Re: [PATCH 9/9] cputlb: Hoist timestamp outside of loops over tlbs

[Philippe Mathieu-Daudé]

On 1/9/20 3:49 AM, Richard Henderson wrote:
> Do not call get_clock_realtime() in tlb_mmu_resize_locked,
> but hoist outside of any loop over a set of tlbs.  This is
> only two (indirect) callers, tlb_flush_by_mmuidx_async_work
> and tlb_flush_page_locked, so not onerous.
>
> Signed-off-by: Richard Henderson <address@hidden>

Reviewed-by: Philippe Mathieu-Daudé <address@hidden>

> ---
>   accel/tcg/cputlb.c | 14 ++++++++------
>   1 file changed, 8 insertions(+), 6 deletions(-)
>
> diff --git a/accel/tcg/cputlb.c b/accel/tcg/cputlb.c
> index 761e9d44d7..9f6cb36921 100644
> --- a/accel/tcg/cputlb.c
> +++ b/accel/tcg/cputlb.c
> @@ -137,12 +137,12 @@ static void tlb_window_reset(CPUTLBDesc *desc, int64_t ns,
>    * high), since otherwise we are likely to have a significant amount of
>    * conflict misses.
>    */
> -static void tlb_mmu_resize_locked(CPUTLBDesc *desc, CPUTLBDescFast *fast)
> +static void tlb_mmu_resize_locked(CPUTLBDesc *desc, CPUTLBDescFast *fast,
> +                                  int64_t now)
>   {
>       size_t old_size = tlb_n_entries(fast);
>       size_t rate;
>       size_t new_size = old_size;
> -    int64_t now = get_clock_realtime();
>       int64_t window_len_ms = 100;
>       int64_t window_len_ns = window_len_ms * 1000 * 1000;
>       bool window_expired = now > desc->window_begin_ns + window_len_ns;
> @@ -222,12 +222,13 @@ static void tlb_mmu_flush_locked(CPUTLBDesc *desc, 
> CPUTLBDescFast *fast)
>       memset(desc->vtable, -1, sizeof(desc->vtable));
>   }
>   
> -static void tlb_flush_one_mmuidx_locked(CPUArchState *env, int mmu_idx)
> +static void tlb_flush_one_mmuidx_locked(CPUArchState *env, int mmu_idx,
> +                                        int64_t now)
>   {
>       CPUTLBDesc *desc = &env_tlb(env)->d[mmu_idx];
>       CPUTLBDescFast *fast = &env_tlb(env)->f[mmu_idx];
>   
> -    tlb_mmu_resize_locked(desc, fast);
> +    tlb_mmu_resize_locked(desc, fast, now);
>       tlb_mmu_flush_locked(desc, fast);
>   }
>   
> @@ -310,6 +311,7 @@ static void tlb_flush_by_mmuidx_async_work(CPUState *cpu, run_on_cpu_data data)
>       CPUArchState *env = cpu->env_ptr;
>       uint16_t asked = data.host_int;
>       uint16_t all_dirty, work, to_clean;
> +    int64_t now = get_clock_realtime();
>   
>       assert_cpu_is_self(cpu);
>   
> @@ -324,7 +326,7 @@ static void tlb_flush_by_mmuidx_async_work(CPUState *cpu, run_on_cpu_data data)
>   
>       for (work = to_clean; work != 0; work &= work - 1) {
>           int mmu_idx = ctz32(work);
> -        tlb_flush_one_mmuidx_locked(env, mmu_idx);
> +        tlb_flush_one_mmuidx_locked(env, mmu_idx, now);
>       }
>   
>       qemu_spin_unlock(&env_tlb(env)->c.lock);
> @@ -446,7 +448,7 @@ static void tlb_flush_page_locked(CPUArchState *env, int 
> midx,
>           tlb_debug("forcing full flush midx %d ("
>                     TARGET_FMT_lx "/" TARGET_FMT_lx ")\n",
>                     midx, lp_addr, lp_mask);
> -        tlb_flush_one_mmuidx_locked(env, midx);
> +        tlb_flush_one_mmuidx_locked(env, midx, get_clock_realtime());
>       } else {
>           if (tlb_flush_entry_locked(tlb_entry(env, midx, page), page)) {
>               tlb_n_used_entries_dec(env, midx);