linux-2.6
17 years agosata_sil24: fix IRQ clearing race when PCIX_IRQ_WOC is used
Tejun Heo [Sun, 23 Sep 2007 03:37:05 +0000 (12:37 +0900)] 
sata_sil24: fix IRQ clearing race when PCIX_IRQ_WOC is used

When PCIX_IRQ_WOC is used, sil24 has an inherent race condition
between clearing IRQ pending and reading IRQ status.  If IRQ pending
is cleared after reading IRQ status, there's possibility of lost IRQ.
If IRQ pending is cleared before reading IRQ status, spurious IRQs
will occur.

sata_sil24 till now cleared IRQ pending after reading IRQ status thus
losing IRQs on machines where PCIX_IRQ_WOC was used.  Reverse the
order and ignore spurious IRQs if PCIX_IRQ_WOC.

Signed-off-by: Tejun Heo <htejun@gmail.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
17 years agoLinux 2.6.23-rc8 v2.6.23-rc8
Linus Torvalds [Tue, 25 Sep 2007 00:33:10 +0000 (17:33 -0700)] 
Linux 2.6.23-rc8

Getting there...

17 years agoMerge branch 'upstream' of git://ftp.linux-mips.org/pub/scm/upstream-linus
Linus Torvalds [Tue, 25 Sep 2007 00:32:16 +0000 (17:32 -0700)] 
Merge branch 'upstream' of git://ftp.linux-mips.org/upstream-linus

* 'upstream' of git://ftp.linux-mips.org/pub/scm/upstream-linus:
  [MIPS] SMTC: Make ack_bad_irq() safe with no IM backstop.

17 years agoMerge branch 'async-tx-fixes-for-linus' of git://lost.foo-projects.org/~dwillia2...
Linus Torvalds [Tue, 25 Sep 2007 00:25:30 +0000 (17:25 -0700)] 
Merge branch 'async-tx-fixes-for-linus' of git://lost.foo-projects.org/~dwillia2/git/iop

* 'async-tx-fixes-for-linus' of git://lost.foo-projects.org/~dwillia2/git/iop:
  raid5: fix 2 bugs in ops_complete_biofill
  async_tx: fix dma_wait_for_async_tx
  async_tx: usage documentation and developer notes (v2)

17 years agoraid5: fix 2 bugs in ops_complete_biofill
Dan Williams [Mon, 24 Sep 2007 17:06:13 +0000 (10:06 -0700)] 
raid5: fix 2 bugs in ops_complete_biofill

1/ ops_complete_biofill tried to avoid calling handle_stripe since all the
state necessary to return read completions is available.  However the
process of determining whether more read requests are pending requires
locking the stripe (to block add_stripe_bio from updating dev->toead).
ops_complete_biofill can run in tasklet context, so rather than upgrading
all the stripe locks from spin_lock to spin_lock_bh this patch just
unconditionally reschedules handle_stripe after completing the read
request.

2/ ops_complete_biofill needlessly qualified processing R5_Wantfill with
dev->toread.  The result being that the 'biofill' pending bit is cleared
before handling the pending read-completions on dev->read.  R5_Wantfill can
be unconditionally handled because the 'biofill' pending bit prevents new
R5_Wantfill requests from being seen by ops_run_biofill and
ops_complete_biofill.

Found-by: Yuri Tikhonov <yur@emcraft.com>
[neilb@suse.de: simpler fix for bug 1 than moving code]
Signed-off-by: NeilBrown <neilb@suse.de>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
17 years agoasync_tx: fix dma_wait_for_async_tx
Dan Williams [Fri, 21 Sep 2007 20:27:04 +0000 (13:27 -0700)] 
async_tx: fix dma_wait_for_async_tx

Fix dma_wait_for_async_tx to not loop forever in the case where a
dependency chain is longer than two entries.  This condition will not
happen with current in-kernel drivers, but fix it for future drivers.

Found-by: Saeed Bishara <saeed.bishara@gmail.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
17 years agoasync_tx: usage documentation and developer notes (v2)
Dan Williams [Thu, 20 Sep 2007 22:49:08 +0000 (15:49 -0700)] 
async_tx: usage documentation and developer notes (v2)

Changes in v2:
* cleanups from Randy and Shannon

Reviewed-by: Randy Dunlap <randy.dunlap@oracle.com>
Reviewed-by: Shannon Nelson <shannon.nelson@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
17 years ago[MIPS] SMTC: Make ack_bad_irq() safe with no IM backstop.
Ralf Baechle [Fri, 21 Sep 2007 16:13:55 +0000 (17:13 +0100)] 
[MIPS] SMTC: Make ack_bad_irq() safe with no IM backstop.

Issue reported and original patch by Kevin Kissel, cleaner (imho)
implementation by me.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
17 years agoConvert snd-page-alloc proc file to use seq_file
Takashi Iwai [Mon, 17 Sep 2007 19:55:10 +0000 (21:55 +0200)] 
Convert snd-page-alloc proc file to use seq_file

Use seq_file for the proc file read/write of snd-page-alloc module.
This automatically fixes bugs in the old proc code.

Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/roland...
Linus Torvalds [Mon, 24 Sep 2007 05:39:57 +0000 (22:39 -0700)] 
Merge branch 'for-linus' of git://git./linux/kernel/git/roland/infiniband

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/roland/infiniband:
  IB/mlx4: Fix data corruption triggered by wrong headroom marking order

17 years agoMerge git://git.infradead.org/battery-2.6
Linus Torvalds [Sun, 23 Sep 2007 20:04:01 +0000 (13:04 -0700)] 
Merge git://git.infradead.org/battery-2.6

* git://git.infradead.org/battery-2.6:
  [PATCH] missing null termination in power supply uevent

17 years agoIB/mlx4: Fix data corruption triggered by wrong headroom marking order
Jack Morgenstein [Wed, 19 Sep 2007 16:52:25 +0000 (09:52 -0700)] 
IB/mlx4: Fix data corruption triggered by wrong headroom marking order

This is an addendum to commit 0e6e7416 ("IB/mlx4: Handle new FW
requirement for send request prefetching").  We also need to handle
prefetch marking properly for S/G segments, or else the HCA may end up
processing S/G segments that are not fully written and end up sending
the wrong data.  This can actually cause data corruption in practice,
especially on systems with relatively slow CPUs (where the HCA is more
likely to prefetch while the CPU is in the middle of writing a work
request into memory).

We write S/G segments in reverse order into the WQE, in order to
guarantee that the first dword of all cachelines containing S/G
segments is written last (overwriting the headroom invalidation
pattern).  The entire cacheline will thus contain valid data when the
invalidation pattern is overwritten.

Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
17 years agoclockevents: remove the suspend/resume workaround^Wthinko
Thomas Gleixner [Sat, 22 Sep 2007 22:29:06 +0000 (22:29 +0000)] 
clockevents: remove the suspend/resume workaround^Wthinko

In a desparate attempt to fix the suspend/resume problem on Andrews
VAIO I added a workaround which enforced the broadcast of the oneshot
timer on resume. This was actually resolving the problem on the VAIO
but was just a stupid workaround, which was not tackling the root
cause: the assignement of lower idle C-States in the ACPI processor_idle
code. The cpuidle patches, which utilize the dynamic tick feature and
go faster into deeper C-states exposed the problem again. The correct
solution is the previous patch, which prevents lower C-states across
the suspend/resume.

Remove the enforcement code, including the conditional broadcast timer
arming, which helped to pamper over the real problem for quite a time.
The oneshot broadcast flag for the cpu, which runs the resume code can
never be set at the time when this code is executed. It only gets set,
when the CPU is entering a lower idle C-State.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Len Brown <lenb@kernel.org>
Cc: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
Cc: Rafael J. Wysocki <rjw@sisk.pl>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoACPI: disable lower idle C-states across suspend/resume
Thomas Gleixner [Sat, 22 Sep 2007 22:29:05 +0000 (22:29 +0000)] 
ACPI: disable lower idle C-states across suspend/resume

device_suspend() calls ACPI suspend functions, which seems to have undesired
side effects on lower idle C-states. It took me some time to realize that
especially the VAIO BIOSes (both Andrews jinxed UP and my elfstruck SMP one)
show this effect. I'm quite sure that other bug reports against suspend/resume
about turning the system into a brick have the same root cause.

After fishing in the dark for quite some time, I realized that removing the ACPI
processor module before suspend (this removes the lower C-state functionality)
made the problem disappear. Interestingly enough the propability of having a
bricked box is influenced by various factors (interrupts, size of the ram image,
...). Even adding a bunch of printks in the wrong places made the problem go
away. The previous periodic tick implementation simply pampered over the
problem, which explains why the dyntick / clockevents changes made this more
prominent.

We avoid complex functionality during the boot process and we have to do the
same during suspend/resume. It is a similar scenario and equaly fragile.

Add suspend / resume functions to the ACPI processor code and disable the lower
idle C-states across suspend/resume. Fall back to the default idle
implementation (halt) instead.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Len Brown <lenb@kernel.org>
Cc: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
Cc: Rafael J. Wysocki <rjw@sisk.pl>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoMerge branch 'release' of git://git.kernel.org/pub/scm/linux/kernel/git/lenb/linux...
Linus Torvalds [Sat, 22 Sep 2007 19:56:48 +0000 (12:56 -0700)] 
Merge branch 'release' of git://git./linux/kernel/git/lenb/linux-acpi-2.6

* 'release' of git://git.kernel.org/pub/scm/linux/kernel/git/lenb/linux-acpi-2.6:
  ACPI: suspend: consolidate handling of Sx states addendum
  ACPI: suspend: consolidate handling of Sx states.
  ACPI: video: remove dmesg spam
  ACPI: video: _DOS=0 by default to prevent hotkey hang

17 years agoMerge branch 'for-linus' of git://oss.sgi.com:8090/xfs/xfs-2.6
Linus Torvalds [Sat, 22 Sep 2007 19:56:13 +0000 (12:56 -0700)] 
Merge branch 'for-linus' of git://oss.sgi.com:8090/xfs/xfs-2.6

* 'for-linus' of git://oss.sgi.com:8090/xfs/xfs-2.6:
  [XFS] fix valid but harmless sparse warning
  [XFS] fix filestreams on 32-bit boxes

17 years agoKVM: Fix virtualization menu help text
Avi Kivity [Sat, 22 Sep 2007 12:43:45 +0000 (14:43 +0200)] 
KVM: Fix virtualization menu help text

What guest drivers?

Cc: Jan Engelhardt <jengelh@computergmbh.de>
Signed-off-by: Avi Kivity <avi@qumranet.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoPull suspend.now into release branch
Len Brown [Sat, 22 Sep 2007 01:55:34 +0000 (21:55 -0400)] 
Pull suspend.now into release branch

17 years agoPull now into release branch
Len Brown [Sat, 22 Sep 2007 01:55:29 +0000 (21:55 -0400)] 
Pull now into release branch

17 years agoACPI: suspend: consolidate handling of Sx states addendum
Frans Pop [Thu, 20 Sep 2007 20:27:44 +0000 (22:27 +0200)] 
ACPI: suspend: consolidate handling of Sx states addendum

Make the S0 state be always reported as supported

Signed-off: Frans Pop <elendil@planet.nl>
Acked-by: Alexey Starikovskiy <astarikovskiy@suse.de>
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Signed-off-by: Len Brown <len.brown@intel.com>
17 years agoMerge master.kernel.org:/home/rmk/linux-2.6-arm
Linus Torvalds [Fri, 21 Sep 2007 21:05:45 +0000 (14:05 -0700)] 
Merge master.kernel.org:/home/rmk/linux-2.6-arm

* master.kernel.org:/home/rmk/linux-2.6-arm:
  [ARM] 4569/1: ep93xx_gpio_irq_type(): fix spurious enumeration offset for FGPIO handling
  [ARM] 4568/1: fix l2x0 cache invalidate handling of unaligned addresses

17 years agoRevert "x86_64: Quicklist support for x86_64"
Linus Torvalds [Fri, 21 Sep 2007 19:09:41 +0000 (12:09 -0700)] 
Revert "x86_64: Quicklist support for x86_64"

This reverts commit 34feb2c83beb3bdf13535a36770f7e50b47ef299.

Suresh Siddha points out that this one breaks the fundamental
requirement that you cannot free page table pages before the TLB caches
are flushed.  The quicklists do not give the same kinds of guarantees
that the mmu_gather structure does, at least not in NUMA configurations.

Requested-by: Suresh Siddha <suresh.b.siddha@intel.com>
Acked-by: Andi Kleen <ak@suse.de>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Christoph Lameter <clameter@sgi.com>
Cc: Asit Mallick <asit.k.mallick@intel.com>
Cc: Tony Luck <tony.luck@intel.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoMerge branch 'upstream' of git://ftp.linux-mips.org/pub/scm/upstream-linus
Linus Torvalds [Fri, 21 Sep 2007 17:00:52 +0000 (10:00 -0700)] 
Merge branch 'upstream' of git://ftp.linux-mips.org/upstream-linus

* 'upstream' of git://ftp.linux-mips.org/pub/scm/upstream-linus:
  [MIPS] BCM1480: include <linux/init.h>.
  [MIPS] BCM1480: Export zbbus_mhz.

17 years ago[MIPS] BCM1480: include <linux/init.h>.
Ralf Baechle [Fri, 21 Sep 2007 08:01:15 +0000 (09:01 +0100)] 
[MIPS] BCM1480: include <linux/init.h>.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
17 years ago[MIPS] BCM1480: Export zbbus_mhz.
Ralf Baechle [Fri, 21 Sep 2007 07:59:07 +0000 (08:59 +0100)] 
[MIPS] BCM1480: Export zbbus_mhz.

Symbol is required by the ZBus profiler.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
17 years agoMerge branch 'upstream-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mfashe...
Linus Torvalds [Fri, 21 Sep 2007 16:52:20 +0000 (09:52 -0700)] 
Merge branch 'upstream-linus' of git://git./linux/kernel/git/mfasheh/ocfs2

* 'upstream-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mfasheh/ocfs2:
  ocfs2: Pack vote message and response structures
  ocfs2: Don't double set write parameters
  ocfs2: Fix pos/len passed to ocfs2_write_cluster
  ocfs2: Allow smaller allocations during large writes

17 years agox86_64: Zero extend all registers after ptrace in 32bit entry path.
Andi Kleen [Fri, 21 Sep 2007 14:16:18 +0000 (16:16 +0200)] 
x86_64: Zero extend all registers after ptrace in 32bit entry path.

Strictly it's only needed for eax.

It actually does a little more than strictly needed -- the other registers
are already zero extended.

Also remove the now unnecessary and non functional compat task check
in ptrace.

This is CVE-2007-4573

Found by Wojciech Purczynski

Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoACPI: suspend: consolidate handling of Sx states.
Alexey Starikovskiy [Thu, 20 Sep 2007 17:32:35 +0000 (21:32 +0400)] 
ACPI: suspend: consolidate handling of Sx states.

Recent changes to sleep initialization in ACPI dropped reporting of supported Sx
states above S3. Fix that and also move S5 init into same file as other Sx.
The only functional change is adding printk() for S4 and S5 cases.

Signed-off-by: Alexey Starikovskiy <astarikovskiy@suse.de>
Acked-by: Rafael J. Wysocki <rjw@sisk.pl>
Signed-off-by: Len Brown <len.brown@intel.com>
17 years agoocfs2: Pack vote message and response structures
Sunil Mushran [Thu, 20 Sep 2007 17:59:48 +0000 (10:59 -0700)] 
ocfs2: Pack vote message and response structures

The ocfs2_vote_msg and ocfs2_response_msg structs needed to be
packed to ensure similar sizeofs in 32-bit and 64-bit arches. Without this,
we had inadvertantly broken 32/64 bit cross mounts.

Signed-off-by: Sunil Mushran <sunil.mushran@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>
17 years agoocfs2: Don't double set write parameters
Mark Fasheh [Wed, 19 Sep 2007 00:49:29 +0000 (17:49 -0700)] 
ocfs2: Don't double set write parameters

The target page offsets were being incorrectly set a second time in
ocfs2_prepare_page_for_write(), which was causing problems on a 16k page
size kernel. Additionally, ocfs2_write_failure() was incorrectly using those
parameters instead of the parameters for the individual page being cleaned
up.

Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>
17 years agoocfs2: Fix pos/len passed to ocfs2_write_cluster
Mark Fasheh [Mon, 17 Sep 2007 16:06:29 +0000 (09:06 -0700)] 
ocfs2: Fix pos/len passed to ocfs2_write_cluster

This was broken for file systems whose cluster size is greater than page
size. Pos needs to be incremented as we loop through the descriptors, and
len needs to be capped to the size of a single cluster.

Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>
17 years agoocfs2: Allow smaller allocations during large writes
Mark Fasheh [Mon, 17 Sep 2007 03:10:16 +0000 (20:10 -0700)] 
ocfs2: Allow smaller allocations during large writes

The ocfs2 write code loops through a page much like the block code, except
that ocfs2 allocation units can be any size, including larger than page
size. Typically it's equal to or larger than page size - most kernels run 4k
pages, the minimum ocfs2 allocation (cluster) size.

Some changes introduced during 2.6.23 changed the way writes to pages are
handled, and inadvertantly broke support for > 4k page size. Instead of just
writing one cluster at a time, we now handle the whole page in one pass.

This means that multiple (small) seperate allocations might happen in the
same pass. The allocation code howver typically optimizes by getting the
maximum which was reserved. This triggered a BUG_ON in the extend code where
it'd ask for a single bit (for one part of a > 4k page) and get back more
than it asked for.

Fix this by providing a variant of the high level allocation function which
allows the caller to specify a maximum. The traditional function remains and
just calls the new one with a maximum determined from the initial
reservation.

Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>
17 years ago[PATCH] missing null termination in power supply uevent
Stephen Hemminger [Thu, 20 Sep 2007 19:06:10 +0000 (12:06 -0700)] 
[PATCH] missing null termination in power supply uevent

Need to null terminate environment. Found by inspection
while looking for similar problems to platform uevent bug

Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
17 years agoMerge branch 'upstream-linus' of master.kernel.org:/pub/scm/linux/kernel/git/jgarzik...
Linus Torvalds [Thu, 20 Sep 2007 20:25:35 +0000 (13:25 -0700)] 
Merge branch 'upstream-linus' of /linux/kernel/git/jgarzik/libata-dev

* 'upstream-linus' of master.kernel.org:/pub/scm/linux/kernel/git/jgarzik/libata-dev:
  [libata] ahci: add ATI SB800 PCI IDs
  libata-sff: Fix documentation
  libata: Update the blacklist with a few more devices

17 years agosignalfd simplification
Davide Libenzi [Thu, 20 Sep 2007 19:40:16 +0000 (12:40 -0700)] 
signalfd simplification

This simplifies signalfd code, by avoiding it to remain attached to the
sighand during its lifetime.

In this way, the signalfd remain attached to the sighand only during
poll(2) (and select and epoll) and read(2).  This also allows to remove
all the custom "tsk == current" checks in kernel/signal.c, since
dequeue_signal() will only be called by "current".

I think this is also what Ben was suggesting time ago.

The external effect of this, is that a thread can extract only its own
private signals and the group ones.  I think this is an acceptable
behaviour, in that those are the signals the thread would be able to
fetch w/out signalfd.

Signed-off-by: Davide Libenzi <davidel@xmailserver.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agorpc: fix garbage in printk in svc_tcp_accept()
Wolfgang Walter [Thu, 20 Sep 2007 19:51:46 +0000 (15:51 -0400)] 
rpc: fix garbage in printk in svc_tcp_accept()

we upgraded the kernel of a nfs-server from 2.6.17.11 to 2.6.22.6. Since
then we get the message

lockd: too many open TCP sockets, consider increasing the number of nfsd threads
lockd: last TCP connect from ^\\236^\É^D

These random characters in the second line are caused by a bug in
svc_tcp_accept.

(Note: there are two previous __svc_print_addr(sin, buf, sizeof(buf))
calls in this function, either of which would initialize buf correctly;
but both are inside "if"'s and are not necessarily executed.  This is
less obvious in the second case, which is inside a dprintk(), which is a
macro which expands to an if statement.)

Signed-off-by: Wolfgang Walter <wolfgang.walter@studentenwerk.mhn.de>
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years ago[libata] ahci: add ATI SB800 PCI IDs
henry su [Thu, 20 Sep 2007 20:07:33 +0000 (16:07 -0400)] 
[libata] ahci: add ATI SB800 PCI IDs

ATI/AMD SB800 shares some device IDs with SB700,
and SB800 adds two more device IDs:0x4394,0x4395.

Signed-off-by: henry su <henry.su.ati@gmail.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
17 years agolibata-sff: Fix documentation
Alan Cox [Thu, 20 Sep 2007 14:03:07 +0000 (15:03 +0100)] 
libata-sff: Fix documentation

Code moved to ioread/iowrite but the comment didn't
Also note a posting issue

Signed-off-by: Alan Cox <alan@redhat.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
17 years agolibata: Update the blacklist with a few more devices
Alan Cox [Thu, 20 Sep 2007 14:22:47 +0000 (15:22 +0100)] 
libata: Update the blacklist with a few more devices

Signed-off-by: Alan Cox <alan@redhat.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
17 years agoMerge branch 'master' of master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6
Linus Torvalds [Thu, 20 Sep 2007 19:42:47 +0000 (12:42 -0700)] 
Merge branch 'master' of /linux/kernel/git/davem/net-2.6

* 'master' of master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6:
  [BNX2]: Add PHY workaround for 5709 A1.
  [PPP] L2TP: Fix skb handling in pppol2tp_xmit
  [PPP] L2TP: Fix skb handling in pppol2tp_recv_core
  [PPP] L2TP: Disallow non-UDP datagram sockets
  [PPP] pppoe: Fix double-free on skb after transmit failure
  [PKT_SCHED]: Fix 'SFQ qdisc crashes with limit of 2 packets'
  [NETFILTER]: MAINTAINERS update
  [NETFILTER]: nfnetlink_log: fix sending of multipart messages

17 years agoMerge branch 'upstream-linus' of master.kernel.org:/pub/scm/linux/kernel/git/jgarzik...
Linus Torvalds [Thu, 20 Sep 2007 19:42:23 +0000 (12:42 -0700)] 
Merge branch 'upstream-linus' of /linux/kernel/git/jgarzik/netdev-2.6

* 'upstream-linus' of master.kernel.org:/pub/scm/linux/kernel/git/jgarzik/netdev-2.6:
  sky2: version 1.18
  sky2: receive FIFO checking
  sky2: fe+ chip support
  sky2: reorganize chip revision features
  sky2: ethtool speed report bug
  sky2: fix VLAN receive processing (resend)
  phy: export phy_mii_ioctl
  myri10ge: Add support for PCI device id 9

17 years agosky2: version 1.18
Stephen Hemminger [Wed, 19 Sep 2007 22:36:47 +0000 (15:36 -0700)] 
sky2: version 1.18

Update version number

Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
17 years agosky2: receive FIFO checking
Stephen Hemminger [Wed, 19 Sep 2007 22:36:46 +0000 (15:36 -0700)] 
sky2: receive FIFO checking

A driver writer from another operating system hinted that
the versions of Yukon 2 chip with rambuffer (EC and XL) have
a hardware bug that if the FIFO ever gets completely full it
will hang. Sounds like a classic ring full vs ring empty wrap around
bug.

As a workaround, use the existing watchdog timer to check for
ring full lockup.

Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
17 years agosky2: fe+ chip support
Stephen Hemminger [Wed, 19 Sep 2007 22:36:45 +0000 (15:36 -0700)] 
sky2: fe+ chip support

Add support for newest Marvell chips.
The Yukon FE plus chip is found in some not yet released laptops.
Tested on hardware evaluation boards.

This version of the patch is for 2.6.23. It supersedes
the two previous patches that are sitting in netdev-2.6 (upstream branch).

Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
17 years agosky2: reorganize chip revision features
Stephen Hemminger [Wed, 19 Sep 2007 22:36:44 +0000 (15:36 -0700)] 
sky2: reorganize chip revision features

This patch should cause no functional changes in driver behaviour.
There are (too) many revisions of the Yukon 2 chip now. Instead of
adding more conditionals based on chip revision; rerganize into a
set of feature flags so adding new versions is less problematic.

Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
17 years agosky2: ethtool speed report bug
Stephen Hemminger [Wed, 19 Sep 2007 22:36:43 +0000 (15:36 -0700)] 
sky2: ethtool speed report bug

On 100mbit versions, the driver always reports gigabit speed
available. The correct modes are already computed, then overwritten.

Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
17 years agosky2: fix VLAN receive processing (resend)
Stephen Hemminger [Wed, 19 Sep 2007 22:36:42 +0000 (15:36 -0700)] 
sky2: fix VLAN receive processing (resend)

The length check for truncated frames was not correctly handling
the case where VLAN acceleration had already read the tag.
Also, the Yukon EX has some features that use high bit of status
as security tag.

Signed-off-by: Pierre-Yves Ritschard <pyr@spootnik.org>
Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
17 years agoieee1394: ohci1394: fix initialization if built non-modular
Stefan Richter [Thu, 20 Sep 2007 19:17:33 +0000 (21:17 +0200)] 
ieee1394: ohci1394: fix initialization if built non-modular

Initialization of ohci1394 was broken according to one reporter if the
driver was statically linked, i.e. not built as loadable module.  Dmesg:

  PCI: Device 0000:02:07.0 not available because of resource collisions
  ohci1394: Failed to enable OHCI hardware.

This was reported for a Toshiba Satellite 5100-503.  The cause is commit
8df4083c5291b3647e0381d3c69ab2196f5dd3b7 in Linux 2.6.19-rc1 which only
served purposes of early remote debugging via FireWire.  This
functionality is better provided by the currently out-of-tree driver
ohci1394_earlyinit.  Reversal of the commit was OK'd by Andi Kleen.

Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
17 years ago[BNX2]: Add PHY workaround for 5709 A1.
Michael Chan [Thu, 20 Sep 2007 18:04:58 +0000 (11:04 -0700)] 
[BNX2]: Add PHY workaround for 5709 A1.

Add the DIS_EARLY_DAC PHY workaround for 5709 A1.  Without it, link
sometimes does not come up.

Update version to 1.6.5.

Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 years ago[PPP] L2TP: Fix skb handling in pppol2tp_xmit
Herbert Xu [Wed, 19 Sep 2007 17:46:28 +0000 (10:46 -0700)] 
[PPP] L2TP: Fix skb handling in pppol2tp_xmit

This patch makes pppol2tp_xmit call skb_cow_head so that we don't modify
cloned skb data.  It also gets rid of skb2 we only need to preserve the
original skb for congestion notification, which is only applicable for
ppp_async and ppp_sync.

The other semantic change made here is the removal of socket accounting
for data tranmitted out of pppol2tp_xmit.  The original code leaked any
existing socket skb accounting.  We could fix this by dropping the
original skb owner.  However, this is undesirable as the packet has not
physically left the host yet.

In fact, all other tunnels in the kernel do not account skb's passing
through to their own socket.  In partciular, ESP over UDP does not do
so and it is the closest tunnel type to PPPoL2TP.  So this patch simply
removes the socket accounting in pppol2tp_xmit.  The accounting still
applies to control packets of course.

I've also added a reminder that the outgoing checksum here doesn't work.
I suppose existing deployments don't actually enable checksums.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 years ago[PPP] L2TP: Fix skb handling in pppol2tp_recv_core
Herbert Xu [Tue, 18 Sep 2007 20:18:42 +0000 (13:18 -0700)] 
[PPP] L2TP: Fix skb handling in pppol2tp_recv_core

The function pppol2tp_recv_core doesn't handle non-linear packets properly.
It also fails to check the remote offset field.

This patch fixes these problems.  It also removes an unnecessary check on
the UDP header which has already been performed by the UDP layer.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 years ago[PPP] L2TP: Disallow non-UDP datagram sockets
Herbert Xu [Tue, 18 Sep 2007 20:18:17 +0000 (13:18 -0700)] 
[PPP] L2TP: Disallow non-UDP datagram sockets

With the addition of UDP-Lite we need to refine the socket check so
that only genuine UDP sockets are allowed through.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 years ago[PPP] pppoe: Fix double-free on skb after transmit failure
Herbert Xu [Wed, 19 Sep 2007 17:45:02 +0000 (10:45 -0700)] 
[PPP] pppoe: Fix double-free on skb after transmit failure

When I got rid of the second packet in __pppoe_xmit I created
a double-free on the skb because of the goto abort on failure.
This patch removes that.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 years ago[PKT_SCHED]: Fix 'SFQ qdisc crashes with limit of 2 packets'
Alexey Kuznetsov [Wed, 19 Sep 2007 17:42:03 +0000 (10:42 -0700)] 
[PKT_SCHED]: Fix 'SFQ qdisc crashes with limit of 2 packets'

Acked-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 years ago[NETFILTER]: MAINTAINERS update
Patrick McHardy [Tue, 18 Sep 2007 20:19:26 +0000 (13:19 -0700)] 
[NETFILTER]: MAINTAINERS update

Update netfilter list addresses and an old email address of myself.

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 years ago[NETFILTER]: nfnetlink_log: fix sending of multipart messages
Eric Leblond [Tue, 18 Sep 2007 20:07:15 +0000 (13:07 -0700)] 
[NETFILTER]: nfnetlink_log: fix sending of multipart messages

The following patch fixes the handling of netlink packets containing
multiple messages.

As exposed during netfilter workshop, nfnetlink_log was overwritten the
message type of the last message (setting it to MSG_DONE) in a multipart
packet. The consequence was libnfnetlink to ignore the last message in the
packet.

The following patch adds a supplementary message (with type MSG_DONE) af
the end of the netlink skb.

Signed-off-by: Eric Leblond <eric@inl.fr>
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 years agoFix CRLF line endings in Documentation/input/iforce-protocol.txt
Linus Torvalds [Thu, 20 Sep 2007 18:33:45 +0000 (11:33 -0700)] 
Fix CRLF line endings in Documentation/input/iforce-protocol.txt

Emil Medve points out that this documentation file uses CRLF line
endings, which means that if you use

[core]
autocrlf=input

(which makes sense if you ever develop under Windows, for example, or if
you use other broken tools) in your git config, git will always complain
about the file being dirty.

This removes the bogus DOS line endings, and removes whitespace at the
end of line.

Cc: Emil Medve <Emilian.Medve@Freescale.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years ago[x86 setup] Fix typo in arch/i386/boot/header.S
Paul Bolle [Mon, 10 Sep 2007 21:39:02 +0000 (23:39 +0200)] 
[x86 setup] Fix typo in arch/i386/boot/header.S

There's an obvious typo in arch/i386/boot/header.S (in your
linux-2.6-x86setup.git) that I noticed by just studying the code.

Signed-off-by: Paul Bolle <pebolle@tiscali.nl>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
17 years ago[acpi] Correct the decoding of video mode numbers in wakeup.S
H. Peter Anvin [Thu, 13 Sep 2007 21:16:37 +0000 (14:16 -0700)] 
[acpi] Correct the decoding of video mode numbers in wakeup.S

wakeup.S looks at the video mode number from the setup header and
looks to see if it is a VESA mode.  Unfortunately, the decoding is
done incorrectly and it will attempt to frob the VESA BIOS for any
mode number 0x0200 or larger.  Correct this, and remove a bunch of #if
0'd code.

Massive thanks to Jeff Chua for reporting the bug, and suffering
though a large number of experiments in order to track this problem
down.

Cc: Pavel Machek <pavel@ucw.cz>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
17 years ago[x86 setup] Present the canonical video mode number to the kernel
H. Peter Anvin [Thu, 13 Sep 2007 21:14:29 +0000 (14:14 -0700)] 
[x86 setup] Present the canonical video mode number to the kernel

Canonicalize the video mode number as presented to the kernel.  The
video mode number may be user-entered (e.g. ASK_VGA), an alias
(e.g. NORMAL_VGA), or a size specification, and that confuses the
suspend wakeup code.

Signed-off-by: H. Peter Anvin <hpa@zytor.com>
17 years ago[XFS] fix valid but harmless sparse warning
Christoph Hellwig [Wed, 19 Sep 2007 05:27:30 +0000 (15:27 +1000)] 
[XFS] fix valid but harmless sparse warning

The new xlog_recover_do_reg_buffer checks call be16_to_cpu on di_gen which
is a 32bit value so sparse rightly complains. Fortunately the warning is
harmless because we don't care for the value, but only whether it's
non-NULL. Due to that fact we can simply kill the endian swaps on this and
the previous di_mode check entirely.

SGI-PV: 969656
SGI-Modid: xfs-linux-melb:xfs-kern:29709a

Signed-off-by: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Lachlan McIlroy <lachlan@sgi.com>
Signed-off-by: Tim Shimmin <tes@sgi.com>
17 years ago[XFS] fix filestreams on 32-bit boxes
Eric Sandeen [Thu, 30 Aug 2007 07:21:38 +0000 (17:21 +1000)] 
[XFS] fix filestreams on 32-bit boxes

xfs_filestream_mount() sets up an mru cache with:
  err = xfs_mru_cache_create(&mp->m_filestream, lifetime, grp_count,
  (xfs_mru_cache_free_func_t)xfs_fstrm_free_func);
but that cast is causing problems...
  typedef void (*xfs_mru_cache_free_func_t)(unsigned long, void*);
but:
  void xfs_fstrm_free_func( xfs_ino_t ino, fstrm_item_t *item)
so on a 32-bit box, it's casting (32, 32) args into (64, 32) and I assume
it's getting garbage for *item, which subsequently causes an explosion.
With this change the filestreams xfsqa tests don't oops on my 32-bit box.

SGI-PV: 967795
SGI-Modid: xfs-linux-melb:xfs-kern:29510a

Signed-off-by: Eric Sandeen <sandeen@sandeen.net>
Signed-off-by: David Chinner <dgc@sgi.com>
Signed-off-by: Tim Shimmin <tes@sgi.com>
17 years ago[ARM] 4569/1: ep93xx_gpio_irq_type(): fix spurious enumeration offset for FGPIO handling
Herbert Valerio Riedel [Sun, 16 Sep 2007 15:44:39 +0000 (16:44 +0100)] 
[ARM] 4569/1: ep93xx_gpio_irq_type(): fix spurious enumeration offset for FGPIO handling

The EP93XX_GPIO_LINE_F() macro is supposed to be called with a line
number between 0 and 7, but the current code causes it to get called
with an spuriously offset number range {16..23}.

Signed-off-by: Herbert Valerio Riedel <hvr@gnu.org>
Signed-off-by: Lennert Buytenhek <kernel@wantstofly.org>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
17 years agophy: export phy_mii_ioctl
Domen Puncer [Mon, 17 Sep 2007 20:21:40 +0000 (22:21 +0200)] 
phy: export phy_mii_ioctl

Export phy_mii_ioctl, so network drivers can use it when built
as modules too.

Signed-off-by: Domen Puncer <domen@coderock.org>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
17 years agoLinux 2.6.23-rc7 v2.6.23-rc7
Linus Torvalds [Wed, 19 Sep 2007 23:01:13 +0000 (16:01 -0700)] 
Linux 2.6.23-rc7

17 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/mingo/linux-2.6-sched
Linus Torvalds [Wed, 19 Sep 2007 22:47:59 +0000 (15:47 -0700)] 
Merge git://git./linux/kernel/git/mingo/linux-2.6-sched

* git://git.kernel.org/pub/scm/linux/kernel/git/mingo/linux-2.6-sched:
  sched: fix invalid sched_class use
  sched: add /proc/sys/kernel/sched_compat_yield

17 years agoSELinux: fix array out of bounds when mounting with selinux options
Eric Paris [Wed, 19 Sep 2007 21:19:12 +0000 (17:19 -0400)] 
SELinux: fix array out of bounds when mounting with selinux options

Given an illegal selinux option it was possible for match_token to work in
random memory at the end of the match_table_t array.

Note that privilege is required to perform a context mount, so this issue is
effectively limited to root only.

Signed-off-by: Eric Paris <eparis@redhat.com>
Acked-by: Stephen Smalley <sds@tycho.nsa.gov>
Signed-off-by: James Morris <jmorris@namei.org>
17 years agosched: fix invalid sched_class use
Hiroshi Shimamoto [Wed, 19 Sep 2007 21:34:46 +0000 (23:34 +0200)] 
sched: fix invalid sched_class use

When using rt_mutex, a NULL pointer dereference is occurred at
enqueue_task_rt. Here is a scenario;
1) there are two threads, the thread A is fair_sched_class and
   thread B is rt_sched_class.
2) Thread A is boosted up to rt_sched_class, because the thread A
   has a rt_mutex lock and the thread B is waiting the lock.
3) At this time, when thread A create a new thread C, the thread
   C has a rt_sched_class.
4) When doing wake_up_new_task() for the thread C, the priority
   of the thread C is out of the RT priority range, because the
   normal priority of thread A is not the RT priority. It makes
   data corruption by overflowing the rt_prio_array.
The new thread C should be fair_sched_class.

The new thread should be valid scheduler class before queuing.
This patch fixes to set the suitable scheduler class.

Signed-off-by: Hiroshi Shimamoto <h-shimamoto@ct.jp.nec.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
17 years agosched: add /proc/sys/kernel/sched_compat_yield
Ingo Molnar [Wed, 19 Sep 2007 21:34:46 +0000 (23:34 +0200)] 
sched: add /proc/sys/kernel/sched_compat_yield

add /proc/sys/kernel/sched_compat_yield to make sys_sched_yield()
more agressive, by moving the yielding task to the last position
in the rbtree.

with sched_compat_yield=0:

   PID USER      PR  NI  VIRT  RES  SHR S %CPU %MEM    TIME+  COMMAND
  2539 mingo     20   0  1576  252  204 R   50  0.0   0:02.03 loop_yield
  2541 mingo     20   0  1576  244  196 R   50  0.0   0:02.05 loop

with sched_compat_yield=1:

   PID USER      PR  NI  VIRT  RES  SHR S %CPU %MEM    TIME+  COMMAND
  2584 mingo     20   0  1576  248  196 R   99  0.0   0:52.45 loop
  2582 mingo     20   0  1576  256  204 R    0  0.0   0:00.00 loop_yield

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
17 years agomyri10ge: Add support for PCI device id 9
Brice Goglin [Thu, 13 Sep 2007 22:40:14 +0000 (00:40 +0200)] 
myri10ge: Add support for PCI device id 9

Add support for new Myri-10G boards with PCI device id 9.

Signed-off-by: Brice Goglin <brice@myri.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
17 years agoMerge branch 'upstream' of git://ftp.linux-mips.org/pub/scm/upstream-linus
Linus Torvalds [Wed, 19 Sep 2007 18:45:32 +0000 (11:45 -0700)] 
Merge branch 'upstream' of git://ftp.linux-mips.org/upstream-linus

* 'upstream' of git://ftp.linux-mips.org/pub/scm/upstream-linus:
  [MIPS] cpu-bugs64.c: GCC 3.3 constraint workaround
  [MIPS] DEC: Initialise ioasic_ssr_lock

17 years agoMerge master.kernel.org:/pub/scm/linux/kernel/git/mchehab/v4l-dvb
Linus Torvalds [Wed, 19 Sep 2007 18:41:15 +0000 (11:41 -0700)] 
Merge /pub/scm/linux/kernel/git/mchehab/v4l-dvb

* master.kernel.org:/pub/scm/linux/kernel/git/mchehab/v4l-dvb:
  V4L/DVB (6173a): Documentation: Remove reference to dead "cpia_pp=" boot-time option
  Revert "V4L/DVB (6173a): Documentation: Remove reference to dead "cpia_pp=" boot-time option"

17 years agoMerge branch 'for-linus' of git://oss.sgi.com:8090/xfs/xfs-2.6
Linus Torvalds [Wed, 19 Sep 2007 18:40:13 +0000 (11:40 -0700)] 
Merge branch 'for-linus' of git://oss.sgi.com:8090/xfs/xfs-2.6

* 'for-linus' of git://oss.sgi.com:8090/xfs/xfs-2.6:
  [XFS] Avoid replaying inode buffer initialisation log items if on-disk version is newer.
  [XFS] Ensure file size updates have been completed before writing inode to disk.
  [XFS] On-demand reaping of the MRU cache

17 years agoMerge master.kernel.org:/pub/scm/linux/kernel/git/davem/sparc-2.6
Linus Torvalds [Wed, 19 Sep 2007 18:39:39 +0000 (11:39 -0700)] 
Merge /pub/scm/linux/kernel/git/davem/sparc-2.6

* master.kernel.org:/pub/scm/linux/kernel/git/davem/sparc-2.6:
  [SUNSAB]: Fix several bugs.

17 years agoMerge master.kernel.org:/pub/scm/linux/kernel/git/bart/ide-2.6
Linus Torvalds [Wed, 19 Sep 2007 18:39:10 +0000 (11:39 -0700)] 
Merge /pub/scm/linux/kernel/git/bart/ide-2.6

* master.kernel.org:/pub/scm/linux/kernel/git/bart/ide-2.6:
  ide: remove unused variables from drivers/ide/ppc/pmac.c
  ide: ST320413A has the same problem as ST340823A

17 years agoMerge branch 'merge' of git://git.kernel.org/pub/scm/linux/kernel/git/paulus/powerpc
Linus Torvalds [Wed, 19 Sep 2007 18:38:25 +0000 (11:38 -0700)] 
Merge branch 'merge' of git://git./linux/kernel/git/paulus/powerpc

* 'merge' of git://git.kernel.org/pub/scm/linux/kernel/git/paulus/powerpc:
  [POWERPC] Fix timekeeping on PowerPC 601
  [POWERPC] Don't expose clock vDSO functions when CPU has no timebase
  [POWERPC] spusched: Fix null pointer dereference in find_victim

17 years agox86-64: page faults from user mode are always user faults
Linus Torvalds [Wed, 19 Sep 2007 18:37:14 +0000 (11:37 -0700)] 
x86-64: page faults from user mode are always user faults

Randy Dunlap noticed an interesting "crashme" behaviour on his dual
Prescott Xeon setup, where he gets page faults with the error code
having a zero "user" bit, but the register state points back to user
mode.

This may be a CPU microcode buglet triggered by some strange instruction
pattern that crashme generates, and loading a microcode update seems to
possibly have fixed it.

Regardless, we really should trust the register state more than the
error code, since it's really the register state that determines whether
we can actually send a signal, or whether we're in kernel mode and need
to oops/kill the process in the case of a page fault.

Cc: Randy Dunlap <rdunlap@xenotime.net>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years ago[MIPS] cpu-bugs64.c: GCC 3.3 constraint workaround
Maciej W. Rozycki [Mon, 17 Sep 2007 16:11:07 +0000 (17:11 +0100)] 
[MIPS] cpu-bugs64.c: GCC 3.3 constraint workaround

Add a workaround to address warnings generated on the "n" constraint by
GCC 3.3 and below.

Signed-off-by: Maciej W. Rozycki <macro@linux-mips.org>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
17 years ago[MIPS] DEC: Initialise ioasic_ssr_lock
Maciej W. Rozycki [Mon, 17 Sep 2007 15:58:18 +0000 (16:58 +0100)] 
[MIPS] DEC: Initialise ioasic_ssr_lock

Fix the definition of the ioasic_ssr_lock spinlock to include a proper
initialisation.

Signed-off-by: Maciej W. Rozycki <macro@linux-mips.org>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
17 years agoDriver core: fix deprectated sysfs structure for nested class devices
Dmitry Torokhov [Wed, 19 Sep 2007 05:46:50 +0000 (22:46 -0700)] 
Driver core: fix deprectated sysfs structure for nested class devices

Nested class devices used to have 'device' symlink point to a real
(physical) device instead of a parent class device.  When converting
subsystems to struct device we need to keep doing what class devices did if
CONFIG_SYSFS_DEPRECATED is Y, otherwise parts of udev break.

Signed-off-by: Dmitry Torokhov <dtor@mail.ru>
Cc: Kay Sievers <kay.sievers@vrfy.org>
Acked-by: Greg KH <greg@kroah.com>
Tested-by: Anssi Hannula <anssi.hannula@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agouml: fix irqstack crash
Jeff Dike [Wed, 19 Sep 2007 05:46:49 +0000 (22:46 -0700)] 
uml: fix irqstack crash

This patch fixes a crash caused by an interrupt coming in when an IRQ stack
is being torn down.  When this happens, handle_signal will loop, setting up
the IRQ stack again because the tearing down had finished, and handling
whatever signals had come in.

However, to_irq_stack returns a mask of pending signals to be handled, plus
bit zero is set if the IRQ stack was already active, and thus shouldn't be
torn down.  This causes a problem because when handle_signal goes around
the loop, sig will be zero, and to_irq_stack will duly set bit zero in the
returned mask, faking handle_signal into believing that it shouldn't tear
down the IRQ stack and return thread_info pointers back to their original
values.

This will eventually cause a crash, as the IRQ stack thread_info will
continue pointing to the original task_struct and an interrupt will look
into it after it has been freed.

The fix is to stop passing a signal number into to_irq_stack.  Rather, the
pending signals mask is initialized beforehand with the bit for sig already
set.  References to sig in to_irq_stack can be replaced with references to
the mask.

[akpm@linux-foundation.org: use UL]
Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoFix NUMA Memory Policy Reference Counting
Lee Schermerhorn [Wed, 19 Sep 2007 05:46:47 +0000 (22:46 -0700)] 
Fix NUMA Memory Policy Reference Counting

This patch proposes fixes to the reference counting of memory policy in the
page allocation paths and in show_numa_map().  Extracted from my "Memory
Policy Cleanups and Enhancements" series as stand-alone.

Shared policy lookup [shmem] has always added a reference to the policy,
but this was never unrefed after page allocation or after formatting the
numa map data.

Default system policy should not require additional ref counting, nor
should the current task's task policy.  However, show_numa_map() calls
get_vma_policy() to examine what may be [likely is] another task's policy.
The latter case needs protection against freeing of the policy.

This patch adds a reference count to a mempolicy returned by
get_vma_policy() when the policy is a vma policy or another task's
mempolicy.  Again, shared policy is already reference counted on lookup.  A
matching "unref" [__mpol_free()] is performed in alloc_page_vma() for
shared and vma policies, and in show_numa_map() for shared and another
task's mempolicy.  We can call __mpol_free() directly, saving an admittedly
inexpensive inline NULL test, because we know we have a non-NULL policy.

Handling policy ref counts for hugepages is a bit trickier.
huge_zonelist() returns a zone list that might come from a shared or vma
'BIND policy.  In this case, we should hold the reference until after the
huge page allocation in dequeue_hugepage().  The patch modifies
huge_zonelist() to return a pointer to the mempolicy if it needs to be
unref'd after allocation.

Kernel Build [16cpu, 32GB, ia64] - average of 10 runs:

w/o patch w/ refcount patch
    Avg   Std Devn    Avg   Std Devn
Real:  100.59     0.38  100.63     0.43
User: 1209.60     0.37 1209.91     0.31
System:   81.52     0.42   81.64     0.34

Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Acked-by: Andi Kleen <ak@suse.de>
Cc: Christoph Lameter <clameter@sgi.com>
Acked-by: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoFix user namespace exiting OOPs
Pavel Emelyanov [Wed, 19 Sep 2007 05:46:45 +0000 (22:46 -0700)] 
Fix user namespace exiting OOPs

It turned out, that the user namespace is released during the do_exit() in
exit_task_namespaces(), but the struct user_struct is released only during the
put_task_struct(), i.e.  MUCH later.

On debug kernels with poisoned slabs this will cause the oops in
uid_hash_remove() because the head of the chain, which resides inside the
struct user_namespace, will be already freed and poisoned.

Since the uid hash itself is required only when someone can search it, i.e.
when the namespace is alive, we can safely unhash all the user_struct-s from
it during the namespace exiting.  The subsequent free_uid() will complete the
user_struct destruction.

For example simple program

   #include <sched.h>

   char stack[2 * 1024 * 1024];

   int f(void *foo)
   {
    return 0;
   }

   int main(void)
   {
    clone(f, stack + 1 * 1024 * 1024, 0x10000000, 0);
    return 0;
   }

run on kernel with CONFIG_USER_NS turned on will oops the
kernel immediately.

This was spotted during OpenVZ kernel testing.

Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Alexey Dobriyan <adobriyan@openvz.org>
Acked-by: "Serge E. Hallyn" <serue@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoConvert uid hash to hlist
Pavel Emelyanov [Wed, 19 Sep 2007 05:46:44 +0000 (22:46 -0700)] 
Convert uid hash to hlist

Surprisingly, but (spotted by Alexey Dobriyan) the uid hash still uses
list_heads, thus occupying twice as much place as it could.  Convert it to
hlist_heads.

Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Alexey Dobriyan <adobriyan@openvz.org>
Acked-by: Serge Hallyn <serue@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agokernel/user.c: Use list_for_each_entry instead of list_for_each
Matthias Kaehlcke [Wed, 19 Sep 2007 05:46:43 +0000 (22:46 -0700)] 
kernel/user.c: Use list_for_each_entry instead of list_for_each

kernel/user.c: Convert list_for_each to list_for_each_entry in
uid_hash_find()

Signed-off-by: Matthias Kaehlcke <matthias.kaehlcke@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoext34: ensure do_split leaves enough free space in both blocks
Eric Sandeen [Wed, 19 Sep 2007 05:46:42 +0000 (22:46 -0700)] 
ext34: ensure do_split leaves enough free space in both blocks

The do_split() function for htree dir blocks is intended to split a leaf
block to make room for a new entry.  It sorts the entries in the original
block by hash value, then moves the last half of the entries to the new
block - without accounting for how much space this actually moves.  (IOW,
it moves half of the entry *count* not half of the entry *space*).  If by
chance we have both large & small entries, and we move only the smallest
entries, and we have a large new entry to insert, we may not have created
enough space for it.

The patch below stores each record size when calculating the dx_map, and
then walks the hash-sorted dx_map, calculating how many entries must be
moved to more evenly split the existing entries between the old block and
the new block, guaranteeing enough space for the new entry.

The dx_map "offs" member is reduced to u16 so that the overall map size
does not change - it is temporarily stored at the end of the new block, and
if it grows too large it may be overwritten.  By making offs and size both
u16, we won't grow the map size.

Also add a few comments to the functions involved.

This fixes the testcase reported by hooanon05@yahoo.co.jp on the
linux-ext4 list, "ext3 dir_index causes an error"

Thanks to Andreas Dilger for discussing the problem & solution with me.

Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Andreas Dilger <adilger@clusterfs.com>
Tested-by: Junjiro Okajima <hooanon05@yahoo.co.jp>
Cc: Theodore Ts'o <tytso@mit.edu>
Cc: <linux-ext4@vger.kernel.org>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agodisable sys_timerfd() for 2.6.23
Andrew Morton [Wed, 19 Sep 2007 05:46:41 +0000 (22:46 -0700)] 
disable sys_timerfd() for 2.6.23

There is still some confusion and disagreement over what this interface should
actually do.  So it is best that we disable it in 2.6.23 until we get that
fully sorted out.

(sys_timerfd() was present in 2.6.22 but it was apparently broken, so here we
assume that nobody is using it yet).

Cc: Michael Kerrisk <mtk-manpages@gmx.net>
Cc: Davide Libenzi <davidel@xmailserver.org>
Acked-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agonfs: fix oops re sysctls and V4 support
Alexey Dobriyan [Wed, 19 Sep 2007 05:46:40 +0000 (22:46 -0700)] 
nfs: fix oops re sysctls and V4 support

NFS unregisters sysctls only if V4 support is compiled in.  However, sysctl
table is not V4 specific, so unregister it always.

Steps to reproduce:

[build nfs.ko with CONFIG_NFS_V4=n]
modrobe nfs
rmmod nfs
ls /proc/sys

Unable to handle kernel paging request at ffffffff880661c0 RIP:
 [<ffffffff802af8e3>] proc_sys_readdir+0xd3/0x350
PGD 203067 PUD 207063 PMD 7e216067 PTE 0
Oops: 0000 [1] SMP
CPU 1
Modules linked in: lockd nfs_acl sunrpc
Pid: 3335, comm: ls Not tainted 2.6.23-rc3-bloat #2
RIP: 0010:[<ffffffff802af8e3>]  [<ffffffff802af8e3>] proc_sys_readdir+0xd3/0x350
RSP: 0018:ffff81007fd93e78  EFLAGS: 00010286
RAX: ffffffff880661c0 RBX: ffffffff80466370 RCX: ffffffff880661c0
RDX: 00000000000014c0 RSI: ffff81007f3ad020 RDI: ffff81007efd8b40
RBP: 0000000000000018 R08: 0000000000000000 R09: 0000000000000000
R10: 0000000000000001 R11: ffffffff802a8570 R12: ffffffff880661c0
R13: ffff81007e219640 R14: ffff81007efd8b40 R15: ffff81007ded7280
FS:  00002ba25ef03060(0000) GS:ffff81007ff81258(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 000000008005003b
CR2: ffffffff880661c0 CR3: 000000007dfaf000 CR4: 00000000000006e0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400
Process ls (pid: 3335, threadinfo ffff81007fd92000, task ffff81007d8a0000)
Stack:  ffff81007f3ad150 ffffffff80283f30 ffff81007fd93f48 ffff81007efd8b40
 ffff81007ee00440 0000000422222222 0000000200035593 ffffffff88037e9a
 2222222222222222 ffffffff80466500 ffff81007e416400 ffff81007e219640
Call Trace:
 [<ffffffff80283f30>] filldir+0x0/0xf0
 [<ffffffff80283f30>] filldir+0x0/0xf0
 [<ffffffff802840c7>] vfs_readdir+0xa7/0xc0
 [<ffffffff80284376>] sys_getdents+0x96/0xe0
 [<ffffffff8020bb3e>] system_call+0x7e/0x83

Code: 41 8b 14 24 85 d2 74 dc 49 8b 44 24 08 48 85 c0 74 e7 49 3b
RIP  [<ffffffff802af8e3>] proc_sys_readdir+0xd3/0x350
 RSP <ffff81007fd93e78>
CR2: ffffffff880661c0
Kernel panic - not syncing: Fatal exception

Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Acked-by: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: "J. Bruce Fields" <bfields@fieldses.org>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agodir_index: error out instead of BUG on corrupt dx dirs
Eric Sandeen [Wed, 19 Sep 2007 05:46:38 +0000 (22:46 -0700)] 
dir_index: error out instead of BUG on corrupt dx dirs

Convert asserts (BUGs) in dx_probe from bad on-disk data to recoverable
errors with helpful warnings.  With help catching other asserts from Duane
Griffin <duaneg@dghda.com>

Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Acked-by: Duane Griffin <duaneg@dghda.com>
Acked-by: Theodore Ts'o <tytso@mit.edu>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agointel-agp: Fix i830 mask variable that changed with G33 support
Dave Airlie [Wed, 19 Sep 2007 05:46:35 +0000 (22:46 -0700)] 
intel-agp: Fix i830 mask variable that changed with G33 support

The mask on i830 should be 0x70 always, later chips 0xF0 should be okay.

Signed-off-by: Dave Airlie <airlied@linux.ie>
Acked-by: Zhenyu Wang <zhenyu.z.wang@intel.com>
Cc: Michael Haas <laga@laga.ath.cx>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agointelfb: Fix bug in DPLL disable
Antonino A. Daplas [Wed, 19 Sep 2007 05:46:34 +0000 (22:46 -0700)] 
intelfb: Fix bug in DPLL disable

Reported in Kernel Bugzilla 9006

Fix an obvious bug in DPLL disable.

Signed-off-by: Antonino Daplas <adaplas@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoxen: don't bother trying to set cr4
Jeremy Fitzhardinge [Wed, 19 Sep 2007 05:46:33 +0000 (22:46 -0700)] 
xen: don't bother trying to set cr4

Xen ignores all updates to cr4, and some versions will kill the domain if
you try to change its value.  Just ignore all changes.

Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agopci: fix unterminated pci_device_id lists
Kees Cook [Wed, 19 Sep 2007 05:46:32 +0000 (22:46 -0700)] 
pci: fix unterminated pci_device_id lists

Fix a couple drivers that do not correctly terminate their pci_device_id
lists.  This results in garbage being spewed into modules.pcimap when the
module happens to not have 28 NULL bytes following the table, and/or the
last PCI ID is actually truncated from the table when calculating the
modules.alias PCI aliases, cause those unfortunate device IDs to not
auto-load.

Signed-off-by: Kees Cook <kees@ubuntu.com>
Acked-by: Corey Minyard <minyard@acm.org>
Cc: David Woodhouse <dwmw2@infradead.org>
Acked-by: Jeff Garzik <jeff@garzik.org>
Cc: Greg KH <greg@kroah.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agomspec: handle shrinking virtual memory areas
Cliff Wickman [Wed, 19 Sep 2007 05:46:31 +0000 (22:46 -0700)] 
mspec: handle shrinking virtual memory areas

The shrinking of a virtual memory area that is mmap(2)'d to a memory
special file (device drivers/char/mspec.c) can cause a panic.

If the mapped size of the vma (vm_area_struct) is very large, mspec allocates
a large vma_data structure with vmalloc(). But such a vma can be shrunk by
an munmap(2).  The current driver uses the current size of each vma to
deduce whether its vma_data structure was allocated by kmalloc() or vmalloc().
So if the vma was shrunk it appears to have been allocated by kmalloc(),
and mspec attempts to free it with kfree().  This results in a panic.

This patch avoids the panic (by preserving the type of the allocation) and
also makes mspec work correctly as the vma is split into pieces by the
munmap(2)'s.

All vma's derived from such a split vma share the same vma_data structure that
represents all the pages mapped into this set of vma's.  The mpec driver
must be made capable of using the right portion of the structure for each
member vma.  In other words, it must index into the array of page addresses
using the portion of the array that represents the current vma. This is
enabled by storing the vma group's vm_start in the vma_data structure.

The shared vma_data's are not protected by mm->mmap_sem in the fork() case
so the reference count is left as atomic_t.

Signed-off-by: Cliff Wickman <cpw@sgi.com>
Acked-by: Jes Sorensen <jes@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agortc: rtc-ds1553.c should use resource_size_t for base address
Atsushi Nemoto [Wed, 19 Sep 2007 05:46:30 +0000 (22:46 -0700)] 
rtc: rtc-ds1553.c should use resource_size_t for base address

Currently the rtc driver, rtc-ds1552.c uses an unsigned long to store the
base mmio address of the NVRAM/RTC.  This breaks on 32-bit systems with
larger physical addresses.

Signed-off-by: Atsushi Nemoto <anemo@mba.ocn.ne.jp>
Cc: David Brownell <david-b@pacbell.net>
Cc: Alessandro Zummo <a.zummo@towertech.it>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agortc-ds1742.c should use resource_size_t for base address
David Gibson [Wed, 19 Sep 2007 05:46:28 +0000 (22:46 -0700)] 
rtc-ds1742.c should use resource_size_t for base address

Currently the rtc driver, rtc-ds1742.c uses an unsigned long to store the
base mmio address of the NVRAM/RTC.  This breaks on systems like PowerPC
440, which is a 32-bit core with 36-bit physical addresses: IO on the
system, including the RTC, is typically above the 4GB point, and cannot fit
into an unsigned long.

This patch fixes the problem by replacing the unsigned long with a
resource_size_t.  Tested on Ebony (PPC440) (with additional patches to
instantiate the ds1742 platform device appropriately).

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Acked-by: Atsushi Nemoto <anemo@mba.ocn.ne.jp>
Cc: Alessandro Zummo <a.zummo@towertech.it>
Cc: David Brownell <david-b@pacbell.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoFix UTS corruption during clone(CLONE_NEWUTS)
Alexey Dobriyan [Wed, 19 Sep 2007 05:46:27 +0000 (22:46 -0700)] 
Fix UTS corruption during clone(CLONE_NEWUTS)

struct utsname is copied from master one without any exclusion.

Here is sample output from one proggie doing

sethostname("aaaaaaaaaaaaaaaaaaaaaaaaaaaaaa");
sethostname("bbbbbbbbbbbbbbbbbbbbbbbbbbbbbb");

and another

clone(,, CLONE_NEWUTS, ...)
uname()

hostname = 'aaaaaaaaaaaaaaaaaaaaaaaaabbbbb'
hostname = 'bbbaaaaaaaaaaaaaaaaaaaaaaaaaaa'
hostname = 'aaaaaaaabbbbbbbbbbbbbbbbbbbbbb'
hostname = 'aaaaaaaaaaaaaaaaaaaaaaaaaabbbb'
hostname = 'aaaaaaaaaaaaaaaaaaaaaaaaaaaabb'
hostname = 'aaabbbbbbbbbbbbbbbbbbbbbbbbbbb'
hostname = 'bbbbbbbbbbbbbbbbaaaaaaaaaaaaaa'

Hostname is sometimes corrupted.

Yes, even _the_ simplest namespace activity had bug in it. :-(

Signed-off-by: Alexey Dobriyan <adobriyan@sw.ru>
Acked-by: Serge Hallyn <serue@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoFix failure to resume from initrds
Nigel Cunningham [Wed, 19 Sep 2007 05:46:25 +0000 (22:46 -0700)] 
Fix failure to resume from initrds

Commit 831441862956fffa17b9801db37e6ea1650b0f69 (Freezer: make kernel
threads nonfreezable by default) breaks freezing when attempting to resume
from an initrd, because the init (which is freezeable) spins while waiting
for another thread to run /linuxrc, but doesn't check whether it has been
told to enter the refrigerator.  The original patch replaced a call to
try_to_freeze() with a call to yield().  I believe a simple reversion is
wrong because if !CONFIG_PM_SLEEP, try_to_freeze() is a noop.  It should
still yield.

Signed-off-by: Nigel Cunningham <nigel@nigel.suspend2.net>
Acked-by: Rafael J. Wysocki <rjw@sisk.pl>
Acked-by: Pavel Machek <pavel@ucw.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agouml: use correct type in BLKGETSIZE ioctl
Nicolas George [Wed, 19 Sep 2007 05:46:21 +0000 (22:46 -0700)] 
uml: use correct type in BLKGETSIZE ioctl

I found a type mismatch in UML that makes host block devices unusable as ubd
devices on x86_64 and other 64 bits systems (segfault of the mm subsystem):

In block/ioctl.c, the following lines show that the BLKGETSIZE ioctl expects
a pointer to a long:

case BLKGETSIZE:
if ((bdev->bd_inode->i_size >> 9) > ~0UL)
return -EFBIG;
return put_ulong(arg, bdev->bd_inode->i_size >> 9);

In arch/um/os-Linux/file.c, os_file_size calls it with an int.

The ioctl_list man page should be fixed as well.

Cc: Jeff Dike <jdike@addtoit.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoFix "Fix DAC960 driver on machines which don't support 64-bit DMA"
Andrew Morton [Wed, 19 Sep 2007 05:46:19 +0000 (22:46 -0700)] 
Fix "Fix DAC960 driver on machines which don't support 64-bit DMA"

sparc32:

drivers/block/DAC960.c: In function 'DAC960_V1_EnableMemoryMailboxInterface':
drivers/block/DAC960.c:1168: error: 'DMA_32BIT_MASK' undeclared (first use in this function)
drivers/block/DAC960.c:1168: error: (Each undeclared identifier is reported only

Cc: <dac@conglom-o.org>
Cc: <stable@kernel.org>
Cc: Alessandro Polverini <alex@nibbles.it>
Cc: Jeff Garzik <jeff@garzik.org>
Cc: Matthew Wilcox <matthew@wil.cx>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>