Re: [PATCH v3 59/66] accel/tcg: Handle SIGBUS in handle_cpu_signal

[Richard Henderson]

On 8/20/21 2:34 AM, Peter Maydell wrote:
> On Wed, 18 Aug 2021 at 21:13, Richard Henderson
> <richard.henderson@linaro.org> wrote:
> > We've been registering host SIGBUS, but then treating it
> > exactly like SIGSEGV.
> >
> > Handle BUS_ADRALN via cpu_unaligned_access, but allow other
> > SIGBUS si_codes to continue into the host-to-guest signal
> > coversion code in host_signal_handler.  Unwind the guest
> > state so that we report the correct guest PC for the fault.
>
> You can't rely on alignment faults being marked by BUS_ADRALN:
> eg MIPS doesn't give you that si_code. How much does that matter
> for our use of it here ?

Aside from whatever tcg/foo relies upon, it doesn't matter.

But: silly mips. I was going to rely on it there -- oh well, I'll undo that section of the 
follow-on tcg/mips patch set.  And it's got a different syscall for disabling the 
unaligned fixup.


r~