Re: [PATCH v2 5/7] target/nios2: Convert to TranslatorOps

[Peter Maydell]

On Sun, 20 Jun 2021 at 23:18, Richard Henderson
<richard.henderson@linaro.org> wrote:
>
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> ---
>  target/nios2/translate.c | 130 ++++++++++++++++++++-------------------
>  1 file changed, 67 insertions(+), 63 deletions(-)
>
> diff --git a/target/nios2/translate.c b/target/nios2/translate.c
> index 31653b7912..06705c894d 100644
> --- a/target/nios2/translate.c
> +++ b/target/nios2/translate.c
> @@ -803,75 +803,72 @@ static void gen_exception(DisasContext *dc, uint32_t 
> excp)
>  }
>
>  /* generate intermediate code for basic block 'tb'.  */
> -void gen_intermediate_code(CPUState *cs, TranslationBlock *tb, int max_insns)
> +static void nios2_tr_init_disas_context(DisasContextBase *dcbase, CPUState 
> *cs)
>  {
> +    DisasContext *dc = container_of(dcbase, DisasContext, base);
>      CPUNios2State *env = cs->env_ptr;
> -    DisasContext dc1, *dc = &dc1;
> -    int num_insns;
> -
> -    /* Initialize DC */
> -
> -    dc->base.tb = tb;
> -    dc->base.singlestep_enabled = cs->singlestep_enabled;
> -    dc->base.is_jmp = DISAS_NEXT;
> -    dc->base.pc_first = tb->pc;
> -    dc->base.pc_next = tb->pc;
> +    target_ulong pc = dc->base.pc_first;

The local variable doesn't really seem necessary -- you could just
write "dc->pc = dc->base.pc_first" and then use "dc->pc" in the
calculation of page_insns.

> +    int page_insns;
>
>      dc->zero    = NULL;
> -    dc->pc      = tb->pc;
> +    dc->pc      = pc;
>      dc->mem_idx = cpu_mmu_index(env, false);
>
> -    /* Set up instruction counts */
> -    num_insns = 0;
> -    if (max_insns > 1) {
> -        int page_insns = (TARGET_PAGE_SIZE - (tb->pc & ~TARGET_PAGE_MASK)) / 
> 4;
> -        if (max_insns > page_insns) {
> -            max_insns = page_insns;
> -        }
> -    }
> +    /* Bound the number of insns to execute to those left on the page.  */
> +    page_insns = -(pc | TARGET_PAGE_MASK) / 4;
> +    dc->base.max_insns = MIN(page_insns, dc->base.max_insns);
> +}
>
> -    gen_tb_start(tb);
> -    do {
> -        tcg_gen_insn_start(dc->pc);
> -        num_insns++;
> +static void nios2_tr_tb_start(DisasContextBase *db, CPUState *cs)
> +{
> +}
>
> -        if (unlikely(cpu_breakpoint_test(cs, dc->pc, BP_ANY))) {
> -            gen_exception(dc, EXCP_DEBUG);
> -            /* The address covered by the breakpoint must be included in
> -               [tb->pc, tb->pc + tb->size) in order to for it to be
> -               properly cleared -- thus we increment the PC here so that
> -               the logic setting tb->size below does the right thing.  */
> -            dc->pc += 4;
> -            break;
> -        }
> +static void nios2_tr_insn_start(DisasContextBase *dcbase, CPUState *cs)
> +{
> +    tcg_gen_insn_start(dcbase->pc_next);
> +}
>
> -        if (num_insns == max_insns && (tb_cflags(tb) & CF_LAST_IO)) {
> -            gen_io_start();
> -        }
> +static bool nios2_tr_breakpoint_check(DisasContextBase *dcbase, CPUState *cs,
> +                                      const CPUBreakpoint *bp)
> +{
> +    DisasContext *dc = container_of(dcbase, DisasContext, base);
>
> -        /* Decode an instruction */
> -        handle_instruction(dc, env);
> +    gen_exception(dc, EXCP_DEBUG);
> +    /*
> +     * The address covered by the breakpoint must be included in
> +     * [tb->pc, tb->pc + tb->size) in order to for it to be
> +     * properly cleared -- thus we increment the PC here so that
> +     * the logic setting tb->size below does the right thing.
> +     */
> +    dc->pc += 4;

Don't we need to increment dc->base.pc_next here, not dc->pc? The
generic setting of tb->size in accel/tcg uses "db->pc_next - db->pc_first".

Side note: that comment about "setting tb->size below" seems to have
been copied-and-pasted into most of the front-ends, but it's wrong:
the setting of tb->size is no longer "below" but in the common code.

(More generally, some followup patches rationalizing the use of
dc->pc along the lines of the Arm dc->pc_curr vs dc->base.pc_next
might help. There seems to be a fair amount of code which uses
"dc->pc + 4" that could be using pc_next instead.)

> +    return true;

The arm versions of the breakpoint_check hook also set dc->base.is_jmp
to DISAS_NORETURN.
Are they doing that unnecessarily, or do we need to do that here ?

> +}
>
> -        dc->pc += 4;
> +static void nios2_tr_translate_insn(DisasContextBase *dcbase, CPUState *cs)
> +{
> +    DisasContext *dc = container_of(dcbase, DisasContext, base);
> +    CPUNios2State *env = cs->env_ptr;
>
> -        /* Translation stops when a conditional branch is encountered.
> -         * Otherwise the subsequent code could get translated several times.
> -         * Also stop translation when a page boundary is reached.  This
> -         * ensures prefetch aborts occur at the right place.  */
> -    } while (!dc->base.is_jmp &&
> -             !tcg_op_buf_full() &&
> -             num_insns < max_insns);
> +    /* Decode an instruction */
> +    handle_instruction(dc, env);
> +
> +    dc->base.pc_next += 4;
> +    dc->pc += 4;

This isn't wrong, but I think that a setup like the Arm translator
that does
       dc->pc_curr = s->base.pc_next;
       code = cpu_ldl_code(env, s->base.pc_next);
       s->base.pc_next += 4;
       /* dispatch to handler function here */

would be nicer (dunno whether clearer to do as a single thing or
first to do this conversion and then do a followup patch).

thanks
-- PMM