Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1443594

Re: Severe performance regression w/ 4.4+ on Android due to cgroup locking changes

From Peter Zijlstra <peterz@infradead.org>
Newsgroups linux.kernel
Subject Re: Severe performance regression w/ 4.4+ on Android due to cgroup locking changes
Date 2016-07-14 18:50 +0200
Message-ID <rURZM-3En-29@gated-at.bofh.it> (permalink)
References (5 earlier) <rUyX7-7ZC-1@gated-at.bofh.it> <rUzgu-87F-9@gated-at.bofh.it> <rUzqa-8bo-19@gated-at.bofh.it> <rUOIy-1JY-1@gated-at.bofh.it> <rURGp-3xe-11@gated-at.bofh.it>
Organization linux.* mail to news gateway

Show all headers | View raw

On Thu, Jul 14, 2016 at 09:23:55AM -0700, Paul E. McKenney wrote:
> Hmmm...  How does this handle the following sequence of events for
> the case where we are not biased towards the reader?
> 
> o	The per-CPU rwsem is set up with RCU_NONE and readers_slow
> 	(as opposed to readers_block).  The rcu_sync ->gp_state is
> 	GP_PENDING, which means that rcu_sync_is_idle() will always
> 	return true.

/false/, rcu_sync_is_idle() will always be false, to force us into the
slowpath.

> o	Task A on CPU 0 runs percpu_down_read() to completion, and remains
> 	within the critical section.  CPU 0's ->refcount is therefore 1.
> 
> o	Task B on CPU 1 does percpu_down_write(), which write-acquires
> 	->rw_sem, does rcu_sync_enter() (which is a no-op due to
> 	RCU_NONE), sets ->state to readers_block, and is just now going
> 	to wait for readers, which first invokes readers_active_check(),
> 	which starts summing the ->refcount for CPUs 0, 1, and 2,
> 	obtaining the value 1 so far.
> 
> o	Task C CPU 2 enters percpu_down_read(), disables preemption,
> 	increments CPU 2's ->refcount, sees rcu_sync_is_idle() return
> 	true (so skips __percpu_down_read()), enables preemption, and
> 	enters its critical section.

false, so does __percpu_down_read()

> 
> o	Task C migrates to CPU 3 and invokes percpu_up_read(), which
> 	disables preemption, sees rcu_sync_is_idle() return true, calls
> 	__this_cpu_dec() on CPU 3's ->refcount, and enables preemption.
> 	The value of CPU 3's ->refcount is thus (unsigned int)-1.

__percpu_up_read()

> 
> o	Task B on CPU 1 continues execution in readers_active_check(), with
> 	the full sum being zero.
> 
> So it looks to me like we have Task A as a writer at the same time that
> Task A is a reader, which would not be so good.
> 
> So what am I missing here?

for RCU_NONE we init rsp->gp_state to !0, which makes:

static inline rcu_sync_is_idle()'s

	return !rsp->gp_state (aka. rsp->gp_state == 0)

return false.

> And a couple of checkpatch nits below.  Yes, I had to apply the patch to
> figure out what it was doing.  ;-)

Yah, too much churn to read :-)

In any case, rest assured you've already gone over this part of the
patch several times. I repurposed an old percpu-rwsem optimization, Oleg
recognised it.

Back to linux.kernel | Previous | Next — Previous in thread | Next in thread | Find similar | Unroll thread

Thread

Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes John Stultz <john.stultz@linaro.org> - 2016-07-13 02:10 +0200
  Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes Peter Zijlstra <peterz@infradead.org> - 2016-07-13 10:30 +0200
    Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes "Paul E. McKenney" <paulmck@linux.vnet.ibm.com> - 2016-07-13 16:50 +0200
      Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes "Paul E. McKenney" <paulmck@linux.vnet.ibm.com> - 2016-07-13 20:40 +0200
  Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes Tejun Heo <tj@kernel.org> - 2016-07-13 20:30 +0200
    Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes Tejun Heo <tj@kernel.org> - 2016-07-13 20:40 +0200
      Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes Tejun Heo <tj@kernel.org> - 2016-07-13 22:20 +0200
        Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes Peter Zijlstra <peterz@infradead.org> - 2016-07-13 22:30 +0200
          Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes Tejun Heo <tj@kernel.org> - 2016-07-13 22:50 +0200
            Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes Peter Zijlstra <peterz@infradead.org> - 2016-07-13 23:00 +0200
              Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes Tejun Heo <tj@kernel.org> - 2016-07-13 23:10 +0200
                Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes "Paul E. McKenney" <paulmck@linux.vnet.ibm.com> - 2016-07-13 23:20 +0200
                Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes "Paul E. McKenney" <paulmck@linux.vnet.ibm.com> - 2016-07-13 23:50 +0200
                Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes John Stultz <john.stultz@linaro.org> - 2016-07-13 23:50 +0200
                Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes "Paul E. McKenney" <paulmck@linux.vnet.ibm.com> - 2016-07-14 00:20 +0200
                Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes John Stultz <john.stultz@linaro.org> - 2016-07-14 00:40 +0200
                Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes "Paul E. McKenney" <paulmck@linux.vnet.ibm.com> - 2016-07-14 01:10 +0200
                Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes "Paul E. McKenney" <paulmck@linux.vnet.ibm.com> - 2016-07-14 01:10 +0200
                Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes Tejun Heo <tj@kernel.org> - 2016-07-14 13:40 +0200
                Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes Peter Zijlstra <peterz@infradead.org> - 2016-07-14 14:10 +0200
                Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes Tejun Heo <tj@kernel.org> - 2016-07-14 14:10 +0200
                Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes Peter Zijlstra <peterz@infradead.org> - 2016-07-14 14:30 +0200
                Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes Tejun Heo <tj@kernel.org> - 2016-07-14 17:10 +0200
                Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes Tejun Heo <tj@kernel.org> - 2016-07-14 17:30 +0200
                Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes Peter Zijlstra <peterz@infradead.org> - 2016-07-14 18:40 +0200
                Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes Oleg Nesterov <oleg@redhat.com> - 2016-07-14 19:40 +0200
                Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes John Stultz <john.stultz@linaro.org> - 2016-07-14 19:00 +0200
                Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes John Stultz <john.stultz@linaro.org> - 2016-07-14 00:30 +0200
                Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes Tejun Heo <tj@kernel.org> - 2016-07-14 00:10 +0200
                Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes "Paul E. McKenney" <paulmck@linux.vnet.ibm.com> - 2016-07-14 00:40 +0200
                Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes Peter Zijlstra <peterz@infradead.org> - 2016-07-14 09:00 +0200
                Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes Tejun Heo <tj@kernel.org> - 2016-07-14 13:30 +0200
                Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes Peter Zijlstra <peterz@infradead.org> - 2016-07-14 14:20 +0200
                Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes Tejun Heo <tj@kernel.org> - 2016-07-14 17:20 +0200
              Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes Tejun Heo <tj@kernel.org> - 2016-07-13 23:10 +0200
              Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes "Paul E. McKenney" <paulmck@linux.vnet.ibm.com> - 2016-07-13 23:10 +0200
              Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes Peter Zijlstra <peterz@infradead.org> - 2016-07-14 15:20 +0200
                Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes Peter Zijlstra <peterz@infradead.org> - 2016-07-14 16:20 +0200
                Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes Oleg Nesterov <oleg@redhat.com> - 2016-07-14 17:00 +0200
                Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes Peter Zijlstra <peterz@infradead.org> - 2016-07-14 18:20 +0200
                Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes Peter Zijlstra <peterz@infradead.org> - 2016-07-14 18:40 +0200
                Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes Oleg Nesterov <oleg@redhat.com> - 2016-07-14 19:10 +0200
                Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes "Paul E. McKenney" <paulmck@linux.vnet.ibm.com> - 2016-07-14 18:30 +0200
                Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes Peter Zijlstra <peterz@infradead.org> - 2016-07-14 18:50 +0200
                Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes "Paul E. McKenney" <paulmck@linux.vnet.ibm.com> - 2016-07-14 19:20 +0200
                Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes John Stultz <john.stultz@linaro.org> - 2016-07-14 18:50 +0200
                Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes Peter Zijlstra <peterz@infradead.org> - 2016-07-14 18:50 +0200
                Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes John Stultz <john.stultz@linaro.org> - 2016-07-14 19:10 +0200
                Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes Oleg Nesterov <oleg@redhat.com> - 2016-07-14 19:20 +0200
                Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes John Stultz <john.stultz@linaro.org> - 2016-07-14 19:40 +0200
                Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes Oleg Nesterov <oleg@redhat.com> - 2016-07-14 19:50 +0200
                Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes John Stultz <john.stultz@linaro.org> - 2016-07-14 20:00 +0200
                Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes Oleg Nesterov <oleg@redhat.com> - 2016-07-14 20:10 +0200
                Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes Peter Zijlstra <peterz@infradead.org> - 2016-07-14 20:40 +0200
                Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes Peter Zijlstra <peterz@infradead.org> - 2016-07-14 21:40 +0200
            Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes John Stultz <john.stultz@linaro.org> - 2016-07-13 23:00 +0200
          Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes Peter Zijlstra <peterz@infradead.org> - 2016-07-13 23:00 +0200
            Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes "Paul E. McKenney" <paulmck@linux.vnet.ibm.com> - 2016-07-13 23:10 +0200
          Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes "Paul E. McKenney" <paulmck@linux.vnet.ibm.com> - 2016-07-13 23:00 +0200
            Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes Dmitry Shmidt <dimitrysh@google.com> - 2016-07-13 23:10 +0200
              Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes "Paul E. McKenney" <paulmck@linux.vnet.ibm.com> - 2016-07-13 23:10 +0200
              Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes John Stultz <john.stultz@linaro.org> - 2016-07-13 23:10 +0200
      Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes John Stultz <john.stultz@linaro.org> - 2016-07-13 22:20 +0200
      Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes Dmitry Shmidt <dimitrysh@google.com> - 2016-07-13 22:40 +0200
    Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes Colin Cross <ccross@google.com> - 2016-07-13 22:50 +0200
      Re: Severe performance regression w/ 4.4+ on Android due to cgroup  locking changes Tejun Heo <tj@kernel.org> - 2016-07-13 23:00 +0200

csiph-web