On Mon, 2023-10-02 at 16:16 +0200, Sebastian Andrzej Siewior wrote: > On 2023-10-02 13:58:57 [+0200], Mike Galbraith wrote: > > > Why not perform all wakes from hardirq then? > > > > Sounds good to me iff we're talking about a dinky irq width delta. > > > > Threads bundling up what are otherwise irq context cycles is loaded > > with goodness, but static priority leaves you holding a bill and paying > > context switch fees on top. Pick your poison carefully applies I > > suppose. A tiny swig of hemlock can't do _too_ much harm, right ;-) > > A swarm of wake-ups for SCHED_OTHER at prio 50+ will delay your threaded > interrupt handler and everything else that runs 50 and lower. > Since SCHED_OTHER is less important than any RT task, the delay is > usually of lower concern. Sure, but you also have to weigh when to stop caring about inversion. I'd personally prefer an ever so slightly wider IRQ, but yeah, there is no free lunch, it's a trade. (wrt SCHED_OTHER wakeup swarm, in an environment where it's a concern, those perturbations are likely the least of your perturbation worries) -Mike
