xref: /openbmc/linux/drivers/dma/sh/rcar-dmac.c (revision 82e6fdd6)
1 /*
2  * Renesas R-Car Gen2 DMA Controller Driver
3  *
4  * Copyright (C) 2014 Renesas Electronics Inc.
5  *
6  * Author: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
7  *
8  * This is free software; you can redistribute it and/or modify
9  * it under the terms of version 2 of the GNU General Public License as
10  * published by the Free Software Foundation.
11  */
12 
13 #include <linux/delay.h>
14 #include <linux/dma-mapping.h>
15 #include <linux/dmaengine.h>
16 #include <linux/interrupt.h>
17 #include <linux/list.h>
18 #include <linux/module.h>
19 #include <linux/mutex.h>
20 #include <linux/of.h>
21 #include <linux/of_dma.h>
22 #include <linux/of_platform.h>
23 #include <linux/platform_device.h>
24 #include <linux/pm_runtime.h>
25 #include <linux/slab.h>
26 #include <linux/spinlock.h>
27 
28 #include "../dmaengine.h"
29 
30 /*
31  * struct rcar_dmac_xfer_chunk - Descriptor for a hardware transfer
32  * @node: entry in the parent's chunks list
33  * @src_addr: device source address
34  * @dst_addr: device destination address
35  * @size: transfer size in bytes
36  */
37 struct rcar_dmac_xfer_chunk {
38 	struct list_head node;
39 
40 	dma_addr_t src_addr;
41 	dma_addr_t dst_addr;
42 	u32 size;
43 };
44 
45 /*
46  * struct rcar_dmac_hw_desc - Hardware descriptor for a transfer chunk
47  * @sar: value of the SAR register (source address)
48  * @dar: value of the DAR register (destination address)
49  * @tcr: value of the TCR register (transfer count)
50  */
51 struct rcar_dmac_hw_desc {
52 	u32 sar;
53 	u32 dar;
54 	u32 tcr;
55 	u32 reserved;
56 } __attribute__((__packed__));
57 
58 /*
59  * struct rcar_dmac_desc - R-Car Gen2 DMA Transfer Descriptor
60  * @async_tx: base DMA asynchronous transaction descriptor
61  * @direction: direction of the DMA transfer
62  * @xfer_shift: log2 of the transfer size
63  * @chcr: value of the channel configuration register for this transfer
64  * @node: entry in the channel's descriptors lists
65  * @chunks: list of transfer chunks for this transfer
66  * @running: the transfer chunk being currently processed
67  * @nchunks: number of transfer chunks for this transfer
68  * @hwdescs.use: whether the transfer descriptor uses hardware descriptors
69  * @hwdescs.mem: hardware descriptors memory for the transfer
70  * @hwdescs.dma: device address of the hardware descriptors memory
71  * @hwdescs.size: size of the hardware descriptors in bytes
72  * @size: transfer size in bytes
73  * @cyclic: when set indicates that the DMA transfer is cyclic
74  */
75 struct rcar_dmac_desc {
76 	struct dma_async_tx_descriptor async_tx;
77 	enum dma_transfer_direction direction;
78 	unsigned int xfer_shift;
79 	u32 chcr;
80 
81 	struct list_head node;
82 	struct list_head chunks;
83 	struct rcar_dmac_xfer_chunk *running;
84 	unsigned int nchunks;
85 
86 	struct {
87 		bool use;
88 		struct rcar_dmac_hw_desc *mem;
89 		dma_addr_t dma;
90 		size_t size;
91 	} hwdescs;
92 
93 	unsigned int size;
94 	bool cyclic;
95 };
96 
97 #define to_rcar_dmac_desc(d)	container_of(d, struct rcar_dmac_desc, async_tx)
98 
99 /*
100  * struct rcar_dmac_desc_page - One page worth of descriptors
101  * @node: entry in the channel's pages list
102  * @descs: array of DMA descriptors
103  * @chunks: array of transfer chunk descriptors
104  */
105 struct rcar_dmac_desc_page {
106 	struct list_head node;
107 
108 	union {
109 		struct rcar_dmac_desc descs[0];
110 		struct rcar_dmac_xfer_chunk chunks[0];
111 	};
112 };
113 
114 #define RCAR_DMAC_DESCS_PER_PAGE					\
115 	((PAGE_SIZE - offsetof(struct rcar_dmac_desc_page, descs)) /	\
116 	sizeof(struct rcar_dmac_desc))
117 #define RCAR_DMAC_XFER_CHUNKS_PER_PAGE					\
118 	((PAGE_SIZE - offsetof(struct rcar_dmac_desc_page, chunks)) /	\
119 	sizeof(struct rcar_dmac_xfer_chunk))
120 
121 /*
122  * struct rcar_dmac_chan_slave - Slave configuration
123  * @slave_addr: slave memory address
124  * @xfer_size: size (in bytes) of hardware transfers
125  */
126 struct rcar_dmac_chan_slave {
127 	phys_addr_t slave_addr;
128 	unsigned int xfer_size;
129 };
130 
131 /*
132  * struct rcar_dmac_chan_map - Map of slave device phys to dma address
133  * @addr: slave dma address
134  * @dir: direction of mapping
135  * @slave: slave configuration that is mapped
136  */
137 struct rcar_dmac_chan_map {
138 	dma_addr_t addr;
139 	enum dma_data_direction dir;
140 	struct rcar_dmac_chan_slave slave;
141 };
142 
143 /*
144  * struct rcar_dmac_chan - R-Car Gen2 DMA Controller Channel
145  * @chan: base DMA channel object
146  * @iomem: channel I/O memory base
147  * @index: index of this channel in the controller
148  * @irq: channel IRQ
149  * @src: slave memory address and size on the source side
150  * @dst: slave memory address and size on the destination side
151  * @mid_rid: hardware MID/RID for the DMA client using this channel
152  * @lock: protects the channel CHCR register and the desc members
153  * @desc.free: list of free descriptors
154  * @desc.pending: list of pending descriptors (submitted with tx_submit)
155  * @desc.active: list of active descriptors (activated with issue_pending)
156  * @desc.done: list of completed descriptors
157  * @desc.wait: list of descriptors waiting for an ack
158  * @desc.running: the descriptor being processed (a member of the active list)
159  * @desc.chunks_free: list of free transfer chunk descriptors
160  * @desc.pages: list of pages used by allocated descriptors
161  */
162 struct rcar_dmac_chan {
163 	struct dma_chan chan;
164 	void __iomem *iomem;
165 	unsigned int index;
166 	int irq;
167 
168 	struct rcar_dmac_chan_slave src;
169 	struct rcar_dmac_chan_slave dst;
170 	struct rcar_dmac_chan_map map;
171 	int mid_rid;
172 
173 	spinlock_t lock;
174 
175 	struct {
176 		struct list_head free;
177 		struct list_head pending;
178 		struct list_head active;
179 		struct list_head done;
180 		struct list_head wait;
181 		struct rcar_dmac_desc *running;
182 
183 		struct list_head chunks_free;
184 
185 		struct list_head pages;
186 	} desc;
187 };
188 
189 #define to_rcar_dmac_chan(c)	container_of(c, struct rcar_dmac_chan, chan)
190 
191 /*
192  * struct rcar_dmac - R-Car Gen2 DMA Controller
193  * @engine: base DMA engine object
194  * @dev: the hardware device
195  * @iomem: remapped I/O memory base
196  * @n_channels: number of available channels
197  * @channels: array of DMAC channels
198  * @modules: bitmask of client modules in use
199  */
200 struct rcar_dmac {
201 	struct dma_device engine;
202 	struct device *dev;
203 	void __iomem *iomem;
204 
205 	unsigned int n_channels;
206 	struct rcar_dmac_chan *channels;
207 
208 	DECLARE_BITMAP(modules, 256);
209 };
210 
211 #define to_rcar_dmac(d)		container_of(d, struct rcar_dmac, engine)
212 
213 /* -----------------------------------------------------------------------------
214  * Registers
215  */
216 
217 #define RCAR_DMAC_CHAN_OFFSET(i)	(0x8000 + 0x80 * (i))
218 
219 #define RCAR_DMAISTA			0x0020
220 #define RCAR_DMASEC			0x0030
221 #define RCAR_DMAOR			0x0060
222 #define RCAR_DMAOR_PRI_FIXED		(0 << 8)
223 #define RCAR_DMAOR_PRI_ROUND_ROBIN	(3 << 8)
224 #define RCAR_DMAOR_AE			(1 << 2)
225 #define RCAR_DMAOR_DME			(1 << 0)
226 #define RCAR_DMACHCLR			0x0080
227 #define RCAR_DMADPSEC			0x00a0
228 
229 #define RCAR_DMASAR			0x0000
230 #define RCAR_DMADAR			0x0004
231 #define RCAR_DMATCR			0x0008
232 #define RCAR_DMATCR_MASK		0x00ffffff
233 #define RCAR_DMATSR			0x0028
234 #define RCAR_DMACHCR			0x000c
235 #define RCAR_DMACHCR_CAE		(1 << 31)
236 #define RCAR_DMACHCR_CAIE		(1 << 30)
237 #define RCAR_DMACHCR_DPM_DISABLED	(0 << 28)
238 #define RCAR_DMACHCR_DPM_ENABLED	(1 << 28)
239 #define RCAR_DMACHCR_DPM_REPEAT		(2 << 28)
240 #define RCAR_DMACHCR_DPM_INFINITE	(3 << 28)
241 #define RCAR_DMACHCR_RPT_SAR		(1 << 27)
242 #define RCAR_DMACHCR_RPT_DAR		(1 << 26)
243 #define RCAR_DMACHCR_RPT_TCR		(1 << 25)
244 #define RCAR_DMACHCR_DPB		(1 << 22)
245 #define RCAR_DMACHCR_DSE		(1 << 19)
246 #define RCAR_DMACHCR_DSIE		(1 << 18)
247 #define RCAR_DMACHCR_TS_1B		((0 << 20) | (0 << 3))
248 #define RCAR_DMACHCR_TS_2B		((0 << 20) | (1 << 3))
249 #define RCAR_DMACHCR_TS_4B		((0 << 20) | (2 << 3))
250 #define RCAR_DMACHCR_TS_16B		((0 << 20) | (3 << 3))
251 #define RCAR_DMACHCR_TS_32B		((1 << 20) | (0 << 3))
252 #define RCAR_DMACHCR_TS_64B		((1 << 20) | (1 << 3))
253 #define RCAR_DMACHCR_TS_8B		((1 << 20) | (3 << 3))
254 #define RCAR_DMACHCR_DM_FIXED		(0 << 14)
255 #define RCAR_DMACHCR_DM_INC		(1 << 14)
256 #define RCAR_DMACHCR_DM_DEC		(2 << 14)
257 #define RCAR_DMACHCR_SM_FIXED		(0 << 12)
258 #define RCAR_DMACHCR_SM_INC		(1 << 12)
259 #define RCAR_DMACHCR_SM_DEC		(2 << 12)
260 #define RCAR_DMACHCR_RS_AUTO		(4 << 8)
261 #define RCAR_DMACHCR_RS_DMARS		(8 << 8)
262 #define RCAR_DMACHCR_IE			(1 << 2)
263 #define RCAR_DMACHCR_TE			(1 << 1)
264 #define RCAR_DMACHCR_DE			(1 << 0)
265 #define RCAR_DMATCRB			0x0018
266 #define RCAR_DMATSRB			0x0038
267 #define RCAR_DMACHCRB			0x001c
268 #define RCAR_DMACHCRB_DCNT(n)		((n) << 24)
269 #define RCAR_DMACHCRB_DPTR_MASK		(0xff << 16)
270 #define RCAR_DMACHCRB_DPTR_SHIFT	16
271 #define RCAR_DMACHCRB_DRST		(1 << 15)
272 #define RCAR_DMACHCRB_DTS		(1 << 8)
273 #define RCAR_DMACHCRB_SLM_NORMAL	(0 << 4)
274 #define RCAR_DMACHCRB_SLM_CLK(n)	((8 | (n)) << 4)
275 #define RCAR_DMACHCRB_PRI(n)		((n) << 0)
276 #define RCAR_DMARS			0x0040
277 #define RCAR_DMABUFCR			0x0048
278 #define RCAR_DMABUFCR_MBU(n)		((n) << 16)
279 #define RCAR_DMABUFCR_ULB(n)		((n) << 0)
280 #define RCAR_DMADPBASE			0x0050
281 #define RCAR_DMADPBASE_MASK		0xfffffff0
282 #define RCAR_DMADPBASE_SEL		(1 << 0)
283 #define RCAR_DMADPCR			0x0054
284 #define RCAR_DMADPCR_DIPT(n)		((n) << 24)
285 #define RCAR_DMAFIXSAR			0x0010
286 #define RCAR_DMAFIXDAR			0x0014
287 #define RCAR_DMAFIXDPBASE		0x0060
288 
289 /* Hardcode the MEMCPY transfer size to 4 bytes. */
290 #define RCAR_DMAC_MEMCPY_XFER_SIZE	4
291 
292 /* -----------------------------------------------------------------------------
293  * Device access
294  */
295 
296 static void rcar_dmac_write(struct rcar_dmac *dmac, u32 reg, u32 data)
297 {
298 	if (reg == RCAR_DMAOR)
299 		writew(data, dmac->iomem + reg);
300 	else
301 		writel(data, dmac->iomem + reg);
302 }
303 
304 static u32 rcar_dmac_read(struct rcar_dmac *dmac, u32 reg)
305 {
306 	if (reg == RCAR_DMAOR)
307 		return readw(dmac->iomem + reg);
308 	else
309 		return readl(dmac->iomem + reg);
310 }
311 
312 static u32 rcar_dmac_chan_read(struct rcar_dmac_chan *chan, u32 reg)
313 {
314 	if (reg == RCAR_DMARS)
315 		return readw(chan->iomem + reg);
316 	else
317 		return readl(chan->iomem + reg);
318 }
319 
320 static void rcar_dmac_chan_write(struct rcar_dmac_chan *chan, u32 reg, u32 data)
321 {
322 	if (reg == RCAR_DMARS)
323 		writew(data, chan->iomem + reg);
324 	else
325 		writel(data, chan->iomem + reg);
326 }
327 
328 /* -----------------------------------------------------------------------------
329  * Initialization and configuration
330  */
331 
332 static bool rcar_dmac_chan_is_busy(struct rcar_dmac_chan *chan)
333 {
334 	u32 chcr = rcar_dmac_chan_read(chan, RCAR_DMACHCR);
335 
336 	return !!(chcr & (RCAR_DMACHCR_DE | RCAR_DMACHCR_TE));
337 }
338 
339 static void rcar_dmac_chan_start_xfer(struct rcar_dmac_chan *chan)
340 {
341 	struct rcar_dmac_desc *desc = chan->desc.running;
342 	u32 chcr = desc->chcr;
343 
344 	WARN_ON_ONCE(rcar_dmac_chan_is_busy(chan));
345 
346 	if (chan->mid_rid >= 0)
347 		rcar_dmac_chan_write(chan, RCAR_DMARS, chan->mid_rid);
348 
349 	if (desc->hwdescs.use) {
350 		struct rcar_dmac_xfer_chunk *chunk =
351 			list_first_entry(&desc->chunks,
352 					 struct rcar_dmac_xfer_chunk, node);
353 
354 		dev_dbg(chan->chan.device->dev,
355 			"chan%u: queue desc %p: %u@%pad\n",
356 			chan->index, desc, desc->nchunks, &desc->hwdescs.dma);
357 
358 #ifdef CONFIG_ARCH_DMA_ADDR_T_64BIT
359 		rcar_dmac_chan_write(chan, RCAR_DMAFIXSAR,
360 				     chunk->src_addr >> 32);
361 		rcar_dmac_chan_write(chan, RCAR_DMAFIXDAR,
362 				     chunk->dst_addr >> 32);
363 		rcar_dmac_chan_write(chan, RCAR_DMAFIXDPBASE,
364 				     desc->hwdescs.dma >> 32);
365 #endif
366 		rcar_dmac_chan_write(chan, RCAR_DMADPBASE,
367 				     (desc->hwdescs.dma & 0xfffffff0) |
368 				     RCAR_DMADPBASE_SEL);
369 		rcar_dmac_chan_write(chan, RCAR_DMACHCRB,
370 				     RCAR_DMACHCRB_DCNT(desc->nchunks - 1) |
371 				     RCAR_DMACHCRB_DRST);
372 
373 		/*
374 		 * Errata: When descriptor memory is accessed through an IOMMU
375 		 * the DMADAR register isn't initialized automatically from the
376 		 * first descriptor at beginning of transfer by the DMAC like it
377 		 * should. Initialize it manually with the destination address
378 		 * of the first chunk.
379 		 */
380 		rcar_dmac_chan_write(chan, RCAR_DMADAR,
381 				     chunk->dst_addr & 0xffffffff);
382 
383 		/*
384 		 * Program the descriptor stage interrupt to occur after the end
385 		 * of the first stage.
386 		 */
387 		rcar_dmac_chan_write(chan, RCAR_DMADPCR, RCAR_DMADPCR_DIPT(1));
388 
389 		chcr |= RCAR_DMACHCR_RPT_SAR | RCAR_DMACHCR_RPT_DAR
390 		     |  RCAR_DMACHCR_RPT_TCR | RCAR_DMACHCR_DPB;
391 
392 		/*
393 		 * If the descriptor isn't cyclic enable normal descriptor mode
394 		 * and the transfer completion interrupt.
395 		 */
396 		if (!desc->cyclic)
397 			chcr |= RCAR_DMACHCR_DPM_ENABLED | RCAR_DMACHCR_IE;
398 		/*
399 		 * If the descriptor is cyclic and has a callback enable the
400 		 * descriptor stage interrupt in infinite repeat mode.
401 		 */
402 		else if (desc->async_tx.callback)
403 			chcr |= RCAR_DMACHCR_DPM_INFINITE | RCAR_DMACHCR_DSIE;
404 		/*
405 		 * Otherwise just select infinite repeat mode without any
406 		 * interrupt.
407 		 */
408 		else
409 			chcr |= RCAR_DMACHCR_DPM_INFINITE;
410 	} else {
411 		struct rcar_dmac_xfer_chunk *chunk = desc->running;
412 
413 		dev_dbg(chan->chan.device->dev,
414 			"chan%u: queue chunk %p: %u@%pad -> %pad\n",
415 			chan->index, chunk, chunk->size, &chunk->src_addr,
416 			&chunk->dst_addr);
417 
418 #ifdef CONFIG_ARCH_DMA_ADDR_T_64BIT
419 		rcar_dmac_chan_write(chan, RCAR_DMAFIXSAR,
420 				     chunk->src_addr >> 32);
421 		rcar_dmac_chan_write(chan, RCAR_DMAFIXDAR,
422 				     chunk->dst_addr >> 32);
423 #endif
424 		rcar_dmac_chan_write(chan, RCAR_DMASAR,
425 				     chunk->src_addr & 0xffffffff);
426 		rcar_dmac_chan_write(chan, RCAR_DMADAR,
427 				     chunk->dst_addr & 0xffffffff);
428 		rcar_dmac_chan_write(chan, RCAR_DMATCR,
429 				     chunk->size >> desc->xfer_shift);
430 
431 		chcr |= RCAR_DMACHCR_DPM_DISABLED | RCAR_DMACHCR_IE;
432 	}
433 
434 	rcar_dmac_chan_write(chan, RCAR_DMACHCR, chcr | RCAR_DMACHCR_DE);
435 }
436 
437 static int rcar_dmac_init(struct rcar_dmac *dmac)
438 {
439 	u16 dmaor;
440 
441 	/* Clear all channels and enable the DMAC globally. */
442 	rcar_dmac_write(dmac, RCAR_DMACHCLR, GENMASK(dmac->n_channels - 1, 0));
443 	rcar_dmac_write(dmac, RCAR_DMAOR,
444 			RCAR_DMAOR_PRI_FIXED | RCAR_DMAOR_DME);
445 
446 	dmaor = rcar_dmac_read(dmac, RCAR_DMAOR);
447 	if ((dmaor & (RCAR_DMAOR_AE | RCAR_DMAOR_DME)) != RCAR_DMAOR_DME) {
448 		dev_warn(dmac->dev, "DMAOR initialization failed.\n");
449 		return -EIO;
450 	}
451 
452 	return 0;
453 }
454 
455 /* -----------------------------------------------------------------------------
456  * Descriptors submission
457  */
458 
459 static dma_cookie_t rcar_dmac_tx_submit(struct dma_async_tx_descriptor *tx)
460 {
461 	struct rcar_dmac_chan *chan = to_rcar_dmac_chan(tx->chan);
462 	struct rcar_dmac_desc *desc = to_rcar_dmac_desc(tx);
463 	unsigned long flags;
464 	dma_cookie_t cookie;
465 
466 	spin_lock_irqsave(&chan->lock, flags);
467 
468 	cookie = dma_cookie_assign(tx);
469 
470 	dev_dbg(chan->chan.device->dev, "chan%u: submit #%d@%p\n",
471 		chan->index, tx->cookie, desc);
472 
473 	list_add_tail(&desc->node, &chan->desc.pending);
474 	desc->running = list_first_entry(&desc->chunks,
475 					 struct rcar_dmac_xfer_chunk, node);
476 
477 	spin_unlock_irqrestore(&chan->lock, flags);
478 
479 	return cookie;
480 }
481 
482 /* -----------------------------------------------------------------------------
483  * Descriptors allocation and free
484  */
485 
486 /*
487  * rcar_dmac_desc_alloc - Allocate a page worth of DMA descriptors
488  * @chan: the DMA channel
489  * @gfp: allocation flags
490  */
491 static int rcar_dmac_desc_alloc(struct rcar_dmac_chan *chan, gfp_t gfp)
492 {
493 	struct rcar_dmac_desc_page *page;
494 	unsigned long flags;
495 	LIST_HEAD(list);
496 	unsigned int i;
497 
498 	page = (void *)get_zeroed_page(gfp);
499 	if (!page)
500 		return -ENOMEM;
501 
502 	for (i = 0; i < RCAR_DMAC_DESCS_PER_PAGE; ++i) {
503 		struct rcar_dmac_desc *desc = &page->descs[i];
504 
505 		dma_async_tx_descriptor_init(&desc->async_tx, &chan->chan);
506 		desc->async_tx.tx_submit = rcar_dmac_tx_submit;
507 		INIT_LIST_HEAD(&desc->chunks);
508 
509 		list_add_tail(&desc->node, &list);
510 	}
511 
512 	spin_lock_irqsave(&chan->lock, flags);
513 	list_splice_tail(&list, &chan->desc.free);
514 	list_add_tail(&page->node, &chan->desc.pages);
515 	spin_unlock_irqrestore(&chan->lock, flags);
516 
517 	return 0;
518 }
519 
520 /*
521  * rcar_dmac_desc_put - Release a DMA transfer descriptor
522  * @chan: the DMA channel
523  * @desc: the descriptor
524  *
525  * Put the descriptor and its transfer chunk descriptors back in the channel's
526  * free descriptors lists. The descriptor's chunks list will be reinitialized to
527  * an empty list as a result.
528  *
529  * The descriptor must have been removed from the channel's lists before calling
530  * this function.
531  */
532 static void rcar_dmac_desc_put(struct rcar_dmac_chan *chan,
533 			       struct rcar_dmac_desc *desc)
534 {
535 	unsigned long flags;
536 
537 	spin_lock_irqsave(&chan->lock, flags);
538 	list_splice_tail_init(&desc->chunks, &chan->desc.chunks_free);
539 	list_add(&desc->node, &chan->desc.free);
540 	spin_unlock_irqrestore(&chan->lock, flags);
541 }
542 
543 static void rcar_dmac_desc_recycle_acked(struct rcar_dmac_chan *chan)
544 {
545 	struct rcar_dmac_desc *desc, *_desc;
546 	unsigned long flags;
547 	LIST_HEAD(list);
548 
549 	/*
550 	 * We have to temporarily move all descriptors from the wait list to a
551 	 * local list as iterating over the wait list, even with
552 	 * list_for_each_entry_safe, isn't safe if we release the channel lock
553 	 * around the rcar_dmac_desc_put() call.
554 	 */
555 	spin_lock_irqsave(&chan->lock, flags);
556 	list_splice_init(&chan->desc.wait, &list);
557 	spin_unlock_irqrestore(&chan->lock, flags);
558 
559 	list_for_each_entry_safe(desc, _desc, &list, node) {
560 		if (async_tx_test_ack(&desc->async_tx)) {
561 			list_del(&desc->node);
562 			rcar_dmac_desc_put(chan, desc);
563 		}
564 	}
565 
566 	if (list_empty(&list))
567 		return;
568 
569 	/* Put the remaining descriptors back in the wait list. */
570 	spin_lock_irqsave(&chan->lock, flags);
571 	list_splice(&list, &chan->desc.wait);
572 	spin_unlock_irqrestore(&chan->lock, flags);
573 }
574 
575 /*
576  * rcar_dmac_desc_get - Allocate a descriptor for a DMA transfer
577  * @chan: the DMA channel
578  *
579  * Locking: This function must be called in a non-atomic context.
580  *
581  * Return: A pointer to the allocated descriptor or NULL if no descriptor can
582  * be allocated.
583  */
584 static struct rcar_dmac_desc *rcar_dmac_desc_get(struct rcar_dmac_chan *chan)
585 {
586 	struct rcar_dmac_desc *desc;
587 	unsigned long flags;
588 	int ret;
589 
590 	/* Recycle acked descriptors before attempting allocation. */
591 	rcar_dmac_desc_recycle_acked(chan);
592 
593 	spin_lock_irqsave(&chan->lock, flags);
594 
595 	while (list_empty(&chan->desc.free)) {
596 		/*
597 		 * No free descriptors, allocate a page worth of them and try
598 		 * again, as someone else could race us to get the newly
599 		 * allocated descriptors. If the allocation fails return an
600 		 * error.
601 		 */
602 		spin_unlock_irqrestore(&chan->lock, flags);
603 		ret = rcar_dmac_desc_alloc(chan, GFP_NOWAIT);
604 		if (ret < 0)
605 			return NULL;
606 		spin_lock_irqsave(&chan->lock, flags);
607 	}
608 
609 	desc = list_first_entry(&chan->desc.free, struct rcar_dmac_desc, node);
610 	list_del(&desc->node);
611 
612 	spin_unlock_irqrestore(&chan->lock, flags);
613 
614 	return desc;
615 }
616 
617 /*
618  * rcar_dmac_xfer_chunk_alloc - Allocate a page worth of transfer chunks
619  * @chan: the DMA channel
620  * @gfp: allocation flags
621  */
622 static int rcar_dmac_xfer_chunk_alloc(struct rcar_dmac_chan *chan, gfp_t gfp)
623 {
624 	struct rcar_dmac_desc_page *page;
625 	unsigned long flags;
626 	LIST_HEAD(list);
627 	unsigned int i;
628 
629 	page = (void *)get_zeroed_page(gfp);
630 	if (!page)
631 		return -ENOMEM;
632 
633 	for (i = 0; i < RCAR_DMAC_XFER_CHUNKS_PER_PAGE; ++i) {
634 		struct rcar_dmac_xfer_chunk *chunk = &page->chunks[i];
635 
636 		list_add_tail(&chunk->node, &list);
637 	}
638 
639 	spin_lock_irqsave(&chan->lock, flags);
640 	list_splice_tail(&list, &chan->desc.chunks_free);
641 	list_add_tail(&page->node, &chan->desc.pages);
642 	spin_unlock_irqrestore(&chan->lock, flags);
643 
644 	return 0;
645 }
646 
647 /*
648  * rcar_dmac_xfer_chunk_get - Allocate a transfer chunk for a DMA transfer
649  * @chan: the DMA channel
650  *
651  * Locking: This function must be called in a non-atomic context.
652  *
653  * Return: A pointer to the allocated transfer chunk descriptor or NULL if no
654  * descriptor can be allocated.
655  */
656 static struct rcar_dmac_xfer_chunk *
657 rcar_dmac_xfer_chunk_get(struct rcar_dmac_chan *chan)
658 {
659 	struct rcar_dmac_xfer_chunk *chunk;
660 	unsigned long flags;
661 	int ret;
662 
663 	spin_lock_irqsave(&chan->lock, flags);
664 
665 	while (list_empty(&chan->desc.chunks_free)) {
666 		/*
667 		 * No free descriptors, allocate a page worth of them and try
668 		 * again, as someone else could race us to get the newly
669 		 * allocated descriptors. If the allocation fails return an
670 		 * error.
671 		 */
672 		spin_unlock_irqrestore(&chan->lock, flags);
673 		ret = rcar_dmac_xfer_chunk_alloc(chan, GFP_NOWAIT);
674 		if (ret < 0)
675 			return NULL;
676 		spin_lock_irqsave(&chan->lock, flags);
677 	}
678 
679 	chunk = list_first_entry(&chan->desc.chunks_free,
680 				 struct rcar_dmac_xfer_chunk, node);
681 	list_del(&chunk->node);
682 
683 	spin_unlock_irqrestore(&chan->lock, flags);
684 
685 	return chunk;
686 }
687 
688 static void rcar_dmac_realloc_hwdesc(struct rcar_dmac_chan *chan,
689 				     struct rcar_dmac_desc *desc, size_t size)
690 {
691 	/*
692 	 * dma_alloc_coherent() allocates memory in page size increments. To
693 	 * avoid reallocating the hardware descriptors when the allocated size
694 	 * wouldn't change align the requested size to a multiple of the page
695 	 * size.
696 	 */
697 	size = PAGE_ALIGN(size);
698 
699 	if (desc->hwdescs.size == size)
700 		return;
701 
702 	if (desc->hwdescs.mem) {
703 		dma_free_coherent(chan->chan.device->dev, desc->hwdescs.size,
704 				  desc->hwdescs.mem, desc->hwdescs.dma);
705 		desc->hwdescs.mem = NULL;
706 		desc->hwdescs.size = 0;
707 	}
708 
709 	if (!size)
710 		return;
711 
712 	desc->hwdescs.mem = dma_alloc_coherent(chan->chan.device->dev, size,
713 					       &desc->hwdescs.dma, GFP_NOWAIT);
714 	if (!desc->hwdescs.mem)
715 		return;
716 
717 	desc->hwdescs.size = size;
718 }
719 
720 static int rcar_dmac_fill_hwdesc(struct rcar_dmac_chan *chan,
721 				 struct rcar_dmac_desc *desc)
722 {
723 	struct rcar_dmac_xfer_chunk *chunk;
724 	struct rcar_dmac_hw_desc *hwdesc;
725 
726 	rcar_dmac_realloc_hwdesc(chan, desc, desc->nchunks * sizeof(*hwdesc));
727 
728 	hwdesc = desc->hwdescs.mem;
729 	if (!hwdesc)
730 		return -ENOMEM;
731 
732 	list_for_each_entry(chunk, &desc->chunks, node) {
733 		hwdesc->sar = chunk->src_addr;
734 		hwdesc->dar = chunk->dst_addr;
735 		hwdesc->tcr = chunk->size >> desc->xfer_shift;
736 		hwdesc++;
737 	}
738 
739 	return 0;
740 }
741 
742 /* -----------------------------------------------------------------------------
743  * Stop and reset
744  */
745 static void rcar_dmac_chcr_de_barrier(struct rcar_dmac_chan *chan)
746 {
747 	u32 chcr;
748 	unsigned int i;
749 
750 	/*
751 	 * Ensure that the setting of the DE bit is actually 0 after
752 	 * clearing it.
753 	 */
754 	for (i = 0; i < 1024; i++) {
755 		chcr = rcar_dmac_chan_read(chan, RCAR_DMACHCR);
756 		if (!(chcr & RCAR_DMACHCR_DE))
757 			return;
758 		udelay(1);
759 	}
760 
761 	dev_err(chan->chan.device->dev, "CHCR DE check error\n");
762 }
763 
764 static void rcar_dmac_sync_tcr(struct rcar_dmac_chan *chan)
765 {
766 	u32 chcr = rcar_dmac_chan_read(chan, RCAR_DMACHCR);
767 
768 	if (!(chcr & RCAR_DMACHCR_DE))
769 		return;
770 
771 	/* set DE=0 and flush remaining data */
772 	rcar_dmac_chan_write(chan, RCAR_DMACHCR, (chcr & ~RCAR_DMACHCR_DE));
773 
774 	/* make sure all remaining data was flushed */
775 	rcar_dmac_chcr_de_barrier(chan);
776 
777 	/* back DE */
778 	rcar_dmac_chan_write(chan, RCAR_DMACHCR, chcr);
779 }
780 
781 static void rcar_dmac_chan_halt(struct rcar_dmac_chan *chan)
782 {
783 	u32 chcr = rcar_dmac_chan_read(chan, RCAR_DMACHCR);
784 
785 	chcr &= ~(RCAR_DMACHCR_DSE | RCAR_DMACHCR_DSIE | RCAR_DMACHCR_IE |
786 		  RCAR_DMACHCR_TE | RCAR_DMACHCR_DE);
787 	rcar_dmac_chan_write(chan, RCAR_DMACHCR, chcr);
788 	rcar_dmac_chcr_de_barrier(chan);
789 }
790 
791 static void rcar_dmac_chan_reinit(struct rcar_dmac_chan *chan)
792 {
793 	struct rcar_dmac_desc *desc, *_desc;
794 	unsigned long flags;
795 	LIST_HEAD(descs);
796 
797 	spin_lock_irqsave(&chan->lock, flags);
798 
799 	/* Move all non-free descriptors to the local lists. */
800 	list_splice_init(&chan->desc.pending, &descs);
801 	list_splice_init(&chan->desc.active, &descs);
802 	list_splice_init(&chan->desc.done, &descs);
803 	list_splice_init(&chan->desc.wait, &descs);
804 
805 	chan->desc.running = NULL;
806 
807 	spin_unlock_irqrestore(&chan->lock, flags);
808 
809 	list_for_each_entry_safe(desc, _desc, &descs, node) {
810 		list_del(&desc->node);
811 		rcar_dmac_desc_put(chan, desc);
812 	}
813 }
814 
815 static void rcar_dmac_stop(struct rcar_dmac *dmac)
816 {
817 	rcar_dmac_write(dmac, RCAR_DMAOR, 0);
818 }
819 
820 static void rcar_dmac_abort(struct rcar_dmac *dmac)
821 {
822 	unsigned int i;
823 
824 	/* Stop all channels. */
825 	for (i = 0; i < dmac->n_channels; ++i) {
826 		struct rcar_dmac_chan *chan = &dmac->channels[i];
827 
828 		/* Stop and reinitialize the channel. */
829 		spin_lock(&chan->lock);
830 		rcar_dmac_chan_halt(chan);
831 		spin_unlock(&chan->lock);
832 
833 		rcar_dmac_chan_reinit(chan);
834 	}
835 }
836 
837 /* -----------------------------------------------------------------------------
838  * Descriptors preparation
839  */
840 
841 static void rcar_dmac_chan_configure_desc(struct rcar_dmac_chan *chan,
842 					  struct rcar_dmac_desc *desc)
843 {
844 	static const u32 chcr_ts[] = {
845 		RCAR_DMACHCR_TS_1B, RCAR_DMACHCR_TS_2B,
846 		RCAR_DMACHCR_TS_4B, RCAR_DMACHCR_TS_8B,
847 		RCAR_DMACHCR_TS_16B, RCAR_DMACHCR_TS_32B,
848 		RCAR_DMACHCR_TS_64B,
849 	};
850 
851 	unsigned int xfer_size;
852 	u32 chcr;
853 
854 	switch (desc->direction) {
855 	case DMA_DEV_TO_MEM:
856 		chcr = RCAR_DMACHCR_DM_INC | RCAR_DMACHCR_SM_FIXED
857 		     | RCAR_DMACHCR_RS_DMARS;
858 		xfer_size = chan->src.xfer_size;
859 		break;
860 
861 	case DMA_MEM_TO_DEV:
862 		chcr = RCAR_DMACHCR_DM_FIXED | RCAR_DMACHCR_SM_INC
863 		     | RCAR_DMACHCR_RS_DMARS;
864 		xfer_size = chan->dst.xfer_size;
865 		break;
866 
867 	case DMA_MEM_TO_MEM:
868 	default:
869 		chcr = RCAR_DMACHCR_DM_INC | RCAR_DMACHCR_SM_INC
870 		     | RCAR_DMACHCR_RS_AUTO;
871 		xfer_size = RCAR_DMAC_MEMCPY_XFER_SIZE;
872 		break;
873 	}
874 
875 	desc->xfer_shift = ilog2(xfer_size);
876 	desc->chcr = chcr | chcr_ts[desc->xfer_shift];
877 }
878 
879 /*
880  * rcar_dmac_chan_prep_sg - prepare transfer descriptors from an SG list
881  *
882  * Common routine for public (MEMCPY) and slave DMA. The MEMCPY case is also
883  * converted to scatter-gather to guarantee consistent locking and a correct
884  * list manipulation. For slave DMA direction carries the usual meaning, and,
885  * logically, the SG list is RAM and the addr variable contains slave address,
886  * e.g., the FIFO I/O register. For MEMCPY direction equals DMA_MEM_TO_MEM
887  * and the SG list contains only one element and points at the source buffer.
888  */
889 static struct dma_async_tx_descriptor *
890 rcar_dmac_chan_prep_sg(struct rcar_dmac_chan *chan, struct scatterlist *sgl,
891 		       unsigned int sg_len, dma_addr_t dev_addr,
892 		       enum dma_transfer_direction dir, unsigned long dma_flags,
893 		       bool cyclic)
894 {
895 	struct rcar_dmac_xfer_chunk *chunk;
896 	struct rcar_dmac_desc *desc;
897 	struct scatterlist *sg;
898 	unsigned int nchunks = 0;
899 	unsigned int max_chunk_size;
900 	unsigned int full_size = 0;
901 	bool cross_boundary = false;
902 	unsigned int i;
903 #ifdef CONFIG_ARCH_DMA_ADDR_T_64BIT
904 	u32 high_dev_addr;
905 	u32 high_mem_addr;
906 #endif
907 
908 	desc = rcar_dmac_desc_get(chan);
909 	if (!desc)
910 		return NULL;
911 
912 	desc->async_tx.flags = dma_flags;
913 	desc->async_tx.cookie = -EBUSY;
914 
915 	desc->cyclic = cyclic;
916 	desc->direction = dir;
917 
918 	rcar_dmac_chan_configure_desc(chan, desc);
919 
920 	max_chunk_size = (RCAR_DMATCR_MASK + 1) << desc->xfer_shift;
921 
922 	/*
923 	 * Allocate and fill the transfer chunk descriptors. We own the only
924 	 * reference to the DMA descriptor, there's no need for locking.
925 	 */
926 	for_each_sg(sgl, sg, sg_len, i) {
927 		dma_addr_t mem_addr = sg_dma_address(sg);
928 		unsigned int len = sg_dma_len(sg);
929 
930 		full_size += len;
931 
932 #ifdef CONFIG_ARCH_DMA_ADDR_T_64BIT
933 		if (i == 0) {
934 			high_dev_addr = dev_addr >> 32;
935 			high_mem_addr = mem_addr >> 32;
936 		}
937 
938 		if ((dev_addr >> 32 != high_dev_addr) ||
939 		    (mem_addr >> 32 != high_mem_addr))
940 			cross_boundary = true;
941 #endif
942 		while (len) {
943 			unsigned int size = min(len, max_chunk_size);
944 
945 #ifdef CONFIG_ARCH_DMA_ADDR_T_64BIT
946 			/*
947 			 * Prevent individual transfers from crossing 4GB
948 			 * boundaries.
949 			 */
950 			if (dev_addr >> 32 != (dev_addr + size - 1) >> 32) {
951 				size = ALIGN(dev_addr, 1ULL << 32) - dev_addr;
952 				cross_boundary = true;
953 			}
954 			if (mem_addr >> 32 != (mem_addr + size - 1) >> 32) {
955 				size = ALIGN(mem_addr, 1ULL << 32) - mem_addr;
956 				cross_boundary = true;
957 			}
958 #endif
959 
960 			chunk = rcar_dmac_xfer_chunk_get(chan);
961 			if (!chunk) {
962 				rcar_dmac_desc_put(chan, desc);
963 				return NULL;
964 			}
965 
966 			if (dir == DMA_DEV_TO_MEM) {
967 				chunk->src_addr = dev_addr;
968 				chunk->dst_addr = mem_addr;
969 			} else {
970 				chunk->src_addr = mem_addr;
971 				chunk->dst_addr = dev_addr;
972 			}
973 
974 			chunk->size = size;
975 
976 			dev_dbg(chan->chan.device->dev,
977 				"chan%u: chunk %p/%p sgl %u@%p, %u/%u %pad -> %pad\n",
978 				chan->index, chunk, desc, i, sg, size, len,
979 				&chunk->src_addr, &chunk->dst_addr);
980 
981 			mem_addr += size;
982 			if (dir == DMA_MEM_TO_MEM)
983 				dev_addr += size;
984 
985 			len -= size;
986 
987 			list_add_tail(&chunk->node, &desc->chunks);
988 			nchunks++;
989 		}
990 	}
991 
992 	desc->nchunks = nchunks;
993 	desc->size = full_size;
994 
995 	/*
996 	 * Use hardware descriptor lists if possible when more than one chunk
997 	 * needs to be transferred (otherwise they don't make much sense).
998 	 *
999 	 * Source/Destination address should be located in same 4GiB region
1000 	 * in the 40bit address space when it uses Hardware descriptor,
1001 	 * and cross_boundary is checking it.
1002 	 */
1003 	desc->hwdescs.use = !cross_boundary && nchunks > 1;
1004 	if (desc->hwdescs.use) {
1005 		if (rcar_dmac_fill_hwdesc(chan, desc) < 0)
1006 			desc->hwdescs.use = false;
1007 	}
1008 
1009 	return &desc->async_tx;
1010 }
1011 
1012 /* -----------------------------------------------------------------------------
1013  * DMA engine operations
1014  */
1015 
1016 static int rcar_dmac_alloc_chan_resources(struct dma_chan *chan)
1017 {
1018 	struct rcar_dmac_chan *rchan = to_rcar_dmac_chan(chan);
1019 	int ret;
1020 
1021 	INIT_LIST_HEAD(&rchan->desc.chunks_free);
1022 	INIT_LIST_HEAD(&rchan->desc.pages);
1023 
1024 	/* Preallocate descriptors. */
1025 	ret = rcar_dmac_xfer_chunk_alloc(rchan, GFP_KERNEL);
1026 	if (ret < 0)
1027 		return -ENOMEM;
1028 
1029 	ret = rcar_dmac_desc_alloc(rchan, GFP_KERNEL);
1030 	if (ret < 0)
1031 		return -ENOMEM;
1032 
1033 	return pm_runtime_get_sync(chan->device->dev);
1034 }
1035 
1036 static void rcar_dmac_free_chan_resources(struct dma_chan *chan)
1037 {
1038 	struct rcar_dmac_chan *rchan = to_rcar_dmac_chan(chan);
1039 	struct rcar_dmac *dmac = to_rcar_dmac(chan->device);
1040 	struct rcar_dmac_chan_map *map = &rchan->map;
1041 	struct rcar_dmac_desc_page *page, *_page;
1042 	struct rcar_dmac_desc *desc;
1043 	LIST_HEAD(list);
1044 
1045 	/* Protect against ISR */
1046 	spin_lock_irq(&rchan->lock);
1047 	rcar_dmac_chan_halt(rchan);
1048 	spin_unlock_irq(&rchan->lock);
1049 
1050 	/*
1051 	 * Now no new interrupts will occur, but one might already be
1052 	 * running. Wait for it to finish before freeing resources.
1053 	 */
1054 	synchronize_irq(rchan->irq);
1055 
1056 	if (rchan->mid_rid >= 0) {
1057 		/* The caller is holding dma_list_mutex */
1058 		clear_bit(rchan->mid_rid, dmac->modules);
1059 		rchan->mid_rid = -EINVAL;
1060 	}
1061 
1062 	list_splice_init(&rchan->desc.free, &list);
1063 	list_splice_init(&rchan->desc.pending, &list);
1064 	list_splice_init(&rchan->desc.active, &list);
1065 	list_splice_init(&rchan->desc.done, &list);
1066 	list_splice_init(&rchan->desc.wait, &list);
1067 
1068 	rchan->desc.running = NULL;
1069 
1070 	list_for_each_entry(desc, &list, node)
1071 		rcar_dmac_realloc_hwdesc(rchan, desc, 0);
1072 
1073 	list_for_each_entry_safe(page, _page, &rchan->desc.pages, node) {
1074 		list_del(&page->node);
1075 		free_page((unsigned long)page);
1076 	}
1077 
1078 	/* Remove slave mapping if present. */
1079 	if (map->slave.xfer_size) {
1080 		dma_unmap_resource(chan->device->dev, map->addr,
1081 				   map->slave.xfer_size, map->dir, 0);
1082 		map->slave.xfer_size = 0;
1083 	}
1084 
1085 	pm_runtime_put(chan->device->dev);
1086 }
1087 
1088 static struct dma_async_tx_descriptor *
1089 rcar_dmac_prep_dma_memcpy(struct dma_chan *chan, dma_addr_t dma_dest,
1090 			  dma_addr_t dma_src, size_t len, unsigned long flags)
1091 {
1092 	struct rcar_dmac_chan *rchan = to_rcar_dmac_chan(chan);
1093 	struct scatterlist sgl;
1094 
1095 	if (!len)
1096 		return NULL;
1097 
1098 	sg_init_table(&sgl, 1);
1099 	sg_set_page(&sgl, pfn_to_page(PFN_DOWN(dma_src)), len,
1100 		    offset_in_page(dma_src));
1101 	sg_dma_address(&sgl) = dma_src;
1102 	sg_dma_len(&sgl) = len;
1103 
1104 	return rcar_dmac_chan_prep_sg(rchan, &sgl, 1, dma_dest,
1105 				      DMA_MEM_TO_MEM, flags, false);
1106 }
1107 
1108 static int rcar_dmac_map_slave_addr(struct dma_chan *chan,
1109 				    enum dma_transfer_direction dir)
1110 {
1111 	struct rcar_dmac_chan *rchan = to_rcar_dmac_chan(chan);
1112 	struct rcar_dmac_chan_map *map = &rchan->map;
1113 	phys_addr_t dev_addr;
1114 	size_t dev_size;
1115 	enum dma_data_direction dev_dir;
1116 
1117 	if (dir == DMA_DEV_TO_MEM) {
1118 		dev_addr = rchan->src.slave_addr;
1119 		dev_size = rchan->src.xfer_size;
1120 		dev_dir = DMA_TO_DEVICE;
1121 	} else {
1122 		dev_addr = rchan->dst.slave_addr;
1123 		dev_size = rchan->dst.xfer_size;
1124 		dev_dir = DMA_FROM_DEVICE;
1125 	}
1126 
1127 	/* Reuse current map if possible. */
1128 	if (dev_addr == map->slave.slave_addr &&
1129 	    dev_size == map->slave.xfer_size &&
1130 	    dev_dir == map->dir)
1131 		return 0;
1132 
1133 	/* Remove old mapping if present. */
1134 	if (map->slave.xfer_size)
1135 		dma_unmap_resource(chan->device->dev, map->addr,
1136 				   map->slave.xfer_size, map->dir, 0);
1137 	map->slave.xfer_size = 0;
1138 
1139 	/* Create new slave address map. */
1140 	map->addr = dma_map_resource(chan->device->dev, dev_addr, dev_size,
1141 				     dev_dir, 0);
1142 
1143 	if (dma_mapping_error(chan->device->dev, map->addr)) {
1144 		dev_err(chan->device->dev,
1145 			"chan%u: failed to map %zx@%pap", rchan->index,
1146 			dev_size, &dev_addr);
1147 		return -EIO;
1148 	}
1149 
1150 	dev_dbg(chan->device->dev, "chan%u: map %zx@%pap to %pad dir: %s\n",
1151 		rchan->index, dev_size, &dev_addr, &map->addr,
1152 		dev_dir == DMA_TO_DEVICE ? "DMA_TO_DEVICE" : "DMA_FROM_DEVICE");
1153 
1154 	map->slave.slave_addr = dev_addr;
1155 	map->slave.xfer_size = dev_size;
1156 	map->dir = dev_dir;
1157 
1158 	return 0;
1159 }
1160 
1161 static struct dma_async_tx_descriptor *
1162 rcar_dmac_prep_slave_sg(struct dma_chan *chan, struct scatterlist *sgl,
1163 			unsigned int sg_len, enum dma_transfer_direction dir,
1164 			unsigned long flags, void *context)
1165 {
1166 	struct rcar_dmac_chan *rchan = to_rcar_dmac_chan(chan);
1167 
1168 	/* Someone calling slave DMA on a generic channel? */
1169 	if (rchan->mid_rid < 0 || !sg_len) {
1170 		dev_warn(chan->device->dev,
1171 			 "%s: bad parameter: len=%d, id=%d\n",
1172 			 __func__, sg_len, rchan->mid_rid);
1173 		return NULL;
1174 	}
1175 
1176 	if (rcar_dmac_map_slave_addr(chan, dir))
1177 		return NULL;
1178 
1179 	return rcar_dmac_chan_prep_sg(rchan, sgl, sg_len, rchan->map.addr,
1180 				      dir, flags, false);
1181 }
1182 
1183 #define RCAR_DMAC_MAX_SG_LEN	32
1184 
1185 static struct dma_async_tx_descriptor *
1186 rcar_dmac_prep_dma_cyclic(struct dma_chan *chan, dma_addr_t buf_addr,
1187 			  size_t buf_len, size_t period_len,
1188 			  enum dma_transfer_direction dir, unsigned long flags)
1189 {
1190 	struct rcar_dmac_chan *rchan = to_rcar_dmac_chan(chan);
1191 	struct dma_async_tx_descriptor *desc;
1192 	struct scatterlist *sgl;
1193 	unsigned int sg_len;
1194 	unsigned int i;
1195 
1196 	/* Someone calling slave DMA on a generic channel? */
1197 	if (rchan->mid_rid < 0 || buf_len < period_len) {
1198 		dev_warn(chan->device->dev,
1199 			"%s: bad parameter: buf_len=%zu, period_len=%zu, id=%d\n",
1200 			__func__, buf_len, period_len, rchan->mid_rid);
1201 		return NULL;
1202 	}
1203 
1204 	if (rcar_dmac_map_slave_addr(chan, dir))
1205 		return NULL;
1206 
1207 	sg_len = buf_len / period_len;
1208 	if (sg_len > RCAR_DMAC_MAX_SG_LEN) {
1209 		dev_err(chan->device->dev,
1210 			"chan%u: sg length %d exceds limit %d",
1211 			rchan->index, sg_len, RCAR_DMAC_MAX_SG_LEN);
1212 		return NULL;
1213 	}
1214 
1215 	/*
1216 	 * Allocate the sg list dynamically as it would consume too much stack
1217 	 * space.
1218 	 */
1219 	sgl = kcalloc(sg_len, sizeof(*sgl), GFP_NOWAIT);
1220 	if (!sgl)
1221 		return NULL;
1222 
1223 	sg_init_table(sgl, sg_len);
1224 
1225 	for (i = 0; i < sg_len; ++i) {
1226 		dma_addr_t src = buf_addr + (period_len * i);
1227 
1228 		sg_set_page(&sgl[i], pfn_to_page(PFN_DOWN(src)), period_len,
1229 			    offset_in_page(src));
1230 		sg_dma_address(&sgl[i]) = src;
1231 		sg_dma_len(&sgl[i]) = period_len;
1232 	}
1233 
1234 	desc = rcar_dmac_chan_prep_sg(rchan, sgl, sg_len, rchan->map.addr,
1235 				      dir, flags, true);
1236 
1237 	kfree(sgl);
1238 	return desc;
1239 }
1240 
1241 static int rcar_dmac_device_config(struct dma_chan *chan,
1242 				   struct dma_slave_config *cfg)
1243 {
1244 	struct rcar_dmac_chan *rchan = to_rcar_dmac_chan(chan);
1245 
1246 	/*
1247 	 * We could lock this, but you shouldn't be configuring the
1248 	 * channel, while using it...
1249 	 */
1250 	rchan->src.slave_addr = cfg->src_addr;
1251 	rchan->dst.slave_addr = cfg->dst_addr;
1252 	rchan->src.xfer_size = cfg->src_addr_width;
1253 	rchan->dst.xfer_size = cfg->dst_addr_width;
1254 
1255 	return 0;
1256 }
1257 
1258 static int rcar_dmac_chan_terminate_all(struct dma_chan *chan)
1259 {
1260 	struct rcar_dmac_chan *rchan = to_rcar_dmac_chan(chan);
1261 	unsigned long flags;
1262 
1263 	spin_lock_irqsave(&rchan->lock, flags);
1264 	rcar_dmac_chan_halt(rchan);
1265 	spin_unlock_irqrestore(&rchan->lock, flags);
1266 
1267 	/*
1268 	 * FIXME: No new interrupt can occur now, but the IRQ thread might still
1269 	 * be running.
1270 	 */
1271 
1272 	rcar_dmac_chan_reinit(rchan);
1273 
1274 	return 0;
1275 }
1276 
1277 static unsigned int rcar_dmac_chan_get_residue(struct rcar_dmac_chan *chan,
1278 					       dma_cookie_t cookie)
1279 {
1280 	struct rcar_dmac_desc *desc = chan->desc.running;
1281 	struct rcar_dmac_xfer_chunk *running = NULL;
1282 	struct rcar_dmac_xfer_chunk *chunk;
1283 	enum dma_status status;
1284 	unsigned int residue = 0;
1285 	unsigned int dptr = 0;
1286 
1287 	if (!desc)
1288 		return 0;
1289 
1290 	/*
1291 	 * If the cookie corresponds to a descriptor that has been completed
1292 	 * there is no residue. The same check has already been performed by the
1293 	 * caller but without holding the channel lock, so the descriptor could
1294 	 * now be complete.
1295 	 */
1296 	status = dma_cookie_status(&chan->chan, cookie, NULL);
1297 	if (status == DMA_COMPLETE)
1298 		return 0;
1299 
1300 	/*
1301 	 * If the cookie doesn't correspond to the currently running transfer
1302 	 * then the descriptor hasn't been processed yet, and the residue is
1303 	 * equal to the full descriptor size.
1304 	 */
1305 	if (cookie != desc->async_tx.cookie) {
1306 		list_for_each_entry(desc, &chan->desc.pending, node) {
1307 			if (cookie == desc->async_tx.cookie)
1308 				return desc->size;
1309 		}
1310 		list_for_each_entry(desc, &chan->desc.active, node) {
1311 			if (cookie == desc->async_tx.cookie)
1312 				return desc->size;
1313 		}
1314 
1315 		/*
1316 		 * No descriptor found for the cookie, there's thus no residue.
1317 		 * This shouldn't happen if the calling driver passes a correct
1318 		 * cookie value.
1319 		 */
1320 		WARN(1, "No descriptor for cookie!");
1321 		return 0;
1322 	}
1323 
1324 	/*
1325 	 * In descriptor mode the descriptor running pointer is not maintained
1326 	 * by the interrupt handler, find the running descriptor from the
1327 	 * descriptor pointer field in the CHCRB register. In non-descriptor
1328 	 * mode just use the running descriptor pointer.
1329 	 */
1330 	if (desc->hwdescs.use) {
1331 		dptr = (rcar_dmac_chan_read(chan, RCAR_DMACHCRB) &
1332 			RCAR_DMACHCRB_DPTR_MASK) >> RCAR_DMACHCRB_DPTR_SHIFT;
1333 		if (dptr == 0)
1334 			dptr = desc->nchunks;
1335 		dptr--;
1336 		WARN_ON(dptr >= desc->nchunks);
1337 	} else {
1338 		running = desc->running;
1339 	}
1340 
1341 	/* Compute the size of all chunks still to be transferred. */
1342 	list_for_each_entry_reverse(chunk, &desc->chunks, node) {
1343 		if (chunk == running || ++dptr == desc->nchunks)
1344 			break;
1345 
1346 		residue += chunk->size;
1347 	}
1348 
1349 	if (desc->direction == DMA_DEV_TO_MEM)
1350 		rcar_dmac_sync_tcr(chan);
1351 
1352 	/* Add the residue for the current chunk. */
1353 	residue += rcar_dmac_chan_read(chan, RCAR_DMATCRB) << desc->xfer_shift;
1354 
1355 	return residue;
1356 }
1357 
1358 static enum dma_status rcar_dmac_tx_status(struct dma_chan *chan,
1359 					   dma_cookie_t cookie,
1360 					   struct dma_tx_state *txstate)
1361 {
1362 	struct rcar_dmac_chan *rchan = to_rcar_dmac_chan(chan);
1363 	enum dma_status status;
1364 	unsigned long flags;
1365 	unsigned int residue;
1366 
1367 	status = dma_cookie_status(chan, cookie, txstate);
1368 	if (status == DMA_COMPLETE || !txstate)
1369 		return status;
1370 
1371 	spin_lock_irqsave(&rchan->lock, flags);
1372 	residue = rcar_dmac_chan_get_residue(rchan, cookie);
1373 	spin_unlock_irqrestore(&rchan->lock, flags);
1374 
1375 	/* if there's no residue, the cookie is complete */
1376 	if (!residue)
1377 		return DMA_COMPLETE;
1378 
1379 	dma_set_residue(txstate, residue);
1380 
1381 	return status;
1382 }
1383 
1384 static void rcar_dmac_issue_pending(struct dma_chan *chan)
1385 {
1386 	struct rcar_dmac_chan *rchan = to_rcar_dmac_chan(chan);
1387 	unsigned long flags;
1388 
1389 	spin_lock_irqsave(&rchan->lock, flags);
1390 
1391 	if (list_empty(&rchan->desc.pending))
1392 		goto done;
1393 
1394 	/* Append the pending list to the active list. */
1395 	list_splice_tail_init(&rchan->desc.pending, &rchan->desc.active);
1396 
1397 	/*
1398 	 * If no transfer is running pick the first descriptor from the active
1399 	 * list and start the transfer.
1400 	 */
1401 	if (!rchan->desc.running) {
1402 		struct rcar_dmac_desc *desc;
1403 
1404 		desc = list_first_entry(&rchan->desc.active,
1405 					struct rcar_dmac_desc, node);
1406 		rchan->desc.running = desc;
1407 
1408 		rcar_dmac_chan_start_xfer(rchan);
1409 	}
1410 
1411 done:
1412 	spin_unlock_irqrestore(&rchan->lock, flags);
1413 }
1414 
1415 static void rcar_dmac_device_synchronize(struct dma_chan *chan)
1416 {
1417 	struct rcar_dmac_chan *rchan = to_rcar_dmac_chan(chan);
1418 
1419 	synchronize_irq(rchan->irq);
1420 }
1421 
1422 /* -----------------------------------------------------------------------------
1423  * IRQ handling
1424  */
1425 
1426 static irqreturn_t rcar_dmac_isr_desc_stage_end(struct rcar_dmac_chan *chan)
1427 {
1428 	struct rcar_dmac_desc *desc = chan->desc.running;
1429 	unsigned int stage;
1430 
1431 	if (WARN_ON(!desc || !desc->cyclic)) {
1432 		/*
1433 		 * This should never happen, there should always be a running
1434 		 * cyclic descriptor when a descriptor stage end interrupt is
1435 		 * triggered. Warn and return.
1436 		 */
1437 		return IRQ_NONE;
1438 	}
1439 
1440 	/* Program the interrupt pointer to the next stage. */
1441 	stage = (rcar_dmac_chan_read(chan, RCAR_DMACHCRB) &
1442 		 RCAR_DMACHCRB_DPTR_MASK) >> RCAR_DMACHCRB_DPTR_SHIFT;
1443 	rcar_dmac_chan_write(chan, RCAR_DMADPCR, RCAR_DMADPCR_DIPT(stage));
1444 
1445 	return IRQ_WAKE_THREAD;
1446 }
1447 
1448 static irqreturn_t rcar_dmac_isr_transfer_end(struct rcar_dmac_chan *chan)
1449 {
1450 	struct rcar_dmac_desc *desc = chan->desc.running;
1451 	irqreturn_t ret = IRQ_WAKE_THREAD;
1452 
1453 	if (WARN_ON_ONCE(!desc)) {
1454 		/*
1455 		 * This should never happen, there should always be a running
1456 		 * descriptor when a transfer end interrupt is triggered. Warn
1457 		 * and return.
1458 		 */
1459 		return IRQ_NONE;
1460 	}
1461 
1462 	/*
1463 	 * The transfer end interrupt isn't generated for each chunk when using
1464 	 * descriptor mode. Only update the running chunk pointer in
1465 	 * non-descriptor mode.
1466 	 */
1467 	if (!desc->hwdescs.use) {
1468 		/*
1469 		 * If we haven't completed the last transfer chunk simply move
1470 		 * to the next one. Only wake the IRQ thread if the transfer is
1471 		 * cyclic.
1472 		 */
1473 		if (!list_is_last(&desc->running->node, &desc->chunks)) {
1474 			desc->running = list_next_entry(desc->running, node);
1475 			if (!desc->cyclic)
1476 				ret = IRQ_HANDLED;
1477 			goto done;
1478 		}
1479 
1480 		/*
1481 		 * We've completed the last transfer chunk. If the transfer is
1482 		 * cyclic, move back to the first one.
1483 		 */
1484 		if (desc->cyclic) {
1485 			desc->running =
1486 				list_first_entry(&desc->chunks,
1487 						 struct rcar_dmac_xfer_chunk,
1488 						 node);
1489 			goto done;
1490 		}
1491 	}
1492 
1493 	/* The descriptor is complete, move it to the done list. */
1494 	list_move_tail(&desc->node, &chan->desc.done);
1495 
1496 	/* Queue the next descriptor, if any. */
1497 	if (!list_empty(&chan->desc.active))
1498 		chan->desc.running = list_first_entry(&chan->desc.active,
1499 						      struct rcar_dmac_desc,
1500 						      node);
1501 	else
1502 		chan->desc.running = NULL;
1503 
1504 done:
1505 	if (chan->desc.running)
1506 		rcar_dmac_chan_start_xfer(chan);
1507 
1508 	return ret;
1509 }
1510 
1511 static irqreturn_t rcar_dmac_isr_channel(int irq, void *dev)
1512 {
1513 	u32 mask = RCAR_DMACHCR_DSE | RCAR_DMACHCR_TE;
1514 	struct rcar_dmac_chan *chan = dev;
1515 	irqreturn_t ret = IRQ_NONE;
1516 	u32 chcr;
1517 
1518 	spin_lock(&chan->lock);
1519 
1520 	chcr = rcar_dmac_chan_read(chan, RCAR_DMACHCR);
1521 	if (chcr & RCAR_DMACHCR_TE)
1522 		mask |= RCAR_DMACHCR_DE;
1523 	rcar_dmac_chan_write(chan, RCAR_DMACHCR, chcr & ~mask);
1524 	if (mask & RCAR_DMACHCR_DE)
1525 		rcar_dmac_chcr_de_barrier(chan);
1526 
1527 	if (chcr & RCAR_DMACHCR_DSE)
1528 		ret |= rcar_dmac_isr_desc_stage_end(chan);
1529 
1530 	if (chcr & RCAR_DMACHCR_TE)
1531 		ret |= rcar_dmac_isr_transfer_end(chan);
1532 
1533 	spin_unlock(&chan->lock);
1534 
1535 	return ret;
1536 }
1537 
1538 static irqreturn_t rcar_dmac_isr_channel_thread(int irq, void *dev)
1539 {
1540 	struct rcar_dmac_chan *chan = dev;
1541 	struct rcar_dmac_desc *desc;
1542 	struct dmaengine_desc_callback cb;
1543 
1544 	spin_lock_irq(&chan->lock);
1545 
1546 	/* For cyclic transfers notify the user after every chunk. */
1547 	if (chan->desc.running && chan->desc.running->cyclic) {
1548 		desc = chan->desc.running;
1549 		dmaengine_desc_get_callback(&desc->async_tx, &cb);
1550 
1551 		if (dmaengine_desc_callback_valid(&cb)) {
1552 			spin_unlock_irq(&chan->lock);
1553 			dmaengine_desc_callback_invoke(&cb, NULL);
1554 			spin_lock_irq(&chan->lock);
1555 		}
1556 	}
1557 
1558 	/*
1559 	 * Call the callback function for all descriptors on the done list and
1560 	 * move them to the ack wait list.
1561 	 */
1562 	while (!list_empty(&chan->desc.done)) {
1563 		desc = list_first_entry(&chan->desc.done, struct rcar_dmac_desc,
1564 					node);
1565 		dma_cookie_complete(&desc->async_tx);
1566 		list_del(&desc->node);
1567 
1568 		dmaengine_desc_get_callback(&desc->async_tx, &cb);
1569 		if (dmaengine_desc_callback_valid(&cb)) {
1570 			spin_unlock_irq(&chan->lock);
1571 			/*
1572 			 * We own the only reference to this descriptor, we can
1573 			 * safely dereference it without holding the channel
1574 			 * lock.
1575 			 */
1576 			dmaengine_desc_callback_invoke(&cb, NULL);
1577 			spin_lock_irq(&chan->lock);
1578 		}
1579 
1580 		list_add_tail(&desc->node, &chan->desc.wait);
1581 	}
1582 
1583 	spin_unlock_irq(&chan->lock);
1584 
1585 	/* Recycle all acked descriptors. */
1586 	rcar_dmac_desc_recycle_acked(chan);
1587 
1588 	return IRQ_HANDLED;
1589 }
1590 
1591 static irqreturn_t rcar_dmac_isr_error(int irq, void *data)
1592 {
1593 	struct rcar_dmac *dmac = data;
1594 
1595 	if (!(rcar_dmac_read(dmac, RCAR_DMAOR) & RCAR_DMAOR_AE))
1596 		return IRQ_NONE;
1597 
1598 	/*
1599 	 * An unrecoverable error occurred on an unknown channel. Halt the DMAC,
1600 	 * abort transfers on all channels, and reinitialize the DMAC.
1601 	 */
1602 	rcar_dmac_stop(dmac);
1603 	rcar_dmac_abort(dmac);
1604 	rcar_dmac_init(dmac);
1605 
1606 	return IRQ_HANDLED;
1607 }
1608 
1609 /* -----------------------------------------------------------------------------
1610  * OF xlate and channel filter
1611  */
1612 
1613 static bool rcar_dmac_chan_filter(struct dma_chan *chan, void *arg)
1614 {
1615 	struct rcar_dmac *dmac = to_rcar_dmac(chan->device);
1616 	struct of_phandle_args *dma_spec = arg;
1617 
1618 	/*
1619 	 * FIXME: Using a filter on OF platforms is a nonsense. The OF xlate
1620 	 * function knows from which device it wants to allocate a channel from,
1621 	 * and would be perfectly capable of selecting the channel it wants.
1622 	 * Forcing it to call dma_request_channel() and iterate through all
1623 	 * channels from all controllers is just pointless.
1624 	 */
1625 	if (chan->device->device_config != rcar_dmac_device_config ||
1626 	    dma_spec->np != chan->device->dev->of_node)
1627 		return false;
1628 
1629 	return !test_and_set_bit(dma_spec->args[0], dmac->modules);
1630 }
1631 
1632 static struct dma_chan *rcar_dmac_of_xlate(struct of_phandle_args *dma_spec,
1633 					   struct of_dma *ofdma)
1634 {
1635 	struct rcar_dmac_chan *rchan;
1636 	struct dma_chan *chan;
1637 	dma_cap_mask_t mask;
1638 
1639 	if (dma_spec->args_count != 1)
1640 		return NULL;
1641 
1642 	/* Only slave DMA channels can be allocated via DT */
1643 	dma_cap_zero(mask);
1644 	dma_cap_set(DMA_SLAVE, mask);
1645 
1646 	chan = dma_request_channel(mask, rcar_dmac_chan_filter, dma_spec);
1647 	if (!chan)
1648 		return NULL;
1649 
1650 	rchan = to_rcar_dmac_chan(chan);
1651 	rchan->mid_rid = dma_spec->args[0];
1652 
1653 	return chan;
1654 }
1655 
1656 /* -----------------------------------------------------------------------------
1657  * Power management
1658  */
1659 
1660 #ifdef CONFIG_PM
1661 static int rcar_dmac_runtime_suspend(struct device *dev)
1662 {
1663 	return 0;
1664 }
1665 
1666 static int rcar_dmac_runtime_resume(struct device *dev)
1667 {
1668 	struct rcar_dmac *dmac = dev_get_drvdata(dev);
1669 
1670 	return rcar_dmac_init(dmac);
1671 }
1672 #endif
1673 
1674 static const struct dev_pm_ops rcar_dmac_pm = {
1675 	/*
1676 	 * TODO for system sleep/resume:
1677 	 *   - Wait for the current transfer to complete and stop the device,
1678 	 *   - Resume transfers, if any.
1679 	 */
1680 	SET_LATE_SYSTEM_SLEEP_PM_OPS(pm_runtime_force_suspend,
1681 				     pm_runtime_force_resume)
1682 	SET_RUNTIME_PM_OPS(rcar_dmac_runtime_suspend, rcar_dmac_runtime_resume,
1683 			   NULL)
1684 };
1685 
1686 /* -----------------------------------------------------------------------------
1687  * Probe and remove
1688  */
1689 
1690 static int rcar_dmac_chan_probe(struct rcar_dmac *dmac,
1691 				struct rcar_dmac_chan *rchan,
1692 				unsigned int index)
1693 {
1694 	struct platform_device *pdev = to_platform_device(dmac->dev);
1695 	struct dma_chan *chan = &rchan->chan;
1696 	char pdev_irqname[5];
1697 	char *irqname;
1698 	int ret;
1699 
1700 	rchan->index = index;
1701 	rchan->iomem = dmac->iomem + RCAR_DMAC_CHAN_OFFSET(index);
1702 	rchan->mid_rid = -EINVAL;
1703 
1704 	spin_lock_init(&rchan->lock);
1705 
1706 	INIT_LIST_HEAD(&rchan->desc.free);
1707 	INIT_LIST_HEAD(&rchan->desc.pending);
1708 	INIT_LIST_HEAD(&rchan->desc.active);
1709 	INIT_LIST_HEAD(&rchan->desc.done);
1710 	INIT_LIST_HEAD(&rchan->desc.wait);
1711 
1712 	/* Request the channel interrupt. */
1713 	sprintf(pdev_irqname, "ch%u", index);
1714 	rchan->irq = platform_get_irq_byname(pdev, pdev_irqname);
1715 	if (rchan->irq < 0) {
1716 		dev_err(dmac->dev, "no IRQ specified for channel %u\n", index);
1717 		return -ENODEV;
1718 	}
1719 
1720 	irqname = devm_kasprintf(dmac->dev, GFP_KERNEL, "%s:%u",
1721 				 dev_name(dmac->dev), index);
1722 	if (!irqname)
1723 		return -ENOMEM;
1724 
1725 	/*
1726 	 * Initialize the DMA engine channel and add it to the DMA engine
1727 	 * channels list.
1728 	 */
1729 	chan->device = &dmac->engine;
1730 	dma_cookie_init(chan);
1731 
1732 	list_add_tail(&chan->device_node, &dmac->engine.channels);
1733 
1734 	ret = devm_request_threaded_irq(dmac->dev, rchan->irq,
1735 					rcar_dmac_isr_channel,
1736 					rcar_dmac_isr_channel_thread, 0,
1737 					irqname, rchan);
1738 	if (ret) {
1739 		dev_err(dmac->dev, "failed to request IRQ %u (%d)\n",
1740 			rchan->irq, ret);
1741 		return ret;
1742 	}
1743 
1744 	return 0;
1745 }
1746 
1747 static int rcar_dmac_parse_of(struct device *dev, struct rcar_dmac *dmac)
1748 {
1749 	struct device_node *np = dev->of_node;
1750 	int ret;
1751 
1752 	ret = of_property_read_u32(np, "dma-channels", &dmac->n_channels);
1753 	if (ret < 0) {
1754 		dev_err(dev, "unable to read dma-channels property\n");
1755 		return ret;
1756 	}
1757 
1758 	if (dmac->n_channels <= 0 || dmac->n_channels >= 100) {
1759 		dev_err(dev, "invalid number of channels %u\n",
1760 			dmac->n_channels);
1761 		return -EINVAL;
1762 	}
1763 
1764 	return 0;
1765 }
1766 
1767 static int rcar_dmac_probe(struct platform_device *pdev)
1768 {
1769 	const enum dma_slave_buswidth widths = DMA_SLAVE_BUSWIDTH_1_BYTE |
1770 		DMA_SLAVE_BUSWIDTH_2_BYTES | DMA_SLAVE_BUSWIDTH_4_BYTES |
1771 		DMA_SLAVE_BUSWIDTH_8_BYTES | DMA_SLAVE_BUSWIDTH_16_BYTES |
1772 		DMA_SLAVE_BUSWIDTH_32_BYTES | DMA_SLAVE_BUSWIDTH_64_BYTES;
1773 	unsigned int channels_offset = 0;
1774 	struct dma_device *engine;
1775 	struct rcar_dmac *dmac;
1776 	struct resource *mem;
1777 	unsigned int i;
1778 	char *irqname;
1779 	int irq;
1780 	int ret;
1781 
1782 	dmac = devm_kzalloc(&pdev->dev, sizeof(*dmac), GFP_KERNEL);
1783 	if (!dmac)
1784 		return -ENOMEM;
1785 
1786 	dmac->dev = &pdev->dev;
1787 	platform_set_drvdata(pdev, dmac);
1788 	dma_set_mask_and_coherent(dmac->dev, DMA_BIT_MASK(40));
1789 
1790 	ret = rcar_dmac_parse_of(&pdev->dev, dmac);
1791 	if (ret < 0)
1792 		return ret;
1793 
1794 	/*
1795 	 * A still unconfirmed hardware bug prevents the IPMMU microTLB 0 to be
1796 	 * flushed correctly, resulting in memory corruption. DMAC 0 channel 0
1797 	 * is connected to microTLB 0 on currently supported platforms, so we
1798 	 * can't use it with the IPMMU. As the IOMMU API operates at the device
1799 	 * level we can't disable it selectively, so ignore channel 0 for now if
1800 	 * the device is part of an IOMMU group.
1801 	 */
1802 	if (pdev->dev.iommu_group) {
1803 		dmac->n_channels--;
1804 		channels_offset = 1;
1805 	}
1806 
1807 	dmac->channels = devm_kcalloc(&pdev->dev, dmac->n_channels,
1808 				      sizeof(*dmac->channels), GFP_KERNEL);
1809 	if (!dmac->channels)
1810 		return -ENOMEM;
1811 
1812 	/* Request resources. */
1813 	mem = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1814 	dmac->iomem = devm_ioremap_resource(&pdev->dev, mem);
1815 	if (IS_ERR(dmac->iomem))
1816 		return PTR_ERR(dmac->iomem);
1817 
1818 	irq = platform_get_irq_byname(pdev, "error");
1819 	if (irq < 0) {
1820 		dev_err(&pdev->dev, "no error IRQ specified\n");
1821 		return -ENODEV;
1822 	}
1823 
1824 	irqname = devm_kasprintf(dmac->dev, GFP_KERNEL, "%s:error",
1825 				 dev_name(dmac->dev));
1826 	if (!irqname)
1827 		return -ENOMEM;
1828 
1829 	/* Enable runtime PM and initialize the device. */
1830 	pm_runtime_enable(&pdev->dev);
1831 	ret = pm_runtime_get_sync(&pdev->dev);
1832 	if (ret < 0) {
1833 		dev_err(&pdev->dev, "runtime PM get sync failed (%d)\n", ret);
1834 		return ret;
1835 	}
1836 
1837 	ret = rcar_dmac_init(dmac);
1838 	pm_runtime_put(&pdev->dev);
1839 
1840 	if (ret) {
1841 		dev_err(&pdev->dev, "failed to reset device\n");
1842 		goto error;
1843 	}
1844 
1845 	/* Initialize engine */
1846 	engine = &dmac->engine;
1847 
1848 	dma_cap_set(DMA_MEMCPY, engine->cap_mask);
1849 	dma_cap_set(DMA_SLAVE, engine->cap_mask);
1850 
1851 	engine->dev		= &pdev->dev;
1852 	engine->copy_align	= ilog2(RCAR_DMAC_MEMCPY_XFER_SIZE);
1853 
1854 	engine->src_addr_widths	= widths;
1855 	engine->dst_addr_widths	= widths;
1856 	engine->directions	= BIT(DMA_MEM_TO_DEV) | BIT(DMA_DEV_TO_MEM);
1857 	engine->residue_granularity = DMA_RESIDUE_GRANULARITY_BURST;
1858 
1859 	engine->device_alloc_chan_resources	= rcar_dmac_alloc_chan_resources;
1860 	engine->device_free_chan_resources	= rcar_dmac_free_chan_resources;
1861 	engine->device_prep_dma_memcpy		= rcar_dmac_prep_dma_memcpy;
1862 	engine->device_prep_slave_sg		= rcar_dmac_prep_slave_sg;
1863 	engine->device_prep_dma_cyclic		= rcar_dmac_prep_dma_cyclic;
1864 	engine->device_config			= rcar_dmac_device_config;
1865 	engine->device_terminate_all		= rcar_dmac_chan_terminate_all;
1866 	engine->device_tx_status		= rcar_dmac_tx_status;
1867 	engine->device_issue_pending		= rcar_dmac_issue_pending;
1868 	engine->device_synchronize		= rcar_dmac_device_synchronize;
1869 
1870 	INIT_LIST_HEAD(&engine->channels);
1871 
1872 	for (i = 0; i < dmac->n_channels; ++i) {
1873 		ret = rcar_dmac_chan_probe(dmac, &dmac->channels[i],
1874 					   i + channels_offset);
1875 		if (ret < 0)
1876 			goto error;
1877 	}
1878 
1879 	ret = devm_request_irq(&pdev->dev, irq, rcar_dmac_isr_error, 0,
1880 			       irqname, dmac);
1881 	if (ret) {
1882 		dev_err(&pdev->dev, "failed to request IRQ %u (%d)\n",
1883 			irq, ret);
1884 		return ret;
1885 	}
1886 
1887 	/* Register the DMAC as a DMA provider for DT. */
1888 	ret = of_dma_controller_register(pdev->dev.of_node, rcar_dmac_of_xlate,
1889 					 NULL);
1890 	if (ret < 0)
1891 		goto error;
1892 
1893 	/*
1894 	 * Register the DMA engine device.
1895 	 *
1896 	 * Default transfer size of 32 bytes requires 32-byte alignment.
1897 	 */
1898 	ret = dma_async_device_register(engine);
1899 	if (ret < 0)
1900 		goto error;
1901 
1902 	return 0;
1903 
1904 error:
1905 	of_dma_controller_free(pdev->dev.of_node);
1906 	pm_runtime_disable(&pdev->dev);
1907 	return ret;
1908 }
1909 
1910 static int rcar_dmac_remove(struct platform_device *pdev)
1911 {
1912 	struct rcar_dmac *dmac = platform_get_drvdata(pdev);
1913 
1914 	of_dma_controller_free(pdev->dev.of_node);
1915 	dma_async_device_unregister(&dmac->engine);
1916 
1917 	pm_runtime_disable(&pdev->dev);
1918 
1919 	return 0;
1920 }
1921 
1922 static void rcar_dmac_shutdown(struct platform_device *pdev)
1923 {
1924 	struct rcar_dmac *dmac = platform_get_drvdata(pdev);
1925 
1926 	rcar_dmac_stop(dmac);
1927 }
1928 
1929 static const struct of_device_id rcar_dmac_of_ids[] = {
1930 	{ .compatible = "renesas,rcar-dmac", },
1931 	{ /* Sentinel */ }
1932 };
1933 MODULE_DEVICE_TABLE(of, rcar_dmac_of_ids);
1934 
1935 static struct platform_driver rcar_dmac_driver = {
1936 	.driver		= {
1937 		.pm	= &rcar_dmac_pm,
1938 		.name	= "rcar-dmac",
1939 		.of_match_table = rcar_dmac_of_ids,
1940 	},
1941 	.probe		= rcar_dmac_probe,
1942 	.remove		= rcar_dmac_remove,
1943 	.shutdown	= rcar_dmac_shutdown,
1944 };
1945 
1946 module_platform_driver(rcar_dmac_driver);
1947 
1948 MODULE_DESCRIPTION("R-Car Gen2 DMA Controller Driver");
1949 MODULE_AUTHOR("Laurent Pinchart <laurent.pinchart@ideasonboard.com>");
1950 MODULE_LICENSE("GPL v2");
1951