xref: /openbmc/linux/drivers/usb/musb/musb_host.c (revision 0edbfea5)
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 	int		retries = 1000;
116 
117 	csr = musb_readw(epio, MUSB_TXCSR);
118 	while (csr & MUSB_TXCSR_FIFONOTEMPTY) {
119 		csr |= MUSB_TXCSR_FLUSHFIFO | MUSB_TXCSR_TXPKTRDY;
120 		musb_writew(epio, MUSB_TXCSR, csr);
121 		csr = musb_readw(epio, MUSB_TXCSR);
122 
123 		/*
124 		 * FIXME: sometimes the tx fifo flush failed, it has been
125 		 * observed during device disconnect on AM335x.
126 		 *
127 		 * To reproduce the issue, ensure tx urb(s) are queued when
128 		 * unplug the usb device which is connected to AM335x usb
129 		 * host port.
130 		 *
131 		 * I found using a usb-ethernet device and running iperf
132 		 * (client on AM335x) has very high chance to trigger it.
133 		 *
134 		 * Better to turn on dev_dbg() in musb_cleanup_urb() with
135 		 * CPPI enabled to see the issue when aborting the tx channel.
136 		 */
137 		if (dev_WARN_ONCE(musb->controller, retries-- < 1,
138 				"Could not flush host TX%d fifo: csr: %04x\n",
139 				ep->epnum, csr))
140 			return;
141 	}
142 }
143 
144 static void musb_h_ep0_flush_fifo(struct musb_hw_ep *ep)
145 {
146 	void __iomem	*epio = ep->regs;
147 	u16		csr;
148 	int		retries = 5;
149 
150 	/* scrub any data left in the fifo */
151 	do {
152 		csr = musb_readw(epio, MUSB_TXCSR);
153 		if (!(csr & (MUSB_CSR0_TXPKTRDY | MUSB_CSR0_RXPKTRDY)))
154 			break;
155 		musb_writew(epio, MUSB_TXCSR, MUSB_CSR0_FLUSHFIFO);
156 		csr = musb_readw(epio, MUSB_TXCSR);
157 		udelay(10);
158 	} while (--retries);
159 
160 	WARN(!retries, "Could not flush host TX%d fifo: csr: %04x\n",
161 			ep->epnum, csr);
162 
163 	/* and reset for the next transfer */
164 	musb_writew(epio, MUSB_TXCSR, 0);
165 }
166 
167 /*
168  * Start transmit. Caller is responsible for locking shared resources.
169  * musb must be locked.
170  */
171 static inline void musb_h_tx_start(struct musb_hw_ep *ep)
172 {
173 	u16	txcsr;
174 
175 	/* NOTE: no locks here; caller should lock and select EP */
176 	if (ep->epnum) {
177 		txcsr = musb_readw(ep->regs, MUSB_TXCSR);
178 		txcsr |= MUSB_TXCSR_TXPKTRDY | MUSB_TXCSR_H_WZC_BITS;
179 		musb_writew(ep->regs, MUSB_TXCSR, txcsr);
180 	} else {
181 		txcsr = MUSB_CSR0_H_SETUPPKT | MUSB_CSR0_TXPKTRDY;
182 		musb_writew(ep->regs, MUSB_CSR0, txcsr);
183 	}
184 
185 }
186 
187 static inline void musb_h_tx_dma_start(struct musb_hw_ep *ep)
188 {
189 	u16	txcsr;
190 
191 	/* NOTE: no locks here; caller should lock and select EP */
192 	txcsr = musb_readw(ep->regs, MUSB_TXCSR);
193 	txcsr |= MUSB_TXCSR_DMAENAB | MUSB_TXCSR_H_WZC_BITS;
194 	if (is_cppi_enabled(ep->musb))
195 		txcsr |= MUSB_TXCSR_DMAMODE;
196 	musb_writew(ep->regs, MUSB_TXCSR, txcsr);
197 }
198 
199 static void musb_ep_set_qh(struct musb_hw_ep *ep, int is_in, struct musb_qh *qh)
200 {
201 	if (is_in != 0 || ep->is_shared_fifo)
202 		ep->in_qh  = qh;
203 	if (is_in == 0 || ep->is_shared_fifo)
204 		ep->out_qh = qh;
205 }
206 
207 static struct musb_qh *musb_ep_get_qh(struct musb_hw_ep *ep, int is_in)
208 {
209 	return is_in ? ep->in_qh : ep->out_qh;
210 }
211 
212 /*
213  * Start the URB at the front of an endpoint's queue
214  * end must be claimed from the caller.
215  *
216  * Context: controller locked, irqs blocked
217  */
218 static void
219 musb_start_urb(struct musb *musb, int is_in, struct musb_qh *qh)
220 {
221 	u16			frame;
222 	u32			len;
223 	void __iomem		*mbase =  musb->mregs;
224 	struct urb		*urb = next_urb(qh);
225 	void			*buf = urb->transfer_buffer;
226 	u32			offset = 0;
227 	struct musb_hw_ep	*hw_ep = qh->hw_ep;
228 	unsigned		pipe = urb->pipe;
229 	u8			address = usb_pipedevice(pipe);
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 	dev_dbg(musb->controller, "qh %p urb %p dev%d ep%d%s%s, hw_ep %d, %p/%d\n",
258 			qh, urb, address, qh->epnum,
259 			is_in ? "in" : "out",
260 			({char *s; switch (qh->type) {
261 			case USB_ENDPOINT_XFER_CONTROL:	s = ""; break;
262 			case USB_ENDPOINT_XFER_BULK:	s = "-bulk"; break;
263 			case USB_ENDPOINT_XFER_ISOC:	s = "-iso"; break;
264 			default:			s = "-intr"; break;
265 			} s; }),
266 			epnum, buf + offset, len);
267 
268 	/* Configure endpoint */
269 	musb_ep_set_qh(hw_ep, is_in, qh);
270 	musb_ep_program(musb, epnum, urb, !is_in, buf, offset, len);
271 
272 	/* transmit may have more work: start it when it is time */
273 	if (is_in)
274 		return;
275 
276 	/* determine if the time is right for a periodic transfer */
277 	switch (qh->type) {
278 	case USB_ENDPOINT_XFER_ISOC:
279 	case USB_ENDPOINT_XFER_INT:
280 		dev_dbg(musb->controller, "check whether there's still time for periodic Tx\n");
281 		frame = musb_readw(mbase, MUSB_FRAME);
282 		/* FIXME this doesn't implement that scheduling policy ...
283 		 * or handle framecounter wrapping
284 		 */
285 		if (1) {	/* Always assume URB_ISO_ASAP */
286 			/* REVISIT the SOF irq handler shouldn't duplicate
287 			 * this code; and we don't init urb->start_frame...
288 			 */
289 			qh->frame = 0;
290 			goto start;
291 		} else {
292 			qh->frame = urb->start_frame;
293 			/* enable SOF interrupt so we can count down */
294 			dev_dbg(musb->controller, "SOF for %d\n", epnum);
295 #if 1 /* ifndef	CONFIG_ARCH_DAVINCI */
296 			musb_writeb(mbase, MUSB_INTRUSBE, 0xff);
297 #endif
298 		}
299 		break;
300 	default:
301 start:
302 		dev_dbg(musb->controller, "Start TX%d %s\n", epnum,
303 			hw_ep->tx_channel ? "dma" : "pio");
304 
305 		if (!hw_ep->tx_channel)
306 			musb_h_tx_start(hw_ep);
307 		else if (is_cppi_enabled(musb) || tusb_dma_omap(musb))
308 			musb_h_tx_dma_start(hw_ep);
309 	}
310 }
311 
312 /* Context: caller owns controller lock, IRQs are blocked */
313 static void musb_giveback(struct musb *musb, struct urb *urb, int status)
314 __releases(musb->lock)
315 __acquires(musb->lock)
316 {
317 	dev_dbg(musb->controller,
318 			"complete %p %pF (%d), dev%d ep%d%s, %d/%d\n",
319 			urb, urb->complete, status,
320 			usb_pipedevice(urb->pipe),
321 			usb_pipeendpoint(urb->pipe),
322 			usb_pipein(urb->pipe) ? "in" : "out",
323 			urb->actual_length, urb->transfer_buffer_length
324 			);
325 
326 	usb_hcd_unlink_urb_from_ep(musb->hcd, urb);
327 	spin_unlock(&musb->lock);
328 	usb_hcd_giveback_urb(musb->hcd, urb, status);
329 	spin_lock(&musb->lock);
330 }
331 
332 /* For bulk/interrupt endpoints only */
333 static inline void musb_save_toggle(struct musb_qh *qh, int is_in,
334 				    struct urb *urb)
335 {
336 	void __iomem		*epio = qh->hw_ep->regs;
337 	u16			csr;
338 
339 	/*
340 	 * FIXME: the current Mentor DMA code seems to have
341 	 * problems getting toggle correct.
342 	 */
343 
344 	if (is_in)
345 		csr = musb_readw(epio, MUSB_RXCSR) & MUSB_RXCSR_H_DATATOGGLE;
346 	else
347 		csr = musb_readw(epio, MUSB_TXCSR) & MUSB_TXCSR_H_DATATOGGLE;
348 
349 	usb_settoggle(urb->dev, qh->epnum, !is_in, csr ? 1 : 0);
350 }
351 
352 /*
353  * Advance this hardware endpoint's queue, completing the specified URB and
354  * advancing to either the next URB queued to that qh, or else invalidating
355  * that qh and advancing to the next qh scheduled after the current one.
356  *
357  * Context: caller owns controller lock, IRQs are blocked
358  */
359 static void musb_advance_schedule(struct musb *musb, struct urb *urb,
360 				  struct musb_hw_ep *hw_ep, int is_in)
361 {
362 	struct musb_qh		*qh = musb_ep_get_qh(hw_ep, is_in);
363 	struct musb_hw_ep	*ep = qh->hw_ep;
364 	int			ready = qh->is_ready;
365 	int			status;
366 
367 	status = (urb->status == -EINPROGRESS) ? 0 : urb->status;
368 
369 	/* save toggle eagerly, for paranoia */
370 	switch (qh->type) {
371 	case USB_ENDPOINT_XFER_BULK:
372 	case USB_ENDPOINT_XFER_INT:
373 		musb_save_toggle(qh, is_in, urb);
374 		break;
375 	case USB_ENDPOINT_XFER_ISOC:
376 		if (status == 0 && urb->error_count)
377 			status = -EXDEV;
378 		break;
379 	}
380 
381 	qh->is_ready = 0;
382 	musb_giveback(musb, urb, status);
383 	qh->is_ready = ready;
384 
385 	/* reclaim resources (and bandwidth) ASAP; deschedule it, and
386 	 * invalidate qh as soon as list_empty(&hep->urb_list)
387 	 */
388 	if (list_empty(&qh->hep->urb_list)) {
389 		struct list_head	*head;
390 		struct dma_controller	*dma = musb->dma_controller;
391 
392 		if (is_in) {
393 			ep->rx_reinit = 1;
394 			if (ep->rx_channel) {
395 				dma->channel_release(ep->rx_channel);
396 				ep->rx_channel = NULL;
397 			}
398 		} else {
399 			ep->tx_reinit = 1;
400 			if (ep->tx_channel) {
401 				dma->channel_release(ep->tx_channel);
402 				ep->tx_channel = NULL;
403 			}
404 		}
405 
406 		/* Clobber old pointers to this qh */
407 		musb_ep_set_qh(ep, is_in, NULL);
408 		qh->hep->hcpriv = NULL;
409 
410 		switch (qh->type) {
411 
412 		case USB_ENDPOINT_XFER_CONTROL:
413 		case USB_ENDPOINT_XFER_BULK:
414 			/* fifo policy for these lists, except that NAKing
415 			 * should rotate a qh to the end (for fairness).
416 			 */
417 			if (qh->mux == 1) {
418 				head = qh->ring.prev;
419 				list_del(&qh->ring);
420 				kfree(qh);
421 				qh = first_qh(head);
422 				break;
423 			}
424 
425 		case USB_ENDPOINT_XFER_ISOC:
426 		case USB_ENDPOINT_XFER_INT:
427 			/* this is where periodic bandwidth should be
428 			 * de-allocated if it's tracked and allocated;
429 			 * and where we'd update the schedule tree...
430 			 */
431 			kfree(qh);
432 			qh = NULL;
433 			break;
434 		}
435 	}
436 
437 	if (qh != NULL && qh->is_ready) {
438 		dev_dbg(musb->controller, "... next ep%d %cX urb %p\n",
439 		    hw_ep->epnum, is_in ? 'R' : 'T', next_urb(qh));
440 		musb_start_urb(musb, is_in, qh);
441 	}
442 }
443 
444 static u16 musb_h_flush_rxfifo(struct musb_hw_ep *hw_ep, u16 csr)
445 {
446 	/* we don't want fifo to fill itself again;
447 	 * ignore dma (various models),
448 	 * leave toggle alone (may not have been saved yet)
449 	 */
450 	csr |= MUSB_RXCSR_FLUSHFIFO | MUSB_RXCSR_RXPKTRDY;
451 	csr &= ~(MUSB_RXCSR_H_REQPKT
452 		| MUSB_RXCSR_H_AUTOREQ
453 		| MUSB_RXCSR_AUTOCLEAR);
454 
455 	/* write 2x to allow double buffering */
456 	musb_writew(hw_ep->regs, MUSB_RXCSR, csr);
457 	musb_writew(hw_ep->regs, MUSB_RXCSR, csr);
458 
459 	/* flush writebuffer */
460 	return musb_readw(hw_ep->regs, MUSB_RXCSR);
461 }
462 
463 /*
464  * PIO RX for a packet (or part of it).
465  */
466 static bool
467 musb_host_packet_rx(struct musb *musb, struct urb *urb, u8 epnum, u8 iso_err)
468 {
469 	u16			rx_count;
470 	u8			*buf;
471 	u16			csr;
472 	bool			done = false;
473 	u32			length;
474 	int			do_flush = 0;
475 	struct musb_hw_ep	*hw_ep = musb->endpoints + epnum;
476 	void __iomem		*epio = hw_ep->regs;
477 	struct musb_qh		*qh = hw_ep->in_qh;
478 	int			pipe = urb->pipe;
479 	void			*buffer = urb->transfer_buffer;
480 
481 	/* musb_ep_select(mbase, epnum); */
482 	rx_count = musb_readw(epio, MUSB_RXCOUNT);
483 	dev_dbg(musb->controller, "RX%d count %d, buffer %p len %d/%d\n", epnum, rx_count,
484 			urb->transfer_buffer, qh->offset,
485 			urb->transfer_buffer_length);
486 
487 	/* unload FIFO */
488 	if (usb_pipeisoc(pipe)) {
489 		int					status = 0;
490 		struct usb_iso_packet_descriptor	*d;
491 
492 		if (iso_err) {
493 			status = -EILSEQ;
494 			urb->error_count++;
495 		}
496 
497 		d = urb->iso_frame_desc + qh->iso_idx;
498 		buf = buffer + d->offset;
499 		length = d->length;
500 		if (rx_count > length) {
501 			if (status == 0) {
502 				status = -EOVERFLOW;
503 				urb->error_count++;
504 			}
505 			dev_dbg(musb->controller, "** OVERFLOW %d into %d\n", rx_count, length);
506 			do_flush = 1;
507 		} else
508 			length = rx_count;
509 		urb->actual_length += length;
510 		d->actual_length = length;
511 
512 		d->status = status;
513 
514 		/* see if we are done */
515 		done = (++qh->iso_idx >= urb->number_of_packets);
516 	} else {
517 		/* non-isoch */
518 		buf = buffer + qh->offset;
519 		length = urb->transfer_buffer_length - qh->offset;
520 		if (rx_count > length) {
521 			if (urb->status == -EINPROGRESS)
522 				urb->status = -EOVERFLOW;
523 			dev_dbg(musb->controller, "** OVERFLOW %d into %d\n", rx_count, length);
524 			do_flush = 1;
525 		} else
526 			length = rx_count;
527 		urb->actual_length += length;
528 		qh->offset += length;
529 
530 		/* see if we are done */
531 		done = (urb->actual_length == urb->transfer_buffer_length)
532 			|| (rx_count < qh->maxpacket)
533 			|| (urb->status != -EINPROGRESS);
534 		if (done
535 				&& (urb->status == -EINPROGRESS)
536 				&& (urb->transfer_flags & URB_SHORT_NOT_OK)
537 				&& (urb->actual_length
538 					< urb->transfer_buffer_length))
539 			urb->status = -EREMOTEIO;
540 	}
541 
542 	musb_read_fifo(hw_ep, length, buf);
543 
544 	csr = musb_readw(epio, MUSB_RXCSR);
545 	csr |= MUSB_RXCSR_H_WZC_BITS;
546 	if (unlikely(do_flush))
547 		musb_h_flush_rxfifo(hw_ep, csr);
548 	else {
549 		/* REVISIT this assumes AUTOCLEAR is never set */
550 		csr &= ~(MUSB_RXCSR_RXPKTRDY | MUSB_RXCSR_H_REQPKT);
551 		if (!done)
552 			csr |= MUSB_RXCSR_H_REQPKT;
553 		musb_writew(epio, MUSB_RXCSR, csr);
554 	}
555 
556 	return done;
557 }
558 
559 /* we don't always need to reinit a given side of an endpoint...
560  * when we do, use tx/rx reinit routine and then construct a new CSR
561  * to address data toggle, NYET, and DMA or PIO.
562  *
563  * it's possible that driver bugs (especially for DMA) or aborting a
564  * transfer might have left the endpoint busier than it should be.
565  * the busy/not-empty tests are basically paranoia.
566  */
567 static void
568 musb_rx_reinit(struct musb *musb, struct musb_qh *qh, u8 epnum)
569 {
570 	struct musb_hw_ep *ep = musb->endpoints + epnum;
571 	u16	csr;
572 
573 	/* NOTE:  we know the "rx" fifo reinit never triggers for ep0.
574 	 * That always uses tx_reinit since ep0 repurposes TX register
575 	 * offsets; the initial SETUP packet is also a kind of OUT.
576 	 */
577 
578 	/* if programmed for Tx, put it in RX mode */
579 	if (ep->is_shared_fifo) {
580 		csr = musb_readw(ep->regs, MUSB_TXCSR);
581 		if (csr & MUSB_TXCSR_MODE) {
582 			musb_h_tx_flush_fifo(ep);
583 			csr = musb_readw(ep->regs, MUSB_TXCSR);
584 			musb_writew(ep->regs, MUSB_TXCSR,
585 				    csr | MUSB_TXCSR_FRCDATATOG);
586 		}
587 
588 		/*
589 		 * Clear the MODE bit (and everything else) to enable Rx.
590 		 * NOTE: we mustn't clear the DMAMODE bit before DMAENAB.
591 		 */
592 		if (csr & MUSB_TXCSR_DMAMODE)
593 			musb_writew(ep->regs, MUSB_TXCSR, MUSB_TXCSR_DMAMODE);
594 		musb_writew(ep->regs, MUSB_TXCSR, 0);
595 
596 	/* scrub all previous state, clearing toggle */
597 	} else {
598 		csr = musb_readw(ep->regs, MUSB_RXCSR);
599 		if (csr & MUSB_RXCSR_RXPKTRDY)
600 			WARNING("rx%d, packet/%d ready?\n", ep->epnum,
601 				musb_readw(ep->regs, MUSB_RXCOUNT));
602 
603 		musb_h_flush_rxfifo(ep, MUSB_RXCSR_CLRDATATOG);
604 	}
605 
606 	/* target addr and (for multipoint) hub addr/port */
607 	if (musb->is_multipoint) {
608 		musb_write_rxfunaddr(musb, epnum, qh->addr_reg);
609 		musb_write_rxhubaddr(musb, epnum, qh->h_addr_reg);
610 		musb_write_rxhubport(musb, epnum, qh->h_port_reg);
611 	} else
612 		musb_writeb(musb->mregs, MUSB_FADDR, qh->addr_reg);
613 
614 	/* protocol/endpoint, interval/NAKlimit, i/o size */
615 	musb_writeb(ep->regs, MUSB_RXTYPE, qh->type_reg);
616 	musb_writeb(ep->regs, MUSB_RXINTERVAL, qh->intv_reg);
617 	/* NOTE: bulk combining rewrites high bits of maxpacket */
618 	/* Set RXMAXP with the FIFO size of the endpoint
619 	 * to disable double buffer mode.
620 	 */
621 	if (musb->double_buffer_not_ok)
622 		musb_writew(ep->regs, MUSB_RXMAXP, ep->max_packet_sz_rx);
623 	else
624 		musb_writew(ep->regs, MUSB_RXMAXP,
625 				qh->maxpacket | ((qh->hb_mult - 1) << 11));
626 
627 	ep->rx_reinit = 0;
628 }
629 
630 static int musb_tx_dma_set_mode_mentor(struct dma_controller *dma,
631 		struct musb_hw_ep *hw_ep, struct musb_qh *qh,
632 		struct urb *urb, u32 offset,
633 		u32 *length, u8 *mode)
634 {
635 	struct dma_channel	*channel = hw_ep->tx_channel;
636 	void __iomem		*epio = hw_ep->regs;
637 	u16			pkt_size = qh->maxpacket;
638 	u16			csr;
639 
640 	if (*length > channel->max_len)
641 		*length = channel->max_len;
642 
643 	csr = musb_readw(epio, MUSB_TXCSR);
644 	if (*length > pkt_size) {
645 		*mode = 1;
646 		csr |= MUSB_TXCSR_DMAMODE | MUSB_TXCSR_DMAENAB;
647 		/* autoset shouldn't be set in high bandwidth */
648 		/*
649 		 * Enable Autoset according to table
650 		 * below
651 		 * bulk_split hb_mult	Autoset_Enable
652 		 *	0	1	Yes(Normal)
653 		 *	0	>1	No(High BW ISO)
654 		 *	1	1	Yes(HS bulk)
655 		 *	1	>1	Yes(FS bulk)
656 		 */
657 		if (qh->hb_mult == 1 || (qh->hb_mult > 1 &&
658 					can_bulk_split(hw_ep->musb, qh->type)))
659 			csr |= MUSB_TXCSR_AUTOSET;
660 	} else {
661 		*mode = 0;
662 		csr &= ~(MUSB_TXCSR_AUTOSET | MUSB_TXCSR_DMAMODE);
663 		csr |= MUSB_TXCSR_DMAENAB; /* against programmer's guide */
664 	}
665 	channel->desired_mode = *mode;
666 	musb_writew(epio, MUSB_TXCSR, csr);
667 
668 	return 0;
669 }
670 
671 static int musb_tx_dma_set_mode_cppi_tusb(struct dma_controller *dma,
672 					  struct musb_hw_ep *hw_ep,
673 					  struct musb_qh *qh,
674 					  struct urb *urb,
675 					  u32 offset,
676 					  u32 *length,
677 					  u8 *mode)
678 {
679 	struct dma_channel *channel = hw_ep->tx_channel;
680 
681 	if (!is_cppi_enabled(hw_ep->musb) && !tusb_dma_omap(hw_ep->musb))
682 		return -ENODEV;
683 
684 	channel->actual_len = 0;
685 
686 	/*
687 	 * TX uses "RNDIS" mode automatically but needs help
688 	 * to identify the zero-length-final-packet case.
689 	 */
690 	*mode = (urb->transfer_flags & URB_ZERO_PACKET) ? 1 : 0;
691 
692 	return 0;
693 }
694 
695 static bool musb_tx_dma_program(struct dma_controller *dma,
696 		struct musb_hw_ep *hw_ep, struct musb_qh *qh,
697 		struct urb *urb, u32 offset, u32 length)
698 {
699 	struct dma_channel	*channel = hw_ep->tx_channel;
700 	u16			pkt_size = qh->maxpacket;
701 	u8			mode;
702 	int			res;
703 
704 	if (musb_dma_inventra(hw_ep->musb) || musb_dma_ux500(hw_ep->musb))
705 		res = musb_tx_dma_set_mode_mentor(dma, hw_ep, qh, urb,
706 						 offset, &length, &mode);
707 	else
708 		res = musb_tx_dma_set_mode_cppi_tusb(dma, hw_ep, qh, urb,
709 						     offset, &length, &mode);
710 	if (res)
711 		return false;
712 
713 	qh->segsize = length;
714 
715 	/*
716 	 * Ensure the data reaches to main memory before starting
717 	 * DMA transfer
718 	 */
719 	wmb();
720 
721 	if (!dma->channel_program(channel, pkt_size, mode,
722 			urb->transfer_dma + offset, length)) {
723 		void __iomem *epio = hw_ep->regs;
724 		u16 csr;
725 
726 		dma->channel_release(channel);
727 		hw_ep->tx_channel = NULL;
728 
729 		csr = musb_readw(epio, MUSB_TXCSR);
730 		csr &= ~(MUSB_TXCSR_AUTOSET | MUSB_TXCSR_DMAENAB);
731 		musb_writew(epio, MUSB_TXCSR, csr | MUSB_TXCSR_H_WZC_BITS);
732 		return false;
733 	}
734 	return true;
735 }
736 
737 /*
738  * Program an HDRC endpoint as per the given URB
739  * Context: irqs blocked, controller lock held
740  */
741 static void musb_ep_program(struct musb *musb, u8 epnum,
742 			struct urb *urb, int is_out,
743 			u8 *buf, u32 offset, u32 len)
744 {
745 	struct dma_controller	*dma_controller;
746 	struct dma_channel	*dma_channel;
747 	u8			dma_ok;
748 	void __iomem		*mbase = musb->mregs;
749 	struct musb_hw_ep	*hw_ep = musb->endpoints + epnum;
750 	void __iomem		*epio = hw_ep->regs;
751 	struct musb_qh		*qh = musb_ep_get_qh(hw_ep, !is_out);
752 	u16			packet_sz = qh->maxpacket;
753 	u8			use_dma = 1;
754 	u16			csr;
755 
756 	dev_dbg(musb->controller, "%s hw%d urb %p spd%d dev%d ep%d%s "
757 				"h_addr%02x h_port%02x bytes %d\n",
758 			is_out ? "-->" : "<--",
759 			epnum, urb, urb->dev->speed,
760 			qh->addr_reg, qh->epnum, is_out ? "out" : "in",
761 			qh->h_addr_reg, qh->h_port_reg,
762 			len);
763 
764 	musb_ep_select(mbase, epnum);
765 
766 	if (is_out && !len) {
767 		use_dma = 0;
768 		csr = musb_readw(epio, MUSB_TXCSR);
769 		csr &= ~MUSB_TXCSR_DMAENAB;
770 		musb_writew(epio, MUSB_TXCSR, csr);
771 		hw_ep->tx_channel = NULL;
772 	}
773 
774 	/* candidate for DMA? */
775 	dma_controller = musb->dma_controller;
776 	if (use_dma && is_dma_capable() && epnum && dma_controller) {
777 		dma_channel = is_out ? hw_ep->tx_channel : hw_ep->rx_channel;
778 		if (!dma_channel) {
779 			dma_channel = dma_controller->channel_alloc(
780 					dma_controller, hw_ep, is_out);
781 			if (is_out)
782 				hw_ep->tx_channel = dma_channel;
783 			else
784 				hw_ep->rx_channel = dma_channel;
785 		}
786 	} else
787 		dma_channel = NULL;
788 
789 	/* make sure we clear DMAEnab, autoSet bits from previous run */
790 
791 	/* OUT/transmit/EP0 or IN/receive? */
792 	if (is_out) {
793 		u16	csr;
794 		u16	int_txe;
795 		u16	load_count;
796 
797 		csr = musb_readw(epio, MUSB_TXCSR);
798 
799 		/* disable interrupt in case we flush */
800 		int_txe = musb->intrtxe;
801 		musb_writew(mbase, MUSB_INTRTXE, int_txe & ~(1 << epnum));
802 
803 		/* general endpoint setup */
804 		if (epnum) {
805 			/* flush all old state, set default */
806 			/*
807 			 * We could be flushing valid
808 			 * packets in double buffering
809 			 * case
810 			 */
811 			if (!hw_ep->tx_double_buffered)
812 				musb_h_tx_flush_fifo(hw_ep);
813 
814 			/*
815 			 * We must not clear the DMAMODE bit before or in
816 			 * the same cycle with the DMAENAB bit, so we clear
817 			 * the latter first...
818 			 */
819 			csr &= ~(MUSB_TXCSR_H_NAKTIMEOUT
820 					| MUSB_TXCSR_AUTOSET
821 					| MUSB_TXCSR_DMAENAB
822 					| MUSB_TXCSR_FRCDATATOG
823 					| MUSB_TXCSR_H_RXSTALL
824 					| MUSB_TXCSR_H_ERROR
825 					| MUSB_TXCSR_TXPKTRDY
826 					);
827 			csr |= MUSB_TXCSR_MODE;
828 
829 			if (!hw_ep->tx_double_buffered) {
830 				if (usb_gettoggle(urb->dev, qh->epnum, 1))
831 					csr |= MUSB_TXCSR_H_WR_DATATOGGLE
832 						| MUSB_TXCSR_H_DATATOGGLE;
833 				else
834 					csr |= MUSB_TXCSR_CLRDATATOG;
835 			}
836 
837 			musb_writew(epio, MUSB_TXCSR, csr);
838 			/* REVISIT may need to clear FLUSHFIFO ... */
839 			csr &= ~MUSB_TXCSR_DMAMODE;
840 			musb_writew(epio, MUSB_TXCSR, csr);
841 			csr = musb_readw(epio, MUSB_TXCSR);
842 		} else {
843 			/* endpoint 0: just flush */
844 			musb_h_ep0_flush_fifo(hw_ep);
845 		}
846 
847 		/* target addr and (for multipoint) hub addr/port */
848 		if (musb->is_multipoint) {
849 			musb_write_txfunaddr(musb, epnum, qh->addr_reg);
850 			musb_write_txhubaddr(musb, epnum, qh->h_addr_reg);
851 			musb_write_txhubport(musb, epnum, qh->h_port_reg);
852 /* FIXME if !epnum, do the same for RX ... */
853 		} else
854 			musb_writeb(mbase, MUSB_FADDR, qh->addr_reg);
855 
856 		/* protocol/endpoint/interval/NAKlimit */
857 		if (epnum) {
858 			musb_writeb(epio, MUSB_TXTYPE, qh->type_reg);
859 			if (musb->double_buffer_not_ok) {
860 				musb_writew(epio, MUSB_TXMAXP,
861 						hw_ep->max_packet_sz_tx);
862 			} else if (can_bulk_split(musb, qh->type)) {
863 				qh->hb_mult = hw_ep->max_packet_sz_tx
864 						/ packet_sz;
865 				musb_writew(epio, MUSB_TXMAXP, packet_sz
866 					| ((qh->hb_mult) - 1) << 11);
867 			} else {
868 				musb_writew(epio, MUSB_TXMAXP,
869 						qh->maxpacket |
870 						((qh->hb_mult - 1) << 11));
871 			}
872 			musb_writeb(epio, MUSB_TXINTERVAL, qh->intv_reg);
873 		} else {
874 			musb_writeb(epio, MUSB_NAKLIMIT0, qh->intv_reg);
875 			if (musb->is_multipoint)
876 				musb_writeb(epio, MUSB_TYPE0,
877 						qh->type_reg);
878 		}
879 
880 		if (can_bulk_split(musb, qh->type))
881 			load_count = min((u32) hw_ep->max_packet_sz_tx,
882 						len);
883 		else
884 			load_count = min((u32) packet_sz, len);
885 
886 		if (dma_channel && musb_tx_dma_program(dma_controller,
887 					hw_ep, qh, urb, offset, len))
888 			load_count = 0;
889 
890 		if (load_count) {
891 			/* PIO to load FIFO */
892 			qh->segsize = load_count;
893 			if (!buf) {
894 				sg_miter_start(&qh->sg_miter, urb->sg, 1,
895 						SG_MITER_ATOMIC
896 						| SG_MITER_FROM_SG);
897 				if (!sg_miter_next(&qh->sg_miter)) {
898 					dev_err(musb->controller,
899 							"error: sg"
900 							"list empty\n");
901 					sg_miter_stop(&qh->sg_miter);
902 					goto finish;
903 				}
904 				buf = qh->sg_miter.addr + urb->sg->offset +
905 					urb->actual_length;
906 				load_count = min_t(u32, load_count,
907 						qh->sg_miter.length);
908 				musb_write_fifo(hw_ep, load_count, buf);
909 				qh->sg_miter.consumed = load_count;
910 				sg_miter_stop(&qh->sg_miter);
911 			} else
912 				musb_write_fifo(hw_ep, load_count, buf);
913 		}
914 finish:
915 		/* re-enable interrupt */
916 		musb_writew(mbase, MUSB_INTRTXE, int_txe);
917 
918 	/* IN/receive */
919 	} else {
920 		u16	csr;
921 
922 		if (hw_ep->rx_reinit) {
923 			musb_rx_reinit(musb, qh, epnum);
924 
925 			/* init new state: toggle and NYET, maybe DMA later */
926 			if (usb_gettoggle(urb->dev, qh->epnum, 0))
927 				csr = MUSB_RXCSR_H_WR_DATATOGGLE
928 					| MUSB_RXCSR_H_DATATOGGLE;
929 			else
930 				csr = 0;
931 			if (qh->type == USB_ENDPOINT_XFER_INT)
932 				csr |= MUSB_RXCSR_DISNYET;
933 
934 		} else {
935 			csr = musb_readw(hw_ep->regs, MUSB_RXCSR);
936 
937 			if (csr & (MUSB_RXCSR_RXPKTRDY
938 					| MUSB_RXCSR_DMAENAB
939 					| MUSB_RXCSR_H_REQPKT))
940 				ERR("broken !rx_reinit, ep%d csr %04x\n",
941 						hw_ep->epnum, csr);
942 
943 			/* scrub any stale state, leaving toggle alone */
944 			csr &= MUSB_RXCSR_DISNYET;
945 		}
946 
947 		/* kick things off */
948 
949 		if ((is_cppi_enabled(musb) || tusb_dma_omap(musb)) && dma_channel) {
950 			/* Candidate for DMA */
951 			dma_channel->actual_len = 0L;
952 			qh->segsize = len;
953 
954 			/* AUTOREQ is in a DMA register */
955 			musb_writew(hw_ep->regs, MUSB_RXCSR, csr);
956 			csr = musb_readw(hw_ep->regs, MUSB_RXCSR);
957 
958 			/*
959 			 * Unless caller treats short RX transfers as
960 			 * errors, we dare not queue multiple transfers.
961 			 */
962 			dma_ok = dma_controller->channel_program(dma_channel,
963 					packet_sz, !(urb->transfer_flags &
964 						     URB_SHORT_NOT_OK),
965 					urb->transfer_dma + offset,
966 					qh->segsize);
967 			if (!dma_ok) {
968 				dma_controller->channel_release(dma_channel);
969 				hw_ep->rx_channel = dma_channel = NULL;
970 			} else
971 				csr |= MUSB_RXCSR_DMAENAB;
972 		}
973 
974 		csr |= MUSB_RXCSR_H_REQPKT;
975 		dev_dbg(musb->controller, "RXCSR%d := %04x\n", epnum, csr);
976 		musb_writew(hw_ep->regs, MUSB_RXCSR, csr);
977 		csr = musb_readw(hw_ep->regs, MUSB_RXCSR);
978 	}
979 }
980 
981 /* Schedule next QH from musb->in_bulk/out_bulk and move the current qh to
982  * the end; avoids starvation for other endpoints.
983  */
984 static void musb_bulk_nak_timeout(struct musb *musb, struct musb_hw_ep *ep,
985 	int is_in)
986 {
987 	struct dma_channel	*dma;
988 	struct urb		*urb;
989 	void __iomem		*mbase = musb->mregs;
990 	void __iomem		*epio = ep->regs;
991 	struct musb_qh		*cur_qh, *next_qh;
992 	u16			rx_csr, tx_csr;
993 
994 	musb_ep_select(mbase, ep->epnum);
995 	if (is_in) {
996 		dma = is_dma_capable() ? ep->rx_channel : NULL;
997 
998 		/* clear nak timeout bit */
999 		rx_csr = musb_readw(epio, MUSB_RXCSR);
1000 		rx_csr |= MUSB_RXCSR_H_WZC_BITS;
1001 		rx_csr &= ~MUSB_RXCSR_DATAERROR;
1002 		musb_writew(epio, MUSB_RXCSR, rx_csr);
1003 
1004 		cur_qh = first_qh(&musb->in_bulk);
1005 	} else {
1006 		dma = is_dma_capable() ? ep->tx_channel : NULL;
1007 
1008 		/* clear nak timeout bit */
1009 		tx_csr = musb_readw(epio, MUSB_TXCSR);
1010 		tx_csr |= MUSB_TXCSR_H_WZC_BITS;
1011 		tx_csr &= ~MUSB_TXCSR_H_NAKTIMEOUT;
1012 		musb_writew(epio, MUSB_TXCSR, tx_csr);
1013 
1014 		cur_qh = first_qh(&musb->out_bulk);
1015 	}
1016 	if (cur_qh) {
1017 		urb = next_urb(cur_qh);
1018 		if (dma_channel_status(dma) == MUSB_DMA_STATUS_BUSY) {
1019 			dma->status = MUSB_DMA_STATUS_CORE_ABORT;
1020 			musb->dma_controller->channel_abort(dma);
1021 			urb->actual_length += dma->actual_len;
1022 			dma->actual_len = 0L;
1023 		}
1024 		musb_save_toggle(cur_qh, is_in, urb);
1025 
1026 		if (is_in) {
1027 			/* move cur_qh to end of queue */
1028 			list_move_tail(&cur_qh->ring, &musb->in_bulk);
1029 
1030 			/* get the next qh from musb->in_bulk */
1031 			next_qh = first_qh(&musb->in_bulk);
1032 
1033 			/* set rx_reinit and schedule the next qh */
1034 			ep->rx_reinit = 1;
1035 		} else {
1036 			/* move cur_qh to end of queue */
1037 			list_move_tail(&cur_qh->ring, &musb->out_bulk);
1038 
1039 			/* get the next qh from musb->out_bulk */
1040 			next_qh = first_qh(&musb->out_bulk);
1041 
1042 			/* set tx_reinit and schedule the next qh */
1043 			ep->tx_reinit = 1;
1044 		}
1045 		musb_start_urb(musb, is_in, next_qh);
1046 	}
1047 }
1048 
1049 /*
1050  * Service the default endpoint (ep0) as host.
1051  * Return true until it's time to start the status stage.
1052  */
1053 static bool musb_h_ep0_continue(struct musb *musb, u16 len, struct urb *urb)
1054 {
1055 	bool			 more = false;
1056 	u8			*fifo_dest = NULL;
1057 	u16			fifo_count = 0;
1058 	struct musb_hw_ep	*hw_ep = musb->control_ep;
1059 	struct musb_qh		*qh = hw_ep->in_qh;
1060 	struct usb_ctrlrequest	*request;
1061 
1062 	switch (musb->ep0_stage) {
1063 	case MUSB_EP0_IN:
1064 		fifo_dest = urb->transfer_buffer + urb->actual_length;
1065 		fifo_count = min_t(size_t, len, urb->transfer_buffer_length -
1066 				   urb->actual_length);
1067 		if (fifo_count < len)
1068 			urb->status = -EOVERFLOW;
1069 
1070 		musb_read_fifo(hw_ep, fifo_count, fifo_dest);
1071 
1072 		urb->actual_length += fifo_count;
1073 		if (len < qh->maxpacket) {
1074 			/* always terminate on short read; it's
1075 			 * rarely reported as an error.
1076 			 */
1077 		} else if (urb->actual_length <
1078 				urb->transfer_buffer_length)
1079 			more = true;
1080 		break;
1081 	case MUSB_EP0_START:
1082 		request = (struct usb_ctrlrequest *) urb->setup_packet;
1083 
1084 		if (!request->wLength) {
1085 			dev_dbg(musb->controller, "start no-DATA\n");
1086 			break;
1087 		} else if (request->bRequestType & USB_DIR_IN) {
1088 			dev_dbg(musb->controller, "start IN-DATA\n");
1089 			musb->ep0_stage = MUSB_EP0_IN;
1090 			more = true;
1091 			break;
1092 		} else {
1093 			dev_dbg(musb->controller, "start OUT-DATA\n");
1094 			musb->ep0_stage = MUSB_EP0_OUT;
1095 			more = true;
1096 		}
1097 		/* FALLTHROUGH */
1098 	case MUSB_EP0_OUT:
1099 		fifo_count = min_t(size_t, qh->maxpacket,
1100 				   urb->transfer_buffer_length -
1101 				   urb->actual_length);
1102 		if (fifo_count) {
1103 			fifo_dest = (u8 *) (urb->transfer_buffer
1104 					+ urb->actual_length);
1105 			dev_dbg(musb->controller, "Sending %d byte%s to ep0 fifo %p\n",
1106 					fifo_count,
1107 					(fifo_count == 1) ? "" : "s",
1108 					fifo_dest);
1109 			musb_write_fifo(hw_ep, fifo_count, fifo_dest);
1110 
1111 			urb->actual_length += fifo_count;
1112 			more = true;
1113 		}
1114 		break;
1115 	default:
1116 		ERR("bogus ep0 stage %d\n", musb->ep0_stage);
1117 		break;
1118 	}
1119 
1120 	return more;
1121 }
1122 
1123 /*
1124  * Handle default endpoint interrupt as host. Only called in IRQ time
1125  * from musb_interrupt().
1126  *
1127  * called with controller irqlocked
1128  */
1129 irqreturn_t musb_h_ep0_irq(struct musb *musb)
1130 {
1131 	struct urb		*urb;
1132 	u16			csr, len;
1133 	int			status = 0;
1134 	void __iomem		*mbase = musb->mregs;
1135 	struct musb_hw_ep	*hw_ep = musb->control_ep;
1136 	void __iomem		*epio = hw_ep->regs;
1137 	struct musb_qh		*qh = hw_ep->in_qh;
1138 	bool			complete = false;
1139 	irqreturn_t		retval = IRQ_NONE;
1140 
1141 	/* ep0 only has one queue, "in" */
1142 	urb = next_urb(qh);
1143 
1144 	musb_ep_select(mbase, 0);
1145 	csr = musb_readw(epio, MUSB_CSR0);
1146 	len = (csr & MUSB_CSR0_RXPKTRDY)
1147 			? musb_readb(epio, MUSB_COUNT0)
1148 			: 0;
1149 
1150 	dev_dbg(musb->controller, "<== csr0 %04x, qh %p, count %d, urb %p, stage %d\n",
1151 		csr, qh, len, urb, musb->ep0_stage);
1152 
1153 	/* if we just did status stage, we are done */
1154 	if (MUSB_EP0_STATUS == musb->ep0_stage) {
1155 		retval = IRQ_HANDLED;
1156 		complete = true;
1157 	}
1158 
1159 	/* prepare status */
1160 	if (csr & MUSB_CSR0_H_RXSTALL) {
1161 		dev_dbg(musb->controller, "STALLING ENDPOINT\n");
1162 		status = -EPIPE;
1163 
1164 	} else if (csr & MUSB_CSR0_H_ERROR) {
1165 		dev_dbg(musb->controller, "no response, csr0 %04x\n", csr);
1166 		status = -EPROTO;
1167 
1168 	} else if (csr & MUSB_CSR0_H_NAKTIMEOUT) {
1169 		dev_dbg(musb->controller, "control NAK timeout\n");
1170 
1171 		/* NOTE:  this code path would be a good place to PAUSE a
1172 		 * control transfer, if another one is queued, so that
1173 		 * ep0 is more likely to stay busy.  That's already done
1174 		 * for bulk RX transfers.
1175 		 *
1176 		 * if (qh->ring.next != &musb->control), then
1177 		 * we have a candidate... NAKing is *NOT* an error
1178 		 */
1179 		musb_writew(epio, MUSB_CSR0, 0);
1180 		retval = IRQ_HANDLED;
1181 	}
1182 
1183 	if (status) {
1184 		dev_dbg(musb->controller, "aborting\n");
1185 		retval = IRQ_HANDLED;
1186 		if (urb)
1187 			urb->status = status;
1188 		complete = true;
1189 
1190 		/* use the proper sequence to abort the transfer */
1191 		if (csr & MUSB_CSR0_H_REQPKT) {
1192 			csr &= ~MUSB_CSR0_H_REQPKT;
1193 			musb_writew(epio, MUSB_CSR0, csr);
1194 			csr &= ~MUSB_CSR0_H_NAKTIMEOUT;
1195 			musb_writew(epio, MUSB_CSR0, csr);
1196 		} else {
1197 			musb_h_ep0_flush_fifo(hw_ep);
1198 		}
1199 
1200 		musb_writeb(epio, MUSB_NAKLIMIT0, 0);
1201 
1202 		/* clear it */
1203 		musb_writew(epio, MUSB_CSR0, 0);
1204 	}
1205 
1206 	if (unlikely(!urb)) {
1207 		/* stop endpoint since we have no place for its data, this
1208 		 * SHOULD NEVER HAPPEN! */
1209 		ERR("no URB for end 0\n");
1210 
1211 		musb_h_ep0_flush_fifo(hw_ep);
1212 		goto done;
1213 	}
1214 
1215 	if (!complete) {
1216 		/* call common logic and prepare response */
1217 		if (musb_h_ep0_continue(musb, len, urb)) {
1218 			/* more packets required */
1219 			csr = (MUSB_EP0_IN == musb->ep0_stage)
1220 				?  MUSB_CSR0_H_REQPKT : MUSB_CSR0_TXPKTRDY;
1221 		} else {
1222 			/* data transfer complete; perform status phase */
1223 			if (usb_pipeout(urb->pipe)
1224 					|| !urb->transfer_buffer_length)
1225 				csr = MUSB_CSR0_H_STATUSPKT
1226 					| MUSB_CSR0_H_REQPKT;
1227 			else
1228 				csr = MUSB_CSR0_H_STATUSPKT
1229 					| MUSB_CSR0_TXPKTRDY;
1230 
1231 			/* disable ping token in status phase */
1232 			csr |= MUSB_CSR0_H_DIS_PING;
1233 
1234 			/* flag status stage */
1235 			musb->ep0_stage = MUSB_EP0_STATUS;
1236 
1237 			dev_dbg(musb->controller, "ep0 STATUS, csr %04x\n", csr);
1238 
1239 		}
1240 		musb_writew(epio, MUSB_CSR0, csr);
1241 		retval = IRQ_HANDLED;
1242 	} else
1243 		musb->ep0_stage = MUSB_EP0_IDLE;
1244 
1245 	/* call completion handler if done */
1246 	if (complete)
1247 		musb_advance_schedule(musb, urb, hw_ep, 1);
1248 done:
1249 	return retval;
1250 }
1251 
1252 
1253 #ifdef CONFIG_USB_INVENTRA_DMA
1254 
1255 /* Host side TX (OUT) using Mentor DMA works as follows:
1256 	submit_urb ->
1257 		- if queue was empty, Program Endpoint
1258 		- ... which starts DMA to fifo in mode 1 or 0
1259 
1260 	DMA Isr (transfer complete) -> TxAvail()
1261 		- Stop DMA (~DmaEnab)	(<--- Alert ... currently happens
1262 					only in musb_cleanup_urb)
1263 		- TxPktRdy has to be set in mode 0 or for
1264 			short packets in mode 1.
1265 */
1266 
1267 #endif
1268 
1269 /* Service a Tx-Available or dma completion irq for the endpoint */
1270 void musb_host_tx(struct musb *musb, u8 epnum)
1271 {
1272 	int			pipe;
1273 	bool			done = false;
1274 	u16			tx_csr;
1275 	size_t			length = 0;
1276 	size_t			offset = 0;
1277 	struct musb_hw_ep	*hw_ep = musb->endpoints + epnum;
1278 	void __iomem		*epio = hw_ep->regs;
1279 	struct musb_qh		*qh = hw_ep->out_qh;
1280 	struct urb		*urb = next_urb(qh);
1281 	u32			status = 0;
1282 	void __iomem		*mbase = musb->mregs;
1283 	struct dma_channel	*dma;
1284 	bool			transfer_pending = false;
1285 
1286 	musb_ep_select(mbase, epnum);
1287 	tx_csr = musb_readw(epio, MUSB_TXCSR);
1288 
1289 	/* with CPPI, DMA sometimes triggers "extra" irqs */
1290 	if (!urb) {
1291 		dev_dbg(musb->controller, "extra TX%d ready, csr %04x\n", epnum, tx_csr);
1292 		return;
1293 	}
1294 
1295 	pipe = urb->pipe;
1296 	dma = is_dma_capable() ? hw_ep->tx_channel : NULL;
1297 	dev_dbg(musb->controller, "OUT/TX%d end, csr %04x%s\n", epnum, tx_csr,
1298 			dma ? ", dma" : "");
1299 
1300 	/* check for errors */
1301 	if (tx_csr & MUSB_TXCSR_H_RXSTALL) {
1302 		/* dma was disabled, fifo flushed */
1303 		dev_dbg(musb->controller, "TX end %d stall\n", epnum);
1304 
1305 		/* stall; record URB status */
1306 		status = -EPIPE;
1307 
1308 	} else if (tx_csr & MUSB_TXCSR_H_ERROR) {
1309 		/* (NON-ISO) dma was disabled, fifo flushed */
1310 		dev_dbg(musb->controller, "TX 3strikes on ep=%d\n", epnum);
1311 
1312 		status = -ETIMEDOUT;
1313 
1314 	} else if (tx_csr & MUSB_TXCSR_H_NAKTIMEOUT) {
1315 		if (USB_ENDPOINT_XFER_BULK == qh->type && qh->mux == 1
1316 				&& !list_is_singular(&musb->out_bulk)) {
1317 			dev_dbg(musb->controller,
1318 				"NAK timeout on TX%d ep\n", epnum);
1319 			musb_bulk_nak_timeout(musb, hw_ep, 0);
1320 		} else {
1321 			dev_dbg(musb->controller,
1322 				"TX end=%d device not responding\n", epnum);
1323 			/* NOTE:  this code path would be a good place to PAUSE a
1324 			 * transfer, if there's some other (nonperiodic) tx urb
1325 			 * that could use this fifo.  (dma complicates it...)
1326 			 * That's already done for bulk RX transfers.
1327 			 *
1328 			 * if (bulk && qh->ring.next != &musb->out_bulk), then
1329 			 * we have a candidate... NAKing is *NOT* an error
1330 			 */
1331 			musb_ep_select(mbase, epnum);
1332 			musb_writew(epio, MUSB_TXCSR,
1333 					MUSB_TXCSR_H_WZC_BITS
1334 					| MUSB_TXCSR_TXPKTRDY);
1335 		}
1336 			return;
1337 	}
1338 
1339 done:
1340 	if (status) {
1341 		if (dma_channel_status(dma) == MUSB_DMA_STATUS_BUSY) {
1342 			dma->status = MUSB_DMA_STATUS_CORE_ABORT;
1343 			musb->dma_controller->channel_abort(dma);
1344 		}
1345 
1346 		/* do the proper sequence to abort the transfer in the
1347 		 * usb core; the dma engine should already be stopped.
1348 		 */
1349 		musb_h_tx_flush_fifo(hw_ep);
1350 		tx_csr &= ~(MUSB_TXCSR_AUTOSET
1351 				| MUSB_TXCSR_DMAENAB
1352 				| MUSB_TXCSR_H_ERROR
1353 				| MUSB_TXCSR_H_RXSTALL
1354 				| MUSB_TXCSR_H_NAKTIMEOUT
1355 				);
1356 
1357 		musb_ep_select(mbase, epnum);
1358 		musb_writew(epio, MUSB_TXCSR, tx_csr);
1359 		/* REVISIT may need to clear FLUSHFIFO ... */
1360 		musb_writew(epio, MUSB_TXCSR, tx_csr);
1361 		musb_writeb(epio, MUSB_TXINTERVAL, 0);
1362 
1363 		done = true;
1364 	}
1365 
1366 	/* second cppi case */
1367 	if (dma_channel_status(dma) == MUSB_DMA_STATUS_BUSY) {
1368 		dev_dbg(musb->controller, "extra TX%d ready, csr %04x\n", epnum, tx_csr);
1369 		return;
1370 	}
1371 
1372 	if (is_dma_capable() && dma && !status) {
1373 		/*
1374 		 * DMA has completed.  But if we're using DMA mode 1 (multi
1375 		 * packet DMA), we need a terminal TXPKTRDY interrupt before
1376 		 * we can consider this transfer completed, lest we trash
1377 		 * its last packet when writing the next URB's data.  So we
1378 		 * switch back to mode 0 to get that interrupt; we'll come
1379 		 * back here once it happens.
1380 		 */
1381 		if (tx_csr & MUSB_TXCSR_DMAMODE) {
1382 			/*
1383 			 * We shouldn't clear DMAMODE with DMAENAB set; so
1384 			 * clear them in a safe order.  That should be OK
1385 			 * once TXPKTRDY has been set (and I've never seen
1386 			 * it being 0 at this moment -- DMA interrupt latency
1387 			 * is significant) but if it hasn't been then we have
1388 			 * no choice but to stop being polite and ignore the
1389 			 * programmer's guide... :-)
1390 			 *
1391 			 * Note that we must write TXCSR with TXPKTRDY cleared
1392 			 * in order not to re-trigger the packet send (this bit
1393 			 * can't be cleared by CPU), and there's another caveat:
1394 			 * TXPKTRDY may be set shortly and then cleared in the
1395 			 * double-buffered FIFO mode, so we do an extra TXCSR
1396 			 * read for debouncing...
1397 			 */
1398 			tx_csr &= musb_readw(epio, MUSB_TXCSR);
1399 			if (tx_csr & MUSB_TXCSR_TXPKTRDY) {
1400 				tx_csr &= ~(MUSB_TXCSR_DMAENAB |
1401 					    MUSB_TXCSR_TXPKTRDY);
1402 				musb_writew(epio, MUSB_TXCSR,
1403 					    tx_csr | MUSB_TXCSR_H_WZC_BITS);
1404 			}
1405 			tx_csr &= ~(MUSB_TXCSR_DMAMODE |
1406 				    MUSB_TXCSR_TXPKTRDY);
1407 			musb_writew(epio, MUSB_TXCSR,
1408 				    tx_csr | MUSB_TXCSR_H_WZC_BITS);
1409 
1410 			/*
1411 			 * There is no guarantee that we'll get an interrupt
1412 			 * after clearing DMAMODE as we might have done this
1413 			 * too late (after TXPKTRDY was cleared by controller).
1414 			 * Re-read TXCSR as we have spoiled its previous value.
1415 			 */
1416 			tx_csr = musb_readw(epio, MUSB_TXCSR);
1417 		}
1418 
1419 		/*
1420 		 * We may get here from a DMA completion or TXPKTRDY interrupt.
1421 		 * In any case, we must check the FIFO status here and bail out
1422 		 * only if the FIFO still has data -- that should prevent the
1423 		 * "missed" TXPKTRDY interrupts and deal with double-buffered
1424 		 * FIFO mode too...
1425 		 */
1426 		if (tx_csr & (MUSB_TXCSR_FIFONOTEMPTY | MUSB_TXCSR_TXPKTRDY)) {
1427 			dev_dbg(musb->controller, "DMA complete but packet still in FIFO, "
1428 			    "CSR %04x\n", tx_csr);
1429 			return;
1430 		}
1431 	}
1432 
1433 	if (!status || dma || usb_pipeisoc(pipe)) {
1434 		if (dma)
1435 			length = dma->actual_len;
1436 		else
1437 			length = qh->segsize;
1438 		qh->offset += length;
1439 
1440 		if (usb_pipeisoc(pipe)) {
1441 			struct usb_iso_packet_descriptor	*d;
1442 
1443 			d = urb->iso_frame_desc + qh->iso_idx;
1444 			d->actual_length = length;
1445 			d->status = status;
1446 			if (++qh->iso_idx >= urb->number_of_packets) {
1447 				done = true;
1448 			} else {
1449 				d++;
1450 				offset = d->offset;
1451 				length = d->length;
1452 			}
1453 		} else if (dma && urb->transfer_buffer_length == qh->offset) {
1454 			done = true;
1455 		} else {
1456 			/* see if we need to send more data, or ZLP */
1457 			if (qh->segsize < qh->maxpacket)
1458 				done = true;
1459 			else if (qh->offset == urb->transfer_buffer_length
1460 					&& !(urb->transfer_flags
1461 						& URB_ZERO_PACKET))
1462 				done = true;
1463 			if (!done) {
1464 				offset = qh->offset;
1465 				length = urb->transfer_buffer_length - offset;
1466 				transfer_pending = true;
1467 			}
1468 		}
1469 	}
1470 
1471 	/* urb->status != -EINPROGRESS means request has been faulted,
1472 	 * so we must abort this transfer after cleanup
1473 	 */
1474 	if (urb->status != -EINPROGRESS) {
1475 		done = true;
1476 		if (status == 0)
1477 			status = urb->status;
1478 	}
1479 
1480 	if (done) {
1481 		/* set status */
1482 		urb->status = status;
1483 		urb->actual_length = qh->offset;
1484 		musb_advance_schedule(musb, urb, hw_ep, USB_DIR_OUT);
1485 		return;
1486 	} else if ((usb_pipeisoc(pipe) || transfer_pending) && dma) {
1487 		if (musb_tx_dma_program(musb->dma_controller, hw_ep, qh, urb,
1488 				offset, length)) {
1489 			if (is_cppi_enabled(musb) || tusb_dma_omap(musb))
1490 				musb_h_tx_dma_start(hw_ep);
1491 			return;
1492 		}
1493 	} else	if (tx_csr & MUSB_TXCSR_DMAENAB) {
1494 		dev_dbg(musb->controller, "not complete, but DMA enabled?\n");
1495 		return;
1496 	}
1497 
1498 	/*
1499 	 * PIO: start next packet in this URB.
1500 	 *
1501 	 * REVISIT: some docs say that when hw_ep->tx_double_buffered,
1502 	 * (and presumably, FIFO is not half-full) we should write *two*
1503 	 * packets before updating TXCSR; other docs disagree...
1504 	 */
1505 	if (length > qh->maxpacket)
1506 		length = qh->maxpacket;
1507 	/* Unmap the buffer so that CPU can use it */
1508 	usb_hcd_unmap_urb_for_dma(musb->hcd, urb);
1509 
1510 	/*
1511 	 * We need to map sg if the transfer_buffer is
1512 	 * NULL.
1513 	 */
1514 	if (!urb->transfer_buffer)
1515 		qh->use_sg = true;
1516 
1517 	if (qh->use_sg) {
1518 		/* sg_miter_start is already done in musb_ep_program */
1519 		if (!sg_miter_next(&qh->sg_miter)) {
1520 			dev_err(musb->controller, "error: sg list empty\n");
1521 			sg_miter_stop(&qh->sg_miter);
1522 			status = -EINVAL;
1523 			goto done;
1524 		}
1525 		urb->transfer_buffer = qh->sg_miter.addr;
1526 		length = min_t(u32, length, qh->sg_miter.length);
1527 		musb_write_fifo(hw_ep, length, urb->transfer_buffer);
1528 		qh->sg_miter.consumed = length;
1529 		sg_miter_stop(&qh->sg_miter);
1530 	} else {
1531 		musb_write_fifo(hw_ep, length, urb->transfer_buffer + offset);
1532 	}
1533 
1534 	qh->segsize = length;
1535 
1536 	if (qh->use_sg) {
1537 		if (offset + length >= urb->transfer_buffer_length)
1538 			qh->use_sg = false;
1539 	}
1540 
1541 	musb_ep_select(mbase, epnum);
1542 	musb_writew(epio, MUSB_TXCSR,
1543 			MUSB_TXCSR_H_WZC_BITS | MUSB_TXCSR_TXPKTRDY);
1544 }
1545 
1546 #ifdef CONFIG_USB_TI_CPPI41_DMA
1547 /* Seems to set up ISO for cppi41 and not advance len. See commit c57c41d */
1548 static int musb_rx_dma_iso_cppi41(struct dma_controller *dma,
1549 				  struct musb_hw_ep *hw_ep,
1550 				  struct musb_qh *qh,
1551 				  struct urb *urb,
1552 				  size_t len)
1553 {
1554 	struct dma_channel *channel = hw_ep->tx_channel;
1555 	void __iomem *epio = hw_ep->regs;
1556 	dma_addr_t *buf;
1557 	u32 length, res;
1558 	u16 val;
1559 
1560 	buf = (void *)urb->iso_frame_desc[qh->iso_idx].offset +
1561 		(u32)urb->transfer_dma;
1562 
1563 	length = urb->iso_frame_desc[qh->iso_idx].length;
1564 
1565 	val = musb_readw(epio, MUSB_RXCSR);
1566 	val |= MUSB_RXCSR_DMAENAB;
1567 	musb_writew(hw_ep->regs, MUSB_RXCSR, val);
1568 
1569 	res = dma->channel_program(channel, qh->maxpacket, 0,
1570 				   (u32)buf, length);
1571 
1572 	return res;
1573 }
1574 #else
1575 static inline int musb_rx_dma_iso_cppi41(struct dma_controller *dma,
1576 					 struct musb_hw_ep *hw_ep,
1577 					 struct musb_qh *qh,
1578 					 struct urb *urb,
1579 					 size_t len)
1580 {
1581 	return false;
1582 }
1583 #endif
1584 
1585 #if defined(CONFIG_USB_INVENTRA_DMA) || defined(CONFIG_USB_UX500_DMA) || \
1586 	defined(CONFIG_USB_TI_CPPI41_DMA)
1587 /* Host side RX (IN) using Mentor DMA works as follows:
1588 	submit_urb ->
1589 		- if queue was empty, ProgramEndpoint
1590 		- first IN token is sent out (by setting ReqPkt)
1591 	LinuxIsr -> RxReady()
1592 	/\	=> first packet is received
1593 	|	- Set in mode 0 (DmaEnab, ~ReqPkt)
1594 	|		-> DMA Isr (transfer complete) -> RxReady()
1595 	|		    - Ack receive (~RxPktRdy), turn off DMA (~DmaEnab)
1596 	|		    - if urb not complete, send next IN token (ReqPkt)
1597 	|			   |		else complete urb.
1598 	|			   |
1599 	---------------------------
1600  *
1601  * Nuances of mode 1:
1602  *	For short packets, no ack (+RxPktRdy) is sent automatically
1603  *	(even if AutoClear is ON)
1604  *	For full packets, ack (~RxPktRdy) and next IN token (+ReqPkt) is sent
1605  *	automatically => major problem, as collecting the next packet becomes
1606  *	difficult. Hence mode 1 is not used.
1607  *
1608  * REVISIT
1609  *	All we care about at this driver level is that
1610  *       (a) all URBs terminate with REQPKT cleared and fifo(s) empty;
1611  *       (b) termination conditions are: short RX, or buffer full;
1612  *       (c) fault modes include
1613  *           - iff URB_SHORT_NOT_OK, short RX status is -EREMOTEIO.
1614  *             (and that endpoint's dma queue stops immediately)
1615  *           - overflow (full, PLUS more bytes in the terminal packet)
1616  *
1617  *	So for example, usb-storage sets URB_SHORT_NOT_OK, and would
1618  *	thus be a great candidate for using mode 1 ... for all but the
1619  *	last packet of one URB's transfer.
1620  */
1621 static int musb_rx_dma_inventra_cppi41(struct dma_controller *dma,
1622 				       struct musb_hw_ep *hw_ep,
1623 				       struct musb_qh *qh,
1624 				       struct urb *urb,
1625 				       size_t len)
1626 {
1627 	struct dma_channel *channel = hw_ep->rx_channel;
1628 	void __iomem *epio = hw_ep->regs;
1629 	u16 val;
1630 	int pipe;
1631 	bool done;
1632 
1633 	pipe = urb->pipe;
1634 
1635 	if (usb_pipeisoc(pipe)) {
1636 		struct usb_iso_packet_descriptor *d;
1637 
1638 		d = urb->iso_frame_desc + qh->iso_idx;
1639 		d->actual_length = len;
1640 
1641 		/* even if there was an error, we did the dma
1642 		 * for iso_frame_desc->length
1643 		 */
1644 		if (d->status != -EILSEQ && d->status != -EOVERFLOW)
1645 			d->status = 0;
1646 
1647 		if (++qh->iso_idx >= urb->number_of_packets) {
1648 			done = true;
1649 		} else {
1650 			/* REVISIT: Why ignore return value here? */
1651 			if (musb_dma_cppi41(hw_ep->musb))
1652 				done = musb_rx_dma_iso_cppi41(dma, hw_ep, qh,
1653 							      urb, len);
1654 			done = false;
1655 		}
1656 
1657 	} else  {
1658 		/* done if urb buffer is full or short packet is recd */
1659 		done = (urb->actual_length + len >=
1660 			urb->transfer_buffer_length
1661 			|| channel->actual_len < qh->maxpacket
1662 			|| channel->rx_packet_done);
1663 	}
1664 
1665 	/* send IN token for next packet, without AUTOREQ */
1666 	if (!done) {
1667 		val = musb_readw(epio, MUSB_RXCSR);
1668 		val |= MUSB_RXCSR_H_REQPKT;
1669 		musb_writew(epio, MUSB_RXCSR, MUSB_RXCSR_H_WZC_BITS | val);
1670 	}
1671 
1672 	return done;
1673 }
1674 
1675 /* Disadvantage of using mode 1:
1676  *	It's basically usable only for mass storage class; essentially all
1677  *	other protocols also terminate transfers on short packets.
1678  *
1679  * Details:
1680  *	An extra IN token is sent at the end of the transfer (due to AUTOREQ)
1681  *	If you try to use mode 1 for (transfer_buffer_length - 512), and try
1682  *	to use the extra IN token to grab the last packet using mode 0, then
1683  *	the problem is that you cannot be sure when the device will send the
1684  *	last packet and RxPktRdy set. Sometimes the packet is recd too soon
1685  *	such that it gets lost when RxCSR is re-set at the end of the mode 1
1686  *	transfer, while sometimes it is recd just a little late so that if you
1687  *	try to configure for mode 0 soon after the mode 1 transfer is
1688  *	completed, you will find rxcount 0. Okay, so you might think why not
1689  *	wait for an interrupt when the pkt is recd. Well, you won't get any!
1690  */
1691 static int musb_rx_dma_in_inventra_cppi41(struct dma_controller *dma,
1692 					  struct musb_hw_ep *hw_ep,
1693 					  struct musb_qh *qh,
1694 					  struct urb *urb,
1695 					  size_t len,
1696 					  u8 iso_err)
1697 {
1698 	struct musb *musb = hw_ep->musb;
1699 	void __iomem *epio = hw_ep->regs;
1700 	struct dma_channel *channel = hw_ep->rx_channel;
1701 	u16 rx_count, val;
1702 	int length, pipe, done;
1703 	dma_addr_t buf;
1704 
1705 	rx_count = musb_readw(epio, MUSB_RXCOUNT);
1706 	pipe = urb->pipe;
1707 
1708 	if (usb_pipeisoc(pipe)) {
1709 		int d_status = 0;
1710 		struct usb_iso_packet_descriptor *d;
1711 
1712 		d = urb->iso_frame_desc + qh->iso_idx;
1713 
1714 		if (iso_err) {
1715 			d_status = -EILSEQ;
1716 			urb->error_count++;
1717 		}
1718 		if (rx_count > d->length) {
1719 			if (d_status == 0) {
1720 				d_status = -EOVERFLOW;
1721 				urb->error_count++;
1722 			}
1723 			dev_dbg(musb->controller, "** OVERFLOW %d into %d\n",
1724 				rx_count, d->length);
1725 
1726 			length = d->length;
1727 		} else
1728 			length = rx_count;
1729 		d->status = d_status;
1730 		buf = urb->transfer_dma + d->offset;
1731 	} else {
1732 		length = rx_count;
1733 		buf = urb->transfer_dma + urb->actual_length;
1734 	}
1735 
1736 	channel->desired_mode = 0;
1737 #ifdef USE_MODE1
1738 	/* because of the issue below, mode 1 will
1739 	 * only rarely behave with correct semantics.
1740 	 */
1741 	if ((urb->transfer_flags & URB_SHORT_NOT_OK)
1742 	    && (urb->transfer_buffer_length - urb->actual_length)
1743 	    > qh->maxpacket)
1744 		channel->desired_mode = 1;
1745 	if (rx_count < hw_ep->max_packet_sz_rx) {
1746 		length = rx_count;
1747 		channel->desired_mode = 0;
1748 	} else {
1749 		length = urb->transfer_buffer_length;
1750 	}
1751 #endif
1752 
1753 	/* See comments above on disadvantages of using mode 1 */
1754 	val = musb_readw(epio, MUSB_RXCSR);
1755 	val &= ~MUSB_RXCSR_H_REQPKT;
1756 
1757 	if (channel->desired_mode == 0)
1758 		val &= ~MUSB_RXCSR_H_AUTOREQ;
1759 	else
1760 		val |= MUSB_RXCSR_H_AUTOREQ;
1761 	val |= MUSB_RXCSR_DMAENAB;
1762 
1763 	/* autoclear shouldn't be set in high bandwidth */
1764 	if (qh->hb_mult == 1)
1765 		val |= MUSB_RXCSR_AUTOCLEAR;
1766 
1767 	musb_writew(epio, MUSB_RXCSR, MUSB_RXCSR_H_WZC_BITS | val);
1768 
1769 	/* REVISIT if when actual_length != 0,
1770 	 * transfer_buffer_length needs to be
1771 	 * adjusted first...
1772 	 */
1773 	done = dma->channel_program(channel, qh->maxpacket,
1774 				   channel->desired_mode,
1775 				   buf, length);
1776 
1777 	if (!done) {
1778 		dma->channel_release(channel);
1779 		hw_ep->rx_channel = NULL;
1780 		channel = NULL;
1781 		val = musb_readw(epio, MUSB_RXCSR);
1782 		val &= ~(MUSB_RXCSR_DMAENAB
1783 			 | MUSB_RXCSR_H_AUTOREQ
1784 			 | MUSB_RXCSR_AUTOCLEAR);
1785 		musb_writew(epio, MUSB_RXCSR, val);
1786 	}
1787 
1788 	return done;
1789 }
1790 #else
1791 static inline int musb_rx_dma_inventra_cppi41(struct dma_controller *dma,
1792 					      struct musb_hw_ep *hw_ep,
1793 					      struct musb_qh *qh,
1794 					      struct urb *urb,
1795 					      size_t len)
1796 {
1797 	return false;
1798 }
1799 
1800 static inline int musb_rx_dma_in_inventra_cppi41(struct dma_controller *dma,
1801 						 struct musb_hw_ep *hw_ep,
1802 						 struct musb_qh *qh,
1803 						 struct urb *urb,
1804 						 size_t len,
1805 						 u8 iso_err)
1806 {
1807 	return false;
1808 }
1809 #endif
1810 
1811 /*
1812  * Service an RX interrupt for the given IN endpoint; docs cover bulk, iso,
1813  * and high-bandwidth IN transfer cases.
1814  */
1815 void musb_host_rx(struct musb *musb, u8 epnum)
1816 {
1817 	struct urb		*urb;
1818 	struct musb_hw_ep	*hw_ep = musb->endpoints + epnum;
1819 	struct dma_controller	*c = musb->dma_controller;
1820 	void __iomem		*epio = hw_ep->regs;
1821 	struct musb_qh		*qh = hw_ep->in_qh;
1822 	size_t			xfer_len;
1823 	void __iomem		*mbase = musb->mregs;
1824 	int			pipe;
1825 	u16			rx_csr, val;
1826 	bool			iso_err = false;
1827 	bool			done = false;
1828 	u32			status;
1829 	struct dma_channel	*dma;
1830 	unsigned int sg_flags = SG_MITER_ATOMIC | SG_MITER_TO_SG;
1831 
1832 	musb_ep_select(mbase, epnum);
1833 
1834 	urb = next_urb(qh);
1835 	dma = is_dma_capable() ? hw_ep->rx_channel : NULL;
1836 	status = 0;
1837 	xfer_len = 0;
1838 
1839 	rx_csr = musb_readw(epio, MUSB_RXCSR);
1840 	val = rx_csr;
1841 
1842 	if (unlikely(!urb)) {
1843 		/* REVISIT -- THIS SHOULD NEVER HAPPEN ... but, at least
1844 		 * usbtest #11 (unlinks) triggers it regularly, sometimes
1845 		 * with fifo full.  (Only with DMA??)
1846 		 */
1847 		dev_dbg(musb->controller, "BOGUS RX%d ready, csr %04x, count %d\n", epnum, val,
1848 			musb_readw(epio, MUSB_RXCOUNT));
1849 		musb_h_flush_rxfifo(hw_ep, MUSB_RXCSR_CLRDATATOG);
1850 		return;
1851 	}
1852 
1853 	pipe = urb->pipe;
1854 
1855 	dev_dbg(musb->controller, "<== hw %d rxcsr %04x, urb actual %d (+dma %zu)\n",
1856 		epnum, rx_csr, urb->actual_length,
1857 		dma ? dma->actual_len : 0);
1858 
1859 	/* check for errors, concurrent stall & unlink is not really
1860 	 * handled yet! */
1861 	if (rx_csr & MUSB_RXCSR_H_RXSTALL) {
1862 		dev_dbg(musb->controller, "RX end %d STALL\n", epnum);
1863 
1864 		/* stall; record URB status */
1865 		status = -EPIPE;
1866 
1867 	} else if (rx_csr & MUSB_RXCSR_H_ERROR) {
1868 		dev_dbg(musb->controller, "end %d RX proto error\n", epnum);
1869 
1870 		status = -EPROTO;
1871 		musb_writeb(epio, MUSB_RXINTERVAL, 0);
1872 
1873 	} else if (rx_csr & MUSB_RXCSR_DATAERROR) {
1874 
1875 		if (USB_ENDPOINT_XFER_ISOC != qh->type) {
1876 			dev_dbg(musb->controller, "RX end %d NAK timeout\n", epnum);
1877 
1878 			/* NOTE: NAKing is *NOT* an error, so we want to
1879 			 * continue.  Except ... if there's a request for
1880 			 * another QH, use that instead of starving it.
1881 			 *
1882 			 * Devices like Ethernet and serial adapters keep
1883 			 * reads posted at all times, which will starve
1884 			 * other devices without this logic.
1885 			 */
1886 			if (usb_pipebulk(urb->pipe)
1887 					&& qh->mux == 1
1888 					&& !list_is_singular(&musb->in_bulk)) {
1889 				musb_bulk_nak_timeout(musb, hw_ep, 1);
1890 				return;
1891 			}
1892 			musb_ep_select(mbase, epnum);
1893 			rx_csr |= MUSB_RXCSR_H_WZC_BITS;
1894 			rx_csr &= ~MUSB_RXCSR_DATAERROR;
1895 			musb_writew(epio, MUSB_RXCSR, rx_csr);
1896 
1897 			goto finish;
1898 		} else {
1899 			dev_dbg(musb->controller, "RX end %d ISO data error\n", epnum);
1900 			/* packet error reported later */
1901 			iso_err = true;
1902 		}
1903 	} else if (rx_csr & MUSB_RXCSR_INCOMPRX) {
1904 		dev_dbg(musb->controller, "end %d high bandwidth incomplete ISO packet RX\n",
1905 				epnum);
1906 		status = -EPROTO;
1907 	}
1908 
1909 	/* faults abort the transfer */
1910 	if (status) {
1911 		/* clean up dma and collect transfer count */
1912 		if (dma_channel_status(dma) == MUSB_DMA_STATUS_BUSY) {
1913 			dma->status = MUSB_DMA_STATUS_CORE_ABORT;
1914 			musb->dma_controller->channel_abort(dma);
1915 			xfer_len = dma->actual_len;
1916 		}
1917 		musb_h_flush_rxfifo(hw_ep, MUSB_RXCSR_CLRDATATOG);
1918 		musb_writeb(epio, MUSB_RXINTERVAL, 0);
1919 		done = true;
1920 		goto finish;
1921 	}
1922 
1923 	if (unlikely(dma_channel_status(dma) == MUSB_DMA_STATUS_BUSY)) {
1924 		/* SHOULD NEVER HAPPEN ... but at least DaVinci has done it */
1925 		ERR("RX%d dma busy, csr %04x\n", epnum, rx_csr);
1926 		goto finish;
1927 	}
1928 
1929 	/* thorough shutdown for now ... given more precise fault handling
1930 	 * and better queueing support, we might keep a DMA pipeline going
1931 	 * while processing this irq for earlier completions.
1932 	 */
1933 
1934 	/* FIXME this is _way_ too much in-line logic for Mentor DMA */
1935 	if (!musb_dma_inventra(musb) && !musb_dma_ux500(musb) &&
1936 	    (rx_csr & MUSB_RXCSR_H_REQPKT)) {
1937 		/* REVISIT this happened for a while on some short reads...
1938 		 * the cleanup still needs investigation... looks bad...
1939 		 * and also duplicates dma cleanup code above ... plus,
1940 		 * shouldn't this be the "half full" double buffer case?
1941 		 */
1942 		if (dma_channel_status(dma) == MUSB_DMA_STATUS_BUSY) {
1943 			dma->status = MUSB_DMA_STATUS_CORE_ABORT;
1944 			musb->dma_controller->channel_abort(dma);
1945 			xfer_len = dma->actual_len;
1946 			done = true;
1947 		}
1948 
1949 		dev_dbg(musb->controller, "RXCSR%d %04x, reqpkt, len %zu%s\n", epnum, rx_csr,
1950 				xfer_len, dma ? ", dma" : "");
1951 		rx_csr &= ~MUSB_RXCSR_H_REQPKT;
1952 
1953 		musb_ep_select(mbase, epnum);
1954 		musb_writew(epio, MUSB_RXCSR,
1955 				MUSB_RXCSR_H_WZC_BITS | rx_csr);
1956 	}
1957 
1958 	if (dma && (rx_csr & MUSB_RXCSR_DMAENAB)) {
1959 		xfer_len = dma->actual_len;
1960 
1961 		val &= ~(MUSB_RXCSR_DMAENAB
1962 			| MUSB_RXCSR_H_AUTOREQ
1963 			| MUSB_RXCSR_AUTOCLEAR
1964 			| MUSB_RXCSR_RXPKTRDY);
1965 		musb_writew(hw_ep->regs, MUSB_RXCSR, val);
1966 
1967 		if (musb_dma_inventra(musb) || musb_dma_ux500(musb) ||
1968 		    musb_dma_cppi41(musb)) {
1969 			    done = musb_rx_dma_inventra_cppi41(c, hw_ep, qh, urb, xfer_len);
1970 			    dev_dbg(hw_ep->musb->controller,
1971 				    "ep %d dma %s, rxcsr %04x, rxcount %d\n",
1972 				    epnum, done ? "off" : "reset",
1973 				    musb_readw(epio, MUSB_RXCSR),
1974 				    musb_readw(epio, MUSB_RXCOUNT));
1975 		} else {
1976 			done = true;
1977 		}
1978 
1979 	} else if (urb->status == -EINPROGRESS) {
1980 		/* if no errors, be sure a packet is ready for unloading */
1981 		if (unlikely(!(rx_csr & MUSB_RXCSR_RXPKTRDY))) {
1982 			status = -EPROTO;
1983 			ERR("Rx interrupt with no errors or packet!\n");
1984 
1985 			/* FIXME this is another "SHOULD NEVER HAPPEN" */
1986 
1987 /* SCRUB (RX) */
1988 			/* do the proper sequence to abort the transfer */
1989 			musb_ep_select(mbase, epnum);
1990 			val &= ~MUSB_RXCSR_H_REQPKT;
1991 			musb_writew(epio, MUSB_RXCSR, val);
1992 			goto finish;
1993 		}
1994 
1995 		/* we are expecting IN packets */
1996 		if ((musb_dma_inventra(musb) || musb_dma_ux500(musb) ||
1997 		    musb_dma_cppi41(musb)) && dma) {
1998 			dev_dbg(hw_ep->musb->controller,
1999 				"RX%d count %d, buffer 0x%llx len %d/%d\n",
2000 				epnum, musb_readw(epio, MUSB_RXCOUNT),
2001 				(unsigned long long) urb->transfer_dma
2002 				+ urb->actual_length,
2003 				qh->offset,
2004 				urb->transfer_buffer_length);
2005 
2006 			if (musb_rx_dma_in_inventra_cppi41(c, hw_ep, qh, urb,
2007 							   xfer_len, iso_err))
2008 				goto finish;
2009 			else
2010 				dev_err(musb->controller, "error: rx_dma failed\n");
2011 		}
2012 
2013 		if (!dma) {
2014 			unsigned int received_len;
2015 
2016 			/* Unmap the buffer so that CPU can use it */
2017 			usb_hcd_unmap_urb_for_dma(musb->hcd, urb);
2018 
2019 			/*
2020 			 * We need to map sg if the transfer_buffer is
2021 			 * NULL.
2022 			 */
2023 			if (!urb->transfer_buffer) {
2024 				qh->use_sg = true;
2025 				sg_miter_start(&qh->sg_miter, urb->sg, 1,
2026 						sg_flags);
2027 			}
2028 
2029 			if (qh->use_sg) {
2030 				if (!sg_miter_next(&qh->sg_miter)) {
2031 					dev_err(musb->controller, "error: sg list empty\n");
2032 					sg_miter_stop(&qh->sg_miter);
2033 					status = -EINVAL;
2034 					done = true;
2035 					goto finish;
2036 				}
2037 				urb->transfer_buffer = qh->sg_miter.addr;
2038 				received_len = urb->actual_length;
2039 				qh->offset = 0x0;
2040 				done = musb_host_packet_rx(musb, urb, epnum,
2041 						iso_err);
2042 				/* Calculate the number of bytes received */
2043 				received_len = urb->actual_length -
2044 					received_len;
2045 				qh->sg_miter.consumed = received_len;
2046 				sg_miter_stop(&qh->sg_miter);
2047 			} else {
2048 				done = musb_host_packet_rx(musb, urb,
2049 						epnum, iso_err);
2050 			}
2051 			dev_dbg(musb->controller, "read %spacket\n", done ? "last " : "");
2052 		}
2053 	}
2054 
2055 finish:
2056 	urb->actual_length += xfer_len;
2057 	qh->offset += xfer_len;
2058 	if (done) {
2059 		if (qh->use_sg)
2060 			qh->use_sg = false;
2061 
2062 		if (urb->status == -EINPROGRESS)
2063 			urb->status = status;
2064 		musb_advance_schedule(musb, urb, hw_ep, USB_DIR_IN);
2065 	}
2066 }
2067 
2068 /* schedule nodes correspond to peripheral endpoints, like an OHCI QH.
2069  * the software schedule associates multiple such nodes with a given
2070  * host side hardware endpoint + direction; scheduling may activate
2071  * that hardware endpoint.
2072  */
2073 static int musb_schedule(
2074 	struct musb		*musb,
2075 	struct musb_qh		*qh,
2076 	int			is_in)
2077 {
2078 	int			idle = 0;
2079 	int			best_diff;
2080 	int			best_end, epnum;
2081 	struct musb_hw_ep	*hw_ep = NULL;
2082 	struct list_head	*head = NULL;
2083 	u8			toggle;
2084 	u8			txtype;
2085 	struct urb		*urb = next_urb(qh);
2086 
2087 	/* use fixed hardware for control and bulk */
2088 	if (qh->type == USB_ENDPOINT_XFER_CONTROL) {
2089 		head = &musb->control;
2090 		hw_ep = musb->control_ep;
2091 		goto success;
2092 	}
2093 
2094 	/* else, periodic transfers get muxed to other endpoints */
2095 
2096 	/*
2097 	 * We know this qh hasn't been scheduled, so all we need to do
2098 	 * is choose which hardware endpoint to put it on ...
2099 	 *
2100 	 * REVISIT what we really want here is a regular schedule tree
2101 	 * like e.g. OHCI uses.
2102 	 */
2103 	best_diff = 4096;
2104 	best_end = -1;
2105 
2106 	for (epnum = 1, hw_ep = musb->endpoints + 1;
2107 			epnum < musb->nr_endpoints;
2108 			epnum++, hw_ep++) {
2109 		int	diff;
2110 
2111 		if (musb_ep_get_qh(hw_ep, is_in) != NULL)
2112 			continue;
2113 
2114 		if (hw_ep == musb->bulk_ep)
2115 			continue;
2116 
2117 		if (is_in)
2118 			diff = hw_ep->max_packet_sz_rx;
2119 		else
2120 			diff = hw_ep->max_packet_sz_tx;
2121 		diff -= (qh->maxpacket * qh->hb_mult);
2122 
2123 		if (diff >= 0 && best_diff > diff) {
2124 
2125 			/*
2126 			 * Mentor controller has a bug in that if we schedule
2127 			 * a BULK Tx transfer on an endpoint that had earlier
2128 			 * handled ISOC then the BULK transfer has to start on
2129 			 * a zero toggle.  If the BULK transfer starts on a 1
2130 			 * toggle then this transfer will fail as the mentor
2131 			 * controller starts the Bulk transfer on a 0 toggle
2132 			 * irrespective of the programming of the toggle bits
2133 			 * in the TXCSR register.  Check for this condition
2134 			 * while allocating the EP for a Tx Bulk transfer.  If
2135 			 * so skip this EP.
2136 			 */
2137 			hw_ep = musb->endpoints + epnum;
2138 			toggle = usb_gettoggle(urb->dev, qh->epnum, !is_in);
2139 			txtype = (musb_readb(hw_ep->regs, MUSB_TXTYPE)
2140 					>> 4) & 0x3;
2141 			if (!is_in && (qh->type == USB_ENDPOINT_XFER_BULK) &&
2142 				toggle && (txtype == USB_ENDPOINT_XFER_ISOC))
2143 				continue;
2144 
2145 			best_diff = diff;
2146 			best_end = epnum;
2147 		}
2148 	}
2149 	/* use bulk reserved ep1 if no other ep is free */
2150 	if (best_end < 0 && qh->type == USB_ENDPOINT_XFER_BULK) {
2151 		hw_ep = musb->bulk_ep;
2152 		if (is_in)
2153 			head = &musb->in_bulk;
2154 		else
2155 			head = &musb->out_bulk;
2156 
2157 		/* Enable bulk RX/TX NAK timeout scheme when bulk requests are
2158 		 * multiplexed. This scheme does not work in high speed to full
2159 		 * speed scenario as NAK interrupts are not coming from a
2160 		 * full speed device connected to a high speed device.
2161 		 * NAK timeout interval is 8 (128 uframe or 16ms) for HS and
2162 		 * 4 (8 frame or 8ms) for FS device.
2163 		 */
2164 		if (qh->dev)
2165 			qh->intv_reg =
2166 				(USB_SPEED_HIGH == qh->dev->speed) ? 8 : 4;
2167 		goto success;
2168 	} else if (best_end < 0) {
2169 		return -ENOSPC;
2170 	}
2171 
2172 	idle = 1;
2173 	qh->mux = 0;
2174 	hw_ep = musb->endpoints + best_end;
2175 	dev_dbg(musb->controller, "qh %p periodic slot %d\n", qh, best_end);
2176 success:
2177 	if (head) {
2178 		idle = list_empty(head);
2179 		list_add_tail(&qh->ring, head);
2180 		qh->mux = 1;
2181 	}
2182 	qh->hw_ep = hw_ep;
2183 	qh->hep->hcpriv = qh;
2184 	if (idle)
2185 		musb_start_urb(musb, is_in, qh);
2186 	return 0;
2187 }
2188 
2189 static int musb_urb_enqueue(
2190 	struct usb_hcd			*hcd,
2191 	struct urb			*urb,
2192 	gfp_t				mem_flags)
2193 {
2194 	unsigned long			flags;
2195 	struct musb			*musb = hcd_to_musb(hcd);
2196 	struct usb_host_endpoint	*hep = urb->ep;
2197 	struct musb_qh			*qh;
2198 	struct usb_endpoint_descriptor	*epd = &hep->desc;
2199 	int				ret;
2200 	unsigned			type_reg;
2201 	unsigned			interval;
2202 
2203 	/* host role must be active */
2204 	if (!is_host_active(musb) || !musb->is_active)
2205 		return -ENODEV;
2206 
2207 	spin_lock_irqsave(&musb->lock, flags);
2208 	ret = usb_hcd_link_urb_to_ep(hcd, urb);
2209 	qh = ret ? NULL : hep->hcpriv;
2210 	if (qh)
2211 		urb->hcpriv = qh;
2212 	spin_unlock_irqrestore(&musb->lock, flags);
2213 
2214 	/* DMA mapping was already done, if needed, and this urb is on
2215 	 * hep->urb_list now ... so we're done, unless hep wasn't yet
2216 	 * scheduled onto a live qh.
2217 	 *
2218 	 * REVISIT best to keep hep->hcpriv valid until the endpoint gets
2219 	 * disabled, testing for empty qh->ring and avoiding qh setup costs
2220 	 * except for the first urb queued after a config change.
2221 	 */
2222 	if (qh || ret)
2223 		return ret;
2224 
2225 	/* Allocate and initialize qh, minimizing the work done each time
2226 	 * hw_ep gets reprogrammed, or with irqs blocked.  Then schedule it.
2227 	 *
2228 	 * REVISIT consider a dedicated qh kmem_cache, so it's harder
2229 	 * for bugs in other kernel code to break this driver...
2230 	 */
2231 	qh = kzalloc(sizeof *qh, mem_flags);
2232 	if (!qh) {
2233 		spin_lock_irqsave(&musb->lock, flags);
2234 		usb_hcd_unlink_urb_from_ep(hcd, urb);
2235 		spin_unlock_irqrestore(&musb->lock, flags);
2236 		return -ENOMEM;
2237 	}
2238 
2239 	qh->hep = hep;
2240 	qh->dev = urb->dev;
2241 	INIT_LIST_HEAD(&qh->ring);
2242 	qh->is_ready = 1;
2243 
2244 	qh->maxpacket = usb_endpoint_maxp(epd);
2245 	qh->type = usb_endpoint_type(epd);
2246 
2247 	/* Bits 11 & 12 of wMaxPacketSize encode high bandwidth multiplier.
2248 	 * Some musb cores don't support high bandwidth ISO transfers; and
2249 	 * we don't (yet!) support high bandwidth interrupt transfers.
2250 	 */
2251 	qh->hb_mult = 1 + ((qh->maxpacket >> 11) & 0x03);
2252 	if (qh->hb_mult > 1) {
2253 		int ok = (qh->type == USB_ENDPOINT_XFER_ISOC);
2254 
2255 		if (ok)
2256 			ok = (usb_pipein(urb->pipe) && musb->hb_iso_rx)
2257 				|| (usb_pipeout(urb->pipe) && musb->hb_iso_tx);
2258 		if (!ok) {
2259 			ret = -EMSGSIZE;
2260 			goto done;
2261 		}
2262 		qh->maxpacket &= 0x7ff;
2263 	}
2264 
2265 	qh->epnum = usb_endpoint_num(epd);
2266 
2267 	/* NOTE: urb->dev->devnum is wrong during SET_ADDRESS */
2268 	qh->addr_reg = (u8) usb_pipedevice(urb->pipe);
2269 
2270 	/* precompute rxtype/txtype/type0 register */
2271 	type_reg = (qh->type << 4) | qh->epnum;
2272 	switch (urb->dev->speed) {
2273 	case USB_SPEED_LOW:
2274 		type_reg |= 0xc0;
2275 		break;
2276 	case USB_SPEED_FULL:
2277 		type_reg |= 0x80;
2278 		break;
2279 	default:
2280 		type_reg |= 0x40;
2281 	}
2282 	qh->type_reg = type_reg;
2283 
2284 	/* Precompute RXINTERVAL/TXINTERVAL register */
2285 	switch (qh->type) {
2286 	case USB_ENDPOINT_XFER_INT:
2287 		/*
2288 		 * Full/low speeds use the  linear encoding,
2289 		 * high speed uses the logarithmic encoding.
2290 		 */
2291 		if (urb->dev->speed <= USB_SPEED_FULL) {
2292 			interval = max_t(u8, epd->bInterval, 1);
2293 			break;
2294 		}
2295 		/* FALLTHROUGH */
2296 	case USB_ENDPOINT_XFER_ISOC:
2297 		/* ISO always uses logarithmic encoding */
2298 		interval = min_t(u8, epd->bInterval, 16);
2299 		break;
2300 	default:
2301 		/* REVISIT we actually want to use NAK limits, hinting to the
2302 		 * transfer scheduling logic to try some other qh, e.g. try
2303 		 * for 2 msec first:
2304 		 *
2305 		 * interval = (USB_SPEED_HIGH == urb->dev->speed) ? 16 : 2;
2306 		 *
2307 		 * The downside of disabling this is that transfer scheduling
2308 		 * gets VERY unfair for nonperiodic transfers; a misbehaving
2309 		 * peripheral could make that hurt.  That's perfectly normal
2310 		 * for reads from network or serial adapters ... so we have
2311 		 * partial NAKlimit support for bulk RX.
2312 		 *
2313 		 * The upside of disabling it is simpler transfer scheduling.
2314 		 */
2315 		interval = 0;
2316 	}
2317 	qh->intv_reg = interval;
2318 
2319 	/* precompute addressing for external hub/tt ports */
2320 	if (musb->is_multipoint) {
2321 		struct usb_device	*parent = urb->dev->parent;
2322 
2323 		if (parent != hcd->self.root_hub) {
2324 			qh->h_addr_reg = (u8) parent->devnum;
2325 
2326 			/* set up tt info if needed */
2327 			if (urb->dev->tt) {
2328 				qh->h_port_reg = (u8) urb->dev->ttport;
2329 				if (urb->dev->tt->hub)
2330 					qh->h_addr_reg =
2331 						(u8) urb->dev->tt->hub->devnum;
2332 				if (urb->dev->tt->multi)
2333 					qh->h_addr_reg |= 0x80;
2334 			}
2335 		}
2336 	}
2337 
2338 	/* invariant: hep->hcpriv is null OR the qh that's already scheduled.
2339 	 * until we get real dma queues (with an entry for each urb/buffer),
2340 	 * we only have work to do in the former case.
2341 	 */
2342 	spin_lock_irqsave(&musb->lock, flags);
2343 	if (hep->hcpriv || !next_urb(qh)) {
2344 		/* some concurrent activity submitted another urb to hep...
2345 		 * odd, rare, error prone, but legal.
2346 		 */
2347 		kfree(qh);
2348 		qh = NULL;
2349 		ret = 0;
2350 	} else
2351 		ret = musb_schedule(musb, qh,
2352 				epd->bEndpointAddress & USB_ENDPOINT_DIR_MASK);
2353 
2354 	if (ret == 0) {
2355 		urb->hcpriv = qh;
2356 		/* FIXME set urb->start_frame for iso/intr, it's tested in
2357 		 * musb_start_urb(), but otherwise only konicawc cares ...
2358 		 */
2359 	}
2360 	spin_unlock_irqrestore(&musb->lock, flags);
2361 
2362 done:
2363 	if (ret != 0) {
2364 		spin_lock_irqsave(&musb->lock, flags);
2365 		usb_hcd_unlink_urb_from_ep(hcd, urb);
2366 		spin_unlock_irqrestore(&musb->lock, flags);
2367 		kfree(qh);
2368 	}
2369 	return ret;
2370 }
2371 
2372 
2373 /*
2374  * abort a transfer that's at the head of a hardware queue.
2375  * called with controller locked, irqs blocked
2376  * that hardware queue advances to the next transfer, unless prevented
2377  */
2378 static int musb_cleanup_urb(struct urb *urb, struct musb_qh *qh)
2379 {
2380 	struct musb_hw_ep	*ep = qh->hw_ep;
2381 	struct musb		*musb = ep->musb;
2382 	void __iomem		*epio = ep->regs;
2383 	unsigned		hw_end = ep->epnum;
2384 	void __iomem		*regs = ep->musb->mregs;
2385 	int			is_in = usb_pipein(urb->pipe);
2386 	int			status = 0;
2387 	u16			csr;
2388 
2389 	musb_ep_select(regs, hw_end);
2390 
2391 	if (is_dma_capable()) {
2392 		struct dma_channel	*dma;
2393 
2394 		dma = is_in ? ep->rx_channel : ep->tx_channel;
2395 		if (dma) {
2396 			status = ep->musb->dma_controller->channel_abort(dma);
2397 			dev_dbg(musb->controller,
2398 				"abort %cX%d DMA for urb %p --> %d\n",
2399 				is_in ? 'R' : 'T', ep->epnum,
2400 				urb, status);
2401 			urb->actual_length += dma->actual_len;
2402 		}
2403 	}
2404 
2405 	/* turn off DMA requests, discard state, stop polling ... */
2406 	if (ep->epnum && is_in) {
2407 		/* giveback saves bulk toggle */
2408 		csr = musb_h_flush_rxfifo(ep, 0);
2409 
2410 		/* REVISIT we still get an irq; should likely clear the
2411 		 * endpoint's irq status here to avoid bogus irqs.
2412 		 * clearing that status is platform-specific...
2413 		 */
2414 	} else if (ep->epnum) {
2415 		musb_h_tx_flush_fifo(ep);
2416 		csr = musb_readw(epio, MUSB_TXCSR);
2417 		csr &= ~(MUSB_TXCSR_AUTOSET
2418 			| MUSB_TXCSR_DMAENAB
2419 			| MUSB_TXCSR_H_RXSTALL
2420 			| MUSB_TXCSR_H_NAKTIMEOUT
2421 			| MUSB_TXCSR_H_ERROR
2422 			| MUSB_TXCSR_TXPKTRDY);
2423 		musb_writew(epio, MUSB_TXCSR, csr);
2424 		/* REVISIT may need to clear FLUSHFIFO ... */
2425 		musb_writew(epio, MUSB_TXCSR, csr);
2426 		/* flush cpu writebuffer */
2427 		csr = musb_readw(epio, MUSB_TXCSR);
2428 	} else  {
2429 		musb_h_ep0_flush_fifo(ep);
2430 	}
2431 	if (status == 0)
2432 		musb_advance_schedule(ep->musb, urb, ep, is_in);
2433 	return status;
2434 }
2435 
2436 static int musb_urb_dequeue(struct usb_hcd *hcd, struct urb *urb, int status)
2437 {
2438 	struct musb		*musb = hcd_to_musb(hcd);
2439 	struct musb_qh		*qh;
2440 	unsigned long		flags;
2441 	int			is_in  = usb_pipein(urb->pipe);
2442 	int			ret;
2443 
2444 	dev_dbg(musb->controller, "urb=%p, dev%d ep%d%s\n", urb,
2445 			usb_pipedevice(urb->pipe),
2446 			usb_pipeendpoint(urb->pipe),
2447 			is_in ? "in" : "out");
2448 
2449 	spin_lock_irqsave(&musb->lock, flags);
2450 	ret = usb_hcd_check_unlink_urb(hcd, urb, status);
2451 	if (ret)
2452 		goto done;
2453 
2454 	qh = urb->hcpriv;
2455 	if (!qh)
2456 		goto done;
2457 
2458 	/*
2459 	 * Any URB not actively programmed into endpoint hardware can be
2460 	 * immediately given back; that's any URB not at the head of an
2461 	 * endpoint queue, unless someday we get real DMA queues.  And even
2462 	 * if it's at the head, it might not be known to the hardware...
2463 	 *
2464 	 * Otherwise abort current transfer, pending DMA, etc.; urb->status
2465 	 * has already been updated.  This is a synchronous abort; it'd be
2466 	 * OK to hold off until after some IRQ, though.
2467 	 *
2468 	 * NOTE: qh is invalid unless !list_empty(&hep->urb_list)
2469 	 */
2470 	if (!qh->is_ready
2471 			|| urb->urb_list.prev != &qh->hep->urb_list
2472 			|| musb_ep_get_qh(qh->hw_ep, is_in) != qh) {
2473 		int	ready = qh->is_ready;
2474 
2475 		qh->is_ready = 0;
2476 		musb_giveback(musb, urb, 0);
2477 		qh->is_ready = ready;
2478 
2479 		/* If nothing else (usually musb_giveback) is using it
2480 		 * and its URB list has emptied, recycle this qh.
2481 		 */
2482 		if (ready && list_empty(&qh->hep->urb_list)) {
2483 			qh->hep->hcpriv = NULL;
2484 			list_del(&qh->ring);
2485 			kfree(qh);
2486 		}
2487 	} else
2488 		ret = musb_cleanup_urb(urb, qh);
2489 done:
2490 	spin_unlock_irqrestore(&musb->lock, flags);
2491 	return ret;
2492 }
2493 
2494 /* disable an endpoint */
2495 static void
2496 musb_h_disable(struct usb_hcd *hcd, struct usb_host_endpoint *hep)
2497 {
2498 	u8			is_in = hep->desc.bEndpointAddress & USB_DIR_IN;
2499 	unsigned long		flags;
2500 	struct musb		*musb = hcd_to_musb(hcd);
2501 	struct musb_qh		*qh;
2502 	struct urb		*urb;
2503 
2504 	spin_lock_irqsave(&musb->lock, flags);
2505 
2506 	qh = hep->hcpriv;
2507 	if (qh == NULL)
2508 		goto exit;
2509 
2510 	/* NOTE: qh is invalid unless !list_empty(&hep->urb_list) */
2511 
2512 	/* Kick the first URB off the hardware, if needed */
2513 	qh->is_ready = 0;
2514 	if (musb_ep_get_qh(qh->hw_ep, is_in) == qh) {
2515 		urb = next_urb(qh);
2516 
2517 		/* make software (then hardware) stop ASAP */
2518 		if (!urb->unlinked)
2519 			urb->status = -ESHUTDOWN;
2520 
2521 		/* cleanup */
2522 		musb_cleanup_urb(urb, qh);
2523 
2524 		/* Then nuke all the others ... and advance the
2525 		 * queue on hw_ep (e.g. bulk ring) when we're done.
2526 		 */
2527 		while (!list_empty(&hep->urb_list)) {
2528 			urb = next_urb(qh);
2529 			urb->status = -ESHUTDOWN;
2530 			musb_advance_schedule(musb, urb, qh->hw_ep, is_in);
2531 		}
2532 	} else {
2533 		/* Just empty the queue; the hardware is busy with
2534 		 * other transfers, and since !qh->is_ready nothing
2535 		 * will activate any of these as it advances.
2536 		 */
2537 		while (!list_empty(&hep->urb_list))
2538 			musb_giveback(musb, next_urb(qh), -ESHUTDOWN);
2539 
2540 		hep->hcpriv = NULL;
2541 		list_del(&qh->ring);
2542 		kfree(qh);
2543 	}
2544 exit:
2545 	spin_unlock_irqrestore(&musb->lock, flags);
2546 }
2547 
2548 static int musb_h_get_frame_number(struct usb_hcd *hcd)
2549 {
2550 	struct musb	*musb = hcd_to_musb(hcd);
2551 
2552 	return musb_readw(musb->mregs, MUSB_FRAME);
2553 }
2554 
2555 static int musb_h_start(struct usb_hcd *hcd)
2556 {
2557 	struct musb	*musb = hcd_to_musb(hcd);
2558 
2559 	/* NOTE: musb_start() is called when the hub driver turns
2560 	 * on port power, or when (OTG) peripheral starts.
2561 	 */
2562 	hcd->state = HC_STATE_RUNNING;
2563 	musb->port1_status = 0;
2564 	return 0;
2565 }
2566 
2567 static void musb_h_stop(struct usb_hcd *hcd)
2568 {
2569 	musb_stop(hcd_to_musb(hcd));
2570 	hcd->state = HC_STATE_HALT;
2571 }
2572 
2573 static int musb_bus_suspend(struct usb_hcd *hcd)
2574 {
2575 	struct musb	*musb = hcd_to_musb(hcd);
2576 	u8		devctl;
2577 
2578 	musb_port_suspend(musb, true);
2579 
2580 	if (!is_host_active(musb))
2581 		return 0;
2582 
2583 	switch (musb->xceiv->otg->state) {
2584 	case OTG_STATE_A_SUSPEND:
2585 		return 0;
2586 	case OTG_STATE_A_WAIT_VRISE:
2587 		/* ID could be grounded even if there's no device
2588 		 * on the other end of the cable.  NOTE that the
2589 		 * A_WAIT_VRISE timers are messy with MUSB...
2590 		 */
2591 		devctl = musb_readb(musb->mregs, MUSB_DEVCTL);
2592 		if ((devctl & MUSB_DEVCTL_VBUS) == MUSB_DEVCTL_VBUS)
2593 			musb->xceiv->otg->state = OTG_STATE_A_WAIT_BCON;
2594 		break;
2595 	default:
2596 		break;
2597 	}
2598 
2599 	if (musb->is_active) {
2600 		WARNING("trying to suspend as %s while active\n",
2601 				usb_otg_state_string(musb->xceiv->otg->state));
2602 		return -EBUSY;
2603 	} else
2604 		return 0;
2605 }
2606 
2607 static int musb_bus_resume(struct usb_hcd *hcd)
2608 {
2609 	struct musb *musb = hcd_to_musb(hcd);
2610 
2611 	if (musb->config &&
2612 	    musb->config->host_port_deassert_reset_at_resume)
2613 		musb_port_reset(musb, false);
2614 
2615 	return 0;
2616 }
2617 
2618 #ifndef CONFIG_MUSB_PIO_ONLY
2619 
2620 #define MUSB_USB_DMA_ALIGN 4
2621 
2622 struct musb_temp_buffer {
2623 	void *kmalloc_ptr;
2624 	void *old_xfer_buffer;
2625 	u8 data[0];
2626 };
2627 
2628 static void musb_free_temp_buffer(struct urb *urb)
2629 {
2630 	enum dma_data_direction dir;
2631 	struct musb_temp_buffer *temp;
2632 	size_t length;
2633 
2634 	if (!(urb->transfer_flags & URB_ALIGNED_TEMP_BUFFER))
2635 		return;
2636 
2637 	dir = usb_urb_dir_in(urb) ? DMA_FROM_DEVICE : DMA_TO_DEVICE;
2638 
2639 	temp = container_of(urb->transfer_buffer, struct musb_temp_buffer,
2640 			    data);
2641 
2642 	if (dir == DMA_FROM_DEVICE) {
2643 		if (usb_pipeisoc(urb->pipe))
2644 			length = urb->transfer_buffer_length;
2645 		else
2646 			length = urb->actual_length;
2647 
2648 		memcpy(temp->old_xfer_buffer, temp->data, length);
2649 	}
2650 	urb->transfer_buffer = temp->old_xfer_buffer;
2651 	kfree(temp->kmalloc_ptr);
2652 
2653 	urb->transfer_flags &= ~URB_ALIGNED_TEMP_BUFFER;
2654 }
2655 
2656 static int musb_alloc_temp_buffer(struct urb *urb, gfp_t mem_flags)
2657 {
2658 	enum dma_data_direction dir;
2659 	struct musb_temp_buffer *temp;
2660 	void *kmalloc_ptr;
2661 	size_t kmalloc_size;
2662 
2663 	if (urb->num_sgs || urb->sg ||
2664 	    urb->transfer_buffer_length == 0 ||
2665 	    !((uintptr_t)urb->transfer_buffer & (MUSB_USB_DMA_ALIGN - 1)))
2666 		return 0;
2667 
2668 	dir = usb_urb_dir_in(urb) ? DMA_FROM_DEVICE : DMA_TO_DEVICE;
2669 
2670 	/* Allocate a buffer with enough padding for alignment */
2671 	kmalloc_size = urb->transfer_buffer_length +
2672 		sizeof(struct musb_temp_buffer) + MUSB_USB_DMA_ALIGN - 1;
2673 
2674 	kmalloc_ptr = kmalloc(kmalloc_size, mem_flags);
2675 	if (!kmalloc_ptr)
2676 		return -ENOMEM;
2677 
2678 	/* Position our struct temp_buffer such that data is aligned */
2679 	temp = PTR_ALIGN(kmalloc_ptr, MUSB_USB_DMA_ALIGN);
2680 
2681 
2682 	temp->kmalloc_ptr = kmalloc_ptr;
2683 	temp->old_xfer_buffer = urb->transfer_buffer;
2684 	if (dir == DMA_TO_DEVICE)
2685 		memcpy(temp->data, urb->transfer_buffer,
2686 		       urb->transfer_buffer_length);
2687 	urb->transfer_buffer = temp->data;
2688 
2689 	urb->transfer_flags |= URB_ALIGNED_TEMP_BUFFER;
2690 
2691 	return 0;
2692 }
2693 
2694 static int musb_map_urb_for_dma(struct usb_hcd *hcd, struct urb *urb,
2695 				      gfp_t mem_flags)
2696 {
2697 	struct musb	*musb = hcd_to_musb(hcd);
2698 	int ret;
2699 
2700 	/*
2701 	 * The DMA engine in RTL1.8 and above cannot handle
2702 	 * DMA addresses that are not aligned to a 4 byte boundary.
2703 	 * For such engine implemented (un)map_urb_for_dma hooks.
2704 	 * Do not use these hooks for RTL<1.8
2705 	 */
2706 	if (musb->hwvers < MUSB_HWVERS_1800)
2707 		return usb_hcd_map_urb_for_dma(hcd, urb, mem_flags);
2708 
2709 	ret = musb_alloc_temp_buffer(urb, mem_flags);
2710 	if (ret)
2711 		return ret;
2712 
2713 	ret = usb_hcd_map_urb_for_dma(hcd, urb, mem_flags);
2714 	if (ret)
2715 		musb_free_temp_buffer(urb);
2716 
2717 	return ret;
2718 }
2719 
2720 static void musb_unmap_urb_for_dma(struct usb_hcd *hcd, struct urb *urb)
2721 {
2722 	struct musb	*musb = hcd_to_musb(hcd);
2723 
2724 	usb_hcd_unmap_urb_for_dma(hcd, urb);
2725 
2726 	/* Do not use this hook for RTL<1.8 (see description above) */
2727 	if (musb->hwvers < MUSB_HWVERS_1800)
2728 		return;
2729 
2730 	musb_free_temp_buffer(urb);
2731 }
2732 #endif /* !CONFIG_MUSB_PIO_ONLY */
2733 
2734 static const struct hc_driver musb_hc_driver = {
2735 	.description		= "musb-hcd",
2736 	.product_desc		= "MUSB HDRC host driver",
2737 	.hcd_priv_size		= sizeof(struct musb *),
2738 	.flags			= HCD_USB2 | HCD_MEMORY,
2739 
2740 	/* not using irq handler or reset hooks from usbcore, since
2741 	 * those must be shared with peripheral code for OTG configs
2742 	 */
2743 
2744 	.start			= musb_h_start,
2745 	.stop			= musb_h_stop,
2746 
2747 	.get_frame_number	= musb_h_get_frame_number,
2748 
2749 	.urb_enqueue		= musb_urb_enqueue,
2750 	.urb_dequeue		= musb_urb_dequeue,
2751 	.endpoint_disable	= musb_h_disable,
2752 
2753 #ifndef CONFIG_MUSB_PIO_ONLY
2754 	.map_urb_for_dma	= musb_map_urb_for_dma,
2755 	.unmap_urb_for_dma	= musb_unmap_urb_for_dma,
2756 #endif
2757 
2758 	.hub_status_data	= musb_hub_status_data,
2759 	.hub_control		= musb_hub_control,
2760 	.bus_suspend		= musb_bus_suspend,
2761 	.bus_resume		= musb_bus_resume,
2762 	/* .start_port_reset	= NULL, */
2763 	/* .hub_irq_enable	= NULL, */
2764 };
2765 
2766 int musb_host_alloc(struct musb *musb)
2767 {
2768 	struct device	*dev = musb->controller;
2769 
2770 	/* usbcore sets dev->driver_data to hcd, and sometimes uses that... */
2771 	musb->hcd = usb_create_hcd(&musb_hc_driver, dev, dev_name(dev));
2772 	if (!musb->hcd)
2773 		return -EINVAL;
2774 
2775 	*musb->hcd->hcd_priv = (unsigned long) musb;
2776 	musb->hcd->self.uses_pio_for_control = 1;
2777 	musb->hcd->uses_new_polling = 1;
2778 	musb->hcd->has_tt = 1;
2779 
2780 	return 0;
2781 }
2782 
2783 void musb_host_cleanup(struct musb *musb)
2784 {
2785 	if (musb->port_mode == MUSB_PORT_MODE_GADGET)
2786 		return;
2787 	usb_remove_hcd(musb->hcd);
2788 }
2789 
2790 void musb_host_free(struct musb *musb)
2791 {
2792 	usb_put_hcd(musb->hcd);
2793 }
2794 
2795 int musb_host_setup(struct musb *musb, int power_budget)
2796 {
2797 	int ret;
2798 	struct usb_hcd *hcd = musb->hcd;
2799 
2800 	MUSB_HST_MODE(musb);
2801 	musb->xceiv->otg->default_a = 1;
2802 	musb->xceiv->otg->state = OTG_STATE_A_IDLE;
2803 
2804 	otg_set_host(musb->xceiv->otg, &hcd->self);
2805 	hcd->self.otg_port = 1;
2806 	musb->xceiv->otg->host = &hcd->self;
2807 	hcd->power_budget = 2 * (power_budget ? : 250);
2808 
2809 	ret = usb_add_hcd(hcd, 0, 0);
2810 	if (ret < 0)
2811 		return ret;
2812 
2813 	device_wakeup_enable(hcd->self.controller);
2814 	return 0;
2815 }
2816 
2817 void musb_host_resume_root_hub(struct musb *musb)
2818 {
2819 	usb_hcd_resume_root_hub(musb->hcd);
2820 }
2821 
2822 void musb_host_poke_root_hub(struct musb *musb)
2823 {
2824 	MUSB_HST_MODE(musb);
2825 	if (musb->hcd->status_urb)
2826 		usb_hcd_poll_rh_status(musb->hcd);
2827 	else
2828 		usb_hcd_resume_root_hub(musb->hcd);
2829 }
2830