xref: /openbmc/linux/drivers/dma/k3dma.c (revision 96ac6d43)
1 /*
2  * Copyright (c) 2013 - 2015 Linaro Ltd.
3  * Copyright (c) 2013 Hisilicon Limited.
4  *
5  * This program is free software; you can redistribute it and/or modify
6  * it under the terms of the GNU General Public License version 2 as
7  * published by the Free Software Foundation.
8  */
9 #include <linux/sched.h>
10 #include <linux/device.h>
11 #include <linux/dma-mapping.h>
12 #include <linux/dmapool.h>
13 #include <linux/dmaengine.h>
14 #include <linux/init.h>
15 #include <linux/interrupt.h>
16 #include <linux/kernel.h>
17 #include <linux/module.h>
18 #include <linux/platform_device.h>
19 #include <linux/slab.h>
20 #include <linux/spinlock.h>
21 #include <linux/of_device.h>
22 #include <linux/of.h>
23 #include <linux/clk.h>
24 #include <linux/of_dma.h>
25 
26 #include "virt-dma.h"
27 
28 #define DRIVER_NAME		"k3-dma"
29 #define DMA_MAX_SIZE		0x1ffc
30 #define DMA_CYCLIC_MAX_PERIOD	0x1000
31 #define LLI_BLOCK_SIZE		(4 * PAGE_SIZE)
32 
33 #define INT_STAT		0x00
34 #define INT_TC1			0x04
35 #define INT_TC2			0x08
36 #define INT_ERR1		0x0c
37 #define INT_ERR2		0x10
38 #define INT_TC1_MASK		0x18
39 #define INT_TC2_MASK		0x1c
40 #define INT_ERR1_MASK		0x20
41 #define INT_ERR2_MASK		0x24
42 #define INT_TC1_RAW		0x600
43 #define INT_TC2_RAW		0x608
44 #define INT_ERR1_RAW		0x610
45 #define INT_ERR2_RAW		0x618
46 #define CH_PRI			0x688
47 #define CH_STAT			0x690
48 #define CX_CUR_CNT		0x704
49 #define CX_LLI			0x800
50 #define CX_CNT1			0x80c
51 #define CX_CNT0			0x810
52 #define CX_SRC			0x814
53 #define CX_DST			0x818
54 #define CX_CFG			0x81c
55 
56 #define CX_LLI_CHAIN_EN		0x2
57 #define CX_CFG_EN		0x1
58 #define CX_CFG_NODEIRQ		BIT(1)
59 #define CX_CFG_MEM2PER		(0x1 << 2)
60 #define CX_CFG_PER2MEM		(0x2 << 2)
61 #define CX_CFG_SRCINCR		(0x1 << 31)
62 #define CX_CFG_DSTINCR		(0x1 << 30)
63 
64 struct k3_desc_hw {
65 	u32 lli;
66 	u32 reserved[3];
67 	u32 count;
68 	u32 saddr;
69 	u32 daddr;
70 	u32 config;
71 } __aligned(32);
72 
73 struct k3_dma_desc_sw {
74 	struct virt_dma_desc	vd;
75 	dma_addr_t		desc_hw_lli;
76 	size_t			desc_num;
77 	size_t			size;
78 	struct k3_desc_hw	*desc_hw;
79 };
80 
81 struct k3_dma_phy;
82 
83 struct k3_dma_chan {
84 	u32			ccfg;
85 	struct virt_dma_chan	vc;
86 	struct k3_dma_phy	*phy;
87 	struct list_head	node;
88 	dma_addr_t		dev_addr;
89 	enum dma_status		status;
90 	bool			cyclic;
91 	struct dma_slave_config	slave_config;
92 };
93 
94 struct k3_dma_phy {
95 	u32			idx;
96 	void __iomem		*base;
97 	struct k3_dma_chan	*vchan;
98 	struct k3_dma_desc_sw	*ds_run;
99 	struct k3_dma_desc_sw	*ds_done;
100 };
101 
102 struct k3_dma_dev {
103 	struct dma_device	slave;
104 	void __iomem		*base;
105 	struct tasklet_struct	task;
106 	spinlock_t		lock;
107 	struct list_head	chan_pending;
108 	struct k3_dma_phy	*phy;
109 	struct k3_dma_chan	*chans;
110 	struct clk		*clk;
111 	struct dma_pool		*pool;
112 	u32			dma_channels;
113 	u32			dma_requests;
114 	u32			dma_channel_mask;
115 	unsigned int		irq;
116 };
117 
118 
119 #define K3_FLAG_NOCLK	BIT(1)
120 
121 struct k3dma_soc_data {
122 	unsigned long flags;
123 };
124 
125 
126 #define to_k3_dma(dmadev) container_of(dmadev, struct k3_dma_dev, slave)
127 
128 static int k3_dma_config_write(struct dma_chan *chan,
129 			       enum dma_transfer_direction dir,
130 			       struct dma_slave_config *cfg);
131 
132 static struct k3_dma_chan *to_k3_chan(struct dma_chan *chan)
133 {
134 	return container_of(chan, struct k3_dma_chan, vc.chan);
135 }
136 
137 static void k3_dma_pause_dma(struct k3_dma_phy *phy, bool on)
138 {
139 	u32 val = 0;
140 
141 	if (on) {
142 		val = readl_relaxed(phy->base + CX_CFG);
143 		val |= CX_CFG_EN;
144 		writel_relaxed(val, phy->base + CX_CFG);
145 	} else {
146 		val = readl_relaxed(phy->base + CX_CFG);
147 		val &= ~CX_CFG_EN;
148 		writel_relaxed(val, phy->base + CX_CFG);
149 	}
150 }
151 
152 static void k3_dma_terminate_chan(struct k3_dma_phy *phy, struct k3_dma_dev *d)
153 {
154 	u32 val = 0;
155 
156 	k3_dma_pause_dma(phy, false);
157 
158 	val = 0x1 << phy->idx;
159 	writel_relaxed(val, d->base + INT_TC1_RAW);
160 	writel_relaxed(val, d->base + INT_TC2_RAW);
161 	writel_relaxed(val, d->base + INT_ERR1_RAW);
162 	writel_relaxed(val, d->base + INT_ERR2_RAW);
163 }
164 
165 static void k3_dma_set_desc(struct k3_dma_phy *phy, struct k3_desc_hw *hw)
166 {
167 	writel_relaxed(hw->lli, phy->base + CX_LLI);
168 	writel_relaxed(hw->count, phy->base + CX_CNT0);
169 	writel_relaxed(hw->saddr, phy->base + CX_SRC);
170 	writel_relaxed(hw->daddr, phy->base + CX_DST);
171 	writel_relaxed(hw->config, phy->base + CX_CFG);
172 }
173 
174 static u32 k3_dma_get_curr_cnt(struct k3_dma_dev *d, struct k3_dma_phy *phy)
175 {
176 	u32 cnt = 0;
177 
178 	cnt = readl_relaxed(d->base + CX_CUR_CNT + phy->idx * 0x10);
179 	cnt &= 0xffff;
180 	return cnt;
181 }
182 
183 static u32 k3_dma_get_curr_lli(struct k3_dma_phy *phy)
184 {
185 	return readl_relaxed(phy->base + CX_LLI);
186 }
187 
188 static u32 k3_dma_get_chan_stat(struct k3_dma_dev *d)
189 {
190 	return readl_relaxed(d->base + CH_STAT);
191 }
192 
193 static void k3_dma_enable_dma(struct k3_dma_dev *d, bool on)
194 {
195 	if (on) {
196 		/* set same priority */
197 		writel_relaxed(0x0, d->base + CH_PRI);
198 
199 		/* unmask irq */
200 		writel_relaxed(0xffff, d->base + INT_TC1_MASK);
201 		writel_relaxed(0xffff, d->base + INT_TC2_MASK);
202 		writel_relaxed(0xffff, d->base + INT_ERR1_MASK);
203 		writel_relaxed(0xffff, d->base + INT_ERR2_MASK);
204 	} else {
205 		/* mask irq */
206 		writel_relaxed(0x0, d->base + INT_TC1_MASK);
207 		writel_relaxed(0x0, d->base + INT_TC2_MASK);
208 		writel_relaxed(0x0, d->base + INT_ERR1_MASK);
209 		writel_relaxed(0x0, d->base + INT_ERR2_MASK);
210 	}
211 }
212 
213 static irqreturn_t k3_dma_int_handler(int irq, void *dev_id)
214 {
215 	struct k3_dma_dev *d = (struct k3_dma_dev *)dev_id;
216 	struct k3_dma_phy *p;
217 	struct k3_dma_chan *c;
218 	u32 stat = readl_relaxed(d->base + INT_STAT);
219 	u32 tc1  = readl_relaxed(d->base + INT_TC1);
220 	u32 tc2  = readl_relaxed(d->base + INT_TC2);
221 	u32 err1 = readl_relaxed(d->base + INT_ERR1);
222 	u32 err2 = readl_relaxed(d->base + INT_ERR2);
223 	u32 i, irq_chan = 0;
224 
225 	while (stat) {
226 		i = __ffs(stat);
227 		stat &= ~BIT(i);
228 		if (likely(tc1 & BIT(i)) || (tc2 & BIT(i))) {
229 			unsigned long flags;
230 
231 			p = &d->phy[i];
232 			c = p->vchan;
233 			if (c && (tc1 & BIT(i))) {
234 				spin_lock_irqsave(&c->vc.lock, flags);
235 				vchan_cookie_complete(&p->ds_run->vd);
236 				p->ds_done = p->ds_run;
237 				p->ds_run = NULL;
238 				spin_unlock_irqrestore(&c->vc.lock, flags);
239 			}
240 			if (c && (tc2 & BIT(i))) {
241 				spin_lock_irqsave(&c->vc.lock, flags);
242 				if (p->ds_run != NULL)
243 					vchan_cyclic_callback(&p->ds_run->vd);
244 				spin_unlock_irqrestore(&c->vc.lock, flags);
245 			}
246 			irq_chan |= BIT(i);
247 		}
248 		if (unlikely((err1 & BIT(i)) || (err2 & BIT(i))))
249 			dev_warn(d->slave.dev, "DMA ERR\n");
250 	}
251 
252 	writel_relaxed(irq_chan, d->base + INT_TC1_RAW);
253 	writel_relaxed(irq_chan, d->base + INT_TC2_RAW);
254 	writel_relaxed(err1, d->base + INT_ERR1_RAW);
255 	writel_relaxed(err2, d->base + INT_ERR2_RAW);
256 
257 	if (irq_chan)
258 		tasklet_schedule(&d->task);
259 
260 	if (irq_chan || err1 || err2)
261 		return IRQ_HANDLED;
262 
263 	return IRQ_NONE;
264 }
265 
266 static int k3_dma_start_txd(struct k3_dma_chan *c)
267 {
268 	struct k3_dma_dev *d = to_k3_dma(c->vc.chan.device);
269 	struct virt_dma_desc *vd = vchan_next_desc(&c->vc);
270 
271 	if (!c->phy)
272 		return -EAGAIN;
273 
274 	if (BIT(c->phy->idx) & k3_dma_get_chan_stat(d))
275 		return -EAGAIN;
276 
277 	if (vd) {
278 		struct k3_dma_desc_sw *ds =
279 			container_of(vd, struct k3_dma_desc_sw, vd);
280 		/*
281 		 * fetch and remove request from vc->desc_issued
282 		 * so vc->desc_issued only contains desc pending
283 		 */
284 		list_del(&ds->vd.node);
285 
286 		c->phy->ds_run = ds;
287 		c->phy->ds_done = NULL;
288 		/* start dma */
289 		k3_dma_set_desc(c->phy, &ds->desc_hw[0]);
290 		return 0;
291 	}
292 	c->phy->ds_run = NULL;
293 	c->phy->ds_done = NULL;
294 	return -EAGAIN;
295 }
296 
297 static void k3_dma_tasklet(unsigned long arg)
298 {
299 	struct k3_dma_dev *d = (struct k3_dma_dev *)arg;
300 	struct k3_dma_phy *p;
301 	struct k3_dma_chan *c, *cn;
302 	unsigned pch, pch_alloc = 0;
303 
304 	/* check new dma request of running channel in vc->desc_issued */
305 	list_for_each_entry_safe(c, cn, &d->slave.channels, vc.chan.device_node) {
306 		spin_lock_irq(&c->vc.lock);
307 		p = c->phy;
308 		if (p && p->ds_done) {
309 			if (k3_dma_start_txd(c)) {
310 				/* No current txd associated with this channel */
311 				dev_dbg(d->slave.dev, "pchan %u: free\n", p->idx);
312 				/* Mark this channel free */
313 				c->phy = NULL;
314 				p->vchan = NULL;
315 			}
316 		}
317 		spin_unlock_irq(&c->vc.lock);
318 	}
319 
320 	/* check new channel request in d->chan_pending */
321 	spin_lock_irq(&d->lock);
322 	for (pch = 0; pch < d->dma_channels; pch++) {
323 		if (!(d->dma_channel_mask & (1 << pch)))
324 			continue;
325 
326 		p = &d->phy[pch];
327 
328 		if (p->vchan == NULL && !list_empty(&d->chan_pending)) {
329 			c = list_first_entry(&d->chan_pending,
330 				struct k3_dma_chan, node);
331 			/* remove from d->chan_pending */
332 			list_del_init(&c->node);
333 			pch_alloc |= 1 << pch;
334 			/* Mark this channel allocated */
335 			p->vchan = c;
336 			c->phy = p;
337 			dev_dbg(d->slave.dev, "pchan %u: alloc vchan %p\n", pch, &c->vc);
338 		}
339 	}
340 	spin_unlock_irq(&d->lock);
341 
342 	for (pch = 0; pch < d->dma_channels; pch++) {
343 		if (!(d->dma_channel_mask & (1 << pch)))
344 			continue;
345 
346 		if (pch_alloc & (1 << pch)) {
347 			p = &d->phy[pch];
348 			c = p->vchan;
349 			if (c) {
350 				spin_lock_irq(&c->vc.lock);
351 				k3_dma_start_txd(c);
352 				spin_unlock_irq(&c->vc.lock);
353 			}
354 		}
355 	}
356 }
357 
358 static void k3_dma_free_chan_resources(struct dma_chan *chan)
359 {
360 	struct k3_dma_chan *c = to_k3_chan(chan);
361 	struct k3_dma_dev *d = to_k3_dma(chan->device);
362 	unsigned long flags;
363 
364 	spin_lock_irqsave(&d->lock, flags);
365 	list_del_init(&c->node);
366 	spin_unlock_irqrestore(&d->lock, flags);
367 
368 	vchan_free_chan_resources(&c->vc);
369 	c->ccfg = 0;
370 }
371 
372 static enum dma_status k3_dma_tx_status(struct dma_chan *chan,
373 	dma_cookie_t cookie, struct dma_tx_state *state)
374 {
375 	struct k3_dma_chan *c = to_k3_chan(chan);
376 	struct k3_dma_dev *d = to_k3_dma(chan->device);
377 	struct k3_dma_phy *p;
378 	struct virt_dma_desc *vd;
379 	unsigned long flags;
380 	enum dma_status ret;
381 	size_t bytes = 0;
382 
383 	ret = dma_cookie_status(&c->vc.chan, cookie, state);
384 	if (ret == DMA_COMPLETE)
385 		return ret;
386 
387 	spin_lock_irqsave(&c->vc.lock, flags);
388 	p = c->phy;
389 	ret = c->status;
390 
391 	/*
392 	 * If the cookie is on our issue queue, then the residue is
393 	 * its total size.
394 	 */
395 	vd = vchan_find_desc(&c->vc, cookie);
396 	if (vd && !c->cyclic) {
397 		bytes = container_of(vd, struct k3_dma_desc_sw, vd)->size;
398 	} else if ((!p) || (!p->ds_run)) {
399 		bytes = 0;
400 	} else {
401 		struct k3_dma_desc_sw *ds = p->ds_run;
402 		u32 clli = 0, index = 0;
403 
404 		bytes = k3_dma_get_curr_cnt(d, p);
405 		clli = k3_dma_get_curr_lli(p);
406 		index = ((clli - ds->desc_hw_lli) /
407 				sizeof(struct k3_desc_hw)) + 1;
408 		for (; index < ds->desc_num; index++) {
409 			bytes += ds->desc_hw[index].count;
410 			/* end of lli */
411 			if (!ds->desc_hw[index].lli)
412 				break;
413 		}
414 	}
415 	spin_unlock_irqrestore(&c->vc.lock, flags);
416 	dma_set_residue(state, bytes);
417 	return ret;
418 }
419 
420 static void k3_dma_issue_pending(struct dma_chan *chan)
421 {
422 	struct k3_dma_chan *c = to_k3_chan(chan);
423 	struct k3_dma_dev *d = to_k3_dma(chan->device);
424 	unsigned long flags;
425 
426 	spin_lock_irqsave(&c->vc.lock, flags);
427 	/* add request to vc->desc_issued */
428 	if (vchan_issue_pending(&c->vc)) {
429 		spin_lock(&d->lock);
430 		if (!c->phy) {
431 			if (list_empty(&c->node)) {
432 				/* if new channel, add chan_pending */
433 				list_add_tail(&c->node, &d->chan_pending);
434 				/* check in tasklet */
435 				tasklet_schedule(&d->task);
436 				dev_dbg(d->slave.dev, "vchan %p: issued\n", &c->vc);
437 			}
438 		}
439 		spin_unlock(&d->lock);
440 	} else
441 		dev_dbg(d->slave.dev, "vchan %p: nothing to issue\n", &c->vc);
442 	spin_unlock_irqrestore(&c->vc.lock, flags);
443 }
444 
445 static void k3_dma_fill_desc(struct k3_dma_desc_sw *ds, dma_addr_t dst,
446 			dma_addr_t src, size_t len, u32 num, u32 ccfg)
447 {
448 	if (num != ds->desc_num - 1)
449 		ds->desc_hw[num].lli = ds->desc_hw_lli + (num + 1) *
450 			sizeof(struct k3_desc_hw);
451 
452 	ds->desc_hw[num].lli |= CX_LLI_CHAIN_EN;
453 	ds->desc_hw[num].count = len;
454 	ds->desc_hw[num].saddr = src;
455 	ds->desc_hw[num].daddr = dst;
456 	ds->desc_hw[num].config = ccfg;
457 }
458 
459 static struct k3_dma_desc_sw *k3_dma_alloc_desc_resource(int num,
460 							struct dma_chan *chan)
461 {
462 	struct k3_dma_chan *c = to_k3_chan(chan);
463 	struct k3_dma_desc_sw *ds;
464 	struct k3_dma_dev *d = to_k3_dma(chan->device);
465 	int lli_limit = LLI_BLOCK_SIZE / sizeof(struct k3_desc_hw);
466 
467 	if (num > lli_limit) {
468 		dev_dbg(chan->device->dev, "vch %p: sg num %d exceed max %d\n",
469 			&c->vc, num, lli_limit);
470 		return NULL;
471 	}
472 
473 	ds = kzalloc(sizeof(*ds), GFP_NOWAIT);
474 	if (!ds)
475 		return NULL;
476 
477 	ds->desc_hw = dma_pool_zalloc(d->pool, GFP_NOWAIT, &ds->desc_hw_lli);
478 	if (!ds->desc_hw) {
479 		dev_dbg(chan->device->dev, "vch %p: dma alloc fail\n", &c->vc);
480 		kfree(ds);
481 		return NULL;
482 	}
483 	ds->desc_num = num;
484 	return ds;
485 }
486 
487 static struct dma_async_tx_descriptor *k3_dma_prep_memcpy(
488 	struct dma_chan *chan,	dma_addr_t dst, dma_addr_t src,
489 	size_t len, unsigned long flags)
490 {
491 	struct k3_dma_chan *c = to_k3_chan(chan);
492 	struct k3_dma_desc_sw *ds;
493 	size_t copy = 0;
494 	int num = 0;
495 
496 	if (!len)
497 		return NULL;
498 
499 	num = DIV_ROUND_UP(len, DMA_MAX_SIZE);
500 
501 	ds = k3_dma_alloc_desc_resource(num, chan);
502 	if (!ds)
503 		return NULL;
504 
505 	c->cyclic = 0;
506 	ds->size = len;
507 	num = 0;
508 
509 	if (!c->ccfg) {
510 		/* default is memtomem, without calling device_config */
511 		c->ccfg = CX_CFG_SRCINCR | CX_CFG_DSTINCR | CX_CFG_EN;
512 		c->ccfg |= (0xf << 20) | (0xf << 24);	/* burst = 16 */
513 		c->ccfg |= (0x3 << 12) | (0x3 << 16);	/* width = 64 bit */
514 	}
515 
516 	do {
517 		copy = min_t(size_t, len, DMA_MAX_SIZE);
518 		k3_dma_fill_desc(ds, dst, src, copy, num++, c->ccfg);
519 
520 		src += copy;
521 		dst += copy;
522 		len -= copy;
523 	} while (len);
524 
525 	ds->desc_hw[num-1].lli = 0;	/* end of link */
526 	return vchan_tx_prep(&c->vc, &ds->vd, flags);
527 }
528 
529 static struct dma_async_tx_descriptor *k3_dma_prep_slave_sg(
530 	struct dma_chan *chan, struct scatterlist *sgl, unsigned int sglen,
531 	enum dma_transfer_direction dir, unsigned long flags, void *context)
532 {
533 	struct k3_dma_chan *c = to_k3_chan(chan);
534 	struct k3_dma_desc_sw *ds;
535 	size_t len, avail, total = 0;
536 	struct scatterlist *sg;
537 	dma_addr_t addr, src = 0, dst = 0;
538 	int num = sglen, i;
539 
540 	if (sgl == NULL)
541 		return NULL;
542 
543 	c->cyclic = 0;
544 
545 	for_each_sg(sgl, sg, sglen, i) {
546 		avail = sg_dma_len(sg);
547 		if (avail > DMA_MAX_SIZE)
548 			num += DIV_ROUND_UP(avail, DMA_MAX_SIZE) - 1;
549 	}
550 
551 	ds = k3_dma_alloc_desc_resource(num, chan);
552 	if (!ds)
553 		return NULL;
554 	num = 0;
555 	k3_dma_config_write(chan, dir, &c->slave_config);
556 
557 	for_each_sg(sgl, sg, sglen, i) {
558 		addr = sg_dma_address(sg);
559 		avail = sg_dma_len(sg);
560 		total += avail;
561 
562 		do {
563 			len = min_t(size_t, avail, DMA_MAX_SIZE);
564 
565 			if (dir == DMA_MEM_TO_DEV) {
566 				src = addr;
567 				dst = c->dev_addr;
568 			} else if (dir == DMA_DEV_TO_MEM) {
569 				src = c->dev_addr;
570 				dst = addr;
571 			}
572 
573 			k3_dma_fill_desc(ds, dst, src, len, num++, c->ccfg);
574 
575 			addr += len;
576 			avail -= len;
577 		} while (avail);
578 	}
579 
580 	ds->desc_hw[num-1].lli = 0;	/* end of link */
581 	ds->size = total;
582 	return vchan_tx_prep(&c->vc, &ds->vd, flags);
583 }
584 
585 static struct dma_async_tx_descriptor *
586 k3_dma_prep_dma_cyclic(struct dma_chan *chan, dma_addr_t buf_addr,
587 		       size_t buf_len, size_t period_len,
588 		       enum dma_transfer_direction dir,
589 		       unsigned long flags)
590 {
591 	struct k3_dma_chan *c = to_k3_chan(chan);
592 	struct k3_dma_desc_sw *ds;
593 	size_t len, avail, total = 0;
594 	dma_addr_t addr, src = 0, dst = 0;
595 	int num = 1, since = 0;
596 	size_t modulo = DMA_CYCLIC_MAX_PERIOD;
597 	u32 en_tc2 = 0;
598 
599 	dev_dbg(chan->device->dev, "%s: buf %pad, dst %pad, buf len %zu, period_len = %zu, dir %d\n",
600 	       __func__, &buf_addr, &to_k3_chan(chan)->dev_addr,
601 	       buf_len, period_len, (int)dir);
602 
603 	avail = buf_len;
604 	if (avail > modulo)
605 		num += DIV_ROUND_UP(avail, modulo) - 1;
606 
607 	ds = k3_dma_alloc_desc_resource(num, chan);
608 	if (!ds)
609 		return NULL;
610 
611 	c->cyclic = 1;
612 	addr = buf_addr;
613 	avail = buf_len;
614 	total = avail;
615 	num = 0;
616 	k3_dma_config_write(chan, dir, &c->slave_config);
617 
618 	if (period_len < modulo)
619 		modulo = period_len;
620 
621 	do {
622 		len = min_t(size_t, avail, modulo);
623 
624 		if (dir == DMA_MEM_TO_DEV) {
625 			src = addr;
626 			dst = c->dev_addr;
627 		} else if (dir == DMA_DEV_TO_MEM) {
628 			src = c->dev_addr;
629 			dst = addr;
630 		}
631 		since += len;
632 		if (since >= period_len) {
633 			/* descriptor asks for TC2 interrupt on completion */
634 			en_tc2 = CX_CFG_NODEIRQ;
635 			since -= period_len;
636 		} else
637 			en_tc2 = 0;
638 
639 		k3_dma_fill_desc(ds, dst, src, len, num++, c->ccfg | en_tc2);
640 
641 		addr += len;
642 		avail -= len;
643 	} while (avail);
644 
645 	/* "Cyclic" == end of link points back to start of link */
646 	ds->desc_hw[num - 1].lli |= ds->desc_hw_lli;
647 
648 	ds->size = total;
649 
650 	return vchan_tx_prep(&c->vc, &ds->vd, flags);
651 }
652 
653 static int k3_dma_config(struct dma_chan *chan,
654 			 struct dma_slave_config *cfg)
655 {
656 	struct k3_dma_chan *c = to_k3_chan(chan);
657 
658 	memcpy(&c->slave_config, cfg, sizeof(*cfg));
659 
660 	return 0;
661 }
662 
663 static int k3_dma_config_write(struct dma_chan *chan,
664 			       enum dma_transfer_direction dir,
665 			       struct dma_slave_config *cfg)
666 {
667 	struct k3_dma_chan *c = to_k3_chan(chan);
668 	u32 maxburst = 0, val = 0;
669 	enum dma_slave_buswidth width = DMA_SLAVE_BUSWIDTH_UNDEFINED;
670 
671 	if (dir == DMA_DEV_TO_MEM) {
672 		c->ccfg = CX_CFG_DSTINCR;
673 		c->dev_addr = cfg->src_addr;
674 		maxburst = cfg->src_maxburst;
675 		width = cfg->src_addr_width;
676 	} else if (dir == DMA_MEM_TO_DEV) {
677 		c->ccfg = CX_CFG_SRCINCR;
678 		c->dev_addr = cfg->dst_addr;
679 		maxburst = cfg->dst_maxburst;
680 		width = cfg->dst_addr_width;
681 	}
682 	switch (width) {
683 	case DMA_SLAVE_BUSWIDTH_1_BYTE:
684 	case DMA_SLAVE_BUSWIDTH_2_BYTES:
685 	case DMA_SLAVE_BUSWIDTH_4_BYTES:
686 	case DMA_SLAVE_BUSWIDTH_8_BYTES:
687 		val =  __ffs(width);
688 		break;
689 	default:
690 		val = 3;
691 		break;
692 	}
693 	c->ccfg |= (val << 12) | (val << 16);
694 
695 	if ((maxburst == 0) || (maxburst > 16))
696 		val = 15;
697 	else
698 		val = maxburst - 1;
699 	c->ccfg |= (val << 20) | (val << 24);
700 	c->ccfg |= CX_CFG_MEM2PER | CX_CFG_EN;
701 
702 	/* specific request line */
703 	c->ccfg |= c->vc.chan.chan_id << 4;
704 
705 	return 0;
706 }
707 
708 static void k3_dma_free_desc(struct virt_dma_desc *vd)
709 {
710 	struct k3_dma_desc_sw *ds =
711 		container_of(vd, struct k3_dma_desc_sw, vd);
712 	struct k3_dma_dev *d = to_k3_dma(vd->tx.chan->device);
713 
714 	dma_pool_free(d->pool, ds->desc_hw, ds->desc_hw_lli);
715 	kfree(ds);
716 }
717 
718 static int k3_dma_terminate_all(struct dma_chan *chan)
719 {
720 	struct k3_dma_chan *c = to_k3_chan(chan);
721 	struct k3_dma_dev *d = to_k3_dma(chan->device);
722 	struct k3_dma_phy *p = c->phy;
723 	unsigned long flags;
724 	LIST_HEAD(head);
725 
726 	dev_dbg(d->slave.dev, "vchan %p: terminate all\n", &c->vc);
727 
728 	/* Prevent this channel being scheduled */
729 	spin_lock(&d->lock);
730 	list_del_init(&c->node);
731 	spin_unlock(&d->lock);
732 
733 	/* Clear the tx descriptor lists */
734 	spin_lock_irqsave(&c->vc.lock, flags);
735 	vchan_get_all_descriptors(&c->vc, &head);
736 	if (p) {
737 		/* vchan is assigned to a pchan - stop the channel */
738 		k3_dma_terminate_chan(p, d);
739 		c->phy = NULL;
740 		p->vchan = NULL;
741 		if (p->ds_run) {
742 			vchan_terminate_vdesc(&p->ds_run->vd);
743 			p->ds_run = NULL;
744 		}
745 		p->ds_done = NULL;
746 	}
747 	spin_unlock_irqrestore(&c->vc.lock, flags);
748 	vchan_dma_desc_free_list(&c->vc, &head);
749 
750 	return 0;
751 }
752 
753 static void k3_dma_synchronize(struct dma_chan *chan)
754 {
755 	struct k3_dma_chan *c = to_k3_chan(chan);
756 
757 	vchan_synchronize(&c->vc);
758 }
759 
760 static int k3_dma_transfer_pause(struct dma_chan *chan)
761 {
762 	struct k3_dma_chan *c = to_k3_chan(chan);
763 	struct k3_dma_dev *d = to_k3_dma(chan->device);
764 	struct k3_dma_phy *p = c->phy;
765 
766 	dev_dbg(d->slave.dev, "vchan %p: pause\n", &c->vc);
767 	if (c->status == DMA_IN_PROGRESS) {
768 		c->status = DMA_PAUSED;
769 		if (p) {
770 			k3_dma_pause_dma(p, false);
771 		} else {
772 			spin_lock(&d->lock);
773 			list_del_init(&c->node);
774 			spin_unlock(&d->lock);
775 		}
776 	}
777 
778 	return 0;
779 }
780 
781 static int k3_dma_transfer_resume(struct dma_chan *chan)
782 {
783 	struct k3_dma_chan *c = to_k3_chan(chan);
784 	struct k3_dma_dev *d = to_k3_dma(chan->device);
785 	struct k3_dma_phy *p = c->phy;
786 	unsigned long flags;
787 
788 	dev_dbg(d->slave.dev, "vchan %p: resume\n", &c->vc);
789 	spin_lock_irqsave(&c->vc.lock, flags);
790 	if (c->status == DMA_PAUSED) {
791 		c->status = DMA_IN_PROGRESS;
792 		if (p) {
793 			k3_dma_pause_dma(p, true);
794 		} else if (!list_empty(&c->vc.desc_issued)) {
795 			spin_lock(&d->lock);
796 			list_add_tail(&c->node, &d->chan_pending);
797 			spin_unlock(&d->lock);
798 		}
799 	}
800 	spin_unlock_irqrestore(&c->vc.lock, flags);
801 
802 	return 0;
803 }
804 
805 static const struct k3dma_soc_data k3_v1_dma_data = {
806 	.flags = 0,
807 };
808 
809 static const struct k3dma_soc_data asp_v1_dma_data = {
810 	.flags = K3_FLAG_NOCLK,
811 };
812 
813 static const struct of_device_id k3_pdma_dt_ids[] = {
814 	{ .compatible = "hisilicon,k3-dma-1.0",
815 	  .data = &k3_v1_dma_data
816 	},
817 	{ .compatible = "hisilicon,hisi-pcm-asp-dma-1.0",
818 	  .data = &asp_v1_dma_data
819 	},
820 	{}
821 };
822 MODULE_DEVICE_TABLE(of, k3_pdma_dt_ids);
823 
824 static struct dma_chan *k3_of_dma_simple_xlate(struct of_phandle_args *dma_spec,
825 						struct of_dma *ofdma)
826 {
827 	struct k3_dma_dev *d = ofdma->of_dma_data;
828 	unsigned int request = dma_spec->args[0];
829 
830 	if (request >= d->dma_requests)
831 		return NULL;
832 
833 	return dma_get_slave_channel(&(d->chans[request].vc.chan));
834 }
835 
836 static int k3_dma_probe(struct platform_device *op)
837 {
838 	const struct k3dma_soc_data *soc_data;
839 	struct k3_dma_dev *d;
840 	const struct of_device_id *of_id;
841 	struct resource *iores;
842 	int i, ret, irq = 0;
843 
844 	iores = platform_get_resource(op, IORESOURCE_MEM, 0);
845 	if (!iores)
846 		return -EINVAL;
847 
848 	d = devm_kzalloc(&op->dev, sizeof(*d), GFP_KERNEL);
849 	if (!d)
850 		return -ENOMEM;
851 
852 	soc_data = device_get_match_data(&op->dev);
853 	if (!soc_data)
854 		return -EINVAL;
855 
856 	d->base = devm_ioremap_resource(&op->dev, iores);
857 	if (IS_ERR(d->base))
858 		return PTR_ERR(d->base);
859 
860 	of_id = of_match_device(k3_pdma_dt_ids, &op->dev);
861 	if (of_id) {
862 		of_property_read_u32((&op->dev)->of_node,
863 				"dma-channels", &d->dma_channels);
864 		of_property_read_u32((&op->dev)->of_node,
865 				"dma-requests", &d->dma_requests);
866 		ret = of_property_read_u32((&op->dev)->of_node,
867 				"dma-channel-mask", &d->dma_channel_mask);
868 		if (ret) {
869 			dev_warn(&op->dev,
870 				 "dma-channel-mask doesn't exist, considering all as available.\n");
871 			d->dma_channel_mask = (u32)~0UL;
872 		}
873 	}
874 
875 	if (!(soc_data->flags & K3_FLAG_NOCLK)) {
876 		d->clk = devm_clk_get(&op->dev, NULL);
877 		if (IS_ERR(d->clk)) {
878 			dev_err(&op->dev, "no dma clk\n");
879 			return PTR_ERR(d->clk);
880 		}
881 	}
882 
883 	irq = platform_get_irq(op, 0);
884 	ret = devm_request_irq(&op->dev, irq,
885 			k3_dma_int_handler, 0, DRIVER_NAME, d);
886 	if (ret)
887 		return ret;
888 
889 	d->irq = irq;
890 
891 	/* A DMA memory pool for LLIs, align on 32-byte boundary */
892 	d->pool = dmam_pool_create(DRIVER_NAME, &op->dev,
893 					LLI_BLOCK_SIZE, 32, 0);
894 	if (!d->pool)
895 		return -ENOMEM;
896 
897 	/* init phy channel */
898 	d->phy = devm_kcalloc(&op->dev,
899 		d->dma_channels, sizeof(struct k3_dma_phy), GFP_KERNEL);
900 	if (d->phy == NULL)
901 		return -ENOMEM;
902 
903 	for (i = 0; i < d->dma_channels; i++) {
904 		struct k3_dma_phy *p;
905 
906 		if (!(d->dma_channel_mask & BIT(i)))
907 			continue;
908 
909 		p = &d->phy[i];
910 		p->idx = i;
911 		p->base = d->base + i * 0x40;
912 	}
913 
914 	INIT_LIST_HEAD(&d->slave.channels);
915 	dma_cap_set(DMA_SLAVE, d->slave.cap_mask);
916 	dma_cap_set(DMA_MEMCPY, d->slave.cap_mask);
917 	dma_cap_set(DMA_CYCLIC, d->slave.cap_mask);
918 	d->slave.dev = &op->dev;
919 	d->slave.device_free_chan_resources = k3_dma_free_chan_resources;
920 	d->slave.device_tx_status = k3_dma_tx_status;
921 	d->slave.device_prep_dma_memcpy = k3_dma_prep_memcpy;
922 	d->slave.device_prep_slave_sg = k3_dma_prep_slave_sg;
923 	d->slave.device_prep_dma_cyclic = k3_dma_prep_dma_cyclic;
924 	d->slave.device_issue_pending = k3_dma_issue_pending;
925 	d->slave.device_config = k3_dma_config;
926 	d->slave.device_pause = k3_dma_transfer_pause;
927 	d->slave.device_resume = k3_dma_transfer_resume;
928 	d->slave.device_terminate_all = k3_dma_terminate_all;
929 	d->slave.device_synchronize = k3_dma_synchronize;
930 	d->slave.copy_align = DMAENGINE_ALIGN_8_BYTES;
931 
932 	/* init virtual channel */
933 	d->chans = devm_kcalloc(&op->dev,
934 		d->dma_requests, sizeof(struct k3_dma_chan), GFP_KERNEL);
935 	if (d->chans == NULL)
936 		return -ENOMEM;
937 
938 	for (i = 0; i < d->dma_requests; i++) {
939 		struct k3_dma_chan *c = &d->chans[i];
940 
941 		c->status = DMA_IN_PROGRESS;
942 		INIT_LIST_HEAD(&c->node);
943 		c->vc.desc_free = k3_dma_free_desc;
944 		vchan_init(&c->vc, &d->slave);
945 	}
946 
947 	/* Enable clock before accessing registers */
948 	ret = clk_prepare_enable(d->clk);
949 	if (ret < 0) {
950 		dev_err(&op->dev, "clk_prepare_enable failed: %d\n", ret);
951 		return ret;
952 	}
953 
954 	k3_dma_enable_dma(d, true);
955 
956 	ret = dma_async_device_register(&d->slave);
957 	if (ret)
958 		goto dma_async_register_fail;
959 
960 	ret = of_dma_controller_register((&op->dev)->of_node,
961 					k3_of_dma_simple_xlate, d);
962 	if (ret)
963 		goto of_dma_register_fail;
964 
965 	spin_lock_init(&d->lock);
966 	INIT_LIST_HEAD(&d->chan_pending);
967 	tasklet_init(&d->task, k3_dma_tasklet, (unsigned long)d);
968 	platform_set_drvdata(op, d);
969 	dev_info(&op->dev, "initialized\n");
970 
971 	return 0;
972 
973 of_dma_register_fail:
974 	dma_async_device_unregister(&d->slave);
975 dma_async_register_fail:
976 	clk_disable_unprepare(d->clk);
977 	return ret;
978 }
979 
980 static int k3_dma_remove(struct platform_device *op)
981 {
982 	struct k3_dma_chan *c, *cn;
983 	struct k3_dma_dev *d = platform_get_drvdata(op);
984 
985 	dma_async_device_unregister(&d->slave);
986 	of_dma_controller_free((&op->dev)->of_node);
987 
988 	devm_free_irq(&op->dev, d->irq, d);
989 
990 	list_for_each_entry_safe(c, cn, &d->slave.channels, vc.chan.device_node) {
991 		list_del(&c->vc.chan.device_node);
992 		tasklet_kill(&c->vc.task);
993 	}
994 	tasklet_kill(&d->task);
995 	clk_disable_unprepare(d->clk);
996 	return 0;
997 }
998 
999 #ifdef CONFIG_PM_SLEEP
1000 static int k3_dma_suspend_dev(struct device *dev)
1001 {
1002 	struct k3_dma_dev *d = dev_get_drvdata(dev);
1003 	u32 stat = 0;
1004 
1005 	stat = k3_dma_get_chan_stat(d);
1006 	if (stat) {
1007 		dev_warn(d->slave.dev,
1008 			"chan %d is running fail to suspend\n", stat);
1009 		return -1;
1010 	}
1011 	k3_dma_enable_dma(d, false);
1012 	clk_disable_unprepare(d->clk);
1013 	return 0;
1014 }
1015 
1016 static int k3_dma_resume_dev(struct device *dev)
1017 {
1018 	struct k3_dma_dev *d = dev_get_drvdata(dev);
1019 	int ret = 0;
1020 
1021 	ret = clk_prepare_enable(d->clk);
1022 	if (ret < 0) {
1023 		dev_err(d->slave.dev, "clk_prepare_enable failed: %d\n", ret);
1024 		return ret;
1025 	}
1026 	k3_dma_enable_dma(d, true);
1027 	return 0;
1028 }
1029 #endif
1030 
1031 static SIMPLE_DEV_PM_OPS(k3_dma_pmops, k3_dma_suspend_dev, k3_dma_resume_dev);
1032 
1033 static struct platform_driver k3_pdma_driver = {
1034 	.driver		= {
1035 		.name	= DRIVER_NAME,
1036 		.pm	= &k3_dma_pmops,
1037 		.of_match_table = k3_pdma_dt_ids,
1038 	},
1039 	.probe		= k3_dma_probe,
1040 	.remove		= k3_dma_remove,
1041 };
1042 
1043 module_platform_driver(k3_pdma_driver);
1044 
1045 MODULE_DESCRIPTION("Hisilicon k3 DMA Driver");
1046 MODULE_ALIAS("platform:k3dma");
1047 MODULE_LICENSE("GPL v2");
1048