On Mon, Apr 20, 2020 at 01:07:09PM -0700, Linus Torvalds wrote: > On Mon, Apr 20, 2020 at 12:29 PM Dan Williams <dan.j.williams@xxxxxxxxx> wrote: > > > > I didn't consider asynchronous to be > > better because that means there is a gap between when the data > > corruption is detected and when it might escape the system that some > > external agent could trust the result and start acting on before the > > asynchronous signal is delivered. > > The thing is, absolutely nobody cares whether you start acting on the > wrong data or not. I think they do. If the result of the wrong data has already been sent out the network before you process the signal, then you will need far smarter application software than has ever been written to hunt it down and stop the spread of the bogus result. Stopping dead on the instruction before it consumes the data means you can "recover" by killing just one process, or just one VMM guest. I'm in total agreement the machine check (especially broadcast) was a bad choice for how to "stop on a dime". But I can't see how you could possibly decide what to do if you let thousands of instructions retire based on a bad data value before you even know that it happened. -Tony
