xref: /openbmc/linux/drivers/dma/k3dma.c (revision aac5987a)
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 #define AXI_CFG			0x820
56 #define AXI_CFG_DEFAULT		0x201201
57 
58 #define CX_LLI_CHAIN_EN		0x2
59 #define CX_CFG_EN		0x1
60 #define CX_CFG_NODEIRQ		BIT(1)
61 #define CX_CFG_MEM2PER		(0x1 << 2)
62 #define CX_CFG_PER2MEM		(0x2 << 2)
63 #define CX_CFG_SRCINCR		(0x1 << 31)
64 #define CX_CFG_DSTINCR		(0x1 << 30)
65 
66 struct k3_desc_hw {
67 	u32 lli;
68 	u32 reserved[3];
69 	u32 count;
70 	u32 saddr;
71 	u32 daddr;
72 	u32 config;
73 } __aligned(32);
74 
75 struct k3_dma_desc_sw {
76 	struct virt_dma_desc	vd;
77 	dma_addr_t		desc_hw_lli;
78 	size_t			desc_num;
79 	size_t			size;
80 	struct k3_desc_hw	*desc_hw;
81 };
82 
83 struct k3_dma_phy;
84 
85 struct k3_dma_chan {
86 	u32			ccfg;
87 	struct virt_dma_chan	vc;
88 	struct k3_dma_phy	*phy;
89 	struct list_head	node;
90 	enum dma_transfer_direction dir;
91 	dma_addr_t		dev_addr;
92 	enum dma_status		status;
93 	bool			cyclic;
94 };
95 
96 struct k3_dma_phy {
97 	u32			idx;
98 	void __iomem		*base;
99 	struct k3_dma_chan	*vchan;
100 	struct k3_dma_desc_sw	*ds_run;
101 	struct k3_dma_desc_sw	*ds_done;
102 };
103 
104 struct k3_dma_dev {
105 	struct dma_device	slave;
106 	void __iomem		*base;
107 	struct tasklet_struct	task;
108 	spinlock_t		lock;
109 	struct list_head	chan_pending;
110 	struct k3_dma_phy	*phy;
111 	struct k3_dma_chan	*chans;
112 	struct clk		*clk;
113 	struct dma_pool		*pool;
114 	u32			dma_channels;
115 	u32			dma_requests;
116 	unsigned int		irq;
117 };
118 
119 #define to_k3_dma(dmadev) container_of(dmadev, struct k3_dma_dev, slave)
120 
121 static struct k3_dma_chan *to_k3_chan(struct dma_chan *chan)
122 {
123 	return container_of(chan, struct k3_dma_chan, vc.chan);
124 }
125 
126 static void k3_dma_pause_dma(struct k3_dma_phy *phy, bool on)
127 {
128 	u32 val = 0;
129 
130 	if (on) {
131 		val = readl_relaxed(phy->base + CX_CFG);
132 		val |= CX_CFG_EN;
133 		writel_relaxed(val, phy->base + CX_CFG);
134 	} else {
135 		val = readl_relaxed(phy->base + CX_CFG);
136 		val &= ~CX_CFG_EN;
137 		writel_relaxed(val, phy->base + CX_CFG);
138 	}
139 }
140 
141 static void k3_dma_terminate_chan(struct k3_dma_phy *phy, struct k3_dma_dev *d)
142 {
143 	u32 val = 0;
144 
145 	k3_dma_pause_dma(phy, false);
146 
147 	val = 0x1 << phy->idx;
148 	writel_relaxed(val, d->base + INT_TC1_RAW);
149 	writel_relaxed(val, d->base + INT_TC2_RAW);
150 	writel_relaxed(val, d->base + INT_ERR1_RAW);
151 	writel_relaxed(val, d->base + INT_ERR2_RAW);
152 }
153 
154 static void k3_dma_set_desc(struct k3_dma_phy *phy, struct k3_desc_hw *hw)
155 {
156 	writel_relaxed(hw->lli, phy->base + CX_LLI);
157 	writel_relaxed(hw->count, phy->base + CX_CNT0);
158 	writel_relaxed(hw->saddr, phy->base + CX_SRC);
159 	writel_relaxed(hw->daddr, phy->base + CX_DST);
160 	writel_relaxed(AXI_CFG_DEFAULT, phy->base + AXI_CFG);
161 	writel_relaxed(hw->config, phy->base + CX_CFG);
162 }
163 
164 static u32 k3_dma_get_curr_cnt(struct k3_dma_dev *d, struct k3_dma_phy *phy)
165 {
166 	u32 cnt = 0;
167 
168 	cnt = readl_relaxed(d->base + CX_CUR_CNT + phy->idx * 0x10);
169 	cnt &= 0xffff;
170 	return cnt;
171 }
172 
173 static u32 k3_dma_get_curr_lli(struct k3_dma_phy *phy)
174 {
175 	return readl_relaxed(phy->base + CX_LLI);
176 }
177 
178 static u32 k3_dma_get_chan_stat(struct k3_dma_dev *d)
179 {
180 	return readl_relaxed(d->base + CH_STAT);
181 }
182 
183 static void k3_dma_enable_dma(struct k3_dma_dev *d, bool on)
184 {
185 	if (on) {
186 		/* set same priority */
187 		writel_relaxed(0x0, d->base + CH_PRI);
188 
189 		/* unmask irq */
190 		writel_relaxed(0xffff, d->base + INT_TC1_MASK);
191 		writel_relaxed(0xffff, d->base + INT_TC2_MASK);
192 		writel_relaxed(0xffff, d->base + INT_ERR1_MASK);
193 		writel_relaxed(0xffff, d->base + INT_ERR2_MASK);
194 	} else {
195 		/* mask irq */
196 		writel_relaxed(0x0, d->base + INT_TC1_MASK);
197 		writel_relaxed(0x0, d->base + INT_TC2_MASK);
198 		writel_relaxed(0x0, d->base + INT_ERR1_MASK);
199 		writel_relaxed(0x0, d->base + INT_ERR2_MASK);
200 	}
201 }
202 
203 static irqreturn_t k3_dma_int_handler(int irq, void *dev_id)
204 {
205 	struct k3_dma_dev *d = (struct k3_dma_dev *)dev_id;
206 	struct k3_dma_phy *p;
207 	struct k3_dma_chan *c;
208 	u32 stat = readl_relaxed(d->base + INT_STAT);
209 	u32 tc1  = readl_relaxed(d->base + INT_TC1);
210 	u32 tc2  = readl_relaxed(d->base + INT_TC2);
211 	u32 err1 = readl_relaxed(d->base + INT_ERR1);
212 	u32 err2 = readl_relaxed(d->base + INT_ERR2);
213 	u32 i, irq_chan = 0;
214 
215 	while (stat) {
216 		i = __ffs(stat);
217 		stat &= ~BIT(i);
218 		if (likely(tc1 & BIT(i)) || (tc2 & BIT(i))) {
219 			unsigned long flags;
220 
221 			p = &d->phy[i];
222 			c = p->vchan;
223 			if (c && (tc1 & BIT(i))) {
224 				spin_lock_irqsave(&c->vc.lock, flags);
225 				vchan_cookie_complete(&p->ds_run->vd);
226 				WARN_ON_ONCE(p->ds_done);
227 				p->ds_done = p->ds_run;
228 				p->ds_run = NULL;
229 				spin_unlock_irqrestore(&c->vc.lock, flags);
230 			}
231 			if (c && (tc2 & BIT(i))) {
232 				spin_lock_irqsave(&c->vc.lock, flags);
233 				if (p->ds_run != NULL)
234 					vchan_cyclic_callback(&p->ds_run->vd);
235 				spin_unlock_irqrestore(&c->vc.lock, flags);
236 			}
237 			irq_chan |= BIT(i);
238 		}
239 		if (unlikely((err1 & BIT(i)) || (err2 & BIT(i))))
240 			dev_warn(d->slave.dev, "DMA ERR\n");
241 	}
242 
243 	writel_relaxed(irq_chan, d->base + INT_TC1_RAW);
244 	writel_relaxed(irq_chan, d->base + INT_TC2_RAW);
245 	writel_relaxed(err1, d->base + INT_ERR1_RAW);
246 	writel_relaxed(err2, d->base + INT_ERR2_RAW);
247 
248 	if (irq_chan)
249 		tasklet_schedule(&d->task);
250 
251 	if (irq_chan || err1 || err2)
252 		return IRQ_HANDLED;
253 
254 	return IRQ_NONE;
255 }
256 
257 static int k3_dma_start_txd(struct k3_dma_chan *c)
258 {
259 	struct k3_dma_dev *d = to_k3_dma(c->vc.chan.device);
260 	struct virt_dma_desc *vd = vchan_next_desc(&c->vc);
261 
262 	if (!c->phy)
263 		return -EAGAIN;
264 
265 	if (BIT(c->phy->idx) & k3_dma_get_chan_stat(d))
266 		return -EAGAIN;
267 
268 	if (vd) {
269 		struct k3_dma_desc_sw *ds =
270 			container_of(vd, struct k3_dma_desc_sw, vd);
271 		/*
272 		 * fetch and remove request from vc->desc_issued
273 		 * so vc->desc_issued only contains desc pending
274 		 */
275 		list_del(&ds->vd.node);
276 
277 		WARN_ON_ONCE(c->phy->ds_run);
278 		WARN_ON_ONCE(c->phy->ds_done);
279 		c->phy->ds_run = ds;
280 		/* start dma */
281 		k3_dma_set_desc(c->phy, &ds->desc_hw[0]);
282 		return 0;
283 	}
284 	return -EAGAIN;
285 }
286 
287 static void k3_dma_tasklet(unsigned long arg)
288 {
289 	struct k3_dma_dev *d = (struct k3_dma_dev *)arg;
290 	struct k3_dma_phy *p;
291 	struct k3_dma_chan *c, *cn;
292 	unsigned pch, pch_alloc = 0;
293 
294 	/* check new dma request of running channel in vc->desc_issued */
295 	list_for_each_entry_safe(c, cn, &d->slave.channels, vc.chan.device_node) {
296 		spin_lock_irq(&c->vc.lock);
297 		p = c->phy;
298 		if (p && p->ds_done) {
299 			if (k3_dma_start_txd(c)) {
300 				/* No current txd associated with this channel */
301 				dev_dbg(d->slave.dev, "pchan %u: free\n", p->idx);
302 				/* Mark this channel free */
303 				c->phy = NULL;
304 				p->vchan = NULL;
305 			}
306 		}
307 		spin_unlock_irq(&c->vc.lock);
308 	}
309 
310 	/* check new channel request in d->chan_pending */
311 	spin_lock_irq(&d->lock);
312 	for (pch = 0; pch < d->dma_channels; pch++) {
313 		p = &d->phy[pch];
314 
315 		if (p->vchan == NULL && !list_empty(&d->chan_pending)) {
316 			c = list_first_entry(&d->chan_pending,
317 				struct k3_dma_chan, node);
318 			/* remove from d->chan_pending */
319 			list_del_init(&c->node);
320 			pch_alloc |= 1 << pch;
321 			/* Mark this channel allocated */
322 			p->vchan = c;
323 			c->phy = p;
324 			dev_dbg(d->slave.dev, "pchan %u: alloc vchan %p\n", pch, &c->vc);
325 		}
326 	}
327 	spin_unlock_irq(&d->lock);
328 
329 	for (pch = 0; pch < d->dma_channels; pch++) {
330 		if (pch_alloc & (1 << pch)) {
331 			p = &d->phy[pch];
332 			c = p->vchan;
333 			if (c) {
334 				spin_lock_irq(&c->vc.lock);
335 				k3_dma_start_txd(c);
336 				spin_unlock_irq(&c->vc.lock);
337 			}
338 		}
339 	}
340 }
341 
342 static void k3_dma_free_chan_resources(struct dma_chan *chan)
343 {
344 	struct k3_dma_chan *c = to_k3_chan(chan);
345 	struct k3_dma_dev *d = to_k3_dma(chan->device);
346 	unsigned long flags;
347 
348 	spin_lock_irqsave(&d->lock, flags);
349 	list_del_init(&c->node);
350 	spin_unlock_irqrestore(&d->lock, flags);
351 
352 	vchan_free_chan_resources(&c->vc);
353 	c->ccfg = 0;
354 }
355 
356 static enum dma_status k3_dma_tx_status(struct dma_chan *chan,
357 	dma_cookie_t cookie, struct dma_tx_state *state)
358 {
359 	struct k3_dma_chan *c = to_k3_chan(chan);
360 	struct k3_dma_dev *d = to_k3_dma(chan->device);
361 	struct k3_dma_phy *p;
362 	struct virt_dma_desc *vd;
363 	unsigned long flags;
364 	enum dma_status ret;
365 	size_t bytes = 0;
366 
367 	ret = dma_cookie_status(&c->vc.chan, cookie, state);
368 	if (ret == DMA_COMPLETE)
369 		return ret;
370 
371 	spin_lock_irqsave(&c->vc.lock, flags);
372 	p = c->phy;
373 	ret = c->status;
374 
375 	/*
376 	 * If the cookie is on our issue queue, then the residue is
377 	 * its total size.
378 	 */
379 	vd = vchan_find_desc(&c->vc, cookie);
380 	if (vd && !c->cyclic) {
381 		bytes = container_of(vd, struct k3_dma_desc_sw, vd)->size;
382 	} else if ((!p) || (!p->ds_run)) {
383 		bytes = 0;
384 	} else {
385 		struct k3_dma_desc_sw *ds = p->ds_run;
386 		u32 clli = 0, index = 0;
387 
388 		bytes = k3_dma_get_curr_cnt(d, p);
389 		clli = k3_dma_get_curr_lli(p);
390 		index = ((clli - ds->desc_hw_lli) /
391 				sizeof(struct k3_desc_hw)) + 1;
392 		for (; index < ds->desc_num; index++) {
393 			bytes += ds->desc_hw[index].count;
394 			/* end of lli */
395 			if (!ds->desc_hw[index].lli)
396 				break;
397 		}
398 	}
399 	spin_unlock_irqrestore(&c->vc.lock, flags);
400 	dma_set_residue(state, bytes);
401 	return ret;
402 }
403 
404 static void k3_dma_issue_pending(struct dma_chan *chan)
405 {
406 	struct k3_dma_chan *c = to_k3_chan(chan);
407 	struct k3_dma_dev *d = to_k3_dma(chan->device);
408 	unsigned long flags;
409 
410 	spin_lock_irqsave(&c->vc.lock, flags);
411 	/* add request to vc->desc_issued */
412 	if (vchan_issue_pending(&c->vc)) {
413 		spin_lock(&d->lock);
414 		if (!c->phy) {
415 			if (list_empty(&c->node)) {
416 				/* if new channel, add chan_pending */
417 				list_add_tail(&c->node, &d->chan_pending);
418 				/* check in tasklet */
419 				tasklet_schedule(&d->task);
420 				dev_dbg(d->slave.dev, "vchan %p: issued\n", &c->vc);
421 			}
422 		}
423 		spin_unlock(&d->lock);
424 	} else
425 		dev_dbg(d->slave.dev, "vchan %p: nothing to issue\n", &c->vc);
426 	spin_unlock_irqrestore(&c->vc.lock, flags);
427 }
428 
429 static void k3_dma_fill_desc(struct k3_dma_desc_sw *ds, dma_addr_t dst,
430 			dma_addr_t src, size_t len, u32 num, u32 ccfg)
431 {
432 	if (num != ds->desc_num - 1)
433 		ds->desc_hw[num].lli = ds->desc_hw_lli + (num + 1) *
434 			sizeof(struct k3_desc_hw);
435 
436 	ds->desc_hw[num].lli |= CX_LLI_CHAIN_EN;
437 	ds->desc_hw[num].count = len;
438 	ds->desc_hw[num].saddr = src;
439 	ds->desc_hw[num].daddr = dst;
440 	ds->desc_hw[num].config = ccfg;
441 }
442 
443 static struct k3_dma_desc_sw *k3_dma_alloc_desc_resource(int num,
444 							struct dma_chan *chan)
445 {
446 	struct k3_dma_chan *c = to_k3_chan(chan);
447 	struct k3_dma_desc_sw *ds;
448 	struct k3_dma_dev *d = to_k3_dma(chan->device);
449 	int lli_limit = LLI_BLOCK_SIZE / sizeof(struct k3_desc_hw);
450 
451 	if (num > lli_limit) {
452 		dev_dbg(chan->device->dev, "vch %p: sg num %d exceed max %d\n",
453 			&c->vc, num, lli_limit);
454 		return NULL;
455 	}
456 
457 	ds = kzalloc(sizeof(*ds), GFP_NOWAIT);
458 	if (!ds)
459 		return NULL;
460 
461 	ds->desc_hw = dma_pool_zalloc(d->pool, GFP_NOWAIT, &ds->desc_hw_lli);
462 	if (!ds->desc_hw) {
463 		dev_dbg(chan->device->dev, "vch %p: dma alloc fail\n", &c->vc);
464 		kfree(ds);
465 		return NULL;
466 	}
467 	ds->desc_num = num;
468 	return ds;
469 }
470 
471 static struct dma_async_tx_descriptor *k3_dma_prep_memcpy(
472 	struct dma_chan *chan,	dma_addr_t dst, dma_addr_t src,
473 	size_t len, unsigned long flags)
474 {
475 	struct k3_dma_chan *c = to_k3_chan(chan);
476 	struct k3_dma_desc_sw *ds;
477 	size_t copy = 0;
478 	int num = 0;
479 
480 	if (!len)
481 		return NULL;
482 
483 	num = DIV_ROUND_UP(len, DMA_MAX_SIZE);
484 
485 	ds = k3_dma_alloc_desc_resource(num, chan);
486 	if (!ds)
487 		return NULL;
488 
489 	c->cyclic = 0;
490 	ds->size = len;
491 	num = 0;
492 
493 	if (!c->ccfg) {
494 		/* default is memtomem, without calling device_config */
495 		c->ccfg = CX_CFG_SRCINCR | CX_CFG_DSTINCR | CX_CFG_EN;
496 		c->ccfg |= (0xf << 20) | (0xf << 24);	/* burst = 16 */
497 		c->ccfg |= (0x3 << 12) | (0x3 << 16);	/* width = 64 bit */
498 	}
499 
500 	do {
501 		copy = min_t(size_t, len, DMA_MAX_SIZE);
502 		k3_dma_fill_desc(ds, dst, src, copy, num++, c->ccfg);
503 
504 		if (c->dir == DMA_MEM_TO_DEV) {
505 			src += copy;
506 		} else if (c->dir == DMA_DEV_TO_MEM) {
507 			dst += copy;
508 		} else {
509 			src += copy;
510 			dst += copy;
511 		}
512 		len -= copy;
513 	} while (len);
514 
515 	ds->desc_hw[num-1].lli = 0;	/* end of link */
516 	return vchan_tx_prep(&c->vc, &ds->vd, flags);
517 }
518 
519 static struct dma_async_tx_descriptor *k3_dma_prep_slave_sg(
520 	struct dma_chan *chan, struct scatterlist *sgl, unsigned int sglen,
521 	enum dma_transfer_direction dir, unsigned long flags, void *context)
522 {
523 	struct k3_dma_chan *c = to_k3_chan(chan);
524 	struct k3_dma_desc_sw *ds;
525 	size_t len, avail, total = 0;
526 	struct scatterlist *sg;
527 	dma_addr_t addr, src = 0, dst = 0;
528 	int num = sglen, i;
529 
530 	if (sgl == NULL)
531 		return NULL;
532 
533 	c->cyclic = 0;
534 
535 	for_each_sg(sgl, sg, sglen, i) {
536 		avail = sg_dma_len(sg);
537 		if (avail > DMA_MAX_SIZE)
538 			num += DIV_ROUND_UP(avail, DMA_MAX_SIZE) - 1;
539 	}
540 
541 	ds = k3_dma_alloc_desc_resource(num, chan);
542 	if (!ds)
543 		return NULL;
544 	num = 0;
545 
546 	for_each_sg(sgl, sg, sglen, i) {
547 		addr = sg_dma_address(sg);
548 		avail = sg_dma_len(sg);
549 		total += avail;
550 
551 		do {
552 			len = min_t(size_t, avail, DMA_MAX_SIZE);
553 
554 			if (dir == DMA_MEM_TO_DEV) {
555 				src = addr;
556 				dst = c->dev_addr;
557 			} else if (dir == DMA_DEV_TO_MEM) {
558 				src = c->dev_addr;
559 				dst = addr;
560 			}
561 
562 			k3_dma_fill_desc(ds, dst, src, len, num++, c->ccfg);
563 
564 			addr += len;
565 			avail -= len;
566 		} while (avail);
567 	}
568 
569 	ds->desc_hw[num-1].lli = 0;	/* end of link */
570 	ds->size = total;
571 	return vchan_tx_prep(&c->vc, &ds->vd, flags);
572 }
573 
574 static struct dma_async_tx_descriptor *
575 k3_dma_prep_dma_cyclic(struct dma_chan *chan, dma_addr_t buf_addr,
576 		       size_t buf_len, size_t period_len,
577 		       enum dma_transfer_direction dir,
578 		       unsigned long flags)
579 {
580 	struct k3_dma_chan *c = to_k3_chan(chan);
581 	struct k3_dma_desc_sw *ds;
582 	size_t len, avail, total = 0;
583 	dma_addr_t addr, src = 0, dst = 0;
584 	int num = 1, since = 0;
585 	size_t modulo = DMA_CYCLIC_MAX_PERIOD;
586 	u32 en_tc2 = 0;
587 
588 	dev_dbg(chan->device->dev, "%s: buf %pad, dst %pad, buf len %zu, period_len = %zu, dir %d\n",
589 	       __func__, &buf_addr, &to_k3_chan(chan)->dev_addr,
590 	       buf_len, period_len, (int)dir);
591 
592 	avail = buf_len;
593 	if (avail > modulo)
594 		num += DIV_ROUND_UP(avail, modulo) - 1;
595 
596 	ds = k3_dma_alloc_desc_resource(num, chan);
597 	if (!ds)
598 		return NULL;
599 
600 	c->cyclic = 1;
601 	addr = buf_addr;
602 	avail = buf_len;
603 	total = avail;
604 	num = 0;
605 
606 	if (period_len < modulo)
607 		modulo = period_len;
608 
609 	do {
610 		len = min_t(size_t, avail, modulo);
611 
612 		if (dir == DMA_MEM_TO_DEV) {
613 			src = addr;
614 			dst = c->dev_addr;
615 		} else if (dir == DMA_DEV_TO_MEM) {
616 			src = c->dev_addr;
617 			dst = addr;
618 		}
619 		since += len;
620 		if (since >= period_len) {
621 			/* descriptor asks for TC2 interrupt on completion */
622 			en_tc2 = CX_CFG_NODEIRQ;
623 			since -= period_len;
624 		} else
625 			en_tc2 = 0;
626 
627 		k3_dma_fill_desc(ds, dst, src, len, num++, c->ccfg | en_tc2);
628 
629 		addr += len;
630 		avail -= len;
631 	} while (avail);
632 
633 	/* "Cyclic" == end of link points back to start of link */
634 	ds->desc_hw[num - 1].lli |= ds->desc_hw_lli;
635 
636 	ds->size = total;
637 
638 	return vchan_tx_prep(&c->vc, &ds->vd, flags);
639 }
640 
641 static int k3_dma_config(struct dma_chan *chan,
642 			 struct dma_slave_config *cfg)
643 {
644 	struct k3_dma_chan *c = to_k3_chan(chan);
645 	u32 maxburst = 0, val = 0;
646 	enum dma_slave_buswidth width = DMA_SLAVE_BUSWIDTH_UNDEFINED;
647 
648 	if (cfg == NULL)
649 		return -EINVAL;
650 	c->dir = cfg->direction;
651 	if (c->dir == DMA_DEV_TO_MEM) {
652 		c->ccfg = CX_CFG_DSTINCR;
653 		c->dev_addr = cfg->src_addr;
654 		maxburst = cfg->src_maxburst;
655 		width = cfg->src_addr_width;
656 	} else if (c->dir == DMA_MEM_TO_DEV) {
657 		c->ccfg = CX_CFG_SRCINCR;
658 		c->dev_addr = cfg->dst_addr;
659 		maxburst = cfg->dst_maxburst;
660 		width = cfg->dst_addr_width;
661 	}
662 	switch (width) {
663 	case DMA_SLAVE_BUSWIDTH_1_BYTE:
664 	case DMA_SLAVE_BUSWIDTH_2_BYTES:
665 	case DMA_SLAVE_BUSWIDTH_4_BYTES:
666 	case DMA_SLAVE_BUSWIDTH_8_BYTES:
667 		val =  __ffs(width);
668 		break;
669 	default:
670 		val = 3;
671 		break;
672 	}
673 	c->ccfg |= (val << 12) | (val << 16);
674 
675 	if ((maxburst == 0) || (maxburst > 16))
676 		val = 15;
677 	else
678 		val = maxburst - 1;
679 	c->ccfg |= (val << 20) | (val << 24);
680 	c->ccfg |= CX_CFG_MEM2PER | CX_CFG_EN;
681 
682 	/* specific request line */
683 	c->ccfg |= c->vc.chan.chan_id << 4;
684 
685 	return 0;
686 }
687 
688 static void k3_dma_free_desc(struct virt_dma_desc *vd)
689 {
690 	struct k3_dma_desc_sw *ds =
691 		container_of(vd, struct k3_dma_desc_sw, vd);
692 	struct k3_dma_dev *d = to_k3_dma(vd->tx.chan->device);
693 
694 	dma_pool_free(d->pool, ds->desc_hw, ds->desc_hw_lli);
695 	kfree(ds);
696 }
697 
698 static int k3_dma_terminate_all(struct dma_chan *chan)
699 {
700 	struct k3_dma_chan *c = to_k3_chan(chan);
701 	struct k3_dma_dev *d = to_k3_dma(chan->device);
702 	struct k3_dma_phy *p = c->phy;
703 	unsigned long flags;
704 	LIST_HEAD(head);
705 
706 	dev_dbg(d->slave.dev, "vchan %p: terminate all\n", &c->vc);
707 
708 	/* Prevent this channel being scheduled */
709 	spin_lock(&d->lock);
710 	list_del_init(&c->node);
711 	spin_unlock(&d->lock);
712 
713 	/* Clear the tx descriptor lists */
714 	spin_lock_irqsave(&c->vc.lock, flags);
715 	vchan_get_all_descriptors(&c->vc, &head);
716 	if (p) {
717 		/* vchan is assigned to a pchan - stop the channel */
718 		k3_dma_terminate_chan(p, d);
719 		c->phy = NULL;
720 		p->vchan = NULL;
721 		if (p->ds_run) {
722 			k3_dma_free_desc(&p->ds_run->vd);
723 			p->ds_run = NULL;
724 		}
725 		if (p->ds_done) {
726 			k3_dma_free_desc(&p->ds_done->vd);
727 			p->ds_done = NULL;
728 		}
729 
730 	}
731 	spin_unlock_irqrestore(&c->vc.lock, flags);
732 	vchan_dma_desc_free_list(&c->vc, &head);
733 
734 	return 0;
735 }
736 
737 static int k3_dma_transfer_pause(struct dma_chan *chan)
738 {
739 	struct k3_dma_chan *c = to_k3_chan(chan);
740 	struct k3_dma_dev *d = to_k3_dma(chan->device);
741 	struct k3_dma_phy *p = c->phy;
742 
743 	dev_dbg(d->slave.dev, "vchan %p: pause\n", &c->vc);
744 	if (c->status == DMA_IN_PROGRESS) {
745 		c->status = DMA_PAUSED;
746 		if (p) {
747 			k3_dma_pause_dma(p, false);
748 		} else {
749 			spin_lock(&d->lock);
750 			list_del_init(&c->node);
751 			spin_unlock(&d->lock);
752 		}
753 	}
754 
755 	return 0;
756 }
757 
758 static int k3_dma_transfer_resume(struct dma_chan *chan)
759 {
760 	struct k3_dma_chan *c = to_k3_chan(chan);
761 	struct k3_dma_dev *d = to_k3_dma(chan->device);
762 	struct k3_dma_phy *p = c->phy;
763 	unsigned long flags;
764 
765 	dev_dbg(d->slave.dev, "vchan %p: resume\n", &c->vc);
766 	spin_lock_irqsave(&c->vc.lock, flags);
767 	if (c->status == DMA_PAUSED) {
768 		c->status = DMA_IN_PROGRESS;
769 		if (p) {
770 			k3_dma_pause_dma(p, true);
771 		} else if (!list_empty(&c->vc.desc_issued)) {
772 			spin_lock(&d->lock);
773 			list_add_tail(&c->node, &d->chan_pending);
774 			spin_unlock(&d->lock);
775 		}
776 	}
777 	spin_unlock_irqrestore(&c->vc.lock, flags);
778 
779 	return 0;
780 }
781 
782 static const struct of_device_id k3_pdma_dt_ids[] = {
783 	{ .compatible = "hisilicon,k3-dma-1.0", },
784 	{}
785 };
786 MODULE_DEVICE_TABLE(of, k3_pdma_dt_ids);
787 
788 static struct dma_chan *k3_of_dma_simple_xlate(struct of_phandle_args *dma_spec,
789 						struct of_dma *ofdma)
790 {
791 	struct k3_dma_dev *d = ofdma->of_dma_data;
792 	unsigned int request = dma_spec->args[0];
793 
794 	if (request > d->dma_requests)
795 		return NULL;
796 
797 	return dma_get_slave_channel(&(d->chans[request].vc.chan));
798 }
799 
800 static int k3_dma_probe(struct platform_device *op)
801 {
802 	struct k3_dma_dev *d;
803 	const struct of_device_id *of_id;
804 	struct resource *iores;
805 	int i, ret, irq = 0;
806 
807 	iores = platform_get_resource(op, IORESOURCE_MEM, 0);
808 	if (!iores)
809 		return -EINVAL;
810 
811 	d = devm_kzalloc(&op->dev, sizeof(*d), GFP_KERNEL);
812 	if (!d)
813 		return -ENOMEM;
814 
815 	d->base = devm_ioremap_resource(&op->dev, iores);
816 	if (IS_ERR(d->base))
817 		return PTR_ERR(d->base);
818 
819 	of_id = of_match_device(k3_pdma_dt_ids, &op->dev);
820 	if (of_id) {
821 		of_property_read_u32((&op->dev)->of_node,
822 				"dma-channels", &d->dma_channels);
823 		of_property_read_u32((&op->dev)->of_node,
824 				"dma-requests", &d->dma_requests);
825 	}
826 
827 	d->clk = devm_clk_get(&op->dev, NULL);
828 	if (IS_ERR(d->clk)) {
829 		dev_err(&op->dev, "no dma clk\n");
830 		return PTR_ERR(d->clk);
831 	}
832 
833 	irq = platform_get_irq(op, 0);
834 	ret = devm_request_irq(&op->dev, irq,
835 			k3_dma_int_handler, 0, DRIVER_NAME, d);
836 	if (ret)
837 		return ret;
838 
839 	d->irq = irq;
840 
841 	/* A DMA memory pool for LLIs, align on 32-byte boundary */
842 	d->pool = dmam_pool_create(DRIVER_NAME, &op->dev,
843 					LLI_BLOCK_SIZE, 32, 0);
844 	if (!d->pool)
845 		return -ENOMEM;
846 
847 	/* init phy channel */
848 	d->phy = devm_kzalloc(&op->dev,
849 		d->dma_channels * sizeof(struct k3_dma_phy), GFP_KERNEL);
850 	if (d->phy == NULL)
851 		return -ENOMEM;
852 
853 	for (i = 0; i < d->dma_channels; i++) {
854 		struct k3_dma_phy *p = &d->phy[i];
855 
856 		p->idx = i;
857 		p->base = d->base + i * 0x40;
858 	}
859 
860 	INIT_LIST_HEAD(&d->slave.channels);
861 	dma_cap_set(DMA_SLAVE, d->slave.cap_mask);
862 	dma_cap_set(DMA_MEMCPY, d->slave.cap_mask);
863 	dma_cap_set(DMA_CYCLIC, d->slave.cap_mask);
864 	d->slave.dev = &op->dev;
865 	d->slave.device_free_chan_resources = k3_dma_free_chan_resources;
866 	d->slave.device_tx_status = k3_dma_tx_status;
867 	d->slave.device_prep_dma_memcpy = k3_dma_prep_memcpy;
868 	d->slave.device_prep_slave_sg = k3_dma_prep_slave_sg;
869 	d->slave.device_prep_dma_cyclic = k3_dma_prep_dma_cyclic;
870 	d->slave.device_issue_pending = k3_dma_issue_pending;
871 	d->slave.device_config = k3_dma_config;
872 	d->slave.device_pause = k3_dma_transfer_pause;
873 	d->slave.device_resume = k3_dma_transfer_resume;
874 	d->slave.device_terminate_all = k3_dma_terminate_all;
875 	d->slave.copy_align = DMAENGINE_ALIGN_8_BYTES;
876 
877 	/* init virtual channel */
878 	d->chans = devm_kzalloc(&op->dev,
879 		d->dma_requests * sizeof(struct k3_dma_chan), GFP_KERNEL);
880 	if (d->chans == NULL)
881 		return -ENOMEM;
882 
883 	for (i = 0; i < d->dma_requests; i++) {
884 		struct k3_dma_chan *c = &d->chans[i];
885 
886 		c->status = DMA_IN_PROGRESS;
887 		INIT_LIST_HEAD(&c->node);
888 		c->vc.desc_free = k3_dma_free_desc;
889 		vchan_init(&c->vc, &d->slave);
890 	}
891 
892 	/* Enable clock before accessing registers */
893 	ret = clk_prepare_enable(d->clk);
894 	if (ret < 0) {
895 		dev_err(&op->dev, "clk_prepare_enable failed: %d\n", ret);
896 		return ret;
897 	}
898 
899 	k3_dma_enable_dma(d, true);
900 
901 	ret = dma_async_device_register(&d->slave);
902 	if (ret)
903 		goto dma_async_register_fail;
904 
905 	ret = of_dma_controller_register((&op->dev)->of_node,
906 					k3_of_dma_simple_xlate, d);
907 	if (ret)
908 		goto of_dma_register_fail;
909 
910 	spin_lock_init(&d->lock);
911 	INIT_LIST_HEAD(&d->chan_pending);
912 	tasklet_init(&d->task, k3_dma_tasklet, (unsigned long)d);
913 	platform_set_drvdata(op, d);
914 	dev_info(&op->dev, "initialized\n");
915 
916 	return 0;
917 
918 of_dma_register_fail:
919 	dma_async_device_unregister(&d->slave);
920 dma_async_register_fail:
921 	clk_disable_unprepare(d->clk);
922 	return ret;
923 }
924 
925 static int k3_dma_remove(struct platform_device *op)
926 {
927 	struct k3_dma_chan *c, *cn;
928 	struct k3_dma_dev *d = platform_get_drvdata(op);
929 
930 	dma_async_device_unregister(&d->slave);
931 	of_dma_controller_free((&op->dev)->of_node);
932 
933 	devm_free_irq(&op->dev, d->irq, d);
934 
935 	list_for_each_entry_safe(c, cn, &d->slave.channels, vc.chan.device_node) {
936 		list_del(&c->vc.chan.device_node);
937 		tasklet_kill(&c->vc.task);
938 	}
939 	tasklet_kill(&d->task);
940 	clk_disable_unprepare(d->clk);
941 	return 0;
942 }
943 
944 #ifdef CONFIG_PM_SLEEP
945 static int k3_dma_suspend_dev(struct device *dev)
946 {
947 	struct k3_dma_dev *d = dev_get_drvdata(dev);
948 	u32 stat = 0;
949 
950 	stat = k3_dma_get_chan_stat(d);
951 	if (stat) {
952 		dev_warn(d->slave.dev,
953 			"chan %d is running fail to suspend\n", stat);
954 		return -1;
955 	}
956 	k3_dma_enable_dma(d, false);
957 	clk_disable_unprepare(d->clk);
958 	return 0;
959 }
960 
961 static int k3_dma_resume_dev(struct device *dev)
962 {
963 	struct k3_dma_dev *d = dev_get_drvdata(dev);
964 	int ret = 0;
965 
966 	ret = clk_prepare_enable(d->clk);
967 	if (ret < 0) {
968 		dev_err(d->slave.dev, "clk_prepare_enable failed: %d\n", ret);
969 		return ret;
970 	}
971 	k3_dma_enable_dma(d, true);
972 	return 0;
973 }
974 #endif
975 
976 static SIMPLE_DEV_PM_OPS(k3_dma_pmops, k3_dma_suspend_dev, k3_dma_resume_dev);
977 
978 static struct platform_driver k3_pdma_driver = {
979 	.driver		= {
980 		.name	= DRIVER_NAME,
981 		.pm	= &k3_dma_pmops,
982 		.of_match_table = k3_pdma_dt_ids,
983 	},
984 	.probe		= k3_dma_probe,
985 	.remove		= k3_dma_remove,
986 };
987 
988 module_platform_driver(k3_pdma_driver);
989 
990 MODULE_DESCRIPTION("Hisilicon k3 DMA Driver");
991 MODULE_ALIAS("platform:k3dma");
992 MODULE_LICENSE("GPL v2");
993