[PATCH 2/2] drm/msm/a5xx: Record per-submission GPU time and cycles

[Fabricio Akio via B4 Relay <devnull+fabricioakio.gmail.com@kernel.org>]

From: Fabricio Akio <fabricioakio@gmail.com>

The GPU time and cycles msm records for each submission, reported in
the msm_gpu_submit_retired trace event and added to fdinfo's
drm-engine-gpu and drm-cycles-gpu, come from counters the CP copies to
memory around each submission. a6xx_submit() emits those copies;
a5xx_submit() does not, so on A5xx all of them read zero.

Copy the CP cycle counter and the always-on counter before and after
the submission's IBs, as a6xx does.

Signed-off-by: Fabricio Akio <fabricioakio@gmail.com>
---
The counters are copied outside the IBs, so a preemption inside a
submission, at a yield point in its IBs, would count the other ring's
time in it. Switches between submissions were tested.
---
 drivers/gpu/drm/msm/adreno/a5xx_gpu.c | 22 ++++++++++++++++++++++
 1 file changed, 22 insertions(+)

diff --git a/drivers/gpu/drm/msm/adreno/a5xx_gpu.c b/drivers/gpu/drm/msm/adreno/a5xx_gpu.c
index 16c7335c4b1f..7d4a9db0a3de 100644
--- a/drivers/gpu/drm/msm/adreno/a5xx_gpu.c
+++ b/drivers/gpu/drm/msm/adreno/a5xx_gpu.c
@@ -124,8 +124,20 @@ static void a5xx_submit_in_rb(struct msm_gpu *gpu, struct msm_gem_submit *submit
 	msm_gpu_retire(gpu);
 }
 
+static void get_stats_counter(struct msm_ringbuffer *ring, u32 counter,
+			      u64 iova)
+{
+	OUT_PKT7(ring, CP_REG_TO_MEM, 3);
+	OUT_RING(ring, CP_REG_TO_MEM_0_REG(counter) |
+		CP_REG_TO_MEM_0_CNT(2) |
+		CP_REG_TO_MEM_0_IS_64B);
+	OUT_RING(ring, lower_32_bits(iova));
+	OUT_RING(ring, upper_32_bits(iova));
+}
+
 static void a5xx_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit)
 {
+	unsigned int index = submit->seqno % MSM_GPU_SUBMIT_STATS_COUNT;
 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
 	struct a5xx_gpu *a5xx_gpu = to_a5xx_gpu(adreno_gpu);
 	struct msm_ringbuffer *ring = submit->ring;
@@ -167,6 +179,11 @@ static void a5xx_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit)
 	OUT_PKT7(ring, CP_YIELD_ENABLE, 1);
 	OUT_RING(ring, 0x02);
 
+	get_stats_counter(ring, REG_A5XX_RBBM_PERFCTR_CP_0_LO,
+			  rbmemptr_stats(ring, index, cpcycles_start));
+	get_stats_counter(ring, REG_A5XX_RBBM_ALWAYSON_COUNTER_LO,
+			  rbmemptr_stats(ring, index, alwayson_start));
+
 	/* Submit the commands */
 	for (i = 0; i < submit->nr_cmds; i++) {
 		switch (submit->cmd[i].type) {
@@ -208,6 +225,11 @@ static void a5xx_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit)
 	OUT_RING(ring, 0);
 	OUT_RING(ring, 0);
 
+	get_stats_counter(ring, REG_A5XX_RBBM_PERFCTR_CP_0_LO,
+			  rbmemptr_stats(ring, index, cpcycles_end));
+	get_stats_counter(ring, REG_A5XX_RBBM_ALWAYSON_COUNTER_LO,
+			  rbmemptr_stats(ring, index, alwayson_end));
+
 	/* Turn off IB level preemptions */
 	OUT_PKT7(ring, CP_YIELD_ENABLE, 1);
 	OUT_RING(ring, 0x01);

-- 
2.56.0