This reverts commit af14cca1.

This patch contains a race condition between packets and ctnetlink
in the conntrack addition. A new patch to fix this issue follows up.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

In the current kernel implementation, the Logitech Harmony 900 remote
control is matched to the cdc_ether driver through the generic
USB_CDC_SUBCLASS_MDLM entry.  However, this device appears to be of the
pseudo-MDLM (Belcarra) type, rather than the standard one.  This patch
blacklists the Harmony 900 from the cdc_ether driver and whitelists it for
the pseudo-MDLM driver in zaurus.
Signed-off-by: Scott Talbert <talbert@techie.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

The intent was to clear out the icount struct here, but we accidentally
clear stack memory instead.  It probably will lead to a NULL dereference
right away.
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ip6_route_output() never returns NULL, so it is wrong to
check if the return value is NULL.
Signed-off-by: RongQing.Li <roy.qing.li@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ip6_route_output() never returns NULL, so it is wrong to
check if the return value is NULL.
Signed-off-by: RongQing.Li <roy.qing.li@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ip6_route_output() never returns NULL, so it is wrong to
check if the return value is NULL.
Signed-off-by: RongQing.Li <roy.qing.li@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Set the RX FIFO flush watermark lower.
According to Federico and JMicron's reply,
setting it to 16QW would be stable on most platforms.
Otherwise, user might experience packet drop issue.

CC: stable@kernel.org
Reported-by: Federico Quagliata <federico@quagliata.org>
Fixed-by: Federico Quagliata <federico@quagliata.org>
Signed-off-by: Guo-Fu Tseng <cooldavid@cooldavid.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

Commit 32092ecf (atm: clip: Use device neigh support on top of
"arp_tbl".) introduced a bug since clip_tbl is zeroed : Crash occurs in
__neigh_for_each_release()

idle_timer_check() must use instead arp_tbl and neigh_check_cb() should
ignore non clip neighbours.

Idea from David Miller.
Reported-by: Meelis Roos <mroos@linux.ee>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Tested-by: Meelis Roos <mroos@linux.ee>
Signed-off-by: David S. Miller <davem@davemloft.net>

Don't return an uninitialized variable as the error, return
-EOPNOTSUPP instead.
Reported-by: Dave Jones <davej@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Implement a new netlink attribute type IFLA_EXT_MASK.  The mask
is a 32 bit value that can be used to indicate to the kernel that
certain extended ifinfo values are requested by the user application.
At this time the only mask value defined is RTEXT_FILTER_VF to
indicate that the user wants the ifinfo dump to send information
about the VFs belonging to the interface.

This patch fixes a bug in which certain applications do not have
large enough buffers to accommodate the extra information returned
by the kernel with large numbers of SR-IOV virtual functions.
Those applications will not send the new netlink attribute with
the interface info dump request netlink messages so they will
not get unexpectedly large request buffers returned by the kernel.

Modifies the rtnl_calcit function to traverse the list of net
devices and compute the minimum buffer size that can hold the
info dumps of all matching devices based upon the filter passed
in via the new netlink attribute filter mask.  If no filter
mask is sent then the buffer allocation defaults to NLMSG_GOODSIZE.

With this change it is possible to add yet to be defined netlink
attributes to the dump request which should make it fairly extensible
in the future.
Signed-off-by: Greg Rose <gregory.v.rose@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

When the fixed race condition happens:

1. While function neigh_periodic_work scans the neighbor hash table
pointed by field tbl->nht, it unlocks and locks tbl->lock between
buckets in order to call cond_resched.

2. Assume that function neigh_periodic_work calls cond_resched, that is,
the lock tbl->lock is available, and function neigh_hash_grow runs.

3. Once function neigh_hash_grow finishes, and RCU calls
neigh_hash_free_rcu, the original struct neigh_hash_table that function
neigh_periodic_work was using doesn't exist anymore.

4. Once back at neigh_periodic_work, whenever the old struct
neigh_hash_table is accessed, things can go badly.
Signed-off-by: Michel Machado <michel@digirati.com.br>
Acked-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

In port type change flow, need to set the new port types only after
all interfaces have finished the unregister process.
Otherwise, during unregister, one of the interfaces might issue a SET_PORT
command with wrong port types, it can cause bad FW behavior.
Signed-off-by: Yevgeny Petrilin <yevgenyp@mellanox.co.il>
Signed-off-by: David S. Miller <davem@davemloft.net>

Under the spinlock we call request_irq(), which allocates memory with GFP_KERNEL,
This causes the following trace when DEBUG_SPINLOCK is enabled, it can cause
the following trace:

 BUG: spinlock wrong CPU on CPU#2, ethtool/2595
 lock: ffff8801f9cbc2b0, .magic: dead4ead, .owner: ethtool/2595, .owner_cpu: 0
 Pid: 2595, comm: ethtool Not tainted 3.0.18 #2
 Call Trace:
 spin_bug+0xa2/0xf0
 do_raw_spin_unlock+0x71/0xa0
 _raw_spin_unlock+0xe/0x10
 mlx4_assign_eq+0x12b/0x190 [mlx4_core]
 mlx4_en_activate_cq+0x252/0x2d0 [mlx4_en]
 ? mlx4_en_activate_rx_rings+0x227/0x370 [mlx4_en]
 mlx4_en_start_port+0x189/0xb90 [mlx4_en]
 mlx4_en_set_ringparam+0x29a/0x340 [mlx4_en]
 dev_ethtool+0x816/0xb10
 ? dev_get_by_name_rcu+0xa4/0xe0
 dev_ioctl+0x2b5/0x470
 handle_mm_fault+0x1cd/0x2d0
 sock_do_ioctl+0x5d/0x70
 sock_ioctl+0x79/0x2f0
 do_vfs_ioctl+0x8c/0x340
 sys_ioctl+0xa1/0xb0
 system_call_fastpath+0x16/0x1b

Replacing with mutex, which is enough in this case.
Signed-off-by: Yevgeny Petrilin <yevgenyp@mellanox.co.il>
Signed-off-by: David S. Miller <davem@davemloft.net>

In 16adf5d0 I removed an over-broad
alias that caused zaurus.ko to bind to unrelated devices.
I had a report that at least one valid case no longer auto-loads because of this.
This patch adds an ID for that case.
Reported-by: Raphael Wimmer <raphael.wimmer@ur.de>
Signed-off-by: Dave Jones <davej@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Marcell Zambo and Janos Farago noticed and reported that when
new conntrack entries are added via netlink and the conntrack table
gets full, soft lockup happens. This is because the nf_conntrack_lock
is held while nf_conntrack_alloc is called, which is in turn wants
to lock nf_conntrack_lock while evicting entries from the full table.

The patch fixes the soft lockup with limiting the holding of the
nf_conntrack_lock to the minimum, where it's absolutely required.
Signed-off-by: Jozsef Kadlecsik <kadlec@blackhole.kfki.hu>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

ebt_among extension of ebtables uses __alignof__(_xt_align) while the
corresponding kernel module uses __alignof__(ebt_replace) to determine
the alignment in EBT_ALIGN().

These are the results of these values on different platforms:

x86 x86_64 ppc
__alignof__(_xt_align) 4 8 8
__alignof__(ebt_replace) 4 8 4

ebtables fails to add rules which use the among extension.

I'm using kernel 2.6.33 and ebtables 2.0.10-4

According to Bart De Schuymer, userspace alignment was changed to
_xt_align to fix an alignment issue on a userspace32-kernel64 system
(he thinks it was for an ARM device). So userspace must be right.
The kernel alignment macro needs to change so it also uses _xt_align
instead of ebt_replace. The userspace changes date back from
June 29, 2009.
Signed-off-by: Joerg Willmann <joe@clnt.de>
Signed-off by: Bart De Schuymer <bdschuym@pandora.be>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

BF can be disabled in some cases, the capability field, bf_reg_size is set
to zero in this case. Don't map the BF area in this case, it would cause
failures.  In addition, leaving the BF area unmapped
also alerts the ETH driver to not use BF.
Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/linville/wireless into for-davem

In the case of hotplug enabled devices (PCMCIA/PCIeC) the removal of the
hardware can cause an infinite loop in the common sja1000 isr.

Use the already retrieved status register to indicate a possible hardware
removal and double check by reading the mode register in sja1000_is_absent.

Cc: stable@kernel.org [3.2+]
Signed-off-by: Oliver Hartkopp <socketcan@hartkopp.net>
Acked-by: Wolfgang Grandegger <wg@grandegger.com>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>

WARNING: drivers/net/ethernet/broadcom/built-in.o(.init.text+0x5d): Section mismatch in reference from the function b44_init() to the function .exit.text:b44_pci_exit()

module exits with b44_cleanup()
Signed-off-by: Nikola Pajkovsky <n.pajkovsky@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The calls to pci_iounmap are in the wrong order, as compared to the
associated calls to pci_iomap.

A simplified version of the semantic match that finds this problem is as
follows: (http://coccinelle.lip6.fr/)

// <smpl>
@@
expression e,x;
statement S,S1;
int ret;
@@
e = pci_iomap(x,...)
... when != pci_iounmap(x,e)
if (<+...e...+>) S
... when any
    when != pci_iounmap(x,e)
*if (...)
   { ... when != pci_iounmap(x,e)
     return ...; }
... when any
pci_iounmap(x,e);
// </smpl>
Signed-off-by: Julia Lawall <Julia.Lawall@lip6.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>

netdev->irq is unsigned, so it's never less than zero.
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Tested-by: Jan Weitzel <j.weitzel@phytec.de>
Signed-off-by: David S. Miller <davem@davemloft.net>

I was recently shown that vmxnet3 devices on transmit, will drop very small udp
frames consistently.  This is due to a regression introduced by commit
39d4a96f.  This commit attempts to introduce an
optimization to the tx path, indicating that the underlying hardware behaves
optimally when at least 54 bytes of header data are available for direct access.
This causes problems however, if the entire frame is less than 54 bytes long.
The subsequent pskb_may_pull in vmxnet3_parse_and_copy_hdr fails, causing an
error return code, which leads to vmxnet3_tq_xmit dropping the frame.

Fix it by placing a cap on the copy length.  For frames longer than 54 bytes, we
do the pull as we normally would.  If the frame is shorter than that, copy the
whole frame, but no more.  This ensures that we still get the optimization for
qualifying frames, but don't do any damange for frames that are too short.

Also, since I'm unable to do this, it wuold be great if vmware could follow up
this patch with some additional code commentary as to why 54 bytes is an optimal
pull length for a virtual NIC driver.  The comment that introduced this was
vague on that.  Thanks!
Signed-off-by: Neil Horman <nhorman@tuxdriver.com>
Reported-by: Max Matveev <mmatveev@redhat.com>
CC: Max Matveev <mmatveev@redhat.com>
CC: "David S. Miller" <davem@davemloft.net>
CC: Shreyas Bhatewara <sbhatewara@vmware.com>
CC: "VMware, Inc." <pv-drivers@vmware.com>
Signed-off-by: Shreyas N Bhatewara <sbhatewara@vmware.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

This driver attempts to use two TX rings but lacks proper support :

1) IRQ handler only takes care of TX completion on first TX ring
2) the stop/start logic uses the legacy functions (for non multiqueue
drivers)

This means all packets witk skb mark set to 1 are sent through high
queue but are never cleaned and queue eventualy fills and block the
device, triggering the infamous "NETDEV WATCHDOG" message.

Lets use a single TX ring to fix the problem, this driver is not a real
multiqueue one yet.

Minimal fix for stable kernels.
Reported-by: Thomas Meyer <thomas@m3y3r.de>
Tested-by: Thomas Meyer <thomas@m3y3r.de>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Jay Cliburn <jcliburn@gmail.com>
Cc: Chris Snook <chris.snook@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

commit 50612537 (netem: fix classful handling) added two errors in
netem_dequeue()

1) After checking skb at the head of tfifo queue for time constraints,
   it dequeues tail skb, thus adding unwanted reordering.

2) qdisc stats are updated twice per packet
   (one when packet dequeued from tfifo, once when delivered)
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

These are the bug fixes that have accumulated since 3.3-rc3 in arm-soc.
The majority of them are regression fixes for stuff that broke during
the merge 3.3 window.

The notable ones are:

* The at91 ata drivers both broke because of an earlier cleanup patch that
  some other patches were based on. Jean-Christophe decided to remove
  the legacy at91_ide driver and fix the new-style at91-pata driver while
  keeping the cleanup patch. I almost rejected the patches for being too
  late and too big but in the end decided to accept them because they
  fix a regression.

* A patch fixing build breakage from the sysdev-to-device conversion
  colliding with other changes touches a number of mach-s3c files.

* b0654037 "ARM: orion: Fix Orion5x GPIO regression from MPP cleanup"
  is a mechanical change that unfortunately touches a lot of lines
  that should up in the diffstat.

* tag 'fixes-3.3-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/arm/arm-soc: (28 commits)
  ARM: at91: drop ide driver in favor of the pata one
  pata/at91: use newly introduced SMC accessors
  ARM: at91: add accessor to manage SMC
  ARM: at91:rtc/rtc-at91sam9: ioremap register bank
  ARM: at91: USB AT91 gadget registration for module
  ep93xx: fix build of vision_ep93xx.c
  ARM: OMAP2xxx: PM: fix OMAP2xxx-specific UART idle bug in v3.3
  ARM: orion: Fix USB phy for orion5x.
  ARM: orion: Fix Orion5x GPIO regression from MPP cleanup
  ARM: EXYNOS: Add cpu-offset property in gic device tree node
  ARM: EXYNOS: Bring exynos4-dt up to date
  ARM: OMAP3: cm-t35: fix section mismatch warning
  ARM: OMAP2: Fix the OMAP2 only build break seen with 2011+ ARM tool-chains
  ARM: tegra: paz00: fix wrong UART port on mini-pcie plug
  ARM: tegra: paz00: fix wrong SD1 power gpio
  i2c: tegra: Add devexit_p() for remove
  ARM: EXYNOS: Correct M-5MOLS sensor clock frequency on Universal C210 board
  ARM: EXYNOS: Correct framebuffer window size on Nuri board
  ARM: SAMSUNG: Fix missing api-change from subsys_interface change
  ARM: EXYNOS: Fix "warning: initialization from incompatible pointer type"
  ...

1) VETH_INFO_PEER netlink attribute needs to have it's size validated,
   from Thomas Graf.

2) 'poll' module option of bnx2x driver crashes the machine, just remove
   it.  From Michal Schmidt.

3) ks8851_mll driver reads the irq number from two places, but only
   initializes one of them, oops.  Use only one location and fix this
   problem, from Jan Weitzel.

4) Fix buffer overrun and unicast sterring bugs in mellanox mlx4 driver,
   from Eugenia Emantayev.

5) Swapped kcalloc() args in RxRPC and mlx4, from Axel Lin.

6) PHY MDIO device name regression fixes from Florian Fainelli.

7) If the wake event IRQ line is different from the netdevice one, we
   have to properly route it to the stmmac interrupt handler.  From
   Francesco Virlinzi.

8) Fix rwlock lock initialization ordering bug in mac80211, from
   Mohammed Shafi Shajakhan.

9) TCP lost_cnt can get out of sync, and in fact go negative, in certain
   circumstances.  Fix the way we specify what sequence range to operate
   on in tcp_sacktag_one() to fix this bug.  From Neal Cardwell.

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: (27 commits)
  net/ethernet: ks8851_mll fix irq handling
  veth: Enforce minimum size of VETH_INFO_PEER
  stmmac: update the driver version to Feb 2012 (v2)
  stmmac: move hw init in the probe (v2)
  stmmac: request_irq when use an ext wake irq line (v2)
  stmmac: do not discard frame on dribbling bit assert
  ipheth: Add iPhone 4S
  mlx4: add unicast steering entries to resource_tracker
  mlx4: fix QP tree trashing
  mlx4: fix buffer overrun
  3c59x: shorten timer period for slave devices
  netpoll: netpoll_poll_dev() should access dev->flags
  RxRPC: Fix kcalloc parameters swapped
  bnx2x: remove the 'poll' module option
  tcp: fix tcp_shifted_skb() adjustment of lost_cnt_hint for FACK
  ks8851: Fix NOHZ local_softirq_pending 08 warning
  bnx2x: fix bnx2x_storm_stats_update() on big endian
  ixp4xx-eth: fix PHY name to match MDIO bus name
  octeon: fix PHY name to match MDIO bus name
  fec: fix PHY name to match fixed MDIO bus name
  ...

Fixes a bootstrapping issue for some registers when a less commonly used
method for register cache initialisation is used.  Only affects a fairly
small proportion of users that both don't use explicit register defaults
and do use the cache.

* tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/broonie/regmap:
  regmap: Fix cache defaults initialization from raw cache defaults

Merge tag 'ecryptfs-3.3-rc4-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/tyhicks/ecryptfs

Fixes maximum filename length and filesystem type reporting in statfs() calls
and also fixes stale inode mode bits on eCryptfs inodes after a POSIX ACL was
set on the lower filesystem's inode.

* tag 'ecryptfs-3.3-rc4-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/tyhicks/ecryptfs:
  ecryptfs: remove the second argument of k[un]map_atomic()
  eCryptfs: Copy up lower inode attrs after setting lower xattr
  eCryptfs: Improve statfs reporting

pinctrl fixes for v3.3

* tag 'pinctrl-for-torvalds-20120216' of git://git.kernel.org/pub/scm/linux/kernel/git/linusw/linux-pinctrl:
  pinctrl: restore pin naming

Here are a few more fixes for powerpc.  Some are regressions, the rest
is simple/obvious/nasty enough that I deemed it good to go now.

Here's also step one of deprecating legacy iSeries support: we are
removing it from the main defconfig.

Nobody seems to be using it anymore and the code is nasty to maintain,
(involves horrible hacks in various low level areas of the kernel) so we
plan to actually rip it out at some point.  For now let's just avoid
building it by default.  Stephen will proceed to do the actual removal
later (probably 3.4 or 3.5).

* 'merge' of git://git.kernel.org/pub/scm/linux/kernel/git/benh/powerpc:
  powerpc/perf: power_pmu_start restores incorrect values, breaking frequency events
  powerpc/adb: Use set_current_state()
  powerpc: Disable interrupts early in Program Check
  powerpc: Remove legacy iSeries from ppc64_defconfig
  powerpc/fsl/pci: Fix PCIe fixup regression
  powerpc: Fix kernel log of oops/panic instruction dump

One regression fix for SR-IOV on PPC and a couple of misc fixes from
Yinghai.

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jbarnes/pci:
  PCI: Fix pci cardbus removal
  PCI: set pci sriov page size before reading SRIOV BAR
  PCI: workaround hard-wired bus number V2

3 radeon fixes, I have some exynos fixes to push later but I'll queue
them separately once I've looked them over a bit.

* 'drm-fixes' of git://people.freedesktop.org/~airlied/linux:
  drm/radeon/kms: fix MSI re-arm on rv370+
  drm/radeon/kms/atom: bios scratch reg handling updates
  drm/radeon/kms: drop lock in return path of radeon_fence_count_emitted.

* git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6:
  crypto: sha512 - use standard ror64()

After all the FPU state cleanups and finally finding the problem that
caused all our FPU save/restore problems, this re-introduces the
preloading of FPU state that was removed in commit b3b0870e ("i387:
do not preload FPU state at task switch time").

However, instead of simply reverting the removal, this reimplements
preloading with several fixes, most notably

 - properly abstracted as a true FPU state switch, rather than as
   open-coded save and restore with various hacks.

   In particular, implementing it as a proper FPU state switch allows us
   to optimize the CR0.TS flag accesses: there is no reason to set the
   TS bit only to then almost immediately clear it again.  CR0 accesses
   are quite slow and expensive, don't flip the bit back and forth for
   no good reason.

 - Make sure that the same model works for both x86-32 and x86-64, so
   that there are no gratuitous differences between the two due to the
   way they save and restore segment state differently due to
   architectural differences that really don't matter to the FPU state.

 - Avoid exposing the "preload" state to the context switch routines,
   and in particular allow the concept of lazy state restore: if nothing
   else has used the FPU in the meantime, and the process is still on
   the same CPU, we can avoid restoring state from memory entirely, just
   re-expose the state that is still in the FPU unit.

   That optimized lazy restore isn't actually implemented here, but the
   infrastructure is set up for it.  Of course, older CPU's that use
   'fnsave' to save the state cannot take advantage of this, since the
   state saving also trashes the state.

In other words, there is now an actual _design_ to the FPU state saving,
rather than just random historical baggage.  Hopefully it's easier to
follow as a result.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

This moves the bit that indicates whether a thread has ownership of the
FPU from the TS_USEDFPU bit in thread_info->status to a word of its own
(called 'has_fpu') in task_struct->thread.has_fpu.

This fixes two independent bugs at the same time:

 - changing 'thread_info->status' from the scheduler causes nasty
   problems for the other users of that variable, since it is defined to
   be thread-synchronous (that's what the "TS_" part of the naming was
   supposed to indicate).

   So perfectly valid code could (and did) do

	ti->status |= TS_RESTORE_SIGMASK;

   and the compiler was free to do that as separate load, or and store
   instructions.  Which can cause problems with preemption, since a task
   switch could happen in between, and change the TS_USEDFPU bit. The
   change to TS_USEDFPU would be overwritten by the final store.

   In practice, this seldom happened, though, because the 'status' field
   was seldom used more than once, so gcc would generally tend to
   generate code that used a read-modify-write instruction and thus
   happened to avoid this problem - RMW instructions are naturally low
   fat and preemption-safe.

 - On x86-32, the current_thread_info() pointer would, during interrupts
   and softirqs, point to a *copy* of the real thread_info, because
   x86-32 uses %esp to calculate the thread_info address, and thus the
   separate irq (and softirq) stacks would cause these kinds of odd
   thread_info copy aliases.

   This is normally not a problem, since interrupts aren't supposed to
   look at thread information anyway (what thread is running at
   interrupt time really isn't very well-defined), but it confused the
   heck out of irq_fpu_usable() and the code that tried to squirrel
   away the FPU state.

   (It also caused untold confusion for us poor kernel developers).

It also turns out that using 'task_struct' is actually much more natural
for most of the call sites that care about the FPU state, since they
tend to work with the task struct for other reasons anyway (ie
scheduling).  And the FPU data that we are going to save/restore is
found there too.

Thanks to Arjan Van De Ven <arjan@linux.intel.com> for pointing us to
the %esp issue.

Cc: Arjan van de Ven <arjan@linux.intel.com>
Reported-and-tested-by: Raphael Prevost <raphael@buro.asia>
Acked-and-tested-by: Suresh Siddha <suresh.b.siddha@intel.com>
Tested-by: Peter Anvin <hpa@zytor.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

The AMD K7/K8 CPUs don't save/restore FDP/FIP/FOP unless an exception is
pending.  In order to not leak FIP state from one process to another, we
need to do a floating point load after the fxsave of the old process,
and before the fxrstor of the new FPU state.  That resets the state to
the (uninteresting) kernel load, rather than some potentially sensitive
user information.

We used to do this directly after the FPU state save, but that is
actually very inconvenient, since it

 (a) corrupts what is potentially perfectly good FPU state that we might
     want to lazy avoid restoring later and

 (b) on x86-64 it resulted in a very annoying ordering constraint, where
     "__unlazy_fpu()" in the task switch needs to be delayed until after
     the DS segment has been reloaded just to get the new DS value.

Coupling it to the fxrstor instead of the fxsave automatically avoids
both of these issues, and also ensures that we only do it when actually
necessary (the FP state after a save may never actually get used).  It's
simply a much more natural place for the leaked state cleanup.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>