Re: [Qemu-devel] [PATCH] improve emulation correctness

[Peter Maydell]

On 27 April 2014 17:46, Michael Tokarev <address@hidden> wrote:
> 25.04.2014 21:24, Peter Maydell wrote:
>> It is always going to be possible to determine that you're
>> running on an emulator rather than real hardware, so changing
>> QEMU behaviour just for this is uninteresting. If QEMU
>> behaves differently from the specification (in this case
>> the x86 hardware and architecture manuals) that's an interesting
>> bug. If we just happen to choose a different undefined
>> behaviour from that which hardware does, that is not in
>> my opinion a problem.
>
> Actually it might be a problem.  We know a ton of examples where
> hardware had bugs and software had to compensate and actually
> _expect_ buggy behavour.  Or when software depends on (reliable
> repeating) certain behavour in undocumented parts, and hell
> breaks when this (undocumented!) behavour changes.

Yes, sometimes you get that sort of thing. But we should only
put that kind of change in when we actually need it (ie when
there's real guest code that hits it, not just silly test programs
deliberately looking for undefined behaviour), and comment
why it's there so people who need to maintain that bit of code
in future know that it's deliberate and not just accidental.

I should perhaps have said "that is not inherently a problem".

thanks
-- PMM