This patch automatically updates a tasks NUMA mempolicy when its cpuset
memory placement changes. It does so within the context of the task,
without any need to support low level external mempolicy manipulation.
If a system is not using cpusets, or if running on a system with just the
root (all-encompassing) cpuset, then this remap is a no-op. Only when a
task is moved between cpusets, or a cpusets memory placement is changed
does the following apply. Otherwise, the main routine below,
rebind_policy() is not even called.
When mixing cpusets, scheduler affinity, and NUMA mempolicies, the
essential role of cpusets is to place jobs (several related tasks) on a set
of CPUs and Memory Nodes, the essential role of sched_setaffinity is to
manage a jobs processor placement within its allowed cpuset, and the
essential role of NUMA mempolicy (mbind, set_mempolicy) is to manage a jobs
memory placement within its allowed cpuset.
However, CPU affinity and NUMA memory placement are managed within the
kernel using absolute system wide numbering, not cpuset relative numbering.
This is ok until a job is migrated to a different cpuset, or what's the
same, a jobs cpuset is moved to different CPUs and Memory Nodes.
Then the CPU affinity and NUMA memory placement of the tasks in the job
need to be updated, to preserve their cpuset-relative position. This can
be done for CPU affinity using sched_setaffinity() from user code, as one
task can modify anothers CPU affinity. This cannot be done from an
external task for NUMA memory placement, as that can only be modified in
the context of the task using it.
However, it easy enough to remap a tasks NUMA mempolicy automatically when
a task is migrated, using the existing cpuset mechanism to trigger a
refresh of a tasks memory placement after its cpuset has changed. All that
is needed is the old and new nodemask, and notice to the task that it needs
to rebind its mempolicy. The tasks mems_allowed has the old mask, the
tasks cpuset has the new mask, and the existing
cpuset_update_current_mems_allowed() mechanism provides the notice. The
bitmap/cpumask/nodemask remap operators provide the cpuset relative
calculations.
This patch leaves open a couple of issues:
1) Updating vma and shmfs/tmpfs/hugetlbfs memory policies:
These mempolicies may reference nodes outside of those allowed to
the current task by its cpuset. Tasks are migrated as part of jobs,
which reside on what might be several cpusets in a subtree. When such
a job is migrated, all NUMA memory policy references to nodes within
that cpuset subtree should be translated, and references to any nodes
outside that subtree should be left untouched. A future patch will
provide the cpuset mechanism needed to mark such subtrees. With that
patch, we will be able to correctly migrate these other memory policies
across a job migration.
2) Updating cpuset, affinity and memory policies in user space:
This is harder. Any placement state stored in user space using
system-wide numbering will be invalidated across a migration. More
work will be required to provide user code with a migration-safe means
to manage its cpuset relative placement, while preserving the current
API's that pass system wide numbers, not cpuset relative numbers across
the kernel-user boundary.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Add support for renaming cpusets. Only allow simple rename of cpuset
directories in place. Don't allow moving cpusets elsewhere in hierarchy or
renaming the special cpuset files in each cpuset directory.
The usefulness of this simple rename became apparent when developing task
migration facilities. It allows building a second cpuset hierarchy using
new names and containing new CPUs and Memory Nodes, moving tasks from the
old to the new cpusets, removing the old cpusets, and then renaming the new
cpusets to be just like the old names, so that any knowledge that the tasks
had of their cpuset names will still be valid.
Leaf node cpusets can be migrated to other CPUs or Memory Nodes by just
updating their 'cpus' and 'mems' files, but because no cpuset can contain
CPUs or Nodes not in its parent cpuset, one cannot do this in a cpuset
hierarchy without first expanding all the non-leaf cpusets to contain the
union of both the old and new CPUs and Nodes, which would obfuscate the
one-to-one migration of a task from one cpuset to another required to
correctly migrate the physical page frames currently allocated to that
task.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Overhaul cpuset locking. Replace single semaphore with two semaphores.
The suggestion to use two locks was made by Roman Zippel.
Both locks are global. Code that wants to modify cpusets must first
acquire the exclusive manage_sem, which allows them read-only access to
cpusets, and holds off other would-be modifiers. Before making actual
changes, the second semaphore, callback_sem must be acquired as well. Code
that needs only to query cpusets must acquire callback_sem, which is also a
global exclusive lock.
The earlier problems with double tripping are avoided, because it is
allowed for holders of manage_sem to nest the second callback_sem lock, and
only callback_sem is needed by code called from within __alloc_pages(),
where the double tripping had been possible.
This is not quite the same as a normal read/write semaphore, because
obtaining read-only access with intent to change must hold off other such
attempts, while allowing read-only access w/o such intention. Changing
cpusets involves several related checks and changes, which must be done
while allowing read-only queries (to avoid the double trip), but while
ensuring nothing changes (holding off other would be modifiers.)
This overhaul of cpuset locking also makes careful use of task_lock() to
guard access to the task->cpuset pointer, closing a couple of race
conditions noticed while reading this code (thanks, Roman). I've never
seen these races fail in any use or test.
See further the comments in the code.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Remove a rather hackish depth counter on cpuset locking. The depth counter
was avoiding a possible double trip on the global cpuset_sem semaphore. It
worked, but now an improved version of cpuset locking is available, to come
in the next patch, using two global semaphores.
This patch reverses "cpuset semaphore depth check deadlock fix"
The kernel still works, even after this patch, except for some rare and
difficult to reproduce race conditions when agressively creating and
destroying cpusets marked with the notify_on_release option, on very large
systems.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Remove one more useless line from cpuset_common_file_read().
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
- added typedef unsigned int __nocast gfp_t;
- replaced __nocast uses for gfp flags with gfp_t - it gives exactly
the same warnings as far as sparse is concerned, doesn't change
generated code (from gcc point of view we replaced unsigned int with
typedef) and documents what's going on far better.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Switched cpuset_common_file_read() to simple_read_from_buffer(), killed
a bunch of useless (and not quite correct - e.g. min(size_t,ssize_t))
code.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Don't leak a page of memory if user reads a cpuset file past eof.
Signed-off-by: KUROSAWA Takahiro <kurosawa@valinux.co.jp>
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Optimize the deadlock avoidance check on the global cpuset
semaphore cpuset_sem. Instead of adding a depth counter to the
task struct of each task, rather just two words are enough, one
to store the depth and the other the current cpuset_sem holder.
Thanks to Nikita Danilov for the idea.
Signed-off-by: Paul Jackson <pj@sgi.com>
[ We may want to change this further, but at least it's now
a totally internal decision to the cpusets code ]
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
The cpusets-formalize-intermediate-gfp_kernel-containment patch
has a deadlock problem.
This patch was part of a set of four patches to make more
extensive use of the cpuset 'mem_exclusive' attribute to
manage kernel GFP_KERNEL memory allocations and to constrain
the out-of-memory (oom) killer.
A task that is changing cpusets in particular ways on a system
when it is very short of free memory could double trip over
the global cpuset_sem semaphore (get the lock and then deadlock
trying to get it again).
The second attempt to get cpuset_sem would be in the routine
cpuset_zone_allowed(). This was discovered by code inspection.
I can not reproduce the problem except with an artifically
hacked kernel and a specialized stress test.
In real life you cannot hit this unless you are manipulating
cpusets, and are very unlikely to hit it unless you are rapidly
modifying cpusets on a memory tight system. Even then it would
be a rare occurence.
If you did hit it, the task double tripping over cpuset_sem
would deadlock in the kernel, and any other task also trying
to manipulate cpusets would deadlock there too, on cpuset_sem.
Your batch manager would be wedged solid (if it was cpuset
savvy), but classic Unix shells and utilities would work well
enough to reboot the system.
The unusual condition that led to this bug is that unlike most
semaphores, cpuset_sem _can_ be acquired while in the page
allocation code, when __alloc_pages() calls cpuset_zone_allowed.
So it easy to mistakenly perform the following sequence:
1) task makes system call to alter a cpuset
2) take cpuset_sem
3) try to allocate memory
4) memory allocator, via cpuset_zone_allowed, trys to take cpuset_sem
5) deadlock
The reason that this is not a serious bug for most users
is that almost all calls to allocate memory don't require
taking cpuset_sem. Only some code paths off the beaten
track require taking cpuset_sem -- which is good. Taking
a global semaphore on the main code path for allocating
memory would not scale well.
This patch fixes this deadlock by wrapping the up() and down()
calls on cpuset_sem in kernel/cpuset.c with code that tracks
the nesting depth of the current task on that semaphore, and
only does the real down() if the task doesn't hold the lock
already, and only does the real up() if the nesting depth
(number of unmatched downs) is exactly one.
The previous required use of refresh_mems(), anytime that
the cpuset_sem semaphore was acquired and the code executed
while holding that semaphore might try to allocate memory, is
no longer required. Two refresh_mems() calls were removed
thanks to this. This is a good change, as failing to get
all the necessary refresh_mems() calls placed was a primary
source of bugs in this cpuset code. The only remaining call
to refresh_mems() is made while doing a memory allocation,
if certain task memory placement data needs to be updated
from its cpuset, due to the cpuset having been changed behind
the tasks back.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
This patch fixes minor problem that the CPUSETS have when files in the
cpuset filesystem are read after lseek()-ed beyond the EOF.
Signed-off-by: KUROSAWA Takahiro <kurosawa@valinux.co.jp>
Acked-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Now the real motivation for this cpuset mem_exclusive patch series seems
trivial.
This patch keeps a task in or under one mem_exclusive cpuset from provoking an
oom kill of a task under a non-overlapping mem_exclusive cpuset. Since only
interrupt and GFP_ATOMIC allocations are allowed to escape mem_exclusive
containment, there is little to gain from oom killing a task under a
non-overlapping mem_exclusive cpuset, as almost all kernel and user memory
allocation must come from disjoint memory nodes.
This patch enables configuring a system so that a runaway job under one
mem_exclusive cpuset cannot cause the killing of a job in another such cpuset
that might be using very high compute and memory resources for a prolonged
time.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
This patch makes use of the previously underutilized cpuset flag
'mem_exclusive' to provide what amounts to another layer of memory placement
resolution. With this patch, there are now the following four layers of
memory placement available:
1) The whole system (interrupt and GFP_ATOMIC allocations can use this),
2) The nearest enclosing mem_exclusive cpuset (GFP_KERNEL allocations can use),
3) The current tasks cpuset (GFP_USER allocations constrained to here), and
4) Specific node placement, using mbind and set_mempolicy.
These nest - each layer is a subset (same or within) of the previous.
Layer (2) above is new, with this patch. The call used to check whether a
zone (its node, actually) is in a cpuset (in its mems_allowed, actually) is
extended to take a gfp_mask argument, and its logic is extended, in the case
that __GFP_HARDWALL is not set in the flag bits, to look up the cpuset
hierarchy for the nearest enclosing mem_exclusive cpuset, to determine if
placement is allowed. The definition of GFP_USER, which used to be identical
to GFP_KERNEL, is changed to also set the __GFP_HARDWALL bit, in the previous
cpuset_gfp_hardwall_flag patch.
GFP_ATOMIC and GFP_KERNEL allocations will stay within the current tasks
cpuset, so long as any node therein is not too tight on memory, but will
escape to the larger layer, if need be.
The intended use is to allow something like a batch manager to handle several
jobs, each job in its own cpuset, but using common kernel memory for caches
and such. Swapper and oom_kill activity is also constrained to Layer (2). A
task in or below one mem_exclusive cpuset should not cause swapping on nodes
in another non-overlapping mem_exclusive cpuset, nor provoke oom_killing of a
task in another such cpuset. Heavy use of kernel memory for i/o caching and
such by one job should not impact the memory available to jobs in other
non-overlapping mem_exclusive cpusets.
This patch enables providing hardwall, inescapable cpusets for memory
allocations of each job, while sharing kernel memory allocations between
several jobs, in an enclosing mem_exclusive cpuset.
Like Dinakar's patch earlier to enable administering sched domains using the
cpu_exclusive flag, this patch also provides a useful meaning to a cpuset flag
that had previously done nothing much useful other than restrict what cpuset
configurations were 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>
At the suggestion of Nick Piggin and Dinakar, totally disable
the facility to allow cpu_exclusive cpusets to define dynamic
sched domains in Linux 2.6.13, in order to avoid problems
first reported by John Hawkes (corrupt sched data structures
and kernel oops).
This has been built for ppc64, i386, ia64, x86_64, sparc, alpha.
It has been built, booted and tested for cpuset functionality
on an SN2 (ia64).
Dinakar or Nick - could you verify that it for sure does avoid
the problems Hawkes reported. Hawkes is out of town, and I don't
have the recipe to reproduce what he found.
Signed-off-by: Paul Jackson <pj@sgi.com>
Acked-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
The partial disabling of Dinakar's new facility to allow
cpu_exclusive cpusets to define dynamic sched domains
doesn't go far enough. At the suggestion of Nick Piggin
and Dinakar, let us instead totally disable this facility
for 2.6.13, in order to avoid problems first reported
by John Hawkes (corrupt sched data structures and kernel oops).
This patch removes the partial disabling code in 2.6.13-rc7,
in anticipation of the next patch, which will totally disable
it instead.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
As reported by Paul Mackerras <paulus@samba.org>, the previous patch
"cpu_exclusive sched domains fix" broke the ppc64 build with
CONFIC_CPUSET, yielding error messages:
kernel/cpuset.c: In function 'update_cpu_domains':
kernel/cpuset.c:648: error: invalid lvalue in unary '&'
kernel/cpuset.c:648: error: invalid lvalue in unary '&'
On some arch's, the node_to_cpumask() is a function, returning
a cpumask_t. But the for_each_cpu_mask() requires an lvalue mask.
The following patch fixes this build failure by making a copy
of the cpumask_t on the stack.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
This keeps the kernel/cpuset.c routine update_cpu_domains() from
invoking the sched.c routine partition_sched_domains() if the cpuset in
question doesn't fall on node boundaries.
I have boot tested this on an SN2, and with the help of a couple of ad
hoc printk's, determined that it does indeed avoid calling the
partition_sched_domains() routine on partial nodes.
I did not directly verify that this avoids setting up bogus sched
domains or avoids the oops that Hawkes saw.
This patch imposes a silent artificial constraint on which cpusets can
be used to define dynamic sched domains.
This patch should allow proceeding with this new feature in 2.6.13 for
the configurations in which it is useful (node alligned sched domains)
while avoiding trying to setup sched domains in the less useful cases
that can cause the kernel corruption and oops.
Signed-off-by: Paul Jackson <pj@sgi.com>
Acked-by: Ingo Molnar <mingo@elte.hu>
Acked-by: Dinakar Guniguntala <dino@in.ibm.com>
Acked-by: John Hawkes <hawkes@sgi.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Fix possible cpuset_sem ABBA deadlock if 'notify_on_release' set.
For a particular usage pattern, creating and destroying cpusets fairly
frequently using notify_on_release, on a very large system, this deadlock
can be seen every few days. If you are not using the cpuset
notify_on_release feature, you will never see this deadlock.
The existing code, on task exit (or cpuset deletion) did:
get cpuset_sem
if cpuset marked notify_on_release and is ready to release:
compute cpuset path relative to /dev/cpuset mount point
call_usermodehelper() forks /sbin/cpuset_release_agent with path
drop cpuset_sem
Unfortunately, the fork in call_usermodehelper can allocate memory, and
allocating memory can require cpuset_sem, if the mems_generation values
changed in the interim. This results in an ABBA deadlock, trying to obtain
cpuset_sem when it is already held by the current task.
To fix this, I put the cpuset path (which must be computed while holding
cpuset_sem) in a temporary buffer, to be used in the call_usermodehelper
call of /sbin/cpuset_release_agent only _after_ dropping cpuset_sem.
So the new logic is:
get cpuset_sem
if cpuset marked notify_on_release and is ready to release:
compute cpuset path relative to /dev/cpuset mount point
stash path in kmalloc'd buffer
drop cpuset_sem
call_usermodehelper() forks /sbin/cpuset_release_agent with path
free path
The sharp eyed reader might notice that this patch does not contain any
calls to kmalloc. The existing code in the check_for_release() routine was
already kmalloc'ing a buffer to hold the cpuset path. In the old code, it
just held the buffer for a few lines, over the cpuset_release_agent() call
that in turn invoked call_usermodehelper(). In the new code, with the
application of this patch, it returns that buffer via the new char
**ppathbuf parameter, for later use and freeing in cpuset_release_agent(),
which is called after cpuset_sem is dropped. Whereas the old code has just
one call to cpuset_release_agent(), right in the check_for_release()
routine, the new code has three calls to cpuset_release_agent(), from the
various places that a cpuset can be released.
This patch has been build and booted on SN2, and passed a stress test that
previously hit the deadlock within a few seconds.
Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Add kerneldoc to kernel/cpuset.c
Fix cpuset typos in init/Kconfig
Signed-off-by: Randy Dunlap <rdunlap@xenotime.net>
Acked-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Adds the core update_cpu_domains code and updated cpusets documentation
Signed-off-by: Dinakar Guniguntala <dino@in.ibm.com>
Acked-by: Paul Jackson <pj@sgi.com>
Acked-by: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Various filesystem drivers have grown a get_dentry() function that's a
duplicate of lookup_one_len, except that it doesn't take a maximum length
argument and doesn't check for \0 or / in the passed in filename.
Switch all these places to use lookup_one_len.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Cc: Greg KH <greg@kroah.com>
Cc: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
There is a race in the kernel cpuset code, between the code
to handle notify_on_release, and the code to remove a cpuset.
The notify_on_release code can end up trying to access a
cpuset that has been removed. In the most common case, this
causes a NULL pointer dereference from the routine cpuset_path.
However all manner of bad things are possible, in theory at least.
The existing code decrements the cpuset use count, and if the
count goes to zero, processes the notify_on_release request,
if appropriate. However, once the count goes to zero, unless we
are holding the global cpuset_sem semaphore, there is nothing to
stop another task from immediately removing the cpuset entirely,
and recycling its memory.
The obvious fix would be to always hold the cpuset_sem
semaphore while decrementing the use count and dealing with
notify_on_release. However we don't want to force a global
semaphore into the mainline task exit path, as that might create
a scaling problem.
The actual fix is almost as easy - since this is only an issue
for cpusets using notify_on_release, which the top level big
cpusets don't normally need to use, only take the cpuset_sem
for cpusets using notify_on_release.
This code has been run for hours without a hiccup, while running
a cpuset create/destroy stress test that could crash the existing
kernel in seconds. This patch applies to the current -linus
git kernel.
Signed-off-by: Paul Jackson <pj@sgi.com>
Acked-by: Simon Derr <simon.derr@bull.net>
Acked-by: Dinakar Guniguntala <dino@in.ibm.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
gcc-4 warns with
include/linux/cpuset.h:21: warning: type qualifiers ignored on function
return type
cpuset_cpus_allowed is declared with const
extern const cpumask_t cpuset_cpus_allowed(const struct task_struct *p);
First const should be __attribute__((const)), but the gcc manual
explains that:
"Note that a function that has pointer arguments and examines the data
pointed to must not be declared const. Likewise, a function that calls a
non-const function usually must not be const. It does not make sense for
a const function to return void."
The following patch remove const from the function declaration.
Signed-off-by: Benoit Boissinot <benoit.boissinot@ens-lyon.org>
Acked-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Initial git repository build. I'm not bothering with the full history,
even though we have it. We can create a separate "historical" git
archive of that later if we want to, and in the meantime it's about
3.2GB when imported into git - space that would just make the early
git days unnecessarily complicated, when we don't have a lot of good
infrastructure for it.
Let it rip!
Paul Jackson