From: Nick Piggin <nickpiggin@yahoo.com.au>

From: Suresh Siddha <suresh.b.siddha@intel.com>

Node max rebalance interval is too large.  It is currently dependent on
number of online cpus.  For 16 cpu system, max node balance interval in
busy case is 32 seconds.  Agreed that it will use max 32 seconds only when
it doesn't find imbalance for a long time.  But this will lead to slow
response time in cases where load runs for a second with no imbalance and
suddently creates an imbalance.  My patch makes the busy max node rebalance
interval equal to the base [scheduler].

From: Nick Piggin <nickpiggin@yahoo.com.au>

Analysis and basic idea from Suresh Siddha <suresh.b.siddha@intel.com>

"This small change in load_balance() brings the performance back upto base
scheduler(infact I see a ~1.5% performance improvement now).  Basically
this fix removes the unnecessary double_lock.."

Workload is SpecJBB on 16-way Altix.

From: Nick Piggin <nickpiggin@yahoo.com.au>

Fine-tune the unsynched sched_clock handling.

Basically, you need to be careful about ensuring timestamps get correctly
adjusted when moving CPUs, and you *can't* look at your unadjusted
sched_clock() and a remote task's ->timestamp and try to come up with
anything meaningful.

I think this second problem will really hit hard in the activate_task path
on systems with unsynched sched_clock when you're waking up a remote task,
which happens very often.  Andi, I thought some Opterons have unsynched
tscs?  Maybe this is causing your unexplained bad interactivity?

Another problem is a fixup in pull_task.  When adjusting ->timestamp from
one processor to another, you must use timestamp_last_tick for the local
processor too.  Using sched_clock() will cause ->timestamp to creep
forward.

A final small fix is for sync wakeups.  They were using __activate_task for
some reason, thus they don't get credited for sleeping at all AFAIKS.

And another thing, do we want to #ifdef timestamp_last_tick so it doesn't
show on UP?

From: Nick Piggin <nickpiggin@yahoo.com.au>

"Siddha, Suresh B" <suresh.b.siddha@intel.com> noticed a problem in the
cpu_load averaging where the integer truncation could sometimes cause cpu_load
to never quite reach its target.

I'm not sure that you could demonstrate a real world problem, but I quite
like this fix.

Fix bug identified by Chris Mason.

If writeback_inodes is left holding a ref on the superblock's last inode then
the superblock list walk can race with umount and the superblock can be
released.

Take and put a ref against the superblock to fix that.

From: Adam Litke <agl@us.ibm.com>

Teach the x86 stack tracing code to use frame pointers, if they are available.
It eliminates all the false-positives in the normal stack traces.

This is a big improvement, and -fomit-frame-pointer seems to make no
difference at all to generated code size.  Maybe we should kill off
-fomit-frame-pointer.

This fixes a problem with multiple IDE controllers in a system.

The problem is that pcibios_fixups table (in arch/i386/pci/fixup.c) uses
the pci_fixup_ide_trash() quirk for Intel's ICH3 (my case specifically
8086:248b).  This clears any bogus BAR information set up by the BIOS.

In a system which has multiple ICH3's can't use any of the IDE
controllers beside the one on the first ICH3.

Anyhow, the fix is to make sure pci_fixup_ide_trash resets the BARs only
for first time being called, so the subsequent IDE controllers will use
the BIOS BARs.  This is better than "loosing" all these IDE controllers
in the case their BARs set right.

The issue discussed and agreed with Bartlomiej Zolnierkiewicz (see below).

This avoids warnings about unused variables.

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

into redhat.com:/spare/repo/i810-audio-2.6

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

x86_64 has an incorrect ffs() implementation.  The asm uses "g" instead
of "rm" for the bsfl instruction.  (This was spotted by Yuri Per.)

bsfl does not accept constant values but only memory or register ones.
On i386 the correct "rm" is used.

This causes NTFS build to fail as gcc optimizes a variable into a
constant and ffs() then fails to assemble.

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

into kernel.bkbits.net:/home/davem/net-2.6

this trivial fix makes the AVM B1 PCMCIA cards work with 2.6

This adds links to the driver and device inside /sys/class/net/ethX for
PowerPC Virtual Ethernet devices.

This fixes a bug where different partitions were assigned the same MAC
address.  Also, according to Anton, gcc 3.5 didn't like our mac_addr_p
gymnastics, so this ends up fixing that as well.

This patch fixes SMP boot on Apple Xserve G5

From: Stephen Rothwell <sfr@canb.auug.org.au>

It is possible to attach a virtual disk to a logical partition on an iSeries
machine so that it is read only to the partition.  This patch allows Linux to
use such virtual disks.

From: Paul Mackerras <paulus@samba.org>

After I sent the recent patch to include/asm-ppc64/io.h which put stronger
barriers in the I/O accessor macros, Paul McKenney pointed out to me that a
writex/outx could still slide out from inside a spinlocked region.  This patch
makes the barriers a bit stronger so that this can't happen.  It means that we
need to use a sync instruction for wmb (a full "heavyweight" sync), since
drivers rely on wmb for ordering between writes to system memory and writes to
a device.

I have left smb_wmb() as a lighter-weight barrier that orders stores, and
doesn't impose an ordering between cacheable and non-cacheable accesses (the
amusingly-named eieio instruction).  I am assuming here that smp_wmb is only
used for ordering stores to system memory so that another cpu will see them in
order.  It can't be used for enforcing any ordering that a device will see,
because it is just a gcc barrier on UP.

This also changes the spinlock/rwlock unlock code to use lwsync ("light-weight
sync") rather than eieio, since eieio doesn't order loads, and we need to
ensure that loads stay inside the spinlocked region.

From: Paul Mackerras <paulus@samba.org>

When I sent the patch to uninline the spinlocks, I inadvertently left out
the change to arch/ppc64/Kconfig which defines the config symbol for
inlining the locks (CONFIG_SPINLINE now).  This patch adds it.  It also
adds a symbol CONFIG_PPC_SPLPAR which enables the code for calling the
hypervisor on shared-processor logically-partitioned system to yield the
physical processor to the lock holder when spinning.  (The code that
depends on this symbol is already present in arch/ppc64/lib/locks.c.)

From: Anton Blanchard <anton@samba.org>

It turns out we do avoid irq recursion on iseries so remove the workaround.

From: Anton Blanchard <anton@samba.org>

We check nvram_fetch/nvram_store against -1, so better not make these
unsigned.

From: Anton Blanchard <anton@samba.org>

Based on feedback from the hardware guys align jiffies and tb_last_stamp.
We update both regularly and there are other read only, heavily accessed
things that share those cachelines.

From: Anton Blanchard <anton@samba.org>

We need to set MSR_RI in iseries exception prolog.

From: Anton Blanchard <anton@samba.org>

We were allocating radix tree nodes in the interrupt code with GFP_KERNEL
under a spinlock.  Change it to GFP_ATOMIC.

From: Paul Mackerras <paulus@samba.org>

The patch below moves some declarations from C files into the appropriate
header file in include/asm-ppc (and removes an unused local variable in a
function).

From: Paul Mackerras <paulus@samba.org>

Matt Porter's recent changes broke the compile for non-4xx ppc32 systems,
unfortunately.  I get an error that mfdcr is not defined in
include/asm-ppc/ocp.h when compiling for powermac (reasonable, since the
mfdcr instruction only exists on 4xx processors).  The patch below fixes
it.

From: Matt Porter <mporter@kernel.crashing.org>

Merge all current PPC44x ports against new OCP.

From: Matt Porter <mporter@kernel.crashing.org>

Merge all current PPC40x ports against new OCP.

From: Matt Porter <mporter@kernel.crashing.org>

Update all current 4xx defconfigs for new OCP.

From: Matt Porter <mporter@kernel.crashing.org>

Merge misc.  4xx core fixes and support for the new cascade scheme in the
440gx.

From: Matt Porter <mporter@kernel.crashing.org>

Merge PPC4xx-specific OCP support for new OCP core.

From: Matt Porter <mporter@kernel.crashing.org>

Merge PPC44x library support against new OCP.

From: Matt Porter <mporter@kernel.crashing.org>

Merge Bubinga/405EP support against new OCP.

From: Matt Porter <mporter@kernel.crashing.org>

New OCP infrastructure ported from 2.4 along with several enhancements. 
Updated patch with comments from hch and Valdis.