Messages in this thread | | | Date | Fri, 9 Oct 2020 13:39:09 +0200 | From | Peter Zijlstra <> | Subject | Re: [RFC PATCH] kernel: allow to configure PREEMPT_NONE, PREEMPT_VOLUNTARY on kernel command line |
| |
On Fri, Oct 09, 2020 at 12:48:08PM +0200, Michal Hocko wrote: > On Fri 09-10-20 12:20:09, Peter Zijlstra wrote: > > On Fri, Oct 09, 2020 at 12:14:05PM +0200, Michal Hocko wrote: > > > On Fri 09-10-20 11:47:41, Peter Zijlstra wrote: > > > > > > That is, work backwards (from PREEMPT back to VOLUNTARY) instead of the > > > > other way around. > > > > > > My original idea was that the config would only define the default > > > preemption mode. preempt_none parameter would then just act as an > > > override. That would mean that CONFIG_PREEMPTION would be effectively > > > gone from the kernel. The reason being that any code outside of the > > > scheduler shouldn't really care about the preemption mode. I suspect > > > this will prevent from dubious hacks and provide a more robust code in > > > the end. > > > > Sure; but the way of arriving at that destination might be easier if > > you work backwards from PREEMPT=y, because while there _should_ not be > > dependencies outside of the scheduler, we both know there are. > > Wouldn't we need to examine each of the CONFIG_PREEMPTION code anyway? > And wouldn't that be even more tricky? The boot time option would result > in a more restrictive preemption mode while the code is actually > assuming a less restrictive one.
Sure, in the end we'll have to look at all of that.
> > This also makes your patches independent of the series that makes > > CONFIG_PREEMPTION unconditional. > > > > It also gives Kconfig space to limit the dynamic thing to archs that > > have sufficient support (we'll be relying on static_call/static_branch, > > and not everybody has that implemented in a way that makes it the > > dynamic change worth-while). > > Hmm, this is actually a good argument. I can imagine that kernels > without CONFIG_JUMP_LABEL might increase a runtime overhead for > something that users of that kernel might be not really interested in. > This would make CONFIG_PREEMPT_DYNAMIC be selected by CONFIG_JUMP_LABEL. > > I will add the CONFIG_PREEMPT_DYNAMIC in the next version. I just have > to think whether flipping the direction is really safe and easier in the > end. For our particular usecase we are more interested in > NONE<->VOLUNTARY at this moment and having full preemption in the mix > later is just fine. If you insist on the other direction then we can > work on that.
Ah, I was purely thinking of the FULL preempt case. For the NONE/VOLATILE case you can probably keep it simpler.
| |