On Thu, 11 Mar 2010, Alan Stern wrote: > If they do come out that way, the only explanation I can think of is a > cache-coherence failure. (Note that the usbmon code runs _after_ > unmap_urb_for_dma().) We'll have to ask an x86_64 architecture person. Wait a minute! I just realized something. Although usbmon _is_ called after unmap_urb_for_dma(), that's not where the unmapping actually takes place! Transfers coming through the scatter-gather library are unmapped when sg_clean() in message.c calls usb_buffer_unmap_sg(), which doesn't happen until after the entire transfer is complete. In particular, if there are multiple scatterlist elements giving rise to multiple URBs, the unmapping doesn't happen until after all the URBs have completed. That's not the case with you, but it's still true that the unmapping occurs after the usbmon and giveback calls. If there was a bounce buffer involved for any reason, that would explain your test results. Offhand, I can't think how to fix this. I'm open to suggestions. Alan Stern -- 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
