This is based on the i2c cvs driver, but ported to 2.6 and rewritten
to use the sysfs binary file interface.

K8 can run with a 32bit kernel too.  If that's the case we have to tell
user space about it.  Remove the unneeded ifdef.

In that case the CPU is called "x86-64/hammer" even on 32bit.  This is
not strictly true on i386, but it's the named used by the current
oprofile user space for K8 performance counters and it would complicate
things to use different names between 32bit and 64bit.

 - Fix -funit-at-a-time compilation and enable it when possible
 - Remove -finline-limit as it should not be needed anymore.
 - Update defconfig
 - Fix Makefile bug that caused a recompilation of vsyscall.so on
   every make (thanks to Sam Ravnborg)
 - Add beginning of asm/dwarf2.h to support assembler CFI directives
   (not complete yet)
 - Remove old PDAREF macro from entry.S
 - Remove clever and buggy code in sys_ioperm/set_bitmap and replace it
   by simple and working code
 - Don't make acpi_disabled __initdata.  It is referenced after boot.
 - Fix TLB size reporting in /proc/cpuinfo
 - Cleanup oops printing a bit
 - Add "executive summary" at end of oopses
 - Reenable interrupts on oopses before calling do_exit
 - Remove some unneeded prefetches.  Just two are enough to kickstart
   the hardware prefetcher.
 - Add prefetch workaround (based on code from Richard Brunner)
 - Clean up signal checking in do_page_fault
 - Don't allow modify_ldt to set 64bit codesegments
 - Readd SIGEV_PAD_SIZE (Stephen Rothwell)
 - Add some likelys to uaccess.h (idea from Manfred Spraul)

into samba.org:/home/paulus/kernel/for-linus-ppc

into samba.org:/home/paulus/kernel/for-linus-ppc

Support an arbitrary number of target parameters.  [Alasdair Kergon]

Fix error message when linear targets gets handed more than 2 arguments.
[Alasdair Kergon]

dev_wait was meant to return table status not dev status.  [Alasdair Kergon]

Move retrieve_status up so dev_wait() can use it.

When multiple load ioctls are issued the reference count on older
'new_tables' wasn't being dropped.  [Christophe Saout]

Use the format_dev_t function for target status functions.

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

This implments the LOCKU operation.  These all pass the connectathon
lock test suite against the solaris nfsv4 client.

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

This implements the LOCKT operation.

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

This implements the nfsv4 LOCK operation.

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

This renames some structures and functions that can be used for
byte-range locking as well as for the exiting open share locks, and does
some cleanup.

This implements the idempotent replay cache need for NFSv4 OPEN state.
each state owner (open owner or lock owner) is required to store the
last sequence number mutating operation, and retransmit it when replayed
sequence number is presented for the operation.

I've implemented the cache as a static buffer of size 112 bytes
(NFSD4_REPLAY_ISIZE) which is large enough to hold the OPEN, the largest
of the sequence mutation operations.  This implements the cache for
OPEN, OPEN_CONFIRM, OPEN_DOWNGRADE, and CLOSE.  LOCK and UNLOCK will be
added when byte-range locking is done (soon!).

There is a warning because of calling return with an argument on a void
function.  The root cause is putting a return statement in a macro (evil
side effect).

Change ENCODE_SEQID_OP_TAIL to be a more proper macro, change all
calling functions to return void.

From: Trond Myklebust <trond.myklebust@fys.uio.no>

... see the code in ip_sockglue.c + the macros in socket.h....
AFAICS the control messages have wierd alignment requirements.

Silly typo.

From: Rusty Russell <rusty@rustcorp.com.au>

Also revert to md_fail_request before stopping an array.

The ->stop method can never fail, so there is not point checking it.

into home.osdl.org:/home/torvalds/v2.5/linux

Found by checker (David Yu Chen @Stanford)

into kroah.com:/home/greg/linux/BK/i2c-2.6

> That's what you are going to have to set the name file to in the
> i2c_client structure, much like your patch did.  Then look at the
> different name files in each device directory to see what kind of device
> it is (chip, subclient, etc.)

OK attached patch sets all names to just chip name for chips themselves and
"chipname subclient" when subclient ios registered.

Fix bug when registering the i2c_adap as the structure was not set to 0.

Elimination of this nonsense allows for the assumption
that a task group shares VM. This lets procps run faster.

Remove direct memory accesses and link up device in the proper place in the
sysfs tree.