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

  condition, we test the preemption count in preempt_schedule(). This reduces the icache
  footprint and the overhead of preemption.

- plus optimize the irq-path preemption check a bit.

this is a fragile piece of the ptrace code, the code relies on a single wakeup coming from the parent.
This fix is necessery after the preempt_schedule() cleanups, it unbreaks 'strace strace ...'.

fixes a preemption race that was introduced by my removal of PREEMPT_ACTIVE.

- the PREEMPT_ACTIVE trick is not needed

- schedule() should check for need_resched, we might miss a
  reschedule otherwise.

the cleanup also fixes the bug. The only reason why i kept
preempt_schedule() was to fix up p->state to TASK_RUNNING,
to make it possible to preempt from places that mark the
task TASK_UNINTERRUPTIBLE before adding the task to a waitqueue,
and thus a preemption in that small window could cause the
task to be removed from the runqueue erroneously.

	BKL shifted into all instances of ->truncate().  Callers updated.

	OK, here comes: ->lookup() had lost BKL, all in-tree instances of
->lookup() converted.

	I'm adding Documentation/filesystems/porting - with the list of
API changes since 2.4.  Are you OK with that format?

(and yes, this sucker is *post*-compile ;-)

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

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

net driver.  Also included are a couple of minor error reporting
updates and the proper detection for cardbus removal.

Contributor:
Mike Phillips
Linux Token Ring Project

the ibmtr token ring net driver. 

Contributor:
Mike Phillips
Linux Token Ring Project

* Only print warning on truly "oversized" packets
* Fix theoretical bug on gigabit cards - return to 1.1.3 behavior

Contributor: Val Henson

from 2.4.x.
Also, remove deprecated SIOCDEVPRIVATE ioctl calls.

Via Dave Jones.

0.13a - optical transceiver support added
	by Michael Clark <michael@metaparadigm.com>
0.13b - call register_netdev earlier in initialization
	suppress duplicate link status messages
0.15	get ppc (big endian) working

Via Dave Jones.

Fix originally by Andreas Dilger IIRC, merged by Dave Jones.

Via Dave Jones.

for dmfe net driver from 2.4.x.

Via Dave Jones.

which prevented the driver from building.

Via Dave Jones.

net drivers.  Merge two sis900 bug fixes from 2.4.x.

Via Dave Jones.

3c59x, eepro100, sis900, and tulip.

Also, update eepro100 Becker URL.

Contributor: Dave Jones

Via Dave Jones.

for net drivers 3c509, acenic, ni52, and skfp.

Via Dave Jones.

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

Now that the input core changes have made it into 2.5 I can finally
update the USB input drivers to their latest versions.

Here is a patch that does that.

In detail:

	HID driver:
		Fix a bug in descriptor parsing (array/variable),
			namely visible with Logitech new joysticks and mice
		Fix bugs in logical/physical min/max parsing
		Fix bugs in exponent parsing
		Remove workaround for low-speed devices with >8 byte
			reports, fix this in a correct way (bigger irq
			request)
		Untangle some code (fetc_item())
		Implement asynchronous input/output/feature report
			reading and writing
		Implement (hopefully) proper locking in the above
		Implement support for devices with an output endpoint
		Add some support functions for force feedback support
			currently in development
		Add entries to the debug dump code, including FF and
			exponents
		Add more mappings into the hid-input interface
		Cleanups here and there

	usbkbd driver:

		Make LED URBS use GFP_ATOMIC, they'll be called from a
			completion handler
		Remove dependency on hid.h

	usbmouse driver:

		Just conversion to the new input core, minor cleanups

	wacom driver:

		Just conversion to the new input core.

this patch somehow didn't get applied to 2.5.4
so i resend it.  It is pretty harmless - only
adds 3 more devices and 2 vendor ids into pegasus.h :-)

mmap currently sets errno to EINVAL when it should be ENOMEM.
SUS/POSIX states that ENOMEM should be returned when:

"MAP_FIXED was specified, and the range [addr, addr + len) exceeds
that allowed for the address space of a process; or if MAP_FIXED was
not specified and there is insufficient room in the address space to
effect the mapping."

The following patch (against 2.4.17) fixes this behaviour:

* fs/ncpfs/inode.c: Return reasonable error codes instead of universal
     -EINVAL. Remove printk() as reasonable code is returned.
     Set maximum file size limit on ncpfs to 4GB-1.

* fs/ncpfs/sock.c: Return correct error code when send() fails.

	Petr Vandrovec

* zisofs_cleanup cannot be __exit, as it is invoked from __init
  section when register_filesystem() fails.

		Petr Vandrovec

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

having the thread_info struct at the base of the stack and
the task_struct elsewhere.

into quango.(none):/home/paulus/kernel/linuxppc-2.5

	Linus, I've got the first of BKL-removal ext2 patches ready to
go.  It removes BKL from ext2_get_block() and guts of ext2_truncate().
The only place where we hold BKL on these paths is in dquot.c - probably
can be easily dealt with, but threading quota is a separate story.

	Inode metadata (pointers to blocks, both in inode itself and in
indirect blocks, preallocation data and allocation goal) are protected
by rwlock - EXT2_I(inode)->i_meta_lock.

	Next steps will involve threading the group descriptors and bitmaps
handling - lock_super() uses in ext2 are going to die.  However, that's
a separate story - let's do that step-by-step.

	I suspect that patch below will take care of almost all BKL contention
from ext2 - we still have BKL held over directory operations, but for regular
files that's it.

   truncate now correctly sets mtime always. Before this fix, mtime was not
   updated if truncated file was of zero length or if new filesize was bigger
   then old.
   Problem was noticed by Matthias Andree <ma@dt.e-technik.uni-dortmund.de>