xref: /openbmc/linux/drivers/usb/musb/musb_host.c (revision ef2b56df)
1 /*
2  * MUSB OTG driver host support
3  *
4  * Copyright 2005 Mentor Graphics Corporation
5  * Copyright (C) 2005-2006 by Texas Instruments
6  * Copyright (C) 2006-2007 Nokia Corporation
7  * Copyright (C) 2008-2009 MontaVista Software, Inc. <source@mvista.com>
8  *
9  * This program is free software; you can redistribute it and/or
10  * modify it under the terms of the GNU General Public License
11  * version 2 as published by the Free Software Foundation.
12  *
13  * This program is distributed in the hope that it will be useful, but
14  * WITHOUT ANY WARRANTY; without even the implied warranty of
15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
16  * General Public License for more details.
17  *
18  * You should have received a copy of the GNU General Public License
19  * along with this program; if not, write to the Free Software
20  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA
21  * 02110-1301 USA
22  *
23  * THIS SOFTWARE IS PROVIDED "AS IS" AND ANY EXPRESS OR IMPLIED
24  * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
25  * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN
26  * NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY DIRECT, INDIRECT,
27  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
28  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF
29  * USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
30  * ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
31  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
32  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
33  *
34  */
35 
36 #include <linux/module.h>
37 #include <linux/kernel.h>
38 #include <linux/delay.h>
39 #include <linux/sched.h>
40 #include <linux/slab.h>
41 #include <linux/errno.h>
42 #include <linux/list.h>
43 #include <linux/dma-mapping.h>
44 
45 #include "musb_core.h"
46 #include "musb_host.h"
47 #include "musb_trace.h"
48 
49 /* MUSB HOST status 22-mar-2006
50  *
51  * - There's still lots of partial code duplication for fault paths, so
52  *   they aren't handled as consistently as they need to be.
53  *
54  * - PIO mostly behaved when last tested.
55  *     + including ep0, with all usbtest cases 9, 10
56  *     + usbtest 14 (ep0out) doesn't seem to run at all
57  *     + double buffered OUT/TX endpoints saw stalls(!) with certain usbtest
58  *       configurations, but otherwise double buffering passes basic tests.
59  *     + for 2.6.N, for N > ~10, needs API changes for hcd framework.
60  *
61  * - DMA (CPPI) ... partially behaves, not currently recommended
62  *     + about 1/15 the speed of typical EHCI implementations (PCI)
63  *     + RX, all too often reqpkt seems to misbehave after tx
64  *     + TX, no known issues (other than evident silicon issue)
65  *
66  * - DMA (Mentor/OMAP) ...has at least toggle update problems
67  *
68  * - [23-feb-2009] minimal traffic scheduling to avoid bulk RX packet
69  *   starvation ... nothing yet for TX, interrupt, or bulk.
70  *
71  * - Not tested with HNP, but some SRP paths seem to behave.
72  *
73  * NOTE 24-August-2006:
74  *
75  * - Bulk traffic finally uses both sides of hardware ep1, freeing up an
76  *   extra endpoint for periodic use enabling hub + keybd + mouse.  That
77  *   mostly works, except that with "usbnet" it's easy to trigger cases
78  *   with "ping" where RX loses.  (a) ping to davinci, even "ping -f",
79  *   fine; but (b) ping _from_ davinci, even "ping -c 1", ICMP RX loses
80  *   although ARP RX wins.  (That test was done with a full speed link.)
81  */
82 
83 
84 /*
85  * NOTE on endpoint usage:
86  *
87  * CONTROL transfers all go through ep0.  BULK ones go through dedicated IN
88  * and OUT endpoints ... hardware is dedicated for those "async" queue(s).
89  * (Yes, bulk _could_ use more of the endpoints than that, and would even
90  * benefit from it.)
91  *
92  * INTERUPPT and ISOCHRONOUS transfers are scheduled to the other endpoints.
93  * So far that scheduling is both dumb and optimistic:  the endpoint will be
94  * "claimed" until its software queue is no longer refilled.  No multiplexing
95  * of transfers between endpoints, or anything clever.
96  */
97 
98 struct musb *hcd_to_musb(struct usb_hcd *hcd)
99 {
100 	return *(struct musb **) hcd->hcd_priv;
101 }
102 
103 
104 static void musb_ep_program(struct musb *musb, u8 epnum,
105 			struct urb *urb, int is_out,
106 			u8 *buf, u32 offset, u32 len);
107 
108 /*
109  * Clear TX fifo. Needed to avoid BABBLE errors.
110  */
111 static void musb_h_tx_flush_fifo(struct musb_hw_ep *ep)
112 {
113 	struct musb	*musb = ep->musb;
114 	void __iomem	*epio = ep->regs;
115 	u16		csr;
116 	int		retries = 1000;
117 
118 	csr = musb_readw(epio, MUSB_TXCSR);
119 	while (csr & MUSB_TXCSR_FIFONOTEMPTY) {
120 		csr |= MUSB_TXCSR_FLUSHFIFO | MUSB_TXCSR_TXPKTRDY;
121 		musb_writew(epio, MUSB_TXCSR, csr);
122 		csr = musb_readw(epio, MUSB_TXCSR);
123 
124 		/*
125 		 * FIXME: sometimes the tx fifo flush failed, it has been
126 		 * observed during device disconnect on AM335x.
127 		 *
128 		 * To reproduce the issue, ensure tx urb(s) are queued when
129 		 * unplug the usb device which is connected to AM335x usb
130 		 * host port.
131 		 *
132 		 * I found using a usb-ethernet device and running iperf
133 		 * (client on AM335x) has very high chance to trigger it.
134 		 *
135 		 * Better to turn on musb_dbg() in musb_cleanup_urb() with
136 		 * CPPI enabled to see the issue when aborting the tx channel.
137 		 */
138 		if (dev_WARN_ONCE(musb->controller, retries-- < 1,
139 				"Could not flush host TX%d fifo: csr: %04x\n",
140 				ep->epnum, csr))
141 			return;
142 		mdelay(1);
143 	}
144 }
145 
146 static void musb_h_ep0_flush_fifo(struct musb_hw_ep *ep)
147 {
148 	void __iomem	*epio = ep->regs;
149 	u16		csr;
150 	int		retries = 5;
151 
152 	/* scrub any data left in the fifo */
153 	do {
154 		csr = musb_readw(epio, MUSB_TXCSR);
155 		if (!(csr & (MUSB_CSR0_TXPKTRDY | MUSB_CSR0_RXPKTRDY)))
156 			break;
157 		musb_writew(epio, MUSB_TXCSR, MUSB_CSR0_FLUSHFIFO);
158 		csr = musb_readw(epio, MUSB_TXCSR);
159 		udelay(10);
160 	} while (--retries);
161 
162 	WARN(!retries, "Could not flush host TX%d fifo: csr: %04x\n",
163 			ep->epnum, csr);
164 
165 	/* and reset for the next transfer */
166 	musb_writew(epio, MUSB_TXCSR, 0);
167 }
168 
169 /*
170  * Start transmit. Caller is responsible for locking shared resources.
171  * musb must be locked.
172  */
173 static inline void musb_h_tx_start(struct musb_hw_ep *ep)
174 {
175 	u16	txcsr;
176 
177 	/* NOTE: no locks here; caller should lock and select EP */
178 	if (ep->epnum) {
179 		txcsr = musb_readw(ep->regs, MUSB_TXCSR);
180 		txcsr |= MUSB_TXCSR_TXPKTRDY | MUSB_TXCSR_H_WZC_BITS;
181 		musb_writew(ep->regs, MUSB_TXCSR, txcsr);
182 	} else {
183 		txcsr = MUSB_CSR0_H_SETUPPKT | MUSB_CSR0_TXPKTRDY;
184 		musb_writew(ep->regs, MUSB_CSR0, txcsr);
185 	}
186 
187 }
188 
189 static inline void musb_h_tx_dma_start(struct musb_hw_ep *ep)
190 {
191 	u16	txcsr;
192 
193 	/* NOTE: no locks here; caller should lock and select EP */
194 	txcsr = musb_readw(ep->regs, MUSB_TXCSR);
195 	txcsr |= MUSB_TXCSR_DMAENAB | MUSB_TXCSR_H_WZC_BITS;
196 	if (is_cppi_enabled(ep->musb))
197 		txcsr |= MUSB_TXCSR_DMAMODE;
198 	musb_writew(ep->regs, MUSB_TXCSR, txcsr);
199 }
200 
201 static void musb_ep_set_qh(struct musb_hw_ep *ep, int is_in, struct musb_qh *qh)
202 {
203 	if (is_in != 0 || ep->is_shared_fifo)
204 		ep->in_qh  = qh;
205 	if (is_in == 0 || ep->is_shared_fifo)
206 		ep->out_qh = qh;
207 }
208 
209 static struct musb_qh *musb_ep_get_qh(struct musb_hw_ep *ep, int is_in)
210 {
211 	return is_in ? ep->in_qh : ep->out_qh;
212 }
213 
214 /*
215  * Start the URB at the front of an endpoint's queue
216  * end must be claimed from the caller.
217  *
218  * Context: controller locked, irqs blocked
219  */
220 static void
221 musb_start_urb(struct musb *musb, int is_in, struct musb_qh *qh)
222 {
223 	u16			frame;
224 	u32			len;
225 	void __iomem		*mbase =  musb->mregs;
226 	struct urb		*urb = next_urb(qh);
227 	void			*buf = urb->transfer_buffer;
228 	u32			offset = 0;
229 	struct musb_hw_ep	*hw_ep = qh->hw_ep;
230 	int			epnum = hw_ep->epnum;
231 
232 	/* initialize software qh state */
233 	qh->offset = 0;
234 	qh->segsize = 0;
235 
236 	/* gather right source of data */
237 	switch (qh->type) {
238 	case USB_ENDPOINT_XFER_CONTROL:
239 		/* control transfers always start with SETUP */
240 		is_in = 0;
241 		musb->ep0_stage = MUSB_EP0_START;
242 		buf = urb->setup_packet;
243 		len = 8;
244 		break;
245 	case USB_ENDPOINT_XFER_ISOC:
246 		qh->iso_idx = 0;
247 		qh->frame = 0;
248 		offset = urb->iso_frame_desc[0].offset;
249 		len = urb->iso_frame_desc[0].length;
250 		break;
251 	default:		/* bulk, interrupt */
252 		/* actual_length may be nonzero on retry paths */
253 		buf = urb->transfer_buffer + urb->actual_length;
254 		len = urb->transfer_buffer_length - urb->actual_length;
255 	}
256 
257 	trace_musb_urb_start(musb, urb);
258 
259 	/* Configure endpoint */
260 	musb_ep_set_qh(hw_ep, is_in, qh);
261 	musb_ep_program(musb, epnum, urb, !is_in, buf, offset, len);
262 
263 	/* transmit may have more work: start it when it is time */
264 	if (is_in)
265 		return;
266 
267 	/* determine if the time is right for a periodic transfer */
268 	switch (qh->type) {
269 	case USB_ENDPOINT_XFER_ISOC:
270 	case USB_ENDPOINT_XFER_INT:
271 		musb_dbg(musb, "check whether there's still time for periodic Tx");
272 		frame = musb_readw(mbase, MUSB_FRAME);
273 		/* FIXME this doesn't implement that scheduling policy ...
274 		 * or handle framecounter wrapping
275 		 */
276 		if (1) {	/* Always assume URB_ISO_ASAP */
277 			/* REVISIT the SOF irq handler shouldn't duplicate
278 			 * this code; and we don't init urb->start_frame...
279 			 */
280 			qh->frame = 0;
281 			goto start;
282 		} else {
283 			qh->frame = urb->start_frame;
284 			/* enable SOF interrupt so we can count down */
285 			musb_dbg(musb, "SOF for %d", epnum);
286 #if 1 /* ifndef	CONFIG_ARCH_DAVINCI */
287 			musb_writeb(mbase, MUSB_INTRUSBE, 0xff);
288 #endif
289 		}
290 		break;
291 	default:
292 start:
293 		musb_dbg(musb, "Start TX%d %s", epnum,
294 			hw_ep->tx_channel ? "dma" : "pio");
295 
296 		if (!hw_ep->tx_channel)
297 			musb_h_tx_start(hw_ep);
298 		else if (is_cppi_enabled(musb) || tusb_dma_omap(musb))
299 			musb_h_tx_dma_start(hw_ep);
300 	}
301 }
302 
303 /* Context: caller owns controller lock, IRQs are blocked */
304 static void musb_giveback(struct musb *musb, struct urb *urb, int status)
305 __releases(musb->lock)
306 __acquires(musb->lock)
307 {
308 	trace_musb_urb_gb(musb, urb);
309 
310 	usb_hcd_unlink_urb_from_ep(musb->hcd, urb);
311 	spin_unlock(&musb->lock);
312 	usb_hcd_giveback_urb(musb->hcd, urb, status);
313 	spin_lock(&musb->lock);
314 }
315 
316 /* For bulk/interrupt endpoints only */
317 static inline void musb_save_toggle(struct musb_qh *qh, int is_in,
318 				    struct urb *urb)
319 {
320 	void __iomem		*epio = qh->hw_ep->regs;
321 	u16			csr;
322 
323 	/*
324 	 * FIXME: the current Mentor DMA code seems to have
325 	 * problems getting toggle correct.
326 	 */
327 
328 	if (is_in)
329 		csr = musb_readw(epio, MUSB_RXCSR) & MUSB_RXCSR_H_DATATOGGLE;
330 	else
331 		csr = musb_readw(epio, MUSB_TXCSR) & MUSB_TXCSR_H_DATATOGGLE;
332 
333 	usb_settoggle(urb->dev, qh->epnum, !is_in, csr ? 1 : 0);
334 }
335 
336 /*
337  * Advance this hardware endpoint's queue, completing the specified URB and
338  * advancing to either the next URB queued to that qh, or else invalidating
339  * that qh and advancing to the next qh scheduled after the current one.
340  *
341  * Context: caller owns controller lock, IRQs are blocked
342  */
343 static void musb_advance_schedule(struct musb *musb, struct urb *urb,
344 				  struct musb_hw_ep *hw_ep, int is_in)
345 {
346 	struct musb_qh		*qh = musb_ep_get_qh(hw_ep, is_in);
347 	struct musb_hw_ep	*ep = qh->hw_ep;
348 	int			ready = qh->is_ready;
349 	int			status;
350 
351 	status = (urb->status == -EINPROGRESS) ? 0 : urb->status;
352 
353 	/* save toggle eagerly, for paranoia */
354 	switch (qh->type) {
355 	case USB_ENDPOINT_XFER_BULK:
356 	case USB_ENDPOINT_XFER_INT:
357 		musb_save_toggle(qh, is_in, urb);
358 		break;
359 	case USB_ENDPOINT_XFER_ISOC:
360 		if (status == 0 && urb->error_count)
361 			status = -EXDEV;
362 		break;
363 	}
364 
365 	qh->is_ready = 0;
366 	musb_giveback(musb, urb, status);
367 	qh->is_ready = ready;
368 
369 	/* reclaim resources (and bandwidth) ASAP; deschedule it, and
370 	 * invalidate qh as soon as list_empty(&hep->urb_list)
371 	 */
372 	if (list_empty(&qh->hep->urb_list)) {
373 		struct list_head	*head;
374 		struct dma_controller	*dma = musb->dma_controller;
375 
376 		if (is_in) {
377 			ep->rx_reinit = 1;
378 			if (ep->rx_channel) {
379 				dma->channel_release(ep->rx_channel);
380 				ep->rx_channel = NULL;
381 			}
382 		} else {
383 			ep->tx_reinit = 1;
384 			if (ep->tx_channel) {
385 				dma->channel_release(ep->tx_channel);
386 				ep->tx_channel = NULL;
387 			}
388 		}
389 
390 		/* Clobber old pointers to this qh */
391 		musb_ep_set_qh(ep, is_in, NULL);
392 		qh->hep->hcpriv = NULL;
393 
394 		switch (qh->type) {
395 
396 		case USB_ENDPOINT_XFER_CONTROL:
397 		case USB_ENDPOINT_XFER_BULK:
398 			/* fifo policy for these lists, except that NAKing
399 			 * should rotate a qh to the end (for fairness).
400 			 */
401 			if (qh->mux == 1) {
402 				head = qh->ring.prev;
403 				list_del(&qh->ring);
404 				kfree(qh);
405 				qh = first_qh(head);
406 				break;
407 			}
408 
409 		case USB_ENDPOINT_XFER_ISOC:
410 		case USB_ENDPOINT_XFER_INT:
411 			/* this is where periodic bandwidth should be
412 			 * de-allocated if it's tracked and allocated;
413 			 * and where we'd update the schedule tree...
414 			 */
415 			kfree(qh);
416 			qh = NULL;
417 			break;
418 		}
419 	}
420 
421 	/*
422 	 * The pipe must be broken if current urb->status is set, so don't
423 	 * start next urb.
424 	 * TODO: to minimize the risk of regression, only check urb->status
425 	 * for RX, until we have a test case to understand the behavior of TX.
426 	 */
427 	if ((!status || !is_in) && qh && qh->is_ready) {
428 		musb_dbg(musb, "... next ep%d %cX urb %p",
429 		    hw_ep->epnum, is_in ? 'R' : 'T', next_urb(qh));
430 		musb_start_urb(musb, is_in, qh);
431 	}
432 }
433 
434 static u16 musb_h_flush_rxfifo(struct musb_hw_ep *hw_ep, u16 csr)
435 {
436 	/* we don't want fifo to fill itself again;
437 	 * ignore dma (various models),
438 	 * leave toggle alone (may not have been saved yet)
439 	 */
440 	csr |= MUSB_RXCSR_FLUSHFIFO | MUSB_RXCSR_RXPKTRDY;
441 	csr &= ~(MUSB_RXCSR_H_REQPKT
442 		| MUSB_RXCSR_H_AUTOREQ
443 		| MUSB_RXCSR_AUTOCLEAR);
444 
445 	/* write 2x to allow double buffering */
446 	musb_writew(hw_ep->regs, MUSB_RXCSR, csr);
447 	musb_writew(hw_ep->regs, MUSB_RXCSR, csr);
448 
449 	/* flush writebuffer */
450 	return musb_readw(hw_ep->regs, MUSB_RXCSR);
451 }
452 
453 /*
454  * PIO RX for a packet (or part of it).
455  */
456 static bool
457 musb_host_packet_rx(struct musb *musb, struct urb *urb, u8 epnum, u8 iso_err)
458 {
459 	u16			rx_count;
460 	u8			*buf;
461 	u16			csr;
462 	bool			done = false;
463 	u32			length;
464 	int			do_flush = 0;
465 	struct musb_hw_ep	*hw_ep = musb->endpoints + epnum;
466 	void __iomem		*epio = hw_ep->regs;
467 	struct musb_qh		*qh = hw_ep->in_qh;
468 	int			pipe = urb->pipe;
469 	void			*buffer = urb->transfer_buffer;
470 
471 	/* musb_ep_select(mbase, epnum); */
472 	rx_count = musb_readw(epio, MUSB_RXCOUNT);
473 	musb_dbg(musb, "RX%d count %d, buffer %p len %d/%d", epnum, rx_count,
474 			urb->transfer_buffer, qh->offset,
475 			urb->transfer_buffer_length);
476 
477 	/* unload FIFO */
478 	if (usb_pipeisoc(pipe)) {
479 		int					status = 0;
480 		struct usb_iso_packet_descriptor	*d;
481 
482 		if (iso_err) {
483 			status = -EILSEQ;
484 			urb->error_count++;
485 		}
486 
487 		d = urb->iso_frame_desc + qh->iso_idx;
488 		buf = buffer + d->offset;
489 		length = d->length;
490 		if (rx_count > length) {
491 			if (status == 0) {
492 				status = -EOVERFLOW;
493 				urb->error_count++;
494 			}
495 			musb_dbg(musb, "OVERFLOW %d into %d", rx_count, length);
496 			do_flush = 1;
497 		} else
498 			length = rx_count;
499 		urb->actual_length += length;
500 		d->actual_length = length;
501 
502 		d->status = status;
503 
504 		/* see if we are done */
505 		done = (++qh->iso_idx >= urb->number_of_packets);
506 	} else {
507 		/* non-isoch */
508 		buf = buffer + qh->offset;
509 		length = urb->transfer_buffer_length - qh->offset;
510 		if (rx_count > length) {
511 			if (urb->status == -EINPROGRESS)
512 				urb->status = -EOVERFLOW;
513 			musb_dbg(musb, "OVERFLOW %d into %d", rx_count, length);
514 			do_flush = 1;
515 		} else
516 			length = rx_count;
517 		urb->actual_length += length;
518 		qh->offset += length;
519 
520 		/* see if we are done */
521 		done = (urb->actual_length == urb->transfer_buffer_length)
522 			|| (rx_count < qh->maxpacket)
523 			|| (urb->status != -EINPROGRESS);
524 		if (done
525 				&& (urb->status == -EINPROGRESS)
526 				&& (urb->transfer_flags & URB_SHORT_NOT_OK)
527 				&& (urb->actual_length
528 					< urb->transfer_buffer_length))
529 			urb->status = -EREMOTEIO;
530 	}
531 
532 	musb_read_fifo(hw_ep, length, buf);
533 
534 	csr = musb_readw(epio, MUSB_RXCSR);
535 	csr |= MUSB_RXCSR_H_WZC_BITS;
536 	if (unlikely(do_flush))
537 		musb_h_flush_rxfifo(hw_ep, csr);
538 	else {
539 		/* REVISIT this assumes AUTOCLEAR is never set */
540 		csr &= ~(MUSB_RXCSR_RXPKTRDY | MUSB_RXCSR_H_REQPKT);
541 		if (!done)
542 			csr |= MUSB_RXCSR_H_REQPKT;
543 		musb_writew(epio, MUSB_RXCSR, csr);
544 	}
545 
546 	return done;
547 }
548 
549 /* we don't always need to reinit a given side of an endpoint...
550  * when we do, use tx/rx reinit routine and then construct a new CSR
551  * to address data toggle, NYET, and DMA or PIO.
552  *
553  * it's possible that driver bugs (especially for DMA) or aborting a
554  * transfer might have left the endpoint busier than it should be.
555  * the busy/not-empty tests are basically paranoia.
556  */
557 static void
558 musb_rx_reinit(struct musb *musb, struct musb_qh *qh, u8 epnum)
559 {
560 	struct musb_hw_ep *ep = musb->endpoints + epnum;
561 	u16	csr;
562 
563 	/* NOTE:  we know the "rx" fifo reinit never triggers for ep0.
564 	 * That always uses tx_reinit since ep0 repurposes TX register
565 	 * offsets; the initial SETUP packet is also a kind of OUT.
566 	 */
567 
568 	/* if programmed for Tx, put it in RX mode */
569 	if (ep->is_shared_fifo) {
570 		csr = musb_readw(ep->regs, MUSB_TXCSR);
571 		if (csr & MUSB_TXCSR_MODE) {
572 			musb_h_tx_flush_fifo(ep);
573 			csr = musb_readw(ep->regs, MUSB_TXCSR);
574 			musb_writew(ep->regs, MUSB_TXCSR,
575 				    csr | MUSB_TXCSR_FRCDATATOG);
576 		}
577 
578 		/*
579 		 * Clear the MODE bit (and everything else) to enable Rx.
580 		 * NOTE: we mustn't clear the DMAMODE bit before DMAENAB.
581 		 */
582 		if (csr & MUSB_TXCSR_DMAMODE)
583 			musb_writew(ep->regs, MUSB_TXCSR, MUSB_TXCSR_DMAMODE);
584 		musb_writew(ep->regs, MUSB_TXCSR, 0);
585 
586 	/* scrub all previous state, clearing toggle */
587 	}
588 	csr = musb_readw(ep->regs, MUSB_RXCSR);
589 	if (csr & MUSB_RXCSR_RXPKTRDY)
590 		WARNING("rx%d, packet/%d ready?\n", ep->epnum,
591 			musb_readw(ep->regs, MUSB_RXCOUNT));
592 
593 	musb_h_flush_rxfifo(ep, MUSB_RXCSR_CLRDATATOG);
594 
595 	/* target addr and (for multipoint) hub addr/port */
596 	if (musb->is_multipoint) {
597 		musb_write_rxfunaddr(musb, epnum, qh->addr_reg);
598 		musb_write_rxhubaddr(musb, epnum, qh->h_addr_reg);
599 		musb_write_rxhubport(musb, epnum, qh->h_port_reg);
600 	} else
601 		musb_writeb(musb->mregs, MUSB_FADDR, qh->addr_reg);
602 
603 	/* protocol/endpoint, interval/NAKlimit, i/o size */
604 	musb_writeb(ep->regs, MUSB_RXTYPE, qh->type_reg);
605 	musb_writeb(ep->regs, MUSB_RXINTERVAL, qh->intv_reg);
606 	/* NOTE: bulk combining rewrites high bits of maxpacket */
607 	/* Set RXMAXP with the FIFO size of the endpoint
608 	 * to disable double buffer mode.
609 	 */
610 	if (musb->double_buffer_not_ok)
611 		musb_writew(ep->regs, MUSB_RXMAXP, ep->max_packet_sz_rx);
612 	else
613 		musb_writew(ep->regs, MUSB_RXMAXP,
614 				qh->maxpacket | ((qh->hb_mult - 1) << 11));
615 
616 	ep->rx_reinit = 0;
617 }
618 
619 static void musb_tx_dma_set_mode_mentor(struct dma_controller *dma,
620 		struct musb_hw_ep *hw_ep, struct musb_qh *qh,
621 		struct urb *urb, u32 offset,
622 		u32 *length, u8 *mode)
623 {
624 	struct dma_channel	*channel = hw_ep->tx_channel;
625 	void __iomem		*epio = hw_ep->regs;
626 	u16			pkt_size = qh->maxpacket;
627 	u16			csr;
628 
629 	if (*length > channel->max_len)
630 		*length = channel->max_len;
631 
632 	csr = musb_readw(epio, MUSB_TXCSR);
633 	if (*length > pkt_size) {
634 		*mode = 1;
635 		csr |= MUSB_TXCSR_DMAMODE | MUSB_TXCSR_DMAENAB;
636 		/* autoset shouldn't be set in high bandwidth */
637 		/*
638 		 * Enable Autoset according to table
639 		 * below
640 		 * bulk_split hb_mult	Autoset_Enable
641 		 *	0	1	Yes(Normal)
642 		 *	0	>1	No(High BW ISO)
643 		 *	1	1	Yes(HS bulk)
644 		 *	1	>1	Yes(FS bulk)
645 		 */
646 		if (qh->hb_mult == 1 || (qh->hb_mult > 1 &&
647 					can_bulk_split(hw_ep->musb, qh->type)))
648 			csr |= MUSB_TXCSR_AUTOSET;
649 	} else {
650 		*mode = 0;
651 		csr &= ~(MUSB_TXCSR_AUTOSET | MUSB_TXCSR_DMAMODE);
652 		csr |= MUSB_TXCSR_DMAENAB; /* against programmer's guide */
653 	}
654 	channel->desired_mode = *mode;
655 	musb_writew(epio, MUSB_TXCSR, csr);
656 }
657 
658 static void musb_tx_dma_set_mode_cppi_tusb(struct dma_controller *dma,
659 					   struct musb_hw_ep *hw_ep,
660 					   struct musb_qh *qh,
661 					   struct urb *urb,
662 					   u32 offset,
663 					   u32 *length,
664 					   u8 *mode)
665 {
666 	struct dma_channel *channel = hw_ep->tx_channel;
667 
668 	channel->actual_len = 0;
669 
670 	/*
671 	 * TX uses "RNDIS" mode automatically but needs help
672 	 * to identify the zero-length-final-packet case.
673 	 */
674 	*mode = (urb->transfer_flags & URB_ZERO_PACKET) ? 1 : 0;
675 }
676 
677 static bool musb_tx_dma_program(struct dma_controller *dma,
678 		struct musb_hw_ep *hw_ep, struct musb_qh *qh,
679 		struct urb *urb, u32 offset, u32 length)
680 {
681 	struct dma_channel	*channel = hw_ep->tx_channel;
682 	u16			pkt_size = qh->maxpacket;
683 	u8			mode;
684 
685 	if (musb_dma_inventra(hw_ep->musb) || musb_dma_ux500(hw_ep->musb))
686 		musb_tx_dma_set_mode_mentor(dma, hw_ep, qh, urb, offset,
687 					    &length, &mode);
688 	else if (is_cppi_enabled(hw_ep->musb) || tusb_dma_omap(hw_ep->musb))
689 		musb_tx_dma_set_mode_cppi_tusb(dma, hw_ep, qh, urb, offset,
690 					       &length, &mode);
691 	else
692 		return false;
693 
694 	qh->segsize = length;
695 
696 	/*
697 	 * Ensure the data reaches to main memory before starting
698 	 * DMA transfer
699 	 */
700 	wmb();
701 
702 	if (!dma->channel_program(channel, pkt_size, mode,
703 			urb->transfer_dma + offset, length)) {
704 		void __iomem *epio = hw_ep->regs;
705 		u16 csr;
706 
707 		dma->channel_release(channel);
708 		hw_ep->tx_channel = NULL;
709 
710 		csr = musb_readw(epio, MUSB_TXCSR);
711 		csr &= ~(MUSB_TXCSR_AUTOSET | MUSB_TXCSR_DMAENAB);
712 		musb_writew(epio, MUSB_TXCSR, csr | MUSB_TXCSR_H_WZC_BITS);
713 		return false;
714 	}
715 	return true;
716 }
717 
718 /*
719  * Program an HDRC endpoint as per the given URB
720  * Context: irqs blocked, controller lock held
721  */
722 static void musb_ep_program(struct musb *musb, u8 epnum,
723 			struct urb *urb, int is_out,
724 			u8 *buf, u32 offset, u32 len)
725 {
726 	struct dma_controller	*dma_controller;
727 	struct dma_channel	*dma_channel;
728 	u8			dma_ok;
729 	void __iomem		*mbase = musb->mregs;
730 	struct musb_hw_ep	*hw_ep = musb->endpoints + epnum;
731 	void __iomem		*epio = hw_ep->regs;
732 	struct musb_qh		*qh = musb_ep_get_qh(hw_ep, !is_out);
733 	u16			packet_sz = qh->maxpacket;
734 	u8			use_dma = 1;
735 	u16			csr;
736 
737 	musb_dbg(musb, "%s hw%d urb %p spd%d dev%d ep%d%s "
738 				"h_addr%02x h_port%02x bytes %d",
739 			is_out ? "-->" : "<--",
740 			epnum, urb, urb->dev->speed,
741 			qh->addr_reg, qh->epnum, is_out ? "out" : "in",
742 			qh->h_addr_reg, qh->h_port_reg,
743 			len);
744 
745 	musb_ep_select(mbase, epnum);
746 
747 	if (is_out && !len) {
748 		use_dma = 0;
749 		csr = musb_readw(epio, MUSB_TXCSR);
750 		csr &= ~MUSB_TXCSR_DMAENAB;
751 		musb_writew(epio, MUSB_TXCSR, csr);
752 		hw_ep->tx_channel = NULL;
753 	}
754 
755 	/* candidate for DMA? */
756 	dma_controller = musb->dma_controller;
757 	if (use_dma && is_dma_capable() && epnum && dma_controller) {
758 		dma_channel = is_out ? hw_ep->tx_channel : hw_ep->rx_channel;
759 		if (!dma_channel) {
760 			dma_channel = dma_controller->channel_alloc(
761 					dma_controller, hw_ep, is_out);
762 			if (is_out)
763 				hw_ep->tx_channel = dma_channel;
764 			else
765 				hw_ep->rx_channel = dma_channel;
766 		}
767 	} else
768 		dma_channel = NULL;
769 
770 	/* make sure we clear DMAEnab, autoSet bits from previous run */
771 
772 	/* OUT/transmit/EP0 or IN/receive? */
773 	if (is_out) {
774 		u16	csr;
775 		u16	int_txe;
776 		u16	load_count;
777 
778 		csr = musb_readw(epio, MUSB_TXCSR);
779 
780 		/* disable interrupt in case we flush */
781 		int_txe = musb->intrtxe;
782 		musb_writew(mbase, MUSB_INTRTXE, int_txe & ~(1 << epnum));
783 
784 		/* general endpoint setup */
785 		if (epnum) {
786 			/* flush all old state, set default */
787 			/*
788 			 * We could be flushing valid
789 			 * packets in double buffering
790 			 * case
791 			 */
792 			if (!hw_ep->tx_double_buffered)
793 				musb_h_tx_flush_fifo(hw_ep);
794 
795 			/*
796 			 * We must not clear the DMAMODE bit before or in
797 			 * the same cycle with the DMAENAB bit, so we clear
798 			 * the latter first...
799 			 */
800 			csr &= ~(MUSB_TXCSR_H_NAKTIMEOUT
801 					| MUSB_TXCSR_AUTOSET
802 					| MUSB_TXCSR_DMAENAB
803 					| MUSB_TXCSR_FRCDATATOG
804 					| MUSB_TXCSR_H_RXSTALL
805 					| MUSB_TXCSR_H_ERROR
806 					| MUSB_TXCSR_TXPKTRDY
807 					);
808 			csr |= MUSB_TXCSR_MODE;
809 
810 			if (!hw_ep->tx_double_buffered) {
811 				if (usb_gettoggle(urb->dev, qh->epnum, 1))
812 					csr |= MUSB_TXCSR_H_WR_DATATOGGLE
813 						| MUSB_TXCSR_H_DATATOGGLE;
814 				else
815 					csr |= MUSB_TXCSR_CLRDATATOG;
816 			}
817 
818 			musb_writew(epio, MUSB_TXCSR, csr);
819 			/* REVISIT may need to clear FLUSHFIFO ... */
820 			csr &= ~MUSB_TXCSR_DMAMODE;
821 			musb_writew(epio, MUSB_TXCSR, csr);
822 			csr = musb_readw(epio, MUSB_TXCSR);
823 		} else {
824 			/* endpoint 0: just flush */
825 			musb_h_ep0_flush_fifo(hw_ep);
826 		}
827 
828 		/* target addr and (for multipoint) hub addr/port */
829 		if (musb->is_multipoint) {
830 			musb_write_txfunaddr(musb, epnum, qh->addr_reg);
831 			musb_write_txhubaddr(musb, epnum, qh->h_addr_reg);
832 			musb_write_txhubport(musb, epnum, qh->h_port_reg);
833 /* FIXME if !epnum, do the same for RX ... */
834 		} else
835 			musb_writeb(mbase, MUSB_FADDR, qh->addr_reg);
836 
837 		/* protocol/endpoint/interval/NAKlimit */
838 		if (epnum) {
839 			musb_writeb(epio, MUSB_TXTYPE, qh->type_reg);
840 			if (musb->double_buffer_not_ok) {
841 				musb_writew(epio, MUSB_TXMAXP,
842 						hw_ep->max_packet_sz_tx);
843 			} else if (can_bulk_split(musb, qh->type)) {
844 				qh->hb_mult = hw_ep->max_packet_sz_tx
845 						/ packet_sz;
846 				musb_writew(epio, MUSB_TXMAXP, packet_sz
847 					| ((qh->hb_mult) - 1) << 11);
848 			} else {
849 				musb_writew(epio, MUSB_TXMAXP,
850 						qh->maxpacket |
851 						((qh->hb_mult - 1) << 11));
852 			}
853 			musb_writeb(epio, MUSB_TXINTERVAL, qh->intv_reg);
854 		} else {
855 			musb_writeb(epio, MUSB_NAKLIMIT0, qh->intv_reg);
856 			if (musb->is_multipoint)
857 				musb_writeb(epio, MUSB_TYPE0,
858 						qh->type_reg);
859 		}
860 
861 		if (can_bulk_split(musb, qh->type))
862 			load_count = min((u32) hw_ep->max_packet_sz_tx,
863 						len);
864 		else
865 			load_count = min((u32) packet_sz, len);
866 
867 		if (dma_channel && musb_tx_dma_program(dma_controller,
868 					hw_ep, qh, urb, offset, len))
869 			load_count = 0;
870 
871 		if (load_count) {
872 			/* PIO to load FIFO */
873 			qh->segsize = load_count;
874 			if (!buf) {
875 				sg_miter_start(&qh->sg_miter, urb->sg, 1,
876 						SG_MITER_ATOMIC
877 						| SG_MITER_FROM_SG);
878 				if (!sg_miter_next(&qh->sg_miter)) {
879 					dev_err(musb->controller,
880 							"error: sg"
881 							"list empty\n");
882 					sg_miter_stop(&qh->sg_miter);
883 					goto finish;
884 				}
885 				buf = qh->sg_miter.addr + urb->sg->offset +
886 					urb->actual_length;
887 				load_count = min_t(u32, load_count,
888 						qh->sg_miter.length);
889 				musb_write_fifo(hw_ep, load_count, buf);
890 				qh->sg_miter.consumed = load_count;
891 				sg_miter_stop(&qh->sg_miter);
892 			} else
893 				musb_write_fifo(hw_ep, load_count, buf);
894 		}
895 finish:
896 		/* re-enable interrupt */
897 		musb_writew(mbase, MUSB_INTRTXE, int_txe);
898 
899 	/* IN/receive */
900 	} else {
901 		u16	csr;
902 
903 		if (hw_ep->rx_reinit) {
904 			musb_rx_reinit(musb, qh, epnum);
905 
906 			/* init new state: toggle and NYET, maybe DMA later */
907 			if (usb_gettoggle(urb->dev, qh->epnum, 0))
908 				csr = MUSB_RXCSR_H_WR_DATATOGGLE
909 					| MUSB_RXCSR_H_DATATOGGLE;
910 			else
911 				csr = 0;
912 			if (qh->type == USB_ENDPOINT_XFER_INT)
913 				csr |= MUSB_RXCSR_DISNYET;
914 
915 		} else {
916 			csr = musb_readw(hw_ep->regs, MUSB_RXCSR);
917 
918 			if (csr & (MUSB_RXCSR_RXPKTRDY
919 					| MUSB_RXCSR_DMAENAB
920 					| MUSB_RXCSR_H_REQPKT))
921 				ERR("broken !rx_reinit, ep%d csr %04x\n",
922 						hw_ep->epnum, csr);
923 
924 			/* scrub any stale state, leaving toggle alone */
925 			csr &= MUSB_RXCSR_DISNYET;
926 		}
927 
928 		/* kick things off */
929 
930 		if ((is_cppi_enabled(musb) || tusb_dma_omap(musb)) && dma_channel) {
931 			/* Candidate for DMA */
932 			dma_channel->actual_len = 0L;
933 			qh->segsize = len;
934 
935 			/* AUTOREQ is in a DMA register */
936 			musb_writew(hw_ep->regs, MUSB_RXCSR, csr);
937 			csr = musb_readw(hw_ep->regs, MUSB_RXCSR);
938 
939 			/*
940 			 * Unless caller treats short RX transfers as
941 			 * errors, we dare not queue multiple transfers.
942 			 */
943 			dma_ok = dma_controller->channel_program(dma_channel,
944 					packet_sz, !(urb->transfer_flags &
945 						     URB_SHORT_NOT_OK),
946 					urb->transfer_dma + offset,
947 					qh->segsize);
948 			if (!dma_ok) {
949 				dma_controller->channel_release(dma_channel);
950 				hw_ep->rx_channel = dma_channel = NULL;
951 			} else
952 				csr |= MUSB_RXCSR_DMAENAB;
953 		}
954 
955 		csr |= MUSB_RXCSR_H_REQPKT;
956 		musb_dbg(musb, "RXCSR%d := %04x", epnum, csr);
957 		musb_writew(hw_ep->regs, MUSB_RXCSR, csr);
958 		csr = musb_readw(hw_ep->regs, MUSB_RXCSR);
959 	}
960 }
961 
962 /* Schedule next QH from musb->in_bulk/out_bulk and move the current qh to
963  * the end; avoids starvation for other endpoints.
964  */
965 static void musb_bulk_nak_timeout(struct musb *musb, struct musb_hw_ep *ep,
966 	int is_in)
967 {
968 	struct dma_channel	*dma;
969 	struct urb		*urb;
970 	void __iomem		*mbase = musb->mregs;
971 	void __iomem		*epio = ep->regs;
972 	struct musb_qh		*cur_qh, *next_qh;
973 	u16			rx_csr, tx_csr;
974 
975 	musb_ep_select(mbase, ep->epnum);
976 	if (is_in) {
977 		dma = is_dma_capable() ? ep->rx_channel : NULL;
978 
979 		/*
980 		 * Need to stop the transaction by clearing REQPKT first
981 		 * then the NAK Timeout bit ref MUSBMHDRC USB 2.0 HIGH-SPEED
982 		 * DUAL-ROLE CONTROLLER Programmer's Guide, section 9.2.2
983 		 */
984 		rx_csr = musb_readw(epio, MUSB_RXCSR);
985 		rx_csr |= MUSB_RXCSR_H_WZC_BITS;
986 		rx_csr &= ~MUSB_RXCSR_H_REQPKT;
987 		musb_writew(epio, MUSB_RXCSR, rx_csr);
988 		rx_csr &= ~MUSB_RXCSR_DATAERROR;
989 		musb_writew(epio, MUSB_RXCSR, rx_csr);
990 
991 		cur_qh = first_qh(&musb->in_bulk);
992 	} else {
993 		dma = is_dma_capable() ? ep->tx_channel : NULL;
994 
995 		/* clear nak timeout bit */
996 		tx_csr = musb_readw(epio, MUSB_TXCSR);
997 		tx_csr |= MUSB_TXCSR_H_WZC_BITS;
998 		tx_csr &= ~MUSB_TXCSR_H_NAKTIMEOUT;
999 		musb_writew(epio, MUSB_TXCSR, tx_csr);
1000 
1001 		cur_qh = first_qh(&musb->out_bulk);
1002 	}
1003 	if (cur_qh) {
1004 		urb = next_urb(cur_qh);
1005 		if (dma_channel_status(dma) == MUSB_DMA_STATUS_BUSY) {
1006 			dma->status = MUSB_DMA_STATUS_CORE_ABORT;
1007 			musb->dma_controller->channel_abort(dma);
1008 			urb->actual_length += dma->actual_len;
1009 			dma->actual_len = 0L;
1010 		}
1011 		musb_save_toggle(cur_qh, is_in, urb);
1012 
1013 		if (is_in) {
1014 			/* move cur_qh to end of queue */
1015 			list_move_tail(&cur_qh->ring, &musb->in_bulk);
1016 
1017 			/* get the next qh from musb->in_bulk */
1018 			next_qh = first_qh(&musb->in_bulk);
1019 
1020 			/* set rx_reinit and schedule the next qh */
1021 			ep->rx_reinit = 1;
1022 		} else {
1023 			/* move cur_qh to end of queue */
1024 			list_move_tail(&cur_qh->ring, &musb->out_bulk);
1025 
1026 			/* get the next qh from musb->out_bulk */
1027 			next_qh = first_qh(&musb->out_bulk);
1028 
1029 			/* set tx_reinit and schedule the next qh */
1030 			ep->tx_reinit = 1;
1031 		}
1032 		musb_start_urb(musb, is_in, next_qh);
1033 	}
1034 }
1035 
1036 /*
1037  * Service the default endpoint (ep0) as host.
1038  * Return true until it's time to start the status stage.
1039  */
1040 static bool musb_h_ep0_continue(struct musb *musb, u16 len, struct urb *urb)
1041 {
1042 	bool			 more = false;
1043 	u8			*fifo_dest = NULL;
1044 	u16			fifo_count = 0;
1045 	struct musb_hw_ep	*hw_ep = musb->control_ep;
1046 	struct musb_qh		*qh = hw_ep->in_qh;
1047 	struct usb_ctrlrequest	*request;
1048 
1049 	switch (musb->ep0_stage) {
1050 	case MUSB_EP0_IN:
1051 		fifo_dest = urb->transfer_buffer + urb->actual_length;
1052 		fifo_count = min_t(size_t, len, urb->transfer_buffer_length -
1053 				   urb->actual_length);
1054 		if (fifo_count < len)
1055 			urb->status = -EOVERFLOW;
1056 
1057 		musb_read_fifo(hw_ep, fifo_count, fifo_dest);
1058 
1059 		urb->actual_length += fifo_count;
1060 		if (len < qh->maxpacket) {
1061 			/* always terminate on short read; it's
1062 			 * rarely reported as an error.
1063 			 */
1064 		} else if (urb->actual_length <
1065 				urb->transfer_buffer_length)
1066 			more = true;
1067 		break;
1068 	case MUSB_EP0_START:
1069 		request = (struct usb_ctrlrequest *) urb->setup_packet;
1070 
1071 		if (!request->wLength) {
1072 			musb_dbg(musb, "start no-DATA");
1073 			break;
1074 		} else if (request->bRequestType & USB_DIR_IN) {
1075 			musb_dbg(musb, "start IN-DATA");
1076 			musb->ep0_stage = MUSB_EP0_IN;
1077 			more = true;
1078 			break;
1079 		} else {
1080 			musb_dbg(musb, "start OUT-DATA");
1081 			musb->ep0_stage = MUSB_EP0_OUT;
1082 			more = true;
1083 		}
1084 		/* FALLTHROUGH */
1085 	case MUSB_EP0_OUT:
1086 		fifo_count = min_t(size_t, qh->maxpacket,
1087 				   urb->transfer_buffer_length -
1088 				   urb->actual_length);
1089 		if (fifo_count) {
1090 			fifo_dest = (u8 *) (urb->transfer_buffer
1091 					+ urb->actual_length);
1092 			musb_dbg(musb, "Sending %d byte%s to ep0 fifo %p",
1093 					fifo_count,
1094 					(fifo_count == 1) ? "" : "s",
1095 					fifo_dest);
1096 			musb_write_fifo(hw_ep, fifo_count, fifo_dest);
1097 
1098 			urb->actual_length += fifo_count;
1099 			more = true;
1100 		}
1101 		break;
1102 	default:
1103 		ERR("bogus ep0 stage %d\n", musb->ep0_stage);
1104 		break;
1105 	}
1106 
1107 	return more;
1108 }
1109 
1110 /*
1111  * Handle default endpoint interrupt as host. Only called in IRQ time
1112  * from musb_interrupt().
1113  *
1114  * called with controller irqlocked
1115  */
1116 irqreturn_t musb_h_ep0_irq(struct musb *musb)
1117 {
1118 	struct urb		*urb;
1119 	u16			csr, len;
1120 	int			status = 0;
1121 	void __iomem		*mbase = musb->mregs;
1122 	struct musb_hw_ep	*hw_ep = musb->control_ep;
1123 	void __iomem		*epio = hw_ep->regs;
1124 	struct musb_qh		*qh = hw_ep->in_qh;
1125 	bool			complete = false;
1126 	irqreturn_t		retval = IRQ_NONE;
1127 
1128 	/* ep0 only has one queue, "in" */
1129 	urb = next_urb(qh);
1130 
1131 	musb_ep_select(mbase, 0);
1132 	csr = musb_readw(epio, MUSB_CSR0);
1133 	len = (csr & MUSB_CSR0_RXPKTRDY)
1134 			? musb_readb(epio, MUSB_COUNT0)
1135 			: 0;
1136 
1137 	musb_dbg(musb, "<== csr0 %04x, qh %p, count %d, urb %p, stage %d",
1138 		csr, qh, len, urb, musb->ep0_stage);
1139 
1140 	/* if we just did status stage, we are done */
1141 	if (MUSB_EP0_STATUS == musb->ep0_stage) {
1142 		retval = IRQ_HANDLED;
1143 		complete = true;
1144 	}
1145 
1146 	/* prepare status */
1147 	if (csr & MUSB_CSR0_H_RXSTALL) {
1148 		musb_dbg(musb, "STALLING ENDPOINT");
1149 		status = -EPIPE;
1150 
1151 	} else if (csr & MUSB_CSR0_H_ERROR) {
1152 		musb_dbg(musb, "no response, csr0 %04x", csr);
1153 		status = -EPROTO;
1154 
1155 	} else if (csr & MUSB_CSR0_H_NAKTIMEOUT) {
1156 		musb_dbg(musb, "control NAK timeout");
1157 
1158 		/* NOTE:  this code path would be a good place to PAUSE a
1159 		 * control transfer, if another one is queued, so that
1160 		 * ep0 is more likely to stay busy.  That's already done
1161 		 * for bulk RX transfers.
1162 		 *
1163 		 * if (qh->ring.next != &musb->control), then
1164 		 * we have a candidate... NAKing is *NOT* an error
1165 		 */
1166 		musb_writew(epio, MUSB_CSR0, 0);
1167 		retval = IRQ_HANDLED;
1168 	}
1169 
1170 	if (status) {
1171 		musb_dbg(musb, "aborting");
1172 		retval = IRQ_HANDLED;
1173 		if (urb)
1174 			urb->status = status;
1175 		complete = true;
1176 
1177 		/* use the proper sequence to abort the transfer */
1178 		if (csr & MUSB_CSR0_H_REQPKT) {
1179 			csr &= ~MUSB_CSR0_H_REQPKT;
1180 			musb_writew(epio, MUSB_CSR0, csr);
1181 			csr &= ~MUSB_CSR0_H_NAKTIMEOUT;
1182 			musb_writew(epio, MUSB_CSR0, csr);
1183 		} else {
1184 			musb_h_ep0_flush_fifo(hw_ep);
1185 		}
1186 
1187 		musb_writeb(epio, MUSB_NAKLIMIT0, 0);
1188 
1189 		/* clear it */
1190 		musb_writew(epio, MUSB_CSR0, 0);
1191 	}
1192 
1193 	if (unlikely(!urb)) {
1194 		/* stop endpoint since we have no place for its data, this
1195 		 * SHOULD NEVER HAPPEN! */
1196 		ERR("no URB for end 0\n");
1197 
1198 		musb_h_ep0_flush_fifo(hw_ep);
1199 		goto done;
1200 	}
1201 
1202 	if (!complete) {
1203 		/* call common logic and prepare response */
1204 		if (musb_h_ep0_continue(musb, len, urb)) {
1205 			/* more packets required */
1206 			csr = (MUSB_EP0_IN == musb->ep0_stage)
1207 				?  MUSB_CSR0_H_REQPKT : MUSB_CSR0_TXPKTRDY;
1208 		} else {
1209 			/* data transfer complete; perform status phase */
1210 			if (usb_pipeout(urb->pipe)
1211 					|| !urb->transfer_buffer_length)
1212 				csr = MUSB_CSR0_H_STATUSPKT
1213 					| MUSB_CSR0_H_REQPKT;
1214 			else
1215 				csr = MUSB_CSR0_H_STATUSPKT
1216 					| MUSB_CSR0_TXPKTRDY;
1217 
1218 			/* disable ping token in status phase */
1219 			csr |= MUSB_CSR0_H_DIS_PING;
1220 
1221 			/* flag status stage */
1222 			musb->ep0_stage = MUSB_EP0_STATUS;
1223 
1224 			musb_dbg(musb, "ep0 STATUS, csr %04x", csr);
1225 
1226 		}
1227 		musb_writew(epio, MUSB_CSR0, csr);
1228 		retval = IRQ_HANDLED;
1229 	} else
1230 		musb->ep0_stage = MUSB_EP0_IDLE;
1231 
1232 	/* call completion handler if done */
1233 	if (complete)
1234 		musb_advance_schedule(musb, urb, hw_ep, 1);
1235 done:
1236 	return retval;
1237 }
1238 
1239 
1240 #ifdef CONFIG_USB_INVENTRA_DMA
1241 
1242 /* Host side TX (OUT) using Mentor DMA works as follows:
1243 	submit_urb ->
1244 		- if queue was empty, Program Endpoint
1245 		- ... which starts DMA to fifo in mode 1 or 0
1246 
1247 	DMA Isr (transfer complete) -> TxAvail()
1248 		- Stop DMA (~DmaEnab)	(<--- Alert ... currently happens
1249 					only in musb_cleanup_urb)
1250 		- TxPktRdy has to be set in mode 0 or for
1251 			short packets in mode 1.
1252 */
1253 
1254 #endif
1255 
1256 /* Service a Tx-Available or dma completion irq for the endpoint */
1257 void musb_host_tx(struct musb *musb, u8 epnum)
1258 {
1259 	int			pipe;
1260 	bool			done = false;
1261 	u16			tx_csr;
1262 	size_t			length = 0;
1263 	size_t			offset = 0;
1264 	struct musb_hw_ep	*hw_ep = musb->endpoints + epnum;
1265 	void __iomem		*epio = hw_ep->regs;
1266 	struct musb_qh		*qh = hw_ep->out_qh;
1267 	struct urb		*urb = next_urb(qh);
1268 	u32			status = 0;
1269 	void __iomem		*mbase = musb->mregs;
1270 	struct dma_channel	*dma;
1271 	bool			transfer_pending = false;
1272 
1273 	musb_ep_select(mbase, epnum);
1274 	tx_csr = musb_readw(epio, MUSB_TXCSR);
1275 
1276 	/* with CPPI, DMA sometimes triggers "extra" irqs */
1277 	if (!urb) {
1278 		musb_dbg(musb, "extra TX%d ready, csr %04x", epnum, tx_csr);
1279 		return;
1280 	}
1281 
1282 	pipe = urb->pipe;
1283 	dma = is_dma_capable() ? hw_ep->tx_channel : NULL;
1284 	trace_musb_urb_tx(musb, urb);
1285 	musb_dbg(musb, "OUT/TX%d end, csr %04x%s", epnum, tx_csr,
1286 			dma ? ", dma" : "");
1287 
1288 	/* check for errors */
1289 	if (tx_csr & MUSB_TXCSR_H_RXSTALL) {
1290 		/* dma was disabled, fifo flushed */
1291 		musb_dbg(musb, "TX end %d stall", epnum);
1292 
1293 		/* stall; record URB status */
1294 		status = -EPIPE;
1295 
1296 	} else if (tx_csr & MUSB_TXCSR_H_ERROR) {
1297 		/* (NON-ISO) dma was disabled, fifo flushed */
1298 		musb_dbg(musb, "TX 3strikes on ep=%d", epnum);
1299 
1300 		status = -ETIMEDOUT;
1301 
1302 	} else if (tx_csr & MUSB_TXCSR_H_NAKTIMEOUT) {
1303 		if (USB_ENDPOINT_XFER_BULK == qh->type && qh->mux == 1
1304 				&& !list_is_singular(&musb->out_bulk)) {
1305 			musb_dbg(musb, "NAK timeout on TX%d ep", epnum);
1306 			musb_bulk_nak_timeout(musb, hw_ep, 0);
1307 		} else {
1308 			musb_dbg(musb, "TX ep%d device not responding", epnum);
1309 			/* NOTE:  this code path would be a good place to PAUSE a
1310 			 * transfer, if there's some other (nonperiodic) tx urb
1311 			 * that could use this fifo.  (dma complicates it...)
1312 			 * That's already done for bulk RX transfers.
1313 			 *
1314 			 * if (bulk && qh->ring.next != &musb->out_bulk), then
1315 			 * we have a candidate... NAKing is *NOT* an error
1316 			 */
1317 			musb_ep_select(mbase, epnum);
1318 			musb_writew(epio, MUSB_TXCSR,
1319 					MUSB_TXCSR_H_WZC_BITS
1320 					| MUSB_TXCSR_TXPKTRDY);
1321 		}
1322 			return;
1323 	}
1324 
1325 done:
1326 	if (status) {
1327 		if (dma_channel_status(dma) == MUSB_DMA_STATUS_BUSY) {
1328 			dma->status = MUSB_DMA_STATUS_CORE_ABORT;
1329 			musb->dma_controller->channel_abort(dma);
1330 		}
1331 
1332 		/* do the proper sequence to abort the transfer in the
1333 		 * usb core; the dma engine should already be stopped.
1334 		 */
1335 		musb_h_tx_flush_fifo(hw_ep);
1336 		tx_csr &= ~(MUSB_TXCSR_AUTOSET
1337 				| MUSB_TXCSR_DMAENAB
1338 				| MUSB_TXCSR_H_ERROR
1339 				| MUSB_TXCSR_H_RXSTALL
1340 				| MUSB_TXCSR_H_NAKTIMEOUT
1341 				);
1342 
1343 		musb_ep_select(mbase, epnum);
1344 		musb_writew(epio, MUSB_TXCSR, tx_csr);
1345 		/* REVISIT may need to clear FLUSHFIFO ... */
1346 		musb_writew(epio, MUSB_TXCSR, tx_csr);
1347 		musb_writeb(epio, MUSB_TXINTERVAL, 0);
1348 
1349 		done = true;
1350 	}
1351 
1352 	/* second cppi case */
1353 	if (dma_channel_status(dma) == MUSB_DMA_STATUS_BUSY) {
1354 		musb_dbg(musb, "extra TX%d ready, csr %04x", epnum, tx_csr);
1355 		return;
1356 	}
1357 
1358 	if (is_dma_capable() && dma && !status) {
1359 		/*
1360 		 * DMA has completed.  But if we're using DMA mode 1 (multi
1361 		 * packet DMA), we need a terminal TXPKTRDY interrupt before
1362 		 * we can consider this transfer completed, lest we trash
1363 		 * its last packet when writing the next URB's data.  So we
1364 		 * switch back to mode 0 to get that interrupt; we'll come
1365 		 * back here once it happens.
1366 		 */
1367 		if (tx_csr & MUSB_TXCSR_DMAMODE) {
1368 			/*
1369 			 * We shouldn't clear DMAMODE with DMAENAB set; so
1370 			 * clear them in a safe order.  That should be OK
1371 			 * once TXPKTRDY has been set (and I've never seen
1372 			 * it being 0 at this moment -- DMA interrupt latency
1373 			 * is significant) but if it hasn't been then we have
1374 			 * no choice but to stop being polite and ignore the
1375 			 * programmer's guide... :-)
1376 			 *
1377 			 * Note that we must write TXCSR with TXPKTRDY cleared
1378 			 * in order not to re-trigger the packet send (this bit
1379 			 * can't be cleared by CPU), and there's another caveat:
1380 			 * TXPKTRDY may be set shortly and then cleared in the
1381 			 * double-buffered FIFO mode, so we do an extra TXCSR
1382 			 * read for debouncing...
1383 			 */
1384 			tx_csr &= musb_readw(epio, MUSB_TXCSR);
1385 			if (tx_csr & MUSB_TXCSR_TXPKTRDY) {
1386 				tx_csr &= ~(MUSB_TXCSR_DMAENAB |
1387 					    MUSB_TXCSR_TXPKTRDY);
1388 				musb_writew(epio, MUSB_TXCSR,
1389 					    tx_csr | MUSB_TXCSR_H_WZC_BITS);
1390 			}
1391 			tx_csr &= ~(MUSB_TXCSR_DMAMODE |
1392 				    MUSB_TXCSR_TXPKTRDY);
1393 			musb_writew(epio, MUSB_TXCSR,
1394 				    tx_csr | MUSB_TXCSR_H_WZC_BITS);
1395 
1396 			/*
1397 			 * There is no guarantee that we'll get an interrupt
1398 			 * after clearing DMAMODE as we might have done this
1399 			 * too late (after TXPKTRDY was cleared by controller).
1400 			 * Re-read TXCSR as we have spoiled its previous value.
1401 			 */
1402 			tx_csr = musb_readw(epio, MUSB_TXCSR);
1403 		}
1404 
1405 		/*
1406 		 * We may get here from a DMA completion or TXPKTRDY interrupt.
1407 		 * In any case, we must check the FIFO status here and bail out
1408 		 * only if the FIFO still has data -- that should prevent the
1409 		 * "missed" TXPKTRDY interrupts and deal with double-buffered
1410 		 * FIFO mode too...
1411 		 */
1412 		if (tx_csr & (MUSB_TXCSR_FIFONOTEMPTY | MUSB_TXCSR_TXPKTRDY)) {
1413 			musb_dbg(musb,
1414 				"DMA complete but FIFO not empty, CSR %04x",
1415 				tx_csr);
1416 			return;
1417 		}
1418 	}
1419 
1420 	if (!status || dma || usb_pipeisoc(pipe)) {
1421 		if (dma)
1422 			length = dma->actual_len;
1423 		else
1424 			length = qh->segsize;
1425 		qh->offset += length;
1426 
1427 		if (usb_pipeisoc(pipe)) {
1428 			struct usb_iso_packet_descriptor	*d;
1429 
1430 			d = urb->iso_frame_desc + qh->iso_idx;
1431 			d->actual_length = length;
1432 			d->status = status;
1433 			if (++qh->iso_idx >= urb->number_of_packets) {
1434 				done = true;
1435 			} else {
1436 				d++;
1437 				offset = d->offset;
1438 				length = d->length;
1439 			}
1440 		} else if (dma && urb->transfer_buffer_length == qh->offset) {
1441 			done = true;
1442 		} else {
1443 			/* see if we need to send more data, or ZLP */
1444 			if (qh->segsize < qh->maxpacket)
1445 				done = true;
1446 			else if (qh->offset == urb->transfer_buffer_length
1447 					&& !(urb->transfer_flags
1448 						& URB_ZERO_PACKET))
1449 				done = true;
1450 			if (!done) {
1451 				offset = qh->offset;
1452 				length = urb->transfer_buffer_length - offset;
1453 				transfer_pending = true;
1454 			}
1455 		}
1456 	}
1457 
1458 	/* urb->status != -EINPROGRESS means request has been faulted,
1459 	 * so we must abort this transfer after cleanup
1460 	 */
1461 	if (urb->status != -EINPROGRESS) {
1462 		done = true;
1463 		if (status == 0)
1464 			status = urb->status;
1465 	}
1466 
1467 	if (done) {
1468 		/* set status */
1469 		urb->status = status;
1470 		urb->actual_length = qh->offset;
1471 		musb_advance_schedule(musb, urb, hw_ep, USB_DIR_OUT);
1472 		return;
1473 	} else if ((usb_pipeisoc(pipe) || transfer_pending) && dma) {
1474 		if (musb_tx_dma_program(musb->dma_controller, hw_ep, qh, urb,
1475 				offset, length)) {
1476 			if (is_cppi_enabled(musb) || tusb_dma_omap(musb))
1477 				musb_h_tx_dma_start(hw_ep);
1478 			return;
1479 		}
1480 	} else	if (tx_csr & MUSB_TXCSR_DMAENAB) {
1481 		musb_dbg(musb, "not complete, but DMA enabled?");
1482 		return;
1483 	}
1484 
1485 	/*
1486 	 * PIO: start next packet in this URB.
1487 	 *
1488 	 * REVISIT: some docs say that when hw_ep->tx_double_buffered,
1489 	 * (and presumably, FIFO is not half-full) we should write *two*
1490 	 * packets before updating TXCSR; other docs disagree...
1491 	 */
1492 	if (length > qh->maxpacket)
1493 		length = qh->maxpacket;
1494 	/* Unmap the buffer so that CPU can use it */
1495 	usb_hcd_unmap_urb_for_dma(musb->hcd, urb);
1496 
1497 	/*
1498 	 * We need to map sg if the transfer_buffer is
1499 	 * NULL.
1500 	 */
1501 	if (!urb->transfer_buffer)
1502 		qh->use_sg = true;
1503 
1504 	if (qh->use_sg) {
1505 		/* sg_miter_start is already done in musb_ep_program */
1506 		if (!sg_miter_next(&qh->sg_miter)) {
1507 			dev_err(musb->controller, "error: sg list empty\n");
1508 			sg_miter_stop(&qh->sg_miter);
1509 			status = -EINVAL;
1510 			goto done;
1511 		}
1512 		urb->transfer_buffer = qh->sg_miter.addr;
1513 		length = min_t(u32, length, qh->sg_miter.length);
1514 		musb_write_fifo(hw_ep, length, urb->transfer_buffer);
1515 		qh->sg_miter.consumed = length;
1516 		sg_miter_stop(&qh->sg_miter);
1517 	} else {
1518 		musb_write_fifo(hw_ep, length, urb->transfer_buffer + offset);
1519 	}
1520 
1521 	qh->segsize = length;
1522 
1523 	if (qh->use_sg) {
1524 		if (offset + length >= urb->transfer_buffer_length)
1525 			qh->use_sg = false;
1526 	}
1527 
1528 	musb_ep_select(mbase, epnum);
1529 	musb_writew(epio, MUSB_TXCSR,
1530 			MUSB_TXCSR_H_WZC_BITS | MUSB_TXCSR_TXPKTRDY);
1531 }
1532 
1533 #ifdef CONFIG_USB_TI_CPPI41_DMA
1534 /* Seems to set up ISO for cppi41 and not advance len. See commit c57c41d */
1535 static int musb_rx_dma_iso_cppi41(struct dma_controller *dma,
1536 				  struct musb_hw_ep *hw_ep,
1537 				  struct musb_qh *qh,
1538 				  struct urb *urb,
1539 				  size_t len)
1540 {
1541 	struct dma_channel *channel = hw_ep->rx_channel;
1542 	void __iomem *epio = hw_ep->regs;
1543 	dma_addr_t *buf;
1544 	u32 length;
1545 	u16 val;
1546 
1547 	buf = (void *)urb->iso_frame_desc[qh->iso_idx].offset +
1548 		(u32)urb->transfer_dma;
1549 
1550 	length = urb->iso_frame_desc[qh->iso_idx].length;
1551 
1552 	val = musb_readw(epio, MUSB_RXCSR);
1553 	val |= MUSB_RXCSR_DMAENAB;
1554 	musb_writew(hw_ep->regs, MUSB_RXCSR, val);
1555 
1556 	return dma->channel_program(channel, qh->maxpacket, 0,
1557 				   (u32)buf, length);
1558 }
1559 #else
1560 static inline int musb_rx_dma_iso_cppi41(struct dma_controller *dma,
1561 					 struct musb_hw_ep *hw_ep,
1562 					 struct musb_qh *qh,
1563 					 struct urb *urb,
1564 					 size_t len)
1565 {
1566 	return false;
1567 }
1568 #endif
1569 
1570 #if defined(CONFIG_USB_INVENTRA_DMA) || defined(CONFIG_USB_UX500_DMA) || \
1571 	defined(CONFIG_USB_TI_CPPI41_DMA)
1572 /* Host side RX (IN) using Mentor DMA works as follows:
1573 	submit_urb ->
1574 		- if queue was empty, ProgramEndpoint
1575 		- first IN token is sent out (by setting ReqPkt)
1576 	LinuxIsr -> RxReady()
1577 	/\	=> first packet is received
1578 	|	- Set in mode 0 (DmaEnab, ~ReqPkt)
1579 	|		-> DMA Isr (transfer complete) -> RxReady()
1580 	|		    - Ack receive (~RxPktRdy), turn off DMA (~DmaEnab)
1581 	|		    - if urb not complete, send next IN token (ReqPkt)
1582 	|			   |		else complete urb.
1583 	|			   |
1584 	---------------------------
1585  *
1586  * Nuances of mode 1:
1587  *	For short packets, no ack (+RxPktRdy) is sent automatically
1588  *	(even if AutoClear is ON)
1589  *	For full packets, ack (~RxPktRdy) and next IN token (+ReqPkt) is sent
1590  *	automatically => major problem, as collecting the next packet becomes
1591  *	difficult. Hence mode 1 is not used.
1592  *
1593  * REVISIT
1594  *	All we care about at this driver level is that
1595  *       (a) all URBs terminate with REQPKT cleared and fifo(s) empty;
1596  *       (b) termination conditions are: short RX, or buffer full;
1597  *       (c) fault modes include
1598  *           - iff URB_SHORT_NOT_OK, short RX status is -EREMOTEIO.
1599  *             (and that endpoint's dma queue stops immediately)
1600  *           - overflow (full, PLUS more bytes in the terminal packet)
1601  *
1602  *	So for example, usb-storage sets URB_SHORT_NOT_OK, and would
1603  *	thus be a great candidate for using mode 1 ... for all but the
1604  *	last packet of one URB's transfer.
1605  */
1606 static int musb_rx_dma_inventra_cppi41(struct dma_controller *dma,
1607 				       struct musb_hw_ep *hw_ep,
1608 				       struct musb_qh *qh,
1609 				       struct urb *urb,
1610 				       size_t len)
1611 {
1612 	struct dma_channel *channel = hw_ep->rx_channel;
1613 	void __iomem *epio = hw_ep->regs;
1614 	u16 val;
1615 	int pipe;
1616 	bool done;
1617 
1618 	pipe = urb->pipe;
1619 
1620 	if (usb_pipeisoc(pipe)) {
1621 		struct usb_iso_packet_descriptor *d;
1622 
1623 		d = urb->iso_frame_desc + qh->iso_idx;
1624 		d->actual_length = len;
1625 
1626 		/* even if there was an error, we did the dma
1627 		 * for iso_frame_desc->length
1628 		 */
1629 		if (d->status != -EILSEQ && d->status != -EOVERFLOW)
1630 			d->status = 0;
1631 
1632 		if (++qh->iso_idx >= urb->number_of_packets) {
1633 			done = true;
1634 		} else {
1635 			/* REVISIT: Why ignore return value here? */
1636 			if (musb_dma_cppi41(hw_ep->musb))
1637 				done = musb_rx_dma_iso_cppi41(dma, hw_ep, qh,
1638 							      urb, len);
1639 			done = false;
1640 		}
1641 
1642 	} else  {
1643 		/* done if urb buffer is full or short packet is recd */
1644 		done = (urb->actual_length + len >=
1645 			urb->transfer_buffer_length
1646 			|| channel->actual_len < qh->maxpacket
1647 			|| channel->rx_packet_done);
1648 	}
1649 
1650 	/* send IN token for next packet, without AUTOREQ */
1651 	if (!done) {
1652 		val = musb_readw(epio, MUSB_RXCSR);
1653 		val |= MUSB_RXCSR_H_REQPKT;
1654 		musb_writew(epio, MUSB_RXCSR, MUSB_RXCSR_H_WZC_BITS | val);
1655 	}
1656 
1657 	return done;
1658 }
1659 
1660 /* Disadvantage of using mode 1:
1661  *	It's basically usable only for mass storage class; essentially all
1662  *	other protocols also terminate transfers on short packets.
1663  *
1664  * Details:
1665  *	An extra IN token is sent at the end of the transfer (due to AUTOREQ)
1666  *	If you try to use mode 1 for (transfer_buffer_length - 512), and try
1667  *	to use the extra IN token to grab the last packet using mode 0, then
1668  *	the problem is that you cannot be sure when the device will send the
1669  *	last packet and RxPktRdy set. Sometimes the packet is recd too soon
1670  *	such that it gets lost when RxCSR is re-set at the end of the mode 1
1671  *	transfer, while sometimes it is recd just a little late so that if you
1672  *	try to configure for mode 0 soon after the mode 1 transfer is
1673  *	completed, you will find rxcount 0. Okay, so you might think why not
1674  *	wait for an interrupt when the pkt is recd. Well, you won't get any!
1675  */
1676 static int musb_rx_dma_in_inventra_cppi41(struct dma_controller *dma,
1677 					  struct musb_hw_ep *hw_ep,
1678 					  struct musb_qh *qh,
1679 					  struct urb *urb,
1680 					  size_t len,
1681 					  u8 iso_err)
1682 {
1683 	struct musb *musb = hw_ep->musb;
1684 	void __iomem *epio = hw_ep->regs;
1685 	struct dma_channel *channel = hw_ep->rx_channel;
1686 	u16 rx_count, val;
1687 	int length, pipe, done;
1688 	dma_addr_t buf;
1689 
1690 	rx_count = musb_readw(epio, MUSB_RXCOUNT);
1691 	pipe = urb->pipe;
1692 
1693 	if (usb_pipeisoc(pipe)) {
1694 		int d_status = 0;
1695 		struct usb_iso_packet_descriptor *d;
1696 
1697 		d = urb->iso_frame_desc + qh->iso_idx;
1698 
1699 		if (iso_err) {
1700 			d_status = -EILSEQ;
1701 			urb->error_count++;
1702 		}
1703 		if (rx_count > d->length) {
1704 			if (d_status == 0) {
1705 				d_status = -EOVERFLOW;
1706 				urb->error_count++;
1707 			}
1708 			musb_dbg(musb, "** OVERFLOW %d into %d",
1709 				rx_count, d->length);
1710 
1711 			length = d->length;
1712 		} else
1713 			length = rx_count;
1714 		d->status = d_status;
1715 		buf = urb->transfer_dma + d->offset;
1716 	} else {
1717 		length = rx_count;
1718 		buf = urb->transfer_dma + urb->actual_length;
1719 	}
1720 
1721 	channel->desired_mode = 0;
1722 #ifdef USE_MODE1
1723 	/* because of the issue below, mode 1 will
1724 	 * only rarely behave with correct semantics.
1725 	 */
1726 	if ((urb->transfer_flags & URB_SHORT_NOT_OK)
1727 	    && (urb->transfer_buffer_length - urb->actual_length)
1728 	    > qh->maxpacket)
1729 		channel->desired_mode = 1;
1730 	if (rx_count < hw_ep->max_packet_sz_rx) {
1731 		length = rx_count;
1732 		channel->desired_mode = 0;
1733 	} else {
1734 		length = urb->transfer_buffer_length;
1735 	}
1736 #endif
1737 
1738 	/* See comments above on disadvantages of using mode 1 */
1739 	val = musb_readw(epio, MUSB_RXCSR);
1740 	val &= ~MUSB_RXCSR_H_REQPKT;
1741 
1742 	if (channel->desired_mode == 0)
1743 		val &= ~MUSB_RXCSR_H_AUTOREQ;
1744 	else
1745 		val |= MUSB_RXCSR_H_AUTOREQ;
1746 	val |= MUSB_RXCSR_DMAENAB;
1747 
1748 	/* autoclear shouldn't be set in high bandwidth */
1749 	if (qh->hb_mult == 1)
1750 		val |= MUSB_RXCSR_AUTOCLEAR;
1751 
1752 	musb_writew(epio, MUSB_RXCSR, MUSB_RXCSR_H_WZC_BITS | val);
1753 
1754 	/* REVISIT if when actual_length != 0,
1755 	 * transfer_buffer_length needs to be
1756 	 * adjusted first...
1757 	 */
1758 	done = dma->channel_program(channel, qh->maxpacket,
1759 				   channel->desired_mode,
1760 				   buf, length);
1761 
1762 	if (!done) {
1763 		dma->channel_release(channel);
1764 		hw_ep->rx_channel = NULL;
1765 		channel = NULL;
1766 		val = musb_readw(epio, MUSB_RXCSR);
1767 		val &= ~(MUSB_RXCSR_DMAENAB
1768 			 | MUSB_RXCSR_H_AUTOREQ
1769 			 | MUSB_RXCSR_AUTOCLEAR);
1770 		musb_writew(epio, MUSB_RXCSR, val);
1771 	}
1772 
1773 	return done;
1774 }
1775 #else
1776 static inline int musb_rx_dma_inventra_cppi41(struct dma_controller *dma,
1777 					      struct musb_hw_ep *hw_ep,
1778 					      struct musb_qh *qh,
1779 					      struct urb *urb,
1780 					      size_t len)
1781 {
1782 	return false;
1783 }
1784 
1785 static inline int musb_rx_dma_in_inventra_cppi41(struct dma_controller *dma,
1786 						 struct musb_hw_ep *hw_ep,
1787 						 struct musb_qh *qh,
1788 						 struct urb *urb,
1789 						 size_t len,
1790 						 u8 iso_err)
1791 {
1792 	return false;
1793 }
1794 #endif
1795 
1796 /*
1797  * Service an RX interrupt for the given IN endpoint; docs cover bulk, iso,
1798  * and high-bandwidth IN transfer cases.
1799  */
1800 void musb_host_rx(struct musb *musb, u8 epnum)
1801 {
1802 	struct urb		*urb;
1803 	struct musb_hw_ep	*hw_ep = musb->endpoints + epnum;
1804 	struct dma_controller	*c = musb->dma_controller;
1805 	void __iomem		*epio = hw_ep->regs;
1806 	struct musb_qh		*qh = hw_ep->in_qh;
1807 	size_t			xfer_len;
1808 	void __iomem		*mbase = musb->mregs;
1809 	int			pipe;
1810 	u16			rx_csr, val;
1811 	bool			iso_err = false;
1812 	bool			done = false;
1813 	u32			status;
1814 	struct dma_channel	*dma;
1815 	unsigned int sg_flags = SG_MITER_ATOMIC | SG_MITER_TO_SG;
1816 
1817 	musb_ep_select(mbase, epnum);
1818 
1819 	urb = next_urb(qh);
1820 	dma = is_dma_capable() ? hw_ep->rx_channel : NULL;
1821 	status = 0;
1822 	xfer_len = 0;
1823 
1824 	rx_csr = musb_readw(epio, MUSB_RXCSR);
1825 	val = rx_csr;
1826 
1827 	if (unlikely(!urb)) {
1828 		/* REVISIT -- THIS SHOULD NEVER HAPPEN ... but, at least
1829 		 * usbtest #11 (unlinks) triggers it regularly, sometimes
1830 		 * with fifo full.  (Only with DMA??)
1831 		 */
1832 		musb_dbg(musb, "BOGUS RX%d ready, csr %04x, count %d",
1833 			epnum, val, musb_readw(epio, MUSB_RXCOUNT));
1834 		musb_h_flush_rxfifo(hw_ep, MUSB_RXCSR_CLRDATATOG);
1835 		return;
1836 	}
1837 
1838 	pipe = urb->pipe;
1839 
1840 	trace_musb_urb_rx(musb, urb);
1841 
1842 	/* check for errors, concurrent stall & unlink is not really
1843 	 * handled yet! */
1844 	if (rx_csr & MUSB_RXCSR_H_RXSTALL) {
1845 		musb_dbg(musb, "RX end %d STALL", epnum);
1846 
1847 		/* stall; record URB status */
1848 		status = -EPIPE;
1849 
1850 	} else if (rx_csr & MUSB_RXCSR_H_ERROR) {
1851 		musb_dbg(musb, "end %d RX proto error", epnum);
1852 
1853 		status = -EPROTO;
1854 		musb_writeb(epio, MUSB_RXINTERVAL, 0);
1855 
1856 		rx_csr &= ~MUSB_RXCSR_H_ERROR;
1857 		musb_writew(epio, MUSB_RXCSR, rx_csr);
1858 
1859 	} else if (rx_csr & MUSB_RXCSR_DATAERROR) {
1860 
1861 		if (USB_ENDPOINT_XFER_ISOC != qh->type) {
1862 			musb_dbg(musb, "RX end %d NAK timeout", epnum);
1863 
1864 			/* NOTE: NAKing is *NOT* an error, so we want to
1865 			 * continue.  Except ... if there's a request for
1866 			 * another QH, use that instead of starving it.
1867 			 *
1868 			 * Devices like Ethernet and serial adapters keep
1869 			 * reads posted at all times, which will starve
1870 			 * other devices without this logic.
1871 			 */
1872 			if (usb_pipebulk(urb->pipe)
1873 					&& qh->mux == 1
1874 					&& !list_is_singular(&musb->in_bulk)) {
1875 				musb_bulk_nak_timeout(musb, hw_ep, 1);
1876 				return;
1877 			}
1878 			musb_ep_select(mbase, epnum);
1879 			rx_csr |= MUSB_RXCSR_H_WZC_BITS;
1880 			rx_csr &= ~MUSB_RXCSR_DATAERROR;
1881 			musb_writew(epio, MUSB_RXCSR, rx_csr);
1882 
1883 			goto finish;
1884 		} else {
1885 			musb_dbg(musb, "RX end %d ISO data error", epnum);
1886 			/* packet error reported later */
1887 			iso_err = true;
1888 		}
1889 	} else if (rx_csr & MUSB_RXCSR_INCOMPRX) {
1890 		musb_dbg(musb, "end %d high bandwidth incomplete ISO packet RX",
1891 				epnum);
1892 		status = -EPROTO;
1893 	}
1894 
1895 	/* faults abort the transfer */
1896 	if (status) {
1897 		/* clean up dma and collect transfer count */
1898 		if (dma_channel_status(dma) == MUSB_DMA_STATUS_BUSY) {
1899 			dma->status = MUSB_DMA_STATUS_CORE_ABORT;
1900 			musb->dma_controller->channel_abort(dma);
1901 			xfer_len = dma->actual_len;
1902 		}
1903 		musb_h_flush_rxfifo(hw_ep, MUSB_RXCSR_CLRDATATOG);
1904 		musb_writeb(epio, MUSB_RXINTERVAL, 0);
1905 		done = true;
1906 		goto finish;
1907 	}
1908 
1909 	if (unlikely(dma_channel_status(dma) == MUSB_DMA_STATUS_BUSY)) {
1910 		/* SHOULD NEVER HAPPEN ... but at least DaVinci has done it */
1911 		ERR("RX%d dma busy, csr %04x\n", epnum, rx_csr);
1912 		goto finish;
1913 	}
1914 
1915 	/* thorough shutdown for now ... given more precise fault handling
1916 	 * and better queueing support, we might keep a DMA pipeline going
1917 	 * while processing this irq for earlier completions.
1918 	 */
1919 
1920 	/* FIXME this is _way_ too much in-line logic for Mentor DMA */
1921 	if (!musb_dma_inventra(musb) && !musb_dma_ux500(musb) &&
1922 	    (rx_csr & MUSB_RXCSR_H_REQPKT)) {
1923 		/* REVISIT this happened for a while on some short reads...
1924 		 * the cleanup still needs investigation... looks bad...
1925 		 * and also duplicates dma cleanup code above ... plus,
1926 		 * shouldn't this be the "half full" double buffer case?
1927 		 */
1928 		if (dma_channel_status(dma) == MUSB_DMA_STATUS_BUSY) {
1929 			dma->status = MUSB_DMA_STATUS_CORE_ABORT;
1930 			musb->dma_controller->channel_abort(dma);
1931 			xfer_len = dma->actual_len;
1932 			done = true;
1933 		}
1934 
1935 		musb_dbg(musb, "RXCSR%d %04x, reqpkt, len %zu%s", epnum, rx_csr,
1936 				xfer_len, dma ? ", dma" : "");
1937 		rx_csr &= ~MUSB_RXCSR_H_REQPKT;
1938 
1939 		musb_ep_select(mbase, epnum);
1940 		musb_writew(epio, MUSB_RXCSR,
1941 				MUSB_RXCSR_H_WZC_BITS | rx_csr);
1942 	}
1943 
1944 	if (dma && (rx_csr & MUSB_RXCSR_DMAENAB)) {
1945 		xfer_len = dma->actual_len;
1946 
1947 		val &= ~(MUSB_RXCSR_DMAENAB
1948 			| MUSB_RXCSR_H_AUTOREQ
1949 			| MUSB_RXCSR_AUTOCLEAR
1950 			| MUSB_RXCSR_RXPKTRDY);
1951 		musb_writew(hw_ep->regs, MUSB_RXCSR, val);
1952 
1953 		if (musb_dma_inventra(musb) || musb_dma_ux500(musb) ||
1954 		    musb_dma_cppi41(musb)) {
1955 			    done = musb_rx_dma_inventra_cppi41(c, hw_ep, qh, urb, xfer_len);
1956 			    musb_dbg(hw_ep->musb,
1957 				    "ep %d dma %s, rxcsr %04x, rxcount %d",
1958 				    epnum, done ? "off" : "reset",
1959 				    musb_readw(epio, MUSB_RXCSR),
1960 				    musb_readw(epio, MUSB_RXCOUNT));
1961 		} else {
1962 			done = true;
1963 		}
1964 
1965 	} else if (urb->status == -EINPROGRESS) {
1966 		/* if no errors, be sure a packet is ready for unloading */
1967 		if (unlikely(!(rx_csr & MUSB_RXCSR_RXPKTRDY))) {
1968 			status = -EPROTO;
1969 			ERR("Rx interrupt with no errors or packet!\n");
1970 
1971 			/* FIXME this is another "SHOULD NEVER HAPPEN" */
1972 
1973 /* SCRUB (RX) */
1974 			/* do the proper sequence to abort the transfer */
1975 			musb_ep_select(mbase, epnum);
1976 			val &= ~MUSB_RXCSR_H_REQPKT;
1977 			musb_writew(epio, MUSB_RXCSR, val);
1978 			goto finish;
1979 		}
1980 
1981 		/* we are expecting IN packets */
1982 		if ((musb_dma_inventra(musb) || musb_dma_ux500(musb) ||
1983 		    musb_dma_cppi41(musb)) && dma) {
1984 			musb_dbg(hw_ep->musb,
1985 				"RX%d count %d, buffer 0x%llx len %d/%d",
1986 				epnum, musb_readw(epio, MUSB_RXCOUNT),
1987 				(unsigned long long) urb->transfer_dma
1988 				+ urb->actual_length,
1989 				qh->offset,
1990 				urb->transfer_buffer_length);
1991 
1992 			if (musb_rx_dma_in_inventra_cppi41(c, hw_ep, qh, urb,
1993 							   xfer_len, iso_err))
1994 				goto finish;
1995 			else
1996 				dev_err(musb->controller, "error: rx_dma failed\n");
1997 		}
1998 
1999 		if (!dma) {
2000 			unsigned int received_len;
2001 
2002 			/* Unmap the buffer so that CPU can use it */
2003 			usb_hcd_unmap_urb_for_dma(musb->hcd, urb);
2004 
2005 			/*
2006 			 * We need to map sg if the transfer_buffer is
2007 			 * NULL.
2008 			 */
2009 			if (!urb->transfer_buffer) {
2010 				qh->use_sg = true;
2011 				sg_miter_start(&qh->sg_miter, urb->sg, 1,
2012 						sg_flags);
2013 			}
2014 
2015 			if (qh->use_sg) {
2016 				if (!sg_miter_next(&qh->sg_miter)) {
2017 					dev_err(musb->controller, "error: sg list empty\n");
2018 					sg_miter_stop(&qh->sg_miter);
2019 					status = -EINVAL;
2020 					done = true;
2021 					goto finish;
2022 				}
2023 				urb->transfer_buffer = qh->sg_miter.addr;
2024 				received_len = urb->actual_length;
2025 				qh->offset = 0x0;
2026 				done = musb_host_packet_rx(musb, urb, epnum,
2027 						iso_err);
2028 				/* Calculate the number of bytes received */
2029 				received_len = urb->actual_length -
2030 					received_len;
2031 				qh->sg_miter.consumed = received_len;
2032 				sg_miter_stop(&qh->sg_miter);
2033 			} else {
2034 				done = musb_host_packet_rx(musb, urb,
2035 						epnum, iso_err);
2036 			}
2037 			musb_dbg(musb, "read %spacket", done ? "last " : "");
2038 		}
2039 	}
2040 
2041 finish:
2042 	urb->actual_length += xfer_len;
2043 	qh->offset += xfer_len;
2044 	if (done) {
2045 		if (qh->use_sg)
2046 			qh->use_sg = false;
2047 
2048 		if (urb->status == -EINPROGRESS)
2049 			urb->status = status;
2050 		musb_advance_schedule(musb, urb, hw_ep, USB_DIR_IN);
2051 	}
2052 }
2053 
2054 /* schedule nodes correspond to peripheral endpoints, like an OHCI QH.
2055  * the software schedule associates multiple such nodes with a given
2056  * host side hardware endpoint + direction; scheduling may activate
2057  * that hardware endpoint.
2058  */
2059 static int musb_schedule(
2060 	struct musb		*musb,
2061 	struct musb_qh		*qh,
2062 	int			is_in)
2063 {
2064 	int			idle = 0;
2065 	int			best_diff;
2066 	int			best_end, epnum;
2067 	struct musb_hw_ep	*hw_ep = NULL;
2068 	struct list_head	*head = NULL;
2069 	u8			toggle;
2070 	u8			txtype;
2071 	struct urb		*urb = next_urb(qh);
2072 
2073 	/* use fixed hardware for control and bulk */
2074 	if (qh->type == USB_ENDPOINT_XFER_CONTROL) {
2075 		head = &musb->control;
2076 		hw_ep = musb->control_ep;
2077 		goto success;
2078 	}
2079 
2080 	/* else, periodic transfers get muxed to other endpoints */
2081 
2082 	/*
2083 	 * We know this qh hasn't been scheduled, so all we need to do
2084 	 * is choose which hardware endpoint to put it on ...
2085 	 *
2086 	 * REVISIT what we really want here is a regular schedule tree
2087 	 * like e.g. OHCI uses.
2088 	 */
2089 	best_diff = 4096;
2090 	best_end = -1;
2091 
2092 	for (epnum = 1, hw_ep = musb->endpoints + 1;
2093 			epnum < musb->nr_endpoints;
2094 			epnum++, hw_ep++) {
2095 		int	diff;
2096 
2097 		if (musb_ep_get_qh(hw_ep, is_in) != NULL)
2098 			continue;
2099 
2100 		if (hw_ep == musb->bulk_ep)
2101 			continue;
2102 
2103 		if (is_in)
2104 			diff = hw_ep->max_packet_sz_rx;
2105 		else
2106 			diff = hw_ep->max_packet_sz_tx;
2107 		diff -= (qh->maxpacket * qh->hb_mult);
2108 
2109 		if (diff >= 0 && best_diff > diff) {
2110 
2111 			/*
2112 			 * Mentor controller has a bug in that if we schedule
2113 			 * a BULK Tx transfer on an endpoint that had earlier
2114 			 * handled ISOC then the BULK transfer has to start on
2115 			 * a zero toggle.  If the BULK transfer starts on a 1
2116 			 * toggle then this transfer will fail as the mentor
2117 			 * controller starts the Bulk transfer on a 0 toggle
2118 			 * irrespective of the programming of the toggle bits
2119 			 * in the TXCSR register.  Check for this condition
2120 			 * while allocating the EP for a Tx Bulk transfer.  If
2121 			 * so skip this EP.
2122 			 */
2123 			hw_ep = musb->endpoints + epnum;
2124 			toggle = usb_gettoggle(urb->dev, qh->epnum, !is_in);
2125 			txtype = (musb_readb(hw_ep->regs, MUSB_TXTYPE)
2126 					>> 4) & 0x3;
2127 			if (!is_in && (qh->type == USB_ENDPOINT_XFER_BULK) &&
2128 				toggle && (txtype == USB_ENDPOINT_XFER_ISOC))
2129 				continue;
2130 
2131 			best_diff = diff;
2132 			best_end = epnum;
2133 		}
2134 	}
2135 	/* use bulk reserved ep1 if no other ep is free */
2136 	if (best_end < 0 && qh->type == USB_ENDPOINT_XFER_BULK) {
2137 		hw_ep = musb->bulk_ep;
2138 		if (is_in)
2139 			head = &musb->in_bulk;
2140 		else
2141 			head = &musb->out_bulk;
2142 
2143 		/* Enable bulk RX/TX NAK timeout scheme when bulk requests are
2144 		 * multiplexed. This scheme does not work in high speed to full
2145 		 * speed scenario as NAK interrupts are not coming from a
2146 		 * full speed device connected to a high speed device.
2147 		 * NAK timeout interval is 8 (128 uframe or 16ms) for HS and
2148 		 * 4 (8 frame or 8ms) for FS device.
2149 		 */
2150 		if (qh->dev)
2151 			qh->intv_reg =
2152 				(USB_SPEED_HIGH == qh->dev->speed) ? 8 : 4;
2153 		goto success;
2154 	} else if (best_end < 0) {
2155 		return -ENOSPC;
2156 	}
2157 
2158 	idle = 1;
2159 	qh->mux = 0;
2160 	hw_ep = musb->endpoints + best_end;
2161 	musb_dbg(musb, "qh %p periodic slot %d", qh, best_end);
2162 success:
2163 	if (head) {
2164 		idle = list_empty(head);
2165 		list_add_tail(&qh->ring, head);
2166 		qh->mux = 1;
2167 	}
2168 	qh->hw_ep = hw_ep;
2169 	qh->hep->hcpriv = qh;
2170 	if (idle)
2171 		musb_start_urb(musb, is_in, qh);
2172 	return 0;
2173 }
2174 
2175 static int musb_urb_enqueue(
2176 	struct usb_hcd			*hcd,
2177 	struct urb			*urb,
2178 	gfp_t				mem_flags)
2179 {
2180 	unsigned long			flags;
2181 	struct musb			*musb = hcd_to_musb(hcd);
2182 	struct usb_host_endpoint	*hep = urb->ep;
2183 	struct musb_qh			*qh;
2184 	struct usb_endpoint_descriptor	*epd = &hep->desc;
2185 	int				ret;
2186 	unsigned			type_reg;
2187 	unsigned			interval;
2188 
2189 	/* host role must be active */
2190 	if (!is_host_active(musb) || !musb->is_active)
2191 		return -ENODEV;
2192 
2193 	trace_musb_urb_enq(musb, urb);
2194 
2195 	spin_lock_irqsave(&musb->lock, flags);
2196 	ret = usb_hcd_link_urb_to_ep(hcd, urb);
2197 	qh = ret ? NULL : hep->hcpriv;
2198 	if (qh)
2199 		urb->hcpriv = qh;
2200 	spin_unlock_irqrestore(&musb->lock, flags);
2201 
2202 	/* DMA mapping was already done, if needed, and this urb is on
2203 	 * hep->urb_list now ... so we're done, unless hep wasn't yet
2204 	 * scheduled onto a live qh.
2205 	 *
2206 	 * REVISIT best to keep hep->hcpriv valid until the endpoint gets
2207 	 * disabled, testing for empty qh->ring and avoiding qh setup costs
2208 	 * except for the first urb queued after a config change.
2209 	 */
2210 	if (qh || ret)
2211 		return ret;
2212 
2213 	/* Allocate and initialize qh, minimizing the work done each time
2214 	 * hw_ep gets reprogrammed, or with irqs blocked.  Then schedule it.
2215 	 *
2216 	 * REVISIT consider a dedicated qh kmem_cache, so it's harder
2217 	 * for bugs in other kernel code to break this driver...
2218 	 */
2219 	qh = kzalloc(sizeof *qh, mem_flags);
2220 	if (!qh) {
2221 		spin_lock_irqsave(&musb->lock, flags);
2222 		usb_hcd_unlink_urb_from_ep(hcd, urb);
2223 		spin_unlock_irqrestore(&musb->lock, flags);
2224 		return -ENOMEM;
2225 	}
2226 
2227 	qh->hep = hep;
2228 	qh->dev = urb->dev;
2229 	INIT_LIST_HEAD(&qh->ring);
2230 	qh->is_ready = 1;
2231 
2232 	qh->maxpacket = usb_endpoint_maxp(epd);
2233 	qh->type = usb_endpoint_type(epd);
2234 
2235 	/* Bits 11 & 12 of wMaxPacketSize encode high bandwidth multiplier.
2236 	 * Some musb cores don't support high bandwidth ISO transfers; and
2237 	 * we don't (yet!) support high bandwidth interrupt transfers.
2238 	 */
2239 	qh->hb_mult = usb_endpoint_maxp_mult(epd);
2240 	if (qh->hb_mult > 1) {
2241 		int ok = (qh->type == USB_ENDPOINT_XFER_ISOC);
2242 
2243 		if (ok)
2244 			ok = (usb_pipein(urb->pipe) && musb->hb_iso_rx)
2245 				|| (usb_pipeout(urb->pipe) && musb->hb_iso_tx);
2246 		if (!ok) {
2247 			ret = -EMSGSIZE;
2248 			goto done;
2249 		}
2250 		qh->maxpacket &= 0x7ff;
2251 	}
2252 
2253 	qh->epnum = usb_endpoint_num(epd);
2254 
2255 	/* NOTE: urb->dev->devnum is wrong during SET_ADDRESS */
2256 	qh->addr_reg = (u8) usb_pipedevice(urb->pipe);
2257 
2258 	/* precompute rxtype/txtype/type0 register */
2259 	type_reg = (qh->type << 4) | qh->epnum;
2260 	switch (urb->dev->speed) {
2261 	case USB_SPEED_LOW:
2262 		type_reg |= 0xc0;
2263 		break;
2264 	case USB_SPEED_FULL:
2265 		type_reg |= 0x80;
2266 		break;
2267 	default:
2268 		type_reg |= 0x40;
2269 	}
2270 	qh->type_reg = type_reg;
2271 
2272 	/* Precompute RXINTERVAL/TXINTERVAL register */
2273 	switch (qh->type) {
2274 	case USB_ENDPOINT_XFER_INT:
2275 		/*
2276 		 * Full/low speeds use the  linear encoding,
2277 		 * high speed uses the logarithmic encoding.
2278 		 */
2279 		if (urb->dev->speed <= USB_SPEED_FULL) {
2280 			interval = max_t(u8, epd->bInterval, 1);
2281 			break;
2282 		}
2283 		/* FALLTHROUGH */
2284 	case USB_ENDPOINT_XFER_ISOC:
2285 		/* ISO always uses logarithmic encoding */
2286 		interval = min_t(u8, epd->bInterval, 16);
2287 		break;
2288 	default:
2289 		/* REVISIT we actually want to use NAK limits, hinting to the
2290 		 * transfer scheduling logic to try some other qh, e.g. try
2291 		 * for 2 msec first:
2292 		 *
2293 		 * interval = (USB_SPEED_HIGH == urb->dev->speed) ? 16 : 2;
2294 		 *
2295 		 * The downside of disabling this is that transfer scheduling
2296 		 * gets VERY unfair for nonperiodic transfers; a misbehaving
2297 		 * peripheral could make that hurt.  That's perfectly normal
2298 		 * for reads from network or serial adapters ... so we have
2299 		 * partial NAKlimit support for bulk RX.
2300 		 *
2301 		 * The upside of disabling it is simpler transfer scheduling.
2302 		 */
2303 		interval = 0;
2304 	}
2305 	qh->intv_reg = interval;
2306 
2307 	/* precompute addressing for external hub/tt ports */
2308 	if (musb->is_multipoint) {
2309 		struct usb_device	*parent = urb->dev->parent;
2310 
2311 		if (parent != hcd->self.root_hub) {
2312 			qh->h_addr_reg = (u8) parent->devnum;
2313 
2314 			/* set up tt info if needed */
2315 			if (urb->dev->tt) {
2316 				qh->h_port_reg = (u8) urb->dev->ttport;
2317 				if (urb->dev->tt->hub)
2318 					qh->h_addr_reg =
2319 						(u8) urb->dev->tt->hub->devnum;
2320 				if (urb->dev->tt->multi)
2321 					qh->h_addr_reg |= 0x80;
2322 			}
2323 		}
2324 	}
2325 
2326 	/* invariant: hep->hcpriv is null OR the qh that's already scheduled.
2327 	 * until we get real dma queues (with an entry for each urb/buffer),
2328 	 * we only have work to do in the former case.
2329 	 */
2330 	spin_lock_irqsave(&musb->lock, flags);
2331 	if (hep->hcpriv || !next_urb(qh)) {
2332 		/* some concurrent activity submitted another urb to hep...
2333 		 * odd, rare, error prone, but legal.
2334 		 */
2335 		kfree(qh);
2336 		qh = NULL;
2337 		ret = 0;
2338 	} else
2339 		ret = musb_schedule(musb, qh,
2340 				epd->bEndpointAddress & USB_ENDPOINT_DIR_MASK);
2341 
2342 	if (ret == 0) {
2343 		urb->hcpriv = qh;
2344 		/* FIXME set urb->start_frame for iso/intr, it's tested in
2345 		 * musb_start_urb(), but otherwise only konicawc cares ...
2346 		 */
2347 	}
2348 	spin_unlock_irqrestore(&musb->lock, flags);
2349 
2350 done:
2351 	if (ret != 0) {
2352 		spin_lock_irqsave(&musb->lock, flags);
2353 		usb_hcd_unlink_urb_from_ep(hcd, urb);
2354 		spin_unlock_irqrestore(&musb->lock, flags);
2355 		kfree(qh);
2356 	}
2357 	return ret;
2358 }
2359 
2360 
2361 /*
2362  * abort a transfer that's at the head of a hardware queue.
2363  * called with controller locked, irqs blocked
2364  * that hardware queue advances to the next transfer, unless prevented
2365  */
2366 static int musb_cleanup_urb(struct urb *urb, struct musb_qh *qh)
2367 {
2368 	struct musb_hw_ep	*ep = qh->hw_ep;
2369 	struct musb		*musb = ep->musb;
2370 	void __iomem		*epio = ep->regs;
2371 	unsigned		hw_end = ep->epnum;
2372 	void __iomem		*regs = ep->musb->mregs;
2373 	int			is_in = usb_pipein(urb->pipe);
2374 	int			status = 0;
2375 	u16			csr;
2376 	struct dma_channel	*dma = NULL;
2377 
2378 	musb_ep_select(regs, hw_end);
2379 
2380 	if (is_dma_capable()) {
2381 		dma = is_in ? ep->rx_channel : ep->tx_channel;
2382 		if (dma) {
2383 			status = ep->musb->dma_controller->channel_abort(dma);
2384 			musb_dbg(musb, "abort %cX%d DMA for urb %p --> %d",
2385 				is_in ? 'R' : 'T', ep->epnum,
2386 				urb, status);
2387 			urb->actual_length += dma->actual_len;
2388 		}
2389 	}
2390 
2391 	/* turn off DMA requests, discard state, stop polling ... */
2392 	if (ep->epnum && is_in) {
2393 		/* giveback saves bulk toggle */
2394 		csr = musb_h_flush_rxfifo(ep, 0);
2395 
2396 		/* clear the endpoint's irq status here to avoid bogus irqs */
2397 		if (is_dma_capable() && dma)
2398 			musb_platform_clear_ep_rxintr(musb, ep->epnum);
2399 	} else if (ep->epnum) {
2400 		musb_h_tx_flush_fifo(ep);
2401 		csr = musb_readw(epio, MUSB_TXCSR);
2402 		csr &= ~(MUSB_TXCSR_AUTOSET
2403 			| MUSB_TXCSR_DMAENAB
2404 			| MUSB_TXCSR_H_RXSTALL
2405 			| MUSB_TXCSR_H_NAKTIMEOUT
2406 			| MUSB_TXCSR_H_ERROR
2407 			| MUSB_TXCSR_TXPKTRDY);
2408 		musb_writew(epio, MUSB_TXCSR, csr);
2409 		/* REVISIT may need to clear FLUSHFIFO ... */
2410 		musb_writew(epio, MUSB_TXCSR, csr);
2411 		/* flush cpu writebuffer */
2412 		csr = musb_readw(epio, MUSB_TXCSR);
2413 	} else  {
2414 		musb_h_ep0_flush_fifo(ep);
2415 	}
2416 	if (status == 0)
2417 		musb_advance_schedule(ep->musb, urb, ep, is_in);
2418 	return status;
2419 }
2420 
2421 static int musb_urb_dequeue(struct usb_hcd *hcd, struct urb *urb, int status)
2422 {
2423 	struct musb		*musb = hcd_to_musb(hcd);
2424 	struct musb_qh		*qh;
2425 	unsigned long		flags;
2426 	int			is_in  = usb_pipein(urb->pipe);
2427 	int			ret;
2428 
2429 	trace_musb_urb_deq(musb, urb);
2430 
2431 	spin_lock_irqsave(&musb->lock, flags);
2432 	ret = usb_hcd_check_unlink_urb(hcd, urb, status);
2433 	if (ret)
2434 		goto done;
2435 
2436 	qh = urb->hcpriv;
2437 	if (!qh)
2438 		goto done;
2439 
2440 	/*
2441 	 * Any URB not actively programmed into endpoint hardware can be
2442 	 * immediately given back; that's any URB not at the head of an
2443 	 * endpoint queue, unless someday we get real DMA queues.  And even
2444 	 * if it's at the head, it might not be known to the hardware...
2445 	 *
2446 	 * Otherwise abort current transfer, pending DMA, etc.; urb->status
2447 	 * has already been updated.  This is a synchronous abort; it'd be
2448 	 * OK to hold off until after some IRQ, though.
2449 	 *
2450 	 * NOTE: qh is invalid unless !list_empty(&hep->urb_list)
2451 	 */
2452 	if (!qh->is_ready
2453 			|| urb->urb_list.prev != &qh->hep->urb_list
2454 			|| musb_ep_get_qh(qh->hw_ep, is_in) != qh) {
2455 		int	ready = qh->is_ready;
2456 
2457 		qh->is_ready = 0;
2458 		musb_giveback(musb, urb, 0);
2459 		qh->is_ready = ready;
2460 
2461 		/* If nothing else (usually musb_giveback) is using it
2462 		 * and its URB list has emptied, recycle this qh.
2463 		 */
2464 		if (ready && list_empty(&qh->hep->urb_list)) {
2465 			qh->hep->hcpriv = NULL;
2466 			list_del(&qh->ring);
2467 			kfree(qh);
2468 		}
2469 	} else
2470 		ret = musb_cleanup_urb(urb, qh);
2471 done:
2472 	spin_unlock_irqrestore(&musb->lock, flags);
2473 	return ret;
2474 }
2475 
2476 /* disable an endpoint */
2477 static void
2478 musb_h_disable(struct usb_hcd *hcd, struct usb_host_endpoint *hep)
2479 {
2480 	u8			is_in = hep->desc.bEndpointAddress & USB_DIR_IN;
2481 	unsigned long		flags;
2482 	struct musb		*musb = hcd_to_musb(hcd);
2483 	struct musb_qh		*qh;
2484 	struct urb		*urb;
2485 
2486 	spin_lock_irqsave(&musb->lock, flags);
2487 
2488 	qh = hep->hcpriv;
2489 	if (qh == NULL)
2490 		goto exit;
2491 
2492 	/* NOTE: qh is invalid unless !list_empty(&hep->urb_list) */
2493 
2494 	/* Kick the first URB off the hardware, if needed */
2495 	qh->is_ready = 0;
2496 	if (musb_ep_get_qh(qh->hw_ep, is_in) == qh) {
2497 		urb = next_urb(qh);
2498 
2499 		/* make software (then hardware) stop ASAP */
2500 		if (!urb->unlinked)
2501 			urb->status = -ESHUTDOWN;
2502 
2503 		/* cleanup */
2504 		musb_cleanup_urb(urb, qh);
2505 
2506 		/* Then nuke all the others ... and advance the
2507 		 * queue on hw_ep (e.g. bulk ring) when we're done.
2508 		 */
2509 		while (!list_empty(&hep->urb_list)) {
2510 			urb = next_urb(qh);
2511 			urb->status = -ESHUTDOWN;
2512 			musb_advance_schedule(musb, urb, qh->hw_ep, is_in);
2513 		}
2514 	} else {
2515 		/* Just empty the queue; the hardware is busy with
2516 		 * other transfers, and since !qh->is_ready nothing
2517 		 * will activate any of these as it advances.
2518 		 */
2519 		while (!list_empty(&hep->urb_list))
2520 			musb_giveback(musb, next_urb(qh), -ESHUTDOWN);
2521 
2522 		hep->hcpriv = NULL;
2523 		list_del(&qh->ring);
2524 		kfree(qh);
2525 	}
2526 exit:
2527 	spin_unlock_irqrestore(&musb->lock, flags);
2528 }
2529 
2530 static int musb_h_get_frame_number(struct usb_hcd *hcd)
2531 {
2532 	struct musb	*musb = hcd_to_musb(hcd);
2533 
2534 	return musb_readw(musb->mregs, MUSB_FRAME);
2535 }
2536 
2537 static int musb_h_start(struct usb_hcd *hcd)
2538 {
2539 	struct musb	*musb = hcd_to_musb(hcd);
2540 
2541 	/* NOTE: musb_start() is called when the hub driver turns
2542 	 * on port power, or when (OTG) peripheral starts.
2543 	 */
2544 	hcd->state = HC_STATE_RUNNING;
2545 	musb->port1_status = 0;
2546 	return 0;
2547 }
2548 
2549 static void musb_h_stop(struct usb_hcd *hcd)
2550 {
2551 	musb_stop(hcd_to_musb(hcd));
2552 	hcd->state = HC_STATE_HALT;
2553 }
2554 
2555 static int musb_bus_suspend(struct usb_hcd *hcd)
2556 {
2557 	struct musb	*musb = hcd_to_musb(hcd);
2558 	u8		devctl;
2559 
2560 	musb_port_suspend(musb, true);
2561 
2562 	if (!is_host_active(musb))
2563 		return 0;
2564 
2565 	switch (musb->xceiv->otg->state) {
2566 	case OTG_STATE_A_SUSPEND:
2567 		return 0;
2568 	case OTG_STATE_A_WAIT_VRISE:
2569 		/* ID could be grounded even if there's no device
2570 		 * on the other end of the cable.  NOTE that the
2571 		 * A_WAIT_VRISE timers are messy with MUSB...
2572 		 */
2573 		devctl = musb_readb(musb->mregs, MUSB_DEVCTL);
2574 		if ((devctl & MUSB_DEVCTL_VBUS) == MUSB_DEVCTL_VBUS)
2575 			musb->xceiv->otg->state = OTG_STATE_A_WAIT_BCON;
2576 		break;
2577 	default:
2578 		break;
2579 	}
2580 
2581 	if (musb->is_active) {
2582 		WARNING("trying to suspend as %s while active\n",
2583 				usb_otg_state_string(musb->xceiv->otg->state));
2584 		return -EBUSY;
2585 	} else
2586 		return 0;
2587 }
2588 
2589 static int musb_bus_resume(struct usb_hcd *hcd)
2590 {
2591 	struct musb *musb = hcd_to_musb(hcd);
2592 
2593 	if (musb->config &&
2594 	    musb->config->host_port_deassert_reset_at_resume)
2595 		musb_port_reset(musb, false);
2596 
2597 	return 0;
2598 }
2599 
2600 #ifndef CONFIG_MUSB_PIO_ONLY
2601 
2602 #define MUSB_USB_DMA_ALIGN 4
2603 
2604 struct musb_temp_buffer {
2605 	void *kmalloc_ptr;
2606 	void *old_xfer_buffer;
2607 	u8 data[0];
2608 };
2609 
2610 static void musb_free_temp_buffer(struct urb *urb)
2611 {
2612 	enum dma_data_direction dir;
2613 	struct musb_temp_buffer *temp;
2614 	size_t length;
2615 
2616 	if (!(urb->transfer_flags & URB_ALIGNED_TEMP_BUFFER))
2617 		return;
2618 
2619 	dir = usb_urb_dir_in(urb) ? DMA_FROM_DEVICE : DMA_TO_DEVICE;
2620 
2621 	temp = container_of(urb->transfer_buffer, struct musb_temp_buffer,
2622 			    data);
2623 
2624 	if (dir == DMA_FROM_DEVICE) {
2625 		if (usb_pipeisoc(urb->pipe))
2626 			length = urb->transfer_buffer_length;
2627 		else
2628 			length = urb->actual_length;
2629 
2630 		memcpy(temp->old_xfer_buffer, temp->data, length);
2631 	}
2632 	urb->transfer_buffer = temp->old_xfer_buffer;
2633 	kfree(temp->kmalloc_ptr);
2634 
2635 	urb->transfer_flags &= ~URB_ALIGNED_TEMP_BUFFER;
2636 }
2637 
2638 static int musb_alloc_temp_buffer(struct urb *urb, gfp_t mem_flags)
2639 {
2640 	enum dma_data_direction dir;
2641 	struct musb_temp_buffer *temp;
2642 	void *kmalloc_ptr;
2643 	size_t kmalloc_size;
2644 
2645 	if (urb->num_sgs || urb->sg ||
2646 	    urb->transfer_buffer_length == 0 ||
2647 	    !((uintptr_t)urb->transfer_buffer & (MUSB_USB_DMA_ALIGN - 1)))
2648 		return 0;
2649 
2650 	dir = usb_urb_dir_in(urb) ? DMA_FROM_DEVICE : DMA_TO_DEVICE;
2651 
2652 	/* Allocate a buffer with enough padding for alignment */
2653 	kmalloc_size = urb->transfer_buffer_length +
2654 		sizeof(struct musb_temp_buffer) + MUSB_USB_DMA_ALIGN - 1;
2655 
2656 	kmalloc_ptr = kmalloc(kmalloc_size, mem_flags);
2657 	if (!kmalloc_ptr)
2658 		return -ENOMEM;
2659 
2660 	/* Position our struct temp_buffer such that data is aligned */
2661 	temp = PTR_ALIGN(kmalloc_ptr, MUSB_USB_DMA_ALIGN);
2662 
2663 
2664 	temp->kmalloc_ptr = kmalloc_ptr;
2665 	temp->old_xfer_buffer = urb->transfer_buffer;
2666 	if (dir == DMA_TO_DEVICE)
2667 		memcpy(temp->data, urb->transfer_buffer,
2668 		       urb->transfer_buffer_length);
2669 	urb->transfer_buffer = temp->data;
2670 
2671 	urb->transfer_flags |= URB_ALIGNED_TEMP_BUFFER;
2672 
2673 	return 0;
2674 }
2675 
2676 static int musb_map_urb_for_dma(struct usb_hcd *hcd, struct urb *urb,
2677 				      gfp_t mem_flags)
2678 {
2679 	struct musb	*musb = hcd_to_musb(hcd);
2680 	int ret;
2681 
2682 	/*
2683 	 * The DMA engine in RTL1.8 and above cannot handle
2684 	 * DMA addresses that are not aligned to a 4 byte boundary.
2685 	 * For such engine implemented (un)map_urb_for_dma hooks.
2686 	 * Do not use these hooks for RTL<1.8
2687 	 */
2688 	if (musb->hwvers < MUSB_HWVERS_1800)
2689 		return usb_hcd_map_urb_for_dma(hcd, urb, mem_flags);
2690 
2691 	ret = musb_alloc_temp_buffer(urb, mem_flags);
2692 	if (ret)
2693 		return ret;
2694 
2695 	ret = usb_hcd_map_urb_for_dma(hcd, urb, mem_flags);
2696 	if (ret)
2697 		musb_free_temp_buffer(urb);
2698 
2699 	return ret;
2700 }
2701 
2702 static void musb_unmap_urb_for_dma(struct usb_hcd *hcd, struct urb *urb)
2703 {
2704 	struct musb	*musb = hcd_to_musb(hcd);
2705 
2706 	usb_hcd_unmap_urb_for_dma(hcd, urb);
2707 
2708 	/* Do not use this hook for RTL<1.8 (see description above) */
2709 	if (musb->hwvers < MUSB_HWVERS_1800)
2710 		return;
2711 
2712 	musb_free_temp_buffer(urb);
2713 }
2714 #endif /* !CONFIG_MUSB_PIO_ONLY */
2715 
2716 static const struct hc_driver musb_hc_driver = {
2717 	.description		= "musb-hcd",
2718 	.product_desc		= "MUSB HDRC host driver",
2719 	.hcd_priv_size		= sizeof(struct musb *),
2720 	.flags			= HCD_USB2 | HCD_MEMORY,
2721 
2722 	/* not using irq handler or reset hooks from usbcore, since
2723 	 * those must be shared with peripheral code for OTG configs
2724 	 */
2725 
2726 	.start			= musb_h_start,
2727 	.stop			= musb_h_stop,
2728 
2729 	.get_frame_number	= musb_h_get_frame_number,
2730 
2731 	.urb_enqueue		= musb_urb_enqueue,
2732 	.urb_dequeue		= musb_urb_dequeue,
2733 	.endpoint_disable	= musb_h_disable,
2734 
2735 #ifndef CONFIG_MUSB_PIO_ONLY
2736 	.map_urb_for_dma	= musb_map_urb_for_dma,
2737 	.unmap_urb_for_dma	= musb_unmap_urb_for_dma,
2738 #endif
2739 
2740 	.hub_status_data	= musb_hub_status_data,
2741 	.hub_control		= musb_hub_control,
2742 	.bus_suspend		= musb_bus_suspend,
2743 	.bus_resume		= musb_bus_resume,
2744 	/* .start_port_reset	= NULL, */
2745 	/* .hub_irq_enable	= NULL, */
2746 };
2747 
2748 int musb_host_alloc(struct musb *musb)
2749 {
2750 	struct device	*dev = musb->controller;
2751 
2752 	/* usbcore sets dev->driver_data to hcd, and sometimes uses that... */
2753 	musb->hcd = usb_create_hcd(&musb_hc_driver, dev, dev_name(dev));
2754 	if (!musb->hcd)
2755 		return -EINVAL;
2756 
2757 	*musb->hcd->hcd_priv = (unsigned long) musb;
2758 	musb->hcd->self.uses_pio_for_control = 1;
2759 	musb->hcd->uses_new_polling = 1;
2760 	musb->hcd->has_tt = 1;
2761 
2762 	return 0;
2763 }
2764 
2765 void musb_host_cleanup(struct musb *musb)
2766 {
2767 	if (musb->port_mode == MUSB_PORT_MODE_GADGET)
2768 		return;
2769 	usb_remove_hcd(musb->hcd);
2770 }
2771 
2772 void musb_host_free(struct musb *musb)
2773 {
2774 	usb_put_hcd(musb->hcd);
2775 }
2776 
2777 int musb_host_setup(struct musb *musb, int power_budget)
2778 {
2779 	int ret;
2780 	struct usb_hcd *hcd = musb->hcd;
2781 
2782 	if (musb->port_mode == MUSB_PORT_MODE_HOST) {
2783 		MUSB_HST_MODE(musb);
2784 		musb->xceiv->otg->default_a = 1;
2785 		musb->xceiv->otg->state = OTG_STATE_A_IDLE;
2786 	}
2787 	otg_set_host(musb->xceiv->otg, &hcd->self);
2788 	hcd->self.otg_port = 1;
2789 	musb->xceiv->otg->host = &hcd->self;
2790 	hcd->power_budget = 2 * (power_budget ? : 250);
2791 
2792 	ret = usb_add_hcd(hcd, 0, 0);
2793 	if (ret < 0)
2794 		return ret;
2795 
2796 	device_wakeup_enable(hcd->self.controller);
2797 	return 0;
2798 }
2799 
2800 void musb_host_resume_root_hub(struct musb *musb)
2801 {
2802 	usb_hcd_resume_root_hub(musb->hcd);
2803 }
2804 
2805 void musb_host_poke_root_hub(struct musb *musb)
2806 {
2807 	MUSB_HST_MODE(musb);
2808 	if (musb->hcd->status_urb)
2809 		usb_hcd_poll_rh_status(musb->hcd);
2810 	else
2811 		usb_hcd_resume_root_hub(musb->hcd);
2812 }
2813