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