musb_gadget.c

/*
 * 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
 */

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

#define is_buffer_mapped(req) (is_dma_capable() && \
					(req->map_state != UN_MAPPED))

/* Maps the buffer to dma  */

static inline void map_dma_buffer(struct musb_request *request,
			struct musb *musb, struct musb_ep *musb_ep)
{
	int compatible = true;
	struct dma_controller *dma = musb->dma_controller;

	request->map_state = UN_MAPPED;

	if (!is_dma_capable() || !musb_ep->dma)
		return;

	/* Check if DMA engine can handle this request.
	 * DMA code must reject the USB request explicitly.
	 * Default behaviour is to map the request.
	 */
	if (dma->is_compatible)
		compatible = dma->is_compatible(musb_ep->dma,
				musb_ep->packet_sz, request->request.buf,
				request->request.length);
	if (!compatible)
		return;

	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->map_state = MUSB_MAPPED;
	} else {
		dma_sync_single_for_device(musb->controller,
			request->request.dma,
			request->request.length,
			request->tx
				? DMA_TO_DEVICE
				: DMA_FROM_DEVICE);
		request->map_state = PRE_MAPPED;
	}
}

/* Unmap the buffer from dma and maps it back to cpu */
static inline void unmap_dma_buffer(struct musb_request *request,
				struct musb *musb)
{
	if (!is_buffer_mapped(request))
		return;

	if (request->request.dma == DMA_ADDR_INVALID) {
		DBG(20, "not unmapping a never mapped buffer\n");
		return;
	}
	if (request->map_state == MUSB_MAPPED) {
		dma_unmap_single(musb->controller,
			request->request.dma,
			request->request.length,
			request->tx
				? DMA_TO_DEVICE
				: DMA_FROM_DEVICE);
		request->request.dma = DMA_ADDR_INVALID;
	} else { /* PRE_MAPPED */
		dma_sync_single_for_cpu(musb->controller,
			request->request.dma,
			request->request.length,
			request->tx
				? DMA_TO_DEVICE
				: DMA_FROM_DEVICE);
	}
	request->map_state = UN_MAPPED;
}

/*
 * 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);
	unmap_dma_buffer(req, musb);
	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_buffer_mapped(req)) {
		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) {
		/*
		 * Unmap the dma buffer back to cpu if dma channel
		 * programming fails
		 */
		unmap_dma_buffer(req, musb);

		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() && is_buffer_mapped(req)) {
		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_buffer_mapped(req)) {
				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;
#ifdef USE_MODE1
				csr |= MUSB_RXCSR_AUTOCLEAR;
				/* 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);
#else
				if (!musb_ep->hb_mult &&
					musb_ep->hw_ep->rx_double_buffered)
					csr |= MUSB_RXCSR_AUTOCLEAR;
#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() && is_buffer_mapped(req)) {
				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
			/*
			 * Unmap the dma buffer back to cpu if dma channel
			 * programming fails. This buffer is mapped if the
			 * channel allocation is successful
			 */
			 if (is_buffer_mapped(req)) {
				unmap_dma_buffer(req, musb);

				/*
				 * Clear DMAENAB and AUTOCLEAR for the
				 * PIO mode transfer
				 */
				csr &= ~(MUSB_RXCSR_DMAENAB | MUSB_RXCSR_AUTOCLEAR);
				musb_writew(epio, MUSB_RXCSR, csr);
			}

			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 && !hw_ep->rx_double_buffered)
				|| (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)) {
			/* In double buffer case, continue to unload fifo if
 			 * there is Rx packet in FIFO.
 			 **/
			csr = musb_readw(epio, MUSB_RXCSR);
			if ((csr & MUSB_RXCSR_RXPKTRDY) &&
				hw_ep->rx_double_buffered)
				goto exit;
			return;
		}
#endif
		musb_g_giveback(musb_ep, request, 0);

		request = next_request(musb_ep);
		if (!request)
			return;
	}
#if defined(CONFIG_USB_INVENTRA_DMA) || defined(CONFIG_USB_TUSB_OMAP_DMA)
exit:
#endif
	/* 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) {