xref: /openbmc/linux/drivers/dma/ti/k3-udma.c (revision afba8b0a)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  Copyright (C) 2019 Texas Instruments Incorporated - http://www.ti.com
4  *  Author: Peter Ujfalusi <peter.ujfalusi@ti.com>
5  */
6 
7 #include <linux/kernel.h>
8 #include <linux/delay.h>
9 #include <linux/dmaengine.h>
10 #include <linux/dma-mapping.h>
11 #include <linux/dmapool.h>
12 #include <linux/err.h>
13 #include <linux/init.h>
14 #include <linux/interrupt.h>
15 #include <linux/list.h>
16 #include <linux/platform_device.h>
17 #include <linux/slab.h>
18 #include <linux/spinlock.h>
19 #include <linux/sys_soc.h>
20 #include <linux/of.h>
21 #include <linux/of_dma.h>
22 #include <linux/of_device.h>
23 #include <linux/of_irq.h>
24 #include <linux/workqueue.h>
25 #include <linux/completion.h>
26 #include <linux/soc/ti/k3-ringacc.h>
27 #include <linux/soc/ti/ti_sci_protocol.h>
28 #include <linux/soc/ti/ti_sci_inta_msi.h>
29 #include <linux/dma/ti-cppi5.h>
30 
31 #include "../virt-dma.h"
32 #include "k3-udma.h"
33 #include "k3-psil-priv.h"
34 
35 struct udma_static_tr {
36 	u8 elsize; /* RPSTR0 */
37 	u16 elcnt; /* RPSTR0 */
38 	u16 bstcnt; /* RPSTR1 */
39 };
40 
41 #define K3_UDMA_MAX_RFLOWS		1024
42 #define K3_UDMA_DEFAULT_RING_SIZE	16
43 
44 /* How SRC/DST tag should be updated by UDMA in the descriptor's Word 3 */
45 #define UDMA_RFLOW_SRCTAG_NONE		0
46 #define UDMA_RFLOW_SRCTAG_CFG_TAG	1
47 #define UDMA_RFLOW_SRCTAG_FLOW_ID	2
48 #define UDMA_RFLOW_SRCTAG_SRC_TAG	4
49 
50 #define UDMA_RFLOW_DSTTAG_NONE		0
51 #define UDMA_RFLOW_DSTTAG_CFG_TAG	1
52 #define UDMA_RFLOW_DSTTAG_FLOW_ID	2
53 #define UDMA_RFLOW_DSTTAG_DST_TAG_LO	4
54 #define UDMA_RFLOW_DSTTAG_DST_TAG_HI	5
55 
56 struct udma_chan;
57 
58 enum udma_mmr {
59 	MMR_GCFG = 0,
60 	MMR_RCHANRT,
61 	MMR_TCHANRT,
62 	MMR_LAST,
63 };
64 
65 static const char * const mmr_names[] = { "gcfg", "rchanrt", "tchanrt" };
66 
67 struct udma_tchan {
68 	void __iomem *reg_rt;
69 
70 	int id;
71 	struct k3_ring *t_ring; /* Transmit ring */
72 	struct k3_ring *tc_ring; /* Transmit Completion ring */
73 };
74 
75 struct udma_rflow {
76 	int id;
77 	struct k3_ring *fd_ring; /* Free Descriptor ring */
78 	struct k3_ring *r_ring; /* Receive ring */
79 };
80 
81 struct udma_rchan {
82 	void __iomem *reg_rt;
83 
84 	int id;
85 };
86 
87 #define UDMA_FLAG_PDMA_ACC32		BIT(0)
88 #define UDMA_FLAG_PDMA_BURST		BIT(1)
89 
90 struct udma_match_data {
91 	u32 psil_base;
92 	bool enable_memcpy_support;
93 	u32 flags;
94 	u32 statictr_z_mask;
95 };
96 
97 struct udma_soc_data {
98 	u32 rchan_oes_offset;
99 };
100 
101 struct udma_hwdesc {
102 	size_t cppi5_desc_size;
103 	void *cppi5_desc_vaddr;
104 	dma_addr_t cppi5_desc_paddr;
105 
106 	/* TR descriptor internal pointers */
107 	void *tr_req_base;
108 	struct cppi5_tr_resp_t *tr_resp_base;
109 };
110 
111 struct udma_rx_flush {
112 	struct udma_hwdesc hwdescs[2];
113 
114 	size_t buffer_size;
115 	void *buffer_vaddr;
116 	dma_addr_t buffer_paddr;
117 };
118 
119 struct udma_dev {
120 	struct dma_device ddev;
121 	struct device *dev;
122 	void __iomem *mmrs[MMR_LAST];
123 	const struct udma_match_data *match_data;
124 	const struct udma_soc_data *soc_data;
125 
126 	u8 tpl_levels;
127 	u32 tpl_start_idx[3];
128 
129 	size_t desc_align; /* alignment to use for descriptors */
130 
131 	struct udma_tisci_rm tisci_rm;
132 
133 	struct k3_ringacc *ringacc;
134 
135 	struct work_struct purge_work;
136 	struct list_head desc_to_purge;
137 	spinlock_t lock;
138 
139 	struct udma_rx_flush rx_flush;
140 
141 	int tchan_cnt;
142 	int echan_cnt;
143 	int rchan_cnt;
144 	int rflow_cnt;
145 	unsigned long *tchan_map;
146 	unsigned long *rchan_map;
147 	unsigned long *rflow_gp_map;
148 	unsigned long *rflow_gp_map_allocated;
149 	unsigned long *rflow_in_use;
150 
151 	struct udma_tchan *tchans;
152 	struct udma_rchan *rchans;
153 	struct udma_rflow *rflows;
154 
155 	struct udma_chan *channels;
156 	u32 psil_base;
157 	u32 atype;
158 };
159 
160 struct udma_desc {
161 	struct virt_dma_desc vd;
162 
163 	bool terminated;
164 
165 	enum dma_transfer_direction dir;
166 
167 	struct udma_static_tr static_tr;
168 	u32 residue;
169 
170 	unsigned int sglen;
171 	unsigned int desc_idx; /* Only used for cyclic in packet mode */
172 	unsigned int tr_idx;
173 
174 	u32 metadata_size;
175 	void *metadata; /* pointer to provided metadata buffer (EPIP, PSdata) */
176 
177 	unsigned int hwdesc_count;
178 	struct udma_hwdesc hwdesc[];
179 };
180 
181 enum udma_chan_state {
182 	UDMA_CHAN_IS_IDLE = 0, /* not active, no teardown is in progress */
183 	UDMA_CHAN_IS_ACTIVE, /* Normal operation */
184 	UDMA_CHAN_IS_TERMINATING, /* channel is being terminated */
185 };
186 
187 struct udma_tx_drain {
188 	struct delayed_work work;
189 	ktime_t tstamp;
190 	u32 residue;
191 };
192 
193 struct udma_chan_config {
194 	bool pkt_mode; /* TR or packet */
195 	bool needs_epib; /* EPIB is needed for the communication or not */
196 	u32 psd_size; /* size of Protocol Specific Data */
197 	u32 metadata_size; /* (needs_epib ? 16:0) + psd_size */
198 	u32 hdesc_size; /* Size of a packet descriptor in packet mode */
199 	bool notdpkt; /* Suppress sending TDC packet */
200 	int remote_thread_id;
201 	u32 atype;
202 	u32 src_thread;
203 	u32 dst_thread;
204 	enum psil_endpoint_type ep_type;
205 	bool enable_acc32;
206 	bool enable_burst;
207 	enum udma_tp_level channel_tpl; /* Channel Throughput Level */
208 
209 	enum dma_transfer_direction dir;
210 };
211 
212 struct udma_chan {
213 	struct virt_dma_chan vc;
214 	struct dma_slave_config	cfg;
215 	struct udma_dev *ud;
216 	struct udma_desc *desc;
217 	struct udma_desc *terminated_desc;
218 	struct udma_static_tr static_tr;
219 	char *name;
220 
221 	struct udma_tchan *tchan;
222 	struct udma_rchan *rchan;
223 	struct udma_rflow *rflow;
224 
225 	bool psil_paired;
226 
227 	int irq_num_ring;
228 	int irq_num_udma;
229 
230 	bool cyclic;
231 	bool paused;
232 
233 	enum udma_chan_state state;
234 	struct completion teardown_completed;
235 
236 	struct udma_tx_drain tx_drain;
237 
238 	u32 bcnt; /* number of bytes completed since the start of the channel */
239 
240 	/* Channel configuration parameters */
241 	struct udma_chan_config config;
242 
243 	/* dmapool for packet mode descriptors */
244 	bool use_dma_pool;
245 	struct dma_pool *hdesc_pool;
246 
247 	u32 id;
248 };
249 
250 static inline struct udma_dev *to_udma_dev(struct dma_device *d)
251 {
252 	return container_of(d, struct udma_dev, ddev);
253 }
254 
255 static inline struct udma_chan *to_udma_chan(struct dma_chan *c)
256 {
257 	return container_of(c, struct udma_chan, vc.chan);
258 }
259 
260 static inline struct udma_desc *to_udma_desc(struct dma_async_tx_descriptor *t)
261 {
262 	return container_of(t, struct udma_desc, vd.tx);
263 }
264 
265 /* Generic register access functions */
266 static inline u32 udma_read(void __iomem *base, int reg)
267 {
268 	return readl(base + reg);
269 }
270 
271 static inline void udma_write(void __iomem *base, int reg, u32 val)
272 {
273 	writel(val, base + reg);
274 }
275 
276 static inline void udma_update_bits(void __iomem *base, int reg,
277 				    u32 mask, u32 val)
278 {
279 	u32 tmp, orig;
280 
281 	orig = readl(base + reg);
282 	tmp = orig & ~mask;
283 	tmp |= (val & mask);
284 
285 	if (tmp != orig)
286 		writel(tmp, base + reg);
287 }
288 
289 /* TCHANRT */
290 static inline u32 udma_tchanrt_read(struct udma_chan *uc, int reg)
291 {
292 	if (!uc->tchan)
293 		return 0;
294 	return udma_read(uc->tchan->reg_rt, reg);
295 }
296 
297 static inline void udma_tchanrt_write(struct udma_chan *uc, int reg, u32 val)
298 {
299 	if (!uc->tchan)
300 		return;
301 	udma_write(uc->tchan->reg_rt, reg, val);
302 }
303 
304 static inline void udma_tchanrt_update_bits(struct udma_chan *uc, int reg,
305 					    u32 mask, u32 val)
306 {
307 	if (!uc->tchan)
308 		return;
309 	udma_update_bits(uc->tchan->reg_rt, reg, mask, val);
310 }
311 
312 /* RCHANRT */
313 static inline u32 udma_rchanrt_read(struct udma_chan *uc, int reg)
314 {
315 	if (!uc->rchan)
316 		return 0;
317 	return udma_read(uc->rchan->reg_rt, reg);
318 }
319 
320 static inline void udma_rchanrt_write(struct udma_chan *uc, int reg, u32 val)
321 {
322 	if (!uc->rchan)
323 		return;
324 	udma_write(uc->rchan->reg_rt, reg, val);
325 }
326 
327 static inline void udma_rchanrt_update_bits(struct udma_chan *uc, int reg,
328 					    u32 mask, u32 val)
329 {
330 	if (!uc->rchan)
331 		return;
332 	udma_update_bits(uc->rchan->reg_rt, reg, mask, val);
333 }
334 
335 static int navss_psil_pair(struct udma_dev *ud, u32 src_thread, u32 dst_thread)
336 {
337 	struct udma_tisci_rm *tisci_rm = &ud->tisci_rm;
338 
339 	dst_thread |= K3_PSIL_DST_THREAD_ID_OFFSET;
340 	return tisci_rm->tisci_psil_ops->pair(tisci_rm->tisci,
341 					      tisci_rm->tisci_navss_dev_id,
342 					      src_thread, dst_thread);
343 }
344 
345 static int navss_psil_unpair(struct udma_dev *ud, u32 src_thread,
346 			     u32 dst_thread)
347 {
348 	struct udma_tisci_rm *tisci_rm = &ud->tisci_rm;
349 
350 	dst_thread |= K3_PSIL_DST_THREAD_ID_OFFSET;
351 	return tisci_rm->tisci_psil_ops->unpair(tisci_rm->tisci,
352 						tisci_rm->tisci_navss_dev_id,
353 						src_thread, dst_thread);
354 }
355 
356 static void udma_reset_uchan(struct udma_chan *uc)
357 {
358 	memset(&uc->config, 0, sizeof(uc->config));
359 	uc->config.remote_thread_id = -1;
360 	uc->state = UDMA_CHAN_IS_IDLE;
361 }
362 
363 static void udma_dump_chan_stdata(struct udma_chan *uc)
364 {
365 	struct device *dev = uc->ud->dev;
366 	u32 offset;
367 	int i;
368 
369 	if (uc->config.dir == DMA_MEM_TO_DEV || uc->config.dir == DMA_MEM_TO_MEM) {
370 		dev_dbg(dev, "TCHAN State data:\n");
371 		for (i = 0; i < 32; i++) {
372 			offset = UDMA_CHAN_RT_STDATA_REG + i * 4;
373 			dev_dbg(dev, "TRT_STDATA[%02d]: 0x%08x\n", i,
374 				udma_tchanrt_read(uc, offset));
375 		}
376 	}
377 
378 	if (uc->config.dir == DMA_DEV_TO_MEM || uc->config.dir == DMA_MEM_TO_MEM) {
379 		dev_dbg(dev, "RCHAN State data:\n");
380 		for (i = 0; i < 32; i++) {
381 			offset = UDMA_CHAN_RT_STDATA_REG + i * 4;
382 			dev_dbg(dev, "RRT_STDATA[%02d]: 0x%08x\n", i,
383 				udma_rchanrt_read(uc, offset));
384 		}
385 	}
386 }
387 
388 static inline dma_addr_t udma_curr_cppi5_desc_paddr(struct udma_desc *d,
389 						    int idx)
390 {
391 	return d->hwdesc[idx].cppi5_desc_paddr;
392 }
393 
394 static inline void *udma_curr_cppi5_desc_vaddr(struct udma_desc *d, int idx)
395 {
396 	return d->hwdesc[idx].cppi5_desc_vaddr;
397 }
398 
399 static struct udma_desc *udma_udma_desc_from_paddr(struct udma_chan *uc,
400 						   dma_addr_t paddr)
401 {
402 	struct udma_desc *d = uc->terminated_desc;
403 
404 	if (d) {
405 		dma_addr_t desc_paddr = udma_curr_cppi5_desc_paddr(d,
406 								   d->desc_idx);
407 
408 		if (desc_paddr != paddr)
409 			d = NULL;
410 	}
411 
412 	if (!d) {
413 		d = uc->desc;
414 		if (d) {
415 			dma_addr_t desc_paddr = udma_curr_cppi5_desc_paddr(d,
416 								d->desc_idx);
417 
418 			if (desc_paddr != paddr)
419 				d = NULL;
420 		}
421 	}
422 
423 	return d;
424 }
425 
426 static void udma_free_hwdesc(struct udma_chan *uc, struct udma_desc *d)
427 {
428 	if (uc->use_dma_pool) {
429 		int i;
430 
431 		for (i = 0; i < d->hwdesc_count; i++) {
432 			if (!d->hwdesc[i].cppi5_desc_vaddr)
433 				continue;
434 
435 			dma_pool_free(uc->hdesc_pool,
436 				      d->hwdesc[i].cppi5_desc_vaddr,
437 				      d->hwdesc[i].cppi5_desc_paddr);
438 
439 			d->hwdesc[i].cppi5_desc_vaddr = NULL;
440 		}
441 	} else if (d->hwdesc[0].cppi5_desc_vaddr) {
442 		struct udma_dev *ud = uc->ud;
443 
444 		dma_free_coherent(ud->dev, d->hwdesc[0].cppi5_desc_size,
445 				  d->hwdesc[0].cppi5_desc_vaddr,
446 				  d->hwdesc[0].cppi5_desc_paddr);
447 
448 		d->hwdesc[0].cppi5_desc_vaddr = NULL;
449 	}
450 }
451 
452 static void udma_purge_desc_work(struct work_struct *work)
453 {
454 	struct udma_dev *ud = container_of(work, typeof(*ud), purge_work);
455 	struct virt_dma_desc *vd, *_vd;
456 	unsigned long flags;
457 	LIST_HEAD(head);
458 
459 	spin_lock_irqsave(&ud->lock, flags);
460 	list_splice_tail_init(&ud->desc_to_purge, &head);
461 	spin_unlock_irqrestore(&ud->lock, flags);
462 
463 	list_for_each_entry_safe(vd, _vd, &head, node) {
464 		struct udma_chan *uc = to_udma_chan(vd->tx.chan);
465 		struct udma_desc *d = to_udma_desc(&vd->tx);
466 
467 		udma_free_hwdesc(uc, d);
468 		list_del(&vd->node);
469 		kfree(d);
470 	}
471 
472 	/* If more to purge, schedule the work again */
473 	if (!list_empty(&ud->desc_to_purge))
474 		schedule_work(&ud->purge_work);
475 }
476 
477 static void udma_desc_free(struct virt_dma_desc *vd)
478 {
479 	struct udma_dev *ud = to_udma_dev(vd->tx.chan->device);
480 	struct udma_chan *uc = to_udma_chan(vd->tx.chan);
481 	struct udma_desc *d = to_udma_desc(&vd->tx);
482 	unsigned long flags;
483 
484 	if (uc->terminated_desc == d)
485 		uc->terminated_desc = NULL;
486 
487 	if (uc->use_dma_pool) {
488 		udma_free_hwdesc(uc, d);
489 		kfree(d);
490 		return;
491 	}
492 
493 	spin_lock_irqsave(&ud->lock, flags);
494 	list_add_tail(&vd->node, &ud->desc_to_purge);
495 	spin_unlock_irqrestore(&ud->lock, flags);
496 
497 	schedule_work(&ud->purge_work);
498 }
499 
500 static bool udma_is_chan_running(struct udma_chan *uc)
501 {
502 	u32 trt_ctl = 0;
503 	u32 rrt_ctl = 0;
504 
505 	if (uc->tchan)
506 		trt_ctl = udma_tchanrt_read(uc, UDMA_CHAN_RT_CTL_REG);
507 	if (uc->rchan)
508 		rrt_ctl = udma_rchanrt_read(uc, UDMA_CHAN_RT_CTL_REG);
509 
510 	if (trt_ctl & UDMA_CHAN_RT_CTL_EN || rrt_ctl & UDMA_CHAN_RT_CTL_EN)
511 		return true;
512 
513 	return false;
514 }
515 
516 static bool udma_is_chan_paused(struct udma_chan *uc)
517 {
518 	u32 val, pause_mask;
519 
520 	switch (uc->config.dir) {
521 	case DMA_DEV_TO_MEM:
522 		val = udma_rchanrt_read(uc, UDMA_CHAN_RT_PEER_RT_EN_REG);
523 		pause_mask = UDMA_PEER_RT_EN_PAUSE;
524 		break;
525 	case DMA_MEM_TO_DEV:
526 		val = udma_tchanrt_read(uc, UDMA_CHAN_RT_PEER_RT_EN_REG);
527 		pause_mask = UDMA_PEER_RT_EN_PAUSE;
528 		break;
529 	case DMA_MEM_TO_MEM:
530 		val = udma_tchanrt_read(uc, UDMA_CHAN_RT_CTL_REG);
531 		pause_mask = UDMA_CHAN_RT_CTL_PAUSE;
532 		break;
533 	default:
534 		return false;
535 	}
536 
537 	if (val & pause_mask)
538 		return true;
539 
540 	return false;
541 }
542 
543 static inline dma_addr_t udma_get_rx_flush_hwdesc_paddr(struct udma_chan *uc)
544 {
545 	return uc->ud->rx_flush.hwdescs[uc->config.pkt_mode].cppi5_desc_paddr;
546 }
547 
548 static int udma_push_to_ring(struct udma_chan *uc, int idx)
549 {
550 	struct udma_desc *d = uc->desc;
551 	struct k3_ring *ring = NULL;
552 	dma_addr_t paddr;
553 
554 	switch (uc->config.dir) {
555 	case DMA_DEV_TO_MEM:
556 		ring = uc->rflow->fd_ring;
557 		break;
558 	case DMA_MEM_TO_DEV:
559 	case DMA_MEM_TO_MEM:
560 		ring = uc->tchan->t_ring;
561 		break;
562 	default:
563 		return -EINVAL;
564 	}
565 
566 	/* RX flush packet: idx == -1 is only passed in case of DEV_TO_MEM */
567 	if (idx == -1) {
568 		paddr = udma_get_rx_flush_hwdesc_paddr(uc);
569 	} else {
570 		paddr = udma_curr_cppi5_desc_paddr(d, idx);
571 
572 		wmb(); /* Ensure that writes are not moved over this point */
573 	}
574 
575 	return k3_ringacc_ring_push(ring, &paddr);
576 }
577 
578 static bool udma_desc_is_rx_flush(struct udma_chan *uc, dma_addr_t addr)
579 {
580 	if (uc->config.dir != DMA_DEV_TO_MEM)
581 		return false;
582 
583 	if (addr == udma_get_rx_flush_hwdesc_paddr(uc))
584 		return true;
585 
586 	return false;
587 }
588 
589 static int udma_pop_from_ring(struct udma_chan *uc, dma_addr_t *addr)
590 {
591 	struct k3_ring *ring = NULL;
592 	int ret;
593 
594 	switch (uc->config.dir) {
595 	case DMA_DEV_TO_MEM:
596 		ring = uc->rflow->r_ring;
597 		break;
598 	case DMA_MEM_TO_DEV:
599 	case DMA_MEM_TO_MEM:
600 		ring = uc->tchan->tc_ring;
601 		break;
602 	default:
603 		return -ENOENT;
604 	}
605 
606 	ret = k3_ringacc_ring_pop(ring, addr);
607 	if (ret)
608 		return ret;
609 
610 	rmb(); /* Ensure that reads are not moved before this point */
611 
612 	/* Teardown completion */
613 	if (cppi5_desc_is_tdcm(*addr))
614 		return 0;
615 
616 	/* Check for flush descriptor */
617 	if (udma_desc_is_rx_flush(uc, *addr))
618 		return -ENOENT;
619 
620 	return 0;
621 }
622 
623 static void udma_reset_rings(struct udma_chan *uc)
624 {
625 	struct k3_ring *ring1 = NULL;
626 	struct k3_ring *ring2 = NULL;
627 
628 	switch (uc->config.dir) {
629 	case DMA_DEV_TO_MEM:
630 		if (uc->rchan) {
631 			ring1 = uc->rflow->fd_ring;
632 			ring2 = uc->rflow->r_ring;
633 		}
634 		break;
635 	case DMA_MEM_TO_DEV:
636 	case DMA_MEM_TO_MEM:
637 		if (uc->tchan) {
638 			ring1 = uc->tchan->t_ring;
639 			ring2 = uc->tchan->tc_ring;
640 		}
641 		break;
642 	default:
643 		break;
644 	}
645 
646 	if (ring1)
647 		k3_ringacc_ring_reset_dma(ring1,
648 					  k3_ringacc_ring_get_occ(ring1));
649 	if (ring2)
650 		k3_ringacc_ring_reset(ring2);
651 
652 	/* make sure we are not leaking memory by stalled descriptor */
653 	if (uc->terminated_desc) {
654 		udma_desc_free(&uc->terminated_desc->vd);
655 		uc->terminated_desc = NULL;
656 	}
657 }
658 
659 static void udma_reset_counters(struct udma_chan *uc)
660 {
661 	u32 val;
662 
663 	if (uc->tchan) {
664 		val = udma_tchanrt_read(uc, UDMA_CHAN_RT_BCNT_REG);
665 		udma_tchanrt_write(uc, UDMA_CHAN_RT_BCNT_REG, val);
666 
667 		val = udma_tchanrt_read(uc, UDMA_CHAN_RT_SBCNT_REG);
668 		udma_tchanrt_write(uc, UDMA_CHAN_RT_SBCNT_REG, val);
669 
670 		val = udma_tchanrt_read(uc, UDMA_CHAN_RT_PCNT_REG);
671 		udma_tchanrt_write(uc, UDMA_CHAN_RT_PCNT_REG, val);
672 
673 		val = udma_tchanrt_read(uc, UDMA_CHAN_RT_PEER_BCNT_REG);
674 		udma_tchanrt_write(uc, UDMA_CHAN_RT_PEER_BCNT_REG, val);
675 	}
676 
677 	if (uc->rchan) {
678 		val = udma_rchanrt_read(uc, UDMA_CHAN_RT_BCNT_REG);
679 		udma_rchanrt_write(uc, UDMA_CHAN_RT_BCNT_REG, val);
680 
681 		val = udma_rchanrt_read(uc, UDMA_CHAN_RT_SBCNT_REG);
682 		udma_rchanrt_write(uc, UDMA_CHAN_RT_SBCNT_REG, val);
683 
684 		val = udma_rchanrt_read(uc, UDMA_CHAN_RT_PCNT_REG);
685 		udma_rchanrt_write(uc, UDMA_CHAN_RT_PCNT_REG, val);
686 
687 		val = udma_rchanrt_read(uc, UDMA_CHAN_RT_PEER_BCNT_REG);
688 		udma_rchanrt_write(uc, UDMA_CHAN_RT_PEER_BCNT_REG, val);
689 	}
690 
691 	uc->bcnt = 0;
692 }
693 
694 static int udma_reset_chan(struct udma_chan *uc, bool hard)
695 {
696 	switch (uc->config.dir) {
697 	case DMA_DEV_TO_MEM:
698 		udma_rchanrt_write(uc, UDMA_CHAN_RT_PEER_RT_EN_REG, 0);
699 		udma_rchanrt_write(uc, UDMA_CHAN_RT_CTL_REG, 0);
700 		break;
701 	case DMA_MEM_TO_DEV:
702 		udma_tchanrt_write(uc, UDMA_CHAN_RT_CTL_REG, 0);
703 		udma_tchanrt_write(uc, UDMA_CHAN_RT_PEER_RT_EN_REG, 0);
704 		break;
705 	case DMA_MEM_TO_MEM:
706 		udma_rchanrt_write(uc, UDMA_CHAN_RT_CTL_REG, 0);
707 		udma_tchanrt_write(uc, UDMA_CHAN_RT_CTL_REG, 0);
708 		break;
709 	default:
710 		return -EINVAL;
711 	}
712 
713 	/* Reset all counters */
714 	udma_reset_counters(uc);
715 
716 	/* Hard reset: re-initialize the channel to reset */
717 	if (hard) {
718 		struct udma_chan_config ucc_backup;
719 		int ret;
720 
721 		memcpy(&ucc_backup, &uc->config, sizeof(uc->config));
722 		uc->ud->ddev.device_free_chan_resources(&uc->vc.chan);
723 
724 		/* restore the channel configuration */
725 		memcpy(&uc->config, &ucc_backup, sizeof(uc->config));
726 		ret = uc->ud->ddev.device_alloc_chan_resources(&uc->vc.chan);
727 		if (ret)
728 			return ret;
729 
730 		/*
731 		 * Setting forced teardown after forced reset helps recovering
732 		 * the rchan.
733 		 */
734 		if (uc->config.dir == DMA_DEV_TO_MEM)
735 			udma_rchanrt_write(uc, UDMA_CHAN_RT_CTL_REG,
736 					   UDMA_CHAN_RT_CTL_EN |
737 					   UDMA_CHAN_RT_CTL_TDOWN |
738 					   UDMA_CHAN_RT_CTL_FTDOWN);
739 	}
740 	uc->state = UDMA_CHAN_IS_IDLE;
741 
742 	return 0;
743 }
744 
745 static void udma_start_desc(struct udma_chan *uc)
746 {
747 	struct udma_chan_config *ucc = &uc->config;
748 
749 	if (ucc->pkt_mode && (uc->cyclic || ucc->dir == DMA_DEV_TO_MEM)) {
750 		int i;
751 
752 		/* Push all descriptors to ring for packet mode cyclic or RX */
753 		for (i = 0; i < uc->desc->sglen; i++)
754 			udma_push_to_ring(uc, i);
755 	} else {
756 		udma_push_to_ring(uc, 0);
757 	}
758 }
759 
760 static bool udma_chan_needs_reconfiguration(struct udma_chan *uc)
761 {
762 	/* Only PDMAs have staticTR */
763 	if (uc->config.ep_type == PSIL_EP_NATIVE)
764 		return false;
765 
766 	/* Check if the staticTR configuration has changed for TX */
767 	if (memcmp(&uc->static_tr, &uc->desc->static_tr, sizeof(uc->static_tr)))
768 		return true;
769 
770 	return false;
771 }
772 
773 static int udma_start(struct udma_chan *uc)
774 {
775 	struct virt_dma_desc *vd = vchan_next_desc(&uc->vc);
776 
777 	if (!vd) {
778 		uc->desc = NULL;
779 		return -ENOENT;
780 	}
781 
782 	list_del(&vd->node);
783 
784 	uc->desc = to_udma_desc(&vd->tx);
785 
786 	/* Channel is already running and does not need reconfiguration */
787 	if (udma_is_chan_running(uc) && !udma_chan_needs_reconfiguration(uc)) {
788 		udma_start_desc(uc);
789 		goto out;
790 	}
791 
792 	/* Make sure that we clear the teardown bit, if it is set */
793 	udma_reset_chan(uc, false);
794 
795 	/* Push descriptors before we start the channel */
796 	udma_start_desc(uc);
797 
798 	switch (uc->desc->dir) {
799 	case DMA_DEV_TO_MEM:
800 		/* Config remote TR */
801 		if (uc->config.ep_type == PSIL_EP_PDMA_XY) {
802 			u32 val = PDMA_STATIC_TR_Y(uc->desc->static_tr.elcnt) |
803 				  PDMA_STATIC_TR_X(uc->desc->static_tr.elsize);
804 			const struct udma_match_data *match_data =
805 							uc->ud->match_data;
806 
807 			if (uc->config.enable_acc32)
808 				val |= PDMA_STATIC_TR_XY_ACC32;
809 			if (uc->config.enable_burst)
810 				val |= PDMA_STATIC_TR_XY_BURST;
811 
812 			udma_rchanrt_write(uc,
813 					   UDMA_CHAN_RT_PEER_STATIC_TR_XY_REG,
814 					   val);
815 
816 			udma_rchanrt_write(uc,
817 				UDMA_CHAN_RT_PEER_STATIC_TR_Z_REG,
818 				PDMA_STATIC_TR_Z(uc->desc->static_tr.bstcnt,
819 						 match_data->statictr_z_mask));
820 
821 			/* save the current staticTR configuration */
822 			memcpy(&uc->static_tr, &uc->desc->static_tr,
823 			       sizeof(uc->static_tr));
824 		}
825 
826 		udma_rchanrt_write(uc, UDMA_CHAN_RT_CTL_REG,
827 				   UDMA_CHAN_RT_CTL_EN);
828 
829 		/* Enable remote */
830 		udma_rchanrt_write(uc, UDMA_CHAN_RT_PEER_RT_EN_REG,
831 				   UDMA_PEER_RT_EN_ENABLE);
832 
833 		break;
834 	case DMA_MEM_TO_DEV:
835 		/* Config remote TR */
836 		if (uc->config.ep_type == PSIL_EP_PDMA_XY) {
837 			u32 val = PDMA_STATIC_TR_Y(uc->desc->static_tr.elcnt) |
838 				  PDMA_STATIC_TR_X(uc->desc->static_tr.elsize);
839 
840 			if (uc->config.enable_acc32)
841 				val |= PDMA_STATIC_TR_XY_ACC32;
842 			if (uc->config.enable_burst)
843 				val |= PDMA_STATIC_TR_XY_BURST;
844 
845 			udma_tchanrt_write(uc,
846 					   UDMA_CHAN_RT_PEER_STATIC_TR_XY_REG,
847 					   val);
848 
849 			/* save the current staticTR configuration */
850 			memcpy(&uc->static_tr, &uc->desc->static_tr,
851 			       sizeof(uc->static_tr));
852 		}
853 
854 		/* Enable remote */
855 		udma_tchanrt_write(uc, UDMA_CHAN_RT_PEER_RT_EN_REG,
856 				   UDMA_PEER_RT_EN_ENABLE);
857 
858 		udma_tchanrt_write(uc, UDMA_CHAN_RT_CTL_REG,
859 				   UDMA_CHAN_RT_CTL_EN);
860 
861 		break;
862 	case DMA_MEM_TO_MEM:
863 		udma_rchanrt_write(uc, UDMA_CHAN_RT_CTL_REG,
864 				   UDMA_CHAN_RT_CTL_EN);
865 		udma_tchanrt_write(uc, UDMA_CHAN_RT_CTL_REG,
866 				   UDMA_CHAN_RT_CTL_EN);
867 
868 		break;
869 	default:
870 		return -EINVAL;
871 	}
872 
873 	uc->state = UDMA_CHAN_IS_ACTIVE;
874 out:
875 
876 	return 0;
877 }
878 
879 static int udma_stop(struct udma_chan *uc)
880 {
881 	enum udma_chan_state old_state = uc->state;
882 
883 	uc->state = UDMA_CHAN_IS_TERMINATING;
884 	reinit_completion(&uc->teardown_completed);
885 
886 	switch (uc->config.dir) {
887 	case DMA_DEV_TO_MEM:
888 		if (!uc->cyclic && !uc->desc)
889 			udma_push_to_ring(uc, -1);
890 
891 		udma_rchanrt_write(uc, UDMA_CHAN_RT_PEER_RT_EN_REG,
892 				   UDMA_PEER_RT_EN_ENABLE |
893 				   UDMA_PEER_RT_EN_TEARDOWN);
894 		break;
895 	case DMA_MEM_TO_DEV:
896 		udma_tchanrt_write(uc, UDMA_CHAN_RT_PEER_RT_EN_REG,
897 				   UDMA_PEER_RT_EN_ENABLE |
898 				   UDMA_PEER_RT_EN_FLUSH);
899 		udma_tchanrt_write(uc, UDMA_CHAN_RT_CTL_REG,
900 				   UDMA_CHAN_RT_CTL_EN |
901 				   UDMA_CHAN_RT_CTL_TDOWN);
902 		break;
903 	case DMA_MEM_TO_MEM:
904 		udma_tchanrt_write(uc, UDMA_CHAN_RT_CTL_REG,
905 				   UDMA_CHAN_RT_CTL_EN |
906 				   UDMA_CHAN_RT_CTL_TDOWN);
907 		break;
908 	default:
909 		uc->state = old_state;
910 		complete_all(&uc->teardown_completed);
911 		return -EINVAL;
912 	}
913 
914 	return 0;
915 }
916 
917 static void udma_cyclic_packet_elapsed(struct udma_chan *uc)
918 {
919 	struct udma_desc *d = uc->desc;
920 	struct cppi5_host_desc_t *h_desc;
921 
922 	h_desc = d->hwdesc[d->desc_idx].cppi5_desc_vaddr;
923 	cppi5_hdesc_reset_to_original(h_desc);
924 	udma_push_to_ring(uc, d->desc_idx);
925 	d->desc_idx = (d->desc_idx + 1) % d->sglen;
926 }
927 
928 static inline void udma_fetch_epib(struct udma_chan *uc, struct udma_desc *d)
929 {
930 	struct cppi5_host_desc_t *h_desc = d->hwdesc[0].cppi5_desc_vaddr;
931 
932 	memcpy(d->metadata, h_desc->epib, d->metadata_size);
933 }
934 
935 static bool udma_is_desc_really_done(struct udma_chan *uc, struct udma_desc *d)
936 {
937 	u32 peer_bcnt, bcnt;
938 
939 	/* Only TX towards PDMA is affected */
940 	if (uc->config.ep_type == PSIL_EP_NATIVE ||
941 	    uc->config.dir != DMA_MEM_TO_DEV)
942 		return true;
943 
944 	peer_bcnt = udma_tchanrt_read(uc, UDMA_CHAN_RT_PEER_BCNT_REG);
945 	bcnt = udma_tchanrt_read(uc, UDMA_CHAN_RT_BCNT_REG);
946 
947 	/* Transfer is incomplete, store current residue and time stamp */
948 	if (peer_bcnt < bcnt) {
949 		uc->tx_drain.residue = bcnt - peer_bcnt;
950 		uc->tx_drain.tstamp = ktime_get();
951 		return false;
952 	}
953 
954 	return true;
955 }
956 
957 static void udma_check_tx_completion(struct work_struct *work)
958 {
959 	struct udma_chan *uc = container_of(work, typeof(*uc),
960 					    tx_drain.work.work);
961 	bool desc_done = true;
962 	u32 residue_diff;
963 	ktime_t time_diff;
964 	unsigned long delay;
965 
966 	while (1) {
967 		if (uc->desc) {
968 			/* Get previous residue and time stamp */
969 			residue_diff = uc->tx_drain.residue;
970 			time_diff = uc->tx_drain.tstamp;
971 			/*
972 			 * Get current residue and time stamp or see if
973 			 * transfer is complete
974 			 */
975 			desc_done = udma_is_desc_really_done(uc, uc->desc);
976 		}
977 
978 		if (!desc_done) {
979 			/*
980 			 * Find the time delta and residue delta w.r.t
981 			 * previous poll
982 			 */
983 			time_diff = ktime_sub(uc->tx_drain.tstamp,
984 					      time_diff) + 1;
985 			residue_diff -= uc->tx_drain.residue;
986 			if (residue_diff) {
987 				/*
988 				 * Try to guess when we should check
989 				 * next time by calculating rate at
990 				 * which data is being drained at the
991 				 * peer device
992 				 */
993 				delay = (time_diff / residue_diff) *
994 					uc->tx_drain.residue;
995 			} else {
996 				/* No progress, check again in 1 second  */
997 				schedule_delayed_work(&uc->tx_drain.work, HZ);
998 				break;
999 			}
1000 
1001 			usleep_range(ktime_to_us(delay),
1002 				     ktime_to_us(delay) + 10);
1003 			continue;
1004 		}
1005 
1006 		if (uc->desc) {
1007 			struct udma_desc *d = uc->desc;
1008 
1009 			uc->bcnt += d->residue;
1010 			udma_start(uc);
1011 			vchan_cookie_complete(&d->vd);
1012 			break;
1013 		}
1014 
1015 		break;
1016 	}
1017 }
1018 
1019 static irqreturn_t udma_ring_irq_handler(int irq, void *data)
1020 {
1021 	struct udma_chan *uc = data;
1022 	struct udma_desc *d;
1023 	unsigned long flags;
1024 	dma_addr_t paddr = 0;
1025 
1026 	if (udma_pop_from_ring(uc, &paddr) || !paddr)
1027 		return IRQ_HANDLED;
1028 
1029 	spin_lock_irqsave(&uc->vc.lock, flags);
1030 
1031 	/* Teardown completion message */
1032 	if (cppi5_desc_is_tdcm(paddr)) {
1033 		complete_all(&uc->teardown_completed);
1034 
1035 		if (uc->terminated_desc) {
1036 			udma_desc_free(&uc->terminated_desc->vd);
1037 			uc->terminated_desc = NULL;
1038 		}
1039 
1040 		if (!uc->desc)
1041 			udma_start(uc);
1042 
1043 		goto out;
1044 	}
1045 
1046 	d = udma_udma_desc_from_paddr(uc, paddr);
1047 
1048 	if (d) {
1049 		dma_addr_t desc_paddr = udma_curr_cppi5_desc_paddr(d,
1050 								   d->desc_idx);
1051 		if (desc_paddr != paddr) {
1052 			dev_err(uc->ud->dev, "not matching descriptors!\n");
1053 			goto out;
1054 		}
1055 
1056 		if (d == uc->desc) {
1057 			/* active descriptor */
1058 			if (uc->cyclic) {
1059 				udma_cyclic_packet_elapsed(uc);
1060 				vchan_cyclic_callback(&d->vd);
1061 			} else {
1062 				if (udma_is_desc_really_done(uc, d)) {
1063 					uc->bcnt += d->residue;
1064 					udma_start(uc);
1065 					vchan_cookie_complete(&d->vd);
1066 				} else {
1067 					schedule_delayed_work(&uc->tx_drain.work,
1068 							      0);
1069 				}
1070 			}
1071 		} else {
1072 			/*
1073 			 * terminated descriptor, mark the descriptor as
1074 			 * completed to update the channel's cookie marker
1075 			 */
1076 			dma_cookie_complete(&d->vd.tx);
1077 		}
1078 	}
1079 out:
1080 	spin_unlock_irqrestore(&uc->vc.lock, flags);
1081 
1082 	return IRQ_HANDLED;
1083 }
1084 
1085 static irqreturn_t udma_udma_irq_handler(int irq, void *data)
1086 {
1087 	struct udma_chan *uc = data;
1088 	struct udma_desc *d;
1089 	unsigned long flags;
1090 
1091 	spin_lock_irqsave(&uc->vc.lock, flags);
1092 	d = uc->desc;
1093 	if (d) {
1094 		d->tr_idx = (d->tr_idx + 1) % d->sglen;
1095 
1096 		if (uc->cyclic) {
1097 			vchan_cyclic_callback(&d->vd);
1098 		} else {
1099 			/* TODO: figure out the real amount of data */
1100 			uc->bcnt += d->residue;
1101 			udma_start(uc);
1102 			vchan_cookie_complete(&d->vd);
1103 		}
1104 	}
1105 
1106 	spin_unlock_irqrestore(&uc->vc.lock, flags);
1107 
1108 	return IRQ_HANDLED;
1109 }
1110 
1111 /**
1112  * __udma_alloc_gp_rflow_range - alloc range of GP RX flows
1113  * @ud: UDMA device
1114  * @from: Start the search from this flow id number
1115  * @cnt: Number of consecutive flow ids to allocate
1116  *
1117  * Allocate range of RX flow ids for future use, those flows can be requested
1118  * only using explicit flow id number. if @from is set to -1 it will try to find
1119  * first free range. if @from is positive value it will force allocation only
1120  * of the specified range of flows.
1121  *
1122  * Returns -ENOMEM if can't find free range.
1123  * -EEXIST if requested range is busy.
1124  * -EINVAL if wrong input values passed.
1125  * Returns flow id on success.
1126  */
1127 static int __udma_alloc_gp_rflow_range(struct udma_dev *ud, int from, int cnt)
1128 {
1129 	int start, tmp_from;
1130 	DECLARE_BITMAP(tmp, K3_UDMA_MAX_RFLOWS);
1131 
1132 	tmp_from = from;
1133 	if (tmp_from < 0)
1134 		tmp_from = ud->rchan_cnt;
1135 	/* default flows can't be allocated and accessible only by id */
1136 	if (tmp_from < ud->rchan_cnt)
1137 		return -EINVAL;
1138 
1139 	if (tmp_from + cnt > ud->rflow_cnt)
1140 		return -EINVAL;
1141 
1142 	bitmap_or(tmp, ud->rflow_gp_map, ud->rflow_gp_map_allocated,
1143 		  ud->rflow_cnt);
1144 
1145 	start = bitmap_find_next_zero_area(tmp,
1146 					   ud->rflow_cnt,
1147 					   tmp_from, cnt, 0);
1148 	if (start >= ud->rflow_cnt)
1149 		return -ENOMEM;
1150 
1151 	if (from >= 0 && start != from)
1152 		return -EEXIST;
1153 
1154 	bitmap_set(ud->rflow_gp_map_allocated, start, cnt);
1155 	return start;
1156 }
1157 
1158 static int __udma_free_gp_rflow_range(struct udma_dev *ud, int from, int cnt)
1159 {
1160 	if (from < ud->rchan_cnt)
1161 		return -EINVAL;
1162 	if (from + cnt > ud->rflow_cnt)
1163 		return -EINVAL;
1164 
1165 	bitmap_clear(ud->rflow_gp_map_allocated, from, cnt);
1166 	return 0;
1167 }
1168 
1169 static struct udma_rflow *__udma_get_rflow(struct udma_dev *ud, int id)
1170 {
1171 	/*
1172 	 * Attempt to request rflow by ID can be made for any rflow
1173 	 * if not in use with assumption that caller knows what's doing.
1174 	 * TI-SCI FW will perform additional permission check ant way, it's
1175 	 * safe
1176 	 */
1177 
1178 	if (id < 0 || id >= ud->rflow_cnt)
1179 		return ERR_PTR(-ENOENT);
1180 
1181 	if (test_bit(id, ud->rflow_in_use))
1182 		return ERR_PTR(-ENOENT);
1183 
1184 	/* GP rflow has to be allocated first */
1185 	if (!test_bit(id, ud->rflow_gp_map) &&
1186 	    !test_bit(id, ud->rflow_gp_map_allocated))
1187 		return ERR_PTR(-EINVAL);
1188 
1189 	dev_dbg(ud->dev, "get rflow%d\n", id);
1190 	set_bit(id, ud->rflow_in_use);
1191 	return &ud->rflows[id];
1192 }
1193 
1194 static void __udma_put_rflow(struct udma_dev *ud, struct udma_rflow *rflow)
1195 {
1196 	if (!test_bit(rflow->id, ud->rflow_in_use)) {
1197 		dev_err(ud->dev, "attempt to put unused rflow%d\n", rflow->id);
1198 		return;
1199 	}
1200 
1201 	dev_dbg(ud->dev, "put rflow%d\n", rflow->id);
1202 	clear_bit(rflow->id, ud->rflow_in_use);
1203 }
1204 
1205 #define UDMA_RESERVE_RESOURCE(res)					\
1206 static struct udma_##res *__udma_reserve_##res(struct udma_dev *ud,	\
1207 					       enum udma_tp_level tpl,	\
1208 					       int id)			\
1209 {									\
1210 	if (id >= 0) {							\
1211 		if (test_bit(id, ud->res##_map)) {			\
1212 			dev_err(ud->dev, "res##%d is in use\n", id);	\
1213 			return ERR_PTR(-ENOENT);			\
1214 		}							\
1215 	} else {							\
1216 		int start;						\
1217 									\
1218 		if (tpl >= ud->tpl_levels)				\
1219 			tpl = ud->tpl_levels - 1;			\
1220 									\
1221 		start = ud->tpl_start_idx[tpl];				\
1222 									\
1223 		id = find_next_zero_bit(ud->res##_map, ud->res##_cnt,	\
1224 					start);				\
1225 		if (id == ud->res##_cnt) {				\
1226 			return ERR_PTR(-ENOENT);			\
1227 		}							\
1228 	}								\
1229 									\
1230 	set_bit(id, ud->res##_map);					\
1231 	return &ud->res##s[id];						\
1232 }
1233 
1234 UDMA_RESERVE_RESOURCE(tchan);
1235 UDMA_RESERVE_RESOURCE(rchan);
1236 
1237 static int udma_get_tchan(struct udma_chan *uc)
1238 {
1239 	struct udma_dev *ud = uc->ud;
1240 
1241 	if (uc->tchan) {
1242 		dev_dbg(ud->dev, "chan%d: already have tchan%d allocated\n",
1243 			uc->id, uc->tchan->id);
1244 		return 0;
1245 	}
1246 
1247 	uc->tchan = __udma_reserve_tchan(ud, uc->config.channel_tpl, -1);
1248 
1249 	return PTR_ERR_OR_ZERO(uc->tchan);
1250 }
1251 
1252 static int udma_get_rchan(struct udma_chan *uc)
1253 {
1254 	struct udma_dev *ud = uc->ud;
1255 
1256 	if (uc->rchan) {
1257 		dev_dbg(ud->dev, "chan%d: already have rchan%d allocated\n",
1258 			uc->id, uc->rchan->id);
1259 		return 0;
1260 	}
1261 
1262 	uc->rchan = __udma_reserve_rchan(ud, uc->config.channel_tpl, -1);
1263 
1264 	return PTR_ERR_OR_ZERO(uc->rchan);
1265 }
1266 
1267 static int udma_get_chan_pair(struct udma_chan *uc)
1268 {
1269 	struct udma_dev *ud = uc->ud;
1270 	int chan_id, end;
1271 
1272 	if ((uc->tchan && uc->rchan) && uc->tchan->id == uc->rchan->id) {
1273 		dev_info(ud->dev, "chan%d: already have %d pair allocated\n",
1274 			 uc->id, uc->tchan->id);
1275 		return 0;
1276 	}
1277 
1278 	if (uc->tchan) {
1279 		dev_err(ud->dev, "chan%d: already have tchan%d allocated\n",
1280 			uc->id, uc->tchan->id);
1281 		return -EBUSY;
1282 	} else if (uc->rchan) {
1283 		dev_err(ud->dev, "chan%d: already have rchan%d allocated\n",
1284 			uc->id, uc->rchan->id);
1285 		return -EBUSY;
1286 	}
1287 
1288 	/* Can be optimized, but let's have it like this for now */
1289 	end = min(ud->tchan_cnt, ud->rchan_cnt);
1290 	/* Try to use the highest TPL channel pair for MEM_TO_MEM channels */
1291 	chan_id = ud->tpl_start_idx[ud->tpl_levels - 1];
1292 	for (; chan_id < end; chan_id++) {
1293 		if (!test_bit(chan_id, ud->tchan_map) &&
1294 		    !test_bit(chan_id, ud->rchan_map))
1295 			break;
1296 	}
1297 
1298 	if (chan_id == end)
1299 		return -ENOENT;
1300 
1301 	set_bit(chan_id, ud->tchan_map);
1302 	set_bit(chan_id, ud->rchan_map);
1303 	uc->tchan = &ud->tchans[chan_id];
1304 	uc->rchan = &ud->rchans[chan_id];
1305 
1306 	return 0;
1307 }
1308 
1309 static int udma_get_rflow(struct udma_chan *uc, int flow_id)
1310 {
1311 	struct udma_dev *ud = uc->ud;
1312 
1313 	if (!uc->rchan) {
1314 		dev_err(ud->dev, "chan%d: does not have rchan??\n", uc->id);
1315 		return -EINVAL;
1316 	}
1317 
1318 	if (uc->rflow) {
1319 		dev_dbg(ud->dev, "chan%d: already have rflow%d allocated\n",
1320 			uc->id, uc->rflow->id);
1321 		return 0;
1322 	}
1323 
1324 	uc->rflow = __udma_get_rflow(ud, flow_id);
1325 
1326 	return PTR_ERR_OR_ZERO(uc->rflow);
1327 }
1328 
1329 static void udma_put_rchan(struct udma_chan *uc)
1330 {
1331 	struct udma_dev *ud = uc->ud;
1332 
1333 	if (uc->rchan) {
1334 		dev_dbg(ud->dev, "chan%d: put rchan%d\n", uc->id,
1335 			uc->rchan->id);
1336 		clear_bit(uc->rchan->id, ud->rchan_map);
1337 		uc->rchan = NULL;
1338 	}
1339 }
1340 
1341 static void udma_put_tchan(struct udma_chan *uc)
1342 {
1343 	struct udma_dev *ud = uc->ud;
1344 
1345 	if (uc->tchan) {
1346 		dev_dbg(ud->dev, "chan%d: put tchan%d\n", uc->id,
1347 			uc->tchan->id);
1348 		clear_bit(uc->tchan->id, ud->tchan_map);
1349 		uc->tchan = NULL;
1350 	}
1351 }
1352 
1353 static void udma_put_rflow(struct udma_chan *uc)
1354 {
1355 	struct udma_dev *ud = uc->ud;
1356 
1357 	if (uc->rflow) {
1358 		dev_dbg(ud->dev, "chan%d: put rflow%d\n", uc->id,
1359 			uc->rflow->id);
1360 		__udma_put_rflow(ud, uc->rflow);
1361 		uc->rflow = NULL;
1362 	}
1363 }
1364 
1365 static void udma_free_tx_resources(struct udma_chan *uc)
1366 {
1367 	if (!uc->tchan)
1368 		return;
1369 
1370 	k3_ringacc_ring_free(uc->tchan->t_ring);
1371 	k3_ringacc_ring_free(uc->tchan->tc_ring);
1372 	uc->tchan->t_ring = NULL;
1373 	uc->tchan->tc_ring = NULL;
1374 
1375 	udma_put_tchan(uc);
1376 }
1377 
1378 static int udma_alloc_tx_resources(struct udma_chan *uc)
1379 {
1380 	struct k3_ring_cfg ring_cfg;
1381 	struct udma_dev *ud = uc->ud;
1382 	int ret;
1383 
1384 	ret = udma_get_tchan(uc);
1385 	if (ret)
1386 		return ret;
1387 
1388 	ret = k3_ringacc_request_rings_pair(ud->ringacc, uc->tchan->id, -1,
1389 					    &uc->tchan->t_ring,
1390 					    &uc->tchan->tc_ring);
1391 	if (ret) {
1392 		ret = -EBUSY;
1393 		goto err_ring;
1394 	}
1395 
1396 	memset(&ring_cfg, 0, sizeof(ring_cfg));
1397 	ring_cfg.size = K3_UDMA_DEFAULT_RING_SIZE;
1398 	ring_cfg.elm_size = K3_RINGACC_RING_ELSIZE_8;
1399 	ring_cfg.mode = K3_RINGACC_RING_MODE_MESSAGE;
1400 
1401 	ret = k3_ringacc_ring_cfg(uc->tchan->t_ring, &ring_cfg);
1402 	ret |= k3_ringacc_ring_cfg(uc->tchan->tc_ring, &ring_cfg);
1403 
1404 	if (ret)
1405 		goto err_ringcfg;
1406 
1407 	return 0;
1408 
1409 err_ringcfg:
1410 	k3_ringacc_ring_free(uc->tchan->tc_ring);
1411 	uc->tchan->tc_ring = NULL;
1412 	k3_ringacc_ring_free(uc->tchan->t_ring);
1413 	uc->tchan->t_ring = NULL;
1414 err_ring:
1415 	udma_put_tchan(uc);
1416 
1417 	return ret;
1418 }
1419 
1420 static void udma_free_rx_resources(struct udma_chan *uc)
1421 {
1422 	if (!uc->rchan)
1423 		return;
1424 
1425 	if (uc->rflow) {
1426 		struct udma_rflow *rflow = uc->rflow;
1427 
1428 		k3_ringacc_ring_free(rflow->fd_ring);
1429 		k3_ringacc_ring_free(rflow->r_ring);
1430 		rflow->fd_ring = NULL;
1431 		rflow->r_ring = NULL;
1432 
1433 		udma_put_rflow(uc);
1434 	}
1435 
1436 	udma_put_rchan(uc);
1437 }
1438 
1439 static int udma_alloc_rx_resources(struct udma_chan *uc)
1440 {
1441 	struct udma_dev *ud = uc->ud;
1442 	struct k3_ring_cfg ring_cfg;
1443 	struct udma_rflow *rflow;
1444 	int fd_ring_id;
1445 	int ret;
1446 
1447 	ret = udma_get_rchan(uc);
1448 	if (ret)
1449 		return ret;
1450 
1451 	/* For MEM_TO_MEM we don't need rflow or rings */
1452 	if (uc->config.dir == DMA_MEM_TO_MEM)
1453 		return 0;
1454 
1455 	ret = udma_get_rflow(uc, uc->rchan->id);
1456 	if (ret) {
1457 		ret = -EBUSY;
1458 		goto err_rflow;
1459 	}
1460 
1461 	rflow = uc->rflow;
1462 	fd_ring_id = ud->tchan_cnt + ud->echan_cnt + uc->rchan->id;
1463 	ret = k3_ringacc_request_rings_pair(ud->ringacc, fd_ring_id, -1,
1464 					    &rflow->fd_ring, &rflow->r_ring);
1465 	if (ret) {
1466 		ret = -EBUSY;
1467 		goto err_ring;
1468 	}
1469 
1470 	memset(&ring_cfg, 0, sizeof(ring_cfg));
1471 
1472 	if (uc->config.pkt_mode)
1473 		ring_cfg.size = SG_MAX_SEGMENTS;
1474 	else
1475 		ring_cfg.size = K3_UDMA_DEFAULT_RING_SIZE;
1476 
1477 	ring_cfg.elm_size = K3_RINGACC_RING_ELSIZE_8;
1478 	ring_cfg.mode = K3_RINGACC_RING_MODE_MESSAGE;
1479 
1480 	ret = k3_ringacc_ring_cfg(rflow->fd_ring, &ring_cfg);
1481 	ring_cfg.size = K3_UDMA_DEFAULT_RING_SIZE;
1482 	ret |= k3_ringacc_ring_cfg(rflow->r_ring, &ring_cfg);
1483 
1484 	if (ret)
1485 		goto err_ringcfg;
1486 
1487 	return 0;
1488 
1489 err_ringcfg:
1490 	k3_ringacc_ring_free(rflow->r_ring);
1491 	rflow->r_ring = NULL;
1492 	k3_ringacc_ring_free(rflow->fd_ring);
1493 	rflow->fd_ring = NULL;
1494 err_ring:
1495 	udma_put_rflow(uc);
1496 err_rflow:
1497 	udma_put_rchan(uc);
1498 
1499 	return ret;
1500 }
1501 
1502 #define TISCI_TCHAN_VALID_PARAMS (				\
1503 	TI_SCI_MSG_VALUE_RM_UDMAP_CH_PAUSE_ON_ERR_VALID |	\
1504 	TI_SCI_MSG_VALUE_RM_UDMAP_CH_TX_FILT_EINFO_VALID |	\
1505 	TI_SCI_MSG_VALUE_RM_UDMAP_CH_TX_FILT_PSWORDS_VALID |	\
1506 	TI_SCI_MSG_VALUE_RM_UDMAP_CH_CHAN_TYPE_VALID |		\
1507 	TI_SCI_MSG_VALUE_RM_UDMAP_CH_TX_SUPR_TDPKT_VALID |	\
1508 	TI_SCI_MSG_VALUE_RM_UDMAP_CH_FETCH_SIZE_VALID |		\
1509 	TI_SCI_MSG_VALUE_RM_UDMAP_CH_CQ_QNUM_VALID |		\
1510 	TI_SCI_MSG_VALUE_RM_UDMAP_CH_ATYPE_VALID)
1511 
1512 #define TISCI_RCHAN_VALID_PARAMS (				\
1513 	TI_SCI_MSG_VALUE_RM_UDMAP_CH_PAUSE_ON_ERR_VALID |	\
1514 	TI_SCI_MSG_VALUE_RM_UDMAP_CH_FETCH_SIZE_VALID |		\
1515 	TI_SCI_MSG_VALUE_RM_UDMAP_CH_CQ_QNUM_VALID |		\
1516 	TI_SCI_MSG_VALUE_RM_UDMAP_CH_CHAN_TYPE_VALID |		\
1517 	TI_SCI_MSG_VALUE_RM_UDMAP_CH_RX_IGNORE_SHORT_VALID |	\
1518 	TI_SCI_MSG_VALUE_RM_UDMAP_CH_RX_IGNORE_LONG_VALID |	\
1519 	TI_SCI_MSG_VALUE_RM_UDMAP_CH_RX_FLOWID_START_VALID |	\
1520 	TI_SCI_MSG_VALUE_RM_UDMAP_CH_RX_FLOWID_CNT_VALID |	\
1521 	TI_SCI_MSG_VALUE_RM_UDMAP_CH_ATYPE_VALID)
1522 
1523 static int udma_tisci_m2m_channel_config(struct udma_chan *uc)
1524 {
1525 	struct udma_dev *ud = uc->ud;
1526 	struct udma_tisci_rm *tisci_rm = &ud->tisci_rm;
1527 	const struct ti_sci_rm_udmap_ops *tisci_ops = tisci_rm->tisci_udmap_ops;
1528 	struct udma_tchan *tchan = uc->tchan;
1529 	struct udma_rchan *rchan = uc->rchan;
1530 	int ret = 0;
1531 
1532 	/* Non synchronized - mem to mem type of transfer */
1533 	int tc_ring = k3_ringacc_get_ring_id(tchan->tc_ring);
1534 	struct ti_sci_msg_rm_udmap_tx_ch_cfg req_tx = { 0 };
1535 	struct ti_sci_msg_rm_udmap_rx_ch_cfg req_rx = { 0 };
1536 
1537 	req_tx.valid_params = TISCI_TCHAN_VALID_PARAMS;
1538 	req_tx.nav_id = tisci_rm->tisci_dev_id;
1539 	req_tx.index = tchan->id;
1540 	req_tx.tx_chan_type = TI_SCI_RM_UDMAP_CHAN_TYPE_3RDP_BCOPY_PBRR;
1541 	req_tx.tx_fetch_size = sizeof(struct cppi5_desc_hdr_t) >> 2;
1542 	req_tx.txcq_qnum = tc_ring;
1543 	req_tx.tx_atype = ud->atype;
1544 
1545 	ret = tisci_ops->tx_ch_cfg(tisci_rm->tisci, &req_tx);
1546 	if (ret) {
1547 		dev_err(ud->dev, "tchan%d cfg failed %d\n", tchan->id, ret);
1548 		return ret;
1549 	}
1550 
1551 	req_rx.valid_params = TISCI_RCHAN_VALID_PARAMS;
1552 	req_rx.nav_id = tisci_rm->tisci_dev_id;
1553 	req_rx.index = rchan->id;
1554 	req_rx.rx_fetch_size = sizeof(struct cppi5_desc_hdr_t) >> 2;
1555 	req_rx.rxcq_qnum = tc_ring;
1556 	req_rx.rx_chan_type = TI_SCI_RM_UDMAP_CHAN_TYPE_3RDP_BCOPY_PBRR;
1557 	req_rx.rx_atype = ud->atype;
1558 
1559 	ret = tisci_ops->rx_ch_cfg(tisci_rm->tisci, &req_rx);
1560 	if (ret)
1561 		dev_err(ud->dev, "rchan%d alloc failed %d\n", rchan->id, ret);
1562 
1563 	return ret;
1564 }
1565 
1566 static int udma_tisci_tx_channel_config(struct udma_chan *uc)
1567 {
1568 	struct udma_dev *ud = uc->ud;
1569 	struct udma_tisci_rm *tisci_rm = &ud->tisci_rm;
1570 	const struct ti_sci_rm_udmap_ops *tisci_ops = tisci_rm->tisci_udmap_ops;
1571 	struct udma_tchan *tchan = uc->tchan;
1572 	int tc_ring = k3_ringacc_get_ring_id(tchan->tc_ring);
1573 	struct ti_sci_msg_rm_udmap_tx_ch_cfg req_tx = { 0 };
1574 	u32 mode, fetch_size;
1575 	int ret = 0;
1576 
1577 	if (uc->config.pkt_mode) {
1578 		mode = TI_SCI_RM_UDMAP_CHAN_TYPE_PKT_PBRR;
1579 		fetch_size = cppi5_hdesc_calc_size(uc->config.needs_epib,
1580 						   uc->config.psd_size, 0);
1581 	} else {
1582 		mode = TI_SCI_RM_UDMAP_CHAN_TYPE_3RDP_PBRR;
1583 		fetch_size = sizeof(struct cppi5_desc_hdr_t);
1584 	}
1585 
1586 	req_tx.valid_params = TISCI_TCHAN_VALID_PARAMS;
1587 	req_tx.nav_id = tisci_rm->tisci_dev_id;
1588 	req_tx.index = tchan->id;
1589 	req_tx.tx_chan_type = mode;
1590 	req_tx.tx_supr_tdpkt = uc->config.notdpkt;
1591 	req_tx.tx_fetch_size = fetch_size >> 2;
1592 	req_tx.txcq_qnum = tc_ring;
1593 	req_tx.tx_atype = uc->config.atype;
1594 
1595 	ret = tisci_ops->tx_ch_cfg(tisci_rm->tisci, &req_tx);
1596 	if (ret)
1597 		dev_err(ud->dev, "tchan%d cfg failed %d\n", tchan->id, ret);
1598 
1599 	return ret;
1600 }
1601 
1602 static int udma_tisci_rx_channel_config(struct udma_chan *uc)
1603 {
1604 	struct udma_dev *ud = uc->ud;
1605 	struct udma_tisci_rm *tisci_rm = &ud->tisci_rm;
1606 	const struct ti_sci_rm_udmap_ops *tisci_ops = tisci_rm->tisci_udmap_ops;
1607 	struct udma_rchan *rchan = uc->rchan;
1608 	int fd_ring = k3_ringacc_get_ring_id(uc->rflow->fd_ring);
1609 	int rx_ring = k3_ringacc_get_ring_id(uc->rflow->r_ring);
1610 	struct ti_sci_msg_rm_udmap_rx_ch_cfg req_rx = { 0 };
1611 	struct ti_sci_msg_rm_udmap_flow_cfg flow_req = { 0 };
1612 	u32 mode, fetch_size;
1613 	int ret = 0;
1614 
1615 	if (uc->config.pkt_mode) {
1616 		mode = TI_SCI_RM_UDMAP_CHAN_TYPE_PKT_PBRR;
1617 		fetch_size = cppi5_hdesc_calc_size(uc->config.needs_epib,
1618 						   uc->config.psd_size, 0);
1619 	} else {
1620 		mode = TI_SCI_RM_UDMAP_CHAN_TYPE_3RDP_PBRR;
1621 		fetch_size = sizeof(struct cppi5_desc_hdr_t);
1622 	}
1623 
1624 	req_rx.valid_params = TISCI_RCHAN_VALID_PARAMS;
1625 	req_rx.nav_id = tisci_rm->tisci_dev_id;
1626 	req_rx.index = rchan->id;
1627 	req_rx.rx_fetch_size =  fetch_size >> 2;
1628 	req_rx.rxcq_qnum = rx_ring;
1629 	req_rx.rx_chan_type = mode;
1630 	req_rx.rx_atype = uc->config.atype;
1631 
1632 	ret = tisci_ops->rx_ch_cfg(tisci_rm->tisci, &req_rx);
1633 	if (ret) {
1634 		dev_err(ud->dev, "rchan%d cfg failed %d\n", rchan->id, ret);
1635 		return ret;
1636 	}
1637 
1638 	flow_req.valid_params =
1639 		TI_SCI_MSG_VALUE_RM_UDMAP_FLOW_EINFO_PRESENT_VALID |
1640 		TI_SCI_MSG_VALUE_RM_UDMAP_FLOW_PSINFO_PRESENT_VALID |
1641 		TI_SCI_MSG_VALUE_RM_UDMAP_FLOW_ERROR_HANDLING_VALID |
1642 		TI_SCI_MSG_VALUE_RM_UDMAP_FLOW_DESC_TYPE_VALID |
1643 		TI_SCI_MSG_VALUE_RM_UDMAP_FLOW_DEST_QNUM_VALID |
1644 		TI_SCI_MSG_VALUE_RM_UDMAP_FLOW_SRC_TAG_HI_SEL_VALID |
1645 		TI_SCI_MSG_VALUE_RM_UDMAP_FLOW_SRC_TAG_LO_SEL_VALID |
1646 		TI_SCI_MSG_VALUE_RM_UDMAP_FLOW_DEST_TAG_HI_SEL_VALID |
1647 		TI_SCI_MSG_VALUE_RM_UDMAP_FLOW_DEST_TAG_LO_SEL_VALID |
1648 		TI_SCI_MSG_VALUE_RM_UDMAP_FLOW_FDQ0_SZ0_QNUM_VALID |
1649 		TI_SCI_MSG_VALUE_RM_UDMAP_FLOW_FDQ1_QNUM_VALID |
1650 		TI_SCI_MSG_VALUE_RM_UDMAP_FLOW_FDQ2_QNUM_VALID |
1651 		TI_SCI_MSG_VALUE_RM_UDMAP_FLOW_FDQ3_QNUM_VALID;
1652 
1653 	flow_req.nav_id = tisci_rm->tisci_dev_id;
1654 	flow_req.flow_index = rchan->id;
1655 
1656 	if (uc->config.needs_epib)
1657 		flow_req.rx_einfo_present = 1;
1658 	else
1659 		flow_req.rx_einfo_present = 0;
1660 	if (uc->config.psd_size)
1661 		flow_req.rx_psinfo_present = 1;
1662 	else
1663 		flow_req.rx_psinfo_present = 0;
1664 	flow_req.rx_error_handling = 1;
1665 	flow_req.rx_dest_qnum = rx_ring;
1666 	flow_req.rx_src_tag_hi_sel = UDMA_RFLOW_SRCTAG_NONE;
1667 	flow_req.rx_src_tag_lo_sel = UDMA_RFLOW_SRCTAG_SRC_TAG;
1668 	flow_req.rx_dest_tag_hi_sel = UDMA_RFLOW_DSTTAG_DST_TAG_HI;
1669 	flow_req.rx_dest_tag_lo_sel = UDMA_RFLOW_DSTTAG_DST_TAG_LO;
1670 	flow_req.rx_fdq0_sz0_qnum = fd_ring;
1671 	flow_req.rx_fdq1_qnum = fd_ring;
1672 	flow_req.rx_fdq2_qnum = fd_ring;
1673 	flow_req.rx_fdq3_qnum = fd_ring;
1674 
1675 	ret = tisci_ops->rx_flow_cfg(tisci_rm->tisci, &flow_req);
1676 
1677 	if (ret)
1678 		dev_err(ud->dev, "flow%d config failed: %d\n", rchan->id, ret);
1679 
1680 	return 0;
1681 }
1682 
1683 static int udma_alloc_chan_resources(struct dma_chan *chan)
1684 {
1685 	struct udma_chan *uc = to_udma_chan(chan);
1686 	struct udma_dev *ud = to_udma_dev(chan->device);
1687 	const struct udma_soc_data *soc_data = ud->soc_data;
1688 	struct k3_ring *irq_ring;
1689 	u32 irq_udma_idx;
1690 	int ret;
1691 
1692 	if (uc->config.pkt_mode || uc->config.dir == DMA_MEM_TO_MEM) {
1693 		uc->use_dma_pool = true;
1694 		/* in case of MEM_TO_MEM we have maximum of two TRs */
1695 		if (uc->config.dir == DMA_MEM_TO_MEM) {
1696 			uc->config.hdesc_size = cppi5_trdesc_calc_size(
1697 					sizeof(struct cppi5_tr_type15_t), 2);
1698 			uc->config.pkt_mode = false;
1699 		}
1700 	}
1701 
1702 	if (uc->use_dma_pool) {
1703 		uc->hdesc_pool = dma_pool_create(uc->name, ud->ddev.dev,
1704 						 uc->config.hdesc_size,
1705 						 ud->desc_align,
1706 						 0);
1707 		if (!uc->hdesc_pool) {
1708 			dev_err(ud->ddev.dev,
1709 				"Descriptor pool allocation failed\n");
1710 			uc->use_dma_pool = false;
1711 			ret = -ENOMEM;
1712 			goto err_cleanup;
1713 		}
1714 	}
1715 
1716 	/*
1717 	 * Make sure that the completion is in a known state:
1718 	 * No teardown, the channel is idle
1719 	 */
1720 	reinit_completion(&uc->teardown_completed);
1721 	complete_all(&uc->teardown_completed);
1722 	uc->state = UDMA_CHAN_IS_IDLE;
1723 
1724 	switch (uc->config.dir) {
1725 	case DMA_MEM_TO_MEM:
1726 		/* Non synchronized - mem to mem type of transfer */
1727 		dev_dbg(uc->ud->dev, "%s: chan%d as MEM-to-MEM\n", __func__,
1728 			uc->id);
1729 
1730 		ret = udma_get_chan_pair(uc);
1731 		if (ret)
1732 			goto err_cleanup;
1733 
1734 		ret = udma_alloc_tx_resources(uc);
1735 		if (ret) {
1736 			udma_put_rchan(uc);
1737 			goto err_cleanup;
1738 		}
1739 
1740 		ret = udma_alloc_rx_resources(uc);
1741 		if (ret) {
1742 			udma_free_tx_resources(uc);
1743 			goto err_cleanup;
1744 		}
1745 
1746 		uc->config.src_thread = ud->psil_base + uc->tchan->id;
1747 		uc->config.dst_thread = (ud->psil_base + uc->rchan->id) |
1748 					K3_PSIL_DST_THREAD_ID_OFFSET;
1749 
1750 		irq_ring = uc->tchan->tc_ring;
1751 		irq_udma_idx = uc->tchan->id;
1752 
1753 		ret = udma_tisci_m2m_channel_config(uc);
1754 		break;
1755 	case DMA_MEM_TO_DEV:
1756 		/* Slave transfer synchronized - mem to dev (TX) trasnfer */
1757 		dev_dbg(uc->ud->dev, "%s: chan%d as MEM-to-DEV\n", __func__,
1758 			uc->id);
1759 
1760 		ret = udma_alloc_tx_resources(uc);
1761 		if (ret)
1762 			goto err_cleanup;
1763 
1764 		uc->config.src_thread = ud->psil_base + uc->tchan->id;
1765 		uc->config.dst_thread = uc->config.remote_thread_id;
1766 		uc->config.dst_thread |= K3_PSIL_DST_THREAD_ID_OFFSET;
1767 
1768 		irq_ring = uc->tchan->tc_ring;
1769 		irq_udma_idx = uc->tchan->id;
1770 
1771 		ret = udma_tisci_tx_channel_config(uc);
1772 		break;
1773 	case DMA_DEV_TO_MEM:
1774 		/* Slave transfer synchronized - dev to mem (RX) trasnfer */
1775 		dev_dbg(uc->ud->dev, "%s: chan%d as DEV-to-MEM\n", __func__,
1776 			uc->id);
1777 
1778 		ret = udma_alloc_rx_resources(uc);
1779 		if (ret)
1780 			goto err_cleanup;
1781 
1782 		uc->config.src_thread = uc->config.remote_thread_id;
1783 		uc->config.dst_thread = (ud->psil_base + uc->rchan->id) |
1784 					K3_PSIL_DST_THREAD_ID_OFFSET;
1785 
1786 		irq_ring = uc->rflow->r_ring;
1787 		irq_udma_idx = soc_data->rchan_oes_offset + uc->rchan->id;
1788 
1789 		ret = udma_tisci_rx_channel_config(uc);
1790 		break;
1791 	default:
1792 		/* Can not happen */
1793 		dev_err(uc->ud->dev, "%s: chan%d invalid direction (%u)\n",
1794 			__func__, uc->id, uc->config.dir);
1795 		ret = -EINVAL;
1796 		goto err_cleanup;
1797 
1798 	}
1799 
1800 	/* check if the channel configuration was successful */
1801 	if (ret)
1802 		goto err_res_free;
1803 
1804 	if (udma_is_chan_running(uc)) {
1805 		dev_warn(ud->dev, "chan%d: is running!\n", uc->id);
1806 		udma_reset_chan(uc, false);
1807 		if (udma_is_chan_running(uc)) {
1808 			dev_err(ud->dev, "chan%d: won't stop!\n", uc->id);
1809 			ret = -EBUSY;
1810 			goto err_res_free;
1811 		}
1812 	}
1813 
1814 	/* PSI-L pairing */
1815 	ret = navss_psil_pair(ud, uc->config.src_thread, uc->config.dst_thread);
1816 	if (ret) {
1817 		dev_err(ud->dev, "PSI-L pairing failed: 0x%04x -> 0x%04x\n",
1818 			uc->config.src_thread, uc->config.dst_thread);
1819 		goto err_res_free;
1820 	}
1821 
1822 	uc->psil_paired = true;
1823 
1824 	uc->irq_num_ring = k3_ringacc_get_ring_irq_num(irq_ring);
1825 	if (uc->irq_num_ring <= 0) {
1826 		dev_err(ud->dev, "Failed to get ring irq (index: %u)\n",
1827 			k3_ringacc_get_ring_id(irq_ring));
1828 		ret = -EINVAL;
1829 		goto err_psi_free;
1830 	}
1831 
1832 	ret = request_irq(uc->irq_num_ring, udma_ring_irq_handler,
1833 			  IRQF_TRIGGER_HIGH, uc->name, uc);
1834 	if (ret) {
1835 		dev_err(ud->dev, "chan%d: ring irq request failed\n", uc->id);
1836 		goto err_irq_free;
1837 	}
1838 
1839 	/* Event from UDMA (TR events) only needed for slave TR mode channels */
1840 	if (is_slave_direction(uc->config.dir) && !uc->config.pkt_mode) {
1841 		uc->irq_num_udma = ti_sci_inta_msi_get_virq(ud->dev,
1842 							    irq_udma_idx);
1843 		if (uc->irq_num_udma <= 0) {
1844 			dev_err(ud->dev, "Failed to get udma irq (index: %u)\n",
1845 				irq_udma_idx);
1846 			free_irq(uc->irq_num_ring, uc);
1847 			ret = -EINVAL;
1848 			goto err_irq_free;
1849 		}
1850 
1851 		ret = request_irq(uc->irq_num_udma, udma_udma_irq_handler, 0,
1852 				  uc->name, uc);
1853 		if (ret) {
1854 			dev_err(ud->dev, "chan%d: UDMA irq request failed\n",
1855 				uc->id);
1856 			free_irq(uc->irq_num_ring, uc);
1857 			goto err_irq_free;
1858 		}
1859 	} else {
1860 		uc->irq_num_udma = 0;
1861 	}
1862 
1863 	udma_reset_rings(uc);
1864 
1865 	return 0;
1866 
1867 err_irq_free:
1868 	uc->irq_num_ring = 0;
1869 	uc->irq_num_udma = 0;
1870 err_psi_free:
1871 	navss_psil_unpair(ud, uc->config.src_thread, uc->config.dst_thread);
1872 	uc->psil_paired = false;
1873 err_res_free:
1874 	udma_free_tx_resources(uc);
1875 	udma_free_rx_resources(uc);
1876 err_cleanup:
1877 	udma_reset_uchan(uc);
1878 
1879 	if (uc->use_dma_pool) {
1880 		dma_pool_destroy(uc->hdesc_pool);
1881 		uc->use_dma_pool = false;
1882 	}
1883 
1884 	return ret;
1885 }
1886 
1887 static int udma_slave_config(struct dma_chan *chan,
1888 			     struct dma_slave_config *cfg)
1889 {
1890 	struct udma_chan *uc = to_udma_chan(chan);
1891 
1892 	memcpy(&uc->cfg, cfg, sizeof(uc->cfg));
1893 
1894 	return 0;
1895 }
1896 
1897 static struct udma_desc *udma_alloc_tr_desc(struct udma_chan *uc,
1898 					    size_t tr_size, int tr_count,
1899 					    enum dma_transfer_direction dir)
1900 {
1901 	struct udma_hwdesc *hwdesc;
1902 	struct cppi5_desc_hdr_t *tr_desc;
1903 	struct udma_desc *d;
1904 	u32 reload_count = 0;
1905 	u32 ring_id;
1906 
1907 	switch (tr_size) {
1908 	case 16:
1909 	case 32:
1910 	case 64:
1911 	case 128:
1912 		break;
1913 	default:
1914 		dev_err(uc->ud->dev, "Unsupported TR size of %zu\n", tr_size);
1915 		return NULL;
1916 	}
1917 
1918 	/* We have only one descriptor containing multiple TRs */
1919 	d = kzalloc(sizeof(*d) + sizeof(d->hwdesc[0]), GFP_NOWAIT);
1920 	if (!d)
1921 		return NULL;
1922 
1923 	d->sglen = tr_count;
1924 
1925 	d->hwdesc_count = 1;
1926 	hwdesc = &d->hwdesc[0];
1927 
1928 	/* Allocate memory for DMA ring descriptor */
1929 	if (uc->use_dma_pool) {
1930 		hwdesc->cppi5_desc_size = uc->config.hdesc_size;
1931 		hwdesc->cppi5_desc_vaddr = dma_pool_zalloc(uc->hdesc_pool,
1932 						GFP_NOWAIT,
1933 						&hwdesc->cppi5_desc_paddr);
1934 	} else {
1935 		hwdesc->cppi5_desc_size = cppi5_trdesc_calc_size(tr_size,
1936 								 tr_count);
1937 		hwdesc->cppi5_desc_size = ALIGN(hwdesc->cppi5_desc_size,
1938 						uc->ud->desc_align);
1939 		hwdesc->cppi5_desc_vaddr = dma_alloc_coherent(uc->ud->dev,
1940 						hwdesc->cppi5_desc_size,
1941 						&hwdesc->cppi5_desc_paddr,
1942 						GFP_NOWAIT);
1943 	}
1944 
1945 	if (!hwdesc->cppi5_desc_vaddr) {
1946 		kfree(d);
1947 		return NULL;
1948 	}
1949 
1950 	/* Start of the TR req records */
1951 	hwdesc->tr_req_base = hwdesc->cppi5_desc_vaddr + tr_size;
1952 	/* Start address of the TR response array */
1953 	hwdesc->tr_resp_base = hwdesc->tr_req_base + tr_size * tr_count;
1954 
1955 	tr_desc = hwdesc->cppi5_desc_vaddr;
1956 
1957 	if (uc->cyclic)
1958 		reload_count = CPPI5_INFO0_TRDESC_RLDCNT_INFINITE;
1959 
1960 	if (dir == DMA_DEV_TO_MEM)
1961 		ring_id = k3_ringacc_get_ring_id(uc->rflow->r_ring);
1962 	else
1963 		ring_id = k3_ringacc_get_ring_id(uc->tchan->tc_ring);
1964 
1965 	cppi5_trdesc_init(tr_desc, tr_count, tr_size, 0, reload_count);
1966 	cppi5_desc_set_pktids(tr_desc, uc->id,
1967 			      CPPI5_INFO1_DESC_FLOWID_DEFAULT);
1968 	cppi5_desc_set_retpolicy(tr_desc, 0, ring_id);
1969 
1970 	return d;
1971 }
1972 
1973 /**
1974  * udma_get_tr_counters - calculate TR counters for a given length
1975  * @len: Length of the trasnfer
1976  * @align_to: Preferred alignment
1977  * @tr0_cnt0: First TR icnt0
1978  * @tr0_cnt1: First TR icnt1
1979  * @tr1_cnt0: Second (if used) TR icnt0
1980  *
1981  * For len < SZ_64K only one TR is enough, tr1_cnt0 is not updated
1982  * For len >= SZ_64K two TRs are used in a simple way:
1983  * First TR: SZ_64K-alignment blocks (tr0_cnt0, tr0_cnt1)
1984  * Second TR: the remaining length (tr1_cnt0)
1985  *
1986  * Returns the number of TRs the length needs (1 or 2)
1987  * -EINVAL if the length can not be supported
1988  */
1989 static int udma_get_tr_counters(size_t len, unsigned long align_to,
1990 				u16 *tr0_cnt0, u16 *tr0_cnt1, u16 *tr1_cnt0)
1991 {
1992 	if (len < SZ_64K) {
1993 		*tr0_cnt0 = len;
1994 		*tr0_cnt1 = 1;
1995 
1996 		return 1;
1997 	}
1998 
1999 	if (align_to > 3)
2000 		align_to = 3;
2001 
2002 realign:
2003 	*tr0_cnt0 = SZ_64K - BIT(align_to);
2004 	if (len / *tr0_cnt0 >= SZ_64K) {
2005 		if (align_to) {
2006 			align_to--;
2007 			goto realign;
2008 		}
2009 		return -EINVAL;
2010 	}
2011 
2012 	*tr0_cnt1 = len / *tr0_cnt0;
2013 	*tr1_cnt0 = len % *tr0_cnt0;
2014 
2015 	return 2;
2016 }
2017 
2018 static struct udma_desc *
2019 udma_prep_slave_sg_tr(struct udma_chan *uc, struct scatterlist *sgl,
2020 		      unsigned int sglen, enum dma_transfer_direction dir,
2021 		      unsigned long tx_flags, void *context)
2022 {
2023 	struct scatterlist *sgent;
2024 	struct udma_desc *d;
2025 	struct cppi5_tr_type1_t *tr_req = NULL;
2026 	u16 tr0_cnt0, tr0_cnt1, tr1_cnt0;
2027 	unsigned int i;
2028 	size_t tr_size;
2029 	int num_tr = 0;
2030 	int tr_idx = 0;
2031 
2032 	/* estimate the number of TRs we will need */
2033 	for_each_sg(sgl, sgent, sglen, i) {
2034 		if (sg_dma_len(sgent) < SZ_64K)
2035 			num_tr++;
2036 		else
2037 			num_tr += 2;
2038 	}
2039 
2040 	/* Now allocate and setup the descriptor. */
2041 	tr_size = sizeof(struct cppi5_tr_type1_t);
2042 	d = udma_alloc_tr_desc(uc, tr_size, num_tr, dir);
2043 	if (!d)
2044 		return NULL;
2045 
2046 	d->sglen = sglen;
2047 
2048 	tr_req = d->hwdesc[0].tr_req_base;
2049 	for_each_sg(sgl, sgent, sglen, i) {
2050 		dma_addr_t sg_addr = sg_dma_address(sgent);
2051 
2052 		num_tr = udma_get_tr_counters(sg_dma_len(sgent), __ffs(sg_addr),
2053 					      &tr0_cnt0, &tr0_cnt1, &tr1_cnt0);
2054 		if (num_tr < 0) {
2055 			dev_err(uc->ud->dev, "size %u is not supported\n",
2056 				sg_dma_len(sgent));
2057 			udma_free_hwdesc(uc, d);
2058 			kfree(d);
2059 			return NULL;
2060 		}
2061 
2062 		cppi5_tr_init(&tr_req[tr_idx].flags, CPPI5_TR_TYPE1, false,
2063 			      false, CPPI5_TR_EVENT_SIZE_COMPLETION, 0);
2064 		cppi5_tr_csf_set(&tr_req[tr_idx].flags, CPPI5_TR_CSF_SUPR_EVT);
2065 
2066 		tr_req[tr_idx].addr = sg_addr;
2067 		tr_req[tr_idx].icnt0 = tr0_cnt0;
2068 		tr_req[tr_idx].icnt1 = tr0_cnt1;
2069 		tr_req[tr_idx].dim1 = tr0_cnt0;
2070 		tr_idx++;
2071 
2072 		if (num_tr == 2) {
2073 			cppi5_tr_init(&tr_req[tr_idx].flags, CPPI5_TR_TYPE1,
2074 				      false, false,
2075 				      CPPI5_TR_EVENT_SIZE_COMPLETION, 0);
2076 			cppi5_tr_csf_set(&tr_req[tr_idx].flags,
2077 					 CPPI5_TR_CSF_SUPR_EVT);
2078 
2079 			tr_req[tr_idx].addr = sg_addr + tr0_cnt1 * tr0_cnt0;
2080 			tr_req[tr_idx].icnt0 = tr1_cnt0;
2081 			tr_req[tr_idx].icnt1 = 1;
2082 			tr_req[tr_idx].dim1 = tr1_cnt0;
2083 			tr_idx++;
2084 		}
2085 
2086 		d->residue += sg_dma_len(sgent);
2087 	}
2088 
2089 	cppi5_tr_csf_set(&tr_req[tr_idx - 1].flags,
2090 			 CPPI5_TR_CSF_SUPR_EVT | CPPI5_TR_CSF_EOP);
2091 
2092 	return d;
2093 }
2094 
2095 static int udma_configure_statictr(struct udma_chan *uc, struct udma_desc *d,
2096 				   enum dma_slave_buswidth dev_width,
2097 				   u16 elcnt)
2098 {
2099 	if (uc->config.ep_type != PSIL_EP_PDMA_XY)
2100 		return 0;
2101 
2102 	/* Bus width translates to the element size (ES) */
2103 	switch (dev_width) {
2104 	case DMA_SLAVE_BUSWIDTH_1_BYTE:
2105 		d->static_tr.elsize = 0;
2106 		break;
2107 	case DMA_SLAVE_BUSWIDTH_2_BYTES:
2108 		d->static_tr.elsize = 1;
2109 		break;
2110 	case DMA_SLAVE_BUSWIDTH_3_BYTES:
2111 		d->static_tr.elsize = 2;
2112 		break;
2113 	case DMA_SLAVE_BUSWIDTH_4_BYTES:
2114 		d->static_tr.elsize = 3;
2115 		break;
2116 	case DMA_SLAVE_BUSWIDTH_8_BYTES:
2117 		d->static_tr.elsize = 4;
2118 		break;
2119 	default: /* not reached */
2120 		return -EINVAL;
2121 	}
2122 
2123 	d->static_tr.elcnt = elcnt;
2124 
2125 	/*
2126 	 * PDMA must to close the packet when the channel is in packet mode.
2127 	 * For TR mode when the channel is not cyclic we also need PDMA to close
2128 	 * the packet otherwise the transfer will stall because PDMA holds on
2129 	 * the data it has received from the peripheral.
2130 	 */
2131 	if (uc->config.pkt_mode || !uc->cyclic) {
2132 		unsigned int div = dev_width * elcnt;
2133 
2134 		if (uc->cyclic)
2135 			d->static_tr.bstcnt = d->residue / d->sglen / div;
2136 		else
2137 			d->static_tr.bstcnt = d->residue / div;
2138 
2139 		if (uc->config.dir == DMA_DEV_TO_MEM &&
2140 		    d->static_tr.bstcnt > uc->ud->match_data->statictr_z_mask)
2141 			return -EINVAL;
2142 	} else {
2143 		d->static_tr.bstcnt = 0;
2144 	}
2145 
2146 	return 0;
2147 }
2148 
2149 static struct udma_desc *
2150 udma_prep_slave_sg_pkt(struct udma_chan *uc, struct scatterlist *sgl,
2151 		       unsigned int sglen, enum dma_transfer_direction dir,
2152 		       unsigned long tx_flags, void *context)
2153 {
2154 	struct scatterlist *sgent;
2155 	struct cppi5_host_desc_t *h_desc = NULL;
2156 	struct udma_desc *d;
2157 	u32 ring_id;
2158 	unsigned int i;
2159 
2160 	d = kzalloc(struct_size(d, hwdesc, sglen), GFP_NOWAIT);
2161 	if (!d)
2162 		return NULL;
2163 
2164 	d->sglen = sglen;
2165 	d->hwdesc_count = sglen;
2166 
2167 	if (dir == DMA_DEV_TO_MEM)
2168 		ring_id = k3_ringacc_get_ring_id(uc->rflow->r_ring);
2169 	else
2170 		ring_id = k3_ringacc_get_ring_id(uc->tchan->tc_ring);
2171 
2172 	for_each_sg(sgl, sgent, sglen, i) {
2173 		struct udma_hwdesc *hwdesc = &d->hwdesc[i];
2174 		dma_addr_t sg_addr = sg_dma_address(sgent);
2175 		struct cppi5_host_desc_t *desc;
2176 		size_t sg_len = sg_dma_len(sgent);
2177 
2178 		hwdesc->cppi5_desc_vaddr = dma_pool_zalloc(uc->hdesc_pool,
2179 						GFP_NOWAIT,
2180 						&hwdesc->cppi5_desc_paddr);
2181 		if (!hwdesc->cppi5_desc_vaddr) {
2182 			dev_err(uc->ud->dev,
2183 				"descriptor%d allocation failed\n", i);
2184 
2185 			udma_free_hwdesc(uc, d);
2186 			kfree(d);
2187 			return NULL;
2188 		}
2189 
2190 		d->residue += sg_len;
2191 		hwdesc->cppi5_desc_size = uc->config.hdesc_size;
2192 		desc = hwdesc->cppi5_desc_vaddr;
2193 
2194 		if (i == 0) {
2195 			cppi5_hdesc_init(desc, 0, 0);
2196 			/* Flow and Packed ID */
2197 			cppi5_desc_set_pktids(&desc->hdr, uc->id,
2198 					      CPPI5_INFO1_DESC_FLOWID_DEFAULT);
2199 			cppi5_desc_set_retpolicy(&desc->hdr, 0, ring_id);
2200 		} else {
2201 			cppi5_hdesc_reset_hbdesc(desc);
2202 			cppi5_desc_set_retpolicy(&desc->hdr, 0, 0xffff);
2203 		}
2204 
2205 		/* attach the sg buffer to the descriptor */
2206 		cppi5_hdesc_attach_buf(desc, sg_addr, sg_len, sg_addr, sg_len);
2207 
2208 		/* Attach link as host buffer descriptor */
2209 		if (h_desc)
2210 			cppi5_hdesc_link_hbdesc(h_desc,
2211 						hwdesc->cppi5_desc_paddr);
2212 
2213 		if (dir == DMA_MEM_TO_DEV)
2214 			h_desc = desc;
2215 	}
2216 
2217 	if (d->residue >= SZ_4M) {
2218 		dev_err(uc->ud->dev,
2219 			"%s: Transfer size %u is over the supported 4M range\n",
2220 			__func__, d->residue);
2221 		udma_free_hwdesc(uc, d);
2222 		kfree(d);
2223 		return NULL;
2224 	}
2225 
2226 	h_desc = d->hwdesc[0].cppi5_desc_vaddr;
2227 	cppi5_hdesc_set_pktlen(h_desc, d->residue);
2228 
2229 	return d;
2230 }
2231 
2232 static int udma_attach_metadata(struct dma_async_tx_descriptor *desc,
2233 				void *data, size_t len)
2234 {
2235 	struct udma_desc *d = to_udma_desc(desc);
2236 	struct udma_chan *uc = to_udma_chan(desc->chan);
2237 	struct cppi5_host_desc_t *h_desc;
2238 	u32 psd_size = len;
2239 	u32 flags = 0;
2240 
2241 	if (!uc->config.pkt_mode || !uc->config.metadata_size)
2242 		return -ENOTSUPP;
2243 
2244 	if (!data || len > uc->config.metadata_size)
2245 		return -EINVAL;
2246 
2247 	if (uc->config.needs_epib && len < CPPI5_INFO0_HDESC_EPIB_SIZE)
2248 		return -EINVAL;
2249 
2250 	h_desc = d->hwdesc[0].cppi5_desc_vaddr;
2251 	if (d->dir == DMA_MEM_TO_DEV)
2252 		memcpy(h_desc->epib, data, len);
2253 
2254 	if (uc->config.needs_epib)
2255 		psd_size -= CPPI5_INFO0_HDESC_EPIB_SIZE;
2256 
2257 	d->metadata = data;
2258 	d->metadata_size = len;
2259 	if (uc->config.needs_epib)
2260 		flags |= CPPI5_INFO0_HDESC_EPIB_PRESENT;
2261 
2262 	cppi5_hdesc_update_flags(h_desc, flags);
2263 	cppi5_hdesc_update_psdata_size(h_desc, psd_size);
2264 
2265 	return 0;
2266 }
2267 
2268 static void *udma_get_metadata_ptr(struct dma_async_tx_descriptor *desc,
2269 				   size_t *payload_len, size_t *max_len)
2270 {
2271 	struct udma_desc *d = to_udma_desc(desc);
2272 	struct udma_chan *uc = to_udma_chan(desc->chan);
2273 	struct cppi5_host_desc_t *h_desc;
2274 
2275 	if (!uc->config.pkt_mode || !uc->config.metadata_size)
2276 		return ERR_PTR(-ENOTSUPP);
2277 
2278 	h_desc = d->hwdesc[0].cppi5_desc_vaddr;
2279 
2280 	*max_len = uc->config.metadata_size;
2281 
2282 	*payload_len = cppi5_hdesc_epib_present(&h_desc->hdr) ?
2283 		       CPPI5_INFO0_HDESC_EPIB_SIZE : 0;
2284 	*payload_len += cppi5_hdesc_get_psdata_size(h_desc);
2285 
2286 	return h_desc->epib;
2287 }
2288 
2289 static int udma_set_metadata_len(struct dma_async_tx_descriptor *desc,
2290 				 size_t payload_len)
2291 {
2292 	struct udma_desc *d = to_udma_desc(desc);
2293 	struct udma_chan *uc = to_udma_chan(desc->chan);
2294 	struct cppi5_host_desc_t *h_desc;
2295 	u32 psd_size = payload_len;
2296 	u32 flags = 0;
2297 
2298 	if (!uc->config.pkt_mode || !uc->config.metadata_size)
2299 		return -ENOTSUPP;
2300 
2301 	if (payload_len > uc->config.metadata_size)
2302 		return -EINVAL;
2303 
2304 	if (uc->config.needs_epib && payload_len < CPPI5_INFO0_HDESC_EPIB_SIZE)
2305 		return -EINVAL;
2306 
2307 	h_desc = d->hwdesc[0].cppi5_desc_vaddr;
2308 
2309 	if (uc->config.needs_epib) {
2310 		psd_size -= CPPI5_INFO0_HDESC_EPIB_SIZE;
2311 		flags |= CPPI5_INFO0_HDESC_EPIB_PRESENT;
2312 	}
2313 
2314 	cppi5_hdesc_update_flags(h_desc, flags);
2315 	cppi5_hdesc_update_psdata_size(h_desc, psd_size);
2316 
2317 	return 0;
2318 }
2319 
2320 static struct dma_descriptor_metadata_ops metadata_ops = {
2321 	.attach = udma_attach_metadata,
2322 	.get_ptr = udma_get_metadata_ptr,
2323 	.set_len = udma_set_metadata_len,
2324 };
2325 
2326 static struct dma_async_tx_descriptor *
2327 udma_prep_slave_sg(struct dma_chan *chan, struct scatterlist *sgl,
2328 		   unsigned int sglen, enum dma_transfer_direction dir,
2329 		   unsigned long tx_flags, void *context)
2330 {
2331 	struct udma_chan *uc = to_udma_chan(chan);
2332 	enum dma_slave_buswidth dev_width;
2333 	struct udma_desc *d;
2334 	u32 burst;
2335 
2336 	if (dir != uc->config.dir) {
2337 		dev_err(chan->device->dev,
2338 			"%s: chan%d is for %s, not supporting %s\n",
2339 			__func__, uc->id,
2340 			dmaengine_get_direction_text(uc->config.dir),
2341 			dmaengine_get_direction_text(dir));
2342 		return NULL;
2343 	}
2344 
2345 	if (dir == DMA_DEV_TO_MEM) {
2346 		dev_width = uc->cfg.src_addr_width;
2347 		burst = uc->cfg.src_maxburst;
2348 	} else if (dir == DMA_MEM_TO_DEV) {
2349 		dev_width = uc->cfg.dst_addr_width;
2350 		burst = uc->cfg.dst_maxburst;
2351 	} else {
2352 		dev_err(chan->device->dev, "%s: bad direction?\n", __func__);
2353 		return NULL;
2354 	}
2355 
2356 	if (!burst)
2357 		burst = 1;
2358 
2359 	if (uc->config.pkt_mode)
2360 		d = udma_prep_slave_sg_pkt(uc, sgl, sglen, dir, tx_flags,
2361 					   context);
2362 	else
2363 		d = udma_prep_slave_sg_tr(uc, sgl, sglen, dir, tx_flags,
2364 					  context);
2365 
2366 	if (!d)
2367 		return NULL;
2368 
2369 	d->dir = dir;
2370 	d->desc_idx = 0;
2371 	d->tr_idx = 0;
2372 
2373 	/* static TR for remote PDMA */
2374 	if (udma_configure_statictr(uc, d, dev_width, burst)) {
2375 		dev_err(uc->ud->dev,
2376 			"%s: StaticTR Z is limited to maximum 4095 (%u)\n",
2377 			__func__, d->static_tr.bstcnt);
2378 
2379 		udma_free_hwdesc(uc, d);
2380 		kfree(d);
2381 		return NULL;
2382 	}
2383 
2384 	if (uc->config.metadata_size)
2385 		d->vd.tx.metadata_ops = &metadata_ops;
2386 
2387 	return vchan_tx_prep(&uc->vc, &d->vd, tx_flags);
2388 }
2389 
2390 static struct udma_desc *
2391 udma_prep_dma_cyclic_tr(struct udma_chan *uc, dma_addr_t buf_addr,
2392 			size_t buf_len, size_t period_len,
2393 			enum dma_transfer_direction dir, unsigned long flags)
2394 {
2395 	struct udma_desc *d;
2396 	size_t tr_size, period_addr;
2397 	struct cppi5_tr_type1_t *tr_req;
2398 	unsigned int periods = buf_len / period_len;
2399 	u16 tr0_cnt0, tr0_cnt1, tr1_cnt0;
2400 	unsigned int i;
2401 	int num_tr;
2402 
2403 	num_tr = udma_get_tr_counters(period_len, __ffs(buf_addr), &tr0_cnt0,
2404 				      &tr0_cnt1, &tr1_cnt0);
2405 	if (num_tr < 0) {
2406 		dev_err(uc->ud->dev, "size %zu is not supported\n",
2407 			period_len);
2408 		return NULL;
2409 	}
2410 
2411 	/* Now allocate and setup the descriptor. */
2412 	tr_size = sizeof(struct cppi5_tr_type1_t);
2413 	d = udma_alloc_tr_desc(uc, tr_size, periods * num_tr, dir);
2414 	if (!d)
2415 		return NULL;
2416 
2417 	tr_req = d->hwdesc[0].tr_req_base;
2418 	period_addr = buf_addr;
2419 	for (i = 0; i < periods; i++) {
2420 		int tr_idx = i * num_tr;
2421 
2422 		cppi5_tr_init(&tr_req[tr_idx].flags, CPPI5_TR_TYPE1, false,
2423 			      false, CPPI5_TR_EVENT_SIZE_COMPLETION, 0);
2424 
2425 		tr_req[tr_idx].addr = period_addr;
2426 		tr_req[tr_idx].icnt0 = tr0_cnt0;
2427 		tr_req[tr_idx].icnt1 = tr0_cnt1;
2428 		tr_req[tr_idx].dim1 = tr0_cnt0;
2429 
2430 		if (num_tr == 2) {
2431 			cppi5_tr_csf_set(&tr_req[tr_idx].flags,
2432 					 CPPI5_TR_CSF_SUPR_EVT);
2433 			tr_idx++;
2434 
2435 			cppi5_tr_init(&tr_req[tr_idx].flags, CPPI5_TR_TYPE1,
2436 				      false, false,
2437 				      CPPI5_TR_EVENT_SIZE_COMPLETION, 0);
2438 
2439 			tr_req[tr_idx].addr = period_addr + tr0_cnt1 * tr0_cnt0;
2440 			tr_req[tr_idx].icnt0 = tr1_cnt0;
2441 			tr_req[tr_idx].icnt1 = 1;
2442 			tr_req[tr_idx].dim1 = tr1_cnt0;
2443 		}
2444 
2445 		if (!(flags & DMA_PREP_INTERRUPT))
2446 			cppi5_tr_csf_set(&tr_req[tr_idx].flags,
2447 					 CPPI5_TR_CSF_SUPR_EVT);
2448 
2449 		period_addr += period_len;
2450 	}
2451 
2452 	return d;
2453 }
2454 
2455 static struct udma_desc *
2456 udma_prep_dma_cyclic_pkt(struct udma_chan *uc, dma_addr_t buf_addr,
2457 			 size_t buf_len, size_t period_len,
2458 			 enum dma_transfer_direction dir, unsigned long flags)
2459 {
2460 	struct udma_desc *d;
2461 	u32 ring_id;
2462 	int i;
2463 	int periods = buf_len / period_len;
2464 
2465 	if (periods > (K3_UDMA_DEFAULT_RING_SIZE - 1))
2466 		return NULL;
2467 
2468 	if (period_len >= SZ_4M)
2469 		return NULL;
2470 
2471 	d = kzalloc(struct_size(d, hwdesc, periods), GFP_NOWAIT);
2472 	if (!d)
2473 		return NULL;
2474 
2475 	d->hwdesc_count = periods;
2476 
2477 	/* TODO: re-check this... */
2478 	if (dir == DMA_DEV_TO_MEM)
2479 		ring_id = k3_ringacc_get_ring_id(uc->rflow->r_ring);
2480 	else
2481 		ring_id = k3_ringacc_get_ring_id(uc->tchan->tc_ring);
2482 
2483 	for (i = 0; i < periods; i++) {
2484 		struct udma_hwdesc *hwdesc = &d->hwdesc[i];
2485 		dma_addr_t period_addr = buf_addr + (period_len * i);
2486 		struct cppi5_host_desc_t *h_desc;
2487 
2488 		hwdesc->cppi5_desc_vaddr = dma_pool_zalloc(uc->hdesc_pool,
2489 						GFP_NOWAIT,
2490 						&hwdesc->cppi5_desc_paddr);
2491 		if (!hwdesc->cppi5_desc_vaddr) {
2492 			dev_err(uc->ud->dev,
2493 				"descriptor%d allocation failed\n", i);
2494 
2495 			udma_free_hwdesc(uc, d);
2496 			kfree(d);
2497 			return NULL;
2498 		}
2499 
2500 		hwdesc->cppi5_desc_size = uc->config.hdesc_size;
2501 		h_desc = hwdesc->cppi5_desc_vaddr;
2502 
2503 		cppi5_hdesc_init(h_desc, 0, 0);
2504 		cppi5_hdesc_set_pktlen(h_desc, period_len);
2505 
2506 		/* Flow and Packed ID */
2507 		cppi5_desc_set_pktids(&h_desc->hdr, uc->id,
2508 				      CPPI5_INFO1_DESC_FLOWID_DEFAULT);
2509 		cppi5_desc_set_retpolicy(&h_desc->hdr, 0, ring_id);
2510 
2511 		/* attach each period to a new descriptor */
2512 		cppi5_hdesc_attach_buf(h_desc,
2513 				       period_addr, period_len,
2514 				       period_addr, period_len);
2515 	}
2516 
2517 	return d;
2518 }
2519 
2520 static struct dma_async_tx_descriptor *
2521 udma_prep_dma_cyclic(struct dma_chan *chan, dma_addr_t buf_addr, size_t buf_len,
2522 		     size_t period_len, enum dma_transfer_direction dir,
2523 		     unsigned long flags)
2524 {
2525 	struct udma_chan *uc = to_udma_chan(chan);
2526 	enum dma_slave_buswidth dev_width;
2527 	struct udma_desc *d;
2528 	u32 burst;
2529 
2530 	if (dir != uc->config.dir) {
2531 		dev_err(chan->device->dev,
2532 			"%s: chan%d is for %s, not supporting %s\n",
2533 			__func__, uc->id,
2534 			dmaengine_get_direction_text(uc->config.dir),
2535 			dmaengine_get_direction_text(dir));
2536 		return NULL;
2537 	}
2538 
2539 	uc->cyclic = true;
2540 
2541 	if (dir == DMA_DEV_TO_MEM) {
2542 		dev_width = uc->cfg.src_addr_width;
2543 		burst = uc->cfg.src_maxburst;
2544 	} else if (dir == DMA_MEM_TO_DEV) {
2545 		dev_width = uc->cfg.dst_addr_width;
2546 		burst = uc->cfg.dst_maxburst;
2547 	} else {
2548 		dev_err(uc->ud->dev, "%s: bad direction?\n", __func__);
2549 		return NULL;
2550 	}
2551 
2552 	if (!burst)
2553 		burst = 1;
2554 
2555 	if (uc->config.pkt_mode)
2556 		d = udma_prep_dma_cyclic_pkt(uc, buf_addr, buf_len, period_len,
2557 					     dir, flags);
2558 	else
2559 		d = udma_prep_dma_cyclic_tr(uc, buf_addr, buf_len, period_len,
2560 					    dir, flags);
2561 
2562 	if (!d)
2563 		return NULL;
2564 
2565 	d->sglen = buf_len / period_len;
2566 
2567 	d->dir = dir;
2568 	d->residue = buf_len;
2569 
2570 	/* static TR for remote PDMA */
2571 	if (udma_configure_statictr(uc, d, dev_width, burst)) {
2572 		dev_err(uc->ud->dev,
2573 			"%s: StaticTR Z is limited to maximum 4095 (%u)\n",
2574 			__func__, d->static_tr.bstcnt);
2575 
2576 		udma_free_hwdesc(uc, d);
2577 		kfree(d);
2578 		return NULL;
2579 	}
2580 
2581 	if (uc->config.metadata_size)
2582 		d->vd.tx.metadata_ops = &metadata_ops;
2583 
2584 	return vchan_tx_prep(&uc->vc, &d->vd, flags);
2585 }
2586 
2587 static struct dma_async_tx_descriptor *
2588 udma_prep_dma_memcpy(struct dma_chan *chan, dma_addr_t dest, dma_addr_t src,
2589 		     size_t len, unsigned long tx_flags)
2590 {
2591 	struct udma_chan *uc = to_udma_chan(chan);
2592 	struct udma_desc *d;
2593 	struct cppi5_tr_type15_t *tr_req;
2594 	int num_tr;
2595 	size_t tr_size = sizeof(struct cppi5_tr_type15_t);
2596 	u16 tr0_cnt0, tr0_cnt1, tr1_cnt0;
2597 
2598 	if (uc->config.dir != DMA_MEM_TO_MEM) {
2599 		dev_err(chan->device->dev,
2600 			"%s: chan%d is for %s, not supporting %s\n",
2601 			__func__, uc->id,
2602 			dmaengine_get_direction_text(uc->config.dir),
2603 			dmaengine_get_direction_text(DMA_MEM_TO_MEM));
2604 		return NULL;
2605 	}
2606 
2607 	num_tr = udma_get_tr_counters(len, __ffs(src | dest), &tr0_cnt0,
2608 				      &tr0_cnt1, &tr1_cnt0);
2609 	if (num_tr < 0) {
2610 		dev_err(uc->ud->dev, "size %zu is not supported\n",
2611 			len);
2612 		return NULL;
2613 	}
2614 
2615 	d = udma_alloc_tr_desc(uc, tr_size, num_tr, DMA_MEM_TO_MEM);
2616 	if (!d)
2617 		return NULL;
2618 
2619 	d->dir = DMA_MEM_TO_MEM;
2620 	d->desc_idx = 0;
2621 	d->tr_idx = 0;
2622 	d->residue = len;
2623 
2624 	tr_req = d->hwdesc[0].tr_req_base;
2625 
2626 	cppi5_tr_init(&tr_req[0].flags, CPPI5_TR_TYPE15, false, true,
2627 		      CPPI5_TR_EVENT_SIZE_COMPLETION, 0);
2628 	cppi5_tr_csf_set(&tr_req[0].flags, CPPI5_TR_CSF_SUPR_EVT);
2629 
2630 	tr_req[0].addr = src;
2631 	tr_req[0].icnt0 = tr0_cnt0;
2632 	tr_req[0].icnt1 = tr0_cnt1;
2633 	tr_req[0].icnt2 = 1;
2634 	tr_req[0].icnt3 = 1;
2635 	tr_req[0].dim1 = tr0_cnt0;
2636 
2637 	tr_req[0].daddr = dest;
2638 	tr_req[0].dicnt0 = tr0_cnt0;
2639 	tr_req[0].dicnt1 = tr0_cnt1;
2640 	tr_req[0].dicnt2 = 1;
2641 	tr_req[0].dicnt3 = 1;
2642 	tr_req[0].ddim1 = tr0_cnt0;
2643 
2644 	if (num_tr == 2) {
2645 		cppi5_tr_init(&tr_req[1].flags, CPPI5_TR_TYPE15, false, true,
2646 			      CPPI5_TR_EVENT_SIZE_COMPLETION, 0);
2647 		cppi5_tr_csf_set(&tr_req[1].flags, CPPI5_TR_CSF_SUPR_EVT);
2648 
2649 		tr_req[1].addr = src + tr0_cnt1 * tr0_cnt0;
2650 		tr_req[1].icnt0 = tr1_cnt0;
2651 		tr_req[1].icnt1 = 1;
2652 		tr_req[1].icnt2 = 1;
2653 		tr_req[1].icnt3 = 1;
2654 
2655 		tr_req[1].daddr = dest + tr0_cnt1 * tr0_cnt0;
2656 		tr_req[1].dicnt0 = tr1_cnt0;
2657 		tr_req[1].dicnt1 = 1;
2658 		tr_req[1].dicnt2 = 1;
2659 		tr_req[1].dicnt3 = 1;
2660 	}
2661 
2662 	cppi5_tr_csf_set(&tr_req[num_tr - 1].flags,
2663 			 CPPI5_TR_CSF_SUPR_EVT | CPPI5_TR_CSF_EOP);
2664 
2665 	if (uc->config.metadata_size)
2666 		d->vd.tx.metadata_ops = &metadata_ops;
2667 
2668 	return vchan_tx_prep(&uc->vc, &d->vd, tx_flags);
2669 }
2670 
2671 static void udma_issue_pending(struct dma_chan *chan)
2672 {
2673 	struct udma_chan *uc = to_udma_chan(chan);
2674 	unsigned long flags;
2675 
2676 	spin_lock_irqsave(&uc->vc.lock, flags);
2677 
2678 	/* If we have something pending and no active descriptor, then */
2679 	if (vchan_issue_pending(&uc->vc) && !uc->desc) {
2680 		/*
2681 		 * start a descriptor if the channel is NOT [marked as
2682 		 * terminating _and_ it is still running (teardown has not
2683 		 * completed yet)].
2684 		 */
2685 		if (!(uc->state == UDMA_CHAN_IS_TERMINATING &&
2686 		      udma_is_chan_running(uc)))
2687 			udma_start(uc);
2688 	}
2689 
2690 	spin_unlock_irqrestore(&uc->vc.lock, flags);
2691 }
2692 
2693 static enum dma_status udma_tx_status(struct dma_chan *chan,
2694 				      dma_cookie_t cookie,
2695 				      struct dma_tx_state *txstate)
2696 {
2697 	struct udma_chan *uc = to_udma_chan(chan);
2698 	enum dma_status ret;
2699 	unsigned long flags;
2700 
2701 	spin_lock_irqsave(&uc->vc.lock, flags);
2702 
2703 	ret = dma_cookie_status(chan, cookie, txstate);
2704 
2705 	if (!udma_is_chan_running(uc))
2706 		ret = DMA_COMPLETE;
2707 
2708 	if (ret == DMA_IN_PROGRESS && udma_is_chan_paused(uc))
2709 		ret = DMA_PAUSED;
2710 
2711 	if (ret == DMA_COMPLETE || !txstate)
2712 		goto out;
2713 
2714 	if (uc->desc && uc->desc->vd.tx.cookie == cookie) {
2715 		u32 peer_bcnt = 0;
2716 		u32 bcnt = 0;
2717 		u32 residue = uc->desc->residue;
2718 		u32 delay = 0;
2719 
2720 		if (uc->desc->dir == DMA_MEM_TO_DEV) {
2721 			bcnt = udma_tchanrt_read(uc, UDMA_CHAN_RT_SBCNT_REG);
2722 
2723 			if (uc->config.ep_type != PSIL_EP_NATIVE) {
2724 				peer_bcnt = udma_tchanrt_read(uc,
2725 						UDMA_CHAN_RT_PEER_BCNT_REG);
2726 
2727 				if (bcnt > peer_bcnt)
2728 					delay = bcnt - peer_bcnt;
2729 			}
2730 		} else if (uc->desc->dir == DMA_DEV_TO_MEM) {
2731 			bcnt = udma_rchanrt_read(uc, UDMA_CHAN_RT_BCNT_REG);
2732 
2733 			if (uc->config.ep_type != PSIL_EP_NATIVE) {
2734 				peer_bcnt = udma_rchanrt_read(uc,
2735 						UDMA_CHAN_RT_PEER_BCNT_REG);
2736 
2737 				if (peer_bcnt > bcnt)
2738 					delay = peer_bcnt - bcnt;
2739 			}
2740 		} else {
2741 			bcnt = udma_tchanrt_read(uc, UDMA_CHAN_RT_BCNT_REG);
2742 		}
2743 
2744 		bcnt -= uc->bcnt;
2745 		if (bcnt && !(bcnt % uc->desc->residue))
2746 			residue = 0;
2747 		else
2748 			residue -= bcnt % uc->desc->residue;
2749 
2750 		if (!residue && (uc->config.dir == DMA_DEV_TO_MEM || !delay)) {
2751 			ret = DMA_COMPLETE;
2752 			delay = 0;
2753 		}
2754 
2755 		dma_set_residue(txstate, residue);
2756 		dma_set_in_flight_bytes(txstate, delay);
2757 
2758 	} else {
2759 		ret = DMA_COMPLETE;
2760 	}
2761 
2762 out:
2763 	spin_unlock_irqrestore(&uc->vc.lock, flags);
2764 	return ret;
2765 }
2766 
2767 static int udma_pause(struct dma_chan *chan)
2768 {
2769 	struct udma_chan *uc = to_udma_chan(chan);
2770 
2771 	/* pause the channel */
2772 	switch (uc->config.dir) {
2773 	case DMA_DEV_TO_MEM:
2774 		udma_rchanrt_update_bits(uc, UDMA_CHAN_RT_PEER_RT_EN_REG,
2775 					 UDMA_PEER_RT_EN_PAUSE,
2776 					 UDMA_PEER_RT_EN_PAUSE);
2777 		break;
2778 	case DMA_MEM_TO_DEV:
2779 		udma_tchanrt_update_bits(uc, UDMA_CHAN_RT_PEER_RT_EN_REG,
2780 					 UDMA_PEER_RT_EN_PAUSE,
2781 					 UDMA_PEER_RT_EN_PAUSE);
2782 		break;
2783 	case DMA_MEM_TO_MEM:
2784 		udma_tchanrt_update_bits(uc, UDMA_CHAN_RT_CTL_REG,
2785 					 UDMA_CHAN_RT_CTL_PAUSE,
2786 					 UDMA_CHAN_RT_CTL_PAUSE);
2787 		break;
2788 	default:
2789 		return -EINVAL;
2790 	}
2791 
2792 	return 0;
2793 }
2794 
2795 static int udma_resume(struct dma_chan *chan)
2796 {
2797 	struct udma_chan *uc = to_udma_chan(chan);
2798 
2799 	/* resume the channel */
2800 	switch (uc->config.dir) {
2801 	case DMA_DEV_TO_MEM:
2802 		udma_rchanrt_update_bits(uc, UDMA_CHAN_RT_PEER_RT_EN_REG,
2803 					 UDMA_PEER_RT_EN_PAUSE, 0);
2804 
2805 		break;
2806 	case DMA_MEM_TO_DEV:
2807 		udma_tchanrt_update_bits(uc, UDMA_CHAN_RT_PEER_RT_EN_REG,
2808 					 UDMA_PEER_RT_EN_PAUSE, 0);
2809 		break;
2810 	case DMA_MEM_TO_MEM:
2811 		udma_tchanrt_update_bits(uc, UDMA_CHAN_RT_CTL_REG,
2812 					 UDMA_CHAN_RT_CTL_PAUSE, 0);
2813 		break;
2814 	default:
2815 		return -EINVAL;
2816 	}
2817 
2818 	return 0;
2819 }
2820 
2821 static int udma_terminate_all(struct dma_chan *chan)
2822 {
2823 	struct udma_chan *uc = to_udma_chan(chan);
2824 	unsigned long flags;
2825 	LIST_HEAD(head);
2826 
2827 	spin_lock_irqsave(&uc->vc.lock, flags);
2828 
2829 	if (udma_is_chan_running(uc))
2830 		udma_stop(uc);
2831 
2832 	if (uc->desc) {
2833 		uc->terminated_desc = uc->desc;
2834 		uc->desc = NULL;
2835 		uc->terminated_desc->terminated = true;
2836 		cancel_delayed_work(&uc->tx_drain.work);
2837 	}
2838 
2839 	uc->paused = false;
2840 
2841 	vchan_get_all_descriptors(&uc->vc, &head);
2842 	spin_unlock_irqrestore(&uc->vc.lock, flags);
2843 	vchan_dma_desc_free_list(&uc->vc, &head);
2844 
2845 	return 0;
2846 }
2847 
2848 static void udma_synchronize(struct dma_chan *chan)
2849 {
2850 	struct udma_chan *uc = to_udma_chan(chan);
2851 	unsigned long timeout = msecs_to_jiffies(1000);
2852 
2853 	vchan_synchronize(&uc->vc);
2854 
2855 	if (uc->state == UDMA_CHAN_IS_TERMINATING) {
2856 		timeout = wait_for_completion_timeout(&uc->teardown_completed,
2857 						      timeout);
2858 		if (!timeout) {
2859 			dev_warn(uc->ud->dev, "chan%d teardown timeout!\n",
2860 				 uc->id);
2861 			udma_dump_chan_stdata(uc);
2862 			udma_reset_chan(uc, true);
2863 		}
2864 	}
2865 
2866 	udma_reset_chan(uc, false);
2867 	if (udma_is_chan_running(uc))
2868 		dev_warn(uc->ud->dev, "chan%d refused to stop!\n", uc->id);
2869 
2870 	cancel_delayed_work_sync(&uc->tx_drain.work);
2871 	udma_reset_rings(uc);
2872 }
2873 
2874 static void udma_desc_pre_callback(struct virt_dma_chan *vc,
2875 				   struct virt_dma_desc *vd,
2876 				   struct dmaengine_result *result)
2877 {
2878 	struct udma_chan *uc = to_udma_chan(&vc->chan);
2879 	struct udma_desc *d;
2880 
2881 	if (!vd)
2882 		return;
2883 
2884 	d = to_udma_desc(&vd->tx);
2885 
2886 	if (d->metadata_size)
2887 		udma_fetch_epib(uc, d);
2888 
2889 	/* Provide residue information for the client */
2890 	if (result) {
2891 		void *desc_vaddr = udma_curr_cppi5_desc_vaddr(d, d->desc_idx);
2892 
2893 		if (cppi5_desc_get_type(desc_vaddr) ==
2894 		    CPPI5_INFO0_DESC_TYPE_VAL_HOST) {
2895 			result->residue = d->residue -
2896 					  cppi5_hdesc_get_pktlen(desc_vaddr);
2897 			if (result->residue)
2898 				result->result = DMA_TRANS_ABORTED;
2899 			else
2900 				result->result = DMA_TRANS_NOERROR;
2901 		} else {
2902 			result->residue = 0;
2903 			result->result = DMA_TRANS_NOERROR;
2904 		}
2905 	}
2906 }
2907 
2908 /*
2909  * This tasklet handles the completion of a DMA descriptor by
2910  * calling its callback and freeing it.
2911  */
2912 static void udma_vchan_complete(struct tasklet_struct *t)
2913 {
2914 	struct virt_dma_chan *vc = from_tasklet(vc, t, task);
2915 	struct virt_dma_desc *vd, *_vd;
2916 	struct dmaengine_desc_callback cb;
2917 	LIST_HEAD(head);
2918 
2919 	spin_lock_irq(&vc->lock);
2920 	list_splice_tail_init(&vc->desc_completed, &head);
2921 	vd = vc->cyclic;
2922 	if (vd) {
2923 		vc->cyclic = NULL;
2924 		dmaengine_desc_get_callback(&vd->tx, &cb);
2925 	} else {
2926 		memset(&cb, 0, sizeof(cb));
2927 	}
2928 	spin_unlock_irq(&vc->lock);
2929 
2930 	udma_desc_pre_callback(vc, vd, NULL);
2931 	dmaengine_desc_callback_invoke(&cb, NULL);
2932 
2933 	list_for_each_entry_safe(vd, _vd, &head, node) {
2934 		struct dmaengine_result result;
2935 
2936 		dmaengine_desc_get_callback(&vd->tx, &cb);
2937 
2938 		list_del(&vd->node);
2939 
2940 		udma_desc_pre_callback(vc, vd, &result);
2941 		dmaengine_desc_callback_invoke(&cb, &result);
2942 
2943 		vchan_vdesc_fini(vd);
2944 	}
2945 }
2946 
2947 static void udma_free_chan_resources(struct dma_chan *chan)
2948 {
2949 	struct udma_chan *uc = to_udma_chan(chan);
2950 	struct udma_dev *ud = to_udma_dev(chan->device);
2951 
2952 	udma_terminate_all(chan);
2953 	if (uc->terminated_desc) {
2954 		udma_reset_chan(uc, false);
2955 		udma_reset_rings(uc);
2956 	}
2957 
2958 	cancel_delayed_work_sync(&uc->tx_drain.work);
2959 
2960 	if (uc->irq_num_ring > 0) {
2961 		free_irq(uc->irq_num_ring, uc);
2962 
2963 		uc->irq_num_ring = 0;
2964 	}
2965 	if (uc->irq_num_udma > 0) {
2966 		free_irq(uc->irq_num_udma, uc);
2967 
2968 		uc->irq_num_udma = 0;
2969 	}
2970 
2971 	/* Release PSI-L pairing */
2972 	if (uc->psil_paired) {
2973 		navss_psil_unpair(ud, uc->config.src_thread,
2974 				  uc->config.dst_thread);
2975 		uc->psil_paired = false;
2976 	}
2977 
2978 	vchan_free_chan_resources(&uc->vc);
2979 	tasklet_kill(&uc->vc.task);
2980 
2981 	udma_free_tx_resources(uc);
2982 	udma_free_rx_resources(uc);
2983 	udma_reset_uchan(uc);
2984 
2985 	if (uc->use_dma_pool) {
2986 		dma_pool_destroy(uc->hdesc_pool);
2987 		uc->use_dma_pool = false;
2988 	}
2989 }
2990 
2991 static struct platform_driver udma_driver;
2992 
2993 struct udma_filter_param {
2994 	int remote_thread_id;
2995 	u32 atype;
2996 };
2997 
2998 static bool udma_dma_filter_fn(struct dma_chan *chan, void *param)
2999 {
3000 	struct udma_chan_config *ucc;
3001 	struct psil_endpoint_config *ep_config;
3002 	struct udma_filter_param *filter_param;
3003 	struct udma_chan *uc;
3004 	struct udma_dev *ud;
3005 
3006 	if (chan->device->dev->driver != &udma_driver.driver)
3007 		return false;
3008 
3009 	uc = to_udma_chan(chan);
3010 	ucc = &uc->config;
3011 	ud = uc->ud;
3012 	filter_param = param;
3013 
3014 	if (filter_param->atype > 2) {
3015 		dev_err(ud->dev, "Invalid channel atype: %u\n",
3016 			filter_param->atype);
3017 		return false;
3018 	}
3019 
3020 	ucc->remote_thread_id = filter_param->remote_thread_id;
3021 	ucc->atype = filter_param->atype;
3022 
3023 	if (ucc->remote_thread_id & K3_PSIL_DST_THREAD_ID_OFFSET)
3024 		ucc->dir = DMA_MEM_TO_DEV;
3025 	else
3026 		ucc->dir = DMA_DEV_TO_MEM;
3027 
3028 	ep_config = psil_get_ep_config(ucc->remote_thread_id);
3029 	if (IS_ERR(ep_config)) {
3030 		dev_err(ud->dev, "No configuration for psi-l thread 0x%04x\n",
3031 			ucc->remote_thread_id);
3032 		ucc->dir = DMA_MEM_TO_MEM;
3033 		ucc->remote_thread_id = -1;
3034 		ucc->atype = 0;
3035 		return false;
3036 	}
3037 
3038 	ucc->pkt_mode = ep_config->pkt_mode;
3039 	ucc->channel_tpl = ep_config->channel_tpl;
3040 	ucc->notdpkt = ep_config->notdpkt;
3041 	ucc->ep_type = ep_config->ep_type;
3042 
3043 	if (ucc->ep_type != PSIL_EP_NATIVE) {
3044 		const struct udma_match_data *match_data = ud->match_data;
3045 
3046 		if (match_data->flags & UDMA_FLAG_PDMA_ACC32)
3047 			ucc->enable_acc32 = ep_config->pdma_acc32;
3048 		if (match_data->flags & UDMA_FLAG_PDMA_BURST)
3049 			ucc->enable_burst = ep_config->pdma_burst;
3050 	}
3051 
3052 	ucc->needs_epib = ep_config->needs_epib;
3053 	ucc->psd_size = ep_config->psd_size;
3054 	ucc->metadata_size =
3055 			(ucc->needs_epib ? CPPI5_INFO0_HDESC_EPIB_SIZE : 0) +
3056 			ucc->psd_size;
3057 
3058 	if (ucc->pkt_mode)
3059 		ucc->hdesc_size = ALIGN(sizeof(struct cppi5_host_desc_t) +
3060 				 ucc->metadata_size, ud->desc_align);
3061 
3062 	dev_dbg(ud->dev, "chan%d: Remote thread: 0x%04x (%s)\n", uc->id,
3063 		ucc->remote_thread_id, dmaengine_get_direction_text(ucc->dir));
3064 
3065 	return true;
3066 }
3067 
3068 static struct dma_chan *udma_of_xlate(struct of_phandle_args *dma_spec,
3069 				      struct of_dma *ofdma)
3070 {
3071 	struct udma_dev *ud = ofdma->of_dma_data;
3072 	dma_cap_mask_t mask = ud->ddev.cap_mask;
3073 	struct udma_filter_param filter_param;
3074 	struct dma_chan *chan;
3075 
3076 	if (dma_spec->args_count != 1 && dma_spec->args_count != 2)
3077 		return NULL;
3078 
3079 	filter_param.remote_thread_id = dma_spec->args[0];
3080 	if (dma_spec->args_count == 2)
3081 		filter_param.atype = dma_spec->args[1];
3082 	else
3083 		filter_param.atype = 0;
3084 
3085 	chan = __dma_request_channel(&mask, udma_dma_filter_fn, &filter_param,
3086 				     ofdma->of_node);
3087 	if (!chan) {
3088 		dev_err(ud->dev, "get channel fail in %s.\n", __func__);
3089 		return ERR_PTR(-EINVAL);
3090 	}
3091 
3092 	return chan;
3093 }
3094 
3095 static struct udma_match_data am654_main_data = {
3096 	.psil_base = 0x1000,
3097 	.enable_memcpy_support = true,
3098 	.statictr_z_mask = GENMASK(11, 0),
3099 };
3100 
3101 static struct udma_match_data am654_mcu_data = {
3102 	.psil_base = 0x6000,
3103 	.enable_memcpy_support = false,
3104 	.statictr_z_mask = GENMASK(11, 0),
3105 };
3106 
3107 static struct udma_match_data j721e_main_data = {
3108 	.psil_base = 0x1000,
3109 	.enable_memcpy_support = true,
3110 	.flags = UDMA_FLAG_PDMA_ACC32 | UDMA_FLAG_PDMA_BURST,
3111 	.statictr_z_mask = GENMASK(23, 0),
3112 };
3113 
3114 static struct udma_match_data j721e_mcu_data = {
3115 	.psil_base = 0x6000,
3116 	.enable_memcpy_support = false, /* MEM_TO_MEM is slow via MCU UDMA */
3117 	.flags = UDMA_FLAG_PDMA_ACC32 | UDMA_FLAG_PDMA_BURST,
3118 	.statictr_z_mask = GENMASK(23, 0),
3119 };
3120 
3121 static const struct of_device_id udma_of_match[] = {
3122 	{
3123 		.compatible = "ti,am654-navss-main-udmap",
3124 		.data = &am654_main_data,
3125 	},
3126 	{
3127 		.compatible = "ti,am654-navss-mcu-udmap",
3128 		.data = &am654_mcu_data,
3129 	}, {
3130 		.compatible = "ti,j721e-navss-main-udmap",
3131 		.data = &j721e_main_data,
3132 	}, {
3133 		.compatible = "ti,j721e-navss-mcu-udmap",
3134 		.data = &j721e_mcu_data,
3135 	},
3136 	{ /* Sentinel */ },
3137 };
3138 
3139 static struct udma_soc_data am654_soc_data = {
3140 	.rchan_oes_offset = 0x200,
3141 };
3142 
3143 static struct udma_soc_data j721e_soc_data = {
3144 	.rchan_oes_offset = 0x400,
3145 };
3146 
3147 static struct udma_soc_data j7200_soc_data = {
3148 	.rchan_oes_offset = 0x80,
3149 };
3150 
3151 static const struct soc_device_attribute k3_soc_devices[] = {
3152 	{ .family = "AM65X", .data = &am654_soc_data },
3153 	{ .family = "J721E", .data = &j721e_soc_data },
3154 	{ .family = "J7200", .data = &j7200_soc_data },
3155 	{ /* sentinel */ }
3156 };
3157 
3158 static int udma_get_mmrs(struct platform_device *pdev, struct udma_dev *ud)
3159 {
3160 	int i;
3161 
3162 	for (i = 0; i < MMR_LAST; i++) {
3163 		ud->mmrs[i] = devm_platform_ioremap_resource_byname(pdev, mmr_names[i]);
3164 		if (IS_ERR(ud->mmrs[i]))
3165 			return PTR_ERR(ud->mmrs[i]);
3166 	}
3167 
3168 	return 0;
3169 }
3170 
3171 static int udma_setup_resources(struct udma_dev *ud)
3172 {
3173 	struct device *dev = ud->dev;
3174 	int ch_count, ret, i, j;
3175 	u32 cap2, cap3;
3176 	struct ti_sci_resource_desc *rm_desc;
3177 	struct ti_sci_resource *rm_res, irq_res;
3178 	struct udma_tisci_rm *tisci_rm = &ud->tisci_rm;
3179 	static const char * const range_names[] = { "ti,sci-rm-range-tchan",
3180 						    "ti,sci-rm-range-rchan",
3181 						    "ti,sci-rm-range-rflow" };
3182 
3183 	cap2 = udma_read(ud->mmrs[MMR_GCFG], UDMA_CAP_REG(2));
3184 	cap3 = udma_read(ud->mmrs[MMR_GCFG], UDMA_CAP_REG(3));
3185 
3186 	ud->rflow_cnt = UDMA_CAP3_RFLOW_CNT(cap3);
3187 	ud->tchan_cnt = UDMA_CAP2_TCHAN_CNT(cap2);
3188 	ud->echan_cnt = UDMA_CAP2_ECHAN_CNT(cap2);
3189 	ud->rchan_cnt = UDMA_CAP2_RCHAN_CNT(cap2);
3190 	ch_count  = ud->tchan_cnt + ud->rchan_cnt;
3191 
3192 	/* Set up the throughput level start indexes */
3193 	if (of_device_is_compatible(dev->of_node,
3194 				    "ti,am654-navss-main-udmap")) {
3195 		ud->tpl_levels = 2;
3196 		ud->tpl_start_idx[0] = 8;
3197 	} else if (of_device_is_compatible(dev->of_node,
3198 					   "ti,am654-navss-mcu-udmap")) {
3199 		ud->tpl_levels = 2;
3200 		ud->tpl_start_idx[0] = 2;
3201 	} else if (UDMA_CAP3_UCHAN_CNT(cap3)) {
3202 		ud->tpl_levels = 3;
3203 		ud->tpl_start_idx[1] = UDMA_CAP3_UCHAN_CNT(cap3);
3204 		ud->tpl_start_idx[0] = ud->tpl_start_idx[1] +
3205 				       UDMA_CAP3_HCHAN_CNT(cap3);
3206 	} else if (UDMA_CAP3_HCHAN_CNT(cap3)) {
3207 		ud->tpl_levels = 2;
3208 		ud->tpl_start_idx[0] = UDMA_CAP3_HCHAN_CNT(cap3);
3209 	} else {
3210 		ud->tpl_levels = 1;
3211 	}
3212 
3213 	ud->tchan_map = devm_kmalloc_array(dev, BITS_TO_LONGS(ud->tchan_cnt),
3214 					   sizeof(unsigned long), GFP_KERNEL);
3215 	ud->tchans = devm_kcalloc(dev, ud->tchan_cnt, sizeof(*ud->tchans),
3216 				  GFP_KERNEL);
3217 	ud->rchan_map = devm_kmalloc_array(dev, BITS_TO_LONGS(ud->rchan_cnt),
3218 					   sizeof(unsigned long), GFP_KERNEL);
3219 	ud->rchans = devm_kcalloc(dev, ud->rchan_cnt, sizeof(*ud->rchans),
3220 				  GFP_KERNEL);
3221 	ud->rflow_gp_map = devm_kmalloc_array(dev, BITS_TO_LONGS(ud->rflow_cnt),
3222 					      sizeof(unsigned long),
3223 					      GFP_KERNEL);
3224 	ud->rflow_gp_map_allocated = devm_kcalloc(dev,
3225 						  BITS_TO_LONGS(ud->rflow_cnt),
3226 						  sizeof(unsigned long),
3227 						  GFP_KERNEL);
3228 	ud->rflow_in_use = devm_kcalloc(dev, BITS_TO_LONGS(ud->rflow_cnt),
3229 					sizeof(unsigned long),
3230 					GFP_KERNEL);
3231 	ud->rflows = devm_kcalloc(dev, ud->rflow_cnt, sizeof(*ud->rflows),
3232 				  GFP_KERNEL);
3233 
3234 	if (!ud->tchan_map || !ud->rchan_map || !ud->rflow_gp_map ||
3235 	    !ud->rflow_gp_map_allocated || !ud->tchans || !ud->rchans ||
3236 	    !ud->rflows || !ud->rflow_in_use)
3237 		return -ENOMEM;
3238 
3239 	/*
3240 	 * RX flows with the same Ids as RX channels are reserved to be used
3241 	 * as default flows if remote HW can't generate flow_ids. Those
3242 	 * RX flows can be requested only explicitly by id.
3243 	 */
3244 	bitmap_set(ud->rflow_gp_map_allocated, 0, ud->rchan_cnt);
3245 
3246 	/* by default no GP rflows are assigned to Linux */
3247 	bitmap_set(ud->rflow_gp_map, 0, ud->rflow_cnt);
3248 
3249 	/* Get resource ranges from tisci */
3250 	for (i = 0; i < RM_RANGE_LAST; i++)
3251 		tisci_rm->rm_ranges[i] =
3252 			devm_ti_sci_get_of_resource(tisci_rm->tisci, dev,
3253 						    tisci_rm->tisci_dev_id,
3254 						    (char *)range_names[i]);
3255 
3256 	/* tchan ranges */
3257 	rm_res = tisci_rm->rm_ranges[RM_RANGE_TCHAN];
3258 	if (IS_ERR(rm_res)) {
3259 		bitmap_zero(ud->tchan_map, ud->tchan_cnt);
3260 	} else {
3261 		bitmap_fill(ud->tchan_map, ud->tchan_cnt);
3262 		for (i = 0; i < rm_res->sets; i++) {
3263 			rm_desc = &rm_res->desc[i];
3264 			bitmap_clear(ud->tchan_map, rm_desc->start,
3265 				     rm_desc->num);
3266 			dev_dbg(dev, "ti-sci-res: tchan: %d:%d\n",
3267 				rm_desc->start, rm_desc->num);
3268 		}
3269 	}
3270 	irq_res.sets = rm_res->sets;
3271 
3272 	/* rchan and matching default flow ranges */
3273 	rm_res = tisci_rm->rm_ranges[RM_RANGE_RCHAN];
3274 	if (IS_ERR(rm_res)) {
3275 		bitmap_zero(ud->rchan_map, ud->rchan_cnt);
3276 	} else {
3277 		bitmap_fill(ud->rchan_map, ud->rchan_cnt);
3278 		for (i = 0; i < rm_res->sets; i++) {
3279 			rm_desc = &rm_res->desc[i];
3280 			bitmap_clear(ud->rchan_map, rm_desc->start,
3281 				     rm_desc->num);
3282 			dev_dbg(dev, "ti-sci-res: rchan: %d:%d\n",
3283 				rm_desc->start, rm_desc->num);
3284 		}
3285 	}
3286 
3287 	irq_res.sets += rm_res->sets;
3288 	irq_res.desc = kcalloc(irq_res.sets, sizeof(*irq_res.desc), GFP_KERNEL);
3289 	rm_res = tisci_rm->rm_ranges[RM_RANGE_TCHAN];
3290 	for (i = 0; i < rm_res->sets; i++) {
3291 		irq_res.desc[i].start = rm_res->desc[i].start;
3292 		irq_res.desc[i].num = rm_res->desc[i].num;
3293 	}
3294 	rm_res = tisci_rm->rm_ranges[RM_RANGE_RCHAN];
3295 	for (j = 0; j < rm_res->sets; j++, i++) {
3296 		irq_res.desc[i].start = rm_res->desc[j].start +
3297 					ud->soc_data->rchan_oes_offset;
3298 		irq_res.desc[i].num = rm_res->desc[j].num;
3299 	}
3300 	ret = ti_sci_inta_msi_domain_alloc_irqs(ud->dev, &irq_res);
3301 	kfree(irq_res.desc);
3302 	if (ret) {
3303 		dev_err(ud->dev, "Failed to allocate MSI interrupts\n");
3304 		return ret;
3305 	}
3306 
3307 	/* GP rflow ranges */
3308 	rm_res = tisci_rm->rm_ranges[RM_RANGE_RFLOW];
3309 	if (IS_ERR(rm_res)) {
3310 		/* all gp flows are assigned exclusively to Linux */
3311 		bitmap_clear(ud->rflow_gp_map, ud->rchan_cnt,
3312 			     ud->rflow_cnt - ud->rchan_cnt);
3313 	} else {
3314 		for (i = 0; i < rm_res->sets; i++) {
3315 			rm_desc = &rm_res->desc[i];
3316 			bitmap_clear(ud->rflow_gp_map, rm_desc->start,
3317 				     rm_desc->num);
3318 			dev_dbg(dev, "ti-sci-res: rflow: %d:%d\n",
3319 				rm_desc->start, rm_desc->num);
3320 		}
3321 	}
3322 
3323 	ch_count -= bitmap_weight(ud->tchan_map, ud->tchan_cnt);
3324 	ch_count -= bitmap_weight(ud->rchan_map, ud->rchan_cnt);
3325 	if (!ch_count)
3326 		return -ENODEV;
3327 
3328 	ud->channels = devm_kcalloc(dev, ch_count, sizeof(*ud->channels),
3329 				    GFP_KERNEL);
3330 	if (!ud->channels)
3331 		return -ENOMEM;
3332 
3333 	dev_info(dev, "Channels: %d (tchan: %u, rchan: %u, gp-rflow: %u)\n",
3334 		 ch_count,
3335 		 ud->tchan_cnt - bitmap_weight(ud->tchan_map, ud->tchan_cnt),
3336 		 ud->rchan_cnt - bitmap_weight(ud->rchan_map, ud->rchan_cnt),
3337 		 ud->rflow_cnt - bitmap_weight(ud->rflow_gp_map,
3338 					       ud->rflow_cnt));
3339 
3340 	return ch_count;
3341 }
3342 
3343 static int udma_setup_rx_flush(struct udma_dev *ud)
3344 {
3345 	struct udma_rx_flush *rx_flush = &ud->rx_flush;
3346 	struct cppi5_desc_hdr_t *tr_desc;
3347 	struct cppi5_tr_type1_t *tr_req;
3348 	struct cppi5_host_desc_t *desc;
3349 	struct device *dev = ud->dev;
3350 	struct udma_hwdesc *hwdesc;
3351 	size_t tr_size;
3352 
3353 	/* Allocate 1K buffer for discarded data on RX channel teardown */
3354 	rx_flush->buffer_size = SZ_1K;
3355 	rx_flush->buffer_vaddr = devm_kzalloc(dev, rx_flush->buffer_size,
3356 					      GFP_KERNEL);
3357 	if (!rx_flush->buffer_vaddr)
3358 		return -ENOMEM;
3359 
3360 	rx_flush->buffer_paddr = dma_map_single(dev, rx_flush->buffer_vaddr,
3361 						rx_flush->buffer_size,
3362 						DMA_TO_DEVICE);
3363 	if (dma_mapping_error(dev, rx_flush->buffer_paddr))
3364 		return -ENOMEM;
3365 
3366 	/* Set up descriptor to be used for TR mode */
3367 	hwdesc = &rx_flush->hwdescs[0];
3368 	tr_size = sizeof(struct cppi5_tr_type1_t);
3369 	hwdesc->cppi5_desc_size = cppi5_trdesc_calc_size(tr_size, 1);
3370 	hwdesc->cppi5_desc_size = ALIGN(hwdesc->cppi5_desc_size,
3371 					ud->desc_align);
3372 
3373 	hwdesc->cppi5_desc_vaddr = devm_kzalloc(dev, hwdesc->cppi5_desc_size,
3374 						GFP_KERNEL);
3375 	if (!hwdesc->cppi5_desc_vaddr)
3376 		return -ENOMEM;
3377 
3378 	hwdesc->cppi5_desc_paddr = dma_map_single(dev, hwdesc->cppi5_desc_vaddr,
3379 						  hwdesc->cppi5_desc_size,
3380 						  DMA_TO_DEVICE);
3381 	if (dma_mapping_error(dev, hwdesc->cppi5_desc_paddr))
3382 		return -ENOMEM;
3383 
3384 	/* Start of the TR req records */
3385 	hwdesc->tr_req_base = hwdesc->cppi5_desc_vaddr + tr_size;
3386 	/* Start address of the TR response array */
3387 	hwdesc->tr_resp_base = hwdesc->tr_req_base + tr_size;
3388 
3389 	tr_desc = hwdesc->cppi5_desc_vaddr;
3390 	cppi5_trdesc_init(tr_desc, 1, tr_size, 0, 0);
3391 	cppi5_desc_set_pktids(tr_desc, 0, CPPI5_INFO1_DESC_FLOWID_DEFAULT);
3392 	cppi5_desc_set_retpolicy(tr_desc, 0, 0);
3393 
3394 	tr_req = hwdesc->tr_req_base;
3395 	cppi5_tr_init(&tr_req->flags, CPPI5_TR_TYPE1, false, false,
3396 		      CPPI5_TR_EVENT_SIZE_COMPLETION, 0);
3397 	cppi5_tr_csf_set(&tr_req->flags, CPPI5_TR_CSF_SUPR_EVT);
3398 
3399 	tr_req->addr = rx_flush->buffer_paddr;
3400 	tr_req->icnt0 = rx_flush->buffer_size;
3401 	tr_req->icnt1 = 1;
3402 
3403 	dma_sync_single_for_device(dev, hwdesc->cppi5_desc_paddr,
3404 				   hwdesc->cppi5_desc_size, DMA_TO_DEVICE);
3405 
3406 	/* Set up descriptor to be used for packet mode */
3407 	hwdesc = &rx_flush->hwdescs[1];
3408 	hwdesc->cppi5_desc_size = ALIGN(sizeof(struct cppi5_host_desc_t) +
3409 					CPPI5_INFO0_HDESC_EPIB_SIZE +
3410 					CPPI5_INFO0_HDESC_PSDATA_MAX_SIZE,
3411 					ud->desc_align);
3412 
3413 	hwdesc->cppi5_desc_vaddr = devm_kzalloc(dev, hwdesc->cppi5_desc_size,
3414 						GFP_KERNEL);
3415 	if (!hwdesc->cppi5_desc_vaddr)
3416 		return -ENOMEM;
3417 
3418 	hwdesc->cppi5_desc_paddr = dma_map_single(dev, hwdesc->cppi5_desc_vaddr,
3419 						  hwdesc->cppi5_desc_size,
3420 						  DMA_TO_DEVICE);
3421 	if (dma_mapping_error(dev, hwdesc->cppi5_desc_paddr))
3422 		return -ENOMEM;
3423 
3424 	desc = hwdesc->cppi5_desc_vaddr;
3425 	cppi5_hdesc_init(desc, 0, 0);
3426 	cppi5_desc_set_pktids(&desc->hdr, 0, CPPI5_INFO1_DESC_FLOWID_DEFAULT);
3427 	cppi5_desc_set_retpolicy(&desc->hdr, 0, 0);
3428 
3429 	cppi5_hdesc_attach_buf(desc,
3430 			       rx_flush->buffer_paddr, rx_flush->buffer_size,
3431 			       rx_flush->buffer_paddr, rx_flush->buffer_size);
3432 
3433 	dma_sync_single_for_device(dev, hwdesc->cppi5_desc_paddr,
3434 				   hwdesc->cppi5_desc_size, DMA_TO_DEVICE);
3435 	return 0;
3436 }
3437 
3438 #ifdef CONFIG_DEBUG_FS
3439 static void udma_dbg_summary_show_chan(struct seq_file *s,
3440 				       struct dma_chan *chan)
3441 {
3442 	struct udma_chan *uc = to_udma_chan(chan);
3443 	struct udma_chan_config *ucc = &uc->config;
3444 
3445 	seq_printf(s, " %-13s| %s", dma_chan_name(chan),
3446 		   chan->dbg_client_name ?: "in-use");
3447 	seq_printf(s, " (%s, ", dmaengine_get_direction_text(uc->config.dir));
3448 
3449 	switch (uc->config.dir) {
3450 	case DMA_MEM_TO_MEM:
3451 		seq_printf(s, "chan%d pair [0x%04x -> 0x%04x], ", uc->tchan->id,
3452 			   ucc->src_thread, ucc->dst_thread);
3453 		break;
3454 	case DMA_DEV_TO_MEM:
3455 		seq_printf(s, "rchan%d [0x%04x -> 0x%04x], ", uc->rchan->id,
3456 			   ucc->src_thread, ucc->dst_thread);
3457 		break;
3458 	case DMA_MEM_TO_DEV:
3459 		seq_printf(s, "tchan%d [0x%04x -> 0x%04x], ", uc->tchan->id,
3460 			   ucc->src_thread, ucc->dst_thread);
3461 		break;
3462 	default:
3463 		seq_printf(s, ")\n");
3464 		return;
3465 	}
3466 
3467 	if (ucc->ep_type == PSIL_EP_NATIVE) {
3468 		seq_printf(s, "PSI-L Native");
3469 		if (ucc->metadata_size) {
3470 			seq_printf(s, "[%s", ucc->needs_epib ? " EPIB" : "");
3471 			if (ucc->psd_size)
3472 				seq_printf(s, " PSDsize:%u", ucc->psd_size);
3473 			seq_printf(s, " ]");
3474 		}
3475 	} else {
3476 		seq_printf(s, "PDMA");
3477 		if (ucc->enable_acc32 || ucc->enable_burst)
3478 			seq_printf(s, "[%s%s ]",
3479 				   ucc->enable_acc32 ? " ACC32" : "",
3480 				   ucc->enable_burst ? " BURST" : "");
3481 	}
3482 
3483 	seq_printf(s, ", %s)\n", ucc->pkt_mode ? "Packet mode" : "TR mode");
3484 }
3485 
3486 static void udma_dbg_summary_show(struct seq_file *s,
3487 				  struct dma_device *dma_dev)
3488 {
3489 	struct dma_chan *chan;
3490 
3491 	list_for_each_entry(chan, &dma_dev->channels, device_node) {
3492 		if (chan->client_count)
3493 			udma_dbg_summary_show_chan(s, chan);
3494 	}
3495 }
3496 #endif /* CONFIG_DEBUG_FS */
3497 
3498 #define TI_UDMAC_BUSWIDTHS	(BIT(DMA_SLAVE_BUSWIDTH_1_BYTE) | \
3499 				 BIT(DMA_SLAVE_BUSWIDTH_2_BYTES) | \
3500 				 BIT(DMA_SLAVE_BUSWIDTH_3_BYTES) | \
3501 				 BIT(DMA_SLAVE_BUSWIDTH_4_BYTES) | \
3502 				 BIT(DMA_SLAVE_BUSWIDTH_8_BYTES))
3503 
3504 static int udma_probe(struct platform_device *pdev)
3505 {
3506 	struct device_node *navss_node = pdev->dev.parent->of_node;
3507 	const struct soc_device_attribute *soc;
3508 	struct device *dev = &pdev->dev;
3509 	struct udma_dev *ud;
3510 	const struct of_device_id *match;
3511 	int i, ret;
3512 	int ch_count;
3513 
3514 	ret = dma_coerce_mask_and_coherent(dev, DMA_BIT_MASK(48));
3515 	if (ret)
3516 		dev_err(dev, "failed to set dma mask stuff\n");
3517 
3518 	ud = devm_kzalloc(dev, sizeof(*ud), GFP_KERNEL);
3519 	if (!ud)
3520 		return -ENOMEM;
3521 
3522 	ret = udma_get_mmrs(pdev, ud);
3523 	if (ret)
3524 		return ret;
3525 
3526 	ud->tisci_rm.tisci = ti_sci_get_by_phandle(dev->of_node, "ti,sci");
3527 	if (IS_ERR(ud->tisci_rm.tisci))
3528 		return PTR_ERR(ud->tisci_rm.tisci);
3529 
3530 	ret = of_property_read_u32(dev->of_node, "ti,sci-dev-id",
3531 				   &ud->tisci_rm.tisci_dev_id);
3532 	if (ret) {
3533 		dev_err(dev, "ti,sci-dev-id read failure %d\n", ret);
3534 		return ret;
3535 	}
3536 	pdev->id = ud->tisci_rm.tisci_dev_id;
3537 
3538 	ret = of_property_read_u32(navss_node, "ti,sci-dev-id",
3539 				   &ud->tisci_rm.tisci_navss_dev_id);
3540 	if (ret) {
3541 		dev_err(dev, "NAVSS ti,sci-dev-id read failure %d\n", ret);
3542 		return ret;
3543 	}
3544 
3545 	ret = of_property_read_u32(dev->of_node, "ti,udma-atype", &ud->atype);
3546 	if (!ret && ud->atype > 2) {
3547 		dev_err(dev, "Invalid atype: %u\n", ud->atype);
3548 		return -EINVAL;
3549 	}
3550 
3551 	ud->tisci_rm.tisci_udmap_ops = &ud->tisci_rm.tisci->ops.rm_udmap_ops;
3552 	ud->tisci_rm.tisci_psil_ops = &ud->tisci_rm.tisci->ops.rm_psil_ops;
3553 
3554 	ud->ringacc = of_k3_ringacc_get_by_phandle(dev->of_node, "ti,ringacc");
3555 	if (IS_ERR(ud->ringacc))
3556 		return PTR_ERR(ud->ringacc);
3557 
3558 	dev->msi_domain = of_msi_get_domain(dev, dev->of_node,
3559 					    DOMAIN_BUS_TI_SCI_INTA_MSI);
3560 	if (!dev->msi_domain) {
3561 		dev_err(dev, "Failed to get MSI domain\n");
3562 		return -EPROBE_DEFER;
3563 	}
3564 
3565 	match = of_match_node(udma_of_match, dev->of_node);
3566 	if (!match) {
3567 		dev_err(dev, "No compatible match found\n");
3568 		return -ENODEV;
3569 	}
3570 	ud->match_data = match->data;
3571 
3572 	soc = soc_device_match(k3_soc_devices);
3573 	if (!soc) {
3574 		dev_err(dev, "No compatible SoC found\n");
3575 		return -ENODEV;
3576 	}
3577 	ud->soc_data = soc->data;
3578 
3579 	dma_cap_set(DMA_SLAVE, ud->ddev.cap_mask);
3580 	dma_cap_set(DMA_CYCLIC, ud->ddev.cap_mask);
3581 
3582 	ud->ddev.device_alloc_chan_resources = udma_alloc_chan_resources;
3583 	ud->ddev.device_config = udma_slave_config;
3584 	ud->ddev.device_prep_slave_sg = udma_prep_slave_sg;
3585 	ud->ddev.device_prep_dma_cyclic = udma_prep_dma_cyclic;
3586 	ud->ddev.device_issue_pending = udma_issue_pending;
3587 	ud->ddev.device_tx_status = udma_tx_status;
3588 	ud->ddev.device_pause = udma_pause;
3589 	ud->ddev.device_resume = udma_resume;
3590 	ud->ddev.device_terminate_all = udma_terminate_all;
3591 	ud->ddev.device_synchronize = udma_synchronize;
3592 #ifdef CONFIG_DEBUG_FS
3593 	ud->ddev.dbg_summary_show = udma_dbg_summary_show;
3594 #endif
3595 
3596 	ud->ddev.device_free_chan_resources = udma_free_chan_resources;
3597 	ud->ddev.src_addr_widths = TI_UDMAC_BUSWIDTHS;
3598 	ud->ddev.dst_addr_widths = TI_UDMAC_BUSWIDTHS;
3599 	ud->ddev.directions = BIT(DMA_DEV_TO_MEM) | BIT(DMA_MEM_TO_DEV);
3600 	ud->ddev.residue_granularity = DMA_RESIDUE_GRANULARITY_BURST;
3601 	ud->ddev.copy_align = DMAENGINE_ALIGN_8_BYTES;
3602 	ud->ddev.desc_metadata_modes = DESC_METADATA_CLIENT |
3603 				       DESC_METADATA_ENGINE;
3604 	if (ud->match_data->enable_memcpy_support) {
3605 		dma_cap_set(DMA_MEMCPY, ud->ddev.cap_mask);
3606 		ud->ddev.device_prep_dma_memcpy = udma_prep_dma_memcpy;
3607 		ud->ddev.directions |= BIT(DMA_MEM_TO_MEM);
3608 	}
3609 
3610 	ud->ddev.dev = dev;
3611 	ud->dev = dev;
3612 	ud->psil_base = ud->match_data->psil_base;
3613 
3614 	INIT_LIST_HEAD(&ud->ddev.channels);
3615 	INIT_LIST_HEAD(&ud->desc_to_purge);
3616 
3617 	ch_count = udma_setup_resources(ud);
3618 	if (ch_count <= 0)
3619 		return ch_count;
3620 
3621 	spin_lock_init(&ud->lock);
3622 	INIT_WORK(&ud->purge_work, udma_purge_desc_work);
3623 
3624 	ud->desc_align = 64;
3625 	if (ud->desc_align < dma_get_cache_alignment())
3626 		ud->desc_align = dma_get_cache_alignment();
3627 
3628 	ret = udma_setup_rx_flush(ud);
3629 	if (ret)
3630 		return ret;
3631 
3632 	for (i = 0; i < ud->tchan_cnt; i++) {
3633 		struct udma_tchan *tchan = &ud->tchans[i];
3634 
3635 		tchan->id = i;
3636 		tchan->reg_rt = ud->mmrs[MMR_TCHANRT] + i * 0x1000;
3637 	}
3638 
3639 	for (i = 0; i < ud->rchan_cnt; i++) {
3640 		struct udma_rchan *rchan = &ud->rchans[i];
3641 
3642 		rchan->id = i;
3643 		rchan->reg_rt = ud->mmrs[MMR_RCHANRT] + i * 0x1000;
3644 	}
3645 
3646 	for (i = 0; i < ud->rflow_cnt; i++) {
3647 		struct udma_rflow *rflow = &ud->rflows[i];
3648 
3649 		rflow->id = i;
3650 	}
3651 
3652 	for (i = 0; i < ch_count; i++) {
3653 		struct udma_chan *uc = &ud->channels[i];
3654 
3655 		uc->ud = ud;
3656 		uc->vc.desc_free = udma_desc_free;
3657 		uc->id = i;
3658 		uc->tchan = NULL;
3659 		uc->rchan = NULL;
3660 		uc->config.remote_thread_id = -1;
3661 		uc->config.dir = DMA_MEM_TO_MEM;
3662 		uc->name = devm_kasprintf(dev, GFP_KERNEL, "%s chan%d",
3663 					  dev_name(dev), i);
3664 
3665 		vchan_init(&uc->vc, &ud->ddev);
3666 		/* Use custom vchan completion handling */
3667 		tasklet_setup(&uc->vc.task, udma_vchan_complete);
3668 		init_completion(&uc->teardown_completed);
3669 		INIT_DELAYED_WORK(&uc->tx_drain.work, udma_check_tx_completion);
3670 	}
3671 
3672 	ret = dma_async_device_register(&ud->ddev);
3673 	if (ret) {
3674 		dev_err(dev, "failed to register slave DMA engine: %d\n", ret);
3675 		return ret;
3676 	}
3677 
3678 	platform_set_drvdata(pdev, ud);
3679 
3680 	ret = of_dma_controller_register(dev->of_node, udma_of_xlate, ud);
3681 	if (ret) {
3682 		dev_err(dev, "failed to register of_dma controller\n");
3683 		dma_async_device_unregister(&ud->ddev);
3684 	}
3685 
3686 	return ret;
3687 }
3688 
3689 static struct platform_driver udma_driver = {
3690 	.driver = {
3691 		.name	= "ti-udma",
3692 		.of_match_table = udma_of_match,
3693 		.suppress_bind_attrs = true,
3694 	},
3695 	.probe		= udma_probe,
3696 };
3697 builtin_platform_driver(udma_driver);
3698 
3699 /* Private interfaces to UDMA */
3700 #include "k3-udma-private.c"
3701