In most cases the fix is to add an struct module * member to the operations
vector instead and manipulate the refcounts in the callers context.

For the ALSA cases it was completly superflous (when will people get it that
using an exported symbol will make it's module unloadable?..)

since we now properly reference-count the allocations and thus can't
be freeing the thread structures from underneath the task running on
another CPU.

Move "free_uid(p->user)" and "security_task_free()" to __put_task_struct(),
so that we're guaranteed to always have a user structure pointer.

Please accept this large patch against 2.5.62bk4 to update the
Fusion-MPT driver from 2.03.01.01 to 2.05.00.03.

This version contains full 2.5 kernel support (add, configuration &
removal of disks, new module hooks), performance enhancements, removal
of obsolete code that is 2.4 (or earlier) specific and removal of blank
spaces at end of lines.  Tested overnight on 2.5.62bk3 and ran some
tests against bk4 this morning on both machines.  Things appear to be
fine - have not tested the two failure conditions today.

The driver has been tested on two different platforms:

- P4 4-processor
   o boot controller
   o built into kernel
   o disk add, configure, remove
   o bus, target reset
   o device offline cases
   o cross channel I/O stress tests w/ Fibre and SCSI

 - Athlon uniprocessor, IDE boot
   o Ran a test that repeatedly loads the driver, mounts disks, runs a
     short I/O test (20mins) then unload the driver components in a
     different order

Failures:
   o Unable to boot if boot controller is built as a module - pivot_root
     fails (Happens if IDE or SCSI boot)

   o Problems when formatting large SCSI disks.  On P4 w/ 2.5.60, there
     were no difficulties formatting a 36GB disk.  On Athlon, mkfs.ext2
     would never send a write command to the driver (verified w/ SCSI
     trace and driver debug).  Reads completed successfully.  mkfs.ext2
     would not exit and is shown as a running process.

     Smaller partitions were formatted correctly.  W/ 2.5.62 failures on
     P4 and Athlon.

     Same error with other drivers. 100% reproducible.

which is needed to keep track of process usage counts correctly and
efficiently.

Bill Irwin was talking about hw_interrupt_type.set_affinity and kirq.
When I went looking, I failed to find this initialization.

Here are some nice, easy-to-find, named initializers.

We're calling mmdrop() under spin_lock_irq(&rq->lock).  But mmdrop
calls vfree(), which calls smp_call_function().

It is not legal to call smp_call_function() with irq's off.  Because
another CPU may be running smp_call_function() against _this_ CPU, which
deadlocks.

So the patch arranges for mmdrop() to not be called under
spin_lock_irq(&rq->lock).

We are leaving local interrupts disabled coming out of exit_notify().

But we are about to call wait_task_inactive() which spins, waiting for
another CPU to end a task.  If that CPU has issued smp_call_function() to
this CPU, deadlock.

So the patch enables interrupts again before returning from exit_notify().

Also, exit_notify() returns with preemption disabled, so there is no
need to perform another preempt_disable() in do_exit().

Manifested as X profile appearing as /dev/mem ...

This potentially helps debugging, since otherwise a double fault
would generate a triple fault and then reboot the machine. Now
instead it can print out a note about where the problem happened,
unless all the kernel data structures are truly buggered.

it's a noop in 2.5 now that the behvaiour of implicitly exporting
all symbols is gone.

This patch removes the devfs interface code in mtrr that has been
stubbed out by an ifdef forever.  It's one of the few remaining users
of regular files on devfs so there's some urge for me to get rid of it :)

Use cdevname instead in the serial driver printk cut & pasted into about
a dozend places.

small updates to bring us nearer the lm_Sensors CVS (docs, comments,
missing statics, named initializers).  I still need to review their
actual code changes before submitting those..

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

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

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

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

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

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

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

into kanga.twiddle.net:/home/rth/linux/axp-2.5

the number of metadata buffers. Use the same techniques as the
regular linux buffer cache here.

use more hash buckets for holding xfs metadata, and use the same
hash algorithm as the regular buffer cache.

SGI Modid: 2.5.x-xfs:slinx:139997a