Big comment, because it wasn't clear why this cast was valid.
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

forte annotated

msnd annotated

ymfpci annotated

maestro annotated

btaudio annotated

ad1889 annotated

Remaining bits and pieces in sound/oss annotated.

At that point sound/* is done - we have no noise warnings left there.

maestro3 annotated

sonicvibes annotated

ali5455 annotated

cs46xx annotated

i810 annotated

rme96xx annotated

esssolo annotated

es1370 annotated

es1371 annotated

cmpci annotated

cs4281 annotated

The rest of annotation for ALSA drivers; only OSS stuff left to do in
sound/*

emu10k1 annotated.

Annotated driver and ioctl structure used by it.

Both ALSA and OSS variants

Both PCI and PCMCIA variants of driver annotated

sb annotation (both ALSA and OSS drivers)

Both ALSA and OSS drivers + wavefront ioctl structure annotated.  NB:
both should be switched to generic firmware loading - as it is, they are
using a homegrown and rather ugly variant

GUS annotated.

The rest of sond/core annotated; reverted bogus addition of __user in
snd_seq_kernel_client_ctl() - I should've guessed from the name alone
;-)

Annotated sound/drivers/opl*

The tricky part here was an iterator that used to take a callback and
argument for that callback as parameters.  Iterator itself didn't care
what type that argument had been; it's entirely up to callback.  The
thing is, two callbacks expect (and get) char __user * while other two
expect (and also get) char __user **.

Iterator used to use void * as "opaque data"; I've switched it to
unsigned long.  Note that there was nothing that said "it's a pointer" -
use of callback that would take e.g.  int is also perfectly legitimate. 

The rest is triviali annotation.

Annotated method prototypes, commonly used ioctl-only structures and
generic helpers.

Instances of methods in drivers are _not_ annotated at that point - they
will go with driver patches

Trivial parts of sound/core annotation.

copy_from_user() where copy_to_user() should've been.  Unfortunately, on
x86 they are almost identical, so it went unnoticed for quite a while.

msnd_pinnacle/msnd_classic do copy_{to,from}_user under a spinlock.

Taken out of spinlock (into a temp. buffer). Calls of msnd_fifo_{read,write}
always go from kernel buffer now, so we can drop the 'int user' argument in
them _and_ simplify error handling - all errors were from copy_..._user() and
now these are called directly by dsp_read()/dsp_write().

	In some cases snd_sb_csp_load() did kmalloc() and copy_from_user()
under a spinlock.  Split into snd_sb_csp_load() and snd_sb_csp_load_user() -
ther former always from kernel pointer, the latter - from userland.
	snd_sb_csp_load_user() doesn't take any locks itself, it just
does kmalloc, copy_from_user and calls snd_sb_csp_load() to do the rest.

Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

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

Patch from Ben Dooks

Fix the minor number of the serial device to be the same
as the configuration in the 2.4.x series kernels.

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