into home.transmeta.com:/home/torvalds/v2.5/linux

into home.transmeta.com:/home/torvalds/v2.5/linux

into halogen.(none):/home/davej/src/kernel/2.5/agpgart

into tetrachloride.(none):/mnt/stuff/kernel/2.5/agpgart

With Martin's blessing, here is the s390x syscall compatability stuff.

Another one that James noticed..

into kroah.com:/home/linux/linux/BK/pnp-2.5

I missed one hunk when splitting the patch for removing
try_inc_mod_count().

The might_sleep() check in kmm_cache_alloc() is currently disabled if slab
debugging is disabled.

Fix that so it's enabled if might_sleep debugging is enabled.

add_disk_randomness() is causing a context switch per disk request.  It
is scheduling process-context work one timer tick in the future for
every request.

But is has a buffer for this, so change it to not set up the
process-context work until that buffer is half full.

The patch reduces the context switch rate during a 20 megabyte/sec
write to scsi from 150/sec to 50/sec.

George Anzinger identified the following problem: when a secondary CPU is
coming up, it calls printk() before it is "online".  It calls the console
drivers before its per-cpu storage has been prepared.  And the vga console
driver does a mod_timer().  This CPU's timers have not yet been initialised;
it is not clear why this doesn't oops - George thinks it is because virtual
address zero is still accessible at that time.

I believe the right way to fix this is to change printk so that a not-online
CPU will not call the console drivers.  Because printk should always be
callable.  If the CPU is not online the message is buffered, so the next
caller to printk who is online will actually display it.

ia64 has been doing exactly this for ages, so we can remove the
arch_consoles_callable() hook and just open-code the cpu_online() test in
printk.



That fixes things up for the secondary CPUs.  But this change causes a
problem for the boot CPU: it is being marked online very late in boot, so the
printk buffer is being displayed much later than we would like.

I believe that the solution to this is to mark the boot CPU online much
earlier.  So in this patch we call the new arch-provided function
smp_prepare_boot_cpu() immediately after the boot CPU's per-cpu areas are set
up.  Its mandate is to (at least) mark the boot CPU "online".

The change has been reviewed by davem and rth.  No comments were received
from the other arch maintainers.

ALIGN() currently has global scope in <linux/cache.h>.  This causes a
compilation error in the defxx driver.

Move ALIGN() to <linux/kernel.h> and change the defxx driver to use the
generic macro in place of its own.

It will leak a bh ref on ENOMEM.

Patch from Ben LaHaise and Bill Irwin.

As discussed a month or so ago, all parties agree that the new
hugetlbfs and shm APIs are sufficient for the 2.6 kernel.

I marked the vacated syscall slots as "available for reuse".  I doubt
that there will be any conflicts when that happens.

Remove unused task_struct.swappable.

Some workloads (Oracle...) use a huge number of VMA's.  They are
currently a tidy 64 bytes in size, and padding them out to 128 on P4's
is not worthwhile.

Patch from Ravikiran G Thirumalai <kiran@in.ibm.com>

Creates a simple "kmalloc for each CPU" API.  This will be used for net
statistics, disk statistics, etc.  (davem has acked the net patches which use
this code).

kmalloc_per_cpu() is available to modules, unlike the current static per-cpu
infrastructure.

swsusp gets confused when pages which it freed do not appear in the
buddy lists.  So provide a function which will drain the calling CPU's
per-cpu-pages into the buddy.

The patch has been tested by Pavel.  Presence of the new code is
conditional on CONFIG_SOFTWARE_SUSPEND.

Patch from Mark Fasheh <mark.fasheh@oracle.com>

A missing chunk from the semtimedop() implementation.  It enables
semtimedop() for 32-bit ia32 applications running on ia64.

It also changes semop() to go direct to sys_semtimedop(), bypassing a
function call.

Cleanup patch from Hollis Blanchard <hollis@austin.ibm.com>

We have a large number of private implementations of BIN_TO_BCD and
BCD_TO_BIN, which are all the same.  And a lot of them are inflexible because
they modify their arg:

	#define BIN_TO_BCD(val) ((val)=(((val)/10)<<4) + (val)%10)

- Create (in <linux/bcd.h> a generic BIN2BCD/BCD2BIN which does not modify
  its arg

- Create generic BIN_TO_BCD/BCD_TO_BIN which uses the above

- Update lots of callers to use the new generic version.

It seems to have been displaying zero for the past several years, and
we can't think of anything useful to put in there.

Patch from Paul Jakma <paulj@alphyra.ie>

Documentation/sysctl/fs.txt is incorrect wrt to the middle number of
/proc/sys/fs/file-nr.  The current docs state it is the number of
in-use file-handles, from observation this number is actually the
number of /unused/ file-handles - we've observe that when it hits 0 we
get file handle problems (rather than when it hits == file-max as docs
imply).

Turns out there was an undocumented "c" flag for MODULE_PARM.

This implementation is a little ugly, but it works, and will do for
compatibility (I haven't implemented such a two-dimensional array
primitive, but the whole point of the module_parm et al is that they are
extensible).

The name is just a bit too generic, and we already use the _X86 prefix
for lots of other stuff in that area.  Dito for the never use CONFIG_PC
and CONFIG_VISWS.

This patch combines my patch which statically initialises the crc32
tables so they can be used at any time (during initialisation) and
Joakim Tjernlund's patch to speed up the crc calculations by doing word
operations instead of exclusively byte.

The crc routines are used extensively in jffs2 where speed is very
important.

I need the crc32 routines to calculate a checksum on values read from an
eeprom which contain cpu speed and memory size information - so they are
needed very much earlier in the initialisation process than they are
currently available.

Clean up searching for best frequency, and add one safety check.