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

What happens when you get a patch that does something an applied patch 
already does, but a little better, and merge it sloppily: You end up calling
the wrong function because you've defined equivalent methods in two places.

Bad Pat, Bad. 

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

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

into osdl.org:/home/mochel/src/kernel/devel/linux-2.5-core

into kroah.com:/home/greg/linux/BK/pci-2.5

into kroah.com:/home/greg/linux/BK/gregkh-2.5

From Mike Anderson: 

I have been using it on an outstanding patch for scsi_set_host_offline. It
appears to work fine in my testing.

into kroah.com:/home/greg/linux/BK/i2c-2.5

There is a typo in the current 2.5.70 bk version of class_device_release that
was not there in my original patch.  By confusing the class and the class_device,
the release function oops.  cd->release is always the function itself (class_device_release),
cls->release is the one setup for the class (net_class in my case).

i2c-core.c contains 2 loops that iterate over the list of the clients attached
to an adapter and detaches them. Detaching the clients will actually remove
them from the list the loop is iterating over. Therefore the
list_for_each_safe() method has to be used.

This patch fixes the various return values in the w83781d_detect()
error paths.  It also cleans up some formatting here and there.
It should be applied on top of the previous one.

It works for me; same caveat as above w.r.t. ISA.

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

Now the only users of this directly should be the pci core and arch specific
pci core code.

This will let not have to export the pci_devices variable.

This makes the driver build properly now, and removes a direct access
of the pci_devices variable.

Yeah, this is commented out code, but just trying to be complete...

The only thing preventing this from happening earlier was the circular sysfs
registration dependency - it would need to be initialized before it was 
registered, but needed to be registered before it was initialized. 

With kobjects gone from struct filesystem_type, the dependency no longer 
exists and we don't have to special-case the possibility that a kobject will
be passed to kobject_get with a refcount == 0. 

Note that a kobject with a count of 0 in that function is still a bug, but 
one in the subsystem making the call. We should add a debugging hook to dump
the stack if it does happen. 

It was initially added for the immediate gain of being able to see what 
filesystems were registered in the system. However, it was done without
proper analysis of the lifetime rules associated with them. 

Someday, we should convert struct filesystem_type and struct super_block to
use kobjects, though it will take sufficent time and effort to get it right.
Until then, we go without..

This patch gets rid of CONFIG_ALL_PPC, which was a very confusing
option, since it didn't actually mean "ALL" at all, it was more a
"common set" thing.

The primary replacement for CONFIG_ALL_PPC is CONFIG_PPC_MULTIPLATFORM.
I have also defined CONFIG_PPC_PMAC, CONFIG_PPC_PREP and CONFIG_PPC_CHRP
for selecting code which is only needed for one of the three platforms
that CONFIG_ALL_PPC represented.  This is something that we (the PPC
community) have been talking about doing for some time.  There is also a
CONFIG_PPC_OF which is for PPC machines with Open Firmware, which is
currently powermacs and CHRP machines.

At the moment, CONFIG_PPC_{PMAC,PREP,CHRP,OF} get unconditionally
defined if CONFIG_PPC_MULTIPLATFORM is selected, but in future this
split will let us have more control over what gets included, so that
for example we don't necessarily have to include powermac bits in a
kernel for a PReP machine.

I have gone through the uses of CONFIG_ALL_PPC one by one and decided
which of the new symbols best represents the set of machines that need
the code in question.  In fact most of the uses of CONFIG_ALL_PPC in
the drivers have been replaced by CONFIG_PPC_PMAC.  The other symbols
are mostly confined to the PPC architecture code.

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

and fix the start/stop thing as well. I think this is all of them.

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

you should not drop errors.
  - proper error detection during initialisation

this cuts out old cruft.

Here's the non-inlined doc for the gadget API.

> Ben, it looks like your patch broke something for USB keyboards, any
> idea?

Yep, my patch killed hid-input from scanning HID_OUTPUT_REPORT's. Fixed
with this patch for 2.5.70+bk. I'll send one for 2.4.x in a few minutes.

A few choice handling fixes:
- only visible choice values define the new state of the complete choice
- improve handling of choices without visible value
- two new warnings

If /proc/mdstat is large, or reads are for a small size,
then the last line of /proc/mdstat is repeated infinitely.

This patch will fix it.

Some paths free things twice, others free un-initialised values :-(
Not any more.

When raid1 writes, it needs to schedule writes to some number
of devices, and when all writes have completed, the r1_bio
structure that holds it all together must be freed.
However we must make sure not to free it before all devices
have been considered for submitting writes to.

This happens in two places: when submitting a normal write request
and when submiting a write as part of resync.

This patch makes both these places:
  the same
  simpler
  more correct.