Re: what are the requirements on target/ code for -icount to work correctly?

[Pavel Dovgalyuk]

On 22.06.2020 08:24, Max Filippov wrote:
> On Fri, Jun 19, 2020 at 10:05 AM Peter Maydell <peter.maydell@linaro.org> wrote:
> > I've just sent a patch that removes the target/arm gen_io_end() calls.
> > I had a quick look at sparc, xtensa and ppc, but they were too complicated
> > for a quick look to be sufficient :-)
>
> I've checked the xtensa translator. The only gen_io_end() is for
> opcodes that end TB when a full instruction is translated, because
> they can change active interrupt requests. So it can be removed.
> I'll send a patch.
>
> One instance of gen_io_start is for the rsr.ccount opcode that reads
> current cycle counter that translates to reading
> QEMU_CLOCK_VIRTUAL clock.

Right, this is documented by Alex in the new document recently sent into 
the mailing list:
7611429f-ec38-554d-3aec-e4a2f456a1ea@linaro.org/T/#mff14f2e3dc2cedd7cc6b1356985a0454e4772202">https://lore.kernel.org/qemu-devel/7611429f-ec38-554d-3aec-e4a2f456a1ea@linaro.org/T/#mff14f2e3dc2cedd7cc6b1356985a0454e4772202

> Looks like this adds one more case to the
> following list:
>
> > As I understand it, the definition of "I/O insn" is anything that can
> > either:
> >
> > - affect the icount deadline (e.g. by setting or removing a
> > QEMU_CLOCK_VIRTUAL timer)
> >
> > - interrupt the current translation block with cpu_loop_exit,
> > cpu_restore_state or similar.