drm/amdgpu/mes11: Use a separate fence per transaction

We can't use a shared fence location because each transaction
should be considered independently.

Reviewed-by: Shaoyun.liu <shaoyunl@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

3 files changed, 33 insertions, 4 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.c
index 310c8f0c21da..21d1d508ee10 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.c
@@ -32,6 +32,18 @@
 #define AMDGPU_MES_MAX_NUM_OF_QUEUES_PER_PROCESS 1024
 #define AMDGPU_ONE_DOORBELL_SIZE 8
 
+signed long amdgpu_mes_fence_wait_polling(u64 *fence,
+					  u64 wait_seq,
+					  signed long timeout)
+{
+
+	while ((s64)(wait_seq - *fence) > 0 && timeout > 0) {
+		udelay(2);
+		timeout -= 2;
+	}
+	return timeout > 0 ? timeout : 0;
+}
+
 int amdgpu_mes_doorbell_process_slice(struct amdgpu_device *adev)
 {
 	return roundup(AMDGPU_ONE_DOORBELL_SIZE *
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h
index 6b3e1844eac5..b99a2b3cffe3 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h
@@ -340,6 +340,10 @@ struct amdgpu_mes_funcs {
 #define amdgpu_mes_kiq_hw_init(adev) (adev)->mes.kiq_hw_init((adev))
 #define amdgpu_mes_kiq_hw_fini(adev) (adev)->mes.kiq_hw_fini((adev))
 
+signed long amdgpu_mes_fence_wait_polling(u64 *fence,
+					  u64 wait_seq,
+					  signed long timeout);
+
 int amdgpu_mes_ctx_get_offs(struct amdgpu_ring *ring, unsigned int id_offs);
 
 int amdgpu_mes_init_microcode(struct amdgpu_device *adev, int pipe);
diff --git a/drivers/gpu/drm/amd/amdgpu/mes_v11_0.c b/drivers/gpu/drm/amd/amdgpu/mes_v11_0.c
index 5c08ad234439..0d1407f25005 100644
--- a/drivers/gpu/drm/amd/amdgpu/mes_v11_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/mes_v11_0.c
@@ -163,6 +163,10 @@ static int mes_v11_0_submit_pkt_and_poll_completion(struct amdgpu_mes *mes,
 	unsigned long flags;
 	signed long timeout = 3000000; /* 3000 ms */
 	const char *op_str, *misc_op_str;
+	u32 fence_offset;
+	u64 fence_gpu_addr;
+	u64 *fence_ptr;
+	int ret;
 
 	if (x_pkt->header.opcode >= MES_SCH_API_MAX)
 		return -EINVAL;
@@ -175,15 +179,24 @@ static int mes_v11_0_submit_pkt_and_poll_completion(struct amdgpu_mes *mes,
 	}
 	BUG_ON(size % 4 != 0);
 
+	ret = amdgpu_device_wb_get(adev, &fence_offset);
+	if (ret)
+		return ret;
+	fence_gpu_addr =
+		adev->wb.gpu_addr + (fence_offset * 4);
+	fence_ptr = (u64 *)&adev->wb.wb[fence_offset];
+	*fence_ptr = 0;
+
 	spin_lock_irqsave(&mes->ring_lock, flags);
 	if (amdgpu_ring_alloc(ring, ndw)) {
 		spin_unlock_irqrestore(&mes->ring_lock, flags);
+		amdgpu_device_wb_free(adev, fence_offset);
 		return -ENOMEM;
 	}
 
 	api_status = (struct MES_API_STATUS *)((char *)pkt + api_status_off);
-	api_status->api_completion_fence_addr = mes->ring.fence_drv.gpu_addr;
-	api_status->api_completion_fence_value = ++mes->ring.fence_drv.sync_seq;
+	api_status->api_completion_fence_addr = fence_gpu_addr;
+	api_status->api_completion_fence_value = 1;
 
 	amdgpu_ring_write_multiple(ring, pkt, ndw);
 	amdgpu_ring_commit(ring);
@@ -199,8 +212,8 @@ static int mes_v11_0_submit_pkt_and_poll_completion(struct amdgpu_mes *mes,
 	else
 		dev_dbg(adev->dev, "MES msg=%d was emitted\n", x_pkt->header.opcode);
 
-	r = amdgpu_fence_wait_polling(ring, ring->fence_drv.sync_seq,
-		      timeout);
+	r = amdgpu_mes_fence_wait_polling(fence_ptr, (u64)1, timeout);
+	amdgpu_device_wb_free(adev, fence_offset);
 	if (r < 1) {
 
 		if (misc_op_str)