Re: [linux-usb-devel] [PATCH 02/02] USB: add dynamic id functionality to USB core

[Greg KH <gregkh@suse.de>]

On Thu, Nov 17, 2005 at 10:55:33AM -0500, Alan Stern wrote:
> A few minor comments...
> 
> On Wed, 16 Nov 2005, Greg KH wrote:
> 
> > +static ssize_t store_new_id(struct device_driver *driver,
> > +			    const char *buf, size_t count)
> > +{
> > +	struct usb_driver *usb_drv = to_usb_driver(driver);
> > +	struct usb_dynid *dynid;
> > +	u32 idVendor = 0;
> > +	u32 idProduct = 0;
> > +	int fields = 0;
> > +
> > +	fields = sscanf(buf, "%x %x", &idVendor, &idProduct);
> > +	if (fields < 0)
> > +		return -EINVAL;
> 
> Don't you want to test (fields < 2)?

No, it's ok to just specify the vendor, if you want a product of 0 :)

PCI does it this way too.

> > +	if (get_driver(&usb_drv->driver)) {
> > +		driver_attach(&usb_drv->driver);
> > +		put_driver(&usb_drv->driver);
> > +	}
> 
> Here you don't have to refer to &usb_drv->driver; you can just refer to 
> driver.

Good point, changed, thanks.

> > +static int usb_create_newid_file(struct usb_driver *usb_drv)
> > +{
> > +	int error = 0;
> > +
> > +	if (usb_drv->probe != NULL)
> > +		error = sysfs_create_file(&usb_drv->driver.kobj,
> > +					  &driver_attr_new_id.attr);
> > +	return error;
> > +}
> 
> This deserves to be an inline function.
> 
> > +static void usb_free_dynids(struct usb_driver *usb_drv)
> > +{
> > +	struct usb_dynid *dynid, *n;
> > +
> > +	spin_lock(&usb_drv->dynids.lock);
> > +	list_for_each_entry_safe(dynid, n, &usb_drv->dynids.list, node) {
> > +		list_del(&dynid->node);
> > +		kfree(dynid);
> > +	}
> > +	spin_unlock(&usb_drv->dynids.lock);
> > +}
> 
> This could be inline as well, although being longer, its overhead is less 
> noticeable.

It's just not worth it to inline these, it's not speed critical at all.
I prefer to not inline stuff unless it help out somehow.

thanks for the review,

greg k-h