drm/radeon/kms: enable writeback (v2)

When writeback is enabled, the GPU shadows writes to certain
registers into a buffer in memory.  The driver can then read
the values from the shadow rather than reading back from the
register across the bus.  Writeback can be disabled by setting
the no_wb module param to 1.

On r6xx/r7xx/evergreen, the following registers are shadowed:
- CP scratch registers
- CP read pointer
- IH write pointer
On r1xx-rr5xx, the following registers are shadowed:
- CP scratch registers
- CP read pointer

v2:
- Combine wb patches for r6xx-evergreen and r1xx-r5xx
- Writeback is disabled on AGP boards since it tends to be
unreliable on AGP using the gart.
- Check radeon_wb_init return values properly.
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drivers/gpu/drm/radeon/evergreen.c

@@ -731,7 +731,7 @@ int evergreen_cp_resume(struct radeon_device *rdev)
 
 	/* Set ring buffer size */
 	rb_bufsz = drm_order(rdev->cp.ring_size / 8);
-	tmp = RB_NO_UPDATE | (drm_order(RADEON_GPU_PAGE_SIZE/8) << 8) | rb_bufsz;
+	tmp = (drm_order(RADEON_GPU_PAGE_SIZE/8) << 8) | rb_bufsz;
 #ifdef __BIG_ENDIAN
 	tmp |= BUF_SWAP_32BIT;
 #endif
@@ -745,8 +745,19 @@ int evergreen_cp_resume(struct radeon_device *rdev)
 	WREG32(CP_RB_CNTL, tmp | RB_RPTR_WR_ENA);
 	WREG32(CP_RB_RPTR_WR, 0);
 	WREG32(CP_RB_WPTR, 0);
-	WREG32(CP_RB_RPTR_ADDR, rdev->cp.gpu_addr & 0xFFFFFFFF);
-	WREG32(CP_RB_RPTR_ADDR_HI, upper_32_bits(rdev->cp.gpu_addr));
+
+	/* set the wb address wether it's enabled or not */
+	WREG32(CP_RB_RPTR_ADDR, (rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET) & 0xFFFFFFFC);
+	WREG32(CP_RB_RPTR_ADDR_HI, upper_32_bits(rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET) & 0xFF);
+	WREG32(SCRATCH_ADDR, ((rdev->wb.gpu_addr + RADEON_WB_SCRATCH_OFFSET) >> 8) & 0xFFFFFFFF);
+
+	if (rdev->wb.enabled)
+		WREG32(SCRATCH_UMSK, 0xff);
+	else {
+		tmp |= RB_NO_UPDATE;
+		WREG32(SCRATCH_UMSK, 0);
+	}
+
 	mdelay(1);
 	WREG32(CP_RB_CNTL, tmp);
 
@@ -1759,7 +1770,9 @@ static inline u32 evergreen_get_ih_wptr(struct radeon_device *rdev)
 {
 	u32 wptr, tmp;
 
-	/* XXX use writeback */
+	if (rdev->wb.enabled)
+		wptr = rdev->wb.wb[R600_WB_IH_WPTR_OFFSET/4];
+	else
 		wptr = RREG32(IH_RB_WPTR);
 
 	if (wptr & RB_OVERFLOW) {
@@ -2068,6 +2081,11 @@ static int evergreen_startup(struct radeon_device *rdev)
 	}
 #endif
 
+	/* allocate wb buffer */
+	r = radeon_wb_init(rdev);
+	if (r)
+		return r;
+
 	/* Enable IRQ */
 	r = r600_irq_init(rdev);
 	if (r) {
@@ -2086,8 +2104,6 @@ static int evergreen_startup(struct radeon_device *rdev)
 	r = evergreen_cp_resume(rdev);
 	if (r)
 		return r;
-	/* write back buffer are not vital so don't worry about failure */
-	r600_wb_enable(rdev);
 
 	return 0;
 }
@@ -2128,7 +2144,7 @@ int evergreen_suspend(struct radeon_device *rdev)
 	r700_cp_stop(rdev);
 	rdev->cp.ready = false;
 	evergreen_irq_suspend(rdev);
-	r600_wb_disable(rdev);
+	radeon_wb_disable(rdev);
 	evergreen_pcie_gart_disable(rdev);
 #if 0
 	/* unpin shaders bo */
@@ -2245,8 +2261,8 @@ int evergreen_init(struct radeon_device *rdev)
 	if (r) {
 		dev_err(rdev->dev, "disabling GPU acceleration\n");
 		r700_cp_fini(rdev);
-		r600_wb_fini(rdev);
 		r600_irq_fini(rdev);
+		radeon_wb_fini(rdev);
 		radeon_irq_kms_fini(rdev);
 		evergreen_pcie_gart_fini(rdev);
 		rdev->accel_working = false;
@@ -2270,8 +2286,8 @@ void evergreen_fini(struct radeon_device *rdev)
 {
 	/*r600_blit_fini(rdev);*/
 	r700_cp_fini(rdev);
-	r600_wb_fini(rdev);
 	r600_irq_fini(rdev);
+	radeon_wb_fini(rdev);
 	radeon_irq_kms_fini(rdev);
 	evergreen_pcie_gart_fini(rdev);
 	radeon_gem_fini(rdev);

drivers/gpu/drm/radeon/r100.c

@@ -675,67 +675,6 @@ void r100_fence_ring_emit(struct radeon_device *rdev,
 	radeon_ring_write(rdev, RADEON_SW_INT_FIRE);
 }
 
-int r100_wb_init(struct radeon_device *rdev)
-{
-	int r;
-
-	if (rdev->wb.wb_obj == NULL) {
-		r = radeon_bo_create(rdev, NULL, RADEON_GPU_PAGE_SIZE, true,
-					RADEON_GEM_DOMAIN_GTT,
-					&rdev->wb.wb_obj);
-		if (r) {
-			dev_err(rdev->dev, "(%d) create WB buffer failed\n", r);
-			return r;
-		}
-		r = radeon_bo_reserve(rdev->wb.wb_obj, false);
-		if (unlikely(r != 0))
-			return r;
-		r = radeon_bo_pin(rdev->wb.wb_obj, RADEON_GEM_DOMAIN_GTT,
-					&rdev->wb.gpu_addr);
-		if (r) {
-			dev_err(rdev->dev, "(%d) pin WB buffer failed\n", r);
-			radeon_bo_unreserve(rdev->wb.wb_obj);
-			return r;
-		}
-		r = radeon_bo_kmap(rdev->wb.wb_obj, (void **)&rdev->wb.wb);
-		radeon_bo_unreserve(rdev->wb.wb_obj);
-		if (r) {
-			dev_err(rdev->dev, "(%d) map WB buffer failed\n", r);
-			return r;
-		}
-	}
-	WREG32(R_000774_SCRATCH_ADDR, rdev->wb.gpu_addr);
-	WREG32(R_00070C_CP_RB_RPTR_ADDR,
-		S_00070C_RB_RPTR_ADDR((rdev->wb.gpu_addr + 1024) >> 2));
-	WREG32(R_000770_SCRATCH_UMSK, 0xff);
-	return 0;
-}
-
-void r100_wb_disable(struct radeon_device *rdev)
-{
-	WREG32(R_000770_SCRATCH_UMSK, 0);
-}
-
-void r100_wb_fini(struct radeon_device *rdev)
-{
-	int r;
-
-	r100_wb_disable(rdev);
-	if (rdev->wb.wb_obj) {
-		r = radeon_bo_reserve(rdev->wb.wb_obj, false);
-		if (unlikely(r != 0)) {
-			dev_err(rdev->dev, "(%d) can't finish WB\n", r);
-			return;
-		}
-		radeon_bo_kunmap(rdev->wb.wb_obj);
-		radeon_bo_unpin(rdev->wb.wb_obj);
-		radeon_bo_unreserve(rdev->wb.wb_obj);
-		radeon_bo_unref(&rdev->wb.wb_obj);
-		rdev->wb.wb = NULL;
-		rdev->wb.wb_obj = NULL;
-	}
-}
-
 int r100_copy_blit(struct radeon_device *rdev,
 		   uint64_t src_offset,
 		   uint64_t dst_offset,
@@ -996,20 +935,32 @@ int r100_cp_init(struct radeon_device *rdev, unsigned ring_size)
 	WREG32(0x718, pre_write_timer | (pre_write_limit << 28));
 	tmp = (REG_SET(RADEON_RB_BUFSZ, rb_bufsz) |
 	       REG_SET(RADEON_RB_BLKSZ, rb_blksz) |
-	       REG_SET(RADEON_MAX_FETCH, max_fetch) |
-	       RADEON_RB_NO_UPDATE);
+	       REG_SET(RADEON_MAX_FETCH, max_fetch));
 #ifdef __BIG_ENDIAN
 	tmp |= RADEON_BUF_SWAP_32BIT;
 #endif
-	WREG32(RADEON_CP_RB_CNTL, tmp);
+	WREG32(RADEON_CP_RB_CNTL, tmp | RADEON_RB_NO_UPDATE);
 
 	/* Set ring address */
 	DRM_INFO("radeon: ring at 0x%016lX\n", (unsigned long)rdev->cp.gpu_addr);
 	WREG32(RADEON_CP_RB_BASE, rdev->cp.gpu_addr);
 	/* Force read & write ptr to 0 */
-	WREG32(RADEON_CP_RB_CNTL, tmp | RADEON_RB_RPTR_WR_ENA);
+	WREG32(RADEON_CP_RB_CNTL, tmp | RADEON_RB_RPTR_WR_ENA | RADEON_RB_NO_UPDATE);
 	WREG32(RADEON_CP_RB_RPTR_WR, 0);
 	WREG32(RADEON_CP_RB_WPTR, 0);
+
+	/* set the wb address whether it's enabled or not */
+	WREG32(R_00070C_CP_RB_RPTR_ADDR,
+		S_00070C_RB_RPTR_ADDR((rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET) >> 2));
+	WREG32(R_000774_SCRATCH_ADDR, rdev->wb.gpu_addr + RADEON_WB_SCRATCH_OFFSET);
+
+	if (rdev->wb.enabled)
+		WREG32(R_000770_SCRATCH_UMSK, 0xff);
+	else {
+		tmp |= RADEON_RB_NO_UPDATE;
+		WREG32(R_000770_SCRATCH_UMSK, 0);
+	}
+
 	WREG32(RADEON_CP_RB_CNTL, tmp);
 	udelay(10);
 	rdev->cp.rptr = RREG32(RADEON_CP_RB_RPTR);
@@ -1050,6 +1001,7 @@ void r100_cp_disable(struct radeon_device *rdev)
 	rdev->cp.ready = false;
 	WREG32(RADEON_CP_CSQ_MODE, 0);
 	WREG32(RADEON_CP_CSQ_CNTL, 0);
+	WREG32(R_000770_SCRATCH_UMSK, 0);
 	if (r100_gui_wait_for_idle(rdev)) {
 		printk(KERN_WARNING "Failed to wait GUI idle while "
 		       "programming pipes. Bad things might happen.\n");
@@ -3734,6 +3686,12 @@ static int r100_startup(struct radeon_device *rdev)
 		if (r)
 			return r;
 	}
+
+	/* allocate wb buffer */
+	r = radeon_wb_init(rdev);
+	if (r)
+		return r;
+
 	/* Enable IRQ */
 	r100_irq_set(rdev);
 	rdev->config.r100.hdp_cntl = RREG32(RADEON_HOST_PATH_CNTL);
@@ -3743,9 +3701,6 @@ static int r100_startup(struct radeon_device *rdev)
 		dev_err(rdev->dev, "failled initializing CP (%d).\n", r);
 		return r;
 	}
-	r = r100_wb_init(rdev);
-	if (r)
-		dev_err(rdev->dev, "failled initializing WB (%d).\n", r);
 	r = r100_ib_init(rdev);
 	if (r) {
 		dev_err(rdev->dev, "failled initializing IB (%d).\n", r);
@@ -3779,7 +3734,7 @@ int r100_resume(struct radeon_device *rdev)
 int r100_suspend(struct radeon_device *rdev)
 {
 	r100_cp_disable(rdev);
-	r100_wb_disable(rdev);
+	radeon_wb_disable(rdev);
 	r100_irq_disable(rdev);
 	if (rdev->flags & RADEON_IS_PCI)
 		r100_pci_gart_disable(rdev);
@@ -3789,7 +3744,7 @@ int r100_suspend(struct radeon_device *rdev)
 void r100_fini(struct radeon_device *rdev)
 {
 	r100_cp_fini(rdev);
-	r100_wb_fini(rdev);
+	radeon_wb_fini(rdev);
 	r100_ib_fini(rdev);
 	radeon_gem_fini(rdev);
 	if (rdev->flags & RADEON_IS_PCI)
@@ -3902,7 +3857,7 @@ int r100_init(struct radeon_device *rdev)
 		/* Somethings want wront with the accel init stop accel */
 		dev_err(rdev->dev, "Disabling GPU acceleration\n");
 		r100_cp_fini(rdev);
-		r100_wb_fini(rdev);
+		radeon_wb_fini(rdev);
 		r100_ib_fini(rdev);
 		radeon_irq_kms_fini(rdev);
 		if (rdev->flags & RADEON_IS_PCI)

drivers/gpu/drm/radeon/r300.c

@@ -1332,6 +1332,12 @@ static int r300_startup(struct radeon_device *rdev)
 		if (r)
 			return r;
 	}
+
+	/* allocate wb buffer */
+	r = radeon_wb_init(rdev);
+	if (r)
+		return r;
+
 	/* Enable IRQ */
 	r100_irq_set(rdev);
 	rdev->config.r300.hdp_cntl = RREG32(RADEON_HOST_PATH_CNTL);
@@ -1341,9 +1347,6 @@ static int r300_startup(struct radeon_device *rdev)
 		dev_err(rdev->dev, "failled initializing CP (%d).\n", r);
 		return r;
 	}
-	r = r100_wb_init(rdev);
-	if (r)
-		dev_err(rdev->dev, "failled initializing WB (%d).\n", r);
 	r = r100_ib_init(rdev);
 	if (r) {
 		dev_err(rdev->dev, "failled initializing IB (%d).\n", r);
@@ -1379,7 +1382,7 @@ int r300_resume(struct radeon_device *rdev)
 int r300_suspend(struct radeon_device *rdev)
 {
 	r100_cp_disable(rdev);
-	r100_wb_disable(rdev);
+	radeon_wb_disable(rdev);
 	r100_irq_disable(rdev);
 	if (rdev->flags & RADEON_IS_PCIE)
 		rv370_pcie_gart_disable(rdev);
@@ -1391,7 +1394,7 @@ int r300_suspend(struct radeon_device *rdev)
 void r300_fini(struct radeon_device *rdev)
 {
 	r100_cp_fini(rdev);
-	r100_wb_fini(rdev);
+	radeon_wb_fini(rdev);
 	r100_ib_fini(rdev);
 	radeon_gem_fini(rdev);
 	if (rdev->flags & RADEON_IS_PCIE)
@@ -1484,7 +1487,7 @@ int r300_init(struct radeon_device *rdev)
 		/* Somethings want wront with the accel init stop accel */
 		dev_err(rdev->dev, "Disabling GPU acceleration\n");
 		r100_cp_fini(rdev);
-		r100_wb_fini(rdev);
+		radeon_wb_fini(rdev);
 		r100_ib_fini(rdev);
 		radeon_irq_kms_fini(rdev);
 		if (rdev->flags & RADEON_IS_PCIE)

drivers/gpu/drm/radeon/r420.c

@@ -248,6 +248,12 @@ static int r420_startup(struct radeon_device *rdev)
 			return r;
 	}
 	r420_pipes_init(rdev);
+
+	/* allocate wb buffer */
+	r = radeon_wb_init(rdev);
+	if (r)
+		return r;
+
 	/* Enable IRQ */
 	r100_irq_set(rdev);
 	rdev->config.r300.hdp_cntl = RREG32(RADEON_HOST_PATH_CNTL);
@@ -258,10 +264,6 @@ static int r420_startup(struct radeon_device *rdev)
 		return r;
 	}
 	r420_cp_errata_init(rdev);
-	r = r100_wb_init(rdev);
-	if (r) {
-		dev_err(rdev->dev, "failled initializing WB (%d).\n", r);
-	}
 	r = r100_ib_init(rdev);
 	if (r) {
 		dev_err(rdev->dev, "failled initializing IB (%d).\n", r);
@@ -302,7 +304,7 @@ int r420_suspend(struct radeon_device *rdev)
 {
 	r420_cp_errata_fini(rdev);
 	r100_cp_disable(rdev);
-	r100_wb_disable(rdev);
+	radeon_wb_disable(rdev);
 	r100_irq_disable(rdev);
 	if (rdev->flags & RADEON_IS_PCIE)
 		rv370_pcie_gart_disable(rdev);
@@ -314,7 +316,7 @@ int r420_suspend(struct radeon_device *rdev)
 void r420_fini(struct radeon_device *rdev)
 {
 	r100_cp_fini(rdev);
-	r100_wb_fini(rdev);
+	radeon_wb_fini(rdev);
 	r100_ib_fini(rdev);
 	radeon_gem_fini(rdev);
 	if (rdev->flags & RADEON_IS_PCIE)
@@ -418,7 +420,7 @@ int r420_init(struct radeon_device *rdev)
 		/* Somethings want wront with the accel init stop accel */
 		dev_err(rdev->dev, "Disabling GPU acceleration\n");
 		r100_cp_fini(rdev);
-		r100_wb_fini(rdev);
+		radeon_wb_fini(rdev);
 		r100_ib_fini(rdev);
 		radeon_irq_kms_fini(rdev);
 		if (rdev->flags & RADEON_IS_PCIE)

drivers/gpu/drm/radeon/r520.c

@@ -181,6 +181,12 @@ static int r520_startup(struct radeon_device *rdev)
 		if (r)
 			return r;
 	}
+
+	/* allocate wb buffer */
+	r = radeon_wb_init(rdev);
+	if (r)
+		return r;
+
 	/* Enable IRQ */
 	rs600_irq_set(rdev);
 	rdev->config.r300.hdp_cntl = RREG32(RADEON_HOST_PATH_CNTL);
@@ -190,9 +196,6 @@ static int r520_startup(struct radeon_device *rdev)
 		dev_err(rdev->dev, "failled initializing CP (%d).\n", r);
 		return r;
 	}
-	r = r100_wb_init(rdev);
-	if (r)
-		dev_err(rdev->dev, "failled initializing WB (%d).\n", r);
 	r = r100_ib_init(rdev);
 	if (r) {
 		dev_err(rdev->dev, "failled initializing IB (%d).\n", r);
@@ -295,7 +298,7 @@ int r520_init(struct radeon_device *rdev)
 		/* Somethings want wront with the accel init stop accel */
 		dev_err(rdev->dev, "Disabling GPU acceleration\n");
 		r100_cp_fini(rdev);
-		r100_wb_fini(rdev);
+		radeon_wb_fini(rdev);
 		r100_ib_fini(rdev);
 		radeon_irq_kms_fini(rdev);
 		rv370_pcie_gart_fini(rdev);

drivers/gpu/drm/radeon/r600.c

@@ -1918,6 +1918,7 @@ void r600_pciep_wreg(struct radeon_device *rdev, u32 reg, u32 v)
 void r600_cp_stop(struct radeon_device *rdev)
 {
 	WREG32(R_0086D8_CP_ME_CNTL, S_0086D8_CP_ME_HALT(1));
+	WREG32(SCRATCH_UMSK, 0);
 }
 
 int r600_init_microcode(struct radeon_device *rdev)
@@ -2150,7 +2151,7 @@ int r600_cp_resume(struct radeon_device *rdev)
 
 	/* Set ring buffer size */
 	rb_bufsz = drm_order(rdev->cp.ring_size / 8);
-	tmp = RB_NO_UPDATE | (drm_order(RADEON_GPU_PAGE_SIZE/8) << 8) | rb_bufsz;
+	tmp = (drm_order(RADEON_GPU_PAGE_SIZE/8) << 8) | rb_bufsz;
 #ifdef __BIG_ENDIAN
 	tmp |= BUF_SWAP_32BIT;
 #endif
@@ -2164,8 +2165,19 @@ int r600_cp_resume(struct radeon_device *rdev)
 	WREG32(CP_RB_CNTL, tmp | RB_RPTR_WR_ENA);
 	WREG32(CP_RB_RPTR_WR, 0);
 	WREG32(CP_RB_WPTR, 0);
-	WREG32(CP_RB_RPTR_ADDR, rdev->cp.gpu_addr & 0xFFFFFFFF);
-	WREG32(CP_RB_RPTR_ADDR_HI, upper_32_bits(rdev->cp.gpu_addr));
+
+	/* set the wb address whether it's enabled or not */
+	WREG32(CP_RB_RPTR_ADDR, (rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET) & 0xFFFFFFFC);
+	WREG32(CP_RB_RPTR_ADDR_HI, upper_32_bits(rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET) & 0xFF);
+	WREG32(SCRATCH_ADDR, ((rdev->wb.gpu_addr + RADEON_WB_SCRATCH_OFFSET) >> 8) & 0xFFFFFFFF);
+
+	if (rdev->wb.enabled)
+		WREG32(SCRATCH_UMSK, 0xff);
+	else {
+		tmp |= RB_NO_UPDATE;
+		WREG32(SCRATCH_UMSK, 0);
+	}
+
 	mdelay(1);
 	WREG32(CP_RB_CNTL, tmp);
 
@@ -2217,9 +2229,10 @@ void r600_scratch_init(struct radeon_device *rdev)
 	int i;
 
 	rdev->scratch.num_reg = 7;
+	rdev->scratch.reg_base = SCRATCH_REG0;
 	for (i = 0; i < rdev->scratch.num_reg; i++) {
 		rdev->scratch.free[i] = true;
-		rdev->scratch.reg[i] = SCRATCH_REG0 + (i * 4);
+		rdev->scratch.reg[i] = rdev->scratch.reg_base + (i * 4);
 	}
 }
 
@@ -2263,70 +2276,6 @@ int r600_ring_test(struct radeon_device *rdev)
 	return r;
 }
 
-void r600_wb_disable(struct radeon_device *rdev)
-{
-	int r;
-
-	WREG32(SCRATCH_UMSK, 0);
-	if (rdev->wb.wb_obj) {
-		r = radeon_bo_reserve(rdev->wb.wb_obj, false);
-		if (unlikely(r != 0))
-			return;
-		radeon_bo_kunmap(rdev->wb.wb_obj);
-		radeon_bo_unpin(rdev->wb.wb_obj);
-		radeon_bo_unreserve(rdev->wb.wb_obj);
-	}
-}
-
-void r600_wb_fini(struct radeon_device *rdev)
-{
-	r600_wb_disable(rdev);
-	if (rdev->wb.wb_obj) {
-		radeon_bo_unref(&rdev->wb.wb_obj);
-		rdev->wb.wb = NULL;
-		rdev->wb.wb_obj = NULL;
-	}
-}
-
-int r600_wb_enable(struct radeon_device *rdev)
-{
-	int r;
-
-	if (rdev->wb.wb_obj == NULL) {
-		r = radeon_bo_create(rdev, NULL, RADEON_GPU_PAGE_SIZE, true,
-				RADEON_GEM_DOMAIN_GTT, &rdev->wb.wb_obj);
-		if (r) {
-			dev_warn(rdev->dev, "(%d) create WB bo failed\n", r);
-			return r;
-		}
-		r = radeon_bo_reserve(rdev->wb.wb_obj, false);
-		if (unlikely(r != 0)) {
-			r600_wb_fini(rdev);
-			return r;
-		}
-		r = radeon_bo_pin(rdev->wb.wb_obj, RADEON_GEM_DOMAIN_GTT,
-				&rdev->wb.gpu_addr);
-		if (r) {
-			radeon_bo_unreserve(rdev->wb.wb_obj);
-			dev_warn(rdev->dev, "(%d) pin WB bo failed\n", r);
-			r600_wb_fini(rdev);
-			return r;
-		}
-		r = radeon_bo_kmap(rdev->wb.wb_obj, (void **)&rdev->wb.wb);
-		radeon_bo_unreserve(rdev->wb.wb_obj);
-		if (r) {
-			dev_warn(rdev->dev, "(%d) map WB bo failed\n", r);
-			r600_wb_fini(rdev);
-			return r;
-		}
-	}
-	WREG32(SCRATCH_ADDR, (rdev->wb.gpu_addr >> 8) & 0xFFFFFFFF);
-	WREG32(CP_RB_RPTR_ADDR, (rdev->wb.gpu_addr + 1024) & 0xFFFFFFFC);
-	WREG32(CP_RB_RPTR_ADDR_HI, upper_32_bits(rdev->wb.gpu_addr + 1024) & 0xFF);
-	WREG32(SCRATCH_UMSK, 0xff);
-	return 0;
-}
-
 void r600_fence_ring_emit(struct radeon_device *rdev,
 			  struct radeon_fence *fence)
 {
@@ -2427,6 +2376,11 @@ int r600_startup(struct radeon_device *rdev)
 		dev_warn(rdev->dev, "failed blitter (%d) falling back to memcpy\n", r);
 	}
 
+	/* allocate wb buffer */
+	r = radeon_wb_init(rdev);
+	if (r)
+		return r;
+
 	/* Enable IRQ */
 	r = r600_irq_init(rdev);
 	if (r) {
@@ -2445,8 +2399,7 @@ int r600_startup(struct radeon_device *rdev)
 	r = r600_cp_resume(rdev);
 	if (r)
 		return r;
-	/* write back buffer are not vital so don't worry about failure */
-	r600_wb_enable(rdev);
+
 	return 0;
 }
 
@@ -2505,7 +2458,7 @@ int r600_suspend(struct radeon_device *rdev)
 	r600_cp_stop(rdev);
 	rdev->cp.ready = false;
 	r600_irq_suspend(rdev);
-	r600_wb_disable(rdev);
+	radeon_wb_disable(rdev);
 	r600_pcie_gart_disable(rdev);
 	/* unpin shaders bo */
 	if (rdev->r600_blit.shader_obj) {
@@ -2602,8 +2555,8 @@ int r600_init(struct radeon_device *rdev)
 	if (r) {
 		dev_err(rdev->dev, "disabling GPU acceleration\n");
 		r600_cp_fini(rdev);
-		r600_wb_fini(rdev);
 		r600_irq_fini(rdev);
+		radeon_wb_fini(rdev);
 		radeon_irq_kms_fini(rdev);
 		r600_pcie_gart_fini(rdev);
 		rdev->accel_working = false;
@@ -2633,8 +2586,8 @@ void r600_fini(struct radeon_device *rdev)
 	r600_audio_fini(rdev);
 	r600_blit_fini(rdev);
 	r600_cp_fini(rdev);
-	r600_wb_fini(rdev);
 	r600_irq_fini(rdev);
+	radeon_wb_fini(rdev);
 	radeon_irq_kms_fini(rdev);
 	r600_pcie_gart_fini(rdev);
 	radeon_agp_fini(rdev);
@@ -2969,10 +2922,13 @@ int r600_irq_init(struct radeon_device *rdev)
 	ih_rb_cntl = (IH_WPTR_OVERFLOW_ENABLE |
 		      IH_WPTR_OVERFLOW_CLEAR |
 		      (rb_bufsz << 1));
-	/* WPTR writeback, not yet */
-	/*ih_rb_cntl |= IH_WPTR_WRITEBACK_ENABLE;*/
-	WREG32(IH_RB_WPTR_ADDR_LO, 0);
-	WREG32(IH_RB_WPTR_ADDR_HI, 0);
+
+	if (rdev->wb.enabled)
+		ih_rb_cntl |= IH_WPTR_WRITEBACK_ENABLE;
+
+	/* set the writeback address whether it's enabled or not */
+	WREG32(IH_RB_WPTR_ADDR_LO, (rdev->wb.gpu_addr + R600_WB_IH_WPTR_OFFSET) & 0xFFFFFFFC);
+	WREG32(IH_RB_WPTR_ADDR_HI, upper_32_bits(rdev->wb.gpu_addr + R600_WB_IH_WPTR_OFFSET) & 0xFF);
 
 	WREG32(IH_RB_CNTL, ih_rb_cntl);
 
@@ -3230,7 +3186,9 @@ static inline u32 r600_get_ih_wptr(struct radeon_device *rdev)
 {
 	u32 wptr, tmp;
 
-	/* XXX use writeback */
+	if (rdev->wb.enabled)
+		wptr = rdev->wb.wb[R600_WB_IH_WPTR_OFFSET/4];
+	else
 		wptr = RREG32(IH_RB_WPTR);
 
 	if (wptr & RB_OVERFLOW) {

drivers/gpu/drm/radeon/radeon.h

@@ -365,6 +365,7 @@ bool radeon_atombios_sideport_present(struct radeon_device *rdev);
  */
 struct radeon_scratch {
 	unsigned		num_reg;
+	uint32_t                reg_base;
 	bool			free[32];
 	uint32_t		reg[32];
 };
@@ -593,8 +594,13 @@ struct radeon_wb {
 	struct radeon_bo	*wb_obj;
 	volatile uint32_t	*wb;
 	uint64_t		gpu_addr;
+	bool                    enabled;
 };
 
+#define RADEON_WB_SCRATCH_OFFSET 0
+#define RADEON_WB_CP_RPTR_OFFSET 1024
+#define R600_WB_IH_WPTR_OFFSET   2048
+
 /**
  * struct radeon_pm - power management datas
  * @max_bandwidth:      maximum bandwidth the gpu has (MByte/s)
@@ -1340,6 +1346,9 @@ extern void radeon_update_bandwidth_info(struct radeon_device *rdev);
 extern void radeon_update_display_priority(struct radeon_device *rdev);
 extern bool radeon_boot_test_post_card(struct radeon_device *rdev);
 extern void radeon_scratch_init(struct radeon_device *rdev);
+extern void radeon_wb_fini(struct radeon_device *rdev);
+extern int radeon_wb_init(struct radeon_device *rdev);
+extern void radeon_wb_disable(struct radeon_device *rdev);
 extern void radeon_surface_init(struct radeon_device *rdev);
 extern int radeon_cs_parser_init(struct radeon_cs_parser *p, void *data);
 extern void radeon_legacy_set_clock_gating(struct radeon_device *rdev, int enable);
@@ -1424,9 +1433,6 @@ extern int r600_pcie_gart_init(struct radeon_device *rdev);
 extern void r600_pcie_gart_tlb_flush(struct radeon_device *rdev);
 extern int r600_ib_test(struct radeon_device *rdev);
 extern int r600_ring_test(struct radeon_device *rdev);
-extern void r600_wb_fini(struct radeon_device *rdev);
-extern int r600_wb_enable(struct radeon_device *rdev);
-extern void r600_wb_disable(struct radeon_device *rdev);
 extern void r600_scratch_init(struct radeon_device *rdev);
 extern int r600_blit_init(struct radeon_device *rdev);
 extern void r600_blit_fini(struct radeon_device *rdev);

drivers/gpu/drm/radeon/radeon_asic.h

@@ -108,9 +108,6 @@ void r100_irq_disable(struct radeon_device *rdev);
 void r100_mc_stop(struct radeon_device *rdev, struct r100_mc_save *save);
 void r100_mc_resume(struct radeon_device *rdev, struct r100_mc_save *save);
 void r100_vram_init_sizes(struct radeon_device *rdev);
-void r100_wb_disable(struct radeon_device *rdev);
-void r100_wb_fini(struct radeon_device *rdev);
-int r100_wb_init(struct radeon_device *rdev);
 int r100_cp_reset(struct radeon_device *rdev);
 void r100_vga_render_disable(struct radeon_device *rdev);
 void r100_restore_sanity(struct radeon_device *rdev);

drivers/gpu/drm/radeon/radeon_device.c

@@ -117,9 +117,10 @@ void radeon_scratch_init(struct radeon_device *rdev)
 	} else {
 		rdev->scratch.num_reg = 7;
 	}
+	rdev->scratch.reg_base = RADEON_SCRATCH_REG0;
 	for (i = 0; i < rdev->scratch.num_reg; i++) {
 		rdev->scratch.free[i] = true;
-		rdev->scratch.reg[i] = RADEON_SCRATCH_REG0 + (i * 4);
+		rdev->scratch.reg[i] = rdev->scratch.reg_base + (i * 4);
 	}
 }
 
@@ -149,6 +150,80 @@ void radeon_scratch_free(struct radeon_device *rdev, uint32_t reg)
 	}
 }
 
+void radeon_wb_disable(struct radeon_device *rdev)
+{
+	int r;
+
+	if (rdev->wb.wb_obj) {
+		r = radeon_bo_reserve(rdev->wb.wb_obj, false);
+		if (unlikely(r != 0))
+			return;
+		radeon_bo_kunmap(rdev->wb.wb_obj);
+		radeon_bo_unpin(rdev->wb.wb_obj);
+		radeon_bo_unreserve(rdev->wb.wb_obj);
+	}
+	rdev->wb.enabled = false;
+}
+
+void radeon_wb_fini(struct radeon_device *rdev)
+{
+	radeon_wb_disable(rdev);
+	if (rdev->wb.wb_obj) {
+		radeon_bo_unref(&rdev->wb.wb_obj);
+		rdev->wb.wb = NULL;
+		rdev->wb.wb_obj = NULL;
+	}
+}
+
+int radeon_wb_init(struct radeon_device *rdev)
+{
+	int r;
+
+	if (rdev->wb.wb_obj == NULL) {
+		r = radeon_bo_create(rdev, NULL, RADEON_GPU_PAGE_SIZE, true,
+				RADEON_GEM_DOMAIN_GTT, &rdev->wb.wb_obj);
+		if (r) {
+			dev_warn(rdev->dev, "(%d) create WB bo failed\n", r);
+			return r;
+		}
+	}
+	r = radeon_bo_reserve(rdev->wb.wb_obj, false);
+	if (unlikely(r != 0)) {
+		radeon_wb_fini(rdev);
+		return r;
+	}
+	r = radeon_bo_pin(rdev->wb.wb_obj, RADEON_GEM_DOMAIN_GTT,
+			  &rdev->wb.gpu_addr);
+	if (r) {
+		radeon_bo_unreserve(rdev->wb.wb_obj);
+		dev_warn(rdev->dev, "(%d) pin WB bo failed\n", r);
+		radeon_wb_fini(rdev);
+		return r;
+	}
+	r = radeon_bo_kmap(rdev->wb.wb_obj, (void **)&rdev->wb.wb);
+	radeon_bo_unreserve(rdev->wb.wb_obj);
+	if (r) {
+		dev_warn(rdev->dev, "(%d) map WB bo failed\n", r);
+		radeon_wb_fini(rdev);
+		return r;
+	}
+
+	/* disabled via module param */
+	if (radeon_no_wb == 1)
+		rdev->wb.enabled = false;
+	else {
+		/* often unreliable on AGP */
+		if (rdev->flags & RADEON_IS_AGP) {
+			rdev->wb.enabled = false;
+		} else
+			rdev->wb.enabled = true;
+	}
+
+	dev_info(rdev->dev, "WB %sabled\n", rdev->wb.enabled ? "en" : "dis");
+
+	return 0;
+}
+
 /**
  * radeon_vram_location - try to find VRAM location
  * @rdev: radeon device structure holding all necessary informations

drivers/gpu/drm/radeon/radeon_fence.c

@@ -72,6 +72,10 @@ static bool radeon_fence_poll_locked(struct radeon_device *rdev)
 	bool wake = false;
 	unsigned long cjiffies;
 
+	if (rdev->wb.enabled) {
+		u32 scratch_index = RADEON_WB_SCRATCH_OFFSET + rdev->fence_drv.scratch_reg - rdev->scratch.reg_base;
+		seq = rdev->wb.wb[scratch_index/4];
+	} else
 		seq = RREG32(rdev->fence_drv.scratch_reg);
 	if (seq != rdev->fence_drv.last_seq) {
 		rdev->fence_drv.last_seq = seq;

drivers/gpu/drm/radeon/radeon_ring.c

@@ -247,10 +247,14 @@ void radeon_ib_pool_fini(struct radeon_device *rdev)
  */
 void radeon_ring_free_size(struct radeon_device *rdev)
 {
+	if (rdev->wb.enabled)
+		rdev->cp.rptr = rdev->wb.wb[RADEON_WB_CP_RPTR_OFFSET/4];
+	else {
 		if (rdev->family >= CHIP_R600)
 			rdev->cp.rptr = RREG32(R600_CP_RB_RPTR);
 		else
 			rdev->cp.rptr = RREG32(RADEON_CP_RB_RPTR);
+	}
 	/* This works because ring_size is a power of 2 */
 	rdev->cp.ring_free_dw = (rdev->cp.rptr + (rdev->cp.ring_size / 4));
 	rdev->cp.ring_free_dw -= rdev->cp.wptr;

drivers/gpu/drm/radeon/rs400.c

@@ -397,6 +397,12 @@ static int rs400_startup(struct radeon_device *rdev)
 	r = rs400_gart_enable(rdev);
 	if (r)
 		return r;
+
+	/* allocate wb buffer */
+	r = radeon_wb_init(rdev);
+	if (r)
+		return r;
+
 	/* Enable IRQ */
 	r100_irq_set(rdev);
 	rdev->config.r300.hdp_cntl = RREG32(RADEON_HOST_PATH_CNTL);
@@ -406,9 +412,6 @@ static int rs400_startup(struct radeon_device *rdev)
 		dev_err(rdev->dev, "failled initializing CP (%d).\n", r);
 		return r;
 	}
-	r = r100_wb_init(rdev);
-	if (r)
-		dev_err(rdev->dev, "failled initializing WB (%d).\n", r);
 	r = r100_ib_init(rdev);
 	if (r) {
 		dev_err(rdev->dev, "failled initializing IB (%d).\n", r);
@@ -443,7 +446,7 @@ int rs400_resume(struct radeon_device *rdev)
 int rs400_suspend(struct radeon_device *rdev)
 {
 	r100_cp_disable(rdev);
-	r100_wb_disable(rdev);
+	radeon_wb_disable(rdev);
 	r100_irq_disable(rdev);
 	rs400_gart_disable(rdev);
 	return 0;
@@ -452,7 +455,7 @@ int rs400_suspend(struct radeon_device *rdev)
 void rs400_fini(struct radeon_device *rdev)
 {
 	r100_cp_fini(rdev);
-	r100_wb_fini(rdev);
+	radeon_wb_fini(rdev);
 	r100_ib_fini(rdev);
 	radeon_gem_fini(rdev);
 	rs400_gart_fini(rdev);
@@ -526,7 +529,7 @@ int rs400_init(struct radeon_device *rdev)
 		/* Somethings want wront with the accel init stop accel */
 		dev_err(rdev->dev, "Disabling GPU acceleration\n");
 		r100_cp_fini(rdev);
-		r100_wb_fini(rdev);
+		radeon_wb_fini(rdev);
 		r100_ib_fini(rdev);
 		rs400_gart_fini(rdev);
 		radeon_irq_kms_fini(rdev);

drivers/gpu/drm/radeon/rs600.c

@@ -795,6 +795,12 @@ static int rs600_startup(struct radeon_device *rdev)
 	r = rs600_gart_enable(rdev);
 	if (r)
 		return r;
+
+	/* allocate wb buffer */
+	r = radeon_wb_init(rdev);
+	if (r)
+		return r;
+
 	/* Enable IRQ */
 	rs600_irq_set(rdev);
 	rdev->config.r300.hdp_cntl = RREG32(RADEON_HOST_PATH_CNTL);
@@ -804,9 +810,6 @@ static int rs600_startup(struct radeon_device *rdev)
 		dev_err(rdev->dev, "failled initializing CP (%d).\n", r);
 		return r;
 	}
-	r = r100_wb_init(rdev);
-	if (r)
-		dev_err(rdev->dev, "failled initializing WB (%d).\n", r);
 	r = r100_ib_init(rdev);
 	if (r) {
 		dev_err(rdev->dev, "failled initializing IB (%d).\n", r);
@@ -847,7 +850,7 @@ int rs600_suspend(struct radeon_device *rdev)
 {
 	r600_audio_fini(rdev);
 	r100_cp_disable(rdev);
-	r100_wb_disable(rdev);
+	radeon_wb_disable(rdev);
 	rs600_irq_disable(rdev);
 	rs600_gart_disable(rdev);
 	return 0;
@@ -857,7 +860,7 @@ void rs600_fini(struct radeon_device *rdev)
 {
 	r600_audio_fini(rdev);
 	r100_cp_fini(rdev);
-	r100_wb_fini(rdev);
+	radeon_wb_fini(rdev);
 	r100_ib_fini(rdev);
 	radeon_gem_fini(rdev);
 	rs600_gart_fini(rdev);
@@ -931,7 +934,7 @@ int rs600_init(struct radeon_device *rdev)
 		/* Somethings want wront with the accel init stop accel */
 		dev_err(rdev->dev, "Disabling GPU acceleration\n");
 		r100_cp_fini(rdev);
-		r100_wb_fini(rdev);
+		radeon_wb_fini(rdev);
 		r100_ib_fini(rdev);
 		rs600_gart_fini(rdev);
 		radeon_irq_kms_fini(rdev);

drivers/gpu/drm/radeon/rs690.c

@@ -615,6 +615,12 @@ static int rs690_startup(struct radeon_device *rdev)
 	r = rs400_gart_enable(rdev);
 	if (r)
 		return r;
+
+	/* allocate wb buffer */
+	r = radeon_wb_init(rdev);
+	if (r)
+		return r;
+
 	/* Enable IRQ */
 	rs600_irq_set(rdev);
 	rdev->config.r300.hdp_cntl = RREG32(RADEON_HOST_PATH_CNTL);
@@ -624,9 +630,6 @@ static int rs690_startup(struct radeon_device *rdev)
 		dev_err(rdev->dev, "failled initializing CP (%d).\n", r);
 		return r;
 	}
-	r = r100_wb_init(rdev);
-	if (r)
-		dev_err(rdev->dev, "failled initializing WB (%d).\n", r);
 	r = r100_ib_init(rdev);
 	if (r) {
 		dev_err(rdev->dev, "failled initializing IB (%d).\n", r);
@@ -667,7 +670,7 @@ int rs690_suspend(struct radeon_device *rdev)
 {
 	r600_audio_fini(rdev);
 	r100_cp_disable(rdev);
-	r100_wb_disable(rdev);
+	radeon_wb_disable(rdev);
 	rs600_irq_disable(rdev);
 	rs400_gart_disable(rdev);
 	return 0;
@@ -677,7 +680,7 @@ void rs690_fini(struct radeon_device *rdev)
 {
 	r600_audio_fini(rdev);
 	r100_cp_fini(rdev);
-	r100_wb_fini(rdev);
+	radeon_wb_fini(rdev);
 	r100_ib_fini(rdev);
 	radeon_gem_fini(rdev);
 	rs400_gart_fini(rdev);
@@ -752,7 +755,7 @@ int rs690_init(struct radeon_device *rdev)
 		/* Somethings want wront with the accel init stop accel */
 		dev_err(rdev->dev, "Disabling GPU acceleration\n");
 		r100_cp_fini(rdev);
-		r100_wb_fini(rdev);
+		radeon_wb_fini(rdev);
 		r100_ib_fini(rdev);
 		rs400_gart_fini(rdev);
 		radeon_irq_kms_fini(rdev);

drivers/gpu/drm/radeon/rv515.c

@@ -386,6 +386,12 @@ static int rv515_startup(struct radeon_device *rdev)
 		if (r)
 			return r;
 	}
+
+	/* allocate wb buffer */
+	r = radeon_wb_init(rdev);
+	if (r)
+		return r;
+
 	/* Enable IRQ */
 	rs600_irq_set(rdev);
 	rdev->config.r300.hdp_cntl = RREG32(RADEON_HOST_PATH_CNTL);
@@ -395,9 +401,6 @@ static int rv515_startup(struct radeon_device *rdev)
 		dev_err(rdev->dev, "failled initializing CP (%d).\n", r);
 		return r;
 	}
-	r = r100_wb_init(rdev);
-	if (r)
-		dev_err(rdev->dev, "failled initializing WB (%d).\n", r);
 	r = r100_ib_init(rdev);
 	if (r) {
 		dev_err(rdev->dev, "failled initializing IB (%d).\n", r);
@@ -431,7 +434,7 @@ int rv515_resume(struct radeon_device *rdev)
 int rv515_suspend(struct radeon_device *rdev)
 {
 	r100_cp_disable(rdev);
-	r100_wb_disable(rdev);
+	radeon_wb_disable(rdev);
 	rs600_irq_disable(rdev);
 	if (rdev->flags & RADEON_IS_PCIE)
 		rv370_pcie_gart_disable(rdev);
@@ -447,7 +450,7 @@ void rv515_set_safe_registers(struct radeon_device *rdev)
 void rv515_fini(struct radeon_device *rdev)
 {
 	r100_cp_fini(rdev);
-	r100_wb_fini(rdev);
+	radeon_wb_fini(rdev);
 	r100_ib_fini(rdev);
 	radeon_gem_fini(rdev);
 	rv370_pcie_gart_fini(rdev);
@@ -527,7 +530,7 @@ int rv515_init(struct radeon_device *rdev)
 		/* Somethings want wront with the accel init stop accel */
 		dev_err(rdev->dev, "Disabling GPU acceleration\n");
 		r100_cp_fini(rdev);
-		r100_wb_fini(rdev);
+		radeon_wb_fini(rdev);
 		r100_ib_fini(rdev);
 		radeon_irq_kms_fini(rdev);
 		rv370_pcie_gart_fini(rdev);

drivers/gpu/drm/radeon/rv770.c

@@ -268,6 +268,7 @@ static void rv770_mc_program(struct radeon_device *rdev)
 void r700_cp_stop(struct radeon_device *rdev)
 {
 	WREG32(CP_ME_CNTL, (CP_ME_HALT | CP_PFP_HALT));
+	WREG32(SCRATCH_UMSK, 0);
 }
 
 static int rv770_cp_load_microcode(struct radeon_device *rdev)
@@ -1029,6 +1030,11 @@ static int rv770_startup(struct radeon_device *rdev)
 		dev_warn(rdev->dev, "failed blitter (%d) falling back to memcpy\n", r);
 	}
 
+	/* allocate wb buffer */
+	r = radeon_wb_init(rdev);
+	if (r)
+		return r;
+
 	/* Enable IRQ */
 	r = r600_irq_init(rdev);
 	if (r) {
@@ -1047,8 +1053,7 @@ static int rv770_startup(struct radeon_device *rdev)
 	r = r600_cp_resume(rdev);
 	if (r)
 		return r;
-	/* write back buffer are not vital so don't worry about failure */
-	r600_wb_enable(rdev);
+
 	return 0;
 }
 
@@ -1094,7 +1099,7 @@ int rv770_suspend(struct radeon_device *rdev)
 	r700_cp_stop(rdev);
 	rdev->cp.ready = false;
 	r600_irq_suspend(rdev);
-	r600_wb_disable(rdev);
+	radeon_wb_disable(rdev);
 	rv770_pcie_gart_disable(rdev);
 	/* unpin shaders bo */
 	if (rdev->r600_blit.shader_obj) {
@@ -1189,8 +1194,8 @@ int rv770_init(struct radeon_device *rdev)
 	if (r) {
 		dev_err(rdev->dev, "disabling GPU acceleration\n");
 		r700_cp_fini(rdev);
-		r600_wb_fini(rdev);
 		r600_irq_fini(rdev);
+		radeon_wb_fini(rdev);
 		radeon_irq_kms_fini(rdev);
 		rv770_pcie_gart_fini(rdev);
 		rdev->accel_working = false;
@@ -1222,8 +1227,8 @@ void rv770_fini(struct radeon_device *rdev)
 {
 	r600_blit_fini(rdev);
 	r700_cp_fini(rdev);
-	r600_wb_fini(rdev);
 	r600_irq_fini(rdev);
+	radeon_wb_fini(rdev);
 	radeon_irq_kms_fini(rdev);
 	rv770_pcie_gart_fini(rdev);
 	rv770_vram_scratch_fini(rdev);