Re: [PATCH kvm-unit-tests] vmexit: measure IPI and EOI cost

[Wanpeng Li <kernellwp@xxxxxxxxx>]

On Fri, 4 Oct 2019 at 00:00, Paolo Bonzini <pbonzini@xxxxxxxxxx> wrote:
>
> Signed-off-by: Paolo Bonzini <pbonzini@xxxxxxxxxx>
> ---
>  x86/vmexit.c | 16 ++++++++++++++++
>  1 file changed, 16 insertions(+)
>
> diff --git a/x86/vmexit.c b/x86/vmexit.c
> index 66d3458..81b743b 100644
> --- a/x86/vmexit.c
> +++ b/x86/vmexit.c
> @@ -65,22 +65,30 @@ static void nop(void *junk)
>  }
>
>  volatile int x = 0;
> +volatile uint64_t tsc_eoi = 0;
> +volatile uint64_t tsc_ipi = 0;
>
>  static void self_ipi_isr(isr_regs_t *regs)
>  {
>         x++;
> +       uint64_t start = rdtsc();
>         eoi();
> +       tsc_eoi += rdtsc() - start;
>  }
>
>  static void x2apic_self_ipi(int vec)
>  {
> +       uint64_t start = rdtsc();
>         wrmsr(0x83f, vec);
> +       tsc_ipi += rdtsc() - start;
>  }
>
>  static void apic_self_ipi(int vec)
>  {
> +       uint64_t start = rdtsc();
>          apic_icr_write(APIC_INT_ASSERT | APIC_DEST_SELF | APIC_DEST_PHYSICAL |
>                        APIC_DM_FIXED | IPI_TEST_VECTOR, vec);
> +       tsc_ipi += rdtsc() - start;
>  }
>
>  static void self_ipi_sti_nop(void)
> @@ -180,7 +188,9 @@ static void x2apic_self_ipi_tpr_sti_hlt(void)
>
>  static void ipi(void)
>  {
> +       uint64_t start = rdtsc();
>         on_cpu(1, nop, 0);
> +       tsc_ipi += rdtsc() - start;
>  }
>
>  static void ipi_halt(void)
> @@ -511,6 +521,7 @@ static bool do_test(struct test *test)
>         }
>
>         do {
> +               tsc_eoi = tsc_ipi = 0;
>                 iterations *= 2;
>                 t1 = rdtsc();
>
> @@ -523,6 +534,11 @@ static bool do_test(struct test *test)
>                 t2 = rdtsc();
>         } while ((t2 - t1) < GOAL);
>         printf("%s %d\n", test->name, (int)((t2 - t1) / iterations));
> +       if (tsc_ipi)
> +               printf("  ipi %s %d\n", test->name, (int)(tsc_ipi / iterations));
> +       if (tsc_eoi)
> +               printf("  eoi %s %d\n", test->name, (int)(tsc_eoi / iterations));
> +
>         return test->next;
>  }

Thanks for this, do you have more idea to optimize the virtual ipi latency?

    Wanpeng