lkml.org 
[lkml]   [2006]   [Jul]   [23]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
SubjectRe: [patch 3/3] [-rt] Fixes the timeout-bug in the rtmutex/PI-futex.
From


On Sun, 23 Jul 2006, Steven Rostedt wrote:

> On Sun, 2006-07-23 at 02:18 +0100, Esben Nielsen wrote:
>> This patch fixes
>>
>> 1) The timeout bug in rtmutexes and PI futexes: When a task is blocked on a
>> lock with timeout and times out it will not wakeup until the owner of the lock
>> is done. This is because the owner is boosted to the same priority as the
>> blocked task and therefore has the CPU such the blocked task never gets around
>> to de-boost it!
>>
>> 2) setscheduler() now does the PI walking - but defers the work to the blocked
>> task.
>>
>> 3) In general it makes sure that a task, which is boosting another has enough
>> priority to do the de-boosting no matter how complicated the lock structure is,
>> or how many times the priorities have changed.
>>
>> The idea behind the patch is simple:
>> If a task is boosting another it is scheduled in LIFO order and it will never
>> loose it's priority. This property lasts until it has left the lock operation
>> (successfully or not).
>>
>> The needed priority to do the unboosting is stored in task->boosting_prio.
>> In the current patch this is always increasing (numerically decreasing) while
>> trying to take a lock. In a future it might be found safe to decrease
>> boosting_prio before finally leaving the lock operation.
>>
>> include/linux/rtmutex.h | 1
>> include/linux/sched.h | 7
>> kernel/fork.c | 1
>> kernel/rtmutex.c | 151 +++++++++++++-----
>
> It is possible that these changes can break the pi code in rtmutex.c.

Hmm, I am quite confident this one doesn't :-) It is tested against my own
user-space unit tester and against Thomas's in-kernel rt-tester. And my
stress PriorityInheritanceTest by the way.

> I
> haven't analyzed it enough yet. But just so that you know that your
> changes don't break the code, and to make it easier for me to look at
> it. Please update Documentation/rt-mutex-design.txt including your
> changes. This will be a good exercise to see if it doesn't really break
> anything, and it will give other reviewers a better starting point for
> review.

Is that up-to-date in the -rt tree? The last patch you sent was to
2.6.18-rc2, right?

Esben

>
>> kernel/rtmutex_common.h | 1
>> kernel/sched.c | 14 +
>> scripts/rt-tester/t5-l4-pi-boost-deboost-setsched.tst | 42 +++--
>> scripts/rt-tester/t5-l4-pi-boost-deboost.tst | 12 -
>> 8 files changed, 167 insertions(+), 62 deletions(-)
>
> -- Steve
>
>
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2006-07-23 18:01    [W:0.061 / U:1.228 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site