xref: /openbmc/linux/drivers/dma/sh/rcar-dmac.c (revision 11a163f2)
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  * @iomem: remapped I/O memory base
193  * @n_channels: number of available channels
194  * @channels: array of DMAC channels
195  * @channels_mask: bitfield of which DMA channels are managed by this driver
196  * @modules: bitmask of client modules in use
197  */
198 struct rcar_dmac {
199 	struct dma_device engine;
200 	struct device *dev;
201 	void __iomem *iomem;
202 
203 	unsigned int n_channels;
204 	struct rcar_dmac_chan *channels;
205 	u32 channels_mask;
206 
207 	DECLARE_BITMAP(modules, 256);
208 };
209 
210 #define to_rcar_dmac(d)		container_of(d, struct rcar_dmac, engine)
211 
212 /*
213  * struct rcar_dmac_of_data - This driver's OF data
214  * @chan_offset_base: DMAC channels base offset
215  * @chan_offset_stride: DMAC channels offset stride
216  */
217 struct rcar_dmac_of_data {
218 	u32 chan_offset_base;
219 	u32 chan_offset_stride;
220 };
221 
222 /* -----------------------------------------------------------------------------
223  * Registers
224  */
225 
226 #define RCAR_DMAISTA			0x0020
227 #define RCAR_DMASEC			0x0030
228 #define RCAR_DMAOR			0x0060
229 #define RCAR_DMAOR_PRI_FIXED		(0 << 8)
230 #define RCAR_DMAOR_PRI_ROUND_ROBIN	(3 << 8)
231 #define RCAR_DMAOR_AE			(1 << 2)
232 #define RCAR_DMAOR_DME			(1 << 0)
233 #define RCAR_DMACHCLR			0x0080
234 #define RCAR_DMADPSEC			0x00a0
235 
236 #define RCAR_DMASAR			0x0000
237 #define RCAR_DMADAR			0x0004
238 #define RCAR_DMATCR			0x0008
239 #define RCAR_DMATCR_MASK		0x00ffffff
240 #define RCAR_DMATSR			0x0028
241 #define RCAR_DMACHCR			0x000c
242 #define RCAR_DMACHCR_CAE		(1 << 31)
243 #define RCAR_DMACHCR_CAIE		(1 << 30)
244 #define RCAR_DMACHCR_DPM_DISABLED	(0 << 28)
245 #define RCAR_DMACHCR_DPM_ENABLED	(1 << 28)
246 #define RCAR_DMACHCR_DPM_REPEAT		(2 << 28)
247 #define RCAR_DMACHCR_DPM_INFINITE	(3 << 28)
248 #define RCAR_DMACHCR_RPT_SAR		(1 << 27)
249 #define RCAR_DMACHCR_RPT_DAR		(1 << 26)
250 #define RCAR_DMACHCR_RPT_TCR		(1 << 25)
251 #define RCAR_DMACHCR_DPB		(1 << 22)
252 #define RCAR_DMACHCR_DSE		(1 << 19)
253 #define RCAR_DMACHCR_DSIE		(1 << 18)
254 #define RCAR_DMACHCR_TS_1B		((0 << 20) | (0 << 3))
255 #define RCAR_DMACHCR_TS_2B		((0 << 20) | (1 << 3))
256 #define RCAR_DMACHCR_TS_4B		((0 << 20) | (2 << 3))
257 #define RCAR_DMACHCR_TS_16B		((0 << 20) | (3 << 3))
258 #define RCAR_DMACHCR_TS_32B		((1 << 20) | (0 << 3))
259 #define RCAR_DMACHCR_TS_64B		((1 << 20) | (1 << 3))
260 #define RCAR_DMACHCR_TS_8B		((1 << 20) | (3 << 3))
261 #define RCAR_DMACHCR_DM_FIXED		(0 << 14)
262 #define RCAR_DMACHCR_DM_INC		(1 << 14)
263 #define RCAR_DMACHCR_DM_DEC		(2 << 14)
264 #define RCAR_DMACHCR_SM_FIXED		(0 << 12)
265 #define RCAR_DMACHCR_SM_INC		(1 << 12)
266 #define RCAR_DMACHCR_SM_DEC		(2 << 12)
267 #define RCAR_DMACHCR_RS_AUTO		(4 << 8)
268 #define RCAR_DMACHCR_RS_DMARS		(8 << 8)
269 #define RCAR_DMACHCR_IE			(1 << 2)
270 #define RCAR_DMACHCR_TE			(1 << 1)
271 #define RCAR_DMACHCR_DE			(1 << 0)
272 #define RCAR_DMATCRB			0x0018
273 #define RCAR_DMATSRB			0x0038
274 #define RCAR_DMACHCRB			0x001c
275 #define RCAR_DMACHCRB_DCNT(n)		((n) << 24)
276 #define RCAR_DMACHCRB_DPTR_MASK		(0xff << 16)
277 #define RCAR_DMACHCRB_DPTR_SHIFT	16
278 #define RCAR_DMACHCRB_DRST		(1 << 15)
279 #define RCAR_DMACHCRB_DTS		(1 << 8)
280 #define RCAR_DMACHCRB_SLM_NORMAL	(0 << 4)
281 #define RCAR_DMACHCRB_SLM_CLK(n)	((8 | (n)) << 4)
282 #define RCAR_DMACHCRB_PRI(n)		((n) << 0)
283 #define RCAR_DMARS			0x0040
284 #define RCAR_DMABUFCR			0x0048
285 #define RCAR_DMABUFCR_MBU(n)		((n) << 16)
286 #define RCAR_DMABUFCR_ULB(n)		((n) << 0)
287 #define RCAR_DMADPBASE			0x0050
288 #define RCAR_DMADPBASE_MASK		0xfffffff0
289 #define RCAR_DMADPBASE_SEL		(1 << 0)
290 #define RCAR_DMADPCR			0x0054
291 #define RCAR_DMADPCR_DIPT(n)		((n) << 24)
292 #define RCAR_DMAFIXSAR			0x0010
293 #define RCAR_DMAFIXDAR			0x0014
294 #define RCAR_DMAFIXDPBASE		0x0060
295 
296 /* Hardcode the MEMCPY transfer size to 4 bytes. */
297 #define RCAR_DMAC_MEMCPY_XFER_SIZE	4
298 
299 /* -----------------------------------------------------------------------------
300  * Device access
301  */
302 
303 static void rcar_dmac_write(struct rcar_dmac *dmac, u32 reg, u32 data)
304 {
305 	if (reg == RCAR_DMAOR)
306 		writew(data, dmac->iomem + reg);
307 	else
308 		writel(data, dmac->iomem + reg);
309 }
310 
311 static u32 rcar_dmac_read(struct rcar_dmac *dmac, u32 reg)
312 {
313 	if (reg == RCAR_DMAOR)
314 		return readw(dmac->iomem + reg);
315 	else
316 		return readl(dmac->iomem + reg);
317 }
318 
319 static u32 rcar_dmac_chan_read(struct rcar_dmac_chan *chan, u32 reg)
320 {
321 	if (reg == RCAR_DMARS)
322 		return readw(chan->iomem + reg);
323 	else
324 		return readl(chan->iomem + reg);
325 }
326 
327 static void rcar_dmac_chan_write(struct rcar_dmac_chan *chan, u32 reg, u32 data)
328 {
329 	if (reg == RCAR_DMARS)
330 		writew(data, chan->iomem + reg);
331 	else
332 		writel(data, chan->iomem + reg);
333 }
334 
335 /* -----------------------------------------------------------------------------
336  * Initialization and configuration
337  */
338 
339 static bool rcar_dmac_chan_is_busy(struct rcar_dmac_chan *chan)
340 {
341 	u32 chcr = rcar_dmac_chan_read(chan, RCAR_DMACHCR);
342 
343 	return !!(chcr & (RCAR_DMACHCR_DE | RCAR_DMACHCR_TE));
344 }
345 
346 static void rcar_dmac_chan_start_xfer(struct rcar_dmac_chan *chan)
347 {
348 	struct rcar_dmac_desc *desc = chan->desc.running;
349 	u32 chcr = desc->chcr;
350 
351 	WARN_ON_ONCE(rcar_dmac_chan_is_busy(chan));
352 
353 	if (chan->mid_rid >= 0)
354 		rcar_dmac_chan_write(chan, RCAR_DMARS, chan->mid_rid);
355 
356 	if (desc->hwdescs.use) {
357 		struct rcar_dmac_xfer_chunk *chunk =
358 			list_first_entry(&desc->chunks,
359 					 struct rcar_dmac_xfer_chunk, node);
360 
361 		dev_dbg(chan->chan.device->dev,
362 			"chan%u: queue desc %p: %u@%pad\n",
363 			chan->index, desc, desc->nchunks, &desc->hwdescs.dma);
364 
365 #ifdef CONFIG_ARCH_DMA_ADDR_T_64BIT
366 		rcar_dmac_chan_write(chan, RCAR_DMAFIXSAR,
367 				     chunk->src_addr >> 32);
368 		rcar_dmac_chan_write(chan, RCAR_DMAFIXDAR,
369 				     chunk->dst_addr >> 32);
370 		rcar_dmac_chan_write(chan, RCAR_DMAFIXDPBASE,
371 				     desc->hwdescs.dma >> 32);
372 #endif
373 		rcar_dmac_chan_write(chan, RCAR_DMADPBASE,
374 				     (desc->hwdescs.dma & 0xfffffff0) |
375 				     RCAR_DMADPBASE_SEL);
376 		rcar_dmac_chan_write(chan, RCAR_DMACHCRB,
377 				     RCAR_DMACHCRB_DCNT(desc->nchunks - 1) |
378 				     RCAR_DMACHCRB_DRST);
379 
380 		/*
381 		 * Errata: When descriptor memory is accessed through an IOMMU
382 		 * the DMADAR register isn't initialized automatically from the
383 		 * first descriptor at beginning of transfer by the DMAC like it
384 		 * should. Initialize it manually with the destination address
385 		 * of the first chunk.
386 		 */
387 		rcar_dmac_chan_write(chan, RCAR_DMADAR,
388 				     chunk->dst_addr & 0xffffffff);
389 
390 		/*
391 		 * Program the descriptor stage interrupt to occur after the end
392 		 * of the first stage.
393 		 */
394 		rcar_dmac_chan_write(chan, RCAR_DMADPCR, RCAR_DMADPCR_DIPT(1));
395 
396 		chcr |= RCAR_DMACHCR_RPT_SAR | RCAR_DMACHCR_RPT_DAR
397 		     |  RCAR_DMACHCR_RPT_TCR | RCAR_DMACHCR_DPB;
398 
399 		/*
400 		 * If the descriptor isn't cyclic enable normal descriptor mode
401 		 * and the transfer completion interrupt.
402 		 */
403 		if (!desc->cyclic)
404 			chcr |= RCAR_DMACHCR_DPM_ENABLED | RCAR_DMACHCR_IE;
405 		/*
406 		 * If the descriptor is cyclic and has a callback enable the
407 		 * descriptor stage interrupt in infinite repeat mode.
408 		 */
409 		else if (desc->async_tx.callback)
410 			chcr |= RCAR_DMACHCR_DPM_INFINITE | RCAR_DMACHCR_DSIE;
411 		/*
412 		 * Otherwise just select infinite repeat mode without any
413 		 * interrupt.
414 		 */
415 		else
416 			chcr |= RCAR_DMACHCR_DPM_INFINITE;
417 	} else {
418 		struct rcar_dmac_xfer_chunk *chunk = desc->running;
419 
420 		dev_dbg(chan->chan.device->dev,
421 			"chan%u: queue chunk %p: %u@%pad -> %pad\n",
422 			chan->index, chunk, chunk->size, &chunk->src_addr,
423 			&chunk->dst_addr);
424 
425 #ifdef CONFIG_ARCH_DMA_ADDR_T_64BIT
426 		rcar_dmac_chan_write(chan, RCAR_DMAFIXSAR,
427 				     chunk->src_addr >> 32);
428 		rcar_dmac_chan_write(chan, RCAR_DMAFIXDAR,
429 				     chunk->dst_addr >> 32);
430 #endif
431 		rcar_dmac_chan_write(chan, RCAR_DMASAR,
432 				     chunk->src_addr & 0xffffffff);
433 		rcar_dmac_chan_write(chan, RCAR_DMADAR,
434 				     chunk->dst_addr & 0xffffffff);
435 		rcar_dmac_chan_write(chan, RCAR_DMATCR,
436 				     chunk->size >> desc->xfer_shift);
437 
438 		chcr |= RCAR_DMACHCR_DPM_DISABLED | RCAR_DMACHCR_IE;
439 	}
440 
441 	rcar_dmac_chan_write(chan, RCAR_DMACHCR,
442 			     chcr | RCAR_DMACHCR_DE | RCAR_DMACHCR_CAIE);
443 }
444 
445 static int rcar_dmac_init(struct rcar_dmac *dmac)
446 {
447 	u16 dmaor;
448 
449 	/* Clear all channels and enable the DMAC globally. */
450 	rcar_dmac_write(dmac, RCAR_DMACHCLR, dmac->channels_mask);
451 	rcar_dmac_write(dmac, RCAR_DMAOR,
452 			RCAR_DMAOR_PRI_FIXED | RCAR_DMAOR_DME);
453 
454 	dmaor = rcar_dmac_read(dmac, RCAR_DMAOR);
455 	if ((dmaor & (RCAR_DMAOR_AE | RCAR_DMAOR_DME)) != RCAR_DMAOR_DME) {
456 		dev_warn(dmac->dev, "DMAOR initialization failed.\n");
457 		return -EIO;
458 	}
459 
460 	return 0;
461 }
462 
463 /* -----------------------------------------------------------------------------
464  * Descriptors submission
465  */
466 
467 static dma_cookie_t rcar_dmac_tx_submit(struct dma_async_tx_descriptor *tx)
468 {
469 	struct rcar_dmac_chan *chan = to_rcar_dmac_chan(tx->chan);
470 	struct rcar_dmac_desc *desc = to_rcar_dmac_desc(tx);
471 	unsigned long flags;
472 	dma_cookie_t cookie;
473 
474 	spin_lock_irqsave(&chan->lock, flags);
475 
476 	cookie = dma_cookie_assign(tx);
477 
478 	dev_dbg(chan->chan.device->dev, "chan%u: submit #%d@%p\n",
479 		chan->index, tx->cookie, desc);
480 
481 	list_add_tail(&desc->node, &chan->desc.pending);
482 	desc->running = list_first_entry(&desc->chunks,
483 					 struct rcar_dmac_xfer_chunk, node);
484 
485 	spin_unlock_irqrestore(&chan->lock, flags);
486 
487 	return cookie;
488 }
489 
490 /* -----------------------------------------------------------------------------
491  * Descriptors allocation and free
492  */
493 
494 /*
495  * rcar_dmac_desc_alloc - Allocate a page worth of DMA descriptors
496  * @chan: the DMA channel
497  * @gfp: allocation flags
498  */
499 static int rcar_dmac_desc_alloc(struct rcar_dmac_chan *chan, gfp_t gfp)
500 {
501 	struct rcar_dmac_desc_page *page;
502 	unsigned long flags;
503 	LIST_HEAD(list);
504 	unsigned int i;
505 
506 	page = (void *)get_zeroed_page(gfp);
507 	if (!page)
508 		return -ENOMEM;
509 
510 	for (i = 0; i < RCAR_DMAC_DESCS_PER_PAGE; ++i) {
511 		struct rcar_dmac_desc *desc = &page->descs[i];
512 
513 		dma_async_tx_descriptor_init(&desc->async_tx, &chan->chan);
514 		desc->async_tx.tx_submit = rcar_dmac_tx_submit;
515 		INIT_LIST_HEAD(&desc->chunks);
516 
517 		list_add_tail(&desc->node, &list);
518 	}
519 
520 	spin_lock_irqsave(&chan->lock, flags);
521 	list_splice_tail(&list, &chan->desc.free);
522 	list_add_tail(&page->node, &chan->desc.pages);
523 	spin_unlock_irqrestore(&chan->lock, flags);
524 
525 	return 0;
526 }
527 
528 /*
529  * rcar_dmac_desc_put - Release a DMA transfer descriptor
530  * @chan: the DMA channel
531  * @desc: the descriptor
532  *
533  * Put the descriptor and its transfer chunk descriptors back in the channel's
534  * free descriptors lists. The descriptor's chunks list will be reinitialized to
535  * an empty list as a result.
536  *
537  * The descriptor must have been removed from the channel's lists before calling
538  * this function.
539  */
540 static void rcar_dmac_desc_put(struct rcar_dmac_chan *chan,
541 			       struct rcar_dmac_desc *desc)
542 {
543 	unsigned long flags;
544 
545 	spin_lock_irqsave(&chan->lock, flags);
546 	list_splice_tail_init(&desc->chunks, &chan->desc.chunks_free);
547 	list_add(&desc->node, &chan->desc.free);
548 	spin_unlock_irqrestore(&chan->lock, flags);
549 }
550 
551 static void rcar_dmac_desc_recycle_acked(struct rcar_dmac_chan *chan)
552 {
553 	struct rcar_dmac_desc *desc, *_desc;
554 	unsigned long flags;
555 	LIST_HEAD(list);
556 
557 	/*
558 	 * We have to temporarily move all descriptors from the wait list to a
559 	 * local list as iterating over the wait list, even with
560 	 * list_for_each_entry_safe, isn't safe if we release the channel lock
561 	 * around the rcar_dmac_desc_put() call.
562 	 */
563 	spin_lock_irqsave(&chan->lock, flags);
564 	list_splice_init(&chan->desc.wait, &list);
565 	spin_unlock_irqrestore(&chan->lock, flags);
566 
567 	list_for_each_entry_safe(desc, _desc, &list, node) {
568 		if (async_tx_test_ack(&desc->async_tx)) {
569 			list_del(&desc->node);
570 			rcar_dmac_desc_put(chan, desc);
571 		}
572 	}
573 
574 	if (list_empty(&list))
575 		return;
576 
577 	/* Put the remaining descriptors back in the wait list. */
578 	spin_lock_irqsave(&chan->lock, flags);
579 	list_splice(&list, &chan->desc.wait);
580 	spin_unlock_irqrestore(&chan->lock, flags);
581 }
582 
583 /*
584  * rcar_dmac_desc_get - Allocate a descriptor for a DMA transfer
585  * @chan: the DMA channel
586  *
587  * Locking: This function must be called in a non-atomic context.
588  *
589  * Return: A pointer to the allocated descriptor or NULL if no descriptor can
590  * be allocated.
591  */
592 static struct rcar_dmac_desc *rcar_dmac_desc_get(struct rcar_dmac_chan *chan)
593 {
594 	struct rcar_dmac_desc *desc;
595 	unsigned long flags;
596 	int ret;
597 
598 	/* Recycle acked descriptors before attempting allocation. */
599 	rcar_dmac_desc_recycle_acked(chan);
600 
601 	spin_lock_irqsave(&chan->lock, flags);
602 
603 	while (list_empty(&chan->desc.free)) {
604 		/*
605 		 * No free descriptors, allocate a page worth of them and try
606 		 * again, as someone else could race us to get the newly
607 		 * allocated descriptors. If the allocation fails return an
608 		 * error.
609 		 */
610 		spin_unlock_irqrestore(&chan->lock, flags);
611 		ret = rcar_dmac_desc_alloc(chan, GFP_NOWAIT);
612 		if (ret < 0)
613 			return NULL;
614 		spin_lock_irqsave(&chan->lock, flags);
615 	}
616 
617 	desc = list_first_entry(&chan->desc.free, struct rcar_dmac_desc, node);
618 	list_del(&desc->node);
619 
620 	spin_unlock_irqrestore(&chan->lock, flags);
621 
622 	return desc;
623 }
624 
625 /*
626  * rcar_dmac_xfer_chunk_alloc - Allocate a page worth of transfer chunks
627  * @chan: the DMA channel
628  * @gfp: allocation flags
629  */
630 static int rcar_dmac_xfer_chunk_alloc(struct rcar_dmac_chan *chan, gfp_t gfp)
631 {
632 	struct rcar_dmac_desc_page *page;
633 	unsigned long flags;
634 	LIST_HEAD(list);
635 	unsigned int i;
636 
637 	page = (void *)get_zeroed_page(gfp);
638 	if (!page)
639 		return -ENOMEM;
640 
641 	for (i = 0; i < RCAR_DMAC_XFER_CHUNKS_PER_PAGE; ++i) {
642 		struct rcar_dmac_xfer_chunk *chunk = &page->chunks[i];
643 
644 		list_add_tail(&chunk->node, &list);
645 	}
646 
647 	spin_lock_irqsave(&chan->lock, flags);
648 	list_splice_tail(&list, &chan->desc.chunks_free);
649 	list_add_tail(&page->node, &chan->desc.pages);
650 	spin_unlock_irqrestore(&chan->lock, flags);
651 
652 	return 0;
653 }
654 
655 /*
656  * rcar_dmac_xfer_chunk_get - Allocate a transfer chunk for a DMA transfer
657  * @chan: the DMA channel
658  *
659  * Locking: This function must be called in a non-atomic context.
660  *
661  * Return: A pointer to the allocated transfer chunk descriptor or NULL if no
662  * descriptor can be allocated.
663  */
664 static struct rcar_dmac_xfer_chunk *
665 rcar_dmac_xfer_chunk_get(struct rcar_dmac_chan *chan)
666 {
667 	struct rcar_dmac_xfer_chunk *chunk;
668 	unsigned long flags;
669 	int ret;
670 
671 	spin_lock_irqsave(&chan->lock, flags);
672 
673 	while (list_empty(&chan->desc.chunks_free)) {
674 		/*
675 		 * No free descriptors, allocate a page worth of them and try
676 		 * again, as someone else could race us to get the newly
677 		 * allocated descriptors. If the allocation fails return an
678 		 * error.
679 		 */
680 		spin_unlock_irqrestore(&chan->lock, flags);
681 		ret = rcar_dmac_xfer_chunk_alloc(chan, GFP_NOWAIT);
682 		if (ret < 0)
683 			return NULL;
684 		spin_lock_irqsave(&chan->lock, flags);
685 	}
686 
687 	chunk = list_first_entry(&chan->desc.chunks_free,
688 				 struct rcar_dmac_xfer_chunk, node);
689 	list_del(&chunk->node);
690 
691 	spin_unlock_irqrestore(&chan->lock, flags);
692 
693 	return chunk;
694 }
695 
696 static void rcar_dmac_realloc_hwdesc(struct rcar_dmac_chan *chan,
697 				     struct rcar_dmac_desc *desc, size_t size)
698 {
699 	/*
700 	 * dma_alloc_coherent() allocates memory in page size increments. To
701 	 * avoid reallocating the hardware descriptors when the allocated size
702 	 * wouldn't change align the requested size to a multiple of the page
703 	 * size.
704 	 */
705 	size = PAGE_ALIGN(size);
706 
707 	if (desc->hwdescs.size == size)
708 		return;
709 
710 	if (desc->hwdescs.mem) {
711 		dma_free_coherent(chan->chan.device->dev, desc->hwdescs.size,
712 				  desc->hwdescs.mem, desc->hwdescs.dma);
713 		desc->hwdescs.mem = NULL;
714 		desc->hwdescs.size = 0;
715 	}
716 
717 	if (!size)
718 		return;
719 
720 	desc->hwdescs.mem = dma_alloc_coherent(chan->chan.device->dev, size,
721 					       &desc->hwdescs.dma, GFP_NOWAIT);
722 	if (!desc->hwdescs.mem)
723 		return;
724 
725 	desc->hwdescs.size = size;
726 }
727 
728 static int rcar_dmac_fill_hwdesc(struct rcar_dmac_chan *chan,
729 				 struct rcar_dmac_desc *desc)
730 {
731 	struct rcar_dmac_xfer_chunk *chunk;
732 	struct rcar_dmac_hw_desc *hwdesc;
733 
734 	rcar_dmac_realloc_hwdesc(chan, desc, desc->nchunks * sizeof(*hwdesc));
735 
736 	hwdesc = desc->hwdescs.mem;
737 	if (!hwdesc)
738 		return -ENOMEM;
739 
740 	list_for_each_entry(chunk, &desc->chunks, node) {
741 		hwdesc->sar = chunk->src_addr;
742 		hwdesc->dar = chunk->dst_addr;
743 		hwdesc->tcr = chunk->size >> desc->xfer_shift;
744 		hwdesc++;
745 	}
746 
747 	return 0;
748 }
749 
750 /* -----------------------------------------------------------------------------
751  * Stop and reset
752  */
753 static void rcar_dmac_chcr_de_barrier(struct rcar_dmac_chan *chan)
754 {
755 	u32 chcr;
756 	unsigned int i;
757 
758 	/*
759 	 * Ensure that the setting of the DE bit is actually 0 after
760 	 * clearing it.
761 	 */
762 	for (i = 0; i < 1024; i++) {
763 		chcr = rcar_dmac_chan_read(chan, RCAR_DMACHCR);
764 		if (!(chcr & RCAR_DMACHCR_DE))
765 			return;
766 		udelay(1);
767 	}
768 
769 	dev_err(chan->chan.device->dev, "CHCR DE check error\n");
770 }
771 
772 static void rcar_dmac_clear_chcr_de(struct rcar_dmac_chan *chan)
773 {
774 	u32 chcr = rcar_dmac_chan_read(chan, RCAR_DMACHCR);
775 
776 	/* set DE=0 and flush remaining data */
777 	rcar_dmac_chan_write(chan, RCAR_DMACHCR, (chcr & ~RCAR_DMACHCR_DE));
778 
779 	/* make sure all remaining data was flushed */
780 	rcar_dmac_chcr_de_barrier(chan);
781 }
782 
783 static void rcar_dmac_chan_halt(struct rcar_dmac_chan *chan)
784 {
785 	u32 chcr = rcar_dmac_chan_read(chan, RCAR_DMACHCR);
786 
787 	chcr &= ~(RCAR_DMACHCR_DSE | RCAR_DMACHCR_DSIE | RCAR_DMACHCR_IE |
788 		  RCAR_DMACHCR_TE | RCAR_DMACHCR_DE |
789 		  RCAR_DMACHCR_CAE | RCAR_DMACHCR_CAIE);
790 	rcar_dmac_chan_write(chan, RCAR_DMACHCR, chcr);
791 	rcar_dmac_chcr_de_barrier(chan);
792 }
793 
794 static void rcar_dmac_chan_reinit(struct rcar_dmac_chan *chan)
795 {
796 	struct rcar_dmac_desc *desc, *_desc;
797 	unsigned long flags;
798 	LIST_HEAD(descs);
799 
800 	spin_lock_irqsave(&chan->lock, flags);
801 
802 	/* Move all non-free descriptors to the local lists. */
803 	list_splice_init(&chan->desc.pending, &descs);
804 	list_splice_init(&chan->desc.active, &descs);
805 	list_splice_init(&chan->desc.done, &descs);
806 	list_splice_init(&chan->desc.wait, &descs);
807 
808 	chan->desc.running = NULL;
809 
810 	spin_unlock_irqrestore(&chan->lock, flags);
811 
812 	list_for_each_entry_safe(desc, _desc, &descs, node) {
813 		list_del(&desc->node);
814 		rcar_dmac_desc_put(chan, desc);
815 	}
816 }
817 
818 static void rcar_dmac_stop_all_chan(struct rcar_dmac *dmac)
819 {
820 	unsigned int i;
821 
822 	/* Stop all channels. */
823 	for (i = 0; i < dmac->n_channels; ++i) {
824 		struct rcar_dmac_chan *chan = &dmac->channels[i];
825 
826 		if (!(dmac->channels_mask & BIT(i)))
827 			continue;
828 
829 		/* Stop and reinitialize the channel. */
830 		spin_lock_irq(&chan->lock);
831 		rcar_dmac_chan_halt(chan);
832 		spin_unlock_irq(&chan->lock);
833 	}
834 }
835 
836 static int rcar_dmac_chan_pause(struct dma_chan *chan)
837 {
838 	unsigned long flags;
839 	struct rcar_dmac_chan *rchan = to_rcar_dmac_chan(chan);
840 
841 	spin_lock_irqsave(&rchan->lock, flags);
842 	rcar_dmac_clear_chcr_de(rchan);
843 	spin_unlock_irqrestore(&rchan->lock, flags);
844 
845 	return 0;
846 }
847 
848 /* -----------------------------------------------------------------------------
849  * Descriptors preparation
850  */
851 
852 static void rcar_dmac_chan_configure_desc(struct rcar_dmac_chan *chan,
853 					  struct rcar_dmac_desc *desc)
854 {
855 	static const u32 chcr_ts[] = {
856 		RCAR_DMACHCR_TS_1B, RCAR_DMACHCR_TS_2B,
857 		RCAR_DMACHCR_TS_4B, RCAR_DMACHCR_TS_8B,
858 		RCAR_DMACHCR_TS_16B, RCAR_DMACHCR_TS_32B,
859 		RCAR_DMACHCR_TS_64B,
860 	};
861 
862 	unsigned int xfer_size;
863 	u32 chcr;
864 
865 	switch (desc->direction) {
866 	case DMA_DEV_TO_MEM:
867 		chcr = RCAR_DMACHCR_DM_INC | RCAR_DMACHCR_SM_FIXED
868 		     | RCAR_DMACHCR_RS_DMARS;
869 		xfer_size = chan->src.xfer_size;
870 		break;
871 
872 	case DMA_MEM_TO_DEV:
873 		chcr = RCAR_DMACHCR_DM_FIXED | RCAR_DMACHCR_SM_INC
874 		     | RCAR_DMACHCR_RS_DMARS;
875 		xfer_size = chan->dst.xfer_size;
876 		break;
877 
878 	case DMA_MEM_TO_MEM:
879 	default:
880 		chcr = RCAR_DMACHCR_DM_INC | RCAR_DMACHCR_SM_INC
881 		     | RCAR_DMACHCR_RS_AUTO;
882 		xfer_size = RCAR_DMAC_MEMCPY_XFER_SIZE;
883 		break;
884 	}
885 
886 	desc->xfer_shift = ilog2(xfer_size);
887 	desc->chcr = chcr | chcr_ts[desc->xfer_shift];
888 }
889 
890 /*
891  * rcar_dmac_chan_prep_sg - prepare transfer descriptors from an SG list
892  *
893  * Common routine for public (MEMCPY) and slave DMA. The MEMCPY case is also
894  * converted to scatter-gather to guarantee consistent locking and a correct
895  * list manipulation. For slave DMA direction carries the usual meaning, and,
896  * logically, the SG list is RAM and the addr variable contains slave address,
897  * e.g., the FIFO I/O register. For MEMCPY direction equals DMA_MEM_TO_MEM
898  * and the SG list contains only one element and points at the source buffer.
899  */
900 static struct dma_async_tx_descriptor *
901 rcar_dmac_chan_prep_sg(struct rcar_dmac_chan *chan, struct scatterlist *sgl,
902 		       unsigned int sg_len, dma_addr_t dev_addr,
903 		       enum dma_transfer_direction dir, unsigned long dma_flags,
904 		       bool cyclic)
905 {
906 	struct rcar_dmac_xfer_chunk *chunk;
907 	struct rcar_dmac_desc *desc;
908 	struct scatterlist *sg;
909 	unsigned int nchunks = 0;
910 	unsigned int max_chunk_size;
911 	unsigned int full_size = 0;
912 	bool cross_boundary = false;
913 	unsigned int i;
914 #ifdef CONFIG_ARCH_DMA_ADDR_T_64BIT
915 	u32 high_dev_addr;
916 	u32 high_mem_addr;
917 #endif
918 
919 	desc = rcar_dmac_desc_get(chan);
920 	if (!desc)
921 		return NULL;
922 
923 	desc->async_tx.flags = dma_flags;
924 	desc->async_tx.cookie = -EBUSY;
925 
926 	desc->cyclic = cyclic;
927 	desc->direction = dir;
928 
929 	rcar_dmac_chan_configure_desc(chan, desc);
930 
931 	max_chunk_size = RCAR_DMATCR_MASK << desc->xfer_shift;
932 
933 	/*
934 	 * Allocate and fill the transfer chunk descriptors. We own the only
935 	 * reference to the DMA descriptor, there's no need for locking.
936 	 */
937 	for_each_sg(sgl, sg, sg_len, i) {
938 		dma_addr_t mem_addr = sg_dma_address(sg);
939 		unsigned int len = sg_dma_len(sg);
940 
941 		full_size += len;
942 
943 #ifdef CONFIG_ARCH_DMA_ADDR_T_64BIT
944 		if (i == 0) {
945 			high_dev_addr = dev_addr >> 32;
946 			high_mem_addr = mem_addr >> 32;
947 		}
948 
949 		if ((dev_addr >> 32 != high_dev_addr) ||
950 		    (mem_addr >> 32 != high_mem_addr))
951 			cross_boundary = true;
952 #endif
953 		while (len) {
954 			unsigned int size = min(len, max_chunk_size);
955 
956 #ifdef CONFIG_ARCH_DMA_ADDR_T_64BIT
957 			/*
958 			 * Prevent individual transfers from crossing 4GB
959 			 * boundaries.
960 			 */
961 			if (dev_addr >> 32 != (dev_addr + size - 1) >> 32) {
962 				size = ALIGN(dev_addr, 1ULL << 32) - dev_addr;
963 				cross_boundary = true;
964 			}
965 			if (mem_addr >> 32 != (mem_addr + size - 1) >> 32) {
966 				size = ALIGN(mem_addr, 1ULL << 32) - mem_addr;
967 				cross_boundary = true;
968 			}
969 #endif
970 
971 			chunk = rcar_dmac_xfer_chunk_get(chan);
972 			if (!chunk) {
973 				rcar_dmac_desc_put(chan, desc);
974 				return NULL;
975 			}
976 
977 			if (dir == DMA_DEV_TO_MEM) {
978 				chunk->src_addr = dev_addr;
979 				chunk->dst_addr = mem_addr;
980 			} else {
981 				chunk->src_addr = mem_addr;
982 				chunk->dst_addr = dev_addr;
983 			}
984 
985 			chunk->size = size;
986 
987 			dev_dbg(chan->chan.device->dev,
988 				"chan%u: chunk %p/%p sgl %u@%p, %u/%u %pad -> %pad\n",
989 				chan->index, chunk, desc, i, sg, size, len,
990 				&chunk->src_addr, &chunk->dst_addr);
991 
992 			mem_addr += size;
993 			if (dir == DMA_MEM_TO_MEM)
994 				dev_addr += size;
995 
996 			len -= size;
997 
998 			list_add_tail(&chunk->node, &desc->chunks);
999 			nchunks++;
1000 		}
1001 	}
1002 
1003 	desc->nchunks = nchunks;
1004 	desc->size = full_size;
1005 
1006 	/*
1007 	 * Use hardware descriptor lists if possible when more than one chunk
1008 	 * needs to be transferred (otherwise they don't make much sense).
1009 	 *
1010 	 * Source/Destination address should be located in same 4GiB region
1011 	 * in the 40bit address space when it uses Hardware descriptor,
1012 	 * and cross_boundary is checking it.
1013 	 */
1014 	desc->hwdescs.use = !cross_boundary && nchunks > 1;
1015 	if (desc->hwdescs.use) {
1016 		if (rcar_dmac_fill_hwdesc(chan, desc) < 0)
1017 			desc->hwdescs.use = false;
1018 	}
1019 
1020 	return &desc->async_tx;
1021 }
1022 
1023 /* -----------------------------------------------------------------------------
1024  * DMA engine operations
1025  */
1026 
1027 static int rcar_dmac_alloc_chan_resources(struct dma_chan *chan)
1028 {
1029 	struct rcar_dmac_chan *rchan = to_rcar_dmac_chan(chan);
1030 	int ret;
1031 
1032 	INIT_LIST_HEAD(&rchan->desc.chunks_free);
1033 	INIT_LIST_HEAD(&rchan->desc.pages);
1034 
1035 	/* Preallocate descriptors. */
1036 	ret = rcar_dmac_xfer_chunk_alloc(rchan, GFP_KERNEL);
1037 	if (ret < 0)
1038 		return -ENOMEM;
1039 
1040 	ret = rcar_dmac_desc_alloc(rchan, GFP_KERNEL);
1041 	if (ret < 0)
1042 		return -ENOMEM;
1043 
1044 	return pm_runtime_get_sync(chan->device->dev);
1045 }
1046 
1047 static void rcar_dmac_free_chan_resources(struct dma_chan *chan)
1048 {
1049 	struct rcar_dmac_chan *rchan = to_rcar_dmac_chan(chan);
1050 	struct rcar_dmac *dmac = to_rcar_dmac(chan->device);
1051 	struct rcar_dmac_chan_map *map = &rchan->map;
1052 	struct rcar_dmac_desc_page *page, *_page;
1053 	struct rcar_dmac_desc *desc;
1054 	LIST_HEAD(list);
1055 
1056 	/* Protect against ISR */
1057 	spin_lock_irq(&rchan->lock);
1058 	rcar_dmac_chan_halt(rchan);
1059 	spin_unlock_irq(&rchan->lock);
1060 
1061 	/*
1062 	 * Now no new interrupts will occur, but one might already be
1063 	 * running. Wait for it to finish before freeing resources.
1064 	 */
1065 	synchronize_irq(rchan->irq);
1066 
1067 	if (rchan->mid_rid >= 0) {
1068 		/* The caller is holding dma_list_mutex */
1069 		clear_bit(rchan->mid_rid, dmac->modules);
1070 		rchan->mid_rid = -EINVAL;
1071 	}
1072 
1073 	list_splice_init(&rchan->desc.free, &list);
1074 	list_splice_init(&rchan->desc.pending, &list);
1075 	list_splice_init(&rchan->desc.active, &list);
1076 	list_splice_init(&rchan->desc.done, &list);
1077 	list_splice_init(&rchan->desc.wait, &list);
1078 
1079 	rchan->desc.running = NULL;
1080 
1081 	list_for_each_entry(desc, &list, node)
1082 		rcar_dmac_realloc_hwdesc(rchan, desc, 0);
1083 
1084 	list_for_each_entry_safe(page, _page, &rchan->desc.pages, node) {
1085 		list_del(&page->node);
1086 		free_page((unsigned long)page);
1087 	}
1088 
1089 	/* Remove slave mapping if present. */
1090 	if (map->slave.xfer_size) {
1091 		dma_unmap_resource(chan->device->dev, map->addr,
1092 				   map->slave.xfer_size, map->dir, 0);
1093 		map->slave.xfer_size = 0;
1094 	}
1095 
1096 	pm_runtime_put(chan->device->dev);
1097 }
1098 
1099 static struct dma_async_tx_descriptor *
1100 rcar_dmac_prep_dma_memcpy(struct dma_chan *chan, dma_addr_t dma_dest,
1101 			  dma_addr_t dma_src, size_t len, unsigned long flags)
1102 {
1103 	struct rcar_dmac_chan *rchan = to_rcar_dmac_chan(chan);
1104 	struct scatterlist sgl;
1105 
1106 	if (!len)
1107 		return NULL;
1108 
1109 	sg_init_table(&sgl, 1);
1110 	sg_set_page(&sgl, pfn_to_page(PFN_DOWN(dma_src)), len,
1111 		    offset_in_page(dma_src));
1112 	sg_dma_address(&sgl) = dma_src;
1113 	sg_dma_len(&sgl) = len;
1114 
1115 	return rcar_dmac_chan_prep_sg(rchan, &sgl, 1, dma_dest,
1116 				      DMA_MEM_TO_MEM, flags, false);
1117 }
1118 
1119 static int rcar_dmac_map_slave_addr(struct dma_chan *chan,
1120 				    enum dma_transfer_direction dir)
1121 {
1122 	struct rcar_dmac_chan *rchan = to_rcar_dmac_chan(chan);
1123 	struct rcar_dmac_chan_map *map = &rchan->map;
1124 	phys_addr_t dev_addr;
1125 	size_t dev_size;
1126 	enum dma_data_direction dev_dir;
1127 
1128 	if (dir == DMA_DEV_TO_MEM) {
1129 		dev_addr = rchan->src.slave_addr;
1130 		dev_size = rchan->src.xfer_size;
1131 		dev_dir = DMA_TO_DEVICE;
1132 	} else {
1133 		dev_addr = rchan->dst.slave_addr;
1134 		dev_size = rchan->dst.xfer_size;
1135 		dev_dir = DMA_FROM_DEVICE;
1136 	}
1137 
1138 	/* Reuse current map if possible. */
1139 	if (dev_addr == map->slave.slave_addr &&
1140 	    dev_size == map->slave.xfer_size &&
1141 	    dev_dir == map->dir)
1142 		return 0;
1143 
1144 	/* Remove old mapping if present. */
1145 	if (map->slave.xfer_size)
1146 		dma_unmap_resource(chan->device->dev, map->addr,
1147 				   map->slave.xfer_size, map->dir, 0);
1148 	map->slave.xfer_size = 0;
1149 
1150 	/* Create new slave address map. */
1151 	map->addr = dma_map_resource(chan->device->dev, dev_addr, dev_size,
1152 				     dev_dir, 0);
1153 
1154 	if (dma_mapping_error(chan->device->dev, map->addr)) {
1155 		dev_err(chan->device->dev,
1156 			"chan%u: failed to map %zx@%pap", rchan->index,
1157 			dev_size, &dev_addr);
1158 		return -EIO;
1159 	}
1160 
1161 	dev_dbg(chan->device->dev, "chan%u: map %zx@%pap to %pad dir: %s\n",
1162 		rchan->index, dev_size, &dev_addr, &map->addr,
1163 		dev_dir == DMA_TO_DEVICE ? "DMA_TO_DEVICE" : "DMA_FROM_DEVICE");
1164 
1165 	map->slave.slave_addr = dev_addr;
1166 	map->slave.xfer_size = dev_size;
1167 	map->dir = dev_dir;
1168 
1169 	return 0;
1170 }
1171 
1172 static struct dma_async_tx_descriptor *
1173 rcar_dmac_prep_slave_sg(struct dma_chan *chan, struct scatterlist *sgl,
1174 			unsigned int sg_len, enum dma_transfer_direction dir,
1175 			unsigned long flags, void *context)
1176 {
1177 	struct rcar_dmac_chan *rchan = to_rcar_dmac_chan(chan);
1178 
1179 	/* Someone calling slave DMA on a generic channel? */
1180 	if (rchan->mid_rid < 0 || !sg_len || !sg_dma_len(sgl)) {
1181 		dev_warn(chan->device->dev,
1182 			 "%s: bad parameter: len=%d, id=%d\n",
1183 			 __func__, sg_len, rchan->mid_rid);
1184 		return NULL;
1185 	}
1186 
1187 	if (rcar_dmac_map_slave_addr(chan, dir))
1188 		return NULL;
1189 
1190 	return rcar_dmac_chan_prep_sg(rchan, sgl, sg_len, rchan->map.addr,
1191 				      dir, flags, false);
1192 }
1193 
1194 #define RCAR_DMAC_MAX_SG_LEN	32
1195 
1196 static struct dma_async_tx_descriptor *
1197 rcar_dmac_prep_dma_cyclic(struct dma_chan *chan, dma_addr_t buf_addr,
1198 			  size_t buf_len, size_t period_len,
1199 			  enum dma_transfer_direction dir, unsigned long flags)
1200 {
1201 	struct rcar_dmac_chan *rchan = to_rcar_dmac_chan(chan);
1202 	struct dma_async_tx_descriptor *desc;
1203 	struct scatterlist *sgl;
1204 	unsigned int sg_len;
1205 	unsigned int i;
1206 
1207 	/* Someone calling slave DMA on a generic channel? */
1208 	if (rchan->mid_rid < 0 || buf_len < period_len) {
1209 		dev_warn(chan->device->dev,
1210 			"%s: bad parameter: buf_len=%zu, period_len=%zu, id=%d\n",
1211 			__func__, buf_len, period_len, rchan->mid_rid);
1212 		return NULL;
1213 	}
1214 
1215 	if (rcar_dmac_map_slave_addr(chan, dir))
1216 		return NULL;
1217 
1218 	sg_len = buf_len / period_len;
1219 	if (sg_len > RCAR_DMAC_MAX_SG_LEN) {
1220 		dev_err(chan->device->dev,
1221 			"chan%u: sg length %d exceeds limit %d",
1222 			rchan->index, sg_len, RCAR_DMAC_MAX_SG_LEN);
1223 		return NULL;
1224 	}
1225 
1226 	/*
1227 	 * Allocate the sg list dynamically as it would consume too much stack
1228 	 * space.
1229 	 */
1230 	sgl = kmalloc_array(sg_len, sizeof(*sgl), GFP_NOWAIT);
1231 	if (!sgl)
1232 		return NULL;
1233 
1234 	sg_init_table(sgl, sg_len);
1235 
1236 	for (i = 0; i < sg_len; ++i) {
1237 		dma_addr_t src = buf_addr + (period_len * i);
1238 
1239 		sg_set_page(&sgl[i], pfn_to_page(PFN_DOWN(src)), period_len,
1240 			    offset_in_page(src));
1241 		sg_dma_address(&sgl[i]) = src;
1242 		sg_dma_len(&sgl[i]) = period_len;
1243 	}
1244 
1245 	desc = rcar_dmac_chan_prep_sg(rchan, sgl, sg_len, rchan->map.addr,
1246 				      dir, flags, true);
1247 
1248 	kfree(sgl);
1249 	return desc;
1250 }
1251 
1252 static int rcar_dmac_device_config(struct dma_chan *chan,
1253 				   struct dma_slave_config *cfg)
1254 {
1255 	struct rcar_dmac_chan *rchan = to_rcar_dmac_chan(chan);
1256 
1257 	/*
1258 	 * We could lock this, but you shouldn't be configuring the
1259 	 * channel, while using it...
1260 	 */
1261 	rchan->src.slave_addr = cfg->src_addr;
1262 	rchan->dst.slave_addr = cfg->dst_addr;
1263 	rchan->src.xfer_size = cfg->src_addr_width;
1264 	rchan->dst.xfer_size = cfg->dst_addr_width;
1265 
1266 	return 0;
1267 }
1268 
1269 static int rcar_dmac_chan_terminate_all(struct dma_chan *chan)
1270 {
1271 	struct rcar_dmac_chan *rchan = to_rcar_dmac_chan(chan);
1272 	unsigned long flags;
1273 
1274 	spin_lock_irqsave(&rchan->lock, flags);
1275 	rcar_dmac_chan_halt(rchan);
1276 	spin_unlock_irqrestore(&rchan->lock, flags);
1277 
1278 	/*
1279 	 * FIXME: No new interrupt can occur now, but the IRQ thread might still
1280 	 * be running.
1281 	 */
1282 
1283 	rcar_dmac_chan_reinit(rchan);
1284 
1285 	return 0;
1286 }
1287 
1288 static unsigned int rcar_dmac_chan_get_residue(struct rcar_dmac_chan *chan,
1289 					       dma_cookie_t cookie)
1290 {
1291 	struct rcar_dmac_desc *desc = chan->desc.running;
1292 	struct rcar_dmac_xfer_chunk *running = NULL;
1293 	struct rcar_dmac_xfer_chunk *chunk;
1294 	enum dma_status status;
1295 	unsigned int residue = 0;
1296 	unsigned int dptr = 0;
1297 	unsigned int chcrb;
1298 	unsigned int tcrb;
1299 	unsigned int i;
1300 
1301 	if (!desc)
1302 		return 0;
1303 
1304 	/*
1305 	 * If the cookie corresponds to a descriptor that has been completed
1306 	 * there is no residue. The same check has already been performed by the
1307 	 * caller but without holding the channel lock, so the descriptor could
1308 	 * now be complete.
1309 	 */
1310 	status = dma_cookie_status(&chan->chan, cookie, NULL);
1311 	if (status == DMA_COMPLETE)
1312 		return 0;
1313 
1314 	/*
1315 	 * If the cookie doesn't correspond to the currently running transfer
1316 	 * then the descriptor hasn't been processed yet, and the residue is
1317 	 * equal to the full descriptor size.
1318 	 * Also, a client driver is possible to call this function before
1319 	 * rcar_dmac_isr_channel_thread() runs. In this case, the "desc.running"
1320 	 * will be the next descriptor, and the done list will appear. So, if
1321 	 * the argument cookie matches the done list's cookie, we can assume
1322 	 * the residue is zero.
1323 	 */
1324 	if (cookie != desc->async_tx.cookie) {
1325 		list_for_each_entry(desc, &chan->desc.done, node) {
1326 			if (cookie == desc->async_tx.cookie)
1327 				return 0;
1328 		}
1329 		list_for_each_entry(desc, &chan->desc.pending, node) {
1330 			if (cookie == desc->async_tx.cookie)
1331 				return desc->size;
1332 		}
1333 		list_for_each_entry(desc, &chan->desc.active, node) {
1334 			if (cookie == desc->async_tx.cookie)
1335 				return desc->size;
1336 		}
1337 
1338 		/*
1339 		 * No descriptor found for the cookie, there's thus no residue.
1340 		 * This shouldn't happen if the calling driver passes a correct
1341 		 * cookie value.
1342 		 */
1343 		WARN(1, "No descriptor for cookie!");
1344 		return 0;
1345 	}
1346 
1347 	/*
1348 	 * We need to read two registers.
1349 	 * Make sure the control register does not skip to next chunk
1350 	 * while reading the counter.
1351 	 * Trying it 3 times should be enough: Initial read, retry, retry
1352 	 * for the paranoid.
1353 	 */
1354 	for (i = 0; i < 3; i++) {
1355 		chcrb = rcar_dmac_chan_read(chan, RCAR_DMACHCRB) &
1356 					    RCAR_DMACHCRB_DPTR_MASK;
1357 		tcrb = rcar_dmac_chan_read(chan, RCAR_DMATCRB);
1358 		/* Still the same? */
1359 		if (chcrb == (rcar_dmac_chan_read(chan, RCAR_DMACHCRB) &
1360 			      RCAR_DMACHCRB_DPTR_MASK))
1361 			break;
1362 	}
1363 	WARN_ONCE(i >= 3, "residue might be not continuous!");
1364 
1365 	/*
1366 	 * In descriptor mode the descriptor running pointer is not maintained
1367 	 * by the interrupt handler, find the running descriptor from the
1368 	 * descriptor pointer field in the CHCRB register. In non-descriptor
1369 	 * mode just use the running descriptor pointer.
1370 	 */
1371 	if (desc->hwdescs.use) {
1372 		dptr = chcrb >> RCAR_DMACHCRB_DPTR_SHIFT;
1373 		if (dptr == 0)
1374 			dptr = desc->nchunks;
1375 		dptr--;
1376 		WARN_ON(dptr >= desc->nchunks);
1377 	} else {
1378 		running = desc->running;
1379 	}
1380 
1381 	/* Compute the size of all chunks still to be transferred. */
1382 	list_for_each_entry_reverse(chunk, &desc->chunks, node) {
1383 		if (chunk == running || ++dptr == desc->nchunks)
1384 			break;
1385 
1386 		residue += chunk->size;
1387 	}
1388 
1389 	/* Add the residue for the current chunk. */
1390 	residue += tcrb << desc->xfer_shift;
1391 
1392 	return residue;
1393 }
1394 
1395 static enum dma_status rcar_dmac_tx_status(struct dma_chan *chan,
1396 					   dma_cookie_t cookie,
1397 					   struct dma_tx_state *txstate)
1398 {
1399 	struct rcar_dmac_chan *rchan = to_rcar_dmac_chan(chan);
1400 	enum dma_status status;
1401 	unsigned long flags;
1402 	unsigned int residue;
1403 	bool cyclic;
1404 
1405 	status = dma_cookie_status(chan, cookie, txstate);
1406 	if (status == DMA_COMPLETE || !txstate)
1407 		return status;
1408 
1409 	spin_lock_irqsave(&rchan->lock, flags);
1410 	residue = rcar_dmac_chan_get_residue(rchan, cookie);
1411 	cyclic = rchan->desc.running ? rchan->desc.running->cyclic : false;
1412 	spin_unlock_irqrestore(&rchan->lock, flags);
1413 
1414 	/* if there's no residue, the cookie is complete */
1415 	if (!residue && !cyclic)
1416 		return DMA_COMPLETE;
1417 
1418 	dma_set_residue(txstate, residue);
1419 
1420 	return status;
1421 }
1422 
1423 static void rcar_dmac_issue_pending(struct dma_chan *chan)
1424 {
1425 	struct rcar_dmac_chan *rchan = to_rcar_dmac_chan(chan);
1426 	unsigned long flags;
1427 
1428 	spin_lock_irqsave(&rchan->lock, flags);
1429 
1430 	if (list_empty(&rchan->desc.pending))
1431 		goto done;
1432 
1433 	/* Append the pending list to the active list. */
1434 	list_splice_tail_init(&rchan->desc.pending, &rchan->desc.active);
1435 
1436 	/*
1437 	 * If no transfer is running pick the first descriptor from the active
1438 	 * list and start the transfer.
1439 	 */
1440 	if (!rchan->desc.running) {
1441 		struct rcar_dmac_desc *desc;
1442 
1443 		desc = list_first_entry(&rchan->desc.active,
1444 					struct rcar_dmac_desc, node);
1445 		rchan->desc.running = desc;
1446 
1447 		rcar_dmac_chan_start_xfer(rchan);
1448 	}
1449 
1450 done:
1451 	spin_unlock_irqrestore(&rchan->lock, flags);
1452 }
1453 
1454 static void rcar_dmac_device_synchronize(struct dma_chan *chan)
1455 {
1456 	struct rcar_dmac_chan *rchan = to_rcar_dmac_chan(chan);
1457 
1458 	synchronize_irq(rchan->irq);
1459 }
1460 
1461 /* -----------------------------------------------------------------------------
1462  * IRQ handling
1463  */
1464 
1465 static irqreturn_t rcar_dmac_isr_desc_stage_end(struct rcar_dmac_chan *chan)
1466 {
1467 	struct rcar_dmac_desc *desc = chan->desc.running;
1468 	unsigned int stage;
1469 
1470 	if (WARN_ON(!desc || !desc->cyclic)) {
1471 		/*
1472 		 * This should never happen, there should always be a running
1473 		 * cyclic descriptor when a descriptor stage end interrupt is
1474 		 * triggered. Warn and return.
1475 		 */
1476 		return IRQ_NONE;
1477 	}
1478 
1479 	/* Program the interrupt pointer to the next stage. */
1480 	stage = (rcar_dmac_chan_read(chan, RCAR_DMACHCRB) &
1481 		 RCAR_DMACHCRB_DPTR_MASK) >> RCAR_DMACHCRB_DPTR_SHIFT;
1482 	rcar_dmac_chan_write(chan, RCAR_DMADPCR, RCAR_DMADPCR_DIPT(stage));
1483 
1484 	return IRQ_WAKE_THREAD;
1485 }
1486 
1487 static irqreturn_t rcar_dmac_isr_transfer_end(struct rcar_dmac_chan *chan)
1488 {
1489 	struct rcar_dmac_desc *desc = chan->desc.running;
1490 	irqreturn_t ret = IRQ_WAKE_THREAD;
1491 
1492 	if (WARN_ON_ONCE(!desc)) {
1493 		/*
1494 		 * This should never happen, there should always be a running
1495 		 * descriptor when a transfer end interrupt is triggered. Warn
1496 		 * and return.
1497 		 */
1498 		return IRQ_NONE;
1499 	}
1500 
1501 	/*
1502 	 * The transfer end interrupt isn't generated for each chunk when using
1503 	 * descriptor mode. Only update the running chunk pointer in
1504 	 * non-descriptor mode.
1505 	 */
1506 	if (!desc->hwdescs.use) {
1507 		/*
1508 		 * If we haven't completed the last transfer chunk simply move
1509 		 * to the next one. Only wake the IRQ thread if the transfer is
1510 		 * cyclic.
1511 		 */
1512 		if (!list_is_last(&desc->running->node, &desc->chunks)) {
1513 			desc->running = list_next_entry(desc->running, node);
1514 			if (!desc->cyclic)
1515 				ret = IRQ_HANDLED;
1516 			goto done;
1517 		}
1518 
1519 		/*
1520 		 * We've completed the last transfer chunk. If the transfer is
1521 		 * cyclic, move back to the first one.
1522 		 */
1523 		if (desc->cyclic) {
1524 			desc->running =
1525 				list_first_entry(&desc->chunks,
1526 						 struct rcar_dmac_xfer_chunk,
1527 						 node);
1528 			goto done;
1529 		}
1530 	}
1531 
1532 	/* The descriptor is complete, move it to the done list. */
1533 	list_move_tail(&desc->node, &chan->desc.done);
1534 
1535 	/* Queue the next descriptor, if any. */
1536 	if (!list_empty(&chan->desc.active))
1537 		chan->desc.running = list_first_entry(&chan->desc.active,
1538 						      struct rcar_dmac_desc,
1539 						      node);
1540 	else
1541 		chan->desc.running = NULL;
1542 
1543 done:
1544 	if (chan->desc.running)
1545 		rcar_dmac_chan_start_xfer(chan);
1546 
1547 	return ret;
1548 }
1549 
1550 static irqreturn_t rcar_dmac_isr_channel(int irq, void *dev)
1551 {
1552 	u32 mask = RCAR_DMACHCR_DSE | RCAR_DMACHCR_TE;
1553 	struct rcar_dmac_chan *chan = dev;
1554 	irqreturn_t ret = IRQ_NONE;
1555 	bool reinit = false;
1556 	u32 chcr;
1557 
1558 	spin_lock(&chan->lock);
1559 
1560 	chcr = rcar_dmac_chan_read(chan, RCAR_DMACHCR);
1561 	if (chcr & RCAR_DMACHCR_CAE) {
1562 		struct rcar_dmac *dmac = to_rcar_dmac(chan->chan.device);
1563 
1564 		/*
1565 		 * We don't need to call rcar_dmac_chan_halt()
1566 		 * because channel is already stopped in error case.
1567 		 * We need to clear register and check DE bit as recovery.
1568 		 */
1569 		rcar_dmac_write(dmac, RCAR_DMACHCLR, 1 << chan->index);
1570 		rcar_dmac_chcr_de_barrier(chan);
1571 		reinit = true;
1572 		goto spin_lock_end;
1573 	}
1574 
1575 	if (chcr & RCAR_DMACHCR_TE)
1576 		mask |= RCAR_DMACHCR_DE;
1577 	rcar_dmac_chan_write(chan, RCAR_DMACHCR, chcr & ~mask);
1578 	if (mask & RCAR_DMACHCR_DE)
1579 		rcar_dmac_chcr_de_barrier(chan);
1580 
1581 	if (chcr & RCAR_DMACHCR_DSE)
1582 		ret |= rcar_dmac_isr_desc_stage_end(chan);
1583 
1584 	if (chcr & RCAR_DMACHCR_TE)
1585 		ret |= rcar_dmac_isr_transfer_end(chan);
1586 
1587 spin_lock_end:
1588 	spin_unlock(&chan->lock);
1589 
1590 	if (reinit) {
1591 		dev_err(chan->chan.device->dev, "Channel Address Error\n");
1592 
1593 		rcar_dmac_chan_reinit(chan);
1594 		ret = IRQ_HANDLED;
1595 	}
1596 
1597 	return ret;
1598 }
1599 
1600 static irqreturn_t rcar_dmac_isr_channel_thread(int irq, void *dev)
1601 {
1602 	struct rcar_dmac_chan *chan = dev;
1603 	struct rcar_dmac_desc *desc;
1604 	struct dmaengine_desc_callback cb;
1605 
1606 	spin_lock_irq(&chan->lock);
1607 
1608 	/* For cyclic transfers notify the user after every chunk. */
1609 	if (chan->desc.running && chan->desc.running->cyclic) {
1610 		desc = chan->desc.running;
1611 		dmaengine_desc_get_callback(&desc->async_tx, &cb);
1612 
1613 		if (dmaengine_desc_callback_valid(&cb)) {
1614 			spin_unlock_irq(&chan->lock);
1615 			dmaengine_desc_callback_invoke(&cb, NULL);
1616 			spin_lock_irq(&chan->lock);
1617 		}
1618 	}
1619 
1620 	/*
1621 	 * Call the callback function for all descriptors on the done list and
1622 	 * move them to the ack wait list.
1623 	 */
1624 	while (!list_empty(&chan->desc.done)) {
1625 		desc = list_first_entry(&chan->desc.done, struct rcar_dmac_desc,
1626 					node);
1627 		dma_cookie_complete(&desc->async_tx);
1628 		list_del(&desc->node);
1629 
1630 		dmaengine_desc_get_callback(&desc->async_tx, &cb);
1631 		if (dmaengine_desc_callback_valid(&cb)) {
1632 			spin_unlock_irq(&chan->lock);
1633 			/*
1634 			 * We own the only reference to this descriptor, we can
1635 			 * safely dereference it without holding the channel
1636 			 * lock.
1637 			 */
1638 			dmaengine_desc_callback_invoke(&cb, NULL);
1639 			spin_lock_irq(&chan->lock);
1640 		}
1641 
1642 		list_add_tail(&desc->node, &chan->desc.wait);
1643 	}
1644 
1645 	spin_unlock_irq(&chan->lock);
1646 
1647 	/* Recycle all acked descriptors. */
1648 	rcar_dmac_desc_recycle_acked(chan);
1649 
1650 	return IRQ_HANDLED;
1651 }
1652 
1653 /* -----------------------------------------------------------------------------
1654  * OF xlate and channel filter
1655  */
1656 
1657 static bool rcar_dmac_chan_filter(struct dma_chan *chan, void *arg)
1658 {
1659 	struct rcar_dmac *dmac = to_rcar_dmac(chan->device);
1660 	struct of_phandle_args *dma_spec = arg;
1661 
1662 	/*
1663 	 * FIXME: Using a filter on OF platforms is a nonsense. The OF xlate
1664 	 * function knows from which device it wants to allocate a channel from,
1665 	 * and would be perfectly capable of selecting the channel it wants.
1666 	 * Forcing it to call dma_request_channel() and iterate through all
1667 	 * channels from all controllers is just pointless.
1668 	 */
1669 	if (chan->device->device_config != rcar_dmac_device_config)
1670 		return false;
1671 
1672 	return !test_and_set_bit(dma_spec->args[0], dmac->modules);
1673 }
1674 
1675 static struct dma_chan *rcar_dmac_of_xlate(struct of_phandle_args *dma_spec,
1676 					   struct of_dma *ofdma)
1677 {
1678 	struct rcar_dmac_chan *rchan;
1679 	struct dma_chan *chan;
1680 	dma_cap_mask_t mask;
1681 
1682 	if (dma_spec->args_count != 1)
1683 		return NULL;
1684 
1685 	/* Only slave DMA channels can be allocated via DT */
1686 	dma_cap_zero(mask);
1687 	dma_cap_set(DMA_SLAVE, mask);
1688 
1689 	chan = __dma_request_channel(&mask, rcar_dmac_chan_filter, dma_spec,
1690 				     ofdma->of_node);
1691 	if (!chan)
1692 		return NULL;
1693 
1694 	rchan = to_rcar_dmac_chan(chan);
1695 	rchan->mid_rid = dma_spec->args[0];
1696 
1697 	return chan;
1698 }
1699 
1700 /* -----------------------------------------------------------------------------
1701  * Power management
1702  */
1703 
1704 #ifdef CONFIG_PM
1705 static int rcar_dmac_runtime_suspend(struct device *dev)
1706 {
1707 	return 0;
1708 }
1709 
1710 static int rcar_dmac_runtime_resume(struct device *dev)
1711 {
1712 	struct rcar_dmac *dmac = dev_get_drvdata(dev);
1713 
1714 	return rcar_dmac_init(dmac);
1715 }
1716 #endif
1717 
1718 static const struct dev_pm_ops rcar_dmac_pm = {
1719 	/*
1720 	 * TODO for system sleep/resume:
1721 	 *   - Wait for the current transfer to complete and stop the device,
1722 	 *   - Resume transfers, if any.
1723 	 */
1724 	SET_NOIRQ_SYSTEM_SLEEP_PM_OPS(pm_runtime_force_suspend,
1725 				      pm_runtime_force_resume)
1726 	SET_RUNTIME_PM_OPS(rcar_dmac_runtime_suspend, rcar_dmac_runtime_resume,
1727 			   NULL)
1728 };
1729 
1730 /* -----------------------------------------------------------------------------
1731  * Probe and remove
1732  */
1733 
1734 static int rcar_dmac_chan_probe(struct rcar_dmac *dmac,
1735 				struct rcar_dmac_chan *rchan,
1736 				const struct rcar_dmac_of_data *data,
1737 				unsigned int index)
1738 {
1739 	struct platform_device *pdev = to_platform_device(dmac->dev);
1740 	struct dma_chan *chan = &rchan->chan;
1741 	char pdev_irqname[5];
1742 	char *irqname;
1743 	int ret;
1744 
1745 	rchan->index = index;
1746 	rchan->iomem = dmac->iomem + data->chan_offset_base +
1747 		       data->chan_offset_stride * index;
1748 	rchan->mid_rid = -EINVAL;
1749 
1750 	spin_lock_init(&rchan->lock);
1751 
1752 	INIT_LIST_HEAD(&rchan->desc.free);
1753 	INIT_LIST_HEAD(&rchan->desc.pending);
1754 	INIT_LIST_HEAD(&rchan->desc.active);
1755 	INIT_LIST_HEAD(&rchan->desc.done);
1756 	INIT_LIST_HEAD(&rchan->desc.wait);
1757 
1758 	/* Request the channel interrupt. */
1759 	sprintf(pdev_irqname, "ch%u", index);
1760 	rchan->irq = platform_get_irq_byname(pdev, pdev_irqname);
1761 	if (rchan->irq < 0)
1762 		return -ENODEV;
1763 
1764 	irqname = devm_kasprintf(dmac->dev, GFP_KERNEL, "%s:%u",
1765 				 dev_name(dmac->dev), index);
1766 	if (!irqname)
1767 		return -ENOMEM;
1768 
1769 	/*
1770 	 * Initialize the DMA engine channel and add it to the DMA engine
1771 	 * channels list.
1772 	 */
1773 	chan->device = &dmac->engine;
1774 	dma_cookie_init(chan);
1775 
1776 	list_add_tail(&chan->device_node, &dmac->engine.channels);
1777 
1778 	ret = devm_request_threaded_irq(dmac->dev, rchan->irq,
1779 					rcar_dmac_isr_channel,
1780 					rcar_dmac_isr_channel_thread, 0,
1781 					irqname, rchan);
1782 	if (ret) {
1783 		dev_err(dmac->dev, "failed to request IRQ %u (%d)\n",
1784 			rchan->irq, ret);
1785 		return ret;
1786 	}
1787 
1788 	return 0;
1789 }
1790 
1791 #define RCAR_DMAC_MAX_CHANNELS	32
1792 
1793 static int rcar_dmac_parse_of(struct device *dev, struct rcar_dmac *dmac)
1794 {
1795 	struct device_node *np = dev->of_node;
1796 	int ret;
1797 
1798 	ret = of_property_read_u32(np, "dma-channels", &dmac->n_channels);
1799 	if (ret < 0) {
1800 		dev_err(dev, "unable to read dma-channels property\n");
1801 		return ret;
1802 	}
1803 
1804 	/* The hardware and driver don't support more than 32 bits in CHCLR */
1805 	if (dmac->n_channels <= 0 ||
1806 	    dmac->n_channels >= RCAR_DMAC_MAX_CHANNELS) {
1807 		dev_err(dev, "invalid number of channels %u\n",
1808 			dmac->n_channels);
1809 		return -EINVAL;
1810 	}
1811 
1812 	/*
1813 	 * If the driver is unable to read dma-channel-mask property,
1814 	 * the driver assumes that it can use all channels.
1815 	 */
1816 	dmac->channels_mask = GENMASK(dmac->n_channels - 1, 0);
1817 	of_property_read_u32(np, "dma-channel-mask", &dmac->channels_mask);
1818 
1819 	/* If the property has out-of-channel mask, this driver clears it */
1820 	dmac->channels_mask &= GENMASK(dmac->n_channels - 1, 0);
1821 
1822 	return 0;
1823 }
1824 
1825 static int rcar_dmac_probe(struct platform_device *pdev)
1826 {
1827 	const enum dma_slave_buswidth widths = DMA_SLAVE_BUSWIDTH_1_BYTE |
1828 		DMA_SLAVE_BUSWIDTH_2_BYTES | DMA_SLAVE_BUSWIDTH_4_BYTES |
1829 		DMA_SLAVE_BUSWIDTH_8_BYTES | DMA_SLAVE_BUSWIDTH_16_BYTES |
1830 		DMA_SLAVE_BUSWIDTH_32_BYTES | DMA_SLAVE_BUSWIDTH_64_BYTES;
1831 	struct dma_device *engine;
1832 	struct rcar_dmac *dmac;
1833 	const struct rcar_dmac_of_data *data;
1834 	unsigned int i;
1835 	int ret;
1836 
1837 	data = of_device_get_match_data(&pdev->dev);
1838 	if (!data)
1839 		return -EINVAL;
1840 
1841 	dmac = devm_kzalloc(&pdev->dev, sizeof(*dmac), GFP_KERNEL);
1842 	if (!dmac)
1843 		return -ENOMEM;
1844 
1845 	dmac->dev = &pdev->dev;
1846 	platform_set_drvdata(pdev, dmac);
1847 	dma_set_max_seg_size(dmac->dev, RCAR_DMATCR_MASK);
1848 	dma_set_mask_and_coherent(dmac->dev, DMA_BIT_MASK(40));
1849 
1850 	ret = rcar_dmac_parse_of(&pdev->dev, dmac);
1851 	if (ret < 0)
1852 		return ret;
1853 
1854 	/*
1855 	 * A still unconfirmed hardware bug prevents the IPMMU microTLB 0 to be
1856 	 * flushed correctly, resulting in memory corruption. DMAC 0 channel 0
1857 	 * is connected to microTLB 0 on currently supported platforms, so we
1858 	 * can't use it with the IPMMU. As the IOMMU API operates at the device
1859 	 * level we can't disable it selectively, so ignore channel 0 for now if
1860 	 * the device is part of an IOMMU group.
1861 	 */
1862 	if (device_iommu_mapped(&pdev->dev))
1863 		dmac->channels_mask &= ~BIT(0);
1864 
1865 	dmac->channels = devm_kcalloc(&pdev->dev, dmac->n_channels,
1866 				      sizeof(*dmac->channels), GFP_KERNEL);
1867 	if (!dmac->channels)
1868 		return -ENOMEM;
1869 
1870 	/* Request resources. */
1871 	dmac->iomem = devm_platform_ioremap_resource(pdev, 0);
1872 	if (IS_ERR(dmac->iomem))
1873 		return PTR_ERR(dmac->iomem);
1874 
1875 	/* Enable runtime PM and initialize the device. */
1876 	pm_runtime_enable(&pdev->dev);
1877 	ret = pm_runtime_get_sync(&pdev->dev);
1878 	if (ret < 0) {
1879 		dev_err(&pdev->dev, "runtime PM get sync failed (%d)\n", ret);
1880 		return ret;
1881 	}
1882 
1883 	ret = rcar_dmac_init(dmac);
1884 	pm_runtime_put(&pdev->dev);
1885 
1886 	if (ret) {
1887 		dev_err(&pdev->dev, "failed to reset device\n");
1888 		goto error;
1889 	}
1890 
1891 	/* Initialize engine */
1892 	engine = &dmac->engine;
1893 
1894 	dma_cap_set(DMA_MEMCPY, engine->cap_mask);
1895 	dma_cap_set(DMA_SLAVE, engine->cap_mask);
1896 
1897 	engine->dev		= &pdev->dev;
1898 	engine->copy_align	= ilog2(RCAR_DMAC_MEMCPY_XFER_SIZE);
1899 
1900 	engine->src_addr_widths	= widths;
1901 	engine->dst_addr_widths	= widths;
1902 	engine->directions	= BIT(DMA_MEM_TO_DEV) | BIT(DMA_DEV_TO_MEM);
1903 	engine->residue_granularity = DMA_RESIDUE_GRANULARITY_BURST;
1904 
1905 	engine->device_alloc_chan_resources	= rcar_dmac_alloc_chan_resources;
1906 	engine->device_free_chan_resources	= rcar_dmac_free_chan_resources;
1907 	engine->device_prep_dma_memcpy		= rcar_dmac_prep_dma_memcpy;
1908 	engine->device_prep_slave_sg		= rcar_dmac_prep_slave_sg;
1909 	engine->device_prep_dma_cyclic		= rcar_dmac_prep_dma_cyclic;
1910 	engine->device_config			= rcar_dmac_device_config;
1911 	engine->device_pause			= rcar_dmac_chan_pause;
1912 	engine->device_terminate_all		= rcar_dmac_chan_terminate_all;
1913 	engine->device_tx_status		= rcar_dmac_tx_status;
1914 	engine->device_issue_pending		= rcar_dmac_issue_pending;
1915 	engine->device_synchronize		= rcar_dmac_device_synchronize;
1916 
1917 	INIT_LIST_HEAD(&engine->channels);
1918 
1919 	for (i = 0; i < dmac->n_channels; ++i) {
1920 		if (!(dmac->channels_mask & BIT(i)))
1921 			continue;
1922 
1923 		ret = rcar_dmac_chan_probe(dmac, &dmac->channels[i], data, i);
1924 		if (ret < 0)
1925 			goto error;
1926 	}
1927 
1928 	/* Register the DMAC as a DMA provider for DT. */
1929 	ret = of_dma_controller_register(pdev->dev.of_node, rcar_dmac_of_xlate,
1930 					 NULL);
1931 	if (ret < 0)
1932 		goto error;
1933 
1934 	/*
1935 	 * Register the DMA engine device.
1936 	 *
1937 	 * Default transfer size of 32 bytes requires 32-byte alignment.
1938 	 */
1939 	ret = dma_async_device_register(engine);
1940 	if (ret < 0)
1941 		goto error;
1942 
1943 	return 0;
1944 
1945 error:
1946 	of_dma_controller_free(pdev->dev.of_node);
1947 	pm_runtime_disable(&pdev->dev);
1948 	return ret;
1949 }
1950 
1951 static int rcar_dmac_remove(struct platform_device *pdev)
1952 {
1953 	struct rcar_dmac *dmac = platform_get_drvdata(pdev);
1954 
1955 	of_dma_controller_free(pdev->dev.of_node);
1956 	dma_async_device_unregister(&dmac->engine);
1957 
1958 	pm_runtime_disable(&pdev->dev);
1959 
1960 	return 0;
1961 }
1962 
1963 static void rcar_dmac_shutdown(struct platform_device *pdev)
1964 {
1965 	struct rcar_dmac *dmac = platform_get_drvdata(pdev);
1966 
1967 	rcar_dmac_stop_all_chan(dmac);
1968 }
1969 
1970 static const struct rcar_dmac_of_data rcar_dmac_data = {
1971 	.chan_offset_base = 0x8000,
1972 	.chan_offset_stride = 0x80,
1973 };
1974 
1975 static const struct of_device_id rcar_dmac_of_ids[] = {
1976 	{
1977 		.compatible = "renesas,rcar-dmac",
1978 		.data = &rcar_dmac_data,
1979 	},
1980 	{ /* Sentinel */ }
1981 };
1982 MODULE_DEVICE_TABLE(of, rcar_dmac_of_ids);
1983 
1984 static struct platform_driver rcar_dmac_driver = {
1985 	.driver		= {
1986 		.pm	= &rcar_dmac_pm,
1987 		.name	= "rcar-dmac",
1988 		.of_match_table = rcar_dmac_of_ids,
1989 	},
1990 	.probe		= rcar_dmac_probe,
1991 	.remove		= rcar_dmac_remove,
1992 	.shutdown	= rcar_dmac_shutdown,
1993 };
1994 
1995 module_platform_driver(rcar_dmac_driver);
1996 
1997 MODULE_DESCRIPTION("R-Car Gen2 DMA Controller Driver");
1998 MODULE_AUTHOR("Laurent Pinchart <laurent.pinchart@ideasonboard.com>");
1999 MODULE_LICENSE("GPL v2");
2000