Question about ack to PC side using read zero length packet

[seago66]

Hello,

Is this a bug for below code that should use write but not read ?
// Just ACK
status = read (ctx->usb_device, &status, 0);

Thanks a lot!

[marcone]

@jfdelnero I was wondering about this too. In change 1e6936a I removed the read() for the MTP_REQ_CANCEL case because it was causing a hang, however I found that changing that 0-length read to a 0-length write also works. I am unfamiliar with the actual behavior of the USB driver though, so don't know if doing an "ACK" really requires doing a read, or if that was perhaps a copy/paste error and it was supposed to be a write all along.

[jfdelnero]

0 byte read is generally used to ack. for example :

In http://www.linux-usb.org/gadget/usb.c

Line 1429 (and some others)

		/* ... ack (a write would stall) */
		status = read (fd, &status, 0);
		if (status)
			perror ("ack SET_CONFIGURATION");
		return;

[marcone]

OK, though given that there's no further explanation or documentation of the reason for the stall, I can't help but wonder if it's just because of a driver quirk in a 20 year old USB driver (that example code was written over 20 years ago).

In any case, that read in the cancel case was definitely causing the hang that caused the Windows playback problems. It does seem to be device-specific though (as I mentioned in another comment it worked OK both with and without the read on a Rock Pi 4C+), so if it turns out to be necessary on some platforms then it should be put back with a compile-time or run-time option to skip the call.