[for-4.2,02/24] cputlb: Add tlb_flush_asid_by_mmuidx and friends

Since we have remembered ASIDs, we can further minimize flushing
by comparing against the one we want to flush.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 include/exec/exec-all.h | 16 +++++++++++++
 include/qom/cpu.h       |  1 +
 accel/tcg/cputlb.c      | 51 +++++++++++++++++++++++++++++++++++++++++
 3 files changed, 68 insertions(+)

Richard Henderson <richard.henderson@linaro.org> writes:

> Since we have remembered ASIDs, we can further minimize flushing
> by comparing against the one we want to flush.
>
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> ---
>  include/exec/exec-all.h | 16 +++++++++++++
>  include/qom/cpu.h       |  1 +
>  accel/tcg/cputlb.c      | 51 +++++++++++++++++++++++++++++++++++++++++
>  3 files changed, 68 insertions(+)
>
> diff --git a/include/exec/exec-all.h b/include/exec/exec-all.h
> index 9c77aa5bf9..0d890e1e60 100644
> --- a/include/exec/exec-all.h
> +++ b/include/exec/exec-all.h
> @@ -240,6 +240,22 @@ void tlb_flush_by_mmuidx_all_cpus_synced(CPUState *cpu, uint16_t idxmap);
>   */
>  void tlb_set_asid_for_mmuidx(CPUState *cpu, uint32_t asid,
>                               uint16_t idxmap, uint16_t dep_idxmap);
> +/**
> + * tlb_flush_asid_by_mmuidx:
> + * @cpu: Originating CPU of the flush
> + * @asid: Address Space Identifier
> + * @idxmap: bitmap of MMU indexes to flush if asid matches
> + *
> + * For each mmu index, if @asid matches the value previously saved via
> + * tlb_set_asid_for_mmuidx, flush the index.
> + */
> +void tlb_flush_asid_by_mmuidx(CPUState *cpu, uint32_t asid, uint16_t idxmap);
> +/* Similarly, broadcasting to all cpus. */
> +void tlb_flush_asid_by_mmuidx_all_cpus(CPUState *cpu, uint32_t asid,
> +                                       uint16_t idxmap);
> +/* Similarly, waiting for the broadcast to complete.  */
> +void tlb_flush_asid_by_mmuidx_all_cpus_synced(CPUState *cpu, uint32_t asid,
> +                                              uint16_t idxmap);
>  /**
>   * tlb_set_page_with_attrs:
>   * @cpu: CPU to add this TLB entry for
> diff --git a/include/qom/cpu.h b/include/qom/cpu.h
> index 5ee0046b62..4ae6ea3e1d 100644
> --- a/include/qom/cpu.h
> +++ b/include/qom/cpu.h
> @@ -283,6 +283,7 @@ struct hax_vcpu_state;
>  typedef union {
>      int           host_int;
>      unsigned long host_ulong;
> +    uint64_t      host_uint64;

Missing an access helper, although host_uint64 doesn't make sense, 64 is
64 always:

  #define RUN_ON_CPU_UINT64(i)    ((run_on_cpu_data){.uint64 = (i)})

>      void         *host_ptr;
>      vaddr         target_ptr;
>  } run_on_cpu_data;
> diff --git a/accel/tcg/cputlb.c b/accel/tcg/cputlb.c
> index c68f57755b..3ef68a11bf 100644
> --- a/accel/tcg/cputlb.c
> +++ b/accel/tcg/cputlb.c
> @@ -540,6 +540,57 @@ void tlb_flush_page_all_cpus_synced(CPUState *src, target_ulong addr)
>      tlb_flush_page_by_mmuidx_all_cpus_synced(src, addr, ALL_MMUIDX_BITS);
>  }
>
> +static void tlb_flush_asid_by_mmuidx_async_work(CPUState *cpu,
> +                                                run_on_cpu_data data)
> +{
> +    CPUTLB *tlb = cpu_tlb(cpu);
> +    uint32_t asid = data.host_uint64;
> +    uint16_t idxmap = data.host_uint64 >> 32;
> +    uint16_t to_flush = 0, work;
> +
> +    assert_cpu_is_self(cpu);
> +
> +    for (work = idxmap; work != 0; work &= work - 1) {
> +        int mmu_idx = ctz32(work);
> +        if (tlb->d[mmu_idx].asid == asid) {
> +            to_flush |= 1 << mmu_idx;
> +        }
> +    }
> +
> +    if (to_flush) {
> +        tlb_flush_by_mmuidx_async_work(cpu, RUN_ON_CPU_HOST_INT(to_flush));
> +    }
> +}
> +
> +void tlb_flush_asid_by_mmuidx(CPUState *cpu, uint32_t asid, uint16_t idxmap)
> +{
> +    run_on_cpu_data data = { .host_uint64 = deposit64(asid, 32, 32,
> idxmap) };

Then this would be:
     uint64_t asid_idx_map = deposit64(asid, 32, 32, idxmap);
     ...
     async_run_on_cpu(cpu, tlb_flush_asid_by_mmuidx_async_work, RUN_ON_CPU_UINT64(data))

Not a massive win but consistent with the other *_run_on calls and
easier to grep.

> +
> +    if (cpu->created && !qemu_cpu_is_self(cpu)) {
> +        async_run_on_cpu(cpu, tlb_flush_asid_by_mmuidx_async_work, data);
> +    } else {
> +        tlb_flush_asid_by_mmuidx_async_work(cpu, data);
> +    }
> +}
> +
> +void tlb_flush_asid_by_mmuidx_all_cpus(CPUState *src_cpu,
> +                                       uint32_t asid, uint16_t idxmap)
> +{
> +    run_on_cpu_data data = { .host_uint64 = deposit64(asid, 32, 32, idxmap) };
> +
> +    flush_all_helper(src_cpu, tlb_flush_asid_by_mmuidx_async_work, data);
> +    tlb_flush_asid_by_mmuidx_async_work(src_cpu, data);
> +}
> +
> +void tlb_flush_asid_by_mmuidx_all_cpus_synced(CPUState *src_cpu,
> +                                              uint32_t asid, uint16_t idxmap)
> +{
> +    run_on_cpu_data data = { .host_uint64 = deposit64(asid, 32, 32, idxmap) };
> +
> +    flush_all_helper(src_cpu, tlb_flush_asid_by_mmuidx_async_work, data);
> +    async_safe_run_on_cpu(src_cpu, tlb_flush_asid_by_mmuidx_async_work, data);
> +}
> +
>  void tlb_set_asid_for_mmuidx(CPUState *cpu, uint32_t asid, uint16_t idxmap,
>                               uint16_t depmap)
>  {

Otherwise:

Reviewed-by: Alex Bennée <alex.bennee@linaro.org>

--
Alex Bennée