pandora-kernel.git
17 years ago[PATCH] m68knommu: fix 68VZ328/config.c asm
Greg Ungerer [Wed, 28 Jun 2006 06:21:56 +0000 (16:21 +1000)]
[PATCH] m68knommu: fix 68VZ328/config.c asm

Fix 68VZ328/config.c asm to be clean for new gcc versions.

Signed-off-by: Greg Ungerer <gerg@uclinux.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] m68knommu: fix 68360/config.c asm
Greg Ungerer [Wed, 28 Jun 2006 06:21:38 +0000 (16:21 +1000)]
[PATCH] m68knommu: fix 68360/config.c asm

Fix 68360/config.c asm to be clean for new gcc versions.

Signed-off-by: Greg Ungerer <gerg@uclinux.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] m68knommu: include irqnode.h in 68360/ints.c
Greg Ungerer [Wed, 28 Jun 2006 06:09:41 +0000 (16:09 +1000)]
[PATCH] m68knommu: include irqnode.h in 68360/ints.c

The irqnode_t struct has moved to irqnode.h, need to include that
68360 ints.c.

Signed-off-by: Greg Ungerer <gerg@uclinux.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] m68knommu: fix clobber list in uCdimm/uCsimm helper asm
Greg Ungerer [Wed, 28 Jun 2006 06:12:45 +0000 (16:12 +1000)]
[PATCH] m68knommu: fix clobber list in uCdimm/uCsimm helper asm

Fix clobber list in uCsimm/uCdimm boot load helper asm.

Signed-off-by: Greg Ungerer <gerg@uclinux.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] m68knommu: build support for 68328 romvec.S
Greg Ungerer [Wed, 28 Jun 2006 05:59:54 +0000 (15:59 +1000)]
[PATCH] m68knommu: build support for 68328 romvec.S

Conditionaly compile the 68328 romvec code based on the ROM configuration
being enabled.

Signed-off-by: Greg Ungerer <gerg@uclinux.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] m68knommu: remove romvec asm code from ints.c
Greg Ungerer [Wed, 28 Jun 2006 05:56:22 +0000 (15:56 +1000)]
[PATCH] m68knommu: remove romvec asm code from ints.c

Remove the inline 68328 romvec section asm code into its own file.

Signed-off-by: Greg Ungerer <gerg@uclinux.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] m68knommu: fix compilation problems with 68328serial driver
Greg Ungerer [Wed, 28 Jun 2006 05:54:22 +0000 (15:54 +1000)]
[PATCH] m68knommu: fix compilation problems with 68328serial driver

Clean out the 68328serial driver:

. remove use of cli/sti
. fix usage of tty_* calls
. remove unused variables

Signed-off-by: Greg Ungerer <gerg@uclinux.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] m68knommu: create romvec.S for 68328
Greg Ungerer [Wed, 28 Jun 2006 05:58:18 +0000 (15:58 +1000)]
[PATCH] m68knommu: create romvec.S for 68328

Create the 68328 romvec section in its own assembler file. It can be
compiled in when required.

Signed-off-by: Greg Ungerer <gerg@uclinux.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years agoMerge branch 'release' of git://git.kernel.org/pub/scm/linux/kernel/git/aegl/linux-2.6
Linus Torvalds [Wed, 28 Jun 2006 22:00:49 +0000 (15:00 -0700)]
Merge branch 'release' of git://git./linux/kernel/git/aegl/linux-2.6

* 'release' of git://git.kernel.org/pub/scm/linux/kernel/git/aegl/linux-2.6:
  [IA64-SGI] fix prom revision checks in SN kernel
  [IA64] tiger_defconfig s/NR_CPUS=4/NR_CPUS=16/
  [IA64-SGI] - Pass OS logical cpu number to the SN prom (bios)
  [IA64] palinfo.c: s/register_cpu_notifier/register_hotcpu_notifier/

17 years ago[PATCH] x86_64: oprofile build fix
Andrew Morton [Wed, 28 Jun 2006 11:27:04 +0000 (04:27 -0700)]
[PATCH] x86_64: oprofile build fix

WARNING: "unset_nmi_callback" [arch/x86_64/oprofile/oprofile.ko] undefined!
WARNING: "set_nmi_callback" [arch/x86_64/oprofile/oprofile.ko] undefined!

Cc: Andi Kleen <ak@muc.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] Set err_stops_fifo for newer Promise as well
Alan Cox [Wed, 28 Jun 2006 11:27:03 +0000 (04:27 -0700)]
[PATCH] Set err_stops_fifo for newer Promise as well

Signed-off-by: Alan Cox <alan@redhat.com>
Cc: Bartlomiej Zolnierkiewicz <B.Zolnierkiewicz@elka.pw.edu.pl>
Cc: Sergei Shtylyov <sshtylyov@ru.mvista.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] ide: clean up pdc202xx_old so its more readable (done so I could work on...
Alan Cox [Wed, 28 Jun 2006 11:27:03 +0000 (04:27 -0700)]
[PATCH] ide: clean up pdc202xx_old so its more readable (done so I could work on libata ports)

Also sets the new fifo flag so that we don't hang on some errors with this
chipset.

Signed-off-by: Alan Cox <alan@redhat.com>
Cc: Bartlomiej Zolnierkiewicz <B.Zolnierkiewicz@elka.pw.edu.pl>
Cc: Sergei Shtylyov <sshtylyov@ru.mvista.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] ide: housekeeping on IDE drivers
Alan Cox [Wed, 28 Jun 2006 11:27:02 +0000 (04:27 -0700)]
[PATCH] ide: housekeeping on IDE drivers

Move auto arrays to static (const).  Clean up using PCI_DEVICE in places,
remove unreachable junk and dead code.

Fix the serverworks cable detect logic (if ordering is wrong).  Backport
from libata.  Plenty of scope for more cleanup left.

Signed-off-by: Alan Cox <alan@redhat.com>
Cc: Bartlomiej Zolnierkiewicz <B.Zolnierkiewicz@elka.pw.edu.pl>
Cc: Sergei Shtylyov <sshtylyov@ru.mvista.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] ide: fix error handling for drives which clear the FIFO on error
Alan Cox [Wed, 28 Jun 2006 11:27:01 +0000 (04:27 -0700)]
[PATCH] ide: fix error handling for drives which clear the FIFO on error

If the controller FIFO cleared automatically on error we must not try
and drain it as this will hang some chips.

Based in concept on a broken patch from -mm some while back

Signed-off-by: Alan Cox <alan@redhat.com>
Cc: Sergei Shtylyov <sshtylyov@ru.mvista.com>
Cc: Bartlomiej Zolnierkiewicz <B.Zolnierkiewicz@elka.pw.edu.pl>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] SC1200 debug printk
Alan Cox [Wed, 28 Jun 2006 11:27:00 +0000 (04:27 -0700)]
[PATCH] SC1200 debug printk

Kill a pair of long escaped debug printk calls

Signed-off-by: Alan Cox <alan@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] ide: clean up siimage
Alan Cox [Wed, 28 Jun 2006 11:26:59 +0000 (04:26 -0700)]
[PATCH] ide: clean up siimage

Remove all the ifdef preparation for enhanced features that never occcurred
and is only in libata.  For the SATA chips (but not yet PATA ones) politely
suggest to the user that libata may offer more features.

Signed-off-by: Alan Cox <alan@redhat.com>
Cc: Sergei Shtylylov <sshtylyov@ru.mvista.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] Old IDE, fix SATA detection for cabling
Alan Cox [Wed, 28 Jun 2006 11:26:58 +0000 (04:26 -0700)]
[PATCH] Old IDE, fix SATA detection for cabling

This is based on the proposed patches flying around but also checks that
the device in question is new enough to have word 93 rather thanb blindly
assuming word 93 == 0 means SATA (see ATA-5, ATA-7)

Signed-off-by: Alan Cox <alan@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] oss/via: make bitfield unsigned
Randy Dunlap [Wed, 28 Jun 2006 11:26:58 +0000 (04:26 -0700)]
[PATCH] oss/via: make bitfield unsigned

Make a 1-bit field unsigned like all of the others near it.
sound/oss/via82cxxx_audio.c:311:21: error: dubious one-bit signed bitfield

Signed-off-by: Randy Dunlap <rdunlap@xenotime.net>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] ac97_codec: make bitfield unsigned
Randy Dunlap [Wed, 28 Jun 2006 11:26:57 +0000 (04:26 -0700)]
[PATCH] ac97_codec: make bitfield unsigned

Make a 1-bit bitfield unsigned (no space for sign bit).
Removes 24 sparse warnings from this one file:
include/linux/ac97_codec.h:262:13: error: dubious one-bit signed bitfield

Signed-off-by: Randy Dunlap <rdunlap@xenotime.net>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] sound: fix cs4232 section mismatch
Randy Dunlap [Wed, 28 Jun 2006 11:26:56 +0000 (04:26 -0700)]
[PATCH] sound: fix cs4232 section mismatch

Fix section mismatch: adds __init to probe function,
frees some init memory, not critical.
WARNING: sound/oss/cs4232.o - Section mismatch: reference to .init.text: from .text after 'cs4232_pnp_probe' (at offset 0x152)

Signed-off-by: Randy Dunlap <rdunlap@xenotime.net>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] powermac backlight fixes
Michael Hanselmann [Wed, 28 Jun 2006 11:26:55 +0000 (04:26 -0700)]
[PATCH] powermac backlight fixes

Fix a erroneous calculation of the legacy brightness values as reported by
Paul Collins.  Additionally, it moves the calculation of the negative value
in the radeonfb driver after the value check.

Signed-off-by: Michael Hanselmann <linux-kernel@hansmi.ch>
Acked-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Acked-by: Paul Collins <paul@briny.ondioline.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] IPMI: watchdog handle panic properly
Corey Minyard [Wed, 28 Jun 2006 11:26:55 +0000 (04:26 -0700)]
[PATCH] IPMI: watchdog handle panic properly

Modify the watchdog timeout in IPMI to only do things at panic/reboot time if
the watchdog timer was already running.  Some BIOSes do not disable the
watchdog timer at startup, and this led to a reboot a while later if the new
OS running didn't start monitoring the watchdog, even if the watchdog was not
running before.

Signed-off-by: Corey Minyard <minyard@acm.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] IPMI: remove high res timer code
Corey Minyard [Wed, 28 Jun 2006 11:26:54 +0000 (04:26 -0700)]
[PATCH] IPMI: remove high res timer code

There was some old high-res-timer code in the IPMI driver that is dead.
Remove it.

Signed-off-by: Corey Minyard <minyard@acm.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] IPMI: tidy msghandler timer
Corey Minyard [Wed, 28 Jun 2006 11:26:53 +0000 (04:26 -0700)]
[PATCH] IPMI: tidy msghandler timer

Tidy up the timer usage in the IPMI driver.

Signed-off-by: Corey Minyard <minyard@acm.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] istallion: cleanups
Alan Cox [Wed, 28 Jun 2006 11:26:52 +0000 (04:26 -0700)]
[PATCH] istallion: cleanups

Turned out to be rather a monster

Signed-off-by: Alan Cox <alan@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] stallion: fix recent SMP locking cleanup
Alan Cox [Wed, 28 Jun 2006 11:26:51 +0000 (04:26 -0700)]
[PATCH] stallion: fix recent SMP locking cleanup

Works better on SMP if...

Signed-off-by: Alan Cox <alan@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] v9fs: fix fid check in v9fs_create
Latchesar Ionkov [Wed, 28 Jun 2006 11:26:51 +0000 (04:26 -0700)]
[PATCH] v9fs: fix fid check in v9fs_create

Fix an incorrect check whether a fid was allocated in v9fs_create and if it
should be freed on error.

Signed-off-by: Latchesar Ionkov <lucho@ionkov.net>
Cc: Eric Van Hensbergen <ericvh@ericvh.myip.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] v9fs: return the correct error when interrupted by signal
Latchesar Ionkov [Wed, 28 Jun 2006 11:26:50 +0000 (04:26 -0700)]
[PATCH] v9fs: return the correct error when interrupted by signal

If a signal interrupts the user process, v9fs sends a flush request to the
file server and waits for its response.  It error code is incorrectly set
to the error code of the flush message instead of ERESTARTSYS.  The patch
sets the error code to the correct value.

Signed-off-by: Latchesar Ionkov <lucho@ionkov.net>
Cc: Eric Van Hensbergen <ericvh@ericvh.myip.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] remove active field from tty buffer structure
Paul Fulghum [Wed, 28 Jun 2006 11:26:49 +0000 (04:26 -0700)]
[PATCH] remove active field from tty buffer structure

Remove 'active' field from tty buffer structure.  This was added in 2.6.16
as part of a patch to make the new tty buffering SMP safe.  This field is
unnecessary with the more intelligently written flush_to_ldisc that adds
receive_room handling.

Removing this field reverts to simpler logic where the tail buffer is
always the 'active' buffer, which should not be freed by flush_to_ldisc.
(active == buffer being filled with new data)

The result is simpler, smaller, and faster tty buffer code.

Signed-off-by: Paul Fulghum <paulkf@microgate.com>
Cc: Alan Cox <alan@lxorguk.ukuu.org.uk>
Cc: Theodore Ts'o <tytso@mit.edu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] add receive_room flow control to flush_to_ldisc
Paul Fulghum [Wed, 28 Jun 2006 11:26:48 +0000 (04:26 -0700)]
[PATCH] add receive_room flow control to flush_to_ldisc

Flush data serially to line discipline in blocks no larger than
tty->receive_room to avoid losing data if line discipline is busy (such as
N_TTY operating at high speed on heavily loaded system) or does not accept
data in large blocks (such as N_MOUSE).

Signed-off-by: Paul Fulghum <paulkf@microgate.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] remove TTY_DONT_FLIP
Paul Fulghum [Wed, 28 Jun 2006 11:26:47 +0000 (04:26 -0700)]
[PATCH] remove TTY_DONT_FLIP

Remove TTY_DONT_FLIP tty flag.  This flag was introduced in 2.1.X kernels
to prevent the N_TTY line discipline functions read_chan() and
n_tty_receive_buf() from running at the same time.  2.2.15 introduced
tty->read_lock to protect access to the N_TTY read buffer, which is the
only state requiring protection between these two functions.

The current TTY_DONT_FLIP implementation is broken for SMP, and is not
universally honored by drivers that send data directly to the line
discipline receive_buf function.

Because TTY_DONT_FLIP is not necessary, is broken in implementation, and is
not universally honored, it is removed.

Signed-off-by: Paul Fulghum <paulkf@microgate.com>
Cc: Alan Cox <alan@lxorguk.ukuu.org.uk>
Cc: Theodore Ts'o <tytso@mit.edu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] RTC: add rtc-rs5c348 driver
Atsushi Nemoto [Wed, 28 Jun 2006 11:26:47 +0000 (04:26 -0700)]
[PATCH] RTC: add rtc-rs5c348 driver

Add an SPI driver for the Ricoh RS5C348 RTC chip.

Signed-off-by: Atsushi Nemoto <anemo@mba.ocn.ne.jp>
Acked-by: Alessandro Zummo <a.zummo@towertech.it>
Cc: David Brownell <david-b@pacbell.net>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] load_module() cleanup
Andrew Morton [Wed, 28 Jun 2006 11:26:46 +0000 (04:26 -0700)]
[PATCH] load_module() cleanup

Undo bizarre declaration in load_module().

Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] Add EXPORT_UNUSED_SYMBOL and EXPORT_UNUSED_SYMBOL_GPL
Arjan van de Ven [Wed, 28 Jun 2006 11:26:45 +0000 (04:26 -0700)]
[PATCH] Add EXPORT_UNUSED_SYMBOL and EXPORT_UNUSED_SYMBOL_GPL

Temporarily add EXPORT_UNUSED_SYMBOL and EXPORT_UNUSED_SYMBOL_GPL.  These
will be used as a transition measure for symbols that aren't used in the
kernel and are on the way out.  When a module uses such a symbol, a warning
is printk'd at modprobe time.

The main reason for removing unused exports is size: eacho export takes
roughly between 100 and 150 bytes of kernel space in the binary.  This
patch gives users the option to immediately get this size gain via a config
option.

Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] mark address_space_operations const
Christoph Hellwig [Wed, 28 Jun 2006 11:26:44 +0000 (04:26 -0700)]
[PATCH] mark address_space_operations const

Same as with already do with the file operations: keep them in .rodata and
prevents people from doing runtime patching.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Cc: Steven French <sfrench@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] x86: do_IRQ(): check irq number
Andrew Morton [Wed, 28 Jun 2006 11:26:43 +0000 (04:26 -0700)]
[PATCH] x86: do_IRQ(): check irq number

We recently changed x86 to handle more than 256 IRQs.  Add a check in do_IRQ()
just to make sure that nothing went wrong with that implementation.

[chrisw@sous-sol.org: do x86_64 too]
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Andi Kleen <ak@muc.de>
Cc: Chris Wright <chrisw@sous-sol.org>
Cc: "Protasevich, Natalie" <Natalie.Protasevich@UNISYS.com>
Cc: <Christian.Limpach@cl.cam.ac.uk>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] small fix for not releasing the mmap semaphore in i386/arch_setup_additional_...
pageexec@freemail.hu [Wed, 28 Jun 2006 18:44:16 +0000 (20:44 +0200)]
[PATCH] small fix for not releasing the mmap semaphore in i386/arch_setup_additional_pages

the VDSO randomization code on i386 fails to release the mmap semaphore
if insert_vm_struct() fails.

[ Made the conditional unlikely. -- Linus ]

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[IA64-SGI] fix prom revision checks in SN kernel
Aaron Young [Wed, 28 Jun 2006 15:34:55 +0000 (08:34 -0700)]
[IA64-SGI] fix prom revision checks in SN kernel

The following patch fixes two spots in the SN kernel
that check a fixed prom revision number to determine prom
feature support. These checks are only valid on shub1 systems.
They are invalid on shub2 systems which have a different prom
with different revision numbers.

Signed-off-by: Aaron Young <ayoung@sgi.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
17 years ago[IA64] tiger_defconfig s/NR_CPUS=4/NR_CPUS=16/
Tony Luck [Tue, 27 Jun 2006 22:32:24 +0000 (15:32 -0700)]
[IA64] tiger_defconfig s/NR_CPUS=4/NR_CPUS=16/

Montecito is coming with dual core and threading, so this
four socket box can now have sixteen logical cpus.

Signed-off-by: Tony Luck <tony.luck@intel.com>
17 years ago[IA64-SGI] - Pass OS logical cpu number to the SN prom (bios)
Jack Steiner [Fri, 28 Apr 2006 01:01:05 +0000 (20:01 -0500)]
[IA64-SGI] - Pass OS logical cpu number to the SN prom (bios)

Pass the OS logical cpu number to the PROM. This allows PROM
to log the OS logical cpu number in error records viewed thru POD.

Signed-off-by: Jack Steiner <steiner@sgi.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
17 years ago[IA64] palinfo.c: s/register_cpu_notifier/register_hotcpu_notifier/
Tony Luck [Wed, 28 Jun 2006 16:55:13 +0000 (09:55 -0700)]
[IA64] palinfo.c: s/register_cpu_notifier/register_hotcpu_notifier/

Chandra Seetharaman missed one place in commit:
 65edc68c345cbe21d0b0375c3452a3ed5e322868
[but it only shows up when building the ski simulator configuration
 of ia64, so thats understandable]

Signed-off-by: Tony Luck <tony.luck@intel.com>
17 years agoMerge branch 'upstream-linus' of master.kernel.org:/pub/scm/linux/kernel/git/jgarzik...
Linus Torvalds [Wed, 28 Jun 2006 02:15:50 +0000 (19:15 -0700)]
Merge branch 'upstream-linus' of /linux/kernel/git/jgarzik/netdev-2.6

* 'upstream-linus' of master.kernel.org:/pub/scm/linux/kernel/git/jgarzik/netdev-2.6:
  [netdrvr] Remove long-unused bits from Becker template drivers
  [netdrvr] natsemi: minor cleanups
  [netdrvr] natsemi: Separate out media initialization code
  [PATCH] WAN: update info page for a bunch of my drivers
  [PATCH] drivers/net/hamradio/dmascc.c: fix section mismatch
  [PATCH] Fix phy id for LXT971A/LXT972A
  [PATCH] DM9000 - minor code cleanups
  [PATCH] DM9000 - do no re-init spin lock
  [PATCH] DM9000 - check for MAC left in by bootloader
  [PATCH] DM9000 - better checks for platform resources

17 years agoMerge git://git.infradead.org/mtd-2.6
Linus Torvalds [Wed, 28 Jun 2006 02:13:56 +0000 (19:13 -0700)]
Merge git://git.infradead.org/mtd-2.6

* git://git.infradead.org/mtd-2.6:
  [MTD] NAND: Select chip before checking write protect status
  [MTD] CORE mtdchar.c: fix off-by-one error in lseek()
  [MTD] NAND: Fix typo in mtd/nand/ts7250.c
  [JFFS2][XATTR] coexistence between xattr and write buffering support.
  [JFFS2][XATTR] Fix wrong copyright
  [JFFS2][XATTR] Re-define xd->refcnt as atomic_t
  [JFFS2][XATTR] Fix memory leak with jffs2_xattr_ref
  [JFFS2][XATTR] rid unnecessary writing of delete marker.
  [JFFS2][XATTR] Fix ACL bug when updating null xattr by null ACL.
  [JFFS2][XATTR] using 'delete marker' for xdatum/xref deletion
  [MTD] Fix off-by-one error in physmap.c
  [MTD] Remove unused 'nr_banks' variable from ixp2000 map driver
  [MTD NAND] s3c2412 support in s3c2410.c
  [MTD] Initialize 'writesize'
  [MTD] NAND: ndfc fix address offset thinko
  [MTD] NAND: S3C2410 convert prinks to dev_*()s
  [MTD] NAND: Missing fixups

17 years agoMerge git://oss.sgi.com:8090/nathans/xfs-2.6
Linus Torvalds [Wed, 28 Jun 2006 02:09:16 +0000 (19:09 -0700)]
Merge git://oss.sgi.com:8090/nathans/xfs-2.6

* git://oss.sgi.com:8090/nathans/xfs-2.6:
  [XFS] Fixup whitespace damage in log_write, remove final warning.
  [XFS] Rework code snippets slightly to remove remaining recent-gcc
  [XFS] Fix realtime subvolume expansion, a porting bug b0rked it.  Coverity
  [XFS] Remove a race condition where a linked inode could BUG_ON in
  [XFS] Remove redundant directory checks from inode link operation.
  [XFS] Remove a couple of no-longer-used macros.
  [XFS] Reduce size of xfs_trans_t structure. * remove ->t_forw, ->t_back --
  [XFS] remove unused behaviour lock - shrink XFS vnode as a side effect.
  [XFS] * There is trivial "inode => vnode => inode" conversion, but only
  [XFS] link(2) on directory is banned in VFS.

17 years agoMerge branch 'upstream-linus' of master.kernel.org:/pub/scm/linux/kernel/git/jgarzik...
Linus Torvalds [Wed, 28 Jun 2006 02:07:21 +0000 (19:07 -0700)]
Merge branch 'upstream-linus' of /linux/kernel/git/jgarzik/libata-dev

* 'upstream-linus' of master.kernel.org:/pub/scm/linux/kernel/git/jgarzik/libata-dev:
  [PATCH] ata_piix: add ICH6/7/8 to Kconfig
  [PATCH] sata_sil: disable hotplug interrupts on two ATI IXPs
  [PATCH] libata: cosmetic updates
  [PATCH] ata: add some NVIDIA chipset IDs
  [PATCH] libata reduce timeouts
  [PATCH] libata: implement ata_port_max_devices()
  [PATCH] libata: make two functions global
  [PATCH] libata: update ata_do_simple_cmd()
  [PATCH] libata: move ata_do_simple_cmd() below ata_exec_internal()
  [PATCH] libata: clear EH action on device detach
  [PATCH] libata: implement and use ata_deh_dev_action()
  [PATCH] libata: move ata_eh_clear_action() upward
  [PATCH] libata.h needs scatterlist.h
  [libata] sata_vsc: partially revert a PCI ID-related commit
  [libata] Bump versions

17 years agoProperly delete sound/ppc/toonie.c
Linus Torvalds [Wed, 28 Jun 2006 01:47:18 +0000 (18:47 -0700)]
Properly delete sound/ppc/toonie.c

The previous "delete" had actually just truncated it to a zero
size, something that can easily happen if you just apply a patch.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] i2c-i801.c: don't pci_disable_device() after it was just enabled
Daniel Ritz [Tue, 27 Jun 2006 16:40:54 +0000 (18:40 +0200)]
[PATCH] i2c-i801.c: don't pci_disable_device() after it was just enabled

Commit 02dd7ae2892e5ceff111d032769c78d3377df970 ("[PATCH] i2c-i801:
Merge setup function") has a missing return 0 in the _probe() function.
This means the error path is always executed and pci_disable_device() is
called even when the device just got successfully enabled.

Having the SMBus device disabled makes some systems (eg.
Fujitsu-Siemens Lifebook E8010) hang hard during power-off.

Intead of reverting the whole commit this patch fixes it up:
- don't ever call pci_disable_device(), also not in the _remove() function
  to avoid hangs
- fix missing pci_release_region() in error path

Signed-off-by: Daniel Ritz <daniel.ritz@gmx.ch>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/sfrench/cifs-2.6
Linus Torvalds [Wed, 28 Jun 2006 01:31:57 +0000 (18:31 -0700)]
Merge git://git./linux/kernel/git/sfrench/cifs-2.6

* git://git.kernel.org/pub/scm/linux/kernel/git/sfrench/cifs-2.6: (25 commits)
  [CIFS] Fix authentication choice so we do not force NTLMv2 unless the
  [CIFS] Fix alignment of unicode strings in previous patch
  [CIFS] Fix allocation of buffers for new session setup routine to allow
  [CIFS] Remove calls to to take f_owner.lock
  [CIFS] remove some redundant null pointer checks
  [CIFS] Fix compile warning when CONFIG_CIFS_EXPERIMENTAL is off
  [CIFS] Enable sec flags on mount for cifs (part one)
  [CIFS] Fix suspend/resume problem which causes EIO on subsequent access to
  [CIFS] fix minor compile warning when config_cifs_weak_security is off
  [CIFS] NTLMv2 support part 5
  [CIFS] Add support for readdir to legacy servers
  [CIFS] NTLMv2 support part 4
  [CIFS] NTLMv2 support part 3
  [CIFS] NTLMv2 support part 2
  [CIFS] Fix mask so can set new cifs security flags properly
  CIFS] Support for older servers which require plaintext passwords - part 2
  [CIFS] Support for older servers which require plaintext passwords
  [CIFS] Fix mapping of old SMB return code Invalid Net Name so it is
  [CIFS] Missing brace
  [CIFS] Do not overwrite aops
  ...

17 years ago[PATCH] m68knommu: use Kconfig RAM config options in 68328 startup code
Greg Ungerer [Tue, 27 Jun 2006 03:27:02 +0000 (13:27 +1000)]
[PATCH] m68knommu: use Kconfig RAM config options in 68328 startup code

Switch to using the new RAM Kconfig settings, instead of linker defined
regions in ROM specific 68328 startup code.

Signed-off-by: Greg Ungerer <gerg@uclinux.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] m68knommu: use Kconfig RAM config options in 68360 ROM startup code
Greg Ungerer [Tue, 27 Jun 2006 03:27:13 +0000 (13:27 +1000)]
[PATCH] m68knommu: use Kconfig RAM config options in 68360 ROM startup code

Switch to using the new RAM Kconfig settings, instead of linker defined
regions in ROM specific 68360 startup code.

Signed-off-by: Greg Ungerer <gerg@uclinux.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] m68knommu: use Kconfig RAM config options in 68360 RAM startup code
Greg Ungerer [Tue, 27 Jun 2006 03:27:08 +0000 (13:27 +1000)]
[PATCH] m68knommu: use Kconfig RAM config options in 68360 RAM startup code

Switch to using the new RAM Kconfig settings, instead of linker defined
regions in RAM specific 68360 startup code.

Signed-off-by: Greg Ungerer <gerg@uclinux.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] m68knommu: remove NO_FORMAT_VECi from ptrace.h header
Greg Ungerer [Tue, 27 Jun 2006 03:22:07 +0000 (13:22 +1000)]
[PATCH] m68knommu: remove NO_FORMAT_VECi from ptrace.h header

Remove NO_FORMAT_VEC conditional check. It is not used or defined anywhere.

Signed-off-by: Greg Ungerer <gerg@uclinux.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] m68knommu: FEC driver event/irq fixes
Greg Ungerer [Tue, 27 Jun 2006 03:19:33 +0000 (13:19 +1000)]
[PATCH] m68knommu: FEC driver event/irq fixes

Collection of fixes for the ColdFire FEC ethernet driver:

. reworked event setting so that it occurs after the MII setup.
  roucaries bastien <roucaries.bastien@gmail.com>
. Do not read cbd_sc in memory for each bit we test. Once per buffer is enough.
. Overrun errors must increase `rx_fifo_errors', not `rx_crc_errors'
. No need for a special value to activate rx or tx.  Only write access matters.
. Simplify parameter of eth_copy_and_sum : `data' has already the right value.
. Some spelling fixes.

Signed-off-by: Philippe De Muyter <phdm@macqel.be>
Signed-off-by: Greg Ungerer <gerg@uclinux.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] m68knommu: FEC driver set different priority/level on each IRQ
Willson Callan [Tue, 27 Jun 2006 03:13:44 +0000 (13:13 +1000)]
[PATCH] m68knommu: FEC driver set different priority/level on each IRQ

Set different irq priority levels for each IRQ requested.
According to the Freescale ColdFire documentation each separate IRQ
must have its own unique priority/level combination.

Signed-off-by: Greg Ungerer <gerg@uclinux.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] m68knommu: FEC driver support for the ColdFire 523x CPU family
Matt Waddel [Tue, 27 Jun 2006 03:10:56 +0000 (13:10 +1000)]
[PATCH] m68knommu: FEC driver support for the ColdFire 523x CPU family

Add support for the FEC module in the ColdFire 532x CPU family.

Signed-off-by: Greg Ungerer <gerg@uclinux.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] m68knommu: avoid fec driver hang when link disappears
Philippe De Muyter [Tue, 27 Jun 2006 03:05:33 +0000 (13:05 +1000)]
[PATCH] m68knommu: avoid fec driver hang when link disappears

Avoid requesting a `Graceful Transmit Stop' when link has disappeared,
because that request cannot complete without link.

Signed-off-by: Greg Ungerer <gerg@uclinux.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] m68knommu: update m68knommu defconfnig
Greg Ungerer [Tue, 27 Jun 2006 02:59:35 +0000 (12:59 +1000)]
[PATCH] m68knommu: update m68knommu defconfnig

Updated defconfig for m68knommu arch. Includes recent changes to the clock
and RAM configuration options.

Signed-off-by: Greg Ungerer <gerg@uclinux.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] m68knommu: build support for the Freescale 532x CPU family
Matt Waddel [Tue, 27 Jun 2006 02:50:53 +0000 (12:50 +1000)]
[PATCH] m68knommu: build support for the Freescale 532x CPU family

Add build support for the M523x ColdFire CPU family.

Signed-off-by: Greg Ungerer <gerg@uclinux.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] m68knommu: build support for the Avnet/5282 board
Daniel Alomar [Tue, 27 Jun 2006 02:49:11 +0000 (12:49 +1000)]
[PATCH] m68knommu: build support for the Avnet/5282 board

Add support for the Avnet/5282 board.

Signed-off-by: Greg Ungerer <gerg@uclinux.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] m68knommu: remove useless compiler args
Philipe De Muyter [Tue, 27 Jun 2006 02:47:00 +0000 (12:47 +1000)]
[PATCH] m68knommu: remove useless compiler args

Here is a small patch that made my kernel .text segment shrink by 8k IIRC
on my 5272-based board, by removing `-Wa,-S' from CFLAGS.

The `-Wa,-S' option prevents `gas' from using short forms of jsr.
Without it, `gas' replaces `jsr xxx.l' (6 bytes) by `jsr xxx@pc'
(4 bytes) when possible.  On 5272, both forms are equally fast.

The `-Wa,-m5307' option is useless, because gcc already gives it
to `gas' from the `-m5307' option.

Signed-off-by: Greg Ungerer <gerg@uclinux.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] voyager: add cpu_present_map
James Bottomley [Tue, 27 Jun 2006 02:33:09 +0000 (21:33 -0500)]
[PATCH] voyager: add cpu_present_map

Voyager stopped booting some time in the 2.6.16-2.6.17 timeframe;
the reason was that it doesn't have a cpu_present_map, so add
one.

Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years agoMerge master.kernel.org:/pub/scm/linux/kernel/git/mchehab/v4l-dvb
Linus Torvalds [Wed, 28 Jun 2006 01:22:13 +0000 (18:22 -0700)]
Merge /pub/scm/linux/kernel/git/mchehab/v4l-dvb

* master.kernel.org:/pub/scm/linux/kernel/git/mchehab/v4l-dvb: (26 commits)
  V4L/DVB (4263): Fix warning when compiling on 64 bit machines
  V4L/DVB (4261): Included required header for in-kernel compilation
  V4L/DVB (4260): Stradis.c: make 2 functions static
  V4L/DVB (4259): Pass an explicit log prefix to cx2341x_log_status
  V4L/DVB (4257): Fix 64-bit compile warnings.
  V4L/DVB (4255): Tda9887 default TOP value is 0x10
  V4L/DVB (4254): Remove obsoleted tuner_debug option.
  V4L/DVB (4253): IVTV VBI format description too long.
  V4L/DVB (4252): Remove duplicate 'tda9887' in info messages.
  V4L/DVB (4245): Reduce the amount of pvrusb2-sourced noise going into the system log
  V4L/DVB (4244): Implement use of cx2341x module in pvrusb2 driver
  V4L/DVB (4243): Exploit new V4L control features in pvrusb2
  V4L/DVB (4242): Don't suspend encoder when changing its attributes (in pvrusb2)
  V4L/DVB (4241): Fix faulty encoder error recovery in pvrusb2
  V4L/DVB (4240): Various V4L control enhancements in pvrusb2
  V4L/DVB (4239): Handle boolean controls in pvrusb2
  V4L/DVB (4238): Make sure flags field is initialized when quering a control in pvrusb2
  V4L/DVB (4237): Move LOG_STATUS bracketing to a different part of the pvrusb2 driver
  V4L/DVB (4236): Rearrange things in pvrusb2 driver in preparation for using cx2341x module
  V4L/DVB (4235): Increase the maximum number of controls that pvrusb2-sysfs.c can handle.
  ...

17 years ago[XFS] Fixup whitespace damage in log_write, remove final warning.
Nathan Scott [Wed, 28 Jun 2006 01:17:28 +0000 (11:17 +1000)]
[XFS] Fixup whitespace damage in log_write, remove final warning.

SGI-PV: 904196
SGI-Modid: xfs-linux-melb:xfs-kern:26366a

Signed-off-by: Nathan Scott <nathans@sgi.com>
17 years ago[PATCH] i4l fix DLE masking in isdn_tty_try_read
Karsten Keil [Tue, 27 Jun 2006 11:01:27 +0000 (13:01 +0200)]
[PATCH] i4l fix DLE masking in isdn_tty_try_read

DLE masking was non-functional since the new tty handling.

Found by Peter Evertz <leo2@pec.homeip.net>

Signed-off-by: Karsten Keil <kkeil@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] do_IRQ() warning fix
Andrew Morton [Tue, 27 Jun 2006 09:55:09 +0000 (02:55 -0700)]
[PATCH] do_IRQ() warning fix

arch/i386/kernel/irq.c: In function 'do_IRQ':
arch/i386/kernel/irq.c:104: warning: suggest parentheses around arithmetic in operand of |

Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] drivers/message/i2o/iop.c: unexport i2o_msg_nop()
Adrian Bunk [Tue, 27 Jun 2006 09:55:08 +0000 (02:55 -0700)]
[PATCH] drivers/message/i2o/iop.c: unexport i2o_msg_nop()

It's available in a header as a static inline - there's no need to export it.

Signed-off-by: Adrian Bunk <bunk@stusta.de>
Cc: Markus Lidel <Markus.Lidel@shadowconnect.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] drivers/char/ipmi/ipmi_msghandler.c: make proc_ipmi_root static
Adrian Bunk [Tue, 27 Jun 2006 09:55:07 +0000 (02:55 -0700)]
[PATCH] drivers/char/ipmi/ipmi_msghandler.c: make proc_ipmi_root static

Make struct proc_ipmi_root static.

Besides this, tremove removes an unused #ifdef CONFIG_PROC_FS from
include/linux/ipmi.h.

Acked-by: Corey Minyard <minyard@acm.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] Remove redundant NULL checks before [kv]free - in drivers/
Jesper Juhl [Tue, 27 Jun 2006 09:55:06 +0000 (02:55 -0700)]
[PATCH] Remove redundant NULL checks before [kv]free - in drivers/

Remove redundant NULL chck before kfree + tiny CodingStyle cleanup for
drivers/

Signed-off-by: Jesper Juhl <jesper.juhl@gmail.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] Remove redundant NULL checks before [kv]free - in kernel/
Jesper Juhl [Tue, 27 Jun 2006 09:55:05 +0000 (02:55 -0700)]
[PATCH] Remove redundant NULL checks before [kv]free - in kernel/

Remove redundant kfree NULL checks from kernel/

Signed-off-by: Jesper Juhl <jesper.juhl@gmail.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] Remove redundant NULL checks before [kv]free - in fs/
Jesper Juhl [Tue, 27 Jun 2006 09:55:04 +0000 (02:55 -0700)]
[PATCH] Remove redundant NULL checks before [kv]free - in fs/

Remove redundant NULL checks before kfree for fs/

Signed-off-by: Jesper Juhl <jesper.juhl@gmail.com>
Acked-by: Mark Fasheh <mark.fasheh@oracle.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] futex_requeue() optimization
Sebastien Dugue [Tue, 27 Jun 2006 09:55:03 +0000 (02:55 -0700)]
[PATCH] futex_requeue() optimization

In futex_requeue(), when the 2 futexes keys hash to the same bucket, there
is no need to move the futex_q to the end of the bucket list.

Signed-off-by: Sebastien Dugue <sebastien.dugue@bull.net>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] rtmutex: Propagate priority settings into PI lock chains
Thomas Gleixner [Tue, 27 Jun 2006 09:55:02 +0000 (02:55 -0700)]
[PATCH] rtmutex: Propagate priority settings into PI lock chains

When the priority of a task, which is blocked on a lock, changes we must
propagate this change into the PI lock chain.  Therefor the chain walk code
is changed to get rid of the references to current to avoid false positives
in the deadlock detector, as setscheduler might be called by a task which
holds the lock on which the task whose priority is changed is blocked.

Also add some comments about the get/put_task_struct usage to avoid
confusion.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Cc: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] rtmutex: Modify rtmutex-tester to test the setscheduler propagation
Thomas Gleixner [Tue, 27 Jun 2006 09:55:01 +0000 (02:55 -0700)]
[PATCH] rtmutex: Modify rtmutex-tester to test the setscheduler propagation

Make test suite setscheduler calls asynchronously.  Remove the waits in the
test cases and add a new testcase to verify the correctness of the
setscheduler priority propagation.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] Drop tasklist lock in do_sched_setscheduler
Thomas Gleixner [Tue, 27 Jun 2006 09:55:00 +0000 (02:55 -0700)]
[PATCH] Drop tasklist lock in do_sched_setscheduler

There is no need to hold tasklist_lock across the setscheduler call, when
we pin the task structure with get_task_struct().  Interrupts are disabled
in setscheduler anyway and the permission checks do not need interrupts
disabled.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] fix rt-mutex defaults and dependencies
Roman Zippel [Tue, 27 Jun 2006 09:55:00 +0000 (02:55 -0700)]
[PATCH] fix rt-mutex defaults and dependencies

Fix defaults and dependencies.

Signed-off-by: Roman Zippel <zippel@linux-m68k.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] pi-futex: futex_lock_pi/futex_unlock_pi support
Ingo Molnar [Tue, 27 Jun 2006 09:54:58 +0000 (02:54 -0700)]
[PATCH] pi-futex: futex_lock_pi/futex_unlock_pi support

This adds the actual pi-futex implementation, based on rt-mutexes.

[dino@in.ibm.com: fix an oops-causing race]
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: Dinakar Guniguntala <dino@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] pi-futex: rt mutex futex api
Ingo Molnar [Tue, 27 Jun 2006 09:54:57 +0000 (02:54 -0700)]
[PATCH] pi-futex: rt mutex futex api

Add proxy-locking rt-mutex functionality needed by pi-futexes.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] pi-futex: rt mutex tester
Thomas Gleixner [Tue, 27 Jun 2006 09:54:56 +0000 (02:54 -0700)]
[PATCH] pi-futex: rt mutex tester

RT-mutex tester: scriptable tester for rt mutexes, which allows userspace
scripting of mutex unit-tests (and dynamic tests as well), using the actual
rt-mutex implementation of the kernel.

[akpm@osdl.org: fixlet]
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] pi-futex: rt mutex debug
Ingo Molnar [Tue, 27 Jun 2006 09:54:55 +0000 (02:54 -0700)]
[PATCH] pi-futex: rt mutex debug

Runtime debugging functionality for rt-mutexes.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] pi-futex: rt mutex docs
Steven Rostedt [Tue, 27 Jun 2006 09:54:54 +0000 (02:54 -0700)]
[PATCH] pi-futex: rt mutex docs

Add rt-mutex documentation.

[rostedt@goodmis.org: Update rt-mutex-design.txt as per Randy Dunlap suggestions]
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Cc: "Randy.Dunlap" <rdunlap@xenotime.net>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] pi-futex: rt mutex core
Ingo Molnar [Tue, 27 Jun 2006 09:54:53 +0000 (02:54 -0700)]
[PATCH] pi-futex: rt mutex core

Core functions for the rt-mutex subsystem.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] pi-futex: scheduler support for pi
Ingo Molnar [Tue, 27 Jun 2006 09:54:51 +0000 (02:54 -0700)]
[PATCH] pi-futex: scheduler support for pi

Add framework to boost/unboost the priority of RT tasks.

This consists of:

 - caching the 'normal' priority in ->normal_prio
 - providing a functions to set/get the priority of the task
 - make sched_setscheduler() aware of boosting

The effective_prio() cleanups also fix a priority-calculation bug pointed out
by Andrey Gelman, in set_user_nice().

has_rt_policy() fix: Peter Williams <pwil3058@bigpond.net.au>

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
Cc: Andrey Gelman <agelman@012.net.il>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] pi-futex: add plist implementation
Ingo Molnar [Tue, 27 Jun 2006 09:54:51 +0000 (02:54 -0700)]
[PATCH] pi-futex: add plist implementation

Add the priority-sorted list (plist) implementation.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] pi-futex: introduce WARN_ON_SMP
Ingo Molnar [Tue, 27 Jun 2006 09:54:50 +0000 (02:54 -0700)]
[PATCH] pi-futex: introduce WARN_ON_SMP

Introduce a new WARN_ON variant: WARN_ON_SMP(cond).

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] pi-futex: introduce debug_check_no_locks_freed()
Ingo Molnar [Tue, 27 Jun 2006 09:54:49 +0000 (02:54 -0700)]
[PATCH] pi-futex: introduce debug_check_no_locks_freed()

Add debug_check_no_locks_freed(), as a central inline to add
bad-lock-free-debugging functionality to.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] pi-futex: robust futex docs fix
Ingo Molnar [Tue, 27 Jun 2006 09:54:48 +0000 (02:54 -0700)]
[PATCH] pi-futex: robust futex docs fix

Fix typo in Documentation/robust-futexes.txt.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] pi-futex: futex code cleanups
Ingo Molnar [Tue, 27 Jun 2006 09:54:47 +0000 (02:54 -0700)]
[PATCH] pi-futex: futex code cleanups

We are pleased to announce "lightweight userspace priority inheritance" (PI)
support for futexes.  The following patchset and glibc patch implements it,
ontop of the robust-futexes patchset which is included in 2.6.16-mm1.

We are calling it lightweight for 3 reasons:

 - in the user-space fastpath a PI-enabled futex involves no kernel work
   (or any other PI complexity) at all.  No registration, no extra kernel
   calls - just pure fast atomic ops in userspace.

 - in the slowpath (in the lock-contention case), the system call and
   scheduling pattern is in fact better than that of normal futexes, due to
   the 'integrated' nature of FUTEX_LOCK_PI.  [more about that further down]

 - the in-kernel PI implementation is streamlined around the mutex
   abstraction, with strict rules that keep the implementation relatively
   simple: only a single owner may own a lock (i.e.  no read-write lock
   support), only the owner may unlock a lock, no recursive locking, etc.

  Priority Inheritance - why, oh why???
  -------------------------------------

Many of you heard the horror stories about the evil PI code circling Linux for
years, which makes no real sense at all and is only used by buggy applications
and which has horrible overhead.  Some of you have dreaded this very moment,
when someone actually submits working PI code ;-)

So why would we like to see PI support for futexes?

We'd like to see it done purely for technological reasons.  We dont think it's
a buggy concept, we think it's useful functionality to offer to applications,
which functionality cannot be achieved in other ways.  We also think it's the
right thing to do, and we think we've got the right arguments and the right
numbers to prove that.  We also believe that we can address all the
counter-arguments as well.  For these reasons (and the reasons outlined below)
we are submitting this patch-set for upstream kernel inclusion.

What are the benefits of PI?

  The short reply:
  ----------------

User-space PI helps achieving/improving determinism for user-space
applications.  In the best-case, it can help achieve determinism and
well-bound latencies.  Even in the worst-case, PI will improve the statistical
distribution of locking related application delays.

  The longer reply:
  -----------------

Firstly, sharing locks between multiple tasks is a common programming
technique that often cannot be replaced with lockless algorithms.  As we can
see it in the kernel [which is a quite complex program in itself], lockless
structures are rather the exception than the norm - the current ratio of
lockless vs.  locky code for shared data structures is somewhere between 1:10
and 1:100.  Lockless is hard, and the complexity of lockless algorithms often
endangers to ability to do robust reviews of said code.  I.e.  critical RT
apps often choose lock structures to protect critical data structures, instead
of lockless algorithms.  Furthermore, there are cases (like shared hardware,
or other resource limits) where lockless access is mathematically impossible.

Media players (such as Jack) are an example of reasonable application design
with multiple tasks (with multiple priority levels) sharing short-held locks:
for example, a highprio audio playback thread is combined with medium-prio
construct-audio-data threads and low-prio display-colory-stuff threads.  Add
video and decoding to the mix and we've got even more priority levels.

So once we accept that synchronization objects (locks) are an unavoidable fact
of life, and once we accept that multi-task userspace apps have a very fair
expectation of being able to use locks, we've got to think about how to offer
the option of a deterministic locking implementation to user-space.

Most of the technical counter-arguments against doing priority inheritance
only apply to kernel-space locks.  But user-space locks are different, there
we cannot disable interrupts or make the task non-preemptible in a critical
section, so the 'use spinlocks' argument does not apply (user-space spinlocks
have the same priority inversion problems as other user-space locking
constructs).  Fact is, pretty much the only technique that currently enables
good determinism for userspace locks (such as futex-based pthread mutexes) is
priority inheritance:

Currently (without PI), if a high-prio and a low-prio task shares a lock [this
is a quite common scenario for most non-trivial RT applications], even if all
critical sections are coded carefully to be deterministic (i.e.  all critical
sections are short in duration and only execute a limited number of
instructions), the kernel cannot guarantee any deterministic execution of the
high-prio task: any medium-priority task could preempt the low-prio task while
it holds the shared lock and executes the critical section, and could delay it
indefinitely.

  Implementation:
  ---------------

As mentioned before, the userspace fastpath of PI-enabled pthread mutexes
involves no kernel work at all - they behave quite similarly to normal
futex-based locks: a 0 value means unlocked, and a value==TID means locked.
(This is the same method as used by list-based robust futexes.) Userspace uses
atomic ops to lock/unlock these mutexes without entering the kernel.

To handle the slowpath, we have added two new futex ops:

  FUTEX_LOCK_PI
  FUTEX_UNLOCK_PI

If the lock-acquire fastpath fails, [i.e.  an atomic transition from 0 to TID
fails], then FUTEX_LOCK_PI is called.  The kernel does all the remaining work:
if there is no futex-queue attached to the futex address yet then the code
looks up the task that owns the futex [it has put its own TID into the futex
value], and attaches a 'PI state' structure to the futex-queue.  The pi_state
includes an rt-mutex, which is a PI-aware, kernel-based synchronization
object.  The 'other' task is made the owner of the rt-mutex, and the
FUTEX_WAITERS bit is atomically set in the futex value.  Then this task tries
to lock the rt-mutex, on which it blocks.  Once it returns, it has the mutex
acquired, and it sets the futex value to its own TID and returns.  Userspace
has no other work to perform - it now owns the lock, and futex value contains
FUTEX_WAITERS|TID.

If the unlock side fastpath succeeds, [i.e.  userspace manages to do a TID ->
0 atomic transition of the futex value], then no kernel work is triggered.

If the unlock fastpath fails (because the FUTEX_WAITERS bit is set), then
FUTEX_UNLOCK_PI is called, and the kernel unlocks the futex on the behalf of
userspace - and it also unlocks the attached pi_state->rt_mutex and thus wakes
up any potential waiters.

Note that under this approach, contrary to other PI-futex approaches, there is
no prior 'registration' of a PI-futex.  [which is not quite possible anyway,
due to existing ABI properties of pthread mutexes.]

Also, under this scheme, 'robustness' and 'PI' are two orthogonal properties
of futexes, and all four combinations are possible: futex, robust-futex,
PI-futex, robust+PI-futex.

  glibc support:
  --------------

Ulrich Drepper and Jakub Jelinek have written glibc support for PI-futexes
(and robust futexes), enabling robust and PI (PTHREAD_PRIO_INHERIT) POSIX
mutexes.  (PTHREAD_PRIO_PROTECT support will be added later on too, no
additional kernel changes are needed for that).  [NOTE: The glibc patch is
obviously inofficial and unsupported without matching upstream kernel
functionality.]

the patch-queue and the glibc patch can also be downloaded from:

  http://redhat.com/~mingo/PI-futex-patches/

Many thanks go to the people who helped us create this kernel feature: Steven
Rostedt, Esben Nielsen, Benedikt Spranger, Daniel Walker, John Cooper, Arjan
van de Ven, Oleg Nesterov and others.  Credits for related prior projects goes
to Dirk Grambow, Inaky Perez-Gonzalez, Bill Huey and many others.

Clean up the futex code, before adding more features to it:

 - use u32 as the futex field type - that's the ABI
 - use __user and pointers to u32 instead of unsigned long
 - code style / comment style cleanups
 - rename hash-bucket name from 'bh' to 'hb'.

I checked the pre and post futex.o object files to make sure this
patch has no code effects.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
Cc: Ulrich Drepper <drepper@redhat.com>
Cc: Jakub Jelinek <jakub@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] BUG() if setscheduler is called from interrupt context
Steven Rostedt [Tue, 27 Jun 2006 09:54:44 +0000 (02:54 -0700)]
[PATCH] BUG() if setscheduler is called from interrupt context

Thomas Gleixner is adding the call to a rtmutex function in setscheduler.
This call grabs a spin_lock that is not always protected by interrupts
disabled.  So this means that setscheduler cant be called from interrupt
context.

To prevent this from happening in the future, this patch adds a
BUG_ON(in_interrupt()) in that function.  (Thanks to akpm <aka.  Andrew
Morton> for this suggestion).

Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] sched: uninline task_rq_lock()
Oleg Nesterov [Tue, 27 Jun 2006 09:54:42 +0000 (02:54 -0700)]
[PATCH] sched: uninline task_rq_lock()

Saves 543 bytes from sched.o (gcc 3.3.3).

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Con Kolivas <kernel@kolivas.org>
Cc: Peter Williams <pwil3058@bigpond.net.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] sched: mc/smt power savings sched policy
Siddha, Suresh B [Tue, 27 Jun 2006 09:54:42 +0000 (02:54 -0700)]
[PATCH] sched: mc/smt power savings sched policy

sysfs entries 'sched_mc_power_savings' and 'sched_smt_power_savings' in
/sys/devices/system/cpu/ control the MC/SMT power savings policy for the
scheduler.

Based on the values (1-enable, 0-disable) for these controls, sched groups
cpu power will be determined for different domains.  When power savings
policy is enabled and under light load conditions, scheduler will minimize
the physical packages/cpu cores carrying the load and thus conserving
power(with a perf impact based on the workload characteristics...  see OLS
2005 CMP kernel scheduler paper for more details..)

Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Con Kolivas <kernel@kolivas.org>
Cc: "Chen, Kenneth W" <kenneth.w.chen@intel.com>
Cc: "David S. Miller" <davem@davemloft.net>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] sched_domai: Allocate sched_group structures dynamically
Srivatsa Vaddagiri [Tue, 27 Jun 2006 09:54:41 +0000 (02:54 -0700)]
[PATCH] sched_domai: Allocate sched_group structures dynamically

As explained here:
http://marc.theaimsgroup.com/?l=linux-kernel&m=114327539012323&w=2

there is a problem with sharing sched_group structures between two
separate sched_group structures for different sched_domains.

The patch has been tested and found to avoid the kernel lockup problem
described in above URL.

Signed-off-by: Srivatsa Vaddagiri <vatsa@in.ibm.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: "Siddha, Suresh B" <suresh.b.siddha@intel.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] sched_domai: Use kmalloc_node
Srivatsa Vaddagiri [Tue, 27 Jun 2006 09:54:40 +0000 (02:54 -0700)]
[PATCH] sched_domai: Use kmalloc_node

The sched group structures used to represent various nodes need to be
allocated from respective nodes (as suggested here also:

http://uwsg.ucs.indiana.edu/hypermail/linux/kernel/0603.3/0051.html)

Signed-off-by: Srivatsa Vaddagiri <vatsa@in.ibm.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: "Siddha, Suresh B" <suresh.b.siddha@intel.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] sched_domai: Don't use GFP_ATOMIC
Srivatsa Vaddagiri [Tue, 27 Jun 2006 09:54:39 +0000 (02:54 -0700)]
[PATCH] sched_domai: Don't use GFP_ATOMIC

Replace GFP_ATOMIC allocation for sched_group_nodes with GFP_KERNEL based
allocation.

Signed-off-by: Srivatsa Vaddagiri <vatsa@in.ibm.com
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: "Siddha, Suresh B" <suresh.b.siddha@intel.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] sched_domain: handle kmalloc failure
Srivatsa Vaddagiri [Tue, 27 Jun 2006 09:54:38 +0000 (02:54 -0700)]
[PATCH] sched_domain: handle kmalloc failure

Try to handle mem allocation failures in build_sched_domains by bailing out
and cleaning up thus-far allocated memory.  The patch has a direct consequence
that we disable load balancing completely (even at sibling level) upon *any*
memory allocation failure.

[Lee.Schermerhorn@hp.com: bugfix]
Signed-off-by: Srivatsa Vaddagir <vatsa@in.ibm.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: "Siddha, Suresh B" <suresh.b.siddha@intel.com>
Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] sched: Avoid unnecessarily moving highest priority task move_tasks()
Peter Williams [Tue, 27 Jun 2006 09:54:37 +0000 (02:54 -0700)]
[PATCH] sched: Avoid unnecessarily moving highest priority task move_tasks()

Problem:

To help distribute high priority tasks evenly across the available CPUs
move_tasks() does not, under some circumstances, skip tasks whose load
weight is bigger than the designated amount.  Because the highest priority
task on the busiest queue may be on the expired array it may be moved as a
result of this mechanism.  Apart from not being the most desirable way to
redistribute the high priority tasks (we'd rather move the second highest
priority task), there is a risk that this could set up a loop with this
task bouncing backwards and forwards between the two queues.  (This latter
possibility can be demonstrated by running a nice==-20 CPU bound task on an
otherwise quiet 2 CPU system.)

Solution:

Modify the mechanism so that it does not override skip for the highest
priority task on the CPU.  Of course, if there are more than one tasks at
the highest priority then it will allow the override for one of them as
this is a desirable redistribution of high priority tasks.

Signed-off-by: Peter Williams <pwil3058@bigpond.com.au>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: "Siddha, Suresh B" <suresh.b.siddha@intel.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] sched: modify move_tasks() to improve load balancing outcomes
Peter Williams [Tue, 27 Jun 2006 09:54:36 +0000 (02:54 -0700)]
[PATCH] sched: modify move_tasks() to improve load balancing outcomes

Problem:

The move_tasks() function is designed to move UP TO the amount of load it
is asked to move and in doing this it skips over tasks looking for ones
whose load weights are less than or equal to the remaining load to be
moved.  This is (in general) a good thing but it has the unfortunate result
of breaking one of the original load balancer's good points: namely, that
(within the limits imposed by the active/expired array model and the fact
the expired is processed first) it moves high priority tasks before low
priority ones and this means there's a good chance (see active/expired
problem for why it's only a chance) that the highest priority task on the
queue but not actually on the CPU will be moved to the other CPU where (as
a high priority task) it may preempt the current task.

Solution:

Modify move_tasks() so that high priority tasks are not skipped when moving
them will make them the highest priority task on their new run queue.

Signed-off-by: Peter Williams <pwil3058@bigpond.com.au>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: "Siddha, Suresh B" <suresh.b.siddha@intel.com>
Cc: "Chen, Kenneth W" <kenneth.w.chen@intel.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] sched: implement smpnice
Peter Williams [Tue, 27 Jun 2006 09:54:34 +0000 (02:54 -0700)]
[PATCH] sched: implement smpnice

Problem:

The introduction of separate run queues per CPU has brought with it "nice"
enforcement problems that are best described by a simple example.

For the sake of argument suppose that on a single CPU machine with a
nice==19 hard spinner and a nice==0 hard spinner running that the nice==0
task gets 95% of the CPU and the nice==19 task gets 5% of the CPU.  Now
suppose that there is a system with 2 CPUs and 2 nice==19 hard spinners and
2 nice==0 hard spinners running.  The user of this system would be entitled
to expect that the nice==0 tasks each get 95% of a CPU and the nice==19
tasks only get 5% each.  However, whether this expectation is met is pretty
much down to luck as there are four equally likely distributions of the
tasks to the CPUs that the load balancing code will consider to be balanced
with loads of 2.0 for each CPU.  Two of these distributions involve one
nice==0 and one nice==19 task per CPU and in these circumstances the users
expectations will be met.  The other two distributions both involve both
nice==0 tasks being on one CPU and both nice==19 being on the other CPU and
each task will get 50% of a CPU and the user's expectations will not be
met.

Solution:

The solution to this problem that is implemented in the attached patch is
to use weighted loads when determining if the system is balanced and, when
an imbalance is detected, to move an amount of weighted load between run
queues (as opposed to a number of tasks) to restore the balance.  Once
again, the easiest way to explain why both of these measures are necessary
is to use a simple example.  Suppose that (in a slight variation of the
above example) that we have a two CPU system with 4 nice==0 and 4 nice=19
hard spinning tasks running and that the 4 nice==0 tasks are on one CPU and
the 4 nice==19 tasks are on the other CPU.  The weighted loads for the two
CPUs would be 4.0 and 0.2 respectively and the load balancing code would
move 2 tasks resulting in one CPU with a load of 2.0 and the other with
load of 2.2.  If this was considered to be a big enough imbalance to
justify moving a task and that task was moved using the current
move_tasks() then it would move the highest priority task that it found and
this would result in one CPU with a load of 3.0 and the other with a load
of 1.2 which would result in the movement of a task in the opposite
direction and so on -- infinite loop.  If, on the other hand, an amount of
load to be moved is calculated from the imbalance (in this case 0.1) and
move_tasks() skips tasks until it find ones whose contributions to the
weighted load are less than this amount it would move two of the nice==19
tasks resulting in a system with 2 nice==0 and 2 nice=19 on each CPU with
loads of 2.1 for each CPU.

One of the advantages of this mechanism is that on a system where all tasks
have nice==0 the load balancing calculations would be mathematically
identical to the current load balancing code.

Notes:

struct task_struct:

has a new field load_weight which (in a trade off of space for speed)
stores the contribution that this task makes to a CPU's weighted load when
it is runnable.

struct runqueue:

has a new field raw_weighted_load which is the sum of the load_weight
values for the currently runnable tasks on this run queue.  This field
always needs to be updated when nr_running is updated so two new inline
functions inc_nr_running() and dec_nr_running() have been created to make
sure that this happens.  This also offers a convenient way to optimize away
this part of the smpnice mechanism when CONFIG_SMP is not defined.

int try_to_wake_up():

in this function the value SCHED_LOAD_BALANCE is used to represent the load
contribution of a single task in various calculations in the code that
decides which CPU to put the waking task on.  While this would be a valid
on a system where the nice values for the runnable tasks were distributed
evenly around zero it will lead to anomalous load balancing if the
distribution is skewed in either direction.  To overcome this problem
SCHED_LOAD_SCALE has been replaced by the load_weight for the relevant task
or by the average load_weight per task for the queue in question (as
appropriate).

int move_tasks():

The modifications to this function were complicated by the fact that
active_load_balance() uses it to move exactly one task without checking
whether an imbalance actually exists.  This precluded the simple
overloading of max_nr_move with max_load_move and necessitated the addition
of the latter as an extra argument to the function.  The internal
implementation is then modified to move up to max_nr_move tasks and
max_load_move of weighted load.  This slightly complicates the code where
move_tasks() is called and if ever active_load_balance() is changed to not
use move_tasks() the implementation of move_tasks() should be simplified
accordingly.

struct sched_group *find_busiest_group():

Similar to try_to_wake_up(), there are places in this function where
SCHED_LOAD_SCALE is used to represent the load contribution of a single
task and the same issues are created.  A similar solution is adopted except
that it is now the average per task contribution to a group's load (as
opposed to a run queue) that is required.  As this value is not directly
available from the group it is calculated on the fly as the queues in the
groups are visited when determining the busiest group.

A key change to this function is that it is no longer to scale down
*imbalance on exit as move_tasks() uses the load in its scaled form.

void set_user_nice():

has been modified to update the task's load_weight field when it's nice
value and also to ensure that its run queue's raw_weighted_load field is
updated if it was runnable.

From: "Siddha, Suresh B" <suresh.b.siddha@intel.com>

With smpnice, sched groups with highest priority tasks can mask the imbalance
between the other sched groups with in the same domain.  This patch fixes some
of the listed down scenarios by not considering the sched groups which are
lightly loaded.

a) on a simple 4-way MP system, if we have one high priority and 4 normal
   priority tasks, with smpnice we would like to see the high priority task
   scheduled on one cpu, two other cpus getting one normal task each and the
   fourth cpu getting the remaining two normal tasks.  but with current
   smpnice extra normal priority task keeps jumping from one cpu to another
   cpu having the normal priority task.  This is because of the
   busiest_has_loaded_cpus, nr_loaded_cpus logic..  We are not including the
   cpu with high priority task in max_load calculations but including that in
   total and avg_load calcuations..  leading to max_load < avg_load and load
   balance between cpus running normal priority tasks(2 Vs 1) will always show
   imbalanace as one normal priority and the extra normal priority task will
   keep moving from one cpu to another cpu having normal priority task..

b) 4-way system with HT (8 logical processors).  Package-P0 T0 has a
   highest priority task, T1 is idle.  Package-P1 Both T0 and T1 have 1 normal
   priority task each..  P2 and P3 are idle.  With this patch, one of the
   normal priority tasks on P1 will be moved to P2 or P3..

c) With the current weighted smp nice calculations, it doesn't always make
   sense to look at the highest weighted runqueue in the busy group..
   Consider a load balance scenario on a DP with HT system, with Package-0
   containing one high priority and one low priority, Package-1 containing one
   low priority(with other thread being idle)..  Package-1 thinks that it need
   to take the low priority thread from Package-0.  And find_busiest_queue()
   returns the cpu thread with highest priority task..  And ultimately(with
   help of active load balance) we move high priority task to Package-1.  And
   same continues with Package-0 now, moving high priority task from package-1
   to package-0..  Even without the presence of active load balance, load
   balance will fail to balance the above scenario..  Fix find_busiest_queue
   to use "imbalance" when it is lightly loaded.

[kernel@kolivas.org: sched: store weighted load on up]
[kernel@kolivas.org: sched: add discrete weighted cpu load function]
[suresh.b.siddha@intel.com: sched: remove dead code]
Signed-off-by: Peter Williams <pwil3058@bigpond.com.au>
Cc: "Siddha, Suresh B" <suresh.b.siddha@intel.com>
Cc: "Chen, Kenneth W" <kenneth.w.chen@intel.com>
Acked-by: Ingo Molnar <mingo@elte.hu>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: Con Kolivas <kernel@kolivas.org>
Cc: John Hawkes <hawkes@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] sched: CPU hotplug race vs. set_cpus_allowed()
Kirill Korotaev [Tue, 27 Jun 2006 09:54:32 +0000 (02:54 -0700)]
[PATCH] sched: CPU hotplug race vs. set_cpus_allowed()

There is a race between set_cpus_allowed() and move_task_off_dead_cpu().
__migrate_task() doesn't report any err code, so task can be left on its
runqueue if its cpus_allowed mask changed so that dest_cpu is not longer a
possible target.  Also, chaning cpus_allowed mask requires rq->lock being
held.

Signed-off-by: Kirill Korotaev <dev@openvz.org>
Acked-By: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] unnecessary long index i in sched
Steven Rostedt [Tue, 27 Jun 2006 09:54:31 +0000 (02:54 -0700)]
[PATCH] unnecessary long index i in sched

Unless we expect to have more than 2G CPUs, there's no reason to have 'i'
as a long long here.

Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
17 years ago[PATCH] sched: fix interactive ceiling code
Con Kolivas [Tue, 27 Jun 2006 09:54:30 +0000 (02:54 -0700)]
[PATCH] sched: fix interactive ceiling code

The relationship between INTERACTIVE_SLEEP and the ceiling is not perfect
and not explicit enough.  The sleep boost is not supposed to be any larger
than without this code and the comment is not clear enough about what
exactly it does, just the reason it does it.  Fix it.

There is a ceiling to the priority beyond which tasks that only ever sleep
for very long periods cannot surpass.  Fix it.

Prevent the on-runqueue bonus logic from defeating the idle sleep logic.

Opportunity to micro-optimise.

Signed-off-by: Con Kolivas <kernel@kolivas.org>
Signed-off-by: Mike Galbraith <efault@gmx.de>
Acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Ken Chen <kenneth.w.chen@intel.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>