> By "asynchronous" I don't mean "hours later". > > Make it be "interrupts are enabled, before serializing instruction". > > Yes, we want bounded error handling latency. But that doesn't mean "synchronous" Another X86 vendor seems to be adding something like that. See MCOMMIT in https://www.amd.com/system/files/TechDocs/24594.pdf But I wonder how an OS will know whether it is running some smart MCOMMIT-aware application that can figure out what to do with bad data, or a legacy application that should probably be stopped before it hurts somebody. I also wonder how expensive MCOMMIT is (since it is essentially polling for "did any errors happen"). -Tony
