On Thu, Apr 28, 2022 at 4:15 AM hasegawa-hitomi@xxxxxxxxxxx <hasegawa-hitomi@xxxxxxxxxxx> wrote: > > Hi Greg, Arnd, and Daniel, > > > I understand that it is not appropriate to hardcode c. > > How about using __setup() to add a new kernel parameter and allow the admin > > to specify the sysrq command when booting? > > I have received a lot of advice regarding sysrq, but after some consideration, > I would like to change to calling panic() directly as in v1 instead of sysrq. > > If the administrator wants to request a diagnostic, I think they usually > expect crash with NMI like x86 and take a dump the kernel. It's not common > to handle diagnostic interrupts with sysrq now, so I don't think > it's necessary to make this driver extensible at this time. Ok, fair enough. Matching x86 behavior sounds like a reasonable outcome. If we want to make this configurable in the future, that can still be done then, and it should work the same across architectures but adding the logic in nmi_panic() directly. Arnd