lkml.org 
[lkml]   [2011]   [Jun]   [19]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH v2 5/7] KVM-GST: KVM Steal time accounting
On 06/19/2011 07:04 AM, Avi Kivity wrote:
> On 06/17/2011 01:20 AM, Glauber Costa wrote:
>> This patch accounts steal time time in kernel/sched.
>> I kept it from last proposal, because I still see advantages
>> in it: Doing it here will give us easier access from scheduler
>> variables such as the cpu rq. The next patch shows an example of
>> usage for it.
>>
>> Since functions like account_idle_time() can be called from
>> multiple places, not only account_process_tick(), steal time
>> grabbing is repeated in each account function separatedely.
>>
>> /*
>> + * We have to at flush steal time information every time something else
>> + * is accounted. Since the accounting functions are all visible to
>> the rest
>> + * of the kernel, it gets tricky to do them in one place. This helper
>> function
>> + * helps us.
>> + *
>> + * When the system is idle, the concept of steal time does not apply.
>> We just
>> + * tell the underlying hypervisor that we grabbed the data, but skip
>> steal time
>> + * accounting
>> + */
>> +static inline bool touch_steal_time(int is_idle)
>> +{
>> + u64 steal, st = 0;
>> +
>> + if (static_branch(&paravirt_steal_enabled)) {
>> +
>> + steal = paravirt_steal_clock(smp_processor_id());
>> +
>> + steal -= this_rq()->prev_steal_time;
>> + if (is_idle) {
>> + this_rq()->prev_steal_time += steal;
>> + return false;
>> + }
>> +
>> + while (steal>= TICK_NSEC) {
>> + /*
>> + * Inline assembly required to prevent the compiler
>> + * optimising this loop into a divmod call.
>> + * See __iter_div_u64_rem() for another example of this.
>> + */
>
> Why not use said function?

because here we want to do work during each loop. The said function
would have to be adapted for that, possibly using a macro, to run
arbitrary code during each loop iteration, in a way that I don't think
it is worthy given the current number of callers (2 counting this new one)

>> + asm("" : "+rm" (steal));
>> +
>> + steal -= TICK_NSEC;
>> + this_rq()->prev_steal_time += TICK_NSEC;
>> + st++;
>
> Suppose a live migration or SIGSTOP causes lots of steal time. How long
> will we spend here?
Silly me. I actually used this same argument with Peter to cap it with
"delta" in the next patch in this series. So I think you are 100 %
right. Here, however, we do want to account all that time, I believe.

How about we do a slow division if we're > 10 sec (unlikely), and
account everything as steal time in this scenario ?

>> + }
>> +
>> + account_steal_time(st);
>> + return !!st;
>
> !! !needed, you're returning a bool.

ah, sure thing.

>> + }
>> + return false;
>> +}
>> +
>
> I'll need Peter's (or another sched maintainer's) review to apply this.
>



\
 
 \ /
  Last update: 2011-06-20 04:41    [W:0.120 / U:0.324 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site