On 6 December 2012 17:31, Liviu Dudau <liviu.du...@arm.com> wrote: > [discussion not relevant to this reply removed] >> >> Liviu, >> >> I'm just saying that the use case above can't happen because the >> fields are not updated by thread but by scheduler and it can't be >> pre-empted by a thread while updating them. We can have the situation >> where cpu A updates its fields and cpu B reads them simultaneously. >> This is the use case that i'm trying to address with this patch. > > And me and Morten are saying that the compiler and/or system can reorder the > reads > on CPU B and you have
That's a new use case ;-) and the result of the compilation had shown that it was not reordered. But I agree that the sequence need to be more deeply studied to ensure that all use case are solved and that's why I had not pushed any improvement/modification yet. > > CPU A (running scheduler) CPU B (running is_buddy_busy) > update rq->avg.runnable_avg_sum period = rq->avg.runnable_avg_period > new_period = > rq->avg.runnable_avg_period > update rq->avg.runnable_avg_period sum = rq->avg.runnable_avg_sum > new_sum = rq->avg.runnable_avg_sum > > while () condition is met and you break out of the loop, using stale value for > one of the variables that you then use to decide if the buddy is busy. > > Regards, > Liviu > > >> >> Regards, >> Vincent >> > > [rest of the thread discussion removed for clarity] > > -- > ==================== > | I would like to | > | fix the world, | > | but they're not | > | giving me the | > \ source code! / > --------------- > ¯\_(ツ)_/¯ > _______________________________________________ linaro-dev mailing list linaro-dev@lists.linaro.org http://lists.linaro.org/mailman/listinfo/linaro-dev
