On Wed, Jan 29, 2020 at 02:45:38PM +0530, Nagarjuna Kristam wrote:

On 28-01-2020 23:02, Thierry Reding wrote:

quoted

On Mon, Dec 30, 2019 at 04:39:40PM +0530, Nagarjuna Kristam wrote:

quoted

If usb-role-switch property is present in USB 2 port, register
usb-role-switch to receive usb role changes.

Signed-off-by: Nagarjuna Kristam<redacted>
---
V3:
  - Driver aborts if usb-role-switch is not added in dt forotg/peripheral
    roles.
  - Added role name strings instead of enum values in debug prints.
  - Updated arguments and variable allignments as per Thierry inputs.
---
V2:
  - Removed dev_set_drvdata for port->dev.
  - Added of_platform_depopulate during error handling and driver removal.
---
  drivers/phy/tegra/Kconfig |  1 +
  drivers/phy/tegra/xusb.c  | 57 +++++++++++++++++++++++++++++++++++++++++++++++
  drivers/phy/tegra/xusb.h  |  3 +++
  3 files changed, 61 insertions(+)

diff --git a/drivers/phy/tegra/Kconfig b/drivers/phy/tegra/Kconfig
index f9817c3..df07c4d 100644
--- a/drivers/phy/tegra/Kconfig
+++ b/drivers/phy/tegra/Kconfig

@@ -2,6 +2,7 @@
  config PHY_TEGRA_XUSB
  	tristate "NVIDIA Tegra XUSB pad controller driver"
  	depends on ARCH_TEGRA
+	select USB_CONN_GPIO
  	help
  	  Choose this option if you have an NVIDIA Tegra SoC.

diff --git a/drivers/phy/tegra/xusb.c b/drivers/phy/tegra/xusb.c
index f98ec39..11ea9b5 100644
--- a/drivers/phy/tegra/xusb.c
+++ b/drivers/phy/tegra/xusb.c

@@ -523,6 +523,7 @@ static int tegra_xusb_port_init(struct tegra_xusb_port *port,
  	port->dev.type = &tegra_xusb_port_type;
  	port->dev.of_node = of_node_get(np);
  	port->dev.parent = padctl->dev;
+	port->dev.driver = padctl->dev->driver;

This looks wrong. I don't think driver's are supposed to set this
because it basically means that the device is being attached to the
driver, but in this case it doesn't get probed by the driver and in
fact the ports don't match the pad controller, so they can't really
be driven by the same driver.

Is there any particular reason why you need this?

Thierry

Yes, port->dev.driver->owner is accessed in USB role switch driver in API
usb_role_switch_get. If driver param is not updated, it causes NULL pointer
exception. Based on your inputs, since this assignment is not supposed to
used, I believe option available is to create a new device_driver structure
and assign the same here.
Please share your thoughts.

Sounds to me more like what we want is to make the module_get() and
module_put() calls conditional to avoid the NULL pointer dereference.
Another common variant that I've seen in other subsystems is to make
drivers pass in an owner explicitly via some operations structure to
allow more fine-grained control over the reference count.

In this particular case one option to do this would be to add a struct
module *owner field to usb_role_switch_desc and that's copied to struct
usb_role_switch in usb_role_switch_register() so that that can be used
for reference counting (perhaps with the current code as fallback).

That would allow using simple devices (i.e. not bound to a driver) to
work with this framework as well.

Thierry