(i) The prototypes for free_vfsmnt(), alloc_vfsmnt(), do_kern_mount()
    so far occurred in several individual c files.  Now they are in
    <linux/mount.h>.

(ii) do_kern_mount() has a third argument name that is typically a
     constant.  It is called with "rootfs", "nfsd", type->name,
     "capifs", "usbdevfs", "binfmt_misc" etc.  So, it should have a
     prototype that expresses this:

	do_kern_mount(const char *fstype, int flags, const char *name, void *data);

This makes the ugly cast

-       return do_kern_mount(type->name, 0, (char *)type->name, NULL);
+       return do_kern_mount(type->name, 0, type->name, NULL);

go away. Now do_kern_mount() calls type->get_sb(), so also get_sb()
must have a const third argument. That is what the patch below does.

If I am not mistaken, precisely two filesystems do not treat this
argument as a constant, namely afs and cifs. A separate patch
gives some cleanup there.

This fixes some ColdFire 5407 cache bogosity.  Previous code was pushing
all cache lines and the invalidating all of the cache.  The push should
be enough, and now with underlying fixes the the cache setup registers
it is.  Removed the whole invalidate cycle.

Original patch cam from Allessandro Rubini.

Correct the wrong cache setup used on the MOTOROLA 5407 based board.
Specifically it enables precise exception mode and write buffering.

Original patch from Alessendra Rubini.

Correct the wrong cache setup used on the CLEOPATRA 5407 based board.
Specifically it enables precise exception mode and write buffering.

Original patch from Alessendra Rubini.

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

This patch takes out the MOD_INC/DEC_USE_COUNT in the airport (Apple
wireless ethernet) driver.  The driver already does SET_MODULE_OWNER
on the netdevice, so the MOD_INC/DEC_USE_COUNT are unnecessary and
just cause warnings.

This patch fixes ppp_synctty.c (used for doing PPP over some synchronous
serial HDLC links) so that it sets the owner field of the tty line
discipline it exports, rather than using MOD_INC/DEC_USE_COUNT.  This
is more or less from Stephen Hemminger.

Please apply.

Thanks,
Paul.

* Cleanup ethtool/mii_ioctl sets of speed/duplex/autoneg by
  stop/set/start driver to ensure sets stick.  Must hold
  xmit_lock around stop/start.

into redhat.com:/garz/repo/net-drivers-2.5

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

into redhat.com:/garz/repo/irda-2.5

This patch fixes the build of drivers/ide/ppc/pmac.c.

While going through sound/ for strncpy replacing, I came across this
routine:

/* return the (first) queue matching with the specified name */
queue_t *snd_seq_queue_find_name(char *name)
{
        int i;
        queue_t *q;

        for (i = 0; i < SNDRV_SEQ_MAX_QUEUES; i++) {
                if ((q = queueptr(i)) != NULL) {
                        if (strncpy(q->name, name, sizeof(q->name)) == 0)
                                return q;
                        queuefree(q);
                }
        }
        return NULL;
}


I'm _really_ sure that they meant to use strncmp() here instead.

Like this.

A couple of updates for the non-MMU memory managment code:

 - declare the overcommit variables
 - return a real vma pointer from get_user_pages (it is used by some code)
 - correct args to vmap()

Use strlcpy in root_dev_setup()

I only converted the cases where it was obvious that the intent was to
truncate on overflow. Lots of places for maxpath/readlink type stuff I
left alone.

This does a lot of cleanup for strncpy->strlcpy, replaces some
sprintf/snprintf's aswell.  There were only two places where things
weren't straight forward.  All-in-all a good cleanup though.

Make the default kernel build less verbose, to make warnings show up
more clearly.

Yet another one in the namespace.c series.

The code in graft_tree() used to be correct, but the code

        err = -ENOENT;
        down(&nd->dentry->d_inode->i_sem);
        if (IS_DEADDIR(nd->dentry->d_inode))
                goto out_unlock;

        spin_lock(&dcache_lock);
        if (IS_ROOT(nd->dentry) || !d_unhashed(nd->dentry)) {
		...
	}
	spin_unlock(&dcache_lock);
 out_unlock:

was made incorrect in 2.5.29 when

        err = security_sb_check_sb(mnt, nd);
        if (err)
                goto out_unlock;

was inserted.  It has happened more often that people overlooked
a preexisting setting of err.

Tell the arch that install_page() has just added a page into the pagetables.

From: William Lee Irwin III <wli@holomorphy.com>

Silence a printk warning.

From: William Lee Irwin III <wli@holomorphy.com>

64-bit ACPI fixes

From: William Lee Irwin III <wli@holomorphy.com>

Fix "cast to pointer from integer of a different size".

From: Martin Hicks <mort@wildopensource.com>

Here is a patch that changes mmu_gathers into a per-cpu resource.  It
includes the changes for all arches except ia64.  I've sent a separate patch
to David Mosberger for ia64.

From: Hollis Blanchard <hollis@austin.ibm.com>

Fix two pnp error-path memory leaks, caught by Stanford memory leak checker
circa 2.5.48.

From: Hollis Blanchard <hollis@austin.ibm.com>

Another potential memory leak the Stanford checker caught at 2.5.48: while
closing and opening floppy disks, buf could be allocated and never freed.

From: Hollis Blanchard <hollis@austin.ibm.com>

Hi, this was caught by the Stanford memory leak checker a while ago
(2.5.48). If the tmp_stats allocation fails, tmp is not being freed.

From: john stultz <johnstul@us.ibm.com>

This is a rework of John's recent change which resolved a circular include
dependency: a function in mach_apic.h requires hard_smp_processor_id() and
hard_smp_processor_id() requires macros from mach_apic.h

So this patch (against bk-current) reverts the previous, and fixes the same
circular dependency in a much cleaner way, by moving a piece of the circular
chain into its own .h file, rather then removing hard_smp_processor_id() and
accessing the apic by hand.

From: Maneesh Soni <maneesh@in.ibm.com>

Provides a separate request queue for each ramdisk instance.  Without this,
the kernel oopses when the block later tries to unregister the same set of
kobject things multiple times.  This makes rd.c consistent with all other
"disk" devices.

From: Duncan Haldane <f.duncan.m.haldane@worldnet.att.net>

Gerd Knorr noticed a small use of floating point math in the cpia driver
updates for 2.5.x I sent you a while ago, and this is not allowed in the
kernel.

This was in some code taken essentially verbatim from the Windows cpia
driver released under the GPL by STM inc.  that had been incorporated in
the later versions of the cpia drivera sourceforge.

It turns out that the use of floating point was quite inessential, and I've
reimplemented the couple of lines of code involved in integer arithmetic.

From: Rick Lindsley <ricklind@us.ibm.com>

Here is a file to add to the Documentation/ directory which describes the
disk statistics fields.

- Add an explanation for clearing the focus bit on P4 (zwane)

- __d_path kerneldoc fix (John Levon)

- generic-hdlc documentation fix (Krzysztof Halasa <khc@pm.waw.pl>)

- cmdline_read_proc cleanup (Oleg Drokin)

- remove a couple of unused vars from drivers/ide/pci/hpt366.c

- sound/core/sgbuf.c needs mm.h at least on alpha, for mem_map and other
  page stuff.  (Ivan Kokshaysky <ink@jurassic.park.msu.ru>)

- Don't use "u32 long" in cs46xx.c (Kevin Puetz <puetzk@puetzk.org>)

- fs/nfs/nfs4xdr.c warning fix: all the `goto out;' statements are
  commented away, so comment away the label too.

- net/ipv6/af_inet6.c: remove unused var

- drivers/media/video/bttv-cards.c: jiffies are unsigned long

- drivers/media/video/saa7134/saa7134-cards.c: unused var

- Fix Documentation/Changes comment wrt sparc compiler version

- drivers/pnp/quirks.c needs slab.h for kfree().  (Daniele Bellucci
  <bellucda@tiscali.it>)

From: David Gibson <david@gibson.dropbear.id.au>


Renames check_valid_hugepage_range() to is_hugepage_only_range(), which makes
more sense.

From: Manfred Spraul <manfred@colorfullife.com>

de_thread is called by exec to kill all threads in the thread group except
the threads required for exec.

The waiting is implemented by waiting for a wakeup from __exit_signal: If
the reference count is less or equal to 2, then the waiter is woken up.  If
exec is called by a non-leader thread, then two threads are required for
exec.

But if a thread group leader calls exec, then only one thread is required
for exec.  Thus the hardcoded "2" leads to a superfluous wakeup.  The patch
fixes that by adding a "notify_count" field to the signal structure.