On 03/24/2014 05:41 AM, Peter Maydell wrote: > On 15 March 2014 02:48, Richard Henderson <r...@twiddle.net> wrote: >> Since the kernel doesn't pass any info on the reason for the fault, >> disassemble the instruction to detect a store. > > Incidentally, I've been wondering if we could improve > handle_cpu_signal so that at least the "check if this > fault was because we write-protected a page when we > translated code out of it" part doesn't depend on the > CPU-specific signal handler setting is_write correctly. > I think most guests don't depend on getting exactly > correct fault information, but if we don't track our > own page protection correctly then even simple guest > binaries don't work.
Indeed. I had wondered if just setting is_write=1 when we don't know wouldn't be a better solution that what we have now. > (Also, shouldn't we ideally speaking see if the SIGSEGV > was the result of attempting to execute from non-executable > memory?) Probably, but I'm not sure we know at this point. Although, honestly, the best fallback would be softmmu. Which we really ought to enable for both 64-on-32 and host > guest page size. Especially with the later we sometimes can't even *load* simple binaries. r~
