llc: fix info leak via getsockname()

[ Upstream commit 3592aaeb80290bda0f2cf0b5456c97bfc638b192 ]

The LLC code wrongly returns 0, i.e. "success", when the socket is
zapped. Together with the uninitialized uaddrlen pointer argument from
sys_getsockname this leads to an arbitrary memory leak of up to 128
bytes kernel stack via the getsockname() syscall.

Return an error instead when the socket is zapped to prevent the info
leak. Also remove the unnecessary memset(0). We don't directly write to
the memory pointed by uaddr but memcpy() a local structure at the end of
the function that is properly initialized.

Signed-off-by: Mathias Krause <minipli@googlemail.com>
Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

Bluetooth: L2CAP - Fix info leak via getsockname()

[ Upstream commit 792039c73cf176c8e39a6e8beef2c94ff46522ed ]

The L2CAP code fails to initialize the l2_bdaddr_type member of struct
sockaddr_l2 and the padding byte added for alignment. It that for leaks
two bytes kernel stack via the getsockname() syscall. Add an explicit
memset(0) before filling the structure to avoid the info leak.

Signed-off-by: Mathias Krause <minipli@googlemail.com>
Cc: Marcel Holtmann <marcel@holtmann.org>
Cc: Gustavo Padovan <gustavo@padovan.org>
Cc: Johan Hedberg <johan.hedberg@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

Bluetooth: RFCOMM - Fix info leak via getsockname()

[ Upstream commit 9344a972961d1a6d2c04d9008b13617bcb6ec2ef ]

The RFCOMM code fails to initialize the trailing padding byte of struct
sockaddr_rc added for alignment. It that for leaks one byte kernel stack
via the getsockname() syscall. Add an explicit memset(0) before filling
the structure to avoid the info leak.

Signed-off-by: Mathias Krause <minipli@googlemail.com>
Cc: Marcel Holtmann <marcel@holtmann.org>
Cc: Gustavo Padovan <gustavo@padovan.org>
Cc: Johan Hedberg <johan.hedberg@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

Bluetooth: RFCOMM - Fix info leak in ioctl(RFCOMMGETDEVLIST)

[ Upstream commit f9432c5ec8b1e9a09b9b0e5569e3c73db8de432a ]

The RFCOMM code fails to initialize the two padding bytes of struct
rfcomm_dev_list_req inserted for alignment before copying it to
userland. Additionally there are two padding bytes in each instance of
struct rfcomm_dev_info. The ioctl() that for disclosures two bytes plus
dev_num times two bytes uninitialized kernel heap memory.

Allocate the memory using kzalloc() to fix this issue.

Signed-off-by: Mathias Krause <minipli@googlemail.com>
Cc: Marcel Holtmann <marcel@holtmann.org>
Cc: Gustavo Padovan <gustavo@padovan.org>
Cc: Johan Hedberg <johan.hedberg@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

Bluetooth: RFCOMM - Fix info leak in getsockopt(BT_SECURITY)

[ Upstream commit 9ad2de43f1aee7e7274a4e0d41465489299e344b ]

The RFCOMM code fails to initialize the key_size member of struct
bt_security before copying it to userland -- that for leaking one
byte kernel stack. Initialize key_size with 0 to avoid the info
leak.

Signed-off-by: Mathias Krause <minipli@googlemail.com>
Cc: Marcel Holtmann <marcel@holtmann.org>
Cc: Gustavo Padovan <gustavo@padovan.org>
Cc: Johan Hedberg <johan.hedberg@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

Bluetooth: HCI - Fix info leak via getsockname()

[ Upstream commit 3f68ba07b1da811bf383b4b701b129bfcb2e4988 ]

The HCI code fails to initialize the hci_channel member of struct
sockaddr_hci and that for leaks two bytes kernel stack via the
getsockname() syscall. Initialize hci_channel with 0 to avoid the
info leak.

Signed-off-by: Mathias Krause <minipli@googlemail.com>
Cc: Marcel Holtmann <marcel@holtmann.org>
Cc: Gustavo Padovan <gustavo@padovan.org>
Cc: Johan Hedberg <johan.hedberg@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

Bluetooth: HCI - Fix info leak in getsockopt(HCI_FILTER)

[ Upstream commit e15ca9a0ef9a86f0477530b0f44a725d67f889ee ]

The HCI code fails to initialize the two padding bytes of struct
hci_ufilter before copying it to userland -- that for leaking two
bytes kernel stack. Add an explicit memset(0) before filling the
structure to avoid the info leak.

Signed-off-by: Mathias Krause <minipli@googlemail.com>
Cc: Marcel Holtmann <marcel@holtmann.org>
Cc: Gustavo Padovan <gustavo@padovan.org>
Cc: Johan Hedberg <johan.hedberg@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

atm: fix info leak via getsockname()

[ Upstream commit 3c0c5cfdcd4d69ffc4b9c0907cec99039f30a50a ]

The ATM code fails to initialize the two padding bytes of struct
sockaddr_atmpvc inserted for alignment. Add an explicit memset(0)
before filling the structure to avoid the info leak.

Signed-off-by: Mathias Krause <minipli@googlemail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

atm: fix info leak in getsockopt(SO_ATMPVC)

[ Upstream commit e862f1a9b7df4e8196ebec45ac62295138aa3fc2 ]

The ATM code fails to initialize the two padding bytes of struct
sockaddr_atmpvc inserted for alignment. Add an explicit memset(0)
before filling the structure to avoid the info leak.

Signed-off-by: Mathias Krause <minipli@googlemail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

ipv6: addrconf: Avoid calling netdevice notifiers with RCU read-side lock

[ Upstream commit 4acd4945cd1e1f92b20d14e349c6c6a52acbd42d ]

Cong Wang reports that lockdep detected suspicious RCU usage while
enabling IPV6 forwarding:

 [ 1123.310275] ===============================
 [ 1123.442202] [ INFO: suspicious RCU usage. ]
 [ 1123.558207] 3.6.0-rc1+ #109 Not tainted
 [ 1123.665204] -------------------------------
 [ 1123.768254] include/linux/rcupdate.h:430 Illegal context switch in RCU read-side critical section!
 [ 1123.992320]
 [ 1123.992320] other info that might help us debug this:
 [ 1123.992320]
 [ 1124.307382]
 [ 1124.307382] rcu_scheduler_active = 1, debug_locks = 0
 [ 1124.522220] 2 locks held by sysctl/5710:
 [ 1124.648364]  #0:  (rtnl_mutex){+.+.+.}, at: [<ffffffff81768498>] rtnl_trylock+0x15/0x17
 [ 1124.882211]  #1:  (rcu_read_lock){.+.+.+}, at: [<ffffffff81871df8>] rcu_lock_acquire+0x0/0x29
 [ 1125.085209]
 [ 1125.085209] stack backtrace:
 [ 1125.332213] Pid: 5710, comm: sysctl Not tainted 3.6.0-rc1+ #109
 [ 1125.441291] Call Trace:
 [ 1125.545281]  [<ffffffff8109d915>] lockdep_rcu_suspicious+0x109/0x112
 [ 1125.667212]  [<ffffffff8107c240>] rcu_preempt_sleep_check+0x45/0x47
 [ 1125.781838]  [<ffffffff8107c260>] __might_sleep+0x1e/0x19b
[...]
 [ 1127.445223]  [<ffffffff81757ac5>] call_netdevice_notifiers+0x4a/0x4f
[...]
 [ 1127.772188]  [<ffffffff8175e125>] dev_disable_lro+0x32/0x6b
 [ 1127.885174]  [<ffffffff81872d26>] dev_forward_change+0x30/0xcb
 [ 1128.013214]  [<ffffffff818738c4>] addrconf_forward_change+0x85/0xc5
[...]

addrconf_forward_change() uses RCU iteration over the netdev list,
which is unnecessary since it already holds the RTNL lock.  We also
cannot reasonably require netdevice notifier functions not to sleep.

Reported-by: Cong Wang <amwang@redhat.com>
Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

af_packet: remove BUG statement in tpacket_destruct_skb

[ Upstream commit 7f5c3e3a80e6654cf48dfba7cf94f88c6b505467 ]

Here's a quote of the comment about the BUG macro from asm-generic/bug.h:

 Don't use BUG() or BUG_ON() unless there's really no way out; one
 example might be detecting data structure corruption in the middle
 of an operation that can't be backed out of.  If the (sub)system
 can somehow continue operating, perhaps with reduced functionality,
 it's probably not BUG-worthy.

 If you're tempted to BUG(), think again:  is completely giving up
 really the *only* solution?  There are usually better options, where
 users don't need to reboot ASAP and can mostly shut down cleanly.

In our case, the status flag of a ring buffer slot is managed from both sides,
the kernel space and the user space. This means that even though the kernel
side might work as expected, the user space screws up and changes this flag
right between the send(2) is triggered when the flag is changed to
TP_STATUS_SENDING and a given skb is destructed after some time. Then, this
will hit the BUG macro. As David suggested, the best solution is to simply
remove this statement since it cannot be used for kernel side internal
consistency checks. I've tested it and the system still behaves /stable/ in
this case, so in accordance with the above comment, we should rather remove it.

Signed-off-by: Daniel Borkmann <daniel.borkmann@tik.ee.ethz.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

net/core: Fix potential memory leak in dev_set_alias()

[ Upstream commit 7364e445f62825758fa61195d237a5b8ecdd06ec ]

Do not leak memory by updating pointer with potentially NULL realloc return value.

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

Signed-off-by: Alexey Khoroshilov <khoroshilov@ispras.ru>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

pptp: lookup route with the proper net namespace

[ Upstream commit 08252b32311c3fa84219ad794d640af7399b5485 ]

pptp always use init_net as the net namespace to lookup
route, this will cause route lookup failed in container.

because we already set the correct net namespace to struct
sock in pptp_create,so fix this by using sock_net(sk) to
replace &init_net.

Signed-off-by: Gao feng <gaofeng@cn.fujitsu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

isdnloop: fix and simplify isdnloop_init()

[ Upstream commit 77f00f6324cb97cf1df6f9c4aaeea6ada23abdb2 ]

Fix a buffer overflow bug by removing the revision and printk.

[   22.016214] isdnloop-ISDN-driver Rev 1.11.6.7
[   22.097508] isdnloop: (loop0) virtual card added
[   22.174400] Kernel panic - not syncing: stack-protector: Kernel stack is corrupted in: ffffffff83244972
[   22.174400]
[   22.436157] Pid: 1, comm: swapper Not tainted 3.5.0-bisect-00018-gfa8bbb1-dirty #129
[   22.624071] Call Trace:
[   22.720558]  [<ffffffff832448c3>] ? CallcNew+0x56/0x56
[   22.815248]  [<ffffffff8222b623>] panic+0x110/0x329
[   22.914330]  [<ffffffff83244972>] ? isdnloop_init+0xaf/0xb1
[   23.014800]  [<ffffffff832448c3>] ? CallcNew+0x56/0x56
[   23.090763]  [<ffffffff8108e24b>] __stack_chk_fail+0x2b/0x30
[   23.185748]  [<ffffffff83244972>] isdnloop_init+0xaf/0xb1

Signed-off-by: Fengguang Wu <fengguang.wu@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

net_sched: gact: Fix potential panic in tcf_gact().

[ Upstream commit 696ecdc10622d86541f2e35cc16e15b6b3b1b67e ]

gact_rand array is accessed by gact->tcfg_ptype whose value
is assumed to less than MAX_RAND, but any range checks are
not performed.

So add a check in tcf_gact_init(). And in tcf_gact(), we can
reduce a branch.

Signed-off-by: Hiroaki SHIMODA <shimoda.hiroaki@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

tcp: Apply device TSO segment limit earlier

[ Upstream commit 1485348d2424e1131ea42efc033cbd9366462b01 ]

Cache the device gso_max_segs in sock::sk_gso_max_segs and use it to
limit the size of TSO skbs.  This avoids the need to fall back to
software GSO for local TCP senders.

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

sfc: Fix maximum number of TSO segments and minimum TX queue size

[ Upstream commit 7e6d06f0de3f74ca929441add094518ae332257c ]

Currently an skb requiring TSO may not fit within a minimum-size TX
queue.  The TX queue selected for the skb may stall and trigger the TX
watchdog repeatedly (since the problem skb will be retried after the
TX reset).  This issue is designated as CVE-2012-3412.

Set the maximum number of TSO segments for our devices to 100.  This
should make no difference to behaviour unless the actual MSS is less
than about 700.  Increase the minimum TX queue size accordingly to
allow for 2 worst-case skbs, so that there will definitely be space
to add an skb after we wake a queue.

To avoid invalidating existing configurations, change
efx_ethtool_set_ringparam() to fix up values that are too small rather
than returning -EINVAL.

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

net: Allow driver to limit number of GSO segments per skb

[ Upstream commit 30b678d844af3305cda5953467005cebb5d7b687 ]

A peer (or local user) may cause TCP to use a nominal MSS of as little
as 88 (actual MSS of 76 with timestamps).  Given that we have a
sufficiently prodigious local sender and the peer ACKs quickly enough,
it is nevertheless possible to grow the window for such a connection
to the point that we will try to send just under 64K at once.  This
results in a single skb that expands to 861 segments.

In some drivers with TSO support, such an skb will require hundreds of
DMA descriptors; a substantial fraction of a TX ring or even more than
a full ring.  The TX queue selected for the skb may stall and trigger
the TX watchdog repeatedly (since the problem skb will be retried
after the TX reset).  This particularly affects sfc, for which the
issue is designated as CVE-2012-3412.

Therefore:
1. Add the field net_device::gso_max_segs holding the device-specific
   limit.
2. In netif_skb_features(), if the number of segments is too high then
   mask out GSO features to force fall back to software GSO.

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

i2c-designware: Fix build error if CONFIG_I2C_DESIGNWARE_PLATFORM=y && CONFIG_I2C_DESIGNWARE_PCI=y

commit e68bb91baa0bb9817567bd45d560919e8e26373b upstream.

This patch adds config I2C_DESIGNWARE_CORE in Kconfig, and let
I2C_DESIGNWARE_PLATFORM and I2C_DESIGNWARE_PCI select I2C_DESIGNWARE_CORE.

Because both I2C_DESIGNWARE_PLATFORM and I2C_DESIGNWARE_PCI can be built as
built-in or module, we also need to export the functions in i2c-designware-core.

This fixes below build error when CONFIG_I2C_DESIGNWARE_PLATFORM=y &&
CONFIG_I2C_DESIGNWARE_PCI=y:

  LD      drivers/i2c/busses/built-in.o
drivers/i2c/busses/i2c-designware-pci.o: In function `i2c_dw_clear_int':
i2c-designware-core.c:(.text+0xa10): multiple definition of `i2c_dw_clear_int'
drivers/i2c/busses/i2c-designware-platform.o:i2c-designware-platdrv.c:(.text+0x928): first defined here
drivers/i2c/busses/i2c-designware-pci.o: In function `i2c_dw_init':
i2c-designware-core.c:(.text+0x178): multiple definition of `i2c_dw_init'
drivers/i2c/busses/i2c-designware-platform.o:i2c-designware-platdrv.c:(.text+0x90): first defined here
drivers/i2c/busses/i2c-designware-pci.o: In function `dw_readl':
i2c-designware-core.c:(.text+0xe8): multiple definition of `dw_readl'
drivers/i2c/busses/i2c-designware-platform.o:i2c-designware-platdrv.c:(.text+0x0): first defined here
drivers/i2c/busses/i2c-designware-pci.o: In function `i2c_dw_isr':
i2c-designware-core.c:(.text+0x724): multiple definition of `i2c_dw_isr'
drivers/i2c/busses/i2c-designware-platform.o:i2c-designware-platdrv.c:(.text+0x63c): first defined here
drivers/i2c/busses/i2c-designware-pci.o: In function `i2c_dw_xfer':
i2c-designware-core.c:(.text+0x4b0): multiple definition of `i2c_dw_xfer'
drivers/i2c/busses/i2c-designware-platform.o:i2c-designware-platdrv.c:(.text+0x3c8): first defined here
drivers/i2c/busses/i2c-designware-pci.o: In function `i2c_dw_is_enabled':
i2c-designware-core.c:(.text+0x9d4): multiple definition of `i2c_dw_is_enabled'
drivers/i2c/busses/i2c-designware-platform.o:i2c-designware-platdrv.c:(.text+0x8ec): first defined here
drivers/i2c/busses/i2c-designware-pci.o: In function `dw_writel':
i2c-designware-core.c:(.text+0x124): multiple definition of `dw_writel'
drivers/i2c/busses/i2c-designware-platform.o:i2c-designware-platdrv.c:(.text+0x3c): first defined here
drivers/i2c/busses/i2c-designware-pci.o: In function `i2c_dw_xfer_msg':
i2c-designware-core.c:(.text+0x2e8): multiple definition of `i2c_dw_xfer_msg'
drivers/i2c/busses/i2c-designware-platform.o:i2c-designware-platdrv.c:(.text+0x200): first defined here
drivers/i2c/busses/i2c-designware-pci.o: In function `i2c_dw_enable':
i2c-designware-core.c:(.text+0x9c8): multiple definition of `i2c_dw_enable'
drivers/i2c/busses/i2c-designware-platform.o:i2c-designware-platdrv.c:(.text+0x8e0): first defined here
drivers/i2c/busses/i2c-designware-pci.o: In function `i2c_dw_read_comp_param':
i2c-designware-core.c:(.text+0xa24): multiple definition of `i2c_dw_read_comp_param'
drivers/i2c/busses/i2c-designware-platform.o:i2c-designware-platdrv.c:(.text+0x93c): first defined here
drivers/i2c/busses/i2c-designware-pci.o: In function `i2c_dw_disable':
i2c-designware-core.c:(.text+0x9dc): multiple definition of `i2c_dw_disable'
drivers/i2c/busses/i2c-designware-platform.o:i2c-designware-platdrv.c:(.text+0x8f4): first defined here
drivers/i2c/busses/i2c-designware-pci.o: In function `i2c_dw_func':
i2c-designware-core.c:(.text+0x710): multiple definition of `i2c_dw_func'
drivers/i2c/busses/i2c-designware-platform.o:i2c-designware-platdrv.c:(.text+0x628): first defined here
drivers/i2c/busses/i2c-designware-pci.o: In function `i2c_dw_disable_int':
i2c-designware-core.c:(.text+0xa18): multiple definition of `i2c_dw_disable_int'
drivers/i2c/busses/i2c-designware-platform.o:i2c-designware-platdrv.c:(.text+0x930): first defined here
make[3]: *** [drivers/i2c/busses/built-in.o] Error 1
make[2]: *** [drivers/i2c/busses] Error 2
make[1]: *** [drivers/i2c] Error 2
make: *** [drivers] Error 2

Signed-off-by: Axel Lin <axel.lin@gmail.com>
Signed-off-by: Jean Delvare <khali@linux-fr.org>
Tested-by: Jiri Slaby <jslaby@suse.cz>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

drm/vmwgfx: add MODULE_DEVICE_TABLE so vmwgfx loads at boot

commit c4903429a92be60e6fe59868924a65eca4cd1a38 upstream.

This will cause udev to load vmwgfx instead of waiting for X
to do it.

Reviewed-by: Jakob Bornecrantz <jakob@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

CIFS: Fix error handling in cifs_push_mandatory_locks

commit e2f2886a824ff0a56da1eaa13019fde86aa89fa6 upstream.

Signed-off-by: Pavel Shilovsky <pshilovsky@etersoft.ru>
Signed-off-by: Steve French <smfrench@gmail.com>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

Remove user-triggerable BUG from mpol_to_str

commit 80de7c3138ee9fd86a98696fd2cf7ad89b995d0a upstream.

Trivially triggerable, found by trinity:

  kernel BUG at mm/mempolicy.c:2546!
  Process trinity-child2 (pid: 23988, threadinfo ffff88010197e000, task ffff88007821a670)
  Call Trace:
    show_numa_map+0xd5/0x450
    show_pid_numa_map+0x13/0x20
    traverse+0xf2/0x230
    seq_read+0x34b/0x3e0
    vfs_read+0xac/0x180
    sys_pread64+0xa2/0xc0
    system_call_fastpath+0x1a/0x1f
  RIP: mpol_to_str+0x156/0x360

Signed-off-by: Dave Jones <davej@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

xen: Use correct masking in xen_swiotlb_alloc_coherent.

commit b5031ed1be0aa419250557123633453753181643 upstream.

When running 32-bit pvops-dom0 and a driver tries to allocate a coherent
DMA-memory the xen swiotlb-implementation returned memory beyond 4GB.

The underlaying reason is that if the supplied driver passes in a
DMA_BIT_MASK(64) ( hwdev->coherent_dma_mask is set to 0xffffffffffffffff)
our dma_mask will be u64 set to 0xffffffffffffffff even if we set it to
DMA_BIT_MASK(32) previously. Meaning we do not reset the upper bits.
By using the dma_alloc_coherent_mask function - it does the proper casting
and we get 0xfffffffff.

This caused not working sound on a system with 4 GB and a 64-bit
compatible sound-card with sets the DMA-mask to 64bit.

On bare-metal and the forward-ported xen-dom0 patches from OpenSuse a coherent
DMA-memory is always allocated inside the 32-bit address-range by calling
dma_alloc_coherent_mask.

This patch adds the same functionality to xen swiotlb and is a rebase of the
original patch from Ronny Hegewald which never got upstream b/c the
underlaying reason was not understood until now.

The original email with the original patch is in:
http://old-list-archives.xen.org/archives/html/xen-devel/2010-02/msg00038.html
the original thread from where the discussion started is in:
http://old-list-archives.xen.org/archives/html/xen-devel/2010-01/msg00928.html

Signed-off-by: Ronny Hegewald <ronny.hegewald@online.de>
Signed-off-by: Stefano Panella <stefano.panella@citrix.com>
Acked-By: David Vrabel <david.vrabel@citrix.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

udf: Fix data corruption for files in ICB

commit 9c2fc0de1a6e638fe58c354a463f544f42a90a09 upstream.

When a file is stored in ICB (inode), we overwrite part of the file, and
the page containing file's data is not in page cache, we end up corrupting
file's data by overwriting them with zeros. The problem is we use
simple_write_begin() which simply zeroes parts of the page which are not
written to. The problem has been introduced by be021ee4 (udf: convert to
new aops).

Fix the problem by providing a ->write_begin function which makes the page
properly uptodate.

Reported-by: Ian Abbott <abbotti@mev.co.uk>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

powerpc: Make sure IPI handlers see data written by IPI senders

commit 9fb1b36ca1234e64a5d1cc573175303395e3354d upstream.

We have been observing hangs, both of KVM guest vcpu tasks and more
generally, where a process that is woken doesn't properly wake up and
continue to run, but instead sticks in TASK_WAKING state.  This
happens because the update of rq->wake_list in ttwu_queue_remote()
is not ordered with the update of ipi_message in
smp_muxed_ipi_message_pass(), and the reading of rq->wake_list in
scheduler_ipi() is not ordered with the reading of ipi_message in
smp_ipi_demux().  Thus it is possible for the IPI receiver not to see
the updated rq->wake_list and therefore conclude that there is nothing
for it to do.

In order to make sure that anything done before smp_send_reschedule()
is ordered before anything done in the resulting call to scheduler_ipi(),
this adds barriers in smp_muxed_message_pass() and smp_ipi_demux().
The barrier in smp_muxed_message_pass() is a full barrier to ensure that
there is a full ordering between the smp_send_reschedule() caller and
scheduler_ipi().  In smp_ipi_demux(), we use xchg() rather than
xchg_local() because xchg() includes release and acquire barriers.
Using xchg() rather than xchg_local() makes sense given that
ipi_message is not just accessed locally.

This moves the barrier between setting the message and calling the
cause_ipi() function into the individual cause_ipi implementations.
Most of them -- those that used outb, out_8 or similar -- already had
a full barrier because out_8 etc. include a sync before the MMIO
store.  This adds an explicit barrier in the two remaining cases.

These changes made no measurable difference to the speed of IPIs as
measured using a simple ping-pong latency test across two CPUs on
different cores of a POWER7 machine.

The analysis of the reason why processes were not waking up properly
is due to Milton Miller.

Reported-by: Milton Miller <miltonm@bga.com>
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

powerpc/xics: Harden xics hypervisor backend

commit 3ce21cdfe93efffa4ffba9cf3ca2576d3d60d6dc upstream.

During kdump stress testing I sometimes see the kdump kernel panic
with:

  Interrupt 0x306 (real) is invalid, disabling it.
  Kernel panic - not syncing: bad return code EOI - rc = -4, value=ff000306

Instead of panicing print the error message, dump the stack the first
time it happens and continue on. Add some more information to the
debug messages as well.

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>

powerpc: Restore correct DSCR in context switch

commit 714332858bfd40dcf8f741498336d93875c23aa7 upstream.

During a context switch we always restore the per thread DSCR value.
If we aren't doing explicit DSCR management
(ie thread.dscr_inherit == 0) and the default DSCR changed while
the process has been sleeping we end up with the wrong value.

Check thread.dscr_inherit and select the default DSCR or per thread
DSCR as required.

This was found with the following test case, when running with
more threads than CPUs (ie forcing context switching):

http://ozlabs.org/~anton/junkcode/dscr_default_test.c

With the four patches applied I can run a combination of all
test cases successfully at the same time:

http://ozlabs.org/~anton/junkcode/dscr_default_test.c
http://ozlabs.org/~anton/junkcode/dscr_explicit_test.c
http://ozlabs.org/~anton/junkcode/dscr_inherit_test.c

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>

powerpc: Fix DSCR inheritance in copy_thread()

commit 1021cb268b3025573c4811f1dee4a11260c4507b upstream.

If the default DSCR is non zero we set thread.dscr_inherit in
copy_thread() meaning the new thread and all its children will ignore
future updates to the default DSCR. This is not intended and is
a change in behaviour that a number of our users have hit.

We just need to inherit thread.dscr and thread.dscr_inherit from
the parent which ends up being much simpler.

This was found with the following test case:

http://ozlabs.org/~anton/junkcode/dscr_default_test.c

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>

powerpc: Keep thread.dscr and thread.dscr_inherit in sync

commit 00ca0de02f80924dfff6b4f630e1dff3db005e35 upstream.

When we update the DSCR either via emulation of mtspr(DSCR) or via
a change to dscr_default in sysfs we don't update thread.dscr.
We will eventually update it at context switch time but there is
a period where thread.dscr is incorrect.

If we fork at this point we will copy the old value of thread.dscr
into the child. To avoid this, always keep thread.dscr in sync with
reality.

This issue was found with the following testcase:

http://ozlabs.org/~anton/junkcode/dscr_inherit_test.c

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>

powerpc: Update DSCR on all CPUs when writing sysfs dscr_default

commit 1b6ca2a6fe56e7697d57348646e07df08f43b1bb upstream.

Writing to dscr_default in sysfs doesn't actually change the DSCR -
we rely on a context switch on each CPU to do the work. There is no
guarantee we will get a context switch in a reasonable amount of time
so fire off an IPI to force an immediate change.

This issue was found with the following test case:

http://ozlabs.org/~anton/junkcode/dscr_explicit_test.c

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

mmc: card: Skip secure erase on MoviNAND; causes unrecoverable corruption.

commit 3550ccdb9d8d350e526b809bf3dd92b550a74fe1 upstream.

For several MoviNAND eMMC parts, there are known issues with secure
erase and secure trim.  For these specific MoviNAND devices, we skip
these operations.

Specifically, there is a bug in the eMMC firmware that causes
unrecoverable corruption when the MMC is erased with MMC_CAP_ERASE
enabled.

References:

http://forum.xda-developers.com/showthread.php?t=1644364
https://plus.google.com/111398485184813224730/posts/21pTYfTsCkB#111398485184813224730/posts/21pTYfTsCkB

Signed-off-by: Ian Chen <ian.cy.chen@samsung.com>
Reviewed-by: Namjae Jeon <linkinjeon@gmail.com>
Acked-by: Jaehoon Chung <jh80.chung@samsung.com>
Reviewed-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: Chris Ball <cjb@laptop.org>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

mmc: sdhci-esdhc: break out early if clock is 0

commit 74f330bceaa7b88d06062e1cac3d519a3dfc041e upstream.

Since commit 30832ab56 ("mmc: sdhci: Always pass clock request value
zero to set_clock host op") was merged, esdhc_set_clock starts hitting
"if (clock == 0)" where ESDHC_SYSTEM_CONTROL has been operated.  This
causes SDHCI card-detection function being broken.  Fix the regression
by moving "if (clock == 0)" above ESDHC_SYSTEM_CONTROL operation.

Signed-off-by: Shawn Guo <shawn.guo@linaro.org>
Signed-off-by: Chris Ball <cjb@laptop.org>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

mmc: mxs-mmc: fix deadlock caused by recursion loop

commit fc108d24d3a6da63576a460e122fa1df0cbdea20 upstream.

Release the lock before mmc_signal_sdio_irq is called by
mxs_mmc_enable_sdio_irq.

Backtrace:
[   65.470000] =============================================
[   65.470000] [ INFO: possible recursive locking detected ]
[   65.470000] 3.5.0-rc5 #2 Not tainted
[   65.470000] ---------------------------------------------
[   65.470000] ksdioirqd/mmc0/73 is trying to acquire lock:
[   65.470000]  (&(&host->lock)->rlock#2){-.-...}, at: [<bf054120>] mxs_mmc_enable_sdio_irq+0x18/0xdc [mxs_mmc]
[   65.470000]
[   65.470000] but task is already holding lock:
[   65.470000]  (&(&host->lock)->rlock#2){-.-...}, at: [<bf054120>] mxs_mmc_enable_sdio_irq+0x18/0xdc [mxs_mmc]
[   65.470000]
[   65.470000] other info that might help us debug this:
[   65.470000]  Possible unsafe locking scenario:
[   65.470000]
[   65.470000]        CPU0
[   65.470000]        ----
[   65.470000]   lock(&(&host->lock)->rlock#2);
[   65.470000]   lock(&(&host->lock)->rlock#2);
[   65.470000]
[   65.470000]  *** DEADLOCK ***
[   65.470000]
[   65.470000]  May be due to missing lock nesting notation
[   65.470000]
[   65.470000] 1 lock held by ksdioirqd/mmc0/73:
[   65.470000]  #0:  (&(&host->lock)->rlock#2){-.-...}, at: [<bf054120>] mxs_mmc_enable_sdio_irq+0x18/0xdc [mxs_mmc]
[   65.470000]
[   65.470000] stack backtrace:
[   65.470000] [<c0014990>] (unwind_backtrace+0x0/0xf4) from [<c005ccb8>] (__lock_acquire+0x14f8/0x1b98)
[   65.470000] [<c005ccb8>] (__lock_acquire+0x14f8/0x1b98) from [<c005d3f8>] (lock_acquire+0xa0/0x108)
[   65.470000] [<c005d3f8>] (lock_acquire+0xa0/0x108) from [<c02f671c>] (_raw_spin_lock_irqsave+0x48/0x5c)
[   65.470000] [<c02f671c>] (_raw_spin_lock_irqsave+0x48/0x5c) from [<bf054120>] (mxs_mmc_enable_sdio_irq+0x18/0xdc [mxs_mmc])
[   65.470000] [<bf054120>] (mxs_mmc_enable_sdio_irq+0x18/0xdc [mxs_mmc]) from [<bf0541d0>] (mxs_mmc_enable_sdio_irq+0xc8/0xdc [mxs_mmc])
[   65.470000] [<bf0541d0>] (mxs_mmc_enable_sdio_irq+0xc8/0xdc [mxs_mmc]) from [<c0219b38>] (sdio_irq_thread+0x1bc/0x274)
[   65.470000] [<c0219b38>] (sdio_irq_thread+0x1bc/0x274) from [<c003c324>] (kthread+0x8c/0x98)
[   65.470000] [<c003c324>] (kthread+0x8c/0x98) from [<c00101ac>] (kernel_thread_exit+0x0/0x8)
[   65.470000] BUG: spinlock lockup suspected on CPU#0, ksdioirqd/mmc0/73
[   65.470000]  lock: 0xc3358724, .magic: dead4ead, .owner: ksdioirqd/mmc0/73, .owner_cpu: 0
[   65.470000] [<c0014990>] (unwind_backtrace+0x0/0xf4) from [<c01b46b0>] (do_raw_spin_lock+0x100/0x144)
[   65.470000] [<c01b46b0>] (do_raw_spin_lock+0x100/0x144) from [<c02f6724>] (_raw_spin_lock_irqsave+0x50/0x5c)
[   65.470000] [<c02f6724>] (_raw_spin_lock_irqsave+0x50/0x5c) from [<bf054120>] (mxs_mmc_enable_sdio_irq+0x18/0xdc [mxs_mmc])
[   65.470000] [<bf054120>] (mxs_mmc_enable_sdio_irq+0x18/0xdc [mxs_mmc]) from [<bf0541d0>] (mxs_mmc_enable_sdio_irq+0xc8/0xdc [mxs_mmc])
[   65.470000] [<bf0541d0>] (mxs_mmc_enable_sdio_irq+0xc8/0xdc [mxs_mmc]) from [<c0219b38>] (sdio_irq_thread+0x1bc/0x274)
[   65.470000] [<c0219b38>] (sdio_irq_thread+0x1bc/0x274) from [<c003c324>] (kthread+0x8c/0x98)
[   65.470000] [<c003c324>] (kthread+0x8c/0x98) from [<c00101ac>] (kernel_thread_exit+0x0/0x8)

Reported-by: Attila Kinali <attila@kinali.ch>
Signed-off-by: Lauri Hintsala <lauri.hintsala@bluegiga.com>
Acked-by: Shawn Guo <shawn.guo@linaro.org>
Signed-off-by: Chris Ball <cjb@laptop.org>
[bwh: Backported to 3.2:
 - Adjust context
 - HW_SSP_STATUS is a simple rather than function-like macro]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

mmc: mxs-mmc: fix deadlock in SDIO IRQ case

commit 1af36b2a993dddfa3d6860ec4879c9e8abc9b976 upstream.

Release the lock before mmc_signal_sdio_irq is called by mxs_mmc_irq_handler.

Backtrace:
[   79.660000] =============================================
[   79.660000] [ INFO: possible recursive locking detected ]
[   79.660000] 3.4.0-00009-g3e96082-dirty #11 Not tainted
[   79.660000] ---------------------------------------------
[   79.660000] swapper/0 is trying to acquire lock:
[   79.660000]  (&(&host->lock)->rlock#2){-.....}, at: [<c026ea3c>] mxs_mmc_enable_sdio_irq+0x18/0xd4
[   79.660000]
[   79.660000] but task is already holding lock:
[   79.660000]  (&(&host->lock)->rlock#2){-.....}, at: [<c026f744>] mxs_mmc_irq_handler+0x1c/0xe8
[   79.660000]
[   79.660000] other info that might help us debug this:
[   79.660000]  Possible unsafe locking scenario:
[   79.660000]
[   79.660000]        CPU0
[   79.660000]        ----
[   79.660000]   lock(&(&host->lock)->rlock#2);
[   79.660000]   lock(&(&host->lock)->rlock#2);
[   79.660000]
[   79.660000]  *** DEADLOCK ***
[   79.660000]
[   79.660000]  May be due to missing lock nesting notation
[   79.660000]
[   79.660000] 1 lock held by swapper/0:
[   79.660000]  #0:  (&(&host->lock)->rlock#2){-.....}, at: [<c026f744>] mxs_mmc_irq_handler+0x1c/0xe8
[   79.660000]
[   79.660000] stack backtrace:
[   79.660000] [<c0014bd0>] (unwind_backtrace+0x0/0xf4) from [<c005f9c0>] (__lock_acquire+0x1948/0x1d48)
[   79.660000] [<c005f9c0>] (__lock_acquire+0x1948/0x1d48) from [<c005fea0>] (lock_acquire+0xe0/0xf8)
[   79.660000] [<c005fea0>] (lock_acquire+0xe0/0xf8) from [<c03a8460>] (_raw_spin_lock_irqsave+0x44/0x58)
[   79.660000] [<c03a8460>] (_raw_spin_lock_irqsave+0x44/0x58) from [<c026ea3c>] (mxs_mmc_enable_sdio_irq+0x18/0xd4)
[   79.660000] [<c026ea3c>] (mxs_mmc_enable_sdio_irq+0x18/0xd4) from [<c026f7fc>] (mxs_mmc_irq_handler+0xd4/0xe8)
[   79.660000] [<c026f7fc>] (mxs_mmc_irq_handler+0xd4/0xe8) from [<c006bdd8>] (handle_irq_event_percpu+0x70/0x254)
[   79.660000] [<c006bdd8>] (handle_irq_event_percpu+0x70/0x254) from [<c006bff8>] (handle_irq_event+0x3c/0x5c)
[   79.660000] [<c006bff8>] (handle_irq_event+0x3c/0x5c) from [<c006e6d0>] (handle_level_irq+0x90/0x110)
[   79.660000] [<c006e6d0>] (handle_level_irq+0x90/0x110) from [<c006b930>] (generic_handle_irq+0x38/0x50)
[   79.660000] [<c006b930>] (generic_handle_irq+0x38/0x50) from [<c00102fc>] (handle_IRQ+0x30/0x84)
[   79.660000] [<c00102fc>] (handle_IRQ+0x30/0x84) from [<c000f058>] (__irq_svc+0x38/0x60)
[   79.660000] [<c000f058>] (__irq_svc+0x38/0x60) from [<c0010520>] (default_idle+0x2c/0x40)
[   79.660000] [<c0010520>] (default_idle+0x2c/0x40) from [<c0010a90>] (cpu_idle+0x64/0xcc)
[   79.660000] [<c0010a90>] (cpu_idle+0x64/0xcc) from [<c04ff858>] (start_kernel+0x244/0x2c8)
[   79.660000] BUG: spinlock lockup on CPU#0, swapper/0
[   79.660000]  lock: c398cb2c, .magic: dead4ead, .owner: swapper/0, .owner_cpu: 0
[   79.660000] [<c0014bd0>] (unwind_backtrace+0x0/0xf4) from [<c01ddb1c>] (do_raw_spin_lock+0xf0/0x144)
[   79.660000] [<c01ddb1c>] (do_raw_spin_lock+0xf0/0x144) from [<c03a8468>] (_raw_spin_lock_irqsave+0x4c/0x58)
[   79.660000] [<c03a8468>] (_raw_spin_lock_irqsave+0x4c/0x58) from [<c026ea3c>] (mxs_mmc_enable_sdio_irq+0x18/0xd4)
[   79.660000] [<c026ea3c>] (mxs_mmc_enable_sdio_irq+0x18/0xd4) from [<c026f7fc>] (mxs_mmc_irq_handler+0xd4/0xe8)
[   79.660000] [<c026f7fc>] (mxs_mmc_irq_handler+0xd4/0xe8) from [<c006bdd8>] (handle_irq_event_percpu+0x70/0x254)
[   79.660000] [<c006bdd8>] (handle_irq_event_percpu+0x70/0x254) from [<c006bff8>] (handle_irq_event+0x3c/0x5c)
[   79.660000] [<c006bff8>] (handle_irq_event+0x3c/0x5c) from [<c006e6d0>] (handle_level_irq+0x90/0x110)
[   79.660000] [<c006e6d0>] (handle_level_irq+0x90/0x110) from [<c006b930>] (generic_handle_irq+0x38/0x50)
[   79.660000] [<c006b930>] (generic_handle_irq+0x38/0x50) from [<c00102fc>] (handle_IRQ+0x30/0x84)
[   79.660000] [<c00102fc>] (handle_IRQ+0x30/0x84) from [<c000f058>] (__irq_svc+0x38/0x60)
[   79.660000] [<c000f058>] (__irq_svc+0x38/0x60) from [<c0010520>] (default_idle+0x2c/0x40)
[   79.660000] [<c0010520>] (default_idle+0x2c/0x40) from [<c0010a90>] (cpu_idle+0x64/0xcc)
[   79.660000] [<c0010a90>] (cpu_idle+0x64/0xcc) from [<c04ff858>] (start_kernel+0x244/0x2c8)

Signed-off-by: Lauri Hintsala <lauri.hintsala@bluegiga.com>
Acked-by: Shawn Guo <shawn.guo@linaro.org>
Signed-off-by: Chris Ball <cjb@laptop.org>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

fuse: fix retrieve length

commit c9e67d483776d8d2a5f3f70491161b205930ffe1 upstream.

In some cases fuse_retrieve() would return a short byte count if offset was
non-zero.  The data returned was correct, though.

Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

UBI: fix a horrible memory deallocation bug

commit 78b495c39add820ab66ab897af9bd77a5f2e91f6 upstream.

UBI was mistakingly using 'kfree()' instead of 'kmem_cache_free()' when
freeing "attach eraseblock" structures in vtbl.c. Thankfully, this happened
only when we were doing auto-format, so many systems were unaffected. However,
there are still many users affected.

It is strange, but the system did not crash and nothing bad happened when
the SLUB memory allocator was used. However, in case of SLOB we observed an
crash right away.

This problem was introduced in 2.6.39 by commit
"6c1e875 UBI: add slab cache for ubi_scan_leb objects"

A note for stable trees:
  Because variable were renamed, this won't cleanly apply to older kernels.
  Changing names like this should help:
1. ai -> si
2. aeb_slab_cache -> seb_slab_cache
3. new_aeb -> new_seb

Reported-by: Richard Genoud <richard.genoud@gmail.com>
Tested-by: Richard Genoud <richard.genoud@gmail.com>
Tested-by: Artem Bityutskiy <artem.bityutskiy@linux.intel.com>
Signed-off-by: Artem Bityutskiy <artem.bityutskiy@linux.intel.com>
[bwh: Backported to 3.2: aeb_slab_cache was actually named scan_leb_slab]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

ext3: Fix fdatasync() for files with only i_size changes

commit 156bddd8e505b295540f3ca0e27dda68cb0d49aa upstream.

Code tracking when transaction needs to be committed on fdatasync(2) forgets
to handle a situation when only inode's i_size is changed. Thus in such
situations fdatasync(2) doesn't force transaction with new i_size to disk
and that can result in wrong i_size after a crash.

Fix the issue by updating inode's i_datasync_tid whenever its size is
updated.

Reported-by: Kristian Nielsen <knielsen@knielsen-hq.org>
Signed-off-by: Jan Kara <jack@suse.cz>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

e1000e: DoS while TSO enabled caused by link partner with small MSS

commit d821a4c4d11ad160925dab2bb009b8444beff484 upstream.

With a low enough MSS on the link partner and TSO enabled locally, the
networking stack can periodically send a very large (e.g.  64KB) TCP
message for which the driver will attempt to use more Tx descriptors than
are available by default in the Tx ring.  This is due to a workaround in
the code that imposes a limit of only 4 MSS-sized segments per descriptor
which appears to be a carry-over from the older e1000 driver and may be
applicable only to some older PCI or PCIx parts which are not supported in
e1000e.  When the driver gets a message that is too large to fit across the
configured number of Tx descriptors, it stops the upper stack from queueing
any more and gets stuck in this state.  After a timeout, the upper stack
assumes the adapter is hung and calls the driver to reset it.

Remove the unnecessary limitation of using up to only 4 MSS-sized segments
per Tx descriptor, and put in a hard failure test to catch when attempting
to check for message sizes larger than would fit in the whole Tx ring.
Refactor the remaining logic that limits the size of data per Tx descriptor
from a seemingly arbitrary 8KB to a limit based on the dynamic size of the
Tx packet buffer as described in the hardware specification.

Also, fix the logic in the check for space in the Tx ring for the next
largest possible packet after the current one has been successfully queued
for transmit, and use the appropriate defines for default ring sizes in
e1000_probe instead of magic values.

This issue goes back to the introduction of e1000e in 2.6.24 when it was
split off from e1000.

Reported-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: Bruce Allan <bruce.w.allan@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
[bwh: Backported to 3.2:
 - Adjust context
 - Adjust for use of net_device vs e1000_ring parameter]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

drm: Add EDID_QUIRK_FORCE_REDUCED_BLANKING for ASUS VW222S

commit 6f33814bd4d9cfe76033a31b1c0c76c960cd8e4b upstream.

Connecting an ASUS VW222S [1] over VGA a garbled screen is shown with
vertical stripes in the top half.

In commit bc42aabc [2]

        commit bc42aabc6a01b92b0f961d65671564e0e1cd7592
        Author: Adam Jackson <ajax@redhat.com>
        Date:   Wed May 23 16:26:54 2012 -0400

            drm/edid/quirks: ViewSonic VA2026w

Adam Jackson added the quirk `EDID_QUIRK_FORCE_REDUCED_BLANKING` which
is also needed for this ASUS monitor.

All log files and output from `xrandr` is included in the referenced
Bugzilla report #17629.

Please note that this monitor only has a VGA (D-Sub) connector [1].

[1] http://www.asus.com/Display/LCD_Monitors/VW222S/
[2] http://git.kernel.org/?p=linux/kernel/git/torvalds/linux.git;a=commit;h=bc42aabc6a01b92b0f961d65671564e0e1cd7592

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=17629
Signed-off-by: Paul Menzel <paulepanter@users.sourceforge.net>
Cc: <dri-devel@lists.freedesktop.org>
Cc: Adam Jackson <ajax@redhat.com>
Cc: Ian Pilcher <arequipeno@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

drm/edid/quirks: ViewSonic VA2026w

commit bc42aabc6a01b92b0f961d65671564e0e1cd7592 upstream.

Entirely new class of fail for this one.  The detailed timings are for
normal CVT but the monitor really wanted CVT-R.

Bugzilla: http://bugzilla.redhat/com/516471
Signed-off-by: Adam Jackson <ajax@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

drm/radeon: force dma32 to fix regression rs4xx,rs6xx,rs740

commit 4a2b6662c3632176b4fdf012243dd3751367bf1f upstream.

It seems some of those IGP dislike non dma32 page despite what
documentation says. Fix regression since we allowed non dma32
pages. It seems it only affect some revision of those IGP chips
as we don't know which one just force dma32 for all of them.

https://bugzilla.redhat.com/show_bug.cgi?id=785375

Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

drm/radeon/atom: rework DIG modesetting on DCE3+

commit 8d1af57ae3c4458ed0de93ef97f388dd1b3239c7 upstream.

The ordering is important and the current drm code
wasn't cutting it for modern DIG encoders.  We need
to have information about crtc before setting up
the encoders so I've shifted the ordering a bit.
Probably we'll need a full rework akin to danvet's
recent intel patchs.  This patch fixes numerous
issues with DP bridge chips and makes link training
much more reliable.

Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[bwh: Backported to 3.2: drop DCE6 cases]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

drm/radeon: don't disable plls that are in use by other crtcs

commit 4e58591c8961b3e31709313f75819f2eec06e322 upstream.

Some plls are shared for DP.

Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
[bwh: Backported to 3.2: add the dev and rdev variables, previously added
 upstream to support DCE6.1 which isn't supported in this version]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

HID: add NOGET quirk for Eaton Ellipse MAX UPS

commit 67ddbb3e6568fb1820b2cc45b00c50702b114801 upstream.

This patch (as1603) adds a NOGET quirk for the Eaton Ellipse MAX UPS
device.  (The USB IDs were already present in hid-ids.h, apparently
under a different name.)

Signed-off-by: Alan Stern <stern@rowland.harvard.edu>
Reported-by: Laurent Bigonville <l.bigonville@edpnet.be>
Signed-off-by: Jiri Kosina <jkosina@suse.cz>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

HID: multitouch: support PixArt optical touch screen

commit b7ea95ff9baab144dacdc30d752307938c5ab6bf upstream.

This patch modifies hid-multitouch driver for supporting PixArt optical touch
screen.  Because of the device does not have to set initial report, we apply
"HID_QUIRK_NO_INIT_REPORTS" quirk and add the device into hid_blacklist[]

Signed-off-by: Aaron Tian <aaron_tian@pixart.com.tw>
Signed-off-by: Jiri Kosina <jkosina@suse.cz>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

drm: Check for invalid cursor flags

commit 7c4eaca4162d0b5ad4fb39f974d7ffd71b9daa09 upstream.

Signed-off-by: Jakob Bornecrantz <jakob@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

drm: remove some potentially dangerous DRM_ERRORs

commit acb4b992d8a13728044f430b47b9199aa45993e9 upstream.

Each of these error messages can be caused by a broken or malicious
userspace wanting to spam the dmesg with useless info.  They're really
not worthy of DRM_DEBUG statements either; those are generally only
useful during bringup of new hardware or versions, and ought to be
removed before going upstream anyway.

Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Reviewed-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
[bwh: Backported to 3.2: s/r\./r->/ in drm_mode_addfb()]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

ARM: imx: select CPU_FREQ_TABLE when needed

commit f637c4c9405e21f44cf0045eaf77eddd3a79ca5a upstream.

The i.MX cpufreq implementation uses the CPU_FREQ_TABLE helpers,
so it needs to select that code to be built. This problem has
apparently existed since the i.MX cpufreq code was first merged
in v2.6.37.

Building IMX without CPU_FREQ_TABLE results in:

arch/arm/plat-mxc/built-in.o: In function `mxc_cpufreq_exit':
arch/arm/plat-mxc/cpufreq.c:173: undefined reference to `cpufreq_frequency_table_put_attr'
arch/arm/plat-mxc/built-in.o: In function `mxc_set_target':
arch/arm/plat-mxc/cpufreq.c:84: undefined reference to `cpufreq_frequency_table_target'
arch/arm/plat-mxc/built-in.o: In function `mxc_verify_speed':
arch/arm/plat-mxc/cpufreq.c:65: undefined reference to `cpufreq_frequency_table_verify'
arch/arm/plat-mxc/built-in.o: In function `mxc_cpufreq_init':
arch/arm/plat-mxc/cpufreq.c:154: undefined reference to `cpufreq_frequency_table_cpuinfo'
arch/arm/plat-mxc/cpufreq.c:162: undefined reference to `cpufreq_frequency_table_get_attr'

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Acked-by: Shawn Guo <shawn.guo@linaro.org>
Cc: Sascha Hauer <s.hauer@pengutronix.de>
Cc: Yong Shen <yong.shen@linaro.org>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

xen/setup: Fix one-off error when adding for-balloon PFNs to the P2M.

commit c96aae1f7f393387d160211f60398d58463a7e65 upstream.

When we are finished with return PFNs to the hypervisor, then
populate it back, and also mark the E820 MMIO and E820 gaps
as IDENTITY_FRAMEs, we then call P2M to set areas that can
be used for ballooning. We were off by one, and ended up
over-writting a P2M entry that most likely was an IDENTITY_FRAME.
For example:

1-1 mapping on 40000->40200
1-1 mapping on bc558->bc5ac
1-1 mapping on bc5b4->bc8c5
1-1 mapping on bc8c6->bcb7c
1-1 mapping on bcd00->100000
Released 614 pages of unused memory
Set 277889 page(s) to 1-1 mapping
Populating 40200-40466 pfn range: 614 pages added

=> here we set from 40466 up to bc559 P2M tree to be
INVALID_P2M_ENTRY. We should have done it up to bc558.

The end result is that if anybody is trying to construct
a PTE for PFN bc558 they end up with ~PAGE_PRESENT.

Reported-by-and-Tested-by: Andre Przywara <andre.przywara@amd.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

ARM: dts: imx51-babbage: fix esdhc cd/wp properties

commit a46d2619d7180bda12bad2bf15bbd0731dfc2dcf upstream.

The binding doc and dts use properties "fsl,{cd,wp}-internal" while
esdhc driver uses "fsl,{cd,wp}-controller".  Fix binding doc and dts
to get them match driver code.

Reported-by: Chris Ball <cjb@laptop.org>
Signed-off-by: Shawn Guo <shawn.guo@linaro.org>
Acked-by: Chris Ball <cjb@laptop.org>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

ARM: imx6: spin the cpu until hardware takes it down

commit c944b0b9354ea06ffb0c8a7178949f1185f9f499 upstream.

Though commit 602bf40 (ARM: imx6: exit coherency when shutting down
a cpu) improves the stability of imx6q cpu hotplug a lot, there are
still hangs seen with a more stressful hotplug testing.

It's expected that once imx_enable_cpu(cpu, false) is called, the cpu
will be taken down by hardware immediately, and the code after that
will not get any chance to execute.  However, this is not always the
case from the testing.  The cpu could possibly be alive for a few
cycles before hardware actually takes it down.  So rather than letting
cpu execute some code that could cause a hang in these cycles, let's
make the cpu spin there and wait for hardware to take it down.

Signed-off-by: Shawn Guo <shawn.guo@linaro.org>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

OMAPFB: fix framebuffer console colors

commit c1c52848cef52e157468b8879fc3cae23b6f3a99 upstream.

omapfb does not currently set pseudo palette correctly for color depths
above 16bpp, making red text invisible, command like
  echo -e '\e[0;31mRED' > /dev/tty1
will display nothing on framebuffer console in 24bpp mode.
This is because temporary variable is declared incorrectly, fix it.

Signed-off-by: Grazvydas Ignotas <notasas@gmail.com>
Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ti.com>
Signed-off-by: Florian Tobias Schandinat <FlorianSchandinat@gmx.de>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

drm/i915: fix wrong order of parameters in port checking functions

commit b70ad586162609141f0aa9eb34790f31a8954f89 upstream.

Wrong order of parameters passed-in when calling hdmi/adpa
/lvds_pipe_enabled(), 2nd and 3rd parameters are reversed.

This bug was indroduced by

commit 1519b9956eb4b4180fa3f47c73341463cdcfaa37
Author: Keith Packard <keithp@keithp.com>
Date:   Sat Aug 6 10:35:34 2011 -0700

    drm/i915: Fix PCH port pipe select in CPT disable paths

The reachable tag for this commit is v3.1-rc1-3-g1519b99

Signed-off-by: Anhua Xu <anhua.xu@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=44876
Tested-by: Daniel Schroeder <sec@dschroeder.info>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

drm/i915: Fix assert_pch_hdmi_disabled to mention HDMI (not DP)

commit 23c99e775d14f01ba45a5affd2fb51af4328359c upstream.

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

hwmon: (asus_atk0110) Add quirk for Asus M5A78L

commit 43ca6cb28c871f2fbad10117b0648e5ae3b0f638 upstream.

The old interface is bugged and reads the wrong sensor when retrieving
the reading for the chassis fan (it reads the CPU sensor); the new
interface works fine.

Reported-by: Göran Uddeborg <goeran@uddeborg.se>
Tested-by: Göran Uddeborg <goeran@uddeborg.se>
Signed-off-by: Luca Tettamanti <kronos.it@gmail.com>
Signed-off-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

Fix 'Device not ready' issue on mpt2sas

commit 14216561e164671ce147458653b1fea06a4ada1e upstream.

This is a particularly nasty SCSI ATA Translation Layer (SATL) problem.

SAT-2 says (section 8.12.2)

        if the device is in the stopped state as the result of
        processing a START STOP UNIT command (see 9.11), then the SATL
        shall terminate the TEST UNIT READY command with CHECK CONDITION
        status with the sense key set to NOT READY and the additional
        sense code of LOGICAL UNIT NOT READY, INITIALIZING COMMAND
        REQUIRED;

mpt2sas internal SATL seems to implement this.  The result is very confusing
standby behaviour (using hdparm -y).  If you suspend a drive and then send
another command, usually it wakes up.  However, if the next command is a TEST
UNIT READY, the SATL sees that the drive is suspended and proceeds to follow
the SATL rules for this, returning NOT READY to all subsequent commands.  This
means that the ordering of TEST UNIT READY is crucial: if you send TUR and
then a command, you get a NOT READY to both back.  If you send a command and
then a TUR, you get GOOD status because the preceeding command woke the drive.

This bit us badly because

commit 85ef06d1d252f6a2e73b678591ab71caad4667bb
Author: Tejun Heo <tj@kernel.org>
Date:   Fri Jul 1 16:17:47 2011 +0200

    block: flush MEDIA_CHANGE from drivers on close(2)

Changed our ordering on TEST UNIT READY commands meaning that SATA drives
connected to an mpt2sas now suspend and refuse to wake (because the mpt2sas
SATL sees the suspend *before* the drives get awoken by the next ATA command)
resulting in lots of failed commands.

The standard is completely nuts forcing this inconsistent behaviour, but we
have to work around it.

The fix for this is twofold:

   1. Set the allow_restart flag so we wake the drive when we see it has been
      suspended

   2. Return all TEST UNIT READY status directly to the mid layer without any
      further error handling which prevents us causing error handling which
      may offline the device just because of a media check TUR.

Reported-by: Matthias Prager <linux@matthiasprager.de>
Signed-off-by: James Bottomley <JBottomley@Parallels.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

megaraid_sas: Move poll_aen_lock initializer

commit bd8d6dd43a77bfd2b8fef5b094b9d6095e169dee upstream.

The following patch moves the poll_aen_lock initializer from
megasas_probe_one() to megasas_init().  This prevents a crash when a user
loads the driver and tries to issue a poll() system call on the ioctl
interface with no adapters present.

Signed-off-by: Kashyap Desai <Kashyap.Desai@lsi.com>
Signed-off-by: Adam Radford <aradford@gmail.com>
Signed-off-by: James Bottomley <JBottomley@Parallels.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

mpt2sas: Fix for Driver oops, when loading driver with max_queue_depth command line option to a very small value

commit 338b131a3269881c7431234855c93c219b0979b6 upstream.

If the specified max_queue_depth setting is less than the
expected number of internal commands, then driver will calculate
the queue depth size to a negitive number. This negitive number
is actually a very large number because variable is unsigned
16bit integer. So, the driver will ask for a very large amount of
memory for message frames and resulting into oops as memory
allocation routines will not able to handle such a large request.

So, in order to limit this kind of oops, The driver need to set
the max_queue_depth to a scsi mid layer's can_queue value. Then
the overall message frames required for IO is minimum of either
(max_queue_depth plus internal commands) or the IOC global
credits.

Signed-off-by: Sreekanth Reddy <sreekanth.reddy@lsi.com>
Signed-off-by: James Bottomley <JBottomley@Parallels.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

Input: i8042 - add Gigabyte T1005 series netbooks to noloop table

commit 7b125b94ca16b7e618c6241cb02c4c8060cea5e3 upstream.

They all define their chassis type as "Other" and therefore are not
categorized as "laptops" by the driver, which tries to perform AUX IRQ
delivery test which fails and causes touchpad not working.

Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=42620
Signed-off-by: Dmitry Torokhov <dmitry.torokhov@gmail.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

iwlwifi: protect SRAM debugfs

commit 4fc79db178f0a0ede479b4713e00df2d106028b3 upstream.

If the device is not started, we can't read its
SRAM and attempting to do so will cause issues.
Protect the debugfs read.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
[bwh: Backported to 3.2:
 - Adjust context
 - Pass iwl_shared not iwl_priv pointer to iwl_is_ready_rf()]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

iwlwifi: fix flow handler debug code

commit 94543a8d4fb302817014981489f15cb3b92ec3c2 upstream.

iwl_dbgfs_fh_reg_read() can cause crashes and/or
BUG_ON in slub because the ifdefs are wrong, the
code in iwl_dump_fh() should use DEBUGFS, not
DEBUG to protect the buffer writing code.

Also, while at it, clean up the arguments to the
function, some code and make it generally safer.

Reported-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
[bwh: Backported to 3.2: adjust filenames and context]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

ARM: Orion: Set eth packet size csum offload limit

commit 58569aee5a1a5dcc25c34a0a2ed9a377874e6b05 upstream.

The mv643xx ethernet controller limits the packet size for the TX
checksum offloading. This patch sets this limits for Kirkwood and
Dove which have smaller limits that the default.

As a side note, this patch is an updated version of a patch sent some years
ago: http://lists.infradead.org/pipermail/linux-arm-kernel/2010-June/017320.html
which seems to have been lost.

Signed-off-by: Arnaud Patard <arnaud.patard@rtp-net.org>
Signed-off-by: Jason Cooper <jason@lakedaemon.net>
[bwh: Backported to 3.2: adjust for the extra two parameters of
 orion_ge0{0,1}_init()]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

ARM: OMAP2+: Fix dmtimer set source clock failure

commit 54f32a35f4d3a653a18a2c8c239f19ae060bd803 upstream.

Calling the dmtimer function omap_dm_timer_set_source() fails if following a
call to pm_runtime_put() to disable the timer. For example the following
sequence would fail to set the parent clock ...

omap_dm_timer_stop(gptimer);
omap_dm_timer_set_source(gptimer, OMAP_TIMER_SRC_32_KHZ);

The following error message would be seen ...

omap_dm_timer_set_source: failed to set timer_32k_ck as parent

The problem is that, by design, pm_runtime_put() simply decrements the usage
count and returns before the timer has actually been disabled. Therefore,
setting the parent clock failed because the timer was still active when the
trying to set the parent clock. Setting a parent clock will fail if the clock
you are setting the parent of has a non-zero usage count. To ensure that this
does not fail use pm_runtime_put_sync() when disabling the timer.

Note that this will not be seen on OMAP1 devices, because these devices do
not use the clock framework for dmtimers.

Signed-off-by: Jon Hunter <jon-hunter@ti.com>
Acked-by: Kevin Hilman <khilman@ti.com>
Signed-off-by: Tony Lindgren <tony@atomide.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

ARM: S3C24XX: Fix s3c2410_dma_enqueue parameters

commit b01858c7806e7e6f6121da2e51c9222fc4d21dc6 upstream.

Commit d670ac019f60 (ARM: SAMSUNG: DMA Cleanup as per sparse) changed the
prototype of the s3c2410_dma_* functions to use the enum dma_ch instead
of an generic unsigned int.

In the s3c24xx dma.c s3c2410_dma_enqueue seems to have been forgotten,
the other functions there were changed correctly.

Signed-off-by: Heiko Stuebner <heiko@sntech.de>
Signed-off-by: Kukjin Kim <kgene.kim@samsung.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

Redefine ATOMIC_INIT and ATOMIC64_INIT to drop the casts

commit bba3d8c3b3c0f2123be5bc687d1cddc13437c923 upstream.

The following build error occured during a parisc build with
swap-over-NFS patches applied.

net/core/sock.c:274:36: error: initializer element is not constant
net/core/sock.c:274:36: error: (near initialization for 'memalloc_socks')
net/core/sock.c:274:36: error: initializer element is not constant

Dave Anglin says:
> Here is the line in sock.i:
>
> struct static_key memalloc_socks = ((struct static_key) { .enabled =
> ((atomic_t) { (0) }) });

The above line contains two compound literals.  It also uses a designated
initializer to initialize the field enabled.  A compound literal is not a
constant expression.

The location of the above statement isn't fully clear, but if a compound
literal occurs outside the body of a function, the initializer list must
consist of constant expressions.

Reported-by: Fengguang Wu <fengguang.wu@intel.com>
Signed-off-by: Mel Gorman <mgorman@suse.de>
Signed-off-by: James Bottomley <JBottomley@Parallels.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

Linux 3.2.29

mm: avoid swapping out with swappiness==0

commit fe35004fbf9eaf67482b074a2e032abb9c89b1dd upstream.

Sometimes we'd like to avoid swapping out anonymous memory.  In
particular, avoid swapping out pages of important process or process
groups while there is a reasonable amount of pagecache on RAM so that we
can satisfy our customers' requirements.

OTOH, we can control how aggressive the kernel will swap memory pages with
/proc/sys/vm/swappiness for global and
/sys/fs/cgroup/memory/memory.swappiness for each memcg.

But with current reclaim implementation, the kernel may swap out even if
we set swappiness=0 and there is pagecache in RAM.

This patch changes the behavior with swappiness==0.  If we set
swappiness==0, the kernel does not swap out completely (for global reclaim
until the amount of free pages and filebacked pages in a zone has been
reduced to something very very small (nr_free + nr_filebacked < high
watermark)).

Signed-off-by: Satoru Moriya <satoru.moriya@hds.com>
Acked-by: Minchan Kim <minchan@kernel.org>
Reviewed-by: Rik van Riel <riel@redhat.com>
Acked-by: Jerome Marchand <jmarchan@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
[bwh: Backported to 3.2:
 - Adjust context
 - vmscan_swappiness() does not have a zone parameter]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

Squashfs: fix mount time sanity check for corrupted superblock

commit cc37f75a9ffbbfcb1c3297534f293c8284e3c5a6 upstream.

A Squashfs filesystem containing nothing but an empty directory,
although unusual and ultimately pointless, is still valid.

The directory_table >= next_table sanity check rejects these
filesystems as invalid because the directory_table is empty and
equal to next_table.

Signed-off-by: Phillip Lougher <phillip@squashfs.org.uk>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

asus-nb-wmi: add some video toggle keys

commit 3766054fff4af1b58a1440a284907887f4d2e8be upstream.

There are some new video switch keys that used by newer machines.
0xA0 - SDSP HDMI only
0xA1 - SDSP LCD + HDMI
0xA2 - SDSP CRT + HDMI
0xA3 - SDSP TV + HDMI
But in Linux, there is no suitable userspace application to handle this,
so, mapping them all to KEY_SWITCHVIDEOMODE.

Signed-off-by: AceLan Kao <acelan.kao@canonical.com>
Signed-off-by: Matthew Garrett <mjg@redhat.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

NFS: Fix Oopses in nfs_lookup_revalidate and nfs4_lookup_revalidate

Fix the following Oops in 3.5.1:

 BUG: unable to handle kernel NULL pointer dereference at 0000000000000038
 IP: [<ffffffffa03789cd>] nfs_lookup_revalidate+0x2d/0x480 [nfs]
 PGD 337c63067 PUD 0
 Oops: 0000 [#1] SMP
 CPU 5
 Modules linked in: nfs fscache nfsd lockd nfs_acl auth_rpcgss sunrpc af_packet binfmt_misc cpufreq_conservative cpufreq_userspace cpufreq_powersave dm_mod acpi_cpufreq mperf coretemp gpio_ich kvm_intel joydev kvm ioatdma hid_generic igb lpc_ich i7core_edac edac_core ptp serio_raw dca pcspkr i2c_i801 mfd_core sg pps_core usbhid crc32c_intel microcode button autofs4 uhci_hcd ttm drm_kms_helper drm i2c_algo_bit sysimgblt sysfillrect syscopyarea ehci_hcd usbcore usb_common scsi_dh_rdac scsi_dh_emc scsi_dh_hp_sw scsi_dh_alua scsi_dh edd fan ata_piix thermal processor thermal_sys

 Pid: 30431, comm: java Not tainted 3.5.1-2-default #1 Supermicro X8DTT/X8DTT
 RIP: 0010:[<ffffffffa03789cd>]  [<ffffffffa03789cd>] nfs_lookup_revalidate+0x2d/0x480 [nfs]
 RSP: 0018:ffff8801b418bd38  EFLAGS: 00010292
 RAX: 00000000fffffff6 RBX: ffff88032016d800 RCX: 0000000000000020
 RDX: ffffffff00000000 RSI: 0000000000000000 RDI: ffff8801824a7b00
 RBP: ffff8801b418bdf8 R08: 7fffff0034323030 R09: fffffffff04c03ed
 R10: ffff8801824a7b00 R11: 0000000000000002 R12: ffff8801824a7b00
 R13: ffff8801824a7b00 R14: 0000000000000000 R15: ffff8803201725d0
 FS:  00002b53a46cb700(0000) GS:ffff88033fc20000(0000) knlGS:0000000000000000
 CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
 CR2: 0000000000000038 CR3: 000000020a426000 CR4: 00000000000007e0
 DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
 DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400
 Process java (pid: 30431, threadinfo ffff8801b418a000, task ffff8801b5d20600)
 Stack:
  ffff8801b418be44 ffff88032016d800 ffff8801b418bdf8 0000000000000000
  ffff8801824a7b00 ffff8801b418bdd7 ffff8803201725d0 ffffffff8116a9c0
  ffff8801b5c38dc0 0000000000000007 ffff88032016d800 0000000000000000
 Call Trace:
  [<ffffffff8116a9c0>] lookup_dcache+0x80/0xe0
  [<ffffffff8116aa43>] __lookup_hash+0x23/0x90
  [<ffffffff8116b4a5>] lookup_one_len+0xc5/0x100
  [<ffffffffa03869a3>] nfs_sillyrename+0xe3/0x210 [nfs]
  [<ffffffff8116cadf>] vfs_unlink.part.25+0x7f/0xe0
  [<ffffffff8116f22c>] do_unlinkat+0x1ac/0x1d0
  [<ffffffff815717b9>] system_call_fastpath+0x16/0x1b
  [<00002b5348b5f527>] 0x2b5348b5f526
 Code: ec 38 b8 f6 ff ff ff 4c 89 64 24 18 4c 89 74 24 28 49 89 fc 48 89 5c 24 08 48 89 6c 24 10 49 89 f6 4c 89 6c 24 20 4c 89 7c 24 30 <f6> 46 38 40 0f 85 d1 00 00 00 e8 c4 c4 df e0 48 8b 58 30 49 89
 RIP  [<ffffffffa03789cd>] nfs_lookup_revalidate+0x2d/0x480 [nfs]
  RSP <ffff8801b418bd38>
 CR2: 0000000000000038
 ---[ end trace 845113ed191985dd ]---

This Oops affects 3.5 kernels and older, and is due to lookup_one_len()
calling down to the dentry revalidation code with a NULL pointer
to struct nameidata.

It is fixed upstream by commit 0b728e1911c (stop passing nameidata *
to ->d_revalidate())

Reported-by: Richard Ems <richard.ems@cape-horn-eng.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

Staging: speakup: fix an improperly-declared variable.

commit 4ea418b8b2fa8a70d0fcc8231b65e67b3a72984b upstream.

A local static variable was declared as a pointer to a string
constant.  We're assigning to the underlying memory, so it
needs to be an array instead.

Signed-off-by: Christopher Brannon <chris@the-brannons.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

pmac_zilog,kdb: Fix console poll hook to return instead of loop

commit 38f8eefccf3a23c4058a570fa2938a4f553cf8e0 upstream.

kdb <-> kgdb transitioning does not work properly with this UART
driver because the get character routine loops indefinitely as opposed
to returning NO_POLL_CHAR per the expectation of the KDB I/O driver
API.

The symptom is a kernel hang when trying to switch debug modes.

Cc: Alan Cox <alan@linux.intel.com>
Signed-off-by: Jason Wessel <jason.wessel@windriver.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

HID: add support for Cypress barcode scanner 04B4:ED81

commit 76c9d8fe2c7fc34ffc387d8022c5828d6ff9df48 upstream.

Add yet another device to the list of Cypress barcode scanners
needing the CP_RDESC_SWAPPED_MIN_MAX quirk.

Signed-off-by: Lionel Vaux (iouri) <lionel.vaux@free.fr>
Signed-off-by: Jiri Kosina <jkosina@suse.cz>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

virtio-blk: Reset device after blk_cleanup_queue()

commit 483001c765af6892b3fc3726576cb42f17d1d6b5 upstream.

blk_cleanup_queue() will call blk_drian_queue() to drain all the
requests before queue DEAD marking. If we reset the device before
blk_cleanup_queue() the drain would fail.

1) if the queue is stopped in do_virtblk_request() because device is
full, the q->request_fn() will not be called.

blk_drain_queue() {
   while(true) {
      ...
      if (!list_empty(&q->queue_head))
        __blk_run_queue(q) {
    if (queue is not stoped)
q->request_fn()
}
      ...
   }
}

Do no reset the device before blk_cleanup_queue() gives the chance to
start the queue in interrupt handler blk_done().

2) In commit b79d866c8b7014a51f611a64c40546109beaf24a, We abort requests
dispatched to driver before blk_cleanup_queue(). There is a race if
requests are dispatched to driver after the abort and before the queue
DEAD mark. To fix this, instead of aborting the requests explicitly, we
can just reset the device after after blk_cleanup_queue so that the
device can complete all the requests before queue DEAD marking in the
drain process.

Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: virtualization@lists.linux-foundation.org
Cc: kvm@vger.kernel.org
Signed-off-by: Asias He <asias@redhat.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

virtio-blk: Call del_gendisk() before disable guest kick

commit 02e2b124943648fba0a2ccee5c3656a5653e0151 upstream.

del_gendisk() might not return due to failing to remove the
/sys/block/vda/serial sysfs entry when another thread (udev) is
trying to read it.

virtblk_remove()
  vdev->config->reset() : guest will not kick us through interrupt
    del_gendisk()
      device_del()
        kobject_del(): got stuck, sysfs entry ref count non zero

sysfs_open_file(): user space process read /sys/block/vda/serial
   sysfs_get_active() : got sysfs entry ref count
      dev_attr_show()
        virtblk_serial_show()
           blk_execute_rq() : got stuck, interrupt is disabled
                              request cannot be finished

This patch fixes it by calling del_gendisk() before we disable guest's
interrupt so that the request sent in virtblk_serial_show() will be
finished and del_gendisk() will success.

This fixes another race in hot-unplug process.

It is save to call del_gendisk(vblk->disk) before
flush_work(&vblk->config_work) which might access vblk->disk, because
vblk->disk is not freed until put_disk(vblk->disk).

Cc: virtualization@lists.linux-foundation.org
Cc: kvm@vger.kernel.org
Signed-off-by: Asias He <asias@redhat.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

virtio-blk: Fix hot-unplug race in remove method

commit b79d866c8b7014a51f611a64c40546109beaf24a upstream.

If we reset the virtio-blk device before the requests already dispatched
to the virtio-blk driver from the block layer are finised, we will stuck
in blk_cleanup_queue() and the remove will fail.

blk_cleanup_queue() calls blk_drain_queue() to drain all requests queued
before DEAD marking. However it will never success if the device is
already stopped. We'll have q->in_flight[] > 0, so the drain will not
finish.

How to reproduce the race:
1. hot-plug a virtio-blk device
2. keep reading/writing the device in guest
3. hot-unplug while the device is busy serving I/O

Test:
~1000 rounds of hot-plug/hot-unplug test passed with this patch.

Changes in v3:
- Drop blk_abort_queue and blk_abort_request
- Use __blk_end_request_all to complete request dispatched to driver

Changes in v2:
- Drop req_in_flight
- Use virtqueue_detach_unused_buf to get request dispatched to driver

Signed-off-by: Asias He <asias@redhat.com>
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

virtio_blk: Drop unused request tracking list

commit f65ca1dc6a8c81c6bd72297d4399ec5f4c1f3a01 upstream.

Benchmark shows small performance improvement on fusion io device.

Before:
  seq-read : io=1,024MB, bw=19,982KB/s, iops=39,964, runt= 52475msec
  seq-write: io=1,024MB, bw=20,321KB/s, iops=40,641, runt= 51601msec
  rnd-read : io=1,024MB, bw=15,404KB/s, iops=30,808, runt= 68070msec
  rnd-write: io=1,024MB, bw=14,776KB/s, iops=29,552, runt= 70963msec

After:
  seq-read : io=1,024MB, bw=20,343KB/s, iops=40,685, runt= 51546msec
  seq-write: io=1,024MB, bw=20,803KB/s, iops=41,606, runt= 50404msec
  rnd-read : io=1,024MB, bw=16,221KB/s, iops=32,442, runt= 64642msec
  rnd-write: io=1,024MB, bw=15,199KB/s, iops=30,397, runt= 68991msec

Signed-off-by: Asias He <asias@redhat.com>
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

virtio_blk: fix config handler race

commit 4678d6f970c2f7c0cbfefc0cc666432d153b321b upstream.

Fix a theoretical race related to config work
handler: a config interrupt might happen
after we flush config work but before we
reset the device. It will then cause the
config work to run during or after reset.

Two problems with this:
- if this runs after device is gone we will get use after free
- access of config while reset is in progress is racy
(as layout is changing).

As a solution
1. flush after reset when we know there will be no more interrupts
2. add a flag to disable config access before reset

Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

dmi: Feed DMI table to /dev/random driver

commit d114a33387472555188f142ed8e98acdb8181c6d upstream.

Send the entire DMI (SMBIOS) table to the /dev/random driver to
help seed its pools.

Signed-off-by: Tony Luck <tony.luck@intel.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

random: Add comment to random_initialize()

commit cbc96b7594b5691d61eba2db8b2ea723645be9ca upstream.

Many platforms have per-machine instance data (serial numbers,
asset tags, etc.) squirreled away in areas that are accessed
during early system bringup. Mixing this data into the random
pools has a very high value in providing better random data,
so we should allow (and even encourage) architecture code to
call add_device_randomness() from the setup_arch() paths.

However, this limits our options for internal structure of
the random driver since random_initialize() is not called
until long after setup_arch().

Add a big fat comment to rand_initialize() spelling out
this requirement.

Suggested-by: Theodore Ts'o <tytso@mit.edu>
Signed-off-by: Tony Luck <tony.luck@intel.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

MAINTAINERS: Theodore Ts'o is taking over the random driver

commit 330e0a01d54c2b8606c56816f99af6ebc58ec92c upstream.

Matt Mackall stepped down as the /dev/random driver maintainer last
year, so Theodore Ts'o is taking back the /dev/random driver.

Cc: Matt Mackall <mpm@selenic.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

drivers/char/random.c: fix boot id uniqueness race

commit 44e4360fa3384850d65dd36fb4e6e5f2f112709b upstream.

/proc/sys/kernel/random/boot_id can be read concurrently by userspace
processes.  If two (or more) user-space processes concurrently read
boot_id when sysctl_bootid is not yet assigned, a race can occur making
boot_id differ between the reads.  Because the whole point of the boot id
is to be unique across a kernel execution, fix this by protecting this
operation with a spinlock.

Given that this operation is not frequently used, hitting the spinlock
on each call should not be an issue.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Cc: "Theodore Ts'o" <tytso@mit.edu>
Cc: Matt Mackall <mpm@selenic.com>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Greg Kroah-Hartman <greg@kroah.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>

Bluetooth: Fix using uninitialized option in RFCMode

commit 8f321f853ea33330c7141977cd34804476e2e07e upstream.

If remote device sends bogus RFC option with invalid length,
undefined options values are used. Fix this by using defaults when
remote misbehaves.

This also fixes the following warning reported by gcc 4.7.0:

net/bluetooth/l2cap_core.c: In function 'l2cap_config_rsp':
net/bluetooth/l2cap_core.c:3302:13: warning: 'rfc.max_pdu_size' may be used uninitialized in this function [-Wmaybe-uninitialized]
net/bluetooth/l2cap_core.c:3266:24: note: 'rfc.max_pdu_size' was declared here
net/bluetooth/l2cap_core.c:3298:25: warning: 'rfc.monitor_timeout' may be used uninitialized in this function [-Wmaybe-uninitialized]
net/bluetooth/l2cap_core.c:3266:24: note: 'rfc.monitor_timeout' was declared here
net/bluetooth/l2cap_core.c:3297:25: warning: 'rfc.retrans_timeout' may be used uninitialized in this function [-Wmaybe-uninitialized]
net/bluetooth/l2cap_core.c:3266:24: note: 'rfc.retrans_timeout' was declared here
net/bluetooth/l2cap_core.c:3295:2: warning: 'rfc.mode' may be used uninitialized in this function [-Wmaybe-uninitialized]
net/bluetooth/l2cap_core.c:3266:24: note: 'rfc.mode' was declared here

Signed-off-by: Szymon Janc <szymon.janc@tieto.com>
Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

block: replace __getblk_slow misfix by grow_dev_page fix

commit 676ce6d5ca3098339c028d44fe0427d1566a4d2d upstream.

Commit 91f68c89d8f3 ("block: fix infinite loop in __getblk_slow")
is not good: a successful call to grow_buffers() cannot guarantee
that the page won't be reclaimed before the immediate next call to
__find_get_block(), which is why there was always a loop there.

Yesterday I got "EXT4-fs error (device loop0): __ext4_get_inode_loc:3595:
inode #19278: block 664: comm cc1: unable to read itable block" on console,
which pointed to this commit.

I've been trying to bisect for weeks, why kbuild-on-ext4-on-loop-on-tmpfs
sometimes fails from a missing header file, under memory pressure on
ppc G5.  I've never seen this on x86, and I've never seen it on 3.5-rc7
itself, despite that commit being in there: bisection pointed to an
irrelevant pinctrl merge, but hard to tell when failure takes between
18 minutes and 38 hours (but so far it's happened quicker on 3.6-rc2).

(I've since found such __ext4_get_inode_loc errors in /var/log/messages
from previous weeks: why the message never appeared on console until
yesterday morning is a mystery for another day.)

Revert 91f68c89d8f3, restoring __getblk_slow() to how it was (plus
a checkpatch nitfix).  Simplify the interface between grow_buffers()
and grow_dev_page(), and avoid the infinite loop beyond end of device
by instead checking init_page_buffers()'s end_block there (I presume
that's more efficient than a repeated call to blkdev_max_block()),
returning -ENXIO to __getblk_slow() in that case.

And remove akpm's ten-year-old "__getblk() cannot fail ... weird"
comment, but that is worrying: are all users of __getblk() really
now prepared for a NULL bh beyond end of device, or will some oops??

Signed-off-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

fs/buffer.c: remove BUG() in possible but rare condition

commit 61065a30af8df4b8989c2ac7a1f4b4034e4df2d5 upstream.

While stressing the kernel with with failing allocations today, I hit the
following chain of events:

alloc_page_buffers():

bh = alloc_buffer_head(GFP_NOFS);
if (!bh)
goto no_grow; <= path taken

grow_dev_page():
        bh = alloc_page_buffers(page, size, 0);
        if (!bh)
                goto failed;  <= taken, consequence of the above

and then the failed path BUG()s the kernel.

The failure is inserted a litte bit artificially, but even then, I see no
reason why it should be deemed impossible in a real box.

Even though this is not a condition that we expect to see around every
time, failed allocations are expected to be handled, and BUG() sounds just
too much.  As a matter of fact, grow_dev_page() can return NULL just fine
in other circumstances, so I propose we just remove it, then.

Signed-off-by: Glauber Costa <glommer@parallels.com>
Cc: Michal Hocko <mhocko@suse.cz>
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>

rapidio/tsi721: fix unused variable compiler warning

commit 9a9a9a7adafe62a34de8b4fb48936c1c5f9bafa5 upstream.

Fix unused variable compiler warning when built with CONFIG_RAPIDIO_DEBUG
option off.

This patch is applicable to kernel versions starting from v3.2

Signed-off-by: Alexandre Bounine <alexandre.bounine@idt.com>
Cc: Matt Porter <mporter@kernel.crashing.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

rapidio/tsi721: fix inbound doorbell interrupt handling

commit 3670e7e12e582c6d67761275d148171feb7a9004 upstream.

Make sure that there is no doorbell messages left behind due to disabled
interrupts during inbound doorbell processing.

The most common case for this bug is loss of rionet JOIN messages in
systems with three or more rionet participants and MSI or MSI-X enabled.
As result, requests for packet transfers may finish with "destination
unreachable" error message.

This patch is applicable to kernel versions starting from v3.2.

Signed-off-by: Alexandre Bounine <alexandre.bounine@idt.com>
Cc: Matt Porter <mporter@kernel.crashing.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

drivers/rtc/rtc-rs5c348.c: fix hour decoding in 12-hour mode

commit 7dbfb315b2aaef0a115765946bf3026d074c33a7 upstream.

Correct the offset by subtracting 20 from tm_hour before taking the
modulo 12.

[ "Why 20?" I hear you ask. Or at least I did.

  Here's the reason why: RS5C348_BIT_PM is 32, and is - stupidly -
  included in the RS5C348_HOURS_MASK define.  So it's really subtracting
  out that bit to get "hour+12".  But then because it does things modulo
  12, it needs to add the 12 in again afterwards anyway.

  This code is confused.  It would be much clearer if RS5C348_HOURS_MASK
  just didn't include the RS5C348_BIT_PM bit at all, then it wouldn't
  need to do the silly subtract either.

  Whatever. It's all just math, the end result is the same.   - Linus ]

Reported-by: James Nute <newten82@gmail.com>
Tested-by: James Nute <newten82@gmail.com>
Signed-off-by: Atsushi Nemoto <anemo@mba.ocn.ne.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>

drivers/misc/sgi-xp/xpc_uv.c: SGI XPC fails to load when cpu 0 is out of IRQ resources

commit 7838f994b4fceff24c343f4e26a6cf4393869579 upstream.

On many of our larger systems, CPU 0 has had all of its IRQ resources
consumed before XPC loads.  Worst cases on machines with multiple 10
GigE cards and multiple IB cards have depleted the entire first socket
of IRQs.

This patch makes selecting the node upon which IRQs are allocated (as
well as all the other GRU Message Queue structures) specifiable as a
module load param and has a default behavior of searching all nodes/cpus
for an available resources.

[akpm@linux-foundation.org: fix build: include cpu.h and module.h]
Signed-off-by: Robin Holt <holt@sgi.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: hugetlbfs: correctly populate shared pmd

commit eb48c071464757414538c68a6033c8f8c15196f8 upstream.

Each page mapped in a process's address space must be correctly
accounted for in _mapcount.  Normally the rules for this are
straightforward but hugetlbfs page table sharing is different.  The page
table pages at the PMD level are reference counted while the mapcount
remains the same.

If this accounting is wrong, it causes bugs like this one reported by
Larry Woodman:

  kernel BUG at mm/filemap.c:135!
  invalid opcode: 0000 [#1] SMP
  CPU 22
  Modules linked in: bridge stp llc sunrpc binfmt_misc dcdbas microcode pcspkr acpi_pad acpi]
  Pid: 18001, comm: mpitest Tainted: G        W    3.3.0+ #4 Dell Inc. PowerEdge R620/07NDJ2
  RIP: 0010:[<ffffffff8112cfed>]  [<ffffffff8112cfed>] __delete_from_page_cache+0x15d/0x170
  Process mpitest (pid: 18001, threadinfo ffff880428972000, task ffff880428b5cc20)
  Call Trace:
    delete_from_page_cache+0x40/0x80
    truncate_hugepages+0x115/0x1f0
    hugetlbfs_evict_inode+0x18/0x30
    evict+0x9f/0x1b0
    iput_final+0xe3/0x1e0
    iput+0x3e/0x50
    d_kill+0xf8/0x110
    dput+0xe2/0x1b0
    __fput+0x162/0x240

During fork(), copy_hugetlb_page_range() detects if huge_pte_alloc()
shared page tables with the check dst_pte == src_pte.  The logic is if
the PMD page is the same, they must be shared.  This assumes that the
sharing is between the parent and child.  However, if the sharing is
with a different process entirely then this check fails as in this
diagram:

  parent
    |
    ------------>pmd
                 src_pte----------> data page
                                        ^
  other--------->pmd--------------------|
                  ^
  child-----------|
                 dst_pte

For this situation to occur, it must be possible for Parent and Other to
have faulted and failed to share page tables with each other.  This is
possible due to the following style of race.

  PROC A                                          PROC B
  copy_hugetlb_page_range                         copy_hugetlb_page_range
    src_pte == huge_pte_offset                      src_pte == huge_pte_offset
    !src_pte so no sharing                          !src_pte so no sharing

  (time passes)

  hugetlb_fault                                   hugetlb_fault
    huge_pte_alloc                                  huge_pte_alloc
      huge_pmd_share                                 huge_pmd_share
        LOCK(i_mmap_mutex)
        find nothing, no sharing
        UNLOCK(i_mmap_mutex)
                                                      LOCK(i_mmap_mutex)
                                                      find nothing, no sharing
                                                      UNLOCK(i_mmap_mutex)
      pmd_alloc                                       pmd_alloc
      LOCK(instantiation_mutex)
      fault
      UNLOCK(instantiation_mutex)
                                                  LOCK(instantiation_mutex)
                                                  fault
                                                  UNLOCK(instantiation_mutex)

These two processes are not poing to the same data page but are not
sharing page tables because the opportunity was missed.  When either
process later forks, the src_pte == dst pte is potentially insufficient.
As the check falls through, the wrong PTE information is copied in
(harmless but wrong) and the mapcount is bumped for a page mapped by a
shared page table leading to the BUG_ON.

This patch addresses the issue by moving pmd_alloc into huge_pmd_share
which guarantees that the shared pud is populated in the same critical
section as pmd.  This also means that huge_pte_offset test in
huge_pmd_share is serialized correctly now which in turn means that the
success of the sharing will be higher as the racing tasks see the pud
and pmd populated together.

Race identified and changelog written mostly by Mel Gorman.

{akpm@linux-foundation.org: attempt to make the huge_pmd_share() comment comprehensible, clean up coding style]
Reported-by: Larry Woodman <lwoodman@redhat.com>
Tested-by: Larry Woodman <lwoodman@redhat.com>
Reviewed-by: Mel Gorman <mgorman@suse.de>
Signed-off-by: Michal Hocko <mhocko@suse.cz>
Reviewed-by: Rik van Riel <riel@redhat.com>
Cc: David Gibson <david@gibson.dropbear.id.au>
Cc: Ken Chen <kenchen@google.com>
Cc: Cong Wang <xiyou.wangcong@gmail.com>
Cc: Hillf Danton <dhillf@gmail.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>

cciss: fix incorrect scsi status reporting

commit b0cf0b118c90477d1a6811f2cd2307f6a5578362 upstream.

Delete code which sets SCSI status incorrectly as it's already been set
correctly above this incorrect code.  The bug was introduced in 2009 by
commit b0e15f6db111 ("cciss: fix typo that causes scsi status to be
lost.")

Signed-off-by: Stephen M. Cameron <scameron@beardog.cce.hp.com>
Reported-by: Roel van Meer <roel.vanmeer@bokxing.nl>
Tested-by: Roel van Meer <roel.vanmeer@bokxing.nl>
Cc: Jens Axboe <axboe@kernel.dk>
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>

fbcon: fix race condition between console lock and cursor timer (v1.1)

commit d8636a2717bb3da2a7ce2154bf08de90bb8c87b0 upstream.

So we've had a fair few reports of fbcon handover breakage between
efi/vesafb and i915 surface recently, so I dedicated a couple of
days to finding the problem.

Essentially the last thing we saw was the conflicting framebuffer
message and that was all.

So after much tracing with direct netconsole writes (printks
under console_lock not so useful), I think I found the race.

Thread A (driver load)    Thread B (timer thread)
  unbind_con_driver ->              |
  bind_con_driver ->                |
  vc->vc_sw->con_deinit ->          |
  fbcon_deinit ->                   |
  console_lock()                    |
      |                             |
      |                       fbcon_flashcursor timer fires
      |                       console_lock() <- blocked for A
      |
      |
fbcon_del_cursor_timer ->
  del_timer_sync
  (BOOM)

Of course because all of this is under the console lock,
we never see anything, also since we also just unbound the active
console guess what we never see anything.

Hopefully this fixes the problem for anyone seeing vesafb->kms
driver handoff.

v1.1: add comment suggestion from Alan.

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

Revert "drm/radeon: fix bo creation retry path"

commit 676bc2e1e4f9072f7a640d5b7c99ffdf9709a6e7 upstream.

This reverts commit d1c7871ddb1f588b8eb35affd9ee1a3d5e11cd0c.

ttm_bo_init() destroys the BO on failure. So this patch makes
the retry path work with freed memory.  This ends up causing
kernel panics when this path is hit.

Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[bwh: Backported to 3.2: adjust context]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

svcrpc: fix svc_xprt_enqueue/svc_recv busy-looping

commit d10f27a750312ed5638c876e4bd6aa83664cccd8 upstream.

The rpc server tries to ensure that there will be room to send a reply
before it receives a request.

It does this by tracking, in xpt_reserved, an upper bound on the total
size of the replies that is has already committed to for the socket.

Currently it is adding in the estimate for a new reply *before* it
checks whether there is space available.  If it finds that there is not
space, it then subtracts the estimate back out.

This may lead the subsequent svc_xprt_enqueue to decide that there is
space after all.

The results is a svc_recv() that will repeatedly return -EAGAIN, causing
server threads to loop without doing any actual work.

Reported-by: Michael Tokarev <mjt@tls.msk.ru>
Tested-by: Michael Tokarev <mjt@tls.msk.ru>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

svcrpc: sends on closed socket should stop immediately

commit f06f00a24d76e168ecb38d352126fd203937b601 upstream.

svc_tcp_sendto sets XPT_CLOSE if we fail to transmit the entire reply.
However, the XPT_CLOSE won't be acted on immediately.  Meanwhile other
threads could send further replies before the socket is really shut
down.  This can manifest as data corruption: for example, if a truncated
read reply is followed by another rpc reply, that second reply will look
to the client like further read data.

Symptoms were data corruption preceded by svc_tcp_sendto logging
something like

kernel: rpc-srv/tcp: nfsd: sent only 963696 when sending 1048708 bytes - shutting down socket

Reported-by: Malahal Naineni <malahal@us.ibm.com>
Tested-by: Malahal Naineni <malahal@us.ibm.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

svcrpc: fix BUG() in svc_tcp_clear_pages

commit be1e44441a560c43c136a562d49a1c9623c91197 upstream.

Examination of svc_tcp_clear_pages shows that it assumes sk_tcplen is
consistent with sk_pages[] (in particular, sk_pages[n] can't be NULL if
sk_tcplen would lead us to expect n pages of data).

svc_tcp_restore_pages zeroes out sk_pages[] while leaving sk_tcplen.
This is OK, since both functions are serialized by XPT_BUSY.  However,
that means the inconsistency must be repaired before dropping XPT_BUSY.

Therefore we should be ensuring that svc_tcp_save_pages repairs the
problem before exiting svc_tcp_recv_record on error.

Symptoms were a BUG() in svc_tcp_clear_pages.

Signed-off-by: J. Bruce Fields <bfields@redhat.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

vfs: missed source of ->f_pos races

commit 0e665d5d1125f9f4ccff56a75e814f10f88861a2 upstream.

compat_sys_{read,write}v() need the same "pass a copy of file->f_pos" thing
as sys_{read,write}{,v}().

Signed-off-by: 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>

NFSv3: Ensure that do_proc_get_root() reports errors correctly

commit 086600430493e04b802bee6e5b3ce0458e4eb77f upstream.

If the rpc call to NFS3PROC_FSINFO fails, then we need to report that
error so that the mount fails. Otherwise we can end up with a
superblock with completely unusable values for block sizes, maxfilesize,
etc.

Reported-by: Yuanming Chen <hikvision_linux@163.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

drm/radeon: split ATRM support out from the ATPX handler (v3)

commit c61e2775873f603148e8e998a938721b7d222d24 upstream.

There are systems that use ATRM, but not ATPX.

Fixes:
https://bugs.freedesktop.org/show_bug.cgi?id=41265

V2: fix #ifdefs as per Greg's comments
V3: fix it harder

Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

drm/radeon: fix use after free in ATRM bios reading code.

commit de47a9cd62771e3e78954d855d2304fbad4c5a44 upstream.

Fixes:
https://bugs.freedesktop.org/show_bug.cgi?id=45503

Reported-and-Debugged-by: mlambda@gmail.com
Signed-off-by: Dave Airlie <airlied@redhat.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>