xref: /linux/drivers/dma/ti/cppi41.c (revision b7019ac550eb3916f34d79db583e9b7ea2524afa)
1 // SPDX-License-Identifier: GPL-2.0-only
2 #include <linux/delay.h>
3 #include <linux/dmaengine.h>
4 #include <linux/dma-mapping.h>
5 #include <linux/platform_device.h>
6 #include <linux/module.h>
7 #include <linux/of.h>
8 #include <linux/slab.h>
9 #include <linux/of_dma.h>
10 #include <linux/of_irq.h>
11 #include <linux/dmapool.h>
12 #include <linux/interrupt.h>
13 #include <linux/of_address.h>
14 #include <linux/pm_runtime.h>
15 #include "../dmaengine.h"
16 
17 #define DESC_TYPE	27
18 #define DESC_TYPE_HOST	0x10
19 #define DESC_TYPE_TEARD	0x13
20 
21 #define TD_DESC_IS_RX	(1 << 16)
22 #define TD_DESC_DMA_NUM	10
23 
24 #define DESC_LENGTH_BITS_NUM	21
25 
26 #define DESC_TYPE_USB	(5 << 26)
27 #define DESC_PD_COMPLETE	(1 << 31)
28 
29 /* DMA engine */
30 #define DMA_TDFDQ	4
31 #define DMA_TXGCR(x)	(0x800 + (x) * 0x20)
32 #define DMA_RXGCR(x)	(0x808 + (x) * 0x20)
33 #define RXHPCRA0		4
34 
35 #define GCR_CHAN_ENABLE		(1 << 31)
36 #define GCR_TEARDOWN		(1 << 30)
37 #define GCR_STARV_RETRY		(1 << 24)
38 #define GCR_DESC_TYPE_HOST	(1 << 14)
39 
40 /* DMA scheduler */
41 #define DMA_SCHED_CTRL		0
42 #define DMA_SCHED_CTRL_EN	(1 << 31)
43 #define DMA_SCHED_WORD(x)	((x) * 4 + 0x800)
44 
45 #define SCHED_ENTRY0_CHAN(x)	((x) << 0)
46 #define SCHED_ENTRY0_IS_RX	(1 << 7)
47 
48 #define SCHED_ENTRY1_CHAN(x)	((x) << 8)
49 #define SCHED_ENTRY1_IS_RX	(1 << 15)
50 
51 #define SCHED_ENTRY2_CHAN(x)	((x) << 16)
52 #define SCHED_ENTRY2_IS_RX	(1 << 23)
53 
54 #define SCHED_ENTRY3_CHAN(x)	((x) << 24)
55 #define SCHED_ENTRY3_IS_RX	(1 << 31)
56 
57 /* Queue manager */
58 /* 4 KiB of memory for descriptors, 2 for each endpoint */
59 #define ALLOC_DECS_NUM		128
60 #define DESCS_AREAS		1
61 #define TOTAL_DESCS_NUM		(ALLOC_DECS_NUM * DESCS_AREAS)
62 #define QMGR_SCRATCH_SIZE	(TOTAL_DESCS_NUM * 4)
63 
64 #define QMGR_LRAM0_BASE		0x80
65 #define QMGR_LRAM_SIZE		0x84
66 #define QMGR_LRAM1_BASE		0x88
67 #define QMGR_MEMBASE(x)		(0x1000 + (x) * 0x10)
68 #define QMGR_MEMCTRL(x)		(0x1004 + (x) * 0x10)
69 #define QMGR_MEMCTRL_IDX_SH	16
70 #define QMGR_MEMCTRL_DESC_SH	8
71 
72 #define QMGR_PEND(x)	(0x90 + (x) * 4)
73 
74 #define QMGR_PENDING_SLOT_Q(x)	(x / 32)
75 #define QMGR_PENDING_BIT_Q(x)	(x % 32)
76 
77 #define QMGR_QUEUE_A(n)	(0x2000 + (n) * 0x10)
78 #define QMGR_QUEUE_B(n)	(0x2004 + (n) * 0x10)
79 #define QMGR_QUEUE_C(n)	(0x2008 + (n) * 0x10)
80 #define QMGR_QUEUE_D(n)	(0x200c + (n) * 0x10)
81 
82 /* Packet Descriptor */
83 #define PD2_ZERO_LENGTH		(1 << 19)
84 
85 struct cppi41_channel {
86 	struct dma_chan chan;
87 	struct dma_async_tx_descriptor txd;
88 	struct cppi41_dd *cdd;
89 	struct cppi41_desc *desc;
90 	dma_addr_t desc_phys;
91 	void __iomem *gcr_reg;
92 	int is_tx;
93 	u32 residue;
94 
95 	unsigned int q_num;
96 	unsigned int q_comp_num;
97 	unsigned int port_num;
98 
99 	unsigned td_retry;
100 	unsigned td_queued:1;
101 	unsigned td_seen:1;
102 	unsigned td_desc_seen:1;
103 
104 	struct list_head node;		/* Node for pending list */
105 };
106 
107 struct cppi41_desc {
108 	u32 pd0;
109 	u32 pd1;
110 	u32 pd2;
111 	u32 pd3;
112 	u32 pd4;
113 	u32 pd5;
114 	u32 pd6;
115 	u32 pd7;
116 } __aligned(32);
117 
118 struct chan_queues {
119 	u16 submit;
120 	u16 complete;
121 };
122 
123 struct cppi41_dd {
124 	struct dma_device ddev;
125 
126 	void *qmgr_scratch;
127 	dma_addr_t scratch_phys;
128 
129 	struct cppi41_desc *cd;
130 	dma_addr_t descs_phys;
131 	u32 first_td_desc;
132 	struct cppi41_channel *chan_busy[ALLOC_DECS_NUM];
133 
134 	void __iomem *ctrl_mem;
135 	void __iomem *sched_mem;
136 	void __iomem *qmgr_mem;
137 	unsigned int irq;
138 	const struct chan_queues *queues_rx;
139 	const struct chan_queues *queues_tx;
140 	struct chan_queues td_queue;
141 	u16 first_completion_queue;
142 	u16 qmgr_num_pend;
143 	u32 n_chans;
144 	u8 platform;
145 
146 	struct list_head pending;	/* Pending queued transfers */
147 	spinlock_t lock;		/* Lock for pending list */
148 
149 	/* context for suspend/resume */
150 	unsigned int dma_tdfdq;
151 
152 	bool is_suspended;
153 };
154 
155 static struct chan_queues am335x_usb_queues_tx[] = {
156 	/* USB0 ENDP 1 */
157 	[ 0] = { .submit = 32, .complete =  93},
158 	[ 1] = { .submit = 34, .complete =  94},
159 	[ 2] = { .submit = 36, .complete =  95},
160 	[ 3] = { .submit = 38, .complete =  96},
161 	[ 4] = { .submit = 40, .complete =  97},
162 	[ 5] = { .submit = 42, .complete =  98},
163 	[ 6] = { .submit = 44, .complete =  99},
164 	[ 7] = { .submit = 46, .complete = 100},
165 	[ 8] = { .submit = 48, .complete = 101},
166 	[ 9] = { .submit = 50, .complete = 102},
167 	[10] = { .submit = 52, .complete = 103},
168 	[11] = { .submit = 54, .complete = 104},
169 	[12] = { .submit = 56, .complete = 105},
170 	[13] = { .submit = 58, .complete = 106},
171 	[14] = { .submit = 60, .complete = 107},
172 
173 	/* USB1 ENDP1 */
174 	[15] = { .submit = 62, .complete = 125},
175 	[16] = { .submit = 64, .complete = 126},
176 	[17] = { .submit = 66, .complete = 127},
177 	[18] = { .submit = 68, .complete = 128},
178 	[19] = { .submit = 70, .complete = 129},
179 	[20] = { .submit = 72, .complete = 130},
180 	[21] = { .submit = 74, .complete = 131},
181 	[22] = { .submit = 76, .complete = 132},
182 	[23] = { .submit = 78, .complete = 133},
183 	[24] = { .submit = 80, .complete = 134},
184 	[25] = { .submit = 82, .complete = 135},
185 	[26] = { .submit = 84, .complete = 136},
186 	[27] = { .submit = 86, .complete = 137},
187 	[28] = { .submit = 88, .complete = 138},
188 	[29] = { .submit = 90, .complete = 139},
189 };
190 
191 static const struct chan_queues am335x_usb_queues_rx[] = {
192 	/* USB0 ENDP 1 */
193 	[ 0] = { .submit =  1, .complete = 109},
194 	[ 1] = { .submit =  2, .complete = 110},
195 	[ 2] = { .submit =  3, .complete = 111},
196 	[ 3] = { .submit =  4, .complete = 112},
197 	[ 4] = { .submit =  5, .complete = 113},
198 	[ 5] = { .submit =  6, .complete = 114},
199 	[ 6] = { .submit =  7, .complete = 115},
200 	[ 7] = { .submit =  8, .complete = 116},
201 	[ 8] = { .submit =  9, .complete = 117},
202 	[ 9] = { .submit = 10, .complete = 118},
203 	[10] = { .submit = 11, .complete = 119},
204 	[11] = { .submit = 12, .complete = 120},
205 	[12] = { .submit = 13, .complete = 121},
206 	[13] = { .submit = 14, .complete = 122},
207 	[14] = { .submit = 15, .complete = 123},
208 
209 	/* USB1 ENDP 1 */
210 	[15] = { .submit = 16, .complete = 141},
211 	[16] = { .submit = 17, .complete = 142},
212 	[17] = { .submit = 18, .complete = 143},
213 	[18] = { .submit = 19, .complete = 144},
214 	[19] = { .submit = 20, .complete = 145},
215 	[20] = { .submit = 21, .complete = 146},
216 	[21] = { .submit = 22, .complete = 147},
217 	[22] = { .submit = 23, .complete = 148},
218 	[23] = { .submit = 24, .complete = 149},
219 	[24] = { .submit = 25, .complete = 150},
220 	[25] = { .submit = 26, .complete = 151},
221 	[26] = { .submit = 27, .complete = 152},
222 	[27] = { .submit = 28, .complete = 153},
223 	[28] = { .submit = 29, .complete = 154},
224 	[29] = { .submit = 30, .complete = 155},
225 };
226 
227 static const struct chan_queues da8xx_usb_queues_tx[] = {
228 	[0] = { .submit =  16, .complete = 24},
229 	[1] = { .submit =  18, .complete = 24},
230 	[2] = { .submit =  20, .complete = 24},
231 	[3] = { .submit =  22, .complete = 24},
232 };
233 
234 static const struct chan_queues da8xx_usb_queues_rx[] = {
235 	[0] = { .submit =  1, .complete = 26},
236 	[1] = { .submit =  3, .complete = 26},
237 	[2] = { .submit =  5, .complete = 26},
238 	[3] = { .submit =  7, .complete = 26},
239 };
240 
241 struct cppi_glue_infos {
242 	const struct chan_queues *queues_rx;
243 	const struct chan_queues *queues_tx;
244 	struct chan_queues td_queue;
245 	u16 first_completion_queue;
246 	u16 qmgr_num_pend;
247 };
248 
249 static struct cppi41_channel *to_cpp41_chan(struct dma_chan *c)
250 {
251 	return container_of(c, struct cppi41_channel, chan);
252 }
253 
254 static struct cppi41_channel *desc_to_chan(struct cppi41_dd *cdd, u32 desc)
255 {
256 	struct cppi41_channel *c;
257 	u32 descs_size;
258 	u32 desc_num;
259 
260 	descs_size = sizeof(struct cppi41_desc) * ALLOC_DECS_NUM;
261 
262 	if (!((desc >= cdd->descs_phys) &&
263 			(desc < (cdd->descs_phys + descs_size)))) {
264 		return NULL;
265 	}
266 
267 	desc_num = (desc - cdd->descs_phys) / sizeof(struct cppi41_desc);
268 	BUG_ON(desc_num >= ALLOC_DECS_NUM);
269 	c = cdd->chan_busy[desc_num];
270 	cdd->chan_busy[desc_num] = NULL;
271 
272 	/* Usecount for chan_busy[], paired with push_desc_queue() */
273 	pm_runtime_put(cdd->ddev.dev);
274 
275 	return c;
276 }
277 
278 static void cppi_writel(u32 val, void *__iomem *mem)
279 {
280 	__raw_writel(val, mem);
281 }
282 
283 static u32 cppi_readl(void *__iomem *mem)
284 {
285 	return __raw_readl(mem);
286 }
287 
288 static u32 pd_trans_len(u32 val)
289 {
290 	return val & ((1 << (DESC_LENGTH_BITS_NUM + 1)) - 1);
291 }
292 
293 static u32 cppi41_pop_desc(struct cppi41_dd *cdd, unsigned queue_num)
294 {
295 	u32 desc;
296 
297 	desc = cppi_readl(cdd->qmgr_mem + QMGR_QUEUE_D(queue_num));
298 	desc &= ~0x1f;
299 	return desc;
300 }
301 
302 static irqreturn_t cppi41_irq(int irq, void *data)
303 {
304 	struct cppi41_dd *cdd = data;
305 	u16 first_completion_queue = cdd->first_completion_queue;
306 	u16 qmgr_num_pend = cdd->qmgr_num_pend;
307 	struct cppi41_channel *c;
308 	int i;
309 
310 	for (i = QMGR_PENDING_SLOT_Q(first_completion_queue); i < qmgr_num_pend;
311 			i++) {
312 		u32 val;
313 		u32 q_num;
314 
315 		val = cppi_readl(cdd->qmgr_mem + QMGR_PEND(i));
316 		if (i == QMGR_PENDING_SLOT_Q(first_completion_queue) && val) {
317 			u32 mask;
318 			/* set corresponding bit for completetion Q 93 */
319 			mask = 1 << QMGR_PENDING_BIT_Q(first_completion_queue);
320 			/* not set all bits for queues less than Q 93 */
321 			mask--;
322 			/* now invert and keep only Q 93+ set */
323 			val &= ~mask;
324 		}
325 
326 		if (val)
327 			__iormb();
328 
329 		while (val) {
330 			u32 desc, len;
331 
332 			/*
333 			 * This should never trigger, see the comments in
334 			 * push_desc_queue()
335 			 */
336 			WARN_ON(cdd->is_suspended);
337 
338 			q_num = __fls(val);
339 			val &= ~(1 << q_num);
340 			q_num += 32 * i;
341 			desc = cppi41_pop_desc(cdd, q_num);
342 			c = desc_to_chan(cdd, desc);
343 			if (WARN_ON(!c)) {
344 				pr_err("%s() q %d desc %08x\n", __func__,
345 						q_num, desc);
346 				continue;
347 			}
348 
349 			if (c->desc->pd2 & PD2_ZERO_LENGTH)
350 				len = 0;
351 			else
352 				len = pd_trans_len(c->desc->pd0);
353 
354 			c->residue = pd_trans_len(c->desc->pd6) - len;
355 			dma_cookie_complete(&c->txd);
356 			dmaengine_desc_get_callback_invoke(&c->txd, NULL);
357 		}
358 	}
359 	return IRQ_HANDLED;
360 }
361 
362 static dma_cookie_t cppi41_tx_submit(struct dma_async_tx_descriptor *tx)
363 {
364 	dma_cookie_t cookie;
365 
366 	cookie = dma_cookie_assign(tx);
367 
368 	return cookie;
369 }
370 
371 static int cppi41_dma_alloc_chan_resources(struct dma_chan *chan)
372 {
373 	struct cppi41_channel *c = to_cpp41_chan(chan);
374 	struct cppi41_dd *cdd = c->cdd;
375 	int error;
376 
377 	error = pm_runtime_get_sync(cdd->ddev.dev);
378 	if (error < 0) {
379 		dev_err(cdd->ddev.dev, "%s pm runtime get: %i\n",
380 			__func__, error);
381 		pm_runtime_put_noidle(cdd->ddev.dev);
382 
383 		return error;
384 	}
385 
386 	dma_cookie_init(chan);
387 	dma_async_tx_descriptor_init(&c->txd, chan);
388 	c->txd.tx_submit = cppi41_tx_submit;
389 
390 	if (!c->is_tx)
391 		cppi_writel(c->q_num, c->gcr_reg + RXHPCRA0);
392 
393 	pm_runtime_mark_last_busy(cdd->ddev.dev);
394 	pm_runtime_put_autosuspend(cdd->ddev.dev);
395 
396 	return 0;
397 }
398 
399 static void cppi41_dma_free_chan_resources(struct dma_chan *chan)
400 {
401 	struct cppi41_channel *c = to_cpp41_chan(chan);
402 	struct cppi41_dd *cdd = c->cdd;
403 	int error;
404 
405 	error = pm_runtime_get_sync(cdd->ddev.dev);
406 	if (error < 0) {
407 		pm_runtime_put_noidle(cdd->ddev.dev);
408 
409 		return;
410 	}
411 
412 	WARN_ON(!list_empty(&cdd->pending));
413 
414 	pm_runtime_mark_last_busy(cdd->ddev.dev);
415 	pm_runtime_put_autosuspend(cdd->ddev.dev);
416 }
417 
418 static enum dma_status cppi41_dma_tx_status(struct dma_chan *chan,
419 	dma_cookie_t cookie, struct dma_tx_state *txstate)
420 {
421 	struct cppi41_channel *c = to_cpp41_chan(chan);
422 	enum dma_status ret;
423 
424 	ret = dma_cookie_status(chan, cookie, txstate);
425 
426 	dma_set_residue(txstate, c->residue);
427 
428 	return ret;
429 }
430 
431 static void push_desc_queue(struct cppi41_channel *c)
432 {
433 	struct cppi41_dd *cdd = c->cdd;
434 	u32 desc_num;
435 	u32 desc_phys;
436 	u32 reg;
437 
438 	c->residue = 0;
439 
440 	reg = GCR_CHAN_ENABLE;
441 	if (!c->is_tx) {
442 		reg |= GCR_STARV_RETRY;
443 		reg |= GCR_DESC_TYPE_HOST;
444 		reg |= c->q_comp_num;
445 	}
446 
447 	cppi_writel(reg, c->gcr_reg);
448 
449 	/*
450 	 * We don't use writel() but __raw_writel() so we have to make sure
451 	 * that the DMA descriptor in coherent memory made to the main memory
452 	 * before starting the dma engine.
453 	 */
454 	__iowmb();
455 
456 	/*
457 	 * DMA transfers can take at least 200ms to complete with USB mass
458 	 * storage connected. To prevent autosuspend timeouts, we must use
459 	 * pm_runtime_get/put() when chan_busy[] is modified. This will get
460 	 * cleared in desc_to_chan() or cppi41_stop_chan() depending on the
461 	 * outcome of the transfer.
462 	 */
463 	pm_runtime_get(cdd->ddev.dev);
464 
465 	desc_phys = lower_32_bits(c->desc_phys);
466 	desc_num = (desc_phys - cdd->descs_phys) / sizeof(struct cppi41_desc);
467 	WARN_ON(cdd->chan_busy[desc_num]);
468 	cdd->chan_busy[desc_num] = c;
469 
470 	reg = (sizeof(struct cppi41_desc) - 24) / 4;
471 	reg |= desc_phys;
472 	cppi_writel(reg, cdd->qmgr_mem + QMGR_QUEUE_D(c->q_num));
473 }
474 
475 /*
476  * Caller must hold cdd->lock to prevent push_desc_queue()
477  * getting called out of order. We have both cppi41_dma_issue_pending()
478  * and cppi41_runtime_resume() call this function.
479  */
480 static void cppi41_run_queue(struct cppi41_dd *cdd)
481 {
482 	struct cppi41_channel *c, *_c;
483 
484 	list_for_each_entry_safe(c, _c, &cdd->pending, node) {
485 		push_desc_queue(c);
486 		list_del(&c->node);
487 	}
488 }
489 
490 static void cppi41_dma_issue_pending(struct dma_chan *chan)
491 {
492 	struct cppi41_channel *c = to_cpp41_chan(chan);
493 	struct cppi41_dd *cdd = c->cdd;
494 	unsigned long flags;
495 	int error;
496 
497 	error = pm_runtime_get(cdd->ddev.dev);
498 	if ((error != -EINPROGRESS) && error < 0) {
499 		pm_runtime_put_noidle(cdd->ddev.dev);
500 		dev_err(cdd->ddev.dev, "Failed to pm_runtime_get: %i\n",
501 			error);
502 
503 		return;
504 	}
505 
506 	spin_lock_irqsave(&cdd->lock, flags);
507 	list_add_tail(&c->node, &cdd->pending);
508 	if (!cdd->is_suspended)
509 		cppi41_run_queue(cdd);
510 	spin_unlock_irqrestore(&cdd->lock, flags);
511 
512 	pm_runtime_mark_last_busy(cdd->ddev.dev);
513 	pm_runtime_put_autosuspend(cdd->ddev.dev);
514 }
515 
516 static u32 get_host_pd0(u32 length)
517 {
518 	u32 reg;
519 
520 	reg = DESC_TYPE_HOST << DESC_TYPE;
521 	reg |= length;
522 
523 	return reg;
524 }
525 
526 static u32 get_host_pd1(struct cppi41_channel *c)
527 {
528 	u32 reg;
529 
530 	reg = 0;
531 
532 	return reg;
533 }
534 
535 static u32 get_host_pd2(struct cppi41_channel *c)
536 {
537 	u32 reg;
538 
539 	reg = DESC_TYPE_USB;
540 	reg |= c->q_comp_num;
541 
542 	return reg;
543 }
544 
545 static u32 get_host_pd3(u32 length)
546 {
547 	u32 reg;
548 
549 	/* PD3 = packet size */
550 	reg = length;
551 
552 	return reg;
553 }
554 
555 static u32 get_host_pd6(u32 length)
556 {
557 	u32 reg;
558 
559 	/* PD6 buffer size */
560 	reg = DESC_PD_COMPLETE;
561 	reg |= length;
562 
563 	return reg;
564 }
565 
566 static u32 get_host_pd4_or_7(u32 addr)
567 {
568 	u32 reg;
569 
570 	reg = addr;
571 
572 	return reg;
573 }
574 
575 static u32 get_host_pd5(void)
576 {
577 	u32 reg;
578 
579 	reg = 0;
580 
581 	return reg;
582 }
583 
584 static struct dma_async_tx_descriptor *cppi41_dma_prep_slave_sg(
585 	struct dma_chan *chan, struct scatterlist *sgl, unsigned sg_len,
586 	enum dma_transfer_direction dir, unsigned long tx_flags, void *context)
587 {
588 	struct cppi41_channel *c = to_cpp41_chan(chan);
589 	struct cppi41_desc *d;
590 	struct scatterlist *sg;
591 	unsigned int i;
592 
593 	d = c->desc;
594 	for_each_sg(sgl, sg, sg_len, i) {
595 		u32 addr;
596 		u32 len;
597 
598 		/* We need to use more than one desc once musb supports sg */
599 		addr = lower_32_bits(sg_dma_address(sg));
600 		len = sg_dma_len(sg);
601 
602 		d->pd0 = get_host_pd0(len);
603 		d->pd1 = get_host_pd1(c);
604 		d->pd2 = get_host_pd2(c);
605 		d->pd3 = get_host_pd3(len);
606 		d->pd4 = get_host_pd4_or_7(addr);
607 		d->pd5 = get_host_pd5();
608 		d->pd6 = get_host_pd6(len);
609 		d->pd7 = get_host_pd4_or_7(addr);
610 
611 		d++;
612 	}
613 
614 	return &c->txd;
615 }
616 
617 static void cppi41_compute_td_desc(struct cppi41_desc *d)
618 {
619 	d->pd0 = DESC_TYPE_TEARD << DESC_TYPE;
620 }
621 
622 static int cppi41_tear_down_chan(struct cppi41_channel *c)
623 {
624 	struct dmaengine_result abort_result;
625 	struct cppi41_dd *cdd = c->cdd;
626 	struct cppi41_desc *td;
627 	u32 reg;
628 	u32 desc_phys;
629 	u32 td_desc_phys;
630 
631 	td = cdd->cd;
632 	td += cdd->first_td_desc;
633 
634 	td_desc_phys = cdd->descs_phys;
635 	td_desc_phys += cdd->first_td_desc * sizeof(struct cppi41_desc);
636 
637 	if (!c->td_queued) {
638 		cppi41_compute_td_desc(td);
639 		__iowmb();
640 
641 		reg = (sizeof(struct cppi41_desc) - 24) / 4;
642 		reg |= td_desc_phys;
643 		cppi_writel(reg, cdd->qmgr_mem +
644 				QMGR_QUEUE_D(cdd->td_queue.submit));
645 
646 		reg = GCR_CHAN_ENABLE;
647 		if (!c->is_tx) {
648 			reg |= GCR_STARV_RETRY;
649 			reg |= GCR_DESC_TYPE_HOST;
650 			reg |= cdd->td_queue.complete;
651 		}
652 		reg |= GCR_TEARDOWN;
653 		cppi_writel(reg, c->gcr_reg);
654 		c->td_queued = 1;
655 		c->td_retry = 500;
656 	}
657 
658 	if (!c->td_seen || !c->td_desc_seen) {
659 
660 		desc_phys = cppi41_pop_desc(cdd, cdd->td_queue.complete);
661 		if (!desc_phys && c->is_tx)
662 			desc_phys = cppi41_pop_desc(cdd, c->q_comp_num);
663 
664 		if (desc_phys == c->desc_phys) {
665 			c->td_desc_seen = 1;
666 
667 		} else if (desc_phys == td_desc_phys) {
668 			u32 pd0;
669 
670 			__iormb();
671 			pd0 = td->pd0;
672 			WARN_ON((pd0 >> DESC_TYPE) != DESC_TYPE_TEARD);
673 			WARN_ON(!c->is_tx && !(pd0 & TD_DESC_IS_RX));
674 			WARN_ON((pd0 & 0x1f) != c->port_num);
675 			c->td_seen = 1;
676 		} else if (desc_phys) {
677 			WARN_ON_ONCE(1);
678 		}
679 	}
680 	c->td_retry--;
681 	/*
682 	 * If the TX descriptor / channel is in use, the caller needs to poke
683 	 * his TD bit multiple times. After that he hardware releases the
684 	 * transfer descriptor followed by TD descriptor. Waiting seems not to
685 	 * cause any difference.
686 	 * RX seems to be thrown out right away. However once the TearDown
687 	 * descriptor gets through we are done. If we have seens the transfer
688 	 * descriptor before the TD we fetch it from enqueue, it has to be
689 	 * there waiting for us.
690 	 */
691 	if (!c->td_seen && c->td_retry) {
692 		udelay(1);
693 		return -EAGAIN;
694 	}
695 	WARN_ON(!c->td_retry);
696 
697 	if (!c->td_desc_seen) {
698 		desc_phys = cppi41_pop_desc(cdd, c->q_num);
699 		if (!desc_phys)
700 			desc_phys = cppi41_pop_desc(cdd, c->q_comp_num);
701 		WARN_ON(!desc_phys);
702 	}
703 
704 	c->td_queued = 0;
705 	c->td_seen = 0;
706 	c->td_desc_seen = 0;
707 	cppi_writel(0, c->gcr_reg);
708 
709 	/* Invoke the callback to do the necessary clean-up */
710 	abort_result.result = DMA_TRANS_ABORTED;
711 	dma_cookie_complete(&c->txd);
712 	dmaengine_desc_get_callback_invoke(&c->txd, &abort_result);
713 
714 	return 0;
715 }
716 
717 static int cppi41_stop_chan(struct dma_chan *chan)
718 {
719 	struct cppi41_channel *c = to_cpp41_chan(chan);
720 	struct cppi41_dd *cdd = c->cdd;
721 	u32 desc_num;
722 	u32 desc_phys;
723 	int ret;
724 
725 	desc_phys = lower_32_bits(c->desc_phys);
726 	desc_num = (desc_phys - cdd->descs_phys) / sizeof(struct cppi41_desc);
727 	if (!cdd->chan_busy[desc_num]) {
728 		struct cppi41_channel *cc, *_ct;
729 
730 		/*
731 		 * channels might still be in the pendling list if
732 		 * cppi41_dma_issue_pending() is called after
733 		 * cppi41_runtime_suspend() is called
734 		 */
735 		list_for_each_entry_safe(cc, _ct, &cdd->pending, node) {
736 			if (cc != c)
737 				continue;
738 			list_del(&cc->node);
739 			break;
740 		}
741 		return 0;
742 	}
743 
744 	ret = cppi41_tear_down_chan(c);
745 	if (ret)
746 		return ret;
747 
748 	WARN_ON(!cdd->chan_busy[desc_num]);
749 	cdd->chan_busy[desc_num] = NULL;
750 
751 	/* Usecount for chan_busy[], paired with push_desc_queue() */
752 	pm_runtime_put(cdd->ddev.dev);
753 
754 	return 0;
755 }
756 
757 static int cppi41_add_chans(struct device *dev, struct cppi41_dd *cdd)
758 {
759 	struct cppi41_channel *cchan, *chans;
760 	int i;
761 	u32 n_chans = cdd->n_chans;
762 
763 	/*
764 	 * The channels can only be used as TX or as RX. So we add twice
765 	 * that much dma channels because USB can only do RX or TX.
766 	 */
767 	n_chans *= 2;
768 
769 	chans = devm_kcalloc(dev, n_chans, sizeof(*chans), GFP_KERNEL);
770 	if (!chans)
771 		return -ENOMEM;
772 
773 	for (i = 0; i < n_chans; i++) {
774 		cchan = &chans[i];
775 
776 		cchan->cdd = cdd;
777 		if (i & 1) {
778 			cchan->gcr_reg = cdd->ctrl_mem + DMA_TXGCR(i >> 1);
779 			cchan->is_tx = 1;
780 		} else {
781 			cchan->gcr_reg = cdd->ctrl_mem + DMA_RXGCR(i >> 1);
782 			cchan->is_tx = 0;
783 		}
784 		cchan->port_num = i >> 1;
785 		cchan->desc = &cdd->cd[i];
786 		cchan->desc_phys = cdd->descs_phys;
787 		cchan->desc_phys += i * sizeof(struct cppi41_desc);
788 		cchan->chan.device = &cdd->ddev;
789 		list_add_tail(&cchan->chan.device_node, &cdd->ddev.channels);
790 	}
791 	cdd->first_td_desc = n_chans;
792 
793 	return 0;
794 }
795 
796 static void purge_descs(struct device *dev, struct cppi41_dd *cdd)
797 {
798 	unsigned int mem_decs;
799 	int i;
800 
801 	mem_decs = ALLOC_DECS_NUM * sizeof(struct cppi41_desc);
802 
803 	for (i = 0; i < DESCS_AREAS; i++) {
804 
805 		cppi_writel(0, cdd->qmgr_mem + QMGR_MEMBASE(i));
806 		cppi_writel(0, cdd->qmgr_mem + QMGR_MEMCTRL(i));
807 
808 		dma_free_coherent(dev, mem_decs, cdd->cd,
809 				cdd->descs_phys);
810 	}
811 }
812 
813 static void disable_sched(struct cppi41_dd *cdd)
814 {
815 	cppi_writel(0, cdd->sched_mem + DMA_SCHED_CTRL);
816 }
817 
818 static void deinit_cppi41(struct device *dev, struct cppi41_dd *cdd)
819 {
820 	disable_sched(cdd);
821 
822 	purge_descs(dev, cdd);
823 
824 	cppi_writel(0, cdd->qmgr_mem + QMGR_LRAM0_BASE);
825 	cppi_writel(0, cdd->qmgr_mem + QMGR_LRAM0_BASE);
826 	dma_free_coherent(dev, QMGR_SCRATCH_SIZE, cdd->qmgr_scratch,
827 			cdd->scratch_phys);
828 }
829 
830 static int init_descs(struct device *dev, struct cppi41_dd *cdd)
831 {
832 	unsigned int desc_size;
833 	unsigned int mem_decs;
834 	int i;
835 	u32 reg;
836 	u32 idx;
837 
838 	BUILD_BUG_ON(sizeof(struct cppi41_desc) &
839 			(sizeof(struct cppi41_desc) - 1));
840 	BUILD_BUG_ON(sizeof(struct cppi41_desc) < 32);
841 	BUILD_BUG_ON(ALLOC_DECS_NUM < 32);
842 
843 	desc_size = sizeof(struct cppi41_desc);
844 	mem_decs = ALLOC_DECS_NUM * desc_size;
845 
846 	idx = 0;
847 	for (i = 0; i < DESCS_AREAS; i++) {
848 
849 		reg = idx << QMGR_MEMCTRL_IDX_SH;
850 		reg |= (ilog2(desc_size) - 5) << QMGR_MEMCTRL_DESC_SH;
851 		reg |= ilog2(ALLOC_DECS_NUM) - 5;
852 
853 		BUILD_BUG_ON(DESCS_AREAS != 1);
854 		cdd->cd = dma_alloc_coherent(dev, mem_decs,
855 				&cdd->descs_phys, GFP_KERNEL);
856 		if (!cdd->cd)
857 			return -ENOMEM;
858 
859 		cppi_writel(cdd->descs_phys, cdd->qmgr_mem + QMGR_MEMBASE(i));
860 		cppi_writel(reg, cdd->qmgr_mem + QMGR_MEMCTRL(i));
861 
862 		idx += ALLOC_DECS_NUM;
863 	}
864 	return 0;
865 }
866 
867 static void init_sched(struct cppi41_dd *cdd)
868 {
869 	unsigned ch;
870 	unsigned word;
871 	u32 reg;
872 
873 	word = 0;
874 	cppi_writel(0, cdd->sched_mem + DMA_SCHED_CTRL);
875 	for (ch = 0; ch < cdd->n_chans; ch += 2) {
876 
877 		reg = SCHED_ENTRY0_CHAN(ch);
878 		reg |= SCHED_ENTRY1_CHAN(ch) | SCHED_ENTRY1_IS_RX;
879 
880 		reg |= SCHED_ENTRY2_CHAN(ch + 1);
881 		reg |= SCHED_ENTRY3_CHAN(ch + 1) | SCHED_ENTRY3_IS_RX;
882 		cppi_writel(reg, cdd->sched_mem + DMA_SCHED_WORD(word));
883 		word++;
884 	}
885 	reg = cdd->n_chans * 2 - 1;
886 	reg |= DMA_SCHED_CTRL_EN;
887 	cppi_writel(reg, cdd->sched_mem + DMA_SCHED_CTRL);
888 }
889 
890 static int init_cppi41(struct device *dev, struct cppi41_dd *cdd)
891 {
892 	int ret;
893 
894 	BUILD_BUG_ON(QMGR_SCRATCH_SIZE > ((1 << 14) - 1));
895 	cdd->qmgr_scratch = dma_alloc_coherent(dev, QMGR_SCRATCH_SIZE,
896 			&cdd->scratch_phys, GFP_KERNEL);
897 	if (!cdd->qmgr_scratch)
898 		return -ENOMEM;
899 
900 	cppi_writel(cdd->scratch_phys, cdd->qmgr_mem + QMGR_LRAM0_BASE);
901 	cppi_writel(TOTAL_DESCS_NUM, cdd->qmgr_mem + QMGR_LRAM_SIZE);
902 	cppi_writel(0, cdd->qmgr_mem + QMGR_LRAM1_BASE);
903 
904 	ret = init_descs(dev, cdd);
905 	if (ret)
906 		goto err_td;
907 
908 	cppi_writel(cdd->td_queue.submit, cdd->ctrl_mem + DMA_TDFDQ);
909 	init_sched(cdd);
910 
911 	return 0;
912 err_td:
913 	deinit_cppi41(dev, cdd);
914 	return ret;
915 }
916 
917 static struct platform_driver cpp41_dma_driver;
918 /*
919  * The param format is:
920  * X Y
921  * X: Port
922  * Y: 0 = RX else TX
923  */
924 #define INFO_PORT	0
925 #define INFO_IS_TX	1
926 
927 static bool cpp41_dma_filter_fn(struct dma_chan *chan, void *param)
928 {
929 	struct cppi41_channel *cchan;
930 	struct cppi41_dd *cdd;
931 	const struct chan_queues *queues;
932 	u32 *num = param;
933 
934 	if (chan->device->dev->driver != &cpp41_dma_driver.driver)
935 		return false;
936 
937 	cchan = to_cpp41_chan(chan);
938 
939 	if (cchan->port_num != num[INFO_PORT])
940 		return false;
941 
942 	if (cchan->is_tx && !num[INFO_IS_TX])
943 		return false;
944 	cdd = cchan->cdd;
945 	if (cchan->is_tx)
946 		queues = cdd->queues_tx;
947 	else
948 		queues = cdd->queues_rx;
949 
950 	BUILD_BUG_ON(ARRAY_SIZE(am335x_usb_queues_rx) !=
951 		     ARRAY_SIZE(am335x_usb_queues_tx));
952 	if (WARN_ON(cchan->port_num >= ARRAY_SIZE(am335x_usb_queues_rx)))
953 		return false;
954 
955 	cchan->q_num = queues[cchan->port_num].submit;
956 	cchan->q_comp_num = queues[cchan->port_num].complete;
957 	return true;
958 }
959 
960 static struct of_dma_filter_info cpp41_dma_info = {
961 	.filter_fn = cpp41_dma_filter_fn,
962 };
963 
964 static struct dma_chan *cppi41_dma_xlate(struct of_phandle_args *dma_spec,
965 		struct of_dma *ofdma)
966 {
967 	int count = dma_spec->args_count;
968 	struct of_dma_filter_info *info = ofdma->of_dma_data;
969 
970 	if (!info || !info->filter_fn)
971 		return NULL;
972 
973 	if (count != 2)
974 		return NULL;
975 
976 	return dma_request_channel(info->dma_cap, info->filter_fn,
977 			&dma_spec->args[0]);
978 }
979 
980 static const struct cppi_glue_infos am335x_usb_infos = {
981 	.queues_rx = am335x_usb_queues_rx,
982 	.queues_tx = am335x_usb_queues_tx,
983 	.td_queue = { .submit = 31, .complete = 0 },
984 	.first_completion_queue = 93,
985 	.qmgr_num_pend = 5,
986 };
987 
988 static const struct cppi_glue_infos da8xx_usb_infos = {
989 	.queues_rx = da8xx_usb_queues_rx,
990 	.queues_tx = da8xx_usb_queues_tx,
991 	.td_queue = { .submit = 31, .complete = 0 },
992 	.first_completion_queue = 24,
993 	.qmgr_num_pend = 2,
994 };
995 
996 static const struct of_device_id cppi41_dma_ids[] = {
997 	{ .compatible = "ti,am3359-cppi41", .data = &am335x_usb_infos},
998 	{ .compatible = "ti,da830-cppi41", .data = &da8xx_usb_infos},
999 	{},
1000 };
1001 MODULE_DEVICE_TABLE(of, cppi41_dma_ids);
1002 
1003 static const struct cppi_glue_infos *get_glue_info(struct device *dev)
1004 {
1005 	const struct of_device_id *of_id;
1006 
1007 	of_id = of_match_node(cppi41_dma_ids, dev->of_node);
1008 	if (!of_id)
1009 		return NULL;
1010 	return of_id->data;
1011 }
1012 
1013 #define CPPI41_DMA_BUSWIDTHS	(BIT(DMA_SLAVE_BUSWIDTH_1_BYTE) | \
1014 				BIT(DMA_SLAVE_BUSWIDTH_2_BYTES) | \
1015 				BIT(DMA_SLAVE_BUSWIDTH_3_BYTES) | \
1016 				BIT(DMA_SLAVE_BUSWIDTH_4_BYTES))
1017 
1018 static int cppi41_dma_probe(struct platform_device *pdev)
1019 {
1020 	struct cppi41_dd *cdd;
1021 	struct device *dev = &pdev->dev;
1022 	const struct cppi_glue_infos *glue_info;
1023 	struct resource *mem;
1024 	int index;
1025 	int irq;
1026 	int ret;
1027 
1028 	glue_info = get_glue_info(dev);
1029 	if (!glue_info)
1030 		return -EINVAL;
1031 
1032 	cdd = devm_kzalloc(&pdev->dev, sizeof(*cdd), GFP_KERNEL);
1033 	if (!cdd)
1034 		return -ENOMEM;
1035 
1036 	dma_cap_set(DMA_SLAVE, cdd->ddev.cap_mask);
1037 	cdd->ddev.device_alloc_chan_resources = cppi41_dma_alloc_chan_resources;
1038 	cdd->ddev.device_free_chan_resources = cppi41_dma_free_chan_resources;
1039 	cdd->ddev.device_tx_status = cppi41_dma_tx_status;
1040 	cdd->ddev.device_issue_pending = cppi41_dma_issue_pending;
1041 	cdd->ddev.device_prep_slave_sg = cppi41_dma_prep_slave_sg;
1042 	cdd->ddev.device_terminate_all = cppi41_stop_chan;
1043 	cdd->ddev.directions = BIT(DMA_DEV_TO_MEM) | BIT(DMA_MEM_TO_DEV);
1044 	cdd->ddev.src_addr_widths = CPPI41_DMA_BUSWIDTHS;
1045 	cdd->ddev.dst_addr_widths = CPPI41_DMA_BUSWIDTHS;
1046 	cdd->ddev.residue_granularity = DMA_RESIDUE_GRANULARITY_BURST;
1047 	cdd->ddev.dev = dev;
1048 	INIT_LIST_HEAD(&cdd->ddev.channels);
1049 	cpp41_dma_info.dma_cap = cdd->ddev.cap_mask;
1050 
1051 	index = of_property_match_string(dev->of_node,
1052 					 "reg-names", "controller");
1053 	if (index < 0)
1054 		return index;
1055 
1056 	mem = platform_get_resource(pdev, IORESOURCE_MEM, index);
1057 	cdd->ctrl_mem = devm_ioremap_resource(dev, mem);
1058 	if (IS_ERR(cdd->ctrl_mem))
1059 		return PTR_ERR(cdd->ctrl_mem);
1060 
1061 	mem = platform_get_resource(pdev, IORESOURCE_MEM, index + 1);
1062 	cdd->sched_mem = devm_ioremap_resource(dev, mem);
1063 	if (IS_ERR(cdd->sched_mem))
1064 		return PTR_ERR(cdd->sched_mem);
1065 
1066 	mem = platform_get_resource(pdev, IORESOURCE_MEM, index + 2);
1067 	cdd->qmgr_mem = devm_ioremap_resource(dev, mem);
1068 	if (IS_ERR(cdd->qmgr_mem))
1069 		return PTR_ERR(cdd->qmgr_mem);
1070 
1071 	spin_lock_init(&cdd->lock);
1072 	INIT_LIST_HEAD(&cdd->pending);
1073 
1074 	platform_set_drvdata(pdev, cdd);
1075 
1076 	pm_runtime_enable(dev);
1077 	pm_runtime_set_autosuspend_delay(dev, 100);
1078 	pm_runtime_use_autosuspend(dev);
1079 	ret = pm_runtime_get_sync(dev);
1080 	if (ret < 0)
1081 		goto err_get_sync;
1082 
1083 	cdd->queues_rx = glue_info->queues_rx;
1084 	cdd->queues_tx = glue_info->queues_tx;
1085 	cdd->td_queue = glue_info->td_queue;
1086 	cdd->qmgr_num_pend = glue_info->qmgr_num_pend;
1087 	cdd->first_completion_queue = glue_info->first_completion_queue;
1088 
1089 	ret = of_property_read_u32(dev->of_node,
1090 				   "#dma-channels", &cdd->n_chans);
1091 	if (ret)
1092 		goto err_get_n_chans;
1093 
1094 	ret = init_cppi41(dev, cdd);
1095 	if (ret)
1096 		goto err_init_cppi;
1097 
1098 	ret = cppi41_add_chans(dev, cdd);
1099 	if (ret)
1100 		goto err_chans;
1101 
1102 	irq = irq_of_parse_and_map(dev->of_node, 0);
1103 	if (!irq) {
1104 		ret = -EINVAL;
1105 		goto err_chans;
1106 	}
1107 
1108 	ret = devm_request_irq(&pdev->dev, irq, cppi41_irq, IRQF_SHARED,
1109 			dev_name(dev), cdd);
1110 	if (ret)
1111 		goto err_chans;
1112 	cdd->irq = irq;
1113 
1114 	ret = dma_async_device_register(&cdd->ddev);
1115 	if (ret)
1116 		goto err_chans;
1117 
1118 	ret = of_dma_controller_register(dev->of_node,
1119 			cppi41_dma_xlate, &cpp41_dma_info);
1120 	if (ret)
1121 		goto err_of;
1122 
1123 	pm_runtime_mark_last_busy(dev);
1124 	pm_runtime_put_autosuspend(dev);
1125 
1126 	return 0;
1127 err_of:
1128 	dma_async_device_unregister(&cdd->ddev);
1129 err_chans:
1130 	deinit_cppi41(dev, cdd);
1131 err_init_cppi:
1132 	pm_runtime_dont_use_autosuspend(dev);
1133 err_get_n_chans:
1134 err_get_sync:
1135 	pm_runtime_put_sync(dev);
1136 	pm_runtime_disable(dev);
1137 	return ret;
1138 }
1139 
1140 static int cppi41_dma_remove(struct platform_device *pdev)
1141 {
1142 	struct cppi41_dd *cdd = platform_get_drvdata(pdev);
1143 	int error;
1144 
1145 	error = pm_runtime_get_sync(&pdev->dev);
1146 	if (error < 0)
1147 		dev_err(&pdev->dev, "%s could not pm_runtime_get: %i\n",
1148 			__func__, error);
1149 	of_dma_controller_free(pdev->dev.of_node);
1150 	dma_async_device_unregister(&cdd->ddev);
1151 
1152 	devm_free_irq(&pdev->dev, cdd->irq, cdd);
1153 	deinit_cppi41(&pdev->dev, cdd);
1154 	pm_runtime_dont_use_autosuspend(&pdev->dev);
1155 	pm_runtime_put_sync(&pdev->dev);
1156 	pm_runtime_disable(&pdev->dev);
1157 	return 0;
1158 }
1159 
1160 static int __maybe_unused cppi41_suspend(struct device *dev)
1161 {
1162 	struct cppi41_dd *cdd = dev_get_drvdata(dev);
1163 
1164 	cdd->dma_tdfdq = cppi_readl(cdd->ctrl_mem + DMA_TDFDQ);
1165 	disable_sched(cdd);
1166 
1167 	return 0;
1168 }
1169 
1170 static int __maybe_unused cppi41_resume(struct device *dev)
1171 {
1172 	struct cppi41_dd *cdd = dev_get_drvdata(dev);
1173 	struct cppi41_channel *c;
1174 	int i;
1175 
1176 	for (i = 0; i < DESCS_AREAS; i++)
1177 		cppi_writel(cdd->descs_phys, cdd->qmgr_mem + QMGR_MEMBASE(i));
1178 
1179 	list_for_each_entry(c, &cdd->ddev.channels, chan.device_node)
1180 		if (!c->is_tx)
1181 			cppi_writel(c->q_num, c->gcr_reg + RXHPCRA0);
1182 
1183 	init_sched(cdd);
1184 
1185 	cppi_writel(cdd->dma_tdfdq, cdd->ctrl_mem + DMA_TDFDQ);
1186 	cppi_writel(cdd->scratch_phys, cdd->qmgr_mem + QMGR_LRAM0_BASE);
1187 	cppi_writel(QMGR_SCRATCH_SIZE, cdd->qmgr_mem + QMGR_LRAM_SIZE);
1188 	cppi_writel(0, cdd->qmgr_mem + QMGR_LRAM1_BASE);
1189 
1190 	return 0;
1191 }
1192 
1193 static int __maybe_unused cppi41_runtime_suspend(struct device *dev)
1194 {
1195 	struct cppi41_dd *cdd = dev_get_drvdata(dev);
1196 	unsigned long flags;
1197 
1198 	spin_lock_irqsave(&cdd->lock, flags);
1199 	cdd->is_suspended = true;
1200 	WARN_ON(!list_empty(&cdd->pending));
1201 	spin_unlock_irqrestore(&cdd->lock, flags);
1202 
1203 	return 0;
1204 }
1205 
1206 static int __maybe_unused cppi41_runtime_resume(struct device *dev)
1207 {
1208 	struct cppi41_dd *cdd = dev_get_drvdata(dev);
1209 	unsigned long flags;
1210 
1211 	spin_lock_irqsave(&cdd->lock, flags);
1212 	cdd->is_suspended = false;
1213 	cppi41_run_queue(cdd);
1214 	spin_unlock_irqrestore(&cdd->lock, flags);
1215 
1216 	return 0;
1217 }
1218 
1219 static const struct dev_pm_ops cppi41_pm_ops = {
1220 	SET_LATE_SYSTEM_SLEEP_PM_OPS(cppi41_suspend, cppi41_resume)
1221 	SET_RUNTIME_PM_OPS(cppi41_runtime_suspend,
1222 			   cppi41_runtime_resume,
1223 			   NULL)
1224 };
1225 
1226 static struct platform_driver cpp41_dma_driver = {
1227 	.probe  = cppi41_dma_probe,
1228 	.remove = cppi41_dma_remove,
1229 	.driver = {
1230 		.name = "cppi41-dma-engine",
1231 		.pm = &cppi41_pm_ops,
1232 		.of_match_table = of_match_ptr(cppi41_dma_ids),
1233 	},
1234 };
1235 
1236 module_platform_driver(cpp41_dma_driver);
1237 MODULE_LICENSE("GPL");
1238 MODULE_AUTHOR("Sebastian Andrzej Siewior <bigeasy@linutronix.de>");
1239