cpufreq doesn't appear to initialise on ARM - cpufreq_register_driver()
fails with -ENODEV.

This appears to be because we register cpus at subsys_initcall() time,
and cpufreq drivers at arch_initcall() time.  We need cpufreq up and
running before any drivers are initialised since their timings depend
on the CPU clock rate.

The following patch appears to fix these problems.

From: Russell King
Signed-off-by: Dave Jones <davej@redhat.com>

Signed-off-by: Bruno Ducrot <ducrot@poupinou.org>
Signed-off-by: Dave Jones <davej@redhat.com>

Remove 'notify' in speedstep_set_state for speedstep-lib
Signed-off-by: Bruno Ducrot <ducrot@poupinou.org>
Signed-off-by: Dave Jones <davej@redhat.com>

Looking around speedstep-smi, there is a lot of call to
speedstep_get_state() which is used as an index for the array
speedstep_freqs.  But speedstep_get_state() will return the value of ebx
after a call to a SMI handler.  It's not safe IMHO.

From: Bruno Ducrot
Signed-off-by: Dave Jones <davej@redhat.com>

From Paul Devriendt.
Signed-off-by: Dave Jones <davej@redhat.com>

Pavel's fix for preempt in the suspend scenario.
The driver protects itself from multiple entry already, and it does not care if
there is a long delay in the middle of a multi-step transition, so there
really is no need for it to disable preempt.
Signed-off-by: Dave Jones <davej@redhat.com>

Ignore two identical low freq entries in the ACPI tables for the powernow-k8
driver.
Signed-off-by: Dominik Brodowski <linux@brodo.de>
Signed-off-by: Dave Jones <davej@redhat.com>
 

Convert gx-suspmod's MODULE_PARM to module_param.
Signed-off-by: Dominik Brodowski <linux@brodo.de>
Signed-off-by: Dave Jones <davej@redhat.com>

                                                                                                                                                                    
Convert elanfreq's MODULE_PARM to module_param, and warn users that the setup parameter
"elanfreq" will go away in favor of "elanfreq.max_freq".
Signed-off-by: Dominik Brodowski <linux@brodo.de>
Signed-off-by: Dave Jones <davej@redhat.com>

From: "Pallipadi, Venkatesh" <venkatesh.pallipadi@intel.com>
                                                                                                                                                                    
I was trying to fix this hotplug force migration issue in
cpu/cpufreq/acpi.c. As you suggested, it made more sense to do this in
cpufreq.c, so that we need not do this in multiple low level drivers.
How does this simple patch look. Beyond this, we may have to change some
documentation, to tell to all low level drivers that they can assume
hotplug lock will be taken, when they are in the target() routine.
Signed-off-by: Dave Jones <davej@redhat.com>

It printed garbage.
Signed-off-by: Dave Jones <davej@redhat.com>

Signed-off-by: Dave Jones <davej@redhat.com>

If we don't do this, and the speed was set to the minimum,
we trigger the lowest==highest assertion on subsequent loads.
Signed-off-by: Dave Jones <davej@redhat.com>

s2io.c: bogus initializer ({{"a"},{"b"}} has too many braces for
two-dimensional array of char - strings are arrays themselves).

de4x5.h: missing __user on a pointer in ioctl structure. 

airo.c: cleaned up and fixed procfs ->read() and ->write()

osst.c, st.c, scsi_debug.c annotated.

selinuxfs.c annotated

	b44.c: ->ioctl() is broken, since it uses &ifr->ifr_data instead of
ifr->ifr_data itself.  Surprise, surprise, copy_from_user() on that address
doesn't do any good...

	baycom_epp.c: does get_user() of the first word of structure, then
immediately does copy_from_user() on the entire thing and completely ignores
the value read by get_user() (it uses the same value in copied structure
instead).  Bogus get_user() call removed.

->ifr_data users annotated.

Some code that is used on iSeries (do_hash_page_DSI in head.S) was
clearing the RI (recoverable interrupt) bit in the MSR when it
shouldn't.  We were getting SLB miss interrupts following that which
were panicking because they appeared to have occurred at a bad place.
This patch fixes the problem.  In fact it isn't necessary for
do_hash_page_DSI to do anything to RI, so the patch changes the code
to not set or clear it.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

net/bridge partially annotated.

There are nasty problems with net/bridge/netfilter/* and they'll need to
be dealt with at some point - it mixes kernel and userland pointers a
lot and while it seems to avoid obvious breakage, it's not a nice code.

econet partially annotated.

It's still badly broken - it mixes userland and kernel chunks in the
same iovec, then does set_fs(KERNEL_FS) and sends that to
sock_sendmsg().  Do we still want to support that protocol family,
anyway?