this fixes the logic kaweth uses to deal with ENOMEM from
usb_submit_urb(). Using the interrupt endpoint's completion
handller is longer an option because automagic resubmission
has been removed. Thus we use workqueues which can be delayed
so that kswapd can do its job and can use GFP_NOIO which is much
likelier to succeed.

 This is for 2.5.72 with the US_PR_DEVICE / US_SC_DEVICE patch in place.
 Tested and working.

this tells the network layer that we can DMA from high memory
if the host controller supports it.

  - enable 64bit DMA on platforms that support it

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

Jeff Garzik points out the initializing the exit completion at
exit time is foolish: we should just initialize it at creation time
live everything else in that structure, and avoid the memory barrier.

into kernel.bkbits.net:/home/davem/sparc-2.5

This makes NBD work with the new linux 2.5 block layer design.
Specifically, it fixes memory corruption that results from module
removal and possible memory corruption from sending or receiving disk
data from the server.

It essentially rolls together the changes from two of the last patchlets
that I emailed: the fix for module removal & the fix for incorrect
struct bio usage.  I believe it's wisest to roll these both together
into this one patch since they both deal with making NBD work better
with the 2.5 linux block layer design and without either of which, it's
possible that NBD will corrupt memory.

Other changes I'd like to see introduced (like in the earlier jumbo
patch) meanwhile are feature enhancements so they can wait.  This patch
also should address all the very helpful concerns that have been raised
so far.  Particularly:

1. that the very first submitted NBD patch was broken down [Andrew]
2. that only 1 spinlock is used for all the NBD request_queue structures
   used [Jens,Al]
3. that kmap() is used in case of highmem pages [Jens]
4. that the allocation of request_queue is dynamic and seperate from
   other allocated objects [Al]

This does the following:
 - remove trailing spaces
 - make loop.h independent by including bio.h, blk.h, spinlock.h
 - replace the lock/unlock functions by module_get/module_put;
   in struct loop this is the change
	-       void (*lock)(struct loop_device *);
	-       void (*unlock)(struct loop_device *);
	+       struct module *owner;
 - replace the integer lo_encrypt_type by the pointer lo_encryption;
   there was a race with loop_unregister_transfer
 - fixed an off-by-one in loop_register_transfer

This is Step 1 of a series of half a dozen or so.

Half of the above is from Jari. Anything that is wrong is mine.

There are some bad compiler issues with section confusion ("const" stuff
likes going into the rodata section).

This corrects the back off count so that write errors will not be
ignored

This patch:
- re-enables cumana partition parsing
- adds eesox partition parsing
- makes the powertec partition parsing fail if sector 0 looks like
  a PC bios partition table

Rather than having a single "acorn_partition" parser for all these
types, we list them explicitly in check.c instead, along with some
explaination about why they're where they are.

From: Rene Scharfe <l.s.r@web.de>

cifs_statfs() is called with a pointer to a struct kstatfs, so let's
propagate this type into the helper function.

Fix a couple of compilation errors in the mouse code.

It was using a non-existent socket[] index and not doing it for all
sockets.

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

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

Also, default to the 10-byte version of mode sense, since a lot
of modern SCSI-like devices don't even support the old version
(we will automatically downgrade to the 6-byte version if the long
version isn't supported).

in VT_RESIZEX handling.

allocation have only the two standard ports by default.

Which was what MELAN really wanted, and others don't
really care.

Pointed out by Roland Dreier

Yet another Asus motherboard hiding features

We currently mask off offline CPUs in both set_cpus_allowed and
sys_sched_setaffinity.  This is firstly redundant, and secondly
erroneous when more CPUs come online (eg. setting affinity to all 1s
should mean all CPUs, including future ones).

We mask with cpu_online_map() in sys_sched_getaffinity *anyway* (which
is another issue, since this is not valid with changing of online
cpus either), so userspace won't see any difference.

This patch makes set_cpus_allowed() return -errno, and check that in
sys_sched_setaffinity.

Matt Fleming points out that returning int from any_online_cpu
where cpu numbers are passed as unsigned ints elsewhere is awkward
and a little dangerous.

Make any_online_cpu() match find_first_bit(), by returning NR_CPUS
when no cpu is found, rather than -1.  This also simplifies the
future case where NR_CPUS > BITS_PER_LONG.

Make the ROMfs copy in the startup code for Motorola/M5272C3 board
conditional on actually using a ROMfs setup.

Allow for hard setting of boot arguments from configuration for
the Motorola Coldfire 5272 CPU targets.

Make the ROMfs copy in the startup code for Motorola/M5249C3 board
conditional on actually using a ROMfs setup.

Allow for hard setting of boot arguments from configuration for
the Motorola Coldfire 5249 CPU targets.

Remove the empty initializer workaround that was added for egcs 1.1.
Only 2.95+ is supported now, so all compilers should support empty
structures.

The if just checked for __GNUC__, which means that 2.95 got
the workaround (and the incompatibility) too even though it didn't need it.

Advantage is that gcc 2.95 and 3.x compiled kernels are now potentially
binary compatible.  Module loading still checks the compiler version,
but it might be removable.

       CPU time while reading PSX pads, and make it a module parameter also,
       for devices which would need the huge value of 60.

       misbehaving 3DPro joysticks, don't trust FreestylePro
       1-bit data packet for data width recognition, invert
       FreestylePro buttons.