drm/amdgpu: simplify IH programming

Calculate all the addresses and pointers in amdgpu_ih.c
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

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

@@ -52,6 +52,8 @@ int amdgpu_ih_ring_init(struct amdgpu_device *adev, struct amdgpu_ih_ring *ih,
 	ih->use_bus_addr = use_bus_addr;
 
 	if (use_bus_addr) {
+		dma_addr_t dma_addr;
+
 		if (ih->ring)
 			return 0;
 
@@ -59,21 +61,26 @@ int amdgpu_ih_ring_init(struct amdgpu_device *adev, struct amdgpu_ih_ring *ih,
 		 * add them to the end of the ring allocation.
 		 */
 		ih->ring = dma_alloc_coherent(adev->dev, ih->ring_size + 8,
-					      &ih->rb_dma_addr, GFP_KERNEL);
+					      &dma_addr, GFP_KERNEL);
 		if (ih->ring == NULL)
 			return -ENOMEM;
 
 		memset((void *)ih->ring, 0, ih->ring_size + 8);
-		ih->wptr_offs = (ih->ring_size / 4) + 0;
-		ih->rptr_offs = (ih->ring_size / 4) + 1;
+		ih->gpu_addr = dma_addr;
+		ih->wptr_addr = dma_addr + ih->ring_size;
+		ih->wptr_cpu = &ih->ring[ih->ring_size / 4];
+		ih->rptr_addr = dma_addr + ih->ring_size + 4;
+		ih->rptr_cpu = &ih->ring[(ih->ring_size / 4) + 1];
 	} else {
-		r = amdgpu_device_wb_get(adev, &ih->wptr_offs);
+		unsigned wptr_offs, rptr_offs;
+
+		r = amdgpu_device_wb_get(adev, &wptr_offs);
 		if (r)
 			return r;
 
-		r = amdgpu_device_wb_get(adev, &ih->rptr_offs);
+		r = amdgpu_device_wb_get(adev, &rptr_offs);
 		if (r) {
-			amdgpu_device_wb_free(adev, ih->wptr_offs);
+			amdgpu_device_wb_free(adev, wptr_offs);
 			return r;
 		}
 
@@ -82,10 +89,15 @@ int amdgpu_ih_ring_init(struct amdgpu_device *adev, struct amdgpu_ih_ring *ih,
 					    &ih->ring_obj, &ih->gpu_addr,
 					    (void **)&ih->ring);
 		if (r) {
-			amdgpu_device_wb_free(adev, ih->rptr_offs);
-			amdgpu_device_wb_free(adev, ih->wptr_offs);
+			amdgpu_device_wb_free(adev, rptr_offs);
+			amdgpu_device_wb_free(adev, wptr_offs);
 			return r;
 		}
+
+		ih->wptr_addr = adev->wb.gpu_addr + wptr_offs * 4;
+		ih->wptr_cpu = &adev->wb.wb[wptr_offs];
+		ih->rptr_addr = adev->wb.gpu_addr + rptr_offs * 4;
+		ih->rptr_cpu = &adev->wb.wb[rptr_offs];
 	}
 	return 0;
 }
@@ -109,13 +121,13 @@ void amdgpu_ih_ring_fini(struct amdgpu_device *adev, struct amdgpu_ih_ring *ih)
 		 * add them to the end of the ring allocation.
 		 */
 		dma_free_coherent(adev->dev, ih->ring_size + 8,
-				  (void *)ih->ring, ih->rb_dma_addr);
+				  (void *)ih->ring, ih->gpu_addr);
 		ih->ring = NULL;
 	} else {
 		amdgpu_bo_free_kernel(&ih->ring_obj, &ih->gpu_addr,
 				      (void **)&ih->ring);
-		amdgpu_device_wb_free(adev, ih->wptr_offs);
-		amdgpu_device_wb_free(adev, ih->rptr_offs);
+		amdgpu_device_wb_free(adev, (ih->wptr_addr - ih->gpu_addr) / 4);
+		amdgpu_device_wb_free(adev, (ih->rptr_addr - ih->gpu_addr) / 4);
 	}
 }
 

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

@@ -31,20 +31,25 @@ struct amdgpu_iv_entry;
  * R6xx+ IH ring
  */
 struct amdgpu_ih_ring {
-	struct amdgpu_bo	*ring_obj;
-	volatile uint32_t	*ring;
-	unsigned		rptr;
 	unsigned		ring_size;
-	uint64_t		gpu_addr;
 	uint32_t		ptr_mask;
-	atomic_t		lock;
-	bool                    enabled;
-	unsigned		wptr_offs;
-	unsigned		rptr_offs;
 	u32			doorbell_index;
 	bool			use_doorbell;
 	bool			use_bus_addr;
-	dma_addr_t		rb_dma_addr; /* only used when use_bus_addr = true */
+
+	struct amdgpu_bo	*ring_obj;
+	volatile uint32_t	*ring;
+	uint64_t		gpu_addr;
+
+	uint64_t		wptr_addr;
+	volatile uint32_t	*wptr_cpu;
+
+	uint64_t		rptr_addr;
+	volatile uint32_t	*rptr_cpu;
+
+	bool                    enabled;
+	unsigned		rptr;
+	atomic_t		lock;
 };
 
 /* provided by the ih block */

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

@@ -103,9 +103,9 @@ static void cik_ih_disable_interrupts(struct amdgpu_device *adev)
  */
 static int cik_ih_irq_init(struct amdgpu_device *adev)
 {
+	struct amdgpu_ih_ring *ih = &adev->irq.ih;
 	int rb_bufsz;
 	u32 interrupt_cntl, ih_cntl, ih_rb_cntl;
-	u64 wptr_off;
 
 	/* disable irqs */
 	cik_ih_disable_interrupts(adev);
@@ -131,9 +131,8 @@ static int cik_ih_irq_init(struct amdgpu_device *adev)
 	ih_rb_cntl |= IH_RB_CNTL__WPTR_WRITEBACK_ENABLE_MASK;
 
 	/* set the writeback address whether it's enabled or not */
-	wptr_off = adev->wb.gpu_addr + (adev->irq.ih.wptr_offs * 4);
-	WREG32(mmIH_RB_WPTR_ADDR_LO, lower_32_bits(wptr_off));
-	WREG32(mmIH_RB_WPTR_ADDR_HI, upper_32_bits(wptr_off) & 0xFF);
+	WREG32(mmIH_RB_WPTR_ADDR_LO, lower_32_bits(ih->wptr_addr));
+	WREG32(mmIH_RB_WPTR_ADDR_HI, upper_32_bits(ih->wptr_addr) & 0xFF);
 
 	WREG32(mmIH_RB_CNTL, ih_rb_cntl);
 
@@ -188,7 +187,7 @@ static u32 cik_ih_get_wptr(struct amdgpu_device *adev,
 {
 	u32 wptr, tmp;
 
-	wptr = le32_to_cpu(adev->wb.wb[ih->wptr_offs]);
+	wptr = le32_to_cpu(*ih->wptr_cpu);
 
 	if (wptr & IH_RB_WPTR__RB_OVERFLOW_MASK) {
 		wptr &= ~IH_RB_WPTR__RB_OVERFLOW_MASK;

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

@@ -103,9 +103,9 @@ static void cz_ih_disable_interrupts(struct amdgpu_device *adev)
  */
 static int cz_ih_irq_init(struct amdgpu_device *adev)
 {
-	int rb_bufsz;
+	struct amdgpu_ih_ring *ih = &adev->irq.ih;
 	u32 interrupt_cntl, ih_cntl, ih_rb_cntl;
-	u64 wptr_off;
+	int rb_bufsz;
 
 	/* disable irqs */
 	cz_ih_disable_interrupts(adev);
@@ -133,9 +133,8 @@ static int cz_ih_irq_init(struct amdgpu_device *adev)
 	ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, WPTR_WRITEBACK_ENABLE, 1);
 
 	/* set the writeback address whether it's enabled or not */
-	wptr_off = adev->wb.gpu_addr + (adev->irq.ih.wptr_offs * 4);
-	WREG32(mmIH_RB_WPTR_ADDR_LO, lower_32_bits(wptr_off));
-	WREG32(mmIH_RB_WPTR_ADDR_HI, upper_32_bits(wptr_off) & 0xFF);
+	WREG32(mmIH_RB_WPTR_ADDR_LO, lower_32_bits(ih->wptr_addr));
+	WREG32(mmIH_RB_WPTR_ADDR_HI, upper_32_bits(ih->wptr_addr) & 0xFF);
 
 	WREG32(mmIH_RB_CNTL, ih_rb_cntl);
 
@@ -190,7 +189,7 @@ static u32 cz_ih_get_wptr(struct amdgpu_device *adev,
 {
 	u32 wptr, tmp;
 
-	wptr = le32_to_cpu(adev->wb.wb[ih->wptr_offs]);
+	wptr = le32_to_cpu(*ih->wptr_cpu);
 
 	if (REG_GET_FIELD(wptr, IH_RB_WPTR, RB_OVERFLOW)) {
 		wptr = REG_SET_FIELD(wptr, IH_RB_WPTR, RB_OVERFLOW, 0);

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

@@ -103,9 +103,9 @@ static void iceland_ih_disable_interrupts(struct amdgpu_device *adev)
  */
 static int iceland_ih_irq_init(struct amdgpu_device *adev)
 {
+	struct amdgpu_ih_ring *ih = &adev->irq.ih;
 	int rb_bufsz;
 	u32 interrupt_cntl, ih_cntl, ih_rb_cntl;
-	u64 wptr_off;
 
 	/* disable irqs */
 	iceland_ih_disable_interrupts(adev);
@@ -133,9 +133,8 @@ static int iceland_ih_irq_init(struct amdgpu_device *adev)
 	ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, WPTR_WRITEBACK_ENABLE, 1);
 
 	/* set the writeback address whether it's enabled or not */
-	wptr_off = adev->wb.gpu_addr + (adev->irq.ih.wptr_offs * 4);
-	WREG32(mmIH_RB_WPTR_ADDR_LO, lower_32_bits(wptr_off));
-	WREG32(mmIH_RB_WPTR_ADDR_HI, upper_32_bits(wptr_off) & 0xFF);
+	WREG32(mmIH_RB_WPTR_ADDR_LO, lower_32_bits(ih->wptr_addr));
+	WREG32(mmIH_RB_WPTR_ADDR_HI, upper_32_bits(ih->wptr_addr) & 0xFF);
 
 	WREG32(mmIH_RB_CNTL, ih_rb_cntl);
 
@@ -190,7 +189,7 @@ static u32 iceland_ih_get_wptr(struct amdgpu_device *adev,
 {
 	u32 wptr, tmp;
 
-	wptr = le32_to_cpu(adev->wb.wb[ih->wptr_offs]);
+	wptr = le32_to_cpu(*ih->wptr_cpu);
 
 	if (REG_GET_FIELD(wptr, IH_RB_WPTR, RB_OVERFLOW)) {
 		wptr = REG_SET_FIELD(wptr, IH_RB_WPTR, RB_OVERFLOW, 0);

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

@@ -57,9 +57,9 @@ static void si_ih_disable_interrupts(struct amdgpu_device *adev)
 
 static int si_ih_irq_init(struct amdgpu_device *adev)
 {
+	struct amdgpu_ih_ring *ih = &adev->irq.ih;
 	int rb_bufsz;
 	u32 interrupt_cntl, ih_cntl, ih_rb_cntl;
-	u64 wptr_off;
 
 	si_ih_disable_interrupts(adev);
 	WREG32(INTERRUPT_CNTL2, adev->irq.ih.gpu_addr >> 8);
@@ -76,9 +76,8 @@ static int si_ih_irq_init(struct amdgpu_device *adev)
 		     (rb_bufsz << 1) |
 		     IH_WPTR_WRITEBACK_ENABLE;
 
-	wptr_off = adev->wb.gpu_addr + (adev->irq.ih.wptr_offs * 4);
-	WREG32(IH_RB_WPTR_ADDR_LO, lower_32_bits(wptr_off));
-	WREG32(IH_RB_WPTR_ADDR_HI, upper_32_bits(wptr_off) & 0xFF);
+	WREG32(IH_RB_WPTR_ADDR_LO, lower_32_bits(ih->wptr_addr));
+	WREG32(IH_RB_WPTR_ADDR_HI, upper_32_bits(ih->wptr_addr) & 0xFF);
 	WREG32(IH_RB_CNTL, ih_rb_cntl);
 	WREG32(IH_RB_RPTR, 0);
 	WREG32(IH_RB_WPTR, 0);
@@ -105,7 +104,7 @@ static u32 si_ih_get_wptr(struct amdgpu_device *adev,
 {
 	u32 wptr, tmp;
 
-	wptr = le32_to_cpu(adev->wb.wb[ih->wptr_offs]);
+	wptr = le32_to_cpu(*ih->wptr_cpu);
 
 	if (wptr & IH_RB_WPTR__RB_OVERFLOW_MASK) {
 		wptr &= ~IH_RB_WPTR__RB_OVERFLOW_MASK;

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

@@ -99,9 +99,9 @@ static void tonga_ih_disable_interrupts(struct amdgpu_device *adev)
  */
 static int tonga_ih_irq_init(struct amdgpu_device *adev)
 {
-	int rb_bufsz;
 	u32 interrupt_cntl, ih_rb_cntl, ih_doorbell_rtpr;
-	u64 wptr_off;
+	struct amdgpu_ih_ring *ih = &adev->irq.ih;
+	int rb_bufsz;
 
 	/* disable irqs */
 	tonga_ih_disable_interrupts(adev);
@@ -118,10 +118,7 @@ static int tonga_ih_irq_init(struct amdgpu_device *adev)
 	WREG32(mmINTERRUPT_CNTL, interrupt_cntl);
 
 	/* Ring Buffer base. [39:8] of 40-bit address of the beginning of the ring buffer*/
-	if (adev->irq.ih.use_bus_addr)
-		WREG32(mmIH_RB_BASE, adev->irq.ih.rb_dma_addr >> 8);
-	else
-		WREG32(mmIH_RB_BASE, adev->irq.ih.gpu_addr >> 8);
+	WREG32(mmIH_RB_BASE, ih->gpu_addr >> 8);
 
 	rb_bufsz = order_base_2(adev->irq.ih.ring_size / 4);
 	ih_rb_cntl = REG_SET_FIELD(0, IH_RB_CNTL, WPTR_OVERFLOW_CLEAR, 1);
@@ -136,12 +133,8 @@ static int tonga_ih_irq_init(struct amdgpu_device *adev)
 	WREG32(mmIH_RB_CNTL, ih_rb_cntl);
 
 	/* set the writeback address whether it's enabled or not */
-	if (adev->irq.ih.use_bus_addr)
-		wptr_off = adev->irq.ih.rb_dma_addr + (adev->irq.ih.wptr_offs * 4);
-	else
-		wptr_off = adev->wb.gpu_addr + (adev->irq.ih.wptr_offs * 4);
-	WREG32(mmIH_RB_WPTR_ADDR_LO, lower_32_bits(wptr_off));
-	WREG32(mmIH_RB_WPTR_ADDR_HI, upper_32_bits(wptr_off) & 0xFF);
+	WREG32(mmIH_RB_WPTR_ADDR_LO, lower_32_bits(ih->wptr_addr));
+	WREG32(mmIH_RB_WPTR_ADDR_HI, upper_32_bits(ih->wptr_addr) & 0xFF);
 
 	/* set rptr, wptr to 0 */
 	WREG32(mmIH_RB_RPTR, 0);
@@ -198,10 +191,7 @@ static u32 tonga_ih_get_wptr(struct amdgpu_device *adev,
 {
 	u32 wptr, tmp;
 
-	if (adev->irq.ih.use_bus_addr)
-		wptr = le32_to_cpu(ih->ring[ih->wptr_offs]);
-	else
-		wptr = le32_to_cpu(adev->wb.wb[ih->wptr_offs]);
+	wptr = le32_to_cpu(*ih->wptr_cpu);
 
 	if (REG_GET_FIELD(wptr, IH_RB_WPTR, RB_OVERFLOW)) {
 		wptr = REG_SET_FIELD(wptr, IH_RB_WPTR, RB_OVERFLOW, 0);
@@ -263,10 +253,7 @@ static void tonga_ih_set_rptr(struct amdgpu_device *adev,
 {
 	if (ih->use_doorbell) {
 		/* XXX check if swapping is necessary on BE */
-		if (ih->use_bus_addr)
-			ih->ring[ih->rptr_offs] = ih->rptr;
-		else
-			adev->wb.wb[ih->rptr_offs] = ih->rptr;
+		*ih->rptr_cpu = ih->rptr;
 		WDOORBELL32(ih->doorbell_index, ih->rptr);
 	} else {
 		WREG32(mmIH_RB_RPTR, ih->rptr);

drivers/gpu/drm/amd/amdgpu/vega10_ih.c

@@ -86,11 +86,11 @@ static void vega10_ih_disable_interrupts(struct amdgpu_device *adev)
  */
 static int vega10_ih_irq_init(struct amdgpu_device *adev)
 {
+	struct amdgpu_ih_ring *ih = &adev->irq.ih;
 	int ret = 0;
 	int rb_bufsz;
 	u32 ih_rb_cntl, ih_doorbell_rtpr;
 	u32 tmp;
-	u64 wptr_off;
 
 	/* disable irqs */
 	vega10_ih_disable_interrupts(adev);
@@ -99,15 +99,11 @@ static int vega10_ih_irq_init(struct amdgpu_device *adev)
 
 	ih_rb_cntl = RREG32_SOC15(OSSSYS, 0, mmIH_RB_CNTL);
 	/* Ring Buffer base. [39:8] of 40-bit address of the beginning of the ring buffer*/
-	if (adev->irq.ih.use_bus_addr) {
-		WREG32_SOC15(OSSSYS, 0, mmIH_RB_BASE, adev->irq.ih.rb_dma_addr >> 8);
-		WREG32_SOC15(OSSSYS, 0, mmIH_RB_BASE_HI, ((u64)adev->irq.ih.rb_dma_addr >> 40) & 0xff);
-		ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, MC_SPACE, 1);
-	} else {
-		WREG32_SOC15(OSSSYS, 0, mmIH_RB_BASE, adev->irq.ih.gpu_addr >> 8);
-		WREG32_SOC15(OSSSYS, 0, mmIH_RB_BASE_HI, (adev->irq.ih.gpu_addr >> 40) & 0xff);
-		ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, MC_SPACE, 4);
-	}
+	WREG32_SOC15(OSSSYS, 0, mmIH_RB_BASE, adev->irq.ih.gpu_addr >> 8);
+	WREG32_SOC15(OSSSYS, 0, mmIH_RB_BASE_HI,
+		     (adev->irq.ih.gpu_addr >> 40) & 0xff);
+	ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, MC_SPACE,
+				   ih->use_bus_addr ? 1 : 4);
 	rb_bufsz = order_base_2(adev->irq.ih.ring_size / 4);
 	ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, WPTR_OVERFLOW_CLEAR, 1);
 	ih_rb_cntl = REG_SET_FIELD(ih_rb_cntl, IH_RB_CNTL, WPTR_OVERFLOW_ENABLE, 1);
@@ -124,12 +120,10 @@ static int vega10_ih_irq_init(struct amdgpu_device *adev)
 	WREG32_SOC15(OSSSYS, 0, mmIH_RB_CNTL, ih_rb_cntl);
 
 	/* set the writeback address whether it's enabled or not */
-	if (adev->irq.ih.use_bus_addr)
-		wptr_off = adev->irq.ih.rb_dma_addr + (adev->irq.ih.wptr_offs * 4);
-	else
-		wptr_off = adev->wb.gpu_addr + (adev->irq.ih.wptr_offs * 4);
-	WREG32_SOC15(OSSSYS, 0, mmIH_RB_WPTR_ADDR_LO, lower_32_bits(wptr_off));
-	WREG32_SOC15(OSSSYS, 0, mmIH_RB_WPTR_ADDR_HI, upper_32_bits(wptr_off) & 0xFFFF);
+	WREG32_SOC15(OSSSYS, 0, mmIH_RB_WPTR_ADDR_LO,
+		     lower_32_bits(ih->wptr_addr));
+	WREG32_SOC15(OSSSYS, 0, mmIH_RB_WPTR_ADDR_HI,
+		     upper_32_bits(ih->wptr_addr) & 0xFFFF);
 
 	/* set rptr, wptr to 0 */
 	WREG32_SOC15(OSSSYS, 0, mmIH_RB_RPTR, 0);
@@ -196,10 +190,7 @@ static u32 vega10_ih_get_wptr(struct amdgpu_device *adev,
 {
 	u32 wptr, tmp;
 
-	if (ih->use_bus_addr)
-		wptr = le32_to_cpu(ih->ring[ih->wptr_offs]);
-	else
-		wptr = le32_to_cpu(adev->wb.wb[ih->wptr_offs]);
+	wptr = le32_to_cpu(*ih->wptr_cpu);
 
 	if (REG_GET_FIELD(wptr, IH_RB_WPTR, RB_OVERFLOW)) {
 		wptr = REG_SET_FIELD(wptr, IH_RB_WPTR, RB_OVERFLOW, 0);
@@ -275,10 +266,7 @@ static void vega10_ih_set_rptr(struct amdgpu_device *adev,
 {
 	if (ih->use_doorbell) {
 		/* XXX check if swapping is necessary on BE */
-		if (ih->use_bus_addr)
-			ih->ring[ih->rptr_offs] = ih->rptr;
-		else
-			adev->wb.wb[ih->rptr_offs] = ih->rptr;
+		*ih->rptr_cpu = ih->rptr;
 		WDOORBELL32(ih->doorbell_index, ih->rptr);
 	} else {
 		WREG32_SOC15(OSSSYS, 0, mmIH_RB_RPTR, ih->rptr);