Paolo Bonzini <[email protected]> writes:
> On 17/03/2017 21:43, Alex Bennée wrote:
>> There is currently cpu->running which is set/cleared by
>> cpu_exec_start/end. Although if we crash between cpu_exec_start and
>> sigsetjmp the same sort of brokenness might happen.
>
> I think cpu_exec_start/end should be moved into cpu_exec itself (but
> probably just in 2.10).
Sure. Although hopefully we can resist the temptation to insert segging
code into that small window in the meantime ;-)
>
> Paolo
>
>> Anyway understood now. If anyone has any suggestions for neater stuff
>> over the weekend please shout, otherwise I'll probably just hack
>> handle_cpu_signal to do:
>>
>> cpu = current_cpu;
>> if (!cpu->running) {
>> /* we weren't running or translating JIT code when the signal came */
>> return 1;
>> }
--
Alex Bennée
