[PATCH 2/2] drm/amdgpu: fix fence calculation

[Christian König]

Am 07.05.21 um 00:37 schrieb David M Nieto:
> The proper metric for fence utilization over several
> contexts is an harmonic mean, but such calculation is
> prohibitive in kernel space, so the code approximates it.
>
> Because the approximation diverges when one context has a
> very small ratio compared with the other context, this change
> filter out ratios smaller that 0.01%
>
> Signed-off-by: David M Nieto <david.nieto at amd.com>
> Change-Id: I5b6e0ce5f489a5f55855d35354a6a3653e9d613b
> ---
>   drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c | 18 +++++++++++++++++-
>   1 file changed, 17 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
> index 9036c93b4a0c..a26496735080 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
> @@ -689,6 +689,8 @@ void amdgpu_ctx_fence_time(struct amdgpu_ctx *ctx, struct amdgpu_ctx_entity *cen
>   	}
>   }
>   
> +#define FENCE_USAGE_MIN_RATIO(max, total) (max > 16384ULL*total)

An AMDGPU_CTX_ prefix looks appropriate here and defines should be at 
the beginning of the file.

> +
>   ktime_t amdgpu_ctx_mgr_fence_usage(struct amdgpu_ctx_mgr *mgr, uint32_t hwip,
>   		uint32_t idx, uint64_t *elapsed)
>   {
> @@ -697,17 +699,29 @@ ktime_t amdgpu_ctx_mgr_fence_usage(struct amdgpu_ctx_mgr *mgr, uint32_t hwip,
>   	uint32_t id;
>   	struct amdgpu_ctx_entity *centity;
>   	ktime_t total = 0, max = 0;
> +	ktime_t ttotal = 0, tmax = 0;
> +
>   
>   	if (idx >= AMDGPU_MAX_ENTITY_NUM)
>   		return 0;
>   	idp = &mgr->ctx_handles;
>   	mutex_lock(&mgr->lock);
>   	idr_for_each_entry(idp, ctx, id) {
> +		ttotal = tmax = ktime_set(0, 0);

Rather define the variable in the loop in the first place.

>   		if (!ctx->entities[hwip][idx])
>   			continue;
>   
>   		centity = ctx->entities[hwip][idx];
> -		amdgpu_ctx_fence_time(ctx, centity, &total, &max);
> +		amdgpu_ctx_fence_time(ctx, centity, &ttotal, &tmax);
> +
> +		/* Harmonic mean approximation diverges for very small
> +		 * values. If ratio < 0.01% ignore
> +		 */
> +		if (FENCE_USAGE_MIN_RATIO(tmax, ttotal))
> +			continue;
> +
> +		total = ktime_add(total, ttotal);
> +		max = ktime_after(tmax, max) ? tmax : max;
>   	}
>   
>   	mutex_unlock(&mgr->lock);
> @@ -716,3 +730,5 @@ ktime_t amdgpu_ctx_mgr_fence_usage(struct amdgpu_ctx_mgr *mgr, uint32_t hwip,
>   
>   	return total;
>   }
> +
> +#undef FENCE_USAGE_MIN_RATIO

Please don't undef macros if not necessary.