linux-toradex.git/kernel/sched, branch v3.9.4

sched: Avoid prev->stime underflow

2013-05-19T18:38:19+00:00

commit 68aa8efcd1ab961e4684ef5af32f72a6ec1911de upstream.

Dave Hansen reported strange utime/stime values on his system:
https://lkml.org/lkml/2013/4/4/435

This happens because prev->stime value is bigger than rtime
value. Root of the problem are non-monotonic rtime values (i.e.
current rtime is smaller than previous rtime) and that should be
debugged and fixed.

But since problem did not manifest itself before commit
62188451f0d63add7ad0cd2a1ae269d600c1663d "cputime: Avoid
multiplication overflow on utime scaling", it should be threated
as regression, which we can easily fixed on cputime_adjust()
function.

For now, let's apply this fix, but further work is needed to fix
root of the problem.

Reported-and-tested-by: Dave Hansen 
Signed-off-by: Stanislaw Gruszka 
Cc: Frederic Weisbecker 
Cc: rostedt@goodmis.org
Cc: Linus Torvalds 
Cc: Dave Hansen 
Cc: Peter Zijlstra 
Link: http://lkml.kernel.org/r/1367314507-9728-3-git-send-email-sgruszka@redhat.com
Signed-off-by: Ingo Molnar 
Signed-off-by: Greg Kroah-Hartman

sched: Do not account bogus utime

2013-05-19T18:38:18+00:00

commit 772c808a252594692972773f6ee41c289b8e0b2a upstream.

Due to rounding in scale_stime(), for big numbers, scaled stime
values will grow in chunks. Since rtime grow in jiffies and we
calculate utime like below:

	prev->stime = max(prev->stime, stime);
	prev->utime = max(prev->utime, rtime - prev->stime);

we could erroneously account stime values as utime. To prevent
that only update prev->{u,s}time values when they are smaller
than current rtime.

Signed-off-by: Stanislaw Gruszka 
Cc: Frederic Weisbecker 
Cc: rostedt@goodmis.org
Cc: Linus Torvalds 
Cc: Dave Hansen 
Cc: Peter Zijlstra 
Link: http://lkml.kernel.org/r/1367314507-9728-2-git-send-email-sgruszka@redhat.com
Signed-off-by: Ingo Molnar 
Signed-off-by: Greg Kroah-Hartman

sched: Avoid cputime scaling overflow

2013-05-19T18:38:17+00:00

commit 55eaa7c1f511af5fb6ef808b5328804f4d4e5243 upstream.

Here is patch, which adds Linus's cputime scaling algorithm to the
kernel.

This is a follow up (well, fix) to commit
d9a3c9823a2e6a543eb7807fb3d15d8233817ec5 ("sched: Lower chances
of cputime scaling overflow") which commit tried to avoid
multiplication overflow, but did not guarantee that the overflow
would not happen.

Linus crated a different algorithm, which completely avoids the
multiplication overflow by dropping precision when numbers are
big.

It was tested by me and it gives good relative error of
scaled numbers. Testing method is described here:
http://marc.info/?l=linux-kernel&m=136733059505406&w=2

Originally-From: Linus Torvalds 
Signed-off-by: Stanislaw Gruszka 
Cc: Frederic Weisbecker 
Cc: rostedt@goodmis.org
Cc: Dave Hansen 
Cc: Peter Zijlstra 
Link: http://lkml.kernel.org/r/20130430151441.GC10465@redhat.com
Signed-off-by: Ingo Molnar 
Signed-off-by: Greg Kroah-Hartman

sched: Lower chances of cputime scaling overflow

2013-05-19T18:38:17+00:00

commit d9a3c9823a2e6a543eb7807fb3d15d8233817ec5 upstream.

Some users have reported that after running a process with
hundreds of threads on intensive CPU-bound loads, the cputime
of the group started to freeze after a few days.

This is due to how we scale the tick-based cputime against
the scheduler precise execution time value.

We add the values of all threads in the group and we multiply
that against the sum of the scheduler exec runtime of the whole
group.

This easily overflows after a few days/weeks of execution.

A proposed solution to solve this was to compute that multiplication
on stime instead of utime:
   62188451f0d63add7ad0cd2a1ae269d600c1663d
   ("cputime: Avoid multiplication overflow on utime scaling")

The rationale behind that was that it's easy for a thread to
spend most of its time in userspace under intensive CPU-bound workload
but it's much harder to do CPU-bound intensive long run in the kernel.

This postulate got defeated when a user recently reported he was still
seeing cputime freezes after the above patch. The workload that
triggers this issue relates to intensive networking workloads where
most of the cputime is consumed in the kernel.

To reduce much more the opportunities for multiplication overflow,
lets reduce the multiplication factors to the remainders of the division
between sched exec runtime and cputime. Assuming the difference between
these shouldn't ever be that large, it could work on many situations.

This gets the same results as in the upstream scaling code except for
a small difference: the upstream code always rounds the results to
the nearest integer not greater to what would be the precise result.
The new code rounds to the nearest integer either greater or not
greater. In practice this difference probably shouldn't matter but
it's worth mentioning.

If this solution appears not to be enough in the end, we'll
need to partly revert back to the behaviour prior to commit
     0cf55e1ec08bb5a22e068309e2d8ba1180ab4239
     ("sched, cputime: Introduce thread_group_times()")

Back then, the scaling was done on exit() time before adding the cputime
of an exiting thread to the signal struct. And then we'll need to
scale one-by-one the live threads cputime in thread_group_cputime(). The
drawback may be a slightly slower code on exit time.

Signed-off-by: Frederic Weisbecker 
Cc: Stanislaw Gruszka 
Cc: Steven Rostedt 
Cc: Peter Zijlstra 
Cc: Ingo Molnar 
Cc: Andrew Morton 
Signed-off-by: Stanislaw Gruszka 
Acked-by: Ingo Molnar 
Signed-off-by: Greg Kroah-Hartman

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

2013-04-14T18:12:17+00:00

Pull scheduler fixes from Ingo Molnar:
 "Misc fixlets"

* 'sched-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  sched/cputime: Fix accounting on multi-threaded processes
  sched/debug: Fix sd->*_idx limit range avoiding overflow
  sched_clock: Prevent 64bit inatomicity on 32bit systems
  sched: Convert BUG_ON()s in try_to_wake_up_local() to WARN_ON_ONCE()s

sched/cputime: Fix accounting on multi-threaded processes

2013-04-08T15:40:52+00:00

Recent commit 6fac4829 ("cputime: Use accessors to read task
cputime stats") introduced a bug, where we account many times
the cputime of the first thread, instead of cputimes of all
the different threads.

Signed-off-by: Stanislaw Gruszka 
Acked-by: Frederic Weisbecker 
Cc: Oleg Nesterov 
Cc: Peter Zijlstra 
Link: http://lkml.kernel.org/r/20130404085740.GA2495@redhat.com
Signed-off-by: Ingo Molnar

sched/debug: Fix sd->*_idx limit range avoiding overflow

2013-04-08T11:23:03+00:00

Commit 201c373e8e ("sched/debug: Limit sd->*_idx range on
sysctl") was an incomplete bug fix.

This patch fixes sd->*_idx limit range to [0 ~ CPU_LOAD_IDX_MAX-1]
avoiding array overflow caused by setting sd->*_idx to CPU_LOAD_IDX_MAX
on sysctl.

Signed-off-by: Libin 
Cc: 
Cc: 
Cc: Peter Zijlstra 
Link: http://lkml.kernel.org/r/51626610.2040607@huawei.com
Signed-off-by: Ingo Molnar

sched_clock: Prevent 64bit inatomicity on 32bit systems

2013-04-08T09:50:44+00:00

The sched_clock_remote() implementation has the following inatomicity
problem on 32bit systems when accessing the remote scd->clock, which
is a 64bit value.

CPU0			CPU1

sched_clock_local()	sched_clock_remote(CPU0)
...
			remote_clock = scd[CPU0]->clock
			    read_low32bit(scd[CPU0]->clock)
cmpxchg64(scd->clock,...)
			    read_high32bit(scd[CPU0]->clock)

While the update of scd->clock is using an atomic64 mechanism, the
readout on the remote cpu is not, which can cause completely bogus
readouts.

It is a quite rare problem, because it requires the update to hit the
narrow race window between the low/high readout and the update must go
across the 32bit boundary.

The resulting misbehaviour is, that CPU1 will see the sched_clock on
CPU1 ~4 seconds ahead of it's own and update CPU1s sched_clock value
to this bogus timestamp. This stays that way due to the clamping
implementation for about 4 seconds until the synchronization with
CLOCK_MONOTONIC undoes the problem.

The issue is hard to observe, because it might only result in a less
accurate SCHED_OTHER timeslicing behaviour. To create observable
damage on realtime scheduling classes, it is necessary that the bogus
update of CPU1 sched_clock happens in the context of an realtime
thread, which then gets charged 4 seconds of RT runtime, which results
in the RT throttler mechanism to trigger and prevent scheduling of RT
tasks for a little less than 4 seconds. So this is quite unlikely as
well.

The issue was quite hard to decode as the reproduction time is between
2 days and 3 weeks and intrusive tracing makes it less likely, but the
following trace recorded with trace_clock=global, which uses
sched_clock_local(), gave the final hint:

  -0   0d..30 400269.477150: hrtimer_cancel: hrtimer=0xf7061e80
  -0   0d..30 400269.477151: hrtimer_start:  hrtimer=0xf7061e80 ...
irq/20-S-587 1d..32 400273.772118: sched_wakeup:   comm= ... target_cpu=0
  -0   0dN.30 400273.772118: hrtimer_cancel: hrtimer=0xf7061e80

What happens is that CPU0 goes idle and invokes
sched_clock_idle_sleep_event() which invokes sched_clock_local() and
CPU1 runs a remote wakeup for CPU0 at the same time, which invokes
sched_remote_clock(). The time jump gets propagated to CPU0 via
sched_remote_clock() and stays stale on both cores for ~4 seconds.

There are only two other possibilities, which could cause a stale
sched clock:

1) ktime_get() which reads out CLOCK_MONOTONIC returns a sporadic
   wrong value.

2) sched_clock() which reads the TSC returns a sporadic wrong value.

#1 can be excluded because sched_clock would continue to increase for
   one jiffy and then go stale.

#2 can be excluded because it would not make the clock jump
   forward. It would just result in a stale sched_clock for one jiffy.

After quite some brain twisting and finding the same pattern on other
traces, sched_clock_remote() remained the only place which could cause
such a problem and as explained above it's indeed racy on 32bit
systems.

So while on 64bit systems the readout is atomic, we need to verify the
remote readout on 32bit machines. We need to protect the local->clock
readout in sched_clock_remote() on 32bit as well because an NMI could
hit between the low and the high readout, call sched_clock_local() and
modify local->clock.

Thanks to Siegfried Wulsch for bearing with my debug requests and
going through the tedious tasks of running a bunch of reproducer
systems to generate the debug information which let me decode the
issue.

Reported-by: Siegfried Wulsch 
Acked-by: Peter Zijlstra 
Cc: Steven Rostedt 
Link: http://lkml.kernel.org/r/alpine.LFD.2.02.1304051544160.21884@ionos
Signed-off-by: Thomas Gleixner 
Cc: stable@vger.kernel.org

sched: Convert BUG_ON()s in try_to_wake_up_local() to WARN_ON_ONCE()s

2013-03-21T10:48:20+00:00

try_to_wake_up_local() should only be invoked to wake up another
task in the same runqueue and BUG_ON()s are used to enforce the
rule. Missing try_to_wake_up_local() can stall workqueue
execution but such stalls are likely to be finite either by
another work item being queued or the one blocked getting
unblocked.  There's no reason to trigger BUG while holding rq
lock crashing the whole system.

Convert BUG_ON()s in try_to_wake_up_local() to WARN_ON_ONCE()s.

Signed-off-by: Tejun Heo 
Acked-by: Steven Rostedt 
Cc: Peter Zijlstra 
Link: http://lkml.kernel.org/r/20130318192234.GD3042@htj.dyndns.org
Signed-off-by: Ingo Molnar

Merge branch 'for-3.9/core' of git://git.kernel.dk/linux-block

2013-02-28T20:52:24+00:00

Pull block IO core bits from Jens Axboe:
 "Below are the core block IO bits for 3.9.  It was delayed a few days
  since my workstation kept crashing every 2-8h after pulling it into
  current -git, but turns out it is a bug in the new pstate code (divide
  by zero, will report separately).  In any case, it contains:

   - The big cfq/blkcg update from Tejun and and Vivek.

   - Additional block and writeback tracepoints from Tejun.

   - Improvement of the should sort (based on queues) logic in the plug
     flushing.

   - _io() variants of the wait_for_completion() interface, using
     io_schedule() instead of schedule() to contribute to io wait
     properly.

   - Various little fixes.

  You'll get two trivial merge conflicts, which should be easy enough to
  fix up"

Fix up the trivial conflicts due to hlist traversal cleanups (commit
b67bfe0d42ca: "hlist: drop the node parameter from iterators").

* 'for-3.9/core' of git://git.kernel.dk/linux-block: (39 commits)
  block: remove redundant check to bd_openers()
  block: use i_size_write() in bd_set_size()
  cfq: fix lock imbalance with failed allocations
  drivers/block/swim3.c: fix null pointer dereference
  block: don't select PERCPU_RWSEM
  block: account iowait time when waiting for completion of IO request
  sched: add wait_for_completion_io[_timeout]
  writeback: add more tracepoints
  block: add block_{touch|dirty}_buffer tracepoint
  buffer: make touch_buffer() an exported function
  block: add @req to bio_{front|back}_merge tracepoints
  block: add missing block_bio_complete() tracepoint
  block: Remove should_sort judgement when flush blk_plug
  block,elevator: use new hashtable implementation
  cfq-iosched: add hierarchical cfq_group statistics
  cfq-iosched: collect stats from dead cfqgs
  cfq-iosched: separate out cfqg_stats_reset() from cfq_pd_reset_stats()
  blkcg: make blkcg_print_blkgs() grab q locks instead of blkcg lock
  block: RCU free request_queue
  blkcg: implement blkg_[rw]stat_recursive_sum() and blkg_[rw]stat_merge()
  ...