> On May 13, 2021, at 10:57 AM, Visa Hankala <v...@hankala.org> wrote: > > On Thu, May 13, 2021 at 12:04:57AM -0500, Scott Cheloha wrote: >> The funky locking dance in softclock() and softclock_thread() is >> needed to keep from violating the locking hierarchy. The >> timeout_mutex is above the kernel lock, so we need to leave the >> timeout_mutex, then drop the kernel lock, and then reenter the >> timeout_mutex to start running TIMEOUT_MPSAFE timeouts. > > Are you sure that dropping the kernel lock in softclock(), in > non-process context, is a good idea? That makes assumptions how the > layers above work.
I figured it was the most obvious way forward. When you say "layers above," do you mean softintr_dispatch()? Or something else? > At the minimum, I think the soft interrupt code has to be changed so > that it is possible to register MP-safe handlers. Some platforms (e.g. arm64) have softintr_establish_flags(), but this is not universally available. Do I need to unlock all the softintr internals before proceeding with this? Or do I just need to bring softrintr_establish_flags() to every platform without it?
