xref: /linux/drivers/dma/mmp_pdma.c (revision 887069f424550ebdcb411166733e1d05002b58e4)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright 2012 Marvell International Ltd.
4  */
5 
6 #include <linux/err.h>
7 #include <linux/module.h>
8 #include <linux/init.h>
9 #include <linux/types.h>
10 #include <linux/interrupt.h>
11 #include <linux/dma-mapping.h>
12 #include <linux/slab.h>
13 #include <linux/dmaengine.h>
14 #include <linux/platform_device.h>
15 #include <linux/device.h>
16 #include <linux/platform_data/mmp_dma.h>
17 #include <linux/dmapool.h>
18 #include <linux/of_device.h>
19 #include <linux/of_dma.h>
20 #include <linux/of.h>
21 
22 #include "dmaengine.h"
23 
24 #define DCSR		0x0000
25 #define DALGN		0x00a0
26 #define DINT		0x00f0
27 #define DDADR		0x0200
28 #define DSADR(n)	(0x0204 + ((n) << 4))
29 #define DTADR(n)	(0x0208 + ((n) << 4))
30 #define DCMD		0x020c
31 
32 #define DCSR_RUN	BIT(31)	/* Run Bit (read / write) */
33 #define DCSR_NODESC	BIT(30)	/* No-Descriptor Fetch (read / write) */
34 #define DCSR_STOPIRQEN	BIT(29)	/* Stop Interrupt Enable (read / write) */
35 #define DCSR_REQPEND	BIT(8)	/* Request Pending (read-only) */
36 #define DCSR_STOPSTATE	BIT(3)	/* Stop State (read-only) */
37 #define DCSR_ENDINTR	BIT(2)	/* End Interrupt (read / write) */
38 #define DCSR_STARTINTR	BIT(1)	/* Start Interrupt (read / write) */
39 #define DCSR_BUSERR	BIT(0)	/* Bus Error Interrupt (read / write) */
40 
41 #define DCSR_EORIRQEN	BIT(28)	/* End of Receive Interrupt Enable (R/W) */
42 #define DCSR_EORJMPEN	BIT(27)	/* Jump to next descriptor on EOR */
43 #define DCSR_EORSTOPEN	BIT(26)	/* STOP on an EOR */
44 #define DCSR_SETCMPST	BIT(25)	/* Set Descriptor Compare Status */
45 #define DCSR_CLRCMPST	BIT(24)	/* Clear Descriptor Compare Status */
46 #define DCSR_CMPST	BIT(10)	/* The Descriptor Compare Status */
47 #define DCSR_EORINTR	BIT(9)	/* The end of Receive */
48 
49 #define DRCMR(n)	((((n) < 64) ? 0x0100 : 0x1100) + (((n) & 0x3f) << 2))
50 #define DRCMR_MAPVLD	BIT(7)	/* Map Valid (read / write) */
51 #define DRCMR_CHLNUM	0x1f	/* mask for Channel Number (read / write) */
52 
53 #define DDADR_DESCADDR	0xfffffff0	/* Address of next descriptor (mask) */
54 #define DDADR_STOP	BIT(0)	/* Stop (read / write) */
55 
56 #define DCMD_INCSRCADDR	BIT(31)	/* Source Address Increment Setting. */
57 #define DCMD_INCTRGADDR	BIT(30)	/* Target Address Increment Setting. */
58 #define DCMD_FLOWSRC	BIT(29)	/* Flow Control by the source. */
59 #define DCMD_FLOWTRG	BIT(28)	/* Flow Control by the target. */
60 #define DCMD_STARTIRQEN	BIT(22)	/* Start Interrupt Enable */
61 #define DCMD_ENDIRQEN	BIT(21)	/* End Interrupt Enable */
62 #define DCMD_ENDIAN	BIT(18)	/* Device Endian-ness. */
63 #define DCMD_BURST8	(1 << 16)	/* 8 byte burst */
64 #define DCMD_BURST16	(2 << 16)	/* 16 byte burst */
65 #define DCMD_BURST32	(3 << 16)	/* 32 byte burst */
66 #define DCMD_WIDTH1	(1 << 14)	/* 1 byte width */
67 #define DCMD_WIDTH2	(2 << 14)	/* 2 byte width (HalfWord) */
68 #define DCMD_WIDTH4	(3 << 14)	/* 4 byte width (Word) */
69 #define DCMD_LENGTH	0x01fff		/* length mask (max = 8K - 1) */
70 
71 #define PDMA_MAX_DESC_BYTES	DCMD_LENGTH
72 
73 struct mmp_pdma_desc_hw {
74 	u32 ddadr;	/* Points to the next descriptor + flags */
75 	u32 dsadr;	/* DSADR value for the current transfer */
76 	u32 dtadr;	/* DTADR value for the current transfer */
77 	u32 dcmd;	/* DCMD value for the current transfer */
78 } __aligned(32);
79 
80 struct mmp_pdma_desc_sw {
81 	struct mmp_pdma_desc_hw desc;
82 	struct list_head node;
83 	struct list_head tx_list;
84 	struct dma_async_tx_descriptor async_tx;
85 };
86 
87 struct mmp_pdma_phy;
88 
89 struct mmp_pdma_chan {
90 	struct device *dev;
91 	struct dma_chan chan;
92 	struct dma_async_tx_descriptor desc;
93 	struct mmp_pdma_phy *phy;
94 	enum dma_transfer_direction dir;
95 	struct dma_slave_config slave_config;
96 
97 	struct mmp_pdma_desc_sw *cyclic_first;	/* first desc_sw if channel
98 						 * is in cyclic mode */
99 
100 	/* channel's basic info */
101 	struct tasklet_struct tasklet;
102 	u32 dcmd;
103 	u32 drcmr;
104 	u32 dev_addr;
105 
106 	/* list for desc */
107 	spinlock_t desc_lock;		/* Descriptor list lock */
108 	struct list_head chain_pending;	/* Link descriptors queue for pending */
109 	struct list_head chain_running;	/* Link descriptors queue for running */
110 	bool idle;			/* channel statue machine */
111 	bool byte_align;
112 
113 	struct dma_pool *desc_pool;	/* Descriptors pool */
114 };
115 
116 struct mmp_pdma_phy {
117 	int idx;
118 	void __iomem *base;
119 	struct mmp_pdma_chan *vchan;
120 };
121 
122 struct mmp_pdma_device {
123 	int				dma_channels;
124 	void __iomem			*base;
125 	struct device			*dev;
126 	struct dma_device		device;
127 	struct mmp_pdma_phy		*phy;
128 	spinlock_t phy_lock; /* protect alloc/free phy channels */
129 };
130 
131 #define tx_to_mmp_pdma_desc(tx)					\
132 	container_of(tx, struct mmp_pdma_desc_sw, async_tx)
133 #define to_mmp_pdma_desc(lh)					\
134 	container_of(lh, struct mmp_pdma_desc_sw, node)
135 #define to_mmp_pdma_chan(dchan)					\
136 	container_of(dchan, struct mmp_pdma_chan, chan)
137 #define to_mmp_pdma_dev(dmadev)					\
138 	container_of(dmadev, struct mmp_pdma_device, device)
139 
140 static int mmp_pdma_config_write(struct dma_chan *dchan,
141 			   struct dma_slave_config *cfg,
142 			   enum dma_transfer_direction direction);
143 
144 static void set_desc(struct mmp_pdma_phy *phy, dma_addr_t addr)
145 {
146 	u32 reg = (phy->idx << 4) + DDADR;
147 
148 	writel(addr, phy->base + reg);
149 }
150 
151 static void enable_chan(struct mmp_pdma_phy *phy)
152 {
153 	u32 reg, dalgn;
154 
155 	if (!phy->vchan)
156 		return;
157 
158 	reg = DRCMR(phy->vchan->drcmr);
159 	writel(DRCMR_MAPVLD | phy->idx, phy->base + reg);
160 
161 	dalgn = readl(phy->base + DALGN);
162 	if (phy->vchan->byte_align)
163 		dalgn |= 1 << phy->idx;
164 	else
165 		dalgn &= ~(1 << phy->idx);
166 	writel(dalgn, phy->base + DALGN);
167 
168 	reg = (phy->idx << 2) + DCSR;
169 	writel(readl(phy->base + reg) | DCSR_RUN, phy->base + reg);
170 }
171 
172 static void disable_chan(struct mmp_pdma_phy *phy)
173 {
174 	u32 reg;
175 
176 	if (!phy)
177 		return;
178 
179 	reg = (phy->idx << 2) + DCSR;
180 	writel(readl(phy->base + reg) & ~DCSR_RUN, phy->base + reg);
181 }
182 
183 static int clear_chan_irq(struct mmp_pdma_phy *phy)
184 {
185 	u32 dcsr;
186 	u32 dint = readl(phy->base + DINT);
187 	u32 reg = (phy->idx << 2) + DCSR;
188 
189 	if (!(dint & BIT(phy->idx)))
190 		return -EAGAIN;
191 
192 	/* clear irq */
193 	dcsr = readl(phy->base + reg);
194 	writel(dcsr, phy->base + reg);
195 	if ((dcsr & DCSR_BUSERR) && (phy->vchan))
196 		dev_warn(phy->vchan->dev, "DCSR_BUSERR\n");
197 
198 	return 0;
199 }
200 
201 static irqreturn_t mmp_pdma_chan_handler(int irq, void *dev_id)
202 {
203 	struct mmp_pdma_phy *phy = dev_id;
204 
205 	if (clear_chan_irq(phy) != 0)
206 		return IRQ_NONE;
207 
208 	tasklet_schedule(&phy->vchan->tasklet);
209 	return IRQ_HANDLED;
210 }
211 
212 static irqreturn_t mmp_pdma_int_handler(int irq, void *dev_id)
213 {
214 	struct mmp_pdma_device *pdev = dev_id;
215 	struct mmp_pdma_phy *phy;
216 	u32 dint = readl(pdev->base + DINT);
217 	int i, ret;
218 	int irq_num = 0;
219 
220 	while (dint) {
221 		i = __ffs(dint);
222 		/* only handle interrupts belonging to pdma driver*/
223 		if (i >= pdev->dma_channels)
224 			break;
225 		dint &= (dint - 1);
226 		phy = &pdev->phy[i];
227 		ret = mmp_pdma_chan_handler(irq, phy);
228 		if (ret == IRQ_HANDLED)
229 			irq_num++;
230 	}
231 
232 	if (irq_num)
233 		return IRQ_HANDLED;
234 
235 	return IRQ_NONE;
236 }
237 
238 /* lookup free phy channel as descending priority */
239 static struct mmp_pdma_phy *lookup_phy(struct mmp_pdma_chan *pchan)
240 {
241 	int prio, i;
242 	struct mmp_pdma_device *pdev = to_mmp_pdma_dev(pchan->chan.device);
243 	struct mmp_pdma_phy *phy, *found = NULL;
244 	unsigned long flags;
245 
246 	/*
247 	 * dma channel priorities
248 	 * ch 0 - 3,  16 - 19  <--> (0)
249 	 * ch 4 - 7,  20 - 23  <--> (1)
250 	 * ch 8 - 11, 24 - 27  <--> (2)
251 	 * ch 12 - 15, 28 - 31  <--> (3)
252 	 */
253 
254 	spin_lock_irqsave(&pdev->phy_lock, flags);
255 	for (prio = 0; prio <= ((pdev->dma_channels - 1) & 0xf) >> 2; prio++) {
256 		for (i = 0; i < pdev->dma_channels; i++) {
257 			if (prio != (i & 0xf) >> 2)
258 				continue;
259 			phy = &pdev->phy[i];
260 			if (!phy->vchan) {
261 				phy->vchan = pchan;
262 				found = phy;
263 				goto out_unlock;
264 			}
265 		}
266 	}
267 
268 out_unlock:
269 	spin_unlock_irqrestore(&pdev->phy_lock, flags);
270 	return found;
271 }
272 
273 static void mmp_pdma_free_phy(struct mmp_pdma_chan *pchan)
274 {
275 	struct mmp_pdma_device *pdev = to_mmp_pdma_dev(pchan->chan.device);
276 	unsigned long flags;
277 	u32 reg;
278 
279 	if (!pchan->phy)
280 		return;
281 
282 	/* clear the channel mapping in DRCMR */
283 	reg = DRCMR(pchan->drcmr);
284 	writel(0, pchan->phy->base + reg);
285 
286 	spin_lock_irqsave(&pdev->phy_lock, flags);
287 	pchan->phy->vchan = NULL;
288 	pchan->phy = NULL;
289 	spin_unlock_irqrestore(&pdev->phy_lock, flags);
290 }
291 
292 /*
293  * start_pending_queue - transfer any pending transactions
294  * pending list ==> running list
295  */
296 static void start_pending_queue(struct mmp_pdma_chan *chan)
297 {
298 	struct mmp_pdma_desc_sw *desc;
299 
300 	/* still in running, irq will start the pending list */
301 	if (!chan->idle) {
302 		dev_dbg(chan->dev, "DMA controller still busy\n");
303 		return;
304 	}
305 
306 	if (list_empty(&chan->chain_pending)) {
307 		/* chance to re-fetch phy channel with higher prio */
308 		mmp_pdma_free_phy(chan);
309 		dev_dbg(chan->dev, "no pending list\n");
310 		return;
311 	}
312 
313 	if (!chan->phy) {
314 		chan->phy = lookup_phy(chan);
315 		if (!chan->phy) {
316 			dev_dbg(chan->dev, "no free dma channel\n");
317 			return;
318 		}
319 	}
320 
321 	/*
322 	 * pending -> running
323 	 * reintilize pending list
324 	 */
325 	desc = list_first_entry(&chan->chain_pending,
326 				struct mmp_pdma_desc_sw, node);
327 	list_splice_tail_init(&chan->chain_pending, &chan->chain_running);
328 
329 	/*
330 	 * Program the descriptor's address into the DMA controller,
331 	 * then start the DMA transaction
332 	 */
333 	set_desc(chan->phy, desc->async_tx.phys);
334 	enable_chan(chan->phy);
335 	chan->idle = false;
336 }
337 
338 
339 /* desc->tx_list ==> pending list */
340 static dma_cookie_t mmp_pdma_tx_submit(struct dma_async_tx_descriptor *tx)
341 {
342 	struct mmp_pdma_chan *chan = to_mmp_pdma_chan(tx->chan);
343 	struct mmp_pdma_desc_sw *desc = tx_to_mmp_pdma_desc(tx);
344 	struct mmp_pdma_desc_sw *child;
345 	unsigned long flags;
346 	dma_cookie_t cookie = -EBUSY;
347 
348 	spin_lock_irqsave(&chan->desc_lock, flags);
349 
350 	list_for_each_entry(child, &desc->tx_list, node) {
351 		cookie = dma_cookie_assign(&child->async_tx);
352 	}
353 
354 	/* softly link to pending list - desc->tx_list ==> pending list */
355 	list_splice_tail_init(&desc->tx_list, &chan->chain_pending);
356 
357 	spin_unlock_irqrestore(&chan->desc_lock, flags);
358 
359 	return cookie;
360 }
361 
362 static struct mmp_pdma_desc_sw *
363 mmp_pdma_alloc_descriptor(struct mmp_pdma_chan *chan)
364 {
365 	struct mmp_pdma_desc_sw *desc;
366 	dma_addr_t pdesc;
367 
368 	desc = dma_pool_zalloc(chan->desc_pool, GFP_ATOMIC, &pdesc);
369 	if (!desc) {
370 		dev_err(chan->dev, "out of memory for link descriptor\n");
371 		return NULL;
372 	}
373 
374 	INIT_LIST_HEAD(&desc->tx_list);
375 	dma_async_tx_descriptor_init(&desc->async_tx, &chan->chan);
376 	/* each desc has submit */
377 	desc->async_tx.tx_submit = mmp_pdma_tx_submit;
378 	desc->async_tx.phys = pdesc;
379 
380 	return desc;
381 }
382 
383 /*
384  * mmp_pdma_alloc_chan_resources - Allocate resources for DMA channel.
385  *
386  * This function will create a dma pool for descriptor allocation.
387  * Request irq only when channel is requested
388  * Return - The number of allocated descriptors.
389  */
390 
391 static int mmp_pdma_alloc_chan_resources(struct dma_chan *dchan)
392 {
393 	struct mmp_pdma_chan *chan = to_mmp_pdma_chan(dchan);
394 
395 	if (chan->desc_pool)
396 		return 1;
397 
398 	chan->desc_pool = dma_pool_create(dev_name(&dchan->dev->device),
399 					  chan->dev,
400 					  sizeof(struct mmp_pdma_desc_sw),
401 					  __alignof__(struct mmp_pdma_desc_sw),
402 					  0);
403 	if (!chan->desc_pool) {
404 		dev_err(chan->dev, "unable to allocate descriptor pool\n");
405 		return -ENOMEM;
406 	}
407 
408 	mmp_pdma_free_phy(chan);
409 	chan->idle = true;
410 	chan->dev_addr = 0;
411 	return 1;
412 }
413 
414 static void mmp_pdma_free_desc_list(struct mmp_pdma_chan *chan,
415 				    struct list_head *list)
416 {
417 	struct mmp_pdma_desc_sw *desc, *_desc;
418 
419 	list_for_each_entry_safe(desc, _desc, list, node) {
420 		list_del(&desc->node);
421 		dma_pool_free(chan->desc_pool, desc, desc->async_tx.phys);
422 	}
423 }
424 
425 static void mmp_pdma_free_chan_resources(struct dma_chan *dchan)
426 {
427 	struct mmp_pdma_chan *chan = to_mmp_pdma_chan(dchan);
428 	unsigned long flags;
429 
430 	spin_lock_irqsave(&chan->desc_lock, flags);
431 	mmp_pdma_free_desc_list(chan, &chan->chain_pending);
432 	mmp_pdma_free_desc_list(chan, &chan->chain_running);
433 	spin_unlock_irqrestore(&chan->desc_lock, flags);
434 
435 	dma_pool_destroy(chan->desc_pool);
436 	chan->desc_pool = NULL;
437 	chan->idle = true;
438 	chan->dev_addr = 0;
439 	mmp_pdma_free_phy(chan);
440 	return;
441 }
442 
443 static struct dma_async_tx_descriptor *
444 mmp_pdma_prep_memcpy(struct dma_chan *dchan,
445 		     dma_addr_t dma_dst, dma_addr_t dma_src,
446 		     size_t len, unsigned long flags)
447 {
448 	struct mmp_pdma_chan *chan;
449 	struct mmp_pdma_desc_sw *first = NULL, *prev = NULL, *new;
450 	size_t copy = 0;
451 
452 	if (!dchan)
453 		return NULL;
454 
455 	if (!len)
456 		return NULL;
457 
458 	chan = to_mmp_pdma_chan(dchan);
459 	chan->byte_align = false;
460 
461 	if (!chan->dir) {
462 		chan->dir = DMA_MEM_TO_MEM;
463 		chan->dcmd = DCMD_INCTRGADDR | DCMD_INCSRCADDR;
464 		chan->dcmd |= DCMD_BURST32;
465 	}
466 
467 	do {
468 		/* Allocate the link descriptor from DMA pool */
469 		new = mmp_pdma_alloc_descriptor(chan);
470 		if (!new) {
471 			dev_err(chan->dev, "no memory for desc\n");
472 			goto fail;
473 		}
474 
475 		copy = min_t(size_t, len, PDMA_MAX_DESC_BYTES);
476 		if (dma_src & 0x7 || dma_dst & 0x7)
477 			chan->byte_align = true;
478 
479 		new->desc.dcmd = chan->dcmd | (DCMD_LENGTH & copy);
480 		new->desc.dsadr = dma_src;
481 		new->desc.dtadr = dma_dst;
482 
483 		if (!first)
484 			first = new;
485 		else
486 			prev->desc.ddadr = new->async_tx.phys;
487 
488 		new->async_tx.cookie = 0;
489 		async_tx_ack(&new->async_tx);
490 
491 		prev = new;
492 		len -= copy;
493 
494 		if (chan->dir == DMA_MEM_TO_DEV) {
495 			dma_src += copy;
496 		} else if (chan->dir == DMA_DEV_TO_MEM) {
497 			dma_dst += copy;
498 		} else if (chan->dir == DMA_MEM_TO_MEM) {
499 			dma_src += copy;
500 			dma_dst += copy;
501 		}
502 
503 		/* Insert the link descriptor to the LD ring */
504 		list_add_tail(&new->node, &first->tx_list);
505 	} while (len);
506 
507 	first->async_tx.flags = flags; /* client is in control of this ack */
508 	first->async_tx.cookie = -EBUSY;
509 
510 	/* last desc and fire IRQ */
511 	new->desc.ddadr = DDADR_STOP;
512 	new->desc.dcmd |= DCMD_ENDIRQEN;
513 
514 	chan->cyclic_first = NULL;
515 
516 	return &first->async_tx;
517 
518 fail:
519 	if (first)
520 		mmp_pdma_free_desc_list(chan, &first->tx_list);
521 	return NULL;
522 }
523 
524 static struct dma_async_tx_descriptor *
525 mmp_pdma_prep_slave_sg(struct dma_chan *dchan, struct scatterlist *sgl,
526 		       unsigned int sg_len, enum dma_transfer_direction dir,
527 		       unsigned long flags, void *context)
528 {
529 	struct mmp_pdma_chan *chan = to_mmp_pdma_chan(dchan);
530 	struct mmp_pdma_desc_sw *first = NULL, *prev = NULL, *new = NULL;
531 	size_t len, avail;
532 	struct scatterlist *sg;
533 	dma_addr_t addr;
534 	int i;
535 
536 	if ((sgl == NULL) || (sg_len == 0))
537 		return NULL;
538 
539 	chan->byte_align = false;
540 
541 	mmp_pdma_config_write(dchan, &chan->slave_config, dir);
542 
543 	for_each_sg(sgl, sg, sg_len, i) {
544 		addr = sg_dma_address(sg);
545 		avail = sg_dma_len(sgl);
546 
547 		do {
548 			len = min_t(size_t, avail, PDMA_MAX_DESC_BYTES);
549 			if (addr & 0x7)
550 				chan->byte_align = true;
551 
552 			/* allocate and populate the descriptor */
553 			new = mmp_pdma_alloc_descriptor(chan);
554 			if (!new) {
555 				dev_err(chan->dev, "no memory for desc\n");
556 				goto fail;
557 			}
558 
559 			new->desc.dcmd = chan->dcmd | (DCMD_LENGTH & len);
560 			if (dir == DMA_MEM_TO_DEV) {
561 				new->desc.dsadr = addr;
562 				new->desc.dtadr = chan->dev_addr;
563 			} else {
564 				new->desc.dsadr = chan->dev_addr;
565 				new->desc.dtadr = addr;
566 			}
567 
568 			if (!first)
569 				first = new;
570 			else
571 				prev->desc.ddadr = new->async_tx.phys;
572 
573 			new->async_tx.cookie = 0;
574 			async_tx_ack(&new->async_tx);
575 			prev = new;
576 
577 			/* Insert the link descriptor to the LD ring */
578 			list_add_tail(&new->node, &first->tx_list);
579 
580 			/* update metadata */
581 			addr += len;
582 			avail -= len;
583 		} while (avail);
584 	}
585 
586 	first->async_tx.cookie = -EBUSY;
587 	first->async_tx.flags = flags;
588 
589 	/* last desc and fire IRQ */
590 	new->desc.ddadr = DDADR_STOP;
591 	new->desc.dcmd |= DCMD_ENDIRQEN;
592 
593 	chan->dir = dir;
594 	chan->cyclic_first = NULL;
595 
596 	return &first->async_tx;
597 
598 fail:
599 	if (first)
600 		mmp_pdma_free_desc_list(chan, &first->tx_list);
601 	return NULL;
602 }
603 
604 static struct dma_async_tx_descriptor *
605 mmp_pdma_prep_dma_cyclic(struct dma_chan *dchan,
606 			 dma_addr_t buf_addr, size_t len, size_t period_len,
607 			 enum dma_transfer_direction direction,
608 			 unsigned long flags)
609 {
610 	struct mmp_pdma_chan *chan;
611 	struct mmp_pdma_desc_sw *first = NULL, *prev = NULL, *new;
612 	dma_addr_t dma_src, dma_dst;
613 
614 	if (!dchan || !len || !period_len)
615 		return NULL;
616 
617 	/* the buffer length must be a multiple of period_len */
618 	if (len % period_len != 0)
619 		return NULL;
620 
621 	if (period_len > PDMA_MAX_DESC_BYTES)
622 		return NULL;
623 
624 	chan = to_mmp_pdma_chan(dchan);
625 	mmp_pdma_config_write(dchan, &chan->slave_config, direction);
626 
627 	switch (direction) {
628 	case DMA_MEM_TO_DEV:
629 		dma_src = buf_addr;
630 		dma_dst = chan->dev_addr;
631 		break;
632 	case DMA_DEV_TO_MEM:
633 		dma_dst = buf_addr;
634 		dma_src = chan->dev_addr;
635 		break;
636 	default:
637 		dev_err(chan->dev, "Unsupported direction for cyclic DMA\n");
638 		return NULL;
639 	}
640 
641 	chan->dir = direction;
642 
643 	do {
644 		/* Allocate the link descriptor from DMA pool */
645 		new = mmp_pdma_alloc_descriptor(chan);
646 		if (!new) {
647 			dev_err(chan->dev, "no memory for desc\n");
648 			goto fail;
649 		}
650 
651 		new->desc.dcmd = (chan->dcmd | DCMD_ENDIRQEN |
652 				  (DCMD_LENGTH & period_len));
653 		new->desc.dsadr = dma_src;
654 		new->desc.dtadr = dma_dst;
655 
656 		if (!first)
657 			first = new;
658 		else
659 			prev->desc.ddadr = new->async_tx.phys;
660 
661 		new->async_tx.cookie = 0;
662 		async_tx_ack(&new->async_tx);
663 
664 		prev = new;
665 		len -= period_len;
666 
667 		if (chan->dir == DMA_MEM_TO_DEV)
668 			dma_src += period_len;
669 		else
670 			dma_dst += period_len;
671 
672 		/* Insert the link descriptor to the LD ring */
673 		list_add_tail(&new->node, &first->tx_list);
674 	} while (len);
675 
676 	first->async_tx.flags = flags; /* client is in control of this ack */
677 	first->async_tx.cookie = -EBUSY;
678 
679 	/* make the cyclic link */
680 	new->desc.ddadr = first->async_tx.phys;
681 	chan->cyclic_first = first;
682 
683 	return &first->async_tx;
684 
685 fail:
686 	if (first)
687 		mmp_pdma_free_desc_list(chan, &first->tx_list);
688 	return NULL;
689 }
690 
691 static int mmp_pdma_config_write(struct dma_chan *dchan,
692 			   struct dma_slave_config *cfg,
693 			   enum dma_transfer_direction direction)
694 {
695 	struct mmp_pdma_chan *chan = to_mmp_pdma_chan(dchan);
696 	u32 maxburst = 0, addr = 0;
697 	enum dma_slave_buswidth width = DMA_SLAVE_BUSWIDTH_UNDEFINED;
698 
699 	if (!dchan)
700 		return -EINVAL;
701 
702 	if (direction == DMA_DEV_TO_MEM) {
703 		chan->dcmd = DCMD_INCTRGADDR | DCMD_FLOWSRC;
704 		maxburst = cfg->src_maxburst;
705 		width = cfg->src_addr_width;
706 		addr = cfg->src_addr;
707 	} else if (direction == DMA_MEM_TO_DEV) {
708 		chan->dcmd = DCMD_INCSRCADDR | DCMD_FLOWTRG;
709 		maxburst = cfg->dst_maxburst;
710 		width = cfg->dst_addr_width;
711 		addr = cfg->dst_addr;
712 	}
713 
714 	if (width == DMA_SLAVE_BUSWIDTH_1_BYTE)
715 		chan->dcmd |= DCMD_WIDTH1;
716 	else if (width == DMA_SLAVE_BUSWIDTH_2_BYTES)
717 		chan->dcmd |= DCMD_WIDTH2;
718 	else if (width == DMA_SLAVE_BUSWIDTH_4_BYTES)
719 		chan->dcmd |= DCMD_WIDTH4;
720 
721 	if (maxburst == 8)
722 		chan->dcmd |= DCMD_BURST8;
723 	else if (maxburst == 16)
724 		chan->dcmd |= DCMD_BURST16;
725 	else if (maxburst == 32)
726 		chan->dcmd |= DCMD_BURST32;
727 
728 	chan->dir = direction;
729 	chan->dev_addr = addr;
730 	/* FIXME: drivers should be ported over to use the filter
731 	 * function. Once that's done, the following two lines can
732 	 * be removed.
733 	 */
734 	if (cfg->slave_id)
735 		chan->drcmr = cfg->slave_id;
736 
737 	return 0;
738 }
739 
740 static int mmp_pdma_config(struct dma_chan *dchan,
741 			   struct dma_slave_config *cfg)
742 {
743 	struct mmp_pdma_chan *chan = to_mmp_pdma_chan(dchan);
744 
745 	memcpy(&chan->slave_config, cfg, sizeof(*cfg));
746 	return 0;
747 }
748 
749 static int mmp_pdma_terminate_all(struct dma_chan *dchan)
750 {
751 	struct mmp_pdma_chan *chan = to_mmp_pdma_chan(dchan);
752 	unsigned long flags;
753 
754 	if (!dchan)
755 		return -EINVAL;
756 
757 	disable_chan(chan->phy);
758 	mmp_pdma_free_phy(chan);
759 	spin_lock_irqsave(&chan->desc_lock, flags);
760 	mmp_pdma_free_desc_list(chan, &chan->chain_pending);
761 	mmp_pdma_free_desc_list(chan, &chan->chain_running);
762 	spin_unlock_irqrestore(&chan->desc_lock, flags);
763 	chan->idle = true;
764 
765 	return 0;
766 }
767 
768 static unsigned int mmp_pdma_residue(struct mmp_pdma_chan *chan,
769 				     dma_cookie_t cookie)
770 {
771 	struct mmp_pdma_desc_sw *sw;
772 	u32 curr, residue = 0;
773 	bool passed = false;
774 	bool cyclic = chan->cyclic_first != NULL;
775 
776 	/*
777 	 * If the channel does not have a phy pointer anymore, it has already
778 	 * been completed. Therefore, its residue is 0.
779 	 */
780 	if (!chan->phy)
781 		return 0;
782 
783 	if (chan->dir == DMA_DEV_TO_MEM)
784 		curr = readl(chan->phy->base + DTADR(chan->phy->idx));
785 	else
786 		curr = readl(chan->phy->base + DSADR(chan->phy->idx));
787 
788 	list_for_each_entry(sw, &chan->chain_running, node) {
789 		u32 start, end, len;
790 
791 		if (chan->dir == DMA_DEV_TO_MEM)
792 			start = sw->desc.dtadr;
793 		else
794 			start = sw->desc.dsadr;
795 
796 		len = sw->desc.dcmd & DCMD_LENGTH;
797 		end = start + len;
798 
799 		/*
800 		 * 'passed' will be latched once we found the descriptor which
801 		 * lies inside the boundaries of the curr pointer. All
802 		 * descriptors that occur in the list _after_ we found that
803 		 * partially handled descriptor are still to be processed and
804 		 * are hence added to the residual bytes counter.
805 		 */
806 
807 		if (passed) {
808 			residue += len;
809 		} else if (curr >= start && curr <= end) {
810 			residue += end - curr;
811 			passed = true;
812 		}
813 
814 		/*
815 		 * Descriptors that have the ENDIRQEN bit set mark the end of a
816 		 * transaction chain, and the cookie assigned with it has been
817 		 * returned previously from mmp_pdma_tx_submit().
818 		 *
819 		 * In case we have multiple transactions in the running chain,
820 		 * and the cookie does not match the one the user asked us
821 		 * about, reset the state variables and start over.
822 		 *
823 		 * This logic does not apply to cyclic transactions, where all
824 		 * descriptors have the ENDIRQEN bit set, and for which we
825 		 * can't have multiple transactions on one channel anyway.
826 		 */
827 		if (cyclic || !(sw->desc.dcmd & DCMD_ENDIRQEN))
828 			continue;
829 
830 		if (sw->async_tx.cookie == cookie) {
831 			return residue;
832 		} else {
833 			residue = 0;
834 			passed = false;
835 		}
836 	}
837 
838 	/* We should only get here in case of cyclic transactions */
839 	return residue;
840 }
841 
842 static enum dma_status mmp_pdma_tx_status(struct dma_chan *dchan,
843 					  dma_cookie_t cookie,
844 					  struct dma_tx_state *txstate)
845 {
846 	struct mmp_pdma_chan *chan = to_mmp_pdma_chan(dchan);
847 	enum dma_status ret;
848 
849 	ret = dma_cookie_status(dchan, cookie, txstate);
850 	if (likely(ret != DMA_ERROR))
851 		dma_set_residue(txstate, mmp_pdma_residue(chan, cookie));
852 
853 	return ret;
854 }
855 
856 /*
857  * mmp_pdma_issue_pending - Issue the DMA start command
858  * pending list ==> running list
859  */
860 static void mmp_pdma_issue_pending(struct dma_chan *dchan)
861 {
862 	struct mmp_pdma_chan *chan = to_mmp_pdma_chan(dchan);
863 	unsigned long flags;
864 
865 	spin_lock_irqsave(&chan->desc_lock, flags);
866 	start_pending_queue(chan);
867 	spin_unlock_irqrestore(&chan->desc_lock, flags);
868 }
869 
870 /*
871  * dma_do_tasklet
872  * Do call back
873  * Start pending list
874  */
875 static void dma_do_tasklet(struct tasklet_struct *t)
876 {
877 	struct mmp_pdma_chan *chan = from_tasklet(chan, t, tasklet);
878 	struct mmp_pdma_desc_sw *desc, *_desc;
879 	LIST_HEAD(chain_cleanup);
880 	unsigned long flags;
881 	struct dmaengine_desc_callback cb;
882 
883 	if (chan->cyclic_first) {
884 		spin_lock_irqsave(&chan->desc_lock, flags);
885 		desc = chan->cyclic_first;
886 		dmaengine_desc_get_callback(&desc->async_tx, &cb);
887 		spin_unlock_irqrestore(&chan->desc_lock, flags);
888 
889 		dmaengine_desc_callback_invoke(&cb, NULL);
890 
891 		return;
892 	}
893 
894 	/* submit pending list; callback for each desc; free desc */
895 	spin_lock_irqsave(&chan->desc_lock, flags);
896 
897 	list_for_each_entry_safe(desc, _desc, &chan->chain_running, node) {
898 		/*
899 		 * move the descriptors to a temporary list so we can drop
900 		 * the lock during the entire cleanup operation
901 		 */
902 		list_move(&desc->node, &chain_cleanup);
903 
904 		/*
905 		 * Look for the first list entry which has the ENDIRQEN flag
906 		 * set. That is the descriptor we got an interrupt for, so
907 		 * complete that transaction and its cookie.
908 		 */
909 		if (desc->desc.dcmd & DCMD_ENDIRQEN) {
910 			dma_cookie_t cookie = desc->async_tx.cookie;
911 			dma_cookie_complete(&desc->async_tx);
912 			dev_dbg(chan->dev, "completed_cookie=%d\n", cookie);
913 			break;
914 		}
915 	}
916 
917 	/*
918 	 * The hardware is idle and ready for more when the
919 	 * chain_running list is empty.
920 	 */
921 	chan->idle = list_empty(&chan->chain_running);
922 
923 	/* Start any pending transactions automatically */
924 	start_pending_queue(chan);
925 	spin_unlock_irqrestore(&chan->desc_lock, flags);
926 
927 	/* Run the callback for each descriptor, in order */
928 	list_for_each_entry_safe(desc, _desc, &chain_cleanup, node) {
929 		struct dma_async_tx_descriptor *txd = &desc->async_tx;
930 
931 		/* Remove from the list of transactions */
932 		list_del(&desc->node);
933 		/* Run the link descriptor callback function */
934 		dmaengine_desc_get_callback(txd, &cb);
935 		dmaengine_desc_callback_invoke(&cb, NULL);
936 
937 		dma_pool_free(chan->desc_pool, desc, txd->phys);
938 	}
939 }
940 
941 static int mmp_pdma_remove(struct platform_device *op)
942 {
943 	struct mmp_pdma_device *pdev = platform_get_drvdata(op);
944 	struct mmp_pdma_phy *phy;
945 	int i, irq = 0, irq_num = 0;
946 
947 	if (op->dev.of_node)
948 		of_dma_controller_free(op->dev.of_node);
949 
950 	for (i = 0; i < pdev->dma_channels; i++) {
951 		if (platform_get_irq(op, i) > 0)
952 			irq_num++;
953 	}
954 
955 	if (irq_num != pdev->dma_channels) {
956 		irq = platform_get_irq(op, 0);
957 		devm_free_irq(&op->dev, irq, pdev);
958 	} else {
959 		for (i = 0; i < pdev->dma_channels; i++) {
960 			phy = &pdev->phy[i];
961 			irq = platform_get_irq(op, i);
962 			devm_free_irq(&op->dev, irq, phy);
963 		}
964 	}
965 
966 	dma_async_device_unregister(&pdev->device);
967 	return 0;
968 }
969 
970 static int mmp_pdma_chan_init(struct mmp_pdma_device *pdev, int idx, int irq)
971 {
972 	struct mmp_pdma_phy *phy  = &pdev->phy[idx];
973 	struct mmp_pdma_chan *chan;
974 	int ret;
975 
976 	chan = devm_kzalloc(pdev->dev, sizeof(*chan), GFP_KERNEL);
977 	if (chan == NULL)
978 		return -ENOMEM;
979 
980 	phy->idx = idx;
981 	phy->base = pdev->base;
982 
983 	if (irq) {
984 		ret = devm_request_irq(pdev->dev, irq, mmp_pdma_chan_handler,
985 				       IRQF_SHARED, "pdma", phy);
986 		if (ret) {
987 			dev_err(pdev->dev, "channel request irq fail!\n");
988 			return ret;
989 		}
990 	}
991 
992 	spin_lock_init(&chan->desc_lock);
993 	chan->dev = pdev->dev;
994 	chan->chan.device = &pdev->device;
995 	tasklet_setup(&chan->tasklet, dma_do_tasklet);
996 	INIT_LIST_HEAD(&chan->chain_pending);
997 	INIT_LIST_HEAD(&chan->chain_running);
998 
999 	/* register virt channel to dma engine */
1000 	list_add_tail(&chan->chan.device_node, &pdev->device.channels);
1001 
1002 	return 0;
1003 }
1004 
1005 static const struct of_device_id mmp_pdma_dt_ids[] = {
1006 	{ .compatible = "marvell,pdma-1.0", },
1007 	{}
1008 };
1009 MODULE_DEVICE_TABLE(of, mmp_pdma_dt_ids);
1010 
1011 static struct dma_chan *mmp_pdma_dma_xlate(struct of_phandle_args *dma_spec,
1012 					   struct of_dma *ofdma)
1013 {
1014 	struct mmp_pdma_device *d = ofdma->of_dma_data;
1015 	struct dma_chan *chan;
1016 
1017 	chan = dma_get_any_slave_channel(&d->device);
1018 	if (!chan)
1019 		return NULL;
1020 
1021 	to_mmp_pdma_chan(chan)->drcmr = dma_spec->args[0];
1022 
1023 	return chan;
1024 }
1025 
1026 static int mmp_pdma_probe(struct platform_device *op)
1027 {
1028 	struct mmp_pdma_device *pdev;
1029 	const struct of_device_id *of_id;
1030 	struct mmp_dma_platdata *pdata = dev_get_platdata(&op->dev);
1031 	struct resource *iores;
1032 	int i, ret, irq = 0;
1033 	int dma_channels = 0, irq_num = 0;
1034 	const enum dma_slave_buswidth widths =
1035 		DMA_SLAVE_BUSWIDTH_1_BYTE   | DMA_SLAVE_BUSWIDTH_2_BYTES |
1036 		DMA_SLAVE_BUSWIDTH_4_BYTES;
1037 
1038 	pdev = devm_kzalloc(&op->dev, sizeof(*pdev), GFP_KERNEL);
1039 	if (!pdev)
1040 		return -ENOMEM;
1041 
1042 	pdev->dev = &op->dev;
1043 
1044 	spin_lock_init(&pdev->phy_lock);
1045 
1046 	iores = platform_get_resource(op, IORESOURCE_MEM, 0);
1047 	pdev->base = devm_ioremap_resource(pdev->dev, iores);
1048 	if (IS_ERR(pdev->base))
1049 		return PTR_ERR(pdev->base);
1050 
1051 	of_id = of_match_device(mmp_pdma_dt_ids, pdev->dev);
1052 	if (of_id)
1053 		of_property_read_u32(pdev->dev->of_node, "#dma-channels",
1054 				     &dma_channels);
1055 	else if (pdata && pdata->dma_channels)
1056 		dma_channels = pdata->dma_channels;
1057 	else
1058 		dma_channels = 32;	/* default 32 channel */
1059 	pdev->dma_channels = dma_channels;
1060 
1061 	for (i = 0; i < dma_channels; i++) {
1062 		if (platform_get_irq_optional(op, i) > 0)
1063 			irq_num++;
1064 	}
1065 
1066 	pdev->phy = devm_kcalloc(pdev->dev, dma_channels, sizeof(*pdev->phy),
1067 				 GFP_KERNEL);
1068 	if (pdev->phy == NULL)
1069 		return -ENOMEM;
1070 
1071 	INIT_LIST_HEAD(&pdev->device.channels);
1072 
1073 	if (irq_num != dma_channels) {
1074 		/* all chan share one irq, demux inside */
1075 		irq = platform_get_irq(op, 0);
1076 		ret = devm_request_irq(pdev->dev, irq, mmp_pdma_int_handler,
1077 				       IRQF_SHARED, "pdma", pdev);
1078 		if (ret)
1079 			return ret;
1080 	}
1081 
1082 	for (i = 0; i < dma_channels; i++) {
1083 		irq = (irq_num != dma_channels) ? 0 : platform_get_irq(op, i);
1084 		ret = mmp_pdma_chan_init(pdev, i, irq);
1085 		if (ret)
1086 			return ret;
1087 	}
1088 
1089 	dma_cap_set(DMA_SLAVE, pdev->device.cap_mask);
1090 	dma_cap_set(DMA_MEMCPY, pdev->device.cap_mask);
1091 	dma_cap_set(DMA_CYCLIC, pdev->device.cap_mask);
1092 	dma_cap_set(DMA_PRIVATE, pdev->device.cap_mask);
1093 	pdev->device.dev = &op->dev;
1094 	pdev->device.device_alloc_chan_resources = mmp_pdma_alloc_chan_resources;
1095 	pdev->device.device_free_chan_resources = mmp_pdma_free_chan_resources;
1096 	pdev->device.device_tx_status = mmp_pdma_tx_status;
1097 	pdev->device.device_prep_dma_memcpy = mmp_pdma_prep_memcpy;
1098 	pdev->device.device_prep_slave_sg = mmp_pdma_prep_slave_sg;
1099 	pdev->device.device_prep_dma_cyclic = mmp_pdma_prep_dma_cyclic;
1100 	pdev->device.device_issue_pending = mmp_pdma_issue_pending;
1101 	pdev->device.device_config = mmp_pdma_config;
1102 	pdev->device.device_terminate_all = mmp_pdma_terminate_all;
1103 	pdev->device.copy_align = DMAENGINE_ALIGN_8_BYTES;
1104 	pdev->device.src_addr_widths = widths;
1105 	pdev->device.dst_addr_widths = widths;
1106 	pdev->device.directions = BIT(DMA_MEM_TO_DEV) | BIT(DMA_DEV_TO_MEM);
1107 	pdev->device.residue_granularity = DMA_RESIDUE_GRANULARITY_DESCRIPTOR;
1108 
1109 	if (pdev->dev->coherent_dma_mask)
1110 		dma_set_mask(pdev->dev, pdev->dev->coherent_dma_mask);
1111 	else
1112 		dma_set_mask(pdev->dev, DMA_BIT_MASK(64));
1113 
1114 	ret = dma_async_device_register(&pdev->device);
1115 	if (ret) {
1116 		dev_err(pdev->device.dev, "unable to register\n");
1117 		return ret;
1118 	}
1119 
1120 	if (op->dev.of_node) {
1121 		/* Device-tree DMA controller registration */
1122 		ret = of_dma_controller_register(op->dev.of_node,
1123 						 mmp_pdma_dma_xlate, pdev);
1124 		if (ret < 0) {
1125 			dev_err(&op->dev, "of_dma_controller_register failed\n");
1126 			return ret;
1127 		}
1128 	}
1129 
1130 	platform_set_drvdata(op, pdev);
1131 	dev_info(pdev->device.dev, "initialized %d channels\n", dma_channels);
1132 	return 0;
1133 }
1134 
1135 static const struct platform_device_id mmp_pdma_id_table[] = {
1136 	{ "mmp-pdma", },
1137 	{ },
1138 };
1139 
1140 static struct platform_driver mmp_pdma_driver = {
1141 	.driver		= {
1142 		.name	= "mmp-pdma",
1143 		.of_match_table = mmp_pdma_dt_ids,
1144 	},
1145 	.id_table	= mmp_pdma_id_table,
1146 	.probe		= mmp_pdma_probe,
1147 	.remove		= mmp_pdma_remove,
1148 };
1149 
1150 module_platform_driver(mmp_pdma_driver);
1151 
1152 MODULE_DESCRIPTION("MARVELL MMP Peripheral DMA Driver");
1153 MODULE_AUTHOR("Marvell International Ltd.");
1154 MODULE_LICENSE("GPL v2");
1155