* 'drm-intel-next' of git://git.kernel.org/pub/scm/linux/kernel/git/anholt/drm-intel: (22 commits)
  drm/i915: Fix for LVDS VBT change on IGDNG
  drm/i915: Zap the GTT mapping when transitioning from untiled to tiled.
  drm/i915: Refactor calls to unmap_mapping_range
  drm/i915: Avoid saving/restore the modesetting registers twice in KMS mode
  drm: Disable the unused connectors explicitly when resuming with KMS.
  drm/i915: Restore the KMS modeset for every activated CRTC
  drm/i915: Fix harmless warning from patch merged after i2c rework.
  drm/i915: Disable GEM when a broken video BIOS takes up the whole aperture.
  drm/i915: Check the LID device to decide whether the LVDS should be initialized
  drm/i915: Move lock to more reasonable location
  drm/i915: Add gtt_offset to gem object list debugfs output
  drm/i915: Remove gtt_bound from drm_i915_gem_object
  drm/i915: Disable VGA output when doing DRM_MODE_DPMS_OFF.
  drm/i915: crt fetch EDID by DVI-I converter on G4x platform
  drm/i915: Don't update display FIFO watermark on IGDNG
  drm/i915: Adjust DisplayPort clocks to use 96MHz reference
  drm/i915: Make driver less chatty
  drm/i915: fix up a raw 64bit divide
  drm/i915: enable sdvo lvds scaling function.
  drm/i915: Set SSC frequency for 8xx chips correctly
  ...

exception.txt only explains the code on x86, so it's better to
move it into Documentation/x86 directory.

And also rename it to exception-tables.txt which looks much
more reasonable.

This patch is on top of the previous one.
Signed-off-by: NWANG Cong <amwang@redhat.com>
Signed-off-by: NRandy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Update Documentation/exception.txt.
Remove trailing whitespaces in it.
Signed-off-by: NWANG Cong <amwang@redhat.com>
Signed-off-by: NRandy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

When building v2.6.31-rc2-344-g69ca06c9, the following build errors are
found due to missing includes:

 CC [M]  fs/fuse/dev.o
fs/fuse/dev.c: In function ‘request_end’:
fs/fuse/dev.c:289: error: ‘BLK_RW_SYNC’ undeclared (first use in this function)
...
fs/nfs/write.c: In function ‘nfs_set_page_writeback’:
fs/nfs/write.c:207: error: ‘BLK_RW_ASYNC’ undeclared (first use in this function)

Signed-off-by: Larry Finger@lwfinger.net>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

* 'for-linus' of git://git.kernel.dk/linux-2.6-block:
  cfq-iosched: reset oom_cfqq in cfq_set_request()
  block: fix sg SG_DXFER_TO_FROM_DEV regression
  block: call blk_scsi_ioctl_init()
  Fix congestion_wait() sync/async vs read/write confusion

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ryusuke/nilfs2:
  nilfs2: fix disorder in cp count on error during deleting checkpoints
  nilfs2: fix lockdep warning between regular file and inode file
  nilfs2: fix incorrect KERN_CRIT messages in case of write failures
  nilfs2: fix hang problem of log writer which occurs after write failures
  nilfs2: remove unlikely directive causing mis-conversion of error code

Merge branch 'core-fixes-for-linus-2' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip

* 'core-fixes-for-linus-2' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
  dma-debug: Fix the overlap() function to be correct and readable
  oprofile: reset bt_lost_no_mapping with other stats
  x86/oprofile: rename kernel parameter for architectural perfmon to arch_perfmon
  signals: declare sys_rt_tgsigqueueinfo in syscalls.h
  rcu: Mark Hierarchical RCU no longer experimental
  dma-debug: Put all hash-chain locks into the same lock class
  dma-debug: fix off-by-one error in overlap function

Merge branch 'perfcounters-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip

* 'perfcounters-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip: (50 commits)
  perf report: Add "Fractal" mode output - support callchains with relative overhead rate
  perf_counter tools: callchains: Manage the cumul hits on the fly
  perf report: Change default callchain parameters
  perf report: Use a modifiable string for default callchain options
  perf report: Warn on callchain output request from non-callchain file
  x86: atomic64: Inline atomic64_read() again
  x86: atomic64: Clean up atomic64_sub_and_test() and atomic64_add_negative()
  x86: atomic64: Improve atomic64_xchg()
  x86: atomic64: Export APIs to modules
  x86: atomic64: Improve atomic64_read()
  x86: atomic64: Code atomic(64)_read and atomic(64)_set in C not CPP
  x86: atomic64: Fix unclean type use in atomic64_xchg()
  x86: atomic64: Make atomic_read() type-safe
  x86: atomic64: Reduce size of functions
  x86: atomic64: Improve atomic64_add_return()
  x86: atomic64: Improve cmpxchg8b()
  x86: atomic64: Improve atomic64_read()
  x86: atomic64: Move the 32-bit atomic64_t implementation to a .c file
  x86: atomic64: The atomic64_t data type should be 8 bytes aligned on 32-bit too
  perf report: Annotate variable initialization
  ...

Optimize cond_resched() by removing one conditional.

Currently cond_resched() checks system_state ==
SYSTEM_RUNNING in order to avoid scheduling before the
scheduler is running.

We can however, as per suggestion of Matt, use
PREEMPT_ACTIVE to accomplish that very same.
Suggested-by: NMatt Mackall <mpm@selenic.com>
Signed-off-by: NPeter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: NMatt Mackall <mpm@selenic.com>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Pull the initial preempt_count value into a single
definition site.

Maintainers for: alpha, ia64 and m68k, please have a look,
your arch code is funny.

The header magic is a bit odd, but similar to the KERNEL_DS
one, CPP waits with expanding these macros until the
INIT_THREAD_INFO macro itself is expanded, which is in
arch/*/kernel/init_task.c where we've already included
sched.h so we're good.

Cc: tony.luck@intel.com
Cc: rth@twiddle.net
Cc: geert@linux-m68k.org
Signed-off-by: NPeter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: NMatt Mackall <mpm@selenic.com>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

IGDNG mobile chip's LVDS data block removes panel fitting
register definition. So this fixes offset for LVDS timing
block parsing. Thanks for Michael Fu to catch this.
Signed-off-by: NZhenyu Wang <zhenyu.z.wang@intel.com>
Signed-off-by: NEric Anholt <eric@anholt.net>

As of 52dc7d32, we could leave an old
linear GTT mapping in place, so that apps trying to GTT-mapped write in
tiled data wouldn't get the fence added, and garbage would get displayed.
Signed-off-by: NEric Anholt <eric@anholt.net>

As we call unmap_mapping_range() twice in identical fashion, refactor
and attempt to explain why we need to call unmap_mapping_range().
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: NEric Anholt <eric@anholt.net>

Linus noticed how unclean and buggy the overlap() function is:

 - It uses convoluted (and bug-causing) positive checks for
   range overlap - instead of using a more natural negative
   check.

 - Even the positive checks are buggy: a positive intersection
   check has four natural cases while we checked only for three,
   missing the (addr < start && addr2 == end) case for example.

 - The variables are mis-named, making it non-obvious how the
   check was done.

 - It needlessly uses u64 instead of unsigned long. Since these
   are kernel memory pointers and we explicitly exclude highmem
   ranges anyway we cannot ever overflow 32 bits, even if we
   could. (and on 64-bit it doesnt matter anyway)

All in one, this function needs a total revamp. I used Linus's
suggestions minus the paranoid checks (we cannot overflow really
because if we get totally bad DMA ranges passed far more things
break in the systems than just DMA debugging). I also fixed a
few other small details i noticed.
Reported-by: NLinus Torvalds <torvalds@linux-foundation.org>
Cc: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

In KMS mode we now use the normal mode-setting paths to set the modes
back to the current configuration, so we don't need to also run the more
limited non-KMS implementation of modesetting for resume.
Signed-off-by: NZhao Yakui <yakui.zhao@intel.com>
Acked-by: NJesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: NEric Anholt <eric@anholt.net>

Signed-off-by: NZhao Yakui <yakui.zhao@intel.com>
Acked-by: NJesse Barnes <jbarnes@virtuousgeek.org>
Acked-by: NDave Airlie <airlied@redhat.com>
Signed-off-by: NEric Anholt <eric@anholt.net>

Restore the modeset for every activated CRTC in course of resume.
This is realized by calling the function of drm_helper_resume_force_mode.
Note: it is meaningful only for the KMS mode.

https://bugs.freedesktop.org/show_bug.cgi?id=21719
https://bugs.freedesktop.org/show_bug.cgi?id=21708
https://bugs.freedesktop.org/show_bug.cgi?id=22285
https://bugs.freedesktop.org/show_bug.cgi?id=22263Signed-off-by: NZhao Yakui <yakui.zhao@intel.com>
Signed-off-by: NEric Anholt <eric@anholt.net>

Signed-off-by: NEric Anholt <eric@anholt.net>

This is seen on some G41 systems, where the BIOS will consume all but
a few KB of the aperture.  This should be bad for all operating systems, as
it means that the OS can't dynamically manage memory between graphics and
the rest of the system, and OSes that did static memory management
statically add memory in addition to the BIOS allocation anyway.  So, instead
of working around it, just fail out verbosely.

fd.o bug #21574
Signed-off-by: NEric Anholt <eric@anholt.net>
Reviewed-by: NIan Romanick <ian.d.romanick@intel.com>

Merge branch 'tracing-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip

* 'tracing-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
  tracing: Fix trace_print_seq()
  kprobes: No need to unlock kprobe_insn_mutex
  tracing/fastboot: Document the need of initcall_debug
  trace_export: Repair missed fields
  tracing: Fix stack tracer sysctl handling

In case memory is scarce, we now default to oom_cfqq. Once memory is
available again, we should allocate a new cfqq and stop using oom_cfqq for
a particular io context.

Once a new request comes in, check if we are using oom_cfqq, and if yes,
try to allocate a new cfqq.

Tested the patch by forcing the use of oom_cfqq and upon next request thread
realized that it was using oom_cfqq and it allocated a new cfqq.
Signed-off-by: NVivek Goyal <vgoyal@redhat.com>
Signed-off-by: NJens Axboe <jens.axboe@oracle.com>

I overlooked SG_DXFER_TO_FROM_DEV support when I converted sg to use
the block layer mapping API (2.6.28).

Douglas Gilbert explained SG_DXFER_TO_FROM_DEV:

http://www.spinics.net/lists/linux-scsi/msg37135.html

=
The semantics of SG_DXFER_TO_FROM_DEV were:
   - copy user space buffer to kernel (LLD) buffer
   - do SCSI command which is assumed to be of the DATA_IN
     (data from device) variety. This would overwrite
     some or all of the kernel buffer
   - copy kernel (LLD) buffer back to the user space.

The idea was to detect short reads by filling the original
user space buffer with some marker bytes ("0xec" it would
seem in this report). The "resid" value is a better way
of detecting short reads but that was only added this century
and requires co-operation from the LLD.
=

This patch changes the block layer mapping API to support this
semantics. This simply adds another field to struct rq_map_data and
enables __bio_copy_iov() to copy data from user space even with READ
requests.

It's better to add the flags field and kills null_mapped and the new
from_user fields in struct rq_map_data but that approach makes it
difficult to send this patch to stable trees because st and osst
drivers use struct rq_map_data (they were converted to use the block
layer in 2.6.29 and 2.6.30). Well, I should clean up the block layer
mapping API.

zhou sf reported this regiression and tested this patch:

http://www.spinics.net/lists/linux-scsi/msg37128.html
http://www.spinics.net/lists/linux-scsi/msg37168.htmlReported-by: Nzhou sf <sxzzsf@gmail.com>
Tested-by: Nzhou sf <sxzzsf@gmail.com>
Cc: stable@kernel.org
Signed-off-by: NFUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Signed-off-by: NJens Axboe <jens.axboe@oracle.com>

Currently, blk_scsi_ioctl_init() is not called since it lacks
an initcall marking. This causes the command table to be
unitialized, hence somce commands are block when they should
not have been.

This fixes a regression introduced by commit
018e0446Signed-off-by: NFUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Signed-off-by: NJens Axboe <jens.axboe@oracle.com>

Commit 1faa16d2 accidentally broke
the bdi congestion wait queue logic, causing us to wait on congestion
for WRITE (== 1) when we really wanted BLK_RW_ASYNC (== 0) instead.
Signed-off-by: NJens Axboe <jens.axboe@oracle.com>

The bt_lost_no_mapping is not getting reset at the start of a
profiling run, thus the oprofiled.log shows erroneous values for this
statistic. The attached patch fixes this problem.
Signed-off-by: NMaynard Johnson <maynardj@us.ibm.com>
Signed-off-by: NRobert Richter <robert.richter@amd.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

* git://git.kernel.org/pub/scm/linux/kernel/git/wim/linux-2.6-watchdog:
  [WATCHDOG] w83627hf_wdt.c: add support for the W83627EHF support
  [WATCHDOG] SA1100 watchdog maximum timeout
  [WATCHDOG] w83697ug, fix lock imbalance
  [WATCHDOG] drivers/watchdog/bcm47xx_wdt.c: Remove unnecessary semicolons

Remove call to the mx3fb_set_par() and the mx3fb_blank() before the
register_framebuffer().

This fixes a problem with uninitialized the fb_info->mm_lock mutex
introduced by the commit 537a1bf0 " fbdev: add mutex for fb_mmap
locking"
Signed-off-by: NKrzysztof Helt <krzysztof.h1@wp.pl>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Remove call to the fb_set_par() before the register_framebuffer().

This fixes a problem with uninitialized the fb_info->mm_lock mutex
introduced by the commit 537a1bf0 " fbdev: add mutex for fb_mmap
locking"
Signed-off-by: NKrzysztof Helt <krzysztof.h1@wp.pl>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

* git://git.kernel.org/pub/scm/linux/kernel/git/sfrench/cifs-2.6:
  cifs: when ATTR_READONLY is set, only clear write bits on non-directories
  cifs: remove cifsInodeInfo->inUse counter
  cifs: convert cifs_get_inode_info and non-posix readdir to use cifs_iget
  [CIFS] update cifs version number
  cifs: add and use CIFSSMBUnixSetFileInfo for setattr calls
  cifs: make a separate function for filling out FILE_UNIX_BASIC_INFO
  cifs: rename CIFSSMBUnixSetInfo to CIFSSMBUnixSetPathInfo
  cifs: add pid of initiating process to spnego upcall info
  cifs: fix regression with O_EXCL creates and optimize away lookup
  cifs: add new cifs_iget function and convert unix codepath to use it

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6: (29 commits)
  cxgb3: Fix crash caused by stashing wrong netdev_queue
  ixgbe: Fix coexistence of FCoE and Flow Director in 82599
  memory barrier: adding smp_mb__after_lock
  net: adding memory barrier to the poll and receive callbacks
  netpoll: Fix carrier detection for drivers that are using phylib
  includecheck fix: include/linux, rfkill.h
  p54: tx refused but queue active
  Atheros Kconfig needs to be dependent on WLAN_80211
  mac80211: fix docbook
  mac80211_hwsim: avoid NULL access
  ssb: Add support for 4318E
  b43: Add support for 4318E
  zd1211rw: adding SONY IFU-WLM2 (054c:0257) as a zd1211b device
  zd1211rw: 07b8:6001 is a ZD1211B
  r6040: bump driver version to 0.24 and date to 08 July 2009
  r6040: restore MIER register correctly when IRQ line is shared
  ipv4: Fix fib_trie rebalancing, part 4 (root thresholds)
  davinci_emac: fix kernel oops when changing MAC address while interface is down
  igb: set lan id prior to configuring phy
  mac80211: minstrel: avoid accessing negative indices in rix_to_ndx()
  ...

Looks like the change in ad361c98
wasn't compile tested.
Signed-off-by: NDave Jones <davej@redhat.com>
Acked-by: NDavid S. Miller <davem@davemloft.net>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

The short name of the achitecture is 'arch_perfmon'. This patch
changes the kernel parameter to use this name.

Cc: Andi Kleen <andi@firstfloor.org>
Signed-off-by: NRobert Richter <robert.richter@amd.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>

Commit c3a8c5b6 ("cxgb3: move away from LLTX") exposed a bug in how
cxgb3 looks up the netdev_queue it stashes away in a qset during
initialization.  For multiport devices, the TX queue index it uses is
offset by the first_qset index of each port.  This leads to a crash
once LLTX is removed, since hard_start_xmit is called with one TX
queue lock held, while the TX reclaim timer task grabs a different
(wrong) TX queue lock when it frees skbs.

Fix this by removing the first_qset offset used to look up the TX
queue passed into t3_sge_alloc_qset() from setup_sge_qsets().
Signed-off-by: NRoland Dreier <rolandd@cisco.com>
Acked-by: NDivy Le Ray <divy@chelsio.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Fix coexistence of Fiber Channel over Ethernet (FCoE) and Flow Director (FDIR)
in 82599 and remove the disabling of FDIR when FCoE is enabled.

Currently, FDIR is turned off when FCoE is enabled under the assumption that
FCoE is always enabled with DCB being turned on. However, FDIR does not have
to be turned off all the time when FCoE is enabled since FCoE can be enabled
without DCB being turned on, e.g., use link pause only. This patch makes sure
that when DCB is turned on or off, FDIR is turned on or off correspondingly;
and when FCoE is enabled, it does not disable FDIR, rather, it will have FDIR
set up properly so FCoE and FDIR can coexist regardless of DCB being on or off.
Signed-off-by: NYi Zou <yi.zou@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Adding smp_mb__after_lock define to be used as a smp_mb call after
a lock.

Making it nop for x86, since {read|write|spin}_lock() on x86 are
full memory barriers.
Signed-off-by: NJiri Olsa <jolsa@redhat.com>
Signed-off-by: NEric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Adding memory barrier after the poll_wait function, paired with
receive callbacks. Adding fuctions sock_poll_wait and sk_has_sleeper
to wrap the memory barrier.

Without the memory barrier, following race can happen.
The race fires, when following code paths meet, and the tp->rcv_nxt
and __add_wait_queue updates stay in CPU caches.

CPU1                         CPU2

sys_select                   receive packet
  ...                        ...
  __add_wait_queue           update tp->rcv_nxt
  ...                        ...
  tp->rcv_nxt check          sock_def_readable
  ...                        {
  schedule                      ...
                                if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
                                        wake_up_interruptible(sk->sk_sleep)
                                ...
                             }

If there was no cache the code would work ok, since the wait_queue and
rcv_nxt are opposit to each other.

Meaning that once tp->rcv_nxt is updated by CPU2, the CPU1 either already
passed the tp->rcv_nxt check and sleeps, or will get the new value for
tp->rcv_nxt and will return with new data mask.
In both cases the process (CPU1) is being added to the wait queue, so the
waitqueue_active (CPU2) call cannot miss and will wake up CPU1.

The bad case is when the __add_wait_queue changes done by CPU1 stay in its
cache, and so does the tp->rcv_nxt update on CPU2 side.  The CPU1 will then
endup calling schedule and sleep forever if there are no more data on the
socket.

Calls to poll_wait in following modules were ommited:
	net/bluetooth/af_bluetooth.c
	net/irda/af_irda.c
	net/irda/irnet/irnet_ppp.c
	net/mac80211/rc80211_pid_debugfs.c
	net/phonet/socket.c
	net/rds/af_rds.c
	net/rfkill/core.c
	net/sunrpc/cache.c
	net/sunrpc/rpc_pipe.c
	net/tipc/socket.c
Signed-off-by: NJiri Olsa <jolsa@redhat.com>
Signed-off-by: NEric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

cifs: when ATTR_READONLY is set, only clear write bits on non-directories

On windows servers, ATTR_READONLY apparently either has no meaning or
serves as some sort of queue to certain applications for unrelated
behavior. This MS kbase article has details:

http://support.microsoft.com/kb/326549/

Don't clear the write bits directory mode when ATTR_READONLY is set.

Reported-by: pouchat@peewiki.net
Signed-off-by: NJeff Layton <jlayton@redhat.com>
Signed-off-by: NSteve French <sfrench@us.ibm.com>

cifs: remove cifsInodeInfo->inUse counter

It was purported to be a refcounter of some sort, but was never
used that way. It never served any purpose that wasn't served equally well
by the I_NEW flag.
Signed-off-by: NJeff Layton <jlayton@redhat.com>
Reviewed-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NSteve French <sfrench@us.ibm.com>

cifs: convert cifs_get_inode_info and non-posix readdir to use cifs_iget

Rather than allocating an inode and filling it out, have
cifs_get_inode_info fill out a cifs_fattr and call cifs_iget. This means
a pretty hefty reorganization of cifs_get_inode_info.

For the readdir codepath, add a couple of new functions for filling out
cifs_fattr's from different FindFile response infolevels.

Finally, remove cifs_new_inode since there are no more callers.
Signed-off-by: NJeff Layton <jlayton@redhat.com>
Reviewed-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NSteve French <sfrench@us.ibm.com>

On some boxes the mobile chipset is used and there is no LVDS device. In such
case we had better not initialize the LVDS output device so that one pipe can
be used for other output device. For example: E-TOP.

But unfortunately the LVDS device is still initialized on the boxes based on
mobile chipset in KMS mode. It brings that this pipe occupied by LVDS can't be
used for other output device.

After checking the acpidump we find that there is no LID device on such boxes.
In such case we can use the LID device to decide whether the LVDS device should
be initialized.

If there is no LID device, we can think that there is no LVDS device. It is
unnecessary to initialize the LVDS output device.
If there exists the LID device, it will continue the current flowchart.

Maybe on some boxes there is no LVDS device but the LID device is found. In
such case it should be added to the quirk list.

http://bugs.freedesktop.org/show_bug.cgi?id=21496
http://bugs.freedesktop.org/show_bug.cgi?id=21856
http://bugs.freedesktop.org/show_bug.cgi?id=21127Signed-off-by: NZhao Yakui <yakui.zhao@intel.com>
Reviewed-by: NJesse Barnes <jbarnes@virtuousgeek.org>
[anholt: squashed in style fixups]
Signed-off-by: NEric Anholt <eric@anholt.net>