Am Dienstag 28 April 2009 04:53:56 schrieb Sarah Sharp:
> +static void inc_enq(struct xhci_hcd *xhci, struct xhci_ring *ring, bool consumer)
> +{
> + u32 chain;
> + union xhci_trb *next;
> +
> + chain = ring->enqueue->generic.field[3] & TRB_CHAIN;
> + next = ++(ring->enqueue);
> +
> + ring->enq_updates++;
> + /* Update the dequeue pointer further if that was a link TRB or we're at
> + * the end of an event ring segment (which doesn't have link TRBS)
> + */
> + while (last_trb(xhci, ring, ring->enq_seg, next)) {
This looks fishy. If this really hands over the TRB to the hardware, we
need to make sure that the TRBs hit RAM in order. You need a wmb().
> + if (!consumer) {
> + if (ring != xhci->event_ring) {
> + /* Give this link TRB to the hardware */
> + if (next->link.control & TRB_CYCLE)
> + next->link.control &= (u32) ~TRB_CYCLE;
> + else
> + next->link.control |= (u32) TRB_CYCLE;
> + next->link.control &= TRB_CHAIN;
> + next->link.control |= chain;
Coded this way, the compiler is free to spill intermediate values into
RAM, so the hardware sees them.
Regards
Oliver
--
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
