Correct the notification of the cpufreq core of the new frequency being
set in ich's speedstep_set, and print accurate debug info in this function.
(both bugs noted by Valdis Kletnieks, thanks!)

Implement per-CPU memory allocation. Is a bit cleaner than the
previous "once and for all" approach.

Remove the final instances of the now-deprecated
cpufreq_driver_sem. Also, the previous "one-frequency change at one
moment" limitation is gone. If it's needed by the cpufreq driver, it
should be implemented in its cpufreq_driver->{target,setpolicy}
callback.

Use the per-CPU policy->lock to lock the per-CPU "policy" data. Also,
use it to serialize the setting of new frequency policies on each CPU.

Split up cpufreq_governor into a locking and non-locking variant.

Finally implement the two different cpufreq_driver_target callbacks --
the one called while the per-CPU lock is held, the other while
not. While we're at it, clean up cpufreq_governor.

Change the function parameter of cpufreq_cpu_put to struct
cpufreq_policy *.

Make cpufreq_set_policy a bit more readable by storing the CPU's
cpufreq_policy into a pointer.

Make cpufreq_remove_dev a bit more readable. Also, assure that the
GOV_STOP and the cpufreq_driver->exit() call are indeed the last calls
done.

Change the return value of cpufreq_cpu_get from a nondescriptent int
to struct cpufreq_policy*. This will allow for cleaner code - and
later it will be expected that anyone who grabs a reference by calling
cpufreq_cpu_get must return this struct cpufreq_policy* to
cpufreq_cpu_put.

As the per-CPU initialization can fail, we need to deal with this
situation in cpufreq_remove_dev as the sysdev core doesn't (yet) do
this for us. The solution: add a "struct cpufreq_policy
*cpufreq_cpu_data[NR_CPUS]" array, which has a NULL pointer for each
CPU not managed by cpufreq (yet) and a pointer to the per-CPU data
for each CPU managed by cpufreq. It is locked by cpufreq_driver_lock.

First part in a series of patches from Dominik which clean up the locking in
cpufreq. It sort of worked, but is full of races. But to keep it
working at least as well it works now, add a new spinlock
cpufreq_driver_lock which will be what cpufreq_driver_sem intended to
be -- but it can indeed be a spinlock instead of a semaphore. 

This driver adds proper protection for struct cpufreq_driver
*cpufreq_driver which can only go away during cpufreq_unregister --
block this by increasing the driver's module reference count.

From Dominik

Instead of trying to guess what the user intended, fail if the user
requests a setting which is not supported by the longrun driver.

From Dominik

- the path to sysfs changed
- add the cpufreq_driver.owner requirement
- clarify another paragraph

from 2.4-ac - it seems to cause hangs for people.

This tunes the volume control of maestro sound boards: The log
conversion seems not good for PCM and CD.  The volume buttons
incrementing / decrementing the main volume by 10 on 100 is too much,
much better set it to 5.

Memory leak fix: the allocated areas weren't referenced any more once the
original error path returned.

It used to depend on us having a signed type (which in turn is
incorrect for the later division).

alpha, ia64, PA, sparc already export the existing "utimes()" system
call (using 'timeval[2]' instead of 'utimebuf' like the old "utime()").

Make it available on x86 too..

Some minimum critical fixes for x86-64:

- Make it compile for uni processor again
- Tell BIOS about long mode
- Fix 32bit IPC (broken by earlier patch)

Deprecate the numerical sysctl name space. People can use /proc/sys
instead.

The numeric name space was never well maintained and especially
in distribution kernels is not very consistent (everybody has their
own extensions, conflicting with others). It's also a great
source of rejects when merging patches.  The name-based /proc/sys
is a much better interface for this, which people should use instead.

Discussion of this on l-k found no advocate for it, so it seems to not
be very popular anyways.

This patch deprecates numerical name space accesses to make it possible
to remove them in the future. The only exception is kernel.version,
which is used by glibc (this one has to be maintained forever)

 - __div64_32(): remove __attribute_pure__ qualifier from the prototype
   since this function obviously clobbers memory through &(n);

 - do_div(): add a check to ensure (n) is type-compatible with uint64_t;

 - as_update_iohist(): Use sector_div() instead of do_div().
   (Whether the result of the addition should always be stored in 64bits
   regardless of CONFIG_LBD is still being discussed, therefore it's
   unadderessed here);

 - Fix all places where do_div() was being called with a bad divisor argument.

Every failed SG_IO request is replayed second time without data
transfer.  Fix the problem by immediately purging the failed SG_IO
request from the request queue.

It fixes (among others)

	http://marc.theaimsgroup.com/?t=105634805100001&r=1&w=2

where induced kernel crash occurs upon packet replay.

people not to remove the HAVE_KERNEL_CTX_SWITCH support that
the sparc drivers require.

Fix a FIXME.

- include/linux/wrapper.h doesn't exist in 2.5.75 and none of it's content
  is used by the driver;
- s/MINOR/minor/;
- KConfig/Makefile update.