[PATCH v7 0/8] media: qcom: venus: add MSM8939 support

[PATCH v7 0/8] media: qcom: venus: add MSM8939 support

[Erikas Bitovtas]

This patch series adds support for Venus on MSM8939. It is mostly
similar to MSM8916 Venus, except it needs two additional cores to be
powered on before it can start decoding.

As per Dmitry's request, I am attaching Fluster results and
v4l2-compliance output. Fluster results were very inconsistent and
caused power collapse fails.
H.264: https://pastebin.com/C15qeq5W
H.265 (HEVC): https://pastebin.com/WDsnxvuk
VP8: https://pastebin.com/egAgEm15
v4l2-compliance: https://pastebin.com/VpBhEFc1
Power collapse fail log: https://pastebin.com/rTivMcpK

Signed-off-by: Erikas Bitovtas <xerikasxx@gmail.com>
---
Changes in v7:
- Moved codec blacklist mechanism to be applied before MSM8939 struct is
  added.
- Vcodec power domains are attached with a managed method now.
- Removed duplicate clock rates in msm8939_freq_tbl.
- Redid the merge of vcodec_clks done in v5.
- Removed the {vdec,venc}_{get,power}() methods and moved them to their
  respective core_* parts. Vcodec cores are now powered up along with
  the core.
- Switched HW_CTRL to HW_CTRL_TRIGGER in Venus core GDSCs. They are now
  powered up first, then their control is passed to hardware.
- For now, document just the change of halt check in the commit message
  of setting GDSCs to hardware controlled. I'm not sure what happens
  exactly, but if they are set to BRANCH_HALT, probing the driver fails:
  https://pastebin.com/EYQekR1p
- Given the power management behavior has changed substantially in the
  power domain enable logic patch, I removed Bryan's "Reviewed-by: "
  tag.
- Link to v6: https://patch.msgid.link/20260510-msm8939-venus-rfc-v6-0-e69465375900@gmail.com

Changes in v6:
- Added a more generic mechanism on blacklisting codecs.
- Reworded a commit for setting Venus core GDSCs to hardware controlled.
- Reworded a commit for adding HEVC decoding to MSM8939.
- Link to v5: https://patch.msgid.link/20260507-msm8939-venus-rfc-v5-0-d7b5ea2ce591@gmail.com

Changes in v5:
- Undid the merge of vcodec_clks, they are split back to
  vcodec{0,1}_clks now.
- Extracted a dev_pm_domain_attach_list() call into a function of its
  own to reduce nesting.
- Added missing "power-domain-names" required property to device tree
  binding.
- Renamed vcodec clocks and power domains to match other Venus bindings.
- Reordered commits and grouped them by subsystems. Now first come DTB
  patches, then clock, then media.
- Removed "status = "disabled"" in the device tree example.
- Link to v4: https://patch.msgid.link/20260506-msm8939-venus-rfc-v4-0-994f5eb22acb@gmail.com

Changes in v4:
- Removed vcodec{0,1}_pmdomains and merged vcodec{0,1}_clks into
  vcodec_clks instead for MSM8939.
- Inlined video decoder and encoder device tree nodes in the driver and
  removed them from the binding.
- Kept vdec and venc methods for HFI v3 separate from HFI v1.
- {vdec,venc}_get() are now called as early as before, since it is no
  longer needed for us to attach power domains to dev_dec and dev_enc.
- Link to v3: https://patch.msgid.link/20260427-msm8939-venus-rfc-v3-0-288195bb7917@gmail.com

Changes in v3:
- Added missing vcodec1_clks to resource struct.
- Removed enc_nodename from resource struct since we include
  video-decoder now.
- Link to v2: https://patch.msgid.link/20260425-msm8939-venus-rfc-v2-0-f69e52b36207@gmail.com

Changes in v2:
- Enabled GDSCs during encoding as well.
- Merged vcodec{0,1}_pmdomains_num into vcodec_pmdomains_num.
- Reworded commit for marking GDSCs as hardware controlled. Same
  situation as in cdc59600bccf ("clk: qcom: gcc-msm8953: fix stuck venus0_core0 clock")
- Clarified the reason for missing firmware-name property in device
  tree.
- Clarified the reason for moving vdec_get and venc_get for later.
- Link to v1: https://patch.msgid.link/20260416-msm8939-venus-rfc-v1-0-a09fcf2c23df@gmail.com

To: Bryan O'Donoghue <bod@kernel.org>
To: Vikash Garodia <vikash.garodia@oss.qualcomm.com>
To: Dikshita Agarwal <dikshita.agarwal@oss.qualcomm.com>
To: Mauro Carvalho Chehab <mchehab@kernel.org>
To: Rob Herring <robh@kernel.org>
To: Krzysztof Kozlowski <krzk+dt@kernel.org>
To: Conor Dooley <conor+dt@kernel.org>
To: André Apitzsch <git@apitzsch.eu>
To: Erikas Bitovtas <xerikasxx@gmail.com>
To: Bjorn Andersson <andersson@kernel.org>
To: Konrad Dybcio <konradybcio@kernel.org>
To: Michael Turquette <mturquette@baylibre.com>
To: Stephen Boyd <sboyd@kernel.org>
To: Brian Masney <bmasney@redhat.com>
Cc: linux-arm-msm@vger.kernel.org
Cc: linux-media@vger.kernel.org
Cc: devicetree@vger.kernel.org
Cc: linux-kernel@vger.kernel.org
Cc: linux-clk@vger.kernel.org

---
André Apitzsch (4):
      media: dt-bindings: venus: Add qcom,msm8939 schema
      arm64: dts: qcom: msm8939: Add venus node
      arm64: dts: qcom: msm8939-longcheer-l9100: Enable venus node
      media: qcom: venus: Add msm8939 resource struct

Erikas Bitovtas (4):
      arm64: dts: qcom: msm8939-asus-z00t: add Venus
      clk: qcom: gcc-msm8939: mark Venus core GDSCs as hardware controlled
      media: qcom: venus: add power domain enable logic for Venus cores
      media: qcom: venus: add codec blacklist mechanism

 .../bindings/media/qcom,msm8939-venus.yaml         |  79 ++++++++++++++
 arch/arm64/boot/dts/qcom/msm8939-asus-z00t.dts     |   8 ++
 .../boot/dts/qcom/msm8939-longcheer-l9100.dts      |   8 ++
 arch/arm64/boot/dts/qcom/msm8939.dtsi              |  23 ++++
 drivers/clk/qcom/gcc-msm8939.c                     |   4 +
 drivers/media/platform/qcom/venus/core.c           |  44 ++++++++
 drivers/media/platform/qcom/venus/core.h           |   5 +
 drivers/media/platform/qcom/venus/hfi_parser.c     |  10 +-
 drivers/media/platform/qcom/venus/pm_helpers.c     | 118 ++++++++++++++++++++-
 9 files changed, 289 insertions(+), 10 deletions(-)
---
base-commit: e98d21c170b01ddef366f023bbfcf6b31509fa83
change-id: 20260416-msm8939-venus-rfc-c025c4c74fae

Best regards,
--  
Erikas Bitovtas <xerikasxx@gmail.com>

[PATCH v7 1/8] media: dt-bindings: venus: Add qcom,msm8939 schema

[Erikas Bitovtas]

From: André Apitzsch <git@apitzsch.eu>

Add a schema description for the Venus video decoder/encoder IP in
MSM8939.

Signed-off-by: André Apitzsch <git@apitzsch.eu>
Reviewed-by: Bryan O'Donoghue <bod@kernel.org>
Signed-off-by: Erikas Bitovtas <xerikasxx@gmail.com>
---
 .../bindings/media/qcom,msm8939-venus.yaml         | 79 ++++++++++++++++++++++
 1 file changed, 79 insertions(+)

diff --git a/Documentation/devicetree/bindings/media/qcom,msm8939-venus.yaml b/Documentation/devicetree/bindings/media/qcom,msm8939-venus.yaml
new file mode 100644
index 000000000000..10a50a410748
--- /dev/null
+++ b/Documentation/devicetree/bindings/media/qcom,msm8939-venus.yaml
@@ -0,0 +1,79 @@
+# SPDX-License-Identifier: (GPL-2.0-only OR BSD-2-Clause)
+%YAML 1.2
+---
+$id: http://devicetree.org/schemas/media/qcom,msm8939-venus.yaml#
+$schema: http://devicetree.org/meta-schemas/core.yaml#
+
+title: Qualcomm MSM8939 Venus video encode and decode accelerators
+
+maintainers:
+  - André Apitzsch <git@apitzsch.eu>
+  - Erikas Bitovtas <xerikasxx@gmail.com>
+
+description:
+  The Venus IP is a video encode and decode accelerator present
+  on Qualcomm platforms
+
+allOf:
+  - $ref: qcom,venus-common.yaml#
+
+properties:
+  compatible:
+    const: qcom,msm8939-venus
+
+  power-domains:
+    maxItems: 3
+
+  power-domain-names:
+    items:
+      - const: venus
+      - const: vcodec0
+      - const: vcodec1
+
+  clocks:
+    maxItems: 5
+
+  clock-names:
+    items:
+      - const: core
+      - const: iface
+      - const: bus
+      - const: vcodec0_core
+      - const: vcodec1_core
+
+  iommus:
+    maxItems: 1
+
+required:
+  - compatible
+  - iommus
+  - power-domain-names
+
+unevaluatedProperties: false
+
+examples:
+  - |
+    #include <dt-bindings/interrupt-controller/arm-gic.h>
+    #include <dt-bindings/clock/qcom,gcc-msm8939.h>
+
+    video-codec@1d00000 {
+      compatible = "qcom,msm8939-venus";
+      reg = <0x01d00000 0xff000>;
+      interrupts = <GIC_SPI 44 IRQ_TYPE_LEVEL_HIGH>;
+      clocks = <&gcc GCC_VENUS0_VCODEC0_CLK>,
+               <&gcc GCC_VENUS0_AHB_CLK>,
+               <&gcc GCC_VENUS0_AXI_CLK>,
+               <&gcc GCC_VENUS0_CORE0_VCODEC0_CLK>,
+               <&gcc GCC_VENUS0_CORE1_VCODEC0_CLK>;
+      clock-names = "core",
+                    "iface",
+                    "bus",
+                    "vcodec0_core",
+                    "vcodec1_core";
+      power-domains = <&gcc VENUS_GDSC>,
+                      <&gcc VENUS_CORE0_GDSC>,
+                      <&gcc VENUS_CORE1_GDSC>;
+      power-domain-names = "venus", "vcodec0", "vcodec1";
+      iommus = <&apps_iommu 5>;
+      memory-region = <&venus_mem>;
+    };

-- 
2.54.0

Re: [PATCH v7 1/8] media: dt-bindings: venus: Add qcom,msm8939 schema

[Krzysztof Kozlowski]

On Thu, May 14, 2026 at 12:24:30AM +0300, Erikas Bitovtas wrote:
> From: André Apitzsch <git@apitzsch.eu>
> 
> Add a schema description for the Venus video decoder/encoder IP in
> MSM8939.
> 
> Signed-off-by: André Apitzsch <git@apitzsch.eu>
> Reviewed-by: Bryan O'Donoghue <bod@kernel.org>
> Signed-off-by: Erikas Bitovtas <xerikasxx@gmail.com>
> ---
>  .../bindings/media/qcom,msm8939-venus.yaml         | 79 ++++++++++++++++++++++
>  1 file changed, 79 insertions(+)

Reviewed-by: Krzysztof Kozlowski <krzysztof.kozlowski@oss.qualcomm.com>

Best regards,
Krzysztof

Re: [PATCH v7 1/8] media: dt-bindings: venus: Add qcom,msm8939 schema

[Bryan O'Donoghue]

On 13/05/2026 22:24, Erikas Bitovtas wrote:
> From: André Apitzsch <git@apitzsch.eu>
> 
> Add a schema description for the Venus video decoder/encoder IP in
> MSM8939.
> 
> Signed-off-by: André Apitzsch <git@apitzsch.eu>
> Reviewed-by: Bryan O'Donoghue <bod@kernel.org>
> Signed-off-by: Erikas Bitovtas <xerikasxx@gmail.com>
> ---
>   .../bindings/media/qcom,msm8939-venus.yaml         | 79 ++++++++++++++++++++++
>   1 file changed, 79 insertions(+)
> 
> diff --git a/Documentation/devicetree/bindings/media/qcom,msm8939-venus.yaml b/Documentation/devicetree/bindings/media/qcom,msm8939-venus.yaml
> new file mode 100644
> index 000000000000..10a50a410748
> --- /dev/null
> +++ b/Documentation/devicetree/bindings/media/qcom,msm8939-venus.yaml
> @@ -0,0 +1,79 @@
> +# SPDX-License-Identifier: (GPL-2.0-only OR BSD-2-Clause)
> +%YAML 1.2
> +---
> +$id: http://devicetree.org/schemas/media/qcom,msm8939-venus.yaml#
> +$schema: http://devicetree.org/meta-schemas/core.yaml#
> +
> +title: Qualcomm MSM8939 Venus video encode and decode accelerators
> +
> +maintainers:
> +  - André Apitzsch <git@apitzsch.eu>
> +  - Erikas Bitovtas <xerikasxx@gmail.com>
> +
> +description:
> +  The Venus IP is a video encode and decode accelerator present
> +  on Qualcomm platforms
> +
> +allOf:
> +  - $ref: qcom,venus-common.yaml#
> +
> +properties:
> +  compatible:
> +    const: qcom,msm8939-venus
> +
> +  power-domains:
> +    maxItems: 3
> +
> +  power-domain-names:
> +    items:
> +      - const: venus
> +      - const: vcodec0
> +      - const: vcodec1
> +
> +  clocks:
> +    maxItems: 5
> +
> +  clock-names:
> +    items:
> +      - const: core
> +      - const: iface
> +      - const: bus
> +      - const: vcodec0_core
> +      - const: vcodec1_core
> +
> +  iommus:
> +    maxItems: 1
> +
> +required:
> +  - compatible
> +  - iommus
> +  - power-domain-names
> +
> +unevaluatedProperties: false
> +
> +examples:
> +  - |
> +    #include <dt-bindings/interrupt-controller/arm-gic.h>
> +    #include <dt-bindings/clock/qcom,gcc-msm8939.h>
> +
> +    video-codec@1d00000 {
> +      compatible = "qcom,msm8939-venus";
> +      reg = <0x01d00000 0xff000>;
> +      interrupts = <GIC_SPI 44 IRQ_TYPE_LEVEL_HIGH>;
> +      clocks = <&gcc GCC_VENUS0_VCODEC0_CLK>,
> +               <&gcc GCC_VENUS0_AHB_CLK>,
> +               <&gcc GCC_VENUS0_AXI_CLK>,
> +               <&gcc GCC_VENUS0_CORE0_VCODEC0_CLK>,
> +               <&gcc GCC_VENUS0_CORE1_VCODEC0_CLK>;
> +      clock-names = "core",
> +                    "iface",
> +                    "bus",
> +                    "vcodec0_core",
> +                    "vcodec1_core";
> +      power-domains = <&gcc VENUS_GDSC>,
> +                      <&gcc VENUS_CORE0_GDSC>,
> +                      <&gcc VENUS_CORE1_GDSC>;
> +      power-domain-names = "venus", "vcodec0", "vcodec1";
> +      iommus = <&apps_iommu 5>;
> +      memory-region = <&venus_mem>;
> +    };
> 
> --
> 2.54.0
> 

Reviewed-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org>

[PATCH v7 2/8] arm64: dts: qcom: msm8939: Add venus node

[Erikas Bitovtas]

From: André Apitzsch <git@apitzsch.eu>

Add DT entries for the msm8939 venus encoder/decoder.

Signed-off-by: André Apitzsch <git@apitzsch.eu>
Reviewed-by: Bryan O'Donoghue <bod@kernel.org>
Signed-off-by: Erikas Bitovtas <xerikasxx@gmail.com>
---
 arch/arm64/boot/dts/qcom/msm8939.dtsi | 23 +++++++++++++++++++++++
 1 file changed, 23 insertions(+)

diff --git a/arch/arm64/boot/dts/qcom/msm8939.dtsi b/arch/arm64/boot/dts/qcom/msm8939.dtsi
index d4d7b0c9206c..71e22584a46b 100644
--- a/arch/arm64/boot/dts/qcom/msm8939.dtsi
+++ b/arch/arm64/boot/dts/qcom/msm8939.dtsi
@@ -1627,6 +1627,29 @@ opp-19200000 {
 			};
 		};
 
+		venus: video-codec@1d00000 {
+			compatible = "qcom,msm8939-venus";
+			reg = <0x01d00000 0xff000>;
+			interrupts = <GIC_SPI 44 IRQ_TYPE_LEVEL_HIGH>;
+			clocks = <&gcc GCC_VENUS0_VCODEC0_CLK>,
+				 <&gcc GCC_VENUS0_AHB_CLK>,
+				 <&gcc GCC_VENUS0_AXI_CLK>,
+				 <&gcc GCC_VENUS0_CORE0_VCODEC0_CLK>,
+				 <&gcc GCC_VENUS0_CORE1_VCODEC0_CLK>;
+			clock-names = "core",
+				      "iface",
+				      "bus",
+				      "vcodec0_core",
+				      "vcodec1_core";
+			power-domains = <&gcc VENUS_GDSC>,
+					<&gcc VENUS_CORE0_GDSC>,
+					<&gcc VENUS_CORE1_GDSC>;
+			power-domain-names = "venus", "vcodec0", "vcodec1";
+			iommus = <&apps_iommu 5>;
+			memory-region = <&venus_mem>;
+			status = "disabled";
+		};
+
 		apps_iommu: iommu@1ef0000 {
 			compatible = "qcom,msm8916-iommu", "qcom,msm-iommu-v1";
 			reg = <0x01ef0000 0x3000>;

-- 
2.54.0

[PATCH v7 3/8] arm64: dts: qcom: msm8939-longcheer-l9100: Enable venus node

[Erikas Bitovtas]

From: André Apitzsch <git@apitzsch.eu>

Enable the venus node so that the video encoder/decoder will start
working.

Default paths for firmware work on BQ Aquaris M5, which means
firmware-name property can be left out.

Signed-off-by: André Apitzsch <git@apitzsch.eu>
Reviewed-by: Bryan O'Donoghue <bod@kernel.org>
---
 arch/arm64/boot/dts/qcom/msm8939-longcheer-l9100.dts | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/arch/arm64/boot/dts/qcom/msm8939-longcheer-l9100.dts b/arch/arm64/boot/dts/qcom/msm8939-longcheer-l9100.dts
index 13422a19c26a..48514c3df718 100644
--- a/arch/arm64/boot/dts/qcom/msm8939-longcheer-l9100.dts
+++ b/arch/arm64/boot/dts/qcom/msm8939-longcheer-l9100.dts
@@ -314,6 +314,14 @@ &usb_hs_phy {
 	extcon = <&usb_id>;
 };
 
+&venus {
+	status = "okay";
+};
+
+&venus_mem {
+	status = "okay";
+};
+
 &wcnss {
 	status = "okay";
 };

-- 
2.54.0

[PATCH v7 4/8] arm64: dts: qcom: msm8939-asus-z00t: add Venus

[Erikas Bitovtas]

Enable Venus video encoder/decoder for Asus ZenFone 2 Laser/Selfie.

Default firmware paths work on Asus ZenFone 2 Laser/Selfie, which
means firmware-name property can be left out.

Reviewed-by: Bryan O'Donoghue <bod@kernel.org>
Signed-off-by: Erikas Bitovtas <xerikasxx@gmail.com>
---
 arch/arm64/boot/dts/qcom/msm8939-asus-z00t.dts | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/arch/arm64/boot/dts/qcom/msm8939-asus-z00t.dts b/arch/arm64/boot/dts/qcom/msm8939-asus-z00t.dts
index 90e966242720..231a3e9c1929 100644
--- a/arch/arm64/boot/dts/qcom/msm8939-asus-z00t.dts
+++ b/arch/arm64/boot/dts/qcom/msm8939-asus-z00t.dts
@@ -267,6 +267,14 @@ &usb_hs_phy {
 	extcon = <&usb_id>;
 };
 
+&venus {
+	status = "okay";
+};
+
+&venus_mem {
+	status = "okay";
+};
+
 &wcnss {
 	status = "okay";
 };

-- 
2.54.0

[PATCH v7 5/8] clk: qcom: gcc-msm8939: mark Venus core GDSCs as hardware controlled

[Erikas Bitovtas]

Since in downstream kernel VENUS_CORE0_GDSC and VENUS_CORE1_GDSC have a
device tree property "qcom,supports-hw-trigger", add a HW_CTRL_TRIGGER
flag to these GDSCs to indicate that their control can be passed to
hardware.

Signed-off-by: Erikas Bitovtas <xerikasxx@gmail.com>
---
 drivers/clk/qcom/gcc-msm8939.c | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/drivers/clk/qcom/gcc-msm8939.c b/drivers/clk/qcom/gcc-msm8939.c
index 45193b3d714b..243d31a32e92 100644
--- a/drivers/clk/qcom/gcc-msm8939.c
+++ b/drivers/clk/qcom/gcc-msm8939.c
@@ -3664,6 +3664,7 @@ static struct clk_branch gcc_venus0_vcodec0_clk = {
 
 static struct clk_branch gcc_venus0_core0_vcodec0_clk = {
 	.halt_reg = 0x4c02c,
+	.halt_check = BRANCH_HALT_SKIP,
 	.clkr = {
 		.enable_reg = 0x4c02c,
 		.enable_mask = BIT(0),
@@ -3681,6 +3682,7 @@ static struct clk_branch gcc_venus0_core0_vcodec0_clk = {
 
 static struct clk_branch gcc_venus0_core1_vcodec0_clk = {
 	.halt_reg = 0x4c034,
+	.halt_check = BRANCH_HALT_SKIP,
 	.clkr = {
 		.enable_reg = 0x4c034,
 		.enable_mask = BIT(0),
@@ -3753,6 +3755,7 @@ static struct gdsc venus_core0_gdsc = {
 	.pd = {
 		.name = "venus_core0",
 	},
+	.flags = HW_CTRL_TRIGGER,
 	.pwrsts = PWRSTS_OFF_ON,
 };
 
@@ -3761,6 +3764,7 @@ static struct gdsc venus_core1_gdsc = {
 	.pd = {
 		.name = "venus_core1",
 	},
+	.flags = HW_CTRL_TRIGGER,
 	.pwrsts = PWRSTS_OFF_ON,
 };
 

-- 
2.54.0

Re: [PATCH v7 5/8] clk: qcom: gcc-msm8939: mark Venus core GDSCs as hardware controlled

[sashiko-bot]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [Medium] The commit message fails to mention and explain the addition of `.halt_check = BRANCH_HALT_SKIP` to `gcc_venus0_core0_vcodec0_clk` and `gcc_venus0_core1_vcodec0_clk`.
--

commit 4d497cce125abb4eb5e06f88b3056c8f33e326e0
Author: Erikas Bitovtas <xerikasxx@gmail.com>

clk: qcom: gcc-msm8939: mark Venus core GDSCs as hardware controlled

This commit adds the HW_CTRL_TRIGGER flag to VENUS_CORE0_GDSC and
VENUS_CORE1_GDSC to indicate their control can be passed to hardware,
based on downstream device tree properties.

> diff --git a/drivers/clk/qcom/gcc-msm8939.c b/drivers/clk/qcom/gcc-msm8939.c
> index 45193b3d714ba..243d31a32e924 100644
> --- a/drivers/clk/qcom/gcc-msm8939.c
> +++ b/drivers/clk/qcom/gcc-msm8939.c
> @@ -3664,6 +3664,7 @@ static struct clk_branch gcc_venus0_vcodec0_clk = {
>  
>  static struct clk_branch gcc_venus0_core0_vcodec0_clk = {
>  	.halt_reg = 0x4c02c,
> +	.halt_check = BRANCH_HALT_SKIP,

This isn't a bug, but the commit message explains the addition of the
HW_CTRL_TRIGGER flag to the GDSCs without mentioning the modifications to the
associated clock branches.

While skipping the halt check is the correct approach for branch clocks tied
to hardware-triggered GDSCs (as software polling would time out while the
hardware keeps the clock gated), could the commit message be updated to also
document this change to .halt_check?

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260514-msm8939-venus-rfc-v7-0-33c6c6fb9285@gmail.com?part=5

Re: [PATCH v7 5/8] clk: qcom: gcc-msm8939: mark Venus core GDSCs as hardware controlled

[Bryan O'Donoghue]

On 13/05/2026 22:24, Erikas Bitovtas wrote:
> Since in downstream kernel VENUS_CORE0_GDSC and VENUS_CORE1_GDSC have a
> device tree property "qcom,supports-hw-trigger", add a HW_CTRL_TRIGGER
> flag to these GDSCs to indicate that their control can be passed to
> hardware.
> 
> Signed-off-by: Erikas Bitovtas<xerikasxx@gmail.com>
> ---

I'm going to push back on this wording.

Take ownership of the proposed solution. "Do what downstream does" is 
cargo-cultism. You have spotted a flag in downstream and tested your 
code with/without the change, clearly demonstrating that HW_CTRL_TRIGGER 
is required.

Downstream is irrelevant. Your testing and imprimatur are all that is 
relevant to the commit log.

Please fix.

---
bod

[PATCH v7 6/8] media: qcom: venus: add power domain enable logic for Venus cores

[Erikas Bitovtas]

Attach power domains for vdec and venc cores and power them up if a vdec
or venc session is started.

Vcodev clocks are added and enabled to the core Venus device both for
vcodec0 and vcodec1. To ensure they are added only once, introduce a new
property "vcodec_clks", which is an array of clocks which are enabled
both during decode and encode and is retrieved from the device tree only
once.

Signed-off-by: Erikas Bitovtas <xerikasxx@gmail.com>
---
 drivers/media/platform/qcom/venus/core.h       |   3 +
 drivers/media/platform/qcom/venus/pm_helpers.c | 118 +++++++++++++++++++++++--
 2 files changed, 116 insertions(+), 5 deletions(-)

diff --git a/drivers/media/platform/qcom/venus/core.h b/drivers/media/platform/qcom/venus/core.h
index 7506f5d0f609..ff9174a4e8c1 100644
--- a/drivers/media/platform/qcom/venus/core.h
+++ b/drivers/media/platform/qcom/venus/core.h
@@ -77,6 +77,7 @@ struct venus_resources {
 	const struct hfi_ubwc_config *ubwc_conf;
 	const char * const clks[VIDC_CLKS_NUM_MAX];
 	unsigned int clks_num;
+	const char * const vcodec_clks[VIDC_VCODEC_CLKS_NUM_MAX];
 	const char * const vcodec0_clks[VIDC_VCODEC_CLKS_NUM_MAX];
 	const char * const vcodec1_clks[VIDC_VCODEC_CLKS_NUM_MAX];
 	unsigned int vcodec_clks_num;
@@ -140,6 +141,7 @@ struct venus_format {
  * @aon_base:	AON base address
  * @irq:		Venus irq
  * @clks:	an array of struct clk pointers
+ * @vcodec_clks: an array of vcodec struct clk pointers
  * @vcodec0_clks: an array of vcodec0 struct clk pointers
  * @vcodec1_clks: an array of vcodec1 struct clk pointers
  * @video_path: an interconnect handle to video to/from memory path
@@ -194,6 +196,7 @@ struct venus_core {
 	void __iomem *aon_base;
 	int irq;
 	struct clk *clks[VIDC_CLKS_NUM_MAX];
+	struct clk *vcodec_clks[VIDC_VCODEC_CLKS_NUM_MAX];
 	struct clk *vcodec0_clks[VIDC_VCODEC_CLKS_NUM_MAX];
 	struct clk *vcodec1_clks[VIDC_VCODEC_CLKS_NUM_MAX];
 	struct icc_path *video_path;
diff --git a/drivers/media/platform/qcom/venus/pm_helpers.c b/drivers/media/platform/qcom/venus/pm_helpers.c
index f0269524ac70..be705d4d64b5 100644
--- a/drivers/media/platform/qcom/venus/pm_helpers.c
+++ b/drivers/media/platform/qcom/venus/pm_helpers.c
@@ -89,12 +89,23 @@ static void core_clks_disable(struct venus_core *core)
 
 static int core_clks_set_rate(struct venus_core *core, unsigned long freq)
 {
-	int ret;
+	const struct venus_resources *res = core->res;
+	int ret, i;
 
 	ret = dev_pm_opp_set_rate(core->dev, freq);
 	if (ret)
 		return ret;
 
+	if (!res->vcodec_clks_num)
+		goto set_rates;
+
+	for (i = 0; i < res->vcodec_clks_num; i++) {
+		ret = clk_set_rate(core->vcodec_clks[i], freq);
+		if (ret)
+			return ret;
+	}
+
+set_rates:
 	ret = clk_set_rate(core->vcodec0_clks[0], freq);
 	if (ret)
 		return ret;
@@ -297,10 +308,33 @@ static int load_scale_v1(struct venus_inst *inst)
 	return ret;
 }
 
+static int vcodec_domains_get_v1(struct venus_core *core)
+{
+	struct device *dev = core->dev;
+	const struct venus_resources *res = core->res;
+	const struct dev_pm_domain_attach_data vcodec_data = {
+		.pd_names = res->vcodec_pmdomains,
+		.num_pd_names = res->vcodec_pmdomains_num,
+		.pd_flags = PD_FLAG_NO_DEV_LINK,
+	};
+
+	if (!res->vcodec_pmdomains)
+		return 0;
+
+	return devm_pm_domain_attach_list(dev, &vcodec_data,
+					  &core->pmdomains);
+}
+
 static int core_get_v1(struct venus_core *core)
 {
+	const struct venus_resources *res = core->res;
+	struct device *dev = core->dev;
 	int ret;
 
+	ret = vcodec_domains_get_v1(core);
+	if (ret < 0)
+		return ret;
+
 	ret = core_clks_get(core);
 	if (ret)
 		return ret;
@@ -309,9 +343,63 @@ static int core_get_v1(struct venus_core *core)
 	if (ret)
 		return ret;
 
+	if (!res->vcodec_pmdomains)
+		return 0;
+
+	ret = vcodec_clks_get(core, dev, core->vcodec_clks,
+			      res->vcodec_clks);
+	if (ret)
+		return ret;
+
 	return 0;
 }
 
+static int vcodec_domains_enable(struct venus_core *core)
+{
+	const struct venus_resources *res = core->res;
+	struct device *pd_dev;
+	int i = 0, ret;
+
+	if (!res->vcodec_pmdomains)
+		return 0;
+
+	for (; i < res->vcodec_pmdomains_num; i++) {
+		pd_dev = core->pmdomains->pd_devs[i];
+		ret = pm_runtime_resume_and_get(pd_dev);
+		if (ret)
+			goto err;
+
+		ret = dev_pm_genpd_set_hwmode(pd_dev, true);
+		if (ret && ret != -EOPNOTSUPP)
+			goto err;
+	}
+
+	return 0;
+err:
+	while (i--) {
+		pd_dev = core->pmdomains->pd_devs[i];
+		dev_pm_genpd_set_hwmode(pd_dev, false);
+		pm_runtime_put_sync(pd_dev);
+	}
+	return ret;
+}
+
+static void vcodec_domains_disable(struct venus_core *core)
+{
+	const struct venus_resources *res = core->res;
+	struct device *pd_dev;
+	int i = res->vcodec_pmdomains_num;
+
+	if (!res->vcodec_pmdomains)
+		return;
+
+	while (i--) {
+		pd_dev = core->pmdomains->pd_devs[i];
+		dev_pm_genpd_set_hwmode(pd_dev, false);
+		pm_runtime_put_sync(pd_dev);
+	}
+}
+
 static void core_put_v1(struct venus_core *core)
 {
 }
@@ -320,11 +408,31 @@ static int core_power_v1(struct venus_core *core, int on)
 {
 	int ret = 0;
 
-	if (on == POWER_ON)
+	if (on == POWER_ON) {
+		ret = vcodec_domains_enable(core);
+		if (ret)
+			return ret;
+
 		ret = core_clks_enable(core);
-	else
+		if (ret)
+			goto fail_pmdomains;
+
+		ret = vcodec_clks_enable(core, core->vcodec_clks);
+		if (ret)
+			goto fail_core_clks;
+
+	} else {
+		vcodec_clks_disable(core, core->vcodec_clks);
 		core_clks_disable(core);
+		vcodec_domains_disable(core);
+	}
+
+	return 0;
 
+fail_core_clks:
+	core_clks_disable(core);
+fail_pmdomains:
+	vcodec_domains_disable(core);
 	return ret;
 }
 
@@ -875,7 +983,7 @@ static int venc_power_v4(struct device *dev, int on)
 	return ret;
 }
 
-static int vcodec_domains_get(struct venus_core *core)
+static int vcodec_domains_get_v4(struct venus_core *core)
 {
 	int ret;
 	struct device *dev = core->dev;
@@ -993,7 +1101,7 @@ static int core_get_v4(struct venus_core *core)
 	if (ret)
 		return ret;
 
-	ret = vcodec_domains_get(core);
+	ret = vcodec_domains_get_v4(core);
 	if (ret)
 		return ret;
 

-- 
2.54.0

Re: [PATCH v7 6/8] media: qcom: venus: add power domain enable logic for Venus cores

[sashiko-bot]

Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:
- [High] Clock initialization is incorrectly gated by the presence of power domains, leading to unclocked hardware access.
- [Medium] Runtime PM reference leak in vcodec_domains_enable() error path.
--

commit 1bdd7b26dbc6049ee656dfce05ea05b682bb5990
Author: Erikas Bitovtas <xerikasxx@gmail.com>

media: qcom: venus: add power domain enable logic for Venus cores

Attach power domains for vdec and venc cores and power them up if a vdec
or venc session is started. Vcodev clocks are added and enabled to the
core Venus device both for vcodec0 and vcodec1.

> diff --git a/drivers/media/platform/qcom/venus/pm_helpers.c b/drivers/media/platform/qcom/venus/pm_helpers.c
> index f0269524ac70e..be705d4d64b53 100644
> --- a/drivers/media/platform/qcom/venus/pm_helpers.c
> +++ b/drivers/media/platform/qcom/venus/pm_helpers.c

[ ... ]

> @@ -309,9 +343,63 @@ static int core_get_v1(struct venus_core *core)
>  	if (ret)
>  		return ret;
>  
> +	if (!res->vcodec_pmdomains)
> +		return 0;
> +
> +	ret = vcodec_clks_get(core, dev, core->vcodec_clks,
> +			      res->vcodec_clks);
> +	if (ret)
> +		return ret;
> +
>  	return 0;
>  }

Will skipping vcodec_clks_get() when power domains are not present leave
the vcodec clocks uninitialized?

If a platform defines vcodec_clks but does not require vcodec_pmdomains,
this early return skips the clock retrieval entirely.

When core_power_v1() later calls vcodec_clks_enable(), it will succeed
on the NULL pointers, but the physical hardware clocks will remain disabled.
Accessing Venus hardware registers without clocks enabled can trigger an
asynchronous external abort and a kernel panic.

> +static int vcodec_domains_enable(struct venus_core *core)
> +{
> +	const struct venus_resources *res = core->res;
> +	struct device *pd_dev;
> +	int i = 0, ret;
> +
> +	if (!res->vcodec_pmdomains)
> +		return 0;
> +
> +	for (; i < res->vcodec_pmdomains_num; i++) {
> +		pd_dev = core->pmdomains->pd_devs[i];
> +		ret = pm_runtime_resume_and_get(pd_dev);
> +		if (ret)
> +			goto err;
> +
> +		ret = dev_pm_genpd_set_hwmode(pd_dev, true);
> +		if (ret && ret != -EOPNOTSUPP)
> +			goto err;
> +	}
> +
> +	return 0;
> +err:
> +	while (i--) {
> +		pd_dev = core->pmdomains->pd_devs[i];
> +		dev_pm_genpd_set_hwmode(pd_dev, false);
> +		pm_runtime_put_sync(pd_dev);
> +	}
> +	return ret;
> +}

Does this error path leak a runtime PM reference if dev_pm_genpd_set_hwmode()
fails?

If pm_runtime_resume_and_get() succeeds but the subsequent call to
dev_pm_genpd_set_hwmode() fails, the code jumps to the err label.

Since while (i--) pre-decrements the index, it skips the current iteration
index i and rolls back from i-1 down to 0.

This misses the pm_runtime_put_sync() call for the current power domain
where dev_pm_genpd_set_hwmode() failed, leaving its reference incremented
and preventing it from ever suspending.

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260514-msm8939-venus-rfc-v7-0-33c6c6fb9285@gmail.com?part=6

Re: [PATCH v7 6/8] media: qcom: venus: add power domain enable logic for Venus cores

[Bryan O'Donoghue]

On 13/05/2026 22:24, Erikas Bitovtas wrote:
> +static int vcodec_domains_enable(struct venus_core *core)
> +{
> +	const struct venus_resources *res = core->res;
> +	struct device *pd_dev;
> +	int i = 0, ret;
> +
> +	if (!res->vcodec_pmdomains)
> +		return 0;
> +
> +	for (; i < res->vcodec_pmdomains_num; i++) {
> +		pd_dev = core->pmdomains->pd_devs[i];
> +		ret = pm_runtime_resume_and_get(pd_dev);
> +		if (ret)
> +			goto err;
> +
> +		ret = dev_pm_genpd_set_hwmode(pd_dev, true);
> +		if (ret && ret != -EOPNOTSUPP)
> +			goto err;
> +	}
> +
> +	return 0;
> +err:
> +	while (i--) {
> +		pd_dev = core->pmdomains->pd_devs[i];
> +		dev_pm_genpd_set_hwmode(pd_dev, false);
> +		pm_runtime_put_sync(pd_dev);
> +	}
\n> +	return ret;
> +}
> +

+

Reviewed-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org>

---
bod

[PATCH v7 7/8] media: qcom: venus: add codec blacklist mechanism

[Erikas Bitovtas]

Add decode and encode blacklist properties to allow disabling different
codecs per Venus device, instead of doing it per HFI version.

Signed-off-by: Erikas Bitovtas <xerikasxx@gmail.com>
---
 drivers/media/platform/qcom/venus/core.c       |  4 ++++
 drivers/media/platform/qcom/venus/core.h       |  2 ++
 drivers/media/platform/qcom/venus/hfi_parser.c | 10 +++++-----
 3 files changed, 11 insertions(+), 5 deletions(-)

diff --git a/drivers/media/platform/qcom/venus/core.c b/drivers/media/platform/qcom/venus/core.c
index 7e639760c41d..1b5e72980702 100644
--- a/drivers/media/platform/qcom/venus/core.c
+++ b/drivers/media/platform/qcom/venus/core.c
@@ -178,6 +178,8 @@ static void venus_sys_error_handler(struct work_struct *work)
 static u32 to_v4l2_codec_type(u32 codec)
 {
 	switch (codec) {
+	case HFI_VIDEO_CODEC_HEVC:
+		return V4L2_PIX_FMT_HEVC;
 	case HFI_VIDEO_CODEC_H264:
 		return V4L2_PIX_FMT_H264;
 	case HFI_VIDEO_CODEC_H263:
@@ -684,6 +686,8 @@ static const struct venus_resources msm8916_res = {
 	.vmem_addr = 0,
 	.dma_mask = 0xddc00000 - 1,
 	.fwname = "qcom/venus-1.8/venus.mbn",
+	.dec_codec_blacklist = HFI_VIDEO_CODEC_HEVC | HFI_VIDEO_CODEC_SPARK,
+	.enc_codec_blacklist = HFI_VIDEO_CODEC_HEVC,
 	.dec_nodename = "video-decoder",
 	.enc_nodename = "video-encoder",
 };
diff --git a/drivers/media/platform/qcom/venus/core.h b/drivers/media/platform/qcom/venus/core.h
index ff9174a4e8c1..3f162b11e669 100644
--- a/drivers/media/platform/qcom/venus/core.h
+++ b/drivers/media/platform/qcom/venus/core.h
@@ -85,6 +85,8 @@ struct venus_resources {
 	unsigned int vcodec_pmdomains_num;
 	const char **opp_pmdomain;
 	unsigned int vcodec_num;
+	const u32 dec_codec_blacklist;
+	const u32 enc_codec_blacklist;
 	const char * const resets[VIDC_RESETS_NUM_MAX];
 	unsigned int resets_num;
 	enum hfi_version hfi_version;
diff --git a/drivers/media/platform/qcom/venus/hfi_parser.c b/drivers/media/platform/qcom/venus/hfi_parser.c
index 92765f9c8873..8e762cf77968 100644
--- a/drivers/media/platform/qcom/venus/hfi_parser.c
+++ b/drivers/media/platform/qcom/venus/hfi_parser.c
@@ -206,11 +206,11 @@ static int parse_codecs(struct venus_core *core, void *data)
 	core->dec_codecs = codecs->dec_codecs;
 	core->enc_codecs = codecs->enc_codecs;
 
-	if (IS_V1(core)) {
-		core->dec_codecs &= ~HFI_VIDEO_CODEC_HEVC;
-		core->dec_codecs &= ~HFI_VIDEO_CODEC_SPARK;
-		core->enc_codecs &= ~HFI_VIDEO_CODEC_HEVC;
-	}
+	if (core->res->dec_codec_blacklist)
+		core->dec_codecs &= ~core->res->dec_codec_blacklist;
+
+	if (core->res->enc_codec_blacklist)
+		core->enc_codecs &= ~core->res->enc_codec_blacklist;
 
 	return sizeof(*codecs);
 }

-- 
2.54.0

Re: [PATCH v7 7/8] media: qcom: venus: add codec blacklist mechanism

[Bryan O'Donoghue]

On 13/05/2026 22:24, Erikas Bitovtas wrote:
>   	case HFI_VIDEO_CODEC_H263:
> @@ -684,6 +686,8 @@ static const struct venus_resources msm8916_res = {
>   	.vmem_addr = 0,
>   	.dma_mask = 0xddc00000 - 1,
>   	.fwname = "qcom/venus-1.8/venus.mbn",
> +	.dec_codec_blacklist = HFI_VIDEO_CODEC_HEVC | HFI_VIDEO_CODEC_SPARK,
> +	.enc_codec_blacklist = HFI_VIDEO_CODEC_HEVC,

This should be separate from the addition of the mechanism.

This is enumeration of a particular plaform and you might want for 
example to revert the 8916 stuff in isolation without yanking the logic 
that goes with it.

--
bod

[PATCH v7 8/8] media: qcom: venus: Add msm8939 resource struct

[Erikas Bitovtas]

From: André Apitzsch <git@apitzsch.eu>

Add msm8939 configuration data and related compatible.

Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Reviewed-by: Bryan O'Donoghue <bod@kernel.org>
Signed-off-by: André Apitzsch <git@apitzsch.eu>
Signed-off-by: Erikas Bitovtas <xerikasxx@gmail.com>
---
 drivers/media/platform/qcom/venus/core.c | 40 ++++++++++++++++++++++++++++++++
 1 file changed, 40 insertions(+)

diff --git a/drivers/media/platform/qcom/venus/core.c b/drivers/media/platform/qcom/venus/core.c
index 1b5e72980702..b6eb96e43319 100644
--- a/drivers/media/platform/qcom/venus/core.c
+++ b/drivers/media/platform/qcom/venus/core.c
@@ -692,6 +692,45 @@ static const struct venus_resources msm8916_res = {
 	.enc_nodename = "video-encoder",
 };
 
+static const struct freq_tbl msm8939_freq_table[] = {
+	{ 489600, 266670000 },	/* 1080p @ 60 */
+	{ 244800, 133330000 },	/* 1080p @ 30 */
+	{ 220800, 133330000 },	/* 720p @ 60 */
+	{ 108000, 133330000 },	/* 720p @ 30 */
+	{ 72000, 133330000 },	/* VGA @ 60 */
+	{ 36000, 133330000 },	/* VGA @ 30 */
+};
+
+static const struct reg_val msm8939_reg_preset[] = {
+	{ 0xe0020, 0x0aaaaaaa },
+	{ 0xe0024, 0x0aaaaaaa },
+	{ 0x80124, 0x00000003 },
+};
+
+static const struct venus_resources msm8939_res = {
+	.freq_tbl = msm8939_freq_table,
+	.freq_tbl_size = ARRAY_SIZE(msm8939_freq_table),
+	.reg_tbl = msm8939_reg_preset,
+	.reg_tbl_size = ARRAY_SIZE(msm8939_reg_preset),
+	.clks = { "core", "iface", "bus", },
+	.clks_num = 3,
+	.vcodec_clks = { "vcodec0_core", "vcodec1_core" },
+	.vcodec_clks_num = 2,
+	.vcodec_pmdomains = (const char *[]) { "venus", "vcodec0", "vcodec1" },
+	.vcodec_pmdomains_num = 3,
+	.max_load = 489600, /* 1080p@30 + 1080p@30 */
+	.hfi_version = HFI_VERSION_1XX,
+	.vmem_id = VIDC_RESOURCE_NONE,
+	.vmem_size = 0,
+	.vmem_addr = 0,
+	.dma_mask = 0xddc00000 - 1,
+	.fwname = "qcom/venus-1.8/venus.mbn",
+	.dec_codec_blacklist = HFI_VIDEO_CODEC_SPARK,
+	.enc_codec_blacklist = HFI_VIDEO_CODEC_HEVC,
+	.dec_nodename = "video-decoder",
+	.enc_nodename = "video-encoder",
+};
+
 static const struct freq_tbl msm8996_freq_table[] = {
 	{ 1944000, 520000000 },	/* 4k UHD @ 60 (decode only) */
 	{  972000, 520000000 },	/* 4k UHD @ 30 */
@@ -1125,6 +1164,7 @@ static const struct venus_resources qcm2290_res = {
 
 static const struct of_device_id venus_dt_match[] = {
 	{ .compatible = "qcom,msm8916-venus", .data = &msm8916_res, },
+	{ .compatible = "qcom,msm8939-venus", .data = &msm8939_res, },
 	{ .compatible = "qcom,msm8996-venus", .data = &msm8996_res, },
 	{ .compatible = "qcom,msm8998-venus", .data = &msm8998_res, },
 	{ .compatible = "qcom,qcm2290-venus", .data = &qcm2290_res, },

-- 
2.54.0

Re: [PATCH v7 0/8] media: qcom: venus: add MSM8939 support

[Bryan O'Donoghue]

On 13/05/2026 22:24, Erikas Bitovtas wrote:
> This patch series adds support for Venus on MSM8939. It is mostly
> similar to MSM8916 Venus, except it needs two additional cores to be
> powered on before it can start decoding.
Stylistic feedback for future reference.

- Schema
- Driver
- DTS

Is the correct stacking for patches. Makes it just a little bit easier 
to pick and apply, no need to resend for this.

---
bod