]> err.no Git - linux-2.6/log
linux-2.6
18 years ago[PATCH] kernel/: small cleanups
Adrian Bunk [Sun, 8 Jan 2006 09:02:17 +0000 (01:02 -0800)]
[PATCH] kernel/: small cleanups

This patch contains the following cleanups:
- make needlessly global functions static
- every file should include the headers containing the prototypes for
  it's global functions

Signed-off-by: Adrian Bunk <bunk@stusta.de>
Acked-by: "Paul E. McKenney" <paulmck@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] drivers/isdn/: "extern inline" -> "static inline"
Adrian Bunk [Sun, 8 Jan 2006 09:02:16 +0000 (01:02 -0800)]
[PATCH] drivers/isdn/: "extern inline" -> "static inline"

"extern inline" -> "static inline"

Since there's no pullphone() function this patch removes the dead
prototype.

Signed-off-by: Adrian Bunk <bunk@stusta.de>
Acked-by: Karsten Keil <kkeil@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] move rtc_interrupt() prototype to rtc.h
Adrian Bunk [Sun, 8 Jan 2006 09:02:15 +0000 (01:02 -0800)]
[PATCH] move rtc_interrupt() prototype to rtc.h

This patch moves the rtc_interrupt() prototype to rtc.h and removes the
prototypes from C files.

It also renames static rtc_interrupt() functions in
arch/arm/mach-integrator/time.c and arch/sh64/kernel/time.c to avoid compile
problems.

Signed-off-by: Adrian Bunk <bunk@stusta.de>
Signed-off-by: Paul Gortmaker <p_gortmaker@yahoo.com>
Acked-by: Paul Mundt <lethal@linux-sh.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] Fix and add EXPORT_SYMBOL(filemap_write_and_wait)
OGAWA Hirofumi [Sun, 8 Jan 2006 09:02:14 +0000 (01:02 -0800)]
[PATCH] Fix and add EXPORT_SYMBOL(filemap_write_and_wait)

This patch add EXPORT_SYMBOL(filemap_write_and_wait) and use it.

See mm/filemap.c:

And changes the filemap_write_and_wait() and filemap_write_and_wait_range().

Current filemap_write_and_wait() doesn't wait if filemap_fdatawrite()
returns error.  However, even if filemap_fdatawrite() returned an
error, it may have submitted the partially data pages to the device.
(e.g. in the case of -ENOSPC)

<quotation>
Andrew Morton writes,

If filemap_fdatawrite() returns an error, this might be due to some
I/O problem: dead disk, unplugged cable, etc.  Given the generally
crappy quality of the kernel's handling of such exceptions, there's a
good chance that the filemap_fdatawait() will get stuck in D state
forever.
</quotation>

So, this patch doesn't wait if filemap_fdatawrite() returns the -EIO.

Trond, could you please review the nfs part?  Especially I'm not sure,
nfs must use the "filemap_fdatawrite(inode->i_mapping) == 0", or not.

Acked-by: Trond Myklebust <trond.myklebust@fys.uio.no>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] fat: support a truncate() for expanding size (generic_cont_expand)
OGAWA Hirofumi [Sun, 8 Jan 2006 09:02:13 +0000 (01:02 -0800)]
[PATCH] fat: support a truncate() for expanding size (generic_cont_expand)

This patch changes generic_cont_expand(), in order to share the code
with fatfs.

  - Use vmtruncate() if ->prepare_write() returns a error.

Even if ->prepare_write() returns an error, it may already have added some
blocks.  So, this truncates blocks outside of ->i_size by vmtruncate().

  - Add generic_cont_expand_simple().

The generic_cont_expand_simple() assumes that ->prepare_write() can handle
the block boundary.  With this, we don't need to care the extra byte.

And for expanding a file size by truncate(), fatfs uses the
added generic_cont_expand_simple().

Signed-off-by: OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] export/change sync_page_range/_nolock()
OGAWA Hirofumi [Sun, 8 Jan 2006 09:02:12 +0000 (01:02 -0800)]
[PATCH] export/change sync_page_range/_nolock()

This exports/changes the sync_page_range/_nolock().  The fatfs needs
sync_page_range/_nolock() for expanding truncate, and changes "size_t count"
to "loff_t count".

Signed-off-by: OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] fat: support ->direct_IO()
OGAWA Hirofumi [Sun, 8 Jan 2006 09:02:11 +0000 (01:02 -0800)]
[PATCH] fat: support ->direct_IO()

This patch add to support of ->direct_IO() for mostly read.

The user of this seems to want to use for streaming read.  So, current direct
I/O has limitation, it can only overwrite.  (For write operation, mainly we
need to handle the hole etc..)

Signed-off-by: OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] fat: s/EXPORT_SYMBOL/EXPORT_SYMBOL_GPL/
OGAWA Hirofumi [Sun, 8 Jan 2006 09:02:10 +0000 (01:02 -0800)]
[PATCH] fat: s/EXPORT_SYMBOL/EXPORT_SYMBOL_GPL/

All EXPORT_SYMBOL of fatfs is only for vfat/msdos. _GPL would be proper.

Signed-off-by: OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] fat: add the read/writepages()
OGAWA Hirofumi [Sun, 8 Jan 2006 09:02:10 +0000 (01:02 -0800)]
[PATCH] fat: add the read/writepages()

Signed-off-by: OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] fat: use sb_find_get_block() instead of sb_getblk()
OGAWA Hirofumi [Sun, 8 Jan 2006 09:02:09 +0000 (01:02 -0800)]
[PATCH] fat: use sb_find_get_block() instead of sb_getblk()

We don't need to allocate buffer for checking the buffer is uptodate.  This
use sb_find_get_block() instead, and if it returns NULL it's not uptodate.

Signed-off-by: OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] fat: move fat_clusters_flush() to write_super()
OGAWA Hirofumi [Sun, 8 Jan 2006 09:02:08 +0000 (01:02 -0800)]
[PATCH] fat: move fat_clusters_flush() to write_super()

It is overkill to update the FS_INFO whenever modifying
prev_free/free_clusters, because those are just a hint.

So, this patch uses ->write_super() for updating FS_INFO instead.

Signed-off-by: OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] IRQ type flags
Russell King [Sun, 8 Jan 2006 09:02:07 +0000 (01:02 -0800)]
[PATCH] IRQ type flags

Some ARM platforms have the ability to program the interrupt controller to
detect various interrupt edges and/or levels.  For some platforms, this is
critical to setup correctly, particularly those which the setting is dependent
on the device.

Currently, ARM drivers do (eg) the following:

err = request_irq(irq, ...);

set_irq_type(irq, IRQT_RISING);

However, if the interrupt has previously been programmed to be level sensitive
(for whatever reason) then this will cause an interrupt storm.

Hence, if we combine set_irq_type() with request_irq(), we can then safely set
the type prior to unmasking the interrupt.  The unfortunate problem is that in
order to support this, these flags need to be visible outside of the ARM
architecture - drivers such as smc91x need these flags and they're
cross-architecture.

Finally, the SA_TRIGGER_* flag passed to request_irq() should reflect the
property that the device would like.  The IRQ controller code should do its
best to select the most appropriate supported mode.

Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] new driver synclink_gt
Paul Fulghum [Sun, 8 Jan 2006 09:02:06 +0000 (01:02 -0800)]
[PATCH] new driver synclink_gt

New character device driver for the SyncLink GT and SyncLink AC families of
synchronous and asynchronous serial adapters

Signed-off-by: Paul Fulghum <paulkf@microgate.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] fix more missing includes
Tim Schmielau [Sun, 8 Jan 2006 09:02:05 +0000 (01:02 -0800)]
[PATCH] fix more missing includes

Include fixes for 2.6.14-git11.  Should allow to remove sched.h from
module.h on i386, x86_64, arm, ia64, ppc, ppc64, and s390.  Probably more
to come since I haven't yet checked the other archs.

Signed-off-by: Tim Schmielau <tim@physik3.uni-rostock.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] cpuset: skip rcu check if task is in root cpuset
Paul Jackson [Sun, 8 Jan 2006 09:02:04 +0000 (01:02 -0800)]
[PATCH] cpuset: skip rcu check if task is in root cpuset

For systems that aren't using cpusets, but have them CONFIG_CPUSET enabled in
their kernel (eventually this may be most distribution kernels), this patch
removes even the minimal rcu_read_lock() from the memory page allocation path.

Actually, it removes that rcu call for any task that is in the root cpuset
(top_cpuset), which on systems not actively using cpusets, is all tasks.

We don't need the rcu check for tasks in the top_cpuset, because the
top_cpuset is statically allocated, so at no risk of being freed out from
underneath us.

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] cpuset: mark number_of_cpusets read_mostly
Paul Jackson [Sun, 8 Jan 2006 09:02:03 +0000 (01:02 -0800)]
[PATCH] cpuset: mark number_of_cpusets read_mostly

Mark cpuset global 'number_of_cpusets' as __read_mostly.

This global is accessed everytime a zone is considered in the zonelist loops
beneath __alloc_pages, looking for a free memory page.  If number_of_cpusets
is just one, then we can short circuit the mems_allowed check.

Since this global is read alot on a hot path, and written rarely, it is an
excellent candidate for __read_mostly.

Thanks to Christoph Lameter for the suggestion.

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] cpuset: use rcu directly optimization
Paul Jackson [Sun, 8 Jan 2006 09:02:02 +0000 (01:02 -0800)]
[PATCH] cpuset: use rcu directly optimization

Optimize the cpuset impact on page allocation, the most performance critical
cpuset hook in the kernel.

On each page allocation, the cpuset hook needs to check for a possible change
in the current tasks cpuset.  It can now handle the common case, of no change,
without taking any spinlock or semaphore, thanks to RCU.

Convert a spinlock on the current task to an rcu_read_lock(), saving
approximately a memory barrier and an atomic op, depending on architecture.

This is done by adding rcu_assign_pointer() and synchronize_rcu() calls to the
write side of the task->cpuset pointer, in cpuset.c:attach_task(), to delay
freeing up a detached cpuset until after any critical sections referencing
that pointer.

Thanks to Andi Kleen, Nick Piggin and Eric Dumazet for ideas.

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] cpuset: remove test for null cpuset from alloc code path
Paul Jackson [Sun, 8 Jan 2006 09:02:01 +0000 (01:02 -0800)]
[PATCH] cpuset: remove test for null cpuset from alloc code path

Remove a couple of more lines of code from the cpuset hooks in the page
allocation code path.

There was a check for a NULL cpuset pointer in the routine
cpuset_update_task_memory_state() that was only needed during system boot,
after the memory subsystem was initialized, before the cpuset subsystem was
initialized, to catch a NULL task->cpuset pointer.

Add a cpuset_init_early() routine, just before the mem_init() call in
init/main.c, that sets up just enough of the init tasks cpuset structure to
render cpuset_update_task_memory_state() calls harmless.

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] cpuset: migrate all tasks in cpuset at once
Paul Jackson [Sun, 8 Jan 2006 09:02:00 +0000 (01:02 -0800)]
[PATCH] cpuset: migrate all tasks in cpuset at once

Given the mechanism in the previous patch to handle rebinding the per-vma
mempolicies of all tasks in a cpuset that changes its memory placement, it is
now easier to handle the page migration requirements of such tasks at the same
time.

The previous code didn't actually attempt to migrate the pages of the tasks in
a cpuset whose memory placement changed until the next time each such task
tried to allocate memory.  This was undesirable, as users invoking memory page
migration exected to happen when the placement changed, not some unspecified
time later when the task needed more memory.

It is now trivial to handle the page migration at the same time as the per-vma
rebinding is done.

The routine cpuset.c:update_nodemask(), which handles changing a cpusets
memory placement ('mems') now checks for the special case of being asked to
write a placement that is the same as before.  It was harmless enough before
to just recompute everything again, even though nothing had changed.  But page
migration is a heavy weight operation - moving pages about.  So now it is
worth avoiding that if asked to move a cpuset to its current location.

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] cpuset: rebind vma mempolicies fix
Paul Jackson [Sun, 8 Jan 2006 09:01:59 +0000 (01:01 -0800)]
[PATCH] cpuset: rebind vma mempolicies fix

Fix more of longstanding bug in cpuset/mempolicy interaction.

NUMA mempolicies (mm/mempolicy.c) are constrained by the current tasks cpuset
to just the Memory Nodes allowed by that cpuset.  The kernel maintains
internal state for each mempolicy, tracking what nodes are used for the
MPOL_INTERLEAVE, MPOL_BIND or MPOL_PREFERRED policies.

When a tasks cpuset memory placement changes, whether because the cpuset
changed, or because the task was attached to a different cpuset, then the
tasks mempolicies have to be rebound to the new cpuset placement, so as to
preserve the cpuset-relative numbering of the nodes in that policy.

An earlier fix handled such mempolicy rebinding for mempolicies attached to a
task.

This fix rebinds mempolicies attached to vma's (address ranges in a tasks
address space.) Due to the need to hold the task->mm->mmap_sem semaphore while
updating vma's, the rebinding of vma mempolicies has to be done when the
cpuset memory placement is changed, at which time mmap_sem can be safely
acquired.  The tasks mempolicy is rebound later, when the task next attempts
to allocate memory and notices that its task->cpuset_mems_generation is
out-of-date with its cpusets mems_generation.

Because walking the tasklist to find all tasks attached to a changing cpuset
requires holding tasklist_lock, a spinlock, one cannot update the vma's of the
affected tasks while doing the tasklist scan.  In general, one cannot acquire
a semaphore (which can sleep) while already holding a spinlock (such as
tasklist_lock).  So a list of mm references has to be built up during the
tasklist scan, then the tasklist lock dropped, then for each mm, its mmap_sem
acquired, and the vma's in that mm rebound.

Once the tasklist lock is dropped, affected tasks may fork new tasks, before
their mm's are rebound.  A kernel global 'cpuset_being_rebound' is set to
point to the cpuset being rebound (there can only be one; cpuset modifications
are done under a global 'manage_sem' semaphore), and the mpol_copy code that
is used to copy a tasks mempolicies during fork catches such forking tasks,
and ensures their children are also rebound.

When a task is moved to a different cpuset, it is easier, as there is only one
task involved.  It's mm->vma's are scanned, using the same
mpol_rebind_policy() as used above.

It may happen that both the mpol_copy hook and the update done via the
tasklist scan update the same mm twice.  This is ok, as the mempolicies of
each vma in an mm keep track of what mems_allowed they are relative to, and
safely no-op a second request to rebind to the same nodes.

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] cpuset: number_of_cpusets optimization
Paul Jackson [Sun, 8 Jan 2006 09:01:57 +0000 (01:01 -0800)]
[PATCH] cpuset: number_of_cpusets optimization

Easy little optimization hack to avoid actually having to call
cpuset_zone_allowed() and check mems_allowed, in the main page allocation
routine, __alloc_pages().  This saves several CPU cycles per page allocation
on systems not using cpusets.

A counter is updated each time a cpuset is created or removed, and whenever
there is only one cpuset in the system, it must be the root cpuset, which
contains all CPUs and all Memory Nodes.  In that case, when the counter is
one, all allocations are allowed.

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] cpuset: numa_policy_rebind cleanup
Paul Jackson [Sun, 8 Jan 2006 09:01:56 +0000 (01:01 -0800)]
[PATCH] cpuset: numa_policy_rebind cleanup

Cleanup, reorganize and make more robust the mempolicy.c code to rebind
mempolicies relative to the containing cpuset after a tasks memory placement
changes.

The real motivator for this cleanup patch is to lay more groundwork for the
upcoming patch to correctly rebind NUMA mempolicies that are attached to vma's
after the containing cpuset memory placement changes.

NUMA mempolicies are constrained by the cpuset their task is a member of.
When either (1) a task is moved to a different cpuset, or (2) the 'mems'
mems_allowed of a cpuset is changed, then the NUMA mempolicies have embedded
node numbers (for MPOL_BIND, MPOL_INTERLEAVE and MPOL_PREFERRED) that need to
be recalculated, relative to their new cpuset placement.

The old code used an unreliable method of determining what was the old
mems_allowed constraining the mempolicy.  It just looked at the tasks
mems_allowed value.  This sort of worked with the present code, that just
rebinds the -task- mempolicy, and leaves any -vma- mempolicies broken,
referring to the old nodes.  But in an upcoming patch, the vma mempolicies
will be rebound as well.  Then the order in which the various task and vma
mempolicies are updated will no longer be deterministic, and one can no longer
count on the task->mems_allowed holding the old value for as long as needed.
It's not even clear if the current code was guaranteed to work reliably for
task mempolicies.

So I added a mems_allowed field to each mempolicy, stating exactly what
mems_allowed the policy is relative to, and updated synchronously and reliably
anytime that the mempolicy is rebound.

Also removed a useless wrapper routine, numa_policy_rebind(), and had its
caller, cpuset_update_task_memory_state(), call directly to the rewritten
policy_rebind() routine, and made that rebind routine extern instead of
static, and added a "mpol_" prefix to its name, making it
mpol_rebind_policy().

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] cpuset: implement cpuset_mems_allowed
Paul Jackson [Sun, 8 Jan 2006 09:01:55 +0000 (01:01 -0800)]
[PATCH] cpuset: implement cpuset_mems_allowed

Provide a cpuset_mems_allowed() method, which the sys_migrate_pages() code
needed, to obtain the mems_allowed vector of a cpuset, and replaced the
workaround in sys_migrate_pages() to call this new method.

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] cpuset: combine refresh_mems and update_mems
Paul Jackson [Sun, 8 Jan 2006 09:01:54 +0000 (01:01 -0800)]
[PATCH] cpuset: combine refresh_mems and update_mems

The important code paths through alloc_pages_current() and alloc_page_vma(),
by which most kernel page allocations go, both called
cpuset_update_current_mems_allowed(), which in turn called refresh_mems().
-Both- of these latter two routines did a tasklock, got the tasks cpuset
pointer, and checked for out of date cpuset->mems_generation.

That was a silly duplication of code and waste of CPU cycles on an important
code path.

Consolidated those two routines into a single routine, called
cpuset_update_task_memory_state(), since it updates more than just
mems_allowed.

Changed all callers of either routine to call the new consolidated routine.

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] cpuset: fork hook fix
Paul Jackson [Sun, 8 Jan 2006 09:01:53 +0000 (01:01 -0800)]
[PATCH] cpuset: fork hook fix

Fix obscure, never seen in real life, cpuset fork race.  The cpuset_fork()
call in fork.c was setting up the correct task->cpuset pointer after the
tasklist_lock was dropped, which briefly exposed the newly forked process with
an unsafe (copied from parent without locks or usage counter increment) cpuset
pointer.

In theory, that exposed cpuset pointer could have been pointing at a cpuset
that was already freed and removed, and in theory another task that had been
sitting on the tasklist_lock waiting to scan the task list could have raced
down the entire tasklist, found our new child at the far end, and dereferenced
that bogus cpuset pointer.

To fix, setup up the correct cpuset pointer in the new child by calling
cpuset_fork() before the new task is linked into the tasklist, and with that,
add a fork failure case, to dereference that cpuset, if the fork fails along
the way, after cpuset_fork() was called.

Had to remove a BUG_ON() from cpuset_exit(), because it was no longer valid -
the call to cpuset_exit() from a failed fork would not have PF_EXITING set.

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] cpuset: update_nodemask code reformat
Paul Jackson [Sun, 8 Jan 2006 09:01:52 +0000 (01:01 -0800)]
[PATCH] cpuset: update_nodemask code reformat

Restructure code layout of the kernel/cpuset.c update_nodemask() routine,
removing embedded returns and nested if's in favor of goto completion labels.
This is being done in anticipation of adding more logic to this routine, which
will favor the goto style structure.

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] cpuset: minor spacing initializer fixes
Paul Jackson [Sun, 8 Jan 2006 09:01:51 +0000 (01:01 -0800)]
[PATCH] cpuset: minor spacing initializer fixes

Four trivial cpuset fixes: remove extra spaces, remove useless initializers,
mark one __read_mostly.

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] cpuset: remove marker_pid documentation
Paul Jackson [Sun, 8 Jan 2006 09:01:51 +0000 (01:01 -0800)]
[PATCH] cpuset: remove marker_pid documentation

Remove documentation for the cpuset 'marker_pid' feature, that was in the
patch "cpuset: change marker for relative numbering" That patch was previously
pulled from *-mm at my (pj) request.

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] cpuset: document additional features
Paul Jackson [Sun, 8 Jan 2006 09:01:50 +0000 (01:01 -0800)]
[PATCH] cpuset: document additional features

Document the additional cpuset features:
notify_on_release
marker_pid
memory_pressure
memory_pressure_enabled

Rearrange and improve formatting of existing documentation for
cpu_exclusive and mem_exclusive features.

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] cpuset: memory pressure meter
Paul Jackson [Sun, 8 Jan 2006 09:01:49 +0000 (01:01 -0800)]
[PATCH] cpuset: memory pressure meter

Provide a simple per-cpuset metric of memory pressure, tracking the -rate-
that the tasks in a cpuset call try_to_free_pages(), the synchronous
(direct) memory reclaim code.

This enables batch managers monitoring jobs running in dedicated cpusets to
efficiently detect what level of memory pressure that job is causing.

This is useful both on tightly managed systems running a wide mix of
submitted jobs, which may choose to terminate or reprioritize jobs that are
trying to use more memory than allowed on the nodes assigned them, and with
tightly coupled, long running, massively parallel scientific computing jobs
that will dramatically fail to meet required performance goals if they
start to use more memory than allowed to them.

This patch just provides a very economical way for the batch manager to
monitor a cpuset for signs of memory pressure.  It's up to the batch
manager or other user code to decide what to do about it and take action.

==> Unless this feature is enabled by writing "1" to the special file
    /dev/cpuset/memory_pressure_enabled, the hook in the rebalance
    code of __alloc_pages() for this metric reduces to simply noticing
    that the cpuset_memory_pressure_enabled flag is zero.  So only
    systems that enable this feature will compute the metric.

Why a per-cpuset, running average:

    Because this meter is per-cpuset, rather than per-task or mm, the
    system load imposed by a batch scheduler monitoring this metric is
    sharply reduced on large systems, because a scan of the tasklist can be
    avoided on each set of queries.

    Because this meter is a running average, instead of an accumulating
    counter, a batch scheduler can detect memory pressure with a single
    read, instead of having to read and accumulate results for a period of
    time.

    Because this meter is per-cpuset rather than per-task or mm, the
    batch scheduler can obtain the key information, memory pressure in a
    cpuset, with a single read, rather than having to query and accumulate
    results over all the (dynamically changing) set of tasks in the cpuset.

A per-cpuset simple digital filter (requires a spinlock and 3 words of data
per-cpuset) is kept, and updated by any task attached to that cpuset, if it
enters the synchronous (direct) page reclaim code.

A per-cpuset file provides an integer number representing the recent
(half-life of 10 seconds) rate of direct page reclaims caused by the tasks
in the cpuset, in units of reclaims attempted per second, times 1000.

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] cpuset: mempolicy one more nodemask conversion
Paul Jackson [Sun, 8 Jan 2006 09:01:47 +0000 (01:01 -0800)]
[PATCH] cpuset: mempolicy one more nodemask conversion

Finish converting mm/mempolicy.c from bitmaps to nodemasks.  The previous
conversion had left one routine using bitmaps, since it involved a
corresponding change to kernel/cpuset.c

Fix that interface by replacing with a simple macro that calls nodes_subset(),
or if !CONFIG_CPUSET, returns (1).

Signed-off-by: Paul Jackson <pj@sgi.com>
Cc: Christoph Lameter <christoph@lameter.com>
Cc: Andi Kleen <ak@muc.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] cpuset: better bitmap remap defaults
Paul Jackson [Sun, 8 Jan 2006 09:01:46 +0000 (01:01 -0800)]
[PATCH] cpuset: better bitmap remap defaults

Fix the default behaviour for the remap operators in bitmap, cpumask and
nodemask.

As previously submitted, the pair of masks <A, B> defined a map of the
positions of the set bits in A to the corresponding bits in B.  This is still
true.

The issue is how to map the other positions, corresponding to the unset (0)
bits in A.  As previously submitted, they were all mapped to the first set bit
position in B, a constant map.

When I tried to code per-vma mempolicy rebinding using these remap operators,
I realized this was wrong.

This patch changes the default to map all the unset bit positions in A to the
same positions in B, the identity map.

For example, if A has bits 4-7 set, and B has bits 9-12 set, then the map
defined by the pair <A, B> maps each bit position in the first 32 bits as
follows:

0 ==> 0
  ...
3 ==> 3
4 ==> 9
  ...
7 ==> 12
8 ==> 8
9 ==> 9
  ...
31 ==> 31

This now corresponds to the typical behaviour desired when migrating pages and
policies from one cpuset to another.

The pages on nodes within the original cpuset, and the references in memory
policies to nodes within the original cpuset, are migrated to the
corresponding cpuset-relative nodes in the destination cpuset.  Other pages
and node references are left untouched.

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] slob: introduce the SLOB allocator
Matt Mackall [Sun, 8 Jan 2006 09:01:45 +0000 (01:01 -0800)]
[PATCH] slob: introduce the SLOB allocator

configurable replacement for slab allocator

This adds a CONFIG_SLAB option under CONFIG_EMBEDDED.  When CONFIG_SLAB is
disabled, the kernel falls back to using the 'SLOB' allocator.

SLOB is a traditional K&R/UNIX allocator with a SLAB emulation layer,
similar to the original Linux kmalloc allocator that SLAB replaced.  It's
signicantly smaller code and is more memory efficient.  But like all
similar allocators, it scales poorly and suffers from fragmentation more
than SLAB, so it's only appropriate for small systems.

It's been tested extensively in the Linux-tiny tree.  I've also
stress-tested it with make -j 8 compiles on a 3G SMP+PREEMPT box (not
recommended).

Here's a comparison for otherwise identical builds, showing SLOB saving
nearly half a megabyte of RAM:

$ size vmlinux*
   text    data     bss     dec     hex filename
3336372  529360  190812 4056544  3de5e0 vmlinux-slab
3323208  527948  190684 4041840  3dac70 vmlinux-slob

$ size mm/{slab,slob}.o
   text    data     bss     dec     hex filename
  13221     752      48   14021    36c5 mm/slab.o
   1896      52       8    1956     7a4 mm/slob.o

/proc/meminfo:
                  SLAB          SLOB      delta
MemTotal:        27964 kB      27980 kB     +16 kB
MemFree:         24596 kB      25092 kB    +496 kB
Buffers:            36 kB         36 kB       0 kB
Cached:           1188 kB       1188 kB       0 kB
SwapCached:          0 kB          0 kB       0 kB
Active:            608 kB        600 kB      -8 kB
Inactive:          808 kB        812 kB      +4 kB
HighTotal:           0 kB          0 kB       0 kB
HighFree:            0 kB          0 kB       0 kB
LowTotal:        27964 kB      27980 kB     +16 kB
LowFree:         24596 kB      25092 kB    +496 kB
SwapTotal:           0 kB          0 kB       0 kB
SwapFree:            0 kB          0 kB       0 kB
Dirty:               4 kB         12 kB      +8 kB
Writeback:           0 kB          0 kB       0 kB
Mapped:            560 kB        556 kB      -4 kB
Slab:             1756 kB          0 kB   -1756 kB
CommitLimit:     13980 kB      13988 kB      +8 kB
Committed_AS:     4208 kB       4208 kB       0 kB
PageTables:         28 kB         28 kB       0 kB
VmallocTotal:  1007312 kB    1007312 kB       0 kB
VmallocUsed:        48 kB         48 kB       0 kB
VmallocChunk:  1007264 kB    1007264 kB       0 kB

(this work has been sponsored in part by CELF)

From: Ingo Molnar <mingo@elte.hu>

   Fix 32-bitness bugs in mm/slob.c.

Signed-off-by: Matt Mackall <mpm@selenic.com>
Signed-off-by: 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] slob: introduce mm/util.c for shared functions
Matt Mackall [Sun, 8 Jan 2006 09:01:43 +0000 (01:01 -0800)]
[PATCH] slob: introduce mm/util.c for shared functions

Add mm/util.c for functions common between SLAB and SLOB.

Signed-off-by: Matt Mackall <mpm@selenic.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] DEBUG_SLAB depends on SLAB
Ingo Molnar [Sun, 8 Jan 2006 09:01:42 +0000 (01:01 -0800)]
[PATCH] DEBUG_SLAB depends on SLAB

Make DEBUG_SLAB depend on SLAB.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Cc: Matt Mackall <mpm@selenic.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] radix-tree: reduce tree height upon partial truncation
Nick Piggin [Sun, 8 Jan 2006 09:01:41 +0000 (01:01 -0800)]
[PATCH] radix-tree: reduce tree height upon partial truncation

Shrink the height of a radix tree when it is partially truncated - we only do
shrinkage of full truncation at present.

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] radix tree: early termination of tag clearing
Nick Piggin [Sun, 8 Jan 2006 09:01:41 +0000 (01:01 -0800)]
[PATCH] radix tree: early termination of tag clearing

Correctly determine the tags to be cleared in radix_tree_delete() so we
don't keep moving up the tree clearing tags that we don't need to.  For
example, if a tag is simply not set in the deleted item, nor anywhere up
the tree, radix_tree_delete() would attempt to clear it up the entire
height of the tree.

Also, tag_set() was made conditional so as not to dirty too many cachelines
high up in the radix tree.  Instead, put this logic into
radix_tree_tag_set().

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] radix tree: code consolidation
Nick Piggin [Sun, 8 Jan 2006 09:01:40 +0000 (01:01 -0800)]
[PATCH] radix tree: code consolidation

Introduce helper any_tag_set() rather than repeat the same code sequence 4
times.

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] remove get_task_struct_rcu()
Paul E. McKenney [Sun, 8 Jan 2006 09:01:39 +0000 (01:01 -0800)]
[PATCH] remove get_task_struct_rcu()

The latest set of signal-RCU patches does not use get_task_struct_rcu().
Attached is a patch that removes it.

Signed-off-by: "Paul E. McKenney" <paulmck@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] Simpler signal-exit concurrency handling
Paul E. McKenney [Sun, 8 Jan 2006 09:01:38 +0000 (01:01 -0800)]
[PATCH] Simpler signal-exit concurrency handling

Some simplification in checking signal delivery against concurrent exit.
Instead of using get_task_struct_rcu(), which increments the task_struct
reference count, check the reference count after acquiring sighand lock.

Signed-off-by: "Paul E. McKenney" <paulmck@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] RCU signal handling
Ingo Molnar [Sun, 8 Jan 2006 09:01:37 +0000 (01:01 -0800)]
[PATCH] RCU signal handling

RCU tasklist_lock and RCU signal handling: send signals RCU-read-locked
instead of tasklist_lock read-locked.  This is a scalability improvement on
SMP and a preemption-latency improvement under PREEMPT_RCU.

Signed-off-by: Paul E. McKenney <paulmck@us.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Acked-by: William Irwin <wli@holomorphy.com>
Cc: Roland McGrath <roland@redhat.com>
Cc: Oleg Nesterov <oleg@tv-sign.ru>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] Make RCU task_struct safe for oprofile
Paul E. McKenney [Sun, 8 Jan 2006 09:01:35 +0000 (01:01 -0800)]
[PATCH] Make RCU task_struct safe for oprofile

Applying RCU to the task structure broke oprofile, because
free_task_notify() can now be called from softirq.  This means that the
task_mortuary lock must be acquired with irq disabled in order to avoid
intermittent self-deadlock.  Since irq is now disabled, the critical
section within process_task_mortuary() has been restructured to be O(1) in
order to maximize scalability and minimize realtime latency degradation.

Kudos to Wu Fengguang for finding this problem!

CC: Wu Fengguang <wfg@mail.ustc.edu.cn>
Cc: Philippe Elie <phil.el@wanadoo.fr>
Cc: John Levon <levon@movementarian.org>
Signed-off-by: "Paul E. McKenney" <paulmck@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] uml: prevent MODE_SKAS=n and MODE_TT=n
Adrian Bunk [Sun, 8 Jan 2006 09:01:34 +0000 (01:01 -0800)]
[PATCH] uml: prevent MODE_SKAS=n and MODE_TT=n

If MODE_TT=n, MODE_SKAS must be y.

Signed-off-by: Adrian Bunk <bunk@stusta.de>
Acked-by: 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] uml: whitespace cleanup
Jeff Dike [Sun, 8 Jan 2006 09:01:33 +0000 (01:01 -0800)]
[PATCH] uml: whitespace cleanup

This fixes some mangled whitespace added by the earlier trap_user.c patch.

Signed-off-by: Jeff Dike <jdike@addtoit.com>
Cc: Paolo 'Blaisorblade' Giarrusso <blaisorblade@yahoo.it>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] consolidate asm/futex.h
Jeff Dike [Sun, 8 Jan 2006 09:01:32 +0000 (01:01 -0800)]
[PATCH] consolidate asm/futex.h

Most of the architectures have the same asm/futex.h.  This consolidates them
into asm-generic, with the arches including it from their own asm/futex.h.

In the case of UML, this reverts the old broken futex.h and goes back to using
the same one as almost everyone else.

Signed-off-by: Jeff Dike <jdike@addtoit.com>
Cc: Paolo 'Blaisorblade' Giarrusso <blaisorblade@yahoo.it>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] uml: merge trap_user.c and trap_kern.c
Gennady Sharapov [Sun, 8 Jan 2006 09:01:32 +0000 (01:01 -0800)]
[PATCH] uml: merge trap_user.c and trap_kern.c

The serial UML OS-abstraction layer patch (um/kernel dir).

This joins trap_user.c and trap_kernel.c files.

Signed-off-by: Gennady Sharapov <Gennady.V.Sharapov@intel.com>
Signed-off-by: Jeff Dike <jdike@addtoit.com>
Cc: Paolo 'Blaisorblade' Giarrusso <blaisorblade@yahoo.it>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] uml: move libc-dependent code from trap_user.c
Gennady Sharapov [Sun, 8 Jan 2006 09:01:31 +0000 (01:01 -0800)]
[PATCH] uml: move libc-dependent code from trap_user.c

The serial UML OS-abstraction layer patch (um/kernel dir).

This moves all systemcalls from trap_user.c file under os-Linux dir

Signed-off-by: Gennady Sharapov <Gennady.V.Sharapov@intel.com>
Signed-off-by: Jeff Dike <jdike@addtoit.com>
Cc: Paolo 'Blaisorblade' Giarrusso <blaisorblade@yahoo.it>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] uml: move libc-dependent code from signal_user.c
Gennady Sharapov [Sun, 8 Jan 2006 09:01:29 +0000 (01:01 -0800)]
[PATCH] uml: move libc-dependent code from signal_user.c

The serial UML OS-abstraction layer patch (um/kernel dir).

This moves all systemcalls from signal_user.c file under os-Linux dir

Signed-off-by: Gennady Sharapov <Gennady.V.Sharapov@intel.com>
Signed-off-by: Jeff Dike <jdike@addtoit.com>
Cc: Paolo 'Blaisorblade' Giarrusso <blaisorblade@yahoo.it>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] ARM: Netwinder ds1620 driver needs an export to be built as module
Woody Suwalski [Sun, 8 Jan 2006 09:01:29 +0000 (01:01 -0800)]
[PATCH] ARM: Netwinder ds1620 driver needs an export to be built as module

ds1620 module is using gpio_read symbol, so works only if "built-in" symbol
needs to be exported from the kernel image

Signed-off-by: Woody Suwalski <woodys@xandros.com>
Cc: Russell King <rmk@arm.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] Kill L1_CACHE_SHIFT_MAX
Ravikiran G Thirumalai [Sun, 8 Jan 2006 09:01:28 +0000 (01:01 -0800)]
[PATCH] Kill L1_CACHE_SHIFT_MAX

Kill L1_CACHE_SHIFT from all arches.  Since L1_CACHE_SHIFT_MAX is not used
anymore with the introduction of INTERNODE_CACHE, kill L1_CACHE_SHIFT_MAX.

Signed-off-by: Ravikiran Thirumalai <kiran@scalex86.org>
Signed-off-by: Shai Fultheim <shai@scalex86.org>
Signed-off-by: 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] Change maxaligned_in_smp alignemnt macros to internodealigned_in_smp macros
Ravikiran G Thirumalai [Sun, 8 Jan 2006 09:01:27 +0000 (01:01 -0800)]
[PATCH] Change maxaligned_in_smp alignemnt macros to internodealigned_in_smp macros

____cacheline_maxaligned_in_smp is currently used to align critical structures
and avoid false sharing.  It uses per-arch L1_CACHE_SHIFT_MAX and people find
L1_CACHE_SHIFT_MAX useless.

However, we have been using ____cacheline_maxaligned_in_smp to align
structures on the internode cacheline size.  As per Andi's suggestion,
following patch kills ____cacheline_maxaligned_in_smp and introduces
INTERNODE_CACHE_SHIFT, which defaults to L1_CACHE_SHIFT for all arches.
Arches needing L3/Internode cacheline alignment can define
INTERNODE_CACHE_SHIFT in the arch asm/cache.h.  Patch replaces
____cacheline_maxaligned_in_smp with ____cacheline_internodealigned_in_smp

With this patch, L1_CACHE_SHIFT_MAX can be killed

Signed-off-by: Ravikiran Thirumalai <kiran@scalex86.org>
Signed-off-by: Shai Fultheim <shai@scalex86.org>
Signed-off-by: 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] frv: fix uninitialised variable in serverworks driver
David Howells [Sun, 8 Jan 2006 09:01:26 +0000 (01:01 -0800)]
[PATCH] frv: fix uninitialised variable in serverworks driver

Fix an uninitialised variable warning in the serverworks driver.

Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] frv: fix uninitialised variable in atm nicstar driver
David Howells [Sun, 8 Jan 2006 09:01:25 +0000 (01:01 -0800)]
[PATCH] frv: fix uninitialised variable in atm nicstar driver

Fix an uninitialised variable warning in the atm nicstar driver.

Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] frv: miscellaneous changes
David Howells [Sun, 8 Jan 2006 09:01:25 +0000 (01:01 -0800)]
[PATCH] frv: miscellaneous changes

Fix a number of miscellanous items:

 (1) Declare lock sections in the linker script.

 (2) Recurse in the correct manner in the arch makefile.

 (3) asm/bug.h requires asm/linkage.h to be included first. One C file puts
     asm/bug.h first.

 (4) Add an empty RTC header file to avoid missing header file errors.

 (5) sg_dma_address() should use the dma_address member of a scatter list.

 (6) Add trivial pci_unmap support.

 (7) Add pgprot_noncached()

 (8) Discard u_quad_t.

 (9) Use ~0UL rather than ULONG_MAX in unistd.h in case the latter isn't
     declared.

(10) Add an empty VGA header file to avoid missing header file errors.

(11) Add an XOR header file to use the generic XOR stuff.

Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] frv: make get_user macro cast pointers
David Howells [Sun, 8 Jan 2006 09:01:24 +0000 (01:01 -0800)]
[PATCH] frv: make get_user macro cast pointers

Make the get_user macro cast the source pointer to an appropriate type for the
specified size.

Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] frv: force serial driver inclusion
David Howells [Sun, 8 Jan 2006 09:01:23 +0000 (01:01 -0800)]
[PATCH] frv: force serial driver inclusion

Force the 8230 serial driver to be built in if the on-CPU UARTs are to be
used.  It can't be used as a module because the arch setup needs to call into
it.

Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] frv: fix PCMCIA configuration
David Howells [Sun, 8 Jan 2006 09:01:22 +0000 (01:01 -0800)]
[PATCH] frv: fix PCMCIA configuration

Fix PCMCIA configuration for FRV by including the stock PCMCIA configuration
description file.

Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] frv: add pci_iomap
David Howells [Sun, 8 Jan 2006 09:01:22 +0000 (01:01 -0800)]
[PATCH] frv: add pci_iomap

Implement pci_iomap() for FRV.

Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] frv: add module support stubs
David Howells [Sun, 8 Jan 2006 09:01:21 +0000 (01:01 -0800)]
[PATCH] frv: add module support stubs

Add stubs for FRV module support.

Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] frv: supply various missing I/O access primitives
David Howells [Sun, 8 Jan 2006 09:01:20 +0000 (01:01 -0800)]
[PATCH] frv: supply various missing I/O access primitives

Supply various I/O access primitives that are missing for the FRV arch:

 (*) mmiowb()

 (*) read*_relaxed()

 (*) ioport_*map()

 (*) ioread*(), iowrite*(), ioread*_rep() and iowrite*_rep()

 (*) pci_io*map()

 (*) check_signature()

The patch also makes __is_PCI_addr() more efficient.

Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] frv: support module exception tables
David Howells [Sun, 8 Jan 2006 09:01:19 +0000 (01:01 -0800)]
[PATCH] frv: support module exception tables

Fix the exception table handling so that modules exceptions are dealt with.

Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] frv: implement and export various things required by modules
David Howells [Sun, 8 Jan 2006 09:01:19 +0000 (01:01 -0800)]
[PATCH] frv: implement and export various things required by modules

Export a number of features required to build all the modules.  It also
implements the following simple features:

 (*) csum_partial_copy_from_user() for MMU as well as no-MMU.

 (*) __ucmpdi2().

so that they can be exported too.

Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] frv: drop unsupported debugging features
David Howells [Sun, 8 Jan 2006 09:01:18 +0000 (01:01 -0800)]
[PATCH] frv: drop unsupported debugging features

Drop support for debugging features that aren't supported on FRV:

 (*) EARLY_PRINTK

The on-chip UARTs are set up early enough that this isn't required,
and VGA support isn't available. There's also a gdbstub available.

 (*) DEBUG_PAGEALLOC

This can't be easily be done since we use huge static mappings to
cover the kernel, not pages.

Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] frv: drop 8/16-bit xchg and cmpxchg
David Howells [Sun, 8 Jan 2006 09:01:17 +0000 (01:01 -0800)]
[PATCH] frv: drop 8/16-bit xchg and cmpxchg

Drop support for 8-bit and 16-bit xchg and cmpxchg emulation and implements
32-bit xchg with the SWAP/SWAPI instruction.

Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] frv: suppress configuration of certain features for FRV
David Howells [Sun, 8 Jan 2006 09:01:16 +0000 (01:01 -0800)]
[PATCH] frv: suppress configuration of certain features for FRV

Suppress configuration of certain features for the FRV arch as they can't be
built for FRV at the moment:

 (*) RTC

 (*) HISAX_*

 (*) PARPORT_PC

 (*) VGA_CONSOLE

 (*) BINFMT_ELF

Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] cs89x0: fix up after pnx0105 Kconfig symbol renaming
Lennert Buytenhek [Sun, 8 Jan 2006 09:01:14 +0000 (01:01 -0800)]
[PATCH] cs89x0: fix up after pnx0105 Kconfig symbol renaming

The Kconfig symbol for pnx0105 was recently renamed to ARCH_PNX010X.

Signed-off-by: Lennert Buytenhek <buytenh@wantstofly.org>
Cc: dmitry pervushin <dpervushin@ru.mvista.com>
Cc: <dsaxena@plexity.net>
Cc: Jeff Garzik <jgarzik@pobox.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] fix Kconfig depends for cs89x0 (PNX010X support)
Lennert Buytenhek [Sun, 8 Jan 2006 09:01:13 +0000 (01:01 -0800)]
[PATCH] fix Kconfig depends for cs89x0 (PNX010X support)

PNX010X support for CS89x0 should be conditional on NET_PCI, as it is an 'on
board controller' and NET_PCI includes that category of NICs.  Since
ARCH_PNX0105 was recently changed to ARCH_PNX010X, incorporate that change as
well while we're at it.

Signed-off-by: Lennert Buytenhek <buytenh@wantstofly.org>
Cc: dmitry pervushin <dpervushin@ru.mvista.com>
Cc: <dsaxena@plexity.net>
Cc: Jeff Garzik <jgarzik@pobox.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] cs89x0: switch {in,out}sw to {read,write}words
Lennert Buytenhek [Sun, 8 Jan 2006 09:01:12 +0000 (01:01 -0800)]
[PATCH] cs89x0: switch {in,out}sw to {read,write}words

Implement readwords/writewords that use readword/writeword, and switch the
rest of the driver over to use these.

Signed-off-by: Lennert Buytenhek <buytenh@wantstofly.org>
Cc: dmitry pervushin <dpervushin@ru.mvista.com>
Cc: <dsaxena@plexity.net>
Cc: Jeff Garzik <jgarzik@pobox.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] cs89x0: cleanly implement ixdp2x01 and pnx0501 support
Lennert Buytenhek [Sun, 8 Jan 2006 09:01:11 +0000 (01:01 -0800)]
[PATCH] cs89x0: cleanly implement ixdp2x01 and pnx0501 support

Implement suitable versions of the readword/writeword macros for ixdp2x01 and
pnx0501.  Handle the 32-bit spacing of the registers in these functions
instead of in the header file.

Signed-off-by: Lennert Buytenhek <buytenh@wantstofly.org>
Cc: dmitry pervushin <dpervushin@ru.mvista.com>
Cc: <dsaxena@plexity.net>
Cc: Jeff Garzik <jgarzik@pobox.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] cs89x0: make {read,write}reg use {read,write}word
Lennert Buytenhek [Sun, 8 Jan 2006 09:01:10 +0000 (01:01 -0800)]
[PATCH] cs89x0: make {read,write}reg use {read,write}word

Make readreg/writereg use readword/writeword.

Signed-off-by: Lennert Buytenhek <buytenh@wantstofly.org>
Cc: dmitry pervushin <dpervushin@ru.mvista.com>
Cc: <dsaxena@plexity.net>
Cc: Jeff Garzik <jgarzik@pobox.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] cs89x0: swap {read,write}reg and {read,write}word
Lennert Buytenhek [Sun, 8 Jan 2006 09:01:09 +0000 (01:01 -0800)]
[PATCH] cs89x0: swap {read,write}reg and {read,write}word

Reverse the order of readreg/writereg and readword/writeword in the
file, so that we can make readreg/writereg use readword/writeword.

Signed-off-by: Lennert Buytenhek <buytenh@wantstofly.org>
Cc: dmitry pervushin <dpervushin@ru.mvista.com>
Cc: <dsaxena@plexity.net>
Cc: Jeff Garzik <jgarzik@pobox.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] cs89x0: convert {inw,outw} calls to {read,write}word
Lennert Buytenhek [Sun, 8 Jan 2006 09:01:08 +0000 (01:01 -0800)]
[PATCH] cs89x0: convert {inw,outw} calls to {read,write}word

Switch all occurences of inw/outw in the driver over to readword/writeword.

Signed-off-by: Lennert Buytenhek <buytenh@wantstofly.org>
Cc: dmitry pervushin <dpervushin@ru.mvista.com>
Cc: <dsaxena@plexity.net>
Cc: Jeff Garzik <jgarzik@pobox.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] cs89x0: make {read,write}word take base_addr
Lennert Buytenhek [Sun, 8 Jan 2006 09:01:06 +0000 (01:01 -0800)]
[PATCH] cs89x0: make {read,write}word take base_addr

readword() and writeword() take a 'struct net_device *' and deref its
->base_addr member.  Make them take the base_addr directly instead, so
that we can switch the other occurences of inw/outw in the file over
to readword/writeword as well.

Signed-off-by: Lennert Buytenhek <buytenh@wantstofly.org>
Cc: dmitry pervushin <dpervushin@ru.mvista.com>
Cc: <dsaxena@plexity.net>
Cc: Jeff Garzik <jgarzik@pobox.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] Optimise oom kill of current task
Kirill Korotaev [Sun, 8 Jan 2006 09:01:05 +0000 (01:01 -0800)]
[PATCH] Optimise oom kill of current task

When oom_killer kills current there's no need to call
schedule_timeout_interruptible() since task must die ASAP.

Signed-Off-By: Pavel Emelianov <xemul@sw.ru>
Signed-Off-By: Kirill Korotaev <dev@openvz.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] Move page migration related functions near do_migrate_pages()
Christoph Lameter [Sun, 8 Jan 2006 09:01:04 +0000 (01:01 -0800)]
[PATCH] Move page migration related functions near do_migrate_pages()

Group page migration functions in mempolicy.c

Add a forward declaration for migrate_page_add (like gather_stats()) and use
our new found mobility to group all page migration related function around
do_migrate_pages().

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Cc: Andi Kleen <ak@muc.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] mempolicies: unexport get_vma_policy()
Christoph Lameter [Sun, 8 Jan 2006 09:01:03 +0000 (01:01 -0800)]
[PATCH] mempolicies: unexport get_vma_policy()

Since the numa_maps functionality is now in mempolicy.c we no longer need to
export get_vma_policy().

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Cc: Andi Kleen <ak@muc.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] Drop page table lock before calling migrate_page_add()
Christoph Lameter [Sun, 8 Jan 2006 09:01:02 +0000 (01:01 -0800)]
[PATCH] Drop page table lock before calling migrate_page_add()

migrate_page_add cannot be called with a spinlock held (calls
isolate_lru_page which calles schedule_on_each_cpu).  Drop ptl lock in
check_pte_range before calling migrate_page_add().

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] Fold numa_maps into mempolicies.c
Christoph Lameter [Sun, 8 Jan 2006 09:01:02 +0000 (01:01 -0800)]
[PATCH] Fold numa_maps into mempolicies.c

First discussed at http://marc.theaimsgroup.com/?t=113149255100001&r=1&w=2

- Use the check_range() in mempolicy.c to gather statistics.

- Improve the numa_maps code in general and fix some comments.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Cc: Andi Kleen <ak@muc.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] mempolicies: private pointer in check_range and MPOL_MF_INVERT
Christoph Lameter [Sun, 8 Jan 2006 09:01:01 +0000 (01:01 -0800)]
[PATCH] mempolicies: private pointer in check_range and MPOL_MF_INVERT

This was was first posted at
http://marc.theaimsgroup.com/?l=linux-mm&m=113149240227584&w=2

(Part of this functionality is also contained in the direct migration
pathset. The functionality here is more generic and independent of that
patchset.)

- Add internal flags MPOL_MF_INVERT to control check_range() behavior.

- Replace the pagelist passed through by check_range by a general
  private pointer that may be used for other purposes.
  (The following patches will use that to merge numa_maps into
  mempolicy.c and to better group the page migration code in
  the policy layer)

- Improve some comments.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Cc: Andi Kleen <ak@muc.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] rmap: additional diagnostics in page_remove_rmap()
Dave Jones [Sun, 8 Jan 2006 09:01:00 +0000 (01:01 -0800)]
[PATCH] rmap: additional diagnostics in page_remove_rmap()

We seem to be hitting this assertion failure too often for it to be
hardware bugs.

Cc: Hugh Dickins <hugh@veritas.com>
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] mm: clean up local variables
Tobias Klauser [Sun, 8 Jan 2006 09:00:59 +0000 (01:00 -0800)]
[PATCH] mm: clean up local variables

Clean up a local variable with the same name as a variable in a larger
block.  Also move a variable into the block where it's actually used.

Spotted by http://linuxicc.sourceforge.net/

Signed-off-by: Tobias Klauser <tklauser@nuerscht.ch>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] set_page_count() macro safety
Avishay Traeger [Sun, 8 Jan 2006 09:00:58 +0000 (01:00 -0800)]
[PATCH] set_page_count() macro safety

Fix set_page_count() macro to handle complex arguments.

Signed-off-by: Avishay Traeger <atraeger@cs.sunysb.edu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] mm: make hugepages obey cpusets.
Christoph Lameter [Sun, 8 Jan 2006 09:00:57 +0000 (01:00 -0800)]
[PATCH] mm: make hugepages obey cpusets.

See http://marc.theaimsgroup.com/?l=linux-kernel&m=113167000201265&w=2
http://marc.theaimsgroup.com/?l=linux-mm&m=113167267527312&w=2

Make hugepages obey cpusets.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Acked-by: William Irwin <wli@holomorphy.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] cpusets: swap migration interface
Paul Jackson [Sun, 8 Jan 2006 09:00:56 +0000 (01:00 -0800)]
[PATCH] cpusets: swap migration interface

Add a boolean "memory_migrate" to each cpuset, represented by a file
containing "0" or "1" in each directory below /dev/cpuset.

It defaults to false (file contains "0").  It can be set true by writing
"1" to the file.

If true, then anytime that a task is attached to the cpuset so marked, the
pages of that task will be moved to that cpuset, preserving, to the extent
practical, the cpuset-relative placement of the pages.

Also anytime that a cpuset so marked has its memory placement changed (by
writing to its "mems" file), the tasks in that cpuset will have their pages
moved to the cpusets new nodes, preserving, to the extent practical, the
cpuset-relative placement of the moved pages.

Signed-off-by: Paul Jackson <pj@sgi.com>
Cc: Christoph Lameter <christoph@lameter.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] SwapMig: Switch error handling in migrate_pages to use -Exx
Christoph Lameter [Sun, 8 Jan 2006 09:00:55 +0000 (01:00 -0800)]
[PATCH] SwapMig: Switch error handling in migrate_pages to use -Exx

Use -Exxx instead of numeric return codes and cleanup the code in
migrate_pages() using -Exx error codes.

Consolidate successful migration handling

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] SwapMig: Extend parameters for migrate_pages()
Christoph Lameter [Sun, 8 Jan 2006 09:00:55 +0000 (01:00 -0800)]
[PATCH] SwapMig: Extend parameters for migrate_pages()

Extend the parameters of migrate_pages() to allow the caller control over the
fate of successfully migrated or impossible to migrate pages.

Swap migration and direct migration will have the same interface after this
patch so that patches can be independently applied to the policy layer and the
core migration code.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Cc: Andi Kleen <ak@muc.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] SwapMig: Drop unused pages immediately
Christoph Lameter [Sun, 8 Jan 2006 09:00:54 +0000 (01:00 -0800)]
[PATCH] SwapMig: Drop unused pages immediately

Drop unused pages immediately

If a page is encountered that is only referenced by the migration code then
there is no reason to swap or migrate the page.  Release the page by calling
move_to_lru().

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] SwapMig: add_to_swap() avoid atomic allocations
Christoph Lameter [Sun, 8 Jan 2006 09:00:53 +0000 (01:00 -0800)]
[PATCH] SwapMig: add_to_swap() avoid atomic allocations

Add gfp_mask to add_to_swap

add_to_swap does allocations with GFP_ATOMIC in order not to interfere with
swapping.  During migration we may have use add_to_swap extensively which may
lead to out of memory errors.

This patch makes add_to_swap take a parameter that specifies the gfp mask.
The page migration code can then make add_to_swap use GFP_KERNEL.

Signed-off-by: Hirokazu Takahashi <taka@valinux.co.jp>
Signed-off-by: Dave Hansen <haveblue@us.ibm.com>
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] SwapMig: CONFIG_MIGRATION fixes
Christoph Lameter [Sun, 8 Jan 2006 09:00:52 +0000 (01:00 -0800)]
[PATCH] SwapMig: CONFIG_MIGRATION fixes

Move move_to_lru, putback_lru_pages and isolate_lru in section surrounded by
CONFIG_MIGRATION saving some codesize for single processor kernels.

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] Swap Migration V5: sys_migrate_pages interface
Christoph Lameter [Sun, 8 Jan 2006 09:00:51 +0000 (01:00 -0800)]
[PATCH] Swap Migration V5: sys_migrate_pages interface

sys_migrate_pages implementation using swap based page migration

This is the original API proposed by Ray Bryant in his posts during the first
half of 2005 on linux-mm@kvack.org and linux-kernel@vger.kernel.org.

The intent of sys_migrate is to migrate memory of a process.  A process may
have migrated to another node.  Memory was allocated optimally for the prior
context.  sys_migrate_pages allows to shift the memory to the new node.

sys_migrate_pages is also useful if the processes available memory nodes have
changed through cpuset operations to manually move the processes memory.  Paul
Jackson is working on an automated mechanism that will allow an automatic
migration if the cpuset of a process is changed.  However, a user may decide
to manually control the migration.

This implementation is put into the policy layer since it uses concepts and
functions that are also needed for mbind and friends.  The patch also provides
a do_migrate_pages function that may be useful for cpusets to automatically
move memory.  sys_migrate_pages does not modify policies in contrast to Ray's
implementation.

The current code here is based on the swap based page migration capability and
thus is not able to preserve the physical layout relative to it containing
nodeset (which may be a cpuset).  When direct page migration becomes available
then the implementation needs to be changed to do a isomorphic move of pages
between different nodesets.  The current implementation simply evicts all
pages in source nodeset that are not in the target nodeset.

Patch supports ia64, i386 and x86_64.

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] Swap Migration V5: MPOL_MF_MOVE interface
Christoph Lameter [Sun, 8 Jan 2006 09:00:50 +0000 (01:00 -0800)]
[PATCH] Swap Migration V5: MPOL_MF_MOVE interface

Add page migration support via swap to the NUMA policy layer

This patch adds page migration support to the NUMA policy layer.  An
additional flag MPOL_MF_MOVE is introduced for mbind.  If MPOL_MF_MOVE is
specified then pages that do not conform to the memory policy will be evicted
from memory.  When they get pages back in new pages will be allocated
following the numa policy.

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] Swap Migration V5: Add CONFIG_MIGRATION for page migration support
Christoph Lameter [Sun, 8 Jan 2006 09:00:49 +0000 (01:00 -0800)]
[PATCH] Swap Migration V5: Add CONFIG_MIGRATION for page migration support

Include page migration if the system is NUMA or having a memory model that
allows distinct areas of memory (SPARSEMEM, DISCONTIGMEM).

And:
- Only include lru_add_drain_per_cpu if building for an SMP system.

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] Swap Migration V5: migrate_pages() function
Christoph Lameter [Sun, 8 Jan 2006 09:00:48 +0000 (01:00 -0800)]
[PATCH] Swap Migration V5: migrate_pages() function

This adds the basic page migration function with a minimal implementation that
only allows the eviction of pages to swap space.

Page eviction and migration may be useful to migrate pages, to suspend
programs or for remapping single pages (useful for faulty pages or pages with
soft ECC failures)

The process is as follows:

The function wanting to migrate pages must first build a list of pages to be
migrated or evicted and take them off the lru lists via isolate_lru_page().
isolate_lru_page determines that a page is freeable based on the LRU bit set.

Then the actual migration or swapout can happen by calling migrate_pages().

migrate_pages does its best to migrate or swapout the pages and does multiple
passes over the list.  Some pages may only be swappable if they are not dirty.
 migrate_pages may start writing out dirty pages in the initial passes over
the pages.  However, migrate_pages may not be able to migrate or evict all
pages for a variety of reasons.

The remaining pages may be returned to the LRU lists using putback_lru_pages().

Changelog V4->V5:
- Use the lru caches to return pages to the LRU

Changelog V3->V4:
- Restructure code so that applying patches to support full migration does
  require minimal changes. Rename swapout_pages() to migrate_pages().

Changelog V2->V3:
- Extract common code from shrink_list() and swapout_pages()

Signed-off-by: Mike Kravetz <kravetz@us.ibm.com>
Signed-off-by: Christoph Lameter <clameter@sgi.com>
Cc: "Michael Kerrisk" <mtk-manpages@gmx.net>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] Swap Migration V5: PF_SWAPWRITE to allow writing to swap
Christoph Lameter [Sun, 8 Jan 2006 09:00:47 +0000 (01:00 -0800)]
[PATCH] Swap Migration V5: PF_SWAPWRITE to allow writing to swap

Add PF_SWAPWRITE to control a processes permission to write to swap.

- Use PF_SWAPWRITE in may_write_to_queue() instead of checking for kswapd
  and pdflush

- Set PF_SWAPWRITE flag for kswapd and pdflush

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] Swap Migration V5: LRU operations
Christoph Lameter [Sun, 8 Jan 2006 09:00:45 +0000 (01:00 -0800)]
[PATCH] Swap Migration V5: LRU operations

This is the start of the `swap migration' patch series.

Swap migration allows the moving of the physical location of pages between
nodes in a numa system while the process is running.  This means that the
virtual addresses that the process sees do not change.  However, the system
rearranges the physical location of those pages.

The main intent of page migration patches here is to reduce the latency of
memory access by moving pages near to the processor where the process
accessing that memory is running.

The patchset allows a process to manually relocate the node on which its
pages are located through the MF_MOVE and MF_MOVE_ALL options while
setting a new memory policy.

The pages of process can also be relocated from another process using the
sys_migrate_pages() function call.  Requires CAP_SYS_ADMIN.  The migrate_pages
function call takes two sets of nodes and moves pages of a process that are
located on the from nodes to the destination nodes.

Manual migration is very useful if for example the scheduler has relocated a
process to a processor on a distant node.  A batch scheduler or an
administrator can detect the situation and move the pages of the process
nearer to the new processor.

sys_migrate_pages() could be used on non-numa machines as well, to force all
of a particualr process's pages out to swap, if someone thinks that's useful.

Larger installations usually partition the system using cpusets into sections
of nodes.  Paul has equipped cpusets with the ability to move pages when a
task is moved to another cpuset.  This allows automatic control over locality
of a process.  If a task is moved to a new cpuset then also all its pages are
moved with it so that the performance of the process does not sink
dramatically (as is the case today).

Swap migration works by simply evicting the page.  The pages must be faulted
back in.  The pages are then typically reallocated by the system near the node
where the process is executing.

For swap migration the destination of the move is controlled by the allocation
policy.  Cpusets set the allocation policy before calling sys_migrate_pages()
in order to move the pages as intended.

No allocation policy changes are performed for sys_migrate_pages().  This
means that the pages may not faulted in to the specified nodes if no
allocation policy was set by other means.  The pages will just end up near the
node where the fault occurred.

There's another patch series in the pipeline which implements "direct
migration".

The direct migration patchset extends the migration functionality to avoid
going through swap.  The destination node of the relation is controllable
during the actual moving of pages.  The crutch of using the allocation policy
to relocate is not necessary and the pages are moved directly to the target.
Its also faster since swap is not used.

And sys_migrate_pages() can then move pages directly to the specified node.
Implement functions to isolate pages from the LRU and put them back later.

This patch:

An earlier implementation was provided by Hirokazu Takahashi
<taka@valinux.co.jp> and IWAMOTO Toshihiro <iwamoto@valinux.co.jp> for the
memory hotplug project.

From: Magnus

This breaks out isolate_lru_page() and putpack_lru_page().  Needed for swap
migration.

Signed-off-by: Magnus Damm <magnus.damm@gmail.com>
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] add schedule_on_each_cpu()
Christoph Lameter [Sun, 8 Jan 2006 09:00:43 +0000 (01:00 -0800)]
[PATCH] add schedule_on_each_cpu()

swap migration's isolate_lru_page() currently uses an IPI to notify other
processors that the lru caches need to be drained if the page cannot be
found on the LRU.  The IPI interrupt may interrupt a processor that is just
processing lru requests and cause a race condition.

This patch introduces a new function run_on_each_cpu() that uses the
keventd() to run the LRU draining on each processor.  Processors disable
preemption when dealing the LRU caches (these are per processor) and thus
executing LRU draining from another process is safe.

Thanks to Lee Schermerhorn <lee.schermerhorn@hp.com> for finding this race
condition.

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] mm: free_pages opt
Nick Piggin [Sun, 8 Jan 2006 09:00:42 +0000 (01:00 -0800)]
[PATCH] mm: free_pages opt

Try to streamline free_pages_bulk by ensuring callers don't pass in a
'count' that exceeds the list size.

Some cleanups:
Rename __free_pages_bulk to __free_one_page.
Put the page list manipulation from __free_pages_ok into free_one_page.
Make __free_pages_ok static.

Signed-off-by: Nick Piggin <npiggin@suse.de>
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] mm: cleanup zone_pcp
Nick Piggin [Sun, 8 Jan 2006 09:00:41 +0000 (01:00 -0800)]
[PATCH] mm: cleanup zone_pcp

Use zone_pcp everywhere even though NUMA code "knows" the internal details
of the zone.  Stop other people trying to copy, and it looks nicer.

Also, only print the pagesets of online cpus in zoneinfo.

Signed-off-by: Nick Piggin <npiggin@suse.de>
Cc: "Seth, Rohit" <rohit.seth@intel.com>
Cc: Christoph Lameter <christoph@lameter.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] Make high and batch sizes of per_cpu_pagelists configurable
Rohit Seth [Sun, 8 Jan 2006 09:00:40 +0000 (01:00 -0800)]
[PATCH] Make high and batch sizes of per_cpu_pagelists configurable

As recently there has been lot of traffic on the right values for batch and
high water marks for per_cpu_pagelists.  This patch makes these two
variables configurable through /proc interface.

A new tunable /proc/sys/vm/percpu_pagelist_fraction is added.  This entry
controls the fraction of pages at most in each zone that are allocated for
each per cpu page list.  The min value for this is 8.  It means that we
don't allow more than 1/8th of pages in each zone to be allocated in any
single per_cpu_pagelist.

The batch value of each per cpu pagelist is also updated as a result.  It
is set to pcp->high/4.  The upper limit of batch is (PAGE_SHIFT * 8)

Signed-off-by: Rohit Seth <rohit.seth@intel.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] drop-pagecache
Andrew Morton [Sun, 8 Jan 2006 09:00:39 +0000 (01:00 -0800)]
[PATCH] drop-pagecache

Add /proc/sys/vm/drop_caches.  When written to, this will cause the kernel to
discard as much pagecache and/or reclaimable slab objects as it can.  THis
operation requires root permissions.

It won't drop dirty data, so the user should run `sync' first.

Caveats:

a) Holds inode_lock for exorbitant amounts of time.

b) Needs to be taught about NUMA nodes: propagate these all the way through
   so the discarding can be controlled on a per-node basis.

This is a debugging feature: useful for getting consistent results between
filesystem benchmarks.  We could possibly put it under a config option, but
it's less than 300 bytes.

Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>