no longer calls the generic VFS read and write routines.
    This allows all application read requests to pass through
    to the server, instead of just the ones that appear to be
    inside the file. this eliminates the requirement to use a
    GETATTR operation before each read or write to determine
    where the EOF is. This is a significant performance and
    scalability win.

    It also removes all requirements for holding the inode
    semaphore during NFS direct reads and writes, as the read
    and write logic no longer needs atomic access to the size
    of the file. this also helps client CPU scalability by
    reducing the serialization of writes against a single file.
                                                                                
    Patch by Chuck Lever

    where NFS_I(inode)->data_updates can get out of sync
    with reality, which will lead to a BUG() in nfs_clear_inode
    later on.

    Patch by Olaf Kirch.

     seconds in order to avoid flooding of servers.

The NFS readlink() methods all take a buffer length argument.  Use that
instead of assuming PAGE_SIZE... 

We need to return ENAMETOOLONG rather than EIO.

Make sure NFS client doesn't see errors from mode setting on new symlinks.

When nfsd creates a symlink, it tries to set the mode as the mode is 
carried in the NFS request and some filesystems store a mode.

If the filesystem refuses to set the mode (e.g. -EOPNOTSUPP), this
error should not be returned to the client.

into ppc970.osdl.org:/home/torvalds/v2.6/linux

into nuts.davemloft.net:/disk1/BK/sparc-2.6

This adds support for the CP-104 Moxa Smartio serial cards.  Just add
the PCI ID information.

When the kmem_bufctl_t typedef got added to include/asm-ppc64/types.h,
it got added outside the #ifndef __ASSEMBLY__ section, causing
assembler errors.  This patch, from David Gibson, moves it inside the
#ifndef __ASSEMBLY__ region.

Add newlines to some printk debug trace of comempci.c driver.

Remove a lot of un-used and un-needed libgcc funstions from export list
for m68knommu syms.

Correct build lines for Motorola Dragonball 68x328 frame buffer driver. 

Patch from Georges Menie <georges@menie.org>

Remove ColdFire specific code sections for atomic_add and atomc_sub.

These are not needed, the m68k asm code for these functions is ColdFire
clean.

We should un-define all the x86 style IO routines when redefining local
versions.

Create init points for the Motorola Dragonball 68x328 frame
buffer driver.

Patch from Georges Menie <georges@menie.org>

A couple of fixups for asm-m68knommu/bitops.h:

. re-order definition of fls(), to be outside __KERNEL__
. add code for find_next_bit()

Big cleanup of the Motorola DragonBall 68x328 frame buffer.  It was
quite broken before. 

Patch from Georges Menie <georges@menie.org>.

Fix the cache flushing code for the ColdFire 5407 CPU.

The cpushl instruction arguments are wrong, causing it to miss some
cache lines.

Added a nommu version of sysctl_max_map_count.

Fix return value from do_munmap(), it should return 0 on success not
EINVAL.

From: Neil Brown <neilb@cse.unsw.edu.au>

Fix bug #2661

Raid currently calls ->unplug_fn under spin_lock_irqsave(), but unplug_fns
can sleep.


After a morning of scratching my head and trying to come up with some that
does less locking, the following is the best I can come up with.  I'm not
proud of it but it should work.

If I move "nr_pending" out or rdev into the per-personality structures
(e.g.  mirror_info), and if I had "atomic_inc_if_nonzero" I could do with
without locking so much, but random atomic* functions don't seem trivial

into ppc970.osdl.org:/home/torvalds/v2.6/linux

        spinlock_t *lp = &((struct netdev_private *)dev->priv)->lock;
doesn't mix with
	spin_unlock_irqrestore(&lp, flags);

From: David Mosberger <davidm@napali.hpl.hp.com>

From: Martin Diehl <lists@mdiehl.de>

There was a spin_unlock missing in the raw mode tx-completion path.  Probably
it slipped through because the raw mode stuff is never reached with my Actisys
hardware.

From: Martin Schwidefsky <schwidefsky@de.ibm.com>

Network driver changes:
 - iucv: Make grab_param function SMP safe.
 - lcs: Fix null-pointer dereference after unsuccessful set_online.
 - qeth: Fix kmalloc flags in qeth_alloc_reply.
 - qeth: Show broadcase capability also in route4/6 sysfs attributes.
 - qeth: Remove debug code.
 - qeth: Add option to qetharp user space interface to strip unused
         fields from query arp records.
 - qeth: Add shortcut in outbound path for HiperSockets.
 - qeth: Add more info to qeth_perf_stats.
 - qeth: Add support for direct SNMP interface to OSA express cards.

From: Martin Schwidefsky <schwidefsky@de.ibm.com>

zfcp host adapter change:
 - Remove misplaced dot in error message.
 - Remove unused performance statistics code.

From: Martin Schwidefsky <schwidefsky@de.ibm.com>

dasd device driver changes:
 - Reset pointer from ccw device to dasd_devmap on device removal.

From: Martin Schwidefsky <schwidefsky@de.ibm.com>

s390 core changes:
 - Fix system call trace / audit interface.
 - Fix find_first_bit / find_next_bit inlines assembly constraints.

From: Mingming Cao <cmm@us.ibm.com>

There is a uninitialized goal value being referenced in both ext3 and ext2
find goal block functions (ext3_find_goal() and ext2_find_goal()).

In the non-sequential write case, these functions check the goal value(non
zero) before calling ext3(2)_find_near() to find the goal block to
allocate.

Since the goal value is uninitialized(non zero), the ext3(2)_find_near() is
never being called in the non-sequential write, thus ext3(2)_find_goal()
failed to guide a goal block in the random write case.

ext3(2)_new_block() takes the junk goal value and will turn it to goal 0
since it's normally beyond the filesystem block number limit.  The fix is
trivial.

From: Rusty Russell <rusty@rustcorp.com.au>

The IA64 hotplug CPU merge seems to have included some core changes: in
particular the recalc_bh_state() needs to sum for all (including offline)
cpus, since we don't empty the counters on CPU down.  The totals printed by
/proc/stat (the first loop) should include offline cpus, too (apparently
printing out the per-cpu lines for offline cpus confuses top).

The recent ARM-specific fix broke ia32.  Hopefully the ARM team can find an
arch-specific implementation of VGA_MAP_MEM() which makes it work.

From: Matt Mackall <mpm@selenic.com>

Doing the algebra:

c = (a - b) * 3/2
a' = a - c = a - 3/2(a - b) = (2a - 3a + 3b)/2 = (3b - a)/2
a' >= 0
3b - a >= 0
3b >= a
b >= a/3
nr_free_pages() >= mempages/3

We can indeed get into trouble if we try to load a large kernel on a very
small box (ie kernel reserves more than 2/3 of usable memory).  Surprisingly I
haven't hit this, but here's a fix.

From: "J. Bruce Fields" <bfields@fieldses.org>

svc_recv may call svc_sock_release before rqstp->rq_res is initialized.