Skip to contentSkip to conversation

drm/bridge: analogix_dp: use devm_phy_optional_get() for the DP PHY

PATCH Superseded Jun Yan 3 participants 6 messages Nov 10 – Nov 14

Subsystems (4)

v3 of 5 Superseded Latest is v5 →

lore
Share & tools
lore

Jun Yan

author
#0 lore raw

PATCH v3 1/1 drm/bridge: analogix_dp: use devm_phy_optional_get() for the DP PHY

The DP PHY is optional: the Rockchip RK3399 eDP controller (used as the internal eDP port) has no separate PHY device and relies on its built-in PHY, which is programmed directly through the GRF registers, so its DT node carries no "phys"/"phy-names" property.

Requesting the PHY with devm_phy_get() therefore returns -ENODEV (or -ENOSYS when CONFIG_GENERIC_PHY is disabled) and the driver logs a misleading "no DP phy configured" error at probe time before falling back to a NULL phy.

Use devm_phy_optional_get() instead, which returns NULL when the PHY is absent and otherwise behaves identically, and drop the manual -ENODEV/ -ENOSYS handling. This also removes the spurious error message.

---
Changes in v3:
- Add missing changelog entries and reference links
- Link to v2: https://lore.kernel.org/dri-devel/[email protected]/

Changes in v2:
- Added comments to explain the changes made.
- Link to v1: https://lore.kernel.org/dri-devel/[email protected]/

 .../drm/bridge/analogix/analogix_dp_core.c    | 22 ++++++-------------
 1 file changed, 7 insertions(+), 15 deletions(-)
drivers/gpu/drm/bridge/analogix/analogix_dp_core.c +7 -15
index 6df6963740d2..637b1bff0f6c 100644
@@ -1426,21 +1426,13 @@ analogix_dp_probe(struct device *dev, struct analogix_dp_plat_data *plat_data)
 	if (ret)
 		return ERR_PTR(ret);
 
-	dp->phy = devm_phy_get(dp->dev, "dp");
-	if (IS_ERR(dp->phy)) {
-		dev_err(dp->dev, "no DP phy configured\n");
-		ret = PTR_ERR(dp->phy);
-		if (ret) {
-			/*
-			 * phy itself is not enabled, so we can move forward
-			 * assigning NULL to phy pointer.
-			 */
-			if (ret == -ENOSYS || ret == -ENODEV)
-				dp->phy = NULL;
-			else
-				return ERR_PTR(ret);
-		}
-	}
+	/*
+	 * RK3288/RK3399 use Analogix DP controller built-in PHY,
+	 * no extra PHY configuration is required.
+	 */
+	dp->phy = devm_phy_optional_get(dp->dev, "dp");
+	if (IS_ERR(dp->phy))
+		return ERR_CAST(dp->phy);
 
 	dp->clock = devm_clk_get(&pdev->dev, "dp");
 	if (IS_ERR(dp->clock)) {
Signature · 1 line
2.56.0
reviewer
#1 lore raw

Hello,

Quoted 48 lines · Jun Yan

The DP PHY is optional: the Rockchip RK3399 eDP controller (used as the internal eDP port) has no separate PHY device and relies on its built-in PHY, which is programmed directly through the GRF registers, so its DT node carries no "phys"/"phy-names" property.

Requesting the PHY with devm_phy_get() therefore returns -ENODEV (or -ENOSYS when CONFIG_GENERIC_PHY is disabled) and the driver logs a misleading "no DP phy configured" error at probe time before falling back to a NULL phy.

Use devm_phy_optional_get() instead, which returns NULL when the PHY is absent and otherwise behaves identically, and drop the manual -ENODEV/ -ENOSYS handling. This also removes the spurious error message.

---
Changes in v3:
- Add missing changelog entries and reference links
- Link to v2: https://lore.kernel.org/dri-devel/[email protected]/

Changes in v2:
- Added comments to explain the changes made.
- Link to v1: https://lore.kernel.org/dri-devel/[email protected]/

 .../drm/bridge/analogix/analogix_dp_core.c    | 22 ++++++-------------
 1 file changed, 7 insertions(+), 15 deletions(-)
drivers/gpu/drm/bridge/analogix/analogix_dp_core.c +0 -12
index 6df6963740d2..637b1bff0f6c 100644
@@ -1426,21 +1426,13 @@ analogix_dp_probe(struct device *dev, struct analogix_dp_plat_data *plat_data)
 	if (ret)
 		return ERR_PTR(ret);
 
-	dp->phy = devm_phy_get(dp->dev, "dp");
-	if (IS_ERR(dp->phy)) {
-		dev_err(dp->dev, "no DP phy configured\n");
-		ret = PTR_ERR(dp->phy);
-		if (ret) {
-			/*
-			 * phy itself is not enabled, so we can move forward
-			 * assigning NULL to phy pointer.
-			 */
-			if (ret == -ENOSYS || ret == -ENODEV)
-				dp->phy = NULL;
-			else

drivers/gpu/drm/bridge/analogix/analogix_dp_core.c analogix_dp_probe(struct device *dev, struct analogix_dp_plat_data *plat_data) · line 1435

Quoted 10 lines
drivers/gpu/drm/bridge/analogix/analogix_dp_core.c +7 -3
-				return ERR_PTR(ret);
-		}
-	}
+	/*
+	 * RK3288/RK3399 use Analogix DP controller built-in PHY,
+	 * no extra PHY configuration is required.
+	 */
+	dp->phy = devm_phy_optional_get(dp->dev, "dp");
+	if (IS_ERR(dp->phy))
+		return ERR_CAST(dp->phy);

The original code ignbores -ENOSYS, devm_phy_optional_get() does not. So, either -ENOSYS was just impossible, or this patch potentially changes the behaviour.

Can you elaborate on that?

Luca

--
Luca Ceresoli, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com

#2 lore raw
Quoted 66 lines · Luca Ceresoli

Hello,

Quoted 48 lines · Jun Yan

The DP PHY is optional: the Rockchip RK3399 eDP controller (used as the internal eDP port) has no separate PHY device and relies on its built-in PHY, which is programmed directly through the GRF registers, so its DT node carries no "phys"/"phy-names" property.

Requesting the PHY with devm_phy_get() therefore returns -ENODEV (or -ENOSYS when CONFIG_GENERIC_PHY is disabled) and the driver logs a misleading "no DP phy configured" error at probe time before falling back to a NULL phy.

Use devm_phy_optional_get() instead, which returns NULL when the PHY is absent and otherwise behaves identically, and drop the manual -ENODEV/ -ENOSYS handling. This also removes the spurious error message.

---
Changes in v3:
- Add missing changelog entries and reference links
- Link to v2: https://lore.kernel.org/dri-devel/[email protected]/

Changes in v2:
- Added comments to explain the changes made.
- Link to v1: https://lore.kernel.org/dri-devel/[email protected]/

  .../drm/bridge/analogix/analogix_dp_core.c    | 22 ++++++-------------
  1 file changed, 7 insertions(+), 15 deletions(-)
drivers/gpu/drm/bridge/analogix/analogix_dp_core.c +0 -12
index 6df6963740d2..637b1bff0f6c 100644
@@ -1426,21 +1426,13 @@ analogix_dp_probe(struct device *dev, struct analogix_dp_plat_data *plat_data)
  	if (ret)
  		return ERR_PTR(ret);
 
-	dp->phy = devm_phy_get(dp->dev, "dp");
-	if (IS_ERR(dp->phy)) {
-		dev_err(dp->dev, "no DP phy configured\n");
-		ret = PTR_ERR(dp->phy);
-		if (ret) {
-			/*
-			 * phy itself is not enabled, so we can move forward
-			 * assigning NULL to phy pointer.
-			 */
-			if (ret == -ENOSYS || ret == -ENODEV)
-				dp->phy = NULL;
-			else

drivers/gpu/drm/bridge/analogix/analogix_dp_core.c analogix_dp_probe(struct device *dev, struct analogix_dp_plat_data *plat_data) · line 1435

Quoted 10 lines
drivers/gpu/drm/bridge/analogix/analogix_dp_core.c +7 -3
-				return ERR_PTR(ret);
-		}
-	}
+	/*
+	 * RK3288/RK3399 use Analogix DP controller built-in PHY,
+	 * no extra PHY configuration is required.
+	 */
+	dp->phy = devm_phy_optional_get(dp->dev, "dp");
+	if (IS_ERR(dp->phy))
+		return ERR_CAST(dp->phy);

The original code ignbores -ENOSYS, devm_phy_optional_get() does not. So, either -ENOSYS was just impossible, or this patch potentially changes the behaviour.

Quoted 3 lines

Can you elaborate on that?

You are right, this is really a valid concern.

If CONFIG_GENERIC_PHY is disabled, the old code could still proceed and keep the driver probing, while the new version would return an error directly.

It should be better to keep the -ENOSYS handling to preserve the original behavior.

Best regards,
Damon

author
#3 lore raw

Hi Damon,

Quoted 5 lines

You are right, this is really a valid concern.

If CONFIG_GENERIC_PHY is disabled, the old code could still proceed and keep the driver probing, while the new version would return an error directly.

-ENOSYS can only come from the !CONFIG_GENERIC_PHY stubs in include/linux/phy/phy.h; the real implementation in drivers/phy/phy-core.c never returns it. And the stub for the optional variant returns NULL, not -ENOSYS:

    static inline struct phy *devm_phy_optional_get(struct device *dev,
                                                    const char *string)
    {
            return NULL;
    }

So with CONFIG_GENERIC_PHY disabled the PHY is still treated as absent (NULL), exactly as the old -ENOSYS branch did. With CONFIG_GENERIC_PHY enabled, -ENOSYS cannot occur, devm_phy_optional_get() turns -ENODEV into NULL, and any other error is still propagated - same as the open-coded check.

Quoted 2 lines

It should be better to keep the -ENOSYS handling to preserve the original behavior.

Overall the behaviour is unchanged in both configurations, so no separate handling is needed.

Best regards,
Jun Yan

reviewer
#4 lore raw

Hello,

Quoted 25 lines · Jun Yan

Hi Damon,

Quoted 5 lines

You are right, this is really a valid concern.

If CONFIG_GENERIC_PHY is disabled, the old code could still proceed and keep the driver probing, while the new version would return an error directly.

-ENOSYS can only come from the !CONFIG_GENERIC_PHY stubs in include/linux/phy/phy.h; the real implementation in drivers/phy/phy-core.c never returns it. And the stub for the optional variant returns NULL, not -ENOSYS:

    static inline struct phy *devm_phy_optional_get(struct device *dev,
                                                    const char *string)
    {
            return NULL;
    }

So with CONFIG_GENERIC_PHY disabled the PHY is still treated as absent (NULL), exactly as the old -ENOSYS branch did. With CONFIG_GENERIC_PHY enabled, -ENOSYS cannot occur, devm_phy_optional_get() turns -ENODEV into NULL, and any other error is still propagated - same as the open-coded check.

Quoted 5 lines
Quoted 2 lines

It should be better to keep the -ENOSYS handling to preserve the original behavior.

Overall the behaviour is unchanged in both configurations, so no separate handling is needed.

Thanks for the analysis! Cna you send a v4 with that explained in the commit message, possibly in a more concise way, maybe with bullets for the various CONFIG_GENERIC_PHY on/off cases or so, for readability?

Reading it now, this v3 commit message kind of said the same things. but it misled both me and Damon, so it's worth a clear explanation.

Thanks!

Luca

--
Luca Ceresoli, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com

author
#5 lore raw

Hi Luca,

Quoted 32 lines

Hello,

Quoted 10 lines · Jun Yan

Hi Damon,

Quoted 5 lines

You are right, this is really a valid concern.

If CONFIG_GENERIC_PHY is disabled, the old code could still proceed and keep the driver probing, while the new version would return an error directly.

-ENOSYS can only come from the !CONFIG_GENERIC_PHY stubs in include/linux/phy/phy.h; the real implementation in drivers/phy/phy-core.=

c

Quoted 18 lines

never returns it. And the stub for the optional variant returns NULL, not -ENOSYS:

    static inline struct phy *devm_phy_optional_get(struct device *dev,
                                                    const char *string)
    {
            return NULL;
    }

So with CONFIG_GENERIC_PHY disabled the PHY is still treated as absent (NULL), exactly as the old -ENOSYS branch did. With CONFIG_GENERIC_PHY enabled, -ENOSYS cannot occur, devm_phy_optional_get() turns -ENODEV into NULL, and any other error is still propagated - same as the open-coded check.

Quoted 2 lines

It should be better to keep the -ENOSYS handling to preserve the original behavior.

Quoted 7 lines
Quoted 2 lines

Overall the behaviour is unchanged in both configurations, so no separate handling is needed.

Thanks for the analysis! Cna you send a v4 with that explained in the commit message, possibly in a more concise way, maybe with bullets for the various CONFIG_GENERIC_PHY on/off cases or so, for readability?

Yes, I'll send a v4.

Quoted 6 lines

Reading it now, this v3 commit message kind of said the same things. but it misled both me and Damon, so it's worth a clear explanation.

Thanks!

Luca

Thanks,
Jun