On Thu, Dec 19, 2019 at 10:23:19AM +0100, Peter Zijlstra wrote: > On Thu, Dec 19, 2019 at 09:59:48AM +0800, Ming Lei wrote: > > > So pray tell, why did you not integrate this with IRQ_TIME_ACCOUNTING ? > > > That already takes a timestamp and does most of what you need. > > > > Yeah, that was the 1st approach I thought of, but IRQ_TIME_ACCOUNTING > > may be disabled, and enabling it may cause observable effect on IO > > performance. > > Is that an actual concern, are people disabling it? For example, it is only enabled for x86 on RHEL8. And the interrupt flood issue is easier to trigger on other ARCH, for example, John reported the issue on arm64: https://lore.kernel.org/lkml/a7ef3810-31af-013a-6d18-ceb6154aa2ef@xxxxxxxxxx/ > > > > > @@ -356,6 +512,7 @@ void irq_enter(void) > > > > } > > > > > > > > __irq_enter(); > > > > + irq_interval_update(); > > > > } > > > > > > Arggh.. you're going to make every single interrupt take at least 2 > > > extra cache misses for this gunk?!? > > > > Could you explain it a bit why two cache misses are involved? > > > > I understand at most one miss is caused, which should only happen in > > irq_interval_update(), and what is the other one? > > The rq clock thing IIRC. OK. But task is often waken up by interrupt event, I guess the rq clock thing should be fine. Thanks, Ming
