1 // SPDX-License-Identifier: GPL-2.0
2 // (C) 2017-2018 Synopsys, Inc. (www.synopsys.com)
3 
4 /*
5  * Synopsys DesignWare AXI DMA Controller driver.
6  *
7  * Author: Eugeniy Paltsev <Eugeniy.Paltsev@synopsys.com>
8  */
9 
10 #include <linux/bitops.h>
11 #include <linux/delay.h>
12 #include <linux/device.h>
13 #include <linux/dmaengine.h>
14 #include <linux/dmapool.h>
15 #include <linux/dma-mapping.h>
16 #include <linux/err.h>
17 #include <linux/interrupt.h>
18 #include <linux/io.h>
19 #include <linux/iopoll.h>
20 #include <linux/io-64-nonatomic-lo-hi.h>
21 #include <linux/kernel.h>
22 #include <linux/module.h>
23 #include <linux/of.h>
24 #include <linux/of_dma.h>
25 #include <linux/platform_device.h>
26 #include <linux/pm_runtime.h>
27 #include <linux/property.h>
28 #include <linux/slab.h>
29 #include <linux/types.h>
30 
31 #include "dw-axi-dmac.h"
32 #include "../dmaengine.h"
33 #include "../virt-dma.h"
34 
35 /*
36  * The set of bus widths supported by the DMA controller. DW AXI DMAC supports
37  * master data bus width up to 512 bits (for both AXI master interfaces), but
38  * it depends on IP block configuration.
39  */
40 #define AXI_DMA_BUSWIDTHS		  \
41 	(DMA_SLAVE_BUSWIDTH_1_BYTE	| \
42 	DMA_SLAVE_BUSWIDTH_2_BYTES	| \
43 	DMA_SLAVE_BUSWIDTH_4_BYTES	| \
44 	DMA_SLAVE_BUSWIDTH_8_BYTES	| \
45 	DMA_SLAVE_BUSWIDTH_16_BYTES	| \
46 	DMA_SLAVE_BUSWIDTH_32_BYTES	| \
47 	DMA_SLAVE_BUSWIDTH_64_BYTES)
48 
49 static inline void
50 axi_dma_iowrite32(struct axi_dma_chip *chip, u32 reg, u32 val)
51 {
52 	iowrite32(val, chip->regs + reg);
53 }
54 
55 static inline u32 axi_dma_ioread32(struct axi_dma_chip *chip, u32 reg)
56 {
57 	return ioread32(chip->regs + reg);
58 }
59 
60 static inline void
61 axi_chan_iowrite32(struct axi_dma_chan *chan, u32 reg, u32 val)
62 {
63 	iowrite32(val, chan->chan_regs + reg);
64 }
65 
66 static inline u32 axi_chan_ioread32(struct axi_dma_chan *chan, u32 reg)
67 {
68 	return ioread32(chan->chan_regs + reg);
69 }
70 
71 static inline void
72 axi_chan_iowrite64(struct axi_dma_chan *chan, u32 reg, u64 val)
73 {
74 	/*
75 	 * We split one 64 bit write for two 32 bit write as some HW doesn't
76 	 * support 64 bit access.
77 	 */
78 	iowrite32(lower_32_bits(val), chan->chan_regs + reg);
79 	iowrite32(upper_32_bits(val), chan->chan_regs + reg + 4);
80 }
81 
82 static inline void axi_chan_config_write(struct axi_dma_chan *chan,
83 					 struct axi_dma_chan_config *config)
84 {
85 	u32 cfg_lo, cfg_hi;
86 
87 	cfg_lo = (config->dst_multblk_type << CH_CFG_L_DST_MULTBLK_TYPE_POS |
88 		  config->src_multblk_type << CH_CFG_L_SRC_MULTBLK_TYPE_POS);
89 	if (chan->chip->dw->hdata->reg_map_8_channels) {
90 		cfg_hi = config->tt_fc << CH_CFG_H_TT_FC_POS |
91 			 config->hs_sel_src << CH_CFG_H_HS_SEL_SRC_POS |
92 			 config->hs_sel_dst << CH_CFG_H_HS_SEL_DST_POS |
93 			 config->src_per << CH_CFG_H_SRC_PER_POS |
94 			 config->dst_per << CH_CFG_H_DST_PER_POS |
95 			 config->prior << CH_CFG_H_PRIORITY_POS;
96 	} else {
97 		cfg_lo |= config->src_per << CH_CFG2_L_SRC_PER_POS |
98 			  config->dst_per << CH_CFG2_L_DST_PER_POS;
99 		cfg_hi = config->tt_fc << CH_CFG2_H_TT_FC_POS |
100 			 config->hs_sel_src << CH_CFG2_H_HS_SEL_SRC_POS |
101 			 config->hs_sel_dst << CH_CFG2_H_HS_SEL_DST_POS |
102 			 config->prior << CH_CFG2_H_PRIORITY_POS;
103 	}
104 	axi_chan_iowrite32(chan, CH_CFG_L, cfg_lo);
105 	axi_chan_iowrite32(chan, CH_CFG_H, cfg_hi);
106 }
107 
108 static inline void axi_dma_disable(struct axi_dma_chip *chip)
109 {
110 	u32 val;
111 
112 	val = axi_dma_ioread32(chip, DMAC_CFG);
113 	val &= ~DMAC_EN_MASK;
114 	axi_dma_iowrite32(chip, DMAC_CFG, val);
115 }
116 
117 static inline void axi_dma_enable(struct axi_dma_chip *chip)
118 {
119 	u32 val;
120 
121 	val = axi_dma_ioread32(chip, DMAC_CFG);
122 	val |= DMAC_EN_MASK;
123 	axi_dma_iowrite32(chip, DMAC_CFG, val);
124 }
125 
126 static inline void axi_dma_irq_disable(struct axi_dma_chip *chip)
127 {
128 	u32 val;
129 
130 	val = axi_dma_ioread32(chip, DMAC_CFG);
131 	val &= ~INT_EN_MASK;
132 	axi_dma_iowrite32(chip, DMAC_CFG, val);
133 }
134 
135 static inline void axi_dma_irq_enable(struct axi_dma_chip *chip)
136 {
137 	u32 val;
138 
139 	val = axi_dma_ioread32(chip, DMAC_CFG);
140 	val |= INT_EN_MASK;
141 	axi_dma_iowrite32(chip, DMAC_CFG, val);
142 }
143 
144 static inline void axi_chan_irq_disable(struct axi_dma_chan *chan, u32 irq_mask)
145 {
146 	u32 val;
147 
148 	if (likely(irq_mask == DWAXIDMAC_IRQ_ALL)) {
149 		axi_chan_iowrite32(chan, CH_INTSTATUS_ENA, DWAXIDMAC_IRQ_NONE);
150 	} else {
151 		val = axi_chan_ioread32(chan, CH_INTSTATUS_ENA);
152 		val &= ~irq_mask;
153 		axi_chan_iowrite32(chan, CH_INTSTATUS_ENA, val);
154 	}
155 }
156 
157 static inline void axi_chan_irq_set(struct axi_dma_chan *chan, u32 irq_mask)
158 {
159 	axi_chan_iowrite32(chan, CH_INTSTATUS_ENA, irq_mask);
160 }
161 
162 static inline void axi_chan_irq_sig_set(struct axi_dma_chan *chan, u32 irq_mask)
163 {
164 	axi_chan_iowrite32(chan, CH_INTSIGNAL_ENA, irq_mask);
165 }
166 
167 static inline void axi_chan_irq_clear(struct axi_dma_chan *chan, u32 irq_mask)
168 {
169 	axi_chan_iowrite32(chan, CH_INTCLEAR, irq_mask);
170 }
171 
172 static inline u32 axi_chan_irq_read(struct axi_dma_chan *chan)
173 {
174 	return axi_chan_ioread32(chan, CH_INTSTATUS);
175 }
176 
177 static inline void axi_chan_disable(struct axi_dma_chan *chan)
178 {
179 	u32 val;
180 
181 	val = axi_dma_ioread32(chan->chip, DMAC_CHEN);
182 	val &= ~(BIT(chan->id) << DMAC_CHAN_EN_SHIFT);
183 	if (chan->chip->dw->hdata->reg_map_8_channels)
184 		val |=   BIT(chan->id) << DMAC_CHAN_EN_WE_SHIFT;
185 	else
186 		val |=   BIT(chan->id) << DMAC_CHAN_EN2_WE_SHIFT;
187 	axi_dma_iowrite32(chan->chip, DMAC_CHEN, val);
188 }
189 
190 static inline void axi_chan_enable(struct axi_dma_chan *chan)
191 {
192 	u32 val;
193 
194 	val = axi_dma_ioread32(chan->chip, DMAC_CHEN);
195 	if (chan->chip->dw->hdata->reg_map_8_channels)
196 		val |= BIT(chan->id) << DMAC_CHAN_EN_SHIFT |
197 			BIT(chan->id) << DMAC_CHAN_EN_WE_SHIFT;
198 	else
199 		val |= BIT(chan->id) << DMAC_CHAN_EN_SHIFT |
200 			BIT(chan->id) << DMAC_CHAN_EN2_WE_SHIFT;
201 	axi_dma_iowrite32(chan->chip, DMAC_CHEN, val);
202 }
203 
204 static inline bool axi_chan_is_hw_enable(struct axi_dma_chan *chan)
205 {
206 	u32 val;
207 
208 	val = axi_dma_ioread32(chan->chip, DMAC_CHEN);
209 
210 	return !!(val & (BIT(chan->id) << DMAC_CHAN_EN_SHIFT));
211 }
212 
213 static void axi_dma_hw_init(struct axi_dma_chip *chip)
214 {
215 	int ret;
216 	u32 i;
217 
218 	for (i = 0; i < chip->dw->hdata->nr_channels; i++) {
219 		axi_chan_irq_disable(&chip->dw->chan[i], DWAXIDMAC_IRQ_ALL);
220 		axi_chan_disable(&chip->dw->chan[i]);
221 	}
222 	ret = dma_set_mask_and_coherent(chip->dev, DMA_BIT_MASK(64));
223 	if (ret)
224 		dev_warn(chip->dev, "Unable to set coherent mask\n");
225 }
226 
227 static u32 axi_chan_get_xfer_width(struct axi_dma_chan *chan, dma_addr_t src,
228 				   dma_addr_t dst, size_t len)
229 {
230 	u32 max_width = chan->chip->dw->hdata->m_data_width;
231 
232 	return __ffs(src | dst | len | BIT(max_width));
233 }
234 
235 static inline const char *axi_chan_name(struct axi_dma_chan *chan)
236 {
237 	return dma_chan_name(&chan->vc.chan);
238 }
239 
240 static struct axi_dma_desc *axi_desc_alloc(u32 num)
241 {
242 	struct axi_dma_desc *desc;
243 
244 	desc = kzalloc(sizeof(*desc), GFP_NOWAIT);
245 	if (!desc)
246 		return NULL;
247 
248 	desc->hw_desc = kcalloc(num, sizeof(*desc->hw_desc), GFP_NOWAIT);
249 	if (!desc->hw_desc) {
250 		kfree(desc);
251 		return NULL;
252 	}
253 
254 	return desc;
255 }
256 
257 static struct axi_dma_lli *axi_desc_get(struct axi_dma_chan *chan,
258 					dma_addr_t *addr)
259 {
260 	struct axi_dma_lli *lli;
261 	dma_addr_t phys;
262 
263 	lli = dma_pool_zalloc(chan->desc_pool, GFP_NOWAIT, &phys);
264 	if (unlikely(!lli)) {
265 		dev_err(chan2dev(chan), "%s: not enough descriptors available\n",
266 			axi_chan_name(chan));
267 		return NULL;
268 	}
269 
270 	atomic_inc(&chan->descs_allocated);
271 	*addr = phys;
272 
273 	return lli;
274 }
275 
276 static void axi_desc_put(struct axi_dma_desc *desc)
277 {
278 	struct axi_dma_chan *chan = desc->chan;
279 	int count = atomic_read(&chan->descs_allocated);
280 	struct axi_dma_hw_desc *hw_desc;
281 	int descs_put;
282 
283 	for (descs_put = 0; descs_put < count; descs_put++) {
284 		hw_desc = &desc->hw_desc[descs_put];
285 		dma_pool_free(chan->desc_pool, hw_desc->lli, hw_desc->llp);
286 	}
287 
288 	kfree(desc->hw_desc);
289 	kfree(desc);
290 	atomic_sub(descs_put, &chan->descs_allocated);
291 	dev_vdbg(chan2dev(chan), "%s: %d descs put, %d still allocated\n",
292 		axi_chan_name(chan), descs_put,
293 		atomic_read(&chan->descs_allocated));
294 }
295 
296 static void vchan_desc_put(struct virt_dma_desc *vdesc)
297 {
298 	axi_desc_put(vd_to_axi_desc(vdesc));
299 }
300 
301 static enum dma_status
302 dma_chan_tx_status(struct dma_chan *dchan, dma_cookie_t cookie,
303 		  struct dma_tx_state *txstate)
304 {
305 	struct axi_dma_chan *chan = dchan_to_axi_dma_chan(dchan);
306 	struct virt_dma_desc *vdesc;
307 	enum dma_status status;
308 	u32 completed_length;
309 	unsigned long flags;
310 	u32 completed_blocks;
311 	size_t bytes = 0;
312 	u32 length;
313 	u32 len;
314 
315 	status = dma_cookie_status(dchan, cookie, txstate);
316 	if (status == DMA_COMPLETE || !txstate)
317 		return status;
318 
319 	spin_lock_irqsave(&chan->vc.lock, flags);
320 
321 	vdesc = vchan_find_desc(&chan->vc, cookie);
322 	if (vdesc) {
323 		length = vd_to_axi_desc(vdesc)->length;
324 		completed_blocks = vd_to_axi_desc(vdesc)->completed_blocks;
325 		len = vd_to_axi_desc(vdesc)->hw_desc[0].len;
326 		completed_length = completed_blocks * len;
327 		bytes = length - completed_length;
328 	} else {
329 		bytes = vd_to_axi_desc(vdesc)->length;
330 	}
331 
332 	spin_unlock_irqrestore(&chan->vc.lock, flags);
333 	dma_set_residue(txstate, bytes);
334 
335 	return status;
336 }
337 
338 static void write_desc_llp(struct axi_dma_hw_desc *desc, dma_addr_t adr)
339 {
340 	desc->lli->llp = cpu_to_le64(adr);
341 }
342 
343 static void write_chan_llp(struct axi_dma_chan *chan, dma_addr_t adr)
344 {
345 	axi_chan_iowrite64(chan, CH_LLP, adr);
346 }
347 
348 static void dw_axi_dma_set_byte_halfword(struct axi_dma_chan *chan, bool set)
349 {
350 	u32 offset = DMAC_APB_BYTE_WR_CH_EN;
351 	u32 reg_width, val;
352 
353 	if (!chan->chip->apb_regs) {
354 		dev_dbg(chan->chip->dev, "apb_regs not initialized\n");
355 		return;
356 	}
357 
358 	reg_width = __ffs(chan->config.dst_addr_width);
359 	if (reg_width == DWAXIDMAC_TRANS_WIDTH_16)
360 		offset = DMAC_APB_HALFWORD_WR_CH_EN;
361 
362 	val = ioread32(chan->chip->apb_regs + offset);
363 
364 	if (set)
365 		val |= BIT(chan->id);
366 	else
367 		val &= ~BIT(chan->id);
368 
369 	iowrite32(val, chan->chip->apb_regs + offset);
370 }
371 /* Called in chan locked context */
372 static void axi_chan_block_xfer_start(struct axi_dma_chan *chan,
373 				      struct axi_dma_desc *first)
374 {
375 	u32 priority = chan->chip->dw->hdata->priority[chan->id];
376 	struct axi_dma_chan_config config = {};
377 	u32 irq_mask;
378 	u8 lms = 0; /* Select AXI0 master for LLI fetching */
379 
380 	if (unlikely(axi_chan_is_hw_enable(chan))) {
381 		dev_err(chan2dev(chan), "%s is non-idle!\n",
382 			axi_chan_name(chan));
383 
384 		return;
385 	}
386 
387 	axi_dma_enable(chan->chip);
388 
389 	config.dst_multblk_type = DWAXIDMAC_MBLK_TYPE_LL;
390 	config.src_multblk_type = DWAXIDMAC_MBLK_TYPE_LL;
391 	config.tt_fc = DWAXIDMAC_TT_FC_MEM_TO_MEM_DMAC;
392 	config.prior = priority;
393 	config.hs_sel_dst = DWAXIDMAC_HS_SEL_HW;
394 	config.hs_sel_src = DWAXIDMAC_HS_SEL_HW;
395 	switch (chan->direction) {
396 	case DMA_MEM_TO_DEV:
397 		dw_axi_dma_set_byte_halfword(chan, true);
398 		config.tt_fc = chan->config.device_fc ?
399 				DWAXIDMAC_TT_FC_MEM_TO_PER_DST :
400 				DWAXIDMAC_TT_FC_MEM_TO_PER_DMAC;
401 		if (chan->chip->apb_regs)
402 			config.dst_per = chan->id;
403 		else
404 			config.dst_per = chan->hw_handshake_num;
405 		break;
406 	case DMA_DEV_TO_MEM:
407 		config.tt_fc = chan->config.device_fc ?
408 				DWAXIDMAC_TT_FC_PER_TO_MEM_SRC :
409 				DWAXIDMAC_TT_FC_PER_TO_MEM_DMAC;
410 		if (chan->chip->apb_regs)
411 			config.src_per = chan->id;
412 		else
413 			config.src_per = chan->hw_handshake_num;
414 		break;
415 	default:
416 		break;
417 	}
418 	axi_chan_config_write(chan, &config);
419 
420 	write_chan_llp(chan, first->hw_desc[0].llp | lms);
421 
422 	irq_mask = DWAXIDMAC_IRQ_DMA_TRF | DWAXIDMAC_IRQ_ALL_ERR;
423 	axi_chan_irq_sig_set(chan, irq_mask);
424 
425 	/* Generate 'suspend' status but don't generate interrupt */
426 	irq_mask |= DWAXIDMAC_IRQ_SUSPENDED;
427 	axi_chan_irq_set(chan, irq_mask);
428 
429 	axi_chan_enable(chan);
430 }
431 
432 static void axi_chan_start_first_queued(struct axi_dma_chan *chan)
433 {
434 	struct axi_dma_desc *desc;
435 	struct virt_dma_desc *vd;
436 
437 	vd = vchan_next_desc(&chan->vc);
438 	if (!vd)
439 		return;
440 
441 	desc = vd_to_axi_desc(vd);
442 	dev_vdbg(chan2dev(chan), "%s: started %u\n", axi_chan_name(chan),
443 		vd->tx.cookie);
444 	axi_chan_block_xfer_start(chan, desc);
445 }
446 
447 static void dma_chan_issue_pending(struct dma_chan *dchan)
448 {
449 	struct axi_dma_chan *chan = dchan_to_axi_dma_chan(dchan);
450 	unsigned long flags;
451 
452 	spin_lock_irqsave(&chan->vc.lock, flags);
453 	if (vchan_issue_pending(&chan->vc))
454 		axi_chan_start_first_queued(chan);
455 	spin_unlock_irqrestore(&chan->vc.lock, flags);
456 }
457 
458 static void dw_axi_dma_synchronize(struct dma_chan *dchan)
459 {
460 	struct axi_dma_chan *chan = dchan_to_axi_dma_chan(dchan);
461 
462 	vchan_synchronize(&chan->vc);
463 }
464 
465 static int dma_chan_alloc_chan_resources(struct dma_chan *dchan)
466 {
467 	struct axi_dma_chan *chan = dchan_to_axi_dma_chan(dchan);
468 
469 	/* ASSERT: channel is idle */
470 	if (axi_chan_is_hw_enable(chan)) {
471 		dev_err(chan2dev(chan), "%s is non-idle!\n",
472 			axi_chan_name(chan));
473 		return -EBUSY;
474 	}
475 
476 	/* LLI address must be aligned to a 64-byte boundary */
477 	chan->desc_pool = dma_pool_create(dev_name(chan2dev(chan)),
478 					  chan->chip->dev,
479 					  sizeof(struct axi_dma_lli),
480 					  64, 0);
481 	if (!chan->desc_pool) {
482 		dev_err(chan2dev(chan), "No memory for descriptors\n");
483 		return -ENOMEM;
484 	}
485 	dev_vdbg(dchan2dev(dchan), "%s: allocating\n", axi_chan_name(chan));
486 
487 	pm_runtime_get(chan->chip->dev);
488 
489 	return 0;
490 }
491 
492 static void dma_chan_free_chan_resources(struct dma_chan *dchan)
493 {
494 	struct axi_dma_chan *chan = dchan_to_axi_dma_chan(dchan);
495 
496 	/* ASSERT: channel is idle */
497 	if (axi_chan_is_hw_enable(chan))
498 		dev_err(dchan2dev(dchan), "%s is non-idle!\n",
499 			axi_chan_name(chan));
500 
501 	axi_chan_disable(chan);
502 	axi_chan_irq_disable(chan, DWAXIDMAC_IRQ_ALL);
503 
504 	vchan_free_chan_resources(&chan->vc);
505 
506 	dma_pool_destroy(chan->desc_pool);
507 	chan->desc_pool = NULL;
508 	dev_vdbg(dchan2dev(dchan),
509 		 "%s: free resources, descriptor still allocated: %u\n",
510 		 axi_chan_name(chan), atomic_read(&chan->descs_allocated));
511 
512 	pm_runtime_put(chan->chip->dev);
513 }
514 
515 static void dw_axi_dma_set_hw_channel(struct axi_dma_chan *chan, bool set)
516 {
517 	struct axi_dma_chip *chip = chan->chip;
518 	unsigned long reg_value, val;
519 
520 	if (!chip->apb_regs) {
521 		dev_err(chip->dev, "apb_regs not initialized\n");
522 		return;
523 	}
524 
525 	/*
526 	 * An unused DMA channel has a default value of 0x3F.
527 	 * Lock the DMA channel by assign a handshake number to the channel.
528 	 * Unlock the DMA channel by assign 0x3F to the channel.
529 	 */
530 	if (set)
531 		val = chan->hw_handshake_num;
532 	else
533 		val = UNUSED_CHANNEL;
534 
535 	reg_value = lo_hi_readq(chip->apb_regs + DMAC_APB_HW_HS_SEL_0);
536 
537 	/* Channel is already allocated, set handshake as per channel ID */
538 	/* 64 bit write should handle for 8 channels */
539 
540 	reg_value &= ~(DMA_APB_HS_SEL_MASK <<
541 			(chan->id * DMA_APB_HS_SEL_BIT_SIZE));
542 	reg_value |= (val << (chan->id * DMA_APB_HS_SEL_BIT_SIZE));
543 	lo_hi_writeq(reg_value, chip->apb_regs + DMAC_APB_HW_HS_SEL_0);
544 
545 	return;
546 }
547 
548 /*
549  * If DW_axi_dmac sees CHx_CTL.ShadowReg_Or_LLI_Last bit of the fetched LLI
550  * as 1, it understands that the current block is the final block in the
551  * transfer and completes the DMA transfer operation at the end of current
552  * block transfer.
553  */
554 static void set_desc_last(struct axi_dma_hw_desc *desc)
555 {
556 	u32 val;
557 
558 	val = le32_to_cpu(desc->lli->ctl_hi);
559 	val |= CH_CTL_H_LLI_LAST;
560 	desc->lli->ctl_hi = cpu_to_le32(val);
561 }
562 
563 static void write_desc_sar(struct axi_dma_hw_desc *desc, dma_addr_t adr)
564 {
565 	desc->lli->sar = cpu_to_le64(adr);
566 }
567 
568 static void write_desc_dar(struct axi_dma_hw_desc *desc, dma_addr_t adr)
569 {
570 	desc->lli->dar = cpu_to_le64(adr);
571 }
572 
573 static void set_desc_src_master(struct axi_dma_hw_desc *desc)
574 {
575 	u32 val;
576 
577 	/* Select AXI0 for source master */
578 	val = le32_to_cpu(desc->lli->ctl_lo);
579 	val &= ~CH_CTL_L_SRC_MAST;
580 	desc->lli->ctl_lo = cpu_to_le32(val);
581 }
582 
583 static void set_desc_dest_master(struct axi_dma_hw_desc *hw_desc,
584 				 struct axi_dma_desc *desc)
585 {
586 	u32 val;
587 
588 	/* Select AXI1 for source master if available */
589 	val = le32_to_cpu(hw_desc->lli->ctl_lo);
590 	if (desc->chan->chip->dw->hdata->nr_masters > 1)
591 		val |= CH_CTL_L_DST_MAST;
592 	else
593 		val &= ~CH_CTL_L_DST_MAST;
594 
595 	hw_desc->lli->ctl_lo = cpu_to_le32(val);
596 }
597 
598 static int dw_axi_dma_set_hw_desc(struct axi_dma_chan *chan,
599 				  struct axi_dma_hw_desc *hw_desc,
600 				  dma_addr_t mem_addr, size_t len)
601 {
602 	unsigned int data_width = BIT(chan->chip->dw->hdata->m_data_width);
603 	unsigned int reg_width;
604 	unsigned int mem_width;
605 	dma_addr_t device_addr;
606 	size_t axi_block_ts;
607 	size_t block_ts;
608 	u32 ctllo, ctlhi;
609 	u32 burst_len;
610 
611 	axi_block_ts = chan->chip->dw->hdata->block_size[chan->id];
612 
613 	mem_width = __ffs(data_width | mem_addr | len);
614 	if (mem_width > DWAXIDMAC_TRANS_WIDTH_32)
615 		mem_width = DWAXIDMAC_TRANS_WIDTH_32;
616 
617 	if (!IS_ALIGNED(mem_addr, 4)) {
618 		dev_err(chan->chip->dev, "invalid buffer alignment\n");
619 		return -EINVAL;
620 	}
621 
622 	switch (chan->direction) {
623 	case DMA_MEM_TO_DEV:
624 		reg_width = __ffs(chan->config.dst_addr_width);
625 		device_addr = chan->config.dst_addr;
626 		ctllo = reg_width << CH_CTL_L_DST_WIDTH_POS |
627 			mem_width << CH_CTL_L_SRC_WIDTH_POS |
628 			DWAXIDMAC_CH_CTL_L_NOINC << CH_CTL_L_DST_INC_POS |
629 			DWAXIDMAC_CH_CTL_L_INC << CH_CTL_L_SRC_INC_POS;
630 		block_ts = len >> mem_width;
631 		break;
632 	case DMA_DEV_TO_MEM:
633 		reg_width = __ffs(chan->config.src_addr_width);
634 		device_addr = chan->config.src_addr;
635 		ctllo = reg_width << CH_CTL_L_SRC_WIDTH_POS |
636 			mem_width << CH_CTL_L_DST_WIDTH_POS |
637 			DWAXIDMAC_CH_CTL_L_INC << CH_CTL_L_DST_INC_POS |
638 			DWAXIDMAC_CH_CTL_L_NOINC << CH_CTL_L_SRC_INC_POS;
639 		block_ts = len >> reg_width;
640 		break;
641 	default:
642 		return -EINVAL;
643 	}
644 
645 	if (block_ts > axi_block_ts)
646 		return -EINVAL;
647 
648 	hw_desc->lli = axi_desc_get(chan, &hw_desc->llp);
649 	if (unlikely(!hw_desc->lli))
650 		return -ENOMEM;
651 
652 	ctlhi = CH_CTL_H_LLI_VALID;
653 
654 	if (chan->chip->dw->hdata->restrict_axi_burst_len) {
655 		burst_len = chan->chip->dw->hdata->axi_rw_burst_len;
656 		ctlhi |= CH_CTL_H_ARLEN_EN | CH_CTL_H_AWLEN_EN |
657 			 burst_len << CH_CTL_H_ARLEN_POS |
658 			 burst_len << CH_CTL_H_AWLEN_POS;
659 	}
660 
661 	hw_desc->lli->ctl_hi = cpu_to_le32(ctlhi);
662 
663 	if (chan->direction == DMA_MEM_TO_DEV) {
664 		write_desc_sar(hw_desc, mem_addr);
665 		write_desc_dar(hw_desc, device_addr);
666 	} else {
667 		write_desc_sar(hw_desc, device_addr);
668 		write_desc_dar(hw_desc, mem_addr);
669 	}
670 
671 	hw_desc->lli->block_ts_lo = cpu_to_le32(block_ts - 1);
672 
673 	ctllo |= DWAXIDMAC_BURST_TRANS_LEN_4 << CH_CTL_L_DST_MSIZE_POS |
674 		 DWAXIDMAC_BURST_TRANS_LEN_4 << CH_CTL_L_SRC_MSIZE_POS;
675 	hw_desc->lli->ctl_lo = cpu_to_le32(ctllo);
676 
677 	set_desc_src_master(hw_desc);
678 
679 	hw_desc->len = len;
680 	return 0;
681 }
682 
683 static size_t calculate_block_len(struct axi_dma_chan *chan,
684 				  dma_addr_t dma_addr, size_t buf_len,
685 				  enum dma_transfer_direction direction)
686 {
687 	u32 data_width, reg_width, mem_width;
688 	size_t axi_block_ts, block_len;
689 
690 	axi_block_ts = chan->chip->dw->hdata->block_size[chan->id];
691 
692 	switch (direction) {
693 	case DMA_MEM_TO_DEV:
694 		data_width = BIT(chan->chip->dw->hdata->m_data_width);
695 		mem_width = __ffs(data_width | dma_addr | buf_len);
696 		if (mem_width > DWAXIDMAC_TRANS_WIDTH_32)
697 			mem_width = DWAXIDMAC_TRANS_WIDTH_32;
698 
699 		block_len = axi_block_ts << mem_width;
700 		break;
701 	case DMA_DEV_TO_MEM:
702 		reg_width = __ffs(chan->config.src_addr_width);
703 		block_len = axi_block_ts << reg_width;
704 		break;
705 	default:
706 		block_len = 0;
707 	}
708 
709 	return block_len;
710 }
711 
712 static struct dma_async_tx_descriptor *
713 dw_axi_dma_chan_prep_cyclic(struct dma_chan *dchan, dma_addr_t dma_addr,
714 			    size_t buf_len, size_t period_len,
715 			    enum dma_transfer_direction direction,
716 			    unsigned long flags)
717 {
718 	struct axi_dma_chan *chan = dchan_to_axi_dma_chan(dchan);
719 	struct axi_dma_hw_desc *hw_desc = NULL;
720 	struct axi_dma_desc *desc = NULL;
721 	dma_addr_t src_addr = dma_addr;
722 	u32 num_periods, num_segments;
723 	size_t axi_block_len;
724 	u32 total_segments;
725 	u32 segment_len;
726 	unsigned int i;
727 	int status;
728 	u64 llp = 0;
729 	u8 lms = 0; /* Select AXI0 master for LLI fetching */
730 
731 	num_periods = buf_len / period_len;
732 
733 	axi_block_len = calculate_block_len(chan, dma_addr, buf_len, direction);
734 	if (axi_block_len == 0)
735 		return NULL;
736 
737 	num_segments = DIV_ROUND_UP(period_len, axi_block_len);
738 	segment_len = DIV_ROUND_UP(period_len, num_segments);
739 
740 	total_segments = num_periods * num_segments;
741 
742 	desc = axi_desc_alloc(total_segments);
743 	if (unlikely(!desc))
744 		goto err_desc_get;
745 
746 	chan->direction = direction;
747 	desc->chan = chan;
748 	chan->cyclic = true;
749 	desc->length = 0;
750 	desc->period_len = period_len;
751 
752 	for (i = 0; i < total_segments; i++) {
753 		hw_desc = &desc->hw_desc[i];
754 
755 		status = dw_axi_dma_set_hw_desc(chan, hw_desc, src_addr,
756 						segment_len);
757 		if (status < 0)
758 			goto err_desc_get;
759 
760 		desc->length += hw_desc->len;
761 		/* Set end-of-link to the linked descriptor, so that cyclic
762 		 * callback function can be triggered during interrupt.
763 		 */
764 		set_desc_last(hw_desc);
765 
766 		src_addr += segment_len;
767 	}
768 
769 	llp = desc->hw_desc[0].llp;
770 
771 	/* Managed transfer list */
772 	do {
773 		hw_desc = &desc->hw_desc[--total_segments];
774 		write_desc_llp(hw_desc, llp | lms);
775 		llp = hw_desc->llp;
776 	} while (total_segments);
777 
778 	dw_axi_dma_set_hw_channel(chan, true);
779 
780 	return vchan_tx_prep(&chan->vc, &desc->vd, flags);
781 
782 err_desc_get:
783 	if (desc)
784 		axi_desc_put(desc);
785 
786 	return NULL;
787 }
788 
789 static struct dma_async_tx_descriptor *
790 dw_axi_dma_chan_prep_slave_sg(struct dma_chan *dchan, struct scatterlist *sgl,
791 			      unsigned int sg_len,
792 			      enum dma_transfer_direction direction,
793 			      unsigned long flags, void *context)
794 {
795 	struct axi_dma_chan *chan = dchan_to_axi_dma_chan(dchan);
796 	struct axi_dma_hw_desc *hw_desc = NULL;
797 	struct axi_dma_desc *desc = NULL;
798 	u32 num_segments, segment_len;
799 	unsigned int loop = 0;
800 	struct scatterlist *sg;
801 	size_t axi_block_len;
802 	u32 len, num_sgs = 0;
803 	unsigned int i;
804 	dma_addr_t mem;
805 	int status;
806 	u64 llp = 0;
807 	u8 lms = 0; /* Select AXI0 master for LLI fetching */
808 
809 	if (unlikely(!is_slave_direction(direction) || !sg_len))
810 		return NULL;
811 
812 	mem = sg_dma_address(sgl);
813 	len = sg_dma_len(sgl);
814 
815 	axi_block_len = calculate_block_len(chan, mem, len, direction);
816 	if (axi_block_len == 0)
817 		return NULL;
818 
819 	for_each_sg(sgl, sg, sg_len, i)
820 		num_sgs += DIV_ROUND_UP(sg_dma_len(sg), axi_block_len);
821 
822 	desc = axi_desc_alloc(num_sgs);
823 	if (unlikely(!desc))
824 		goto err_desc_get;
825 
826 	desc->chan = chan;
827 	desc->length = 0;
828 	chan->direction = direction;
829 
830 	for_each_sg(sgl, sg, sg_len, i) {
831 		mem = sg_dma_address(sg);
832 		len = sg_dma_len(sg);
833 		num_segments = DIV_ROUND_UP(sg_dma_len(sg), axi_block_len);
834 		segment_len = DIV_ROUND_UP(sg_dma_len(sg), num_segments);
835 
836 		do {
837 			hw_desc = &desc->hw_desc[loop++];
838 			status = dw_axi_dma_set_hw_desc(chan, hw_desc, mem, segment_len);
839 			if (status < 0)
840 				goto err_desc_get;
841 
842 			desc->length += hw_desc->len;
843 			len -= segment_len;
844 			mem += segment_len;
845 		} while (len >= segment_len);
846 	}
847 
848 	/* Set end-of-link to the last link descriptor of list */
849 	set_desc_last(&desc->hw_desc[num_sgs - 1]);
850 
851 	/* Managed transfer list */
852 	do {
853 		hw_desc = &desc->hw_desc[--num_sgs];
854 		write_desc_llp(hw_desc, llp | lms);
855 		llp = hw_desc->llp;
856 	} while (num_sgs);
857 
858 	dw_axi_dma_set_hw_channel(chan, true);
859 
860 	return vchan_tx_prep(&chan->vc, &desc->vd, flags);
861 
862 err_desc_get:
863 	if (desc)
864 		axi_desc_put(desc);
865 
866 	return NULL;
867 }
868 
869 static struct dma_async_tx_descriptor *
870 dma_chan_prep_dma_memcpy(struct dma_chan *dchan, dma_addr_t dst_adr,
871 			 dma_addr_t src_adr, size_t len, unsigned long flags)
872 {
873 	struct axi_dma_chan *chan = dchan_to_axi_dma_chan(dchan);
874 	size_t block_ts, max_block_ts, xfer_len;
875 	struct axi_dma_hw_desc *hw_desc = NULL;
876 	struct axi_dma_desc *desc = NULL;
877 	u32 xfer_width, reg, num;
878 	u64 llp = 0;
879 	u8 lms = 0; /* Select AXI0 master for LLI fetching */
880 
881 	dev_dbg(chan2dev(chan), "%s: memcpy: src: %pad dst: %pad length: %zd flags: %#lx",
882 		axi_chan_name(chan), &src_adr, &dst_adr, len, flags);
883 
884 	max_block_ts = chan->chip->dw->hdata->block_size[chan->id];
885 	xfer_width = axi_chan_get_xfer_width(chan, src_adr, dst_adr, len);
886 	num = DIV_ROUND_UP(len, max_block_ts << xfer_width);
887 	desc = axi_desc_alloc(num);
888 	if (unlikely(!desc))
889 		goto err_desc_get;
890 
891 	desc->chan = chan;
892 	num = 0;
893 	desc->length = 0;
894 	while (len) {
895 		xfer_len = len;
896 
897 		hw_desc = &desc->hw_desc[num];
898 		/*
899 		 * Take care for the alignment.
900 		 * Actually source and destination widths can be different, but
901 		 * make them same to be simpler.
902 		 */
903 		xfer_width = axi_chan_get_xfer_width(chan, src_adr, dst_adr, xfer_len);
904 
905 		/*
906 		 * block_ts indicates the total number of data of width
907 		 * to be transferred in a DMA block transfer.
908 		 * BLOCK_TS register should be set to block_ts - 1
909 		 */
910 		block_ts = xfer_len >> xfer_width;
911 		if (block_ts > max_block_ts) {
912 			block_ts = max_block_ts;
913 			xfer_len = max_block_ts << xfer_width;
914 		}
915 
916 		hw_desc->lli = axi_desc_get(chan, &hw_desc->llp);
917 		if (unlikely(!hw_desc->lli))
918 			goto err_desc_get;
919 
920 		write_desc_sar(hw_desc, src_adr);
921 		write_desc_dar(hw_desc, dst_adr);
922 		hw_desc->lli->block_ts_lo = cpu_to_le32(block_ts - 1);
923 
924 		reg = CH_CTL_H_LLI_VALID;
925 		if (chan->chip->dw->hdata->restrict_axi_burst_len) {
926 			u32 burst_len = chan->chip->dw->hdata->axi_rw_burst_len;
927 
928 			reg |= (CH_CTL_H_ARLEN_EN |
929 				burst_len << CH_CTL_H_ARLEN_POS |
930 				CH_CTL_H_AWLEN_EN |
931 				burst_len << CH_CTL_H_AWLEN_POS);
932 		}
933 		hw_desc->lli->ctl_hi = cpu_to_le32(reg);
934 
935 		reg = (DWAXIDMAC_BURST_TRANS_LEN_4 << CH_CTL_L_DST_MSIZE_POS |
936 		       DWAXIDMAC_BURST_TRANS_LEN_4 << CH_CTL_L_SRC_MSIZE_POS |
937 		       xfer_width << CH_CTL_L_DST_WIDTH_POS |
938 		       xfer_width << CH_CTL_L_SRC_WIDTH_POS |
939 		       DWAXIDMAC_CH_CTL_L_INC << CH_CTL_L_DST_INC_POS |
940 		       DWAXIDMAC_CH_CTL_L_INC << CH_CTL_L_SRC_INC_POS);
941 		hw_desc->lli->ctl_lo = cpu_to_le32(reg);
942 
943 		set_desc_src_master(hw_desc);
944 		set_desc_dest_master(hw_desc, desc);
945 
946 		hw_desc->len = xfer_len;
947 		desc->length += hw_desc->len;
948 		/* update the length and addresses for the next loop cycle */
949 		len -= xfer_len;
950 		dst_adr += xfer_len;
951 		src_adr += xfer_len;
952 		num++;
953 	}
954 
955 	/* Set end-of-link to the last link descriptor of list */
956 	set_desc_last(&desc->hw_desc[num - 1]);
957 	/* Managed transfer list */
958 	do {
959 		hw_desc = &desc->hw_desc[--num];
960 		write_desc_llp(hw_desc, llp | lms);
961 		llp = hw_desc->llp;
962 	} while (num);
963 
964 	return vchan_tx_prep(&chan->vc, &desc->vd, flags);
965 
966 err_desc_get:
967 	if (desc)
968 		axi_desc_put(desc);
969 	return NULL;
970 }
971 
972 static int dw_axi_dma_chan_slave_config(struct dma_chan *dchan,
973 					struct dma_slave_config *config)
974 {
975 	struct axi_dma_chan *chan = dchan_to_axi_dma_chan(dchan);
976 
977 	memcpy(&chan->config, config, sizeof(*config));
978 
979 	return 0;
980 }
981 
982 static void axi_chan_dump_lli(struct axi_dma_chan *chan,
983 			      struct axi_dma_hw_desc *desc)
984 {
985 	if (!desc->lli) {
986 		dev_err(dchan2dev(&chan->vc.chan), "NULL LLI\n");
987 		return;
988 	}
989 
990 	dev_err(dchan2dev(&chan->vc.chan),
991 		"SAR: 0x%llx DAR: 0x%llx LLP: 0x%llx BTS 0x%x CTL: 0x%x:%08x",
992 		le64_to_cpu(desc->lli->sar),
993 		le64_to_cpu(desc->lli->dar),
994 		le64_to_cpu(desc->lli->llp),
995 		le32_to_cpu(desc->lli->block_ts_lo),
996 		le32_to_cpu(desc->lli->ctl_hi),
997 		le32_to_cpu(desc->lli->ctl_lo));
998 }
999 
1000 static void axi_chan_list_dump_lli(struct axi_dma_chan *chan,
1001 				   struct axi_dma_desc *desc_head)
1002 {
1003 	int count = atomic_read(&chan->descs_allocated);
1004 	int i;
1005 
1006 	for (i = 0; i < count; i++)
1007 		axi_chan_dump_lli(chan, &desc_head->hw_desc[i]);
1008 }
1009 
1010 static noinline void axi_chan_handle_err(struct axi_dma_chan *chan, u32 status)
1011 {
1012 	struct virt_dma_desc *vd;
1013 	unsigned long flags;
1014 
1015 	spin_lock_irqsave(&chan->vc.lock, flags);
1016 
1017 	axi_chan_disable(chan);
1018 
1019 	/* The bad descriptor currently is in the head of vc list */
1020 	vd = vchan_next_desc(&chan->vc);
1021 	if (!vd) {
1022 		dev_err(chan2dev(chan), "BUG: %s, IRQ with no descriptors\n",
1023 			axi_chan_name(chan));
1024 		goto out;
1025 	}
1026 	/* Remove the completed descriptor from issued list */
1027 	list_del(&vd->node);
1028 
1029 	/* WARN about bad descriptor */
1030 	dev_err(chan2dev(chan),
1031 		"Bad descriptor submitted for %s, cookie: %d, irq: 0x%08x\n",
1032 		axi_chan_name(chan), vd->tx.cookie, status);
1033 	axi_chan_list_dump_lli(chan, vd_to_axi_desc(vd));
1034 
1035 	vchan_cookie_complete(vd);
1036 
1037 	/* Try to restart the controller */
1038 	axi_chan_start_first_queued(chan);
1039 
1040 out:
1041 	spin_unlock_irqrestore(&chan->vc.lock, flags);
1042 }
1043 
1044 static void axi_chan_block_xfer_complete(struct axi_dma_chan *chan)
1045 {
1046 	int count = atomic_read(&chan->descs_allocated);
1047 	struct axi_dma_hw_desc *hw_desc;
1048 	struct axi_dma_desc *desc;
1049 	struct virt_dma_desc *vd;
1050 	unsigned long flags;
1051 	u64 llp;
1052 	int i;
1053 
1054 	spin_lock_irqsave(&chan->vc.lock, flags);
1055 	if (unlikely(axi_chan_is_hw_enable(chan))) {
1056 		dev_err(chan2dev(chan), "BUG: %s caught DWAXIDMAC_IRQ_DMA_TRF, but channel not idle!\n",
1057 			axi_chan_name(chan));
1058 		axi_chan_disable(chan);
1059 	}
1060 
1061 	/* The completed descriptor currently is in the head of vc list */
1062 	vd = vchan_next_desc(&chan->vc);
1063 	if (!vd) {
1064 		dev_err(chan2dev(chan), "BUG: %s, IRQ with no descriptors\n",
1065 			axi_chan_name(chan));
1066 		goto out;
1067 	}
1068 
1069 	if (chan->cyclic) {
1070 		desc = vd_to_axi_desc(vd);
1071 		if (desc) {
1072 			llp = lo_hi_readq(chan->chan_regs + CH_LLP);
1073 			for (i = 0; i < count; i++) {
1074 				hw_desc = &desc->hw_desc[i];
1075 				if (hw_desc->llp == llp) {
1076 					axi_chan_irq_clear(chan, hw_desc->lli->status_lo);
1077 					hw_desc->lli->ctl_hi |= CH_CTL_H_LLI_VALID;
1078 					desc->completed_blocks = i;
1079 
1080 					if (((hw_desc->len * (i + 1)) % desc->period_len) == 0)
1081 						vchan_cyclic_callback(vd);
1082 					break;
1083 				}
1084 			}
1085 
1086 			axi_chan_enable(chan);
1087 		}
1088 	} else {
1089 		/* Remove the completed descriptor from issued list before completing */
1090 		list_del(&vd->node);
1091 		vchan_cookie_complete(vd);
1092 
1093 		/* Submit queued descriptors after processing the completed ones */
1094 		axi_chan_start_first_queued(chan);
1095 	}
1096 
1097 out:
1098 	spin_unlock_irqrestore(&chan->vc.lock, flags);
1099 }
1100 
1101 static irqreturn_t dw_axi_dma_interrupt(int irq, void *dev_id)
1102 {
1103 	struct axi_dma_chip *chip = dev_id;
1104 	struct dw_axi_dma *dw = chip->dw;
1105 	struct axi_dma_chan *chan;
1106 
1107 	u32 status, i;
1108 
1109 	/* Disable DMAC interrupts. We'll enable them after processing channels */
1110 	axi_dma_irq_disable(chip);
1111 
1112 	/* Poll, clear and process every channel interrupt status */
1113 	for (i = 0; i < dw->hdata->nr_channels; i++) {
1114 		chan = &dw->chan[i];
1115 		status = axi_chan_irq_read(chan);
1116 		axi_chan_irq_clear(chan, status);
1117 
1118 		dev_vdbg(chip->dev, "%s %u IRQ status: 0x%08x\n",
1119 			axi_chan_name(chan), i, status);
1120 
1121 		if (status & DWAXIDMAC_IRQ_ALL_ERR)
1122 			axi_chan_handle_err(chan, status);
1123 		else if (status & DWAXIDMAC_IRQ_DMA_TRF)
1124 			axi_chan_block_xfer_complete(chan);
1125 	}
1126 
1127 	/* Re-enable interrupts */
1128 	axi_dma_irq_enable(chip);
1129 
1130 	return IRQ_HANDLED;
1131 }
1132 
1133 static int dma_chan_terminate_all(struct dma_chan *dchan)
1134 {
1135 	struct axi_dma_chan *chan = dchan_to_axi_dma_chan(dchan);
1136 	u32 chan_active = BIT(chan->id) << DMAC_CHAN_EN_SHIFT;
1137 	unsigned long flags;
1138 	u32 val;
1139 	int ret;
1140 	LIST_HEAD(head);
1141 
1142 	axi_chan_disable(chan);
1143 
1144 	ret = readl_poll_timeout_atomic(chan->chip->regs + DMAC_CHEN, val,
1145 					!(val & chan_active), 1000, 10000);
1146 	if (ret == -ETIMEDOUT)
1147 		dev_warn(dchan2dev(dchan),
1148 			 "%s failed to stop\n", axi_chan_name(chan));
1149 
1150 	if (chan->direction != DMA_MEM_TO_MEM)
1151 		dw_axi_dma_set_hw_channel(chan, false);
1152 	if (chan->direction == DMA_MEM_TO_DEV)
1153 		dw_axi_dma_set_byte_halfword(chan, false);
1154 
1155 	spin_lock_irqsave(&chan->vc.lock, flags);
1156 
1157 	vchan_get_all_descriptors(&chan->vc, &head);
1158 
1159 	chan->cyclic = false;
1160 	spin_unlock_irqrestore(&chan->vc.lock, flags);
1161 
1162 	vchan_dma_desc_free_list(&chan->vc, &head);
1163 
1164 	dev_vdbg(dchan2dev(dchan), "terminated: %s\n", axi_chan_name(chan));
1165 
1166 	return 0;
1167 }
1168 
1169 static int dma_chan_pause(struct dma_chan *dchan)
1170 {
1171 	struct axi_dma_chan *chan = dchan_to_axi_dma_chan(dchan);
1172 	unsigned long flags;
1173 	unsigned int timeout = 20; /* timeout iterations */
1174 	u32 val;
1175 
1176 	spin_lock_irqsave(&chan->vc.lock, flags);
1177 
1178 	if (chan->chip->dw->hdata->reg_map_8_channels) {
1179 		val = axi_dma_ioread32(chan->chip, DMAC_CHEN);
1180 		val |= BIT(chan->id) << DMAC_CHAN_SUSP_SHIFT |
1181 			BIT(chan->id) << DMAC_CHAN_SUSP_WE_SHIFT;
1182 		axi_dma_iowrite32(chan->chip, DMAC_CHEN, val);
1183 	} else {
1184 		val = axi_dma_ioread32(chan->chip, DMAC_CHSUSPREG);
1185 		val |= BIT(chan->id) << DMAC_CHAN_SUSP2_SHIFT |
1186 			BIT(chan->id) << DMAC_CHAN_SUSP2_WE_SHIFT;
1187 		axi_dma_iowrite32(chan->chip, DMAC_CHSUSPREG, val);
1188 	}
1189 
1190 	do  {
1191 		if (axi_chan_irq_read(chan) & DWAXIDMAC_IRQ_SUSPENDED)
1192 			break;
1193 
1194 		udelay(2);
1195 	} while (--timeout);
1196 
1197 	axi_chan_irq_clear(chan, DWAXIDMAC_IRQ_SUSPENDED);
1198 
1199 	chan->is_paused = true;
1200 
1201 	spin_unlock_irqrestore(&chan->vc.lock, flags);
1202 
1203 	return timeout ? 0 : -EAGAIN;
1204 }
1205 
1206 /* Called in chan locked context */
1207 static inline void axi_chan_resume(struct axi_dma_chan *chan)
1208 {
1209 	u32 val;
1210 
1211 	if (chan->chip->dw->hdata->reg_map_8_channels) {
1212 		val = axi_dma_ioread32(chan->chip, DMAC_CHEN);
1213 		val &= ~(BIT(chan->id) << DMAC_CHAN_SUSP_SHIFT);
1214 		val |=  (BIT(chan->id) << DMAC_CHAN_SUSP_WE_SHIFT);
1215 		axi_dma_iowrite32(chan->chip, DMAC_CHEN, val);
1216 	} else {
1217 		val = axi_dma_ioread32(chan->chip, DMAC_CHSUSPREG);
1218 		val &= ~(BIT(chan->id) << DMAC_CHAN_SUSP2_SHIFT);
1219 		val |=  (BIT(chan->id) << DMAC_CHAN_SUSP2_WE_SHIFT);
1220 		axi_dma_iowrite32(chan->chip, DMAC_CHSUSPREG, val);
1221 	}
1222 
1223 	chan->is_paused = false;
1224 }
1225 
1226 static int dma_chan_resume(struct dma_chan *dchan)
1227 {
1228 	struct axi_dma_chan *chan = dchan_to_axi_dma_chan(dchan);
1229 	unsigned long flags;
1230 
1231 	spin_lock_irqsave(&chan->vc.lock, flags);
1232 
1233 	if (chan->is_paused)
1234 		axi_chan_resume(chan);
1235 
1236 	spin_unlock_irqrestore(&chan->vc.lock, flags);
1237 
1238 	return 0;
1239 }
1240 
1241 static int axi_dma_suspend(struct axi_dma_chip *chip)
1242 {
1243 	axi_dma_irq_disable(chip);
1244 	axi_dma_disable(chip);
1245 
1246 	clk_disable_unprepare(chip->core_clk);
1247 	clk_disable_unprepare(chip->cfgr_clk);
1248 
1249 	return 0;
1250 }
1251 
1252 static int axi_dma_resume(struct axi_dma_chip *chip)
1253 {
1254 	int ret;
1255 
1256 	ret = clk_prepare_enable(chip->cfgr_clk);
1257 	if (ret < 0)
1258 		return ret;
1259 
1260 	ret = clk_prepare_enable(chip->core_clk);
1261 	if (ret < 0)
1262 		return ret;
1263 
1264 	axi_dma_enable(chip);
1265 	axi_dma_irq_enable(chip);
1266 
1267 	return 0;
1268 }
1269 
1270 static int __maybe_unused axi_dma_runtime_suspend(struct device *dev)
1271 {
1272 	struct axi_dma_chip *chip = dev_get_drvdata(dev);
1273 
1274 	return axi_dma_suspend(chip);
1275 }
1276 
1277 static int __maybe_unused axi_dma_runtime_resume(struct device *dev)
1278 {
1279 	struct axi_dma_chip *chip = dev_get_drvdata(dev);
1280 
1281 	return axi_dma_resume(chip);
1282 }
1283 
1284 static struct dma_chan *dw_axi_dma_of_xlate(struct of_phandle_args *dma_spec,
1285 					    struct of_dma *ofdma)
1286 {
1287 	struct dw_axi_dma *dw = ofdma->of_dma_data;
1288 	struct axi_dma_chan *chan;
1289 	struct dma_chan *dchan;
1290 
1291 	dchan = dma_get_any_slave_channel(&dw->dma);
1292 	if (!dchan)
1293 		return NULL;
1294 
1295 	chan = dchan_to_axi_dma_chan(dchan);
1296 	chan->hw_handshake_num = dma_spec->args[0];
1297 	return dchan;
1298 }
1299 
1300 static int parse_device_properties(struct axi_dma_chip *chip)
1301 {
1302 	struct device *dev = chip->dev;
1303 	u32 tmp, carr[DMAC_MAX_CHANNELS];
1304 	int ret;
1305 
1306 	ret = device_property_read_u32(dev, "dma-channels", &tmp);
1307 	if (ret)
1308 		return ret;
1309 	if (tmp == 0 || tmp > DMAC_MAX_CHANNELS)
1310 		return -EINVAL;
1311 
1312 	chip->dw->hdata->nr_channels = tmp;
1313 	if (tmp <= DMA_REG_MAP_CH_REF)
1314 		chip->dw->hdata->reg_map_8_channels = true;
1315 
1316 	ret = device_property_read_u32(dev, "snps,dma-masters", &tmp);
1317 	if (ret)
1318 		return ret;
1319 	if (tmp == 0 || tmp > DMAC_MAX_MASTERS)
1320 		return -EINVAL;
1321 
1322 	chip->dw->hdata->nr_masters = tmp;
1323 
1324 	ret = device_property_read_u32(dev, "snps,data-width", &tmp);
1325 	if (ret)
1326 		return ret;
1327 	if (tmp > DWAXIDMAC_TRANS_WIDTH_MAX)
1328 		return -EINVAL;
1329 
1330 	chip->dw->hdata->m_data_width = tmp;
1331 
1332 	ret = device_property_read_u32_array(dev, "snps,block-size", carr,
1333 					     chip->dw->hdata->nr_channels);
1334 	if (ret)
1335 		return ret;
1336 	for (tmp = 0; tmp < chip->dw->hdata->nr_channels; tmp++) {
1337 		if (carr[tmp] == 0 || carr[tmp] > DMAC_MAX_BLK_SIZE)
1338 			return -EINVAL;
1339 
1340 		chip->dw->hdata->block_size[tmp] = carr[tmp];
1341 	}
1342 
1343 	ret = device_property_read_u32_array(dev, "snps,priority", carr,
1344 					     chip->dw->hdata->nr_channels);
1345 	if (ret)
1346 		return ret;
1347 	/* Priority value must be programmed within [0:nr_channels-1] range */
1348 	for (tmp = 0; tmp < chip->dw->hdata->nr_channels; tmp++) {
1349 		if (carr[tmp] >= chip->dw->hdata->nr_channels)
1350 			return -EINVAL;
1351 
1352 		chip->dw->hdata->priority[tmp] = carr[tmp];
1353 	}
1354 
1355 	/* axi-max-burst-len is optional property */
1356 	ret = device_property_read_u32(dev, "snps,axi-max-burst-len", &tmp);
1357 	if (!ret) {
1358 		if (tmp > DWAXIDMAC_ARWLEN_MAX + 1)
1359 			return -EINVAL;
1360 		if (tmp < DWAXIDMAC_ARWLEN_MIN + 1)
1361 			return -EINVAL;
1362 
1363 		chip->dw->hdata->restrict_axi_burst_len = true;
1364 		chip->dw->hdata->axi_rw_burst_len = tmp;
1365 	}
1366 
1367 	return 0;
1368 }
1369 
1370 static int dw_probe(struct platform_device *pdev)
1371 {
1372 	struct device_node *node = pdev->dev.of_node;
1373 	struct axi_dma_chip *chip;
1374 	struct resource *mem;
1375 	struct dw_axi_dma *dw;
1376 	struct dw_axi_dma_hcfg *hdata;
1377 	u32 i;
1378 	int ret;
1379 
1380 	chip = devm_kzalloc(&pdev->dev, sizeof(*chip), GFP_KERNEL);
1381 	if (!chip)
1382 		return -ENOMEM;
1383 
1384 	dw = devm_kzalloc(&pdev->dev, sizeof(*dw), GFP_KERNEL);
1385 	if (!dw)
1386 		return -ENOMEM;
1387 
1388 	hdata = devm_kzalloc(&pdev->dev, sizeof(*hdata), GFP_KERNEL);
1389 	if (!hdata)
1390 		return -ENOMEM;
1391 
1392 	chip->dw = dw;
1393 	chip->dev = &pdev->dev;
1394 	chip->dw->hdata = hdata;
1395 
1396 	chip->irq = platform_get_irq(pdev, 0);
1397 	if (chip->irq < 0)
1398 		return chip->irq;
1399 
1400 	mem = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1401 	chip->regs = devm_ioremap_resource(chip->dev, mem);
1402 	if (IS_ERR(chip->regs))
1403 		return PTR_ERR(chip->regs);
1404 
1405 	if (of_device_is_compatible(node, "intel,kmb-axi-dma")) {
1406 		chip->apb_regs = devm_platform_ioremap_resource(pdev, 1);
1407 		if (IS_ERR(chip->apb_regs))
1408 			return PTR_ERR(chip->apb_regs);
1409 	}
1410 
1411 	chip->core_clk = devm_clk_get(chip->dev, "core-clk");
1412 	if (IS_ERR(chip->core_clk))
1413 		return PTR_ERR(chip->core_clk);
1414 
1415 	chip->cfgr_clk = devm_clk_get(chip->dev, "cfgr-clk");
1416 	if (IS_ERR(chip->cfgr_clk))
1417 		return PTR_ERR(chip->cfgr_clk);
1418 
1419 	ret = parse_device_properties(chip);
1420 	if (ret)
1421 		return ret;
1422 
1423 	dw->chan = devm_kcalloc(chip->dev, hdata->nr_channels,
1424 				sizeof(*dw->chan), GFP_KERNEL);
1425 	if (!dw->chan)
1426 		return -ENOMEM;
1427 
1428 	ret = devm_request_irq(chip->dev, chip->irq, dw_axi_dma_interrupt,
1429 			       IRQF_SHARED, KBUILD_MODNAME, chip);
1430 	if (ret)
1431 		return ret;
1432 
1433 	INIT_LIST_HEAD(&dw->dma.channels);
1434 	for (i = 0; i < hdata->nr_channels; i++) {
1435 		struct axi_dma_chan *chan = &dw->chan[i];
1436 
1437 		chan->chip = chip;
1438 		chan->id = i;
1439 		chan->chan_regs = chip->regs + COMMON_REG_LEN + i * CHAN_REG_LEN;
1440 		atomic_set(&chan->descs_allocated, 0);
1441 
1442 		chan->vc.desc_free = vchan_desc_put;
1443 		vchan_init(&chan->vc, &dw->dma);
1444 	}
1445 
1446 	/* Set capabilities */
1447 	dma_cap_set(DMA_MEMCPY, dw->dma.cap_mask);
1448 	dma_cap_set(DMA_SLAVE, dw->dma.cap_mask);
1449 	dma_cap_set(DMA_CYCLIC, dw->dma.cap_mask);
1450 
1451 	/* DMA capabilities */
1452 	dw->dma.chancnt = hdata->nr_channels;
1453 	dw->dma.max_burst = hdata->axi_rw_burst_len;
1454 	dw->dma.src_addr_widths = AXI_DMA_BUSWIDTHS;
1455 	dw->dma.dst_addr_widths = AXI_DMA_BUSWIDTHS;
1456 	dw->dma.directions = BIT(DMA_MEM_TO_MEM);
1457 	dw->dma.directions |= BIT(DMA_MEM_TO_DEV) | BIT(DMA_DEV_TO_MEM);
1458 	dw->dma.residue_granularity = DMA_RESIDUE_GRANULARITY_BURST;
1459 
1460 	dw->dma.dev = chip->dev;
1461 	dw->dma.device_tx_status = dma_chan_tx_status;
1462 	dw->dma.device_issue_pending = dma_chan_issue_pending;
1463 	dw->dma.device_terminate_all = dma_chan_terminate_all;
1464 	dw->dma.device_pause = dma_chan_pause;
1465 	dw->dma.device_resume = dma_chan_resume;
1466 
1467 	dw->dma.device_alloc_chan_resources = dma_chan_alloc_chan_resources;
1468 	dw->dma.device_free_chan_resources = dma_chan_free_chan_resources;
1469 
1470 	dw->dma.device_prep_dma_memcpy = dma_chan_prep_dma_memcpy;
1471 	dw->dma.device_synchronize = dw_axi_dma_synchronize;
1472 	dw->dma.device_config = dw_axi_dma_chan_slave_config;
1473 	dw->dma.device_prep_slave_sg = dw_axi_dma_chan_prep_slave_sg;
1474 	dw->dma.device_prep_dma_cyclic = dw_axi_dma_chan_prep_cyclic;
1475 
1476 	/*
1477 	 * Synopsis DesignWare AxiDMA datasheet mentioned Maximum
1478 	 * supported blocks is 1024. Device register width is 4 bytes.
1479 	 * Therefore, set constraint to 1024 * 4.
1480 	 */
1481 	dw->dma.dev->dma_parms = &dw->dma_parms;
1482 	dma_set_max_seg_size(&pdev->dev, MAX_BLOCK_SIZE);
1483 	platform_set_drvdata(pdev, chip);
1484 
1485 	pm_runtime_enable(chip->dev);
1486 
1487 	/*
1488 	 * We can't just call pm_runtime_get here instead of
1489 	 * pm_runtime_get_noresume + axi_dma_resume because we need
1490 	 * driver to work also without Runtime PM.
1491 	 */
1492 	pm_runtime_get_noresume(chip->dev);
1493 	ret = axi_dma_resume(chip);
1494 	if (ret < 0)
1495 		goto err_pm_disable;
1496 
1497 	axi_dma_hw_init(chip);
1498 
1499 	pm_runtime_put(chip->dev);
1500 
1501 	ret = dmaenginem_async_device_register(&dw->dma);
1502 	if (ret)
1503 		goto err_pm_disable;
1504 
1505 	/* Register with OF helpers for DMA lookups */
1506 	ret = of_dma_controller_register(pdev->dev.of_node,
1507 					 dw_axi_dma_of_xlate, dw);
1508 	if (ret < 0)
1509 		dev_warn(&pdev->dev,
1510 			 "Failed to register OF DMA controller, fallback to MEM_TO_MEM mode\n");
1511 
1512 	dev_info(chip->dev, "DesignWare AXI DMA Controller, %d channels\n",
1513 		 dw->hdata->nr_channels);
1514 
1515 	return 0;
1516 
1517 err_pm_disable:
1518 	pm_runtime_disable(chip->dev);
1519 
1520 	return ret;
1521 }
1522 
1523 static int dw_remove(struct platform_device *pdev)
1524 {
1525 	struct axi_dma_chip *chip = platform_get_drvdata(pdev);
1526 	struct dw_axi_dma *dw = chip->dw;
1527 	struct axi_dma_chan *chan, *_chan;
1528 	u32 i;
1529 
1530 	/* Enable clk before accessing to registers */
1531 	clk_prepare_enable(chip->cfgr_clk);
1532 	clk_prepare_enable(chip->core_clk);
1533 	axi_dma_irq_disable(chip);
1534 	for (i = 0; i < dw->hdata->nr_channels; i++) {
1535 		axi_chan_disable(&chip->dw->chan[i]);
1536 		axi_chan_irq_disable(&chip->dw->chan[i], DWAXIDMAC_IRQ_ALL);
1537 	}
1538 	axi_dma_disable(chip);
1539 
1540 	pm_runtime_disable(chip->dev);
1541 	axi_dma_suspend(chip);
1542 
1543 	devm_free_irq(chip->dev, chip->irq, chip);
1544 
1545 	of_dma_controller_free(chip->dev->of_node);
1546 
1547 	list_for_each_entry_safe(chan, _chan, &dw->dma.channels,
1548 			vc.chan.device_node) {
1549 		list_del(&chan->vc.chan.device_node);
1550 		tasklet_kill(&chan->vc.task);
1551 	}
1552 
1553 	return 0;
1554 }
1555 
1556 static const struct dev_pm_ops dw_axi_dma_pm_ops = {
1557 	SET_RUNTIME_PM_OPS(axi_dma_runtime_suspend, axi_dma_runtime_resume, NULL)
1558 };
1559 
1560 static const struct of_device_id dw_dma_of_id_table[] = {
1561 	{ .compatible = "snps,axi-dma-1.01a" },
1562 	{ .compatible = "intel,kmb-axi-dma" },
1563 	{}
1564 };
1565 MODULE_DEVICE_TABLE(of, dw_dma_of_id_table);
1566 
1567 static struct platform_driver dw_driver = {
1568 	.probe		= dw_probe,
1569 	.remove		= dw_remove,
1570 	.driver = {
1571 		.name	= KBUILD_MODNAME,
1572 		.of_match_table = dw_dma_of_id_table,
1573 		.pm = &dw_axi_dma_pm_ops,
1574 	},
1575 };
1576 module_platform_driver(dw_driver);
1577 
1578 MODULE_LICENSE("GPL v2");
1579 MODULE_DESCRIPTION("Synopsys DesignWare AXI DMA Controller platform driver");
1580 MODULE_AUTHOR("Eugeniy Paltsev <Eugeniy.Paltsev@synopsys.com>");
1581