xref: /linux/drivers/dma/nbpfaxi.c (revision 4949009eb8d40a441dcddcd96e101e77d31cf1b2)
1 /*
2  * Copyright (C) 2013-2014 Renesas Electronics Europe Ltd.
3  * Author: Guennadi Liakhovetski <g.liakhovetski@gmx.de>
4  *
5  * This program is free software; you can redistribute it and/or modify
6  * it under the terms of version 2 of the GNU General Public License as
7  * published by the Free Software Foundation.
8  */
9 
10 #include <linux/bitmap.h>
11 #include <linux/bitops.h>
12 #include <linux/clk.h>
13 #include <linux/dma-mapping.h>
14 #include <linux/dmaengine.h>
15 #include <linux/err.h>
16 #include <linux/interrupt.h>
17 #include <linux/io.h>
18 #include <linux/log2.h>
19 #include <linux/module.h>
20 #include <linux/of.h>
21 #include <linux/of_device.h>
22 #include <linux/of_dma.h>
23 #include <linux/platform_device.h>
24 #include <linux/slab.h>
25 
26 #include <dt-bindings/dma/nbpfaxi.h>
27 
28 #include "dmaengine.h"
29 
30 #define NBPF_REG_CHAN_OFFSET	0
31 #define NBPF_REG_CHAN_SIZE	0x40
32 
33 /* Channel Current Transaction Byte register */
34 #define NBPF_CHAN_CUR_TR_BYTE	0x20
35 
36 /* Channel Status register */
37 #define NBPF_CHAN_STAT	0x24
38 #define NBPF_CHAN_STAT_EN	1
39 #define NBPF_CHAN_STAT_TACT	4
40 #define NBPF_CHAN_STAT_ERR	0x10
41 #define NBPF_CHAN_STAT_END	0x20
42 #define NBPF_CHAN_STAT_TC	0x40
43 #define NBPF_CHAN_STAT_DER	0x400
44 
45 /* Channel Control register */
46 #define NBPF_CHAN_CTRL	0x28
47 #define NBPF_CHAN_CTRL_SETEN	1
48 #define NBPF_CHAN_CTRL_CLREN	2
49 #define NBPF_CHAN_CTRL_STG	4
50 #define NBPF_CHAN_CTRL_SWRST	8
51 #define NBPF_CHAN_CTRL_CLRRQ	0x10
52 #define NBPF_CHAN_CTRL_CLREND	0x20
53 #define NBPF_CHAN_CTRL_CLRTC	0x40
54 #define NBPF_CHAN_CTRL_SETSUS	0x100
55 #define NBPF_CHAN_CTRL_CLRSUS	0x200
56 
57 /* Channel Configuration register */
58 #define NBPF_CHAN_CFG	0x2c
59 #define NBPF_CHAN_CFG_SEL	7		/* terminal SELect: 0..7 */
60 #define NBPF_CHAN_CFG_REQD	8		/* REQuest Direction: DMAREQ is 0: input, 1: output */
61 #define NBPF_CHAN_CFG_LOEN	0x10		/* LOw ENable: low DMA request line is: 0: inactive, 1: active */
62 #define NBPF_CHAN_CFG_HIEN	0x20		/* HIgh ENable: high DMA request line is: 0: inactive, 1: active */
63 #define NBPF_CHAN_CFG_LVL	0x40		/* LeVeL: DMA request line is sensed as 0: edge, 1: level */
64 #define NBPF_CHAN_CFG_AM	0x700		/* ACK Mode: 0: Pulse mode, 1: Level mode, b'1x: Bus Cycle */
65 #define NBPF_CHAN_CFG_SDS	0xf000		/* Source Data Size: 0: 8 bits,... , 7: 1024 bits */
66 #define NBPF_CHAN_CFG_DDS	0xf0000		/* Destination Data Size: as above */
67 #define NBPF_CHAN_CFG_SAD	0x100000	/* Source ADdress counting: 0: increment, 1: fixed */
68 #define NBPF_CHAN_CFG_DAD	0x200000	/* Destination ADdress counting: 0: increment, 1: fixed */
69 #define NBPF_CHAN_CFG_TM	0x400000	/* Transfer Mode: 0: single, 1: block TM */
70 #define NBPF_CHAN_CFG_DEM	0x1000000	/* DMAEND interrupt Mask */
71 #define NBPF_CHAN_CFG_TCM	0x2000000	/* DMATCO interrupt Mask */
72 #define NBPF_CHAN_CFG_SBE	0x8000000	/* Sweep Buffer Enable */
73 #define NBPF_CHAN_CFG_RSEL	0x10000000	/* RM: Register Set sELect */
74 #define NBPF_CHAN_CFG_RSW	0x20000000	/* RM: Register Select sWitch */
75 #define NBPF_CHAN_CFG_REN	0x40000000	/* RM: Register Set Enable */
76 #define NBPF_CHAN_CFG_DMS	0x80000000	/* 0: register mode (RM), 1: link mode (LM) */
77 
78 #define NBPF_CHAN_NXLA	0x38
79 #define NBPF_CHAN_CRLA	0x3c
80 
81 /* Link Header field */
82 #define NBPF_HEADER_LV	1
83 #define NBPF_HEADER_LE	2
84 #define NBPF_HEADER_WBD	4
85 #define NBPF_HEADER_DIM	8
86 
87 #define NBPF_CTRL	0x300
88 #define NBPF_CTRL_PR	1		/* 0: fixed priority, 1: round robin */
89 #define NBPF_CTRL_LVINT	2		/* DMAEND and DMAERR signalling: 0: pulse, 1: level */
90 
91 #define NBPF_DSTAT_ER	0x314
92 #define NBPF_DSTAT_END	0x318
93 
94 #define NBPF_DMA_BUSWIDTHS \
95 	(BIT(DMA_SLAVE_BUSWIDTH_UNDEFINED) | \
96 	 BIT(DMA_SLAVE_BUSWIDTH_1_BYTE) | \
97 	 BIT(DMA_SLAVE_BUSWIDTH_2_BYTES) | \
98 	 BIT(DMA_SLAVE_BUSWIDTH_4_BYTES) | \
99 	 BIT(DMA_SLAVE_BUSWIDTH_8_BYTES))
100 
101 struct nbpf_config {
102 	int num_channels;
103 	int buffer_size;
104 };
105 
106 /*
107  * We've got 3 types of objects, used to describe DMA transfers:
108  * 1. high-level descriptor, containing a struct dma_async_tx_descriptor object
109  *	in it, used to communicate with the user
110  * 2. hardware DMA link descriptors, that we pass to DMAC for DMA transfer
111  *	queuing, these must be DMAable, using either the streaming DMA API or
112  *	allocated from coherent memory - one per SG segment
113  * 3. one per SG segment descriptors, used to manage HW link descriptors from
114  *	(2). They do not have to be DMAable. They can either be (a) allocated
115  *	together with link descriptors as mixed (DMA / CPU) objects, or (b)
116  *	separately. Even if allocated separately it would be best to link them
117  *	to link descriptors once during channel resource allocation and always
118  *	use them as a single object.
119  * Therefore for both cases (a) and (b) at run-time objects (2) and (3) shall be
120  * treated as a single SG segment descriptor.
121  */
122 
123 struct nbpf_link_reg {
124 	u32	header;
125 	u32	src_addr;
126 	u32	dst_addr;
127 	u32	transaction_size;
128 	u32	config;
129 	u32	interval;
130 	u32	extension;
131 	u32	next;
132 } __packed;
133 
134 struct nbpf_device;
135 struct nbpf_channel;
136 struct nbpf_desc;
137 
138 struct nbpf_link_desc {
139 	struct nbpf_link_reg *hwdesc;
140 	dma_addr_t hwdesc_dma_addr;
141 	struct nbpf_desc *desc;
142 	struct list_head node;
143 };
144 
145 /**
146  * struct nbpf_desc - DMA transfer descriptor
147  * @async_tx:	dmaengine object
148  * @user_wait:	waiting for a user ack
149  * @length:	total transfer length
150  * @sg:		list of hardware descriptors, represented by struct nbpf_link_desc
151  * @node:	member in channel descriptor lists
152  */
153 struct nbpf_desc {
154 	struct dma_async_tx_descriptor async_tx;
155 	bool user_wait;
156 	size_t length;
157 	struct nbpf_channel *chan;
158 	struct list_head sg;
159 	struct list_head node;
160 };
161 
162 /* Take a wild guess: allocate 4 segments per descriptor */
163 #define NBPF_SEGMENTS_PER_DESC 4
164 #define NBPF_DESCS_PER_PAGE ((PAGE_SIZE - sizeof(struct list_head)) /	\
165 	(sizeof(struct nbpf_desc) +					\
166 	 NBPF_SEGMENTS_PER_DESC *					\
167 	 (sizeof(struct nbpf_link_desc) + sizeof(struct nbpf_link_reg))))
168 #define NBPF_SEGMENTS_PER_PAGE (NBPF_SEGMENTS_PER_DESC * NBPF_DESCS_PER_PAGE)
169 
170 struct nbpf_desc_page {
171 	struct list_head node;
172 	struct nbpf_desc desc[NBPF_DESCS_PER_PAGE];
173 	struct nbpf_link_desc ldesc[NBPF_SEGMENTS_PER_PAGE];
174 	struct nbpf_link_reg hwdesc[NBPF_SEGMENTS_PER_PAGE];
175 };
176 
177 /**
178  * struct nbpf_channel - one DMAC channel
179  * @dma_chan:	standard dmaengine channel object
180  * @base:	register address base
181  * @nbpf:	DMAC
182  * @name:	IRQ name
183  * @irq:	IRQ number
184  * @slave_addr:	address for slave DMA
185  * @slave_width:slave data size in bytes
186  * @slave_burst:maximum slave burst size in bytes
187  * @terminal:	DMA terminal, assigned to this channel
188  * @dmarq_cfg:	DMA request line configuration - high / low, edge / level for NBPF_CHAN_CFG
189  * @flags:	configuration flags from DT
190  * @lock:	protect descriptor lists
191  * @free_links:	list of free link descriptors
192  * @free:	list of free descriptors
193  * @queued:	list of queued descriptors
194  * @active:	list of descriptors, scheduled for processing
195  * @done:	list of completed descriptors, waiting post-processing
196  * @desc_page:	list of additionally allocated descriptor pages - if any
197  */
198 struct nbpf_channel {
199 	struct dma_chan dma_chan;
200 	struct tasklet_struct tasklet;
201 	void __iomem *base;
202 	struct nbpf_device *nbpf;
203 	char name[16];
204 	int irq;
205 	dma_addr_t slave_src_addr;
206 	size_t slave_src_width;
207 	size_t slave_src_burst;
208 	dma_addr_t slave_dst_addr;
209 	size_t slave_dst_width;
210 	size_t slave_dst_burst;
211 	unsigned int terminal;
212 	u32 dmarq_cfg;
213 	unsigned long flags;
214 	spinlock_t lock;
215 	struct list_head free_links;
216 	struct list_head free;
217 	struct list_head queued;
218 	struct list_head active;
219 	struct list_head done;
220 	struct list_head desc_page;
221 	struct nbpf_desc *running;
222 	bool paused;
223 };
224 
225 struct nbpf_device {
226 	struct dma_device dma_dev;
227 	void __iomem *base;
228 	struct clk *clk;
229 	const struct nbpf_config *config;
230 	struct nbpf_channel chan[];
231 };
232 
233 enum nbpf_model {
234 	NBPF1B4,
235 	NBPF1B8,
236 	NBPF1B16,
237 	NBPF4B4,
238 	NBPF4B8,
239 	NBPF4B16,
240 	NBPF8B4,
241 	NBPF8B8,
242 	NBPF8B16,
243 };
244 
245 static struct nbpf_config nbpf_cfg[] = {
246 	[NBPF1B4] = {
247 		.num_channels = 1,
248 		.buffer_size = 4,
249 	},
250 	[NBPF1B8] = {
251 		.num_channels = 1,
252 		.buffer_size = 8,
253 	},
254 	[NBPF1B16] = {
255 		.num_channels = 1,
256 		.buffer_size = 16,
257 	},
258 	[NBPF4B4] = {
259 		.num_channels = 4,
260 		.buffer_size = 4,
261 	},
262 	[NBPF4B8] = {
263 		.num_channels = 4,
264 		.buffer_size = 8,
265 	},
266 	[NBPF4B16] = {
267 		.num_channels = 4,
268 		.buffer_size = 16,
269 	},
270 	[NBPF8B4] = {
271 		.num_channels = 8,
272 		.buffer_size = 4,
273 	},
274 	[NBPF8B8] = {
275 		.num_channels = 8,
276 		.buffer_size = 8,
277 	},
278 	[NBPF8B16] = {
279 		.num_channels = 8,
280 		.buffer_size = 16,
281 	},
282 };
283 
284 #define nbpf_to_chan(d) container_of(d, struct nbpf_channel, dma_chan)
285 
286 /*
287  * dmaengine drivers seem to have a lot in common and instead of sharing more
288  * code, they reimplement those common algorithms independently. In this driver
289  * we try to separate the hardware-specific part from the (largely) generic
290  * part. This improves code readability and makes it possible in the future to
291  * reuse the generic code in form of a helper library. That generic code should
292  * be suitable for various DMA controllers, using transfer descriptors in RAM
293  * and pushing one SG list at a time to the DMA controller.
294  */
295 
296 /*		Hardware-specific part		*/
297 
298 static inline u32 nbpf_chan_read(struct nbpf_channel *chan,
299 				 unsigned int offset)
300 {
301 	u32 data = ioread32(chan->base + offset);
302 	dev_dbg(chan->dma_chan.device->dev, "%s(0x%p + 0x%x) = 0x%x\n",
303 		__func__, chan->base, offset, data);
304 	return data;
305 }
306 
307 static inline void nbpf_chan_write(struct nbpf_channel *chan,
308 				   unsigned int offset, u32 data)
309 {
310 	iowrite32(data, chan->base + offset);
311 	dev_dbg(chan->dma_chan.device->dev, "%s(0x%p + 0x%x) = 0x%x\n",
312 		__func__, chan->base, offset, data);
313 }
314 
315 static inline u32 nbpf_read(struct nbpf_device *nbpf,
316 			    unsigned int offset)
317 {
318 	u32 data = ioread32(nbpf->base + offset);
319 	dev_dbg(nbpf->dma_dev.dev, "%s(0x%p + 0x%x) = 0x%x\n",
320 		__func__, nbpf->base, offset, data);
321 	return data;
322 }
323 
324 static inline void nbpf_write(struct nbpf_device *nbpf,
325 			      unsigned int offset, u32 data)
326 {
327 	iowrite32(data, nbpf->base + offset);
328 	dev_dbg(nbpf->dma_dev.dev, "%s(0x%p + 0x%x) = 0x%x\n",
329 		__func__, nbpf->base, offset, data);
330 }
331 
332 static void nbpf_chan_halt(struct nbpf_channel *chan)
333 {
334 	nbpf_chan_write(chan, NBPF_CHAN_CTRL, NBPF_CHAN_CTRL_CLREN);
335 }
336 
337 static bool nbpf_status_get(struct nbpf_channel *chan)
338 {
339 	u32 status = nbpf_read(chan->nbpf, NBPF_DSTAT_END);
340 
341 	return status & BIT(chan - chan->nbpf->chan);
342 }
343 
344 static void nbpf_status_ack(struct nbpf_channel *chan)
345 {
346 	nbpf_chan_write(chan, NBPF_CHAN_CTRL, NBPF_CHAN_CTRL_CLREND);
347 }
348 
349 static u32 nbpf_error_get(struct nbpf_device *nbpf)
350 {
351 	return nbpf_read(nbpf, NBPF_DSTAT_ER);
352 }
353 
354 static struct nbpf_channel *nbpf_error_get_channel(struct nbpf_device *nbpf, u32 error)
355 {
356 	return nbpf->chan + __ffs(error);
357 }
358 
359 static void nbpf_error_clear(struct nbpf_channel *chan)
360 {
361 	u32 status;
362 	int i;
363 
364 	/* Stop the channel, make sure DMA has been aborted */
365 	nbpf_chan_halt(chan);
366 
367 	for (i = 1000; i; i--) {
368 		status = nbpf_chan_read(chan, NBPF_CHAN_STAT);
369 		if (!(status & NBPF_CHAN_STAT_TACT))
370 			break;
371 		cpu_relax();
372 	}
373 
374 	if (!i)
375 		dev_err(chan->dma_chan.device->dev,
376 			"%s(): abort timeout, channel status 0x%x\n", __func__, status);
377 
378 	nbpf_chan_write(chan, NBPF_CHAN_CTRL, NBPF_CHAN_CTRL_SWRST);
379 }
380 
381 static int nbpf_start(struct nbpf_desc *desc)
382 {
383 	struct nbpf_channel *chan = desc->chan;
384 	struct nbpf_link_desc *ldesc = list_first_entry(&desc->sg, struct nbpf_link_desc, node);
385 
386 	nbpf_chan_write(chan, NBPF_CHAN_NXLA, (u32)ldesc->hwdesc_dma_addr);
387 	nbpf_chan_write(chan, NBPF_CHAN_CTRL, NBPF_CHAN_CTRL_SETEN | NBPF_CHAN_CTRL_CLRSUS);
388 	chan->paused = false;
389 
390 	/* Software trigger MEMCPY - only MEMCPY uses the block mode */
391 	if (ldesc->hwdesc->config & NBPF_CHAN_CFG_TM)
392 		nbpf_chan_write(chan, NBPF_CHAN_CTRL, NBPF_CHAN_CTRL_STG);
393 
394 	dev_dbg(chan->nbpf->dma_dev.dev, "%s(): next 0x%x, cur 0x%x\n", __func__,
395 		nbpf_chan_read(chan, NBPF_CHAN_NXLA), nbpf_chan_read(chan, NBPF_CHAN_CRLA));
396 
397 	return 0;
398 }
399 
400 static void nbpf_chan_prepare(struct nbpf_channel *chan)
401 {
402 	chan->dmarq_cfg = (chan->flags & NBPF_SLAVE_RQ_HIGH ? NBPF_CHAN_CFG_HIEN : 0) |
403 		(chan->flags & NBPF_SLAVE_RQ_LOW ? NBPF_CHAN_CFG_LOEN : 0) |
404 		(chan->flags & NBPF_SLAVE_RQ_LEVEL ?
405 		 NBPF_CHAN_CFG_LVL | (NBPF_CHAN_CFG_AM & 0x200) : 0) |
406 		chan->terminal;
407 }
408 
409 static void nbpf_chan_prepare_default(struct nbpf_channel *chan)
410 {
411 	/* Don't output DMAACK */
412 	chan->dmarq_cfg = NBPF_CHAN_CFG_AM & 0x400;
413 	chan->terminal = 0;
414 	chan->flags = 0;
415 }
416 
417 static void nbpf_chan_configure(struct nbpf_channel *chan)
418 {
419 	/*
420 	 * We assume, that only the link mode and DMA request line configuration
421 	 * have to be set in the configuration register manually. Dynamic
422 	 * per-transfer configuration will be loaded from transfer descriptors.
423 	 */
424 	nbpf_chan_write(chan, NBPF_CHAN_CFG, NBPF_CHAN_CFG_DMS | chan->dmarq_cfg);
425 }
426 
427 static u32 nbpf_xfer_ds(struct nbpf_device *nbpf, size_t size)
428 {
429 	/* Maximum supported bursts depend on the buffer size */
430 	return min_t(int, __ffs(size), ilog2(nbpf->config->buffer_size * 8));
431 }
432 
433 static size_t nbpf_xfer_size(struct nbpf_device *nbpf,
434 			     enum dma_slave_buswidth width, u32 burst)
435 {
436 	size_t size;
437 
438 	if (!burst)
439 		burst = 1;
440 
441 	switch (width) {
442 	case DMA_SLAVE_BUSWIDTH_8_BYTES:
443 		size = 8 * burst;
444 		break;
445 
446 	case DMA_SLAVE_BUSWIDTH_4_BYTES:
447 		size = 4 * burst;
448 		break;
449 
450 	case DMA_SLAVE_BUSWIDTH_2_BYTES:
451 		size = 2 * burst;
452 		break;
453 
454 	default:
455 		pr_warn("%s(): invalid bus width %u\n", __func__, width);
456 	case DMA_SLAVE_BUSWIDTH_1_BYTE:
457 		size = burst;
458 	}
459 
460 	return nbpf_xfer_ds(nbpf, size);
461 }
462 
463 /*
464  * We need a way to recognise slaves, whose data is sent "raw" over the bus,
465  * i.e. it isn't known in advance how many bytes will be received. Therefore
466  * the slave driver has to provide a "large enough" buffer and either read the
467  * buffer, when it is full, or detect, that some data has arrived, then wait for
468  * a timeout, if no more data arrives - receive what's already there. We want to
469  * handle such slaves in a special way to allow an optimised mode for other
470  * users, for whom the amount of data is known in advance. So far there's no way
471  * to recognise such slaves. We use a data-width check to distinguish between
472  * the SD host and the PL011 UART.
473  */
474 
475 static int nbpf_prep_one(struct nbpf_link_desc *ldesc,
476 			 enum dma_transfer_direction direction,
477 			 dma_addr_t src, dma_addr_t dst, size_t size, bool last)
478 {
479 	struct nbpf_link_reg *hwdesc = ldesc->hwdesc;
480 	struct nbpf_desc *desc = ldesc->desc;
481 	struct nbpf_channel *chan = desc->chan;
482 	struct device *dev = chan->dma_chan.device->dev;
483 	size_t mem_xfer, slave_xfer;
484 	bool can_burst;
485 
486 	hwdesc->header = NBPF_HEADER_WBD | NBPF_HEADER_LV |
487 		(last ? NBPF_HEADER_LE : 0);
488 
489 	hwdesc->src_addr = src;
490 	hwdesc->dst_addr = dst;
491 	hwdesc->transaction_size = size;
492 
493 	/*
494 	 * set config: SAD, DAD, DDS, SDS, etc.
495 	 * Note on transfer sizes: the DMAC can perform unaligned DMA transfers,
496 	 * but it is important to have transaction size a multiple of both
497 	 * receiver and transmitter transfer sizes. It is also possible to use
498 	 * different RAM and device transfer sizes, and it does work well with
499 	 * some devices, e.g. with V08R07S01E SD host controllers, which can use
500 	 * 128 byte transfers. But this doesn't work with other devices,
501 	 * especially when the transaction size is unknown. This is the case,
502 	 * e.g. with serial drivers like amba-pl011.c. For reception it sets up
503 	 * the transaction size of 4K and if fewer bytes are received, it
504 	 * pauses DMA and reads out data received via DMA as well as those left
505 	 * in the Rx FIFO. For this to work with the RAM side using burst
506 	 * transfers we enable the SBE bit and terminate the transfer in our
507 	 * DMA_PAUSE handler.
508 	 */
509 	mem_xfer = nbpf_xfer_ds(chan->nbpf, size);
510 
511 	switch (direction) {
512 	case DMA_DEV_TO_MEM:
513 		can_burst = chan->slave_src_width >= 3;
514 		slave_xfer = min(mem_xfer, can_burst ?
515 				 chan->slave_src_burst : chan->slave_src_width);
516 		/*
517 		 * Is the slave narrower than 64 bits, i.e. isn't using the full
518 		 * bus width and cannot use bursts?
519 		 */
520 		if (mem_xfer > chan->slave_src_burst && !can_burst)
521 			mem_xfer = chan->slave_src_burst;
522 		/* Device-to-RAM DMA is unreliable without REQD set */
523 		hwdesc->config = NBPF_CHAN_CFG_SAD | (NBPF_CHAN_CFG_DDS & (mem_xfer << 16)) |
524 			(NBPF_CHAN_CFG_SDS & (slave_xfer << 12)) | NBPF_CHAN_CFG_REQD |
525 			NBPF_CHAN_CFG_SBE;
526 		break;
527 
528 	case DMA_MEM_TO_DEV:
529 		slave_xfer = min(mem_xfer, chan->slave_dst_width >= 3 ?
530 				 chan->slave_dst_burst : chan->slave_dst_width);
531 		hwdesc->config = NBPF_CHAN_CFG_DAD | (NBPF_CHAN_CFG_SDS & (mem_xfer << 12)) |
532 			(NBPF_CHAN_CFG_DDS & (slave_xfer << 16)) | NBPF_CHAN_CFG_REQD;
533 		break;
534 
535 	case DMA_MEM_TO_MEM:
536 		hwdesc->config = NBPF_CHAN_CFG_TCM | NBPF_CHAN_CFG_TM |
537 			(NBPF_CHAN_CFG_SDS & (mem_xfer << 12)) |
538 			(NBPF_CHAN_CFG_DDS & (mem_xfer << 16));
539 		break;
540 
541 	default:
542 		return -EINVAL;
543 	}
544 
545 	hwdesc->config |= chan->dmarq_cfg | (last ? 0 : NBPF_CHAN_CFG_DEM) |
546 		NBPF_CHAN_CFG_DMS;
547 
548 	dev_dbg(dev, "%s(): desc @ %pad: hdr 0x%x, cfg 0x%x, %zu @ %pad -> %pad\n",
549 		__func__, &ldesc->hwdesc_dma_addr, hwdesc->header,
550 		hwdesc->config, size, &src, &dst);
551 
552 	dma_sync_single_for_device(dev, ldesc->hwdesc_dma_addr, sizeof(*hwdesc),
553 				   DMA_TO_DEVICE);
554 
555 	return 0;
556 }
557 
558 static size_t nbpf_bytes_left(struct nbpf_channel *chan)
559 {
560 	return nbpf_chan_read(chan, NBPF_CHAN_CUR_TR_BYTE);
561 }
562 
563 static void nbpf_configure(struct nbpf_device *nbpf)
564 {
565 	nbpf_write(nbpf, NBPF_CTRL, NBPF_CTRL_LVINT);
566 }
567 
568 static void nbpf_pause(struct nbpf_channel *chan)
569 {
570 	nbpf_chan_write(chan, NBPF_CHAN_CTRL, NBPF_CHAN_CTRL_SETSUS);
571 	/* See comment in nbpf_prep_one() */
572 	nbpf_chan_write(chan, NBPF_CHAN_CTRL, NBPF_CHAN_CTRL_CLREN);
573 }
574 
575 /*		Generic part			*/
576 
577 /* DMA ENGINE functions */
578 static void nbpf_issue_pending(struct dma_chan *dchan)
579 {
580 	struct nbpf_channel *chan = nbpf_to_chan(dchan);
581 	unsigned long flags;
582 
583 	dev_dbg(dchan->device->dev, "Entry %s()\n", __func__);
584 
585 	spin_lock_irqsave(&chan->lock, flags);
586 	if (list_empty(&chan->queued))
587 		goto unlock;
588 
589 	list_splice_tail_init(&chan->queued, &chan->active);
590 
591 	if (!chan->running) {
592 		struct nbpf_desc *desc = list_first_entry(&chan->active,
593 						struct nbpf_desc, node);
594 		if (!nbpf_start(desc))
595 			chan->running = desc;
596 	}
597 
598 unlock:
599 	spin_unlock_irqrestore(&chan->lock, flags);
600 }
601 
602 static enum dma_status nbpf_tx_status(struct dma_chan *dchan,
603 		dma_cookie_t cookie, struct dma_tx_state *state)
604 {
605 	struct nbpf_channel *chan = nbpf_to_chan(dchan);
606 	enum dma_status status = dma_cookie_status(dchan, cookie, state);
607 
608 	if (state) {
609 		dma_cookie_t running;
610 		unsigned long flags;
611 
612 		spin_lock_irqsave(&chan->lock, flags);
613 		running = chan->running ? chan->running->async_tx.cookie : -EINVAL;
614 
615 		if (cookie == running) {
616 			state->residue = nbpf_bytes_left(chan);
617 			dev_dbg(dchan->device->dev, "%s(): residue %u\n", __func__,
618 				state->residue);
619 		} else if (status == DMA_IN_PROGRESS) {
620 			struct nbpf_desc *desc;
621 			bool found = false;
622 
623 			list_for_each_entry(desc, &chan->active, node)
624 				if (desc->async_tx.cookie == cookie) {
625 					found = true;
626 					break;
627 				}
628 
629 			if (!found)
630 				list_for_each_entry(desc, &chan->queued, node)
631 					if (desc->async_tx.cookie == cookie) {
632 						found = true;
633 						break;
634 
635 					}
636 
637 			state->residue = found ? desc->length : 0;
638 		}
639 
640 		spin_unlock_irqrestore(&chan->lock, flags);
641 	}
642 
643 	if (chan->paused)
644 		status = DMA_PAUSED;
645 
646 	return status;
647 }
648 
649 static dma_cookie_t nbpf_tx_submit(struct dma_async_tx_descriptor *tx)
650 {
651 	struct nbpf_desc *desc = container_of(tx, struct nbpf_desc, async_tx);
652 	struct nbpf_channel *chan = desc->chan;
653 	unsigned long flags;
654 	dma_cookie_t cookie;
655 
656 	spin_lock_irqsave(&chan->lock, flags);
657 	cookie = dma_cookie_assign(tx);
658 	list_add_tail(&desc->node, &chan->queued);
659 	spin_unlock_irqrestore(&chan->lock, flags);
660 
661 	dev_dbg(chan->dma_chan.device->dev, "Entry %s(%d)\n", __func__, cookie);
662 
663 	return cookie;
664 }
665 
666 static int nbpf_desc_page_alloc(struct nbpf_channel *chan)
667 {
668 	struct dma_chan *dchan = &chan->dma_chan;
669 	struct nbpf_desc_page *dpage = (void *)get_zeroed_page(GFP_KERNEL | GFP_DMA);
670 	struct nbpf_link_desc *ldesc;
671 	struct nbpf_link_reg *hwdesc;
672 	struct nbpf_desc *desc;
673 	LIST_HEAD(head);
674 	LIST_HEAD(lhead);
675 	int i;
676 	struct device *dev = dchan->device->dev;
677 
678 	if (!dpage)
679 		return -ENOMEM;
680 
681 	dev_dbg(dev, "%s(): alloc %lu descriptors, %lu segments, total alloc %zu\n",
682 		__func__, NBPF_DESCS_PER_PAGE, NBPF_SEGMENTS_PER_PAGE, sizeof(*dpage));
683 
684 	for (i = 0, ldesc = dpage->ldesc, hwdesc = dpage->hwdesc;
685 	     i < ARRAY_SIZE(dpage->ldesc);
686 	     i++, ldesc++, hwdesc++) {
687 		ldesc->hwdesc = hwdesc;
688 		list_add_tail(&ldesc->node, &lhead);
689 		ldesc->hwdesc_dma_addr = dma_map_single(dchan->device->dev,
690 					hwdesc, sizeof(*hwdesc), DMA_TO_DEVICE);
691 
692 		dev_dbg(dev, "%s(): mapped 0x%p to %pad\n", __func__,
693 			hwdesc, &ldesc->hwdesc_dma_addr);
694 	}
695 
696 	for (i = 0, desc = dpage->desc;
697 	     i < ARRAY_SIZE(dpage->desc);
698 	     i++, desc++) {
699 		dma_async_tx_descriptor_init(&desc->async_tx, dchan);
700 		desc->async_tx.tx_submit = nbpf_tx_submit;
701 		desc->chan = chan;
702 		INIT_LIST_HEAD(&desc->sg);
703 		list_add_tail(&desc->node, &head);
704 	}
705 
706 	/*
707 	 * This function cannot be called from interrupt context, so, no need to
708 	 * save flags
709 	 */
710 	spin_lock_irq(&chan->lock);
711 	list_splice_tail(&lhead, &chan->free_links);
712 	list_splice_tail(&head, &chan->free);
713 	list_add(&dpage->node, &chan->desc_page);
714 	spin_unlock_irq(&chan->lock);
715 
716 	return ARRAY_SIZE(dpage->desc);
717 }
718 
719 static void nbpf_desc_put(struct nbpf_desc *desc)
720 {
721 	struct nbpf_channel *chan = desc->chan;
722 	struct nbpf_link_desc *ldesc, *tmp;
723 	unsigned long flags;
724 
725 	spin_lock_irqsave(&chan->lock, flags);
726 	list_for_each_entry_safe(ldesc, tmp, &desc->sg, node)
727 		list_move(&ldesc->node, &chan->free_links);
728 
729 	list_add(&desc->node, &chan->free);
730 	spin_unlock_irqrestore(&chan->lock, flags);
731 }
732 
733 static void nbpf_scan_acked(struct nbpf_channel *chan)
734 {
735 	struct nbpf_desc *desc, *tmp;
736 	unsigned long flags;
737 	LIST_HEAD(head);
738 
739 	spin_lock_irqsave(&chan->lock, flags);
740 	list_for_each_entry_safe(desc, tmp, &chan->done, node)
741 		if (async_tx_test_ack(&desc->async_tx) && desc->user_wait) {
742 			list_move(&desc->node, &head);
743 			desc->user_wait = false;
744 		}
745 	spin_unlock_irqrestore(&chan->lock, flags);
746 
747 	list_for_each_entry_safe(desc, tmp, &head, node) {
748 		list_del(&desc->node);
749 		nbpf_desc_put(desc);
750 	}
751 }
752 
753 /*
754  * We have to allocate descriptors with the channel lock dropped. This means,
755  * before we re-acquire the lock buffers can be taken already, so we have to
756  * re-check after re-acquiring the lock and possibly retry, if buffers are gone
757  * again.
758  */
759 static struct nbpf_desc *nbpf_desc_get(struct nbpf_channel *chan, size_t len)
760 {
761 	struct nbpf_desc *desc = NULL;
762 	struct nbpf_link_desc *ldesc, *prev = NULL;
763 
764 	nbpf_scan_acked(chan);
765 
766 	spin_lock_irq(&chan->lock);
767 
768 	do {
769 		int i = 0, ret;
770 
771 		if (list_empty(&chan->free)) {
772 			/* No more free descriptors */
773 			spin_unlock_irq(&chan->lock);
774 			ret = nbpf_desc_page_alloc(chan);
775 			if (ret < 0)
776 				return NULL;
777 			spin_lock_irq(&chan->lock);
778 			continue;
779 		}
780 		desc = list_first_entry(&chan->free, struct nbpf_desc, node);
781 		list_del(&desc->node);
782 
783 		do {
784 			if (list_empty(&chan->free_links)) {
785 				/* No more free link descriptors */
786 				spin_unlock_irq(&chan->lock);
787 				ret = nbpf_desc_page_alloc(chan);
788 				if (ret < 0) {
789 					nbpf_desc_put(desc);
790 					return NULL;
791 				}
792 				spin_lock_irq(&chan->lock);
793 				continue;
794 			}
795 
796 			ldesc = list_first_entry(&chan->free_links,
797 						 struct nbpf_link_desc, node);
798 			ldesc->desc = desc;
799 			if (prev)
800 				prev->hwdesc->next = (u32)ldesc->hwdesc_dma_addr;
801 
802 			prev = ldesc;
803 			list_move_tail(&ldesc->node, &desc->sg);
804 
805 			i++;
806 		} while (i < len);
807 	} while (!desc);
808 
809 	prev->hwdesc->next = 0;
810 
811 	spin_unlock_irq(&chan->lock);
812 
813 	return desc;
814 }
815 
816 static void nbpf_chan_idle(struct nbpf_channel *chan)
817 {
818 	struct nbpf_desc *desc, *tmp;
819 	unsigned long flags;
820 	LIST_HEAD(head);
821 
822 	spin_lock_irqsave(&chan->lock, flags);
823 
824 	list_splice_init(&chan->done, &head);
825 	list_splice_init(&chan->active, &head);
826 	list_splice_init(&chan->queued, &head);
827 
828 	chan->running = NULL;
829 
830 	spin_unlock_irqrestore(&chan->lock, flags);
831 
832 	list_for_each_entry_safe(desc, tmp, &head, node) {
833 		dev_dbg(chan->nbpf->dma_dev.dev, "%s(): force-free desc %p cookie %d\n",
834 			__func__, desc, desc->async_tx.cookie);
835 		list_del(&desc->node);
836 		nbpf_desc_put(desc);
837 	}
838 }
839 
840 static int nbpf_control(struct dma_chan *dchan, enum dma_ctrl_cmd cmd,
841 			unsigned long arg)
842 {
843 	struct nbpf_channel *chan = nbpf_to_chan(dchan);
844 	struct dma_slave_config *config;
845 
846 	dev_dbg(dchan->device->dev, "Entry %s(%d)\n", __func__, cmd);
847 
848 	switch (cmd) {
849 	case DMA_TERMINATE_ALL:
850 		dev_dbg(dchan->device->dev, "Terminating\n");
851 		nbpf_chan_halt(chan);
852 		nbpf_chan_idle(chan);
853 		break;
854 
855 	case DMA_SLAVE_CONFIG:
856 		if (!arg)
857 			return -EINVAL;
858 		config = (struct dma_slave_config *)arg;
859 
860 		/*
861 		 * We could check config->slave_id to match chan->terminal here,
862 		 * but with DT they would be coming from the same source, so
863 		 * such a check would be superflous
864 		 */
865 
866 		chan->slave_dst_addr = config->dst_addr;
867 		chan->slave_dst_width = nbpf_xfer_size(chan->nbpf,
868 						       config->dst_addr_width, 1);
869 		chan->slave_dst_burst = nbpf_xfer_size(chan->nbpf,
870 						       config->dst_addr_width,
871 						       config->dst_maxburst);
872 		chan->slave_src_addr = config->src_addr;
873 		chan->slave_src_width = nbpf_xfer_size(chan->nbpf,
874 						       config->src_addr_width, 1);
875 		chan->slave_src_burst = nbpf_xfer_size(chan->nbpf,
876 						       config->src_addr_width,
877 						       config->src_maxburst);
878 		break;
879 
880 	case DMA_PAUSE:
881 		chan->paused = true;
882 		nbpf_pause(chan);
883 		break;
884 
885 	default:
886 		return -ENXIO;
887 	}
888 
889 	return 0;
890 }
891 
892 static struct dma_async_tx_descriptor *nbpf_prep_sg(struct nbpf_channel *chan,
893 		struct scatterlist *src_sg, struct scatterlist *dst_sg,
894 		size_t len, enum dma_transfer_direction direction,
895 		unsigned long flags)
896 {
897 	struct nbpf_link_desc *ldesc;
898 	struct scatterlist *mem_sg;
899 	struct nbpf_desc *desc;
900 	bool inc_src, inc_dst;
901 	size_t data_len = 0;
902 	int i = 0;
903 
904 	switch (direction) {
905 	case DMA_DEV_TO_MEM:
906 		mem_sg = dst_sg;
907 		inc_src = false;
908 		inc_dst = true;
909 		break;
910 
911 	case DMA_MEM_TO_DEV:
912 		mem_sg = src_sg;
913 		inc_src = true;
914 		inc_dst = false;
915 		break;
916 
917 	default:
918 	case DMA_MEM_TO_MEM:
919 		mem_sg = src_sg;
920 		inc_src = true;
921 		inc_dst = true;
922 	}
923 
924 	desc = nbpf_desc_get(chan, len);
925 	if (!desc)
926 		return NULL;
927 
928 	desc->async_tx.flags = flags;
929 	desc->async_tx.cookie = -EBUSY;
930 	desc->user_wait = false;
931 
932 	/*
933 	 * This is a private descriptor list, and we own the descriptor. No need
934 	 * to lock.
935 	 */
936 	list_for_each_entry(ldesc, &desc->sg, node) {
937 		int ret = nbpf_prep_one(ldesc, direction,
938 					sg_dma_address(src_sg),
939 					sg_dma_address(dst_sg),
940 					sg_dma_len(mem_sg),
941 					i == len - 1);
942 		if (ret < 0) {
943 			nbpf_desc_put(desc);
944 			return NULL;
945 		}
946 		data_len += sg_dma_len(mem_sg);
947 		if (inc_src)
948 			src_sg = sg_next(src_sg);
949 		if (inc_dst)
950 			dst_sg = sg_next(dst_sg);
951 		mem_sg = direction == DMA_DEV_TO_MEM ? dst_sg : src_sg;
952 		i++;
953 	}
954 
955 	desc->length = data_len;
956 
957 	/* The user has to return the descriptor to us ASAP via .tx_submit() */
958 	return &desc->async_tx;
959 }
960 
961 static struct dma_async_tx_descriptor *nbpf_prep_memcpy(
962 	struct dma_chan *dchan, dma_addr_t dst, dma_addr_t src,
963 	size_t len, unsigned long flags)
964 {
965 	struct nbpf_channel *chan = nbpf_to_chan(dchan);
966 	struct scatterlist dst_sg;
967 	struct scatterlist src_sg;
968 
969 	sg_init_table(&dst_sg, 1);
970 	sg_init_table(&src_sg, 1);
971 
972 	sg_dma_address(&dst_sg) = dst;
973 	sg_dma_address(&src_sg) = src;
974 
975 	sg_dma_len(&dst_sg) = len;
976 	sg_dma_len(&src_sg) = len;
977 
978 	dev_dbg(dchan->device->dev, "%s(): %zu @ %pad -> %pad\n",
979 		__func__, len, &src, &dst);
980 
981 	return nbpf_prep_sg(chan, &src_sg, &dst_sg, 1,
982 			    DMA_MEM_TO_MEM, flags);
983 }
984 
985 static struct dma_async_tx_descriptor *nbpf_prep_memcpy_sg(
986 	struct dma_chan *dchan,
987 	struct scatterlist *dst_sg, unsigned int dst_nents,
988 	struct scatterlist *src_sg, unsigned int src_nents,
989 	unsigned long flags)
990 {
991 	struct nbpf_channel *chan = nbpf_to_chan(dchan);
992 
993 	if (dst_nents != src_nents)
994 		return NULL;
995 
996 	return nbpf_prep_sg(chan, src_sg, dst_sg, src_nents,
997 			    DMA_MEM_TO_MEM, flags);
998 }
999 
1000 static struct dma_async_tx_descriptor *nbpf_prep_slave_sg(
1001 	struct dma_chan *dchan, struct scatterlist *sgl, unsigned int sg_len,
1002 	enum dma_transfer_direction direction, unsigned long flags, void *context)
1003 {
1004 	struct nbpf_channel *chan = nbpf_to_chan(dchan);
1005 	struct scatterlist slave_sg;
1006 
1007 	dev_dbg(dchan->device->dev, "Entry %s()\n", __func__);
1008 
1009 	sg_init_table(&slave_sg, 1);
1010 
1011 	switch (direction) {
1012 	case DMA_MEM_TO_DEV:
1013 		sg_dma_address(&slave_sg) = chan->slave_dst_addr;
1014 		return nbpf_prep_sg(chan, sgl, &slave_sg, sg_len,
1015 				    direction, flags);
1016 
1017 	case DMA_DEV_TO_MEM:
1018 		sg_dma_address(&slave_sg) = chan->slave_src_addr;
1019 		return nbpf_prep_sg(chan, &slave_sg, sgl, sg_len,
1020 				    direction, flags);
1021 
1022 	default:
1023 		return NULL;
1024 	}
1025 }
1026 
1027 static int nbpf_alloc_chan_resources(struct dma_chan *dchan)
1028 {
1029 	struct nbpf_channel *chan = nbpf_to_chan(dchan);
1030 	int ret;
1031 
1032 	INIT_LIST_HEAD(&chan->free);
1033 	INIT_LIST_HEAD(&chan->free_links);
1034 	INIT_LIST_HEAD(&chan->queued);
1035 	INIT_LIST_HEAD(&chan->active);
1036 	INIT_LIST_HEAD(&chan->done);
1037 
1038 	ret = nbpf_desc_page_alloc(chan);
1039 	if (ret < 0)
1040 		return ret;
1041 
1042 	dev_dbg(dchan->device->dev, "Entry %s(): terminal %u\n", __func__,
1043 		chan->terminal);
1044 
1045 	nbpf_chan_configure(chan);
1046 
1047 	return ret;
1048 }
1049 
1050 static void nbpf_free_chan_resources(struct dma_chan *dchan)
1051 {
1052 	struct nbpf_channel *chan = nbpf_to_chan(dchan);
1053 	struct nbpf_desc_page *dpage, *tmp;
1054 
1055 	dev_dbg(dchan->device->dev, "Entry %s()\n", __func__);
1056 
1057 	nbpf_chan_halt(chan);
1058 	nbpf_chan_idle(chan);
1059 	/* Clean up for if a channel is re-used for MEMCPY after slave DMA */
1060 	nbpf_chan_prepare_default(chan);
1061 
1062 	list_for_each_entry_safe(dpage, tmp, &chan->desc_page, node) {
1063 		struct nbpf_link_desc *ldesc;
1064 		int i;
1065 		list_del(&dpage->node);
1066 		for (i = 0, ldesc = dpage->ldesc;
1067 		     i < ARRAY_SIZE(dpage->ldesc);
1068 		     i++, ldesc++)
1069 			dma_unmap_single(dchan->device->dev, ldesc->hwdesc_dma_addr,
1070 					 sizeof(*ldesc->hwdesc), DMA_TO_DEVICE);
1071 		free_page((unsigned long)dpage);
1072 	}
1073 }
1074 
1075 static int nbpf_slave_caps(struct dma_chan *dchan,
1076 			   struct dma_slave_caps *caps)
1077 {
1078 	caps->src_addr_widths = NBPF_DMA_BUSWIDTHS;
1079 	caps->dstn_addr_widths = NBPF_DMA_BUSWIDTHS;
1080 	caps->directions = BIT(DMA_DEV_TO_MEM) | BIT(DMA_MEM_TO_DEV);
1081 	caps->cmd_pause = false;
1082 	caps->cmd_terminate = true;
1083 
1084 	return 0;
1085 }
1086 
1087 static struct dma_chan *nbpf_of_xlate(struct of_phandle_args *dma_spec,
1088 				      struct of_dma *ofdma)
1089 {
1090 	struct nbpf_device *nbpf = ofdma->of_dma_data;
1091 	struct dma_chan *dchan;
1092 	struct nbpf_channel *chan;
1093 
1094 	if (dma_spec->args_count != 2)
1095 		return NULL;
1096 
1097 	dchan = dma_get_any_slave_channel(&nbpf->dma_dev);
1098 	if (!dchan)
1099 		return NULL;
1100 
1101 	dev_dbg(dchan->device->dev, "Entry %s(%s)\n", __func__,
1102 		dma_spec->np->name);
1103 
1104 	chan = nbpf_to_chan(dchan);
1105 
1106 	chan->terminal = dma_spec->args[0];
1107 	chan->flags = dma_spec->args[1];
1108 
1109 	nbpf_chan_prepare(chan);
1110 	nbpf_chan_configure(chan);
1111 
1112 	return dchan;
1113 }
1114 
1115 static void nbpf_chan_tasklet(unsigned long data)
1116 {
1117 	struct nbpf_channel *chan = (struct nbpf_channel *)data;
1118 	struct nbpf_desc *desc, *tmp;
1119 	dma_async_tx_callback callback;
1120 	void *param;
1121 
1122 	while (!list_empty(&chan->done)) {
1123 		bool found = false, must_put, recycling = false;
1124 
1125 		spin_lock_irq(&chan->lock);
1126 
1127 		list_for_each_entry_safe(desc, tmp, &chan->done, node) {
1128 			if (!desc->user_wait) {
1129 				/* Newly completed descriptor, have to process */
1130 				found = true;
1131 				break;
1132 			} else if (async_tx_test_ack(&desc->async_tx)) {
1133 				/*
1134 				 * This descriptor was waiting for a user ACK,
1135 				 * it can be recycled now.
1136 				 */
1137 				list_del(&desc->node);
1138 				spin_unlock_irq(&chan->lock);
1139 				nbpf_desc_put(desc);
1140 				recycling = true;
1141 				break;
1142 			}
1143 		}
1144 
1145 		if (recycling)
1146 			continue;
1147 
1148 		if (!found) {
1149 			/* This can happen if TERMINATE_ALL has been called */
1150 			spin_unlock_irq(&chan->lock);
1151 			break;
1152 		}
1153 
1154 		dma_cookie_complete(&desc->async_tx);
1155 
1156 		/*
1157 		 * With released lock we cannot dereference desc, maybe it's
1158 		 * still on the "done" list
1159 		 */
1160 		if (async_tx_test_ack(&desc->async_tx)) {
1161 			list_del(&desc->node);
1162 			must_put = true;
1163 		} else {
1164 			desc->user_wait = true;
1165 			must_put = false;
1166 		}
1167 
1168 		callback = desc->async_tx.callback;
1169 		param = desc->async_tx.callback_param;
1170 
1171 		/* ack and callback completed descriptor */
1172 		spin_unlock_irq(&chan->lock);
1173 
1174 		if (callback)
1175 			callback(param);
1176 
1177 		if (must_put)
1178 			nbpf_desc_put(desc);
1179 	}
1180 }
1181 
1182 static irqreturn_t nbpf_chan_irq(int irq, void *dev)
1183 {
1184 	struct nbpf_channel *chan = dev;
1185 	bool done = nbpf_status_get(chan);
1186 	struct nbpf_desc *desc;
1187 	irqreturn_t ret;
1188 	bool bh = false;
1189 
1190 	if (!done)
1191 		return IRQ_NONE;
1192 
1193 	nbpf_status_ack(chan);
1194 
1195 	dev_dbg(&chan->dma_chan.dev->device, "%s()\n", __func__);
1196 
1197 	spin_lock(&chan->lock);
1198 	desc = chan->running;
1199 	if (WARN_ON(!desc)) {
1200 		ret = IRQ_NONE;
1201 		goto unlock;
1202 	} else {
1203 		ret = IRQ_HANDLED;
1204 		bh = true;
1205 	}
1206 
1207 	list_move_tail(&desc->node, &chan->done);
1208 	chan->running = NULL;
1209 
1210 	if (!list_empty(&chan->active)) {
1211 		desc = list_first_entry(&chan->active,
1212 					struct nbpf_desc, node);
1213 		if (!nbpf_start(desc))
1214 			chan->running = desc;
1215 	}
1216 
1217 unlock:
1218 	spin_unlock(&chan->lock);
1219 
1220 	if (bh)
1221 		tasklet_schedule(&chan->tasklet);
1222 
1223 	return ret;
1224 }
1225 
1226 static irqreturn_t nbpf_err_irq(int irq, void *dev)
1227 {
1228 	struct nbpf_device *nbpf = dev;
1229 	u32 error = nbpf_error_get(nbpf);
1230 
1231 	dev_warn(nbpf->dma_dev.dev, "DMA error IRQ %u\n", irq);
1232 
1233 	if (!error)
1234 		return IRQ_NONE;
1235 
1236 	do {
1237 		struct nbpf_channel *chan = nbpf_error_get_channel(nbpf, error);
1238 		/* On error: abort all queued transfers, no callback */
1239 		nbpf_error_clear(chan);
1240 		nbpf_chan_idle(chan);
1241 		error = nbpf_error_get(nbpf);
1242 	} while (error);
1243 
1244 	return IRQ_HANDLED;
1245 }
1246 
1247 static int nbpf_chan_probe(struct nbpf_device *nbpf, int n)
1248 {
1249 	struct dma_device *dma_dev = &nbpf->dma_dev;
1250 	struct nbpf_channel *chan = nbpf->chan + n;
1251 	int ret;
1252 
1253 	chan->nbpf = nbpf;
1254 	chan->base = nbpf->base + NBPF_REG_CHAN_OFFSET + NBPF_REG_CHAN_SIZE * n;
1255 	INIT_LIST_HEAD(&chan->desc_page);
1256 	spin_lock_init(&chan->lock);
1257 	chan->dma_chan.device = dma_dev;
1258 	dma_cookie_init(&chan->dma_chan);
1259 	nbpf_chan_prepare_default(chan);
1260 
1261 	dev_dbg(dma_dev->dev, "%s(): channel %d: -> %p\n", __func__, n, chan->base);
1262 
1263 	snprintf(chan->name, sizeof(chan->name), "nbpf %d", n);
1264 
1265 	tasklet_init(&chan->tasklet, nbpf_chan_tasklet, (unsigned long)chan);
1266 	ret = devm_request_irq(dma_dev->dev, chan->irq,
1267 			nbpf_chan_irq, IRQF_SHARED,
1268 			chan->name, chan);
1269 	if (ret < 0)
1270 		return ret;
1271 
1272 	/* Add the channel to DMA device channel list */
1273 	list_add_tail(&chan->dma_chan.device_node,
1274 		      &dma_dev->channels);
1275 
1276 	return 0;
1277 }
1278 
1279 static const struct of_device_id nbpf_match[] = {
1280 	{.compatible = "renesas,nbpfaxi64dmac1b4",	.data = &nbpf_cfg[NBPF1B4]},
1281 	{.compatible = "renesas,nbpfaxi64dmac1b8",	.data = &nbpf_cfg[NBPF1B8]},
1282 	{.compatible = "renesas,nbpfaxi64dmac1b16",	.data = &nbpf_cfg[NBPF1B16]},
1283 	{.compatible = "renesas,nbpfaxi64dmac4b4",	.data = &nbpf_cfg[NBPF4B4]},
1284 	{.compatible = "renesas,nbpfaxi64dmac4b8",	.data = &nbpf_cfg[NBPF4B8]},
1285 	{.compatible = "renesas,nbpfaxi64dmac4b16",	.data = &nbpf_cfg[NBPF4B16]},
1286 	{.compatible = "renesas,nbpfaxi64dmac8b4",	.data = &nbpf_cfg[NBPF8B4]},
1287 	{.compatible = "renesas,nbpfaxi64dmac8b8",	.data = &nbpf_cfg[NBPF8B8]},
1288 	{.compatible = "renesas,nbpfaxi64dmac8b16",	.data = &nbpf_cfg[NBPF8B16]},
1289 	{}
1290 };
1291 MODULE_DEVICE_TABLE(of, nbpf_match);
1292 
1293 static int nbpf_probe(struct platform_device *pdev)
1294 {
1295 	struct device *dev = &pdev->dev;
1296 	const struct of_device_id *of_id = of_match_device(nbpf_match, dev);
1297 	struct device_node *np = dev->of_node;
1298 	struct nbpf_device *nbpf;
1299 	struct dma_device *dma_dev;
1300 	struct resource *iomem, *irq_res;
1301 	const struct nbpf_config *cfg;
1302 	int num_channels;
1303 	int ret, irq, eirq, i;
1304 	int irqbuf[9] /* maximum 8 channels + error IRQ */;
1305 	unsigned int irqs = 0;
1306 
1307 	BUILD_BUG_ON(sizeof(struct nbpf_desc_page) > PAGE_SIZE);
1308 
1309 	/* DT only */
1310 	if (!np || !of_id || !of_id->data)
1311 		return -ENODEV;
1312 
1313 	cfg = of_id->data;
1314 	num_channels = cfg->num_channels;
1315 
1316 	nbpf = devm_kzalloc(dev, sizeof(*nbpf) + num_channels *
1317 			    sizeof(nbpf->chan[0]), GFP_KERNEL);
1318 	if (!nbpf) {
1319 		dev_err(dev, "Memory allocation failed\n");
1320 		return -ENOMEM;
1321 	}
1322 	dma_dev = &nbpf->dma_dev;
1323 	dma_dev->dev = dev;
1324 
1325 	iomem = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1326 	nbpf->base = devm_ioremap_resource(dev, iomem);
1327 	if (IS_ERR(nbpf->base))
1328 		return PTR_ERR(nbpf->base);
1329 
1330 	nbpf->clk = devm_clk_get(dev, NULL);
1331 	if (IS_ERR(nbpf->clk))
1332 		return PTR_ERR(nbpf->clk);
1333 
1334 	nbpf->config = cfg;
1335 
1336 	for (i = 0; irqs < ARRAY_SIZE(irqbuf); i++) {
1337 		irq_res = platform_get_resource(pdev, IORESOURCE_IRQ, i);
1338 		if (!irq_res)
1339 			break;
1340 
1341 		for (irq = irq_res->start; irq <= irq_res->end;
1342 		     irq++, irqs++)
1343 			irqbuf[irqs] = irq;
1344 	}
1345 
1346 	/*
1347 	 * 3 IRQ resource schemes are supported:
1348 	 * 1. 1 shared IRQ for error and all channels
1349 	 * 2. 2 IRQs: one for error and one shared for all channels
1350 	 * 3. 1 IRQ for error and an own IRQ for each channel
1351 	 */
1352 	if (irqs != 1 && irqs != 2 && irqs != num_channels + 1)
1353 		return -ENXIO;
1354 
1355 	if (irqs == 1) {
1356 		eirq = irqbuf[0];
1357 
1358 		for (i = 0; i <= num_channels; i++)
1359 			nbpf->chan[i].irq = irqbuf[0];
1360 	} else {
1361 		eirq = platform_get_irq_byname(pdev, "error");
1362 		if (eirq < 0)
1363 			return eirq;
1364 
1365 		if (irqs == num_channels + 1) {
1366 			struct nbpf_channel *chan;
1367 
1368 			for (i = 0, chan = nbpf->chan; i <= num_channels;
1369 			     i++, chan++) {
1370 				/* Skip the error IRQ */
1371 				if (irqbuf[i] == eirq)
1372 					i++;
1373 				chan->irq = irqbuf[i];
1374 			}
1375 
1376 			if (chan != nbpf->chan + num_channels)
1377 				return -EINVAL;
1378 		} else {
1379 			/* 2 IRQs and more than one channel */
1380 			if (irqbuf[0] == eirq)
1381 				irq = irqbuf[1];
1382 			else
1383 				irq = irqbuf[0];
1384 
1385 			for (i = 0; i <= num_channels; i++)
1386 				nbpf->chan[i].irq = irq;
1387 		}
1388 	}
1389 
1390 	ret = devm_request_irq(dev, eirq, nbpf_err_irq,
1391 			       IRQF_SHARED, "dma error", nbpf);
1392 	if (ret < 0)
1393 		return ret;
1394 
1395 	INIT_LIST_HEAD(&dma_dev->channels);
1396 
1397 	/* Create DMA Channel */
1398 	for (i = 0; i < num_channels; i++) {
1399 		ret = nbpf_chan_probe(nbpf, i);
1400 		if (ret < 0)
1401 			return ret;
1402 	}
1403 
1404 	dma_cap_set(DMA_MEMCPY, dma_dev->cap_mask);
1405 	dma_cap_set(DMA_SLAVE, dma_dev->cap_mask);
1406 	dma_cap_set(DMA_PRIVATE, dma_dev->cap_mask);
1407 	dma_cap_set(DMA_SG, dma_dev->cap_mask);
1408 
1409 	/* Common and MEMCPY operations */
1410 	dma_dev->device_alloc_chan_resources
1411 		= nbpf_alloc_chan_resources;
1412 	dma_dev->device_free_chan_resources = nbpf_free_chan_resources;
1413 	dma_dev->device_prep_dma_sg = nbpf_prep_memcpy_sg;
1414 	dma_dev->device_prep_dma_memcpy = nbpf_prep_memcpy;
1415 	dma_dev->device_tx_status = nbpf_tx_status;
1416 	dma_dev->device_issue_pending = nbpf_issue_pending;
1417 	dma_dev->device_slave_caps = nbpf_slave_caps;
1418 
1419 	/*
1420 	 * If we drop support for unaligned MEMCPY buffer addresses and / or
1421 	 * lengths by setting
1422 	 * dma_dev->copy_align = 4;
1423 	 * then we can set transfer length to 4 bytes in nbpf_prep_one() for
1424 	 * DMA_MEM_TO_MEM
1425 	 */
1426 
1427 	/* Compulsory for DMA_SLAVE fields */
1428 	dma_dev->device_prep_slave_sg = nbpf_prep_slave_sg;
1429 	dma_dev->device_control = nbpf_control;
1430 
1431 	platform_set_drvdata(pdev, nbpf);
1432 
1433 	ret = clk_prepare_enable(nbpf->clk);
1434 	if (ret < 0)
1435 		return ret;
1436 
1437 	nbpf_configure(nbpf);
1438 
1439 	ret = dma_async_device_register(dma_dev);
1440 	if (ret < 0)
1441 		goto e_clk_off;
1442 
1443 	ret = of_dma_controller_register(np, nbpf_of_xlate, nbpf);
1444 	if (ret < 0)
1445 		goto e_dma_dev_unreg;
1446 
1447 	return 0;
1448 
1449 e_dma_dev_unreg:
1450 	dma_async_device_unregister(dma_dev);
1451 e_clk_off:
1452 	clk_disable_unprepare(nbpf->clk);
1453 
1454 	return ret;
1455 }
1456 
1457 static int nbpf_remove(struct platform_device *pdev)
1458 {
1459 	struct nbpf_device *nbpf = platform_get_drvdata(pdev);
1460 
1461 	of_dma_controller_free(pdev->dev.of_node);
1462 	dma_async_device_unregister(&nbpf->dma_dev);
1463 	clk_disable_unprepare(nbpf->clk);
1464 
1465 	return 0;
1466 }
1467 
1468 static struct platform_device_id nbpf_ids[] = {
1469 	{"nbpfaxi64dmac1b4",	(kernel_ulong_t)&nbpf_cfg[NBPF1B4]},
1470 	{"nbpfaxi64dmac1b8",	(kernel_ulong_t)&nbpf_cfg[NBPF1B8]},
1471 	{"nbpfaxi64dmac1b16",	(kernel_ulong_t)&nbpf_cfg[NBPF1B16]},
1472 	{"nbpfaxi64dmac4b4",	(kernel_ulong_t)&nbpf_cfg[NBPF4B4]},
1473 	{"nbpfaxi64dmac4b8",	(kernel_ulong_t)&nbpf_cfg[NBPF4B8]},
1474 	{"nbpfaxi64dmac4b16",	(kernel_ulong_t)&nbpf_cfg[NBPF4B16]},
1475 	{"nbpfaxi64dmac8b4",	(kernel_ulong_t)&nbpf_cfg[NBPF8B4]},
1476 	{"nbpfaxi64dmac8b8",	(kernel_ulong_t)&nbpf_cfg[NBPF8B8]},
1477 	{"nbpfaxi64dmac8b16",	(kernel_ulong_t)&nbpf_cfg[NBPF8B16]},
1478 	{},
1479 };
1480 MODULE_DEVICE_TABLE(platform, nbpf_ids);
1481 
1482 #ifdef CONFIG_PM
1483 static int nbpf_runtime_suspend(struct device *dev)
1484 {
1485 	struct nbpf_device *nbpf = platform_get_drvdata(to_platform_device(dev));
1486 	clk_disable_unprepare(nbpf->clk);
1487 	return 0;
1488 }
1489 
1490 static int nbpf_runtime_resume(struct device *dev)
1491 {
1492 	struct nbpf_device *nbpf = platform_get_drvdata(to_platform_device(dev));
1493 	return clk_prepare_enable(nbpf->clk);
1494 }
1495 #endif
1496 
1497 static const struct dev_pm_ops nbpf_pm_ops = {
1498 	SET_RUNTIME_PM_OPS(nbpf_runtime_suspend, nbpf_runtime_resume, NULL)
1499 };
1500 
1501 static struct platform_driver nbpf_driver = {
1502 	.driver = {
1503 		.name = "dma-nbpf",
1504 		.of_match_table = nbpf_match,
1505 		.pm = &nbpf_pm_ops,
1506 	},
1507 	.id_table = nbpf_ids,
1508 	.probe = nbpf_probe,
1509 	.remove = nbpf_remove,
1510 };
1511 
1512 module_platform_driver(nbpf_driver);
1513 
1514 MODULE_AUTHOR("Guennadi Liakhovetski <g.liakhovetski@gmx.de>");
1515 MODULE_DESCRIPTION("dmaengine driver for NBPFAXI64* DMACs");
1516 MODULE_LICENSE("GPL v2");
1517