On Fri 2015-11-27 17:26:16, Max Filippov wrote: Hi Max, > > Another exception is Xtensa architecture that uses just a > > fake NMI. > > It's called fake because it's actually maskable, but sometimes > it is safe to use it as NMI (when there are no other IRQs at the > same priority level and that level equals EXCM level). That > condition is checked in arch/xtensa/include/asm/processor.h > So 'fake' here is to avoid confusion with real NMI that exists > on xtensa (and is not currently used in linux), otherwise code > that runs in fake NMI must follow the NMI rules. > > To make xtensa compatible with your change we can add a > choice whether fake NMI should be used to kconfig. It can > then set HAVE_NMI accordingly. I'll post a patch for xtensa. Thanks a lot for explanation. I'll wait for the destiny of the patch adding CONFIG_XTENSA_FAKE_NMI. It is not easy for me to review. Anyway, we could select HAVE_NMI for Xtensa anytime later if this patchset goes in earlier. Best Regards, Petr
