Add qemu 2.4.0
[kvmfornfv.git] / qemu / roms / u-boot / drivers / usb / musb-new / musb_host.c
diff --git a/qemu/roms/u-boot/drivers/usb/musb-new/musb_host.c b/qemu/roms/u-boot/drivers/usb/musb-new/musb_host.c
new file mode 100644 (file)
index 0000000..9a2cf59
--- /dev/null
@@ -0,0 +1,2400 @@
+/*
+ * MUSB OTG driver host support
+ *
+ * Copyright 2005 Mentor Graphics Corporation
+ * Copyright (C) 2005-2006 by Texas Instruments
+ * Copyright (C) 2006-2007 Nokia Corporation
+ * Copyright (C) 2008-2009 MontaVista Software, Inc. <source@mvista.com>
+ *
+ * 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., 51 Franklin St, Fifth Floor, Boston, MA
+ * 02110-1301 USA
+ *
+ * THIS SOFTWARE IS PROVIDED "AS IS" AND ANY EXPRESS OR IMPLIED
+ * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN
+ * NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
+ * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF
+ * USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ * ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
+ * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ */
+
+#define __UBOOT__
+#ifndef __UBOOT__
+#include <linux/module.h>
+#include <linux/kernel.h>
+#include <linux/delay.h>
+#include <linux/sched.h>
+#include <linux/slab.h>
+#include <linux/errno.h>
+#include <linux/init.h>
+#include <linux/list.h>
+#include <linux/dma-mapping.h>
+#else
+#include <common.h>
+#include <usb.h>
+#include "linux-compat.h"
+#include "usb-compat.h"
+#endif
+
+#include "musb_core.h"
+#include "musb_host.h"
+
+
+/* MUSB HOST status 22-mar-2006
+ *
+ * - There's still lots of partial code duplication for fault paths, so
+ *   they aren't handled as consistently as they need to be.
+ *
+ * - PIO mostly behaved when last tested.
+ *     + including ep0, with all usbtest cases 9, 10
+ *     + usbtest 14 (ep0out) doesn't seem to run at all
+ *     + double buffered OUT/TX endpoints saw stalls(!) with certain usbtest
+ *       configurations, but otherwise double buffering passes basic tests.
+ *     + for 2.6.N, for N > ~10, needs API changes for hcd framework.
+ *
+ * - DMA (CPPI) ... partially behaves, not currently recommended
+ *     + about 1/15 the speed of typical EHCI implementations (PCI)
+ *     + RX, all too often reqpkt seems to misbehave after tx
+ *     + TX, no known issues (other than evident silicon issue)
+ *
+ * - DMA (Mentor/OMAP) ...has at least toggle update problems
+ *
+ * - [23-feb-2009] minimal traffic scheduling to avoid bulk RX packet
+ *   starvation ... nothing yet for TX, interrupt, or bulk.
+ *
+ * - Not tested with HNP, but some SRP paths seem to behave.
+ *
+ * NOTE 24-August-2006:
+ *
+ * - Bulk traffic finally uses both sides of hardware ep1, freeing up an
+ *   extra endpoint for periodic use enabling hub + keybd + mouse.  That
+ *   mostly works, except that with "usbnet" it's easy to trigger cases
+ *   with "ping" where RX loses.  (a) ping to davinci, even "ping -f",
+ *   fine; but (b) ping _from_ davinci, even "ping -c 1", ICMP RX loses
+ *   although ARP RX wins.  (That test was done with a full speed link.)
+ */
+
+
+/*
+ * NOTE on endpoint usage:
+ *
+ * CONTROL transfers all go through ep0.  BULK ones go through dedicated IN
+ * and OUT endpoints ... hardware is dedicated for those "async" queue(s).
+ * (Yes, bulk _could_ use more of the endpoints than that, and would even
+ * benefit from it.)
+ *
+ * INTERUPPT and ISOCHRONOUS transfers are scheduled to the other endpoints.
+ * So far that scheduling is both dumb and optimistic:  the endpoint will be
+ * "claimed" until its software queue is no longer refilled.  No multiplexing
+ * of transfers between endpoints, or anything clever.
+ */
+
+
+static void musb_ep_program(struct musb *musb, u8 epnum,
+                       struct urb *urb, int is_out,
+                       u8 *buf, u32 offset, u32 len);
+
+/*
+ * Clear TX fifo. Needed to avoid BABBLE errors.
+ */
+static void musb_h_tx_flush_fifo(struct musb_hw_ep *ep)
+{
+       struct musb     *musb = ep->musb;
+       void __iomem    *epio = ep->regs;
+       u16             csr;
+       u16             lastcsr = 0;
+       int             retries = 1000;
+
+       csr = musb_readw(epio, MUSB_TXCSR);
+       while (csr & MUSB_TXCSR_FIFONOTEMPTY) {
+               if (csr != lastcsr)
+                       dev_dbg(musb->controller, "Host TX FIFONOTEMPTY csr: %02x\n", csr);
+               lastcsr = csr;
+               csr |= MUSB_TXCSR_FLUSHFIFO;
+               musb_writew(epio, MUSB_TXCSR, csr);
+               csr = musb_readw(epio, MUSB_TXCSR);
+               if (WARN(retries-- < 1,
+                               "Could not flush host TX%d fifo: csr: %04x\n",
+                               ep->epnum, csr))
+                       return;
+               mdelay(1);
+       }
+}
+
+static void musb_h_ep0_flush_fifo(struct musb_hw_ep *ep)
+{
+       void __iomem    *epio = ep->regs;
+       u16             csr;
+       int             retries = 5;
+
+       /* scrub any data left in the fifo */
+       do {
+               csr = musb_readw(epio, MUSB_TXCSR);
+               if (!(csr & (MUSB_CSR0_TXPKTRDY | MUSB_CSR0_RXPKTRDY)))
+                       break;
+               musb_writew(epio, MUSB_TXCSR, MUSB_CSR0_FLUSHFIFO);
+               csr = musb_readw(epio, MUSB_TXCSR);
+               udelay(10);
+       } while (--retries);
+
+       WARN(!retries, "Could not flush host TX%d fifo: csr: %04x\n",
+                       ep->epnum, csr);
+
+       /* and reset for the next transfer */
+       musb_writew(epio, MUSB_TXCSR, 0);
+}
+
+/*
+ * Start transmit. Caller is responsible for locking shared resources.
+ * musb must be locked.
+ */
+static inline void musb_h_tx_start(struct musb_hw_ep *ep)
+{
+       u16     txcsr;
+
+       /* NOTE: no locks here; caller should lock and select EP */
+       if (ep->epnum) {
+               txcsr = musb_readw(ep->regs, MUSB_TXCSR);
+               txcsr |= MUSB_TXCSR_TXPKTRDY | MUSB_TXCSR_H_WZC_BITS;
+               musb_writew(ep->regs, MUSB_TXCSR, txcsr);
+       } else {
+               txcsr = MUSB_CSR0_H_SETUPPKT | MUSB_CSR0_TXPKTRDY;
+               musb_writew(ep->regs, MUSB_CSR0, txcsr);
+       }
+
+}
+
+static inline void musb_h_tx_dma_start(struct musb_hw_ep *ep)
+{
+       u16     txcsr;
+
+       /* NOTE: no locks here; caller should lock and select EP */
+       txcsr = musb_readw(ep->regs, MUSB_TXCSR);
+       txcsr |= MUSB_TXCSR_DMAENAB | MUSB_TXCSR_H_WZC_BITS;
+       if (is_cppi_enabled())
+               txcsr |= MUSB_TXCSR_DMAMODE;
+       musb_writew(ep->regs, MUSB_TXCSR, txcsr);
+}
+
+static void musb_ep_set_qh(struct musb_hw_ep *ep, int is_in, struct musb_qh *qh)
+{
+       if (is_in != 0 || ep->is_shared_fifo)
+               ep->in_qh  = qh;
+       if (is_in == 0 || ep->is_shared_fifo)
+               ep->out_qh = qh;
+}
+
+static struct musb_qh *musb_ep_get_qh(struct musb_hw_ep *ep, int is_in)
+{
+       return is_in ? ep->in_qh : ep->out_qh;
+}
+
+/*
+ * Start the URB at the front of an endpoint's queue
+ * end must be claimed from the caller.
+ *
+ * Context: controller locked, irqs blocked
+ */
+static void
+musb_start_urb(struct musb *musb, int is_in, struct musb_qh *qh)
+{
+       u16                     frame;
+       u32                     len;
+       void __iomem            *mbase =  musb->mregs;
+       struct urb              *urb = next_urb(qh);
+       void                    *buf = urb->transfer_buffer;
+       u32                     offset = 0;
+       struct musb_hw_ep       *hw_ep = qh->hw_ep;
+       unsigned                pipe = urb->pipe;
+       u8                      address = usb_pipedevice(pipe);
+       int                     epnum = hw_ep->epnum;
+
+       /* initialize software qh state */
+       qh->offset = 0;
+       qh->segsize = 0;
+
+       /* gather right source of data */
+       switch (qh->type) {
+       case USB_ENDPOINT_XFER_CONTROL:
+               /* control transfers always start with SETUP */
+               is_in = 0;
+               musb->ep0_stage = MUSB_EP0_START;
+               buf = urb->setup_packet;
+               len = 8;
+               break;
+#ifndef __UBOOT__
+       case USB_ENDPOINT_XFER_ISOC:
+               qh->iso_idx = 0;
+               qh->frame = 0;
+               offset = urb->iso_frame_desc[0].offset;
+               len = urb->iso_frame_desc[0].length;
+               break;
+#endif
+       default:                /* bulk, interrupt */
+               /* actual_length may be nonzero on retry paths */
+               buf = urb->transfer_buffer + urb->actual_length;
+               len = urb->transfer_buffer_length - urb->actual_length;
+       }
+
+       dev_dbg(musb->controller, "qh %p urb %p dev%d ep%d%s%s, hw_ep %d, %p/%d\n",
+                       qh, urb, address, qh->epnum,
+                       is_in ? "in" : "out",
+                       ({char *s; switch (qh->type) {
+                       case USB_ENDPOINT_XFER_CONTROL: s = ""; break;
+                       case USB_ENDPOINT_XFER_BULK:    s = "-bulk"; break;
+#ifndef __UBOOT__
+                       case USB_ENDPOINT_XFER_ISOC:    s = "-iso"; break;
+#endif
+                       default:                        s = "-intr"; break;
+                       }; s; }),
+                       epnum, buf + offset, len);
+
+       /* Configure endpoint */
+       musb_ep_set_qh(hw_ep, is_in, qh);
+       musb_ep_program(musb, epnum, urb, !is_in, buf, offset, len);
+
+       /* transmit may have more work: start it when it is time */
+       if (is_in)
+               return;
+
+       /* determine if the time is right for a periodic transfer */
+       switch (qh->type) {
+#ifndef __UBOOT__
+       case USB_ENDPOINT_XFER_ISOC:
+#endif
+       case USB_ENDPOINT_XFER_INT:
+               dev_dbg(musb->controller, "check whether there's still time for periodic Tx\n");
+               frame = musb_readw(mbase, MUSB_FRAME);
+               /* FIXME this doesn't implement that scheduling policy ...
+                * or handle framecounter wrapping
+                */
+#ifndef __UBOOT__
+               if ((urb->transfer_flags & URB_ISO_ASAP)
+                               || (frame >= urb->start_frame)) {
+                       /* REVISIT the SOF irq handler shouldn't duplicate
+                        * this code; and we don't init urb->start_frame...
+                        */
+                       qh->frame = 0;
+                       goto start;
+               } else {
+#endif
+                       qh->frame = urb->start_frame;
+                       /* enable SOF interrupt so we can count down */
+                       dev_dbg(musb->controller, "SOF for %d\n", epnum);
+#if 1 /* ifndef        CONFIG_ARCH_DAVINCI */
+                       musb_writeb(mbase, MUSB_INTRUSBE, 0xff);
+#endif
+#ifndef __UBOOT__
+               }
+#endif
+               break;
+       default:
+start:
+               dev_dbg(musb->controller, "Start TX%d %s\n", epnum,
+                       hw_ep->tx_channel ? "dma" : "pio");
+
+               if (!hw_ep->tx_channel)
+                       musb_h_tx_start(hw_ep);
+               else if (is_cppi_enabled() || tusb_dma_omap())
+                       musb_h_tx_dma_start(hw_ep);
+       }
+}
+
+/* Context: caller owns controller lock, IRQs are blocked */
+static void musb_giveback(struct musb *musb, struct urb *urb, int status)
+__releases(musb->lock)
+__acquires(musb->lock)
+{
+       dev_dbg(musb->controller,
+                       "complete %p %pF (%d), dev%d ep%d%s, %d/%d\n",
+                       urb, urb->complete, status,
+                       usb_pipedevice(urb->pipe),
+                       usb_pipeendpoint(urb->pipe),
+                       usb_pipein(urb->pipe) ? "in" : "out",
+                       urb->actual_length, urb->transfer_buffer_length
+                       );
+
+       usb_hcd_unlink_urb_from_ep(musb_to_hcd(musb), urb);
+       spin_unlock(&musb->lock);
+       usb_hcd_giveback_urb(musb_to_hcd(musb), urb, status);
+       spin_lock(&musb->lock);
+}
+
+/* For bulk/interrupt endpoints only */
+static inline void musb_save_toggle(struct musb_qh *qh, int is_in,
+                                   struct urb *urb)
+{
+       void __iomem            *epio = qh->hw_ep->regs;
+       u16                     csr;
+
+       /*
+        * FIXME: the current Mentor DMA code seems to have
+        * problems getting toggle correct.
+        */
+
+       if (is_in)
+               csr = musb_readw(epio, MUSB_RXCSR) & MUSB_RXCSR_H_DATATOGGLE;
+       else
+               csr = musb_readw(epio, MUSB_TXCSR) & MUSB_TXCSR_H_DATATOGGLE;
+
+       usb_settoggle(urb->dev, qh->epnum, !is_in, csr ? 1 : 0);
+}
+
+/*
+ * Advance this hardware endpoint's queue, completing the specified URB and
+ * advancing to either the next URB queued to that qh, or else invalidating
+ * that qh and advancing to the next qh scheduled after the current one.
+ *
+ * Context: caller owns controller lock, IRQs are blocked
+ */
+static void musb_advance_schedule(struct musb *musb, struct urb *urb,
+                                 struct musb_hw_ep *hw_ep, int is_in)
+{
+       struct musb_qh          *qh = musb_ep_get_qh(hw_ep, is_in);
+       struct musb_hw_ep       *ep = qh->hw_ep;
+       int                     ready = qh->is_ready;
+       int                     status;
+
+       status = (urb->status == -EINPROGRESS) ? 0 : urb->status;
+
+       /* save toggle eagerly, for paranoia */
+       switch (qh->type) {
+       case USB_ENDPOINT_XFER_BULK:
+       case USB_ENDPOINT_XFER_INT:
+               musb_save_toggle(qh, is_in, urb);
+               break;
+#ifndef __UBOOT__
+       case USB_ENDPOINT_XFER_ISOC:
+               if (status == 0 && urb->error_count)
+                       status = -EXDEV;
+               break;
+#endif
+       }
+
+       qh->is_ready = 0;
+       musb_giveback(musb, urb, status);
+       qh->is_ready = ready;
+
+       /* reclaim resources (and bandwidth) ASAP; deschedule it, and
+        * invalidate qh as soon as list_empty(&hep->urb_list)
+        */
+       if (list_empty(&qh->hep->urb_list)) {
+               struct list_head        *head;
+               struct dma_controller   *dma = musb->dma_controller;
+
+               if (is_in) {
+                       ep->rx_reinit = 1;
+                       if (ep->rx_channel) {
+                               dma->channel_release(ep->rx_channel);
+                               ep->rx_channel = NULL;
+                       }
+               } else {
+                       ep->tx_reinit = 1;
+                       if (ep->tx_channel) {
+                               dma->channel_release(ep->tx_channel);
+                               ep->tx_channel = NULL;
+                       }
+               }
+
+               /* Clobber old pointers to this qh */
+               musb_ep_set_qh(ep, is_in, NULL);
+               qh->hep->hcpriv = NULL;
+
+               switch (qh->type) {
+
+               case USB_ENDPOINT_XFER_CONTROL:
+               case USB_ENDPOINT_XFER_BULK:
+                       /* fifo policy for these lists, except that NAKing
+                        * should rotate a qh to the end (for fairness).
+                        */
+                       if (qh->mux == 1) {
+                               head = qh->ring.prev;
+                               list_del(&qh->ring);
+                               kfree(qh);
+                               qh = first_qh(head);
+                               break;
+                       }
+
+               case USB_ENDPOINT_XFER_ISOC:
+               case USB_ENDPOINT_XFER_INT:
+                       /* this is where periodic bandwidth should be
+                        * de-allocated if it's tracked and allocated;
+                        * and where we'd update the schedule tree...
+                        */
+                       kfree(qh);
+                       qh = NULL;
+                       break;
+               }
+       }
+
+       if (qh != NULL && qh->is_ready) {
+               dev_dbg(musb->controller, "... next ep%d %cX urb %p\n",
+                   hw_ep->epnum, is_in ? 'R' : 'T', next_urb(qh));
+               musb_start_urb(musb, is_in, qh);
+       }
+}
+
+static u16 musb_h_flush_rxfifo(struct musb_hw_ep *hw_ep, u16 csr)
+{
+       /* we don't want fifo to fill itself again;
+        * ignore dma (various models),
+        * leave toggle alone (may not have been saved yet)
+        */
+       csr |= MUSB_RXCSR_FLUSHFIFO | MUSB_RXCSR_RXPKTRDY;
+       csr &= ~(MUSB_RXCSR_H_REQPKT
+               | MUSB_RXCSR_H_AUTOREQ
+               | MUSB_RXCSR_AUTOCLEAR);
+
+       /* write 2x to allow double buffering */
+       musb_writew(hw_ep->regs, MUSB_RXCSR, csr);
+       musb_writew(hw_ep->regs, MUSB_RXCSR, csr);
+
+       /* flush writebuffer */
+       return musb_readw(hw_ep->regs, MUSB_RXCSR);
+}
+
+/*
+ * PIO RX for a packet (or part of it).
+ */
+static bool
+musb_host_packet_rx(struct musb *musb, struct urb *urb, u8 epnum, u8 iso_err)
+{
+       u16                     rx_count;
+       u8                      *buf;
+       u16                     csr;
+       bool                    done = false;
+       u32                     length;
+       int                     do_flush = 0;
+       struct musb_hw_ep       *hw_ep = musb->endpoints + epnum;
+       void __iomem            *epio = hw_ep->regs;
+       struct musb_qh          *qh = hw_ep->in_qh;
+       int                     pipe = urb->pipe;
+       void                    *buffer = urb->transfer_buffer;
+
+       /* musb_ep_select(mbase, epnum); */
+       rx_count = musb_readw(epio, MUSB_RXCOUNT);
+       dev_dbg(musb->controller, "RX%d count %d, buffer %p len %d/%d\n", epnum, rx_count,
+                       urb->transfer_buffer, qh->offset,
+                       urb->transfer_buffer_length);
+
+       /* unload FIFO */
+#ifndef __UBOOT__
+       if (usb_pipeisoc(pipe)) {
+               int                                     status = 0;
+               struct usb_iso_packet_descriptor        *d;
+
+               if (iso_err) {
+                       status = -EILSEQ;
+                       urb->error_count++;
+               }
+
+               d = urb->iso_frame_desc + qh->iso_idx;
+               buf = buffer + d->offset;
+               length = d->length;
+               if (rx_count > length) {
+                       if (status == 0) {
+                               status = -EOVERFLOW;
+                               urb->error_count++;
+                       }
+                       dev_dbg(musb->controller, "** OVERFLOW %d into %d\n", rx_count, length);
+                       do_flush = 1;
+               } else
+                       length = rx_count;
+               urb->actual_length += length;
+               d->actual_length = length;
+
+               d->status = status;
+
+               /* see if we are done */
+               done = (++qh->iso_idx >= urb->number_of_packets);
+       } else {
+#endif
+               /* non-isoch */
+               buf = buffer + qh->offset;
+               length = urb->transfer_buffer_length - qh->offset;
+               if (rx_count > length) {
+                       if (urb->status == -EINPROGRESS)
+                               urb->status = -EOVERFLOW;
+                       dev_dbg(musb->controller, "** OVERFLOW %d into %d\n", rx_count, length);
+                       do_flush = 1;
+               } else
+                       length = rx_count;
+               urb->actual_length += length;
+               qh->offset += length;
+
+               /* see if we are done */
+               done = (urb->actual_length == urb->transfer_buffer_length)
+                       || (rx_count < qh->maxpacket)
+                       || (urb->status != -EINPROGRESS);
+               if (done
+                               && (urb->status == -EINPROGRESS)
+                               && (urb->transfer_flags & URB_SHORT_NOT_OK)
+                               && (urb->actual_length
+                                       < urb->transfer_buffer_length))
+                       urb->status = -EREMOTEIO;
+#ifndef __UBOOT__
+       }
+#endif
+
+       musb_read_fifo(hw_ep, length, buf);
+
+       csr = musb_readw(epio, MUSB_RXCSR);
+       csr |= MUSB_RXCSR_H_WZC_BITS;
+       if (unlikely(do_flush))
+               musb_h_flush_rxfifo(hw_ep, csr);
+       else {
+               /* REVISIT this assumes AUTOCLEAR is never set */
+               csr &= ~(MUSB_RXCSR_RXPKTRDY | MUSB_RXCSR_H_REQPKT);
+               if (!done)
+                       csr |= MUSB_RXCSR_H_REQPKT;
+               musb_writew(epio, MUSB_RXCSR, csr);
+       }
+
+       return done;
+}
+
+/* we don't always need to reinit a given side of an endpoint...
+ * when we do, use tx/rx reinit routine and then construct a new CSR
+ * to address data toggle, NYET, and DMA or PIO.
+ *
+ * it's possible that driver bugs (especially for DMA) or aborting a
+ * transfer might have left the endpoint busier than it should be.
+ * the busy/not-empty tests are basically paranoia.
+ */
+static void
+musb_rx_reinit(struct musb *musb, struct musb_qh *qh, struct musb_hw_ep *ep)
+{
+       u16     csr;
+
+       /* NOTE:  we know the "rx" fifo reinit never triggers for ep0.
+        * That always uses tx_reinit since ep0 repurposes TX register
+        * offsets; the initial SETUP packet is also a kind of OUT.
+        */
+
+       /* if programmed for Tx, put it in RX mode */
+       if (ep->is_shared_fifo) {
+               csr = musb_readw(ep->regs, MUSB_TXCSR);
+               if (csr & MUSB_TXCSR_MODE) {
+                       musb_h_tx_flush_fifo(ep);
+                       csr = musb_readw(ep->regs, MUSB_TXCSR);
+                       musb_writew(ep->regs, MUSB_TXCSR,
+                                   csr | MUSB_TXCSR_FRCDATATOG);
+               }
+
+               /*
+                * Clear the MODE bit (and everything else) to enable Rx.
+                * NOTE: we mustn't clear the DMAMODE bit before DMAENAB.
+                */
+               if (csr & MUSB_TXCSR_DMAMODE)
+                       musb_writew(ep->regs, MUSB_TXCSR, MUSB_TXCSR_DMAMODE);
+               musb_writew(ep->regs, MUSB_TXCSR, 0);
+
+       /* scrub all previous state, clearing toggle */
+       } else {
+               csr = musb_readw(ep->regs, MUSB_RXCSR);
+               if (csr & MUSB_RXCSR_RXPKTRDY)
+                       WARNING("rx%d, packet/%d ready?\n", ep->epnum,
+                               musb_readw(ep->regs, MUSB_RXCOUNT));
+
+               musb_h_flush_rxfifo(ep, MUSB_RXCSR_CLRDATATOG);
+       }
+
+       /* target addr and (for multipoint) hub addr/port */
+       if (musb->is_multipoint) {
+               musb_write_rxfunaddr(ep->target_regs, qh->addr_reg);
+               musb_write_rxhubaddr(ep->target_regs, qh->h_addr_reg);
+               musb_write_rxhubport(ep->target_regs, qh->h_port_reg);
+
+       } else
+               musb_writeb(musb->mregs, MUSB_FADDR, qh->addr_reg);
+
+       /* protocol/endpoint, interval/NAKlimit, i/o size */
+       musb_writeb(ep->regs, MUSB_RXTYPE, qh->type_reg);
+       musb_writeb(ep->regs, MUSB_RXINTERVAL, qh->intv_reg);
+       /* NOTE: bulk combining rewrites high bits of maxpacket */
+       /* Set RXMAXP with the FIFO size of the endpoint
+        * to disable double buffer mode.
+        */
+       if (musb->double_buffer_not_ok)
+               musb_writew(ep->regs, MUSB_RXMAXP, ep->max_packet_sz_rx);
+       else
+               musb_writew(ep->regs, MUSB_RXMAXP,
+                               qh->maxpacket | ((qh->hb_mult - 1) << 11));
+
+       ep->rx_reinit = 0;
+}
+
+static bool musb_tx_dma_program(struct dma_controller *dma,
+               struct musb_hw_ep *hw_ep, struct musb_qh *qh,
+               struct urb *urb, u32 offset, u32 length)
+{
+       struct dma_channel      *channel = hw_ep->tx_channel;
+       void __iomem            *epio = hw_ep->regs;
+       u16                     pkt_size = qh->maxpacket;
+       u16                     csr;
+       u8                      mode;
+
+#ifdef CONFIG_USB_INVENTRA_DMA
+       if (length > channel->max_len)
+               length = channel->max_len;
+
+       csr = musb_readw(epio, MUSB_TXCSR);
+       if (length > pkt_size) {
+               mode = 1;
+               csr |= MUSB_TXCSR_DMAMODE | MUSB_TXCSR_DMAENAB;
+               /* autoset shouldn't be set in high bandwidth */
+               if (qh->hb_mult == 1)
+                       csr |= MUSB_TXCSR_AUTOSET;
+       } else {
+               mode = 0;
+               csr &= ~(MUSB_TXCSR_AUTOSET | MUSB_TXCSR_DMAMODE);
+               csr |= MUSB_TXCSR_DMAENAB; /* against programmer's guide */
+       }
+       channel->desired_mode = mode;
+       musb_writew(epio, MUSB_TXCSR, csr);
+#else
+       if (!is_cppi_enabled() && !tusb_dma_omap())
+               return false;
+
+       channel->actual_len = 0;
+
+       /*
+        * TX uses "RNDIS" mode automatically but needs help
+        * to identify the zero-length-final-packet case.
+        */
+       mode = (urb->transfer_flags & URB_ZERO_PACKET) ? 1 : 0;
+#endif
+
+       qh->segsize = length;
+
+       /*
+        * Ensure the data reaches to main memory before starting
+        * DMA transfer
+        */
+       wmb();
+
+       if (!dma->channel_program(channel, pkt_size, mode,
+                       urb->transfer_dma + offset, length)) {
+               dma->channel_release(channel);
+               hw_ep->tx_channel = NULL;
+
+               csr = musb_readw(epio, MUSB_TXCSR);
+               csr &= ~(MUSB_TXCSR_AUTOSET | MUSB_TXCSR_DMAENAB);
+               musb_writew(epio, MUSB_TXCSR, csr | MUSB_TXCSR_H_WZC_BITS);
+               return false;
+       }
+       return true;
+}
+
+/*
+ * Program an HDRC endpoint as per the given URB
+ * Context: irqs blocked, controller lock held
+ */
+static void musb_ep_program(struct musb *musb, u8 epnum,
+                       struct urb *urb, int is_out,
+                       u8 *buf, u32 offset, u32 len)
+{
+       struct dma_controller   *dma_controller;
+       struct dma_channel      *dma_channel;
+       u8                      dma_ok;
+       void __iomem            *mbase = musb->mregs;
+       struct musb_hw_ep       *hw_ep = musb->endpoints + epnum;
+       void __iomem            *epio = hw_ep->regs;
+       struct musb_qh          *qh = musb_ep_get_qh(hw_ep, !is_out);
+       u16                     packet_sz = qh->maxpacket;
+
+       dev_dbg(musb->controller, "%s hw%d urb %p spd%d dev%d ep%d%s "
+                               "h_addr%02x h_port%02x bytes %d\n",
+                       is_out ? "-->" : "<--",
+                       epnum, urb, urb->dev->speed,
+                       qh->addr_reg, qh->epnum, is_out ? "out" : "in",
+                       qh->h_addr_reg, qh->h_port_reg,
+                       len);
+
+       musb_ep_select(mbase, epnum);
+
+       /* candidate for DMA? */
+       dma_controller = musb->dma_controller;
+       if (is_dma_capable() && epnum && dma_controller) {
+               dma_channel = is_out ? hw_ep->tx_channel : hw_ep->rx_channel;
+               if (!dma_channel) {
+                       dma_channel = dma_controller->channel_alloc(
+                                       dma_controller, hw_ep, is_out);
+                       if (is_out)
+                               hw_ep->tx_channel = dma_channel;
+                       else
+                               hw_ep->rx_channel = dma_channel;
+               }
+       } else
+               dma_channel = NULL;
+
+       /* make sure we clear DMAEnab, autoSet bits from previous run */
+
+       /* OUT/transmit/EP0 or IN/receive? */
+       if (is_out) {
+               u16     csr;
+               u16     int_txe;
+               u16     load_count;
+
+               csr = musb_readw(epio, MUSB_TXCSR);
+
+               /* disable interrupt in case we flush */
+               int_txe = musb_readw(mbase, MUSB_INTRTXE);
+               musb_writew(mbase, MUSB_INTRTXE, int_txe & ~(1 << epnum));
+
+               /* general endpoint setup */
+               if (epnum) {
+                       /* flush all old state, set default */
+                       musb_h_tx_flush_fifo(hw_ep);
+
+                       /*
+                        * We must not clear the DMAMODE bit before or in
+                        * the same cycle with the DMAENAB bit, so we clear
+                        * the latter first...
+                        */
+                       csr &= ~(MUSB_TXCSR_H_NAKTIMEOUT
+                                       | MUSB_TXCSR_AUTOSET
+                                       | MUSB_TXCSR_DMAENAB
+                                       | MUSB_TXCSR_FRCDATATOG
+                                       | MUSB_TXCSR_H_RXSTALL
+                                       | MUSB_TXCSR_H_ERROR
+                                       | MUSB_TXCSR_TXPKTRDY
+                                       );
+                       csr |= MUSB_TXCSR_MODE;
+
+                       if (usb_gettoggle(urb->dev, qh->epnum, 1))
+                               csr |= MUSB_TXCSR_H_WR_DATATOGGLE
+                                       | MUSB_TXCSR_H_DATATOGGLE;
+                       else
+                               csr |= MUSB_TXCSR_CLRDATATOG;
+
+                       musb_writew(epio, MUSB_TXCSR, csr);
+                       /* REVISIT may need to clear FLUSHFIFO ... */
+                       csr &= ~MUSB_TXCSR_DMAMODE;
+                       musb_writew(epio, MUSB_TXCSR, csr);
+                       csr = musb_readw(epio, MUSB_TXCSR);
+               } else {
+                       /* endpoint 0: just flush */
+                       musb_h_ep0_flush_fifo(hw_ep);
+               }
+
+               /* target addr and (for multipoint) hub addr/port */
+               if (musb->is_multipoint) {
+                       musb_write_txfunaddr(mbase, epnum, qh->addr_reg);
+                       musb_write_txhubaddr(mbase, epnum, qh->h_addr_reg);
+                       musb_write_txhubport(mbase, epnum, qh->h_port_reg);
+/* FIXME if !epnum, do the same for RX ... */
+               } else
+                       musb_writeb(mbase, MUSB_FADDR, qh->addr_reg);
+
+               /* protocol/endpoint/interval/NAKlimit */
+               if (epnum) {
+                       musb_writeb(epio, MUSB_TXTYPE, qh->type_reg);
+                       if (musb->double_buffer_not_ok)
+                               musb_writew(epio, MUSB_TXMAXP,
+                                               hw_ep->max_packet_sz_tx);
+                       else if (can_bulk_split(musb, qh->type))
+                               musb_writew(epio, MUSB_TXMAXP, packet_sz
+                                       | ((hw_ep->max_packet_sz_tx /
+                                               packet_sz) - 1) << 11);
+                       else
+                               musb_writew(epio, MUSB_TXMAXP,
+                                               qh->maxpacket |
+                                               ((qh->hb_mult - 1) << 11));
+                       musb_writeb(epio, MUSB_TXINTERVAL, qh->intv_reg);
+               } else {
+                       musb_writeb(epio, MUSB_NAKLIMIT0, qh->intv_reg);
+                       if (musb->is_multipoint)
+                               musb_writeb(epio, MUSB_TYPE0,
+                                               qh->type_reg);
+               }
+
+               if (can_bulk_split(musb, qh->type))
+                       load_count = min((u32) hw_ep->max_packet_sz_tx,
+                                               len);
+               else
+                       load_count = min((u32) packet_sz, len);
+
+               if (dma_channel && musb_tx_dma_program(dma_controller,
+                                       hw_ep, qh, urb, offset, len))
+                       load_count = 0;
+
+               if (load_count) {
+                       /* PIO to load FIFO */
+                       qh->segsize = load_count;
+                       musb_write_fifo(hw_ep, load_count, buf);
+               }
+
+               /* re-enable interrupt */
+               musb_writew(mbase, MUSB_INTRTXE, int_txe);
+
+       /* IN/receive */
+       } else {
+               u16     csr;
+
+               if (hw_ep->rx_reinit) {
+                       musb_rx_reinit(musb, qh, hw_ep);
+
+                       /* init new state: toggle and NYET, maybe DMA later */
+                       if (usb_gettoggle(urb->dev, qh->epnum, 0))
+                               csr = MUSB_RXCSR_H_WR_DATATOGGLE
+                                       | MUSB_RXCSR_H_DATATOGGLE;
+                       else
+                               csr = 0;
+                       if (qh->type == USB_ENDPOINT_XFER_INT)
+                               csr |= MUSB_RXCSR_DISNYET;
+
+               } else {
+                       csr = musb_readw(hw_ep->regs, MUSB_RXCSR);
+
+                       if (csr & (MUSB_RXCSR_RXPKTRDY
+                                       | MUSB_RXCSR_DMAENAB
+                                       | MUSB_RXCSR_H_REQPKT))
+                               ERR("broken !rx_reinit, ep%d csr %04x\n",
+                                               hw_ep->epnum, csr);
+
+                       /* scrub any stale state, leaving toggle alone */
+                       csr &= MUSB_RXCSR_DISNYET;
+               }
+
+               /* kick things off */
+
+               if ((is_cppi_enabled() || tusb_dma_omap()) && dma_channel) {
+                       /* Candidate for DMA */
+                       dma_channel->actual_len = 0L;
+                       qh->segsize = len;
+
+                       /* AUTOREQ is in a DMA register */
+                       musb_writew(hw_ep->regs, MUSB_RXCSR, csr);
+                       csr = musb_readw(hw_ep->regs, MUSB_RXCSR);
+
+                       /*
+                        * Unless caller treats short RX transfers as
+                        * errors, we dare not queue multiple transfers.
+                        */
+                       dma_ok = dma_controller->channel_program(dma_channel,
+                                       packet_sz, !(urb->transfer_flags &
+                                                    URB_SHORT_NOT_OK),
+                                       urb->transfer_dma + offset,
+                                       qh->segsize);
+                       if (!dma_ok) {
+                               dma_controller->channel_release(dma_channel);
+                               hw_ep->rx_channel = dma_channel = NULL;
+                       } else
+                               csr |= MUSB_RXCSR_DMAENAB;
+               }
+
+               csr |= MUSB_RXCSR_H_REQPKT;
+               dev_dbg(musb->controller, "RXCSR%d := %04x\n", epnum, csr);
+               musb_writew(hw_ep->regs, MUSB_RXCSR, csr);
+               csr = musb_readw(hw_ep->regs, MUSB_RXCSR);
+       }
+}
+
+
+/*
+ * Service the default endpoint (ep0) as host.
+ * Return true until it's time to start the status stage.
+ */
+static bool musb_h_ep0_continue(struct musb *musb, u16 len, struct urb *urb)
+{
+       bool                     more = false;
+       u8                      *fifo_dest = NULL;
+       u16                     fifo_count = 0;
+       struct musb_hw_ep       *hw_ep = musb->control_ep;
+       struct musb_qh          *qh = hw_ep->in_qh;
+       struct usb_ctrlrequest  *request;
+
+       switch (musb->ep0_stage) {
+       case MUSB_EP0_IN:
+               fifo_dest = urb->transfer_buffer + urb->actual_length;
+               fifo_count = min_t(size_t, len, urb->transfer_buffer_length -
+                                  urb->actual_length);
+               if (fifo_count < len)
+                       urb->status = -EOVERFLOW;
+
+               musb_read_fifo(hw_ep, fifo_count, fifo_dest);
+
+               urb->actual_length += fifo_count;
+               if (len < qh->maxpacket) {
+                       /* always terminate on short read; it's
+                        * rarely reported as an error.
+                        */
+               } else if (urb->actual_length <
+                               urb->transfer_buffer_length)
+                       more = true;
+               break;
+       case MUSB_EP0_START:
+               request = (struct usb_ctrlrequest *) urb->setup_packet;
+
+               if (!request->wLength) {
+                       dev_dbg(musb->controller, "start no-DATA\n");
+                       break;
+               } else if (request->bRequestType & USB_DIR_IN) {
+                       dev_dbg(musb->controller, "start IN-DATA\n");
+                       musb->ep0_stage = MUSB_EP0_IN;
+                       more = true;
+                       break;
+               } else {
+                       dev_dbg(musb->controller, "start OUT-DATA\n");
+                       musb->ep0_stage = MUSB_EP0_OUT;
+                       more = true;
+               }
+               /* FALLTHROUGH */
+       case MUSB_EP0_OUT:
+               fifo_count = min_t(size_t, qh->maxpacket,
+                                  urb->transfer_buffer_length -
+                                  urb->actual_length);
+               if (fifo_count) {
+                       fifo_dest = (u8 *) (urb->transfer_buffer
+                                       + urb->actual_length);
+                       dev_dbg(musb->controller, "Sending %d byte%s to ep0 fifo %p\n",
+                                       fifo_count,
+                                       (fifo_count == 1) ? "" : "s",
+                                       fifo_dest);
+                       musb_write_fifo(hw_ep, fifo_count, fifo_dest);
+
+                       urb->actual_length += fifo_count;
+                       more = true;
+               }
+               break;
+       default:
+               ERR("bogus ep0 stage %d\n", musb->ep0_stage);
+               break;
+       }
+
+       return more;
+}
+
+/*
+ * Handle default endpoint interrupt as host. Only called in IRQ time
+ * from musb_interrupt().
+ *
+ * called with controller irqlocked
+ */
+irqreturn_t musb_h_ep0_irq(struct musb *musb)
+{
+       struct urb              *urb;
+       u16                     csr, len;
+       int                     status = 0;
+       void __iomem            *mbase = musb->mregs;
+       struct musb_hw_ep       *hw_ep = musb->control_ep;
+       void __iomem            *epio = hw_ep->regs;
+       struct musb_qh          *qh = hw_ep->in_qh;
+       bool                    complete = false;
+       irqreturn_t             retval = IRQ_NONE;
+
+       /* ep0 only has one queue, "in" */
+       urb = next_urb(qh);
+
+       musb_ep_select(mbase, 0);
+       csr = musb_readw(epio, MUSB_CSR0);
+       len = (csr & MUSB_CSR0_RXPKTRDY)
+                       ? musb_readb(epio, MUSB_COUNT0)
+                       : 0;
+
+       dev_dbg(musb->controller, "<== csr0 %04x, qh %p, count %d, urb %p, stage %d\n",
+               csr, qh, len, urb, musb->ep0_stage);
+
+       /* if we just did status stage, we are done */
+       if (MUSB_EP0_STATUS == musb->ep0_stage) {
+               retval = IRQ_HANDLED;
+               complete = true;
+       }
+
+       /* prepare status */
+       if (csr & MUSB_CSR0_H_RXSTALL) {
+               dev_dbg(musb->controller, "STALLING ENDPOINT\n");
+               status = -EPIPE;
+
+       } else if (csr & MUSB_CSR0_H_ERROR) {
+               dev_dbg(musb->controller, "no response, csr0 %04x\n", csr);
+               status = -EPROTO;
+
+       } else if (csr & MUSB_CSR0_H_NAKTIMEOUT) {
+               dev_dbg(musb->controller, "control NAK timeout\n");
+
+               /* NOTE:  this code path would be a good place to PAUSE a
+                * control transfer, if another one is queued, so that
+                * ep0 is more likely to stay busy.  That's already done
+                * for bulk RX transfers.
+                *
+                * if (qh->ring.next != &musb->control), then
+                * we have a candidate... NAKing is *NOT* an error
+                */
+               musb_writew(epio, MUSB_CSR0, 0);
+               retval = IRQ_HANDLED;
+       }
+
+       if (status) {
+               dev_dbg(musb->controller, "aborting\n");
+               retval = IRQ_HANDLED;
+               if (urb)
+                       urb->status = status;
+               complete = true;
+
+               /* use the proper sequence to abort the transfer */
+               if (csr & MUSB_CSR0_H_REQPKT) {
+                       csr &= ~MUSB_CSR0_H_REQPKT;
+                       musb_writew(epio, MUSB_CSR0, csr);
+                       csr &= ~MUSB_CSR0_H_NAKTIMEOUT;
+                       musb_writew(epio, MUSB_CSR0, csr);
+               } else {
+                       musb_h_ep0_flush_fifo(hw_ep);
+               }
+
+               musb_writeb(epio, MUSB_NAKLIMIT0, 0);
+
+               /* clear it */
+               musb_writew(epio, MUSB_CSR0, 0);
+       }
+
+       if (unlikely(!urb)) {
+               /* stop endpoint since we have no place for its data, this
+                * SHOULD NEVER HAPPEN! */
+               ERR("no URB for end 0\n");
+
+               musb_h_ep0_flush_fifo(hw_ep);
+               goto done;
+       }
+
+       if (!complete) {
+               /* call common logic and prepare response */
+               if (musb_h_ep0_continue(musb, len, urb)) {
+                       /* more packets required */
+                       csr = (MUSB_EP0_IN == musb->ep0_stage)
+                               ?  MUSB_CSR0_H_REQPKT : MUSB_CSR0_TXPKTRDY;
+               } else {
+                       /* data transfer complete; perform status phase */
+                       if (usb_pipeout(urb->pipe)
+                                       || !urb->transfer_buffer_length)
+                               csr = MUSB_CSR0_H_STATUSPKT
+                                       | MUSB_CSR0_H_REQPKT;
+                       else
+                               csr = MUSB_CSR0_H_STATUSPKT
+                                       | MUSB_CSR0_TXPKTRDY;
+
+                       /* flag status stage */
+                       musb->ep0_stage = MUSB_EP0_STATUS;
+
+                       dev_dbg(musb->controller, "ep0 STATUS, csr %04x\n", csr);
+
+               }
+               musb_writew(epio, MUSB_CSR0, csr);
+               retval = IRQ_HANDLED;
+       } else
+               musb->ep0_stage = MUSB_EP0_IDLE;
+
+       /* call completion handler if done */
+       if (complete)
+               musb_advance_schedule(musb, urb, hw_ep, 1);
+done:
+       return retval;
+}
+
+
+#ifdef CONFIG_USB_INVENTRA_DMA
+
+/* Host side TX (OUT) using Mentor DMA works as follows:
+       submit_urb ->
+               - if queue was empty, Program Endpoint
+               - ... which starts DMA to fifo in mode 1 or 0
+
+       DMA Isr (transfer complete) -> TxAvail()
+               - Stop DMA (~DmaEnab)   (<--- Alert ... currently happens
+                                       only in musb_cleanup_urb)
+               - TxPktRdy has to be set in mode 0 or for
+                       short packets in mode 1.
+*/
+
+#endif
+
+/* Service a Tx-Available or dma completion irq for the endpoint */
+void musb_host_tx(struct musb *musb, u8 epnum)
+{
+       int                     pipe;
+       bool                    done = false;
+       u16                     tx_csr;
+       size_t                  length = 0;
+       size_t                  offset = 0;
+       struct musb_hw_ep       *hw_ep = musb->endpoints + epnum;
+       void __iomem            *epio = hw_ep->regs;
+       struct musb_qh          *qh = hw_ep->out_qh;
+       struct urb              *urb = next_urb(qh);
+       u32                     status = 0;
+       void __iomem            *mbase = musb->mregs;
+       struct dma_channel      *dma;
+       bool                    transfer_pending = false;
+
+       musb_ep_select(mbase, epnum);
+       tx_csr = musb_readw(epio, MUSB_TXCSR);
+
+       /* with CPPI, DMA sometimes triggers "extra" irqs */
+       if (!urb) {
+               dev_dbg(musb->controller, "extra TX%d ready, csr %04x\n", epnum, tx_csr);
+               return;
+       }
+
+       pipe = urb->pipe;
+       dma = is_dma_capable() ? hw_ep->tx_channel : NULL;
+       dev_dbg(musb->controller, "OUT/TX%d end, csr %04x%s\n", epnum, tx_csr,
+                       dma ? ", dma" : "");
+
+       /* check for errors */
+       if (tx_csr & MUSB_TXCSR_H_RXSTALL) {
+               /* dma was disabled, fifo flushed */
+               dev_dbg(musb->controller, "TX end %d stall\n", epnum);
+
+               /* stall; record URB status */
+               status = -EPIPE;
+
+       } else if (tx_csr & MUSB_TXCSR_H_ERROR) {
+               /* (NON-ISO) dma was disabled, fifo flushed */
+               dev_dbg(musb->controller, "TX 3strikes on ep=%d\n", epnum);
+
+               status = -ETIMEDOUT;
+
+       } else if (tx_csr & MUSB_TXCSR_H_NAKTIMEOUT) {
+               dev_dbg(musb->controller, "TX end=%d device not responding\n", epnum);
+
+               /* NOTE:  this code path would be a good place to PAUSE a
+                * transfer, if there's some other (nonperiodic) tx urb
+                * that could use this fifo.  (dma complicates it...)
+                * That's already done for bulk RX transfers.
+                *
+                * if (bulk && qh->ring.next != &musb->out_bulk), then
+                * we have a candidate... NAKing is *NOT* an error
+                */
+               musb_ep_select(mbase, epnum);
+               musb_writew(epio, MUSB_TXCSR,
+                               MUSB_TXCSR_H_WZC_BITS
+                               | MUSB_TXCSR_TXPKTRDY);
+               return;
+       }
+
+       if (status) {
+               if (dma_channel_status(dma) == MUSB_DMA_STATUS_BUSY) {
+                       dma->status = MUSB_DMA_STATUS_CORE_ABORT;
+                       (void) musb->dma_controller->channel_abort(dma);
+               }
+
+               /* do the proper sequence to abort the transfer in the
+                * usb core; the dma engine should already be stopped.
+                */
+               musb_h_tx_flush_fifo(hw_ep);
+               tx_csr &= ~(MUSB_TXCSR_AUTOSET
+                               | MUSB_TXCSR_DMAENAB
+                               | MUSB_TXCSR_H_ERROR
+                               | MUSB_TXCSR_H_RXSTALL
+                               | MUSB_TXCSR_H_NAKTIMEOUT
+                               );
+
+               musb_ep_select(mbase, epnum);
+               musb_writew(epio, MUSB_TXCSR, tx_csr);
+               /* REVISIT may need to clear FLUSHFIFO ... */
+               musb_writew(epio, MUSB_TXCSR, tx_csr);
+               musb_writeb(epio, MUSB_TXINTERVAL, 0);
+
+               done = true;
+       }
+
+       /* second cppi case */
+       if (dma_channel_status(dma) == MUSB_DMA_STATUS_BUSY) {
+               dev_dbg(musb->controller, "extra TX%d ready, csr %04x\n", epnum, tx_csr);
+               return;
+       }
+
+       if (is_dma_capable() && dma && !status) {
+               /*
+                * DMA has completed.  But if we're using DMA mode 1 (multi
+                * packet DMA), we need a terminal TXPKTRDY interrupt before
+                * we can consider this transfer completed, lest we trash
+                * its last packet when writing the next URB's data.  So we
+                * switch back to mode 0 to get that interrupt; we'll come
+                * back here once it happens.
+                */
+               if (tx_csr & MUSB_TXCSR_DMAMODE) {
+                       /*
+                        * We shouldn't clear DMAMODE with DMAENAB set; so
+                        * clear them in a safe order.  That should be OK
+                        * once TXPKTRDY has been set (and I've never seen
+                        * it being 0 at this moment -- DMA interrupt latency
+                        * is significant) but if it hasn't been then we have
+                        * no choice but to stop being polite and ignore the
+                        * programmer's guide... :-)
+                        *
+                        * Note that we must write TXCSR with TXPKTRDY cleared
+                        * in order not to re-trigger the packet send (this bit
+                        * can't be cleared by CPU), and there's another caveat:
+                        * TXPKTRDY may be set shortly and then cleared in the
+                        * double-buffered FIFO mode, so we do an extra TXCSR
+                        * read for debouncing...
+                        */
+                       tx_csr &= musb_readw(epio, MUSB_TXCSR);
+                       if (tx_csr & MUSB_TXCSR_TXPKTRDY) {
+                               tx_csr &= ~(MUSB_TXCSR_DMAENAB |
+                                           MUSB_TXCSR_TXPKTRDY);
+                               musb_writew(epio, MUSB_TXCSR,
+                                           tx_csr | MUSB_TXCSR_H_WZC_BITS);
+                       }
+                       tx_csr &= ~(MUSB_TXCSR_DMAMODE |
+                                   MUSB_TXCSR_TXPKTRDY);
+                       musb_writew(epio, MUSB_TXCSR,
+                                   tx_csr | MUSB_TXCSR_H_WZC_BITS);
+
+                       /*
+                        * There is no guarantee that we'll get an interrupt
+                        * after clearing DMAMODE as we might have done this
+                        * too late (after TXPKTRDY was cleared by controller).
+                        * Re-read TXCSR as we have spoiled its previous value.
+                        */
+                       tx_csr = musb_readw(epio, MUSB_TXCSR);
+               }
+
+               /*
+                * We may get here from a DMA completion or TXPKTRDY interrupt.
+                * In any case, we must check the FIFO status here and bail out
+                * only if the FIFO still has data -- that should prevent the
+                * "missed" TXPKTRDY interrupts and deal with double-buffered
+                * FIFO mode too...
+                */
+               if (tx_csr & (MUSB_TXCSR_FIFONOTEMPTY | MUSB_TXCSR_TXPKTRDY)) {
+                       dev_dbg(musb->controller, "DMA complete but packet still in FIFO, "
+                           "CSR %04x\n", tx_csr);
+                       return;
+               }
+       }
+
+       if (!status || dma || usb_pipeisoc(pipe)) {
+               if (dma)
+                       length = dma->actual_len;
+               else
+                       length = qh->segsize;
+               qh->offset += length;
+
+               if (usb_pipeisoc(pipe)) {
+#ifndef __UBOOT__
+                       struct usb_iso_packet_descriptor        *d;
+
+                       d = urb->iso_frame_desc + qh->iso_idx;
+                       d->actual_length = length;
+                       d->status = status;
+                       if (++qh->iso_idx >= urb->number_of_packets) {
+                               done = true;
+                       } else {
+                               d++;
+                               offset = d->offset;
+                               length = d->length;
+                       }
+#endif
+               } else if (dma && urb->transfer_buffer_length == qh->offset) {
+                       done = true;
+               } else {
+                       /* see if we need to send more data, or ZLP */
+                       if (qh->segsize < qh->maxpacket)
+                               done = true;
+                       else if (qh->offset == urb->transfer_buffer_length
+                                       && !(urb->transfer_flags
+                                               & URB_ZERO_PACKET))
+                               done = true;
+                       if (!done) {
+                               offset = qh->offset;
+                               length = urb->transfer_buffer_length - offset;
+                               transfer_pending = true;
+                       }
+               }
+       }
+
+       /* urb->status != -EINPROGRESS means request has been faulted,
+        * so we must abort this transfer after cleanup
+        */
+       if (urb->status != -EINPROGRESS) {
+               done = true;
+               if (status == 0)
+                       status = urb->status;
+       }
+
+       if (done) {
+               /* set status */
+               urb->status = status;
+               urb->actual_length = qh->offset;
+               musb_advance_schedule(musb, urb, hw_ep, USB_DIR_OUT);
+               return;
+       } else if ((usb_pipeisoc(pipe) || transfer_pending) && dma) {
+               if (musb_tx_dma_program(musb->dma_controller, hw_ep, qh, urb,
+                               offset, length)) {
+                       if (is_cppi_enabled() || tusb_dma_omap())
+                               musb_h_tx_dma_start(hw_ep);
+                       return;
+               }
+       } else  if (tx_csr & MUSB_TXCSR_DMAENAB) {
+               dev_dbg(musb->controller, "not complete, but DMA enabled?\n");
+               return;
+       }
+
+       /*
+        * PIO: start next packet in this URB.
+        *
+        * REVISIT: some docs say that when hw_ep->tx_double_buffered,
+        * (and presumably, FIFO is not half-full) we should write *two*
+        * packets before updating TXCSR; other docs disagree...
+        */
+       if (length > qh->maxpacket)
+               length = qh->maxpacket;
+       /* Unmap the buffer so that CPU can use it */
+       usb_hcd_unmap_urb_for_dma(musb_to_hcd(musb), urb);
+       musb_write_fifo(hw_ep, length, urb->transfer_buffer + offset);
+       qh->segsize = length;
+
+       musb_ep_select(mbase, epnum);
+       musb_writew(epio, MUSB_TXCSR,
+                       MUSB_TXCSR_H_WZC_BITS | MUSB_TXCSR_TXPKTRDY);
+}
+
+
+#ifdef CONFIG_USB_INVENTRA_DMA
+
+/* Host side RX (IN) using Mentor DMA works as follows:
+       submit_urb ->
+               - if queue was empty, ProgramEndpoint
+               - first IN token is sent out (by setting ReqPkt)
+       LinuxIsr -> RxReady()
+       /\      => first packet is received
+       |       - Set in mode 0 (DmaEnab, ~ReqPkt)
+       |               -> DMA Isr (transfer complete) -> RxReady()
+       |                   - Ack receive (~RxPktRdy), turn off DMA (~DmaEnab)
+       |                   - if urb not complete, send next IN token (ReqPkt)
+       |                          |            else complete urb.
+       |                          |
+       ---------------------------
+ *
+ * Nuances of mode 1:
+ *     For short packets, no ack (+RxPktRdy) is sent automatically
+ *     (even if AutoClear is ON)
+ *     For full packets, ack (~RxPktRdy) and next IN token (+ReqPkt) is sent
+ *     automatically => major problem, as collecting the next packet becomes
+ *     difficult. Hence mode 1 is not used.
+ *
+ * REVISIT
+ *     All we care about at this driver level is that
+ *       (a) all URBs terminate with REQPKT cleared and fifo(s) empty;
+ *       (b) termination conditions are: short RX, or buffer full;
+ *       (c) fault modes include
+ *           - iff URB_SHORT_NOT_OK, short RX status is -EREMOTEIO.
+ *             (and that endpoint's dma queue stops immediately)
+ *           - overflow (full, PLUS more bytes in the terminal packet)
+ *
+ *     So for example, usb-storage sets URB_SHORT_NOT_OK, and would
+ *     thus be a great candidate for using mode 1 ... for all but the
+ *     last packet of one URB's transfer.
+ */
+
+#endif
+
+/* Schedule next QH from musb->in_bulk and move the current qh to
+ * the end; avoids starvation for other endpoints.
+ */
+static void musb_bulk_rx_nak_timeout(struct musb *musb, struct musb_hw_ep *ep)
+{
+       struct dma_channel      *dma;
+       struct urb              *urb;
+       void __iomem            *mbase = musb->mregs;
+       void __iomem            *epio = ep->regs;
+       struct musb_qh          *cur_qh, *next_qh;
+       u16                     rx_csr;
+
+       musb_ep_select(mbase, ep->epnum);
+       dma = is_dma_capable() ? ep->rx_channel : NULL;
+
+       /* clear nak timeout bit */
+       rx_csr = musb_readw(epio, MUSB_RXCSR);
+       rx_csr |= MUSB_RXCSR_H_WZC_BITS;
+       rx_csr &= ~MUSB_RXCSR_DATAERROR;
+       musb_writew(epio, MUSB_RXCSR, rx_csr);
+
+       cur_qh = first_qh(&musb->in_bulk);
+       if (cur_qh) {
+               urb = next_urb(cur_qh);
+               if (dma_channel_status(dma) == MUSB_DMA_STATUS_BUSY) {
+                       dma->status = MUSB_DMA_STATUS_CORE_ABORT;
+                       musb->dma_controller->channel_abort(dma);
+                       urb->actual_length += dma->actual_len;
+                       dma->actual_len = 0L;
+               }
+               musb_save_toggle(cur_qh, 1, urb);
+
+               /* move cur_qh to end of queue */
+               list_move_tail(&cur_qh->ring, &musb->in_bulk);
+
+               /* get the next qh from musb->in_bulk */
+               next_qh = first_qh(&musb->in_bulk);
+
+               /* set rx_reinit and schedule the next qh */
+               ep->rx_reinit = 1;
+               musb_start_urb(musb, 1, next_qh);
+       }
+}
+
+/*
+ * Service an RX interrupt for the given IN endpoint; docs cover bulk, iso,
+ * and high-bandwidth IN transfer cases.
+ */
+void musb_host_rx(struct musb *musb, u8 epnum)
+{
+       struct urb              *urb;
+       struct musb_hw_ep       *hw_ep = musb->endpoints + epnum;
+       void __iomem            *epio = hw_ep->regs;
+       struct musb_qh          *qh = hw_ep->in_qh;
+       size_t                  xfer_len;
+       void __iomem            *mbase = musb->mregs;
+       int                     pipe;
+       u16                     rx_csr, val;
+       bool                    iso_err = false;
+       bool                    done = false;
+       u32                     status;
+       struct dma_channel      *dma;
+
+       musb_ep_select(mbase, epnum);
+
+       urb = next_urb(qh);
+       dma = is_dma_capable() ? hw_ep->rx_channel : NULL;
+       status = 0;
+       xfer_len = 0;
+
+       rx_csr = musb_readw(epio, MUSB_RXCSR);
+       val = rx_csr;
+
+       if (unlikely(!urb)) {
+               /* REVISIT -- THIS SHOULD NEVER HAPPEN ... but, at least
+                * usbtest #11 (unlinks) triggers it regularly, sometimes
+                * with fifo full.  (Only with DMA??)
+                */
+               dev_dbg(musb->controller, "BOGUS RX%d ready, csr %04x, count %d\n", epnum, val,
+                       musb_readw(epio, MUSB_RXCOUNT));
+               musb_h_flush_rxfifo(hw_ep, MUSB_RXCSR_CLRDATATOG);
+               return;
+       }
+
+       pipe = urb->pipe;
+
+       dev_dbg(musb->controller, "<== hw %d rxcsr %04x, urb actual %d (+dma %zu)\n",
+               epnum, rx_csr, urb->actual_length,
+               dma ? dma->actual_len : 0);
+
+       /* check for errors, concurrent stall & unlink is not really
+        * handled yet! */
+       if (rx_csr & MUSB_RXCSR_H_RXSTALL) {
+               dev_dbg(musb->controller, "RX end %d STALL\n", epnum);
+
+               /* stall; record URB status */
+               status = -EPIPE;
+
+       } else if (rx_csr & MUSB_RXCSR_H_ERROR) {
+               dev_dbg(musb->controller, "end %d RX proto error\n", epnum);
+
+               status = -EPROTO;
+               musb_writeb(epio, MUSB_RXINTERVAL, 0);
+
+       } else if (rx_csr & MUSB_RXCSR_DATAERROR) {
+
+               if (USB_ENDPOINT_XFER_ISOC != qh->type) {
+                       dev_dbg(musb->controller, "RX end %d NAK timeout\n", epnum);
+
+                       /* NOTE: NAKing is *NOT* an error, so we want to
+                        * continue.  Except ... if there's a request for
+                        * another QH, use that instead of starving it.
+                        *
+                        * Devices like Ethernet and serial adapters keep
+                        * reads posted at all times, which will starve
+                        * other devices without this logic.
+                        */
+                       if (usb_pipebulk(urb->pipe)
+                                       && qh->mux == 1
+                                       && !list_is_singular(&musb->in_bulk)) {
+                               musb_bulk_rx_nak_timeout(musb, hw_ep);
+                               return;
+                       }
+                       musb_ep_select(mbase, epnum);
+                       rx_csr |= MUSB_RXCSR_H_WZC_BITS;
+                       rx_csr &= ~MUSB_RXCSR_DATAERROR;
+                       musb_writew(epio, MUSB_RXCSR, rx_csr);
+
+                       goto finish;
+               } else {
+                       dev_dbg(musb->controller, "RX end %d ISO data error\n", epnum);
+                       /* packet error reported later */
+                       iso_err = true;
+               }
+       } else if (rx_csr & MUSB_RXCSR_INCOMPRX) {
+               dev_dbg(musb->controller, "end %d high bandwidth incomplete ISO packet RX\n",
+                               epnum);
+               status = -EPROTO;
+       }
+
+       /* faults abort the transfer */
+       if (status) {
+               /* clean up dma and collect transfer count */
+               if (dma_channel_status(dma) == MUSB_DMA_STATUS_BUSY) {
+                       dma->status = MUSB_DMA_STATUS_CORE_ABORT;
+                       (void) musb->dma_controller->channel_abort(dma);
+                       xfer_len = dma->actual_len;
+               }
+               musb_h_flush_rxfifo(hw_ep, MUSB_RXCSR_CLRDATATOG);
+               musb_writeb(epio, MUSB_RXINTERVAL, 0);
+               done = true;
+               goto finish;
+       }
+
+       if (unlikely(dma_channel_status(dma) == MUSB_DMA_STATUS_BUSY)) {
+               /* SHOULD NEVER HAPPEN ... but at least DaVinci has done it */
+               ERR("RX%d dma busy, csr %04x\n", epnum, rx_csr);
+               goto finish;
+       }
+
+       /* thorough shutdown for now ... given more precise fault handling
+        * and better queueing support, we might keep a DMA pipeline going
+        * while processing this irq for earlier completions.
+        */
+
+       /* FIXME this is _way_ too much in-line logic for Mentor DMA */
+
+#ifndef CONFIG_USB_INVENTRA_DMA
+       if (rx_csr & MUSB_RXCSR_H_REQPKT)  {
+               /* REVISIT this happened for a while on some short reads...
+                * the cleanup still needs investigation... looks bad...
+                * and also duplicates dma cleanup code above ... plus,
+                * shouldn't this be the "half full" double buffer case?
+                */
+               if (dma_channel_status(dma) == MUSB_DMA_STATUS_BUSY) {
+                       dma->status = MUSB_DMA_STATUS_CORE_ABORT;
+                       (void) musb->dma_controller->channel_abort(dma);
+                       xfer_len = dma->actual_len;
+                       done = true;
+               }
+
+               dev_dbg(musb->controller, "RXCSR%d %04x, reqpkt, len %zu%s\n", epnum, rx_csr,
+                               xfer_len, dma ? ", dma" : "");
+               rx_csr &= ~MUSB_RXCSR_H_REQPKT;
+
+               musb_ep_select(mbase, epnum);
+               musb_writew(epio, MUSB_RXCSR,
+                               MUSB_RXCSR_H_WZC_BITS | rx_csr);
+       }
+#endif
+       if (dma && (rx_csr & MUSB_RXCSR_DMAENAB)) {
+               xfer_len = dma->actual_len;
+
+               val &= ~(MUSB_RXCSR_DMAENAB
+                       | MUSB_RXCSR_H_AUTOREQ
+                       | MUSB_RXCSR_AUTOCLEAR
+                       | MUSB_RXCSR_RXPKTRDY);
+               musb_writew(hw_ep->regs, MUSB_RXCSR, val);
+
+#ifdef CONFIG_USB_INVENTRA_DMA
+               if (usb_pipeisoc(pipe)) {
+                       struct usb_iso_packet_descriptor *d;
+
+                       d = urb->iso_frame_desc + qh->iso_idx;
+                       d->actual_length = xfer_len;
+
+                       /* even if there was an error, we did the dma
+                        * for iso_frame_desc->length
+                        */
+                       if (d->status != -EILSEQ && d->status != -EOVERFLOW)
+                               d->status = 0;
+
+                       if (++qh->iso_idx >= urb->number_of_packets)
+                               done = true;
+                       else
+                               done = false;
+
+               } else  {
+               /* done if urb buffer is full or short packet is recd */
+               done = (urb->actual_length + xfer_len >=
+                               urb->transfer_buffer_length
+                       || dma->actual_len < qh->maxpacket);
+               }
+
+               /* send IN token for next packet, without AUTOREQ */
+               if (!done) {
+                       val |= MUSB_RXCSR_H_REQPKT;
+                       musb_writew(epio, MUSB_RXCSR,
+                               MUSB_RXCSR_H_WZC_BITS | val);
+               }
+
+               dev_dbg(musb->controller, "ep %d dma %s, rxcsr %04x, rxcount %d\n", epnum,
+                       done ? "off" : "reset",
+                       musb_readw(epio, MUSB_RXCSR),
+                       musb_readw(epio, MUSB_RXCOUNT));
+#else
+               done = true;
+#endif
+       } else if (urb->status == -EINPROGRESS) {
+               /* if no errors, be sure a packet is ready for unloading */
+               if (unlikely(!(rx_csr & MUSB_RXCSR_RXPKTRDY))) {
+                       status = -EPROTO;
+                       ERR("Rx interrupt with no errors or packet!\n");
+
+                       /* FIXME this is another "SHOULD NEVER HAPPEN" */
+
+/* SCRUB (RX) */
+                       /* do the proper sequence to abort the transfer */
+                       musb_ep_select(mbase, epnum);
+                       val &= ~MUSB_RXCSR_H_REQPKT;
+                       musb_writew(epio, MUSB_RXCSR, val);
+                       goto finish;
+               }
+
+               /* we are expecting IN packets */
+#ifdef CONFIG_USB_INVENTRA_DMA
+               if (dma) {
+                       struct dma_controller   *c;
+                       u16                     rx_count;
+                       int                     ret, length;
+                       dma_addr_t              buf;
+
+                       rx_count = musb_readw(epio, MUSB_RXCOUNT);
+
+                       dev_dbg(musb->controller, "RX%d count %d, buffer 0x%x len %d/%d\n",
+                                       epnum, rx_count,
+                                       urb->transfer_dma
+                                               + urb->actual_length,
+                                       qh->offset,
+                                       urb->transfer_buffer_length);
+
+                       c = musb->dma_controller;
+
+                       if (usb_pipeisoc(pipe)) {
+                               int d_status = 0;
+                               struct usb_iso_packet_descriptor *d;
+
+                               d = urb->iso_frame_desc + qh->iso_idx;
+
+                               if (iso_err) {
+                                       d_status = -EILSEQ;
+                                       urb->error_count++;
+                               }
+                               if (rx_count > d->length) {
+                                       if (d_status == 0) {
+                                               d_status = -EOVERFLOW;
+                                               urb->error_count++;
+                                       }
+                                       dev_dbg(musb->controller, "** OVERFLOW %d into %d\n",\
+                                           rx_count, d->length);
+
+                                       length = d->length;
+                               } else
+                                       length = rx_count;
+                               d->status = d_status;
+                               buf = urb->transfer_dma + d->offset;
+                       } else {
+                               length = rx_count;
+                               buf = urb->transfer_dma +
+                                               urb->actual_length;
+                       }
+
+                       dma->desired_mode = 0;
+#ifdef USE_MODE1
+                       /* because of the issue below, mode 1 will
+                        * only rarely behave with correct semantics.
+                        */
+                       if ((urb->transfer_flags &
+                                               URB_SHORT_NOT_OK)
+                               && (urb->transfer_buffer_length -
+                                               urb->actual_length)
+                                       > qh->maxpacket)
+                               dma->desired_mode = 1;
+                       if (rx_count < hw_ep->max_packet_sz_rx) {
+                               length = rx_count;
+                               dma->desired_mode = 0;
+                       } else {
+                               length = urb->transfer_buffer_length;
+                       }
+#endif
+
+/* Disadvantage of using mode 1:
+ *     It's basically usable only for mass storage class; essentially all
+ *     other protocols also terminate transfers on short packets.
+ *
+ * Details:
+ *     An extra IN token is sent at the end of the transfer (due to AUTOREQ)
+ *     If you try to use mode 1 for (transfer_buffer_length - 512), and try
+ *     to use the extra IN token to grab the last packet using mode 0, then
+ *     the problem is that you cannot be sure when the device will send the
+ *     last packet and RxPktRdy set. Sometimes the packet is recd too soon
+ *     such that it gets lost when RxCSR is re-set at the end of the mode 1
+ *     transfer, while sometimes it is recd just a little late so that if you
+ *     try to configure for mode 0 soon after the mode 1 transfer is
+ *     completed, you will find rxcount 0. Okay, so you might think why not
+ *     wait for an interrupt when the pkt is recd. Well, you won't get any!
+ */
+
+                       val = musb_readw(epio, MUSB_RXCSR);
+                       val &= ~MUSB_RXCSR_H_REQPKT;
+
+                       if (dma->desired_mode == 0)
+                               val &= ~MUSB_RXCSR_H_AUTOREQ;
+                       else
+                               val |= MUSB_RXCSR_H_AUTOREQ;
+                       val |= MUSB_RXCSR_DMAENAB;
+
+                       /* autoclear shouldn't be set in high bandwidth */
+                       if (qh->hb_mult == 1)
+                               val |= MUSB_RXCSR_AUTOCLEAR;
+
+                       musb_writew(epio, MUSB_RXCSR,
+                               MUSB_RXCSR_H_WZC_BITS | val);
+
+                       /* REVISIT if when actual_length != 0,
+                        * transfer_buffer_length needs to be
+                        * adjusted first...
+                        */
+                       ret = c->channel_program(
+                               dma, qh->maxpacket,
+                               dma->desired_mode, buf, length);
+
+                       if (!ret) {
+                               c->channel_release(dma);
+                               hw_ep->rx_channel = NULL;
+                               dma = NULL;
+                               val = musb_readw(epio, MUSB_RXCSR);
+                               val &= ~(MUSB_RXCSR_DMAENAB
+                                       | MUSB_RXCSR_H_AUTOREQ
+                                       | MUSB_RXCSR_AUTOCLEAR);
+                               musb_writew(epio, MUSB_RXCSR, val);
+                       }
+               }
+#endif /* Mentor DMA */
+
+               if (!dma) {
+                       /* Unmap the buffer so that CPU can use it */
+                       usb_hcd_unmap_urb_for_dma(musb_to_hcd(musb), urb);
+                       done = musb_host_packet_rx(musb, urb,
+                                       epnum, iso_err);
+                       dev_dbg(musb->controller, "read %spacket\n", done ? "last " : "");
+               }
+       }
+
+finish:
+       urb->actual_length += xfer_len;
+       qh->offset += xfer_len;
+       if (done) {
+               if (urb->status == -EINPROGRESS)
+                       urb->status = status;
+               musb_advance_schedule(musb, urb, hw_ep, USB_DIR_IN);
+       }
+}
+
+/* schedule nodes correspond to peripheral endpoints, like an OHCI QH.
+ * the software schedule associates multiple such nodes with a given
+ * host side hardware endpoint + direction; scheduling may activate
+ * that hardware endpoint.
+ */
+static int musb_schedule(
+       struct musb             *musb,
+       struct musb_qh          *qh,
+       int                     is_in)
+{
+       int                     idle;
+       int                     best_diff;
+       int                     best_end, epnum;
+       struct musb_hw_ep       *hw_ep = NULL;
+       struct list_head        *head = NULL;
+       u8                      toggle;
+       u8                      txtype;
+       struct urb              *urb = next_urb(qh);
+
+       /* use fixed hardware for control and bulk */
+       if (qh->type == USB_ENDPOINT_XFER_CONTROL) {
+               head = &musb->control;
+               hw_ep = musb->control_ep;
+               goto success;
+       }
+
+       /* else, periodic transfers get muxed to other endpoints */
+
+       /*
+        * We know this qh hasn't been scheduled, so all we need to do
+        * is choose which hardware endpoint to put it on ...
+        *
+        * REVISIT what we really want here is a regular schedule tree
+        * like e.g. OHCI uses.
+        */
+       best_diff = 4096;
+       best_end = -1;
+
+       for (epnum = 1, hw_ep = musb->endpoints + 1;
+                       epnum < musb->nr_endpoints;
+                       epnum++, hw_ep++) {
+               int     diff;
+
+               if (musb_ep_get_qh(hw_ep, is_in) != NULL)
+                       continue;
+
+               if (hw_ep == musb->bulk_ep)
+                       continue;
+
+               if (is_in)
+                       diff = hw_ep->max_packet_sz_rx;
+               else
+                       diff = hw_ep->max_packet_sz_tx;
+               diff -= (qh->maxpacket * qh->hb_mult);
+
+               if (diff >= 0 && best_diff > diff) {
+
+                       /*
+                        * Mentor controller has a bug in that if we schedule
+                        * a BULK Tx transfer on an endpoint that had earlier
+                        * handled ISOC then the BULK transfer has to start on
+                        * a zero toggle.  If the BULK transfer starts on a 1
+                        * toggle then this transfer will fail as the mentor
+                        * controller starts the Bulk transfer on a 0 toggle
+                        * irrespective of the programming of the toggle bits
+                        * in the TXCSR register.  Check for this condition
+                        * while allocating the EP for a Tx Bulk transfer.  If
+                        * so skip this EP.
+                        */
+                       hw_ep = musb->endpoints + epnum;
+                       toggle = usb_gettoggle(urb->dev, qh->epnum, !is_in);
+                       txtype = (musb_readb(hw_ep->regs, MUSB_TXTYPE)
+                                       >> 4) & 0x3;
+                       if (!is_in && (qh->type == USB_ENDPOINT_XFER_BULK) &&
+                               toggle && (txtype == USB_ENDPOINT_XFER_ISOC))
+                               continue;
+
+                       best_diff = diff;
+                       best_end = epnum;
+               }
+       }
+       /* use bulk reserved ep1 if no other ep is free */
+       if (best_end < 0 && qh->type == USB_ENDPOINT_XFER_BULK) {
+               hw_ep = musb->bulk_ep;
+               if (is_in)
+                       head = &musb->in_bulk;
+               else
+                       head = &musb->out_bulk;
+
+               /* Enable bulk RX NAK timeout scheme when bulk requests are
+                * multiplexed.  This scheme doen't work in high speed to full
+                * speed scenario as NAK interrupts are not coming from a
+                * full speed device connected to a high speed device.
+                * NAK timeout interval is 8 (128 uframe or 16ms) for HS and
+                * 4 (8 frame or 8ms) for FS device.
+                */
+               if (is_in && qh->dev)
+                       qh->intv_reg =
+                               (USB_SPEED_HIGH == qh->dev->speed) ? 8 : 4;
+               goto success;
+       } else if (best_end < 0) {
+               return -ENOSPC;
+       }
+
+       idle = 1;
+       qh->mux = 0;
+       hw_ep = musb->endpoints + best_end;
+       dev_dbg(musb->controller, "qh %p periodic slot %d\n", qh, best_end);
+success:
+       if (head) {
+               idle = list_empty(head);
+               list_add_tail(&qh->ring, head);
+               qh->mux = 1;
+       }
+       qh->hw_ep = hw_ep;
+       qh->hep->hcpriv = qh;
+       if (idle)
+               musb_start_urb(musb, is_in, qh);
+       return 0;
+}
+
+#ifdef __UBOOT__
+/* check if transaction translator is needed for device */
+static int tt_needed(struct musb *musb, struct usb_device *dev)
+{
+       if ((musb_readb(musb->mregs, MUSB_POWER) & MUSB_POWER_HSMODE) &&
+                       (dev->speed < USB_SPEED_HIGH))
+               return 1;
+       return 0;
+}
+#endif
+
+#ifndef __UBOOT__
+static int musb_urb_enqueue(
+#else
+int musb_urb_enqueue(
+#endif
+       struct usb_hcd                  *hcd,
+       struct urb                      *urb,
+       gfp_t                           mem_flags)
+{
+       unsigned long                   flags;
+       struct musb                     *musb = hcd_to_musb(hcd);
+       struct usb_host_endpoint        *hep = urb->ep;
+       struct musb_qh                  *qh;
+       struct usb_endpoint_descriptor  *epd = &hep->desc;
+       int                             ret;
+       unsigned                        type_reg;
+       unsigned                        interval;
+
+       /* host role must be active */
+       if (!is_host_active(musb) || !musb->is_active)
+               return -ENODEV;
+
+       spin_lock_irqsave(&musb->lock, flags);
+       ret = usb_hcd_link_urb_to_ep(hcd, urb);
+       qh = ret ? NULL : hep->hcpriv;
+       if (qh)
+               urb->hcpriv = qh;
+       spin_unlock_irqrestore(&musb->lock, flags);
+
+       /* DMA mapping was already done, if needed, and this urb is on
+        * hep->urb_list now ... so we're done, unless hep wasn't yet
+        * scheduled onto a live qh.
+        *
+        * REVISIT best to keep hep->hcpriv valid until the endpoint gets
+        * disabled, testing for empty qh->ring and avoiding qh setup costs
+        * except for the first urb queued after a config change.
+        */
+       if (qh || ret)
+               return ret;
+
+       /* Allocate and initialize qh, minimizing the work done each time
+        * hw_ep gets reprogrammed, or with irqs blocked.  Then schedule it.
+        *
+        * REVISIT consider a dedicated qh kmem_cache, so it's harder
+        * for bugs in other kernel code to break this driver...
+        */
+       qh = kzalloc(sizeof *qh, mem_flags);
+       if (!qh) {
+               spin_lock_irqsave(&musb->lock, flags);
+               usb_hcd_unlink_urb_from_ep(hcd, urb);
+               spin_unlock_irqrestore(&musb->lock, flags);
+               return -ENOMEM;
+       }
+
+       qh->hep = hep;
+       qh->dev = urb->dev;
+       INIT_LIST_HEAD(&qh->ring);
+       qh->is_ready = 1;
+
+       qh->maxpacket = usb_endpoint_maxp(epd);
+       qh->type = usb_endpoint_type(epd);
+
+       /* Bits 11 & 12 of wMaxPacketSize encode high bandwidth multiplier.
+        * Some musb cores don't support high bandwidth ISO transfers; and
+        * we don't (yet!) support high bandwidth interrupt transfers.
+        */
+       qh->hb_mult = 1 + ((qh->maxpacket >> 11) & 0x03);
+       if (qh->hb_mult > 1) {
+               int ok = (qh->type == USB_ENDPOINT_XFER_ISOC);
+
+               if (ok)
+                       ok = (usb_pipein(urb->pipe) && musb->hb_iso_rx)
+                               || (usb_pipeout(urb->pipe) && musb->hb_iso_tx);
+               if (!ok) {
+                       ret = -EMSGSIZE;
+                       goto done;
+               }
+               qh->maxpacket &= 0x7ff;
+       }
+
+       qh->epnum = usb_endpoint_num(epd);
+
+       /* NOTE: urb->dev->devnum is wrong during SET_ADDRESS */
+       qh->addr_reg = (u8) usb_pipedevice(urb->pipe);
+
+       /* precompute rxtype/txtype/type0 register */
+       type_reg = (qh->type << 4) | qh->epnum;
+       switch (urb->dev->speed) {
+       case USB_SPEED_LOW:
+               type_reg |= 0xc0;
+               break;
+       case USB_SPEED_FULL:
+               type_reg |= 0x80;
+               break;
+       default:
+               type_reg |= 0x40;
+       }
+       qh->type_reg = type_reg;
+
+       /* Precompute RXINTERVAL/TXINTERVAL register */
+       switch (qh->type) {
+       case USB_ENDPOINT_XFER_INT:
+               /*
+                * Full/low speeds use the  linear encoding,
+                * high speed uses the logarithmic encoding.
+                */
+               if (urb->dev->speed <= USB_SPEED_FULL) {
+                       interval = max_t(u8, epd->bInterval, 1);
+                       break;
+               }
+               /* FALLTHROUGH */
+       case USB_ENDPOINT_XFER_ISOC:
+               /* ISO always uses logarithmic encoding */
+               interval = min_t(u8, epd->bInterval, 16);
+               break;
+       default:
+               /* REVISIT we actually want to use NAK limits, hinting to the
+                * transfer scheduling logic to try some other qh, e.g. try
+                * for 2 msec first:
+                *
+                * interval = (USB_SPEED_HIGH == urb->dev->speed) ? 16 : 2;
+                *
+                * The downside of disabling this is that transfer scheduling
+                * gets VERY unfair for nonperiodic transfers; a misbehaving
+                * peripheral could make that hurt.  That's perfectly normal
+                * for reads from network or serial adapters ... so we have
+                * partial NAKlimit support for bulk RX.
+                *
+                * The upside of disabling it is simpler transfer scheduling.
+                */
+               interval = 0;
+       }
+       qh->intv_reg = interval;
+
+       /* precompute addressing for external hub/tt ports */
+       if (musb->is_multipoint) {
+               struct usb_device       *parent = urb->dev->parent;
+
+#ifndef __UBOOT__
+               if (parent != hcd->self.root_hub) {
+#else
+               if (parent) {
+#endif
+                       qh->h_addr_reg = (u8) parent->devnum;
+
+#ifndef __UBOOT__
+                       /* set up tt info if needed */
+                       if (urb->dev->tt) {
+                               qh->h_port_reg = (u8) urb->dev->ttport;
+                               if (urb->dev->tt->hub)
+                                       qh->h_addr_reg =
+                                               (u8) urb->dev->tt->hub->devnum;
+                               if (urb->dev->tt->multi)
+                                       qh->h_addr_reg |= 0x80;
+                       }
+#else
+                       if (tt_needed(musb, urb->dev)) {
+                               u16 hub_port = find_tt(urb->dev);
+                               qh->h_addr_reg = (u8) (hub_port >> 8);
+                               qh->h_port_reg = (u8) (hub_port & 0xff);
+                       }
+#endif
+               }
+       }
+
+       /* invariant: hep->hcpriv is null OR the qh that's already scheduled.
+        * until we get real dma queues (with an entry for each urb/buffer),
+        * we only have work to do in the former case.
+        */
+       spin_lock_irqsave(&musb->lock, flags);
+       if (hep->hcpriv) {
+               /* some concurrent activity submitted another urb to hep...
+                * odd, rare, error prone, but legal.
+                */
+               kfree(qh);
+               qh = NULL;
+               ret = 0;
+       } else
+               ret = musb_schedule(musb, qh,
+                               epd->bEndpointAddress & USB_ENDPOINT_DIR_MASK);
+
+       if (ret == 0) {
+               urb->hcpriv = qh;
+               /* FIXME set urb->start_frame for iso/intr, it's tested in
+                * musb_start_urb(), but otherwise only konicawc cares ...
+                */
+       }
+       spin_unlock_irqrestore(&musb->lock, flags);
+
+done:
+       if (ret != 0) {
+               spin_lock_irqsave(&musb->lock, flags);
+               usb_hcd_unlink_urb_from_ep(hcd, urb);
+               spin_unlock_irqrestore(&musb->lock, flags);
+               kfree(qh);
+       }
+       return ret;
+}
+
+
+#ifndef __UBOOT__
+/*
+ * abort a transfer that's at the head of a hardware queue.
+ * called with controller locked, irqs blocked
+ * that hardware queue advances to the next transfer, unless prevented
+ */
+static int musb_cleanup_urb(struct urb *urb, struct musb_qh *qh)
+{
+       struct musb_hw_ep       *ep = qh->hw_ep;
+       struct musb             *musb = ep->musb;
+       void __iomem            *epio = ep->regs;
+       unsigned                hw_end = ep->epnum;
+       void __iomem            *regs = ep->musb->mregs;
+       int                     is_in = usb_pipein(urb->pipe);
+       int                     status = 0;
+       u16                     csr;
+
+       musb_ep_select(regs, hw_end);
+
+       if (is_dma_capable()) {
+               struct dma_channel      *dma;
+
+               dma = is_in ? ep->rx_channel : ep->tx_channel;
+               if (dma) {
+                       status = ep->musb->dma_controller->channel_abort(dma);
+                       dev_dbg(musb->controller,
+                               "abort %cX%d DMA for urb %p --> %d\n",
+                               is_in ? 'R' : 'T', ep->epnum,
+                               urb, status);
+                       urb->actual_length += dma->actual_len;
+               }
+       }
+
+       /* turn off DMA requests, discard state, stop polling ... */
+       if (ep->epnum && is_in) {
+               /* giveback saves bulk toggle */
+               csr = musb_h_flush_rxfifo(ep, 0);
+
+               /* REVISIT we still get an irq; should likely clear the
+                * endpoint's irq status here to avoid bogus irqs.
+                * clearing that status is platform-specific...
+                */
+       } else if (ep->epnum) {
+               musb_h_tx_flush_fifo(ep);
+               csr = musb_readw(epio, MUSB_TXCSR);
+               csr &= ~(MUSB_TXCSR_AUTOSET
+                       | MUSB_TXCSR_DMAENAB
+                       | MUSB_TXCSR_H_RXSTALL
+                       | MUSB_TXCSR_H_NAKTIMEOUT
+                       | MUSB_TXCSR_H_ERROR
+                       | MUSB_TXCSR_TXPKTRDY);
+               musb_writew(epio, MUSB_TXCSR, csr);
+               /* REVISIT may need to clear FLUSHFIFO ... */
+               musb_writew(epio, MUSB_TXCSR, csr);
+               /* flush cpu writebuffer */
+               csr = musb_readw(epio, MUSB_TXCSR);
+       } else  {
+               musb_h_ep0_flush_fifo(ep);
+       }
+       if (status == 0)
+               musb_advance_schedule(ep->musb, urb, ep, is_in);
+       return status;
+}
+
+static int musb_urb_dequeue(struct usb_hcd *hcd, struct urb *urb, int status)
+{
+       struct musb             *musb = hcd_to_musb(hcd);
+       struct musb_qh          *qh;
+       unsigned long           flags;
+       int                     is_in  = usb_pipein(urb->pipe);
+       int                     ret;
+
+       dev_dbg(musb->controller, "urb=%p, dev%d ep%d%s\n", urb,
+                       usb_pipedevice(urb->pipe),
+                       usb_pipeendpoint(urb->pipe),
+                       is_in ? "in" : "out");
+
+       spin_lock_irqsave(&musb->lock, flags);
+       ret = usb_hcd_check_unlink_urb(hcd, urb, status);
+       if (ret)
+               goto done;
+
+       qh = urb->hcpriv;
+       if (!qh)
+               goto done;
+
+       /*
+        * Any URB not actively programmed into endpoint hardware can be
+        * immediately given back; that's any URB not at the head of an
+        * endpoint queue, unless someday we get real DMA queues.  And even
+        * if it's at the head, it might not be known to the hardware...
+        *
+        * Otherwise abort current transfer, pending DMA, etc.; urb->status
+        * has already been updated.  This is a synchronous abort; it'd be
+        * OK to hold off until after some IRQ, though.
+        *
+        * NOTE: qh is invalid unless !list_empty(&hep->urb_list)
+        */
+       if (!qh->is_ready
+                       || urb->urb_list.prev != &qh->hep->urb_list
+                       || musb_ep_get_qh(qh->hw_ep, is_in) != qh) {
+               int     ready = qh->is_ready;
+
+               qh->is_ready = 0;
+               musb_giveback(musb, urb, 0);
+               qh->is_ready = ready;
+
+               /* If nothing else (usually musb_giveback) is using it
+                * and its URB list has emptied, recycle this qh.
+                */
+               if (ready && list_empty(&qh->hep->urb_list)) {
+                       qh->hep->hcpriv = NULL;
+                       list_del(&qh->ring);
+                       kfree(qh);
+               }
+       } else
+               ret = musb_cleanup_urb(urb, qh);
+done:
+       spin_unlock_irqrestore(&musb->lock, flags);
+       return ret;
+}
+
+/* disable an endpoint */
+static void
+musb_h_disable(struct usb_hcd *hcd, struct usb_host_endpoint *hep)
+{
+       u8                      is_in = hep->desc.bEndpointAddress & USB_DIR_IN;
+       unsigned long           flags;
+       struct musb             *musb = hcd_to_musb(hcd);
+       struct musb_qh          *qh;
+       struct urb              *urb;
+
+       spin_lock_irqsave(&musb->lock, flags);
+
+       qh = hep->hcpriv;
+       if (qh == NULL)
+               goto exit;
+
+       /* NOTE: qh is invalid unless !list_empty(&hep->urb_list) */
+
+       /* Kick the first URB off the hardware, if needed */
+       qh->is_ready = 0;
+       if (musb_ep_get_qh(qh->hw_ep, is_in) == qh) {
+               urb = next_urb(qh);
+
+               /* make software (then hardware) stop ASAP */
+               if (!urb->unlinked)
+                       urb->status = -ESHUTDOWN;
+
+               /* cleanup */
+               musb_cleanup_urb(urb, qh);
+
+               /* Then nuke all the others ... and advance the
+                * queue on hw_ep (e.g. bulk ring) when we're done.
+                */
+               while (!list_empty(&hep->urb_list)) {
+                       urb = next_urb(qh);
+                       urb->status = -ESHUTDOWN;
+                       musb_advance_schedule(musb, urb, qh->hw_ep, is_in);
+               }
+       } else {
+               /* Just empty the queue; the hardware is busy with
+                * other transfers, and since !qh->is_ready nothing
+                * will activate any of these as it advances.
+                */
+               while (!list_empty(&hep->urb_list))
+                       musb_giveback(musb, next_urb(qh), -ESHUTDOWN);
+
+               hep->hcpriv = NULL;
+               list_del(&qh->ring);
+               kfree(qh);
+       }
+exit:
+       spin_unlock_irqrestore(&musb->lock, flags);
+}
+
+static int musb_h_get_frame_number(struct usb_hcd *hcd)
+{
+       struct musb     *musb = hcd_to_musb(hcd);
+
+       return musb_readw(musb->mregs, MUSB_FRAME);
+}
+
+static int musb_h_start(struct usb_hcd *hcd)
+{
+       struct musb     *musb = hcd_to_musb(hcd);
+
+       /* NOTE: musb_start() is called when the hub driver turns
+        * on port power, or when (OTG) peripheral starts.
+        */
+       hcd->state = HC_STATE_RUNNING;
+       musb->port1_status = 0;
+       return 0;
+}
+
+static void musb_h_stop(struct usb_hcd *hcd)
+{
+       musb_stop(hcd_to_musb(hcd));
+       hcd->state = HC_STATE_HALT;
+}
+
+static int musb_bus_suspend(struct usb_hcd *hcd)
+{
+       struct musb     *musb = hcd_to_musb(hcd);
+       u8              devctl;
+
+       if (!is_host_active(musb))
+               return 0;
+
+       switch (musb->xceiv->state) {
+       case OTG_STATE_A_SUSPEND:
+               return 0;
+       case OTG_STATE_A_WAIT_VRISE:
+               /* ID could be grounded even if there's no device
+                * on the other end of the cable.  NOTE that the
+                * A_WAIT_VRISE timers are messy with MUSB...
+                */
+               devctl = musb_readb(musb->mregs, MUSB_DEVCTL);
+               if ((devctl & MUSB_DEVCTL_VBUS) == MUSB_DEVCTL_VBUS)
+                       musb->xceiv->state = OTG_STATE_A_WAIT_BCON;
+               break;
+       default:
+               break;
+       }
+
+       if (musb->is_active) {
+               WARNING("trying to suspend as %s while active\n",
+                               otg_state_string(musb->xceiv->state));
+               return -EBUSY;
+       } else
+               return 0;
+}
+
+static int musb_bus_resume(struct usb_hcd *hcd)
+{
+       /* resuming child port does the work */
+       return 0;
+}
+
+const struct hc_driver musb_hc_driver = {
+       .description            = "musb-hcd",
+       .product_desc           = "MUSB HDRC host driver",
+       .hcd_priv_size          = sizeof(struct musb),
+       .flags                  = HCD_USB2 | HCD_MEMORY,
+
+       /* not using irq handler or reset hooks from usbcore, since
+        * those must be shared with peripheral code for OTG configs
+        */
+
+       .start                  = musb_h_start,
+       .stop                   = musb_h_stop,
+
+       .get_frame_number       = musb_h_get_frame_number,
+
+       .urb_enqueue            = musb_urb_enqueue,
+       .urb_dequeue            = musb_urb_dequeue,
+       .endpoint_disable       = musb_h_disable,
+
+       .hub_status_data        = musb_hub_status_data,
+       .hub_control            = musb_hub_control,
+       .bus_suspend            = musb_bus_suspend,
+       .bus_resume             = musb_bus_resume,
+       /* .start_port_reset    = NULL, */
+       /* .hub_irq_enable      = NULL, */
+};
+#endif