Fix two OHCI unlink issues.

   * All EDs now get a 1 msec delay before re-linking, even those
     which were seemingly "clean" unlink cases.  This gets rid of
     some list corruption issues ("bad entry") by getting rid of
     a fast-path carried over from 2.4 usb-ohci.

   * In case of unlink-during-submit, we must giveback() right away.
     This is a reasonably rare case.

There have been recent reports of problems here.  The "bad entry"
showed up with usbtest tests #11 and #12, or "stir4200", and maybe
in other cases.  The unlink-during-submit shows up in usbtest.

The DACA sound driver (early iBook models) doesn't clear the i2c_client
structure. That cause the embedded struct device (and thus kobject) to
contain garbage in the "k_name" field, which kobject_set_name will
later try to kfree...

Also remove references to unused struct data_data.

Here is a small ISDN fix for the current tree.

There is a compiler inlining/optimation problem with strpbrk, if it has
only a one character search string.  This results in a missing strchr
because the compiler internally replace strpbrk with strchr in this
case, but did so after inline handling stage.

A small birdie tells us that in the long run it may not be a good idea
to write the APIC ID register.  It might be read-only in some
hypothetical situation down the road.

into ppc970.osdl.org:/home/torvalds/v2.5/linux

into ppc970.osdl.org:/home/torvalds/v2.5/linux

From: Philippe Elie <phil.el@wanadoo.fr>

When I debugged P4 ht oprofile a few month ago I noticed that but though it
wasn't a problem...  The fix I propose is not completely clean.

With P4 HT we split msr in two subset, one for each logical processor.  The
msrs subset used in op_model_p4.c at save and setup point of view are
distinct (*), it means we must serialize setup and save operation else a
logical processor can save some msr value already setup by the other thread
then when oprofile shutdown we restore wrong msrs values.

Nobody noticed the problem because after restoring the msrs we call
enable_lapic_nmi_watchdog() -> setup_p4_watchdog() wich clear all the msrs
but it's a bit fragile.  If nmi watchdog is not enabled nothing bad occurs
because the LVTPC remains disabled.

(*) this is done in this way because it allows a lot of simplification in
op_model_p4.c, yes it isn't clean but it's not fixable w/o rewriting 75% of
op_model_p4.c and I think the code will be bigger and more complex.

From: Geert Uytterhoeven <Geert.Uytterhoeven@sonycom.com>

I need the following patch to kill a warning (__endian() may be unused)
when cross-compiling m68k kernels on an ia32 box.

From: Geert Uytterhoeven <geert@linux-m68k.org>

The syscalls.h change broke the m68k interrupt management code, since
sys_{request,free}_irq() are not syscalls, but routines to manage system
(CPU) interrupts.

From: Pavel Machek <pavel@ucw.cz>

Without this, if there's not enough swapspace, suspend fails, but leaves
devices suspended, leading to dead machine.

From: Dmitry Torokhov <dtor_core@ameritech.net>

Do not call hotplug until firmware class device is completely instantiated.

From: Dmitry Torokhov <dtor_core@ameritech.net>

We need to pin the firmware loader module until the last reference to the
firmware class device is dropped and the class device is destroyed.

From: Keith Owens <kaos@sgi.com>

New pdflush threads are launched on-demand by pdflush.

It turns out that on some architectures (eg, ia64) a kernel thread inherits
its parent's stack utilisation.  So after the thread-launches-a-thread
cycle has progressed sufficiently far we run out of stack space and crash.

Simple fix: convert pdflush to use kthreads.  All kthreads are parented by
keventd so there is no stack windup as a result of pdflush launching
pdflush.

From: Wim Van Sebroeck <wim@iguana.be>

<wim@iguana.be> (04/02/29 1.1628)
   [WATCHDOG] v2.6.3 pcwd_usb-watchdog

   Add the Berkshire Products USB-PC Watchdog driver

<wim@iguana.be> (04/02/29 1.1629)
   [WATCHDOG] v2.6.3 MODULE_*-patch

   Add MODULE_* info

<wim@iguana.be> (04/02/29 1.1630)
   [WATCHDOG/SPARC64] v2.6.3 Kconfig-WATCHDOG_CP1XXX/WATCHDOG_RIO-patch

   Move WATCHDOG_CP1XXX and WATCHDOG_RIO for SPARC64 architecture
   from arch/sparc64/Kconfig to drivers/char/watchdog/Kconfig
   and made them dependant of WATCHDOG also

From: Andi Kleen <ak@suse.de>

Fix a few issues on x86-64 for 2.6.4rc1.

The 32bit emulation used 4GB/3 for the mmap break. This actually gave
programs less sbrk space than with a standard 32bit kernel. Move the 32bit
mmap break to 0xc5000000.

Uses the correct gcc option to optimize for Prescott now (requires a very
new mainline gcc)

Also merge the i386 sched_clock().  I hope this will fix some interactivity
problems with the scheduler.

- Change initializer to new style (Arnd Bergmann)
- Remove 2 sibling limit in HT support (from i386)
- Always log RIP in MCE records even when not exact
- Move 32bit program task break up to 0xc5000000 by default
- Use -march=prescott for Prescott optimized kernel if possible
- Don't divide by zero with report_lost_ticks on and HPET off
- Merge preempt/smp/debug_pagealloc oops printing from i386
- Add pfn_to_nid()
- Merge more accurate sched_clock from i386
- Remove traces of debugging code in mce.c
- Update defconfig

From: Pavel Machek <pavel@suse.cz>

We curretly have an off-by-factor-of-1000:

cat /proc/cpufreq

          minimum CPU frequency  -  maximum CPU frequency  -  policy
CPU  0         2000 kHz (100 %)  -       2000 kHz (100 %)  -  powersave


I do not have explanation why it is 2MHz - 2MHz.  On my system I removed
bug which prevented my system from being reported as 0.8MHz - 1.8MHz.  (And
modulo cosmetic uglyness, it worked).

This fixes cosmetic uglyness, adds proper copyrights, removes warning
"untested on PREEMPT" (someone tested it, and makes it easier to override
PST by user (eMachine users will need that one).

From: Adrian Bunk <bunk@fs.tum.de>

"swap" is more known than "Support for paging of anonymous memory".  The
patch below adds "(swap)" to the prompt of CONFIG_SWAP.

From: Adrian Bunk <bunk@fs.tum.de>

The patch below is an updated version of the patch that removes some
remaining mentions of "make dep".

From: Bob Doyle <doyle@primenet.com>

BusLogic_UnregisterHostAdapter() is referenced from __exit code, so it cannot
be __init.

The new networking hashtable sizing is all bollixed up.  The logic is wrong
and tcp is setting it to the logarithm of what was intended.  Heaven knows.

Fix it up so that the sizing is the same as it used to be in 2.6, with a
boot-time override.  Which was what was intended.

From: Kurt Garloff <garloff@suse.de>

when calling

alarm(1); alarm(0);

the second alarm can wrongly return 2. This makes an LSB test fail.

It happens due to rounding errors in the timeval to jiffie conversion and
back.  On i386 with HZ =3D=3D 1000, there would not need to be rounding
error IMVHO, but they even occur there.  On HZ=3D1024 platforms, they may
even be unavoidable.

Attached patch fixes the return value of alarm().

From: Benjamin Herrenschmidt <benh@kernel.crashing.org>

This patch implements what we discussed earlier to fix the switch bewteen
KD_GRAPHICS and KD_TEXT.

It has been tested for a few days now and appear to resolve the problem for
affected users.

James: I know you have some objections, I don't fully agree with them, and
I want that in asap now, that bug has been plaguing fbdev since the very
beginning and it's time to get rid of that and my corresponding todolist
entry.  You are welcome to propose a patch on top of this one if you feel
you can make things cleaner.  The approach of adding a parameter to
con_blank() is Linus idea btw :)

I didn't add a separate function as that would have made the butchering of
drivers/char/vt beyond what I want to deal with in 2.6.

From: Marcelo Tosatti <marcelo.tosatti@cyclades.com>

The following patches fixes tty drivers which dont set devfs_name. Not
doing so will cause the tty layer to create "/dev/<NULL>x" entries when
devfs is being used.

I used "drivername/" in isicom and pcxe because the letter used to
identify them are already used by other drivers.

From: James Morris <jmorris@redhat.com>

The patch below fixes a bug in the error handling code of
xprt_create_socket().  If sock_create() fails, we should not try and
release the non existent socket.

This fix is by James Carter <jwcart2@epoch.ncsc.mil>.

From: Dave Jones <davej@redhat.com>

This is probably useless on x86 for eg..  (Everything else in this file is
dependant on some other s390 feature, so only this one shows up).  Too bad
the drivers/s390/block stuff gets source'd at all on !s390.

From: Zwane Mwaikambo <zwane@linuxpower.ca>

Patch to resolve http://bugzilla.kernel.org/show_bug.cgi?id=1644

The bug reporter pointed out a bit of outdated information in the README
file.

Test booted on a 32x NUMAQ with 10,000 disks

From: Neil Brown <neilb@cse.unsw.edu.au>

Fix an off-by-one in the r1buf_pool_alloc() ENOMEM error recovery path.

From: "Art Haas" <ahaas@airmail.net>

Here is a small patch changing the GNU-style initializers to C99
initializers.

From: "Art Haas" <ahaas@airmail.net>

Here's a small patch changing the GNU-style initializers to C99
initializers.

From: "Ronald S. Bultje" <rbultje@ronald.bitfreak.net>

Fix the zoran driver (zr36067) for the fact that we did not handle
bitrate-conversion at all in the zr36050 MJPEG codec (on DC30 cards), with
the result being that at high-resolution, we'd overload the PCI bus and
drop half of our video capture data into /dev/null'ishness.

Also update Ronald's email address in MAINTAINERS.

It appears that gdb, perfmon and perhaps other applications are already
opening /proc/<tid> even though it does not appear in /proc readdir output
and is available under its group leader's directory anyway.

Apparently ascertaining the group leader is hard for gdb and thsi trick
always worked for LinuxThreads apps.

So revert this change.

From: Christophe Saout <christophe@saout.de>

We should copy the bvec array for read requests so that we still
have the unmodified bvec array to decrypt the data afterwards.

(as discussed earlier this day for highmem bounces)

From: Christophe Saout <christophe@saout.de>

These are some dm-crypt cleanups.  Use a #define PFX "crypt: " for all the
places where something gets logged as suggested by Jeff Garzik.  It also
adds a small additional security check and fixed header includes.

From: Christophe Saout <christophe@saout.de>

This patch fixes the bug where in-place encryption was not detected when
the same highmem pages is mapped twice to different virtual addresses.

This adds a parameter to xxx_process to indicate whether this is an
in-place encryption and moves the responsability to the caller using a
helper function scatterwalk.h.

From: Christophe Saout <christophe@saout.de>

I've cleaned up the latest patches and adjusted the header files.

This patch moves the scatterwalk functions from cipher.c to
scatterwalk.c and adds a header file.

From: Christophe Saout <christophe@saout.de>

__end_that_request_first might modify the bv_offset and bv_len if the
segment was partially completed.  The bio-read-bounce-back code should use
the unmodified bv_offset when copying the segment data: