struct block_device made the private part of bdevfs inodes; bd_count
is gone, we use ->i_count of inode now; separate hash is also gone and we
are using iget5_locked()/igrab()/iput() instead.

Now that floppy_open() stores bdev in opened_bdevs[drive], we can remove
crap from floppy_read_block_0() and have it use that bdev instead of
messing with bdget_disk()/setting ->bd_disk by hand/bdput().

removed dead code from sparc64 hugetlbpage.c

 - hpfs_unlink() should not try to truncate the victim unless it's a
   regular file; truncate will not help for anything else and it will
   screw the page cache if victim happens to be a block device.
 - network filesystems should *not* invalidate page cache of block
   device node when inode gets invalidated; fixed, added a new helper -
   invalidate_remote_inode(). 
 - nfs setattr syncs the file before sending SETATTR to server; that
   makes a lot of sense for regular files, but not for anything else.
   Fixed.

removed bogus uses of ->i_sb->s_dev in intermezzo (comparizons can and
should simply compare ->i_sb, printks should use ->s_id).

jffs used to put kdev_t values on disk - blind copy of in-core
representation.

Switched to explicit use of u16 (which is what kdev_t currently is),
with appropriate conversion

killed gratitious uses of kdev_t in tpqic02

tty_paranoia_check() switched from kdev_t to struct inode.

removed unused kdev_t stuff, fixed a typo left from the
console->device() conversion.

Fix up interrupt handler type to be irqreturn_t.

Fix up interrupt handler type to be irqreturn_t.

Fix up interrupt handler type to be irqreturn_t.

From: "William A.(Andy) Adamson" <andros@citi.umich.edu>

this patch fixes the share state sequenceid bookeeping.

- increment the sequence id on an open that is confirmed
- increment the sequence id on close

2.4 kernels are very fussy about some values in the superblock, and
2.6 got them wrong.  This fixes it.

raid1 and multipath have not concept of a chunksize, so basing
max_sectors on it is obviously wrong.

Similary 'linear' has a very different concept of chunksize and
max_sectors doesn't apply.

raid5 does have relevant chunk_size concept, but it has code to
effectively handle any chunksize.

So we only need to set max_sectors based on chunk_size in raid0.

If we get a failure trying to allocate a stripe_head for a read-ahead
request (the only time we can get a failure), we skip the rest of the
request and fail the whole bio.

We don't need to explicitly count references as:
 - refcounting already happens for opens of /dev/md?
 -  when an array is active, a daughter module is loaded which
      locks "md" in.
We just need to make sure we clean up properly on unload. (export_array)

Also, xor needs a null module_exit so that it can be unloaded.

The queue init is really far too verbose at boot time.  I don't think
these messages add anything to either the end user experience or debug
ability. 

Acked by Jens

This assigns the queue properly.

Make everything compile and boot again.

 - Update defconfig
 - Some minor cleanup
 - Introduce physid_t for APIC masks (fixes UP kernels)
 - Add bandaid for CardBus bridges and broken BIOS (Vojtech)
 - Add bandaid for unsynchronized TSCs  (Vojtech)
 - Fix ffs(0) return value (fixes XFS) 
 - Fix compilation with software suspend

synchronize_irq() requires an argument when built with CONFIG_SMP.

As discussed earlier. This implements Linus' idea of printing
the addresses in /proc/<pid>/maps as 32bit if possible.

This works around some broken 32bit programs that cannot parse
64bit addresses as generated by x86-64 kernels.

Also simplifies the code slightly.

CONFIG_ACPI_SLEEP doesn't compile without SOFTWARE_SUSPEND.
Make it select it automagically.

This fixes some bugzilla bug whose number I forgot.

export_ide_init_queue() and export_probe_for_drive()

also fix comment in init_irq()

From Benjamin Herrenschmidt <benh@kernel.crashing.org>.

From Mikael Pettersson <mikpe@csd.uu.se>.

From Stephane Ouellette <ouellettes@videotron.ca>.

From Mikael Pettersson <mikpe@csd.uu.se>.

From Duncan Laurie <duncan@sun.com>

This same patch made its way into 2.4 via the -ac tree but hasn't
been put in 2.6 yet.

It fixes some cable detect issues that stem from the fact that the
cable detect pins are also used as address/data lines, so they need
to first be configured as inputs to read valid cable detect state.

One more overlooked area where the proper process ID has to be used:
SysV IPC "pid" values should use the thread group ID, not the per-thread
one.

sound/pcmcia/vx/vx_entry.c:

This patch removes the PCMCIA timer release functionality which is no
longer required. Without this the module does not compile.

Too much cut-and-paste, noticed by Dmitry Torokhov <dtor_core@ameritech.net>