There are races WRT walking the lists of the buses and classes and the 
device being unregistered while it's the current node we're accessing. If 
the node was deleted, ->next would point to itself, and we'd sit in an
inifinite loop (until that memory was freed from underneath us and we Oops'd).

This changes device_unregister() to only mark the device as not present. put_device()
deletes the device from all the lists it belongs to, while holding device_lock,
then drops it to call device_del().

This should guarantee that everything in the device is valid until the last reference
goes away, regardless of whether it's still physically present or not. 

Here's a patch to add an Andrew File System (AFS) driver to the kernel.
Currently it only provides read-only, uncached, non-automounted and
unsecured support.

This adds RxRPC support to Linux for use by the AFS filesystem

Someone adding tests for failed kmalloc's (which is fine in itself)
did so in a way that would leave some PCMCIA data structures in
inconsistent states... and also introduced a fatal bug affecting
PCMCIA memory cards even when there are no kmalloc failures.

This is the RCU core patch from akpm's tree. It has been in his
tree since about 2.5.37-mm1 along with dcache_rcu and so far it has
worked fine. For 2.5, I am hoping that we might get the following
RCU patches included -

1. rt_rcu - ipv4 routecache lookup. Davem agreed to include this patch
   if and when you include RCU core in your tree.

2. dcache_rcu (by Maneesh Soni) - dcache lookup avoiding dcache_lock as
   much as possible. This has been akpm's tree - stable and gives us
   good yield. I have been submitting this to Viro and I will publish
   some more benchmark numbers later to help decide on this.

This RCU core implements RCU APIs, call_rcu() and synchronize_kernel(),
by monitoring a per-CPU quiescent state (idle/user etc.) counter.
call_rcu() queues a callback to be invoked after all the CPUs have
gone through a quiescent state. Queuing is per-CPU and each per-CPU
batch gets a batch number. As batches get their turn, a global
cpu mask is used to keep track of CPUs pending quiescent state.
Checking for quiescent cycle is done by saving the per-CPU
counter at the beginning of the batch and then monitoring it for change
through the local timer interrupt handler.

Adds a raw_phys_apicid array that maps from the mps cpu number
to the apicid - this is needed because the apicids for Summit can be
larger than 32, and thus won't fit into the bitmap. Also adds little wrappers
to map neatly between the two.

Bumps up MAX_APICS for Summit.

Adds detection for summit machines from the MPS tables.
Prints a handy-dandy debug message telling you what kind of twisted
machine the kernel thinks you have.

This one sets up the apic broadcast id (the maximum allowable apic address)
properly for whichever platform. It also abstracts out check_apicid_used,
because that check doesn't work on Summit. Oh, and I bumped up
MAX_IO_APICS, but only for NUMA x86 platforms.

This puts the DFR (desination format register) value into a #define, and
calculates the LDR (logical desitination register) correctly dependant
on platform. Similarly for TARGET_CPUS.

This just adds the config option for summit, and it's Config.help entry,
puts the hooks for the new mach_apic.h subarch file into the right
places, and creates empty shells of the header files.

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

  - fixed returned structure in ctl_read/write

  - fixed compilation of ioctl32 from David S. Miller

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

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

This patch defines a new switch in fs/Config.in -
  CONFIG_NFS_V4:  enables nfsv4 client

Now that all the hooks are in place, this large patch imports all
of the new code for the NFSv4 client.
  nfs4proc.c   - procedure vectors
  nfs4xdr.c    - XDR
  nfs4state.c  - state bookkeeping (very minimal for now)
  nfs4renewd.c - a daemon (implemented as an rpc_task) to keep
                 state from expiring on the server

Note: The RPCSEC_GSS authentication code is not yet included here.
  For the moment we make do with AUTH_UNIX aka. AUTH_SYS.

  Neither is the code to do upcalls to userland in order to do
  uid/gid <-> name mappings. Instead, stubs have been added to
  translate everything to 'nobody:nobody' == '-2:-2'

Define the new NFSv4 data structure for passing user information
from the 'mount' program in nfs4_mount.h.

If CONFIG_NFS_V4 is defined
        Add code to parse the mount structure into the superblock.
        Declare the NFSv4 filesystem to the VFS.

Further cleanups

Separate the static and dynamic filesystem data retrieval calls as per the
NFSv3 spec. This also simplifies things for NFSv4, since many of the
attributes in the fsinfo+fstat combined call are not mandatory to
implement.

Clean up nfs_fill_super().

Separate the parsing of the nfs_mount_data struct and the
initialization + sanity checking of the resulting super_block.
The latter is now performed in the new function nfs_sb_init().

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

It is just a missing include, fixed thus..

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

Updates/corrects Documentation/kernel-parameters.txt file.

Yes it's a GCC optimiser bug.  I'm surprised I didn't see it: I tried
a whole heap of different compilers on that code, and had problems
only on the earlier similar code in raid0_run().

I didn't try redhat's compilers (I run debian) but didn't expect the
behaviour to be that different.

Anyway, please apply this patch (which also fixes the chunk overlap
problems).

there was a bug in fysnc_buffers_list() in 2.4 (fixed in 2.4.19) that
could cause the function to return without having written the current
contents of all the buffers.

Obviously, this could be bad for anybody relying on ordering using
O_SYNC or fsync().  If an I/O was already in flight for a particular bh
at the time of the call to fsync_buffers_list(), ll_rw_block() will not
initiate a new I/O even though the contents may have changed.  It is
therefore necessary to wait before the call.  Here's a patch against
2.5.42 that applies the same fix.

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

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

Rename queue_depth to current_queue_depth so that:

A) we represent the true purpose of the variable
B) we catch anyone using it wrongly

This patch for dpt_i2o fixes TCQ stuff, but doesn't solve DMA mapping
issues (so it still doesn't work, but it's not because of TCQ).

Getting around to fixing DMA mapping API issues is much more work per
driver than the TCQ stuff, so I'm putting it off until later.

dpti.h:
  Update template
dpt_i2o.c:
  Update to new TCQ scheme

aic7xxx_linux_host.h:
  Update host template
aic7xxx_linux.c:
  Update to new TCQ scheme

linit.c:
  Update to use slave_attach

eata.h:
  Update template
eata.c:
  Add slave_attach and new TCQ method

qla1280.h:
  Update device template
qla1280.c:
  Update to use slave_attach for setting queue depth

advansys.h:
  Update host template to include slave_attach
advansys.c:
  Update to use slave_attach instead of select_queue_depths

drivers/scsi/esp.c:
  Fix the build.
scsi.h:
  Add struct list_head items for future (but soon) use
hosts.h:
  Remove select_queue_depths from host struct, add struct list
  stuff for proper list linking of host structs
hosts.c:
  Don't touch select_queue_depths any longer

ips.h:
  Since we now have proper tagged depth setting, make
  the cmd_per_lun value reasonable for untagged devices
  like it is suppossed to be.