctp (1) [SACK] [cum ack 3092969732] [a_rwnd 106486] [#gap acks 0] [#dup tsns 0]
14:28:59.103330 IP > sctp (1) [SHUTDOWN]
14:28:59.107793 IP > sctp (1) [SHUTDOWN ACK]
14:28:59.107890 IP > sctp (1) [SHUTDOWN COMPLETE]

Looks like this bug is from the pre-git history museum. ;)

Fixes: 08707d5482df ("lksctp-2_5_31-0_5_1.patch")
Signed-off-by: Daniel Borkmann <>
Acked-by: Vlad Yasevich <>
Signed-off-by: David S. Miller <>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <>
8 years agogpio: sysfs: fix gpio attribute-creation race
Johan Hovold [Tue, 13 Jan 2015 12:00:06 +0000 (13:00 +0100)]
gpio: sysfs: fix gpio attribute-creation race

commit ebbeba120ab2ec6ac5f3afc1425ec6ff0b77ad6f upstream.

Fix attribute-creation race with userspace by using the default group
to create also the contingent gpio device attributes.

Fixes: d8f388d8dc8d ("gpio: sysfs interface")
Signed-off-by: Johan Hovold <>
Signed-off-by: Linus Walleij <>
[bwh: Backported to 3.2:
 - Adjust filenames, context
 - Use gpio_to_desc(), not gpiod_to_desc(), in gpio_is_visible()
 - gpio_is_visible() must return mode_t]
Signed-off-by: Ben Hutchings <>
8 years agogpio: sysfs: fix gpio device-attribute leak
Johan Hovold [Tue, 13 Jan 2015 12:00:05 +0000 (13:00 +0100)]
gpio: sysfs: fix gpio device-attribute leak

commit 0915e6feb38de8d3601819992a5bd050201a56fa upstream.

The gpio device attributes were never destroyed when the gpio was
unexported (or on export failures).

Use device_create_with_groups() to create the default device attributes
of the gpio class device. Note that this also fixes the
attribute-creation race with userspace for these attributes.

Remove contingent attributes in export error path and on unexport.

Fixes: d8f388d8dc8d ("gpio: sysfs interface")
Signed-off-by: Johan Hovold <>
Signed-off-by: Linus Walleij <>
[bwh: Backported to 3.2: adjust filename, context]
Signed-off-by: Ben Hutchings <>
8 years agoFix circular locking dependency (3.3-rc2)
Ming Lei [Mon, 13 Feb 2012 14:53:20 +0000 (22:53 +0800)]
Fix circular locking dependency (3.3-rc2)

commit 864533ceb6db336dead389577c102a8b792a121a upstream.


On Wed, Feb 8, 2012 at 8:41 PM, Felipe Balbi <> wrote:
> Hi guys,
> I have just triggered the folllowing:
> [   84.860321] ======================================================
> [   84.860321] [ INFO: possible circular locking dependency detected ]
> [   84.860321] 3.3.0-rc2-00026-ge4e8a39 #474 Not tainted
> [   84.860321] -------------------------------------------------------
> [   84.860321] bash/949 is trying to acquire lock:
> [   84.860321]  (sysfs_lock){+.+.+.}, at: [<c0275358>] gpio_value_store+0x24/0xcc
> [   84.860321]
> [   84.860321] but task is already holding lock:
> [   84.860321]  (s_active#22){++++.+}, at: [<c016996c>] sysfs_write_file+0xdc/0x184
> [   84.911468]
> [   84.911468] which lock already depends on the new lock.
> [   84.911468]
> [   84.920043]
> [   84.920043] the existing dependency chain (in reverse order) is:
> [   84.920043]
> [   84.927886] -> #1 (s_active#22){++++.+}:
> [   84.927886]        [<c008f640>] check_prevs_add+0xdc/0x150
> [   84.927886]        [<c008fc18>] validate_chain.clone.24+0x564/0x694
> [   84.927886]        [<c0090cdc>] __lock_acquire+0x49c/0x980
> [   84.951660]        [<c0091838>] lock_acquire+0x98/0x100
> [   84.951660]        [<c016a8e8>] sysfs_deactivate+0xb0/0x100
> [   84.962982]        [<c016b1b4>] sysfs_addrm_finish+0x2c/0x6c
> [   84.962982]        [<c016b8bc>] sysfs_remove_dir+0x84/0x98
> [   84.962982]        [<c02590d8>] kobject_del+0x10/0x78
> [   84.974670]        [<c02c29e8>] device_del+0x140/0x170
> [   84.974670]        [<c02c2a24>] device_unregister+0xc/0x18
> [   84.985382]        [<c0276894>] gpio_unexport+0xbc/0xdc
> [   84.985382]        [<c02768c8>] gpio_free+0x14/0xfc
> [   85.001708]        [<c0276a28>] unexport_store+0x78/0x8c
> [   85.001708]        [<c02c5af8>] class_attr_store+0x18/0x24
> [   85.007293]        [<c0169990>] sysfs_write_file+0x100/0x184
> [   85.018981]        [<c0109d48>] vfs_write+0xb4/0x148
> [   85.018981]        [<c0109fd0>] sys_write+0x40/0x70
> [   85.018981]        [<c0013cc0>] ret_fast_syscall+0x0/0x3c
> [   85.035003]
> [   85.035003] -> #0 (sysfs_lock){+.+.+.}:
> [   85.035003]        [<c008f54c>] check_prev_add+0x680/0x698
> [   85.035003]        [<c008f640>] check_prevs_add+0xdc/0x150
> [   85.052093]        [<c008fc18>] validate_chain.clone.24+0x564/0x694
> [   85.052093]        [<c0090cdc>] __lock_acquire+0x49c/0x980
> [   85.052093]        [<c0091838>] lock_acquire+0x98/0x100
> [   85.069885]        [<c047e280>] mutex_lock_nested+0x3c/0x2f4
> [   85.069885]        [<c0275358>] gpio_value_store+0x24/0xcc
> [   85.069885]        [<c02c18dc>] dev_attr_store+0x18/0x24
> [   85.087158]        [<c0169990>] sysfs_write_file+0x100/0x184
> [   85.087158]        [<c0109d48>] vfs_write+0xb4/0x148
> [   85.098297]        [<c0109fd0>] sys_write+0x40/0x70
> [   85.098297]        [<c0013cc0>] ret_fast_syscall+0x0/0x3c
> [   85.109069]
> [   85.109069] other info that might help us debug this:
> [   85.109069]
> [   85.117462]  Possible unsafe locking scenario:
> [   85.117462]
> [   85.117462]        CPU0                    CPU1
> [   85.128417]        ----                    ----
> [   85.128417]   lock(s_active#22);
> [   85.128417]                                lock(sysfs_lock);
> [   85.128417]                                lock(s_active#22);
> [   85.142486]   lock(sysfs_lock);
> [   85.151794]
> [   85.151794]  *** DEADLOCK ***
> [   85.151794]
> [   85.151794] 2 locks held by bash/949:
> [   85.158020]  #0:  (&buffer->mutex){+.+.+.}, at: [<c01698b8>] sysfs_write_file+0x28/0x184
> [   85.170349]  #1:  (s_active#22){++++.+}, at: [<c016996c>] sysfs_write_file+0xdc/0x184
> [   85.170349]
> [   85.178588] stack backtrace:
> [   85.178588] [<c001b824>] (unwind_backtrace+0x0/0xf0) from [<c008de64>] (print_circular_bug+0x100/0x114)
> [   85.193023] [<c008de64>] (print_circular_bug+0x100/0x114) from [<c008f54c>] (check_prev_add+0x680/0x698)
> [   85.193023] [<c008f54c>] (check_prev_add+0x680/0x698) from [<c008f640>] (check_prevs_add+0xdc/0x150)
> [   85.212524] [<c008f640>] (check_prevs_add+0xdc/0x150) from [<c008fc18>] (validate_chain.clone.24+0x564/0x694)
> [   85.212524] [<c008fc18>] (validate_chain.clone.24+0x564/0x694) from [<c0090cdc>] (__lock_acquire+0x49c/0x980)
> [   85.233306] [<c0090cdc>] (__lock_acquire+0x49c/0x980) from [<c0091838>] (lock_acquire+0x98/0x100)
> [   85.233306] [<c0091838>] (lock_acquire+0x98/0x100) from [<c047e280>] (mutex_lock_nested+0x3c/0x2f4)
> [   85.242614] [<c047e280>] (mutex_lock_nested+0x3c/0x2f4) from [<c0275358>] (gpio_value_store+0x24/0xcc)
> [   85.261840] [<c0275358>] (gpio_value_store+0x24/0xcc) from [<c02c18dc>] (dev_attr_store+0x18/0x24)
> [   85.261840] [<c02c18dc>] (dev_attr_store+0x18/0x24) from [<c0169990>] (sysfs_write_file+0x100/0x184)
> [   85.271240] [<c0169990>] (sysfs_write_file+0x100/0x184) from [<c0109d48>] (vfs_write+0xb4/0x148)
> [   85.290008] [<c0109d48>] (vfs_write+0xb4/0x148) from [<c0109fd0>] (sys_write+0x40/0x70)
> [   85.298400] [<c0109fd0>] (sys_write+0x40/0x70) from [<c0013cc0>] (ret_fast_syscall+0x0/0x3c)
> -bash: echo: write error: Operation not permitted
> the way to trigger is:
> root@legolas:~# cd /sys/class/gpio/
> root@legolas:/sys/class/gpio# echo 2 > export
> root@legolas:/sys/class/gpio# echo 2 > unexport
> root@legolas:/sys/class/gpio# echo 2 > export
> root@legolas:/sys/class/gpio# cd gpio2/
> root@legolas:/sys/class/gpio/gpio2# echo 1 > value

Looks 'sysfs_lock' needn't to be held for unregister, so the patch below may
fix the problem.

Acked-by: Linus Walleij <>
Signed-off-by: Grant Likely <>
Signed-off-by: Ben Hutchings <>
8 years agogpiolib: Refactor gpio_export
Ryan Mallon [Mon, 22 Oct 2012 00:39:12 +0000 (11:39 +1100)]
gpiolib: Refactor gpio_export

commit fc4e2514995d9cd7f3e1a67098ce65d72acf8ec7 upstream.

The gpio_export function uses nested if statements and the status
variable to handle the failure cases. This makes the function logic
difficult to follow. Refactor the code to abort immediately on failure
using goto. This makes the code slightly longer, but significantly
reduces the nesting and number of split lines and makes the code easier
to read.

Signed-off-by: Ryan Mallon <>
Signed-off-by: Linus Walleij <>
Signed-off-by: Ben Hutchings <>