This fixes a *very* obscure bug, which only applies to VIA chipsets that
support UDMA-133 mode, and which is only known to be tickled by one
UDMA-66 hard drive (Maxtor 91360U4) that happens to report 80-wire cable
detection opposite to the ATA standard.

The bug appears in a test to see how the BIOS set up UDMA timing.  This
test is only reached when the drive says 80-wire *and* the chipset says
40-wire (which is only known to happen with this drive).

The timing bits that are checked represent clocks T minus 2, i.e., ((N -
2) * T).  But Vojtech forgot to subtract 2, and applied N = 8 rather
than N = 6 in the test.  Since the test masks the bits at 7, they are
always less than 8, and the test always succeeds, even though the BIOS
set UDMA-33.

This allows SCSI to survive the I/O queueing stress harness with AS.

Jens has signed off on it, and Mark Havercamp confirms it also
eliminates the test induced hangs for him too.

Update support for .incbin in /usr. No longer generate .S files
from within the Makefile.

Also deleted the assignment to LDFLAGS_BLOB for most architectures.
I did not touch ARM and cris:
 arm: Russell told me they did not have a new as that could be used
 cris: Looked like it was used for more than just usr/

Simplify Makefiles in fs/ by utilising the new syntax for composite
objects. The new syntax allowed a couple of ifeq ($(CONFIG_FOO),y)
to be deleted, resulting in more readable Makefiles.

No functional changes introduced.

Patch by Herbert Xu <herbert@gondor.apana.org.au>:

As of 2.5.74, make oldconfig always disables existing optional choices
even if they were selected previously.  For example, if all the EICON
ISDN drivers were selected as modules, then make oldconfig will turn
them off.

Part of the problem is that the choice value itself is computed before
the SYMBOL_NEW flag is turned off.  This patch addresses that particular
problem.

During an erlier update the call to generate the .config dependency
got lost.
Add a missing warning.

correctly select a nested menuconfig entry

Patch by Romain Lievin <roms@tilp.info>
various fixes to handle menuconfig options

A spinlock was recently added to struct skmca_priv so as to remove
save_flags()/cli() but nothing in the current code initializes it.

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

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

This updates generic HDLC from 1.14 to 1.15.

 - fix a kernel panic caused by a recent change to unregister_netdevice()
   (struct net_device * can't be kfreed before rtnl_unlock())
 - adds carrier_* support - hw drivers report DCD status and higher level
   protocols use that info, and do netif_carrier_{on,off}() according to
   DCD and (Cisco and FR) link management status.
 - moves Frame-Relay constants etc. from include/linux/hdlc.h to hdlc_fr.c.
   They are internal FR things and are not needed in the global header.
 - protocol hooks are slighty changed to allow zeroing (memset).
 - removes CONFIG_HDLC_DEBUG_* variables. Users tend to make very wrong
   use of them. Now setting them requires changing .c #define. Anyway they
   are development-only things.
 - misc style corrections etc.

Forward-port of 2.4.21-ac4 patch by Wilfried Weissmann.

ide_get_or_set_dma_base() may be called via ide_setup_pci_device()
when a PCI controller module is loaded.  However, it is defined with __init
and the function is called when its memory is already deallocated.

 misc:      - Consolidate and make consistent, the NODE_BUS_{FMT,ARGS}
	      macros.
	    - Add HPSB_VERBOSE() macro to make VERBOSEDEBUG code a
	      little cleaner.
 ieee1394 : - Replace panic() with BUG and return value in dma routine.
	    - Reset reset_retries on success or loop recover.
	    - Add RECV_FLUSH to flush buffers on > 1 irq interval or in
	      buffer-fill mode.
	    - Probe logic-unit-subdirectories as children of the
	      unit-dir.
	    - Fix compile warning in nodemgr thread.
	    - Replace tightly scheduled timeout loop with a 50ms timer.
	    - Better return and usage of BUG_ON() in dma handler.
	    - Calculate timeout expiry interval at change instead of at
	      every timeout check.
 SBP2     : - Fix logical-unit-directory LUN's.
	    - Remove procfs interface.
	    - Fix usage of initdata after init.
 Host con : - Change irq output to use __irq_itoa() for sparc.
	    - Do more to notify bus when module is unloaded, to clear
	      IRM, etc.
 ohci1394 : Fix endianess of async stream packet headers.

Here's the patch to enable failfast flag in the bio submission code, and
use it for multipath and readahead.

Long overdue, make James the listed maintainer. He _is_ doing the
maintenance, lets put him on the front line :-)

A recent change moved the tcq init too early in the init sequence, the
queue isn't initialized at this point which causes the task file
executions to kill elevator with a NULL EIP (->add_req_fn()).  This
patch fixes it, acked by Bart.

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

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

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

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

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

The module load code adds a reference to the owner field before calling
the modules init function so it isn't necessary for the init function
to do a similar action.

The module load code adds a reference to the owner field before calling
the modules init function so it isn't necessary for the init function
to do a similar action.

The module load code adds a reference to the owner field before calling
the modules init function so it isn't necessary for the init function
to do a similar action.

Module init methods are called with a reference already taken by the
module code so the MOD_*_USE_COUNT can be deleted.

Module init methods are called with a reference already taken by the
module code so the MOD_*_USE_COUNT can be deleted.

The module load code adds a reference to the owner field before calling
the modules init function so it isn't necessary for the init function
to do a similar action.

Also fixed the racy check_region()/request_region() calls.

This patch exports the kstat per-cpu variable, needed for
hisax, which uses kstat_irqs() during card probing to make sure
that irqs actually work. This could possibly replaced by a
private counter in the hisax ISRs, but that's really just
unnecessary overhead, since the core kernel already does the work
anyway.

Instead of faking a hisax-internal card, just register a card
driven by a submodule directly, which fixes the current crashes
when loading such a driver. This makes use of the just introduced
do_register_isdn() and do_init() helpers.

Again, just introduce do_init() and move some code into it.
It'll be needed in the next cset.

This cset just introduces a new function do_register_isdn() which
takes care of what's necessary to register with the i4l link level.

In the unload path, we forgot to free the D-Channel state machines.

FsmNew() is exported, so it sure can't be __init (and since we now
discard __init in modules, it in fact crashes).