linux-2.6-microblaze.git
16 years agoblock: unify sysfs size node handling
Tejun Heo [Mon, 25 Aug 2008 10:56:09 +0000 (19:56 +0900)]
block: unify sysfs size node handling

Now that capacity and __dev are moved to part0, part0 and others can
share the same method.

Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: move __dev from disk to part0
Tejun Heo [Fri, 29 Aug 2008 07:01:47 +0000 (09:01 +0200)]
block: move __dev from disk to part0

Move disk->__dev to part0->__dev.  This simplifies bdget_disk() and
lookup_devt() and allows common sysfs attributes to be unified.
part_to_disk() is updated to handle part0 -> disk.

Updated to include a fix from Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>,
he writes:

"part0 is a "special" partition and doesn't need to have capacity set - this
fixes regression caused by "block: move __dev from disk to part0" commit."

Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: move capacity from disk to part0
Tejun Heo [Mon, 25 Aug 2008 10:56:07 +0000 (19:56 +0900)]
block: move capacity from disk to part0

Move disk->capacity to part0->nr_sects and convert all users who
directly accessed the field to use {get|set}_capacity().  This is done
early to allow the __dev field to be moved.

Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: introduce partition 0
Tejun Heo [Wed, 3 Sep 2008 07:06:42 +0000 (09:06 +0200)]
block: introduce partition 0

genhd and partition code handled disk and partitions separately.  All
information about the whole disk was in struct genhd and partitions in
struct hd_struct.  However, the whole disk (part0) and other
partitions have a lot in common and the data structures end up having
good number of common fields and thus separate code paths doing the
same thing.  Also, the partition array was indexed by partno - 1 which
gets pretty confusing at times.

This patch introduces partition 0 and makes the partition array
indexed by partno.  Following patches will unify the handling of disk
and parts piece-by-piece.

This patch also implements disk_partitionable() which tests whether a
disk is partitionable.  With coming dynamic partition array change,
the most common usage of disk_max_parts() will be testing whether a
disk is partitionable and the number of max partitions will become
much less important.

Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: implement and use {disk|part}_to_dev()
Tejun Heo [Mon, 25 Aug 2008 10:56:05 +0000 (19:56 +0900)]
block: implement and use {disk|part}_to_dev()

Implement {disk|part}_to_dev() and use them to access generic device
instead of directly dereferencing {disk|part}->dev.  To make sure no
user is left behind, rename generic devices fields to __dev.

This is in preparation of unifying partition 0 handling with other
partitions.

Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: implement CONFIG_DEBUG_BLOCK_EXT_DEVT
Tejun Heo [Mon, 25 Aug 2008 10:47:25 +0000 (19:47 +0900)]
block: implement CONFIG_DEBUG_BLOCK_EXT_DEVT

Extended devt introduces non-contiguos device numbers.  This patch
implements a debug option which forces most devt allocations to be
from the extended area and spreads them out.  This is enabled by
default if DEBUG_KERNEL is set and achieves...

1. Detects code paths in kernel or userland which expect predetermined
   consecutive device numbers.

2. When something goes wrong, avoid corruption as adding to the minor
   of earlier partition won't lead to the wrong but valid device.

Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agosd/ide-disk: apply extended minors to sd and ide
Tejun Heo [Mon, 25 Aug 2008 10:47:24 +0000 (19:47 +0900)]
sd/ide-disk: apply extended minors to sd and ide

Update sd and ide-disk such that they can take advantage of extended
minors.

ide-disk already has 64 minors per device and currently doesn't use
extended minors although after this patch it can be turned on by
simply tweaking constants.

sd only had 16 minors per device causing problems on certain peculiar
configurations.  This patch lifts the restriction and enables it to
use upto 64 minors.

Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: adjust formatting for large minors and add ext_range sysfs attr
Tejun Heo [Mon, 25 Aug 2008 10:47:23 +0000 (19:47 +0900)]
block: adjust formatting for large minors and add ext_range sysfs attr

With extended minors and the soon-to-follow debug feature, large minor
numbers for block devices will be common.  This patch does the
followings to make printouts pretty.

* Adapt print formats such that large minors don't break the
  formatting.

* For extended MAJ:MIN, %02x%02x for MAJ:MIN used in
  printk_all_partitions() doesn't cut it anymore.  Update it such that
  %03x:%05x is used if either MAJ or MIN doesn't fit in %02x.

* Implement ext_range sysfs attribute which shows total minors the
  device can use including both conventional minor space and the
  extended one.

Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: implement extended dev numbers
Tejun Heo [Mon, 25 Aug 2008 10:47:22 +0000 (19:47 +0900)]
block: implement extended dev numbers

Implement extended device numbers.  A block driver can tell block
layer that it wants to use extended device numbers.  After the usual
minor space is used up, block layer automatically allocates devt's
from EXT_BLOCK_MAJOR.

Currently only one major number is allocated for this but as the
allocation is strictly on-demand, ~1mil minor space under it should
suffice unless the system actually has more than ~1mil partitions and
if that ever happens adding more majors to the extended devt area is
easy.

Due to internal implementation issues, the first partition can't be
allocated on the extended area.  In other words, genhd->minors should
at least be 1.  This limitation will be lifted by later changes.

Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: fix diskstats access
Tejun Heo [Mon, 25 Aug 2008 10:47:21 +0000 (19:47 +0900)]
block: fix diskstats access

There are two variants of stat functions - ones prefixed with double
underbars which don't care about preemption and ones without which
disable preemption before manipulating per-cpu counters.  It's unclear
whether the underbarred ones assume that preemtion is disabled on
entry as some callers don't do that.

This patch unifies diskstats access by implementing disk_stat_lock()
and disk_stat_unlock() which take care of both RCU (for partition
access) and preemption (for per-cpu counter access).  diskstats access
should always be enclosed between the two functions.  As such, there's
no need for the versions which disables preemption.  They're removed
and double underbars ones are renamed to drop the underbars.  As an
extra argument is added, there's no danger of using the old version
unconverted.

disk_stat_lock() uses get_cpu() and returns the cpu index and all
diskstat functions which access per-cpu counters now has @cpu
argument to help RT.

This change adds RCU or preemption operations at some places but also
collapses several preemption ops into one at others.  Overall, the
performance difference should be negligible as all involved ops are
very lightweight per-cpu ones.

Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: fix disk->part[] dereferencing race
Tejun Heo [Wed, 3 Sep 2008 07:03:02 +0000 (09:03 +0200)]
block: fix disk->part[] dereferencing race

disk->part[] is protected by its matching bdev's lock.  However,
non-critical accesses like collecting stats and printing out sysfs and
proc information used to be performed without any locking.  As
partitions can come and go dynamically, partitions can go away
underneath those non-critical accesses.  As some of those accesses are
writes, this theoretically can lead to silent corruption.

This patch fixes the race by using RCU for the partition array and dev
reference counter to hold partitions.

* Rename disk->part[] to disk->__part[] to make sure no one outside
  genhd layer proper accesses it directly.

* Use RCU for disk->__part[] dereferencing.

* Implement disk_{get|put}_part() which can be used to get and put
  partitions from gendisk respectively.

* Iterators are implemented to help iterate through all partitions
  safely.

* Functions which require RCU readlock are marked with _rcu suffix.

* Use disk_put_part() in __blkdev_put() instead of directly putting
  the contained kobject.

Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: don't depend on consecutive minor space
Tejun Heo [Wed, 3 Sep 2008 07:01:48 +0000 (09:01 +0200)]
block: don't depend on consecutive minor space

* Implement disk_devt() and part_devt() and use them to directly
  access devt instead of computing it from ->major and ->first_minor.

  Note that all references to ->major and ->first_minor outside of
  block layer is used to determine devt of the disk (the part0) and as
  ->major and ->first_minor will continue to represent devt for the
  disk, converting these users aren't strictly necessary.  However,
  convert them for consistency.

* Implement disk_max_parts() to avoid directly deferencing
  genhd->minors.

* Update bdget_disk() such that it doesn't assume consecutive minor
  space.

* Move devt computation from register_disk() to add_disk() and make it
  the only one (all other usages use the initially determined value).

These changes clean up the code and will help disk->part dereference
fix and extended block device numbers.

Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: make variable and argument names more consistent
Tejun Heo [Wed, 3 Sep 2008 07:01:09 +0000 (09:01 +0200)]
block: make variable and argument names more consistent

In hd_struct, @partno is used to denote partition number and a number
of other places use @part to denote hd_struct.  Functions use @part
and @index instead.  This causes confusion and makes it difficult to
use consistent variable names for hd_struct.  Always use @partno if a
variable represents partition number.

Also, print out functions use @f or @part for seq_file argument.  Use
@seqf uniformly instead.

Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: misc updates
Tejun Heo [Mon, 25 Aug 2008 10:47:17 +0000 (19:47 +0900)]
block: misc updates

This patch makes the following misc updates in preparation for
disk->part dereference fix and extended block devt support.

* implment part_to_disk()

* fix comment about gendisk->part indexing

* rename get_part() to disk_map_sector()

* don't use n which is always zero while printing disk information in
  diskstats_show()

Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: update add_partition() error handling
Tejun Heo [Mon, 25 Aug 2008 10:30:16 +0000 (19:30 +0900)]
block: update add_partition() error handling

d805dda4 tried to fix error case handling in add_partition() but had a
few problems.

* disk->part[] entry is set early and left dangling if operation
  fails.

* Once device initialized, the last put_device() is responsible for
  freeing all the resources.  The failure path freed part_stats and p
  regardless of put_device() causing double free.

* holders subdir holds reference to the disk device, so failure path
  should remove it to release resources properly which was missing.

This patch fixes the above problems and while at it move partition
slot busy check into add_partition() for completeness and inlines
holders subdirectory creation.  Using separate function for it just
obfuscates the code.

Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Abdel Benamrouche <draconux@gmail.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: allow deleting zero length partition
Tejun Heo [Mon, 25 Aug 2008 10:30:15 +0000 (19:30 +0900)]
block: allow deleting zero length partition

delete_partition() was noop for zero length partition.  As the
addition code allows creating zero lenght partition and deletion is
assumed to always succeed, this causes memory leak for zero length
partitions.  Allow zero length partitions to end their meaningless
lives.

While at it, allow deleting zero lenght partition via
BLKPG_DEL_PARTITION ioctl too.

Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: use class_dev_iterator instead of class_for_each_device()
Tejun Heo [Wed, 3 Sep 2008 06:57:12 +0000 (08:57 +0200)]
block: use class_dev_iterator instead of class_for_each_device()

Recent block_class iteration updates 5c6f35c5..27f3025 converted all
class device iteration to class_for_each_device() and
class_find_device(), which are correct but pain in the ass to use.
This pach converts them to newly introduced class_dev_iterator so that
they can use more natural control structures instead of separate
callbacks and struct to pass parameters to them.

This results in smaller and easier code.

This patch also restores the original behavior of not printing header
in /proc/partitions if there's no partition to print.  This is trivial
but still user-visible behavior.

Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Greg Kroah-Hartman <gregkh@suse.de>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: don't grab block_class_lock unnecessarily
Tejun Heo [Wed, 3 Sep 2008 06:53:37 +0000 (08:53 +0200)]
block: don't grab block_class_lock unnecessarily

block_class_lock protects major_names array and bdev_map and doesn't
have anything to do with block class devices.  Don't grab them while
iterating over block class devices.

Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: fix partition info printouts
Tejun Heo [Mon, 25 Aug 2008 10:30:12 +0000 (19:30 +0900)]
block: fix partition info printouts

Recent block_class iteration updates 5c6f35c5..27f3025 broke partition
info printouts.

* printk_all_partitions(): Partition print out stops when it meets a
  partition hole.  Partition printing inner loop should continue
  instead of exiting on empty partition slot.

* /proc/partitions and /proc/diskstats: If all information can't be
  read in single read(), the information is truncated.  This is
  because find_start() doesn't actually update the counter containing
  the initial seek.  It runs to the end and ends up always reporting
  EOF on the second read.

This patch fixes both problems.

Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Greg Kroah-Hartman <gregkh@suse.de>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agodriver-core: use klist for class device list and implement iterator
Tejun Heo [Mon, 25 Aug 2008 17:50:19 +0000 (19:50 +0200)]
driver-core: use klist for class device list and implement iterator

Iterating over entries using callback usually isn't too fun especially
when the entry being iterated over can't be manipulated freely.  This
patch converts class->p->class_devices to klist and implements class
device iterator so that the users can freely build their own control
structure.  The users are also free to call back into class code
without worrying about locking.

class_for_each_device() and class_find_device() are converted to use
the new iterators, so their users don't have to worry about locking
anymore either.

Note: This depends on klist-dont-iterate-over-deleted-entries patch
because class_intf->add/remove_dev() depends on proper synchronization
with device removal.

Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Greg Kroah-Hartman <gregkh@suse.de>
Cc: Jens Axboe <jens.axboe@oracle.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoklist: don't iterate over deleted entries
Tejun Heo [Mon, 25 Aug 2008 17:50:16 +0000 (19:50 +0200)]
klist: don't iterate over deleted entries

A klist entry is kept on the list till all its current iterations are
finished; however, a new iteration after deletion also iterates over
deleted entries as long as their reference count stays above zero.
This causes problems for cases where there are users which iterate
over the list while synchronized against list manipulations and
natuarally expect already deleted entries to not show up during
iteration.

This patch implements dead flag which gets set on deletion so that
iteration can skip already deleted entries.  The dead flag piggy backs
on the lowest bit of knode->n_klist and only visible to klist
implementation proper.

While at it, drop klist_iter->i_head as it's redundant and doesn't
offer anything in semantics or performance wise as klist_iter->i_klist
is dereferenced on every iteration anyway.

Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Greg Kroah-Hartman <gregkh@suse.de>
Cc: Alan Stern <stern@rowland.harvard.edu>
Cc: Jens Axboe <jens.axboe@oracle.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoAdd some block/ source files to the kernel-api docbook. Fix kernel-doc notation in...
Randy Dunlap [Tue, 19 Aug 2008 18:13:11 +0000 (20:13 +0200)]
Add some block/ source files to the kernel-api docbook. Fix kernel-doc notation in them as needed. Fix changed function parameter names. Fix typos/spellos. In comments, change REQ_SPECIAL to REQ_TYPE_SPECIAL and REQ_BLOCK_PC to REQ_TYPE_BLOCK_PC.

Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: make bi_phys_segments an unsigned int instead of short
Jens Axboe [Fri, 15 Aug 2008 08:56:11 +0000 (10:56 +0200)]
block: make bi_phys_segments an unsigned int instead of short

raid5 can overflow with more than 255 stripes, and we can increase it
to an int for free on both 32 and 64-bit archs due to the padding.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: raid fixups for removal of bi_hw_segments
Jens Axboe [Fri, 15 Aug 2008 08:41:18 +0000 (10:41 +0200)]
block: raid fixups for removal of bi_hw_segments

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agodrop vmerge accounting
Mikulas Patocka [Fri, 15 Aug 2008 08:20:02 +0000 (10:20 +0200)]
drop vmerge accounting

Remove hw_segments field from struct bio and struct request. Without virtual
merge accounting they have no purpose.

Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: drop virtual merging accounting
Mikulas Patocka [Fri, 15 Aug 2008 08:15:19 +0000 (10:15 +0200)]
block: drop virtual merging accounting

Remove virtual merge accounting.

Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: update documentation for deadline fifo_batch tunable
Aaron Carroll [Thu, 14 Aug 2008 08:17:15 +0000 (18:17 +1000)]
block: update documentation for deadline fifo_batch tunable

Update the description of fifo_batch to match the current implementation,
and include a description of how to tune it.

Signed-off-by: Aaron Carroll <aaronc@gelato.unsw.edu.au>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agodeadline-iosched: non-functional fixes
Aaron Carroll [Thu, 14 Aug 2008 08:17:14 +0000 (18:17 +1000)]
deadline-iosched: non-functional fixes

* convert goto to simpler while loop;
 * use rq_end_sector() instead of computing manually;
 * fix false comments;
 * remove spurious whitespace;
 * convert rq_rb_root macro to an inline function.

Signed-off-by: Aaron Carroll <aaronc@gelato.unsw.edu.au>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agodeadline-iosched: allow non-sequential batching
Aaron Carroll [Thu, 14 Aug 2008 08:17:13 +0000 (18:17 +1000)]
deadline-iosched: allow non-sequential batching

Deadline currently only batches sector-contiguous requests, so except
for a few circumstances (e.g. requests in a single direction), it is
essentially first come first served.  This is bad for throughput, so
change it to CSCAN, which means requests in a batch do not need to be
sequential and are issued in increasing sector order.

Signed-off-by: Aaron Carroll <aaronc@gelato.unsw.edu.au>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agovirtio_blk: use a wrapper function to access io context information of IO requests
Fernando Luis Vázquez Cao [Thu, 14 Aug 2008 07:59:13 +0000 (09:59 +0200)]
virtio_blk: use a wrapper function to access io context information of IO requests

struct request has an ioprio member but it is never updated because
currently bios do not hold io context information. The implication of
this is that virtio_blk ends up passing useless information to the
backend driver.

That said, some IO schedulers such as CFQ do store io context
information in struct request, but use private members for that, which
means that that information cannot be directly accessed in a IO
scheduler-independent way.

This patch adds a function to obtain the ioprio of a request. We should
avoid accessing ioprio directly and use this function instead, so that
its users do not have to care about future changes in block layer
structures or what the currently active IO controller is.

This patch does not introduce any functional changes but paves the way
for future clean-ups and enhancements.

Signed-off-by: Fernando Luis Vazquez Cao <fernando@oss.ntt.co.jp>
Acked-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoKill REQ_TYPE_FLUSH
David Woodhouse [Wed, 13 Aug 2008 11:35:09 +0000 (12:35 +0100)]
Kill REQ_TYPE_FLUSH

It was only used by ps3disk, and it should probably have been
REQ_TYPE_LINUX_BLOCK + REQ_LB_OP_FLUSH.

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoAllow elevators to sort/merge discard requests
David Woodhouse [Sat, 9 Aug 2008 15:42:20 +0000 (16:42 +0100)]
Allow elevators to sort/merge discard requests

But blkdev_issue_discard() still emits requests which are interpreted as
soft barriers, because naïve callers might otherwise issue subsequent
writes to those same sectors, which might cross on the queue (if they're
reallocated quickly enough).

Callers still _can_ issue non-barrier discard requests, but they have to
take care of queue ordering for themselves.

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoAdd BLKDISCARD ioctl to allow userspace to discard sectors
David Woodhouse [Mon, 11 Aug 2008 14:58:42 +0000 (15:58 +0100)]
Add BLKDISCARD ioctl to allow userspace to discard sectors

We may well want mkfs tools to use this to mark the whole device as
unwanted before they format it, for example.

The ioctl takes a pair of uint64_ts, which are start offset and length
in _bytes_. Although at the moment it might make sense for them both to
be in 512-byte sectors, I don't want to limit the ABI to that.

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoUse WRITE_BARRIER in blkdev_issue_flush(), not (1<<BIO_RW_BARRIER)
OGAWA Hirofumi [Mon, 11 Aug 2008 16:07:08 +0000 (17:07 +0100)]
Use WRITE_BARRIER in blkdev_issue_flush(), not (1<<BIO_RW_BARRIER)

Barriers should be submitted with the WRITE flag set.

Signed-off-by: OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblktrace: simplify flags handling in __blk_add_trace
David Woodhouse [Sun, 10 Aug 2008 11:33:00 +0000 (12:33 +0100)]
blktrace: simplify flags handling in __blk_add_trace

Let the compiler see what's going on, and it can all get a lot simpler.
On PPC64 this reduces the size of the code calculating these bits by
about 60%. On x86_64 it's less of a win -- only 40%.

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblktrace: support discard requests
David Woodhouse [Sun, 10 Aug 2008 10:21:57 +0000 (11:21 +0100)]
blktrace: support discard requests

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoSupport 'discard sectors' operation.
David Woodhouse [Tue, 5 Aug 2008 17:08:56 +0000 (18:08 +0100)]
Support 'discard sectors' operation.

We can benefit from knowing that the file system no longer cares about
the contents of certain sectors, by throwing them away immediately and
then never having to garbage collect them, and using the extra free
space to make our operations more efficient. Do so.

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoSupport 'discard sectors' operation in translation layer support core
David Woodhouse [Tue, 5 Aug 2008 17:08:25 +0000 (18:08 +0100)]
Support 'discard sectors' operation in translation layer support core

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoLet the block device know when sectors can be discarded
David Woodhouse [Tue, 5 Aug 2008 17:05:46 +0000 (18:05 +0100)]
Let the block device know when sectors can be discarded

[hirofumi@mail.parknet.co.jp: discard _after_ checking for corrupt chains]

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>
Acked-by: OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoAdd 'discard' request handling
David Woodhouse [Tue, 5 Aug 2008 17:01:53 +0000 (18:01 +0100)]
Add 'discard' request handling

Some block devices benefit from a hint that they can forget the contents
of certain sectors. Add basic support for this to the block core, along
with a 'blkdev_issue_discard()' helper function which issues such
requests.

The caller doesn't get to provide an end_io functio, since
blkdev_issue_discard() will automatically split the request up into
multiple bios if appropriate. Neither does the function wait for
completion -- it's expected that callers won't care about when, or even
_if_, the request completes. It's only a hint to the device anyway. By
definition, the file system doesn't _care_ about these sectors any more.

[With feedback from OGAWA Hirofumi <hirofumi@mail.parknet.co.jp> and
Jens Axboe <jens.axboe@oracle.com]

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoFix up comments about matching flags between bio and rq
David Woodhouse [Sat, 9 Aug 2008 15:22:17 +0000 (16:22 +0100)]
Fix up comments about matching flags between bio and rq

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agohighmem: use bio_has_data() in the bounce path
Jens Axboe [Thu, 14 Aug 2008 11:12:15 +0000 (13:12 +0200)]
highmem: use bio_has_data() in the bounce path

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: use bio_has_data() in the IO completion path
Jens Axboe [Fri, 8 Aug 2008 09:06:45 +0000 (11:06 +0200)]
block: use bio_has_data() in the IO completion path

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: use bio_has_data() to check for data carrying bio
Jens Axboe [Fri, 8 Aug 2008 09:04:44 +0000 (11:04 +0200)]
block: use bio_has_data() to check for data carrying bio

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoblock: add bio_has_data() to detect whether a bio carries data or not
Jens Axboe [Fri, 8 Aug 2008 09:17:12 +0000 (11:17 +0200)]
block: add bio_has_data() to detect whether a bio carries data or not

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoSG_IO block filter whitelist missing MMC SET READ AHEAD command
xiphmont@xiph.org [Fri, 22 Aug 2008 09:12:21 +0000 (11:12 +0200)]
SG_IO block filter whitelist missing MMC SET READ AHEAD command

I have another request for the block filter SG_IO command whitelist,
specifically the MMC streaming command set SET READ AHEAD command.
The command applies only to MMC CDROM/DVDROM drives with the streaming
optional feature set.  The command is useful to cdparanoia in that it
allows explicit cache control side effects that are, on many drives,
cdparanoia's most efficient way to flush/disable the media cache on
cdrom drives. I am aware of no reason why it should not be accessible
from usespace.

Also note that the command is already fully accessible through the
SCSI-native version of the SG_IO ioctl as well as the traditional SG
interface.  The command is only being refused on block devices.  That
means that on a typical stock distro, the command is available through
/dev/sg* but not /dev/scd* although both are typically available and
accessible.  Filtering the command is not providing any protection,
only a confusing inconsistency.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
16 years agoMerge branch 'upstream' of git://ftp.linux-mips.org/pub/scm/upstream-linus
Linus Torvalds [Wed, 8 Oct 2008 18:41:10 +0000 (11:41 -0700)]
Merge branch 'upstream' of git://ftp.linux-mips.org/upstream-linus

* 'upstream' of git://ftp.linux-mips.org/pub/scm/upstream-linus:
  [MIPS] Sibyte: Register PIO PATA device only for Swarm and Litte Sur

16 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6
Linus Torvalds [Wed, 8 Oct 2008 18:40:19 +0000 (11:40 -0700)]
Merge git://git./linux/kernel/git/davem/net-2.6

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6:
  tcp: Fix tcp_hybla zero congestion window growth with small rho and large cwnd.
  net: Fix netdev_run_todo dead-lock
  tcp: Fix possible double-ack w/ user dma
  net: only invoke dev->change_rx_flags when device is UP
  netrom: Fix sock_orphan() use in nr_release
  ax25: Quick fix for making sure unaccepted sockets get destroyed.
  Revert "ax25: Fix std timer socket destroy handling."
  [Bluetooth] Add reset quirk for A-Link BlueUSB21 dongle
  [Bluetooth] Add reset quirk for new Targus and Belkin dongles
  [Bluetooth] Fix double frees on error paths of btusb and bpa10x drivers

16 years ago[MIPS] Sibyte: Register PIO PATA device only for Swarm and Litte Sur
Ralf Baechle [Wed, 8 Oct 2008 18:14:33 +0000 (19:14 +0100)]
[MIPS] Sibyte: Register PIO PATA device only for Swarm and Litte Sur

Symbol name spaghetti which is too complicated to cleanup on this stage
of the release cycle breaks the build on BCM1480 platforms.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
16 years agotcp: Fix tcp_hybla zero congestion window growth with small rho and large cwnd.
Daniele Lacamera [Tue, 7 Oct 2008 22:58:17 +0000 (15:58 -0700)]
tcp: Fix tcp_hybla zero congestion window growth with small rho and large cwnd.

Because of rounding, in certain conditions, i.e. when in congestion
avoidance state rho is smaller than 1/128 of the current cwnd, TCP
Hybla congestion control starves and the cwnd is kept constant
forever.

This patch forces an increment by one segment after #send_cwnd calls
without increments(newreno behavior).

Signed-off-by: Daniele Lacamera <root@danielinux.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 years agonet: Fix netdev_run_todo dead-lock
Herbert Xu [Tue, 7 Oct 2008 22:50:03 +0000 (15:50 -0700)]
net: Fix netdev_run_todo dead-lock

Benjamin Thery tracked down a bug that explains many instances
of the error

unregister_netdevice: waiting for %s to become free. Usage count = %d

It turns out that netdev_run_todo can dead-lock with itself if
a second instance of it is run in a thread that will then free
a reference to the device waited on by the first instance.

The problem is really quite silly.  We were trying to create
parallelism where none was required.  As netdev_run_todo always
follows a RTNL section, and that todo tasks can only be added
with the RTNL held, by definition you should only need to wait
for the very ones that you've added and be done with it.

There is no need for a second mutex or spinlock.

This is exactly what the following patch does.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 years agoMerge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/holtmann/bluet...
David S. Miller [Tue, 7 Oct 2008 22:32:20 +0000 (15:32 -0700)]
Merge branch 'master' of git://git./linux/kernel/git/holtmann/bluetooth-2.6

16 years agotcp: Fix possible double-ack w/ user dma
Ali Saidi [Tue, 7 Oct 2008 22:31:19 +0000 (15:31 -0700)]
tcp: Fix possible double-ack w/ user dma

From: Ali Saidi <saidi@engin.umich.edu>

When TCP receive copy offload is enabled it's possible that
tcp_rcv_established() will cause two acks to be sent for a single
packet. In the case that a tcp_dma_early_copy() is successful,
copied_early is set to true which causes tcp_cleanup_rbuf() to be
called early which can send an ack. Further along in
tcp_rcv_established(), __tcp_ack_snd_check() is called and will
schedule a delayed ACK. If no packets are processed before the delayed
ack timer expires the packet will be acked twice.

Signed-off-by: David S. Miller <davem@davemloft.net>
16 years agonet: only invoke dev->change_rx_flags when device is UP
Patrick McHardy [Tue, 7 Oct 2008 22:26:48 +0000 (15:26 -0700)]
net: only invoke dev->change_rx_flags when device is UP

Jesper Dangaard Brouer <hawk@comx.dk> reported a bug when setting a VLAN
device down that is in promiscous mode:

When the VLAN device is set down, the promiscous count on the real
device is decremented by one by vlan_dev_stop(). When removing the
promiscous flag from the VLAN device afterwards, the promiscous
count on the real device is decremented a second time by the
vlan_change_rx_flags() callback.

The root cause for this is that the ->change_rx_flags() callback is
invoked while the device is down. The synchronization is meant to mirror
the behaviour of the ->set_rx_mode callbacks, meaning the ->open function
is responsible for doing a full sync on open, the ->close() function is
responsible for doing full cleanup on ->stop() and ->change_rx_flags()
is meant to do incremental changes while the device is UP.

Only invoke ->change_rx_flags() while the device is UP to provide the
intended behaviour.

Tested-by: Jesper Dangaard Brouer <jdb@comx.dk>
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 years agoSLOB: fix bogus ksize calculation
Matt Mackall [Tue, 7 Oct 2008 16:37:35 +0000 (11:37 -0500)]
SLOB: fix bogus ksize calculation

SLOB's ksize calculation was braindamaged and generally harmlessly
underreported the allocation size. But for very small buffers, it could
in fact overreport them, leading code depending on krealloc to overrun
the allocation and trample other data.

Signed-off-by: Matt Mackall <mpm@selenic.com>
Tested-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agoRevert "V4L/DVB (8904): cx88: add missing unlock_kernel"
Linus Torvalds [Tue, 7 Oct 2008 14:54:34 +0000 (07:54 -0700)]
Revert "V4L/DVB (8904): cx88: add missing unlock_kernel"

This reverts commit 135aedc38e812b922aa56096f36a3d72ffbcf2fb, as
requested by Hans Verkuil.

It was a patch for 2.6.28 where the BKL was pushed down from v4l core to
the drivers, not for 2.6.27!

Requested-by: Hans Verkuil <hverkuil@xs4all.nl>
Cc: Mauro Carvalho Chehab <mchehab@infradead.org>
Signed-of-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agoLinux 2.6.27-rc9
Linus Torvalds [Mon, 6 Oct 2008 23:39:58 +0000 (16:39 -0700)]
Linux 2.6.27-rc9

16 years agoMarker depmod fix core kernel list
Mathieu Desnoyers [Mon, 6 Oct 2008 13:30:12 +0000 (09:30 -0400)]
Marker depmod fix core kernel list

* Theodore Ts'o (tytso@mit.edu) wrote:
>
> I've been playing with adding some markers into ext4 to see if they
> could be useful in solving some problems along with Systemtap.  It
> appears, though, that as of 2.6.27-rc8, markers defined in code which is
> compiled directly into the kernel (i.e., not as modules) don't show up
> in Module.markers:
>
> kvm_trace_entryexit arch/x86/kvm/kvm-intel  %u %p %u %u %u %u %u %u
> kvm_trace_handler arch/x86/kvm/kvm-intel  %u %p %u %u %u %u %u %u
> kvm_trace_entryexit arch/x86/kvm/kvm-amd  %u %p %u %u %u %u %u %u
> kvm_trace_handler arch/x86/kvm/kvm-amd  %u %p %u %u %u %u %u %u
>
> (Note the lack of any of the kernel_sched_* markers, and the markers I
> added for ext4_* and jbd2_* are missing as wel.)
>
> Systemtap apparently depends on in-kernel trace_mark being recorded in
> Module.markers, and apparently it's been claimed that it used to be
> there.  Is this a bug in systemtap, or in how Module.markers is getting
> built?   And is there a file that contains the equivalent information
> for markers located in non-modules code?

I think the problem comes from "markers: fix duplicate modpost entry"
(commit d35cb360c29956510b2fe1a953bd4968536f7216)

Especially :

  -   add_marker(mod, marker, fmt);
  +   if (!mod->skip)
  +     add_marker(mod, marker, fmt);
    }
    return;
   fail:

Here is a fix that should take care if this problem.

Thanks for the bug report!

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
Tested-by: "Theodore Ts'o" <tytso@mit.edu>
CC: Greg KH <greg@kroah.com>
CC: David Smith <dsmith@redhat.com>
CC: Roland McGrath <roland@redhat.com>
CC: Sam Ravnborg <sam@ravnborg.org>
CC: Wenji Huang <wenji.huang@oracle.com>
CC: Takashi Nishiie <t-nishiie@np.css.fujitsu.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agoMerge branch 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jwessel...
Linus Torvalds [Mon, 6 Oct 2008 21:30:02 +0000 (14:30 -0700)]
Merge branch 'for_linus' of git://git./linux/kernel/git/jwessel/linux-2.6-kgdb

* 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jwessel/linux-2.6-kgdb:
  kgdb: call touch_softlockup_watchdog on resume
  kgdb, x86: Avoid invoking kgdb_nmicallback twice per NMI

16 years agoMerge branch 'x86-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Mon, 6 Oct 2008 21:29:16 +0000 (14:29 -0700)]
Merge branch 'x86-fixes-for-linus' of git://git./linux/kernel/git/tip/linux-2.6-tip

* 'x86-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
  x86: gart iommu have direct mapping when agp is present too

16 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/bart/ide-2.6
Linus Torvalds [Mon, 6 Oct 2008 21:27:57 +0000 (14:27 -0700)]
Merge git://git./linux/kernel/git/bart/ide-2.6

* git://git.kernel.org/pub/scm/linux/kernel/git/bart/ide-2.6:
  ide: workaround for bogus gcc warning in ide_sysfs_register_port()
  ide-cd: Optiarc DVD RW AD-7200A does play audio
  IDE: Fix platform device registration in Swarm IDE driver (v2)
  ide-dma: fix ide_build_dmatable() for TRM290
  ide-cd: temporary tray close fix

16 years agoMerge branch 'upstream' of git://ftp.linux-mips.org/pub/scm/upstream-linus
Linus Torvalds [Mon, 6 Oct 2008 21:27:39 +0000 (14:27 -0700)]
Merge branch 'upstream' of git://ftp.linux-mips.org/upstream-linus

* 'upstream' of git://ftp.linux-mips.org/pub/scm/upstream-linus:
  [MIPS] IP27: Fix build errors if CONFIG_MAPPED_KERNEL=y
  [MIPS] Fix CMP Kconfig configuration and mark as broken.

16 years agoMerge branch 'fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/mchehab/v4l-dvb
Linus Torvalds [Mon, 6 Oct 2008 21:27:15 +0000 (14:27 -0700)]
Merge branch 'fixes' of git://git./linux/kernel/git/mchehab/v4l-dvb

* 'fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/mchehab/v4l-dvb: (33 commits)
  V4L/DVB (9103): em28xx: HVR-900 B3C0 - fix audio clicking issue
  V4L/DVB (9099): em28xx: Add detection for K-WORLD DVB-T 310U
  V4L/DVB (9092): gspca: Bad init values for sonixj ov7660.
  V4L/DVB (9080): gspca: Add a delay after writing to the sonixj sensors.
  V4L/DVB (9075): gspca: Bad check of returned status in i2c_read() spca561.
  V4L/DVB (9053): fix buffer overflow in uvc-video
  V4L/DVB (9043): S5H1420: Fix size of shadow-array to avoid overflow
  V4L/DVB (9037): Fix support for Hauppauge Nova-S SE
  V4L/DVB (9029): Fix deadlock in demux code
  V4L/DVB (8979): sms1xxx: Add new USB product ID for Hauppauge WinTV MiniStick
  V4L/DVB (8978): sms1xxx: fix product name for Hauppauge WinTV MiniStick
  V4L/DVB (8967): Use correct XC3028L firmware for AMD ATI TV Wonder 600
  V4L/DVB (8963): s2255drv field count fix
  V4L/DVB (8961): zr36067: Fix RGBR pixel format
  V4L/DVB (8960): drivers/media/video/cafe_ccic.c needs mm.h
  V4L/DVB (8958): zr36067: Return proper bytes-per-line value
  V4L/DVB (8957): zr36067: Restore the default pixel format
  V4L/DVB (8955): bttv: Prevent NULL pointer dereference in radio_open
  V4L/DVB (8935): em28xx-cards: Remove duplicate entry (EM2800_BOARD_KWORLD_USB2800)
  V4L/DVB (8933): gspca: Disable light frquency for zc3xx cs2102 Kokom.
  ...

16 years agoatmel-mci: Initialize BLKR before sending data transfer command
Haavard Skinnemoen [Mon, 6 Oct 2008 18:09:53 +0000 (20:09 +0200)]
atmel-mci: Initialize BLKR before sending data transfer command

The atmel-mci driver sometimes fails data transfers like this:

   mmcblk0: error -5 transferring data
   end_request: I/O error, dev mmcblk0, sector 2749769
   end_request: I/O error, dev mmcblk0, sector 2749777

It turns out that this might be caused by the BLKR register (which
contains the block size and the number of blocks being transfered) being
initialized too late. This patch moves the initialization of BLKR so
that it contains the correct value before the block transfer command is
sent.

This error is difficult to reproduce, but if you insert a long delay
(mdelay(10) or thereabouts) between the calls to atmci_start_command()
and atmci_submit_data(), all transfers seem to fail without this patch,
while I haven't seen any failures with this patch.

Reported-by: Hein_Tibosch <hein_tibosch@yahoo.es>
Signed-off-by: Haavard Skinnemoen <haavard.skinnemoen@atmel.com>
Signed-off-by: Pierre Ossman <drzeus@drzeus.cx>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years agonetrom: Fix sock_orphan() use in nr_release
Jarek Poplawski [Mon, 6 Oct 2008 19:54:57 +0000 (12:54 -0700)]
netrom: Fix sock_orphan() use in nr_release

While debugging another bug it was found that NetRom socks
are sometimes seen unorphaned in sk_free(). This patch moves
sock_orphan() in nr_release() to the beginning (like in ax25,
or rose).

Reported-and-tested-by: Bernard Pidoux f6bvp <f6bvp@free.fr>
Signed-off-by: Jarek Poplawski <jarkao2@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
16 years agoax25: Quick fix for making sure unaccepted sockets get destroyed.
David S. Miller [Mon, 6 Oct 2008 19:53:50 +0000 (12:53 -0700)]
ax25: Quick fix for making sure unaccepted sockets get destroyed.

Since we reverted 30902dc3cb0ea1cfc7ac2b17bcf478ff98420d74 ("ax25: Fix
std timer socket destroy handling.") we have to put some kind of fix
in to cure the issue whereby unaccepted connections do not get destroyed.

The approach used here is from Tihomir Heidelberg - 9a4gl

Signed-off-by: David S. Miller <davem@davemloft.net>
16 years agoRevert "ax25: Fix std timer socket destroy handling."
David S. Miller [Mon, 6 Oct 2008 19:48:29 +0000 (12:48 -0700)]
Revert "ax25: Fix std timer socket destroy handling."

This reverts commit 30902dc3cb0ea1cfc7ac2b17bcf478ff98420d74.

It causes all kinds of problems, based upon a report by
Bernard (f6bvp) and analysis by Jarek Poplawski.

Signed-off-by: David S. Miller <davem@davemloft.net>
16 years agokgdb: call touch_softlockup_watchdog on resume
Jason Wessel [Mon, 6 Oct 2008 18:50:59 +0000 (13:50 -0500)]
kgdb: call touch_softlockup_watchdog on resume

The softlockup watchdog needs to be touched when resuming the from the
kgdb stopped state to avoid the printk that a CPU is stuck if the
debugger was active for longer than the softlockup threshold.

Signed-off-by: Jason Wessel <jason.wessel@windriver.com>
16 years agokgdb, x86: Avoid invoking kgdb_nmicallback twice per NMI
Jan Kiszka [Mon, 6 Oct 2008 18:50:59 +0000 (13:50 -0500)]
kgdb, x86: Avoid invoking kgdb_nmicallback twice per NMI

Stress-testing KVM's latest NMI support with kgdbts inside an SMP guest,
I came across spurious unhandled NMIs while running the singlestep test.
Looking closer at the code path each NMI takes when KGDB is enabled, I
noticed that kgdb_nmicallback is called twice per event: One time via
DIE_NMI_IPI notification, the second time on DIE_NMI. Removing the first
invocation cures the unhandled NMIs here.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Jason Wessel <jason.wessel@windriver.com>
16 years agox86 ACPI: Blacklist two HP machines with buggy BIOSes
Rafael J. Wysocki [Mon, 6 Oct 2008 09:59:29 +0000 (11:59 +0200)]
x86 ACPI: Blacklist two HP machines with buggy BIOSes

There is a bug in the BIOSes of some HP boxes with AMD Turions which
connects IO-APIC pins with ACPI thermal trip points in such a way that
if the state of the IO-APIC is not as expected by the (buggy) BIOS, the
thermal trip points are set to insanely low values (usually all of them
become 16 degrees Celsius).  As a result, thermal throttling kicks in
and knock the system down to its shoes.

Unfortunately some of the recent IO-APIC changes made the bug show up.
To prevent this from happening, blacklist machines that are known to be
affected (nx6115 and 6715b in this particular case).

This fixes http://bugzilla.kernel.org/show_bug.cgi?id=11516 listed as
a regression from 2.6.26.

On my box it was caused by:

commit 691874fa96d6349a8b60f8ea9c2bae52ece79941
Author: Maciej W. Rozycki <macro@linux-mips.org>
Date:   Tue May 27 21:19:51 2008 +0100

    x86: I/O APIC: timer through 8259A second-chance

Signed-off-by: Maciej W. Rozycki <macro@linux-mips.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
and the whole story is described in this (huge) thread:

    http://marc.info/?l=linux-kernel&m=121358440508410&w=4

Matthew Garrett told us about that happening on the nx6125:

    http://marc.info/?l=linux-kernel&m=121396307411930&w=4

and then Maciej analysed the breakage on the basis of a DSDT from the
nx6325:

    http://marc.info/?l=linux-kernel&m=121401068718826&w=4

As far as the Dmitry's and Jason's boxes are concerned, I recognized the
symptoms and asked them to verify that the blacklisting helped.

It appears that the buggy BIOS code has been copy-pasted to the entire
range of machines, for no good reason.

Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Tested-by: Dmitry Torokhov <dmitry.torokhov@gmail.com>
Tested-by: Jason Vas Dias <jason.vas.dias@gmail.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
16 years ago[Bluetooth] Add reset quirk for A-Link BlueUSB21 dongle
Marcel Holtmann [Mon, 6 Oct 2008 10:22:52 +0000 (12:22 +0200)]
[Bluetooth] Add reset quirk for A-Link BlueUSB21 dongle

The new A-Link Bluetooth dongle is another one based on the BCM2046 chip
from Broadcom and it also needs to send HCI_Reset before it becomes fully
operational. Without the quirk it will show a lot of I/O errors.

Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
16 years ago[Bluetooth] Add reset quirk for new Targus and Belkin dongles
Marcel Holtmann [Mon, 6 Oct 2008 10:22:51 +0000 (12:22 +0200)]
[Bluetooth] Add reset quirk for new Targus and Belkin dongles

Targus and Belkin have come out with new Bluetooth 2.1 capable dongles
using the latest BCM2046 chip from Broadcom. Both of them are so called
HID proxy dongles and they need to send HCI_Reset before they become
fully operational.

Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
16 years ago[Bluetooth] Fix double frees on error paths of btusb and bpa10x drivers
Marcel Holtmann [Mon, 6 Oct 2008 10:22:51 +0000 (12:22 +0200)]
[Bluetooth] Fix double frees on error paths of btusb and bpa10x drivers

The transfer buffer of an URB will be automatically freed when using
the URB_FREE_BUFFER transfer_flag. So the extra calls to kfree() will
cause a double free.

Reported-by: Justin Mattock <justinmattock@gmail.com>
Signed-off-by: Rabin Vincent <rabin@rab.in>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
16 years ago[MIPS] IP27: Fix build errors if CONFIG_MAPPED_KERNEL=y
Ralf Baechle [Fri, 3 Oct 2008 21:43:38 +0000 (22:43 +0100)]
[MIPS] IP27: Fix build errors if CONFIG_MAPPED_KERNEL=y

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
16 years ago[MIPS] Fix CMP Kconfig configuration and mark as broken.
Ralf Baechle [Fri, 3 Oct 2008 23:06:29 +0000 (00:06 +0100)]
[MIPS] Fix CMP Kconfig configuration and mark as broken.

Because sync-r4k.c doesn't build.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
16 years agoide: workaround for bogus gcc warning in ide_sysfs_register_port()
Bartlomiej Zolnierkiewicz [Sun, 5 Oct 2008 16:23:28 +0000 (18:23 +0200)]
ide: workaround for bogus gcc warning in ide_sysfs_register_port()

Reported-by: "Steven Noonan" <steven@uplinklabs.net>
Suggested-by: "Elias Oltmanns" <eo@nebensachen.de>
Cc: mingo@elte.hu
Signed-off-by: Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>
16 years agoide-cd: Optiarc DVD RW AD-7200A does play audio
Bodo Eggert [Sun, 5 Oct 2008 16:23:28 +0000 (18:23 +0200)]
ide-cd: Optiarc DVD RW AD-7200A does play audio

The Optiarc DVD RW AD-7200A can play audio, but tells it could not.

Signed-off-by: Bodo Eggert <7eggert@gmx.de>
Tested-by: Nick Warne <nick@ukfsn.org>
Received-from: Borislav Petkov <petkovbb@googlemail.com>
[bart: keep "audio" quirks together]
Signed-off-by: Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>
16 years agoIDE: Fix platform device registration in Swarm IDE driver (v2)
Ralf Baechle [Sun, 5 Oct 2008 16:23:28 +0000 (18:23 +0200)]
IDE: Fix platform device registration in Swarm IDE driver (v2)

The Swarm IDE driver uses a release method which is defined in the driver
itself thus potentially oopsable.  The simple fix would be to just leak
the device but this patch goes the full length and moves the entire
handling of the platform device in the platform code and retains only
the platform driver code in drivers/ide/mips/swarm.c.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Cc: "Maciej W. Rozycki" <macro@linux-mips.org>
Acked-by: Sergei Shtylyov <sshtylyov@ru.mvista.com>
[bart: remove no longer needed BLK_DEV_IDE_SWARM from ide/Kconfig]
Signed-off-by: Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>
16 years agoide-dma: fix ide_build_dmatable() for TRM290
Sergei Shtylylov [Sun, 5 Oct 2008 16:23:27 +0000 (18:23 +0200)]
ide-dma: fix ide_build_dmatable() for TRM290

Apparently, 'xcount' being 0 does not mean 0 bytes for TRM290; it means 4 bytes,
judging from the code immediately preceding this check.  So, we must never try
to "split" the PRD for TRM290.

This is probably never hit anyway -- with the DMA buffers aligned to at least
512 bytes and ATAPI DMA not being used for non block I/O commands...

Signed-off-by: Sergei Shtylyov <sshtylyov@ru.mvista.com>
Signed-off-by: Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>
16 years agoide-cd: temporary tray close fix
Borislav Petkov [Sun, 5 Oct 2008 16:23:27 +0000 (18:23 +0200)]
ide-cd: temporary tray close fix

This one fixes http://bugzilla.kernel.org/show_bug.cgi?id=11602.

A more generic fix for drives which cannot autoclose tray will follow.

Signed-off-by: Borislav Petkov <petkovbb@gmail.com>
Cc: Jens Axboe <jens.axboe@oracle.com>
[bart: add an extra parentheses for consistency with the rest of kernel code]
Signed-off-by: Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>
16 years agox86: gart iommu have direct mapping when agp is present too
Yinghai Lu [Sat, 4 Oct 2008 22:55:12 +0000 (15:55 -0700)]
x86: gart iommu have direct mapping when agp is present too

move init_memory_mapping() out of init_k8_gatt.

for: http://bugzilla.kernel.org/show_bug.cgi?id=11676
    2.6.27-rc2 to rc8, apgart fails, iommu=soft works, regression

This is needed because we need to map the GART aperture even
if the GATT is not initialized.

Signed-off-by: Yinghai Lu <yhlu.kernel@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
16 years agoV4L/DVB (9103): em28xx: HVR-900 B3C0 - fix audio clicking issue
Wiktor Grebla [Sun, 28 Sep 2008 04:37:48 +0000 (01:37 -0300)]
V4L/DVB (9103): em28xx: HVR-900 B3C0 - fix audio clicking issue

Fixed audio clicking problem which could be heard when using analog tv or composite input

Signed-off-by: Wiktor Grebla <greblus@gmail.com>
Signed-off-by: Douglas Schilling Landgraf <dougsland@linuxtv.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
16 years agoV4L/DVB (9099): em28xx: Add detection for K-WORLD DVB-T 310U
Darron Broad [Tue, 30 Sep 2008 05:46:41 +0000 (02:46 -0300)]
V4L/DVB (9099): em28xx: Add detection for K-WORLD DVB-T 310U

Correct firmware type to MTS
Correct audio routing for composite/s-video
Add DVB-T detection.

This patch uses the eeprom hash method for detection as the vendor/product
ids are also used for the DIGIVOX_AD. This may be a clone of the same
product. Explanatory text has been added prior to the hask look-up in
anticipation that it may help others.

The following has been tested to work:
Analogue TV (PAL-I)
Composite In
DVB-T (UK Crystal Palace)
USB AUDIO

The following has not been tested but probably works:
S-Video In

Signed-off-by: Darron Broad <darron@kewl.org>
Signed-off-by: Douglas Schilling Landgraf <dougsland@linuxtv.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
16 years agoV4L/DVB (9092): gspca: Bad init values for sonixj ov7660.
Jean-Francois Moine [Thu, 2 Oct 2008 11:06:59 +0000 (08:06 -0300)]
V4L/DVB (9092): gspca: Bad init values for sonixj ov7660.

Signed-off-by: Jean-Francois Moine <moinejf@free.fr>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
16 years agoV4L/DVB (9080): gspca: Add a delay after writing to the sonixj sensors.
Jean-Francois Moine [Sun, 21 Sep 2008 06:28:55 +0000 (03:28 -0300)]
V4L/DVB (9080): gspca: Add a delay after writing to the sonixj sensors.

Signed-off-by: Jean-Francois Moine <moinejf@free.fr>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
16 years agoV4L/DVB (9075): gspca: Bad check of returned status in i2c_read() spca561.
Shane [Sun, 14 Sep 2008 07:17:46 +0000 (04:17 -0300)]
V4L/DVB (9075): gspca: Bad check of returned status in i2c_read() spca561.

This makes auto gain functional on 04fc:0561.

Signed-off-by: Shane <gnome42@gmail.com>
Signed-off-by: Jean-Francois Moine <moinejf@free.fr>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
16 years agoV4L/DVB (9053): fix buffer overflow in uvc-video
Ralph Loader [Tue, 23 Sep 2008 00:06:48 +0000 (21:06 -0300)]
V4L/DVB (9053): fix buffer overflow in uvc-video

There is a buffer overflow in drivers/media/video/uvc/uvc_ctrl.c:

INFO: 0xf2c5ce08-0xf2c5ce0b. First byte 0xa1 instead of 0xcc
INFO: Allocated in uvc_query_v4l2_ctrl+0x3c/0x239 [uvcvideo] age=13 cpu=1 pid=4975
...

A fixed size 8-byte buffer is allocated, and a variable size field is read
into it; there is no particular bound on the size of the field (it is
dependent on hardware and configuration) and it can overflow [also
verified by inserting printk's.]

The patch attempts to size the buffer to the correctly.

Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Laurent Pinchart <laurent.pinchart@skynet.be>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
16 years agoV4L/DVB (9043): S5H1420: Fix size of shadow-array to avoid overflow
Patrick Boettcher [Sun, 7 Sep 2008 19:04:38 +0000 (16:04 -0300)]
V4L/DVB (9043): S5H1420: Fix size of shadow-array to avoid overflow

The array size of 'shadow' still needs to be fixed in order to not overflow when reading register 0x00.

Thanks to Oliver Endriss for pointing that out.

Signed-off-by: Patrick Boettcher <pb@linuxtv.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
16 years agoV4L/DVB (9037): Fix support for Hauppauge Nova-S SE
Patrick Boettcher [Sat, 6 Sep 2008 16:31:58 +0000 (13:31 -0300)]
V4L/DVB (9037): Fix support for Hauppauge Nova-S SE

Different backends have different input busses (saa7146, flexcop).
To reflect that a config-option to the s5h1420-driver was added which makes
the output mode selectable.

Furthermore the s5h1420-driver is now doing the same i2c-method as it was done
before adding support for other i2c-users.

This patch needs to go into the current release of the kernel, as this driver
is currently broken.

(Thanks to Eberhard Kaltenhaeuser for helping out to debug this issue.)

Signed-off-by: Patrick Boettcher <pb@linuxtv.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
16 years agoV4L/DVB (9029): Fix deadlock in demux code
Andreas Oberritter [Wed, 24 Sep 2008 08:00:37 +0000 (05:00 -0300)]
V4L/DVB (9029): Fix deadlock in demux code

The functions dvb_dmxdev_section_callback, dvb_dmxdev_ts_callback,
dvb_dmx_swfilter_packet, dvb_dmx_swfilter_packets, dvb_dmx_swfilter and
dvb_dmx_swfilter_204 may be called from both interrupt and process
context. Therefore they need to be protected by spin_lock_irqsave()
instead of spin_lock().

This fixes a deadlock discovered by lockdep.

Signed-off-by: Andreas Oberritter <obi@linuxtv.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
16 years agoV4L/DVB (8979): sms1xxx: Add new USB product ID for Hauppauge WinTV MiniStick
Michael Krufky [Tue, 16 Sep 2008 19:46:42 +0000 (16:46 -0300)]
V4L/DVB (8979): sms1xxx: Add new USB product ID for Hauppauge WinTV MiniStick

2040:5510 is the same hardware as 2040:5500

Signed-off-by: Michael Krufky <mkrufky@linuxtv.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
16 years agoV4L/DVB (8978): sms1xxx: fix product name for Hauppauge WinTV MiniStick
Michael Krufky [Fri, 12 Sep 2008 02:02:12 +0000 (23:02 -0300)]
V4L/DVB (8978): sms1xxx: fix product name for Hauppauge WinTV MiniStick

Signed-off-by: Michael Krufky <mkrufky@linuxtv.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
16 years agoV4L/DVB (8967): Use correct XC3028L firmware for AMD ATI TV Wonder 600
Devin Heitmueller [Tue, 16 Sep 2008 05:02:12 +0000 (02:02 -0300)]
V4L/DVB (8967): Use correct XC3028L firmware for AMD ATI TV Wonder 600

The AMD ATI TV Wonder 600 has an XC3028L and *not* an XC3028, so we need to
load the proper firmware to prevent the device from overheating.

Signed-off-by: Devin Heitmueller <devin.heitmueller@gmail.com>
Signed-off-by: Michael Krufky <mkrufky@linuxtv.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
16 years agoV4L/DVB (8963): s2255drv field count fix
Dean Anderson [Tue, 9 Sep 2008 15:29:56 +0000 (12:29 -0300)]
V4L/DVB (8963): s2255drv field count fix

Fixes videobuf field_count

Signed-off-by: Dean Anderson <dean@sensoray.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
16 years agoV4L/DVB (8961): zr36067: Fix RGBR pixel format
Jean Delvare [Fri, 5 Sep 2008 13:39:27 +0000 (10:39 -0300)]
V4L/DVB (8961): zr36067: Fix RGBR pixel format

The zr36067 driver is improperly declaring pixel format RGBP twice,
once as "16-bit RGB LE" and once as "16-bit RGB BE". The latter is
actually RGBR. Fix the code to properly map both pixel formats.

Signed-off-by: Jean Delvare <khali@linux-fr.org>
Acked-by: Trent Piepho <xyzzy@speakeasy.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
16 years agoV4L/DVB (8960): drivers/media/video/cafe_ccic.c needs mm.h
Andrew Morton [Wed, 3 Sep 2008 05:15:39 +0000 (02:15 -0300)]
V4L/DVB (8960): drivers/media/video/cafe_ccic.c needs mm.h

sparc32 allmodconfig:

drivers/media/video/cafe_ccic.c: In function 'cafe_setup_siobuf':
drivers/media/video/cafe_ccic.c:1192: error: implicit declaration of function 'PAGE_ALIGN'
drivers/media/video/cafe_ccic.c: At top level:
drivers/media/video/cafe_ccic.c:1430: error: variable 'cafe_v4l_vm_ops' has initializer but incomplete type
drivers/media/video/cafe_ccic.c:1431: error: unknown field 'open' specified in initializer
drivers/media/video/cafe_ccic.c:1431: warning: excess elements in struct initializer
drivers/media/video/cafe_ccic.c:1431: warning: (near initialization for 'cafe_v4l_vm_ops')
drivers/media/video/cafe_ccic.c:1432: error: unknown field 'close' specified in initializer
drivers/media/video/cafe_ccic.c:1433: warning: excess elements in struct initializer
drivers/media/video/cafe_ccic.c:1433: warning: (near initialization for 'cafe_v4l_vm_ops')
drivers/media/video/cafe_ccic.c: In function 'cafe_v4l_mmap':
drivers/media/video/cafe_ccic.c:1444: error: 'VM_WRITE' undeclared (first use in this function)
drivers/media/video/cafe_ccic.c:1444: error: (Each undeclared identifier is reported only once
drivers/media/video/cafe_ccic.c:1444: error: for each function it appears in.)
drivers/media/video/cafe_ccic.c:1444: error: 'VM_SHARED' undeclared (first use in this function)
drivers/media/video/cafe_ccic.c:1461: error: 'VM_DONTEXPAND' undeclared (first use in this function)

This build breakage is caused by some header file shuffle in linux-next.  But
I suggest that this patch be merged ahead of linux-next to avoid bisection
breakage.

Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
16 years agoV4L/DVB (8958): zr36067: Return proper bytes-per-line value
Jean Delvare [Sun, 7 Sep 2008 08:56:55 +0000 (05:56 -0300)]
V4L/DVB (8958): zr36067: Return proper bytes-per-line value

The zr36067 driver should return the actual bytes-per-line value when
queried with ioctl VIDIOC_G_FMT, instead of 0. Otherwise user-space
applications can get confused.

Likewise, with ioctl VIDIOC_S_FMT, we are supposed to fill the
bytes-per-line value. And we shouldn't fail if the caller sets the
initial value to something different from 0. This is perfectly valid
for applications to pre-fill this field with the value they expect.

Signed-off-by: Jean Delvare <khali@linux-fr.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
16 years agoV4L/DVB (8957): zr36067: Restore the default pixel format
Jean Delvare [Sun, 7 Sep 2008 08:21:34 +0000 (05:21 -0300)]
V4L/DVB (8957): zr36067: Restore the default pixel format

Restore the default pixel format to YUYV as it used to be before
kernel 2.6.23. It was accidentally changed to BGR3 by commit
603d6f2c8f9f3604f9c6c1f8903efc2df30a000f.

Signed-off-by: Jean Delvare <khali@linux-fr.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
16 years agoV4L/DVB (8955): bttv: Prevent NULL pointer dereference in radio_open
Jean Delvare [Sun, 7 Sep 2008 16:47:03 +0000 (13:47 -0300)]
V4L/DVB (8955): bttv: Prevent NULL pointer dereference in radio_open

Fix the following crash in the bttv driver:

BUG: unable to handle kernel NULL pointer dereference at 000000000000036c
IP: [<ffffffffa037860a>] radio_open+0x3a/0x170 [bttv]

This happens because radio_open assumes that all present bttv devices
have a radio function. If a bttv device without radio and one with
radio are installed on the same system, and the one without radio is
registered first, then radio_open checks for the radio device number
of a bttv device that has no radio function, and this breaks. All we
have to do to fix it is to skip bttv devices without a radio function.

Signed-off-by: Jean Delvare <khali@linux-fr.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
16 years agoV4L/DVB (8935): em28xx-cards: Remove duplicate entry (EM2800_BOARD_KWORLD_USB2800)
Douglas Schilling Landgraf [Mon, 8 Sep 2008 06:22:55 +0000 (03:22 -0300)]
V4L/DVB (8935): em28xx-cards: Remove duplicate entry (EM2800_BOARD_KWORLD_USB2800)

Removed duplicated entry for EM2800_BOARD_KWORLD_USB2800

Signed-off-by: Douglas Schilling Landgraf <dougsland@linuxtv.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>