The existing allocator is first-fit with wraparound.  This allows
a large number of small holes to accumulate in the early part of the
region, leading to heavy fragmentation.  This adjusts the algorithm
to rescan the region when smaller sizes are requested, reducing
early fragmentation.

From Art Haas <ahaas@airmail.net>

Previously num_physpages only took into consideration low memory.
Since nr_free_pages returned something much larger, the caches
init would oops.

into foobazco.org:/sources/2.5-sparc-todave

From: Patrick Wildi <patrick@wildi.com>

On OSB4 the hwif->ultra_mask is set to not support UDMA.
Unfortunately in that case svwks_config_drive_xfer_rate()
falls through to the end of the function, instead of trying
other DMA modes.

 - Correctly handle wraparound on offset+len
 - fix FADV_WILLNEED handling of non-page-aligned (offset+len)

Let's hope we don't need to fix the fixed fix.

On sparc64 toolchain:

drivers/built-in.o(.init.text+0xaf8c): In function `usb_init':
: undefined reference to `usbfs_cleanup'

usb_init() is __init and usbfs_cleanup() is __exit.  No can do.

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

Patch from Nicolas Pitre

Patch from Nicolas Pitre

... including the new PXA270 aka Bulverde.

Patch from Nicolas Pitre

Before previous patch this driver compiled OK but was buggy.
Now it doesn't compile anymore as the bogus macro has been
deleted.  Fix that in any case.

The same fix has been committed to the MTD CVS already, but please forward 
this to Linus otherwise Lubbock won't compile from kernel.org tree anymore
(waiting for dwmw2 to update this might prove ... hrm ... long) 

Patch from Nicolas Pitre

Too many macro with too generic names.  Let's remove unneeded code and
redundant/unused macros.  This also prevent namespace clash with upcoming 
patches.

Patch from Nicolas Pitre

Minor cleanup of Lubbock specific code, like removal of
redundant mappings. 
Also a prerequisite for some upcoming patches.

Patch from Nicolas Pitre

virtual address mapping can change.

The recent slab alignment changes broke an unknown number of architectures
(parisc and x86_64 for sure) by causing task_structs to be insufficiently
aligned.

We need good alignemnt because architectures do things like dumping FP state
into the task_struct with instructions which require particular alignment (I
think).

So change the default alignment to L1_CACHE_BYTES, which is what we used to
have, via SLAB_HW_CACHE_ALIGN.

This patch updates the MTD concatenating driver from MTD CVS, which
fixes issues found with this driver which concatenates multiple MTD
devices into one MTD device.

From David Woodhouse, through CVS:

	revision 1.8
	date: 2003/06/30 11:01:26;  author: dwmw2;  state: Exp;  lines: +5 -5
	I will not commit stuff whilst pissed
	I will not commit stuff whilst pissed

	revision 1.7
	date: 2003/06/29 21:26:34;  author: dwmw2;  state: Exp;  lines: +9 -9
	Fix ecc/oob subdev comparisions

	revision 1.6
	date: 2003/06/25 12:37:50;  author: dwmw2;  state: Exp;  lines: +14 -6
	Don't pretend to have {read,write}_{oob,ecc} functions if subdevices don't

	revision 1.5
	date: 2003/06/25 12:21:16;  author: dwmw2;  state: Exp;  lines: +390 -397
	coding style cleanup

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

POSIX sez: "If len is zero, all data following offset is specified."

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

The 32->64 bit syscall translation layer on ppc64 incorrectly sign-
extends rather than zero-extending the second parameter to shmget(),
which should be a size_t.  This means that it is impossible to shmget()
more 2GB or more from a 32-bit process.

From: "Randy.Dunlap" <rddunlap@osdl.org>

From: Pavel Machek <pavel@ucw.cz>

Kill uninformative boot-time message.

From: Arnd Bergmann <arnd@arndb.de>

Enable basic profiling code on s390 depending on CONFIG_PROFILING, not
CONFIG_OPROFILE.

CONFIG_PROFILING should enable the generic profiling code here, even if
CONFIG_OPROFILE is not set.  Note that the identical code on i386 is always
compiled in, regardless of CONFIG_PROFILING and CONFIG_OPROFILE.

From: Michael Hunold <hunold@convergence.de>

unfortunately it's possible to remove a DVB adapter module even if a DVB
network device has been set up using this adapter.

The attached patch fixes this problem.

From: Matt Tolentino <metolent@snoqualmie.dp.intel.com>

This trailing space in sysfs contents is not needed.

From: Trond Myklebust <trond.myklebust@fys.uio.no>

fs/nfs/direct.c: In function `nfs_direct_IO':
fs/nfs/direct.c:458: warning: int format, different type arg (arg 2)

From: Adam Belay <ambx1@neo.rr.com>

Fix http://bugme.osdl.org/show_bug.cgi?id=2540

It prevents PnP detection if devices were already detected by SuperIO.

Ensure that when we send the SETATTR call after doing an O_EXCL create,
we always set the atime and ctime fields.

See RFC1813 for details on why the server is allowed to clobber these
two fields in order to cache a verifier that protects CREATE in case of
a timeout+resend from the client.

Patch is by Olaf Kirch.

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

Move HW interrupt register init. to the proper place
and don't force an interrupt if the IRQ is disabled
or in progress.

No pun intended, but X can't work if it can't turn on eXecute-permission
on the some data pages...

Just found I never changed type of move_page_tables when I changed it to
return offset: einormous mremap moves would fail on 64-bit.

into tiger.hpl.hp.com:/data1/bk/lia64/to-linus-2.5

To avoid various livelocks, the writeback code parks all the dirty inodes onto
sb->s_io and then works through that list until it is empty.  This assumes
that each inode will be moved to some other list as it is processed.

But there's a loophole: if the ->writepages() implementation does nothing at
all, the inode is not redirtied (which would move it to s_dirty).  This causes
s_io to not empty and pdflush goes nuts.

So when this happens, move the inode onto s_dirty within
__sync_single_inode().  Use list_move_tail() to attempt to preserve the
time-ordering of the s_dirty list.

From: Mikael Pettersson <mikpe@user.it.uu.se>

This patch fixes three warnings from gcc-3.4.0 in 2.6.6-rc3:

- arch/i386/pci/pcbios.c: use of "+m" constraint

- drivers/char/ftape/: use of cast-as-lvalue

- drivers/char/ftape/: __attribute__((packed)) on something containing only
  bytes

From: Tom Rini <trini@kernel.crashing.org>

Makes the Motorola PrPMC800 platform functional again.  This comes from Randy
Vinson <rvinson@mvista.com>.

From: Stephen Hemminger <shemminger@osdl.org>

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

Add oprofile support for s/390.

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

zfcp host adapter change:
 - Fix addressing exception due to uninitialized host_scribble pointer.

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

3270 device driver change:
 - Add missing irb error checking.

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

Network driver changes:
 - ctc: Add missing irb error checking.
 - iucv: Add name of net_device to iucvMagic to more than one
         connection between two guests.
 - qeth: Don't send IPA command if card is not in state SOFTSETUP or UP.
 - qeth: Fix number base in simple_strtoul call for buffer_count attribute.
 - qeth: Fix reallocating of buffers when buffer_count attribute is changed.
 - qeth: Correct handling of return codes in qeth_realloc_buffer_pool.
 - qeth: Don't call dev_close/dev_open on STOPLAN/STARTLAN commands.
         Use netif_carrier_off/netif_carrier_on instead.