linux-2.6
18 years ago[PATCH] slab: remove SLAB_DMA
Christoph Lameter [Thu, 7 Dec 2006 04:33:19 +0000 (20:33 -0800)] 
[PATCH] slab: remove SLAB_DMA

SLAB_DMA is an alias of GFP_DMA. This is the last one so we
remove the leftover comment too.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] slab: remove SLAB_KERNEL
Christoph Lameter [Thu, 7 Dec 2006 04:33:17 +0000 (20:33 -0800)] 
[PATCH] slab: remove SLAB_KERNEL

SLAB_KERNEL is an alias of GFP_KERNEL.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] slab: remove SLAB_ATOMIC
Christoph Lameter [Thu, 7 Dec 2006 04:33:16 +0000 (20:33 -0800)] 
[PATCH] slab: remove SLAB_ATOMIC

SLAB_ATOMIC is an alias of GFP_ATOMIC

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] slab: remove SLAB_USER
Christoph Lameter [Thu, 7 Dec 2006 04:33:15 +0000 (20:33 -0800)] 
[PATCH] slab: remove SLAB_USER

SLAB_USER is an alias of GFP_USER

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] slab: remove SLAB_NOFS
Christoph Lameter [Thu, 7 Dec 2006 04:33:14 +0000 (20:33 -0800)] 
[PATCH] slab: remove SLAB_NOFS

SLAB_NOFS is an alias of GFP_NOFS.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] slab: remove SLAB_NOIO
Christoph Lameter [Thu, 7 Dec 2006 04:33:13 +0000 (20:33 -0800)] 
[PATCH] slab: remove SLAB_NOIO

SLAB_NOIO is an alias of GFP_NOIO with a single instance of use.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] slab: remove SLAB_LEVEL_MASK
Christoph Lameter [Thu, 7 Dec 2006 04:33:12 +0000 (20:33 -0800)] 
[PATCH] slab: remove SLAB_LEVEL_MASK

SLAB_LEVEL_MASK is only used internally to the slab and is
and alias of GFP_LEVEL_MASK.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] slab: remove SLAB_NO_GROW
Christoph Lameter [Thu, 7 Dec 2006 04:33:10 +0000 (20:33 -0800)] 
[PATCH] slab: remove SLAB_NO_GROW

It is only used internally in the slab.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] kill install_file_pte's pte_val
Hugh Dickins [Thu, 7 Dec 2006 04:33:09 +0000 (20:33 -0800)] 
[PATCH] kill install_file_pte's pte_val

David Binderman and his Intel C compiler rightly observe that
install_file_pte no longer has any use for its pte_val.

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Cc: d binderman <dcb314@hotmail.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] mm: cleanup indentation on switch for CPU operations
Andy Whitcroft [Thu, 7 Dec 2006 04:33:08 +0000 (20:33 -0800)] 
[PATCH] mm: cleanup indentation on switch for CPU operations

These patches introduced new switch statements which are indented contrary
to the concensus in mm/*.c.  Fix them up to match that concensus.

    [PATCH] node local per-cpu-pages
    [PATCH] ZVC: Scale thresholds depending on the size of the system
    commit e7c8d5c9955a4d2e88e36b640563f5d6d5aba48a
    commit df9ecaba3f152d1ea79f2a5e0b87505e03f47590

Signed-off-by: Andy Whitcroft <apw@shadowen.org>
Cc: Christoph Lameter <clameter@engr.sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] reject corrupt swapfiles earlier
Eric Sandeen [Thu, 7 Dec 2006 04:33:06 +0000 (20:33 -0800)] 
[PATCH] reject corrupt swapfiles earlier

The fsfuzzer found this; with a corrupt small swapfile that claims to have
many pages:

  [root]# file swap.741.img
  swap.741.img: Linux/i386 swap file (new style) 1 (4K pages) size 1040191487 pages
  [root]# ls -l swap.741.img
  -rw-r--r-- 1 root root 16777216 Nov 22 05:18 swap.741.img

sys_swapon() will try to vmalloc all those pages, and -then- check to see if
the file is actually that large:

                if (!(p->swap_map = vmalloc(maxpages * sizeof(short)))) {
  <snip>
        if (swapfilesize && maxpages > swapfilesize) {
                printk(KERN_WARNING
                       "Swap area shorter than signature indicates\n");

It seems to me that it would make more sense to move this test up before
the vmalloc, with the other checks, to avoid the OOM-killer in this
situation...

Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Cc: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] silence unused pgdat warning from alloc_bootmem_node and friends
Andy Whitcroft [Thu, 7 Dec 2006 04:33:04 +0000 (20:33 -0800)] 
[PATCH] silence unused pgdat warning from alloc_bootmem_node and friends

x86 NUMA systems only define bootmem for node 0.  alloc_bootmem_node() and
friends therefore ignore the passed pgdat and use NODE_DATA(0) in all
cases.  This leads to the following warnings as we are not using the passed
parameter:

  .../mm/page_alloc.c: In function 'zone_wait_table_init':
  .../mm/page_alloc.c:2259: warning: unused variable 'pgdat'

One option would be to define all variables used with these macros
__attribute__ ((unused)), but this would leave us exposed should these
become genuinely unused.

The key here is that we _are_ using the value, we ignore it but that is a
deliberate action.  This patch adds a nested local variable within the
alloc_bootmem_node helper to which the pgdat parameter is assigned making
it 'used'.  The nested local is marked __attribute__ ((unused)) to silence
this same warning for it.

Signed-off-by: Andy Whitcroft <apw@shadowen.org>
Cc: Christoph Lameter <clameter@engr.sgi.com>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] numa node ids are int, page_to_nid and zone_to_nid should return int
Andy Whitcroft [Thu, 7 Dec 2006 04:33:03 +0000 (20:33 -0800)] 
[PATCH] numa node ids are int, page_to_nid and zone_to_nid should return int

NUMA node ids are passed as either int or unsigned int almost exclusivly
page_to_nid and zone_to_nid both return unsigned long.  This is a throw
back to when page_to_nid was a #define and was thus exposing the real type
of the page flags field.

In addition to fixing up the definitions of page_to_nid and zone_to_nid I
audited the users of these functions identifying the following incorrect
uses:

1) mm/page_alloc.c show_node() -- printk dumping the node id,
2) include/asm-ia64/pgalloc.h pgtable_quicklist_free() -- comparison
   against numa_node_id() which returns an int from cpu_to_node(), and
3) mm/mpolicy.c check_pte_range -- used as an index in node_isset which
   uses bit_set which in generic code takes an int.

Signed-off-by: Andy Whitcroft <apw@shadowen.org>
Cc: Christoph Lameter <clameter@engr.sgi.com>
Cc: "Luck, Tony" <tony.luck@intel.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] drain_node_page(): Drain pages in batch units
Christoph Lameter [Thu, 7 Dec 2006 04:33:02 +0000 (20:33 -0800)] 
[PATCH] drain_node_page(): Drain pages in batch units

drain_node_pages() currently drains the complete pageset of all pages.  If
there are a large number of pages in the queues then we may hold off
interrupts for too long.

Duplicate the method used in free_hot_cold_page.  Only drain pcp->batch
pages at one time.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] Remove uses of kmem_cache_t from mm/* and include/linux/slab.h
Christoph Lameter [Thu, 7 Dec 2006 04:32:59 +0000 (20:32 -0800)] 
[PATCH] Remove uses of kmem_cache_t from mm/* and include/linux/slab.h

Remove all uses of kmem_cache_t (the most were left in slab.h).  The
typedef for kmem_cache_t is then only necessary for other kernel
subsystems.  Add a comment to that effect.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] Move names_cachep to linux/fs.h
Christoph Lameter [Thu, 7 Dec 2006 04:32:57 +0000 (20:32 -0800)] 
[PATCH] Move names_cachep to linux/fs.h

The names_cachep is used for getname() and putname().  So lets put it into
fs.h near those two definitions.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] Move fs_cachep to linux/fs_struct.h
Christoph Lameter [Thu, 7 Dec 2006 04:32:54 +0000 (20:32 -0800)] 
[PATCH] Move fs_cachep to linux/fs_struct.h

fs_cachep is only used in kernel/exit.c and in kernel/fork.c.

It is used to store fs_struct items so it should be placed in linux/fs_struct.h

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] Move filep_cachep to include/file.h
Christoph Lameter [Thu, 7 Dec 2006 04:32:52 +0000 (20:32 -0800)] 
[PATCH] Move filep_cachep to include/file.h

filp_cachep is only used in fs/file_table.c and in fs/dcache.c where
it is defined.

Move it to related definitions in linux/file.h.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] Move files_cachep to include/file.h
Christoph Lameter [Thu, 7 Dec 2006 04:32:50 +0000 (20:32 -0800)] 
[PATCH] Move files_cachep to include/file.h

Proper place is in file.h since files_cachep uses are rated to file I/O.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] Move vm_area_cachep to include/mm.h
Christoph Lameter [Thu, 7 Dec 2006 04:32:48 +0000 (20:32 -0800)] 
[PATCH] Move vm_area_cachep to include/mm.h

vm_area_cachep is used to store vm_area_structs. So move to mm.h.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] Move sighand_cachep to include/signal.h
Christoph Lameter [Thu, 7 Dec 2006 04:32:47 +0000 (20:32 -0800)] 
[PATCH] Move sighand_cachep to include/signal.h

Move sighand_cachep definitioni to linux/signal.h

The sighand cache is only used in fs/exec.c and kernel/fork.c.  It is defined
in kernel/fork.c but only used in fs/exec.c.

The sighand_cachep is related to signal processing.  So add the definition to
signal.h.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] Remove bio_cachep from slab.h
Christoph Lameter [Thu, 7 Dec 2006 04:32:45 +0000 (20:32 -0800)] 
[PATCH] Remove bio_cachep from slab.h

Remove bio_cachep from slab.h - it no longer exists.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] make mm/thrash.c:global_faults static
Adrian Bunk [Thu, 7 Dec 2006 04:32:43 +0000 (20:32 -0800)] 
[PATCH] make mm/thrash.c:global_faults static

This patch makes the needlessly global "global_faults" static.

Signed-off-by: Adrian Bunk <bunk@stusta.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] enable booting a NUMA system where some nodes have no memory
Christian Krafft [Thu, 7 Dec 2006 04:32:41 +0000 (20:32 -0800)] 
[PATCH] enable booting a NUMA system where some nodes have no memory

When booting a NUMA system with nodes that have no memory (eg by limiting
memory), bootmem_alloc_core tried to find pages in an uninitialized
bootmem_map.  This caused a null pointer access.  This fix adds a check, so
that NULL is returned.  That will enable the caller (bootmem_alloc_nopanic)
to alloc memory on other without a panic.

Signed-off-by: Christian Krafft <krafft@de.ibm.com>
Cc: Christoph Lameter <clameter@engr.sgi.com>
Cc: Andy Whitcroft <apw@shadowen.org>
Cc: Martin Bligh <mbligh@google.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] Allow NULL pointers in percpu_free
Alan Stern [Thu, 7 Dec 2006 04:32:37 +0000 (20:32 -0800)] 
[PATCH] Allow NULL pointers in percpu_free

The patch (as824b) makes percpu_free() ignore NULL arguments, as one would
expect for a deallocation routine.  (Note that free_percpu is #defined as
percpu_free in include/linux/percpu.h.) A few callers are updated to remove
now-unneeded tests for NULL.  A few other callers already seem to assume
that passing a NULL pointer to percpu_free() is okay!

The patch also removes an unnecessary NULL check in percpu_depopulate().

Signed-off-by: Alan Stern <stern@rowland.harvard.edu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] node-aware skb allocation
Christoph Hellwig [Thu, 7 Dec 2006 04:32:36 +0000 (20:32 -0800)] 
[PATCH] node-aware skb allocation

Node-aware allocation of skbs for the receive path.

Details:

  - __alloc_skb gets a new node argument and cals the node-aware
    slab functions with it.
  - netdev_alloc_skb passed the node number it gets from dev_to_node
    to it, everyone else passes -1 (any node)

Signed-off-by: Christoph Hellwig <hch@lst.de>
Cc: Christoph Lameter <clameter@engr.sgi.com>
Cc: "David S. Miller" <davem@davemloft.net>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] add numa node information to struct device
Christoph Hellwig [Thu, 7 Dec 2006 04:32:33 +0000 (20:32 -0800)] 
[PATCH] add numa node information to struct device

For node-aware skb allocations we need information about the node in struct
net_device or struct device.  Davem suggested to put it into struct device
which this patch does.

In particular:

 - struct device gets a new int numa_node member if CONFIG_NUMA is set
 - there are two new helpers, dev_to_node and set_dev_node to
   transparently deal with the non-numa case
 - for pci devices the node-info is set to the value we get from
   pcibus_to_node.

Note that for some architectures pcibus_to_node doesn't work yet at the time
we call it currently.  This is harmless and will just mean skb allocations
aren't node-local on this architectures until the implementation of
pcibus_to_node on these architectures have been updated (There are patches for
x86 and x86_64 floating around)

[akpm@osdl.org: cleanup]
Signed-off-by: Christoph Hellwig <hch@lst.de>
Cc: Christoph Lameter <clameter@engr.sgi.com>
Cc: Greg KH <greg@kroah.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] leak tracking for kmalloc_node
Christoph Hellwig [Thu, 7 Dec 2006 04:32:30 +0000 (20:32 -0800)] 
[PATCH] leak tracking for kmalloc_node

We have variants of kmalloc and kmem_cache_alloc that leave leak tracking to
the caller.  This is used for subsystem-specific allocators like skb_alloc.

To make skb_alloc node-aware we need similar routines for the node-aware slab
allocator, which this patch adds.

Note that the code is rather ugly, but it mirrors the non-node-aware code 1:1:

[akpm@osdl.org: add module export]
Signed-off-by: Christoph Hellwig <hch@lst.de>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] Always print out the header line in /proc/swaps
Suleiman Souhlal [Thu, 7 Dec 2006 04:32:28 +0000 (20:32 -0800)] 
[PATCH] Always print out the header line in /proc/swaps

It would be possible for /proc/swaps to not always print out the header:

swapon /dev/hdc2
swapon /dev/hde2
swapoff /dev/hdc2

At this point /proc/swaps would not have a header.

Signed-off-by: Suleiman Souhlal <suleiman@google.com>
Cc: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] OOM can panic due to processes stuck in __alloc_pages()
Kirill Korotaev [Thu, 7 Dec 2006 04:32:27 +0000 (20:32 -0800)] 
[PATCH] OOM can panic due to processes stuck in __alloc_pages()

OOM can panic due to the processes stuck in __alloc_pages() doing infinite
rebalance loop while no memory can be reclaimed.  OOM killer tries to kill
some processes, but unfortunetaly, rebalance label was moved by someone
below the TIF_MEMDIE check, so buddy allocator doesn't see that process is
OOM-killed and it can simply fail the allocation :/

Observed in reality on RHEL4(2.6.9)+OpenVZ kernel when a user doing some
memory allocation tricks triggered OOM panic.

Signed-off-by: Denis Lunev <den@sw.ru>
Signed-off-by: Kirill Korotaev <dev@openvz.org>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] mlock cleanup
Rik Bobbaers [Thu, 7 Dec 2006 04:32:25 +0000 (20:32 -0800)] 
[PATCH] mlock cleanup

mm is defined as vma->vm_mm, so use that.

Acked-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] Allow user processes to raise their oom_adj value
Guillem Jover [Thu, 7 Dec 2006 04:32:24 +0000 (20:32 -0800)] 
[PATCH] Allow user processes to raise their oom_adj value

Currently a user process cannot rise its own oom_adj value (i.e.
unprotecting itself from the OOM killer).  As this value is stored in the
task structure it gets inherited and the unprivileged childs will be unable
to rise it.

The EPERM will be handled by the generic proc fs layer, as only processes
with the proper caps or the owner of the process will be able to write to
the file.  So we allow only the processes with CAP_SYS_RESOURCE to lower
the value, otherwise it will get an EACCES which seems more appropriate
than EPERM.

Signed-off-by: Guillem Jover <guillem.jover@nokia.com>
Acked-by: Andrea Arcangeli <andrea@novell.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] Fix kunmap_atomic's use of kpte_clear_flush()
Jeremy Fitzhardinge [Thu, 7 Dec 2006 04:32:22 +0000 (20:32 -0800)] 
[PATCH] Fix kunmap_atomic's use of kpte_clear_flush()

kunmap_atomic() will call kpte_clear_flush with vaddr/ptep arguments which
don't correspond if the vaddr is just a normal lowmem address (ie, not in
the KMAP area).  This patch makes sure that the pte is only cleared if kmap
area was actually used for the mapping.

Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Cc: Zachary Amsden <zach@vmware.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] mm: k{,um}map_atomic() vs in_atomic()
Peter Zijlstra [Thu, 7 Dec 2006 04:32:21 +0000 (20:32 -0800)] 
[PATCH] mm: k{,um}map_atomic() vs in_atomic()

Make kmap_atomic/kunmap_atomic denote a pagefault disabled scope.  All non
trivial implementations already do this anyway.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] mm: pagefault_{disable,enable}()
Peter Zijlstra [Thu, 7 Dec 2006 04:32:20 +0000 (20:32 -0800)] 
[PATCH] mm: pagefault_{disable,enable}()

Introduce pagefault_{disable,enable}() and use these where previously we did
manual preempt increments/decrements to make the pagefault handler do the
atomic thing.

Currently they still rely on the increased preempt count, but do not rely on
the disabled preemption, this might go away in the future.

(NOTE: the extra barrier() in pagefault_disable might fix some holes on
       machines which have too many registers for their own good)

[heiko.carstens@de.ibm.com: s390 fix]
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Nick Piggin <npiggin@suse.de>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] mm: arch do_page_fault() vs in_atomic()
Peter Zijlstra [Thu, 7 Dec 2006 04:32:18 +0000 (20:32 -0800)] 
[PATCH] mm: arch do_page_fault() vs in_atomic()

In light of the recent pagefault and filemap_copy_from_user work I've gone
through all the arch pagefault handlers to make sure the inc_preempt_count()
'feature' works as expected.

Several sections of code (including the new filemap_copy_from_user) rely on
the fact that faults do not take locks under increased preempt count.

arch/x86_64 - good
arch/powerpc - good
arch/cris - fixed
arch/i386 - good
arch/parisc - fixed
arch/sh - good
arch/sparc - good
arch/s390 - good
arch/m68k - fixed
arch/ppc - good
arch/alpha - fixed
arch/mips - good
arch/sparc64 - good
arch/ia64 - good
arch/arm - fixed
arch/um - good
arch/avr32 - good
arch/h8300 - NA
arch/m32r - good
arch/v850 - good
arch/frv - fixed
arch/m68knommu - NA
arch/arm26 - fixed
arch/sh64 - fixed
arch/xtensa - good

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] mm: add noaliencache boot option to disable numa alien caches
Paul Menage [Thu, 7 Dec 2006 04:32:16 +0000 (20:32 -0800)] 
[PATCH] mm: add noaliencache boot option to disable numa alien caches

When using numa=fake on non-NUMA hardware there is no benefit to having the
alien caches, and they consume much memory.

Add a kernel boot option to disable them.

Christoph sayeth "This is good to have even on large NUMA.  The problem is
that the alien caches grow by the square of the size of the system in terms of
nodes."

Cc: Christoph Lameter <clameter@engr.sgi.com>
Cc: Pekka Enberg <penberg@cs.helsinki.fi>
Cc: Manfred Spraul <manfred@colorfullife.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] mm: slab: eliminate lock_cpu_hotplug from slab
Ravikiran G Thirumalai [Thu, 7 Dec 2006 04:32:14 +0000 (20:32 -0800)] 
[PATCH] mm: slab: eliminate lock_cpu_hotplug from slab

Here's an attempt towards doing away with lock_cpu_hotplug in the slab
subsystem.  This approach also fixes a bug which shows up when cpus are
being offlined/onlined and slab caches are being tuned simultaneously.

http://marc.theaimsgroup.com/?l=linux-kernel&m=116098888100481&w=2

The patch has been stress tested overnight on a 2 socket 4 core AMD box with
repeated cpu online and offline, while dbench and kernbench process are
running, and slab caches being tuned at the same time.
There were no lockdep warnings either.  (This test on 2,6.18 as 2.6.19-rc
crashes at __drain_pages
http://marc.theaimsgroup.com/?l=linux-kernel&m=116172164217678&w=2 )

The approach here is to hold cache_chain_mutex from CPU_UP_PREPARE until
CPU_ONLINE (similar in approach as worqueue_mutex) .  Slab code sensitive
to cpu_online_map (kmem_cache_create, kmem_cache_destroy, slabinfo_write,
__cache_shrink) is already serialized with cache_chain_mutex.  (This patch
lengthens cache_chain_mutex hold time at kmem_cache_destroy to cover this).
 This patch also takes the cache_chain_sem at kmem_cache_shrink to protect
sanity of cpu_online_map at __cache_shrink, as viewed by slab.
(kmem_cache_shrink->__cache_shrink->drain_cpu_caches).  But, really,
kmem_cache_shrink is used at just one place in the acpi subsystem!  Do we
really need to keep kmem_cache_shrink at all?

Another note.  Looks like a cpu hotplug event can send  CPU_UP_CANCELED to
a registered subsystem even if the subsystem did not receive CPU_UP_PREPARE.
This could be due to a subsystem registered for notification earlier than
the current subsystem crapping out with NOTIFY_BAD. Badness can occur with
in the CPU_UP_CANCELED code path at slab if this happens (The same would
apply for workqueue.c as well).  To overcome this, we might have to use either
a) a per subsystem flag and avoid handling of CPU_UP_CANCELED, or
b) Use a special notifier events like LOCK_ACQUIRE/RELEASE as Gautham was
   using in his experiments, or
c) Do not send CPU_UP_CANCELED to a subsystem which did not receive
   CPU_UP_PREPARE.

I would prefer c).

Signed-off-by: Ravikiran Thirumalai <kiran@scalex86.org>
Signed-off-by: Shai Fultheim <shai@scalex86.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] slab debug and ARCH_SLAB_MINALIGN don't get along
Kevin Hilman [Thu, 7 Dec 2006 04:32:11 +0000 (20:32 -0800)] 
[PATCH] slab debug and ARCH_SLAB_MINALIGN don't get along

When CONFIG_SLAB_DEBUG is used in combination with ARCH_SLAB_MINALIGN, some
debug flags should be disabled which depend on BYTES_PER_WORD alignment.

The disabling of these debug flags is not properly handled when
BYTES_PER_WORD < ARCH_SLAB_MEMALIGN < cache_line_size()

This patch fixes that and also adds an alignment check to
cache_alloc_debugcheck_after() when ARCH_SLAB_MINALIGN is used.

Signed-off-by: Kevin Hilman <khilman@mvista.com>
Cc: Pekka Enberg <penberg@cs.helsinki.fi>
Cc: Christoph Lameter <clameter@engr.sgi.com>
Cc: Manfred Spraul <manfred@colorfullife.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] htlb forget rss with pt sharing
Chen, Kenneth W [Thu, 7 Dec 2006 04:32:07 +0000 (20:32 -0800)] 
[PATCH] htlb forget rss with pt sharing

Imprecise RSS accounting is an irritating ill effect with pt sharing.  After
consulted with several VM experts, I have tried various methods to solve that
problem: (1) iterate through all mm_structs that share the PT and increment
count; (2) keep RSS count in page table structure and then sum them up at
reporting time.  None of the above methods yield any satisfactory
implementation.

Since process RSS accounting is pure information only, I propose we don't
count them at all for hugetlb page.  rlimit has such field, though there is
absolutely no enforcement on limiting that resource.  One other method is to
account all RSS at hugetlb mmap time regardless they are faulted or not.  I
opt for the simplicity of no accounting at all.

Hugetlb page are special, they are reserved up front in global reservation
pool and is not reclaimable.  From physical memory resource point of view, it
is already consumed regardless whether there are users using them.

If the concern is that RSS can be used to control resource allocation, we
already can specify hugetlb fs size limit and sysadmin can enforce that at
mount time.  Combined with the two points mentioned above, I fail to see if
there is anything got affected because of this patch.

Signed-off-by: Ken Chen <kenneth.w.chen@intel.com>
Acked-by: Hugh Dickins <hugh@veritas.com>
Cc: Dave McCracken <dmccr@us.ibm.com>
Cc: William Lee Irwin III <wli@holomorphy.com>
Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: David Gibson <david@gibson.dropbear.id.au>
Cc: Adam Litke <agl@us.ibm.com>
Cc: Paul Mundt <lethal@linux-sh.org>
Cc: "David S. Miller" <davem@davemloft.net>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] shared page table for hugetlb page
Chen, Kenneth W [Thu, 7 Dec 2006 04:32:03 +0000 (20:32 -0800)] 
[PATCH] shared page table for hugetlb page

Following up with the work on shared page table done by Dave McCracken.  This
set of patch target shared page table for hugetlb memory only.

The shared page table is particular useful in the situation of large number of
independent processes sharing large shared memory segments.  In the normal
page case, the amount of memory saved from process' page table is quite
significant.  For hugetlb, the saving on page table memory is not the primary
objective (as hugetlb itself already cuts down page table overhead
significantly), instead, the purpose of using shared page table on hugetlb is
to allow faster TLB refill and smaller cache pollution upon TLB miss.

With PT sharing, pte entries are shared among hundreds of processes, the cache
consumption used by all the page table is smaller and in return, application
gets much higher cache hit ratio.  One other effect is that cache hit ratio
with hardware page walker hitting on pte in cache will be higher and this
helps to reduce tlb miss latency.  These two effects contribute to higher
application performance.

Signed-off-by: Ken Chen <kenneth.w.chen@intel.com>
Acked-by: Hugh Dickins <hugh@veritas.com>
Cc: Dave McCracken <dmccr@us.ibm.com>
Cc: William Lee Irwin III <wli@holomorphy.com>
Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: David Gibson <david@gibson.dropbear.id.au>
Cc: Adam Litke <agl@us.ibm.com>
Cc: Paul Mundt <lethal@linux-sh.org>
Cc: "David S. Miller" <davem@davemloft.net>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] balance_pdgat() cleanup
Andrew Morton [Thu, 7 Dec 2006 04:32:01 +0000 (20:32 -0800)] 
[PATCH] balance_pdgat() cleanup

Despaghettify balance_pdgat() a bit.

Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] mm: add arch_alloc_page
Nick Piggin [Thu, 7 Dec 2006 04:32:00 +0000 (20:32 -0800)] 
[PATCH] mm: add arch_alloc_page

Add an arch_alloc_page to match arch_free_page.

Signed-off-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] new scheme to preempt swap token
Ashwin Chaugule [Thu, 7 Dec 2006 04:31:57 +0000 (20:31 -0800)] 
[PATCH] new scheme to preempt swap token

The new swap token patches replace the current token traversal algo.  The old
algo had a crude timeout parameter that was used to handover the token from
one task to another.  This algo, transfers the token to the tasks that are in
need of the token.  The urgency for the token is based on the number of times
a task is required to swap-in pages.  Accordingly, the priority of a task is
incremented if it has been badly affected due to swap-outs.  To ensure that
the token doesnt bounce around rapidly, the token holders are given a priority
boost.  The priority of tasks is also decremented, if their rate of swap-in's
keeps reducing.  This way, the condition to check whether to pre-empt the swap
token, is a matter of comparing two task's priority fields.

[akpm@osdl.org: cleanups]
Signed-off-by: Ashwin Chaugule <ashwin.chaugule@celunite.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] grab swap token reordered
Ashwin Chaugule [Thu, 7 Dec 2006 04:31:54 +0000 (20:31 -0800)] 
[PATCH] grab swap token reordered

Make sure the contention for the token happens _before_ any read-in and
kicks the swap-token algo only when the VM is under pressure.

Signed-off-by: Ashwin Chaugule <ashwin.chaugule@celunite.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] mm: incorrect VM_FAULT_OOM returns from drivers
Nick Piggin [Thu, 7 Dec 2006 04:31:53 +0000 (20:31 -0800)] 
[PATCH] mm: incorrect VM_FAULT_OOM returns from drivers

Some drivers are returning OOM when it is not in response to a memory
shortage.

Signed-off-by: Nick Piggin <npiggin@suse.de>
Cc: Dave Airlie <airlied@linux.ie>
Cc: Jaroslav Kysela <perex@suse.cz>
Cc: Takashi Iwai <tiwai@suse.de>
Cc: Greg KH <greg@kroah.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] oom: less memdie
Nick Piggin [Thu, 7 Dec 2006 04:31:52 +0000 (20:31 -0800)] 
[PATCH] oom: less memdie

Don't cause all threads in all other thread groups to gain TIF_MEMDIE
otherwise we'll get a thundering herd eating our memory reserve.  This may not
be the optimal scheme, but it fits our policy of allowing just one TIF_MEMDIE
in the system at once.

Signed-off-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] oom: cleanup messages
Nick Piggin [Thu, 7 Dec 2006 04:31:51 +0000 (20:31 -0800)] 
[PATCH] oom: cleanup messages

Clean up the OOM killer messages to be more consistent.

Signed-off-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] oom: don't kill unkillable children or siblings
Nick Piggin [Thu, 7 Dec 2006 04:31:50 +0000 (20:31 -0800)] 
[PATCH] oom: don't kill unkillable children or siblings

Abort the kill if any of our threads have OOM_DISABLE set.  Having this
test here also prevents any OOM_DISABLE child of the "selected" process
from being killed.

Signed-off-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] memory page_alloc zonelist caching reorder structure
Paul Jackson [Thu, 7 Dec 2006 04:31:49 +0000 (20:31 -0800)] 
[PATCH] memory page_alloc zonelist caching reorder structure

Rearrange the struct members in the 'struct zonelist_cache' structure, so
as to put the readonly (once initialized) z_to_n[] array first, where it
will come right after the zones[] array in struct zonelist.

This pretty much eliminates the chance that the two frequently written
elements of 'struct zonelist_cache', the fullzones bitmap and last_full_zap
times, will end up on the same cache line as the performance sensitive,
frequently read, never (after init) written zones[] array.

Keeping frequently written data off frequently read cache lines is good for
performance.

Thanks to Rohit Seth for the suggestion.

Signed-off-by: Paul Jackson <pj@sgi.com>
Cc: Rohit Seth <rohitseth@google.com>
Cc: Paul Menage <menage@google.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] memory page_alloc zonelist caching speedup
Paul Jackson [Thu, 7 Dec 2006 04:31:48 +0000 (20:31 -0800)] 
[PATCH] memory page_alloc zonelist caching speedup

Optimize the critical zonelist scanning for free pages in the kernel memory
allocator by caching the zones that were found to be full recently, and
skipping them.

Remembers the zones in a zonelist that were short of free memory in the
last second.  And it stashes a zone-to-node table in the zonelist struct,
to optimize that conversion (minimize its cache footprint.)

Recent changes:

    This differs in a significant way from a similar patch that I
    posted a week ago.  Now, instead of having a nodemask_t of
    recently full nodes, I have a bitmask of recently full zones.
    This solves a problem that last weeks patch had, which on
    systems with multiple zones per node (such as DMA zone) would
    take seeing any of these zones full as meaning that all zones
    on that node were full.

    Also I changed names - from "zonelist faster" to "zonelist cache",
    as that seemed to better convey what we're doing here - caching
    some of the key zonelist state (for faster access.)

    See below for some performance benchmark results.  After all that
    discussion with David on why I didn't need them, I went and got
    some ;).  I wanted to verify that I had not hurt the normal case
    of memory allocation noticeably.  At least for my one little
    microbenchmark, I found (1) the normal case wasn't affected, and
    (2) workloads that forced scanning across multiple nodes for
    memory improved up to 10% fewer System CPU cycles and lower
    elapsed clock time ('sys' and 'real').  Good.  See details, below.

    I didn't have the logic in get_page_from_freelist() for various
    full nodes and zone reclaim failures correct.  That should be
    fixed up now - notice the new goto labels zonelist_scan,
    this_zone_full, and try_next_zone, in get_page_from_freelist().

There are two reasons I persued this alternative, over some earlier
proposals that would have focused on optimizing the fake numa
emulation case by caching the last useful zone:

 1) Contrary to what I said before, we (SGI, on large ia64 sn2 systems)
    have seen real customer loads where the cost to scan the zonelist
    was a problem, due to many nodes being full of memory before
    we got to a node we could use.  Or at least, I think we have.
    This was related to me by another engineer, based on experiences
    from some time past.  So this is not guaranteed.  Most likely, though.

    The following approach should help such real numa systems just as
    much as it helps fake numa systems, or any combination thereof.

 2) The effort to distinguish fake from real numa, using node_distance,
    so that we could cache a fake numa node and optimize choosing
    it over equivalent distance fake nodes, while continuing to
    properly scan all real nodes in distance order, was going to
    require a nasty blob of zonelist and node distance munging.

    The following approach has no new dependency on node distances or
    zone sorting.

See comment in the patch below for a description of what it actually does.

Technical details of note (or controversy):

 - See the use of "zlc_active" and "did_zlc_setup" below, to delay
   adding any work for this new mechanism until we've looked at the
   first zone in zonelist.  I figured the odds of the first zone
   having the memory we needed were high enough that we should just
   look there, first, then get fancy only if we need to keep looking.

 - Some odd hackery was needed to add items to struct zonelist, while
   not tripping up the custom zonelists built by the mm/mempolicy.c
   code for MPOL_BIND.  My usual wordy comments below explain this.
   Search for "MPOL_BIND".

 - Some per-node data in the struct zonelist is now modified frequently,
   with no locking.  Multiple CPU cores on a node could hit and mangle
   this data.  The theory is that this is just performance hint data,
   and the memory allocator will work just fine despite any such mangling.
   The fields at risk are the struct 'zonelist_cache' fields 'fullzones'
   (a bitmask) and 'last_full_zap' (unsigned long jiffies).  It should
   all be self correcting after at most a one second delay.

 - This still does a linear scan of the same lengths as before.  All
   I've optimized is making the scan faster, not algorithmically
   shorter.  It is now able to scan a compact array of 'unsigned
   short' in the case of many full nodes, so one cache line should
   cover quite a few nodes, rather than each node hitting another
   one or two new and distinct cache lines.

 - If both Andi and Nick don't find this too complicated, I will be
   (pleasantly) flabbergasted.

 - I removed the comment claiming we only use one cachline's worth of
   zonelist.  We seem, at least in the fake numa case, to have put the
   lie to that claim.

 - I pay no attention to the various watermarks and such in this performance
   hint.  A node could be marked full for one watermark, and then skipped
   over when searching for a page using a different watermark.  I think
   that's actually quite ok, as it will tend to slightly increase the
   spreading of memory over other nodes, away from a memory stressed node.

===============

Performance - some benchmark results and analysis:

This benchmark runs a memory hog program that uses multiple
threads to touch alot of memory as quickly as it can.

Multiple runs were made, touching 12, 38, 64 or 90 GBytes out of
the total 96 GBytes on the system, and using 1, 19, 37, or 55
threads (on a 56 CPU system.)  System, user and real (elapsed)
timings were recorded for each run, shown in units of seconds,
in the table below.

Two kernels were tested - 2.6.18-mm3 and the same kernel with
this zonelist caching patch added.  The table also shows the
percentage improvement the zonelist caching sys time is over
(lower than) the stock *-mm kernel.

      number     2.6.18-mm3    zonelist-cache    delta (< 0 good) percent
 GBs    N   ------------    --------------    ---------------- systime
 mem threads   sys user  real   sys  user  real     sys  user  real  better
  12  1     153   24   177   151  24   176      -2     0    -1    1%
  12 19 99   22     8    99  22 8 0     0     0    0%
  12 37     111   25     6   112  25 6 1     0     0   -0%
  12 55     115   25     5   110  23 5      -5    -2     0    4%
  38  1     502   74   576   497  73   570      -5    -1    -6    0%
  38 19     426   78    48   373  76    39     -53    -2    -9   12%
  38 37     544   83    36   547  82    36 3    -1     0   -0%
  38 55     501   77    23   511  80    24      10     3     1   -1%
  64  1     917  125  1042   890 124  1014     -27    -1   -28    2%
  64 19    1118  138   119   965 141   103    -153     3   -16   13%
  64 37    1202  151    94  1136 150    81     -66    -1   -13    5%
  64 55    1118  141    61  1072 140    58     -46    -1    -3    4%
  90  1    1342  177  1519  1275 174  1450     -67    -3   -69    4%
  90 19    2392  199   192  2116 189   176    -276   -10   -16   11%
  90 37    3313  238   175  2972 225   145    -341   -13   -30   10%
  90 55    1948  210   104  1843 213   100    -105     3    -4    5%

Notes:
 1) This test ran a memory hog program that started a specified number N of
    threads, and had each thread allocate and touch 1/N'th of
    the total memory to be used in the test run in a single loop,
    writing a constant word to memory, one store every 4096 bytes.
    Watching this test during some earlier trial runs, I would see
    each of these threads sit down on one CPU and stay there, for
    the remainder of the pass, a different CPU for each thread.

 2) The 'real' column is not comparable to the 'sys' or 'user' columns.
    The 'real' column is seconds wall clock time elapsed, from beginning
    to end of that test pass.  The 'sys' and 'user' columns are total
    CPU seconds spent on that test pass.  For a 19 thread test run,
    for example, the sum of 'sys' and 'user' could be up to 19 times the
    number of 'real' elapsed wall clock seconds.

 3) Tests were run on a fresh, single-user boot, to minimize the amount
    of memory already in use at the start of the test, and to minimize
    the amount of background activity that might interfere.

 4) Tests were done on a 56 CPU, 28 Node system with 96 GBytes of RAM.

 5) Notice that the 'real' time gets large for the single thread runs, even
    though the measured 'sys' and 'user' times are modest.  I'm not sure what
    that means - probably something to do with it being slow for one thread to
    be accessing memory along ways away.  Perhaps the fake numa system, running
    ostensibly the same workload, would not show this substantial degradation
    of 'real' time for one thread on many nodes -- lets hope not.

 6) The high thread count passes (one thread per CPU - on 55 of 56 CPUs)
    ran quite efficiently, as one might expect.  Each pair of threads needed
    to allocate and touch the memory on the node the two threads shared, a
    pleasantly parallizable workload.

 7) The intermediate thread count passes, when asking for alot of memory forcing
    them to go to a few neighboring nodes, improved the most with this zonelist
    caching patch.

Conclusions:
 * This zonelist cache patch probably makes little difference one way or the
   other for most workloads on real numa hardware, if those workloads avoid
   heavy off node allocations.
 * For memory intensive workloads requiring substantial off-node allocations
   on real numa hardware, this patch improves both kernel and elapsed timings
   up to ten per-cent.
 * For fake numa systems, I'm optimistic, but will have to leave that up to
   Rohit Seth to actually test (once I get him a 2.6.18 backport.)

Signed-off-by: Paul Jackson <pj@sgi.com>
Cc: Rohit Seth <rohitseth@google.com>
Cc: Christoph Lameter <clameter@engr.sgi.com>
Cc: David Rientjes <rientjes@cs.washington.edu>
Cc: Paul Menage <menage@google.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] Get rid of zone_table[]
Christoph Lameter [Thu, 7 Dec 2006 04:31:45 +0000 (20:31 -0800)] 
[PATCH] Get rid of zone_table[]

The zone table is mostly not needed.  If we have a node in the page flags
then we can get to the zone via NODE_DATA() which is much more likely to be
already in the cpu cache.

In case of SMP and UP NODE_DATA() is a constant pointer which allows us to
access an exact replica of zonetable in the node_zones field.  In all of
the above cases there will be no need at all for the zone table.

The only remaining case is if in a NUMA system the node numbers do not fit
into the page flags.  In that case we make sparse generate a table that
maps sections to nodes and use that table to to figure out the node number.
 This table is sized to fit in a single cache line for the known 32 bit
NUMA platform which makes it very likely that the information can be
obtained without a cache miss.

For sparsemem the zone table seems to be have been fairly large based on
the maximum possible number of sections and the number of zones per node.
There is some memory saving by removing zone_table.  The main benefit is to
reduce the cache foootprint of the VM from the frequent lookups of zones.
Plus it simplifies the page allocator.

[akpm@osdl.org: build fix]
Signed-off-by: Christoph Lameter <clameter@sgi.com>
Cc: Dave Hansen <haveblue@us.ibm.com>
Cc: Andy Whitcroft <apw@shadowen.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] __unmap_hugepage_range(): add comment
Chen, Kenneth W [Thu, 7 Dec 2006 04:31:39 +0000 (20:31 -0800)] 
[PATCH] __unmap_hugepage_range(): add comment

Signed-off-by: Ken Chen <kenneth.w.chen@intel.com>
Cc: David Gibson <david@gibson.dropbear.id.au>
Cc: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] memory page alloc minor cleanups
Paul Jackson [Thu, 7 Dec 2006 04:31:38 +0000 (20:31 -0800)] 
[PATCH] memory page alloc minor cleanups

- s/freeliest/freelist/ spelling fix

- Check for NULL *z zone seems useless - even if it could happen, so
  what?  Perhaps we should have a check later on if we are faced with an
  allocation request that is not allowed to fail - shouldn't that be a
  serious kernel error, passing an empty zonelist with a mandate to not
  fail?

- Initializing 'z' to zonelist->zones can wait until after the first
  get_page_from_freelist() fails; we only use 'z' in the wakeup_kswapd()
  loop, so let's initialize 'z' there, in a 'for' loop.  Seems clearer.

- Remove superfluous braces around a break

- Fix a couple errant spaces

- Adjust indentation on the cpuset_zone_allowed() check, to match the
  lines just before it -- seems easier to read in this case.

- Add another set of braces to the zone_watermark_ok logic

From: Paul Jackson <pj@sgi.com>

  Backout one item from a previous "memory page_alloc minor cleanups" patch.
   Until and unless we are certain that no one can ever pass an empty zonelist
  to __alloc_pages(), this check for an empty zonelist (or some BUG
  equivalent) is essential.  The code in get_page_from_freelist() blow ups if
  passed an empty zonelist.

Signed-off-by: Paul Jackson <pj@sgi.com>
Acked-by: Christoph Lameter <clameter@sgi.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] uml: workqueue build fix
Andrew Morton [Thu, 7 Dec 2006 04:31:36 +0000 (20:31 -0800)] 
[PATCH] uml: workqueue build fix

  arch/um/drivers/chan_kern.c:643: error: conflicting types for 'chan_interrupt'
  arch/um/include/chan_kern.h:31: error: previous declaration of 'chan_interrupt'

Cc: David Howells <dhowells@redhat.com>
Cc: Jeff Dike <jdike@addtoit.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] skip data conversion in compat_sys_mount when data_page is NULL
Andrey Mirkin [Thu, 7 Dec 2006 04:31:35 +0000 (20:31 -0800)] 
[PATCH] skip data conversion in compat_sys_mount when data_page is NULL

OpenVZ Linux kernel team has found a problem with mounting in compat mode.

Simple command "mount -t smbfs ..." on Fedora Core 5 distro in 32-bit mode
leads to oops:

  Unable to handle kernel NULL pointer dereference at 0000000000000000 RIP: compat_sys_mount+0xd6/0x290
  Process mount (pid: 14656, veid=300, threadinfo ffff810034d30000, task ffff810034c86bc0)
  Call Trace: ia32_sysret+0x0/0xa

The problem is that data_page pointer can be NULL, so we should skip data
conversion in this case.

Signed-off-by: Andrey Mirkin <amirkin@openvz.org>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] drm-sis linkage fix
Andrew Morton [Thu, 7 Dec 2006 04:31:33 +0000 (20:31 -0800)] 
[PATCH] drm-sis linkage fix

Fix http://bugzilla.kernel.org/show_bug.cgi?id=7606

WARNING: "drm_sman_set_manager" [drivers/char/drm/sis.ko] undefined!

Cc: <daniel-silveira@gee.inatel.br>
Cc: Dave Airlie <airlied@linux.ie>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] add bottom_half.h
Andrew Morton [Thu, 7 Dec 2006 04:31:30 +0000 (20:31 -0800)] 
[PATCH] add bottom_half.h

With CONFIG_SMP=n:

  drivers/input/ff-memless.c:384: warning: implicit declaration of function 'local_bh_disable'
  drivers/input/ff-memless.c:393: warning: implicit declaration of function 'local_bh_enable'

Really linux/spinlock.h should include linux/interrupt.h.  But interrupt.h
includes sched.h which will need spinlock.h.

So the patch breaks the _bh declarations out into a separate header and
includes it in both interrupt.h and spinlock.h.

Cc: "Randy.Dunlap" <rdunlap@xenotime.net>
Cc: Andi Kleen <ak@suse.de>
Cc: <stable@kernel.org>
Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] A few small additions and corrections to README
Jesper Juhl [Wed, 6 Dec 2006 23:45:58 +0000 (00:45 +0100)] 
[PATCH] A few small additions and corrections to README

Here's a small patch which

 - adds a few archs to the current list of supported platforms.
 - adds a few missing slashes at the end of URLs.
 - adds a few references to additional documentation.
 - adds "make config" to the list of possible configuration targets.
 - makes a few other minor changes.

Signed-off-by: Jesper Juhl <jesper.juhl@gmail.com>
[ Ben Nizette <ben.nizette@iinet.net.au> points out AVR32 arch too ]
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] Clean up 'make help' output for documentation targets.
Jesper Juhl [Wed, 6 Dec 2006 23:16:26 +0000 (00:16 +0100)] 
[PATCH] Clean up 'make help' output for documentation targets.

Here's a patch that cleans up the "make help" output a bit for the
documentation targets.

Currently the documentation targets are listed completely different than
all the other targets :

  Documentation targets:
    Linux kernel internal documentation in different formats:
    xmldocs (XML DocBook), psdocs (Postscript), pdfdocs (PDF)
    htmldocs (HTML), mandocs (man pages, use installmandocs to install)

with this patch they are more in line with the rest of the output :

  Documentation targets:
   Linux kernel internal documentation in different formats:
    htmldocs        - HTML
    installmandocs  - install man pages generated by mandocs
    mandocs         - man pages
    pdfdocs         - PDF
    psdocs          - Postscript
    xmldocs         - XML DocBook

Signed-off-by: Jesper Juhl <jesper.juhl@gmail.com>
Acked-by: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years agoMerge branch 'upstream' of git://ftp.linux-mips.org/pub/scm/upstream-linus
Linus Torvalds [Thu, 7 Dec 2006 00:17:37 +0000 (16:17 -0800)] 
Merge branch 'upstream' of git://ftp.linux-mips.org/upstream-linus

* 'upstream' of git://ftp.linux-mips.org/pub/scm/upstream-linus:
  [MIPS] Import updates from i386's i8259.c
  [MIPS] *-berr: Header inclusions for DEC bus error handlers
  [MIPS] Compile __do_IRQ() when really needed
  [MIPS] genirq: use name instead of typename
  [MIPS] Do not use handle_level_irq for ioasic_dma_irq_type.
  [MIPS] pte_offset(dir,addr): parenthesis fix

18 years agoMerge branch 'release' of master.kernel.org:/home/ftp/pub/scm/linux/kernel/git/aegl...
Linus Torvalds [Thu, 7 Dec 2006 00:16:35 +0000 (16:16 -0800)] 
Merge branch 'release' of /home/ftp/linux/kernel/git/aegl/linux-2.6

* 'release' of master.kernel.org:/home/ftp/pub/scm/linux/kernel/git/aegl/linux-2.6:
  [IA64] Fix pci.c kernel compilation breakage.

18 years ago[PATCH] ... and then some more work_struct-induced breakage (ibmvscsi)
Al Viro [Wed, 6 Dec 2006 21:15:22 +0000 (21:15 +0000)] 
[PATCH] ... and then some more work_struct-induced breakage (ibmvscsi)

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] ... and more work_struct-induced breakage (mips)
Al Viro [Wed, 6 Dec 2006 19:50:06 +0000 (19:50 +0000)] 
[PATCH] ... and more work_struct-induced breakage (mips)

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] More work_struct induced breakage (s390)
Al Viro [Wed, 6 Dec 2006 19:18:20 +0000 (19:18 +0000)] 
[PATCH] More work_struct induced breakage (s390)

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years agox86[-64]:Remove 'volatile' from atomic_t
Linus Torvalds [Wed, 6 Dec 2006 22:42:57 +0000 (14:42 -0800)] 
x86[-64]:Remove 'volatile' from atomic_t

Any code that relies on the volatile would be a bug waiting to happen
anyway.

Don't encourage people to think that putting 'volatile' on data
structures somehow fixes problems.  We should always use proper locking
(and other serialization) techniques.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] Remove 'volatile' from spinlock_types
Art Haas [Wed, 6 Dec 2006 20:45:53 +0000 (14:45 -0600)] 
[PATCH] Remove 'volatile' from spinlock_types

This is a resubmission of patches originally created by Ingo Molnar.
The link below is the initial (?) posting of the patch.

  http://marc.theaimsgroup.com/?l=linux-kernel&m=115217423929806&w=2

Remove 'volatile' from spinlock_types as it causes GCC to generate bad
code (see link) and locking should be used on kernel data.

Signed-off-by: Art Haas <ahaas@airmail.net>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[IA64] Fix pci.c kernel compilation breakage.
Peter Chubb [Tue, 5 Dec 2006 01:25:31 +0000 (12:25 +1100)] 
[IA64] Fix pci.c kernel compilation breakage.

The recent change to convert the is_enabled flag in the PCI device to an
atomic count broke the IA64 compilation.

As pcibios_disable_device is only ever called if the reference count
is zero, convert the if to a BUG_ON.

Signed-off-by: Peter Chubb <peterc@gelato.unsw.edu.au>
Signed-off-by: Tony Luck <tony.luck@intel.com>
18 years ago[MIPS] Import updates from i386's i8259.c
Atsushi Nemoto [Wed, 6 Dec 2006 17:04:17 +0000 (02:04 +0900)] 
[MIPS] Import updates from i386's i8259.c

Import many updates from i386's i8259.c, especially genirq transitions.

Signed-off-by: Atsushi Nemoto <anemo@mba.ocn.ne.jp>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] *-berr: Header inclusions for DEC bus error handlers
Maciej W. Rozycki [Wed, 6 Dec 2006 11:50:23 +0000 (11:50 +0000)] 
[MIPS] *-berr: Header inclusions for DEC bus error handlers

 A fixup to add missing header inclusions for bus error handlers for
DECstation system after the recent switch to get_irq_regs().

Signed-off-by: Maciej W. Rozycki <macro@linux-mips.org>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Compile __do_IRQ() when really needed
Franck Bui-Huu [Fri, 1 Dec 2006 17:22:27 +0000 (18:22 +0100)] 
[MIPS] Compile __do_IRQ() when really needed

__do_IRQ() is needed only by irq handlers that can't use
default handlers defined in kernel/irq/chip.c.

For others platforms there's no need to compile this function
since it won't be used. For those platforms this patch defines
GENERIC_HARDIRQS_NO__DO_IRQ symbol which is used exactly for
this purpose.

Futhermore for platforms which do not use __do_IRQ(), end()
method which is part of the 'irq_chip' structure is not used.
This patch simply removes this method in this case.

Signed-off-by: Franck Bui-Huu <fbuihuu@gmail.com>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] genirq: use name instead of typename
Atsushi Nemoto [Tue, 5 Dec 2006 16:20:57 +0000 (01:20 +0900)] 
[MIPS] genirq: use name instead of typename

The "typename" field was obsoleted by the "name" field.

Signed-off-by: Atsushi Nemoto <anemo@mba.ocn.ne.jp>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Do not use handle_level_irq for ioasic_dma_irq_type.
Atsushi Nemoto [Fri, 1 Dec 2006 15:08:03 +0000 (00:08 +0900)] 
[MIPS] Do not use handle_level_irq for ioasic_dma_irq_type.

Signed-off-by: Atsushi Nemoto <anemo@mba.ocn.ne.jp>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] pte_offset(dir,addr): parenthesis fix
Franck Bui-Huu [Tue, 5 Dec 2006 09:39:56 +0000 (10:39 +0100)] 
[MIPS] pte_offset(dir,addr): parenthesis fix

This patch adds missing parenthesis around 'dir' argument in pte_offset()
macro definition.

It also removes an extra space in the definition of pte_offset_kernel()
macro.

Signed-off-by: Franck Bui-Huu <fbuihuu@gmail.com>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[PATCH] drivers/{char|isdn}: work_struct-induced breakage
Al Viro [Wed, 6 Dec 2006 18:41:45 +0000 (18:41 +0000)] 
[PATCH] drivers/{char|isdn}: work_struct-induced breakage

part 1 of fsck-knows-how-many

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] hamradio/dmascc: fix up work_struct-induced breakage
Al Viro [Wed, 6 Dec 2006 18:51:40 +0000 (18:51 +0000)] 
[PATCH] hamradio/dmascc: fix up work_struct-induced breakage

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years agoMerge master.kernel.org:/pub/scm/linux/kernel/git/lethal/sh-2.6
Linus Torvalds [Wed, 6 Dec 2006 16:10:55 +0000 (08:10 -0800)] 
Merge /pub/scm/linux/kernel/git/lethal/sh-2.6

* master.kernel.org:/pub/scm/linux/kernel/git/lethal/sh-2.6: (43 commits)
  sh: sh775x/titan fixes for irq header changes.
  sh: update r7780rp defconfig.
  sh: compile fixes for header cleanup.
  sh: Fixup pte_mkhuge() build failure.
  sh: set KBUILD_IMAGE to something sensible.
  sh: show held locks in stack trace with lockdep.
  sh: platform_pata support for R7780RP
  sh: stacktrace/lockdep/irqflags tracing support.
  sh: Fixup movli.l/movco.l atomic ops for gcc4.
  sh: dyntick infrastructure.
  sh: Clock framework tidying.
  sh: Turn off IRQs around get_timer_offset() calls.
  sh: Get the PGD right in oops case with 64-bit PTEs.
  sh: Fix store queue bitmap end.
  sh: More flexible + SH7780 earlyprintk SCIF support.
  sh: Fixup various PAGE_SIZE == 4096 assumptions.
  sh: Fixup 4K irq stacks.
  sh: dma-api channel capability extensions.
  sh: Drop name overload in dma-sh.
  sh: Make dma-isa depend on ISA_DMA_API.
  ...

18 years agoMerge git://git.infradead.org/users/dhowells/workq-2.6
Linus Torvalds [Wed, 6 Dec 2006 16:01:37 +0000 (08:01 -0800)] 
Merge git://git.infradead.org/users/dhowells/workq-2.6

* git://git.infradead.org/users/dhowells/workq-2.6:
  Actually update the fixed up compile failures.
  WorkQueue: Fix up arch-specific work items where possible
  WorkStruct: make allyesconfig
  WorkStruct: Pass the work_struct pointer instead of context data
  WorkStruct: Merge the pending bit into the wq_data pointer
  WorkStruct: Typedef the work function prototype
  WorkStruct: Separate delayable and non-delayable events.

18 years ago[PATCH] uclinux: fix mmap() of directory for nommu case
Mike Frysinger [Wed, 6 Dec 2006 02:02:59 +0000 (12:02 +1000)] 
[PATCH] uclinux: fix mmap() of directory for nommu case

I was playing with blackfin when i hit a neat bug ... doing an open() on a
directory and then passing that fd to mmap() would cause the kernel to hang

after poking into the code a bit more, i found that
mm/nommu.c:validate_mmap_request() checks the length and if it is 0, just
returns the address ... this is in stark contrast to mmu's
mm/mmap.c:do_mmap_pgoff() where it returns -EINVAL for 0 length requests ...
i then noticed that some other parts of the logic is out of date between the
two funcs, so perhaps that's the easy fix ?

Signed-off-by: Greg Ungerer <gerg@uclinux.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] m68knommu: remove FP conditionals in ucontext struct
Gavin Lambert [Wed, 6 Dec 2006 01:57:53 +0000 (11:57 +1000)] 
[PATCH] m68knommu: remove FP conditionals in ucontext struct

The first patch is to the 2.6 kernel include file (for m68knommu), to get
rid of the conditional definitions, otherwise the structures have different
sizes depending on whether there's an FPU or not.

Signed-off-by: Greg Ungerer <gerg@uclinux.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] m68knommu: ColdFire serial driver fixes
Greg Ungerer [Wed, 6 Dec 2006 01:49:34 +0000 (11:49 +1000)] 
[PATCH] m68knommu: ColdFire serial driver fixes

Some updates for the old ColdFire serial driver:

 . support 3 and 4 UARTs on some ColdFire parts that have them
 . enable multifunction pins to serial for 527x CPU's
 . support the 5272 UART's fractional baud rate divisor
 . switch driver name to "mcfserial"

Signed-off-by: Greg Ungerer <gerg@uclinux.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] m68knommu: switch 68360 to using rtc_time
Greg Ungerer [Wed, 6 Dec 2006 01:43:14 +0000 (11:43 +1000)] 
[PATCH] m68knommu: switch 68360 to using rtc_time

Adds support for RTCs (through genrtc) for M68KNOMMU.

Board-specific code will have to link the appropriate RTC driver to the
mach_hwclk callback, at minimum.

This patch switches the 68360 code over to using rtc_time.

Signed-off-by: Gavin Lambert <gavinl@compacsort.com>
Signed-off-by: Greg Ungerer <gerg@uclinux.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] m68knommu: fix timer register access on 523x ColdFire platforms
Greg Ungerer [Wed, 6 Dec 2006 01:36:59 +0000 (11:36 +1000)] 
[PATCH] m68knommu: fix timer register access on 523x ColdFire platforms

The 523x timer TRR register is a full 32bits, the older register (on
other ColdFire parts) was only 16 bits.  Use the right type of
__raw_read when accessing it.

Problem found by Yaroslav Vinogradov <yaroslav.vinogradov@freescale.com>

Signed-off-by: Greg Ungerer <gerg@uclinux.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] m68knommu: implement irq_canonicalize()
Greg Ungerer [Wed, 6 Dec 2006 01:36:13 +0000 (11:36 +1000)] 
[PATCH] m68knommu: implement irq_canonicalize()

Add a null definition for irq_canonicalize(). It is used in the gerneric
serial subsystem code, can't compile without it.

Signed-off-by: Greg Ungerer <gerg@uclinux.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] m68knommu: create rtc.h
Greg Ungerer [Wed, 6 Dec 2006 01:35:16 +0000 (11:35 +1000)] 
[PATCH] m68knommu: create rtc.h

This adds support for RTCs (through genrtc) for M68KNOMMU.

Board-specific code will have to link the appropriate RTC driver to the
mach_hwclk callback, at minimum.

Signed-off-by: Gavin Lambert <gavinl@compacsort.com>
Signed-off-by: Greg Ungerer <gerg@uclinux.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years agoActually update the fixed up compile failures.
David Howells [Wed, 6 Dec 2006 15:02:26 +0000 (15:02 +0000)] 
Actually update the fixed up compile failures.

Signed-Off-By: David Howells <dhowells@redhat.com>
18 years agoMerge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux-2.6
David Howells [Wed, 6 Dec 2006 15:01:18 +0000 (15:01 +0000)] 
Merge branch 'master' of git://git./linux/kernel/git/torvalds/linux-2.6

Conflicts:

drivers/pcmcia/ds.c

Fix up merge failures with Linus's head and fix new compile failures.

Signed-Off-By: David Howells <dhowells@redhat.com>
18 years agosh: sh775x/titan fixes for irq header changes.
Jamie Lenehan [Wed, 6 Dec 2006 03:05:02 +0000 (12:05 +0900)] 
sh: sh775x/titan fixes for irq header changes.

The following moves the creation of IPR interupts into setup-7750.c
and updates a few other things to make it all work after the "Drop
CPU subtype IRQ headers" commit. It boots and runs fine on my titan
board.

 - adds an ipr_idx to the ipr_data and uses a function in the subtype
   code to calculate the address of the IPR registers

 - adds a function to enable individual interrupt mode for externals
   in the subtype code and calls that from the titan board code
   instead of doing it directly.

 - I changed the shift in the ipr_data to be the actual # of bits to
   shift, instead of the numnber / 4 - made it easier to match with
   the manual.

Signed-off-by: Jamie Lenehan <lenehan@twibble.org>
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
18 years agosh: update r7780rp defconfig.
Paul Mundt [Wed, 6 Dec 2006 03:02:01 +0000 (12:02 +0900)] 
sh: update r7780rp defconfig.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
18 years agosh: compile fixes for header cleanup.
Paul Mundt [Wed, 6 Dec 2006 02:24:48 +0000 (11:24 +0900)] 
sh: compile fixes for header cleanup.

Since some header inclusion paths were cleaned up, compilation
broke. Add in the headers we need directly to build again.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
18 years agosh: Fixup pte_mkhuge() build failure.
Paul Mundt [Wed, 6 Dec 2006 02:20:53 +0000 (11:20 +0900)] 
sh: Fixup pte_mkhuge() build failure.

When hugetlbpage support isn't enabled, this can be bogus.
Wrap it back in _PAGE_FLAGS_HARD to avoid changes to the
base PTE when not aiming for larger sizes.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
18 years agosh: set KBUILD_IMAGE to something sensible.
Paul Mundt [Wed, 6 Dec 2006 02:08:49 +0000 (11:08 +0900)] 
sh: set KBUILD_IMAGE to something sensible.

This was missing for sh too, wire it up..

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
18 years agosh: show held locks in stack trace with lockdep.
Paul Mundt [Wed, 6 Dec 2006 02:07:51 +0000 (11:07 +0900)] 
sh: show held locks in stack trace with lockdep.

Follows the same change as other architectures..

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
18 years agosh: platform_pata support for R7780RP
Paul Mundt [Wed, 6 Dec 2006 01:43:44 +0000 (10:43 +0900)] 
sh: platform_pata support for R7780RP

This adds a platform device for the directly connected
CF interface on R7780RP boards, for use with the
pata_platform libata driver.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
18 years agosh: stacktrace/lockdep/irqflags tracing support.
Paul Mundt [Mon, 4 Dec 2006 09:17:28 +0000 (18:17 +0900)] 
sh: stacktrace/lockdep/irqflags tracing support.

Wire up all of the essentials for lockdep..

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
18 years agosh: Fixup movli.l/movco.l atomic ops for gcc4.
Paul Mundt [Fri, 1 Dec 2006 05:32:54 +0000 (14:32 +0900)] 
sh: Fixup movli.l/movco.l atomic ops for gcc4.

gcc4 gets a bit pissy about the outputs:

include/asm/atomic.h: In function 'atomic_add':
include/asm/atomic.h:37: error: invalid lvalue in asm statement
include/asm/atomic.h:30: error: invalid lvalue in asm output 1
...

this ended up being a thinko anyways, so just fix it up.

Verified for proper behaviour with the older toolchains, too.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
18 years agosh: dyntick infrastructure.
Paul Mundt [Fri, 1 Dec 2006 04:23:47 +0000 (13:23 +0900)] 
sh: dyntick infrastructure.

This adds basic NO_IDLE_HZ support to the SH timer API so timers
are able to wire it up. Taken from the ARM version, as it fit in
to our API with very few changes needed.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
18 years agosh: Clock framework tidying.
Paul Mundt [Fri, 1 Dec 2006 04:15:14 +0000 (13:15 +0900)] 
sh: Clock framework tidying.

This syncs up the SH clock framework with the linux/clk.h API,
for which there were only some minor changes required, namely
the clk_get() dev_id and subsequent callsites.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
18 years agosh: Turn off IRQs around get_timer_offset() calls.
Paul Mundt [Fri, 1 Dec 2006 04:12:05 +0000 (13:12 +0900)] 
sh: Turn off IRQs around get_timer_offset() calls.

Since all of the sys_timer sources currently do this on their own
within the ->get_offset() path, it's more sensible to just have
the caller take care of it when grabbing xtime_lock. Incidentally,
this is more in line with what others (ie, ARM) are doing already.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
18 years agosh: Get the PGD right in oops case with 64-bit PTEs.
Paul Mundt [Fri, 1 Dec 2006 03:14:11 +0000 (12:14 +0900)] 
sh: Get the PGD right in oops case with 64-bit PTEs.

Previously this was using a static pgd shift in the reporting
code, simply flip this to PGDIR_SHIFT which does the right
thing depending on varying PTE magnitudes on the SH-X2 MMU.

While we're at it, and since it's been recently added, use
get_TTB() for fetching the TTB, rather than the open coded
instructions.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>