xref: /openbmc/linux/drivers/usb/musb/musb_host.c (revision c0e297dc)
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 
48 /* MUSB HOST status 22-mar-2006
49  *
50  * - There's still lots of partial code duplication for fault paths, so
51  *   they aren't handled as consistently as they need to be.
52  *
53  * - PIO mostly behaved when last tested.
54  *     + including ep0, with all usbtest cases 9, 10
55  *     + usbtest 14 (ep0out) doesn't seem to run at all
56  *     + double buffered OUT/TX endpoints saw stalls(!) with certain usbtest
57  *       configurations, but otherwise double buffering passes basic tests.
58  *     + for 2.6.N, for N > ~10, needs API changes for hcd framework.
59  *
60  * - DMA (CPPI) ... partially behaves, not currently recommended
61  *     + about 1/15 the speed of typical EHCI implementations (PCI)
62  *     + RX, all too often reqpkt seems to misbehave after tx
63  *     + TX, no known issues (other than evident silicon issue)
64  *
65  * - DMA (Mentor/OMAP) ...has at least toggle update problems
66  *
67  * - [23-feb-2009] minimal traffic scheduling to avoid bulk RX packet
68  *   starvation ... nothing yet for TX, interrupt, or bulk.
69  *
70  * - Not tested with HNP, but some SRP paths seem to behave.
71  *
72  * NOTE 24-August-2006:
73  *
74  * - Bulk traffic finally uses both sides of hardware ep1, freeing up an
75  *   extra endpoint for periodic use enabling hub + keybd + mouse.  That
76  *   mostly works, except that with "usbnet" it's easy to trigger cases
77  *   with "ping" where RX loses.  (a) ping to davinci, even "ping -f",
78  *   fine; but (b) ping _from_ davinci, even "ping -c 1", ICMP RX loses
79  *   although ARP RX wins.  (That test was done with a full speed link.)
80  */
81 
82 
83 /*
84  * NOTE on endpoint usage:
85  *
86  * CONTROL transfers all go through ep0.  BULK ones go through dedicated IN
87  * and OUT endpoints ... hardware is dedicated for those "async" queue(s).
88  * (Yes, bulk _could_ use more of the endpoints than that, and would even
89  * benefit from it.)
90  *
91  * INTERUPPT and ISOCHRONOUS transfers are scheduled to the other endpoints.
92  * So far that scheduling is both dumb and optimistic:  the endpoint will be
93  * "claimed" until its software queue is no longer refilled.  No multiplexing
94  * of transfers between endpoints, or anything clever.
95  */
96 
97 struct musb *hcd_to_musb(struct usb_hcd *hcd)
98 {
99 	return *(struct musb **) hcd->hcd_priv;
100 }
101 
102 
103 static void musb_ep_program(struct musb *musb, u8 epnum,
104 			struct urb *urb, int is_out,
105 			u8 *buf, u32 offset, u32 len);
106 
107 /*
108  * Clear TX fifo. Needed to avoid BABBLE errors.
109  */
110 static void musb_h_tx_flush_fifo(struct musb_hw_ep *ep)
111 {
112 	struct musb	*musb = ep->musb;
113 	void __iomem	*epio = ep->regs;
114 	u16		csr;
115 	u16		lastcsr = 0;
116 	int		retries = 1000;
117 
118 	csr = musb_readw(epio, MUSB_TXCSR);
119 	while (csr & MUSB_TXCSR_FIFONOTEMPTY) {
120 		if (csr != lastcsr)
121 			dev_dbg(musb->controller, "Host TX FIFONOTEMPTY csr: %02x\n", csr);
122 		lastcsr = csr;
123 		csr |= MUSB_TXCSR_FLUSHFIFO | MUSB_TXCSR_TXPKTRDY;
124 		musb_writew(epio, MUSB_TXCSR, csr);
125 		csr = musb_readw(epio, MUSB_TXCSR);
126 		if (WARN(retries-- < 1,
127 				"Could not flush host TX%d fifo: csr: %04x\n",
128 				ep->epnum, csr))
129 			return;
130 		mdelay(1);
131 	}
132 }
133 
134 static void musb_h_ep0_flush_fifo(struct musb_hw_ep *ep)
135 {
136 	void __iomem	*epio = ep->regs;
137 	u16		csr;
138 	int		retries = 5;
139 
140 	/* scrub any data left in the fifo */
141 	do {
142 		csr = musb_readw(epio, MUSB_TXCSR);
143 		if (!(csr & (MUSB_CSR0_TXPKTRDY | MUSB_CSR0_RXPKTRDY)))
144 			break;
145 		musb_writew(epio, MUSB_TXCSR, MUSB_CSR0_FLUSHFIFO);
146 		csr = musb_readw(epio, MUSB_TXCSR);
147 		udelay(10);
148 	} while (--retries);
149 
150 	WARN(!retries, "Could not flush host TX%d fifo: csr: %04x\n",
151 			ep->epnum, csr);
152 
153 	/* and reset for the next transfer */
154 	musb_writew(epio, MUSB_TXCSR, 0);
155 }
156 
157 /*
158  * Start transmit. Caller is responsible for locking shared resources.
159  * musb must be locked.
160  */
161 static inline void musb_h_tx_start(struct musb_hw_ep *ep)
162 {
163 	u16	txcsr;
164 
165 	/* NOTE: no locks here; caller should lock and select EP */
166 	if (ep->epnum) {
167 		txcsr = musb_readw(ep->regs, MUSB_TXCSR);
168 		txcsr |= MUSB_TXCSR_TXPKTRDY | MUSB_TXCSR_H_WZC_BITS;
169 		musb_writew(ep->regs, MUSB_TXCSR, txcsr);
170 	} else {
171 		txcsr = MUSB_CSR0_H_SETUPPKT | MUSB_CSR0_TXPKTRDY;
172 		musb_writew(ep->regs, MUSB_CSR0, txcsr);
173 	}
174 
175 }
176 
177 static inline void musb_h_tx_dma_start(struct musb_hw_ep *ep)
178 {
179 	u16	txcsr;
180 
181 	/* NOTE: no locks here; caller should lock and select EP */
182 	txcsr = musb_readw(ep->regs, MUSB_TXCSR);
183 	txcsr |= MUSB_TXCSR_DMAENAB | MUSB_TXCSR_H_WZC_BITS;
184 	if (is_cppi_enabled(ep->musb))
185 		txcsr |= MUSB_TXCSR_DMAMODE;
186 	musb_writew(ep->regs, MUSB_TXCSR, txcsr);
187 }
188 
189 static void musb_ep_set_qh(struct musb_hw_ep *ep, int is_in, struct musb_qh *qh)
190 {
191 	if (is_in != 0 || ep->is_shared_fifo)
192 		ep->in_qh  = qh;
193 	if (is_in == 0 || ep->is_shared_fifo)
194 		ep->out_qh = qh;
195 }
196 
197 static struct musb_qh *musb_ep_get_qh(struct musb_hw_ep *ep, int is_in)
198 {
199 	return is_in ? ep->in_qh : ep->out_qh;
200 }
201 
202 /*
203  * Start the URB at the front of an endpoint's queue
204  * end must be claimed from the caller.
205  *
206  * Context: controller locked, irqs blocked
207  */
208 static void
209 musb_start_urb(struct musb *musb, int is_in, struct musb_qh *qh)
210 {
211 	u16			frame;
212 	u32			len;
213 	void __iomem		*mbase =  musb->mregs;
214 	struct urb		*urb = next_urb(qh);
215 	void			*buf = urb->transfer_buffer;
216 	u32			offset = 0;
217 	struct musb_hw_ep	*hw_ep = qh->hw_ep;
218 	unsigned		pipe = urb->pipe;
219 	u8			address = usb_pipedevice(pipe);
220 	int			epnum = hw_ep->epnum;
221 
222 	/* initialize software qh state */
223 	qh->offset = 0;
224 	qh->segsize = 0;
225 
226 	/* gather right source of data */
227 	switch (qh->type) {
228 	case USB_ENDPOINT_XFER_CONTROL:
229 		/* control transfers always start with SETUP */
230 		is_in = 0;
231 		musb->ep0_stage = MUSB_EP0_START;
232 		buf = urb->setup_packet;
233 		len = 8;
234 		break;
235 	case USB_ENDPOINT_XFER_ISOC:
236 		qh->iso_idx = 0;
237 		qh->frame = 0;
238 		offset = urb->iso_frame_desc[0].offset;
239 		len = urb->iso_frame_desc[0].length;
240 		break;
241 	default:		/* bulk, interrupt */
242 		/* actual_length may be nonzero on retry paths */
243 		buf = urb->transfer_buffer + urb->actual_length;
244 		len = urb->transfer_buffer_length - urb->actual_length;
245 	}
246 
247 	dev_dbg(musb->controller, "qh %p urb %p dev%d ep%d%s%s, hw_ep %d, %p/%d\n",
248 			qh, urb, address, qh->epnum,
249 			is_in ? "in" : "out",
250 			({char *s; switch (qh->type) {
251 			case USB_ENDPOINT_XFER_CONTROL:	s = ""; break;
252 			case USB_ENDPOINT_XFER_BULK:	s = "-bulk"; break;
253 			case USB_ENDPOINT_XFER_ISOC:	s = "-iso"; break;
254 			default:			s = "-intr"; break;
255 			} s; }),
256 			epnum, buf + offset, len);
257 
258 	/* Configure endpoint */
259 	musb_ep_set_qh(hw_ep, is_in, qh);
260 	musb_ep_program(musb, epnum, urb, !is_in, buf, offset, len);
261 
262 	/* transmit may have more work: start it when it is time */
263 	if (is_in)
264 		return;
265 
266 	/* determine if the time is right for a periodic transfer */
267 	switch (qh->type) {
268 	case USB_ENDPOINT_XFER_ISOC:
269 	case USB_ENDPOINT_XFER_INT:
270 		dev_dbg(musb->controller, "check whether there's still time for periodic Tx\n");
271 		frame = musb_readw(mbase, MUSB_FRAME);
272 		/* FIXME this doesn't implement that scheduling policy ...
273 		 * or handle framecounter wrapping
274 		 */
275 		if (1) {	/* Always assume URB_ISO_ASAP */
276 			/* REVISIT the SOF irq handler shouldn't duplicate
277 			 * this code; and we don't init urb->start_frame...
278 			 */
279 			qh->frame = 0;
280 			goto start;
281 		} else {
282 			qh->frame = urb->start_frame;
283 			/* enable SOF interrupt so we can count down */
284 			dev_dbg(musb->controller, "SOF for %d\n", epnum);
285 #if 1 /* ifndef	CONFIG_ARCH_DAVINCI */
286 			musb_writeb(mbase, MUSB_INTRUSBE, 0xff);
287 #endif
288 		}
289 		break;
290 	default:
291 start:
292 		dev_dbg(musb->controller, "Start TX%d %s\n", epnum,
293 			hw_ep->tx_channel ? "dma" : "pio");
294 
295 		if (!hw_ep->tx_channel)
296 			musb_h_tx_start(hw_ep);
297 		else if (is_cppi_enabled(musb) || tusb_dma_omap(musb))
298 			musb_h_tx_dma_start(hw_ep);
299 	}
300 }
301 
302 /* Context: caller owns controller lock, IRQs are blocked */
303 static void musb_giveback(struct musb *musb, struct urb *urb, int status)
304 __releases(musb->lock)
305 __acquires(musb->lock)
306 {
307 	dev_dbg(musb->controller,
308 			"complete %p %pF (%d), dev%d ep%d%s, %d/%d\n",
309 			urb, urb->complete, status,
310 			usb_pipedevice(urb->pipe),
311 			usb_pipeendpoint(urb->pipe),
312 			usb_pipein(urb->pipe) ? "in" : "out",
313 			urb->actual_length, urb->transfer_buffer_length
314 			);
315 
316 	usb_hcd_unlink_urb_from_ep(musb->hcd, urb);
317 	spin_unlock(&musb->lock);
318 	usb_hcd_giveback_urb(musb->hcd, urb, status);
319 	spin_lock(&musb->lock);
320 }
321 
322 /* For bulk/interrupt endpoints only */
323 static inline void musb_save_toggle(struct musb_qh *qh, int is_in,
324 				    struct urb *urb)
325 {
326 	void __iomem		*epio = qh->hw_ep->regs;
327 	u16			csr;
328 
329 	/*
330 	 * FIXME: the current Mentor DMA code seems to have
331 	 * problems getting toggle correct.
332 	 */
333 
334 	if (is_in)
335 		csr = musb_readw(epio, MUSB_RXCSR) & MUSB_RXCSR_H_DATATOGGLE;
336 	else
337 		csr = musb_readw(epio, MUSB_TXCSR) & MUSB_TXCSR_H_DATATOGGLE;
338 
339 	usb_settoggle(urb->dev, qh->epnum, !is_in, csr ? 1 : 0);
340 }
341 
342 /*
343  * Advance this hardware endpoint's queue, completing the specified URB and
344  * advancing to either the next URB queued to that qh, or else invalidating
345  * that qh and advancing to the next qh scheduled after the current one.
346  *
347  * Context: caller owns controller lock, IRQs are blocked
348  */
349 static void musb_advance_schedule(struct musb *musb, struct urb *urb,
350 				  struct musb_hw_ep *hw_ep, int is_in)
351 {
352 	struct musb_qh		*qh = musb_ep_get_qh(hw_ep, is_in);
353 	struct musb_hw_ep	*ep = qh->hw_ep;
354 	int			ready = qh->is_ready;
355 	int			status;
356 
357 	status = (urb->status == -EINPROGRESS) ? 0 : urb->status;
358 
359 	/* save toggle eagerly, for paranoia */
360 	switch (qh->type) {
361 	case USB_ENDPOINT_XFER_BULK:
362 	case USB_ENDPOINT_XFER_INT:
363 		musb_save_toggle(qh, is_in, urb);
364 		break;
365 	case USB_ENDPOINT_XFER_ISOC:
366 		if (status == 0 && urb->error_count)
367 			status = -EXDEV;
368 		break;
369 	}
370 
371 	qh->is_ready = 0;
372 	musb_giveback(musb, urb, status);
373 	qh->is_ready = ready;
374 
375 	/* reclaim resources (and bandwidth) ASAP; deschedule it, and
376 	 * invalidate qh as soon as list_empty(&hep->urb_list)
377 	 */
378 	if (list_empty(&qh->hep->urb_list)) {
379 		struct list_head	*head;
380 		struct dma_controller	*dma = musb->dma_controller;
381 
382 		if (is_in) {
383 			ep->rx_reinit = 1;
384 			if (ep->rx_channel) {
385 				dma->channel_release(ep->rx_channel);
386 				ep->rx_channel = NULL;
387 			}
388 		} else {
389 			ep->tx_reinit = 1;
390 			if (ep->tx_channel) {
391 				dma->channel_release(ep->tx_channel);
392 				ep->tx_channel = NULL;
393 			}
394 		}
395 
396 		/* Clobber old pointers to this qh */
397 		musb_ep_set_qh(ep, is_in, NULL);
398 		qh->hep->hcpriv = NULL;
399 
400 		switch (qh->type) {
401 
402 		case USB_ENDPOINT_XFER_CONTROL:
403 		case USB_ENDPOINT_XFER_BULK:
404 			/* fifo policy for these lists, except that NAKing
405 			 * should rotate a qh to the end (for fairness).
406 			 */
407 			if (qh->mux == 1) {
408 				head = qh->ring.prev;
409 				list_del(&qh->ring);
410 				kfree(qh);
411 				qh = first_qh(head);
412 				break;
413 			}
414 
415 		case USB_ENDPOINT_XFER_ISOC:
416 		case USB_ENDPOINT_XFER_INT:
417 			/* this is where periodic bandwidth should be
418 			 * de-allocated if it's tracked and allocated;
419 			 * and where we'd update the schedule tree...
420 			 */
421 			kfree(qh);
422 			qh = NULL;
423 			break;
424 		}
425 	}
426 
427 	if (qh != NULL && qh->is_ready) {
428 		dev_dbg(musb->controller, "... next ep%d %cX urb %p\n",
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 	dev_dbg(musb->controller, "RX%d count %d, buffer %p len %d/%d\n", 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 			dev_dbg(musb->controller, "** OVERFLOW %d into %d\n", 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 			dev_dbg(musb->controller, "** OVERFLOW %d into %d\n", 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 	} else {
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 
596 	/* target addr and (for multipoint) hub addr/port */
597 	if (musb->is_multipoint) {
598 		musb_write_rxfunaddr(musb, epnum, qh->addr_reg);
599 		musb_write_rxhubaddr(musb, epnum, qh->h_addr_reg);
600 		musb_write_rxhubport(musb, epnum, qh->h_port_reg);
601 	} else
602 		musb_writeb(musb->mregs, MUSB_FADDR, qh->addr_reg);
603 
604 	/* protocol/endpoint, interval/NAKlimit, i/o size */
605 	musb_writeb(ep->regs, MUSB_RXTYPE, qh->type_reg);
606 	musb_writeb(ep->regs, MUSB_RXINTERVAL, qh->intv_reg);
607 	/* NOTE: bulk combining rewrites high bits of maxpacket */
608 	/* Set RXMAXP with the FIFO size of the endpoint
609 	 * to disable double buffer mode.
610 	 */
611 	if (musb->double_buffer_not_ok)
612 		musb_writew(ep->regs, MUSB_RXMAXP, ep->max_packet_sz_rx);
613 	else
614 		musb_writew(ep->regs, MUSB_RXMAXP,
615 				qh->maxpacket | ((qh->hb_mult - 1) << 11));
616 
617 	ep->rx_reinit = 0;
618 }
619 
620 static int musb_tx_dma_set_mode_mentor(struct dma_controller *dma,
621 		struct musb_hw_ep *hw_ep, struct musb_qh *qh,
622 		struct urb *urb, u32 offset,
623 		u32 *length, u8 *mode)
624 {
625 	struct dma_channel	*channel = hw_ep->tx_channel;
626 	void __iomem		*epio = hw_ep->regs;
627 	u16			pkt_size = qh->maxpacket;
628 	u16			csr;
629 
630 	if (*length > channel->max_len)
631 		*length = channel->max_len;
632 
633 	csr = musb_readw(epio, MUSB_TXCSR);
634 	if (*length > pkt_size) {
635 		*mode = 1;
636 		csr |= MUSB_TXCSR_DMAMODE | MUSB_TXCSR_DMAENAB;
637 		/* autoset shouldn't be set in high bandwidth */
638 		/*
639 		 * Enable Autoset according to table
640 		 * below
641 		 * bulk_split hb_mult	Autoset_Enable
642 		 *	0	1	Yes(Normal)
643 		 *	0	>1	No(High BW ISO)
644 		 *	1	1	Yes(HS bulk)
645 		 *	1	>1	Yes(FS bulk)
646 		 */
647 		if (qh->hb_mult == 1 || (qh->hb_mult > 1 &&
648 					can_bulk_split(hw_ep->musb, qh->type)))
649 			csr |= MUSB_TXCSR_AUTOSET;
650 	} else {
651 		*mode = 0;
652 		csr &= ~(MUSB_TXCSR_AUTOSET | MUSB_TXCSR_DMAMODE);
653 		csr |= MUSB_TXCSR_DMAENAB; /* against programmer's guide */
654 	}
655 	channel->desired_mode = mode;
656 	musb_writew(epio, MUSB_TXCSR, csr);
657 
658 	return 0;
659 }
660 
661 static int musb_tx_dma_set_mode_cppi_tusb(struct dma_controller *dma,
662 					  struct musb_hw_ep *hw_ep,
663 					  struct musb_qh *qh,
664 					  struct urb *urb,
665 					  u32 offset,
666 					  u32 *length,
667 					  u8 *mode)
668 {
669 	struct dma_channel *channel = hw_ep->tx_channel;
670 
671 	if (!is_cppi_enabled(hw_ep->musb) && !tusb_dma_omap(hw_ep->musb))
672 		return -ENODEV;
673 
674 	channel->actual_len = 0;
675 
676 	/*
677 	 * TX uses "RNDIS" mode automatically but needs help
678 	 * to identify the zero-length-final-packet case.
679 	 */
680 	*mode = (urb->transfer_flags & URB_ZERO_PACKET) ? 1 : 0;
681 
682 	return 0;
683 }
684 
685 static bool musb_tx_dma_program(struct dma_controller *dma,
686 		struct musb_hw_ep *hw_ep, struct musb_qh *qh,
687 		struct urb *urb, u32 offset, u32 length)
688 {
689 	struct dma_channel	*channel = hw_ep->tx_channel;
690 	u16			pkt_size = qh->maxpacket;
691 	u8			mode;
692 	int			res;
693 
694 	if (musb_dma_inventra(hw_ep->musb) || musb_dma_ux500(hw_ep->musb))
695 		res = musb_tx_dma_set_mode_mentor(dma, hw_ep, qh, urb,
696 						 offset, &length, &mode);
697 	else
698 		res = musb_tx_dma_set_mode_cppi_tusb(dma, hw_ep, qh, urb,
699 						     offset, &length, &mode);
700 	if (res)
701 		return false;
702 
703 	qh->segsize = length;
704 
705 	/*
706 	 * Ensure the data reaches to main memory before starting
707 	 * DMA transfer
708 	 */
709 	wmb();
710 
711 	if (!dma->channel_program(channel, pkt_size, mode,
712 			urb->transfer_dma + offset, length)) {
713 		void __iomem *epio = hw_ep->regs;
714 		u16 csr;
715 
716 		dma->channel_release(channel);
717 		hw_ep->tx_channel = NULL;
718 
719 		csr = musb_readw(epio, MUSB_TXCSR);
720 		csr &= ~(MUSB_TXCSR_AUTOSET | MUSB_TXCSR_DMAENAB);
721 		musb_writew(epio, MUSB_TXCSR, csr | MUSB_TXCSR_H_WZC_BITS);
722 		return false;
723 	}
724 	return true;
725 }
726 
727 /*
728  * Program an HDRC endpoint as per the given URB
729  * Context: irqs blocked, controller lock held
730  */
731 static void musb_ep_program(struct musb *musb, u8 epnum,
732 			struct urb *urb, int is_out,
733 			u8 *buf, u32 offset, u32 len)
734 {
735 	struct dma_controller	*dma_controller;
736 	struct dma_channel	*dma_channel;
737 	u8			dma_ok;
738 	void __iomem		*mbase = musb->mregs;
739 	struct musb_hw_ep	*hw_ep = musb->endpoints + epnum;
740 	void __iomem		*epio = hw_ep->regs;
741 	struct musb_qh		*qh = musb_ep_get_qh(hw_ep, !is_out);
742 	u16			packet_sz = qh->maxpacket;
743 	u8			use_dma = 1;
744 	u16			csr;
745 
746 	dev_dbg(musb->controller, "%s hw%d urb %p spd%d dev%d ep%d%s "
747 				"h_addr%02x h_port%02x bytes %d\n",
748 			is_out ? "-->" : "<--",
749 			epnum, urb, urb->dev->speed,
750 			qh->addr_reg, qh->epnum, is_out ? "out" : "in",
751 			qh->h_addr_reg, qh->h_port_reg,
752 			len);
753 
754 	musb_ep_select(mbase, epnum);
755 
756 	if (is_out && !len) {
757 		use_dma = 0;
758 		csr = musb_readw(epio, MUSB_TXCSR);
759 		csr &= ~MUSB_TXCSR_DMAENAB;
760 		musb_writew(epio, MUSB_TXCSR, csr);
761 		hw_ep->tx_channel = NULL;
762 	}
763 
764 	/* candidate for DMA? */
765 	dma_controller = musb->dma_controller;
766 	if (use_dma && is_dma_capable() && epnum && dma_controller) {
767 		dma_channel = is_out ? hw_ep->tx_channel : hw_ep->rx_channel;
768 		if (!dma_channel) {
769 			dma_channel = dma_controller->channel_alloc(
770 					dma_controller, hw_ep, is_out);
771 			if (is_out)
772 				hw_ep->tx_channel = dma_channel;
773 			else
774 				hw_ep->rx_channel = dma_channel;
775 		}
776 	} else
777 		dma_channel = NULL;
778 
779 	/* make sure we clear DMAEnab, autoSet bits from previous run */
780 
781 	/* OUT/transmit/EP0 or IN/receive? */
782 	if (is_out) {
783 		u16	csr;
784 		u16	int_txe;
785 		u16	load_count;
786 
787 		csr = musb_readw(epio, MUSB_TXCSR);
788 
789 		/* disable interrupt in case we flush */
790 		int_txe = musb->intrtxe;
791 		musb_writew(mbase, MUSB_INTRTXE, int_txe & ~(1 << epnum));
792 
793 		/* general endpoint setup */
794 		if (epnum) {
795 			/* flush all old state, set default */
796 			/*
797 			 * We could be flushing valid
798 			 * packets in double buffering
799 			 * case
800 			 */
801 			if (!hw_ep->tx_double_buffered)
802 				musb_h_tx_flush_fifo(hw_ep);
803 
804 			/*
805 			 * We must not clear the DMAMODE bit before or in
806 			 * the same cycle with the DMAENAB bit, so we clear
807 			 * the latter first...
808 			 */
809 			csr &= ~(MUSB_TXCSR_H_NAKTIMEOUT
810 					| MUSB_TXCSR_AUTOSET
811 					| MUSB_TXCSR_DMAENAB
812 					| MUSB_TXCSR_FRCDATATOG
813 					| MUSB_TXCSR_H_RXSTALL
814 					| MUSB_TXCSR_H_ERROR
815 					| MUSB_TXCSR_TXPKTRDY
816 					);
817 			csr |= MUSB_TXCSR_MODE;
818 
819 			if (!hw_ep->tx_double_buffered) {
820 				if (usb_gettoggle(urb->dev, qh->epnum, 1))
821 					csr |= MUSB_TXCSR_H_WR_DATATOGGLE
822 						| MUSB_TXCSR_H_DATATOGGLE;
823 				else
824 					csr |= MUSB_TXCSR_CLRDATATOG;
825 			}
826 
827 			musb_writew(epio, MUSB_TXCSR, csr);
828 			/* REVISIT may need to clear FLUSHFIFO ... */
829 			csr &= ~MUSB_TXCSR_DMAMODE;
830 			musb_writew(epio, MUSB_TXCSR, csr);
831 			csr = musb_readw(epio, MUSB_TXCSR);
832 		} else {
833 			/* endpoint 0: just flush */
834 			musb_h_ep0_flush_fifo(hw_ep);
835 		}
836 
837 		/* target addr and (for multipoint) hub addr/port */
838 		if (musb->is_multipoint) {
839 			musb_write_txfunaddr(musb, epnum, qh->addr_reg);
840 			musb_write_txhubaddr(musb, epnum, qh->h_addr_reg);
841 			musb_write_txhubport(musb, epnum, qh->h_port_reg);
842 /* FIXME if !epnum, do the same for RX ... */
843 		} else
844 			musb_writeb(mbase, MUSB_FADDR, qh->addr_reg);
845 
846 		/* protocol/endpoint/interval/NAKlimit */
847 		if (epnum) {
848 			musb_writeb(epio, MUSB_TXTYPE, qh->type_reg);
849 			if (musb->double_buffer_not_ok) {
850 				musb_writew(epio, MUSB_TXMAXP,
851 						hw_ep->max_packet_sz_tx);
852 			} else if (can_bulk_split(musb, qh->type)) {
853 				qh->hb_mult = hw_ep->max_packet_sz_tx
854 						/ packet_sz;
855 				musb_writew(epio, MUSB_TXMAXP, packet_sz
856 					| ((qh->hb_mult) - 1) << 11);
857 			} else {
858 				musb_writew(epio, MUSB_TXMAXP,
859 						qh->maxpacket |
860 						((qh->hb_mult - 1) << 11));
861 			}
862 			musb_writeb(epio, MUSB_TXINTERVAL, qh->intv_reg);
863 		} else {
864 			musb_writeb(epio, MUSB_NAKLIMIT0, qh->intv_reg);
865 			if (musb->is_multipoint)
866 				musb_writeb(epio, MUSB_TYPE0,
867 						qh->type_reg);
868 		}
869 
870 		if (can_bulk_split(musb, qh->type))
871 			load_count = min((u32) hw_ep->max_packet_sz_tx,
872 						len);
873 		else
874 			load_count = min((u32) packet_sz, len);
875 
876 		if (dma_channel && musb_tx_dma_program(dma_controller,
877 					hw_ep, qh, urb, offset, len))
878 			load_count = 0;
879 
880 		if (load_count) {
881 			/* PIO to load FIFO */
882 			qh->segsize = load_count;
883 			if (!buf) {
884 				sg_miter_start(&qh->sg_miter, urb->sg, 1,
885 						SG_MITER_ATOMIC
886 						| SG_MITER_FROM_SG);
887 				if (!sg_miter_next(&qh->sg_miter)) {
888 					dev_err(musb->controller,
889 							"error: sg"
890 							"list empty\n");
891 					sg_miter_stop(&qh->sg_miter);
892 					goto finish;
893 				}
894 				buf = qh->sg_miter.addr + urb->sg->offset +
895 					urb->actual_length;
896 				load_count = min_t(u32, load_count,
897 						qh->sg_miter.length);
898 				musb_write_fifo(hw_ep, load_count, buf);
899 				qh->sg_miter.consumed = load_count;
900 				sg_miter_stop(&qh->sg_miter);
901 			} else
902 				musb_write_fifo(hw_ep, load_count, buf);
903 		}
904 finish:
905 		/* re-enable interrupt */
906 		musb_writew(mbase, MUSB_INTRTXE, int_txe);
907 
908 	/* IN/receive */
909 	} else {
910 		u16	csr;
911 
912 		if (hw_ep->rx_reinit) {
913 			musb_rx_reinit(musb, qh, epnum);
914 
915 			/* init new state: toggle and NYET, maybe DMA later */
916 			if (usb_gettoggle(urb->dev, qh->epnum, 0))
917 				csr = MUSB_RXCSR_H_WR_DATATOGGLE
918 					| MUSB_RXCSR_H_DATATOGGLE;
919 			else
920 				csr = 0;
921 			if (qh->type == USB_ENDPOINT_XFER_INT)
922 				csr |= MUSB_RXCSR_DISNYET;
923 
924 		} else {
925 			csr = musb_readw(hw_ep->regs, MUSB_RXCSR);
926 
927 			if (csr & (MUSB_RXCSR_RXPKTRDY
928 					| MUSB_RXCSR_DMAENAB
929 					| MUSB_RXCSR_H_REQPKT))
930 				ERR("broken !rx_reinit, ep%d csr %04x\n",
931 						hw_ep->epnum, csr);
932 
933 			/* scrub any stale state, leaving toggle alone */
934 			csr &= MUSB_RXCSR_DISNYET;
935 		}
936 
937 		/* kick things off */
938 
939 		if ((is_cppi_enabled(musb) || tusb_dma_omap(musb)) && dma_channel) {
940 			/* Candidate for DMA */
941 			dma_channel->actual_len = 0L;
942 			qh->segsize = len;
943 
944 			/* AUTOREQ is in a DMA register */
945 			musb_writew(hw_ep->regs, MUSB_RXCSR, csr);
946 			csr = musb_readw(hw_ep->regs, MUSB_RXCSR);
947 
948 			/*
949 			 * Unless caller treats short RX transfers as
950 			 * errors, we dare not queue multiple transfers.
951 			 */
952 			dma_ok = dma_controller->channel_program(dma_channel,
953 					packet_sz, !(urb->transfer_flags &
954 						     URB_SHORT_NOT_OK),
955 					urb->transfer_dma + offset,
956 					qh->segsize);
957 			if (!dma_ok) {
958 				dma_controller->channel_release(dma_channel);
959 				hw_ep->rx_channel = dma_channel = NULL;
960 			} else
961 				csr |= MUSB_RXCSR_DMAENAB;
962 		}
963 
964 		csr |= MUSB_RXCSR_H_REQPKT;
965 		dev_dbg(musb->controller, "RXCSR%d := %04x\n", epnum, csr);
966 		musb_writew(hw_ep->regs, MUSB_RXCSR, csr);
967 		csr = musb_readw(hw_ep->regs, MUSB_RXCSR);
968 	}
969 }
970 
971 /* Schedule next QH from musb->in_bulk/out_bulk and move the current qh to
972  * the end; avoids starvation for other endpoints.
973  */
974 static void musb_bulk_nak_timeout(struct musb *musb, struct musb_hw_ep *ep,
975 	int is_in)
976 {
977 	struct dma_channel	*dma;
978 	struct urb		*urb;
979 	void __iomem		*mbase = musb->mregs;
980 	void __iomem		*epio = ep->regs;
981 	struct musb_qh		*cur_qh, *next_qh;
982 	u16			rx_csr, tx_csr;
983 
984 	musb_ep_select(mbase, ep->epnum);
985 	if (is_in) {
986 		dma = is_dma_capable() ? ep->rx_channel : NULL;
987 
988 		/* clear nak timeout bit */
989 		rx_csr = musb_readw(epio, MUSB_RXCSR);
990 		rx_csr |= MUSB_RXCSR_H_WZC_BITS;
991 		rx_csr &= ~MUSB_RXCSR_DATAERROR;
992 		musb_writew(epio, MUSB_RXCSR, rx_csr);
993 
994 		cur_qh = first_qh(&musb->in_bulk);
995 	} else {
996 		dma = is_dma_capable() ? ep->tx_channel : NULL;
997 
998 		/* clear nak timeout bit */
999 		tx_csr = musb_readw(epio, MUSB_TXCSR);
1000 		tx_csr |= MUSB_TXCSR_H_WZC_BITS;
1001 		tx_csr &= ~MUSB_TXCSR_H_NAKTIMEOUT;
1002 		musb_writew(epio, MUSB_TXCSR, tx_csr);
1003 
1004 		cur_qh = first_qh(&musb->out_bulk);
1005 	}
1006 	if (cur_qh) {
1007 		urb = next_urb(cur_qh);
1008 		if (dma_channel_status(dma) == MUSB_DMA_STATUS_BUSY) {
1009 			dma->status = MUSB_DMA_STATUS_CORE_ABORT;
1010 			musb->dma_controller->channel_abort(dma);
1011 			urb->actual_length += dma->actual_len;
1012 			dma->actual_len = 0L;
1013 		}
1014 		musb_save_toggle(cur_qh, is_in, urb);
1015 
1016 		if (is_in) {
1017 			/* move cur_qh to end of queue */
1018 			list_move_tail(&cur_qh->ring, &musb->in_bulk);
1019 
1020 			/* get the next qh from musb->in_bulk */
1021 			next_qh = first_qh(&musb->in_bulk);
1022 
1023 			/* set rx_reinit and schedule the next qh */
1024 			ep->rx_reinit = 1;
1025 		} else {
1026 			/* move cur_qh to end of queue */
1027 			list_move_tail(&cur_qh->ring, &musb->out_bulk);
1028 
1029 			/* get the next qh from musb->out_bulk */
1030 			next_qh = first_qh(&musb->out_bulk);
1031 
1032 			/* set tx_reinit and schedule the next qh */
1033 			ep->tx_reinit = 1;
1034 		}
1035 		musb_start_urb(musb, is_in, next_qh);
1036 	}
1037 }
1038 
1039 /*
1040  * Service the default endpoint (ep0) as host.
1041  * Return true until it's time to start the status stage.
1042  */
1043 static bool musb_h_ep0_continue(struct musb *musb, u16 len, struct urb *urb)
1044 {
1045 	bool			 more = false;
1046 	u8			*fifo_dest = NULL;
1047 	u16			fifo_count = 0;
1048 	struct musb_hw_ep	*hw_ep = musb->control_ep;
1049 	struct musb_qh		*qh = hw_ep->in_qh;
1050 	struct usb_ctrlrequest	*request;
1051 
1052 	switch (musb->ep0_stage) {
1053 	case MUSB_EP0_IN:
1054 		fifo_dest = urb->transfer_buffer + urb->actual_length;
1055 		fifo_count = min_t(size_t, len, urb->transfer_buffer_length -
1056 				   urb->actual_length);
1057 		if (fifo_count < len)
1058 			urb->status = -EOVERFLOW;
1059 
1060 		musb_read_fifo(hw_ep, fifo_count, fifo_dest);
1061 
1062 		urb->actual_length += fifo_count;
1063 		if (len < qh->maxpacket) {
1064 			/* always terminate on short read; it's
1065 			 * rarely reported as an error.
1066 			 */
1067 		} else if (urb->actual_length <
1068 				urb->transfer_buffer_length)
1069 			more = true;
1070 		break;
1071 	case MUSB_EP0_START:
1072 		request = (struct usb_ctrlrequest *) urb->setup_packet;
1073 
1074 		if (!request->wLength) {
1075 			dev_dbg(musb->controller, "start no-DATA\n");
1076 			break;
1077 		} else if (request->bRequestType & USB_DIR_IN) {
1078 			dev_dbg(musb->controller, "start IN-DATA\n");
1079 			musb->ep0_stage = MUSB_EP0_IN;
1080 			more = true;
1081 			break;
1082 		} else {
1083 			dev_dbg(musb->controller, "start OUT-DATA\n");
1084 			musb->ep0_stage = MUSB_EP0_OUT;
1085 			more = true;
1086 		}
1087 		/* FALLTHROUGH */
1088 	case MUSB_EP0_OUT:
1089 		fifo_count = min_t(size_t, qh->maxpacket,
1090 				   urb->transfer_buffer_length -
1091 				   urb->actual_length);
1092 		if (fifo_count) {
1093 			fifo_dest = (u8 *) (urb->transfer_buffer
1094 					+ urb->actual_length);
1095 			dev_dbg(musb->controller, "Sending %d byte%s to ep0 fifo %p\n",
1096 					fifo_count,
1097 					(fifo_count == 1) ? "" : "s",
1098 					fifo_dest);
1099 			musb_write_fifo(hw_ep, fifo_count, fifo_dest);
1100 
1101 			urb->actual_length += fifo_count;
1102 			more = true;
1103 		}
1104 		break;
1105 	default:
1106 		ERR("bogus ep0 stage %d\n", musb->ep0_stage);
1107 		break;
1108 	}
1109 
1110 	return more;
1111 }
1112 
1113 /*
1114  * Handle default endpoint interrupt as host. Only called in IRQ time
1115  * from musb_interrupt().
1116  *
1117  * called with controller irqlocked
1118  */
1119 irqreturn_t musb_h_ep0_irq(struct musb *musb)
1120 {
1121 	struct urb		*urb;
1122 	u16			csr, len;
1123 	int			status = 0;
1124 	void __iomem		*mbase = musb->mregs;
1125 	struct musb_hw_ep	*hw_ep = musb->control_ep;
1126 	void __iomem		*epio = hw_ep->regs;
1127 	struct musb_qh		*qh = hw_ep->in_qh;
1128 	bool			complete = false;
1129 	irqreturn_t		retval = IRQ_NONE;
1130 
1131 	/* ep0 only has one queue, "in" */
1132 	urb = next_urb(qh);
1133 
1134 	musb_ep_select(mbase, 0);
1135 	csr = musb_readw(epio, MUSB_CSR0);
1136 	len = (csr & MUSB_CSR0_RXPKTRDY)
1137 			? musb_readb(epio, MUSB_COUNT0)
1138 			: 0;
1139 
1140 	dev_dbg(musb->controller, "<== csr0 %04x, qh %p, count %d, urb %p, stage %d\n",
1141 		csr, qh, len, urb, musb->ep0_stage);
1142 
1143 	/* if we just did status stage, we are done */
1144 	if (MUSB_EP0_STATUS == musb->ep0_stage) {
1145 		retval = IRQ_HANDLED;
1146 		complete = true;
1147 	}
1148 
1149 	/* prepare status */
1150 	if (csr & MUSB_CSR0_H_RXSTALL) {
1151 		dev_dbg(musb->controller, "STALLING ENDPOINT\n");
1152 		status = -EPIPE;
1153 
1154 	} else if (csr & MUSB_CSR0_H_ERROR) {
1155 		dev_dbg(musb->controller, "no response, csr0 %04x\n", csr);
1156 		status = -EPROTO;
1157 
1158 	} else if (csr & MUSB_CSR0_H_NAKTIMEOUT) {
1159 		dev_dbg(musb->controller, "control NAK timeout\n");
1160 
1161 		/* NOTE:  this code path would be a good place to PAUSE a
1162 		 * control transfer, if another one is queued, so that
1163 		 * ep0 is more likely to stay busy.  That's already done
1164 		 * for bulk RX transfers.
1165 		 *
1166 		 * if (qh->ring.next != &musb->control), then
1167 		 * we have a candidate... NAKing is *NOT* an error
1168 		 */
1169 		musb_writew(epio, MUSB_CSR0, 0);
1170 		retval = IRQ_HANDLED;
1171 	}
1172 
1173 	if (status) {
1174 		dev_dbg(musb->controller, "aborting\n");
1175 		retval = IRQ_HANDLED;
1176 		if (urb)
1177 			urb->status = status;
1178 		complete = true;
1179 
1180 		/* use the proper sequence to abort the transfer */
1181 		if (csr & MUSB_CSR0_H_REQPKT) {
1182 			csr &= ~MUSB_CSR0_H_REQPKT;
1183 			musb_writew(epio, MUSB_CSR0, csr);
1184 			csr &= ~MUSB_CSR0_H_NAKTIMEOUT;
1185 			musb_writew(epio, MUSB_CSR0, csr);
1186 		} else {
1187 			musb_h_ep0_flush_fifo(hw_ep);
1188 		}
1189 
1190 		musb_writeb(epio, MUSB_NAKLIMIT0, 0);
1191 
1192 		/* clear it */
1193 		musb_writew(epio, MUSB_CSR0, 0);
1194 	}
1195 
1196 	if (unlikely(!urb)) {
1197 		/* stop endpoint since we have no place for its data, this
1198 		 * SHOULD NEVER HAPPEN! */
1199 		ERR("no URB for end 0\n");
1200 
1201 		musb_h_ep0_flush_fifo(hw_ep);
1202 		goto done;
1203 	}
1204 
1205 	if (!complete) {
1206 		/* call common logic and prepare response */
1207 		if (musb_h_ep0_continue(musb, len, urb)) {
1208 			/* more packets required */
1209 			csr = (MUSB_EP0_IN == musb->ep0_stage)
1210 				?  MUSB_CSR0_H_REQPKT : MUSB_CSR0_TXPKTRDY;
1211 		} else {
1212 			/* data transfer complete; perform status phase */
1213 			if (usb_pipeout(urb->pipe)
1214 					|| !urb->transfer_buffer_length)
1215 				csr = MUSB_CSR0_H_STATUSPKT
1216 					| MUSB_CSR0_H_REQPKT;
1217 			else
1218 				csr = MUSB_CSR0_H_STATUSPKT
1219 					| MUSB_CSR0_TXPKTRDY;
1220 
1221 			/* disable ping token in status phase */
1222 			csr |= MUSB_CSR0_H_DIS_PING;
1223 
1224 			/* flag status stage */
1225 			musb->ep0_stage = MUSB_EP0_STATUS;
1226 
1227 			dev_dbg(musb->controller, "ep0 STATUS, csr %04x\n", csr);
1228 
1229 		}
1230 		musb_writew(epio, MUSB_CSR0, csr);
1231 		retval = IRQ_HANDLED;
1232 	} else
1233 		musb->ep0_stage = MUSB_EP0_IDLE;
1234 
1235 	/* call completion handler if done */
1236 	if (complete)
1237 		musb_advance_schedule(musb, urb, hw_ep, 1);
1238 done:
1239 	return retval;
1240 }
1241 
1242 
1243 #ifdef CONFIG_USB_INVENTRA_DMA
1244 
1245 /* Host side TX (OUT) using Mentor DMA works as follows:
1246 	submit_urb ->
1247 		- if queue was empty, Program Endpoint
1248 		- ... which starts DMA to fifo in mode 1 or 0
1249 
1250 	DMA Isr (transfer complete) -> TxAvail()
1251 		- Stop DMA (~DmaEnab)	(<--- Alert ... currently happens
1252 					only in musb_cleanup_urb)
1253 		- TxPktRdy has to be set in mode 0 or for
1254 			short packets in mode 1.
1255 */
1256 
1257 #endif
1258 
1259 /* Service a Tx-Available or dma completion irq for the endpoint */
1260 void musb_host_tx(struct musb *musb, u8 epnum)
1261 {
1262 	int			pipe;
1263 	bool			done = false;
1264 	u16			tx_csr;
1265 	size_t			length = 0;
1266 	size_t			offset = 0;
1267 	struct musb_hw_ep	*hw_ep = musb->endpoints + epnum;
1268 	void __iomem		*epio = hw_ep->regs;
1269 	struct musb_qh		*qh = hw_ep->out_qh;
1270 	struct urb		*urb = next_urb(qh);
1271 	u32			status = 0;
1272 	void __iomem		*mbase = musb->mregs;
1273 	struct dma_channel	*dma;
1274 	bool			transfer_pending = false;
1275 
1276 	musb_ep_select(mbase, epnum);
1277 	tx_csr = musb_readw(epio, MUSB_TXCSR);
1278 
1279 	/* with CPPI, DMA sometimes triggers "extra" irqs */
1280 	if (!urb) {
1281 		dev_dbg(musb->controller, "extra TX%d ready, csr %04x\n", epnum, tx_csr);
1282 		return;
1283 	}
1284 
1285 	pipe = urb->pipe;
1286 	dma = is_dma_capable() ? hw_ep->tx_channel : NULL;
1287 	dev_dbg(musb->controller, "OUT/TX%d end, csr %04x%s\n", epnum, tx_csr,
1288 			dma ? ", dma" : "");
1289 
1290 	/* check for errors */
1291 	if (tx_csr & MUSB_TXCSR_H_RXSTALL) {
1292 		/* dma was disabled, fifo flushed */
1293 		dev_dbg(musb->controller, "TX end %d stall\n", epnum);
1294 
1295 		/* stall; record URB status */
1296 		status = -EPIPE;
1297 
1298 	} else if (tx_csr & MUSB_TXCSR_H_ERROR) {
1299 		/* (NON-ISO) dma was disabled, fifo flushed */
1300 		dev_dbg(musb->controller, "TX 3strikes on ep=%d\n", epnum);
1301 
1302 		status = -ETIMEDOUT;
1303 
1304 	} else if (tx_csr & MUSB_TXCSR_H_NAKTIMEOUT) {
1305 		if (USB_ENDPOINT_XFER_BULK == qh->type && qh->mux == 1
1306 				&& !list_is_singular(&musb->out_bulk)) {
1307 			dev_dbg(musb->controller,
1308 				"NAK timeout on TX%d ep\n", epnum);
1309 			musb_bulk_nak_timeout(musb, hw_ep, 0);
1310 		} else {
1311 			dev_dbg(musb->controller,
1312 				"TX end=%d device not responding\n", epnum);
1313 			/* NOTE:  this code path would be a good place to PAUSE a
1314 			 * transfer, if there's some other (nonperiodic) tx urb
1315 			 * that could use this fifo.  (dma complicates it...)
1316 			 * That's already done for bulk RX transfers.
1317 			 *
1318 			 * if (bulk && qh->ring.next != &musb->out_bulk), then
1319 			 * we have a candidate... NAKing is *NOT* an error
1320 			 */
1321 			musb_ep_select(mbase, epnum);
1322 			musb_writew(epio, MUSB_TXCSR,
1323 					MUSB_TXCSR_H_WZC_BITS
1324 					| MUSB_TXCSR_TXPKTRDY);
1325 		}
1326 			return;
1327 	}
1328 
1329 done:
1330 	if (status) {
1331 		if (dma_channel_status(dma) == MUSB_DMA_STATUS_BUSY) {
1332 			dma->status = MUSB_DMA_STATUS_CORE_ABORT;
1333 			musb->dma_controller->channel_abort(dma);
1334 		}
1335 
1336 		/* do the proper sequence to abort the transfer in the
1337 		 * usb core; the dma engine should already be stopped.
1338 		 */
1339 		musb_h_tx_flush_fifo(hw_ep);
1340 		tx_csr &= ~(MUSB_TXCSR_AUTOSET
1341 				| MUSB_TXCSR_DMAENAB
1342 				| MUSB_TXCSR_H_ERROR
1343 				| MUSB_TXCSR_H_RXSTALL
1344 				| MUSB_TXCSR_H_NAKTIMEOUT
1345 				);
1346 
1347 		musb_ep_select(mbase, epnum);
1348 		musb_writew(epio, MUSB_TXCSR, tx_csr);
1349 		/* REVISIT may need to clear FLUSHFIFO ... */
1350 		musb_writew(epio, MUSB_TXCSR, tx_csr);
1351 		musb_writeb(epio, MUSB_TXINTERVAL, 0);
1352 
1353 		done = true;
1354 	}
1355 
1356 	/* second cppi case */
1357 	if (dma_channel_status(dma) == MUSB_DMA_STATUS_BUSY) {
1358 		dev_dbg(musb->controller, "extra TX%d ready, csr %04x\n", epnum, tx_csr);
1359 		return;
1360 	}
1361 
1362 	if (is_dma_capable() && dma && !status) {
1363 		/*
1364 		 * DMA has completed.  But if we're using DMA mode 1 (multi
1365 		 * packet DMA), we need a terminal TXPKTRDY interrupt before
1366 		 * we can consider this transfer completed, lest we trash
1367 		 * its last packet when writing the next URB's data.  So we
1368 		 * switch back to mode 0 to get that interrupt; we'll come
1369 		 * back here once it happens.
1370 		 */
1371 		if (tx_csr & MUSB_TXCSR_DMAMODE) {
1372 			/*
1373 			 * We shouldn't clear DMAMODE with DMAENAB set; so
1374 			 * clear them in a safe order.  That should be OK
1375 			 * once TXPKTRDY has been set (and I've never seen
1376 			 * it being 0 at this moment -- DMA interrupt latency
1377 			 * is significant) but if it hasn't been then we have
1378 			 * no choice but to stop being polite and ignore the
1379 			 * programmer's guide... :-)
1380 			 *
1381 			 * Note that we must write TXCSR with TXPKTRDY cleared
1382 			 * in order not to re-trigger the packet send (this bit
1383 			 * can't be cleared by CPU), and there's another caveat:
1384 			 * TXPKTRDY may be set shortly and then cleared in the
1385 			 * double-buffered FIFO mode, so we do an extra TXCSR
1386 			 * read for debouncing...
1387 			 */
1388 			tx_csr &= musb_readw(epio, MUSB_TXCSR);
1389 			if (tx_csr & MUSB_TXCSR_TXPKTRDY) {
1390 				tx_csr &= ~(MUSB_TXCSR_DMAENAB |
1391 					    MUSB_TXCSR_TXPKTRDY);
1392 				musb_writew(epio, MUSB_TXCSR,
1393 					    tx_csr | MUSB_TXCSR_H_WZC_BITS);
1394 			}
1395 			tx_csr &= ~(MUSB_TXCSR_DMAMODE |
1396 				    MUSB_TXCSR_TXPKTRDY);
1397 			musb_writew(epio, MUSB_TXCSR,
1398 				    tx_csr | MUSB_TXCSR_H_WZC_BITS);
1399 
1400 			/*
1401 			 * There is no guarantee that we'll get an interrupt
1402 			 * after clearing DMAMODE as we might have done this
1403 			 * too late (after TXPKTRDY was cleared by controller).
1404 			 * Re-read TXCSR as we have spoiled its previous value.
1405 			 */
1406 			tx_csr = musb_readw(epio, MUSB_TXCSR);
1407 		}
1408 
1409 		/*
1410 		 * We may get here from a DMA completion or TXPKTRDY interrupt.
1411 		 * In any case, we must check the FIFO status here and bail out
1412 		 * only if the FIFO still has data -- that should prevent the
1413 		 * "missed" TXPKTRDY interrupts and deal with double-buffered
1414 		 * FIFO mode too...
1415 		 */
1416 		if (tx_csr & (MUSB_TXCSR_FIFONOTEMPTY | MUSB_TXCSR_TXPKTRDY)) {
1417 			dev_dbg(musb->controller, "DMA complete but packet still in FIFO, "
1418 			    "CSR %04x\n", tx_csr);
1419 			return;
1420 		}
1421 	}
1422 
1423 	if (!status || dma || usb_pipeisoc(pipe)) {
1424 		if (dma)
1425 			length = dma->actual_len;
1426 		else
1427 			length = qh->segsize;
1428 		qh->offset += length;
1429 
1430 		if (usb_pipeisoc(pipe)) {
1431 			struct usb_iso_packet_descriptor	*d;
1432 
1433 			d = urb->iso_frame_desc + qh->iso_idx;
1434 			d->actual_length = length;
1435 			d->status = status;
1436 			if (++qh->iso_idx >= urb->number_of_packets) {
1437 				done = true;
1438 			} else {
1439 				d++;
1440 				offset = d->offset;
1441 				length = d->length;
1442 			}
1443 		} else if (dma && urb->transfer_buffer_length == qh->offset) {
1444 			done = true;
1445 		} else {
1446 			/* see if we need to send more data, or ZLP */
1447 			if (qh->segsize < qh->maxpacket)
1448 				done = true;
1449 			else if (qh->offset == urb->transfer_buffer_length
1450 					&& !(urb->transfer_flags
1451 						& URB_ZERO_PACKET))
1452 				done = true;
1453 			if (!done) {
1454 				offset = qh->offset;
1455 				length = urb->transfer_buffer_length - offset;
1456 				transfer_pending = true;
1457 			}
1458 		}
1459 	}
1460 
1461 	/* urb->status != -EINPROGRESS means request has been faulted,
1462 	 * so we must abort this transfer after cleanup
1463 	 */
1464 	if (urb->status != -EINPROGRESS) {
1465 		done = true;
1466 		if (status == 0)
1467 			status = urb->status;
1468 	}
1469 
1470 	if (done) {
1471 		/* set status */
1472 		urb->status = status;
1473 		urb->actual_length = qh->offset;
1474 		musb_advance_schedule(musb, urb, hw_ep, USB_DIR_OUT);
1475 		return;
1476 	} else if ((usb_pipeisoc(pipe) || transfer_pending) && dma) {
1477 		if (musb_tx_dma_program(musb->dma_controller, hw_ep, qh, urb,
1478 				offset, length)) {
1479 			if (is_cppi_enabled(musb) || tusb_dma_omap(musb))
1480 				musb_h_tx_dma_start(hw_ep);
1481 			return;
1482 		}
1483 	} else	if (tx_csr & MUSB_TXCSR_DMAENAB) {
1484 		dev_dbg(musb->controller, "not complete, but DMA enabled?\n");
1485 		return;
1486 	}
1487 
1488 	/*
1489 	 * PIO: start next packet in this URB.
1490 	 *
1491 	 * REVISIT: some docs say that when hw_ep->tx_double_buffered,
1492 	 * (and presumably, FIFO is not half-full) we should write *two*
1493 	 * packets before updating TXCSR; other docs disagree...
1494 	 */
1495 	if (length > qh->maxpacket)
1496 		length = qh->maxpacket;
1497 	/* Unmap the buffer so that CPU can use it */
1498 	usb_hcd_unmap_urb_for_dma(musb->hcd, urb);
1499 
1500 	/*
1501 	 * We need to map sg if the transfer_buffer is
1502 	 * NULL.
1503 	 */
1504 	if (!urb->transfer_buffer)
1505 		qh->use_sg = true;
1506 
1507 	if (qh->use_sg) {
1508 		/* sg_miter_start is already done in musb_ep_program */
1509 		if (!sg_miter_next(&qh->sg_miter)) {
1510 			dev_err(musb->controller, "error: sg list empty\n");
1511 			sg_miter_stop(&qh->sg_miter);
1512 			status = -EINVAL;
1513 			goto done;
1514 		}
1515 		urb->transfer_buffer = qh->sg_miter.addr;
1516 		length = min_t(u32, length, qh->sg_miter.length);
1517 		musb_write_fifo(hw_ep, length, urb->transfer_buffer);
1518 		qh->sg_miter.consumed = length;
1519 		sg_miter_stop(&qh->sg_miter);
1520 	} else {
1521 		musb_write_fifo(hw_ep, length, urb->transfer_buffer + offset);
1522 	}
1523 
1524 	qh->segsize = length;
1525 
1526 	if (qh->use_sg) {
1527 		if (offset + length >= urb->transfer_buffer_length)
1528 			qh->use_sg = false;
1529 	}
1530 
1531 	musb_ep_select(mbase, epnum);
1532 	musb_writew(epio, MUSB_TXCSR,
1533 			MUSB_TXCSR_H_WZC_BITS | MUSB_TXCSR_TXPKTRDY);
1534 }
1535 
1536 #ifdef CONFIG_USB_TI_CPPI41_DMA
1537 /* Seems to set up ISO for cppi41 and not advance len. See commit c57c41d */
1538 static int musb_rx_dma_iso_cppi41(struct dma_controller *dma,
1539 				  struct musb_hw_ep *hw_ep,
1540 				  struct musb_qh *qh,
1541 				  struct urb *urb,
1542 				  size_t len)
1543 {
1544 	struct dma_channel *channel = hw_ep->tx_channel;
1545 	void __iomem *epio = hw_ep->regs;
1546 	dma_addr_t *buf;
1547 	u32 length, res;
1548 	u16 val;
1549 
1550 	buf = (void *)urb->iso_frame_desc[qh->iso_idx].offset +
1551 		(u32)urb->transfer_dma;
1552 
1553 	length = urb->iso_frame_desc[qh->iso_idx].length;
1554 
1555 	val = musb_readw(epio, MUSB_RXCSR);
1556 	val |= MUSB_RXCSR_DMAENAB;
1557 	musb_writew(hw_ep->regs, MUSB_RXCSR, val);
1558 
1559 	res = dma->channel_program(channel, qh->maxpacket, 0,
1560 				   (u32)buf, length);
1561 
1562 	return res;
1563 }
1564 #else
1565 static inline int musb_rx_dma_iso_cppi41(struct dma_controller *dma,
1566 					 struct musb_hw_ep *hw_ep,
1567 					 struct musb_qh *qh,
1568 					 struct urb *urb,
1569 					 size_t len)
1570 {
1571 	return false;
1572 }
1573 #endif
1574 
1575 #if defined(CONFIG_USB_INVENTRA_DMA) || defined(CONFIG_USB_UX500_DMA) || \
1576 	defined(CONFIG_USB_TI_CPPI41_DMA)
1577 /* Host side RX (IN) using Mentor DMA works as follows:
1578 	submit_urb ->
1579 		- if queue was empty, ProgramEndpoint
1580 		- first IN token is sent out (by setting ReqPkt)
1581 	LinuxIsr -> RxReady()
1582 	/\	=> first packet is received
1583 	|	- Set in mode 0 (DmaEnab, ~ReqPkt)
1584 	|		-> DMA Isr (transfer complete) -> RxReady()
1585 	|		    - Ack receive (~RxPktRdy), turn off DMA (~DmaEnab)
1586 	|		    - if urb not complete, send next IN token (ReqPkt)
1587 	|			   |		else complete urb.
1588 	|			   |
1589 	---------------------------
1590  *
1591  * Nuances of mode 1:
1592  *	For short packets, no ack (+RxPktRdy) is sent automatically
1593  *	(even if AutoClear is ON)
1594  *	For full packets, ack (~RxPktRdy) and next IN token (+ReqPkt) is sent
1595  *	automatically => major problem, as collecting the next packet becomes
1596  *	difficult. Hence mode 1 is not used.
1597  *
1598  * REVISIT
1599  *	All we care about at this driver level is that
1600  *       (a) all URBs terminate with REQPKT cleared and fifo(s) empty;
1601  *       (b) termination conditions are: short RX, or buffer full;
1602  *       (c) fault modes include
1603  *           - iff URB_SHORT_NOT_OK, short RX status is -EREMOTEIO.
1604  *             (and that endpoint's dma queue stops immediately)
1605  *           - overflow (full, PLUS more bytes in the terminal packet)
1606  *
1607  *	So for example, usb-storage sets URB_SHORT_NOT_OK, and would
1608  *	thus be a great candidate for using mode 1 ... for all but the
1609  *	last packet of one URB's transfer.
1610  */
1611 static int musb_rx_dma_inventra_cppi41(struct dma_controller *dma,
1612 				       struct musb_hw_ep *hw_ep,
1613 				       struct musb_qh *qh,
1614 				       struct urb *urb,
1615 				       size_t len)
1616 {
1617 	struct dma_channel *channel = hw_ep->rx_channel;
1618 	void __iomem *epio = hw_ep->regs;
1619 	u16 val;
1620 	int pipe;
1621 	bool done;
1622 
1623 	pipe = urb->pipe;
1624 
1625 	if (usb_pipeisoc(pipe)) {
1626 		struct usb_iso_packet_descriptor *d;
1627 
1628 		d = urb->iso_frame_desc + qh->iso_idx;
1629 		d->actual_length = len;
1630 
1631 		/* even if there was an error, we did the dma
1632 		 * for iso_frame_desc->length
1633 		 */
1634 		if (d->status != -EILSEQ && d->status != -EOVERFLOW)
1635 			d->status = 0;
1636 
1637 		if (++qh->iso_idx >= urb->number_of_packets) {
1638 			done = true;
1639 		} else {
1640 			/* REVISIT: Why ignore return value here? */
1641 			if (musb_dma_cppi41(hw_ep->musb))
1642 				done = musb_rx_dma_iso_cppi41(dma, hw_ep, qh,
1643 							      urb, len);
1644 			done = false;
1645 		}
1646 
1647 	} else  {
1648 		/* done if urb buffer is full or short packet is recd */
1649 		done = (urb->actual_length + len >=
1650 			urb->transfer_buffer_length
1651 			|| channel->actual_len < qh->maxpacket
1652 			|| channel->rx_packet_done);
1653 	}
1654 
1655 	/* send IN token for next packet, without AUTOREQ */
1656 	if (!done) {
1657 		val = musb_readw(epio, MUSB_RXCSR);
1658 		val |= MUSB_RXCSR_H_REQPKT;
1659 		musb_writew(epio, MUSB_RXCSR, MUSB_RXCSR_H_WZC_BITS | val);
1660 	}
1661 
1662 	return done;
1663 }
1664 
1665 /* Disadvantage of using mode 1:
1666  *	It's basically usable only for mass storage class; essentially all
1667  *	other protocols also terminate transfers on short packets.
1668  *
1669  * Details:
1670  *	An extra IN token is sent at the end of the transfer (due to AUTOREQ)
1671  *	If you try to use mode 1 for (transfer_buffer_length - 512), and try
1672  *	to use the extra IN token to grab the last packet using mode 0, then
1673  *	the problem is that you cannot be sure when the device will send the
1674  *	last packet and RxPktRdy set. Sometimes the packet is recd too soon
1675  *	such that it gets lost when RxCSR is re-set at the end of the mode 1
1676  *	transfer, while sometimes it is recd just a little late so that if you
1677  *	try to configure for mode 0 soon after the mode 1 transfer is
1678  *	completed, you will find rxcount 0. Okay, so you might think why not
1679  *	wait for an interrupt when the pkt is recd. Well, you won't get any!
1680  */
1681 static int musb_rx_dma_in_inventra_cppi41(struct dma_controller *dma,
1682 					  struct musb_hw_ep *hw_ep,
1683 					  struct musb_qh *qh,
1684 					  struct urb *urb,
1685 					  size_t len,
1686 					  u8 iso_err)
1687 {
1688 	struct musb *musb = hw_ep->musb;
1689 	void __iomem *epio = hw_ep->regs;
1690 	struct dma_channel *channel = hw_ep->rx_channel;
1691 	u16 rx_count, val;
1692 	int length, pipe, done;
1693 	dma_addr_t buf;
1694 
1695 	rx_count = musb_readw(epio, MUSB_RXCOUNT);
1696 	pipe = urb->pipe;
1697 
1698 	if (usb_pipeisoc(pipe)) {
1699 		int d_status = 0;
1700 		struct usb_iso_packet_descriptor *d;
1701 
1702 		d = urb->iso_frame_desc + qh->iso_idx;
1703 
1704 		if (iso_err) {
1705 			d_status = -EILSEQ;
1706 			urb->error_count++;
1707 		}
1708 		if (rx_count > d->length) {
1709 			if (d_status == 0) {
1710 				d_status = -EOVERFLOW;
1711 				urb->error_count++;
1712 			}
1713 			dev_dbg(musb->controller, "** OVERFLOW %d into %d\n",
1714 				rx_count, d->length);
1715 
1716 			length = d->length;
1717 		} else
1718 			length = rx_count;
1719 		d->status = d_status;
1720 		buf = urb->transfer_dma + d->offset;
1721 	} else {
1722 		length = rx_count;
1723 		buf = urb->transfer_dma + urb->actual_length;
1724 	}
1725 
1726 	channel->desired_mode = 0;
1727 #ifdef USE_MODE1
1728 	/* because of the issue below, mode 1 will
1729 	 * only rarely behave with correct semantics.
1730 	 */
1731 	if ((urb->transfer_flags & URB_SHORT_NOT_OK)
1732 	    && (urb->transfer_buffer_length - urb->actual_length)
1733 	    > qh->maxpacket)
1734 		channel->desired_mode = 1;
1735 	if (rx_count < hw_ep->max_packet_sz_rx) {
1736 		length = rx_count;
1737 		channel->desired_mode = 0;
1738 	} else {
1739 		length = urb->transfer_buffer_length;
1740 	}
1741 #endif
1742 
1743 	/* See comments above on disadvantages of using mode 1 */
1744 	val = musb_readw(epio, MUSB_RXCSR);
1745 	val &= ~MUSB_RXCSR_H_REQPKT;
1746 
1747 	if (channel->desired_mode == 0)
1748 		val &= ~MUSB_RXCSR_H_AUTOREQ;
1749 	else
1750 		val |= MUSB_RXCSR_H_AUTOREQ;
1751 	val |= MUSB_RXCSR_DMAENAB;
1752 
1753 	/* autoclear shouldn't be set in high bandwidth */
1754 	if (qh->hb_mult == 1)
1755 		val |= MUSB_RXCSR_AUTOCLEAR;
1756 
1757 	musb_writew(epio, MUSB_RXCSR, MUSB_RXCSR_H_WZC_BITS | val);
1758 
1759 	/* REVISIT if when actual_length != 0,
1760 	 * transfer_buffer_length needs to be
1761 	 * adjusted first...
1762 	 */
1763 	done = dma->channel_program(channel, qh->maxpacket,
1764 				   channel->desired_mode,
1765 				   buf, length);
1766 
1767 	if (!done) {
1768 		dma->channel_release(channel);
1769 		hw_ep->rx_channel = NULL;
1770 		channel = NULL;
1771 		val = musb_readw(epio, MUSB_RXCSR);
1772 		val &= ~(MUSB_RXCSR_DMAENAB
1773 			 | MUSB_RXCSR_H_AUTOREQ
1774 			 | MUSB_RXCSR_AUTOCLEAR);
1775 		musb_writew(epio, MUSB_RXCSR, val);
1776 	}
1777 
1778 	return done;
1779 }
1780 #else
1781 static inline int musb_rx_dma_inventra_cppi41(struct dma_controller *dma,
1782 					      struct musb_hw_ep *hw_ep,
1783 					      struct musb_qh *qh,
1784 					      struct urb *urb,
1785 					      size_t len)
1786 {
1787 	return false;
1788 }
1789 
1790 static inline int musb_rx_dma_in_inventra_cppi41(struct dma_controller *dma,
1791 						 struct musb_hw_ep *hw_ep,
1792 						 struct musb_qh *qh,
1793 						 struct urb *urb,
1794 						 size_t len,
1795 						 u8 iso_err)
1796 {
1797 	return false;
1798 }
1799 #endif
1800 
1801 /*
1802  * Service an RX interrupt for the given IN endpoint; docs cover bulk, iso,
1803  * and high-bandwidth IN transfer cases.
1804  */
1805 void musb_host_rx(struct musb *musb, u8 epnum)
1806 {
1807 	struct urb		*urb;
1808 	struct musb_hw_ep	*hw_ep = musb->endpoints + epnum;
1809 	struct dma_controller	*c = musb->dma_controller;
1810 	void __iomem		*epio = hw_ep->regs;
1811 	struct musb_qh		*qh = hw_ep->in_qh;
1812 	size_t			xfer_len;
1813 	void __iomem		*mbase = musb->mregs;
1814 	int			pipe;
1815 	u16			rx_csr, val;
1816 	bool			iso_err = false;
1817 	bool			done = false;
1818 	u32			status;
1819 	struct dma_channel	*dma;
1820 	unsigned int sg_flags = SG_MITER_ATOMIC | SG_MITER_TO_SG;
1821 
1822 	musb_ep_select(mbase, epnum);
1823 
1824 	urb = next_urb(qh);
1825 	dma = is_dma_capable() ? hw_ep->rx_channel : NULL;
1826 	status = 0;
1827 	xfer_len = 0;
1828 
1829 	rx_csr = musb_readw(epio, MUSB_RXCSR);
1830 	val = rx_csr;
1831 
1832 	if (unlikely(!urb)) {
1833 		/* REVISIT -- THIS SHOULD NEVER HAPPEN ... but, at least
1834 		 * usbtest #11 (unlinks) triggers it regularly, sometimes
1835 		 * with fifo full.  (Only with DMA??)
1836 		 */
1837 		dev_dbg(musb->controller, "BOGUS RX%d ready, csr %04x, count %d\n", epnum, val,
1838 			musb_readw(epio, MUSB_RXCOUNT));
1839 		musb_h_flush_rxfifo(hw_ep, MUSB_RXCSR_CLRDATATOG);
1840 		return;
1841 	}
1842 
1843 	pipe = urb->pipe;
1844 
1845 	dev_dbg(musb->controller, "<== hw %d rxcsr %04x, urb actual %d (+dma %zu)\n",
1846 		epnum, rx_csr, urb->actual_length,
1847 		dma ? dma->actual_len : 0);
1848 
1849 	/* check for errors, concurrent stall & unlink is not really
1850 	 * handled yet! */
1851 	if (rx_csr & MUSB_RXCSR_H_RXSTALL) {
1852 		dev_dbg(musb->controller, "RX end %d STALL\n", epnum);
1853 
1854 		/* stall; record URB status */
1855 		status = -EPIPE;
1856 
1857 	} else if (rx_csr & MUSB_RXCSR_H_ERROR) {
1858 		dev_dbg(musb->controller, "end %d RX proto error\n", epnum);
1859 
1860 		status = -EPROTO;
1861 		musb_writeb(epio, MUSB_RXINTERVAL, 0);
1862 
1863 	} else if (rx_csr & MUSB_RXCSR_DATAERROR) {
1864 
1865 		if (USB_ENDPOINT_XFER_ISOC != qh->type) {
1866 			dev_dbg(musb->controller, "RX end %d NAK timeout\n", epnum);
1867 
1868 			/* NOTE: NAKing is *NOT* an error, so we want to
1869 			 * continue.  Except ... if there's a request for
1870 			 * another QH, use that instead of starving it.
1871 			 *
1872 			 * Devices like Ethernet and serial adapters keep
1873 			 * reads posted at all times, which will starve
1874 			 * other devices without this logic.
1875 			 */
1876 			if (usb_pipebulk(urb->pipe)
1877 					&& qh->mux == 1
1878 					&& !list_is_singular(&musb->in_bulk)) {
1879 				musb_bulk_nak_timeout(musb, hw_ep, 1);
1880 				return;
1881 			}
1882 			musb_ep_select(mbase, epnum);
1883 			rx_csr |= MUSB_RXCSR_H_WZC_BITS;
1884 			rx_csr &= ~MUSB_RXCSR_DATAERROR;
1885 			musb_writew(epio, MUSB_RXCSR, rx_csr);
1886 
1887 			goto finish;
1888 		} else {
1889 			dev_dbg(musb->controller, "RX end %d ISO data error\n", epnum);
1890 			/* packet error reported later */
1891 			iso_err = true;
1892 		}
1893 	} else if (rx_csr & MUSB_RXCSR_INCOMPRX) {
1894 		dev_dbg(musb->controller, "end %d high bandwidth incomplete ISO packet RX\n",
1895 				epnum);
1896 		status = -EPROTO;
1897 	}
1898 
1899 	/* faults abort the transfer */
1900 	if (status) {
1901 		/* clean up dma and collect transfer count */
1902 		if (dma_channel_status(dma) == MUSB_DMA_STATUS_BUSY) {
1903 			dma->status = MUSB_DMA_STATUS_CORE_ABORT;
1904 			musb->dma_controller->channel_abort(dma);
1905 			xfer_len = dma->actual_len;
1906 		}
1907 		musb_h_flush_rxfifo(hw_ep, MUSB_RXCSR_CLRDATATOG);
1908 		musb_writeb(epio, MUSB_RXINTERVAL, 0);
1909 		done = true;
1910 		goto finish;
1911 	}
1912 
1913 	if (unlikely(dma_channel_status(dma) == MUSB_DMA_STATUS_BUSY)) {
1914 		/* SHOULD NEVER HAPPEN ... but at least DaVinci has done it */
1915 		ERR("RX%d dma busy, csr %04x\n", epnum, rx_csr);
1916 		goto finish;
1917 	}
1918 
1919 	/* thorough shutdown for now ... given more precise fault handling
1920 	 * and better queueing support, we might keep a DMA pipeline going
1921 	 * while processing this irq for earlier completions.
1922 	 */
1923 
1924 	/* FIXME this is _way_ too much in-line logic for Mentor DMA */
1925 	if (!musb_dma_inventra(musb) && !musb_dma_ux500(musb) &&
1926 	    (rx_csr & MUSB_RXCSR_H_REQPKT)) {
1927 		/* REVISIT this happened for a while on some short reads...
1928 		 * the cleanup still needs investigation... looks bad...
1929 		 * and also duplicates dma cleanup code above ... plus,
1930 		 * shouldn't this be the "half full" double buffer case?
1931 		 */
1932 		if (dma_channel_status(dma) == MUSB_DMA_STATUS_BUSY) {
1933 			dma->status = MUSB_DMA_STATUS_CORE_ABORT;
1934 			musb->dma_controller->channel_abort(dma);
1935 			xfer_len = dma->actual_len;
1936 			done = true;
1937 		}
1938 
1939 		dev_dbg(musb->controller, "RXCSR%d %04x, reqpkt, len %zu%s\n", epnum, rx_csr,
1940 				xfer_len, dma ? ", dma" : "");
1941 		rx_csr &= ~MUSB_RXCSR_H_REQPKT;
1942 
1943 		musb_ep_select(mbase, epnum);
1944 		musb_writew(epio, MUSB_RXCSR,
1945 				MUSB_RXCSR_H_WZC_BITS | rx_csr);
1946 	}
1947 
1948 	if (dma && (rx_csr & MUSB_RXCSR_DMAENAB)) {
1949 		xfer_len = dma->actual_len;
1950 
1951 		val &= ~(MUSB_RXCSR_DMAENAB
1952 			| MUSB_RXCSR_H_AUTOREQ
1953 			| MUSB_RXCSR_AUTOCLEAR
1954 			| MUSB_RXCSR_RXPKTRDY);
1955 		musb_writew(hw_ep->regs, MUSB_RXCSR, val);
1956 
1957 		if (musb_dma_inventra(musb) || musb_dma_ux500(musb) ||
1958 		    musb_dma_cppi41(musb)) {
1959 			    done = musb_rx_dma_inventra_cppi41(c, hw_ep, qh, urb, xfer_len);
1960 			    dev_dbg(hw_ep->musb->controller,
1961 				    "ep %d dma %s, rxcsr %04x, rxcount %d\n",
1962 				    epnum, done ? "off" : "reset",
1963 				    musb_readw(epio, MUSB_RXCSR),
1964 				    musb_readw(epio, MUSB_RXCOUNT));
1965 		} else {
1966 			done = true;
1967 		}
1968 
1969 	} else if (urb->status == -EINPROGRESS) {
1970 		/* if no errors, be sure a packet is ready for unloading */
1971 		if (unlikely(!(rx_csr & MUSB_RXCSR_RXPKTRDY))) {
1972 			status = -EPROTO;
1973 			ERR("Rx interrupt with no errors or packet!\n");
1974 
1975 			/* FIXME this is another "SHOULD NEVER HAPPEN" */
1976 
1977 /* SCRUB (RX) */
1978 			/* do the proper sequence to abort the transfer */
1979 			musb_ep_select(mbase, epnum);
1980 			val &= ~MUSB_RXCSR_H_REQPKT;
1981 			musb_writew(epio, MUSB_RXCSR, val);
1982 			goto finish;
1983 		}
1984 
1985 		/* we are expecting IN packets */
1986 		if ((musb_dma_inventra(musb) || musb_dma_ux500(musb) ||
1987 		    musb_dma_cppi41(musb)) && dma) {
1988 			dev_dbg(hw_ep->musb->controller,
1989 				"RX%d count %d, buffer 0x%llx len %d/%d\n",
1990 				epnum, musb_readw(epio, MUSB_RXCOUNT),
1991 				(unsigned long long) urb->transfer_dma
1992 				+ urb->actual_length,
1993 				qh->offset,
1994 				urb->transfer_buffer_length);
1995 
1996 			done = musb_rx_dma_in_inventra_cppi41(c, hw_ep, qh,
1997 							      urb, xfer_len,
1998 							      iso_err);
1999 			if (done)
2000 				goto finish;
2001 			else
2002 				dev_err(musb->controller, "error: rx_dma failed\n");
2003 		}
2004 
2005 		if (!dma) {
2006 			unsigned int received_len;
2007 
2008 			/* Unmap the buffer so that CPU can use it */
2009 			usb_hcd_unmap_urb_for_dma(musb->hcd, urb);
2010 
2011 			/*
2012 			 * We need to map sg if the transfer_buffer is
2013 			 * NULL.
2014 			 */
2015 			if (!urb->transfer_buffer) {
2016 				qh->use_sg = true;
2017 				sg_miter_start(&qh->sg_miter, urb->sg, 1,
2018 						sg_flags);
2019 			}
2020 
2021 			if (qh->use_sg) {
2022 				if (!sg_miter_next(&qh->sg_miter)) {
2023 					dev_err(musb->controller, "error: sg list empty\n");
2024 					sg_miter_stop(&qh->sg_miter);
2025 					status = -EINVAL;
2026 					done = true;
2027 					goto finish;
2028 				}
2029 				urb->transfer_buffer = qh->sg_miter.addr;
2030 				received_len = urb->actual_length;
2031 				qh->offset = 0x0;
2032 				done = musb_host_packet_rx(musb, urb, epnum,
2033 						iso_err);
2034 				/* Calculate the number of bytes received */
2035 				received_len = urb->actual_length -
2036 					received_len;
2037 				qh->sg_miter.consumed = received_len;
2038 				sg_miter_stop(&qh->sg_miter);
2039 			} else {
2040 				done = musb_host_packet_rx(musb, urb,
2041 						epnum, iso_err);
2042 			}
2043 			dev_dbg(musb->controller, "read %spacket\n", done ? "last " : "");
2044 		}
2045 	}
2046 
2047 finish:
2048 	urb->actual_length += xfer_len;
2049 	qh->offset += xfer_len;
2050 	if (done) {
2051 		if (qh->use_sg)
2052 			qh->use_sg = false;
2053 
2054 		if (urb->status == -EINPROGRESS)
2055 			urb->status = status;
2056 		musb_advance_schedule(musb, urb, hw_ep, USB_DIR_IN);
2057 	}
2058 }
2059 
2060 /* schedule nodes correspond to peripheral endpoints, like an OHCI QH.
2061  * the software schedule associates multiple such nodes with a given
2062  * host side hardware endpoint + direction; scheduling may activate
2063  * that hardware endpoint.
2064  */
2065 static int musb_schedule(
2066 	struct musb		*musb,
2067 	struct musb_qh		*qh,
2068 	int			is_in)
2069 {
2070 	int			idle = 0;
2071 	int			best_diff;
2072 	int			best_end, epnum;
2073 	struct musb_hw_ep	*hw_ep = NULL;
2074 	struct list_head	*head = NULL;
2075 	u8			toggle;
2076 	u8			txtype;
2077 	struct urb		*urb = next_urb(qh);
2078 
2079 	/* use fixed hardware for control and bulk */
2080 	if (qh->type == USB_ENDPOINT_XFER_CONTROL) {
2081 		head = &musb->control;
2082 		hw_ep = musb->control_ep;
2083 		goto success;
2084 	}
2085 
2086 	/* else, periodic transfers get muxed to other endpoints */
2087 
2088 	/*
2089 	 * We know this qh hasn't been scheduled, so all we need to do
2090 	 * is choose which hardware endpoint to put it on ...
2091 	 *
2092 	 * REVISIT what we really want here is a regular schedule tree
2093 	 * like e.g. OHCI uses.
2094 	 */
2095 	best_diff = 4096;
2096 	best_end = -1;
2097 
2098 	for (epnum = 1, hw_ep = musb->endpoints + 1;
2099 			epnum < musb->nr_endpoints;
2100 			epnum++, hw_ep++) {
2101 		int	diff;
2102 
2103 		if (musb_ep_get_qh(hw_ep, is_in) != NULL)
2104 			continue;
2105 
2106 		if (hw_ep == musb->bulk_ep)
2107 			continue;
2108 
2109 		if (is_in)
2110 			diff = hw_ep->max_packet_sz_rx;
2111 		else
2112 			diff = hw_ep->max_packet_sz_tx;
2113 		diff -= (qh->maxpacket * qh->hb_mult);
2114 
2115 		if (diff >= 0 && best_diff > diff) {
2116 
2117 			/*
2118 			 * Mentor controller has a bug in that if we schedule
2119 			 * a BULK Tx transfer on an endpoint that had earlier
2120 			 * handled ISOC then the BULK transfer has to start on
2121 			 * a zero toggle.  If the BULK transfer starts on a 1
2122 			 * toggle then this transfer will fail as the mentor
2123 			 * controller starts the Bulk transfer on a 0 toggle
2124 			 * irrespective of the programming of the toggle bits
2125 			 * in the TXCSR register.  Check for this condition
2126 			 * while allocating the EP for a Tx Bulk transfer.  If
2127 			 * so skip this EP.
2128 			 */
2129 			hw_ep = musb->endpoints + epnum;
2130 			toggle = usb_gettoggle(urb->dev, qh->epnum, !is_in);
2131 			txtype = (musb_readb(hw_ep->regs, MUSB_TXTYPE)
2132 					>> 4) & 0x3;
2133 			if (!is_in && (qh->type == USB_ENDPOINT_XFER_BULK) &&
2134 				toggle && (txtype == USB_ENDPOINT_XFER_ISOC))
2135 				continue;
2136 
2137 			best_diff = diff;
2138 			best_end = epnum;
2139 		}
2140 	}
2141 	/* use bulk reserved ep1 if no other ep is free */
2142 	if (best_end < 0 && qh->type == USB_ENDPOINT_XFER_BULK) {
2143 		hw_ep = musb->bulk_ep;
2144 		if (is_in)
2145 			head = &musb->in_bulk;
2146 		else
2147 			head = &musb->out_bulk;
2148 
2149 		/* Enable bulk RX/TX NAK timeout scheme when bulk requests are
2150 		 * multiplexed. This scheme does not work in high speed to full
2151 		 * speed scenario as NAK interrupts are not coming from a
2152 		 * full speed device connected to a high speed device.
2153 		 * NAK timeout interval is 8 (128 uframe or 16ms) for HS and
2154 		 * 4 (8 frame or 8ms) for FS device.
2155 		 */
2156 		if (qh->dev)
2157 			qh->intv_reg =
2158 				(USB_SPEED_HIGH == qh->dev->speed) ? 8 : 4;
2159 		goto success;
2160 	} else if (best_end < 0) {
2161 		return -ENOSPC;
2162 	}
2163 
2164 	idle = 1;
2165 	qh->mux = 0;
2166 	hw_ep = musb->endpoints + best_end;
2167 	dev_dbg(musb->controller, "qh %p periodic slot %d\n", qh, best_end);
2168 success:
2169 	if (head) {
2170 		idle = list_empty(head);
2171 		list_add_tail(&qh->ring, head);
2172 		qh->mux = 1;
2173 	}
2174 	qh->hw_ep = hw_ep;
2175 	qh->hep->hcpriv = qh;
2176 	if (idle)
2177 		musb_start_urb(musb, is_in, qh);
2178 	return 0;
2179 }
2180 
2181 static int musb_urb_enqueue(
2182 	struct usb_hcd			*hcd,
2183 	struct urb			*urb,
2184 	gfp_t				mem_flags)
2185 {
2186 	unsigned long			flags;
2187 	struct musb			*musb = hcd_to_musb(hcd);
2188 	struct usb_host_endpoint	*hep = urb->ep;
2189 	struct musb_qh			*qh;
2190 	struct usb_endpoint_descriptor	*epd = &hep->desc;
2191 	int				ret;
2192 	unsigned			type_reg;
2193 	unsigned			interval;
2194 
2195 	/* host role must be active */
2196 	if (!is_host_active(musb) || !musb->is_active)
2197 		return -ENODEV;
2198 
2199 	spin_lock_irqsave(&musb->lock, flags);
2200 	ret = usb_hcd_link_urb_to_ep(hcd, urb);
2201 	qh = ret ? NULL : hep->hcpriv;
2202 	if (qh)
2203 		urb->hcpriv = qh;
2204 	spin_unlock_irqrestore(&musb->lock, flags);
2205 
2206 	/* DMA mapping was already done, if needed, and this urb is on
2207 	 * hep->urb_list now ... so we're done, unless hep wasn't yet
2208 	 * scheduled onto a live qh.
2209 	 *
2210 	 * REVISIT best to keep hep->hcpriv valid until the endpoint gets
2211 	 * disabled, testing for empty qh->ring and avoiding qh setup costs
2212 	 * except for the first urb queued after a config change.
2213 	 */
2214 	if (qh || ret)
2215 		return ret;
2216 
2217 	/* Allocate and initialize qh, minimizing the work done each time
2218 	 * hw_ep gets reprogrammed, or with irqs blocked.  Then schedule it.
2219 	 *
2220 	 * REVISIT consider a dedicated qh kmem_cache, so it's harder
2221 	 * for bugs in other kernel code to break this driver...
2222 	 */
2223 	qh = kzalloc(sizeof *qh, mem_flags);
2224 	if (!qh) {
2225 		spin_lock_irqsave(&musb->lock, flags);
2226 		usb_hcd_unlink_urb_from_ep(hcd, urb);
2227 		spin_unlock_irqrestore(&musb->lock, flags);
2228 		return -ENOMEM;
2229 	}
2230 
2231 	qh->hep = hep;
2232 	qh->dev = urb->dev;
2233 	INIT_LIST_HEAD(&qh->ring);
2234 	qh->is_ready = 1;
2235 
2236 	qh->maxpacket = usb_endpoint_maxp(epd);
2237 	qh->type = usb_endpoint_type(epd);
2238 
2239 	/* Bits 11 & 12 of wMaxPacketSize encode high bandwidth multiplier.
2240 	 * Some musb cores don't support high bandwidth ISO transfers; and
2241 	 * we don't (yet!) support high bandwidth interrupt transfers.
2242 	 */
2243 	qh->hb_mult = 1 + ((qh->maxpacket >> 11) & 0x03);
2244 	if (qh->hb_mult > 1) {
2245 		int ok = (qh->type == USB_ENDPOINT_XFER_ISOC);
2246 
2247 		if (ok)
2248 			ok = (usb_pipein(urb->pipe) && musb->hb_iso_rx)
2249 				|| (usb_pipeout(urb->pipe) && musb->hb_iso_tx);
2250 		if (!ok) {
2251 			ret = -EMSGSIZE;
2252 			goto done;
2253 		}
2254 		qh->maxpacket &= 0x7ff;
2255 	}
2256 
2257 	qh->epnum = usb_endpoint_num(epd);
2258 
2259 	/* NOTE: urb->dev->devnum is wrong during SET_ADDRESS */
2260 	qh->addr_reg = (u8) usb_pipedevice(urb->pipe);
2261 
2262 	/* precompute rxtype/txtype/type0 register */
2263 	type_reg = (qh->type << 4) | qh->epnum;
2264 	switch (urb->dev->speed) {
2265 	case USB_SPEED_LOW:
2266 		type_reg |= 0xc0;
2267 		break;
2268 	case USB_SPEED_FULL:
2269 		type_reg |= 0x80;
2270 		break;
2271 	default:
2272 		type_reg |= 0x40;
2273 	}
2274 	qh->type_reg = type_reg;
2275 
2276 	/* Precompute RXINTERVAL/TXINTERVAL register */
2277 	switch (qh->type) {
2278 	case USB_ENDPOINT_XFER_INT:
2279 		/*
2280 		 * Full/low speeds use the  linear encoding,
2281 		 * high speed uses the logarithmic encoding.
2282 		 */
2283 		if (urb->dev->speed <= USB_SPEED_FULL) {
2284 			interval = max_t(u8, epd->bInterval, 1);
2285 			break;
2286 		}
2287 		/* FALLTHROUGH */
2288 	case USB_ENDPOINT_XFER_ISOC:
2289 		/* ISO always uses logarithmic encoding */
2290 		interval = min_t(u8, epd->bInterval, 16);
2291 		break;
2292 	default:
2293 		/* REVISIT we actually want to use NAK limits, hinting to the
2294 		 * transfer scheduling logic to try some other qh, e.g. try
2295 		 * for 2 msec first:
2296 		 *
2297 		 * interval = (USB_SPEED_HIGH == urb->dev->speed) ? 16 : 2;
2298 		 *
2299 		 * The downside of disabling this is that transfer scheduling
2300 		 * gets VERY unfair for nonperiodic transfers; a misbehaving
2301 		 * peripheral could make that hurt.  That's perfectly normal
2302 		 * for reads from network or serial adapters ... so we have
2303 		 * partial NAKlimit support for bulk RX.
2304 		 *
2305 		 * The upside of disabling it is simpler transfer scheduling.
2306 		 */
2307 		interval = 0;
2308 	}
2309 	qh->intv_reg = interval;
2310 
2311 	/* precompute addressing for external hub/tt ports */
2312 	if (musb->is_multipoint) {
2313 		struct usb_device	*parent = urb->dev->parent;
2314 
2315 		if (parent != hcd->self.root_hub) {
2316 			qh->h_addr_reg = (u8) parent->devnum;
2317 
2318 			/* set up tt info if needed */
2319 			if (urb->dev->tt) {
2320 				qh->h_port_reg = (u8) urb->dev->ttport;
2321 				if (urb->dev->tt->hub)
2322 					qh->h_addr_reg =
2323 						(u8) urb->dev->tt->hub->devnum;
2324 				if (urb->dev->tt->multi)
2325 					qh->h_addr_reg |= 0x80;
2326 			}
2327 		}
2328 	}
2329 
2330 	/* invariant: hep->hcpriv is null OR the qh that's already scheduled.
2331 	 * until we get real dma queues (with an entry for each urb/buffer),
2332 	 * we only have work to do in the former case.
2333 	 */
2334 	spin_lock_irqsave(&musb->lock, flags);
2335 	if (hep->hcpriv || !next_urb(qh)) {
2336 		/* some concurrent activity submitted another urb to hep...
2337 		 * odd, rare, error prone, but legal.
2338 		 */
2339 		kfree(qh);
2340 		qh = NULL;
2341 		ret = 0;
2342 	} else
2343 		ret = musb_schedule(musb, qh,
2344 				epd->bEndpointAddress & USB_ENDPOINT_DIR_MASK);
2345 
2346 	if (ret == 0) {
2347 		urb->hcpriv = qh;
2348 		/* FIXME set urb->start_frame for iso/intr, it's tested in
2349 		 * musb_start_urb(), but otherwise only konicawc cares ...
2350 		 */
2351 	}
2352 	spin_unlock_irqrestore(&musb->lock, flags);
2353 
2354 done:
2355 	if (ret != 0) {
2356 		spin_lock_irqsave(&musb->lock, flags);
2357 		usb_hcd_unlink_urb_from_ep(hcd, urb);
2358 		spin_unlock_irqrestore(&musb->lock, flags);
2359 		kfree(qh);
2360 	}
2361 	return ret;
2362 }
2363 
2364 
2365 /*
2366  * abort a transfer that's at the head of a hardware queue.
2367  * called with controller locked, irqs blocked
2368  * that hardware queue advances to the next transfer, unless prevented
2369  */
2370 static int musb_cleanup_urb(struct urb *urb, struct musb_qh *qh)
2371 {
2372 	struct musb_hw_ep	*ep = qh->hw_ep;
2373 	struct musb		*musb = ep->musb;
2374 	void __iomem		*epio = ep->regs;
2375 	unsigned		hw_end = ep->epnum;
2376 	void __iomem		*regs = ep->musb->mregs;
2377 	int			is_in = usb_pipein(urb->pipe);
2378 	int			status = 0;
2379 	u16			csr;
2380 
2381 	musb_ep_select(regs, hw_end);
2382 
2383 	if (is_dma_capable()) {
2384 		struct dma_channel	*dma;
2385 
2386 		dma = is_in ? ep->rx_channel : ep->tx_channel;
2387 		if (dma) {
2388 			status = ep->musb->dma_controller->channel_abort(dma);
2389 			dev_dbg(musb->controller,
2390 				"abort %cX%d DMA for urb %p --> %d\n",
2391 				is_in ? 'R' : 'T', ep->epnum,
2392 				urb, status);
2393 			urb->actual_length += dma->actual_len;
2394 		}
2395 	}
2396 
2397 	/* turn off DMA requests, discard state, stop polling ... */
2398 	if (ep->epnum && is_in) {
2399 		/* giveback saves bulk toggle */
2400 		csr = musb_h_flush_rxfifo(ep, 0);
2401 
2402 		/* REVISIT we still get an irq; should likely clear the
2403 		 * endpoint's irq status here to avoid bogus irqs.
2404 		 * clearing that status is platform-specific...
2405 		 */
2406 	} else if (ep->epnum) {
2407 		musb_h_tx_flush_fifo(ep);
2408 		csr = musb_readw(epio, MUSB_TXCSR);
2409 		csr &= ~(MUSB_TXCSR_AUTOSET
2410 			| MUSB_TXCSR_DMAENAB
2411 			| MUSB_TXCSR_H_RXSTALL
2412 			| MUSB_TXCSR_H_NAKTIMEOUT
2413 			| MUSB_TXCSR_H_ERROR
2414 			| MUSB_TXCSR_TXPKTRDY);
2415 		musb_writew(epio, MUSB_TXCSR, csr);
2416 		/* REVISIT may need to clear FLUSHFIFO ... */
2417 		musb_writew(epio, MUSB_TXCSR, csr);
2418 		/* flush cpu writebuffer */
2419 		csr = musb_readw(epio, MUSB_TXCSR);
2420 	} else  {
2421 		musb_h_ep0_flush_fifo(ep);
2422 	}
2423 	if (status == 0)
2424 		musb_advance_schedule(ep->musb, urb, ep, is_in);
2425 	return status;
2426 }
2427 
2428 static int musb_urb_dequeue(struct usb_hcd *hcd, struct urb *urb, int status)
2429 {
2430 	struct musb		*musb = hcd_to_musb(hcd);
2431 	struct musb_qh		*qh;
2432 	unsigned long		flags;
2433 	int			is_in  = usb_pipein(urb->pipe);
2434 	int			ret;
2435 
2436 	dev_dbg(musb->controller, "urb=%p, dev%d ep%d%s\n", urb,
2437 			usb_pipedevice(urb->pipe),
2438 			usb_pipeendpoint(urb->pipe),
2439 			is_in ? "in" : "out");
2440 
2441 	spin_lock_irqsave(&musb->lock, flags);
2442 	ret = usb_hcd_check_unlink_urb(hcd, urb, status);
2443 	if (ret)
2444 		goto done;
2445 
2446 	qh = urb->hcpriv;
2447 	if (!qh)
2448 		goto done;
2449 
2450 	/*
2451 	 * Any URB not actively programmed into endpoint hardware can be
2452 	 * immediately given back; that's any URB not at the head of an
2453 	 * endpoint queue, unless someday we get real DMA queues.  And even
2454 	 * if it's at the head, it might not be known to the hardware...
2455 	 *
2456 	 * Otherwise abort current transfer, pending DMA, etc.; urb->status
2457 	 * has already been updated.  This is a synchronous abort; it'd be
2458 	 * OK to hold off until after some IRQ, though.
2459 	 *
2460 	 * NOTE: qh is invalid unless !list_empty(&hep->urb_list)
2461 	 */
2462 	if (!qh->is_ready
2463 			|| urb->urb_list.prev != &qh->hep->urb_list
2464 			|| musb_ep_get_qh(qh->hw_ep, is_in) != qh) {
2465 		int	ready = qh->is_ready;
2466 
2467 		qh->is_ready = 0;
2468 		musb_giveback(musb, urb, 0);
2469 		qh->is_ready = ready;
2470 
2471 		/* If nothing else (usually musb_giveback) is using it
2472 		 * and its URB list has emptied, recycle this qh.
2473 		 */
2474 		if (ready && list_empty(&qh->hep->urb_list)) {
2475 			qh->hep->hcpriv = NULL;
2476 			list_del(&qh->ring);
2477 			kfree(qh);
2478 		}
2479 	} else
2480 		ret = musb_cleanup_urb(urb, qh);
2481 done:
2482 	spin_unlock_irqrestore(&musb->lock, flags);
2483 	return ret;
2484 }
2485 
2486 /* disable an endpoint */
2487 static void
2488 musb_h_disable(struct usb_hcd *hcd, struct usb_host_endpoint *hep)
2489 {
2490 	u8			is_in = hep->desc.bEndpointAddress & USB_DIR_IN;
2491 	unsigned long		flags;
2492 	struct musb		*musb = hcd_to_musb(hcd);
2493 	struct musb_qh		*qh;
2494 	struct urb		*urb;
2495 
2496 	spin_lock_irqsave(&musb->lock, flags);
2497 
2498 	qh = hep->hcpriv;
2499 	if (qh == NULL)
2500 		goto exit;
2501 
2502 	/* NOTE: qh is invalid unless !list_empty(&hep->urb_list) */
2503 
2504 	/* Kick the first URB off the hardware, if needed */
2505 	qh->is_ready = 0;
2506 	if (musb_ep_get_qh(qh->hw_ep, is_in) == qh) {
2507 		urb = next_urb(qh);
2508 
2509 		/* make software (then hardware) stop ASAP */
2510 		if (!urb->unlinked)
2511 			urb->status = -ESHUTDOWN;
2512 
2513 		/* cleanup */
2514 		musb_cleanup_urb(urb, qh);
2515 
2516 		/* Then nuke all the others ... and advance the
2517 		 * queue on hw_ep (e.g. bulk ring) when we're done.
2518 		 */
2519 		while (!list_empty(&hep->urb_list)) {
2520 			urb = next_urb(qh);
2521 			urb->status = -ESHUTDOWN;
2522 			musb_advance_schedule(musb, urb, qh->hw_ep, is_in);
2523 		}
2524 	} else {
2525 		/* Just empty the queue; the hardware is busy with
2526 		 * other transfers, and since !qh->is_ready nothing
2527 		 * will activate any of these as it advances.
2528 		 */
2529 		while (!list_empty(&hep->urb_list))
2530 			musb_giveback(musb, next_urb(qh), -ESHUTDOWN);
2531 
2532 		hep->hcpriv = NULL;
2533 		list_del(&qh->ring);
2534 		kfree(qh);
2535 	}
2536 exit:
2537 	spin_unlock_irqrestore(&musb->lock, flags);
2538 }
2539 
2540 static int musb_h_get_frame_number(struct usb_hcd *hcd)
2541 {
2542 	struct musb	*musb = hcd_to_musb(hcd);
2543 
2544 	return musb_readw(musb->mregs, MUSB_FRAME);
2545 }
2546 
2547 static int musb_h_start(struct usb_hcd *hcd)
2548 {
2549 	struct musb	*musb = hcd_to_musb(hcd);
2550 
2551 	/* NOTE: musb_start() is called when the hub driver turns
2552 	 * on port power, or when (OTG) peripheral starts.
2553 	 */
2554 	hcd->state = HC_STATE_RUNNING;
2555 	musb->port1_status = 0;
2556 	return 0;
2557 }
2558 
2559 static void musb_h_stop(struct usb_hcd *hcd)
2560 {
2561 	musb_stop(hcd_to_musb(hcd));
2562 	hcd->state = HC_STATE_HALT;
2563 }
2564 
2565 static int musb_bus_suspend(struct usb_hcd *hcd)
2566 {
2567 	struct musb	*musb = hcd_to_musb(hcd);
2568 	u8		devctl;
2569 
2570 	musb_port_suspend(musb, true);
2571 
2572 	if (!is_host_active(musb))
2573 		return 0;
2574 
2575 	switch (musb->xceiv->otg->state) {
2576 	case OTG_STATE_A_SUSPEND:
2577 		return 0;
2578 	case OTG_STATE_A_WAIT_VRISE:
2579 		/* ID could be grounded even if there's no device
2580 		 * on the other end of the cable.  NOTE that the
2581 		 * A_WAIT_VRISE timers are messy with MUSB...
2582 		 */
2583 		devctl = musb_readb(musb->mregs, MUSB_DEVCTL);
2584 		if ((devctl & MUSB_DEVCTL_VBUS) == MUSB_DEVCTL_VBUS)
2585 			musb->xceiv->otg->state = OTG_STATE_A_WAIT_BCON;
2586 		break;
2587 	default:
2588 		break;
2589 	}
2590 
2591 	if (musb->is_active) {
2592 		WARNING("trying to suspend as %s while active\n",
2593 				usb_otg_state_string(musb->xceiv->otg->state));
2594 		return -EBUSY;
2595 	} else
2596 		return 0;
2597 }
2598 
2599 static int musb_bus_resume(struct usb_hcd *hcd)
2600 {
2601 	struct musb *musb = hcd_to_musb(hcd);
2602 
2603 	if (musb->config &&
2604 	    musb->config->host_port_deassert_reset_at_resume)
2605 		musb_port_reset(musb, false);
2606 
2607 	return 0;
2608 }
2609 
2610 #ifndef CONFIG_MUSB_PIO_ONLY
2611 
2612 #define MUSB_USB_DMA_ALIGN 4
2613 
2614 struct musb_temp_buffer {
2615 	void *kmalloc_ptr;
2616 	void *old_xfer_buffer;
2617 	u8 data[0];
2618 };
2619 
2620 static void musb_free_temp_buffer(struct urb *urb)
2621 {
2622 	enum dma_data_direction dir;
2623 	struct musb_temp_buffer *temp;
2624 	size_t length;
2625 
2626 	if (!(urb->transfer_flags & URB_ALIGNED_TEMP_BUFFER))
2627 		return;
2628 
2629 	dir = usb_urb_dir_in(urb) ? DMA_FROM_DEVICE : DMA_TO_DEVICE;
2630 
2631 	temp = container_of(urb->transfer_buffer, struct musb_temp_buffer,
2632 			    data);
2633 
2634 	if (dir == DMA_FROM_DEVICE) {
2635 		if (usb_pipeisoc(urb->pipe))
2636 			length = urb->transfer_buffer_length;
2637 		else
2638 			length = urb->actual_length;
2639 
2640 		memcpy(temp->old_xfer_buffer, temp->data, length);
2641 	}
2642 	urb->transfer_buffer = temp->old_xfer_buffer;
2643 	kfree(temp->kmalloc_ptr);
2644 
2645 	urb->transfer_flags &= ~URB_ALIGNED_TEMP_BUFFER;
2646 }
2647 
2648 static int musb_alloc_temp_buffer(struct urb *urb, gfp_t mem_flags)
2649 {
2650 	enum dma_data_direction dir;
2651 	struct musb_temp_buffer *temp;
2652 	void *kmalloc_ptr;
2653 	size_t kmalloc_size;
2654 
2655 	if (urb->num_sgs || urb->sg ||
2656 	    urb->transfer_buffer_length == 0 ||
2657 	    !((uintptr_t)urb->transfer_buffer & (MUSB_USB_DMA_ALIGN - 1)))
2658 		return 0;
2659 
2660 	dir = usb_urb_dir_in(urb) ? DMA_FROM_DEVICE : DMA_TO_DEVICE;
2661 
2662 	/* Allocate a buffer with enough padding for alignment */
2663 	kmalloc_size = urb->transfer_buffer_length +
2664 		sizeof(struct musb_temp_buffer) + MUSB_USB_DMA_ALIGN - 1;
2665 
2666 	kmalloc_ptr = kmalloc(kmalloc_size, mem_flags);
2667 	if (!kmalloc_ptr)
2668 		return -ENOMEM;
2669 
2670 	/* Position our struct temp_buffer such that data is aligned */
2671 	temp = PTR_ALIGN(kmalloc_ptr, MUSB_USB_DMA_ALIGN);
2672 
2673 
2674 	temp->kmalloc_ptr = kmalloc_ptr;
2675 	temp->old_xfer_buffer = urb->transfer_buffer;
2676 	if (dir == DMA_TO_DEVICE)
2677 		memcpy(temp->data, urb->transfer_buffer,
2678 		       urb->transfer_buffer_length);
2679 	urb->transfer_buffer = temp->data;
2680 
2681 	urb->transfer_flags |= URB_ALIGNED_TEMP_BUFFER;
2682 
2683 	return 0;
2684 }
2685 
2686 static int musb_map_urb_for_dma(struct usb_hcd *hcd, struct urb *urb,
2687 				      gfp_t mem_flags)
2688 {
2689 	struct musb	*musb = hcd_to_musb(hcd);
2690 	int ret;
2691 
2692 	/*
2693 	 * The DMA engine in RTL1.8 and above cannot handle
2694 	 * DMA addresses that are not aligned to a 4 byte boundary.
2695 	 * For such engine implemented (un)map_urb_for_dma hooks.
2696 	 * Do not use these hooks for RTL<1.8
2697 	 */
2698 	if (musb->hwvers < MUSB_HWVERS_1800)
2699 		return usb_hcd_map_urb_for_dma(hcd, urb, mem_flags);
2700 
2701 	ret = musb_alloc_temp_buffer(urb, mem_flags);
2702 	if (ret)
2703 		return ret;
2704 
2705 	ret = usb_hcd_map_urb_for_dma(hcd, urb, mem_flags);
2706 	if (ret)
2707 		musb_free_temp_buffer(urb);
2708 
2709 	return ret;
2710 }
2711 
2712 static void musb_unmap_urb_for_dma(struct usb_hcd *hcd, struct urb *urb)
2713 {
2714 	struct musb	*musb = hcd_to_musb(hcd);
2715 
2716 	usb_hcd_unmap_urb_for_dma(hcd, urb);
2717 
2718 	/* Do not use this hook for RTL<1.8 (see description above) */
2719 	if (musb->hwvers < MUSB_HWVERS_1800)
2720 		return;
2721 
2722 	musb_free_temp_buffer(urb);
2723 }
2724 #endif /* !CONFIG_MUSB_PIO_ONLY */
2725 
2726 static const struct hc_driver musb_hc_driver = {
2727 	.description		= "musb-hcd",
2728 	.product_desc		= "MUSB HDRC host driver",
2729 	.hcd_priv_size		= sizeof(struct musb *),
2730 	.flags			= HCD_USB2 | HCD_MEMORY | HCD_BH,
2731 
2732 	/* not using irq handler or reset hooks from usbcore, since
2733 	 * those must be shared with peripheral code for OTG configs
2734 	 */
2735 
2736 	.start			= musb_h_start,
2737 	.stop			= musb_h_stop,
2738 
2739 	.get_frame_number	= musb_h_get_frame_number,
2740 
2741 	.urb_enqueue		= musb_urb_enqueue,
2742 	.urb_dequeue		= musb_urb_dequeue,
2743 	.endpoint_disable	= musb_h_disable,
2744 
2745 #ifndef CONFIG_MUSB_PIO_ONLY
2746 	.map_urb_for_dma	= musb_map_urb_for_dma,
2747 	.unmap_urb_for_dma	= musb_unmap_urb_for_dma,
2748 #endif
2749 
2750 	.hub_status_data	= musb_hub_status_data,
2751 	.hub_control		= musb_hub_control,
2752 	.bus_suspend		= musb_bus_suspend,
2753 	.bus_resume		= musb_bus_resume,
2754 	/* .start_port_reset	= NULL, */
2755 	/* .hub_irq_enable	= NULL, */
2756 };
2757 
2758 int musb_host_alloc(struct musb *musb)
2759 {
2760 	struct device	*dev = musb->controller;
2761 
2762 	/* usbcore sets dev->driver_data to hcd, and sometimes uses that... */
2763 	musb->hcd = usb_create_hcd(&musb_hc_driver, dev, dev_name(dev));
2764 	if (!musb->hcd)
2765 		return -EINVAL;
2766 
2767 	*musb->hcd->hcd_priv = (unsigned long) musb;
2768 	musb->hcd->self.uses_pio_for_control = 1;
2769 	musb->hcd->uses_new_polling = 1;
2770 	musb->hcd->has_tt = 1;
2771 
2772 	return 0;
2773 }
2774 
2775 void musb_host_cleanup(struct musb *musb)
2776 {
2777 	if (musb->port_mode == MUSB_PORT_MODE_GADGET)
2778 		return;
2779 	usb_remove_hcd(musb->hcd);
2780 }
2781 
2782 void musb_host_free(struct musb *musb)
2783 {
2784 	usb_put_hcd(musb->hcd);
2785 }
2786 
2787 int musb_host_setup(struct musb *musb, int power_budget)
2788 {
2789 	int ret;
2790 	struct usb_hcd *hcd = musb->hcd;
2791 
2792 	MUSB_HST_MODE(musb);
2793 	musb->xceiv->otg->default_a = 1;
2794 	musb->xceiv->otg->state = OTG_STATE_A_IDLE;
2795 
2796 	otg_set_host(musb->xceiv->otg, &hcd->self);
2797 	hcd->self.otg_port = 1;
2798 	musb->xceiv->otg->host = &hcd->self;
2799 	hcd->power_budget = 2 * (power_budget ? : 250);
2800 
2801 	ret = usb_add_hcd(hcd, 0, 0);
2802 	if (ret < 0)
2803 		return ret;
2804 
2805 	device_wakeup_enable(hcd->self.controller);
2806 	return 0;
2807 }
2808 
2809 void musb_host_resume_root_hub(struct musb *musb)
2810 {
2811 	usb_hcd_resume_root_hub(musb->hcd);
2812 }
2813 
2814 void musb_host_poke_root_hub(struct musb *musb)
2815 {
2816 	MUSB_HST_MODE(musb);
2817 	if (musb->hcd->status_urb)
2818 		usb_hcd_poll_rh_status(musb->hcd);
2819 	else
2820 		usb_hcd_resume_root_hub(musb->hcd);
2821 }
2822