Skip to contentSkip to conversation

phy: mtk-mipi-csi: add driver for CSI phy

PATCH Superseded Julien Stephan 6 participants 24 messages May 15, 2023 – Jun 8, 2023

Subsystems (6)

v2 of 4 Superseded Latest is v4 →

lore
Share & tools
lore

Julien Stephan

author maintainer
#0 lore raw

v2 cover phy: mtk-mipi-csi: add driver for CSI phy

Adding a new driver for the MIPI CSI CD-PHY module v 0.5 embedded in some Mediatek soc, such as the MT8365

This driver was adapted from https://patchwork.kernel.org/project/linux-mediatek/cover/[email protected]/

v1 can be found here: https://lore.kernel.org/all/[email protected]/

Changelog
Changes in v2:
- fix all comments on bindings
- move the binding chunk from driver to binding commit
- fix dt_binding_check error (reported by DT_CHECKER_FLAGS)
- use a more generic compatible string
- add a new dt properties to simplify the driver "mediatek,is_cdphy"
- rename the driver and the corresponding file to include version
- drop of_match_ptr()
- use devm_platform_ioremap_resource
- use phy-mtk-io.h api instead of regmap
- rework the driver to use dt nodes to declare PHY instead of an array in the driver
- remove useless define for unused registers
- remove support for CSI0A/B because it cannot be tested, and it simplifies the driver for a first review
- edit commit message and bindings to be more descriptive about the hardware

Florian Sylvestre (1):
dt-bindings: phy: add mediatek mipi csi driver v 0.5

Phi-bang Nguyen (1):
phy: mtk-mipi-csi: add driver for CSI phy

 .../phy/mediatek,phy-mipi-csi-0-5.yaml        |  62 +++++
 MAINTAINERS                                   |   7 +
 drivers/phy/mediatek/Kconfig                  |   8 +
 drivers/phy/mediatek/Makefile                 |   2 +
 .../mediatek/phy-mtk-mipi-csi-0-5-rx-reg.h    |  58 ++++
 drivers/phy/mediatek/phy-mtk-mipi-csi-0-5.c   | 257 ++++++++++++++++++
 .../phy/mediatek,phy-mipi-csi-0-5.h           |  15 +
 7 files changed, 409 insertions(+)
 create mode 100644 Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml
 create mode 100644 drivers/phy/mediatek/phy-mtk-mipi-csi-0-5-rx-reg.h
 create mode 100644 drivers/phy/mediatek/phy-mtk-mipi-csi-0-5.c
 create mode 100644 include/dt-bindings/phy/mediatek,phy-mipi-csi-0-5.h

--
2.40.0

author maintainer
#1 lore raw

PATCH v2 1/2 dt-bindings: phy: add mediatek mipi csi driver v 0.5

From: Florian Sylvestre <[email protected]>

This adds the bindings, for the MIPI CD-PHY module v 0.5 embedded in some Mediatek soc, such as the mt8365

---
 .../phy/mediatek,phy-mipi-csi-0-5.yaml        | 62 +++++++++++++++++++
 MAINTAINERS                                   |  6 ++
 2 files changed, 68 insertions(+)
 create mode 100644 Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml
Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml new file +62 -0
new file mode 100644
index 000000000000..5aa8c0b41cdf
@@ -0,0 +1,62 @@
+# SPDX-License-Identifier: (GPL-2.0-Only OR BSD-2-Clause)
+%YAML 1.2
+---
+$id: http://devicetree.org/schemas/phy/mediatek,phy-mipi-csi-0-5.yaml#
+$schema: http://devicetree.org/meta-schemas/core.yaml#
+
+title: Mediatek Sensor Interface MIPI CSI CD-PHY
+
+maintainers:
+  - Julien Stephan <[email protected]>
+  - Andy Hsieh <[email protected]>
+
+description:
+  The SENINF CD-PHY is a set of CD-PHY connected to the SENINF CSI-2
+  receivers. The number of PHYs depends on the SoC model.
+  Depending on the soc model, each PHYs can support CDPHY or DPHY only
+
+properties:
+  compatible:
+    enum:
+      - mediatek,phy-mipi-csi-0-5
+
+  reg:
+    maxItems: 1
+
+  '#phy-cells':
+    const: 0
+
+  mediatek,is_cdphy:
+    description:
+      Specify if the current phy support CDPHY configuration
+    type: boolean
+
+required:
+  - compatible
+  - reg
+  - '#phy-cells'
+
+additionalProperties: false
+
+examples:
+  - |
+    soc {
+      #address-cells = <2>;
+      #size-cells = <2>;
+
+      mipi_rx_csi0: mipi_rx_csi0@11c10000 {
+        compatible = "mediatek,phy-mipi-csi-0-5";
+        reg = <0 0x11C10000 0 0x2000>;
+        status = "okay";
+        mediatek,is_cdphy;
+        #phy-cells = <0>;
+      };
+
+      mipi_rx_csi1: mipi-rx-csi1@11c12000 {
+        compatible = "mediatek,phy-mipi-csi-0-5";
+        reg = <0 0x11C12000 0 0x2000>;
+        status = "disabled";
+        #phy-cells = <0>;
+      };
+    };
+...
MAINTAINERS +6 -0
index 6d54f3193075..44f0ff11e984 100644
@@ -13098,6 +13098,12 @@ F:	Documentation/devicetree/bindings/media/mediatek-vpu.txt
 F:	drivers/media/platform/mediatek/vcodec/
 F:	drivers/media/platform/mediatek/vpu/
 
+MEDIATEK MIPI-CSI CDPHY DRIVER
+M:	Julien Stephan <[email protected]>
+M:	Andy Hsieh <[email protected]>
+S:	Supported
+F:	Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml
+
 MEDIATEK MMC/SD/SDIO DRIVER
 M:	Chaotian Jing <[email protected]>
 S:	Maintained
Signature · 1 line
2.40.0
#2 lore raw

Re: [PATCH v2 1/2] dt-bindings: phy: add mediatek mipi csi driver v 0.5

Il 15/05/23 11:05, Julien Stephan ha scritto:

Quoted 30 lines

From: Florian Sylvestre <[email protected]>

This adds the bindings, for the MIPI CD-PHY module v 0.5 embedded in some Mediatek soc, such as the mt8365

---
  .../phy/mediatek,phy-mipi-csi-0-5.yaml        | 62 +++++++++++++++++++
  MAINTAINERS                                   |  6 ++
  2 files changed, 68 insertions(+)

create mode 100644 Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml

Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml new file +11 -0
new file mode 100644
index 000000000000..5aa8c0b41cdf
@@ -0,0 +1,62 @@
+# SPDX-License-Identifier: (GPL-2.0-Only OR BSD-2-Clause)
+%YAML 1.2
+---
+$id: http://devicetree.org/schemas/phy/mediatek,phy-mipi-csi-0-5.yaml#
+$schema: http://devicetree.org/meta-schemas/core.yaml#
+
+title: Mediatek Sensor Interface MIPI CSI CD-PHY
+
+maintainers:
+  - Julien Stephan <[email protected]>
+  - Andy Hsieh <[email protected]>

Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml line 21

Quoted 10 lines
Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml new file +10 -0
+
+description:
+  The SENINF CD-PHY is a set of CD-PHY connected to the SENINF CSI-2
+  receivers. The number of PHYs depends on the SoC model.
+  Depending on the soc model, each PHYs can support CDPHY or DPHY only
+
+properties:
+  compatible:
+    enum:
+      - mediatek,phy-mipi-csi-0-5

mediatek,mtXXXX-csi-rx sounds great, doesn't it?

Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml line 29

Quoted 8 lines
Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml new file +8 -0
+
+  reg:
+    maxItems: 1
+
+  '#phy-cells':
+    const: 0
+
+  mediatek,is_cdphy:

No underscores please: mediatek,is-cdphy

Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml line 31

Quoted 2 lines
Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml new file +2 -0
+    description:
+      Specify if the current phy support CDPHY configuration

Description fits in one line.

description: Specify if the current phy support CDPHY configuration

Quoted 6 lines
Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml new file +6 -0
+    type: boolean
+
+required:
+  - compatible
+  - reg
+  - '#phy-cells'

Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml line 47

Quoted 10 lines
Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml new file +10 -0
+
+additionalProperties: false
+
+examples:
+  - |
+    soc {
+      #address-cells = <2>;
+      #size-cells = <2>;
+
+      mipi_rx_csi0: mipi_rx_csi0@11c10000 {

csi0_rx: phy@11c10000

Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml line 55

Quoted 8 lines
Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml new file +8 -0
+        compatible = "mediatek,phy-mipi-csi-0-5";
+        reg = <0 0x11C10000 0 0x2000>;
+        status = "okay";
+        mediatek,is_cdphy;
+        #phy-cells = <0>;
+      };
+
+      mipi_rx_csi1: mipi-rx-csi1@11c12000 {

csi1_rx: phy@11c20000

Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml line 62

Quoted 7 lines
Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml new file +7 -0
+        compatible = "mediatek,phy-mipi-csi-0-5";
+        reg = <0 0x11C12000 0 0x2000>;
+        status = "disabled";
+        #phy-cells = <0>;
+      };
+    };
+...

Regards,
Angelo

#3 lore raw

Re: [PATCH v2 1/2] dt-bindings: phy: add mediatek mipi csi driver v 0.5

In reply to Julien Stephan
Quoted 7 lines · Julien Stephan

From: Florian Sylvestre <[email protected]>

This adds the bindings, for the MIPI CD-PHY module v 0.5 embedded in some Mediatek soc, such as the mt8365

What are the changes? IOW: changelog here or in cover letter.

Subject: you have some multiple spaces.

Subject: drop driver. Bindings are not for drivers.

Quoted 23 lines
---
 .../phy/mediatek,phy-mipi-csi-0-5.yaml        | 62 +++++++++++++++++++
 MAINTAINERS                                   |  6 ++
 2 files changed, 68 insertions(+)
 create mode 100644 Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml
Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml new file +11 -0
new file mode 100644
index 000000000000..5aa8c0b41cdf
@@ -0,0 +1,62 @@
+# SPDX-License-Identifier: (GPL-2.0-Only OR BSD-2-Clause)
+%YAML 1.2
+---
+$id: http://devicetree.org/schemas/phy/mediatek,phy-mipi-csi-0-5.yaml#
+$schema: http://devicetree.org/meta-schemas/core.yaml#
+
+title: Mediatek Sensor Interface MIPI CSI CD-PHY
+
+maintainers:
+  - Julien Stephan <[email protected]>
+  - Andy Hsieh <[email protected]>

Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml line 21

Quoted 10 lines
Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml new file +10 -0
+
+description:
+  The SENINF CD-PHY is a set of CD-PHY connected to the SENINF CSI-2
+  receivers. The number of PHYs depends on the SoC model.
+  Depending on the soc model, each PHYs can support CDPHY or DPHY only
+
+properties:
+  compatible:
+    enum:
+      - mediatek,phy-mipi-csi-0-5

SoC based compatibles. 0-5 is odd.

Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml line 29

Quoted 8 lines
Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml new file +8 -0
+
+  reg:
+    maxItems: 1
+
+  '#phy-cells':
+    const: 0
+
+  mediatek,is_cdphy:

No underscores in node names.

Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml line 31

Quoted 2 lines
Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml new file +2 -0
+    description:
+      Specify if the current phy support CDPHY configuration

Why this cannot be implied from compatible? Add specific compatibles.

Quoted 6 lines
Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml new file +6 -0
+    type: boolean
+
+required:
+  - compatible
+  - reg
+  - '#phy-cells'

Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml line 47

Quoted 10 lines
Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml new file +10 -0
+
+additionalProperties: false
+
+examples:
+  - |
+    soc {
+      #address-cells = <2>;
+      #size-cells = <2>;
+
+      mipi_rx_csi0: mipi_rx_csi0@11c10000 {

No underscores in node names. How this is v2?

Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml line 50

Quoted 3 lines
Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml new file +3 -0
+        compatible = "mediatek,phy-mipi-csi-0-5";
+        reg = <0 0x11C10000 0 0x2000>;
+        status = "okay";

Drop

Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml line 58

Quoted 8 lines
Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml new file +8 -0
+        mediatek,is_cdphy;
+        #phy-cells = <0>;
+      };
+
+      mipi_rx_csi1: mipi-rx-csi1@11c12000 {
+        compatible = "mediatek,phy-mipi-csi-0-5";
+        reg = <0 0x11C12000 0 0x2000>;
+        status = "disabled";

???

Best regards,
Krzysztof

author maintainer
#4 lore raw

Re: [PATCH v2 1/2] dt-bindings: phy: add mediatek mipi csi driver v 0.5

Quoted 11 lines · Krzysztof Kozlowski
Quoted 7 lines · Julien Stephan

From: Florian Sylvestre <[email protected]>

This adds the bindings, for the MIPI CD-PHY module v 0.5 embedded in some Mediatek soc, such as the mt8365

What are the changes? IOW: changelog here or in cover letter.

Hi Krzysztof,
I added a changelog in the cover letter, but I will try to be more descritpive next time. Changes from v1 are mainly style issues fixed (mostly from your first review)

Quoted 51 lines

Subject: you have some multiple spaces.

Subject: drop driver. Bindings are not for drivers.

Quoted 23 lines
---
 .../phy/mediatek,phy-mipi-csi-0-5.yaml        | 62 +++++++++++++++++++
 MAINTAINERS                                   |  6 ++
 2 files changed, 68 insertions(+)
 create mode 100644 Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml
Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml new file +11 -0
new file mode 100644
index 000000000000..5aa8c0b41cdf
@@ -0,0 +1,62 @@
+# SPDX-License-Identifier: (GPL-2.0-Only OR BSD-2-Clause)
+%YAML 1.2
+---
+$id: http://devicetree.org/schemas/phy/mediatek,phy-mipi-csi-0-5.yaml#
+$schema: http://devicetree.org/meta-schemas/core.yaml#
+
+title: Mediatek Sensor Interface MIPI CSI CD-PHY
+
+maintainers:
+  - Julien Stephan <[email protected]>
+  - Andy Hsieh <[email protected]>

Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml line 21

Quoted 10 lines
Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml new file +10 -0
+
+description:
+  The SENINF CD-PHY is a set of CD-PHY connected to the SENINF CSI-2
+  receivers. The number of PHYs depends on the SoC model.
+  Depending on the soc model, each PHYs can support CDPHY or DPHY only
+
+properties:
+  compatible:
+    enum:
+      - mediatek,phy-mipi-csi-0-5

SoC based compatibles. 0-5 is odd.

Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml line 29

Quoted 8 lines
Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml new file +8 -0
+
+  reg:
+    maxItems: 1
+
+  '#phy-cells':
+    const: 0
+
+  mediatek,is_cdphy:

No underscores in node names.

Quoted 6 lines

Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml line 31

Quoted 2 lines
Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml new file +2 -0
+    description:
+      Specify if the current phy support CDPHY configuration

Why this cannot be implied from compatible? Add specific compatibles.

This cannot be implied by compatible because the number of phys depends on the soc and each phy can be either D-PHY only or CD-PHY capable. For example mt8365 has 2 phy: CSI0 and CSI1. CSI1 is DPHY only and CSI0 is CD-PHY

Regards,
Julien

Quoted 40 lines
Quoted 6 lines
Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml new file +6 -0
+    type: boolean
+
+required:
+  - compatible
+  - reg
+  - '#phy-cells'

Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml line 47

Quoted 10 lines
Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml new file +10 -0
+
+additionalProperties: false
+
+examples:
+  - |
+    soc {
+      #address-cells = <2>;
+      #size-cells = <2>;
+
+      mipi_rx_csi0: mipi_rx_csi0@11c10000 {

No underscores in node names. How this is v2?

Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml line 50

Quoted 3 lines
Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml new file +3 -0
+        compatible = "mediatek,phy-mipi-csi-0-5";
+        reg = <0 0x11C10000 0 0x2000>;
+        status = "okay";

Drop

Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml line 58

Quoted 8 lines
Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml new file +8 -0
+        mediatek,is_cdphy;
+        #phy-cells = <0>;
+      };
+
+      mipi_rx_csi1: mipi-rx-csi1@11c12000 {
+        compatible = "mediatek,phy-mipi-csi-0-5";
+        reg = <0 0x11C12000 0 0x2000>;
+        status = "disabled";

???

Best regards,
Krzysztof

Signature · 3 lines
linux-phy mailing list
[email protected]
https://lists.infradead.org/mailman/listinfo/linux-phy
#5 lore raw

Re: [PATCH v2 1/2] dt-bindings: phy: add mediatek mipi csi driver v 0.5

Quoted 10 lines · Julien Stephan
Quoted 9 lines · Krzysztof Kozlowski
Quoted 7 lines · Julien Stephan

From: Florian Sylvestre <[email protected]>

This adds the bindings, for the MIPI CD-PHY module v 0.5 embedded in some Mediatek soc, such as the mt8365

Quoted 6 lines
Quoted 2 lines

What are the changes? IOW: changelog here or in cover letter.

Hi Krzysztof,
I added a changelog in the cover letter, but I will try to be more descritpive next time. Changes from v1 are mainly style issues fixed (mostly from your first review)

What do you mean by "in cover letter"? There is no cover letter.

Quoted 55 lines
Quoted 55 lines

Subject: you have some multiple spaces.

Subject: drop driver. Bindings are not for drivers.

Quoted 23 lines
---
 .../phy/mediatek,phy-mipi-csi-0-5.yaml        | 62 +++++++++++++++++++
 MAINTAINERS                                   |  6 ++
 2 files changed, 68 insertions(+)
 create mode 100644 Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml
Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml new file +11 -0
new file mode 100644
index 000000000000..5aa8c0b41cdf
@@ -0,0 +1,62 @@
+# SPDX-License-Identifier: (GPL-2.0-Only OR BSD-2-Clause)
+%YAML 1.2
+---
+$id: http://devicetree.org/schemas/phy/mediatek,phy-mipi-csi-0-5.yaml#
+$schema: http://devicetree.org/meta-schemas/core.yaml#
+
+title: Mediatek Sensor Interface MIPI CSI CD-PHY
+
+maintainers:
+  - Julien Stephan <[email protected]>
+  - Andy Hsieh <[email protected]>

Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml line 21

Quoted 10 lines
Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml new file +10 -0
+
+description:
+  The SENINF CD-PHY is a set of CD-PHY connected to the SENINF CSI-2
+  receivers. The number of PHYs depends on the SoC model.
+  Depending on the soc model, each PHYs can support CDPHY or DPHY only
+
+properties:
+  compatible:
+    enum:
+      - mediatek,phy-mipi-csi-0-5

SoC based compatibles. 0-5 is odd.

Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml line 29

Quoted 8 lines
Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml new file +8 -0
+
+  reg:
+    maxItems: 1
+
+  '#phy-cells':
+    const: 0
+
+  mediatek,is_cdphy:

No underscores in node names.

Quoted 2 lines
Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml new file +2 -0
+    description:
+      Specify if the current phy support CDPHY configuration
Quoted 6 lines
Quoted 3 lines

Why this cannot be implied from compatible? Add specific compatibles.

This cannot be implied by compatible because the number of phys depends on the soc and each phy can be either D-PHY only or CD-PHY capable. For example mt8365 has 2 phy: CSI0 and CSI1. CSI1 is DPHY only and CSI0 is CD-PHY

So it is SoC specific so why it cannot be implied by compatible? I don't understand. You will have SoC specific compatibles, right? or you just ignored my comments here?

Quoted 1 line
Quoted 1 line

Best regards,
Krzysztof

Signature · 3 lines
linux-phy mailing list
[email protected]
https://lists.infradead.org/mailman/listinfo/linux-phy
#6 lore raw

Re: [PATCH v2 1/2] dt-bindings: phy: add mediatek mipi csi driver v 0.5

Quoted 11 lines · Krzysztof Kozlowski
Quoted 10 lines · Julien Stephan
Quoted 9 lines · Krzysztof Kozlowski
Quoted 7 lines · Julien Stephan

From: Florian Sylvestre <[email protected]>

This adds the bindings, for the MIPI CD-PHY module v 0.5 embedded in some Mediatek soc, such as the mt8365

Quoted 8 lines
Quoted 6 lines
Quoted 2 lines

What are the changes? IOW: changelog here or in cover letter.

Hi Krzysztof,
I added a changelog in the cover letter, but I will try to be more descritpive next time. Changes from v1 are mainly style issues fixed (mostly from your first review)

What do you mean by "in cover letter"? There is no cover letter.

Julien, your cover letter[1] was sent to a a different list of recipients than the patches, and most important for this thread, it was *not* sent to the devictree list. So I'm guessing that's why Krzysztof doesn't see it in his devicetree review queue. Generally, you should have the same list of recipients for the cover letter as the patches since reviewers/maintainers generally filter mail based on which mailing lists are in to/cc.

Quoted 58 lines
Quoted 58 lines
Quoted 58 lines

Subject: you have some multiple spaces.

Subject: drop driver. Bindings are not for drivers.

Quoted 23 lines
---
 .../phy/mediatek,phy-mipi-csi-0-5.yaml        | 62 +++++++++++++++++++
 MAINTAINERS                                   |  6 ++
 2 files changed, 68 insertions(+)
 create mode 100644 Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml
Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml new file +11 -0
new file mode 100644
index 000000000000..5aa8c0b41cdf
@@ -0,0 +1,62 @@
+# SPDX-License-Identifier: (GPL-2.0-Only OR BSD-2-Clause)
+%YAML 1.2
+---
+$id: http://devicetree.org/schemas/phy/mediatek,phy-mipi-csi-0-5.yaml#
+$schema: http://devicetree.org/meta-schemas/core.yaml#
+
+title: Mediatek Sensor Interface MIPI CSI CD-PHY
+
+maintainers:
+  - Julien Stephan <[email protected]>
+  - Andy Hsieh <[email protected]>

Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml line 21

Quoted 10 lines
Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml new file +10 -0
+
+description:
+  The SENINF CD-PHY is a set of CD-PHY connected to the SENINF CSI-2
+  receivers. The number of PHYs depends on the SoC model.
+  Depending on the soc model, each PHYs can support CDPHY or DPHY only
+
+properties:
+  compatible:
+    enum:
+      - mediatek,phy-mipi-csi-0-5

SoC based compatibles. 0-5 is odd.

Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml line 29

Quoted 8 lines
Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml new file +8 -0
+
+  reg:
+    maxItems: 1
+
+  '#phy-cells':
+    const: 0
+
+  mediatek,is_cdphy:

No underscores in node names.

Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml line 31

Quoted 2 lines
Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml new file +2 -0
+    description:
+      Specify if the current phy support CDPHY configuration

Why this cannot be implied from compatible? Add specific compatibles.

Quoted 7 lines
Quoted 3 lines

This cannot be implied by compatible because the number of phys depends on the soc and each phy can be either D-PHY only or CD-PHY capable. For example mt8365 has 2 phy: CSI0 and CSI1. CSI1 is DPHY only and CSI0 is CD-PHY

So it is SoC specific so why it cannot be implied by compatible? I don't understand. You will have SoC specific compatibles, right? or you just ignored my comments here?

Julien, I think you had SoC specific compatibles in an earlier version but then changed it to be generic based on reviewer feedback. However, that earlier version of the driver was trying to do a bunch of SoC specific logic internally and support multiple SoCs. You've now greatly simplified the driver, with only a few SoC specific decisions needed. These can be implied by the driver based SoC specific compatible, as Krzysztof suggests, so you should just go back to having SoC specific compatibles.

Kevin

[1] https://lore.kernel.org/linux-mediatek/[email protected]/#r

Signature · 3 lines
linux-phy mailing list
[email protected]
https://lists.infradead.org/mailman/listinfo/linux-phy
#7 lore raw

Re: [PATCH v2 1/2] dt-bindings: phy: add mediatek mipi csi driver v 0.5

Quoted 31 lines · Kevin Hilman
Quoted 23 lines
Quoted 23 lines
Quoted 23 lines

Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml line 21

Quoted 3 lines
Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml new file +3 -0
+  compatible:
+    enum:
+      - mediatek,phy-mipi-csi-0-5

SoC based compatibles. 0-5 is odd.

Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml line 29

Quoted 8 lines
Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml new file +8 -0
+
+  reg:
+    maxItems: 1
+
+  '#phy-cells':
+    const: 0
+
+  mediatek,is_cdphy:

No underscores in node names.

Review

Quoted 2 lines
(quoted diff) +2 -0
+    description:
+      Specify if the current phy support CDPHY configuration

Why this cannot be implied from compatible? Add specific compatibles.

Quoted 7 lines
Quoted 3 lines

This cannot be implied by compatible because the number of phys depends on the soc and each phy can be either D-PHY only or CD-PHY capable. For example mt8365 has 2 phy: CSI0 and CSI1. CSI1 is DPHY only and CSI0 is CD-PHY

So it is SoC specific so why it cannot be implied by compatible? I don't understand. You will have SoC specific compatibles, right? or you just ignored my comments here?

Quoted 9 lines

Julien, I think you had SoC specific compatibles in an earlier version but then changed it to be generic based on reviewer feedback. However, that earlier version of the driver was trying to do a bunch of SoC specific logic internally and support multiple SoCs. You've now greatly simplified the driver, with only a few SoC specific decisions needed. These can be implied by the driver based SoC specific compatible, as Krzysztof suggests, so you should just go back to having SoC specific compatibles.

Yes. If there is common part, e.g. several SoCs use the same device with same programming model, then the generic recommendation is to have SoC-based fallback (used also in the driver) and SoC-specific compatibles.

Second accepted solution is to have generic fallback which does not use SoC in the compatible (and of course mandatory SoC-specific comaptibles).

Third is to use versioned IP blocks.

The second case also would work, if it is applicable to you (you really have fallback matching all devices). Third solution depends on your versioning and Rob expressed dislike about it many times.

We had many discussions on mailing lists, thus simplifying the review - I recommend the first choice. For a better recommendation you should say a bit more about the block in different SoCs.

Best regards,
Krzysztof

#8 lore raw

Re: [PATCH v2 1/2] dt-bindings: phy: add mediatek mipi csi driver v 0.5

Hi Krzysztof,

Quoted 51 lines · Krzysztof Kozlowski
Quoted 31 lines · Kevin Hilman
Quoted 23 lines
Quoted 23 lines
Quoted 23 lines

Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml line 21

Quoted 3 lines
Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml new file +3 -0
+  compatible:
+    enum:
+      - mediatek,phy-mipi-csi-0-5

SoC based compatibles. 0-5 is odd.

Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml line 29

Quoted 8 lines
Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml new file +8 -0
+
+  reg:
+    maxItems: 1
+
+  '#phy-cells':
+    const: 0
+
+  mediatek,is_cdphy:

No underscores in node names.

Review

Quoted 2 lines
(quoted diff) +2 -0
+    description:
+      Specify if the current phy support CDPHY configuration

Why this cannot be implied from compatible? Add specific compatibles.

Quoted 7 lines
Quoted 3 lines

This cannot be implied by compatible because the number of phys depends on the soc and each phy can be either D-PHY only or CD-PHY capable. For example mt8365 has 2 phy: CSI0 and CSI1. CSI1 is DPHY only and CSI0 is CD-PHY

So it is SoC specific so why it cannot be implied by compatible? I don't understand. You will have SoC specific compatibles, right? or you just ignored my comments here?

Quoted 9 lines

Julien, I think you had SoC specific compatibles in an earlier version but then changed it to be generic based on reviewer feedback. However, that earlier version of the driver was trying to do a bunch of SoC specific logic internally and support multiple SoCs. You've now greatly simplified the driver, with only a few SoC specific decisions needed. These can be implied by the driver based SoC specific compatible, as Krzysztof suggests, so you should just go back to having SoC specific compatibles.

Yes. If there is common part, e.g. several SoCs use the same device with same programming model, then the generic recommendation is to have SoC-based fallback (used also in the driver) and SoC-specific compatibles.

Second accepted solution is to have generic fallback which does not use SoC in the compatible (and of course mandatory SoC-specific comaptibles).

Third is to use versioned IP blocks.

Quoted 7 lines

The second case also would work, if it is applicable to you (you really have fallback matching all devices). Third solution depends on your versioning and Rob expressed dislike about it many times.

We had many discussions on mailing lists, thus simplifying the review - I recommend the first choice. For a better recommendation you should say a bit more about the block in different SoCs.

I'll try to say a bit more about the PHY block, but in fact, it's not just about differences between SoCs. On the same SoC, 2 different PHYs may have different features/capabilities.

For example, on MT8365, There are 2 PHYs: CSI0 and CSI1. CSI0 can function as a C-PHY or a D-PHY, but CSI1 can only function as D-PHY (used as the example in the binding patch[1].) On another related SoC, there are 3 PHYs, where CSI0 is C-D but CSI1 & CSI2 are only D.

So that's why it seems (at least to me) that while we need SoC compatible, it's not enough. We also need properties to describe PHY-specific features (e.g. C-D PHY)

Of course, we could rely only on SoC-specific compatibles describe this. But then driver will need an SoC-specific table with the number of PHYs and per-PHY features for each SoC encoded in the driver. Since the driver otherwise doesn't (and shouldn't, IMHO) need to know how many PHYs are on each SoC, I suggested to Julien that perhaps the additional propery was the better solution.

To me it seems redundant to have the driver encode PHYs-per-SoC info, when the per-SoC DT is going to have the same info, so my suggestion was to simplify the driver and have this kind of hardware description in the DT, and keep the driver simple, but we are definitely open to learning the "right way" of doing this.

Thanks for your review and guidance,

Kevin

[1] https://lore.kernel.org/linux-mediatek/[email protected]/

#9 lore raw

Re: [PATCH v2 1/2] dt-bindings: phy: add mediatek mipi csi driver v 0.5

Quoted 14 lines · Kevin Hilman
Quoted 9 lines

Third is to use versioned IP blocks.

The second case also would work, if it is applicable to you (you really have fallback matching all devices). Third solution depends on your versioning and Rob expressed dislike about it many times.

We had many discussions on mailing lists, thus simplifying the review - I recommend the first choice. For a better recommendation you should say a bit more about the block in different SoCs.

I'll try to say a bit more about the PHY block, but in fact, it's not just about differences between SoCs. On the same SoC, 2 different PHYs may have different features/capabilities.

Quoted 8 lines

For example, on MT8365, There are 2 PHYs: CSI0 and CSI1. CSI0 can function as a C-PHY or a D-PHY, but CSI1 can only function as D-PHY (used as the example in the binding patch[1].) On another related SoC, there are 3 PHYs, where CSI0 is C-D but CSI1 & CSI2 are only D.

So that's why it seems (at least to me) that while we need SoC compatible, it's not enough. We also need properties to describe PHY-specific features (e.g. C-D PHY)

I recall the same or very similar case... It bugs me now, but unfortunately I cannot find it.

Quoted 7 lines

Of course, we could rely only on SoC-specific compatibles describe this. But then driver will need an SoC-specific table with the number of PHYs and per-PHY features for each SoC encoded in the driver. Since the driver otherwise doesn't (and shouldn't, IMHO) need to know how many PHYs are on each SoC, I suggested to Julien that perhaps the additional propery was the better solution.

Phys were modeled as separate device instances, so you would need difference in compatible to figure out which phy is it.

Other way could be to create device for all phys and use phy-cells=1. Whether it makes sense, depends on the actual datasheet - maybe the split phy per device is artificial? There is one PHY block with two address ranges for each PHY - CSI0 and CSI1 - but it is actually one block? You should carefully check this because once design is chosen, you won't be able to go back to other and it might be a problem (e.g. there is some top-level block for powering on all CSI instances).

Quoted 6 lines

To me it seems redundant to have the driver encode PHYs-per-SoC info, when the per-SoC DT is going to have the same info, so my suggestion was to simplify the driver and have this kind of hardware description in the DT, and keep the driver simple, but we are definitely open to learning the "right way" of doing this.

The property then is reasonable. It should not be bool, though, because it does not scale. There can be next block which supports only D-PHY on CSI0 and C-PHY on CSI1? Maybe some enum or list, depending on possible configurations.

Best regards,
Krzysztof

#10 lore raw

Re: [PATCH v2 1/2] dt-bindings: phy: add mediatek mipi csi driver v 0.5

Quoted 39 lines · Krzysztof Kozlowski
Quoted 14 lines · Kevin Hilman
Quoted 9 lines

Third is to use versioned IP blocks.

The second case also would work, if it is applicable to you (you really have fallback matching all devices). Third solution depends on your versioning and Rob expressed dislike about it many times.

We had many discussions on mailing lists, thus simplifying the review - I recommend the first choice. For a better recommendation you should say a bit more about the block in different SoCs.

I'll try to say a bit more about the PHY block, but in fact, it's not just about differences between SoCs. On the same SoC, 2 different PHYs may have different features/capabilities.

Quoted 8 lines

For example, on MT8365, There are 2 PHYs: CSI0 and CSI1. CSI0 can function as a C-PHY or a D-PHY, but CSI1 can only function as D-PHY (used as the example in the binding patch[1].) On another related SoC, there are 3 PHYs, where CSI0 is C-D but CSI1 & CSI2 are only D.

So that's why it seems (at least to me) that while we need SoC compatible, it's not enough. We also need properties to describe PHY-specific features (e.g. C-D PHY)

I recall the same or very similar case... It bugs me now, but unfortunately I cannot find it.

Quoted 7 lines

Of course, we could rely only on SoC-specific compatibles describe this. But then driver will need an SoC-specific table with the number of PHYs and per-PHY features for each SoC encoded in the driver. Since the driver otherwise doesn't (and shouldn't, IMHO) need to know how many PHYs are on each SoC, I suggested to Julien that perhaps the additional propery was the better solution.

Phys were modeled as separate device instances, so you would need difference in compatible to figure out which phy is it.

Quoted 7 lines

Other way could be to create device for all phys and use phy-cells=1. Whether it makes sense, depends on the actual datasheet - maybe the split phy per device is artificial? There is one PHY block with two address ranges for each PHY - CSI0 and CSI1 - but it is actually one block? You should carefully check this because once design is chosen, you won't be able to go back to other and it might be a problem (e.g. there is some top-level block for powering on all CSI instances).

We're pretty sure these are multiple instances of the IP block as they can operate completely independently.

Quoted 11 lines
Quoted 6 lines

To me it seems redundant to have the driver encode PHYs-per-SoC info, when the per-SoC DT is going to have the same info, so my suggestion was to simplify the driver and have this kind of hardware description in the DT, and keep the driver simple, but we are definitely open to learning the "right way" of doing this.

The property then is reasonable. It should not be bool, though, because it does not scale. There can be next block which supports only D-PHY on CSI0 and C-PHY on CSI1? Maybe some enum or list, depending on possible configurations.

OK, looks like include/dt-bindings/phy/phy.y already has

  #define PHY_TYPE_DPHY		10
  #define PHY_TYPE_CPHY		11

we'll add a PHY_TYPE_CDPHY and use that. Sound reasonable?

Kevin

Signature · 3 lines
linux-phy mailing list
[email protected]
https://lists.infradead.org/mailman/listinfo/linux-phy
#11 lore raw

Re: [PATCH v2 1/2] dt-bindings: phy: add mediatek mipi csi driver v 0.5

Quoted 64 lines · Kevin Hilman
Quoted 39 lines · Krzysztof Kozlowski
Quoted 14 lines · Kevin Hilman
Quoted 9 lines

Third is to use versioned IP blocks.

The second case also would work, if it is applicable to you (you really have fallback matching all devices). Third solution depends on your versioning and Rob expressed dislike about it many times.

We had many discussions on mailing lists, thus simplifying the review - I recommend the first choice. For a better recommendation you should say a bit more about the block in different SoCs.

I'll try to say a bit more about the PHY block, but in fact, it's not just about differences between SoCs. On the same SoC, 2 different PHYs may have different features/capabilities.

Quoted 8 lines

For example, on MT8365, There are 2 PHYs: CSI0 and CSI1. CSI0 can function as a C-PHY or a D-PHY, but CSI1 can only function as D-PHY (used as the example in the binding patch[1].) On another related SoC, there are 3 PHYs, where CSI0 is C-D but CSI1 & CSI2 are only D.

So that's why it seems (at least to me) that while we need SoC compatible, it's not enough. We also need properties to describe PHY-specific features (e.g. C-D PHY)

I recall the same or very similar case... It bugs me now, but unfortunately I cannot find it.

Quoted 7 lines

Of course, we could rely only on SoC-specific compatibles describe this. But then driver will need an SoC-specific table with the number of PHYs and per-PHY features for each SoC encoded in the driver. Since the driver otherwise doesn't (and shouldn't, IMHO) need to know how many PHYs are on each SoC, I suggested to Julien that perhaps the additional propery was the better solution.

Phys were modeled as separate device instances, so you would need difference in compatible to figure out which phy is it.

Quoted 7 lines

Other way could be to create device for all phys and use phy-cells=1. Whether it makes sense, depends on the actual datasheet - maybe the split phy per device is artificial? There is one PHY block with two address ranges for each PHY - CSI0 and CSI1 - but it is actually one block? You should carefully check this because once design is chosen, you won't be able to go back to other and it might be a problem (e.g. there is some top-level block for powering on all CSI instances).

We're pretty sure these are multiple instances of the IP block as they can operate completely independently.

Quoted 11 lines
Quoted 6 lines

To me it seems redundant to have the driver encode PHYs-per-SoC info, when the per-SoC DT is going to have the same info, so my suggestion was to simplify the driver and have this kind of hardware description in the DT, and keep the driver simple, but we are definitely open to learning the "right way" of doing this.

The property then is reasonable. It should not be bool, though, because it does not scale. There can be next block which supports only D-PHY on CSI0 and C-PHY on CSI1? Maybe some enum or list, depending on possible configurations.

Quoted 6 lines

OK, looks like include/dt-bindings/phy/phy.y already has

  #define PHY_TYPE_DPHY		10
  #define PHY_TYPE_CPHY		11

we'll add a PHY_TYPE_CDPHY and use that. Sound reasonable?

Yes. Currently it is usually used as phy-cells argument (after the phy number/lane/ID), but cdns,phy-type and intel,phy-mode use it directly as property in provider. In both cases they have a bit different meaning than yours. You want to list all supported modes or narrow/restrict them. Maybe hisilicon,fixed-mode fits your purpose?

Best regards,
Krzysztof

author maintainer
#12 lore raw

Re: [PATCH v2 1/2] dt-bindings: phy: add mediatek mipi csi driver v 0.5

Quoted 70 lines · Krzysztof Kozlowski
Quoted 69 lines · Kevin Hilman
Quoted 39 lines · Krzysztof Kozlowski
Quoted 14 lines · Kevin Hilman
Quoted 9 lines

Third is to use versioned IP blocks.

The second case also would work, if it is applicable to you (you really have fallback matching all devices). Third solution depends on your versioning and Rob expressed dislike about it many times.

We had many discussions on mailing lists, thus simplifying the review - I recommend the first choice. For a better recommendation you should say a bit more about the block in different SoCs.

I'll try to say a bit more about the PHY block, but in fact, it's not just about differences between SoCs. On the same SoC, 2 different PHYs may have different features/capabilities.

Quoted 8 lines

For example, on MT8365, There are 2 PHYs: CSI0 and CSI1. CSI0 can function as a C-PHY or a D-PHY, but CSI1 can only function as D-PHY (used as the example in the binding patch[1].) On another related SoC, there are 3 PHYs, where CSI0 is C-D but CSI1 & CSI2 are only D.

So that's why it seems (at least to me) that while we need SoC compatible, it's not enough. We also need properties to describe PHY-specific features (e.g. C-D PHY)

I recall the same or very similar case... It bugs me now, but unfortunately I cannot find it.

Quoted 7 lines

Of course, we could rely only on SoC-specific compatibles describe this. But then driver will need an SoC-specific table with the number of PHYs and per-PHY features for each SoC encoded in the driver. Since the driver otherwise doesn't (and shouldn't, IMHO) need to know how many PHYs are on each SoC, I suggested to Julien that perhaps the additional propery was the better solution.

Phys were modeled as separate device instances, so you would need difference in compatible to figure out which phy is it.

Quoted 7 lines

Other way could be to create device for all phys and use phy-cells=1. Whether it makes sense, depends on the actual datasheet - maybe the split phy per device is artificial? There is one PHY block with two address ranges for each PHY - CSI0 and CSI1 - but it is actually one block? You should carefully check this because once design is chosen, you won't be able to go back to other and it might be a problem (e.g. there is some top-level block for powering on all CSI instances).

We're pretty sure these are multiple instances of the IP block as they can operate completely independently.

Quoted 11 lines
Quoted 6 lines

To me it seems redundant to have the driver encode PHYs-per-SoC info, when the per-SoC DT is going to have the same info, so my suggestion was to simplify the driver and have this kind of hardware description in the DT, and keep the driver simple, but we are definitely open to learning the "right way" of doing this.

The property then is reasonable. It should not be bool, though, because it does not scale. There can be next block which supports only D-PHY on CSI0 and C-PHY on CSI1? Maybe some enum or list, depending on possible configurations.

OK, looks like include/dt-bindings/phy/phy.y already has

  #define PHY_TYPE_DPHY		10
  #define PHY_TYPE_CPHY		11
Quoted 8 lines
Quoted 1 line

we'll add a PHY_TYPE_CDPHY and use that. Sound reasonable?

Yes. Currently it is usually used as phy-cells argument (after the phy number/lane/ID), but cdns,phy-type and intel,phy-mode use it directly as property in provider. In both cases they have a bit different meaning than yours. You want to list all supported modes or narrow/restrict them. Maybe hisilicon,fixed-mode fits your purpose?

Hi Krzysztof,

Thanks for the suggestion, using something like hisilicon,fixed-node looks like a good fit.
With mediatek,fixed-node, by default CSI node will be considered as CD-PHY capable (unless the fixed-mode property is set.) so I won't need anymore the new define PHY_TYPE_CDPHY introduced in v3.

Also introducing mediatek,fixed-mode suggests that PHYs not declaring the fixed mode property support mode selection, so I suggest to add a phy argument (#phy-cells = <1>) to select the mode (D or C mode). Exactly what is done by the hsilicon driver.

How does that sound to you?

Best,
Julien

Quoted 3 lines

Best regards,
Krzysztof

Signature · 3 lines
linux-phy mailing list
[email protected]
https://lists.infradead.org/mailman/listinfo/linux-phy
maintainer
#13 lore raw

Re: [PATCH v2 1/2] dt-bindings: phy: add mediatek mipi csi driver v 0.5

Quoted 66 lines · Julien Stephan
Quoted 65 lines · Krzysztof Kozlowski
Quoted 64 lines · Kevin Hilman
Quoted 39 lines · Krzysztof Kozlowski
Quoted 14 lines · Kevin Hilman
Quoted 9 lines

Third is to use versioned IP blocks.

The second case also would work, if it is applicable to you (you really have fallback matching all devices). Third solution depends on your versioning and Rob expressed dislike about it many times.

We had many discussions on mailing lists, thus simplifying the review - I recommend the first choice. For a better recommendation you should say a bit more about the block in different SoCs.

I'll try to say a bit more about the PHY block, but in fact, it's not just about differences between SoCs. On the same SoC, 2 different PHYs may have different features/capabilities.

Quoted 8 lines

For example, on MT8365, There are 2 PHYs: CSI0 and CSI1. CSI0 can function as a C-PHY or a D-PHY, but CSI1 can only function as D-PHY (used as the example in the binding patch[1].) On another related SoC, there are 3 PHYs, where CSI0 is C-D but CSI1 & CSI2 are only D.

So that's why it seems (at least to me) that while we need SoC compatible, it's not enough. We also need properties to describe PHY-specific features (e.g. C-D PHY)

I recall the same or very similar case... It bugs me now, but unfortunately I cannot find it.

Quoted 7 lines

Of course, we could rely only on SoC-specific compatibles describe this. But then driver will need an SoC-specific table with the number of PHYs and per-PHY features for each SoC encoded in the driver. Since the driver otherwise doesn't (and shouldn't, IMHO) need to know how many PHYs are on each SoC, I suggested to Julien that perhaps the additional propery was the better solution.

Phys were modeled as separate device instances, so you would need difference in compatible to figure out which phy is it.

Quoted 7 lines

Other way could be to create device for all phys and use phy-cells=1. Whether it makes sense, depends on the actual datasheet - maybe the split phy per device is artificial? There is one PHY block with two address ranges for each PHY - CSI0 and CSI1 - but it is actually one block? You should carefully check this because once design is chosen, you won't be able to go back to other and it might be a problem (e.g. there is some top-level block for powering on all CSI instances).

We're pretty sure these are multiple instances of the IP block as they can operate completely independently.

Quoted 11 lines
Quoted 6 lines

To me it seems redundant to have the driver encode PHYs-per-SoC info, when the per-SoC DT is going to have the same info, so my suggestion was to simplify the driver and have this kind of hardware description in the DT, and keep the driver simple, but we are definitely open to learning the "right way" of doing this.

The property then is reasonable. It should not be bool, though, because it does not scale. There can be next block which supports only D-PHY on CSI0 and C-PHY on CSI1? Maybe some enum or list, depending on possible configurations.

Quoted 6 lines
Quoted 6 lines
Quoted 6 lines

OK, looks like include/dt-bindings/phy/phy.y already has

  #define PHY_TYPE_DPHY		10
  #define PHY_TYPE_CPHY		11

we'll add a PHY_TYPE_CDPHY and use that. Sound reasonable?

No, because these defines are the 1 mode used for a specific connection, not the set of supported modes.

Quoted 6 lines
Quoted 6 lines

Yes. Currently it is usually used as phy-cells argument (after the phy number/lane/ID), but cdns,phy-type and intel,phy-mode use it directly as property in provider. In both cases they have a bit different meaning than yours. You want to list all supported modes or narrow/restrict them. Maybe hisilicon,fixed-mode fits your purpose?

There is also 'phy-type' to set the mode/type in the provider.

If we need to list all possible modes, then you need to justify why that's needed and then define a property which is a mask of the existing type defines.

Quoted 10 lines
Quoted 1 line

Hi Krzysztof,

Thanks for the suggestion, using something like hisilicon,fixed-node looks like a good fit.
With mediatek,fixed-node, by default CSI node will be considered as CD-PHY capable (unless the fixed-mode property is set.) so I won't need anymore the new define PHY_TYPE_CDPHY introduced in v3.

Quoted 6 lines

Also introducing mediatek,fixed-mode suggests that PHYs not declaring the fixed mode property support mode selection, so I suggest to add a phy argument (#phy-cells = <1>) to select the mode (D or C mode). Exactly what is done by the hsilicon driver.

How does that sound to you?

I don't follow the need for fixed-mode, but agree you should use a phy arg. Can't you just assume the D-PHY only instance will only have D-PHY for the arg?

Rob

#14 lore raw

Re: [PATCH v2 1/2] dt-bindings: phy: add mediatek mipi csi driver v 0.5

In reply to Julien Stephan
Quoted 93 lines · Julien Stephan

External email : Please do not click links or open attachments until you have verified the sender or the content.

Quoted 12 lines · Krzysztof Kozlowski
Quoted 8 lines · Julien Stephan

From: Florian Sylvestre <[email protected]>

This adds the bindings, for the MIPI CD-PHY module v 0.5 embedded in
some Mediatek soc, such as the mt8365

What are the changes? IOW: changelog here or in cover letter.

Hi Krzysztof,
I added a changelog in the cover letter, but I will try to be more descritpive next time. Changes from v1 are mainly style issues fixed (mostly from your first review)

Quoted 62 lines

Subject: you have some multiple spaces.

Subject: drop driver. Bindings are not for drivers.

Quoted 39 lines
---
 .../phy/mediatek,phy-mipi-csi-0-5.yaml        | 62
+++++++++++++++++++
 MAINTAINERS                                   |  6 ++
 2 files changed, 68 insertions(+)
 create mode 100644

Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0- 5.yaml

(quoted diff) +0 -0

mipi-csi-0-5.yaml b/Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0- 5.yaml
new file mode 100644
index 000000000000..5aa8c0b41cdf

Documentation/devicetree/bindings/phy/mediatek,phy-mipi- new file +0 -0

csi-0-5.yaml

Documentation/devicetree/bindings/phy/mediatek,phy-mipi- new file +4 -0
@@ -0,0 +1,62 @@
+# SPDX-License-Identifier: (GPL-2.0-Only OR BSD-2-Clause)
+%YAML 1.2
+---
+$id: 

https://urldefense.com/v3/__http://devicetree.org/schemas/phy/mediatek,phy-mipi-csi-0-5.yaml*__;Iw!!CTRNKA9wMg0ARbw!kUABjbQO6-_jrANKiqoLPxZkh_0LAplQwlgCG9lhmCKm2ec8pCItYUB3-zQ2PGBnlAcgsBhwrOgi5qeilrddZ5s$
+$schema:
https://urldefense.com/v3/__http://devicetree.org/meta-schemas/core.yaml*__;Iw!!CTRNKA9wMg0ARbw!kUABjbQO6-_jrANKiqoLPxZkh_0LAplQwlgCG9lhmCKm2ec8pCItYUB3-zQ2PGBnlAcgsBhwrOgi5qeiABoSu-M$

Documentation/devicetree/bindings/phy/mediatek,phy-mipi- new file +9 -0
+
+title: Mediatek Sensor Interface MIPI CSI CD-PHY
+
+maintainers:
+  - Julien Stephan <[email protected]>
+  - Andy Hsieh <[email protected]>
+
+description:
+  The SENINF CD-PHY is a set of CD-PHY connected to the SENINF

CSI-2

Documentation/devicetree/bindings/phy/mediatek,phy-mipi- new file +2 -0
+  receivers. The number of PHYs depends on the SoC model.
+  Depending on the soc model, each PHYs can support CDPHY or

DPHY only

Documentation/devicetree/bindings/phy/mediatek,phy-mipi-

Quoted 5 lines
Documentation/devicetree/bindings/phy/mediatek,phy-mipi- new file +5 -0
+
+properties:
+  compatible:
+    enum:
+      - mediatek,phy-mipi-csi-0-5

SoC based compatibles. 0-5 is odd.

Documentation/devicetree/bindings/phy/mediatek,phy-mipi-

Quoted 8 lines
Documentation/devicetree/bindings/phy/mediatek,phy-mipi- new file +8 -0
+
+  reg:
+    maxItems: 1
+
+  '#phy-cells':
+    const: 0
+
+  mediatek,is_cdphy:

No underscores in node names.

Quoted 7 lines

Documentation/devicetree/bindings/phy/mediatek,phy-mipi-

Quoted 2 lines
Documentation/devicetree/bindings/phy/mediatek,phy-mipi- new file +2 -0
+    description:
+      Specify if the current phy support CDPHY configuration

Why this cannot be implied from compatible? Add specific compatibles.

Quoted 5 lines

This cannot be implied by compatible because the number of phys depends
on the soc and each phy can be either D-PHY only or CD-PHY capable. For example mt8365 has 2 phy: CSI0 and CSI1. CSI1 is DPHY only and CSI0 is CD-PHY

Do the user need to know which type phy he will use?

Quoted 43 lines

Regards,
Julien

Quoted 40 lines
Quoted 6 lines
Documentation/devicetree/bindings/phy/mediatek,phy-mipi- new file +6 -0
+    type: boolean
+
+required:
+  - compatible
+  - reg
+  - '#phy-cells'

Documentation/devicetree/bindings/phy/mediatek,phy-mipi-

Quoted 10 lines
Documentation/devicetree/bindings/phy/mediatek,phy-mipi- new file +10 -0
+
+additionalProperties: false
+
+examples:
+  - |
+    soc {
+      #address-cells = <2>;
+      #size-cells = <2>;
+
+      mipi_rx_csi0: mipi_rx_csi0@11c10000 {

No underscores in node names. How this is v2?

Documentation/devicetree/bindings/phy/mediatek,phy-mipi-

Quoted 3 lines
Documentation/devicetree/bindings/phy/mediatek,phy-mipi- new file +3 -0
+        compatible = "mediatek,phy-mipi-csi-0-5";
+        reg = <0 0x11C10000 0 0x2000>;
+        status = "okay";

Drop

Documentation/devicetree/bindings/phy/mediatek,phy-mipi-

Quoted 8 lines
Documentation/devicetree/bindings/phy/mediatek,phy-mipi- new file +8 -0
+        mediatek,is_cdphy;
+        #phy-cells = <0>;
+      };
+
+      mipi_rx_csi1: mipi-rx-csi1@11c12000 {
+        compatible = "mediatek,phy-mipi-csi-0-5";
+        reg = <0 0x11C12000 0 0x2000>;
+        status = "disabled";

???

Best regards,
Krzysztof

author maintainer
#15 lore raw

PATCH v2 2/2 phy: mtk-mipi-csi: add driver for CSI phy

In reply to Julien Stephan

From: Phi-bang Nguyen <[email protected]>

This is a new driver that supports the MIPI CSI CD-PHY version 0.5

The number of PHYs depend on the soc.

[Julien Stephan: use GENMASK]
[Julien Stephan: refactor code]
[Julien Stephan: update device tree support and probe function]
---
 MAINTAINERS                                   |   1 +
 drivers/phy/mediatek/Kconfig                  |   8 +
 drivers/phy/mediatek/Makefile                 |   2 +
 .../mediatek/phy-mtk-mipi-csi-0-5-rx-reg.h    |  58 ++++
 drivers/phy/mediatek/phy-mtk-mipi-csi-0-5.c   | 257 ++++++++++++++++++
 5 files changed, 326 insertions(+)
 create mode 100644 drivers/phy/mediatek/phy-mtk-mipi-csi-0-5-rx-reg.h
 create mode 100644 drivers/phy/mediatek/phy-mtk-mipi-csi-0-5.c
MAINTAINERS +1 -0
index 44f0ff11e984..fc2766cb50d3 100644
@@ -13103,6 +13103,7 @@ M:	Julien Stephan <[email protected]>
 M:	Andy Hsieh <[email protected]>
 S:	Supported
 F:	Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml
+F:	drivers/phy/mediatek/phy-mtk-mipi-csi-0-5*
 
 MEDIATEK MMC/SD/SDIO DRIVER
 M:	Chaotian Jing <[email protected]>
drivers/phy/mediatek/Kconfig +8 -0
index 3125ecb5d119..452bc7ac5ce5 100644
@@ -74,3 +74,11 @@ config PHY_MTK_DP
 	select GENERIC_PHY
 	help
 	  Support DisplayPort PHY for MediaTek SoCs.
+
+config PHY_MTK_MIPI_CSI_0_5
+	tristate "MediaTek CSI CD-PHY v 0.5 Driver"
+	depends on ARCH_MEDIATEK && OF
+	select GENERIC_PHY
+	help
+	  Enable this to support the MIPI CSI CD-PHY receiver version 0.5.
+	  The driver supports multiple CSI cdphy ports simultaneously.
drivers/phy/mediatek/Makefile +2 -0
index fb1f8edaffa7..8eb7b8747c67 100644
@@ -18,3 +18,5 @@ phy-mtk-mipi-dsi-drv-y			:= phy-mtk-mipi-dsi.o
 phy-mtk-mipi-dsi-drv-y			+= phy-mtk-mipi-dsi-mt8173.o
 phy-mtk-mipi-dsi-drv-y			+= phy-mtk-mipi-dsi-mt8183.o
 obj-$(CONFIG_PHY_MTK_MIPI_DSI)		+= phy-mtk-mipi-dsi-drv.o
+
+obj-$(CONFIG_PHY_MTK_MIPI_CSI_0_5)	+= phy-mtk-mipi-csi-0-5.o
drivers/phy/mediatek/phy-mtk-mipi-csi-0-5-rx-reg.h new file +58 -0

58 lines not on this page · raw message

drivers/phy/mediatek/phy-mtk-mipi-csi-0-5.c new file +257 -0

257 lines not on this page · raw message

Signature · 7 lines
2.40.0


-- 
linux-phy mailing list
[email protected]
https://lists.infradead.org/mailman/listinfo/linux-phy
#16 lore raw

Il 15/05/23 11:05, Julien Stephan ha scritto:

Quoted 40 lines

From: Phi-bang Nguyen <[email protected]>

This is a new driver that supports the MIPI CSI CD-PHY version 0.5

The number of PHYs depend on the soc.

[Julien Stephan: use GENMASK]
[Julien Stephan: refactor code]
[Julien Stephan: update device tree support and probe function]
---
  MAINTAINERS                                   |   1 +
  drivers/phy/mediatek/Kconfig                  |   8 +
  drivers/phy/mediatek/Makefile                 |   2 +
  .../mediatek/phy-mtk-mipi-csi-0-5-rx-reg.h    |  58 ++++
  drivers/phy/mediatek/phy-mtk-mipi-csi-0-5.c   | 257 ++++++++++++++++++
  5 files changed, 326 insertions(+)

create mode 100644 drivers/phy/mediatek/phy-mtk-mipi-csi-0-5-rx-reg.h create mode 100644 drivers/phy/mediatek/phy-mtk-mipi-csi-0-5.c

MAINTAINERS +1 -0
index 44f0ff11e984..fc2766cb50d3 100644
@@ -13103,6 +13103,7 @@ M:	Julien Stephan <[email protected]>
  M:	Andy Hsieh <[email protected]>
  S:	Supported
  F:	Documentation/devicetree/bindings/phy/mediatek,phy-mipi-csi-0-5.yaml
+F:	drivers/phy/mediatek/phy-mtk-mipi-csi-0-5*
 
  MEDIATEK MMC/SD/SDIO DRIVER
  M:	Chaotian Jing <[email protected]>
drivers/phy/mediatek/Kconfig +0 -0
index 3125ecb5d119..452bc7ac5ce5 100644

drivers/phy/mediatek/Kconfig config PHY_MTK_DP · line 79

Quoted 6 lines
drivers/phy/mediatek/Kconfig +3 -0
@@ -74,3 +74,11 @@ config PHY_MTK_DP
  	select GENERIC_PHY
  	help
  	  Support DisplayPort PHY for MediaTek SoCs.
+
+config PHY_MTK_MIPI_CSI_0_5
+	tristate "MediaTek CSI CD-PHY v 0.5 Driver"

"MediaTek CSI CD-PHY v0.5 Driver"

Quoted 21 lines
drivers/phy/mediatek/Kconfig +5 -0
+	depends on ARCH_MEDIATEK && OF
+	select GENERIC_PHY
+	help
+	  Enable this to support the MIPI CSI CD-PHY receiver version 0.5.
+	  The driver supports multiple CSI cdphy ports simultaneously.
drivers/phy/mediatek/Makefile +2 -0
index fb1f8edaffa7..8eb7b8747c67 100644
@@ -18,3 +18,5 @@ phy-mtk-mipi-dsi-drv-y			:= phy-mtk-mipi-dsi.o
  phy-mtk-mipi-dsi-drv-y			+= phy-mtk-mipi-dsi-mt8173.o
  phy-mtk-mipi-dsi-drv-y			+= phy-mtk-mipi-dsi-mt8183.o
  obj-$(CONFIG_PHY_MTK_MIPI_DSI)		+= phy-mtk-mipi-dsi-drv.o
+
+obj-$(CONFIG_PHY_MTK_MIPI_CSI_0_5)	+= phy-mtk-mipi-csi-0-5.o
drivers/phy/mediatek/phy-mtk-mipi-csi-0-5-rx-reg.h new file +0 -0
new file mode 100644
index 000000000000..e9a7f1ab3e2f

drivers/phy/mediatek/phy-mtk-mipi-csi-0-5-rx-reg.h line 4

Quoted 4 lines
drivers/phy/mediatek/phy-mtk-mipi-csi-0-5-rx-reg.h new file +4 -0
@@ -0,0 +1,58 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+
+#ifndef __PHY_MTK__MIPI_CSI__C_0_5_RX_REG_H__
+#define __PHY_MTK__MIPI_CSI__C_0_5_RX_REG_H__

What about....

__PHY_MTK_MIPI_CSI_V_0_5_RX_REG_H ?

drivers/phy/mediatek/phy-mtk-mipi-csi-0-5-rx-reg.h line 13

Quoted 9 lines
drivers/phy/mediatek/phy-mtk-mipi-csi-0-5-rx-reg.h new file +9 -0
+
+/*
+ * CSI1 and CSI2 are identical, and similar to CSI0. All CSIx macros are
+ * applicable to the three PHYs. Where differences exist, they are denoted by
+ * macro names using CSI0 and CSI1, the latter being applicable to CSI1 and
+ * CSI2 alike.
+ */
+
+#define MIPI_RX_ANA00_CSIxA			0x0000

I would rename all those from "CSIx" to "CSIX" (so, just toupper('x')).

Quoted 1 line
drivers/phy/mediatek/phy-mtk-mipi-csi-0-5-rx-reg.h new file +1 -0
+#define RG_CSI0A_CPHY_EN			BIT(0)

..snip..

Quoted 13 lines
drivers/phy/mediatek/phy-mtk-mipi-csi-0-5-rx-reg.h new file +2 -0
+
+#endif
drivers/phy/mediatek/phy-mtk-mipi-csi-0-5.c new file +5 -0
new file mode 100644
index 000000000000..ae2d3dc9631d
@@ -0,0 +1,257 @@
+// SPDX-License-Identifier: GPL-2.0
+
+#include <linux/bitfield.h>
+#include <linux/delay.h>
+#include <linux/io.h>

drivers/phy/mediatek/phy-mtk-mipi-csi-0-5.c line 15

Quoted 10 lines
drivers/phy/mediatek/phy-mtk-mipi-csi-0-5.c new file +10 -0
+#include <linux/module.h>
+#include <linux/mutex.h>
+#include <linux/phy/phy.h>
+#include <linux/platform_device.h>
+#include <linux/slab.h>
+
+#include "phy-mtk-io.h"
+#include "phy-mtk-mipi-csi-0-5-rx-reg.h"
+
+#define CSIxB_OFFSET		0x1000

What if we grab two (or three?) iospaces from devicetree?

- base (global)
- csi_a
- csi_b

That would make it possible to maybe eventually extend this driver to more versions (older or newer) of the CSI PHY IP without putting fixes offsets inside of platform data structures and such.

Quoted 70 lines
drivers/phy/mediatek/phy-mtk-mipi-csi-0-5.c new file +70 -0
+
+struct mtk_mipi_dphy;
+
+struct mtk_mipi_dphy_port {
+	struct device *dev;
+	void __iomem *base;
+	struct phy *phy;
+	bool is_cdphy;
+};
+
+static int mtk_mipi_phy_power_on(struct phy *phy)
+{
+	struct mtk_mipi_dphy_port *port = phy_get_drvdata(phy);
+	void __iomem *base = port->base;
+
+	/*
+	 * Only DPHY is supported for now,
+	 * so set analog phy mode to DPHY in CDPHY compatible PHYs
+	 */
+	if (port->is_cdphy) {
+		mtk_phy_update_field(base + MIPI_RX_ANA00_CSIxA,
+				     RG_CSI0A_CPHY_EN, 0);
+		mtk_phy_update_field(base + CSIxB_OFFSET + MIPI_RX_ANA00_CSIxA,
+				     RG_CSI0A_CPHY_EN, 0);
+	}
+
+	/*
+	 * Lane configuration:
+	 *
+	 * Only 4 data + 1 clock is supported for now with the following mapping:
+	 *
+	 * CSIxA_LNR0 --> D2
+	 * CSIxA_LNR1 --> D0
+	 * CSIxA_LNR2 --> C
+	 * CSIxB_LNR0 --> D1
+	 * CSIxB_LNR1 --> D3
+	 */
+	mtk_phy_update_field(base + MIPI_RX_ANA00_CSIxA,
+			     RG_CSIxA_DPHY_L0_CKMODE_EN, 0);
+	mtk_phy_update_field(base + MIPI_RX_ANA00_CSIxA,
+			     RG_CSIxA_DPHY_L0_CKSEL, 1);
+	mtk_phy_update_field(base + MIPI_RX_ANA00_CSIxA,
+			     RG_CSIxA_DPHY_L1_CKMODE_EN, 0);
+	mtk_phy_update_field(base + MIPI_RX_ANA00_CSIxA,
+			     RG_CSIxA_DPHY_L1_CKSEL, 1);
+	mtk_phy_update_field(base + MIPI_RX_ANA00_CSIxA,
+			     RG_CSIxA_DPHY_L2_CKMODE_EN, 1);
+	mtk_phy_update_field(base + MIPI_RX_ANA00_CSIxA,
+			     RG_CSIxA_DPHY_L2_CKSEL, 1);
+
+	mtk_phy_update_field(base + CSIxB_OFFSET + MIPI_RX_ANA00_CSIxA,
+			     RG_CSIxA_DPHY_L0_CKMODE_EN, 0);
+	mtk_phy_update_field(base + CSIxB_OFFSET + MIPI_RX_ANA00_CSIxA,
+			     RG_CSIxA_DPHY_L0_CKSEL, 1);
+	mtk_phy_update_field(base + CSIxB_OFFSET + MIPI_RX_ANA00_CSIxA,
+			     RG_CSIxA_DPHY_L1_CKMODE_EN, 0);
+	mtk_phy_update_field(base + CSIxB_OFFSET + MIPI_RX_ANA00_CSIxA,
+			     RG_CSIxA_DPHY_L1_CKSEL, 1);
+	mtk_phy_update_field(base + CSIxB_OFFSET + MIPI_RX_ANA00_CSIxA,
+			     RG_CSIxA_DPHY_L2_CKMODE_EN, 0);
+	mtk_phy_update_field(base + CSIxB_OFFSET + MIPI_RX_ANA00_CSIxA,
+			     RG_CSIxA_DPHY_L2_CKSEL, 1);
+
+	/* Byte clock invert */
+	mtk_phy_update_field(base + MIPI_RX_ANAA8_CSIxA,
+			     RG_CSIxA_CDPHY_L0_T0_BYTECK_INVERT, 1);
+	mtk_phy_update_field(base + MIPI_RX_ANAA8_CSIxA,
+			     RG_CSIxA_DPHY_L1_BYTECK_INVERT, 1);
+	mtk_phy_update_field(base + MIPI_RX_ANAA8_CSIxA,
+			     RG_CSIxA_CDPHY_L2_T1_BYTECK_INVERT, 1);

drivers/phy/mediatek/phy-mtk-mipi-csi-0-5.c line 95

Quoted 10 lines
drivers/phy/mediatek/phy-mtk-mipi-csi-0-5.c new file +10 -0
+
+	mtk_phy_update_field(base + CSIxB_OFFSET + MIPI_RX_ANAA8_CSIxA,
+			     RG_CSIxA_CDPHY_L0_T0_BYTECK_INVERT, 1);
+	mtk_phy_update_field(base + CSIxB_OFFSET + MIPI_RX_ANAA8_CSIxA,
+			     RG_CSIxA_DPHY_L1_BYTECK_INVERT, 1);
+	mtk_phy_update_field(base + CSIxB_OFFSET + MIPI_RX_ANAA8_CSIxA,
+			     RG_CSIxA_CDPHY_L2_T1_BYTECK_INVERT, 1);
+
+	/* Start ANA EQ tuning */
+	if (port->is_cdphy) {

statid void mtk_phy_csi_analog_eq_tune(struct mtk_mipi_dphy_port *port)

{
	if (port->is_cdphy)
		mtk_phy_csi_dphy_ana_eq_tune(...)
	else
		mtk_phy_csi_cphy_ana_eq_tune(...)

	/* CPHY/DPHY common "end of tuning" sequence below */
	mtk_phy_update_field( ... stuff ...);
}

...then all those calls will also fit in one line due to the reduced indentation, improving readability and reducing line count.

Quoted 113 lines
drivers/phy/mediatek/phy-mtk-mipi-csi-0-5.c new file +113 -0
+		mtk_phy_update_field(base + MIPI_RX_ANA18_CSIxA,
+				     RG_CSI0A_L0_T0AB_EQ_IS, 1);
+		mtk_phy_update_field(base + MIPI_RX_ANA18_CSIxA,
+				     RG_CSI0A_L0_T0AB_EQ_BW, 1);
+		mtk_phy_update_field(base + MIPI_RX_ANA1C_CSIxA,
+				     RG_CSI0A_L1_T1AB_EQ_IS, 1);
+		mtk_phy_update_field(base + MIPI_RX_ANA1C_CSIxA,
+				     RG_CSI0A_L1_T1AB_EQ_BW, 1);
+		mtk_phy_update_field(base + MIPI_RX_ANA20_CSI0A,
+				     RG_CSI0A_L2_T1BC_EQ_IS, 1);
+		mtk_phy_update_field(base + MIPI_RX_ANA20_CSI0A,
+				     RG_CSI0A_L2_T1BC_EQ_BW, 1);
+
+		mtk_phy_update_field(base + CSIxB_OFFSET + MIPI_RX_ANA18_CSIxA,
+				     RG_CSI0A_L0_T0AB_EQ_IS, 1);
+		mtk_phy_update_field(base + CSIxB_OFFSET + MIPI_RX_ANA18_CSIxA,
+				     RG_CSI0A_L0_T0AB_EQ_BW, 1);
+		mtk_phy_update_field(base + CSIxB_OFFSET + MIPI_RX_ANA1C_CSIxA,
+				     RG_CSI0A_L1_T1AB_EQ_IS, 1);
+		mtk_phy_update_field(base + CSIxB_OFFSET + MIPI_RX_ANA1C_CSIxA,
+				     RG_CSI0A_L1_T1AB_EQ_BW, 1);
+		mtk_phy_update_field(base + CSIxB_OFFSET + MIPI_RX_ANA20_CSI0A,
+				     RG_CSI0A_L2_T1BC_EQ_IS, 1);
+		mtk_phy_update_field(base + CSIxB_OFFSET + MIPI_RX_ANA20_CSI0A,
+				     RG_CSI0A_L2_T1BC_EQ_BW, 1);
+	} else {
+		mtk_phy_update_field(base + MIPI_RX_ANA18_CSIxA,
+				     RG_CSI1A_L0_EQ_IS, 1);
+		mtk_phy_update_field(base + MIPI_RX_ANA18_CSIxA,
+				     RG_CSI1A_L0_EQ_BW, 1);
+		mtk_phy_update_field(base + MIPI_RX_ANA18_CSIxA,
+				     RG_CSI1A_L1_EQ_IS, 1);
+		mtk_phy_update_field(base + MIPI_RX_ANA18_CSIxA,
+				     RG_CSI1A_L1_EQ_BW, 1);
+		mtk_phy_update_field(base + MIPI_RX_ANA1C_CSIxA,
+				     RG_CSI1A_L2_EQ_IS, 1);
+		mtk_phy_update_field(base + MIPI_RX_ANA1C_CSIxA,
+				     RG_CSI1A_L2_EQ_BW, 1);
+
+		mtk_phy_update_field(base + CSIxB_OFFSET + MIPI_RX_ANA18_CSIxA,
+				     RG_CSI1A_L0_EQ_IS, 1);
+		mtk_phy_update_field(base + CSIxB_OFFSET + MIPI_RX_ANA18_CSIxA,
+				     RG_CSI1A_L0_EQ_BW, 1);
+		mtk_phy_update_field(base + CSIxB_OFFSET + MIPI_RX_ANA18_CSIxA,
+				     RG_CSI1A_L1_EQ_IS, 1);
+		mtk_phy_update_field(base + CSIxB_OFFSET + MIPI_RX_ANA18_CSIxA,
+				     RG_CSI1A_L1_EQ_BW, 1);
+		mtk_phy_update_field(base + CSIxB_OFFSET + MIPI_RX_ANA1C_CSIxA,
+				     RG_CSI1A_L2_EQ_IS, 1);
+		mtk_phy_update_field(base + CSIxB_OFFSET + MIPI_RX_ANA1C_CSIxA,
+				     RG_CSI1A_L2_EQ_BW, 1);
+	}
+
+	/* End ANA EQ tuning */
+	mtk_phy_set_bits(base + MIPI_RX_ANA40_CSIxA, 0x90);
+
+	mtk_phy_update_field(base + MIPI_RX_ANA24_CSIxA,
+			     RG_CSIxA_RESERVE, 0x40);
+	mtk_phy_update_field(base + CSIxB_OFFSET + MIPI_RX_ANA24_CSIxA,
+			     RG_CSIxA_RESERVE, 0x40);
+	mtk_phy_update_field(base + MIPI_RX_WRAPPER80_CSIxA,
+			     CSR_CSI_RST_MODE, 0);
+	mtk_phy_update_field(base + CSIxB_OFFSET + MIPI_RX_WRAPPER80_CSIxA,
+			     CSR_CSI_RST_MODE, 0);
+	/* ANA power on */
+	mtk_phy_update_field(base + MIPI_RX_ANA00_CSIxA,
+			     RG_CSIxA_BG_CORE_EN, 1);
+	mtk_phy_update_field(base + CSIxB_OFFSET + MIPI_RX_ANA00_CSIxA,
+			     RG_CSIxA_BG_CORE_EN, 1);
+	usleep_range(20, 40);
+	mtk_phy_update_field(base + MIPI_RX_ANA00_CSIxA,
+			     RG_CSIxA_BG_LPF_EN, 1);
+	mtk_phy_update_field(base + CSIxB_OFFSET + MIPI_RX_ANA00_CSIxA,
+			     RG_CSIxA_BG_LPF_EN, 1);
+
+	return 0;
+}
+
+static int mtk_mipi_phy_power_off(struct phy *phy)
+{
+	struct mtk_mipi_dphy_port *port = phy_get_drvdata(phy);
+	void __iomem *base = port->base;
+
+	/* Disable MIPI BG. */
+	mtk_phy_update_field(base + MIPI_RX_ANA00_CSIxA,
+			     RG_CSIxA_BG_CORE_EN, 0);
+	mtk_phy_update_field(base + MIPI_RX_ANA00_CSIxA,
+			     RG_CSIxA_BG_LPF_EN, 0);
+
+	mtk_phy_update_field(base + CSIxB_OFFSET + MIPI_RX_ANA00_CSIxA,
+			     RG_CSIxA_BG_CORE_EN, 0);
+	mtk_phy_update_field(base + CSIxB_OFFSET + MIPI_RX_ANA00_CSIxA,
+			     RG_CSIxA_BG_LPF_EN, 0);
+
+	return 0;
+}
+
+static const struct phy_ops mtk_dphy_ops = {
+	.power_on	= mtk_mipi_phy_power_on,
+	.power_off	= mtk_mipi_phy_power_off,
+	.owner		= THIS_MODULE,
+};
+
+static int mtk_mipi_dphy_probe(struct platform_device *pdev)
+{
+	struct device *dev = &pdev->dev;
+	struct phy_provider *phy_provider;
+	struct mtk_mipi_dphy_port *port;
+	struct phy *phy;
+
+	port = devm_kzalloc(dev, sizeof(*port), GFP_KERNEL);
+	if (!port)
+		return -ENOMEM;

drivers/phy/mediatek/phy-mtk-mipi-csi-0-5.c line 218

Quoted 10 lines
drivers/phy/mediatek/phy-mtk-mipi-csi-0-5.c new file +10 -0
+
+	dev_set_drvdata(dev, port);
+
+	port->dev = dev;
+
+	port->base = devm_platform_ioremap_resource(pdev, 0);
+	if (IS_ERR(port->base))
+		return PTR_ERR(port->base);
+
+	port->is_cdphy = of_property_read_bool(dev->of_node, "mediatek,is_cdphy");

This driver doesn't support C-PHY mode, so you either add support for that, or in my opinion you should simply refuse to probe it, as it is *dysfunctional* for the unsupported case (and might even introduce unstabilities).

	/* At the moment, only D-PHY mode is supported */
	if (!port->is_cdphy)
		return -EINVAL;

Also, please don't use underscores for devicetree properties: "mediatek,is-cdphy" is fine.

Quoted 12 lines
drivers/phy/mediatek/phy-mtk-mipi-csi-0-5.c new file +12 -0
+
+	phy = devm_phy_create(dev, NULL, &mtk_dphy_ops);
+	if (IS_ERR(phy)) {
+		dev_err(dev, "Failed to create PHY: %ld\n", PTR_ERR(phy));
+		return PTR_ERR(phy);
+	}
+
+	port->phy = phy;
+	phy_set_drvdata(phy, port);
+
+	phy_provider = devm_of_phy_provider_register(dev, of_phy_simple_xlate);
+	if (IS_ERR(phy_provider)) {

drivers/phy/mediatek/phy-mtk-mipi-csi-0-5.c line 240

Quoted 10 lines
drivers/phy/mediatek/phy-mtk-mipi-csi-0-5.c new file +10 -0
+		dev_err(dev, "Failed to register PHY provider: %ld\n",
+			PTR_ERR(phy_provider));
+		return PTR_ERR(phy_provider);
+	}
+
+	return 0;
+}
+
+static const struct of_device_id mtk_mipi_dphy_of_match[] = {
+	{.compatible = "mediatek,phy-mipi-csi-0-5"},

leave spaces.

	{ .comp... " },

...and always end with

	{ /* sentinel */ }

Also, please follow what the other PHY drivers do and use a SoC model, example:

"mediatek,mt7777-csi-phy", or "mediatek,mt8888-csi-rx"

where the latter would make more sense imo.

drivers/phy/mediatek/phy-mtk-mipi-csi-0-5.c line 252

Quoted 12 lines
drivers/phy/mediatek/phy-mtk-mipi-csi-0-5.c new file +12 -0
+	{},
+};
+MODULE_DEVICE_TABLE(of, mtk_mipi_dphy_of_match);
+
+static struct platform_driver mipi_dphy_pdrv = {
+	.probe = mtk_mipi_dphy_probe,
+	.driver	= {
+		.name	= "mtk-mipi-csi-0-5",
+		.of_match_table = mtk_mipi_dphy_of_match,
+	},
+};
+

remove extra blank line here.

drivers/phy/mediatek/phy-mtk-mipi-csi-0-5.c line 255

Quoted 3 lines
drivers/phy/mediatek/phy-mtk-mipi-csi-0-5.c new file +3 -0
+module_platform_driver(mipi_dphy_pdrv);
+
+MODULE_DESCRIPTION("MTK mipi csi cdphy driver");

"MediaTek MIPI CSI CDPHY Driver"

drivers/phy/mediatek/phy-mtk-mipi-csi-0-5.c line 257

Quoted 2 lines
drivers/phy/mediatek/phy-mtk-mipi-csi-0-5.c new file +2 -0
+MODULE_AUTHOR("Louis Kuo <[email protected]>");
+MODULE_LICENSE("GPL");

Regards,
Angelo

author maintainer
#17 lore raw
Quoted 1 line · AngeloGioacchino Del Regno

Il 15/05/23 11:05, Julien Stephan ha scritto:

..snip..

Quoted 6 lines

drivers/phy/mediatek/phy-mtk-mipi-csi-0-5.c line 218

Quoted 1 line
drivers/phy/mediatek/phy-mtk-mipi-csi-0-5.c new file +1 -0
+	port->is_cdphy = of_property_read_bool(dev->of_node, "mediatek,is_cdphy");

This driver doesn't support C-PHY mode, so you either add support for that, or in my opinion you should simply refuse to probe it, as it is *dysfunctional* for the unsupported case (and might even introduce unstabilities).

Quoted 7 lines
	/* At the moment, only D-PHY mode is supported */
	if (!port->is_cdphy)
		return -EINVAL;

Also, please don't use underscores for devicetree properties: "mediatek,is-cdphy" is fine.

Hi Angelo,
You are right this driver does not support C-PHY mode, but some of the PHYs themselves support BOTH C-PHY AND D-PHY. The idea of is_cdphy variable is to know if the CSI port supports BOTH C-PHY AND D-PHY or only DPHY. For example mt8365 has 2 PHYs: CSI0 and CSI1. CSI1 support only D-PHY, while CSI0 can be configured in C-PHY or D-PHY. Registers for CD-PHY and D-PHY are almost identical, except that CD-PHY compatible has some extra bitfields to configure properly the mode and the lanes (because supporting trios for CD-PHY).
If C-PHY support is eventually added into the driver, I think we will need another variable such as mode to know the mode. I was also thinking of adding a phy argument to determine if the mode is C-PHY or D-PHY.

So here, I don't want to stop the probe if is_cdphy variable is set to true. Does it make sense ?

Regards
Julien

.. snip..

Quoted 10 lines

drivers/phy/mediatek/phy-mtk-mipi-csi-0-5.c line 224

Quoted 6 lines
drivers/phy/mediatek/phy-mtk-mipi-csi-0-5.c new file +6 -0
+
+	phy = devm_phy_create(dev, NULL, &mtk_dphy_ops);
+	if (IS_ERR(phy)) {
+		dev_err(dev, "Failed to create PHY: %ld\n", PTR_ERR(phy));
+		return PTR_ERR(phy);
+	}

Regards,
Angelo

#18 lore raw

Il 15/05/23 15:36, Julien Stephan ha scritto:

Quoted 29 lines · AngeloGioacchino Del Regno
Quoted 1 line

Il 15/05/23 11:05, Julien Stephan ha scritto:

..snip..

Quoted 6 lines

drivers/phy/mediatek/phy-mtk-mipi-csi-0-5.c line 218

Quoted 1 line
drivers/phy/mediatek/phy-mtk-mipi-csi-0-5.c new file +1 -0
+	port->is_cdphy = of_property_read_bool(dev->of_node, "mediatek,is_cdphy");

This driver doesn't support C-PHY mode, so you either add support for that, or in my opinion you should simply refuse to probe it, as it is *dysfunctional* for the unsupported case (and might even introduce unstabilities).

Quoted 7 lines
	/* At the moment, only D-PHY mode is supported */
	if (!port->is_cdphy)
		return -EINVAL;

Also, please don't use underscores for devicetree properties: "mediatek,is-cdphy" is fine.

Hi Angelo,
You are right this driver does not support C-PHY mode, but some of the PHYs themselves support BOTH C-PHY AND D-PHY. The idea of is_cdphy variable is to know if the CSI port supports BOTH C-PHY AND D-PHY or only DPHY. For example mt8365 has 2 PHYs: CSI0 and CSI1. CSI1 support only D-PHY, while CSI0 can be configured in C-PHY or D-PHY. Registers for CD-PHY and D-PHY are almost identical, except that CD-PHY compatible has some extra bitfields to configure properly the mode and the lanes (because supporting trios for CD-PHY).
If C-PHY support is eventually added into the driver, I think we will need another variable such as mode to know the mode. I was also thinking of adding a phy argument to determine if the mode is C-PHY or D-PHY.

Quoted 3 lines

So here, I don't want to stop the probe if is_cdphy variable is set to true. Does it make sense ?

Comments in the code convinced me that the other PHYs providing only C or D PHY support weren't compatible at all with this driver.

I got it now - but at this point can you please add a comment in the code actually clarifying that this driver supports both PHYs providing *only* D-PHY and ones providing selectable C-or-D PHY?

That clarified, it would not make sense to stop probing if it's not a CDPHY because as you said there might be a D-only PHY that would be actually supported here.

Regards,
Angelo

author maintainer
#19 lore raw
Quoted 41 lines · AngeloGioacchino Del Regno

Il 15/05/23 15:36, Julien Stephan ha scritto:

Quoted 29 lines · AngeloGioacchino Del Regno
Quoted 1 line

Il 15/05/23 11:05, Julien Stephan ha scritto:

..snip..

Quoted 6 lines

drivers/phy/mediatek/phy-mtk-mipi-csi-0-5.c line 218

Quoted 1 line
drivers/phy/mediatek/phy-mtk-mipi-csi-0-5.c new file +1 -0
+	port->is_cdphy = of_property_read_bool(dev->of_node, "mediatek,is_cdphy");

This driver doesn't support C-PHY mode, so you either add support for that, or in my opinion you should simply refuse to probe it, as it is *dysfunctional* for the unsupported case (and might even introduce unstabilities).

Quoted 7 lines
	/* At the moment, only D-PHY mode is supported */
	if (!port->is_cdphy)
		return -EINVAL;

Also, please don't use underscores for devicetree properties: "mediatek,is-cdphy" is fine.

Hi Angelo,
You are right this driver does not support C-PHY mode, but some of the PHYs themselves support BOTH C-PHY AND D-PHY. The idea of is_cdphy variable is to know if the CSI port supports BOTH C-PHY AND D-PHY or only DPHY. For example mt8365 has 2 PHYs: CSI0 and CSI1. CSI1 support only D-PHY, while CSI0 can be configured in C-PHY or D-PHY. Registers for CD-PHY and D-PHY are almost identical, except that CD-PHY compatible has some extra bitfields to configure properly the mode and the lanes (because supporting trios for CD-PHY).
If C-PHY support is eventually added into the driver, I think we will need another variable such as mode to know the mode. I was also thinking of adding a phy argument to determine if the mode is C-PHY or D-PHY.

Quoted 3 lines

So here, I don't want to stop the probe if is_cdphy variable is set to true. Does it make sense ?

Comments in the code convinced me that the other PHYs providing only C or D PHY support weren't compatible at all with this driver.

I got it now - but at this point can you please add a comment in the code actually clarifying that this driver supports both PHYs providing *only* D-PHY and ones providing selectable C-or-D PHY?

Quoted 7 lines

That clarified, it would not make sense to stop probing if it's not a CDPHY because as you said there might be a D-only PHY that would be actually supported here.

Regards,
Angelo

Ok, I will add a comment in the code to make it more clear.

Regards
Julien

author maintainer
#20 lore raw
In reply to AngeloGioacchino Del Regno
Quoted 4 lines · AngeloGioacchino Del Regno
Quoted 1 line

+#define CSIxB_OFFSET 0x1000

What if we grab two (or three?) iospaces from devicetree?

Quoted 8 lines

- base (global)
- csi_a
- csi_b

That would make it possible to maybe eventually extend this driver to more versions (older or newer) of the CSI PHY IP without putting fixes offsets inside of platform data structures and such.

Hi Angelo,
The register bank of the CSI port is divided into 2:
* from base address to base + 0x1000 (port A)
* from base + 0x1000 to base +0x2000 (port B)
Some CSI port can be configured in 4D1C mode (4 data + 1 clock) using the whole register bank from base to base + 0x2000 or in 2D1C mode (2 data + 1 clock) and use either port A or port B.

For example mt8365 has CSI0 that can be used either in 4D1C mode or in 2 * 2D1C and CSI1 which can use only 4D1C mode

2D1C mode can not be tested and is not implemented in the driver so I guess adding csi_a and csi_b reg value may be confusing?

What do you think?

Regards,
Julien

#21 lore raw

Il 15/05/23 16:07, Julien Stephan ha scritto:

Quoted 21 lines · AngeloGioacchino Del Regno
Quoted 4 lines
Quoted 1 line

+#define CSIxB_OFFSET 0x1000

What if we grab two (or three?) iospaces from devicetree?

Quoted 8 lines

- base (global)
- csi_a
- csi_b

That would make it possible to maybe eventually extend this driver to more versions (older or newer) of the CSI PHY IP without putting fixes offsets inside of platform data structures and such.

Hi Angelo,
The register bank of the CSI port is divided into 2:
* from base address to base + 0x1000 (port A)
* from base + 0x1000 to base +0x2000 (port B)
Some CSI port can be configured in 4D1C mode (4 data + 1 clock) using the whole register bank from base to base + 0x2000 or in 2D1C mode (2 data + 1 clock) and use either port A or port B.

Quoted 7 lines

For example mt8365 has CSI0 that can be used either in 4D1C mode or in 2 * 2D1C and CSI1 which can use only 4D1C mode

2D1C mode can not be tested and is not implemented in the driver so I guess adding csi_a and csi_b reg value may be confusing?

What do you think?

Ok so we're talking about two data lanes per CSI port... it may still be beneficial to split the two register regions as

reg-names = "csi-a", "csi-b"; (whoops, I actually used underscores before,
and that was a mistake, sorry!)

....but that would be actually good only if we are expecting to get a CSI PHY in the future with four data lanes per port.

If you do *not* expect at all such a CSI PHY, or you do *not* expect such a PHY to ever be compatible with this driver (read as: if you expect such a PHY to be literally completely different from this one), then it would not change much to have the registers split in two.

Another case in which it would make sense is if we were to get a PHY that provides more than two CSI ports: in that case, we'd avoid platform data machinery to check the number of actual ports in the IP, as we would be just checking how many register regions we were given from the devicetree, meaning that if we got "csi-a", "csi-b", "csi-c", "csi-d", we have four ports.

Besides, another thing to think about is... yes you cannot test nor implement 2D1C mode in your submission, but this doesn't mean that others won't ever be interested in this and that other people won't be actually implementing that; Providing them with the right initial driver structure will surely make things easier, encouraging other people from the community to spend their precious time on the topic.

Regards,
Angelo

author maintainer
#22 lore raw
Quoted 51 lines · AngeloGioacchino Del Regno

Il 15/05/23 16:07, Julien Stephan ha scritto:

Quoted 21 lines · AngeloGioacchino Del Regno
Quoted 4 lines
Quoted 1 line

+#define CSIxB_OFFSET 0x1000

What if we grab two (or three?) iospaces from devicetree?

Quoted 8 lines

- base (global)
- csi_a
- csi_b

That would make it possible to maybe eventually extend this driver to more versions (older or newer) of the CSI PHY IP without putting fixes offsets inside of platform data structures and such.

Hi Angelo,
The register bank of the CSI port is divided into 2:
* from base address to base + 0x1000 (port A)
* from base + 0x1000 to base +0x2000 (port B)
Some CSI port can be configured in 4D1C mode (4 data + 1 clock) using the whole register bank from base to base + 0x2000 or in 2D1C mode (2 data + 1 clock) and use either port A or port B.

Quoted 7 lines

For example mt8365 has CSI0 that can be used either in 4D1C mode or in 2 * 2D1C and CSI1 which can use only 4D1C mode

2D1C mode can not be tested and is not implemented in the driver so I guess adding csi_a and csi_b reg value may be confusing?

What do you think?

Ok so we're talking about two data lanes per CSI port... it may still be beneficial to split the two register regions as

reg-names = "csi-a", "csi-b"; (whoops, I actually used underscores before,
and that was a mistake, sorry!)

....but that would be actually good only if we are expecting to get a CSI PHY in the future with four data lanes per port.

If you do *not* expect at all such a CSI PHY, or you do *not* expect such a PHY to ever be compatible with this driver (read as: if you expect such a PHY to be literally completely different from this one), then it would not change much to have the registers split in two.

Another case in which it would make sense is if we were to get a PHY that provides more than two CSI ports: in that case, we'd avoid platform data machinery to check the number of actual ports in the IP, as we would be just checking how many register regions we were given from the devicetree, meaning that if we got "csi-a", "csi-b", "csi-c", "csi-d", we have four ports.

Quoted 7 lines

Besides, another thing to think about is... yes you cannot test nor implement 2D1C mode in your submission, but this doesn't mean that others won't ever be interested in this and that other people won't be actually implementing that; Providing them with the right initial driver structure will surely make things easier, encouraging other people from the community to spend their precious time on the topic.

Hi Angelo,
Ok, I see your point, but for future potential upgrade to support A/B ports I was thinking of something else: adding independent nodes for csixA and csixB such as:

csi0_rx: phy@11c10000 {
reg = <0 0x11C10000 0 0x2000>;
mediatek,mode = <4D1c>;
...
};

csi0a_rx: phy@11c10000 {
reg = <0 0x11C10000 0 0x1000>;
mediatek,mode = <2D1c>;
...
};
csi0b_rx: phy@11c11000 {
reg = <0 0x11C11000 0 0x1000>;
mediatek,mode = <2D1c>;
...
};

giving the correct register range. One thing I did not mention is that if csi0_rx is used csi0a_rx and csi0b_rx cannot be used (they share same physical lanes as csio_rx), but csi0a_rx and csi0b_rx can be used simultaneously. So platform device will enable only the node(s) it needs and enabling csi0_rx and csioa/b_rx will fail because they share the same register region and map will fail and it does not have any sense because you either have a camera using the whole port or sub port but you cannot have both plugged in. What do you think about it?

Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

+#define CSIxB_OFFSET 0x1000

Maybe moving this declaration in phy-mtk-mipi-csi-0-5-rx-reg.h would be better?

Regards,
Julien

Quoted 3 lines

Regards,
Angelo

#23 lore raw

Il 16/05/23 11:30, Julien Stephan ha scritto:

Quoted 81 lines · AngeloGioacchino Del Regno
Quoted 51 lines

Il 15/05/23 16:07, Julien Stephan ha scritto:

Quoted 21 lines · AngeloGioacchino Del Regno
Quoted 4 lines
Quoted 1 line

+#define CSIxB_OFFSET 0x1000

What if we grab two (or three?) iospaces from devicetree?

Quoted 8 lines

- base (global)
- csi_a
- csi_b

That would make it possible to maybe eventually extend this driver to more versions (older or newer) of the CSI PHY IP without putting fixes offsets inside of platform data structures and such.

Hi Angelo,
The register bank of the CSI port is divided into 2:
* from base address to base + 0x1000 (port A)
* from base + 0x1000 to base +0x2000 (port B)
Some CSI port can be configured in 4D1C mode (4 data + 1 clock) using the whole register bank from base to base + 0x2000 or in 2D1C mode (2 data + 1 clock) and use either port A or port B.

Quoted 7 lines

For example mt8365 has CSI0 that can be used either in 4D1C mode or in 2 * 2D1C and CSI1 which can use only 4D1C mode

2D1C mode can not be tested and is not implemented in the driver so I guess adding csi_a and csi_b reg value may be confusing?

What do you think?

Ok so we're talking about two data lanes per CSI port... it may still be beneficial to split the two register regions as

reg-names = "csi-a", "csi-b"; (whoops, I actually used underscores before,
and that was a mistake, sorry!)

....but that would be actually good only if we are expecting to get a CSI PHY in the future with four data lanes per port.

If you do *not* expect at all such a CSI PHY, or you do *not* expect such a PHY to ever be compatible with this driver (read as: if you expect such a PHY to be literally completely different from this one), then it would not change much to have the registers split in two.

Another case in which it would make sense is if we were to get a PHY that provides more than two CSI ports: in that case, we'd avoid platform data machinery to check the number of actual ports in the IP, as we would be just checking how many register regions we were given from the devicetree, meaning that if we got "csi-a", "csi-b", "csi-c", "csi-d", we have four ports.

Quoted 7 lines

Besides, another thing to think about is... yes you cannot test nor implement 2D1C mode in your submission, but this doesn't mean that others won't ever be interested in this and that other people won't be actually implementing that; Providing them with the right initial driver structure will surely make things easier, encouraging other people from the community to spend their precious time on the topic.

Hi Angelo,
Ok, I see your point, but for future potential upgrade to support A/B ports I was thinking of something else: adding independent nodes for csixA and csixB such as:

csi0_rx: phy@11c10000 {
   reg = <0 0x11C10000 0 0x2000>;
   mediatek,mode = <4D1c>;
   ...
};
csi0a_rx: phy@11c10000 {
   reg = <0 0x11C10000 0 0x1000>;
   mediatek,mode = <2D1c>;
   ...
};
csi0b_rx: phy@11c11000 {
   reg = <0 0x11C11000 0 0x1000>;
   mediatek,mode = <2D1c>;
   ...
};
Quoted 9 lines

giving the correct register range. One thing I did not mention is that if csi0_rx is used csi0a_rx and csi0b_rx cannot be used (they share same physical lanes as csio_rx), but csi0a_rx and csi0b_rx can be used simultaneously. So platform device will enable only the node(s) it needs and enabling csi0_rx and csioa/b_rx will fail because they share the same register region and map will fail and it does not have any sense because you either have a camera using the whole port or sub port but you cannot have both plugged in. What do you think about it?

Your description of the hardware makes me even more confident in pushing for having one single node with multiple iospaces.

You could have a node such as:

csi0_rx: phy@11c10000 {
	compatible = ....
	reg = <0 0x11c10000 0 0x1000>, <0 0x11c20000 0 0x1000>;
	reg-names = "csi-a", "csi-b";
	/* 4 means 4D1C */
	num-lanes = <4>;
		or
	/* 2 means 2D1C */
	num-lanes = <2>;
};

You would then reference the csi0_rx node as:

	/* PHY is configured as 4 lanes (4D1C) */
	something = <&csi0_rx 0>;

	or

	/* First two lanes (CSI0 PORT-A) */
	something = <&csi0_rx 0>;
	/* Second two lanes (CSI0 PORT-B) */
	something = <&csi0_rx 1>;

Preferrably, you should (or shall?) use a graph to describe such connections, anyway.

This is because overriding the number of lanes on a per-board basis becomes *otherwise* difficult, in the sense of human readability issues, other than duplicated nodes being a real issue.

Regards,
Angelo