From: Rusty Russell <rusty@rustcorp.com.au>

From: Andrea Arcangeli <andrea@suse.de>

We should drop module sem before calling mod->exit, for practical reasons:
too many module exit functions oops or hang, resulting in a permenantly held
module sem, which blocks all module ops including lsmod.

From: "David S. Miller" <davem@redhat.com>

Currently, for an existing DMA mapping, there is a way to transfer buffer
ownership back to the cpu, yet there is no way to give it back to the device
again explicitly.  The latter really is needed on platforms where the PCI
subsystem does not snoop the cpu caches, MIPS is one example.

Many drivers were expecting the existing DMA sync interface to handle both
directions, which was wrong.

Now, with this change, we have explicit interfaces for DMA syncing to/from
the device and the cpu.

From: Andi Kleen <ak@suse.de>

This patch has been in the SuSE 2.4 kernel forever, but for some reason
never made it mainline.

It works around the infamous "only works stable when a mouse is plugged in"
problem some AMD 768MPX Dual Athlon chipsets have.  The problem happens
because the chipset can hang when PCI prefetch strides from a RAM page into
the VGA text buffer.  When a PS2 mouse is plugged in the BIOS reserves a
page before the VGA text buffer, which stops the prefetch early.

This patch always reserves this page when the chipset could be AMD768MPX.
This can be only done early in bootmem setup.  Because it's difficult to
scan the PCI bus that early it just always reserves this page when the CPU
is an Athlon.  Normally it should not make a difference because the BIOS
will have reserved that page anyways when a PS/2 mouse is plugged in.

From: NeilBrown <neilb@cse.unsw.edu.au>

kernel parameters:

   raid=partitionable
     will make all auto-detected md arrays partitionable

   md=d....
     will assemble an array as a partitionable array.

From: NeilBrown <neilb@cse.unsw.edu.au>

Use "shedule_timeout()" instead of yield() as it seems to wait for less
time.

From: Rusty Russell <rusty@rustcorp.com.au>

Make it clear that the reason we do wait_task_inactive is because
kthread_bind frobs with k->thread_info->cpu, which is only legal because
the task is definitely not running.

We can't use the normal migration thread code here, because it doesn't let
us bind to cpus which are offline yet, and also because we use this to
start the migration threads themselves.

From: Arnd Bergmann <arnd@arndb.de>

Dave Jones already removed some of the useless __KERNEL_SYSCALLS__ defines
in various files, this gets rid of almost all the others.  Replacing
execve() is nontrivial, so I left those in for now.

For all the other system calls that are currently used from inside the
kernel, calling the sys_* function directly should always have an identical
effect.

From: Anton Blanchard <anton@samba.org>

Distribute boot time memory allocations across all nodes, from Manfred
Spraul.

We want to spread memory across nodes to avoid all allocations ending
up on node 0.

Spreading boot time allocations around also helps us to avoid node 0
becoming the hot node.

I took it for a spin:

buddyinfo before:
Node 7, 0    2    1    1    0    2    1    2    1    2    1    2    741
Node 6, 0    0    0    2    0    2    1    1    2    2    2    2   1002
Node 5, 0    0    0    2    0    2    1    2    1    2    2    2   2006
Node 4, 0    0    0    2    0    2    1    2    1    2    2    2   2006
Node 3, 0    0    0    2    0    2    1    2    1    2    2    2   2006
Node 2, 0    0    0    2    0    2    1    2    1    2    2    2   2006
Node 1, 0    0    0    2    0    2    1    1    2    2    2    2   1002
Node 0, 0    0   38    7    0    1    1    1    0    0    0    0   1998

buddyinfo after:
Node 7, 0    1    0    1    1    1    1    0    0    0    1    2    738
Node 6, 0    1    0    1    1    1    0    1    0    0    2    2   1002
Node 5, 0    0    0    1    1    1    1    0    0    0    2    2   2006
Node 4, 0    1    0    1    0    1    1    0    0    0    2    2   2006
Node 3, 0    0    0    1    0    1    1    0    0    0    2    2   2005
Node 2, 0    1    0    0    0    0    0    1    0    0    2    2   2006
Node 1, 0    2    1    1    0    1    1    1    0    0    2    2   1002
Node 0, 0   20   45    8    3    0    1    1    1    1    0    1   2004

Change in free memory due to patch:

Node 7 -54.08 MB
Node 6  -6.33 MB
Node 5  -6.09 MB
Node 4  -6.14 MB
Node 3 -22.15 MB
Node 2  -6.05 MB
Node 1  -6.12 MB
Node 0 107.35 MB

As you can see we gained over 100MB on node 0.

From: Kingsley Cheung <kingsley@aurema.com>

A number of scheduler counters wrap around after 47 days.  The context-switch
counter can wrap around after considerably less time.

Convert them to 64-bit values.

From: Anton Blanchard <anton@samba.org>

The recent NUMA changes fail to compile with large cpumasks, we need to use
a temporary to get around the type checking.

Remove unnecessary initialization in notifier_block

Version 0.03 of wdt977.c - Changes that were made are:
* Extract the stop code in a seperate function (wdt977_stop)
* Extract the start code in a seperate function (wdt977_start)
* Rename kick_wdog to wdt977_keepalive for consistency
* Extract the watchdog's status code to a seperate function (wdt977_get_status)
* Change the way we deal with the watchdog timeout:
   Up till now we used timeoutM (in minutes) as the correct value and then
   calculated timeout as being timeoutM*60 or *timeoutM*120 (depending on
   wether or not we have the netwinder hardware bug).

   From now on timeout is the correct value and we calculate timeoutM out
   of it. Because of this we start with checking wether or not we have a
   correct timeout value (if not we reset it to the default value) and we
   automatically calculate timeoutM. Each time we change timeout with a
   correct timeout value, we recalculate timeoutM.
* Extended ioctl code with WDIOC_SETOPTIONS and updated the watchdog_info structure
* Added notifier support

Code has been tested by Woody

Two small fixes:
* Make cards_found a global variable so that if we remove the
  pci device we can count down.
* If we can't find a correct I/O address for the card, then we
should disable the card again.

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

into redhat.com:/spare/repo/net-drivers-2.5

* the scan-channels message seemed to always give invalid output.
  Look at the constant, and discover we are sending another
  message entirely.  Fix the constant (CARM_MSG_IOCTL).

* s/MISC_SYNC_TIME/MISC_SET_TIME/

* list some additional messages

* bump version number

* Not setting cb->skb = NULL after releasing skb to OS or during
  initialization of cbs.  Reported by Deepak Saxena
  [dsaxena@plexity.net].

into redhat.com:/spare/repo/net-drivers-2.5

patch below adds bus_info for xirc2ps_cs; anaconda depends on this info to
be there.

This patch adds a helper function to initialize the debug bit mask
for use with netif_msg_*.  When the debug_value is out of range
it returns the default_msg_enable_bits.  Tested IA32.

This patch corrects the names of contributors of changes to the pcnet32
driver.

smctr_chk_mca() can be __init.

eepro_print_info() can be __init.

Andrew de Quincey added wol support to forcedeth.
The patch also renames additional function, to help analyzing backtraces. 

 From:  <adobriyan@mail.ru>

The sense of the comparison was signed, but the code was testing
an unsigned variable for less-than-zero.