[Intel-gfx] [PATCH i-g-t] intel_gpu_top: Use actual period when calculating client busyness

[Tvrtko Ursulin]

On 15/03/2023 19:56, Umesh Nerlige Ramappa wrote:
> On Wed, Mar 15, 2023 at 09:20:49AM +0000, Tvrtko Ursulin wrote:
>>
>> Hi Umesh,
>>
>> On 14/03/2023 18:25, Umesh Nerlige Ramappa wrote:
>>> lgtm,
>>>
>>> Reviewed-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa at intel.com>
>>
>> Thanks - I had one second thought though. See below please.
>>
>>> On Tue, Mar 14, 2023 at 12:17:40PM +0000, Tvrtko Ursulin wrote:
>>>> From: Tvrtko Ursulin <tvrtko.ursulin at intel.com>
>>>>
>>>> On a slow machine, or with many processes and/or file descriptors to
>>>> parse, the period of the scanning loop can drift significantly from the
>>>> assumed value. This results in artificially inflated client busyness
>>>> percentages.
>>>>
>>>> To alleviate the issue take some real timestamps and use actual elapsed
>>>> time when calculating relative busyness.
>>>>
>>>> Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin at intel.com>
>>>> ---
>>>> tools/intel_gpu_top.c | 39 ++++++++++++++++++++++++++++++++++++++-
>>>> 1 file changed, 38 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/tools/intel_gpu_top.c b/tools/intel_gpu_top.c
>>>> index e13e35b71f4b..af4b350da8e4 100644
>>>> --- a/tools/intel_gpu_top.c
>>>> +++ b/tools/intel_gpu_top.c
>>>> @@ -43,6 +43,7 @@
>>>> #include <sys/types.h>
>>>> #include <unistd.h>
>>>> #include <termios.h>
>>>> +#include <time.h>
>>>> #include <sys/sysmacros.h>
>>>>
>>>> #include "igt_perf.h"
>>>> @@ -2524,6 +2525,38 @@ static void show_help_screen(void)
>>>> "\n");
>>>> }
>>>>
>>>> +static int gettime(struct timespec *ts)
>>>> +{
>>>> +    memset(ts, 0, sizeof(*ts));
>>>> +
>>>> +#ifdef CLOCK_MONOTONIC_RAW
>>>> +    if (!clock_gettime(CLOCK_MONOTONIC_RAW, ts))
>>>> +        return 0;
>>>> +#endif
>>>> +#ifdef CLOCK_MONOTONIC_COARSE
>>>> +    if (!clock_gettime(CLOCK_MONOTONIC_COARSE, ts))
>>>> +        return 0;
>>>> +#endif
>>
>> So I copied this (with some edits) from igt_core.c but I think I 
>> should actually remove the CLOCK_MONOTONIC_COARSE option. The usage in 
>> intel_gpu_top is not performance sensitive and tick granularity 
>> actually defeats to point of this patch.
>>
>> Okay to keep the r-b if I remove it?
> 
> Sure, okay to keep the R-b.

Thanks, pushed!

Regards,

Tvrtko