]> err.no Git - linux-2.6/log
linux-2.6
17 years ago[GFS2] Fix sign problem in quota/statfs and cleanup _host structures
Steven Whitehouse [Fri, 1 Jun 2007 13:11:58 +0000 (14:11 +0100)]
[GFS2] Fix sign problem in quota/statfs and cleanup _host structures

This patch fixes some sign issues which were accidentally introduced
into the quota & statfs code during the endianess annotation process.
Also included is a general clean up which moves all of the _host
structures out of gfs2_ondisk.h (where they should not have been to
start with) and into the places where they are actually used (often only
one place). Also those _host structures which are not required any more
are removed entirely (which is the eventual plan for all of them).

The conversion routines from ondisk.c are also moved into the places
where they are actually used, which for almost every one, was just one
single place, so all those are now static functions. This also cleans up
the end of gfs2_ondisk.h which no longer needs the #ifdef __KERNEL__.

The net result is a reduction of about 100 lines of code, many functions
now marked static plus the bug fixes as mentioned above. For good
measure I ran the code through sparse after making these changes to
check that there are no warnings generated.

This fixes Red Hat bz #239686

Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
17 years ago[GFS2] fix jdata issues
Benjamin Marzinski [Fri, 1 Jun 2007 19:21:38 +0000 (14:21 -0500)]
[GFS2] fix jdata issues

This is a patch for the first three issues of RHBZ #238162

The first issue is that when you allocate a new page for a file, it will not
start off uptodate. This makes sense, since you haven't written anything to that
part of the file yet.  Unfortunately, gfs2_pin() checks to make sure that the
buffers are uptodate.  The solution to this is to mark the buffers uptodate in
gfs2_commit_write(), after they have been zeroed out and have the data written
into them.  I'm pretty confident with this fix, although it's not completely
obvious that there is no problem with marking the buffers uptodate here.

The second issue is simply that you can try to pin a data buffer that is already
on the incore log, and thus, already pinned. This patch checks to see if this
buffer is already on the log, and exits databuf_lo_add() if it is, just like
buf_lo_add() does.

The third issue is that gfs2_log_flush() doesn't do it's block accounting
correctly.  Both metadata and journaled data are logged, but gfs2_log_flush()
only compares the number of metadata blocks with the number of blocks to commit
to the ondisk journal.  This patch also counts the journaled data blocks.

Signed-off-by: Benjamin Marzinski <bmarzins@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
17 years ago[DLM] fix socket shutdown
Patrick Caulfield [Fri, 1 Jun 2007 15:07:26 +0000 (10:07 -0500)]
[DLM] fix socket shutdown

This patch clears the user_data of active sockets as part of cleanup.
This prevents any late-arriving data from trying to add jobs to the work
queue while we are tidying up.

Signed-Off-By: Patrick Caulfield <pcaulfie@redhat.com>
Signed-Off-By: David Teigland <teigland@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
17 years ago[GFS2] Make the log reserved blocks depend on block size
Steven Whitehouse [Fri, 1 Jun 2007 14:19:33 +0000 (15:19 +0100)]
[GFS2] Make the log reserved blocks depend on block size

The number of blocks which we reserve in the log at the start of each
transaction needs to depends upon the block size since the overhead is
related to the number of "pointers" which can be fitted into a single
block.

This relates to Red Hat bz #240435

Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
17 years ago[GFS2] Quotas non-functional - fix another bug
Abhijith Das [Thu, 31 May 2007 22:52:02 +0000 (17:52 -0500)]
[GFS2] Quotas non-functional - fix another bug

This patch fixes a bug where gfs2 was writing update quota usage
information to the wrong location in the quota file.

Signed-off-by: Abhijith Das <adas@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
17 years ago[DLM] show default protocol
David Teigland [Tue, 29 May 2007 13:47:51 +0000 (08:47 -0500)]
[DLM] show default protocol

Display the initial value of the "protocol" config value in configfs.
The default value has always been 0 in the past anyway, so it's always
appeared to be correct.

Signed-off-by: David Teigland <teigland@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
17 years ago[DLM] dumping master locks
David Teigland [Tue, 29 May 2007 13:47:04 +0000 (08:47 -0500)]
[DLM] dumping master locks

Add a new debugfs file that dumps a compact list of mastered locks.
This will be used by a userland daemon to collect state for deadlock
detection.

Also, for the existing function that prints all lock state, lock the rsb
before going through the lock lists since they can be changing in the
course of normal dlm activity.

Signed-off-by: David Teigland <teigland@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
17 years ago[DLM] canceling deadlocked lock
David Teigland [Tue, 29 May 2007 13:46:00 +0000 (08:46 -0500)]
[DLM] canceling deadlocked lock

Add a function that can be used through libdlm by a system daemon to cancel
another process's deadlocked lock.  A completion ast with EDEADLK is returned
to the process waiting for the lock.

Signed-off-by: David Teigland <teigland@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
17 years ago[DLM] timeout fixes
David Teigland [Tue, 29 May 2007 13:44:23 +0000 (08:44 -0500)]
[DLM] timeout fixes

Various fixes related to the new timeout feature:
- add_timeout() missed setting TIMEWARN flag on lkb's when the
  TIMEOUT flag was already set
- clear_proc_locks should remove a dead process's locks from the
  timeout list
- the end-of-life calculation for user locks needs to consider that
  ETIMEDOUT is equivalent to -DLM_ECANCEL
- make initial default timewarn_cs config value visible in configfs
- change bit position of TIMEOUT_CANCEL flag so it's not copied to
  a remote master node
- set timestamp on remote lkb's so a lock dump will display the time
  they've been waiting

Signed-off-by: David Teigland <teigland@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
17 years ago[DLM] Compile fix
Steven Whitehouse [Tue, 29 May 2007 10:14:21 +0000 (11:14 +0100)]
[DLM] Compile fix

A one liner fix which got missed from the earlier patches.

Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
Cc: Fabio Massimo Di Nitto <fabbione@ubuntu.com>
Cc: David Teigland <teigland@redhat.com>
17 years ago[GFS2] latest gfs2-nmw headers break userland build
Fabio Massimo Di Nitto [Tue, 22 May 2007 07:00:24 +0000 (09:00 +0200)]
[GFS2] latest gfs2-nmw headers break userland build

2e8701a15cd6f7c95e74d6660615a69b09e453ef commit breaks libgfs2 build:

gcc -Wall -I/usr/src/ubuntu/mypkgs/rhcluster/cluster/config -DHELPER_PROGRAM
-D_FILE_OFFSET_BITS=64 -DGFS2_RELEASE_NAME=\"2.0\" -ggdb  -I/usr/include
-I../include -I../libgfs2 -c -o gfs2hex.o gfs2hex.c
In file included from hexedit.h:22,
                  from gfs2hex.c:27:
/usr/include/linux/gfs2_ondisk.h:505: error: expected specifier-qualifier-list
before ‘u32’
make[2]: *** [gfs2hex.o] Error 1
make[2]: Leaving directory `/usr/src/ubuntu/mypkgs/rhcluster/cluster/gfs2/edit'
make[1]: *** [all] Error 2
make[1]: Leaving directory `/usr/src/ubuntu/mypkgs/rhcluster/cluster/gfs2'
make: *** [gfs2] Error 2

Signed-off-by: Fabio Massimo Di Nitto <fabbione@ubuntu.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
17 years ago[DLM] fix compile breakage
David Teigland [Fri, 18 May 2007 21:02:57 +0000 (16:02 -0500)]
[DLM] fix compile breakage

In the rush to get the previous patch set sent, a compilation bug I fixed
shortly before sending somehow got clobbered, probably by a missed quilt
refresh or something.

Signed-off-by: David Teigland <teigland@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
17 years ago[DLM] wait for config check during join [6/6]
David Teigland [Fri, 18 May 2007 14:03:35 +0000 (09:03 -0500)]
[DLM] wait for config check during join [6/6]

Joining the lockspace should wait for the initial round of inter-node
config checks to complete before returning.  This way, if there's a
configuration mismatch between the joining node and the existing nodes,
the join can fail and return an error to the application.

Signed-off-by: David Teigland <teigland@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
17 years ago[DLM] fix new_lockspace error exit [5/6]
David Teigland [Fri, 18 May 2007 14:02:20 +0000 (09:02 -0500)]
[DLM] fix new_lockspace error exit [5/6]

Fix the error path when exiting new_lockspace().  It was kfree'ing the
lockspace struct at the end, but that's only valid if it exits before
kobject_register occured.  After kobject_register we have to let the
kobject do the freeing.

Signed-off-by: David Teigland <teigland@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
17 years ago[DLM] cancel in conversion deadlock [4/6]
David Teigland [Fri, 18 May 2007 14:01:26 +0000 (09:01 -0500)]
[DLM] cancel in conversion deadlock [4/6]

When conversion deadlock is detected, cancel the conversion and return
EDEADLK to the application.  This is a new default behavior where before
the dlm would allow the deadlock to exist indefinately.

The DLM_LKF_NODLCKWT flag can now be used in a conversion to prevent the
dlm from performing conversion deadlock detection/cancelation on it.
The DLM_LKF_CONVDEADLK flag can continue to be used as before to tell the
dlm to demote the granted mode of the lock being converted if it gets into
a conversion deadlock.

Signed-off-by: David Teigland <teigland@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
17 years ago[DLM] dlm_device interface changes [3/6]
David Teigland [Fri, 18 May 2007 14:00:32 +0000 (09:00 -0500)]
[DLM] dlm_device interface changes [3/6]

Change the user/kernel device interface used by libdlm:
- Add ability for userspace to check the version of the interface.  libdlm
  can now adapt to different versions of the kernel interface.
- Increase the size of the flags passed in a lock request so all possible
  flags can be used from userspace.
- Add an opaque "xid" value for each lock.  This "transaction id" will be
  used later to associate locks with each other during deadlock detection.
- Add a "timeout" value for each lock.  This is used along with the
  DLM_LKF_TIMEOUT flag.

Also, remove a fragment of unused code in device_read().

This patch requires updating libdlm which is backward compatible with
older kernels.

Signed-off-by: David Teigland <teigland@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
17 years ago[DLM] add lock timeouts and warnings [2/6]
David Teigland [Fri, 18 May 2007 13:59:31 +0000 (08:59 -0500)]
[DLM] add lock timeouts and warnings [2/6]

New features: lock timeouts and time warnings.  If the DLM_LKF_TIMEOUT
flag is set, then the request/conversion will be canceled after waiting
the specified number of centiseconds (specified per lock).  This feature
is only available for locks requested through libdlm (can be enabled for
kernel dlm users if there's a use for it.)

If the new DLM_LSFL_TIMEWARN flag is set when creating the lockspace, then
a warning message will be sent to userspace (using genetlink) after a
request/conversion has been waiting for a given number of centiseconds
(configurable per node).  The time warnings will be used in the future
to do deadlock detection in userspace.

Signed-off-by: David Teigland <teigland@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
17 years ago[DLM] block scand during recovery [1/6]
David Teigland [Fri, 18 May 2007 13:58:15 +0000 (08:58 -0500)]
[DLM] block scand during recovery [1/6]

Don't let dlm_scand run during recovery since it may try to do a resource
directory removal while the directory nodes are changing.

Signed-off-by: David Teigland <teigland@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
17 years ago[DLM] keep dlm from panicing when traversing rsb list in debugfs
Josef Bacik [Wed, 16 May 2007 19:56:13 +0000 (15:56 -0400)]
[DLM] keep dlm from panicing when traversing rsb list in debugfs

This problem was originally reported against GFS6.1, but the same issue exists
in upstream DLM.  This patch keeps the rsb iterator assigning under the rsbtbl
list lock.  Each time we process an rsb we grab a reference to it to make sure
it is not freed out from underneath us, and then put it when we get the next rsb
in the list or move onto another list.

Signed-off-by: Josef Bacik <jwhiter@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
17 years ago[GFS2] Quotas non-functional - fix bug
Abhijith Das [Wed, 16 May 2007 22:02:19 +0000 (17:02 -0500)]
[GFS2] Quotas non-functional - fix bug

This patch fixes an error in the quota code where a 'struct
gfs2_quota_lvb*' was being passed to gfs2_adjust_quota() instead of a
'struct gfs2_quota_data*'. Also moved 'struct gfs2_quota_lvb' from
fs/gfs2/incore.h to include/linux/gfs2_ondisk.h as per Steve's suggestion.

Signed-off-by: Abhijith Das <adas@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
17 years ago[GFS2] Clean up inode number handling
Steven Whitehouse [Tue, 15 May 2007 14:37:50 +0000 (15:37 +0100)]
[GFS2] Clean up inode number handling

This patch cleans up the inode number handling code. The main difference
is that instead of looking up the inodes using a struct gfs2_inum_host
we now use just the no_addr member of this structure. The tests relating
to no_formal_ino can then be done by the calling code. This has
advantages in that we want to do different things in different code
paths if the no_formal_ino doesn't match. In the NFS patch we want to
return -ESTALE, but in the ->lookup() path, its a bug in the fs if the
no_formal_ino doesn't match and thus we can withdraw in this case.

In order to later fix bz #201012, we need to be able to look up an inode
without knowing no_formal_ino, as the only information that is known to
us is the on-disk location of the inode in question.

This patch will also help us to fix bz #236099 at a later date by
cleaning up a lot of the code in that area.

There are no user visible changes as a result of this patch and there
are no changes to the on-disk format either.

Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
17 years ago[GFS2] Reduce size of struct gdlm_lock
Steven Whitehouse [Mon, 14 May 2007 16:43:26 +0000 (17:43 +0100)]
[GFS2] Reduce size of struct gdlm_lock

This patch removes the completion (which is rather large) from struct
gdlm_lock in favour of using the wait_on_bit() functions. We don't need
to add any extra fields to the structure to do this, so we save 32 bytes
(on x86_64) per structure. This adds up to quite a lot when we may
potentially have millions of these lock structures,

Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
Acked-by: David Teigland <teigland@redhat.com>
17 years ago[GFS2] Addendum patch 2 for gfs2_grow
Robert Peterson [Mon, 14 May 2007 17:42:18 +0000 (12:42 -0500)]
[GFS2] Addendum patch 2 for gfs2_grow

This addendum patch 2 corrects three things:

1. It fixes a stupid mistake in the previous addendum that broke gfs2.
   Ref: https://www.redhat.com/archives/cluster-devel/2007-May/msg00162.html
2. It fixes a problem that Dave Teigland pointed out regarding the
   external declarations in ops_address.h being in the wrong place.
3. It recasts a couple more %llu printks to (unsigned long long)
   as requested by Steve Whitehouse.

I would have loved to put this all in one revised patch, but there was
a rush to get some patches for RHEL5. Therefore, the previous patches
were applied to the git tree "as is" and therefore, I'm posting another
addendum.  Sorry.

Signed-off-by: Bob Peterson <rpeterso@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
17 years ago[GFS2] use zero_user_page
Nate Diller [Fri, 11 May 2007 05:41:28 +0000 (22:41 -0700)]
[GFS2] use zero_user_page

Use zero_user_page() instead of open-coding it.

Signed-off-by: Nate Diller <nate.diller@gmail.com>
Cc: Steven Whitehouse <swhiteho@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
17 years ago[GFS2] Kernel changes to support new gfs2_grow command (part 2)
Robert Peterson [Thu, 10 May 2007 21:54:38 +0000 (16:54 -0500)]
[GFS2] Kernel changes to support new gfs2_grow command (part 2)

To avoid code redundancy, I separated out the operational "guts" into
a new function called read_rindex_entry.  Then I made two functions:
the closer-to-original gfs2_ri_update (without the special condition
checks) and gfs2_ri_update_special that's designed with that condition
in mind.  (I don't like the name, but if you have a suggestion, I'm
all ears).

Oh, and there's an added benefit:  we don't need all the ugly gotos
anymore.  ;)

This patch has been tested with gfs2_fsck_hellfire (which runs for
three and a half hours, btw).

Signed-off-By: Bob Peterson <rpeterso@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
17 years ago[GFS2] kernel changes to support new gfs2_grow command
Robert Peterson [Wed, 9 May 2007 14:37:57 +0000 (09:37 -0500)]
[GFS2] kernel changes to support new gfs2_grow command

This is another revision of my gfs2 kernel patch that allows
gfs2_grow to function properly.

Steve Whitehouse expressed some concerns about the previous
patch and I restructured it based on his comments.
The previous patch was doing the statfs_change at file close time,
under its own transaction.  The current patch does the statfs_change
inside the gfs2_commit_write function, which keeps it under the
umbrella of the inode transaction.

I can't call ri_update to re-read the rindex file during the
transaction because the transaction may have outstanding unwritten
buffers attached to the rgrps that would be otherwise blown away.
So instead, I created a new function, gfs2_ri_total, that will
re-read the rindex file just to total the file system space
for the sake of the statfs_change.  The ri_update will happen
later, when gfs2 realizes the version number has changed, as it
happened before my patch.

Since the statfs_change is happening at write_commit time and there
may be multiple writes to the rindex file for one grow operation.
So one consequence of this restructuring is that instead of getting
one kernel message to indicate the change, you may see several.
For example, before when you did a gfs2_grow, you'd get a single
message like:

GFS2: File system extended by 247876 blocks (968MB)

Now you get something like:

GFS2: File system extended by 207896 blocks (812MB)
GFS2: File system extended by 39980 blocks (156MB)

This version has also been successfully run against the hours-long
"gfs2_fsck_hellfire" test that does several gfs2_grow and gfs2_fsck
while interjecting file system damage.  It does this repeatedly
under a variety Resource Group conditions.

Signed-off-By: Bob Peterson <rpeterso@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
17 years ago[DLM] fix a couple of races
Satyam Sharma [Tue, 8 May 2007 08:18:58 +0000 (09:18 +0100)]
[DLM] fix a couple of races

Fix two races in fs/dlm/config.c:

(1) Grab the configfs subsystem semaphore before calling
config_group_find_obj() in get_space(). This solves a potential race
between get_space() and concurrent mkdir(2) or rmdir(2).

(2) Grab a reference on the found config_item _while_ holding the configfs
subsystem semaphore in get_comm(), and not after it. This solves a
potential race between get_comm() and concurrent rmdir(2).

Signed-off-by: Satyam Sharma <ssatyam@cse.iitk.ac.in>
Signed-off-by: David Teigland <teigland@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
17 years ago[GFS2] flush the glock completely in inode_go_sync
Benjamin Marzinski [Wed, 2 May 2007 14:44:03 +0000 (09:44 -0500)]
[GFS2] flush the glock completely in inode_go_sync

Fix for bz #231910
When filemap_fdatawrite() is called on the inode mapping in data=ordered mode,
it will add the glock to the log. In inode_go_sync(), if you do the
gfs2_log_flush() before this, after the filemap_fdatawrite() call, the glock
and its associated data buffers will be on the log again. This means you can
demote a lock from exclusive, without having it flushed from the log. The
attached patch simply moves the gfs2_log_flush up to after the
filemap_fdatawrite() call.

Originally, I tried moving the gfs2_log_flush to after gfs2_meta_sync(), but
that caused me to trip the following assert.

GFS2: fsid=cypher-36:test.0: fatal: assertion "!buffer_busy(bh)" failed
GFS2: fsid=cypher-36:test.0:   function = gfs2_ail_empty_gl, file = fs/gfs2/glops.c, line = 61

It appears that gfs2_log_flush() puts some of the glocks buffers in the busy
state and the filemap_fdatawrite() call is necessary to flush them. This makes
me worry slightly that a related problem could happen because of moving the
gfs2_log_flush() after the initial filemap_fdatawrite(), but I assume that
gfs2_ail_empty_gl() would catch that case as well.

Signed-off-by: Benjamin E. Marzinski <bmarzins@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
17 years agoLinux 2.6.22 v2.6.22
Linus Torvalds [Sun, 8 Jul 2007 23:32:17 +0000 (16:32 -0700)]
Linux 2.6.22

Woo-hoo. I'm sure somebody will report a "this doesn't compile, and
I have a new root exploit" five minutes after release, but it still
feels good ;)

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoMerge master.kernel.org:/pub/scm/linux/kernel/git/bart/ide-2.6
Linus Torvalds [Sun, 8 Jul 2007 19:14:27 +0000 (12:14 -0700)]
Merge master.kernel.org:/pub/scm/linux/kernel/git/bart/ide-2.6

* master.kernel.org:/pub/scm/linux/kernel/git/bart/ide-2.6:
  qd65xx: fix PIO mode selection
  sis5513: adding PCI-ID

17 years agoFix permission checking for the new utimensat() system call
Linus Torvalds [Sun, 8 Jul 2007 19:02:55 +0000 (12:02 -0700)]
Fix permission checking for the new utimensat() system call

Commit 1c710c896eb461895d3c399e15bb5f20b39c9073 added the utimensat()
system call, but didn't handle the case of checking for the writability
of the target right, when the target was a file descriptor, not a
filename.

We cannot use vfs_permission(MAY_WRITE) for that case, and need to
simply check whether the file descriptor is writable.  The oops from
using the wrong function was noticed and narrowed down by Markus
Trippelsdorf.

Cc: Ulrich Drepper <drepper@redhat.com>
Cc: Markus Trippelsdorf <markus@trippelsdorf.de>
Cc: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Al Viro <viro@ftp.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agomm: double mark_page_accessed() in read_cache_page_async()
Peter Zijlstra [Sun, 8 Jul 2007 08:13:06 +0000 (01:13 -0700)]
mm: double mark_page_accessed() in read_cache_page_async()

Fix a post-2.6.21 regression.

read_cache_page_async() has two invocations of mark_page_accessed() which will
launch pages right onto the active list.

Remove the first one, keeping the latter one.  This avoids marking unwanted
pages active (in the retry loop).

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoqd65xx: fix PIO mode selection
Bartlomiej Zolnierkiewicz [Sun, 8 Jul 2007 13:21:58 +0000 (15:21 +0200)]
qd65xx: fix PIO mode selection

PIO4 is a maximum PIO mode supported by a driver.  Using "255" as a max_mode
argument to ide_get_best_pio_mode() could result in wrong timings being used
by a driver (for "pio" equal to 5) or OOPS (for "pio" values > 5 && < 255).

Signed-off-by: Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>
Acked-by: Sergei Shtylyov <sshtylyov@ru.mvista.com>
Reviewed-by: Alan Cox <alan@lxorguk.ukuu.org.uk>
17 years agosis5513: adding PCI-ID
Uwe Koziolek [Sun, 8 Jul 2007 13:21:58 +0000 (15:21 +0200)]
sis5513: adding PCI-ID

The SiS966 has one additional PCI-ID 1180.

If the chipset is using this PCI-ID, the primary channel is connected to the
first PATA-port. The secondary channel is connected to SATA-ports in IDE
emulation mode.  The legacy IO-ports are used.

The including of the PCI-ID into pata_sis is not sufficient, because the legacy
driver in drivers/ide is initialized before pata_sis.

Signed-off-by: Uwe Koziolek <uwe.koziolek@gmx.net>
Signed-off-by: Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>
17 years agoDLM must depend on SYSFS
Adrian Bunk [Sat, 7 Jul 2007 04:14:02 +0000 (06:14 +0200)]
DLM must depend on SYSFS

The dependency of DLM on SYSFS got lost in
commit 6ed7257b46709e87d79ac2b6b819b7e0c9184998 resulting in the
following compile error with CONFIG_DLM=y, CONFIG_SYSFS=n:

<--  snip  -->

...
  LD      .tmp_vmlinux1
fs/built-in.o: In function `dlm_lockspace_init':
/home/bunk/linux/kernel-2.6/linux-2.6.22-rc6-mm1/fs/dlm/lockspace.c:231: undefined reference to `kernel_subsys'
fs/built-in.o: In function `configfs_init':
/home/bunk/linux/kernel-2.6/linux-2.6.22-rc6-mm1/fs/configfs/mount.c:143: undefined reference to `kernel_subsys'
make[1]: *** [.tmp_vmlinux1] Error 1

<--  snip  -->

Signed-off-by: Adrian Bunk <bunk@stusta.de>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoClean up E7520/7320/7525 quirk printk.
Dave Jones [Fri, 6 Jul 2007 18:59:43 +0000 (14:59 -0400)]
Clean up E7520/7320/7525 quirk printk.

The printk level in this printk is bogus, as the previous printk
didn't have a terminating \n resulting in ..

Intel E7520/7320/7525 detected.<6>Disabling irq balancing and affinity

It also never printed a \n at all in the case where we didn't do
the quirk.

Change it to only make noise if it actually does something useful.

Signed-off-by: Dave Jones <davej@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoinclude/linux/kallsyms.h must #include <linux/errno.h>
Adrian Bunk [Fri, 6 Jul 2007 22:54:09 +0000 (00:54 +0200)]
include/linux/kallsyms.h must #include <linux/errno.h>

This patch fixes the following 2.6.22 regression with CONFIG_KALLSYMS=n:

<--  snip  -->

...
  CC      arch/m32r/kernel/traps.o
In file included from /home/bunk/linux/kernel-2.6/linux-2.6.22-rc6-mm1/arch/m32r/kernel/traps.c:14:
/home/bunk/linux/kernel-2.6/linux-2.6.22-rc6-mm1/include/linux/kallsyms.h: In function 'lookup_symbol_name':
/home/bunk/linux/kernel-2.6/linux-2.6.22-rc6-mm1/include/linux/kallsyms.h:66: error: 'ERANGE' undeclared (first use in this function)
/home/bunk/linux/kernel-2.6/linux-2.6.22-rc6-mm1/include/linux/kallsyms.h:66: error: (Each undeclared identifier is reported only once
/home/bunk/linux/kernel-2.6/linux-2.6.22-rc6-mm1/include/linux/kallsyms.h:66: error: for each function it appears in.)
/home/bunk/linux/kernel-2.6/linux-2.6.22-rc6-mm1/include/linux/kallsyms.h: In function 'lookup_symbol_attrs':
/home/bunk/linux/kernel-2.6/linux-2.6.22-rc6-mm1/include/linux/kallsyms.h:71: error: 'ERANGE' undeclared (first use in this function)
make[2]: *** [arch/m32r/kernel/traps.o] Error 1

<--  snip  -->

Signed-off-by: Adrian Bunk <bunk@stusta.de>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoFix use-after-free oops in Bluetooth HID.
David Woodhouse [Sat, 7 Jul 2007 18:58:39 +0000 (14:58 -0400)]
Fix use-after-free oops in Bluetooth HID.

When cleaning up HIDP sessions, we currently close the ACL connection
before deregistering the input device. Closing the ACL connection
schedules a workqueue to remove the associated objects from sysfs, but
the input device still refers to them -- and if the workqueue happens to
run before the input device removal, the kernel will oops when trying to
look up PHYSDEVPATH for the removed input device.

Fix this by deregistering the input device before closing the
connections.

Signed-off-by: David Woodhouse <dwmw2@infradead.org>
Acked-by: Marcel Holtmann <marcel@holtmann.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoslub: remove useless EXPORT_SYMBOL
Christoph Lameter [Fri, 6 Jul 2007 18:17:32 +0000 (11:17 -0700)]
slub: remove useless EXPORT_SYMBOL

kmem_cache_open is static. EXPORT_SYMBOL was leftover from some earlier
time period where kmem_cache_open was usable outside of slub.

(Fixes powerpc build error)

Signed-off-by: Chrsitoph Lameter <clameter@sgi.com>
Cc: Johannes Berg <johannes@sipsolutions.net>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoMAINTAINERS new kernel janitors ml
maximilian attems [Fri, 6 Jul 2007 18:17:32 +0000 (11:17 -0700)]
MAINTAINERS new kernel janitors ml

davem kindly moved the list from osdl to vger.

Signed-of-by: maximilian attems <max@stro.at>
Cc: Alexey Dobriyan <adobriyan@gmail.com>
Cc: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoGEODE: reboot fixup for geode machines with CS5536 boards
Andres Salomon [Fri, 6 Jul 2007 18:17:30 +0000 (11:17 -0700)]
GEODE: reboot fixup for geode machines with CS5536 boards

Writing to MSR 0x51400017 forces a hard reset on CS5536-based machines,
this has the reboot fixup do just that if such a board is detected.

Acked-by: Jordan Crouse <jordan.crouse@amd.com>
Signed-off-by: Andres Salomon <dilinger@debian.org>
Cc: Alan Cox <alan@lxorguk.ukuu.org.uk>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoMerge branch 'master' of master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6
Linus Torvalds [Fri, 6 Jul 2007 17:30:12 +0000 (10:30 -0700)]
Merge branch 'master' of master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6

* 'master' of master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6:
  [NETPOLL]: Fixups for 'fix soft lockup when removing module'
  [NET]: net/core/netevent.c should #include <net/netevent.h>
  [NETFILTER]: nf_conntrack_h323: add checking of out-of-range on choices' index values
  [NET] skbuff: remove export of static symbol
  SCTP: Add scope_id validation for link-local binds
  SCTP: Check to make sure file is valid before setting timeout
  SCTP: Fix thinko in sctp_copy_laddrs()

17 years agoMerge branch 'upstream' of git://ftp.linux-mips.org/pub/scm/upstream-linus
Linus Torvalds [Fri, 6 Jul 2007 17:29:33 +0000 (10:29 -0700)]
Merge branch 'upstream' of git://ftp.linux-mips.org/pub/scm/upstream-linus

* 'upstream' of git://ftp.linux-mips.org/pub/scm/upstream-linus:
  [MIPS] Fix scheduling latency issue on 24K, 34K and 74K cores
  [MIPS] Add macros to encode processor revisions.
  [MIPS] RM7000: Enable ICACHE_REFILLS_WORKAROUND_WAR.
  [MIPS] SMTC: Fix cut'n'paste bug in Kconfig.debug
  [MIPS] Change libgcc-style functions from lib-y to obj-y
  [MIPS] Fix timer/performance interrupt detection
  [MIPS] AP/SP: Avoid triggering the 34K E125 performance issue
  [MIPS] 64-bit TO_PHYS_MASK macro for RM9000 processors

17 years agomm: fixup /proc/vmstat output
Peter Zijlstra [Fri, 6 Jul 2007 11:35:34 +0000 (13:35 +0200)]
mm: fixup /proc/vmstat output

Line up the vmstat_text with zone_stat_item

enum zone_stat_item {
/* First 128 byte cacheline (assuming 64 bit words) */
NR_FREE_PAGES,
NR_INACTIVE,
NR_ACTIVE,

We current have nr_active and nr_inactive reversed.

[ "OK with patch, though using initializers canbe handy to prevent such
   things in future:

static const char * const vmstat_text[] = {
[NR_FREE_PAGES] = "nr_free_pages",
..."
 - Alexey ]

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Alexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agopotential compiler error, irqfunc caller sites update
Yoann Padioleau [Fri, 6 Jul 2007 09:39:56 +0000 (02:39 -0700)]
potential compiler error, irqfunc caller sites update

In 7d12e780e003f93433d49ce78cfedf4b4c52adc5 David Howells performed
this evolution:
 "IRQ: Maintain regs pointer globally rather than passing to IRQ handlers"

He correctly updated many of the function definitions that were using this
extra regs pointer parameter but forgot to update some caller sites of
those functions.  The reason the modifications was not properly done on all
drivers is that some drivers were rarely compiled because they are for
AMIGA, or that some code sites were inside #ifdefs where the option is not
set or inside #if 0.

Here is the semantic patch that found the occurences
and fixed the problem.

@ rule1 @
identifier fn;
identifier irq, dev_id;
typedef irqreturn_t;
@@

static irqreturn_t fn(int irq, void *dev_id)
{
   ...
}

@@
identifier rule1.fn;
expression E1, E2, E3;
@@

 fn(E1, E2
-   ,E3
   )

Signed-off-by: Yoann Padioleau <padator@wanadoo.fr>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Jeff Garzik <jeff@garzik.org>
Cc: Greg KH <greg@kroah.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoi386: es7000 build breakage fix
Vivek Goyal [Fri, 6 Jul 2007 09:39:55 +0000 (02:39 -0700)]
i386: es7000 build breakage fix

o Commit 1833d6bc72893265f22addd79cf52e6987496e0f broke the build if
  compiled with CONFIG_ES7000=y and CONFIG_X86_GENERICARCH=n

arch/i386/kernel/built-in.o(.init.text+0x4fa9): In function `acpi_parse_madt':
: undefined reference to `acpi_madt_oem_check'
arch/i386/kernel/built-in.o(.init.text+0x7406): In function `smp_read_mpc':
: undefined reference to `mps_oem_check'
arch/i386/kernel/built-in.o(.init.text+0x8990): In function
`connect_bsp_APIC':
: undefined reference to `enable_apic_mode'
make: *** [.tmp_vmlinux1] Error 1

o Fix the build issue. Provided the definitions of missing functions.

o Don't have ES7000 machine. Only compile tested.

Cc: Len Brown <lenb@kernel.org>
Cc: Natalie Protasevich <protasnb@gmail.com>
Cc: Roland Dreier <rolandd@cisco.com>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoPNP SMCf010 quirk: work around Toshiba Portege 4000 ACPI issues
Bjorn Helgaas [Fri, 6 Jul 2007 09:39:54 +0000 (02:39 -0700)]
PNP SMCf010 quirk: work around Toshiba Portege 4000 ACPI issues

When we enable the SMCf010 IR device, the Toshiba Portege 4000 BIOS claims
the device is working, but it really isn't configured correctly.  The BIOS
*will* configure it, but only if we call _SRS after (1) reversing the order
of the SIR and FIR I/O port regions and (2) changing the IRQ from
active-high to active-low.

This patch addresses the 2.6.22 regression:
    "no irda0 interface (2.6.21 was OK), smsc does not find chip"

I tested this on a Portege 4000.  The smsc-ircc2 driver correctly detects
the device, and "irattach irda0 -s && irdadump" shows transmitted and
received packets.

Signed-off-by: Bjorn Helgaas <bjorn.helgaas@hp.com>
Cc: Andrey Borzenkov <arvidjaar@mail.ru>
Cc: Samuel Ortiz <samuel@sortiz.org>
Cc: "Linus Walleij (LD/EAB)" <linus.walleij@ericsson.com>
Cc: Michal Piotrowski <michal.k.k.piotrowski@gmail.com>
Cc: Adam Belay <ambx1@neo.rr.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agofix logic error in ipc compat semctl()
Alexander Graf [Fri, 6 Jul 2007 09:39:53 +0000 (02:39 -0700)]
fix logic error in ipc compat semctl()

When calling a semctl(IPC_STAT) without IPC_64 the check if the memory is
unevaluated.  This patch fixes this.

Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agox86_64: fix headers_install
David Woodhouse [Fri, 6 Jul 2007 09:39:52 +0000 (02:39 -0700)]
x86_64: fix headers_install

A bug in headers_install for ARCH=x86_64 yields an asm/ directory full of
files all of which are using the same #ifdef guard, "__ASM_STUB_" with no
postfix.  So the second and later asm files #included in the same C file
(often through standard headers like ioctl.h) yields no symbols.

Strangeness with the Ubuntu 'tell me if I support something that's not
explcitly mentioned in POSIX, and I'll strip it out' shell, I believe.

We don't need the 'export' but we do need a semicolon at the end of the
FNAME line:

Signed-off-by: David Woodhouse <dwmw2@infradead.org>
Signed-off-by: Rob Landley <rob@landley.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoMTRR: Fix race causing set_mtrr to go into infinite loop
Loic Prylli [Fri, 6 Jul 2007 09:39:52 +0000 (02:39 -0700)]
MTRR: Fix race causing set_mtrr to go into infinite loop

Processors synchronization in set_mtrr requires the .gate field to be set
after .count field is properly initialized.  Without an explicit barrier,
the compiler was reordering those memory stores.  That was sometimes
causing a processor (in ipi_handler) to see the .gate change and decrement
.count before the latter is set by set_mtrr() (which then hangs in a
infinite loop with irqs disabled).

Signed-off-by: Loic Prylli <loic@myri.com>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoi386: fix regression, endless loop in ptrace singlestep over an int80
Jason Wessel [Fri, 6 Jul 2007 09:39:50 +0000 (02:39 -0700)]
i386: fix regression, endless loop in ptrace singlestep over an int80

The commit 635cf99a80f4ebee59d70eb64bb85ce829e4591f introduced a
regression.  Executing a ptrace single step after certain int80
accesses will infinitely loop and never advance the PC.

The TIF_SINGLESTEP check should be done on the return from the syscall
and not before it.

I loops on each single step on the pop right after the int80 which writes out
to the console.  At that point you can issue as many single steps as you want
and it will not advance any further.

The test case is below:

/* Test whether singlestep through an int80 syscall works.
 */
#define _GNU_SOURCE
#include <stdio.h>
#include <unistd.h>
#include <fcntl.h>
#include <sys/ptrace.h>
#include <sys/wait.h>
#include <sys/mman.h>
#include <asm/user.h>
#include <string.h>

static int child, status;
static struct user_regs_struct regs;

static void do_child()
{
char str[80] = "child: int80 test\n";

ptrace(PTRACE_TRACEME, 0, 0, 0);
kill(getpid(), SIGUSR1);
write(fileno(stdout),str,strlen(str));
asm ("int $0x80" : : "a" (20)); /* getpid */
}

static void do_parent()
{
unsigned long eip, expected = 0;
again:
waitpid(child, &status, 0);
if (WIFEXITED(status) || WIFSIGNALED(status))
return;

if (WIFSTOPPED(status)) {
ptrace(PTRACE_GETREGS, child, 0, &regs);
eip = regs.eip;
if (expected)
fprintf(stderr, "child stop @ %08lx, expected %08lx %s\n",
eip, expected,
eip == expected ? "" : " <== ERROR");

if (*(unsigned short *)eip == 0x80cd) {
fprintf(stderr, "int 0x80 at %08x\n", (unsigned int)eip);
expected = eip + 2;
} else
expected = 0;

ptrace(PTRACE_SINGLESTEP, child, NULL, NULL);
}
goto again;
}

int main(int argc, char * const argv[])
{
child = fork();
if (child)
do_parent();
else
do_child();
return 0;
}

Signed-off-by: Jason Wessel <jason.wessel@windriver.com>
Cc: Jeremy Fitzhardinge <jeremy@goop.org>
Cc: <stable@kernel.org>
Cc: Chuck Ebbert <76306.1226@compuserve.com>
Acked-by: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoFix elf_core_dump() when writing arch specific notes (spu coredumps)
Michael Ellerman [Fri, 6 Jul 2007 09:39:49 +0000 (02:39 -0700)]
Fix elf_core_dump() when writing arch specific notes (spu coredumps)

elf_core_dump() supports dumping arch specific ELF notes, via the #define
ELF_CORE_WRITE_EXTRA_NOTES.  Currently the only user of this is the powerpc
spu coredump code.

There is a bug in the handling of foffset WRT the arch notes, which causes
us to erroneously increment foffset by the size of the arch notes, leaving
a block of zeroes in the file, and causing all subsequent data in the file
to be at <supposed position> + <arch note size>.  eg:

  LOAD  0x050000 0x00100000 0x00000000 0x20000 0x20000 R E 0x10000

Tells us we should have a chunk of data at 0x50000.  The truth is the data
is at 0x90dbc = 0x50000 + 0x40dbc (the size of the arch notes).

This bug prevents gdb from reading the core file correctly.

The simplest fix is to simply remember the size of the arch notes, and add
it to foffset after we've written the arch notes.  The only drawback is
that if the arch code doesn't write as many bytes as it said it would, we
end up with a broken core dump again.  For now I think that's a reasonable
requirement.

Tested on a Cell blade, gdb no longer complains about the core file being
bogus.

While I'm here I should point out that the spu coredump code does not work
if we're dumping to a pipe - we'll have to wait for 23 to fix that.

Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Acked-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Acked-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years ago[MIPS] Fix scheduling latency issue on 24K, 34K and 74K cores
Ralf Baechle [Wed, 20 Jun 2007 23:22:34 +0000 (00:22 +0100)]
[MIPS] Fix scheduling latency issue on 24K, 34K and 74K cores

The idle loop goes to sleep using the WAIT instruction if !need_resched().
This has is suffering from from a race condition that if if just after
need_resched has returned 0 an interrupt might set TIF_NEED_RESCHED but
we've just completed the test so go to sleep anyway.  This would be
trivial to fix by just disabling interrupts during that sequence as in:

        local_irq_disable();
        if (!need_resched())
                __asm__("wait");
        local_irq_enable();

but the processor architecture leaves it undefined if a processor calling
WAIT with interrupts disabled will ever restart its pipeline and indeed
some processors have made use of the freedom provided by the architecture
definition.  This has been resolved and the Config7.WII bit indicates that
the use of WAIT is safe on 24K, 24KE and 34K cores.  It also is safe on
74K starting revision 2.1.0 so enable the use of WAIT with interrupts
disabled for 74K based on a c0_prid of at least that.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
17 years ago[MIPS] Add macros to encode processor revisions.
Ralf Baechle [Fri, 6 Jul 2007 13:40:05 +0000 (14:40 +0100)]
[MIPS] Add macros to encode processor revisions.

Older processors used to encode processor version and revision in two
4-bit bitfields, the 4K seems to simply count up and even newer MTI cores
have switched to use the 8-bits as 3:3:2 bitfield with the last field as
the patch number.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
17 years ago[MIPS] RM7000: Enable ICACHE_REFILLS_WORKAROUND_WAR.
Ralf Baechle [Thu, 5 Jul 2007 07:14:21 +0000 (08:14 +0100)]
[MIPS] RM7000: Enable ICACHE_REFILLS_WORKAROUND_WAR.

The RM7000 processors and the E9000 cores have a bug (though PMC-Sierra
opposes it being called that) where invalid instructions in the same
I-cache line worth of instructions being fetched may case spurious
exceptions.

The workaround for this was only enabled for E9000 cores; enable it also
for all RM7000-based platforms.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
17 years ago[MIPS] SMTC: Fix cut'n'paste bug in Kconfig.debug
Ralf Baechle [Thu, 5 Jul 2007 16:39:48 +0000 (17:39 +0100)]
[MIPS] SMTC: Fix cut'n'paste bug in Kconfig.debug

This effectivly turned the SMTC_IDLE_HOOK_DEBUG debug option into a no-op.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
17 years ago[MIPS] Change libgcc-style functions from lib-y to obj-y
Ralf Baechle [Tue, 26 Jun 2007 18:19:00 +0000 (20:19 +0200)]
[MIPS] Change libgcc-style functions from lib-y to obj-y

Reported by Eugene Surovegin <ebs@ebshome.net>.

If only modules were users of these functions they did not get linked into
the kernel proper, so later module loads would fail as well.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
17 years ago[MIPS] Fix timer/performance interrupt detection
Chris Dearman [Thu, 21 Jun 2007 11:59:57 +0000 (12:59 +0100)]
[MIPS] Fix timer/performance interrupt detection

Signed-off-by: Chris Dearman <chris@mips.com>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
17 years ago[MIPS] AP/SP: Avoid triggering the 34K E125 performance issue
Ralf Baechle [Thu, 7 Jun 2007 07:44:32 +0000 (08:44 +0100)]
[MIPS] AP/SP: Avoid triggering the 34K E125 performance issue

C0_status doesn't need to be initialized at this point anyway; the register
will be initialized later.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
17 years ago[MIPS] 64-bit TO_PHYS_MASK macro for RM9000 processors
Andrew Sharp [Fri, 23 Mar 2007 19:15:18 +0000 (12:15 -0700)]
[MIPS] 64-bit TO_PHYS_MASK macro for RM9000 processors

Signed-off-by: Andrew Sharp <tigerand@gmail.com>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
17 years ago[NETPOLL]: Fixups for 'fix soft lockup when removing module'
Jarek Poplawski [Fri, 6 Jul 2007 00:42:44 +0000 (17:42 -0700)]
[NETPOLL]: Fixups for 'fix soft lockup when removing module'

>From my recent patch:

> >    #1
> >    Until kernel ver. 2.6.21 (including) cancel_rearming_delayed_work()
> >    required a work function should always (unconditionally) rearm with
> >    delay > 0 - otherwise it would endlessly loop. This patch replaces
> >    this function with cancel_delayed_work(). Later kernel versions don't
> >    require this, so here it's only for uniformity.

But Oleg Nesterov <oleg@tv-sign.ru> found:

> But 2.6.22 doesn't need this change, why it was merged?
>
> In fact, I suspect this change adds a race,
...

His description was right (thanks), so this patch reverts #1.

Signed-off-by: Jarek Poplawski <jarkao2@o2.pl>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 years ago[NET]: net/core/netevent.c should #include <net/netevent.h>
Adrian Bunk [Fri, 6 Jul 2007 00:06:21 +0000 (17:06 -0700)]
[NET]: net/core/netevent.c should #include <net/netevent.h>

Every file should include the headers containing the prototypes for
its global functions.

Signed-off-by: Adrian Bunk <bunk@stusta.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 years ago[NETFILTER]: nf_conntrack_h323: add checking of out-of-range on choices' index values
Jing Min Zhao [Fri, 6 Jul 2007 00:05:01 +0000 (17:05 -0700)]
[NETFILTER]: nf_conntrack_h323: add checking of out-of-range on choices' index values

Choices' index values may be out of range while still encoded in the fixed
length bit-field. This bug may cause access to undefined types (NULL
pointers) and thus crashes (Reported by Zhongling Wen).

This patch also adds checking of decode flag when decoding SEQUENCEs.

Signed-off-by: Jing Min Zhao <zhaojingmin@vivecode.com>
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 years ago[NET] skbuff: remove export of static symbol
Johannes Berg [Fri, 6 Jul 2007 00:03:09 +0000 (17:03 -0700)]
[NET] skbuff: remove export of static symbol

skb_clone_fraglist is static so it shouldn't be exported.

Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 years agoSCTP: Add scope_id validation for link-local binds
Vlad Yasevich [Tue, 3 Jul 2007 18:29:23 +0000 (14:29 -0400)]
SCTP: Add scope_id validation for link-local binds

SCTP currently permits users to bind to link-local addresses,
but doesn't verify that the scope id specified at bind matches
the interface that the address is configured on.  It was report
that this can hang a system.

Signed-off-by: Vlad Yasevich <vladislav.yasevich@hp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 years agoSCTP: Check to make sure file is valid before setting timeout
Vlad Yasevich [Tue, 3 Jul 2007 16:47:40 +0000 (12:47 -0400)]
SCTP: Check to make sure file is valid before setting timeout

In-kernel sockets created with sock_create_kern don't usually
have a file and file descriptor allocated to them.  As a result,
when SCTP tries to check the non-blocking flag, we Oops when
dereferencing a NULL file pointer.

Signed-off-by: Vlad Yasevich <vladislav.yasevich@hp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 years agoSCTP: Fix thinko in sctp_copy_laddrs()
Vlad Yasevich [Tue, 3 Jul 2007 16:43:12 +0000 (12:43 -0400)]
SCTP: Fix thinko in sctp_copy_laddrs()

Correctly dereference bytes_copied in sctp_copy_laddrs().
I totally must have spaced when doing this.

Signed-off-by: Vlad Yasevich <vladislav.yasevich@hp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
17 years agoMerge master.kernel.org:/home/rmk/linux-2.6-arm
Linus Torvalds [Thu, 5 Jul 2007 23:10:16 +0000 (16:10 -0700)]
Merge master.kernel.org:/home/rmk/linux-2.6-arm

* master.kernel.org:/home/rmk/linux-2.6-arm:
  [ARM] always allow dump_stack() to produce a backtrace
  [ARM] Fix non-page aligned boot time mappings
  [ARM] 4458/1: pxa: Fix CKEN usage and hence fix pxa suspend/resume
  [ARM] 4454/1: Use word accesses in Versatile PCI config reads

17 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input
Linus Torvalds [Thu, 5 Jul 2007 22:55:37 +0000 (15:55 -0700)]
Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input:
  Input: document some of keycodes
  Input: add a new EV_SW SW_RADIO event, for radio switches on laptops
  Input: serio - take drv_mutex in serio_cleanup()
  Input: atkbd - use printk_ratelimit for spurious ACK messages
  Input: atkbd - throttle LED switching
  Input: i8042 - add HP Pavilion ZT1000 to the MUX blacklist

17 years agoMerge branch 'merge' of git://git.kernel.org/pub/scm/linux/kernel/git/paulus/powerpc
Linus Torvalds [Thu, 5 Jul 2007 22:55:00 +0000 (15:55 -0700)]
Merge branch 'merge' of git://git.kernel.org/pub/scm/linux/kernel/git/paulus/powerpc

* 'merge' of git://git.kernel.org/pub/scm/linux/kernel/git/paulus/powerpc:
  [POWERPC] Update defconfigs
  [POWERPC] Uninline and export virq_to_hw() for the pasemi_mac driver
  [POWERPC] Fix PMI breakage in cbe_cbufreq driver
  [POWERPC] Disable old EMAC driver in arch/powerpc

17 years agoFix slab redzone alignment
David Woodhouse [Thu, 5 Jul 2007 01:26:44 +0000 (21:26 -0400)]
Fix slab redzone alignment

Commit b46b8f19c9cd435ecac4d9d12b39d78c137ecd66 fixed a couple of bugs
by switching the redzone to 64 bits. Unfortunately, it neglected to
ensure that the _second_ redzone, after the slab object, is aligned
correctly. This caused illegal instruction faults on sparc32, which for
some reason not entirely clear to me are not trapped and fixed up.

Two things need to be done to fix this:
  - increase the object size, rounding up to alignof(long long) so
    that the second redzone can be aligned correctly.
  - If SLAB_STORE_USER is set but alignof(long long)==8, allow a
    full 64 bits of space for the user word at the end of the buffer,
    even though we may not _use_ the whole 64 bits.

This patch should be a no-op on any 64-bit architecture or any 32-bit
architecture where alignof(long long) == 4. Of the others, it's tested
on ppc32 by myself and a very similar patch was tested on sparc32 by
Mark Fortescue, who reported the new problem.

Also, fix the conditions for FORCED_DEBUG, which hadn't been adjusted to
the new sizes. Again noticed by Mark.

Signed-off-by: David Woodhouse <dwmw2@infradead.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years ago[ARM] always allow dump_stack() to produce a backtrace
Russell King [Thu, 5 Jul 2007 18:59:51 +0000 (19:59 +0100)]
[ARM] always allow dump_stack() to produce a backtrace

Don't make this dependent on CONFIG_DEBUG_KERNEL - if we hit a WARN_ON
we need the stack trace to work out how we got to that point.

Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
17 years agoRemove the blink driver
Linus Torvalds [Wed, 4 Jul 2007 22:24:39 +0000 (15:24 -0700)]
Remove the blink driver

Yeah, we could have just disabled it, but there's work on a new one that
isn't as fundamentally broken, so there really doesn't seem to be any
point in keeping it around.

The recent timer cleanup broke the only valid use, and when I say
"valid", I obviously mean "totally broken".  So it's not like it works,
or really even can work in the current format that uses the unsafe
"panic" LED blinking routines..

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years ago[ARM] Fix non-page aligned boot time mappings
Russell King [Wed, 4 Jul 2007 20:16:33 +0000 (21:16 +0100)]
[ARM] Fix non-page aligned boot time mappings

AT91SAM9260 stopped booting with the recent changes to MM
initialisation - it was asking for a non-aligned virtual address
which caused loops to be non-terminal.  Fix this by rounding
virtual addresses down, but remember to include the offset in
the length, and round the length up to the following page.

This means that asking for a mapping of 4K starting at 2K into
a page maps two pages as one would expect.

Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
17 years agoMerge branch 'upstream' of git://ftp.linux-mips.org/pub/scm/upstream-linus
Linus Torvalds [Wed, 4 Jul 2007 16:26:44 +0000 (09:26 -0700)]
Merge branch 'upstream' of git://ftp.linux-mips.org/pub/scm/upstream-linus

* 'upstream' of git://ftp.linux-mips.org/pub/scm/upstream-linus:
  [MIPS] VSMP: Fix initialization ordering bug.
  [MIPS] Add whitelists for checksyscalls.sh
  [MIPS] die(): Properly declare as non-returning
  [MIPS] Fix include wrapper symbol definitions in IP32 code.

17 years ago[MIPS] VSMP: Fix initialization ordering bug.
Ralf Baechle [Tue, 3 Jul 2007 16:25:58 +0000 (18:25 +0200)]
[MIPS] VSMP: Fix initialization ordering bug.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
17 years ago[MIPS] Add whitelists for checksyscalls.sh
Atsushi Nemoto [Tue, 29 May 2007 14:30:04 +0000 (23:30 +0900)]
[MIPS] Add whitelists for checksyscalls.sh

Signed-off-by: Atsushi Nemoto <anemo@mba.ocn.ne.jp>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
17 years ago[MIPS] die(): Properly declare as non-returning
Maciej W. Rozycki [Tue, 29 May 2007 14:03:56 +0000 (15:03 +0100)]
[MIPS] die(): Properly declare as non-returning

 This marks the declaration of die() correctly, removing "control reaches
end of non-void function" warnings from non-void functions that die() at
the end.

Signed-off-by: Maciej W. Rozycki <macro@linux-mips.org>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
17 years ago[MIPS] Fix include wrapper symbol definitions in IP32 code.
Kumba [Fri, 25 May 2007 06:26:47 +0000 (02:26 -0400)]
[MIPS] Fix include wrapper symbol definitions in IP32 code.

Some IP35 defines snuck into some IP32-specific code during the DMA re-write.

Signed-off-by: Joshua Kinard <kumba@gentoo.org>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
17 years ago[JFFS2] Fix readinode failure when read_dnode() detects CRC failure.
David Woodhouse [Tue, 3 Jul 2007 20:51:19 +0000 (16:51 -0400)]
[JFFS2] Fix readinode failure when read_dnode() detects CRC failure.

We should have stopped returning 1 from read_dnode() to indicate
failure. We can just mark the damn thing obsolete immediately. But I
missed a case where we don't.

Signed-off-by: David Woodhouse <dwmw2@infradead.org>
17 years agoRemove some unused variables
Linus Torvalds [Wed, 4 Jul 2007 01:27:53 +0000 (18:27 -0700)]
Remove some unused variables

When Andi reverted the HPET resource reservation (in commit
0f8dc2f06560e2ca126d1670a24126ba08357d38), he didn't remove the now
unused variables, which just causes gcc to be noisy.

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agodio: remove bogus refcounting BUG_ON
Zach Brown [Tue, 3 Jul 2007 22:28:55 +0000 (15:28 -0700)]
dio: remove bogus refcounting BUG_ON

Badari Pulavarty reported a case of this BUG_ON is triggering during
testing.  It's completely bogus and should be removed.

It's trying to notice if we left references to the dio hanging around in
the sync case.  They should have been dropped as IO completed while this
path was in dio_await_completion().  This condition will also be
checked, via some twisty logic, by the BUG_ON(ret != -EIOCBQUEUED) a few
lines lower.  So to start this BUG_ON() is redundant.

More fatally, it's dereferencing dio-> after having dropped its
reference.  It's only safe to dereference the dio after releasing the
lock if the final reference was just dropped.  Another CPU might free
the dio in bio completion and reuse the memory after this path drops the
dio lock but before the BUG_ON() is evaluated.

This patch passed aio+dio regression unit tests and aio-stress on ext3.

Signed-off-by: Zach Brown <zach.brown@oracle.com>
Cc: Badari Pulavarty <pbadari@us.ibm.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoRevert perfctr reservation to 2.6.21 state
Andi Kleen [Tue, 3 Jul 2007 23:38:13 +0000 (01:38 +0200)]
Revert perfctr reservation to 2.6.21 state

With this change it works again when the nmi watchdog is disabled.

Signed-off-by: Andi Kleen <ak@suse.de>
Cc: Björn Steinbrink <B.Steinbrink@gmx.de>
Cc: Stephane Eranian <eranian@hpl.hp.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoRevert HPET resource reservation
Andi Kleen [Fri, 29 Jun 2007 14:16:36 +0000 (16:16 +0200)]
Revert HPET resource reservation

Matthias Lenk reports that the PCI subsystem would move the HPET on
SB400/SB600-based systems, where the HPET is in BAR1 of the SMbus
controller.

The reason? The ACPI layer registered the PCI MMIO range as being busy
too early, before PCI enumeration had happened, causing the PCI layer to
decide that it should relocate the resources somewhere else.

Firmware resources should be marked busy _after_ the PCI enumeration and
probing has happened, not before.

Remove the too-early reservation, we'll fix it up to do it properly
later.  In the meantime, this solves the regression.

Tested-by: Matthias Lenk <matthias.lenk@amd.com>
Cc: Aaron Durbin <adurbin@google.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoMerge master.kernel.org:/pub/scm/linux/kernel/git/bart/ide-2.6
Linus Torvalds [Tue, 3 Jul 2007 20:58:49 +0000 (13:58 -0700)]
Merge master.kernel.org:/pub/scm/linux/kernel/git/bart/ide-2.6

* master.kernel.org:/pub/scm/linux/kernel/git/bart/ide-2.6:
  ide: ide_scan_pcibus(): check __pci_register_driver return value
  ide: pdc202xx_new PLL input clock fix
  it821x: fix incorrect SWDMA mask
  amd74xx: resume fix
  hpt366: use correct enablebits for HPT36x
  hpt366: blacklist MAXTOR STM3320620A for UltraDMA/66
  ide: Fix a theoretical Ooops case
  ide: never called printk statement in ide-taskfile.c::wait_drive_not_busy

17 years agoMerge branch 'master' of ssh://master.kernel.org/pub/scm/linux/kernel/git/mchehab...
Linus Torvalds [Tue, 3 Jul 2007 20:58:03 +0000 (13:58 -0700)]
Merge branch 'master' of ssh://master.kernel.org/pub/scm/linux/kernel/git/mchehab/v4l-dvb

* 'master' of ssh://master.kernel.org/pub/scm/linux/kernel/git/mchehab/v4l-dvb:
  V4L/DVB (5822): Fix the return value in ttpci_budget_init()
  V4L/DVB (5818): CinergyT2: fix flush_workqueue() vs work->func() deadlock
  V4L/DVB (5816): Cx88-blackbird: fix vidioc_g_tuner never ending list of tuners
  V4L/DVB (5808): Bttv: fix v4l1 breaking the driver

17 years agoSLUB: Make lockdep happy by not calling add_partial with interrupts enabled during...
Christoph Lameter [Tue, 3 Jul 2007 16:31:04 +0000 (09:31 -0700)]
SLUB: Make lockdep happy by not calling add_partial with interrupts enabled during bootstrap

If we move the local_irq_enable() to the end of the function then
add_partial() in early_kmem_cache_node_alloc() will be called
with interrupts disabled like during regular operations.

This makes lockdep happy.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Tested-by: Andre Noll <maan@systemlinux.org>
Acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoNTP: remove clock_was_set() call to prevent deadlock
Thomas Gleixner [Tue, 3 Jul 2007 18:05:20 +0000 (20:05 +0200)]
NTP: remove clock_was_set() call to prevent deadlock

The clock_was_set() call in seconds_overflow() which happens only when
leap seconds are inserted / deleted is wrong in two aspects:

1. it results in a call to on_each_cpu() with interrupts disabled
2. it is potential deadlock source vs. call_lock in smp_call_function()

The only possible side effect of the removal might be, that an absolute
CLOCK_REALTIME timer fires 1 second too late, in the rare case of leap
second deletion and an absolute CLOCK_REALTIME timer which expires in
the affected time frame. It will never fire too early.

This was probably observed by the reporter of a June 30th -> July 1st
hang: http://lkml.org/lkml/2007/7/3/103

A similar problem was observed by Dave Jones, who provided a screen shot
with a lockdep back trace, which allowed to analyse the problem.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoide: ide_scan_pcibus(): check __pci_register_driver return value
Andrew Morton [Tue, 3 Jul 2007 20:28:36 +0000 (22:28 +0200)]
ide: ide_scan_pcibus(): check __pci_register_driver return value

drivers/ide/setup-pci.c: In function 'ide_scan_pcibus':
drivers/ide/setup-pci.c:879: warning: ignoring return value of '__pci_register_driver', declared with attribute warn_unused_result

Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>
17 years agoide: pdc202xx_new PLL input clock fix
Albert Lee [Tue, 3 Jul 2007 20:28:36 +0000 (22:28 +0200)]
ide: pdc202xx_new PLL input clock fix

Recently the PLL input clock of Promise 2027x is sometimes detected
higher than expected (e.g. 20.027 MHz compared to 16.714 MHz).
It seems sometimes the mdelay() function is not as precise as it
used to be. Per Alan's advice, HT or power management might affect
the precision of mdelay().

This patch calls gettimeofday() to measure the time elapsed and
calculate the PLL input clock accordingly.

Signed-off-by: Albert Lee <albertcc@tw.ibm.com>
Cc: Alan Cox <alan@lxorguk.ukuu.org.uk>
Cc: Bahadir Balban <bahadir.balban@gmail.com>
Acked-by: Sergei Shtylyov <sshtylyov@ru.mvista.com>
Signed-off-by: Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>
17 years agoit821x: fix incorrect SWDMA mask
Bartlomiej Zolnierkiewicz [Tue, 3 Jul 2007 20:28:35 +0000 (22:28 +0200)]
it821x: fix incorrect SWDMA mask

SWDMA modes are unsupported by it821x.  Attempts to tune SWDMA modes always
fail (due to sanity check in ->speedproc) and result in PIO being tuned.

* Fix incorrect SWDMA mask so core code won't try these modes and will just
  tune PIO if no other DMA modes are available.

* Bump driver version.

Signed-off-by: Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>
Acked-by: Sergei Shtylyov <sshtylyov@ru.mvista.com>
17 years agoamd74xx: resume fix
Bartlomiej Zolnierkiewicz [Tue, 3 Jul 2007 20:28:35 +0000 (22:28 +0200)]
amd74xx: resume fix

* Driver can't skip programming transfer mode on the device in amd_set_drive()
  (similar fix has been applied to via82cxxx driver ages ago).

* While at it remove redundant warning (ide_config_drive_speed() already
  produces more valuable one).

* Bump driver version.

Signed-off-by: Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>
17 years agohpt366: use correct enablebits for HPT36x
Sergei Shtylyov [Tue, 3 Jul 2007 20:28:35 +0000 (22:28 +0200)]
hpt366: use correct enablebits for HPT36x

The HPT36x chips finally turned out to have the channel enable bits -- however,
badly implemented.  Make use of them despite it's probably only going to burden
the driver's code -- assuming both channels are always enabled by the HighPoint
BIOS anyway...

Signed-off-by: Sergei Shtylyov <sshtylyov@ru.mvista.com>
Acked-by: Linas Vepstas <linas@austin.ibm.com>
Cc: michal.kepien@poczta.onet.pl
Signed-off-by: Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>
17 years agohpt366: blacklist MAXTOR STM3320620A for UltraDMA/66
Sergei Shtylyov [Tue, 3 Jul 2007 20:28:35 +0000 (22:28 +0200)]
hpt366: blacklist MAXTOR STM3320620A for UltraDMA/66

Add the MAXTOR STM3320620A drive into the UltraDMA/66 mode blacklist
for the HPT36x chips.

Signed-off-by: Sergei Shtylyov <sshtylyov@ru.mvista.com>
Acked-by: Linas Vepstas <linas@austin.ibm.com>
Signed-off-by: Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>
17 years agoide: Fix a theoretical Ooops case
Alan Cox [Tue, 3 Jul 2007 20:28:35 +0000 (22:28 +0200)]
ide: Fix a theoretical Ooops case

Found by a static analyser. It is in theory possible we dereference
dev->id when it has become invalid. Re-order to avoid this.

Not needed for new-ide as we no longer support the crazy exabyte nest stuff

Signed-off-by: Alan Cox <alan@redhat.com>
Cc: Andrew Morton <akpm@osdl.org>
Signed-off-by: Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>
17 years agoide: never called printk statement in ide-taskfile.c::wait_drive_not_busy
Masatake YAMATO [Tue, 3 Jul 2007 20:28:34 +0000 (22:28 +0200)]
ide: never called printk statement in ide-taskfile.c::wait_drive_not_busy

Look at wait_drive_not_busy in drivers/ide/ide-taskfile.c:

    static u8 wait_drive_not_busy(ide_drive_t *drive)
    {
            ide_hwif_t *hwif = HWIF(drive);
            int retries = 100;
            u8 stat;

            /*
             * Last sector was transfered, wait until drive is ready.
             * This can take up to 10 usec, but we will wait max 1 ms
             * (drive_cmd_intr() waits that long).
             */
            while (((stat = hwif->INB(IDE_STATUS_REG)) & BUSY_STAT) && retries--)
                    udelay(10);

            if (!retries)
                    printk(KERN_ERR "%s: drive still BUSY!\n", drive->name);

            return stat;
    }

`printk' is never called because `retries' never holds zero at the
outside of `while' loop: when `retries' holds zero at the while's loop
condition, `retries' will hold -1 at the if condition.

Signed-off-by: Masatake YAMATO <jet@gyve.org>
Cc: Chuck Ebbert <cebbert@redhat.com>
Cc: joe@perches.com
Signed-off-by: Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>
17 years agoV4L/DVB (5822): Fix the return value in ttpci_budget_init()
Hartmut Birr [Mon, 2 Jul 2007 16:39:41 +0000 (13:39 -0300)]
V4L/DVB (5822): Fix the return value in ttpci_budget_init()

if the call to budget_register() fails in ttpci_budget_int(),
ttpci_budget_init() returns success. The attached patch will
fix this problem.

Signed-off-by: Hartmut Birr <e9hack@googlemail.com>
Signed-off-by: Michael Krufky <mkrufky@linuxtv.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab@infradead.org>
17 years agoV4L/DVB (5818): CinergyT2: fix flush_workqueue() vs work->func() deadlock
Oleg Nesterov [Mon, 2 Jul 2007 15:26:20 +0000 (12:26 -0300)]
V4L/DVB (5818): CinergyT2: fix flush_workqueue() vs work->func() deadlock

Spotted and tested by Thomas Sattler <tsattler@gmx.de>.

cinergyT2.c does cancel_delayed_work() + flush_scheduled_work() while
holding cinergyt2->sem. This leads to deadlock because work->func()
needs the same mutex to complete. Another bug is that this code in fact
can't reliably stop the re-arming delayed_work.

Convert this code to use cancel_rearming_delayed_work() and move it
out of ->sem. Another mutex, ->wq_sem, was added to protect against the
concurrent open/resume.

This patch is a horrible hack to fix the lockup which happens in practice.
As Dmitry Torokhov pointed out this driver has other problems and needs
further changes.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Signed-off-by: Mauro Carvalho Chehab <mchehab@infradead.org>
17 years agoV4L/DVB (5816): Cx88-blackbird: fix vidioc_g_tuner never ending list of tuners
Jelle Foks [Mon, 21 May 2007 17:56:17 +0000 (14:56 -0300)]
V4L/DVB (5816): Cx88-blackbird: fix vidioc_g_tuner never ending list of tuners

v4l-info and other programs would loop indefinitely while querying the
tuners for cx88-blackbird cards.

The cause was that vidioc_g_tuner didn't return an error value for
qctrl->id != 0, making the application think there is a never ending
list of tuners...

This patch adds the same index check as done in vidioc_g_tuner() in
cx88-video.

Signed-off-by: Jelle Foks <jelle@foks.8m.com>
Signed-off-by: Michael Krufky <mkrufky@linuxtv.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab@infradead.org>
17 years agoV4L/DVB (5808): Bttv: fix v4l1 breaking the driver
Trent Piepho [Tue, 3 Jul 2007 18:08:10 +0000 (15:08 -0300)]
V4L/DVB (5808): Bttv: fix v4l1 breaking the driver

If one uses a V4L *one* application, such as vlc or mplayer's v4l driver, as
the first user after the driver is loaded, the driver wedges itself and will
never capture properly.  Even if one uses a V4L2 application later, it still
won't work.

If one uses a V4L *two* application first, such as tvtime or mplayer's v4l2
driver, then the driver will be ok.  One can then run a V4L1 application, and
it will work.

It turns out the problem is with norm changing and the crop support that was
added in 2.6.21.  The driver defaults to PAL, and keeps the last norm it was
set too across opens.  If one changes the norm via V4L1, the cropping
parameters are not reset like they should be, and they'll remain broken across
device opens.

This patch removes the direct setting of btv->tvnorm in the V4L1 ioctl
VIDIOCSCHAN handler.  The norm is set via the existing call to set_input(),
which calls set_tvnorm(), which will reset the cropping values now that it is
able to detect the norm change.

Signed-off-by: Trent Piepho <xyzzy@speakeasy.org>
Signed-off-by: Michael Krufky <mkrufky@linuxtv.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab@infradead.org>