1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * Copyright (C) 2013-2014 Allwinner Tech Co., Ltd
4 * Author: Sugar <shuge@allwinnertech.com>
5 *
6 * Copyright (C) 2014 Maxime Ripard
7 * Maxime Ripard <maxime.ripard@free-electrons.com>
8 */
9
10#include <linux/clk.h>
11#include <linux/delay.h>
12#include <linux/dma-mapping.h>
13#include <linux/dmaengine.h>
14#include <linux/dmapool.h>
15#include <linux/interrupt.h>
16#include <linux/module.h>
17#include <linux/of.h>
18#include <linux/of_dma.h>
19#include <linux/platform_device.h>
20#include <linux/reset.h>
21#include <linux/slab.h>
22#include <linux/types.h>
23
24#include "virt-dma.h"
25
26/*
27 * Common registers
28 */
29#define DMA_IRQ_EN(x)		((x) * 0x04)
30#define DMA_IRQ_HALF			BIT(0)
31#define DMA_IRQ_PKG			BIT(1)
32#define DMA_IRQ_QUEUE			BIT(2)
33
34#define DMA_IRQ_CHAN_NR			8
35#define DMA_IRQ_CHAN_WIDTH		4
36
37
38#define DMA_IRQ_STAT(x)		((x) * 0x04 + 0x10)
39
40#define DMA_STAT		0x30
41
42/* Offset between DMA_IRQ_EN and DMA_IRQ_STAT limits number of channels */
43#define DMA_MAX_CHANNELS	(DMA_IRQ_CHAN_NR * 0x10 / 4)
44
45/*
46 * sun8i specific registers
47 */
48#define SUN8I_DMA_GATE		0x20
49#define SUN8I_DMA_GATE_ENABLE	0x4
50
51#define SUNXI_H3_SECURE_REG		0x20
52#define SUNXI_H3_DMA_GATE		0x28
53#define SUNXI_H3_DMA_GATE_ENABLE	0x4
54/*
55 * Channels specific registers
56 */
57#define DMA_CHAN_ENABLE		0x00
58#define DMA_CHAN_ENABLE_START		BIT(0)
59#define DMA_CHAN_ENABLE_STOP		0
60
61#define DMA_CHAN_PAUSE		0x04
62#define DMA_CHAN_PAUSE_PAUSE		BIT(1)
63#define DMA_CHAN_PAUSE_RESUME		0
64
65#define DMA_CHAN_LLI_ADDR	0x08
66
67#define DMA_CHAN_CUR_CFG	0x0c
68#define DMA_CHAN_MAX_DRQ_A31		0x1f
69#define DMA_CHAN_MAX_DRQ_H6		0x3f
70#define DMA_CHAN_CFG_SRC_DRQ_A31(x)	((x) & DMA_CHAN_MAX_DRQ_A31)
71#define DMA_CHAN_CFG_SRC_DRQ_H6(x)	((x) & DMA_CHAN_MAX_DRQ_H6)
72#define DMA_CHAN_CFG_SRC_MODE_A31(x)	(((x) & 0x1) << 5)
73#define DMA_CHAN_CFG_SRC_MODE_H6(x)	(((x) & 0x1) << 8)
74#define DMA_CHAN_CFG_SRC_BURST_A31(x)	(((x) & 0x3) << 7)
75#define DMA_CHAN_CFG_SRC_BURST_H3(x)	(((x) & 0x3) << 6)
76#define DMA_CHAN_CFG_SRC_WIDTH(x)	(((x) & 0x3) << 9)
77
78#define DMA_CHAN_CFG_DST_DRQ_A31(x)	(DMA_CHAN_CFG_SRC_DRQ_A31(x) << 16)
79#define DMA_CHAN_CFG_DST_DRQ_H6(x)	(DMA_CHAN_CFG_SRC_DRQ_H6(x) << 16)
80#define DMA_CHAN_CFG_DST_MODE_A31(x)	(DMA_CHAN_CFG_SRC_MODE_A31(x) << 16)
81#define DMA_CHAN_CFG_DST_MODE_H6(x)	(DMA_CHAN_CFG_SRC_MODE_H6(x) << 16)
82#define DMA_CHAN_CFG_DST_BURST_A31(x)	(DMA_CHAN_CFG_SRC_BURST_A31(x) << 16)
83#define DMA_CHAN_CFG_DST_BURST_H3(x)	(DMA_CHAN_CFG_SRC_BURST_H3(x) << 16)
84#define DMA_CHAN_CFG_DST_WIDTH(x)	(DMA_CHAN_CFG_SRC_WIDTH(x) << 16)
85
86#define DMA_CHAN_CUR_SRC	0x10
87
88#define DMA_CHAN_CUR_DST	0x14
89
90#define DMA_CHAN_CUR_CNT	0x18
91
92#define DMA_CHAN_CUR_PARA	0x1c
93
94/*
95 * LLI address mangling
96 *
97 * The LLI link physical address is also mangled, but we avoid dealing
98 * with that by allocating LLIs from the DMA32 zone.
99 */
100#define SRC_HIGH_ADDR(x)		(((x) & 0x3U) << 16)
101#define DST_HIGH_ADDR(x)		(((x) & 0x3U) << 18)
102
103/*
104 * Various hardware related defines
105 */
106#define LLI_LAST_ITEM	0xfffff800
107#define NORMAL_WAIT	8
108#define DRQ_SDRAM	1
109#define LINEAR_MODE     0
110#define IO_MODE         1
111
112/* forward declaration */
113struct sun6i_dma_dev;
114
115/*
116 * Hardware channels / ports representation
117 *
118 * The hardware is used in several SoCs, with differing numbers
119 * of channels and endpoints. This structure ties those numbers
120 * to a certain compatible string.
121 */
122struct sun6i_dma_config {
123	u32 nr_max_channels;
124	u32 nr_max_requests;
125	u32 nr_max_vchans;
126	/*
127	 * In the datasheets/user manuals of newer Allwinner SoCs, a special
128	 * bit (bit 2 at register 0x20) is present.
129	 * It's named "DMA MCLK interface circuit auto gating bit" in the
130	 * documents, and the footnote of this register says that this bit
131	 * should be set up when initializing the DMA controller.
132	 * Allwinner A23/A33 user manuals do not have this bit documented,
133	 * however these SoCs really have and need this bit, as seen in the
134	 * BSP kernel source code.
135	 */
136	void (*clock_autogate_enable)(struct sun6i_dma_dev *);
137	void (*set_burst_length)(u32 *p_cfg, s8 src_burst, s8 dst_burst);
138	void (*set_drq)(u32 *p_cfg, s8 src_drq, s8 dst_drq);
139	void (*set_mode)(u32 *p_cfg, s8 src_mode, s8 dst_mode);
140	u32 src_burst_lengths;
141	u32 dst_burst_lengths;
142	u32 src_addr_widths;
143	u32 dst_addr_widths;
144	bool has_high_addr;
145	bool has_mbus_clk;
146};
147
148/*
149 * Hardware representation of the LLI
150 *
151 * The hardware will be fed the physical address of this structure,
152 * and read its content in order to start the transfer.
153 */
154struct sun6i_dma_lli {
155	u32			cfg;
156	u32			src;
157	u32			dst;
158	u32			len;
159	u32			para;
160	u32			p_lli_next;
161
162	/*
163	 * This field is not used by the DMA controller, but will be
164	 * used by the CPU to go through the list (mostly for dumping
165	 * or freeing it).
166	 */
167	struct sun6i_dma_lli	*v_lli_next;
168};
169
170
171struct sun6i_desc {
172	struct virt_dma_desc	vd;
173	dma_addr_t		p_lli;
174	struct sun6i_dma_lli	*v_lli;
175};
176
177struct sun6i_pchan {
178	u32			idx;
179	void __iomem		*base;
180	struct sun6i_vchan	*vchan;
181	struct sun6i_desc	*desc;
182	struct sun6i_desc	*done;
183};
184
185struct sun6i_vchan {
186	struct virt_dma_chan	vc;
187	struct list_head	node;
188	struct dma_slave_config	cfg;
189	struct sun6i_pchan	*phy;
190	u8			port;
191	u8			irq_type;
192	bool			cyclic;
193};
194
195struct sun6i_dma_dev {
196	struct dma_device	slave;
197	void __iomem		*base;
198	struct clk		*clk;
199	struct clk		*clk_mbus;
200	int			irq;
201	spinlock_t		lock;
202	struct reset_control	*rstc;
203	struct tasklet_struct	task;
204	atomic_t		tasklet_shutdown;
205	struct list_head	pending;
206	struct dma_pool		*pool;
207	struct sun6i_pchan	*pchans;
208	struct sun6i_vchan	*vchans;
209	const struct sun6i_dma_config *cfg;
210	u32			num_pchans;
211	u32			num_vchans;
212	u32			max_request;
213};
214
215static struct device *chan2dev(struct dma_chan *chan)
216{
217	return &chan->dev->device;
218}
219
220static inline struct sun6i_dma_dev *to_sun6i_dma_dev(struct dma_device *d)
221{
222	return container_of(d, struct sun6i_dma_dev, slave);
223}
224
225static inline struct sun6i_vchan *to_sun6i_vchan(struct dma_chan *chan)
226{
227	return container_of(chan, struct sun6i_vchan, vc.chan);
228}
229
230static inline struct sun6i_desc *
231to_sun6i_desc(struct dma_async_tx_descriptor *tx)
232{
233	return container_of(tx, struct sun6i_desc, vd.tx);
234}
235
236static inline void sun6i_dma_dump_com_regs(struct sun6i_dma_dev *sdev)
237{
238	dev_dbg(sdev->slave.dev, "Common register:\n"
239		"\tmask0(%04x): 0x%08x\n"
240		"\tmask1(%04x): 0x%08x\n"
241		"\tpend0(%04x): 0x%08x\n"
242		"\tpend1(%04x): 0x%08x\n"
243		"\tstats(%04x): 0x%08x\n",
244		DMA_IRQ_EN(0), readl(sdev->base + DMA_IRQ_EN(0)),
245		DMA_IRQ_EN(1), readl(sdev->base + DMA_IRQ_EN(1)),
246		DMA_IRQ_STAT(0), readl(sdev->base + DMA_IRQ_STAT(0)),
247		DMA_IRQ_STAT(1), readl(sdev->base + DMA_IRQ_STAT(1)),
248		DMA_STAT, readl(sdev->base + DMA_STAT));
249}
250
251static inline void sun6i_dma_dump_chan_regs(struct sun6i_dma_dev *sdev,
252					    struct sun6i_pchan *pchan)
253{
254	dev_dbg(sdev->slave.dev, "Chan %d reg:\n"
255		"\t___en(%04x): \t0x%08x\n"
256		"\tpause(%04x): \t0x%08x\n"
257		"\tstart(%04x): \t0x%08x\n"
258		"\t__cfg(%04x): \t0x%08x\n"
259		"\t__src(%04x): \t0x%08x\n"
260		"\t__dst(%04x): \t0x%08x\n"
261		"\tcount(%04x): \t0x%08x\n"
262		"\t_para(%04x): \t0x%08x\n\n",
263		pchan->idx,
264		DMA_CHAN_ENABLE,
265		readl(pchan->base + DMA_CHAN_ENABLE),
266		DMA_CHAN_PAUSE,
267		readl(pchan->base + DMA_CHAN_PAUSE),
268		DMA_CHAN_LLI_ADDR,
269		readl(pchan->base + DMA_CHAN_LLI_ADDR),
270		DMA_CHAN_CUR_CFG,
271		readl(pchan->base + DMA_CHAN_CUR_CFG),
272		DMA_CHAN_CUR_SRC,
273		readl(pchan->base + DMA_CHAN_CUR_SRC),
274		DMA_CHAN_CUR_DST,
275		readl(pchan->base + DMA_CHAN_CUR_DST),
276		DMA_CHAN_CUR_CNT,
277		readl(pchan->base + DMA_CHAN_CUR_CNT),
278		DMA_CHAN_CUR_PARA,
279		readl(pchan->base + DMA_CHAN_CUR_PARA));
280}
281
282static inline s8 convert_burst(u32 maxburst)
283{
284	switch (maxburst) {
285	case 1:
286		return 0;
287	case 4:
288		return 1;
289	case 8:
290		return 2;
291	case 16:
292		return 3;
293	default:
294		return -EINVAL;
295	}
296}
297
298static inline s8 convert_buswidth(enum dma_slave_buswidth addr_width)
299{
300	return ilog2(addr_width);
301}
302
303static void sun6i_enable_clock_autogate_a23(struct sun6i_dma_dev *sdev)
304{
305	writel(SUN8I_DMA_GATE_ENABLE, sdev->base + SUN8I_DMA_GATE);
306}
307
308static void sun6i_enable_clock_autogate_h3(struct sun6i_dma_dev *sdev)
309{
310	writel(SUNXI_H3_DMA_GATE_ENABLE, sdev->base + SUNXI_H3_DMA_GATE);
311}
312
313static void sun6i_set_burst_length_a31(u32 *p_cfg, s8 src_burst, s8 dst_burst)
314{
315	*p_cfg |= DMA_CHAN_CFG_SRC_BURST_A31(src_burst) |
316		  DMA_CHAN_CFG_DST_BURST_A31(dst_burst);
317}
318
319static void sun6i_set_burst_length_h3(u32 *p_cfg, s8 src_burst, s8 dst_burst)
320{
321	*p_cfg |= DMA_CHAN_CFG_SRC_BURST_H3(src_burst) |
322		  DMA_CHAN_CFG_DST_BURST_H3(dst_burst);
323}
324
325static void sun6i_set_drq_a31(u32 *p_cfg, s8 src_drq, s8 dst_drq)
326{
327	*p_cfg |= DMA_CHAN_CFG_SRC_DRQ_A31(src_drq) |
328		  DMA_CHAN_CFG_DST_DRQ_A31(dst_drq);
329}
330
331static void sun6i_set_drq_h6(u32 *p_cfg, s8 src_drq, s8 dst_drq)
332{
333	*p_cfg |= DMA_CHAN_CFG_SRC_DRQ_H6(src_drq) |
334		  DMA_CHAN_CFG_DST_DRQ_H6(dst_drq);
335}
336
337static void sun6i_set_mode_a31(u32 *p_cfg, s8 src_mode, s8 dst_mode)
338{
339	*p_cfg |= DMA_CHAN_CFG_SRC_MODE_A31(src_mode) |
340		  DMA_CHAN_CFG_DST_MODE_A31(dst_mode);
341}
342
343static void sun6i_set_mode_h6(u32 *p_cfg, s8 src_mode, s8 dst_mode)
344{
345	*p_cfg |= DMA_CHAN_CFG_SRC_MODE_H6(src_mode) |
346		  DMA_CHAN_CFG_DST_MODE_H6(dst_mode);
347}
348
349static size_t sun6i_get_chan_size(struct sun6i_pchan *pchan)
350{
351	struct sun6i_desc *txd = pchan->desc;
352	struct sun6i_dma_lli *lli;
353	size_t bytes;
354	dma_addr_t pos;
355
356	pos = readl(pchan->base + DMA_CHAN_LLI_ADDR);
357	bytes = readl(pchan->base + DMA_CHAN_CUR_CNT);
358
359	if (pos == LLI_LAST_ITEM)
360		return bytes;
361
362	for (lli = txd->v_lli; lli; lli = lli->v_lli_next) {
363		if (lli->p_lli_next == pos) {
364			for (lli = lli->v_lli_next; lli; lli = lli->v_lli_next)
365				bytes += lli->len;
366			break;
367		}
368	}
369
370	return bytes;
371}
372
373static void *sun6i_dma_lli_add(struct sun6i_dma_lli *prev,
374			       struct sun6i_dma_lli *next,
375			       dma_addr_t next_phy,
376			       struct sun6i_desc *txd)
377{
378	if ((!prev && !txd) || !next)
379		return NULL;
380
381	if (!prev) {
382		txd->p_lli = next_phy;
383		txd->v_lli = next;
384	} else {
385		prev->p_lli_next = next_phy;
386		prev->v_lli_next = next;
387	}
388
389	next->p_lli_next = LLI_LAST_ITEM;
390	next->v_lli_next = NULL;
391
392	return next;
393}
394
395static inline void sun6i_dma_dump_lli(struct sun6i_vchan *vchan,
396				      struct sun6i_dma_lli *v_lli,
397				      dma_addr_t p_lli)
398{
399	dev_dbg(chan2dev(&vchan->vc.chan),
400		"\n\tdesc:\tp - %pad v - 0x%p\n"
401		"\t\tc - 0x%08x s - 0x%08x d - 0x%08x\n"
402		"\t\tl - 0x%08x p - 0x%08x n - 0x%08x\n",
403		&p_lli, v_lli,
404		v_lli->cfg, v_lli->src, v_lli->dst,
405		v_lli->len, v_lli->para, v_lli->p_lli_next);
406}
407
408static void sun6i_dma_free_desc(struct virt_dma_desc *vd)
409{
410	struct sun6i_desc *txd = to_sun6i_desc(&vd->tx);
411	struct sun6i_dma_dev *sdev = to_sun6i_dma_dev(vd->tx.chan->device);
412	struct sun6i_dma_lli *v_lli, *v_next;
413	dma_addr_t p_lli, p_next;
414
415	if (unlikely(!txd))
416		return;
417
418	p_lli = txd->p_lli;
419	v_lli = txd->v_lli;
420
421	while (v_lli) {
422		v_next = v_lli->v_lli_next;
423		p_next = v_lli->p_lli_next;
424
425		dma_pool_free(sdev->pool, v_lli, p_lli);
426
427		v_lli = v_next;
428		p_lli = p_next;
429	}
430
431	kfree(txd);
432}
433
434static int sun6i_dma_start_desc(struct sun6i_vchan *vchan)
435{
436	struct sun6i_dma_dev *sdev = to_sun6i_dma_dev(vchan->vc.chan.device);
437	struct virt_dma_desc *desc = vchan_next_desc(&vchan->vc);
438	struct sun6i_pchan *pchan = vchan->phy;
439	u32 irq_val, irq_reg, irq_offset;
440
441	if (!pchan)
442		return -EAGAIN;
443
444	if (!desc) {
445		pchan->desc = NULL;
446		pchan->done = NULL;
447		return -EAGAIN;
448	}
449
450	list_del(&desc->node);
451
452	pchan->desc = to_sun6i_desc(&desc->tx);
453	pchan->done = NULL;
454
455	sun6i_dma_dump_lli(vchan, pchan->desc->v_lli, pchan->desc->p_lli);
456
457	irq_reg = pchan->idx / DMA_IRQ_CHAN_NR;
458	irq_offset = pchan->idx % DMA_IRQ_CHAN_NR;
459
460	vchan->irq_type = vchan->cyclic ? DMA_IRQ_PKG : DMA_IRQ_QUEUE;
461
462	irq_val = readl(sdev->base + DMA_IRQ_EN(irq_reg));
463	irq_val &= ~((DMA_IRQ_HALF | DMA_IRQ_PKG | DMA_IRQ_QUEUE) <<
464			(irq_offset * DMA_IRQ_CHAN_WIDTH));
465	irq_val |= vchan->irq_type << (irq_offset * DMA_IRQ_CHAN_WIDTH);
466	writel(irq_val, sdev->base + DMA_IRQ_EN(irq_reg));
467
468	writel(pchan->desc->p_lli, pchan->base + DMA_CHAN_LLI_ADDR);
469	writel(DMA_CHAN_ENABLE_START, pchan->base + DMA_CHAN_ENABLE);
470
471	sun6i_dma_dump_com_regs(sdev);
472	sun6i_dma_dump_chan_regs(sdev, pchan);
473
474	return 0;
475}
476
477static void sun6i_dma_tasklet(struct tasklet_struct *t)
478{
479	struct sun6i_dma_dev *sdev = from_tasklet(sdev, t, task);
480	struct sun6i_vchan *vchan;
481	struct sun6i_pchan *pchan;
482	unsigned int pchan_alloc = 0;
483	unsigned int pchan_idx;
484
485	list_for_each_entry(vchan, &sdev->slave.channels, vc.chan.device_node) {
486		spin_lock_irq(&vchan->vc.lock);
487
488		pchan = vchan->phy;
489
490		if (pchan && pchan->done) {
491			if (sun6i_dma_start_desc(vchan)) {
492				/*
493				 * No current txd associated with this channel
494				 */
495				dev_dbg(sdev->slave.dev, "pchan %u: free\n",
496					pchan->idx);
497
498				/* Mark this channel free */
499				vchan->phy = NULL;
500				pchan->vchan = NULL;
501			}
502		}
503		spin_unlock_irq(&vchan->vc.lock);
504	}
505
506	spin_lock_irq(&sdev->lock);
507	for (pchan_idx = 0; pchan_idx < sdev->num_pchans; pchan_idx++) {
508		pchan = &sdev->pchans[pchan_idx];
509
510		if (pchan->vchan || list_empty(&sdev->pending))
511			continue;
512
513		vchan = list_first_entry(&sdev->pending,
514					 struct sun6i_vchan, node);
515
516		/* Remove from pending channels */
517		list_del_init(&vchan->node);
518		pchan_alloc |= BIT(pchan_idx);
519
520		/* Mark this channel allocated */
521		pchan->vchan = vchan;
522		vchan->phy = pchan;
523		dev_dbg(sdev->slave.dev, "pchan %u: alloc vchan %p\n",
524			pchan->idx, &vchan->vc);
525	}
526	spin_unlock_irq(&sdev->lock);
527
528	for (pchan_idx = 0; pchan_idx < sdev->num_pchans; pchan_idx++) {
529		if (!(pchan_alloc & BIT(pchan_idx)))
530			continue;
531
532		pchan = sdev->pchans + pchan_idx;
533		vchan = pchan->vchan;
534		if (vchan) {
535			spin_lock_irq(&vchan->vc.lock);
536			sun6i_dma_start_desc(vchan);
537			spin_unlock_irq(&vchan->vc.lock);
538		}
539	}
540}
541
542static irqreturn_t sun6i_dma_interrupt(int irq, void *dev_id)
543{
544	struct sun6i_dma_dev *sdev = dev_id;
545	struct sun6i_vchan *vchan;
546	struct sun6i_pchan *pchan;
547	int i, j, ret = IRQ_NONE;
548	u32 status;
549
550	for (i = 0; i < sdev->num_pchans / DMA_IRQ_CHAN_NR; i++) {
551		status = readl(sdev->base + DMA_IRQ_STAT(i));
552		if (!status)
553			continue;
554
555		dev_dbg(sdev->slave.dev, "DMA irq status %s: 0x%x\n",
556			i ? "high" : "low", status);
557
558		writel(status, sdev->base + DMA_IRQ_STAT(i));
559
560		for (j = 0; (j < DMA_IRQ_CHAN_NR) && status; j++) {
561			pchan = sdev->pchans + j;
562			vchan = pchan->vchan;
563			if (vchan && (status & vchan->irq_type)) {
564				if (vchan->cyclic) {
565					vchan_cyclic_callback(&pchan->desc->vd);
566				} else {
567					spin_lock(&vchan->vc.lock);
568					vchan_cookie_complete(&pchan->desc->vd);
569					pchan->done = pchan->desc;
570					spin_unlock(&vchan->vc.lock);
571				}
572			}
573
574			status = status >> DMA_IRQ_CHAN_WIDTH;
575		}
576
577		if (!atomic_read(&sdev->tasklet_shutdown))
578			tasklet_schedule(&sdev->task);
579		ret = IRQ_HANDLED;
580	}
581
582	return ret;
583}
584
585static int set_config(struct sun6i_dma_dev *sdev,
586			struct dma_slave_config *sconfig,
587			enum dma_transfer_direction direction,
588			u32 *p_cfg)
589{
590	enum dma_slave_buswidth src_addr_width, dst_addr_width;
591	u32 src_maxburst, dst_maxburst;
592	s8 src_width, dst_width, src_burst, dst_burst;
593
594	src_addr_width = sconfig->src_addr_width;
595	dst_addr_width = sconfig->dst_addr_width;
596	src_maxburst = sconfig->src_maxburst;
597	dst_maxburst = sconfig->dst_maxburst;
598
599	switch (direction) {
600	case DMA_MEM_TO_DEV:
601		if (src_addr_width == DMA_SLAVE_BUSWIDTH_UNDEFINED)
602			src_addr_width = DMA_SLAVE_BUSWIDTH_4_BYTES;
603		src_maxburst = src_maxburst ? src_maxburst : 8;
604		break;
605	case DMA_DEV_TO_MEM:
606		if (dst_addr_width == DMA_SLAVE_BUSWIDTH_UNDEFINED)
607			dst_addr_width = DMA_SLAVE_BUSWIDTH_4_BYTES;
608		dst_maxburst = dst_maxburst ? dst_maxburst : 8;
609		break;
610	default:
611		return -EINVAL;
612	}
613
614	if (!(BIT(src_addr_width) & sdev->slave.src_addr_widths))
615		return -EINVAL;
616	if (!(BIT(dst_addr_width) & sdev->slave.dst_addr_widths))
617		return -EINVAL;
618	if (!(BIT(src_maxburst) & sdev->cfg->src_burst_lengths))
619		return -EINVAL;
620	if (!(BIT(dst_maxburst) & sdev->cfg->dst_burst_lengths))
621		return -EINVAL;
622
623	src_width = convert_buswidth(src_addr_width);
624	dst_width = convert_buswidth(dst_addr_width);
625	dst_burst = convert_burst(dst_maxburst);
626	src_burst = convert_burst(src_maxburst);
627
628	*p_cfg = DMA_CHAN_CFG_SRC_WIDTH(src_width) |
629		DMA_CHAN_CFG_DST_WIDTH(dst_width);
630
631	sdev->cfg->set_burst_length(p_cfg, src_burst, dst_burst);
632
633	return 0;
634}
635
636static inline void sun6i_dma_set_addr(struct sun6i_dma_dev *sdev,
637				      struct sun6i_dma_lli *v_lli,
638				      dma_addr_t src, dma_addr_t dst)
639{
640	v_lli->src = lower_32_bits(src);
641	v_lli->dst = lower_32_bits(dst);
642
643	if (sdev->cfg->has_high_addr)
644		v_lli->para |= SRC_HIGH_ADDR(upper_32_bits(src)) |
645			       DST_HIGH_ADDR(upper_32_bits(dst));
646}
647
648static struct dma_async_tx_descriptor *sun6i_dma_prep_dma_memcpy(
649		struct dma_chan *chan, dma_addr_t dest, dma_addr_t src,
650		size_t len, unsigned long flags)
651{
652	struct sun6i_dma_dev *sdev = to_sun6i_dma_dev(chan->device);
653	struct sun6i_vchan *vchan = to_sun6i_vchan(chan);
654	struct sun6i_dma_lli *v_lli;
655	struct sun6i_desc *txd;
656	dma_addr_t p_lli;
657	s8 burst, width;
658
659	dev_dbg(chan2dev(chan),
660		"%s; chan: %d, dest: %pad, src: %pad, len: %zu. flags: 0x%08lx\n",
661		__func__, vchan->vc.chan.chan_id, &dest, &src, len, flags);
662
663	if (!len)
664		return NULL;
665
666	txd = kzalloc(sizeof(*txd), GFP_NOWAIT);
667	if (!txd)
668		return NULL;
669
670	v_lli = dma_pool_alloc(sdev->pool, GFP_DMA32 | GFP_NOWAIT, &p_lli);
671	if (!v_lli) {
672		dev_err(sdev->slave.dev, "Failed to alloc lli memory\n");
673		goto err_txd_free;
674	}
675
676	v_lli->len = len;
677	v_lli->para = NORMAL_WAIT;
678	sun6i_dma_set_addr(sdev, v_lli, src, dest);
679
680	burst = convert_burst(8);
681	width = convert_buswidth(DMA_SLAVE_BUSWIDTH_4_BYTES);
682	v_lli->cfg = DMA_CHAN_CFG_SRC_WIDTH(width) |
683		DMA_CHAN_CFG_DST_WIDTH(width);
684
685	sdev->cfg->set_burst_length(&v_lli->cfg, burst, burst);
686	sdev->cfg->set_drq(&v_lli->cfg, DRQ_SDRAM, DRQ_SDRAM);
687	sdev->cfg->set_mode(&v_lli->cfg, LINEAR_MODE, LINEAR_MODE);
688
689	sun6i_dma_lli_add(NULL, v_lli, p_lli, txd);
690
691	sun6i_dma_dump_lli(vchan, v_lli, p_lli);
692
693	return vchan_tx_prep(&vchan->vc, &txd->vd, flags);
694
695err_txd_free:
696	kfree(txd);
697	return NULL;
698}
699
700static struct dma_async_tx_descriptor *sun6i_dma_prep_slave_sg(
701		struct dma_chan *chan, struct scatterlist *sgl,
702		unsigned int sg_len, enum dma_transfer_direction dir,
703		unsigned long flags, void *context)
704{
705	struct sun6i_dma_dev *sdev = to_sun6i_dma_dev(chan->device);
706	struct sun6i_vchan *vchan = to_sun6i_vchan(chan);
707	struct dma_slave_config *sconfig = &vchan->cfg;
708	struct sun6i_dma_lli *v_lli, *prev = NULL;
709	struct sun6i_desc *txd;
710	struct scatterlist *sg;
711	dma_addr_t p_lli;
712	u32 lli_cfg;
713	int i, ret;
714
715	if (!sgl)
716		return NULL;
717
718	ret = set_config(sdev, sconfig, dir, &lli_cfg);
719	if (ret) {
720		dev_err(chan2dev(chan), "Invalid DMA configuration\n");
721		return NULL;
722	}
723
724	txd = kzalloc(sizeof(*txd), GFP_NOWAIT);
725	if (!txd)
726		return NULL;
727
728	for_each_sg(sgl, sg, sg_len, i) {
729		v_lli = dma_pool_alloc(sdev->pool, GFP_DMA32 | GFP_NOWAIT, &p_lli);
730		if (!v_lli)
731			goto err_lli_free;
732
733		v_lli->len = sg_dma_len(sg);
734		v_lli->para = NORMAL_WAIT;
735
736		if (dir == DMA_MEM_TO_DEV) {
737			sun6i_dma_set_addr(sdev, v_lli,
738					   sg_dma_address(sg),
739					   sconfig->dst_addr);
740			v_lli->cfg = lli_cfg;
741			sdev->cfg->set_drq(&v_lli->cfg, DRQ_SDRAM, vchan->port);
742			sdev->cfg->set_mode(&v_lli->cfg, LINEAR_MODE, IO_MODE);
743
744			dev_dbg(chan2dev(chan),
745				"%s; chan: %d, dest: %pad, src: %pad, len: %u. flags: 0x%08lx\n",
746				__func__, vchan->vc.chan.chan_id,
747				&sconfig->dst_addr, &sg_dma_address(sg),
748				sg_dma_len(sg), flags);
749
750		} else {
751			sun6i_dma_set_addr(sdev, v_lli,
752					   sconfig->src_addr,
753					   sg_dma_address(sg));
754			v_lli->cfg = lli_cfg;
755			sdev->cfg->set_drq(&v_lli->cfg, vchan->port, DRQ_SDRAM);
756			sdev->cfg->set_mode(&v_lli->cfg, IO_MODE, LINEAR_MODE);
757
758			dev_dbg(chan2dev(chan),
759				"%s; chan: %d, dest: %pad, src: %pad, len: %u. flags: 0x%08lx\n",
760				__func__, vchan->vc.chan.chan_id,
761				&sg_dma_address(sg), &sconfig->src_addr,
762				sg_dma_len(sg), flags);
763		}
764
765		prev = sun6i_dma_lli_add(prev, v_lli, p_lli, txd);
766	}
767
768	dev_dbg(chan2dev(chan), "First: %pad\n", &txd->p_lli);
769	for (p_lli = txd->p_lli, v_lli = txd->v_lli; v_lli;
770	     p_lli = v_lli->p_lli_next, v_lli = v_lli->v_lli_next)
771		sun6i_dma_dump_lli(vchan, v_lli, p_lli);
772
773	return vchan_tx_prep(&vchan->vc, &txd->vd, flags);
774
775err_lli_free:
776	for (p_lli = txd->p_lli, v_lli = txd->v_lli; v_lli;
777	     p_lli = v_lli->p_lli_next, v_lli = v_lli->v_lli_next)
778		dma_pool_free(sdev->pool, v_lli, p_lli);
779	kfree(txd);
780	return NULL;
781}
782
783static struct dma_async_tx_descriptor *sun6i_dma_prep_dma_cyclic(
784					struct dma_chan *chan,
785					dma_addr_t buf_addr,
786					size_t buf_len,
787					size_t period_len,
788					enum dma_transfer_direction dir,
789					unsigned long flags)
790{
791	struct sun6i_dma_dev *sdev = to_sun6i_dma_dev(chan->device);
792	struct sun6i_vchan *vchan = to_sun6i_vchan(chan);
793	struct dma_slave_config *sconfig = &vchan->cfg;
794	struct sun6i_dma_lli *v_lli, *prev = NULL;
795	struct sun6i_desc *txd;
796	dma_addr_t p_lli;
797	u32 lli_cfg;
798	unsigned int i, periods = buf_len / period_len;
799	int ret;
800
801	ret = set_config(sdev, sconfig, dir, &lli_cfg);
802	if (ret) {
803		dev_err(chan2dev(chan), "Invalid DMA configuration\n");
804		return NULL;
805	}
806
807	txd = kzalloc(sizeof(*txd), GFP_NOWAIT);
808	if (!txd)
809		return NULL;
810
811	for (i = 0; i < periods; i++) {
812		v_lli = dma_pool_alloc(sdev->pool, GFP_DMA32 | GFP_NOWAIT, &p_lli);
813		if (!v_lli) {
814			dev_err(sdev->slave.dev, "Failed to alloc lli memory\n");
815			goto err_lli_free;
816		}
817
818		v_lli->len = period_len;
819		v_lli->para = NORMAL_WAIT;
820
821		if (dir == DMA_MEM_TO_DEV) {
822			sun6i_dma_set_addr(sdev, v_lli,
823					   buf_addr + period_len * i,
824					   sconfig->dst_addr);
825			v_lli->cfg = lli_cfg;
826			sdev->cfg->set_drq(&v_lli->cfg, DRQ_SDRAM, vchan->port);
827			sdev->cfg->set_mode(&v_lli->cfg, LINEAR_MODE, IO_MODE);
828		} else {
829			sun6i_dma_set_addr(sdev, v_lli,
830					   sconfig->src_addr,
831					   buf_addr + period_len * i);
832			v_lli->cfg = lli_cfg;
833			sdev->cfg->set_drq(&v_lli->cfg, vchan->port, DRQ_SDRAM);
834			sdev->cfg->set_mode(&v_lli->cfg, IO_MODE, LINEAR_MODE);
835		}
836
837		prev = sun6i_dma_lli_add(prev, v_lli, p_lli, txd);
838	}
839
840	prev->p_lli_next = txd->p_lli;		/* cyclic list */
841
842	vchan->cyclic = true;
843
844	return vchan_tx_prep(&vchan->vc, &txd->vd, flags);
845
846err_lli_free:
847	for (p_lli = txd->p_lli, v_lli = txd->v_lli; v_lli;
848	     p_lli = v_lli->p_lli_next, v_lli = v_lli->v_lli_next)
849		dma_pool_free(sdev->pool, v_lli, p_lli);
850	kfree(txd);
851	return NULL;
852}
853
854static int sun6i_dma_config(struct dma_chan *chan,
855			    struct dma_slave_config *config)
856{
857	struct sun6i_vchan *vchan = to_sun6i_vchan(chan);
858
859	memcpy(&vchan->cfg, config, sizeof(*config));
860
861	return 0;
862}
863
864static int sun6i_dma_pause(struct dma_chan *chan)
865{
866	struct sun6i_dma_dev *sdev = to_sun6i_dma_dev(chan->device);
867	struct sun6i_vchan *vchan = to_sun6i_vchan(chan);
868	struct sun6i_pchan *pchan = vchan->phy;
869
870	dev_dbg(chan2dev(chan), "vchan %p: pause\n", &vchan->vc);
871
872	if (pchan) {
873		writel(DMA_CHAN_PAUSE_PAUSE,
874		       pchan->base + DMA_CHAN_PAUSE);
875	} else {
876		spin_lock(&sdev->lock);
877		list_del_init(&vchan->node);
878		spin_unlock(&sdev->lock);
879	}
880
881	return 0;
882}
883
884static int sun6i_dma_resume(struct dma_chan *chan)
885{
886	struct sun6i_dma_dev *sdev = to_sun6i_dma_dev(chan->device);
887	struct sun6i_vchan *vchan = to_sun6i_vchan(chan);
888	struct sun6i_pchan *pchan = vchan->phy;
889	unsigned long flags;
890
891	dev_dbg(chan2dev(chan), "vchan %p: resume\n", &vchan->vc);
892
893	spin_lock_irqsave(&vchan->vc.lock, flags);
894
895	if (pchan) {
896		writel(DMA_CHAN_PAUSE_RESUME,
897		       pchan->base + DMA_CHAN_PAUSE);
898	} else if (!list_empty(&vchan->vc.desc_issued)) {
899		spin_lock(&sdev->lock);
900		list_add_tail(&vchan->node, &sdev->pending);
901		spin_unlock(&sdev->lock);
902	}
903
904	spin_unlock_irqrestore(&vchan->vc.lock, flags);
905
906	return 0;
907}
908
909static int sun6i_dma_terminate_all(struct dma_chan *chan)
910{
911	struct sun6i_dma_dev *sdev = to_sun6i_dma_dev(chan->device);
912	struct sun6i_vchan *vchan = to_sun6i_vchan(chan);
913	struct sun6i_pchan *pchan = vchan->phy;
914	unsigned long flags;
915	LIST_HEAD(head);
916
917	spin_lock(&sdev->lock);
918	list_del_init(&vchan->node);
919	spin_unlock(&sdev->lock);
920
921	spin_lock_irqsave(&vchan->vc.lock, flags);
922
923	if (vchan->cyclic) {
924		vchan->cyclic = false;
925		if (pchan && pchan->desc) {
926			struct virt_dma_desc *vd = &pchan->desc->vd;
927			struct virt_dma_chan *vc = &vchan->vc;
928
929			list_add_tail(&vd->node, &vc->desc_completed);
930		}
931	}
932
933	vchan_get_all_descriptors(&vchan->vc, &head);
934
935	if (pchan) {
936		writel(DMA_CHAN_ENABLE_STOP, pchan->base + DMA_CHAN_ENABLE);
937		writel(DMA_CHAN_PAUSE_RESUME, pchan->base + DMA_CHAN_PAUSE);
938
939		vchan->phy = NULL;
940		pchan->vchan = NULL;
941		pchan->desc = NULL;
942		pchan->done = NULL;
943	}
944
945	spin_unlock_irqrestore(&vchan->vc.lock, flags);
946
947	vchan_dma_desc_free_list(&vchan->vc, &head);
948
949	return 0;
950}
951
952static enum dma_status sun6i_dma_tx_status(struct dma_chan *chan,
953					   dma_cookie_t cookie,
954					   struct dma_tx_state *state)
955{
956	struct sun6i_vchan *vchan = to_sun6i_vchan(chan);
957	struct sun6i_pchan *pchan = vchan->phy;
958	struct sun6i_dma_lli *lli;
959	struct virt_dma_desc *vd;
960	struct sun6i_desc *txd;
961	enum dma_status ret;
962	unsigned long flags;
963	size_t bytes = 0;
964
965	ret = dma_cookie_status(chan, cookie, state);
966	if (ret == DMA_COMPLETE || !state)
967		return ret;
968
969	spin_lock_irqsave(&vchan->vc.lock, flags);
970
971	vd = vchan_find_desc(&vchan->vc, cookie);
972	txd = to_sun6i_desc(&vd->tx);
973
974	if (vd) {
975		for (lli = txd->v_lli; lli != NULL; lli = lli->v_lli_next)
976			bytes += lli->len;
977	} else if (!pchan || !pchan->desc) {
978		bytes = 0;
979	} else {
980		bytes = sun6i_get_chan_size(pchan);
981	}
982
983	spin_unlock_irqrestore(&vchan->vc.lock, flags);
984
985	dma_set_residue(state, bytes);
986
987	return ret;
988}
989
990static void sun6i_dma_issue_pending(struct dma_chan *chan)
991{
992	struct sun6i_dma_dev *sdev = to_sun6i_dma_dev(chan->device);
993	struct sun6i_vchan *vchan = to_sun6i_vchan(chan);
994	unsigned long flags;
995
996	spin_lock_irqsave(&vchan->vc.lock, flags);
997
998	if (vchan_issue_pending(&vchan->vc)) {
999		spin_lock(&sdev->lock);
1000
1001		if (!vchan->phy && list_empty(&vchan->node)) {
1002			list_add_tail(&vchan->node, &sdev->pending);
1003			tasklet_schedule(&sdev->task);
1004			dev_dbg(chan2dev(chan), "vchan %p: issued\n",
1005				&vchan->vc);
1006		}
1007
1008		spin_unlock(&sdev->lock);
1009	} else {
1010		dev_dbg(chan2dev(chan), "vchan %p: nothing to issue\n",
1011			&vchan->vc);
1012	}
1013
1014	spin_unlock_irqrestore(&vchan->vc.lock, flags);
1015}
1016
1017static void sun6i_dma_free_chan_resources(struct dma_chan *chan)
1018{
1019	struct sun6i_dma_dev *sdev = to_sun6i_dma_dev(chan->device);
1020	struct sun6i_vchan *vchan = to_sun6i_vchan(chan);
1021	unsigned long flags;
1022
1023	spin_lock_irqsave(&sdev->lock, flags);
1024	list_del_init(&vchan->node);
1025	spin_unlock_irqrestore(&sdev->lock, flags);
1026
1027	vchan_free_chan_resources(&vchan->vc);
1028}
1029
1030static struct dma_chan *sun6i_dma_of_xlate(struct of_phandle_args *dma_spec,
1031					   struct of_dma *ofdma)
1032{
1033	struct sun6i_dma_dev *sdev = ofdma->of_dma_data;
1034	struct sun6i_vchan *vchan;
1035	struct dma_chan *chan;
1036	u8 port = dma_spec->args[0];
1037
1038	if (port > sdev->max_request)
1039		return NULL;
1040
1041	chan = dma_get_any_slave_channel(&sdev->slave);
1042	if (!chan)
1043		return NULL;
1044
1045	vchan = to_sun6i_vchan(chan);
1046	vchan->port = port;
1047
1048	return chan;
1049}
1050
1051static inline void sun6i_kill_tasklet(struct sun6i_dma_dev *sdev)
1052{
1053	/* Disable all interrupts from DMA */
1054	writel(0, sdev->base + DMA_IRQ_EN(0));
1055	writel(0, sdev->base + DMA_IRQ_EN(1));
1056
1057	/* Prevent spurious interrupts from scheduling the tasklet */
1058	atomic_inc(&sdev->tasklet_shutdown);
1059
1060	/* Make sure we won't have any further interrupts */
1061	devm_free_irq(sdev->slave.dev, sdev->irq, sdev);
1062
1063	/* Actually prevent the tasklet from being scheduled */
1064	tasklet_kill(&sdev->task);
1065}
1066
1067static inline void sun6i_dma_free(struct sun6i_dma_dev *sdev)
1068{
1069	int i;
1070
1071	for (i = 0; i < sdev->num_vchans; i++) {
1072		struct sun6i_vchan *vchan = &sdev->vchans[i];
1073
1074		list_del(&vchan->vc.chan.device_node);
1075		tasklet_kill(&vchan->vc.task);
1076	}
1077}
1078
1079/*
1080 * For A31:
1081 *
1082 * There's 16 physical channels that can work in parallel.
1083 *
1084 * However we have 30 different endpoints for our requests.
1085 *
1086 * Since the channels are able to handle only an unidirectional
1087 * transfer, we need to allocate more virtual channels so that
1088 * everyone can grab one channel.
1089 *
1090 * Some devices can't work in both direction (mostly because it
1091 * wouldn't make sense), so we have a bit fewer virtual channels than
1092 * 2 channels per endpoints.
1093 */
1094
1095static struct sun6i_dma_config sun6i_a31_dma_cfg = {
1096	.nr_max_channels = 16,
1097	.nr_max_requests = 30,
1098	.nr_max_vchans   = 53,
1099	.set_burst_length = sun6i_set_burst_length_a31,
1100	.set_drq          = sun6i_set_drq_a31,
1101	.set_mode         = sun6i_set_mode_a31,
1102	.src_burst_lengths = BIT(1) | BIT(8),
1103	.dst_burst_lengths = BIT(1) | BIT(8),
1104	.src_addr_widths   = BIT(DMA_SLAVE_BUSWIDTH_1_BYTE) |
1105			     BIT(DMA_SLAVE_BUSWIDTH_2_BYTES) |
1106			     BIT(DMA_SLAVE_BUSWIDTH_4_BYTES),
1107	.dst_addr_widths   = BIT(DMA_SLAVE_BUSWIDTH_1_BYTE) |
1108			     BIT(DMA_SLAVE_BUSWIDTH_2_BYTES) |
1109			     BIT(DMA_SLAVE_BUSWIDTH_4_BYTES),
1110};
1111
1112/*
1113 * The A23 only has 8 physical channels, a maximum DRQ port id of 24,
1114 * and a total of 37 usable source and destination endpoints.
1115 */
1116
1117static struct sun6i_dma_config sun8i_a23_dma_cfg = {
1118	.nr_max_channels = 8,
1119	.nr_max_requests = 24,
1120	.nr_max_vchans   = 37,
1121	.clock_autogate_enable = sun6i_enable_clock_autogate_a23,
1122	.set_burst_length = sun6i_set_burst_length_a31,
1123	.set_drq          = sun6i_set_drq_a31,
1124	.set_mode         = sun6i_set_mode_a31,
1125	.src_burst_lengths = BIT(1) | BIT(8),
1126	.dst_burst_lengths = BIT(1) | BIT(8),
1127	.src_addr_widths   = BIT(DMA_SLAVE_BUSWIDTH_1_BYTE) |
1128			     BIT(DMA_SLAVE_BUSWIDTH_2_BYTES) |
1129			     BIT(DMA_SLAVE_BUSWIDTH_4_BYTES),
1130	.dst_addr_widths   = BIT(DMA_SLAVE_BUSWIDTH_1_BYTE) |
1131			     BIT(DMA_SLAVE_BUSWIDTH_2_BYTES) |
1132			     BIT(DMA_SLAVE_BUSWIDTH_4_BYTES),
1133};
1134
1135static struct sun6i_dma_config sun8i_a83t_dma_cfg = {
1136	.nr_max_channels = 8,
1137	.nr_max_requests = 28,
1138	.nr_max_vchans   = 39,
1139	.clock_autogate_enable = sun6i_enable_clock_autogate_a23,
1140	.set_burst_length = sun6i_set_burst_length_a31,
1141	.set_drq          = sun6i_set_drq_a31,
1142	.set_mode         = sun6i_set_mode_a31,
1143	.src_burst_lengths = BIT(1) | BIT(8),
1144	.dst_burst_lengths = BIT(1) | BIT(8),
1145	.src_addr_widths   = BIT(DMA_SLAVE_BUSWIDTH_1_BYTE) |
1146			     BIT(DMA_SLAVE_BUSWIDTH_2_BYTES) |
1147			     BIT(DMA_SLAVE_BUSWIDTH_4_BYTES),
1148	.dst_addr_widths   = BIT(DMA_SLAVE_BUSWIDTH_1_BYTE) |
1149			     BIT(DMA_SLAVE_BUSWIDTH_2_BYTES) |
1150			     BIT(DMA_SLAVE_BUSWIDTH_4_BYTES),
1151};
1152
1153/*
1154 * The H3 has 12 physical channels, a maximum DRQ port id of 27,
1155 * and a total of 34 usable source and destination endpoints.
1156 * It also supports additional burst lengths and bus widths,
1157 * and the burst length fields have different offsets.
1158 */
1159
1160static struct sun6i_dma_config sun8i_h3_dma_cfg = {
1161	.nr_max_channels = 12,
1162	.nr_max_requests = 27,
1163	.nr_max_vchans   = 34,
1164	.clock_autogate_enable = sun6i_enable_clock_autogate_h3,
1165	.set_burst_length = sun6i_set_burst_length_h3,
1166	.set_drq          = sun6i_set_drq_a31,
1167	.set_mode         = sun6i_set_mode_a31,
1168	.src_burst_lengths = BIT(1) | BIT(4) | BIT(8) | BIT(16),
1169	.dst_burst_lengths = BIT(1) | BIT(4) | BIT(8) | BIT(16),
1170	.src_addr_widths   = BIT(DMA_SLAVE_BUSWIDTH_1_BYTE) |
1171			     BIT(DMA_SLAVE_BUSWIDTH_2_BYTES) |
1172			     BIT(DMA_SLAVE_BUSWIDTH_4_BYTES) |
1173			     BIT(DMA_SLAVE_BUSWIDTH_8_BYTES),
1174	.dst_addr_widths   = BIT(DMA_SLAVE_BUSWIDTH_1_BYTE) |
1175			     BIT(DMA_SLAVE_BUSWIDTH_2_BYTES) |
1176			     BIT(DMA_SLAVE_BUSWIDTH_4_BYTES) |
1177			     BIT(DMA_SLAVE_BUSWIDTH_8_BYTES),
1178};
1179
1180/*
1181 * The A64 binding uses the number of dma channels from the
1182 * device tree node.
1183 */
1184static struct sun6i_dma_config sun50i_a64_dma_cfg = {
1185	.clock_autogate_enable = sun6i_enable_clock_autogate_h3,
1186	.set_burst_length = sun6i_set_burst_length_h3,
1187	.set_drq          = sun6i_set_drq_a31,
1188	.set_mode         = sun6i_set_mode_a31,
1189	.src_burst_lengths = BIT(1) | BIT(4) | BIT(8) | BIT(16),
1190	.dst_burst_lengths = BIT(1) | BIT(4) | BIT(8) | BIT(16),
1191	.src_addr_widths   = BIT(DMA_SLAVE_BUSWIDTH_1_BYTE) |
1192			     BIT(DMA_SLAVE_BUSWIDTH_2_BYTES) |
1193			     BIT(DMA_SLAVE_BUSWIDTH_4_BYTES) |
1194			     BIT(DMA_SLAVE_BUSWIDTH_8_BYTES),
1195	.dst_addr_widths   = BIT(DMA_SLAVE_BUSWIDTH_1_BYTE) |
1196			     BIT(DMA_SLAVE_BUSWIDTH_2_BYTES) |
1197			     BIT(DMA_SLAVE_BUSWIDTH_4_BYTES) |
1198			     BIT(DMA_SLAVE_BUSWIDTH_8_BYTES),
1199};
1200
1201/*
1202 * The A100 binding uses the number of dma channels from the
1203 * device tree node.
1204 */
1205static struct sun6i_dma_config sun50i_a100_dma_cfg = {
1206	.clock_autogate_enable = sun6i_enable_clock_autogate_h3,
1207	.set_burst_length = sun6i_set_burst_length_h3,
1208	.set_drq          = sun6i_set_drq_h6,
1209	.set_mode         = sun6i_set_mode_h6,
1210	.src_burst_lengths = BIT(1) | BIT(4) | BIT(8) | BIT(16),
1211	.dst_burst_lengths = BIT(1) | BIT(4) | BIT(8) | BIT(16),
1212	.src_addr_widths   = BIT(DMA_SLAVE_BUSWIDTH_1_BYTE) |
1213			     BIT(DMA_SLAVE_BUSWIDTH_2_BYTES) |
1214			     BIT(DMA_SLAVE_BUSWIDTH_4_BYTES) |
1215			     BIT(DMA_SLAVE_BUSWIDTH_8_BYTES),
1216	.dst_addr_widths   = BIT(DMA_SLAVE_BUSWIDTH_1_BYTE) |
1217			     BIT(DMA_SLAVE_BUSWIDTH_2_BYTES) |
1218			     BIT(DMA_SLAVE_BUSWIDTH_4_BYTES) |
1219			     BIT(DMA_SLAVE_BUSWIDTH_8_BYTES),
1220	.has_high_addr = true,
1221	.has_mbus_clk = true,
1222};
1223
1224/*
1225 * The H6 binding uses the number of dma channels from the
1226 * device tree node.
1227 */
1228static struct sun6i_dma_config sun50i_h6_dma_cfg = {
1229	.clock_autogate_enable = sun6i_enable_clock_autogate_h3,
1230	.set_burst_length = sun6i_set_burst_length_h3,
1231	.set_drq          = sun6i_set_drq_h6,
1232	.set_mode         = sun6i_set_mode_h6,
1233	.src_burst_lengths = BIT(1) | BIT(4) | BIT(8) | BIT(16),
1234	.dst_burst_lengths = BIT(1) | BIT(4) | BIT(8) | BIT(16),
1235	.src_addr_widths   = BIT(DMA_SLAVE_BUSWIDTH_1_BYTE) |
1236			     BIT(DMA_SLAVE_BUSWIDTH_2_BYTES) |
1237			     BIT(DMA_SLAVE_BUSWIDTH_4_BYTES) |
1238			     BIT(DMA_SLAVE_BUSWIDTH_8_BYTES),
1239	.dst_addr_widths   = BIT(DMA_SLAVE_BUSWIDTH_1_BYTE) |
1240			     BIT(DMA_SLAVE_BUSWIDTH_2_BYTES) |
1241			     BIT(DMA_SLAVE_BUSWIDTH_4_BYTES) |
1242			     BIT(DMA_SLAVE_BUSWIDTH_8_BYTES),
1243	.has_mbus_clk = true,
1244};
1245
1246/*
1247 * The V3s have only 8 physical channels, a maximum DRQ port id of 23,
1248 * and a total of 24 usable source and destination endpoints.
1249 */
1250
1251static struct sun6i_dma_config sun8i_v3s_dma_cfg = {
1252	.nr_max_channels = 8,
1253	.nr_max_requests = 23,
1254	.nr_max_vchans   = 24,
1255	.clock_autogate_enable = sun6i_enable_clock_autogate_a23,
1256	.set_burst_length = sun6i_set_burst_length_a31,
1257	.set_drq          = sun6i_set_drq_a31,
1258	.set_mode         = sun6i_set_mode_a31,
1259	.src_burst_lengths = BIT(1) | BIT(8),
1260	.dst_burst_lengths = BIT(1) | BIT(8),
1261	.src_addr_widths   = BIT(DMA_SLAVE_BUSWIDTH_1_BYTE) |
1262			     BIT(DMA_SLAVE_BUSWIDTH_2_BYTES) |
1263			     BIT(DMA_SLAVE_BUSWIDTH_4_BYTES),
1264	.dst_addr_widths   = BIT(DMA_SLAVE_BUSWIDTH_1_BYTE) |
1265			     BIT(DMA_SLAVE_BUSWIDTH_2_BYTES) |
1266			     BIT(DMA_SLAVE_BUSWIDTH_4_BYTES),
1267};
1268
1269static const struct of_device_id sun6i_dma_match[] = {
1270	{ .compatible = "allwinner,sun6i-a31-dma", .data = &sun6i_a31_dma_cfg },
1271	{ .compatible = "allwinner,sun8i-a23-dma", .data = &sun8i_a23_dma_cfg },
1272	{ .compatible = "allwinner,sun8i-a83t-dma", .data = &sun8i_a83t_dma_cfg },
1273	{ .compatible = "allwinner,sun8i-h3-dma", .data = &sun8i_h3_dma_cfg },
1274	{ .compatible = "allwinner,sun8i-v3s-dma", .data = &sun8i_v3s_dma_cfg },
1275	{ .compatible = "allwinner,sun20i-d1-dma", .data = &sun50i_a100_dma_cfg },
1276	{ .compatible = "allwinner,sun50i-a64-dma", .data = &sun50i_a64_dma_cfg },
1277	{ .compatible = "allwinner,sun50i-a100-dma", .data = &sun50i_a100_dma_cfg },
1278	{ .compatible = "allwinner,sun50i-h6-dma", .data = &sun50i_h6_dma_cfg },
1279	{ /* sentinel */ }
1280};
1281MODULE_DEVICE_TABLE(of, sun6i_dma_match);
1282
1283static int sun6i_dma_probe(struct platform_device *pdev)
1284{
1285	struct device_node *np = pdev->dev.of_node;
1286	struct sun6i_dma_dev *sdc;
1287	int ret, i;
1288
1289	sdc = devm_kzalloc(&pdev->dev, sizeof(*sdc), GFP_KERNEL);
1290	if (!sdc)
1291		return -ENOMEM;
1292
1293	sdc->cfg = of_device_get_match_data(&pdev->dev);
1294	if (!sdc->cfg)
1295		return -ENODEV;
1296
1297	sdc->base = devm_platform_ioremap_resource(pdev, 0);
1298	if (IS_ERR(sdc->base))
1299		return PTR_ERR(sdc->base);
1300
1301	sdc->irq = platform_get_irq(pdev, 0);
1302	if (sdc->irq < 0)
1303		return sdc->irq;
1304
1305	sdc->clk = devm_clk_get(&pdev->dev, NULL);
1306	if (IS_ERR(sdc->clk)) {
1307		dev_err(&pdev->dev, "No clock specified\n");
1308		return PTR_ERR(sdc->clk);
1309	}
1310
1311	if (sdc->cfg->has_mbus_clk) {
1312		sdc->clk_mbus = devm_clk_get(&pdev->dev, "mbus");
1313		if (IS_ERR(sdc->clk_mbus)) {
1314			dev_err(&pdev->dev, "No mbus clock specified\n");
1315			return PTR_ERR(sdc->clk_mbus);
1316		}
1317	}
1318
1319	sdc->rstc = devm_reset_control_get(&pdev->dev, NULL);
1320	if (IS_ERR(sdc->rstc)) {
1321		dev_err(&pdev->dev, "No reset controller specified\n");
1322		return PTR_ERR(sdc->rstc);
1323	}
1324
1325	sdc->pool = dmam_pool_create(dev_name(&pdev->dev), &pdev->dev,
1326				     sizeof(struct sun6i_dma_lli), 4, 0);
1327	if (!sdc->pool) {
1328		dev_err(&pdev->dev, "No memory for descriptors dma pool\n");
1329		return -ENOMEM;
1330	}
1331
1332	platform_set_drvdata(pdev, sdc);
1333	INIT_LIST_HEAD(&sdc->pending);
1334	spin_lock_init(&sdc->lock);
1335
1336	dma_set_max_seg_size(&pdev->dev, SZ_32M - 1);
1337
1338	dma_cap_set(DMA_PRIVATE, sdc->slave.cap_mask);
1339	dma_cap_set(DMA_MEMCPY, sdc->slave.cap_mask);
1340	dma_cap_set(DMA_SLAVE, sdc->slave.cap_mask);
1341	dma_cap_set(DMA_CYCLIC, sdc->slave.cap_mask);
1342
1343	INIT_LIST_HEAD(&sdc->slave.channels);
1344	sdc->slave.device_free_chan_resources	= sun6i_dma_free_chan_resources;
1345	sdc->slave.device_tx_status		= sun6i_dma_tx_status;
1346	sdc->slave.device_issue_pending		= sun6i_dma_issue_pending;
1347	sdc->slave.device_prep_slave_sg		= sun6i_dma_prep_slave_sg;
1348	sdc->slave.device_prep_dma_memcpy	= sun6i_dma_prep_dma_memcpy;
1349	sdc->slave.device_prep_dma_cyclic	= sun6i_dma_prep_dma_cyclic;
1350	sdc->slave.copy_align			= DMAENGINE_ALIGN_4_BYTES;
1351	sdc->slave.device_config		= sun6i_dma_config;
1352	sdc->slave.device_pause			= sun6i_dma_pause;
1353	sdc->slave.device_resume		= sun6i_dma_resume;
1354	sdc->slave.device_terminate_all		= sun6i_dma_terminate_all;
1355	sdc->slave.src_addr_widths		= sdc->cfg->src_addr_widths;
1356	sdc->slave.dst_addr_widths		= sdc->cfg->dst_addr_widths;
1357	sdc->slave.directions			= BIT(DMA_DEV_TO_MEM) |
1358						  BIT(DMA_MEM_TO_DEV);
1359	sdc->slave.residue_granularity		= DMA_RESIDUE_GRANULARITY_BURST;
1360	sdc->slave.dev = &pdev->dev;
1361
1362	sdc->num_pchans = sdc->cfg->nr_max_channels;
1363	sdc->num_vchans = sdc->cfg->nr_max_vchans;
1364	sdc->max_request = sdc->cfg->nr_max_requests;
1365
1366	ret = of_property_read_u32(np, "dma-channels", &sdc->num_pchans);
1367	if (ret && !sdc->num_pchans) {
1368		dev_err(&pdev->dev, "Can't get dma-channels.\n");
1369		return ret;
1370	}
1371
1372	ret = of_property_read_u32(np, "dma-requests", &sdc->max_request);
1373	if (ret && !sdc->max_request) {
1374		dev_info(&pdev->dev, "Missing dma-requests, using %u.\n",
1375			 DMA_CHAN_MAX_DRQ_A31);
1376		sdc->max_request = DMA_CHAN_MAX_DRQ_A31;
1377	}
1378
1379	/*
1380	 * If the number of vchans is not specified, derive it from the
1381	 * highest port number, at most one channel per port and direction.
1382	 */
1383	if (!sdc->num_vchans)
1384		sdc->num_vchans = 2 * (sdc->max_request + 1);
1385
1386	sdc->pchans = devm_kcalloc(&pdev->dev, sdc->num_pchans,
1387				   sizeof(struct sun6i_pchan), GFP_KERNEL);
1388	if (!sdc->pchans)
1389		return -ENOMEM;
1390
1391	sdc->vchans = devm_kcalloc(&pdev->dev, sdc->num_vchans,
1392				   sizeof(struct sun6i_vchan), GFP_KERNEL);
1393	if (!sdc->vchans)
1394		return -ENOMEM;
1395
1396	tasklet_setup(&sdc->task, sun6i_dma_tasklet);
1397
1398	for (i = 0; i < sdc->num_pchans; i++) {
1399		struct sun6i_pchan *pchan = &sdc->pchans[i];
1400
1401		pchan->idx = i;
1402		pchan->base = sdc->base + 0x100 + i * 0x40;
1403	}
1404
1405	for (i = 0; i < sdc->num_vchans; i++) {
1406		struct sun6i_vchan *vchan = &sdc->vchans[i];
1407
1408		INIT_LIST_HEAD(&vchan->node);
1409		vchan->vc.desc_free = sun6i_dma_free_desc;
1410		vchan_init(&vchan->vc, &sdc->slave);
1411	}
1412
1413	ret = reset_control_deassert(sdc->rstc);
1414	if (ret) {
1415		dev_err(&pdev->dev, "Couldn't deassert the device from reset\n");
1416		goto err_chan_free;
1417	}
1418
1419	ret = clk_prepare_enable(sdc->clk);
1420	if (ret) {
1421		dev_err(&pdev->dev, "Couldn't enable the clock\n");
1422		goto err_reset_assert;
1423	}
1424
1425	if (sdc->cfg->has_mbus_clk) {
1426		ret = clk_prepare_enable(sdc->clk_mbus);
1427		if (ret) {
1428			dev_err(&pdev->dev, "Couldn't enable mbus clock\n");
1429			goto err_clk_disable;
1430		}
1431	}
1432
1433	ret = devm_request_irq(&pdev->dev, sdc->irq, sun6i_dma_interrupt, 0,
1434			       dev_name(&pdev->dev), sdc);
1435	if (ret) {
1436		dev_err(&pdev->dev, "Cannot request IRQ\n");
1437		goto err_mbus_clk_disable;
1438	}
1439
1440	ret = dma_async_device_register(&sdc->slave);
1441	if (ret) {
1442		dev_warn(&pdev->dev, "Failed to register DMA engine device\n");
1443		goto err_irq_disable;
1444	}
1445
1446	ret = of_dma_controller_register(pdev->dev.of_node, sun6i_dma_of_xlate,
1447					 sdc);
1448	if (ret) {
1449		dev_err(&pdev->dev, "of_dma_controller_register failed\n");
1450		goto err_dma_unregister;
1451	}
1452
1453	if (sdc->cfg->clock_autogate_enable)
1454		sdc->cfg->clock_autogate_enable(sdc);
1455
1456	return 0;
1457
1458err_dma_unregister:
1459	dma_async_device_unregister(&sdc->slave);
1460err_irq_disable:
1461	sun6i_kill_tasklet(sdc);
1462err_mbus_clk_disable:
1463	clk_disable_unprepare(sdc->clk_mbus);
1464err_clk_disable:
1465	clk_disable_unprepare(sdc->clk);
1466err_reset_assert:
1467	reset_control_assert(sdc->rstc);
1468err_chan_free:
1469	sun6i_dma_free(sdc);
1470	return ret;
1471}
1472
1473static void sun6i_dma_remove(struct platform_device *pdev)
1474{
1475	struct sun6i_dma_dev *sdc = platform_get_drvdata(pdev);
1476
1477	of_dma_controller_free(pdev->dev.of_node);
1478	dma_async_device_unregister(&sdc->slave);
1479
1480	sun6i_kill_tasklet(sdc);
1481
1482	clk_disable_unprepare(sdc->clk_mbus);
1483	clk_disable_unprepare(sdc->clk);
1484	reset_control_assert(sdc->rstc);
1485
1486	sun6i_dma_free(sdc);
1487}
1488
1489static struct platform_driver sun6i_dma_driver = {
1490	.probe		= sun6i_dma_probe,
1491	.remove_new	= sun6i_dma_remove,
1492	.driver = {
1493		.name		= "sun6i-dma",
1494		.of_match_table	= sun6i_dma_match,
1495	},
1496};
1497module_platform_driver(sun6i_dma_driver);
1498
1499MODULE_DESCRIPTION("Allwinner A31 DMA Controller Driver");
1500MODULE_AUTHOR("Sugar <shuge@allwinnertech.com>");
1501MODULE_AUTHOR("Maxime Ripard <maxime.ripard@free-electrons.com>");
1502MODULE_LICENSE("GPL");
1503