Re: [Qemu-devel] [PATCH v1 1/1] RISC-V: Workaround for critical mstatus.FS MTTCG bug

[Richard Henderson]

On 03/28/2018 08:15 AM, Michael Clark wrote:
> On looking at this again, I think we may need to remove
> the qemu_tcg_mttcg_enabled conditional and always return dirty if the state is
> initial or clean, but not off.

Yes.

> While testing on uniprocessor worked okay, it's likely because we were lucky
> and there was no task migration or multiple FPU tasks working.

Also yes.

>     > +        if (qemu_tcg_mttcg_enabled()) {
>     > +            /* FP is always dirty or off */
>     > +            if (mstatus & MSTATUS_FS) {
>     > +                mstatus |= MSTATUS_FS;
>     > +            }
>     > +        }

I've just posted an alternate patch set to track MSTATUS_FS more exactly, but
if that's thought to be too much this late in the cycle, then I'll sign off on
this patch without the mttcg test.


r~