1 files changed, 124 insertions, 121 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/vce_v3_0.c b/drivers/gpu/drm/amd/amdgpu/vce_v3_0.c
index ce468ee5da2a..c271abffd8dd 100644
--- a/drivers/gpu/drm/amd/amdgpu/vce_v3_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/vce_v3_0.c
@@ -40,9 +40,10 @@
 
 #define GRBM_GFX_INDEX__VCE_INSTANCE__SHIFT	0x04
 #define GRBM_GFX_INDEX__VCE_INSTANCE_MASK	0x10
-#define mmVCE_LMI_VCPU_CACHE_40BIT_BAR0 	0x8616
-#define mmVCE_LMI_VCPU_CACHE_40BIT_BAR1 	0x8617
-#define mmVCE_LMI_VCPU_CACHE_40BIT_BAR2 	0x8618
+#define mmVCE_LMI_VCPU_CACHE_40BIT_BAR0	0x8616
+#define mmVCE_LMI_VCPU_CACHE_40BIT_BAR1	0x8617
+#define mmVCE_LMI_VCPU_CACHE_40BIT_BAR2	0x8618
+#define VCE_STATUS_VCPU_REPORT_FW_LOADED_MASK	0x02
 
 #define VCE_V3_0_FW_SIZE	(384 * 1024)
 #define VCE_V3_0_STACK_SIZE	(64 * 1024)
@@ -51,6 +52,7 @@
 static void vce_v3_0_mc_resume(struct amdgpu_device *adev, int idx);
 static void vce_v3_0_set_ring_funcs(struct amdgpu_device *adev);
 static void vce_v3_0_set_irq_funcs(struct amdgpu_device *adev);
+static int vce_v3_0_wait_for_idle(void *handle);
 
 /**
  * vce_v3_0_ring_get_rptr - get read pointer
@@ -205,6 +207,32 @@ static void vce_v3_0_set_vce_sw_clock_gating(struct amdgpu_device *adev,
 	vce_v3_0_override_vce_clock_gating(adev, false);
 }
 
+static int vce_v3_0_firmware_loaded(struct amdgpu_device *adev)
+{
+	int i, j;
+
+	for (i = 0; i < 10; ++i) {
+		for (j = 0; j < 100; ++j) {
+			uint32_t status = RREG32(mmVCE_STATUS);
+
+			if (status & VCE_STATUS_VCPU_REPORT_FW_LOADED_MASK)
+				return 0;
+			mdelay(10);
+		}
+
+		DRM_ERROR("VCE not responding, trying to reset the ECPU!!!\n");
+		WREG32_P(mmVCE_SOFT_RESET,
+			VCE_SOFT_RESET__ECPU_SOFT_RESET_MASK,
+			~VCE_SOFT_RESET__ECPU_SOFT_RESET_MASK);
+		mdelay(10);
+		WREG32_P(mmVCE_SOFT_RESET, 0,
+			~VCE_SOFT_RESET__ECPU_SOFT_RESET_MASK);
+		mdelay(10);
+	}
+
+	return -ETIMEDOUT;
+}
+
 /**
  * vce_v3_0_start - start VCE block
  *
@@ -215,11 +243,24 @@ static void vce_v3_0_set_vce_sw_clock_gating(struct amdgpu_device *adev,
 static int vce_v3_0_start(struct amdgpu_device *adev)
 {
 	struct amdgpu_ring *ring;
-	int idx, i, j, r;
+	int idx, r;
+
+	ring = &adev->vce.ring[0];
+	WREG32(mmVCE_RB_RPTR, ring->wptr);
+	WREG32(mmVCE_RB_WPTR, ring->wptr);
+	WREG32(mmVCE_RB_BASE_LO, ring->gpu_addr);
+	WREG32(mmVCE_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
+	WREG32(mmVCE_RB_SIZE, ring->ring_size / 4);
+
+	ring = &adev->vce.ring[1];
+	WREG32(mmVCE_RB_RPTR2, ring->wptr);
+	WREG32(mmVCE_RB_WPTR2, ring->wptr);
+	WREG32(mmVCE_RB_BASE_LO2, ring->gpu_addr);
+	WREG32(mmVCE_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
+	WREG32(mmVCE_RB_SIZE2, ring->ring_size / 4);
 
 	mutex_lock(&adev->grbm_idx_mutex);
 	for (idx = 0; idx < 2; ++idx) {
-
 		if (adev->vce.harvest_config & (1 << idx))
 			continue;
 
@@ -233,48 +274,24 @@ static int vce_v3_0_start(struct amdgpu_device *adev)
 
 		vce_v3_0_mc_resume(adev, idx);
 
-		/* set BUSY flag */
-		WREG32_P(mmVCE_STATUS, 1, ~1);
+		WREG32_P(mmVCE_STATUS, VCE_STATUS__JOB_BUSY_MASK,
+		         ~VCE_STATUS__JOB_BUSY_MASK);
+
 		if (adev->asic_type >= CHIP_STONEY)
 			WREG32_P(mmVCE_VCPU_CNTL, 1, ~0x200001);
 		else
 			WREG32_P(mmVCE_VCPU_CNTL, VCE_VCPU_CNTL__CLK_EN_MASK,
 				~VCE_VCPU_CNTL__CLK_EN_MASK);
 
-		WREG32_P(mmVCE_SOFT_RESET,
-			 VCE_SOFT_RESET__ECPU_SOFT_RESET_MASK,
-			 ~VCE_SOFT_RESET__ECPU_SOFT_RESET_MASK);
-
-		mdelay(100);
-
 		WREG32_P(mmVCE_SOFT_RESET, 0,
 			~VCE_SOFT_RESET__ECPU_SOFT_RESET_MASK);
 
-		for (i = 0; i < 10; ++i) {
-			uint32_t status;
-			for (j = 0; j < 100; ++j) {
-				status = RREG32(mmVCE_STATUS);
-				if (status & 2)
-					break;
-				mdelay(10);
-			}
-			r = 0;
-			if (status & 2)
-				break;
-
-			DRM_ERROR("VCE not responding, trying to reset the ECPU!!!\n");
-			WREG32_P(mmVCE_SOFT_RESET,
-				VCE_SOFT_RESET__ECPU_SOFT_RESET_MASK,
-				~VCE_SOFT_RESET__ECPU_SOFT_RESET_MASK);
-			mdelay(10);
-			WREG32_P(mmVCE_SOFT_RESET, 0,
-				~VCE_SOFT_RESET__ECPU_SOFT_RESET_MASK);
-			mdelay(10);
-			r = -1;
-		}
+		mdelay(100);
+
+		r = vce_v3_0_firmware_loaded(adev);
 
 		/* clear BUSY flag */
-		WREG32_P(mmVCE_STATUS, 0, ~1);
+		WREG32_P(mmVCE_STATUS, 0, ~VCE_STATUS__JOB_BUSY_MASK);
 
 		/* Set Clock-Gating off */
 		if (adev->cg_flags & AMD_CG_SUPPORT_VCE_MGCG)
@@ -290,19 +307,46 @@ static int vce_v3_0_start(struct amdgpu_device *adev)
 	WREG32_P(mmGRBM_GFX_INDEX, 0, ~GRBM_GFX_INDEX__VCE_INSTANCE_MASK);
 	mutex_unlock(&adev->grbm_idx_mutex);
 
-	ring = &adev->vce.ring[0];
-	WREG32(mmVCE_RB_RPTR, ring->wptr);
-	WREG32(mmVCE_RB_WPTR, ring->wptr);
-	WREG32(mmVCE_RB_BASE_LO, ring->gpu_addr);
-	WREG32(mmVCE_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
-	WREG32(mmVCE_RB_SIZE, ring->ring_size / 4);
+	return 0;
+}
 
-	ring = &adev->vce.ring[1];
-	WREG32(mmVCE_RB_RPTR2, ring->wptr);
-	WREG32(mmVCE_RB_WPTR2, ring->wptr);
-	WREG32(mmVCE_RB_BASE_LO2, ring->gpu_addr);
-	WREG32(mmVCE_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
-	WREG32(mmVCE_RB_SIZE2, ring->ring_size / 4);
+static int vce_v3_0_stop(struct amdgpu_device *adev)
+{
+	int idx;
+
+	mutex_lock(&adev->grbm_idx_mutex);
+	for (idx = 0; idx < 2; ++idx) {
+		if (adev->vce.harvest_config & (1 << idx))
+			continue;
+
+		if (idx == 0)
+			WREG32_P(mmGRBM_GFX_INDEX, 0,
+				~GRBM_GFX_INDEX__VCE_INSTANCE_MASK);
+		else
+			WREG32_P(mmGRBM_GFX_INDEX,
+				GRBM_GFX_INDEX__VCE_INSTANCE_MASK,
+				~GRBM_GFX_INDEX__VCE_INSTANCE_MASK);
+
+		if (adev->asic_type >= CHIP_STONEY)
+			WREG32_P(mmVCE_VCPU_CNTL, 0, ~0x200001);
+		else
+			WREG32_P(mmVCE_VCPU_CNTL, 0,
+				~VCE_VCPU_CNTL__CLK_EN_MASK);
+		/* hold on ECPU */
+		WREG32_P(mmVCE_SOFT_RESET,
+			 VCE_SOFT_RESET__ECPU_SOFT_RESET_MASK,
+			 ~VCE_SOFT_RESET__ECPU_SOFT_RESET_MASK);
+
+		/* clear BUSY flag */
+		WREG32_P(mmVCE_STATUS, 0, ~VCE_STATUS__JOB_BUSY_MASK);
+
+		/* Set Clock-Gating off */
+		if (adev->cg_flags & AMD_CG_SUPPORT_VCE_MGCG)
+			vce_v3_0_set_vce_sw_clock_gating(adev, false);
+	}
+
+	WREG32_P(mmGRBM_GFX_INDEX, 0, ~GRBM_GFX_INDEX__VCE_INSTANCE_MASK);
+	mutex_unlock(&adev->grbm_idx_mutex);
 
 	return 0;
 }
@@ -315,9 +359,11 @@ static unsigned vce_v3_0_get_harvest_config(struct amdgpu_device *adev)
 {
 	u32 tmp;
 
-	/* Fiji, Stoney are single pipe */
+	/* Fiji, Stoney, Polaris10, Polaris11 are single pipe */
 	if ((adev->asic_type == CHIP_FIJI) ||
-	    (adev->asic_type == CHIP_STONEY))
+	    (adev->asic_type == CHIP_STONEY) ||
+	    (adev->asic_type == CHIP_POLARIS10) ||
+	    (adev->asic_type == CHIP_POLARIS11))
 		return AMDGPU_VCE_HARVEST_VCE1;
 
 	/* Tonga and CZ are dual or single pipe */
@@ -381,14 +427,14 @@ static int vce_v3_0_sw_init(void *handle)
 
 	ring = &adev->vce.ring[0];
 	sprintf(ring->name, "vce0");
-	r = amdgpu_ring_init(adev, ring, 4096, VCE_CMD_NO_OP, 0xf,
+	r = amdgpu_ring_init(adev, ring, 512, VCE_CMD_NO_OP, 0xf,
 			     &adev->vce.irq, 0, AMDGPU_RING_TYPE_VCE);
 	if (r)
 		return r;
 
 	ring = &adev->vce.ring[1];
 	sprintf(ring->name, "vce1");
-	r = amdgpu_ring_init(adev, ring, 4096, VCE_CMD_NO_OP, 0xf,
+	r = amdgpu_ring_init(adev, ring, 512, VCE_CMD_NO_OP, 0xf,
 			     &adev->vce.irq, 0, AMDGPU_RING_TYPE_VCE);
 	if (r)
 		return r;
@@ -439,7 +485,14 @@ static int vce_v3_0_hw_init(void *handle)
 
 static int vce_v3_0_hw_fini(void *handle)
 {
-	return 0;
+	int r;
+	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+
+	r = vce_v3_0_wait_for_idle(handle);
+	if (r)
+		return r;
+
+	return vce_v3_0_stop(adev);
 }
 
 static int vce_v3_0_suspend(void *handle)
@@ -564,73 +617,6 @@ static int vce_v3_0_soft_reset(void *handle)
 	return vce_v3_0_start(adev);
 }
 
-static void vce_v3_0_print_status(void *handle)
-{
-	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
-
-	dev_info(adev->dev, "VCE 3.0 registers\n");
-	dev_info(adev->dev, "  VCE_STATUS=0x%08X\n",
-		 RREG32(mmVCE_STATUS));
-	dev_info(adev->dev, "  VCE_VCPU_CNTL=0x%08X\n",
-		 RREG32(mmVCE_VCPU_CNTL));
-	dev_info(adev->dev, "  VCE_VCPU_CACHE_OFFSET0=0x%08X\n",
-		 RREG32(mmVCE_VCPU_CACHE_OFFSET0));
-	dev_info(adev->dev, "  VCE_VCPU_CACHE_SIZE0=0x%08X\n",
-		 RREG32(mmVCE_VCPU_CACHE_SIZE0));
-	dev_info(adev->dev, "  VCE_VCPU_CACHE_OFFSET1=0x%08X\n",
-		 RREG32(mmVCE_VCPU_CACHE_OFFSET1));
-	dev_info(adev->dev, "  VCE_VCPU_CACHE_SIZE1=0x%08X\n",
-		 RREG32(mmVCE_VCPU_CACHE_SIZE1));
-	dev_info(adev->dev, "  VCE_VCPU_CACHE_OFFSET2=0x%08X\n",
-		 RREG32(mmVCE_VCPU_CACHE_OFFSET2));
-	dev_info(adev->dev, "  VCE_VCPU_CACHE_SIZE2=0x%08X\n",
-		 RREG32(mmVCE_VCPU_CACHE_SIZE2));
-	dev_info(adev->dev, "  VCE_SOFT_RESET=0x%08X\n",
-		 RREG32(mmVCE_SOFT_RESET));
-	dev_info(adev->dev, "  VCE_RB_BASE_LO2=0x%08X\n",
-		 RREG32(mmVCE_RB_BASE_LO2));
-	dev_info(adev->dev, "  VCE_RB_BASE_HI2=0x%08X\n",
-		 RREG32(mmVCE_RB_BASE_HI2));
-	dev_info(adev->dev, "  VCE_RB_SIZE2=0x%08X\n",
-		 RREG32(mmVCE_RB_SIZE2));
-	dev_info(adev->dev, "  VCE_RB_RPTR2=0x%08X\n",
-		 RREG32(mmVCE_RB_RPTR2));
-	dev_info(adev->dev, "  VCE_RB_WPTR2=0x%08X\n",
-		 RREG32(mmVCE_RB_WPTR2));
-	dev_info(adev->dev, "  VCE_RB_BASE_LO=0x%08X\n",
-		 RREG32(mmVCE_RB_BASE_LO));
-	dev_info(adev->dev, "  VCE_RB_BASE_HI=0x%08X\n",
-		 RREG32(mmVCE_RB_BASE_HI));
-	dev_info(adev->dev, "  VCE_RB_SIZE=0x%08X\n",
-		 RREG32(mmVCE_RB_SIZE));
-	dev_info(adev->dev, "  VCE_RB_RPTR=0x%08X\n",
-		 RREG32(mmVCE_RB_RPTR));
-	dev_info(adev->dev, "  VCE_RB_WPTR=0x%08X\n",
-		 RREG32(mmVCE_RB_WPTR));
-	dev_info(adev->dev, "  VCE_CLOCK_GATING_A=0x%08X\n",
-		 RREG32(mmVCE_CLOCK_GATING_A));
-	dev_info(adev->dev, "  VCE_CLOCK_GATING_B=0x%08X\n",
-		 RREG32(mmVCE_CLOCK_GATING_B));
-	dev_info(adev->dev, "  VCE_UENC_CLOCK_GATING=0x%08X\n",
-		 RREG32(mmVCE_UENC_CLOCK_GATING));
-	dev_info(adev->dev, "  VCE_UENC_REG_CLOCK_GATING=0x%08X\n",
-		 RREG32(mmVCE_UENC_REG_CLOCK_GATING));
-	dev_info(adev->dev, "  VCE_SYS_INT_EN=0x%08X\n",
-		 RREG32(mmVCE_SYS_INT_EN));
-	dev_info(adev->dev, "  VCE_LMI_CTRL2=0x%08X\n",
-		 RREG32(mmVCE_LMI_CTRL2));
-	dev_info(adev->dev, "  VCE_LMI_CTRL=0x%08X\n",
-		 RREG32(mmVCE_LMI_CTRL));
-	dev_info(adev->dev, "  VCE_LMI_VM_CTRL=0x%08X\n",
-		 RREG32(mmVCE_LMI_VM_CTRL));
-	dev_info(adev->dev, "  VCE_LMI_SWAP_CNTL=0x%08X\n",
-		 RREG32(mmVCE_LMI_SWAP_CNTL));
-	dev_info(adev->dev, "  VCE_LMI_SWAP_CNTL1=0x%08X\n",
-		 RREG32(mmVCE_LMI_SWAP_CNTL1));
-	dev_info(adev->dev, "  VCE_LMI_CACHE_CTRL=0x%08X\n",
-		 RREG32(mmVCE_LMI_CACHE_CTRL));
-}
-
 static int vce_v3_0_set_interrupt_state(struct amdgpu_device *adev,
 					struct amdgpu_irq_src *source,
 					unsigned type,
@@ -669,6 +655,18 @@ static int vce_v3_0_process_interrupt(struct amdgpu_device *adev,
 	return 0;
 }
 
+static void vce_v3_set_bypass_mode(struct amdgpu_device *adev, bool enable)
+{
+	u32 tmp = RREG32_SMC(ixGCK_DFS_BYPASS_CNTL);
+
+	if (enable)
+		tmp |= GCK_DFS_BYPASS_CNTL__BYPASSECLK_MASK;
+	else
+		tmp &= ~GCK_DFS_BYPASS_CNTL__BYPASSECLK_MASK;
+
+	WREG32_SMC(ixGCK_DFS_BYPASS_CNTL, tmp);
+}
+
 static int vce_v3_0_set_clockgating_state(void *handle,
 					  enum amd_clockgating_state state)
 {
@@ -676,6 +674,9 @@ static int vce_v3_0_set_clockgating_state(void *handle,
 	bool enable = (state == AMD_CG_STATE_GATE) ? true : false;
 	int i;
 
+	if (adev->asic_type == CHIP_POLARIS10)
+		vce_v3_set_bypass_mode(adev, enable);
+
 	if (!(adev->cg_flags & AMD_CG_SUPPORT_VCE_MGCG))
 		return 0;
 
@@ -739,6 +740,7 @@ static int vce_v3_0_set_powergating_state(void *handle,
 }
 
 const struct amd_ip_funcs vce_v3_0_ip_funcs = {
+	.name = "vce_v3_0",
 	.early_init = vce_v3_0_early_init,
 	.late_init = NULL,
 	.sw_init = vce_v3_0_sw_init,
@@ -750,7 +752,6 @@ const struct amd_ip_funcs vce_v3_0_ip_funcs = {
 	.is_idle = vce_v3_0_is_idle,
 	.wait_for_idle = vce_v3_0_wait_for_idle,
 	.soft_reset = vce_v3_0_soft_reset,
-	.print_status = vce_v3_0_print_status,
 	.set_clockgating_state = vce_v3_0_set_clockgating_state,
 	.set_powergating_state = vce_v3_0_set_powergating_state,
 };
@@ -766,6 +767,8 @@ static const struct amdgpu_ring_funcs vce_v3_0_ring_funcs = {
 	.test_ib = amdgpu_vce_ring_test_ib,
 	.insert_nop = amdgpu_ring_insert_nop,
 	.pad_ib = amdgpu_ring_generic_pad_ib,
+	.begin_use = amdgpu_vce_ring_begin_use,
+	.end_use = amdgpu_vce_ring_end_use,
 };
 
 static void vce_v3_0_set_ring_funcs(struct amdgpu_device *adev)