On Tuesday, August 24, 2021 7:44:40 AM CEST Christophe JAILLET wrote:

Le 24/08/2021 à 04:01, Fabio M. De Francesco a écrit :

quoted

On Tuesday, August 24, 2021 3:38:03 AM CEST Fabio M. De Francesco wrote:

quoted

I think that I've inadvertently switched the order by which usb_control_msg_send()
and memcpy() are called. I'm very sorry for not doing my tests, but (as I had said
before) at the moment I don't have my device with me.

No, I did not switch them. There must be something else...
Sorry for the noise.

Fabio

Hi,

'usb_control_msg_recv()' looks like:

int usb_control_msg_recv(struct usb_device *dev, __u8 endpoint, ...)
{
	unsigned int pipe = usb_rcvctrlpipe(dev, endpoint);
	...
	ret = usb_control_msg(dev, pipe, ...);


'usb_control_msg()' looks like:
int usb_control_msg(struct usb_device *dev, unsigned int pipe, ...)
{

The difference is that one expect an 'endpoint' (and compute the pipe 
from it), and the other expect a 'pipe'.

Hi Christophe,

Yes, correct. That's why I changed the type of 'pipe' from "unsigned int"
to "u8". I also saw that usb_control_msg_recv/send take care of calling 
usb_rcvctrpipe() and usb_sndctrlpipe(); so, in my patch I deleted 
those calls.

Not related to my patch... why Linux has u8 and __u8? What are the  
different use cases they are meant for? 

Also, in your code, 'pipe' looks un-initialized.

Oh yes, good catch.  Thanks!

So, my guess is that you should rename 'pipe' into 'endpoint' (to keep 
the semantic),
have "endpoint = 0;" somewhere and pass it to 
usb_control_msg_{recv|send}.
Or just remove 'pipe' and pass an explicit 0 directly.

I've just seen that in other drivers the code passes an explicit 0.
So, also according to your suggestion, I'll remove "pipe/endpoint".

Not sure it is enough, but it looks like a difference between before and 
after your patch.

Since I cannot see other issues, I'm about to fix the code as said above and
then submit a v2 series.

Your 2c are worth much more than how much you think :)

Thanks very much,

Fabio

just my 2c,
CJ