padding to ic_roundoff to cater for pad in reservation cursor updates.

SGI Modid: xfs-linux:xfs-kern:167580a

SGI Modid: xfs-linux:xfs-kern:167547a

SGI Modid: xfs-linux:xfs-kern:167442a

SGI Modid: xfs-linux:xfs-kern:166619a

SGI Modid: xfs-linux:xfs-kern:167279a

SGI Modid: xfs-linux:xfs-kern:167273a

SGI Modid: xfs-linux:xfs-kern:167222a

Log Record wrapping around the physical log end. Need to reset
the pb size back to what we were using and NOT just 32K.

SGI Modid: xfs-linux:xfs-kern:167196a

SGI Modid: xfs-linux:xfs-kern:167192a

SGI Modid: xfs-linux:xfs-kern:167181a

SGI Modid: xfs-linux:xfs-kern:167054a

SGI Modid: xfs-linux:xfs-kern:167038a

SGI Modid: xfs-linux:xfs-kern:166614a

SGI Modid: xfs-linux:xfs-kern:166828a

SGI Modid: xfs-linux:xfs-kern:166665a

SGI Modid: xfs-linux:xfs-kern:166629a

SGI Modid: xfs-linux:xfs-kern:166562a

SGI Modid: xfs-linux:xfs-kern:166505a

SGI Modid: xfs-linux:xfs-kern:166497a

SGI Modid: xfs-linux:xfs-kern:166489a

SGI Modid: xfs-linux:xfs-kern:166409a

Since we do push/pops, a stack-based memory argument would
do the wrong thing by using the wrong offsets.

The x86-64 implementation of csum-partial.c causes it to compute the
checksum incorrectly.

As a result, multicast doesn't work.  It looks as though iptables is
also affected.

Here's a simple patch.

This patch is the first of several planned fixes for the cyclades
multiserial cards driver.

Its mostly a sync with in-house driver:

- Prevent users from opening non-existing Z ports
- Implement special XON/XOFF character handling in Z cards
- Prevent data-loss on Z cards
- Throttling fix for Z card
- Only throttle if CTS/RTS are set
- Fix accounting of received data

Kudos to Cyclades R&D

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

PRD_ENTRIES is specified to be precisely 256; on platforms where
PAGE_SIZE varies from 4KB the calculation in the current expression
defining it is inaccurate, which may cause crashes. This patch changes
it to the constant literal 256.

From: Andrew Morton <akpm@osdl.org>

Use sector_t for the block number.

Noticed by Marcel Cotta <mc123@mail.ru>.

ide_init() always initializes system_bus_speed variable
so system_bus_clock() should never call ide_system_bus_speed()
and no driver is calling ide_system_bus_speed() directly.

Bug was that if no IDE kernel parameter was given during boot
system_bus_speed will be zeroed in init_ide_data().

This patch should fix the problem
(as a bonus -> no need to zero these variables they are static).

From: Vojtech Pavlik <vojtech@suse.cz>

Add UDMA6 support to ALi PCI IDE driver.
Forward port from 2.4 and a patch from ALi.

Was in -mm for a long time...

Fix from jgarzik's sata_sil.c libata driver.

This was in -mm for a long time...

This switches alpha to use drivers/Kconfig instead of including most of
files included by drivers/Kconfig, thus enabling previously omitted i2c
(already used in some drivers available on alpha) and telephony drivers.

Because drivers/message/fusion/Kconfig was included only conditionally
("if PCI"), this patch also changes CONFIG_FUSION to depend on PCI.

(these changes were consulted with Christoph Hellwig in "i2c on alpha
- used but not available in 2.6.3" thread on LKML)

into ppc970.osdl.org:/home/torvalds/v2.5/linux

into ppc970.osdl.org:/home/torvalds/v2.5/linux

arm-26, ppc, sparc, sparc64 and sh have per-arch definitions of
__cacheline_aligned that are identical to default.  And yes, removal is
safe - all users of __cacheline_aligned actually pull linux/cache.h in.

We put ____cacheline_aligned things in their own section, simply
because we waste less space that way.  Otherwise we end up padding
innocent variables to the next cacheline to get the required
alignment.

There's no reason not to do this in modules, too.

	Removes gratitious ifdefs - get_nodes() is only called from __init
and by itself, check_space() is only called by get_nodes(), so they can
be made unconditionally __init.

into ppc970.osdl.org:/home/torvalds/v2.5/linux