firefly-linux-kernel-4.4.55.git
14 years agoxfs: Extend project quotas to support 32bit project ids
Arkadiusz Mi?kiewicz [Sun, 26 Sep 2010 06:10:18 +0000 (06:10 +0000)]
xfs: Extend project quotas to support 32bit project ids

This patch adds support for 32bit project quota identifiers.

On disk format is backward compatible with 16bit projid numbers. projid
on disk is now kept in two 16bit values - di_projid_lo (which holds the
same position as old 16bit projid value) and new di_projid_hi (takes
existing padding) and converts from/to 32bit value on the fly.

xfs_admin (for existing fs), mkfs.xfs (for new fs) needs to be used
to enable PROJID32BIT support.

Signed-off-by: Arkadiusz Miƛkiewicz <arekm@maven.pl>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Alex Elder <aelder@sgi.com>
14 years agoxfs: remove xfs_buf wrappers
Christoph Hellwig [Wed, 6 Oct 2010 18:41:18 +0000 (18:41 +0000)]
xfs: remove xfs_buf wrappers

Stop having two different names for many buffer functions and use
the more descriptive xfs_buf_* names directly.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Alex Elder <aelder@sgi.com>
14 years agoxfs: remove xfs_cred.h
Christoph Hellwig [Wed, 6 Oct 2010 18:41:17 +0000 (18:41 +0000)]
xfs: remove xfs_cred.h

We're not actually passing around credentials inside XFS for a while
now, so remove all xfs_cred.h with it's cred_t typedef and all
instances of it.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Alex Elder <aelder@sgi.com>
14 years agoxfs: remove xfs_globals.h
Christoph Hellwig [Wed, 6 Oct 2010 18:41:16 +0000 (18:41 +0000)]
xfs: remove xfs_globals.h

This header only provides one extern that isn't actually declared
anywhere, and shadowed by a macro.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Alex Elder <aelder@sgi.com>
14 years agoxfs: remove xfs_version.h
Christoph Hellwig [Wed, 6 Oct 2010 18:41:15 +0000 (18:41 +0000)]
xfs: remove xfs_version.h

It used to have a place when it contained an automatically generated
CVS version, but these days it's entirely superflous.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Alex Elder <aelder@sgi.com>
14 years agoxfs: remove xfs_refcache.h
Christoph Hellwig [Wed, 6 Oct 2010 18:41:14 +0000 (18:41 +0000)]
xfs: remove xfs_refcache.h

This header has been completely unused for a couple of years.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Alex Elder <aelder@sgi.com>
14 years agoxfs: fix the xfs_trans_committed
Christoph Hellwig [Wed, 6 Oct 2010 18:41:13 +0000 (18:41 +0000)]
xfs: fix the xfs_trans_committed

Use the correct prototype for xfs_trans_committed instead of casting it.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Alex Elder <aelder@sgi.com>
14 years agoxfs: remove unused t_callback field in struct xfs_trans
Christoph Hellwig [Wed, 6 Oct 2010 18:41:12 +0000 (18:41 +0000)]
xfs: remove unused t_callback field in struct xfs_trans

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Alex Elder <aelder@sgi.com>
14 years agoxfs: fix bogus m_maxagi check in xfs_iget
Christoph Hellwig [Wed, 6 Oct 2010 18:31:23 +0000 (18:31 +0000)]
xfs: fix bogus m_maxagi check in xfs_iget

These days inode64 should only control which AGs we allocate new
inodes from, while we still try to support reading all existing
inodes.  To make this actually work the check ontop of xfs_iget
needs to be relaxed to allow inodes in all allocation groups instead
of just those that we allow allocating inodes from.  Note that we
can't simply remove the check - it prevents us from accessing
invalid data when fed invalid inode numbers from NFS or bulkstat.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Alex Elder <aelder@sgi.com>
14 years agoxfs: do not use xfs_mod_incore_sb_batch for per-cpu counters
Christoph Hellwig [Thu, 30 Sep 2010 02:25:56 +0000 (02:25 +0000)]
xfs: do not use xfs_mod_incore_sb_batch for per-cpu counters

Update the per-cpu counters manually in xfs_trans_unreserve_and_mod_sb
and remove support for per-cpu counters from xfs_mod_incore_sb_batch
to simplify it.  And added benefit is that we don't have to take
m_sb_lock for transactions that only modify per-cpu counters.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Alex Elder <aelder@sgi.com>
14 years agoxfs: do not use xfs_mod_incore_sb for per-cpu counters
Christoph Hellwig [Thu, 30 Sep 2010 02:25:55 +0000 (02:25 +0000)]
xfs: do not use xfs_mod_incore_sb for per-cpu counters

Export xfs_icsb_modify_counters and always use it for modifying
the per-cpu counters.  Remove support for per-cpu counters from
xfs_mod_incore_sb to simplify it.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Alex Elder <aelder@sgi.com>
14 years agoxfs: remove XFS_MOUNT_NO_PERCPU_SB
Christoph Hellwig [Thu, 30 Sep 2010 02:25:54 +0000 (02:25 +0000)]
xfs: remove XFS_MOUNT_NO_PERCPU_SB

Fail the mount if we can't allocate memory for the per-CPU counters.
This is consistent with how we handle everything else in the mount
path and makes the superblock counter modification a lot simpler.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Alex Elder <aelder@sgi.com>
14 years agoxfs: pack xfs_buf structure more tightly
Dave Chinner [Fri, 24 Sep 2010 09:59:15 +0000 (19:59 +1000)]
xfs: pack xfs_buf structure more tightly

pahole reports the struct xfs_buf has quite a few holes in it, so
packing the structure better will reduce the size of it by 16 bytes.
Also, move all the fields used in cache lookups into the first
cacheline.

Before on x86_64:

        /* size: 320, cachelines: 5 */
/* sum members: 298, holes: 6, sum holes: 22 */

After on x86_64:

        /* size: 304, cachelines: 5 */
/* padding: 6 */
/* last cacheline: 48 bytes */

Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Alex Elder <aelder@sgi.com>
14 years agoxfs: convert buffer cache hash to rbtree
Dave Chinner [Fri, 24 Sep 2010 09:59:04 +0000 (19:59 +1000)]
xfs: convert buffer cache hash to rbtree

The buffer cache hash is showing typical hash scalability problems.
In large scale testing the number of cached items growing far larger
than the hash can efficiently handle. Hence we need to move to a
self-scaling cache indexing mechanism.

I have selected rbtrees for indexing becuse they can have O(log n)
search scalability, and insert and remove cost is not excessive,
even on large trees. Hence we should be able to cache large numbers
of buffers without incurring the excessive cache miss search
penalties that the hash is imposing on us.

To ensure we still have parallel access to the cache, we need
multiple trees. Rather than hashing the buffers by disk address to
select a tree, it seems more sensible to separate trees by typical
access patterns. Most operations use buffers from within a single AG
at a time, so rather than searching lots of different lists,
separate the buffer indexes out into per-AG rbtrees. This means that
searches during metadata operation have a much higher chance of
hitting cache resident nodes, and that updates of the tree are less
likely to disturb trees being accessed on other CPUs doing
independent operations.

Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Alex Elder <aelder@sgi.com>
14 years agoxfs: serialise inode reclaim within an AG
Dave Chinner [Mon, 27 Sep 2010 01:09:51 +0000 (11:09 +1000)]
xfs: serialise inode reclaim within an AG

Memory reclaim via shrinkers has a terrible habit of having N+M
concurrent shrinker executions (N = num CPUs, M = num kswapds) all
trying to shrink the same cache. When the cache they are all working
on is protected by a single spinlock, massive contention an
slowdowns occur.

Wrap the per-ag inode caches with a reclaim mutex to serialise
reclaim access to the AG. This will block concurrent reclaim in each
AG but still allow reclaim to scan multiple AGs concurrently. Allow
shrinkers to move on to the next AG if it can't get the lock, and if
we can't get any AG, then start blocking on locks.

To prevent reclaimers from continually scanning the same inodes in
each AG, add a cursor that tracks where the last reclaim got up to
and start from that point on the next reclaim. This should avoid
only ever scanning a small number of inodes at the satart of each AG
and not making progress. If we have a non-shrinker based reclaim
pass, ignore the cursor and reset it to zero once we are done.

Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Alex Elder <aelder@sgi.com>
14 years agoxfs: batch inode reclaim lookup
Dave Chinner [Fri, 24 Sep 2010 09:51:50 +0000 (19:51 +1000)]
xfs: batch inode reclaim lookup

Batch and optimise the per-ag inode lookup for reclaim to minimise
scanning overhead. This involves gang lookups on the radix trees to
get multiple inodes during each tree walk, and tighter validation of
what inodes can be reclaimed without blocking befor we take any
locks.

This is based on ideas suggested in a proof-of-concept patch
posted by Nick Piggin.

Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Alex Elder <aelder@sgi.com>
14 years agoxfs: implement batched inode lookups for AG walking
Dave Chinner [Tue, 28 Sep 2010 02:28:19 +0000 (12:28 +1000)]
xfs: implement batched inode lookups for AG walking

With the reclaim code separated from the generic walking code, it is
simple to implement batched lookups for the generic walk code.
Separate out the inode validation from the execute operations and
modify the tree lookups to get a batch of inodes at a time.

Reclaim operations will be optimised separately.

Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Alex Elder <aelder@sgi.com>
14 years agoxfs: split out inode walk inode grabbing
Dave Chinner [Tue, 28 Sep 2010 02:28:06 +0000 (12:28 +1000)]
xfs: split out inode walk inode grabbing

When doing read side inode cache walks, the code to validate and
grab an inode is common to all callers. Split it out of the execute
callbacks in preparation for batching lookups. Similarly, split out
the inode reference dropping from the execute callbacks into the
main lookup look to be symmetric with the grab.

Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Alex Elder <aelder@sgi.com>
14 years agoxfs: split inode AG walking into separate code for reclaim
Dave Chinner [Fri, 24 Sep 2010 08:40:15 +0000 (18:40 +1000)]
xfs: split inode AG walking into separate code for reclaim

The reclaim walk requires different locking and has a slightly
different walk algorithm, so separate it out so that it can be
optimised separately.

Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Alex Elder <aelder@sgi.com>
14 years agoxfs: remove buftarg hash for external devices
Dave Chinner [Wed, 22 Sep 2010 00:47:20 +0000 (10:47 +1000)]
xfs: remove buftarg hash for external devices

For RT and external log devices, we never use hashed buffers on them
now.  Remove the buftarg hash tables that are set up for them.

Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Alex Elder <aelder@sgi.com>
14 years agoxfs: use unhashed buffers for size checks
Dave Chinner [Wed, 22 Sep 2010 00:47:20 +0000 (10:47 +1000)]
xfs: use unhashed buffers for size checks

When we are checking we can access the last block of each device, we
do not need to use cached buffers as they will be tossed away
immediately. Use uncached buffers for size checks so that all IO
prior to full in-memory structure initialisation does not use the
buffer cache.

Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Alex Elder <aelder@sgi.com>
14 years agoxfs: kill XBF_FS_MANAGED buffers
Dave Chinner [Wed, 22 Sep 2010 00:47:20 +0000 (10:47 +1000)]
xfs: kill XBF_FS_MANAGED buffers

Filesystem level managed buffers are buffers that have their
lifecycle controlled by the filesystem layer, not the buffer cache.
We currently cache these buffers, which makes cleanup and cache
walking somewhat troublesome. Convert the fs managed buffers to
uncached buffers obtained by via xfs_buf_get_uncached(), and remove
the XBF_FS_MANAGED special cases from the buffer cache.

Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Alex Elder <aelder@sgi.com>
14 years agoxfs: store xfs_mount in the buftarg instead of in the xfs_buf
Dave Chinner [Wed, 22 Sep 2010 00:47:20 +0000 (10:47 +1000)]
xfs: store xfs_mount in the buftarg instead of in the xfs_buf

Each buffer contains both a buftarg pointer and a mount pointer. If
we add a mount pointer into the buftarg, we can avoid needing the
b_mount field in every buffer and grab it from the buftarg when
needed instead. This shrinks the xfs_buf by 8 bytes.

Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Alex Elder <aelder@sgi.com>
14 years agoxfs: introduced uncached buffer read primitve
Dave Chinner [Fri, 24 Sep 2010 11:58:31 +0000 (21:58 +1000)]
xfs: introduced uncached buffer read primitve

To avoid the need to use cached buffers for single-shot or buffers
cached at the filesystem level, introduce a new buffer read
primitive that bypasses the cache an reads directly from disk.

Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Alex Elder <aelder@sgi.com>
14 years agoxfs: rename xfs_buf_get_nodaddr to be more appropriate
Dave Chinner [Fri, 24 Sep 2010 10:07:47 +0000 (20:07 +1000)]
xfs: rename xfs_buf_get_nodaddr to be more appropriate

xfs_buf_get_nodaddr() is really used to allocate a buffer that is
uncached. While it is not directly assigned a disk address, the fact
that they are not cached is a more important distinction. With the
upcoming uncached buffer read primitive, we should be consistent
with this disctinction.

While there, make page allocation in xfs_buf_get_nodaddr() safe
against memory reclaim re-entrancy into the filesystem by allowing
a flags parameter to be passed.

Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Alex Elder <aelder@sgi.com>
14 years agoxfs: don't use vfs writeback for pure metadata modifications
Dave Chinner [Tue, 28 Sep 2010 02:27:25 +0000 (12:27 +1000)]
xfs: don't use vfs writeback for pure metadata modifications

Under heavy multi-way parallel create workloads, the VFS struggles
to write back all the inodes that have been changed in age order.
The bdi flusher thread becomes CPU bound, spending 85% of it's time
in the VFS code, mostly traversing the superblock dirty inode list
to separate dirty inodes old enough to flush.

We already keep an index of all metadata changes in age order - in
the AIL - and continued log pressure will do age ordered writeback
without any extra overhead at all. If there is no pressure on the
log, the xfssyncd will periodically write back metadata in ascending
disk address offset order so will be very efficient.

Hence we can stop marking VFS inodes dirty during transaction commit
or when changing timestamps during transactions. This will keep the
inodes in the superblock dirty list to those containing data or
unlogged metadata changes.

However, the timstamp changes are slightly more complex than this -
there are a couple of places that do unlogged updates of the
timestamps, and the VFS need to be informed of these. Hence add a
new function xfs_trans_ichgtime() for transactional changes,
and leave xfs_ichgtime() for the non-transactional changes.

Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Alex Elder <aelder@sgi.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
14 years agoxfs: lockless per-ag lookups
Dave Chinner [Wed, 22 Sep 2010 00:47:20 +0000 (10:47 +1000)]
xfs: lockless per-ag lookups

When we start taking a reference to the per-ag for every cached
buffer in the system, kernel lockstat profiling on an 8-way create
workload shows the mp->m_perag_lock has higher acquisition rates
than the inode lock and has significantly more contention. That is,
it becomes the highest contended lock in the system.

The perag lookup is trivial to convert to lock-less RCU lookups
because perag structures never go away. Hence the only thing we need
to protect against is tree structure changes during a grow. This can
be done simply by replacing the locking in xfs_perag_get() with RCU
read locking. This removes the mp->m_perag_lock completely from this
path.

Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Alex Elder <aelder@sgi.com>
14 years agoxfs: remove debug assert for per-ag reference counting
Dave Chinner [Wed, 22 Sep 2010 00:47:20 +0000 (10:47 +1000)]
xfs: remove debug assert for per-ag reference counting

When we start taking references per cached buffer to the the perag
it is cached on, it will blow the current debug maximum reference
count assert out of the water. The assert has never caught a bug,
and we have tracing to track changes if there ever is a problem,
so just remove it.

Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Alex Elder <aelder@sgi.com>
14 years agoxfs: reduce the number of CIL lock round trips during commit
Dave Chinner [Fri, 24 Sep 2010 08:14:13 +0000 (18:14 +1000)]
xfs: reduce the number of CIL lock round trips during commit

When commiting a transaction, we do a lock CIL state lock round trip
on every single log vector we insert into the CIL. This is resulting
in the lock being as hot as the inode and dcache locks on 8-way
create workloads. Rework the insertion loops to bring the number
of lock round trips to one per transaction for log vectors, and one
more do the busy extents.

Also change the allocation of the log vector buffer not to zero it
as we copy over the entire allocated buffer anyway.

This patch also includes a structural cleanup to the CIL item
insertion provided by Christoph Hellwig.

Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Alex Elder <aelder@sgi.com>
14 years agoxfs: eliminate some newly-reported gcc warnings
Poyo VL [Thu, 2 Sep 2010 07:41:55 +0000 (07:41 +0000)]
xfs: eliminate some newly-reported gcc warnings

Ionut Gabriel Popescu <poyo_vl@yahoo.com> submitted a simple change
to eliminate some "may be used uninitialized" warnings when building
XFS.  The reported condition seems to be something that GCC did not
used to recognize or report.  The warnings were produced by:

    gcc version 4.5.0 20100604
    [gcc-4_5-branch revision 160292] (SUSE Linux)

Signed-off-by: Ionut Gabriel Popescu <poyo_vl@yahoo.com>
Signed-off-by: Alex Elder <aelder@sgi.com>
14 years agoxfs: remove the ->kill_root btree operation
Christoph Hellwig [Tue, 7 Sep 2010 23:34:07 +0000 (23:34 +0000)]
xfs: remove the ->kill_root btree operation

The implementation os ->kill_root only differ by either simply
zeroing out the now unused buffer in the btree cursor in the inode
allocation btree or using xfs_btree_setbuf in the allocation btree.

Initially both of them used xfs_btree_setbuf, but the use in the
ialloc btree was removed early on because it interacted badly with
xfs_trans_binval.

In addition to zeroing out the buffer in the cursor xfs_btree_setbuf
updates the bc_ra array in the btree cursor, and calls
xfs_trans_brelse on the buffer previous occupying the slot.

The bc_ra update should be done for the alloc btree updated too,
although the lack of it does not cause serious problems.  The
xfs_trans_brelse call on the other hand is effectively a no-op in
the end - it keeps decrementing the bli_recur refcount until it hits
zero, and then just skips out because the buffer will always be
dirty at this point.  So removing it for the allocation btree is
just fine.

So unify the code and move it to xfs_btree.c.  While we're at it
also replace the call to xfs_btree_setbuf with a NULL bp argument in
xfs_btree_del_cursor with a direct call to xfs_trans_brelse given
that the cursor is beeing freed just after this and the state
updates are superflous.  After this xfs_btree_setbuf is only used
with a non-NULL bp argument and can thus be simplified.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Alex Elder <aelder@sgi.com>
14 years agoxfs: stop using xfs_qm_dqtobp in xfs_qm_dqflush
Christoph Hellwig [Mon, 6 Sep 2010 01:44:45 +0000 (01:44 +0000)]
xfs: stop using xfs_qm_dqtobp in xfs_qm_dqflush

In xfs_qm_dqflush we know that q_blkno must be initialized already from a
previous xfs_qm_dqread.  So instead of calling xfs_qm_dqtobp we can
simply read the quota buffer directly.  This also saves us from a duplicate
xfs_qm_dqcheck call check and allows xfs_qm_dqtobp to be simplified now
that it is always called for a newly initialized inode.  In addition to
that properly unwind all locks in xfs_qm_dqflush when xfs_qm_dqcheck
fails.

This mirrors a similar cleanup in the inode lookup done earlier.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Alex Elder <aelder@sgi.com>
14 years agoxfs: simplify xfs_qm_dqusage_adjust
Christoph Hellwig [Mon, 6 Sep 2010 01:44:22 +0000 (01:44 +0000)]
xfs: simplify xfs_qm_dqusage_adjust

There is no need to have the users and group/project quota locked at the
same time.  Get rid of xfs_qm_dqget_noattach and just do a xfs_qm_dqget
inside xfs_qm_quotacheck_dqadjust for the quota we are operating on
right now.  The new version of xfs_qm_quotacheck_dqadjust holds the
inode lock over it's operations, which is not a problem as it simply
increments counters and there is no concern about log contention
during mount time.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Alex Elder <aelder@sgi.com>
14 years agoxfs: Introduce XFS_IOC_ZERO_RANGE
Dave Chinner [Tue, 24 Aug 2010 02:02:11 +0000 (12:02 +1000)]
xfs: Introduce XFS_IOC_ZERO_RANGE

XFS_IOC_ZERO_RANGE is the equivalent of an atomic XFS_IOC_UNRESVSP/
XFS_IOC_RESVSP call pair. It enabled ranges of written data to be
turned into zeroes without requiring IO or having to free and
reallocate the extents in the range given as would occur if we had
to punch and then preallocate them separately.  This enables
applications to zero parts of files very quickly without changing
the layout of the files in any way.

Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
14 years agoxfs: use range primitives for xfs page cache operations
Dave Chinner [Tue, 24 Aug 2010 02:01:50 +0000 (12:01 +1000)]
xfs: use range primitives for xfs page cache operations

While XFS passes ranges to operate on from the core code, the
functions being called ignore the either the entire range or the end
of the range. This is historical because when the function were
written linux didn't have the necessary range operations. Update the
functions to use the correct operations.

Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
14 years agoLinux 2.6.36-rc8
Linus Torvalds [Thu, 14 Oct 2010 23:26:43 +0000 (16:26 -0700)]
Linux 2.6.36-rc8

14 years agoUn-inline the core-dump helper functions
Linus Torvalds [Thu, 14 Oct 2010 21:32:06 +0000 (14:32 -0700)]
Un-inline the core-dump helper functions

Tony Luck reports that the addition of the access_ok() check in commit
0eead9ab41da ("Don't dump task struct in a.out core-dumps") broke the
ia64 compile due to missing the necessary header file includes.

Rather than add yet another include (<asm/unistd.h>) to make everything
happy, just uninline the silly core dump helper functions and move the
bodies to fs/exec.c where they make a lot more sense.

dump_seek() in particular was too big to be an inline function anyway,
and none of them are in any way performance-critical.  And we really
don't need to mess up our include file headers more than they already
are.

Reported-and-tested-by: Tony Luck <tony.luck@gmail.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
14 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6
Linus Torvalds [Thu, 14 Oct 2010 18:19:44 +0000 (11:19 -0700)]
Merge git://git./linux/kernel/git/davem/net-2.6

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6:
  ehea: Fix a checksum issue on the receive path
  net: allow FEC driver to use fixed PHY support
  tg3: restore rx_dropped accounting
  b44: fix carrier detection on bind
  net: clear heap allocations for privileged ethtool actions
  NET: wimax, fix use after free
  ATM: iphase, remove sleep-inside-atomic
  ATM: mpc, fix use after free
  ATM: solos-pci, remove use after free
  net/fec: carrier off initially to avoid root mount failure
  r8169: use device model DMA API
  r8169: allocate with GFP_KERNEL flag when able to sleep

14 years agoDon't dump task struct in a.out core-dumps
Linus Torvalds [Thu, 14 Oct 2010 17:57:40 +0000 (10:57 -0700)]
Don't dump task struct in a.out core-dumps

akiphie points out that a.out core-dumps have that odd task struct
dumping that was never used and was never really a good idea (it goes
back into the mists of history, probably the original core-dumping
code).  Just remove it.

Also do the access_ok() check on dump_write().  It probably doesn't
matter (since normal filesystems all seem to do it anyway), but he
points out that it's normally done by the VFS layer, so ...

[ I suspect that we should possibly do "vfs_write()" instead of
  calling ->write directly.  That also does the whole fsnotify and write
  statistics thing, which may or may not be a good idea. ]

And just to be anal, do this all for the x86-64 32-bit a.out emulation
code too, even though it's not enabled (and won't currently even
compile)

Reported-by: akiphie <akiphie@lavabit.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
14 years agoMerge branch 'fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/djbw/async_tx
Linus Torvalds [Wed, 13 Oct 2010 23:51:59 +0000 (16:51 -0700)]
Merge branch 'fixes' of git://git./linux/kernel/git/djbw/async_tx

* 'fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/djbw/async_tx:
  ioat2: fix performance regression

14 years agoMerge branch 'for-2.6.36' of git://linux-nfs.org/~bfields/linux
Linus Torvalds [Wed, 13 Oct 2010 23:51:29 +0000 (16:51 -0700)]
Merge branch 'for-2.6.36' of git://linux-nfs.org/~bfields/linux

* 'for-2.6.36' of git://linux-nfs.org/~bfields/linux:
  nfsd: fix BUG at fs/nfsd/nfsfh.h:199 on unlink

14 years agoMerge branch 'perf-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Wed, 13 Oct 2010 23:50:23 +0000 (16:50 -0700)]
Merge branch 'perf-fixes-for-linus' of git://git./linux/kernel/git/tip/linux-2.6-tip

* 'perf-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
  ring-buffer: Fix typo of time extends per page
  perf, MIPS: Support cross compiling of tools/perf for MIPS
  perf: Fix incorrect copy_from_user() usage

14 years agoMerge master.kernel.org:/home/rmk/linux-2.6-arm
Linus Torvalds [Wed, 13 Oct 2010 23:35:33 +0000 (16:35 -0700)]
Merge master.kernel.org:/home/rmk/linux-2.6-arm

* master.kernel.org:/home/rmk/linux-2.6-arm:
  ARM: relax ioremap prohibition (309caa9) for -final and -stable
  ARM: 6440/1: ep93xx: DMA: fix channel_disable
  cpuimx27: fix i2c bus selection
  cpuimx27: fix compile when ULPI is selected
  ARM: 6435/1: Fix HWCAP_TLS flag for ARM11MPCore/Cortex-A9
  ARM: 6436/1: AT91: Fix power-saving in idle-mode on 926T processors
  ARM: fix section mismatch warnings in Versatile Express
  ARM: 6412/1: kprobes-decode: add support for MOVW instruction
  ARM: 6419/1: mmu: Fix MT_MEMORY and MT_MEMORY_NONCACHED pte flags
  ARM: 6416/1: errata: faulty hazard checking in the Store Buffer may lead to data corruption

14 years agoMerge branch 'omap-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Wed, 13 Oct 2010 23:35:05 +0000 (16:35 -0700)]
Merge branch 'omap-fixes-for-linus' of git://git./linux/kernel/git/tmlind/linux-omap-2.6

* 'omap-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tmlind/linux-omap-2.6:
  omap: iommu-load cam register before flushing the entry

14 years agoMerge branch 'drm-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/airlied...
Linus Torvalds [Wed, 13 Oct 2010 23:34:46 +0000 (16:34 -0700)]
Merge branch 'drm-fixes' of git://git./linux/kernel/git/airlied/drm-2.6

* 'drm-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/airlied/drm-2.6:
  drm/radeon/kms: Silent spurious error message
  drm/radeon/kms: fix bad cast/shift in evergreen.c
  drm/radeon/kms: make TV/DFP table info less verbose
  drm/radeon/kms: leave certain CP int bits enabled
  drm/radeon/kms: avoid corner case issue with unmappable vram V2

14 years agoMerge branch 'x86-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Wed, 13 Oct 2010 23:34:23 +0000 (16:34 -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, numa: For each node, register the memory blocks actually used
  x86, AMD, MCE thresholding: Fix the MCi_MISCj iteration order
  x86, mce, therm_throt.c: Fix missing curly braces in error handling logic

14 years agoioat2: fix performance regression
Dan Williams [Wed, 13 Oct 2010 22:43:10 +0000 (15:43 -0700)]
ioat2: fix performance regression

Commit 0793448 "DMAENGINE: generic channel status v2" changed the interface for
how dma channel progress is retrieved.  It inadvertently exported an internal
helper function ioat_tx_status() instead of ioat_dma_tx_status().  The latter
polls the hardware to get the latest completion state, while the helper just
evaluates the current state without touching hardware.  The effect is that we
end up waiting for completion timeouts or descriptor allocation errors before
the completion state is updated.

iperf (before fix):
[SUM]  0.0-41.3 sec   364 MBytes  73.9 Mbits/sec

iperf (after fix):
[SUM]  0.0- 4.5 sec   499 MBytes   940 Mbits/sec

This is a regression starting with 2.6.35.

Cc: <stable@kernel.org>
Cc: Dave Jiang <dave.jiang@intel.com>
Cc: Jesse Brandeburg <jesse.brandeburg@intel.com>
Cc: Linus Walleij <linus.walleij@stericsson.com>
Cc: Maciej Sosnowski <maciej.sosnowski@intel.com>
Reported-by: Richard Scobie <richard@sauce.co.nz>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
14 years agoehea: Fix a checksum issue on the receive path
Breno Leitao [Thu, 7 Oct 2010 13:17:33 +0000 (13:17 +0000)]
ehea: Fix a checksum issue on the receive path

Currently we set all skbs with CHECKSUM_UNNECESSARY, even
those whose protocol we don't know. This patch just
add the CHECKSUM_COMPLETE tag for non TCP/UDP packets.

Reported-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: Breno Leitao <leitao@linux.vnet.ibm.com>
Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
14 years agonfsd: fix BUG at fs/nfsd/nfsfh.h:199 on unlink
J. Bruce Fields [Wed, 13 Oct 2010 18:46:17 +0000 (14:46 -0400)]
nfsd: fix BUG at fs/nfsd/nfsfh.h:199 on unlink

As of commit 43a9aa64a2f4330a9cb59aaf5c5636566bce067c "NFSD:
Fill in WCC data for REMOVE, RMDIR, MKNOD, and MKDIR", we sometimes call
fh_unlock on a filehandle that isn't fully initialized.

We should fix up the callers, but as a quick fix it is also sufficient
just to remove this assertion.

Reported-by: Marius Tolzmann <tolzmann@molgen.mpg.de>
Cc: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
14 years agonet: allow FEC driver to use fixed PHY support
Greg Ungerer [Mon, 11 Oct 2010 21:03:05 +0000 (21:03 +0000)]
net: allow FEC driver to use fixed PHY support

At least one board using the FEC driver does not have a conventional
PHY attached to it, it is directly connected to a somewhat simple
ethernet switch (the board is the SnapGear/LITE, and the attached
4-port ethernet switch is a RealTek RTL8305). This switch does not
present the usual register interface of a PHY, it presents nothing.
So a PHY scan will find nothing - it finds ID's of 0 for each PHY
on the attached MII bus.

After the FEC driver was changed to use phylib for supporting PHYs
it no longer works on this particular board/switch setup.

Add code support to use a fixed phy if no PHY is found on the MII bus.
This is based on the way the cpmac.c driver solved this same problem.

Signed-off-by: Greg Ungerer <gerg@uclinux.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
14 years agoARM: relax ioremap prohibition (309caa9) for -final and -stable
Russell King [Tue, 12 Oct 2010 23:15:25 +0000 (00:15 +0100)]
ARM: relax ioremap prohibition (309caa9) for -final and -stable

... but produce a big warning about the problem as encouragement
for people to fix their drivers.

Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
14 years agoMerge branch 'for-rmk' of git://git.pengutronix.de/git/imx/linux-2.6
Russell King [Tue, 12 Oct 2010 21:43:26 +0000 (22:43 +0100)]
Merge branch 'for-rmk' of git://git.pengutronix.de/git/imx/linux-2.6

14 years agoARM: 6440/1: ep93xx: DMA: fix channel_disable
Mika Westerberg [Tue, 12 Oct 2010 09:37:59 +0000 (10:37 +0100)]
ARM: 6440/1: ep93xx: DMA: fix channel_disable

When channel_disable() is called, it disables per channel interrupts and
waits until channels state becomes STATE_STALL, and then disables the
channel. Now, if the DMA transfer is disabled while the channel is in
STATE_NEXT we will not wait anything and disable the channel immediately.
This seems to cause weird data corruption for example in audio transfers.

Fix is to wait while we are in STATE_NEXT or STATE_ON and only then
disable the channel.

Signed-off-by: Mika Westerberg <mika.westerberg@iki.fi>
Acked-by: Ryan Mallon <ryan@bluewatersys.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
14 years agoMerge branch 'kvm-updates/2.6.36' of git://git.kernel.org/pub/scm/virt/kvm/kvm
Linus Torvalds [Tue, 12 Oct 2010 16:16:01 +0000 (09:16 -0700)]
Merge branch 'kvm-updates/2.6.36' of git://git./virt/kvm/kvm

* 'kvm-updates/2.6.36' of git://git.kernel.org/pub/scm/virt/kvm/kvm:
  KVM: x86: Move TSC reset out of vmcb_init
  KVM: x86: Fix SVM VMCB reset

14 years agoring-buffer: Fix typo of time extends per page
Steven Rostedt [Tue, 12 Oct 2010 16:06:43 +0000 (12:06 -0400)]
ring-buffer: Fix typo of time extends per page

Time stamps for the ring buffer are created by the difference between
two events. Each page of the ring buffer holds a full 64 bit timestamp.
Each event has a 27 bit delta stamp from the last event. The unit of time
is nanoseconds, so 27 bits can hold ~134 milliseconds. If two events
happen more than 134 milliseconds apart, a time extend is inserted
to add more bits for the delta. The time extend has 59 bits, which
is good for ~18 years.

Currently the time extend is committed separately from the event.
If an event is discarded before it is committed, due to filtering,
the time extend still exists. If all events are being filtered, then
after ~134 milliseconds a new time extend will be added to the buffer.

This can only happen till the end of the page. Since each page holds
a full timestamp, there is no reason to add a time extend to the
beginning of a page. Time extends can only fill a page that has actual
data at the beginning, so there is no fear that time extends will fill
more than a page without any data.

When reading an event, a loop is made to skip over time extends
since they are only used to maintain the time stamp and are never
given to the caller. As a paranoid check to prevent the loop running
forever, with the knowledge that time extends may only fill a page,
a check is made that tests the iteration of the loop, and if the
iteration is more than the number of time extends that can fit in a page
a warning is printed and the ring buffer is disabled (all of ftrace
is also disabled with it).

There is another event type that is called a TIMESTAMP which can
hold 64 bits of data in the theoretical case that two events happen
18 years apart. This code has not been implemented, but the name
of this event exists, as well as the structure for it. The
size of a TIMESTAMP is 16 bytes, where as a time extend is only
8 bytes. The macro used to calculate how many time extends can fit on
a page used the TIMESTAMP size instead of the time extend size
cutting the amount in half.

The following test case can easily trigger the warning since we only
need to have half the page filled with time extends to trigger the
warning:

 # cd /sys/kernel/debug/tracing/
 # echo function > current_tracer
 # echo 'common_pid < 0' > events/ftrace/function/filter
 # echo > trace
 # echo 1 > trace_marker
 # sleep 120
 # cat trace

Enabling the function tracer and then setting the filter to only trace
functions where the process id is negative (no events), then clearing
the trace buffer to ensure that we have nothing in the buffer,
then write to trace_marker to add an event to the beginning of a page,
sleep for 2 minutes (only 35 seconds is probably needed, but this
guarantees the bug), and then finally reading the trace which will
trigger the bug.

This patch fixes the typo and prevents the false positive of that warning.

Reported-by: Hans J. Koch <hjk@linutronix.de>
Tested-by: Hans J. Koch <hjk@linutronix.de>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Stable Kernel <stable@kernel.org>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
14 years agoperf, MIPS: Support cross compiling of tools/perf for MIPS
Deng-Cheng Zhu [Tue, 12 Oct 2010 11:33:33 +0000 (19:33 +0800)]
perf, MIPS: Support cross compiling of tools/perf for MIPS

Changes:
 v4: Fix the cosmetic issue of redundant dot-ops
 v3: Change rmb() to use SYNC
 v2: Include mips unistd.h and define rmb()/cpu_relax() in tools/perf/perf.h

Signed-off-by: Deng-Cheng Zhu <dengcheng.zhu@gmail.com>
Acked-by: Ralf Baechle <ralf@linux-mips.org>
Cc: David Daney <ddaney@caviumnetworks.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agodrm/radeon/kms: Silent spurious error message
Jean Delvare [Fri, 8 Oct 2010 12:34:49 +0000 (14:34 +0200)]
drm/radeon/kms: Silent spurious error message

I see the following error message in my kernel log from time to time:
radeon 0000:07:00.0: ffff88007c334000 reserve failed for wait
radeon 0000:07:00.0: ffff88007c334000 reserve failed for wait

After investigation, it turns out that there's nothing to be afraid of
and everything works as intended. So remove the spurious log message.

Signed-off-by: Jean Delvare <khali@linux-fr.org>
Reviewed-by: Jerome Glisse <jglisse@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
14 years agodrm/radeon/kms: fix bad cast/shift in evergreen.c
Alex Deucher [Mon, 11 Oct 2010 16:41:32 +0000 (12:41 -0400)]
drm/radeon/kms: fix bad cast/shift in evergreen.c

Missing parens.

fixes:
https://bugs.freedesktop.org/show_bug.cgi?id=30718

Reported-by: Dave Gilbert <freedesktop@treblig.org>
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Cc: stable@kernel.org
Signed-off-by: Dave Airlie <airlied@redhat.com>
14 years agodrm/radeon/kms: make TV/DFP table info less verbose
Alex Deucher [Fri, 8 Oct 2010 02:38:42 +0000 (22:38 -0400)]
drm/radeon/kms: make TV/DFP table info less verbose

Make TV standard and DFP table revisions debug only.

Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
14 years agodrm/radeon/kms: leave certain CP int bits enabled
Alex Deucher [Fri, 8 Oct 2010 16:09:12 +0000 (12:09 -0400)]
drm/radeon/kms: leave certain CP int bits enabled

These bits are used for internal communication and should
be left enabled.  This may fix s/r issues on some systems.

Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
14 years agodrm/radeon/kms: avoid corner case issue with unmappable vram V2
Jerome Glisse [Tue, 10 Aug 2010 21:41:31 +0000 (17:41 -0400)]
drm/radeon/kms: avoid corner case issue with unmappable vram V2

We should not allocate any object into unmappable vram if we
have no means to access them which on all GPU means having the
CP running and on newer GPU having the blit utility working.

This patch limit the vram allocation to visible vram until
we have acceleration up and running.

Note that it's more than unlikely that we run into any issue
related to that as when acceleration is not woring userspace
should allocate any object in vram beside front buffer which
should fit in visible vram.

V2 use real_vram_size as mc_vram_size could be bigger than
   the actual amount of vram

[airlied: fixup r700_cp_stop case]

Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
14 years agoperf: Fix incorrect copy_from_user() usage
John Blackwood [Tue, 28 Sep 2010 22:03:11 +0000 (18:03 -0400)]
perf: Fix incorrect copy_from_user() usage

perf events: repair incorrect use of copy_from_user

This makes the perf_event_period() return 0 instead of
-EFAULT on success.

Signed-off-by: John Blackwood<john.blackwood@ccur.com>
Signed-off-by: Joe Korty <joe.korty@ccur.com>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <20100928220311.GA18145@tsunami.ccur.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agofanotify: disable fanotify syscalls
Eric Paris [Mon, 11 Oct 2010 22:13:31 +0000 (18:13 -0400)]
fanotify: disable fanotify syscalls

This patch disables the fanotify syscalls by just not building them and
letting the cond_syscall() statements in kernel/sys_ni.c redirect them
to sys_ni_syscall().

It was pointed out by Tvrtko Ursulin that the fanotify interface did not
include an explicit prioritization between groups.  This is necessary
for fanotify to be usable for hierarchical storage management software,
as they must get first access to the file, before inotify-like notifiers
see the file.

This feature can be added in an ABI compatible way in the next release
(by using a number of bits in the flags field to carry the info) but it
was suggested by Alan that maybe we should just hold off and do it in
the next cycle, likely with an (new) explicit argument to the syscall.
I don't like this approach best as I know people are already starting to
use the current interface, but Alan is all wise and noone on list backed
me up with just using what we have.  I feel this is needlessly ripping
the rug out from under people at the last minute, but if others think it
needs to be a new argument it might be the best way forward.

Three choices:
Go with what we got (and implement the new feature next cycle).  Add a
new field right now (and implement the new feature next cycle).  Wait
till next cycle to release the ABI (and implement the new feature next
cycle).  This is number 3.

Signed-off-by: Eric Paris <eparis@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
14 years agotg3: restore rx_dropped accounting
Eric Dumazet [Sun, 10 Oct 2010 19:55:52 +0000 (19:55 +0000)]
tg3: restore rx_dropped accounting

commit 511d22247be7 (tg3: 64 bit stats on all arches), overlooked the
rx_dropped accounting.

We use a full "struct rtnl_link_stats64" to hold rx_dropped value, but
forgot to report it in tg3_get_stats64().

Use an "unsigned long" instead to shrink "struct tg3" by 176 bytes, and
report this value to stats readers.

Increment rx_dropped counter for oversized frames.

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
CC: Michael Chan <mchan@broadcom.com>
CC: Matt Carlson <mcarlson@broadcom.com>
Acked-by: Matt Carlson <mcarlson@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
14 years agob44: fix carrier detection on bind
Paul Fertser [Mon, 11 Oct 2010 22:45:35 +0000 (15:45 -0700)]
b44: fix carrier detection on bind

For carrier detection to work properly when binding the driver with a cable
unplugged, netif_carrier_off() should be called after register_netdev(),
not before.

Signed-off-by: Paul Fertser <fercerpav@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
14 years agox86, numa: For each node, register the memory blocks actually used
Yinghai Lu [Mon, 11 Oct 2010 02:52:15 +0000 (19:52 -0700)]
x86, numa: For each node, register the memory blocks actually used

Russ reported SGI UV is broken recently. He said:

| The SRAT table shows that memory range is spread over two nodes.
|
| SRAT: Node 0 PXM 0 100000000-800000000
| SRAT: Node 1 PXM 1 800000000-1000000000
| SRAT: Node 0 PXM 0 1000000000-1080000000
|
|Previously, the kernel early_node_map[] would show three entries
|with the proper node.
|
|[    0.000000]     0: 0x00100000 -> 0x00800000
|[    0.000000]     1: 0x00800000 -> 0x01000000
|[    0.000000]     0: 0x01000000 -> 0x01080000
|
|The problem is recent community kernel early_node_map[] shows
|only two entries with the node 0 entry overlapping the node 1
|entry.
|
|    0: 0x00100000 -> 0x01080000
|    1: 0x00800000 -> 0x01000000

After looking at the changelog, Found out that it has been broken for a while by
following commit

|commit 8716273caef7f55f39fe4fc6c69c5f9f197f41f1
|Author: David Rientjes <rientjes@google.com>
|Date:   Fri Sep 25 15:20:04 2009 -0700
|
|    x86: Export srat physical topology

Before that commit, register_active_regions() is called for every SRAT memory
entry right away.

Use nodememblk_range[] instead of nodes[] in order to make sure we
capture the actual memory blocks registered with each node.  nodes[]
contains an extended range which spans all memory regions associated
with a node, but that does not mean that all the memory in between are
included.

Reported-by: Russ Anderson <rja@sgi.com>
Tested-by: Russ Anderson <rja@sgi.com>
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
LKML-Reference: <4CB27BDF.5000800@kernel.org>
Acked-by: David Rientjes <rientjes@google.com>
Cc: <stable@kernel.org> 2.6.33 .34 .35 .36
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
14 years agonet: clear heap allocations for privileged ethtool actions
Kees Cook [Mon, 11 Oct 2010 19:23:25 +0000 (12:23 -0700)]
net: clear heap allocations for privileged ethtool actions

Several other ethtool functions leave heap uncleared (potentially) by
drivers. Some interfaces appear safe (eeprom, etc), in that the sizes
are well controlled. In some situations (e.g. unchecked error conditions),
the heap will remain unchanged in areas before copying back to userspace.
Note that these are less of an issue since these all require CAP_NET_ADMIN.

Cc: stable@kernel.org
Signed-off-by: Kees Cook <kees.cook@canonical.com>
Acked-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
14 years agoNET: wimax, fix use after free
Jiri Slaby [Sun, 10 Oct 2010 23:26:58 +0000 (23:26 +0000)]
NET: wimax, fix use after free

Stanse found that i2400m_rx frees skb, but still uses skb->len even
though it has skb_len defined. So use skb_len properly in the code.

And also define it unsinged int rather than size_t to solve
compilation warnings.

Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Cc: Inaky Perez-Gonzalez <inaky.perez-gonzalez@intel.com>
Cc: linux-wimax@intel.com
Acked-by: Inaky Perez-Gonzalez <inaky.perez-gonzalez@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
14 years agoATM: iphase, remove sleep-inside-atomic
Jiri Slaby [Sun, 10 Oct 2010 23:26:57 +0000 (23:26 +0000)]
ATM: iphase, remove sleep-inside-atomic

Stanse found that ia_init_one locks a spinlock and inside of that it
calls ia_start which calls:
* request_irq
* tx_init which does kmalloc(GFP_KERNEL)

Both of them can thus sleep and result in a deadlock. I don't see a
reason to have a per-device spinlock there which is used only there
and inited right before the lock location. So remove it completely.

Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Cc: Chas Williams <chas@cmf.nrl.navy.mil>
Signed-off-by: David S. Miller <davem@davemloft.net>
14 years agoATM: mpc, fix use after free
Jiri Slaby [Sun, 10 Oct 2010 22:46:34 +0000 (22:46 +0000)]
ATM: mpc, fix use after free

Stanse found that mpc_push frees skb and then it dereferences it. It
is a typo, new_skb should be dereferenced there.

Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Acked-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
14 years agoATM: solos-pci, remove use after free
Jiri Slaby [Sun, 10 Oct 2010 21:50:44 +0000 (21:50 +0000)]
ATM: solos-pci, remove use after free

Stanse found we do in console_show:
  kfree_skb(skb);
  return skb->len;
which is not good. Fix that by remembering the len and use it in the
function instead.

Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Cc: Chas Williams <chas@cmf.nrl.navy.mil>
Acked-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
14 years agoMerge branch 'rc-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/mmarek/kbuil...
Linus Torvalds [Mon, 11 Oct 2010 17:19:24 +0000 (10:19 -0700)]
Merge branch 'rc-fixes' of git://git./linux/kernel/git/mmarek/kbuild-2.6

* 'rc-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/mmarek/kbuild-2.6:
  kbuild: fix oldnoconfig to do the right thing
  kconfig: Temporarily disable dependency warnings
  kconfig: delay symbol direct dependency initialization

14 years agoMerge branch 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mjg59/platf...
Linus Torvalds [Mon, 11 Oct 2010 17:05:05 +0000 (10:05 -0700)]
Merge branch 'for_linus' of git://git./linux/kernel/git/mjg59/platform-drivers-x86

* 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mjg59/platform-drivers-x86:
  IPS driver: Fix limit clamping when reducing CPU power
  [PATCH 2/2] IPS driver: disable CPU turbo
  IPS driver: apply BIOS provided CPU limit if different from default
  intel_ips -- ensure we do not enable gpu turbo mode without driver linkage
  intel_ips: Print MCP limit exceeded values.
  IPS driver: verify BIOS provided limits
  IPS driver: don't toggle CPU turbo on unsupported CPUs
  NULL pointer might be used in ips_monitor()
  Release symbol on error-handling path of ips_get_i915_syms()
  old_cpu_power is wrongly divided by 65535 in ips_monitor()
  seqno mask of THM_ITV register is 16bit

14 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input
Linus Torvalds [Mon, 11 Oct 2010 17:03:19 +0000 (10:03 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/dtor/input

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input:
  Input: wacom - fix pressure in Cintiq 21UX2

14 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound-2.6
Linus Torvalds [Mon, 11 Oct 2010 17:02:57 +0000 (10:02 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/tiwai/sound-2.6

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound-2.6:
  ALSA: hda - Add another HP DV6 quirk
  OSS: soundcard: locking bug in sound_ioctl()
  ASoC: Update links for Wolfson MAINTAINERS entry
  ASoC: Add Dimitris Papastamos to Wolfson maintainers
  ASoC: Add Jassi Brar as Samsung maintainer

14 years agoMerge branch 'fix/asoc' into for-linus
Takashi Iwai [Mon, 11 Oct 2010 12:36:25 +0000 (14:36 +0200)]
Merge branch 'fix/asoc' into for-linus

14 years agoALSA: hda - Add another HP DV6 quirk
Luke Yelavich [Wed, 6 Oct 2010 04:45:46 +0000 (15:45 +1100)]
ALSA: hda - Add another HP DV6 quirk

BugLink: https://bugs.launchpad.net/bugs/653420
Add another HP DV6 notebook (103c:363e) to use STAC_HP_DV5.

Signed-off-by: Luke Yelavich <luke.yelavich@canonical.com>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
14 years agoOSS: soundcard: locking bug in sound_ioctl()
Dan Carpenter [Sun, 10 Oct 2010 17:33:52 +0000 (19:33 +0200)]
OSS: soundcard: locking bug in sound_ioctl()

We shouldn't return directly here because we're still holding the
&soundcard_mutex.

This bug goes all the way back to the start of git.  It's strange that
no one has complained about it as a runtime bug.

CC: stable@kernel.org
Signed-off-by: Dan Carpenter <error27@gmail.com>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
14 years agoKVM: x86: Move TSC reset out of vmcb_init
Zachary Amsden [Fri, 20 Aug 2010 08:07:19 +0000 (22:07 -1000)]
KVM: x86: Move TSC reset out of vmcb_init

The VMCB is reset whenever we receive a startup IPI, so Linux is setting
TSC back to zero happens very late in the boot process and destabilizing
the TSC.  Instead, just set TSC to zero once at VCPU creation time.

Why the separate patch?  So git-bisect is your friend.

Signed-off-by: Zachary Amsden <zamsden@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
14 years agoKVM: x86: Fix SVM VMCB reset
Zachary Amsden [Fri, 20 Aug 2010 08:07:18 +0000 (22:07 -1000)]
KVM: x86: Fix SVM VMCB reset

On reset, VMCB TSC should be set to zero.  Instead, code was setting
tsc_offset to zero, which passes through the underlying TSC.

Signed-off-by: Zachary Amsden <zamsden@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
14 years agox86, AMD, MCE thresholding: Fix the MCi_MISCj iteration order
Borislav Petkov [Fri, 8 Oct 2010 10:08:34 +0000 (12:08 +0200)]
x86, AMD, MCE thresholding: Fix the MCi_MISCj iteration order

This fixes possible cases of not collecting valid error info in
the MCE error thresholding groups on F10h hardware.

The current code contains a subtle problem of checking only the
Valid bit of MSR0000_0413 (which is MC4_MISC0 - DRAM
thresholding group) in its first iteration and breaking out if
the bit is cleared.

But (!), this MSR contains an offset value, BlkPtr[31:24], which
points to the remaining MSRs in this thresholding group which
might contain valid information too. But if we bail out only
after we checked the valid bit in the first MSR and not the
block pointer too, we miss that other information.

The thing is, MC4_MISC0[BlkPtr] is not predicated on
MCi_STATUS[MiscV] or MC4_MISC0[Valid] and should be checked
prior to iterating over the MCI_MISCj thresholding group,
irrespective of the MC4_MISC0[Valid] setting.

Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
Cc: <stable@kernel.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agonet/fec: carrier off initially to avoid root mount failure
Oskar Schirmer [Thu, 7 Oct 2010 02:30:30 +0000 (02:30 +0000)]
net/fec: carrier off initially to avoid root mount failure

with hardware slow in negotiation, the system did freeze
while trying to mount root on nfs at boot time.

the link state has not been initialised so network stack
tried to start transmission right away. this caused instant
retries, as the driver solely stated business upon link down,
rendering the system unusable.

notify carrier off initially to prevent transmission until
phylib will report link up.

Signed-off-by: Oskar Schirmer <oskar@linutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
14 years agoInput: wacom - fix pressure in Cintiq 21UX2
Aristeu Rozanski [Sun, 10 Oct 2010 21:12:33 +0000 (14:12 -0700)]
Input: wacom - fix pressure in Cintiq 21UX2

Currently the pressure range in Cintiq 21UX2 is limited to half of the
supported. This patch fixes the problem.

Signed-off-by: Aristeu Rozanski <aris@redhat.com>
Acked-by: Ping Cheng <pingc@wacom.com>
CC: stable@kernel.org
Signed-off-by: Dmitry Torokhov <dtor@mail.ru>
14 years agokbuild: fix oldnoconfig to do the right thing
Kyle McMartin [Wed, 1 Sep 2010 15:21:30 +0000 (11:21 -0400)]
kbuild: fix oldnoconfig to do the right thing

Commit 861b4ea4 broke oldnoconfig when removed the oldnoconfig checks on
                       if (input_mode == nonint_oldconfig ||
                           input_mode == oldnoconfig) {
                               if (input_mode == nonint_oldconfig &&
                                   sym->name &&
                                    !sym_is_choice_value(sym)) {
to avoid oldnoconfig chugging through the else stanza.

Fix that to restore expected behaviour (which I've confirmed in the
Fedora kernel build that the configs end up looking the same.)

Signed-off-by: Kyle McMartin <kyle@redhat.com>
Acked-by: Sam Ravnborg <sam@ravnborg.org>
Signed-off-by: Michal Marek <mmarek@suse.cz>
14 years agokconfig: Temporarily disable dependency warnings
Michal Marek [Fri, 8 Oct 2010 14:40:27 +0000 (16:40 +0200)]
kconfig: Temporarily disable dependency warnings

After fixing a use-after-free bug in kconfig, a 'make defconfig' or
'make allmodconfig' fills the screen with warnings that were not
detected before. Given that we are close to the release now, disable the
warnings temporarily and deal with them after 2.6.36.

Signed-off-by: Michal Marek <mmarek@suse.cz>
14 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6
Linus Torvalds [Sat, 9 Oct 2010 19:06:26 +0000 (12:06 -0700)]
Merge git://git./linux/kernel/git/davem/net-2.6

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6: (27 commits)
  net: clear heap allocation for ETHTOOL_GRXCLSRLALL
  isdn: strcpy() => strlcpy()
  Revert "mac80211: use netif_receive_skb in ieee80211_tx_status callpath"
  mac80211: delete AddBA response timer
  ath9k_hw: fix regression in ANI listen time calculation
  caif: fix two caif_connect() bugs
  bonding: fix WARN_ON when writing to bond_master sysfs file
  skge: add quirk to limit DMA
  MAINTAINERS: update Intel LAN Ethernet info
  e1000e.txt: Add e1000e documentation
  e1000.txt: Update e1000 documentation
  ixgbevf.txt: Update ixgbevf documentation
  cls_u32: signedness bug
  Bluetooth: Disallow to change L2CAP_OPTIONS values when connected
  sctp: Fix out-of-bounds reading in sctp_asoc_get_hmac()
  sctp: prevent reading out-of-bounds memory
  ipv4: correct IGMP behavior on v3 query during v2-compatibility mode
  netdev: Depend on INET before selecting INET_LRO
  Revert "ipv4: Make INET_LRO a bool instead of tristate."
  net: Fix the condition passed to sk_wait_event()
  ...

14 years agoMerge branch 'idle-release' of git://git.kernel.org/pub/scm/linux/kernel/git/lenb...
Linus Torvalds [Sat, 9 Oct 2010 19:05:09 +0000 (12:05 -0700)]
Merge branch 'idle-release' of git://git./linux/kernel/git/lenb/linux-idle-2.6

* 'idle-release' of git://git.kernel.org/pub/scm/linux/kernel/git/lenb/linux-idle-2.6:
  intel_idle: enable Atom C6

14 years agoMerge branch 'release' of git://git.kernel.org/pub/scm/linux/kernel/git/lenb/linux...
Linus Torvalds [Sat, 9 Oct 2010 19:04:38 +0000 (12:04 -0700)]
Merge branch 'release' of git://git./linux/kernel/git/lenb/linux-acpi-2.6

* 'release' of git://git.kernel.org/pub/scm/linux/kernel/git/lenb/linux-acpi-2.6:
  ACPI: add DMI to disable AML Vista compatibility on MSI GX723 Notebook
  ACPI: Handle ACPI0007 Device in acpi_early_set_pdc

14 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/sage/ceph...
Linus Torvalds [Sat, 9 Oct 2010 19:03:46 +0000 (12:03 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/sage/ceph-client

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/sage/ceph-client:
  ceph: update issue_seq on cap grant
  ceph: send cap release message early on failed revoke.
  ceph: Update max_len with minimum required size
  ceph: Fix return value of encode_fh function
  ceph: avoid null deref in osd request error path
  ceph: fix list_add usage on unsafe_writes list

14 years agoMerge branch 'for-linus' of git://git.open-osd.org/linux-open-osd
Linus Torvalds [Sat, 9 Oct 2010 19:03:23 +0000 (12:03 -0700)]
Merge branch 'for-linus' of git://git.open-osd.org/linux-open-osd

* 'for-linus' of git://git.open-osd.org/linux-open-osd:
  exofs: Fix double page_unlock BUG in write_begin/end

14 years agoMerge branch 'drm-intel-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/ickle...
Linus Torvalds [Sat, 9 Oct 2010 18:43:40 +0000 (11:43 -0700)]
Merge branch 'drm-intel-fixes' of git://git./linux/kernel/git/ickle/drm-intel

* 'drm-intel-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/ickle/drm-intel:
  drm/i915: Prevent module unload to avoid random memory corruption

14 years agoMerge branch 's5p-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sat, 9 Oct 2010 18:43:18 +0000 (11:43 -0700)]
Merge branch 's5p-fixes-for-linus' of git://git./linux/kernel/git/kgene/linux-samsung

* 's5p-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/kgene/linux-samsung:
  ARM: SAMSUNG: Add a workaround for get_clock() for serial driver
  ARM: S5P: Bug fix on errors of build with CONFIG_PREEMPT_NONE
  ARM: SAMSUNG: Fix build warnings because of unused codes

14 years agoDefine _addr_lsb in siginfo_t for mips
Andi Kleen [Fri, 8 Oct 2010 18:36:15 +0000 (20:36 +0200)]
Define _addr_lsb in siginfo_t for mips

Define an _addr_lsb field in the mips and ia64 siginfo_ts, following
the asm-generic version. This just puts the field over padding.

This fixes a compilation problem introduced with a337fda.

Cc: ralf@linux-mips.org
Cc: tony.luck@intel.com
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
14 years agovirtio-blk: fix request leak.
Mike Snitzer [Sat, 9 Oct 2010 01:42:13 +0000 (12:12 +1030)]
virtio-blk: fix request leak.

Must drop reference taken by blk_make_request().

Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Cc: stable@kernel.org # .35.x
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
14 years agor8169: use device model DMA API
Stanislaw Gruszka [Fri, 8 Oct 2010 04:25:01 +0000 (04:25 +0000)]
r8169: use device model DMA API

Use DMA API as PCI equivalents will be deprecated. This change also
allow to allocate with GFP_KERNEL where possible.

Tested-by: Neal Becker <ndbecker2@gmail.com>
Signed-off-by: Stanislaw Gruszka <sgruszka@redhat.com>
Acked-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
14 years agor8169: allocate with GFP_KERNEL flag when able to sleep
Stanislaw Gruszka [Fri, 8 Oct 2010 04:25:00 +0000 (04:25 +0000)]
r8169: allocate with GFP_KERNEL flag when able to sleep

We have fedora bug report where driver fail to initialize after
suspend/resume because of memory allocation errors:
https://bugzilla.redhat.com/show_bug.cgi?id=629158

To fix use GFP_KERNEL allocation where possible.

Tested-by: Neal Becker <ndbecker2@gmail.com>
Signed-off-by: Stanislaw Gruszka <sgruszka@redhat.com>
Acked-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
14 years agoMerge branch 'msi-dmi' into release
Len Brown [Sat, 9 Oct 2010 02:37:46 +0000 (22:37 -0400)]
Merge branch 'msi-dmi' into release

14 years agoMerge branch 'pdc-regression' into release
Len Brown [Sat, 9 Oct 2010 02:35:02 +0000 (22:35 -0400)]
Merge branch 'pdc-regression' into release

14 years agointel_idle: enable Atom C6
Len Brown [Tue, 5 Oct 2010 17:43:14 +0000 (13:43 -0400)]
intel_idle: enable Atom C6

ATM-C6 was commented out, pending public documentation.

https://bugzilla.kernel.org/show_bug.cgi?id=19762

Tested-by: Dennis Jansen <Dennis.Jansen@...>
Signed-off-by: Len Brown <len.brown@intel.com>
14 years agonet: clear heap allocation for ETHTOOL_GRXCLSRLALL
Kees Cook [Thu, 7 Oct 2010 10:03:48 +0000 (10:03 +0000)]
net: clear heap allocation for ETHTOOL_GRXCLSRLALL

Calling ETHTOOL_GRXCLSRLALL with a large rule_cnt will allocate kernel
heap without clearing it. For the one driver (niu) that implements it,
it will leave the unused portion of heap unchanged and copy the full
contents back to userspace.

Signed-off-by: Kees Cook <kees.cook@canonical.com>
Acked-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>