NFSv4: Rate limit the state manager for lock reclaim warning messages

commit 96dcadc2fdd111dca90d559f189a30c65394451a upstream.

Adding rate limit on `Lock reclaim failed` messages since it could fill
up system logs
Signed-off-by: William Dauchy <wdauchy@gmail.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
[bwh: Backported to 3.2: add the 'NFS:' prefix at the same time]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

brcmsmac: "INTERMEDIATE but not AMPDU" only when tracing

commit 6ead629b27269c553c9092c47cd8f5ab0309ee3b upstream.

I keep getting the following messages on the log buffer:
[ 2167.097507] ieee80211 phy0: brcms_c_dotxstatus: INTERMEDIATE but not AMPDU
[ 2281.331305] ieee80211 phy0: brcms_c_dotxstatus: INTERMEDIATE but not AMPDU
[ 2281.332539] ieee80211 phy0: brcms_c_dotxstatus: INTERMEDIATE but not AMPDU
[ 2329.876605] ieee80211 phy0: brcms_c_dotxstatus: INTERMEDIATE but not AMPDU
[ 2329.877354] ieee80211 phy0: brcms_c_dotxstatus: INTERMEDIATE but not AMPDU
[ 2462.280756] ieee80211 phy0: brcms_c_dotxstatus: INTERMEDIATE but not AMPDU
[ 2615.651689] ieee80211 phy0: brcms_c_dotxstatus: INTERMEDIATE but not AMPDU

From the code comment I understand that this something that can -
and does, quite frequently - happen.

Signed-off-by: Eldad Zack <eldad@fogrefinery.com>
Acked-by: Franky Lin<frankyl@broadcom.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

kbuild: do not check for ancient modutils tools

commit 620c231c7a7f48745094727bb612f6321cfc8844 upstream.

scripts/depmod.sh checks for the output of '-V' expecting that it has
module-init-tools in it. It's a hack to prevent users from using
modutils instead of module-init-tools, that only works with 2.4.x
kernels. This however prints an annoying warning for kmod tool, that is
currently replacing module-init-tools.

Rather than putting another check for kmod's version, just remove it
since users of 2.4.x kernel are unlikely to upgrade to 3.x, and if they
do, let depmod fail in that case because they should know what they are
doing.

Signed-off-by: Lucas De Marchi <lucas.demarchi@profusion.mobi>
Acked-by: WANG Cong <amwang@redhat.com>
Acked-By: Kay Sievers <kay.sievers@vrfy.org>
Signed-off-by: Michal Marek <mmarek@suse.cz>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

drm/i915: fix operator precedence when enabling RC6p

commit c0e2ee1bc0cf82eec89e26b7afe7e4db0561b7d9 upstream.

As noticed by Torsten Kaiser, the operator precedence can play tricks with
us here.

CC: Dave Airlie <airlied@redhat.com>
Signed-off-by: Eugeni Dodonov <eugeni.dodonov@intel.com>
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

drm/i915: do not enable RC6p on Sandy Bridge

commit 1c8ecf80fdee4e7b23a9e7da7ff9bd59ba2dcf96 upstream.

With base on latest findings, RC6p seems to be respondible for RC6-related
issues on Sandy Bridge platform. To work-around those issues, the previous
solution was to completely disable RC6 on Sandy Bridge for the past few
releases, even if plain RC6 was not giving any issues.

What this patch does is preventing RC6p from being enabled on Sandy Bridge
even if users enable RC6 via a kernel parameter. So it won't change the
defaults in any way, but will ensure that if users do enable RC6 manually
it won't break their machines by enabling this extra state.

Proper fix for this (enabling specific RC6 states according to the GPU
generation) were proposed for the -next kernel, but we are too late in the
release process now to pick such changes.

Acked-by: Keith Packard <keithp@keithp.com>
Signed-off-by: Eugeni Dodonov <eugeni.dodonov@intel.com>
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

net/wireless: ipw2x00: add supported cipher suites to wiphy initialization

commit a141e6a0097118bb35024485f1faffc0d9042f5c upstream.

Driver doesn't report its supported cipher suites through cfg80211
interface. It still uses wext interface and probably will not work
through nl80211, but will at least correctly advertise supported
features.

Bug was reported by Omar Siam.
https://bugzilla.kernel.org/show_bug.cgi?id=43049

Signed-off-by: Stanislav Yakovlev <stas.yakovlev@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

rtl8187: ->brightness_set can not sleep

commit 0fde0a8cfd0ede7f310d6a681c8e5a7cb3e32406 upstream.

Fix:

BUG: sleeping function called from invalid context at kernel/workqueue.c:2547
in_atomic(): 1, irqs_disabled(): 0, pid: 629, name: wpa_supplicant
2 locks held by wpa_supplicant/629:
 #0:  (rtnl_mutex){+.+.+.}, at: [<c08b2b84>] rtnl_lock+0x14/0x20
 #1:  (&trigger->leddev_list_lock){.+.?..}, at: [<c0867f41>] led_trigger_event+0x21/0x80
Pid: 629, comm: wpa_supplicant Not tainted 3.3.0-0.rc3.git5.1.fc17.i686
Call Trace:
 [<c046a9f6>] __might_sleep+0x126/0x1d0
 [<c0457d6c>] wait_on_work+0x2c/0x1d0
 [<c045a09a>] __cancel_work_timer+0x6a/0x120
 [<c045a160>] cancel_delayed_work_sync+0x10/0x20
 [<f7dd3c22>] rtl8187_led_brightness_set+0x82/0xf0 [rtl8187]
 [<c0867f7c>] led_trigger_event+0x5c/0x80
 [<f7ff5e6d>] ieee80211_led_radio+0x1d/0x40 [mac80211]
 [<f7ff3583>] ieee80211_stop_device+0x13/0x230 [mac80211]

Removing _sync is ok, because if led_on work is currently running
it will be finished before led_off work start to perform, since
they are always queued on the same mac80211 local->workqueue.

Bugzilla: https://bugzilla.redhat.com/show_bug.cgi?id=795176

Signed-off-by: Stanislaw Gruszka <sgruszka@redhat.com>
Acked-by: Larry Finger <Larry.Finger@lwfinger.net>
Acked-by: Hin-Tak Leung <htl10@users.sourceforge.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

tg3: Apply short DMA frag workaround to 5906

commit b7abee6ef888117f92db370620ebf116a38e3f4d upstream.

5906 devices also need the short DMA fragment workaround.  This patch
makes the necessary change.

Signed-off-by: Matt Carlson <mcarlson@broadcom.com>
Tested-by: Christian Kujau <lists@nerdbynature.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

tcp: drop SYN+FIN messages

commit fdf5af0daf8019cec2396cdef8fb042d80fe71fa upstream.

Denys Fedoryshchenko reported that SYN+FIN attacks were bringing his
linux machines to their limits.

Dont call conn_request() if the TCP flags includes SYN flag

Reported-by: Denys Fedoryshchenko <denys@visp.net.lb>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

raid5: delayed stripe fix

commit fab363b5ff502d1b39ddcfec04271f5858d9f26e upstream.

There isn't locking setting STRIPE_DELAYED and STRIPE_PREREAD_ACTIVE bits, but
the two bits have relationship. A delayed stripe can be moved to hold list only
when preread active stripe count is below IO_THRESHOLD. If a stripe has both
the bits set, such stripe will be in delayed list and preread count not 0,
which will make such stripe never leave delayed list.

Signed-off-by: Shaohua Li <shli@fusionio.com>
Signed-off-by: NeilBrown <neilb@suse.de>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

samsung-laptop: make the dmi check less strict

commit 3be324a94df0c3f032178d04549dbfbf6cccb09a upstream.

This enable the driver for everything that look like
a laptop and is from vendor "SAMSUNG ELECTRONICS CO., LTD.".
Note that laptop supported by samsung-q10 seem to have a different
vendor strict.

Also remove every log output until we know that we have a SABI interface
(except if the driver is forced to load, or debug is enabled).

Keeping a whitelist of laptop with a model granularity is something that can't
work without close vendor cooperation (and we don't have that).

Signed-off-by: Corentin Chary <corentincj@iksaif.net>
Acked-by: Greg Kroah-Hartman <gregkh@suse.de>
Signed-off-by: Matthew Garrett <mjg@redhat.com>
[bwh: Backported to 3.2:
 - Adjust context
 - Drop changes relating to ACPI video]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

Linux 3.2.23

vfs: make O_PATH file descriptors usable for 'fchdir()'

commit 332a2e1244bd08b9e3ecd378028513396a004a24 upstream.

We already use them for openat() and friends, but fchdir() also wants to
be able to use O_PATH file descriptors.  This should make it comparable
to the O_SEARCH of Solaris.  In particular, O_PATH allows you to access
(not-quite-open) a directory you don't have read persmission to, only
execute permission.

Noticed during development of multithread support for ksh93.

Reported-by: ольга крыжановская <olga.kryzhanovska@gmail.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

tcm_fc: Resolve suspicious RCU usage warnings

commit 863555be0c81558b1af277addcf68acb8f778860 upstream.

Use rcu_dereference_protected to tell rcu that the ft_lport_lock
is held during ft_lport_create. This resolved "suspicious RCU usage"
warnings when debugging options are turned on.

Signed-off-by: Mark Rustad <mark.d.rustad@intel.com>
Tested-by: Ross Brattain <ross.b.brattain@intel.com>
Signed-off-by: Nicholas Bellinger <nab@linux-iscsi.org>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

mm: Hold a file reference in madvise_remove

commit 9ab4233dd08036fe34a89c7dc6f47a8bf2eb29eb upstream.

Otherwise the code races with munmap (causing a use-after-free
of the vma) or with close (causing a use-after-free of the struct
file).

The bug was introduced by commit 90ed52ebe481 ("[PATCH] holepunch: fix
mmap_sem i_mutex deadlock")

Cc: Hugh Dickins <hugh@veritas.com>
Cc: Miklos Szeredi <mszeredi@suse.cz>
Cc: Badari Pulavarty <pbadari@us.ibm.com>
Cc: Nick Piggin <npiggin@suse.de>
Signed-off-by: Andy Lutomirski <luto@amacapital.net>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
[bwh: Backported to 3.2:
 - Adjust context
 - madvise_remove() calls vmtruncate_range(), not do_fallocate()]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

mtd: cafe_nand: fix an & vs | mistake

commit 48f8b641297df49021093763a3271119a84990a2 upstream.

The intent here was clearly to set result to true if the 0x40000000 flag
was set.  But instead there was a | vs & typo and we always set result
to true.

Artem: check the spec at
wiki.laptop.org/images/5/5c/88ALP01_Datasheet_July_2007.pdf
and this fix looks correct.

Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Artem Bityutskiy <artem.bityutskiy@linux.intel.com>
Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

aio: make kiocb->private NUll in init_sync_kiocb()

commit 2dfd06036ba7ae8e7be2daf5a2fff1dac42390bf upstream.

Ocfs2 uses kiocb.*private as a flag of unsigned long size. In
commit a11f7e6 ocfs2: serialize unaligned aio, the unaligned
io flag is involved in it to serialize the unaligned aio. As
*private is not initialized in init_sync_kiocb() of do_sync_write(),
this unaligned io flag may be unexpectly set in an aligned dio.
And this will cause OCFS2_I(inode)->ip_unaligned_aio decreased
to -1 in ocfs2_dio_end_io(), thus the following unaligned dio
will hang forever at ocfs2_aiodio_wait() in ocfs2_file_aio_write().

Signed-off-by: Junxiao Bi <junxiao.bi@oracle.com>
Acked-by: Jeff Moyer <jmoyer@redhat.com>
Signed-off-by: Joel Becker <jlbec@evilplan.org>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

ocfs2: clear unaligned io flag when dio fails

commit 3e5d3c35a68c9a933bdbdd8685bd1a205b57e806 upstream.

The unaligned io flag is set in the kiocb when an unaligned
dio is issued, it should be cleared even when the dio fails,
or it may affect the following io which are using the same
kiocb.

Signed-off-by: Junxiao Bi <junxiao.bi@oracle.com>
Signed-off-by: Joel Becker <jlbec@evilplan.org>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

cifs: when server doesn't set CAP_LARGE_READ_X, cap default rsize at MaxBufferSize

commit ec01d738a1691dfc85b96b9f796020267a7be577 upstream.

When the server doesn't advertise CAP_LARGE_READ_X, then MS-CIFS states
that you must cap the size of the read at the client's MaxBufferSize.
Unfortunately, testing with many older servers shows that they often
can't service a read larger than their own MaxBufferSize.

Since we can't assume what the server will do in this situation, we must
be conservative here for the default. When the server can't do large
reads, then assume that it can't satisfy any read larger than its
MaxBufferSize either.

Luckily almost all modern servers can do large reads, so this won't
affect them. This is really just for older win9x and OS/2 era servers.
Also, note that this patch just governs the default rsize. The admin can
always override this if he so chooses.

Reported-by: David H. Durgee <dhdurgee@acm.org>
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Steven French <sfrench@w500smf.(none)>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

Btrfs: run delayed directory updates during log replay

commit b6305567e7d31b0bec1b8cb9ec0cadd7f7086f5f upstream.

While we are resolving directory modifications in the
tree log, we are triggering delayed metadata updates to
the filesystem btrees.

This commit forces the delayed updates to run so the
replay code can find any modifications done.  It stops
us from crashing because the directory deleltion replay
expects items to be removed immediately from the tree.

Signed-off-by: Chris Mason <chris.mason@fusionio.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

bridge: Assign rtnl_link_ops to bridge devices created via ioctl (v2)

[ Upstream commit 149ddd83a92b02c658d6c61f3276eb6500d585e8 ]

This ensures that bridges created with brctl(8) or ioctl(2) directly
also carry IFLA_LINKINFO when dumped over netlink. This also allows
to create a bridge with ioctl(2) and delete it with RTM_DELLINK.

Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

ipv6: Move ipv6 proc file registration to end of init order

[ Upstream commit d189634ecab947c10f6f832258b103d0bbfe73cc ]

/proc/net/ipv6_route reflects the contents of fib_table_hash. The proc
handler is installed in ip6_route_net_init() whereas fib_table_hash is
allocated in fib6_net_init() _after_ the proc handler has been installed.

This opens up a short time frame to access fib_table_hash with its pants
down.

Move the registration of the proc files to a later point in the init
order to avoid the race.

Tested :-)

Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

netpoll: fix netpoll_send_udp() bugs

[ Upstream commit 954fba0274058d27c7c07b5ea07c41b3b7477894 ]

Bogdan Hamciuc diagnosed and fixed following bug in netpoll_send_udp() :

"skb->len += len;" instead of "skb_put(skb, len);"

Meaning that _if_ a network driver needs to call skb_realloc_headroom(),
only packet headers would be copied, leaving garbage in the payload.

However the skb_realloc_headroom() must be avoided as much as possible
since it requires memory and netpoll tries hard to work even if memory
is exhausted (using a pool of preallocated skbs)

It appears netpoll_send_udp() reserved 16 bytes for the ethernet header,
which happens to work for typicall drivers but not all.

Right thing is to use LL_RESERVED_SPACE(dev)
(And also add dev->needed_tailroom of tailroom)

This patch combines both fixes.

Many thanks to Bogdan for raising this issue.

Reported-by: Bogdan Hamciuc <bogdan.hamciuc@freescale.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Tested-by: Bogdan Hamciuc <bogdan.hamciuc@freescale.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: Neil Horman <nhorman@tuxdriver.com>
Reviewed-by: Neil Horman <nhorman@tuxdriver.com>
Reviewed-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

ethtool: allow ETHTOOL_GSSET_INFO for users

[ Upstream commit f80400a26a2e8bff541de12834a1134358bb6642 ]

Allow ETHTOOL_GSSET_INFO ethtool ioctl() for unprivileged users.
ETHTOOL_GSTRINGS is already allowed, but is unusable without this one.

Signed-off-by: Michał Mirosław <mirq-linux@rere.qmqm.pl>
Acked-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

bonding: Fix corrupted queue_mapping

[ Upstream commit 5ee31c6898ea5537fcea160999d60dc63bc0c305 ]

In the transmit path of the bonding driver, skb->cb is used to
stash the skb->queue_mapping so that the bonding device can set its
own queue mapping.  This value becomes corrupted since the skb->cb is
also used in __dev_xmit_skb.

When transmitting through bonding driver, bond_select_queue is
called from dev_queue_xmit.  In bond_select_queue the original
skb->queue_mapping is copied into skb->cb (via bond_queue_mapping)
and skb->queue_mapping is overwritten with the bond driver queue.

Subsequently in dev_queue_xmit, __dev_xmit_skb is called which writes
the packet length into skb->cb, thereby overwriting the stashed
queue mappping.  In bond_dev_queue_xmit (called from hard_start_xmit),
the queue mapping for the skb is set to the stashed value which is now
the skb length and hence is an invalid queue for the slave device.

If we want to save skb->queue_mapping into skb->cb[], best place is to
add a field in struct qdisc_skb_cb, to make sure it wont conflict with
other layers (eg : Qdiscc, Infiniband...)

This patchs also makes sure (struct qdisc_skb_cb)->data is aligned on 8
bytes :

netem qdisc for example assumes it can store an u64 in it, without
misalignment penalty.

Note : we only have 20 bytes left in (struct qdisc_skb_cb)->data[].
The largest user is CHOKe and it fills it.

Based on a previous patch from Tom Herbert.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: Tom Herbert <therbert@google.com>
Cc: John Fastabend <john.r.fastabend@intel.com>
Cc: Roland Dreier <roland@kernel.org>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

dummy: fix rcu_sched self-detected stalls

[ Upstream commit 16b0dc29c1af9df341428f4c49ada4f626258082 ]

Trying to "modprobe dummy numdummies=30000" triggers :

INFO: rcu_sched self-detected stall on CPU { 8} (t=60000 jiffies)

After this splat, RTNL is locked and reboot is needed.

We must call cond_resched() to avoid this, even holding RTNL.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

be2net: fix a race in be_xmit()

[ Upstream commit cd8f76c0a0c6fce0b2cf23c9bd0123f91453f46d ]

As soon as hardware is notified of a transmit, we no longer can assume
skb can be dereferenced, as TX completion might have freed the packet.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Sathya Perla <sathya.perla@emulex.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

sky2: fix checksum bit management on some chips

[ Upstream commit 5ff0feac88ced864f44adb145142269196fa79d9 ]

The newer flavors of Yukon II use a different method for receive
checksum offload. This is indicated in the driver by the SKY2_HW_NEW_LE
flag. On these newer chips, the BMU_ENA_RX_CHKSUM should not be set.

The driver would get incorrectly toggle the bit, enabling the old
checksum logic on these chips and cause a BUG_ON() assertion. If
receive checksum was toggled via ethtool.

Reported-by: Kirill Smelkov <kirr@mns.spb.ru>
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

l2tp: fix a race in l2tp_ip_sendmsg()

[ Upstream commit 4399a4df98a63e30fd16e9d0cecc46ea92269e8f ]

Commit 081b1b1bb27f (l2tp: fix l2tp_ip_sendmsg() route handling) added
a race, in case IP route cache is disabled.

In this case, we should not do the dst_release(&rt->dst), since it'll
free the dst immediately, instead of waiting a RCU grace period.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: James Chapman <jchapman@katalix.com>
Cc: Denys Fedoryshchenko <denys@visp.net.lb>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

net: l2tp_eth: fix kernel panic on rmmod l2tp_eth

[ Upstream commit a06998b88b1651c5f71c0e35f528bf2057188ead ]

We must prevent module unloading if some devices are still attached to
l2tp_eth driver.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: Denys Fedoryshchenko <denys@visp.net.lb>
Tested-by: Denys Fedoryshchenko <denys@visp.net.lb>
Cc: James Chapman <jchapman@katalix.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

cipso: handle CIPSO options correctly when NetLabel is disabled

[ Upstream commit 20e2a86485967c385d7c7befc1646e4d1d39362e ]

When NetLabel is not enabled, e.g. CONFIG_NETLABEL=n, and the system
receives a CIPSO tagged packet it is dropped (cipso_v4_validate()
returns non-zero).  In most cases this is the correct and desired
behavior, however, in the case where we are simply forwarding the
traffic, e.g. acting as a network bridge, this becomes a problem.

This patch fixes the forwarding problem by providing the basic CIPSO
validation code directly in ip_options_compile() without the need for
the NetLabel or CIPSO code.  The new validation code can not perform
any of the CIPSO option label/value verification that
cipso_v4_validate() does, but it can verify the basic CIPSO option
format.

The behavior when NetLabel is enabled is unchanged.

Signed-off-by: Paul Moore <pmoore@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

net: sock: validate data_len before allocating skb in sock_alloc_send_pskb()

[ Upstream commit cc9b17ad29ecaa20bfe426a8d4dbfb94b13ff1cc ]

We need to validate the number of pages consumed by data_len, otherwise frags
array could be overflowed by userspace. So this patch validate data_len and
return -EMSGSIZE when data_len may occupies more frags than MAX_SKB_FRAGS.

Signed-off-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

ARM: fix rcu stalls on SMP platforms

commit 7deabca0acfe02b8e18f59a4c95676012f49a304 upstream.

We can stall RCU processing on SMP platforms if a CPU sits in its idle
loop for a long time.  This happens because we don't call irq_enter()
and irq_exit() around generic_smp_call_function_interrupt() and
friends.  Add the necessary calls, and remove the one from within
ipi_timer(), so that they're all in a common place.

Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

udf: Fortify loading of sparing table

commit 1df2ae31c724e57be9d7ac00d78db8a5dabdd050 upstream.

Add sanity checks when loading sparing table from disk to avoid accessing
unallocated memory or writing to it.

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

udf: Avoid run away loop when partition table length is corrupted

commit adee11b2085bee90bd8f4f52123ffb07882d6256 upstream.

Check provided length of partition table so that (possibly maliciously)
corrupted partition table cannot cause accessing data beyond current buffer.

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

udf: Use 'ret' instead of abusing 'i' in udf_load_logicalvol()

commit cb14d340ef1737c24125dd663eff77734a482d47 upstream.

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

mwifiex: fix wrong return values in add_virtual_intf() error cases

commit 858faa57dd9e2b91f3f870fbb1185982e42f5a2b upstream

backported for linux-3.2.y, linux-3.3.y, linux-3.4.y

add_virtual_intf() needs to return an ERR_PTR(), instead of NULL,
on errors, otherwise cfg80211 will crash.

Reported-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: Bing Zhao <bzhao@marvell.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

tracing: change CPU ring buffer state from tracing_cpumask

commit 71babb2705e2203a64c27ede13ae3508a0d2c16c upstream.

According to Documentation/trace/ftrace.txt:

tracing_cpumask:

        This is a mask that lets the user only trace
        on specified CPUS. The format is a hex string
        representing the CPUS.

The tracing_cpumask currently doesn't affect the tracing state of
per-CPU ring buffers.

This patch enables/disables CPU recording as its corresponding bit in
tracing_cpumask is set/unset.

Link: http://lkml.kernel.org/r/1336096792-25373-3-git-send-email-vnagarnaik@google.com
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Laurent Chavey <chavey@google.com>
Cc: Justin Teravest <teravest@google.com>
Cc: David Sharp <dhsharp@google.com>
Signed-off-by: Vaibhav Nagarnaik <vnagarnaik@google.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

ALSA: hda - Fix power-map regression for HP dv6 & co

commit 6e1c39c6b00d9141a82c231ba7c5e5b1716974b2 upstream.

The recent fix for power-map controls (commit b0791dda813) caused
regressions on some other HP laptops.  They have fixed pins but these
pins are exposed as jack-detectable.  Thus the driver tries to control
the power-map dynamically per jack detection where it never gets on.

This patch adds the check of connection and it assumes the no jack
detection is available for fixed pins no matter what pin capability
says.

BugLink: http://bugs.launchpad.net/bugs/1013183
Reported-by: Luis Henriques <luis.henriques@canonical.com>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

dm persistent data: fix allocation failure in space map checker init

commit b0239faaf87c38bb419c9264bf20817438ddc3a9 upstream.

If CONFIG_DM_DEBUG_SPACE_MAPS is enabled and memory is fragmented and a
sufficiently-large metadata device is used in a thin pool then the space
map checker will fail to allocate the memory it requires.

Switch from kmalloc to vmalloc to allow larger virtually contiguous
allocations for the space map checker's internal count arrays.

Reported-by: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

dm persistent data: handle space map checker creation failure

commit 62662303e7f590fdfbb0070ab820a0ad4267c119 upstream.

If CONFIG_DM_DEBUG_SPACE_MAPS is enabled and dm_sm_checker_create()
fails, dm_tm_create_internal() would still return success even though it
cleaned up all resources it was supposed to have created.  This will
lead to a kernel crash:

general protection fault: 0000 [#1] SMP DEBUG_PAGEALLOC
...
RIP: 0010:[<ffffffff81593659>]  [<ffffffff81593659>] dm_bufio_get_block_size+0x9/0x20
Call Trace:
  [<ffffffff81599bae>] dm_bm_block_size+0xe/0x10
  [<ffffffff8159b8b8>] sm_ll_init+0x78/0xd0
  [<ffffffff8159c1a6>] sm_ll_new_disk+0x16/0xa0
  [<ffffffff8159c98e>] dm_sm_disk_create+0xfe/0x160
  [<ffffffff815abf6e>] dm_pool_metadata_open+0x16e/0x6a0
  [<ffffffff815aa010>] pool_ctr+0x3f0/0x900
  [<ffffffff8158d565>] dm_table_add_target+0x195/0x450
  [<ffffffff815904c4>] table_load+0xe4/0x330
  [<ffffffff815917ea>] ctl_ioctl+0x15a/0x2c0
  [<ffffffff81591963>] dm_ctl_ioctl+0x13/0x20
  [<ffffffff8116a4f8>] do_vfs_ioctl+0x98/0x560
  [<ffffffff8116aa51>] sys_ioctl+0x91/0xa0
  [<ffffffff81869f52>] system_call_fastpath+0x16/0x1b

Fix the space map checker code to return an appropriate ERR_PTR and have
dm_sm_disk_create() and dm_tm_create_internal() check for it with
IS_ERR.

Reported-by: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

dm persistent data: fix shadow_info_leak on dm_tm_destroy

commit 25d7cd6faa7ae6ed2565617c3ee2500ccb8a9f7f upstream.

Cleanup the shadow table before destroying the transaction manager.

Reference: leak was identified with kmemleak when running
test_discard_random_sectors in the thinp-test-suite.

Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

drm/i915: kick any firmware framebuffers before claiming the gtt

commit 9f846a16d213523fbe6daea17e20df6b8ac5a1e5 upstream.

Especially vesafb likes to map everything as uc- (yikes), and if that
mapping hangs around still while we try to map the gtt as wc the
kernel will downgrade our request to uc-, resulting in abyssal
performance.

Unfortunately we can't do this as early as readon does (i.e. as the
first thing we do when initializing the hw) because our fb/mmio space
region moves around on a per-gen basis. So I've had to move it below
the gtt initialization, but that seems to work, too. The important
thing is that we do this before we set up the gtt wc mapping.

Now an altogether different question is why people compile their
kernels with vesafb enabled, but I guess making things just work isn't
bad per se ...

v2:
- s/radeondrmfb/inteldrmfb/
- fix up error handling

v3: Kill #ifdef X86, this is Intel after all. Noticed by Ben Widawsky.

v4: Jani Nikula complained about the pointless bool primary
initialization.

v5: Don't oops if we can't allocate, noticed by Chris Wilson.

v6: Resolve conflicts with agp rework and fixup whitespace.

This is commit e188719a2891f01b3100d in drm-next.

Backport to 3.5 -fixes queue requested by Dave Airlie - due to grub
using vesa on fedora their initrd seems to load vesafb before loading
the real kms driver. So tons more people actually experience a
dead-slow gpu. Hence also the Cc: stable.

Reported-and-tested-by: "Kilarski, Bernard R" <bernard.r.kilarski@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Dave Airlie <airlied@redhat.com>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

md/raid10: fix failure when trying to repair a read error.

commit 055d3747dbf00ce85c6872ecca4d466638e80c22 upstream.

commit 58c54fcca3bac5bf9290cfed31c76e4c4bfbabaf
     md/raid10: handle further errors during fix_read_error better.

in 3.1 added "r10_sync_page_io" which takes an IO size in sectors.
But we were passing the IO size in bytes!!!
This resulting in bio_add_page failing, and empty request being sent
down, and a consequent BUG_ON in scsi_lib.

[fix missing space in error message at same time]

This fix is suitable for 3.1.y and later.

Reported-by: Christian Balzer <chibi@gol.com>
Signed-off-by: NeilBrown <neilb@suse.de>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

md/raid5: In ops_run_io, inc nr_pending before calling md_wait_for_blocked_rdev

commit 1850753d2e6d9ca7856581ca5d3cf09521e6a5d7 upstream.

In ops_run_io(), the call to md_wait_for_blocked_rdev will decrement
nr_pending so we lose the reference we hold on the rdev.
So atomic_inc it first to maintain the reference.

This bug was introduced by commit  73e92e51b7969ef5477d
    md/raid5.  Don't write to known bad block on doubtful devices.

which appeared in 3.0, so patch is suitable for stable kernels since
then.

Signed-off-by: majianpeng <majianpeng@gmail.com>
Signed-off-by: NeilBrown <neilb@suse.de>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

md/raid5: Do not add data_offset before call to is_badblock

commit 6c0544e255dd6582a9899572e120fb55d9f672a4 upstream.

In chunk_aligned_read() we are adding data_offset before calling
is_badblock.  But is_badblock also adds data_offset, so that is bad.

So move the addition of data_offset to after the call to
is_badblock.

This bug was introduced by commit 31c176ecdf3563140e639
     md/raid5: avoid reading from known bad blocks.
which first appeared in 3.0.  So that patch is suitable for any
-stable kernel from 3.0.y onwards.  However it will need minor
revision for most of those (as the comment didn't appear until
recently).

Signed-off-by: majianpeng <majianpeng@gmail.com>
Signed-off-by: NeilBrown <neilb@suse.de>
[bwh: Backported to 3.2: ignored missing comment]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

md/raid10: Don't try to recovery unmatched (and unused) chunks.

commit fc448a18ae6219af9a73257b1fbcd009efab4a81 upstream.

If a RAID10 has an odd number of chunks - as might happen when there
are an odd number of devices - the last chunk has no pair and so is
not mirrored.  We don't store data there, but when recovering the last
device in an array we retry to recover that last chunk from a
non-existent location.  This results in an error, and the recovery
aborts.

When we get to that last chunk we should just stop - there is nothing
more to do anyway.

This bug has been present since the introduction of RAID10, so the
patch is appropriate for any -stable kernel.

Reported-by: Christian Balzer <chibi@gol.com>
Tested-by: Christian Balzer <chibi@gol.com>
Signed-off-by: NeilBrown <neilb@suse.de>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

powerpc/kvm: sldi should be sld

commit 2f584a146a2965b82fce89b8d2f95dc5cfe468d0 upstream.

Since we are taking a registers, this should never have been an sldi.
Talking to paulus offline, this is the correct fix.

Was introduced by:
 commit 19ccb76a1938ab364a412253daec64613acbf3df
 Author: Paul Mackerras <paulus@samba.org>
 Date:   Sat Jul 23 17:42:46 2011 +1000

Talking to paulus, this shouldn't be a literal.

Signed-off-by: Michael Neuling <mikey@neuling.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

powerpc/xmon: Use cpumask iterator to avoid warning

commit bc1d7702910c7c7e88eb60b58429dbfe293683ce upstream.

We have a bug report where the kernel hits a warning in the cpumask
code:

WARNING: at include/linux/cpumask.h:107

Which is:
        WARN_ON_ONCE(cpu >= nr_cpumask_bits);

The backtrace is:
        cpu_cmd
        cmds
        xmon_core
        xmon
        die

xmon is iterating through 0 to NR_CPUS. I'm not sure why we are still
open coding this but iterating above nr_cpu_ids is definitely a bug.

This patch iterates through all possible cpus, in case we issue a
system reset and CPUs in an offline state call in.

Perhaps the old code was trying to handle CPUs that were in the
partition but were never started (eg kexec into a kernel with an
nr_cpus= boot option). They are going to die way before we get into
xmon since we haven't set any kernel state up for them.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

ASoC: tlv320aic3x: Fix codec pll configure bug

commit c9fe573a6584034670c1a55ee8162d623519cbbf upstream.

In sound/soc/codecs/tlv320aic3x.c

        data = snd_soc_read(codec, AIC3X_PLL_PROGA_REG);
        snd_soc_write(codec, AIC3X_PLL_PROGA_REG,
                      data | (pll_p << PLLP_SHIFT));

In the above code, pll-p value is OR'ed with previous value without
clearing it. Bug is not seen if pll-p value doesn't change across
Sampling frequency.

However on some platforms (like AM335x EVM-SK), pll-p may have different
values across different sampling frequencies. In such case, above code
configures the pll with a wrong value.
Because of this bug, when a audio stream is played with pll value
different from previous stream, audio is heard as differently(like its
stretched).

Signed-off-by: Hebbar, Gururaja <gururaja.hebbar@ti.com>
Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

mac80211: correct behaviour on unrecognised action frames

commit 4b5ebccc40843104d980f0714bc86bfcd5568941 upstream.

When receiving an "individually addressed" action frame, the
receiver is required to return it to the sender. mac80211
gets this wrong as it also returns group addressed (mcast)
frames to the sender. Fix this and update the reference to
the new 802.11 standards version since things were shuffled
around significantly.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

ath9k: enable serialize_regmode for non-PCIE AR9287

commit 7508b657967cf664b5aa0f6367d05016e7e3bc2a upstream.

https://bugzilla.kernel.org/show_bug.cgi?id=42903

Based on the work of <fynivx@gmail.com>

Signed-off-by: Panayiotis Karabassis <panayk@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

rtlwifi: rtl8192cu: New USB IDs

commit f63d7dabd5da9ef41f28f6d69b29bc084db0ca5a upstream.

The latest Realtek driver for the RTL8188CU and RTL8192CU chips adds three
new USB IDs.

Reported-by: Xose Vazquez Perez <xose.vazquez@gmail.com>
Signed-off-by: Larry Finger <Larry.Finger@lwfinger.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

NFC: Return from rawsock_release when sk is NULL

commit 03e934f620101ca2cfc9383bd76172dd3e1f8567 upstream.

Sasha Levin reported following panic :

[ 2136.383310] BUG: unable to handle kernel NULL pointer dereference at
00000000000003b0
[ 2136.384022] IP: [<ffffffff8114e400>] __lock_acquire+0xc0/0x4b0
[ 2136.384022] PGD 131c4067 PUD 11c0c067 PMD 0
[ 2136.388106] Oops: 0000 [#1] PREEMPT SMP DEBUG_PAGEALLOC
[ 2136.388106] CPU 1
[ 2136.388106] Pid: 24855, comm: trinity-child1 Tainted: G        W
3.5.0-rc2-sasha-00015-g7b268f7 #374
[ 2136.388106] RIP: 0010:[<ffffffff8114e400>]  [<ffffffff8114e400>]
__lock_acquire+0xc0/0x4b0
[ 2136.388106] RSP: 0018:ffff8800130b3ca8  EFLAGS: 00010046
[ 2136.388106] RAX: 0000000000000086 RBX: ffff88001186b000 RCX:
0000000000000000
[ 2136.388106] RDX: 0000000000000000 RSI: 0000000000000000 RDI:
0000000000000000
[ 2136.388106] RBP: ffff8800130b3d08 R08: 0000000000000001 R09:
0000000000000000
[ 2136.388106] R10: 0000000000000000 R11: 0000000000000001 R12:
0000000000000002
[ 2136.388106] R13: 00000000000003b0 R14: 0000000000000000 R15:
0000000000000000
[ 2136.388106] FS:  00007fa5b1bd4700(0000) GS:ffff88001b800000(0000)
knlGS:0000000000000000
[ 2136.388106] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 2136.388106] CR2: 00000000000003b0 CR3: 0000000011d1f000 CR4:
00000000000406e0
[ 2136.388106] DR0: 0000000000000000 DR1: 0000000000000000 DR2:
0000000000000000
[ 2136.388106] DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7:
0000000000000400
[ 2136.388106] Process trinity-child1 (pid: 24855, threadinfo
ffff8800130b2000, task ffff88001186b000)
[ 2136.388106] Stack:
[ 2136.388106]  ffff8800130b3cd8 ffffffff81121785 ffffffff81236774
000080d000000001
[ 2136.388106]  ffff88001b9d6c00 00000000001d6c00 ffffffff130b3d08
ffff88001186b000
[ 2136.388106]  0000000000000000 0000000000000002 0000000000000000
0000000000000000
[ 2136.388106] Call Trace:
[ 2136.388106]  [<ffffffff81121785>] ? sched_clock_local+0x25/0x90
[ 2136.388106]  [<ffffffff81236774>] ? get_empty_filp+0x74/0x220
[ 2136.388106]  [<ffffffff8114e97a>] lock_acquire+0x18a/0x1e0
[ 2136.388106]  [<ffffffff836b37df>] ? rawsock_release+0x4f/0xa0
[ 2136.388106]  [<ffffffff837c0ef0>] _raw_write_lock_bh+0x40/0x80
[ 2136.388106]  [<ffffffff836b37df>] ? rawsock_release+0x4f/0xa0
[ 2136.388106]  [<ffffffff836b37df>] rawsock_release+0x4f/0xa0
[ 2136.388106]  [<ffffffff8321cfe8>] sock_release+0x18/0x70
[ 2136.388106]  [<ffffffff8321d069>] sock_close+0x29/0x30
[ 2136.388106]  [<ffffffff81236bca>] __fput+0x11a/0x2c0
[ 2136.388106]  [<ffffffff81236d85>] fput+0x15/0x20
[ 2136.388106]  [<ffffffff8321de34>] sys_accept4+0x1b4/0x200
[ 2136.388106]  [<ffffffff837c165c>] ? _raw_spin_unlock_irq+0x4c/0x80
[ 2136.388106]  [<ffffffff837c1669>] ? _raw_spin_unlock_irq+0x59/0x80
[ 2136.388106]  [<ffffffff837c2565>] ? sysret_check+0x22/0x5d
[ 2136.388106]  [<ffffffff8321de8b>] sys_accept+0xb/0x10
[ 2136.388106]  [<ffffffff837c2539>] system_call_fastpath+0x16/0x1b
[ 2136.388106] Code: ec 04 00 0f 85 ea 03 00 00 be d5 0b 00 00 48 c7 c7
8a c1 40 84 e8 b1 a5 f8 ff 31 c0 e9 d4 03 00 00 66 2e 0f 1f 84 00 00 00
00 00 <49> 81 7d 00 60 73 5e 85 b8 01 00 00 00 44 0f 44 e0 83 fe 01 77
[ 2136.388106] RIP  [<ffffffff8114e400>] __lock_acquire+0xc0/0x4b0
[ 2136.388106]  RSP <ffff8800130b3ca8>
[ 2136.388106] CR2: 00000000000003b0
[ 2136.388106] ---[ end trace 6d450e935ee18982 ]---
[ 2136.388106] Kernel panic - not syncing: Fatal exception in interrupt

rawsock_release() should test if sock->sk is NULL before calling
sock_orphan()/sock_put()

Reported-by: Sasha Levin <levinsasha928@gmail.com>
Tested-by: Sasha Levin <levinsasha928@gmail.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Samuel Ortiz <sameo@linux.intel.com>
[bwh: Backported to 3.2: keep using nfc_dbg(), not pr_debug()]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

ath9k: fix dynamic WEP related regression

commit bed3d9c0b71f9afbfec905cb6db3b9f16be29d4d upstream.

commit 7a532fe7131216a02c81a6c1b1f8632da1195a58
ath9k_hw: fix interpretation of the rx KeyMiss flag

This commit used the rx key miss indication to detect packets that were
passed from the hardware without being decrypted, however it seems that
this bit is not only undefined in the static WEP case, but also for
dynamically allocated WEP keys. This caused a regression when using
WEP-LEAP.

This patch fixes the regression by keeping track of which key indexes
refer to CCMP keys and only using the key miss indication for those.

Reported-by: Stanislaw Gruszka <sgruszka@redhat.com>
Signed-off-by: Felix Fietkau <nbd@openwrt.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

NFC: Prevent multiple buffer overflows in NCI

commit 67de956ff5dc1d4f321e16cfbd63f5be3b691b43 upstream.

Fix multiple remotely-exploitable stack-based buffer overflows due to
the NCI code pulling length fields directly from incoming frames and
copying too much data into statically-sized arrays.

Signed-off-by: Dan Rosenberg <dan.j.rosenberg@gmail.com>
Cc: security@kernel.org
Cc: Lauro Ramos Venancio <lauro.venancio@openbossa.org>
Cc: Aloisio Almeida Jr <aloisio.almeida@openbossa.org>
Cc: Samuel Ortiz <sameo@linux.intel.com>
Cc: David S. Miller <davem@davemloft.net>
Acked-by: Ilan Elias <ilane@ti.com>
Signed-off-by: Samuel Ortiz <sameo@linux.intel.com>
[bwh: Backported to 3.2:
 - Drop changes to parsing of tech B and tech F parameters
 - Various renaming]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

mwifiex: fix WPS eapol handshake failure

commit f03ba7e9a24e5e9efaad56bd1713b994ea556b16 upstream.

After association, STA will go through eapol handshake with WPS
enabled AP. It's observed that WPS handshake fails with some 11n
AP. The reason for the failure is that the eapol packet is sent
via 11n frame aggregation.

The eapol packet should be sent directly without 11n aggregation.

This patch fixes the problem by adding WPS session control while
dequeuing Tx packets for transmission.

Signed-off-by: Stone Piao <piaoyun@marvell.com>
Signed-off-by: Avinash Patil <patila@marvell.com>
Signed-off-by: Bing Zhao <bzhao@marvell.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
[bwh: Backported to 3.2: reformat the if-statement per earlier
 upstream commit c65a30f35f938b421ac67c34a9e70b0e49e6019a]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

mwifiex: fix 11n rx packet drop issue

commit 925839243dc9aa4ef25305f5afd10ed18258a4ac upstream.

Currently we check the sequence number of last packet received
against start_win. If a sequence hole is detected, start_win is
updated to next sequence number.

Since the rx sequence number is initialized to 0, a corner case
exists when BA setup happens immediately after association. As
0 is a valid sequence number, start_win gets increased to 1
incorrectly. This causes the first packet with sequence number 0
being dropped.

Initialize rx sequence number as 0xffff and skip adjusting
start_win if the sequence number remains 0xffff. The sequence
number will be updated once the first packet is received.

Signed-off-by: Stone Piao <piaoyun@marvell.com>
Signed-off-by: Avinash Patil <patila@marvell.com>
Signed-off-by: Kiran Divekar <dkiran@marvell.com>
Signed-off-by: Bing Zhao <bzhao@marvell.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

umem: fix up unplugging

commit 32587371ad3db2f9d335de10dbd8cffd4fff5669 upstream.

Fix a regression introduced by 7eaceaccab5f40 ("block: remove per-queue
plugging").  In that patch, Jens removed the whole mm_unplug_device()
function, which used to be the trigger to make umem start to work.

We need to implement unplugging to make umem start to work, or I/O will
never be triggered.

Signed-off-by: Tao Guo <Tao.Guo@emc.com>
Cc: Neil Brown <neilb@suse.de>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Shaohua Li <shli@kernel.org>
Acked-by: NeilBrown <neilb@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

splice: fix racy pipe->buffers uses

commit 047fe3605235888f3ebcda0c728cb31937eadfe6 upstream.

Dave Jones reported a kernel BUG at mm/slub.c:3474! triggered
by splice_shrink_spd() called from vmsplice_to_pipe()

commit 35f3d14dbbc5 (pipe: add support for shrinking and growing pipes)
added capability to adjust pipe->buffers.

Problem is some paths don't hold pipe mutex and assume pipe->buffers
doesn't change for their duration.

Fix this by adding nr_pages_max field in struct splice_pipe_desc, and
use it in place of pipe->buffers where appropriate.

splice_shrink_spd() loses its struct pipe_inode_info argument.

Reported-by: Dave Jones <davej@redhat.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Tom Herbert <therbert@google.com>
Tested-by: Dave Jones <davej@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
[bwh: Backported to 3.2:
 - Adjust context in vmsplice_to_pipe()
 - Update one more call to splice_shrink_spd(), from skb_splice_bits()]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

Linux 3.2.22

batman-adv: only drop packets of known wifi clients

commit 5870adc68fc39d81089f1e80efdf64b97e5c37a1 upstream.

bug introduced with 59b699cdee039d75915c354da06937102d1f9a84

If the source or destination mac address of an ethernet packet
could not be found in the translation table the packet was
dropped if AP isolation was turned on. This behavior would
make it impossible to send broadcast packets over the mesh as
the broadcast address will never enter the translation table.

Signed-off-by: Marek Lindner <lindner_marek@yahoo.de>
Acked-by: Antonio Quartulli <ordex@autistici.org>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

acpi_pad: fix power_saving thread deadlock

commit 5f1601261050251a5ca293378b492a69d590dacb upstream.

The acpi_pad driver can get stuck in destroy_power_saving_task()
waiting for kthread_stop() to stop a power_saving thread.  The problem
is that the isolated_cpus_lock mutex is owned when
destroy_power_saving_task() calls kthread_stop(), which waits for a
power_saving thread to end, and the power_saving thread tries to
acquire the isolated_cpus_lock when it calls round_robin_cpu().  This
patch fixes the issue by making round_robin_cpu() use its own mutex.

https://bugzilla.kernel.org/show_bug.cgi?id=42981

Signed-off-by: Stuart Hayes <Stuart_Hayes@Dell.com>
Signed-off-by: Len Brown <len.brown@intel.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

can: flexcan: use be32_to_cpup to handle the value of dt entry

commit 85f2f834e85517307f13e30e630a5fc86f757cb5 upstream.

The freescale arm i.MX series platform can support this driver, and
usually the arm cpu works in the little endian mode by default, while
device tree entry value is stored in big endian format, we should use
be32_to_cpup() to handle them, after modification, it can work well
both on the le cpu and be cpu.

Cc: Shawn Guo <shawn.guo@linaro.org>
Signed-off-by: Hui Wang <jason77.wang@gmail.com>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

xen/netfront: teardown the device before unregistering it.

commit 6bc96d047fe32d76ef79f3195c52a542edf7c705 upstream.

Fixes:
[   15.470311] WARNING: at /local/scratch/ianc/devel/kernels/linux/fs/sysfs/file.c:498 sysfs_attr_ns+0x95/0xa0()
[   15.470326] sysfs: kobject eth0 without dirent
[   15.470333] Modules linked in:
[   15.470342] Pid: 12, comm: xenwatch Not tainted 3.4.0-x86_32p-xenU #93
and
[    9.150554] BUG: unable to handle kernel paging request at 2b359000
[    9.150577] IP: [<c1279561>] linkwatch_do_dev+0x81/0xc0
[    9.150592] *pdpt = 000000002c3c9027 *pde = 0000000000000000
[    9.150604] Oops: 0002 [#1] SMP
[    9.150613] Modules linked in:

This is http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=675190

Reported-by: George Shuklin <george.shuklin@gmail.com>
Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Tested-by: William Dauchy <wdauchy@gmail.com>
Cc: 675190@bugs.debian.org
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

USB: CP210x Add 10 Device IDs

commit 3fcc8f96829776cf181918461923d1e3bbb831a2 upstream.

This patch adds 10 device IDs for CP210x based devices from the following manufacturers:
Timewave
Clipsal
Festo
Link Instruments

Signed-off-by: Craig Shelley <craig@microtron.org.uk>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

USB: option: Add USB ID for Novatel Ovation MC551

commit 065b07e7a14676f4138ce4619d229c0be5a74230 upstream.

This device is also known as the Verizon USB551L.

Signed-off-by: Forest Bond <forest.bond@rapidrollout.com>
Acked-by: Dan Williams <dcbw@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

ALSA: hda - Add Realtek ALC280 codec support

commit befae82e2906cb7155020876a531b0b8c6c8d8c8 upstream.

This chip looks very similar to ALC269 and ALC27* variants. The bug reporter
has verified that sound was working after this patch had been applied.

BugLink: https://bugs.launchpad.net/bugs/1017017
Tested-by: Richard Crossley <richardcrossley@o2.co.uk>
Signed-off-by: David Henningsson <david.henningsson@canonical.com>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

stable: Allow merging of backports for serious user-visible performance issues

commit eb3979f64d25120d60b9e761a4c58f70b1a02f86 upstream.

Distribution kernel maintainers routinely backport fixes for users that
were deemed important but not "something critical" as defined by the
rules. To users of these kernels they are very serious and failing to fix
them reduces the value of -stable.

The problem is that the patches fixing these issues are often subtle and
prone to regressions in other ways and need greater care and attention.
To combat this, these "serious" backports should have a higher barrier
to entry.

This patch relaxes the rules to allow a distribution maintainer to merge
to -stable a backported patch or small series that fixes a "serious"
user-visible performance issue. They should include additional information on
the user-visible bug affected and a link to the bugzilla entry if available.
The same rules about the patch being already in mainline still apply.

Signed-off-by: Mel Gorman <mgorman@suse.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

x86, cpufeature: Rename X86_FEATURE_DTS to X86_FEATURE_DTHERM

commit 4ad33411308596f2f918603509729922a1ec4411 upstream.

It makes sense to label "Digital Thermal Sensor" as "DTS", but
unfortunately the string "dts" was already used for "Debug Store", and
/proc/cpuinfo is a user space ABI.

Therefore, rename this to "dtherm".

This conflict went into mainline via the hwmon tree without any x86
maintainer ack, and without any kind of hint in the subject.

    a4659053 x86/hwmon: fix initialization of coretemp

Reported-by: Jean Delvare <khali@linux-fr.org>
Link: http://lkml.kernel.org/r/4FE34BCB.5050305@linux.intel.com
Cc: Jan Beulich <JBeulich@suse.com>
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
[bwh: Backported to 3.2: drop the coretemp device table change]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

PM / Sleep: Prevent waiting forever on asynchronous suspend after abort

commit 1f758b23177d588a71b96ad02990e715949bb82f upstream.

__device_suspend() must always send a completion. Otherwise, parent
devices will wait forever.

Commit 1e2ef05b, "PM: Limit race conditions between runtime PM and
system sleep (v2)", introduced a regression by short-circuiting the
complete_all() for certain error cases.

This patch fixes the bug by always signalling a completion.

Addresses http://crosbug.com/31972

Tested by injecting an abort.

Signed-off-by: Mandeep Singh Baines <msb@chromium.org>
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

drm/i915: Fix eDP blank screen after S3 resume on HP desktops

commit 6db65cbb941f9d433659bdad02b307f6d94465df upstream.

This patch fixes the problem on some HP desktop machines with eDP
which give blank screens after S3 resume.

It turned out that BLC_PWM_CPU_CTL must be written after
BLC_PWM_CPU_CTL2.  Otherwise it doesn't take effect on these
SNB machines.

Tested with 3.5-rc3 kernel.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=49233

Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

drm/i915: rip out the PM_IIR WARN

commit 58bf8062d0b293b8e1028e5b0342082002886bd4 upstream.

After banging my head against this for the past few months, I still
don't see how this could possible race under the premise that once an
irq bit is masked in PM_IMR and reset in PM_IIR it won't show up again
until we unmask it in PM_IMR.

Still, we have reports of this being seen in the wild. Now Bspec has
this little bit of lovely language in the PMIIR register:

Public SNB Docs, Vol3Part2, 2.5.14 "PMIIR":

"For each bit, the IIR can store a second pending interrupt if two or
more of the same interrupt conditions occur before the first condition
is cleared. Upon clearing the interrupt, the IIR bit will momentarily
go low, then return high to indicate there is another interrupt
pending."

Now if we presume that PMIMR only prevent new interrupts from being
queued, we could easily end up masking an interrupt and clearing it,
but the 2nd pending interrupt setting the bit in PMIIR right away
again. Which leads, the next time the irq handler runs, to hitting the
WARN.

Also, no bad side effects of this have ever been reported. And we've
tracked down our issues with the gpu turbo getting stuck to bogus
interrupt generation limits in th RPLIMIT register.

So let's just rip out this WARN as bogus and call it a day. The only
shallow thing here is that this 2-deep irq queue in the hw makes you
wonder how racy the windows irq handler is ...

Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=42907
Acked-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

drm/i915: Refactor the deferred PM_IIR handling into a single function

commit fc6826d1dcd65f3d1e9a5377678882e4e08f02be upstream.

This function, along with the registers and deferred work hander, are
all shared with SandyBridge, IvyBridge and their variants. So remove the
duplicate code into a single function.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Ben Widawsky <ben@bwidawsk.net>
Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
[bwh: Backported to 3.2: adjust context; drop changes for Valley View]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

oprofile: perf: use NR_CPUS instead or nr_cpumask_bits for static array

commit e734568b675c985db2026848fefaac01c22977a5 upstream.

The OProfile perf backend uses a static array to keep track of the
perf events on the system. When compiling with CONFIG_CPUMASK_OFFSTACK=y
&& SMP, nr_cpumask_bits is not a compile-time constant and the build
will fail with:

oprofile_perf.c:28: error: variably modified 'perf_events' at file scope

This patch uses NR_CPUs instead of nr_cpumask_bits for the array
initialisation. If this causes space problems in the future, we can
always move to dynamic allocation for the events array.

Cc: Matt Fleming <matt@console-pimps.org>
Reported-by: Russell King - ARM Linux <linux@arm.linux.org.uk>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Robert Richter <robert.richter@amd.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

USB: option: add id for Cellient MEN-200

commit 1e2c4e59d2b8797973471b4a287a43eac12a0f40 upstream.

Add vendor and product ID to option.c driver
for Cellient MEN-200 EVDO Rev.B 450MHz data module.
http://cellient.com

Signed-off-by: Dmitry Shmygov <shmygov@rambler.ru>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

ARM: SAMSUNG: Fix for S3C2412 EBI memory mapping

commit 3dca938656c7b0ff6b0717a5dde0f5f45e592be5 upstream.

While upgrading the kernel on a S3C2412 based board I've noted
that it was impossible to boot the board with a 2.6.32 or upper
kernel. I've tracked down the problem to the EBI virtual memory
mapping that is in conflict with the IO mapping definition in
arch/arm/mach-s3c24xx/s3c2412.c.

Signed-off-by: Jose Miguel Goncalves <jose.goncalves@inov.pt>
Signed-off-by: Kukjin Kim <kgene.kim@samsung.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

iwlwifi: remove log_event debugfs file debugging is disabled

commit 882b7b7d11d65e8eccce738f1ce97cdfdb998f9f upstream.

When debugging is disabled, the event log functions aren't
functional in the way that the debugfs file expects. This
leads to the debugfs access crashing. Since the event log
functions aren't functional then, remove the debugfs file
when CONFIG_IWLWIFI_DEBUG is not set.

Reported-by: Lekensteyn <lekensteyn@gmail.com>
Reviewed-by: Emmanuel Grumbach <emmanuel.grumbach@intel.com>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
[bwh: Backported to 3.2: adjust filename, context]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

ath9k_hw: avoid possible infinite loop in ar9003_get_pll_sqsum_dvc

commit f18e3c6b67f448ec47b3a5b242789bd3d5644879 upstream.

"ath9k: Fix softlockup in AR9485" with commit id
64bc1239c790e051ff677e023435d770d2ffa174 fixed the reported
issue, yet its better to avoid the possible infinite loop
in ar9003_get_pll_sqsum_dvc by having a timeout as suggested
by ath9k maintainers.
http://www.spinics.net/lists/linux-wireless/msg92126.html.
Based on my testing PLL's locking measurement is done in
~200us (2 iterations).

Cc: Rolf Offermanns <rolf.offermanns@gmx.net>
Cc: Sujith Manoharan <c_manoha@qca.qualcomm.com>
Cc: Senthil Balasubramanian <senthilb@qca.qualcomm.com>
Signed-off-by: Mohammed Shafi Shajakhan <mohammed@qca.qualcomm.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

ARM: SAMSUNG: Should check for IS_ERR(clk) instead of NULL

commit a5d8f4765f0e92ef027492a8cb979c5b8d45f2c3 upstream.

On the error condition clk_get() returns ERR_PTR().

Signed-off-by: Jonghwan Choi <jhbird.choi@samsung.com>
Signed-off-by: Kukjin Kim <kgene.kim@samsung.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

batman-adv: fix skb->data assignment

commit 2c995ff892313009e336ecc8ec3411022f5b1c39 upstream.

skb_linearize(skb) possibly rearranges the skb internal data and then changes
the skb->data pointer value. For this reason any other pointer in the code that
was assigned skb->data before invoking skb_linearise(skb) must be re-assigned.

In the current tt_query message handling code this is not done and therefore, in
case of skb linearization, the pointer used to handle the packet header ends up
in pointing to free'd memory.

This bug was introduced by a73105b8d4c765d9ebfb664d0a66802127d8e4c7
(batman-adv: improved client announcement mechanism)

Signed-off-by: Antonio Quartulli <ordex@autistici.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
[This patch is a backport for kernel versions 3.1 and 3.2 - Antonio]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

ath9k: fix a tx rate duration calculation bug

commit 76591bea9714a58d8924154068c78d702eb2cb17 upstream.

The rate pointer variable for a rate series is used in a loop before it is
initialized. This went unnoticed because it was used earlier for the RTS/CTS
rate. This bug can lead to the wrong PHY type being passed to the
duration calculation function.

Signed-off-by: Felix Fietkau <nbd@openwrt.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

can: c_can: precedence error in c_can_chip_config()

commit d9cb9bd63eb27ac19f26a8547128c053f43a5da8 upstream.

(CAN_CTRLMODE_LISTENONLY & CAN_CTRLMODE_LOOPBACK) is (0x02 & 0x01) which
is zero so the condition is never true.  The intent here was to test
that both flags were set.

Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Acked-by: Oliver Hartkopp <socketcan@hartkopp.net>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

ath9k: Fix softlockup in AR9485

commit bcb7ad7bcbef030e6ba71ede1f9866368aca7c99 upstream.

steps to recreate:
load latest ath9k driver with AR9485
stop the network-manager and wpa_supplicant
bring the interface up

Call Trace:
[<ffffffffa0517490>] ? ath_hw_check+0xe0/0xe0 [ath9k]
[<ffffffff812cd1e8>] __const_udelay+0x28/0x30
[<ffffffffa03bae7a>] ar9003_get_pll_sqsum_dvc+0x4a/0x80 [ath9k_hw]
[<ffffffffa05174eb>] ath_hw_pll_work+0x5b/0xe0 [ath9k]
[<ffffffff810744fe>] process_one_work+0x11e/0x470
[<ffffffff8107530f>] worker_thread+0x15f/0x360
[<ffffffff810751b0>] ? manage_workers+0x230/0x230
[<ffffffff81079af3>] kthread+0x93/0xa0
[<ffffffff815fd3a4>] kernel_thread_helper+0x4/0x10
[<ffffffff81079a60>] ? kthread_freezable_should_stop+0x70/0x70
[<ffffffff815fd3a0>] ? gs_change+0x13/0x13

ensure that the PLL-WAR for AR9485/AR9340 is executed only if the STA is
associated (or) IBSS/AP mode had started beaconing. Ideally this WAR
is needed to recover from some rare beacon stuck during stress testing.
Before the STA is associated/IBSS had started beaconing, PLL4(0x1618c)
always seem to have zero even though we had configured PLL3(0x16188) to
query about PLL's locking status. When we keep on polling infinitely PLL4's
8th bit(ie check for PLL locking measurements is done), machine hangs
due to softlockup.

fixes https://bugzilla.redhat.com/show_bug.cgi?id=811142

Reported-by: Rolf Offermanns <rolf.offermanns@gmx.net>
Tested-by: Mohammed Shafi Shajakhan <mohammed@qca.qualcomm.com>
Signed-off-by: Mohammed Shafi Shajakhan <mohammed@qca.qualcomm.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

cfg80211: fix potential deadlock in regulatory

commit fe20b39ec32e975f1054c0b7866c873a954adf05 upstream.

reg_timeout_work() calls restore_regulatory_settings() which
takes cfg80211_mutex.

reg_set_request_processed() already holds cfg80211_mutex
before calling cancel_delayed_work_sync(reg_timeout),
so it might deadlock.

Call the async cancel_delayed_work instead, in order
to avoid the potential deadlock.

This is the relevant lockdep warning:

cfg80211: Calling CRDA for country: XX

======================================================
[ INFO: possible circular locking dependency detected ]
3.4.0-rc5-wl+ #26 Not tainted
-------------------------------------------------------
kworker/0:2/1391 is trying to acquire lock:
 (cfg80211_mutex){+.+.+.}, at: [<bf28ae00>] restore_regulatory_settings+0x34/0x418 [cfg80211]

but task is already holding lock:
 ((reg_timeout).work){+.+...}, at: [<c0059e94>] process_one_work+0x1f0/0x480

which lock already depends on the new lock.

the existing dependency chain (in reverse order) is:

-> #2 ((reg_timeout).work){+.+...}:
       [<c008fd44>] validate_chain+0xb94/0x10f0
       [<c0090b68>] __lock_acquire+0x8c8/0x9b0
       [<c0090d40>] lock_acquire+0xf0/0x114
       [<c005b600>] wait_on_work+0x4c/0x154
       [<c005c000>] __cancel_work_timer+0xd4/0x11c
       [<c005c064>] cancel_delayed_work_sync+0x1c/0x20
       [<bf28b274>] reg_set_request_processed+0x50/0x78 [cfg80211]
       [<bf28bd84>] set_regdom+0x550/0x600 [cfg80211]
       [<bf294cd8>] nl80211_set_reg+0x218/0x258 [cfg80211]
       [<c03c7738>] genl_rcv_msg+0x1a8/0x1e8
       [<c03c6a00>] netlink_rcv_skb+0x5c/0xc0
       [<c03c7584>] genl_rcv+0x28/0x34
       [<c03c6720>] netlink_unicast+0x15c/0x228
       [<c03c6c7c>] netlink_sendmsg+0x218/0x298
       [<c03933c8>] sock_sendmsg+0xa4/0xc0
       [<c039406c>] __sys_sendmsg+0x1e4/0x268
       [<c0394228>] sys_sendmsg+0x4c/0x70
       [<c0013840>] ret_fast_syscall+0x0/0x3c

-> #1 (reg_mutex){+.+.+.}:
       [<c008fd44>] validate_chain+0xb94/0x10f0
       [<c0090b68>] __lock_acquire+0x8c8/0x9b0
       [<c0090d40>] lock_acquire+0xf0/0x114
       [<c04734dc>] mutex_lock_nested+0x48/0x320
       [<bf28b2cc>] reg_todo+0x30/0x538 [cfg80211]
       [<c0059f44>] process_one_work+0x2a0/0x480
       [<c005a4b4>] worker_thread+0x1bc/0x2bc
       [<c0061148>] kthread+0x98/0xa4
       [<c0014af4>] kernel_thread_exit+0x0/0x8

-> #0 (cfg80211_mutex){+.+.+.}:
       [<c008ed58>] print_circular_bug+0x68/0x2cc
       [<c008fb28>] validate_chain+0x978/0x10f0
       [<c0090b68>] __lock_acquire+0x8c8/0x9b0
       [<c0090d40>] lock_acquire+0xf0/0x114
       [<c04734dc>] mutex_lock_nested+0x48/0x320
       [<bf28ae00>] restore_regulatory_settings+0x34/0x418 [cfg80211]
       [<bf28b200>] reg_timeout_work+0x1c/0x20 [cfg80211]
       [<c0059f44>] process_one_work+0x2a0/0x480
       [<c005a4b4>] worker_thread+0x1bc/0x2bc
       [<c0061148>] kthread+0x98/0xa4
       [<c0014af4>] kernel_thread_exit+0x0/0x8

other info that might help us debug this:

Chain exists of:
  cfg80211_mutex --> reg_mutex --> (reg_timeout).work

 Possible unsafe locking scenario:

       CPU0                    CPU1
       ----                    ----
  lock((reg_timeout).work);
                               lock(reg_mutex);
                               lock((reg_timeout).work);
  lock(cfg80211_mutex);

 *** DEADLOCK ***

2 locks held by kworker/0:2/1391:
 #0:  (events){.+.+.+}, at: [<c0059e94>] process_one_work+0x1f0/0x480
 #1:  ((reg_timeout).work){+.+...}, at: [<c0059e94>] process_one_work+0x1f0/0x480

stack backtrace:
[<c001b928>] (unwind_backtrace+0x0/0x12c) from [<c0471d3c>] (dump_stack+0x20/0x24)
[<c0471d3c>] (dump_stack+0x20/0x24) from [<c008ef70>] (print_circular_bug+0x280/0x2cc)
[<c008ef70>] (print_circular_bug+0x280/0x2cc) from [<c008fb28>] (validate_chain+0x978/0x10f0)
[<c008fb28>] (validate_chain+0x978/0x10f0) from [<c0090b68>] (__lock_acquire+0x8c8/0x9b0)
[<c0090b68>] (__lock_acquire+0x8c8/0x9b0) from [<c0090d40>] (lock_acquire+0xf0/0x114)
[<c0090d40>] (lock_acquire+0xf0/0x114) from [<c04734dc>] (mutex_lock_nested+0x48/0x320)
[<c04734dc>] (mutex_lock_nested+0x48/0x320) from [<bf28ae00>] (restore_regulatory_settings+0x34/0x418 [cfg80211])
[<bf28ae00>] (restore_regulatory_settings+0x34/0x418 [cfg80211]) from [<bf28b200>] (reg_timeout_work+0x1c/0x20 [cfg80211])
[<bf28b200>] (reg_timeout_work+0x1c/0x20 [cfg80211]) from [<c0059f44>] (process_one_work+0x2a0/0x480)
[<c0059f44>] (process_one_work+0x2a0/0x480) from [<c005a4b4>] (worker_thread+0x1bc/0x2bc)
[<c005a4b4>] (worker_thread+0x1bc/0x2bc) from [<c0061148>] (kthread+0x98/0xa4)
[<c0061148>] (kthread+0x98/0xa4) from [<c0014af4>] (kernel_thread_exit+0x0/0x8)
cfg80211: Calling CRDA to update world regulatory domain
cfg80211: World regulatory domain updated:
cfg80211:   (start_freq - end_freq @ bandwidth), (max_antenna_gain, max_eirp)
cfg80211:   (2402000 KHz - 2472000 KHz @ 40000 KHz), (300 mBi, 2000 mBm)
cfg80211:   (2457000 KHz - 2482000 KHz @ 20000 KHz), (300 mBi, 2000 mBm)
cfg80211:   (2474000 KHz - 2494000 KHz @ 20000 KHz), (300 mBi, 2000 mBm)
cfg80211:   (5170000 KHz - 5250000 KHz @ 40000 KHz), (300 mBi, 2000 mBm)
cfg80211:   (5735000 KHz - 5835000 KHz @ 40000 KHz), (300 mBi, 2000 mBm)

Signed-off-by: Eliad Peller <eliad@wizery.com>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

ath9k: Fix a WARNING on suspend/resume with IBSS

commit 2031b4c2b4904f7448ab9e4bc6b9bf16e32709f5 upstream.

this patch is dependent on the patch "cfg80211: fix interface
combinations"

In ath9k currently we have ADHOC interface as a single incompatible
interface. when drv_add_interface is called during resume we got to
consider number of vifs already present in addition to checking the
drivers 'opmode' information about ADHOC.  we incorrectly assume
an ADHOC interface is already present. Then we may miss some driver
specific data for the ADHOC interface after resume.

The above mentioned checks can be removed from the driver,
as the patch 'cfg80211: fix interface combinations' ensures that
if an interface type is not advertised by the driver in any of the
interface combinations(via ieee80211_iface_combination) then it shall
be treated as a single incompatible interface. Fixes the following
warning on suspend/resume with ibss interface.

        ath: phy0: Cannot create ADHOC interface when other
        interfaces already exist.
        WARNING: at net/mac80211/driver-ops.h:12
        ieee80211_reconfig+0x1882/0x1ca0 [mac80211]()
        Hardware name: 2842RK1
        wlan2:  Failed check-sdata-in-driver check, flags: 0x0

        Call Trace:
        [<c01361b2>] warn_slowpath_common+0x72/0xa0
        [<f8aaa7c2>] ? ieee80211_reconfig+0x1882/0x1ca0
        [mac80211]
        [<f8aaa7c2>] ? ieee80211_reconfig+0x1882/0x1ca0
        [mac80211]
        [<c0136283>] warn_slowpath_fmt+0x33/0x40
        [<f8aaa7c2>] ieee80211_reconfig+0x1882/0x1ca0 [mac80211]
        [<c06c1d1a>] ? mutex_lock_nested+0x23a/0x2f0
        [<f8a95097>] ieee80211_resume+0x27/0x70 [mac80211]
        [<fd177edf>] wiphy_resume+0x8f/0xa0 [cfg80211]

Cc: Rajkumar Manoharan <rmanohar@qca.qualcomm.com>
Signed-off-by: Mohammed Shafi Shajakhan <mohammed@qca.qualcomm.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

dm thin: reinstate missing mempool_free in cell_release_singleton

commit 03aaae7cdc71bc306888440b1f569d463e917b6d upstream.

Fix a significant memory leak inadvertently introduced during
simplification of cell_release_singleton() in commit
6f94a4c45a6f744383f9f695dde019998db3df55 ("dm thin: fix stacked bi_next
usage").

A cell's hlist_del() must be accompanied by a mempool_free().
Use __cell_release() to do this, like before.

Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

drm/nouveau/fbcon: using nv_two_heads is not a good idea

commit 9bd0c15fcfb42f6245447c53347d65ad9e72080b upstream.

nv_two_heads() was never meant to be used outside of pre-nv50 code.  The
code checks for >= NV_10 for 2 CRTCs, then downgrades a few specific
chipsets to 1 CRTC based on (pci_device & 0x0ff0).

The breakage example seen is on GTX 560Ti, with a pciid of 0x1200, which
gets detected as an NV20 (0x020x) with 1 CRTC by nv_two_heads(), causing
memory corruption because there's actually 2 CRTCs..

This switches fbcon to use the CRTC count directly from the mode_config
structure, which will also fix the same issue on Kepler boards which have
4 CRTCs.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

drm/edid: don't return stack garbage from supports_rb

commit b196a4980ff7bb54db478e2a408dc8b12be15304 upstream.

We need to initialize this to false, because the is_rb callback only
ever sets it to true.

Noticed while reading through the code.

Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

smsusb: add autodetection support for USB ID 2040:f5a0

commit 3e1141e2ce5667301a74ca2ef396d9bd5e995f7f upstream.

Signed-off-by: Michael Krufky <mkrufky@linuxtv.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

gspca-core: Fix buffers staying in queued state after a stream_off

commit af05ef01e9cde84620c6855a8d8ab9c8a1db9009 upstream.

This fixes a regression introduced by commit f7059ea and should be
backported to all supported stable kernels which have this commit.

Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Tested-by: Antonio Ospite <ospite@studenti.unina.it>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

staging:rts_pstor:Fix possible panic by NULL pointer dereference

commit 0d05568ac79bfc595f1eadc3e0fd7a20a45f7b69 upstream.

rtsx_transport.c (rtsx_transfer_sglist_adma_partial):
pointer struct scatterlist *sg, which is mapped in dma_map_sg,
is used as an iterator in later transfer operation. It is corrupted and
passed to dma_unmap_sg, thus causing fatal unmap of some erroneous address.
Fix it by duplicating *sg_ptr for iterating.

Signed-off-by: wwang <wei_wang@realsil.com.cn>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

drm/i915: Do the fallback non-IRQ wait in ring throttle, too.

commit 7ea29b13e5e3e1e61e612349eb0366efdb6457f3 upstream.

As a workaround for IRQ synchronization issues in the gen7 BLT ring,
we want to turn the two wait functions into polling loops.

Signed-off-by: Eric Anholt <eric@anholt.net>
Tested-by: Eugeni Dodonov <eugeni.dodonov@intel.com>
Reviewed-by: Eugeni Dodonov <eugeni.dodonov@intel.com>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>
Signed-off-by: Keith Packard <keithp@keithp.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

e1000e: Remove special case for 82573/82574 ASPM L1 disablement

commit 59aed95263bdd0e2b48eb9be5a94346d2d4abf90 upstream.

For the 82573, ASPM L1 gets disabled wholesale so this special-case code
is not required. For the 82574 the previous patch does the same as for
the 82573, disabling L1 on the adapter. Thus, this code is no longer
required and can be removed.

Signed-off-by: Chris Boot <bootc@bootc.net>
Tested-by: Jeff Pieper <jeffrey.e.pieper@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

e1000e: Disable ASPM L1 on 82574

commit id d4a4206ebbaf48b55803a7eb34e330530d83a889

ASPM on the 82574 causes trouble. Currently the driver disables L0s for
this NIC but only disables L1 if the MTU is >1500. This patch simply
causes L1 to be disabled regardless of the MTU setting.

Signed-off-by: Chris Boot <bootc@bootc.net>
Cc: "Wyborny, Carolyn" <carolyn.wyborny@intel.com>
Cc: Nix <nix@esperi.org.uk>
Link: https://lkml.org/lkml/2012/3/19/362
Tested-by: Jeff Pieper <jeffrey.e.pieper@intel.com>
[Jeff Kirsher: Backport to 3.2-3.4 kernels]
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

drm/i915: Remove use of the autoreported ringbuffer HEAD position

This is a revert of 6aa56062eaba67adfb247cded244fd877329588d.

This was originally introduced to workaround reads of the ringbuffer
registers returning 0 on SandyBridge causing hangs due to ringbuffer
overflow. The root cause here was reads through the GT powerwell require
the forcewake dance, something we only learnt of later. Now it appears
that reading the reported head position from the HWS is returning
garbage, leading once again to hangs.

For example, on q35 the autoreported head reports:
  [  217.975608] head now 00010000, actual 00010000
  [  436.725613] head now 00200000, actual 00200000
  [  462.956033] head now 00210000, actual 00210010
  [  485.501409] head now 00400000, actual 00400020
  [  508.064280] head now 00410000, actual 00410000
  [  530.576078] head now 00600000, actual 00600020
  [  553.273489] head now 00610000, actual 00610018
which appears reasonably sane. In contrast, if we look at snb:
  [  141.970680] head now 00e10000, actual 00008238
  [  141.974062] head now 02734000, actual 000083c8
  [  141.974425] head now 00e10000, actual 00008488
  [  141.980374] head now 032b5000, actual 000088b8
  [  141.980885] head now 03271000, actual 00008950
  [  142.040628] head now 02101000, actual 00008b40
  [  142.180173] head now 02734000, actual 00009050
  [  142.181090] head now 00000000, actual 00000ae0
  [  142.183737] head now 02734000, actual 00009050

In addition, the automatic reporting of the head position is scheduled
to be defeatured in the future. It has no more utility, remove it.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=45492
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Tested-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
(cherry picked from commit 5d031e5b633d910f35e6e0abce94d9d842390006)
Signed-off-by: Timo Aaltonen <timo.aaltonen@canonical.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

drm/i915: Finish any pending operations on the framebuffer before disabling

Similar to the case where we are changing from one framebuffer to
another, we need to be sure that there are no pending WAIT_FOR_EVENTs on
the pipe for the current framebuffer before switching. If we disable the
pipe, and then try to execute a WAIT_FOR_EVENT it will block
indefinitely and cause a GPU hang.

We attempted to fix this in commit 85345517fe6d4de27b0d6ca19fef9d28ac947c4a
(drm/i915: Retire any pending operations on the old scanout when switching)
for the case of mode switching, but this leaves the condition where we
are switching off the pipe vulnerable.

There still remains the race condition were a display may be unplugged,
switched off by the core, a uevent sent to notify the DDX and the DDX
may issue a WAIT_FOR_EVENT before it processes the uevent. This window
does not exist if the pipe is only switched off in response to the
uevent. Time to make sure that is so...

Reported-by: Francis Leblanc <Francis.Leblanc-Lebeau@verint.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=36515
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=45413
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Eugeni Dodonov <eugeni.dodonov@intel.com>
[danvet: fixup spelling in comment, noticed by Eugeni.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
(cherry picked from commit 14667a4bde4361b7ac420d68a2e9e9b9b2df5231)
Signed-off-by: Timo Aaltonen <timo.aaltonen@canonical.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

nilfs2: ensure proper cache clearing for gc-inodes

commit fbb24a3a915f105016f1c828476be11aceac8504 upstream.

A gc-inode is a pseudo inode used to buffer the blocks to be moved by
garbage collection.

Block caches of gc-inodes must be cleared every time a garbage collection
function (nilfs_clean_segments) completes.  Otherwise, stale blocks
buffered in the caches may be wrongly reused in successive calls of the GC
function.

For user files, this is not a problem because their gc-inodes are
distinguished by a checkpoint number as well as an inode number.  They
never buffer different blocks if either an inode number, a checkpoint
number, or a block offset differs.

However, gc-inodes of sufile, cpfile and DAT file can store different data
for the same block offset.  Thus, the nilfs_clean_segments function can
move incorrect block for these meta-data files if an old block is cached.
I found this is really causing meta-data corruption in nilfs.

This fixes the issue by ensuring cache clear of gc-inodes and resolves
reported GC problems including checkpoint file corruption, b-tree
corruption, and the following warning during GC.

  nilfs_palloc_freev: entry number 307234 already freed.
  ...

Signed-off-by: Ryusuke Konishi <konishi.ryusuke@lab.ntt.co.jp>
Tested-by: Ryusuke Konishi <konishi.ryusuke@lab.ntt.co.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

thp: avoid atomic64_read in pmd_read_atomic for 32bit PAE

commit e4eed03fd06578571c01d4f1478c874bb432c815 upstream.

In the x86 32bit PAE CONFIG_TRANSPARENT_HUGEPAGE=y case while holding the
mmap_sem for reading, cmpxchg8b cannot be used to read pmd contents under
Xen.

So instead of dealing only with "consistent" pmdvals in
pmd_none_or_trans_huge_or_clear_bad() (which would be conceptually
simpler) we let pmd_none_or_trans_huge_or_clear_bad() deal with pmdvals
where the low 32bit and high 32bit could be inconsistent (to avoid having
to use cmpxchg8b).

The only guarantee we get from pmd_read_atomic is that if the low part of
the pmd was found null, the high part will be null too (so the pmd will be
considered unstable).  And if the low part of the pmd is found "stable"
later, then it means the whole pmd was read atomically (because after a
pmd is stable, neither MADV_DONTNEED nor page faults can alter it anymore,
and we read the high part after the low part).

In the 32bit PAE x86 case, it is enough to read the low part of the pmdval
atomically to declare the pmd as "stable" and that's true for THP and no
THP, furthermore in the THP case we also have a barrier() that will
prevent any inconsistent pmdvals to be cached by a later re-read of the
*pmd.

Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Cc: Jonathan Nieder <jrnieder@gmail.com>
Cc: Ulrich Obergfell <uobergfe@redhat.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Hugh Dickins <hughd@google.com>
Cc: Larry Woodman <lwoodman@redhat.com>
Cc: Petr Matousek <pmatouse@redhat.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Jan Beulich <jbeulich@suse.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@gmail.com>
Tested-by: Andrew Jones <drjones@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

mm: pmd_read_atomic: fix 32bit PAE pmd walk vs pmd_populate SMP race condition

commit 26c191788f18129af0eb32a358cdaea0c7479626 upstream.

When holding the mmap_sem for reading, pmd_offset_map_lock should only
run on a pmd_t that has been read atomically from the pmdp pointer,
otherwise we may read only half of it leading to this crash.

PID: 11679  TASK: f06e8000  CPU: 3   COMMAND: "do_race_2_panic"
 #0 [f06a9dd8] crash_kexec at c049b5ec
 #1 [f06a9e2c] oops_end at c083d1c2
 #2 [f06a9e40] no_context at c0433ded
 #3 [f06a9e64] bad_area_nosemaphore at c043401a
 #4 [f06a9e6c] __do_page_fault at c0434493
 #5 [f06a9eec] do_page_fault at c083eb45
 #6 [f06a9f04] error_code (via page_fault) at c083c5d5
    EAX: 01fb470c EBX: fff35000 ECX: 00000003 EDX: 00000100 EBP:
    00000000
    DS:  007b     ESI: 9e201000 ES:  007b     EDI: 01fb4700 GS:  00e0
    CS:  0060     EIP: c083bc14 ERR: ffffffff EFLAGS: 00010246
 #7 [f06a9f38] _spin_lock at c083bc14
 #8 [f06a9f44] sys_mincore at c0507b7d
 #9 [f06a9fb0] system_call at c083becd
                         start           len
    EAX: ffffffda  EBX: 9e200000  ECX: 00001000  EDX: 6228537f
    DS:  007b      ESI: 00000000  ES:  007b      EDI: 003d0f00
    SS:  007b      ESP: 62285354  EBP: 62285388  GS:  0033
    CS:  0073      EIP: 00291416  ERR: 000000da  EFLAGS: 00000286

This should be a longstanding bug affecting x86 32bit PAE without THP.
Only archs with 64bit large pmd_t and 32bit unsigned long should be
affected.

With THP enabled the barrier() in pmd_none_or_trans_huge_or_clear_bad()
would partly hide the bug when the pmd transition from none to stable,
by forcing a re-read of the *pmd in pmd_offset_map_lock, but when THP is
enabled a new set of problem arises by the fact could then transition
freely in any of the none, pmd_trans_huge or pmd_trans_stable states.
So making the barrier in pmd_none_or_trans_huge_or_clear_bad()
unconditional isn't good idea and it would be a flakey solution.

This should be fully fixed by introducing a pmd_read_atomic that reads
the pmd in order with THP disabled, or by reading the pmd atomically
with cmpxchg8b with THP enabled.

Luckily this new race condition only triggers in the places that must
already be covered by pmd_none_or_trans_huge_or_clear_bad() so the fix
is localized there but this bug is not related to THP.

NOTE: this can trigger on x86 32bit systems with PAE enabled with more
than 4G of ram, otherwise the high part of the pmd will never risk to be
truncated because it would be zero at all times, in turn so hiding the
SMP race.

This bug was discovered and fully debugged by Ulrich, quote:

----
[..]
pmd_none_or_trans_huge_or_clear_bad() loads the content of edx and
eax.

    496 static inline int pmd_none_or_trans_huge_or_clear_bad(pmd_t
    *pmd)
    497 {
    498         /* depend on compiler for an atomic pmd read */
    499         pmd_t pmdval = *pmd;

                                // edi = pmd pointer
0xc0507a74 <sys_mincore+548>:   mov    0x8(%esp),%edi
...
                                // edx = PTE page table high address
0xc0507a84 <sys_mincore+564>:   mov    0x4(%edi),%edx
...
                                // eax = PTE page table low address
0xc0507a8e <sys_mincore+574>:   mov    (%edi),%eax

[..]

Please note that the PMD is not read atomically. These are two "mov"
instructions where the high order bits of the PMD entry are fetched
first. Hence, the above machine code is prone to the following race.

-  The PMD entry {high|low} is 0x0000000000000000.
   The "mov" at 0xc0507a84 loads 0x00000000 into edx.

-  A page fault (on another CPU) sneaks in between the two "mov"
   instructions and instantiates the PMD.

-  The PMD entry {high|low} is now 0x00000003fda38067.
   The "mov" at 0xc0507a8e loads 0xfda38067 into eax.
----

Reported-by: Ulrich Obergfell <uobergfe@redhat.com>
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Hugh Dickins <hughd@google.com>
Cc: Larry Woodman <lwoodman@redhat.com>
Cc: Petr Matousek <pmatouse@redhat.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>