Skip to contentSkip to conversation

support Arcturus IFM workaround

PATCH In review James Zhu 2 participants 12 messages Jan 14, 2020 – Jan 15, 2020

Subsystems (2)
lore
Share & tools
lore

James Zhu

author
#0 lore raw

v3 cover support Arcturus IFM workaround

Add vcn2.5 dpg mode/dpg pause mode/dpg sram mode to support acturus IFM(instruction fetch monitor) work around.

v2: Correct firmware ucode index in vcn_v2_5_mc_resume_dpg_mode
v3: Share multiple instance indirect DPG SRAM mode support for vcn2

James Zhu (5):
drm/amdgpu/vcn: support multiple-instance dpg pause mode drm/amdgpu/vcn2.5: support multiple instance direct SRAM read and
write
drm/amdgpu/vcn2.5: add DPG mode start and stop drm/amdgpu/vcn2.5: add dpg pause mode drm/amdgpu/vcn2.5: implement indirect DPG SRAM mode

 drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c |  34 +--
 drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h |  48 ++--
 drivers/gpu/drm/amd/amdgpu/vcn_v1_0.c   |   8 +-
 drivers/gpu/drm/amd/amdgpu/vcn_v2_0.c   |  98 ++++----
 drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c   | 394 +++++++++++++++++++++++++++++++-
 5 files changed, 487 insertions(+), 95 deletions(-)
Signature · 6 lines
2.7.4

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
author
#1 lore raw

PATCH v3 1/5 drm/amdgpu/vcn: support multiple-instance dpg pause mode

Add multiple-instance dpg pause mode support for VCN2.5

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c | 4 ++--
 drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h | 2 +-
 drivers/gpu/drm/amd/amdgpu/vcn_v1_0.c   | 8 ++++----
 drivers/gpu/drm/amd/amdgpu/vcn_v2_0.c   | 4 ++--
 4 files changed, 9 insertions(+), 9 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c +2 -2
index ed106d9..99df693 100644
@@ -298,7 +298,7 @@ static void amdgpu_vcn_idle_work_handler(struct work_struct *work)
 			else
 				new_state.fw_based = VCN_DPG_STATE__UNPAUSE;
 
-			adev->vcn.pause_dpg_mode(adev, &new_state);
+			adev->vcn.pause_dpg_mode(adev, j, &new_state);
 		}
 
 		fence[j] += amdgpu_fence_count_emitted(&adev->vcn.inst[j].ring_dec);
@@ -341,7 +341,7 @@ void amdgpu_vcn_ring_begin_use(struct amdgpu_ring *ring)
 		if (ring->funcs->type == AMDGPU_RING_TYPE_VCN_ENC)
 			new_state.fw_based = VCN_DPG_STATE__PAUSE;
 
-		adev->vcn.pause_dpg_mode(adev, &new_state);
+		adev->vcn.pause_dpg_mode(adev, ring->me, &new_state);
 	}
 }
 
drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h +1 -1
index e6dee82..26c6623 100644
@@ -199,7 +199,7 @@ struct amdgpu_vcn {
 
 	unsigned	harvest_config;
 	int (*pause_dpg_mode)(struct amdgpu_device *adev,
-		struct dpg_pause_state *new_state);
+		int inst_idx, struct dpg_pause_state *new_state);
 };
 
 int amdgpu_vcn_sw_init(struct amdgpu_device *adev);
drivers/gpu/drm/amd/amdgpu/vcn_v1_0.c +4 -4
index 3b025a3..a70351f 100644
@@ -50,7 +50,7 @@ static void vcn_v1_0_set_enc_ring_funcs(struct amdgpu_device *adev);
 static void vcn_v1_0_set_irq_funcs(struct amdgpu_device *adev);
 static int vcn_v1_0_set_powergating_state(void *handle, enum amd_powergating_state state);
 static int vcn_v1_0_pause_dpg_mode(struct amdgpu_device *adev,
-				struct dpg_pause_state *new_state);
+				int inst_idx, struct dpg_pause_state *new_state);
 
 static void vcn_v1_0_idle_work_handler(struct work_struct *work);
 
@@ -1199,7 +1199,7 @@ static int vcn_v1_0_stop(struct amdgpu_device *adev)
 }
 
 static int vcn_v1_0_pause_dpg_mode(struct amdgpu_device *adev,
-				struct dpg_pause_state *new_state)
+				int inst_idx, struct dpg_pause_state *new_state)
 {
 	int ret_code;
 	uint32_t reg_data = 0;
@@ -1786,7 +1786,7 @@ static void vcn_v1_0_idle_work_handler(struct work_struct *work)
 		else
 			new_state.jpeg = VCN_DPG_STATE__UNPAUSE;
 
-		adev->vcn.pause_dpg_mode(adev, &new_state);
+		adev->vcn.pause_dpg_mode(adev, 0, &new_state);
 	}
 
 	fences += amdgpu_fence_count_emitted(&adev->jpeg.inst->ring_dec);
@@ -1840,7 +1840,7 @@ void vcn_v1_0_ring_begin_use(struct amdgpu_ring *ring)
 		else if (ring->funcs->type == AMDGPU_RING_TYPE_VCN_JPEG)
 			new_state.jpeg = VCN_DPG_STATE__PAUSE;
 
-		adev->vcn.pause_dpg_mode(adev, &new_state);
+		adev->vcn.pause_dpg_mode(adev, 0, &new_state);
 	}
 }
 
drivers/gpu/drm/amd/amdgpu/vcn_v2_0.c +2 -2
index d76ece3..dcdc7ad 100644
@@ -58,7 +58,7 @@ static void vcn_v2_0_set_irq_funcs(struct amdgpu_device *adev);
 static int vcn_v2_0_set_powergating_state(void *handle,
 				enum amd_powergating_state state);
 static int vcn_v2_0_pause_dpg_mode(struct amdgpu_device *adev,
-				struct dpg_pause_state *new_state);
+				int inst_idx, struct dpg_pause_state *new_state);
 
 /**
  * vcn_v2_0_early_init - set function pointers
@@ -1135,7 +1135,7 @@ static int vcn_v2_0_stop(struct amdgpu_device *adev)
 }
 
 static int vcn_v2_0_pause_dpg_mode(struct amdgpu_device *adev,
-				struct dpg_pause_state *new_state)
+				int inst_idx, struct dpg_pause_state *new_state)
 {
 	struct amdgpu_ring *ring;
 	uint32_t reg_data = 0;
Signature · 6 lines
2.7.4

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
author
#2 lore raw

PATCH v3 2/5 drm/amdgpu/vcn2.5: support multiple instance direct SRAM read and write

In reply to James Zhu

Add multiple instance direct SRAM read and write support for vcn2.5

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c | 27 +++++-----
 drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h | 46 ++++++++--------
 drivers/gpu/drm/amd/amdgpu/vcn_v2_0.c   | 94 ++++++++++++++++-----------------
 3 files changed, 83 insertions(+), 84 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c +13 -14
index 99df693..ca62d99 100644
@@ -165,15 +165,15 @@ int amdgpu_vcn_sw_init(struct amdgpu_device *adev)
 			dev_err(adev->dev, "(%d) failed to allocate vcn bo\n", r);
 			return r;
 		}
-	}
 
-	if (adev->vcn.indirect_sram) {
-		r = amdgpu_bo_create_kernel(adev, 64 * 2 * 4, PAGE_SIZE,
-			    AMDGPU_GEM_DOMAIN_VRAM, &adev->vcn.dpg_sram_bo,
-			    &adev->vcn.dpg_sram_gpu_addr, &adev->vcn.dpg_sram_cpu_addr);
-		if (r) {
-			dev_err(adev->dev, "(%d) failed to allocate DPG bo\n", r);
-			return r;
+		if (adev->vcn.indirect_sram) {
+			r = amdgpu_bo_create_kernel(adev, 64 * 2 * 4, PAGE_SIZE,
+					AMDGPU_GEM_DOMAIN_VRAM, &adev->vcn.inst[i].dpg_sram_bo,
+					&adev->vcn.inst[i].dpg_sram_gpu_addr, &adev->vcn.inst[i].dpg_sram_cpu_addr);
+			if (r) {
+				dev_err(adev->dev, "VCN %d (%d) failed to allocate DPG bo\n", i, r);
+				return r;
+			}
 		}
 	}
 
@@ -186,15 +186,14 @@ int amdgpu_vcn_sw_fini(struct amdgpu_device *adev)
 
 	cancel_delayed_work_sync(&adev->vcn.idle_work);
 
-	if (adev->vcn.indirect_sram) {
-		amdgpu_bo_free_kernel(&adev->vcn.dpg_sram_bo,
-				      &adev->vcn.dpg_sram_gpu_addr,
-				      (void **)&adev->vcn.dpg_sram_cpu_addr);
-	}
-
 	for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
 		if (adev->vcn.harvest_config & (1 << j))
 			continue;
+		if (adev->vcn.indirect_sram) {
+			amdgpu_bo_free_kernel(&adev->vcn.inst[i].dpg_sram_bo,
+						  &adev->vcn.inst[i].dpg_sram_gpu_addr,
+						  (void **)&adev->vcn.inst[i].dpg_sram_cpu_addr);
+		}
 		kvfree(adev->vcn.inst[j].saved_bo);
 
 		amdgpu_bo_free_kernel(&adev->vcn.inst[j].vcpu_bo,
drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h +23 -23

66 lines not on this page · raw message

drivers/gpu/drm/amd/amdgpu/vcn_v2_0.c +47 -47

229 lines not on this page · raw message

Signature · 6 lines
2.7.4

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
#3 lore raw

Re: [PATCH v3 2/5] drm/amdgpu/vcn2.5: support multiple instance direct SRAM read and write

Quoted 70 lines · James Zhu

Add multiple instance direct SRAM read and write support for vcn2.5

---
  drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c | 27 +++++-----
  drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h | 46 ++++++++--------
  drivers/gpu/drm/amd/amdgpu/vcn_v2_0.c   | 94 ++++++++++++++++-----------------
  3 files changed, 83 insertions(+), 84 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c +13 -14
index 99df693..ca62d99 100644
@@ -165,15 +165,15 @@ int amdgpu_vcn_sw_init(struct amdgpu_device *adev)
  			dev_err(adev->dev, "(%d) failed to allocate vcn bo\n", r);
  			return r;
  		}
-	}
 
-	if (adev->vcn.indirect_sram) {
-		r = amdgpu_bo_create_kernel(adev, 64 * 2 * 4, PAGE_SIZE,
-			    AMDGPU_GEM_DOMAIN_VRAM, &adev->vcn.dpg_sram_bo,
-			    &adev->vcn.dpg_sram_gpu_addr, &adev->vcn.dpg_sram_cpu_addr);
-		if (r) {
-			dev_err(adev->dev, "(%d) failed to allocate DPG bo\n", r);
-			return r;
+		if (adev->vcn.indirect_sram) {
+			r = amdgpu_bo_create_kernel(adev, 64 * 2 * 4, PAGE_SIZE,
+					AMDGPU_GEM_DOMAIN_VRAM, &adev->vcn.inst[i].dpg_sram_bo,
+					&adev->vcn.inst[i].dpg_sram_gpu_addr, &adev->vcn.inst[i].dpg_sram_cpu_addr);
+			if (r) {
+				dev_err(adev->dev, "VCN %d (%d) failed to allocate DPG bo\n", i, r);
+				return r;
+			}
  		}
  	}
 
@@ -186,15 +186,14 @@ int amdgpu_vcn_sw_fini(struct amdgpu_device *adev)
 
  	cancel_delayed_work_sync(&adev->vcn.idle_work);
 
-	if (adev->vcn.indirect_sram) {
-		amdgpu_bo_free_kernel(&adev->vcn.dpg_sram_bo,
-				      &adev->vcn.dpg_sram_gpu_addr,
-				      (void **)&adev->vcn.dpg_sram_cpu_addr);
-	}
-
  	for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
  		if (adev->vcn.harvest_config & (1 << j))
  			continue;
+		if (adev->vcn.indirect_sram) {
+			amdgpu_bo_free_kernel(&adev->vcn.inst[i].dpg_sram_bo,
+						  &adev->vcn.inst[i].dpg_sram_gpu_addr,
+						  (void **)&adev->vcn.inst[i].dpg_sram_cpu_addr);
+		}
  		kvfree(adev->vcn.inst[j].saved_bo);
 
  		amdgpu_bo_free_kernel(&adev->vcn.inst[j].vcpu_bo,
drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h +0 -4
index 26c6623..63c55bc 100644
@@ -104,27 +104,27 @@
  		internal_reg_offset >>= 2;							\
  	})
 
-#define RREG32_SOC15_DPG_MODE_2_0(offset, mask_en) 						\
-	({ 											\
-		WREG32_SOC15(VCN, 0, mmUVD_DPG_LMA_CTL, 					\
-			(0x0 << UVD_DPG_LMA_CTL__READ_WRITE__SHIFT | 				\

drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h line 113

Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h +7 -3
-			mask_en << UVD_DPG_LMA_CTL__MASK_EN__SHIFT | 				\
-			offset << UVD_DPG_LMA_CTL__READ_WRITE_ADDR__SHIFT)); 			\
-		RREG32_SOC15(VCN, 0, mmUVD_DPG_LMA_DATA); 					\
+#define RREG32_SOC15_DPG_MODE_2_0(inst_idx, offset, mask_en) 						\
+	({											\
+		WREG32_SOC15(VCN, inst, 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(VCN, inst_idx, mmUVD_DPG_LMA_DATA);					\

Please keep the backslash aligned with those above and below.

Regards,

Leo

Quoted 280 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h +16 -16
  	})
 
-#define WREG32_SOC15_DPG_MODE_2_0(offset, value, mask_en, indirect)				\
-	do { 											\
-		if (!indirect) { 								\
-			WREG32_SOC15(VCN, 0, mmUVD_DPG_LMA_DATA, value); 			\
-			WREG32_SOC15(VCN, 0, mmUVD_DPG_LMA_CTL, 				\
-				(0x1 << 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)); 		\
-		} else { 									\
-			*adev->vcn.dpg_sram_curr_addr++ = offset; 				\
-			*adev->vcn.dpg_sram_curr_addr++ = value; 				\
-		} 										\
+#define WREG32_SOC15_DPG_MODE_2_0(inst_idx, offset, value, mask_en, indirect)				\
+	do {											\
+		if (!indirect) {								\
+			WREG32_SOC15(VCN, inst_idx, mmUVD_DPG_LMA_DATA, value);			\
+			WREG32_SOC15(VCN, inst_idx, mmUVD_DPG_LMA_CTL, 				\
+				(0x1 << 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));		\
+		} else {									\
+			*adev->vcn.inst[inst_idx].dpg_sram_curr_addr++ = offset;				\
+			*adev->vcn.inst[inst_idx].dpg_sram_curr_addr++ = value;				\
+		}										\
  	} while (0)
 
  enum engine_status_constants {
@@ -173,6 +173,10 @@ struct amdgpu_vcn_inst {
  	struct amdgpu_ring	ring_enc[AMDGPU_VCN_MAX_ENC_RINGS];
  	struct amdgpu_irq_src	irq;
  	struct amdgpu_vcn_reg	external;
+	struct amdgpu_bo	*dpg_sram_bo;
+	void			*dpg_sram_cpu_addr;
+	uint64_t		dpg_sram_gpu_addr;
+	uint32_t		*dpg_sram_curr_addr;
  };
 
  struct amdgpu_vcn {
@@ -184,10 +188,6 @@ struct amdgpu_vcn {
  	struct dpg_pause_state pause_state;
 
  	bool			indirect_sram;
-	struct amdgpu_bo	*dpg_sram_bo;
-	void			*dpg_sram_cpu_addr;
-	uint64_t		dpg_sram_gpu_addr;
-	uint32_t		*dpg_sram_curr_addr;
 
  	uint8_t	num_vcn_inst;
  	struct amdgpu_vcn_inst	 inst[AMDGPU_MAX_VCN_INSTANCES];
drivers/gpu/drm/amd/amdgpu/vcn_v2_0.c +44 -44
index dcdc7ad..9ff59ac 100644
@@ -356,88 +356,88 @@ static void vcn_v2_0_mc_resume_dpg_mode(struct amdgpu_device *adev, bool indirec
  	/* cache window 0: fw */
  	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
  		if (!indirect) {
-			WREG32_SOC15_DPG_MODE_2_0(SOC15_DPG_MODE_OFFSET_2_0(
+			WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
  				UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
  				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].tmr_mc_addr_lo), 0, indirect);
-			WREG32_SOC15_DPG_MODE_2_0(SOC15_DPG_MODE_OFFSET_2_0(
+			WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
  				UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
  				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].tmr_mc_addr_hi), 0, indirect);
-			WREG32_SOC15_DPG_MODE_2_0(SOC15_DPG_MODE_OFFSET_2_0(
+			WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
  				UVD, 0, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
  		} else {
-			WREG32_SOC15_DPG_MODE_2_0(SOC15_DPG_MODE_OFFSET_2_0(
+			WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
  				UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), 0, 0, indirect);
-			WREG32_SOC15_DPG_MODE_2_0(SOC15_DPG_MODE_OFFSET_2_0(
+			WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
  				UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), 0, 0, indirect);
-			WREG32_SOC15_DPG_MODE_2_0(SOC15_DPG_MODE_OFFSET_2_0(
+			WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
  				UVD, 0, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
  		}
  		offset = 0;
  	} else {
-		WREG32_SOC15_DPG_MODE_2_0(SOC15_DPG_MODE_OFFSET_2_0(
+		WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
  			UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
  			lower_32_bits(adev->vcn.inst->gpu_addr), 0, indirect);
-		WREG32_SOC15_DPG_MODE_2_0(SOC15_DPG_MODE_OFFSET_2_0(
+		WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
  			UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
  			upper_32_bits(adev->vcn.inst->gpu_addr), 0, indirect);
  		offset = size;
-		WREG32_SOC15_DPG_MODE_2_0(SOC15_DPG_MODE_OFFSET_2_0(
+		WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
  			UVD, 0, mmUVD_VCPU_CACHE_OFFSET0),
  			AMDGPU_UVD_FIRMWARE_OFFSET >> 3, 0, indirect);
  	}
 
  	if (!indirect)
-		WREG32_SOC15_DPG_MODE_2_0(SOC15_DPG_MODE_OFFSET_2_0(
+		WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
  			UVD, 0, mmUVD_VCPU_CACHE_SIZE0), size, 0, indirect);
  	else
-		WREG32_SOC15_DPG_MODE_2_0(SOC15_DPG_MODE_OFFSET_2_0(
+		WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
  			UVD, 0, mmUVD_VCPU_CACHE_SIZE0), 0, 0, indirect);
 
  	/* cache window 1: stack */
  	if (!indirect) {
-		WREG32_SOC15_DPG_MODE_2_0(SOC15_DPG_MODE_OFFSET_2_0(
+		WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
  			UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
  			lower_32_bits(adev->vcn.inst->gpu_addr + offset), 0, indirect);
-		WREG32_SOC15_DPG_MODE_2_0(SOC15_DPG_MODE_OFFSET_2_0(
+		WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
  			UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
  			upper_32_bits(adev->vcn.inst->gpu_addr + offset), 0, indirect);
-		WREG32_SOC15_DPG_MODE_2_0(SOC15_DPG_MODE_OFFSET_2_0(
+		WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
  			UVD, 0, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
  	} else {
-		WREG32_SOC15_DPG_MODE_2_0(SOC15_DPG_MODE_OFFSET_2_0(
+		WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
  			UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW), 0, 0, indirect);
-		WREG32_SOC15_DPG_MODE_2_0(SOC15_DPG_MODE_OFFSET_2_0(
+		WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
  			UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH), 0, 0, indirect);
-		WREG32_SOC15_DPG_MODE_2_0(SOC15_DPG_MODE_OFFSET_2_0(
+		WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
  			UVD, 0, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
  	}
-	WREG32_SOC15_DPG_MODE_2_0(SOC15_DPG_MODE_OFFSET_2_0(
+	WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
  		UVD, 0, mmUVD_VCPU_CACHE_SIZE1), AMDGPU_VCN_STACK_SIZE, 0, indirect);
 
  	/* cache window 2: context */
-	WREG32_SOC15_DPG_MODE_2_0(SOC15_DPG_MODE_OFFSET_2_0(
+	WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
  		UVD, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
  		lower_32_bits(adev->vcn.inst->gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
-	WREG32_SOC15_DPG_MODE_2_0(SOC15_DPG_MODE_OFFSET_2_0(
+	WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
  		UVD, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
  		upper_32_bits(adev->vcn.inst->gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
-	WREG32_SOC15_DPG_MODE_2_0(SOC15_DPG_MODE_OFFSET_2_0(
+	WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
  		UVD, 0, mmUVD_VCPU_CACHE_OFFSET2), 0, 0, indirect);
-	WREG32_SOC15_DPG_MODE_2_0(SOC15_DPG_MODE_OFFSET_2_0(
+	WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
  		UVD, 0, mmUVD_VCPU_CACHE_SIZE2), AMDGPU_VCN_CONTEXT_SIZE, 0, indirect);
 
  	/* non-cache window */
-	WREG32_SOC15_DPG_MODE_2_0(SOC15_DPG_MODE_OFFSET_2_0(
+	WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
  		UVD, 0, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW), 0, 0, indirect);
-	WREG32_SOC15_DPG_MODE_2_0(SOC15_DPG_MODE_OFFSET_2_0(
+	WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
  		UVD, 0, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH), 0, 0, indirect);
-	WREG32_SOC15_DPG_MODE_2_0(SOC15_DPG_MODE_OFFSET_2_0(
+	WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
  		UVD, 0, mmUVD_VCPU_NONCACHE_OFFSET0), 0, 0, indirect);
-	WREG32_SOC15_DPG_MODE_2_0(SOC15_DPG_MODE_OFFSET_2_0(
+	WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
  		UVD, 0, mmUVD_VCPU_NONCACHE_SIZE0), 0, 0, indirect);
 
  	/* VCN global tiling registers */
-	WREG32_SOC15_DPG_MODE_2_0(SOC15_DPG_MODE_OFFSET_2_0(
+	WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
  		UVD, 0, mmUVD_GFX10_ADDR_CONFIG), adev->gfx.config.gb_addr_config, 0, indirect);
  }
 
@@ -583,19 +583,19 @@ static void vcn_v2_0_clock_gating_dpg_mode(struct amdgpu_device *adev,
  		 UVD_CGC_CTRL__WCB_MODE_MASK |
  		 UVD_CGC_CTRL__VCPU_MODE_MASK |
  		 UVD_CGC_CTRL__SCPU_MODE_MASK);
-	WREG32_SOC15_DPG_MODE_2_0(SOC15_DPG_MODE_OFFSET_2_0(
+	WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
  		UVD, 0, mmUVD_CGC_CTRL), reg_data, sram_sel, indirect);
 
  	/* turn off clock gating */
-	WREG32_SOC15_DPG_MODE_2_0(SOC15_DPG_MODE_OFFSET_2_0(
+	WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
  		UVD, 0, mmUVD_CGC_GATE), 0, sram_sel, indirect);
 
  	/* turn on SUVD clock gating */
-	WREG32_SOC15_DPG_MODE_2_0(SOC15_DPG_MODE_OFFSET_2_0(
+	WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
  		UVD, 0, mmUVD_SUVD_CGC_GATE), 1, sram_sel, indirect);
 
  	/* turn on sw mode in UVD_SUVD_CGC_CTRL */
-	WREG32_SOC15_DPG_MODE_2_0(SOC15_DPG_MODE_OFFSET_2_0(
+	WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
  		UVD, 0, mmUVD_SUVD_CGC_CTRL), 0, sram_sel, indirect);
  }
 
@@ -759,7 +759,7 @@ static int vcn_v2_0_start_dpg_mode(struct amdgpu_device *adev, bool indirect)
  	WREG32_SOC15(UVD, 0, mmUVD_POWER_STATUS, tmp);
 
  	if (indirect)
-		adev->vcn.dpg_sram_curr_addr = (uint32_t*)adev->vcn.dpg_sram_cpu_addr;
+		adev->vcn.inst->dpg_sram_curr_addr = (uint32_t*)adev->vcn.inst->dpg_sram_cpu_addr;
 
  	/* enable clock gating */
  	vcn_v2_0_clock_gating_dpg_mode(adev, 0, indirect);
@@ -768,11 +768,11 @@ static int vcn_v2_0_start_dpg_mode(struct amdgpu_device *adev, bool indirect)
  	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
  	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
  	tmp |= UVD_VCPU_CNTL__MIF_WR_LOW_THRESHOLD_BP_MASK;
-	WREG32_SOC15_DPG_MODE_2_0(SOC15_DPG_MODE_OFFSET_2_0(
+	WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
  		UVD, 0, mmUVD_VCPU_CNTL), tmp, 0, indirect);
 
  	/* disable master interupt */
-	WREG32_SOC15_DPG_MODE_2_0(SOC15_DPG_MODE_OFFSET_2_0(
+	WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
  		UVD, 0, mmUVD_MASTINT_EN), 0, 0, indirect);
 
  	/* setup mmUVD_LMI_CTRL */
@@ -784,28 +784,28 @@ static int vcn_v2_0_start_dpg_mode(struct amdgpu_device *adev, bool indirect)
  		UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK |
  		(8 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) |
  		0x00100000L);
-	WREG32_SOC15_DPG_MODE_2_0(SOC15_DPG_MODE_OFFSET_2_0(
+	WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
  		UVD, 0, mmUVD_LMI_CTRL), tmp, 0, indirect);
 
-	WREG32_SOC15_DPG_MODE_2_0(SOC15_DPG_MODE_OFFSET_2_0(
+	WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
  		UVD, 0, mmUVD_MPC_CNTL),
  		0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT, 0, indirect);
 
-	WREG32_SOC15_DPG_MODE_2_0(SOC15_DPG_MODE_OFFSET_2_0(
+	WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
  		UVD, 0, mmUVD_MPC_SET_MUXA0),
  		((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
  		 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
  		 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
  		 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)), 0, indirect);
 
-	WREG32_SOC15_DPG_MODE_2_0(SOC15_DPG_MODE_OFFSET_2_0(
+	WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
  		UVD, 0, mmUVD_MPC_SET_MUXB0),
  		((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
  		 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
  		 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
  		 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)), 0, indirect);
 
-	WREG32_SOC15_DPG_MODE_2_0(SOC15_DPG_MODE_OFFSET_2_0(
+	WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
  		UVD, 0, mmUVD_MPC_SET_MUX),
  		((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
  		 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
@@ -813,29 +813,29 @@ static int vcn_v2_0_start_dpg_mode(struct amdgpu_device *adev, bool indirect)
 
  	vcn_v2_0_mc_resume_dpg_mode(adev, indirect);
 
-	WREG32_SOC15_DPG_MODE_2_0(SOC15_DPG_MODE_OFFSET_2_0(
+	WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
  		UVD, 0, mmUVD_REG_XX_MASK), 0x10, 0, indirect);
-	WREG32_SOC15_DPG_MODE_2_0(SOC15_DPG_MODE_OFFSET_2_0(
+	WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
  		UVD, 0, mmUVD_RBC_XX_IB_REG_CHECK), 0x3, 0, indirect);
 
  	/* release VCPU reset to boot */
-	WREG32_SOC15_DPG_MODE_2_0(SOC15_DPG_MODE_OFFSET_2_0(
+	WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
  		UVD, 0, mmUVD_SOFT_RESET), 0, 0, indirect);
 
  	/* enable LMI MC and UMC channels */
-	WREG32_SOC15_DPG_MODE_2_0(SOC15_DPG_MODE_OFFSET_2_0(
+	WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
  		UVD, 0, mmUVD_LMI_CTRL2),
  		0x1F << UVD_LMI_CTRL2__RE_OFLD_MIF_WR_REQ_NUM__SHIFT, 0, indirect);
 
  	/* enable master interrupt */
-	WREG32_SOC15_DPG_MODE_2_0(SOC15_DPG_MODE_OFFSET_2_0(
+	WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
  		UVD, 0, mmUVD_MASTINT_EN),
  		UVD_MASTINT_EN__VCPU_EN_MASK, 0, indirect);
 

drivers/gpu/drm/amd/amdgpu/vcn_v2_0.c static int vcn_v2_0_start_dpg_mode(struct amdgpu_device *adev, bool indirect) · line 841

Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/vcn_v2_0.c +3 -3
  	if (indirect)
-		psp_update_vcn_sram(adev, 0, adev->vcn.dpg_sram_gpu_addr,
-				    (uint32_t)((uintptr_t)adev->vcn.dpg_sram_curr_addr -
-					       (uintptr_t)adev->vcn.dpg_sram_cpu_addr));
+		psp_update_vcn_sram(adev, 0, adev->vcn.inst->dpg_sram_gpu_addr,
+				    (uint32_t)((uintptr_t)adev->vcn.inst->dpg_sram_curr_addr -
+					       (uintptr_t)adev->vcn.inst->dpg_sram_cpu_addr));
 
  	/* force RBC into idle state */
  	rb_bufsz = order_base_2(ring->ring_size);

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

author
#4 lore raw

PATCH v4 2/6 drm/amdgpu/vcn: support multiple instance direct SRAM read and write

In reply to James Zhu

Add multiple instance direct SRAM read and write support for vcn2.5

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c | 27 +++++-----
 drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h | 46 ++++++++--------
 drivers/gpu/drm/amd/amdgpu/vcn_v2_0.c   | 94 ++++++++++++++++-----------------
 3 files changed, 83 insertions(+), 84 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c +13 -14
index 99df693..ca62d99 100644
@@ -165,15 +165,15 @@ int amdgpu_vcn_sw_init(struct amdgpu_device *adev)
 			dev_err(adev->dev, "(%d) failed to allocate vcn bo\n", r);
 			return r;
 		}
-	}
 
-	if (adev->vcn.indirect_sram) {
-		r = amdgpu_bo_create_kernel(adev, 64 * 2 * 4, PAGE_SIZE,
-			    AMDGPU_GEM_DOMAIN_VRAM, &adev->vcn.dpg_sram_bo,
-			    &adev->vcn.dpg_sram_gpu_addr, &adev->vcn.dpg_sram_cpu_addr);
-		if (r) {
-			dev_err(adev->dev, "(%d) failed to allocate DPG bo\n", r);
-			return r;
+		if (adev->vcn.indirect_sram) {
+			r = amdgpu_bo_create_kernel(adev, 64 * 2 * 4, PAGE_SIZE,
+					AMDGPU_GEM_DOMAIN_VRAM, &adev->vcn.inst[i].dpg_sram_bo,
+					&adev->vcn.inst[i].dpg_sram_gpu_addr, &adev->vcn.inst[i].dpg_sram_cpu_addr);
+			if (r) {
+				dev_err(adev->dev, "VCN %d (%d) failed to allocate DPG bo\n", i, r);
+				return r;
+			}
 		}
 	}
 
@@ -186,15 +186,14 @@ int amdgpu_vcn_sw_fini(struct amdgpu_device *adev)
 
 	cancel_delayed_work_sync(&adev->vcn.idle_work);
 
-	if (adev->vcn.indirect_sram) {
-		amdgpu_bo_free_kernel(&adev->vcn.dpg_sram_bo,
-				      &adev->vcn.dpg_sram_gpu_addr,
-				      (void **)&adev->vcn.dpg_sram_cpu_addr);
-	}
-
 	for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
 		if (adev->vcn.harvest_config & (1 << j))
 			continue;
+		if (adev->vcn.indirect_sram) {
+			amdgpu_bo_free_kernel(&adev->vcn.inst[i].dpg_sram_bo,
+						  &adev->vcn.inst[i].dpg_sram_gpu_addr,
+						  (void **)&adev->vcn.inst[i].dpg_sram_cpu_addr);
+		}
 		kvfree(adev->vcn.inst[j].saved_bo);
 
 		amdgpu_bo_free_kernel(&adev->vcn.inst[j].vcpu_bo,
drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h +23 -23

66 lines not on this page · raw message

drivers/gpu/drm/amd/amdgpu/vcn_v2_0.c +47 -47

229 lines not on this page · raw message

Signature · 6 lines
2.7.4

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
author
#5 lore raw

PATCH v3 3/5 drm/amdgpu/vcn2.5: add DPG mode start and stop

In reply to James Zhu

Add DPG mode start and stop functions for vcn2.5

v2: Correct firmware ucode index in vcn_v2_5_mc_resume_dpg_mode

---
 drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c | 295 +++++++++++++++++++++++++++++++++-
 1 file changed, 293 insertions(+), 2 deletions(-)
drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c +293 -2
index fa90249..b3ddf68 100644
@@ -48,6 +48,11 @@
 #define mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH_INTERNAL_OFFSET 	0x3b5
 #define mmUVD_RBC_IB_SIZE_INTERNAL_OFFSET			0x25c
 
+#define mmUVD_RBC_XX_IB_REG_CHECK 					0x026b
+#define mmUVD_RBC_XX_IB_REG_CHECK_BASE_IDX 				1
+#define mmUVD_REG_XX_MASK 						0x026c
+#define mmUVD_REG_XX_MASK_BASE_IDX 					1
+
 #define VCN25_MAX_HW_INSTANCES_ARCTURUS			2
 
 static void vcn_v2_5_set_dec_ring_funcs(struct amdgpu_device *adev);
@@ -286,7 +291,8 @@ static int vcn_v2_5_hw_init(void *handle)
 
 done:
 	if (!r)
-		DRM_INFO("VCN decode and encode initialized successfully.\n");
+		DRM_INFO("VCN decode and encode initialized successfully(under %s).\n",
+			(adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)?"DPG Mode":"SPG Mode");
 
 	return r;
 }
@@ -309,7 +315,9 @@ static int vcn_v2_5_hw_fini(void *handle)
 			continue;
 		ring = &adev->vcn.inst[i].ring_dec;
 
-		if (RREG32_SOC15(VCN, i, mmUVD_STATUS))
+		if ((adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) ||
+	        (adev->vcn.cur_state != AMD_PG_STATE_GATE &&
+	         RREG32_SOC15(VCN, i, mmUVD_STATUS)))
 			vcn_v2_5_set_powergating_state(adev, AMD_PG_STATE_GATE);
 
 		ring->sched.ready = false;
@@ -418,6 +426,78 @@ static void vcn_v2_5_mc_resume(struct amdgpu_device *adev)
 	}
 }
 
+static void vcn_v2_5_mc_resume_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
+{
+	uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
+	uint32_t offset;
+
+	/* cache window 0: fw */
+	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
+		WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+			UVD, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
+			(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_lo), 0, indirect);
+		WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+			UVD, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
+			(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_hi), 0, indirect);
+		WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+			UVD, inst_idx, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
+		offset = 0;
+	} else {
+		WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+			UVD, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
+			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
+		WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+			UVD, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
+			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
+		offset = size;
+		WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+			UVD, inst_idx, mmUVD_VCPU_CACHE_OFFSET0),
+			AMDGPU_UVD_FIRMWARE_OFFSET >> 3, 0, indirect);
+	}
+
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_VCPU_CACHE_SIZE0), size, 0, indirect);
+
+	/* cache window 1: stack */
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
+		lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
+		upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
+
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_VCPU_CACHE_SIZE1), AMDGPU_VCN_STACK_SIZE, 0, indirect);
+
+	/* cache window 2: context */
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
+		lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
+		upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_VCPU_CACHE_OFFSET2), 0, 0, indirect);
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_VCPU_CACHE_SIZE2), AMDGPU_VCN_CONTEXT_SIZE, 0, indirect);
+
+	/* non-cache window */
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW), 0, 0, indirect);
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH), 0, 0, indirect);
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_VCPU_NONCACHE_OFFSET0), 0, 0, indirect);
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_VCPU_NONCACHE_SIZE0), 0, 0, indirect);
+
+	/* VCN global tiling registers */
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_GFX8_ADDR_CONFIG), adev->gfx.config.gb_addr_config, 0, indirect);
+}
+
 /**
  * vcn_v2_5_disable_clock_gating - disable VCN clock gating
  *
@@ -536,6 +616,54 @@ static void vcn_v2_5_disable_clock_gating(struct amdgpu_device *adev)
 	}
 }
 
+static void vcn_v2_5_clock_gating_dpg_mode(struct amdgpu_device *adev,
+		uint8_t sram_sel, int inst_idx, uint8_t indirect)
+{
+	uint32_t reg_data = 0;
+
+	/* enable sw clock gating control */
+	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
+		reg_data = 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
+	else
+		reg_data = 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
+	reg_data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
+	reg_data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
+	reg_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__MMSCH_MODE_MASK);
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_CGC_CTRL), reg_data, sram_sel, indirect);
+
+	/* turn off clock gating */
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_CGC_GATE), 0, sram_sel, indirect);
+
+	/* turn on SUVD clock gating */
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_SUVD_CGC_GATE), 1, sram_sel, indirect);
+
+	/* turn on sw mode in UVD_SUVD_CGC_CTRL */
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_SUVD_CGC_CTRL), 0, sram_sel, indirect);
+}
+
 /**
  * vcn_v2_5_enable_clock_gating - enable VCN clock gating
  *
@@ -598,6 +726,130 @@ static void vcn_v2_5_enable_clock_gating(struct amdgpu_device *adev)
 	}
 }
 
+static int vcn_v2_5_start_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
+{
+	struct amdgpu_ring *ring;
+	uint32_t rb_bufsz, tmp;
+
+	/* disable register anti-hang mechanism */
+	WREG32_P(SOC15_REG_OFFSET(UVD, inst_idx, mmUVD_POWER_STATUS), 1,
+		~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
+	/* enable dynamic power gating mode */
+	tmp = RREG32_SOC15(UVD, inst_idx, mmUVD_POWER_STATUS);
+	tmp |= UVD_POWER_STATUS__UVD_PG_MODE_MASK;
+	tmp |= UVD_POWER_STATUS__UVD_PG_EN_MASK;
+	WREG32_SOC15(UVD, inst_idx, mmUVD_POWER_STATUS, tmp);
+
+	/* enable clock gating */
+	vcn_v2_5_clock_gating_dpg_mode(adev, 0, inst_idx, indirect);
+
+	/* enable VCPU clock */
+	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
+	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
+	tmp |= UVD_VCPU_CNTL__BLK_RST_MASK;
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_VCPU_CNTL), tmp, 0, indirect);
+
+	/* disable master interupt */
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_MASTINT_EN), 0, 0, indirect);
+
+	/* setup mmUVD_LMI_CTRL */
+	tmp = (0x8 | UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
+		UVD_LMI_CTRL__REQ_MODE_MASK |
+		UVD_LMI_CTRL__CRC_RESET_MASK |
+		UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
+		UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
+		UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK |
+		(8 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) |
+		0x00100000L);
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_LMI_CTRL), tmp, 0, indirect);
+
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_MPC_CNTL),
+		0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT, 0, indirect);
+
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_MPC_SET_MUXA0),
+		((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
+		 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
+		 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
+		 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)), 0, indirect);
+
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_MPC_SET_MUXB0),
+		((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
+		 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
+		 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
+		 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)), 0, indirect);
+
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_MPC_SET_MUX),
+		((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
+		 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
+		 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)), 0, indirect);
+
+	vcn_v2_5_mc_resume_dpg_mode(adev, inst_idx, indirect);
+
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_REG_XX_MASK), 0x10, 0, indirect);
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_RBC_XX_IB_REG_CHECK), 0x3, 0, indirect);
+
+	/* enable LMI MC and UMC channels */
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_LMI_CTRL2), 0, 0, indirect);
+
+	/* unblock VCPU register access */
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_RB_ARB_CTRL), 0, 0, indirect);
+
+	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
+	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_VCPU_CNTL), tmp, 0, indirect);
+
+	/* enable master interrupt */
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_MASTINT_EN),
+		UVD_MASTINT_EN__VCPU_EN_MASK, 0, indirect);
+
+	ring = &adev->vcn.inst[inst_idx].ring_dec;
+	/* force RBC into idle state */
+	rb_bufsz = order_base_2(ring->ring_size);
+	tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
+	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
+	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
+	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
+	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
+	WREG32_SOC15(UVD, inst_idx, mmUVD_RBC_RB_CNTL, tmp);
+
+	/* set the write pointer delay */
+	WREG32_SOC15(UVD, inst_idx, mmUVD_RBC_RB_WPTR_CNTL, 0);
+
+	/* set the wb address */
+	WREG32_SOC15(UVD, inst_idx, mmUVD_RBC_RB_RPTR_ADDR,
+		(upper_32_bits(ring->gpu_addr) >> 2));
+
+	/* programm the RB_BASE for ring buffer */
+	WREG32_SOC15(UVD, inst_idx, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,
+		lower_32_bits(ring->gpu_addr));
+	WREG32_SOC15(UVD, inst_idx, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH,
+		upper_32_bits(ring->gpu_addr));
+
+	/* Initialize the ring buffer's read and write pointers */
+	WREG32_SOC15(UVD, inst_idx, mmUVD_RBC_RB_RPTR, 0);
+
+	WREG32_SOC15(UVD, inst_idx, mmUVD_SCRATCH2, 0);
+
+	ring->wptr = RREG32_SOC15(UVD, inst_idx, mmUVD_RBC_RB_RPTR);
+	WREG32_SOC15(UVD, inst_idx, mmUVD_RBC_RB_WPTR,
+		lower_32_bits(ring->wptr));
+
+	return 0;
+}
+
 static int vcn_v2_5_start(struct amdgpu_device *adev)
 {
 	struct amdgpu_ring *ring;
@@ -610,6 +862,9 @@ static int vcn_v2_5_start(struct amdgpu_device *adev)
 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
 		if (adev->vcn.harvest_config & (1 << i))
 			continue;
+		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
+			return vcn_v2_5_start_dpg_mode(adev, i, 0);
+
 		/* disable register anti-hang mechanism */
 		WREG32_P(SOC15_REG_OFFSET(UVD, i, mmUVD_POWER_STATUS), 0,
 			~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
@@ -973,6 +1228,35 @@ static int vcn_v2_5_sriov_start(struct amdgpu_device *adev)
 	return vcn_v2_5_mmsch_start(adev, &adev->virt.mm_table);
 }
 
+static int vcn_v2_5_stop_dpg_mode(struct amdgpu_device *adev, int inst_idx)
+{
+	int ret_code = 0;
+	uint32_t tmp;
+
+	/* Wait for power status to be 1 */
+	SOC15_WAIT_ON_RREG(UVD, inst_idx, mmUVD_POWER_STATUS, 1,
+		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK, ret_code);
+
+	/* wait for read ptr to be equal to write ptr */
+	tmp = RREG32_SOC15(UVD, inst_idx, mmUVD_RB_WPTR);
+	SOC15_WAIT_ON_RREG(UVD, inst_idx, mmUVD_RB_RPTR, tmp, 0xFFFFFFFF, ret_code);
+
+	tmp = RREG32_SOC15(UVD, inst_idx, mmUVD_RB_WPTR2);
+	SOC15_WAIT_ON_RREG(UVD, inst_idx, mmUVD_RB_RPTR2, tmp, 0xFFFFFFFF, ret_code);
+
+	tmp = RREG32_SOC15(UVD, inst_idx, mmUVD_RBC_RB_WPTR) & 0x7FFFFFFF;
+	SOC15_WAIT_ON_RREG(UVD, inst_idx, mmUVD_RBC_RB_RPTR, tmp, 0xFFFFFFFF, ret_code);
+
+	SOC15_WAIT_ON_RREG(UVD, inst_idx, mmUVD_POWER_STATUS, 1,
+		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK, ret_code);
+
+	/* disable dynamic power gating mode */
+	WREG32_P(SOC15_REG_OFFSET(UVD, inst_idx, mmUVD_POWER_STATUS), 0,
+			~UVD_POWER_STATUS__UVD_PG_MODE_MASK);
+
+	return 0;
+}
+
 static int vcn_v2_5_stop(struct amdgpu_device *adev)
 {
 	uint32_t tmp;
@@ -981,6 +1265,12 @@ static int vcn_v2_5_stop(struct amdgpu_device *adev)
 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
 		if (adev->vcn.harvest_config & (1 << i))
 			continue;
+
+		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
+			r = vcn_v2_5_stop_dpg_mode(adev, i);
+			goto power_off;
+		}
+
 		/* wait for vcn idle */
 		SOC15_WAIT_ON_RREG(VCN, i, mmUVD_STATUS, UVD_STATUS__IDLE, 0x7, r);
 		if (r)
@@ -1030,6 +1320,7 @@ static int vcn_v2_5_stop(struct amdgpu_device *adev)
 			~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
 	}
 
+power_off:
 	if (adev->pm.dpm_enabled)
 		amdgpu_dpm_enable_uvd(adev, false);
 
Signature · 6 lines
2.7.4

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
#6 lore raw

Re: [PATCH v3 3/5] drm/amdgpu/vcn2.5: add DPG mode start and stop

Quoted 14 lines · James Zhu

Add DPG mode start and stop functions for vcn2.5

v2: Correct firmware ucode index in vcn_v2_5_mc_resume_dpg_mode

---
  drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c | 295 +++++++++++++++++++++++++++++++++-
  1 file changed, 293 insertions(+), 2 deletions(-)
drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c +0 -0
index fa90249..b3ddf68 100644

drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c line 55

Quoted 8 lines
drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c +5 -0
@@ -48,6 +48,11 @@
  #define mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH_INTERNAL_OFFSET 	0x3b5
  #define mmUVD_RBC_IB_SIZE_INTERNAL_OFFSET			0x25c
 
+#define mmUVD_RBC_XX_IB_REG_CHECK 					0x026b
+#define mmUVD_RBC_XX_IB_REG_CHECK_BASE_IDX 				1
+#define mmUVD_REG_XX_MASK 						0x026c
+#define mmUVD_REG_XX_MASK_BASE_IDX 					1
+

If these are the same as VCN2.0, please move them to amdgpu_vcn.h

Quoted 14 lines
drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c +2 -1
  #define VCN25_MAX_HW_INSTANCES_ARCTURUS			2
 
  static void vcn_v2_5_set_dec_ring_funcs(struct amdgpu_device *adev);
@@ -286,7 +291,8 @@ static int vcn_v2_5_hw_init(void *handle)
 
  done:
  	if (!r)
-		DRM_INFO("VCN decode and encode initialized successfully.\n");
+		DRM_INFO("VCN decode and encode initialized successfully(under %s).\n",
+			(adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)?"DPG Mode":"SPG Mode");
 
  	return r;
  }

drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c static int vcn_v2_5_hw_fini(void *handle) · line 319

Quoted 6 lines
drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c +2 -1
@@ -309,7 +315,9 @@ static int vcn_v2_5_hw_fini(void *handle)
  			continue;
  		ring = &adev->vcn.inst[i].ring_dec;
 
-		if (RREG32_SOC15(VCN, i, mmUVD_STATUS))
+		if ((adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) ||
+	        (adev->vcn.cur_state != AMD_PG_STATE_GATE &&

Should it add some indentation here?

Regards,

Leo

Quoted 329 lines
drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c +283 -0
+	         RREG32_SOC15(VCN, i, mmUVD_STATUS)))
  			vcn_v2_5_set_powergating_state(adev, AMD_PG_STATE_GATE);
 
  		ring->sched.ready = false;
@@ -418,6 +426,78 @@ static void vcn_v2_5_mc_resume(struct amdgpu_device *adev)
  	}
  }
 
+static void vcn_v2_5_mc_resume_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
+{
+	uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
+	uint32_t offset;
+
+	/* cache window 0: fw */
+	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
+		WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+			UVD, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
+			(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_lo), 0, indirect);
+		WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+			UVD, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
+			(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_hi), 0, indirect);
+		WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+			UVD, inst_idx, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
+		offset = 0;
+	} else {
+		WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+			UVD, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
+			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
+		WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+			UVD, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
+			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
+		offset = size;
+		WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+			UVD, inst_idx, mmUVD_VCPU_CACHE_OFFSET0),
+			AMDGPU_UVD_FIRMWARE_OFFSET >> 3, 0, indirect);
+	}
+
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_VCPU_CACHE_SIZE0), size, 0, indirect);
+
+	/* cache window 1: stack */
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
+		lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
+		upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
+
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_VCPU_CACHE_SIZE1), AMDGPU_VCN_STACK_SIZE, 0, indirect);
+
+	/* cache window 2: context */
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
+		lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
+		upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_VCPU_CACHE_OFFSET2), 0, 0, indirect);
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_VCPU_CACHE_SIZE2), AMDGPU_VCN_CONTEXT_SIZE, 0, indirect);
+
+	/* non-cache window */
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW), 0, 0, indirect);
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH), 0, 0, indirect);
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_VCPU_NONCACHE_OFFSET0), 0, 0, indirect);
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_VCPU_NONCACHE_SIZE0), 0, 0, indirect);
+
+	/* VCN global tiling registers */
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_GFX8_ADDR_CONFIG), adev->gfx.config.gb_addr_config, 0, indirect);
+}
+
  /**
   * vcn_v2_5_disable_clock_gating - disable VCN clock gating
   *
@@ -536,6 +616,54 @@ static void vcn_v2_5_disable_clock_gating(struct amdgpu_device *adev)
  	}
  }
 
+static void vcn_v2_5_clock_gating_dpg_mode(struct amdgpu_device *adev,
+		uint8_t sram_sel, int inst_idx, uint8_t indirect)
+{
+	uint32_t reg_data = 0;
+
+	/* enable sw clock gating control */
+	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
+		reg_data = 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
+	else
+		reg_data = 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
+	reg_data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
+	reg_data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
+	reg_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__MMSCH_MODE_MASK);
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_CGC_CTRL), reg_data, sram_sel, indirect);
+
+	/* turn off clock gating */
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_CGC_GATE), 0, sram_sel, indirect);
+
+	/* turn on SUVD clock gating */
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_SUVD_CGC_GATE), 1, sram_sel, indirect);
+
+	/* turn on sw mode in UVD_SUVD_CGC_CTRL */
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_SUVD_CGC_CTRL), 0, sram_sel, indirect);
+}
+
  /**
   * vcn_v2_5_enable_clock_gating - enable VCN clock gating
   *
@@ -598,6 +726,130 @@ static void vcn_v2_5_enable_clock_gating(struct amdgpu_device *adev)
  	}
  }
 
+static int vcn_v2_5_start_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
+{
+	struct amdgpu_ring *ring;
+	uint32_t rb_bufsz, tmp;
+
+	/* disable register anti-hang mechanism */
+	WREG32_P(SOC15_REG_OFFSET(UVD, inst_idx, mmUVD_POWER_STATUS), 1,
+		~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
+	/* enable dynamic power gating mode */
+	tmp = RREG32_SOC15(UVD, inst_idx, mmUVD_POWER_STATUS);
+	tmp |= UVD_POWER_STATUS__UVD_PG_MODE_MASK;
+	tmp |= UVD_POWER_STATUS__UVD_PG_EN_MASK;
+	WREG32_SOC15(UVD, inst_idx, mmUVD_POWER_STATUS, tmp);
+
+	/* enable clock gating */
+	vcn_v2_5_clock_gating_dpg_mode(adev, 0, inst_idx, indirect);
+
+	/* enable VCPU clock */
+	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
+	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
+	tmp |= UVD_VCPU_CNTL__BLK_RST_MASK;
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_VCPU_CNTL), tmp, 0, indirect);
+
+	/* disable master interupt */
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_MASTINT_EN), 0, 0, indirect);
+
+	/* setup mmUVD_LMI_CTRL */
+	tmp = (0x8 | UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
+		UVD_LMI_CTRL__REQ_MODE_MASK |
+		UVD_LMI_CTRL__CRC_RESET_MASK |
+		UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
+		UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
+		UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK |
+		(8 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) |
+		0x00100000L);
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_LMI_CTRL), tmp, 0, indirect);
+
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_MPC_CNTL),
+		0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT, 0, indirect);
+
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_MPC_SET_MUXA0),
+		((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
+		 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
+		 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
+		 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)), 0, indirect);
+
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_MPC_SET_MUXB0),
+		((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
+		 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
+		 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
+		 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)), 0, indirect);
+
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_MPC_SET_MUX),
+		((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
+		 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
+		 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)), 0, indirect);
+
+	vcn_v2_5_mc_resume_dpg_mode(adev, inst_idx, indirect);
+
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_REG_XX_MASK), 0x10, 0, indirect);
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_RBC_XX_IB_REG_CHECK), 0x3, 0, indirect);
+
+	/* enable LMI MC and UMC channels */
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_LMI_CTRL2), 0, 0, indirect);
+
+	/* unblock VCPU register access */
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_RB_ARB_CTRL), 0, 0, indirect);
+
+	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
+	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_VCPU_CNTL), tmp, 0, indirect);
+
+	/* enable master interrupt */
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_MASTINT_EN),
+		UVD_MASTINT_EN__VCPU_EN_MASK, 0, indirect);
+
+	ring = &adev->vcn.inst[inst_idx].ring_dec;
+	/* force RBC into idle state */
+	rb_bufsz = order_base_2(ring->ring_size);
+	tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
+	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
+	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
+	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
+	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
+	WREG32_SOC15(UVD, inst_idx, mmUVD_RBC_RB_CNTL, tmp);
+
+	/* set the write pointer delay */
+	WREG32_SOC15(UVD, inst_idx, mmUVD_RBC_RB_WPTR_CNTL, 0);
+
+	/* set the wb address */
+	WREG32_SOC15(UVD, inst_idx, mmUVD_RBC_RB_RPTR_ADDR,
+		(upper_32_bits(ring->gpu_addr) >> 2));
+
+	/* programm the RB_BASE for ring buffer */
+	WREG32_SOC15(UVD, inst_idx, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,
+		lower_32_bits(ring->gpu_addr));
+	WREG32_SOC15(UVD, inst_idx, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH,
+		upper_32_bits(ring->gpu_addr));
+
+	/* Initialize the ring buffer's read and write pointers */
+	WREG32_SOC15(UVD, inst_idx, mmUVD_RBC_RB_RPTR, 0);
+
+	WREG32_SOC15(UVD, inst_idx, mmUVD_SCRATCH2, 0);
+
+	ring->wptr = RREG32_SOC15(UVD, inst_idx, mmUVD_RBC_RB_RPTR);
+	WREG32_SOC15(UVD, inst_idx, mmUVD_RBC_RB_WPTR,
+		lower_32_bits(ring->wptr));
+
+	return 0;
+}
+
  static int vcn_v2_5_start(struct amdgpu_device *adev)
  {
  	struct amdgpu_ring *ring;
@@ -610,6 +862,9 @@ static int vcn_v2_5_start(struct amdgpu_device *adev)
  	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
  		if (adev->vcn.harvest_config & (1 << i))
  			continue;
+		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
+			return vcn_v2_5_start_dpg_mode(adev, i, 0);
+
  		/* disable register anti-hang mechanism */
  		WREG32_P(SOC15_REG_OFFSET(UVD, i, mmUVD_POWER_STATUS), 0,
  			~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
@@ -973,6 +1228,35 @@ static int vcn_v2_5_sriov_start(struct amdgpu_device *adev)
  	return vcn_v2_5_mmsch_start(adev, &adev->virt.mm_table);
  }
 
+static int vcn_v2_5_stop_dpg_mode(struct amdgpu_device *adev, int inst_idx)
+{
+	int ret_code = 0;
+	uint32_t tmp;
+
+	/* Wait for power status to be 1 */
+	SOC15_WAIT_ON_RREG(UVD, inst_idx, mmUVD_POWER_STATUS, 1,
+		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK, ret_code);
+
+	/* wait for read ptr to be equal to write ptr */
+	tmp = RREG32_SOC15(UVD, inst_idx, mmUVD_RB_WPTR);
+	SOC15_WAIT_ON_RREG(UVD, inst_idx, mmUVD_RB_RPTR, tmp, 0xFFFFFFFF, ret_code);
+
+	tmp = RREG32_SOC15(UVD, inst_idx, mmUVD_RB_WPTR2);
+	SOC15_WAIT_ON_RREG(UVD, inst_idx, mmUVD_RB_RPTR2, tmp, 0xFFFFFFFF, ret_code);
+
+	tmp = RREG32_SOC15(UVD, inst_idx, mmUVD_RBC_RB_WPTR) & 0x7FFFFFFF;
+	SOC15_WAIT_ON_RREG(UVD, inst_idx, mmUVD_RBC_RB_RPTR, tmp, 0xFFFFFFFF, ret_code);
+
+	SOC15_WAIT_ON_RREG(UVD, inst_idx, mmUVD_POWER_STATUS, 1,
+		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK, ret_code);
+
+	/* disable dynamic power gating mode */
+	WREG32_P(SOC15_REG_OFFSET(UVD, inst_idx, mmUVD_POWER_STATUS), 0,
+			~UVD_POWER_STATUS__UVD_PG_MODE_MASK);
+
+	return 0;
+}
+
  static int vcn_v2_5_stop(struct amdgpu_device *adev)
  {
  	uint32_t tmp;
@@ -981,6 +1265,12 @@ static int vcn_v2_5_stop(struct amdgpu_device *adev)
  	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
  		if (adev->vcn.harvest_config & (1 << i))
  			continue;
+
+		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
+			r = vcn_v2_5_stop_dpg_mode(adev, i);
+			goto power_off;
+		}
+
  		/* wait for vcn idle */
  		SOC15_WAIT_ON_RREG(VCN, i, mmUVD_STATUS, UVD_STATUS__IDLE, 0x7, r);
  		if (r)

drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c static int vcn_v2_5_stop(struct amdgpu_device *adev) · line 1326

Quoted 7 lines
drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c +1 -0
@@ -1030,6 +1320,7 @@ static int vcn_v2_5_stop(struct amdgpu_device *adev)
  			~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
  	}
 
+power_off:
  	if (adev->pm.dpm_enabled)
  		amdgpu_dpm_enable_uvd(adev, false);
 

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

author
#7 lore raw

PATCH v4 3/6 drm/amdgpu/vcn: move macro from vcn2.0 to share amdgpu_vcn

In reply to James Zhu

Move macro from vcn2.0 to amdgpu_vcn to share with vcn2.5

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h | 5 +++++
 drivers/gpu/drm/amd/amdgpu/vcn_v2_0.c   | 5 -----
 2 files changed, 5 insertions(+), 5 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h +5 -0
index 5ce13c0..c4984c5 100644
@@ -57,6 +57,11 @@
 #define VCN_VID_IP_ADDRESS_2_0		0x0
 #define VCN_AON_IP_ADDRESS_2_0		0x30000
 
+#define mmUVD_RBC_XX_IB_REG_CHECK 					0x026b
+#define mmUVD_RBC_XX_IB_REG_CHECK_BASE_IDX 				1
+#define mmUVD_REG_XX_MASK 						0x026c
+#define mmUVD_REG_XX_MASK_BASE_IDX 					1
+
 /* 1 second timeout */
 #define VCN_IDLE_TIMEOUT	msecs_to_jiffies(1000)
 
drivers/gpu/drm/amd/amdgpu/vcn_v2_0.c +0 -5
index 9ff59ac..f4db8af6 100644
@@ -47,11 +47,6 @@
 #define mmUVD_LMI_RBC_IB_64BIT_BAR_LOW_INTERNAL_OFFSET		0x5a7
 #define mmUVD_RBC_IB_SIZE_INTERNAL_OFFSET			0x1e2
 
-#define mmUVD_RBC_XX_IB_REG_CHECK 					0x026b
-#define mmUVD_RBC_XX_IB_REG_CHECK_BASE_IDX 				1
-#define mmUVD_REG_XX_MASK 						0x026c
-#define mmUVD_REG_XX_MASK_BASE_IDX 					1
-
 static void vcn_v2_0_set_dec_ring_funcs(struct amdgpu_device *adev);
 static void vcn_v2_0_set_enc_ring_funcs(struct amdgpu_device *adev);
 static void vcn_v2_0_set_irq_funcs(struct amdgpu_device *adev);
Signature · 6 lines
2.7.4

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
author
#8 lore raw

PATCH v4 4/6 drm/amdgpu/vcn2.5: add DPG mode start and stop

Add DPG mode start and stop functions for vcn2.5

v2: Correct firmware ucode index in vcn_v2_5_mc_resume_dpg_mode

---
 drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c | 290 +++++++++++++++++++++++++++++++++-
 1 file changed, 288 insertions(+), 2 deletions(-)
drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c +288 -2
index fa90249..c3ff336 100644
@@ -286,7 +286,8 @@ static int vcn_v2_5_hw_init(void *handle)
 
 done:
 	if (!r)
-		DRM_INFO("VCN decode and encode initialized successfully.\n");
+		DRM_INFO("VCN decode and encode initialized successfully(under %s).\n",
+			(adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)?"DPG Mode":"SPG Mode");
 
 	return r;
 }
@@ -309,7 +310,9 @@ static int vcn_v2_5_hw_fini(void *handle)
 			continue;
 		ring = &adev->vcn.inst[i].ring_dec;
 
-		if (RREG32_SOC15(VCN, i, mmUVD_STATUS))
+		if ((adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) ||
+	        (adev->vcn.cur_state != AMD_PG_STATE_GATE &&
+	         RREG32_SOC15(VCN, i, mmUVD_STATUS)))
 			vcn_v2_5_set_powergating_state(adev, AMD_PG_STATE_GATE);
 
 		ring->sched.ready = false;
@@ -418,6 +421,78 @@ static void vcn_v2_5_mc_resume(struct amdgpu_device *adev)
 	}
 }
 
+static void vcn_v2_5_mc_resume_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
+{
+	uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
+	uint32_t offset;
+
+	/* cache window 0: fw */
+	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
+		WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+			UVD, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
+			(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_lo), 0, indirect);
+		WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+			UVD, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
+			(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_hi), 0, indirect);
+		WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+			UVD, inst_idx, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
+		offset = 0;
+	} else {
+		WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+			UVD, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
+			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
+		WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+			UVD, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
+			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
+		offset = size;
+		WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+			UVD, inst_idx, mmUVD_VCPU_CACHE_OFFSET0),
+			AMDGPU_UVD_FIRMWARE_OFFSET >> 3, 0, indirect);
+	}
+
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_VCPU_CACHE_SIZE0), size, 0, indirect);
+
+	/* cache window 1: stack */
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
+		lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
+		upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
+
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_VCPU_CACHE_SIZE1), AMDGPU_VCN_STACK_SIZE, 0, indirect);
+
+	/* cache window 2: context */
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
+		lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
+		upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_VCPU_CACHE_OFFSET2), 0, 0, indirect);
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_VCPU_CACHE_SIZE2), AMDGPU_VCN_CONTEXT_SIZE, 0, indirect);
+
+	/* non-cache window */
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW), 0, 0, indirect);
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH), 0, 0, indirect);
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_VCPU_NONCACHE_OFFSET0), 0, 0, indirect);
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_VCPU_NONCACHE_SIZE0), 0, 0, indirect);
+
+	/* VCN global tiling registers */
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_GFX8_ADDR_CONFIG), adev->gfx.config.gb_addr_config, 0, indirect);
+}
+
 /**
  * vcn_v2_5_disable_clock_gating - disable VCN clock gating
  *
@@ -536,6 +611,54 @@ static void vcn_v2_5_disable_clock_gating(struct amdgpu_device *adev)
 	}
 }
 
+static void vcn_v2_5_clock_gating_dpg_mode(struct amdgpu_device *adev,
+		uint8_t sram_sel, int inst_idx, uint8_t indirect)
+{
+	uint32_t reg_data = 0;
+
+	/* enable sw clock gating control */
+	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
+		reg_data = 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
+	else
+		reg_data = 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
+	reg_data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
+	reg_data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
+	reg_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__MMSCH_MODE_MASK);
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_CGC_CTRL), reg_data, sram_sel, indirect);
+
+	/* turn off clock gating */
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_CGC_GATE), 0, sram_sel, indirect);
+
+	/* turn on SUVD clock gating */
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_SUVD_CGC_GATE), 1, sram_sel, indirect);
+
+	/* turn on sw mode in UVD_SUVD_CGC_CTRL */
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_SUVD_CGC_CTRL), 0, sram_sel, indirect);
+}
+
 /**
  * vcn_v2_5_enable_clock_gating - enable VCN clock gating
  *
@@ -598,6 +721,130 @@ static void vcn_v2_5_enable_clock_gating(struct amdgpu_device *adev)
 	}
 }
 
+static int vcn_v2_5_start_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
+{
+	struct amdgpu_ring *ring;
+	uint32_t rb_bufsz, tmp;
+
+	/* disable register anti-hang mechanism */
+	WREG32_P(SOC15_REG_OFFSET(UVD, inst_idx, mmUVD_POWER_STATUS), 1,
+		~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
+	/* enable dynamic power gating mode */
+	tmp = RREG32_SOC15(UVD, inst_idx, mmUVD_POWER_STATUS);
+	tmp |= UVD_POWER_STATUS__UVD_PG_MODE_MASK;
+	tmp |= UVD_POWER_STATUS__UVD_PG_EN_MASK;
+	WREG32_SOC15(UVD, inst_idx, mmUVD_POWER_STATUS, tmp);
+
+	/* enable clock gating */
+	vcn_v2_5_clock_gating_dpg_mode(adev, 0, inst_idx, indirect);
+
+	/* enable VCPU clock */
+	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
+	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
+	tmp |= UVD_VCPU_CNTL__BLK_RST_MASK;
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_VCPU_CNTL), tmp, 0, indirect);
+
+	/* disable master interupt */
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_MASTINT_EN), 0, 0, indirect);
+
+	/* setup mmUVD_LMI_CTRL */
+	tmp = (0x8 | UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
+		UVD_LMI_CTRL__REQ_MODE_MASK |
+		UVD_LMI_CTRL__CRC_RESET_MASK |
+		UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
+		UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
+		UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK |
+		(8 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) |
+		0x00100000L);
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_LMI_CTRL), tmp, 0, indirect);
+
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_MPC_CNTL),
+		0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT, 0, indirect);
+
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_MPC_SET_MUXA0),
+		((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
+		 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
+		 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
+		 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)), 0, indirect);
+
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_MPC_SET_MUXB0),
+		((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
+		 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
+		 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
+		 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)), 0, indirect);
+
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_MPC_SET_MUX),
+		((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
+		 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
+		 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)), 0, indirect);
+
+	vcn_v2_5_mc_resume_dpg_mode(adev, inst_idx, indirect);
+
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_REG_XX_MASK), 0x10, 0, indirect);
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_RBC_XX_IB_REG_CHECK), 0x3, 0, indirect);
+
+	/* enable LMI MC and UMC channels */
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_LMI_CTRL2), 0, 0, indirect);
+
+	/* unblock VCPU register access */
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_RB_ARB_CTRL), 0, 0, indirect);
+
+	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
+	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_VCPU_CNTL), tmp, 0, indirect);
+
+	/* enable master interrupt */
+	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
+		UVD, inst_idx, mmUVD_MASTINT_EN),
+		UVD_MASTINT_EN__VCPU_EN_MASK, 0, indirect);
+
+	ring = &adev->vcn.inst[inst_idx].ring_dec;
+	/* force RBC into idle state */
+	rb_bufsz = order_base_2(ring->ring_size);
+	tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
+	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
+	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
+	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
+	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
+	WREG32_SOC15(UVD, inst_idx, mmUVD_RBC_RB_CNTL, tmp);
+
+	/* set the write pointer delay */
+	WREG32_SOC15(UVD, inst_idx, mmUVD_RBC_RB_WPTR_CNTL, 0);
+
+	/* set the wb address */
+	WREG32_SOC15(UVD, inst_idx, mmUVD_RBC_RB_RPTR_ADDR,
+		(upper_32_bits(ring->gpu_addr) >> 2));
+
+	/* programm the RB_BASE for ring buffer */
+	WREG32_SOC15(UVD, inst_idx, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,
+		lower_32_bits(ring->gpu_addr));
+	WREG32_SOC15(UVD, inst_idx, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH,
+		upper_32_bits(ring->gpu_addr));
+
+	/* Initialize the ring buffer's read and write pointers */
+	WREG32_SOC15(UVD, inst_idx, mmUVD_RBC_RB_RPTR, 0);
+
+	WREG32_SOC15(UVD, inst_idx, mmUVD_SCRATCH2, 0);
+
+	ring->wptr = RREG32_SOC15(UVD, inst_idx, mmUVD_RBC_RB_RPTR);
+	WREG32_SOC15(UVD, inst_idx, mmUVD_RBC_RB_WPTR,
+		lower_32_bits(ring->wptr));
+
+	return 0;
+}
+
 static int vcn_v2_5_start(struct amdgpu_device *adev)
 {
 	struct amdgpu_ring *ring;
@@ -610,6 +857,9 @@ static int vcn_v2_5_start(struct amdgpu_device *adev)
 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
 		if (adev->vcn.harvest_config & (1 << i))
 			continue;
+		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
+			return vcn_v2_5_start_dpg_mode(adev, i, 0);
+
 		/* disable register anti-hang mechanism */
 		WREG32_P(SOC15_REG_OFFSET(UVD, i, mmUVD_POWER_STATUS), 0,
 			~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
@@ -973,6 +1223,35 @@ static int vcn_v2_5_sriov_start(struct amdgpu_device *adev)
 	return vcn_v2_5_mmsch_start(adev, &adev->virt.mm_table);
 }
 
+static int vcn_v2_5_stop_dpg_mode(struct amdgpu_device *adev, int inst_idx)
+{
+	int ret_code = 0;
+	uint32_t tmp;
+
+	/* Wait for power status to be 1 */
+	SOC15_WAIT_ON_RREG(UVD, inst_idx, mmUVD_POWER_STATUS, 1,
+		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK, ret_code);
+
+	/* wait for read ptr to be equal to write ptr */
+	tmp = RREG32_SOC15(UVD, inst_idx, mmUVD_RB_WPTR);
+	SOC15_WAIT_ON_RREG(UVD, inst_idx, mmUVD_RB_RPTR, tmp, 0xFFFFFFFF, ret_code);
+
+	tmp = RREG32_SOC15(UVD, inst_idx, mmUVD_RB_WPTR2);
+	SOC15_WAIT_ON_RREG(UVD, inst_idx, mmUVD_RB_RPTR2, tmp, 0xFFFFFFFF, ret_code);
+
+	tmp = RREG32_SOC15(UVD, inst_idx, mmUVD_RBC_RB_WPTR) & 0x7FFFFFFF;
+	SOC15_WAIT_ON_RREG(UVD, inst_idx, mmUVD_RBC_RB_RPTR, tmp, 0xFFFFFFFF, ret_code);
+
+	SOC15_WAIT_ON_RREG(UVD, inst_idx, mmUVD_POWER_STATUS, 1,
+		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK, ret_code);
+
+	/* disable dynamic power gating mode */
+	WREG32_P(SOC15_REG_OFFSET(UVD, inst_idx, mmUVD_POWER_STATUS), 0,
+			~UVD_POWER_STATUS__UVD_PG_MODE_MASK);
+
+	return 0;
+}
+
 static int vcn_v2_5_stop(struct amdgpu_device *adev)
 {
 	uint32_t tmp;
@@ -981,6 +1260,12 @@ static int vcn_v2_5_stop(struct amdgpu_device *adev)
 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
 		if (adev->vcn.harvest_config & (1 << i))
 			continue;
+
+		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
+			r = vcn_v2_5_stop_dpg_mode(adev, i);
+			goto power_off;
+		}
+
 		/* wait for vcn idle */
 		SOC15_WAIT_ON_RREG(VCN, i, mmUVD_STATUS, UVD_STATUS__IDLE, 0x7, r);
 		if (r)
@@ -1030,6 +1315,7 @@ static int vcn_v2_5_stop(struct amdgpu_device *adev)
 			~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
 	}
 
+power_off:
 	if (adev->pm.dpm_enabled)
 		amdgpu_dpm_enable_uvd(adev, false);
 
Signature · 6 lines
2.7.4

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
author
#9 lore raw

PATCH v3 4/5 drm/amdgpu/vcn2.5: add dpg pause mode

In reply to James Zhu

Add dpg pause mode support for vcn2.5

---
 drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c | 70 +++++++++++++++++++++++++++++++++++
 1 file changed, 70 insertions(+)
drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c +70 -0
index b3ddf68..7ce5a9e 100644
@@ -60,6 +60,8 @@ static void vcn_v2_5_set_enc_ring_funcs(struct amdgpu_device *adev);
 static void vcn_v2_5_set_irq_funcs(struct amdgpu_device *adev);
 static int vcn_v2_5_set_powergating_state(void *handle,
 				enum amd_powergating_state state);
+static int vcn_v2_5_pause_dpg_mode(struct amdgpu_device *adev,
+				int inst_idx, struct dpg_pause_state *new_state);
 static int vcn_v2_5_sriov_start(struct amdgpu_device *adev);
 
 static int amdgpu_ih_clientid_vcns[] = {
@@ -217,6 +219,9 @@ static int vcn_v2_5_sw_init(void *handle)
 			return r;
 	}
 
+	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
+		adev->vcn.pause_dpg_mode = vcn_v2_5_pause_dpg_mode;
+
 	return 0;
 }
 
@@ -1327,6 +1332,67 @@ static int vcn_v2_5_stop(struct amdgpu_device *adev)
 	return 0;
 }
 
+static int vcn_v2_5_pause_dpg_mode(struct amdgpu_device *adev,
+				int inst_idx, struct dpg_pause_state *new_state)
+{
+	struct amdgpu_ring *ring;
+	uint32_t reg_data = 0;
+	int ret_code;
+
+	/* pause/unpause if state is changed */
+	if (adev->vcn.pause_state.fw_based != new_state->fw_based) {
+		DRM_DEBUG("dpg pause state changed %d -> %d",
+			adev->vcn.pause_state.fw_based,	new_state->fw_based);
+		reg_data = RREG32_SOC15(UVD, inst_idx, mmUVD_DPG_PAUSE) &
+			(~UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
+
+		if (new_state->fw_based == VCN_DPG_STATE__PAUSE) {
+			ret_code = 0;
+			SOC15_WAIT_ON_RREG(UVD, inst_idx, mmUVD_POWER_STATUS, 0x1,
+				UVD_POWER_STATUS__UVD_POWER_STATUS_MASK, ret_code);
+
+			if (!ret_code) {
+				/* pause DPG */
+				reg_data |= UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
+				WREG32_SOC15(UVD, inst_idx, mmUVD_DPG_PAUSE, reg_data);
+
+				/* wait for ACK */
+				SOC15_WAIT_ON_RREG(UVD, inst_idx, mmUVD_DPG_PAUSE,
+					   UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK,
+					   UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK, ret_code);
+
+				/* Restore */
+				ring = &adev->vcn.inst[inst_idx].ring_enc[0];
+				WREG32_SOC15(UVD, inst_idx, mmUVD_RB_BASE_LO, ring->gpu_addr);
+				WREG32_SOC15(UVD, inst_idx, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
+				WREG32_SOC15(UVD, inst_idx, mmUVD_RB_SIZE, ring->ring_size / 4);
+				WREG32_SOC15(UVD, inst_idx, mmUVD_RB_RPTR, lower_32_bits(ring->wptr));
+				WREG32_SOC15(UVD, inst_idx, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
+
+				ring = &adev->vcn.inst[inst_idx].ring_enc[1];
+				WREG32_SOC15(UVD, inst_idx, mmUVD_RB_BASE_LO2, ring->gpu_addr);
+				WREG32_SOC15(UVD, inst_idx, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
+				WREG32_SOC15(UVD, inst_idx, mmUVD_RB_SIZE2, ring->ring_size / 4);
+				WREG32_SOC15(UVD, inst_idx, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr));
+				WREG32_SOC15(UVD, inst_idx, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
+
+				WREG32_SOC15(UVD, inst_idx, mmUVD_RBC_RB_WPTR,
+					   RREG32_SOC15(UVD, inst_idx, mmUVD_SCRATCH2) & 0x7FFFFFFF);
+
+				SOC15_WAIT_ON_RREG(UVD, inst_idx, mmUVD_POWER_STATUS,
+					   0x0, UVD_POWER_STATUS__UVD_POWER_STATUS_MASK, ret_code);
+			}
+		} else {
+			/* unpause dpg, no need to wait */
+			reg_data &= ~UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
+			WREG32_SOC15(UVD, inst_idx, mmUVD_DPG_PAUSE, reg_data);
+		}
+		adev->vcn.pause_state.fw_based = new_state->fw_based;
+	}
+
+	return 0;
+}
+
 /**
  * vcn_v2_5_dec_ring_get_rptr - get read pointer
  *
@@ -1369,6 +1435,10 @@ static void vcn_v2_5_dec_ring_set_wptr(struct amdgpu_ring *ring)
 {
 	struct amdgpu_device *adev = ring->adev;
 
+	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
+		WREG32_SOC15(UVD, ring->me, mmUVD_SCRATCH2,
+			lower_32_bits(ring->wptr) | 0x80000000);
+
 	if (ring->use_doorbell) {
 		adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr);
 		WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
Signature · 6 lines
2.7.4

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
author
#10 lore raw

PATCH v3 5/5 drm/amdgpu/vcn2.5: implement indirect DPG SRAM mode

In reply to James Zhu

Implement indirect DPG SRAM mode for vcn2.5

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c |  3 ++
 drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c   | 69 +++++++++++++++++++++++----------
 2 files changed, 52 insertions(+), 20 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c +3 -0
index ca62d99..ab51f0e 100644
@@ -75,6 +75,9 @@ int amdgpu_vcn_sw_init(struct amdgpu_device *adev)
 		break;
 	case CHIP_ARCTURUS:
 		fw_name = FIRMWARE_ARCTURUS;
+		if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
+		    (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
+			adev->vcn.indirect_sram = true;
 		break;
 	case CHIP_RENOIR:
 		fw_name = FIRMWARE_RENOIR;
drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c +49 -20

101 lines not on this page · raw message

Signature · 6 lines
2.7.4

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
Reviewed-by on the v3 series
#11 lore raw
In reply to James Zhu
Quoted 5 lines · James Zhu

Add vcn2.5 dpg mode/dpg pause mode/dpg sram mode to support acturus IFM(instruction fetch monitor) work around.

v2: Correct firmware ucode index in vcn_v2_5_mc_resume_dpg_mode
v3: Share multiple instance indirect DPG SRAM mode support for vcn2

With those fixed in patch2&3, the series are:

Regards,

Leo

Quoted 9 lines

James Zhu (5):
drm/amdgpu/vcn: support multiple-instance dpg pause mode drm/amdgpu/vcn2.5: support multiple instance direct SRAM read and
write
drm/amdgpu/vcn2.5: add DPG mode start and stop drm/amdgpu/vcn2.5: add dpg pause mode drm/amdgpu/vcn2.5: implement indirect DPG SRAM mode

Quoted 7 lines
  drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c |  34 +--
  drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h |  48 ++--
  drivers/gpu/drm/amd/amdgpu/vcn_v1_0.c   |   8 +-
  drivers/gpu/drm/amd/amdgpu/vcn_v2_0.c   |  98 ++++----
  drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c   | 394 +++++++++++++++++++++++++++++++-
  5 files changed, 487 insertions(+), 95 deletions(-)

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx