On Tue, 23 Mar 2010, Pete Zaitcev wrote: > On Tue, 23 Mar 2010 10:11:39 -0400 (EDT) > Alan Stern <stern@xxxxxxxxxxxxxxxxxxx> wrote: > > > Host controller hardware can't handle scatter-gather transfers unless > > the boundary between scatterlist elements occurs either at a packet > > boundary or at a page boundary -- and in fact, the host controller > > drivers don't handle the case where a page boundary isn't also a packet > > boundary. > > > > Because of this restriction, and because high-speed bulk endpoints must > > have a maxpacket length of 512, usb-storage (currently the only > > scatter-gather USB driver) requires 512-byte alignment of the data > > buffers. Any future driver using the scatter-gather library can be > > expected to include the same requirement. Hence it is not possible for > > 2 bytes to cross a page boundary. > > Hmm, that's an interesting chain of reasoning. I see no fault in it, > so we just need to make sure we google it out when someone other than > usb-storage somehow manages to submit a tricky buffer and crashes > usbmon... I'm saving it locally too. Messing up usbmon wouldn't be the only symptom -- high-speed transfers would fail because one or more of the non-terminal packets would be smaller than the maxpacket size. Similar problems have already been observed in wusb, where the bulk maxpacket sizes aren't multiples of 512 (and don't ask me why they aren't; that design decision seems remarkably foolish). Adding a comment explaining all this to usb_sg_init() seems like a good idea (another item to add to my todo list)... 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
