Re: [PATCH v2 06/10] accel/tcg: Use CF_NO_GOTO_{TB, PTR} in cpu_exec_step_atomic

[Richard Henderson]

On 7/17/21 10:43 AM, Peter Maydell wrote:
> On Mon, 12 Jul 2021 at 16:46, Richard Henderson
> <richard.henderson@linaro.org> wrote:
> > Request that the one TB returns immediately, so that
> > we release the exclusive lock as soon as possible.
> >
> > Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> > ---
> >   accel/tcg/cpu-exec.c | 11 ++++++++---
> >   1 file changed, 8 insertions(+), 3 deletions(-)
> >
> > diff --git a/accel/tcg/cpu-exec.c b/accel/tcg/cpu-exec.c
> > index 2206c463f5..5bb099174f 100644
> > --- a/accel/tcg/cpu-exec.c
> > +++ b/accel/tcg/cpu-exec.c
> > @@ -330,8 +330,7 @@ void cpu_exec_step_atomic(CPUState *cpu)
> >       CPUArchState *env = (CPUArchState *)cpu->env_ptr;
> >       TranslationBlock *tb;
> >       target_ulong cs_base, pc;
> > -    uint32_t flags;
> > -    uint32_t cflags = (curr_cflags(cpu) & ~CF_PARALLEL) | 1;
> > +    uint32_t flags, cflags;
> >       int tb_exit;
> >
> >       if (sigsetjmp(cpu->jmp_env, 0) == 0) {
> > @@ -341,8 +340,14 @@ void cpu_exec_step_atomic(CPUState *cpu)
> >           cpu->running = true;
> >
> >           cpu_get_tb_cpu_state(env, &pc, &cs_base, &flags);
> > -        tb = tb_lookup(cpu, pc, cs_base, flags, cflags);
> >
> > +        cflags = curr_cflags(cpu);
> > +        /* Execute in a serial context. */
> > +        cflags &= ~CF_PARALLEL;
> > +        /* After 1 insn, return and release the exclusive lock. */
> > +        cflags |= CF_NO_GOTO_TB | CF_NO_GOTO_PTR | 1;
> > +
> > +        tb = tb_lookup(cpu, pc, cs_base, flags, cflags);
> >           if (tb == NULL) {
> >               mmap_lock();
> >               tb = tb_gen_code(cpu, pc, cs_base, flags, cflags);
>
> So previously we would have executed possibly a chain of TBs
> before releasing the lock, and now we definitely execute just one?

Correct.

> (I guess the execute-a-chain case is unlikely given the TB
> only has one insn and we know it's an exclusive insn...)

I think it's actually likely.  While the tb would definitely end after one insn, we had 
passed nothing down that would lead to returning to the main loop.


r~

> Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
>
> thanks
> -- PMM