[5/6] phy: qcom-qmp-combo: fix drm bridge registration

Message ID 20240217150228.5788-6-johan+linaro@kernel.org
State New
Headers
Series soc: qcom: pmic_glink_altmode: fix drm bridge use-after-free |

Commit Message

Johan Hovold Feb. 17, 2024, 3:02 p.m. UTC
  Due to a long-standing issue in driver core, drivers may not probe defer
after having registered child devices to avoid triggering a probe
deferral loop (see fbc35b45f9f6 ("Add documentation on meaning of
-EPROBE_DEFER")).

This could potentially also trigger a bug in the DRM bridge
implementation which does not expect bridges to go away even if device
links may avoid triggering this (when enabled).

Move registration of the DRM aux bridge to after looking up clocks and
other resources.

Note that PHY creation can in theory also trigger a probe deferral when
a 'phy' supply is used. This does not seem to affect the QMP PHY driver
but the PHY subsystem should be reworked to address this (i.e. by
separating initialisation and registration of the PHY).

Fixes: 35921910bbd0 ("phy: qcom: qmp-combo: switch to DRM_AUX_BRIDGE")
Fixes: 1904c3f578dc ("phy: qcom-qmp-combo: Introduce drm_bridge")
Cc: stable@vger.kernel.org      # 6.5
Cc: Bjorn Andersson <quic_bjorande@quicinc.com>
Cc: Dmitry Baryshkov <dmitry.baryshkov@linaro.org>
Signed-off-by: Johan Hovold <johan+linaro@kernel.org>
---
 drivers/phy/qualcomm/phy-qcom-qmp-combo.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)
  

Comments

Neil Armstrong Feb. 19, 2024, 9:03 a.m. UTC | #1
On 17/02/2024 16:02, Johan Hovold wrote:
> Due to a long-standing issue in driver core, drivers may not probe defer
> after having registered child devices to avoid triggering a probe
> deferral loop (see fbc35b45f9f6 ("Add documentation on meaning of
> -EPROBE_DEFER")).
> 
> This could potentially also trigger a bug in the DRM bridge
> implementation which does not expect bridges to go away even if device
> links may avoid triggering this (when enabled).
> 
> Move registration of the DRM aux bridge to after looking up clocks and
> other resources.
> 
> Note that PHY creation can in theory also trigger a probe deferral when
> a 'phy' supply is used. This does not seem to affect the QMP PHY driver
> but the PHY subsystem should be reworked to address this (i.e. by
> separating initialisation and registration of the PHY).
> 
> Fixes: 35921910bbd0 ("phy: qcom: qmp-combo: switch to DRM_AUX_BRIDGE")
> Fixes: 1904c3f578dc ("phy: qcom-qmp-combo: Introduce drm_bridge")
> Cc: stable@vger.kernel.org      # 6.5
> Cc: Bjorn Andersson <quic_bjorande@quicinc.com>
> Cc: Dmitry Baryshkov <dmitry.baryshkov@linaro.org>
> Signed-off-by: Johan Hovold <johan+linaro@kernel.org>
> ---
>   drivers/phy/qualcomm/phy-qcom-qmp-combo.c | 8 ++++----
>   1 file changed, 4 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/phy/qualcomm/phy-qcom-qmp-combo.c b/drivers/phy/qualcomm/phy-qcom-qmp-combo.c
> index 1ad10110dd25..e19d6a084f10 100644
> --- a/drivers/phy/qualcomm/phy-qcom-qmp-combo.c
> +++ b/drivers/phy/qualcomm/phy-qcom-qmp-combo.c
> @@ -3566,10 +3566,6 @@ static int qmp_combo_probe(struct platform_device *pdev)
>   	if (ret)
>   		return ret;
>   
> -	ret = drm_aux_bridge_register(dev);
> -	if (ret)
> -		return ret;
> -
>   	/* Check for legacy binding with child nodes. */
>   	usb_np = of_get_child_by_name(dev->of_node, "usb3-phy");
>   	if (usb_np) {
> @@ -3589,6 +3585,10 @@ static int qmp_combo_probe(struct platform_device *pdev)
>   	if (ret)
>   		goto err_node_put;
>   
> +	ret = drm_aux_bridge_register(dev);
> +	if (ret)
> +		goto err_node_put;
> +
>   	pm_runtime_set_active(dev);
>   	ret = devm_pm_runtime_enable(dev);
>   	if (ret)

Reviewed-by: Neil Armstrong <neil.armstrong@linaro.org>
  
Bjorn Andersson Feb. 22, 2024, 2:21 a.m. UTC | #2
On Sat, Feb 17, 2024 at 04:02:27PM +0100, Johan Hovold wrote:
> Due to a long-standing issue in driver core, drivers may not probe defer
> after having registered child devices to avoid triggering a probe
> deferral loop (see fbc35b45f9f6 ("Add documentation on meaning of
> -EPROBE_DEFER")).
> 
> This could potentially also trigger a bug in the DRM bridge
> implementation which does not expect bridges to go away even if device
> links may avoid triggering this (when enabled).
> 
> Move registration of the DRM aux bridge to after looking up clocks and
> other resources.
> 
> Note that PHY creation can in theory also trigger a probe deferral when
> a 'phy' supply is used. This does not seem to affect the QMP PHY driver
> but the PHY subsystem should be reworked to address this (i.e. by
> separating initialisation and registration of the PHY).
> 
> Fixes: 35921910bbd0 ("phy: qcom: qmp-combo: switch to DRM_AUX_BRIDGE")
> Fixes: 1904c3f578dc ("phy: qcom-qmp-combo: Introduce drm_bridge")
> Cc: stable@vger.kernel.org      # 6.5
> Cc: Bjorn Andersson <quic_bjorande@quicinc.com>
> Cc: Dmitry Baryshkov <dmitry.baryshkov@linaro.org>
> Signed-off-by: Johan Hovold <johan+linaro@kernel.org>

Reviewed-by: Bjorn Andersson <andersson@kernel.org>

Regards,
Bjorn

> ---
>  drivers/phy/qualcomm/phy-qcom-qmp-combo.c | 8 ++++----
>  1 file changed, 4 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/phy/qualcomm/phy-qcom-qmp-combo.c b/drivers/phy/qualcomm/phy-qcom-qmp-combo.c
> index 1ad10110dd25..e19d6a084f10 100644
> --- a/drivers/phy/qualcomm/phy-qcom-qmp-combo.c
> +++ b/drivers/phy/qualcomm/phy-qcom-qmp-combo.c
> @@ -3566,10 +3566,6 @@ static int qmp_combo_probe(struct platform_device *pdev)
>  	if (ret)
>  		return ret;
>  
> -	ret = drm_aux_bridge_register(dev);
> -	if (ret)
> -		return ret;
> -
>  	/* Check for legacy binding with child nodes. */
>  	usb_np = of_get_child_by_name(dev->of_node, "usb3-phy");
>  	if (usb_np) {
> @@ -3589,6 +3585,10 @@ static int qmp_combo_probe(struct platform_device *pdev)
>  	if (ret)
>  		goto err_node_put;
>  
> +	ret = drm_aux_bridge_register(dev);
> +	if (ret)
> +		goto err_node_put;
> +
>  	pm_runtime_set_active(dev);
>  	ret = devm_pm_runtime_enable(dev);
>  	if (ret)
> -- 
> 2.43.0
>
  
Dmitry Baryshkov Feb. 22, 2024, 9:11 p.m. UTC | #3
On Sat, 17 Feb 2024 at 17:03, Johan Hovold <johan+linaro@kernel.org> wrote:
>
> Due to a long-standing issue in driver core, drivers may not probe defer
> after having registered child devices to avoid triggering a probe
> deferral loop (see fbc35b45f9f6 ("Add documentation on meaning of
> -EPROBE_DEFER")).
>
> This could potentially also trigger a bug in the DRM bridge
> implementation which does not expect bridges to go away even if device
> links may avoid triggering this (when enabled).
>
> Move registration of the DRM aux bridge to after looking up clocks and
> other resources.
>
> Note that PHY creation can in theory also trigger a probe deferral when
> a 'phy' supply is used. This does not seem to affect the QMP PHY driver
> but the PHY subsystem should be reworked to address this (i.e. by
> separating initialisation and registration of the PHY).
>
> Fixes: 35921910bbd0 ("phy: qcom: qmp-combo: switch to DRM_AUX_BRIDGE")
> Fixes: 1904c3f578dc ("phy: qcom-qmp-combo: Introduce drm_bridge")
> Cc: stable@vger.kernel.org      # 6.5
> Cc: Bjorn Andersson <quic_bjorande@quicinc.com>
> Cc: Dmitry Baryshkov <dmitry.baryshkov@linaro.org>
> Signed-off-by: Johan Hovold <johan+linaro@kernel.org>
> ---
>  drivers/phy/qualcomm/phy-qcom-qmp-combo.c | 8 ++++----
>  1 file changed, 4 insertions(+), 4 deletions(-)
>
Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@linaro.org>
  
Vinod Koul Feb. 23, 2024, 12:09 p.m. UTC | #4
On 17-02-24, 16:02, Johan Hovold wrote:
> Due to a long-standing issue in driver core, drivers may not probe defer
> after having registered child devices to avoid triggering a probe
> deferral loop (see fbc35b45f9f6 ("Add documentation on meaning of
> -EPROBE_DEFER")).
> 
> This could potentially also trigger a bug in the DRM bridge
> implementation which does not expect bridges to go away even if device
> links may avoid triggering this (when enabled).
> 
> Move registration of the DRM aux bridge to after looking up clocks and
> other resources.
> 
> Note that PHY creation can in theory also trigger a probe deferral when
> a 'phy' supply is used. This does not seem to affect the QMP PHY driver
> but the PHY subsystem should be reworked to address this (i.e. by
> separating initialisation and registration of the PHY).

Acked-by: Vinod Koul <vkoul@kernel.org>
  

Patch

diff --git a/drivers/phy/qualcomm/phy-qcom-qmp-combo.c b/drivers/phy/qualcomm/phy-qcom-qmp-combo.c
index 1ad10110dd25..e19d6a084f10 100644
--- a/drivers/phy/qualcomm/phy-qcom-qmp-combo.c
+++ b/drivers/phy/qualcomm/phy-qcom-qmp-combo.c
@@ -3566,10 +3566,6 @@  static int qmp_combo_probe(struct platform_device *pdev)
 	if (ret)
 		return ret;
 
-	ret = drm_aux_bridge_register(dev);
-	if (ret)
-		return ret;
-
 	/* Check for legacy binding with child nodes. */
 	usb_np = of_get_child_by_name(dev->of_node, "usb3-phy");
 	if (usb_np) {
@@ -3589,6 +3585,10 @@  static int qmp_combo_probe(struct platform_device *pdev)
 	if (ret)
 		goto err_node_put;
 
+	ret = drm_aux_bridge_register(dev);
+	if (ret)
+		goto err_node_put;
+
 	pm_runtime_set_active(dev);
 	ret = devm_pm_runtime_enable(dev);
 	if (ret)