Hello, Hugh. On Wed, Mar 29, 2023 at 01:38:48PM -0700, Hugh Dickins wrote: > > So, in general, there's a trade off between local irq service latency and > > inducing global lock contention when using unprotected locks. With more and > > more CPUs, the balance keeps shifting. The balance still very much depends > > on the specifics of a given lock but yeah I think it's something we need to > > be a lot more careful about now. > > And this looks a very plausible argument to me: I'll let it sink in. Another somewhat relevant change is that flipping irq on/off used to be relatively expensive on older x86 cpus. I forget all details about when and how much but they should be a lot cheaper now. No idea about !x86 cpus tho. > But I hadn't heard that the RT folks were clamouring for more irq disabling: > perhaps they partition their machines with more care, and are not devotees > of high CPU counts. I think RT folks care a lot more about raw IRQ disables. These shouldn't actually disable IRQs on RT kernels. Thanks. -- tejun
