On Tue, Sep 18, 2018 at 14:04:35 -0700, Richard Henderson wrote: > On 9/17/18 9:30 AM, Emilio G. Cota wrote: > > cpu_reset(cs); > > - cs->interrupt_request = sipi; > > + atomic_mb_set(&cs->interrupt_request, sipi); > > memcpy(&env->start_init_save, &save->start_init_save, > > Why does this need a memory barrier? > > Anyway, I think a bare mechanical conversion would be best > for the first patch and then extra barriers added separately > and with a description of why. Almost no corresponding read has a barrier so it's hard to justify this one. I'll drop it. Thanks, Emilio
