xref: /openbmc/linux/drivers/dma/sh/rcar-dmac.c (revision cbdf59ad)
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  * @modules: bitmask of client modules in use
196  */
197 struct rcar_dmac {
198 	struct dma_device engine;
199 	struct device *dev;
200 	void __iomem *iomem;
201 	struct device_dma_parameters parms;
202 
203 	unsigned int n_channels;
204 	struct rcar_dmac_chan *channels;
205 
206 	DECLARE_BITMAP(modules, 256);
207 };
208 
209 #define to_rcar_dmac(d)		container_of(d, struct rcar_dmac, engine)
210 
211 /* -----------------------------------------------------------------------------
212  * Registers
213  */
214 
215 #define RCAR_DMAC_CHAN_OFFSET(i)	(0x8000 + 0x80 * (i))
216 
217 #define RCAR_DMAISTA			0x0020
218 #define RCAR_DMASEC			0x0030
219 #define RCAR_DMAOR			0x0060
220 #define RCAR_DMAOR_PRI_FIXED		(0 << 8)
221 #define RCAR_DMAOR_PRI_ROUND_ROBIN	(3 << 8)
222 #define RCAR_DMAOR_AE			(1 << 2)
223 #define RCAR_DMAOR_DME			(1 << 0)
224 #define RCAR_DMACHCLR			0x0080
225 #define RCAR_DMADPSEC			0x00a0
226 
227 #define RCAR_DMASAR			0x0000
228 #define RCAR_DMADAR			0x0004
229 #define RCAR_DMATCR			0x0008
230 #define RCAR_DMATCR_MASK		0x00ffffff
231 #define RCAR_DMATSR			0x0028
232 #define RCAR_DMACHCR			0x000c
233 #define RCAR_DMACHCR_CAE		(1 << 31)
234 #define RCAR_DMACHCR_CAIE		(1 << 30)
235 #define RCAR_DMACHCR_DPM_DISABLED	(0 << 28)
236 #define RCAR_DMACHCR_DPM_ENABLED	(1 << 28)
237 #define RCAR_DMACHCR_DPM_REPEAT		(2 << 28)
238 #define RCAR_DMACHCR_DPM_INFINITE	(3 << 28)
239 #define RCAR_DMACHCR_RPT_SAR		(1 << 27)
240 #define RCAR_DMACHCR_RPT_DAR		(1 << 26)
241 #define RCAR_DMACHCR_RPT_TCR		(1 << 25)
242 #define RCAR_DMACHCR_DPB		(1 << 22)
243 #define RCAR_DMACHCR_DSE		(1 << 19)
244 #define RCAR_DMACHCR_DSIE		(1 << 18)
245 #define RCAR_DMACHCR_TS_1B		((0 << 20) | (0 << 3))
246 #define RCAR_DMACHCR_TS_2B		((0 << 20) | (1 << 3))
247 #define RCAR_DMACHCR_TS_4B		((0 << 20) | (2 << 3))
248 #define RCAR_DMACHCR_TS_16B		((0 << 20) | (3 << 3))
249 #define RCAR_DMACHCR_TS_32B		((1 << 20) | (0 << 3))
250 #define RCAR_DMACHCR_TS_64B		((1 << 20) | (1 << 3))
251 #define RCAR_DMACHCR_TS_8B		((1 << 20) | (3 << 3))
252 #define RCAR_DMACHCR_DM_FIXED		(0 << 14)
253 #define RCAR_DMACHCR_DM_INC		(1 << 14)
254 #define RCAR_DMACHCR_DM_DEC		(2 << 14)
255 #define RCAR_DMACHCR_SM_FIXED		(0 << 12)
256 #define RCAR_DMACHCR_SM_INC		(1 << 12)
257 #define RCAR_DMACHCR_SM_DEC		(2 << 12)
258 #define RCAR_DMACHCR_RS_AUTO		(4 << 8)
259 #define RCAR_DMACHCR_RS_DMARS		(8 << 8)
260 #define RCAR_DMACHCR_IE			(1 << 2)
261 #define RCAR_DMACHCR_TE			(1 << 1)
262 #define RCAR_DMACHCR_DE			(1 << 0)
263 #define RCAR_DMATCRB			0x0018
264 #define RCAR_DMATSRB			0x0038
265 #define RCAR_DMACHCRB			0x001c
266 #define RCAR_DMACHCRB_DCNT(n)		((n) << 24)
267 #define RCAR_DMACHCRB_DPTR_MASK		(0xff << 16)
268 #define RCAR_DMACHCRB_DPTR_SHIFT	16
269 #define RCAR_DMACHCRB_DRST		(1 << 15)
270 #define RCAR_DMACHCRB_DTS		(1 << 8)
271 #define RCAR_DMACHCRB_SLM_NORMAL	(0 << 4)
272 #define RCAR_DMACHCRB_SLM_CLK(n)	((8 | (n)) << 4)
273 #define RCAR_DMACHCRB_PRI(n)		((n) << 0)
274 #define RCAR_DMARS			0x0040
275 #define RCAR_DMABUFCR			0x0048
276 #define RCAR_DMABUFCR_MBU(n)		((n) << 16)
277 #define RCAR_DMABUFCR_ULB(n)		((n) << 0)
278 #define RCAR_DMADPBASE			0x0050
279 #define RCAR_DMADPBASE_MASK		0xfffffff0
280 #define RCAR_DMADPBASE_SEL		(1 << 0)
281 #define RCAR_DMADPCR			0x0054
282 #define RCAR_DMADPCR_DIPT(n)		((n) << 24)
283 #define RCAR_DMAFIXSAR			0x0010
284 #define RCAR_DMAFIXDAR			0x0014
285 #define RCAR_DMAFIXDPBASE		0x0060
286 
287 /* Hardcode the MEMCPY transfer size to 4 bytes. */
288 #define RCAR_DMAC_MEMCPY_XFER_SIZE	4
289 
290 /* -----------------------------------------------------------------------------
291  * Device access
292  */
293 
294 static void rcar_dmac_write(struct rcar_dmac *dmac, u32 reg, u32 data)
295 {
296 	if (reg == RCAR_DMAOR)
297 		writew(data, dmac->iomem + reg);
298 	else
299 		writel(data, dmac->iomem + reg);
300 }
301 
302 static u32 rcar_dmac_read(struct rcar_dmac *dmac, u32 reg)
303 {
304 	if (reg == RCAR_DMAOR)
305 		return readw(dmac->iomem + reg);
306 	else
307 		return readl(dmac->iomem + reg);
308 }
309 
310 static u32 rcar_dmac_chan_read(struct rcar_dmac_chan *chan, u32 reg)
311 {
312 	if (reg == RCAR_DMARS)
313 		return readw(chan->iomem + reg);
314 	else
315 		return readl(chan->iomem + reg);
316 }
317 
318 static void rcar_dmac_chan_write(struct rcar_dmac_chan *chan, u32 reg, u32 data)
319 {
320 	if (reg == RCAR_DMARS)
321 		writew(data, chan->iomem + reg);
322 	else
323 		writel(data, chan->iomem + reg);
324 }
325 
326 /* -----------------------------------------------------------------------------
327  * Initialization and configuration
328  */
329 
330 static bool rcar_dmac_chan_is_busy(struct rcar_dmac_chan *chan)
331 {
332 	u32 chcr = rcar_dmac_chan_read(chan, RCAR_DMACHCR);
333 
334 	return !!(chcr & (RCAR_DMACHCR_DE | RCAR_DMACHCR_TE));
335 }
336 
337 static void rcar_dmac_chan_start_xfer(struct rcar_dmac_chan *chan)
338 {
339 	struct rcar_dmac_desc *desc = chan->desc.running;
340 	u32 chcr = desc->chcr;
341 
342 	WARN_ON_ONCE(rcar_dmac_chan_is_busy(chan));
343 
344 	if (chan->mid_rid >= 0)
345 		rcar_dmac_chan_write(chan, RCAR_DMARS, chan->mid_rid);
346 
347 	if (desc->hwdescs.use) {
348 		struct rcar_dmac_xfer_chunk *chunk =
349 			list_first_entry(&desc->chunks,
350 					 struct rcar_dmac_xfer_chunk, node);
351 
352 		dev_dbg(chan->chan.device->dev,
353 			"chan%u: queue desc %p: %u@%pad\n",
354 			chan->index, desc, desc->nchunks, &desc->hwdescs.dma);
355 
356 #ifdef CONFIG_ARCH_DMA_ADDR_T_64BIT
357 		rcar_dmac_chan_write(chan, RCAR_DMAFIXSAR,
358 				     chunk->src_addr >> 32);
359 		rcar_dmac_chan_write(chan, RCAR_DMAFIXDAR,
360 				     chunk->dst_addr >> 32);
361 		rcar_dmac_chan_write(chan, RCAR_DMAFIXDPBASE,
362 				     desc->hwdescs.dma >> 32);
363 #endif
364 		rcar_dmac_chan_write(chan, RCAR_DMADPBASE,
365 				     (desc->hwdescs.dma & 0xfffffff0) |
366 				     RCAR_DMADPBASE_SEL);
367 		rcar_dmac_chan_write(chan, RCAR_DMACHCRB,
368 				     RCAR_DMACHCRB_DCNT(desc->nchunks - 1) |
369 				     RCAR_DMACHCRB_DRST);
370 
371 		/*
372 		 * Errata: When descriptor memory is accessed through an IOMMU
373 		 * the DMADAR register isn't initialized automatically from the
374 		 * first descriptor at beginning of transfer by the DMAC like it
375 		 * should. Initialize it manually with the destination address
376 		 * of the first chunk.
377 		 */
378 		rcar_dmac_chan_write(chan, RCAR_DMADAR,
379 				     chunk->dst_addr & 0xffffffff);
380 
381 		/*
382 		 * Program the descriptor stage interrupt to occur after the end
383 		 * of the first stage.
384 		 */
385 		rcar_dmac_chan_write(chan, RCAR_DMADPCR, RCAR_DMADPCR_DIPT(1));
386 
387 		chcr |= RCAR_DMACHCR_RPT_SAR | RCAR_DMACHCR_RPT_DAR
388 		     |  RCAR_DMACHCR_RPT_TCR | RCAR_DMACHCR_DPB;
389 
390 		/*
391 		 * If the descriptor isn't cyclic enable normal descriptor mode
392 		 * and the transfer completion interrupt.
393 		 */
394 		if (!desc->cyclic)
395 			chcr |= RCAR_DMACHCR_DPM_ENABLED | RCAR_DMACHCR_IE;
396 		/*
397 		 * If the descriptor is cyclic and has a callback enable the
398 		 * descriptor stage interrupt in infinite repeat mode.
399 		 */
400 		else if (desc->async_tx.callback)
401 			chcr |= RCAR_DMACHCR_DPM_INFINITE | RCAR_DMACHCR_DSIE;
402 		/*
403 		 * Otherwise just select infinite repeat mode without any
404 		 * interrupt.
405 		 */
406 		else
407 			chcr |= RCAR_DMACHCR_DPM_INFINITE;
408 	} else {
409 		struct rcar_dmac_xfer_chunk *chunk = desc->running;
410 
411 		dev_dbg(chan->chan.device->dev,
412 			"chan%u: queue chunk %p: %u@%pad -> %pad\n",
413 			chan->index, chunk, chunk->size, &chunk->src_addr,
414 			&chunk->dst_addr);
415 
416 #ifdef CONFIG_ARCH_DMA_ADDR_T_64BIT
417 		rcar_dmac_chan_write(chan, RCAR_DMAFIXSAR,
418 				     chunk->src_addr >> 32);
419 		rcar_dmac_chan_write(chan, RCAR_DMAFIXDAR,
420 				     chunk->dst_addr >> 32);
421 #endif
422 		rcar_dmac_chan_write(chan, RCAR_DMASAR,
423 				     chunk->src_addr & 0xffffffff);
424 		rcar_dmac_chan_write(chan, RCAR_DMADAR,
425 				     chunk->dst_addr & 0xffffffff);
426 		rcar_dmac_chan_write(chan, RCAR_DMATCR,
427 				     chunk->size >> desc->xfer_shift);
428 
429 		chcr |= RCAR_DMACHCR_DPM_DISABLED | RCAR_DMACHCR_IE;
430 	}
431 
432 	rcar_dmac_chan_write(chan, RCAR_DMACHCR,
433 			     chcr | RCAR_DMACHCR_DE | RCAR_DMACHCR_CAIE);
434 }
435 
436 static int rcar_dmac_init(struct rcar_dmac *dmac)
437 {
438 	u16 dmaor;
439 
440 	/* Clear all channels and enable the DMAC globally. */
441 	rcar_dmac_write(dmac, RCAR_DMACHCLR, GENMASK(dmac->n_channels - 1, 0));
442 	rcar_dmac_write(dmac, RCAR_DMAOR,
443 			RCAR_DMAOR_PRI_FIXED | RCAR_DMAOR_DME);
444 
445 	dmaor = rcar_dmac_read(dmac, RCAR_DMAOR);
446 	if ((dmaor & (RCAR_DMAOR_AE | RCAR_DMAOR_DME)) != RCAR_DMAOR_DME) {
447 		dev_warn(dmac->dev, "DMAOR initialization failed.\n");
448 		return -EIO;
449 	}
450 
451 	return 0;
452 }
453 
454 /* -----------------------------------------------------------------------------
455  * Descriptors submission
456  */
457 
458 static dma_cookie_t rcar_dmac_tx_submit(struct dma_async_tx_descriptor *tx)
459 {
460 	struct rcar_dmac_chan *chan = to_rcar_dmac_chan(tx->chan);
461 	struct rcar_dmac_desc *desc = to_rcar_dmac_desc(tx);
462 	unsigned long flags;
463 	dma_cookie_t cookie;
464 
465 	spin_lock_irqsave(&chan->lock, flags);
466 
467 	cookie = dma_cookie_assign(tx);
468 
469 	dev_dbg(chan->chan.device->dev, "chan%u: submit #%d@%p\n",
470 		chan->index, tx->cookie, desc);
471 
472 	list_add_tail(&desc->node, &chan->desc.pending);
473 	desc->running = list_first_entry(&desc->chunks,
474 					 struct rcar_dmac_xfer_chunk, node);
475 
476 	spin_unlock_irqrestore(&chan->lock, flags);
477 
478 	return cookie;
479 }
480 
481 /* -----------------------------------------------------------------------------
482  * Descriptors allocation and free
483  */
484 
485 /*
486  * rcar_dmac_desc_alloc - Allocate a page worth of DMA descriptors
487  * @chan: the DMA channel
488  * @gfp: allocation flags
489  */
490 static int rcar_dmac_desc_alloc(struct rcar_dmac_chan *chan, gfp_t gfp)
491 {
492 	struct rcar_dmac_desc_page *page;
493 	unsigned long flags;
494 	LIST_HEAD(list);
495 	unsigned int i;
496 
497 	page = (void *)get_zeroed_page(gfp);
498 	if (!page)
499 		return -ENOMEM;
500 
501 	for (i = 0; i < RCAR_DMAC_DESCS_PER_PAGE; ++i) {
502 		struct rcar_dmac_desc *desc = &page->descs[i];
503 
504 		dma_async_tx_descriptor_init(&desc->async_tx, &chan->chan);
505 		desc->async_tx.tx_submit = rcar_dmac_tx_submit;
506 		INIT_LIST_HEAD(&desc->chunks);
507 
508 		list_add_tail(&desc->node, &list);
509 	}
510 
511 	spin_lock_irqsave(&chan->lock, flags);
512 	list_splice_tail(&list, &chan->desc.free);
513 	list_add_tail(&page->node, &chan->desc.pages);
514 	spin_unlock_irqrestore(&chan->lock, flags);
515 
516 	return 0;
517 }
518 
519 /*
520  * rcar_dmac_desc_put - Release a DMA transfer descriptor
521  * @chan: the DMA channel
522  * @desc: the descriptor
523  *
524  * Put the descriptor and its transfer chunk descriptors back in the channel's
525  * free descriptors lists. The descriptor's chunks list will be reinitialized to
526  * an empty list as a result.
527  *
528  * The descriptor must have been removed from the channel's lists before calling
529  * this function.
530  */
531 static void rcar_dmac_desc_put(struct rcar_dmac_chan *chan,
532 			       struct rcar_dmac_desc *desc)
533 {
534 	unsigned long flags;
535 
536 	spin_lock_irqsave(&chan->lock, flags);
537 	list_splice_tail_init(&desc->chunks, &chan->desc.chunks_free);
538 	list_add(&desc->node, &chan->desc.free);
539 	spin_unlock_irqrestore(&chan->lock, flags);
540 }
541 
542 static void rcar_dmac_desc_recycle_acked(struct rcar_dmac_chan *chan)
543 {
544 	struct rcar_dmac_desc *desc, *_desc;
545 	unsigned long flags;
546 	LIST_HEAD(list);
547 
548 	/*
549 	 * We have to temporarily move all descriptors from the wait list to a
550 	 * local list as iterating over the wait list, even with
551 	 * list_for_each_entry_safe, isn't safe if we release the channel lock
552 	 * around the rcar_dmac_desc_put() call.
553 	 */
554 	spin_lock_irqsave(&chan->lock, flags);
555 	list_splice_init(&chan->desc.wait, &list);
556 	spin_unlock_irqrestore(&chan->lock, flags);
557 
558 	list_for_each_entry_safe(desc, _desc, &list, node) {
559 		if (async_tx_test_ack(&desc->async_tx)) {
560 			list_del(&desc->node);
561 			rcar_dmac_desc_put(chan, desc);
562 		}
563 	}
564 
565 	if (list_empty(&list))
566 		return;
567 
568 	/* Put the remaining descriptors back in the wait list. */
569 	spin_lock_irqsave(&chan->lock, flags);
570 	list_splice(&list, &chan->desc.wait);
571 	spin_unlock_irqrestore(&chan->lock, flags);
572 }
573 
574 /*
575  * rcar_dmac_desc_get - Allocate a descriptor for a DMA transfer
576  * @chan: the DMA channel
577  *
578  * Locking: This function must be called in a non-atomic context.
579  *
580  * Return: A pointer to the allocated descriptor or NULL if no descriptor can
581  * be allocated.
582  */
583 static struct rcar_dmac_desc *rcar_dmac_desc_get(struct rcar_dmac_chan *chan)
584 {
585 	struct rcar_dmac_desc *desc;
586 	unsigned long flags;
587 	int ret;
588 
589 	/* Recycle acked descriptors before attempting allocation. */
590 	rcar_dmac_desc_recycle_acked(chan);
591 
592 	spin_lock_irqsave(&chan->lock, flags);
593 
594 	while (list_empty(&chan->desc.free)) {
595 		/*
596 		 * No free descriptors, allocate a page worth of them and try
597 		 * again, as someone else could race us to get the newly
598 		 * allocated descriptors. If the allocation fails return an
599 		 * error.
600 		 */
601 		spin_unlock_irqrestore(&chan->lock, flags);
602 		ret = rcar_dmac_desc_alloc(chan, GFP_NOWAIT);
603 		if (ret < 0)
604 			return NULL;
605 		spin_lock_irqsave(&chan->lock, flags);
606 	}
607 
608 	desc = list_first_entry(&chan->desc.free, struct rcar_dmac_desc, node);
609 	list_del(&desc->node);
610 
611 	spin_unlock_irqrestore(&chan->lock, flags);
612 
613 	return desc;
614 }
615 
616 /*
617  * rcar_dmac_xfer_chunk_alloc - Allocate a page worth of transfer chunks
618  * @chan: the DMA channel
619  * @gfp: allocation flags
620  */
621 static int rcar_dmac_xfer_chunk_alloc(struct rcar_dmac_chan *chan, gfp_t gfp)
622 {
623 	struct rcar_dmac_desc_page *page;
624 	unsigned long flags;
625 	LIST_HEAD(list);
626 	unsigned int i;
627 
628 	page = (void *)get_zeroed_page(gfp);
629 	if (!page)
630 		return -ENOMEM;
631 
632 	for (i = 0; i < RCAR_DMAC_XFER_CHUNKS_PER_PAGE; ++i) {
633 		struct rcar_dmac_xfer_chunk *chunk = &page->chunks[i];
634 
635 		list_add_tail(&chunk->node, &list);
636 	}
637 
638 	spin_lock_irqsave(&chan->lock, flags);
639 	list_splice_tail(&list, &chan->desc.chunks_free);
640 	list_add_tail(&page->node, &chan->desc.pages);
641 	spin_unlock_irqrestore(&chan->lock, flags);
642 
643 	return 0;
644 }
645 
646 /*
647  * rcar_dmac_xfer_chunk_get - Allocate a transfer chunk for a DMA transfer
648  * @chan: the DMA channel
649  *
650  * Locking: This function must be called in a non-atomic context.
651  *
652  * Return: A pointer to the allocated transfer chunk descriptor or NULL if no
653  * descriptor can be allocated.
654  */
655 static struct rcar_dmac_xfer_chunk *
656 rcar_dmac_xfer_chunk_get(struct rcar_dmac_chan *chan)
657 {
658 	struct rcar_dmac_xfer_chunk *chunk;
659 	unsigned long flags;
660 	int ret;
661 
662 	spin_lock_irqsave(&chan->lock, flags);
663 
664 	while (list_empty(&chan->desc.chunks_free)) {
665 		/*
666 		 * No free descriptors, allocate a page worth of them and try
667 		 * again, as someone else could race us to get the newly
668 		 * allocated descriptors. If the allocation fails return an
669 		 * error.
670 		 */
671 		spin_unlock_irqrestore(&chan->lock, flags);
672 		ret = rcar_dmac_xfer_chunk_alloc(chan, GFP_NOWAIT);
673 		if (ret < 0)
674 			return NULL;
675 		spin_lock_irqsave(&chan->lock, flags);
676 	}
677 
678 	chunk = list_first_entry(&chan->desc.chunks_free,
679 				 struct rcar_dmac_xfer_chunk, node);
680 	list_del(&chunk->node);
681 
682 	spin_unlock_irqrestore(&chan->lock, flags);
683 
684 	return chunk;
685 }
686 
687 static void rcar_dmac_realloc_hwdesc(struct rcar_dmac_chan *chan,
688 				     struct rcar_dmac_desc *desc, size_t size)
689 {
690 	/*
691 	 * dma_alloc_coherent() allocates memory in page size increments. To
692 	 * avoid reallocating the hardware descriptors when the allocated size
693 	 * wouldn't change align the requested size to a multiple of the page
694 	 * size.
695 	 */
696 	size = PAGE_ALIGN(size);
697 
698 	if (desc->hwdescs.size == size)
699 		return;
700 
701 	if (desc->hwdescs.mem) {
702 		dma_free_coherent(chan->chan.device->dev, desc->hwdescs.size,
703 				  desc->hwdescs.mem, desc->hwdescs.dma);
704 		desc->hwdescs.mem = NULL;
705 		desc->hwdescs.size = 0;
706 	}
707 
708 	if (!size)
709 		return;
710 
711 	desc->hwdescs.mem = dma_alloc_coherent(chan->chan.device->dev, size,
712 					       &desc->hwdescs.dma, GFP_NOWAIT);
713 	if (!desc->hwdescs.mem)
714 		return;
715 
716 	desc->hwdescs.size = size;
717 }
718 
719 static int rcar_dmac_fill_hwdesc(struct rcar_dmac_chan *chan,
720 				 struct rcar_dmac_desc *desc)
721 {
722 	struct rcar_dmac_xfer_chunk *chunk;
723 	struct rcar_dmac_hw_desc *hwdesc;
724 
725 	rcar_dmac_realloc_hwdesc(chan, desc, desc->nchunks * sizeof(*hwdesc));
726 
727 	hwdesc = desc->hwdescs.mem;
728 	if (!hwdesc)
729 		return -ENOMEM;
730 
731 	list_for_each_entry(chunk, &desc->chunks, node) {
732 		hwdesc->sar = chunk->src_addr;
733 		hwdesc->dar = chunk->dst_addr;
734 		hwdesc->tcr = chunk->size >> desc->xfer_shift;
735 		hwdesc++;
736 	}
737 
738 	return 0;
739 }
740 
741 /* -----------------------------------------------------------------------------
742  * Stop and reset
743  */
744 static void rcar_dmac_chcr_de_barrier(struct rcar_dmac_chan *chan)
745 {
746 	u32 chcr;
747 	unsigned int i;
748 
749 	/*
750 	 * Ensure that the setting of the DE bit is actually 0 after
751 	 * clearing it.
752 	 */
753 	for (i = 0; i < 1024; i++) {
754 		chcr = rcar_dmac_chan_read(chan, RCAR_DMACHCR);
755 		if (!(chcr & RCAR_DMACHCR_DE))
756 			return;
757 		udelay(1);
758 	}
759 
760 	dev_err(chan->chan.device->dev, "CHCR DE check error\n");
761 }
762 
763 static void rcar_dmac_clear_chcr_de(struct rcar_dmac_chan *chan)
764 {
765 	u32 chcr = rcar_dmac_chan_read(chan, RCAR_DMACHCR);
766 
767 	/* set DE=0 and flush remaining data */
768 	rcar_dmac_chan_write(chan, RCAR_DMACHCR, (chcr & ~RCAR_DMACHCR_DE));
769 
770 	/* make sure all remaining data was flushed */
771 	rcar_dmac_chcr_de_barrier(chan);
772 }
773 
774 static void rcar_dmac_chan_halt(struct rcar_dmac_chan *chan)
775 {
776 	u32 chcr = rcar_dmac_chan_read(chan, RCAR_DMACHCR);
777 
778 	chcr &= ~(RCAR_DMACHCR_DSE | RCAR_DMACHCR_DSIE | RCAR_DMACHCR_IE |
779 		  RCAR_DMACHCR_TE | RCAR_DMACHCR_DE |
780 		  RCAR_DMACHCR_CAE | RCAR_DMACHCR_CAIE);
781 	rcar_dmac_chan_write(chan, RCAR_DMACHCR, chcr);
782 	rcar_dmac_chcr_de_barrier(chan);
783 }
784 
785 static void rcar_dmac_chan_reinit(struct rcar_dmac_chan *chan)
786 {
787 	struct rcar_dmac_desc *desc, *_desc;
788 	unsigned long flags;
789 	LIST_HEAD(descs);
790 
791 	spin_lock_irqsave(&chan->lock, flags);
792 
793 	/* Move all non-free descriptors to the local lists. */
794 	list_splice_init(&chan->desc.pending, &descs);
795 	list_splice_init(&chan->desc.active, &descs);
796 	list_splice_init(&chan->desc.done, &descs);
797 	list_splice_init(&chan->desc.wait, &descs);
798 
799 	chan->desc.running = NULL;
800 
801 	spin_unlock_irqrestore(&chan->lock, flags);
802 
803 	list_for_each_entry_safe(desc, _desc, &descs, node) {
804 		list_del(&desc->node);
805 		rcar_dmac_desc_put(chan, desc);
806 	}
807 }
808 
809 static void rcar_dmac_stop_all_chan(struct rcar_dmac *dmac)
810 {
811 	unsigned int i;
812 
813 	/* Stop all channels. */
814 	for (i = 0; i < dmac->n_channels; ++i) {
815 		struct rcar_dmac_chan *chan = &dmac->channels[i];
816 
817 		/* Stop and reinitialize the channel. */
818 		spin_lock_irq(&chan->lock);
819 		rcar_dmac_chan_halt(chan);
820 		spin_unlock_irq(&chan->lock);
821 	}
822 }
823 
824 static int rcar_dmac_chan_pause(struct dma_chan *chan)
825 {
826 	unsigned long flags;
827 	struct rcar_dmac_chan *rchan = to_rcar_dmac_chan(chan);
828 
829 	spin_lock_irqsave(&rchan->lock, flags);
830 	rcar_dmac_clear_chcr_de(rchan);
831 	spin_unlock_irqrestore(&rchan->lock, flags);
832 
833 	return 0;
834 }
835 
836 /* -----------------------------------------------------------------------------
837  * Descriptors preparation
838  */
839 
840 static void rcar_dmac_chan_configure_desc(struct rcar_dmac_chan *chan,
841 					  struct rcar_dmac_desc *desc)
842 {
843 	static const u32 chcr_ts[] = {
844 		RCAR_DMACHCR_TS_1B, RCAR_DMACHCR_TS_2B,
845 		RCAR_DMACHCR_TS_4B, RCAR_DMACHCR_TS_8B,
846 		RCAR_DMACHCR_TS_16B, RCAR_DMACHCR_TS_32B,
847 		RCAR_DMACHCR_TS_64B,
848 	};
849 
850 	unsigned int xfer_size;
851 	u32 chcr;
852 
853 	switch (desc->direction) {
854 	case DMA_DEV_TO_MEM:
855 		chcr = RCAR_DMACHCR_DM_INC | RCAR_DMACHCR_SM_FIXED
856 		     | RCAR_DMACHCR_RS_DMARS;
857 		xfer_size = chan->src.xfer_size;
858 		break;
859 
860 	case DMA_MEM_TO_DEV:
861 		chcr = RCAR_DMACHCR_DM_FIXED | RCAR_DMACHCR_SM_INC
862 		     | RCAR_DMACHCR_RS_DMARS;
863 		xfer_size = chan->dst.xfer_size;
864 		break;
865 
866 	case DMA_MEM_TO_MEM:
867 	default:
868 		chcr = RCAR_DMACHCR_DM_INC | RCAR_DMACHCR_SM_INC
869 		     | RCAR_DMACHCR_RS_AUTO;
870 		xfer_size = RCAR_DMAC_MEMCPY_XFER_SIZE;
871 		break;
872 	}
873 
874 	desc->xfer_shift = ilog2(xfer_size);
875 	desc->chcr = chcr | chcr_ts[desc->xfer_shift];
876 }
877 
878 /*
879  * rcar_dmac_chan_prep_sg - prepare transfer descriptors from an SG list
880  *
881  * Common routine for public (MEMCPY) and slave DMA. The MEMCPY case is also
882  * converted to scatter-gather to guarantee consistent locking and a correct
883  * list manipulation. For slave DMA direction carries the usual meaning, and,
884  * logically, the SG list is RAM and the addr variable contains slave address,
885  * e.g., the FIFO I/O register. For MEMCPY direction equals DMA_MEM_TO_MEM
886  * and the SG list contains only one element and points at the source buffer.
887  */
888 static struct dma_async_tx_descriptor *
889 rcar_dmac_chan_prep_sg(struct rcar_dmac_chan *chan, struct scatterlist *sgl,
890 		       unsigned int sg_len, dma_addr_t dev_addr,
891 		       enum dma_transfer_direction dir, unsigned long dma_flags,
892 		       bool cyclic)
893 {
894 	struct rcar_dmac_xfer_chunk *chunk;
895 	struct rcar_dmac_desc *desc;
896 	struct scatterlist *sg;
897 	unsigned int nchunks = 0;
898 	unsigned int max_chunk_size;
899 	unsigned int full_size = 0;
900 	bool cross_boundary = false;
901 	unsigned int i;
902 #ifdef CONFIG_ARCH_DMA_ADDR_T_64BIT
903 	u32 high_dev_addr;
904 	u32 high_mem_addr;
905 #endif
906 
907 	desc = rcar_dmac_desc_get(chan);
908 	if (!desc)
909 		return NULL;
910 
911 	desc->async_tx.flags = dma_flags;
912 	desc->async_tx.cookie = -EBUSY;
913 
914 	desc->cyclic = cyclic;
915 	desc->direction = dir;
916 
917 	rcar_dmac_chan_configure_desc(chan, desc);
918 
919 	max_chunk_size = RCAR_DMATCR_MASK << desc->xfer_shift;
920 
921 	/*
922 	 * Allocate and fill the transfer chunk descriptors. We own the only
923 	 * reference to the DMA descriptor, there's no need for locking.
924 	 */
925 	for_each_sg(sgl, sg, sg_len, i) {
926 		dma_addr_t mem_addr = sg_dma_address(sg);
927 		unsigned int len = sg_dma_len(sg);
928 
929 		full_size += len;
930 
931 #ifdef CONFIG_ARCH_DMA_ADDR_T_64BIT
932 		if (i == 0) {
933 			high_dev_addr = dev_addr >> 32;
934 			high_mem_addr = mem_addr >> 32;
935 		}
936 
937 		if ((dev_addr >> 32 != high_dev_addr) ||
938 		    (mem_addr >> 32 != high_mem_addr))
939 			cross_boundary = true;
940 #endif
941 		while (len) {
942 			unsigned int size = min(len, max_chunk_size);
943 
944 #ifdef CONFIG_ARCH_DMA_ADDR_T_64BIT
945 			/*
946 			 * Prevent individual transfers from crossing 4GB
947 			 * boundaries.
948 			 */
949 			if (dev_addr >> 32 != (dev_addr + size - 1) >> 32) {
950 				size = ALIGN(dev_addr, 1ULL << 32) - dev_addr;
951 				cross_boundary = true;
952 			}
953 			if (mem_addr >> 32 != (mem_addr + size - 1) >> 32) {
954 				size = ALIGN(mem_addr, 1ULL << 32) - mem_addr;
955 				cross_boundary = true;
956 			}
957 #endif
958 
959 			chunk = rcar_dmac_xfer_chunk_get(chan);
960 			if (!chunk) {
961 				rcar_dmac_desc_put(chan, desc);
962 				return NULL;
963 			}
964 
965 			if (dir == DMA_DEV_TO_MEM) {
966 				chunk->src_addr = dev_addr;
967 				chunk->dst_addr = mem_addr;
968 			} else {
969 				chunk->src_addr = mem_addr;
970 				chunk->dst_addr = dev_addr;
971 			}
972 
973 			chunk->size = size;
974 
975 			dev_dbg(chan->chan.device->dev,
976 				"chan%u: chunk %p/%p sgl %u@%p, %u/%u %pad -> %pad\n",
977 				chan->index, chunk, desc, i, sg, size, len,
978 				&chunk->src_addr, &chunk->dst_addr);
979 
980 			mem_addr += size;
981 			if (dir == DMA_MEM_TO_MEM)
982 				dev_addr += size;
983 
984 			len -= size;
985 
986 			list_add_tail(&chunk->node, &desc->chunks);
987 			nchunks++;
988 		}
989 	}
990 
991 	desc->nchunks = nchunks;
992 	desc->size = full_size;
993 
994 	/*
995 	 * Use hardware descriptor lists if possible when more than one chunk
996 	 * needs to be transferred (otherwise they don't make much sense).
997 	 *
998 	 * Source/Destination address should be located in same 4GiB region
999 	 * in the 40bit address space when it uses Hardware descriptor,
1000 	 * and cross_boundary is checking it.
1001 	 */
1002 	desc->hwdescs.use = !cross_boundary && nchunks > 1;
1003 	if (desc->hwdescs.use) {
1004 		if (rcar_dmac_fill_hwdesc(chan, desc) < 0)
1005 			desc->hwdescs.use = false;
1006 	}
1007 
1008 	return &desc->async_tx;
1009 }
1010 
1011 /* -----------------------------------------------------------------------------
1012  * DMA engine operations
1013  */
1014 
1015 static int rcar_dmac_alloc_chan_resources(struct dma_chan *chan)
1016 {
1017 	struct rcar_dmac_chan *rchan = to_rcar_dmac_chan(chan);
1018 	int ret;
1019 
1020 	INIT_LIST_HEAD(&rchan->desc.chunks_free);
1021 	INIT_LIST_HEAD(&rchan->desc.pages);
1022 
1023 	/* Preallocate descriptors. */
1024 	ret = rcar_dmac_xfer_chunk_alloc(rchan, GFP_KERNEL);
1025 	if (ret < 0)
1026 		return -ENOMEM;
1027 
1028 	ret = rcar_dmac_desc_alloc(rchan, GFP_KERNEL);
1029 	if (ret < 0)
1030 		return -ENOMEM;
1031 
1032 	return pm_runtime_get_sync(chan->device->dev);
1033 }
1034 
1035 static void rcar_dmac_free_chan_resources(struct dma_chan *chan)
1036 {
1037 	struct rcar_dmac_chan *rchan = to_rcar_dmac_chan(chan);
1038 	struct rcar_dmac *dmac = to_rcar_dmac(chan->device);
1039 	struct rcar_dmac_chan_map *map = &rchan->map;
1040 	struct rcar_dmac_desc_page *page, *_page;
1041 	struct rcar_dmac_desc *desc;
1042 	LIST_HEAD(list);
1043 
1044 	/* Protect against ISR */
1045 	spin_lock_irq(&rchan->lock);
1046 	rcar_dmac_chan_halt(rchan);
1047 	spin_unlock_irq(&rchan->lock);
1048 
1049 	/*
1050 	 * Now no new interrupts will occur, but one might already be
1051 	 * running. Wait for it to finish before freeing resources.
1052 	 */
1053 	synchronize_irq(rchan->irq);
1054 
1055 	if (rchan->mid_rid >= 0) {
1056 		/* The caller is holding dma_list_mutex */
1057 		clear_bit(rchan->mid_rid, dmac->modules);
1058 		rchan->mid_rid = -EINVAL;
1059 	}
1060 
1061 	list_splice_init(&rchan->desc.free, &list);
1062 	list_splice_init(&rchan->desc.pending, &list);
1063 	list_splice_init(&rchan->desc.active, &list);
1064 	list_splice_init(&rchan->desc.done, &list);
1065 	list_splice_init(&rchan->desc.wait, &list);
1066 
1067 	rchan->desc.running = NULL;
1068 
1069 	list_for_each_entry(desc, &list, node)
1070 		rcar_dmac_realloc_hwdesc(rchan, desc, 0);
1071 
1072 	list_for_each_entry_safe(page, _page, &rchan->desc.pages, node) {
1073 		list_del(&page->node);
1074 		free_page((unsigned long)page);
1075 	}
1076 
1077 	/* Remove slave mapping if present. */
1078 	if (map->slave.xfer_size) {
1079 		dma_unmap_resource(chan->device->dev, map->addr,
1080 				   map->slave.xfer_size, map->dir, 0);
1081 		map->slave.xfer_size = 0;
1082 	}
1083 
1084 	pm_runtime_put(chan->device->dev);
1085 }
1086 
1087 static struct dma_async_tx_descriptor *
1088 rcar_dmac_prep_dma_memcpy(struct dma_chan *chan, dma_addr_t dma_dest,
1089 			  dma_addr_t dma_src, size_t len, unsigned long flags)
1090 {
1091 	struct rcar_dmac_chan *rchan = to_rcar_dmac_chan(chan);
1092 	struct scatterlist sgl;
1093 
1094 	if (!len)
1095 		return NULL;
1096 
1097 	sg_init_table(&sgl, 1);
1098 	sg_set_page(&sgl, pfn_to_page(PFN_DOWN(dma_src)), len,
1099 		    offset_in_page(dma_src));
1100 	sg_dma_address(&sgl) = dma_src;
1101 	sg_dma_len(&sgl) = len;
1102 
1103 	return rcar_dmac_chan_prep_sg(rchan, &sgl, 1, dma_dest,
1104 				      DMA_MEM_TO_MEM, flags, false);
1105 }
1106 
1107 static int rcar_dmac_map_slave_addr(struct dma_chan *chan,
1108 				    enum dma_transfer_direction dir)
1109 {
1110 	struct rcar_dmac_chan *rchan = to_rcar_dmac_chan(chan);
1111 	struct rcar_dmac_chan_map *map = &rchan->map;
1112 	phys_addr_t dev_addr;
1113 	size_t dev_size;
1114 	enum dma_data_direction dev_dir;
1115 
1116 	if (dir == DMA_DEV_TO_MEM) {
1117 		dev_addr = rchan->src.slave_addr;
1118 		dev_size = rchan->src.xfer_size;
1119 		dev_dir = DMA_TO_DEVICE;
1120 	} else {
1121 		dev_addr = rchan->dst.slave_addr;
1122 		dev_size = rchan->dst.xfer_size;
1123 		dev_dir = DMA_FROM_DEVICE;
1124 	}
1125 
1126 	/* Reuse current map if possible. */
1127 	if (dev_addr == map->slave.slave_addr &&
1128 	    dev_size == map->slave.xfer_size &&
1129 	    dev_dir == map->dir)
1130 		return 0;
1131 
1132 	/* Remove old mapping if present. */
1133 	if (map->slave.xfer_size)
1134 		dma_unmap_resource(chan->device->dev, map->addr,
1135 				   map->slave.xfer_size, map->dir, 0);
1136 	map->slave.xfer_size = 0;
1137 
1138 	/* Create new slave address map. */
1139 	map->addr = dma_map_resource(chan->device->dev, dev_addr, dev_size,
1140 				     dev_dir, 0);
1141 
1142 	if (dma_mapping_error(chan->device->dev, map->addr)) {
1143 		dev_err(chan->device->dev,
1144 			"chan%u: failed to map %zx@%pap", rchan->index,
1145 			dev_size, &dev_addr);
1146 		return -EIO;
1147 	}
1148 
1149 	dev_dbg(chan->device->dev, "chan%u: map %zx@%pap to %pad dir: %s\n",
1150 		rchan->index, dev_size, &dev_addr, &map->addr,
1151 		dev_dir == DMA_TO_DEVICE ? "DMA_TO_DEVICE" : "DMA_FROM_DEVICE");
1152 
1153 	map->slave.slave_addr = dev_addr;
1154 	map->slave.xfer_size = dev_size;
1155 	map->dir = dev_dir;
1156 
1157 	return 0;
1158 }
1159 
1160 static struct dma_async_tx_descriptor *
1161 rcar_dmac_prep_slave_sg(struct dma_chan *chan, struct scatterlist *sgl,
1162 			unsigned int sg_len, enum dma_transfer_direction dir,
1163 			unsigned long flags, void *context)
1164 {
1165 	struct rcar_dmac_chan *rchan = to_rcar_dmac_chan(chan);
1166 
1167 	/* Someone calling slave DMA on a generic channel? */
1168 	if (rchan->mid_rid < 0 || !sg_len || !sg_dma_len(sgl)) {
1169 		dev_warn(chan->device->dev,
1170 			 "%s: bad parameter: len=%d, id=%d\n",
1171 			 __func__, sg_len, rchan->mid_rid);
1172 		return NULL;
1173 	}
1174 
1175 	if (rcar_dmac_map_slave_addr(chan, dir))
1176 		return NULL;
1177 
1178 	return rcar_dmac_chan_prep_sg(rchan, sgl, sg_len, rchan->map.addr,
1179 				      dir, flags, false);
1180 }
1181 
1182 #define RCAR_DMAC_MAX_SG_LEN	32
1183 
1184 static struct dma_async_tx_descriptor *
1185 rcar_dmac_prep_dma_cyclic(struct dma_chan *chan, dma_addr_t buf_addr,
1186 			  size_t buf_len, size_t period_len,
1187 			  enum dma_transfer_direction dir, unsigned long flags)
1188 {
1189 	struct rcar_dmac_chan *rchan = to_rcar_dmac_chan(chan);
1190 	struct dma_async_tx_descriptor *desc;
1191 	struct scatterlist *sgl;
1192 	unsigned int sg_len;
1193 	unsigned int i;
1194 
1195 	/* Someone calling slave DMA on a generic channel? */
1196 	if (rchan->mid_rid < 0 || buf_len < period_len) {
1197 		dev_warn(chan->device->dev,
1198 			"%s: bad parameter: buf_len=%zu, period_len=%zu, id=%d\n",
1199 			__func__, buf_len, period_len, rchan->mid_rid);
1200 		return NULL;
1201 	}
1202 
1203 	if (rcar_dmac_map_slave_addr(chan, dir))
1204 		return NULL;
1205 
1206 	sg_len = buf_len / period_len;
1207 	if (sg_len > RCAR_DMAC_MAX_SG_LEN) {
1208 		dev_err(chan->device->dev,
1209 			"chan%u: sg length %d exceds limit %d",
1210 			rchan->index, sg_len, RCAR_DMAC_MAX_SG_LEN);
1211 		return NULL;
1212 	}
1213 
1214 	/*
1215 	 * Allocate the sg list dynamically as it would consume too much stack
1216 	 * space.
1217 	 */
1218 	sgl = kcalloc(sg_len, sizeof(*sgl), GFP_NOWAIT);
1219 	if (!sgl)
1220 		return NULL;
1221 
1222 	sg_init_table(sgl, sg_len);
1223 
1224 	for (i = 0; i < sg_len; ++i) {
1225 		dma_addr_t src = buf_addr + (period_len * i);
1226 
1227 		sg_set_page(&sgl[i], pfn_to_page(PFN_DOWN(src)), period_len,
1228 			    offset_in_page(src));
1229 		sg_dma_address(&sgl[i]) = src;
1230 		sg_dma_len(&sgl[i]) = period_len;
1231 	}
1232 
1233 	desc = rcar_dmac_chan_prep_sg(rchan, sgl, sg_len, rchan->map.addr,
1234 				      dir, flags, true);
1235 
1236 	kfree(sgl);
1237 	return desc;
1238 }
1239 
1240 static int rcar_dmac_device_config(struct dma_chan *chan,
1241 				   struct dma_slave_config *cfg)
1242 {
1243 	struct rcar_dmac_chan *rchan = to_rcar_dmac_chan(chan);
1244 
1245 	/*
1246 	 * We could lock this, but you shouldn't be configuring the
1247 	 * channel, while using it...
1248 	 */
1249 	rchan->src.slave_addr = cfg->src_addr;
1250 	rchan->dst.slave_addr = cfg->dst_addr;
1251 	rchan->src.xfer_size = cfg->src_addr_width;
1252 	rchan->dst.xfer_size = cfg->dst_addr_width;
1253 
1254 	return 0;
1255 }
1256 
1257 static int rcar_dmac_chan_terminate_all(struct dma_chan *chan)
1258 {
1259 	struct rcar_dmac_chan *rchan = to_rcar_dmac_chan(chan);
1260 	unsigned long flags;
1261 
1262 	spin_lock_irqsave(&rchan->lock, flags);
1263 	rcar_dmac_chan_halt(rchan);
1264 	spin_unlock_irqrestore(&rchan->lock, flags);
1265 
1266 	/*
1267 	 * FIXME: No new interrupt can occur now, but the IRQ thread might still
1268 	 * be running.
1269 	 */
1270 
1271 	rcar_dmac_chan_reinit(rchan);
1272 
1273 	return 0;
1274 }
1275 
1276 static unsigned int rcar_dmac_chan_get_residue(struct rcar_dmac_chan *chan,
1277 					       dma_cookie_t cookie)
1278 {
1279 	struct rcar_dmac_desc *desc = chan->desc.running;
1280 	struct rcar_dmac_xfer_chunk *running = NULL;
1281 	struct rcar_dmac_xfer_chunk *chunk;
1282 	enum dma_status status;
1283 	unsigned int residue = 0;
1284 	unsigned int dptr = 0;
1285 	unsigned int chcrb;
1286 	unsigned int tcrb;
1287 	unsigned int i;
1288 
1289 	if (!desc)
1290 		return 0;
1291 
1292 	/*
1293 	 * If the cookie corresponds to a descriptor that has been completed
1294 	 * there is no residue. The same check has already been performed by the
1295 	 * caller but without holding the channel lock, so the descriptor could
1296 	 * now be complete.
1297 	 */
1298 	status = dma_cookie_status(&chan->chan, cookie, NULL);
1299 	if (status == DMA_COMPLETE)
1300 		return 0;
1301 
1302 	/*
1303 	 * If the cookie doesn't correspond to the currently running transfer
1304 	 * then the descriptor hasn't been processed yet, and the residue is
1305 	 * equal to the full descriptor size.
1306 	 * Also, a client driver is possible to call this function before
1307 	 * rcar_dmac_isr_channel_thread() runs. In this case, the "desc.running"
1308 	 * will be the next descriptor, and the done list will appear. So, if
1309 	 * the argument cookie matches the done list's cookie, we can assume
1310 	 * the residue is zero.
1311 	 */
1312 	if (cookie != desc->async_tx.cookie) {
1313 		list_for_each_entry(desc, &chan->desc.done, node) {
1314 			if (cookie == desc->async_tx.cookie)
1315 				return 0;
1316 		}
1317 		list_for_each_entry(desc, &chan->desc.pending, node) {
1318 			if (cookie == desc->async_tx.cookie)
1319 				return desc->size;
1320 		}
1321 		list_for_each_entry(desc, &chan->desc.active, node) {
1322 			if (cookie == desc->async_tx.cookie)
1323 				return desc->size;
1324 		}
1325 
1326 		/*
1327 		 * No descriptor found for the cookie, there's thus no residue.
1328 		 * This shouldn't happen if the calling driver passes a correct
1329 		 * cookie value.
1330 		 */
1331 		WARN(1, "No descriptor for cookie!");
1332 		return 0;
1333 	}
1334 
1335 	/*
1336 	 * We need to read two registers.
1337 	 * Make sure the control register does not skip to next chunk
1338 	 * while reading the counter.
1339 	 * Trying it 3 times should be enough: Initial read, retry, retry
1340 	 * for the paranoid.
1341 	 */
1342 	for (i = 0; i < 3; i++) {
1343 		chcrb = rcar_dmac_chan_read(chan, RCAR_DMACHCRB) &
1344 					    RCAR_DMACHCRB_DPTR_MASK;
1345 		tcrb = rcar_dmac_chan_read(chan, RCAR_DMATCRB);
1346 		/* Still the same? */
1347 		if (chcrb == (rcar_dmac_chan_read(chan, RCAR_DMACHCRB) &
1348 			      RCAR_DMACHCRB_DPTR_MASK))
1349 			break;
1350 	}
1351 	WARN_ONCE(i >= 3, "residue might be not continuous!");
1352 
1353 	/*
1354 	 * In descriptor mode the descriptor running pointer is not maintained
1355 	 * by the interrupt handler, find the running descriptor from the
1356 	 * descriptor pointer field in the CHCRB register. In non-descriptor
1357 	 * mode just use the running descriptor pointer.
1358 	 */
1359 	if (desc->hwdescs.use) {
1360 		dptr = chcrb >> RCAR_DMACHCRB_DPTR_SHIFT;
1361 		if (dptr == 0)
1362 			dptr = desc->nchunks;
1363 		dptr--;
1364 		WARN_ON(dptr >= desc->nchunks);
1365 	} else {
1366 		running = desc->running;
1367 	}
1368 
1369 	/* Compute the size of all chunks still to be transferred. */
1370 	list_for_each_entry_reverse(chunk, &desc->chunks, node) {
1371 		if (chunk == running || ++dptr == desc->nchunks)
1372 			break;
1373 
1374 		residue += chunk->size;
1375 	}
1376 
1377 	/* Add the residue for the current chunk. */
1378 	residue += tcrb << desc->xfer_shift;
1379 
1380 	return residue;
1381 }
1382 
1383 static enum dma_status rcar_dmac_tx_status(struct dma_chan *chan,
1384 					   dma_cookie_t cookie,
1385 					   struct dma_tx_state *txstate)
1386 {
1387 	struct rcar_dmac_chan *rchan = to_rcar_dmac_chan(chan);
1388 	enum dma_status status;
1389 	unsigned long flags;
1390 	unsigned int residue;
1391 	bool cyclic;
1392 
1393 	status = dma_cookie_status(chan, cookie, txstate);
1394 	if (status == DMA_COMPLETE || !txstate)
1395 		return status;
1396 
1397 	spin_lock_irqsave(&rchan->lock, flags);
1398 	residue = rcar_dmac_chan_get_residue(rchan, cookie);
1399 	cyclic = rchan->desc.running ? rchan->desc.running->cyclic : false;
1400 	spin_unlock_irqrestore(&rchan->lock, flags);
1401 
1402 	/* if there's no residue, the cookie is complete */
1403 	if (!residue && !cyclic)
1404 		return DMA_COMPLETE;
1405 
1406 	dma_set_residue(txstate, residue);
1407 
1408 	return status;
1409 }
1410 
1411 static void rcar_dmac_issue_pending(struct dma_chan *chan)
1412 {
1413 	struct rcar_dmac_chan *rchan = to_rcar_dmac_chan(chan);
1414 	unsigned long flags;
1415 
1416 	spin_lock_irqsave(&rchan->lock, flags);
1417 
1418 	if (list_empty(&rchan->desc.pending))
1419 		goto done;
1420 
1421 	/* Append the pending list to the active list. */
1422 	list_splice_tail_init(&rchan->desc.pending, &rchan->desc.active);
1423 
1424 	/*
1425 	 * If no transfer is running pick the first descriptor from the active
1426 	 * list and start the transfer.
1427 	 */
1428 	if (!rchan->desc.running) {
1429 		struct rcar_dmac_desc *desc;
1430 
1431 		desc = list_first_entry(&rchan->desc.active,
1432 					struct rcar_dmac_desc, node);
1433 		rchan->desc.running = desc;
1434 
1435 		rcar_dmac_chan_start_xfer(rchan);
1436 	}
1437 
1438 done:
1439 	spin_unlock_irqrestore(&rchan->lock, flags);
1440 }
1441 
1442 static void rcar_dmac_device_synchronize(struct dma_chan *chan)
1443 {
1444 	struct rcar_dmac_chan *rchan = to_rcar_dmac_chan(chan);
1445 
1446 	synchronize_irq(rchan->irq);
1447 }
1448 
1449 /* -----------------------------------------------------------------------------
1450  * IRQ handling
1451  */
1452 
1453 static irqreturn_t rcar_dmac_isr_desc_stage_end(struct rcar_dmac_chan *chan)
1454 {
1455 	struct rcar_dmac_desc *desc = chan->desc.running;
1456 	unsigned int stage;
1457 
1458 	if (WARN_ON(!desc || !desc->cyclic)) {
1459 		/*
1460 		 * This should never happen, there should always be a running
1461 		 * cyclic descriptor when a descriptor stage end interrupt is
1462 		 * triggered. Warn and return.
1463 		 */
1464 		return IRQ_NONE;
1465 	}
1466 
1467 	/* Program the interrupt pointer to the next stage. */
1468 	stage = (rcar_dmac_chan_read(chan, RCAR_DMACHCRB) &
1469 		 RCAR_DMACHCRB_DPTR_MASK) >> RCAR_DMACHCRB_DPTR_SHIFT;
1470 	rcar_dmac_chan_write(chan, RCAR_DMADPCR, RCAR_DMADPCR_DIPT(stage));
1471 
1472 	return IRQ_WAKE_THREAD;
1473 }
1474 
1475 static irqreturn_t rcar_dmac_isr_transfer_end(struct rcar_dmac_chan *chan)
1476 {
1477 	struct rcar_dmac_desc *desc = chan->desc.running;
1478 	irqreturn_t ret = IRQ_WAKE_THREAD;
1479 
1480 	if (WARN_ON_ONCE(!desc)) {
1481 		/*
1482 		 * This should never happen, there should always be a running
1483 		 * descriptor when a transfer end interrupt is triggered. Warn
1484 		 * and return.
1485 		 */
1486 		return IRQ_NONE;
1487 	}
1488 
1489 	/*
1490 	 * The transfer end interrupt isn't generated for each chunk when using
1491 	 * descriptor mode. Only update the running chunk pointer in
1492 	 * non-descriptor mode.
1493 	 */
1494 	if (!desc->hwdescs.use) {
1495 		/*
1496 		 * If we haven't completed the last transfer chunk simply move
1497 		 * to the next one. Only wake the IRQ thread if the transfer is
1498 		 * cyclic.
1499 		 */
1500 		if (!list_is_last(&desc->running->node, &desc->chunks)) {
1501 			desc->running = list_next_entry(desc->running, node);
1502 			if (!desc->cyclic)
1503 				ret = IRQ_HANDLED;
1504 			goto done;
1505 		}
1506 
1507 		/*
1508 		 * We've completed the last transfer chunk. If the transfer is
1509 		 * cyclic, move back to the first one.
1510 		 */
1511 		if (desc->cyclic) {
1512 			desc->running =
1513 				list_first_entry(&desc->chunks,
1514 						 struct rcar_dmac_xfer_chunk,
1515 						 node);
1516 			goto done;
1517 		}
1518 	}
1519 
1520 	/* The descriptor is complete, move it to the done list. */
1521 	list_move_tail(&desc->node, &chan->desc.done);
1522 
1523 	/* Queue the next descriptor, if any. */
1524 	if (!list_empty(&chan->desc.active))
1525 		chan->desc.running = list_first_entry(&chan->desc.active,
1526 						      struct rcar_dmac_desc,
1527 						      node);
1528 	else
1529 		chan->desc.running = NULL;
1530 
1531 done:
1532 	if (chan->desc.running)
1533 		rcar_dmac_chan_start_xfer(chan);
1534 
1535 	return ret;
1536 }
1537 
1538 static irqreturn_t rcar_dmac_isr_channel(int irq, void *dev)
1539 {
1540 	u32 mask = RCAR_DMACHCR_DSE | RCAR_DMACHCR_TE;
1541 	struct rcar_dmac_chan *chan = dev;
1542 	irqreturn_t ret = IRQ_NONE;
1543 	bool reinit = false;
1544 	u32 chcr;
1545 
1546 	spin_lock(&chan->lock);
1547 
1548 	chcr = rcar_dmac_chan_read(chan, RCAR_DMACHCR);
1549 	if (chcr & RCAR_DMACHCR_CAE) {
1550 		struct rcar_dmac *dmac = to_rcar_dmac(chan->chan.device);
1551 
1552 		/*
1553 		 * We don't need to call rcar_dmac_chan_halt()
1554 		 * because channel is already stopped in error case.
1555 		 * We need to clear register and check DE bit as recovery.
1556 		 */
1557 		rcar_dmac_write(dmac, RCAR_DMACHCLR, 1 << chan->index);
1558 		rcar_dmac_chcr_de_barrier(chan);
1559 		reinit = true;
1560 		goto spin_lock_end;
1561 	}
1562 
1563 	if (chcr & RCAR_DMACHCR_TE)
1564 		mask |= RCAR_DMACHCR_DE;
1565 	rcar_dmac_chan_write(chan, RCAR_DMACHCR, chcr & ~mask);
1566 	if (mask & RCAR_DMACHCR_DE)
1567 		rcar_dmac_chcr_de_barrier(chan);
1568 
1569 	if (chcr & RCAR_DMACHCR_DSE)
1570 		ret |= rcar_dmac_isr_desc_stage_end(chan);
1571 
1572 	if (chcr & RCAR_DMACHCR_TE)
1573 		ret |= rcar_dmac_isr_transfer_end(chan);
1574 
1575 spin_lock_end:
1576 	spin_unlock(&chan->lock);
1577 
1578 	if (reinit) {
1579 		dev_err(chan->chan.device->dev, "Channel Address Error\n");
1580 
1581 		rcar_dmac_chan_reinit(chan);
1582 		ret = IRQ_HANDLED;
1583 	}
1584 
1585 	return ret;
1586 }
1587 
1588 static irqreturn_t rcar_dmac_isr_channel_thread(int irq, void *dev)
1589 {
1590 	struct rcar_dmac_chan *chan = dev;
1591 	struct rcar_dmac_desc *desc;
1592 	struct dmaengine_desc_callback cb;
1593 
1594 	spin_lock_irq(&chan->lock);
1595 
1596 	/* For cyclic transfers notify the user after every chunk. */
1597 	if (chan->desc.running && chan->desc.running->cyclic) {
1598 		desc = chan->desc.running;
1599 		dmaengine_desc_get_callback(&desc->async_tx, &cb);
1600 
1601 		if (dmaengine_desc_callback_valid(&cb)) {
1602 			spin_unlock_irq(&chan->lock);
1603 			dmaengine_desc_callback_invoke(&cb, NULL);
1604 			spin_lock_irq(&chan->lock);
1605 		}
1606 	}
1607 
1608 	/*
1609 	 * Call the callback function for all descriptors on the done list and
1610 	 * move them to the ack wait list.
1611 	 */
1612 	while (!list_empty(&chan->desc.done)) {
1613 		desc = list_first_entry(&chan->desc.done, struct rcar_dmac_desc,
1614 					node);
1615 		dma_cookie_complete(&desc->async_tx);
1616 		list_del(&desc->node);
1617 
1618 		dmaengine_desc_get_callback(&desc->async_tx, &cb);
1619 		if (dmaengine_desc_callback_valid(&cb)) {
1620 			spin_unlock_irq(&chan->lock);
1621 			/*
1622 			 * We own the only reference to this descriptor, we can
1623 			 * safely dereference it without holding the channel
1624 			 * lock.
1625 			 */
1626 			dmaengine_desc_callback_invoke(&cb, NULL);
1627 			spin_lock_irq(&chan->lock);
1628 		}
1629 
1630 		list_add_tail(&desc->node, &chan->desc.wait);
1631 	}
1632 
1633 	spin_unlock_irq(&chan->lock);
1634 
1635 	/* Recycle all acked descriptors. */
1636 	rcar_dmac_desc_recycle_acked(chan);
1637 
1638 	return IRQ_HANDLED;
1639 }
1640 
1641 /* -----------------------------------------------------------------------------
1642  * OF xlate and channel filter
1643  */
1644 
1645 static bool rcar_dmac_chan_filter(struct dma_chan *chan, void *arg)
1646 {
1647 	struct rcar_dmac *dmac = to_rcar_dmac(chan->device);
1648 	struct of_phandle_args *dma_spec = arg;
1649 
1650 	/*
1651 	 * FIXME: Using a filter on OF platforms is a nonsense. The OF xlate
1652 	 * function knows from which device it wants to allocate a channel from,
1653 	 * and would be perfectly capable of selecting the channel it wants.
1654 	 * Forcing it to call dma_request_channel() and iterate through all
1655 	 * channels from all controllers is just pointless.
1656 	 */
1657 	if (chan->device->device_config != rcar_dmac_device_config)
1658 		return false;
1659 
1660 	return !test_and_set_bit(dma_spec->args[0], dmac->modules);
1661 }
1662 
1663 static struct dma_chan *rcar_dmac_of_xlate(struct of_phandle_args *dma_spec,
1664 					   struct of_dma *ofdma)
1665 {
1666 	struct rcar_dmac_chan *rchan;
1667 	struct dma_chan *chan;
1668 	dma_cap_mask_t mask;
1669 
1670 	if (dma_spec->args_count != 1)
1671 		return NULL;
1672 
1673 	/* Only slave DMA channels can be allocated via DT */
1674 	dma_cap_zero(mask);
1675 	dma_cap_set(DMA_SLAVE, mask);
1676 
1677 	chan = __dma_request_channel(&mask, rcar_dmac_chan_filter, dma_spec,
1678 				     ofdma->of_node);
1679 	if (!chan)
1680 		return NULL;
1681 
1682 	rchan = to_rcar_dmac_chan(chan);
1683 	rchan->mid_rid = dma_spec->args[0];
1684 
1685 	return chan;
1686 }
1687 
1688 /* -----------------------------------------------------------------------------
1689  * Power management
1690  */
1691 
1692 #ifdef CONFIG_PM
1693 static int rcar_dmac_runtime_suspend(struct device *dev)
1694 {
1695 	return 0;
1696 }
1697 
1698 static int rcar_dmac_runtime_resume(struct device *dev)
1699 {
1700 	struct rcar_dmac *dmac = dev_get_drvdata(dev);
1701 
1702 	return rcar_dmac_init(dmac);
1703 }
1704 #endif
1705 
1706 static const struct dev_pm_ops rcar_dmac_pm = {
1707 	/*
1708 	 * TODO for system sleep/resume:
1709 	 *   - Wait for the current transfer to complete and stop the device,
1710 	 *   - Resume transfers, if any.
1711 	 */
1712 	SET_NOIRQ_SYSTEM_SLEEP_PM_OPS(pm_runtime_force_suspend,
1713 				      pm_runtime_force_resume)
1714 	SET_RUNTIME_PM_OPS(rcar_dmac_runtime_suspend, rcar_dmac_runtime_resume,
1715 			   NULL)
1716 };
1717 
1718 /* -----------------------------------------------------------------------------
1719  * Probe and remove
1720  */
1721 
1722 static int rcar_dmac_chan_probe(struct rcar_dmac *dmac,
1723 				struct rcar_dmac_chan *rchan,
1724 				unsigned int index)
1725 {
1726 	struct platform_device *pdev = to_platform_device(dmac->dev);
1727 	struct dma_chan *chan = &rchan->chan;
1728 	char pdev_irqname[5];
1729 	char *irqname;
1730 	int ret;
1731 
1732 	rchan->index = index;
1733 	rchan->iomem = dmac->iomem + RCAR_DMAC_CHAN_OFFSET(index);
1734 	rchan->mid_rid = -EINVAL;
1735 
1736 	spin_lock_init(&rchan->lock);
1737 
1738 	INIT_LIST_HEAD(&rchan->desc.free);
1739 	INIT_LIST_HEAD(&rchan->desc.pending);
1740 	INIT_LIST_HEAD(&rchan->desc.active);
1741 	INIT_LIST_HEAD(&rchan->desc.done);
1742 	INIT_LIST_HEAD(&rchan->desc.wait);
1743 
1744 	/* Request the channel interrupt. */
1745 	sprintf(pdev_irqname, "ch%u", index);
1746 	rchan->irq = platform_get_irq_byname(pdev, pdev_irqname);
1747 	if (rchan->irq < 0) {
1748 		dev_err(dmac->dev, "no IRQ specified for channel %u\n", index);
1749 		return -ENODEV;
1750 	}
1751 
1752 	irqname = devm_kasprintf(dmac->dev, GFP_KERNEL, "%s:%u",
1753 				 dev_name(dmac->dev), index);
1754 	if (!irqname)
1755 		return -ENOMEM;
1756 
1757 	/*
1758 	 * Initialize the DMA engine channel and add it to the DMA engine
1759 	 * channels list.
1760 	 */
1761 	chan->device = &dmac->engine;
1762 	dma_cookie_init(chan);
1763 
1764 	list_add_tail(&chan->device_node, &dmac->engine.channels);
1765 
1766 	ret = devm_request_threaded_irq(dmac->dev, rchan->irq,
1767 					rcar_dmac_isr_channel,
1768 					rcar_dmac_isr_channel_thread, 0,
1769 					irqname, rchan);
1770 	if (ret) {
1771 		dev_err(dmac->dev, "failed to request IRQ %u (%d)\n",
1772 			rchan->irq, ret);
1773 		return ret;
1774 	}
1775 
1776 	return 0;
1777 }
1778 
1779 static int rcar_dmac_parse_of(struct device *dev, struct rcar_dmac *dmac)
1780 {
1781 	struct device_node *np = dev->of_node;
1782 	int ret;
1783 
1784 	ret = of_property_read_u32(np, "dma-channels", &dmac->n_channels);
1785 	if (ret < 0) {
1786 		dev_err(dev, "unable to read dma-channels property\n");
1787 		return ret;
1788 	}
1789 
1790 	if (dmac->n_channels <= 0 || dmac->n_channels >= 100) {
1791 		dev_err(dev, "invalid number of channels %u\n",
1792 			dmac->n_channels);
1793 		return -EINVAL;
1794 	}
1795 
1796 	return 0;
1797 }
1798 
1799 static int rcar_dmac_probe(struct platform_device *pdev)
1800 {
1801 	const enum dma_slave_buswidth widths = DMA_SLAVE_BUSWIDTH_1_BYTE |
1802 		DMA_SLAVE_BUSWIDTH_2_BYTES | DMA_SLAVE_BUSWIDTH_4_BYTES |
1803 		DMA_SLAVE_BUSWIDTH_8_BYTES | DMA_SLAVE_BUSWIDTH_16_BYTES |
1804 		DMA_SLAVE_BUSWIDTH_32_BYTES | DMA_SLAVE_BUSWIDTH_64_BYTES;
1805 	unsigned int channels_offset = 0;
1806 	struct dma_device *engine;
1807 	struct rcar_dmac *dmac;
1808 	struct resource *mem;
1809 	unsigned int i;
1810 	int ret;
1811 
1812 	dmac = devm_kzalloc(&pdev->dev, sizeof(*dmac), GFP_KERNEL);
1813 	if (!dmac)
1814 		return -ENOMEM;
1815 
1816 	dmac->dev = &pdev->dev;
1817 	platform_set_drvdata(pdev, dmac);
1818 	dmac->dev->dma_parms = &dmac->parms;
1819 	dma_set_max_seg_size(dmac->dev, RCAR_DMATCR_MASK);
1820 	dma_set_mask_and_coherent(dmac->dev, DMA_BIT_MASK(40));
1821 
1822 	ret = rcar_dmac_parse_of(&pdev->dev, dmac);
1823 	if (ret < 0)
1824 		return ret;
1825 
1826 	/*
1827 	 * A still unconfirmed hardware bug prevents the IPMMU microTLB 0 to be
1828 	 * flushed correctly, resulting in memory corruption. DMAC 0 channel 0
1829 	 * is connected to microTLB 0 on currently supported platforms, so we
1830 	 * can't use it with the IPMMU. As the IOMMU API operates at the device
1831 	 * level we can't disable it selectively, so ignore channel 0 for now if
1832 	 * the device is part of an IOMMU group.
1833 	 */
1834 	if (device_iommu_mapped(&pdev->dev)) {
1835 		dmac->n_channels--;
1836 		channels_offset = 1;
1837 	}
1838 
1839 	dmac->channels = devm_kcalloc(&pdev->dev, dmac->n_channels,
1840 				      sizeof(*dmac->channels), GFP_KERNEL);
1841 	if (!dmac->channels)
1842 		return -ENOMEM;
1843 
1844 	/* Request resources. */
1845 	mem = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1846 	dmac->iomem = devm_ioremap_resource(&pdev->dev, mem);
1847 	if (IS_ERR(dmac->iomem))
1848 		return PTR_ERR(dmac->iomem);
1849 
1850 	/* Enable runtime PM and initialize the device. */
1851 	pm_runtime_enable(&pdev->dev);
1852 	ret = pm_runtime_get_sync(&pdev->dev);
1853 	if (ret < 0) {
1854 		dev_err(&pdev->dev, "runtime PM get sync failed (%d)\n", ret);
1855 		return ret;
1856 	}
1857 
1858 	ret = rcar_dmac_init(dmac);
1859 	pm_runtime_put(&pdev->dev);
1860 
1861 	if (ret) {
1862 		dev_err(&pdev->dev, "failed to reset device\n");
1863 		goto error;
1864 	}
1865 
1866 	/* Initialize engine */
1867 	engine = &dmac->engine;
1868 
1869 	dma_cap_set(DMA_MEMCPY, engine->cap_mask);
1870 	dma_cap_set(DMA_SLAVE, engine->cap_mask);
1871 
1872 	engine->dev		= &pdev->dev;
1873 	engine->copy_align	= ilog2(RCAR_DMAC_MEMCPY_XFER_SIZE);
1874 
1875 	engine->src_addr_widths	= widths;
1876 	engine->dst_addr_widths	= widths;
1877 	engine->directions	= BIT(DMA_MEM_TO_DEV) | BIT(DMA_DEV_TO_MEM);
1878 	engine->residue_granularity = DMA_RESIDUE_GRANULARITY_BURST;
1879 
1880 	engine->device_alloc_chan_resources	= rcar_dmac_alloc_chan_resources;
1881 	engine->device_free_chan_resources	= rcar_dmac_free_chan_resources;
1882 	engine->device_prep_dma_memcpy		= rcar_dmac_prep_dma_memcpy;
1883 	engine->device_prep_slave_sg		= rcar_dmac_prep_slave_sg;
1884 	engine->device_prep_dma_cyclic		= rcar_dmac_prep_dma_cyclic;
1885 	engine->device_config			= rcar_dmac_device_config;
1886 	engine->device_pause			= rcar_dmac_chan_pause;
1887 	engine->device_terminate_all		= rcar_dmac_chan_terminate_all;
1888 	engine->device_tx_status		= rcar_dmac_tx_status;
1889 	engine->device_issue_pending		= rcar_dmac_issue_pending;
1890 	engine->device_synchronize		= rcar_dmac_device_synchronize;
1891 
1892 	INIT_LIST_HEAD(&engine->channels);
1893 
1894 	for (i = 0; i < dmac->n_channels; ++i) {
1895 		ret = rcar_dmac_chan_probe(dmac, &dmac->channels[i],
1896 					   i + channels_offset);
1897 		if (ret < 0)
1898 			goto error;
1899 	}
1900 
1901 	/* Register the DMAC as a DMA provider for DT. */
1902 	ret = of_dma_controller_register(pdev->dev.of_node, rcar_dmac_of_xlate,
1903 					 NULL);
1904 	if (ret < 0)
1905 		goto error;
1906 
1907 	/*
1908 	 * Register the DMA engine device.
1909 	 *
1910 	 * Default transfer size of 32 bytes requires 32-byte alignment.
1911 	 */
1912 	ret = dma_async_device_register(engine);
1913 	if (ret < 0)
1914 		goto error;
1915 
1916 	return 0;
1917 
1918 error:
1919 	of_dma_controller_free(pdev->dev.of_node);
1920 	pm_runtime_disable(&pdev->dev);
1921 	return ret;
1922 }
1923 
1924 static int rcar_dmac_remove(struct platform_device *pdev)
1925 {
1926 	struct rcar_dmac *dmac = platform_get_drvdata(pdev);
1927 
1928 	of_dma_controller_free(pdev->dev.of_node);
1929 	dma_async_device_unregister(&dmac->engine);
1930 
1931 	pm_runtime_disable(&pdev->dev);
1932 
1933 	return 0;
1934 }
1935 
1936 static void rcar_dmac_shutdown(struct platform_device *pdev)
1937 {
1938 	struct rcar_dmac *dmac = platform_get_drvdata(pdev);
1939 
1940 	rcar_dmac_stop_all_chan(dmac);
1941 }
1942 
1943 static const struct of_device_id rcar_dmac_of_ids[] = {
1944 	{ .compatible = "renesas,rcar-dmac", },
1945 	{ /* Sentinel */ }
1946 };
1947 MODULE_DEVICE_TABLE(of, rcar_dmac_of_ids);
1948 
1949 static struct platform_driver rcar_dmac_driver = {
1950 	.driver		= {
1951 		.pm	= &rcar_dmac_pm,
1952 		.name	= "rcar-dmac",
1953 		.of_match_table = rcar_dmac_of_ids,
1954 	},
1955 	.probe		= rcar_dmac_probe,
1956 	.remove		= rcar_dmac_remove,
1957 	.shutdown	= rcar_dmac_shutdown,
1958 };
1959 
1960 module_platform_driver(rcar_dmac_driver);
1961 
1962 MODULE_DESCRIPTION("R-Car Gen2 DMA Controller Driver");
1963 MODULE_AUTHOR("Laurent Pinchart <laurent.pinchart@ideasonboard.com>");
1964 MODULE_LICENSE("GPL v2");
1965