+/* This is very similar to what ehci-q.c qtd_fill() does */
+int queue_bulk_tx(struct xhci_hcd *xhci, gfp_t mem_flags,
+ struct urb *urb, int slot_id, unsigned int ep_index)
+{
+ struct xhci_ring *ep_ring;
+ struct xhci_td *td;
+ int num_trbs;
+ struct xhci_generic_trb *start_trb;
+ bool first_trb;
+ int start_cycle;
+ u32 field;
+
+ int running_total, trb_buff_len, ret;
+ u64 addr;
+
+ ep_ring = xhci->devs[slot_id]->ep_rings[ep_index];
+
+ num_trbs = 0;
+ /* How much data is (potentially) left before the 64KB boundary? */
+ running_total = TRB_MAX_BUFF_SIZE -
+ (urb->transfer_dma & ((1 << TRB_MAX_BUFF_SHIFT) - 1));
+
+ /* If there's some data on this 64KB chunk, or we have to send a
+ * zero-length transfer, we need at least one TRB
+ */
+ if (running_total != 0 || urb->transfer_buffer_length == 0)
+ num_trbs++;
+ /* How many more 64KB chunks to transfer, how many more TRBs? */
+ while (running_total < urb->transfer_buffer_length) {
+ num_trbs++;
+ running_total += TRB_MAX_BUFF_SIZE;
+ }
+ /* FIXME: this doesn't deal with URB_ZERO_PACKET - need one more */
+
+ if (!in_interrupt())
+ dev_dbg(&urb->dev->dev, "ep %#x - urb len = %d, addr = %#x, num_trbs = %d\n",
+ urb->ep->desc.bEndpointAddress,
+ urb->transfer_buffer_length, urb->transfer_dma,
+ num_trbs);
+ ret = xhci_prepare_transfer(xhci, xhci->devs[slot_id], ep_index,
+ num_trbs, urb, &td, mem_flags);
+ if (ret < 0)
+ return ret;
+
+ /*
+ * Don't give the first TRB to the hardware (by toggling the cycle bit)
+ * until we've finished creating all the other TRBs. The ring's cycle
+ * state may change as we enqueue the other TRBs, so save it too.
+ */
+ start_trb = &ep_ring->enqueue->generic;
+ start_cycle = ep_ring->cycle_state;
+
+ running_total = 0;
+ /* How much data is in the first TRB? */
+ addr = (u64) urb->transfer_dma;
+ trb_buff_len = TRB_MAX_BUFF_SIZE -
+ (urb->transfer_dma & ((1 << TRB_MAX_BUFF_SHIFT) - 1));
+ if (urb->transfer_buffer_length < trb_buff_len)
+ trb_buff_len = urb->transfer_buffer_length;
+
+ first_trb = true;
+
+ /* Queue the first TRB, even if it's zero-length */
+ do {
+ field = 0;
+
+ /* Don't change the cycle bit of the first TRB until later */
+ if (first_trb)
+ first_trb = false;
+ else
+ field |= ep_ring->cycle_state;
+
+ /* Chain all the TRBs together; clear the chain bit in the last
+ * TRB to indicate it's the last TRB in the chain.
+ */
+ if (num_trbs > 1) {
+ field |= TRB_CHAIN;
+ } else {
+ /* FIXME - add check for ZERO_PACKET flag before this */
+ td->last_trb = ep_ring->enqueue;
+ field |= TRB_IOC;
+ }
+ queue_trb(xhci, ep_ring, false,
+ (u32) addr,
+ (u32) ((u64) addr >> 32),
+ TRB_LEN(trb_buff_len) | TRB_INTR_TARGET(0),
+ /* We always want to know if the TRB was short,
+ * or we won't get an event when it completes.
+ * (Unless we use event data TRBs, which are a
+ * waste of space and HC resources.)
+ */
+ field | TRB_ISP | TRB_TYPE(TRB_NORMAL));
+ --num_trbs;
+ running_total += trb_buff_len;
+
+ /* Calculate length for next transfer */
+ addr += trb_buff_len;
+ trb_buff_len = urb->transfer_buffer_length - running_total;
+ if (trb_buff_len > TRB_MAX_BUFF_SIZE)
+ trb_buff_len = TRB_MAX_BUFF_SIZE;
+ } while (running_total < urb->transfer_buffer_length);
+
+ if (num_trbs != 0)
+ dev_dbg(&urb->dev->dev, "%s - ep %#x - Miscalculated number of "
+ "TRBs, %d left\n", __FUNCTION__,
+ urb->ep->desc.bEndpointAddress, num_trbs);
+ /*
+ * Pass all the TRBs to the hardware at once and make sure this write
+ * isn't reordered.
+ */
+ wmb();
+ start_trb->field[3] |= start_cycle;
+ field = xhci_readl(xhci, &xhci->dba->doorbell[slot_id]) & DB_MASK;
+ xhci_writel(xhci, field | EPI_TO_DB(ep_index), &xhci->dba->doorbell[slot_id]);
+ /* Flush PCI posted writes */
+ xhci_readl(xhci, &xhci->dba->doorbell[slot_id]);
+
+ return 0;
+}
+