xref: /openbmc/linux/drivers/dma/k3dma.c (revision afba8b0a)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (c) 2013 - 2015 Linaro Ltd.
4  * Copyright (c) 2013 Hisilicon Limited.
5  */
6 #include <linux/sched.h>
7 #include <linux/device.h>
8 #include <linux/dma-mapping.h>
9 #include <linux/dmapool.h>
10 #include <linux/dmaengine.h>
11 #include <linux/init.h>
12 #include <linux/interrupt.h>
13 #include <linux/kernel.h>
14 #include <linux/module.h>
15 #include <linux/platform_device.h>
16 #include <linux/slab.h>
17 #include <linux/spinlock.h>
18 #include <linux/of_device.h>
19 #include <linux/of.h>
20 #include <linux/clk.h>
21 #include <linux/of_dma.h>
22 
23 #include "virt-dma.h"
24 
25 #define DRIVER_NAME		"k3-dma"
26 #define DMA_MAX_SIZE		0x1ffc
27 #define DMA_CYCLIC_MAX_PERIOD	0x1000
28 #define LLI_BLOCK_SIZE		(4 * PAGE_SIZE)
29 
30 #define INT_STAT		0x00
31 #define INT_TC1			0x04
32 #define INT_TC2			0x08
33 #define INT_ERR1		0x0c
34 #define INT_ERR2		0x10
35 #define INT_TC1_MASK		0x18
36 #define INT_TC2_MASK		0x1c
37 #define INT_ERR1_MASK		0x20
38 #define INT_ERR2_MASK		0x24
39 #define INT_TC1_RAW		0x600
40 #define INT_TC2_RAW		0x608
41 #define INT_ERR1_RAW		0x610
42 #define INT_ERR2_RAW		0x618
43 #define CH_PRI			0x688
44 #define CH_STAT			0x690
45 #define CX_CUR_CNT		0x704
46 #define CX_LLI			0x800
47 #define CX_CNT1			0x80c
48 #define CX_CNT0			0x810
49 #define CX_SRC			0x814
50 #define CX_DST			0x818
51 #define CX_CFG			0x81c
52 
53 #define CX_LLI_CHAIN_EN		0x2
54 #define CX_CFG_EN		0x1
55 #define CX_CFG_NODEIRQ		BIT(1)
56 #define CX_CFG_MEM2PER		(0x1 << 2)
57 #define CX_CFG_PER2MEM		(0x2 << 2)
58 #define CX_CFG_SRCINCR		(0x1 << 31)
59 #define CX_CFG_DSTINCR		(0x1 << 30)
60 
61 struct k3_desc_hw {
62 	u32 lli;
63 	u32 reserved[3];
64 	u32 count;
65 	u32 saddr;
66 	u32 daddr;
67 	u32 config;
68 } __aligned(32);
69 
70 struct k3_dma_desc_sw {
71 	struct virt_dma_desc	vd;
72 	dma_addr_t		desc_hw_lli;
73 	size_t			desc_num;
74 	size_t			size;
75 	struct k3_desc_hw	*desc_hw;
76 };
77 
78 struct k3_dma_phy;
79 
80 struct k3_dma_chan {
81 	u32			ccfg;
82 	struct virt_dma_chan	vc;
83 	struct k3_dma_phy	*phy;
84 	struct list_head	node;
85 	dma_addr_t		dev_addr;
86 	enum dma_status		status;
87 	bool			cyclic;
88 	struct dma_slave_config	slave_config;
89 };
90 
91 struct k3_dma_phy {
92 	u32			idx;
93 	void __iomem		*base;
94 	struct k3_dma_chan	*vchan;
95 	struct k3_dma_desc_sw	*ds_run;
96 	struct k3_dma_desc_sw	*ds_done;
97 };
98 
99 struct k3_dma_dev {
100 	struct dma_device	slave;
101 	void __iomem		*base;
102 	struct tasklet_struct	task;
103 	spinlock_t		lock;
104 	struct list_head	chan_pending;
105 	struct k3_dma_phy	*phy;
106 	struct k3_dma_chan	*chans;
107 	struct clk		*clk;
108 	struct dma_pool		*pool;
109 	u32			dma_channels;
110 	u32			dma_requests;
111 	u32			dma_channel_mask;
112 	unsigned int		irq;
113 };
114 
115 
116 #define K3_FLAG_NOCLK	BIT(1)
117 
118 struct k3dma_soc_data {
119 	unsigned long flags;
120 };
121 
122 
123 #define to_k3_dma(dmadev) container_of(dmadev, struct k3_dma_dev, slave)
124 
125 static int k3_dma_config_write(struct dma_chan *chan,
126 			       enum dma_transfer_direction dir,
127 			       struct dma_slave_config *cfg);
128 
129 static struct k3_dma_chan *to_k3_chan(struct dma_chan *chan)
130 {
131 	return container_of(chan, struct k3_dma_chan, vc.chan);
132 }
133 
134 static void k3_dma_pause_dma(struct k3_dma_phy *phy, bool on)
135 {
136 	u32 val = 0;
137 
138 	if (on) {
139 		val = readl_relaxed(phy->base + CX_CFG);
140 		val |= CX_CFG_EN;
141 		writel_relaxed(val, phy->base + CX_CFG);
142 	} else {
143 		val = readl_relaxed(phy->base + CX_CFG);
144 		val &= ~CX_CFG_EN;
145 		writel_relaxed(val, phy->base + CX_CFG);
146 	}
147 }
148 
149 static void k3_dma_terminate_chan(struct k3_dma_phy *phy, struct k3_dma_dev *d)
150 {
151 	u32 val = 0;
152 
153 	k3_dma_pause_dma(phy, false);
154 
155 	val = 0x1 << phy->idx;
156 	writel_relaxed(val, d->base + INT_TC1_RAW);
157 	writel_relaxed(val, d->base + INT_TC2_RAW);
158 	writel_relaxed(val, d->base + INT_ERR1_RAW);
159 	writel_relaxed(val, d->base + INT_ERR2_RAW);
160 }
161 
162 static void k3_dma_set_desc(struct k3_dma_phy *phy, struct k3_desc_hw *hw)
163 {
164 	writel_relaxed(hw->lli, phy->base + CX_LLI);
165 	writel_relaxed(hw->count, phy->base + CX_CNT0);
166 	writel_relaxed(hw->saddr, phy->base + CX_SRC);
167 	writel_relaxed(hw->daddr, phy->base + CX_DST);
168 	writel_relaxed(hw->config, phy->base + CX_CFG);
169 }
170 
171 static u32 k3_dma_get_curr_cnt(struct k3_dma_dev *d, struct k3_dma_phy *phy)
172 {
173 	u32 cnt = 0;
174 
175 	cnt = readl_relaxed(d->base + CX_CUR_CNT + phy->idx * 0x10);
176 	cnt &= 0xffff;
177 	return cnt;
178 }
179 
180 static u32 k3_dma_get_curr_lli(struct k3_dma_phy *phy)
181 {
182 	return readl_relaxed(phy->base + CX_LLI);
183 }
184 
185 static u32 k3_dma_get_chan_stat(struct k3_dma_dev *d)
186 {
187 	return readl_relaxed(d->base + CH_STAT);
188 }
189 
190 static void k3_dma_enable_dma(struct k3_dma_dev *d, bool on)
191 {
192 	if (on) {
193 		/* set same priority */
194 		writel_relaxed(0x0, d->base + CH_PRI);
195 
196 		/* unmask irq */
197 		writel_relaxed(0xffff, d->base + INT_TC1_MASK);
198 		writel_relaxed(0xffff, d->base + INT_TC2_MASK);
199 		writel_relaxed(0xffff, d->base + INT_ERR1_MASK);
200 		writel_relaxed(0xffff, d->base + INT_ERR2_MASK);
201 	} else {
202 		/* mask irq */
203 		writel_relaxed(0x0, d->base + INT_TC1_MASK);
204 		writel_relaxed(0x0, d->base + INT_TC2_MASK);
205 		writel_relaxed(0x0, d->base + INT_ERR1_MASK);
206 		writel_relaxed(0x0, d->base + INT_ERR2_MASK);
207 	}
208 }
209 
210 static irqreturn_t k3_dma_int_handler(int irq, void *dev_id)
211 {
212 	struct k3_dma_dev *d = (struct k3_dma_dev *)dev_id;
213 	struct k3_dma_phy *p;
214 	struct k3_dma_chan *c;
215 	u32 stat = readl_relaxed(d->base + INT_STAT);
216 	u32 tc1  = readl_relaxed(d->base + INT_TC1);
217 	u32 tc2  = readl_relaxed(d->base + INT_TC2);
218 	u32 err1 = readl_relaxed(d->base + INT_ERR1);
219 	u32 err2 = readl_relaxed(d->base + INT_ERR2);
220 	u32 i, irq_chan = 0;
221 
222 	while (stat) {
223 		i = __ffs(stat);
224 		stat &= ~BIT(i);
225 		if (likely(tc1 & BIT(i)) || (tc2 & BIT(i))) {
226 			unsigned long flags;
227 
228 			p = &d->phy[i];
229 			c = p->vchan;
230 			if (c && (tc1 & BIT(i))) {
231 				spin_lock_irqsave(&c->vc.lock, flags);
232 				if (p->ds_run != NULL) {
233 					vchan_cookie_complete(&p->ds_run->vd);
234 					p->ds_done = p->ds_run;
235 					p->ds_run = NULL;
236 				}
237 				spin_unlock_irqrestore(&c->vc.lock, flags);
238 			}
239 			if (c && (tc2 & BIT(i))) {
240 				spin_lock_irqsave(&c->vc.lock, flags);
241 				if (p->ds_run != NULL)
242 					vchan_cyclic_callback(&p->ds_run->vd);
243 				spin_unlock_irqrestore(&c->vc.lock, flags);
244 			}
245 			irq_chan |= BIT(i);
246 		}
247 		if (unlikely((err1 & BIT(i)) || (err2 & BIT(i))))
248 			dev_warn(d->slave.dev, "DMA ERR\n");
249 	}
250 
251 	writel_relaxed(irq_chan, d->base + INT_TC1_RAW);
252 	writel_relaxed(irq_chan, d->base + INT_TC2_RAW);
253 	writel_relaxed(err1, d->base + INT_ERR1_RAW);
254 	writel_relaxed(err2, d->base + INT_ERR2_RAW);
255 
256 	if (irq_chan)
257 		tasklet_schedule(&d->task);
258 
259 	if (irq_chan || err1 || err2)
260 		return IRQ_HANDLED;
261 
262 	return IRQ_NONE;
263 }
264 
265 static int k3_dma_start_txd(struct k3_dma_chan *c)
266 {
267 	struct k3_dma_dev *d = to_k3_dma(c->vc.chan.device);
268 	struct virt_dma_desc *vd = vchan_next_desc(&c->vc);
269 
270 	if (!c->phy)
271 		return -EAGAIN;
272 
273 	if (BIT(c->phy->idx) & k3_dma_get_chan_stat(d))
274 		return -EAGAIN;
275 
276 	/* Avoid losing track of  ds_run if a transaction is in flight */
277 	if (c->phy->ds_run)
278 		return -EAGAIN;
279 
280 	if (vd) {
281 		struct k3_dma_desc_sw *ds =
282 			container_of(vd, struct k3_dma_desc_sw, vd);
283 		/*
284 		 * fetch and remove request from vc->desc_issued
285 		 * so vc->desc_issued only contains desc pending
286 		 */
287 		list_del(&ds->vd.node);
288 
289 		c->phy->ds_run = ds;
290 		c->phy->ds_done = NULL;
291 		/* start dma */
292 		k3_dma_set_desc(c->phy, &ds->desc_hw[0]);
293 		return 0;
294 	}
295 	c->phy->ds_run = NULL;
296 	c->phy->ds_done = NULL;
297 	return -EAGAIN;
298 }
299 
300 static void k3_dma_tasklet(struct tasklet_struct *t)
301 {
302 	struct k3_dma_dev *d = from_tasklet(d, t, task);
303 	struct k3_dma_phy *p;
304 	struct k3_dma_chan *c, *cn;
305 	unsigned pch, pch_alloc = 0;
306 
307 	/* check new dma request of running channel in vc->desc_issued */
308 	list_for_each_entry_safe(c, cn, &d->slave.channels, vc.chan.device_node) {
309 		spin_lock_irq(&c->vc.lock);
310 		p = c->phy;
311 		if (p && p->ds_done) {
312 			if (k3_dma_start_txd(c)) {
313 				/* No current txd associated with this channel */
314 				dev_dbg(d->slave.dev, "pchan %u: free\n", p->idx);
315 				/* Mark this channel free */
316 				c->phy = NULL;
317 				p->vchan = NULL;
318 			}
319 		}
320 		spin_unlock_irq(&c->vc.lock);
321 	}
322 
323 	/* check new channel request in d->chan_pending */
324 	spin_lock_irq(&d->lock);
325 	for (pch = 0; pch < d->dma_channels; pch++) {
326 		if (!(d->dma_channel_mask & (1 << pch)))
327 			continue;
328 
329 		p = &d->phy[pch];
330 
331 		if (p->vchan == NULL && !list_empty(&d->chan_pending)) {
332 			c = list_first_entry(&d->chan_pending,
333 				struct k3_dma_chan, node);
334 			/* remove from d->chan_pending */
335 			list_del_init(&c->node);
336 			pch_alloc |= 1 << pch;
337 			/* Mark this channel allocated */
338 			p->vchan = c;
339 			c->phy = p;
340 			dev_dbg(d->slave.dev, "pchan %u: alloc vchan %p\n", pch, &c->vc);
341 		}
342 	}
343 	spin_unlock_irq(&d->lock);
344 
345 	for (pch = 0; pch < d->dma_channels; pch++) {
346 		if (!(d->dma_channel_mask & (1 << pch)))
347 			continue;
348 
349 		if (pch_alloc & (1 << pch)) {
350 			p = &d->phy[pch];
351 			c = p->vchan;
352 			if (c) {
353 				spin_lock_irq(&c->vc.lock);
354 				k3_dma_start_txd(c);
355 				spin_unlock_irq(&c->vc.lock);
356 			}
357 		}
358 	}
359 }
360 
361 static void k3_dma_free_chan_resources(struct dma_chan *chan)
362 {
363 	struct k3_dma_chan *c = to_k3_chan(chan);
364 	struct k3_dma_dev *d = to_k3_dma(chan->device);
365 	unsigned long flags;
366 
367 	spin_lock_irqsave(&d->lock, flags);
368 	list_del_init(&c->node);
369 	spin_unlock_irqrestore(&d->lock, flags);
370 
371 	vchan_free_chan_resources(&c->vc);
372 	c->ccfg = 0;
373 }
374 
375 static enum dma_status k3_dma_tx_status(struct dma_chan *chan,
376 	dma_cookie_t cookie, struct dma_tx_state *state)
377 {
378 	struct k3_dma_chan *c = to_k3_chan(chan);
379 	struct k3_dma_dev *d = to_k3_dma(chan->device);
380 	struct k3_dma_phy *p;
381 	struct virt_dma_desc *vd;
382 	unsigned long flags;
383 	enum dma_status ret;
384 	size_t bytes = 0;
385 
386 	ret = dma_cookie_status(&c->vc.chan, cookie, state);
387 	if (ret == DMA_COMPLETE)
388 		return ret;
389 
390 	spin_lock_irqsave(&c->vc.lock, flags);
391 	p = c->phy;
392 	ret = c->status;
393 
394 	/*
395 	 * If the cookie is on our issue queue, then the residue is
396 	 * its total size.
397 	 */
398 	vd = vchan_find_desc(&c->vc, cookie);
399 	if (vd && !c->cyclic) {
400 		bytes = container_of(vd, struct k3_dma_desc_sw, vd)->size;
401 	} else if ((!p) || (!p->ds_run)) {
402 		bytes = 0;
403 	} else {
404 		struct k3_dma_desc_sw *ds = p->ds_run;
405 		u32 clli = 0, index = 0;
406 
407 		bytes = k3_dma_get_curr_cnt(d, p);
408 		clli = k3_dma_get_curr_lli(p);
409 		index = ((clli - ds->desc_hw_lli) /
410 				sizeof(struct k3_desc_hw)) + 1;
411 		for (; index < ds->desc_num; index++) {
412 			bytes += ds->desc_hw[index].count;
413 			/* end of lli */
414 			if (!ds->desc_hw[index].lli)
415 				break;
416 		}
417 	}
418 	spin_unlock_irqrestore(&c->vc.lock, flags);
419 	dma_set_residue(state, bytes);
420 	return ret;
421 }
422 
423 static void k3_dma_issue_pending(struct dma_chan *chan)
424 {
425 	struct k3_dma_chan *c = to_k3_chan(chan);
426 	struct k3_dma_dev *d = to_k3_dma(chan->device);
427 	unsigned long flags;
428 
429 	spin_lock_irqsave(&c->vc.lock, flags);
430 	/* add request to vc->desc_issued */
431 	if (vchan_issue_pending(&c->vc)) {
432 		spin_lock(&d->lock);
433 		if (!c->phy) {
434 			if (list_empty(&c->node)) {
435 				/* if new channel, add chan_pending */
436 				list_add_tail(&c->node, &d->chan_pending);
437 				/* check in tasklet */
438 				tasklet_schedule(&d->task);
439 				dev_dbg(d->slave.dev, "vchan %p: issued\n", &c->vc);
440 			}
441 		}
442 		spin_unlock(&d->lock);
443 	} else
444 		dev_dbg(d->slave.dev, "vchan %p: nothing to issue\n", &c->vc);
445 	spin_unlock_irqrestore(&c->vc.lock, flags);
446 }
447 
448 static void k3_dma_fill_desc(struct k3_dma_desc_sw *ds, dma_addr_t dst,
449 			dma_addr_t src, size_t len, u32 num, u32 ccfg)
450 {
451 	if (num != ds->desc_num - 1)
452 		ds->desc_hw[num].lli = ds->desc_hw_lli + (num + 1) *
453 			sizeof(struct k3_desc_hw);
454 
455 	ds->desc_hw[num].lli |= CX_LLI_CHAIN_EN;
456 	ds->desc_hw[num].count = len;
457 	ds->desc_hw[num].saddr = src;
458 	ds->desc_hw[num].daddr = dst;
459 	ds->desc_hw[num].config = ccfg;
460 }
461 
462 static struct k3_dma_desc_sw *k3_dma_alloc_desc_resource(int num,
463 							struct dma_chan *chan)
464 {
465 	struct k3_dma_chan *c = to_k3_chan(chan);
466 	struct k3_dma_desc_sw *ds;
467 	struct k3_dma_dev *d = to_k3_dma(chan->device);
468 	int lli_limit = LLI_BLOCK_SIZE / sizeof(struct k3_desc_hw);
469 
470 	if (num > lli_limit) {
471 		dev_dbg(chan->device->dev, "vch %p: sg num %d exceed max %d\n",
472 			&c->vc, num, lli_limit);
473 		return NULL;
474 	}
475 
476 	ds = kzalloc(sizeof(*ds), GFP_NOWAIT);
477 	if (!ds)
478 		return NULL;
479 
480 	ds->desc_hw = dma_pool_zalloc(d->pool, GFP_NOWAIT, &ds->desc_hw_lli);
481 	if (!ds->desc_hw) {
482 		dev_dbg(chan->device->dev, "vch %p: dma alloc fail\n", &c->vc);
483 		kfree(ds);
484 		return NULL;
485 	}
486 	ds->desc_num = num;
487 	return ds;
488 }
489 
490 static struct dma_async_tx_descriptor *k3_dma_prep_memcpy(
491 	struct dma_chan *chan,	dma_addr_t dst, dma_addr_t src,
492 	size_t len, unsigned long flags)
493 {
494 	struct k3_dma_chan *c = to_k3_chan(chan);
495 	struct k3_dma_desc_sw *ds;
496 	size_t copy = 0;
497 	int num = 0;
498 
499 	if (!len)
500 		return NULL;
501 
502 	num = DIV_ROUND_UP(len, DMA_MAX_SIZE);
503 
504 	ds = k3_dma_alloc_desc_resource(num, chan);
505 	if (!ds)
506 		return NULL;
507 
508 	c->cyclic = 0;
509 	ds->size = len;
510 	num = 0;
511 
512 	if (!c->ccfg) {
513 		/* default is memtomem, without calling device_config */
514 		c->ccfg = CX_CFG_SRCINCR | CX_CFG_DSTINCR | CX_CFG_EN;
515 		c->ccfg |= (0xf << 20) | (0xf << 24);	/* burst = 16 */
516 		c->ccfg |= (0x3 << 12) | (0x3 << 16);	/* width = 64 bit */
517 	}
518 
519 	do {
520 		copy = min_t(size_t, len, DMA_MAX_SIZE);
521 		k3_dma_fill_desc(ds, dst, src, copy, num++, c->ccfg);
522 
523 		src += copy;
524 		dst += copy;
525 		len -= copy;
526 	} while (len);
527 
528 	ds->desc_hw[num-1].lli = 0;	/* end of link */
529 	return vchan_tx_prep(&c->vc, &ds->vd, flags);
530 }
531 
532 static struct dma_async_tx_descriptor *k3_dma_prep_slave_sg(
533 	struct dma_chan *chan, struct scatterlist *sgl, unsigned int sglen,
534 	enum dma_transfer_direction dir, unsigned long flags, void *context)
535 {
536 	struct k3_dma_chan *c = to_k3_chan(chan);
537 	struct k3_dma_desc_sw *ds;
538 	size_t len, avail, total = 0;
539 	struct scatterlist *sg;
540 	dma_addr_t addr, src = 0, dst = 0;
541 	int num = sglen, i;
542 
543 	if (sgl == NULL)
544 		return NULL;
545 
546 	c->cyclic = 0;
547 
548 	for_each_sg(sgl, sg, sglen, i) {
549 		avail = sg_dma_len(sg);
550 		if (avail > DMA_MAX_SIZE)
551 			num += DIV_ROUND_UP(avail, DMA_MAX_SIZE) - 1;
552 	}
553 
554 	ds = k3_dma_alloc_desc_resource(num, chan);
555 	if (!ds)
556 		return NULL;
557 	num = 0;
558 	k3_dma_config_write(chan, dir, &c->slave_config);
559 
560 	for_each_sg(sgl, sg, sglen, i) {
561 		addr = sg_dma_address(sg);
562 		avail = sg_dma_len(sg);
563 		total += avail;
564 
565 		do {
566 			len = min_t(size_t, avail, DMA_MAX_SIZE);
567 
568 			if (dir == DMA_MEM_TO_DEV) {
569 				src = addr;
570 				dst = c->dev_addr;
571 			} else if (dir == DMA_DEV_TO_MEM) {
572 				src = c->dev_addr;
573 				dst = addr;
574 			}
575 
576 			k3_dma_fill_desc(ds, dst, src, len, num++, c->ccfg);
577 
578 			addr += len;
579 			avail -= len;
580 		} while (avail);
581 	}
582 
583 	ds->desc_hw[num-1].lli = 0;	/* end of link */
584 	ds->size = total;
585 	return vchan_tx_prep(&c->vc, &ds->vd, flags);
586 }
587 
588 static struct dma_async_tx_descriptor *
589 k3_dma_prep_dma_cyclic(struct dma_chan *chan, dma_addr_t buf_addr,
590 		       size_t buf_len, size_t period_len,
591 		       enum dma_transfer_direction dir,
592 		       unsigned long flags)
593 {
594 	struct k3_dma_chan *c = to_k3_chan(chan);
595 	struct k3_dma_desc_sw *ds;
596 	size_t len, avail, total = 0;
597 	dma_addr_t addr, src = 0, dst = 0;
598 	int num = 1, since = 0;
599 	size_t modulo = DMA_CYCLIC_MAX_PERIOD;
600 	u32 en_tc2 = 0;
601 
602 	dev_dbg(chan->device->dev, "%s: buf %pad, dst %pad, buf len %zu, period_len = %zu, dir %d\n",
603 	       __func__, &buf_addr, &to_k3_chan(chan)->dev_addr,
604 	       buf_len, period_len, (int)dir);
605 
606 	avail = buf_len;
607 	if (avail > modulo)
608 		num += DIV_ROUND_UP(avail, modulo) - 1;
609 
610 	ds = k3_dma_alloc_desc_resource(num, chan);
611 	if (!ds)
612 		return NULL;
613 
614 	c->cyclic = 1;
615 	addr = buf_addr;
616 	avail = buf_len;
617 	total = avail;
618 	num = 0;
619 	k3_dma_config_write(chan, dir, &c->slave_config);
620 
621 	if (period_len < modulo)
622 		modulo = period_len;
623 
624 	do {
625 		len = min_t(size_t, avail, modulo);
626 
627 		if (dir == DMA_MEM_TO_DEV) {
628 			src = addr;
629 			dst = c->dev_addr;
630 		} else if (dir == DMA_DEV_TO_MEM) {
631 			src = c->dev_addr;
632 			dst = addr;
633 		}
634 		since += len;
635 		if (since >= period_len) {
636 			/* descriptor asks for TC2 interrupt on completion */
637 			en_tc2 = CX_CFG_NODEIRQ;
638 			since -= period_len;
639 		} else
640 			en_tc2 = 0;
641 
642 		k3_dma_fill_desc(ds, dst, src, len, num++, c->ccfg | en_tc2);
643 
644 		addr += len;
645 		avail -= len;
646 	} while (avail);
647 
648 	/* "Cyclic" == end of link points back to start of link */
649 	ds->desc_hw[num - 1].lli |= ds->desc_hw_lli;
650 
651 	ds->size = total;
652 
653 	return vchan_tx_prep(&c->vc, &ds->vd, flags);
654 }
655 
656 static int k3_dma_config(struct dma_chan *chan,
657 			 struct dma_slave_config *cfg)
658 {
659 	struct k3_dma_chan *c = to_k3_chan(chan);
660 
661 	memcpy(&c->slave_config, cfg, sizeof(*cfg));
662 
663 	return 0;
664 }
665 
666 static int k3_dma_config_write(struct dma_chan *chan,
667 			       enum dma_transfer_direction dir,
668 			       struct dma_slave_config *cfg)
669 {
670 	struct k3_dma_chan *c = to_k3_chan(chan);
671 	u32 maxburst = 0, val = 0;
672 	enum dma_slave_buswidth width = DMA_SLAVE_BUSWIDTH_UNDEFINED;
673 
674 	if (dir == DMA_DEV_TO_MEM) {
675 		c->ccfg = CX_CFG_DSTINCR;
676 		c->dev_addr = cfg->src_addr;
677 		maxburst = cfg->src_maxburst;
678 		width = cfg->src_addr_width;
679 	} else if (dir == DMA_MEM_TO_DEV) {
680 		c->ccfg = CX_CFG_SRCINCR;
681 		c->dev_addr = cfg->dst_addr;
682 		maxburst = cfg->dst_maxburst;
683 		width = cfg->dst_addr_width;
684 	}
685 	switch (width) {
686 	case DMA_SLAVE_BUSWIDTH_1_BYTE:
687 	case DMA_SLAVE_BUSWIDTH_2_BYTES:
688 	case DMA_SLAVE_BUSWIDTH_4_BYTES:
689 	case DMA_SLAVE_BUSWIDTH_8_BYTES:
690 		val =  __ffs(width);
691 		break;
692 	default:
693 		val = 3;
694 		break;
695 	}
696 	c->ccfg |= (val << 12) | (val << 16);
697 
698 	if ((maxburst == 0) || (maxburst > 16))
699 		val = 15;
700 	else
701 		val = maxburst - 1;
702 	c->ccfg |= (val << 20) | (val << 24);
703 	c->ccfg |= CX_CFG_MEM2PER | CX_CFG_EN;
704 
705 	/* specific request line */
706 	c->ccfg |= c->vc.chan.chan_id << 4;
707 
708 	return 0;
709 }
710 
711 static void k3_dma_free_desc(struct virt_dma_desc *vd)
712 {
713 	struct k3_dma_desc_sw *ds =
714 		container_of(vd, struct k3_dma_desc_sw, vd);
715 	struct k3_dma_dev *d = to_k3_dma(vd->tx.chan->device);
716 
717 	dma_pool_free(d->pool, ds->desc_hw, ds->desc_hw_lli);
718 	kfree(ds);
719 }
720 
721 static int k3_dma_terminate_all(struct dma_chan *chan)
722 {
723 	struct k3_dma_chan *c = to_k3_chan(chan);
724 	struct k3_dma_dev *d = to_k3_dma(chan->device);
725 	struct k3_dma_phy *p = c->phy;
726 	unsigned long flags;
727 	LIST_HEAD(head);
728 
729 	dev_dbg(d->slave.dev, "vchan %p: terminate all\n", &c->vc);
730 
731 	/* Prevent this channel being scheduled */
732 	spin_lock(&d->lock);
733 	list_del_init(&c->node);
734 	spin_unlock(&d->lock);
735 
736 	/* Clear the tx descriptor lists */
737 	spin_lock_irqsave(&c->vc.lock, flags);
738 	vchan_get_all_descriptors(&c->vc, &head);
739 	if (p) {
740 		/* vchan is assigned to a pchan - stop the channel */
741 		k3_dma_terminate_chan(p, d);
742 		c->phy = NULL;
743 		p->vchan = NULL;
744 		if (p->ds_run) {
745 			vchan_terminate_vdesc(&p->ds_run->vd);
746 			p->ds_run = NULL;
747 		}
748 		p->ds_done = NULL;
749 	}
750 	spin_unlock_irqrestore(&c->vc.lock, flags);
751 	vchan_dma_desc_free_list(&c->vc, &head);
752 
753 	return 0;
754 }
755 
756 static void k3_dma_synchronize(struct dma_chan *chan)
757 {
758 	struct k3_dma_chan *c = to_k3_chan(chan);
759 
760 	vchan_synchronize(&c->vc);
761 }
762 
763 static int k3_dma_transfer_pause(struct dma_chan *chan)
764 {
765 	struct k3_dma_chan *c = to_k3_chan(chan);
766 	struct k3_dma_dev *d = to_k3_dma(chan->device);
767 	struct k3_dma_phy *p = c->phy;
768 
769 	dev_dbg(d->slave.dev, "vchan %p: pause\n", &c->vc);
770 	if (c->status == DMA_IN_PROGRESS) {
771 		c->status = DMA_PAUSED;
772 		if (p) {
773 			k3_dma_pause_dma(p, false);
774 		} else {
775 			spin_lock(&d->lock);
776 			list_del_init(&c->node);
777 			spin_unlock(&d->lock);
778 		}
779 	}
780 
781 	return 0;
782 }
783 
784 static int k3_dma_transfer_resume(struct dma_chan *chan)
785 {
786 	struct k3_dma_chan *c = to_k3_chan(chan);
787 	struct k3_dma_dev *d = to_k3_dma(chan->device);
788 	struct k3_dma_phy *p = c->phy;
789 	unsigned long flags;
790 
791 	dev_dbg(d->slave.dev, "vchan %p: resume\n", &c->vc);
792 	spin_lock_irqsave(&c->vc.lock, flags);
793 	if (c->status == DMA_PAUSED) {
794 		c->status = DMA_IN_PROGRESS;
795 		if (p) {
796 			k3_dma_pause_dma(p, true);
797 		} else if (!list_empty(&c->vc.desc_issued)) {
798 			spin_lock(&d->lock);
799 			list_add_tail(&c->node, &d->chan_pending);
800 			spin_unlock(&d->lock);
801 		}
802 	}
803 	spin_unlock_irqrestore(&c->vc.lock, flags);
804 
805 	return 0;
806 }
807 
808 static const struct k3dma_soc_data k3_v1_dma_data = {
809 	.flags = 0,
810 };
811 
812 static const struct k3dma_soc_data asp_v1_dma_data = {
813 	.flags = K3_FLAG_NOCLK,
814 };
815 
816 static const struct of_device_id k3_pdma_dt_ids[] = {
817 	{ .compatible = "hisilicon,k3-dma-1.0",
818 	  .data = &k3_v1_dma_data
819 	},
820 	{ .compatible = "hisilicon,hisi-pcm-asp-dma-1.0",
821 	  .data = &asp_v1_dma_data
822 	},
823 	{}
824 };
825 MODULE_DEVICE_TABLE(of, k3_pdma_dt_ids);
826 
827 static struct dma_chan *k3_of_dma_simple_xlate(struct of_phandle_args *dma_spec,
828 						struct of_dma *ofdma)
829 {
830 	struct k3_dma_dev *d = ofdma->of_dma_data;
831 	unsigned int request = dma_spec->args[0];
832 
833 	if (request >= d->dma_requests)
834 		return NULL;
835 
836 	return dma_get_slave_channel(&(d->chans[request].vc.chan));
837 }
838 
839 static int k3_dma_probe(struct platform_device *op)
840 {
841 	const struct k3dma_soc_data *soc_data;
842 	struct k3_dma_dev *d;
843 	const struct of_device_id *of_id;
844 	int i, ret, irq = 0;
845 
846 	d = devm_kzalloc(&op->dev, sizeof(*d), GFP_KERNEL);
847 	if (!d)
848 		return -ENOMEM;
849 
850 	soc_data = device_get_match_data(&op->dev);
851 	if (!soc_data)
852 		return -EINVAL;
853 
854 	d->base = devm_platform_ioremap_resource(op, 0);
855 	if (IS_ERR(d->base))
856 		return PTR_ERR(d->base);
857 
858 	of_id = of_match_device(k3_pdma_dt_ids, &op->dev);
859 	if (of_id) {
860 		of_property_read_u32((&op->dev)->of_node,
861 				"dma-channels", &d->dma_channels);
862 		of_property_read_u32((&op->dev)->of_node,
863 				"dma-requests", &d->dma_requests);
864 		ret = of_property_read_u32((&op->dev)->of_node,
865 				"dma-channel-mask", &d->dma_channel_mask);
866 		if (ret) {
867 			dev_warn(&op->dev,
868 				 "dma-channel-mask doesn't exist, considering all as available.\n");
869 			d->dma_channel_mask = (u32)~0UL;
870 		}
871 	}
872 
873 	if (!(soc_data->flags & K3_FLAG_NOCLK)) {
874 		d->clk = devm_clk_get(&op->dev, NULL);
875 		if (IS_ERR(d->clk)) {
876 			dev_err(&op->dev, "no dma clk\n");
877 			return PTR_ERR(d->clk);
878 		}
879 	}
880 
881 	irq = platform_get_irq(op, 0);
882 	ret = devm_request_irq(&op->dev, irq,
883 			k3_dma_int_handler, 0, DRIVER_NAME, d);
884 	if (ret)
885 		return ret;
886 
887 	d->irq = irq;
888 
889 	/* A DMA memory pool for LLIs, align on 32-byte boundary */
890 	d->pool = dmam_pool_create(DRIVER_NAME, &op->dev,
891 					LLI_BLOCK_SIZE, 32, 0);
892 	if (!d->pool)
893 		return -ENOMEM;
894 
895 	/* init phy channel */
896 	d->phy = devm_kcalloc(&op->dev,
897 		d->dma_channels, sizeof(struct k3_dma_phy), GFP_KERNEL);
898 	if (d->phy == NULL)
899 		return -ENOMEM;
900 
901 	for (i = 0; i < d->dma_channels; i++) {
902 		struct k3_dma_phy *p;
903 
904 		if (!(d->dma_channel_mask & BIT(i)))
905 			continue;
906 
907 		p = &d->phy[i];
908 		p->idx = i;
909 		p->base = d->base + i * 0x40;
910 	}
911 
912 	INIT_LIST_HEAD(&d->slave.channels);
913 	dma_cap_set(DMA_SLAVE, d->slave.cap_mask);
914 	dma_cap_set(DMA_MEMCPY, d->slave.cap_mask);
915 	dma_cap_set(DMA_CYCLIC, d->slave.cap_mask);
916 	d->slave.dev = &op->dev;
917 	d->slave.device_free_chan_resources = k3_dma_free_chan_resources;
918 	d->slave.device_tx_status = k3_dma_tx_status;
919 	d->slave.device_prep_dma_memcpy = k3_dma_prep_memcpy;
920 	d->slave.device_prep_slave_sg = k3_dma_prep_slave_sg;
921 	d->slave.device_prep_dma_cyclic = k3_dma_prep_dma_cyclic;
922 	d->slave.device_issue_pending = k3_dma_issue_pending;
923 	d->slave.device_config = k3_dma_config;
924 	d->slave.device_pause = k3_dma_transfer_pause;
925 	d->slave.device_resume = k3_dma_transfer_resume;
926 	d->slave.device_terminate_all = k3_dma_terminate_all;
927 	d->slave.device_synchronize = k3_dma_synchronize;
928 	d->slave.copy_align = DMAENGINE_ALIGN_8_BYTES;
929 
930 	/* init virtual channel */
931 	d->chans = devm_kcalloc(&op->dev,
932 		d->dma_requests, sizeof(struct k3_dma_chan), GFP_KERNEL);
933 	if (d->chans == NULL)
934 		return -ENOMEM;
935 
936 	for (i = 0; i < d->dma_requests; i++) {
937 		struct k3_dma_chan *c = &d->chans[i];
938 
939 		c->status = DMA_IN_PROGRESS;
940 		INIT_LIST_HEAD(&c->node);
941 		c->vc.desc_free = k3_dma_free_desc;
942 		vchan_init(&c->vc, &d->slave);
943 	}
944 
945 	/* Enable clock before accessing registers */
946 	ret = clk_prepare_enable(d->clk);
947 	if (ret < 0) {
948 		dev_err(&op->dev, "clk_prepare_enable failed: %d\n", ret);
949 		return ret;
950 	}
951 
952 	k3_dma_enable_dma(d, true);
953 
954 	ret = dma_async_device_register(&d->slave);
955 	if (ret)
956 		goto dma_async_register_fail;
957 
958 	ret = of_dma_controller_register((&op->dev)->of_node,
959 					k3_of_dma_simple_xlate, d);
960 	if (ret)
961 		goto of_dma_register_fail;
962 
963 	spin_lock_init(&d->lock);
964 	INIT_LIST_HEAD(&d->chan_pending);
965 	tasklet_setup(&d->task, k3_dma_tasklet);
966 	platform_set_drvdata(op, d);
967 	dev_info(&op->dev, "initialized\n");
968 
969 	return 0;
970 
971 of_dma_register_fail:
972 	dma_async_device_unregister(&d->slave);
973 dma_async_register_fail:
974 	clk_disable_unprepare(d->clk);
975 	return ret;
976 }
977 
978 static int k3_dma_remove(struct platform_device *op)
979 {
980 	struct k3_dma_chan *c, *cn;
981 	struct k3_dma_dev *d = platform_get_drvdata(op);
982 
983 	dma_async_device_unregister(&d->slave);
984 	of_dma_controller_free((&op->dev)->of_node);
985 
986 	devm_free_irq(&op->dev, d->irq, d);
987 
988 	list_for_each_entry_safe(c, cn, &d->slave.channels, vc.chan.device_node) {
989 		list_del(&c->vc.chan.device_node);
990 		tasklet_kill(&c->vc.task);
991 	}
992 	tasklet_kill(&d->task);
993 	clk_disable_unprepare(d->clk);
994 	return 0;
995 }
996 
997 #ifdef CONFIG_PM_SLEEP
998 static int k3_dma_suspend_dev(struct device *dev)
999 {
1000 	struct k3_dma_dev *d = dev_get_drvdata(dev);
1001 	u32 stat = 0;
1002 
1003 	stat = k3_dma_get_chan_stat(d);
1004 	if (stat) {
1005 		dev_warn(d->slave.dev,
1006 			"chan %d is running fail to suspend\n", stat);
1007 		return -1;
1008 	}
1009 	k3_dma_enable_dma(d, false);
1010 	clk_disable_unprepare(d->clk);
1011 	return 0;
1012 }
1013 
1014 static int k3_dma_resume_dev(struct device *dev)
1015 {
1016 	struct k3_dma_dev *d = dev_get_drvdata(dev);
1017 	int ret = 0;
1018 
1019 	ret = clk_prepare_enable(d->clk);
1020 	if (ret < 0) {
1021 		dev_err(d->slave.dev, "clk_prepare_enable failed: %d\n", ret);
1022 		return ret;
1023 	}
1024 	k3_dma_enable_dma(d, true);
1025 	return 0;
1026 }
1027 #endif
1028 
1029 static SIMPLE_DEV_PM_OPS(k3_dma_pmops, k3_dma_suspend_dev, k3_dma_resume_dev);
1030 
1031 static struct platform_driver k3_pdma_driver = {
1032 	.driver		= {
1033 		.name	= DRIVER_NAME,
1034 		.pm	= &k3_dma_pmops,
1035 		.of_match_table = k3_pdma_dt_ids,
1036 	},
1037 	.probe		= k3_dma_probe,
1038 	.remove		= k3_dma_remove,
1039 };
1040 
1041 module_platform_driver(k3_pdma_driver);
1042 
1043 MODULE_DESCRIPTION("Hisilicon k3 DMA Driver");
1044 MODULE_ALIAS("platform:k3dma");
1045 MODULE_LICENSE("GPL v2");
1046