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

We recently changed the VSID allocation on PPC64 to use a new scheme
based on a multiplicative hash.  It turns out our choice of multiplier
(the largest 28-bit prime) wasn't so great: with large contiguous
mappings, we can get very poor hash scattering.  In particular earlier
machines (without 16M pages) which had a reasonable about of RAM (>2G
or so) wouldn't boot, because the linear mapping overflowed some hash
buckets.

This patch changes the multiplier to something which seems to work
better (it is, rather arbitrarily, the median of the primes between
2^27 and 2^28).  Some more theory should almost certainly go into the
choice of this constant, to avoid more pathological cases.  But for
now, this choice fixes a serious bug, and seems to do at least as well
at scattering as the old choice on a handful of simple testcases.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

The two patches below (compile)fix some fall-out from the tty cleanups.
Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Move extern declarations of ia64_{reg,unreg}_MCA_extension()
to mca.h.  Delete declaration of ia64_mca_ucmc_other_recover_fp()
which doesn't exist.

Patch supplied by Hidetoshi Seto.
Signed-off-by: Tony Luck <tony.luck@intel.com>

This is the latest/Updated OS_MCA handler which try to do recovery
from multibit-ECC/poisoned memory-read error on user-land.
(Thank you very much for comments, Keith and Grant!)

I'd still appreciate it if anyone having good test environment
could apply my patch and could report how it works.
(especially reports on non-Tiger/non-Intel platform are welcome.)
Signed-off-by: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>

into agluck-lia64.sc.intel.com:/data/home/aegl/BK/linux-ia64-release-2.6.9

change the return value of PFM_CREATE_CONTEXT from
EAGAIN to ENOMEM when the sampling buffer size is
larger than the RLIMIT_MEMLOCK limit of the task.
signed-off-by: Stephane Eranian <eranian@hpl.hp.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>

fix a problem in pfm_check_task_state() and pfm_task_incompatible()
which was caused by the introduction of the new TASK_TRACED state.
Tool would fail to attach to a process,i.e., PFM_LOAD_CONTEXT would
fail. With the fix perfmon now accepts to operate on tasks which are
in either TASK_STOPPED or TASK_TRACED state. The problem was tracked
down by Alex Williamson from HP who also submitted the patch.
signed-off-by: Stephane Eranian <eranian@hpl.hp.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>

Some dual use type errors still exist, where the
bios mapping is type-confused.

A few one-liners removed hundreds of lines of warnings.
The driver was clean, just not using the proper types.

It was almost correct, apart from some silly details.

The x86 ROM probing is still wrong, and doesn't use the proper
PCI MMIO accessor functions. Sparse (correctly) warns about
it.

Remove absolutely tons of totally unnecessary casts,
once the types are done right in the first place.

The driver is still ugly as hell and needs some serious
indentation fixing, but this makes it not spew hundreds
of warnings any more.

Signed-off-by: Greg Kroah-Hartman <greg@kroah.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Signed-off-by: Greg Kroah-Hartman <greg@kroah.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

We register it with the ACPI layer, and it's alive long
after init.

Noted by Bjorn Helgaas.

Only recently digitized for our edification.

Basic cleanup - replace untyped/wrongly typed "dev->base_addr" with
use of a strongly typed "ioaddr".

Fixed up resulting mii_delay() search-and-replace error noticed by
Andrey Klochko.

Verified by Franz Pletz.

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

Played a bit with overcommit the past hour.

Am not entirely satisfied with the no overcommit mode 2 -
programs segfault when the system is close to that boundary.
So, instead of the somewhat larger patch that I planned to send,
just symbolic names for the modes.

I didn't have ISDN builds on in my tree for some reason hence missing these
two from the tidy ups at the end. Marcel Holtmann also came up with the
same fixes although I didnt find that email until I did these.

Signed-off-by: Alan Cox

The hash table wasn't fully initialized on some pSeries that had
the workaround for no batching.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Remove an extra left-paren.
Acked-by: Marcelo Tosatti <marcelo.tosatti@cyclades.com>
Signed-off-by: Randy Dunlap <rddunlap@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

and do not clear dirty flag if page only partially written.

SGI Modid: xfs-linux:xfs-kern:19622a
Signed-off-by: Nathan Scott <nathans@sgi.com>

SGI Modid: xfs-linux:xfs-kern:19619a
Signed-off-by: Nathan Scott <nathans@sgi.com>

SGI Modid: xfs-linux:xfs-kern:19601a
Signed-off-by: Nathan Scott <nathans@sgi.com>

This only causes a problem in the -mm tree now, but Zou Nan hai and
Shaohua Li sent me this fix so that SMP systems compiled with the
generic kernel (which turns on NUMA and CPUSETS in -mm) won't hit
a BUG_ON in kernel/cpuset.c guarantee_online_mems() which is called
from acpi_early_init()
Signed-off-by: Tony Luck <tony.luck@intel.com>

Here's a defconfig that comes up on my box and has a reasonable set of modules
enabled.
Signed-off-by: Jesse Barnes <jbarnes@sgi.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>

 
If we read and write the PROM chips at the same time, as might happen at boot
when salinfo extracts MCA records and a user is checking the PROM revision
in /proc/sgi_prominfo, an MCA might occur, since the PROM chips can't be
accessed that way.  This patch fixes the problem for systems with new PROMs
(>= 3.50) by using the SAL to do PROM reads.
Signed-off-by: Jesse Barnes <jbarnes@sgi.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>

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

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

Bartlomiej Zolnierkiewicz wanted to remove the "cbio" structure
from struct request.  Unfortunately, MMC was using the same
"workaround" as IDE to walk the BIO list.  With this change, we
stop using "cbio" and instead use proper scatter-gather lists.

This patch adds a 10 ms delay in the power up loop. Some cards fail
to power up in time without it.

Pierre Ossman reported a problem where his cards do not correctly
initialise with the current MMC implementation.  However, if we
tell the card to go back to the idle state between probing the
supply voltage and telling the card about the chosen voltage, his
cards work - and we don't seem to upset existing cards.

This allows users to provide a fixed major number - early 2.4 MMC
drivers used block device major 60, which is "local/experimental
use"