The following patch in the driver is required to avoid USB 1.1 device
failures that may occur due to requests from USB OHCI controllers may
be overwritten if the latency for any pending request by the USB
controller is very long (in the range of milliseconds).
Signed-off-by: Libin Yang <libin.yang@amd.com> Cc: David Brownell <dbrownell@users.sourceforge.net> Cc: Alan Stern <stern@rowland.harvard.edu> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Original discussion:
http://thread.gmane.org/gmane.linux.usb.general/23217/focus=23248
or
http://marc.info/?l=linux-usb&m=125553790714133&w=2
9a68e39d4a701fb3be03cae9b462408664ebd205 broke carrier handling so that a
cp210x setup which needed the carrier lines set up (non CLOCAL) which did
not make a call which set the termios bits left the lines down even if
CLOCAL was not asserted.
Fix this not by reverting but by adding the proper dtr_rts and
carrier_raised methods. This both sets the modem lines properly and also
implements the correct blocking semantics for the port as required by
POSIX.
Signed-off-by: Alan Cox <alan@linux.intel.com> Reported-by: Karl Hiramoto <karl@hiramoto.org> Tested-by: Karl Hiramoto <karl@hiramoto.org> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Original discussion:
http://thread.gmane.org/gmane.linux.usb.general/23217/focus=23248
or
http://marc.info/?l=linux-usb&m=125553790714133&w=2
The tty_port code inherited a bug common to various drivers it was based
upon. If the tty is opened O_NONBLOCK we do not wait for the carrier to be
raised but we must still raise our modem lines if appropriate.
(There is a second question here about whether we should do so if CLOCAL is
set but that can wait)
Signed-off-by: Alan Cox <alan@linux.intel.com> Reported-by: Karl Hiramoto <karl@hiramoto.org> Tested-by: Karl Hiramoto <karl@hiramoto.org> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
If a direct reclaim makes no forward progress, it considers whether it
should go OOM or not. Whether OOM is triggered or not, it may retry the
allocation afterwards. In times past, this would always wake kswapd as
well but currently, kswapd is not woken up after direct reclaim fails.
For order-0 allocations, this makes little difference but if there is a
heavy mix of higher-order allocations that direct reclaim is failing for,
it might mean that kswapd is not rewoken for higher orders as much as it
did previously.
This patch wakes up kswapd when an allocation is being retried after a
direct reclaim failure. It would be expected that kswapd is already
awake, but this has the effect of telling kswapd to reclaim at the higher
order as well.
Signed-off-by: Mel Gorman <mel@csn.ul.ie> Reviewed-by: Christoph Lameter <cl@linux-foundation.org> Reviewed-by: Pekka Enberg <penberg@cs.helsinki.fi> Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Commit 341ce06f69abfafa31b9468410a13dbd60e2b237 ("page allocator:
calculate the alloc_flags for allocation only once") altered watermark
logic slightly by allowing rt_tasks that are handling an interrupt to set
ALLOC_HARDER. This patch brings the watermark logic more in line with
2.6.30.
This change results in a reduction of the number high-order GFP_ATOMIC
allocation failures reported. See
http://www.gossamer-threads.com/lists/linux/kernel/1144153
[rientjes@google.com: Spotted the problem] Signed-off-by: Mel Gorman <mel@csn.ul.ie> Reviewed-by: Pekka Enberg <penberg@cs.helsinki.fi> Reviewed-by: Rik van Riel <riel@redhat.com> Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Daniel Lezcano reported a leak in 'struct pid' and 'struct pid_namespace'
that is discussed in:
http://lkml.org/lkml/2009/10/2/159.
To summarize the thread, when container-init is terminated, it sets the
PF_EXITING flag, zaps other processes in the container and waits to reap
them. As a part of reaping, the container-init should flush any /proc
dentries associated with the processes. But because the container-init is
itself exiting and the following PF_EXITING check, the dentries are not
flushed, resulting in leak in /proc inodes and dentries.
This fix reverts the commit 7766755a2f249e7e0 ("Fix /proc dcache deadlock
in do_exit") which introduced the check for PF_EXITING. At the time of
the commit, shrink_dcache_parent() flushed dentries from other filesystems
also and could have caused a deadlock which the commit fixed. But as
pointed out by Eric Biederman, after commit 0feae5c47aabdde59,
shrink_dcache_parent() no longer affects other filesystems. So reverting
the commit is now safe.
As pointed out by Jan Kara, the leak is not as critical since the
unclaimed space will be reclaimed under memory pressure or by:
echo 3 > /proc/sys/vm/drop_caches
But since this check is no longer required, its best to remove it.
Signed-off-by: Sukadev Bhattiprolu <sukadev@us.ibm.com> Reported-by: Daniel Lezcano <dlezcano@fr.ibm.com> Acked-by: Eric W. Biederman <ebiederm@xmission.com> Acked-by: Jan Kara <jack@ucw.cz> Cc: Andrea Arcangeli <andrea@cpushare.com> Cc: Serge Hallyn <serue@us.ibm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
For FS_IOC_RESVSP and FS_IOC_RESVSP64 compat_sys_ioctl() uses its
arg argument as a pointer to userspace. However it is missing a
a call to compat_ptr() which will do a proper pointer conversion.
This was introduced with 3e63cbb1 "fs: Add new pre-allocation ioctls
to vfs for compatibility with legacy xfs ioctls".
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com> Cc: Ankit Jain <me@ankitjain.org> Acked-by: Christoph Hellwig <hch@lst.de> Cc: Al Viro <viro@zeniv.linux.org.uk> Acked-by: Arnd Bergmann <arndbergmann@googlemail.com> Acked-by: David S. Miller <davem@davemloft.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
v3020_mmio_read_bit() always returns 0 when left_shift > 7.
v3020_mmio_read_bit()'s return type is (unsigned char). The code returns
a value masked by (1 << left_shift) that is casted to the return type. If
left_shift is larger than 7, the cast will always result in a 0 return
value. The problem was discovered with left_shift = 16, and the included
patch corrects the problem.
The bug was introduced in the last (Apr 3 2009) commit of the file, kernel
versions 2.6.30 and later.
PPS events must be recorded according to PPS's mode settings.
If a process asks for (i.e.) capture-assert events only, when the PPS
client calls the pps_event() function to save the current PPS event, we
should verify the event type and then discard unwanted ones.
Also, without this patch userland processes waiting for a specific PPS
event (assert or clear but not both) may be awakened at wrong time.
Signed-off-by: Rodolfo Giometti <giometti@linux.it> Tested-by: William S. Brasher <billb958@door.net> Tested-by: Reg Clemens <clemens@dwf.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
WARNING: at lib/debugobjects.c:255 debug_print_object+0x42/0x50()
Hardware name: System Product Name
ODEBUG: init active object type: timer_list
Modules linked in:
Pid: 2619, comm: dmesg Tainted: G W 2.6.32-rc5-tip+ #5298
Call Trace:
[<81035443>] warn_slowpath_common+0x6a/0x81
[<8120e483>] ? debug_print_object+0x42/0x50
[<81035498>] warn_slowpath_fmt+0x29/0x2c
[<8120e483>] debug_print_object+0x42/0x50
[<8120ec2a>] __debug_object_init+0x279/0x2d7
[<8120ecb3>] debug_object_init+0x13/0x18
[<810409d2>] init_timer_key+0x17/0x6f
[<81041526>] free_uid+0x50/0x6c
[<8104ed2d>] put_cred_rcu+0x61/0x72
[<81067fac>] rcu_do_batch+0x70/0x121
debugobjects warns about an enqueued timer being initialized. If
CONFIG_USER_SCHED=y the user management code uses delayed work to
remove the user from the hash table and tear down the sysfs objects.
free_uid is called from RCU and initializes/schedules delayed work if
the usage count of the user_struct is 0. The init/schedule happens
outside of the uidhash_lock protected region which allows a concurrent
caller of find_user() to reference the about to be destroyed
user_struct w/o preventing the work from being scheduled. If the next
free_uid call happens before the work timer expired then the active
timer is initialized and the work scheduled again.
The race was introduced in commit 5cb350ba (sched: group scheduling,
sysfs tunables) and made more prominent by commit 3959214f (sched:
delayed cleanup of user_struct)
Move the init/schedule_delayed_work inside of the uidhash_lock
protected region to prevent the race.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Acked-by: Dhaval Giani <dhaval@linux.vnet.ibm.com> Cc: Paul E. McKenney <paulmck@us.ibm.com> Cc: Kay Sievers <kay.sievers@vrfy.org> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
The s2255 driver had logic which aborted processing of a video frame
if there was no process waiting on the video buffer in question. That
simply doesn't work when the application is doing things in an
asynchronous manner. If the application went to the trouble to queue
the buffer in the first place, then the driver should always attempt
to complete it - even if the application at that moment has its
attention turned elsewhere. Applications which always blocked waiting
for I/O on the capture device would not have been affected by this.
Applications which *mostly* blocked waiting for I/O on the capture
device probably only would have been somewhat affected (frame lossage,
at a rate which goes up as the application blocks less). Applications
which never blocked on the capture device (e.g. polling only) however
would never have been able to receive any video frames, since in that
case this "is anyone waiting on this?" check on the buffer never would
have evalutated true. This patch just deletes that harmful check
against the buffer's wait queue.
Signed-off-by: Mike Isely <isely@pobox.com> Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com> Signed-off-by: Michael Krufky <mkrufky@kernellabs.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Fixing kernel oops when driver attemps to load xc2028 firmware.
Note by djh: the patch contribute by Martin is a port of a fix I made during
the PCTV 340e development. It's a temporary workaround that fixes a regression
(an OOPS condition) and the real fix should be in the code that manages the
i2c master on the dib7000p. But this fix does address the immmediate
regression and should be merged upstream until we do a cleaner fix.
Because the counters were not reset when starting up streaming, they would
be reused from the previous run. This can result in cases such that when the
second instance of streaming starts up, the "cnt" variable in
em28xx_audio_isocirq() can end up being negative, resulting in attempting to
write to memory before the start of runtime->dma_area (as well as having a
negative number of bytes to copy).
Multiplication by 62500 causes an overflow in the 32 bit freq variable,
which is later divided by 1000 when using FM radio.
This patch prevents the overflow by scaling the frequency value correctly
upfront. Thanks to Henk Vergonet for spotting the problem and providing
a preliminary patch, which this changeset was based upon.
While having tda18271 module set with debug=17 (cal & info prints) and
cal=0 (delay calibration process until first use) - I discovered that
during the calibration process, if the frequency test for 69750000
returned a bcal of 0 (see tda18721-fe.c in tda18271_powerscan func) that
the tuner wouldn't be able to pickup any of the frequencies in the range
(all the other frequencies bands returned bcal=1). I spent some time
going over the code and the NXP's tda18271 spec (ver.4 of it i think) and
adding a lot of debug prints and walking/stepping through the calibration
process. I found that when the powerscan fails to find a frequency, the
rf calibration is not run and the default value is supposed to be used in
its place (pulled from the RF_CAL_map table) - but something was getting
goofed up there.
Now, my c coding skills are very rusty, but i think root of the problem is
a signedness issue with the math operation for calculating the rf_a1 and
rf_a2 values in tda18271_rf_tracking_filters_init func, which results in
values like 20648 for rf_a1 (when it should probably have a value like 0,
or so slightly negative that it should be zero - this bad value for rf_a1
would in turn makes the approx calc within
tda18271c2_rf_tracking_filters_correction go out of whack). The simplest
solution i found was to explicitly convert the signedness of the
denominator to avoid the implicit conversion. The values placed into the
u32 rf_freq array should never exceed about 900mhz, so i think the s32 max
value shouldn't be an issue in this case.
I've tested it out a little, and even when i get a bcal=0 with the
modified code, the default calibration value gets used, rf_a1 is zero, and
the tuner seems to lock on the stream and mythtv seems to play it fine.
Signed-off-by: Seth Barry <seth@cyberseth.com> Signed-off-by: Michael Krufky <mkrufky@kernellabs.com> Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
The bttv driver function which handles switching of the video standard
(set_tvnorm() in bttv-driver.c) includes a check which can optionally
also reset the cropping configuration to a default value. It is
"optional" based on a comparison of the cropcap parameters of the
previous vs the newly requested video standard. The comparison is
being done with a memcmp(), a function which only returns a true value
if the comparison actually fails.
This if-statement appears to have been written to assume wrong
memcmp() semantics. That is, it was re-initializing the cropping
configuration only if the new video standard did NOT have different
cropcap values. That doesn't make any sense. One definitely should
reset things if the cropcap parameters are different - if there's any
comparison to made at all.
The effect of this problem was that a transition from, say, PAL to
NTSC would leave in place old cropping setup that made sense for the
PAL geometry but not for NTSC. If the application doesn't care about
cropping it also won't try to reset the cropping configuration,
resulting in an improperly cropped video frame. In the case I was
testing this actually caused black video frames to be displayed.
Another interesting effect of this bug is that if one does something
which does NOT change the video standard and this function is run,
then the cropping setup gets reset anyway - again because of the
backwards comparison. It turns out that just running anything which
merely opens and closes the video device node (e.g. v4l-info) will
cause this to happen. One can argue that simply opening the device
node and not doing anything to it should not mess with any of its
state - but because of this behavior, any TV app which does such
things (e.g. xawtv) probably therefore doesn't see the problem.
The solution is to fix the sense of the if-statement. It's easy to
see how this mistake could have been made given how memcmp() works.
The patch is therefore removal of a single "!" character from the
if-statement in set_tvnorm in bttv-driver.c.
Signed-off-by: Mike Isely <isely@pobox.com> Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com> Signed-off-by: Michael Krufky <mkrufky@kernellabs.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
There is a subtle interaction in the bttv driver which can result in
fields being repeatedly processed out of order. This is a problem
specifically when running in V4L2_FIELD_ALTERNATE mode (probably the
most common case).
1. The determination of which fields are associated with which buffers
happens in videobuf, before the bttv driver gets a chance to queue the
corresponding DMA. Thus by the point when the DMA is queued for a
given buffer, the algorithm has to do the queuing based on the
buffer's already assigned field type - not based on which field is
"next" in the video stream.
2. The driver normally tries to queue both the top and bottom fields
at the same time (see bttv_irq_next_video()). It tries to sort out
top vs bottom by looking at the field type for the next 2 available
buffers and assigning them appropriately.
3. However the bttv driver *always* actually processes the top field
first. There's even an interrupt set aside for specifically
recognizing when the top field has been processed so that it can be
marked done even while the bottom field is still being DMAed.
Given all of the above, if one gets into a situation where
bttv_irq_next_video() gets entered when the first available buffer has
been pre-associated as a bottom field, then the function is going to
process the buffers out of order. That first available buffer will be
put into the bottom field slot and the buffer after that will be put
into the top field slot. Problem is, since the top field is always
processed first by the driver, then that second buffer (the one after
the first available buffer) will be the first one to be finished.
Because of the strict fifo handling of all video buffers, then that
top field won't be seen by the app until after the bottom field is
also processed. Worse still, the app will get back the
chronologically later bottom field first, *before* the top field is
received. The buffer's timestamps will even be backwards.
While not fatal to most TV apps, this behavior can subtlely degrade
userspace deinterlacing (probably will cause jitter). That's probably
why it has gone unnoticed. But it will also cause serious problems if
the app in question discards all but the latest received buffer (a
latency minimizing tactic) - causing one field to only ever be
displayed since the other is now always late. Unfortunately once you
get into this state, you're stuck this way - because having consumed
two buffers, now the next time around the "first" available buffer
will again be a bottom field and the same thing happens.
How can we get into this state? In a perfect world, where there's
always a few free buffers queued to the driver, it should be
impossible. However if something disrupts streaming, e.g. if the
userspace app can't queue free buffers fast enough for a moment due
perhaps to a CPU scheduling glitch, then the driver can get
momentarily starved and some number of fields will be dropped. That's
OK. But if an odd number of fields get dropped, then that "first"
available buffer might be the bottom field and now we're stuck...
This patch fixes that problem by deliberately only setting up a single
field for one frame if we don't get a top field as the first available
buffer. By purposely skipping the other field, then we only handle a
single buffer thus bringing things back into proper sync (i.e. top
field first) for the next frame. To do this we just drop the few
lines in bttv_irq_next_video() that attempt to set up the second
buffer when that second buffer isn't for the bottom field.
This is definitely a problem in when in V4L2_FIELD_ALTERNATE mode. In
the other modes this change either has no effect or doesn't harm
things any further anyway.
Signed-off-by: Mike Isely <isely@pobox.com> Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com> Signed-off-by: Michael Krufky <mkrufky@kernellabs.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
d451564 broke ARM by requiring KM_IRQ_PTE, KM_NMI and KM_NMI_PTE to
always be defined. Solve this by providing invalid definitions for
these constants, but only if CONFIG_DEBUG_HIGHMEM is enabled.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Code was added to mm/higmem.c that depends on several
kmap types that powerpc does not support. We add dummy
invalid definitions for KM_NMI, KM_NM_PTE, and KM_IRQ_PTE.
According to list discussion, this fix should not be needed
anymore starting with 2.6.33. The code is commented to this
effect so hopefully we will remember to remove this.
Signed-off-by: Becky Bruce <beckyb@kernel.crashing.org> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
debug_kmap_atomic() tries to prevent ever printing more than 10
warnings, but it does so by testing whether an unsigned integer
is equal to 0. However, if the warning is caused by a nested
IRQ, then this counter may underflow and the stream of warnings
will never end.
If only the output substream of a RawMIDI device has been opened and
if this device is then opened with O_RDWR | O_APPEND and if the
initialization of the input substream fails (either because of low
memory or because the device driver's open callback fails), then the
runtime structure of the already open output substream will be freed
and all following writes through the first handle will cause
snd_rawmidi_write() to use the NULL runtime pointer.
Commit 9a1b64caac82aa02cb74587ffc798e6f42c6170a in 2.6.30 moved the
substream initialization code to where it would be executed every time
the substream is opened.
This had the consequence that any further opening would drop and leak
the data in the existing buffer, and that the device driver's open
callback would be called multiple times, unexpectedly.
Commit 9a1b64caac82aa02cb74587ffc798e6f42c6170a in 2.6.30 dropped the
check that a substream must already have been opened with O_APPEND to be
able to open it a second time.
This would make it possible for a substream to be switched to append
mode, which would mean that non-atomic writes would fail unexpectedly.
Sending an Active Sensing message when closing a port can interfere with
the following data if the port is reopened and a note-on is sent before
the device's timeout has elapsed. Therefore, it is better to disable
this setting by default.
Signed-off-by: Clemens Ladisch <clemens@ladisch.de> Signed-off-by: Takashi Iwai <tiwai@suse.de>
In 2.6.23 kernel, commit a32ea1e1f925399e0d81ca3f7394a44a6dafa12c
("Fix read/truncate race") fixed a race in the generic code, and as a
side effect, now do_generic_file_read() can ask us to readpage() past
the i_size. This seems to be correctly handled by the block routines
(e.g. block_read_full_page() fills the page with zeroes in case if
somebody is trying to read past the last inode's block).
JFFS2 doesn't handle this; it assumes that it won't be asked to read
pages which don't exist -- and thus that there will be at least _one_
valid 'frag' on the page it's being asked to read. It will fill any
holes with the following memset:
When the 'closest smaller match' returned by jffs2_lookup_node_frag() is
actually on a previous page and ends before 'offset', that results in:
memset(buf, 0, <huge unsigned negative>);
Hopefully, in most cases the corruption is fatal, and quickly causing
random oopses, like this:
root@10.0.0.4:~/ltp-fs-20090531# ./testcases/kernel/fs/ftest/ftest01
Unable to handle kernel paging request for data at address 0x00000008
Faulting instruction address: 0xc01cd980
Oops: Kernel access of bad area, sig: 11 [#1]
[...]
NIP [c01cd980] rb_insert_color+0x38/0x184
LR [c0043978] enqueue_hrtimer+0x88/0xc4
Call Trace:
[c6c63b60] [c004f9a8] tick_sched_timer+0xa0/0xe4 (unreliable)
[c6c63b80] [c0043978] enqueue_hrtimer+0x88/0xc4
[c6c63b90] [c0043a48] __run_hrtimer+0x94/0xbc
[c6c63bb0] [c0044628] hrtimer_interrupt+0x140/0x2b8
[c6c63c10] [c000f8e8] timer_interrupt+0x13c/0x254
[c6c63c30] [c001352c] ret_from_except+0x0/0x14
--- Exception: 901 at memset+0x38/0x5c
LR = jffs2_read_inode_range+0x144/0x17c
[c6c63cf0] [00000000] (null) (unreliable)
This patch fixes the issue, plus fixes all LTP tests on NAND/UBI with
JFFS2 filesystem that were failing since 2.6.23 (seems like the bug
above also broke the truncation).
Reported-By: Anton Vorontsov <avorontsov@ru.mvista.com> Tested-By: Anton Vorontsov <avorontsov@ru.mvista.com> Signed-off-by: David Woodhouse <David.Woodhouse@intel.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Fix combine_word problem where first octet is not
read properly. The only affected place seems to be the
INPUT_TERMINAL type. Before now, sound controls can be created
with the output terminal's name which is a fallback mechanism
used only for unknown input terminal types. For example,
Line can wrongly appear as Speaker. After the change it
should appear as Line.
The side effect of this change can be that users
can expect the wrong control name in their scripts or
programs while now we return the correct one.
Probably, these defines should use get_unaligned_le16 and
friends.
During 'check' of a raid1 or raid10 it is possible for the management
thread to spend a lot of time running 'memcmp' on blocks from
different devices, so make sure the thread has a chance to schedule.
raid5d already has a cond_resched (in process_stripe).
Reported-By: Lee Howard <faxguy@howardsilvan.com> Signed-off-by: NeilBrown <neilb@suse.de> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
This value is visible through sysfs and is used by mdadm
when it manages a reshape (backing up data that is about to be
rearranged). So it is important that it is always correct.
Current it does not get updated properly when a reshape
starts which can cause problems when assembling an array
that is in the middle of being reshaped.
If a 'sync_max' has been set (via sysfs), it is wrong to clear it
until a resync (or reshape or recovery ...) actually reached that
point.
So if a resync is interrupted (e.g. by device failure),
leave 'resync_max' unchanged.
This is particularly important for 'reshape' operations that do not
change the size of the array. For such operations mdadm needs to
monitor the reshape taking rolling backups of the section being
reshaped. If resync_max gets cleared, the reshape can get ahead of
mdadm and then the backups that mdadm creates are useless.
As reported by Rick Farina (sidhayn@gmail.com), removing the RTL8187
USB stick, or unloading the driver rtl8187 using rmmod will cause a
kernel oops. There are at least two forms of the failure, (1) BUG:
Scheduling while atomic, and (2) a fatal kernel page fault. This
problem is reported in Bugzilla #14539.
This problem does not occur for kernel 2.6.31, but does for 2.6.32-rc2,
thus it is technically a regression; however, bisection did not locate
any faulty patch. The fix was found by comparing the faulty code in
rtl8187 with p54usb. My interpretation is that the handling of work
queues in mac80211 changed enough to the LEDs to be unregistered
before tasks on the work queues are cancelled. Previously, these
actions could be done in either order.
(Herton Ronaldo Krzesinski <herton@mandriva.com.br> reports that the
code is the same in 2.6.31, so this may be a candidate for 2.6.31.x.
-- JWL)
Signed-off-by: Larry Finger <Larry.Finger@lwfinger.net> Reported-by: Rick Farina <sidhayn@gmail.com> Tested-by: Rick Farina <sidhayn@gmail.com> Signed-off-by: John W. Linville <linville@tuxdriver.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
A negative offset could be used to index before the event buffer and
lead to a security breach.
Signed-off-by: Dave Jones <davej@redhat.com> Signed-off-by: James Bottomley <James.Bottomley@suse.de> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
SMB writes are sent with a starting offset and length. When the server
supports the newer SMB trans2 posix open (rather than using the SMB
NTCreateX) a file can be opened with SMB_O_APPEND flag, and for that
case Samba server assumes that the offset sent in SMBWriteX is unneeded
since the write should go to the end of the file - which can cause
problems if the write was cached (since the beginning part of a
page could be written twice by the client mm). Jeff suggested that
masking the flag on posix open on the client is easiest for the time
being. Note that recent Samba server also had an unrelated problem with
SMB NTCreateX and append (see samba bugzilla bug number 6898) which
should not affect current Linux clients (unless cifs Unix Extensions
are disabled).
The cifs client did not send the O_APPEND flag on posix open
before 2.6.29 so the fix is unneeded on early kernels.
In the future, for the non-cached case (O_DIRECT, and forcedirectio mounts)
it would be possible and useful to send O_APPEND on posix open (for Windows
case: FILE_APPEND_DATA but not FILE_WRITE_DATA on SMB NTCreateX) but for
cached writes although the vfs sets the offset to end of file it
may fragment a write across pages - so we can't send O_APPEND on
open (could result in sending part of a page twice).
Reviewed-by: Shirish Pargaonkar <shirishp@us.ibm.com> Signed-off-by: Jeff Layton <jlayton@redhat.com> Signed-off-by: Steve French <sfrench@us.ibm.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Lookup called during network boot (network root filesystem
for diskless workstation) has case where nd is null in
lookup. This patch fixes that in cifs_lookup.
(Shirish noted that 2.6.30 and 2.6.31 stable need the same check)
Signed-off-by: Shirish Pargaonkar <shirishp@us.ibm.com> Acked-by: Jeff Layton <jlayton@redhat.com> Tested-by: Vladimir Stavrinov <vs@inist.ru> Signed-off-by: Steve French <sfrench@us.ibm.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Fix the commit ec06aedd44 that intended to turn off querying for server inode
numbers when server doesn't consistently support inode numbers. Presumably
the commit didn't actually clear the CIFS_MOUNT_SERVER_INUM flag, perhaps a
typo.
Signed-off-by: Suresh Jayaraman <sjayaraman@suse.de> Acked-by: Jeff Layton <jlayton@redhat.com> Signed-off-by: Steve French <sfrench@us.ibm.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
It's possible that a server will return a valid FileID when we query the
FILE_INTERNAL_INFO for the root inode, but then zeroed out inode numbers
when we do a FindFile with an infolevel of
SMB_FIND_FILE_ID_FULL_DIR_INFO.
In this situation turn off querying for server inode numbers, generate a
warning for the user and just generate an inode number using iunique.
Once we generate any inode number with iunique we can no longer use any
server inode numbers or we risk collisions, so ensure that we don't do
that in cifs_get_inode_info either.
Reported-by: Timothy Normand Miller <theosib@gmail.com> Signed-off-by: Jeff Layton <jlayton@redhat.com> Signed-off-by: Steve French <sfrench@us.ibm.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Because it's lighter weight, CIFS tries to use CIFSGetSrvInodeNumber to
verify the accessibility of the root inode and then falls back to doing a
full QPathInfo if that fails with -EOPNOTSUPP. I have at least a report
of a server that returns NT_STATUS_INTERNAL_ERROR rather than something
that translates to EOPNOTSUPP.
Rather than trying to be clever with that call, just have
is_path_accessible do a normal QPathInfo. That call is widely
supported and it shouldn't increase the overhead significantly.
Signed-off-by: Jeff Layton <jlayton@redhat.com> Signed-off-by: Steve French <sfrench@us.ibm.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
This fixes a kernel oops reported by Markus Trippelsdorf in the email
titled "[NILFS users] kernel Oops while running nilfs_cleanerd".
The oops was caused by a bug of error path in
nilfs_ioctl_move_blocks() function, which was inlined in
nilfs_ioctl_clean_segments().
nilfs_ioctl_move_blocks checks duplication of blocks which will be
moved in garbage collection. But, the check should have be done
within nilfs_ioctl_move_inode_block() to prevent list corruption among
buffers storing the target blocks.
To fix the kernel oops, this moves forward the duplication check
before the list insertion.
I also tested this for stable trees [2.6.30, 2.6.31].
Commit 0c570cdeb8fdfcb354a3e9cd81bfc6a09c19de0c
(PM / yenta: Fix cardbus suspend/resume regression) caused resume to
fail on systems with two CardBus bridges. While the exact nature
of the failure is not known at the moment, it can be worked around by
splitting the yenta resume into an early part, executed during the
early phase of resume, that will only resume the socket and power it
up if there was a card in it during suspend, and a late part,
executed during "regular" resume, that will carry out all of the
remaining yenta resume operations.
Fixes http://bugzilla.kernel.org/show_bug.cgi?id=14334, which is a
listed regression from 2.6.31.
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl> Acked-by: Dominik Brodowski <linux@dominikbrodowski.net> Reported-by: Stephen J. Gowdy <gowdy@cern.ch> Tested-by: Jose Marino <braket@hotmail.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
A misconfiguration by the firmware of the U4 PCIe bridge on PowerMac G5
with the U4 bridge (latest generations, may also affect the iMac G5
"iSight") is causing us to re-assign the PCI BARs of the video card,
which can get it out of sync with the firmware, thus breaking offb.
This works around it by fixing up the bridge configuration properly
at boot time. It also fixes a bug where the firmware provides us with
an incorrect set of accessible regions in the device-tree.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Patch f598282f5145036312d90875d0ed5c14b49fd8a7 exposed a problem in
powerpc MSI-X functionality, making network interfaces such as ixgbe
and cxgb3 stop to work when MSI-X is enabled. RX interrupts were not
being generated.
The problem was caused because MSI irq was not being effectively
unmasked after device initialization.
Signed-off-by: Andre Detsch <adetsch@br.ibm.com> Signed-off-by: Michael Ellerman <michael@ellerman.id.au> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Since the change of how interrupts are disabled during suspend,
certain PowerBook models started exhibiting various issues during
suspend or resume from sleep.
I finally tracked it down to the code that runs various "platform"
functions (kind of little scripts extracted from the device-tree),
which uses our i2c and PMU drivers expecting interrutps to work,
and at a time where with the new scheme, they have been disabled.
This causes timeouts internally which for some reason results in
the PMU being unable to see the trackpad, among other issues, really
it depends on the machine. Most of the time, we fail to properly adjust
some clocks for suspend/resume so the results are not always
predictable.
This patch fixes it by using IRQF_TIMER for both the PMU and the I2C
interrupts. I prefer doing it this way than moving the call sites since
I really want those platform functions to still be called after all
drivers (and before sysdevs).
We also do a slight cleanup to via-pmu.c driver to make sure the
ADB autopoll mask is handled correctly when doing bus resets
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
# modprobe kvm-amd
kvm_init()
kvm_init_debug() <-- second initialization clobbers kvm's
internal pointers to dentries
kvm_arch_init()
kvm_exit_debug() <-- and frees them
If execution gets to the end of kvm_init(), then the calling module has been
established as the kvm provider. Move the debugfs initialization to the end of
the function, and remove the now-unnecessary call to kvm_exit_debug() from the
error path. That way we avoid trampling on the debugfs entries and freeing
them twice.
Signed-off-by: Darrick J. Wong <djwong@us.ibm.com> Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
The function iommu_feature_disable is required on system
shutdown to disable the IOMMU but it is marked as __init.
This may result in a panic if the memory is reused. This
patch fixes this bug.
Replenishment of receive buffers is done in the tasklet handling
received frames as well as in a workqueue. When we are in the tasklet
we cannot sleep and thus attempt atomic skb allocations. It is generally
not a big problem if this fails since iwl_rx_allocate is always followed
by a call to iwl_rx_queue_restock which will queue the work to replenish
the buffers at a time when sleeping is allowed.
We thus add the __GFP_NOWARN to the skb allocation in iwl_rx_allocate to
reduce the noise if such an allocation fails while we still have enough
buffers. We do maintain the warning and the error message when we are low
on buffers to communicate to the user that there is a potential problem with
memory availability on system
This addresses issue reported upstream in thread "iwlagn: order 2 page
allocation failures" in
http://thread.gmane.org/gmane.linux.kernel.wireless.general/39187
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com> Acked-by: Mel Gorman <mel@csn.ul.ie> Signed-off-by: John W. Linville <linville@tuxdriver.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
RX handling maintains a few lists that keep track of the RX buffers.
Buffers move from one list to the other as they are used, replenished, and
again made available for usage. In one such instance, when a buffer is used
it enters the "rx_used" list. When buffers are replenished an skb is
attached to the buffer and it is moved to the "rx_free" list. The problem
here is that the buffer is first removed from the "rx_used" list _before_ the
skb is allocated. Thus, if the skb allocation fails this buffer remains
removed from the "rx_used" list and is thus lost for future usage.
Fix this by first allocating the skb before trying to attach it to a list.
We add an additional check to not do this unnecessarily.
Reported-by: Rick Farrington <rickdic@hotmail.com> Signed-off-by: Reinette Chatre <reinette.chatre@intel.com> Signed-off-by: John W. Linville <linville@tuxdriver.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
If a system switches back and forth between hot and cold mode,
the MCE code will print a stream of critical kernel messages.
Extend the throttling code to properly notice this, by
only printing the first hot + cold transition and omitting
the rest up to CHECK_INTERVAL (5 minutes).
This way we'll only get a single incident of:
[ 102.356584] CPU0: Temperature above threshold, cpu clock throttled (total events = 1)
[ 102.357000] Disabling lock debugging due to kernel taint
[ 102.369223] CPU0: Temperature/speed normal
Every 5 minutes. The 'total events' count tells the number of cold/hot
transitions detected, should overheating occur after 5 minutes again:
[ 402.357580] CPU0: Temperature above threshold, cpu clock throttled (total events = 24891)
[ 402.358001] CPU0: Temperature/speed normal
[ 450.704142] Machine check events logged
It is possible to have !Anon but SwapBacked pages, and some apps could
create huge number of such pages with MAP_SHARED|MAP_ANONYMOUS. These
pages go into the ANON lru list, and hence shall not be protected: we only
care mapped executable files. Failing to do so may trigger OOM.
Tested-by: Christian Borntraeger <borntraeger@de.ibm.com> Reviewed-by: Rik van Riel <riel@redhat.com> Signed-off-by: Wu Fengguang <fengguang.wu@intel.com> Cc: Hugh Dickins <hugh.dickins@tiscali.co.uk> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
On Thu, Aug 13, 2009 at 04:14:58PM +1000, Benjamin Herrenschmidt wrote:
> On Tue, 2009-08-11 at 11:39 +0200, Bastian Blank wrote:
> > This patch just disables this driver on SMP kernels, as it is obviously
> > not supported.
> Why not remove the #error instead ? :-) I don't think it's still
> meaningful, especially since we use the timebase for delays nowadays
> which doesn't depend on the CPU frequency...
Your call. Take this one:
The build of a PowerMac 32bit kernel currently fails with
error: #warning "WARNING, CPUFREQ not recommended on SMP kernels"
Thie patch removes the not longer applicable SMP warning from the
PowerMac cpufreq code.
The NFSv4 renew daemon is shared between all active super blocks that refer
to a particular NFS server, so it is wrong to be shutting it down in
nfs4_kill_super every time a super block is destroyed.
This patch therefore kills nfs4_renewd_prepare_shutdown altogether, and
leaves it up to nfs4_shutdown_client() to also shut down the renew daemon
by means of the existing call to nfs4_kill_renewd().
Commits 29fba38b (nfs41: lease renewal) and fc01cea9 (nfs41: sequence
operation) introduce a couple of put_rpccred() calls on credentials for
which there is no corresponding get_rpccred().
See http://bugzilla.kernel.org/show_bug.cgi?id=14249
RFC 3530 states that when we recieve the error NFS4ERR_RESOURCE, we are not
supposed to bump the sequence number on OPEN, LOCK, LOCKU, CLOSE, etc
operations. The problem is that we map that error into EREMOTEIO in the XDR
layer, and so the NFSv4 middle-layer routines like seqid_mutating_err(),
and nfs_increment_seqid() don't recognise it.
The fix is to defer the mapping until after the middle layers have
processed the error.
Actually pass the NFS_FILE_SYNC option to the server to avoid a
Panic in nfs_direct_write_complete() when a commit fails.
At the end of an nfs write, if the nfs commit fails, all the writes
will be rescheduled. They are supposed to be rescheduled as NFS_FILE_SYNC
writes, but the rpc_task structure is not completely intialized and so
the option is not passed. When the rescheduled writes complete, the
return indicates that they are NFS_UNSTABLE and we try to do another
commit. This leads to a Panic because the commit data structure pointer
was set to null in the initial (failed) commit attempt.
As seen in <http://bugs.debian.org/549002>, nfs4_init_client() can
overrun the source string when copying the client IP address from
nfs_parsed_mount_data::client_address to nfs_client::cl_ipaddr. Since
these are both treated as null-terminated strings elsewhere, the copy
should be done with strlcpy() not memcpy().
Commit 9ef1d4c7c7aca1cd436612b6ca785b726ffb8ed8 ("[NETLINK]: Missing
initializations in dumped data") introduced a typo in
initialization. This patch fixes this.
Signed-off-by: Jiri Pirko <jpirko@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net> Cc: Chuck Ebbert <cebbert@redhat.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
In mii monitor mode, bond_check_dev_link() calls the the ioctl
handler of slave devices. It stores the ndo_do_ioctl function
pointer to a static (!) ioctl variable and later uses it to call the
handler with the IOCTL macro.
If another thread executes bond_check_dev_link() at the same time
(even with a different bond, which none of the locks prevent), a
race condition occurs. If the two racing slaves have different
drivers, this may result in one driver's ioctl handler being
called with a pointer to a net_device controlled with a different
driver, resulting in unpredictable breakage.
Unless I am overlooking something, the "static" must be a
copy'n'paste error (?).
Signed-off-by: Jiri Bohac <jbohac@suse.cz> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
If two streams are started immediately after one another (such as a
playback and a recording stream), the call to set hw params fails with
EBUSY. This patch makes the call succeed, so playback and recording will
work properly.
I found a deadlock bug in UNIX domain socket, which makes able to DoS
attack against the local machine by non-root users.
How to reproduce:
1. Make a listening AF_UNIX/SOCK_STREAM socket with an abstruct
namespace(*), and shutdown(2) it.
2. Repeat connect(2)ing to the listening socket from the other sockets
until the connection backlog is full-filled.
3. connect(2) takes the CPU forever. If every core is taken, the
system hangs.
PoC code: (Run as many times as cores on SMP machines.)
int main(void)
{
int ret;
int csd;
int lsd;
struct sockaddr_un sun;
/* make an abstruct name address (*) */
memset(&sun, 0, sizeof(sun));
sun.sun_family = PF_UNIX;
sprintf(&sun.sun_path[1], "%d", getpid());
/* connect loop */
alarm(15); /* forcely exit the loop after 15 sec */
for (;;) {
csd = socket(AF_UNIX, SOCK_STREAM, 0);
ret = connect(csd, (struct sockaddr *)&sun, sizeof(sun));
if (-1 == ret) {
perror("connect()");
break;
}
puts("Connection OK");
}
return 0;
}
(*) Make sun_path[0] = 0 to use the abstruct namespace.
If a file-based socket is used, the system doesn't deadlock because
of context switches in the file system layer.
Why this happens:
Error checks between unix_socket_connect() and unix_wait_for_peer() are
inconsistent. The former calls the latter to wait until the backlog is
processed. Despite the latter returns without doing anything when the
socket is shutdown, the former doesn't check the shutdown state and
just retries calling the latter forever.
Patch:
The patch below adds shutdown check into unix_socket_connect(), so
connect(2) to the shutdown socket will return -ECONREFUSED.
Signed-off-by: Tomoki Sekiyama <tomoki.sekiyama.qu@hitachi.com> Signed-off-by: Masanori Yoshida <masanori.yoshida.tv@hitachi.com> Cc: Chuck Ebbert <cebbert@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
"b43: Fix PPC crash in rfkill polling on unload" fixed the bug reported
in Bugzilla No. 14181; however, it introduced a new bug. Whenever the
radio switch was turned off, it was necessary to unload and reload
the driver for it to recognize the switch again.
This patch fixes both the original bug in #14181 and the bug introduced by
the previous patch. It must be stated, however, that if there is a BCM4306/3
with an rfkill switch (not yet proven), then the driver will need an
unload/reload cycle to turn the device back on.
Signed-off-by: Larry Finger <Larry.Finger@lwfinger.net> Signed-off-by: John W. Linville <linville@tuxdriver.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
The destination keyring specified to request_key() and co. is made available to
the process that instantiates the key (the slave process started by
/sbin/request-key typically). This is passed in the request_key_auth struct as
the dest_keyring member.
keyctl_instantiate_key and keyctl_negate_key() call get_instantiation_keyring()
to get the keyring to attach the newly constructed key to at the end of
instantiation. This may be given a specific keyring into which a link will be
made later, or it may be asked to find the keyring passed to request_key(). In
the former case, it returns a keyring with the refcount incremented by
lookup_user_key(); in the latter case, it returns the keyring from the
request_key_auth struct - and does _not_ increment the refcount.
The latter case will eventually result in an oops when the keyring prematurely
runs out of references and gets destroyed. The effect may take some time to
show up as the key is destroyed lazily.
To fix this, the keyring returned by get_instantiation_keyring() must always
have its refcount incremented, no matter where it comes from.
This can be tested by setting /etc/request-key.conf to:
keyctl add user _display aaaaaaaa @u
while keyctl request2 user test:x test:x @u &&
keyctl list @u;
do
keyctl request2 user test:x test:x @u;
sleep 31;
keyctl list @u;
done
which will oops eventually. Changing the negate line to have @u rather than
%S at the end is important as that forces the latter case by passing a special
keyring ID rather than an actual keyring ID.
Reported-by: Alexander Zangerl <az@bond.edu.au> Signed-off-by: David Howells <dhowells@redhat.com> Tested-by: Alexander Zangerl <az@bond.edu.au> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Cc: Chuck Ebbert <cebbert@redhat.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
acpi_get_pci_dev() may be called for a non-PCI device, in which case
it should return NULL. However, it assumes that every handle it
finds in the ACPI CA name space, between given device handle and the
PCI root bridge handle, corresponds to a PCI-to-PCI bridge with an
existing secondary bus. For this reason, when it finds a struct
pci_dev object corresponding to one of them, it doesn't check if
its 'subordinate' field is a valid pointer. This obviously leads to
a NULL pointer dereference if acpi_get_pci_dev() is called for a
non-PCI device with a PCI parent which is not a bridge.
To fix this issue make acpi_get_pci_dev() check if pdev->subordinate
is not NULL for every device it finds on the path between the root
bridge and the device it's supposed to get to and return NULL if the
"target" device cannot be found.
http://bugzilla.kernel.org/show_bug.cgi?id=14129
(worked in 2.6.30, regression in 2.6.31)
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl> Reported-by: Danny Feng <dfeng@redhat.com> Reviewed-by: Alex Chiang <achiang@hp.com> Tested-by: chepioq <chepioq@gmail.com> Signed-off-by: Len Brown <len.brown@intel.com> Cc: Chuck Ebbert <cebbert@redhat.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Signed-off-by: Sam Ravnborg <sam@ravnborg.org> Cc: Tim Abbott <tabbott@ksplice.com> Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru> Cc: Richard Henderson <rth@twiddle.net> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Commit 51b563fc93c8cb5bff1d67a0a71c374e4a4ea049 ("arm, cris, mips,
sparc, powerpc, um, xtensa: fix build with bash 4.0") removed a few
CPPFLAGS with vital include paths necessary to build vmlinux.lds
on MIPS, and moved the calculation of the 'jiffies' symbol
directly to vmlinux.lds.S but forgot to change make ifdef/... to
cpp macros.
Signed-off-by: Manuel Lauss <manuel.lauss@gmail.com>
[sam: moved assignment of CPPFLAGS arch/mips/kernel/Makefile] Signed-off-by: Sam Ravnborg <sam@ravnborg.org> Acked-by: Dmitri Vorobiev <dmitri.vorobiev@movial.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
fsnotify_add_mark is supposed to add a mark to the g_list and i_list and to
set the group and inode for the mark. fsnotify_destroy_mark_by_entry uses
the fact that ->group != NULL to know if this group should be destroyed or
if it's already been done.
But fsnotify_add_mark sets the group and inode before it actually adds the
mark to the i_list and g_list. This can result in a race in inotify, it
requires 3 threads.
sys_inotify_add_watch("file") sys_inotify_add_watch("file") sys_inotify_rm_watch([a])
inotify_update_watch()
inotify_new_watch()
inotify_add_to_idr()
^--- returns wd = [a]
inotfiy_update_watch()
inotify_new_watch()
inotify_add_to_idr()
fsnotify_add_mark()
^--- returns wd = [b]
returns to userspace;
inotify_idr_find([a])
^--- gives us the pointer from task 1
fsnotify_add_mark()
^--- this is going to set the mark->group and mark->inode fields, but will
return -EEXIST because of the race with [b].
fsnotify_destroy_mark()
^--- since ->group != NULL we call back
into inotify_freeing_mark() which calls
inotify_remove_from_idr([a])
since fsnotify_add_mark() failed we call:
inotify_remove_from_idr([a]) <------WHOOPS it's not in the idr, this could
have been any entry added later!
The fix is to make sure we don't set mark->group until we are sure the mark is
on the inode and fsnotify_add_mark will return success.
Signed-off-by: Eric Paris <eparis@redhat.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
There is an erratum for IOMMU hardware which documents
undefined behavior when forwarding SMI requests from
peripherals and the DTE of that peripheral has a sysmgt
value of 01b. This problem caused weird IO_PAGE_FAULTS in my
case.
This patch implements the suggested workaround for that
erratum into the AMD IOMMU driver. The erratum is
documented with number 63.
fuse_direct_io() has a loop where requests are allocated in each
iteration. if allocation fails, the loop is broken out and follows
into an unconditional fuse_put_request() on that invalid pointer.
Signed-off-by: Anand V. Avati <avati@gluster.com> Signed-off-by: Miklos Szeredi <mszeredi@suse.cz> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
If TSS we are switching to resides in high memory task switch will fail
since address will be truncated. Windows2k3 does this sometimes when
running with more then 4G
Not a single line of actual code in the function was really
fundamentally correct.
Problems ranged from lack of proper range checking, to removing the last
character written (which admittedly is usually '\n'), to not accepting
hex numbers even though the 'show' routine would show the data in that
format.
This tries to do better.
Acked-by: Michael Buesch <mb@bu3sch.de> Tested-and-acked-by: Jack Steiner <steiner@sgi.com> Cc: Jiri Kosina <jkosina@suse.cz> Cc: Michael Gilbert <michael.s.gilbert@gmail.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
We never want to rely on the hvc workqueue to emit output, because the
most interesting output is when the kernel is broken. This will
improve oops/crash/console message for better debugging.
Instead, we force-poll until all output is emitted.
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
For new stepping of PCH, the display reference clock
is fully under driver's control. This one trys to setup
all needed reference clock for different outputs. Older
stepping of PCH chipset should be ignoring this.
This fixes output failure issue on newer PCH which requires
driver to take control of reference clock enabling.
Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com> Signed-off-by: Eric Anholt <eric@anholt.net> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
FDI M/N calculation hasn't taken the current pipe color depth into account,
but always set as 24bpp. This one checks current pipe color depth setting,
and change FDI M/N calculation a little to use bits_per_pixel first, then
convert to bytes_per_pixel later.
This fixes display corrupt issue on Arrandle LVDS with 1600x900 panel
in 18bpp dual-channel mode.
Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com> Signed-off-by: Eric Anholt <eric@anholt.net> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Bruno Prémont and Dunphy, Bill noticed me that NILFS will certainly
hang on ARM-based targets.
I found this was caused by an underflow of dirty pages counter. A
b-tree cache routine was marking page dirty without adjusting page
account information.
This fixes the dirty page accounting leak and resolves the hang on
arm-based targets.
Reported-by: Bruno Prémont <bonbons@linux-vserver.org> Reported-by: Dunphy, Bill <WDunphy@tandbergdata.com> Signed-off-by: Ryusuke Konishi <konishi.ryusuke@lab.ntt.co.jp> Tested-by: Bruno Prémont <bonbons@linux-vserver.org> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Restoring %ebp after the call to audit_syscall_exit() is not
only unnecessary (because the register didn't get clobbered),
but in the sysenter case wasn't even doing the right thing: It
loaded %ebp from a location below the top of stack (RBP <
ARGOFFSET), i.e. arbitrary kernel data got passed back to user
mode in the register.
Signed-off-by: Jan Beulich <jbeulich@novell.com> Acked-by: Roland McGrath <roland@redhat.com>
LKML-Reference: <4AE5CC4D020000780001BD13@vpn.id2.novell.com> Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
In try_to_unuse(), swcount is a local copy of *swap_map, including the
SWAP_HAS_CACHE bit; but a wrong comparison against swap_count(*swap_map),
which masks off the SWAP_HAS_CACHE bit, succeeded where it should fail.
That had the effect of resetting the mm from which to start searching
for the next swap page, to an irrelevant mm instead of to an mm in which
this swap page had been found: which may increase search time by ~20%.
But we're used to swapoff being slow, so never noticed the slowdown.
Remove that one spurious use of swap_count(): Bo Liu thought it merely
redundant, Hugh rewrote the description since it was measurably wrong.
Signed-off-by: Bo Liu <bo-liu@hotmail.com> Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk> Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Don't pass NULL pointers to fput() in the error handling paths of the NOMMU
do_mmap_pgoff() as it can't handle it.
The following can be used as a test program:
int main() { static long long a[1024 * 1024 * 20] = { 0 }; return a;}
Without the patch, the code oopses in atomic_long_dec_and_test() as called by
fput() after the kernel complains that it can't allocate that big a chunk of
memory. With the patch, the kernel just complains about the allocation size
and then the program segfaults during execve() as execve() can't complete the
allocation of all the new ELF program segments.
Reported-by: Robin Getz <rgetz@blackfin.uclinux.org> Signed-off-by: David Howells <dhowells@redhat.com> Acked-by: Robin Getz <rgetz@blackfin.uclinux.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
A few parts of the uv_hub_info structure are initialized
incorrectly.
- n_val is being loaded with m_val.
- gpa_mask is initialized with a bytes instead of an unsigned long.
- Handle the case where none of the alias registers are used.
Lastly I converted the bau over to using the uv_hub_info->m_val
which is the correct value.
Without this patch, booting a large configuration hits a
problem where the upper bits of the gnode affect the pnode
and the bau will not operate.
For some strange reason the netif_running() check
ended up after the actual type change instead of
before, potentially causing all kinds of problems
if the interface is up while changing the type;
one of the problems manifests itself as a warning:
WARNING: at net/mac80211/iface.c:651 ieee80211_teardown_sdata+0xda/0x1a0 [mac80211]()
Hardware name: Aspire one
Pid: 2596, comm: wpa_supplicant Tainted: G W 2.6.31-10-generic #32-Ubuntu
Call Trace:
[] warn_slowpath_common+0x6d/0xa0
[] warn_slowpath_null+0x15/0x20
[] ieee80211_teardown_sdata+0xda/0x1a0 [mac80211]
[] ieee80211_if_change_type+0x4a/0xc0 [mac80211]
[] ieee80211_change_iface+0x61/0xa0 [mac80211]
[] cfg80211_wext_siwmode+0xc7/0x120 [cfg80211]
[] ioctl_standard_call+0x58/0xf0
Cc: Arjan van de Ven <arjan@infradead.org> Signed-off-by: Johannes Berg <johannes@sipsolutions.net> Signed-off-by: John W. Linville <linville@tuxdriver.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
When hostapd injects a frame, e.g. an authentication or association
response, mac80211 looks for a suitable access point virtual interface
to associate the frame with based on its source address. This makes it
possible e.g. to correctly assign sequence numbers to the frames.
A small typo in the ethernet address comparison statement caused a
failure to find a suitable ap interface. Sequence numbers on such
frames where therefore left unassigned causing some clients
(especially windows-based 11b/g clients) to reject them and fail to
authenticate or associate with the access point. This patch fixes the
typo in the address comparison statement.
Signed-off-by: Björn Smedman <bjorn.smedman@venatech.se> Reviewed-by: Johannes Berg <johannes@sipsolutions.net> Signed-off-by: John W. Linville <linville@tuxdriver.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
The reason of this bug is that the error handling code of
cifs_get_tcp_session()
calls kfree() when corresponding kmalloc() failed.
(The kmalloc() is called by extract_hostname().)
Signed-off-by: Hitoshi Mitake <mitake@dcl.info.waseda.ac.jp> Reviewed-by: Jeff Layton <jlayton@redhat.com> Signed-off-by: Steve French <sfrench@us.ibm.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
On SMP guests, reads from the ring might bypass used index reads. This
causes guest crashes because host writes to used index to signal ring
data readiness. Fix this by inserting rmb before used ring reads.
Signed-off-by: Michael S. Tsirkin <mst@redhat.com> Signed-off-by: Rusty Russell <rusty@rustcorp.com.au> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
projects / karo-tx-linux.git / log