/*
 * MUSB OTG driver peripheral support
 *
 * Copyright 2005 Mentor Graphics Corporation
 * Copyright (C) 2005-2006 by Texas Instruments
 * Copyright (C) 2006-2007 Nokia Corporation
 * Copyright (C) 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.
 *
 */

#include <linux/kernel.h>
#include <linux/list.h>
#include <linux/timer.h>
#include <linux/module.h>
#include <linux/smp.h>
#include <linux/spinlock.h>
#include <linux/delay.h>
#include <linux/moduleparam.h>
#include <linux/stat.h>
#include <linux/dma-mapping.h>
#include <linux/slab.h>

#include "musb_core.h"


/* MUSB PERIPHERAL status 3-mar-2006:
 *
 * - EP0 seems solid.  It passes both USBCV and usbtest control cases.
 *   Minor glitches:
 *
 *     + remote wakeup to Linux hosts work, but saw USBCV failures;
 *       in one test run (operator error?)
 *     + endpoint halt tests -- in both usbtest and usbcv -- seem
 *       to break when dma is enabled ... is something wrongly
 *       clearing SENDSTALL?
 *
 * - Mass storage behaved ok when last tested.  Network traffic patterns
 *   (with lots of short transfers etc) need retesting; they turn up the
 *   worst cases of the DMA, since short packets are typical but are not
 *   required.
 *
 * - TX/IN
 *     + both pio and dma behave in with network and g_zero tests
 *     + no cppi throughput issues other than no-hw-queueing
 *     + failed with FLAT_REG (DaVinci)
 *     + seems to behave with double buffering, PIO -and- CPPI
 *     + with gadgetfs + AIO, requests got lost?
 *
 * - RX/OUT
 *     + both pio and dma behave in with network and g_zero tests
 *     + dma is slow in typical case (short_not_ok is clear)
 *     + double buffering ok with PIO
 *     + double buffering *FAILS* with CPPI, wrong data bytes sometimes
 *     + request lossage observed with gadgetfs
 *
 * - ISO not tested ... might work, but only weakly isochronous
 *
 * - Gadget driver disabling of softconnect during bind() is ignored; so
 *   drivers can't hold off host requests until userspace is ready.
 *   (Workaround:  they can turn it off later.)
 *
 * - PORTABILITY (assumes PIO works):
 *     + DaVinci, basically works with cppi dma
 *     + OMAP 2430, ditto with mentor dma
 *     + TUSB 6010, platform-specific dma in the works
 */

/* ----------------------------------------------------------------------- */

/*
 * Immediately complete a request.
 *
 * @param request the request to complete
 * @param status the status to complete the request with
 * Context: controller locked, IRQs blocked.
 */
void musb_g_giveback(
	struct musb_ep		*ep,
	struct usb_request	*request,
	int			status)
__releases(ep->musb->lock)
__acquires(ep->musb->lock)
{
	struct musb_request	*req;
	struct musb		*musb;
	int			busy = ep->busy;

	req = to_musb_request(request);

	list_del(&request->list);
	if (req->request.status == -EINPROGRESS)
		req->request.status = status;
	musb = req->musb;

	ep->busy = 1;
	spin_unlock(&musb->lock);
	if (is_dma_capable()) {
		if (req->mapped) {
			dma_unmap_single(musb->controller,
					req->request.dma,
					req->request.length,
					req->tx
						? DMA_TO_DEVICE
						: DMA_FROM_DEVICE);
			req->request.dma = DMA_ADDR_INVALID;
			req->mapped = 0;
		} else if (req->request.dma != DMA_ADDR_INVALID)
			dma_sync_single_for_cpu(musb->controller,
					req->request.dma,
					req->request.length,
					req->tx
						? DMA_TO_DEVICE
						: DMA_FROM_DEVICE);
	}
	if (request->status == 0)
		DBG(5, "%s done request %p,  %d/%d\n",
				ep->end_point.name, request,
				req->request.actual, req->request.length);
	else
		DBG(2, "%s request %p, %d/%d fault %d\n",
				ep->end_point.name, request,
				req->request.actual, req->request.length,
				request->status);
	req->request.complete(&req->ep->end_point, &req->request);
	spin_lock(&musb->lock);
	ep->busy = busy;
}

/* ----------------------------------------------------------------------- */

/*
 * Abort requests queued to an endpoint using the status. Synchronous.
 * caller locked controller and blocked irqs, and selected this ep.
 */
static void nuke(struct musb_ep *ep, const int status)
{
	struct musb_request	*req = NULL;
	void __iomem *epio = ep->musb->endpoints[ep->current_epnum].regs;

	ep->busy = 1;

	if (is_dma_capable() && ep->dma) {
		struct dma_controller	*c = ep->musb->dma_controller;
		int value;

		if (ep->is_in) {
			/*
			 * The programming guide says that we must not clear
			 * the DMAMODE bit before DMAENAB, so we only
			 * clear it in the second write...
			 */
			musb_writew(epio, MUSB_TXCSR,
				    MUSB_TXCSR_DMAMODE | MUSB_TXCSR_FLUSHFIFO);
			musb_writew(epio, MUSB_TXCSR,
					0 | MUSB_TXCSR_FLUSHFIFO);
		} else {
			musb_writew(epio, MUSB_RXCSR,
					0 | MUSB_RXCSR_FLUSHFIFO);
			musb_writew(epio, MUSB_RXCSR,
					0 | MUSB_RXCSR_FLUSHFIFO);
		}

		value = c->channel_abort(ep->dma);
		DBG(value ? 1 : 6, "%s: abort DMA --> %d\n", ep->name, value);
		c->channel_release(ep->dma);
		ep->dma = NULL;
	}

	while (!list_empty(&(ep->req_list))) {
		req = container_of(ep->req_list.next, struct musb_request,
				request.list);
		musb_g_giveback(ep, &req->request, status);
	}
}

/* ----------------------------------------------------------------------- */

/* Data transfers - pure PIO, pure DMA, or mixed mode */

/*
 * This assumes the separate CPPI engine is responding to DMA requests
 * from the usb core ... sequenced a bit differently from mentor dma.
 */

static inline int max_ep_writesize(struct musb *musb, struct musb_ep *ep)
{
	if (can_bulk_split(musb, ep->type))
		return ep->hw_ep->max_packet_sz_tx;
	else
		return ep->packet_sz;
}


#ifdef CONFIG_USB_INVENTRA_DMA

/* Peripheral tx (IN) using Mentor DMA works as follows:
	Only mode 0 is used for transfers <= wPktSize,
	mode 1 is used for larger transfers,

	One of the following happens:
	- Host sends IN token which causes an endpoint interrupt
		-> TxAvail
			-> if DMA is currently busy, exit.
			-> if queue is non-empty, txstate().

	- Request is queued by the gadget driver.
		-> if queue was previously empty, txstate()

	txstate()
		-> start
		  /\	-> setup DMA
		  |     (data is transferred to the FIFO, then sent out when
		  |	IN token(s) are recd from Host.
		  |		-> DMA interrupt on completion
		  |		   calls TxAvail.
		  |		      -> stop DMA, ~DMAENAB,
		  |		      -> set TxPktRdy for last short pkt or zlp
		  |		      -> Complete Request
		  |		      -> Continue next request (call txstate)
		  |___________________________________|

 * Non-Mentor DMA engines can of course work differently, such as by
 * upleveling from irq-per-packet to irq-per-buffer.
 */

#endif

/*
 * An endpoint is transmitting data. This can be called either from
 * the IRQ routine or from ep.queue() to kickstart a request on an
 * endpoint.
 *
 * Context: controller locked, IRQs blocked, endpoint selected
 */
static void txstate(struct musb *musb, struct musb_request *req)
{
	u8			epnum = req->epnum;
	struct musb_ep		*musb_ep;
	void __iomem		*epio = musb->endpoints[epnum].regs;
	struct usb_request	*request;
	u16			fifo_count = 0, csr;
	int			use_dma = 0;

	musb_ep = req->ep;

	/* we shouldn't get here while DMA is active ... but we do ... */
	if (dma_channel_status(musb_ep->dma) == MUSB_DMA_STATUS_BUSY) {
		DBG(4, "dma pending...\n");
		return;
	}

	/* read TXCSR before */
	csr = musb_readw(epio, MUSB_TXCSR);

	request = &req->request;
	fifo_count = min(max_ep_writesize(musb, musb_ep),
			(int)(request->length - request->actual));

	if (csr & MUSB_TXCSR_TXPKTRDY) {
		DBG(5, "%s old packet still ready , txcsr %03x\n",
				musb_ep->end_point.name, csr);
		return;
	}

	if (csr & MUSB_TXCSR_P_SENDSTALL) {
		DBG(5, "%s stalling, txcsr %03x\n",
				musb_ep->end_point.name, csr);
		return;
	}

	DBG(4, "hw_ep%d, maxpacket %d, fifo count %d, txcsr %03x\n",
			epnum, musb_ep->packet_sz, fifo_count,
			csr);

#ifndef	CONFIG_MUSB_PIO_ONLY
	if (is_dma_capable() && musb_ep->dma) {
		struct dma_controller	*c = musb->dma_controller;
		size_t request_size;

		/* setup DMA, then program endpoint CSR */
		request_size = min_t(size_t, request->length - request->actual,
					musb_ep->dma->max_len);

		use_dma = (request->dma != DMA_ADDR_INVALID);

		/* MUSB_TXCSR_P_ISO is still set correctly */

#ifdef CONFIG_USB_INVENTRA_DMA
		{
			if (request_size < musb_ep->packet_sz)
				musb_ep->dma->desired_mode = 0;
			else
				musb_ep->dma->desired_mode = 1;

			use_dma = use_dma && c->channel_program(
					musb_ep->dma, musb_ep->packet_sz,
					musb_ep->dma->desired_mode,
					request->dma + request->actual, request_size);
			if (use_dma) {
				if (musb_ep->dma->desired_mode == 0) {
					/*
					 * We must not clear the DMAMODE bit
					 * before the DMAENAB bit -- and the
					 * latter doesn't always get cleared
					 * before we get here...
					 */
					csr &= ~(MUSB_TXCSR_AUTOSET
						| MUSB_TXCSR_DMAENAB);
					musb_writew(epio, MUSB_TXCSR, csr
						| MUSB_TXCSR_P_WZC_BITS);
					csr &= ~MUSB_TXCSR_DMAMODE;
					csr |= (MUSB_TXCSR_DMAENAB |
							MUSB_TXCSR_MODE);
					/* against programming guide */
				} else {
					csr |= (MUSB_TXCSR_DMAENAB
							| MUSB_TXCSR_DMAMODE
							| MUSB_TXCSR_MODE);
					if (!musb_ep->hb_mult)
						csr |= MUSB_TXCSR_AUTOSET;
				}
				csr &= ~MUSB_TXCSR_P_UNDERRUN;

				musb_writew(epio, MUSB_TXCSR, csr);
			}
		}

#elif defined(CONFIG_USB_TI_CPPI_DMA)
		/* program endpoint CSR first, then setup DMA */
		csr &= ~(MUSB_TXCSR_P_UNDERRUN | MUSB_TXCSR_TXPKTRDY);
		csr |= MUSB_TXCSR_DMAENAB | MUSB_TXCSR_DMAMODE |
		       MUSB_TXCSR_MODE;
		musb_writew(epio, MUSB_TXCSR,
			(MUSB_TXCSR_P_WZC_BITS & ~MUSB_TXCSR_P_UNDERRUN)
				| csr);

		/* ensure writebuffer is empty */
		csr = musb_readw(epio, MUSB_TXCSR);

		/* NOTE host side sets DMAENAB later than this; both are
		 * OK since the transfer dma glue (between CPPI and Mentor
		 * fifos) just tells CPPI it could start.  Data only moves
		 * to the USB TX fifo when both fifos are ready.
		 */

		/* "mode" is irrelevant here; handle terminating ZLPs like
		 * PIO does, since the hardware RNDIS mode seems unreliable
		 * except for the last-packet-is-already-short case.
		 */
		use_dma = use_dma && c->channel_program(
				musb_ep->dma, musb_ep->packet_sz,
				0,
				request->dma + request->actual,
				request_size);
		if (!use_dma) {
			c->channel_release(musb_ep->dma);
			musb_ep->dma = NULL;
			csr &= ~MUSB_TXCSR_DMAENAB;
			musb_writew(epio, MUSB_TXCSR, csr);
			/* invariant: prequest->buf is non-null */
		}
#elif defined(CONFIG_USB_TUSB_OMAP_DMA)
		use_dma = use_dma && c->channel_program(
				musb_ep->dma, musb_ep->packet_sz,
				request->zero,
				request->dma + request->actual,
				request_size);
#endif
	}
#endif

	if (!use_dma) {
		musb_write_fifo(musb_ep->hw_ep, fifo_count,
				(u8 *) (request->buf + request->actual));
		request->actual += fifo_count;
		csr |= MUSB_TXCSR_TXPKTRDY;
		csr &= ~MUSB_TXCSR_P_UNDERRUN;
		musb_writew(epio, MUSB_TXCSR, csr);
	}

	/* host may already have the data when this message shows... */
	DBG(3, "%s TX/IN %s len %d/%d, txcsr %04x, fifo %d/%d\n",
			musb_ep->end_point.name, use_dma ? "dma" : "pio",
			request->actual, request->length,
			musb_readw(epio, MUSB_TXCSR),
			fifo_count,
			musb_readw(epio, MUSB_TXMAXP));
}

/*
 * FIFO state update (e.g. data ready).
 * Called from IRQ,  with controller locked.
 */
void musb_g_tx(struct musb *musb, u8 epnum)
{
	u16			csr;
	struct usb_request	*request;
	u8 __iomem		*mbase = musb->mregs;
	struct musb_ep		*musb_ep = &musb->endpoints[epnum].ep_in;
	void __iomem		*epio = musb->endpoints[epnum].regs;
	struct dma_channel	*dma;

	musb_ep_select(mbase, epnum);
	request = next_request(musb_ep);

	csr = musb_readw(epio, MUSB_TXCSR);
	DBG(4, "<== %s, txcsr %04x\n", musb_ep->end_point.name, csr);

	dma = is_dma_capable() ? musb_ep->dma : NULL;

	/*
	 * REVISIT: for high bandwidth, MUSB_TXCSR_P_INCOMPTX
	 * probably rates reporting as a host error.
	 */
	if (csr & MUSB_TXCSR_P_SENTSTALL) {
		csr |=	MUSB_TXCSR_P_WZC_BITS;
		csr &= ~MUSB_TXCSR_P_SENTSTALL;
		musb_writew(epio, MUSB_TXCSR, csr);
		return;
	}

	if (csr & MUSB_TXCSR_P_UNDERRUN) {
		/* We NAKed, no big deal... little reason to care. */
		csr |=	 MUSB_TXCSR_P_WZC_BITS;
		csr &= ~(MUSB_TXCSR_P_UNDERRUN | MUSB_TXCSR_TXPKTRDY);
		musb_writew(epio, MUSB_TXCSR, csr);
		DBG(20, "underrun on ep%d, req %p\n", epnum, request);
	}

	if (dma_channel_status(dma) == MUSB_DMA_STATUS_BUSY) {
		/*
		 * SHOULD NOT HAPPEN... has with CPPI though, after
		 * changing SENDSTALL (and other cases); harmless?
		 */
		DBG(5, "%s dma still busy?\n", musb_ep->end_point.name);
		return;
	}

	if (request) {
		u8	is_dma = 0;

		if (dma && (csr & MUSB_TXCSR_DMAENAB)) {
			is_dma = 1;
			csr |= MUSB_TXCSR_P_WZC_BITS;
			csr &= ~(MUSB_TXCSR_DMAENAB | MUSB_TXCSR_P_UNDERRUN |
				 MUSB_TXCSR_TXPKTRDY);
			musb_writew(epio, MUSB_TXCSR, csr);
			/* Ensure writebuffer is empty. */
			csr = musb_readw(epio, MUSB_TXCSR);
			request->actual += musb_ep->dma->actual_len;
			DBG(4, "TXCSR%d %04x, DMA off, len %zu, req %p\n",
				epnum, csr, musb_ep->dma->actual_len, request);
		}

		/*
		 * First, maybe a terminating short packet. Some DMA
		 * engines might handle this by themselves.
		 */
		if ((request->zero && request->length
			&& (request->length % musb_ep->packet_sz == 0)
			&& (request->actual == request->length))
#ifdef CONFIG_USB_INVENTRA_DMA
			|| (is_dma && (!dma->desired_mode ||
				(request->actual &
					(musb_ep->packet_sz - 1))))
#endif
		) {
			/*
			 * On DMA completion, FIFO may not be
			 * available yet...
			 */
			if (csr & MUSB_TXCSR_TXPKTRDY)
				return;

			DBG(4, "sending zero pkt\n");
			musb_writew(epio, MUSB_TXCSR, MUSB_TXCSR_MODE
					| MUSB_TXCSR_TXPKTRDY);
			request->zero = 0;
		}

		if (request->actual == request->length) {
			musb_g_giveback(musb_ep, request, 0);
			request = musb_ep->desc ? next_request(musb_ep) : NULL;
			if (!request) {
				DBG(4, "%s idle now\n",
					musb_ep->end_point.name);
				return;
			}
		}

		txstate(musb, to_musb_request(request));
	}
}

/* ------------------------------------------------------------ */

#ifdef CONFIG_USB_INVENTRA_DMA

/* Peripheral rx (OUT) using Mentor DMA works as follows:
	- Only mode 0 is used.

	- Request is queued by the gadget class driver.
		-> if queue was previously empty, rxstate()

	- Host sends OUT token which causes an endpoint interrupt
	  /\      -> RxReady
	  |	      -> if request queued, call rxstate
	  |		/\	-> setup DMA
	  |		|	     -> DMA interrupt on completion
	  |		|		-> RxReady
	  |		|		      -> stop DMA
	  |		|		      -> ack the read
	  |		|		      -> if data recd = max expected
	  |		|				by the request, or host
	  |		|				sent a short packet,
	  |		|				complete the request,
	  |		|				and start the next one.
	  |		|_____________________________________|
	  |					 else just wait for the host
	  |					    to send the next OUT token.
	  |__________________________________________________|

 * Non-Mentor DMA engines can of course work differently.
 */

#endif

/*
 * Context: controller locked, IRQs blocked, endpoint selected
 */
static void rxstate(struct musb *musb, struct musb_request *req)
{
	const u8		epnum = req->epnum;
	struct usb_request	*request = &req->request;
	struct musb_ep		*musb_ep;
	void __iomem		*epio = musb->endpoints[epnum].regs;
	unsigned		fifo_count = 0;
	u16			len;
	u16			csr = musb_readw(epio, MUSB_RXCSR);
	struct musb_hw_ep	*hw_ep = &musb->endpoints[epnum];

	if (hw_ep->is_shared_fifo)
		musb_ep = &hw_ep->ep_in;
	else
		musb_ep = &hw_ep->ep_out;

	len = musb_ep->packet_sz;

	/* We shouldn't get here while DMA is active, but we do... */
	if (dma_channel_status(musb_ep->dma) == MUSB_DMA_STATUS_BUSY) {
		DBG(4, "DMA pending...\n");
		return;
	}

	if (csr & MUSB_RXCSR_P_SENDSTALL) {
		DBG(5, "%s stalling, RXCSR %04x\n",
		    musb_ep->end_point.name, csr);
		return;
	}

	if (is_cppi_enabled() && musb_ep->dma) {
		struct dma_controller	*c = musb->dma_controller;
		struct dma_channel	*channel = musb_ep->dma;

		/* NOTE:  CPPI won't actually stop advancing the DMA
		 * queue after short packet transfers, so this is almost
		 * always going to run as IRQ-per-packet DMA so that
		 * faults will be handled correctly.
		 */
		if (c->channel_program(channel,
				musb_ep->packet_sz,
				!request->short_not_ok,
				request->dma + request->actual,
				request->length - request->actual)) {

			/* make sure that if an rxpkt arrived after the irq,
			 * the cppi engine will be ready to take it as soon
			 * as DMA is enabled
			 */
			csr &= ~(MUSB_RXCSR_AUTOCLEAR
					| MUSB_RXCSR_DMAMODE);
			csr |= MUSB_RXCSR_DMAENAB | MUSB_RXCSR_P_WZC_BITS;
			musb_writew(epio, MUSB_RXCSR, csr);
			return;
		}
	}

	if (csr & MUSB_RXCSR_RXPKTRDY) {
		len = musb_readw(epio, MUSB_RXCOUNT);
		if (request->actual < request->length) {
#ifdef CONFIG_USB_INVENTRA_DMA
			if (is_dma_capable() && musb_ep->dma) {
				struct dma_controller	*c;
				struct dma_channel	*channel;
				int			use_dma = 0;

				c = musb->dma_controller;
				channel = musb_ep->dma;

	/* We use DMA Req mode 0 in rx_csr, and DMA controller operates in
	 * mode 0 only. So we do not get endpoint interrupts due to DMA
	 * completion. We only get interrupts from DMA controller.
	 *
	 * We could operate in DMA mode 1 if we knew the size of the tranfer
	 * in advance. For mass storage class, request->length = what the host
	 * sends, so that'd work.  But for pretty much everything else,
	 * request->length is routinely more than what the host sends. For
	 * most these gadgets, end of is signified either by a short packet,
	 * or filling the last byte of the buffer.  (Sending extra data in
	 * that last pckate should trigger an overflow fault.)  But in mode 1,
	 * we don't get DMA completion interrrupt for short packets.
	 *
	 * Theoretically, we could enable DMAReq irq (MUSB_RXCSR_DMAMODE = 1),
	 * to get endpoint interrupt on every DMA req, but that didn't seem
	 * to work reliably.
	 *
	 * REVISIT an updated g_file_storage can set req->short_not_ok, which
	 * then becomes usable as a runtime "use mode 1" hint...
	 */

				csr |= MUSB_RXCSR_DMAENAB;
				if (!musb_ep->hb_mult &&
					musb_ep->hw_ep->rx_double_buffered)
					csr |= MUSB_RXCSR_AUTOCLEAR;
#ifdef USE_MODE1
				/* csr |= MUSB_RXCSR_DMAMODE; */

				/* this special sequence (enabling and then
				 * disabling MUSB_RXCSR_DMAMODE) is required
				 * to get DMAReq to activate
				 */
				musb_writew(epio, MUSB_RXCSR,
					csr | MUSB_RXCSR_DMAMODE);
#endif
				musb_writew(epio, MUSB_RXCSR, csr);

				if (request->actual < request->length) {
					int transfer_size = 0;
#ifdef USE_MODE1
					transfer_size = min(request->length - request->actual,
							channel->max_len);
#else
					transfer_size = min(request->length - request->actual,
							(unsigned)len);
#endif
					if (transfer_size <= musb_ep->packet_sz)
						musb_ep->dma->desired_mode = 0;
					else
						musb_ep->dma->desired_mode = 1;

					use_dma = c->channel_program(
							channel,
							musb_ep->packet_sz,
							channel->desired_mode,
							request->dma
							+ request->actual,
							transfer_size);
				}

				if (use_dma)
					return;
			}
#endif	/* Mentor's DMA */

			fifo_count = request->length - request->actual;
			DBG(3, "%s OUT/RX pio fifo %d/%d, maxpacket %d\n",
					musb_ep->end_point.name,
					len, fifo_count,
					musb_ep->packet_sz);

			fifo_count = min_t(unsigned, len, fifo_count);

#ifdef	CONFIG_USB_TUSB_OMAP_DMA
			if (tusb_dma_omap() && musb_ep->dma) {
				struct dma_controller *c = musb->dma_controller;
				struct dma_channel *channel = musb_ep->dma;
				u32 dma_addr = request->dma + request->actual;
				int ret;

				ret = c->channel_program(channel,
						musb_ep->packet_sz,
						channel->desired_mode,
						dma_addr,
						fifo_count);
				if (ret)
					return;
			}
#endif

			musb_read_fifo(musb_ep->hw_ep, fifo_count, (u8 *)
					(request->buf + request->actual));
			request->actual += fifo_count;

			/* REVISIT if we left anything in the fifo, flush
			 * it and report -EOVERFLOW
			 */

			/* ack the read! */
			csr |= MUSB_RXCSR_P_WZC_BITS;
			csr &= ~MUSB_RXCSR_RXPKTRDY;
			musb_writew(epio, MUSB_RXCSR, csr);
		}
	}

	/* reach the end or short packet detected */
	if (request->actual == request->length || len < musb_ep->packet_sz)
		musb_g_giveback(musb_ep, request, 0);
}

/*
 * Data ready for a request; called from IRQ
 */
void musb_g_rx(struct musb *musb, u8 epnum)
{
	u16			csr;
	struct usb_request	*request;
	void __iomem		*mbase = musb->mregs;
	struct musb_ep		*musb_ep;
	void __iomem		*epio = musb->endpoints[epnum].regs;
	struct dma_channel	*dma;
	struct musb_hw_ep	*hw_ep = &musb->endpoints[epnum];

	if (hw_ep->is_shared_fifo)
		musb_ep = &hw_ep->ep_in;
	else
		musb_ep = &hw_ep->ep_out;

	musb_ep_select(mbase, epnum);

	request = next_request(musb_ep);
	if (!request)
		return;

	csr = musb_readw(epio, MUSB_RXCSR);
	dma = is_dma_capable() ? musb_ep->dma : NULL;

	DBG(4, "<== %s, rxcsr %04x%s %p\n", musb_ep->end_point.name,
			csr, dma ? " (dma)" : "", request);

	if (csr & MUSB_RXCSR_P_SENTSTALL) {
		csr |= MUSB_RXCSR_P_WZC_BITS;
		csr &= ~MUSB_RXCSR_P_SENTSTALL;
		musb_writew(epio, MUSB_RXCSR, csr);
		return;
	}

	if (csr & MUSB_RXCSR_P_OVERRUN) {
		/* csr |= MUSB_RXCSR_P_WZC_BITS; */
		csr &= ~MUSB_RXCSR_P_OVERRUN;
		musb_writew(epio, MUSB_RXCSR, csr);

		DBG(3, "%s iso overrun on %p\n", musb_ep->name, request);
		if (request->status == -EINPROGRESS)
			request->status = -EOVERFLOW;
	}
	if (csr & MUSB_RXCSR_INCOMPRX) {
		/* REVISIT not necessarily an error */
		DBG(4, "%s, incomprx\n", musb_ep->end_point.name);
	}

	if (dma_channel_status(dma) == MUSB_DMA_STATUS_BUSY) {
		/* "should not happen"; likely RXPKTRDY pending for DMA */
		DBG((csr & MUSB_RXCSR_DMAENAB) ? 4 : 1,
			"%s busy, csr %04x\n",
			musb_ep->end_point.name, csr);
		return;
	}

	if (dma && (csr & MUSB_RXCSR_DMAENAB)) {
		csr &= ~(MUSB_RXCSR_AUTOCLEAR
				| MUSB_RXCSR_DMAENAB
				| MUSB_RXCSR_DMAMODE);
		musb_writew(epio, MUSB_RXCSR,
			MUSB_RXCSR_P_WZC_BITS | csr);

		request->actual += musb_ep->dma->actual_len;

		DBG(4, "RXCSR%d %04x, dma off, %04x, len %zu, req %p\n",
			epnum, csr,
			musb_readw(epio, MUSB_RXCSR),
			musb_ep->dma->actual_len, request);

#if defined(CONFIG_USB_INVENTRA_DMA) || defined(CONFIG_USB_TUSB_OMAP_DMA)
		/* Autoclear doesn't clear RxPktRdy for short packets */
		if ((dma->desired_mode == 0)
				|| (dma->actual_len
					& (musb_ep->packet_sz - 1))) {
			/* ack the read! */
			csr &= ~MUSB_RXCSR_RXPKTRDY;
			musb_writew(epio, MUSB_RXCSR, csr);
		}

		/* incomplete, and not short? wait for next IN packet */
		if ((request->actual < request->length)
				&& (musb_ep->dma->actual_len
					== musb_ep->packet_sz))
			return;
#endif
		musb_g_giveback(musb_ep, request, 0);

		request = next_request(musb_ep);
		if (!request)
			return;
	}

	/* Analyze request */
	rxstate(musb, to_musb_request(request));
}

/* ------------------------------------------------------------ */

static int musb_gadget_enable(struct usb_ep *ep,
			const struct usb_endpoint_descriptor *desc)
{
	unsigned long		flags;
	struct musb_ep		*musb_ep;
	struct musb_hw_ep	*hw_ep;
	void __iomem		*regs;
	struct musb		*musb;
	void __iomem	*mbase;
	u8		epnum;
	u16		csr;
	unsigned	tmp;
	int		status = -EINVAL;

	if (!ep || !desc)
		return -EINVAL;

	musb_ep = to_musb_ep(ep);
	hw_ep = musb_ep->hw_ep;
	regs = hw_ep->regs;
	musb = musb_ep->musb;
	mbase = musb->mregs;
	epnum = musb_ep->current_epnum;

	spin_lock_irqsave(&musb->lock, flags);

	if (musb_ep->desc) {
		status = -EBUSY;
		goto fail;
	}
	musb_ep->type = usb_endpoint_type(desc);

	/* check direction and (later) maxpacket size against endpoint */
	if (usb_endpoint_num(desc) != epnum)
		goto fail;

	/* REVISIT this rules out high bandwidth periodic transfers */
	tmp = le16_to_cpu(desc->wMaxPacketSize);
	if (tmp & ~0x07ff) {
		int ok;

		if (usb_endpoint_dir_in(desc))
			ok = musb->hb_iso_tx;
		else
			ok = musb->hb_iso_rx;

		if (!ok) {
			DBG(4, "%s: not support ISO high bandwidth\n", __func__);
			goto fail;
		}
		musb_ep->hb_mult = (tmp >> 11) & 3;
	} else {
		musb_ep->hb_mult = 0;
	}

	musb_ep->packet_sz = tmp & 0x7ff;
	tmp = musb_ep->packet_sz * (musb_ep->hb_mult + 1);

	/* enable the interrupts for the endpoint, set the endpoint
	 * packet size (or fail), set the mode, clear the fifo
	 */
	musb_ep_select(mbase, epnum);
	if (usb_endpoint_dir_in(desc)) {
		u16 int_txe = musb_readw(mbase, MUSB_INTRTXE);

		if (hw_ep->is_shared_fifo)
			musb_ep->is_in = 1;
		if (!musb_ep->is_in)
			goto fail;

		if (tmp > hw_ep->max_packet_sz_tx) {
			DBG(4, "%s: packet size beyond hw fifo size\n", __func__);
			goto fail;
		}

		int_txe |= (1 << epnum);
		musb_writew(mbase, MUSB_INTRTXE, int_txe);

		/* REVISIT if can_bulk_split(), use by updating "tmp";
		 * likewise high bandwidth periodic tx
		 */
		/* Set TXMAXP with the FIFO size of the endpoint
		 * to disable double buffering mode. Currently, It seems that double
		 * buffering has problem if musb RTL revision number < 2.0.
		 */
		if (musb->hwvers < MUSB_HWVERS_2000)
			musb_writew(regs, MUSB_TXMAXP, hw_ep->max_packet_sz_tx);
		else
			musb_writew(regs, MUSB_TXMAXP, musb_ep->packet_sz | (musb_ep->hb_mult << 11));

		csr = MUSB_TXCSR_MODE | MUSB_TXCSR_CLRDATATOG;
		if (musb_readw(regs, MUSB_TXCSR)
				& MUSB_TXCSR_FIFONOTEMPTY)
			csr |= MUSB_TXCSR_FLUSHFIFO;
		if (musb_ep->type == USB_ENDPOINT_XFER_ISOC)
			csr |= MUSB_TXCSR_P_ISO;

		/* set twice in case of double buffering */
		musb_writew(regs, MUSB_TXCSR, csr);
		/* REVISIT may be inappropriate w/o FIFONOTEMPTY ... */
		musb_writew(regs, MUSB_TXCSR, csr);

	} else {
		u16 int_rxe = musb_readw(mbase, MUSB_INTRRXE);

		if (hw_ep->is_shared_fifo)
			musb_ep->is_in = 0;
		if (musb_ep->is_in)
			goto fail;

		if (tmp > hw_ep->max_packet_sz_rx) {
			DBG(4, "%s: packet size beyond hw fifo size\n", __func__);
			goto fail;
		}

		int_rxe |= (1 << epnum);
		musb_writew(mbase, MUSB_INTRRXE, int_rxe);

		/* REVISIT if can_bulk_combine() use by updating "tmp"
		 * likewise high bandwidth periodic rx
		 */
		/* Set RXMAXP with the FIFO size of the endpoint
		 * to disable double buffering mode.
		 */
		if (musb->hwvers < MUSB_HWVERS_2000)
			musb_writew(regs, MUSB_RXMAXP, hw_ep->max_packet_sz_rx);
		else
			musb_writew(regs, MUSB_RXMAXP, musb_ep->packet_sz | (musb_ep->hb_mult << 11));

		/* force shared fifo to OUT-only mode */
		if (hw_ep->is_shared_fifo) {
			csr = musb_readw(regs, MUSB_TXCSR);
			csr &= ~(MUSB_TXCSR_MODE | MUSB_TXCSR_TXPKTRDY);
			musb_writew(regs, MUSB_TXCSR, csr);
		}

		csr = MUSB_RXCSR_FLUSHFIFO | MUSB_RXCSR_CLRDATATOG;
		if (musb_ep->type == USB_ENDPOINT_XFER_ISOC)
			csr |= MUSB_RXCSR_P_ISO;
		else if (musb_ep->type == USB_ENDPOINT_XFER_INT)
			csr |= MUSB_RXCSR_DISNYET;

		/* set twice in case of double buffering */
		musb_writew(regs, MUSB_RXCSR, csr);
		musb_writew(regs, MUSB_RXCSR, csr);
	}

	/* NOTE:  all the I/O code _should_ work fine without DMA, in case
	 * for some reason you run out of channels here.
	 */
	if (is_dma_capable() && musb->dma_controller) {
		struct dma_controller	*c = musb->dma_controller;

		musb_ep->dma = c->channel_alloc(c, hw_ep,
				(desc->bEndpointAddress & USB_DIR_IN));
	} else
		musb_ep->dma = NULL;

	musb_ep->desc = desc;
	musb_ep->busy = 0;
	musb_ep->wedged = 0;
	status = 0;

	pr_debug("%s periph: enabled %s for %s %s, %smaxpacket %d\n",
			musb_driver_name, musb_ep->end_point.name,
			({ char *s; switch (musb_ep->type) {
			case USB_ENDPOINT_XFER_BULK:	s = "bulk"; break;
			case USB_ENDPOINT_XFER_INT:	s = "int"; break;
			default:			s = "iso"; break;
			}; s; }),
			musb_ep->is_in ? "IN" : "OUT",
			musb_ep->dma ? "dma, " : "",
			musb_ep->packet_sz);

	schedule_work(&musb->irq_work);

fail:
	spin_unlock_irqrestore(&musb->lock, flags);
	return status;
}

/*
 * Disable an endpoint flushing all requests queued.
 */
static int musb_gadget_disable(struct usb_ep *ep)
{
	unsigned long	flags;
	struct musb	*musb;
	u8		epnum;
	struct musb_ep	*musb_ep;
	void __iomem	*epio;
	int		status = 0;

	musb_ep = to_musb_ep(ep);
	musb = musb_ep->musb;
	epnum = musb_ep->current_epnum;
	epio = musb->endpoints[epnum].regs;

	spin_lock_irqsave(&musb->lock, flags);
	musb_ep_select(musb->mregs, epnum);

	/* zero the endpoint sizes */
	if (musb_ep->is_in) {
		u16 int_txe = musb_readw(musb->mregs, MUSB_INTRTXE);
		int_txe &= ~(1 << epnum);
		musb_writew(musb->mregs, MUSB_INTRTXE, int_txe);
		musb_writew(epio, MUSB_TXMAXP, 0);
	} else {
		u16 int_rxe = musb_readw(musb->mregs, MUSB_INTRRXE);
		int_rxe &= ~(1 << epnum);
		musb_writew(musb->mregs, MUSB_INTRRXE, int_rxe);
		musb_writew(epio, MUSB_RXMAXP, 0);
	}

	musb_ep->desc = NULL;

	/* abort all pending DMA and requests */
	nuke(musb_ep, -ESHUTDOWN);

	schedule_work(&musb->irq_work);

	spin_unlock_irqrestore(&(musb->lock), flags);

	DBG(2, "%s\n", musb_ep->end_point.name);

	return status;
}

/*
 * Allocate a request for an endpoint.
 * Reused by ep0 code.
 */
struct usb_request *musb_alloc_request(struct usb_ep *ep, gfp_t gfp_flags)
{
	struct musb_ep		*musb_ep = to_musb_ep(ep);
	struct musb_request	*request = NULL;

	request = kzalloc(sizeof *request, gfp_flags);
	if (request) {
		INIT_LIST_HEAD(&request->request.list);
		request->request.dma = DMA_ADDR_INVALID;
		request->epnum = musb_ep->current_epnum;
		request->ep = musb_ep;
	}

	return &request->request;
}

/*
 * Free a request
 * Reused by ep0 code.
 */
void musb_free_request(struct usb_ep *ep, struct usb_request *req)
{
	kfree(to_musb_request(req));
}

static LIST_HEAD(buffers);

struct free_record {
	struct list_head	list;
	struct device		*dev;
	unsigned		bytes;
	dma_addr_t		dma;
};

/*
 * Context: controller locked, IRQs blocked.
 */
void musb_ep_restart(struct musb *musb, struct musb_request *req)
{
	DBG(3, "<== %s request %p len %u on hw_ep%d\n",
		req->tx ? "TX/IN" : "RX/OUT",
		&req->request, req->request.length, req->epnum);

	musb_ep_select(musb->mregs, req->epnum);
	if (req->tx)
		txstate(musb, req);
	else
		rxstate(musb, req);
}

static int musb_gadget_queue(struct usb_ep *ep, struct usb_request *req,
			gfp_t gfp_flags)
{
	struct musb_ep		*musb_ep;
	struct musb_request	*request;
	struct musb		*musb;
	int			status = 0;
	unsigned long		lockflags;

	if (!ep || !req)
		return -EINVAL;
	if (!req->buf)
		return -ENODATA;

	musb_ep = to_musb_ep(ep);
	musb = musb_ep->musb;

	request = to_musb_request(req);
	request->musb = musb;

	if (request->ep != musb_ep)
		return -EINVAL;

	DBG(4, "<== to %s request=%p\n", ep->name, req);

	/* request is mine now... */
	request->request.actual = 0;
	request->request.status = -EINPROGRESS;
	request->epnum = musb_ep->current_epnum;
	request->tx = musb_ep->is_in;

	if (is_dma_capable() && musb_ep->dma) {
		if (request->request.dma == DMA_ADDR_INVALID) {
			request->request.dma = dma_map_single(
					musb->controller,
					request->request.buf,
					request->request.length,
					request->tx
						? DMA_TO_DEVICE
						: DMA_FROM_DEVICE);
			request->mapped = 1;
		} else {
			dma_sync_single_for_device(musb->controller,
					request->request.dma,
					request->request.length,
					request->tx
						? DMA_TO_DEVICE
						: DMA_FROM_DEVICE);
			request->mapped = 0;
		}
	} else if (!req->buf) {
		return -ENODATA;
	} else
		request->mapped = 0;

	spin_lock_irqsave(&musb->lock, lockflags);

	/* don't queue if the ep is down */
	if (!musb_ep->desc) {
		DBG(4, "req %p queued to %s while ep %s\n",
				req, ep->name, "disabled");
		status = -ESHUTDOWN;
		goto cleanup;
	}

	/* add request to the list */
	list_add_tail(&(request->request.list), &(musb_ep->req_list));

	/* it this is the head of the queue, start i/o ... */
	if (!musb_ep->busy && &request->request.list == musb_ep->req_list.next)
		musb_ep_restart(musb, request);

cleanup:
	spin_unlock_irqrestore(&musb->lock, lockflags);
	return status;
}

static int musb_gadget_dequeue(struct usb_ep *ep, struct usb_request *request)
{
	struct musb_ep		*musb_ep = to_musb_ep(ep);
	struct usb_request	*r;
	unsigned long		flags;
	int			status = 0;
	struct musb		*musb = musb_ep->musb;

	if (!ep || !request || to_musb_request(request)->ep != musb_ep)
		return -EINVAL;

	spin_lock_irqsave(&musb->lock, flags);

	list_for_each_entry(r, &musb_ep->req_list, list) {
		if (r == request)
			break;
	}
	if (r != request) {
		DBG(3, "request %p not queued to %s\n", request, ep->name);
		status = -EINVAL;
		goto done;
	}

	/* if the hardware doesn't have the request, easy ... */
	if (musb_ep->req_list.next != &request->list || musb_ep->busy)
		musb_g_giveback(musb_ep, request, -ECONNRESET);

	/* ... else abort the dma transfer ... */
	else if (is_dma_capable() && musb_ep->dma) {
		struct dma_controller	*c = musb->dma_controller;

		musb_ep_select(musb->mregs, musb_ep->current_epnum);
		if (c->channel_abort)
			status = c->channel_abort(musb_ep->dma);
		else
			status = -EBUSY;
		if (status == 0)
			musb_g_giveback(musb_ep, request, -ECONNRESET);
	} else {
		/* NOTE: by sticking to easily tested hardware/driver states,
		 * we leave counting of in-flight packets imprecise.
		 */
		musb_g_giveback(musb_ep, request, -ECONNRESET);
	}

done:
	spin_unlock_irqrestore(&musb->lock, flags);
	return status;
}

/*
 * Set or clear the halt bit of an endpoint. A halted enpoint won't tx/rx any
 * data but will queue requests.
 *
 * exported to ep0 code
 */
static int musb_gadget_set_halt(struct usb_ep *ep, int value)
{
	struct musb_ep		*musb_ep = to_musb_ep(ep);
	u8			epnum = musb_ep->current_epnum;
	struct musb		*musb = musb_ep->musb;
	void __iomem		*epio = musb->endpoints[epnum].regs;
	void __iomem		*mbase;
	unsigned long		flags;
	u16			csr;
	struct musb_request	*request;
	int			status = 0;

	if (!ep)
		return -EINVAL;
	mbase = musb->mregs;

	spin_lock_irqsave(&musb->lock, flags);

	if ((USB_ENDPOINT_XFER_ISOC == musb_ep->type)) {
		status = -EINVAL;
		goto done;
	}

	musb_ep_select(mbase, epnum);

	request = to_musb_request(next_request(musb_ep));
	if (value) {
		if (request) {
			DBG(3, "request in progress, cannot halt %s\n",
			    ep->name);
			status = -EAGAIN;
			goto done;
		}
		/* Cannot portably stall with non-empty FIFO */
		if (musb_ep->is_in) {
			csr = musb_readw(epio, MUSB_TXCSR);
			if (csr & MUSB_TXCSR_FIFONOTEMPTY) {
				DBG(3, "FIFO busy, cannot halt %s\n", ep->name);
				status = -EAGAIN;
				goto done;
			}
		}
	} else
		musb_ep->wedged = 0;

	/* set/clear the stall and toggle bits */
	DBG(2, "%s: %s stall\n", ep->name, value ? "set" : "clear");
	if (musb_ep->is_in) {
		csr = musb_readw(epio, MUSB_TXCSR);
		csr |= MUSB_TXCSR_P_WZC_BITS
			| MUSB_TXCSR_CLRDATATOG;
		if (value)
			csr |= MUSB_TXCSR_P_SENDSTALL;
		else
			csr &= ~(MUSB_TXCSR_P_SENDSTALL
				| MUSB_TXCSR_P_SENTSTALL);
		csr &= ~MUSB_TXCSR_TXPKTRDY;
		musb_writew(epio, MUSB_TXCSR, csr);
	} else {
		csr = musb_readw(epio, MUSB_RXCSR);
		csr |= MUSB_RXCSR_P_WZC_BITS
			| MUSB_RXCSR_FLUSHFIFO
			| MUSB_RXCSR_CLRDATATOG;
		if (value)
			csr |= MUSB_RXCSR_P_SENDSTALL;
		else
			csr &= ~(MUSB_RXCSR_P_SENDSTALL
				| MUSB_RXCSR_P_SENTSTALL);
		musb_writew(epio, MUSB_RXCSR, csr);
	}

	/* maybe start the first request in the queue */
	if (!musb_ep->busy && !value && request) {
		DBG(3, "restarting the request\n");
		musb_ep_restart(musb, request);
	}

done:
	spin_unlock_irqrestore(&musb->lock, flags);
	return status;
}

/*
 * Sets the halt feature with the clear requests ignored
 */
static int musb_gadget_set_wedge(struct usb_ep *ep)
{
	struct musb_ep		*musb_ep = to_musb_ep(ep);

	if (!ep)
		return -EINVAL;

	musb_ep->wedged = 1;

	return usb_ep_set_halt(ep);
}

static int musb_gadget_fifo_status(struct usb_ep *ep)
{
	struct musb_ep		*musb_ep = to_musb_ep(ep);
	void __iomem		*epio = musb_ep->hw_ep->regs;
	int			retval = -EINVAL;

	if (musb_ep->desc && !musb_ep->is_in) {
		struct musb		*musb = musb_ep->musb;
		int			epnum = musb_ep->current_epnum;
		void __iomem		*mbase = musb->mregs;
		unsigned long		flags;

		spin_lock_irqsave(&musb->lock, flags);

		musb_ep_select(mbase, epnum);
		/* FIXME return zero unless RXPKTRDY is set */
		retval = musb_readw(epio, MUSB_RXCOUNT);

		spin_unlock_irqrestore(&musb->lock, flags);
	}
	return retval;
}

static void musb_gadget_fifo_flush(struct usb_ep *ep)
{
	struct musb_ep	*musb_ep = to_musb_ep(ep);
	struct musb	*musb = musb_ep->musb;
	u8		epnum = musb_ep->current_epnum;
	void __iomem	*epio = musb->endpoints[epnum].regs;
	void __iomem	*mbase;
	unsigned long	flags;
	u16		csr, int_txe;

	mbase = musb->mregs;

	spin_lock_irqsave(&musb->lock, flags);
	musb_ep_select(mbase, (u8) epnum);

	/* disable interrupts */
	int_txe = musb_readw(mbase, MUSB_INTRTXE);
	musb_writew(mbase, MUSB_INTRTXE, int_txe & ~(1 << epnum));

	if (musb_ep->is_in) {
		csr = musb_readw(epio, MUSB_TXCSR);
		if (csr & MUSB_TXCSR_FIFONOTEMPTY) {
			csr |= MUSB_TXCSR_FLUSHFIFO | MUSB_TXCSR_P_WZC_BITS;
			musb_writew(epio, MUSB_TXCSR, csr);
			/* REVISIT may be inappropriate w/o FIFONOTEMPTY ... */
			musb_writew(epio, MUSB_TXCSR, csr);
		}
	} else {
		csr = musb_readw(epio, MUSB_RXCSR);
		csr |= MUSB_RXCSR_FLUSHFIFO | MUSB_RXCSR_P_WZC_BITS;
		musb_writew(epio, MUSB_RXCSR, csr);
		musb_writew(epio, MUSB_RXCSR, csr);
	}

	/* re-enable interrupt */
	musb_writew(mbase, MUSB_INTRTXE, int_txe);
	spin_unlock_irqrestore(&musb->lock, flags);
}

static const struct usb_ep_ops musb_ep_ops = {
	.enable		= musb_gadget_enable,
	.disable	= musb_gadget_disable,
	.alloc_request	= musb_alloc_request,
	.free_request	= musb_free_request,
	.queue		= musb_gadget_queue,
	.dequeue	= musb_gadget_dequeue,
	.set_halt	= musb_gadget_set_halt,
	.set_wedge	= musb_gadget_set_wedge,
	.fifo_status	= musb_gadget_fifo_status,
	.fifo_flush	= musb_gadget_fifo_flush
};

/* ----------------------------------------------------------------------- */

static int musb_gadget_get_frame(struct usb_gadget *gadget)
{
	struct musb	*musb = gadget_to_musb(gadget);

	return (int)musb_readw(musb->mregs, MUSB_FRAME);
}

static int musb_gadget_wakeup(struct usb_gadget *gadget)
{
	struct musb	*musb = gadget_to_musb(gadget);
	void __iomem	*mregs = musb->mregs;
	unsigned long	flags;
	int		status = -EINVAL;
	u8		power, devctl;
	int		retries;

	spin_lock_irqsave(&musb->lock, flags);

	switch (musb->xceiv->state) {
	case OTG_STATE_B_PERIPHERAL:
		/* NOTE:  OTG state machine doesn't include B_SUSPENDED;
		 * that's part of the standard usb 1.1 state machine, and
		 * doesn't affect OTG transitions.
		 */
		if (musb->may_wakeup && musb->is_suspended)
			break;
		goto done;
	case OTG_STATE_B_IDLE:
		/* Start SRP ... OTG not required. */
		devctl = musb_readb(mregs, MUSB_DEVCTL);
		DBG(2, "Sending SRP: devctl: %02x\n", devctl);
		devctl |= MUSB_DEVCTL_SESSION;
		musb_writeb(mregs, MUSB_DEVCTL, devctl);
		devctl = musb_readb(mregs, MUSB_DEVCTL);
		retries = 100;
		while (!(devctl & MUSB_DEVCTL_SESSION)) {
			devctl = musb_readb(mregs, MUSB_DEVCTL);
			if (retries-- < 1)
				break;
		}
		retries = 10000;
		while (devctl & MUSB_DEVCTL_SESSION) {
			devctl = musb_readb(mregs, MUSB_DEVCTL);
			if (retries-- < 1)
				break;
		}

		/* Block idling for at least 1s */
		musb_platform_try_idle(musb,
			jiffies + msecs_to_jiffies(1 * HZ));

		status = 0;
		goto done;
	default:
		DBG(2, "Unhandled wake: %s\n", otg_state_string(musb));
		goto done;
	}

	status = 0;

	power = musb_readb(mregs, MUSB_POWER);
	power |= MUSB_POWER_RESUME;
	musb_writeb(mregs, MUSB_POWER, power);
	DBG(2, "issue wakeup\n");

	/* FIXME do this next chunk in a timer callback, no udelay */
	mdelay(2);

	power = musb_readb(mregs, MUSB_POWER);
	power &= ~MUSB_POWER_RESUME;
	musb_writeb(mregs, MUSB_POWER, power);
done:
	spin_unlock_irqrestore(&musb->lock, flags);
	return status;
}

static int
musb_gadget_set_self_powered(struct usb_gadget *gadget, int is_selfpowered)
{
	struct musb	*musb = gadget_to_musb(gadget);

	musb->is_self_powered = !!is_selfpowered;
	return 0;
}

static void musb_pullup(struct musb *musb, int is_on)
{
	u8 power;

	power = musb_readb(musb->mregs, MUSB_POWER);
	if (is_on)
		power |= MUSB_POWER_SOFTCONN;
	else
		power &= ~MUSB_POWER_SOFTCONN;

	/* FIXME if on, HdrcStart; if off, HdrcStop */

	DBG(3, "gadget %s D+ pullup %s\n",
		musb->gadget_driver->function, is_on ? "on" : "off");
	musb_writeb(musb->mregs, MUSB_POWER, power);
}

#if 0
static int musb_gadget_vbus_session(struct usb_gadget *gadget, int is_active)
{
	DBG(2, "<= %s =>\n", __func__);

	/*
	 * FIXME iff driver's softconnect flag is set (as it is during probe,
	 * though that can clear it), just musb_pullup().
	 */

	return -EINVAL;
}
#endif

static int musb_gadget_vbus_draw(struct usb_gadget *gadget, unsigned mA)
{
	struct musb	*musb = gadget_to_musb(gadget);

	if (!musb->xceiv->set_power)
		return -EOPNOTSUPP;
	return otg_set_power(musb->xceiv, mA);
}

static int musb_gadget_pullup(struct usb_gadget *gadget, int is_on)
{
	struct musb	*musb = gadget_to_musb(gadget);
	unsigned long	flags;

	is_on = !!is_on;

	/* NOTE: this assumes we are sensing vbus; we'd rather
	 * not pullup unless the B-session is active.
	 */
	spin_lock_irqsave(&musb->lock, flags);
	if (is_on != musb->softconnect) {
		musb->softconnect = is_on;
		musb_pullup(musb, is_on);
	}
	spin_unlock_irqrestore(&musb->lock, flags);
	return 0;
}

static const struct usb_gadget_ops musb_gadget_operations = {
	.get_frame		= musb_gadget_get_frame,
	.wakeup			= musb_gadget_wakeup,
	.set_selfpowered	= musb_gadget_set_self_powered,
	/* .vbus_session		= musb_gadget_vbus_session, */
	.vbus_draw		= musb_gadget_vbus_draw,
	.pullup			= musb_gadget_pullup,
};

/* ----------------------------------------------------------------------- */

/* Registration */

/* Only this registration code "knows" the rule (from USB standards)
 * about there being only one external upstream port.  It assumes
 * all peripheral ports are external...
 */
static struct musb *the_gadget;

static void musb_gadget_release(struct device *dev)
{
	/* kref_put(WHAT) */
	dev_dbg(dev, "%s\n", __func__);
}


static void __init
init_peripheral_ep(struct musb *musb, struct musb_ep *ep, u8 epnum, int is_in)
{
	struct musb_hw_ep	*hw_ep = musb->endpoints + epnum;

	memset(ep, 0, sizeof *ep);

	ep->current_epnum = epnum;
	ep->musb = musb;
	ep->hw_ep = hw_ep;
	ep->is_in = is_in;

	INIT_LIST_HEAD(&ep->req_list);

	sprintf(ep->name, "ep%d%s", epnum,
			(!epnum || hw_ep->is_shared_fifo) ? "" : (
				is_in ? "in" : "out"));
	ep->end_point.name = ep->name;
	INIT_LIST_HEAD(&ep->end_point.ep_list);
	if (!epnum) {
		ep->end_point.maxpacket = 64;
		ep->end_point.ops = &musb_g_ep0_ops;
		musb->g.ep0 = &ep->end_point;
	} else {
		if (is_in)
			ep->end_point.maxpacket = hw_ep->max_packet_sz_tx;
		else
			ep->end_point.maxpacket = hw_ep->max_packet_sz_rx;
		ep->end_point.ops = &musb_ep_ops;
		list_add_tail(&ep->end_point.ep_list, &musb->g.ep_list);
	}
}

/*
 * Initialize the endpoints exposed to peripheral drivers, with backlinks
 * to the rest of the driver state.
 */
static inline void __init musb_g_init_endpoints(struct musb *musb)
{
	u8			epnum;
	struct musb_hw_ep	*hw_ep;
	unsigned		count = 0;

	/* intialize endpoint list just once */
	INIT_LIST_HEAD(&(musb->g.ep_list));

	for (epnum = 0, hw_ep = musb->endpoints;
			epnum < musb->nr_endpoints;
			epnum++, hw_ep++) {
		if (hw_ep->is_shared_fifo /* || !epnum */) {
			init_peripheral_ep(musb, &hw_ep->ep_in, epnum, 0);
			count++;
		} else {
			if (hw_ep->max_packet_sz_tx) {
				init_peripheral_ep(musb, &hw_ep->ep_in,
							epnum, 1);
				count++;
			}
			if (hw_ep->max_packet_sz_rx) {
				init_peripheral_ep(musb, &hw_ep->ep_out,
							epnum, 0);
				count++;
			}
		}
	}
}

/* called once during driver setup to initialize and link into
 * the driver model; memory is zeroed.
 */
int __init musb_gadget_setup(struct musb *musb)
{
	int status;

	/* REVISIT minor race:  if (erroneously) setting up two
	 * musb peripherals at the same time, only the bus lock
	 * is probably held.
	 */
	if (the_gadget)
		return -EBUSY;
	the_gadget = musb;

	musb->g.ops = &musb_gadget_operations;
	musb->g.is_dualspeed = 1;
	musb->g.speed = USB_SPEED_UNKNOWN;

	/* this "gadget" abstracts/virtualizes the controller */
	dev_set_name(&musb->g.dev, "gadget");
	musb->g.dev.parent = musb->controller;
	musb->g.dev.dma_mask = musb->controller->dma_mask;
	musb->g.dev.release = musb_gadget_release;
	musb->g.name = musb_driver_name;

	if (is_otg_enabled(musb))
		musb->g.is_otg = 1;

	musb_g_init_endpoints(musb);

	musb->is_active = 0;
	musb_platform_try_idle(musb, 0);

	status = device_register(&musb->g.dev);
	if (status != 0)
		the_gadget = NULL;
	return status;
}

void musb_gadget_cleanup(struct musb *musb)
{
	if (musb != the_gadget)
		return;

	device_unregister(&musb->g.dev);
	the_gadget = NULL;
}

/*
 * Register the gadget driver. Used by gadget drivers when
 * registering themselves with the controller.
 *
 * -EINVAL something went wrong (not driver)
 * -EBUSY another gadget is already using the controller
 * -ENOMEM no memeory to perform the operation
 *
 * @param driver the gadget driver
 * @param bind the driver's bind function
 * @return <0 if error, 0 if everything is fine
 */
int usb_gadget_probe_driver(struct usb_gadget_driver *driver,
		int (*bind)(struct usb_gadget *))
{
	int retval;
	unsigned long flags;
	struct musb *musb = the_gadget;

	if (!driver
			|| driver->speed != USB_SPEED_HIGH
			|| !bind || !driver->setup)
		return -EINVAL;

	/* driver must be initialized to support peripheral mode */
	if (!musb) {
		DBG(1, "%s, no dev??\n", __func__);
		return -ENODEV;
	}

	DBG(3, "registering driver %s\n", driver->function);
	spin_lock_irqsave(&musb->lock, flags);

	if (musb->gadget_driver) {
		DBG(1, "%s is already bound to %s\n",
				musb_driver_name,
				musb->gadget_driver->driver.name);
		retval = -EBUSY;
	} else {
		musb->gadget_driver = driver;
		musb->g.dev.driver = &driver->driver;
		driver->driver.bus = NULL;
		musb->softconnect = 1;
		retval = 0;
	}

	spin_unlock_irqrestore(&musb->lock, flags);

	if (retval == 0) {
		retval = bind(&musb->g);
		if (retval != 0) {
			DBG(3, "bind to driver %s failed --> %d\n",
					driver->driver.name, retval);
			musb->gadget_driver = NULL;
			musb->g.dev.driver = NULL;
		}

		spin_lock_irqsave(&musb->lock, flags);

		otg_set_peripheral(musb->xceiv, &musb->g);
		musb->xceiv->state = OTG_STATE_B_IDLE;
		musb->is_active = 1;

		/* FIXME this ignores the softconnect flag.  Drivers are
		 * allowed hold the peripheral inactive until for example
		 * userspace hooks up printer hardware or DSP codecs, so
		 * hosts only see fully functional devices.
		 */

		if (!is_otg_enabled(musb))
			musb_start(musb);

		otg_set_peripheral(musb->xceiv, &musb->g);

		spin_unlock_irqrestore(&musb->lock, flags);

		if (is_otg_enabled(musb)) {
			DBG(3, "OTG startup...\n");

			/* REVISIT:  funcall to other code, which also
			 * handles power budgeting ... this way also
			 * ensures HdrcStart is indirectly called.
			 */
			retval = usb_add_hcd(musb_to_hcd(musb), -1, 0);
			if (retval < 0) {
				DBG(1, "add_hcd failed, %d\n", retval);
				spin_lock_irqsave(&musb->lock, flags);
				otg_set_peripheral(musb->xceiv, NULL);
				musb->gadget_driver = NULL;
				musb->g.dev.driver = NULL;
				spin_unlock_irqrestore(&musb->lock, flags);
			}
		}
	}

	return retval;
}
EXPORT_SYMBOL(usb_gadget_probe_driver);

static void stop_activity(struct musb *musb, struct usb_gadget_driver *driver)
{
	int			i;
	struct musb_hw_ep	*hw_ep;

	/* don't disconnect if it's not connected */
	if (musb->g.speed == USB_SPEED_UNKNOWN)
		driver = NULL;
	else
		musb->g.speed = USB_SPEED_UNKNOWN;

	/* deactivate the hardware */
	if (musb->softconnect) {
		musb->softconnect = 0;
		musb_pullup(musb, 0);
	}
	musb_stop(musb);

	/* killing any outstanding requests will quiesce the driver;
	 * then report disconnect
	 */
	if (driver) {
		for (i = 0, hw_ep = musb->endpoints;
				i < musb->nr_endpoints;
				i++, hw_ep++) {
			musb_ep_select(musb->mregs, i);
			if (hw_ep->is_shared_fifo /* || !epnum */) {
				nuke(&hw_ep->ep_in, -ESHUTDOWN);
			} else {
				if (hw_ep->max_packet_sz_tx)
					nuke(&hw_ep->ep_in, -ESHUTDOWN);
				if (hw_ep->max_packet_sz_rx)
					nuke(&hw_ep->ep_out, -ESHUTDOWN);
			}
		}

		spin_unlock(&musb->lock);
		driver->disconnect(&musb->g);
		spin_lock(&musb->lock);
	}
}

/*
 * Unregister the gadget driver. Used by gadget drivers when
 * unregistering themselves from the controller.
 *
 * @param driver the gadget driver to unregister
 */
int usb_gadget_unregister_driver(struct usb_gadget_driver *driver)
{
	unsigned long	flags;
	int		retval = 0;
	struct musb	*musb = the_gadget;

	if (!driver || !driver->unbind || !musb)
		return -EINVAL;

	/* REVISIT always use otg_set_peripheral() here too;
	 * this needs to shut down the OTG engine.
	 */

	spin_lock_irqsave(&musb->lock, flags);

#ifdef	CONFIG_USB_MUSB_OTG
	musb_hnp_stop(musb);
#endif

	if (musb->gadget_driver == driver) {

		(void) musb_gadget_vbus_draw(&musb->g, 0);

		musb->xceiv->state = OTG_STATE_UNDEFINED;
		stop_activity(musb, driver);
		otg_set_peripheral(musb->xceiv, NULL);

		DBG(3, "unregistering driver %s\n", driver->function);
		spin_unlock_irqrestore(&musb->lock, flags);
		driver->unbind(&musb->g);
		spin_lock_irqsave(&musb->lock, flags);

		musb->gadget_driver = NULL;
		musb->g.dev.driver = NULL;

		musb->is_active = 0;
		musb_platform_try_idle(musb, 0);
	} else
		retval = -EINVAL;
	spin_unlock_irqrestore(&musb->lock, flags);

	if (is_otg_enabled(musb) && retval == 0) {
		usb_remove_hcd(musb_to_hcd(musb));
		/* FIXME we need to be able to register another
		 * gadget driver here and have everything work;
		 * that currently misbehaves.
		 */
	}

	return retval;
}
EXPORT_SYMBOL(usb_gadget_unregister_driver);


/* ----------------------------------------------------------------------- */

/* lifecycle operations called through plat_uds.c */

void musb_g_resume(struct musb *musb)
{
	musb->is_suspended = 0;
	switch (musb->xceiv->state) {
	case OTG_STATE_B_IDLE:
		break;
	case OTG_STATE_B_WAIT_ACON:
	case OTG_STATE_B_PERIPHERAL:
		musb->is_active = 1;
		if (musb->gadget_driver && musb->gadget_driver->resume) {
			spin_unlock(&musb->lock);
			musb->gadget_driver->resume(&musb->g);
			spin_lock(&musb->lock);
		}
		break;
	default:
		WARNING("unhandled RESUME transition (%s)\n",
				otg_state_string(musb));
	}
}

/* called when SOF packets stop for 3+ msec */
void musb_g_suspend(struct musb *musb)
{
	u8	devctl;

	devctl = musb_readb(musb->mregs, MUSB_DEVCTL);
	DBG(3, "devctl %02x\n", devctl);

	switch (musb->xceiv->state) {
	case OTG_STATE_B_IDLE:
		if ((devctl & MUSB_DEVCTL_VBUS) == MUSB_DEVCTL_VBUS)
			musb->xceiv->state = OTG_STATE_B_PERIPHERAL;
		break;
	case OTG_STATE_B_PERIPHERAL:
		musb->is_suspended = 1;
		if (musb->gadget_driver && musb->gadget_driver->suspend) {
			spin_unlock(&musb->lock);
			musb->gadget_driver->suspend(&musb->g);
			spin_lock(&musb->lock);
		}
		break;
	default:
		/* REVISIT if B_HOST, clear DEVCTL.HOSTREQ;
		 * A_PERIPHERAL may need care too
		 */
		WARNING("unhandled SUSPEND transition (%s)\n",
				otg_state_string(musb));
	}
}

/* Called during SRP */
void musb_g_wakeup(struct musb *musb)
{
	musb_gadget_wakeup(&musb->g);
}

/* called when VBUS drops below session threshold, and in other cases */
void musb_g_disconnect(struct musb *musb)
{
	void __iomem	*mregs = musb->mregs;
	u8	devctl = musb_readb(mregs, MUSB_DEVCTL);

	DBG(3, "devctl %02x\n", devctl);

	/* clear HR */
	musb_writeb(mregs, MUSB_DEVCTL, devctl & MUSB_DEVCTL_SESSION);

	/* don't draw vbus until new b-default session */
	(void) musb_gadget_vbus_draw(&musb->g, 0);

	musb->g.speed = USB_SPEED_UNKNOWN;
	if (musb->gadget_driver && musb->gadget_driver->disconnect) {
		spin_unlock(&musb->lock);
		musb->gadget_driver->disconnect(&musb->g);
		spin_lock(&musb->lock);
	}

	switch (musb->xceiv->state) {
	default:
#ifdef	CONFIG_USB_MUSB_OTG
		DBG(2, "Unhandled disconnect %s, setting a_idle\n",
			otg_state_string(musb));
		musb->xceiv->state = OTG_STATE_A_IDLE;
		MUSB_HST_MODE(musb);
		break;
	case OTG_STATE_A_PERIPHERAL:
		musb->xceiv->state = OTG_STATE_A_WAIT_BCON;
		MUSB_HST_MODE(musb);
		break;
	case OTG_STATE_B_WAIT_ACON:
	case OTG_STATE_B_HOST:
#endif
	case OTG_STATE_B_PERIPHERAL:
	case OTG_STATE_B_IDLE:
		musb->xceiv->state = OTG_STATE_B_IDLE;
		break;
	case OTG_STATE_B_SRP_INIT:
		break;
	}

	musb->is_active = 0;
}

void musb_g_reset(struct musb *musb)
__releases(musb->lock)
__acquires(musb->lock)
{
	void __iomem	*mbase = musb->mregs;
	u8		devctl = musb_readb(mbase, MUSB_DEVCTL);
	u8		power;

	DBG(3, "<== %s addr=%x driver '%s'\n",
			(devctl & MUSB_DEVCTL_BDEVICE)
				? "B-Device" : "A-Device",
			musb_readb(mbase, MUSB_FADDR),
			musb->gadget_driver
				? musb->gadget_driver->driver.name
				: NULL
			);

	/* report disconnect, if we didn't already (flushing EP state) */
	if (musb->g.speed != USB_SPEED_UNKNOWN)
		musb_g_disconnect(musb);

	/* clear HR */
	else if (devctl & MUSB_DEVCTL_HR)
		musb_writeb(mbase, MUSB_DEVCTL, MUSB_DEVCTL_SESSION);


	/* what speed did we negotiate? */
	power = musb_readb(mbase, MUSB_POWER);
	musb->g.speed = (power & MUSB_POWER_HSMODE)
			? USB_SPEED_HIGH : USB_SPEED_FULL;

	/* start in USB_STATE_DEFAULT */
	musb->is_active = 1;
	musb->is_suspended = 0;
	MUSB_DEV_MODE(musb);
	musb->address = 0;
	musb->ep0_state = MUSB_EP0_STAGE_SETUP;

	musb->may_wakeup = 0;
	musb->g.b_hnp_enable = 0;
	musb->g.a_alt_hnp_support = 0;
	musb->g.a_hnp_support = 0;

	/* Normal reset, as B-Device;
	 * or else after HNP, as A-Device
	 */
	if (devctl & MUSB_DEVCTL_BDEVICE) {
		musb->xceiv->state = OTG_STATE_B_PERIPHERAL;
		musb->g.is_a_peripheral = 0;
	} else if (is_otg_enabled(musb)) {
		musb->xceiv->state = OTG_STATE_A_PERIPHERAL;
		musb->g.is_a_peripheral = 1;
	} else
		WARN_ON(1);

	/* start with default limits on VBUS power draw */
	(void) musb_gadget_vbus_draw(&musb->g,
			is_otg_enabled(musb) ? 8 : 100);
}