[PATCH RFC v3 0/7] Add support for Adreno 810 GPU

[PATCH RFC v3 0/7] Add support for Adreno 810 GPU

[Alexander Koskovich]

Adreno 810 is present in the Milos SoC and is the first GPU to be released in
the A8x family.

Note that the OPP table is limited to 1050MHz to start with as the only Milos
device I have is limited to that speed in GPU_CC_FREQ_LIMIT_VAL.

This series is marked as RFC because it depends on a couple other in review
series, GPU GX GDSC handling [1] and the GXCLKCTL block for Milos [2].

Also depends on A8x batch 2 but it looks like that made it into linux-next.

[1]: https://lore.kernel.org/linux-arm-msm/20260407-gfx-clk-fixes-v1-0-4bb5583a5054@oss.qualcomm.com
[2]: https://lore.kernel.org/linux-arm-msm/20260403-milos-gxclkctl-v2-0-95eb94a7d0a4@fairphone.com

Signed-off-by: Alexander Koskovich <akoskovich@pm.me>
---
Changes in v3:
- Drop DEMET from GMU clocks (not required on A810)
- Document qcom,adreno-44010000 compatible (regex is gone in 7.0+)
- Drop zeroed out CP_PROTECT_REG[46, 62] range, not required
- Add a810_protect to __build_asserts
- Add UCHE_CCHE_TRAP_BASE_[LO|HI] and UCHE_CCHE_WRITE_THRU_BASE_[LO|HI] to a810_pwrup_reglist_regs
- Move TPL1 registers to a810_pwrup_reglist_regs
- Include all protect registers in a810_ifpc_reglist_regs
- Revert pipe reg comment, just copied it from downstream but original also works
- Link to v2: https://lore.kernel.org/r/20260402-adreno-810-v2-0-ce337ca87a9e@pm.me

Changes in v2:
- Mark as RFC due to dependency on in-review changes
- Explain in DTS commit why qcom,kaanapali-gxclkctl.h and not qcom,milos-gxclkctl.h
- cx_mmio -> cx_misc_mmio
- Sync a810_nonctxt_regs with GRAPHICS.LA.14.0.r5-03100-lanai.0
- Link to v1: https://lore.kernel.org/r/20260331-adreno-810-v1-0-725801dbb12b@pm.me

---
Alexander Koskovich (7):
      dt-bindings: display/msm/gmu: Document Adreno 810 GMU
      dt-bindings: display/msm/gpu: Document A810 GPU
      drm/msm/adreno: rename llc_mmio to cx_misc_mmio
      drm/msm/adreno: set cx_misc_mmio regardless of if platform has LLCC
      drm/msm/a8xx: use pipe protect slot 15 for last-span-unbound feature
      drm/msm/adreno: add Adreno 810 GPU support
      arm64: dts: qcom: milos: Add Adreno 810 GPU and GMU nodes

 .../devicetree/bindings/display/msm/gmu.yaml       |  30 +++
 .../devicetree/bindings/display/msm/gpu.yaml       |   1 +
 arch/arm64/boot/dts/qcom/milos.dtsi                | 146 +++++++++++
 drivers/gpu/drm/msm/adreno/a6xx_catalog.c          | 291 +++++++++++++++++++++
 drivers/gpu/drm/msm/adreno/a6xx_gmu.c              |   8 +-
 drivers/gpu/drm/msm/adreno/a6xx_gpu.c              |  44 ++--
 drivers/gpu/drm/msm/adreno/a6xx_gpu.h              |  14 +-
 drivers/gpu/drm/msm/adreno/a8xx_gpu.c              |   6 +-
 drivers/gpu/drm/msm/adreno/adreno_gpu.h            |   5 +
 9 files changed, 507 insertions(+), 38 deletions(-)
---
base-commit: afed2755f671a38114a907c0bff6d35c3a76ef40
change-id: 20260330-adreno-810-5a47525522cd

Best regards,
-- 
Alexander Koskovich <akoskovich@pm.me>

[PATCH RFC v3 1/7] dt-bindings: display/msm/gmu: Document Adreno 810 GMU

[Alexander Koskovich]

Document Adreno 810 GMU in the dt-binding specification.

Reviewed-by: Krzysztof Kozlowski <krzysztof.kozlowski@oss.qualcomm.com>
Signed-off-by: Alexander Koskovich <akoskovich@pm.me>
---
 .../devicetree/bindings/display/msm/gmu.yaml       | 30 ++++++++++++++++++++++
 1 file changed, 30 insertions(+)

diff --git a/Documentation/devicetree/bindings/display/msm/gmu.yaml b/Documentation/devicetree/bindings/display/msm/gmu.yaml
index e32056ae0f5d..cc03cda67ea7 100644
--- a/Documentation/devicetree/bindings/display/msm/gmu.yaml
+++ b/Documentation/devicetree/bindings/display/msm/gmu.yaml
@@ -299,6 +299,36 @@ allOf:
       required:
         - qcom,qmp
 
+  - if:
+      properties:
+        compatible:
+          contains:
+            const: qcom,adreno-gmu-810.0
+    then:
+      properties:
+        reg:
+          items:
+            - description: Core GMU registers
+        reg-names:
+          items:
+            - const: gmu
+        clocks:
+          items:
+            - description: GPU AHB clock
+            - description: GMU clock
+            - description: GPU CX clock
+            - description: GPU AXI clock
+            - description: GPU MEMNOC clock
+            - description: GMU HUB clock
+        clock-names:
+          items:
+            - const: ahb
+            - const: gmu
+            - const: cxo
+            - const: axi
+            - const: memnoc
+            - const: hub
+
   - if:
       properties:
         compatible:

-- 
2.53.0

Claude review: dt-bindings: display/msm/gmu: Document Adreno 810 GMU

[Claude Code Review Bot]

Patch Review

Straightforward binding addition for `qcom,adreno-gmu-810.0`. The clock list has 6 entries (ahb, gmu, cxo, axi, memnoc, hub) — no DEMET, which matches the v3 changelog. Only one `reg` region (gmu) is specified, consistent with A810 not needing the rscc region that some other GMUs use.

Already has Reviewed-by from Krzysztof Kozlowski. Looks correct.

---
Generated by Claude Code Patch Reviewer

[PATCH RFC v3 2/7] dt-bindings: display/msm/gpu: Document A810 GPU

[Alexander Koskovich]

Document the GPU compatible string used for the Adreno 810.

Signed-off-by: Alexander Koskovich <akoskovich@pm.me>
---
 Documentation/devicetree/bindings/display/msm/gpu.yaml | 1 +
 1 file changed, 1 insertion(+)

diff --git a/Documentation/devicetree/bindings/display/msm/gpu.yaml b/Documentation/devicetree/bindings/display/msm/gpu.yaml
index ec84b64d4c00..90ae435146d5 100644
--- a/Documentation/devicetree/bindings/display/msm/gpu.yaml
+++ b/Documentation/devicetree/bindings/display/msm/gpu.yaml
@@ -434,6 +434,7 @@ allOf:
               - qcom,adreno-43050a01
               - qcom,adreno-43050c01
               - qcom,adreno-43051401
+              - qcom,adreno-44010000
 
     then: # Starting with A6xx, the clocks are usually defined in the GMU node
       properties:

-- 
2.53.0

Claude review: dt-bindings: display/msm/gpu: Document A810 GPU

[Claude Code Review Bot]

Patch Review

Single-line addition of `qcom,adreno-44010000` to the compatible list. The v3 changelog notes the regex-based compatible is gone in 7.0+, so this explicit entry is correct.

No issues.

---
Generated by Claude Code Patch Reviewer

[PATCH RFC v3 3/7] drm/msm/adreno: rename llc_mmio to cx_misc_mmio

[Alexander Koskovich]

This region is used for more than just LLCC, it also provides access to
software fuse values (raytracing, etc).

Rename relevant symbols from _llc to _cx_misc for use in a follow up
change that decouples this from LLCC.

Reviewed-by: Akhil P Oommen <akhilpo@oss.qualcomm.com>
Signed-off-by: Alexander Koskovich <akoskovich@pm.me>
---
 drivers/gpu/drm/msm/adreno/a6xx_gmu.c |  8 ++++----
 drivers/gpu/drm/msm/adreno/a6xx_gpu.c | 16 ++++++++--------
 drivers/gpu/drm/msm/adreno/a6xx_gpu.h | 14 +++++++-------
 drivers/gpu/drm/msm/adreno/a8xx_gpu.c |  2 +-
 4 files changed, 20 insertions(+), 20 deletions(-)

diff --git a/drivers/gpu/drm/msm/adreno/a6xx_gmu.c b/drivers/gpu/drm/msm/adreno/a6xx_gmu.c
index b7166a883b01..6a369682bb80 100644
--- a/drivers/gpu/drm/msm/adreno/a6xx_gmu.c
+++ b/drivers/gpu/drm/msm/adreno/a6xx_gmu.c
@@ -947,7 +947,7 @@ static int a6xx_gmu_fw_start(struct a6xx_gmu *gmu, unsigned int state)
 
 	/* Turn on TCM (Tightly Coupled Memory) retention */
 	if (adreno_is_a7xx(adreno_gpu))
-		a6xx_llc_write(a6xx_gpu, REG_A7XX_CX_MISC_TCM_RET_CNTL, 1);
+		a6xx_cx_misc_write(a6xx_gpu, REG_A7XX_CX_MISC_TCM_RET_CNTL, 1);
 	else if (!adreno_is_a8xx(adreno_gpu))
 		gmu_write(gmu, REG_A6XX_GMU_GENERAL_7, 1);
 
@@ -1215,7 +1215,7 @@ static int a6xx_gmu_secure_init(struct a6xx_gpu *a6xx_gpu)
 		if (!qcom_scm_is_available()) {
 			dev_warn_once(gpu->dev->dev,
 				"SCM is not available, poking fuse register\n");
-			a6xx_llc_write(a6xx_gpu, REG_A7XX_CX_MISC_SW_FUSE_VALUE,
+			a6xx_cx_misc_write(a6xx_gpu, REG_A7XX_CX_MISC_SW_FUSE_VALUE,
 				A7XX_CX_MISC_SW_FUSE_VALUE_RAYTRACING |
 				A7XX_CX_MISC_SW_FUSE_VALUE_FASTBLEND |
 				A7XX_CX_MISC_SW_FUSE_VALUE_LPAC);
@@ -1236,7 +1236,7 @@ static int a6xx_gmu_secure_init(struct a6xx_gpu *a6xx_gpu)
 		 * firmware, find out whether that's the case. The scm call
 		 * above sets the fuse register.
 		 */
-		fuse_val = a6xx_llc_read(a6xx_gpu,
+		fuse_val = a6xx_cx_misc_read(a6xx_gpu,
 					 REG_A7XX_CX_MISC_SW_FUSE_VALUE);
 		adreno_gpu->has_ray_tracing =
 			!!(fuse_val & A7XX_CX_MISC_SW_FUSE_VALUE_RAYTRACING);
@@ -1343,7 +1343,7 @@ int a6xx_gmu_resume(struct a6xx_gpu *a6xx_gpu)
 
 	/* Check to see if we are doing a cold or warm boot */
 	if (adreno_is_a7xx(adreno_gpu) || adreno_is_a8xx(adreno_gpu)) {
-		status = a6xx_llc_read(a6xx_gpu, REG_A7XX_CX_MISC_TCM_RET_CNTL) == 1 ?
+		status = a6xx_cx_misc_read(a6xx_gpu, REG_A7XX_CX_MISC_TCM_RET_CNTL) == 1 ?
 			GMU_WARM_BOOT : GMU_COLD_BOOT;
 	} else if (gmu->legacy) {
 		status = gmu_read(gmu, REG_A6XX_GMU_GENERAL_7) == 1 ?
diff --git a/drivers/gpu/drm/msm/adreno/a6xx_gpu.c b/drivers/gpu/drm/msm/adreno/a6xx_gpu.c
index e1eae6cb1e40..9847f83b92af 100644
--- a/drivers/gpu/drm/msm/adreno/a6xx_gpu.c
+++ b/drivers/gpu/drm/msm/adreno/a6xx_gpu.c
@@ -2039,7 +2039,7 @@ static void a6xx_llc_activate(struct a6xx_gpu *a6xx_gpu)
 	struct msm_gpu *gpu = &adreno_gpu->base;
 	u32 cntl1_regval = 0;
 
-	if (IS_ERR(a6xx_gpu->llc_mmio))
+	if (IS_ERR(a6xx_gpu->cx_misc_mmio))
 		return;
 
 	if (!llcc_slice_activate(a6xx_gpu->llc_slice)) {
@@ -2078,14 +2078,14 @@ static void a6xx_llc_activate(struct a6xx_gpu *a6xx_gpu)
 	 * pagetables
 	 */
 	if (!a6xx_gpu->have_mmu500) {
-		a6xx_llc_write(a6xx_gpu,
+		a6xx_cx_misc_write(a6xx_gpu,
 			REG_A6XX_CX_MISC_SYSTEM_CACHE_CNTL_1, cntl1_regval);
 
 		/*
 		 * Program cacheability overrides to not allocate cache
 		 * lines on a write miss
 		 */
-		a6xx_llc_rmw(a6xx_gpu,
+		a6xx_cx_misc_rmw(a6xx_gpu,
 			REG_A6XX_CX_MISC_SYSTEM_CACHE_CNTL_0, 0xF, 0x03);
 		return;
 	}
@@ -2098,7 +2098,7 @@ static void a7xx_llc_activate(struct a6xx_gpu *a6xx_gpu)
 	struct adreno_gpu *adreno_gpu = &a6xx_gpu->base;
 	struct msm_gpu *gpu = &adreno_gpu->base;
 
-	if (IS_ERR(a6xx_gpu->llc_mmio))
+	if (IS_ERR(a6xx_gpu->cx_misc_mmio))
 		return;
 
 	if (!llcc_slice_activate(a6xx_gpu->llc_slice)) {
@@ -2151,15 +2151,15 @@ static void a6xx_llc_slices_init(struct platform_device *pdev,
 	of_node_put(phandle);
 
 	if (is_a7xx || !a6xx_gpu->have_mmu500)
-		a6xx_gpu->llc_mmio = msm_ioremap(pdev, "cx_mem");
+		a6xx_gpu->cx_misc_mmio = msm_ioremap(pdev, "cx_mem");
 	else
-		a6xx_gpu->llc_mmio = NULL;
+		a6xx_gpu->cx_misc_mmio = NULL;
 
 	a6xx_gpu->llc_slice = llcc_slice_getd(LLCC_GPU);
 	a6xx_gpu->htw_llc_slice = llcc_slice_getd(LLCC_GPUHTW);
 
 	if (IS_ERR_OR_NULL(a6xx_gpu->llc_slice) && IS_ERR_OR_NULL(a6xx_gpu->htw_llc_slice))
-		a6xx_gpu->llc_mmio = ERR_PTR(-EINVAL);
+		a6xx_gpu->cx_misc_mmio = ERR_PTR(-EINVAL);
 }
 
 #define GBIF_CLIENT_HALT_MASK		BIT(0)
@@ -2560,7 +2560,7 @@ static int a6xx_read_speedbin(struct device *dev, struct a6xx_gpu *a6xx_gpu,
 		return ret;
 
 	if (info->quirks & ADRENO_QUIRK_SOFTFUSE) {
-		*speedbin = a6xx_llc_read(a6xx_gpu, REG_A8XX_CX_MISC_SW_FUSE_FREQ_LIMIT_STATUS);
+		*speedbin = a6xx_cx_misc_read(a6xx_gpu, REG_A8XX_CX_MISC_SW_FUSE_FREQ_LIMIT_STATUS);
 		*speedbin = A8XX_CX_MISC_SW_FUSE_FREQ_LIMIT_STATUS_FINALFREQLIMIT(*speedbin);
 		return 0;
 	}
diff --git a/drivers/gpu/drm/msm/adreno/a6xx_gpu.h b/drivers/gpu/drm/msm/adreno/a6xx_gpu.h
index eb431e5e00b1..648608c1c98e 100644
--- a/drivers/gpu/drm/msm/adreno/a6xx_gpu.h
+++ b/drivers/gpu/drm/msm/adreno/a6xx_gpu.h
@@ -102,7 +102,7 @@ struct a6xx_gpu {
 
 	bool has_whereami;
 
-	void __iomem *llc_mmio;
+	void __iomem *cx_misc_mmio;
 	void *llc_slice;
 	void *htw_llc_slice;
 	bool have_mmu500;
@@ -240,19 +240,19 @@ static inline bool a6xx_has_gbif(struct adreno_gpu *gpu)
 	return true;
 }
 
-static inline void a6xx_llc_rmw(struct a6xx_gpu *a6xx_gpu, u32 reg, u32 mask, u32 or)
+static inline void a6xx_cx_misc_rmw(struct a6xx_gpu *a6xx_gpu, u32 reg, u32 mask, u32 or)
 {
-	return msm_rmw(a6xx_gpu->llc_mmio + (reg << 2), mask, or);
+	return msm_rmw(a6xx_gpu->cx_misc_mmio + (reg << 2), mask, or);
 }
 
-static inline u32 a6xx_llc_read(struct a6xx_gpu *a6xx_gpu, u32 reg)
+static inline u32 a6xx_cx_misc_read(struct a6xx_gpu *a6xx_gpu, u32 reg)
 {
-	return readl(a6xx_gpu->llc_mmio + (reg << 2));
+	return readl(a6xx_gpu->cx_misc_mmio + (reg << 2));
 }
 
-static inline void a6xx_llc_write(struct a6xx_gpu *a6xx_gpu, u32 reg, u32 value)
+static inline void a6xx_cx_misc_write(struct a6xx_gpu *a6xx_gpu, u32 reg, u32 value)
 {
-	writel(value, a6xx_gpu->llc_mmio + (reg << 2));
+	writel(value, a6xx_gpu->cx_misc_mmio + (reg << 2));
 }
 
 #define shadowptr(_a6xx_gpu, _ring) ((_a6xx_gpu)->shadow_iova + \
diff --git a/drivers/gpu/drm/msm/adreno/a8xx_gpu.c b/drivers/gpu/drm/msm/adreno/a8xx_gpu.c
index edfd04bf6cd1..1515ade22578 100644
--- a/drivers/gpu/drm/msm/adreno/a8xx_gpu.c
+++ b/drivers/gpu/drm/msm/adreno/a8xx_gpu.c
@@ -103,7 +103,7 @@ void a8xx_gpu_get_slice_info(struct msm_gpu *gpu)
 		return;
 	}
 
-	slice_mask &= a6xx_llc_read(a6xx_gpu,
+	slice_mask &= a6xx_cx_misc_read(a6xx_gpu,
 			REG_A8XX_CX_MISC_SLICE_ENABLE_FINAL);
 
 	a6xx_gpu->slice_mask = slice_mask;

-- 
2.53.0

Re: [PATCH RFC v3 3/7] drm/msm/adreno: rename llc_mmio to cx_misc_mmio

[Konrad Dybcio]

On 4/8/26 3:45 AM, Alexander Koskovich wrote:
> This region is used for more than just LLCC, it also provides access to
> software fuse values (raytracing, etc).
> 
> Rename relevant symbols from _llc to _cx_misc for use in a follow up
> change that decouples this from LLCC.
> 
> Reviewed-by: Akhil P Oommen <akhilpo@oss.qualcomm.com>
> Signed-off-by: Alexander Koskovich <akoskovich@pm.me>
> ---

Seems you lost my tag

Reviewed-by: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>

Konrad

Claude review: drm/msm/adreno: rename llc_mmio to cx_misc_mmio

[Claude Code Review Bot]

Patch Review

Clean mechanical rename of `llc_mmio` -> `cx_misc_mmio` and corresponding accessor functions `a6xx_llc_{read,write,rmw}` -> `a6xx_cx_misc_{read,write,rmw}`. The rename is well-motivated: the CX_MISC region is used for more than just LLCC (software fuses for raytracing, TCM retention, etc.).

All callers are updated consistently across a6xx_gmu.c, a6xx_gpu.c, a6xx_gpu.h, and a8xx_gpu.c. Already has Reviewed-by from Akhil P Oommen. No issues.

---
Generated by Claude Code Patch Reviewer

[PATCH RFC v3 4/7] drm/msm/adreno: set cx_misc_mmio regardless of if platform has LLCC

[Alexander Koskovich]

Platforms without a LLCC (e.g. milos) still need to be able to read and
write to the cx_mem region. Previously if LLCC slices were unavailable
the cx_misc_mmio mapping was overwritten with ERR_PTR, causing a crash
when the GMU later accessed cx_mem.

Move the cx_misc_mmio mapping out of a6xx_llc_slices_init() into
a6xx_gpu_init() so that cx_mem mapping is independent of LLCC.

Reviewed-by: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>
Reviewed-by: Akhil P Oommen <akhilpo@oss.qualcomm.com>
Signed-off-by: Alexander Koskovich <akoskovich@pm.me>
---
 drivers/gpu/drm/msm/adreno/a6xx_gpu.c | 38 ++++++++++++++++-------------------
 1 file changed, 17 insertions(+), 21 deletions(-)

diff --git a/drivers/gpu/drm/msm/adreno/a6xx_gpu.c b/drivers/gpu/drm/msm/adreno/a6xx_gpu.c
index 9847f83b92af..d691ad1f88b3 100644
--- a/drivers/gpu/drm/msm/adreno/a6xx_gpu.c
+++ b/drivers/gpu/drm/msm/adreno/a6xx_gpu.c
@@ -2039,7 +2039,7 @@ static void a6xx_llc_activate(struct a6xx_gpu *a6xx_gpu)
 	struct msm_gpu *gpu = &adreno_gpu->base;
 	u32 cntl1_regval = 0;
 
-	if (IS_ERR(a6xx_gpu->cx_misc_mmio))
+	if (IS_ERR_OR_NULL(a6xx_gpu->llc_slice) && IS_ERR_OR_NULL(a6xx_gpu->htw_llc_slice))
 		return;
 
 	if (!llcc_slice_activate(a6xx_gpu->llc_slice)) {
@@ -2098,7 +2098,7 @@ static void a7xx_llc_activate(struct a6xx_gpu *a6xx_gpu)
 	struct adreno_gpu *adreno_gpu = &a6xx_gpu->base;
 	struct msm_gpu *gpu = &adreno_gpu->base;
 
-	if (IS_ERR(a6xx_gpu->cx_misc_mmio))
+	if (IS_ERR_OR_NULL(a6xx_gpu->llc_slice) && IS_ERR_OR_NULL(a6xx_gpu->htw_llc_slice))
 		return;
 
 	if (!llcc_slice_activate(a6xx_gpu->llc_slice)) {
@@ -2135,31 +2135,12 @@ static void a6xx_llc_slices_destroy(struct a6xx_gpu *a6xx_gpu)
 static void a6xx_llc_slices_init(struct platform_device *pdev,
 		struct a6xx_gpu *a6xx_gpu, bool is_a7xx)
 {
-	struct device_node *phandle;
-
 	/* No LLCC on non-RPMh (and by extension, non-GMU) SoCs */
 	if (adreno_has_gmu_wrapper(&a6xx_gpu->base))
 		return;
 
-	/*
-	 * There is a different programming path for A6xx targets with an
-	 * mmu500 attached, so detect if that is the case
-	 */
-	phandle = of_parse_phandle(pdev->dev.of_node, "iommus", 0);
-	a6xx_gpu->have_mmu500 = (phandle &&
-		of_device_is_compatible(phandle, "arm,mmu-500"));
-	of_node_put(phandle);
-
-	if (is_a7xx || !a6xx_gpu->have_mmu500)
-		a6xx_gpu->cx_misc_mmio = msm_ioremap(pdev, "cx_mem");
-	else
-		a6xx_gpu->cx_misc_mmio = NULL;
-
 	a6xx_gpu->llc_slice = llcc_slice_getd(LLCC_GPU);
 	a6xx_gpu->htw_llc_slice = llcc_slice_getd(LLCC_GPUHTW);
-
-	if (IS_ERR_OR_NULL(a6xx_gpu->llc_slice) && IS_ERR_OR_NULL(a6xx_gpu->htw_llc_slice))
-		a6xx_gpu->cx_misc_mmio = ERR_PTR(-EINVAL);
 }
 
 #define GBIF_CLIENT_HALT_MASK		BIT(0)
@@ -2621,6 +2602,7 @@ static struct msm_gpu *a6xx_gpu_init(struct drm_device *dev)
 	struct platform_device *pdev = priv->gpu_pdev;
 	struct adreno_platform_config *config = pdev->dev.platform_data;
 	const struct adreno_info *info = config->info;
+	struct device_node *phandle;
 	struct device_node *node;
 	struct a6xx_gpu *a6xx_gpu;
 	struct adreno_gpu *adreno_gpu;
@@ -2656,6 +2638,20 @@ static struct msm_gpu *a6xx_gpu_init(struct drm_device *dev)
 
 	a6xx_llc_slices_init(pdev, a6xx_gpu, is_a7xx);
 
+	/*
+	 * There is a different programming path for A6xx targets with an
+	 * mmu500 attached, so detect if that is the case
+	 */
+	phandle = of_parse_phandle(pdev->dev.of_node, "iommus", 0);
+	a6xx_gpu->have_mmu500 = (phandle &&
+		of_device_is_compatible(phandle, "arm,mmu-500"));
+	of_node_put(phandle);
+
+	if (is_a7xx || !a6xx_gpu->have_mmu500)
+		a6xx_gpu->cx_misc_mmio = msm_ioremap(pdev, "cx_mem");
+	else
+		a6xx_gpu->cx_misc_mmio = NULL;
+
 	ret = a6xx_set_supported_hw(&pdev->dev, a6xx_gpu, info);
 	if (ret) {
 		a6xx_llc_slices_destroy(a6xx_gpu);

-- 
2.53.0

Claude review: drm/msm/adreno: set cx_misc_mmio regardless of if platform has LLCC

[Claude Code Review Bot]

Patch Review

This patch decouples the `cx_misc_mmio` mapping from LLCC slice availability. The key motivation is that Milos has no LLCC but still needs access to CX_MISC registers (for TCM retention, software fuses, etc.).

The approach is sound:
- MMU500 detection and `cx_misc_mmio` mapping are moved from `a6xx_llc_slices_init()` into `a6xx_gpu_init()`.
- The early-return guards in `a6xx_llc_activate()` and `a7xx_llc_activate()` are changed from checking `cx_misc_mmio` to checking whether LLC slices are available directly:

```c
-	if (IS_ERR(a6xx_gpu->cx_misc_mmio))
+	if (IS_ERR_OR_NULL(a6xx_gpu->llc_slice) && IS_ERR_OR_NULL(a6xx_gpu->htw_llc_slice))
```

This is safe because `a7xx_llc_activate()` (the path used for A8xx) does not use `cx_misc_mmio` — it only writes GPU registers via `gpu_write()` and calls `llcc_slice_activate()`. The `a6xx_llc_activate()` path does use `cx_misc_mmio` but only inside the `if (!a6xx_gpu->have_mmu500)` block, and for platforms that reach that code, `cx_misc_mmio` will be mapped (since `!have_mmu500` is one of the conditions for mapping it).

Already has Reviewed-by from both Konrad Dybcio and Akhil P Oommen. Looks correct.

---
Generated by Claude Code Patch Reviewer

[PATCH RFC v3 5/7] drm/msm/a8xx: use pipe protect slot 15 for last-span-unbound feature

[Alexander Koskovich]

A8XX GPUs have two sets of protect registers: 64 global slots and 16
pipe specific slots. The last-span-unbound feature is only available
on pipe protect registers, and should always target pipe slot 15.

This matches the downstream driver which hardcodes pipe slot 15 for
all A8XX GPUs (GRAPHICS.LA.15.0.r1) and resolves protect errors on
A810.

Reviewed-by: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>
Reviewed-by: Akhil P Oommen <akhilpo@oss.qualcomm.com>
Signed-off-by: Alexander Koskovich <akoskovich@pm.me>
---
 drivers/gpu/drm/msm/adreno/a8xx_gpu.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/msm/adreno/a8xx_gpu.c b/drivers/gpu/drm/msm/adreno/a8xx_gpu.c
index 1515ade22578..89a494bcea74 100644
--- a/drivers/gpu/drm/msm/adreno/a8xx_gpu.c
+++ b/drivers/gpu/drm/msm/adreno/a8xx_gpu.c
@@ -255,8 +255,8 @@ static void a8xx_set_cp_protect(struct msm_gpu *gpu)
 	 * Last span feature is only supported on PIPE specific register.
 	 * So update those here
 	 */
-	a8xx_write_pipe(gpu, PIPE_BR, REG_A8XX_CP_PROTECT_PIPE(protect->count_max), final_cfg);
-	a8xx_write_pipe(gpu, PIPE_BV, REG_A8XX_CP_PROTECT_PIPE(protect->count_max), final_cfg);
+	a8xx_write_pipe(gpu, PIPE_BR, REG_A8XX_CP_PROTECT_PIPE(15), final_cfg);
+	a8xx_write_pipe(gpu, PIPE_BV, REG_A8XX_CP_PROTECT_PIPE(15), final_cfg);
 
 	a8xx_aperture_clear(gpu);
 }

-- 
2.53.0

Claude review: drm/msm/a8xx: use pipe protect slot 15 for last-span-unbound feature

[Claude Code Review Bot]

Patch Review

This patch hardcodes pipe protect slot 15 for the last-span-unbound feature:

```c
-	a8xx_write_pipe(gpu, PIPE_BR, REG_A8XX_CP_PROTECT_PIPE(protect->count_max), final_cfg);
-	a8xx_write_pipe(gpu, PIPE_BV, REG_A8XX_CP_PROTECT_PIPE(protect->count_max), final_cfg);
+	a8xx_write_pipe(gpu, PIPE_BR, REG_A8XX_CP_PROTECT_PIPE(15), final_cfg);
+	a8xx_write_pipe(gpu, PIPE_BV, REG_A8XX_CP_PROTECT_PIPE(15), final_cfg);
```

This is necessary because A8xx GPUs have 16 pipe protect slots (0-15) and 64 global protect slots. The existing x285 and a840 entries have `count_max = 15`, which coincidentally matched the last pipe slot. But A810 needs `count_max = 64` (its actual global slot count), which would cause an out-of-bounds pipe slot access without this fix.

The commit message correctly explains this matches downstream behavior. Already has Reviewed-by from Konrad Dybcio and Akhil P Oommen.

**Suggestion:** Since `count_max` is no longer used for the pipe slot, consider also updating x285 and a840 to `count_max = 64` for consistency, and adding them to `__build_asserts`. Currently x285 has 52 global entries and a840 has 54, both with `count_max = 15` — which would fail BUILD_BUG_ON if added to asserts. This isn't introduced by this series, but it's a good opportunity to clean it up.

---
Generated by Claude Code Patch Reviewer

[PATCH RFC v3 6/7] drm/msm/adreno: add Adreno 810 GPU support

[Alexander Koskovich]

Add catalog entry and register configuration for the Adreno 810
found in Qualcomm SM7635 (Milos) based devices.

Signed-off-by: Alexander Koskovich <akoskovich@pm.me>
---
 drivers/gpu/drm/msm/adreno/a6xx_catalog.c | 291 ++++++++++++++++++++++++++++++
 drivers/gpu/drm/msm/adreno/adreno_gpu.h   |   5 +
 2 files changed, 296 insertions(+)

diff --git a/drivers/gpu/drm/msm/adreno/a6xx_catalog.c b/drivers/gpu/drm/msm/adreno/a6xx_catalog.c
index 550ff3a9b82e..328c624db7cb 100644
--- a/drivers/gpu/drm/msm/adreno/a6xx_catalog.c
+++ b/drivers/gpu/drm/msm/adreno/a6xx_catalog.c
@@ -1799,6 +1799,259 @@ static const struct adreno_reglist_pipe x285_dyn_pwrup_reglist_regs[] = {
 };
 DECLARE_ADRENO_REGLIST_PIPE_LIST(x285_dyn_pwrup_reglist);
 
+static const struct adreno_reglist_pipe a810_nonctxt_regs[] = {
+	{ REG_A8XX_CP_SMMU_STREAM_ID_LPAC, 0x00000101, BIT(PIPE_NONE) },
+	{ REG_A8XX_GRAS_DBG_ECO_CNTL, 0x00f80800, BIT(PIPE_BV) | BIT(PIPE_BR) },
+	{ REG_A6XX_PC_AUTO_VERTEX_STRIDE, 0x00000001, BIT(PIPE_BV) | BIT(PIPE_BR) },
+	{ REG_A8XX_PC_VIS_STREAM_CNTL, 0x10010000, BIT(PIPE_BV) | BIT(PIPE_BR) },
+	{ REG_A8XX_PC_CONTEXT_SWITCH_STABILIZE_CNTL_1, 0x00000002, BIT(PIPE_BV) | BIT(PIPE_BR) },
+	{ REG_A8XX_PC_CHICKEN_BITS_1, 0x00000003, BIT(PIPE_BR) },
+	{ REG_A8XX_PC_CHICKEN_BITS_1, 0x00000023, BIT(PIPE_BV) }, /* Avoid partial waves at VFD */
+	{ REG_A8XX_PC_CHICKEN_BITS_2, 0x00000200, BIT(PIPE_BV) | BIT(PIPE_BR) },
+	{ REG_A8XX_PC_CHICKEN_BITS_3, 0x00500000, BIT(PIPE_BV) | BIT(PIPE_BR) },
+	{ REG_A8XX_PC_CHICKEN_BITS_4, 0x00500050, BIT(PIPE_BV) | BIT(PIPE_BR) },
+	{ REG_A7XX_RB_CCU_CNTL, 0x00000068, BIT(PIPE_BR) },
+	{ REG_A8XX_RB_RESOLVE_PREFETCH_CNTL, 0x00000007, BIT(PIPE_BR) },
+	{ REG_A8XX_RB_CMP_DBG_ECO_CNTL, 0x00004000, BIT(PIPE_BR) },
+	{ REG_A8XX_RBBM_NC_MODE_CNTL, 0x00000001, BIT(PIPE_NONE) },
+	{ REG_A8XX_RBBM_SLICE_NC_MODE_CNTL, 0x00000001, BIT(PIPE_NONE) },
+	{ REG_A8XX_RBBM_WAIT_IDLE_CLOCKS_CNTL, 0x00000030, BIT(PIPE_NONE) },
+	{ REG_A8XX_RBBM_WAIT_IDLE_CLOCKS_CNTL2, 0x00000030, BIT(PIPE_NONE) },
+	{ REG_A8XX_UCHE_GBIF_GX_CONFIG, 0x010240e0, BIT(PIPE_NONE) },
+	{ REG_A8XX_RBBM_GBIF_CLIENT_QOS_CNTL, 0x22122212, BIT(PIPE_NONE) },
+	{ REG_A8XX_RBBM_CGC_P2S_CNTL, 0x00000040, BIT(PIPE_NONE) },
+	/*
+	 * BIT(22): Disable PS out of order retire
+	 * BIT(23): Enable half wave mode and MM instruction src&dst is half precision
+	 */
+	{ REG_A7XX_SP_CHICKEN_BITS_2, BIT(22) | BIT(23), BIT(PIPE_NONE) },
+	{ REG_A7XX_SP_CHICKEN_BITS_3, 0x00300000, BIT(PIPE_NONE) },
+	{ REG_A6XX_SP_PERFCTR_SHADER_MASK, 0x0000003f, BIT(PIPE_NONE) },
+	{ REG_A7XX_SP_HLSQ_TIMEOUT_THRESHOLD_DP, 0x00000080, BIT(PIPE_NONE) },
+	{ REG_A7XX_SP_READ_SEL, 0x0001ff00, BIT(PIPE_NONE) },
+	{ REG_A6XX_TPL1_DBG_ECO_CNTL, 0x10000000, BIT(PIPE_NONE) },
+	{ REG_A6XX_TPL1_DBG_ECO_CNTL1, 0x00000724, BIT(PIPE_NONE) },
+	{ REG_A6XX_UCHE_MODE_CNTL, 0x00020000, BIT(PIPE_NONE) },
+	{ REG_A8XX_UCHE_CCHE_MODE_CNTL, 0x00001000, BIT(PIPE_NONE) },
+	{ REG_A8XX_UCHE_CCHE_CACHE_WAYS, 0x00000800, BIT(PIPE_NONE) },
+	{ REG_A8XX_UCHE_CACHE_WAYS, 0x00080000, BIT(PIPE_NONE) },
+	{ REG_A8XX_UCHE_VARB_IDLE_TIMEOUT, 0x00000020, BIT(PIPE_NONE) },
+	{ REG_A7XX_VFD_DBG_ECO_CNTL, 0x00008000, BIT(PIPE_BV) | BIT(PIPE_BR) },
+	{ REG_A8XX_VFD_CB_BV_THRESHOLD, 0x00500050, BIT(PIPE_BV) | BIT(PIPE_BR) },
+	{ REG_A8XX_VFD_CB_BR_THRESHOLD, 0x00600060, BIT(PIPE_BV) | BIT(PIPE_BR) },
+	{ REG_A8XX_VFD_CB_BUSY_REQ_CNT, 0x00200020, BIT(PIPE_BV) | BIT(PIPE_BR) },
+	{ REG_A8XX_VFD_CB_LP_REQ_CNT, 0x00100020, BIT(PIPE_BV) | BIT(PIPE_BR) },
+	{ REG_A8XX_VPC_FLATSHADE_MODE_CNTL, 0x00000001, BIT(PIPE_BV) | BIT(PIPE_BR) },
+	{ REG_A8XX_VSC_BIN_SIZE, 0x00010001, BIT(PIPE_NONE) },
+	{ REG_A8XX_RB_GC_GMEM_PROTECT, 0x00900000, BIT(PIPE_BR) },
+	{ },
+};
+
+static const u32 a810_protect_regs[] = {
+	A6XX_PROTECT_RDONLY(0x00000, 0x03a3),
+	A6XX_PROTECT_RDONLY(0x003b4, 0x008b),
+	A6XX_PROTECT_NORDWR(0x00440, 0x001f),
+	A6XX_PROTECT_RDONLY(0x00580, 0x005f),
+	A6XX_PROTECT_NORDWR(0x005e0, 0x011f),
+	A6XX_PROTECT_RDONLY(0x0074a, 0x0005),
+	A6XX_PROTECT_RDONLY(0x00759, 0x0026),
+	A6XX_PROTECT_RDONLY(0x00789, 0x0000),
+	A6XX_PROTECT_RDONLY(0x0078c, 0x0013),
+	A6XX_PROTECT_NORDWR(0x00800, 0x0029),
+	A6XX_PROTECT_NORDWR(0x00837, 0x00af),
+	A6XX_PROTECT_RDONLY(0x008e7, 0x00c9),
+	A6XX_PROTECT_NORDWR(0x008ec, 0x00c3),
+	A6XX_PROTECT_NORDWR(0x009b1, 0x0250),
+	A6XX_PROTECT_RDONLY(0x00ce0, 0x0001),
+	A6XX_PROTECT_RDONLY(0x00df0, 0x0000),
+	A6XX_PROTECT_NORDWR(0x00df1, 0x0000),
+	A6XX_PROTECT_NORDWR(0x00e01, 0x0000),
+	A6XX_PROTECT_NORDWR(0x00e03, 0x1fff),
+	A6XX_PROTECT_NORDWR(0x03c00, 0x00c5),
+	A6XX_PROTECT_RDONLY(0x03cc6, 0x1fff),
+	A6XX_PROTECT_NORDWR(0x08600, 0x01ff),
+	A6XX_PROTECT_NORDWR(0x08e00, 0x00ff),
+	A6XX_PROTECT_RDONLY(0x08f00, 0x0000),
+	A6XX_PROTECT_NORDWR(0x08f01, 0x01be),
+	A6XX_PROTECT_NORDWR(0x09600, 0x01ff),
+	A6XX_PROTECT_RDONLY(0x0981a, 0x02e5),
+	A6XX_PROTECT_NORDWR(0x09e00, 0x01ff),
+	A6XX_PROTECT_NORDWR(0x0a600, 0x01ff),
+	A6XX_PROTECT_NORDWR(0x0ae00, 0x0006),
+	A6XX_PROTECT_NORDWR(0x0ae08, 0x0006),
+	A6XX_PROTECT_NORDWR(0x0ae10, 0x036f),
+	A6XX_PROTECT_NORDWR(0x0b600, 0x1fff),
+	A6XX_PROTECT_NORDWR(0x0dc00, 0x1fff),
+	A6XX_PROTECT_RDONLY(0x0fc00, 0x1fff),
+	A6XX_PROTECT_NORDWR(0x18400, 0x003f),
+	A6XX_PROTECT_RDONLY(0x18440, 0x013f),
+	A6XX_PROTECT_NORDWR(0x18580, 0x1fff),
+	A6XX_PROTECT_NORDWR(0x1b400, 0x1fff),
+	A6XX_PROTECT_NORDWR(0x1f400, 0x0477),
+	A6XX_PROTECT_RDONLY(0x1f878, 0x0787),
+	A6XX_PROTECT_NORDWR(0x1f930, 0x0329),
+	A6XX_PROTECT_NORDWR(0x20000, 0x1fff),
+	A6XX_PROTECT_NORDWR(0x27800, 0x007f),
+	A6XX_PROTECT_RDONLY(0x27880, 0x0381),
+	A6XX_PROTECT_NORDWR(0x27882, 0x0001),
+	A6XX_PROTECT_NORDWR(0x27c02, 0x0000),
+};
+DECLARE_ADRENO_PROTECT(a810_protect, 64);
+
+static const uint32_t a810_pwrup_reglist_regs[] = {
+	REG_A6XX_UCHE_MODE_CNTL,
+	REG_A8XX_UCHE_VARB_IDLE_TIMEOUT,
+	REG_A8XX_UCHE_GBIF_GX_CONFIG,
+	REG_A8XX_UCHE_CACHE_WAYS,
+	REG_A8XX_UCHE_CCHE_MODE_CNTL,
+	REG_A8XX_UCHE_CCHE_CACHE_WAYS,
+	REG_A8XX_UCHE_CCHE_GC_GMEM_RANGE_MIN,
+	REG_A8XX_UCHE_CCHE_GC_GMEM_RANGE_MIN + 1,
+	REG_A8XX_UCHE_CCHE_TRAP_BASE,
+	REG_A8XX_UCHE_CCHE_TRAP_BASE + 1,
+	REG_A8XX_UCHE_CCHE_WRITE_THRU_BASE,
+	REG_A8XX_UCHE_CCHE_WRITE_THRU_BASE + 1,
+	REG_A8XX_UCHE_WRITE_THRU_BASE,
+	REG_A8XX_UCHE_WRITE_THRU_BASE + 1,
+	REG_A8XX_UCHE_TRAP_BASE,
+	REG_A8XX_UCHE_TRAP_BASE + 1,
+	REG_A8XX_UCHE_CLIENT_PF,
+	REG_A8XX_VSC_BIN_SIZE,
+	REG_A8XX_RB_CMP_NC_MODE_CNTL,
+	REG_A7XX_SP_HLSQ_TIMEOUT_THRESHOLD_DP,
+	REG_A8XX_SP_HLSQ_GC_GMEM_RANGE_MIN,
+	REG_A8XX_SP_HLSQ_GC_GMEM_RANGE_MIN + 1,
+	REG_A7XX_SP_READ_SEL,
+	REG_A8XX_TPL1_BICUBIC_WEIGHTS_TABLE(1),
+	REG_A8XX_TPL1_BICUBIC_WEIGHTS_TABLE(2),
+	REG_A8XX_TPL1_BICUBIC_WEIGHTS_TABLE(3),
+	REG_A8XX_TPL1_BICUBIC_WEIGHTS_TABLE(4),
+	REG_A8XX_TPL1_BICUBIC_WEIGHTS_TABLE(5),
+	REG_A8XX_TPL1_BICUBIC_WEIGHTS_TABLE(6),
+	REG_A8XX_TPL1_BICUBIC_WEIGHTS_TABLE(7),
+	REG_A8XX_TPL1_BICUBIC_WEIGHTS_TABLE(8),
+	REG_A8XX_TPL1_BICUBIC_WEIGHTS_TABLE(9),
+	REG_A8XX_TPL1_BICUBIC_WEIGHTS_TABLE(10),
+	REG_A8XX_TPL1_BICUBIC_WEIGHTS_TABLE(11),
+	REG_A8XX_TPL1_BICUBIC_WEIGHTS_TABLE(12),
+	REG_A8XX_TPL1_BICUBIC_WEIGHTS_TABLE(13),
+	REG_A8XX_TPL1_BICUBIC_WEIGHTS_TABLE(14),
+	REG_A8XX_TPL1_BICUBIC_WEIGHTS_TABLE(15),
+	REG_A8XX_TPL1_BICUBIC_WEIGHTS_TABLE(16),
+	REG_A8XX_TPL1_BICUBIC_WEIGHTS_TABLE(17),
+	REG_A8XX_TPL1_BICUBIC_WEIGHTS_TABLE(18),
+	REG_A8XX_TPL1_BICUBIC_WEIGHTS_TABLE(19),
+};
+DECLARE_ADRENO_REGLIST_LIST(a810_pwrup_reglist);
+
+static const u32 a810_ifpc_reglist_regs[] = {
+	REG_A8XX_RBBM_NC_MODE_CNTL,
+	REG_A8XX_RBBM_PERFCTR_CNTL,
+	REG_A8XX_RBBM_SLICE_INTERFACE_HANG_INT_CNTL,
+	REG_A8XX_RBBM_SLICE_NC_MODE_CNTL,
+	REG_A6XX_SP_NC_MODE_CNTL,
+	REG_A7XX_SP_CHICKEN_BITS_2,
+	REG_A7XX_SP_CHICKEN_BITS_3,
+	REG_A6XX_SP_PERFCTR_SHADER_MASK,
+	REG_A6XX_TPL1_NC_MODE_CNTL,
+	REG_A6XX_TPL1_DBG_ECO_CNTL,
+	REG_A6XX_TPL1_DBG_ECO_CNTL1,
+	REG_A8XX_CP_PROTECT_GLOBAL(0),
+	REG_A8XX_CP_PROTECT_GLOBAL(1),
+	REG_A8XX_CP_PROTECT_GLOBAL(2),
+	REG_A8XX_CP_PROTECT_GLOBAL(3),
+	REG_A8XX_CP_PROTECT_GLOBAL(4),
+	REG_A8XX_CP_PROTECT_GLOBAL(5),
+	REG_A8XX_CP_PROTECT_GLOBAL(6),
+	REG_A8XX_CP_PROTECT_GLOBAL(7),
+	REG_A8XX_CP_PROTECT_GLOBAL(8),
+	REG_A8XX_CP_PROTECT_GLOBAL(9),
+	REG_A8XX_CP_PROTECT_GLOBAL(10),
+	REG_A8XX_CP_PROTECT_GLOBAL(11),
+	REG_A8XX_CP_PROTECT_GLOBAL(12),
+	REG_A8XX_CP_PROTECT_GLOBAL(13),
+	REG_A8XX_CP_PROTECT_GLOBAL(14),
+	REG_A8XX_CP_PROTECT_GLOBAL(15),
+	REG_A8XX_CP_PROTECT_GLOBAL(16),
+	REG_A8XX_CP_PROTECT_GLOBAL(17),
+	REG_A8XX_CP_PROTECT_GLOBAL(18),
+	REG_A8XX_CP_PROTECT_GLOBAL(19),
+	REG_A8XX_CP_PROTECT_GLOBAL(20),
+	REG_A8XX_CP_PROTECT_GLOBAL(21),
+	REG_A8XX_CP_PROTECT_GLOBAL(22),
+	REG_A8XX_CP_PROTECT_GLOBAL(23),
+	REG_A8XX_CP_PROTECT_GLOBAL(24),
+	REG_A8XX_CP_PROTECT_GLOBAL(25),
+	REG_A8XX_CP_PROTECT_GLOBAL(26),
+	REG_A8XX_CP_PROTECT_GLOBAL(27),
+	REG_A8XX_CP_PROTECT_GLOBAL(28),
+	REG_A8XX_CP_PROTECT_GLOBAL(29),
+	REG_A8XX_CP_PROTECT_GLOBAL(30),
+	REG_A8XX_CP_PROTECT_GLOBAL(31),
+	REG_A8XX_CP_PROTECT_GLOBAL(32),
+	REG_A8XX_CP_PROTECT_GLOBAL(33),
+	REG_A8XX_CP_PROTECT_GLOBAL(34),
+	REG_A8XX_CP_PROTECT_GLOBAL(35),
+	REG_A8XX_CP_PROTECT_GLOBAL(36),
+	REG_A8XX_CP_PROTECT_GLOBAL(37),
+	REG_A8XX_CP_PROTECT_GLOBAL(38),
+	REG_A8XX_CP_PROTECT_GLOBAL(39),
+	REG_A8XX_CP_PROTECT_GLOBAL(40),
+	REG_A8XX_CP_PROTECT_GLOBAL(41),
+	REG_A8XX_CP_PROTECT_GLOBAL(42),
+	REG_A8XX_CP_PROTECT_GLOBAL(43),
+	REG_A8XX_CP_PROTECT_GLOBAL(44),
+	REG_A8XX_CP_PROTECT_GLOBAL(45),
+	REG_A8XX_CP_PROTECT_GLOBAL(46),
+	REG_A8XX_CP_PROTECT_GLOBAL(47),
+	REG_A8XX_CP_PROTECT_GLOBAL(48),
+	REG_A8XX_CP_PROTECT_GLOBAL(49),
+	REG_A8XX_CP_PROTECT_GLOBAL(50),
+	REG_A8XX_CP_PROTECT_GLOBAL(51),
+	REG_A8XX_CP_PROTECT_GLOBAL(52),
+	REG_A8XX_CP_PROTECT_GLOBAL(53),
+	REG_A8XX_CP_PROTECT_GLOBAL(54),
+	REG_A8XX_CP_PROTECT_GLOBAL(55),
+	REG_A8XX_CP_PROTECT_GLOBAL(56),
+	REG_A8XX_CP_PROTECT_GLOBAL(57),
+	REG_A8XX_CP_PROTECT_GLOBAL(58),
+	REG_A8XX_CP_PROTECT_GLOBAL(59),
+	REG_A8XX_CP_PROTECT_GLOBAL(60),
+	REG_A8XX_CP_PROTECT_GLOBAL(61),
+	REG_A8XX_CP_PROTECT_GLOBAL(62),
+	REG_A8XX_CP_PROTECT_GLOBAL(63),
+};
+DECLARE_ADRENO_REGLIST_LIST(a810_ifpc_reglist);
+
+static const struct adreno_reglist_pipe a810_dyn_pwrup_reglist_regs[] = {
+	{ REG_A8XX_CP_PROTECT_CNTL_PIPE, 0, BIT(PIPE_BR) | BIT(PIPE_BV) },
+	{ REG_A8XX_CP_PROTECT_PIPE(15), 0, BIT(PIPE_BR) | BIT(PIPE_BV) },
+	{ REG_A8XX_GRAS_TSEFE_DBG_ECO_CNTL, 0, BIT(PIPE_BV) | BIT(PIPE_BR) },
+	{ REG_A8XX_GRAS_NC_MODE_CNTL, 0, BIT(PIPE_BV) | BIT(PIPE_BR) },
+	{ REG_A8XX_GRAS_DBG_ECO_CNTL, 0, BIT(PIPE_BV) | BIT(PIPE_BR) },
+	{ REG_A7XX_RB_CCU_CNTL, 0, BIT(PIPE_BR) },
+	{ REG_A8XX_RB_CCU_NC_MODE_CNTL, 0, BIT(PIPE_BR) },
+	{ REG_A8XX_RB_CMP_NC_MODE_CNTL, 0, BIT(PIPE_BR) },
+	{ REG_A8XX_RB_RESOLVE_PREFETCH_CNTL, 0, BIT(PIPE_BR) },
+	{ REG_A8XX_RB_CMP_DBG_ECO_CNTL, 0, BIT(PIPE_BR) },
+	{ REG_A8XX_RB_GC_GMEM_PROTECT, 0, BIT(PIPE_BR) },
+	{ REG_A6XX_RB_CONTEXT_SWITCH_GMEM_SAVE_RESTORE_ENABLE, 0, BIT(PIPE_BR) },
+	{ REG_A8XX_VPC_FLATSHADE_MODE_CNTL, 0, BIT(PIPE_BV) | BIT(PIPE_BR) },
+	{ REG_A8XX_PC_CHICKEN_BITS_1, 0, BIT(PIPE_BV) | BIT(PIPE_BR) },
+	{ REG_A8XX_PC_CHICKEN_BITS_2, 0, BIT(PIPE_BV) | BIT(PIPE_BR) },
+	{ REG_A8XX_PC_CHICKEN_BITS_3, 0, BIT(PIPE_BV) | BIT(PIPE_BR) },
+	{ REG_A8XX_PC_CHICKEN_BITS_4, 0, BIT(PIPE_BV) | BIT(PIPE_BR) },
+	{ REG_A6XX_PC_AUTO_VERTEX_STRIDE, 0, BIT(PIPE_BR) | BIT(PIPE_BV) },
+	{ REG_A8XX_PC_VIS_STREAM_CNTL, 0, BIT(PIPE_BR) | BIT(PIPE_BV) },
+	{ REG_A8XX_PC_CONTEXT_SWITCH_STABILIZE_CNTL_1, 0, BIT(PIPE_BR) | BIT(PIPE_BV) },
+	{ REG_A8XX_VFD_CB_BV_THRESHOLD, 0, BIT(PIPE_BV) | BIT(PIPE_BR) },
+	{ REG_A8XX_VFD_CB_BR_THRESHOLD, 0, BIT(PIPE_BV) | BIT(PIPE_BR) },
+	{ REG_A8XX_VFD_CB_BUSY_REQ_CNT, 0, BIT(PIPE_BV) | BIT(PIPE_BR) },
+	{ REG_A8XX_VFD_CB_LP_REQ_CNT, 0, BIT(PIPE_BV) | BIT(PIPE_BR) },
+	{ REG_A7XX_VFD_DBG_ECO_CNTL, 0, BIT(PIPE_BR) | BIT(PIPE_BV) },
+};
+DECLARE_ADRENO_REGLIST_PIPE_LIST(a810_dyn_pwrup_reglist);
+
 static const struct adreno_reglist_pipe a840_nonctxt_regs[] = {
 	{ REG_A8XX_CP_SMMU_STREAM_ID_LPAC, 0x00000101, BIT(PIPE_NONE) },
 	{ REG_A8XX_GRAS_DBG_ECO_CNTL, 0x00000800, BIT(PIPE_BV) | BIT(PIPE_BR) },
@@ -2193,6 +2446,43 @@ static const struct adreno_info a8xx_gpus[] = {
 			{ 252, 2 },
 			{ 221, 3 },
 		),
+	}, {
+		.chip_ids = ADRENO_CHIP_IDS(0x44010000),
+		.family = ADRENO_8XX_GEN1,
+		.fw = {
+			[ADRENO_FW_SQE] = "gen80300_sqe.fw",
+			[ADRENO_FW_GMU] = "gen80300_gmu.bin",
+		},
+		.gmem = SZ_512K + SZ_64K,
+		.inactive_period = DRM_MSM_INACTIVE_PERIOD,
+		.quirks = ADRENO_QUIRK_HAS_CACHED_COHERENT |
+			  ADRENO_QUIRK_HAS_HW_APRIV |
+			  ADRENO_QUIRK_PREEMPTION |
+			  ADRENO_QUIRK_IFPC,
+		.funcs = &a8xx_gpu_funcs,
+		.zapfw = "gen80300_zap.mbn",
+		.a6xx = &(const struct a6xx_info) {
+			.protect = &a810_protect,
+			.nonctxt_reglist = a810_nonctxt_regs,
+			.pwrup_reglist = &a810_pwrup_reglist,
+			.dyn_pwrup_reglist = &a810_dyn_pwrup_reglist,
+			.ifpc_reglist = &a810_ifpc_reglist,
+			.gbif_cx = a840_gbif,
+			.max_slices = 1,
+			.gmu_chipid = 0x8030000,
+			.bcms = (const struct a6xx_bcm[]) {
+				{ .name = "SH0", .buswidth = 16 },
+				{ .name = "MC0", .buswidth = 4 },
+				{
+					.name = "ACV",
+					.fixed = true,
+					.perfmode = BIT(2),
+					.perfmode_bw = 10687500,
+				},
+				{ /* sentinel */ },
+			},
+		},
+		.preempt_record_size = 4558 * SZ_1K,
 	}
 };
 
@@ -2205,4 +2495,5 @@ static inline __always_unused void __build_asserts(void)
 	BUILD_BUG_ON(a660_protect.count > a660_protect.count_max);
 	BUILD_BUG_ON(a690_protect.count > a690_protect.count_max);
 	BUILD_BUG_ON(a730_protect.count > a730_protect.count_max);
+	BUILD_BUG_ON(a810_protect.count > a810_protect.count_max);
 }
diff --git a/drivers/gpu/drm/msm/adreno/adreno_gpu.h b/drivers/gpu/drm/msm/adreno/adreno_gpu.h
index c0ee544ce257..d474d88b9152 100644
--- a/drivers/gpu/drm/msm/adreno/adreno_gpu.h
+++ b/drivers/gpu/drm/msm/adreno/adreno_gpu.h
@@ -596,6 +596,11 @@ static inline int adreno_is_a8xx(struct adreno_gpu *gpu)
 	return gpu->info->family >= ADRENO_8XX_GEN1;
 }
 
+static inline int adreno_is_a810(struct adreno_gpu *gpu)
+{
+	return gpu->info->chip_ids[0] == 0x44010000;
+}
+
 static inline int adreno_is_x285(struct adreno_gpu *gpu)
 {
 	return gpu->info->chip_ids[0] == 0x44070001;

-- 
2.53.0

Claude review: drm/msm/adreno: add Adreno 810 GPU support

[Claude Code Review Bot]

Patch Review

This is the main catalog entry patch. Several sub-items:

**adreno_is_a810():** Added in `adreno_gpu.h` but not used anywhere in this series. If no callers are needed yet, consider deferring this addition until it's actually needed to avoid dead code. The pattern does match existing `adreno_is_x285()` and `adreno_is_a840()` helpers though, so adding it preemptively is consistent.

**a810_nonctxt_regs:** The register values look reasonable. A couple of notes:
- `REG_A8XX_GRAS_DBG_ECO_CNTL` is set to `0x00f80800` for A810 vs `0x00000800` for A840 — the extra bits (0x00f80000) are presumably A810-specific hardware requirements.
- `REG_A7XX_SP_CHICKEN_BITS_2` uses `BIT(22) | BIT(23)` — using BIT macros here is cleaner than a hex value. Good.

**a810_protect_regs:** 47 entries with `count_max = 64`. The BUILD_BUG_ON in `__build_asserts` correctly validates 47 <= 64.

**a810_pwrup_reglist_regs:** Includes the UCHE_CCHE_TRAP_BASE and UCHE_CCHE_WRITE_THRU_BASE entries mentioned in the v3 changelog. The `+ 1` pattern for 64-bit register pairs (LO/HI) is standard:

```c
REG_A8XX_UCHE_CCHE_TRAP_BASE,
REG_A8XX_UCHE_CCHE_TRAP_BASE + 1,
```

**a810_ifpc_reglist_regs:** Includes all 64 CP_PROTECT_GLOBAL slots (0-63) as noted in the v3 changelog.

**a810_dyn_pwrup_reglist_regs:** References `REG_A8XX_CP_PROTECT_PIPE(15)`, consistent with the hardcoded slot 15 from patch 5.

**Catalog entry:**
- `family = ADRENO_8XX_GEN1` — makes sense, A810 is gen1 A8xx.
- Firmware names `gen80300_*` match GMU chipid `0x8030000`.
- `gmem = SZ_512K + SZ_64K` (576K) — smaller than A840's GMEM, appropriate for the lower-tier part.
- `max_slices = 1` — appropriate for a single-slice GPU.
- `preempt_record_size = 4558 * SZ_1K` — much smaller than A840's 19708K, consistent with 1 slice vs 3.
- Quirks include `ADRENO_QUIRK_IFPC` but not `ADRENO_QUIRK_SOFTFUSE`, matching A840.
- BCM entries (SH0/16, MC0/4, ACV) look standard.

**Minor:** The entry is placed after the A840 entry but the family is `ADRENO_8XX_GEN1` while A840 is `ADRENO_8XX_GEN2`. It might be more natural to place A810 (GEN1) before the GEN2 entries, but this is cosmetic and shouldn't affect functionality since lookups are by chip_id.

---
Generated by Claude Code Patch Reviewer

[PATCH RFC v3 7/7] arm64: dts: qcom: milos: Add Adreno 810 GPU and GMU nodes

[Alexander Koskovich]

Add GPU and GMU devicetree nodes for the Adreno 810 GPU found on
Qualcomm SM7635 (Milos) based devices.

The qcom,kaanapali-gxclkctl.h header can be reused here because
Milos uses the same driver and the GX_CLKCTL_GX_GDSC definition
is identical.

Reviewed-by: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>
Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Signed-off-by: Alexander Koskovich <akoskovich@pm.me>
---
 arch/arm64/boot/dts/qcom/milos.dtsi | 146 ++++++++++++++++++++++++++++++++++++
 1 file changed, 146 insertions(+)

diff --git a/arch/arm64/boot/dts/qcom/milos.dtsi b/arch/arm64/boot/dts/qcom/milos.dtsi
index 621f05820826..0ffd77048659 100644
--- a/arch/arm64/boot/dts/qcom/milos.dtsi
+++ b/arch/arm64/boot/dts/qcom/milos.dtsi
@@ -3,6 +3,7 @@
  * Copyright (c) 2025, Luca Weiss <luca.weiss@fairphone.com>
  */
 
+#include <dt-bindings/clock/qcom,kaanapali-gxclkctl.h>
 #include <dt-bindings/clock/qcom,milos-camcc.h>
 #include <dt-bindings/clock/qcom,milos-dispcc.h>
 #include <dt-bindings/clock/qcom,milos-gcc.h>
@@ -1224,6 +1225,151 @@ lpass_ag_noc: interconnect@3c40000 {
 			qcom,bcm-voters = <&apps_bcm_voter>;
 		};
 
+		gpu: gpu@3d00000 {
+			compatible = "qcom,adreno-44010000", "qcom,adreno";
+			reg = <0x0 0x03d00000 0x0 0x40000>,
+			      <0x0 0x03d9e000 0x0 0x2000>,
+			      <0x0 0x03d61000 0x0 0x800>;
+			reg-names = "kgsl_3d0_reg_memory",
+				    "cx_mem",
+				    "cx_dbgc";
+
+			interrupts = <GIC_SPI 300 IRQ_TYPE_LEVEL_HIGH 0>;
+
+			iommus = <&adreno_smmu 0 0x0>;
+
+			operating-points-v2 = <&gpu_opp_table>;
+
+			qcom,gmu = <&gmu>;
+			#cooling-cells = <2>;
+
+			interconnects = <&gem_noc MASTER_GFX3D QCOM_ICC_TAG_ALWAYS
+					 &mc_virt SLAVE_EBI1 QCOM_ICC_TAG_ALWAYS>;
+			interconnect-names = "gfx-mem";
+
+			status = "disabled";
+
+			gpu_zap_shader: zap-shader {
+				memory-region = <&gpu_microcode_mem>;
+			};
+
+			gpu_opp_table: opp-table {
+				compatible = "operating-points-v2-adreno",
+					     "operating-points-v2";
+
+				opp-264000000 {
+					opp-hz = /bits/ 64 <264000000>;
+					opp-level = <RPMH_REGULATOR_LEVEL_LOW_SVS_D1>;
+					opp-peak-kBps = <2136718>;
+					qcom,opp-acd-level = <0xc8295ffd>;
+				};
+
+				opp-362000000 {
+					opp-hz = /bits/ 64 <362000000>;
+					opp-level = <RPMH_REGULATOR_LEVEL_LOW_SVS>;
+					opp-peak-kBps = <2136718>;
+					qcom,opp-acd-level = <0xc02c5ffd>;
+				};
+
+				opp-510000000 {
+					opp-hz = /bits/ 64 <510000000>;
+					opp-level = <RPMH_REGULATOR_LEVEL_SVS>;
+					opp-peak-kBps = <3972656>;
+					qcom,opp-acd-level = <0x882b5ffd>;
+				};
+
+				opp-644000000 {
+					opp-hz = /bits/ 64 <644000000>;
+					opp-level = <RPMH_REGULATOR_LEVEL_SVS_L1>;
+					opp-peak-kBps = <5285156>;
+					qcom,opp-acd-level = <0x882a5ffd>;
+				};
+
+				opp-688000000 {
+					opp-hz = /bits/ 64 <688000000>;
+					opp-level = <RPMH_REGULATOR_LEVEL_SVS_L2>;
+					opp-peak-kBps = <6074218>;
+					qcom,opp-acd-level = <0x882a5ffd>;
+				};
+
+				opp-763000000 {
+					opp-hz = /bits/ 64 <763000000>;
+					opp-level = <RPMH_REGULATOR_LEVEL_NOM>;
+					opp-peak-kBps = <6671875>;
+					qcom,opp-acd-level = <0xa8295ffd>;
+				};
+
+				opp-895000000 {
+					opp-hz = /bits/ 64 <895000000>;
+					opp-level = <RPMH_REGULATOR_LEVEL_NOM_L1>;
+					opp-peak-kBps = <8171875>;
+					qcom,opp-acd-level = <0x88295ffd>;
+				};
+
+				opp-960000000 {
+					opp-hz = /bits/ 64 <960000000>;
+					opp-level = <RPMH_REGULATOR_LEVEL_TURBO>;
+					opp-peak-kBps = <8171875>;
+					qcom,opp-acd-level = <0xa8285ffd>;
+				};
+
+				opp-1050000000 {
+					opp-hz = /bits/ 64 <1050000000>;
+					opp-level = <RPMH_REGULATOR_LEVEL_TURBO_L1>;
+					opp-peak-kBps = <18597656>;
+					qcom,opp-acd-level = <0x88285ffd>;
+				};
+			};
+		};
+
+		gmu: gmu@3d37000 {
+			compatible = "qcom,adreno-gmu-810.0", "qcom,adreno-gmu";
+			reg = <0x0 0x03d37000 0x0 0x68000>;
+			reg-names = "gmu";
+
+			interrupts = <GIC_SPI 304 IRQ_TYPE_LEVEL_HIGH 0>,
+				     <GIC_SPI 305 IRQ_TYPE_LEVEL_HIGH 0>;
+			interrupt-names = "hfi", "gmu";
+
+			clocks = <&gpucc GPU_CC_AHB_CLK>,
+				 <&gpucc GPU_CC_CX_GMU_CLK>,
+				 <&gpucc GPU_CC_CXO_CLK>,
+				 <&gcc GCC_DDRSS_GPU_AXI_CLK>,
+				 <&gcc GCC_GPU_MEMNOC_GFX_CLK>,
+				 <&gpucc GPU_CC_HUB_CX_INT_CLK>;
+			clock-names = "ahb",
+				      "gmu",
+				      "cxo",
+				      "axi",
+				      "memnoc",
+				      "hub";
+
+			power-domains = <&gpucc GPU_CC_CX_GDSC>,
+					<&gxclkctl GX_CLKCTL_GX_GDSC>;
+			power-domain-names = "cx",
+					     "gx";
+
+			iommus = <&adreno_smmu 5 0x0>;
+
+			qcom,qmp = <&aoss_qmp>;
+
+			operating-points-v2 = <&gmu_opp_table>;
+
+			gmu_opp_table: opp-table {
+				compatible = "operating-points-v2";
+
+				opp-350000000 {
+					opp-hz = /bits/ 64 <350000000>;
+					opp-level = <RPMH_REGULATOR_LEVEL_LOW_SVS>;
+				};
+
+				opp-650000000 {
+					opp-hz = /bits/ 64 <650000000>;
+					opp-level = <RPMH_REGULATOR_LEVEL_SVS>;
+				};
+			};
+		};
+
 		gxclkctl: clock-controller@3d64000 {
 			compatible = "qcom,milos-gxclkctl";
 			reg = <0x0 0x03d64000 0x0 0x6000>;

-- 
2.53.0

Claude review: arm64: dts: qcom: milos: Add Adreno 810 GPU and GMU nodes

[Claude Code Review Bot]

Patch Review

DTS addition for the Milos SoC. A few observations:

**GPU node:** Three reg regions (kgsl_3d0_reg_memory, cx_mem, cx_dbgc) with standard names. The OPP table ranges from 264MHz to 1050MHz, matching the cover letter's note about GPU_CC_FREQ_LIMIT_VAL limiting to 1050MHz.

**GMU node:** Uses `qcom,adreno-gmu-810.0` compatible with 6 clocks matching the binding from patch 1. Power domains include both CX (from gpucc) and GX (from gxclkctl using the kaanapali header, explained in the commit message as having identical definitions).

**GPU OPP table:**
- Uses `qcom,opp-acd-level` properties for ACD (Adaptive Clock Distribution).
- Has `operating-points-v2-adreno` compatible.
- 9 OPP entries spanning LOW_SVS_D1 through TURBO_L1.

**Status:** `status = "disabled"` on the GPU node is correct — it should be enabled per-board.

Already has Reviewed-by from Konrad Dybcio and Dmitry Baryshkov. The DTS looks reasonable and complete.

---
Generated by Claude Code Patch Reviewer

Claude review: Add support for Adreno 810 GPU

[Claude Code Review Bot]

Overall Series Review

Subject: Add support for Adreno 810 GPU
Author: Alexander Koskovich <akoskovich@pm.me>
Patches: 9
Reviewed: 2026-04-12T13:17:00.559830

---

This 7-patch RFC series adds support for the Adreno 810 GPU (chip ID 0x44010000) found in Qualcomm SM7635 (Milos). The series is well-structured: patches 1-2 add DT bindings, patches 3-4 refactor cx_misc_mmio handling to decouple it from LLCC (needed because Milos has no LLCC), patch 5 fixes the pipe protect slot addressing to accommodate A810's different protect count_max, patch 6 adds the main catalog entry, and patch 7 adds the DTS nodes.

The series is logically ordered and the prerequisite patches (3-5) are correctly sequenced before the GPU support patch (6). The RFC status is appropriate given the external dependencies on GX GDSC and GXCLKCTL series.

**One significant concern:** Patch 5 changes the meaning of `count_max` for A8xx GPUs. Before patch 5, `count_max` was used as the pipe protect slot index (always 15). After patch 5, it's no longer used for the pipe slot. The A810 entry then sets `count_max = 64` (the global protect slot maximum), which is semantically different from the existing x285/a840 `count_max = 15`. The x285 and a840 entries should probably be updated to `count_max = 64` as well for consistency, and their BUILD_BUG_ON checks should be added to `__build_asserts`.

Overall the series looks reasonable for an RFC. A few minor points noted per-patch below.

---
Generated by Claude Code Patch Reviewer