xref: /openbmc/linux/drivers/dma/sa11x0-dma.c (revision b9b77222)
1 /*
2  * SA11x0 DMAengine support
3  *
4  * Copyright (C) 2012 Russell King
5  *   Derived in part from arch/arm/mach-sa1100/dma.c,
6  *   Copyright (C) 2000, 2001 by Nicolas Pitre
7  *
8  * This program is free software; you can redistribute it and/or modify
9  * it under the terms of the GNU General Public License version 2 as
10  * published by the Free Software Foundation.
11  */
12 #include <linux/sched.h>
13 #include <linux/device.h>
14 #include <linux/dmaengine.h>
15 #include <linux/init.h>
16 #include <linux/interrupt.h>
17 #include <linux/kernel.h>
18 #include <linux/module.h>
19 #include <linux/platform_device.h>
20 #include <linux/sa11x0-dma.h>
21 #include <linux/slab.h>
22 #include <linux/spinlock.h>
23 
24 #include "virt-dma.h"
25 
26 #define NR_PHY_CHAN	6
27 #define DMA_ALIGN	3
28 #define DMA_MAX_SIZE	0x1fff
29 #define DMA_CHUNK_SIZE	0x1000
30 
31 #define DMA_DDAR	0x00
32 #define DMA_DCSR_S	0x04
33 #define DMA_DCSR_C	0x08
34 #define DMA_DCSR_R	0x0c
35 #define DMA_DBSA	0x10
36 #define DMA_DBTA	0x14
37 #define DMA_DBSB	0x18
38 #define DMA_DBTB	0x1c
39 #define DMA_SIZE	0x20
40 
41 #define DCSR_RUN	(1 << 0)
42 #define DCSR_IE		(1 << 1)
43 #define DCSR_ERROR	(1 << 2)
44 #define DCSR_DONEA	(1 << 3)
45 #define DCSR_STRTA	(1 << 4)
46 #define DCSR_DONEB	(1 << 5)
47 #define DCSR_STRTB	(1 << 6)
48 #define DCSR_BIU	(1 << 7)
49 
50 #define DDAR_RW		(1 << 0)	/* 0 = W, 1 = R */
51 #define DDAR_E		(1 << 1)	/* 0 = LE, 1 = BE */
52 #define DDAR_BS		(1 << 2)	/* 0 = BS4, 1 = BS8 */
53 #define DDAR_DW		(1 << 3)	/* 0 = 8b, 1 = 16b */
54 #define DDAR_Ser0UDCTr	(0x0 << 4)
55 #define DDAR_Ser0UDCRc	(0x1 << 4)
56 #define DDAR_Ser1SDLCTr	(0x2 << 4)
57 #define DDAR_Ser1SDLCRc	(0x3 << 4)
58 #define DDAR_Ser1UARTTr	(0x4 << 4)
59 #define DDAR_Ser1UARTRc	(0x5 << 4)
60 #define DDAR_Ser2ICPTr	(0x6 << 4)
61 #define DDAR_Ser2ICPRc	(0x7 << 4)
62 #define DDAR_Ser3UARTTr	(0x8 << 4)
63 #define DDAR_Ser3UARTRc	(0x9 << 4)
64 #define DDAR_Ser4MCP0Tr	(0xa << 4)
65 #define DDAR_Ser4MCP0Rc	(0xb << 4)
66 #define DDAR_Ser4MCP1Tr	(0xc << 4)
67 #define DDAR_Ser4MCP1Rc	(0xd << 4)
68 #define DDAR_Ser4SSPTr	(0xe << 4)
69 #define DDAR_Ser4SSPRc	(0xf << 4)
70 
71 struct sa11x0_dma_sg {
72 	u32			addr;
73 	u32			len;
74 };
75 
76 struct sa11x0_dma_desc {
77 	struct virt_dma_desc	vd;
78 
79 	u32			ddar;
80 	size_t			size;
81 	unsigned		period;
82 	bool			cyclic;
83 
84 	unsigned		sglen;
85 	struct sa11x0_dma_sg	sg[0];
86 };
87 
88 struct sa11x0_dma_phy;
89 
90 struct sa11x0_dma_chan {
91 	struct virt_dma_chan	vc;
92 
93 	/* protected by c->vc.lock */
94 	struct sa11x0_dma_phy	*phy;
95 	enum dma_status		status;
96 
97 	/* protected by d->lock */
98 	struct list_head	node;
99 
100 	u32			ddar;
101 	const char		*name;
102 };
103 
104 struct sa11x0_dma_phy {
105 	void __iomem		*base;
106 	struct sa11x0_dma_dev	*dev;
107 	unsigned		num;
108 
109 	struct sa11x0_dma_chan	*vchan;
110 
111 	/* Protected by c->vc.lock */
112 	unsigned		sg_load;
113 	struct sa11x0_dma_desc	*txd_load;
114 	unsigned		sg_done;
115 	struct sa11x0_dma_desc	*txd_done;
116 	u32			dbs[2];
117 	u32			dbt[2];
118 	u32			dcsr;
119 };
120 
121 struct sa11x0_dma_dev {
122 	struct dma_device	slave;
123 	void __iomem		*base;
124 	spinlock_t		lock;
125 	struct tasklet_struct	task;
126 	struct list_head	chan_pending;
127 	struct sa11x0_dma_phy	phy[NR_PHY_CHAN];
128 };
129 
130 static struct sa11x0_dma_chan *to_sa11x0_dma_chan(struct dma_chan *chan)
131 {
132 	return container_of(chan, struct sa11x0_dma_chan, vc.chan);
133 }
134 
135 static struct sa11x0_dma_dev *to_sa11x0_dma(struct dma_device *dmadev)
136 {
137 	return container_of(dmadev, struct sa11x0_dma_dev, slave);
138 }
139 
140 static struct sa11x0_dma_desc *sa11x0_dma_next_desc(struct sa11x0_dma_chan *c)
141 {
142 	struct virt_dma_desc *vd = vchan_next_desc(&c->vc);
143 
144 	return vd ? container_of(vd, struct sa11x0_dma_desc, vd) : NULL;
145 }
146 
147 static void sa11x0_dma_free_desc(struct virt_dma_desc *vd)
148 {
149 	kfree(container_of(vd, struct sa11x0_dma_desc, vd));
150 }
151 
152 static void sa11x0_dma_start_desc(struct sa11x0_dma_phy *p, struct sa11x0_dma_desc *txd)
153 {
154 	list_del(&txd->vd.node);
155 	p->txd_load = txd;
156 	p->sg_load = 0;
157 
158 	dev_vdbg(p->dev->slave.dev, "pchan %u: txd %p[%x]: starting: DDAR:%x\n",
159 		p->num, &txd->vd, txd->vd.tx.cookie, txd->ddar);
160 }
161 
162 static void noinline sa11x0_dma_start_sg(struct sa11x0_dma_phy *p,
163 	struct sa11x0_dma_chan *c)
164 {
165 	struct sa11x0_dma_desc *txd = p->txd_load;
166 	struct sa11x0_dma_sg *sg;
167 	void __iomem *base = p->base;
168 	unsigned dbsx, dbtx;
169 	u32 dcsr;
170 
171 	if (!txd)
172 		return;
173 
174 	dcsr = readl_relaxed(base + DMA_DCSR_R);
175 
176 	/* Don't try to load the next transfer if both buffers are started */
177 	if ((dcsr & (DCSR_STRTA | DCSR_STRTB)) == (DCSR_STRTA | DCSR_STRTB))
178 		return;
179 
180 	if (p->sg_load == txd->sglen) {
181 		if (!txd->cyclic) {
182 			struct sa11x0_dma_desc *txn = sa11x0_dma_next_desc(c);
183 
184 			/*
185 			 * We have reached the end of the current descriptor.
186 			 * Peek at the next descriptor, and if compatible with
187 			 * the current, start processing it.
188 			 */
189 			if (txn && txn->ddar == txd->ddar) {
190 				txd = txn;
191 				sa11x0_dma_start_desc(p, txn);
192 			} else {
193 				p->txd_load = NULL;
194 				return;
195 			}
196 		} else {
197 			/* Cyclic: reset back to beginning */
198 			p->sg_load = 0;
199 		}
200 	}
201 
202 	sg = &txd->sg[p->sg_load++];
203 
204 	/* Select buffer to load according to channel status */
205 	if (((dcsr & (DCSR_BIU | DCSR_STRTB)) == (DCSR_BIU | DCSR_STRTB)) ||
206 	    ((dcsr & (DCSR_BIU | DCSR_STRTA)) == 0)) {
207 		dbsx = DMA_DBSA;
208 		dbtx = DMA_DBTA;
209 		dcsr = DCSR_STRTA | DCSR_IE | DCSR_RUN;
210 	} else {
211 		dbsx = DMA_DBSB;
212 		dbtx = DMA_DBTB;
213 		dcsr = DCSR_STRTB | DCSR_IE | DCSR_RUN;
214 	}
215 
216 	writel_relaxed(sg->addr, base + dbsx);
217 	writel_relaxed(sg->len, base + dbtx);
218 	writel(dcsr, base + DMA_DCSR_S);
219 
220 	dev_dbg(p->dev->slave.dev, "pchan %u: load: DCSR:%02x DBS%c:%08x DBT%c:%08x\n",
221 		p->num, dcsr,
222 		'A' + (dbsx == DMA_DBSB), sg->addr,
223 		'A' + (dbtx == DMA_DBTB), sg->len);
224 }
225 
226 static void noinline sa11x0_dma_complete(struct sa11x0_dma_phy *p,
227 	struct sa11x0_dma_chan *c)
228 {
229 	struct sa11x0_dma_desc *txd = p->txd_done;
230 
231 	if (++p->sg_done == txd->sglen) {
232 		if (!txd->cyclic) {
233 			vchan_cookie_complete(&txd->vd);
234 
235 			p->sg_done = 0;
236 			p->txd_done = p->txd_load;
237 
238 			if (!p->txd_done)
239 				tasklet_schedule(&p->dev->task);
240 		} else {
241 			if ((p->sg_done % txd->period) == 0)
242 				vchan_cyclic_callback(&txd->vd);
243 
244 			/* Cyclic: reset back to beginning */
245 			p->sg_done = 0;
246 		}
247 	}
248 
249 	sa11x0_dma_start_sg(p, c);
250 }
251 
252 static irqreturn_t sa11x0_dma_irq(int irq, void *dev_id)
253 {
254 	struct sa11x0_dma_phy *p = dev_id;
255 	struct sa11x0_dma_dev *d = p->dev;
256 	struct sa11x0_dma_chan *c;
257 	u32 dcsr;
258 
259 	dcsr = readl_relaxed(p->base + DMA_DCSR_R);
260 	if (!(dcsr & (DCSR_ERROR | DCSR_DONEA | DCSR_DONEB)))
261 		return IRQ_NONE;
262 
263 	/* Clear reported status bits */
264 	writel_relaxed(dcsr & (DCSR_ERROR | DCSR_DONEA | DCSR_DONEB),
265 		p->base + DMA_DCSR_C);
266 
267 	dev_dbg(d->slave.dev, "pchan %u: irq: DCSR:%02x\n", p->num, dcsr);
268 
269 	if (dcsr & DCSR_ERROR) {
270 		dev_err(d->slave.dev, "pchan %u: error. DCSR:%02x DDAR:%08x DBSA:%08x DBTA:%08x DBSB:%08x DBTB:%08x\n",
271 			p->num, dcsr,
272 			readl_relaxed(p->base + DMA_DDAR),
273 			readl_relaxed(p->base + DMA_DBSA),
274 			readl_relaxed(p->base + DMA_DBTA),
275 			readl_relaxed(p->base + DMA_DBSB),
276 			readl_relaxed(p->base + DMA_DBTB));
277 	}
278 
279 	c = p->vchan;
280 	if (c) {
281 		unsigned long flags;
282 
283 		spin_lock_irqsave(&c->vc.lock, flags);
284 		/*
285 		 * Now that we're holding the lock, check that the vchan
286 		 * really is associated with this pchan before touching the
287 		 * hardware.  This should always succeed, because we won't
288 		 * change p->vchan or c->phy while the channel is actively
289 		 * transferring.
290 		 */
291 		if (c->phy == p) {
292 			if (dcsr & DCSR_DONEA)
293 				sa11x0_dma_complete(p, c);
294 			if (dcsr & DCSR_DONEB)
295 				sa11x0_dma_complete(p, c);
296 		}
297 		spin_unlock_irqrestore(&c->vc.lock, flags);
298 	}
299 
300 	return IRQ_HANDLED;
301 }
302 
303 static void sa11x0_dma_start_txd(struct sa11x0_dma_chan *c)
304 {
305 	struct sa11x0_dma_desc *txd = sa11x0_dma_next_desc(c);
306 
307 	/* If the issued list is empty, we have no further txds to process */
308 	if (txd) {
309 		struct sa11x0_dma_phy *p = c->phy;
310 
311 		sa11x0_dma_start_desc(p, txd);
312 		p->txd_done = txd;
313 		p->sg_done = 0;
314 
315 		/* The channel should not have any transfers started */
316 		WARN_ON(readl_relaxed(p->base + DMA_DCSR_R) &
317 				      (DCSR_STRTA | DCSR_STRTB));
318 
319 		/* Clear the run and start bits before changing DDAR */
320 		writel_relaxed(DCSR_RUN | DCSR_STRTA | DCSR_STRTB,
321 			       p->base + DMA_DCSR_C);
322 		writel_relaxed(txd->ddar, p->base + DMA_DDAR);
323 
324 		/* Try to start both buffers */
325 		sa11x0_dma_start_sg(p, c);
326 		sa11x0_dma_start_sg(p, c);
327 	}
328 }
329 
330 static void sa11x0_dma_tasklet(unsigned long arg)
331 {
332 	struct sa11x0_dma_dev *d = (struct sa11x0_dma_dev *)arg;
333 	struct sa11x0_dma_phy *p;
334 	struct sa11x0_dma_chan *c;
335 	unsigned pch, pch_alloc = 0;
336 
337 	dev_dbg(d->slave.dev, "tasklet enter\n");
338 
339 	list_for_each_entry(c, &d->slave.channels, vc.chan.device_node) {
340 		spin_lock_irq(&c->vc.lock);
341 		p = c->phy;
342 		if (p && !p->txd_done) {
343 			sa11x0_dma_start_txd(c);
344 			if (!p->txd_done) {
345 				/* No current txd associated with this channel */
346 				dev_dbg(d->slave.dev, "pchan %u: free\n", p->num);
347 
348 				/* Mark this channel free */
349 				c->phy = NULL;
350 				p->vchan = NULL;
351 			}
352 		}
353 		spin_unlock_irq(&c->vc.lock);
354 	}
355 
356 	spin_lock_irq(&d->lock);
357 	for (pch = 0; pch < NR_PHY_CHAN; pch++) {
358 		p = &d->phy[pch];
359 
360 		if (p->vchan == NULL && !list_empty(&d->chan_pending)) {
361 			c = list_first_entry(&d->chan_pending,
362 				struct sa11x0_dma_chan, node);
363 			list_del_init(&c->node);
364 
365 			pch_alloc |= 1 << pch;
366 
367 			/* Mark this channel allocated */
368 			p->vchan = c;
369 
370 			dev_dbg(d->slave.dev, "pchan %u: alloc vchan %p\n", pch, &c->vc);
371 		}
372 	}
373 	spin_unlock_irq(&d->lock);
374 
375 	for (pch = 0; pch < NR_PHY_CHAN; pch++) {
376 		if (pch_alloc & (1 << pch)) {
377 			p = &d->phy[pch];
378 			c = p->vchan;
379 
380 			spin_lock_irq(&c->vc.lock);
381 			c->phy = p;
382 
383 			sa11x0_dma_start_txd(c);
384 			spin_unlock_irq(&c->vc.lock);
385 		}
386 	}
387 
388 	dev_dbg(d->slave.dev, "tasklet exit\n");
389 }
390 
391 
392 static void sa11x0_dma_free_chan_resources(struct dma_chan *chan)
393 {
394 	struct sa11x0_dma_chan *c = to_sa11x0_dma_chan(chan);
395 	struct sa11x0_dma_dev *d = to_sa11x0_dma(chan->device);
396 	unsigned long flags;
397 
398 	spin_lock_irqsave(&d->lock, flags);
399 	list_del_init(&c->node);
400 	spin_unlock_irqrestore(&d->lock, flags);
401 
402 	vchan_free_chan_resources(&c->vc);
403 }
404 
405 static dma_addr_t sa11x0_dma_pos(struct sa11x0_dma_phy *p)
406 {
407 	unsigned reg;
408 	u32 dcsr;
409 
410 	dcsr = readl_relaxed(p->base + DMA_DCSR_R);
411 
412 	if ((dcsr & (DCSR_BIU | DCSR_STRTA)) == DCSR_STRTA ||
413 	    (dcsr & (DCSR_BIU | DCSR_STRTB)) == DCSR_BIU)
414 		reg = DMA_DBSA;
415 	else
416 		reg = DMA_DBSB;
417 
418 	return readl_relaxed(p->base + reg);
419 }
420 
421 static enum dma_status sa11x0_dma_tx_status(struct dma_chan *chan,
422 	dma_cookie_t cookie, struct dma_tx_state *state)
423 {
424 	struct sa11x0_dma_chan *c = to_sa11x0_dma_chan(chan);
425 	struct sa11x0_dma_dev *d = to_sa11x0_dma(chan->device);
426 	struct sa11x0_dma_phy *p;
427 	struct virt_dma_desc *vd;
428 	unsigned long flags;
429 	enum dma_status ret;
430 
431 	ret = dma_cookie_status(&c->vc.chan, cookie, state);
432 	if (ret == DMA_COMPLETE)
433 		return ret;
434 
435 	if (!state)
436 		return c->status;
437 
438 	spin_lock_irqsave(&c->vc.lock, flags);
439 	p = c->phy;
440 
441 	/*
442 	 * If the cookie is on our issue queue, then the residue is
443 	 * its total size.
444 	 */
445 	vd = vchan_find_desc(&c->vc, cookie);
446 	if (vd) {
447 		state->residue = container_of(vd, struct sa11x0_dma_desc, vd)->size;
448 	} else if (!p) {
449 		state->residue = 0;
450 	} else {
451 		struct sa11x0_dma_desc *txd;
452 		size_t bytes = 0;
453 
454 		if (p->txd_done && p->txd_done->vd.tx.cookie == cookie)
455 			txd = p->txd_done;
456 		else if (p->txd_load && p->txd_load->vd.tx.cookie == cookie)
457 			txd = p->txd_load;
458 		else
459 			txd = NULL;
460 
461 		ret = c->status;
462 		if (txd) {
463 			dma_addr_t addr = sa11x0_dma_pos(p);
464 			unsigned i;
465 
466 			dev_vdbg(d->slave.dev, "tx_status: addr:%pad\n", &addr);
467 
468 			for (i = 0; i < txd->sglen; i++) {
469 				dev_vdbg(d->slave.dev, "tx_status: [%u] %x+%x\n",
470 					i, txd->sg[i].addr, txd->sg[i].len);
471 				if (addr >= txd->sg[i].addr &&
472 				    addr < txd->sg[i].addr + txd->sg[i].len) {
473 					unsigned len;
474 
475 					len = txd->sg[i].len -
476 						(addr - txd->sg[i].addr);
477 					dev_vdbg(d->slave.dev, "tx_status: [%u] +%x\n",
478 						i, len);
479 					bytes += len;
480 					i++;
481 					break;
482 				}
483 			}
484 			for (; i < txd->sglen; i++) {
485 				dev_vdbg(d->slave.dev, "tx_status: [%u] %x+%x ++\n",
486 					i, txd->sg[i].addr, txd->sg[i].len);
487 				bytes += txd->sg[i].len;
488 			}
489 		}
490 		state->residue = bytes;
491 	}
492 	spin_unlock_irqrestore(&c->vc.lock, flags);
493 
494 	dev_vdbg(d->slave.dev, "tx_status: bytes 0x%x\n", state->residue);
495 
496 	return ret;
497 }
498 
499 /*
500  * Move pending txds to the issued list, and re-init pending list.
501  * If not already pending, add this channel to the list of pending
502  * channels and trigger the tasklet to run.
503  */
504 static void sa11x0_dma_issue_pending(struct dma_chan *chan)
505 {
506 	struct sa11x0_dma_chan *c = to_sa11x0_dma_chan(chan);
507 	struct sa11x0_dma_dev *d = to_sa11x0_dma(chan->device);
508 	unsigned long flags;
509 
510 	spin_lock_irqsave(&c->vc.lock, flags);
511 	if (vchan_issue_pending(&c->vc)) {
512 		if (!c->phy) {
513 			spin_lock(&d->lock);
514 			if (list_empty(&c->node)) {
515 				list_add_tail(&c->node, &d->chan_pending);
516 				tasklet_schedule(&d->task);
517 				dev_dbg(d->slave.dev, "vchan %p: issued\n", &c->vc);
518 			}
519 			spin_unlock(&d->lock);
520 		}
521 	} else
522 		dev_dbg(d->slave.dev, "vchan %p: nothing to issue\n", &c->vc);
523 	spin_unlock_irqrestore(&c->vc.lock, flags);
524 }
525 
526 static struct dma_async_tx_descriptor *sa11x0_dma_prep_slave_sg(
527 	struct dma_chan *chan, struct scatterlist *sg, unsigned int sglen,
528 	enum dma_transfer_direction dir, unsigned long flags, void *context)
529 {
530 	struct sa11x0_dma_chan *c = to_sa11x0_dma_chan(chan);
531 	struct sa11x0_dma_desc *txd;
532 	struct scatterlist *sgent;
533 	unsigned i, j = sglen;
534 	size_t size = 0;
535 
536 	/* SA11x0 channels can only operate in their native direction */
537 	if (dir != (c->ddar & DDAR_RW ? DMA_DEV_TO_MEM : DMA_MEM_TO_DEV)) {
538 		dev_err(chan->device->dev, "vchan %p: bad DMA direction: DDAR:%08x dir:%u\n",
539 			&c->vc, c->ddar, dir);
540 		return NULL;
541 	}
542 
543 	/* Do not allow zero-sized txds */
544 	if (sglen == 0)
545 		return NULL;
546 
547 	for_each_sg(sg, sgent, sglen, i) {
548 		dma_addr_t addr = sg_dma_address(sgent);
549 		unsigned int len = sg_dma_len(sgent);
550 
551 		if (len > DMA_MAX_SIZE)
552 			j += DIV_ROUND_UP(len, DMA_MAX_SIZE & ~DMA_ALIGN) - 1;
553 		if (addr & DMA_ALIGN) {
554 			dev_dbg(chan->device->dev, "vchan %p: bad buffer alignment: %pad\n",
555 				&c->vc, &addr);
556 			return NULL;
557 		}
558 	}
559 
560 	txd = kzalloc(struct_size(txd, sg, j), GFP_ATOMIC);
561 	if (!txd) {
562 		dev_dbg(chan->device->dev, "vchan %p: kzalloc failed\n", &c->vc);
563 		return NULL;
564 	}
565 
566 	j = 0;
567 	for_each_sg(sg, sgent, sglen, i) {
568 		dma_addr_t addr = sg_dma_address(sgent);
569 		unsigned len = sg_dma_len(sgent);
570 
571 		size += len;
572 
573 		do {
574 			unsigned tlen = len;
575 
576 			/*
577 			 * Check whether the transfer will fit.  If not, try
578 			 * to split the transfer up such that we end up with
579 			 * equal chunks - but make sure that we preserve the
580 			 * alignment.  This avoids small segments.
581 			 */
582 			if (tlen > DMA_MAX_SIZE) {
583 				unsigned mult = DIV_ROUND_UP(tlen,
584 					DMA_MAX_SIZE & ~DMA_ALIGN);
585 
586 				tlen = (tlen / mult) & ~DMA_ALIGN;
587 			}
588 
589 			txd->sg[j].addr = addr;
590 			txd->sg[j].len = tlen;
591 
592 			addr += tlen;
593 			len -= tlen;
594 			j++;
595 		} while (len);
596 	}
597 
598 	txd->ddar = c->ddar;
599 	txd->size = size;
600 	txd->sglen = j;
601 
602 	dev_dbg(chan->device->dev, "vchan %p: txd %p: size %zu nr %u\n",
603 		&c->vc, &txd->vd, txd->size, txd->sglen);
604 
605 	return vchan_tx_prep(&c->vc, &txd->vd, flags);
606 }
607 
608 static struct dma_async_tx_descriptor *sa11x0_dma_prep_dma_cyclic(
609 	struct dma_chan *chan, dma_addr_t addr, size_t size, size_t period,
610 	enum dma_transfer_direction dir, unsigned long flags)
611 {
612 	struct sa11x0_dma_chan *c = to_sa11x0_dma_chan(chan);
613 	struct sa11x0_dma_desc *txd;
614 	unsigned i, j, k, sglen, sgperiod;
615 
616 	/* SA11x0 channels can only operate in their native direction */
617 	if (dir != (c->ddar & DDAR_RW ? DMA_DEV_TO_MEM : DMA_MEM_TO_DEV)) {
618 		dev_err(chan->device->dev, "vchan %p: bad DMA direction: DDAR:%08x dir:%u\n",
619 			&c->vc, c->ddar, dir);
620 		return NULL;
621 	}
622 
623 	sgperiod = DIV_ROUND_UP(period, DMA_MAX_SIZE & ~DMA_ALIGN);
624 	sglen = size * sgperiod / period;
625 
626 	/* Do not allow zero-sized txds */
627 	if (sglen == 0)
628 		return NULL;
629 
630 	txd = kzalloc(struct_size(txd, sg, sglen), GFP_ATOMIC);
631 	if (!txd) {
632 		dev_dbg(chan->device->dev, "vchan %p: kzalloc failed\n", &c->vc);
633 		return NULL;
634 	}
635 
636 	for (i = k = 0; i < size / period; i++) {
637 		size_t tlen, len = period;
638 
639 		for (j = 0; j < sgperiod; j++, k++) {
640 			tlen = len;
641 
642 			if (tlen > DMA_MAX_SIZE) {
643 				unsigned mult = DIV_ROUND_UP(tlen, DMA_MAX_SIZE & ~DMA_ALIGN);
644 				tlen = (tlen / mult) & ~DMA_ALIGN;
645 			}
646 
647 			txd->sg[k].addr = addr;
648 			txd->sg[k].len = tlen;
649 			addr += tlen;
650 			len -= tlen;
651 		}
652 
653 		WARN_ON(len != 0);
654 	}
655 
656 	WARN_ON(k != sglen);
657 
658 	txd->ddar = c->ddar;
659 	txd->size = size;
660 	txd->sglen = sglen;
661 	txd->cyclic = 1;
662 	txd->period = sgperiod;
663 
664 	return vchan_tx_prep(&c->vc, &txd->vd, DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
665 }
666 
667 static int sa11x0_dma_device_config(struct dma_chan *chan,
668 				    struct dma_slave_config *cfg)
669 {
670 	struct sa11x0_dma_chan *c = to_sa11x0_dma_chan(chan);
671 	u32 ddar = c->ddar & ((0xf << 4) | DDAR_RW);
672 	dma_addr_t addr;
673 	enum dma_slave_buswidth width;
674 	u32 maxburst;
675 
676 	if (ddar & DDAR_RW) {
677 		addr = cfg->src_addr;
678 		width = cfg->src_addr_width;
679 		maxburst = cfg->src_maxburst;
680 	} else {
681 		addr = cfg->dst_addr;
682 		width = cfg->dst_addr_width;
683 		maxburst = cfg->dst_maxburst;
684 	}
685 
686 	if ((width != DMA_SLAVE_BUSWIDTH_1_BYTE &&
687 	     width != DMA_SLAVE_BUSWIDTH_2_BYTES) ||
688 	    (maxburst != 4 && maxburst != 8))
689 		return -EINVAL;
690 
691 	if (width == DMA_SLAVE_BUSWIDTH_2_BYTES)
692 		ddar |= DDAR_DW;
693 	if (maxburst == 8)
694 		ddar |= DDAR_BS;
695 
696 	dev_dbg(c->vc.chan.device->dev, "vchan %p: dma_slave_config addr %pad width %u burst %u\n",
697 		&c->vc, &addr, width, maxburst);
698 
699 	c->ddar = ddar | (addr & 0xf0000000) | (addr & 0x003ffffc) << 6;
700 
701 	return 0;
702 }
703 
704 static int sa11x0_dma_device_pause(struct dma_chan *chan)
705 {
706 	struct sa11x0_dma_chan *c = to_sa11x0_dma_chan(chan);
707 	struct sa11x0_dma_dev *d = to_sa11x0_dma(chan->device);
708 	struct sa11x0_dma_phy *p;
709 	LIST_HEAD(head);
710 	unsigned long flags;
711 
712 	dev_dbg(d->slave.dev, "vchan %p: pause\n", &c->vc);
713 	spin_lock_irqsave(&c->vc.lock, flags);
714 	if (c->status == DMA_IN_PROGRESS) {
715 		c->status = DMA_PAUSED;
716 
717 		p = c->phy;
718 		if (p) {
719 			writel(DCSR_RUN | DCSR_IE, p->base + DMA_DCSR_C);
720 		} else {
721 			spin_lock(&d->lock);
722 			list_del_init(&c->node);
723 			spin_unlock(&d->lock);
724 		}
725 	}
726 	spin_unlock_irqrestore(&c->vc.lock, flags);
727 
728 	return 0;
729 }
730 
731 static int sa11x0_dma_device_resume(struct dma_chan *chan)
732 {
733 	struct sa11x0_dma_chan *c = to_sa11x0_dma_chan(chan);
734 	struct sa11x0_dma_dev *d = to_sa11x0_dma(chan->device);
735 	struct sa11x0_dma_phy *p;
736 	LIST_HEAD(head);
737 	unsigned long flags;
738 
739 	dev_dbg(d->slave.dev, "vchan %p: resume\n", &c->vc);
740 	spin_lock_irqsave(&c->vc.lock, flags);
741 	if (c->status == DMA_PAUSED) {
742 		c->status = DMA_IN_PROGRESS;
743 
744 		p = c->phy;
745 		if (p) {
746 			writel(DCSR_RUN | DCSR_IE, p->base + DMA_DCSR_S);
747 		} else if (!list_empty(&c->vc.desc_issued)) {
748 			spin_lock(&d->lock);
749 			list_add_tail(&c->node, &d->chan_pending);
750 			spin_unlock(&d->lock);
751 		}
752 	}
753 	spin_unlock_irqrestore(&c->vc.lock, flags);
754 
755 	return 0;
756 }
757 
758 static int sa11x0_dma_device_terminate_all(struct dma_chan *chan)
759 {
760 	struct sa11x0_dma_chan *c = to_sa11x0_dma_chan(chan);
761 	struct sa11x0_dma_dev *d = to_sa11x0_dma(chan->device);
762 	struct sa11x0_dma_phy *p;
763 	LIST_HEAD(head);
764 	unsigned long flags;
765 
766 	dev_dbg(d->slave.dev, "vchan %p: terminate all\n", &c->vc);
767 	/* Clear the tx descriptor lists */
768 	spin_lock_irqsave(&c->vc.lock, flags);
769 	vchan_get_all_descriptors(&c->vc, &head);
770 
771 	p = c->phy;
772 	if (p) {
773 		dev_dbg(d->slave.dev, "pchan %u: terminating\n", p->num);
774 		/* vchan is assigned to a pchan - stop the channel */
775 		writel(DCSR_RUN | DCSR_IE |
776 		       DCSR_STRTA | DCSR_DONEA |
777 		       DCSR_STRTB | DCSR_DONEB,
778 		       p->base + DMA_DCSR_C);
779 
780 		if (p->txd_load) {
781 			if (p->txd_load != p->txd_done)
782 				list_add_tail(&p->txd_load->vd.node, &head);
783 			p->txd_load = NULL;
784 		}
785 		if (p->txd_done) {
786 			list_add_tail(&p->txd_done->vd.node, &head);
787 			p->txd_done = NULL;
788 		}
789 		c->phy = NULL;
790 		spin_lock(&d->lock);
791 		p->vchan = NULL;
792 		spin_unlock(&d->lock);
793 		tasklet_schedule(&d->task);
794 	}
795 	spin_unlock_irqrestore(&c->vc.lock, flags);
796 	vchan_dma_desc_free_list(&c->vc, &head);
797 
798 	return 0;
799 }
800 
801 struct sa11x0_dma_channel_desc {
802 	u32 ddar;
803 	const char *name;
804 };
805 
806 #define CD(d1, d2) { .ddar = DDAR_##d1 | d2, .name = #d1 }
807 static const struct sa11x0_dma_channel_desc chan_desc[] = {
808 	CD(Ser0UDCTr, 0),
809 	CD(Ser0UDCRc, DDAR_RW),
810 	CD(Ser1SDLCTr, 0),
811 	CD(Ser1SDLCRc, DDAR_RW),
812 	CD(Ser1UARTTr, 0),
813 	CD(Ser1UARTRc, DDAR_RW),
814 	CD(Ser2ICPTr, 0),
815 	CD(Ser2ICPRc, DDAR_RW),
816 	CD(Ser3UARTTr, 0),
817 	CD(Ser3UARTRc, DDAR_RW),
818 	CD(Ser4MCP0Tr, 0),
819 	CD(Ser4MCP0Rc, DDAR_RW),
820 	CD(Ser4MCP1Tr, 0),
821 	CD(Ser4MCP1Rc, DDAR_RW),
822 	CD(Ser4SSPTr, 0),
823 	CD(Ser4SSPRc, DDAR_RW),
824 };
825 
826 static const struct dma_slave_map sa11x0_dma_map[] = {
827 	{ "sa11x0-ir", "tx", "Ser2ICPTr" },
828 	{ "sa11x0-ir", "rx", "Ser2ICPRc" },
829 	{ "sa11x0-ssp", "tx", "Ser4SSPTr" },
830 	{ "sa11x0-ssp", "rx", "Ser4SSPRc" },
831 };
832 
833 static int sa11x0_dma_init_dmadev(struct dma_device *dmadev,
834 	struct device *dev)
835 {
836 	unsigned i;
837 
838 	INIT_LIST_HEAD(&dmadev->channels);
839 	dmadev->dev = dev;
840 	dmadev->device_free_chan_resources = sa11x0_dma_free_chan_resources;
841 	dmadev->device_config = sa11x0_dma_device_config;
842 	dmadev->device_pause = sa11x0_dma_device_pause;
843 	dmadev->device_resume = sa11x0_dma_device_resume;
844 	dmadev->device_terminate_all = sa11x0_dma_device_terminate_all;
845 	dmadev->device_tx_status = sa11x0_dma_tx_status;
846 	dmadev->device_issue_pending = sa11x0_dma_issue_pending;
847 
848 	for (i = 0; i < ARRAY_SIZE(chan_desc); i++) {
849 		struct sa11x0_dma_chan *c;
850 
851 		c = kzalloc(sizeof(*c), GFP_KERNEL);
852 		if (!c) {
853 			dev_err(dev, "no memory for channel %u\n", i);
854 			return -ENOMEM;
855 		}
856 
857 		c->status = DMA_IN_PROGRESS;
858 		c->ddar = chan_desc[i].ddar;
859 		c->name = chan_desc[i].name;
860 		INIT_LIST_HEAD(&c->node);
861 
862 		c->vc.desc_free = sa11x0_dma_free_desc;
863 		vchan_init(&c->vc, dmadev);
864 	}
865 
866 	return dma_async_device_register(dmadev);
867 }
868 
869 static int sa11x0_dma_request_irq(struct platform_device *pdev, int nr,
870 	void *data)
871 {
872 	int irq = platform_get_irq(pdev, nr);
873 
874 	if (irq <= 0)
875 		return -ENXIO;
876 
877 	return request_irq(irq, sa11x0_dma_irq, 0, dev_name(&pdev->dev), data);
878 }
879 
880 static void sa11x0_dma_free_irq(struct platform_device *pdev, int nr,
881 	void *data)
882 {
883 	int irq = platform_get_irq(pdev, nr);
884 	if (irq > 0)
885 		free_irq(irq, data);
886 }
887 
888 static void sa11x0_dma_free_channels(struct dma_device *dmadev)
889 {
890 	struct sa11x0_dma_chan *c, *cn;
891 
892 	list_for_each_entry_safe(c, cn, &dmadev->channels, vc.chan.device_node) {
893 		list_del(&c->vc.chan.device_node);
894 		tasklet_kill(&c->vc.task);
895 		kfree(c);
896 	}
897 }
898 
899 static int sa11x0_dma_probe(struct platform_device *pdev)
900 {
901 	struct sa11x0_dma_dev *d;
902 	struct resource *res;
903 	unsigned i;
904 	int ret;
905 
906 	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
907 	if (!res)
908 		return -ENXIO;
909 
910 	d = kzalloc(sizeof(*d), GFP_KERNEL);
911 	if (!d) {
912 		ret = -ENOMEM;
913 		goto err_alloc;
914 	}
915 
916 	spin_lock_init(&d->lock);
917 	INIT_LIST_HEAD(&d->chan_pending);
918 
919 	d->slave.filter.fn = sa11x0_dma_filter_fn;
920 	d->slave.filter.mapcnt = ARRAY_SIZE(sa11x0_dma_map);
921 	d->slave.filter.map = sa11x0_dma_map;
922 
923 	d->base = ioremap(res->start, resource_size(res));
924 	if (!d->base) {
925 		ret = -ENOMEM;
926 		goto err_ioremap;
927 	}
928 
929 	tasklet_init(&d->task, sa11x0_dma_tasklet, (unsigned long)d);
930 
931 	for (i = 0; i < NR_PHY_CHAN; i++) {
932 		struct sa11x0_dma_phy *p = &d->phy[i];
933 
934 		p->dev = d;
935 		p->num = i;
936 		p->base = d->base + i * DMA_SIZE;
937 		writel_relaxed(DCSR_RUN | DCSR_IE | DCSR_ERROR |
938 			DCSR_DONEA | DCSR_STRTA | DCSR_DONEB | DCSR_STRTB,
939 			p->base + DMA_DCSR_C);
940 		writel_relaxed(0, p->base + DMA_DDAR);
941 
942 		ret = sa11x0_dma_request_irq(pdev, i, p);
943 		if (ret) {
944 			while (i) {
945 				i--;
946 				sa11x0_dma_free_irq(pdev, i, &d->phy[i]);
947 			}
948 			goto err_irq;
949 		}
950 	}
951 
952 	dma_cap_set(DMA_SLAVE, d->slave.cap_mask);
953 	dma_cap_set(DMA_CYCLIC, d->slave.cap_mask);
954 	d->slave.device_prep_slave_sg = sa11x0_dma_prep_slave_sg;
955 	d->slave.device_prep_dma_cyclic = sa11x0_dma_prep_dma_cyclic;
956 	d->slave.directions = BIT(DMA_DEV_TO_MEM) | BIT(DMA_MEM_TO_DEV);
957 	d->slave.residue_granularity = DMA_RESIDUE_GRANULARITY_BURST;
958 	d->slave.src_addr_widths = BIT(DMA_SLAVE_BUSWIDTH_1_BYTE) |
959 				   BIT(DMA_SLAVE_BUSWIDTH_2_BYTES);
960 	d->slave.dst_addr_widths = BIT(DMA_SLAVE_BUSWIDTH_1_BYTE) |
961 				   BIT(DMA_SLAVE_BUSWIDTH_2_BYTES);
962 	ret = sa11x0_dma_init_dmadev(&d->slave, &pdev->dev);
963 	if (ret) {
964 		dev_warn(d->slave.dev, "failed to register slave async device: %d\n",
965 			ret);
966 		goto err_slave_reg;
967 	}
968 
969 	platform_set_drvdata(pdev, d);
970 	return 0;
971 
972  err_slave_reg:
973 	sa11x0_dma_free_channels(&d->slave);
974 	for (i = 0; i < NR_PHY_CHAN; i++)
975 		sa11x0_dma_free_irq(pdev, i, &d->phy[i]);
976  err_irq:
977 	tasklet_kill(&d->task);
978 	iounmap(d->base);
979  err_ioremap:
980 	kfree(d);
981  err_alloc:
982 	return ret;
983 }
984 
985 static int sa11x0_dma_remove(struct platform_device *pdev)
986 {
987 	struct sa11x0_dma_dev *d = platform_get_drvdata(pdev);
988 	unsigned pch;
989 
990 	dma_async_device_unregister(&d->slave);
991 
992 	sa11x0_dma_free_channels(&d->slave);
993 	for (pch = 0; pch < NR_PHY_CHAN; pch++)
994 		sa11x0_dma_free_irq(pdev, pch, &d->phy[pch]);
995 	tasklet_kill(&d->task);
996 	iounmap(d->base);
997 	kfree(d);
998 
999 	return 0;
1000 }
1001 
1002 static int sa11x0_dma_suspend(struct device *dev)
1003 {
1004 	struct sa11x0_dma_dev *d = dev_get_drvdata(dev);
1005 	unsigned pch;
1006 
1007 	for (pch = 0; pch < NR_PHY_CHAN; pch++) {
1008 		struct sa11x0_dma_phy *p = &d->phy[pch];
1009 		u32 dcsr, saved_dcsr;
1010 
1011 		dcsr = saved_dcsr = readl_relaxed(p->base + DMA_DCSR_R);
1012 		if (dcsr & DCSR_RUN) {
1013 			writel(DCSR_RUN | DCSR_IE, p->base + DMA_DCSR_C);
1014 			dcsr = readl_relaxed(p->base + DMA_DCSR_R);
1015 		}
1016 
1017 		saved_dcsr &= DCSR_RUN | DCSR_IE;
1018 		if (dcsr & DCSR_BIU) {
1019 			p->dbs[0] = readl_relaxed(p->base + DMA_DBSB);
1020 			p->dbt[0] = readl_relaxed(p->base + DMA_DBTB);
1021 			p->dbs[1] = readl_relaxed(p->base + DMA_DBSA);
1022 			p->dbt[1] = readl_relaxed(p->base + DMA_DBTA);
1023 			saved_dcsr |= (dcsr & DCSR_STRTA ? DCSR_STRTB : 0) |
1024 				      (dcsr & DCSR_STRTB ? DCSR_STRTA : 0);
1025 		} else {
1026 			p->dbs[0] = readl_relaxed(p->base + DMA_DBSA);
1027 			p->dbt[0] = readl_relaxed(p->base + DMA_DBTA);
1028 			p->dbs[1] = readl_relaxed(p->base + DMA_DBSB);
1029 			p->dbt[1] = readl_relaxed(p->base + DMA_DBTB);
1030 			saved_dcsr |= dcsr & (DCSR_STRTA | DCSR_STRTB);
1031 		}
1032 		p->dcsr = saved_dcsr;
1033 
1034 		writel(DCSR_STRTA | DCSR_STRTB, p->base + DMA_DCSR_C);
1035 	}
1036 
1037 	return 0;
1038 }
1039 
1040 static int sa11x0_dma_resume(struct device *dev)
1041 {
1042 	struct sa11x0_dma_dev *d = dev_get_drvdata(dev);
1043 	unsigned pch;
1044 
1045 	for (pch = 0; pch < NR_PHY_CHAN; pch++) {
1046 		struct sa11x0_dma_phy *p = &d->phy[pch];
1047 		struct sa11x0_dma_desc *txd = NULL;
1048 		u32 dcsr = readl_relaxed(p->base + DMA_DCSR_R);
1049 
1050 		WARN_ON(dcsr & (DCSR_BIU | DCSR_STRTA | DCSR_STRTB | DCSR_RUN));
1051 
1052 		if (p->txd_done)
1053 			txd = p->txd_done;
1054 		else if (p->txd_load)
1055 			txd = p->txd_load;
1056 
1057 		if (!txd)
1058 			continue;
1059 
1060 		writel_relaxed(txd->ddar, p->base + DMA_DDAR);
1061 
1062 		writel_relaxed(p->dbs[0], p->base + DMA_DBSA);
1063 		writel_relaxed(p->dbt[0], p->base + DMA_DBTA);
1064 		writel_relaxed(p->dbs[1], p->base + DMA_DBSB);
1065 		writel_relaxed(p->dbt[1], p->base + DMA_DBTB);
1066 		writel_relaxed(p->dcsr, p->base + DMA_DCSR_S);
1067 	}
1068 
1069 	return 0;
1070 }
1071 
1072 static const struct dev_pm_ops sa11x0_dma_pm_ops = {
1073 	.suspend_noirq = sa11x0_dma_suspend,
1074 	.resume_noirq = sa11x0_dma_resume,
1075 	.freeze_noirq = sa11x0_dma_suspend,
1076 	.thaw_noirq = sa11x0_dma_resume,
1077 	.poweroff_noirq = sa11x0_dma_suspend,
1078 	.restore_noirq = sa11x0_dma_resume,
1079 };
1080 
1081 static struct platform_driver sa11x0_dma_driver = {
1082 	.driver = {
1083 		.name	= "sa11x0-dma",
1084 		.pm	= &sa11x0_dma_pm_ops,
1085 	},
1086 	.probe		= sa11x0_dma_probe,
1087 	.remove		= sa11x0_dma_remove,
1088 };
1089 
1090 bool sa11x0_dma_filter_fn(struct dma_chan *chan, void *param)
1091 {
1092 	if (chan->device->dev->driver == &sa11x0_dma_driver.driver) {
1093 		struct sa11x0_dma_chan *c = to_sa11x0_dma_chan(chan);
1094 		const char *p = param;
1095 
1096 		return !strcmp(c->name, p);
1097 	}
1098 	return false;
1099 }
1100 EXPORT_SYMBOL(sa11x0_dma_filter_fn);
1101 
1102 static int __init sa11x0_dma_init(void)
1103 {
1104 	return platform_driver_register(&sa11x0_dma_driver);
1105 }
1106 subsys_initcall(sa11x0_dma_init);
1107 
1108 static void __exit sa11x0_dma_exit(void)
1109 {
1110 	platform_driver_unregister(&sa11x0_dma_driver);
1111 }
1112 module_exit(sa11x0_dma_exit);
1113 
1114 MODULE_AUTHOR("Russell King");
1115 MODULE_DESCRIPTION("SA-11x0 DMA driver");
1116 MODULE_LICENSE("GPL v2");
1117 MODULE_ALIAS("platform:sa11x0-dma");
1118