On Mon, Jan 07, 2013 at 03:28:33PM +0530, Viresh Kumar wrote:
Firstly the root cause of this patchset.
Myself and some others in Linaro are working on ARM future cores: big.LITTLE systems. Here we have few very powerful, high power consuming cores (big, currently A15's) and few very efficient ones (LITTLE, currently A7's).
The ultimate goal is to save as much power as possible without compromising much with performance. For, that we wanted most of the stuff to run on LITTLE cores and some performance-demanding stuff on big Cores. There are multiple things going around in this direction. Now, we thought A15's or big cores shouldn't be used for running small tasks like timers/workqueues and hence this patch is an attempt towards reaching that goal.
I see. My understanding of big.little is very superficial so there are very good chances that I'm utterly wrong, but to me it seems like more of a stop-gap solution than something which will have staying power in the sense that the complexity doesn't seem to have any inherent reason other than the failure to make the big ones power efficient enough while idle or under minor load.
Maybe this isn't specific to big.little and heterogeneous cores will be the way of future with big.little being just a forefront of the trend. And even if that's not the case, it definitely doesn't mean that we can't accomodate big.little, but, at this point, it does make me a bit more reluctant to make wholesale changes specifically for big.little.
Over that we can do some load balancing of works within multiple alive cpus, so that it can get done quickly. Also, we shouldn't start using an idle cpu just for processing work :)
The latter part "not using idle cpu just for processing work" does apply to homogeneous systems too but as I wrote earlier work items don't spontaneously happen on an idle core. Something, including timer, needs to kick it. So, by definition, a CPU already isn't idle when a work item starts execution on it. What am I missing here?
I have another idea that we can try:
With this we would not break any existing code and can try to migrate old users to this new infrastructure (atleast the ones which are rearming works from their work_handlers). What do you say?
Yeah, this could be a better solution, I think. Plus, it's not like finding the optimal cpu is free.
To take care of the cache locality issue, we can pass an argument to this routine, that can provide
- the mask of cpus to schedule this work on OR
- Sched Level (SD_LEVEL) of cpus to run it.
Let's start simple for now. If we really need it, we can always add more later.