linux-toradex.git/include/linux/cpuset.h, branch v3.6-rc1

cpusets, hotplug: Restructure functions that are invoked during hotplug

2012-07-24T11:53:22+00:00

Separate out the cpuset related handling for CPU/Memory online/offline.
This also helps us exploit the most obvious and basic level of optimization
that any notification mechanism (CPU/Mem online/offline) has to offer us:
"We *know* why we have been invoked. So stop pretending that we are lost,
and do only the necessary amount of processing!".

And while at it, rename scan_for_empty_cpusets() to
scan_cpusets_upon_hotplug(), which is more appropriate considering how
it is restructured.

Signed-off-by: Srivatsa S. Bhat 
Signed-off-by: Peter Zijlstra 
Cc: Linus Torvalds 
Cc: Andrew Morton 
Cc: Thomas Gleixner 
Link: http://lkml.kernel.org/r/20120524141650.3692.48637.stgit@srivatsabhat.in.ibm.com
Signed-off-by: Ingo Molnar

Merge branch 'sched-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip

2012-03-29T21:46:05+00:00

Pull scheduler fixes from Ingo Molnar.

* 'sched-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  cpusets: Remove an unused variable
  sched/rt: Improve pick_next_highest_task_rt()
  sched: Fix select_fallback_rq() vs cpu_active/cpu_online
  sched/x86/smp: Do not enable IRQs over calibrate_delay()
  sched: Fix compiler warning about declared inline after use
  MAINTAINERS: Update email address for SCHEDULER and PERF EVENTS

sched: Fix select_fallback_rq() vs cpu_active/cpu_online

2012-03-27T12:50:14+00:00

Commit 5fbd036b55 ("sched: Cleanup cpu_active madness"), which was
supposed to finally sort the cpu_active mess, instead uncovered more.

Since CPU_STARTING is ran before setting the cpu online, there's a
(small) window where the cpu has active,!online.

If during this time there's a wakeup of a task that used to reside on
that cpu select_task_rq() will use select_fallback_rq() to compute an
alternative cpu to run on since we find !online.

select_fallback_rq() however will compute the new cpu against
cpu_active, this means that it can return the same cpu it started out
with, the !online one, since that cpu is in fact marked active.

This results in us trying to scheduling a task on an offline cpu and
triggering a WARN in the IPI code.

The solution proposed by Chuansheng Liu of setting cpu_active in
set_cpu_online() is buggy, firstly not all archs actually use
set_cpu_online(), secondly, not all archs call set_cpu_online() with
IRQs disabled, this means we would introduce either the same race or
the race from fd8a7de17 ("x86: cpu-hotplug: Prevent softirq wakeup on
wrong CPU") -- albeit much narrower.

[ By setting online first and active later we have a window of
  online,!active, fresh and bound kthreads have task_cpu() of 0 and
  since cpu0 isn't in tsk_cpus_allowed() we end up in
  select_fallback_rq() which excludes !active, resulting in a reset
  of ->cpus_allowed and the thread running all over the place. ]

The solution is to re-work select_fallback_rq() to require active
_and_ online. This makes the active,!online case work as expected,
OTOH archs running CPU_STARTING after setting online are now
vulnerable to the issue from fd8a7de17 -- these are alpha and
blackfin.

Reported-by: Chuansheng Liu 
Signed-off-by: Peter Zijlstra 
Cc: Mike Frysinger 
Cc: linux-alpha@vger.kernel.org
Link: http://lkml.kernel.org/n/tip-hubqk1i10o4dpvlm06gq7v6j@git.kernel.org
Signed-off-by: Ingo Molnar

cpuset: mm: reduce large amounts of memory barrier related damage v3

2012-03-22T00:54:59+00:00

Commit c0ff7453bb5c ("cpuset,mm: fix no node to alloc memory when
changing cpuset's mems") wins a super prize for the largest number of
memory barriers entered into fast paths for one commit.

[get|put]_mems_allowed is incredibly heavy with pairs of full memory
barriers inserted into a number of hot paths.  This was detected while
investigating at large page allocator slowdown introduced some time
after 2.6.32.  The largest portion of this overhead was shown by
oprofile to be at an mfence introduced by this commit into the page
allocator hot path.

For extra style points, the commit introduced the use of yield() in an
implementation of what looks like a spinning mutex.

This patch replaces the full memory barriers on both read and write
sides with a sequence counter with just read barriers on the fast path
side.  This is much cheaper on some architectures, including x86.  The
main bulk of the patch is the retry logic if the nodemask changes in a
manner that can cause a false failure.

While updating the nodemask, a check is made to see if a false failure
is a risk.  If it is, the sequence number gets bumped and parallel
allocators will briefly stall while the nodemask update takes place.

In a page fault test microbenchmark, oprofile samples from
__alloc_pages_nodemask went from 4.53% of all samples to 1.15%.  The
actual results were

                             3.3.0-rc3          3.3.0-rc3
                             rc3-vanilla        nobarrier-v2r1
    Clients   1 UserTime       0.07 (  0.00%)   0.08 (-14.19%)
    Clients   2 UserTime       0.07 (  0.00%)   0.07 (  2.72%)
    Clients   4 UserTime       0.08 (  0.00%)   0.07 (  3.29%)
    Clients   1 SysTime        0.70 (  0.00%)   0.65 (  6.65%)
    Clients   2 SysTime        0.85 (  0.00%)   0.82 (  3.65%)
    Clients   4 SysTime        1.41 (  0.00%)   1.41 (  0.32%)
    Clients   1 WallTime       0.77 (  0.00%)   0.74 (  4.19%)
    Clients   2 WallTime       0.47 (  0.00%)   0.45 (  3.73%)
    Clients   4 WallTime       0.38 (  0.00%)   0.37 (  1.58%)
    Clients   1 Flt/sec/cpu  497620.28 (  0.00%) 520294.53 (  4.56%)
    Clients   2 Flt/sec/cpu  414639.05 (  0.00%) 429882.01 (  3.68%)
    Clients   4 Flt/sec/cpu  257959.16 (  0.00%) 258761.48 (  0.31%)
    Clients   1 Flt/sec      495161.39 (  0.00%) 517292.87 (  4.47%)
    Clients   2 Flt/sec      820325.95 (  0.00%) 850289.77 (  3.65%)
    Clients   4 Flt/sec      1020068.93 (  0.00%) 1022674.06 (  0.26%)
    MMTests Statistics: duration
    Sys Time Running Test (seconds)             135.68    132.17
    User+Sys Time Running Test (seconds)         164.2    160.13
    Total Elapsed Time (seconds)                123.46    120.87

The overall improvement is small but the System CPU time is much
improved and roughly in correlation to what oprofile reported (these
performance figures are without profiling so skew is expected).  The
actual number of page faults is noticeably improved.

For benchmarks like kernel builds, the overall benefit is marginal but
the system CPU time is slightly reduced.

To test the actual bug the commit fixed I opened two terminals.  The
first ran within a cpuset and continually ran a small program that
faulted 100M of anonymous data.  In a second window, the nodemask of the
cpuset was continually randomised in a loop.

Without the commit, the program would fail every so often (usually
within 10 seconds) and obviously with the commit everything worked fine.
With this patch applied, it also worked fine so the fix should be
functionally equivalent.

Signed-off-by: Mel Gorman 
Cc: Miao Xie 
Cc: David Rientjes 
Cc: Peter Zijlstra 
Cc: Christoph Lameter 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds

cpuset: Fix cpuset_cpus_allowed_fallback(), don't update tsk->rt.nr_cpus_allowed

2011-05-28T15:02:57+00:00

The rule is, we have to update tsk->rt.nr_cpus_allowed if we change
tsk->cpus_allowed. Otherwise RT scheduler may confuse.

Signed-off-by: KOSAKI Motohiro 
Cc: Oleg Nesterov 
Signed-off-by: Peter Zijlstra 
Link: http://lkml.kernel.org/r/4DD4B3FA.5060901@jp.fujitsu.com
Signed-off-by: Ingo Molnar

sched: adjust when cpu_active and cpuset configurations are updated during cpu on/offlining

2010-06-08T19:40:36+00:00

Currently, when a cpu goes down, cpu_active is cleared before
CPU_DOWN_PREPARE starts and cpuset configuration is updated from a
default priority cpu notifier.  When a cpu is coming up, it's set
before CPU_ONLINE but cpuset configuration again is updated from the
same cpu notifier.

For cpu notifiers, this presents an inconsistent state.  Threads which
a CPU_DOWN_PREPARE notifier expects to be bound to the CPU can be
migrated to other cpus because the cpu is no more inactive.

Fix it by updating cpu_active in the highest priority cpu notifier and
cpuset configuration in the second highest when a cpu is coming up.
Down path is updated similarly.  This guarantees that all other cpu
notifiers see consistent cpu_active and cpuset configuration.

cpuset_track_online_cpus() notifier is converted to
cpuset_update_active_cpus() which just updates the configuration and
now called from cpuset_cpu_[in]active() notifiers registered from
sched_init_smp().  If cpuset is disabled, cpuset_update_active_cpus()
degenerates into partition_sched_domains() making separate notifier
for !CONFIG_CPUSETS unnecessary.

This problem is triggered by cmwq.  During CPU_DOWN_PREPARE, hotplug
callback creates a kthread and kthread_bind()s it to the target cpu,
and the thread is expected to run on that cpu.

* Ingo's test discovered __cpuinit/exit markups were incorrect.
  Fixed.

Signed-off-by: Tejun Heo 
Acked-by: Peter Zijlstra 
Cc: Rusty Russell 
Cc: Ingo Molnar 
Cc: Paul Menage

cpusets: new round-robin rotor for SLAB allocations

2010-05-27T16:12:44+00:00

We have observed several workloads running on multi-node systems where
memory is assigned unevenly across the nodes in the system.  There are
numerous reasons for this but one is the round-robin rotor in
cpuset_mem_spread_node().

For example, a simple test that writes a multi-page file will allocate
pages on nodes 0 2 4 6 ...  Odd nodes are skipped.  (Sometimes it
allocates on odd nodes & skips even nodes).

An example is shown below.  The program "lfile" writes a file consisting
of 10 pages.  The program then mmaps the file & uses get_mempolicy(...,
MPOL_F_NODE) to determine the nodes where the file pages were allocated.
The output is shown below:

	# ./lfile
	 allocated on nodes: 2 4 6 0 1 2 6 0 2

There is a single rotor that is used for allocating both file pages & slab
pages.  Writing the file allocates both a data page & a slab page
(buffer_head).  This advances the RR rotor 2 nodes for each page
allocated.

A quick confirmation seems to confirm this is the cause of the uneven
allocation:

	# echo 0 >/dev/cpuset/memory_spread_slab
	# ./lfile
	 allocated on nodes: 6 7 8 9 0 1 2 3 4 5

This patch introduces a second rotor that is used for slab allocations.

Signed-off-by: Jack Steiner 
Acked-by: Christoph Lameter 
Cc: Pekka Enberg 
Cc: Paul Menage 
Cc: Jack Steiner 
Cc: Robin Holt 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds

cpuset,mm: fix no node to alloc memory when changing cpuset's mems

2010-05-25T15:06:57+00:00

Before applying this patch, cpuset updates task->mems_allowed and
mempolicy by setting all new bits in the nodemask first, and clearing all
old unallowed bits later.  But in the way, the allocator may find that
there is no node to alloc memory.

The reason is that cpuset rebinds the task's mempolicy, it cleans the
nodes which the allocater can alloc pages on, for example:

(mpol: mempolicy)
	task1			task1's mpol	task2
	alloc page		1
	  alloc on node0? NO	1
				1		change mems from 1 to 0
				1		rebind task1's mpol
				0-1		  set new bits
				0	  	  clear disallowed bits
	  alloc on node1? NO	0
	  ...
	can't alloc page
	  goto oom

This patch fixes this problem by expanding the nodes range first(set newly
allowed bits) and shrink it lazily(clear newly disallowed bits).  So we
use a variable to tell the write-side task that read-side task is reading
nodemask, and the write-side task clears newly disallowed nodes after
read-side task ends the current memory allocation.

[akpm@linux-foundation.org: fix spello]
Signed-off-by: Miao Xie 
Cc: David Rientjes 
Cc: Nick Piggin 
Cc: Paul Menage 
Cc: Lee Schermerhorn 
Cc: Hugh Dickins 
Cc: Ravikiran Thirumalai 
Cc: KOSAKI Motohiro 
Cc: Christoph Lameter 
Cc: Andi Kleen 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds

sched: Make select_fallback_rq() cpuset friendly

2010-04-02T18:12:03+00:00

Introduce cpuset_cpus_allowed_fallback() helper to fix the cpuset problems
with select_fallback_rq(). It can be called from any context and can't use
any cpuset locks including task_lock(). It is called when the task doesn't
have online cpus in ->cpus_allowed but ttwu/etc must be able to find a
suitable cpu.

I am not proud of this patch. Everything which needs such a fat comment
can't be good even if correct. But I'd prefer to not change the locking
rules in the code I hardly understand, and in any case I believe this
simple change make the code much more correct compared to deadlocks we
currently have.

Signed-off-by: Oleg Nesterov 
Signed-off-by: Peter Zijlstra 
LKML-Reference: <20100315091027.GA9155@redhat.com>
Signed-off-by: Ingo Molnar

sched: Kill the broken and deadlockable cpuset_lock/cpuset_cpus_allowed_locked code

2010-04-02T18:12:01+00:00

This patch just states the fact the cpusets/cpuhotplug interaction is
broken and removes the deadlockable code which only pretends to work.

- cpuset_lock() doesn't really work. It is needed for
  cpuset_cpus_allowed_locked() but we can't take this lock in
  try_to_wake_up()->select_fallback_rq() path.

- cpuset_lock() is deadlockable. Suppose that a task T bound to CPU takes
  callback_mutex. If cpu_down(CPU) happens before T drops callback_mutex
  stop_machine() preempts T, then migration_call(CPU_DEAD) tries to take
  cpuset_lock() and hangs forever because CPU is already dead and thus
  T can't be scheduled.

- cpuset_cpus_allowed_locked() is deadlockable too. It takes task_lock()
  which is not irq-safe, but try_to_wake_up() can be called from irq.

Kill them, and change select_fallback_rq() to use cpu_possible_mask, like
we currently do without CONFIG_CPUSETS.

Also, with or without this patch, with or without CONFIG_CPUSETS, the
callers of select_fallback_rq() can race with each other or with
set_cpus_allowed() pathes.

The subsequent patches try to to fix these problems.

Signed-off-by: Oleg Nesterov 
Signed-off-by: Peter Zijlstra 
LKML-Reference: <20100315091003.GA9123@redhat.com>
Signed-off-by: Ingo Molnar