xref: /openbmc/linux/drivers/dma/sh/rcar-dmac.c (revision 710b797c)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Renesas R-Car Gen2/Gen3 DMA Controller Driver
4  *
5  * Copyright (C) 2014-2019 Renesas Electronics Inc.
6  *
7  * Author: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
8  */
9 
10 #include <linux/delay.h>
11 #include <linux/dma-mapping.h>
12 #include <linux/dmaengine.h>
13 #include <linux/interrupt.h>
14 #include <linux/list.h>
15 #include <linux/module.h>
16 #include <linux/mutex.h>
17 #include <linux/of.h>
18 #include <linux/of_dma.h>
19 #include <linux/of_platform.h>
20 #include <linux/platform_device.h>
21 #include <linux/pm_runtime.h>
22 #include <linux/slab.h>
23 #include <linux/spinlock.h>
24 
25 #include "../dmaengine.h"
26 
27 /*
28  * struct rcar_dmac_xfer_chunk - Descriptor for a hardware transfer
29  * @node: entry in the parent's chunks list
30  * @src_addr: device source address
31  * @dst_addr: device destination address
32  * @size: transfer size in bytes
33  */
34 struct rcar_dmac_xfer_chunk {
35 	struct list_head node;
36 
37 	dma_addr_t src_addr;
38 	dma_addr_t dst_addr;
39 	u32 size;
40 };
41 
42 /*
43  * struct rcar_dmac_hw_desc - Hardware descriptor for a transfer chunk
44  * @sar: value of the SAR register (source address)
45  * @dar: value of the DAR register (destination address)
46  * @tcr: value of the TCR register (transfer count)
47  */
48 struct rcar_dmac_hw_desc {
49 	u32 sar;
50 	u32 dar;
51 	u32 tcr;
52 	u32 reserved;
53 } __attribute__((__packed__));
54 
55 /*
56  * struct rcar_dmac_desc - R-Car Gen2 DMA Transfer Descriptor
57  * @async_tx: base DMA asynchronous transaction descriptor
58  * @direction: direction of the DMA transfer
59  * @xfer_shift: log2 of the transfer size
60  * @chcr: value of the channel configuration register for this transfer
61  * @node: entry in the channel's descriptors lists
62  * @chunks: list of transfer chunks for this transfer
63  * @running: the transfer chunk being currently processed
64  * @nchunks: number of transfer chunks for this transfer
65  * @hwdescs.use: whether the transfer descriptor uses hardware descriptors
66  * @hwdescs.mem: hardware descriptors memory for the transfer
67  * @hwdescs.dma: device address of the hardware descriptors memory
68  * @hwdescs.size: size of the hardware descriptors in bytes
69  * @size: transfer size in bytes
70  * @cyclic: when set indicates that the DMA transfer is cyclic
71  */
72 struct rcar_dmac_desc {
73 	struct dma_async_tx_descriptor async_tx;
74 	enum dma_transfer_direction direction;
75 	unsigned int xfer_shift;
76 	u32 chcr;
77 
78 	struct list_head node;
79 	struct list_head chunks;
80 	struct rcar_dmac_xfer_chunk *running;
81 	unsigned int nchunks;
82 
83 	struct {
84 		bool use;
85 		struct rcar_dmac_hw_desc *mem;
86 		dma_addr_t dma;
87 		size_t size;
88 	} hwdescs;
89 
90 	unsigned int size;
91 	bool cyclic;
92 };
93 
94 #define to_rcar_dmac_desc(d)	container_of(d, struct rcar_dmac_desc, async_tx)
95 
96 /*
97  * struct rcar_dmac_desc_page - One page worth of descriptors
98  * @node: entry in the channel's pages list
99  * @descs: array of DMA descriptors
100  * @chunks: array of transfer chunk descriptors
101  */
102 struct rcar_dmac_desc_page {
103 	struct list_head node;
104 
105 	union {
106 		struct rcar_dmac_desc descs[0];
107 		struct rcar_dmac_xfer_chunk chunks[0];
108 	};
109 };
110 
111 #define RCAR_DMAC_DESCS_PER_PAGE					\
112 	((PAGE_SIZE - offsetof(struct rcar_dmac_desc_page, descs)) /	\
113 	sizeof(struct rcar_dmac_desc))
114 #define RCAR_DMAC_XFER_CHUNKS_PER_PAGE					\
115 	((PAGE_SIZE - offsetof(struct rcar_dmac_desc_page, chunks)) /	\
116 	sizeof(struct rcar_dmac_xfer_chunk))
117 
118 /*
119  * struct rcar_dmac_chan_slave - Slave configuration
120  * @slave_addr: slave memory address
121  * @xfer_size: size (in bytes) of hardware transfers
122  */
123 struct rcar_dmac_chan_slave {
124 	phys_addr_t slave_addr;
125 	unsigned int xfer_size;
126 };
127 
128 /*
129  * struct rcar_dmac_chan_map - Map of slave device phys to dma address
130  * @addr: slave dma address
131  * @dir: direction of mapping
132  * @slave: slave configuration that is mapped
133  */
134 struct rcar_dmac_chan_map {
135 	dma_addr_t addr;
136 	enum dma_data_direction dir;
137 	struct rcar_dmac_chan_slave slave;
138 };
139 
140 /*
141  * struct rcar_dmac_chan - R-Car Gen2 DMA Controller Channel
142  * @chan: base DMA channel object
143  * @iomem: channel I/O memory base
144  * @index: index of this channel in the controller
145  * @irq: channel IRQ
146  * @src: slave memory address and size on the source side
147  * @dst: slave memory address and size on the destination side
148  * @mid_rid: hardware MID/RID for the DMA client using this channel
149  * @lock: protects the channel CHCR register and the desc members
150  * @desc.free: list of free descriptors
151  * @desc.pending: list of pending descriptors (submitted with tx_submit)
152  * @desc.active: list of active descriptors (activated with issue_pending)
153  * @desc.done: list of completed descriptors
154  * @desc.wait: list of descriptors waiting for an ack
155  * @desc.running: the descriptor being processed (a member of the active list)
156  * @desc.chunks_free: list of free transfer chunk descriptors
157  * @desc.pages: list of pages used by allocated descriptors
158  */
159 struct rcar_dmac_chan {
160 	struct dma_chan chan;
161 	void __iomem *iomem;
162 	unsigned int index;
163 	int irq;
164 
165 	struct rcar_dmac_chan_slave src;
166 	struct rcar_dmac_chan_slave dst;
167 	struct rcar_dmac_chan_map map;
168 	int mid_rid;
169 
170 	spinlock_t lock;
171 
172 	struct {
173 		struct list_head free;
174 		struct list_head pending;
175 		struct list_head active;
176 		struct list_head done;
177 		struct list_head wait;
178 		struct rcar_dmac_desc *running;
179 
180 		struct list_head chunks_free;
181 
182 		struct list_head pages;
183 	} desc;
184 };
185 
186 #define to_rcar_dmac_chan(c)	container_of(c, struct rcar_dmac_chan, chan)
187 
188 /*
189  * struct rcar_dmac - R-Car Gen2 DMA Controller
190  * @engine: base DMA engine object
191  * @dev: the hardware device
192  * @dmac_base: remapped base register block
193  * @chan_base: remapped channel register block (optional)
194  * @n_channels: number of available channels
195  * @channels: array of DMAC channels
196  * @channels_mask: bitfield of which DMA channels are managed by this driver
197  * @modules: bitmask of client modules in use
198  */
199 struct rcar_dmac {
200 	struct dma_device engine;
201 	struct device *dev;
202 	void __iomem *dmac_base;
203 	void __iomem *chan_base;
204 
205 	unsigned int n_channels;
206 	struct rcar_dmac_chan *channels;
207 	u32 channels_mask;
208 
209 	DECLARE_BITMAP(modules, 256);
210 };
211 
212 #define to_rcar_dmac(d)		container_of(d, struct rcar_dmac, engine)
213 
214 #define for_each_rcar_dmac_chan(i, dmac, chan)						\
215 	for (i = 0, chan = &(dmac)->channels[0]; i < (dmac)->n_channels; i++, chan++)	\
216 		if (!((dmac)->channels_mask & BIT(i))) continue; else
217 
218 /*
219  * struct rcar_dmac_of_data - This driver's OF data
220  * @chan_offset_base: DMAC channels base offset
221  * @chan_offset_stride: DMAC channels offset stride
222  */
223 struct rcar_dmac_of_data {
224 	u32 chan_offset_base;
225 	u32 chan_offset_stride;
226 };
227 
228 /* -----------------------------------------------------------------------------
229  * Registers
230  */
231 
232 #define RCAR_DMAISTA			0x0020
233 #define RCAR_DMASEC			0x0030
234 #define RCAR_DMAOR			0x0060
235 #define RCAR_DMAOR_PRI_FIXED		(0 << 8)
236 #define RCAR_DMAOR_PRI_ROUND_ROBIN	(3 << 8)
237 #define RCAR_DMAOR_AE			(1 << 2)
238 #define RCAR_DMAOR_DME			(1 << 0)
239 #define RCAR_DMACHCLR			0x0080	/* Not on R-Car V3U */
240 #define RCAR_DMADPSEC			0x00a0
241 
242 #define RCAR_DMASAR			0x0000
243 #define RCAR_DMADAR			0x0004
244 #define RCAR_DMATCR			0x0008
245 #define RCAR_DMATCR_MASK		0x00ffffff
246 #define RCAR_DMATSR			0x0028
247 #define RCAR_DMACHCR			0x000c
248 #define RCAR_DMACHCR_CAE		(1 << 31)
249 #define RCAR_DMACHCR_CAIE		(1 << 30)
250 #define RCAR_DMACHCR_DPM_DISABLED	(0 << 28)
251 #define RCAR_DMACHCR_DPM_ENABLED	(1 << 28)
252 #define RCAR_DMACHCR_DPM_REPEAT		(2 << 28)
253 #define RCAR_DMACHCR_DPM_INFINITE	(3 << 28)
254 #define RCAR_DMACHCR_RPT_SAR		(1 << 27)
255 #define RCAR_DMACHCR_RPT_DAR		(1 << 26)
256 #define RCAR_DMACHCR_RPT_TCR		(1 << 25)
257 #define RCAR_DMACHCR_DPB		(1 << 22)
258 #define RCAR_DMACHCR_DSE		(1 << 19)
259 #define RCAR_DMACHCR_DSIE		(1 << 18)
260 #define RCAR_DMACHCR_TS_1B		((0 << 20) | (0 << 3))
261 #define RCAR_DMACHCR_TS_2B		((0 << 20) | (1 << 3))
262 #define RCAR_DMACHCR_TS_4B		((0 << 20) | (2 << 3))
263 #define RCAR_DMACHCR_TS_16B		((0 << 20) | (3 << 3))
264 #define RCAR_DMACHCR_TS_32B		((1 << 20) | (0 << 3))
265 #define RCAR_DMACHCR_TS_64B		((1 << 20) | (1 << 3))
266 #define RCAR_DMACHCR_TS_8B		((1 << 20) | (3 << 3))
267 #define RCAR_DMACHCR_DM_FIXED		(0 << 14)
268 #define RCAR_DMACHCR_DM_INC		(1 << 14)
269 #define RCAR_DMACHCR_DM_DEC		(2 << 14)
270 #define RCAR_DMACHCR_SM_FIXED		(0 << 12)
271 #define RCAR_DMACHCR_SM_INC		(1 << 12)
272 #define RCAR_DMACHCR_SM_DEC		(2 << 12)
273 #define RCAR_DMACHCR_RS_AUTO		(4 << 8)
274 #define RCAR_DMACHCR_RS_DMARS		(8 << 8)
275 #define RCAR_DMACHCR_IE			(1 << 2)
276 #define RCAR_DMACHCR_TE			(1 << 1)
277 #define RCAR_DMACHCR_DE			(1 << 0)
278 #define RCAR_DMATCRB			0x0018
279 #define RCAR_DMATSRB			0x0038
280 #define RCAR_DMACHCRB			0x001c
281 #define RCAR_DMACHCRB_DCNT(n)		((n) << 24)
282 #define RCAR_DMACHCRB_DPTR_MASK		(0xff << 16)
283 #define RCAR_DMACHCRB_DPTR_SHIFT	16
284 #define RCAR_DMACHCRB_DRST		(1 << 15)
285 #define RCAR_DMACHCRB_DTS		(1 << 8)
286 #define RCAR_DMACHCRB_SLM_NORMAL	(0 << 4)
287 #define RCAR_DMACHCRB_SLM_CLK(n)	((8 | (n)) << 4)
288 #define RCAR_DMACHCRB_PRI(n)		((n) << 0)
289 #define RCAR_DMARS			0x0040
290 #define RCAR_DMABUFCR			0x0048
291 #define RCAR_DMABUFCR_MBU(n)		((n) << 16)
292 #define RCAR_DMABUFCR_ULB(n)		((n) << 0)
293 #define RCAR_DMADPBASE			0x0050
294 #define RCAR_DMADPBASE_MASK		0xfffffff0
295 #define RCAR_DMADPBASE_SEL		(1 << 0)
296 #define RCAR_DMADPCR			0x0054
297 #define RCAR_DMADPCR_DIPT(n)		((n) << 24)
298 #define RCAR_DMAFIXSAR			0x0010
299 #define RCAR_DMAFIXDAR			0x0014
300 #define RCAR_DMAFIXDPBASE		0x0060
301 
302 /* For R-Car V3U */
303 #define RCAR_V3U_DMACHCLR		0x0100
304 
305 /* Hardcode the MEMCPY transfer size to 4 bytes. */
306 #define RCAR_DMAC_MEMCPY_XFER_SIZE	4
307 
308 /* -----------------------------------------------------------------------------
309  * Device access
310  */
311 
312 static void rcar_dmac_write(struct rcar_dmac *dmac, u32 reg, u32 data)
313 {
314 	if (reg == RCAR_DMAOR)
315 		writew(data, dmac->dmac_base + reg);
316 	else
317 		writel(data, dmac->dmac_base + reg);
318 }
319 
320 static u32 rcar_dmac_read(struct rcar_dmac *dmac, u32 reg)
321 {
322 	if (reg == RCAR_DMAOR)
323 		return readw(dmac->dmac_base + reg);
324 	else
325 		return readl(dmac->dmac_base + reg);
326 }
327 
328 static u32 rcar_dmac_chan_read(struct rcar_dmac_chan *chan, u32 reg)
329 {
330 	if (reg == RCAR_DMARS)
331 		return readw(chan->iomem + reg);
332 	else
333 		return readl(chan->iomem + reg);
334 }
335 
336 static void rcar_dmac_chan_write(struct rcar_dmac_chan *chan, u32 reg, u32 data)
337 {
338 	if (reg == RCAR_DMARS)
339 		writew(data, chan->iomem + reg);
340 	else
341 		writel(data, chan->iomem + reg);
342 }
343 
344 static void rcar_dmac_chan_clear(struct rcar_dmac *dmac,
345 				 struct rcar_dmac_chan *chan)
346 {
347 	if (dmac->chan_base)
348 		rcar_dmac_chan_write(chan, RCAR_V3U_DMACHCLR, 1);
349 	else
350 		rcar_dmac_write(dmac, RCAR_DMACHCLR, BIT(chan->index));
351 }
352 
353 static void rcar_dmac_chan_clear_all(struct rcar_dmac *dmac)
354 {
355 	struct rcar_dmac_chan *chan;
356 	unsigned int i;
357 
358 	if (dmac->chan_base) {
359 		for_each_rcar_dmac_chan(i, dmac, chan)
360 			rcar_dmac_chan_write(chan, RCAR_V3U_DMACHCLR, 1);
361 	} else {
362 		rcar_dmac_write(dmac, RCAR_DMACHCLR, dmac->channels_mask);
363 	}
364 }
365 
366 /* -----------------------------------------------------------------------------
367  * Initialization and configuration
368  */
369 
370 static bool rcar_dmac_chan_is_busy(struct rcar_dmac_chan *chan)
371 {
372 	u32 chcr = rcar_dmac_chan_read(chan, RCAR_DMACHCR);
373 
374 	return !!(chcr & (RCAR_DMACHCR_DE | RCAR_DMACHCR_TE));
375 }
376 
377 static void rcar_dmac_chan_start_xfer(struct rcar_dmac_chan *chan)
378 {
379 	struct rcar_dmac_desc *desc = chan->desc.running;
380 	u32 chcr = desc->chcr;
381 
382 	WARN_ON_ONCE(rcar_dmac_chan_is_busy(chan));
383 
384 	if (chan->mid_rid >= 0)
385 		rcar_dmac_chan_write(chan, RCAR_DMARS, chan->mid_rid);
386 
387 	if (desc->hwdescs.use) {
388 		struct rcar_dmac_xfer_chunk *chunk =
389 			list_first_entry(&desc->chunks,
390 					 struct rcar_dmac_xfer_chunk, node);
391 
392 		dev_dbg(chan->chan.device->dev,
393 			"chan%u: queue desc %p: %u@%pad\n",
394 			chan->index, desc, desc->nchunks, &desc->hwdescs.dma);
395 
396 #ifdef CONFIG_ARCH_DMA_ADDR_T_64BIT
397 		rcar_dmac_chan_write(chan, RCAR_DMAFIXSAR,
398 				     chunk->src_addr >> 32);
399 		rcar_dmac_chan_write(chan, RCAR_DMAFIXDAR,
400 				     chunk->dst_addr >> 32);
401 		rcar_dmac_chan_write(chan, RCAR_DMAFIXDPBASE,
402 				     desc->hwdescs.dma >> 32);
403 #endif
404 		rcar_dmac_chan_write(chan, RCAR_DMADPBASE,
405 				     (desc->hwdescs.dma & 0xfffffff0) |
406 				     RCAR_DMADPBASE_SEL);
407 		rcar_dmac_chan_write(chan, RCAR_DMACHCRB,
408 				     RCAR_DMACHCRB_DCNT(desc->nchunks - 1) |
409 				     RCAR_DMACHCRB_DRST);
410 
411 		/*
412 		 * Errata: When descriptor memory is accessed through an IOMMU
413 		 * the DMADAR register isn't initialized automatically from the
414 		 * first descriptor at beginning of transfer by the DMAC like it
415 		 * should. Initialize it manually with the destination address
416 		 * of the first chunk.
417 		 */
418 		rcar_dmac_chan_write(chan, RCAR_DMADAR,
419 				     chunk->dst_addr & 0xffffffff);
420 
421 		/*
422 		 * Program the descriptor stage interrupt to occur after the end
423 		 * of the first stage.
424 		 */
425 		rcar_dmac_chan_write(chan, RCAR_DMADPCR, RCAR_DMADPCR_DIPT(1));
426 
427 		chcr |= RCAR_DMACHCR_RPT_SAR | RCAR_DMACHCR_RPT_DAR
428 		     |  RCAR_DMACHCR_RPT_TCR | RCAR_DMACHCR_DPB;
429 
430 		/*
431 		 * If the descriptor isn't cyclic enable normal descriptor mode
432 		 * and the transfer completion interrupt.
433 		 */
434 		if (!desc->cyclic)
435 			chcr |= RCAR_DMACHCR_DPM_ENABLED | RCAR_DMACHCR_IE;
436 		/*
437 		 * If the descriptor is cyclic and has a callback enable the
438 		 * descriptor stage interrupt in infinite repeat mode.
439 		 */
440 		else if (desc->async_tx.callback)
441 			chcr |= RCAR_DMACHCR_DPM_INFINITE | RCAR_DMACHCR_DSIE;
442 		/*
443 		 * Otherwise just select infinite repeat mode without any
444 		 * interrupt.
445 		 */
446 		else
447 			chcr |= RCAR_DMACHCR_DPM_INFINITE;
448 	} else {
449 		struct rcar_dmac_xfer_chunk *chunk = desc->running;
450 
451 		dev_dbg(chan->chan.device->dev,
452 			"chan%u: queue chunk %p: %u@%pad -> %pad\n",
453 			chan->index, chunk, chunk->size, &chunk->src_addr,
454 			&chunk->dst_addr);
455 
456 #ifdef CONFIG_ARCH_DMA_ADDR_T_64BIT
457 		rcar_dmac_chan_write(chan, RCAR_DMAFIXSAR,
458 				     chunk->src_addr >> 32);
459 		rcar_dmac_chan_write(chan, RCAR_DMAFIXDAR,
460 				     chunk->dst_addr >> 32);
461 #endif
462 		rcar_dmac_chan_write(chan, RCAR_DMASAR,
463 				     chunk->src_addr & 0xffffffff);
464 		rcar_dmac_chan_write(chan, RCAR_DMADAR,
465 				     chunk->dst_addr & 0xffffffff);
466 		rcar_dmac_chan_write(chan, RCAR_DMATCR,
467 				     chunk->size >> desc->xfer_shift);
468 
469 		chcr |= RCAR_DMACHCR_DPM_DISABLED | RCAR_DMACHCR_IE;
470 	}
471 
472 	rcar_dmac_chan_write(chan, RCAR_DMACHCR,
473 			     chcr | RCAR_DMACHCR_DE | RCAR_DMACHCR_CAIE);
474 }
475 
476 static int rcar_dmac_init(struct rcar_dmac *dmac)
477 {
478 	u16 dmaor;
479 
480 	/* Clear all channels and enable the DMAC globally. */
481 	rcar_dmac_chan_clear_all(dmac);
482 	rcar_dmac_write(dmac, RCAR_DMAOR,
483 			RCAR_DMAOR_PRI_FIXED | RCAR_DMAOR_DME);
484 
485 	dmaor = rcar_dmac_read(dmac, RCAR_DMAOR);
486 	if ((dmaor & (RCAR_DMAOR_AE | RCAR_DMAOR_DME)) != RCAR_DMAOR_DME) {
487 		dev_warn(dmac->dev, "DMAOR initialization failed.\n");
488 		return -EIO;
489 	}
490 
491 	return 0;
492 }
493 
494 /* -----------------------------------------------------------------------------
495  * Descriptors submission
496  */
497 
498 static dma_cookie_t rcar_dmac_tx_submit(struct dma_async_tx_descriptor *tx)
499 {
500 	struct rcar_dmac_chan *chan = to_rcar_dmac_chan(tx->chan);
501 	struct rcar_dmac_desc *desc = to_rcar_dmac_desc(tx);
502 	unsigned long flags;
503 	dma_cookie_t cookie;
504 
505 	spin_lock_irqsave(&chan->lock, flags);
506 
507 	cookie = dma_cookie_assign(tx);
508 
509 	dev_dbg(chan->chan.device->dev, "chan%u: submit #%d@%p\n",
510 		chan->index, tx->cookie, desc);
511 
512 	list_add_tail(&desc->node, &chan->desc.pending);
513 	desc->running = list_first_entry(&desc->chunks,
514 					 struct rcar_dmac_xfer_chunk, node);
515 
516 	spin_unlock_irqrestore(&chan->lock, flags);
517 
518 	return cookie;
519 }
520 
521 /* -----------------------------------------------------------------------------
522  * Descriptors allocation and free
523  */
524 
525 /*
526  * rcar_dmac_desc_alloc - Allocate a page worth of DMA descriptors
527  * @chan: the DMA channel
528  * @gfp: allocation flags
529  */
530 static int rcar_dmac_desc_alloc(struct rcar_dmac_chan *chan, gfp_t gfp)
531 {
532 	struct rcar_dmac_desc_page *page;
533 	unsigned long flags;
534 	LIST_HEAD(list);
535 	unsigned int i;
536 
537 	page = (void *)get_zeroed_page(gfp);
538 	if (!page)
539 		return -ENOMEM;
540 
541 	for (i = 0; i < RCAR_DMAC_DESCS_PER_PAGE; ++i) {
542 		struct rcar_dmac_desc *desc = &page->descs[i];
543 
544 		dma_async_tx_descriptor_init(&desc->async_tx, &chan->chan);
545 		desc->async_tx.tx_submit = rcar_dmac_tx_submit;
546 		INIT_LIST_HEAD(&desc->chunks);
547 
548 		list_add_tail(&desc->node, &list);
549 	}
550 
551 	spin_lock_irqsave(&chan->lock, flags);
552 	list_splice_tail(&list, &chan->desc.free);
553 	list_add_tail(&page->node, &chan->desc.pages);
554 	spin_unlock_irqrestore(&chan->lock, flags);
555 
556 	return 0;
557 }
558 
559 /*
560  * rcar_dmac_desc_put - Release a DMA transfer descriptor
561  * @chan: the DMA channel
562  * @desc: the descriptor
563  *
564  * Put the descriptor and its transfer chunk descriptors back in the channel's
565  * free descriptors lists. The descriptor's chunks list will be reinitialized to
566  * an empty list as a result.
567  *
568  * The descriptor must have been removed from the channel's lists before calling
569  * this function.
570  */
571 static void rcar_dmac_desc_put(struct rcar_dmac_chan *chan,
572 			       struct rcar_dmac_desc *desc)
573 {
574 	unsigned long flags;
575 
576 	spin_lock_irqsave(&chan->lock, flags);
577 	list_splice_tail_init(&desc->chunks, &chan->desc.chunks_free);
578 	list_add(&desc->node, &chan->desc.free);
579 	spin_unlock_irqrestore(&chan->lock, flags);
580 }
581 
582 static void rcar_dmac_desc_recycle_acked(struct rcar_dmac_chan *chan)
583 {
584 	struct rcar_dmac_desc *desc, *_desc;
585 	unsigned long flags;
586 	LIST_HEAD(list);
587 
588 	/*
589 	 * We have to temporarily move all descriptors from the wait list to a
590 	 * local list as iterating over the wait list, even with
591 	 * list_for_each_entry_safe, isn't safe if we release the channel lock
592 	 * around the rcar_dmac_desc_put() call.
593 	 */
594 	spin_lock_irqsave(&chan->lock, flags);
595 	list_splice_init(&chan->desc.wait, &list);
596 	spin_unlock_irqrestore(&chan->lock, flags);
597 
598 	list_for_each_entry_safe(desc, _desc, &list, node) {
599 		if (async_tx_test_ack(&desc->async_tx)) {
600 			list_del(&desc->node);
601 			rcar_dmac_desc_put(chan, desc);
602 		}
603 	}
604 
605 	if (list_empty(&list))
606 		return;
607 
608 	/* Put the remaining descriptors back in the wait list. */
609 	spin_lock_irqsave(&chan->lock, flags);
610 	list_splice(&list, &chan->desc.wait);
611 	spin_unlock_irqrestore(&chan->lock, flags);
612 }
613 
614 /*
615  * rcar_dmac_desc_get - Allocate a descriptor for a DMA transfer
616  * @chan: the DMA channel
617  *
618  * Locking: This function must be called in a non-atomic context.
619  *
620  * Return: A pointer to the allocated descriptor or NULL if no descriptor can
621  * be allocated.
622  */
623 static struct rcar_dmac_desc *rcar_dmac_desc_get(struct rcar_dmac_chan *chan)
624 {
625 	struct rcar_dmac_desc *desc;
626 	unsigned long flags;
627 	int ret;
628 
629 	/* Recycle acked descriptors before attempting allocation. */
630 	rcar_dmac_desc_recycle_acked(chan);
631 
632 	spin_lock_irqsave(&chan->lock, flags);
633 
634 	while (list_empty(&chan->desc.free)) {
635 		/*
636 		 * No free descriptors, allocate a page worth of them and try
637 		 * again, as someone else could race us to get the newly
638 		 * allocated descriptors. If the allocation fails return an
639 		 * error.
640 		 */
641 		spin_unlock_irqrestore(&chan->lock, flags);
642 		ret = rcar_dmac_desc_alloc(chan, GFP_NOWAIT);
643 		if (ret < 0)
644 			return NULL;
645 		spin_lock_irqsave(&chan->lock, flags);
646 	}
647 
648 	desc = list_first_entry(&chan->desc.free, struct rcar_dmac_desc, node);
649 	list_del(&desc->node);
650 
651 	spin_unlock_irqrestore(&chan->lock, flags);
652 
653 	return desc;
654 }
655 
656 /*
657  * rcar_dmac_xfer_chunk_alloc - Allocate a page worth of transfer chunks
658  * @chan: the DMA channel
659  * @gfp: allocation flags
660  */
661 static int rcar_dmac_xfer_chunk_alloc(struct rcar_dmac_chan *chan, gfp_t gfp)
662 {
663 	struct rcar_dmac_desc_page *page;
664 	unsigned long flags;
665 	LIST_HEAD(list);
666 	unsigned int i;
667 
668 	page = (void *)get_zeroed_page(gfp);
669 	if (!page)
670 		return -ENOMEM;
671 
672 	for (i = 0; i < RCAR_DMAC_XFER_CHUNKS_PER_PAGE; ++i) {
673 		struct rcar_dmac_xfer_chunk *chunk = &page->chunks[i];
674 
675 		list_add_tail(&chunk->node, &list);
676 	}
677 
678 	spin_lock_irqsave(&chan->lock, flags);
679 	list_splice_tail(&list, &chan->desc.chunks_free);
680 	list_add_tail(&page->node, &chan->desc.pages);
681 	spin_unlock_irqrestore(&chan->lock, flags);
682 
683 	return 0;
684 }
685 
686 /*
687  * rcar_dmac_xfer_chunk_get - Allocate a transfer chunk for a DMA transfer
688  * @chan: the DMA channel
689  *
690  * Locking: This function must be called in a non-atomic context.
691  *
692  * Return: A pointer to the allocated transfer chunk descriptor or NULL if no
693  * descriptor can be allocated.
694  */
695 static struct rcar_dmac_xfer_chunk *
696 rcar_dmac_xfer_chunk_get(struct rcar_dmac_chan *chan)
697 {
698 	struct rcar_dmac_xfer_chunk *chunk;
699 	unsigned long flags;
700 	int ret;
701 
702 	spin_lock_irqsave(&chan->lock, flags);
703 
704 	while (list_empty(&chan->desc.chunks_free)) {
705 		/*
706 		 * No free descriptors, allocate a page worth of them and try
707 		 * again, as someone else could race us to get the newly
708 		 * allocated descriptors. If the allocation fails return an
709 		 * error.
710 		 */
711 		spin_unlock_irqrestore(&chan->lock, flags);
712 		ret = rcar_dmac_xfer_chunk_alloc(chan, GFP_NOWAIT);
713 		if (ret < 0)
714 			return NULL;
715 		spin_lock_irqsave(&chan->lock, flags);
716 	}
717 
718 	chunk = list_first_entry(&chan->desc.chunks_free,
719 				 struct rcar_dmac_xfer_chunk, node);
720 	list_del(&chunk->node);
721 
722 	spin_unlock_irqrestore(&chan->lock, flags);
723 
724 	return chunk;
725 }
726 
727 static void rcar_dmac_realloc_hwdesc(struct rcar_dmac_chan *chan,
728 				     struct rcar_dmac_desc *desc, size_t size)
729 {
730 	/*
731 	 * dma_alloc_coherent() allocates memory in page size increments. To
732 	 * avoid reallocating the hardware descriptors when the allocated size
733 	 * wouldn't change align the requested size to a multiple of the page
734 	 * size.
735 	 */
736 	size = PAGE_ALIGN(size);
737 
738 	if (desc->hwdescs.size == size)
739 		return;
740 
741 	if (desc->hwdescs.mem) {
742 		dma_free_coherent(chan->chan.device->dev, desc->hwdescs.size,
743 				  desc->hwdescs.mem, desc->hwdescs.dma);
744 		desc->hwdescs.mem = NULL;
745 		desc->hwdescs.size = 0;
746 	}
747 
748 	if (!size)
749 		return;
750 
751 	desc->hwdescs.mem = dma_alloc_coherent(chan->chan.device->dev, size,
752 					       &desc->hwdescs.dma, GFP_NOWAIT);
753 	if (!desc->hwdescs.mem)
754 		return;
755 
756 	desc->hwdescs.size = size;
757 }
758 
759 static int rcar_dmac_fill_hwdesc(struct rcar_dmac_chan *chan,
760 				 struct rcar_dmac_desc *desc)
761 {
762 	struct rcar_dmac_xfer_chunk *chunk;
763 	struct rcar_dmac_hw_desc *hwdesc;
764 
765 	rcar_dmac_realloc_hwdesc(chan, desc, desc->nchunks * sizeof(*hwdesc));
766 
767 	hwdesc = desc->hwdescs.mem;
768 	if (!hwdesc)
769 		return -ENOMEM;
770 
771 	list_for_each_entry(chunk, &desc->chunks, node) {
772 		hwdesc->sar = chunk->src_addr;
773 		hwdesc->dar = chunk->dst_addr;
774 		hwdesc->tcr = chunk->size >> desc->xfer_shift;
775 		hwdesc++;
776 	}
777 
778 	return 0;
779 }
780 
781 /* -----------------------------------------------------------------------------
782  * Stop and reset
783  */
784 static void rcar_dmac_chcr_de_barrier(struct rcar_dmac_chan *chan)
785 {
786 	u32 chcr;
787 	unsigned int i;
788 
789 	/*
790 	 * Ensure that the setting of the DE bit is actually 0 after
791 	 * clearing it.
792 	 */
793 	for (i = 0; i < 1024; i++) {
794 		chcr = rcar_dmac_chan_read(chan, RCAR_DMACHCR);
795 		if (!(chcr & RCAR_DMACHCR_DE))
796 			return;
797 		udelay(1);
798 	}
799 
800 	dev_err(chan->chan.device->dev, "CHCR DE check error\n");
801 }
802 
803 static void rcar_dmac_clear_chcr_de(struct rcar_dmac_chan *chan)
804 {
805 	u32 chcr = rcar_dmac_chan_read(chan, RCAR_DMACHCR);
806 
807 	/* set DE=0 and flush remaining data */
808 	rcar_dmac_chan_write(chan, RCAR_DMACHCR, (chcr & ~RCAR_DMACHCR_DE));
809 
810 	/* make sure all remaining data was flushed */
811 	rcar_dmac_chcr_de_barrier(chan);
812 }
813 
814 static void rcar_dmac_chan_halt(struct rcar_dmac_chan *chan)
815 {
816 	u32 chcr = rcar_dmac_chan_read(chan, RCAR_DMACHCR);
817 
818 	chcr &= ~(RCAR_DMACHCR_DSE | RCAR_DMACHCR_DSIE | RCAR_DMACHCR_IE |
819 		  RCAR_DMACHCR_TE | RCAR_DMACHCR_DE |
820 		  RCAR_DMACHCR_CAE | RCAR_DMACHCR_CAIE);
821 	rcar_dmac_chan_write(chan, RCAR_DMACHCR, chcr);
822 	rcar_dmac_chcr_de_barrier(chan);
823 }
824 
825 static void rcar_dmac_chan_reinit(struct rcar_dmac_chan *chan)
826 {
827 	struct rcar_dmac_desc *desc, *_desc;
828 	unsigned long flags;
829 	LIST_HEAD(descs);
830 
831 	spin_lock_irqsave(&chan->lock, flags);
832 
833 	/* Move all non-free descriptors to the local lists. */
834 	list_splice_init(&chan->desc.pending, &descs);
835 	list_splice_init(&chan->desc.active, &descs);
836 	list_splice_init(&chan->desc.done, &descs);
837 	list_splice_init(&chan->desc.wait, &descs);
838 
839 	chan->desc.running = NULL;
840 
841 	spin_unlock_irqrestore(&chan->lock, flags);
842 
843 	list_for_each_entry_safe(desc, _desc, &descs, node) {
844 		list_del(&desc->node);
845 		rcar_dmac_desc_put(chan, desc);
846 	}
847 }
848 
849 static void rcar_dmac_stop_all_chan(struct rcar_dmac *dmac)
850 {
851 	struct rcar_dmac_chan *chan;
852 	unsigned int i;
853 
854 	/* Stop all channels. */
855 	for_each_rcar_dmac_chan(i, dmac, chan) {
856 		/* Stop and reinitialize the channel. */
857 		spin_lock_irq(&chan->lock);
858 		rcar_dmac_chan_halt(chan);
859 		spin_unlock_irq(&chan->lock);
860 	}
861 }
862 
863 static int rcar_dmac_chan_pause(struct dma_chan *chan)
864 {
865 	unsigned long flags;
866 	struct rcar_dmac_chan *rchan = to_rcar_dmac_chan(chan);
867 
868 	spin_lock_irqsave(&rchan->lock, flags);
869 	rcar_dmac_clear_chcr_de(rchan);
870 	spin_unlock_irqrestore(&rchan->lock, flags);
871 
872 	return 0;
873 }
874 
875 /* -----------------------------------------------------------------------------
876  * Descriptors preparation
877  */
878 
879 static void rcar_dmac_chan_configure_desc(struct rcar_dmac_chan *chan,
880 					  struct rcar_dmac_desc *desc)
881 {
882 	static const u32 chcr_ts[] = {
883 		RCAR_DMACHCR_TS_1B, RCAR_DMACHCR_TS_2B,
884 		RCAR_DMACHCR_TS_4B, RCAR_DMACHCR_TS_8B,
885 		RCAR_DMACHCR_TS_16B, RCAR_DMACHCR_TS_32B,
886 		RCAR_DMACHCR_TS_64B,
887 	};
888 
889 	unsigned int xfer_size;
890 	u32 chcr;
891 
892 	switch (desc->direction) {
893 	case DMA_DEV_TO_MEM:
894 		chcr = RCAR_DMACHCR_DM_INC | RCAR_DMACHCR_SM_FIXED
895 		     | RCAR_DMACHCR_RS_DMARS;
896 		xfer_size = chan->src.xfer_size;
897 		break;
898 
899 	case DMA_MEM_TO_DEV:
900 		chcr = RCAR_DMACHCR_DM_FIXED | RCAR_DMACHCR_SM_INC
901 		     | RCAR_DMACHCR_RS_DMARS;
902 		xfer_size = chan->dst.xfer_size;
903 		break;
904 
905 	case DMA_MEM_TO_MEM:
906 	default:
907 		chcr = RCAR_DMACHCR_DM_INC | RCAR_DMACHCR_SM_INC
908 		     | RCAR_DMACHCR_RS_AUTO;
909 		xfer_size = RCAR_DMAC_MEMCPY_XFER_SIZE;
910 		break;
911 	}
912 
913 	desc->xfer_shift = ilog2(xfer_size);
914 	desc->chcr = chcr | chcr_ts[desc->xfer_shift];
915 }
916 
917 /*
918  * rcar_dmac_chan_prep_sg - prepare transfer descriptors from an SG list
919  *
920  * Common routine for public (MEMCPY) and slave DMA. The MEMCPY case is also
921  * converted to scatter-gather to guarantee consistent locking and a correct
922  * list manipulation. For slave DMA direction carries the usual meaning, and,
923  * logically, the SG list is RAM and the addr variable contains slave address,
924  * e.g., the FIFO I/O register. For MEMCPY direction equals DMA_MEM_TO_MEM
925  * and the SG list contains only one element and points at the source buffer.
926  */
927 static struct dma_async_tx_descriptor *
928 rcar_dmac_chan_prep_sg(struct rcar_dmac_chan *chan, struct scatterlist *sgl,
929 		       unsigned int sg_len, dma_addr_t dev_addr,
930 		       enum dma_transfer_direction dir, unsigned long dma_flags,
931 		       bool cyclic)
932 {
933 	struct rcar_dmac_xfer_chunk *chunk;
934 	struct rcar_dmac_desc *desc;
935 	struct scatterlist *sg;
936 	unsigned int nchunks = 0;
937 	unsigned int max_chunk_size;
938 	unsigned int full_size = 0;
939 	bool cross_boundary = false;
940 	unsigned int i;
941 #ifdef CONFIG_ARCH_DMA_ADDR_T_64BIT
942 	u32 high_dev_addr;
943 	u32 high_mem_addr;
944 #endif
945 
946 	desc = rcar_dmac_desc_get(chan);
947 	if (!desc)
948 		return NULL;
949 
950 	desc->async_tx.flags = dma_flags;
951 	desc->async_tx.cookie = -EBUSY;
952 
953 	desc->cyclic = cyclic;
954 	desc->direction = dir;
955 
956 	rcar_dmac_chan_configure_desc(chan, desc);
957 
958 	max_chunk_size = RCAR_DMATCR_MASK << desc->xfer_shift;
959 
960 	/*
961 	 * Allocate and fill the transfer chunk descriptors. We own the only
962 	 * reference to the DMA descriptor, there's no need for locking.
963 	 */
964 	for_each_sg(sgl, sg, sg_len, i) {
965 		dma_addr_t mem_addr = sg_dma_address(sg);
966 		unsigned int len = sg_dma_len(sg);
967 
968 		full_size += len;
969 
970 #ifdef CONFIG_ARCH_DMA_ADDR_T_64BIT
971 		if (i == 0) {
972 			high_dev_addr = dev_addr >> 32;
973 			high_mem_addr = mem_addr >> 32;
974 		}
975 
976 		if ((dev_addr >> 32 != high_dev_addr) ||
977 		    (mem_addr >> 32 != high_mem_addr))
978 			cross_boundary = true;
979 #endif
980 		while (len) {
981 			unsigned int size = min(len, max_chunk_size);
982 
983 #ifdef CONFIG_ARCH_DMA_ADDR_T_64BIT
984 			/*
985 			 * Prevent individual transfers from crossing 4GB
986 			 * boundaries.
987 			 */
988 			if (dev_addr >> 32 != (dev_addr + size - 1) >> 32) {
989 				size = ALIGN(dev_addr, 1ULL << 32) - dev_addr;
990 				cross_boundary = true;
991 			}
992 			if (mem_addr >> 32 != (mem_addr + size - 1) >> 32) {
993 				size = ALIGN(mem_addr, 1ULL << 32) - mem_addr;
994 				cross_boundary = true;
995 			}
996 #endif
997 
998 			chunk = rcar_dmac_xfer_chunk_get(chan);
999 			if (!chunk) {
1000 				rcar_dmac_desc_put(chan, desc);
1001 				return NULL;
1002 			}
1003 
1004 			if (dir == DMA_DEV_TO_MEM) {
1005 				chunk->src_addr = dev_addr;
1006 				chunk->dst_addr = mem_addr;
1007 			} else {
1008 				chunk->src_addr = mem_addr;
1009 				chunk->dst_addr = dev_addr;
1010 			}
1011 
1012 			chunk->size = size;
1013 
1014 			dev_dbg(chan->chan.device->dev,
1015 				"chan%u: chunk %p/%p sgl %u@%p, %u/%u %pad -> %pad\n",
1016 				chan->index, chunk, desc, i, sg, size, len,
1017 				&chunk->src_addr, &chunk->dst_addr);
1018 
1019 			mem_addr += size;
1020 			if (dir == DMA_MEM_TO_MEM)
1021 				dev_addr += size;
1022 
1023 			len -= size;
1024 
1025 			list_add_tail(&chunk->node, &desc->chunks);
1026 			nchunks++;
1027 		}
1028 	}
1029 
1030 	desc->nchunks = nchunks;
1031 	desc->size = full_size;
1032 
1033 	/*
1034 	 * Use hardware descriptor lists if possible when more than one chunk
1035 	 * needs to be transferred (otherwise they don't make much sense).
1036 	 *
1037 	 * Source/Destination address should be located in same 4GiB region
1038 	 * in the 40bit address space when it uses Hardware descriptor,
1039 	 * and cross_boundary is checking it.
1040 	 */
1041 	desc->hwdescs.use = !cross_boundary && nchunks > 1;
1042 	if (desc->hwdescs.use) {
1043 		if (rcar_dmac_fill_hwdesc(chan, desc) < 0)
1044 			desc->hwdescs.use = false;
1045 	}
1046 
1047 	return &desc->async_tx;
1048 }
1049 
1050 /* -----------------------------------------------------------------------------
1051  * DMA engine operations
1052  */
1053 
1054 static int rcar_dmac_alloc_chan_resources(struct dma_chan *chan)
1055 {
1056 	struct rcar_dmac_chan *rchan = to_rcar_dmac_chan(chan);
1057 	int ret;
1058 
1059 	INIT_LIST_HEAD(&rchan->desc.chunks_free);
1060 	INIT_LIST_HEAD(&rchan->desc.pages);
1061 
1062 	/* Preallocate descriptors. */
1063 	ret = rcar_dmac_xfer_chunk_alloc(rchan, GFP_KERNEL);
1064 	if (ret < 0)
1065 		return -ENOMEM;
1066 
1067 	ret = rcar_dmac_desc_alloc(rchan, GFP_KERNEL);
1068 	if (ret < 0)
1069 		return -ENOMEM;
1070 
1071 	return pm_runtime_get_sync(chan->device->dev);
1072 }
1073 
1074 static void rcar_dmac_free_chan_resources(struct dma_chan *chan)
1075 {
1076 	struct rcar_dmac_chan *rchan = to_rcar_dmac_chan(chan);
1077 	struct rcar_dmac *dmac = to_rcar_dmac(chan->device);
1078 	struct rcar_dmac_chan_map *map = &rchan->map;
1079 	struct rcar_dmac_desc_page *page, *_page;
1080 	struct rcar_dmac_desc *desc;
1081 	LIST_HEAD(list);
1082 
1083 	/* Protect against ISR */
1084 	spin_lock_irq(&rchan->lock);
1085 	rcar_dmac_chan_halt(rchan);
1086 	spin_unlock_irq(&rchan->lock);
1087 
1088 	/*
1089 	 * Now no new interrupts will occur, but one might already be
1090 	 * running. Wait for it to finish before freeing resources.
1091 	 */
1092 	synchronize_irq(rchan->irq);
1093 
1094 	if (rchan->mid_rid >= 0) {
1095 		/* The caller is holding dma_list_mutex */
1096 		clear_bit(rchan->mid_rid, dmac->modules);
1097 		rchan->mid_rid = -EINVAL;
1098 	}
1099 
1100 	list_splice_init(&rchan->desc.free, &list);
1101 	list_splice_init(&rchan->desc.pending, &list);
1102 	list_splice_init(&rchan->desc.active, &list);
1103 	list_splice_init(&rchan->desc.done, &list);
1104 	list_splice_init(&rchan->desc.wait, &list);
1105 
1106 	rchan->desc.running = NULL;
1107 
1108 	list_for_each_entry(desc, &list, node)
1109 		rcar_dmac_realloc_hwdesc(rchan, desc, 0);
1110 
1111 	list_for_each_entry_safe(page, _page, &rchan->desc.pages, node) {
1112 		list_del(&page->node);
1113 		free_page((unsigned long)page);
1114 	}
1115 
1116 	/* Remove slave mapping if present. */
1117 	if (map->slave.xfer_size) {
1118 		dma_unmap_resource(chan->device->dev, map->addr,
1119 				   map->slave.xfer_size, map->dir, 0);
1120 		map->slave.xfer_size = 0;
1121 	}
1122 
1123 	pm_runtime_put(chan->device->dev);
1124 }
1125 
1126 static struct dma_async_tx_descriptor *
1127 rcar_dmac_prep_dma_memcpy(struct dma_chan *chan, dma_addr_t dma_dest,
1128 			  dma_addr_t dma_src, size_t len, unsigned long flags)
1129 {
1130 	struct rcar_dmac_chan *rchan = to_rcar_dmac_chan(chan);
1131 	struct scatterlist sgl;
1132 
1133 	if (!len)
1134 		return NULL;
1135 
1136 	sg_init_table(&sgl, 1);
1137 	sg_set_page(&sgl, pfn_to_page(PFN_DOWN(dma_src)), len,
1138 		    offset_in_page(dma_src));
1139 	sg_dma_address(&sgl) = dma_src;
1140 	sg_dma_len(&sgl) = len;
1141 
1142 	return rcar_dmac_chan_prep_sg(rchan, &sgl, 1, dma_dest,
1143 				      DMA_MEM_TO_MEM, flags, false);
1144 }
1145 
1146 static int rcar_dmac_map_slave_addr(struct dma_chan *chan,
1147 				    enum dma_transfer_direction dir)
1148 {
1149 	struct rcar_dmac_chan *rchan = to_rcar_dmac_chan(chan);
1150 	struct rcar_dmac_chan_map *map = &rchan->map;
1151 	phys_addr_t dev_addr;
1152 	size_t dev_size;
1153 	enum dma_data_direction dev_dir;
1154 
1155 	if (dir == DMA_DEV_TO_MEM) {
1156 		dev_addr = rchan->src.slave_addr;
1157 		dev_size = rchan->src.xfer_size;
1158 		dev_dir = DMA_TO_DEVICE;
1159 	} else {
1160 		dev_addr = rchan->dst.slave_addr;
1161 		dev_size = rchan->dst.xfer_size;
1162 		dev_dir = DMA_FROM_DEVICE;
1163 	}
1164 
1165 	/* Reuse current map if possible. */
1166 	if (dev_addr == map->slave.slave_addr &&
1167 	    dev_size == map->slave.xfer_size &&
1168 	    dev_dir == map->dir)
1169 		return 0;
1170 
1171 	/* Remove old mapping if present. */
1172 	if (map->slave.xfer_size)
1173 		dma_unmap_resource(chan->device->dev, map->addr,
1174 				   map->slave.xfer_size, map->dir, 0);
1175 	map->slave.xfer_size = 0;
1176 
1177 	/* Create new slave address map. */
1178 	map->addr = dma_map_resource(chan->device->dev, dev_addr, dev_size,
1179 				     dev_dir, 0);
1180 
1181 	if (dma_mapping_error(chan->device->dev, map->addr)) {
1182 		dev_err(chan->device->dev,
1183 			"chan%u: failed to map %zx@%pap", rchan->index,
1184 			dev_size, &dev_addr);
1185 		return -EIO;
1186 	}
1187 
1188 	dev_dbg(chan->device->dev, "chan%u: map %zx@%pap to %pad dir: %s\n",
1189 		rchan->index, dev_size, &dev_addr, &map->addr,
1190 		dev_dir == DMA_TO_DEVICE ? "DMA_TO_DEVICE" : "DMA_FROM_DEVICE");
1191 
1192 	map->slave.slave_addr = dev_addr;
1193 	map->slave.xfer_size = dev_size;
1194 	map->dir = dev_dir;
1195 
1196 	return 0;
1197 }
1198 
1199 static struct dma_async_tx_descriptor *
1200 rcar_dmac_prep_slave_sg(struct dma_chan *chan, struct scatterlist *sgl,
1201 			unsigned int sg_len, enum dma_transfer_direction dir,
1202 			unsigned long flags, void *context)
1203 {
1204 	struct rcar_dmac_chan *rchan = to_rcar_dmac_chan(chan);
1205 
1206 	/* Someone calling slave DMA on a generic channel? */
1207 	if (rchan->mid_rid < 0 || !sg_len || !sg_dma_len(sgl)) {
1208 		dev_warn(chan->device->dev,
1209 			 "%s: bad parameter: len=%d, id=%d\n",
1210 			 __func__, sg_len, rchan->mid_rid);
1211 		return NULL;
1212 	}
1213 
1214 	if (rcar_dmac_map_slave_addr(chan, dir))
1215 		return NULL;
1216 
1217 	return rcar_dmac_chan_prep_sg(rchan, sgl, sg_len, rchan->map.addr,
1218 				      dir, flags, false);
1219 }
1220 
1221 #define RCAR_DMAC_MAX_SG_LEN	32
1222 
1223 static struct dma_async_tx_descriptor *
1224 rcar_dmac_prep_dma_cyclic(struct dma_chan *chan, dma_addr_t buf_addr,
1225 			  size_t buf_len, size_t period_len,
1226 			  enum dma_transfer_direction dir, unsigned long flags)
1227 {
1228 	struct rcar_dmac_chan *rchan = to_rcar_dmac_chan(chan);
1229 	struct dma_async_tx_descriptor *desc;
1230 	struct scatterlist *sgl;
1231 	unsigned int sg_len;
1232 	unsigned int i;
1233 
1234 	/* Someone calling slave DMA on a generic channel? */
1235 	if (rchan->mid_rid < 0 || buf_len < period_len) {
1236 		dev_warn(chan->device->dev,
1237 			"%s: bad parameter: buf_len=%zu, period_len=%zu, id=%d\n",
1238 			__func__, buf_len, period_len, rchan->mid_rid);
1239 		return NULL;
1240 	}
1241 
1242 	if (rcar_dmac_map_slave_addr(chan, dir))
1243 		return NULL;
1244 
1245 	sg_len = buf_len / period_len;
1246 	if (sg_len > RCAR_DMAC_MAX_SG_LEN) {
1247 		dev_err(chan->device->dev,
1248 			"chan%u: sg length %d exceeds limit %d",
1249 			rchan->index, sg_len, RCAR_DMAC_MAX_SG_LEN);
1250 		return NULL;
1251 	}
1252 
1253 	/*
1254 	 * Allocate the sg list dynamically as it would consume too much stack
1255 	 * space.
1256 	 */
1257 	sgl = kmalloc_array(sg_len, sizeof(*sgl), GFP_NOWAIT);
1258 	if (!sgl)
1259 		return NULL;
1260 
1261 	sg_init_table(sgl, sg_len);
1262 
1263 	for (i = 0; i < sg_len; ++i) {
1264 		dma_addr_t src = buf_addr + (period_len * i);
1265 
1266 		sg_set_page(&sgl[i], pfn_to_page(PFN_DOWN(src)), period_len,
1267 			    offset_in_page(src));
1268 		sg_dma_address(&sgl[i]) = src;
1269 		sg_dma_len(&sgl[i]) = period_len;
1270 	}
1271 
1272 	desc = rcar_dmac_chan_prep_sg(rchan, sgl, sg_len, rchan->map.addr,
1273 				      dir, flags, true);
1274 
1275 	kfree(sgl);
1276 	return desc;
1277 }
1278 
1279 static int rcar_dmac_device_config(struct dma_chan *chan,
1280 				   struct dma_slave_config *cfg)
1281 {
1282 	struct rcar_dmac_chan *rchan = to_rcar_dmac_chan(chan);
1283 
1284 	/*
1285 	 * We could lock this, but you shouldn't be configuring the
1286 	 * channel, while using it...
1287 	 */
1288 	rchan->src.slave_addr = cfg->src_addr;
1289 	rchan->dst.slave_addr = cfg->dst_addr;
1290 	rchan->src.xfer_size = cfg->src_addr_width;
1291 	rchan->dst.xfer_size = cfg->dst_addr_width;
1292 
1293 	return 0;
1294 }
1295 
1296 static int rcar_dmac_chan_terminate_all(struct dma_chan *chan)
1297 {
1298 	struct rcar_dmac_chan *rchan = to_rcar_dmac_chan(chan);
1299 	unsigned long flags;
1300 
1301 	spin_lock_irqsave(&rchan->lock, flags);
1302 	rcar_dmac_chan_halt(rchan);
1303 	spin_unlock_irqrestore(&rchan->lock, flags);
1304 
1305 	/*
1306 	 * FIXME: No new interrupt can occur now, but the IRQ thread might still
1307 	 * be running.
1308 	 */
1309 
1310 	rcar_dmac_chan_reinit(rchan);
1311 
1312 	return 0;
1313 }
1314 
1315 static unsigned int rcar_dmac_chan_get_residue(struct rcar_dmac_chan *chan,
1316 					       dma_cookie_t cookie)
1317 {
1318 	struct rcar_dmac_desc *desc = chan->desc.running;
1319 	struct rcar_dmac_xfer_chunk *running = NULL;
1320 	struct rcar_dmac_xfer_chunk *chunk;
1321 	enum dma_status status;
1322 	unsigned int residue = 0;
1323 	unsigned int dptr = 0;
1324 	unsigned int chcrb;
1325 	unsigned int tcrb;
1326 	unsigned int i;
1327 
1328 	if (!desc)
1329 		return 0;
1330 
1331 	/*
1332 	 * If the cookie corresponds to a descriptor that has been completed
1333 	 * there is no residue. The same check has already been performed by the
1334 	 * caller but without holding the channel lock, so the descriptor could
1335 	 * now be complete.
1336 	 */
1337 	status = dma_cookie_status(&chan->chan, cookie, NULL);
1338 	if (status == DMA_COMPLETE)
1339 		return 0;
1340 
1341 	/*
1342 	 * If the cookie doesn't correspond to the currently running transfer
1343 	 * then the descriptor hasn't been processed yet, and the residue is
1344 	 * equal to the full descriptor size.
1345 	 * Also, a client driver is possible to call this function before
1346 	 * rcar_dmac_isr_channel_thread() runs. In this case, the "desc.running"
1347 	 * will be the next descriptor, and the done list will appear. So, if
1348 	 * the argument cookie matches the done list's cookie, we can assume
1349 	 * the residue is zero.
1350 	 */
1351 	if (cookie != desc->async_tx.cookie) {
1352 		list_for_each_entry(desc, &chan->desc.done, node) {
1353 			if (cookie == desc->async_tx.cookie)
1354 				return 0;
1355 		}
1356 		list_for_each_entry(desc, &chan->desc.pending, node) {
1357 			if (cookie == desc->async_tx.cookie)
1358 				return desc->size;
1359 		}
1360 		list_for_each_entry(desc, &chan->desc.active, node) {
1361 			if (cookie == desc->async_tx.cookie)
1362 				return desc->size;
1363 		}
1364 
1365 		/*
1366 		 * No descriptor found for the cookie, there's thus no residue.
1367 		 * This shouldn't happen if the calling driver passes a correct
1368 		 * cookie value.
1369 		 */
1370 		WARN(1, "No descriptor for cookie!");
1371 		return 0;
1372 	}
1373 
1374 	/*
1375 	 * We need to read two registers.
1376 	 * Make sure the control register does not skip to next chunk
1377 	 * while reading the counter.
1378 	 * Trying it 3 times should be enough: Initial read, retry, retry
1379 	 * for the paranoid.
1380 	 */
1381 	for (i = 0; i < 3; i++) {
1382 		chcrb = rcar_dmac_chan_read(chan, RCAR_DMACHCRB) &
1383 					    RCAR_DMACHCRB_DPTR_MASK;
1384 		tcrb = rcar_dmac_chan_read(chan, RCAR_DMATCRB);
1385 		/* Still the same? */
1386 		if (chcrb == (rcar_dmac_chan_read(chan, RCAR_DMACHCRB) &
1387 			      RCAR_DMACHCRB_DPTR_MASK))
1388 			break;
1389 	}
1390 	WARN_ONCE(i >= 3, "residue might be not continuous!");
1391 
1392 	/*
1393 	 * In descriptor mode the descriptor running pointer is not maintained
1394 	 * by the interrupt handler, find the running descriptor from the
1395 	 * descriptor pointer field in the CHCRB register. In non-descriptor
1396 	 * mode just use the running descriptor pointer.
1397 	 */
1398 	if (desc->hwdescs.use) {
1399 		dptr = chcrb >> RCAR_DMACHCRB_DPTR_SHIFT;
1400 		if (dptr == 0)
1401 			dptr = desc->nchunks;
1402 		dptr--;
1403 		WARN_ON(dptr >= desc->nchunks);
1404 	} else {
1405 		running = desc->running;
1406 	}
1407 
1408 	/* Compute the size of all chunks still to be transferred. */
1409 	list_for_each_entry_reverse(chunk, &desc->chunks, node) {
1410 		if (chunk == running || ++dptr == desc->nchunks)
1411 			break;
1412 
1413 		residue += chunk->size;
1414 	}
1415 
1416 	/* Add the residue for the current chunk. */
1417 	residue += tcrb << desc->xfer_shift;
1418 
1419 	return residue;
1420 }
1421 
1422 static enum dma_status rcar_dmac_tx_status(struct dma_chan *chan,
1423 					   dma_cookie_t cookie,
1424 					   struct dma_tx_state *txstate)
1425 {
1426 	struct rcar_dmac_chan *rchan = to_rcar_dmac_chan(chan);
1427 	enum dma_status status;
1428 	unsigned long flags;
1429 	unsigned int residue;
1430 	bool cyclic;
1431 
1432 	status = dma_cookie_status(chan, cookie, txstate);
1433 	if (status == DMA_COMPLETE || !txstate)
1434 		return status;
1435 
1436 	spin_lock_irqsave(&rchan->lock, flags);
1437 	residue = rcar_dmac_chan_get_residue(rchan, cookie);
1438 	cyclic = rchan->desc.running ? rchan->desc.running->cyclic : false;
1439 	spin_unlock_irqrestore(&rchan->lock, flags);
1440 
1441 	/* if there's no residue, the cookie is complete */
1442 	if (!residue && !cyclic)
1443 		return DMA_COMPLETE;
1444 
1445 	dma_set_residue(txstate, residue);
1446 
1447 	return status;
1448 }
1449 
1450 static void rcar_dmac_issue_pending(struct dma_chan *chan)
1451 {
1452 	struct rcar_dmac_chan *rchan = to_rcar_dmac_chan(chan);
1453 	unsigned long flags;
1454 
1455 	spin_lock_irqsave(&rchan->lock, flags);
1456 
1457 	if (list_empty(&rchan->desc.pending))
1458 		goto done;
1459 
1460 	/* Append the pending list to the active list. */
1461 	list_splice_tail_init(&rchan->desc.pending, &rchan->desc.active);
1462 
1463 	/*
1464 	 * If no transfer is running pick the first descriptor from the active
1465 	 * list and start the transfer.
1466 	 */
1467 	if (!rchan->desc.running) {
1468 		struct rcar_dmac_desc *desc;
1469 
1470 		desc = list_first_entry(&rchan->desc.active,
1471 					struct rcar_dmac_desc, node);
1472 		rchan->desc.running = desc;
1473 
1474 		rcar_dmac_chan_start_xfer(rchan);
1475 	}
1476 
1477 done:
1478 	spin_unlock_irqrestore(&rchan->lock, flags);
1479 }
1480 
1481 static void rcar_dmac_device_synchronize(struct dma_chan *chan)
1482 {
1483 	struct rcar_dmac_chan *rchan = to_rcar_dmac_chan(chan);
1484 
1485 	synchronize_irq(rchan->irq);
1486 }
1487 
1488 /* -----------------------------------------------------------------------------
1489  * IRQ handling
1490  */
1491 
1492 static irqreturn_t rcar_dmac_isr_desc_stage_end(struct rcar_dmac_chan *chan)
1493 {
1494 	struct rcar_dmac_desc *desc = chan->desc.running;
1495 	unsigned int stage;
1496 
1497 	if (WARN_ON(!desc || !desc->cyclic)) {
1498 		/*
1499 		 * This should never happen, there should always be a running
1500 		 * cyclic descriptor when a descriptor stage end interrupt is
1501 		 * triggered. Warn and return.
1502 		 */
1503 		return IRQ_NONE;
1504 	}
1505 
1506 	/* Program the interrupt pointer to the next stage. */
1507 	stage = (rcar_dmac_chan_read(chan, RCAR_DMACHCRB) &
1508 		 RCAR_DMACHCRB_DPTR_MASK) >> RCAR_DMACHCRB_DPTR_SHIFT;
1509 	rcar_dmac_chan_write(chan, RCAR_DMADPCR, RCAR_DMADPCR_DIPT(stage));
1510 
1511 	return IRQ_WAKE_THREAD;
1512 }
1513 
1514 static irqreturn_t rcar_dmac_isr_transfer_end(struct rcar_dmac_chan *chan)
1515 {
1516 	struct rcar_dmac_desc *desc = chan->desc.running;
1517 	irqreturn_t ret = IRQ_WAKE_THREAD;
1518 
1519 	if (WARN_ON_ONCE(!desc)) {
1520 		/*
1521 		 * This should never happen, there should always be a running
1522 		 * descriptor when a transfer end interrupt is triggered. Warn
1523 		 * and return.
1524 		 */
1525 		return IRQ_NONE;
1526 	}
1527 
1528 	/*
1529 	 * The transfer end interrupt isn't generated for each chunk when using
1530 	 * descriptor mode. Only update the running chunk pointer in
1531 	 * non-descriptor mode.
1532 	 */
1533 	if (!desc->hwdescs.use) {
1534 		/*
1535 		 * If we haven't completed the last transfer chunk simply move
1536 		 * to the next one. Only wake the IRQ thread if the transfer is
1537 		 * cyclic.
1538 		 */
1539 		if (!list_is_last(&desc->running->node, &desc->chunks)) {
1540 			desc->running = list_next_entry(desc->running, node);
1541 			if (!desc->cyclic)
1542 				ret = IRQ_HANDLED;
1543 			goto done;
1544 		}
1545 
1546 		/*
1547 		 * We've completed the last transfer chunk. If the transfer is
1548 		 * cyclic, move back to the first one.
1549 		 */
1550 		if (desc->cyclic) {
1551 			desc->running =
1552 				list_first_entry(&desc->chunks,
1553 						 struct rcar_dmac_xfer_chunk,
1554 						 node);
1555 			goto done;
1556 		}
1557 	}
1558 
1559 	/* The descriptor is complete, move it to the done list. */
1560 	list_move_tail(&desc->node, &chan->desc.done);
1561 
1562 	/* Queue the next descriptor, if any. */
1563 	if (!list_empty(&chan->desc.active))
1564 		chan->desc.running = list_first_entry(&chan->desc.active,
1565 						      struct rcar_dmac_desc,
1566 						      node);
1567 	else
1568 		chan->desc.running = NULL;
1569 
1570 done:
1571 	if (chan->desc.running)
1572 		rcar_dmac_chan_start_xfer(chan);
1573 
1574 	return ret;
1575 }
1576 
1577 static irqreturn_t rcar_dmac_isr_channel(int irq, void *dev)
1578 {
1579 	u32 mask = RCAR_DMACHCR_DSE | RCAR_DMACHCR_TE;
1580 	struct rcar_dmac_chan *chan = dev;
1581 	irqreturn_t ret = IRQ_NONE;
1582 	bool reinit = false;
1583 	u32 chcr;
1584 
1585 	spin_lock(&chan->lock);
1586 
1587 	chcr = rcar_dmac_chan_read(chan, RCAR_DMACHCR);
1588 	if (chcr & RCAR_DMACHCR_CAE) {
1589 		struct rcar_dmac *dmac = to_rcar_dmac(chan->chan.device);
1590 
1591 		/*
1592 		 * We don't need to call rcar_dmac_chan_halt()
1593 		 * because channel is already stopped in error case.
1594 		 * We need to clear register and check DE bit as recovery.
1595 		 */
1596 		rcar_dmac_chan_clear(dmac, chan);
1597 		rcar_dmac_chcr_de_barrier(chan);
1598 		reinit = true;
1599 		goto spin_lock_end;
1600 	}
1601 
1602 	if (chcr & RCAR_DMACHCR_TE)
1603 		mask |= RCAR_DMACHCR_DE;
1604 	rcar_dmac_chan_write(chan, RCAR_DMACHCR, chcr & ~mask);
1605 	if (mask & RCAR_DMACHCR_DE)
1606 		rcar_dmac_chcr_de_barrier(chan);
1607 
1608 	if (chcr & RCAR_DMACHCR_DSE)
1609 		ret |= rcar_dmac_isr_desc_stage_end(chan);
1610 
1611 	if (chcr & RCAR_DMACHCR_TE)
1612 		ret |= rcar_dmac_isr_transfer_end(chan);
1613 
1614 spin_lock_end:
1615 	spin_unlock(&chan->lock);
1616 
1617 	if (reinit) {
1618 		dev_err(chan->chan.device->dev, "Channel Address Error\n");
1619 
1620 		rcar_dmac_chan_reinit(chan);
1621 		ret = IRQ_HANDLED;
1622 	}
1623 
1624 	return ret;
1625 }
1626 
1627 static irqreturn_t rcar_dmac_isr_channel_thread(int irq, void *dev)
1628 {
1629 	struct rcar_dmac_chan *chan = dev;
1630 	struct rcar_dmac_desc *desc;
1631 	struct dmaengine_desc_callback cb;
1632 
1633 	spin_lock_irq(&chan->lock);
1634 
1635 	/* For cyclic transfers notify the user after every chunk. */
1636 	if (chan->desc.running && chan->desc.running->cyclic) {
1637 		desc = chan->desc.running;
1638 		dmaengine_desc_get_callback(&desc->async_tx, &cb);
1639 
1640 		if (dmaengine_desc_callback_valid(&cb)) {
1641 			spin_unlock_irq(&chan->lock);
1642 			dmaengine_desc_callback_invoke(&cb, NULL);
1643 			spin_lock_irq(&chan->lock);
1644 		}
1645 	}
1646 
1647 	/*
1648 	 * Call the callback function for all descriptors on the done list and
1649 	 * move them to the ack wait list.
1650 	 */
1651 	while (!list_empty(&chan->desc.done)) {
1652 		desc = list_first_entry(&chan->desc.done, struct rcar_dmac_desc,
1653 					node);
1654 		dma_cookie_complete(&desc->async_tx);
1655 		list_del(&desc->node);
1656 
1657 		dmaengine_desc_get_callback(&desc->async_tx, &cb);
1658 		if (dmaengine_desc_callback_valid(&cb)) {
1659 			spin_unlock_irq(&chan->lock);
1660 			/*
1661 			 * We own the only reference to this descriptor, we can
1662 			 * safely dereference it without holding the channel
1663 			 * lock.
1664 			 */
1665 			dmaengine_desc_callback_invoke(&cb, NULL);
1666 			spin_lock_irq(&chan->lock);
1667 		}
1668 
1669 		list_add_tail(&desc->node, &chan->desc.wait);
1670 	}
1671 
1672 	spin_unlock_irq(&chan->lock);
1673 
1674 	/* Recycle all acked descriptors. */
1675 	rcar_dmac_desc_recycle_acked(chan);
1676 
1677 	return IRQ_HANDLED;
1678 }
1679 
1680 /* -----------------------------------------------------------------------------
1681  * OF xlate and channel filter
1682  */
1683 
1684 static bool rcar_dmac_chan_filter(struct dma_chan *chan, void *arg)
1685 {
1686 	struct rcar_dmac *dmac = to_rcar_dmac(chan->device);
1687 	struct of_phandle_args *dma_spec = arg;
1688 
1689 	/*
1690 	 * FIXME: Using a filter on OF platforms is a nonsense. The OF xlate
1691 	 * function knows from which device it wants to allocate a channel from,
1692 	 * and would be perfectly capable of selecting the channel it wants.
1693 	 * Forcing it to call dma_request_channel() and iterate through all
1694 	 * channels from all controllers is just pointless.
1695 	 */
1696 	if (chan->device->device_config != rcar_dmac_device_config)
1697 		return false;
1698 
1699 	return !test_and_set_bit(dma_spec->args[0], dmac->modules);
1700 }
1701 
1702 static struct dma_chan *rcar_dmac_of_xlate(struct of_phandle_args *dma_spec,
1703 					   struct of_dma *ofdma)
1704 {
1705 	struct rcar_dmac_chan *rchan;
1706 	struct dma_chan *chan;
1707 	dma_cap_mask_t mask;
1708 
1709 	if (dma_spec->args_count != 1)
1710 		return NULL;
1711 
1712 	/* Only slave DMA channels can be allocated via DT */
1713 	dma_cap_zero(mask);
1714 	dma_cap_set(DMA_SLAVE, mask);
1715 
1716 	chan = __dma_request_channel(&mask, rcar_dmac_chan_filter, dma_spec,
1717 				     ofdma->of_node);
1718 	if (!chan)
1719 		return NULL;
1720 
1721 	rchan = to_rcar_dmac_chan(chan);
1722 	rchan->mid_rid = dma_spec->args[0];
1723 
1724 	return chan;
1725 }
1726 
1727 /* -----------------------------------------------------------------------------
1728  * Power management
1729  */
1730 
1731 #ifdef CONFIG_PM
1732 static int rcar_dmac_runtime_suspend(struct device *dev)
1733 {
1734 	return 0;
1735 }
1736 
1737 static int rcar_dmac_runtime_resume(struct device *dev)
1738 {
1739 	struct rcar_dmac *dmac = dev_get_drvdata(dev);
1740 
1741 	return rcar_dmac_init(dmac);
1742 }
1743 #endif
1744 
1745 static const struct dev_pm_ops rcar_dmac_pm = {
1746 	/*
1747 	 * TODO for system sleep/resume:
1748 	 *   - Wait for the current transfer to complete and stop the device,
1749 	 *   - Resume transfers, if any.
1750 	 */
1751 	SET_NOIRQ_SYSTEM_SLEEP_PM_OPS(pm_runtime_force_suspend,
1752 				      pm_runtime_force_resume)
1753 	SET_RUNTIME_PM_OPS(rcar_dmac_runtime_suspend, rcar_dmac_runtime_resume,
1754 			   NULL)
1755 };
1756 
1757 /* -----------------------------------------------------------------------------
1758  * Probe and remove
1759  */
1760 
1761 static int rcar_dmac_chan_probe(struct rcar_dmac *dmac,
1762 				struct rcar_dmac_chan *rchan)
1763 {
1764 	struct platform_device *pdev = to_platform_device(dmac->dev);
1765 	struct dma_chan *chan = &rchan->chan;
1766 	char pdev_irqname[5];
1767 	char *irqname;
1768 	int ret;
1769 
1770 	rchan->mid_rid = -EINVAL;
1771 
1772 	spin_lock_init(&rchan->lock);
1773 
1774 	INIT_LIST_HEAD(&rchan->desc.free);
1775 	INIT_LIST_HEAD(&rchan->desc.pending);
1776 	INIT_LIST_HEAD(&rchan->desc.active);
1777 	INIT_LIST_HEAD(&rchan->desc.done);
1778 	INIT_LIST_HEAD(&rchan->desc.wait);
1779 
1780 	/* Request the channel interrupt. */
1781 	sprintf(pdev_irqname, "ch%u", rchan->index);
1782 	rchan->irq = platform_get_irq_byname(pdev, pdev_irqname);
1783 	if (rchan->irq < 0)
1784 		return -ENODEV;
1785 
1786 	irqname = devm_kasprintf(dmac->dev, GFP_KERNEL, "%s:%u",
1787 				 dev_name(dmac->dev), rchan->index);
1788 	if (!irqname)
1789 		return -ENOMEM;
1790 
1791 	/*
1792 	 * Initialize the DMA engine channel and add it to the DMA engine
1793 	 * channels list.
1794 	 */
1795 	chan->device = &dmac->engine;
1796 	dma_cookie_init(chan);
1797 
1798 	list_add_tail(&chan->device_node, &dmac->engine.channels);
1799 
1800 	ret = devm_request_threaded_irq(dmac->dev, rchan->irq,
1801 					rcar_dmac_isr_channel,
1802 					rcar_dmac_isr_channel_thread, 0,
1803 					irqname, rchan);
1804 	if (ret) {
1805 		dev_err(dmac->dev, "failed to request IRQ %u (%d)\n",
1806 			rchan->irq, ret);
1807 		return ret;
1808 	}
1809 
1810 	return 0;
1811 }
1812 
1813 #define RCAR_DMAC_MAX_CHANNELS	32
1814 
1815 static int rcar_dmac_parse_of(struct device *dev, struct rcar_dmac *dmac)
1816 {
1817 	struct device_node *np = dev->of_node;
1818 	int ret;
1819 
1820 	ret = of_property_read_u32(np, "dma-channels", &dmac->n_channels);
1821 	if (ret < 0) {
1822 		dev_err(dev, "unable to read dma-channels property\n");
1823 		return ret;
1824 	}
1825 
1826 	/* The hardware and driver don't support more than 32 bits in CHCLR */
1827 	if (dmac->n_channels <= 0 ||
1828 	    dmac->n_channels >= RCAR_DMAC_MAX_CHANNELS) {
1829 		dev_err(dev, "invalid number of channels %u\n",
1830 			dmac->n_channels);
1831 		return -EINVAL;
1832 	}
1833 
1834 	/*
1835 	 * If the driver is unable to read dma-channel-mask property,
1836 	 * the driver assumes that it can use all channels.
1837 	 */
1838 	dmac->channels_mask = GENMASK(dmac->n_channels - 1, 0);
1839 	of_property_read_u32(np, "dma-channel-mask", &dmac->channels_mask);
1840 
1841 	/* If the property has out-of-channel mask, this driver clears it */
1842 	dmac->channels_mask &= GENMASK(dmac->n_channels - 1, 0);
1843 
1844 	return 0;
1845 }
1846 
1847 static int rcar_dmac_probe(struct platform_device *pdev)
1848 {
1849 	const enum dma_slave_buswidth widths = DMA_SLAVE_BUSWIDTH_1_BYTE |
1850 		DMA_SLAVE_BUSWIDTH_2_BYTES | DMA_SLAVE_BUSWIDTH_4_BYTES |
1851 		DMA_SLAVE_BUSWIDTH_8_BYTES | DMA_SLAVE_BUSWIDTH_16_BYTES |
1852 		DMA_SLAVE_BUSWIDTH_32_BYTES | DMA_SLAVE_BUSWIDTH_64_BYTES;
1853 	const struct rcar_dmac_of_data *data;
1854 	struct rcar_dmac_chan *chan;
1855 	struct dma_device *engine;
1856 	void __iomem *chan_base;
1857 	struct rcar_dmac *dmac;
1858 	unsigned int i;
1859 	int ret;
1860 
1861 	data = of_device_get_match_data(&pdev->dev);
1862 	if (!data)
1863 		return -EINVAL;
1864 
1865 	dmac = devm_kzalloc(&pdev->dev, sizeof(*dmac), GFP_KERNEL);
1866 	if (!dmac)
1867 		return -ENOMEM;
1868 
1869 	dmac->dev = &pdev->dev;
1870 	platform_set_drvdata(pdev, dmac);
1871 	dma_set_max_seg_size(dmac->dev, RCAR_DMATCR_MASK);
1872 	dma_set_mask_and_coherent(dmac->dev, DMA_BIT_MASK(40));
1873 
1874 	ret = rcar_dmac_parse_of(&pdev->dev, dmac);
1875 	if (ret < 0)
1876 		return ret;
1877 
1878 	/*
1879 	 * A still unconfirmed hardware bug prevents the IPMMU microTLB 0 to be
1880 	 * flushed correctly, resulting in memory corruption. DMAC 0 channel 0
1881 	 * is connected to microTLB 0 on currently supported platforms, so we
1882 	 * can't use it with the IPMMU. As the IOMMU API operates at the device
1883 	 * level we can't disable it selectively, so ignore channel 0 for now if
1884 	 * the device is part of an IOMMU group.
1885 	 */
1886 	if (device_iommu_mapped(&pdev->dev))
1887 		dmac->channels_mask &= ~BIT(0);
1888 
1889 	dmac->channels = devm_kcalloc(&pdev->dev, dmac->n_channels,
1890 				      sizeof(*dmac->channels), GFP_KERNEL);
1891 	if (!dmac->channels)
1892 		return -ENOMEM;
1893 
1894 	/* Request resources. */
1895 	dmac->dmac_base = devm_platform_ioremap_resource(pdev, 0);
1896 	if (IS_ERR(dmac->dmac_base))
1897 		return PTR_ERR(dmac->dmac_base);
1898 
1899 	if (!data->chan_offset_base) {
1900 		dmac->chan_base = devm_platform_ioremap_resource(pdev, 1);
1901 		if (IS_ERR(dmac->chan_base))
1902 			return PTR_ERR(dmac->chan_base);
1903 
1904 		chan_base = dmac->chan_base;
1905 	} else {
1906 		chan_base = dmac->dmac_base + data->chan_offset_base;
1907 	}
1908 
1909 	for_each_rcar_dmac_chan(i, dmac, chan) {
1910 		chan->index = i;
1911 		chan->iomem = chan_base + i * data->chan_offset_stride;
1912 	}
1913 
1914 	/* Enable runtime PM and initialize the device. */
1915 	pm_runtime_enable(&pdev->dev);
1916 	ret = pm_runtime_get_sync(&pdev->dev);
1917 	if (ret < 0) {
1918 		dev_err(&pdev->dev, "runtime PM get sync failed (%d)\n", ret);
1919 		return ret;
1920 	}
1921 
1922 	ret = rcar_dmac_init(dmac);
1923 	pm_runtime_put(&pdev->dev);
1924 
1925 	if (ret) {
1926 		dev_err(&pdev->dev, "failed to reset device\n");
1927 		goto error;
1928 	}
1929 
1930 	/* Initialize engine */
1931 	engine = &dmac->engine;
1932 
1933 	dma_cap_set(DMA_MEMCPY, engine->cap_mask);
1934 	dma_cap_set(DMA_SLAVE, engine->cap_mask);
1935 
1936 	engine->dev		= &pdev->dev;
1937 	engine->copy_align	= ilog2(RCAR_DMAC_MEMCPY_XFER_SIZE);
1938 
1939 	engine->src_addr_widths	= widths;
1940 	engine->dst_addr_widths	= widths;
1941 	engine->directions	= BIT(DMA_MEM_TO_DEV) | BIT(DMA_DEV_TO_MEM);
1942 	engine->residue_granularity = DMA_RESIDUE_GRANULARITY_BURST;
1943 
1944 	engine->device_alloc_chan_resources	= rcar_dmac_alloc_chan_resources;
1945 	engine->device_free_chan_resources	= rcar_dmac_free_chan_resources;
1946 	engine->device_prep_dma_memcpy		= rcar_dmac_prep_dma_memcpy;
1947 	engine->device_prep_slave_sg		= rcar_dmac_prep_slave_sg;
1948 	engine->device_prep_dma_cyclic		= rcar_dmac_prep_dma_cyclic;
1949 	engine->device_config			= rcar_dmac_device_config;
1950 	engine->device_pause			= rcar_dmac_chan_pause;
1951 	engine->device_terminate_all		= rcar_dmac_chan_terminate_all;
1952 	engine->device_tx_status		= rcar_dmac_tx_status;
1953 	engine->device_issue_pending		= rcar_dmac_issue_pending;
1954 	engine->device_synchronize		= rcar_dmac_device_synchronize;
1955 
1956 	INIT_LIST_HEAD(&engine->channels);
1957 
1958 	for_each_rcar_dmac_chan(i, dmac, chan) {
1959 		ret = rcar_dmac_chan_probe(dmac, chan);
1960 		if (ret < 0)
1961 			goto error;
1962 	}
1963 
1964 	/* Register the DMAC as a DMA provider for DT. */
1965 	ret = of_dma_controller_register(pdev->dev.of_node, rcar_dmac_of_xlate,
1966 					 NULL);
1967 	if (ret < 0)
1968 		goto error;
1969 
1970 	/*
1971 	 * Register the DMA engine device.
1972 	 *
1973 	 * Default transfer size of 32 bytes requires 32-byte alignment.
1974 	 */
1975 	ret = dma_async_device_register(engine);
1976 	if (ret < 0)
1977 		goto error;
1978 
1979 	return 0;
1980 
1981 error:
1982 	of_dma_controller_free(pdev->dev.of_node);
1983 	pm_runtime_disable(&pdev->dev);
1984 	return ret;
1985 }
1986 
1987 static int rcar_dmac_remove(struct platform_device *pdev)
1988 {
1989 	struct rcar_dmac *dmac = platform_get_drvdata(pdev);
1990 
1991 	of_dma_controller_free(pdev->dev.of_node);
1992 	dma_async_device_unregister(&dmac->engine);
1993 
1994 	pm_runtime_disable(&pdev->dev);
1995 
1996 	return 0;
1997 }
1998 
1999 static void rcar_dmac_shutdown(struct platform_device *pdev)
2000 {
2001 	struct rcar_dmac *dmac = platform_get_drvdata(pdev);
2002 
2003 	rcar_dmac_stop_all_chan(dmac);
2004 }
2005 
2006 static const struct rcar_dmac_of_data rcar_dmac_data = {
2007 	.chan_offset_base	= 0x8000,
2008 	.chan_offset_stride	= 0x80,
2009 };
2010 
2011 static const struct rcar_dmac_of_data rcar_v3u_dmac_data = {
2012 	.chan_offset_base	= 0x0,
2013 	.chan_offset_stride	= 0x1000,
2014 };
2015 
2016 static const struct of_device_id rcar_dmac_of_ids[] = {
2017 	{
2018 		.compatible = "renesas,rcar-dmac",
2019 		.data = &rcar_dmac_data,
2020 	}, {
2021 		.compatible = "renesas,dmac-r8a779a0",
2022 		.data = &rcar_v3u_dmac_data,
2023 	},
2024 	{ /* Sentinel */ }
2025 };
2026 MODULE_DEVICE_TABLE(of, rcar_dmac_of_ids);
2027 
2028 static struct platform_driver rcar_dmac_driver = {
2029 	.driver		= {
2030 		.pm	= &rcar_dmac_pm,
2031 		.name	= "rcar-dmac",
2032 		.of_match_table = rcar_dmac_of_ids,
2033 	},
2034 	.probe		= rcar_dmac_probe,
2035 	.remove		= rcar_dmac_remove,
2036 	.shutdown	= rcar_dmac_shutdown,
2037 };
2038 
2039 module_platform_driver(rcar_dmac_driver);
2040 
2041 MODULE_DESCRIPTION("R-Car Gen2 DMA Controller Driver");
2042 MODULE_AUTHOR("Laurent Pinchart <laurent.pinchart@ideasonboard.com>");
2043 MODULE_LICENSE("GPL v2");
2044