- The balance_dirty_pages() logic is simply wrong.  It goes:

	if (value > threshold)
		go_and_write(value - threshold);

  which is just fine for a single process writing data.  But
  for many processes, they *all* go and bring things back into
  balance, and too much data gets written out.

- The

	go_and_write(this much)

  logic is inoperative, because I turned off the ->writeback_mapping()
  function in ext2.  So a call to writeback_unlocked_inodes(this_much)
  doesn't actually decrement and test *this_much.  It will walk every
  inode, all the time.  Silly.

So quickly fixing the above things, the amount of dirty+writeback
memory in the machine nicely stabilises at 500 megabytes across
the run.

The new consistency checks in buffer.c are complaining about
block_symlink's ->readpage against an already uptodate
page.

- Fix compilation of piix.c

- Remove the PADAM_ prefix from sleep hwgroup member.

- Fix Pacific Digital host chip driver API.

- Fix Tekram host chip driver API.

- Fold hwif_unregister() directly in to channel code.

Add support for request tagging of the block queue.

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

- Fix macro varargs usage, you need to specify a variable name
before the triple dot or else most current compilers complain.
See include/linux/ext2_fs.h:ext2_debug for another example where
it is done correctly.

- Fix bug in strsep/strchr changes, dereference *this_char
not *value at top of while loop.  This matches how the same
code in fs/autofs4/inode.c looks right now.
Withtout this autofs loading causes an OOPS as the first
time through the loop *value is dereferencing a NULL pointer.

Fix some set_bit() warnings.

The real rule is not "block size is at most 32Kb", it's "block size is
at most 8 times the fragment size".

into penguin.transmeta.com:/home/penguin/torvalds/repositories/kernel/linux

(originally caught by Paul Larson <plars@austin.ibm.com>)

	Patch fixes a couple of idiotic bugs in notify_change() (my
fault).  Unary operations have higher priority than binary ones, so
if (!valid & ATTR_MODE) doesn't do the right thing.  I plead temporary
braindamage...

The below patch on 2.5.12 takes the 'dbench 32' throughput on a
1024 megabyte machine from 35 megs/sec up to 125 megs/sec.
That's on the second run.  The first run is much slower because
the ext2 bitmaps aren't in cache.

Presumably, s/60/80/ will take it even higher.

Remove bogus BUG_ON().  On UP kernels spinlocks become NOPs so it is
not possible to test the state of a lock.

Patch to the aha152x driver, which makes it work under 2.5.

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

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

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

Drop back to eieio in spinlocks for the moment due to performance
issues of sync on power3

into samba.org:/scratch/anton/linux-2.5_ppc64

Pretty boring stuff.  Four compile fixes for 2.5.12.

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

If snd-pcm.o is selected built-in, snd-timer.o needs to be get built-in
as well, since snd-pcm.o needs snd-timer.o's symbols.

Remove description of long obsolete (pre-2.4) Makefiles and describe the
current system.

Get rid of long obsolete (and unused) MOD_IN_SUBDIRS and unnecessary
MOD_DIRS variable.

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

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

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

block_flushpage() used to be a macro which pointed at the
exported discard_bh_page().  I turned block_flushpage() into
a real function but forgot the export.

path_lookup() is no longer an inline function, and needs to be exported
for jffs2, nfsd and af_unix to load as modules.

Douglas Gilbert's SCSI cleanups

This patch has the last bit of Justin Gibb's patch
described in:
http://marc.theaimsgroup.com/?l=linux-scsi&m=101200279101550&w=2
[this bit for the sd driver]

There is also a major code cleanup of the sd driver with
documentation headers added and a few obvious bugs fixed
described in:
http://marc.theaimsgroup.com/?l=linux-scsi&m=101798201714399&w=2
I did this cleanup.

Justin's patch has been in Dave's tree for several months while
my code cleanup patch has been there since 2.5.9-dj1 .

The attachment is part of a patch from Justin Gibbs
described in:
http://marc.theaimsgroup.com/?l=linux-scsi&m=101200279101550&w=2

The original patch was targeted at lk 2.4 and Dave forwarded
ported it into 2.5. Other bits (e.g. sr) have already found
there way into your tree. One bit in the sd driver will be
included in my following patch.

	Missing checks added...

 - switch blk_get_ra_pages() to struct block_device *, move its use
   from bdget() to do_open().