RTEMS POSIX Priority Inversion Issue.
Joel Sherrill
joel at rtems.org
Sat Dec 29 21:31:58 UTC 2018
Hi
I couldn't resist looking into this a bit more to confirm my suspicion. The
code is in
cpukit/libmisc/monitor. All tasks and threads are dumped by the same method
and
priority is handled exactly the same independent of the API. For sure, the
SuperCore
priority is always being printed.
I think it would be simple enough to call rtems_object_id_get_api() and see
if it was
a POSIX thread before calling the helper to print priority.and then call
_POSIX_Priority_From_core() to print it per the API view. There is code in
pthreadgetschedparam.c to provide the POSIX view from the SuperCore
settings.
There are probably other ways to fix this but it is a bug in the task
dumping command
and it already knows a lot about the internals so this isn't a horrible
solution.
You aren't seeing priority inversion for sure though. So you can sleep
easier on that one.
--joel
On Fri, Dec 28, 2018 at 11:24 PM JunBeom Kim (EmbedCoreTech) <
jbkim at e-coretech.kr> wrote:
> Dear Joel,
>
>
>
> Thank you for your advice.
>
>
>
> I will check this from RTEMS 4.12 and RTEMS 5.0.
>
>
>
> After I check this, I will share update information.
>
>
>
> Have a good weekend and Happy new year.
>
>
>
> Best Regards,
>
> JunBeom
>
>
>
> *From:* Joel Sherrill <joel at rtems.org>
> *Sent:* Saturday, December 29, 2018 1:46 AM
> *To:* JunBeom Kim (EmbedCoreTech) <jbkim at e-coretech.kr>
> *Cc:* 구철회 <chkoo at kari.re.kr>; rtems-users at rtems.org <users at rtems.org>
> *Subject:* Re: RTEMS POSIX Priority Inversion Issue.
>
>
>
> Hi
>
>
>
> I'm off for the holidays and answering from my phone. My suspicion is that
> the pthread command is an alias for the tasks command. Thus it is not
> reporting the priority from the view of the POSIX API but from the view of
> the Classic API. Internally all threads and tasks are instances of
> SuperCore threads and the priorities are mapped to what is required by the
> API.
>
>
>
> Priority 1 in Classic API is 254 in POSIX.
>
>
>
> Please investigate the code for the pthread command and see if I am right.
> This would be a minor bug in the shell rather than a priority inversion
> issue.
>
>
>
> --joel
>
>
>
> On Fri, Dec 28, 2018, 1:21 AM JunBeom Kim (EmbedCoreTech) <
> jbkim at e-coretech.kr wrote:
>
> Dear Mr. Koo.
>
>
>
> There is not any relationship between main thread and pthread1/2/3.
>
> Main thread is run on core 1 of Cortex-A9.
>
> Pthread1/2/3 threads is run on core 2 of Cortex-A9.
>
> That is, both main thread and pthread1/2/3 is run on each cores
> independently.
>
> Also, I used manual affinity using pthread_setaffinity_np().
>
>
>
> Pthread1/2/3 is created by main thread.
>
> and, actual operation is using Qt Thread API. I attached simple example
> code.
>
> Qt Thread API is calling RTEMS POSIX API.
>
>
>
> Best Regards,
>
> JunBeom
>
>
>
> *From:* chkoo at kari.re.kr <chkoo at kari.re.kr>
> *Sent:* Friday, December 28, 2018 4:05 PM
> *To:* JunBeom Kim (EmbedCoreTech) <jbkim at e-coretech.kr>
> *Cc:* users at rtems.org
> *Subject:* RE: RTEMS POSIX Priority Inversion Issue.
>
>
>
> What is the relation between main thread and pthread1/2/3?
>
> If the main thread has a control to wake up each pthread1/2/3, it gonna be
> in trouble when the prioirty of the main thread is lower than
> pthread1/2/3s because the main thread has never chance to run it, which is
> required to other threads.
>
>
>
> However why do you assign so much long integer value to priority?
>
>
>
> Cheol
>
> *--------- **원본 메일 ---------*
>
> *보낸사람* : "JunBeom Kim (EmbedCoreTech)" <jbkim at e-coretech.kr>
> *받는사람* : <users at rtems.org>
> *받은날짜* : 2018-12-28 (금) 15:45:48
> *제목* : RTEMS POSIX Priority Inversion Issue.
>
> Dear Sir,
>
>
>
> I am testing RTEMS POSIX thread priority.
>
> I am still using RTEMS 4.12 in 2017 year.
>
>
>
> There are four pthreads(main, pthread1, pthread2, pthread3) on two cores
> in my test case.
>
> - main pthread is running on core 1.
>
> - pthread1, pthread2, pthread3 is running on core 2.
>
> - All threads are doing GPIO toggling action in while(1) loop.
>
> - my configuration for timeslice quantum is 25ms.
>
>
>
> As I know from RTEMS POSIX manual, there is below statement.
>
>
>
> In the RTEMS implementation of the POSIX API, the priorities range from
> the low priority of
>
> sched_get_priority_min() to the highest priority of
> sched_get_priority_max(). Numerically
>
> higher values represent higher priorities.
>
>
>
> But, when I test this, lower value is doing as higher priority. Higher
> value is doing as lower priority.
>
> Is this correct operation ?
>
>
>
> Case 0) All threads are same priority.
>
> [/] # pthread
>
> ID NAME CPU SHED PRI STATE MODES EVENTS WAITINFO
>
> ------------------------------------------------------------------------------
>
>
> 0b010001 main_thread 1 MEDF 2147483645 READY P:T:nA NONE
>
> 0b010002 pthread1 2 MEDF 2147483645 READY P:T:nA NONE
>
> 0b010003 pthread2 2 MEDF 2147483645 READY P:T:nA NONE
>
> 0b010004 pthread3 2 MEDF 2147483645 READY P:T:nA NONE
>
>
>
> Case 1) pthread3 is lower priority according to manual. But, pthread3 is
> higher priority in my case.
>
> [/] # pthread
>
> ID NAME CPU SHED PRI STATE MODES EVENTS WAITINFO
>
> ------------------------------------------------------------------------------
>
>
> 0b010001 main_thread 1 MEDF 2147483645 READY P:T:nA NONE
>
> 0b010002 pthread1 2 MEDF 2147483646 READY P:T:nA NONE
>
> 0b010003 pthread2 2 MEDF 2147483646 READY P:T:nA NONE
>
> 0b010004 pthread3 2 MEDF 1789569706 READY P:T:nA NONE
>
>
>
> Case 2) pthread3 is higher priority according to manual. pthread1 and
> pthread2 are higher priority than pthread3.
>
> [/] # pthread
>
> ID NAME CPU SHED PRI STATE MODES EVENTS WAITINFO
>
> ------------------------------------------------------------------------------
>
>
> 0b010001 main_thread 1 MEDF 2147483645 READY P:T:nA NONE
>
> 0b010002 pthread1 2 MEDF 1789569706 READY P:T:nA NONE
>
> 0b010003 pthread2 2 MEDF 1789569706 READY P:T:nA NONE
>
> 0b010004 pthread3 2 MEDF 2147483646 READY P:T:nA NONE
>
>
>
> Please advise.
>
>
>
> Best Regards,
>
> JunBeom Kim
>
> ~~~~~~~~~~~~~~~~~~~~~~
>
> President / EmbedCoreTech
>
> Phone: +82-31-396-5584
>
> Fax: +82-504-065-5720
>
> Mobile:+82-10-6425-5720
>
> Email: jbkim at e-coretech.kr
>
> Web: www.e-coretech.kr
>
> ~~~~~~~~~~~~~~~~~~~~~~
>
>
>
>
>
> _______________________________________________
> users mailing list
> users at rtems.org
> http://lists.rtems.org/mailman/listinfo/users
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.rtems.org/pipermail/users/attachments/20181229/41e6f789/attachment-0001.html>
More information about the users
mailing list