drm/amdgpu/jpeg: add support for jpeg DPG mode

Jpeg DPG support for GC IP v11_5_0
Signed-off-by: Saleemkhan Jamadar <saleemkhan.jamadar@amd.com>
Reviewed-by: Leo Liu <leo.liu@amd.com>
Reviewed-by: Veerabadhran Gopalakrishnan <Veerabadhran.Gopalakrishnan@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

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

@@ -36,10 +36,35 @@ static void amdgpu_jpeg_idle_work_handler(struct work_struct *work);
 
 int amdgpu_jpeg_sw_init(struct amdgpu_device *adev)
 {
+	int i, r;
+
 	INIT_DELAYED_WORK(&adev->jpeg.idle_work, amdgpu_jpeg_idle_work_handler);
 	mutex_init(&adev->jpeg.jpeg_pg_lock);
 	atomic_set(&adev->jpeg.total_submission_cnt, 0);
 
+	if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
+	    (adev->pg_flags & AMD_PG_SUPPORT_JPEG_DPG))
+		adev->jpeg.indirect_sram = true;
+
+	for (i = 0; i < adev->jpeg.num_jpeg_inst; i++) {
+		if (adev->jpeg.harvest_config & (1 << i))
+			continue;
+
+		if (adev->jpeg.indirect_sram) {
+			r = amdgpu_bo_create_kernel(adev, 64 * 2 * 4, PAGE_SIZE,
+					AMDGPU_GEM_DOMAIN_VRAM |
+					AMDGPU_GEM_DOMAIN_GTT,
+					&adev->jpeg.inst[i].dpg_sram_bo,
+					&adev->jpeg.inst[i].dpg_sram_gpu_addr,
+					&adev->jpeg.inst[i].dpg_sram_cpu_addr);
+			if (r) {
+				dev_err(adev->dev,
+				"JPEG %d (%d) failed to allocate DPG bo\n", i, r);
+				return r;
+			}
+		}
+	}
+
 	return 0;
 }
 
@@ -51,6 +76,11 @@ int amdgpu_jpeg_sw_fini(struct amdgpu_device *adev)
 		if (adev->jpeg.harvest_config & (1 << i))
 			continue;
 
+		amdgpu_bo_free_kernel(
+			&adev->jpeg.inst[i].dpg_sram_bo,
+			&adev->jpeg.inst[i].dpg_sram_gpu_addr,
+			(void **)&adev->jpeg.inst[i].dpg_sram_cpu_addr);
+
 		for (j = 0; j < adev->jpeg.num_jpeg_rings; ++j)
 			amdgpu_ring_fini(&adev->jpeg.inst[i].ring_dec[j]);
 	}
@@ -210,6 +240,7 @@ int amdgpu_jpeg_dec_ring_test_ib(struct amdgpu_ring *ring, long timeout)
 	} else {
 		r = 0;
 	}
+
 	if (!amdgpu_sriov_vf(adev)) {
 		for (i = 0; i < adev->usec_timeout; i++) {
 			tmp = RREG32(adev->jpeg.inst[ring->me].external.jpeg_pitch[ring->pipe]);
@@ -296,3 +327,16 @@ int amdgpu_jpeg_ras_sw_init(struct amdgpu_device *adev)
 
 	return 0;
 }
+
+int amdgpu_jpeg_psp_update_sram(struct amdgpu_device *adev, int inst_idx,
+			       enum AMDGPU_UCODE_ID ucode_id)
+{
+	struct amdgpu_firmware_info ucode = {
+		.ucode_id = AMDGPU_UCODE_ID_JPEG_RAM,
+		.mc_addr = adev->jpeg.inst[inst_idx].dpg_sram_gpu_addr,
+		.ucode_size = ((uintptr_t)adev->jpeg.inst[inst_idx].dpg_sram_curr_addr -
+			      (uintptr_t)adev->jpeg.inst[inst_idx].dpg_sram_cpu_addr),
+	};
+
+	return psp_execute_ip_fw_load(&adev->psp, &ucode);
+}

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

@@ -32,6 +32,34 @@
 #define AMDGPU_JPEG_HARVEST_JPEG0 (1 << 0)
 #define AMDGPU_JPEG_HARVEST_JPEG1 (1 << 1)
 
+#define WREG32_SOC15_JPEG_DPG_MODE(inst_idx, offset, value, indirect)			\
+	do {										\
+		if (!indirect) {							\
+			WREG32_SOC15(JPEG, GET_INST(JPEG, inst_idx),			\
+				     mmUVD_DPG_LMA_DATA, value);			\
+			WREG32_SOC15(							\
+				JPEG, GET_INST(JPEG, inst_idx),				\
+				mmUVD_DPG_LMA_CTL,					\
+				(UVD_DPG_LMA_CTL__READ_WRITE_MASK |			\
+				 offset << UVD_DPG_LMA_CTL__READ_WRITE_ADDR__SHIFT |	\
+				 indirect << UVD_DPG_LMA_CTL__SRAM_SEL__SHIFT));	\
+		} else {								\
+			*adev->jpeg.inst[inst_idx].dpg_sram_curr_addr++ =		\
+				offset;							\
+			*adev->jpeg.inst[inst_idx].dpg_sram_curr_addr++ =		\
+				value;							\
+		}									\
+	} while (0)
+
+#define RREG32_SOC15_JPEG_DPG_MODE(inst_idx, offset, mask_en)					\
+	({											\
+		WREG32_SOC15(JPEG, inst_idx, mmUVD_DPG_LMA_CTL,					\
+			(0x0 << UVD_DPG_LMA_CTL__READ_WRITE__SHIFT |				\
+			mask_en << UVD_DPG_LMA_CTL__MASK_EN__SHIFT |				\
+			offset << UVD_DPG_LMA_CTL__READ_WRITE_ADDR__SHIFT));			\
+		RREG32_SOC15(JPEG, inst_idx, mmUVD_DPG_LMA_DATA);				\
+	})
+
 struct amdgpu_jpeg_reg{
 	unsigned jpeg_pitch[AMDGPU_MAX_JPEG_RINGS];
 };
@@ -41,6 +69,11 @@ struct amdgpu_jpeg_inst {
 	struct amdgpu_irq_src irq;
 	struct amdgpu_irq_src ras_poison_irq;
 	struct amdgpu_jpeg_reg external;
+	struct amdgpu_bo	*dpg_sram_bo;
+	struct dpg_pause_state	pause_state;
+	void			*dpg_sram_cpu_addr;
+	uint64_t		dpg_sram_gpu_addr;
+	uint32_t		*dpg_sram_curr_addr;
 	uint8_t aid_id;
 };
 
@@ -63,6 +96,7 @@ struct amdgpu_jpeg {
 
 	uint16_t inst_mask;
 	uint8_t num_inst_per_aid;
+	bool	indirect_sram;
 };
 
 int amdgpu_jpeg_sw_init(struct amdgpu_device *adev);
@@ -82,5 +116,7 @@ int amdgpu_jpeg_process_poison_irq(struct amdgpu_device *adev,
 int amdgpu_jpeg_ras_late_init(struct amdgpu_device *adev,
 				struct ras_common_if *ras_block);
 int amdgpu_jpeg_ras_sw_init(struct amdgpu_device *adev);
+int amdgpu_jpeg_psp_update_sram(struct amdgpu_device *adev, int inst_idx,
+			       enum AMDGPU_UCODE_ID ucode_id);
 
 #endif /*__AMDGPU_JPEG_H__*/

drivers/gpu/drm/amd/amdgpu/soc21.c

@@ -711,6 +711,7 @@ static int soc21_common_early_init(void *handle)
 			AMD_CG_SUPPORT_BIF_MGCG |
 			AMD_CG_SUPPORT_BIF_LS;
 		adev->pg_flags = AMD_PG_SUPPORT_VCN_DPG |
+			AMD_PG_SUPPORT_JPEG_DPG |
 			AMD_PG_SUPPORT_VCN |
 			AMD_PG_SUPPORT_JPEG |
 			AMD_PG_SUPPORT_GFX_PG;

drivers/gpu/drm/amd/include/amd_shared.h

@@ -174,6 +174,7 @@ enum amd_powergating_state {
 #define AMD_PG_SUPPORT_ATHUB			(1 << 16)
 #define AMD_PG_SUPPORT_JPEG			(1 << 17)
 #define AMD_PG_SUPPORT_IH_SRAM_PG		(1 << 18)
+#define AMD_PG_SUPPORT_JPEG_DPG		(1 << 19)
 
 /**
  * enum PP_FEATURE_MASK - Used to mask power play features.