[Priority Inheritance] SCHED_FIFO or SCHED_RR?

[Priority Inheritance] SCHED_FIFO or SCHED_RR?

[Tadeus (Eus) Prastowo]

Hi!

When a SCHED_RR task with RT priority X blocks on a
priority-inheriting mutex held by a SCHED_NORMAL task with nice value
Y, RT-Mutex code will change the SCHED_NORMAL task priority to that of
the SCHED_RR task and based on the new priority, will change the
scheduling class too. But, each task retains its own scheduling
policy.

This means that the task inheriting the SCHED_RR priority will be
scheduled using SCHED_FIFO policy because task_tick_rt does not
enforce the SCHED_RR time slice when the task policy is not SCHED_RR.

Why should a SCHED_NORMAL task inheriting the priority of a SCHED_RR
task get the privilege of SCHED_FIFO task for running as long as it
wishes even when the task giving the inheritance does not have such
privilege?

Thank you very much.

Please kindly put me in the CC since I am not subscribed to the list.

--
Eus

Re: [Priority Inheritance] SCHED_FIFO or SCHED_RR?

[Tadeus (Eus) Prastowo]

Ping :)

On Wed, Sep 28, 2011 at 1:06 AM, Tadeus (Eus) Prastowo
<0x657573@googlemail.com> wrote:
> Hi!
>
> When a SCHED_RR task with RT priority X blocks on a
> priority-inheriting mutex held by a SCHED_NORMAL task with nice value
> Y, RT-Mutex code will change the SCHED_NORMAL task priority to that of
> the SCHED_RR task and based on the new priority, will change the
> scheduling class too. But, each task retains its own scheduling
> policy.
>
> This means that the task inheriting the SCHED_RR priority will be
> scheduled using SCHED_FIFO policy because task_tick_rt does not
> enforce the SCHED_RR time slice when the task policy is not SCHED_RR.
>
> Why should a SCHED_NORMAL task inheriting the priority of a SCHED_RR
> task get the privilege of SCHED_FIFO task for running as long as it
> wishes even when the task giving the inheritance does not have such
> privilege?
>
> Thank you very much.
>
> Please kindly put me in the CC since I am not subscribed to the list.
>
> --
> Eus

Re: [Priority Inheritance] SCHED_FIFO or SCHED_RR?

[Tadeus (Eus) Prastowo]

Anyone familiar with the Linux scheduler, please?

On Wed, Sep 28, 2011 at 1:06 AM, Tadeus (Eus) Prastowo
<0x657573@googlemail.com> wrote:
> Hi!
>
> When a SCHED_RR task with RT priority X blocks on a
> priority-inheriting mutex held by a SCHED_NORMAL task with nice value
> Y, RT-Mutex code will change the SCHED_NORMAL task priority to that of
> the SCHED_RR task and based on the new priority, will change the
> scheduling class too. But, each task retains its own scheduling
> policy.
>
> This means that the task inheriting the SCHED_RR priority will be
> scheduled using SCHED_FIFO policy because task_tick_rt does not
> enforce the SCHED_RR time slice when the task policy is not SCHED_RR.
>
> Why should a SCHED_NORMAL task inheriting the priority of a SCHED_RR
> task get the privilege of SCHED_FIFO task for running as long as it
> wishes even when the task giving the inheritance does not have such
> privilege?
>
> Thank you very much.
>
> Please kindly put me in the CC since I am not subscribed to the list.
>
> --
> Eus

Re: [Priority Inheritance] SCHED_FIFO or SCHED_RR?

[Mike Galbraith]

On Fri, 2011-09-30 at 21:12 +0200, Tadeus (Eus) Prastowo wrote:
> Anyone familiar with the Linux scheduler, please?
> 
> On Wed, Sep 28, 2011 at 1:06 AM, Tadeus (Eus) Prastowo
> <0x657573@googlemail.com> wrote:
> > Hi!
> >
> > When a SCHED_RR task with RT priority X blocks on a
> > priority-inheriting mutex held by a SCHED_NORMAL task with nice value
> > Y, RT-Mutex code will change the SCHED_NORMAL task priority to that of
> > the SCHED_RR task and based on the new priority, will change the
> > scheduling class too. But, each task retains its own scheduling
> > policy.
> >
> > This means that the task inheriting the SCHED_RR priority will be
> > scheduled using SCHED_FIFO policy because task_tick_rt does not
> > enforce the SCHED_RR time slice when the task policy is not SCHED_RR.
> >
> > Why should a SCHED_NORMAL task inheriting the priority of a SCHED_RR
> > task get the privilege of SCHED_FIFO task for running as long as it
> > wishes even when the task giving the inheritance does not have such
> > privilege?

Existence of a critical section > slice would be the real problem, no?

	-Mike

Re: [Priority Inheritance] SCHED_FIFO or SCHED_RR?

[Tadeus (Eus) Prastowo]

On Sat, Oct 1, 2011 at 7:41 AM, Mike Galbraith <efault@gmx.de> wrote:
> On Fri, 2011-09-30 at 21:12 +0200, Tadeus (Eus) Prastowo wrote:
>> > Why should a SCHED_NORMAL task inheriting the priority of a SCHED_RR
>> > task get the privilege of SCHED_FIFO task for running as long as it
>> > wishes even when the task giving the inheritance does not have such
>> > privilege?
>
> Existence of a critical section > slice would be the real problem, no?

No, it's not a problem because if a real-time system designer puts a
real-time task under SCHED_RR, he will calculate the length of the
critical section by taking into account the slice.

For the case of one SCHED_RR task sharing a critical section with one
SCHED_NORMAL task, the worst-case blocking time of the SCHED_RR task
is found by multiplying the critical section execution time taking
into account the slice by two.

>        -Mike

-- 
Eus

Re: [Priority Inheritance] SCHED_FIFO or SCHED_RR?

[Tadeus (Eus) Prastowo]

On Sat, Oct 1, 2011 at 10:29 PM, Tadeus (Eus) Prastowo
<0x657573@googlemail.com> wrote:
> For the case of one SCHED_RR task sharing a critical section with one
> SCHED_NORMAL task, the worst-case blocking time of the SCHED_RR task
> is found by multiplying the critical section execution time taking
> into account the slice by two.

Sorry, the worst-case blocking time is just the critical section
execution time taking into account the slice.

>>        -Mike
>
> --
> Eus

Re: [Priority Inheritance] SCHED_FIFO or SCHED_RR?

[Mike Galbraith]

On Sat, 2011-10-01 at 22:29 +0200, Tadeus (Eus) Prastowo wrote:
> On Sat, Oct 1, 2011 at 7:41 AM, Mike Galbraith <efault@gmx.de> wrote:
> > On Fri, 2011-09-30 at 21:12 +0200, Tadeus (Eus) Prastowo wrote:
> >> > Why should a SCHED_NORMAL task inheriting the priority of a SCHED_RR
> >> > task get the privilege of SCHED_FIFO task for running as long as it
> >> > wishes even when the task giving the inheritance does not have such
> >> > privilege?
> >
> > Existence of a critical section > slice would be the real problem, no?
> 
> No, it's not a problem because if a real-time system designer puts a
> real-time task under SCHED_RR, he will calculate the length of the
> critical section by taking into account the slice.

Submit a patch, and see if it flies.

	-Mike

Re: [Priority Inheritance] SCHED_FIFO or SCHED_RR?

[Tadeus (Eus) Prastowo]

On Sun, Oct 2, 2011 at 9:50 AM, Mike Galbraith <efault@gmx.de> wrote:
> On Sat, 2011-10-01 at 22:29 +0200, Tadeus (Eus) Prastowo wrote:
>> On Sat, Oct 1, 2011 at 7:41 AM, Mike Galbraith <efault@gmx.de> wrote:
>> > On Fri, 2011-09-30 at 21:12 +0200, Tadeus (Eus) Prastowo wrote:
>> >> > Why should a SCHED_NORMAL task inheriting the priority of a SCHED_RR
>> >> > task get the privilege of SCHED_FIFO task for running as long as it
>> >> > wishes even when the task giving the inheritance does not have such
>> >> > privilege?
>> >
>> > Existence of a critical section > slice would be the real problem, no?
>>
>> No, it's not a problem because if a real-time system designer puts a
>> real-time task under SCHED_RR, he will calculate the length of the
>> critical section by taking into account the slice.
>
> Submit a patch, and see if it flies.

Okay, thank you for the direction. It seems that only with a patch I
can get the answer ;-)

>        -Mike

--
Eus