Re: pthread_mutex_unlock (was Re: sched_yield() makes OpenLDAP slow)

[Howard Chu <hyc@symas.com>]

Lee Revell wrote:
> On Wed, 2006-01-25 at 16:49 -0800, Howard Chu wrote:
>   
>> Basic "fairness" isn't the issue. Fairness is concerned with which of 
>> *multiple waiting threads* gets the mutex, and that is certainly 
>> irrelevant here. The issue is that the releasing thread should not be
>> a candidate.
>>
>>     
>
> You seem to be making 2 controversial assertions:
>
> 1. pthread_mutex_unlock must cause an immediate reschedule if other
> threads are blocked on the mutex, and 
> 2. if the unlocking thread immediately tries to relock the mutex,
> another thread must get it first
>
> I disagree with #1, which makes #2 irrelevant.  It would lead to
> obviously incorrect behavior, pthread_mutex_unlock would no longer be an
> RT safe operation for example.
>   

Actually no, I see that #1 is unnecessary, and already acknowledged as such
http://groups.google.com/group/fa.linux.kernel/msg/89da66017d53d496

But #2 still holds.

> Also consider a SCHED_FIFO policy - static priorities and the scheduler
> always runs the highest priority runnable thread - under your
> interpretation of POSIX a high priority thread unlocking a mutex would
> require the scheduler to run a lower priority thread which violates
> SCHED_FIFO semantics

See the Mutex Initialization Scheduling Attributes section which 
specifically addresses priority inversion:
http://www.opengroup.org/onlinepubs/000095399/xrat/xsh_chap02.html#tag_03_02_09

If point #2 were not true, then there would be no need to bother with 
any of that. Instead that text ends with "it is important that 
IEEE Std 1003.1-2001 provide these interfaces for those cases in which 
it is necessary."

-- 
  -- Howard Chu
  Chief Architect, Symas Corp.  http://www.symas.com
  Director, Highland Sun        http://highlandsun.com/hyc
  OpenLDAP Core Team            http://www.openldap.org/project/