This creates a sturcture to mirror the m68knommu/ColdFire CPU timer
hardware.  This is a much cleaner approach to accessing the timer
registers than the previous set of register address defines.

This modifies the m68knommu siginfo include file to use
asm-generic/siginfo.h instead of the asm-m68k/siginfo.h.

The asm-generic/siginfo.h is more appropriate here, and
the m68k is out of date.

This includes asm/unistd.h in the system call table setup code for
m68knommu.  It needs the local NR_syscalls define.  linux/unistd.h is
not assembler clean, so it must be asm/unistd.h.

    dont -> don't (135 occurrences)

    wont -> won't (21 occurrences)

    doesnt -> doesn't (35 occurrences)

    couldnt -> couldn't (4 occurrences)

    cant -> can't (28 occurrences)

The patch below does the following:
- remove #if'd kernel 2.2 code
- changes one MIN to min

Someone nuked the 2nd arg in 2.5.62, but didn't fix up
all the callers. Lazy.

(based on lm_sensors CVS)

(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- removed outdated comments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.