drm/amdgpu: enable RAP TA load

Enable the RAP TA loading path and add RAP test
trigger interface.

v2: fix potential mem leak issue
Signed-off-by: Wenhui Sheng <Wenhui.Sheng@amd.com>
Reviewed-by: Guchun Chen <Guchun.Chen@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c

@@ -1430,6 +1430,168 @@ static int psp_dtm_terminate(struct psp_context *psp)
 }
 // DTM end
 
+// RAP start
+static int psp_rap_init_shared_buf(struct psp_context *psp)
+{
+	int ret;
+
+	/*
+	 * Allocate 16k memory aligned to 4k from Frame Buffer (local
+	 * physical) for rap ta <-> Driver
+	 */
+	ret = amdgpu_bo_create_kernel(psp->adev, PSP_RAP_SHARED_MEM_SIZE,
+				      PAGE_SIZE, AMDGPU_GEM_DOMAIN_VRAM,
+				      &psp->rap_context.rap_shared_bo,
+				      &psp->rap_context.rap_shared_mc_addr,
+				      &psp->rap_context.rap_shared_buf);
+
+	return ret;
+}
+
+static int psp_rap_load(struct psp_context *psp)
+{
+	int ret;
+	struct psp_gfx_cmd_resp *cmd;
+
+	cmd = kzalloc(sizeof(struct psp_gfx_cmd_resp), GFP_KERNEL);
+	if (!cmd)
+		return -ENOMEM;
+
+	memset(psp->fw_pri_buf, 0, PSP_1_MEG);
+	memcpy(psp->fw_pri_buf, psp->ta_rap_start_addr, psp->ta_rap_ucode_size);
+
+	psp_prep_ta_load_cmd_buf(cmd,
+				 psp->fw_pri_mc_addr,
+				 psp->ta_rap_ucode_size,
+				 psp->rap_context.rap_shared_mc_addr,
+				 PSP_RAP_SHARED_MEM_SIZE);
+
+	ret = psp_cmd_submit_buf(psp, NULL, cmd, psp->fence_buf_mc_addr);
+
+	if (!ret) {
+		psp->rap_context.rap_initialized = true;
+		psp->rap_context.session_id = cmd->resp.session_id;
+		mutex_init(&psp->rap_context.mutex);
+	}
+
+	kfree(cmd);
+
+	return ret;
+}
+
+static int psp_rap_unload(struct psp_context *psp)
+{
+	int ret;
+	struct psp_gfx_cmd_resp *cmd;
+
+	cmd = kzalloc(sizeof(struct psp_gfx_cmd_resp), GFP_KERNEL);
+	if (!cmd)
+		return -ENOMEM;
+
+	psp_prep_ta_unload_cmd_buf(cmd, psp->rap_context.session_id);
+
+	ret = psp_cmd_submit_buf(psp, NULL, cmd, psp->fence_buf_mc_addr);
+
+	kfree(cmd);
+
+	return ret;
+}
+
+static int psp_rap_initialize(struct psp_context *psp)
+{
+	int ret;
+
+	/*
+	 * TODO: bypass the initialize in sriov for now
+	 */
+	if (amdgpu_sriov_vf(psp->adev))
+		return 0;
+
+	if (!psp->adev->psp.ta_rap_ucode_size ||
+	    !psp->adev->psp.ta_rap_start_addr) {
+		dev_info(psp->adev->dev, "RAP: optional rap ta ucode is not available\n");
+		return 0;
+	}
+
+	if (!psp->rap_context.rap_initialized) {
+		ret = psp_rap_init_shared_buf(psp);
+		if (ret)
+			return ret;
+	}
+
+	ret = psp_rap_load(psp);
+	if (ret)
+		return ret;
+
+	ret = psp_rap_invoke(psp, TA_CMD_RAP__INITIALIZE);
+	if (ret != TA_RAP_STATUS__SUCCESS) {
+		psp_rap_unload(psp);
+
+		amdgpu_bo_free_kernel(&psp->rap_context.rap_shared_bo,
+			      &psp->rap_context.rap_shared_mc_addr,
+			      &psp->rap_context.rap_shared_buf);
+
+		psp->rap_context.rap_initialized = false;
+
+		dev_warn(psp->adev->dev, "RAP TA initialize fail.\n");
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+static int psp_rap_terminate(struct psp_context *psp)
+{
+	int ret;
+
+	if (!psp->rap_context.rap_initialized)
+		return 0;
+
+	ret = psp_rap_unload(psp);
+
+	psp->rap_context.rap_initialized = false;
+
+	/* free rap shared memory */
+	amdgpu_bo_free_kernel(&psp->rap_context.rap_shared_bo,
+			      &psp->rap_context.rap_shared_mc_addr,
+			      &psp->rap_context.rap_shared_buf);
+
+	return ret;
+}
+
+int psp_rap_invoke(struct psp_context *psp, uint32_t ta_cmd_id)
+{
+	struct ta_rap_shared_memory *rap_cmd;
+	int ret;
+
+	if (!psp->rap_context.rap_initialized)
+		return -EINVAL;
+
+	if (ta_cmd_id != TA_CMD_RAP__INITIALIZE &&
+	    ta_cmd_id != TA_CMD_RAP__VALIDATE_L0)
+		return -EINVAL;
+
+	mutex_lock(&psp->rap_context.mutex);
+
+	rap_cmd = (struct ta_rap_shared_memory *)
+		  psp->rap_context.rap_shared_buf;
+	memset(rap_cmd, 0, sizeof(struct ta_rap_shared_memory));
+
+	rap_cmd->cmd_id = ta_cmd_id;
+	rap_cmd->validation_method_id = METHOD_A;
+
+	ret = psp_ta_invoke(psp, rap_cmd->cmd_id, psp->rap_context.session_id);
+	if (ret) {
+		mutex_unlock(&psp->rap_context.mutex);
+		return ret;
+	}
+
+	mutex_unlock(&psp->rap_context.mutex);
+
+	return rap_cmd->rap_status;
+}
+// RAP end
+
 static int psp_hw_start(struct psp_context *psp)
 {
 	struct amdgpu_device *adev = psp->adev;
@@ -1892,6 +2054,11 @@ static int psp_load_fw(struct amdgpu_device *adev)
 		if (ret)
 			dev_err(psp->adev->dev,
 				"DTM: Failed to initialize DTM\n");
+
+		ret = psp_rap_initialize(psp);
+		if (ret)
+			dev_err(psp->adev->dev,
+				"RAP: Failed to initialize RAP\n");
 	}
 
 	return 0;
@@ -1942,6 +2109,7 @@ static int psp_hw_fini(void *handle)
 
 	if (psp->adev->psp.ta_fw) {
 		psp_ras_terminate(psp);
+		psp_rap_terminate(psp);
 		psp_dtm_terminate(psp);
 		psp_hdcp_terminate(psp);
 	}
@@ -2000,6 +2168,11 @@ static int psp_suspend(void *handle)
 			DRM_ERROR("Failed to terminate dtm ta\n");
 			return ret;
 		}
+		ret = psp_rap_terminate(psp);
+		if (ret) {
+			DRM_ERROR("Failed to terminate rap ta\n");
+			return ret;
+		}
 	}
 
 	ret = psp_asd_unload(psp);
@@ -2078,6 +2251,11 @@ static int psp_resume(void *handle)
 		if (ret)
 			dev_err(psp->adev->dev,
 				"DTM: Failed to initialize DTM\n");
+
+		ret = psp_rap_initialize(psp);
+		if (ret)
+			dev_err(psp->adev->dev,
+				"RAP: Failed to initialize RAP\n");
 	}
 
 	mutex_unlock(&adev->firmware.mutex);
@@ -2343,6 +2521,11 @@ int parse_ta_bin_descriptor(struct psp_context *psp,
 		psp->ta_dtm_ucode_size     = le32_to_cpu(desc->size_bytes);
 		psp->ta_dtm_start_addr     = ucode_start_addr;
 		break;
+	case TA_FW_TYPE_PSP_RAP:
+		psp->ta_rap_ucode_version  = le32_to_cpu(desc->fw_version);
+		psp->ta_rap_ucode_size     = le32_to_cpu(desc->size_bytes);
+		psp->ta_rap_start_addr     = ucode_start_addr;
+		break;
 	default:
 		dev_warn(psp->adev->dev, "Unsupported TA type: %d\n", desc->fw_type);
 		break;

drivers/gpu/drm/amd/amdgpu/amdgpu_psp.h

@@ -29,6 +29,7 @@
 #include "psp_gfx_if.h"
 #include "ta_xgmi_if.h"
 #include "ta_ras_if.h"
+#include "ta_rap_if.h"
 
 #define PSP_FENCE_BUFFER_SIZE	0x1000
 #define PSP_CMD_BUFFER_SIZE	0x1000
@@ -38,6 +39,7 @@
 #define PSP_TMR_SIZE	0x400000
 #define PSP_HDCP_SHARED_MEM_SIZE	0x4000
 #define PSP_DTM_SHARED_MEM_SIZE	0x4000
+#define PSP_RAP_SHARED_MEM_SIZE	0x4000
 #define PSP_SHARED_MEM_SIZE		0x4000
 
 struct psp_context;
@@ -159,6 +161,15 @@ struct psp_dtm_context {
 	struct mutex		mutex;
 };
 
+struct psp_rap_context {
+	bool			rap_initialized;
+	uint32_t		session_id;
+	struct amdgpu_bo	*rap_shared_bo;
+	uint64_t		rap_shared_mc_addr;
+	void			*rap_shared_buf;
+	struct mutex		mutex;
+};
+
 #define MEM_TRAIN_SYSTEM_SIGNATURE		0x54534942
 #define GDDR6_MEM_TRAINING_DATA_SIZE_IN_BYTES	0x1000
 #define GDDR6_MEM_TRAINING_OFFSET		0x8000
@@ -277,11 +288,16 @@ struct psp_context
 	uint32_t			ta_dtm_ucode_size;
 	uint8_t				*ta_dtm_start_addr;
 
+	uint32_t			ta_rap_ucode_version;
+	uint32_t			ta_rap_ucode_size;
+	uint8_t				*ta_rap_start_addr;
+
 	struct psp_asd_context		asd_context;
 	struct psp_xgmi_context		xgmi_context;
 	struct psp_ras_context		ras;
 	struct psp_hdcp_context 	hdcp_context;
 	struct psp_dtm_context		dtm_context;
+	struct psp_rap_context		rap_context;
 	struct mutex			mutex;
 	struct psp_memory_training_context mem_train_ctx;
 };
@@ -357,6 +373,7 @@ int psp_ras_trigger_error(struct psp_context *psp,
 
 int psp_hdcp_invoke(struct psp_context *psp, uint32_t ta_cmd_id);
 int psp_dtm_invoke(struct psp_context *psp, uint32_t ta_cmd_id);
+int psp_rap_invoke(struct psp_context *psp, uint32_t ta_cmd_id);
 
 int psp_rlc_autoload_start(struct psp_context *psp);
 

drivers/gpu/drm/amd/amdgpu/amdgpu_ucode.h

@@ -131,6 +131,7 @@ enum ta_fw_type {
 	TA_FW_TYPE_PSP_RAS,
 	TA_FW_TYPE_PSP_HDCP,
 	TA_FW_TYPE_PSP_DTM,
+	TA_FW_TYPE_PSP_RAP,
 };
 
 struct ta_fw_bin_desc {