xref: /openbmc/linux/drivers/usb/musb/musb_host.c (revision be709d48)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * MUSB OTG driver host support
4  *
5  * Copyright 2005 Mentor Graphics Corporation
6  * Copyright (C) 2005-2006 by Texas Instruments
7  * Copyright (C) 2006-2007 Nokia Corporation
8  * Copyright (C) 2008-2009 MontaVista Software, Inc. <source@mvista.com>
9  */
10 
11 #include <linux/module.h>
12 #include <linux/kernel.h>
13 #include <linux/delay.h>
14 #include <linux/sched.h>
15 #include <linux/slab.h>
16 #include <linux/errno.h>
17 #include <linux/list.h>
18 #include <linux/dma-mapping.h>
19 
20 #include "musb_core.h"
21 #include "musb_host.h"
22 #include "musb_trace.h"
23 
24 /* MUSB HOST status 22-mar-2006
25  *
26  * - There's still lots of partial code duplication for fault paths, so
27  *   they aren't handled as consistently as they need to be.
28  *
29  * - PIO mostly behaved when last tested.
30  *     + including ep0, with all usbtest cases 9, 10
31  *     + usbtest 14 (ep0out) doesn't seem to run at all
32  *     + double buffered OUT/TX endpoints saw stalls(!) with certain usbtest
33  *       configurations, but otherwise double buffering passes basic tests.
34  *     + for 2.6.N, for N > ~10, needs API changes for hcd framework.
35  *
36  * - DMA (CPPI) ... partially behaves, not currently recommended
37  *     + about 1/15 the speed of typical EHCI implementations (PCI)
38  *     + RX, all too often reqpkt seems to misbehave after tx
39  *     + TX, no known issues (other than evident silicon issue)
40  *
41  * - DMA (Mentor/OMAP) ...has at least toggle update problems
42  *
43  * - [23-feb-2009] minimal traffic scheduling to avoid bulk RX packet
44  *   starvation ... nothing yet for TX, interrupt, or bulk.
45  *
46  * - Not tested with HNP, but some SRP paths seem to behave.
47  *
48  * NOTE 24-August-2006:
49  *
50  * - Bulk traffic finally uses both sides of hardware ep1, freeing up an
51  *   extra endpoint for periodic use enabling hub + keybd + mouse.  That
52  *   mostly works, except that with "usbnet" it's easy to trigger cases
53  *   with "ping" where RX loses.  (a) ping to davinci, even "ping -f",
54  *   fine; but (b) ping _from_ davinci, even "ping -c 1", ICMP RX loses
55  *   although ARP RX wins.  (That test was done with a full speed link.)
56  */
57 
58 
59 /*
60  * NOTE on endpoint usage:
61  *
62  * CONTROL transfers all go through ep0.  BULK ones go through dedicated IN
63  * and OUT endpoints ... hardware is dedicated for those "async" queue(s).
64  * (Yes, bulk _could_ use more of the endpoints than that, and would even
65  * benefit from it.)
66  *
67  * INTERUPPT and ISOCHRONOUS transfers are scheduled to the other endpoints.
68  * So far that scheduling is both dumb and optimistic:  the endpoint will be
69  * "claimed" until its software queue is no longer refilled.  No multiplexing
70  * of transfers between endpoints, or anything clever.
71  */
72 
73 struct musb *hcd_to_musb(struct usb_hcd *hcd)
74 {
75 	return *(struct musb **) hcd->hcd_priv;
76 }
77 
78 
79 static void musb_ep_program(struct musb *musb, u8 epnum,
80 			struct urb *urb, int is_out,
81 			u8 *buf, u32 offset, u32 len);
82 
83 /*
84  * Clear TX fifo. Needed to avoid BABBLE errors.
85  */
86 static void musb_h_tx_flush_fifo(struct musb_hw_ep *ep)
87 {
88 	struct musb	*musb = ep->musb;
89 	void __iomem	*epio = ep->regs;
90 	u16		csr;
91 	int		retries = 1000;
92 
93 	csr = musb_readw(epio, MUSB_TXCSR);
94 	while (csr & MUSB_TXCSR_FIFONOTEMPTY) {
95 		csr |= MUSB_TXCSR_FLUSHFIFO | MUSB_TXCSR_TXPKTRDY;
96 		musb_writew(epio, MUSB_TXCSR, csr);
97 		csr = musb_readw(epio, MUSB_TXCSR);
98 
99 		/*
100 		 * FIXME: sometimes the tx fifo flush failed, it has been
101 		 * observed during device disconnect on AM335x.
102 		 *
103 		 * To reproduce the issue, ensure tx urb(s) are queued when
104 		 * unplug the usb device which is connected to AM335x usb
105 		 * host port.
106 		 *
107 		 * I found using a usb-ethernet device and running iperf
108 		 * (client on AM335x) has very high chance to trigger it.
109 		 *
110 		 * Better to turn on musb_dbg() in musb_cleanup_urb() with
111 		 * CPPI enabled to see the issue when aborting the tx channel.
112 		 */
113 		if (dev_WARN_ONCE(musb->controller, retries-- < 1,
114 				"Could not flush host TX%d fifo: csr: %04x\n",
115 				ep->epnum, csr))
116 			return;
117 		mdelay(1);
118 	}
119 }
120 
121 static void musb_h_ep0_flush_fifo(struct musb_hw_ep *ep)
122 {
123 	void __iomem	*epio = ep->regs;
124 	u16		csr;
125 	int		retries = 5;
126 
127 	/* scrub any data left in the fifo */
128 	do {
129 		csr = musb_readw(epio, MUSB_TXCSR);
130 		if (!(csr & (MUSB_CSR0_TXPKTRDY | MUSB_CSR0_RXPKTRDY)))
131 			break;
132 		musb_writew(epio, MUSB_TXCSR, MUSB_CSR0_FLUSHFIFO);
133 		csr = musb_readw(epio, MUSB_TXCSR);
134 		udelay(10);
135 	} while (--retries);
136 
137 	WARN(!retries, "Could not flush host TX%d fifo: csr: %04x\n",
138 			ep->epnum, csr);
139 
140 	/* and reset for the next transfer */
141 	musb_writew(epio, MUSB_TXCSR, 0);
142 }
143 
144 /*
145  * Start transmit. Caller is responsible for locking shared resources.
146  * musb must be locked.
147  */
148 static inline void musb_h_tx_start(struct musb_hw_ep *ep)
149 {
150 	u16	txcsr;
151 
152 	/* NOTE: no locks here; caller should lock and select EP */
153 	if (ep->epnum) {
154 		txcsr = musb_readw(ep->regs, MUSB_TXCSR);
155 		txcsr |= MUSB_TXCSR_TXPKTRDY | MUSB_TXCSR_H_WZC_BITS;
156 		musb_writew(ep->regs, MUSB_TXCSR, txcsr);
157 	} else {
158 		txcsr = MUSB_CSR0_H_SETUPPKT | MUSB_CSR0_TXPKTRDY;
159 		musb_writew(ep->regs, MUSB_CSR0, txcsr);
160 	}
161 
162 }
163 
164 static inline void musb_h_tx_dma_start(struct musb_hw_ep *ep)
165 {
166 	u16	txcsr;
167 
168 	/* NOTE: no locks here; caller should lock and select EP */
169 	txcsr = musb_readw(ep->regs, MUSB_TXCSR);
170 	txcsr |= MUSB_TXCSR_DMAENAB | MUSB_TXCSR_H_WZC_BITS;
171 	if (is_cppi_enabled(ep->musb))
172 		txcsr |= MUSB_TXCSR_DMAMODE;
173 	musb_writew(ep->regs, MUSB_TXCSR, txcsr);
174 }
175 
176 static void musb_ep_set_qh(struct musb_hw_ep *ep, int is_in, struct musb_qh *qh)
177 {
178 	if (is_in != 0 || ep->is_shared_fifo)
179 		ep->in_qh  = qh;
180 	if (is_in == 0 || ep->is_shared_fifo)
181 		ep->out_qh = qh;
182 }
183 
184 static struct musb_qh *musb_ep_get_qh(struct musb_hw_ep *ep, int is_in)
185 {
186 	return is_in ? ep->in_qh : ep->out_qh;
187 }
188 
189 /*
190  * Start the URB at the front of an endpoint's queue
191  * end must be claimed from the caller.
192  *
193  * Context: controller locked, irqs blocked
194  */
195 static void
196 musb_start_urb(struct musb *musb, int is_in, struct musb_qh *qh)
197 {
198 	u32			len;
199 	void __iomem		*mbase =  musb->mregs;
200 	struct urb		*urb = next_urb(qh);
201 	void			*buf = urb->transfer_buffer;
202 	u32			offset = 0;
203 	struct musb_hw_ep	*hw_ep = qh->hw_ep;
204 	int			epnum = hw_ep->epnum;
205 
206 	/* initialize software qh state */
207 	qh->offset = 0;
208 	qh->segsize = 0;
209 
210 	/* gather right source of data */
211 	switch (qh->type) {
212 	case USB_ENDPOINT_XFER_CONTROL:
213 		/* control transfers always start with SETUP */
214 		is_in = 0;
215 		musb->ep0_stage = MUSB_EP0_START;
216 		buf = urb->setup_packet;
217 		len = 8;
218 		break;
219 	case USB_ENDPOINT_XFER_ISOC:
220 		qh->iso_idx = 0;
221 		qh->frame = 0;
222 		offset = urb->iso_frame_desc[0].offset;
223 		len = urb->iso_frame_desc[0].length;
224 		break;
225 	default:		/* bulk, interrupt */
226 		/* actual_length may be nonzero on retry paths */
227 		buf = urb->transfer_buffer + urb->actual_length;
228 		len = urb->transfer_buffer_length - urb->actual_length;
229 	}
230 
231 	trace_musb_urb_start(musb, urb);
232 
233 	/* Configure endpoint */
234 	musb_ep_set_qh(hw_ep, is_in, qh);
235 	musb_ep_program(musb, epnum, urb, !is_in, buf, offset, len);
236 
237 	/* transmit may have more work: start it when it is time */
238 	if (is_in)
239 		return;
240 
241 	/* determine if the time is right for a periodic transfer */
242 	switch (qh->type) {
243 	case USB_ENDPOINT_XFER_ISOC:
244 	case USB_ENDPOINT_XFER_INT:
245 		musb_dbg(musb, "check whether there's still time for periodic Tx");
246 		/* FIXME this doesn't implement that scheduling policy ...
247 		 * or handle framecounter wrapping
248 		 */
249 		if (1) {	/* Always assume URB_ISO_ASAP */
250 			/* REVISIT the SOF irq handler shouldn't duplicate
251 			 * this code; and we don't init urb->start_frame...
252 			 */
253 			qh->frame = 0;
254 			goto start;
255 		} else {
256 			qh->frame = urb->start_frame;
257 			/* enable SOF interrupt so we can count down */
258 			musb_dbg(musb, "SOF for %d", epnum);
259 #if 1 /* ifndef	CONFIG_ARCH_DAVINCI */
260 			musb_writeb(mbase, MUSB_INTRUSBE, 0xff);
261 #endif
262 		}
263 		break;
264 	default:
265 start:
266 		musb_dbg(musb, "Start TX%d %s", epnum,
267 			hw_ep->tx_channel ? "dma" : "pio");
268 
269 		if (!hw_ep->tx_channel)
270 			musb_h_tx_start(hw_ep);
271 		else if (is_cppi_enabled(musb) || tusb_dma_omap(musb))
272 			musb_h_tx_dma_start(hw_ep);
273 	}
274 }
275 
276 /* Context: caller owns controller lock, IRQs are blocked */
277 static void musb_giveback(struct musb *musb, struct urb *urb, int status)
278 __releases(musb->lock)
279 __acquires(musb->lock)
280 {
281 	trace_musb_urb_gb(musb, urb);
282 
283 	usb_hcd_unlink_urb_from_ep(musb->hcd, urb);
284 	spin_unlock(&musb->lock);
285 	usb_hcd_giveback_urb(musb->hcd, urb, status);
286 	spin_lock(&musb->lock);
287 }
288 
289 /* For bulk/interrupt endpoints only */
290 static inline void musb_save_toggle(struct musb_qh *qh, int is_in,
291 				    struct urb *urb)
292 {
293 	void __iomem		*epio = qh->hw_ep->regs;
294 	u16			csr;
295 
296 	/*
297 	 * FIXME: the current Mentor DMA code seems to have
298 	 * problems getting toggle correct.
299 	 */
300 
301 	if (is_in)
302 		csr = musb_readw(epio, MUSB_RXCSR) & MUSB_RXCSR_H_DATATOGGLE;
303 	else
304 		csr = musb_readw(epio, MUSB_TXCSR) & MUSB_TXCSR_H_DATATOGGLE;
305 
306 	usb_settoggle(urb->dev, qh->epnum, !is_in, csr ? 1 : 0);
307 }
308 
309 /*
310  * Advance this hardware endpoint's queue, completing the specified URB and
311  * advancing to either the next URB queued to that qh, or else invalidating
312  * that qh and advancing to the next qh scheduled after the current one.
313  *
314  * Context: caller owns controller lock, IRQs are blocked
315  */
316 static void musb_advance_schedule(struct musb *musb, struct urb *urb,
317 				  struct musb_hw_ep *hw_ep, int is_in)
318 {
319 	struct musb_qh		*qh = musb_ep_get_qh(hw_ep, is_in);
320 	struct musb_hw_ep	*ep = qh->hw_ep;
321 	int			ready = qh->is_ready;
322 	int			status;
323 
324 	status = (urb->status == -EINPROGRESS) ? 0 : urb->status;
325 
326 	/* save toggle eagerly, for paranoia */
327 	switch (qh->type) {
328 	case USB_ENDPOINT_XFER_BULK:
329 	case USB_ENDPOINT_XFER_INT:
330 		musb_save_toggle(qh, is_in, urb);
331 		break;
332 	case USB_ENDPOINT_XFER_ISOC:
333 		if (status == 0 && urb->error_count)
334 			status = -EXDEV;
335 		break;
336 	}
337 
338 	qh->is_ready = 0;
339 	musb_giveback(musb, urb, status);
340 	qh->is_ready = ready;
341 
342 	/* reclaim resources (and bandwidth) ASAP; deschedule it, and
343 	 * invalidate qh as soon as list_empty(&hep->urb_list)
344 	 */
345 	if (list_empty(&qh->hep->urb_list)) {
346 		struct list_head	*head;
347 		struct dma_controller	*dma = musb->dma_controller;
348 
349 		if (is_in) {
350 			ep->rx_reinit = 1;
351 			if (ep->rx_channel) {
352 				dma->channel_release(ep->rx_channel);
353 				ep->rx_channel = NULL;
354 			}
355 		} else {
356 			ep->tx_reinit = 1;
357 			if (ep->tx_channel) {
358 				dma->channel_release(ep->tx_channel);
359 				ep->tx_channel = NULL;
360 			}
361 		}
362 
363 		/* Clobber old pointers to this qh */
364 		musb_ep_set_qh(ep, is_in, NULL);
365 		qh->hep->hcpriv = NULL;
366 
367 		switch (qh->type) {
368 
369 		case USB_ENDPOINT_XFER_CONTROL:
370 		case USB_ENDPOINT_XFER_BULK:
371 			/* fifo policy for these lists, except that NAKing
372 			 * should rotate a qh to the end (for fairness).
373 			 */
374 			if (qh->mux == 1) {
375 				head = qh->ring.prev;
376 				list_del(&qh->ring);
377 				kfree(qh);
378 				qh = first_qh(head);
379 				break;
380 			}
381 			/* fall through */
382 
383 		case USB_ENDPOINT_XFER_ISOC:
384 		case USB_ENDPOINT_XFER_INT:
385 			/* this is where periodic bandwidth should be
386 			 * de-allocated if it's tracked and allocated;
387 			 * and where we'd update the schedule tree...
388 			 */
389 			kfree(qh);
390 			qh = NULL;
391 			break;
392 		}
393 	}
394 
395 	if (qh != NULL && qh->is_ready) {
396 		musb_dbg(musb, "... next ep%d %cX urb %p",
397 		    hw_ep->epnum, is_in ? 'R' : 'T', next_urb(qh));
398 		musb_start_urb(musb, is_in, qh);
399 	}
400 }
401 
402 static u16 musb_h_flush_rxfifo(struct musb_hw_ep *hw_ep, u16 csr)
403 {
404 	/* we don't want fifo to fill itself again;
405 	 * ignore dma (various models),
406 	 * leave toggle alone (may not have been saved yet)
407 	 */
408 	csr |= MUSB_RXCSR_FLUSHFIFO | MUSB_RXCSR_RXPKTRDY;
409 	csr &= ~(MUSB_RXCSR_H_REQPKT
410 		| MUSB_RXCSR_H_AUTOREQ
411 		| MUSB_RXCSR_AUTOCLEAR);
412 
413 	/* write 2x to allow double buffering */
414 	musb_writew(hw_ep->regs, MUSB_RXCSR, csr);
415 	musb_writew(hw_ep->regs, MUSB_RXCSR, csr);
416 
417 	/* flush writebuffer */
418 	return musb_readw(hw_ep->regs, MUSB_RXCSR);
419 }
420 
421 /*
422  * PIO RX for a packet (or part of it).
423  */
424 static bool
425 musb_host_packet_rx(struct musb *musb, struct urb *urb, u8 epnum, u8 iso_err)
426 {
427 	u16			rx_count;
428 	u8			*buf;
429 	u16			csr;
430 	bool			done = false;
431 	u32			length;
432 	int			do_flush = 0;
433 	struct musb_hw_ep	*hw_ep = musb->endpoints + epnum;
434 	void __iomem		*epio = hw_ep->regs;
435 	struct musb_qh		*qh = hw_ep->in_qh;
436 	int			pipe = urb->pipe;
437 	void			*buffer = urb->transfer_buffer;
438 
439 	/* musb_ep_select(mbase, epnum); */
440 	rx_count = musb_readw(epio, MUSB_RXCOUNT);
441 	musb_dbg(musb, "RX%d count %d, buffer %p len %d/%d", epnum, rx_count,
442 			urb->transfer_buffer, qh->offset,
443 			urb->transfer_buffer_length);
444 
445 	/* unload FIFO */
446 	if (usb_pipeisoc(pipe)) {
447 		int					status = 0;
448 		struct usb_iso_packet_descriptor	*d;
449 
450 		if (iso_err) {
451 			status = -EILSEQ;
452 			urb->error_count++;
453 		}
454 
455 		d = urb->iso_frame_desc + qh->iso_idx;
456 		buf = buffer + d->offset;
457 		length = d->length;
458 		if (rx_count > length) {
459 			if (status == 0) {
460 				status = -EOVERFLOW;
461 				urb->error_count++;
462 			}
463 			musb_dbg(musb, "OVERFLOW %d into %d", rx_count, length);
464 			do_flush = 1;
465 		} else
466 			length = rx_count;
467 		urb->actual_length += length;
468 		d->actual_length = length;
469 
470 		d->status = status;
471 
472 		/* see if we are done */
473 		done = (++qh->iso_idx >= urb->number_of_packets);
474 	} else {
475 		/* non-isoch */
476 		buf = buffer + qh->offset;
477 		length = urb->transfer_buffer_length - qh->offset;
478 		if (rx_count > length) {
479 			if (urb->status == -EINPROGRESS)
480 				urb->status = -EOVERFLOW;
481 			musb_dbg(musb, "OVERFLOW %d into %d", rx_count, length);
482 			do_flush = 1;
483 		} else
484 			length = rx_count;
485 		urb->actual_length += length;
486 		qh->offset += length;
487 
488 		/* see if we are done */
489 		done = (urb->actual_length == urb->transfer_buffer_length)
490 			|| (rx_count < qh->maxpacket)
491 			|| (urb->status != -EINPROGRESS);
492 		if (done
493 				&& (urb->status == -EINPROGRESS)
494 				&& (urb->transfer_flags & URB_SHORT_NOT_OK)
495 				&& (urb->actual_length
496 					< urb->transfer_buffer_length))
497 			urb->status = -EREMOTEIO;
498 	}
499 
500 	musb_read_fifo(hw_ep, length, buf);
501 
502 	csr = musb_readw(epio, MUSB_RXCSR);
503 	csr |= MUSB_RXCSR_H_WZC_BITS;
504 	if (unlikely(do_flush))
505 		musb_h_flush_rxfifo(hw_ep, csr);
506 	else {
507 		/* REVISIT this assumes AUTOCLEAR is never set */
508 		csr &= ~(MUSB_RXCSR_RXPKTRDY | MUSB_RXCSR_H_REQPKT);
509 		if (!done)
510 			csr |= MUSB_RXCSR_H_REQPKT;
511 		musb_writew(epio, MUSB_RXCSR, csr);
512 	}
513 
514 	return done;
515 }
516 
517 /* we don't always need to reinit a given side of an endpoint...
518  * when we do, use tx/rx reinit routine and then construct a new CSR
519  * to address data toggle, NYET, and DMA or PIO.
520  *
521  * it's possible that driver bugs (especially for DMA) or aborting a
522  * transfer might have left the endpoint busier than it should be.
523  * the busy/not-empty tests are basically paranoia.
524  */
525 static void
526 musb_rx_reinit(struct musb *musb, struct musb_qh *qh, u8 epnum)
527 {
528 	struct musb_hw_ep *ep = musb->endpoints + epnum;
529 	u16	csr;
530 
531 	/* NOTE:  we know the "rx" fifo reinit never triggers for ep0.
532 	 * That always uses tx_reinit since ep0 repurposes TX register
533 	 * offsets; the initial SETUP packet is also a kind of OUT.
534 	 */
535 
536 	/* if programmed for Tx, put it in RX mode */
537 	if (ep->is_shared_fifo) {
538 		csr = musb_readw(ep->regs, MUSB_TXCSR);
539 		if (csr & MUSB_TXCSR_MODE) {
540 			musb_h_tx_flush_fifo(ep);
541 			csr = musb_readw(ep->regs, MUSB_TXCSR);
542 			musb_writew(ep->regs, MUSB_TXCSR,
543 				    csr | MUSB_TXCSR_FRCDATATOG);
544 		}
545 
546 		/*
547 		 * Clear the MODE bit (and everything else) to enable Rx.
548 		 * NOTE: we mustn't clear the DMAMODE bit before DMAENAB.
549 		 */
550 		if (csr & MUSB_TXCSR_DMAMODE)
551 			musb_writew(ep->regs, MUSB_TXCSR, MUSB_TXCSR_DMAMODE);
552 		musb_writew(ep->regs, MUSB_TXCSR, 0);
553 
554 	/* scrub all previous state, clearing toggle */
555 	}
556 	csr = musb_readw(ep->regs, MUSB_RXCSR);
557 	if (csr & MUSB_RXCSR_RXPKTRDY)
558 		WARNING("rx%d, packet/%d ready?\n", ep->epnum,
559 			musb_readw(ep->regs, MUSB_RXCOUNT));
560 
561 	musb_h_flush_rxfifo(ep, MUSB_RXCSR_CLRDATATOG);
562 
563 	/* target addr and (for multipoint) hub addr/port */
564 	if (musb->is_multipoint) {
565 		musb_write_rxfunaddr(musb, epnum, qh->addr_reg);
566 		musb_write_rxhubaddr(musb, epnum, qh->h_addr_reg);
567 		musb_write_rxhubport(musb, epnum, qh->h_port_reg);
568 	} else
569 		musb_writeb(musb->mregs, MUSB_FADDR, qh->addr_reg);
570 
571 	/* protocol/endpoint, interval/NAKlimit, i/o size */
572 	musb_writeb(ep->regs, MUSB_RXTYPE, qh->type_reg);
573 	musb_writeb(ep->regs, MUSB_RXINTERVAL, qh->intv_reg);
574 	/* NOTE: bulk combining rewrites high bits of maxpacket */
575 	/* Set RXMAXP with the FIFO size of the endpoint
576 	 * to disable double buffer mode.
577 	 */
578 	musb_writew(ep->regs, MUSB_RXMAXP,
579 			qh->maxpacket | ((qh->hb_mult - 1) << 11));
580 
581 	ep->rx_reinit = 0;
582 }
583 
584 static void musb_tx_dma_set_mode_mentor(struct dma_controller *dma,
585 		struct musb_hw_ep *hw_ep, struct musb_qh *qh,
586 		struct urb *urb, u32 offset,
587 		u32 *length, u8 *mode)
588 {
589 	struct dma_channel	*channel = hw_ep->tx_channel;
590 	void __iomem		*epio = hw_ep->regs;
591 	u16			pkt_size = qh->maxpacket;
592 	u16			csr;
593 
594 	if (*length > channel->max_len)
595 		*length = channel->max_len;
596 
597 	csr = musb_readw(epio, MUSB_TXCSR);
598 	if (*length > pkt_size) {
599 		*mode = 1;
600 		csr |= MUSB_TXCSR_DMAMODE | MUSB_TXCSR_DMAENAB;
601 		/* autoset shouldn't be set in high bandwidth */
602 		/*
603 		 * Enable Autoset according to table
604 		 * below
605 		 * bulk_split hb_mult	Autoset_Enable
606 		 *	0	1	Yes(Normal)
607 		 *	0	>1	No(High BW ISO)
608 		 *	1	1	Yes(HS bulk)
609 		 *	1	>1	Yes(FS bulk)
610 		 */
611 		if (qh->hb_mult == 1 || (qh->hb_mult > 1 &&
612 					can_bulk_split(hw_ep->musb, qh->type)))
613 			csr |= MUSB_TXCSR_AUTOSET;
614 	} else {
615 		*mode = 0;
616 		csr &= ~(MUSB_TXCSR_AUTOSET | MUSB_TXCSR_DMAMODE);
617 		csr |= MUSB_TXCSR_DMAENAB; /* against programmer's guide */
618 	}
619 	channel->desired_mode = *mode;
620 	musb_writew(epio, MUSB_TXCSR, csr);
621 }
622 
623 static void musb_tx_dma_set_mode_cppi_tusb(struct dma_controller *dma,
624 					   struct musb_hw_ep *hw_ep,
625 					   struct musb_qh *qh,
626 					   struct urb *urb,
627 					   u32 offset,
628 					   u32 *length,
629 					   u8 *mode)
630 {
631 	struct dma_channel *channel = hw_ep->tx_channel;
632 
633 	channel->actual_len = 0;
634 
635 	/*
636 	 * TX uses "RNDIS" mode automatically but needs help
637 	 * to identify the zero-length-final-packet case.
638 	 */
639 	*mode = (urb->transfer_flags & URB_ZERO_PACKET) ? 1 : 0;
640 }
641 
642 static bool musb_tx_dma_program(struct dma_controller *dma,
643 		struct musb_hw_ep *hw_ep, struct musb_qh *qh,
644 		struct urb *urb, u32 offset, u32 length)
645 {
646 	struct dma_channel	*channel = hw_ep->tx_channel;
647 	u16			pkt_size = qh->maxpacket;
648 	u8			mode;
649 
650 	if (musb_dma_inventra(hw_ep->musb) || musb_dma_ux500(hw_ep->musb))
651 		musb_tx_dma_set_mode_mentor(dma, hw_ep, qh, urb, offset,
652 					    &length, &mode);
653 	else if (is_cppi_enabled(hw_ep->musb) || tusb_dma_omap(hw_ep->musb))
654 		musb_tx_dma_set_mode_cppi_tusb(dma, hw_ep, qh, urb, offset,
655 					       &length, &mode);
656 	else
657 		return false;
658 
659 	qh->segsize = length;
660 
661 	/*
662 	 * Ensure the data reaches to main memory before starting
663 	 * DMA transfer
664 	 */
665 	wmb();
666 
667 	if (!dma->channel_program(channel, pkt_size, mode,
668 			urb->transfer_dma + offset, length)) {
669 		void __iomem *epio = hw_ep->regs;
670 		u16 csr;
671 
672 		dma->channel_release(channel);
673 		hw_ep->tx_channel = NULL;
674 
675 		csr = musb_readw(epio, MUSB_TXCSR);
676 		csr &= ~(MUSB_TXCSR_AUTOSET | MUSB_TXCSR_DMAENAB);
677 		musb_writew(epio, MUSB_TXCSR, csr | MUSB_TXCSR_H_WZC_BITS);
678 		return false;
679 	}
680 	return true;
681 }
682 
683 /*
684  * Program an HDRC endpoint as per the given URB
685  * Context: irqs blocked, controller lock held
686  */
687 static void musb_ep_program(struct musb *musb, u8 epnum,
688 			struct urb *urb, int is_out,
689 			u8 *buf, u32 offset, u32 len)
690 {
691 	struct dma_controller	*dma_controller;
692 	struct dma_channel	*dma_channel;
693 	u8			dma_ok;
694 	void __iomem		*mbase = musb->mregs;
695 	struct musb_hw_ep	*hw_ep = musb->endpoints + epnum;
696 	void __iomem		*epio = hw_ep->regs;
697 	struct musb_qh		*qh = musb_ep_get_qh(hw_ep, !is_out);
698 	u16			packet_sz = qh->maxpacket;
699 	u8			use_dma = 1;
700 	u16			csr;
701 
702 	musb_dbg(musb, "%s hw%d urb %p spd%d dev%d ep%d%s "
703 				"h_addr%02x h_port%02x bytes %d",
704 			is_out ? "-->" : "<--",
705 			epnum, urb, urb->dev->speed,
706 			qh->addr_reg, qh->epnum, is_out ? "out" : "in",
707 			qh->h_addr_reg, qh->h_port_reg,
708 			len);
709 
710 	musb_ep_select(mbase, epnum);
711 
712 	if (is_out && !len) {
713 		use_dma = 0;
714 		csr = musb_readw(epio, MUSB_TXCSR);
715 		csr &= ~MUSB_TXCSR_DMAENAB;
716 		musb_writew(epio, MUSB_TXCSR, csr);
717 		hw_ep->tx_channel = NULL;
718 	}
719 
720 	/* candidate for DMA? */
721 	dma_controller = musb->dma_controller;
722 	if (use_dma && is_dma_capable() && epnum && dma_controller) {
723 		dma_channel = is_out ? hw_ep->tx_channel : hw_ep->rx_channel;
724 		if (!dma_channel) {
725 			dma_channel = dma_controller->channel_alloc(
726 					dma_controller, hw_ep, is_out);
727 			if (is_out)
728 				hw_ep->tx_channel = dma_channel;
729 			else
730 				hw_ep->rx_channel = dma_channel;
731 		}
732 	} else
733 		dma_channel = NULL;
734 
735 	/* make sure we clear DMAEnab, autoSet bits from previous run */
736 
737 	/* OUT/transmit/EP0 or IN/receive? */
738 	if (is_out) {
739 		u16	csr;
740 		u16	int_txe;
741 		u16	load_count;
742 
743 		csr = musb_readw(epio, MUSB_TXCSR);
744 
745 		/* disable interrupt in case we flush */
746 		int_txe = musb->intrtxe;
747 		musb_writew(mbase, MUSB_INTRTXE, int_txe & ~(1 << epnum));
748 
749 		/* general endpoint setup */
750 		if (epnum) {
751 			/* flush all old state, set default */
752 			/*
753 			 * We could be flushing valid
754 			 * packets in double buffering
755 			 * case
756 			 */
757 			if (!hw_ep->tx_double_buffered)
758 				musb_h_tx_flush_fifo(hw_ep);
759 
760 			/*
761 			 * We must not clear the DMAMODE bit before or in
762 			 * the same cycle with the DMAENAB bit, so we clear
763 			 * the latter first...
764 			 */
765 			csr &= ~(MUSB_TXCSR_H_NAKTIMEOUT
766 					| MUSB_TXCSR_AUTOSET
767 					| MUSB_TXCSR_DMAENAB
768 					| MUSB_TXCSR_FRCDATATOG
769 					| MUSB_TXCSR_H_RXSTALL
770 					| MUSB_TXCSR_H_ERROR
771 					| MUSB_TXCSR_TXPKTRDY
772 					);
773 			csr |= MUSB_TXCSR_MODE;
774 
775 			if (!hw_ep->tx_double_buffered) {
776 				if (usb_gettoggle(urb->dev, qh->epnum, 1))
777 					csr |= MUSB_TXCSR_H_WR_DATATOGGLE
778 						| MUSB_TXCSR_H_DATATOGGLE;
779 				else
780 					csr |= MUSB_TXCSR_CLRDATATOG;
781 			}
782 
783 			musb_writew(epio, MUSB_TXCSR, csr);
784 			/* REVISIT may need to clear FLUSHFIFO ... */
785 			csr &= ~MUSB_TXCSR_DMAMODE;
786 			musb_writew(epio, MUSB_TXCSR, csr);
787 			csr = musb_readw(epio, MUSB_TXCSR);
788 		} else {
789 			/* endpoint 0: just flush */
790 			musb_h_ep0_flush_fifo(hw_ep);
791 		}
792 
793 		/* target addr and (for multipoint) hub addr/port */
794 		if (musb->is_multipoint) {
795 			musb_write_txfunaddr(musb, epnum, qh->addr_reg);
796 			musb_write_txhubaddr(musb, epnum, qh->h_addr_reg);
797 			musb_write_txhubport(musb, epnum, qh->h_port_reg);
798 /* FIXME if !epnum, do the same for RX ... */
799 		} else
800 			musb_writeb(mbase, MUSB_FADDR, qh->addr_reg);
801 
802 		/* protocol/endpoint/interval/NAKlimit */
803 		if (epnum) {
804 			musb_writeb(epio, MUSB_TXTYPE, qh->type_reg);
805 			if (can_bulk_split(musb, qh->type)) {
806 				qh->hb_mult = hw_ep->max_packet_sz_tx
807 						/ packet_sz;
808 				musb_writew(epio, MUSB_TXMAXP, packet_sz
809 					| ((qh->hb_mult) - 1) << 11);
810 			} else {
811 				musb_writew(epio, MUSB_TXMAXP,
812 						qh->maxpacket |
813 						((qh->hb_mult - 1) << 11));
814 			}
815 			musb_writeb(epio, MUSB_TXINTERVAL, qh->intv_reg);
816 		} else {
817 			musb_writeb(epio, MUSB_NAKLIMIT0, qh->intv_reg);
818 			if (musb->is_multipoint)
819 				musb_writeb(epio, MUSB_TYPE0,
820 						qh->type_reg);
821 		}
822 
823 		if (can_bulk_split(musb, qh->type))
824 			load_count = min((u32) hw_ep->max_packet_sz_tx,
825 						len);
826 		else
827 			load_count = min((u32) packet_sz, len);
828 
829 		if (dma_channel && musb_tx_dma_program(dma_controller,
830 					hw_ep, qh, urb, offset, len))
831 			load_count = 0;
832 
833 		if (load_count) {
834 			/* PIO to load FIFO */
835 			qh->segsize = load_count;
836 			if (!buf) {
837 				sg_miter_start(&qh->sg_miter, urb->sg, 1,
838 						SG_MITER_ATOMIC
839 						| SG_MITER_FROM_SG);
840 				if (!sg_miter_next(&qh->sg_miter)) {
841 					dev_err(musb->controller,
842 							"error: sg"
843 							"list empty\n");
844 					sg_miter_stop(&qh->sg_miter);
845 					goto finish;
846 				}
847 				buf = qh->sg_miter.addr + urb->sg->offset +
848 					urb->actual_length;
849 				load_count = min_t(u32, load_count,
850 						qh->sg_miter.length);
851 				musb_write_fifo(hw_ep, load_count, buf);
852 				qh->sg_miter.consumed = load_count;
853 				sg_miter_stop(&qh->sg_miter);
854 			} else
855 				musb_write_fifo(hw_ep, load_count, buf);
856 		}
857 finish:
858 		/* re-enable interrupt */
859 		musb_writew(mbase, MUSB_INTRTXE, int_txe);
860 
861 	/* IN/receive */
862 	} else {
863 		u16	csr;
864 
865 		if (hw_ep->rx_reinit) {
866 			musb_rx_reinit(musb, qh, epnum);
867 
868 			/* init new state: toggle and NYET, maybe DMA later */
869 			if (usb_gettoggle(urb->dev, qh->epnum, 0))
870 				csr = MUSB_RXCSR_H_WR_DATATOGGLE
871 					| MUSB_RXCSR_H_DATATOGGLE;
872 			else
873 				csr = 0;
874 			if (qh->type == USB_ENDPOINT_XFER_INT)
875 				csr |= MUSB_RXCSR_DISNYET;
876 
877 		} else {
878 			csr = musb_readw(hw_ep->regs, MUSB_RXCSR);
879 
880 			if (csr & (MUSB_RXCSR_RXPKTRDY
881 					| MUSB_RXCSR_DMAENAB
882 					| MUSB_RXCSR_H_REQPKT))
883 				ERR("broken !rx_reinit, ep%d csr %04x\n",
884 						hw_ep->epnum, csr);
885 
886 			/* scrub any stale state, leaving toggle alone */
887 			csr &= MUSB_RXCSR_DISNYET;
888 		}
889 
890 		/* kick things off */
891 
892 		if ((is_cppi_enabled(musb) || tusb_dma_omap(musb)) && dma_channel) {
893 			/* Candidate for DMA */
894 			dma_channel->actual_len = 0L;
895 			qh->segsize = len;
896 
897 			/* AUTOREQ is in a DMA register */
898 			musb_writew(hw_ep->regs, MUSB_RXCSR, csr);
899 			csr = musb_readw(hw_ep->regs, MUSB_RXCSR);
900 
901 			/*
902 			 * Unless caller treats short RX transfers as
903 			 * errors, we dare not queue multiple transfers.
904 			 */
905 			dma_ok = dma_controller->channel_program(dma_channel,
906 					packet_sz, !(urb->transfer_flags &
907 						     URB_SHORT_NOT_OK),
908 					urb->transfer_dma + offset,
909 					qh->segsize);
910 			if (!dma_ok) {
911 				dma_controller->channel_release(dma_channel);
912 				hw_ep->rx_channel = dma_channel = NULL;
913 			} else
914 				csr |= MUSB_RXCSR_DMAENAB;
915 		}
916 
917 		csr |= MUSB_RXCSR_H_REQPKT;
918 		musb_dbg(musb, "RXCSR%d := %04x", epnum, csr);
919 		musb_writew(hw_ep->regs, MUSB_RXCSR, csr);
920 		csr = musb_readw(hw_ep->regs, MUSB_RXCSR);
921 	}
922 }
923 
924 /* Schedule next QH from musb->in_bulk/out_bulk and move the current qh to
925  * the end; avoids starvation for other endpoints.
926  */
927 static void musb_bulk_nak_timeout(struct musb *musb, struct musb_hw_ep *ep,
928 	int is_in)
929 {
930 	struct dma_channel	*dma;
931 	struct urb		*urb;
932 	void __iomem		*mbase = musb->mregs;
933 	void __iomem		*epio = ep->regs;
934 	struct musb_qh		*cur_qh, *next_qh;
935 	u16			rx_csr, tx_csr;
936 
937 	musb_ep_select(mbase, ep->epnum);
938 	if (is_in) {
939 		dma = is_dma_capable() ? ep->rx_channel : NULL;
940 
941 		/*
942 		 * Need to stop the transaction by clearing REQPKT first
943 		 * then the NAK Timeout bit ref MUSBMHDRC USB 2.0 HIGH-SPEED
944 		 * DUAL-ROLE CONTROLLER Programmer's Guide, section 9.2.2
945 		 */
946 		rx_csr = musb_readw(epio, MUSB_RXCSR);
947 		rx_csr |= MUSB_RXCSR_H_WZC_BITS;
948 		rx_csr &= ~MUSB_RXCSR_H_REQPKT;
949 		musb_writew(epio, MUSB_RXCSR, rx_csr);
950 		rx_csr &= ~MUSB_RXCSR_DATAERROR;
951 		musb_writew(epio, MUSB_RXCSR, rx_csr);
952 
953 		cur_qh = first_qh(&musb->in_bulk);
954 	} else {
955 		dma = is_dma_capable() ? ep->tx_channel : NULL;
956 
957 		/* clear nak timeout bit */
958 		tx_csr = musb_readw(epio, MUSB_TXCSR);
959 		tx_csr |= MUSB_TXCSR_H_WZC_BITS;
960 		tx_csr &= ~MUSB_TXCSR_H_NAKTIMEOUT;
961 		musb_writew(epio, MUSB_TXCSR, tx_csr);
962 
963 		cur_qh = first_qh(&musb->out_bulk);
964 	}
965 	if (cur_qh) {
966 		urb = next_urb(cur_qh);
967 		if (dma_channel_status(dma) == MUSB_DMA_STATUS_BUSY) {
968 			dma->status = MUSB_DMA_STATUS_CORE_ABORT;
969 			musb->dma_controller->channel_abort(dma);
970 			urb->actual_length += dma->actual_len;
971 			dma->actual_len = 0L;
972 		}
973 		musb_save_toggle(cur_qh, is_in, urb);
974 
975 		if (is_in) {
976 			/* move cur_qh to end of queue */
977 			list_move_tail(&cur_qh->ring, &musb->in_bulk);
978 
979 			/* get the next qh from musb->in_bulk */
980 			next_qh = first_qh(&musb->in_bulk);
981 
982 			/* set rx_reinit and schedule the next qh */
983 			ep->rx_reinit = 1;
984 		} else {
985 			/* move cur_qh to end of queue */
986 			list_move_tail(&cur_qh->ring, &musb->out_bulk);
987 
988 			/* get the next qh from musb->out_bulk */
989 			next_qh = first_qh(&musb->out_bulk);
990 
991 			/* set tx_reinit and schedule the next qh */
992 			ep->tx_reinit = 1;
993 		}
994 
995 		if (next_qh)
996 			musb_start_urb(musb, is_in, next_qh);
997 	}
998 }
999 
1000 /*
1001  * Service the default endpoint (ep0) as host.
1002  * Return true until it's time to start the status stage.
1003  */
1004 static bool musb_h_ep0_continue(struct musb *musb, u16 len, struct urb *urb)
1005 {
1006 	bool			 more = false;
1007 	u8			*fifo_dest = NULL;
1008 	u16			fifo_count = 0;
1009 	struct musb_hw_ep	*hw_ep = musb->control_ep;
1010 	struct musb_qh		*qh = hw_ep->in_qh;
1011 	struct usb_ctrlrequest	*request;
1012 
1013 	switch (musb->ep0_stage) {
1014 	case MUSB_EP0_IN:
1015 		fifo_dest = urb->transfer_buffer + urb->actual_length;
1016 		fifo_count = min_t(size_t, len, urb->transfer_buffer_length -
1017 				   urb->actual_length);
1018 		if (fifo_count < len)
1019 			urb->status = -EOVERFLOW;
1020 
1021 		musb_read_fifo(hw_ep, fifo_count, fifo_dest);
1022 
1023 		urb->actual_length += fifo_count;
1024 		if (len < qh->maxpacket) {
1025 			/* always terminate on short read; it's
1026 			 * rarely reported as an error.
1027 			 */
1028 		} else if (urb->actual_length <
1029 				urb->transfer_buffer_length)
1030 			more = true;
1031 		break;
1032 	case MUSB_EP0_START:
1033 		request = (struct usb_ctrlrequest *) urb->setup_packet;
1034 
1035 		if (!request->wLength) {
1036 			musb_dbg(musb, "start no-DATA");
1037 			break;
1038 		} else if (request->bRequestType & USB_DIR_IN) {
1039 			musb_dbg(musb, "start IN-DATA");
1040 			musb->ep0_stage = MUSB_EP0_IN;
1041 			more = true;
1042 			break;
1043 		} else {
1044 			musb_dbg(musb, "start OUT-DATA");
1045 			musb->ep0_stage = MUSB_EP0_OUT;
1046 			more = true;
1047 		}
1048 		/* FALLTHROUGH */
1049 	case MUSB_EP0_OUT:
1050 		fifo_count = min_t(size_t, qh->maxpacket,
1051 				   urb->transfer_buffer_length -
1052 				   urb->actual_length);
1053 		if (fifo_count) {
1054 			fifo_dest = (u8 *) (urb->transfer_buffer
1055 					+ urb->actual_length);
1056 			musb_dbg(musb, "Sending %d byte%s to ep0 fifo %p",
1057 					fifo_count,
1058 					(fifo_count == 1) ? "" : "s",
1059 					fifo_dest);
1060 			musb_write_fifo(hw_ep, fifo_count, fifo_dest);
1061 
1062 			urb->actual_length += fifo_count;
1063 			more = true;
1064 		}
1065 		break;
1066 	default:
1067 		ERR("bogus ep0 stage %d\n", musb->ep0_stage);
1068 		break;
1069 	}
1070 
1071 	return more;
1072 }
1073 
1074 /*
1075  * Handle default endpoint interrupt as host. Only called in IRQ time
1076  * from musb_interrupt().
1077  *
1078  * called with controller irqlocked
1079  */
1080 irqreturn_t musb_h_ep0_irq(struct musb *musb)
1081 {
1082 	struct urb		*urb;
1083 	u16			csr, len;
1084 	int			status = 0;
1085 	void __iomem		*mbase = musb->mregs;
1086 	struct musb_hw_ep	*hw_ep = musb->control_ep;
1087 	void __iomem		*epio = hw_ep->regs;
1088 	struct musb_qh		*qh = hw_ep->in_qh;
1089 	bool			complete = false;
1090 	irqreturn_t		retval = IRQ_NONE;
1091 
1092 	/* ep0 only has one queue, "in" */
1093 	urb = next_urb(qh);
1094 
1095 	musb_ep_select(mbase, 0);
1096 	csr = musb_readw(epio, MUSB_CSR0);
1097 	len = (csr & MUSB_CSR0_RXPKTRDY)
1098 			? musb_readb(epio, MUSB_COUNT0)
1099 			: 0;
1100 
1101 	musb_dbg(musb, "<== csr0 %04x, qh %p, count %d, urb %p, stage %d",
1102 		csr, qh, len, urb, musb->ep0_stage);
1103 
1104 	/* if we just did status stage, we are done */
1105 	if (MUSB_EP0_STATUS == musb->ep0_stage) {
1106 		retval = IRQ_HANDLED;
1107 		complete = true;
1108 	}
1109 
1110 	/* prepare status */
1111 	if (csr & MUSB_CSR0_H_RXSTALL) {
1112 		musb_dbg(musb, "STALLING ENDPOINT");
1113 		status = -EPIPE;
1114 
1115 	} else if (csr & MUSB_CSR0_H_ERROR) {
1116 		musb_dbg(musb, "no response, csr0 %04x", csr);
1117 		status = -EPROTO;
1118 
1119 	} else if (csr & MUSB_CSR0_H_NAKTIMEOUT) {
1120 		musb_dbg(musb, "control NAK timeout");
1121 
1122 		/* NOTE:  this code path would be a good place to PAUSE a
1123 		 * control transfer, if another one is queued, so that
1124 		 * ep0 is more likely to stay busy.  That's already done
1125 		 * for bulk RX transfers.
1126 		 *
1127 		 * if (qh->ring.next != &musb->control), then
1128 		 * we have a candidate... NAKing is *NOT* an error
1129 		 */
1130 		musb_writew(epio, MUSB_CSR0, 0);
1131 		retval = IRQ_HANDLED;
1132 	}
1133 
1134 	if (status) {
1135 		musb_dbg(musb, "aborting");
1136 		retval = IRQ_HANDLED;
1137 		if (urb)
1138 			urb->status = status;
1139 		complete = true;
1140 
1141 		/* use the proper sequence to abort the transfer */
1142 		if (csr & MUSB_CSR0_H_REQPKT) {
1143 			csr &= ~MUSB_CSR0_H_REQPKT;
1144 			musb_writew(epio, MUSB_CSR0, csr);
1145 			csr &= ~MUSB_CSR0_H_NAKTIMEOUT;
1146 			musb_writew(epio, MUSB_CSR0, csr);
1147 		} else {
1148 			musb_h_ep0_flush_fifo(hw_ep);
1149 		}
1150 
1151 		musb_writeb(epio, MUSB_NAKLIMIT0, 0);
1152 
1153 		/* clear it */
1154 		musb_writew(epio, MUSB_CSR0, 0);
1155 	}
1156 
1157 	if (unlikely(!urb)) {
1158 		/* stop endpoint since we have no place for its data, this
1159 		 * SHOULD NEVER HAPPEN! */
1160 		ERR("no URB for end 0\n");
1161 
1162 		musb_h_ep0_flush_fifo(hw_ep);
1163 		goto done;
1164 	}
1165 
1166 	if (!complete) {
1167 		/* call common logic and prepare response */
1168 		if (musb_h_ep0_continue(musb, len, urb)) {
1169 			/* more packets required */
1170 			csr = (MUSB_EP0_IN == musb->ep0_stage)
1171 				?  MUSB_CSR0_H_REQPKT : MUSB_CSR0_TXPKTRDY;
1172 		} else {
1173 			/* data transfer complete; perform status phase */
1174 			if (usb_pipeout(urb->pipe)
1175 					|| !urb->transfer_buffer_length)
1176 				csr = MUSB_CSR0_H_STATUSPKT
1177 					| MUSB_CSR0_H_REQPKT;
1178 			else
1179 				csr = MUSB_CSR0_H_STATUSPKT
1180 					| MUSB_CSR0_TXPKTRDY;
1181 
1182 			/* disable ping token in status phase */
1183 			csr |= MUSB_CSR0_H_DIS_PING;
1184 
1185 			/* flag status stage */
1186 			musb->ep0_stage = MUSB_EP0_STATUS;
1187 
1188 			musb_dbg(musb, "ep0 STATUS, csr %04x", csr);
1189 
1190 		}
1191 		musb_writew(epio, MUSB_CSR0, csr);
1192 		retval = IRQ_HANDLED;
1193 	} else
1194 		musb->ep0_stage = MUSB_EP0_IDLE;
1195 
1196 	/* call completion handler if done */
1197 	if (complete)
1198 		musb_advance_schedule(musb, urb, hw_ep, 1);
1199 done:
1200 	return retval;
1201 }
1202 
1203 
1204 #ifdef CONFIG_USB_INVENTRA_DMA
1205 
1206 /* Host side TX (OUT) using Mentor DMA works as follows:
1207 	submit_urb ->
1208 		- if queue was empty, Program Endpoint
1209 		- ... which starts DMA to fifo in mode 1 or 0
1210 
1211 	DMA Isr (transfer complete) -> TxAvail()
1212 		- Stop DMA (~DmaEnab)	(<--- Alert ... currently happens
1213 					only in musb_cleanup_urb)
1214 		- TxPktRdy has to be set in mode 0 or for
1215 			short packets in mode 1.
1216 */
1217 
1218 #endif
1219 
1220 /* Service a Tx-Available or dma completion irq for the endpoint */
1221 void musb_host_tx(struct musb *musb, u8 epnum)
1222 {
1223 	int			pipe;
1224 	bool			done = false;
1225 	u16			tx_csr;
1226 	size_t			length = 0;
1227 	size_t			offset = 0;
1228 	struct musb_hw_ep	*hw_ep = musb->endpoints + epnum;
1229 	void __iomem		*epio = hw_ep->regs;
1230 	struct musb_qh		*qh = hw_ep->out_qh;
1231 	struct urb		*urb = next_urb(qh);
1232 	u32			status = 0;
1233 	void __iomem		*mbase = musb->mregs;
1234 	struct dma_channel	*dma;
1235 	bool			transfer_pending = false;
1236 
1237 	musb_ep_select(mbase, epnum);
1238 	tx_csr = musb_readw(epio, MUSB_TXCSR);
1239 
1240 	/* with CPPI, DMA sometimes triggers "extra" irqs */
1241 	if (!urb) {
1242 		musb_dbg(musb, "extra TX%d ready, csr %04x", epnum, tx_csr);
1243 		return;
1244 	}
1245 
1246 	pipe = urb->pipe;
1247 	dma = is_dma_capable() ? hw_ep->tx_channel : NULL;
1248 	trace_musb_urb_tx(musb, urb);
1249 	musb_dbg(musb, "OUT/TX%d end, csr %04x%s", epnum, tx_csr,
1250 			dma ? ", dma" : "");
1251 
1252 	/* check for errors */
1253 	if (tx_csr & MUSB_TXCSR_H_RXSTALL) {
1254 		/* dma was disabled, fifo flushed */
1255 		musb_dbg(musb, "TX end %d stall", epnum);
1256 
1257 		/* stall; record URB status */
1258 		status = -EPIPE;
1259 
1260 	} else if (tx_csr & MUSB_TXCSR_H_ERROR) {
1261 		/* (NON-ISO) dma was disabled, fifo flushed */
1262 		musb_dbg(musb, "TX 3strikes on ep=%d", epnum);
1263 
1264 		status = -ETIMEDOUT;
1265 
1266 	} else if (tx_csr & MUSB_TXCSR_H_NAKTIMEOUT) {
1267 		if (USB_ENDPOINT_XFER_BULK == qh->type && qh->mux == 1
1268 				&& !list_is_singular(&musb->out_bulk)) {
1269 			musb_dbg(musb, "NAK timeout on TX%d ep", epnum);
1270 			musb_bulk_nak_timeout(musb, hw_ep, 0);
1271 		} else {
1272 			musb_dbg(musb, "TX ep%d device not responding", epnum);
1273 			/* NOTE:  this code path would be a good place to PAUSE a
1274 			 * transfer, if there's some other (nonperiodic) tx urb
1275 			 * that could use this fifo.  (dma complicates it...)
1276 			 * That's already done for bulk RX transfers.
1277 			 *
1278 			 * if (bulk && qh->ring.next != &musb->out_bulk), then
1279 			 * we have a candidate... NAKing is *NOT* an error
1280 			 */
1281 			musb_ep_select(mbase, epnum);
1282 			musb_writew(epio, MUSB_TXCSR,
1283 					MUSB_TXCSR_H_WZC_BITS
1284 					| MUSB_TXCSR_TXPKTRDY);
1285 		}
1286 		return;
1287 	}
1288 
1289 done:
1290 	if (status) {
1291 		if (dma_channel_status(dma) == MUSB_DMA_STATUS_BUSY) {
1292 			dma->status = MUSB_DMA_STATUS_CORE_ABORT;
1293 			musb->dma_controller->channel_abort(dma);
1294 		}
1295 
1296 		/* do the proper sequence to abort the transfer in the
1297 		 * usb core; the dma engine should already be stopped.
1298 		 */
1299 		musb_h_tx_flush_fifo(hw_ep);
1300 		tx_csr &= ~(MUSB_TXCSR_AUTOSET
1301 				| MUSB_TXCSR_DMAENAB
1302 				| MUSB_TXCSR_H_ERROR
1303 				| MUSB_TXCSR_H_RXSTALL
1304 				| MUSB_TXCSR_H_NAKTIMEOUT
1305 				);
1306 
1307 		musb_ep_select(mbase, epnum);
1308 		musb_writew(epio, MUSB_TXCSR, tx_csr);
1309 		/* REVISIT may need to clear FLUSHFIFO ... */
1310 		musb_writew(epio, MUSB_TXCSR, tx_csr);
1311 		musb_writeb(epio, MUSB_TXINTERVAL, 0);
1312 
1313 		done = true;
1314 	}
1315 
1316 	/* second cppi case */
1317 	if (dma_channel_status(dma) == MUSB_DMA_STATUS_BUSY) {
1318 		musb_dbg(musb, "extra TX%d ready, csr %04x", epnum, tx_csr);
1319 		return;
1320 	}
1321 
1322 	if (is_dma_capable() && dma && !status) {
1323 		/*
1324 		 * DMA has completed.  But if we're using DMA mode 1 (multi
1325 		 * packet DMA), we need a terminal TXPKTRDY interrupt before
1326 		 * we can consider this transfer completed, lest we trash
1327 		 * its last packet when writing the next URB's data.  So we
1328 		 * switch back to mode 0 to get that interrupt; we'll come
1329 		 * back here once it happens.
1330 		 */
1331 		if (tx_csr & MUSB_TXCSR_DMAMODE) {
1332 			/*
1333 			 * We shouldn't clear DMAMODE with DMAENAB set; so
1334 			 * clear them in a safe order.  That should be OK
1335 			 * once TXPKTRDY has been set (and I've never seen
1336 			 * it being 0 at this moment -- DMA interrupt latency
1337 			 * is significant) but if it hasn't been then we have
1338 			 * no choice but to stop being polite and ignore the
1339 			 * programmer's guide... :-)
1340 			 *
1341 			 * Note that we must write TXCSR with TXPKTRDY cleared
1342 			 * in order not to re-trigger the packet send (this bit
1343 			 * can't be cleared by CPU), and there's another caveat:
1344 			 * TXPKTRDY may be set shortly and then cleared in the
1345 			 * double-buffered FIFO mode, so we do an extra TXCSR
1346 			 * read for debouncing...
1347 			 */
1348 			tx_csr &= musb_readw(epio, MUSB_TXCSR);
1349 			if (tx_csr & MUSB_TXCSR_TXPKTRDY) {
1350 				tx_csr &= ~(MUSB_TXCSR_DMAENAB |
1351 					    MUSB_TXCSR_TXPKTRDY);
1352 				musb_writew(epio, MUSB_TXCSR,
1353 					    tx_csr | MUSB_TXCSR_H_WZC_BITS);
1354 			}
1355 			tx_csr &= ~(MUSB_TXCSR_DMAMODE |
1356 				    MUSB_TXCSR_TXPKTRDY);
1357 			musb_writew(epio, MUSB_TXCSR,
1358 				    tx_csr | MUSB_TXCSR_H_WZC_BITS);
1359 
1360 			/*
1361 			 * There is no guarantee that we'll get an interrupt
1362 			 * after clearing DMAMODE as we might have done this
1363 			 * too late (after TXPKTRDY was cleared by controller).
1364 			 * Re-read TXCSR as we have spoiled its previous value.
1365 			 */
1366 			tx_csr = musb_readw(epio, MUSB_TXCSR);
1367 		}
1368 
1369 		/*
1370 		 * We may get here from a DMA completion or TXPKTRDY interrupt.
1371 		 * In any case, we must check the FIFO status here and bail out
1372 		 * only if the FIFO still has data -- that should prevent the
1373 		 * "missed" TXPKTRDY interrupts and deal with double-buffered
1374 		 * FIFO mode too...
1375 		 */
1376 		if (tx_csr & (MUSB_TXCSR_FIFONOTEMPTY | MUSB_TXCSR_TXPKTRDY)) {
1377 			musb_dbg(musb,
1378 				"DMA complete but FIFO not empty, CSR %04x",
1379 				tx_csr);
1380 			return;
1381 		}
1382 	}
1383 
1384 	if (!status || dma || usb_pipeisoc(pipe)) {
1385 		if (dma)
1386 			length = dma->actual_len;
1387 		else
1388 			length = qh->segsize;
1389 		qh->offset += length;
1390 
1391 		if (usb_pipeisoc(pipe)) {
1392 			struct usb_iso_packet_descriptor	*d;
1393 
1394 			d = urb->iso_frame_desc + qh->iso_idx;
1395 			d->actual_length = length;
1396 			d->status = status;
1397 			if (++qh->iso_idx >= urb->number_of_packets) {
1398 				done = true;
1399 			} else {
1400 				d++;
1401 				offset = d->offset;
1402 				length = d->length;
1403 			}
1404 		} else if (dma && urb->transfer_buffer_length == qh->offset) {
1405 			done = true;
1406 		} else {
1407 			/* see if we need to send more data, or ZLP */
1408 			if (qh->segsize < qh->maxpacket)
1409 				done = true;
1410 			else if (qh->offset == urb->transfer_buffer_length
1411 					&& !(urb->transfer_flags
1412 						& URB_ZERO_PACKET))
1413 				done = true;
1414 			if (!done) {
1415 				offset = qh->offset;
1416 				length = urb->transfer_buffer_length - offset;
1417 				transfer_pending = true;
1418 			}
1419 		}
1420 	}
1421 
1422 	/* urb->status != -EINPROGRESS means request has been faulted,
1423 	 * so we must abort this transfer after cleanup
1424 	 */
1425 	if (urb->status != -EINPROGRESS) {
1426 		done = true;
1427 		if (status == 0)
1428 			status = urb->status;
1429 	}
1430 
1431 	if (done) {
1432 		/* set status */
1433 		urb->status = status;
1434 		urb->actual_length = qh->offset;
1435 		musb_advance_schedule(musb, urb, hw_ep, USB_DIR_OUT);
1436 		return;
1437 	} else if ((usb_pipeisoc(pipe) || transfer_pending) && dma) {
1438 		if (musb_tx_dma_program(musb->dma_controller, hw_ep, qh, urb,
1439 				offset, length)) {
1440 			if (is_cppi_enabled(musb) || tusb_dma_omap(musb))
1441 				musb_h_tx_dma_start(hw_ep);
1442 			return;
1443 		}
1444 	} else	if (tx_csr & MUSB_TXCSR_DMAENAB) {
1445 		musb_dbg(musb, "not complete, but DMA enabled?");
1446 		return;
1447 	}
1448 
1449 	/*
1450 	 * PIO: start next packet in this URB.
1451 	 *
1452 	 * REVISIT: some docs say that when hw_ep->tx_double_buffered,
1453 	 * (and presumably, FIFO is not half-full) we should write *two*
1454 	 * packets before updating TXCSR; other docs disagree...
1455 	 */
1456 	if (length > qh->maxpacket)
1457 		length = qh->maxpacket;
1458 	/* Unmap the buffer so that CPU can use it */
1459 	usb_hcd_unmap_urb_for_dma(musb->hcd, urb);
1460 
1461 	/*
1462 	 * We need to map sg if the transfer_buffer is
1463 	 * NULL.
1464 	 */
1465 	if (!urb->transfer_buffer)
1466 		qh->use_sg = true;
1467 
1468 	if (qh->use_sg) {
1469 		/* sg_miter_start is already done in musb_ep_program */
1470 		if (!sg_miter_next(&qh->sg_miter)) {
1471 			dev_err(musb->controller, "error: sg list empty\n");
1472 			sg_miter_stop(&qh->sg_miter);
1473 			status = -EINVAL;
1474 			goto done;
1475 		}
1476 		urb->transfer_buffer = qh->sg_miter.addr;
1477 		length = min_t(u32, length, qh->sg_miter.length);
1478 		musb_write_fifo(hw_ep, length, urb->transfer_buffer);
1479 		qh->sg_miter.consumed = length;
1480 		sg_miter_stop(&qh->sg_miter);
1481 	} else {
1482 		musb_write_fifo(hw_ep, length, urb->transfer_buffer + offset);
1483 	}
1484 
1485 	qh->segsize = length;
1486 
1487 	if (qh->use_sg) {
1488 		if (offset + length >= urb->transfer_buffer_length)
1489 			qh->use_sg = false;
1490 	}
1491 
1492 	musb_ep_select(mbase, epnum);
1493 	musb_writew(epio, MUSB_TXCSR,
1494 			MUSB_TXCSR_H_WZC_BITS | MUSB_TXCSR_TXPKTRDY);
1495 }
1496 
1497 #ifdef CONFIG_USB_TI_CPPI41_DMA
1498 /* Seems to set up ISO for cppi41 and not advance len. See commit c57c41d */
1499 static int musb_rx_dma_iso_cppi41(struct dma_controller *dma,
1500 				  struct musb_hw_ep *hw_ep,
1501 				  struct musb_qh *qh,
1502 				  struct urb *urb,
1503 				  size_t len)
1504 {
1505 	struct dma_channel *channel = hw_ep->rx_channel;
1506 	void __iomem *epio = hw_ep->regs;
1507 	dma_addr_t *buf;
1508 	u32 length;
1509 	u16 val;
1510 
1511 	buf = (void *)urb->iso_frame_desc[qh->iso_idx].offset +
1512 		(u32)urb->transfer_dma;
1513 
1514 	length = urb->iso_frame_desc[qh->iso_idx].length;
1515 
1516 	val = musb_readw(epio, MUSB_RXCSR);
1517 	val |= MUSB_RXCSR_DMAENAB;
1518 	musb_writew(hw_ep->regs, MUSB_RXCSR, val);
1519 
1520 	return dma->channel_program(channel, qh->maxpacket, 0,
1521 				   (u32)buf, length);
1522 }
1523 #else
1524 static inline int musb_rx_dma_iso_cppi41(struct dma_controller *dma,
1525 					 struct musb_hw_ep *hw_ep,
1526 					 struct musb_qh *qh,
1527 					 struct urb *urb,
1528 					 size_t len)
1529 {
1530 	return false;
1531 }
1532 #endif
1533 
1534 #if defined(CONFIG_USB_INVENTRA_DMA) || defined(CONFIG_USB_UX500_DMA) || \
1535 	defined(CONFIG_USB_TI_CPPI41_DMA)
1536 /* Host side RX (IN) using Mentor DMA works as follows:
1537 	submit_urb ->
1538 		- if queue was empty, ProgramEndpoint
1539 		- first IN token is sent out (by setting ReqPkt)
1540 	LinuxIsr -> RxReady()
1541 	/\	=> first packet is received
1542 	|	- Set in mode 0 (DmaEnab, ~ReqPkt)
1543 	|		-> DMA Isr (transfer complete) -> RxReady()
1544 	|		    - Ack receive (~RxPktRdy), turn off DMA (~DmaEnab)
1545 	|		    - if urb not complete, send next IN token (ReqPkt)
1546 	|			   |		else complete urb.
1547 	|			   |
1548 	---------------------------
1549  *
1550  * Nuances of mode 1:
1551  *	For short packets, no ack (+RxPktRdy) is sent automatically
1552  *	(even if AutoClear is ON)
1553  *	For full packets, ack (~RxPktRdy) and next IN token (+ReqPkt) is sent
1554  *	automatically => major problem, as collecting the next packet becomes
1555  *	difficult. Hence mode 1 is not used.
1556  *
1557  * REVISIT
1558  *	All we care about at this driver level is that
1559  *       (a) all URBs terminate with REQPKT cleared and fifo(s) empty;
1560  *       (b) termination conditions are: short RX, or buffer full;
1561  *       (c) fault modes include
1562  *           - iff URB_SHORT_NOT_OK, short RX status is -EREMOTEIO.
1563  *             (and that endpoint's dma queue stops immediately)
1564  *           - overflow (full, PLUS more bytes in the terminal packet)
1565  *
1566  *	So for example, usb-storage sets URB_SHORT_NOT_OK, and would
1567  *	thus be a great candidate for using mode 1 ... for all but the
1568  *	last packet of one URB's transfer.
1569  */
1570 static int musb_rx_dma_inventra_cppi41(struct dma_controller *dma,
1571 				       struct musb_hw_ep *hw_ep,
1572 				       struct musb_qh *qh,
1573 				       struct urb *urb,
1574 				       size_t len)
1575 {
1576 	struct dma_channel *channel = hw_ep->rx_channel;
1577 	void __iomem *epio = hw_ep->regs;
1578 	u16 val;
1579 	int pipe;
1580 	bool done;
1581 
1582 	pipe = urb->pipe;
1583 
1584 	if (usb_pipeisoc(pipe)) {
1585 		struct usb_iso_packet_descriptor *d;
1586 
1587 		d = urb->iso_frame_desc + qh->iso_idx;
1588 		d->actual_length = len;
1589 
1590 		/* even if there was an error, we did the dma
1591 		 * for iso_frame_desc->length
1592 		 */
1593 		if (d->status != -EILSEQ && d->status != -EOVERFLOW)
1594 			d->status = 0;
1595 
1596 		if (++qh->iso_idx >= urb->number_of_packets) {
1597 			done = true;
1598 		} else {
1599 			/* REVISIT: Why ignore return value here? */
1600 			if (musb_dma_cppi41(hw_ep->musb))
1601 				done = musb_rx_dma_iso_cppi41(dma, hw_ep, qh,
1602 							      urb, len);
1603 			done = false;
1604 		}
1605 
1606 	} else  {
1607 		/* done if urb buffer is full or short packet is recd */
1608 		done = (urb->actual_length + len >=
1609 			urb->transfer_buffer_length
1610 			|| channel->actual_len < qh->maxpacket
1611 			|| channel->rx_packet_done);
1612 	}
1613 
1614 	/* send IN token for next packet, without AUTOREQ */
1615 	if (!done) {
1616 		val = musb_readw(epio, MUSB_RXCSR);
1617 		val |= MUSB_RXCSR_H_REQPKT;
1618 		musb_writew(epio, MUSB_RXCSR, MUSB_RXCSR_H_WZC_BITS | val);
1619 	}
1620 
1621 	return done;
1622 }
1623 
1624 /* Disadvantage of using mode 1:
1625  *	It's basically usable only for mass storage class; essentially all
1626  *	other protocols also terminate transfers on short packets.
1627  *
1628  * Details:
1629  *	An extra IN token is sent at the end of the transfer (due to AUTOREQ)
1630  *	If you try to use mode 1 for (transfer_buffer_length - 512), and try
1631  *	to use the extra IN token to grab the last packet using mode 0, then
1632  *	the problem is that you cannot be sure when the device will send the
1633  *	last packet and RxPktRdy set. Sometimes the packet is recd too soon
1634  *	such that it gets lost when RxCSR is re-set at the end of the mode 1
1635  *	transfer, while sometimes it is recd just a little late so that if you
1636  *	try to configure for mode 0 soon after the mode 1 transfer is
1637  *	completed, you will find rxcount 0. Okay, so you might think why not
1638  *	wait for an interrupt when the pkt is recd. Well, you won't get any!
1639  */
1640 static int musb_rx_dma_in_inventra_cppi41(struct dma_controller *dma,
1641 					  struct musb_hw_ep *hw_ep,
1642 					  struct musb_qh *qh,
1643 					  struct urb *urb,
1644 					  size_t len,
1645 					  u8 iso_err)
1646 {
1647 	struct musb *musb = hw_ep->musb;
1648 	void __iomem *epio = hw_ep->regs;
1649 	struct dma_channel *channel = hw_ep->rx_channel;
1650 	u16 rx_count, val;
1651 	int length, pipe, done;
1652 	dma_addr_t buf;
1653 
1654 	rx_count = musb_readw(epio, MUSB_RXCOUNT);
1655 	pipe = urb->pipe;
1656 
1657 	if (usb_pipeisoc(pipe)) {
1658 		int d_status = 0;
1659 		struct usb_iso_packet_descriptor *d;
1660 
1661 		d = urb->iso_frame_desc + qh->iso_idx;
1662 
1663 		if (iso_err) {
1664 			d_status = -EILSEQ;
1665 			urb->error_count++;
1666 		}
1667 		if (rx_count > d->length) {
1668 			if (d_status == 0) {
1669 				d_status = -EOVERFLOW;
1670 				urb->error_count++;
1671 			}
1672 			musb_dbg(musb, "** OVERFLOW %d into %d",
1673 				rx_count, d->length);
1674 
1675 			length = d->length;
1676 		} else
1677 			length = rx_count;
1678 		d->status = d_status;
1679 		buf = urb->transfer_dma + d->offset;
1680 	} else {
1681 		length = rx_count;
1682 		buf = urb->transfer_dma + urb->actual_length;
1683 	}
1684 
1685 	channel->desired_mode = 0;
1686 #ifdef USE_MODE1
1687 	/* because of the issue below, mode 1 will
1688 	 * only rarely behave with correct semantics.
1689 	 */
1690 	if ((urb->transfer_flags & URB_SHORT_NOT_OK)
1691 	    && (urb->transfer_buffer_length - urb->actual_length)
1692 	    > qh->maxpacket)
1693 		channel->desired_mode = 1;
1694 	if (rx_count < hw_ep->max_packet_sz_rx) {
1695 		length = rx_count;
1696 		channel->desired_mode = 0;
1697 	} else {
1698 		length = urb->transfer_buffer_length;
1699 	}
1700 #endif
1701 
1702 	/* See comments above on disadvantages of using mode 1 */
1703 	val = musb_readw(epio, MUSB_RXCSR);
1704 	val &= ~MUSB_RXCSR_H_REQPKT;
1705 
1706 	if (channel->desired_mode == 0)
1707 		val &= ~MUSB_RXCSR_H_AUTOREQ;
1708 	else
1709 		val |= MUSB_RXCSR_H_AUTOREQ;
1710 	val |= MUSB_RXCSR_DMAENAB;
1711 
1712 	/* autoclear shouldn't be set in high bandwidth */
1713 	if (qh->hb_mult == 1)
1714 		val |= MUSB_RXCSR_AUTOCLEAR;
1715 
1716 	musb_writew(epio, MUSB_RXCSR, MUSB_RXCSR_H_WZC_BITS | val);
1717 
1718 	/* REVISIT if when actual_length != 0,
1719 	 * transfer_buffer_length needs to be
1720 	 * adjusted first...
1721 	 */
1722 	done = dma->channel_program(channel, qh->maxpacket,
1723 				   channel->desired_mode,
1724 				   buf, length);
1725 
1726 	if (!done) {
1727 		dma->channel_release(channel);
1728 		hw_ep->rx_channel = NULL;
1729 		channel = NULL;
1730 		val = musb_readw(epio, MUSB_RXCSR);
1731 		val &= ~(MUSB_RXCSR_DMAENAB
1732 			 | MUSB_RXCSR_H_AUTOREQ
1733 			 | MUSB_RXCSR_AUTOCLEAR);
1734 		musb_writew(epio, MUSB_RXCSR, val);
1735 	}
1736 
1737 	return done;
1738 }
1739 #else
1740 static inline int musb_rx_dma_inventra_cppi41(struct dma_controller *dma,
1741 					      struct musb_hw_ep *hw_ep,
1742 					      struct musb_qh *qh,
1743 					      struct urb *urb,
1744 					      size_t len)
1745 {
1746 	return false;
1747 }
1748 
1749 static inline int musb_rx_dma_in_inventra_cppi41(struct dma_controller *dma,
1750 						 struct musb_hw_ep *hw_ep,
1751 						 struct musb_qh *qh,
1752 						 struct urb *urb,
1753 						 size_t len,
1754 						 u8 iso_err)
1755 {
1756 	return false;
1757 }
1758 #endif
1759 
1760 /*
1761  * Service an RX interrupt for the given IN endpoint; docs cover bulk, iso,
1762  * and high-bandwidth IN transfer cases.
1763  */
1764 void musb_host_rx(struct musb *musb, u8 epnum)
1765 {
1766 	struct urb		*urb;
1767 	struct musb_hw_ep	*hw_ep = musb->endpoints + epnum;
1768 	struct dma_controller	*c = musb->dma_controller;
1769 	void __iomem		*epio = hw_ep->regs;
1770 	struct musb_qh		*qh = hw_ep->in_qh;
1771 	size_t			xfer_len;
1772 	void __iomem		*mbase = musb->mregs;
1773 	u16			rx_csr, val;
1774 	bool			iso_err = false;
1775 	bool			done = false;
1776 	u32			status;
1777 	struct dma_channel	*dma;
1778 	unsigned int sg_flags = SG_MITER_ATOMIC | SG_MITER_TO_SG;
1779 
1780 	musb_ep_select(mbase, epnum);
1781 
1782 	urb = next_urb(qh);
1783 	dma = is_dma_capable() ? hw_ep->rx_channel : NULL;
1784 	status = 0;
1785 	xfer_len = 0;
1786 
1787 	rx_csr = musb_readw(epio, MUSB_RXCSR);
1788 	val = rx_csr;
1789 
1790 	if (unlikely(!urb)) {
1791 		/* REVISIT -- THIS SHOULD NEVER HAPPEN ... but, at least
1792 		 * usbtest #11 (unlinks) triggers it regularly, sometimes
1793 		 * with fifo full.  (Only with DMA??)
1794 		 */
1795 		musb_dbg(musb, "BOGUS RX%d ready, csr %04x, count %d",
1796 			epnum, val, musb_readw(epio, MUSB_RXCOUNT));
1797 		musb_h_flush_rxfifo(hw_ep, MUSB_RXCSR_CLRDATATOG);
1798 		return;
1799 	}
1800 
1801 	trace_musb_urb_rx(musb, urb);
1802 
1803 	/* check for errors, concurrent stall & unlink is not really
1804 	 * handled yet! */
1805 	if (rx_csr & MUSB_RXCSR_H_RXSTALL) {
1806 		musb_dbg(musb, "RX end %d STALL", epnum);
1807 
1808 		/* stall; record URB status */
1809 		status = -EPIPE;
1810 
1811 	} else if (rx_csr & MUSB_RXCSR_H_ERROR) {
1812 		musb_dbg(musb, "end %d RX proto error", epnum);
1813 
1814 		status = -EPROTO;
1815 		musb_writeb(epio, MUSB_RXINTERVAL, 0);
1816 
1817 		rx_csr &= ~MUSB_RXCSR_H_ERROR;
1818 		musb_writew(epio, MUSB_RXCSR, rx_csr);
1819 
1820 	} else if (rx_csr & MUSB_RXCSR_DATAERROR) {
1821 
1822 		if (USB_ENDPOINT_XFER_ISOC != qh->type) {
1823 			musb_dbg(musb, "RX end %d NAK timeout", epnum);
1824 
1825 			/* NOTE: NAKing is *NOT* an error, so we want to
1826 			 * continue.  Except ... if there's a request for
1827 			 * another QH, use that instead of starving it.
1828 			 *
1829 			 * Devices like Ethernet and serial adapters keep
1830 			 * reads posted at all times, which will starve
1831 			 * other devices without this logic.
1832 			 */
1833 			if (usb_pipebulk(urb->pipe)
1834 					&& qh->mux == 1
1835 					&& !list_is_singular(&musb->in_bulk)) {
1836 				musb_bulk_nak_timeout(musb, hw_ep, 1);
1837 				return;
1838 			}
1839 			musb_ep_select(mbase, epnum);
1840 			rx_csr |= MUSB_RXCSR_H_WZC_BITS;
1841 			rx_csr &= ~MUSB_RXCSR_DATAERROR;
1842 			musb_writew(epio, MUSB_RXCSR, rx_csr);
1843 
1844 			goto finish;
1845 		} else {
1846 			musb_dbg(musb, "RX end %d ISO data error", epnum);
1847 			/* packet error reported later */
1848 			iso_err = true;
1849 		}
1850 	} else if (rx_csr & MUSB_RXCSR_INCOMPRX) {
1851 		musb_dbg(musb, "end %d high bandwidth incomplete ISO packet RX",
1852 				epnum);
1853 		status = -EPROTO;
1854 	}
1855 
1856 	/* faults abort the transfer */
1857 	if (status) {
1858 		/* clean up dma and collect transfer count */
1859 		if (dma_channel_status(dma) == MUSB_DMA_STATUS_BUSY) {
1860 			dma->status = MUSB_DMA_STATUS_CORE_ABORT;
1861 			musb->dma_controller->channel_abort(dma);
1862 			xfer_len = dma->actual_len;
1863 		}
1864 		musb_h_flush_rxfifo(hw_ep, MUSB_RXCSR_CLRDATATOG);
1865 		musb_writeb(epio, MUSB_RXINTERVAL, 0);
1866 		done = true;
1867 		goto finish;
1868 	}
1869 
1870 	if (unlikely(dma_channel_status(dma) == MUSB_DMA_STATUS_BUSY)) {
1871 		/* SHOULD NEVER HAPPEN ... but at least DaVinci has done it */
1872 		ERR("RX%d dma busy, csr %04x\n", epnum, rx_csr);
1873 		goto finish;
1874 	}
1875 
1876 	/* thorough shutdown for now ... given more precise fault handling
1877 	 * and better queueing support, we might keep a DMA pipeline going
1878 	 * while processing this irq for earlier completions.
1879 	 */
1880 
1881 	/* FIXME this is _way_ too much in-line logic for Mentor DMA */
1882 	if (!musb_dma_inventra(musb) && !musb_dma_ux500(musb) &&
1883 	    (rx_csr & MUSB_RXCSR_H_REQPKT)) {
1884 		/* REVISIT this happened for a while on some short reads...
1885 		 * the cleanup still needs investigation... looks bad...
1886 		 * and also duplicates dma cleanup code above ... plus,
1887 		 * shouldn't this be the "half full" double buffer case?
1888 		 */
1889 		if (dma_channel_status(dma) == MUSB_DMA_STATUS_BUSY) {
1890 			dma->status = MUSB_DMA_STATUS_CORE_ABORT;
1891 			musb->dma_controller->channel_abort(dma);
1892 			xfer_len = dma->actual_len;
1893 			done = true;
1894 		}
1895 
1896 		musb_dbg(musb, "RXCSR%d %04x, reqpkt, len %zu%s", epnum, rx_csr,
1897 				xfer_len, dma ? ", dma" : "");
1898 		rx_csr &= ~MUSB_RXCSR_H_REQPKT;
1899 
1900 		musb_ep_select(mbase, epnum);
1901 		musb_writew(epio, MUSB_RXCSR,
1902 				MUSB_RXCSR_H_WZC_BITS | rx_csr);
1903 	}
1904 
1905 	if (dma && (rx_csr & MUSB_RXCSR_DMAENAB)) {
1906 		xfer_len = dma->actual_len;
1907 
1908 		val &= ~(MUSB_RXCSR_DMAENAB
1909 			| MUSB_RXCSR_H_AUTOREQ
1910 			| MUSB_RXCSR_AUTOCLEAR
1911 			| MUSB_RXCSR_RXPKTRDY);
1912 		musb_writew(hw_ep->regs, MUSB_RXCSR, val);
1913 
1914 		if (musb_dma_inventra(musb) || musb_dma_ux500(musb) ||
1915 		    musb_dma_cppi41(musb)) {
1916 			    done = musb_rx_dma_inventra_cppi41(c, hw_ep, qh, urb, xfer_len);
1917 			    musb_dbg(hw_ep->musb,
1918 				    "ep %d dma %s, rxcsr %04x, rxcount %d",
1919 				    epnum, done ? "off" : "reset",
1920 				    musb_readw(epio, MUSB_RXCSR),
1921 				    musb_readw(epio, MUSB_RXCOUNT));
1922 		} else {
1923 			done = true;
1924 		}
1925 
1926 	} else if (urb->status == -EINPROGRESS) {
1927 		/* if no errors, be sure a packet is ready for unloading */
1928 		if (unlikely(!(rx_csr & MUSB_RXCSR_RXPKTRDY))) {
1929 			status = -EPROTO;
1930 			ERR("Rx interrupt with no errors or packet!\n");
1931 
1932 			/* FIXME this is another "SHOULD NEVER HAPPEN" */
1933 
1934 /* SCRUB (RX) */
1935 			/* do the proper sequence to abort the transfer */
1936 			musb_ep_select(mbase, epnum);
1937 			val &= ~MUSB_RXCSR_H_REQPKT;
1938 			musb_writew(epio, MUSB_RXCSR, val);
1939 			goto finish;
1940 		}
1941 
1942 		/* we are expecting IN packets */
1943 		if ((musb_dma_inventra(musb) || musb_dma_ux500(musb) ||
1944 		    musb_dma_cppi41(musb)) && dma) {
1945 			musb_dbg(hw_ep->musb,
1946 				"RX%d count %d, buffer 0x%llx len %d/%d",
1947 				epnum, musb_readw(epio, MUSB_RXCOUNT),
1948 				(unsigned long long) urb->transfer_dma
1949 				+ urb->actual_length,
1950 				qh->offset,
1951 				urb->transfer_buffer_length);
1952 
1953 			if (musb_rx_dma_in_inventra_cppi41(c, hw_ep, qh, urb,
1954 							   xfer_len, iso_err))
1955 				goto finish;
1956 			else
1957 				dev_err(musb->controller, "error: rx_dma failed\n");
1958 		}
1959 
1960 		if (!dma) {
1961 			unsigned int received_len;
1962 
1963 			/* Unmap the buffer so that CPU can use it */
1964 			usb_hcd_unmap_urb_for_dma(musb->hcd, urb);
1965 
1966 			/*
1967 			 * We need to map sg if the transfer_buffer is
1968 			 * NULL.
1969 			 */
1970 			if (!urb->transfer_buffer) {
1971 				qh->use_sg = true;
1972 				sg_miter_start(&qh->sg_miter, urb->sg, 1,
1973 						sg_flags);
1974 			}
1975 
1976 			if (qh->use_sg) {
1977 				if (!sg_miter_next(&qh->sg_miter)) {
1978 					dev_err(musb->controller, "error: sg list empty\n");
1979 					sg_miter_stop(&qh->sg_miter);
1980 					status = -EINVAL;
1981 					done = true;
1982 					goto finish;
1983 				}
1984 				urb->transfer_buffer = qh->sg_miter.addr;
1985 				received_len = urb->actual_length;
1986 				qh->offset = 0x0;
1987 				done = musb_host_packet_rx(musb, urb, epnum,
1988 						iso_err);
1989 				/* Calculate the number of bytes received */
1990 				received_len = urb->actual_length -
1991 					received_len;
1992 				qh->sg_miter.consumed = received_len;
1993 				sg_miter_stop(&qh->sg_miter);
1994 			} else {
1995 				done = musb_host_packet_rx(musb, urb,
1996 						epnum, iso_err);
1997 			}
1998 			musb_dbg(musb, "read %spacket", done ? "last " : "");
1999 		}
2000 	}
2001 
2002 finish:
2003 	urb->actual_length += xfer_len;
2004 	qh->offset += xfer_len;
2005 	if (done) {
2006 		if (qh->use_sg)
2007 			qh->use_sg = false;
2008 
2009 		if (urb->status == -EINPROGRESS)
2010 			urb->status = status;
2011 		musb_advance_schedule(musb, urb, hw_ep, USB_DIR_IN);
2012 	}
2013 }
2014 
2015 /* schedule nodes correspond to peripheral endpoints, like an OHCI QH.
2016  * the software schedule associates multiple such nodes with a given
2017  * host side hardware endpoint + direction; scheduling may activate
2018  * that hardware endpoint.
2019  */
2020 static int musb_schedule(
2021 	struct musb		*musb,
2022 	struct musb_qh		*qh,
2023 	int			is_in)
2024 {
2025 	int			idle = 0;
2026 	int			best_diff;
2027 	int			best_end, epnum;
2028 	struct musb_hw_ep	*hw_ep = NULL;
2029 	struct list_head	*head = NULL;
2030 	u8			toggle;
2031 	u8			txtype;
2032 	struct urb		*urb = next_urb(qh);
2033 
2034 	/* use fixed hardware for control and bulk */
2035 	if (qh->type == USB_ENDPOINT_XFER_CONTROL) {
2036 		head = &musb->control;
2037 		hw_ep = musb->control_ep;
2038 		goto success;
2039 	}
2040 
2041 	/* else, periodic transfers get muxed to other endpoints */
2042 
2043 	/*
2044 	 * We know this qh hasn't been scheduled, so all we need to do
2045 	 * is choose which hardware endpoint to put it on ...
2046 	 *
2047 	 * REVISIT what we really want here is a regular schedule tree
2048 	 * like e.g. OHCI uses.
2049 	 */
2050 	best_diff = 4096;
2051 	best_end = -1;
2052 
2053 	for (epnum = 1, hw_ep = musb->endpoints + 1;
2054 			epnum < musb->nr_endpoints;
2055 			epnum++, hw_ep++) {
2056 		int	diff;
2057 
2058 		if (musb_ep_get_qh(hw_ep, is_in) != NULL)
2059 			continue;
2060 
2061 		if (hw_ep == musb->bulk_ep)
2062 			continue;
2063 
2064 		if (is_in)
2065 			diff = hw_ep->max_packet_sz_rx;
2066 		else
2067 			diff = hw_ep->max_packet_sz_tx;
2068 		diff -= (qh->maxpacket * qh->hb_mult);
2069 
2070 		if (diff >= 0 && best_diff > diff) {
2071 
2072 			/*
2073 			 * Mentor controller has a bug in that if we schedule
2074 			 * a BULK Tx transfer on an endpoint that had earlier
2075 			 * handled ISOC then the BULK transfer has to start on
2076 			 * a zero toggle.  If the BULK transfer starts on a 1
2077 			 * toggle then this transfer will fail as the mentor
2078 			 * controller starts the Bulk transfer on a 0 toggle
2079 			 * irrespective of the programming of the toggle bits
2080 			 * in the TXCSR register.  Check for this condition
2081 			 * while allocating the EP for a Tx Bulk transfer.  If
2082 			 * so skip this EP.
2083 			 */
2084 			hw_ep = musb->endpoints + epnum;
2085 			toggle = usb_gettoggle(urb->dev, qh->epnum, !is_in);
2086 			txtype = (musb_readb(hw_ep->regs, MUSB_TXTYPE)
2087 					>> 4) & 0x3;
2088 			if (!is_in && (qh->type == USB_ENDPOINT_XFER_BULK) &&
2089 				toggle && (txtype == USB_ENDPOINT_XFER_ISOC))
2090 				continue;
2091 
2092 			best_diff = diff;
2093 			best_end = epnum;
2094 		}
2095 	}
2096 	/* use bulk reserved ep1 if no other ep is free */
2097 	if (best_end < 0 && qh->type == USB_ENDPOINT_XFER_BULK) {
2098 		hw_ep = musb->bulk_ep;
2099 		if (is_in)
2100 			head = &musb->in_bulk;
2101 		else
2102 			head = &musb->out_bulk;
2103 
2104 		/* Enable bulk RX/TX NAK timeout scheme when bulk requests are
2105 		 * multiplexed. This scheme does not work in high speed to full
2106 		 * speed scenario as NAK interrupts are not coming from a
2107 		 * full speed device connected to a high speed device.
2108 		 * NAK timeout interval is 8 (128 uframe or 16ms) for HS and
2109 		 * 4 (8 frame or 8ms) for FS device.
2110 		 */
2111 		if (qh->dev)
2112 			qh->intv_reg =
2113 				(USB_SPEED_HIGH == qh->dev->speed) ? 8 : 4;
2114 		goto success;
2115 	} else if (best_end < 0) {
2116 		dev_err(musb->controller,
2117 				"%s hwep alloc failed for %dx%d\n",
2118 				musb_ep_xfertype_string(qh->type),
2119 				qh->hb_mult, qh->maxpacket);
2120 		return -ENOSPC;
2121 	}
2122 
2123 	idle = 1;
2124 	qh->mux = 0;
2125 	hw_ep = musb->endpoints + best_end;
2126 	musb_dbg(musb, "qh %p periodic slot %d", qh, best_end);
2127 success:
2128 	if (head) {
2129 		idle = list_empty(head);
2130 		list_add_tail(&qh->ring, head);
2131 		qh->mux = 1;
2132 	}
2133 	qh->hw_ep = hw_ep;
2134 	qh->hep->hcpriv = qh;
2135 	if (idle)
2136 		musb_start_urb(musb, is_in, qh);
2137 	return 0;
2138 }
2139 
2140 static int musb_urb_enqueue(
2141 	struct usb_hcd			*hcd,
2142 	struct urb			*urb,
2143 	gfp_t				mem_flags)
2144 {
2145 	unsigned long			flags;
2146 	struct musb			*musb = hcd_to_musb(hcd);
2147 	struct usb_host_endpoint	*hep = urb->ep;
2148 	struct musb_qh			*qh;
2149 	struct usb_endpoint_descriptor	*epd = &hep->desc;
2150 	int				ret;
2151 	unsigned			type_reg;
2152 	unsigned			interval;
2153 
2154 	/* host role must be active */
2155 	if (!is_host_active(musb) || !musb->is_active)
2156 		return -ENODEV;
2157 
2158 	trace_musb_urb_enq(musb, urb);
2159 
2160 	spin_lock_irqsave(&musb->lock, flags);
2161 	ret = usb_hcd_link_urb_to_ep(hcd, urb);
2162 	qh = ret ? NULL : hep->hcpriv;
2163 	if (qh)
2164 		urb->hcpriv = qh;
2165 	spin_unlock_irqrestore(&musb->lock, flags);
2166 
2167 	/* DMA mapping was already done, if needed, and this urb is on
2168 	 * hep->urb_list now ... so we're done, unless hep wasn't yet
2169 	 * scheduled onto a live qh.
2170 	 *
2171 	 * REVISIT best to keep hep->hcpriv valid until the endpoint gets
2172 	 * disabled, testing for empty qh->ring and avoiding qh setup costs
2173 	 * except for the first urb queued after a config change.
2174 	 */
2175 	if (qh || ret)
2176 		return ret;
2177 
2178 	/* Allocate and initialize qh, minimizing the work done each time
2179 	 * hw_ep gets reprogrammed, or with irqs blocked.  Then schedule it.
2180 	 *
2181 	 * REVISIT consider a dedicated qh kmem_cache, so it's harder
2182 	 * for bugs in other kernel code to break this driver...
2183 	 */
2184 	qh = kzalloc(sizeof *qh, mem_flags);
2185 	if (!qh) {
2186 		spin_lock_irqsave(&musb->lock, flags);
2187 		usb_hcd_unlink_urb_from_ep(hcd, urb);
2188 		spin_unlock_irqrestore(&musb->lock, flags);
2189 		return -ENOMEM;
2190 	}
2191 
2192 	qh->hep = hep;
2193 	qh->dev = urb->dev;
2194 	INIT_LIST_HEAD(&qh->ring);
2195 	qh->is_ready = 1;
2196 
2197 	qh->maxpacket = usb_endpoint_maxp(epd);
2198 	qh->type = usb_endpoint_type(epd);
2199 
2200 	/* Bits 11 & 12 of wMaxPacketSize encode high bandwidth multiplier.
2201 	 * Some musb cores don't support high bandwidth ISO transfers; and
2202 	 * we don't (yet!) support high bandwidth interrupt transfers.
2203 	 */
2204 	qh->hb_mult = usb_endpoint_maxp_mult(epd);
2205 	if (qh->hb_mult > 1) {
2206 		int ok = (qh->type == USB_ENDPOINT_XFER_ISOC);
2207 
2208 		if (ok)
2209 			ok = (usb_pipein(urb->pipe) && musb->hb_iso_rx)
2210 				|| (usb_pipeout(urb->pipe) && musb->hb_iso_tx);
2211 		if (!ok) {
2212 			dev_err(musb->controller,
2213 				"high bandwidth %s (%dx%d) not supported\n",
2214 				musb_ep_xfertype_string(qh->type),
2215 				qh->hb_mult, qh->maxpacket & 0x7ff);
2216 			ret = -EMSGSIZE;
2217 			goto done;
2218 		}
2219 		qh->maxpacket &= 0x7ff;
2220 	}
2221 
2222 	qh->epnum = usb_endpoint_num(epd);
2223 
2224 	/* NOTE: urb->dev->devnum is wrong during SET_ADDRESS */
2225 	qh->addr_reg = (u8) usb_pipedevice(urb->pipe);
2226 
2227 	/* precompute rxtype/txtype/type0 register */
2228 	type_reg = (qh->type << 4) | qh->epnum;
2229 	switch (urb->dev->speed) {
2230 	case USB_SPEED_LOW:
2231 		type_reg |= 0xc0;
2232 		break;
2233 	case USB_SPEED_FULL:
2234 		type_reg |= 0x80;
2235 		break;
2236 	default:
2237 		type_reg |= 0x40;
2238 	}
2239 	qh->type_reg = type_reg;
2240 
2241 	/* Precompute RXINTERVAL/TXINTERVAL register */
2242 	switch (qh->type) {
2243 	case USB_ENDPOINT_XFER_INT:
2244 		/*
2245 		 * Full/low speeds use the  linear encoding,
2246 		 * high speed uses the logarithmic encoding.
2247 		 */
2248 		if (urb->dev->speed <= USB_SPEED_FULL) {
2249 			interval = max_t(u8, epd->bInterval, 1);
2250 			break;
2251 		}
2252 		/* FALLTHROUGH */
2253 	case USB_ENDPOINT_XFER_ISOC:
2254 		/* ISO always uses logarithmic encoding */
2255 		interval = min_t(u8, epd->bInterval, 16);
2256 		break;
2257 	default:
2258 		/* REVISIT we actually want to use NAK limits, hinting to the
2259 		 * transfer scheduling logic to try some other qh, e.g. try
2260 		 * for 2 msec first:
2261 		 *
2262 		 * interval = (USB_SPEED_HIGH == urb->dev->speed) ? 16 : 2;
2263 		 *
2264 		 * The downside of disabling this is that transfer scheduling
2265 		 * gets VERY unfair for nonperiodic transfers; a misbehaving
2266 		 * peripheral could make that hurt.  That's perfectly normal
2267 		 * for reads from network or serial adapters ... so we have
2268 		 * partial NAKlimit support for bulk RX.
2269 		 *
2270 		 * The upside of disabling it is simpler transfer scheduling.
2271 		 */
2272 		interval = 0;
2273 	}
2274 	qh->intv_reg = interval;
2275 
2276 	/* precompute addressing for external hub/tt ports */
2277 	if (musb->is_multipoint) {
2278 		struct usb_device	*parent = urb->dev->parent;
2279 
2280 		if (parent != hcd->self.root_hub) {
2281 			qh->h_addr_reg = (u8) parent->devnum;
2282 
2283 			/* set up tt info if needed */
2284 			if (urb->dev->tt) {
2285 				qh->h_port_reg = (u8) urb->dev->ttport;
2286 				if (urb->dev->tt->hub)
2287 					qh->h_addr_reg =
2288 						(u8) urb->dev->tt->hub->devnum;
2289 				if (urb->dev->tt->multi)
2290 					qh->h_addr_reg |= 0x80;
2291 			}
2292 		}
2293 	}
2294 
2295 	/* invariant: hep->hcpriv is null OR the qh that's already scheduled.
2296 	 * until we get real dma queues (with an entry for each urb/buffer),
2297 	 * we only have work to do in the former case.
2298 	 */
2299 	spin_lock_irqsave(&musb->lock, flags);
2300 	if (hep->hcpriv || !next_urb(qh)) {
2301 		/* some concurrent activity submitted another urb to hep...
2302 		 * odd, rare, error prone, but legal.
2303 		 */
2304 		kfree(qh);
2305 		qh = NULL;
2306 		ret = 0;
2307 	} else
2308 		ret = musb_schedule(musb, qh,
2309 				epd->bEndpointAddress & USB_ENDPOINT_DIR_MASK);
2310 
2311 	if (ret == 0) {
2312 		urb->hcpriv = qh;
2313 		/* FIXME set urb->start_frame for iso/intr, it's tested in
2314 		 * musb_start_urb(), but otherwise only konicawc cares ...
2315 		 */
2316 	}
2317 	spin_unlock_irqrestore(&musb->lock, flags);
2318 
2319 done:
2320 	if (ret != 0) {
2321 		spin_lock_irqsave(&musb->lock, flags);
2322 		usb_hcd_unlink_urb_from_ep(hcd, urb);
2323 		spin_unlock_irqrestore(&musb->lock, flags);
2324 		kfree(qh);
2325 	}
2326 	return ret;
2327 }
2328 
2329 
2330 /*
2331  * abort a transfer that's at the head of a hardware queue.
2332  * called with controller locked, irqs blocked
2333  * that hardware queue advances to the next transfer, unless prevented
2334  */
2335 static int musb_cleanup_urb(struct urb *urb, struct musb_qh *qh)
2336 {
2337 	struct musb_hw_ep	*ep = qh->hw_ep;
2338 	struct musb		*musb = ep->musb;
2339 	void __iomem		*epio = ep->regs;
2340 	unsigned		hw_end = ep->epnum;
2341 	void __iomem		*regs = ep->musb->mregs;
2342 	int			is_in = usb_pipein(urb->pipe);
2343 	int			status = 0;
2344 	u16			csr;
2345 	struct dma_channel	*dma = NULL;
2346 
2347 	musb_ep_select(regs, hw_end);
2348 
2349 	if (is_dma_capable()) {
2350 		dma = is_in ? ep->rx_channel : ep->tx_channel;
2351 		if (dma) {
2352 			status = ep->musb->dma_controller->channel_abort(dma);
2353 			musb_dbg(musb, "abort %cX%d DMA for urb %p --> %d",
2354 				is_in ? 'R' : 'T', ep->epnum,
2355 				urb, status);
2356 			urb->actual_length += dma->actual_len;
2357 		}
2358 	}
2359 
2360 	/* turn off DMA requests, discard state, stop polling ... */
2361 	if (ep->epnum && is_in) {
2362 		/* giveback saves bulk toggle */
2363 		csr = musb_h_flush_rxfifo(ep, 0);
2364 
2365 		/* clear the endpoint's irq status here to avoid bogus irqs */
2366 		if (is_dma_capable() && dma)
2367 			musb_platform_clear_ep_rxintr(musb, ep->epnum);
2368 	} else if (ep->epnum) {
2369 		musb_h_tx_flush_fifo(ep);
2370 		csr = musb_readw(epio, MUSB_TXCSR);
2371 		csr &= ~(MUSB_TXCSR_AUTOSET
2372 			| MUSB_TXCSR_DMAENAB
2373 			| MUSB_TXCSR_H_RXSTALL
2374 			| MUSB_TXCSR_H_NAKTIMEOUT
2375 			| MUSB_TXCSR_H_ERROR
2376 			| MUSB_TXCSR_TXPKTRDY);
2377 		musb_writew(epio, MUSB_TXCSR, csr);
2378 		/* REVISIT may need to clear FLUSHFIFO ... */
2379 		musb_writew(epio, MUSB_TXCSR, csr);
2380 		/* flush cpu writebuffer */
2381 		csr = musb_readw(epio, MUSB_TXCSR);
2382 	} else  {
2383 		musb_h_ep0_flush_fifo(ep);
2384 	}
2385 	if (status == 0)
2386 		musb_advance_schedule(ep->musb, urb, ep, is_in);
2387 	return status;
2388 }
2389 
2390 static int musb_urb_dequeue(struct usb_hcd *hcd, struct urb *urb, int status)
2391 {
2392 	struct musb		*musb = hcd_to_musb(hcd);
2393 	struct musb_qh		*qh;
2394 	unsigned long		flags;
2395 	int			is_in  = usb_pipein(urb->pipe);
2396 	int			ret;
2397 
2398 	trace_musb_urb_deq(musb, urb);
2399 
2400 	spin_lock_irqsave(&musb->lock, flags);
2401 	ret = usb_hcd_check_unlink_urb(hcd, urb, status);
2402 	if (ret)
2403 		goto done;
2404 
2405 	qh = urb->hcpriv;
2406 	if (!qh)
2407 		goto done;
2408 
2409 	/*
2410 	 * Any URB not actively programmed into endpoint hardware can be
2411 	 * immediately given back; that's any URB not at the head of an
2412 	 * endpoint queue, unless someday we get real DMA queues.  And even
2413 	 * if it's at the head, it might not be known to the hardware...
2414 	 *
2415 	 * Otherwise abort current transfer, pending DMA, etc.; urb->status
2416 	 * has already been updated.  This is a synchronous abort; it'd be
2417 	 * OK to hold off until after some IRQ, though.
2418 	 *
2419 	 * NOTE: qh is invalid unless !list_empty(&hep->urb_list)
2420 	 */
2421 	if (!qh->is_ready
2422 			|| urb->urb_list.prev != &qh->hep->urb_list
2423 			|| musb_ep_get_qh(qh->hw_ep, is_in) != qh) {
2424 		int	ready = qh->is_ready;
2425 
2426 		qh->is_ready = 0;
2427 		musb_giveback(musb, urb, 0);
2428 		qh->is_ready = ready;
2429 
2430 		/* If nothing else (usually musb_giveback) is using it
2431 		 * and its URB list has emptied, recycle this qh.
2432 		 */
2433 		if (ready && list_empty(&qh->hep->urb_list)) {
2434 			qh->hep->hcpriv = NULL;
2435 			list_del(&qh->ring);
2436 			kfree(qh);
2437 		}
2438 	} else
2439 		ret = musb_cleanup_urb(urb, qh);
2440 done:
2441 	spin_unlock_irqrestore(&musb->lock, flags);
2442 	return ret;
2443 }
2444 
2445 /* disable an endpoint */
2446 static void
2447 musb_h_disable(struct usb_hcd *hcd, struct usb_host_endpoint *hep)
2448 {
2449 	u8			is_in = hep->desc.bEndpointAddress & USB_DIR_IN;
2450 	unsigned long		flags;
2451 	struct musb		*musb = hcd_to_musb(hcd);
2452 	struct musb_qh		*qh;
2453 	struct urb		*urb;
2454 
2455 	spin_lock_irqsave(&musb->lock, flags);
2456 
2457 	qh = hep->hcpriv;
2458 	if (qh == NULL)
2459 		goto exit;
2460 
2461 	/* NOTE: qh is invalid unless !list_empty(&hep->urb_list) */
2462 
2463 	/* Kick the first URB off the hardware, if needed */
2464 	qh->is_ready = 0;
2465 	if (musb_ep_get_qh(qh->hw_ep, is_in) == qh) {
2466 		urb = next_urb(qh);
2467 
2468 		/* make software (then hardware) stop ASAP */
2469 		if (!urb->unlinked)
2470 			urb->status = -ESHUTDOWN;
2471 
2472 		/* cleanup */
2473 		musb_cleanup_urb(urb, qh);
2474 
2475 		/* Then nuke all the others ... and advance the
2476 		 * queue on hw_ep (e.g. bulk ring) when we're done.
2477 		 */
2478 		while (!list_empty(&hep->urb_list)) {
2479 			urb = next_urb(qh);
2480 			urb->status = -ESHUTDOWN;
2481 			musb_advance_schedule(musb, urb, qh->hw_ep, is_in);
2482 		}
2483 	} else {
2484 		/* Just empty the queue; the hardware is busy with
2485 		 * other transfers, and since !qh->is_ready nothing
2486 		 * will activate any of these as it advances.
2487 		 */
2488 		while (!list_empty(&hep->urb_list))
2489 			musb_giveback(musb, next_urb(qh), -ESHUTDOWN);
2490 
2491 		hep->hcpriv = NULL;
2492 		list_del(&qh->ring);
2493 		kfree(qh);
2494 	}
2495 exit:
2496 	spin_unlock_irqrestore(&musb->lock, flags);
2497 }
2498 
2499 static int musb_h_get_frame_number(struct usb_hcd *hcd)
2500 {
2501 	struct musb	*musb = hcd_to_musb(hcd);
2502 
2503 	return musb_readw(musb->mregs, MUSB_FRAME);
2504 }
2505 
2506 static int musb_h_start(struct usb_hcd *hcd)
2507 {
2508 	struct musb	*musb = hcd_to_musb(hcd);
2509 
2510 	/* NOTE: musb_start() is called when the hub driver turns
2511 	 * on port power, or when (OTG) peripheral starts.
2512 	 */
2513 	hcd->state = HC_STATE_RUNNING;
2514 	musb->port1_status = 0;
2515 	return 0;
2516 }
2517 
2518 static void musb_h_stop(struct usb_hcd *hcd)
2519 {
2520 	musb_stop(hcd_to_musb(hcd));
2521 	hcd->state = HC_STATE_HALT;
2522 }
2523 
2524 static int musb_bus_suspend(struct usb_hcd *hcd)
2525 {
2526 	struct musb	*musb = hcd_to_musb(hcd);
2527 	u8		devctl;
2528 	int		ret;
2529 
2530 	ret = musb_port_suspend(musb, true);
2531 	if (ret)
2532 		return ret;
2533 
2534 	if (!is_host_active(musb))
2535 		return 0;
2536 
2537 	switch (musb->xceiv->otg->state) {
2538 	case OTG_STATE_A_SUSPEND:
2539 		return 0;
2540 	case OTG_STATE_A_WAIT_VRISE:
2541 		/* ID could be grounded even if there's no device
2542 		 * on the other end of the cable.  NOTE that the
2543 		 * A_WAIT_VRISE timers are messy with MUSB...
2544 		 */
2545 		devctl = musb_readb(musb->mregs, MUSB_DEVCTL);
2546 		if ((devctl & MUSB_DEVCTL_VBUS) == MUSB_DEVCTL_VBUS)
2547 			musb->xceiv->otg->state = OTG_STATE_A_WAIT_BCON;
2548 		break;
2549 	default:
2550 		break;
2551 	}
2552 
2553 	if (musb->is_active) {
2554 		WARNING("trying to suspend as %s while active\n",
2555 				usb_otg_state_string(musb->xceiv->otg->state));
2556 		return -EBUSY;
2557 	} else
2558 		return 0;
2559 }
2560 
2561 static int musb_bus_resume(struct usb_hcd *hcd)
2562 {
2563 	struct musb *musb = hcd_to_musb(hcd);
2564 
2565 	if (musb->config &&
2566 	    musb->config->host_port_deassert_reset_at_resume)
2567 		musb_port_reset(musb, false);
2568 
2569 	return 0;
2570 }
2571 
2572 #ifndef CONFIG_MUSB_PIO_ONLY
2573 
2574 #define MUSB_USB_DMA_ALIGN 4
2575 
2576 struct musb_temp_buffer {
2577 	void *kmalloc_ptr;
2578 	void *old_xfer_buffer;
2579 	u8 data[0];
2580 };
2581 
2582 static void musb_free_temp_buffer(struct urb *urb)
2583 {
2584 	enum dma_data_direction dir;
2585 	struct musb_temp_buffer *temp;
2586 	size_t length;
2587 
2588 	if (!(urb->transfer_flags & URB_ALIGNED_TEMP_BUFFER))
2589 		return;
2590 
2591 	dir = usb_urb_dir_in(urb) ? DMA_FROM_DEVICE : DMA_TO_DEVICE;
2592 
2593 	temp = container_of(urb->transfer_buffer, struct musb_temp_buffer,
2594 			    data);
2595 
2596 	if (dir == DMA_FROM_DEVICE) {
2597 		if (usb_pipeisoc(urb->pipe))
2598 			length = urb->transfer_buffer_length;
2599 		else
2600 			length = urb->actual_length;
2601 
2602 		memcpy(temp->old_xfer_buffer, temp->data, length);
2603 	}
2604 	urb->transfer_buffer = temp->old_xfer_buffer;
2605 	kfree(temp->kmalloc_ptr);
2606 
2607 	urb->transfer_flags &= ~URB_ALIGNED_TEMP_BUFFER;
2608 }
2609 
2610 static int musb_alloc_temp_buffer(struct urb *urb, gfp_t mem_flags)
2611 {
2612 	enum dma_data_direction dir;
2613 	struct musb_temp_buffer *temp;
2614 	void *kmalloc_ptr;
2615 	size_t kmalloc_size;
2616 
2617 	if (urb->num_sgs || urb->sg ||
2618 	    urb->transfer_buffer_length == 0 ||
2619 	    !((uintptr_t)urb->transfer_buffer & (MUSB_USB_DMA_ALIGN - 1)))
2620 		return 0;
2621 
2622 	dir = usb_urb_dir_in(urb) ? DMA_FROM_DEVICE : DMA_TO_DEVICE;
2623 
2624 	/* Allocate a buffer with enough padding for alignment */
2625 	kmalloc_size = urb->transfer_buffer_length +
2626 		sizeof(struct musb_temp_buffer) + MUSB_USB_DMA_ALIGN - 1;
2627 
2628 	kmalloc_ptr = kmalloc(kmalloc_size, mem_flags);
2629 	if (!kmalloc_ptr)
2630 		return -ENOMEM;
2631 
2632 	/* Position our struct temp_buffer such that data is aligned */
2633 	temp = PTR_ALIGN(kmalloc_ptr, MUSB_USB_DMA_ALIGN);
2634 
2635 
2636 	temp->kmalloc_ptr = kmalloc_ptr;
2637 	temp->old_xfer_buffer = urb->transfer_buffer;
2638 	if (dir == DMA_TO_DEVICE)
2639 		memcpy(temp->data, urb->transfer_buffer,
2640 		       urb->transfer_buffer_length);
2641 	urb->transfer_buffer = temp->data;
2642 
2643 	urb->transfer_flags |= URB_ALIGNED_TEMP_BUFFER;
2644 
2645 	return 0;
2646 }
2647 
2648 static int musb_map_urb_for_dma(struct usb_hcd *hcd, struct urb *urb,
2649 				      gfp_t mem_flags)
2650 {
2651 	struct musb	*musb = hcd_to_musb(hcd);
2652 	int ret;
2653 
2654 	/*
2655 	 * The DMA engine in RTL1.8 and above cannot handle
2656 	 * DMA addresses that are not aligned to a 4 byte boundary.
2657 	 * For such engine implemented (un)map_urb_for_dma hooks.
2658 	 * Do not use these hooks for RTL<1.8
2659 	 */
2660 	if (musb->hwvers < MUSB_HWVERS_1800)
2661 		return usb_hcd_map_urb_for_dma(hcd, urb, mem_flags);
2662 
2663 	ret = musb_alloc_temp_buffer(urb, mem_flags);
2664 	if (ret)
2665 		return ret;
2666 
2667 	ret = usb_hcd_map_urb_for_dma(hcd, urb, mem_flags);
2668 	if (ret)
2669 		musb_free_temp_buffer(urb);
2670 
2671 	return ret;
2672 }
2673 
2674 static void musb_unmap_urb_for_dma(struct usb_hcd *hcd, struct urb *urb)
2675 {
2676 	struct musb	*musb = hcd_to_musb(hcd);
2677 
2678 	usb_hcd_unmap_urb_for_dma(hcd, urb);
2679 
2680 	/* Do not use this hook for RTL<1.8 (see description above) */
2681 	if (musb->hwvers < MUSB_HWVERS_1800)
2682 		return;
2683 
2684 	musb_free_temp_buffer(urb);
2685 }
2686 #endif /* !CONFIG_MUSB_PIO_ONLY */
2687 
2688 static const struct hc_driver musb_hc_driver = {
2689 	.description		= "musb-hcd",
2690 	.product_desc		= "MUSB HDRC host driver",
2691 	.hcd_priv_size		= sizeof(struct musb *),
2692 	.flags			= HCD_USB2 | HCD_MEMORY,
2693 
2694 	/* not using irq handler or reset hooks from usbcore, since
2695 	 * those must be shared with peripheral code for OTG configs
2696 	 */
2697 
2698 	.start			= musb_h_start,
2699 	.stop			= musb_h_stop,
2700 
2701 	.get_frame_number	= musb_h_get_frame_number,
2702 
2703 	.urb_enqueue		= musb_urb_enqueue,
2704 	.urb_dequeue		= musb_urb_dequeue,
2705 	.endpoint_disable	= musb_h_disable,
2706 
2707 #ifndef CONFIG_MUSB_PIO_ONLY
2708 	.map_urb_for_dma	= musb_map_urb_for_dma,
2709 	.unmap_urb_for_dma	= musb_unmap_urb_for_dma,
2710 #endif
2711 
2712 	.hub_status_data	= musb_hub_status_data,
2713 	.hub_control		= musb_hub_control,
2714 	.bus_suspend		= musb_bus_suspend,
2715 	.bus_resume		= musb_bus_resume,
2716 	/* .start_port_reset	= NULL, */
2717 	/* .hub_irq_enable	= NULL, */
2718 };
2719 
2720 int musb_host_alloc(struct musb *musb)
2721 {
2722 	struct device	*dev = musb->controller;
2723 
2724 	/* usbcore sets dev->driver_data to hcd, and sometimes uses that... */
2725 	musb->hcd = usb_create_hcd(&musb_hc_driver, dev, dev_name(dev));
2726 	if (!musb->hcd)
2727 		return -EINVAL;
2728 
2729 	*musb->hcd->hcd_priv = (unsigned long) musb;
2730 	musb->hcd->self.uses_pio_for_control = 1;
2731 	musb->hcd->uses_new_polling = 1;
2732 	musb->hcd->has_tt = 1;
2733 
2734 	return 0;
2735 }
2736 
2737 void musb_host_cleanup(struct musb *musb)
2738 {
2739 	if (musb->port_mode == MUSB_PERIPHERAL)
2740 		return;
2741 	usb_remove_hcd(musb->hcd);
2742 }
2743 
2744 void musb_host_free(struct musb *musb)
2745 {
2746 	usb_put_hcd(musb->hcd);
2747 }
2748 
2749 int musb_host_setup(struct musb *musb, int power_budget)
2750 {
2751 	int ret;
2752 	struct usb_hcd *hcd = musb->hcd;
2753 
2754 	if (musb->port_mode == MUSB_HOST) {
2755 		MUSB_HST_MODE(musb);
2756 		musb->xceiv->otg->state = OTG_STATE_A_IDLE;
2757 	}
2758 	otg_set_host(musb->xceiv->otg, &hcd->self);
2759 	/* don't support otg protocols */
2760 	hcd->self.otg_port = 0;
2761 	musb->xceiv->otg->host = &hcd->self;
2762 	hcd->power_budget = 2 * (power_budget ? : 250);
2763 	hcd->skip_phy_initialization = 1;
2764 
2765 	ret = usb_add_hcd(hcd, 0, 0);
2766 	if (ret < 0)
2767 		return ret;
2768 
2769 	device_wakeup_enable(hcd->self.controller);
2770 	return 0;
2771 }
2772 
2773 void musb_host_resume_root_hub(struct musb *musb)
2774 {
2775 	usb_hcd_resume_root_hub(musb->hcd);
2776 }
2777 
2778 void musb_host_poke_root_hub(struct musb *musb)
2779 {
2780 	MUSB_HST_MODE(musb);
2781 	if (musb->hcd->status_urb)
2782 		usb_hcd_poll_rh_status(musb->hcd);
2783 	else
2784 		usb_hcd_resume_root_hub(musb->hcd);
2785 }
2786