(from lm_sensors CVS)

(based on lm_sensors CVS)

use bdevname() for block devices.

(from lm_sensors CVS)

(from lm_sensors CVS)

dito

into home.transmeta.com:/home/torvalds/v2.5/linux

SGI Modid: 2.5.x-xfs:slinx:140255a

namespace which other filesystems are also supporting.

SGI Modid: 2.5.x-xfs:slinx:140237a

SGI Modid: 2.5.x-xfs:slinx:140187a

SGI Modid: 2.5.x-xfs:slinx:140173a

SGI Modid: 2.5.x-xfs:slinx:140135a

certain benchmarks.  This is reverted to how it was, except the number
of hash buckets is larger than previously to attempt to account for the
workload Steve was originally targetting with that change.

SGI Modid: 2.5.x-xfs:slinx:140053a

clear of pb_locking field if IO completion handled in pagebuf_iorequest.

SGI Modid: 2.5.x-xfs:slinx:140047a

after some careful analysis.

SGI Modid: 2.5.x-xfs:slinx:140039a

SGI Modid: 2.5.x-xfs:slinx:140015a

It it is using a pointer to a ctl_table for its proc_handler function
pointer.  Goes oops.

Patch from Andries.Brouwer@cwi.nl

A patch for genhd.c:

- removed outdated comments
- removed MAX_BLKDEV

In genhd.c the variable MAX_BLKDEV was only the size of a hash
table, so I made it MAX_PROBE_HASH. It can be 1, or 23, or 256, or
whatever one wants.

Note that the current setup requires that every device number
in a given range is mapped by dev_to_index() to the same index
in the hash table, so this routine will have to be adapted in
case one wants to register multimajor ranges.

Discussion is possible about whether struct blk_probe needs
a dev_t or a kdev_t, but I left things this time.

If a range can end at precisely the end of [k]dev_t space,
the old code was wrong since (p->dev + p->range) would be 0.
That is why "p->dev + p->range <= dev" was replaced by
"p->dev + p->range - 1 < dev".

This is a forward-port of a 2.4 change from Stephen.  The (old) 2.5 code is
forcing a commit on every write by artificially dirtying the inode.

But generic_file_aio_write() has called generic_osync_inode() for us, which
has synced the file data.  There is no need to force the extra commit.

Patch from Thomas Schlichter <schlicht@uni-mannheim.de>

Adds some missing global_flush_tlb() calls, which are requried after a call
to change_page_attr().

Patch from Rik van Riel <riel@imladris.surriel.com>

If a process cannot exit because it's stuck in eg.  a driver, it doesn't make
sense to have the OOM killer kill it repeatedly; that could lead to a hung
system.

Instead, kill another process if the first process we tried to kill hasn't
made any move to exit within 5 seconds.  This way we have a much better
chance of recovering the system.

Patch from "Martin J. Bligh" <mbligh@aracnet.com>

Fixes up the logic and code which is used to suppress the IRQ balancing code.

We now just have a single boolean, "irqbalance_disabled".  The initial value
comes from the per-platform "NO_BALANCE_IRQ" constant.

If the platform defaults to "on", users can override this with the
"noirqbalance" kernel boot option.

Patch from Alex Tomas <bzzz@tmi.comex.ru>

Fixes a couple of memory leaks in the htree code.

Patch from John Levon <levon@movementarian.org>

Looked like this condition was previously always false ...

Patch from Hugh Dickins <hugh@veritas.com>

When prune_icache coincides with unmounting, invalidate_inodes notices
the inode it's working on as busy but doesn't wait: Self-destruct in 5
seconds message, and later iput oopses on freed super_block.

Neither end is a fast path, so the patch just adds iprune_sem for exclusion.

The semaphore is held across dispose_list so that
dispose_list->clear_inode->destroy_inode cannot reference a destroyed
superblock.

Patch from Rick Lindsley <ricklind@us.ibm.com>

Fixes a couple of odd cases in which disk requests were not being accounted
for.

This check is not needed - the PMD is known to be present.

Patch from Maneesh Soni <maneesh@in.ibm.com>

Turns out that sysfs is doing dget() on a zero-ref dentry.  That's a bug, but
dcache is no longer detecting it.

The check was removed because with lockless d_lookup, there can be cases when
d_lookup and dput are going on concurrently, If d_lookup happens earlier then
it may do dget() on a dentry for which dput() has decremented the ref count
to zero.  This race is handled by taking the per dentry lock and checking the
DCACHE_UNHASHED flag.

The patch open-codes that part of d_lookup(), and restores the BUG check in
dget().

Use interruptible sleep rather than uninterruptible to avoid perturbing load
average.

do_generic_mapping_read() has an open-coded version of find_get_page() hidden
inside it.

Use the early ioremap code to parse the Static Resource Affinity Table on
x440 machines.

Patch from Patricia Gaughen <gone@us.ibm.com>,
		Dave Hansen <haveblue@us.ibm.com>

It provides a very early sort of kmap-by-hand.

The patch is used by the x440 discontigmem to map the srat tables into low
memory so that the memory can be setup.  This remap function is used very
early in the boot process...  at the start of setup_arch().

This functionality is only available to Summit and NUMAQ.  It will work on
other platforms, but they do not need it.

Patch from Thomas Schlichter <schlicht@uni-mannheim.de>

Make flush_tlb_all() preempt-safe.  Same as the ia32 fix.

Patch from Joakim Tjernlund <joakim.tjernlund@lumentis.se>

Here is another update(against BK curr) for crc32(). A kind soul pointed out
the optimizations below.

lib/crc32defs.h:
 - Make it possible to define new values for CRC_LE_BITS/CRC_BE_BITS without
   modifying the source.

lib/crc32.c:
 - Eliminate the need for ENDIAN_SHIFT. Saves a 24 bit shift in the byte
   loops.

 - Swap the XOR expression in DO_CRC. gcc for x86 can not do that simple
   optimization itself(gcc 3.2.2 and RH gcc 2.96 tested). Will improve
   performance with 20-25% on x86.

Spotted by Anton Blanchard: Our attempt to limit the interrupts-off time in
slab wasn't very effective.  #ifndef DEBUG is never true.

Rework this function so that we only make the indirect call to the
page-freeing function on the final put_page(), rather than on every
invokation.

Patch from Hugh Dickins <hugh@veritas.com>

Userspace shows huge elapsed time across jiffies wrap: with USER_HZ less then
HZ, sys_times needs jiffies_64 to calculate its retval.

From Tim Schmielau <tim@physik3.uni-rostock.de>

Force jiffies to start out at five-minutes-before-wrap.  To find
jiffy-wrapping bugs.

This patch provides the following spelling fixes:

paticular   -> particular
usefull     -> useful
occurance   -> occurrence
occurances  -> occurrences
successfull -> successful

This patch provides spelling fixes for the following:

shold      -> should
Docement   -> Document
docomented -> documented
whic       -> which
thresold   -> threshold
asociation -> association