Re: [Xenomai-core] Potential problem with rt_eepro100

[Gilles Chanteperdrix <gilles.chanteperdrix@xenomai.org>]

Jan Kiszka wrote:
> Am 04.11.2010 23:06, Gilles Chanteperdrix wrote:
>> Jan Kiszka wrote:
>>>>> At first sight, here you are more breaking things than cleaning them.
>>>> Still, it has the SMP record for my test program, still runs with ftrace 
>>>> on (after 2 hours, where it previously failed after maximum 23 minutes).
>>> My version was indeed still buggy, I'm reworking it ATM.
>>>
>>>> If I get the gist of Jan's changes, they are (using the IPI to transfer 
>>>> one bit of information: your cpu needs to reschedule):
>>>>
>>>> xnsched_set_resched:
>>>> -      setbits((__sched__)->status, XNRESCHED);
>>>>
>>>> xnpod_schedule_handler:
>>>> +	xnsched_set_resched(sched);
>>>> 	
>>>> If you (we?) decide to keep the debug checks, under what circumstances 
>>>> would the current check trigger (in laymans language, that I'll be able 
>>>> to understand)?
>>> That's actually what /me is wondering as well. I do not see yet how you
>>> can reliably detect a missed reschedule reliably (that was the purpose
>>> of the debug check) given the racy nature between signaling resched and
>>> processing the resched hints.
>> The purpose of the debugging change is to detect a change of the
>> scheduler state which was not followed by setting the XNRESCHED bit.
> 
> But that is nucleus business, nothing skins can screw up (as long as
> they do not misuse APIs).

Yes, but it happens that we modify the nucleus from time to time.

> 
>> Getting it to work is relatively simple: we add a "scheduler change set
>> remotely" bit to the sched structure which is NOT in the status bit, set
>> this bit when changing a remote sched (under nklock). In the debug check
>> code, if the scheduler state changed, and the XNRESCHED bit is not set,
>> only consider this a but if this new bit is not set. All this is
>> compiled out if the debug is not enabled.
> 
> I still see no benefit in this check. Where to you want to place the bit
> set? Aren't that just the same locations where
> xnsched_set_[self_]resched already is today?

Well no, that would be another bit in the sched structure which would
allow us to manipulate the status bits from the local cpu. That
supplementary bit would only be changed from a distant CPU, and serve to
detect the race which causes the false positive. The resched bits are
set on the local cpu to get xnpod_schedule to trigger a rescheduling on
the distance cpu. That bit would be set on the remote cpu's sched. Only
when debugging is enabled.

> 
> But maybe you can provide some motivating bug scenarios, real ones of
> the past or realistic ones of the future.

Of course. The bug is anything which changes the scheduler state but
does not set the XNRESCHED bit. This happened when we started the SMP
port. New scheduling policies would be good candidates for a revival of
this bug.

-- 
                                                                Gilles.