lkml.org 
[lkml]   [2016]   [Sep]   [12]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
From
Date
SubjectRe: [PATCH] android: binder: Disable preemption while holding the global binder lock
Thanks for the reviews. We'll come up with a different solution.

On Sat, Sep 10, 2016 at 10:28 AM, Greg Kroah-Hartman
<gregkh@linuxfoundation.org> wrote:
> On Sat, Sep 10, 2016 at 06:37:29PM +0200, Thomas Gleixner wrote:
>> On Sat, 10 Sep 2016, Peter Zijlstra wrote:
>>
>> > On Sat, Sep 10, 2016 at 09:16:59AM -0700, Christoph Hellwig wrote:
>> > > On Thu, Sep 08, 2016 at 09:12:50AM -0700, Todd Kjos wrote:
>> > > > In Android systems, the display pipeline relies on low
>> > > > latency binder transactions and is therefore sensitive to
>> > > > delays caused by contention for the global binder lock.
>> > > > Jank is siginificantly reduced by disabling preemption
>> > > > while the global binder lock is held.
>> > >
>> > > That's now how preempt_disable is supposed to use. It is for critical
>> >
>> > not, that's supposed to be _not_. Just to be absolutely clear, this is
>> > NOT how you're supposed to use preempt_disable().
>> >
>> > > sections that use per-cpu or similar resources.
>> > >
>> > > >
>> > > > Originally-from: Riley Andrews <riandrews@google.com>
>> > > > Signed-off-by: Todd Kjos <tkjos@google.com>
>> >
>> > > > @@ -389,7 +390,11 @@ static int task_get_unused_fd_flags(struct
>> > > > binder_proc *proc, int flags)
>> > > > rlim_cur = task_rlimit(proc->tsk, RLIMIT_NOFILE);
>> > > > unlock_task_sighand(proc->tsk, &irqs);
>> > > >
>> > > > - return __alloc_fd(files, 0, rlim_cur, flags);
>> > > > + preempt_enable_no_resched();
>> > > > + ret = __alloc_fd(files, 0, rlim_cur, flags);
>> > > > + preempt_disable();
>> >
>> > And the fact that people want to use preempt_enable_no_resched() shows
>> > that they're absolutely clueless.
>> >
>> > This is so broken its not funny.
>> >
>> > NAK NAK NAK
>>
>> Indeed. Sprinkling random preempt_enabe/disable() pairs all over the place
>> documents clearly that this is tinkering and not proper software
>> engineering.
>
> I have pointed out in the other thread for this patch (the one that had
> a patch that could be applied) that the single lock in the binder code
> is the main problem here, it should be solved instead of this messing
> around with priorities.
>
> So don't worry, I'm not taking this change :)
>
> thanks,
>
> greg k-h

\
 
 \ /
  Last update: 2016-09-17 09:59    [W:0.196 / U:0.272 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site