openwrt/staging/blogic.git
17 years agofreezer: fix racy usage of try_to_freeze in kswapd
Rafael J. Wysocki [Sun, 6 May 2007 21:50:48 +0000 (14:50 -0700)]
freezer: fix racy usage of try_to_freeze in kswapd

Currently we can miss freeze_process()->signal_wake_up() in kswapd() if it
happens between try_to_freeze() and prepare_to_wait().  To prevent this
from happening we should check freezing(current) before calling schedule().

Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Cc: Pavel Machek <pavel@ucw.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoswsusp: use rbtree for tracking allocated swap
Rafael J. Wysocki [Sun, 6 May 2007 21:50:47 +0000 (14:50 -0700)]
swsusp: use rbtree for tracking allocated swap

Make swsusp use extents instead of a bitmap to trace swap pages allocated
for saving the image (the tracking is only needed in case there's an error,
so that the allocated swap pages can be released).

This should allow us to reduce the memory usage, practically always, and
improve performance.

Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Cc: Nigel Cunningham <nigel@nigel.suspend2.net>
Cc: Pavel Machek <pavel@ucw.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agofreezer: remove PF_NOFREEZE from handle_initrd
Rafael J. Wysocki [Sun, 6 May 2007 21:50:46 +0000 (14:50 -0700)]
freezer: remove PF_NOFREEZE from handle_initrd

Make handle_initrd() call try_to_freeze() in a suitable place instead of setting
PF_NOFREEZE for the current task.

Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Acked-by: Nigel Cunningham <nigel@nigel.suspend2.net>
Acked-by: Pavel Machek <pavel@ucw.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoswsusp: use GFP_KERNEL for creating basic data structures
Rafael J. Wysocki [Sun, 6 May 2007 21:50:45 +0000 (14:50 -0700)]
swsusp: use GFP_KERNEL for creating basic data structures

Make swsusp call create_basic_memory_bitmaps() before processes are frozen, so
that GFP_KERNEL allocations can be made in it.  Additionally, ensure that the
swsusp's userland interface won't be used while either pm_suspend_disk() or
software_resume() is being executed.

Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Acked-by: Pavel Machek <pavel@ucw.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoswsusp: fix error paths in snapshot_open
Rafael J. Wysocki [Sun, 6 May 2007 21:50:44 +0000 (14:50 -0700)]
swsusp: fix error paths in snapshot_open

We forget to increase device_available if there's an error in snapshot_open(),
so the snapshot device cannot be open at all after snapshot_open() has
returned an error.

Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Acked-by: Pavel Machek <pavel@ucw.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agomm: remove unused page flags
Rafael J. Wysocki [Sun, 6 May 2007 21:50:43 +0000 (14:50 -0700)]
mm: remove unused page flags

Remove the two page flags that were previously used by swsusp and are no
longer needed.

Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Acked-by: Pavel Machek <pavel@ucw.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoswsusp: do not use page flags
Rafael J. Wysocki [Sun, 6 May 2007 21:50:43 +0000 (14:50 -0700)]
swsusp: do not use page flags

Make swsusp use memory bitmaps instead of page flags for marking 'nosave' and
free pages.  This allows us to 'recycle' two page flags that can be used for
other purposes.  Also, the memory needed to store the bitmaps is allocated
when necessary (ie.  before the suspend) and freed after the resume which is
more reasonable.

The patch is designed to minimize the amount of changes and there are some
nice simplifications and optimizations possible on top of it.  I am going to
implement them separately in the future.

Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Acked-by: Pavel Machek <pavel@ucw.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoswsusp: use inline functions for changing page flags
Rafael J. Wysocki [Sun, 6 May 2007 21:50:42 +0000 (14:50 -0700)]
swsusp: use inline functions for changing page flags

Replace direct invocations of SetPageNosave(), SetPageNosaveFree() etc.  with
calls to inline functions that can be changed in subsequent patches without
modifying the code calling them.

Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Acked-by: Pavel Machek <pavel@ucw.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agofix refrigerator() vs thaw_process() race
Oleg Nesterov [Sun, 6 May 2007 21:50:40 +0000 (14:50 -0700)]
fix refrigerator() vs thaw_process() race

refrigerator() can miss a wakeup, "wait event" loop needs a proper memory
ordering.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Acked-by: "Rafael J. Wysocki" <rjw@sisk.pl>
Cc: Pavel Machek <pavel@ucw.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoARM26: remove useless config option GENERIC_BUST_SPINLOCK.
Robert P. J. Day [Sun, 6 May 2007 21:50:39 +0000 (14:50 -0700)]
ARM26: remove useless config option GENERIC_BUST_SPINLOCK.

Remove the apparently useless config option GENERIC_BUST_SPINLOCK,
since nothing in the source tree refers to it.

Signed-off-by: Robert P. J. Day <rpjday@mindspring.com>
Acked-by: Ian Molton <spyro@f2s.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agosrmcons: fix kmalloc(GFP_KERNEL) inside spinlock
Andrew Morton [Sun, 6 May 2007 21:50:39 +0000 (14:50 -0700)]
srmcons: fix kmalloc(GFP_KERNEL) inside spinlock

Fixes http://bugzilla.kernel.org/show_bug.cgi?id=8341

Cc: <matthias.kaehlcke@gmail.com>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoALPHA: "prctl" macros
Ivan Kokshaysky [Sun, 6 May 2007 21:50:38 +0000 (14:50 -0700)]
ALPHA: "prctl" macros

Files:

include/asm-alpha/thread_info.h

Provide "prctl" macros for ALPHA.

Signed-off-by: Jay Estabrook <jay.estabrook@hp.com>
Signed-off-by: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Cc: Richard Henderson <rth@twiddle.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoALPHA: fix BOOTP image creation
Ivan Kokshaysky [Sun, 6 May 2007 21:50:37 +0000 (14:50 -0700)]
ALPHA: fix BOOTP image creation

Files:

arch/alpha/boot/bootpz.c

Create a dummy "__kmalloc()" to satisfy the loader; never called.

arch/alpha/boot/tools/objstrip.c

Remove an include that is now (2.6.x) unnecessary.

Signed-off-by: Jay Estabrook <jay.estabrook@hp.com>
Signed-off-by: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Cc: Richard Henderson <rth@twiddle.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoROUND_UP macro cleanup in arch/alpha/kernel/osf_sys.c
Milind Arun Choudhary [Sun, 6 May 2007 21:50:36 +0000 (14:50 -0700)]
ROUND_UP macro cleanup in arch/alpha/kernel/osf_sys.c

ROUND_UP macro cleanup use ALIGN

Signed-off-by: Milind Arun Choudhary <milindchoudhary@gmail.com>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Cc: Richard Henderson <rth@twiddle.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoh8300: add zImage support
Yoshinori Sato [Sun, 6 May 2007 21:50:36 +0000 (14:50 -0700)]
h8300: add zImage support

h8300 zImage target support.

Signed-off-by: Yoshinori Sato <ysato@users.sourceforge.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoh8300 generic irq
Yoshinori Sato [Sun, 6 May 2007 21:50:35 +0000 (14:50 -0700)]
h8300 generic irq

h8300 using generic irq handler patch.

Signed-off-by: Yoshinori Sato <ysato@users.sourceforge.jp>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoConvert h8/300 to generic timekeeping
john stultz [Sun, 6 May 2007 21:50:34 +0000 (14:50 -0700)]
Convert h8/300 to generic timekeeping

Currently h8/300 does not implement sub-jiffy timekeeping, so there is no
benefit to having arch specific timekeeping code.

This patch simply removes those functions and enables the generic
timekeeping code.

Signed-off-by: John Stultz <johnstul@us.ibm.com>
Acked-by: Yoshinori Sato <ysato@users.sourceforge.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoBlackfin: blackfin on-chip SPI controller driver
Wu, Bryan [Sun, 6 May 2007 21:50:34 +0000 (14:50 -0700)]
Blackfin: blackfin on-chip SPI controller driver

This patch implements the driver necessary use the Analog Devices Blackfin
processor's SPI Port.

Signed-off-by: Bryan Wu <bryan.wu@analog.com>
Cc: David Brownell <david-b@pacbell.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoBlackfin: on-chip RTC controller driver
Wu, Bryan [Sun, 6 May 2007 21:50:32 +0000 (14:50 -0700)]
Blackfin: on-chip RTC controller driver

This patch implements the driver necessary use the Analog Devices Blackfin
processor's on-chip RTC controller.

Signed-off-by: Bryan Wu <bryan.wu@analog.com>
Cc: Alessandro Zummo <a.zummo@towertech.it>
Cc: David Brownell <david-b@pacbell.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoBlackfin: add blackfin support in smc91x ethernet controller driver
Wu, Bryan [Sun, 6 May 2007 21:50:32 +0000 (14:50 -0700)]
Blackfin: add blackfin support in smc91x ethernet controller driver

As SMC91X ethernet controller are used in blackfin STAMP 533 development
board, this patch add blackfin support to the smc91x linux driver.

Signed-off-by: Bryan Wu <bryan.wu@analog.com>
Acked-by: Jeff Garzik <jeff@garzik.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoblackfin: serial driver
Bryan Wu [Sun, 6 May 2007 21:50:30 +0000 (14:50 -0700)]
blackfin: serial driver

This patch implements the driver necessary use the Analog Devices Blackfin
processor's Serial Port.

Signed-off-by: Bryan Wu <bryan.wu@analog.com>
Cc: Alan Cox <alan@lxorguk.ukuu.org.uk>
Cc: Russell King <rmk+lkml@arm.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoblackfin architecture
Bryan Wu [Sun, 6 May 2007 21:50:22 +0000 (14:50 -0700)]
blackfin architecture

This adds support for the Analog Devices Blackfin processor architecture, and
currently supports the BF533, BF532, BF531, BF537, BF536, BF534, and BF561
(Dual Core) devices, with a variety of development platforms including those
avaliable from Analog Devices (BF533-EZKit, BF533-STAMP, BF537-STAMP,
BF561-EZKIT), and Bluetechnix!  Tinyboards.

The Blackfin architecture was jointly developed by Intel and Analog Devices
Inc.  (ADI) as the Micro Signal Architecture (MSA) core and introduced it in
December of 2000.  Since then ADI has put this core into its Blackfin
processor family of devices.  The Blackfin core has the advantages of a clean,
orthogonal,RISC-like microprocessor instruction set.  It combines a dual-MAC
(Multiply/Accumulate), state-of-the-art signal processing engine and
single-instruction, multiple-data (SIMD) multimedia capabilities into a single
instruction-set architecture.

The Blackfin architecture, including the instruction set, is described by the
ADSP-BF53x/BF56x Blackfin Processor Programming Reference
http://blackfin.uclinux.org/gf/download/frsrelease/29/2549/Blackfin_PRM.pdf

The Blackfin processor is already supported by major releases of gcc, and
there are binary and source rpms/tarballs for many architectures at:
http://blackfin.uclinux.org/gf/project/toolchain/frs There is complete
documentation, including "getting started" guides available at:
http://docs.blackfin.uclinux.org/ which provides links to the sources and
patches you will need in order to set up a cross-compiling environment for
bfin-linux-uclibc

This patch, as well as the other patches (toolchain, distribution,
uClibc) are actively supported by Analog Devices Inc, at:
http://blackfin.uclinux.org/

We have tested this on LTP, and our test plan (including pass/fails) can
be found at:
http://docs.blackfin.uclinux.org/doku.php?id=testing_the_linux_kernel

[m.kozlowski@tuxland.pl: balance parenthesis in blackfin header files]
Signed-off-by: Bryan Wu <bryan.wu@analog.com>
Signed-off-by: Mariusz Kozlowski <m.kozlowski@tuxland.pl>
Signed-off-by: Aubrey Li <aubrey.li@analog.com>
Signed-off-by: Jie Zhang <jie.zhang@analog.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoReturn EPERM not ECHILD on security_task_wait failure
Roland McGrath [Sun, 6 May 2007 21:50:20 +0000 (14:50 -0700)]
Return EPERM not ECHILD on security_task_wait failure

wait* syscalls return -ECHILD even when an individual PID of a live child
was requested explicitly, when security_task_wait denies the operation.
This means that something like a broken SELinux policy can produce an
unexpected failure that looks just like a bug with wait or ptrace or
something.

This patch makes do_wait return -EACCES (or other appropriate error returned
from security_task_wait() instead of -ECHILD if some children were ruled out
solely because security_task_wait failed.

[jmorris@namei.org: switch error code to EACCES]
Signed-off-by: Roland McGrath <roland@redhat.com>
Acked-by: Stephen Smalley <sds@tycho.nsa.gov>
Cc: Chris Wright <chrisw@sous-sol.org>
Cc: James Morris <jmorris@namei.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agopage migration: Only migrate pages if allocation in the highest zone is possible
Christoph Lameter [Sun, 6 May 2007 21:50:20 +0000 (14:50 -0700)]
page migration: Only migrate pages if allocation in the highest zone is possible

Address spaces contain an allocation flag that specifies restriction on the
zone for pages placed in the mapping.  I.e.  some device may require pages
to be allocated from a DMA zone.  Block devices may not be able to use
pages from HIGHMEM.

Memory policies and the common use of page migration works only on the
highest zone.  If the address space does not allow allocation from the
highest zone then the pages in the address space are not migratable simply
because we can only allocate memory for a specified node if we allow
allocation for the highest zone on each node.

Acked-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoslob: fix page order calculation on not 4KB page
Akinobu Mita [Sun, 6 May 2007 21:50:19 +0000 (14:50 -0700)]
slob: fix page order calculation on not 4KB page

SLOB doesn't calculate correct page order when page size is not 4KB.  This
patch fixes it with using get_order() instead of find_order() which is SLOB
version of get_order().

Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Acked-by: Matt Mackall <mpm@selenic.com>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agohugetlbfs: add NULL check in hugetlb_zero_setup()
Akinobu Mita [Sun, 6 May 2007 21:50:18 +0000 (14:50 -0700)]
hugetlbfs: add NULL check in hugetlb_zero_setup()

If hugetlbfs module_init() fails, hugetlbfs_vfsmount is not initialized and
shmget() with SHM_HUGETLB flag will cause NULL pointer dereference.

Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Acked-by: William Irwin <wli@holomorphy.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoSlab allocators: remove useless __GFP_NO_GROW flag
Christoph Lameter [Sun, 6 May 2007 21:50:17 +0000 (14:50 -0700)]
Slab allocators: remove useless __GFP_NO_GROW flag

There is no user remaining and I have never seen any use of that flag.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoslab allocators: Remove SLAB_CTOR_ATOMIC
Christoph Lameter [Sun, 6 May 2007 21:50:17 +0000 (14:50 -0700)]
slab allocators: Remove SLAB_CTOR_ATOMIC

SLAB_CTOR atomic is never used which is no surprise since I cannot imagine
that one would want to do something serious in a constructor or destructor.
 In particular given that the slab allocators run with interrupts disabled.
 Actions in constructors and destructors are by their nature very limited
and usually do not go beyond initializing variables and list operations.

(The i386 pgd ctor and dtors do take a spinlock in constructor and
destructor.....  I think that is the furthest we go at this point.)

There is no flag passed to the destructor so removing SLAB_CTOR_ATOMIC also
establishes a certain symmetry.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoslab allocators: Remove SLAB_DEBUG_INITIAL flag
Christoph Lameter [Sun, 6 May 2007 21:50:16 +0000 (14:50 -0700)]
slab allocators: Remove SLAB_DEBUG_INITIAL flag

I have never seen a use of SLAB_DEBUG_INITIAL.  It is only supported by
SLAB.

I think its purpose was to have a callback after an object has been freed
to verify that the state is the constructor state again?  The callback is
performed before each freeing of an object.

I would think that it is much easier to check the object state manually
before the free.  That also places the check near the code object
manipulation of the object.

Also the SLAB_DEBUG_INITIAL callback is only performed if the kernel was
compiled with SLAB debugging on.  If there would be code in a constructor
handling SLAB_DEBUG_INITIAL then it would have to be conditional on
SLAB_DEBUG otherwise it would just be dead code.  But there is no such code
in the kernel.  I think SLUB_DEBUG_INITIAL is too problematic to make real
use of, difficult to understand and there are easier ways to accomplish the
same effect (i.e.  add debug code before kfree).

There is a related flag SLAB_CTOR_VERIFY that is frequently checked to be
clear in fs inode caches.  Remove the pointless checks (they would even be
pointless without removeal of SLAB_DEBUG_INITIAL) from the fs constructors.

This is the last slab flag that SLUB did not support.  Remove the check for
unimplemented flags from SLUB.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoget_unmapped_area doesn't need hugetlbfs hacks anymore
Benjamin Herrenschmidt [Sun, 6 May 2007 21:50:14 +0000 (14:50 -0700)]
get_unmapped_area doesn't need hugetlbfs hacks anymore

Remove the hugetlbfs specific hacks in toplevel get_unmapped_area() now that
all archs and hugetlbfs itself do the right thing for both cases.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Acked-by: William Irwin <bill.irwin@oracle.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Cc: Russell King <rmk+kernel@arm.linux.org.uk>
Cc: David Howells <dhowells@redhat.com>
Cc: Andi Kleen <ak@suse.de>
Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: Kyle McMartin <kyle@mcmartin.ca>
Cc: Grant Grundler <grundler@parisc-linux.org>
Cc: Matthew Wilcox <willy@debian.org>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Adam Litke <agl@us.ibm.com>
Cc: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoget_unmapped_area handles MAP_FIXED in generic code
Benjamin Herrenschmidt [Sun, 6 May 2007 21:50:13 +0000 (14:50 -0700)]
get_unmapped_area handles MAP_FIXED in generic code

generic arch_get_unmapped_area() now handles MAP_FIXED.  Now that all
implementations have been fixed, change the toplevel get_unmapped_area() to
call into arch or drivers for the MAP_FIXED case.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Cc: Russell King <rmk+kernel@arm.linux.org.uk>
Cc: David Howells <dhowells@redhat.com>
Cc: Andi Kleen <ak@suse.de>
Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: Kyle McMartin <kyle@mcmartin.ca>
Cc: Grant Grundler <grundler@parisc-linux.org>
Cc: Matthew Wilcox <willy@debian.org>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: William Irwin <bill.irwin@oracle.com>
Cc: Adam Litke <agl@us.ibm.com>
Cc: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoget_unmapped_area handles MAP_FIXED in hugetlbfs
Benjamin Herrenschmidt [Sun, 6 May 2007 21:50:12 +0000 (14:50 -0700)]
get_unmapped_area handles MAP_FIXED in hugetlbfs

Generic hugetlb_get_unmapped_area() now handles MAP_FIXED by just calling
prepare_hugepage_range()

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Acked-by: William Irwin <bill.irwin@oracle.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Cc: Russell King <rmk+kernel@arm.linux.org.uk>
Cc: David Howells <dhowells@redhat.com>
Cc: Andi Kleen <ak@suse.de>
Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: Kyle McMartin <kyle@mcmartin.ca>
Cc: Grant Grundler <grundler@parisc-linux.org>
Cc: Matthew Wilcox <willy@debian.org>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Adam Litke <agl@us.ibm.com>
Cc: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoget_unmapped_area handles MAP_FIXED on x86_64
Benjamin Herrenschmidt [Sun, 6 May 2007 21:50:11 +0000 (14:50 -0700)]
get_unmapped_area handles MAP_FIXED on x86_64

Handle MAP_FIXED in x86_64 arch_get_unmapped_area(), simple case, just return
the address as passed in

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoget_unmapped_area handles MAP_FIXED on sparc64
Benjamin Herrenschmidt [Sun, 6 May 2007 21:50:10 +0000 (14:50 -0700)]
get_unmapped_area handles MAP_FIXED on sparc64

Handle MAP_FIXED in hugetlb_get_unmapped_area on sparc64 by just using
prepare_hugepage_range()

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Acked-by: William Irwin <bill.irwin@oracle.com>
Cc: "David S. Miller" <davem@davemloft.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoget_unmapped_area handles MAP_FIXED on parisc
Benjamin Herrenschmidt [Sun, 6 May 2007 21:50:09 +0000 (14:50 -0700)]
get_unmapped_area handles MAP_FIXED on parisc

Handle MAP_FIXED in parisc arch_get_unmapped_area(), just return the address.
We might want to also check for possible cache aliasing issues now that we get
called in that case (like ARM or MIPS), leave a comment for the maintainers to
pick up.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Kyle McMartin <kyle@mcmartin.ca>
Cc: Grant Grundler <grundler@parisc-linux.org>
Cc: Matthew Wilcox <willy@debian.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoget_unmapped_area handles MAP_FIXED on ia64
Benjamin Herrenschmidt [Sun, 6 May 2007 21:50:09 +0000 (14:50 -0700)]
get_unmapped_area handles MAP_FIXED on ia64

Handle MAP_FIXED in ia64 arch_get_unmapped_area and
hugetlb_get_unmapped_area(), just call prepare_hugepage_range in the later and
is_hugepage_only_range() in the former.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Acked-by: William Irwin <bill.irwin@oracle.com>
Cc: "Luck, Tony" <tony.luck@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoget_unmapped_area handles MAP_FIXED on i386
Benjamin Herrenschmidt [Sun, 6 May 2007 21:50:08 +0000 (14:50 -0700)]
get_unmapped_area handles MAP_FIXED on i386

Handle MAP_FIXED in i386 hugetlb_get_unmapped_area(), just call
prepare_hugepage_range.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Acked-by: William Irwin <bill.irwin@oracle.com>
Cc: Andi Kleen <ak@suse.de>
Cc: Adam Litke <agl@us.ibm.com>
Cc: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoget_unmapped_area handles MAP_FIXED on frv
Benjamin Herrenschmidt [Sun, 6 May 2007 21:50:07 +0000 (14:50 -0700)]
get_unmapped_area handles MAP_FIXED on frv

Handle MAP_FIXED in arch_get_unmapped_area on frv.  Trivial case, just return
the address.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoget_unmapped_area handles MAP_FIXED on arm
Benjamin Herrenschmidt [Sun, 6 May 2007 21:50:07 +0000 (14:50 -0700)]
get_unmapped_area handles MAP_FIXED on arm

ARM already had a case for MAP_FIXED in arch_get_unmapped_area() though it was
not called before.  Fix the comment to reflect that it will now be called.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Acked-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoget_unmapped_area handles MAP_FIXED on alpha
Benjamin Herrenschmidt [Sun, 6 May 2007 21:50:06 +0000 (14:50 -0700)]
get_unmapped_area handles MAP_FIXED on alpha

Handle MAP_FIXED in alpha's arch_get_unmapped_area(), simple case, just return
the address as passed in

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoget_unmapped_area handles MAP_FIXED on powerpc
Benjamin Herrenschmidt [Sun, 6 May 2007 21:50:02 +0000 (14:50 -0700)]
get_unmapped_area handles MAP_FIXED on powerpc

The current get_unmapped_area code calls the f_ops->get_unmapped_area or the
arch one (via the mm) only when MAP_FIXED is not passed.  That makes it
impossible for archs to impose proper constraints on regions of the virtual
address space.  To work around that, get_unmapped_area() then calls some
hugetlbfs specific hacks.

This cause several problems, among others:

- It makes it impossible for a driver or filesystem to do the same thing
  that hugetlbfs does (for example, to allow a driver to use larger page sizes
  to map external hardware) if that requires applying a constraint on the
  addresses (constraining that mapping in certain regions and other mappings
  out of those regions).

- Some archs like arm, mips, sparc, sparc64, sh and sh64 already want
  MAP_FIXED to be passed down in order to deal with aliasing issues.  The code
  is there to handle it...  but is never called.

This series of patches moves the logic to handle MAP_FIXED down to the various
arch/driver get_unmapped_area() implementations, and then changes the generic
code to always call them.  The hugetlbfs hacks then disappear from the generic
code.

Since I need to do some special 64K pages mappings for SPEs on cell, I need to
work around the first problem at least.  I have further patches thus
implementing a "slices" layer that handles multiple page sizes through slices
of the address space for use by hugetlbfs, the SPE code, and possibly others,
but it requires that serie of patches first/

There is still a potential (but not practical) issue due to the fact that
filesystems/drivers implemeting g_u_a will effectively bypass all arch checks.
 This is not an issue in practice as the only filesystems/drivers using that
hook are doing so for arch specific purposes in the first place.

There is also a problem with mremap that will completely bypass all arch
checks.  I'll try to address that separately, I'm not 100% certain yet how,
possibly by making it not work when the vma has a file whose f_ops has a
get_unmapped_area callback, and by making it use is_hugepage_only_range()
before expanding into a new area.

Also, I want to turn is_hugepage_only_range() into a more generic
is_normal_page_range() as that's really what it will end up meaning when used
in stack grow, brk grow and mremap.

None of the above "issues" however are introduced by this patch, they are
already there, so I think the patch can go ini for 2.6.22.

This patch:

Handle MAP_FIXED in powerpc's arch_get_unmapped_area() in all 3
implementations of it.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Acked-by: William Irwin <bill.irwin@oracle.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Cc: Russell King <rmk+kernel@arm.linux.org.uk>
Cc: David Howells <dhowells@redhat.com>
Cc: Andi Kleen <ak@suse.de>
Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: Kyle McMartin <kyle@mcmartin.ca>
Cc: Grant Grundler <grundler@parisc-linux.org>
Cc: Matthew Wilcox <willy@debian.org>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Adam Litke <agl@us.ibm.com>
Cc: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agooom: fix constraint deadlock
David Rientjes [Sun, 6 May 2007 21:50:00 +0000 (14:50 -0700)]
oom: fix constraint deadlock

Fixes a deadlock in the OOM killer for allocations that are not
__GFP_HARDWALL.

Before the OOM killer checks for the allocation constraint, it takes
callback_mutex.

constrained_alloc() iterates through each zone in the allocation zonelist
and calls cpuset_zone_allowed_softwall() to determine whether an allocation
for gfp_mask is possible.  If a zone's node is not in the OOM-triggering
task's mems_allowed, it is not exiting, and we did not fail on a
__GFP_HARDWALL allocation, cpuset_zone_allowed_softwall() attempts to take
callback_mutex to check the nearest exclusive ancestor of current's cpuset.
 This results in deadlock.

We now take callback_mutex after iterating through the zonelist since we
don't need it yet.

Cc: Andi Kleen <ak@suse.de>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Christoph Lameter <clameter@engr.sgi.com>
Cc: Martin J. Bligh <mbligh@mbligh.org>
Signed-off-by: David Rientjes <rientjes@google.com>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agomm: fix handling of panic_on_oom when cpusets are in use
Yasunori Goto [Sun, 6 May 2007 21:49:59 +0000 (14:49 -0700)]
mm: fix handling of panic_on_oom when cpusets are in use

The current panic_on_oom may not work if there is a process using
cpusets/mempolicy, because other nodes' memory may remain.  But some people
want failover by panic ASAP even if they are used.  This patch makes new
setting for its request.

This is tested on my ia64 box which has 3 nodes.

Signed-off-by: Yasunori Goto <y-goto@jp.fujitsu.com>
Signed-off-by: Benjamin LaHaise <bcrl@kvack.org>
Cc: Christoph Lameter <clameter@sgi.com>
Cc: Paul Jackson <pj@sgi.com>
Cc: Ethan Solomita <solo@google.com>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agofault injection: fix failslab with CONFIG_NUMA
Akinobu Mita [Sun, 6 May 2007 21:49:58 +0000 (14:49 -0700)]
fault injection: fix failslab with CONFIG_NUMA

Currently failslab injects failures into ____cache_alloc().  But with enabling
CONFIG_NUMA it's not enough to let actual slab allocator functions (kmalloc,
kmem_cache_alloc, ...) return NULL.

This patch moves fault injection hook inside of __cache_alloc() and
__cache_alloc_node().  These are lower call path than ____cache_alloc() and
enable to inject faulures to slab allocators with CONFIG_NUMA.

Acked-by: Pekka Enberg <penberg@cs.helsinki.fi>
Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Cc: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoslab allocators: remove multiple alignment specifications
Christoph Lameter [Sun, 6 May 2007 21:49:58 +0000 (14:49 -0700)]
slab allocators: remove multiple alignment specifications

It is not necessary to tell the slab allocators to align to a cacheline
if an explicit alignment was already specified. It is rather confusing
to specify multiple alignments.

Make sure that the call sites only use one form of alignment.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoKMEM_CACHE(): simplify slab cache creation
Christoph Lameter [Sun, 6 May 2007 21:49:57 +0000 (14:49 -0700)]
KMEM_CACHE(): simplify slab cache creation

This patch provides a new macro

KMEM_CACHE(<struct>, <flags>)

to simplify slab creation. KMEM_CACHE creates a slab with the name of the
struct, with the size of the struct and with the alignment of the struct.
Additional slab flags may be specified if necessary.

Example

struct test_slab {
int a,b,c;
struct list_head;
} __cacheline_aligned_in_smp;

test_slab_cache = KMEM_CACHE(test_slab, SLAB_PANIC)

will create a new slab named "test_slab" of the size sizeof(struct
test_slab) and aligned to the alignment of test slab.  If it fails then we
panic.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoslab allocators: Remove obsolete SLAB_MUST_HWCACHE_ALIGN
Christoph Lameter [Sun, 6 May 2007 21:49:56 +0000 (14:49 -0700)]
slab allocators: Remove obsolete SLAB_MUST_HWCACHE_ALIGN

This patch was recently posted to lkml and acked by Pekka.

The flag SLAB_MUST_HWCACHE_ALIGN is

1. Never checked by SLAB at all.

2. A duplicate of SLAB_HWCACHE_ALIGN for SLUB

3. Fulfills the role of SLAB_HWCACHE_ALIGN for SLOB.

The only remaining use is in sparc64 and ppc64 and their use there
reflects some earlier role that the slab flag once may have had. If
its specified then SLAB_HWCACHE_ALIGN is also specified.

The flag is confusing, inconsistent and has no purpose.

Remove it.

Acked-by: Pekka Enberg <penberg@cs.helsinki.fi>
Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agomm: optimize acorn partition truncate
Peter Zijlstra [Sun, 6 May 2007 21:49:55 +0000 (14:49 -0700)]
mm: optimize acorn partition truncate

invalidate_bdev() is superfluous when truncate_inode_pages() is also
called.  do call invalidate_bh_lrus() though, to avoid stale pointers.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agomm: optimize kill_bdev()
Peter Zijlstra [Sun, 6 May 2007 21:49:55 +0000 (14:49 -0700)]
mm: optimize kill_bdev()

Remove duplicate work in kill_bdev().

It currently invalidates and then truncates the bdev's mapping.
invalidate_mapping_pages() will opportunistically remove pages from the
mapping.  And truncate_inode_pages() will forcefully remove all pages.

The only thing truncate doesn't do is flush the bh lrus.  So do that
explicitly.  This avoids (very unlikely) but possible invalid lookup
results if the same bdev is quickly re-issued.

It also will prevent extreme kernel latencies which are observed when
blockdevs which have a large amount of pagecache are unmounted, by avoiding
invalidate_mapping_pages() on that path.  invalidate_mapping_pages() has no
cond_resched (it can be called under spinlock), whereas truncate_inode_pages()
has one.

[akpm@linux-foundation.org: restore nrpages==0 optimisation]
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agomm: remove destroy_dirty_buffers from invalidate_bdev()
Peter Zijlstra [Sun, 6 May 2007 21:49:54 +0000 (14:49 -0700)]
mm: remove destroy_dirty_buffers from invalidate_bdev()

Remove the destroy_dirty_buffers argument from invalidate_bdev(), it hasn't
been used in 6 years (so akpm says).

find * -name \*.[ch] | xargs grep -l invalidate_bdev |
while read file; do
quilt add $file;
sed -ie 's/invalidate_bdev(\([^,]*\),[^)]*)/invalidate_bdev(\1)/g' $file;
done

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agomm: madvise avoid exclusive mmap_sem
Nick Piggin [Sun, 6 May 2007 21:49:53 +0000 (14:49 -0700)]
mm: madvise avoid exclusive mmap_sem

Avoid down_write of the mmap_sem in madvise when we can help it.

Acked-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Nick Piggin <npiggin@suse.de>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoinclude KERN_* constant in printk() calls in mm/slab.c
matze [Sun, 6 May 2007 21:49:52 +0000 (14:49 -0700)]
include KERN_* constant in printk() calls in mm/slab.c

Signed-off-by: Matthias Kaehlcke <matthias.kaehlcke@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoslob: handle SLAB_PANIC flag
Akinobu Mita [Sun, 6 May 2007 21:49:52 +0000 (14:49 -0700)]
slob: handle SLAB_PANIC flag

kmem_cache_create() for slob doesn't handle SLAB_PANIC.

Signed-off-by: Matt Mackall <mpm@selenic.com>
Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoQuicklist support for sparc64
David Miller [Sun, 6 May 2007 21:49:51 +0000 (14:49 -0700)]
Quicklist support for sparc64

I ported this to sparc64 as per the patch below, tested on UP SunBlade1500 and
24 cpu Niagara T1000.

Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Christoph Lameter <clameter@sgi.com>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Andi Kleen <ak@suse.de>
Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: William Lee Irwin III <wli@holomorphy.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoQuicklists for page table pages
Christoph Lameter [Sun, 6 May 2007 21:49:50 +0000 (14:49 -0700)]
Quicklists for page table pages

On x86_64 this cuts allocation overhead for page table pages down to a
fraction (kernel compile / editing load.  TSC based measurement of times spend
in each function):

no quicklist

pte_alloc               1569048 4.3s(401ns/2.7us/179.7us)
pmd_alloc                780988 2.1s(337ns/2.7us/86.1us)
pud_alloc                780072 2.2s(424ns/2.8us/300.6us)
pgd_alloc                260022 1s(920ns/4us/263.1us)

quicklist:

pte_alloc                452436 573.4ms(8ns/1.3us/121.1us)
pmd_alloc                196204 174.5ms(7ns/889ns/46.1us)
pud_alloc                195688 172.4ms(7ns/881ns/151.3us)
pgd_alloc                 65228 9.8ms(8ns/150ns/6.1us)

pgd allocations are the most complex and there we see the most dramatic
improvement (may be we can cut down the amount of pgds cached somewhat?).  But
even the pte allocations still see a doubling of performance.

1. Proven code from the IA64 arch.

The method used here has been fine tuned for years and
is NUMA aware. It is based on the knowledge that accesses
to page table pages are sparse in nature. Taking a page
off the freelists instead of allocating a zeroed pages
allows a reduction of number of cachelines touched
in addition to getting rid of the slab overhead. So
performance improves. This is particularly useful if pgds
contain standard mappings. We can save on the teardown
and setup of such a page if we have some on the quicklists.
This includes avoiding lists operations that are otherwise
necessary on alloc and free to track pgds.

2. Light weight alternative to use slab to manage page size pages

Slab overhead is significant and even page allocator use
is pretty heavy weight. The use of a per cpu quicklist
means that we touch only two cachelines for an allocation.
There is no need to access the page_struct (unless arch code
needs to fiddle around with it). So the fast past just
means bringing in one cacheline at the beginning of the
page. That same cacheline may then be used to store the
page table entry. Or a second cacheline may be used
if the page table entry is not in the first cacheline of
the page. The current code will zero the page which means
touching 32 cachelines (assuming 128 byte). We get down
from 32 to 2 cachelines in the fast path.

3. x86_64 gets lightweight page table page management.

This will allow x86_64 arch code to faster repopulate pgds
and other page table entries. The list operations for pgds
are reduced in the same way as for i386 to the point where
a pgd is allocated from the page allocator and when it is
freed back to the page allocator. A pgd can pass through
the quicklists without having to be reinitialized.

64 Consolidation of code from multiple arches

So far arches have their own implementation of quicklist
management. This patch moves that feature into the core allowing
an easier maintenance and consistent management of quicklists.

Page table pages have the characteristics that they are typically zero or in a
known state when they are freed.  This is usually the exactly same state as
needed after allocation.  So it makes sense to build a list of freed page
table pages and then consume the pages already in use first.  Those pages have
already been initialized correctly (thus no need to zero them) and are likely
already cached in such a way that the MMU can use them most effectively.  Page
table pages are used in a sparse way so zeroing them on allocation is not too
useful.

Such an implementation already exits for ia64.  Howver, that implementation
did not support constructors and destructors as needed by i386 / x86_64.  It
also only supported a single quicklist.  The implementation here has
constructor and destructor support as well as the ability for an arch to
specify how many quicklists are needed.

Quicklists are defined by an arch defining CONFIG_QUICKLIST.  If more than one
quicklist is necessary then we can define NR_QUICK for additional lists.  F.e.
 i386 needs two and thus has

config NR_QUICK
int
default 2

If an arch has requested quicklist support then pages can be allocated
from the quicklist (or from the page allocator if the quicklist is
empty) via:

quicklist_alloc(<quicklist-nr>, <gfpflags>, <constructor>)

Page table pages can be freed using:

quicklist_free(<quicklist-nr>, <destructor>, <page>)

Pages must have a definite state after allocation and before
they are freed. If no constructor is specified then pages
will be zeroed on allocation and must be zeroed before they are
freed.

If a constructor is used then the constructor will establish
a definite page state. F.e. the i386 and x86_64 pgd constructors
establish certain mappings.

Constructors and destructors can also be used to track the pages.
i386 and x86_64 use a list of pgds in order to be able to dynamically
update standard mappings.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Andi Kleen <ak@suse.de>
Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: William Lee Irwin III <wli@holomorphy.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoslub: add slabinfo tool
Christoph Lameter [Sun, 6 May 2007 21:49:48 +0000 (14:49 -0700)]
slub: add slabinfo tool

Add the tool which gets reports about slabs to the VM documentation directory.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoslub: user documentation
Christoph Lameter [Sun, 6 May 2007 21:49:47 +0000 (14:49 -0700)]
slub: user documentation

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoslub: remove object activities out of checking functions
Christoph Lameter [Sun, 6 May 2007 21:49:47 +0000 (14:49 -0700)]
slub: remove object activities out of checking functions

Make sure that the check function really only check things and do not perform
activities.  Extract the tracing and object seeding out of the two check
functions and place them into slab_alloc and slab_free

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoSLUB: Free slabs and sort partial slab lists in kmem_cache_shrink
Christoph Lameter [Sun, 6 May 2007 21:49:46 +0000 (14:49 -0700)]
SLUB: Free slabs and sort partial slab lists in kmem_cache_shrink

At kmem_cache_shrink check if we have any empty slabs on the partial
if so then remove them.

Also--as an anti-fragmentation measure--sort the partial slabs so that
the most fully allocated ones come first and the least allocated last.

The next allocations may fill up the nearly full slabs. Having the
least allocated slabs last gives them the maximum chance that their
remaining objects may be freed. Thus we can hopefully minimize the
partial slabs.

I think this is the best one can do in terms antifragmentation
measures. Real defragmentation (meaning moving objects out of slabs with
the least free objects to those that are almost full) can be implemted
by reverse scanning through the list produced here but that would mean
that we need to provide a callback at slab cache creation that allows
the deletion or moving of an object. This will involve slab API
changes, so defer for now.

Cc: Mel Gorman <mel@skynet.ie>
Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoslub: add ability to list alloc / free callers per slab
Christoph Lameter [Sun, 6 May 2007 21:49:45 +0000 (14:49 -0700)]
slub: add ability to list alloc / free callers per slab

This patch enables listing the callers who allocated or freed objects in a
cache.

For example to list the allocators for kmalloc-128 do

cat /sys/slab/kmalloc-128/alloc_calls
      7 sn_io_slot_fixup+0x40/0x700
      7 sn_io_slot_fixup+0x80/0x700
      9 sn_bus_fixup+0xe0/0x380
      6 param_sysfs_setup+0xf0/0x280
    276 percpu_populate+0xf0/0x1a0
     19 __register_chrdev_region+0x30/0x360
      8 expand_files+0x2e0/0x6e0
      1 sys_epoll_create+0x60/0x200
      1 __mounts_open+0x140/0x2c0
     65 kmem_alloc+0x110/0x280
      3 alloc_disk_node+0xe0/0x200
     33 as_get_io_context+0x90/0x280
     74 kobject_kset_add_dir+0x40/0x140
     12 pci_create_bus+0x2a0/0x5c0
      1 acpi_ev_create_gpe_block+0x120/0x9e0
     41 con_insert_unipair+0x100/0x1c0
      1 uart_open+0x1c0/0xba0
      1 dma_pool_create+0xe0/0x340
      2 neigh_table_init_no_netlink+0x260/0x4c0
      6 neigh_parms_alloc+0x30/0x200
      1 netlink_kernel_create+0x130/0x320
      5 fz_hash_alloc+0x50/0xe0
      2 sn_common_hubdev_init+0xd0/0x6e0
     28 kernel_param_sysfs_setup+0x30/0x180
     72 process_zones+0x70/0x2e0

cat /sys/slab/kmalloc-128/free_calls
    558 <not-available>
      3 sn_io_slot_fixup+0x600/0x700
     84 free_fdtable_rcu+0x120/0x260
      2 seq_release+0x40/0x60
      6 kmem_free+0x70/0xc0
     24 free_as_io_context+0x20/0x200
      1 acpi_get_object_info+0x3a0/0x3e0
      1 acpi_add_single_object+0xcf0/0x1e40
      2 con_release_unimap+0x80/0x140
      1 free+0x20/0x40

SLAB_STORE_USER must be enabled for a slab cache by either booting with
"slab_debug" or enabling user tracking specifically for the slab of interest.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoSLUB: Add MIN_PARTIAL
Christoph Lameter [Sun, 6 May 2007 21:49:44 +0000 (14:49 -0700)]
SLUB: Add MIN_PARTIAL

We leave a mininum of partial slabs on nodes when we search for
partial slabs on other node. Define a constant for that value.

Then modify slub to keep MIN_PARTIAL slabs around.

This avoids bad situations where a function frees the last object
in a slab (which results in the page being returned to the page
allocator) only to then allocate one again (which requires getting
a page back from the page allocator if the partial list was empty).
Keeping a couple of slabs on the partial list reduces overhead.

Empty slabs are added to the end of the partial list to insure that
partially allocated slabs are consumed first (defragmentation).

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoslub: validation of slabs (metadata and guard zones)
Christoph Lameter [Sun, 6 May 2007 21:49:43 +0000 (14:49 -0700)]
slub: validation of slabs (metadata and guard zones)

This enables validation of slab.  Validation means that all objects are
checked to see if there are redzone violations, if padding has been
overwritten or any pointers have been corrupted.  Also checks the consistency
of slab counters.

Validation enables the detection of metadata corruption without the kernel
having to execute code that actually uses (allocs/frees) and object.  It
allows one to make sure that the slab metainformation and the guard values
around an object have not been compromised.

A single slabcache can be checked by writing a 1 to the "validate" file.

i.e.

echo 1 >/sys/slab/kmalloc-128/validate

or use the slabinfo tool to check all slabs

slabinfo -v

Error messages will show up in the syslog.

Note that validation can only reach slabs that are on a list.  This means that
we are usually restricted to partial slabs and active slabs unless
SLAB_STORE_USER is active which will build a full slab list and allows
validation of slabs that are fully in use.  Booting with "slub_debug" set will
enable SLAB_STORE_USER and then full diagnostic are available.

Note that we attempt to push cpu slabs back to the lists when we start the
check.  If the cpu slab is reactivated before we get to it (another processor
grabs it before we get to it) then it cannot be checked.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoslub: enable tracking of full slabs
Christoph Lameter [Sun, 6 May 2007 21:49:42 +0000 (14:49 -0700)]
slub: enable tracking of full slabs

If slab tracking is on then build a list of full slabs so that we can verify
the integrity of all slabs and are also able to built list of alloc/free
callers.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoslub: fix object tracking
Christoph Lameter [Sun, 6 May 2007 21:49:42 +0000 (14:49 -0700)]
slub: fix object tracking

Object tracking did not work the right way for several call chains. Fix this up
by adding a new parameter to slub_alloc and slub_free that specifies the
caller address explicitly.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoAdd virt_to_head_page and consolidate code in slab and slub
Christoph Lameter [Sun, 6 May 2007 21:49:41 +0000 (14:49 -0700)]
Add virt_to_head_page and consolidate code in slab and slub

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agomm: optimize compound_head() by avoiding a shared page flag
Christoph Lameter [Sun, 6 May 2007 21:49:40 +0000 (14:49 -0700)]
mm: optimize compound_head() by avoiding a shared page flag

The patch adds PageTail(page) and PageHead(page) to check if a page is the
head or the tail of a compound page.  This is done by masking the two bits
describing the state of a compound page and then comparing them.  So one
comparision and a branch instead of two bit checks and two branches.

Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoMake page->private usable in compound pages
Christoph Lameter [Sun, 6 May 2007 21:49:39 +0000 (14:49 -0700)]
Make page->private usable in compound pages

If we add a new flag so that we can distinguish between the first page and the
tail pages then we can avoid to use page->private in the first page.
page->private == page for the first page, so there is no real information in
there.

Freeing up page->private makes the use of compound pages more transparent.
They become more usable like real pages.  Right now we have to be careful f.e.
 if we are going beyond PAGE_SIZE allocations in the slab on i386 because we
can then no longer use the private field.  This is one of the issues that
cause us not to support debugging for page size slabs in SLAB.

Having page->private available for SLUB would allow more meta information in
the page struct.  I can probably avoid the 16 bit ints that I have in there
right now.

Also if page->private is available then a compound page may be equipped with
buffer heads.  This may free up the way for filesystems to support larger
blocks than page size.

We add PageTail as an alias of PageReclaim.  Compound pages cannot currently
be reclaimed.  Because of the alias one needs to check PageCompound first.

The RFC for the this approach was discussed at
http://marc.info/?t=117574302800001&r=1&w=2

[nacc@us.ibm.com: fix hugetlbfs]
Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Nishanth Aravamudan <nacc@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoPowerPC: Disable SLUB for configurations in which slab page structs are modified
Christoph Lameter [Sun, 6 May 2007 21:49:38 +0000 (14:49 -0700)]
PowerPC: Disable SLUB for configurations in which slab page structs are modified

PowerPC uses the slab allocator to manage the lowest level of the page
table.  In high cpu configurations we also use the page struct to split the
page table lock.  Disallow the selection of SLUB for that case.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: Paul Mackerras <paulus@samba.org>
Acked-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoSLUB: allocate smallest object size if the user asks for 0 bytes
Christoph Lameter [Sun, 6 May 2007 21:49:38 +0000 (14:49 -0700)]
SLUB: allocate smallest object size if the user asks for 0 bytes

Makes SLUB behave like SLAB in this area to avoid issues....

Throw a stack dump to alert people.

At some point the behavior should be switched back.  NULL is no memory as
far as I can tell and if the use asked for 0 bytes then he need to get no
memory.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoSLUB: change default alignments
Christoph Lameter [Sun, 6 May 2007 21:49:37 +0000 (14:49 -0700)]
SLUB: change default alignments

Structures may contain u64 items on 32 bit platforms that are only able to
address 64 bit items on 64 bit boundaries.  Change the mininum alignment of
slabs to conform to those expectations.

ARCH_KMALLOC_MINALIGN must be changed for good since a variety of structure
are mixed in the general slabs.

ARCH_SLAB_MINALIGN is changed because currently there is no consistent
specification of object alignment.  We may have that in the future when the
KMEM_CACHE and related macros are used to generate slabs.  These pass the
alignment of the structure generated by the compiler to the slab.

With KMEM_CACHE etc we could align structures that do not contain 64
bit values to 32 bit boundaries potentially saving some memory.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoSLUB core
Christoph Lameter [Sun, 6 May 2007 21:49:36 +0000 (14:49 -0700)]
SLUB core

This is a new slab allocator which was motivated by the complexity of the
existing code in mm/slab.c. It attempts to address a variety of concerns
with the existing implementation.

A. Management of object queues

   A particular concern was the complex management of the numerous object
   queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for
   each allocating CPU and use objects from a slab directly instead of
   queueing them up.

B. Storage overhead of object queues

   SLAB Object queues exist per node, per CPU. The alien cache queue even
   has a queue array that contain a queue for each processor on each
   node. For very large systems the number of queues and the number of
   objects that may be caught in those queues grows exponentially. On our
   systems with 1k nodes / processors we have several gigabytes just tied up
   for storing references to objects for those queues  This does not include
   the objects that could be on those queues. One fears that the whole
   memory of the machine could one day be consumed by those queues.

C. SLAB meta data overhead

   SLAB has overhead at the beginning of each slab. This means that data
   cannot be naturally aligned at the beginning of a slab block. SLUB keeps
   all meta data in the corresponding page_struct. Objects can be naturally
   aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte
   boundaries and can fit tightly into a 4k page with no bytes left over.
   SLAB cannot do this.

D. SLAB has a complex cache reaper

   SLUB does not need a cache reaper for UP systems. On SMP systems
   the per CPU slab may be pushed back into partial list but that
   operation is simple and does not require an iteration over a list
   of objects. SLAB expires per CPU, shared and alien object queues
   during cache reaping which may cause strange hold offs.

E. SLAB has complex NUMA policy layer support

   SLUB pushes NUMA policy handling into the page allocator. This means that
   allocation is coarser (SLUB does interleave on a page level) but that
   situation was also present before 2.6.13. SLABs application of
   policies to individual slab objects allocated in SLAB is
   certainly a performance concern due to the frequent references to
   memory policies which may lead a sequence of objects to come from
   one node after another. SLUB will get a slab full of objects
   from one node and then will switch to the next.

F. Reduction of the size of partial slab lists

   SLAB has per node partial lists. This means that over time a large
   number of partial slabs may accumulate on those lists. These can
   only be reused if allocator occur on specific nodes. SLUB has a global
   pool of partial slabs and will consume slabs from that pool to
   decrease fragmentation.

G. Tunables

   SLAB has sophisticated tuning abilities for each slab cache. One can
   manipulate the queue sizes in detail. However, filling the queues still
   requires the uses of the spin lock to check out slabs. SLUB has a global
   parameter (min_slab_order) for tuning. Increasing the minimum slab
   order can decrease the locking overhead. The bigger the slab order the
   less motions of pages between per CPU and partial lists occur and the
   better SLUB will be scaling.

G. Slab merging

   We often have slab caches with similar parameters. SLUB detects those
   on boot up and merges them into the corresponding general caches. This
   leads to more effective memory use. About 50% of all caches can
   be eliminated through slab merging. This will also decrease
   slab fragmentation because partial allocated slabs can be filled
   up again. Slab merging can be switched off by specifying
   slub_nomerge on boot up.

   Note that merging can expose heretofore unknown bugs in the kernel
   because corrupted objects may now be placed differently and corrupt
   differing neighboring objects. Enable sanity checks to find those.

H. Diagnostics

   The current slab diagnostics are difficult to use and require a
   recompilation of the kernel. SLUB contains debugging code that
   is always available (but is kept out of the hot code paths).
   SLUB diagnostics can be enabled via the "slab_debug" option.
   Parameters can be specified to select a single or a group of
   slab caches for diagnostics. This means that the system is running
   with the usual performance and it is much more likely that
   race conditions can be reproduced.

I. Resiliency

   If basic sanity checks are on then SLUB is capable of detecting
   common error conditions and recover as best as possible to allow the
   system to continue.

J. Tracing

   Tracing can be enabled via the slab_debug=T,<slabcache> option
   during boot. SLUB will then protocol all actions on that slabcache
   and dump the object contents on free.

K. On demand DMA cache creation.

   Generally DMA caches are not needed. If a kmalloc is used with
   __GFP_DMA then just create this single slabcache that is needed.
   For systems that have no ZONE_DMA requirement the support is
   completely eliminated.

L. Performance increase

   Some benchmarks have shown speed improvements on kernbench in the
   range of 5-10%. The locking overhead of slub is based on the
   underlying base allocation size. If we can reliably allocate
   larger order pages then it is possible to increase slub
   performance much further. The anti-fragmentation patches may
   enable further performance increases.

Tested on:
i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator

SLUB Boot options

slub_nomerge Disable merging of slabs
slub_min_order=x Require a minimum order for slab caches. This
increases the managed chunk size and therefore
reduces meta data and locking overhead.
slub_min_objects=x Mininum objects per slab. Default is 8.
slub_max_order=x Avoid generating slabs larger than order specified.
slub_debug Enable all diagnostics for all caches
slub_debug=<options> Enable selective options for all caches
slub_debug=<o>,<cache> Enable selective options for a certain set of
caches

Available Debug options
F Double Free checking, sanity and resiliency
R Red zoning
P Object / padding poisoning
U Track last free / alloc
T Trace all allocs / frees (only use for individual slabs).

To use SLUB: Apply this patch and then select SLUB as the default slab
allocator.

[hugh@veritas.com: fix an oops-causing locking error]
[akpm@linux-foundation.org: various stupid cleanups and small fixes]
Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agotty_register_driver: only allocate tty instances when defined
Andy Whitcroft [Sun, 6 May 2007 21:49:33 +0000 (14:49 -0700)]
tty_register_driver: only allocate tty instances when defined

If device->num is zero we attempt to kmalloc() zero bytes.  When SLUB is
enabled this returns a null pointer and take that as an allocation failure
and fail the device register.  Check for no devices and avoid the
allocation.

[akpm: opportunistic kzalloc() conversion]
Signed-off-by: Andy Whitcroft <apw@shadowen.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoi386: use page allocator to allocate thread_info structure
Christoph Lameter [Sun, 6 May 2007 21:49:33 +0000 (14:49 -0700)]
i386: use page allocator to allocate thread_info structure

i386 uses kmalloc to allocate the threadinfo structure assuming that the
allocations result in a page sized aligned allocation.  That has worked so
far because SLAB exempts page sized slabs from debugging and aligns them in
special ways that goes beyond the restrictions imposed by
KMALLOC_ARCH_MINALIGN valid for other slabs in the kmalloc array.

SLUB also works fine without debugging since page sized allocations neatly
align at page boundaries.  However, if debugging is switched on then SLUB
will extend the slab with debug information.  The resulting slab is not
longer of page size.  It will only be aligned following the requirements
imposed by KMALLOC_ARCH_MINALIGN.  As a result the threadinfo structure may
not be page aligned which makes i386 fail to boot with SLUB debug on.

Replace the calls to kmalloc with calls into the page allocator.

An alternate solution may be to create a custom slab cache where the
alignment is set to PAGE_SIZE.  That would allow slub debugging to be
applied to the threadinfo structure.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Cc: William Lee Irwin III <wli@holomorphy.com>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agocpusets: allow TIF_MEMDIE threads to allocate anywhere
David Rientjes [Sun, 6 May 2007 21:49:32 +0000 (14:49 -0700)]
cpusets: allow TIF_MEMDIE threads to allocate anywhere

OOM killed tasks have access to memory reserves as specified by the
TIF_MEMDIE flag in the hopes that it will quickly exit.  If such a task has
memory allocations constrained by cpusets, we may encounter a deadlock if a
blocking task cannot exit because it cannot allocate the necessary memory.

We allow tasks that have the TIF_MEMDIE flag to allocate memory anywhere,
including outside its cpuset restriction, so that it can quickly die
regardless of whether it is __GFP_HARDWALL.

Cc: Andi Kleen <ak@suse.de>
Cc: Paul Jackson <pj@sgi.com>
Cc: Christoph Lameter <clameter@engr.sgi.com>
Signed-off-by: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoslab: mark set_up_list3s() __init
Andrew Morton [Sun, 6 May 2007 21:49:31 +0000 (14:49 -0700)]
slab: mark set_up_list3s() __init

It is only ever used prior to free_initmem().

(It will cause a warning when we run the section checking, but that's a
false-positive and it simply changes the source of an existing warning, which
is also a false-positive)

Cc: Christoph Lameter <clameter@engr.sgi.com>
Cc: Pekka Enberg <penberg@cs.helsinki.fi>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoDo not disable interrupts when reading min_free_kbytes
Mel Gorman [Sun, 6 May 2007 21:49:30 +0000 (14:49 -0700)]
Do not disable interrupts when reading min_free_kbytes

The sysctl handler for min_free_kbytes calls setup_per_zone_pages_min() on
read or write.  This function iterates through every zone and calls
spin_lock_irqsave() on the zone LRU lock.  When reading min_free_kbytes,
this is a total waste of time that disables interrupts on the local
processor.  It might even be noticable machines with large numbers of zones
if a process started constantly reading min_free_kbytes.

This patch only calls setup_per_zone_pages_min() only on write. Tested on
an x86 laptop and it did the right thing.

Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Acked-by: Christoph Lameter <clameter@engr.sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoslab: NUMA kmem_cache diet
Eric Dumazet [Sun, 6 May 2007 21:49:29 +0000 (14:49 -0700)]
slab: NUMA kmem_cache diet

Some NUMA machines have a big MAX_NUMNODES (possibly 1024), but fewer
possible nodes.  This patch dynamically sizes the 'struct kmem_cache' to
allocate only needed space.

I moved nodelists[] field at the end of struct kmem_cache, and use the
following computation in kmem_cache_init()

cache_cache.buffer_size = offsetof(struct kmem_cache, nodelists) +
                                 nr_node_ids * sizeof(struct kmem_list3 *);

On my two nodes x86_64 machine, kmem_cache.obj_size is now 192 instead of 704
(This is because on x86_64, MAX_NUMNODES is 64)

On bigger NUMA setups, this might reduce the gfporder of "cache_cache"

Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Cc: Pekka Enberg <penberg@cs.helsinki.fi>
Cc: Andy Whitcroft <apw@shadowen.org>
Cc: Christoph Lameter <clameter@engr.sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoSLAB: don't allocate empty shared caches
Eric Dumazet [Sun, 6 May 2007 21:49:28 +0000 (14:49 -0700)]
SLAB: don't allocate empty shared caches

We can avoid allocating empty shared caches and avoid unecessary check of
cache->limit.  We save some memory.  We avoid bringing into CPU cache
unecessary cache lines.

All accesses to l3->shared are already checking NULL pointers so this patch is
safe.

Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Acked-by: Pekka Enberg <penberg@cs.helsinki.fi>
Cc: Christoph Lameter <clameter@engr.sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoSLAB: use num_possible_cpus() in enable_cpucache()
Eric Dumazet [Sun, 6 May 2007 21:49:27 +0000 (14:49 -0700)]
SLAB: use num_possible_cpus() in enable_cpucache()

The existing comment in mm/slab.c is *perfect*, so I reproduce it :

         /*
          * CPU bound tasks (e.g. network routing) can exhibit cpu bound
          * allocation behaviour: Most allocs on one cpu, most free operations
          * on another cpu. For these cases, an efficient object passing between
          * cpus is necessary. This is provided by a shared array. The array
          * replaces Bonwick's magazine layer.
          * On uniprocessor, it's functionally equivalent (but less efficient)
          * to a larger limit. Thus disabled by default.
          */

As most shiped linux kernels are now compiled with CONFIG_SMP, there is no way
a preprocessor #if can detect if the machine is UP or SMP. Better to use
num_possible_cpus().

This means on UP we allocate a 'size=0 shared array', to be more efficient.

Another patch can later avoid the allocations of 'empty shared arrays', to
save some memory.

Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Acked-by: Pekka Enberg <penberg@cs.helsinki.fi>
Acked-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoreadahead: code cleanup
Jan Kara [Sun, 6 May 2007 21:49:26 +0000 (14:49 -0700)]
readahead: code cleanup

Rename file_ra_state.prev_page to prev_index and file_ra_state.offset to
prev_offset.  Also update of prev_index in do_generic_mapping_read() is now
moved close to the update of prev_offset.

[wfg@mail.ustc.edu.cn: fix it]
Signed-off-by: Jan Kara <jack@suse.cz>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: WU Fengguang <wfg@mail.ustc.edu.cn>
Signed-off-by: Fengguang Wu <wfg@mail.ustc.edu.cn>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoreadahead: improve heuristic detecting sequential reads
Jan Kara [Sun, 6 May 2007 21:49:25 +0000 (14:49 -0700)]
readahead: improve heuristic detecting sequential reads

Introduce ra.offset and store in it an offset where the previous read
ended.  This way we can detect whether reads are really sequential (and
thus we should not mark the page as accessed repeatedly) or whether they
are random and just happen to be in the same page (and the page should
really be marked accessed again).

Signed-off-by: Jan Kara <jack@suse.cz>
Acked-by: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: WU Fengguang <wfg@mail.ustc.edu.cn>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agosmaps: add clear_refs file to clear reference
David Rientjes [Sun, 6 May 2007 21:49:24 +0000 (14:49 -0700)]
smaps: add clear_refs file to clear reference

Adds /proc/pid/clear_refs.  When any non-zero number is written to this file,
pte_mkold() and ClearPageReferenced() is called for each pte and its
corresponding page, respectively, in that task's VMAs.  This file is only
writable by the user who owns the task.

It is now possible to measure _approximately_ how much memory a task is using
by clearing the reference bits with

echo 1 > /proc/pid/clear_refs

and checking the reference count for each VMA from the /proc/pid/smaps output
at a measured time interval.  For example, to observe the approximate change
in memory footprint for a task, write a script that clears the references
(echo 1 > /proc/pid/clear_refs), sleeps, and then greps for Pgs_Referenced and
extracts the size in kB.  Add the sizes for each VMA together for the total
referenced footprint.  Moments later, repeat the process and observe the
difference.

For example, using an efficient Mozilla:

accumulated time referenced memory
---------------- -----------------
 0 s  408 kB
 1 s  408 kB
 2 s  556 kB
 3 s 1028 kB
 4 s  872 kB
 5 s 1956 kB
 6 s  416 kB
 7 s 1560 kB
 8 s 2336 kB
 9 s 1044 kB
10 s  416 kB

This is a valuable tool to get an approximate measurement of the memory
footprint for a task.

Cc: Hugh Dickins <hugh@veritas.com>
Cc: Paul Mundt <lethal@linux-sh.org>
Cc: Christoph Lameter <clameter@sgi.com>
Signed-off-by: David Rientjes <rientjes@google.com>
[akpm@linux-foundation.org: build fixes]
[mpm@selenic.com: rename for_each_pmd]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agosmaps: add pages referenced count to smaps
David Rientjes [Sun, 6 May 2007 21:49:22 +0000 (14:49 -0700)]
smaps: add pages referenced count to smaps

Adds an additional unsigned long field to struct mem_size_stats called
'referenced'.  For each pte walked in the smaps code, this field is
incremented by PAGE_SIZE if it has pte-reference bits.

An additional line was added to the /proc/pid/smaps output for each VMA to
indicate how many pages within it are currently marked as referenced or
accessed.

Cc: Hugh Dickins <hugh@veritas.com>
Cc: Paul Mundt <lethal@linux-sh.org>
Cc: Christoph Lameter <clameter@sgi.com>
Signed-off-by: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agosmaps: extract pmd walker from smaps code
David Rientjes [Sun, 6 May 2007 21:49:21 +0000 (14:49 -0700)]
smaps: extract pmd walker from smaps code

Extracts the pmd walker from smaps-specific code in fs/proc/task_mmu.c.

The new struct pmd_walker includes the struct vm_area_struct of the memory to
walk over.  Iteration begins at the vma->vm_start and completes at
vma->vm_end.  A pointer to another data structure may be stored in the private
field such as struct mem_size_stats, which acts as the smaps accumulator.  For
each pmd in the VMA, the action function is called with a pointer to its
struct vm_area_struct, a pointer to the pmd_t, its start and end addresses,
and the private field.

The interface for walking pmd's in a VMA for fs/proc/task_mmu.c is now:

void for_each_pmd(struct vm_area_struct *vma,
  void (*action)(struct vm_area_struct *vma,
 pmd_t *pmd, unsigned long addr,
 unsigned long end,
 void *private),
  void *private);

Since the pmd walker is now extracted from the smaps code, smaps_one_pmd() is
invoked for each pmd in the VMA.  Its behavior and efficiency is identical to
the existing implementation.

Cc: Hugh Dickins <hugh@veritas.com>
Cc: Paul Mundt <lethal@linux-sh.org>
Cc: Christoph Lameter <clameter@sgi.com>
Signed-off-by: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoi386: use pte_update_defer in ptep_test_and_clear_{dirty,young}
Zachary Amsden [Sun, 6 May 2007 21:49:20 +0000 (14:49 -0700)]
i386: use pte_update_defer in ptep_test_and_clear_{dirty,young}

If you actually clear the bit, you need to:

+         pte_update_defer(vma->vm_mm, addr, ptep);

The reason is, when updating PTEs, the hypervisor must be notified.  Using
atomic operations to do this is fine for all hypervisors I am aware of.
However, for hypervisors which shadow page tables, if these PTE
modifications are not trapped, you need a post-modification call to fulfill
the update of the shadow page table.

Acked-by: Zachary Amsden <zach@vmware.com>
Cc: Hugh Dickins <hugh@veritas.com>
Signed-off-by: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoi386: add ptep_test_and_clear_{dirty,young}
David Rientjes [Sun, 6 May 2007 21:49:19 +0000 (14:49 -0700)]
i386: add ptep_test_and_clear_{dirty,young}

Add ptep_test_and_clear_{dirty,young} to i386.  They advertise that they
have it and there is at least one place where it needs to be called without
the page table lock: to clear the accessed bit on write to
/proc/pid/clear_refs.

ptep_clear_flush_{dirty,young} are updated to use the new functions.  The
overall net effect to current users of ptep_clear_flush_{dirty,young} is
that we introduce an additional branch.

Cc: Hugh Dickins <hugh@veritas.com>
Cc: Ingo Molnar <mingo@redhat.com>
Signed-off-by: David Rientjes <rientjes@google.com>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoAdd unitialized_var() macro for suppressing gcc warnings
Borislav Petkov [Sun, 6 May 2007 21:49:17 +0000 (14:49 -0700)]
Add unitialized_var() macro for suppressing gcc warnings

Introduce a macro for suppressing gcc from generating a warning about a
probable uninitialized state of a variable.

Example:

- spinlock_t *ptl;
+ spinlock_t *uninitialized_var(ptl);

Not a happy solution, but those warnings are obnoxious.

- Using the usual pointlessly-set-it-to-zero approach wastes several
  bytes of text.

- Using a macro means we can (hopefully) do something else if gcc changes
  cause the `x = x' hack to stop working

- Using a macro means that people who are worried about hiding true bugs
  can easily turn it off.

Signed-off-by: Borislav Petkov <bbpetkov@yahoo.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agomm: simplify filemap_nopage
Nick Piggin [Sun, 6 May 2007 21:49:16 +0000 (14:49 -0700)]
mm: simplify filemap_nopage

Identical block is duplicated twice: contrary to the comment, we have been
re-reading the page *twice* in filemap_nopage rather than once.

If any retry logic or anything is needed, it belongs in lower levels anyway.
Only retry once.  Linus agrees.

Signed-off-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoadd pfn_valid_within helper for sub-MAX_ORDER hole detection
Andy Whitcroft [Sun, 6 May 2007 21:49:14 +0000 (14:49 -0700)]
add pfn_valid_within helper for sub-MAX_ORDER hole detection

Generally we work under the assumption that memory the mem_map array is
contigious and valid out to MAX_ORDER_NR_PAGES block of pages, ie.  that if we
have validated any page within this MAX_ORDER_NR_PAGES block we need not check
any other.  This is not true when CONFIG_HOLES_IN_ZONE is set and we must
check each and every reference we make from a pfn.

Add a pfn_valid_within() helper which should be used when scanning pages
within a MAX_ORDER_NR_PAGES block when we have already checked the validility
of the block normally with pfn_valid().  This can then be optimised away when
we do not have holes within a MAX_ORDER_NR_PAGES block of pages.

Signed-off-by: Andy Whitcroft <apw@shadowen.org>
Acked-by: Mel Gorman <mel@csn.ul.ie>
Acked-by: Bob Picco <bob.picco@hp.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agomm/slab.c: proper prototypes
Adrian Bunk [Sun, 6 May 2007 21:49:12 +0000 (14:49 -0700)]
mm/slab.c: proper prototypes

Add proper prototypes in include/linux/slab.h.

Signed-off-by: Adrian Bunk <bunk@stusta.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoIntroduce CONFIG_HAS_DMA
Heiko Carstens [Sun, 6 May 2007 21:49:09 +0000 (14:49 -0700)]
Introduce CONFIG_HAS_DMA

Architectures that don't support DMA can say so by adding a config NO_DMA
to their Kconfig file.  This will prevent compilation of some dma specific
driver code.  Also dma-mapping-broken.h isn't needed anymore on at least
s390.  This avoids compilation and linking of otherwise dead/broken code.

Other architectures that include dma-mapping-broken.h are arm26, h8300,
m68k, m68knommu and v850.  If these could be converted as well we could get
rid of the header file.

Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
"John W. Linville" <linville@tuxdriver.com>
Cc: Kyle McMartin <kyle@parisc-linux.org>
Cc: <James.Bottomley@SteelEye.com>
Cc: Tejun Heo <htejun@gmail.com>
Cc: Jeff Garzik <jeff@garzik.org>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: <geert@linux-m68k.org>
Cc: <zippel@linux-m68k.org>
Cc: <spyro@f2s.com>
Cc: <uclinux-v850@lsi.nec.co.jp>
Cc: <ysato@users.sourceforge.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoallow oom_adj of saintly processes
Joshua N Pritikin [Sun, 6 May 2007 21:49:07 +0000 (14:49 -0700)]
allow oom_adj of saintly processes

If the badness of a process is zero then oom_adj>0 has no effect.  This
patch makes sure that the oom_adj shift actually increases badness points
appropriately.

Signed-off-by: Joshua N. Pritikin <jpritikin@pobox.com>
Cc: Andrea Arcangeli <andrea@novell.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agofs: buffer don't PageUptodate without page locked
Nick Piggin [Sun, 6 May 2007 21:49:05 +0000 (14:49 -0700)]
fs: buffer don't PageUptodate without page locked

__block_write_full_page is calling SetPageUptodate without the page locked.
This is unusual, but not incorrect, as PG_writeback is still set.

However the next patch will require that SetPageUptodate always be called with
the page locked.  Simply don't bother setting the page uptodate in this case
(it is unusual that the write path does such a thing anyway).  Instead just
leave it to the read side to bring the page uptodate when it notices that all
buffers are uptodate.

Signed-off-by: Nick Piggin <npiggin@suse.de>
Cc: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agomm: make read_cache_page synchronous
Nick Piggin [Sun, 6 May 2007 21:49:04 +0000 (14:49 -0700)]
mm: make read_cache_page synchronous

Ensure pages are uptodate after returning from read_cache_page, which allows
us to cut out most of the filesystem-internal PageUptodate calls.

I didn't have a great look down the call chains, but this appears to fixes 7
possible use-before uptodate in hfs, 2 in hfsplus, 1 in jfs, a few in
ecryptfs, 1 in jffs2, and a possible cleared data overwritten with readpage in
block2mtd.  All depending on whether the filler is async and/or can return
with a !uptodate page.

Signed-off-by: Nick Piggin <npiggin@suse.de>
Cc: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoslab: ensure cache_alloc_refill terminates
Pekka Enberg [Sun, 6 May 2007 21:49:03 +0000 (14:49 -0700)]
slab: ensure cache_alloc_refill terminates

If slab->inuse is corrupted, cache_alloc_refill can enter an infinite
loop as detailed by Michael Richardson in the following post:
<http://lkml.org/lkml/2007/2/16/292>. This adds a BUG_ON to catch
those cases.

Cc: Michael Richardson <mcr@sandelman.ca>
Acked-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agomm: remove gcc workaround
Nick Piggin [Sun, 6 May 2007 21:49:02 +0000 (14:49 -0700)]
mm: remove gcc workaround

Minimum gcc version is 3.2 now.  However, with likely profiling, even
modern gcc versions cannot always eliminate the call.

Replace the placeholder functions with the more conventional empty static
inlines, which should be optimal for everyone.

Signed-off-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoproper prototype for hugetlb_get_unmapped_area()
Adrian Bunk [Sun, 6 May 2007 21:49:00 +0000 (14:49 -0700)]
proper prototype for hugetlb_get_unmapped_area()

Add a proper prototype for hugetlb_get_unmapped_area() in
include/linux/hugetlb.h.

Signed-off-by: Adrian Bunk <bunk@stusta.de>
Acked-by: William Irwin <wli@holomorphy.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoUse ZVC counters to establish exact size of dirtyable pages
Christoph Lameter [Sun, 6 May 2007 21:48:59 +0000 (14:48 -0700)]
Use ZVC counters to establish exact size of dirtyable pages

We can use the global ZVC counters to establish the exact size of the LRU
and the free pages.  This allows a more accurate determination of the dirty
ratio.

This patch will fix the broken ratio calculations if large amounts of
memory are allocated to huge pags or other consumers that do not put the
pages on to the LRU.

Notes:
- I did not add NR_SLAB_RECLAIMABLE to the calculation of the
  dirtyable pages. Those may be reclaimable but they are at this
  point not dirtyable. If NR_SLAB_RECLAIMABLE would be considered
  then a huge number of reclaimable pages would stop writeback
  from occurring.

- This patch used to be in mm as the last one in a series of patches.
  It was removed when Linus updated the treatment of highmem because
  there was a conflict. I updated the patch to follow Linus' approach.
  This patch is neede to fulfill the claims made in the beginning of the
  patchset that is now in Linus' tree.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoSafer nr_node_ids and nr_node_ids determination and initial values
Christoph Lameter [Sun, 6 May 2007 21:48:58 +0000 (14:48 -0700)]
Safer nr_node_ids and nr_node_ids determination and initial values

The nr_cpu_ids value is currently only calculated in smp_init.  However, it
may be needed before (SLUB needs it on kmem_cache_init!) and other kernel
components may also want to allocate dynamically sized per cpu array before
smp_init.  So move the determination of possible cpus into sched_init()
where we already loop over all possible cpus early in boot.

Also initialize both nr_node_ids and nr_cpu_ids with the highest value they
could take.  If we have accidental users before these values are determined
then the current valud of 0 may cause too small per cpu and per node arrays
to be allocated.  If it is set to the maximum possible then we only waste
some memory for early boot users.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
17 years agoAdd apply_to_page_range() which applies a function to a pte range
Jeremy Fitzhardinge [Sun, 6 May 2007 21:48:54 +0000 (14:48 -0700)]
Add apply_to_page_range() which applies a function to a pte range

Add a new mm function apply_to_page_range() which applies a given function to
every pte in a given virtual address range in a given mm structure.  This is a
generic alternative to cut-and-pasting the Linux idiomatic pagetable walking
code in every place that a sequence of PTEs must be accessed.

Although this interface is intended to be useful in a wide range of
situations, it is currently used specifically by several Xen subsystems, for
example: to ensure that pagetables have been allocated for a virtual address
range, and to construct batched special pagetable update requests to map I/O
memory (in ioremap()).

[akpm@linux-foundation.org: fix warning, unpleasantly]
Signed-off-by: Ian Pratt <ian.pratt@xensource.com>
Signed-off-by: Christian Limpach <Christian.Limpach@cl.cam.ac.uk>
Signed-off-by: Chris Wright <chrisw@sous-sol.org>
Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Cc: Christoph Lameter <clameter@sgi.com>
Cc: Matt Mackall <mpm@waste.org>
Acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>