Re: Handling incoming ZLP in cdc-wdm

[Oliver Neukum <oneukum@suse.com>]

Hi,

On 27.03.25 14:01, Robert Hodaszi wrote:

>>
> Following on this: returning 0 bytes for read would kill libqmi, as that handles that as an error as well ("connection broken").
> 
> So what about this patch?
> 
>      diff --git a/drivers/usb/class/cdc-wdm.c b/drivers/usb/class/cdc-wdm.c
>      index 86ee39db013f..37873acd18f4 100644
>      --- a/drivers/usb/class/cdc-wdm.c
>      +++ b/drivers/usb/class/cdc-wdm.c
>      @@ -214,6 +214,11 @@ static void wdm_in_callback(struct urb *urb)
>              if (desc->rerr == 0 && status != -EPIPE)
>                      desc->rerr = status;
>       
>      +       if (length == 0) {
>      +               dev_dbg(&desc->intf->dev, "received ZLP\n");
>      +               goto skip_error;
>      +       }
>      +
>              if (length + desc->length > desc->wMaxCommand) {
>                      /* The buffer would overflow */
>                      set_bit(WDM_OVERFLOW, &desc->flags);
>      @@ -227,10 +232,10 @@ static void wdm_in_callback(struct urb *urb)
>              }
>       skip_error:
>       
>      -       if (desc->rerr) {
>      +       if (desc->rerr || length == 0) {
>                      /*
>      -                * Since there was an error, userspace may decide to not read
>      -                * any data after poll'ing.
>      +                * If there was a ZLP or an error, userspace may decide to not
>      +                * read any data after poll'ing.
>                       * We should respond to further attempts from the device to send
>                       * data, so that we can get unstuck.
>                       */

Why do you wish to react to this like an error?
It seems to me that we indeed need to wait for the device in this case.

	Regards
		Oliver