On 25/01/2017 12:33, Pavel Dovgalyuk wrote: >> From: Paolo Bonzini [mailto:[email protected]] >> On 25/01/2017 12:12, Pavel Dovgalyuk wrote: >>>> From: Paolo Bonzini [mailto:[email protected]] >>>> On 24/01/2017 08:17, Pavel Dovgalyuk wrote: >>>>> @@ -451,6 +451,10 @@ static inline bool cpu_handle_exception(CPUState >>>>> *cpu, int *ret) >>>>> #ifndef CONFIG_USER_ONLY >>>>> } else if (replay_has_exception() >>>>> && cpu->icount_decr.u16.low + cpu->icount_extra == 0) { >>>>> + /* Break the execution loop in case of running out of TB cache. >>>>> + This is needed to make flushing of the TB cache, because >>>>> + real flush is queued to be executed outside the cpu loop. */ >>>>> + cpu->exception_index = EXCP_INTERRUPT; >>>>> /* try to cause an exception pending in the log */ >>>>> cpu_exec_nocache(cpu, 1, tb_find(cpu, NULL, 0), true); >>>>> *ret = -1; >>>> >>>> Why is replay_has_exception() related to be running out of TB cache? >>> >>> It doesn't. >>> Calling tb_find when there is not space in cache causes tb_flush and >>> cpu_loop_exit. >>> But execution loop will continue, because there is no reason to break it >>> (like setting exception_index). >> >> What about setting cpu->exit_request? queue_work_on_cpu calls >> qemu_cpu_kick. > > cpu->exit_request does not checked in this loop. > We have to add this checking somewhere then?
It's checked by cpu_handle_interrupt. Are you not reaching cpu_handle_interrupt then? Why? Or perhaps cpu_handle_interrupt should not be testing cpu->exit_request, but cpu->exception_index != -1 (and cpu_exit can cmpxchg cpu->exception_index from -1 to EXCP_INTERRUPT)? Again, it's hard to follow without knowing the invariants. :( Paolo
