On 04/29/2014 12:19 PM, Viresh Kumar wrote: > On 29 April 2014 11:46, Srivatsa S. Bhat > <srivatsa.bhat@xxxxxxxxxxxxxxxxxx> wrote: >> Yes, I'm aware that this corner case doesn't work well with my debug > > Don't know if its a corner case, it may be the most obvious case for > some :) > Yeah, it could be. >> patch. I tried to avoid this but couldn't think of any solution. > > The problem is not that it wouldn't work for these systems, but we will > get WARN_ON() when they shouldn't have come :) > Yes, I thought about this, and I agree that this is not acceptable. >> (One big-hammer way to avoid this is to exclude this infrastructure >> for all ASYNC_NOTIFICATION drivers, but I didn't want to go with that >> approach, since it makes it look ugly). Do you have any better ideas >> to deal with this scenario? > > Can't think of anything better than this: > > + WARN_ON(!(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION) > && (current == policy->transition_task)); > > which you already mentioned. Yeah, I think we should just go with this. I thought we needed lots of if-conditions to do exclude these drivers (which would have made it ugly), but as you pointed above, just this one would suffice. Besides, the cpufreq core doesn't automatically invoke _begin() and _end() for ASYNC_NOTIFICATION drivers. So that means the probability that such drivers will hit this problem is extremely low, since the driver alone is responsible for invoking _begin/_end and hence there shouldn't be much of a conflict. So I think we should really just skip ASYNC_NOTIFICATION drivers in this debug infrastructure. > >> Also, do we really have cases in mind where a single thread does >> multiple frequency transitions in one go? That too in such quick >> successions? Echo's to sysfs, changing of governors from userspace etc >> all do one frequency transition at a time per-task... > > Its not really about if we can think of a real use case or not. The point > is, governor is free to call transition calls one after the other (will always > happen from a single thread) and it isn't supposed to wait for drivers > to finish earlier transitions as ->target() has already returned. > Yes, I agree now. Making bold assumptions in the cpufreq core about how many frequency transitions a single task will do etc is potentially *very* dangerous. Let's not do it that way. I'll send a v2 excluding the ASYNC_NOTIFICATION drivers. Thanks a lot for your inputs! Regards, Srivatsa S. Bhat -- To unsubscribe from this list: send the line "unsubscribe cpufreq" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
