bypassing on again, which is more beneficial, performance-wise.

	Reported by Richard Henderson.

	out by Jesse Barnes.  Also, drop the old lock code.

I can see a NaT consumption fault every time I cat /proc/net/dev
OK, the problem is in the simulated ethernet driver.

	tell the compiler that per-CPU variables can be addressed
	with "addl".

	On the simulator kernel, this saves about 2776 bytes in the
	kernel image.  On a zx1 kernel, it saves about 5593 bytes.
	As per-CPU variables are used more, these savings will
	increase, of course.

This will fix the generic compile by removing sn2 from the list of
targets, since it depends on the new discontig code, which hasn't gone
in yet.

	This ensures that per-CPU data is "addl"-addressable even
	on UP.

	The impact on native ia64 fork+execve performance as measured by
	LMbench is in the noise range (plus or minus 0.1%). The first two
	hunks in the patch are not together because we wanted to hide the
	memory latency.

	canonical address (l-value).  To get the virtually mapped
	alias (which is more efficient), use __ia64_per_cpu_var().  The
	latter is safe only if the address of the l-value is never passed
	to another CPU (i.e., not stored in any global place).
	For extremely efficient, portable per-CPU variables, there is
	now a new API local.h which was introduced by Rusty Russell.
	To use this, declare a variable of type local_t as a per-CPU
	variable and then use {__,}cpu_local_FOO() to manipulate such
	variables.  This patch also updated the atomic interface with
	a 64-bit counter.

acpi_madt is marked as __initdata, so an attempt to use
acpi_register_irq() after we discarded init sections would cause us to
dereference a random location.  Everywhere else uses has_8259 anyway.

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

These functions can all be devinit as they're all called from functions
which are themselves devinit.

This patch includes quite a few changes and updates for the shwdt driver (which
brings it in sync with LinuxSH CVS HEAD). This fixes up support for the SH-2,
and also fixes up some timer brain-damage.

This patch updates the sh-specific hd64465 pcmcia driver for the new API, as
well as fixing up some other issues (such as remap_page_range() abuse -- we
remap through P3 area instead) which caused it not to compile. Changes by
Dominik Brodowski and myself.

Here's an update for pvr2fb (with a patch this time), which makes it compile
again.

ppc64 machines can have PCI bus numbers larger than 8 bits, but it looks
like pci_device_to_OF_node already handles this case, in both 2.4 and
2.5.

We just need the space to store them.

This solves the unresolved symbol problem with modular intermezzo.  Also
update the MAINTAINERS entry.

send_IPI_self is needed to resend irqs with IRQ_PENDING status when
enabled.

Checked with Ingo, and it's in 2.4-ac for some time.

This should fix ide lost interrupts on UP with IO-APIC

Ide trigers it this way:
 - disable_irq
 - do stuff that triggers IRQ.
 - irq is IRQ_PENDING
 - enable_irq
 - IRQ is lost, needs to be resend.

I'll send the patch to fixup IDE disable_irq logic to Bart.

From Flavio B. Leitner. <fbl@netbank.com.br>

Fix a memory leak and an unchecked return code in the busmouse driver.

From Daniele Bellucci <bellucda@tiscali.it>.

Check the copy_to_user() return code in read_profile().

From Jorn Engel <joern@wohnheim.fh-wedel.de> and Randy Dunlap.

It reduces stack usage in the WAN router by about 0x500 bytes
(from 0x520 to 0x24 when I did it in April).

This fixes this warning:

  syncppp.c:165: warning: function declaration isn't a prototype

by adding "void" as the function parameter list.

This fixes one place where I missed the replacing dcache_lock with
vfsmount_lock in put_namespace().

Tested with CLONE_NEWNS flag also.

into kernel.bkbits.net:/home/davem/net-2.5

First "test" kernel. Same naming we used for 2.4.0 - there it took
from May to December to get to the real version. Let's see if we
can do it faster this time.

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

Switching the PCMCIA interrupts from serial PCI interrupts to parallel
PCI interrupts causes machines to hang solid.  If we leave such setups
well alone, and only change the interrupt mode of parallel PCI systems
between PCI+ISA and PCI only.

into nuts.ninka.net:/home/davem/src/BK/net-2.5

From Matthew Wilcox <willy@debian.org>.

into kanga.twiddle.net:/home/rth/work/linux/axp-2.5

The accounting works now, but I still think this last bit should be
added as well.  This makes absolutely sure we only account fs requests,
at both ends. Better safe than sorry.