drm/amd/amdgpu: add uvd6.0 clock gating support. (v2)

v2: fix bug in register mask setting.

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Acked-by: Jammy Zhou <Jammy.Zhou@amd.com>
Signed-off-by: Eric Huang <JinHuiEric.Huang@amd.com>

drivers/gpu/drm/amd/amdgpu/uvd_v6_0.c

@@ -279,6 +279,234 @@ static void uvd_v6_0_mc_resume(struct amdgpu_device *adev)
 	WREG32(mmUVD_VCPU_CACHE_SIZE2, size);
 }
 
+static void cz_set_uvd_clock_gating_branches(struct amdgpu_device *adev,
+		bool enable)
+{
+	u32 data, data1;
+
+	data = RREG32(mmUVD_CGC_GATE);
+	data1 = RREG32(mmUVD_SUVD_CGC_GATE);
+	if (enable) {
+		data |= UVD_CGC_GATE__SYS_MASK |
+				UVD_CGC_GATE__UDEC_MASK |
+				UVD_CGC_GATE__MPEG2_MASK |
+				UVD_CGC_GATE__RBC_MASK |
+				UVD_CGC_GATE__LMI_MC_MASK |
+				UVD_CGC_GATE__IDCT_MASK |
+				UVD_CGC_GATE__MPRD_MASK |
+				UVD_CGC_GATE__MPC_MASK |
+				UVD_CGC_GATE__LBSI_MASK |
+				UVD_CGC_GATE__LRBBM_MASK |
+				UVD_CGC_GATE__UDEC_RE_MASK |
+				UVD_CGC_GATE__UDEC_CM_MASK |
+				UVD_CGC_GATE__UDEC_IT_MASK |
+				UVD_CGC_GATE__UDEC_DB_MASK |
+				UVD_CGC_GATE__UDEC_MP_MASK |
+				UVD_CGC_GATE__WCB_MASK |
+				UVD_CGC_GATE__VCPU_MASK |
+				UVD_CGC_GATE__SCPU_MASK;
+		data1 |= UVD_SUVD_CGC_GATE__SRE_MASK |
+				UVD_SUVD_CGC_GATE__SIT_MASK |
+				UVD_SUVD_CGC_GATE__SMP_MASK |
+				UVD_SUVD_CGC_GATE__SCM_MASK |
+				UVD_SUVD_CGC_GATE__SDB_MASK |
+				UVD_SUVD_CGC_GATE__SRE_H264_MASK |
+				UVD_SUVD_CGC_GATE__SRE_HEVC_MASK |
+				UVD_SUVD_CGC_GATE__SIT_H264_MASK |
+				UVD_SUVD_CGC_GATE__SIT_HEVC_MASK |
+				UVD_SUVD_CGC_GATE__SCM_H264_MASK |
+				UVD_SUVD_CGC_GATE__SCM_HEVC_MASK |
+				UVD_SUVD_CGC_GATE__SDB_H264_MASK |
+				UVD_SUVD_CGC_GATE__SDB_HEVC_MASK;
+	} else {
+		data &= ~(UVD_CGC_GATE__SYS_MASK |
+				UVD_CGC_GATE__UDEC_MASK |
+				UVD_CGC_GATE__MPEG2_MASK |
+				UVD_CGC_GATE__RBC_MASK |
+				UVD_CGC_GATE__LMI_MC_MASK |
+				UVD_CGC_GATE__LMI_UMC_MASK |
+				UVD_CGC_GATE__IDCT_MASK |
+				UVD_CGC_GATE__MPRD_MASK |
+				UVD_CGC_GATE__MPC_MASK |
+				UVD_CGC_GATE__LBSI_MASK |
+				UVD_CGC_GATE__LRBBM_MASK |
+				UVD_CGC_GATE__UDEC_RE_MASK |
+				UVD_CGC_GATE__UDEC_CM_MASK |
+				UVD_CGC_GATE__UDEC_IT_MASK |
+				UVD_CGC_GATE__UDEC_DB_MASK |
+				UVD_CGC_GATE__UDEC_MP_MASK |
+				UVD_CGC_GATE__WCB_MASK |
+				UVD_CGC_GATE__VCPU_MASK |
+				UVD_CGC_GATE__SCPU_MASK);
+		data1 &= ~(UVD_SUVD_CGC_GATE__SRE_MASK |
+				UVD_SUVD_CGC_GATE__SIT_MASK |
+				UVD_SUVD_CGC_GATE__SMP_MASK |
+				UVD_SUVD_CGC_GATE__SCM_MASK |
+				UVD_SUVD_CGC_GATE__SDB_MASK |
+				UVD_SUVD_CGC_GATE__SRE_H264_MASK |
+				UVD_SUVD_CGC_GATE__SRE_HEVC_MASK |
+				UVD_SUVD_CGC_GATE__SIT_H264_MASK |
+				UVD_SUVD_CGC_GATE__SIT_HEVC_MASK |
+				UVD_SUVD_CGC_GATE__SCM_H264_MASK |
+				UVD_SUVD_CGC_GATE__SCM_HEVC_MASK |
+				UVD_SUVD_CGC_GATE__SDB_H264_MASK |
+				UVD_SUVD_CGC_GATE__SDB_HEVC_MASK);
+	}
+	WREG32(mmUVD_CGC_GATE, data);
+	WREG32(mmUVD_SUVD_CGC_GATE, data1);
+}
+
+static void tonga_set_uvd_clock_gating_branches(struct amdgpu_device *adev,
+		bool enable)
+{
+	u32 data, data1;
+
+	data = RREG32(mmUVD_CGC_GATE);
+	data1 = RREG32(mmUVD_SUVD_CGC_GATE);
+	if (enable) {
+		data |= UVD_CGC_GATE__SYS_MASK |
+				UVD_CGC_GATE__UDEC_MASK |
+				UVD_CGC_GATE__MPEG2_MASK |
+				UVD_CGC_GATE__RBC_MASK |
+				UVD_CGC_GATE__LMI_MC_MASK |
+				UVD_CGC_GATE__IDCT_MASK |
+				UVD_CGC_GATE__MPRD_MASK |
+				UVD_CGC_GATE__MPC_MASK |
+				UVD_CGC_GATE__LBSI_MASK |
+				UVD_CGC_GATE__LRBBM_MASK |
+				UVD_CGC_GATE__UDEC_RE_MASK |
+				UVD_CGC_GATE__UDEC_CM_MASK |
+				UVD_CGC_GATE__UDEC_IT_MASK |
+				UVD_CGC_GATE__UDEC_DB_MASK |
+				UVD_CGC_GATE__UDEC_MP_MASK |
+				UVD_CGC_GATE__WCB_MASK |
+				UVD_CGC_GATE__VCPU_MASK |
+				UVD_CGC_GATE__SCPU_MASK;
+		data1 |= UVD_SUVD_CGC_GATE__SRE_MASK |
+				UVD_SUVD_CGC_GATE__SIT_MASK |
+				UVD_SUVD_CGC_GATE__SMP_MASK |
+				UVD_SUVD_CGC_GATE__SCM_MASK |
+				UVD_SUVD_CGC_GATE__SDB_MASK;
+	} else {
+		data &= ~(UVD_CGC_GATE__SYS_MASK |
+				UVD_CGC_GATE__UDEC_MASK |
+				UVD_CGC_GATE__MPEG2_MASK |
+				UVD_CGC_GATE__RBC_MASK |
+				UVD_CGC_GATE__LMI_MC_MASK |
+				UVD_CGC_GATE__LMI_UMC_MASK |
+				UVD_CGC_GATE__IDCT_MASK |
+				UVD_CGC_GATE__MPRD_MASK |
+				UVD_CGC_GATE__MPC_MASK |
+				UVD_CGC_GATE__LBSI_MASK |
+				UVD_CGC_GATE__LRBBM_MASK |
+				UVD_CGC_GATE__UDEC_RE_MASK |
+				UVD_CGC_GATE__UDEC_CM_MASK |
+				UVD_CGC_GATE__UDEC_IT_MASK |
+				UVD_CGC_GATE__UDEC_DB_MASK |
+				UVD_CGC_GATE__UDEC_MP_MASK |
+				UVD_CGC_GATE__WCB_MASK |
+				UVD_CGC_GATE__VCPU_MASK |
+				UVD_CGC_GATE__SCPU_MASK);
+		data1 &= ~(UVD_SUVD_CGC_GATE__SRE_MASK |
+				UVD_SUVD_CGC_GATE__SIT_MASK |
+				UVD_SUVD_CGC_GATE__SMP_MASK |
+				UVD_SUVD_CGC_GATE__SCM_MASK |
+				UVD_SUVD_CGC_GATE__SDB_MASK);
+	}
+	WREG32(mmUVD_CGC_GATE, data);
+	WREG32(mmUVD_SUVD_CGC_GATE, data1);
+}
+
+static void uvd_v6_0_set_uvd_dynamic_clock_mode(struct amdgpu_device *adev,
+		bool swmode)
+{
+	u32 data, data1 = 0, data2;
+
+	/* Always un-gate UVD REGS bit */
+	data = RREG32(mmUVD_CGC_GATE);
+	data &= ~(UVD_CGC_GATE__REGS_MASK);
+	WREG32(mmUVD_CGC_GATE, data);
+
+	data = RREG32(mmUVD_CGC_CTRL);
+	data &= ~(UVD_CGC_CTRL__CLK_OFF_DELAY_MASK |
+			UVD_CGC_CTRL__CLK_GATE_DLY_TIMER_MASK);
+	data |= UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK |
+			1 << REG_FIELD_SHIFT(UVD_CGC_CTRL, CLK_GATE_DLY_TIMER) |
+			4 << REG_FIELD_SHIFT(UVD_CGC_CTRL, CLK_OFF_DELAY);
+
+	data2 = RREG32(mmUVD_SUVD_CGC_CTRL);
+	if (swmode) {
+		data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK |
+				UVD_CGC_CTRL__UDEC_CM_MODE_MASK |
+				UVD_CGC_CTRL__UDEC_IT_MODE_MASK |
+				UVD_CGC_CTRL__UDEC_DB_MODE_MASK |
+				UVD_CGC_CTRL__UDEC_MP_MODE_MASK |
+				UVD_CGC_CTRL__SYS_MODE_MASK |
+				UVD_CGC_CTRL__UDEC_MODE_MASK |
+				UVD_CGC_CTRL__MPEG2_MODE_MASK |
+				UVD_CGC_CTRL__REGS_MODE_MASK |
+				UVD_CGC_CTRL__RBC_MODE_MASK |
+				UVD_CGC_CTRL__LMI_MC_MODE_MASK |
+				UVD_CGC_CTRL__LMI_UMC_MODE_MASK |
+				UVD_CGC_CTRL__IDCT_MODE_MASK |
+				UVD_CGC_CTRL__MPRD_MODE_MASK |
+				UVD_CGC_CTRL__MPC_MODE_MASK |
+				UVD_CGC_CTRL__LBSI_MODE_MASK |
+				UVD_CGC_CTRL__LRBBM_MODE_MASK |
+				UVD_CGC_CTRL__WCB_MODE_MASK |
+				UVD_CGC_CTRL__VCPU_MODE_MASK |
+				UVD_CGC_CTRL__JPEG_MODE_MASK |
+				UVD_CGC_CTRL__SCPU_MODE_MASK);
+		data1 |= UVD_CGC_CTRL2__DYN_OCLK_RAMP_EN_MASK |
+				UVD_CGC_CTRL2__DYN_RCLK_RAMP_EN_MASK;
+		data1 &= ~UVD_CGC_CTRL2__GATER_DIV_ID_MASK;
+		data1 |= 7 << REG_FIELD_SHIFT(UVD_CGC_CTRL2, GATER_DIV_ID);
+		data2 &= ~(UVD_SUVD_CGC_CTRL__SRE_MODE_MASK |
+				UVD_SUVD_CGC_CTRL__SIT_MODE_MASK |
+				UVD_SUVD_CGC_CTRL__SMP_MODE_MASK |
+				UVD_SUVD_CGC_CTRL__SCM_MODE_MASK |
+				UVD_SUVD_CGC_CTRL__SDB_MODE_MASK);
+	} else {
+		data |= UVD_CGC_CTRL__UDEC_RE_MODE_MASK |
+				UVD_CGC_CTRL__UDEC_CM_MODE_MASK |
+				UVD_CGC_CTRL__UDEC_IT_MODE_MASK |
+				UVD_CGC_CTRL__UDEC_DB_MODE_MASK |
+				UVD_CGC_CTRL__UDEC_MP_MODE_MASK |
+				UVD_CGC_CTRL__SYS_MODE_MASK |
+				UVD_CGC_CTRL__UDEC_MODE_MASK |
+				UVD_CGC_CTRL__MPEG2_MODE_MASK |
+				UVD_CGC_CTRL__REGS_MODE_MASK |
+				UVD_CGC_CTRL__RBC_MODE_MASK |
+				UVD_CGC_CTRL__LMI_MC_MODE_MASK |
+				UVD_CGC_CTRL__LMI_UMC_MODE_MASK |
+				UVD_CGC_CTRL__IDCT_MODE_MASK |
+				UVD_CGC_CTRL__MPRD_MODE_MASK |
+				UVD_CGC_CTRL__MPC_MODE_MASK |
+				UVD_CGC_CTRL__LBSI_MODE_MASK |
+				UVD_CGC_CTRL__LRBBM_MODE_MASK |
+				UVD_CGC_CTRL__WCB_MODE_MASK |
+				UVD_CGC_CTRL__VCPU_MODE_MASK |
+				UVD_CGC_CTRL__SCPU_MODE_MASK;
+		data2 |= UVD_SUVD_CGC_CTRL__SRE_MODE_MASK |
+				UVD_SUVD_CGC_CTRL__SIT_MODE_MASK |
+				UVD_SUVD_CGC_CTRL__SMP_MODE_MASK |
+				UVD_SUVD_CGC_CTRL__SCM_MODE_MASK |
+				UVD_SUVD_CGC_CTRL__SDB_MODE_MASK;
+	}
+	WREG32(mmUVD_CGC_CTRL, data);
+	WREG32(mmUVD_SUVD_CGC_CTRL, data2);
+
+	data = RREG32_UVD_CTX(ixUVD_CGC_CTRL2);
+	data &= ~(REG_FIELD_MASK(UVD_CGC_CTRL2, DYN_OCLK_RAMP_EN) |
+			REG_FIELD_MASK(UVD_CGC_CTRL2, DYN_RCLK_RAMP_EN) |
+			REG_FIELD_MASK(UVD_CGC_CTRL2, GATER_DIV_ID));
+	data1 &= (REG_FIELD_MASK(UVD_CGC_CTRL2, DYN_OCLK_RAMP_EN) |
+			REG_FIELD_MASK(UVD_CGC_CTRL2, DYN_RCLK_RAMP_EN) |
+			REG_FIELD_MASK(UVD_CGC_CTRL2, GATER_DIV_ID));
+	data |= data1;
+	WREG32_UVD_CTX(ixUVD_CGC_CTRL2, data);
+}
+
 /**
  * uvd_v6_0_start - start UVD block
  *
@@ -303,8 +531,19 @@ static int uvd_v6_0_start(struct amdgpu_device *adev)
 
 	uvd_v6_0_mc_resume(adev);
 
-	/* disable clock gating */
-	WREG32(mmUVD_CGC_GATE, 0);
+	/* Set dynamic clock gating in S/W control mode */
+	if (adev->cg_flags & AMDGPU_CG_SUPPORT_UVD_MGCG) {
+		if (adev->flags & AMD_IS_APU)
+			cz_set_uvd_clock_gating_branches(adev, false);
+		else
+			tonga_set_uvd_clock_gating_branches(adev, false);
+		uvd_v6_0_set_uvd_dynamic_clock_mode(adev, true);
+	} else {
+		/* disable clock gating */
+		uint32_t data = RREG32(mmUVD_CGC_CTRL);
+		data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
+		WREG32(mmUVD_CGC_CTRL, data);
+	}
 
 	/* disable interupt */
 	WREG32_P(mmUVD_MASTINT_EN, 0, ~(1 << 1));
@@ -758,6 +997,24 @@ static int uvd_v6_0_process_interrupt(struct amdgpu_device *adev,
 static int uvd_v6_0_set_clockgating_state(void *handle,
 					  enum amd_clockgating_state state)
 {
+	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+	bool enable = (state == AMD_CG_STATE_GATE) ? true : false;
+
+	if (!(adev->cg_flags & AMDGPU_CG_SUPPORT_UVD_MGCG))
+		return 0;
+
+	if (enable) {
+		if (adev->flags & AMD_IS_APU)
+			cz_set_uvd_clock_gating_branches(adev, enable);
+		else
+			tonga_set_uvd_clock_gating_branches(adev, enable);
+		uvd_v6_0_set_uvd_dynamic_clock_mode(adev, true);
+	} else {
+		uint32_t data = RREG32(mmUVD_CGC_CTRL);
+		data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
+		WREG32(mmUVD_CGC_CTRL, data);
+	}
+
 	return 0;
 }