Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / drivers / usb / host / xhci-ring.c
diff --git a/kernel/drivers/usb/host/xhci-ring.c b/kernel/drivers/usb/host/xhci-ring.c
new file mode 100644 (file)
index 0000000..7d34cbf
--- /dev/null
@@ -0,0 +1,3962 @@
+/*
+ * xHCI host controller driver
+ *
+ * Copyright (C) 2008 Intel Corp.
+ *
+ * Author: Sarah Sharp
+ * Some code borrowed from the Linux EHCI driver.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+ * or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
+ * for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software Foundation,
+ * Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ */
+
+/*
+ * Ring initialization rules:
+ * 1. Each segment is initialized to zero, except for link TRBs.
+ * 2. Ring cycle state = 0.  This represents Producer Cycle State (PCS) or
+ *    Consumer Cycle State (CCS), depending on ring function.
+ * 3. Enqueue pointer = dequeue pointer = address of first TRB in the segment.
+ *
+ * Ring behavior rules:
+ * 1. A ring is empty if enqueue == dequeue.  This means there will always be at
+ *    least one free TRB in the ring.  This is useful if you want to turn that
+ *    into a link TRB and expand the ring.
+ * 2. When incrementing an enqueue or dequeue pointer, if the next TRB is a
+ *    link TRB, then load the pointer with the address in the link TRB.  If the
+ *    link TRB had its toggle bit set, you may need to update the ring cycle
+ *    state (see cycle bit rules).  You may have to do this multiple times
+ *    until you reach a non-link TRB.
+ * 3. A ring is full if enqueue++ (for the definition of increment above)
+ *    equals the dequeue pointer.
+ *
+ * Cycle bit rules:
+ * 1. When a consumer increments a dequeue pointer and encounters a toggle bit
+ *    in a link TRB, it must toggle the ring cycle state.
+ * 2. When a producer increments an enqueue pointer and encounters a toggle bit
+ *    in a link TRB, it must toggle the ring cycle state.
+ *
+ * Producer rules:
+ * 1. Check if ring is full before you enqueue.
+ * 2. Write the ring cycle state to the cycle bit in the TRB you're enqueuing.
+ *    Update enqueue pointer between each write (which may update the ring
+ *    cycle state).
+ * 3. Notify consumer.  If SW is producer, it rings the doorbell for command
+ *    and endpoint rings.  If HC is the producer for the event ring,
+ *    and it generates an interrupt according to interrupt modulation rules.
+ *
+ * Consumer rules:
+ * 1. Check if TRB belongs to you.  If the cycle bit == your ring cycle state,
+ *    the TRB is owned by the consumer.
+ * 2. Update dequeue pointer (which may update the ring cycle state) and
+ *    continue processing TRBs until you reach a TRB which is not owned by you.
+ * 3. Notify the producer.  SW is the consumer for the event ring, and it
+ *   updates event ring dequeue pointer.  HC is the consumer for the command and
+ *   endpoint rings; it generates events on the event ring for these.
+ */
+
+#include <linux/scatterlist.h>
+#include <linux/slab.h>
+#include "xhci.h"
+#include "xhci-trace.h"
+
+/*
+ * Returns zero if the TRB isn't in this segment, otherwise it returns the DMA
+ * address of the TRB.
+ */
+dma_addr_t xhci_trb_virt_to_dma(struct xhci_segment *seg,
+               union xhci_trb *trb)
+{
+       unsigned long segment_offset;
+
+       if (!seg || !trb || trb < seg->trbs)
+               return 0;
+       /* offset in TRBs */
+       segment_offset = trb - seg->trbs;
+       if (segment_offset > TRBS_PER_SEGMENT)
+               return 0;
+       return seg->dma + (segment_offset * sizeof(*trb));
+}
+
+/* Does this link TRB point to the first segment in a ring,
+ * or was the previous TRB the last TRB on the last segment in the ERST?
+ */
+static bool last_trb_on_last_seg(struct xhci_hcd *xhci, struct xhci_ring *ring,
+               struct xhci_segment *seg, union xhci_trb *trb)
+{
+       if (ring == xhci->event_ring)
+               return (trb == &seg->trbs[TRBS_PER_SEGMENT]) &&
+                       (seg->next == xhci->event_ring->first_seg);
+       else
+               return le32_to_cpu(trb->link.control) & LINK_TOGGLE;
+}
+
+/* Is this TRB a link TRB or was the last TRB the last TRB in this event ring
+ * segment?  I.e. would the updated event TRB pointer step off the end of the
+ * event seg?
+ */
+static int last_trb(struct xhci_hcd *xhci, struct xhci_ring *ring,
+               struct xhci_segment *seg, union xhci_trb *trb)
+{
+       if (ring == xhci->event_ring)
+               return trb == &seg->trbs[TRBS_PER_SEGMENT];
+       else
+               return TRB_TYPE_LINK_LE32(trb->link.control);
+}
+
+static int enqueue_is_link_trb(struct xhci_ring *ring)
+{
+       struct xhci_link_trb *link = &ring->enqueue->link;
+       return TRB_TYPE_LINK_LE32(link->control);
+}
+
+/* Updates trb to point to the next TRB in the ring, and updates seg if the next
+ * TRB is in a new segment.  This does not skip over link TRBs, and it does not
+ * effect the ring dequeue or enqueue pointers.
+ */
+static void next_trb(struct xhci_hcd *xhci,
+               struct xhci_ring *ring,
+               struct xhci_segment **seg,
+               union xhci_trb **trb)
+{
+       if (last_trb(xhci, ring, *seg, *trb)) {
+               *seg = (*seg)->next;
+               *trb = ((*seg)->trbs);
+       } else {
+               (*trb)++;
+       }
+}
+
+/*
+ * See Cycle bit rules. SW is the consumer for the event ring only.
+ * Don't make a ring full of link TRBs.  That would be dumb and this would loop.
+ */
+static void inc_deq(struct xhci_hcd *xhci, struct xhci_ring *ring)
+{
+       ring->deq_updates++;
+
+       /*
+        * If this is not event ring, and the dequeue pointer
+        * is not on a link TRB, there is one more usable TRB
+        */
+       if (ring->type != TYPE_EVENT &&
+                       !last_trb(xhci, ring, ring->deq_seg, ring->dequeue))
+               ring->num_trbs_free++;
+
+       do {
+               /*
+                * 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)
+                */
+               if (last_trb(xhci, ring, ring->deq_seg, ring->dequeue)) {
+                       if (ring->type == TYPE_EVENT &&
+                                       last_trb_on_last_seg(xhci, ring,
+                                               ring->deq_seg, ring->dequeue)) {
+                               ring->cycle_state ^= 1;
+                       }
+                       ring->deq_seg = ring->deq_seg->next;
+                       ring->dequeue = ring->deq_seg->trbs;
+               } else {
+                       ring->dequeue++;
+               }
+       } while (last_trb(xhci, ring, ring->deq_seg, ring->dequeue));
+}
+
+/*
+ * See Cycle bit rules. SW is the consumer for the event ring only.
+ * Don't make a ring full of link TRBs.  That would be dumb and this would loop.
+ *
+ * If we've just enqueued a TRB that is in the middle of a TD (meaning the
+ * chain bit is set), then set the chain bit in all the following link TRBs.
+ * If we've enqueued the last TRB in a TD, make sure the following link TRBs
+ * have their chain bit cleared (so that each Link TRB is a separate TD).
+ *
+ * Section 6.4.4.1 of the 0.95 spec says link TRBs cannot have the chain bit
+ * set, but other sections talk about dealing with the chain bit set.  This was
+ * fixed in the 0.96 specification errata, but we have to assume that all 0.95
+ * xHCI hardware can't handle the chain bit being cleared on a link TRB.
+ *
+ * @more_trbs_coming:  Will you enqueue more TRBs before calling
+ *                     prepare_transfer()?
+ */
+static void inc_enq(struct xhci_hcd *xhci, struct xhci_ring *ring,
+                       bool more_trbs_coming)
+{
+       u32 chain;
+       union xhci_trb *next;
+
+       chain = le32_to_cpu(ring->enqueue->generic.field[3]) & TRB_CHAIN;
+       /* If this is not event ring, there is one less usable TRB */
+       if (ring->type != TYPE_EVENT &&
+                       !last_trb(xhci, ring, ring->enq_seg, ring->enqueue))
+               ring->num_trbs_free--;
+       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)) {
+               if (ring->type != TYPE_EVENT) {
+                       /*
+                        * If the caller doesn't plan on enqueueing more
+                        * TDs before ringing the doorbell, then we
+                        * don't want to give the link TRB to the
+                        * hardware just yet.  We'll give the link TRB
+                        * back in prepare_ring() just before we enqueue
+                        * the TD at the top of the ring.
+                        */
+                       if (!chain && !more_trbs_coming)
+                               break;
+
+                       /* If we're not dealing with 0.95 hardware or
+                        * isoc rings on AMD 0.96 host,
+                        * carry over the chain bit of the previous TRB
+                        * (which may mean the chain bit is cleared).
+                        */
+                       if (!(ring->type == TYPE_ISOC &&
+                                       (xhci->quirks & XHCI_AMD_0x96_HOST))
+                                               && !xhci_link_trb_quirk(xhci)) {
+                               next->link.control &=
+                                       cpu_to_le32(~TRB_CHAIN);
+                               next->link.control |=
+                                       cpu_to_le32(chain);
+                       }
+                       /* Give this link TRB to the hardware */
+                       wmb();
+                       next->link.control ^= cpu_to_le32(TRB_CYCLE);
+
+                       /* Toggle the cycle bit after the last ring segment. */
+                       if (last_trb_on_last_seg(xhci, ring, ring->enq_seg, next)) {
+                               ring->cycle_state ^= 1;
+                       }
+               }
+               ring->enq_seg = ring->enq_seg->next;
+               ring->enqueue = ring->enq_seg->trbs;
+               next = ring->enqueue;
+       }
+}
+
+/*
+ * Check to see if there's room to enqueue num_trbs on the ring and make sure
+ * enqueue pointer will not advance into dequeue segment. See rules above.
+ */
+static inline int room_on_ring(struct xhci_hcd *xhci, struct xhci_ring *ring,
+               unsigned int num_trbs)
+{
+       int num_trbs_in_deq_seg;
+
+       if (ring->num_trbs_free < num_trbs)
+               return 0;
+
+       if (ring->type != TYPE_COMMAND && ring->type != TYPE_EVENT) {
+               num_trbs_in_deq_seg = ring->dequeue - ring->deq_seg->trbs;
+               if (ring->num_trbs_free < num_trbs + num_trbs_in_deq_seg)
+                       return 0;
+       }
+
+       return 1;
+}
+
+/* Ring the host controller doorbell after placing a command on the ring */
+void xhci_ring_cmd_db(struct xhci_hcd *xhci)
+{
+       if (!(xhci->cmd_ring_state & CMD_RING_STATE_RUNNING))
+               return;
+
+       xhci_dbg(xhci, "// Ding dong!\n");
+       writel(DB_VALUE_HOST, &xhci->dba->doorbell[0]);
+       /* Flush PCI posted writes */
+       readl(&xhci->dba->doorbell[0]);
+}
+
+static int xhci_abort_cmd_ring(struct xhci_hcd *xhci)
+{
+       u64 temp_64;
+       int ret;
+
+       xhci_dbg(xhci, "Abort command ring\n");
+
+       temp_64 = xhci_read_64(xhci, &xhci->op_regs->cmd_ring);
+       xhci->cmd_ring_state = CMD_RING_STATE_ABORTED;
+       xhci_write_64(xhci, temp_64 | CMD_RING_ABORT,
+                       &xhci->op_regs->cmd_ring);
+
+       /* Section 4.6.1.2 of xHCI 1.0 spec says software should
+        * time the completion od all xHCI commands, including
+        * the Command Abort operation. If software doesn't see
+        * CRR negated in a timely manner (e.g. longer than 5
+        * seconds), then it should assume that the there are
+        * larger problems with the xHC and assert HCRST.
+        */
+       ret = xhci_handshake(&xhci->op_regs->cmd_ring,
+                       CMD_RING_RUNNING, 0, 5 * 1000 * 1000);
+       if (ret < 0) {
+               xhci_err(xhci, "Stopped the command ring failed, "
+                               "maybe the host is dead\n");
+               xhci->xhc_state |= XHCI_STATE_DYING;
+               xhci_quiesce(xhci);
+               xhci_halt(xhci);
+               return -ESHUTDOWN;
+       }
+
+       return 0;
+}
+
+void xhci_ring_ep_doorbell(struct xhci_hcd *xhci,
+               unsigned int slot_id,
+               unsigned int ep_index,
+               unsigned int stream_id)
+{
+       __le32 __iomem *db_addr = &xhci->dba->doorbell[slot_id];
+       struct xhci_virt_ep *ep = &xhci->devs[slot_id]->eps[ep_index];
+       unsigned int ep_state = ep->ep_state;
+
+       /* Don't ring the doorbell for this endpoint if there are pending
+        * cancellations because we don't want to interrupt processing.
+        * We don't want to restart any stream rings if there's a set dequeue
+        * pointer command pending because the device can choose to start any
+        * stream once the endpoint is on the HW schedule.
+        */
+       if ((ep_state & EP_HALT_PENDING) || (ep_state & SET_DEQ_PENDING) ||
+           (ep_state & EP_HALTED))
+               return;
+       writel(DB_VALUE(ep_index, stream_id), db_addr);
+       /* The CPU has better things to do at this point than wait for a
+        * write-posting flush.  It'll get there soon enough.
+        */
+}
+
+/* Ring the doorbell for any rings with pending URBs */
+static void ring_doorbell_for_active_rings(struct xhci_hcd *xhci,
+               unsigned int slot_id,
+               unsigned int ep_index)
+{
+       unsigned int stream_id;
+       struct xhci_virt_ep *ep;
+
+       ep = &xhci->devs[slot_id]->eps[ep_index];
+
+       /* A ring has pending URBs if its TD list is not empty */
+       if (!(ep->ep_state & EP_HAS_STREAMS)) {
+               if (ep->ring && !(list_empty(&ep->ring->td_list)))
+                       xhci_ring_ep_doorbell(xhci, slot_id, ep_index, 0);
+               return;
+       }
+
+       for (stream_id = 1; stream_id < ep->stream_info->num_streams;
+                       stream_id++) {
+               struct xhci_stream_info *stream_info = ep->stream_info;
+               if (!list_empty(&stream_info->stream_rings[stream_id]->td_list))
+                       xhci_ring_ep_doorbell(xhci, slot_id, ep_index,
+                                               stream_id);
+       }
+}
+
+static struct xhci_ring *xhci_triad_to_transfer_ring(struct xhci_hcd *xhci,
+               unsigned int slot_id, unsigned int ep_index,
+               unsigned int stream_id)
+{
+       struct xhci_virt_ep *ep;
+
+       ep = &xhci->devs[slot_id]->eps[ep_index];
+       /* Common case: no streams */
+       if (!(ep->ep_state & EP_HAS_STREAMS))
+               return ep->ring;
+
+       if (stream_id == 0) {
+               xhci_warn(xhci,
+                               "WARN: Slot ID %u, ep index %u has streams, "
+                               "but URB has no stream ID.\n",
+                               slot_id, ep_index);
+               return NULL;
+       }
+
+       if (stream_id < ep->stream_info->num_streams)
+               return ep->stream_info->stream_rings[stream_id];
+
+       xhci_warn(xhci,
+                       "WARN: Slot ID %u, ep index %u has "
+                       "stream IDs 1 to %u allocated, "
+                       "but stream ID %u is requested.\n",
+                       slot_id, ep_index,
+                       ep->stream_info->num_streams - 1,
+                       stream_id);
+       return NULL;
+}
+
+/* Get the right ring for the given URB.
+ * If the endpoint supports streams, boundary check the URB's stream ID.
+ * If the endpoint doesn't support streams, return the singular endpoint ring.
+ */
+static struct xhci_ring *xhci_urb_to_transfer_ring(struct xhci_hcd *xhci,
+               struct urb *urb)
+{
+       return xhci_triad_to_transfer_ring(xhci, urb->dev->slot_id,
+               xhci_get_endpoint_index(&urb->ep->desc), urb->stream_id);
+}
+
+/*
+ * Move the xHC's endpoint ring dequeue pointer past cur_td.
+ * Record the new state of the xHC's endpoint ring dequeue segment,
+ * dequeue pointer, and new consumer cycle state in state.
+ * Update our internal representation of the ring's dequeue pointer.
+ *
+ * We do this in three jumps:
+ *  - First we update our new ring state to be the same as when the xHC stopped.
+ *  - Then we traverse the ring to find the segment that contains
+ *    the last TRB in the TD.  We toggle the xHC's new cycle state when we pass
+ *    any link TRBs with the toggle cycle bit set.
+ *  - Finally we move the dequeue state one TRB further, toggling the cycle bit
+ *    if we've moved it past a link TRB with the toggle cycle bit set.
+ *
+ * Some of the uses of xhci_generic_trb are grotty, but if they're done
+ * with correct __le32 accesses they should work fine.  Only users of this are
+ * in here.
+ */
+void xhci_find_new_dequeue_state(struct xhci_hcd *xhci,
+               unsigned int slot_id, unsigned int ep_index,
+               unsigned int stream_id, struct xhci_td *cur_td,
+               struct xhci_dequeue_state *state)
+{
+       struct xhci_virt_device *dev = xhci->devs[slot_id];
+       struct xhci_virt_ep *ep = &dev->eps[ep_index];
+       struct xhci_ring *ep_ring;
+       struct xhci_segment *new_seg;
+       union xhci_trb *new_deq;
+       dma_addr_t addr;
+       u64 hw_dequeue;
+       bool cycle_found = false;
+       bool td_last_trb_found = false;
+
+       ep_ring = xhci_triad_to_transfer_ring(xhci, slot_id,
+                       ep_index, stream_id);
+       if (!ep_ring) {
+               xhci_warn(xhci, "WARN can't find new dequeue state "
+                               "for invalid stream ID %u.\n",
+                               stream_id);
+               return;
+       }
+
+       /* Dig out the cycle state saved by the xHC during the stop ep cmd */
+       xhci_dbg_trace(xhci, trace_xhci_dbg_cancel_urb,
+                       "Finding endpoint context");
+       /* 4.6.9 the css flag is written to the stream context for streams */
+       if (ep->ep_state & EP_HAS_STREAMS) {
+               struct xhci_stream_ctx *ctx =
+                       &ep->stream_info->stream_ctx_array[stream_id];
+               hw_dequeue = le64_to_cpu(ctx->stream_ring);
+       } else {
+               struct xhci_ep_ctx *ep_ctx
+                       = xhci_get_ep_ctx(xhci, dev->out_ctx, ep_index);
+               hw_dequeue = le64_to_cpu(ep_ctx->deq);
+       }
+
+       new_seg = ep_ring->deq_seg;
+       new_deq = ep_ring->dequeue;
+       state->new_cycle_state = hw_dequeue & 0x1;
+
+       /*
+        * We want to find the pointer, segment and cycle state of the new trb
+        * (the one after current TD's last_trb). We know the cycle state at
+        * hw_dequeue, so walk the ring until both hw_dequeue and last_trb are
+        * found.
+        */
+       do {
+               if (!cycle_found && xhci_trb_virt_to_dma(new_seg, new_deq)
+                   == (dma_addr_t)(hw_dequeue & ~0xf)) {
+                       cycle_found = true;
+                       if (td_last_trb_found)
+                               break;
+               }
+               if (new_deq == cur_td->last_trb)
+                       td_last_trb_found = true;
+
+               if (cycle_found &&
+                   TRB_TYPE_LINK_LE32(new_deq->generic.field[3]) &&
+                   new_deq->generic.field[3] & cpu_to_le32(LINK_TOGGLE))
+                       state->new_cycle_state ^= 0x1;
+
+               next_trb(xhci, ep_ring, &new_seg, &new_deq);
+
+               /* Search wrapped around, bail out */
+               if (new_deq == ep->ring->dequeue) {
+                       xhci_err(xhci, "Error: Failed finding new dequeue state\n");
+                       state->new_deq_seg = NULL;
+                       state->new_deq_ptr = NULL;
+                       return;
+               }
+
+       } while (!cycle_found || !td_last_trb_found);
+
+       state->new_deq_seg = new_seg;
+       state->new_deq_ptr = new_deq;
+
+       /* Don't update the ring cycle state for the producer (us). */
+       xhci_dbg_trace(xhci, trace_xhci_dbg_cancel_urb,
+                       "Cycle state = 0x%x", state->new_cycle_state);
+
+       xhci_dbg_trace(xhci, trace_xhci_dbg_cancel_urb,
+                       "New dequeue segment = %p (virtual)",
+                       state->new_deq_seg);
+       addr = xhci_trb_virt_to_dma(state->new_deq_seg, state->new_deq_ptr);
+       xhci_dbg_trace(xhci, trace_xhci_dbg_cancel_urb,
+                       "New dequeue pointer = 0x%llx (DMA)",
+                       (unsigned long long) addr);
+}
+
+/* flip_cycle means flip the cycle bit of all but the first and last TRB.
+ * (The last TRB actually points to the ring enqueue pointer, which is not part
+ * of this TD.)  This is used to remove partially enqueued isoc TDs from a ring.
+ */
+static void td_to_noop(struct xhci_hcd *xhci, struct xhci_ring *ep_ring,
+               struct xhci_td *cur_td, bool flip_cycle)
+{
+       struct xhci_segment *cur_seg;
+       union xhci_trb *cur_trb;
+
+       for (cur_seg = cur_td->start_seg, cur_trb = cur_td->first_trb;
+                       true;
+                       next_trb(xhci, ep_ring, &cur_seg, &cur_trb)) {
+               if (TRB_TYPE_LINK_LE32(cur_trb->generic.field[3])) {
+                       /* Unchain any chained Link TRBs, but
+                        * leave the pointers intact.
+                        */
+                       cur_trb->generic.field[3] &= cpu_to_le32(~TRB_CHAIN);
+                       /* Flip the cycle bit (link TRBs can't be the first
+                        * or last TRB).
+                        */
+                       if (flip_cycle)
+                               cur_trb->generic.field[3] ^=
+                                       cpu_to_le32(TRB_CYCLE);
+                       xhci_dbg_trace(xhci, trace_xhci_dbg_cancel_urb,
+                                       "Cancel (unchain) link TRB");
+                       xhci_dbg_trace(xhci, trace_xhci_dbg_cancel_urb,
+                                       "Address = %p (0x%llx dma); "
+                                       "in seg %p (0x%llx dma)",
+                                       cur_trb,
+                                       (unsigned long long)xhci_trb_virt_to_dma(cur_seg, cur_trb),
+                                       cur_seg,
+                                       (unsigned long long)cur_seg->dma);
+               } else {
+                       cur_trb->generic.field[0] = 0;
+                       cur_trb->generic.field[1] = 0;
+                       cur_trb->generic.field[2] = 0;
+                       /* Preserve only the cycle bit of this TRB */
+                       cur_trb->generic.field[3] &= cpu_to_le32(TRB_CYCLE);
+                       /* Flip the cycle bit except on the first or last TRB */
+                       if (flip_cycle && cur_trb != cur_td->first_trb &&
+                                       cur_trb != cur_td->last_trb)
+                               cur_trb->generic.field[3] ^=
+                                       cpu_to_le32(TRB_CYCLE);
+                       cur_trb->generic.field[3] |= cpu_to_le32(
+                               TRB_TYPE(TRB_TR_NOOP));
+                       xhci_dbg_trace(xhci, trace_xhci_dbg_cancel_urb,
+                                       "TRB to noop at offset 0x%llx",
+                                       (unsigned long long)
+                                       xhci_trb_virt_to_dma(cur_seg, cur_trb));
+               }
+               if (cur_trb == cur_td->last_trb)
+                       break;
+       }
+}
+
+static void xhci_stop_watchdog_timer_in_irq(struct xhci_hcd *xhci,
+               struct xhci_virt_ep *ep)
+{
+       ep->ep_state &= ~EP_HALT_PENDING;
+       /* Can't del_timer_sync in interrupt, so we attempt to cancel.  If the
+        * timer is running on another CPU, we don't decrement stop_cmds_pending
+        * (since we didn't successfully stop the watchdog timer).
+        */
+       if (del_timer(&ep->stop_cmd_timer))
+               ep->stop_cmds_pending--;
+}
+
+/* Must be called with xhci->lock held in interrupt context */
+static void xhci_giveback_urb_in_irq(struct xhci_hcd *xhci,
+               struct xhci_td *cur_td, int status)
+{
+       struct usb_hcd *hcd;
+       struct urb      *urb;
+       struct urb_priv *urb_priv;
+
+       urb = cur_td->urb;
+       urb_priv = urb->hcpriv;
+       urb_priv->td_cnt++;
+       hcd = bus_to_hcd(urb->dev->bus);
+
+       /* Only giveback urb when this is the last td in urb */
+       if (urb_priv->td_cnt == urb_priv->length) {
+               if (usb_pipetype(urb->pipe) == PIPE_ISOCHRONOUS) {
+                       xhci_to_hcd(xhci)->self.bandwidth_isoc_reqs--;
+                       if (xhci_to_hcd(xhci)->self.bandwidth_isoc_reqs == 0) {
+                               if (xhci->quirks & XHCI_AMD_PLL_FIX)
+                                       usb_amd_quirk_pll_enable();
+                       }
+               }
+               usb_hcd_unlink_urb_from_ep(hcd, urb);
+
+               spin_unlock(&xhci->lock);
+               usb_hcd_giveback_urb(hcd, urb, status);
+               xhci_urb_free_priv(urb_priv);
+               spin_lock(&xhci->lock);
+       }
+}
+
+/*
+ * When we get a command completion for a Stop Endpoint Command, we need to
+ * unlink any cancelled TDs from the ring.  There are two ways to do that:
+ *
+ *  1. If the HW was in the middle of processing the TD that needs to be
+ *     cancelled, then we must move the ring's dequeue pointer past the last TRB
+ *     in the TD with a Set Dequeue Pointer Command.
+ *  2. Otherwise, we turn all the TRBs in the TD into No-op TRBs (with the chain
+ *     bit cleared) so that the HW will skip over them.
+ */
+static void xhci_handle_cmd_stop_ep(struct xhci_hcd *xhci, int slot_id,
+               union xhci_trb *trb, struct xhci_event_cmd *event)
+{
+       unsigned int ep_index;
+       struct xhci_ring *ep_ring;
+       struct xhci_virt_ep *ep;
+       struct list_head *entry;
+       struct xhci_td *cur_td = NULL;
+       struct xhci_td *last_unlinked_td;
+
+       struct xhci_dequeue_state deq_state;
+
+       if (unlikely(TRB_TO_SUSPEND_PORT(le32_to_cpu(trb->generic.field[3])))) {
+               if (!xhci->devs[slot_id])
+                       xhci_warn(xhci, "Stop endpoint command "
+                               "completion for disabled slot %u\n",
+                               slot_id);
+               return;
+       }
+
+       memset(&deq_state, 0, sizeof(deq_state));
+       ep_index = TRB_TO_EP_INDEX(le32_to_cpu(trb->generic.field[3]));
+       ep = &xhci->devs[slot_id]->eps[ep_index];
+
+       if (list_empty(&ep->cancelled_td_list)) {
+               xhci_stop_watchdog_timer_in_irq(xhci, ep);
+               ep->stopped_td = NULL;
+               ring_doorbell_for_active_rings(xhci, slot_id, ep_index);
+               return;
+       }
+
+       /* Fix up the ep ring first, so HW stops executing cancelled TDs.
+        * We have the xHCI lock, so nothing can modify this list until we drop
+        * it.  We're also in the event handler, so we can't get re-interrupted
+        * if another Stop Endpoint command completes
+        */
+       list_for_each(entry, &ep->cancelled_td_list) {
+               cur_td = list_entry(entry, struct xhci_td, cancelled_td_list);
+               xhci_dbg_trace(xhci, trace_xhci_dbg_cancel_urb,
+                               "Removing canceled TD starting at 0x%llx (dma).",
+                               (unsigned long long)xhci_trb_virt_to_dma(
+                                       cur_td->start_seg, cur_td->first_trb));
+               ep_ring = xhci_urb_to_transfer_ring(xhci, cur_td->urb);
+               if (!ep_ring) {
+                       /* This shouldn't happen unless a driver is mucking
+                        * with the stream ID after submission.  This will
+                        * leave the TD on the hardware ring, and the hardware
+                        * will try to execute it, and may access a buffer
+                        * that has already been freed.  In the best case, the
+                        * hardware will execute it, and the event handler will
+                        * ignore the completion event for that TD, since it was
+                        * removed from the td_list for that endpoint.  In
+                        * short, don't muck with the stream ID after
+                        * submission.
+                        */
+                       xhci_warn(xhci, "WARN Cancelled URB %p "
+                                       "has invalid stream ID %u.\n",
+                                       cur_td->urb,
+                                       cur_td->urb->stream_id);
+                       goto remove_finished_td;
+               }
+               /*
+                * If we stopped on the TD we need to cancel, then we have to
+                * move the xHC endpoint ring dequeue pointer past this TD.
+                */
+               if (cur_td == ep->stopped_td)
+                       xhci_find_new_dequeue_state(xhci, slot_id, ep_index,
+                                       cur_td->urb->stream_id,
+                                       cur_td, &deq_state);
+               else
+                       td_to_noop(xhci, ep_ring, cur_td, false);
+remove_finished_td:
+               /*
+                * The event handler won't see a completion for this TD anymore,
+                * so remove it from the endpoint ring's TD list.  Keep it in
+                * the cancelled TD list for URB completion later.
+                */
+               list_del_init(&cur_td->td_list);
+       }
+       last_unlinked_td = cur_td;
+       xhci_stop_watchdog_timer_in_irq(xhci, ep);
+
+       /* If necessary, queue a Set Transfer Ring Dequeue Pointer command */
+       if (deq_state.new_deq_ptr && deq_state.new_deq_seg) {
+               xhci_queue_new_dequeue_state(xhci, slot_id, ep_index,
+                               ep->stopped_td->urb->stream_id, &deq_state);
+               xhci_ring_cmd_db(xhci);
+       } else {
+               /* Otherwise ring the doorbell(s) to restart queued transfers */
+               ring_doorbell_for_active_rings(xhci, slot_id, ep_index);
+       }
+
+       ep->stopped_td = NULL;
+
+       /*
+        * Drop the lock and complete the URBs in the cancelled TD list.
+        * New TDs to be cancelled might be added to the end of the list before
+        * we can complete all the URBs for the TDs we already unlinked.
+        * So stop when we've completed the URB for the last TD we unlinked.
+        */
+       do {
+               cur_td = list_entry(ep->cancelled_td_list.next,
+                               struct xhci_td, cancelled_td_list);
+               list_del_init(&cur_td->cancelled_td_list);
+
+               /* Clean up the cancelled URB */
+               /* Doesn't matter what we pass for status, since the core will
+                * just overwrite it (because the URB has been unlinked).
+                */
+               xhci_giveback_urb_in_irq(xhci, cur_td, 0);
+
+               /* Stop processing the cancelled list if the watchdog timer is
+                * running.
+                */
+               if (xhci->xhc_state & XHCI_STATE_DYING)
+                       return;
+       } while (cur_td != last_unlinked_td);
+
+       /* Return to the event handler with xhci->lock re-acquired */
+}
+
+static void xhci_kill_ring_urbs(struct xhci_hcd *xhci, struct xhci_ring *ring)
+{
+       struct xhci_td *cur_td;
+
+       while (!list_empty(&ring->td_list)) {
+               cur_td = list_first_entry(&ring->td_list,
+                               struct xhci_td, td_list);
+               list_del_init(&cur_td->td_list);
+               if (!list_empty(&cur_td->cancelled_td_list))
+                       list_del_init(&cur_td->cancelled_td_list);
+               xhci_giveback_urb_in_irq(xhci, cur_td, -ESHUTDOWN);
+       }
+}
+
+static void xhci_kill_endpoint_urbs(struct xhci_hcd *xhci,
+               int slot_id, int ep_index)
+{
+       struct xhci_td *cur_td;
+       struct xhci_virt_ep *ep;
+       struct xhci_ring *ring;
+
+       ep = &xhci->devs[slot_id]->eps[ep_index];
+       if ((ep->ep_state & EP_HAS_STREAMS) ||
+                       (ep->ep_state & EP_GETTING_NO_STREAMS)) {
+               int stream_id;
+
+               for (stream_id = 0; stream_id < ep->stream_info->num_streams;
+                               stream_id++) {
+                       xhci_dbg_trace(xhci, trace_xhci_dbg_cancel_urb,
+                                       "Killing URBs for slot ID %u, ep index %u, stream %u",
+                                       slot_id, ep_index, stream_id + 1);
+                       xhci_kill_ring_urbs(xhci,
+                                       ep->stream_info->stream_rings[stream_id]);
+               }
+       } else {
+               ring = ep->ring;
+               if (!ring)
+                       return;
+               xhci_dbg_trace(xhci, trace_xhci_dbg_cancel_urb,
+                               "Killing URBs for slot ID %u, ep index %u",
+                               slot_id, ep_index);
+               xhci_kill_ring_urbs(xhci, ring);
+       }
+       while (!list_empty(&ep->cancelled_td_list)) {
+               cur_td = list_first_entry(&ep->cancelled_td_list,
+                               struct xhci_td, cancelled_td_list);
+               list_del_init(&cur_td->cancelled_td_list);
+               xhci_giveback_urb_in_irq(xhci, cur_td, -ESHUTDOWN);
+       }
+}
+
+/* Watchdog timer function for when a stop endpoint command fails to complete.
+ * In this case, we assume the host controller is broken or dying or dead.  The
+ * host may still be completing some other events, so we have to be careful to
+ * let the event ring handler and the URB dequeueing/enqueueing functions know
+ * through xhci->state.
+ *
+ * The timer may also fire if the host takes a very long time to respond to the
+ * command, and the stop endpoint command completion handler cannot delete the
+ * timer before the timer function is called.  Another endpoint cancellation may
+ * sneak in before the timer function can grab the lock, and that may queue
+ * another stop endpoint command and add the timer back.  So we cannot use a
+ * simple flag to say whether there is a pending stop endpoint command for a
+ * particular endpoint.
+ *
+ * Instead we use a combination of that flag and a counter for the number of
+ * pending stop endpoint commands.  If the timer is the tail end of the last
+ * stop endpoint command, and the endpoint's command is still pending, we assume
+ * the host is dying.
+ */
+void xhci_stop_endpoint_command_watchdog(unsigned long arg)
+{
+       struct xhci_hcd *xhci;
+       struct xhci_virt_ep *ep;
+       int ret, i, j;
+       unsigned long flags;
+
+       ep = (struct xhci_virt_ep *) arg;
+       xhci = ep->xhci;
+
+       spin_lock_irqsave(&xhci->lock, flags);
+
+       ep->stop_cmds_pending--;
+       if (xhci->xhc_state & XHCI_STATE_DYING) {
+               xhci_dbg_trace(xhci, trace_xhci_dbg_cancel_urb,
+                               "Stop EP timer ran, but another timer marked "
+                               "xHCI as DYING, exiting.");
+               spin_unlock_irqrestore(&xhci->lock, flags);
+               return;
+       }
+       if (!(ep->stop_cmds_pending == 0 && (ep->ep_state & EP_HALT_PENDING))) {
+               xhci_dbg_trace(xhci, trace_xhci_dbg_cancel_urb,
+                               "Stop EP timer ran, but no command pending, "
+                               "exiting.");
+               spin_unlock_irqrestore(&xhci->lock, flags);
+               return;
+       }
+
+       xhci_warn(xhci, "xHCI host not responding to stop endpoint command.\n");
+       xhci_warn(xhci, "Assuming host is dying, halting host.\n");
+       /* Oops, HC is dead or dying or at least not responding to the stop
+        * endpoint command.
+        */
+       xhci->xhc_state |= XHCI_STATE_DYING;
+       /* Disable interrupts from the host controller and start halting it */
+       xhci_quiesce(xhci);
+       spin_unlock_irqrestore(&xhci->lock, flags);
+
+       ret = xhci_halt(xhci);
+
+       spin_lock_irqsave(&xhci->lock, flags);
+       if (ret < 0) {
+               /* This is bad; the host is not responding to commands and it's
+                * not allowing itself to be halted.  At least interrupts are
+                * disabled. If we call usb_hc_died(), it will attempt to
+                * disconnect all device drivers under this host.  Those
+                * disconnect() methods will wait for all URBs to be unlinked,
+                * so we must complete them.
+                */
+               xhci_warn(xhci, "Non-responsive xHCI host is not halting.\n");
+               xhci_warn(xhci, "Completing active URBs anyway.\n");
+               /* We could turn all TDs on the rings to no-ops.  This won't
+                * help if the host has cached part of the ring, and is slow if
+                * we want to preserve the cycle bit.  Skip it and hope the host
+                * doesn't touch the memory.
+                */
+       }
+       for (i = 0; i < MAX_HC_SLOTS; i++) {
+               if (!xhci->devs[i])
+                       continue;
+               for (j = 0; j < 31; j++)
+                       xhci_kill_endpoint_urbs(xhci, i, j);
+       }
+       spin_unlock_irqrestore(&xhci->lock, flags);
+       xhci_dbg_trace(xhci, trace_xhci_dbg_cancel_urb,
+                       "Calling usb_hc_died()");
+       usb_hc_died(xhci_to_hcd(xhci)->primary_hcd);
+       xhci_dbg_trace(xhci, trace_xhci_dbg_cancel_urb,
+                       "xHCI host controller is dead.");
+}
+
+
+static void update_ring_for_set_deq_completion(struct xhci_hcd *xhci,
+               struct xhci_virt_device *dev,
+               struct xhci_ring *ep_ring,
+               unsigned int ep_index)
+{
+       union xhci_trb *dequeue_temp;
+       int num_trbs_free_temp;
+       bool revert = false;
+
+       num_trbs_free_temp = ep_ring->num_trbs_free;
+       dequeue_temp = ep_ring->dequeue;
+
+       /* If we get two back-to-back stalls, and the first stalled transfer
+        * ends just before a link TRB, the dequeue pointer will be left on
+        * the link TRB by the code in the while loop.  So we have to update
+        * the dequeue pointer one segment further, or we'll jump off
+        * the segment into la-la-land.
+        */
+       if (last_trb(xhci, ep_ring, ep_ring->deq_seg, ep_ring->dequeue)) {
+               ep_ring->deq_seg = ep_ring->deq_seg->next;
+               ep_ring->dequeue = ep_ring->deq_seg->trbs;
+       }
+
+       while (ep_ring->dequeue != dev->eps[ep_index].queued_deq_ptr) {
+               /* We have more usable TRBs */
+               ep_ring->num_trbs_free++;
+               ep_ring->dequeue++;
+               if (last_trb(xhci, ep_ring, ep_ring->deq_seg,
+                               ep_ring->dequeue)) {
+                       if (ep_ring->dequeue ==
+                                       dev->eps[ep_index].queued_deq_ptr)
+                               break;
+                       ep_ring->deq_seg = ep_ring->deq_seg->next;
+                       ep_ring->dequeue = ep_ring->deq_seg->trbs;
+               }
+               if (ep_ring->dequeue == dequeue_temp) {
+                       revert = true;
+                       break;
+               }
+       }
+
+       if (revert) {
+               xhci_dbg(xhci, "Unable to find new dequeue pointer\n");
+               ep_ring->num_trbs_free = num_trbs_free_temp;
+       }
+}
+
+/*
+ * When we get a completion for a Set Transfer Ring Dequeue Pointer command,
+ * we need to clear the set deq pending flag in the endpoint ring state, so that
+ * the TD queueing code can ring the doorbell again.  We also need to ring the
+ * endpoint doorbell to restart the ring, but only if there aren't more
+ * cancellations pending.
+ */
+static void xhci_handle_cmd_set_deq(struct xhci_hcd *xhci, int slot_id,
+               union xhci_trb *trb, u32 cmd_comp_code)
+{
+       unsigned int ep_index;
+       unsigned int stream_id;
+       struct xhci_ring *ep_ring;
+       struct xhci_virt_device *dev;
+       struct xhci_virt_ep *ep;
+       struct xhci_ep_ctx *ep_ctx;
+       struct xhci_slot_ctx *slot_ctx;
+
+       ep_index = TRB_TO_EP_INDEX(le32_to_cpu(trb->generic.field[3]));
+       stream_id = TRB_TO_STREAM_ID(le32_to_cpu(trb->generic.field[2]));
+       dev = xhci->devs[slot_id];
+       ep = &dev->eps[ep_index];
+
+       ep_ring = xhci_stream_id_to_ring(dev, ep_index, stream_id);
+       if (!ep_ring) {
+               xhci_warn(xhci, "WARN Set TR deq ptr command for freed stream ID %u\n",
+                               stream_id);
+               /* XXX: Harmless??? */
+               goto cleanup;
+       }
+
+       ep_ctx = xhci_get_ep_ctx(xhci, dev->out_ctx, ep_index);
+       slot_ctx = xhci_get_slot_ctx(xhci, dev->out_ctx);
+
+       if (cmd_comp_code != COMP_SUCCESS) {
+               unsigned int ep_state;
+               unsigned int slot_state;
+
+               switch (cmd_comp_code) {
+               case COMP_TRB_ERR:
+                       xhci_warn(xhci, "WARN Set TR Deq Ptr cmd invalid because of stream ID configuration\n");
+                       break;
+               case COMP_CTX_STATE:
+                       xhci_warn(xhci, "WARN Set TR Deq Ptr cmd failed due to incorrect slot or ep state.\n");
+                       ep_state = le32_to_cpu(ep_ctx->ep_info);
+                       ep_state &= EP_STATE_MASK;
+                       slot_state = le32_to_cpu(slot_ctx->dev_state);
+                       slot_state = GET_SLOT_STATE(slot_state);
+                       xhci_dbg_trace(xhci, trace_xhci_dbg_cancel_urb,
+                                       "Slot state = %u, EP state = %u",
+                                       slot_state, ep_state);
+                       break;
+               case COMP_EBADSLT:
+                       xhci_warn(xhci, "WARN Set TR Deq Ptr cmd failed because slot %u was not enabled.\n",
+                                       slot_id);
+                       break;
+               default:
+                       xhci_warn(xhci, "WARN Set TR Deq Ptr cmd with unknown completion code of %u.\n",
+                                       cmd_comp_code);
+                       break;
+               }
+               /* OK what do we do now?  The endpoint state is hosed, and we
+                * should never get to this point if the synchronization between
+                * queueing, and endpoint state are correct.  This might happen
+                * if the device gets disconnected after we've finished
+                * cancelling URBs, which might not be an error...
+                */
+       } else {
+               u64 deq;
+               /* 4.6.10 deq ptr is written to the stream ctx for streams */
+               if (ep->ep_state & EP_HAS_STREAMS) {
+                       struct xhci_stream_ctx *ctx =
+                               &ep->stream_info->stream_ctx_array[stream_id];
+                       deq = le64_to_cpu(ctx->stream_ring) & SCTX_DEQ_MASK;
+               } else {
+                       deq = le64_to_cpu(ep_ctx->deq) & ~EP_CTX_CYCLE_MASK;
+               }
+               xhci_dbg_trace(xhci, trace_xhci_dbg_cancel_urb,
+                       "Successful Set TR Deq Ptr cmd, deq = @%08llx", deq);
+               if (xhci_trb_virt_to_dma(ep->queued_deq_seg,
+                                        ep->queued_deq_ptr) == deq) {
+                       /* Update the ring's dequeue segment and dequeue pointer
+                        * to reflect the new position.
+                        */
+                       update_ring_for_set_deq_completion(xhci, dev,
+                               ep_ring, ep_index);
+               } else {
+                       xhci_warn(xhci, "Mismatch between completed Set TR Deq Ptr command & xHCI internal state.\n");
+                       xhci_warn(xhci, "ep deq seg = %p, deq ptr = %p\n",
+                                 ep->queued_deq_seg, ep->queued_deq_ptr);
+               }
+       }
+
+cleanup:
+       dev->eps[ep_index].ep_state &= ~SET_DEQ_PENDING;
+       dev->eps[ep_index].queued_deq_seg = NULL;
+       dev->eps[ep_index].queued_deq_ptr = NULL;
+       /* Restart any rings with pending URBs */
+       ring_doorbell_for_active_rings(xhci, slot_id, ep_index);
+}
+
+static void xhci_handle_cmd_reset_ep(struct xhci_hcd *xhci, int slot_id,
+               union xhci_trb *trb, u32 cmd_comp_code)
+{
+       unsigned int ep_index;
+
+       ep_index = TRB_TO_EP_INDEX(le32_to_cpu(trb->generic.field[3]));
+       /* This command will only fail if the endpoint wasn't halted,
+        * but we don't care.
+        */
+       xhci_dbg_trace(xhci, trace_xhci_dbg_reset_ep,
+               "Ignoring reset ep completion code of %u", cmd_comp_code);
+
+       /* HW with the reset endpoint quirk needs to have a configure endpoint
+        * command complete before the endpoint can be used.  Queue that here
+        * because the HW can't handle two commands being queued in a row.
+        */
+       if (xhci->quirks & XHCI_RESET_EP_QUIRK) {
+               struct xhci_command *command;
+               command = xhci_alloc_command(xhci, false, false, GFP_ATOMIC);
+               if (!command) {
+                       xhci_warn(xhci, "WARN Cannot submit cfg ep: ENOMEM\n");
+                       return;
+               }
+               xhci_dbg_trace(xhci, trace_xhci_dbg_quirks,
+                               "Queueing configure endpoint command");
+               xhci_queue_configure_endpoint(xhci, command,
+                               xhci->devs[slot_id]->in_ctx->dma, slot_id,
+                               false);
+               xhci_ring_cmd_db(xhci);
+       } else {
+               /* Clear our internal halted state */
+               xhci->devs[slot_id]->eps[ep_index].ep_state &= ~EP_HALTED;
+       }
+}
+
+static void xhci_handle_cmd_enable_slot(struct xhci_hcd *xhci, int slot_id,
+               u32 cmd_comp_code)
+{
+       if (cmd_comp_code == COMP_SUCCESS)
+               xhci->slot_id = slot_id;
+       else
+               xhci->slot_id = 0;
+}
+
+static void xhci_handle_cmd_disable_slot(struct xhci_hcd *xhci, int slot_id)
+{
+       struct xhci_virt_device *virt_dev;
+
+       virt_dev = xhci->devs[slot_id];
+       if (!virt_dev)
+               return;
+       if (xhci->quirks & XHCI_EP_LIMIT_QUIRK)
+               /* Delete default control endpoint resources */
+               xhci_free_device_endpoint_resources(xhci, virt_dev, true);
+       xhci_free_virt_device(xhci, slot_id);
+}
+
+static void xhci_handle_cmd_config_ep(struct xhci_hcd *xhci, int slot_id,
+               struct xhci_event_cmd *event, u32 cmd_comp_code)
+{
+       struct xhci_virt_device *virt_dev;
+       struct xhci_input_control_ctx *ctrl_ctx;
+       unsigned int ep_index;
+       unsigned int ep_state;
+       u32 add_flags, drop_flags;
+
+       /*
+        * Configure endpoint commands can come from the USB core
+        * configuration or alt setting changes, or because the HW
+        * needed an extra configure endpoint command after a reset
+        * endpoint command or streams were being configured.
+        * If the command was for a halted endpoint, the xHCI driver
+        * is not waiting on the configure endpoint command.
+        */
+       virt_dev = xhci->devs[slot_id];
+       ctrl_ctx = xhci_get_input_control_ctx(virt_dev->in_ctx);
+       if (!ctrl_ctx) {
+               xhci_warn(xhci, "Could not get input context, bad type.\n");
+               return;
+       }
+
+       add_flags = le32_to_cpu(ctrl_ctx->add_flags);
+       drop_flags = le32_to_cpu(ctrl_ctx->drop_flags);
+       /* Input ctx add_flags are the endpoint index plus one */
+       ep_index = xhci_last_valid_endpoint(add_flags) - 1;
+
+       /* A usb_set_interface() call directly after clearing a halted
+        * condition may race on this quirky hardware.  Not worth
+        * worrying about, since this is prototype hardware.  Not sure
+        * if this will work for streams, but streams support was
+        * untested on this prototype.
+        */
+       if (xhci->quirks & XHCI_RESET_EP_QUIRK &&
+                       ep_index != (unsigned int) -1 &&
+                       add_flags - SLOT_FLAG == drop_flags) {
+               ep_state = virt_dev->eps[ep_index].ep_state;
+               if (!(ep_state & EP_HALTED))
+                       return;
+               xhci_dbg_trace(xhci, trace_xhci_dbg_quirks,
+                               "Completed config ep cmd - "
+                               "last ep index = %d, state = %d",
+                               ep_index, ep_state);
+               /* Clear internal halted state and restart ring(s) */
+               virt_dev->eps[ep_index].ep_state &= ~EP_HALTED;
+               ring_doorbell_for_active_rings(xhci, slot_id, ep_index);
+               return;
+       }
+       return;
+}
+
+static void xhci_handle_cmd_reset_dev(struct xhci_hcd *xhci, int slot_id,
+               struct xhci_event_cmd *event)
+{
+       xhci_dbg(xhci, "Completed reset device command.\n");
+       if (!xhci->devs[slot_id])
+               xhci_warn(xhci, "Reset device command completion "
+                               "for disabled slot %u\n", slot_id);
+}
+
+static void xhci_handle_cmd_nec_get_fw(struct xhci_hcd *xhci,
+               struct xhci_event_cmd *event)
+{
+       if (!(xhci->quirks & XHCI_NEC_HOST)) {
+               xhci->error_bitmask |= 1 << 6;
+               return;
+       }
+       xhci_dbg_trace(xhci, trace_xhci_dbg_quirks,
+                       "NEC firmware version %2x.%02x",
+                       NEC_FW_MAJOR(le32_to_cpu(event->status)),
+                       NEC_FW_MINOR(le32_to_cpu(event->status)));
+}
+
+static void xhci_complete_del_and_free_cmd(struct xhci_command *cmd, u32 status)
+{
+       list_del(&cmd->cmd_list);
+
+       if (cmd->completion) {
+               cmd->status = status;
+               complete(cmd->completion);
+       } else {
+               kfree(cmd);
+       }
+}
+
+void xhci_cleanup_command_queue(struct xhci_hcd *xhci)
+{
+       struct xhci_command *cur_cmd, *tmp_cmd;
+       list_for_each_entry_safe(cur_cmd, tmp_cmd, &xhci->cmd_list, cmd_list)
+               xhci_complete_del_and_free_cmd(cur_cmd, COMP_CMD_ABORT);
+}
+
+/*
+ * Turn all commands on command ring with status set to "aborted" to no-op trbs.
+ * If there are other commands waiting then restart the ring and kick the timer.
+ * This must be called with command ring stopped and xhci->lock held.
+ */
+static void xhci_handle_stopped_cmd_ring(struct xhci_hcd *xhci,
+                                        struct xhci_command *cur_cmd)
+{
+       struct xhci_command *i_cmd, *tmp_cmd;
+       u32 cycle_state;
+
+       /* Turn all aborted commands in list to no-ops, then restart */
+       list_for_each_entry_safe(i_cmd, tmp_cmd, &xhci->cmd_list,
+                                cmd_list) {
+
+               if (i_cmd->status != COMP_CMD_ABORT)
+                       continue;
+
+               i_cmd->status = COMP_CMD_STOP;
+
+               xhci_dbg(xhci, "Turn aborted command %p to no-op\n",
+                        i_cmd->command_trb);
+               /* get cycle state from the original cmd trb */
+               cycle_state = le32_to_cpu(
+                       i_cmd->command_trb->generic.field[3]) & TRB_CYCLE;
+               /* modify the command trb to no-op command */
+               i_cmd->command_trb->generic.field[0] = 0;
+               i_cmd->command_trb->generic.field[1] = 0;
+               i_cmd->command_trb->generic.field[2] = 0;
+               i_cmd->command_trb->generic.field[3] = cpu_to_le32(
+                       TRB_TYPE(TRB_CMD_NOOP) | cycle_state);
+
+               /*
+                * caller waiting for completion is called when command
+                *  completion event is received for these no-op commands
+                */
+       }
+
+       xhci->cmd_ring_state = CMD_RING_STATE_RUNNING;
+
+       /* ring command ring doorbell to restart the command ring */
+       if ((xhci->cmd_ring->dequeue != xhci->cmd_ring->enqueue) &&
+           !(xhci->xhc_state & XHCI_STATE_DYING)) {
+               xhci->current_cmd = cur_cmd;
+               mod_timer(&xhci->cmd_timer, jiffies + XHCI_CMD_DEFAULT_TIMEOUT);
+               xhci_ring_cmd_db(xhci);
+       }
+       return;
+}
+
+
+void xhci_handle_command_timeout(unsigned long data)
+{
+       struct xhci_hcd *xhci;
+       int ret;
+       unsigned long flags;
+       u64 hw_ring_state;
+       struct xhci_command *cur_cmd = NULL;
+       xhci = (struct xhci_hcd *) data;
+
+       /* mark this command to be cancelled */
+       spin_lock_irqsave(&xhci->lock, flags);
+       if (xhci->current_cmd) {
+               cur_cmd = xhci->current_cmd;
+               cur_cmd->status = COMP_CMD_ABORT;
+       }
+
+
+       /* Make sure command ring is running before aborting it */
+       hw_ring_state = xhci_read_64(xhci, &xhci->op_regs->cmd_ring);
+       if ((xhci->cmd_ring_state & CMD_RING_STATE_RUNNING) &&
+           (hw_ring_state & CMD_RING_RUNNING))  {
+
+               spin_unlock_irqrestore(&xhci->lock, flags);
+               xhci_dbg(xhci, "Command timeout\n");
+               ret = xhci_abort_cmd_ring(xhci);
+               if (unlikely(ret == -ESHUTDOWN)) {
+                       xhci_err(xhci, "Abort command ring failed\n");
+                       xhci_cleanup_command_queue(xhci);
+                       usb_hc_died(xhci_to_hcd(xhci)->primary_hcd);
+                       xhci_dbg(xhci, "xHCI host controller is dead.\n");
+               }
+               return;
+       }
+       /* command timeout on stopped ring, ring can't be aborted */
+       xhci_dbg(xhci, "Command timeout on stopped ring\n");
+       xhci_handle_stopped_cmd_ring(xhci, xhci->current_cmd);
+       spin_unlock_irqrestore(&xhci->lock, flags);
+       return;
+}
+
+static void handle_cmd_completion(struct xhci_hcd *xhci,
+               struct xhci_event_cmd *event)
+{
+       int slot_id = TRB_TO_SLOT_ID(le32_to_cpu(event->flags));
+       u64 cmd_dma;
+       dma_addr_t cmd_dequeue_dma;
+       u32 cmd_comp_code;
+       union xhci_trb *cmd_trb;
+       struct xhci_command *cmd;
+       u32 cmd_type;
+
+       cmd_dma = le64_to_cpu(event->cmd_trb);
+       cmd_trb = xhci->cmd_ring->dequeue;
+       cmd_dequeue_dma = xhci_trb_virt_to_dma(xhci->cmd_ring->deq_seg,
+                       cmd_trb);
+       /* Is the command ring deq ptr out of sync with the deq seg ptr? */
+       if (cmd_dequeue_dma == 0) {
+               xhci->error_bitmask |= 1 << 4;
+               return;
+       }
+       /* Does the DMA address match our internal dequeue pointer address? */
+       if (cmd_dma != (u64) cmd_dequeue_dma) {
+               xhci->error_bitmask |= 1 << 5;
+               return;
+       }
+
+       cmd = list_entry(xhci->cmd_list.next, struct xhci_command, cmd_list);
+
+       if (cmd->command_trb != xhci->cmd_ring->dequeue) {
+               xhci_err(xhci,
+                        "Command completion event does not match command\n");
+               return;
+       }
+
+       del_timer(&xhci->cmd_timer);
+
+       trace_xhci_cmd_completion(cmd_trb, (struct xhci_generic_trb *) event);
+
+       cmd_comp_code = GET_COMP_CODE(le32_to_cpu(event->status));
+
+       /* If CMD ring stopped we own the trbs between enqueue and dequeue */
+       if (cmd_comp_code == COMP_CMD_STOP) {
+               xhci_handle_stopped_cmd_ring(xhci, cmd);
+               return;
+       }
+       /*
+        * Host aborted the command ring, check if the current command was
+        * supposed to be aborted, otherwise continue normally.
+        * The command ring is stopped now, but the xHC will issue a Command
+        * Ring Stopped event which will cause us to restart it.
+        */
+       if (cmd_comp_code == COMP_CMD_ABORT) {
+               xhci->cmd_ring_state = CMD_RING_STATE_STOPPED;
+               if (cmd->status == COMP_CMD_ABORT)
+                       goto event_handled;
+       }
+
+       cmd_type = TRB_FIELD_TO_TYPE(le32_to_cpu(cmd_trb->generic.field[3]));
+       switch (cmd_type) {
+       case TRB_ENABLE_SLOT:
+               xhci_handle_cmd_enable_slot(xhci, slot_id, cmd_comp_code);
+               break;
+       case TRB_DISABLE_SLOT:
+               xhci_handle_cmd_disable_slot(xhci, slot_id);
+               break;
+       case TRB_CONFIG_EP:
+               if (!cmd->completion)
+                       xhci_handle_cmd_config_ep(xhci, slot_id, event,
+                                                 cmd_comp_code);
+               break;
+       case TRB_EVAL_CONTEXT:
+               break;
+       case TRB_ADDR_DEV:
+               break;
+       case TRB_STOP_RING:
+               WARN_ON(slot_id != TRB_TO_SLOT_ID(
+                               le32_to_cpu(cmd_trb->generic.field[3])));
+               xhci_handle_cmd_stop_ep(xhci, slot_id, cmd_trb, event);
+               break;
+       case TRB_SET_DEQ:
+               WARN_ON(slot_id != TRB_TO_SLOT_ID(
+                               le32_to_cpu(cmd_trb->generic.field[3])));
+               xhci_handle_cmd_set_deq(xhci, slot_id, cmd_trb, cmd_comp_code);
+               break;
+       case TRB_CMD_NOOP:
+               /* Is this an aborted command turned to NO-OP? */
+               if (cmd->status == COMP_CMD_STOP)
+                       cmd_comp_code = COMP_CMD_STOP;
+               break;
+       case TRB_RESET_EP:
+               WARN_ON(slot_id != TRB_TO_SLOT_ID(
+                               le32_to_cpu(cmd_trb->generic.field[3])));
+               xhci_handle_cmd_reset_ep(xhci, slot_id, cmd_trb, cmd_comp_code);
+               break;
+       case TRB_RESET_DEV:
+               /* SLOT_ID field in reset device cmd completion event TRB is 0.
+                * Use the SLOT_ID from the command TRB instead (xhci 4.6.11)
+                */
+               slot_id = TRB_TO_SLOT_ID(
+                               le32_to_cpu(cmd_trb->generic.field[3]));
+               xhci_handle_cmd_reset_dev(xhci, slot_id, event);
+               break;
+       case TRB_NEC_GET_FW:
+               xhci_handle_cmd_nec_get_fw(xhci, event);
+               break;
+       default:
+               /* Skip over unknown commands on the event ring */
+               xhci->error_bitmask |= 1 << 6;
+               break;
+       }
+
+       /* restart timer if this wasn't the last command */
+       if (cmd->cmd_list.next != &xhci->cmd_list) {
+               xhci->current_cmd = list_entry(cmd->cmd_list.next,
+                                              struct xhci_command, cmd_list);
+               mod_timer(&xhci->cmd_timer, jiffies + XHCI_CMD_DEFAULT_TIMEOUT);
+       }
+
+event_handled:
+       xhci_complete_del_and_free_cmd(cmd, cmd_comp_code);
+
+       inc_deq(xhci, xhci->cmd_ring);
+}
+
+static void handle_vendor_event(struct xhci_hcd *xhci,
+               union xhci_trb *event)
+{
+       u32 trb_type;
+
+       trb_type = TRB_FIELD_TO_TYPE(le32_to_cpu(event->generic.field[3]));
+       xhci_dbg(xhci, "Vendor specific event TRB type = %u\n", trb_type);
+       if (trb_type == TRB_NEC_CMD_COMP && (xhci->quirks & XHCI_NEC_HOST))
+               handle_cmd_completion(xhci, &event->event_cmd);
+}
+
+/* @port_id: the one-based port ID from the hardware (indexed from array of all
+ * port registers -- USB 3.0 and USB 2.0).
+ *
+ * Returns a zero-based port number, which is suitable for indexing into each of
+ * the split roothubs' port arrays and bus state arrays.
+ * Add one to it in order to call xhci_find_slot_id_by_port.
+ */
+static unsigned int find_faked_portnum_from_hw_portnum(struct usb_hcd *hcd,
+               struct xhci_hcd *xhci, u32 port_id)
+{
+       unsigned int i;
+       unsigned int num_similar_speed_ports = 0;
+
+       /* port_id from the hardware is 1-based, but port_array[], usb3_ports[],
+        * and usb2_ports are 0-based indexes.  Count the number of similar
+        * speed ports, up to 1 port before this port.
+        */
+       for (i = 0; i < (port_id - 1); i++) {
+               u8 port_speed = xhci->port_array[i];
+
+               /*
+                * Skip ports that don't have known speeds, or have duplicate
+                * Extended Capabilities port speed entries.
+                */
+               if (port_speed == 0 || port_speed == DUPLICATE_ENTRY)
+                       continue;
+
+               /*
+                * USB 3.0 ports are always under a USB 3.0 hub.  USB 2.0 and
+                * 1.1 ports are under the USB 2.0 hub.  If the port speed
+                * matches the device speed, it's a similar speed port.
+                */
+               if ((port_speed == 0x03) == (hcd->speed == HCD_USB3))
+                       num_similar_speed_ports++;
+       }
+       return num_similar_speed_ports;
+}
+
+static void handle_device_notification(struct xhci_hcd *xhci,
+               union xhci_trb *event)
+{
+       u32 slot_id;
+       struct usb_device *udev;
+
+       slot_id = TRB_TO_SLOT_ID(le32_to_cpu(event->generic.field[3]));
+       if (!xhci->devs[slot_id]) {
+               xhci_warn(xhci, "Device Notification event for "
+                               "unused slot %u\n", slot_id);
+               return;
+       }
+
+       xhci_dbg(xhci, "Device Wake Notification event for slot ID %u\n",
+                       slot_id);
+       udev = xhci->devs[slot_id]->udev;
+       if (udev && udev->parent)
+               usb_wakeup_notification(udev->parent, udev->portnum);
+}
+
+static void handle_port_status(struct xhci_hcd *xhci,
+               union xhci_trb *event)
+{
+       struct usb_hcd *hcd;
+       u32 port_id;
+       u32 temp, temp1;
+       int max_ports;
+       int slot_id;
+       unsigned int faked_port_index;
+       u8 major_revision;
+       struct xhci_bus_state *bus_state;
+       __le32 __iomem **port_array;
+       bool bogus_port_status = false;
+
+       /* Port status change events always have a successful completion code */
+       if (GET_COMP_CODE(le32_to_cpu(event->generic.field[2])) != COMP_SUCCESS) {
+               xhci_warn(xhci, "WARN: xHC returned failed port status event\n");
+               xhci->error_bitmask |= 1 << 8;
+       }
+       port_id = GET_PORT_ID(le32_to_cpu(event->generic.field[0]));
+       xhci_dbg(xhci, "Port Status Change Event for port %d\n", port_id);
+
+       max_ports = HCS_MAX_PORTS(xhci->hcs_params1);
+       if ((port_id <= 0) || (port_id > max_ports)) {
+               xhci_warn(xhci, "Invalid port id %d\n", port_id);
+               inc_deq(xhci, xhci->event_ring);
+               return;
+       }
+
+       /* Figure out which usb_hcd this port is attached to:
+        * is it a USB 3.0 port or a USB 2.0/1.1 port?
+        */
+       major_revision = xhci->port_array[port_id - 1];
+
+       /* Find the right roothub. */
+       hcd = xhci_to_hcd(xhci);
+       if ((major_revision == 0x03) != (hcd->speed == HCD_USB3))
+               hcd = xhci->shared_hcd;
+
+       if (major_revision == 0) {
+               xhci_warn(xhci, "Event for port %u not in "
+                               "Extended Capabilities, ignoring.\n",
+                               port_id);
+               bogus_port_status = true;
+               goto cleanup;
+       }
+       if (major_revision == DUPLICATE_ENTRY) {
+               xhci_warn(xhci, "Event for port %u duplicated in"
+                               "Extended Capabilities, ignoring.\n",
+                               port_id);
+               bogus_port_status = true;
+               goto cleanup;
+       }
+
+       /*
+        * Hardware port IDs reported by a Port Status Change Event include USB
+        * 3.0 and USB 2.0 ports.  We want to check if the port has reported a
+        * resume event, but we first need to translate the hardware port ID
+        * into the index into the ports on the correct split roothub, and the
+        * correct bus_state structure.
+        */
+       bus_state = &xhci->bus_state[hcd_index(hcd)];
+       if (hcd->speed == HCD_USB3)
+               port_array = xhci->usb3_ports;
+       else
+               port_array = xhci->usb2_ports;
+       /* Find the faked port hub number */
+       faked_port_index = find_faked_portnum_from_hw_portnum(hcd, xhci,
+                       port_id);
+
+       temp = readl(port_array[faked_port_index]);
+       if (hcd->state == HC_STATE_SUSPENDED) {
+               xhci_dbg(xhci, "resume root hub\n");
+               usb_hcd_resume_root_hub(hcd);
+       }
+
+       if ((temp & PORT_PLC) && (temp & PORT_PLS_MASK) == XDEV_RESUME) {
+               xhci_dbg(xhci, "port resume event for port %d\n", port_id);
+
+               temp1 = readl(&xhci->op_regs->command);
+               if (!(temp1 & CMD_RUN)) {
+                       xhci_warn(xhci, "xHC is not running.\n");
+                       goto cleanup;
+               }
+
+               if (DEV_SUPERSPEED(temp)) {
+                       xhci_dbg(xhci, "remote wake SS port %d\n", port_id);
+                       /* Set a flag to say the port signaled remote wakeup,
+                        * so we can tell the difference between the end of
+                        * device and host initiated resume.
+                        */
+                       bus_state->port_remote_wakeup |= 1 << faked_port_index;
+                       xhci_test_and_clear_bit(xhci, port_array,
+                                       faked_port_index, PORT_PLC);
+                       xhci_set_link_state(xhci, port_array, faked_port_index,
+                                               XDEV_U0);
+                       /* Need to wait until the next link state change
+                        * indicates the device is actually in U0.
+                        */
+                       bogus_port_status = true;
+                       goto cleanup;
+               } else {
+                       xhci_dbg(xhci, "resume HS port %d\n", port_id);
+                       bus_state->resume_done[faked_port_index] = jiffies +
+                               msecs_to_jiffies(USB_RESUME_TIMEOUT);
+                       set_bit(faked_port_index, &bus_state->resuming_ports);
+                       mod_timer(&hcd->rh_timer,
+                                 bus_state->resume_done[faked_port_index]);
+                       /* Do the rest in GetPortStatus */
+               }
+       }
+
+       if ((temp & PORT_PLC) && (temp & PORT_PLS_MASK) == XDEV_U0 &&
+                       DEV_SUPERSPEED(temp)) {
+               xhci_dbg(xhci, "resume SS port %d finished\n", port_id);
+               /* We've just brought the device into U0 through either the
+                * Resume state after a device remote wakeup, or through the
+                * U3Exit state after a host-initiated resume.  If it's a device
+                * initiated remote wake, don't pass up the link state change,
+                * so the roothub behavior is consistent with external
+                * USB 3.0 hub behavior.
+                */
+               slot_id = xhci_find_slot_id_by_port(hcd, xhci,
+                               faked_port_index + 1);
+               if (slot_id && xhci->devs[slot_id])
+                       xhci_ring_device(xhci, slot_id);
+               if (bus_state->port_remote_wakeup & (1 << faked_port_index)) {
+                       bus_state->port_remote_wakeup &=
+                               ~(1 << faked_port_index);
+                       xhci_test_and_clear_bit(xhci, port_array,
+                                       faked_port_index, PORT_PLC);
+                       usb_wakeup_notification(hcd->self.root_hub,
+                                       faked_port_index + 1);
+                       bogus_port_status = true;
+                       goto cleanup;
+               }
+       }
+
+       /*
+        * Check to see if xhci-hub.c is waiting on RExit to U0 transition (or
+        * RExit to a disconnect state).  If so, let the the driver know it's
+        * out of the RExit state.
+        */
+       if (!DEV_SUPERSPEED(temp) &&
+                       test_and_clear_bit(faked_port_index,
+                               &bus_state->rexit_ports)) {
+               complete(&bus_state->rexit_done[faked_port_index]);
+               bogus_port_status = true;
+               goto cleanup;
+       }
+
+       if (hcd->speed != HCD_USB3)
+               xhci_test_and_clear_bit(xhci, port_array, faked_port_index,
+                                       PORT_PLC);
+
+cleanup:
+       /* Update event ring dequeue pointer before dropping the lock */
+       inc_deq(xhci, xhci->event_ring);
+
+       /* Don't make the USB core poll the roothub if we got a bad port status
+        * change event.  Besides, at that point we can't tell which roothub
+        * (USB 2.0 or USB 3.0) to kick.
+        */
+       if (bogus_port_status)
+               return;
+
+       /*
+        * xHCI port-status-change events occur when the "or" of all the
+        * status-change bits in the portsc register changes from 0 to 1.
+        * New status changes won't cause an event if any other change
+        * bits are still set.  When an event occurs, switch over to
+        * polling to avoid losing status changes.
+        */
+       xhci_dbg(xhci, "%s: starting port polling.\n", __func__);
+       set_bit(HCD_FLAG_POLL_RH, &hcd->flags);
+       spin_unlock(&xhci->lock);
+       /* Pass this up to the core */
+       usb_hcd_poll_rh_status(hcd);
+       spin_lock(&xhci->lock);
+}
+
+/*
+ * This TD is defined by the TRBs starting at start_trb in start_seg and ending
+ * at end_trb, which may be in another segment.  If the suspect DMA address is a
+ * TRB in this TD, this function returns that TRB's segment.  Otherwise it
+ * returns 0.
+ */
+struct xhci_segment *trb_in_td(struct xhci_hcd *xhci,
+               struct xhci_segment *start_seg,
+               union xhci_trb  *start_trb,
+               union xhci_trb  *end_trb,
+               dma_addr_t      suspect_dma,
+               bool            debug)
+{
+       dma_addr_t start_dma;
+       dma_addr_t end_seg_dma;
+       dma_addr_t end_trb_dma;
+       struct xhci_segment *cur_seg;
+
+       start_dma = xhci_trb_virt_to_dma(start_seg, start_trb);
+       cur_seg = start_seg;
+
+       do {
+               if (start_dma == 0)
+                       return NULL;
+               /* We may get an event for a Link TRB in the middle of a TD */
+               end_seg_dma = xhci_trb_virt_to_dma(cur_seg,
+                               &cur_seg->trbs[TRBS_PER_SEGMENT - 1]);
+               /* If the end TRB isn't in this segment, this is set to 0 */
+               end_trb_dma = xhci_trb_virt_to_dma(cur_seg, end_trb);
+
+               if (debug)
+                       xhci_warn(xhci,
+                               "Looking for event-dma %016llx trb-start %016llx trb-end %016llx seg-start %016llx seg-end %016llx\n",
+                               (unsigned long long)suspect_dma,
+                               (unsigned long long)start_dma,
+                               (unsigned long long)end_trb_dma,
+                               (unsigned long long)cur_seg->dma,
+                               (unsigned long long)end_seg_dma);
+
+               if (end_trb_dma > 0) {
+                       /* The end TRB is in this segment, so suspect should be here */
+                       if (start_dma <= end_trb_dma) {
+                               if (suspect_dma >= start_dma && suspect_dma <= end_trb_dma)
+                                       return cur_seg;
+                       } else {
+                               /* Case for one segment with
+                                * a TD wrapped around to the top
+                                */
+                               if ((suspect_dma >= start_dma &&
+                                                       suspect_dma <= end_seg_dma) ||
+                                               (suspect_dma >= cur_seg->dma &&
+                                                suspect_dma <= end_trb_dma))
+                                       return cur_seg;
+                       }
+                       return NULL;
+               } else {
+                       /* Might still be somewhere in this segment */
+                       if (suspect_dma >= start_dma && suspect_dma <= end_seg_dma)
+                               return cur_seg;
+               }
+               cur_seg = cur_seg->next;
+               start_dma = xhci_trb_virt_to_dma(cur_seg, &cur_seg->trbs[0]);
+       } while (cur_seg != start_seg);
+
+       return NULL;
+}
+
+static void xhci_cleanup_halted_endpoint(struct xhci_hcd *xhci,
+               unsigned int slot_id, unsigned int ep_index,
+               unsigned int stream_id,
+               struct xhci_td *td, union xhci_trb *event_trb)
+{
+       struct xhci_virt_ep *ep = &xhci->devs[slot_id]->eps[ep_index];
+       struct xhci_command *command;
+       command = xhci_alloc_command(xhci, false, false, GFP_ATOMIC);
+       if (!command)
+               return;
+
+       ep->ep_state |= EP_HALTED;
+       ep->stopped_stream = stream_id;
+
+       xhci_queue_reset_ep(xhci, command, slot_id, ep_index);
+       xhci_cleanup_stalled_ring(xhci, ep_index, td);
+
+       ep->stopped_stream = 0;
+
+       xhci_ring_cmd_db(xhci);
+}
+
+/* Check if an error has halted the endpoint ring.  The class driver will
+ * cleanup the halt for a non-default control endpoint if we indicate a stall.
+ * However, a babble and other errors also halt the endpoint ring, and the class
+ * driver won't clear the halt in that case, so we need to issue a Set Transfer
+ * Ring Dequeue Pointer command manually.
+ */
+static int xhci_requires_manual_halt_cleanup(struct xhci_hcd *xhci,
+               struct xhci_ep_ctx *ep_ctx,
+               unsigned int trb_comp_code)
+{
+       /* TRB completion codes that may require a manual halt cleanup */
+       if (trb_comp_code == COMP_TX_ERR ||
+                       trb_comp_code == COMP_BABBLE ||
+                       trb_comp_code == COMP_SPLIT_ERR)
+               /* The 0.96 spec says a babbling control endpoint
+                * is not halted. The 0.96 spec says it is.  Some HW
+                * claims to be 0.95 compliant, but it halts the control
+                * endpoint anyway.  Check if a babble halted the
+                * endpoint.
+                */
+               if ((ep_ctx->ep_info & cpu_to_le32(EP_STATE_MASK)) ==
+                   cpu_to_le32(EP_STATE_HALTED))
+                       return 1;
+
+       return 0;
+}
+
+int xhci_is_vendor_info_code(struct xhci_hcd *xhci, unsigned int trb_comp_code)
+{
+       if (trb_comp_code >= 224 && trb_comp_code <= 255) {
+               /* Vendor defined "informational" completion code,
+                * treat as not-an-error.
+                */
+               xhci_dbg(xhci, "Vendor defined info completion code %u\n",
+                               trb_comp_code);
+               xhci_dbg(xhci, "Treating code as success.\n");
+               return 1;
+       }
+       return 0;
+}
+
+/*
+ * Finish the td processing, remove the td from td list;
+ * Return 1 if the urb can be given back.
+ */
+static int finish_td(struct xhci_hcd *xhci, struct xhci_td *td,
+       union xhci_trb *event_trb, struct xhci_transfer_event *event,
+       struct xhci_virt_ep *ep, int *status, bool skip)
+{
+       struct xhci_virt_device *xdev;
+       struct xhci_ring *ep_ring;
+       unsigned int slot_id;
+       int ep_index;
+       struct urb *urb = NULL;
+       struct xhci_ep_ctx *ep_ctx;
+       int ret = 0;
+       struct urb_priv *urb_priv;
+       u32 trb_comp_code;
+
+       slot_id = TRB_TO_SLOT_ID(le32_to_cpu(event->flags));
+       xdev = xhci->devs[slot_id];
+       ep_index = TRB_TO_EP_ID(le32_to_cpu(event->flags)) - 1;
+       ep_ring = xhci_dma_to_transfer_ring(ep, le64_to_cpu(event->buffer));
+       ep_ctx = xhci_get_ep_ctx(xhci, xdev->out_ctx, ep_index);
+       trb_comp_code = GET_COMP_CODE(le32_to_cpu(event->transfer_len));
+
+       if (skip)
+               goto td_cleanup;
+
+       if (trb_comp_code == COMP_STOP_INVAL || trb_comp_code == COMP_STOP) {
+               /* The Endpoint Stop Command completion will take care of any
+                * stopped TDs.  A stopped TD may be restarted, so don't update
+                * the ring dequeue pointer or take this TD off any lists yet.
+                */
+               ep->stopped_td = td;
+               return 0;
+       }
+       if (trb_comp_code == COMP_STALL ||
+               xhci_requires_manual_halt_cleanup(xhci, ep_ctx,
+                                               trb_comp_code)) {
+               /* Issue a reset endpoint command to clear the host side
+                * halt, followed by a set dequeue command to move the
+                * dequeue pointer past the TD.
+                * The class driver clears the device side halt later.
+                */
+               xhci_cleanup_halted_endpoint(xhci, slot_id, ep_index,
+                                       ep_ring->stream_id, td, event_trb);
+       } else {
+               /* Update ring dequeue pointer */
+               while (ep_ring->dequeue != td->last_trb)
+                       inc_deq(xhci, ep_ring);
+               inc_deq(xhci, ep_ring);
+       }
+
+td_cleanup:
+       /* Clean up the endpoint's TD list */
+       urb = td->urb;
+       urb_priv = urb->hcpriv;
+
+       /* Do one last check of the actual transfer length.
+        * If the host controller said we transferred more data than the buffer
+        * length, urb->actual_length will be a very big number (since it's
+        * unsigned).  Play it safe and say we didn't transfer anything.
+        */
+       if (urb->actual_length > urb->transfer_buffer_length) {
+               xhci_warn(xhci, "URB transfer length is wrong, xHC issue? req. len = %u, act. len = %u\n",
+                       urb->transfer_buffer_length,
+                       urb->actual_length);
+               urb->actual_length = 0;
+               if (td->urb->transfer_flags & URB_SHORT_NOT_OK)
+                       *status = -EREMOTEIO;
+               else
+                       *status = 0;
+       }
+       list_del_init(&td->td_list);
+       /* Was this TD slated to be cancelled but completed anyway? */
+       if (!list_empty(&td->cancelled_td_list))
+               list_del_init(&td->cancelled_td_list);
+
+       urb_priv->td_cnt++;
+       /* Giveback the urb when all the tds are completed */
+       if (urb_priv->td_cnt == urb_priv->length) {
+               ret = 1;
+               if (usb_pipetype(urb->pipe) == PIPE_ISOCHRONOUS) {
+                       xhci_to_hcd(xhci)->self.bandwidth_isoc_reqs--;
+                       if (xhci_to_hcd(xhci)->self.bandwidth_isoc_reqs == 0) {
+                               if (xhci->quirks & XHCI_AMD_PLL_FIX)
+                                       usb_amd_quirk_pll_enable();
+                       }
+               }
+       }
+
+       return ret;
+}
+
+/*
+ * Process control tds, update urb status and actual_length.
+ */
+static int process_ctrl_td(struct xhci_hcd *xhci, struct xhci_td *td,
+       union xhci_trb *event_trb, struct xhci_transfer_event *event,
+       struct xhci_virt_ep *ep, int *status)
+{
+       struct xhci_virt_device *xdev;
+       struct xhci_ring *ep_ring;
+       unsigned int slot_id;
+       int ep_index;
+       struct xhci_ep_ctx *ep_ctx;
+       u32 trb_comp_code;
+
+       slot_id = TRB_TO_SLOT_ID(le32_to_cpu(event->flags));
+       xdev = xhci->devs[slot_id];
+       ep_index = TRB_TO_EP_ID(le32_to_cpu(event->flags)) - 1;
+       ep_ring = xhci_dma_to_transfer_ring(ep, le64_to_cpu(event->buffer));
+       ep_ctx = xhci_get_ep_ctx(xhci, xdev->out_ctx, ep_index);
+       trb_comp_code = GET_COMP_CODE(le32_to_cpu(event->transfer_len));
+
+       switch (trb_comp_code) {
+       case COMP_SUCCESS:
+               if (event_trb == ep_ring->dequeue) {
+                       xhci_warn(xhci, "WARN: Success on ctrl setup TRB "
+                                       "without IOC set??\n");
+                       *status = -ESHUTDOWN;
+               } else if (event_trb != td->last_trb) {
+                       xhci_warn(xhci, "WARN: Success on ctrl data TRB "
+                                       "without IOC set??\n");
+                       *status = -ESHUTDOWN;
+               } else {
+                       *status = 0;
+               }
+               break;
+       case COMP_SHORT_TX:
+               if (td->urb->transfer_flags & URB_SHORT_NOT_OK)
+                       *status = -EREMOTEIO;
+               else
+                       *status = 0;
+               break;
+       case COMP_STOP_INVAL:
+       case COMP_STOP:
+               return finish_td(xhci, td, event_trb, event, ep, status, false);
+       default:
+               if (!xhci_requires_manual_halt_cleanup(xhci,
+                                       ep_ctx, trb_comp_code))
+                       break;
+               xhci_dbg(xhci, "TRB error code %u, "
+                               "halted endpoint index = %u\n",
+                               trb_comp_code, ep_index);
+               /* else fall through */
+       case COMP_STALL:
+               /* Did we transfer part of the data (middle) phase? */
+               if (event_trb != ep_ring->dequeue &&
+                               event_trb != td->last_trb)
+                       td->urb->actual_length =
+                               td->urb->transfer_buffer_length -
+                               EVENT_TRB_LEN(le32_to_cpu(event->transfer_len));
+               else
+                       td->urb->actual_length = 0;
+
+               return finish_td(xhci, td, event_trb, event, ep, status, false);
+       }
+       /*
+        * Did we transfer any data, despite the errors that might have
+        * happened?  I.e. did we get past the setup stage?
+        */
+       if (event_trb != ep_ring->dequeue) {
+               /* The event was for the status stage */
+               if (event_trb == td->last_trb) {
+                       if (td->urb_length_set) {
+                               /* Don't overwrite a previously set error code
+                                */
+                               if ((*status == -EINPROGRESS || *status == 0) &&
+                                               (td->urb->transfer_flags
+                                                & URB_SHORT_NOT_OK))
+                                       /* Did we already see a short data
+                                        * stage? */
+                                       *status = -EREMOTEIO;
+                       } else {
+                               td->urb->actual_length =
+                                       td->urb->transfer_buffer_length;
+                       }
+               } else {
+                       /*
+                        * Maybe the event was for the data stage? If so, update
+                        * already the actual_length of the URB and flag it as
+                        * set, so that it is not overwritten in the event for
+                        * the last TRB.
+                        */
+                       td->urb_length_set = true;
+                       td->urb->actual_length =
+                               td->urb->transfer_buffer_length -
+                               EVENT_TRB_LEN(le32_to_cpu(event->transfer_len));
+                       xhci_dbg(xhci, "Waiting for status "
+                                       "stage event\n");
+                       return 0;
+               }
+       }
+
+       return finish_td(xhci, td, event_trb, event, ep, status, false);
+}
+
+/*
+ * Process isochronous tds, update urb packet status and actual_length.
+ */
+static int process_isoc_td(struct xhci_hcd *xhci, struct xhci_td *td,
+       union xhci_trb *event_trb, struct xhci_transfer_event *event,
+       struct xhci_virt_ep *ep, int *status)
+{
+       struct xhci_ring *ep_ring;
+       struct urb_priv *urb_priv;
+       int idx;
+       int len = 0;
+       union xhci_trb *cur_trb;
+       struct xhci_segment *cur_seg;
+       struct usb_iso_packet_descriptor *frame;
+       u32 trb_comp_code;
+       bool skip_td = false;
+
+       ep_ring = xhci_dma_to_transfer_ring(ep, le64_to_cpu(event->buffer));
+       trb_comp_code = GET_COMP_CODE(le32_to_cpu(event->transfer_len));
+       urb_priv = td->urb->hcpriv;
+       idx = urb_priv->td_cnt;
+       frame = &td->urb->iso_frame_desc[idx];
+
+       /* handle completion code */
+       switch (trb_comp_code) {
+       case COMP_SUCCESS:
+               if (EVENT_TRB_LEN(le32_to_cpu(event->transfer_len)) == 0) {
+                       frame->status = 0;
+                       break;
+               }
+               if ((xhci->quirks & XHCI_TRUST_TX_LENGTH))
+                       trb_comp_code = COMP_SHORT_TX;
+       case COMP_SHORT_TX:
+               frame->status = td->urb->transfer_flags & URB_SHORT_NOT_OK ?
+                               -EREMOTEIO : 0;
+               break;
+       case COMP_BW_OVER:
+               frame->status = -ECOMM;
+               skip_td = true;
+               break;
+       case COMP_BUFF_OVER:
+       case COMP_BABBLE:
+               frame->status = -EOVERFLOW;
+               skip_td = true;
+               break;
+       case COMP_DEV_ERR:
+       case COMP_STALL:
+               frame->status = -EPROTO;
+               skip_td = true;
+               break;
+       case COMP_TX_ERR:
+               frame->status = -EPROTO;
+               if (event_trb != td->last_trb)
+                       return 0;
+               skip_td = true;
+               break;
+       case COMP_STOP:
+       case COMP_STOP_INVAL:
+               break;
+       default:
+               frame->status = -1;
+               break;
+       }
+
+       if (trb_comp_code == COMP_SUCCESS || skip_td) {
+               frame->actual_length = frame->length;
+               td->urb->actual_length += frame->length;
+       } else {
+               for (cur_trb = ep_ring->dequeue,
+                    cur_seg = ep_ring->deq_seg; cur_trb != event_trb;
+                    next_trb(xhci, ep_ring, &cur_seg, &cur_trb)) {
+                       if (!TRB_TYPE_NOOP_LE32(cur_trb->generic.field[3]) &&
+                           !TRB_TYPE_LINK_LE32(cur_trb->generic.field[3]))
+                               len += TRB_LEN(le32_to_cpu(cur_trb->generic.field[2]));
+               }
+               len += TRB_LEN(le32_to_cpu(cur_trb->generic.field[2])) -
+                       EVENT_TRB_LEN(le32_to_cpu(event->transfer_len));
+
+               if (trb_comp_code != COMP_STOP_INVAL) {
+                       frame->actual_length = len;
+                       td->urb->actual_length += len;
+               }
+       }
+
+       return finish_td(xhci, td, event_trb, event, ep, status, false);
+}
+
+static int skip_isoc_td(struct xhci_hcd *xhci, struct xhci_td *td,
+                       struct xhci_transfer_event *event,
+                       struct xhci_virt_ep *ep, int *status)
+{
+       struct xhci_ring *ep_ring;
+       struct urb_priv *urb_priv;
+       struct usb_iso_packet_descriptor *frame;
+       int idx;
+
+       ep_ring = xhci_dma_to_transfer_ring(ep, le64_to_cpu(event->buffer));
+       urb_priv = td->urb->hcpriv;
+       idx = urb_priv->td_cnt;
+       frame = &td->urb->iso_frame_desc[idx];
+
+       /* The transfer is partly done. */
+       frame->status = -EXDEV;
+
+       /* calc actual length */
+       frame->actual_length = 0;
+
+       /* Update ring dequeue pointer */
+       while (ep_ring->dequeue != td->last_trb)
+               inc_deq(xhci, ep_ring);
+       inc_deq(xhci, ep_ring);
+
+       return finish_td(xhci, td, NULL, event, ep, status, true);
+}
+
+/*
+ * Process bulk and interrupt tds, update urb status and actual_length.
+ */
+static int process_bulk_intr_td(struct xhci_hcd *xhci, struct xhci_td *td,
+       union xhci_trb *event_trb, struct xhci_transfer_event *event,
+       struct xhci_virt_ep *ep, int *status)
+{
+       struct xhci_ring *ep_ring;
+       union xhci_trb *cur_trb;
+       struct xhci_segment *cur_seg;
+       u32 trb_comp_code;
+
+       ep_ring = xhci_dma_to_transfer_ring(ep, le64_to_cpu(event->buffer));
+       trb_comp_code = GET_COMP_CODE(le32_to_cpu(event->transfer_len));
+
+       switch (trb_comp_code) {
+       case COMP_SUCCESS:
+               /* Double check that the HW transferred everything. */
+               if (event_trb != td->last_trb ||
+                   EVENT_TRB_LEN(le32_to_cpu(event->transfer_len)) != 0) {
+                       xhci_warn(xhci, "WARN Successful completion "
+                                       "on short TX\n");
+                       if (td->urb->transfer_flags & URB_SHORT_NOT_OK)
+                               *status = -EREMOTEIO;
+                       else
+                               *status = 0;
+                       if ((xhci->quirks & XHCI_TRUST_TX_LENGTH))
+                               trb_comp_code = COMP_SHORT_TX;
+               } else {
+                       *status = 0;
+               }
+               break;
+       case COMP_SHORT_TX:
+               if (td->urb->transfer_flags & URB_SHORT_NOT_OK)
+                       *status = -EREMOTEIO;
+               else
+                       *status = 0;
+               break;
+       default:
+               /* Others already handled above */
+               break;
+       }
+       if (trb_comp_code == COMP_SHORT_TX)
+               xhci_dbg(xhci, "ep %#x - asked for %d bytes, "
+                               "%d bytes untransferred\n",
+                               td->urb->ep->desc.bEndpointAddress,
+                               td->urb->transfer_buffer_length,
+                               EVENT_TRB_LEN(le32_to_cpu(event->transfer_len)));
+       /* Fast path - was this the last TRB in the TD for this URB? */
+       if (event_trb == td->last_trb) {
+               if (EVENT_TRB_LEN(le32_to_cpu(event->transfer_len)) != 0) {
+                       td->urb->actual_length =
+                               td->urb->transfer_buffer_length -
+                               EVENT_TRB_LEN(le32_to_cpu(event->transfer_len));
+                       if (td->urb->transfer_buffer_length <
+                                       td->urb->actual_length) {
+                               xhci_warn(xhci, "HC gave bad length "
+                                               "of %d bytes left\n",
+                                         EVENT_TRB_LEN(le32_to_cpu(event->transfer_len)));
+                               td->urb->actual_length = 0;
+                               if (td->urb->transfer_flags & URB_SHORT_NOT_OK)
+                                       *status = -EREMOTEIO;
+                               else
+                                       *status = 0;
+                       }
+                       /* Don't overwrite a previously set error code */
+                       if (*status == -EINPROGRESS) {
+                               if (td->urb->transfer_flags & URB_SHORT_NOT_OK)
+                                       *status = -EREMOTEIO;
+                               else
+                                       *status = 0;
+                       }
+               } else {
+                       td->urb->actual_length =
+                               td->urb->transfer_buffer_length;
+                       /* Ignore a short packet completion if the
+                        * untransferred length was zero.
+                        */
+                       if (*status == -EREMOTEIO)
+                               *status = 0;
+               }
+       } else {
+               /* Slow path - walk the list, starting from the dequeue
+                * pointer, to get the actual length transferred.
+                */
+               td->urb->actual_length = 0;
+               for (cur_trb = ep_ring->dequeue, cur_seg = ep_ring->deq_seg;
+                               cur_trb != event_trb;
+                               next_trb(xhci, ep_ring, &cur_seg, &cur_trb)) {
+                       if (!TRB_TYPE_NOOP_LE32(cur_trb->generic.field[3]) &&
+                           !TRB_TYPE_LINK_LE32(cur_trb->generic.field[3]))
+                               td->urb->actual_length +=
+                                       TRB_LEN(le32_to_cpu(cur_trb->generic.field[2]));
+               }
+               /* If the ring didn't stop on a Link or No-op TRB, add
+                * in the actual bytes transferred from the Normal TRB
+                */
+               if (trb_comp_code != COMP_STOP_INVAL)
+                       td->urb->actual_length +=
+                               TRB_LEN(le32_to_cpu(cur_trb->generic.field[2])) -
+                               EVENT_TRB_LEN(le32_to_cpu(event->transfer_len));
+       }
+
+       return finish_td(xhci, td, event_trb, event, ep, status, false);
+}
+
+/*
+ * If this function returns an error condition, it means it got a Transfer
+ * event with a corrupted Slot ID, Endpoint ID, or TRB DMA address.
+ * At this point, the host controller is probably hosed and should be reset.
+ */
+static int handle_tx_event(struct xhci_hcd *xhci,
+               struct xhci_transfer_event *event)
+       __releases(&xhci->lock)
+       __acquires(&xhci->lock)
+{
+       struct xhci_virt_device *xdev;
+       struct xhci_virt_ep *ep;
+       struct xhci_ring *ep_ring;
+       unsigned int slot_id;
+       int ep_index;
+       struct xhci_td *td = NULL;
+       dma_addr_t event_dma;
+       struct xhci_segment *event_seg;
+       union xhci_trb *event_trb;
+       struct urb *urb = NULL;
+       int status = -EINPROGRESS;
+       struct urb_priv *urb_priv;
+       struct xhci_ep_ctx *ep_ctx;
+       struct list_head *tmp;
+       u32 trb_comp_code;
+       int ret = 0;
+       int td_num = 0;
+
+       slot_id = TRB_TO_SLOT_ID(le32_to_cpu(event->flags));
+       xdev = xhci->devs[slot_id];
+       if (!xdev) {
+               xhci_err(xhci, "ERROR Transfer event pointed to bad slot\n");
+               xhci_err(xhci, "@%016llx %08x %08x %08x %08x\n",
+                        (unsigned long long) xhci_trb_virt_to_dma(
+                                xhci->event_ring->deq_seg,
+                                xhci->event_ring->dequeue),
+                        lower_32_bits(le64_to_cpu(event->buffer)),
+                        upper_32_bits(le64_to_cpu(event->buffer)),
+                        le32_to_cpu(event->transfer_len),
+                        le32_to_cpu(event->flags));
+               xhci_dbg(xhci, "Event ring:\n");
+               xhci_debug_segment(xhci, xhci->event_ring->deq_seg);
+               return -ENODEV;
+       }
+
+       /* Endpoint ID is 1 based, our index is zero based */
+       ep_index = TRB_TO_EP_ID(le32_to_cpu(event->flags)) - 1;
+       ep = &xdev->eps[ep_index];
+       ep_ring = xhci_dma_to_transfer_ring(ep, le64_to_cpu(event->buffer));
+       ep_ctx = xhci_get_ep_ctx(xhci, xdev->out_ctx, ep_index);
+       if (!ep_ring ||
+           (le32_to_cpu(ep_ctx->ep_info) & EP_STATE_MASK) ==
+           EP_STATE_DISABLED) {
+               xhci_err(xhci, "ERROR Transfer event for disabled endpoint "
+                               "or incorrect stream ring\n");
+               xhci_err(xhci, "@%016llx %08x %08x %08x %08x\n",
+                        (unsigned long long) xhci_trb_virt_to_dma(
+                                xhci->event_ring->deq_seg,
+                                xhci->event_ring->dequeue),
+                        lower_32_bits(le64_to_cpu(event->buffer)),
+                        upper_32_bits(le64_to_cpu(event->buffer)),
+                        le32_to_cpu(event->transfer_len),
+                        le32_to_cpu(event->flags));
+               xhci_dbg(xhci, "Event ring:\n");
+               xhci_debug_segment(xhci, xhci->event_ring->deq_seg);
+               return -ENODEV;
+       }
+
+       /* Count current td numbers if ep->skip is set */
+       if (ep->skip) {
+               list_for_each(tmp, &ep_ring->td_list)
+                       td_num++;
+       }
+
+       event_dma = le64_to_cpu(event->buffer);
+       trb_comp_code = GET_COMP_CODE(le32_to_cpu(event->transfer_len));
+       /* Look for common error cases */
+       switch (trb_comp_code) {
+       /* Skip codes that require special handling depending on
+        * transfer type
+        */
+       case COMP_SUCCESS:
+               if (EVENT_TRB_LEN(le32_to_cpu(event->transfer_len)) == 0)
+                       break;
+               if (xhci->quirks & XHCI_TRUST_TX_LENGTH)
+                       trb_comp_code = COMP_SHORT_TX;
+               else
+                       xhci_warn_ratelimited(xhci,
+                                       "WARN Successful completion on short TX: needs XHCI_TRUST_TX_LENGTH quirk?\n");
+       case COMP_SHORT_TX:
+               break;
+       case COMP_STOP:
+               xhci_dbg(xhci, "Stopped on Transfer TRB\n");
+               break;
+       case COMP_STOP_INVAL:
+               xhci_dbg(xhci, "Stopped on No-op or Link TRB\n");
+               break;
+       case COMP_STALL:
+               xhci_dbg(xhci, "Stalled endpoint\n");
+               ep->ep_state |= EP_HALTED;
+               status = -EPIPE;
+               break;
+       case COMP_TRB_ERR:
+               xhci_warn(xhci, "WARN: TRB error on endpoint\n");
+               status = -EILSEQ;
+               break;
+       case COMP_SPLIT_ERR:
+       case COMP_TX_ERR:
+               xhci_dbg(xhci, "Transfer error on endpoint\n");
+               status = -EPROTO;
+               break;
+       case COMP_BABBLE:
+               xhci_dbg(xhci, "Babble error on endpoint\n");
+               status = -EOVERFLOW;
+               break;
+       case COMP_DB_ERR:
+               xhci_warn(xhci, "WARN: HC couldn't access mem fast enough\n");
+               status = -ENOSR;
+               break;
+       case COMP_BW_OVER:
+               xhci_warn(xhci, "WARN: bandwidth overrun event on endpoint\n");
+               break;
+       case COMP_BUFF_OVER:
+               xhci_warn(xhci, "WARN: buffer overrun event on endpoint\n");
+               break;
+       case COMP_UNDERRUN:
+               /*
+                * When the Isoch ring is empty, the xHC will generate
+                * a Ring Overrun Event for IN Isoch endpoint or Ring
+                * Underrun Event for OUT Isoch endpoint.
+                */
+               xhci_dbg(xhci, "underrun event on endpoint\n");
+               if (!list_empty(&ep_ring->td_list))
+                       xhci_dbg(xhci, "Underrun Event for slot %d ep %d "
+                                       "still with TDs queued?\n",
+                                TRB_TO_SLOT_ID(le32_to_cpu(event->flags)),
+                                ep_index);
+               goto cleanup;
+       case COMP_OVERRUN:
+               xhci_dbg(xhci, "overrun event on endpoint\n");
+               if (!list_empty(&ep_ring->td_list))
+                       xhci_dbg(xhci, "Overrun Event for slot %d ep %d "
+                                       "still with TDs queued?\n",
+                                TRB_TO_SLOT_ID(le32_to_cpu(event->flags)),
+                                ep_index);
+               goto cleanup;
+       case COMP_DEV_ERR:
+               xhci_warn(xhci, "WARN: detect an incompatible device");
+               status = -EPROTO;
+               break;
+       case COMP_MISSED_INT:
+               /*
+                * When encounter missed service error, one or more isoc tds
+                * may be missed by xHC.
+                * Set skip flag of the ep_ring; Complete the missed tds as
+                * short transfer when process the ep_ring next time.
+                */
+               ep->skip = true;
+               xhci_dbg(xhci, "Miss service interval error, set skip flag\n");
+               goto cleanup;
+       default:
+               if (xhci_is_vendor_info_code(xhci, trb_comp_code)) {
+                       status = 0;
+                       break;
+               }
+               xhci_warn(xhci, "ERROR Unknown event condition %u, HC probably busted\n",
+                         trb_comp_code);
+               goto cleanup;
+       }
+
+       do {
+               /* This TRB should be in the TD at the head of this ring's
+                * TD list.
+                */
+               if (list_empty(&ep_ring->td_list)) {
+                       /*
+                        * A stopped endpoint may generate an extra completion
+                        * event if the device was suspended.  Don't print
+                        * warnings.
+                        */
+                       if (!(trb_comp_code == COMP_STOP ||
+                                               trb_comp_code == COMP_STOP_INVAL)) {
+                               xhci_warn(xhci, "WARN Event TRB for slot %d ep %d with no TDs queued?\n",
+                                               TRB_TO_SLOT_ID(le32_to_cpu(event->flags)),
+                                               ep_index);
+                               xhci_dbg(xhci, "Event TRB with TRB type ID %u\n",
+                                               (le32_to_cpu(event->flags) &
+                                                TRB_TYPE_BITMASK)>>10);
+                               xhci_print_trb_offsets(xhci, (union xhci_trb *) event);
+                       }
+                       if (ep->skip) {
+                               ep->skip = false;
+                               xhci_dbg(xhci, "td_list is empty while skip "
+                                               "flag set. Clear skip flag.\n");
+                       }
+                       ret = 0;
+                       goto cleanup;
+               }
+
+               /* We've skipped all the TDs on the ep ring when ep->skip set */
+               if (ep->skip && td_num == 0) {
+                       ep->skip = false;
+                       xhci_dbg(xhci, "All tds on the ep_ring skipped. "
+                                               "Clear skip flag.\n");
+                       ret = 0;
+                       goto cleanup;
+               }
+
+               td = list_entry(ep_ring->td_list.next, struct xhci_td, td_list);
+               if (ep->skip)
+                       td_num--;
+
+               /* Is this a TRB in the currently executing TD? */
+               event_seg = trb_in_td(xhci, ep_ring->deq_seg, ep_ring->dequeue,
+                               td->last_trb, event_dma, false);
+
+               /*
+                * Skip the Force Stopped Event. The event_trb(event_dma) of FSE
+                * is not in the current TD pointed by ep_ring->dequeue because
+                * that the hardware dequeue pointer still at the previous TRB
+                * of the current TD. The previous TRB maybe a Link TD or the
+                * last TRB of the previous TD. The command completion handle
+                * will take care the rest.
+                */
+               if (!event_seg && (trb_comp_code == COMP_STOP ||
+                                  trb_comp_code == COMP_STOP_INVAL)) {
+                       ret = 0;
+                       goto cleanup;
+               }
+
+               if (!event_seg) {
+                       if (!ep->skip ||
+                           !usb_endpoint_xfer_isoc(&td->urb->ep->desc)) {
+                               /* Some host controllers give a spurious
+                                * successful event after a short transfer.
+                                * Ignore it.
+                                */
+                               if ((xhci->quirks & XHCI_SPURIOUS_SUCCESS) &&
+                                               ep_ring->last_td_was_short) {
+                                       ep_ring->last_td_was_short = false;
+                                       ret = 0;
+                                       goto cleanup;
+                               }
+                               /* HC is busted, give up! */
+                               xhci_err(xhci,
+                                       "ERROR Transfer event TRB DMA ptr not "
+                                       "part of current TD ep_index %d "
+                                       "comp_code %u\n", ep_index,
+                                       trb_comp_code);
+                               trb_in_td(xhci, ep_ring->deq_seg,
+                                         ep_ring->dequeue, td->last_trb,
+                                         event_dma, true);
+                               return -ESHUTDOWN;
+                       }
+
+                       ret = skip_isoc_td(xhci, td, event, ep, &status);
+                       goto cleanup;
+               }
+               if (trb_comp_code == COMP_SHORT_TX)
+                       ep_ring->last_td_was_short = true;
+               else
+                       ep_ring->last_td_was_short = false;
+
+               if (ep->skip) {
+                       xhci_dbg(xhci, "Found td. Clear skip flag.\n");
+                       ep->skip = false;
+               }
+
+               event_trb = &event_seg->trbs[(event_dma - event_seg->dma) /
+                                               sizeof(*event_trb)];
+               /*
+                * No-op TRB should not trigger interrupts.
+                * If event_trb is a no-op TRB, it means the
+                * corresponding TD has been cancelled. Just ignore
+                * the TD.
+                */
+               if (TRB_TYPE_NOOP_LE32(event_trb->generic.field[3])) {
+                       xhci_dbg(xhci,
+                                "event_trb is a no-op TRB. Skip it\n");
+                       goto cleanup;
+               }
+
+               /* Now update the urb's actual_length and give back to
+                * the core
+                */
+               if (usb_endpoint_xfer_control(&td->urb->ep->desc))
+                       ret = process_ctrl_td(xhci, td, event_trb, event, ep,
+                                                &status);
+               else if (usb_endpoint_xfer_isoc(&td->urb->ep->desc))
+                       ret = process_isoc_td(xhci, td, event_trb, event, ep,
+                                                &status);
+               else
+                       ret = process_bulk_intr_td(xhci, td, event_trb, event,
+                                                ep, &status);
+
+cleanup:
+               /*
+                * Do not update event ring dequeue pointer if ep->skip is set.
+                * Will roll back to continue process missed tds.
+                */
+               if (trb_comp_code == COMP_MISSED_INT || !ep->skip) {
+                       inc_deq(xhci, xhci->event_ring);
+               }
+
+               if (ret) {
+                       urb = td->urb;
+                       urb_priv = urb->hcpriv;
+
+                       xhci_urb_free_priv(urb_priv);
+
+                       usb_hcd_unlink_urb_from_ep(bus_to_hcd(urb->dev->bus), urb);
+                       if ((urb->actual_length != urb->transfer_buffer_length &&
+                                               (urb->transfer_flags &
+                                                URB_SHORT_NOT_OK)) ||
+                                       (status != 0 &&
+                                        !usb_endpoint_xfer_isoc(&urb->ep->desc)))
+                               xhci_dbg(xhci, "Giveback URB %p, len = %d, "
+                                               "expected = %d, status = %d\n",
+                                               urb, urb->actual_length,
+                                               urb->transfer_buffer_length,
+                                               status);
+                       spin_unlock(&xhci->lock);
+                       /* EHCI, UHCI, and OHCI always unconditionally set the
+                        * urb->status of an isochronous endpoint to 0.
+                        */
+                       if (usb_pipetype(urb->pipe) == PIPE_ISOCHRONOUS)
+                               status = 0;
+                       usb_hcd_giveback_urb(bus_to_hcd(urb->dev->bus), urb, status);
+                       spin_lock(&xhci->lock);
+               }
+
+       /*
+        * If ep->skip is set, it means there are missed tds on the
+        * endpoint ring need to take care of.
+        * Process them as short transfer until reach the td pointed by
+        * the event.
+        */
+       } while (ep->skip && trb_comp_code != COMP_MISSED_INT);
+
+       return 0;
+}
+
+/*
+ * This function handles all OS-owned events on the event ring.  It may drop
+ * xhci->lock between event processing (e.g. to pass up port status changes).
+ * Returns >0 for "possibly more events to process" (caller should call again),
+ * otherwise 0 if done.  In future, <0 returns should indicate error code.
+ */
+static int xhci_handle_event(struct xhci_hcd *xhci)
+{
+       union xhci_trb *event;
+       int update_ptrs = 1;
+       int ret;
+
+       if (!xhci->event_ring || !xhci->event_ring->dequeue) {
+               xhci->error_bitmask |= 1 << 1;
+               return 0;
+       }
+
+       event = xhci->event_ring->dequeue;
+       /* Does the HC or OS own the TRB? */
+       if ((le32_to_cpu(event->event_cmd.flags) & TRB_CYCLE) !=
+           xhci->event_ring->cycle_state) {
+               xhci->error_bitmask |= 1 << 2;
+               return 0;
+       }
+
+       /*
+        * Barrier between reading the TRB_CYCLE (valid) flag above and any
+        * speculative reads of the event's flags/data below.
+        */
+       rmb();
+       /* FIXME: Handle more event types. */
+       switch ((le32_to_cpu(event->event_cmd.flags) & TRB_TYPE_BITMASK)) {
+       case TRB_TYPE(TRB_COMPLETION):
+               handle_cmd_completion(xhci, &event->event_cmd);
+               break;
+       case TRB_TYPE(TRB_PORT_STATUS):
+               handle_port_status(xhci, event);
+               update_ptrs = 0;
+               break;
+       case TRB_TYPE(TRB_TRANSFER):
+               ret = handle_tx_event(xhci, &event->trans_event);
+               if (ret < 0)
+                       xhci->error_bitmask |= 1 << 9;
+               else
+                       update_ptrs = 0;
+               break;
+       case TRB_TYPE(TRB_DEV_NOTE):
+               handle_device_notification(xhci, event);
+               break;
+       default:
+               if ((le32_to_cpu(event->event_cmd.flags) & TRB_TYPE_BITMASK) >=
+                   TRB_TYPE(48))
+                       handle_vendor_event(xhci, event);
+               else
+                       xhci->error_bitmask |= 1 << 3;
+       }
+       /* Any of the above functions may drop and re-acquire the lock, so check
+        * to make sure a watchdog timer didn't mark the host as non-responsive.
+        */
+       if (xhci->xhc_state & XHCI_STATE_DYING) {
+               xhci_dbg(xhci, "xHCI host dying, returning from "
+                               "event handler.\n");
+               return 0;
+       }
+
+       if (update_ptrs)
+               /* Update SW event ring dequeue pointer */
+               inc_deq(xhci, xhci->event_ring);
+
+       /* Are there more items on the event ring?  Caller will call us again to
+        * check.
+        */
+       return 1;
+}
+
+/*
+ * xHCI spec says we can get an interrupt, and if the HC has an error condition,
+ * we might get bad data out of the event ring.  Section 4.10.2.7 has a list of
+ * indicators of an event TRB error, but we check the status *first* to be safe.
+ */
+irqreturn_t xhci_irq(struct usb_hcd *hcd)
+{
+       struct xhci_hcd *xhci = hcd_to_xhci(hcd);
+       u32 status;
+       u64 temp_64;
+       union xhci_trb *event_ring_deq;
+       dma_addr_t deq;
+
+       spin_lock(&xhci->lock);
+       /* Check if the xHC generated the interrupt, or the irq is shared */
+       status = readl(&xhci->op_regs->status);
+       if (status == 0xffffffff)
+               goto hw_died;
+
+       if (!(status & STS_EINT)) {
+               spin_unlock(&xhci->lock);
+               return IRQ_NONE;
+       }
+       if (status & STS_FATAL) {
+               xhci_warn(xhci, "WARNING: Host System Error\n");
+               xhci_halt(xhci);
+hw_died:
+               spin_unlock(&xhci->lock);
+               return IRQ_HANDLED;
+       }
+
+       /*
+        * Clear the op reg interrupt status first,
+        * so we can receive interrupts from other MSI-X interrupters.
+        * Write 1 to clear the interrupt status.
+        */
+       status |= STS_EINT;
+       writel(status, &xhci->op_regs->status);
+       /* FIXME when MSI-X is supported and there are multiple vectors */
+       /* Clear the MSI-X event interrupt status */
+
+       if (hcd->irq) {
+               u32 irq_pending;
+               /* Acknowledge the PCI interrupt */
+               irq_pending = readl(&xhci->ir_set->irq_pending);
+               irq_pending |= IMAN_IP;
+               writel(irq_pending, &xhci->ir_set->irq_pending);
+       }
+
+       if (xhci->xhc_state & XHCI_STATE_DYING) {
+               xhci_dbg(xhci, "xHCI dying, ignoring interrupt. "
+                               "Shouldn't IRQs be disabled?\n");
+               /* Clear the event handler busy flag (RW1C);
+                * the event ring should be empty.
+                */
+               temp_64 = xhci_read_64(xhci, &xhci->ir_set->erst_dequeue);
+               xhci_write_64(xhci, temp_64 | ERST_EHB,
+                               &xhci->ir_set->erst_dequeue);
+               spin_unlock(&xhci->lock);
+
+               return IRQ_HANDLED;
+       }
+
+       event_ring_deq = xhci->event_ring->dequeue;
+       /* FIXME this should be a delayed service routine
+        * that clears the EHB.
+        */
+       while (xhci_handle_event(xhci) > 0) {}
+
+       temp_64 = xhci_read_64(xhci, &xhci->ir_set->erst_dequeue);
+       /* If necessary, update the HW's version of the event ring deq ptr. */
+       if (event_ring_deq != xhci->event_ring->dequeue) {
+               deq = xhci_trb_virt_to_dma(xhci->event_ring->deq_seg,
+                               xhci->event_ring->dequeue);
+               if (deq == 0)
+                       xhci_warn(xhci, "WARN something wrong with SW event "
+                                       "ring dequeue ptr.\n");
+               /* Update HC event ring dequeue pointer */
+               temp_64 &= ERST_PTR_MASK;
+               temp_64 |= ((u64) deq & (u64) ~ERST_PTR_MASK);
+       }
+
+       /* Clear the event handler busy flag (RW1C); event ring is empty. */
+       temp_64 |= ERST_EHB;
+       xhci_write_64(xhci, temp_64, &xhci->ir_set->erst_dequeue);
+
+       spin_unlock(&xhci->lock);
+
+       return IRQ_HANDLED;
+}
+
+irqreturn_t xhci_msi_irq(int irq, void *hcd)
+{
+       return xhci_irq(hcd);
+}
+
+/****          Endpoint Ring Operations        ****/
+
+/*
+ * Generic function for queueing a TRB on a ring.
+ * The caller must have checked to make sure there's room on the ring.
+ *
+ * @more_trbs_coming:  Will you enqueue more TRBs before calling
+ *                     prepare_transfer()?
+ */
+static void queue_trb(struct xhci_hcd *xhci, struct xhci_ring *ring,
+               bool more_trbs_coming,
+               u32 field1, u32 field2, u32 field3, u32 field4)
+{
+       struct xhci_generic_trb *trb;
+
+       trb = &ring->enqueue->generic;
+       trb->field[0] = cpu_to_le32(field1);
+       trb->field[1] = cpu_to_le32(field2);
+       trb->field[2] = cpu_to_le32(field3);
+       trb->field[3] = cpu_to_le32(field4);
+       inc_enq(xhci, ring, more_trbs_coming);
+}
+
+/*
+ * Does various checks on the endpoint ring, and makes it ready to queue num_trbs.
+ * FIXME allocate segments if the ring is full.
+ */
+static int prepare_ring(struct xhci_hcd *xhci, struct xhci_ring *ep_ring,
+               u32 ep_state, unsigned int num_trbs, gfp_t mem_flags)
+{
+       unsigned int num_trbs_needed;
+
+       /* Make sure the endpoint has been added to xHC schedule */
+       switch (ep_state) {
+       case EP_STATE_DISABLED:
+               /*
+                * USB core changed config/interfaces without notifying us,
+                * or hardware is reporting the wrong state.
+                */
+               xhci_warn(xhci, "WARN urb submitted to disabled ep\n");
+               return -ENOENT;
+       case EP_STATE_ERROR:
+               xhci_warn(xhci, "WARN waiting for error on ep to be cleared\n");
+               /* FIXME event handling code for error needs to clear it */
+               /* XXX not sure if this should be -ENOENT or not */
+               return -EINVAL;
+       case EP_STATE_HALTED:
+               xhci_dbg(xhci, "WARN halted endpoint, queueing URB anyway.\n");
+       case EP_STATE_STOPPED:
+       case EP_STATE_RUNNING:
+               break;
+       default:
+               xhci_err(xhci, "ERROR unknown endpoint state for ep\n");
+               /*
+                * FIXME issue Configure Endpoint command to try to get the HC
+                * back into a known state.
+                */
+               return -EINVAL;
+       }
+
+       while (1) {
+               if (room_on_ring(xhci, ep_ring, num_trbs))
+                       break;
+
+               if (ep_ring == xhci->cmd_ring) {
+                       xhci_err(xhci, "Do not support expand command ring\n");
+                       return -ENOMEM;
+               }
+
+               xhci_dbg_trace(xhci, trace_xhci_dbg_ring_expansion,
+                               "ERROR no room on ep ring, try ring expansion");
+               num_trbs_needed = num_trbs - ep_ring->num_trbs_free;
+               if (xhci_ring_expansion(xhci, ep_ring, num_trbs_needed,
+                                       mem_flags)) {
+                       xhci_err(xhci, "Ring expansion failed\n");
+                       return -ENOMEM;
+               }
+       }
+
+       if (enqueue_is_link_trb(ep_ring)) {
+               struct xhci_ring *ring = ep_ring;
+               union xhci_trb *next;
+
+               next = ring->enqueue;
+
+               while (last_trb(xhci, ring, ring->enq_seg, next)) {
+                       /* If we're not dealing with 0.95 hardware or isoc rings
+                        * on AMD 0.96 host, clear the chain bit.
+                        */
+                       if (!xhci_link_trb_quirk(xhci) &&
+                                       !(ring->type == TYPE_ISOC &&
+                                        (xhci->quirks & XHCI_AMD_0x96_HOST)))
+                               next->link.control &= cpu_to_le32(~TRB_CHAIN);
+                       else
+                               next->link.control |= cpu_to_le32(TRB_CHAIN);
+
+                       wmb();
+                       next->link.control ^= cpu_to_le32(TRB_CYCLE);
+
+                       /* Toggle the cycle bit after the last ring segment. */
+                       if (last_trb_on_last_seg(xhci, ring, ring->enq_seg, next)) {
+                               ring->cycle_state ^= 1;
+                       }
+                       ring->enq_seg = ring->enq_seg->next;
+                       ring->enqueue = ring->enq_seg->trbs;
+                       next = ring->enqueue;
+               }
+       }
+
+       return 0;
+}
+
+static int prepare_transfer(struct xhci_hcd *xhci,
+               struct xhci_virt_device *xdev,
+               unsigned int ep_index,
+               unsigned int stream_id,
+               unsigned int num_trbs,
+               struct urb *urb,
+               unsigned int td_index,
+               gfp_t mem_flags)
+{
+       int ret;
+       struct urb_priv *urb_priv;
+       struct xhci_td  *td;
+       struct xhci_ring *ep_ring;
+       struct xhci_ep_ctx *ep_ctx = xhci_get_ep_ctx(xhci, xdev->out_ctx, ep_index);
+
+       ep_ring = xhci_stream_id_to_ring(xdev, ep_index, stream_id);
+       if (!ep_ring) {
+               xhci_dbg(xhci, "Can't prepare ring for bad stream ID %u\n",
+                               stream_id);
+               return -EINVAL;
+       }
+
+       ret = prepare_ring(xhci, ep_ring,
+                          le32_to_cpu(ep_ctx->ep_info) & EP_STATE_MASK,
+                          num_trbs, mem_flags);
+       if (ret)
+               return ret;
+
+       urb_priv = urb->hcpriv;
+       td = urb_priv->td[td_index];
+
+       INIT_LIST_HEAD(&td->td_list);
+       INIT_LIST_HEAD(&td->cancelled_td_list);
+
+       if (td_index == 0) {
+               ret = usb_hcd_link_urb_to_ep(bus_to_hcd(urb->dev->bus), urb);
+               if (unlikely(ret))
+                       return ret;
+       }
+
+       td->urb = urb;
+       /* Add this TD to the tail of the endpoint ring's TD list */
+       list_add_tail(&td->td_list, &ep_ring->td_list);
+       td->start_seg = ep_ring->enq_seg;
+       td->first_trb = ep_ring->enqueue;
+
+       urb_priv->td[td_index] = td;
+
+       return 0;
+}
+
+static unsigned int count_sg_trbs_needed(struct xhci_hcd *xhci, struct urb *urb)
+{
+       int num_sgs, num_trbs, running_total, temp, i;
+       struct scatterlist *sg;
+
+       sg = NULL;
+       num_sgs = urb->num_mapped_sgs;
+       temp = urb->transfer_buffer_length;
+
+       num_trbs = 0;
+       for_each_sg(urb->sg, sg, num_sgs, i) {
+               unsigned int len = sg_dma_len(sg);
+
+               /* Scatter gather list entries may cross 64KB boundaries */
+               running_total = TRB_MAX_BUFF_SIZE -
+                       (sg_dma_address(sg) & (TRB_MAX_BUFF_SIZE - 1));
+               running_total &= TRB_MAX_BUFF_SIZE - 1;
+               if (running_total != 0)
+                       num_trbs++;
+
+               /* How many more 64KB chunks to transfer, how many more TRBs? */
+               while (running_total < sg_dma_len(sg) && running_total < temp) {
+                       num_trbs++;
+                       running_total += TRB_MAX_BUFF_SIZE;
+               }
+               len = min_t(int, len, temp);
+               temp -= len;
+               if (temp == 0)
+                       break;
+       }
+       return num_trbs;
+}
+
+static void check_trb_math(struct urb *urb, int num_trbs, int running_total)
+{
+       if (num_trbs != 0)
+               dev_err(&urb->dev->dev, "%s - ep %#x - Miscalculated number of "
+                               "TRBs, %d left\n", __func__,
+                               urb->ep->desc.bEndpointAddress, num_trbs);
+       if (running_total != urb->transfer_buffer_length)
+               dev_err(&urb->dev->dev, "%s - ep %#x - Miscalculated tx length, "
+                               "queued %#x (%d), asked for %#x (%d)\n",
+                               __func__,
+                               urb->ep->desc.bEndpointAddress,
+                               running_total, running_total,
+                               urb->transfer_buffer_length,
+                               urb->transfer_buffer_length);
+}
+
+static void giveback_first_trb(struct xhci_hcd *xhci, int slot_id,
+               unsigned int ep_index, unsigned int stream_id, int start_cycle,
+               struct xhci_generic_trb *start_trb)
+{
+       /*
+        * Pass all the TRBs to the hardware at once and make sure this write
+        * isn't reordered.
+        */
+       wmb();
+       if (start_cycle)
+               start_trb->field[3] |= cpu_to_le32(start_cycle);
+       else
+               start_trb->field[3] &= cpu_to_le32(~TRB_CYCLE);
+       xhci_ring_ep_doorbell(xhci, slot_id, ep_index, stream_id);
+}
+
+/*
+ * xHCI uses normal TRBs for both bulk and interrupt.  When the interrupt
+ * endpoint is to be serviced, the xHC will consume (at most) one TD.  A TD
+ * (comprised of sg list entries) can take several service intervals to
+ * transmit.
+ */
+int xhci_queue_intr_tx(struct xhci_hcd *xhci, gfp_t mem_flags,
+               struct urb *urb, int slot_id, unsigned int ep_index)
+{
+       struct xhci_ep_ctx *ep_ctx = xhci_get_ep_ctx(xhci,
+                       xhci->devs[slot_id]->out_ctx, ep_index);
+       int xhci_interval;
+       int ep_interval;
+
+       xhci_interval = EP_INTERVAL_TO_UFRAMES(le32_to_cpu(ep_ctx->ep_info));
+       ep_interval = urb->interval;
+       /* Convert to microframes */
+       if (urb->dev->speed == USB_SPEED_LOW ||
+                       urb->dev->speed == USB_SPEED_FULL)
+               ep_interval *= 8;
+       /* FIXME change this to a warning and a suggestion to use the new API
+        * to set the polling interval (once the API is added).
+        */
+       if (xhci_interval != ep_interval) {
+               dev_dbg_ratelimited(&urb->dev->dev,
+                               "Driver uses different interval (%d microframe%s) than xHCI (%d microframe%s)\n",
+                               ep_interval, ep_interval == 1 ? "" : "s",
+                               xhci_interval, xhci_interval == 1 ? "" : "s");
+               urb->interval = xhci_interval;
+               /* Convert back to frames for LS/FS devices */
+               if (urb->dev->speed == USB_SPEED_LOW ||
+                               urb->dev->speed == USB_SPEED_FULL)
+                       urb->interval /= 8;
+       }
+       return xhci_queue_bulk_tx(xhci, mem_flags, urb, slot_id, ep_index);
+}
+
+/*
+ * The TD size is the number of bytes remaining in the TD (including this TRB),
+ * right shifted by 10.
+ * It must fit in bits 21:17, so it can't be bigger than 31.
+ */
+static u32 xhci_td_remainder(unsigned int remainder)
+{
+       u32 max = (1 << (21 - 17 + 1)) - 1;
+
+       if ((remainder >> 10) >= max)
+               return max << 17;
+       else
+               return (remainder >> 10) << 17;
+}
+
+/*
+ * For xHCI 1.0 host controllers, TD size is the number of max packet sized
+ * packets remaining in the TD (*not* including this TRB).
+ *
+ * Total TD packet count = total_packet_count =
+ *     DIV_ROUND_UP(TD size in bytes / wMaxPacketSize)
+ *
+ * Packets transferred up to and including this TRB = packets_transferred =
+ *     rounddown(total bytes transferred including this TRB / wMaxPacketSize)
+ *
+ * TD size = total_packet_count - packets_transferred
+ *
+ * It must fit in bits 21:17, so it can't be bigger than 31.
+ * The last TRB in a TD must have the TD size set to zero.
+ */
+static u32 xhci_v1_0_td_remainder(int running_total, int trb_buff_len,
+               unsigned int total_packet_count, struct urb *urb,
+               unsigned int num_trbs_left)
+{
+       int packets_transferred;
+
+       /* One TRB with a zero-length data packet. */
+       if (num_trbs_left == 0 || (running_total == 0 && trb_buff_len == 0))
+               return 0;
+
+       /* All the TRB queueing functions don't count the current TRB in
+        * running_total.
+        */
+       packets_transferred = (running_total + trb_buff_len) /
+               GET_MAX_PACKET(usb_endpoint_maxp(&urb->ep->desc));
+
+       if ((total_packet_count - packets_transferred) > 31)
+               return 31 << 17;
+       return (total_packet_count - packets_transferred) << 17;
+}
+
+static int queue_bulk_sg_tx(struct xhci_hcd *xhci, gfp_t mem_flags,
+               struct urb *urb, int slot_id, unsigned int ep_index)
+{
+       struct xhci_ring *ep_ring;
+       unsigned int num_trbs;
+       struct urb_priv *urb_priv;
+       struct xhci_td *td;
+       struct scatterlist *sg;
+       int num_sgs;
+       int trb_buff_len, this_sg_len, running_total;
+       unsigned int total_packet_count;
+       bool first_trb;
+       u64 addr;
+       bool more_trbs_coming;
+
+       struct xhci_generic_trb *start_trb;
+       int start_cycle;
+
+       ep_ring = xhci_urb_to_transfer_ring(xhci, urb);
+       if (!ep_ring)
+               return -EINVAL;
+
+       num_trbs = count_sg_trbs_needed(xhci, urb);
+       num_sgs = urb->num_mapped_sgs;
+       total_packet_count = DIV_ROUND_UP(urb->transfer_buffer_length,
+                       usb_endpoint_maxp(&urb->ep->desc));
+
+       trb_buff_len = prepare_transfer(xhci, xhci->devs[slot_id],
+                       ep_index, urb->stream_id,
+                       num_trbs, urb, 0, mem_flags);
+       if (trb_buff_len < 0)
+               return trb_buff_len;
+
+       urb_priv = urb->hcpriv;
+       td = urb_priv->td[0];
+
+       /*
+        * 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?
+        *
+        * There are three forces at work for TRB buffer pointers and lengths:
+        * 1. We don't want to walk off the end of this sg-list entry buffer.
+        * 2. The transfer length that the driver requested may be smaller than
+        *    the amount of memory allocated for this scatter-gather list.
+        * 3. TRBs buffers can't cross 64KB boundaries.
+        */
+       sg = urb->sg;
+       addr = (u64) sg_dma_address(sg);
+       this_sg_len = sg_dma_len(sg);
+       trb_buff_len = TRB_MAX_BUFF_SIZE - (addr & (TRB_MAX_BUFF_SIZE - 1));
+       trb_buff_len = min_t(int, trb_buff_len, this_sg_len);
+       if (trb_buff_len > urb->transfer_buffer_length)
+               trb_buff_len = urb->transfer_buffer_length;
+
+       first_trb = true;
+       /* Queue the first TRB, even if it's zero-length */
+       do {
+               u32 field = 0;
+               u32 length_field = 0;
+               u32 remainder = 0;
+
+               /* Don't change the cycle bit of the first TRB until later */
+               if (first_trb) {
+                       first_trb = false;
+                       if (start_cycle == 0)
+                               field |= 0x1;
+               } 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;
+               }
+
+               /* Only set interrupt on short packet for IN endpoints */
+               if (usb_urb_dir_in(urb))
+                       field |= TRB_ISP;
+
+               if (TRB_MAX_BUFF_SIZE -
+                               (addr & (TRB_MAX_BUFF_SIZE - 1)) < trb_buff_len) {
+                       xhci_warn(xhci, "WARN: sg dma xfer crosses 64KB boundaries!\n");
+                       xhci_dbg(xhci, "Next boundary at %#x, end dma = %#x\n",
+                                       (unsigned int) (addr + TRB_MAX_BUFF_SIZE) & ~(TRB_MAX_BUFF_SIZE - 1),
+                                       (unsigned int) addr + trb_buff_len);
+               }
+
+               /* Set the TRB length, TD size, and interrupter fields. */
+               if (xhci->hci_version < 0x100) {
+                       remainder = xhci_td_remainder(
+                                       urb->transfer_buffer_length -
+                                       running_total);
+               } else {
+                       remainder = xhci_v1_0_td_remainder(running_total,
+                                       trb_buff_len, total_packet_count, urb,
+                                       num_trbs - 1);
+               }
+               length_field = TRB_LEN(trb_buff_len) |
+                       remainder |
+                       TRB_INTR_TARGET(0);
+
+               if (num_trbs > 1)
+                       more_trbs_coming = true;
+               else
+                       more_trbs_coming = false;
+               queue_trb(xhci, ep_ring, more_trbs_coming,
+                               lower_32_bits(addr),
+                               upper_32_bits(addr),
+                               length_field,
+                               field | TRB_TYPE(TRB_NORMAL));
+               --num_trbs;
+               running_total += trb_buff_len;
+
+               /* Calculate length for next transfer --
+                * Are we done queueing all the TRBs for this sg entry?
+                */
+               this_sg_len -= trb_buff_len;
+               if (this_sg_len == 0) {
+                       --num_sgs;
+                       if (num_sgs == 0)
+                               break;
+                       sg = sg_next(sg);
+                       addr = (u64) sg_dma_address(sg);
+                       this_sg_len = sg_dma_len(sg);
+               } else {
+                       addr += trb_buff_len;
+               }
+
+               trb_buff_len = TRB_MAX_BUFF_SIZE -
+                       (addr & (TRB_MAX_BUFF_SIZE - 1));
+               trb_buff_len = min_t(int, trb_buff_len, this_sg_len);
+               if (running_total + trb_buff_len > urb->transfer_buffer_length)
+                       trb_buff_len =
+                               urb->transfer_buffer_length - running_total;
+       } while (running_total < urb->transfer_buffer_length);
+
+       check_trb_math(urb, num_trbs, running_total);
+       giveback_first_trb(xhci, slot_id, ep_index, urb->stream_id,
+                       start_cycle, start_trb);
+       return 0;
+}
+
+/* This is very similar to what ehci-q.c qtd_fill() does */
+int xhci_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 urb_priv *urb_priv;
+       struct xhci_td *td;
+       int num_trbs;
+       struct xhci_generic_trb *start_trb;
+       bool first_trb;
+       bool more_trbs_coming;
+       int start_cycle;
+       u32 field, length_field;
+
+       int running_total, trb_buff_len, ret;
+       unsigned int total_packet_count;
+       u64 addr;
+
+       if (urb->num_sgs)
+               return queue_bulk_sg_tx(xhci, mem_flags, urb, slot_id, ep_index);
+
+       ep_ring = xhci_urb_to_transfer_ring(xhci, urb);
+       if (!ep_ring)
+               return -EINVAL;
+
+       num_trbs = 0;
+       /* How much data is (potentially) left before the 64KB boundary? */
+       running_total = TRB_MAX_BUFF_SIZE -
+               (urb->transfer_dma & (TRB_MAX_BUFF_SIZE - 1));
+       running_total &= TRB_MAX_BUFF_SIZE - 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 */
+
+       ret = prepare_transfer(xhci, xhci->devs[slot_id],
+                       ep_index, urb->stream_id,
+                       num_trbs, urb, 0, mem_flags);
+       if (ret < 0)
+               return ret;
+
+       urb_priv = urb->hcpriv;
+       td = urb_priv->td[0];
+
+       /*
+        * 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;
+       total_packet_count = DIV_ROUND_UP(urb->transfer_buffer_length,
+                       usb_endpoint_maxp(&urb->ep->desc));
+       /* How much data is in the first TRB? */
+       addr = (u64) urb->transfer_dma;
+       trb_buff_len = TRB_MAX_BUFF_SIZE -
+               (urb->transfer_dma & (TRB_MAX_BUFF_SIZE - 1));
+       if (trb_buff_len > urb->transfer_buffer_length)
+               trb_buff_len = urb->transfer_buffer_length;
+
+       first_trb = true;
+
+       /* Queue the first TRB, even if it's zero-length */
+       do {
+               u32 remainder = 0;
+               field = 0;
+
+               /* Don't change the cycle bit of the first TRB until later */
+               if (first_trb) {
+                       first_trb = false;
+                       if (start_cycle == 0)
+                               field |= 0x1;
+               } 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;
+               }
+
+               /* Only set interrupt on short packet for IN endpoints */
+               if (usb_urb_dir_in(urb))
+                       field |= TRB_ISP;
+
+               /* Set the TRB length, TD size, and interrupter fields. */
+               if (xhci->hci_version < 0x100) {
+                       remainder = xhci_td_remainder(
+                                       urb->transfer_buffer_length -
+                                       running_total);
+               } else {
+                       remainder = xhci_v1_0_td_remainder(running_total,
+                                       trb_buff_len, total_packet_count, urb,
+                                       num_trbs - 1);
+               }
+               length_field = TRB_LEN(trb_buff_len) |
+                       remainder |
+                       TRB_INTR_TARGET(0);
+
+               if (num_trbs > 1)
+                       more_trbs_coming = true;
+               else
+                       more_trbs_coming = false;
+               queue_trb(xhci, ep_ring, more_trbs_coming,
+                               lower_32_bits(addr),
+                               upper_32_bits(addr),
+                               length_field,
+                               field | 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);
+
+       check_trb_math(urb, num_trbs, running_total);
+       giveback_first_trb(xhci, slot_id, ep_index, urb->stream_id,
+                       start_cycle, start_trb);
+       return 0;
+}
+
+/* Caller must have locked xhci->lock */
+int xhci_queue_ctrl_tx(struct xhci_hcd *xhci, gfp_t mem_flags,
+               struct urb *urb, int slot_id, unsigned int ep_index)
+{
+       struct xhci_ring *ep_ring;
+       int num_trbs;
+       int ret;
+       struct usb_ctrlrequest *setup;
+       struct xhci_generic_trb *start_trb;
+       int start_cycle;
+       u32 field, length_field;
+       struct urb_priv *urb_priv;
+       struct xhci_td *td;
+
+       ep_ring = xhci_urb_to_transfer_ring(xhci, urb);
+       if (!ep_ring)
+               return -EINVAL;
+
+       /*
+        * Need to copy setup packet into setup TRB, so we can't use the setup
+        * DMA address.
+        */
+       if (!urb->setup_packet)
+               return -EINVAL;
+
+       /* 1 TRB for setup, 1 for status */
+       num_trbs = 2;
+       /*
+        * Don't need to check if we need additional event data and normal TRBs,
+        * since data in control transfers will never get bigger than 16MB
+        * XXX: can we get a buffer that crosses 64KB boundaries?
+        */
+       if (urb->transfer_buffer_length > 0)
+               num_trbs++;
+       ret = prepare_transfer(xhci, xhci->devs[slot_id],
+                       ep_index, urb->stream_id,
+                       num_trbs, urb, 0, mem_flags);
+       if (ret < 0)
+               return ret;
+
+       urb_priv = urb->hcpriv;
+       td = urb_priv->td[0];
+
+       /*
+        * 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;
+
+       /* Queue setup TRB - see section 6.4.1.2.1 */
+       /* FIXME better way to translate setup_packet into two u32 fields? */
+       setup = (struct usb_ctrlrequest *) urb->setup_packet;
+       field = 0;
+       field |= TRB_IDT | TRB_TYPE(TRB_SETUP);
+       if (start_cycle == 0)
+               field |= 0x1;
+
+       /* xHCI 1.0 6.4.1.2.1: Transfer Type field */
+       if (xhci->hci_version == 0x100) {
+               if (urb->transfer_buffer_length > 0) {
+                       if (setup->bRequestType & USB_DIR_IN)
+                               field |= TRB_TX_TYPE(TRB_DATA_IN);
+                       else
+                               field |= TRB_TX_TYPE(TRB_DATA_OUT);
+               }
+       }
+
+       queue_trb(xhci, ep_ring, true,
+                 setup->bRequestType | setup->bRequest << 8 | le16_to_cpu(setup->wValue) << 16,
+                 le16_to_cpu(setup->wIndex) | le16_to_cpu(setup->wLength) << 16,
+                 TRB_LEN(8) | TRB_INTR_TARGET(0),
+                 /* Immediate data in pointer */
+                 field);
+
+       /* If there's data, queue data TRBs */
+       /* Only set interrupt on short packet for IN endpoints */
+       if (usb_urb_dir_in(urb))
+               field = TRB_ISP | TRB_TYPE(TRB_DATA);
+       else
+               field = TRB_TYPE(TRB_DATA);
+
+       length_field = TRB_LEN(urb->transfer_buffer_length) |
+               xhci_td_remainder(urb->transfer_buffer_length) |
+               TRB_INTR_TARGET(0);
+       if (urb->transfer_buffer_length > 0) {
+               if (setup->bRequestType & USB_DIR_IN)
+                       field |= TRB_DIR_IN;
+               queue_trb(xhci, ep_ring, true,
+                               lower_32_bits(urb->transfer_dma),
+                               upper_32_bits(urb->transfer_dma),
+                               length_field,
+                               field | ep_ring->cycle_state);
+       }
+
+       /* Save the DMA address of the last TRB in the TD */
+       td->last_trb = ep_ring->enqueue;
+
+       /* Queue status TRB - see Table 7 and sections 4.11.2.2 and 6.4.1.2.3 */
+       /* If the device sent data, the status stage is an OUT transfer */
+       if (urb->transfer_buffer_length > 0 && setup->bRequestType & USB_DIR_IN)
+               field = 0;
+       else
+               field = TRB_DIR_IN;
+       queue_trb(xhci, ep_ring, false,
+                       0,
+                       0,
+                       TRB_INTR_TARGET(0),
+                       /* Event on completion */
+                       field | TRB_IOC | TRB_TYPE(TRB_STATUS) | ep_ring->cycle_state);
+
+       giveback_first_trb(xhci, slot_id, ep_index, 0,
+                       start_cycle, start_trb);
+       return 0;
+}
+
+static int count_isoc_trbs_needed(struct xhci_hcd *xhci,
+               struct urb *urb, int i)
+{
+       int num_trbs = 0;
+       u64 addr, td_len;
+
+       addr = (u64) (urb->transfer_dma + urb->iso_frame_desc[i].offset);
+       td_len = urb->iso_frame_desc[i].length;
+
+       num_trbs = DIV_ROUND_UP(td_len + (addr & (TRB_MAX_BUFF_SIZE - 1)),
+                       TRB_MAX_BUFF_SIZE);
+       if (num_trbs == 0)
+               num_trbs++;
+
+       return num_trbs;
+}
+
+/*
+ * The transfer burst count field of the isochronous TRB defines the number of
+ * bursts that are required to move all packets in this TD.  Only SuperSpeed
+ * devices can burst up to bMaxBurst number of packets per service interval.
+ * This field is zero based, meaning a value of zero in the field means one
+ * burst.  Basically, for everything but SuperSpeed devices, this field will be
+ * zero.  Only xHCI 1.0 host controllers support this field.
+ */
+static unsigned int xhci_get_burst_count(struct xhci_hcd *xhci,
+               struct usb_device *udev,
+               struct urb *urb, unsigned int total_packet_count)
+{
+       unsigned int max_burst;
+
+       if (xhci->hci_version < 0x100 || udev->speed != USB_SPEED_SUPER)
+               return 0;
+
+       max_burst = urb->ep->ss_ep_comp.bMaxBurst;
+       return DIV_ROUND_UP(total_packet_count, max_burst + 1) - 1;
+}
+
+/*
+ * Returns the number of packets in the last "burst" of packets.  This field is
+ * valid for all speeds of devices.  USB 2.0 devices can only do one "burst", so
+ * the last burst packet count is equal to the total number of packets in the
+ * TD.  SuperSpeed endpoints can have up to 3 bursts.  All but the last burst
+ * must contain (bMaxBurst + 1) number of packets, but the last burst can
+ * contain 1 to (bMaxBurst + 1) packets.
+ */
+static unsigned int xhci_get_last_burst_packet_count(struct xhci_hcd *xhci,
+               struct usb_device *udev,
+               struct urb *urb, unsigned int total_packet_count)
+{
+       unsigned int max_burst;
+       unsigned int residue;
+
+       if (xhci->hci_version < 0x100)
+               return 0;
+
+       switch (udev->speed) {
+       case USB_SPEED_SUPER:
+               /* bMaxBurst is zero based: 0 means 1 packet per burst */
+               max_burst = urb->ep->ss_ep_comp.bMaxBurst;
+               residue = total_packet_count % (max_burst + 1);
+               /* If residue is zero, the last burst contains (max_burst + 1)
+                * number of packets, but the TLBPC field is zero-based.
+                */
+               if (residue == 0)
+                       return max_burst;
+               return residue - 1;
+       default:
+               if (total_packet_count == 0)
+                       return 0;
+               return total_packet_count - 1;
+       }
+}
+
+/* This is for isoc transfer */
+static int xhci_queue_isoc_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 urb_priv *urb_priv;
+       struct xhci_td *td;
+       int num_tds, trbs_per_td;
+       struct xhci_generic_trb *start_trb;
+       bool first_trb;
+       int start_cycle;
+       u32 field, length_field;
+       int running_total, trb_buff_len, td_len, td_remain_len, ret;
+       u64 start_addr, addr;
+       int i, j;
+       bool more_trbs_coming;
+
+       ep_ring = xhci->devs[slot_id]->eps[ep_index].ring;
+
+       num_tds = urb->number_of_packets;
+       if (num_tds < 1) {
+               xhci_dbg(xhci, "Isoc URB with zero packets?\n");
+               return -EINVAL;
+       }
+
+       start_addr = (u64) urb->transfer_dma;
+       start_trb = &ep_ring->enqueue->generic;
+       start_cycle = ep_ring->cycle_state;
+
+       urb_priv = urb->hcpriv;
+       /* Queue the first TRB, even if it's zero-length */
+       for (i = 0; i < num_tds; i++) {
+               unsigned int total_packet_count;
+               unsigned int burst_count;
+               unsigned int residue;
+
+               first_trb = true;
+               running_total = 0;
+               addr = start_addr + urb->iso_frame_desc[i].offset;
+               td_len = urb->iso_frame_desc[i].length;
+               td_remain_len = td_len;
+               total_packet_count = DIV_ROUND_UP(td_len,
+                               GET_MAX_PACKET(
+                                       usb_endpoint_maxp(&urb->ep->desc)));
+               /* A zero-length transfer still involves at least one packet. */
+               if (total_packet_count == 0)
+                       total_packet_count++;
+               burst_count = xhci_get_burst_count(xhci, urb->dev, urb,
+                               total_packet_count);
+               residue = xhci_get_last_burst_packet_count(xhci,
+                               urb->dev, urb, total_packet_count);
+
+               trbs_per_td = count_isoc_trbs_needed(xhci, urb, i);
+
+               ret = prepare_transfer(xhci, xhci->devs[slot_id], ep_index,
+                               urb->stream_id, trbs_per_td, urb, i, mem_flags);
+               if (ret < 0) {
+                       if (i == 0)
+                               return ret;
+                       goto cleanup;
+               }
+
+               td = urb_priv->td[i];
+               for (j = 0; j < trbs_per_td; j++) {
+                       u32 remainder = 0;
+                       field = 0;
+
+                       if (first_trb) {
+                               field = TRB_TBC(burst_count) |
+                                       TRB_TLBPC(residue);
+                               /* Queue the isoc TRB */
+                               field |= TRB_TYPE(TRB_ISOC);
+                               /* Assume URB_ISO_ASAP is set */
+                               field |= TRB_SIA;
+                               if (i == 0) {
+                                       if (start_cycle == 0)
+                                               field |= 0x1;
+                               } else
+                                       field |= ep_ring->cycle_state;
+                               first_trb = false;
+                       } else {
+                               /* Queue other normal TRBs */
+                               field |= TRB_TYPE(TRB_NORMAL);
+                               field |= ep_ring->cycle_state;
+                       }
+
+                       /* Only set interrupt on short packet for IN EPs */
+                       if (usb_urb_dir_in(urb))
+                               field |= TRB_ISP;
+
+                       /* Chain all the TRBs together; clear the chain bit in
+                        * the last TRB to indicate it's the last TRB in the
+                        * chain.
+                        */
+                       if (j < trbs_per_td - 1) {
+                               field |= TRB_CHAIN;
+                               more_trbs_coming = true;
+                       } else {
+                               td->last_trb = ep_ring->enqueue;
+                               field |= TRB_IOC;
+                               if (xhci->hci_version == 0x100 &&
+                                               !(xhci->quirks &
+                                                       XHCI_AVOID_BEI)) {
+                                       /* Set BEI bit except for the last td */
+                                       if (i < num_tds - 1)
+                                               field |= TRB_BEI;
+                               }
+                               more_trbs_coming = false;
+                       }
+
+                       /* Calculate TRB length */
+                       trb_buff_len = TRB_MAX_BUFF_SIZE -
+                               (addr & ((1 << TRB_MAX_BUFF_SHIFT) - 1));
+                       if (trb_buff_len > td_remain_len)
+                               trb_buff_len = td_remain_len;
+
+                       /* Set the TRB length, TD size, & interrupter fields. */
+                       if (xhci->hci_version < 0x100) {
+                               remainder = xhci_td_remainder(
+                                               td_len - running_total);
+                       } else {
+                               remainder = xhci_v1_0_td_remainder(
+                                               running_total, trb_buff_len,
+                                               total_packet_count, urb,
+                                               (trbs_per_td - j - 1));
+                       }
+                       length_field = TRB_LEN(trb_buff_len) |
+                               remainder |
+                               TRB_INTR_TARGET(0);
+
+                       queue_trb(xhci, ep_ring, more_trbs_coming,
+                               lower_32_bits(addr),
+                               upper_32_bits(addr),
+                               length_field,
+                               field);
+                       running_total += trb_buff_len;
+
+                       addr += trb_buff_len;
+                       td_remain_len -= trb_buff_len;
+               }
+
+               /* Check TD length */
+               if (running_total != td_len) {
+                       xhci_err(xhci, "ISOC TD length unmatch\n");
+                       ret = -EINVAL;
+                       goto cleanup;
+               }
+       }
+
+       if (xhci_to_hcd(xhci)->self.bandwidth_isoc_reqs == 0) {
+               if (xhci->quirks & XHCI_AMD_PLL_FIX)
+                       usb_amd_quirk_pll_disable();
+       }
+       xhci_to_hcd(xhci)->self.bandwidth_isoc_reqs++;
+
+       giveback_first_trb(xhci, slot_id, ep_index, urb->stream_id,
+                       start_cycle, start_trb);
+       return 0;
+cleanup:
+       /* Clean up a partially enqueued isoc transfer. */
+
+       for (i--; i >= 0; i--)
+               list_del_init(&urb_priv->td[i]->td_list);
+
+       /* Use the first TD as a temporary variable to turn the TDs we've queued
+        * into No-ops with a software-owned cycle bit. That way the hardware
+        * won't accidentally start executing bogus TDs when we partially
+        * overwrite them.  td->first_trb and td->start_seg are already set.
+        */
+       urb_priv->td[0]->last_trb = ep_ring->enqueue;
+       /* Every TRB except the first & last will have its cycle bit flipped. */
+       td_to_noop(xhci, ep_ring, urb_priv->td[0], true);
+
+       /* Reset the ring enqueue back to the first TRB and its cycle bit. */
+       ep_ring->enqueue = urb_priv->td[0]->first_trb;
+       ep_ring->enq_seg = urb_priv->td[0]->start_seg;
+       ep_ring->cycle_state = start_cycle;
+       ep_ring->num_trbs_free = ep_ring->num_trbs_free_temp;
+       usb_hcd_unlink_urb_from_ep(bus_to_hcd(urb->dev->bus), urb);
+       return ret;
+}
+
+/*
+ * Check transfer ring to guarantee there is enough room for the urb.
+ * Update ISO URB start_frame and interval.
+ * Update interval as xhci_queue_intr_tx does. Just use xhci frame_index to
+ * update the urb->start_frame by now.
+ * Always assume URB_ISO_ASAP set, and NEVER use urb->start_frame as input.
+ */
+int xhci_queue_isoc_tx_prepare(struct xhci_hcd *xhci, gfp_t mem_flags,
+               struct urb *urb, int slot_id, unsigned int ep_index)
+{
+       struct xhci_virt_device *xdev;
+       struct xhci_ring *ep_ring;
+       struct xhci_ep_ctx *ep_ctx;
+       int start_frame;
+       int xhci_interval;
+       int ep_interval;
+       int num_tds, num_trbs, i;
+       int ret;
+
+       xdev = xhci->devs[slot_id];
+       ep_ring = xdev->eps[ep_index].ring;
+       ep_ctx = xhci_get_ep_ctx(xhci, xdev->out_ctx, ep_index);
+
+       num_trbs = 0;
+       num_tds = urb->number_of_packets;
+       for (i = 0; i < num_tds; i++)
+               num_trbs += count_isoc_trbs_needed(xhci, urb, i);
+
+       /* Check the ring to guarantee there is enough room for the whole urb.
+        * Do not insert any td of the urb to the ring if the check failed.
+        */
+       ret = prepare_ring(xhci, ep_ring, le32_to_cpu(ep_ctx->ep_info) & EP_STATE_MASK,
+                          num_trbs, mem_flags);
+       if (ret)
+               return ret;
+
+       start_frame = readl(&xhci->run_regs->microframe_index);
+       start_frame &= 0x3fff;
+
+       urb->start_frame = start_frame;
+       if (urb->dev->speed == USB_SPEED_LOW ||
+                       urb->dev->speed == USB_SPEED_FULL)
+               urb->start_frame >>= 3;
+
+       xhci_interval = EP_INTERVAL_TO_UFRAMES(le32_to_cpu(ep_ctx->ep_info));
+       ep_interval = urb->interval;
+       /* Convert to microframes */
+       if (urb->dev->speed == USB_SPEED_LOW ||
+                       urb->dev->speed == USB_SPEED_FULL)
+               ep_interval *= 8;
+       /* FIXME change this to a warning and a suggestion to use the new API
+        * to set the polling interval (once the API is added).
+        */
+       if (xhci_interval != ep_interval) {
+               dev_dbg_ratelimited(&urb->dev->dev,
+                               "Driver uses different interval (%d microframe%s) than xHCI (%d microframe%s)\n",
+                               ep_interval, ep_interval == 1 ? "" : "s",
+                               xhci_interval, xhci_interval == 1 ? "" : "s");
+               urb->interval = xhci_interval;
+               /* Convert back to frames for LS/FS devices */
+               if (urb->dev->speed == USB_SPEED_LOW ||
+                               urb->dev->speed == USB_SPEED_FULL)
+                       urb->interval /= 8;
+       }
+       ep_ring->num_trbs_free_temp = ep_ring->num_trbs_free;
+
+       return xhci_queue_isoc_tx(xhci, mem_flags, urb, slot_id, ep_index);
+}
+
+/****          Command Ring Operations         ****/
+
+/* Generic function for queueing a command TRB on the command ring.
+ * Check to make sure there's room on the command ring for one command TRB.
+ * Also check that there's room reserved for commands that must not fail.
+ * If this is a command that must not fail, meaning command_must_succeed = TRUE,
+ * then only check for the number of reserved spots.
+ * Don't decrement xhci->cmd_ring_reserved_trbs after we've queued the TRB
+ * because the command event handler may want to resubmit a failed command.
+ */
+static int queue_command(struct xhci_hcd *xhci, struct xhci_command *cmd,
+                        u32 field1, u32 field2,
+                        u32 field3, u32 field4, bool command_must_succeed)
+{
+       int reserved_trbs = xhci->cmd_ring_reserved_trbs;
+       int ret;
+       if (xhci->xhc_state & XHCI_STATE_DYING)
+               return -ESHUTDOWN;
+
+       if (!command_must_succeed)
+               reserved_trbs++;
+
+       ret = prepare_ring(xhci, xhci->cmd_ring, EP_STATE_RUNNING,
+                       reserved_trbs, GFP_ATOMIC);
+       if (ret < 0) {
+               xhci_err(xhci, "ERR: No room for command on command ring\n");
+               if (command_must_succeed)
+                       xhci_err(xhci, "ERR: Reserved TRB counting for "
+                                       "unfailable commands failed.\n");
+               return ret;
+       }
+
+       cmd->command_trb = xhci->cmd_ring->enqueue;
+       list_add_tail(&cmd->cmd_list, &xhci->cmd_list);
+
+       /* if there are no other commands queued we start the timeout timer */
+       if (xhci->cmd_list.next == &cmd->cmd_list &&
+           !timer_pending(&xhci->cmd_timer)) {
+               xhci->current_cmd = cmd;
+               mod_timer(&xhci->cmd_timer, jiffies + XHCI_CMD_DEFAULT_TIMEOUT);
+       }
+
+       queue_trb(xhci, xhci->cmd_ring, false, field1, field2, field3,
+                       field4 | xhci->cmd_ring->cycle_state);
+       return 0;
+}
+
+/* Queue a slot enable or disable request on the command ring */
+int xhci_queue_slot_control(struct xhci_hcd *xhci, struct xhci_command *cmd,
+               u32 trb_type, u32 slot_id)
+{
+       return queue_command(xhci, cmd, 0, 0, 0,
+                       TRB_TYPE(trb_type) | SLOT_ID_FOR_TRB(slot_id), false);
+}
+
+/* Queue an address device command TRB */
+int xhci_queue_address_device(struct xhci_hcd *xhci, struct xhci_command *cmd,
+               dma_addr_t in_ctx_ptr, u32 slot_id, enum xhci_setup_dev setup)
+{
+       return queue_command(xhci, cmd, lower_32_bits(in_ctx_ptr),
+                       upper_32_bits(in_ctx_ptr), 0,
+                       TRB_TYPE(TRB_ADDR_DEV) | SLOT_ID_FOR_TRB(slot_id)
+                       | (setup == SETUP_CONTEXT_ONLY ? TRB_BSR : 0), false);
+}
+
+int xhci_queue_vendor_command(struct xhci_hcd *xhci, struct xhci_command *cmd,
+               u32 field1, u32 field2, u32 field3, u32 field4)
+{
+       return queue_command(xhci, cmd, field1, field2, field3, field4, false);
+}
+
+/* Queue a reset device command TRB */
+int xhci_queue_reset_device(struct xhci_hcd *xhci, struct xhci_command *cmd,
+               u32 slot_id)
+{
+       return queue_command(xhci, cmd, 0, 0, 0,
+                       TRB_TYPE(TRB_RESET_DEV) | SLOT_ID_FOR_TRB(slot_id),
+                       false);
+}
+
+/* Queue a configure endpoint command TRB */
+int xhci_queue_configure_endpoint(struct xhci_hcd *xhci,
+               struct xhci_command *cmd, dma_addr_t in_ctx_ptr,
+               u32 slot_id, bool command_must_succeed)
+{
+       return queue_command(xhci, cmd, lower_32_bits(in_ctx_ptr),
+                       upper_32_bits(in_ctx_ptr), 0,
+                       TRB_TYPE(TRB_CONFIG_EP) | SLOT_ID_FOR_TRB(slot_id),
+                       command_must_succeed);
+}
+
+/* Queue an evaluate context command TRB */
+int xhci_queue_evaluate_context(struct xhci_hcd *xhci, struct xhci_command *cmd,
+               dma_addr_t in_ctx_ptr, u32 slot_id, bool command_must_succeed)
+{
+       return queue_command(xhci, cmd, lower_32_bits(in_ctx_ptr),
+                       upper_32_bits(in_ctx_ptr), 0,
+                       TRB_TYPE(TRB_EVAL_CONTEXT) | SLOT_ID_FOR_TRB(slot_id),
+                       command_must_succeed);
+}
+
+/*
+ * Suspend is set to indicate "Stop Endpoint Command" is being issued to stop
+ * activity on an endpoint that is about to be suspended.
+ */
+int xhci_queue_stop_endpoint(struct xhci_hcd *xhci, struct xhci_command *cmd,
+                            int slot_id, unsigned int ep_index, int suspend)
+{
+       u32 trb_slot_id = SLOT_ID_FOR_TRB(slot_id);
+       u32 trb_ep_index = EP_ID_FOR_TRB(ep_index);
+       u32 type = TRB_TYPE(TRB_STOP_RING);
+       u32 trb_suspend = SUSPEND_PORT_FOR_TRB(suspend);
+
+       return queue_command(xhci, cmd, 0, 0, 0,
+                       trb_slot_id | trb_ep_index | type | trb_suspend, false);
+}
+
+/* Set Transfer Ring Dequeue Pointer command */
+void xhci_queue_new_dequeue_state(struct xhci_hcd *xhci,
+               unsigned int slot_id, unsigned int ep_index,
+               unsigned int stream_id,
+               struct xhci_dequeue_state *deq_state)
+{
+       dma_addr_t addr;
+       u32 trb_slot_id = SLOT_ID_FOR_TRB(slot_id);
+       u32 trb_ep_index = EP_ID_FOR_TRB(ep_index);
+       u32 trb_stream_id = STREAM_ID_FOR_TRB(stream_id);
+       u32 trb_sct = 0;
+       u32 type = TRB_TYPE(TRB_SET_DEQ);
+       struct xhci_virt_ep *ep;
+       struct xhci_command *cmd;
+       int ret;
+
+       xhci_dbg_trace(xhci, trace_xhci_dbg_cancel_urb,
+               "Set TR Deq Ptr cmd, new deq seg = %p (0x%llx dma), new deq ptr = %p (0x%llx dma), new cycle = %u",
+               deq_state->new_deq_seg,
+               (unsigned long long)deq_state->new_deq_seg->dma,
+               deq_state->new_deq_ptr,
+               (unsigned long long)xhci_trb_virt_to_dma(
+                       deq_state->new_deq_seg, deq_state->new_deq_ptr),
+               deq_state->new_cycle_state);
+
+       addr = xhci_trb_virt_to_dma(deq_state->new_deq_seg,
+                                   deq_state->new_deq_ptr);
+       if (addr == 0) {
+               xhci_warn(xhci, "WARN Cannot submit Set TR Deq Ptr\n");
+               xhci_warn(xhci, "WARN deq seg = %p, deq pt = %p\n",
+                         deq_state->new_deq_seg, deq_state->new_deq_ptr);
+               return;
+       }
+       ep = &xhci->devs[slot_id]->eps[ep_index];
+       if ((ep->ep_state & SET_DEQ_PENDING)) {
+               xhci_warn(xhci, "WARN Cannot submit Set TR Deq Ptr\n");
+               xhci_warn(xhci, "A Set TR Deq Ptr command is pending.\n");
+               return;
+       }
+
+       /* This function gets called from contexts where it cannot sleep */
+       cmd = xhci_alloc_command(xhci, false, false, GFP_ATOMIC);
+       if (!cmd) {
+               xhci_warn(xhci, "WARN Cannot submit Set TR Deq Ptr: ENOMEM\n");
+               return;
+       }
+
+       ep->queued_deq_seg = deq_state->new_deq_seg;
+       ep->queued_deq_ptr = deq_state->new_deq_ptr;
+       if (stream_id)
+               trb_sct = SCT_FOR_TRB(SCT_PRI_TR);
+       ret = queue_command(xhci, cmd,
+               lower_32_bits(addr) | trb_sct | deq_state->new_cycle_state,
+               upper_32_bits(addr), trb_stream_id,
+               trb_slot_id | trb_ep_index | type, false);
+       if (ret < 0) {
+               xhci_free_command(xhci, cmd);
+               return;
+       }
+
+       /* Stop the TD queueing code from ringing the doorbell until
+        * this command completes.  The HC won't set the dequeue pointer
+        * if the ring is running, and ringing the doorbell starts the
+        * ring running.
+        */
+       ep->ep_state |= SET_DEQ_PENDING;
+}
+
+int xhci_queue_reset_ep(struct xhci_hcd *xhci, struct xhci_command *cmd,
+                       int slot_id, unsigned int ep_index)
+{
+       u32 trb_slot_id = SLOT_ID_FOR_TRB(slot_id);
+       u32 trb_ep_index = EP_ID_FOR_TRB(ep_index);
+       u32 type = TRB_TYPE(TRB_RESET_EP);
+
+       return queue_command(xhci, cmd, 0, 0, 0,
+                       trb_slot_id | trb_ep_index | type, false);
+}