On Fri, Jun 26, 2015 at 8:44 PM, David Laight <David.Laight@xxxxxxxxxx> wrote: > From: Steve Calfee >> Sent: 26 June 2015 15:59 >> > On the host (Wandboard iMX6q) the test app opens /dev/hidraw0 and >> > write 64 bytes with report ID (1). The HID device has no Interrupt OUT >> > ep, therefore uses control endpoint ep0 for the 64 bytes transfer to >> > gadget (Wandboard iMX6q) using set_report. >> > >> > The setup phase is OK and the 64 bytes is written to gadget. However >> > the chipidea interrupt (ci_irq()) and resulting udc interrupt >> > (udc_irq()) is invoked. This indicates that the 64 bytes transaction >> > is not completed over ep0 from host to gadget. >> > >> > **this issue is reproducible for all data transfers that aligns on 64 bytes** >> > >> > ~jayan >> > >> Hi Jayan, >> >> This sounds like a Zero Length Transfer issue. This applies to any >> endpoint including EP0. >> >> A ZLT is needed to end any transfer IFF the length is not already >> known by the protocol. So if the hid URB requests say 1024 bytes and >> the gadget only has 64, it must first send the 64 byte maxpacket and >> then send a zero length packet. This way the host knows the transfer >> is complete. > > Is this using the xhci driver? > The ZLP sending code used to be broken. > A couple of patches have been submitted but I don't remember one being applied. > > David > Thanks for the pointers. This is not using the xhci driver. - This is a ep0 OUT transfer. Does the host need to add a ZLP for 64 byte aligned data transfer? - Is it not expected for the completion interrupt be raised on receiving the 64 bytes of data, irrespective of whether or not a ZLP has been sent. ~Jayan -- 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
