12 years agofs: introduce some page/buffer invariants
Nick Piggin [Tue, 17 Jul 2007 11:03:34 +0000]
fs: introduce some page/buffer invariants

It is a bug to set a page dirty if it is not uptodate unless it has
buffers.  If the page has buffers, then the page may be dirty (some buffers
dirty) but not uptodate (some buffers not uptodate).  The exception to this
rule is if the set_page_dirty caller is racing with truncate or invalidate.

A buffer can not be set dirty if it is not uptodate.

If either of these situations occurs, it indicates there could be some data
loss problem.  Some of these warnings could be a harmless one where the
page or buffer is set uptodate immediately after it is dirtied, however we
should fix those up, and enforce this ordering.

Bring the order of operations for truncate into line with those of
invalidate.  This will prevent a page from being able to go !uptodate while
we're holding the tree_lock, which is probably a good thing anyway.

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>

12 years agoMM: Make needlessly global hugetlb_no_page() static.
Robert P. J. Day [Tue, 17 Jul 2007 11:03:33 +0000]
MM: Make needlessly global hugetlb_no_page() static.

Signed-off-by: Robert P. J. Day <rpjday@mindspring.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

12 years agoAdd VM_BUG_ON in case someone uses page_mapping on a slab page
Christoph Lameter [Tue, 17 Jul 2007 11:03:33 +0000]
Add VM_BUG_ON in case someone uses page_mapping on a slab page

Detect slab objects being passed to the page oriented functions of the VM.

It is not sufficient to simply return NULL because the functions calling
page_mapping may depend on other items of the page_struct also to be setup
properly.  Moreover slab object may not be properly aligned.  The page
oriented functions of the VM expect to operate on page aligned, page sized
objects.  Operations on object straddling page boundaries may only affect the
objects partially which may lead to surprising results.

It is better to detect eventually remaining uses and eliminate them.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Cc: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

12 years agoMake SLUB the default allocator
Christoph Lameter [Tue, 17 Jul 2007 11:03:32 +0000]
Make SLUB the default allocator

There are some reports that 2.6.22 has SLUB as the default. Not
true!

This will make SLUB the default for 2.6.23.

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>

12 years agoSLUB: Fix CONFIG_SLUB_DEBUG use for CONFIG_NUMA
Christoph Lameter [Tue, 17 Jul 2007 11:03:32 +0000]
SLUB: Fix CONFIG_SLUB_DEBUG use for CONFIG_NUMA

We currently cannot disable CONFIG_SLUB_DEBUG for CONFIG_NUMA.  Now that
embedded systems start to use NUMA we may need this.

Put an #ifdef around places where NUMA only code uses fields only valid
for CONFIG_SLUB_DEBUG.

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>

12 years agoSLUB: Move sysfs operations outside of slub_lock
Christoph Lameter [Tue, 17 Jul 2007 11:03:31 +0000]
SLUB: Move sysfs operations outside of slub_lock

Sysfs can do a gazillion things when called.  Make sure that we do not call
any sysfs functions while holding the slub_lock.

Just protect the essentials:

1. The list of all slab caches
2. The kmalloc_dma array
3. The ref counters of the slabs.

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>

12 years agoSLUB: Do not allocate object bit array on stack
Christoph Lameter [Tue, 17 Jul 2007 11:03:30 +0000]
SLUB: Do not allocate object bit array on stack

The objects per slab increase with the current patches in mm since we allow up
to order 3 allocs by default.  More patches in mm actually allow to use 2M or
higher sized slabs.  For slab validation we need per object bitmaps in order
to check a slab.  We end up with up to 64k objects per slab resulting in a
potential requirement of 8K stack space.  That does not look good.

Allocate the bit arrays via kmalloc.

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>

12 years agoSlab allocators: Replace explicit zeroing with __GFP_ZERO
Christoph Lameter [Tue, 17 Jul 2007 11:03:29 +0000]
Slab allocators: Replace explicit zeroing with __GFP_ZERO

kmalloc_node() and kmem_cache_alloc_node() were not available in a zeroing
variant in the past.  But with __GFP_ZERO it is possible now to do zeroing
while allocating.

Use __GFP_ZERO to remove the explicit clearing of memory via memset whereever
we can.

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>

12 years agoSlab allocators: Cleanup zeroing allocations
Christoph Lameter [Tue, 17 Jul 2007 11:03:29 +0000]
Slab allocators: Cleanup zeroing allocations

It becomes now easy to support the zeroing allocs with generic inline
functions in slab.h.  Provide inline definitions to allow the continued use of
kzalloc, kmem_cache_zalloc etc but remove other definitions of zeroing
functions from the slab allocators and util.c.

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>

12 years agoSLUB: Do not use length parameter in slab_alloc()
Christoph Lameter [Tue, 17 Jul 2007 11:03:28 +0000]
SLUB: Do not use length parameter in slab_alloc()

We can get to the length of the object through the kmem_cache_structure.  The
additional parameter does no good and causes the compiler to generate bad
code.

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>

12 years agoSLUB: Style fix up the loop to disable small slabs
Christoph Lameter [Tue, 17 Jul 2007 11:03:28 +0000]
SLUB: Style fix up the loop to disable small slabs

Do proper spacing and we only need to do this in steps of 8.

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>

12 years agomm/slub.c: make code static
Adrian Bunk [Tue, 17 Jul 2007 11:03:27 +0000]
mm/slub.c: make code static

Signed-off-by: Adrian Bunk <bunk@stusta.de>
Cc: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

12 years agoSLUB: Simplify dma index -> size calculation
Christoph Lameter [Tue, 17 Jul 2007 11:03:27 +0000]
SLUB: Simplify dma index -> size calculation

There is no need to caculate the dma slab size ourselves. We can simply
lookup the size of the corresponding non dma slab.

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>

12 years agoSLUB: faster more efficient slab determination for __kmalloc
Christoph Lameter [Tue, 17 Jul 2007 11:03:26 +0000]
SLUB: faster more efficient slab determination for __kmalloc

kmalloc_index is a long series of comparisons.  The attempt to replace
kmalloc_index with something more efficient like ilog2 failed due to compiler
issues with constant folding on gcc 3.3 / powerpc.

kmalloc_index()'es long list of comparisons works fine for constant folding
since all the comparisons are optimized away.  However, SLUB also uses
kmalloc_index to determine the slab to use for the __kmalloc_xxx functions.
This leads to a large set of comparisons in get_slab().

The patch here allows to get rid of that list of comparisons in get_slab():

1. If the requested size is larger than 192 then we can simply use
   fls to determine the slab index since all larger slabs are
   of the power of two type.

2. If the requested size is smaller then we cannot use fls since there
   are non power of two caches to be considered. However, the sizes are
   in a managable range. So we divide the size by 8. Then we have only
   24 possibilities left and then we simply look up the kmalloc index
   in a table.

Code size of slub.o decreases by more than 200 bytes through this patch.

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>

12 years agoSLUB: do proper locking during dma slab creation
Christoph Lameter [Tue, 17 Jul 2007 11:03:25 +0000]
SLUB: do proper locking during dma slab creation

We modify the kmalloc_cache_dma[] array without proper locking.  Do the proper
locking and undo the dma cache creation if another processor has already
created it.

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>

12 years agoSLUB: extract dma_kmalloc_cache from get_cache.
Christoph Lameter [Tue, 17 Jul 2007 11:03:24 +0000]
SLUB: extract dma_kmalloc_cache from get_cache.

The rarely used dma functionality in get_slab() makes the function too
complex.  The compiler begins to spill variables from the working set onto the
stack.  The created function is only used in extremely rare cases so make sure
that the compiler does not decide on its own to merge it back into get_slab().

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>

12 years agoSLUB: add some more inlines and #ifdef CONFIG_SLUB_DEBUG
Christoph Lameter [Tue, 17 Jul 2007 11:03:24 +0000]
SLUB: add some more inlines and #ifdef CONFIG_SLUB_DEBUG

Add #ifdefs around data structures only needed if debugging is compiled into
SLUB.

Add inlines to small functions to reduce code size.

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>

12 years agoSlab allocators: support __GFP_ZERO in all allocators
Christoph Lameter [Tue, 17 Jul 2007 11:03:23 +0000]
Slab allocators: support __GFP_ZERO in all allocators

A kernel convention for many allocators is that if __GFP_ZERO is passed to an
allocator then the allocated memory should be zeroed.

This is currently not supported by the slab allocators.  The inconsistency
makes it difficult to implement in derived allocators such as in the uncached
allocator and the pool allocators.

In addition the support zeroed allocations in the slab allocators does not
have a consistent API.  There are no zeroing allocator functions for NUMA node
placement (kmalloc_node, kmem_cache_alloc_node).  The zeroing allocations are
only provided for default allocs (kzalloc, kmem_cache_zalloc_node).
__GFP_ZERO will make zeroing universally available and does not require any
addititional functions.

So add the necessary logic to all slab allocators to support __GFP_ZERO.

The code is added to the hot path.  The gfp flags are on the stack and so the
cacheline is readily available for checking if we want a zeroed object.

Zeroing while allocating is now a frequent operation and we seem to be
gradually approaching a 1-1 parity between zeroing and not zeroing allocs.
The current tree has 3476 uses of kmalloc vs 2731 uses of kzalloc.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Acked-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>

12 years agoSlab allocators: consistent ZERO_SIZE_PTR support and NULL result semantics
Christoph Lameter [Tue, 17 Jul 2007 11:03:22 +0000]
Slab allocators: consistent ZERO_SIZE_PTR support and NULL result semantics

Define ZERO_OR_NULL_PTR macro to be able to remove the checks from the
allocators.  Move ZERO_SIZE_PTR related stuff into slab.h.

Make ZERO_SIZE_PTR work for all slab allocators and get rid of the
WARN_ON_ONCE(size == 0) that is still remaining in SLAB.

Make slub return NULL like the other allocators if a too large memory segment
is requested via __kmalloc.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Acked-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>

12 years agoSlab allocators: consolidate code for krealloc in mm/util.c
Christoph Lameter [Tue, 17 Jul 2007 11:03:21 +0000]
Slab allocators: consolidate code for krealloc in mm/util.c

The size of a kmalloc object is readily available via ksize().  ksize is
provided by all allocators and thus we can implement krealloc in a generic
way.

Implement krealloc in mm/util.c and drop slab specific implementations of
krealloc.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Acked-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>

12 years agoSLUB Debug: fix initial object debug state of NUMA bootstrap objects
Christoph Lameter [Tue, 17 Jul 2007 11:03:21 +0000]
SLUB Debug: fix initial object debug state of NUMA bootstrap objects

The function we are calling to initialize object debug state during early NUMA
bootstrap sets up an inactive object giving it the wrong redzone signature.
The bootstrap nodes are active objects and should have active redzone
signatures.

Currently slab validation complains and reverts the object to active state.

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>

12 years agoSLUB: ensure that the number of objects per slab stays low for high orders
Christoph Lameter [Tue, 17 Jul 2007 11:03:20 +0000]
SLUB: ensure that the number of objects per slab stays low for high orders

Currently SLUB has no provision to deal with too high page orders that may
be specified on the kernel boot line.  If an order higher than 6 (on a 4k
platform) is generated then we will BUG() because slabs get more than 65535
objects.

Add some logic that decreases order for slabs that have too many objects.
This allow booting with slab sizes up to MAX_ORDER.

For example

slub_min_order=10

will boot with a default slab size of 4M and reduce slab sizes for small
object sizes to lower orders if the number of objects becomes too big.
Large slab sizes like that allow a concentration of objects of the same
slab cache under as few as possible TLB entries and thus potentially
reduces TLB pressure.

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>

12 years agoSLUB slab validation: Move tracking information alloc outside of lock
Christoph Lameter [Tue, 17 Jul 2007 11:03:20 +0000]
SLUB slab validation: Move tracking information alloc outside of lock

We currently have to do an GFP_ATOMIC allocation because the list_lock is
already taken when we first allocate memory for tracking allocation
information.  It would be better if we could avoid atomic allocations.

Allocate a size of the tracking table that is usually sufficient (one page)
before we take the list lock.  We will then only do the atomic allocation
if we need to resize the table to become larger than a page (mostly only
needed under large NUMA because of the tracking of cpus and nodes otherwise
the table stays small).

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>

12 years agoSLUB: use list_for_each_entry for loops over all slabs
Christoph Lameter [Tue, 17 Jul 2007 11:03:19 +0000]
SLUB: use list_for_each_entry for loops over all slabs

Use list_for_each_entry() instead of list_for_each().

Get rid of for_all_slabs(). It had only one user. So fold it into the
callback. This also gets rid of cpu_slab_flush.

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>

12 years agoSLUB: change error reporting format to follow lockdep loosely
Christoph Lameter [Tue, 17 Jul 2007 11:03:18 +0000]
SLUB: change error reporting format to follow lockdep loosely

Changes the error reporting format to loosely follow lockdep.

If data corruption is detected then we generate the following lines:

============================================
BUG <slab-cache>: <problem>
--------------------------------------------

INFO: <more information> [possibly multiple times]

<object dump>

FIX <slab-cache>: <remedial action>

This also adds some more intelligence to the data corruption detection. Its
now capable of figuring out the start and end.

Add a comment on how to configure SLUB so that a production system may
continue to operate even though occasional slab corruption occur through
a misbehaving kernel component. See "Emergency operations" in
Documentation/vm/slub.txt.

[akpm@linux-foundation.org: build fix]
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>

12 years agomm: clean up and kernelify shrinker registration
Rusty Russell [Tue, 17 Jul 2007 11:03:17 +0000]
mm: clean up and kernelify shrinker registration

I can never remember what the function to register to receive VM pressure
is called.  I have to trace down from __alloc_pages() to find it.

It's called "set_shrinker()", and it needs Your Help.

1) Don't hide struct shrinker.  It contains no magic.
2) Don't allocate "struct shrinker".  It's not helpful.
3) Call them "register_shrinker" and "unregister_shrinker".
4) Call the function "shrink" not "shrinker".
5) Reduce the 17 lines of waffly comments to 13, but document it properly.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Cc: David Chinner <dgc@sgi.com>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

12 years agoLumpy Reclaim V4
Andy Whitcroft [Tue, 17 Jul 2007 11:03:16 +0000]
Lumpy Reclaim V4

When we are out of memory of a suitable size we enter reclaim.  The current
reclaim algorithm targets pages in LRU order, which is great for fairness at
order-0 but highly unsuitable if you desire pages at higher orders.  To get
pages of higher order we must shoot down a very high proportion of memory;
>95% in a lot of cases.

This patch set adds a lumpy reclaim algorithm to the allocator.  It targets
groups of pages at the specified order anchored at the end of the active and
inactive lists.  This encourages groups of pages at the requested orders to
move from active to inactive, and active to free lists.  This behaviour is
only triggered out of direct reclaim when higher order pages have been
requested.

This patch set is particularly effective when utilised with an
anti-fragmentation scheme which groups pages of similar reclaimability
together.

This patch set is based on Peter Zijlstra's lumpy reclaim V2 patch which forms
the foundation.  Credit to Mel Gorman for sanitity checking.

Mel said:

  The patches have an application with hugepage pool resizing.

  When lumpy-reclaim is used used with ZONE_MOVABLE, the hugepages pool can
  be resized with greater reliability.  Testing on a desktop machine with 2GB
  of RAM showed that growing the hugepage pool with ZONE_MOVABLE on it's own
  was very slow as the success rate was quite low.  Without lumpy-reclaim,
  each attempt to grow the pool by 100 pages would yield 1 or 2 hugepages.
  With lumpy-reclaim, getting 40 to 70 hugepages on each attempt was typical.

[akpm@osdl.org: ia64 pfn_to_nid fixes and loop cleanup]
[bunk@stusta.de: static declarations for internal functions]
[a.p.zijlstra@chello.nl: initial lumpy V2 implementation]
Signed-off-by: Andy Whitcroft <apw@shadowen.org>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Mel Gorman <mel@csn.ul.ie>
Acked-by: Mel Gorman <mel@csn.ul.ie>
Cc: Bob Picco <bob.picco@hp.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

12 years agoAdd a movablecore= parameter for sizing ZONE_MOVABLE
Mel Gorman [Tue, 17 Jul 2007 11:03:15 +0000]
Add a movablecore= parameter for sizing ZONE_MOVABLE

This patch adds a new parameter for sizing ZONE_MOVABLE called
movablecore=.  While kernelcore= is used to specify the minimum amount of
memory that must be available for all allocation types, movablecore= is
used to specify the minimum amount of memory that is used for migratable
allocations.  The amount of memory used for migratable allocations
determines how large the huge page pool could be dynamically resized to at
runtime for example.

How movablecore is actually handled is that the total number of pages in
the system is calculated and a value is set for kernelcore that is

kernelcore == totalpages - movablecore

Both kernelcore= and movablecore= can be safely specified at the same time.

Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Acked-by: Andy Whitcroft <apw@shadowen.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

12 years agohandle kernelcore=: generic
Mel Gorman [Tue, 17 Jul 2007 11:03:14 +0000]
handle kernelcore=: generic

This patch adds the kernelcore= parameter for x86.

Once all patches are applied, a new command-line parameter exist and a new
sysctl.  This patch adds the necessary documentation.

From: Yasunori Goto <y-goto@jp.fujitsu.com>

  When "kernelcore" boot option is specified, kernel can't boot up on ia64
  because of an infinite loop.  In addition, the parsing code can be handled
  in an architecture-independent manner.

  This patch uses common code to handle the kernelcore= parameter.  It is
  only available to architectures that support arch-independent zone-sizing
  (i.e.  define CONFIG_ARCH_POPULATES_NODE_MAP).  Other architectures will
  ignore the boot parameter.

[bunk@stusta.de: make cmdline_parse_kernelcore() static]
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Yasunori Goto <y-goto@jp.fujitsu.com>
Acked-by: Andy Whitcroft <apw@shadowen.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

12 years agoAllow huge page allocations to use GFP_HIGH_MOVABLE
Mel Gorman [Tue, 17 Jul 2007 11:03:13 +0000]
Allow huge page allocations to use GFP_HIGH_MOVABLE

Huge pages are not movable so are not allocated from ZONE_MOVABLE.  However,
as ZONE_MOVABLE will always have pages that can be migrated or reclaimed, it
can be used to satisfy hugepage allocations even when the system has been
running a long time.  This allows an administrator to resize the hugepage pool
at runtime depending on the size of ZONE_MOVABLE.

This patch adds a new sysctl called hugepages_treat_as_movable.  When a
non-zero value is written to it, future allocations for the huge page pool
will use ZONE_MOVABLE.  Despite huge pages being non-movable, we do not
introduce additional external fragmentation of note as huge pages are always
the largest contiguous block we care about.

[akpm@linux-foundation.org: various fixes]
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

12 years agoCreate the ZONE_MOVABLE zone
Mel Gorman [Tue, 17 Jul 2007 11:03:12 +0000]
Create the ZONE_MOVABLE zone

The following 8 patches against 2.6.20-mm2 create a zone called ZONE_MOVABLE
that is only usable by allocations that specify both __GFP_HIGHMEM and
__GFP_MOVABLE.  This has the effect of keeping all non-movable pages within a
single memory partition while allowing movable allocations to be satisfied
from either partition.  The patches may be applied with the list-based
anti-fragmentation patches that groups pages together based on mobility.

The size of the zone is determined by a kernelcore= parameter specified at
boot-time.  This specifies how much memory is usable by non-movable
allocations and the remainder is used for ZONE_MOVABLE.  Any range of pages
within ZONE_MOVABLE can be released by migrating the pages or by reclaiming.

When selecting a zone to take pages from for ZONE_MOVABLE, there are two
things to consider.  First, only memory from the highest populated zone is
used for ZONE_MOVABLE.  On the x86, this is probably going to be ZONE_HIGHMEM
but it would be ZONE_DMA on ppc64 or possibly ZONE_DMA32 on x86_64.  Second,
the amount of memory usable by the kernel will be spread evenly throughout
NUMA nodes where possible.  If the nodes are not of equal size, the amount of
memory usable by the kernel on some nodes may be greater than others.

By default, the zone is not as useful for hugetlb allocations because they are
pinned and non-migratable (currently at least).  A sysctl is provided that
allows huge pages to be allocated from that zone.  This means that the huge
page pool can be resized to the size of ZONE_MOVABLE during the lifetime of
the system assuming that pages are not mlocked.  Despite huge pages being
non-movable, we do not introduce additional external fragmentation of note as
huge pages are always the largest contiguous block we care about.

Credit goes to Andy Whitcroft for catching a large variety of problems during
review of the patches.

This patch creates an additional zone, ZONE_MOVABLE.  This zone is only usable
by allocations which specify both __GFP_HIGHMEM and __GFP_MOVABLE.  Hot-added
memory continues to be placed in their existing destination as there is no
mechanism to redirect them to a specific zone.

[y-goto@jp.fujitsu.com: Fix section mismatch of memory hotplug related code]
[akpm@linux-foundation.org: various fixes]
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Cc: Andy Whitcroft <apw@shadowen.org>
Signed-off-by: Yasunori Goto <y-goto@jp.fujitsu.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>

12 years agoAdd __GFP_MOVABLE for callers to flag allocations from high memory that may be migrated
Mel Gorman [Tue, 17 Jul 2007 11:03:05 +0000]
Add __GFP_MOVABLE for callers to flag allocations from high memory that may be migrated

It is often known at allocation time whether a page may be migrated or not.
This patch adds a flag called __GFP_MOVABLE and a new mask called
GFP_HIGH_MOVABLE.  Allocations using the __GFP_MOVABLE can be either migrated
using the page migration mechanism or reclaimed by syncing with backing
storage and discarding.

An API function very similar to alloc_zeroed_user_highpage() is added for
__GFP_MOVABLE allocations called alloc_zeroed_user_highpage_movable().  The
flags used by alloc_zeroed_user_highpage() are not changed because it would
change the semantics of an existing API.  After this patch is applied there
are no in-kernel users of alloc_zeroed_user_highpage() so it probably should
be marked deprecated if this patch is merged.

Note that this patch includes a minor cleanup to the use of __GFP_ZERO in
shmem.c to keep all flag modifications to inode->mapping in the
shmem_dir_alloc() helper function.  This clean-up suggestion is courtesy of
Hugh Dickens.

Additional credit goes to Christoph Lameter and Linus Torvalds for shaping the
concept.  Credit to Hugh Dickens for catching issues with shmem swap vector
and ramfs allocations.

[akpm@linux-foundation.org: build fix]
[hugh@veritas.com: __GFP_ZERO cleanup]
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Cc: Andy Whitcroft <apw@shadowen.org>
Cc: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

12 years agoFix read/truncate race
NeilBrown [Tue, 17 Jul 2007 11:03:04 +0000]
Fix read/truncate race

do_generic_mapping_read currently samples the i_size at the start and doesn't
do so again unless it needs to call ->readpage to load a page.  After
->readpage it has to re-sample i_size as a truncate may have caused that page
to be filled with zeros, and the read() call should not see these.

However there are other activities that might cause ->readpage to be called on
a page between the time that do_generic_mapping_read samples i_size and when
it finds that it has an uptodate page.  These include at least read-ahead and
possibly another thread performing a read.

So do_generic_mapping_read must sample i_size *after* it has an uptodate page.
 Thus the current sampling at the start and after a read can be replaced with
a sampling before the copy-out.

The same change applied to __generic_file_splice_read.

Note that this fixes any race with truncate_complete_page, but does not fix a
possible race with truncate_partial_page.  If a partial truncate happens after
do_generic_mapping_read samples i_size and before the copy_out, the nuls that
truncate_partial_page place in the page could be copied out incorrectly.

I think the best fix for that is to *not* zero out parts of the page in
truncate_partial_page, but rather to zero out the tail of a page when
increasing i_size.

Signed-off-by: Neil Brown <neilb@suse.de>
Cc: Jens Axboe <jens.axboe@oracle.com>
Acked-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

12 years agomm: remove ptep_test_and_clear_dirty and ptep_clear_flush_dirty
Martin Schwidefsky [Tue, 17 Jul 2007 11:03:03 +0000]
mm: remove ptep_test_and_clear_dirty and ptep_clear_flush_dirty

Nobody is using ptep_test_and_clear_dirty and ptep_clear_flush_dirty.  Remove
the functions from all architectures.

Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

12 years agomm: remove ptep_establish()
Martin Schwidefsky [Tue, 17 Jul 2007 11:03:03 +0000]
mm: remove ptep_establish()

The last user of ptep_establish in mm/ is long gone.  Remove the architecture
primitive as well.

Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

12 years agoparse error, drivers/i2c/busses/i2c-pmcmsp.c
Yoann Padioleau [Tue, 17 Jul 2007 11:03:01 +0000]
parse error, drivers/i2c/busses/i2c-pmcmsp.c

Signed-off-by: Yoann Padioleau <padator@wanadoo.fr>
Cc: Jean Delvare <khali@linux-fr.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

12 years agoMerge branch 'drm-patches' of ssh://master.kernel.org/pub/scm/linux/kernel/git/airlie...
Linus Torvalds [Tue, 17 Jul 2007 15:44:27 +0000]
Merge branch 'drm-patches' of ssh:///linux/kernel/git/airlied/drm-2.6

* 'drm-patches' of ssh://master.kernel.org/pub/scm/linux/kernel/git/airlied/drm-2.6:
  drm: add idr_init to drm_stub.c
  drm: fix problem with SiS typedef with sisfb enabled.

12 years agodrm: add idr_init to drm_stub.c
Dave Airlie [Tue, 17 Jul 2007 04:20:07 +0000]
drm: add idr_init to drm_stub.c

Brown paper bag for me this patch chunk didn't make it in the first application

Signed-off-by: Dave Airlie <airlied@linux.ie>

12 years agodrm: fix problem with SiS typedef with sisfb enabled.
Dave Airlie [Tue, 17 Jul 2007 02:55:58 +0000]
drm: fix problem with SiS typedef with sisfb enabled.

Reported by: Avuton Olrich <avuton@gmail.com>

Signed-off-by: Dave Airlie <airlied@linux.ie>

12 years agoMerge branch 'drm-patches' of ssh://master.kernel.org/pub/scm/linux/kernel/git/airlie...
Linus Torvalds [Tue, 17 Jul 2007 01:24:37 +0000]
Merge branch 'drm-patches' of ssh:///linux/kernel/git/airlied/drm-2.6

* 'drm-patches' of ssh://master.kernel.org/pub/scm/linux/kernel/git/airlied/drm-2.6:
  drm: convert drawable code to using idr
  drm: convert drm context code to use Linux idr

12 years agodrm: convert drawable code to using idr
Dave Airlie [Tue, 17 Jul 2007 00:55:47 +0000]
drm: convert drawable code to using idr

This converts the code for allocating drawables to the Linux idr,

Fixes from: Michel Dänzer <michel@tungstengraphics.com>, Kristian Høgsberg <krh@redhat.com>

Signed-off-by: Dave Airlie <airlied@linux.ie>

12 years agodrm: convert drm context code to use Linux idr
Dave Airlie [Tue, 17 Jul 2007 00:46:52 +0000]
drm: convert drm context code to use Linux idr

This converts the drm context allocator to an idr, using the new idr
interface features from Kristian.

Fixes from Kristian Hoegsberg <krh@redhat.com>

Signed-off-by: Dave Airlie <airlied@linux.ie>

12 years agoMerge branch 'merge' of git://git.kernel.org/pub/scm/linux/kernel/git/paulus/powerpc
Linus Torvalds [Tue, 17 Jul 2007 00:58:08 +0000]
Merge branch 'merge' of git://git./linux/kernel/git/paulus/powerpc

* 'merge' of git://git.kernel.org/pub/scm/linux/kernel/git/paulus/powerpc: (209 commits)
  [POWERPC] Create add_rtc() function to enable the RTC CMOS driver
  [POWERPC] Add H_ILLAN_ATTRIBUTES hcall number
  [POWERPC] xilinxfb: Parameterize xilinxfb platform device registration
  [POWERPC] Oprofile support for Power 5++
  [POWERPC] Enable arbitary speed tty ioctls and split input/output speed
  [POWERPC] Make drivers/char/hvc_console.c:khvcd() static
  [POWERPC] Remove dead code for preventing pread() and pwrite() calls
  [POWERPC] Remove unnecessary #undef printk from prom.c
  [POWERPC] Fix typo in Ebony default DTS
  [POWERPC] Check for NULL ppc_md.init_IRQ() before calling
  [POWERPC] Remove extra return statement
  [POWERPC] pasemi: Don't auto-select CONFIG_EMBEDDED
  [POWERPC] pasemi: Rename platform
  [POWERPC] arch/powerpc/kernel/sysfs.c: Move NUMA exports
  [POWERPC] Add __read_mostly support for powerpc
  [POWERPC] Modify sched_clock() to make CONFIG_PRINTK_TIME more sane
  [POWERPC] Create a dummy zImage if no valid platform has been selected
  [POWERPC] PS3: Bootwrapper support.
  [POWERPC] powermac i2c: Use mutex
  [POWERPC] Schedule removal of arch/ppc
  ...

Fixed up conflicts manually in:

Documentation/feature-removal-schedule.txt
arch/powerpc/kernel/pci_32.c
arch/powerpc/kernel/pci_64.c
include/asm-powerpc/pci.h

and asked the powerpc people to double-check the result..

12 years agoMerge branch 'upstream-linus' of master.kernel.org:/pub/scm/linux/kernel/git/jgarzik...
Linus Torvalds [Tue, 17 Jul 2007 00:48:54 +0000]
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: (37 commits)
  forcedeth bug fix: realtek phy
  forcedeth bug fix: vitesse phy
  forcedeth bug fix: cicada phy
  atl1: reorder atl1_main functions
  atl1: fix excessively indented code
  atl1: cleanup atl1_main
  atl1: header file cleanup
  atl1: remove irq_sem
  cdc-subset to support new vendor/product ID
  8139cp: implement the missing dev->tx_timeout
  myri10ge: Remove nonsensical limit in the tx done routine
  gianfar: kill unused header
  EP93XX_ETH must select MII
  macb: Add multicast capability
  macb: Use generic PHY layer
  s390: add barriers to qeth driver
  s390: scatter-gather for inbound traffic in qeth driver
  eHEA: Introducing support vor DLPAR memory add
  Fix a potential NULL pointer dereference in free_shared_mem() in drivers/net/s2io.c
  [PATCH] softmac: Fix ESSID problem
  ...

12 years agoMerge branch 'master' of master.kernel.org:/pub/scm/linux/kernel/git/davem/sparc-2.6
Linus Torvalds [Tue, 17 Jul 2007 00:33:17 +0000]
Merge branch 'master' of /linux/kernel/git/davem/sparc-2.6

* 'master' of master.kernel.org:/pub/scm/linux/kernel/git/davem/sparc-2.6:
  [SERIAL] SUNHV: Fix jerky console on LDOM guests.
  [SPARC64]: Fix race between MD update and dr-cpu add.
  [SPARC64]: SMP build fix.

12 years ago[HRTIMER] Fix cpu pointer arg to clockevents_notify()
David Miller [Tue, 17 Jul 2007 00:17:44 +0000]
[HRTIMER] Fix cpu pointer arg to clockevents_notify()

All of the clockevent notifiers expect a pointer to
an "unsigned int" cpu argument, but hrtimer_cpu_notify()
passes in a pointer to a long.

[ Discussed with and ok by Thomas Gleixner ]

Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

12 years ago[SERIAL] SUNHV: Fix jerky console on LDOM guests.
David S. Miller [Tue, 17 Jul 2007 00:05:11 +0000]
[SERIAL] SUNHV: Fix jerky console on LDOM guests.

Mixing putchar() and write() hvcalls does not work %100
correctly.  But we should be using write() all the time
if we can, even from ->start_tx(), anyways.

Signed-off-by: David S. Miller <davem@davemloft.net>

12 years ago[SPARC64]: Fix race between MD update and dr-cpu add.
David S. Miller [Mon, 16 Jul 2007 23:50:36 +0000]
[SPARC64]: Fix race between MD update and dr-cpu add.

We need to make sure the MD update occurs before we try to
process dr-cpu configure requests.  MD update and dr-cpu
were being processed by seperate threads so that did not
happen occaisionally.

Fix this by executing all domain services data packets from
a single thread, in order.

This will help simplify some other things as well.

Signed-off-by: David S. Miller <davem@davemloft.net>

12 years ago[SPARC64]: SMP build fix.
Fabio Massimo Di Nitto [Mon, 16 Jul 2007 21:15:39 +0000]
[SPARC64]: SMP build fix.

The UP build fix had some unintended consequences.

Signed-off-by: David S. Miller <davem@davemloft.net>

12 years agoMake BLK_DEV_BSG depend strictly on SCSI=y
Linus Torvalds [Mon, 16 Jul 2007 23:52:44 +0000]
Make BLK_DEV_BSG depend strictly on SCSI=y

The SCSI code can be compiled modular, but BLK_DEV_BSG currently cannot,
and depends on the SCSI layer.  So make sure that it depends on the SCSI
layer being compiled in, not just available as a module.

Noticed by Jeff Garzik and S.Çağlar Onur.

Cc: Jeff Garzik <jeff@garzik.org>
Cc: Jens Axboe <jens.axboe@oracle.com>
Cc: S.Çağlar Onur <caglar@pardus.org.tr>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

12 years agoMake check_signature depend on CONFIG_HAS_IOMEM
Linus Torvalds [Mon, 16 Jul 2007 23:50:01 +0000]
Make check_signature depend on CONFIG_HAS_IOMEM

This should avoid build problems on architectures without a "readb()",
that got bitten by check_signature() being uninlined.

Noted by Heiko Carstens.

Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

12 years agoforcedeth bug fix: realtek phy
Ayaz Abdulla [Sun, 15 Jul 2007 10:51:03 +0000]
forcedeth bug fix: realtek phy

This patch contains errata fixes for the realtek phy.

Signed-off-by: Ayaz Abdulla <aabdulla@nvidia.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

12 years agoforcedeth bug fix: vitesse phy
Ayaz Abdulla [Sun, 15 Jul 2007 10:50:53 +0000]
forcedeth bug fix: vitesse phy

This patch contains errata fixes for the vitesse phy.

Signed-off-by: Ayaz Abdulla <aabdulla@nvidia.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

12 years agoforcedeth bug fix: cicada phy
Ayaz Abdulla [Sun, 15 Jul 2007 10:50:28 +0000]
forcedeth bug fix: cicada phy

This patch contains errata fixes for the cicada phy. It only renamed the
defines to be phy specific.

Signed-off-by: Ayaz Abdulla <aabdulla@nvidia.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

12 years agoatl1: reorder atl1_main functions
Jay Cliburn [Sun, 15 Jul 2007 16:03:29 +0000]
atl1: reorder atl1_main functions

Reorder functions in atl1_main into more logical groupings to make the
code easier to follow. This patch is large, but it's harmless; it neither
adds nor removes any functionality whatsoever.

Signed-off-by: Jay Cliburn <jacliburn@bellsouth.net>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

12 years agoatl1: fix excessively indented code
Jay Cliburn [Sun, 15 Jul 2007 16:03:28 +0000]
atl1: fix excessively indented code

Move excessively indented code to separate functions.  Also move ring
pointer initialization to its own function.

Signed-off-by: Jay Cliburn <jacliburn@bellsouth.net>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

12 years agoatl1: cleanup atl1_main
Jay Cliburn [Sun, 15 Jul 2007 16:03:27 +0000]
atl1: cleanup atl1_main

Fix indentation, remove dead code, improve some comments, change dev_dbg to
dev_printk.

Signed-off-by: Jay Cliburn <jacliburn@bellsouth.net>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

12 years agoatl1: header file cleanup
Jay Cliburn [Sun, 15 Jul 2007 16:03:26 +0000]
atl1: header file cleanup

Remove unused structure members, improve comments, break long comment lines,
rename a constant to be consistent with others in the file.

Signed-off-by: Jay Cliburn <jacliburn@bellsouth.net>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

12 years agoatl1: remove irq_sem
Jay Cliburn [Sun, 15 Jul 2007 16:03:25 +0000]
atl1: remove irq_sem

Remove unnecessary irq_sem code.

Signed-off-by: Chris Snook <csnook@redhat.com>
Signed-off-by: Jay Cliburn <jacliburn@bellsouth.net>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

12 years agocdc-subset to support new vendor/product ID
jing xiang [Sat, 14 Jul 2007 06:13:24 +0000]
cdc-subset to support new vendor/product ID

This patch is for cdc subset to support Mavell vendor/product ID.

Signed-off-by: Jing Xiang <everxiang@gmail.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

12 years ago8139cp: implement the missing dev->tx_timeout
Francois Romieu [Fri, 13 Jul 2007 21:05:35 +0000]
8139cp: implement the missing dev->tx_timeout

Signed-off-by: Mika Lansirinne <mika.lansirinne@stonesoft.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

12 years agomyri10ge: Remove nonsensical limit in the tx done routine
Brice Goglin [Fri, 13 Jul 2007 18:15:13 +0000]
myri10ge: Remove nonsensical limit in the tx done routine

Remove nonsensical limit in the tx done routine. Specifically,
the loop will always terminate after processing <= 1 rings worth
of frames, as the mcp index is not refetched, so the removed
conditional could never be true.

Signed-off-by: Brice Goglin <brice@myri.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

12 years agogianfar: kill unused header
Kumar Gala [Fri, 13 Jul 2007 05:38:47 +0000]
gianfar: kill unused header

A long time ago we used OCP with the gianfar driver.  Eventually when
we kill arch/ppc including this will cause issues so lets just kill it now.

Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

12 years agoEP93XX_ETH must select MII
John Donoghue [Fri, 13 Jul 2007 00:12:08 +0000]
EP93XX_ETH must select MII

CONFIG_EP93XX_ETH=y, CONFIG_MII=n results in an obvious link error.

Signed-off-by: Adrian Bunk <bunk@stusta.de>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

12 years agomacb: Add multicast capability
Patrice Vilchez [Thu, 12 Jul 2007 17:07:25 +0000]
macb: Add multicast capability

Add multicast capability to Atmel ethernet macb driver.

Signed-off-by: Patrice Vilchez <patrice.vilchez@rfo.atmel.com>
Signed-off-by: Haavard Skinnemoen <hskinnemoen@atmel.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

12 years agomacb: Use generic PHY layer
frederic RODO [Thu, 12 Jul 2007 17:07:24 +0000]
macb: Use generic PHY layer

Convert the macb driver to use the generic PHY layer in
drivers/net/phy.

Signed-off-by: Frederic RODO <f.rodo@til-technologies.fr>
Signed-off-by: Haavard Skinnemoen <hskinnemoen@atmel.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

12 years agos390: add barriers to qeth driver
Frank Blaschka [Thu, 12 Jul 2007 10:51:35 +0000]
s390: add barriers to qeth driver

Add barrier to loop where atomic variable is evaluated.

Signed-off-by: Frank Blaschka <frank.blaschka@de.ibm.com>
Signed-off-by: Ursula Braun <braunu@de.ibm.com>
Signed-off-by: Frank Pavlic <fpavlic@de.ibm.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

12 years agos390: scatter-gather for inbound traffic in qeth driver
Frank Blaschka [Thu, 12 Jul 2007 10:51:34 +0000]
s390: scatter-gather for inbound traffic in qeth driver

For large incoming packets > PAGE_SIZE/2 qeth creates a fragmented skb
by adding pointers to qdio pages to the fragment list of the skb.
This avoids allocating big chunks of consecutive memory. Also copying
data from the qdio buffer to the skb is economized.

Signed-off-by: Frank Blaschka <frank.blaschka@de.ibm.com>
Signed-off-by: Ursula Braun <braunu@de.ibm.com>
Signed-off-by: Frank Pavlic <fpavlic@de.ibm.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

12 years agoeHEA: Introducing support vor DLPAR memory add
Thomas Klein [Wed, 11 Jul 2007 14:32:00 +0000]
eHEA: Introducing support vor DLPAR memory add

This patch adds support for DLPAR memory add to the eHEA driver. To detect
whether memory was added the driver uses its own memory mapping table and
checks for kernel addresses whether they're located in already known memory
sections. If not the function ehea_rereg_mrs() is triggered which performs
a rebuild of the mapping table and a re-registration of the global memory
region.

Signed-off-by: Thomas Klein <tklein@de.ibm.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

12 years agoMerge branch 'upstream-jgarzik' of git://git.kernel.org/pub/scm/linux/kernel/git...
Jeff Garzik [Mon, 16 Jul 2007 21:59:03 +0000]
Merge branch 'upstream-jgarzik' of git://git./linux/kernel/git/linville/wireless-2.6 into upstream

12 years agoFix a potential NULL pointer dereference in free_shared_mem() in drivers/net/s2io.c
Micah Gruber [Mon, 9 Jul 2007 03:29:04 +0000]
Fix a potential NULL pointer dereference in free_shared_mem() in drivers/net/s2io.c

This patch fixes a potential null dereference bug where we dereference
nic before a null check. This patch simply moves the dereferencing
after the null check.

Signed-off-by: Micah Gruber < micah.gruber@gmail.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

12 years agoRevert "[NET]: Fix races in net_rx_action vs netpoll."
Linus Torvalds [Mon, 16 Jul 2007 21:31:08 +0000]
Revert "[NET]: Fix races in net_rx_action vs netpoll."

This reverts commit 29578624e354f56143d92510fff33a8b2aaa2c03.

Ingo Molnar reports complete breakage with his e1000 card (no
networking, card reports transmit timeouts), and bisected it down to
this commit.  Let's figure out what went wrong, but not keep breaking
machines until we do.

Cc: Ingo Molnar <mingo@elte.hu>
Cc: Olaf Kirch <olaf.kirch@oracle.com>
Cc: David Miller <davem@davemloft.net>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

12 years agoMerge branch 'drm-patches' of ssh://master.kernel.org/pub/scm/linux/kernel/git/airlie...
Linus Torvalds [Mon, 16 Jul 2007 19:15:50 +0000]
Merge branch 'drm-patches' of ssh:///linux/kernel/git/airlied/drm-2.6

* 'drm-patches' of ssh://master.kernel.org/pub/scm/linux/kernel/git/airlied/drm-2.6:
  drm: remove core typedefs from the ioc32 wrappers
  drm: remove sarea typedefs
  drm: detypedef the hashtab and more of sman
  drm: de-typedef sman
  drm: detypedeffing continues...
  drm: detypef waitlist/freelist/buf_entry/device_dma/drm_queue structs
  drm: drop drm_vma_entry_t, drm_magic_entry_t
  drm: drop drm_buf_t typedef
  drm: fixup other drivers for typedef removals
  drm: remove drm_file_t, drm_device_t and drm_head_t typedefs
  drm: remove a bunch of typedefs on the userspace interface
  r300: updates register header
  radeon: add support for vblank on crtc2
  drm: cleanup list initialisation
  drm: fix typo on code drm getsarea
  drm: remove DRM_GETSAREA and replace with drm_getsarea function
  drm: cleanup use of Linux list handling macros

12 years agoutime(s): Honour CAP_FOWNER when times==NULL
Satyam Sharma [Mon, 16 Jul 2007 18:54:23 +0000]
utime(s): Honour CAP_FOWNER when times==NULL

do_utimes() does not honour CAP_FOWNER when times==NULL.
Trivial and obvious one-line fix.

Signed-off-by: Satyam Sharma <ssatyam@cse.iitk.ac.in>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

12 years agoFix too few arguments to function `scsi_cmd_ioctl'
S.Çağlar Onur [Mon, 16 Jul 2007 18:56:51 +0000]
Fix too few arguments to function `scsi_cmd_ioctl'

This corrects the following compile error introduced by the merge of the
new bsg layer in commit e245befce7af0a1e1347079ed62695b059594bd4:

  caglar@zangetsu linux-2.6 $ make
    CHK     include/linux/version.h
    CHK     include/linux/utsrelease.h
    CALL    scripts/checksyscalls.sh
    CHK     include/linux/compile.h
    LD      drivers/block/built-in.o
CC [M] drivers/block/cciss.o
  drivers/block/cciss.c: In function `cciss_ioctl':
  drivers/block/cciss.c:1173: warning: passing arg 2 of `scsi_cmd_ioctl' from incompatible pointer type
  drivers/block/cciss.c:1173: warning: passing arg 3 of `scsi_cmd_ioctl' makes pointer from integer without a cast
  drivers/block/cciss.c:1173: warning: passing arg 4 of `scsi_cmd_ioctl' makes integer from pointer without a cast
  drivers/block/cciss.c:1173: error: too few arguments to function `scsi_cmd_ioctl'
  ...
  make[2]: *** [drivers/block/cciss.o] Hata 1
  make[1]: *** [drivers/block] Hata 2
  make: *** [drivers] Hata 2

Signed-off-by: S.Çağlar Onur <caglar@pardus.org.tr>
Cc: Jens Axboe <jens.axboe@oracle.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

12 years agoFix LDM for new field in the VOL5 VBLK.
Anton Altaparmakov [Mon, 16 Jul 2007 18:39:02 +0000]
Fix LDM for new field in the VOL5 VBLK.

Teach LDM about a new field encountered with Windows Vista.

This fixes LDM for people using Vista who have disabled drive letter
assignment from one or more volumes.  Doing this introduces a so far
unknown field in the LDM database in the VOL5 VBLK structure which
causes the LDM driver to fail to parse the VBLK structure and hence LDM
fails to parse the disk altogether.  This patch teaches the driver about
this field.

Thanks got to Ashton Mills <amills@iinet.com.au> for reporting the
problem and working with me on getting it fixed.  It is now working for
him.

Signed-off-by: Anton Altaparmakov <aia21@cantab.net>
CC: Richard Russon <ldm@flatcap.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

12 years agoRemove duplicate comments from sysctl.c
Linus Torvalds [Mon, 16 Jul 2007 18:50:38 +0000]
Remove duplicate comments from sysctl.c

Randy Dunlap noticed that the recent comment clarifications from Andrew
had somehow gotten duplicated.  Quoth Andrew: "hm, that could have been
some late-night reject-fixing."

Fix it up.

Cc: From: Andrew Morton <akpm@linux-foundation.org>
Cc: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

12 years agoFix new generic block device SG compile
Linus Torvalds [Mon, 16 Jul 2007 18:18:23 +0000]
Fix new generic block device SG compile

We had a merge issue with the "dentry" field going away from the
kobject, and being replaced by a sysfs_dirent field (named "sd")
instead.  That broke the BSG compile.

Cc: Jens Axboe <jens.axboe@oracle.com>
Cc: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Cc: James Bottomley <James.Bottomley@SteelEye.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

12 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/mingo/linux-2.6-sched
Linus Torvalds [Mon, 16 Jul 2007 18:02:49 +0000]
Merge git://git./linux/kernel/git/mingo/linux-2.6-sched

* git://git.kernel.org/pub/scm/linux/kernel/git/mingo/linux-2.6-sched:
  [PATCH] sched: fix up fs/proc/array.c whitespace problems
  [PATCH] sched: prettify prio_to_wmult[]
  [PATCH] sched: document prio_to_wmult[]
  [PATCH] sched: improve weight-array comments
  [PATCH] sched: remove dead code from task_stime()

Fixed up trivial conflict in fs/proc/array.c

12 years agoMerge branch 'upstream-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mfashe...
Linus Torvalds [Mon, 16 Jul 2007 17:52:55 +0000]
Merge branch 'upstream-linus' of git://git./linux/kernel/git/mfasheh/ocfs2

* 'upstream-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mfasheh/ocfs2: (32 commits)
  [PATCH] ocfs2: zero_user_page conversion
  ocfs2: Support xfs style space reservation ioctls
  ocfs2: support for removing file regions
  ocfs2: update truncate handling of partial clusters
  ocfs2: btree support for removal of arbirtrary extents
  ocfs2: Support creation of unwritten extents
  ocfs2: support writing of unwritten extents
  ocfs2: small cleanup of ocfs2_write_begin_nolock()
  ocfs2: btree changes for unwritten extents
  ocfs2: abstract btree growing calls
  ocfs2: use all extent block suballocators
  ocfs2: plug truncate into cached dealloc routines
  ocfs2: simplify deallocation locking
  ocfs2: harden buffer check during mapping of page blocks
  ocfs2: shared writeable mmap
  ocfs2: factor out write aops into nolock variants
  ocfs2: rework ocfs2_buffered_write_cluster()
  ocfs2: take ip_alloc_sem during entire truncate
  ocfs2: Add "preferred slot" mount option
  [KJ PATCH] Replacing memset(<addr>,0,PAGE_SIZE) with clear_page() in fs/ocfs2/dlm/dlmrecovery.c
  ...

12 years agoMerge branch 'bsg' of git://git.kernel.dk/data/git/linux-2.6-block
Linus Torvalds [Mon, 16 Jul 2007 17:50:19 +0000]
Merge branch 'bsg' of git://git.kernel.dk/data/git/linux-2.6-block

* 'bsg' of git://git.kernel.dk/data/git/linux-2.6-block: (25 commits)
  bsg: Kconfig updates
  bsg: add SCSI transport-level request support
  bsg: add bidi support
  add a struct request pointer to the request structure
  bsg: fix the deadlock on discarding done commands
  bsg: fix a blocking read bug
  bsg: minor bug fixes
  improve bsg device allocation
  bind bsg to all SCSI devices
  bsg: bind bsg to request_queue instead of gendisk
  bsg: add a request_queue argument to scsi_cmd_ioctl()
  bsg: simplify __bsg_alloc_command failpath
  bsg: add cheasy error checks for sysfs stuff
  Add queue resizing support
  Replace s32, u32 and u64 with __s32, __u32 and __u64 in bsg.h for userspace
  bsg: silence a bogus gcc warning
  bsg: style cleanup
  bsg: use u32 etc instead of uint32_t
  bsg: add SG_IO to SG v4
  bsg: replace SG v3 with SG v4
  ...

12 years agoMerge branch 'for-linus' of git://git.kernel.dk/data/git/linux-2.6-block
Linus Torvalds [Mon, 16 Jul 2007 17:48:20 +0000]
Merge branch 'for-linus' of git://git.kernel.dk/data/git/linux-2.6-block

* 'for-linus' of git://git.kernel.dk/data/git/linux-2.6-block:
  splice: direct splicing updates ppos twice
  more ACSI removal
  umem: Fix match of pci_ids in umem driver
  umem: Remove references to dead CONFIG_MM_MAP_MEMORY variable
  remove the documentation for the legacy CDROM drivers

12 years agoMerge branch 'master' of master.kernel.org:/pub/scm/linux/kernel/git/davem/sparc-2.6
Linus Torvalds [Mon, 16 Jul 2007 17:45:23 +0000]
Merge branch 'master' of /linux/kernel/git/davem/sparc-2.6

* 'master' of master.kernel.org:/pub/scm/linux/kernel/git/davem/sparc-2.6: (26 commits)
  [SPARC64]: Fix UP build.
  [SPARC64]: dr-cpu unconfigure support.
  [SERIAL]: Fix console write locking in sparc drivers.
  [SPARC64]: Give more accurate errors in dr_cpu_configure().
  [SPARC64]: Clear cpu_{core,sibling}_map[] in smp_fill_in_sib_core_maps()
  [SPARC64]: Fix leak when DR added cpu does not bootup.
  [SPARC64]: Add ->set_affinity IRQ handlers.
  [SPARC64]: Process dr-cpu events in a kthread instead of workqueue.
  [SPARC64]: More sensible udelay implementation.
  [SPARC64]: SMP build fixes.
  [SPARC64]: mdesc.c needs linux/mm.h
  [SPARC64]: Fix build regressions added by dr-cpu changes.
  [SPARC64]: Unconditionally register vio_bus_type.
  [SPARC64]: Initial LDOM cpu hotplug support.
  [SPARC64]: Fix setting of variables in LDOM guest.
  [SPARC64]: Fix MD property lifetime bugs.
  [SPARC64]: Abstract out mdesc accesses for better MD update handling.
  [SPARC64]: Use more mearningful names for IRQ registry.
  [SPARC64]: Initial domain-services driver.
  [SPARC64]: Export powerd facilities for external entities.
  ...

12 years agoMerge master.kernel.org:/pub/scm/linux/kernel/git/lethal/sh-2.6
Linus Torvalds [Mon, 16 Jul 2007 17:32:02 +0000]
Merge /pub/scm/linux/kernel/git/lethal/sh-2.6

* master.kernel.org:/pub/scm/linux/kernel/git/lethal/sh-2.6: (68 commits)
  sh: sh-rtc support for SH7709.
  sh: Revert __xdiv64_32 size change.
  sh: Update r7785rp defconfig.
  sh: Export div symbols for GCC 4.2 and ST GCC.
  sh: fix race in parallel out-of-tree build
  sh: Kill off dead mach.c for hp6xx.
  sh: hd64461.h cleanup and added comments.
  sh: Update the alignment when 4K stacks are used.
  sh: Add a .bss.page_aligned section for 4K stacks.
  sh: Don't let SH-4A clobber SH-4 CFLAGS.
  sh: Add parport stub for SuperIO ports.
  sh: Drop -Wa,-dsp for DSP tuning.
  sh: Update dreamcast defconfig.
  fb: pvr2fb: A few more __devinit annotations for PCI.
  fb: pvr2fb: Fix up section mismatch warnings.
  sh: Select IPR-IRQ for SH7091.
  sh: Correct __xdiv64_32/div64_32 return value size.
  sh: Fix timer-tmu build for SH-3.
  sh: Add cpu and mach links to CLEAN_FILES.
  sh: Preliminary support for the SH-X3 CPU.
  ...

12 years agofat: Fix the race of read/write the FAT12 entry
OGAWA Hirofumi [Mon, 16 Jul 2007 00:40:05 +0000]
fat: Fix the race of read/write the FAT12 entry

FAT12 entry is 12bits, so it needs 2 phase to update the value.  And
writer and reader access it without any lock, so reader can get the
half updated value.

This fixes the long standing race condition by adding a global
spinlock to only FAT12 for avoiding any impact against FAT16/32.

Signed-off-by: OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

12 years agocpwatchdog build fix
Andrew Morton [Mon, 16 Jul 2007 06:42:03 +0000]
cpwatchdog build fix

sparc64:

drivers/sbus/char/cpwatchdog.c: In function `wd_toggleintr':
drivers/sbus/char/cpwatchdog.c:523: error: implicit declaration of function `readb'
drivers/sbus/char/cpwatchdog.c:533: error: implicit declaration of function `writeb'
drivers/sbus/char/cpwatchdog.c: In function `wd_pingtimer':
drivers/sbus/char/cpwatchdog.c:545: error: implicit declaration of function `readw'
drivers/sbus/char/cpwatchdog.c: In function `wd_starttimer':
drivers/sbus/char/cpwatchdog.c:584: error: implicit declaration of function `writew'
drivers/sbus/char/cpwatchdog.c: In function `wd_init':
drivers/sbus/char/cpwatchdog.c:767: error: implicit declaration of function `ioremap'
drivers/sbus/char/cpwatchdog.c:767: warning: assignment makes pointer from integer without a cast
drivers/sbus/char/cpwatchdog.c: In function `wd_cleanup':
drivers/sbus/char/cpwatchdog.c:849: error: implicit declaration of function `iounmap'

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>

12 years agosn_console section mismatch warning
Jes Sorensen [Mon, 16 Jul 2007 06:42:02 +0000]
sn_console section mismatch warning

Do not mark sn_sal_console_setup as __init since it's referenced from
non init data structures.

Signed-off-by: Jes Sorensen <jes@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

12 years agocompat32: ignore the LOOP_CLR_FD ioctl
Geert Uytterhoeven [Mon, 16 Jul 2007 06:42:01 +0000]
compat32: ignore the LOOP_CLR_FD ioctl

compat32: Ignore the LOOP_CLR_FD ioctl for the loop block device, to kill an
annoying kernel message when e.g. busybox umount is used.

Signed-off-by: Geert Uytterhoeven <Geert.Uytterhoeven@sonycom.com>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

12 years agoNLS: Remove obsolete Makefile entries
Robert P. J. Day [Mon, 16 Jul 2007 06:42:01 +0000]
NLS: Remove obsolete Makefile entries

Since the corresponding source files no longer exist, remove the
irrelevant Makefile entries for them.

Signed-off-by: Robert P. J. Day <rpjday@mindspring.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

12 years agopermit mempool_free(NULL)
Rusty Russell [Mon, 16 Jul 2007 06:42:00 +0000]
permit mempool_free(NULL)

Christian Borntraeger points out that mempool_free() doesn't noop when
handed NULL.  This is inconsistent with the other free-like functions
in the kernel.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Cc: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

12 years agoext4: statfs speed up
Badari Pulavarty [Mon, 16 Jul 2007 06:42:00 +0000]
ext4: statfs speed up

This is a patch that speeds up statfs.  It is very simple - the "overhead"
calculation, which takes a huge amount of time for large filesystems, never
changes unless the size of the filesystem itself changes.  That means we can
store it in memory and only recalculate if the filesystem has been resized
(almost never).

It also fixes a minor problem that we never update the on-disk superblock free
blocks/inodes counts until the filesystem is unmounted.  While not fatal, we
may as well update that on disk when we have the information, and it makes
things like debugfs and dumpe2fs report a bit more accurate info.

Signed-off-by: Badari Pulavarty <pbadari@us.ibm.com>
Signed-off-by: Andreas Dilger <adilger@clusterfs.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

12 years agoext3: statfs speed up
Badari Pulavarty [Mon, 16 Jul 2007 06:41:59 +0000]
ext3: statfs speed up

This is a patch that speeds up statfs.  It is very simple - the "overhead"
calculation, which takes a huge amount of time for large filesystems, never
changes unless the size of the filesystem itself changes.  That means we can
store it in memory and only recalculate if the filesystem has been resized
(almost never).

It also fixes a minor problem that we never update the on-disk superblock free
blocks/inodes counts until the filesystem is unmounted.  While not fatal, we
may as well update that on disk when we have the information, and it makes
things like debugfs and dumpe2fs report a bit more accurate info.

Signed-off-by: Badari Pulavarty <pbadari@us.ibm.com>
Signed-off-by: Andreas Dilger <adilger@clusterfs.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

12 years agoext2: statfs speed up
Badari Pulavarty [Mon, 16 Jul 2007 06:41:58 +0000]
ext2: statfs speed up

This is a patch that speeds up statfs.  It is very simple - the "overhead"
calculation, which takes a huge amount of time for large filesystems, never
changes unless the size of the filesystem itself changes.  That means we can
store it in memory and only recalculate if the filesystem has been resized
(almost never).

It also fixes a minor problem that we never update the on-disk superblock free
blocks/inodes counts until the filesystem is unmounted.  While not fatal, we
may as well update that on disk when we have the information, and it makes
things like debugfs and dumpe2fs report a bit more accurate info.

Signed-off-by: Badari Pulavarty <pbadari@us.ibm.com>
Signed-off-by: Andreas Dilger <adilger@clusterfs.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

12 years agodrivers/char/ipmi/ipmi_si_intf.c: lower printk severity
Dan Aloni [Mon, 16 Jul 2007 06:41:57 +0000]
drivers/char/ipmi/ipmi_si_intf.c: lower printk severity

Signed-off-by: Dan Aloni <da-x@monatomic.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

12 years agodrivers/char/ipmi/ipmi_poweroff.c: lower printk severity
Dan Aloni [Mon, 16 Jul 2007 06:41:56 +0000]
drivers/char/ipmi/ipmi_poweroff.c: lower printk severity

Signed-off-by: Dan Aloni <da-x@monatomic.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

12 years agovsprintf.c: optimizing, part 2: base 10 conversion speedup, v2
Denis Vlasenko [Mon, 16 Jul 2007 06:41:56 +0000]
vsprintf.c: optimizing, part 2: base 10 conversion speedup, v2

Optimize integer-to-string conversion in vsprintf.c for base 10.  This is
by far the most used conversion, and in some use cases it impacts
performance.  For example, top reads /proc/$PID/stat for every process, and
with 4000 processes decimal conversion alone takes noticeable time.

Using code from

http://www.cs.uiowa.edu/~jones/bcd/decimal.html
(with permission from the author, Douglas W. Jones)

binary-to-decimal-string conversion is done in groups of five digits at
once, using only additions/subtractions/shifts (with -O2; -Os throws in
some multiply instructions).

On i386 arch gcc 4.1.2 -O2 generates ~500 bytes of code.

This patch is run tested. Userspace benchmark/test is also attached.
I tested it on PIII and AMD64 and new code is generally ~2.5 times
faster. On AMD64:

# ./vsprintf_verify-O2
Original decimal conv: .......... 151 ns per iteration
Patched decimal conv:  .......... 62 ns per iteration
Testing correctness
12895992590592 ok...        [Ctrl-C]
# ./vsprintf_verify-O2
Original decimal conv: .......... 151 ns per iteration
Patched decimal conv:  .......... 62 ns per iteration
Testing correctness
26025406464 ok...        [Ctrl-C]

More realistic test: top from busybox project was modified to
report how many us it took to scan /proc (this does not account
any processing done after that, like sorting process list),
and then I test it with 4000 processes:

#!/bin/sh
i=4000
while test $i != 0; do
    sleep 30 &
    let i--
done
busybox top -b -n3 >/dev/null

on unpatched kernel:

top: 4120 processes took 102864 microseconds to scan
top: 4120 processes took 91757 microseconds to scan
top: 4120 processes took 92517 microseconds to scan
top: 4120 processes took 92581 microseconds to scan

on patched kernel:

top: 4120 processes took 75460 microseconds to scan
top: 4120 processes took 66451 microseconds to scan
top: 4120 processes took 67267 microseconds to scan
top: 4120 processes took 67618 microseconds to scan

The speedup comes from much faster generation of /proc/PID/stat
by sprintf() calls inside the kernel.

Signed-off-by: Douglas W Jones <jones@cs.uiowa.edu>
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

12 years agovsprintf.c: optimizing, part 1 (easy and obvious stuff)
Denis Vlasenko [Mon, 16 Jul 2007 06:41:54 +0000]
vsprintf.c: optimizing, part 1 (easy and obvious stuff)

* There is no point in having full "0...9a...z" constant vector,
  if we use only "0...9a...f" (and "x" for "0x").

* Post-decrement usually needs a few more instructions, so use
  pre decrement instead where makes sense:
-       while (i < precision--) {
+       while (i <= --precision) {

* if base != 10 (=> base 8 or 16), we can avoid using division
  in a loop and use mask/shift, obtaining much faster conversion.
  (More complex optimization for base 10 case is in the second patch).

Overall, size vsprintf.o shows ~80 bytes smaller text section
with this patch applied.

Signed-off-by: Douglas W Jones <jones@cs.uiowa.edu>
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

12 years agoFix trivial typos in anon_inodes.c comments
J. Bruce Fields [Mon, 16 Jul 2007 06:41:53 +0000]
Fix trivial typos in anon_inodes.c comments

Trivial typo and grammar fixes.

Signed-off-by: "J. Bruce Fields" <bfields@citi.umich.edu>
Cc: Davide Libenzi <davidel@xmailserver.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

12 years agoisapnp: remove pointless check of 'type' against 0 in isapnp_read_tag()
Jesper Juhl [Mon, 16 Jul 2007 06:41:53 +0000]
isapnp: remove pointless check of 'type' against 0 in isapnp_read_tag()

In drivers/pnp/isapnp/core.c::isapnp_read_tag() there is a test of 'type'
being == 0 a bit down in the function.  That test doesn't make any sense.
If 'type' could indeed be NULL, then the test happens way too late as we'd
already have tried to dereference the pointer earlier and looking at the
callers it also turns out that there is no way type can ever actually be
NULL.

So the test is completely pointless and should just be removed.

Signed-off-by: Jesper Juhl <jesper.juhl@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

12 years ago.gitignore update
Alexey Dobriyan [Mon, 16 Jul 2007 06:41:52 +0000]
.gitignore update

headers_install by default puts headers into usr/include/ .
They're auto-generated, so should be ignored.

Same for *.orig, *.rej .

Signed-off-by: Alexey Dobriyan <adobriyan@sw.ru>
Cc: David Woodhouse <dwmw2@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>