On Tue, 7 Dec 2010, Richard Retanubun wrote:
> Here is the more detailed question:
>
> - My linux baseline is 2.6.32-5
>
> - I've ported the isp1763-hcd driver from isp1760-hcd driver
> (and trying to mainline a unified one on a separate thread of work).
>
> - The platform I am working on (MPC8360E CPU) will have isp1763 chips on both the main board
> and some hotswap modules.
>
> - For the mainboard one, things are simple, declare stuff in dtb and up it goes.
>
> - For the hcd on the hostswap modules, some help from external program is needed to
> realize a new module is plugged in and trigger calling usb_create_hcd and usb_add_hcd
> for the driver (using generic netlink), modprobe with arguments may also work, but
> generic netlink feels like a more structured approach.
Once the kernel knows a new module has been plugged in, the module's
driver should be responsible for registering all the devices contained
in the module.
> - I am mostly wondering about the use of struct device *dev; In the examples I have looked at
> most of these comes wrapped as a member of a larger struct (struct of_device, struct pci_dev,
> struct usb_dev, struct platform_device) I don't know if it is legal to do something like this:
It is legal but it is rarely useful.
> static int isp1763_genl_lmb_dev_add(struct sk_buff *skbuff, struct genl_info *info)
> {
> struct device dev; /* local device structure, initialize somehow */
> struct usb_hcd *hcd;
>
> /* Get mem resource and irq resource from genl_info fill it into hcd struct members */
>
> hcd = usb_create_hcd(&isp1763_hc_driver, &dev, dev_name(&dev));
> -----------------------------------------^^^^^^^^^^^^^^^^^^^^
> /* Is passing a local (non-wrapped) device stucture here legal? is it a good practice?
> * or is it better to use a static platform_device per driver instance?
> */
> }
This is definitely wrong, for several reasons. For one thing, the hcd
structure will contain a pointer to the struct device on the stack.
When this function returns, it will immediately become a stale pointer.
> I can also create a local platform_device structure in the function above (or is it better that this be a static filescope structure?)
> and use the struct device *dev that is included inside the struct platform_device but that feels awkward, and yet I haven't found another
> way to cleanly remove the device later like this:
>
> /* static filescope platform_device, filled in during isp1763_genl_lmb_dev_add */
> static struct platform_device pdev;
>
> static int isp1763_genl_lmb_dev_remove(struct sk_buff *skbuff, struct genl_info *info)
> {
> /* Get mem resource and irq resource from genl_info fill it into hcd struct members */
>
> struct usb_hcd *hcd = platform_get_drvdata(pdev);
>
> usb_remove_hcd(hcd);
> /* Free memory and irq resources */
> usb_put_hcd(hcd);
> return 0;
> }
>
> The only driver I've found that does something similar (having a static struct platform_device) is /usb/host/sl811_cs.c
> but I am not sure if there are some special pcmcia concepts I am missing,
What does pcmcia have to do with any of this?
> for example it calls platform_device_register and platform_device_unregister.
>
> Any comments is much appreciated.
Look at drivers/usb/gadget/dummy_hcd.c for an example much like what
you are doing.
Alan Stern
--
To unsubscribe from this list: send the line "unsubscribe linux-usb" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
