Re: Question on Linux and SCHED_FIFO scheduling for POSIX threads

[Jakub Jelinek]

On Fri, Jul 16, 2004 at 11:11:54AM +0200, Eirik Nordbr?den wrote:
> Can anybody clarify how SCHED_FIFO scheduling and thread priorities works
> on Linux? We are novices in this field in the Linux environment and needs
> help to understand how it works. To verify the behaviour we made up a
> small test program consisting of four threads and a mutex. We have run the
> program on both the 2.6.5 and 2.6.7 kernels with same behaviour.
> 
> Program:
> 
> T-MAIN: scheduling policy=SCHED_FIFO, priority=1
> T-LP:   scheduling policy=SCHED_FIFO, priority=10
> T-MP:   scheduling policy=SCHED_FIFO, priority=20
> T-HP:   scheduling policy=SCHED_FIFO, priority=30
> 
> The program runs like this:
> 
> T-MAIN locks mutex => T-MAIN runs.
> T-MAIN creates T-LP => T-LP runs.
> T-LP waits for mutex => T-MAIN runs.
> T-MAIN creates T-MP => T-MP runs.
> T-MP waits for mutex => T-MAIN runs.
> T-MAIN creates T-HP => T-HP runs.
> T-HP waits for mutex => T-MAIN runs.
> T-MAIN waits 3 seconds and unlocks mutex => T-LP runs.
> T-LP waits 3 seconds and unlocks mutex => T-MP runs.
> T-MP waits 3 seconds and unlocks mutex => T-HP runs.
> :
> :
> 
> For us this is unexpected behaviour. We would expect that the thread with
> the highest priority would be scheduled to run when a number of threads is
> waiting for a mutex and the mutex is unlocked. Can anyone clarify this?
> Have we missed something?

NPTL locking is implemented on top of futex(2).
futex(2) queues are ATM FIFOs, not priority based queues.
Check http://developer.osdl.org/dev/robustmutexes/
for some patches which introduce priority based queues for futexes (well,
AFAIK they introduce new syscalls and call the primitive fusyn instead).

	Jakub
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/