lkml.org 
[lkml]   [2010]   [Dec]   [15]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectRe: [PATCH] rtmutex: multiple candidate owners without unrelated boosting
From
Date
On Wed, 2010-12-15 at 09:02 +0100, Thomas Gleixner wrote:
> On Tue, 14 Dec 2010, Steven Rostedt wrote:
>
> > On Tue, 2010-12-14 at 17:04 +0800, Lai Jiangshan wrote:
> >
> > OK, I was looking at this in a bit more detail (the coffee finally set
> > in) and I was at first looking to nuke the cand_owner since it is
> > redundant to cand_seq. But I think we can nuke the cand_seq instead and
> > use the top_waiter as the decider.
>
> So you just use cand_owner (the name sucks) to flag that the waiter
> has been woken up either by the boost code or by an unlock. The waiter
> clears that flag with waiter->lock->wait_lock held before calling
> schedule().
>
> Though I think we do need it at all. wakeup of an already running task
> is almost a nop, so having one less state to worry about is good.

I was hoping to remove it completely, and yes I was hoping we could
because a wakeup of a woken task is almost a nop. But then I saw this in
rt_mutex_adjust_prio_chain():


> /*
> * Check the orig_waiter state. After we dropped the locks,
> * the previous owner of the lock might have released the lock
> - * and made us the pending owner:
> + * and made us candidate owner:
> */
> - if (orig_waiter && !orig_waiter->task)
> + if (orig_waiter && orig_waiter->cand_owner)
> goto out_unlock_pi;
>

I'm not sure what else we could use to check if the original waiter has
been given the lock.

-- Steve




\
 
 \ /
  Last update: 2010-12-15 15:05    [W:0.116 / U:1.132 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site