kvm.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: "Radim Krčmář" <rkrcmar@redhat.com>
To: Wanpeng Li <kernellwp@gmail.com>
Cc: kvm <kvm@vger.kernel.org>, Wanpeng Li <wanpeng.li@hotmail.com>,
	Paolo Bonzini <pbonzini@redhat.com>
Subject: Re: [PATCH] x86: apic: add APIC Timer periodic/oneshot mode latency test
Date: Wed, 26 Oct 2016 16:13:06 +0200	[thread overview]
Message-ID: <20161026141305.GA4212@potion> (raw)
In-Reply-To: <CANRm+CxiFmCvP8gAwtKJe6qK2m5YamOBEs_neCL7hw83gQVTdg@mail.gmail.com>

2016-10-26 14:35+0800, Wanpeng Li:
> 2016-10-25 19:55 GMT+08:00 Radim Krčmář <rkrcmar@redhat.com>:
>> 2016-10-17 16:28+0800, Wanpeng Li:
> [...]
>>> +static void apic_timer_isr(isr_regs_t *regs)
>>> +{
>>> +    u64 now = rdtsc();
>>> +    ++tdt_count;
>>> +
>>> +    if (table_idx < TABLE_SIZE && tdt_count > 1)
>>> +        table[table_idx++] = now - exptime;
>>> +
>>> +    if (breakmax && tdt_count > 1 && (now - exptime) > breakmax) {
>>> +        hitmax = 1;
>>> +        apic_write(APIC_EOI, 0);
>>> +        return;
>>> +    }
>>> +
>>> +    exptime = now+delta;
>>
>> Two problems on this line:
>>
>> 1) with periodic timer, the delta is added to last expiration time, not
>>    to now().
>>    What you are measuring now is the stability of the period -- very
>>    important as well, but if we used now() only once, you could also
>>    measure the latency.
>>
>> 2) delta is in nanoseconds while exptime is in cycles.  Type error.
> 
> There needs a method to convert the value which is loaded to
> initial-counter register to TSC.

Yes, you need to know the conversion between APIC timer frequency and
TSC frequency.  APIC timer in KVM uses nanoseconds, but TSC is variable.

>>
>>> +
>>> +    if (mode == APIC_LVT_TIMER_ONE_SHOT)
>>> +        /* Set "Initial Counter Register", which starts the timer */
>>> +        apic_write(APIC_TMICT, delta);
>>
>> This is not deadline, so it would be better to read now() for exptime
>> right before the apic_write, so as little time as possible passes in
>> between.
> 
> Do you mean something like this?
> 
> static void apic_timer_isr(isr_regs_t *regs)
> {
>     uint64_t now = rdtsc();
> 
>     if (breakmax && (now - exptime) > breakmax) {

Note that you must have now and exptime in the same unit to have
meaningful operations on.  (especially the "exptime += delta" would bad
without conversion).

>         hitmax = 1;
>         apic_write(APIC_EOI, 0);
>         return;
>     }
> 
>     exptime += delta;


The exptime for oneshot is 'exptime = now + delta', only periodic can do
exptime += delta, so I'd rather read the time again just before writing
the delta

>     if (mode == APIC_LVT_TIMER_ONE_SHOT) {
          time_t now = now();
>         /* Set "Initial Counter Register", which starts the timer */
>         apic_write(APIC_TMICT, delta);

          exptime = now + delta;
      }

>     if (table_idx < TABLE_SIZE)
>         table[table_idx++] = now - exptime;
> 
>     apic_write(APIC_EOI, 0);
> }
> 
> static void test_apic_timer(int mode)
> {
>     handle_irq(APIC_LVT_TIMER_VECTOR, apic_timer_isr);
>     irq_enable();
> 
>          /* Periodic mode */
>     apic_write(APIC_LVTT, mode | APIC_LVT_TIMER_VECTOR);
>     /* Divider == 1 */
>     apic_write(APIC_TDCR, 0x0000000b);
>     exptime = rdtsc() + delta;

Yep, if we convert rdtsc() to the same unit that delta uses.

Thanks.

>     apic_write(APIC_TMICT, delta);
> }
> 
> Regards,
> Wanpeng Li

  reply	other threads:[~2016-10-26 14:13 UTC|newest]

Thread overview: 5+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2016-10-17  8:28 [PATCH] x86: apic: add APIC Timer periodic/oneshot mode latency test Wanpeng Li
2016-10-25 11:55 ` Radim Krčmář
2016-10-26  6:35   ` Wanpeng Li
2016-10-26 14:13     ` Radim Krčmář [this message]
2016-10-27  8:51       ` Wanpeng Li

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20161026141305.GA4212@potion \
    --to=rkrcmar@redhat.com \
    --cc=kernellwp@gmail.com \
    --cc=kvm@vger.kernel.org \
    --cc=pbonzini@redhat.com \
    --cc=wanpeng.li@hotmail.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).