xref: /linux/drivers/dma/ti/k3-udma.c (revision 7a309195d11cde854eb75559fbd6b48f9e518f25)
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/of.h>
20 #include <linux/of_dma.h>
21 #include <linux/of_device.h>
22 #include <linux/of_irq.h>
23 #include <linux/workqueue.h>
24 #include <linux/completion.h>
25 #include <linux/soc/ti/k3-ringacc.h>
26 #include <linux/soc/ti/ti_sci_protocol.h>
27 #include <linux/soc/ti/ti_sci_inta_msi.h>
28 #include <linux/dma/ti-cppi5.h>
29 
30 #include "../virt-dma.h"
31 #include "k3-udma.h"
32 #include "k3-psil-priv.h"
33 
34 struct udma_static_tr {
35 	u8 elsize; /* RPSTR0 */
36 	u16 elcnt; /* RPSTR0 */
37 	u16 bstcnt; /* RPSTR1 */
38 };
39 
40 #define K3_UDMA_MAX_RFLOWS		1024
41 #define K3_UDMA_DEFAULT_RING_SIZE	16
42 
43 /* How SRC/DST tag should be updated by UDMA in the descriptor's Word 3 */
44 #define UDMA_RFLOW_SRCTAG_NONE		0
45 #define UDMA_RFLOW_SRCTAG_CFG_TAG	1
46 #define UDMA_RFLOW_SRCTAG_FLOW_ID	2
47 #define UDMA_RFLOW_SRCTAG_SRC_TAG	4
48 
49 #define UDMA_RFLOW_DSTTAG_NONE		0
50 #define UDMA_RFLOW_DSTTAG_CFG_TAG	1
51 #define UDMA_RFLOW_DSTTAG_FLOW_ID	2
52 #define UDMA_RFLOW_DSTTAG_DST_TAG_LO	4
53 #define UDMA_RFLOW_DSTTAG_DST_TAG_HI	5
54 
55 struct udma_chan;
56 
57 enum udma_mmr {
58 	MMR_GCFG = 0,
59 	MMR_RCHANRT,
60 	MMR_TCHANRT,
61 	MMR_LAST,
62 };
63 
64 static const char * const mmr_names[] = { "gcfg", "rchanrt", "tchanrt" };
65 
66 struct udma_tchan {
67 	void __iomem *reg_rt;
68 
69 	int id;
70 	struct k3_ring *t_ring; /* Transmit ring */
71 	struct k3_ring *tc_ring; /* Transmit Completion ring */
72 };
73 
74 struct udma_rflow {
75 	int id;
76 	struct k3_ring *fd_ring; /* Free Descriptor ring */
77 	struct k3_ring *r_ring; /* Receive ring */
78 };
79 
80 struct udma_rchan {
81 	void __iomem *reg_rt;
82 
83 	int id;
84 };
85 
86 #define UDMA_FLAG_PDMA_ACC32		BIT(0)
87 #define UDMA_FLAG_PDMA_BURST		BIT(1)
88 
89 struct udma_match_data {
90 	u32 psil_base;
91 	bool enable_memcpy_support;
92 	u32 flags;
93 	u32 statictr_z_mask;
94 	u32 rchan_oes_offset;
95 
96 	u8 tpl_levels;
97 	u32 level_start_idx[];
98 };
99 
100 struct udma_hwdesc {
101 	size_t cppi5_desc_size;
102 	void *cppi5_desc_vaddr;
103 	dma_addr_t cppi5_desc_paddr;
104 
105 	/* TR descriptor internal pointers */
106 	void *tr_req_base;
107 	struct cppi5_tr_resp_t *tr_resp_base;
108 };
109 
110 struct udma_rx_flush {
111 	struct udma_hwdesc hwdescs[2];
112 
113 	size_t buffer_size;
114 	void *buffer_vaddr;
115 	dma_addr_t buffer_paddr;
116 };
117 
118 struct udma_dev {
119 	struct dma_device ddev;
120 	struct device *dev;
121 	void __iomem *mmrs[MMR_LAST];
122 	const struct udma_match_data *match_data;
123 
124 	size_t desc_align; /* alignment to use for descriptors */
125 
126 	struct udma_tisci_rm tisci_rm;
127 
128 	struct k3_ringacc *ringacc;
129 
130 	struct work_struct purge_work;
131 	struct list_head desc_to_purge;
132 	spinlock_t lock;
133 
134 	struct udma_rx_flush rx_flush;
135 
136 	int tchan_cnt;
137 	int echan_cnt;
138 	int rchan_cnt;
139 	int rflow_cnt;
140 	unsigned long *tchan_map;
141 	unsigned long *rchan_map;
142 	unsigned long *rflow_gp_map;
143 	unsigned long *rflow_gp_map_allocated;
144 	unsigned long *rflow_in_use;
145 
146 	struct udma_tchan *tchans;
147 	struct udma_rchan *rchans;
148 	struct udma_rflow *rflows;
149 
150 	struct udma_chan *channels;
151 	u32 psil_base;
152 	u32 atype;
153 };
154 
155 struct udma_desc {
156 	struct virt_dma_desc vd;
157 
158 	bool terminated;
159 
160 	enum dma_transfer_direction dir;
161 
162 	struct udma_static_tr static_tr;
163 	u32 residue;
164 
165 	unsigned int sglen;
166 	unsigned int desc_idx; /* Only used for cyclic in packet mode */
167 	unsigned int tr_idx;
168 
169 	u32 metadata_size;
170 	void *metadata; /* pointer to provided metadata buffer (EPIP, PSdata) */
171 
172 	unsigned int hwdesc_count;
173 	struct udma_hwdesc hwdesc[];
174 };
175 
176 enum udma_chan_state {
177 	UDMA_CHAN_IS_IDLE = 0, /* not active, no teardown is in progress */
178 	UDMA_CHAN_IS_ACTIVE, /* Normal operation */
179 	UDMA_CHAN_IS_TERMINATING, /* channel is being terminated */
180 };
181 
182 struct udma_tx_drain {
183 	struct delayed_work work;
184 	ktime_t tstamp;
185 	u32 residue;
186 };
187 
188 struct udma_chan_config {
189 	bool pkt_mode; /* TR or packet */
190 	bool needs_epib; /* EPIB is needed for the communication or not */
191 	u32 psd_size; /* size of Protocol Specific Data */
192 	u32 metadata_size; /* (needs_epib ? 16:0) + psd_size */
193 	u32 hdesc_size; /* Size of a packet descriptor in packet mode */
194 	bool notdpkt; /* Suppress sending TDC packet */
195 	int remote_thread_id;
196 	u32 atype;
197 	u32 src_thread;
198 	u32 dst_thread;
199 	enum psil_endpoint_type ep_type;
200 	bool enable_acc32;
201 	bool enable_burst;
202 	enum udma_tp_level channel_tpl; /* Channel Throughput Level */
203 
204 	enum dma_transfer_direction dir;
205 };
206 
207 struct udma_chan {
208 	struct virt_dma_chan vc;
209 	struct dma_slave_config	cfg;
210 	struct udma_dev *ud;
211 	struct udma_desc *desc;
212 	struct udma_desc *terminated_desc;
213 	struct udma_static_tr static_tr;
214 	char *name;
215 
216 	struct udma_tchan *tchan;
217 	struct udma_rchan *rchan;
218 	struct udma_rflow *rflow;
219 
220 	bool psil_paired;
221 
222 	int irq_num_ring;
223 	int irq_num_udma;
224 
225 	bool cyclic;
226 	bool paused;
227 
228 	enum udma_chan_state state;
229 	struct completion teardown_completed;
230 
231 	struct udma_tx_drain tx_drain;
232 
233 	u32 bcnt; /* number of bytes completed since the start of the channel */
234 
235 	/* Channel configuration parameters */
236 	struct udma_chan_config config;
237 
238 	/* dmapool for packet mode descriptors */
239 	bool use_dma_pool;
240 	struct dma_pool *hdesc_pool;
241 
242 	u32 id;
243 };
244 
245 static inline struct udma_dev *to_udma_dev(struct dma_device *d)
246 {
247 	return container_of(d, struct udma_dev, ddev);
248 }
249 
250 static inline struct udma_chan *to_udma_chan(struct dma_chan *c)
251 {
252 	return container_of(c, struct udma_chan, vc.chan);
253 }
254 
255 static inline struct udma_desc *to_udma_desc(struct dma_async_tx_descriptor *t)
256 {
257 	return container_of(t, struct udma_desc, vd.tx);
258 }
259 
260 /* Generic register access functions */
261 static inline u32 udma_read(void __iomem *base, int reg)
262 {
263 	return readl(base + reg);
264 }
265 
266 static inline void udma_write(void __iomem *base, int reg, u32 val)
267 {
268 	writel(val, base + reg);
269 }
270 
271 static inline void udma_update_bits(void __iomem *base, int reg,
272 				    u32 mask, u32 val)
273 {
274 	u32 tmp, orig;
275 
276 	orig = readl(base + reg);
277 	tmp = orig & ~mask;
278 	tmp |= (val & mask);
279 
280 	if (tmp != orig)
281 		writel(tmp, base + reg);
282 }
283 
284 /* TCHANRT */
285 static inline u32 udma_tchanrt_read(struct udma_tchan *tchan, int reg)
286 {
287 	if (!tchan)
288 		return 0;
289 	return udma_read(tchan->reg_rt, reg);
290 }
291 
292 static inline void udma_tchanrt_write(struct udma_tchan *tchan, int reg,
293 				      u32 val)
294 {
295 	if (!tchan)
296 		return;
297 	udma_write(tchan->reg_rt, reg, val);
298 }
299 
300 static inline void udma_tchanrt_update_bits(struct udma_tchan *tchan, int reg,
301 					    u32 mask, u32 val)
302 {
303 	if (!tchan)
304 		return;
305 	udma_update_bits(tchan->reg_rt, reg, mask, val);
306 }
307 
308 /* RCHANRT */
309 static inline u32 udma_rchanrt_read(struct udma_rchan *rchan, int reg)
310 {
311 	if (!rchan)
312 		return 0;
313 	return udma_read(rchan->reg_rt, reg);
314 }
315 
316 static inline void udma_rchanrt_write(struct udma_rchan *rchan, int reg,
317 				      u32 val)
318 {
319 	if (!rchan)
320 		return;
321 	udma_write(rchan->reg_rt, reg, val);
322 }
323 
324 static inline void udma_rchanrt_update_bits(struct udma_rchan *rchan, int reg,
325 					    u32 mask, u32 val)
326 {
327 	if (!rchan)
328 		return;
329 	udma_update_bits(rchan->reg_rt, reg, mask, val);
330 }
331 
332 static int navss_psil_pair(struct udma_dev *ud, u32 src_thread, u32 dst_thread)
333 {
334 	struct udma_tisci_rm *tisci_rm = &ud->tisci_rm;
335 
336 	dst_thread |= K3_PSIL_DST_THREAD_ID_OFFSET;
337 	return tisci_rm->tisci_psil_ops->pair(tisci_rm->tisci,
338 					      tisci_rm->tisci_navss_dev_id,
339 					      src_thread, dst_thread);
340 }
341 
342 static int navss_psil_unpair(struct udma_dev *ud, u32 src_thread,
343 			     u32 dst_thread)
344 {
345 	struct udma_tisci_rm *tisci_rm = &ud->tisci_rm;
346 
347 	dst_thread |= K3_PSIL_DST_THREAD_ID_OFFSET;
348 	return tisci_rm->tisci_psil_ops->unpair(tisci_rm->tisci,
349 						tisci_rm->tisci_navss_dev_id,
350 						src_thread, dst_thread);
351 }
352 
353 static void udma_reset_uchan(struct udma_chan *uc)
354 {
355 	memset(&uc->config, 0, sizeof(uc->config));
356 	uc->config.remote_thread_id = -1;
357 	uc->state = UDMA_CHAN_IS_IDLE;
358 }
359 
360 static void udma_dump_chan_stdata(struct udma_chan *uc)
361 {
362 	struct device *dev = uc->ud->dev;
363 	u32 offset;
364 	int i;
365 
366 	if (uc->config.dir == DMA_MEM_TO_DEV || uc->config.dir == DMA_MEM_TO_MEM) {
367 		dev_dbg(dev, "TCHAN State data:\n");
368 		for (i = 0; i < 32; i++) {
369 			offset = UDMA_TCHAN_RT_STDATA_REG + i * 4;
370 			dev_dbg(dev, "TRT_STDATA[%02d]: 0x%08x\n", i,
371 				udma_tchanrt_read(uc->tchan, offset));
372 		}
373 	}
374 
375 	if (uc->config.dir == DMA_DEV_TO_MEM || uc->config.dir == DMA_MEM_TO_MEM) {
376 		dev_dbg(dev, "RCHAN State data:\n");
377 		for (i = 0; i < 32; i++) {
378 			offset = UDMA_RCHAN_RT_STDATA_REG + i * 4;
379 			dev_dbg(dev, "RRT_STDATA[%02d]: 0x%08x\n", i,
380 				udma_rchanrt_read(uc->rchan, offset));
381 		}
382 	}
383 }
384 
385 static inline dma_addr_t udma_curr_cppi5_desc_paddr(struct udma_desc *d,
386 						    int idx)
387 {
388 	return d->hwdesc[idx].cppi5_desc_paddr;
389 }
390 
391 static inline void *udma_curr_cppi5_desc_vaddr(struct udma_desc *d, int idx)
392 {
393 	return d->hwdesc[idx].cppi5_desc_vaddr;
394 }
395 
396 static struct udma_desc *udma_udma_desc_from_paddr(struct udma_chan *uc,
397 						   dma_addr_t paddr)
398 {
399 	struct udma_desc *d = uc->terminated_desc;
400 
401 	if (d) {
402 		dma_addr_t desc_paddr = udma_curr_cppi5_desc_paddr(d,
403 								   d->desc_idx);
404 
405 		if (desc_paddr != paddr)
406 			d = NULL;
407 	}
408 
409 	if (!d) {
410 		d = uc->desc;
411 		if (d) {
412 			dma_addr_t desc_paddr = udma_curr_cppi5_desc_paddr(d,
413 								d->desc_idx);
414 
415 			if (desc_paddr != paddr)
416 				d = NULL;
417 		}
418 	}
419 
420 	return d;
421 }
422 
423 static void udma_free_hwdesc(struct udma_chan *uc, struct udma_desc *d)
424 {
425 	if (uc->use_dma_pool) {
426 		int i;
427 
428 		for (i = 0; i < d->hwdesc_count; i++) {
429 			if (!d->hwdesc[i].cppi5_desc_vaddr)
430 				continue;
431 
432 			dma_pool_free(uc->hdesc_pool,
433 				      d->hwdesc[i].cppi5_desc_vaddr,
434 				      d->hwdesc[i].cppi5_desc_paddr);
435 
436 			d->hwdesc[i].cppi5_desc_vaddr = NULL;
437 		}
438 	} else if (d->hwdesc[0].cppi5_desc_vaddr) {
439 		struct udma_dev *ud = uc->ud;
440 
441 		dma_free_coherent(ud->dev, d->hwdesc[0].cppi5_desc_size,
442 				  d->hwdesc[0].cppi5_desc_vaddr,
443 				  d->hwdesc[0].cppi5_desc_paddr);
444 
445 		d->hwdesc[0].cppi5_desc_vaddr = NULL;
446 	}
447 }
448 
449 static void udma_purge_desc_work(struct work_struct *work)
450 {
451 	struct udma_dev *ud = container_of(work, typeof(*ud), purge_work);
452 	struct virt_dma_desc *vd, *_vd;
453 	unsigned long flags;
454 	LIST_HEAD(head);
455 
456 	spin_lock_irqsave(&ud->lock, flags);
457 	list_splice_tail_init(&ud->desc_to_purge, &head);
458 	spin_unlock_irqrestore(&ud->lock, flags);
459 
460 	list_for_each_entry_safe(vd, _vd, &head, node) {
461 		struct udma_chan *uc = to_udma_chan(vd->tx.chan);
462 		struct udma_desc *d = to_udma_desc(&vd->tx);
463 
464 		udma_free_hwdesc(uc, d);
465 		list_del(&vd->node);
466 		kfree(d);
467 	}
468 
469 	/* If more to purge, schedule the work again */
470 	if (!list_empty(&ud->desc_to_purge))
471 		schedule_work(&ud->purge_work);
472 }
473 
474 static void udma_desc_free(struct virt_dma_desc *vd)
475 {
476 	struct udma_dev *ud = to_udma_dev(vd->tx.chan->device);
477 	struct udma_chan *uc = to_udma_chan(vd->tx.chan);
478 	struct udma_desc *d = to_udma_desc(&vd->tx);
479 	unsigned long flags;
480 
481 	if (uc->terminated_desc == d)
482 		uc->terminated_desc = NULL;
483 
484 	if (uc->use_dma_pool) {
485 		udma_free_hwdesc(uc, d);
486 		kfree(d);
487 		return;
488 	}
489 
490 	spin_lock_irqsave(&ud->lock, flags);
491 	list_add_tail(&vd->node, &ud->desc_to_purge);
492 	spin_unlock_irqrestore(&ud->lock, flags);
493 
494 	schedule_work(&ud->purge_work);
495 }
496 
497 static bool udma_is_chan_running(struct udma_chan *uc)
498 {
499 	u32 trt_ctl = 0;
500 	u32 rrt_ctl = 0;
501 
502 	if (uc->tchan)
503 		trt_ctl = udma_tchanrt_read(uc->tchan, UDMA_TCHAN_RT_CTL_REG);
504 	if (uc->rchan)
505 		rrt_ctl = udma_rchanrt_read(uc->rchan, UDMA_RCHAN_RT_CTL_REG);
506 
507 	if (trt_ctl & UDMA_CHAN_RT_CTL_EN || rrt_ctl & UDMA_CHAN_RT_CTL_EN)
508 		return true;
509 
510 	return false;
511 }
512 
513 static bool udma_is_chan_paused(struct udma_chan *uc)
514 {
515 	u32 val, pause_mask;
516 
517 	switch (uc->config.dir) {
518 	case DMA_DEV_TO_MEM:
519 		val = udma_rchanrt_read(uc->rchan,
520 					UDMA_RCHAN_RT_PEER_RT_EN_REG);
521 		pause_mask = UDMA_PEER_RT_EN_PAUSE;
522 		break;
523 	case DMA_MEM_TO_DEV:
524 		val = udma_tchanrt_read(uc->tchan,
525 					UDMA_TCHAN_RT_PEER_RT_EN_REG);
526 		pause_mask = UDMA_PEER_RT_EN_PAUSE;
527 		break;
528 	case DMA_MEM_TO_MEM:
529 		val = udma_tchanrt_read(uc->tchan, UDMA_TCHAN_RT_CTL_REG);
530 		pause_mask = UDMA_CHAN_RT_CTL_PAUSE;
531 		break;
532 	default:
533 		return false;
534 	}
535 
536 	if (val & pause_mask)
537 		return true;
538 
539 	return false;
540 }
541 
542 static void udma_sync_for_device(struct udma_chan *uc, int idx)
543 {
544 	struct udma_desc *d = uc->desc;
545 
546 	if (uc->cyclic && uc->config.pkt_mode) {
547 		dma_sync_single_for_device(uc->ud->dev,
548 					   d->hwdesc[idx].cppi5_desc_paddr,
549 					   d->hwdesc[idx].cppi5_desc_size,
550 					   DMA_TO_DEVICE);
551 	} else {
552 		int i;
553 
554 		for (i = 0; i < d->hwdesc_count; i++) {
555 			if (!d->hwdesc[i].cppi5_desc_vaddr)
556 				continue;
557 
558 			dma_sync_single_for_device(uc->ud->dev,
559 						d->hwdesc[i].cppi5_desc_paddr,
560 						d->hwdesc[i].cppi5_desc_size,
561 						DMA_TO_DEVICE);
562 		}
563 	}
564 }
565 
566 static inline dma_addr_t udma_get_rx_flush_hwdesc_paddr(struct udma_chan *uc)
567 {
568 	return uc->ud->rx_flush.hwdescs[uc->config.pkt_mode].cppi5_desc_paddr;
569 }
570 
571 static int udma_push_to_ring(struct udma_chan *uc, int idx)
572 {
573 	struct udma_desc *d = uc->desc;
574 	struct k3_ring *ring = NULL;
575 	dma_addr_t paddr;
576 
577 	switch (uc->config.dir) {
578 	case DMA_DEV_TO_MEM:
579 		ring = uc->rflow->fd_ring;
580 		break;
581 	case DMA_MEM_TO_DEV:
582 	case DMA_MEM_TO_MEM:
583 		ring = uc->tchan->t_ring;
584 		break;
585 	default:
586 		return -EINVAL;
587 	}
588 
589 	/* RX flush packet: idx == -1 is only passed in case of DEV_TO_MEM */
590 	if (idx == -1) {
591 		paddr = udma_get_rx_flush_hwdesc_paddr(uc);
592 	} else {
593 		paddr = udma_curr_cppi5_desc_paddr(d, idx);
594 
595 		wmb(); /* Ensure that writes are not moved over this point */
596 		udma_sync_for_device(uc, idx);
597 	}
598 
599 	return k3_ringacc_ring_push(ring, &paddr);
600 }
601 
602 static bool udma_desc_is_rx_flush(struct udma_chan *uc, dma_addr_t addr)
603 {
604 	if (uc->config.dir != DMA_DEV_TO_MEM)
605 		return false;
606 
607 	if (addr == udma_get_rx_flush_hwdesc_paddr(uc))
608 		return true;
609 
610 	return false;
611 }
612 
613 static int udma_pop_from_ring(struct udma_chan *uc, dma_addr_t *addr)
614 {
615 	struct k3_ring *ring = NULL;
616 	int ret = -ENOENT;
617 
618 	switch (uc->config.dir) {
619 	case DMA_DEV_TO_MEM:
620 		ring = uc->rflow->r_ring;
621 		break;
622 	case DMA_MEM_TO_DEV:
623 	case DMA_MEM_TO_MEM:
624 		ring = uc->tchan->tc_ring;
625 		break;
626 	default:
627 		break;
628 	}
629 
630 	if (ring && k3_ringacc_ring_get_occ(ring)) {
631 		struct udma_desc *d = NULL;
632 
633 		ret = k3_ringacc_ring_pop(ring, addr);
634 		if (ret)
635 			return ret;
636 
637 		/* Teardown completion */
638 		if (cppi5_desc_is_tdcm(*addr))
639 			return ret;
640 
641 		/* Check for flush descriptor */
642 		if (udma_desc_is_rx_flush(uc, *addr))
643 			return -ENOENT;
644 
645 		d = udma_udma_desc_from_paddr(uc, *addr);
646 
647 		if (d)
648 			dma_sync_single_for_cpu(uc->ud->dev, *addr,
649 						d->hwdesc[0].cppi5_desc_size,
650 						DMA_FROM_DEVICE);
651 		rmb(); /* Ensure that reads are not moved before this point */
652 	}
653 
654 	return ret;
655 }
656 
657 static void udma_reset_rings(struct udma_chan *uc)
658 {
659 	struct k3_ring *ring1 = NULL;
660 	struct k3_ring *ring2 = NULL;
661 
662 	switch (uc->config.dir) {
663 	case DMA_DEV_TO_MEM:
664 		if (uc->rchan) {
665 			ring1 = uc->rflow->fd_ring;
666 			ring2 = uc->rflow->r_ring;
667 		}
668 		break;
669 	case DMA_MEM_TO_DEV:
670 	case DMA_MEM_TO_MEM:
671 		if (uc->tchan) {
672 			ring1 = uc->tchan->t_ring;
673 			ring2 = uc->tchan->tc_ring;
674 		}
675 		break;
676 	default:
677 		break;
678 	}
679 
680 	if (ring1)
681 		k3_ringacc_ring_reset_dma(ring1,
682 					  k3_ringacc_ring_get_occ(ring1));
683 	if (ring2)
684 		k3_ringacc_ring_reset(ring2);
685 
686 	/* make sure we are not leaking memory by stalled descriptor */
687 	if (uc->terminated_desc) {
688 		udma_desc_free(&uc->terminated_desc->vd);
689 		uc->terminated_desc = NULL;
690 	}
691 }
692 
693 static void udma_reset_counters(struct udma_chan *uc)
694 {
695 	u32 val;
696 
697 	if (uc->tchan) {
698 		val = udma_tchanrt_read(uc->tchan, UDMA_TCHAN_RT_BCNT_REG);
699 		udma_tchanrt_write(uc->tchan, UDMA_TCHAN_RT_BCNT_REG, val);
700 
701 		val = udma_tchanrt_read(uc->tchan, UDMA_TCHAN_RT_SBCNT_REG);
702 		udma_tchanrt_write(uc->tchan, UDMA_TCHAN_RT_SBCNT_REG, val);
703 
704 		val = udma_tchanrt_read(uc->tchan, UDMA_TCHAN_RT_PCNT_REG);
705 		udma_tchanrt_write(uc->tchan, UDMA_TCHAN_RT_PCNT_REG, val);
706 
707 		val = udma_tchanrt_read(uc->tchan, UDMA_TCHAN_RT_PEER_BCNT_REG);
708 		udma_tchanrt_write(uc->tchan, UDMA_TCHAN_RT_PEER_BCNT_REG, val);
709 	}
710 
711 	if (uc->rchan) {
712 		val = udma_rchanrt_read(uc->rchan, UDMA_RCHAN_RT_BCNT_REG);
713 		udma_rchanrt_write(uc->rchan, UDMA_RCHAN_RT_BCNT_REG, val);
714 
715 		val = udma_rchanrt_read(uc->rchan, UDMA_RCHAN_RT_SBCNT_REG);
716 		udma_rchanrt_write(uc->rchan, UDMA_RCHAN_RT_SBCNT_REG, val);
717 
718 		val = udma_rchanrt_read(uc->rchan, UDMA_RCHAN_RT_PCNT_REG);
719 		udma_rchanrt_write(uc->rchan, UDMA_RCHAN_RT_PCNT_REG, val);
720 
721 		val = udma_rchanrt_read(uc->rchan, UDMA_RCHAN_RT_PEER_BCNT_REG);
722 		udma_rchanrt_write(uc->rchan, UDMA_RCHAN_RT_PEER_BCNT_REG, val);
723 	}
724 
725 	uc->bcnt = 0;
726 }
727 
728 static int udma_reset_chan(struct udma_chan *uc, bool hard)
729 {
730 	switch (uc->config.dir) {
731 	case DMA_DEV_TO_MEM:
732 		udma_rchanrt_write(uc->rchan, UDMA_RCHAN_RT_PEER_RT_EN_REG, 0);
733 		udma_rchanrt_write(uc->rchan, UDMA_RCHAN_RT_CTL_REG, 0);
734 		break;
735 	case DMA_MEM_TO_DEV:
736 		udma_tchanrt_write(uc->tchan, UDMA_TCHAN_RT_CTL_REG, 0);
737 		udma_tchanrt_write(uc->tchan, UDMA_TCHAN_RT_PEER_RT_EN_REG, 0);
738 		break;
739 	case DMA_MEM_TO_MEM:
740 		udma_rchanrt_write(uc->rchan, UDMA_RCHAN_RT_CTL_REG, 0);
741 		udma_tchanrt_write(uc->tchan, UDMA_TCHAN_RT_CTL_REG, 0);
742 		break;
743 	default:
744 		return -EINVAL;
745 	}
746 
747 	/* Reset all counters */
748 	udma_reset_counters(uc);
749 
750 	/* Hard reset: re-initialize the channel to reset */
751 	if (hard) {
752 		struct udma_chan_config ucc_backup;
753 		int ret;
754 
755 		memcpy(&ucc_backup, &uc->config, sizeof(uc->config));
756 		uc->ud->ddev.device_free_chan_resources(&uc->vc.chan);
757 
758 		/* restore the channel configuration */
759 		memcpy(&uc->config, &ucc_backup, sizeof(uc->config));
760 		ret = uc->ud->ddev.device_alloc_chan_resources(&uc->vc.chan);
761 		if (ret)
762 			return ret;
763 
764 		/*
765 		 * Setting forced teardown after forced reset helps recovering
766 		 * the rchan.
767 		 */
768 		if (uc->config.dir == DMA_DEV_TO_MEM)
769 			udma_rchanrt_write(uc->rchan, UDMA_RCHAN_RT_CTL_REG,
770 					   UDMA_CHAN_RT_CTL_EN |
771 					   UDMA_CHAN_RT_CTL_TDOWN |
772 					   UDMA_CHAN_RT_CTL_FTDOWN);
773 	}
774 	uc->state = UDMA_CHAN_IS_IDLE;
775 
776 	return 0;
777 }
778 
779 static void udma_start_desc(struct udma_chan *uc)
780 {
781 	struct udma_chan_config *ucc = &uc->config;
782 
783 	if (ucc->pkt_mode && (uc->cyclic || ucc->dir == DMA_DEV_TO_MEM)) {
784 		int i;
785 
786 		/* Push all descriptors to ring for packet mode cyclic or RX */
787 		for (i = 0; i < uc->desc->sglen; i++)
788 			udma_push_to_ring(uc, i);
789 	} else {
790 		udma_push_to_ring(uc, 0);
791 	}
792 }
793 
794 static bool udma_chan_needs_reconfiguration(struct udma_chan *uc)
795 {
796 	/* Only PDMAs have staticTR */
797 	if (uc->config.ep_type == PSIL_EP_NATIVE)
798 		return false;
799 
800 	/* Check if the staticTR configuration has changed for TX */
801 	if (memcmp(&uc->static_tr, &uc->desc->static_tr, sizeof(uc->static_tr)))
802 		return true;
803 
804 	return false;
805 }
806 
807 static int udma_start(struct udma_chan *uc)
808 {
809 	struct virt_dma_desc *vd = vchan_next_desc(&uc->vc);
810 
811 	if (!vd) {
812 		uc->desc = NULL;
813 		return -ENOENT;
814 	}
815 
816 	list_del(&vd->node);
817 
818 	uc->desc = to_udma_desc(&vd->tx);
819 
820 	/* Channel is already running and does not need reconfiguration */
821 	if (udma_is_chan_running(uc) && !udma_chan_needs_reconfiguration(uc)) {
822 		udma_start_desc(uc);
823 		goto out;
824 	}
825 
826 	/* Make sure that we clear the teardown bit, if it is set */
827 	udma_reset_chan(uc, false);
828 
829 	/* Push descriptors before we start the channel */
830 	udma_start_desc(uc);
831 
832 	switch (uc->desc->dir) {
833 	case DMA_DEV_TO_MEM:
834 		/* Config remote TR */
835 		if (uc->config.ep_type == PSIL_EP_PDMA_XY) {
836 			u32 val = PDMA_STATIC_TR_Y(uc->desc->static_tr.elcnt) |
837 				  PDMA_STATIC_TR_X(uc->desc->static_tr.elsize);
838 			const struct udma_match_data *match_data =
839 							uc->ud->match_data;
840 
841 			if (uc->config.enable_acc32)
842 				val |= PDMA_STATIC_TR_XY_ACC32;
843 			if (uc->config.enable_burst)
844 				val |= PDMA_STATIC_TR_XY_BURST;
845 
846 			udma_rchanrt_write(uc->rchan,
847 				UDMA_RCHAN_RT_PEER_STATIC_TR_XY_REG, val);
848 
849 			udma_rchanrt_write(uc->rchan,
850 				UDMA_RCHAN_RT_PEER_STATIC_TR_Z_REG,
851 				PDMA_STATIC_TR_Z(uc->desc->static_tr.bstcnt,
852 						 match_data->statictr_z_mask));
853 
854 			/* save the current staticTR configuration */
855 			memcpy(&uc->static_tr, &uc->desc->static_tr,
856 			       sizeof(uc->static_tr));
857 		}
858 
859 		udma_rchanrt_write(uc->rchan, UDMA_RCHAN_RT_CTL_REG,
860 				   UDMA_CHAN_RT_CTL_EN);
861 
862 		/* Enable remote */
863 		udma_rchanrt_write(uc->rchan, UDMA_RCHAN_RT_PEER_RT_EN_REG,
864 				   UDMA_PEER_RT_EN_ENABLE);
865 
866 		break;
867 	case DMA_MEM_TO_DEV:
868 		/* Config remote TR */
869 		if (uc->config.ep_type == PSIL_EP_PDMA_XY) {
870 			u32 val = PDMA_STATIC_TR_Y(uc->desc->static_tr.elcnt) |
871 				  PDMA_STATIC_TR_X(uc->desc->static_tr.elsize);
872 
873 			if (uc->config.enable_acc32)
874 				val |= PDMA_STATIC_TR_XY_ACC32;
875 			if (uc->config.enable_burst)
876 				val |= PDMA_STATIC_TR_XY_BURST;
877 
878 			udma_tchanrt_write(uc->tchan,
879 				UDMA_TCHAN_RT_PEER_STATIC_TR_XY_REG, val);
880 
881 			/* save the current staticTR configuration */
882 			memcpy(&uc->static_tr, &uc->desc->static_tr,
883 			       sizeof(uc->static_tr));
884 		}
885 
886 		/* Enable remote */
887 		udma_tchanrt_write(uc->tchan, UDMA_TCHAN_RT_PEER_RT_EN_REG,
888 				   UDMA_PEER_RT_EN_ENABLE);
889 
890 		udma_tchanrt_write(uc->tchan, UDMA_TCHAN_RT_CTL_REG,
891 				   UDMA_CHAN_RT_CTL_EN);
892 
893 		break;
894 	case DMA_MEM_TO_MEM:
895 		udma_rchanrt_write(uc->rchan, UDMA_RCHAN_RT_CTL_REG,
896 				   UDMA_CHAN_RT_CTL_EN);
897 		udma_tchanrt_write(uc->tchan, UDMA_TCHAN_RT_CTL_REG,
898 				   UDMA_CHAN_RT_CTL_EN);
899 
900 		break;
901 	default:
902 		return -EINVAL;
903 	}
904 
905 	uc->state = UDMA_CHAN_IS_ACTIVE;
906 out:
907 
908 	return 0;
909 }
910 
911 static int udma_stop(struct udma_chan *uc)
912 {
913 	enum udma_chan_state old_state = uc->state;
914 
915 	uc->state = UDMA_CHAN_IS_TERMINATING;
916 	reinit_completion(&uc->teardown_completed);
917 
918 	switch (uc->config.dir) {
919 	case DMA_DEV_TO_MEM:
920 		if (!uc->cyclic && !uc->desc)
921 			udma_push_to_ring(uc, -1);
922 
923 		udma_rchanrt_write(uc->rchan, UDMA_RCHAN_RT_PEER_RT_EN_REG,
924 				   UDMA_PEER_RT_EN_ENABLE |
925 				   UDMA_PEER_RT_EN_TEARDOWN);
926 		break;
927 	case DMA_MEM_TO_DEV:
928 		udma_tchanrt_write(uc->tchan, UDMA_TCHAN_RT_PEER_RT_EN_REG,
929 				   UDMA_PEER_RT_EN_ENABLE |
930 				   UDMA_PEER_RT_EN_FLUSH);
931 		udma_tchanrt_write(uc->tchan, UDMA_TCHAN_RT_CTL_REG,
932 				   UDMA_CHAN_RT_CTL_EN |
933 				   UDMA_CHAN_RT_CTL_TDOWN);
934 		break;
935 	case DMA_MEM_TO_MEM:
936 		udma_tchanrt_write(uc->tchan, UDMA_TCHAN_RT_CTL_REG,
937 				   UDMA_CHAN_RT_CTL_EN |
938 				   UDMA_CHAN_RT_CTL_TDOWN);
939 		break;
940 	default:
941 		uc->state = old_state;
942 		complete_all(&uc->teardown_completed);
943 		return -EINVAL;
944 	}
945 
946 	return 0;
947 }
948 
949 static void udma_cyclic_packet_elapsed(struct udma_chan *uc)
950 {
951 	struct udma_desc *d = uc->desc;
952 	struct cppi5_host_desc_t *h_desc;
953 
954 	h_desc = d->hwdesc[d->desc_idx].cppi5_desc_vaddr;
955 	cppi5_hdesc_reset_to_original(h_desc);
956 	udma_push_to_ring(uc, d->desc_idx);
957 	d->desc_idx = (d->desc_idx + 1) % d->sglen;
958 }
959 
960 static inline void udma_fetch_epib(struct udma_chan *uc, struct udma_desc *d)
961 {
962 	struct cppi5_host_desc_t *h_desc = d->hwdesc[0].cppi5_desc_vaddr;
963 
964 	memcpy(d->metadata, h_desc->epib, d->metadata_size);
965 }
966 
967 static bool udma_is_desc_really_done(struct udma_chan *uc, struct udma_desc *d)
968 {
969 	u32 peer_bcnt, bcnt;
970 
971 	/* Only TX towards PDMA is affected */
972 	if (uc->config.ep_type == PSIL_EP_NATIVE ||
973 	    uc->config.dir != DMA_MEM_TO_DEV)
974 		return true;
975 
976 	peer_bcnt = udma_tchanrt_read(uc->tchan, UDMA_TCHAN_RT_PEER_BCNT_REG);
977 	bcnt = udma_tchanrt_read(uc->tchan, UDMA_TCHAN_RT_BCNT_REG);
978 
979 	/* Transfer is incomplete, store current residue and time stamp */
980 	if (peer_bcnt < bcnt) {
981 		uc->tx_drain.residue = bcnt - peer_bcnt;
982 		uc->tx_drain.tstamp = ktime_get();
983 		return false;
984 	}
985 
986 	return true;
987 }
988 
989 static void udma_check_tx_completion(struct work_struct *work)
990 {
991 	struct udma_chan *uc = container_of(work, typeof(*uc),
992 					    tx_drain.work.work);
993 	bool desc_done = true;
994 	u32 residue_diff;
995 	ktime_t time_diff;
996 	unsigned long delay;
997 
998 	while (1) {
999 		if (uc->desc) {
1000 			/* Get previous residue and time stamp */
1001 			residue_diff = uc->tx_drain.residue;
1002 			time_diff = uc->tx_drain.tstamp;
1003 			/*
1004 			 * Get current residue and time stamp or see if
1005 			 * transfer is complete
1006 			 */
1007 			desc_done = udma_is_desc_really_done(uc, uc->desc);
1008 		}
1009 
1010 		if (!desc_done) {
1011 			/*
1012 			 * Find the time delta and residue delta w.r.t
1013 			 * previous poll
1014 			 */
1015 			time_diff = ktime_sub(uc->tx_drain.tstamp,
1016 					      time_diff) + 1;
1017 			residue_diff -= uc->tx_drain.residue;
1018 			if (residue_diff) {
1019 				/*
1020 				 * Try to guess when we should check
1021 				 * next time by calculating rate at
1022 				 * which data is being drained at the
1023 				 * peer device
1024 				 */
1025 				delay = (time_diff / residue_diff) *
1026 					uc->tx_drain.residue;
1027 			} else {
1028 				/* No progress, check again in 1 second  */
1029 				schedule_delayed_work(&uc->tx_drain.work, HZ);
1030 				break;
1031 			}
1032 
1033 			usleep_range(ktime_to_us(delay),
1034 				     ktime_to_us(delay) + 10);
1035 			continue;
1036 		}
1037 
1038 		if (uc->desc) {
1039 			struct udma_desc *d = uc->desc;
1040 
1041 			uc->bcnt += d->residue;
1042 			udma_start(uc);
1043 			vchan_cookie_complete(&d->vd);
1044 			break;
1045 		}
1046 
1047 		break;
1048 	}
1049 }
1050 
1051 static irqreturn_t udma_ring_irq_handler(int irq, void *data)
1052 {
1053 	struct udma_chan *uc = data;
1054 	struct udma_desc *d;
1055 	unsigned long flags;
1056 	dma_addr_t paddr = 0;
1057 
1058 	if (udma_pop_from_ring(uc, &paddr) || !paddr)
1059 		return IRQ_HANDLED;
1060 
1061 	spin_lock_irqsave(&uc->vc.lock, flags);
1062 
1063 	/* Teardown completion message */
1064 	if (cppi5_desc_is_tdcm(paddr)) {
1065 		complete_all(&uc->teardown_completed);
1066 
1067 		if (uc->terminated_desc) {
1068 			udma_desc_free(&uc->terminated_desc->vd);
1069 			uc->terminated_desc = NULL;
1070 		}
1071 
1072 		if (!uc->desc)
1073 			udma_start(uc);
1074 
1075 		goto out;
1076 	}
1077 
1078 	d = udma_udma_desc_from_paddr(uc, paddr);
1079 
1080 	if (d) {
1081 		dma_addr_t desc_paddr = udma_curr_cppi5_desc_paddr(d,
1082 								   d->desc_idx);
1083 		if (desc_paddr != paddr) {
1084 			dev_err(uc->ud->dev, "not matching descriptors!\n");
1085 			goto out;
1086 		}
1087 
1088 		if (d == uc->desc) {
1089 			/* active descriptor */
1090 			if (uc->cyclic) {
1091 				udma_cyclic_packet_elapsed(uc);
1092 				vchan_cyclic_callback(&d->vd);
1093 			} else {
1094 				if (udma_is_desc_really_done(uc, d)) {
1095 					uc->bcnt += d->residue;
1096 					udma_start(uc);
1097 					vchan_cookie_complete(&d->vd);
1098 				} else {
1099 					schedule_delayed_work(&uc->tx_drain.work,
1100 							      0);
1101 				}
1102 			}
1103 		} else {
1104 			/*
1105 			 * terminated descriptor, mark the descriptor as
1106 			 * completed to update the channel's cookie marker
1107 			 */
1108 			dma_cookie_complete(&d->vd.tx);
1109 		}
1110 	}
1111 out:
1112 	spin_unlock_irqrestore(&uc->vc.lock, flags);
1113 
1114 	return IRQ_HANDLED;
1115 }
1116 
1117 static irqreturn_t udma_udma_irq_handler(int irq, void *data)
1118 {
1119 	struct udma_chan *uc = data;
1120 	struct udma_desc *d;
1121 	unsigned long flags;
1122 
1123 	spin_lock_irqsave(&uc->vc.lock, flags);
1124 	d = uc->desc;
1125 	if (d) {
1126 		d->tr_idx = (d->tr_idx + 1) % d->sglen;
1127 
1128 		if (uc->cyclic) {
1129 			vchan_cyclic_callback(&d->vd);
1130 		} else {
1131 			/* TODO: figure out the real amount of data */
1132 			uc->bcnt += d->residue;
1133 			udma_start(uc);
1134 			vchan_cookie_complete(&d->vd);
1135 		}
1136 	}
1137 
1138 	spin_unlock_irqrestore(&uc->vc.lock, flags);
1139 
1140 	return IRQ_HANDLED;
1141 }
1142 
1143 /**
1144  * __udma_alloc_gp_rflow_range - alloc range of GP RX flows
1145  * @ud: UDMA device
1146  * @from: Start the search from this flow id number
1147  * @cnt: Number of consecutive flow ids to allocate
1148  *
1149  * Allocate range of RX flow ids for future use, those flows can be requested
1150  * only using explicit flow id number. if @from is set to -1 it will try to find
1151  * first free range. if @from is positive value it will force allocation only
1152  * of the specified range of flows.
1153  *
1154  * Returns -ENOMEM if can't find free range.
1155  * -EEXIST if requested range is busy.
1156  * -EINVAL if wrong input values passed.
1157  * Returns flow id on success.
1158  */
1159 static int __udma_alloc_gp_rflow_range(struct udma_dev *ud, int from, int cnt)
1160 {
1161 	int start, tmp_from;
1162 	DECLARE_BITMAP(tmp, K3_UDMA_MAX_RFLOWS);
1163 
1164 	tmp_from = from;
1165 	if (tmp_from < 0)
1166 		tmp_from = ud->rchan_cnt;
1167 	/* default flows can't be allocated and accessible only by id */
1168 	if (tmp_from < ud->rchan_cnt)
1169 		return -EINVAL;
1170 
1171 	if (tmp_from + cnt > ud->rflow_cnt)
1172 		return -EINVAL;
1173 
1174 	bitmap_or(tmp, ud->rflow_gp_map, ud->rflow_gp_map_allocated,
1175 		  ud->rflow_cnt);
1176 
1177 	start = bitmap_find_next_zero_area(tmp,
1178 					   ud->rflow_cnt,
1179 					   tmp_from, cnt, 0);
1180 	if (start >= ud->rflow_cnt)
1181 		return -ENOMEM;
1182 
1183 	if (from >= 0 && start != from)
1184 		return -EEXIST;
1185 
1186 	bitmap_set(ud->rflow_gp_map_allocated, start, cnt);
1187 	return start;
1188 }
1189 
1190 static int __udma_free_gp_rflow_range(struct udma_dev *ud, int from, int cnt)
1191 {
1192 	if (from < ud->rchan_cnt)
1193 		return -EINVAL;
1194 	if (from + cnt > ud->rflow_cnt)
1195 		return -EINVAL;
1196 
1197 	bitmap_clear(ud->rflow_gp_map_allocated, from, cnt);
1198 	return 0;
1199 }
1200 
1201 static struct udma_rflow *__udma_get_rflow(struct udma_dev *ud, int id)
1202 {
1203 	/*
1204 	 * Attempt to request rflow by ID can be made for any rflow
1205 	 * if not in use with assumption that caller knows what's doing.
1206 	 * TI-SCI FW will perform additional permission check ant way, it's
1207 	 * safe
1208 	 */
1209 
1210 	if (id < 0 || id >= ud->rflow_cnt)
1211 		return ERR_PTR(-ENOENT);
1212 
1213 	if (test_bit(id, ud->rflow_in_use))
1214 		return ERR_PTR(-ENOENT);
1215 
1216 	/* GP rflow has to be allocated first */
1217 	if (!test_bit(id, ud->rflow_gp_map) &&
1218 	    !test_bit(id, ud->rflow_gp_map_allocated))
1219 		return ERR_PTR(-EINVAL);
1220 
1221 	dev_dbg(ud->dev, "get rflow%d\n", id);
1222 	set_bit(id, ud->rflow_in_use);
1223 	return &ud->rflows[id];
1224 }
1225 
1226 static void __udma_put_rflow(struct udma_dev *ud, struct udma_rflow *rflow)
1227 {
1228 	if (!test_bit(rflow->id, ud->rflow_in_use)) {
1229 		dev_err(ud->dev, "attempt to put unused rflow%d\n", rflow->id);
1230 		return;
1231 	}
1232 
1233 	dev_dbg(ud->dev, "put rflow%d\n", rflow->id);
1234 	clear_bit(rflow->id, ud->rflow_in_use);
1235 }
1236 
1237 #define UDMA_RESERVE_RESOURCE(res)					\
1238 static struct udma_##res *__udma_reserve_##res(struct udma_dev *ud,	\
1239 					       enum udma_tp_level tpl,	\
1240 					       int id)			\
1241 {									\
1242 	if (id >= 0) {							\
1243 		if (test_bit(id, ud->res##_map)) {			\
1244 			dev_err(ud->dev, "res##%d is in use\n", id);	\
1245 			return ERR_PTR(-ENOENT);			\
1246 		}							\
1247 	} else {							\
1248 		int start;						\
1249 									\
1250 		if (tpl >= ud->match_data->tpl_levels)			\
1251 			tpl = ud->match_data->tpl_levels - 1;		\
1252 									\
1253 		start = ud->match_data->level_start_idx[tpl];		\
1254 									\
1255 		id = find_next_zero_bit(ud->res##_map, ud->res##_cnt,	\
1256 					start);				\
1257 		if (id == ud->res##_cnt) {				\
1258 			return ERR_PTR(-ENOENT);			\
1259 		}							\
1260 	}								\
1261 									\
1262 	set_bit(id, ud->res##_map);					\
1263 	return &ud->res##s[id];						\
1264 }
1265 
1266 UDMA_RESERVE_RESOURCE(tchan);
1267 UDMA_RESERVE_RESOURCE(rchan);
1268 
1269 static int udma_get_tchan(struct udma_chan *uc)
1270 {
1271 	struct udma_dev *ud = uc->ud;
1272 
1273 	if (uc->tchan) {
1274 		dev_dbg(ud->dev, "chan%d: already have tchan%d allocated\n",
1275 			uc->id, uc->tchan->id);
1276 		return 0;
1277 	}
1278 
1279 	uc->tchan = __udma_reserve_tchan(ud, uc->config.channel_tpl, -1);
1280 
1281 	return PTR_ERR_OR_ZERO(uc->tchan);
1282 }
1283 
1284 static int udma_get_rchan(struct udma_chan *uc)
1285 {
1286 	struct udma_dev *ud = uc->ud;
1287 
1288 	if (uc->rchan) {
1289 		dev_dbg(ud->dev, "chan%d: already have rchan%d allocated\n",
1290 			uc->id, uc->rchan->id);
1291 		return 0;
1292 	}
1293 
1294 	uc->rchan = __udma_reserve_rchan(ud, uc->config.channel_tpl, -1);
1295 
1296 	return PTR_ERR_OR_ZERO(uc->rchan);
1297 }
1298 
1299 static int udma_get_chan_pair(struct udma_chan *uc)
1300 {
1301 	struct udma_dev *ud = uc->ud;
1302 	const struct udma_match_data *match_data = ud->match_data;
1303 	int chan_id, end;
1304 
1305 	if ((uc->tchan && uc->rchan) && uc->tchan->id == uc->rchan->id) {
1306 		dev_info(ud->dev, "chan%d: already have %d pair allocated\n",
1307 			 uc->id, uc->tchan->id);
1308 		return 0;
1309 	}
1310 
1311 	if (uc->tchan) {
1312 		dev_err(ud->dev, "chan%d: already have tchan%d allocated\n",
1313 			uc->id, uc->tchan->id);
1314 		return -EBUSY;
1315 	} else if (uc->rchan) {
1316 		dev_err(ud->dev, "chan%d: already have rchan%d allocated\n",
1317 			uc->id, uc->rchan->id);
1318 		return -EBUSY;
1319 	}
1320 
1321 	/* Can be optimized, but let's have it like this for now */
1322 	end = min(ud->tchan_cnt, ud->rchan_cnt);
1323 	/* Try to use the highest TPL channel pair for MEM_TO_MEM channels */
1324 	chan_id = match_data->level_start_idx[match_data->tpl_levels - 1];
1325 	for (; chan_id < end; chan_id++) {
1326 		if (!test_bit(chan_id, ud->tchan_map) &&
1327 		    !test_bit(chan_id, ud->rchan_map))
1328 			break;
1329 	}
1330 
1331 	if (chan_id == end)
1332 		return -ENOENT;
1333 
1334 	set_bit(chan_id, ud->tchan_map);
1335 	set_bit(chan_id, ud->rchan_map);
1336 	uc->tchan = &ud->tchans[chan_id];
1337 	uc->rchan = &ud->rchans[chan_id];
1338 
1339 	return 0;
1340 }
1341 
1342 static int udma_get_rflow(struct udma_chan *uc, int flow_id)
1343 {
1344 	struct udma_dev *ud = uc->ud;
1345 
1346 	if (!uc->rchan) {
1347 		dev_err(ud->dev, "chan%d: does not have rchan??\n", uc->id);
1348 		return -EINVAL;
1349 	}
1350 
1351 	if (uc->rflow) {
1352 		dev_dbg(ud->dev, "chan%d: already have rflow%d allocated\n",
1353 			uc->id, uc->rflow->id);
1354 		return 0;
1355 	}
1356 
1357 	uc->rflow = __udma_get_rflow(ud, flow_id);
1358 
1359 	return PTR_ERR_OR_ZERO(uc->rflow);
1360 }
1361 
1362 static void udma_put_rchan(struct udma_chan *uc)
1363 {
1364 	struct udma_dev *ud = uc->ud;
1365 
1366 	if (uc->rchan) {
1367 		dev_dbg(ud->dev, "chan%d: put rchan%d\n", uc->id,
1368 			uc->rchan->id);
1369 		clear_bit(uc->rchan->id, ud->rchan_map);
1370 		uc->rchan = NULL;
1371 	}
1372 }
1373 
1374 static void udma_put_tchan(struct udma_chan *uc)
1375 {
1376 	struct udma_dev *ud = uc->ud;
1377 
1378 	if (uc->tchan) {
1379 		dev_dbg(ud->dev, "chan%d: put tchan%d\n", uc->id,
1380 			uc->tchan->id);
1381 		clear_bit(uc->tchan->id, ud->tchan_map);
1382 		uc->tchan = NULL;
1383 	}
1384 }
1385 
1386 static void udma_put_rflow(struct udma_chan *uc)
1387 {
1388 	struct udma_dev *ud = uc->ud;
1389 
1390 	if (uc->rflow) {
1391 		dev_dbg(ud->dev, "chan%d: put rflow%d\n", uc->id,
1392 			uc->rflow->id);
1393 		__udma_put_rflow(ud, uc->rflow);
1394 		uc->rflow = NULL;
1395 	}
1396 }
1397 
1398 static void udma_free_tx_resources(struct udma_chan *uc)
1399 {
1400 	if (!uc->tchan)
1401 		return;
1402 
1403 	k3_ringacc_ring_free(uc->tchan->t_ring);
1404 	k3_ringacc_ring_free(uc->tchan->tc_ring);
1405 	uc->tchan->t_ring = NULL;
1406 	uc->tchan->tc_ring = NULL;
1407 
1408 	udma_put_tchan(uc);
1409 }
1410 
1411 static int udma_alloc_tx_resources(struct udma_chan *uc)
1412 {
1413 	struct k3_ring_cfg ring_cfg;
1414 	struct udma_dev *ud = uc->ud;
1415 	int ret;
1416 
1417 	ret = udma_get_tchan(uc);
1418 	if (ret)
1419 		return ret;
1420 
1421 	ret = k3_ringacc_request_rings_pair(ud->ringacc, uc->tchan->id, -1,
1422 					    &uc->tchan->t_ring,
1423 					    &uc->tchan->tc_ring);
1424 	if (ret) {
1425 		ret = -EBUSY;
1426 		goto err_ring;
1427 	}
1428 
1429 	memset(&ring_cfg, 0, sizeof(ring_cfg));
1430 	ring_cfg.size = K3_UDMA_DEFAULT_RING_SIZE;
1431 	ring_cfg.elm_size = K3_RINGACC_RING_ELSIZE_8;
1432 	ring_cfg.mode = K3_RINGACC_RING_MODE_MESSAGE;
1433 
1434 	ret = k3_ringacc_ring_cfg(uc->tchan->t_ring, &ring_cfg);
1435 	ret |= k3_ringacc_ring_cfg(uc->tchan->tc_ring, &ring_cfg);
1436 
1437 	if (ret)
1438 		goto err_ringcfg;
1439 
1440 	return 0;
1441 
1442 err_ringcfg:
1443 	k3_ringacc_ring_free(uc->tchan->tc_ring);
1444 	uc->tchan->tc_ring = NULL;
1445 	k3_ringacc_ring_free(uc->tchan->t_ring);
1446 	uc->tchan->t_ring = NULL;
1447 err_ring:
1448 	udma_put_tchan(uc);
1449 
1450 	return ret;
1451 }
1452 
1453 static void udma_free_rx_resources(struct udma_chan *uc)
1454 {
1455 	if (!uc->rchan)
1456 		return;
1457 
1458 	if (uc->rflow) {
1459 		struct udma_rflow *rflow = uc->rflow;
1460 
1461 		k3_ringacc_ring_free(rflow->fd_ring);
1462 		k3_ringacc_ring_free(rflow->r_ring);
1463 		rflow->fd_ring = NULL;
1464 		rflow->r_ring = NULL;
1465 
1466 		udma_put_rflow(uc);
1467 	}
1468 
1469 	udma_put_rchan(uc);
1470 }
1471 
1472 static int udma_alloc_rx_resources(struct udma_chan *uc)
1473 {
1474 	struct udma_dev *ud = uc->ud;
1475 	struct k3_ring_cfg ring_cfg;
1476 	struct udma_rflow *rflow;
1477 	int fd_ring_id;
1478 	int ret;
1479 
1480 	ret = udma_get_rchan(uc);
1481 	if (ret)
1482 		return ret;
1483 
1484 	/* For MEM_TO_MEM we don't need rflow or rings */
1485 	if (uc->config.dir == DMA_MEM_TO_MEM)
1486 		return 0;
1487 
1488 	ret = udma_get_rflow(uc, uc->rchan->id);
1489 	if (ret) {
1490 		ret = -EBUSY;
1491 		goto err_rflow;
1492 	}
1493 
1494 	rflow = uc->rflow;
1495 	fd_ring_id = ud->tchan_cnt + ud->echan_cnt + uc->rchan->id;
1496 	ret = k3_ringacc_request_rings_pair(ud->ringacc, fd_ring_id, -1,
1497 					    &rflow->fd_ring, &rflow->r_ring);
1498 	if (ret) {
1499 		ret = -EBUSY;
1500 		goto err_ring;
1501 	}
1502 
1503 	memset(&ring_cfg, 0, sizeof(ring_cfg));
1504 
1505 	if (uc->config.pkt_mode)
1506 		ring_cfg.size = SG_MAX_SEGMENTS;
1507 	else
1508 		ring_cfg.size = K3_UDMA_DEFAULT_RING_SIZE;
1509 
1510 	ring_cfg.elm_size = K3_RINGACC_RING_ELSIZE_8;
1511 	ring_cfg.mode = K3_RINGACC_RING_MODE_MESSAGE;
1512 
1513 	ret = k3_ringacc_ring_cfg(rflow->fd_ring, &ring_cfg);
1514 	ring_cfg.size = K3_UDMA_DEFAULT_RING_SIZE;
1515 	ret |= k3_ringacc_ring_cfg(rflow->r_ring, &ring_cfg);
1516 
1517 	if (ret)
1518 		goto err_ringcfg;
1519 
1520 	return 0;
1521 
1522 err_ringcfg:
1523 	k3_ringacc_ring_free(rflow->r_ring);
1524 	rflow->r_ring = NULL;
1525 	k3_ringacc_ring_free(rflow->fd_ring);
1526 	rflow->fd_ring = NULL;
1527 err_ring:
1528 	udma_put_rflow(uc);
1529 err_rflow:
1530 	udma_put_rchan(uc);
1531 
1532 	return ret;
1533 }
1534 
1535 #define TISCI_TCHAN_VALID_PARAMS (				\
1536 	TI_SCI_MSG_VALUE_RM_UDMAP_CH_PAUSE_ON_ERR_VALID |	\
1537 	TI_SCI_MSG_VALUE_RM_UDMAP_CH_TX_FILT_EINFO_VALID |	\
1538 	TI_SCI_MSG_VALUE_RM_UDMAP_CH_TX_FILT_PSWORDS_VALID |	\
1539 	TI_SCI_MSG_VALUE_RM_UDMAP_CH_CHAN_TYPE_VALID |		\
1540 	TI_SCI_MSG_VALUE_RM_UDMAP_CH_TX_SUPR_TDPKT_VALID |	\
1541 	TI_SCI_MSG_VALUE_RM_UDMAP_CH_FETCH_SIZE_VALID |		\
1542 	TI_SCI_MSG_VALUE_RM_UDMAP_CH_CQ_QNUM_VALID |		\
1543 	TI_SCI_MSG_VALUE_RM_UDMAP_CH_ATYPE_VALID)
1544 
1545 #define TISCI_RCHAN_VALID_PARAMS (				\
1546 	TI_SCI_MSG_VALUE_RM_UDMAP_CH_PAUSE_ON_ERR_VALID |	\
1547 	TI_SCI_MSG_VALUE_RM_UDMAP_CH_FETCH_SIZE_VALID |		\
1548 	TI_SCI_MSG_VALUE_RM_UDMAP_CH_CQ_QNUM_VALID |		\
1549 	TI_SCI_MSG_VALUE_RM_UDMAP_CH_CHAN_TYPE_VALID |		\
1550 	TI_SCI_MSG_VALUE_RM_UDMAP_CH_RX_IGNORE_SHORT_VALID |	\
1551 	TI_SCI_MSG_VALUE_RM_UDMAP_CH_RX_IGNORE_LONG_VALID |	\
1552 	TI_SCI_MSG_VALUE_RM_UDMAP_CH_RX_FLOWID_START_VALID |	\
1553 	TI_SCI_MSG_VALUE_RM_UDMAP_CH_RX_FLOWID_CNT_VALID |	\
1554 	TI_SCI_MSG_VALUE_RM_UDMAP_CH_ATYPE_VALID)
1555 
1556 static int udma_tisci_m2m_channel_config(struct udma_chan *uc)
1557 {
1558 	struct udma_dev *ud = uc->ud;
1559 	struct udma_tisci_rm *tisci_rm = &ud->tisci_rm;
1560 	const struct ti_sci_rm_udmap_ops *tisci_ops = tisci_rm->tisci_udmap_ops;
1561 	struct udma_tchan *tchan = uc->tchan;
1562 	struct udma_rchan *rchan = uc->rchan;
1563 	int ret = 0;
1564 
1565 	/* Non synchronized - mem to mem type of transfer */
1566 	int tc_ring = k3_ringacc_get_ring_id(tchan->tc_ring);
1567 	struct ti_sci_msg_rm_udmap_tx_ch_cfg req_tx = { 0 };
1568 	struct ti_sci_msg_rm_udmap_rx_ch_cfg req_rx = { 0 };
1569 
1570 	req_tx.valid_params = TISCI_TCHAN_VALID_PARAMS;
1571 	req_tx.nav_id = tisci_rm->tisci_dev_id;
1572 	req_tx.index = tchan->id;
1573 	req_tx.tx_chan_type = TI_SCI_RM_UDMAP_CHAN_TYPE_3RDP_BCOPY_PBRR;
1574 	req_tx.tx_fetch_size = sizeof(struct cppi5_desc_hdr_t) >> 2;
1575 	req_tx.txcq_qnum = tc_ring;
1576 	req_tx.tx_atype = ud->atype;
1577 
1578 	ret = tisci_ops->tx_ch_cfg(tisci_rm->tisci, &req_tx);
1579 	if (ret) {
1580 		dev_err(ud->dev, "tchan%d cfg failed %d\n", tchan->id, ret);
1581 		return ret;
1582 	}
1583 
1584 	req_rx.valid_params = TISCI_RCHAN_VALID_PARAMS;
1585 	req_rx.nav_id = tisci_rm->tisci_dev_id;
1586 	req_rx.index = rchan->id;
1587 	req_rx.rx_fetch_size = sizeof(struct cppi5_desc_hdr_t) >> 2;
1588 	req_rx.rxcq_qnum = tc_ring;
1589 	req_rx.rx_chan_type = TI_SCI_RM_UDMAP_CHAN_TYPE_3RDP_BCOPY_PBRR;
1590 	req_rx.rx_atype = ud->atype;
1591 
1592 	ret = tisci_ops->rx_ch_cfg(tisci_rm->tisci, &req_rx);
1593 	if (ret)
1594 		dev_err(ud->dev, "rchan%d alloc failed %d\n", rchan->id, ret);
1595 
1596 	return ret;
1597 }
1598 
1599 static int udma_tisci_tx_channel_config(struct udma_chan *uc)
1600 {
1601 	struct udma_dev *ud = uc->ud;
1602 	struct udma_tisci_rm *tisci_rm = &ud->tisci_rm;
1603 	const struct ti_sci_rm_udmap_ops *tisci_ops = tisci_rm->tisci_udmap_ops;
1604 	struct udma_tchan *tchan = uc->tchan;
1605 	int tc_ring = k3_ringacc_get_ring_id(tchan->tc_ring);
1606 	struct ti_sci_msg_rm_udmap_tx_ch_cfg req_tx = { 0 };
1607 	u32 mode, fetch_size;
1608 	int ret = 0;
1609 
1610 	if (uc->config.pkt_mode) {
1611 		mode = TI_SCI_RM_UDMAP_CHAN_TYPE_PKT_PBRR;
1612 		fetch_size = cppi5_hdesc_calc_size(uc->config.needs_epib,
1613 						   uc->config.psd_size, 0);
1614 	} else {
1615 		mode = TI_SCI_RM_UDMAP_CHAN_TYPE_3RDP_PBRR;
1616 		fetch_size = sizeof(struct cppi5_desc_hdr_t);
1617 	}
1618 
1619 	req_tx.valid_params = TISCI_TCHAN_VALID_PARAMS;
1620 	req_tx.nav_id = tisci_rm->tisci_dev_id;
1621 	req_tx.index = tchan->id;
1622 	req_tx.tx_chan_type = mode;
1623 	req_tx.tx_supr_tdpkt = uc->config.notdpkt;
1624 	req_tx.tx_fetch_size = fetch_size >> 2;
1625 	req_tx.txcq_qnum = tc_ring;
1626 	req_tx.tx_atype = uc->config.atype;
1627 
1628 	ret = tisci_ops->tx_ch_cfg(tisci_rm->tisci, &req_tx);
1629 	if (ret)
1630 		dev_err(ud->dev, "tchan%d cfg failed %d\n", tchan->id, ret);
1631 
1632 	return ret;
1633 }
1634 
1635 static int udma_tisci_rx_channel_config(struct udma_chan *uc)
1636 {
1637 	struct udma_dev *ud = uc->ud;
1638 	struct udma_tisci_rm *tisci_rm = &ud->tisci_rm;
1639 	const struct ti_sci_rm_udmap_ops *tisci_ops = tisci_rm->tisci_udmap_ops;
1640 	struct udma_rchan *rchan = uc->rchan;
1641 	int fd_ring = k3_ringacc_get_ring_id(uc->rflow->fd_ring);
1642 	int rx_ring = k3_ringacc_get_ring_id(uc->rflow->r_ring);
1643 	struct ti_sci_msg_rm_udmap_rx_ch_cfg req_rx = { 0 };
1644 	struct ti_sci_msg_rm_udmap_flow_cfg flow_req = { 0 };
1645 	u32 mode, fetch_size;
1646 	int ret = 0;
1647 
1648 	if (uc->config.pkt_mode) {
1649 		mode = TI_SCI_RM_UDMAP_CHAN_TYPE_PKT_PBRR;
1650 		fetch_size = cppi5_hdesc_calc_size(uc->config.needs_epib,
1651 						   uc->config.psd_size, 0);
1652 	} else {
1653 		mode = TI_SCI_RM_UDMAP_CHAN_TYPE_3RDP_PBRR;
1654 		fetch_size = sizeof(struct cppi5_desc_hdr_t);
1655 	}
1656 
1657 	req_rx.valid_params = TISCI_RCHAN_VALID_PARAMS;
1658 	req_rx.nav_id = tisci_rm->tisci_dev_id;
1659 	req_rx.index = rchan->id;
1660 	req_rx.rx_fetch_size =  fetch_size >> 2;
1661 	req_rx.rxcq_qnum = rx_ring;
1662 	req_rx.rx_chan_type = mode;
1663 	req_rx.rx_atype = uc->config.atype;
1664 
1665 	ret = tisci_ops->rx_ch_cfg(tisci_rm->tisci, &req_rx);
1666 	if (ret) {
1667 		dev_err(ud->dev, "rchan%d cfg failed %d\n", rchan->id, ret);
1668 		return ret;
1669 	}
1670 
1671 	flow_req.valid_params =
1672 		TI_SCI_MSG_VALUE_RM_UDMAP_FLOW_EINFO_PRESENT_VALID |
1673 		TI_SCI_MSG_VALUE_RM_UDMAP_FLOW_PSINFO_PRESENT_VALID |
1674 		TI_SCI_MSG_VALUE_RM_UDMAP_FLOW_ERROR_HANDLING_VALID |
1675 		TI_SCI_MSG_VALUE_RM_UDMAP_FLOW_DESC_TYPE_VALID |
1676 		TI_SCI_MSG_VALUE_RM_UDMAP_FLOW_DEST_QNUM_VALID |
1677 		TI_SCI_MSG_VALUE_RM_UDMAP_FLOW_SRC_TAG_HI_SEL_VALID |
1678 		TI_SCI_MSG_VALUE_RM_UDMAP_FLOW_SRC_TAG_LO_SEL_VALID |
1679 		TI_SCI_MSG_VALUE_RM_UDMAP_FLOW_DEST_TAG_HI_SEL_VALID |
1680 		TI_SCI_MSG_VALUE_RM_UDMAP_FLOW_DEST_TAG_LO_SEL_VALID |
1681 		TI_SCI_MSG_VALUE_RM_UDMAP_FLOW_FDQ0_SZ0_QNUM_VALID |
1682 		TI_SCI_MSG_VALUE_RM_UDMAP_FLOW_FDQ1_QNUM_VALID |
1683 		TI_SCI_MSG_VALUE_RM_UDMAP_FLOW_FDQ2_QNUM_VALID |
1684 		TI_SCI_MSG_VALUE_RM_UDMAP_FLOW_FDQ3_QNUM_VALID;
1685 
1686 	flow_req.nav_id = tisci_rm->tisci_dev_id;
1687 	flow_req.flow_index = rchan->id;
1688 
1689 	if (uc->config.needs_epib)
1690 		flow_req.rx_einfo_present = 1;
1691 	else
1692 		flow_req.rx_einfo_present = 0;
1693 	if (uc->config.psd_size)
1694 		flow_req.rx_psinfo_present = 1;
1695 	else
1696 		flow_req.rx_psinfo_present = 0;
1697 	flow_req.rx_error_handling = 1;
1698 	flow_req.rx_dest_qnum = rx_ring;
1699 	flow_req.rx_src_tag_hi_sel = UDMA_RFLOW_SRCTAG_NONE;
1700 	flow_req.rx_src_tag_lo_sel = UDMA_RFLOW_SRCTAG_SRC_TAG;
1701 	flow_req.rx_dest_tag_hi_sel = UDMA_RFLOW_DSTTAG_DST_TAG_HI;
1702 	flow_req.rx_dest_tag_lo_sel = UDMA_RFLOW_DSTTAG_DST_TAG_LO;
1703 	flow_req.rx_fdq0_sz0_qnum = fd_ring;
1704 	flow_req.rx_fdq1_qnum = fd_ring;
1705 	flow_req.rx_fdq2_qnum = fd_ring;
1706 	flow_req.rx_fdq3_qnum = fd_ring;
1707 
1708 	ret = tisci_ops->rx_flow_cfg(tisci_rm->tisci, &flow_req);
1709 
1710 	if (ret)
1711 		dev_err(ud->dev, "flow%d config failed: %d\n", rchan->id, ret);
1712 
1713 	return 0;
1714 }
1715 
1716 static int udma_alloc_chan_resources(struct dma_chan *chan)
1717 {
1718 	struct udma_chan *uc = to_udma_chan(chan);
1719 	struct udma_dev *ud = to_udma_dev(chan->device);
1720 	const struct udma_match_data *match_data = ud->match_data;
1721 	struct k3_ring *irq_ring;
1722 	u32 irq_udma_idx;
1723 	int ret;
1724 
1725 	if (uc->config.pkt_mode || uc->config.dir == DMA_MEM_TO_MEM) {
1726 		uc->use_dma_pool = true;
1727 		/* in case of MEM_TO_MEM we have maximum of two TRs */
1728 		if (uc->config.dir == DMA_MEM_TO_MEM) {
1729 			uc->config.hdesc_size = cppi5_trdesc_calc_size(
1730 					sizeof(struct cppi5_tr_type15_t), 2);
1731 			uc->config.pkt_mode = false;
1732 		}
1733 	}
1734 
1735 	if (uc->use_dma_pool) {
1736 		uc->hdesc_pool = dma_pool_create(uc->name, ud->ddev.dev,
1737 						 uc->config.hdesc_size,
1738 						 ud->desc_align,
1739 						 0);
1740 		if (!uc->hdesc_pool) {
1741 			dev_err(ud->ddev.dev,
1742 				"Descriptor pool allocation failed\n");
1743 			uc->use_dma_pool = false;
1744 			ret = -ENOMEM;
1745 			goto err_cleanup;
1746 		}
1747 	}
1748 
1749 	/*
1750 	 * Make sure that the completion is in a known state:
1751 	 * No teardown, the channel is idle
1752 	 */
1753 	reinit_completion(&uc->teardown_completed);
1754 	complete_all(&uc->teardown_completed);
1755 	uc->state = UDMA_CHAN_IS_IDLE;
1756 
1757 	switch (uc->config.dir) {
1758 	case DMA_MEM_TO_MEM:
1759 		/* Non synchronized - mem to mem type of transfer */
1760 		dev_dbg(uc->ud->dev, "%s: chan%d as MEM-to-MEM\n", __func__,
1761 			uc->id);
1762 
1763 		ret = udma_get_chan_pair(uc);
1764 		if (ret)
1765 			goto err_cleanup;
1766 
1767 		ret = udma_alloc_tx_resources(uc);
1768 		if (ret) {
1769 			udma_put_rchan(uc);
1770 			goto err_cleanup;
1771 		}
1772 
1773 		ret = udma_alloc_rx_resources(uc);
1774 		if (ret) {
1775 			udma_free_tx_resources(uc);
1776 			goto err_cleanup;
1777 		}
1778 
1779 		uc->config.src_thread = ud->psil_base + uc->tchan->id;
1780 		uc->config.dst_thread = (ud->psil_base + uc->rchan->id) |
1781 					K3_PSIL_DST_THREAD_ID_OFFSET;
1782 
1783 		irq_ring = uc->tchan->tc_ring;
1784 		irq_udma_idx = uc->tchan->id;
1785 
1786 		ret = udma_tisci_m2m_channel_config(uc);
1787 		break;
1788 	case DMA_MEM_TO_DEV:
1789 		/* Slave transfer synchronized - mem to dev (TX) trasnfer */
1790 		dev_dbg(uc->ud->dev, "%s: chan%d as MEM-to-DEV\n", __func__,
1791 			uc->id);
1792 
1793 		ret = udma_alloc_tx_resources(uc);
1794 		if (ret)
1795 			goto err_cleanup;
1796 
1797 		uc->config.src_thread = ud->psil_base + uc->tchan->id;
1798 		uc->config.dst_thread = uc->config.remote_thread_id;
1799 		uc->config.dst_thread |= K3_PSIL_DST_THREAD_ID_OFFSET;
1800 
1801 		irq_ring = uc->tchan->tc_ring;
1802 		irq_udma_idx = uc->tchan->id;
1803 
1804 		ret = udma_tisci_tx_channel_config(uc);
1805 		break;
1806 	case DMA_DEV_TO_MEM:
1807 		/* Slave transfer synchronized - dev to mem (RX) trasnfer */
1808 		dev_dbg(uc->ud->dev, "%s: chan%d as DEV-to-MEM\n", __func__,
1809 			uc->id);
1810 
1811 		ret = udma_alloc_rx_resources(uc);
1812 		if (ret)
1813 			goto err_cleanup;
1814 
1815 		uc->config.src_thread = uc->config.remote_thread_id;
1816 		uc->config.dst_thread = (ud->psil_base + uc->rchan->id) |
1817 					K3_PSIL_DST_THREAD_ID_OFFSET;
1818 
1819 		irq_ring = uc->rflow->r_ring;
1820 		irq_udma_idx = match_data->rchan_oes_offset + uc->rchan->id;
1821 
1822 		ret = udma_tisci_rx_channel_config(uc);
1823 		break;
1824 	default:
1825 		/* Can not happen */
1826 		dev_err(uc->ud->dev, "%s: chan%d invalid direction (%u)\n",
1827 			__func__, uc->id, uc->config.dir);
1828 		ret = -EINVAL;
1829 		goto err_cleanup;
1830 
1831 	}
1832 
1833 	/* check if the channel configuration was successful */
1834 	if (ret)
1835 		goto err_res_free;
1836 
1837 	if (udma_is_chan_running(uc)) {
1838 		dev_warn(ud->dev, "chan%d: is running!\n", uc->id);
1839 		udma_reset_chan(uc, false);
1840 		if (udma_is_chan_running(uc)) {
1841 			dev_err(ud->dev, "chan%d: won't stop!\n", uc->id);
1842 			ret = -EBUSY;
1843 			goto err_res_free;
1844 		}
1845 	}
1846 
1847 	/* PSI-L pairing */
1848 	ret = navss_psil_pair(ud, uc->config.src_thread, uc->config.dst_thread);
1849 	if (ret) {
1850 		dev_err(ud->dev, "PSI-L pairing failed: 0x%04x -> 0x%04x\n",
1851 			uc->config.src_thread, uc->config.dst_thread);
1852 		goto err_res_free;
1853 	}
1854 
1855 	uc->psil_paired = true;
1856 
1857 	uc->irq_num_ring = k3_ringacc_get_ring_irq_num(irq_ring);
1858 	if (uc->irq_num_ring <= 0) {
1859 		dev_err(ud->dev, "Failed to get ring irq (index: %u)\n",
1860 			k3_ringacc_get_ring_id(irq_ring));
1861 		ret = -EINVAL;
1862 		goto err_psi_free;
1863 	}
1864 
1865 	ret = request_irq(uc->irq_num_ring, udma_ring_irq_handler,
1866 			  IRQF_TRIGGER_HIGH, uc->name, uc);
1867 	if (ret) {
1868 		dev_err(ud->dev, "chan%d: ring irq request failed\n", uc->id);
1869 		goto err_irq_free;
1870 	}
1871 
1872 	/* Event from UDMA (TR events) only needed for slave TR mode channels */
1873 	if (is_slave_direction(uc->config.dir) && !uc->config.pkt_mode) {
1874 		uc->irq_num_udma = ti_sci_inta_msi_get_virq(ud->dev,
1875 							    irq_udma_idx);
1876 		if (uc->irq_num_udma <= 0) {
1877 			dev_err(ud->dev, "Failed to get udma irq (index: %u)\n",
1878 				irq_udma_idx);
1879 			free_irq(uc->irq_num_ring, uc);
1880 			ret = -EINVAL;
1881 			goto err_irq_free;
1882 		}
1883 
1884 		ret = request_irq(uc->irq_num_udma, udma_udma_irq_handler, 0,
1885 				  uc->name, uc);
1886 		if (ret) {
1887 			dev_err(ud->dev, "chan%d: UDMA irq request failed\n",
1888 				uc->id);
1889 			free_irq(uc->irq_num_ring, uc);
1890 			goto err_irq_free;
1891 		}
1892 	} else {
1893 		uc->irq_num_udma = 0;
1894 	}
1895 
1896 	udma_reset_rings(uc);
1897 
1898 	return 0;
1899 
1900 err_irq_free:
1901 	uc->irq_num_ring = 0;
1902 	uc->irq_num_udma = 0;
1903 err_psi_free:
1904 	navss_psil_unpair(ud, uc->config.src_thread, uc->config.dst_thread);
1905 	uc->psil_paired = false;
1906 err_res_free:
1907 	udma_free_tx_resources(uc);
1908 	udma_free_rx_resources(uc);
1909 err_cleanup:
1910 	udma_reset_uchan(uc);
1911 
1912 	if (uc->use_dma_pool) {
1913 		dma_pool_destroy(uc->hdesc_pool);
1914 		uc->use_dma_pool = false;
1915 	}
1916 
1917 	return ret;
1918 }
1919 
1920 static int udma_slave_config(struct dma_chan *chan,
1921 			     struct dma_slave_config *cfg)
1922 {
1923 	struct udma_chan *uc = to_udma_chan(chan);
1924 
1925 	memcpy(&uc->cfg, cfg, sizeof(uc->cfg));
1926 
1927 	return 0;
1928 }
1929 
1930 static struct udma_desc *udma_alloc_tr_desc(struct udma_chan *uc,
1931 					    size_t tr_size, int tr_count,
1932 					    enum dma_transfer_direction dir)
1933 {
1934 	struct udma_hwdesc *hwdesc;
1935 	struct cppi5_desc_hdr_t *tr_desc;
1936 	struct udma_desc *d;
1937 	u32 reload_count = 0;
1938 	u32 ring_id;
1939 
1940 	switch (tr_size) {
1941 	case 16:
1942 	case 32:
1943 	case 64:
1944 	case 128:
1945 		break;
1946 	default:
1947 		dev_err(uc->ud->dev, "Unsupported TR size of %zu\n", tr_size);
1948 		return NULL;
1949 	}
1950 
1951 	/* We have only one descriptor containing multiple TRs */
1952 	d = kzalloc(sizeof(*d) + sizeof(d->hwdesc[0]), GFP_NOWAIT);
1953 	if (!d)
1954 		return NULL;
1955 
1956 	d->sglen = tr_count;
1957 
1958 	d->hwdesc_count = 1;
1959 	hwdesc = &d->hwdesc[0];
1960 
1961 	/* Allocate memory for DMA ring descriptor */
1962 	if (uc->use_dma_pool) {
1963 		hwdesc->cppi5_desc_size = uc->config.hdesc_size;
1964 		hwdesc->cppi5_desc_vaddr = dma_pool_zalloc(uc->hdesc_pool,
1965 						GFP_NOWAIT,
1966 						&hwdesc->cppi5_desc_paddr);
1967 	} else {
1968 		hwdesc->cppi5_desc_size = cppi5_trdesc_calc_size(tr_size,
1969 								 tr_count);
1970 		hwdesc->cppi5_desc_size = ALIGN(hwdesc->cppi5_desc_size,
1971 						uc->ud->desc_align);
1972 		hwdesc->cppi5_desc_vaddr = dma_alloc_coherent(uc->ud->dev,
1973 						hwdesc->cppi5_desc_size,
1974 						&hwdesc->cppi5_desc_paddr,
1975 						GFP_NOWAIT);
1976 	}
1977 
1978 	if (!hwdesc->cppi5_desc_vaddr) {
1979 		kfree(d);
1980 		return NULL;
1981 	}
1982 
1983 	/* Start of the TR req records */
1984 	hwdesc->tr_req_base = hwdesc->cppi5_desc_vaddr + tr_size;
1985 	/* Start address of the TR response array */
1986 	hwdesc->tr_resp_base = hwdesc->tr_req_base + tr_size * tr_count;
1987 
1988 	tr_desc = hwdesc->cppi5_desc_vaddr;
1989 
1990 	if (uc->cyclic)
1991 		reload_count = CPPI5_INFO0_TRDESC_RLDCNT_INFINITE;
1992 
1993 	if (dir == DMA_DEV_TO_MEM)
1994 		ring_id = k3_ringacc_get_ring_id(uc->rflow->r_ring);
1995 	else
1996 		ring_id = k3_ringacc_get_ring_id(uc->tchan->tc_ring);
1997 
1998 	cppi5_trdesc_init(tr_desc, tr_count, tr_size, 0, reload_count);
1999 	cppi5_desc_set_pktids(tr_desc, uc->id,
2000 			      CPPI5_INFO1_DESC_FLOWID_DEFAULT);
2001 	cppi5_desc_set_retpolicy(tr_desc, 0, ring_id);
2002 
2003 	return d;
2004 }
2005 
2006 /**
2007  * udma_get_tr_counters - calculate TR counters for a given length
2008  * @len: Length of the trasnfer
2009  * @align_to: Preferred alignment
2010  * @tr0_cnt0: First TR icnt0
2011  * @tr0_cnt1: First TR icnt1
2012  * @tr1_cnt0: Second (if used) TR icnt0
2013  *
2014  * For len < SZ_64K only one TR is enough, tr1_cnt0 is not updated
2015  * For len >= SZ_64K two TRs are used in a simple way:
2016  * First TR: SZ_64K-alignment blocks (tr0_cnt0, tr0_cnt1)
2017  * Second TR: the remaining length (tr1_cnt0)
2018  *
2019  * Returns the number of TRs the length needs (1 or 2)
2020  * -EINVAL if the length can not be supported
2021  */
2022 static int udma_get_tr_counters(size_t len, unsigned long align_to,
2023 				u16 *tr0_cnt0, u16 *tr0_cnt1, u16 *tr1_cnt0)
2024 {
2025 	if (len < SZ_64K) {
2026 		*tr0_cnt0 = len;
2027 		*tr0_cnt1 = 1;
2028 
2029 		return 1;
2030 	}
2031 
2032 	if (align_to > 3)
2033 		align_to = 3;
2034 
2035 realign:
2036 	*tr0_cnt0 = SZ_64K - BIT(align_to);
2037 	if (len / *tr0_cnt0 >= SZ_64K) {
2038 		if (align_to) {
2039 			align_to--;
2040 			goto realign;
2041 		}
2042 		return -EINVAL;
2043 	}
2044 
2045 	*tr0_cnt1 = len / *tr0_cnt0;
2046 	*tr1_cnt0 = len % *tr0_cnt0;
2047 
2048 	return 2;
2049 }
2050 
2051 static struct udma_desc *
2052 udma_prep_slave_sg_tr(struct udma_chan *uc, struct scatterlist *sgl,
2053 		      unsigned int sglen, enum dma_transfer_direction dir,
2054 		      unsigned long tx_flags, void *context)
2055 {
2056 	struct scatterlist *sgent;
2057 	struct udma_desc *d;
2058 	struct cppi5_tr_type1_t *tr_req = NULL;
2059 	u16 tr0_cnt0, tr0_cnt1, tr1_cnt0;
2060 	unsigned int i;
2061 	size_t tr_size;
2062 	int num_tr = 0;
2063 	int tr_idx = 0;
2064 
2065 	if (!is_slave_direction(dir)) {
2066 		dev_err(uc->ud->dev, "Only slave cyclic is supported\n");
2067 		return NULL;
2068 	}
2069 
2070 	/* estimate the number of TRs we will need */
2071 	for_each_sg(sgl, sgent, sglen, i) {
2072 		if (sg_dma_len(sgent) < SZ_64K)
2073 			num_tr++;
2074 		else
2075 			num_tr += 2;
2076 	}
2077 
2078 	/* Now allocate and setup the descriptor. */
2079 	tr_size = sizeof(struct cppi5_tr_type1_t);
2080 	d = udma_alloc_tr_desc(uc, tr_size, num_tr, dir);
2081 	if (!d)
2082 		return NULL;
2083 
2084 	d->sglen = sglen;
2085 
2086 	tr_req = d->hwdesc[0].tr_req_base;
2087 	for_each_sg(sgl, sgent, sglen, i) {
2088 		dma_addr_t sg_addr = sg_dma_address(sgent);
2089 
2090 		num_tr = udma_get_tr_counters(sg_dma_len(sgent), __ffs(sg_addr),
2091 					      &tr0_cnt0, &tr0_cnt1, &tr1_cnt0);
2092 		if (num_tr < 0) {
2093 			dev_err(uc->ud->dev, "size %u is not supported\n",
2094 				sg_dma_len(sgent));
2095 			udma_free_hwdesc(uc, d);
2096 			kfree(d);
2097 			return NULL;
2098 		}
2099 
2100 		cppi5_tr_init(&tr_req[i].flags, CPPI5_TR_TYPE1, false, false,
2101 			      CPPI5_TR_EVENT_SIZE_COMPLETION, 0);
2102 		cppi5_tr_csf_set(&tr_req[i].flags, CPPI5_TR_CSF_SUPR_EVT);
2103 
2104 		tr_req[tr_idx].addr = sg_addr;
2105 		tr_req[tr_idx].icnt0 = tr0_cnt0;
2106 		tr_req[tr_idx].icnt1 = tr0_cnt1;
2107 		tr_req[tr_idx].dim1 = tr0_cnt0;
2108 		tr_idx++;
2109 
2110 		if (num_tr == 2) {
2111 			cppi5_tr_init(&tr_req[tr_idx].flags, CPPI5_TR_TYPE1,
2112 				      false, false,
2113 				      CPPI5_TR_EVENT_SIZE_COMPLETION, 0);
2114 			cppi5_tr_csf_set(&tr_req[tr_idx].flags,
2115 					 CPPI5_TR_CSF_SUPR_EVT);
2116 
2117 			tr_req[tr_idx].addr = sg_addr + tr0_cnt1 * tr0_cnt0;
2118 			tr_req[tr_idx].icnt0 = tr1_cnt0;
2119 			tr_req[tr_idx].icnt1 = 1;
2120 			tr_req[tr_idx].dim1 = tr1_cnt0;
2121 			tr_idx++;
2122 		}
2123 
2124 		d->residue += sg_dma_len(sgent);
2125 	}
2126 
2127 	cppi5_tr_csf_set(&tr_req[tr_idx - 1].flags,
2128 			 CPPI5_TR_CSF_SUPR_EVT | CPPI5_TR_CSF_EOP);
2129 
2130 	return d;
2131 }
2132 
2133 static int udma_configure_statictr(struct udma_chan *uc, struct udma_desc *d,
2134 				   enum dma_slave_buswidth dev_width,
2135 				   u16 elcnt)
2136 {
2137 	if (uc->config.ep_type != PSIL_EP_PDMA_XY)
2138 		return 0;
2139 
2140 	/* Bus width translates to the element size (ES) */
2141 	switch (dev_width) {
2142 	case DMA_SLAVE_BUSWIDTH_1_BYTE:
2143 		d->static_tr.elsize = 0;
2144 		break;
2145 	case DMA_SLAVE_BUSWIDTH_2_BYTES:
2146 		d->static_tr.elsize = 1;
2147 		break;
2148 	case DMA_SLAVE_BUSWIDTH_3_BYTES:
2149 		d->static_tr.elsize = 2;
2150 		break;
2151 	case DMA_SLAVE_BUSWIDTH_4_BYTES:
2152 		d->static_tr.elsize = 3;
2153 		break;
2154 	case DMA_SLAVE_BUSWIDTH_8_BYTES:
2155 		d->static_tr.elsize = 4;
2156 		break;
2157 	default: /* not reached */
2158 		return -EINVAL;
2159 	}
2160 
2161 	d->static_tr.elcnt = elcnt;
2162 
2163 	/*
2164 	 * PDMA must to close the packet when the channel is in packet mode.
2165 	 * For TR mode when the channel is not cyclic we also need PDMA to close
2166 	 * the packet otherwise the transfer will stall because PDMA holds on
2167 	 * the data it has received from the peripheral.
2168 	 */
2169 	if (uc->config.pkt_mode || !uc->cyclic) {
2170 		unsigned int div = dev_width * elcnt;
2171 
2172 		if (uc->cyclic)
2173 			d->static_tr.bstcnt = d->residue / d->sglen / div;
2174 		else
2175 			d->static_tr.bstcnt = d->residue / div;
2176 
2177 		if (uc->config.dir == DMA_DEV_TO_MEM &&
2178 		    d->static_tr.bstcnt > uc->ud->match_data->statictr_z_mask)
2179 			return -EINVAL;
2180 	} else {
2181 		d->static_tr.bstcnt = 0;
2182 	}
2183 
2184 	return 0;
2185 }
2186 
2187 static struct udma_desc *
2188 udma_prep_slave_sg_pkt(struct udma_chan *uc, struct scatterlist *sgl,
2189 		       unsigned int sglen, enum dma_transfer_direction dir,
2190 		       unsigned long tx_flags, void *context)
2191 {
2192 	struct scatterlist *sgent;
2193 	struct cppi5_host_desc_t *h_desc = NULL;
2194 	struct udma_desc *d;
2195 	u32 ring_id;
2196 	unsigned int i;
2197 
2198 	d = kzalloc(sizeof(*d) + sglen * sizeof(d->hwdesc[0]), GFP_NOWAIT);
2199 	if (!d)
2200 		return NULL;
2201 
2202 	d->sglen = sglen;
2203 	d->hwdesc_count = sglen;
2204 
2205 	if (dir == DMA_DEV_TO_MEM)
2206 		ring_id = k3_ringacc_get_ring_id(uc->rflow->r_ring);
2207 	else
2208 		ring_id = k3_ringacc_get_ring_id(uc->tchan->tc_ring);
2209 
2210 	for_each_sg(sgl, sgent, sglen, i) {
2211 		struct udma_hwdesc *hwdesc = &d->hwdesc[i];
2212 		dma_addr_t sg_addr = sg_dma_address(sgent);
2213 		struct cppi5_host_desc_t *desc;
2214 		size_t sg_len = sg_dma_len(sgent);
2215 
2216 		hwdesc->cppi5_desc_vaddr = dma_pool_zalloc(uc->hdesc_pool,
2217 						GFP_NOWAIT,
2218 						&hwdesc->cppi5_desc_paddr);
2219 		if (!hwdesc->cppi5_desc_vaddr) {
2220 			dev_err(uc->ud->dev,
2221 				"descriptor%d allocation failed\n", i);
2222 
2223 			udma_free_hwdesc(uc, d);
2224 			kfree(d);
2225 			return NULL;
2226 		}
2227 
2228 		d->residue += sg_len;
2229 		hwdesc->cppi5_desc_size = uc->config.hdesc_size;
2230 		desc = hwdesc->cppi5_desc_vaddr;
2231 
2232 		if (i == 0) {
2233 			cppi5_hdesc_init(desc, 0, 0);
2234 			/* Flow and Packed ID */
2235 			cppi5_desc_set_pktids(&desc->hdr, uc->id,
2236 					      CPPI5_INFO1_DESC_FLOWID_DEFAULT);
2237 			cppi5_desc_set_retpolicy(&desc->hdr, 0, ring_id);
2238 		} else {
2239 			cppi5_hdesc_reset_hbdesc(desc);
2240 			cppi5_desc_set_retpolicy(&desc->hdr, 0, 0xffff);
2241 		}
2242 
2243 		/* attach the sg buffer to the descriptor */
2244 		cppi5_hdesc_attach_buf(desc, sg_addr, sg_len, sg_addr, sg_len);
2245 
2246 		/* Attach link as host buffer descriptor */
2247 		if (h_desc)
2248 			cppi5_hdesc_link_hbdesc(h_desc,
2249 						hwdesc->cppi5_desc_paddr);
2250 
2251 		if (dir == DMA_MEM_TO_DEV)
2252 			h_desc = desc;
2253 	}
2254 
2255 	if (d->residue >= SZ_4M) {
2256 		dev_err(uc->ud->dev,
2257 			"%s: Transfer size %u is over the supported 4M range\n",
2258 			__func__, d->residue);
2259 		udma_free_hwdesc(uc, d);
2260 		kfree(d);
2261 		return NULL;
2262 	}
2263 
2264 	h_desc = d->hwdesc[0].cppi5_desc_vaddr;
2265 	cppi5_hdesc_set_pktlen(h_desc, d->residue);
2266 
2267 	return d;
2268 }
2269 
2270 static int udma_attach_metadata(struct dma_async_tx_descriptor *desc,
2271 				void *data, size_t len)
2272 {
2273 	struct udma_desc *d = to_udma_desc(desc);
2274 	struct udma_chan *uc = to_udma_chan(desc->chan);
2275 	struct cppi5_host_desc_t *h_desc;
2276 	u32 psd_size = len;
2277 	u32 flags = 0;
2278 
2279 	if (!uc->config.pkt_mode || !uc->config.metadata_size)
2280 		return -ENOTSUPP;
2281 
2282 	if (!data || len > uc->config.metadata_size)
2283 		return -EINVAL;
2284 
2285 	if (uc->config.needs_epib && len < CPPI5_INFO0_HDESC_EPIB_SIZE)
2286 		return -EINVAL;
2287 
2288 	h_desc = d->hwdesc[0].cppi5_desc_vaddr;
2289 	if (d->dir == DMA_MEM_TO_DEV)
2290 		memcpy(h_desc->epib, data, len);
2291 
2292 	if (uc->config.needs_epib)
2293 		psd_size -= CPPI5_INFO0_HDESC_EPIB_SIZE;
2294 
2295 	d->metadata = data;
2296 	d->metadata_size = len;
2297 	if (uc->config.needs_epib)
2298 		flags |= CPPI5_INFO0_HDESC_EPIB_PRESENT;
2299 
2300 	cppi5_hdesc_update_flags(h_desc, flags);
2301 	cppi5_hdesc_update_psdata_size(h_desc, psd_size);
2302 
2303 	return 0;
2304 }
2305 
2306 static void *udma_get_metadata_ptr(struct dma_async_tx_descriptor *desc,
2307 				   size_t *payload_len, size_t *max_len)
2308 {
2309 	struct udma_desc *d = to_udma_desc(desc);
2310 	struct udma_chan *uc = to_udma_chan(desc->chan);
2311 	struct cppi5_host_desc_t *h_desc;
2312 
2313 	if (!uc->config.pkt_mode || !uc->config.metadata_size)
2314 		return ERR_PTR(-ENOTSUPP);
2315 
2316 	h_desc = d->hwdesc[0].cppi5_desc_vaddr;
2317 
2318 	*max_len = uc->config.metadata_size;
2319 
2320 	*payload_len = cppi5_hdesc_epib_present(&h_desc->hdr) ?
2321 		       CPPI5_INFO0_HDESC_EPIB_SIZE : 0;
2322 	*payload_len += cppi5_hdesc_get_psdata_size(h_desc);
2323 
2324 	return h_desc->epib;
2325 }
2326 
2327 static int udma_set_metadata_len(struct dma_async_tx_descriptor *desc,
2328 				 size_t payload_len)
2329 {
2330 	struct udma_desc *d = to_udma_desc(desc);
2331 	struct udma_chan *uc = to_udma_chan(desc->chan);
2332 	struct cppi5_host_desc_t *h_desc;
2333 	u32 psd_size = payload_len;
2334 	u32 flags = 0;
2335 
2336 	if (!uc->config.pkt_mode || !uc->config.metadata_size)
2337 		return -ENOTSUPP;
2338 
2339 	if (payload_len > uc->config.metadata_size)
2340 		return -EINVAL;
2341 
2342 	if (uc->config.needs_epib && payload_len < CPPI5_INFO0_HDESC_EPIB_SIZE)
2343 		return -EINVAL;
2344 
2345 	h_desc = d->hwdesc[0].cppi5_desc_vaddr;
2346 
2347 	if (uc->config.needs_epib) {
2348 		psd_size -= CPPI5_INFO0_HDESC_EPIB_SIZE;
2349 		flags |= CPPI5_INFO0_HDESC_EPIB_PRESENT;
2350 	}
2351 
2352 	cppi5_hdesc_update_flags(h_desc, flags);
2353 	cppi5_hdesc_update_psdata_size(h_desc, psd_size);
2354 
2355 	return 0;
2356 }
2357 
2358 static struct dma_descriptor_metadata_ops metadata_ops = {
2359 	.attach = udma_attach_metadata,
2360 	.get_ptr = udma_get_metadata_ptr,
2361 	.set_len = udma_set_metadata_len,
2362 };
2363 
2364 static struct dma_async_tx_descriptor *
2365 udma_prep_slave_sg(struct dma_chan *chan, struct scatterlist *sgl,
2366 		   unsigned int sglen, enum dma_transfer_direction dir,
2367 		   unsigned long tx_flags, void *context)
2368 {
2369 	struct udma_chan *uc = to_udma_chan(chan);
2370 	enum dma_slave_buswidth dev_width;
2371 	struct udma_desc *d;
2372 	u32 burst;
2373 
2374 	if (dir != uc->config.dir) {
2375 		dev_err(chan->device->dev,
2376 			"%s: chan%d is for %s, not supporting %s\n",
2377 			__func__, uc->id,
2378 			dmaengine_get_direction_text(uc->config.dir),
2379 			dmaengine_get_direction_text(dir));
2380 		return NULL;
2381 	}
2382 
2383 	if (dir == DMA_DEV_TO_MEM) {
2384 		dev_width = uc->cfg.src_addr_width;
2385 		burst = uc->cfg.src_maxburst;
2386 	} else if (dir == DMA_MEM_TO_DEV) {
2387 		dev_width = uc->cfg.dst_addr_width;
2388 		burst = uc->cfg.dst_maxburst;
2389 	} else {
2390 		dev_err(chan->device->dev, "%s: bad direction?\n", __func__);
2391 		return NULL;
2392 	}
2393 
2394 	if (!burst)
2395 		burst = 1;
2396 
2397 	if (uc->config.pkt_mode)
2398 		d = udma_prep_slave_sg_pkt(uc, sgl, sglen, dir, tx_flags,
2399 					   context);
2400 	else
2401 		d = udma_prep_slave_sg_tr(uc, sgl, sglen, dir, tx_flags,
2402 					  context);
2403 
2404 	if (!d)
2405 		return NULL;
2406 
2407 	d->dir = dir;
2408 	d->desc_idx = 0;
2409 	d->tr_idx = 0;
2410 
2411 	/* static TR for remote PDMA */
2412 	if (udma_configure_statictr(uc, d, dev_width, burst)) {
2413 		dev_err(uc->ud->dev,
2414 			"%s: StaticTR Z is limited to maximum 4095 (%u)\n",
2415 			__func__, d->static_tr.bstcnt);
2416 
2417 		udma_free_hwdesc(uc, d);
2418 		kfree(d);
2419 		return NULL;
2420 	}
2421 
2422 	if (uc->config.metadata_size)
2423 		d->vd.tx.metadata_ops = &metadata_ops;
2424 
2425 	return vchan_tx_prep(&uc->vc, &d->vd, tx_flags);
2426 }
2427 
2428 static struct udma_desc *
2429 udma_prep_dma_cyclic_tr(struct udma_chan *uc, dma_addr_t buf_addr,
2430 			size_t buf_len, size_t period_len,
2431 			enum dma_transfer_direction dir, unsigned long flags)
2432 {
2433 	struct udma_desc *d;
2434 	size_t tr_size, period_addr;
2435 	struct cppi5_tr_type1_t *tr_req;
2436 	unsigned int periods = buf_len / period_len;
2437 	u16 tr0_cnt0, tr0_cnt1, tr1_cnt0;
2438 	unsigned int i;
2439 	int num_tr;
2440 
2441 	if (!is_slave_direction(dir)) {
2442 		dev_err(uc->ud->dev, "Only slave cyclic is supported\n");
2443 		return NULL;
2444 	}
2445 
2446 	num_tr = udma_get_tr_counters(period_len, __ffs(buf_addr), &tr0_cnt0,
2447 				      &tr0_cnt1, &tr1_cnt0);
2448 	if (num_tr < 0) {
2449 		dev_err(uc->ud->dev, "size %zu is not supported\n",
2450 			period_len);
2451 		return NULL;
2452 	}
2453 
2454 	/* Now allocate and setup the descriptor. */
2455 	tr_size = sizeof(struct cppi5_tr_type1_t);
2456 	d = udma_alloc_tr_desc(uc, tr_size, periods * num_tr, dir);
2457 	if (!d)
2458 		return NULL;
2459 
2460 	tr_req = d->hwdesc[0].tr_req_base;
2461 	period_addr = buf_addr;
2462 	for (i = 0; i < periods; i++) {
2463 		int tr_idx = i * num_tr;
2464 
2465 		cppi5_tr_init(&tr_req[tr_idx].flags, CPPI5_TR_TYPE1, false,
2466 			      false, CPPI5_TR_EVENT_SIZE_COMPLETION, 0);
2467 
2468 		tr_req[tr_idx].addr = period_addr;
2469 		tr_req[tr_idx].icnt0 = tr0_cnt0;
2470 		tr_req[tr_idx].icnt1 = tr0_cnt1;
2471 		tr_req[tr_idx].dim1 = tr0_cnt0;
2472 
2473 		if (num_tr == 2) {
2474 			cppi5_tr_csf_set(&tr_req[tr_idx].flags,
2475 					 CPPI5_TR_CSF_SUPR_EVT);
2476 			tr_idx++;
2477 
2478 			cppi5_tr_init(&tr_req[tr_idx].flags, CPPI5_TR_TYPE1,
2479 				      false, false,
2480 				      CPPI5_TR_EVENT_SIZE_COMPLETION, 0);
2481 
2482 			tr_req[tr_idx].addr = period_addr + tr0_cnt1 * tr0_cnt0;
2483 			tr_req[tr_idx].icnt0 = tr1_cnt0;
2484 			tr_req[tr_idx].icnt1 = 1;
2485 			tr_req[tr_idx].dim1 = tr1_cnt0;
2486 		}
2487 
2488 		if (!(flags & DMA_PREP_INTERRUPT))
2489 			cppi5_tr_csf_set(&tr_req[tr_idx].flags,
2490 					 CPPI5_TR_CSF_SUPR_EVT);
2491 
2492 		period_addr += period_len;
2493 	}
2494 
2495 	return d;
2496 }
2497 
2498 static struct udma_desc *
2499 udma_prep_dma_cyclic_pkt(struct udma_chan *uc, dma_addr_t buf_addr,
2500 			 size_t buf_len, size_t period_len,
2501 			 enum dma_transfer_direction dir, unsigned long flags)
2502 {
2503 	struct udma_desc *d;
2504 	u32 ring_id;
2505 	int i;
2506 	int periods = buf_len / period_len;
2507 
2508 	if (periods > (K3_UDMA_DEFAULT_RING_SIZE - 1))
2509 		return NULL;
2510 
2511 	if (period_len >= SZ_4M)
2512 		return NULL;
2513 
2514 	d = kzalloc(sizeof(*d) + periods * sizeof(d->hwdesc[0]), GFP_NOWAIT);
2515 	if (!d)
2516 		return NULL;
2517 
2518 	d->hwdesc_count = periods;
2519 
2520 	/* TODO: re-check this... */
2521 	if (dir == DMA_DEV_TO_MEM)
2522 		ring_id = k3_ringacc_get_ring_id(uc->rflow->r_ring);
2523 	else
2524 		ring_id = k3_ringacc_get_ring_id(uc->tchan->tc_ring);
2525 
2526 	for (i = 0; i < periods; i++) {
2527 		struct udma_hwdesc *hwdesc = &d->hwdesc[i];
2528 		dma_addr_t period_addr = buf_addr + (period_len * i);
2529 		struct cppi5_host_desc_t *h_desc;
2530 
2531 		hwdesc->cppi5_desc_vaddr = dma_pool_zalloc(uc->hdesc_pool,
2532 						GFP_NOWAIT,
2533 						&hwdesc->cppi5_desc_paddr);
2534 		if (!hwdesc->cppi5_desc_vaddr) {
2535 			dev_err(uc->ud->dev,
2536 				"descriptor%d allocation failed\n", i);
2537 
2538 			udma_free_hwdesc(uc, d);
2539 			kfree(d);
2540 			return NULL;
2541 		}
2542 
2543 		hwdesc->cppi5_desc_size = uc->config.hdesc_size;
2544 		h_desc = hwdesc->cppi5_desc_vaddr;
2545 
2546 		cppi5_hdesc_init(h_desc, 0, 0);
2547 		cppi5_hdesc_set_pktlen(h_desc, period_len);
2548 
2549 		/* Flow and Packed ID */
2550 		cppi5_desc_set_pktids(&h_desc->hdr, uc->id,
2551 				      CPPI5_INFO1_DESC_FLOWID_DEFAULT);
2552 		cppi5_desc_set_retpolicy(&h_desc->hdr, 0, ring_id);
2553 
2554 		/* attach each period to a new descriptor */
2555 		cppi5_hdesc_attach_buf(h_desc,
2556 				       period_addr, period_len,
2557 				       period_addr, period_len);
2558 	}
2559 
2560 	return d;
2561 }
2562 
2563 static struct dma_async_tx_descriptor *
2564 udma_prep_dma_cyclic(struct dma_chan *chan, dma_addr_t buf_addr, size_t buf_len,
2565 		     size_t period_len, enum dma_transfer_direction dir,
2566 		     unsigned long flags)
2567 {
2568 	struct udma_chan *uc = to_udma_chan(chan);
2569 	enum dma_slave_buswidth dev_width;
2570 	struct udma_desc *d;
2571 	u32 burst;
2572 
2573 	if (dir != uc->config.dir) {
2574 		dev_err(chan->device->dev,
2575 			"%s: chan%d is for %s, not supporting %s\n",
2576 			__func__, uc->id,
2577 			dmaengine_get_direction_text(uc->config.dir),
2578 			dmaengine_get_direction_text(dir));
2579 		return NULL;
2580 	}
2581 
2582 	uc->cyclic = true;
2583 
2584 	if (dir == DMA_DEV_TO_MEM) {
2585 		dev_width = uc->cfg.src_addr_width;
2586 		burst = uc->cfg.src_maxburst;
2587 	} else if (dir == DMA_MEM_TO_DEV) {
2588 		dev_width = uc->cfg.dst_addr_width;
2589 		burst = uc->cfg.dst_maxburst;
2590 	} else {
2591 		dev_err(uc->ud->dev, "%s: bad direction?\n", __func__);
2592 		return NULL;
2593 	}
2594 
2595 	if (!burst)
2596 		burst = 1;
2597 
2598 	if (uc->config.pkt_mode)
2599 		d = udma_prep_dma_cyclic_pkt(uc, buf_addr, buf_len, period_len,
2600 					     dir, flags);
2601 	else
2602 		d = udma_prep_dma_cyclic_tr(uc, buf_addr, buf_len, period_len,
2603 					    dir, flags);
2604 
2605 	if (!d)
2606 		return NULL;
2607 
2608 	d->sglen = buf_len / period_len;
2609 
2610 	d->dir = dir;
2611 	d->residue = buf_len;
2612 
2613 	/* static TR for remote PDMA */
2614 	if (udma_configure_statictr(uc, d, dev_width, burst)) {
2615 		dev_err(uc->ud->dev,
2616 			"%s: StaticTR Z is limited to maximum 4095 (%u)\n",
2617 			__func__, d->static_tr.bstcnt);
2618 
2619 		udma_free_hwdesc(uc, d);
2620 		kfree(d);
2621 		return NULL;
2622 	}
2623 
2624 	if (uc->config.metadata_size)
2625 		d->vd.tx.metadata_ops = &metadata_ops;
2626 
2627 	return vchan_tx_prep(&uc->vc, &d->vd, flags);
2628 }
2629 
2630 static struct dma_async_tx_descriptor *
2631 udma_prep_dma_memcpy(struct dma_chan *chan, dma_addr_t dest, dma_addr_t src,
2632 		     size_t len, unsigned long tx_flags)
2633 {
2634 	struct udma_chan *uc = to_udma_chan(chan);
2635 	struct udma_desc *d;
2636 	struct cppi5_tr_type15_t *tr_req;
2637 	int num_tr;
2638 	size_t tr_size = sizeof(struct cppi5_tr_type15_t);
2639 	u16 tr0_cnt0, tr0_cnt1, tr1_cnt0;
2640 
2641 	if (uc->config.dir != DMA_MEM_TO_MEM) {
2642 		dev_err(chan->device->dev,
2643 			"%s: chan%d is for %s, not supporting %s\n",
2644 			__func__, uc->id,
2645 			dmaengine_get_direction_text(uc->config.dir),
2646 			dmaengine_get_direction_text(DMA_MEM_TO_MEM));
2647 		return NULL;
2648 	}
2649 
2650 	num_tr = udma_get_tr_counters(len, __ffs(src | dest), &tr0_cnt0,
2651 				      &tr0_cnt1, &tr1_cnt0);
2652 	if (num_tr < 0) {
2653 		dev_err(uc->ud->dev, "size %zu is not supported\n",
2654 			len);
2655 		return NULL;
2656 	}
2657 
2658 	d = udma_alloc_tr_desc(uc, tr_size, num_tr, DMA_MEM_TO_MEM);
2659 	if (!d)
2660 		return NULL;
2661 
2662 	d->dir = DMA_MEM_TO_MEM;
2663 	d->desc_idx = 0;
2664 	d->tr_idx = 0;
2665 	d->residue = len;
2666 
2667 	tr_req = d->hwdesc[0].tr_req_base;
2668 
2669 	cppi5_tr_init(&tr_req[0].flags, CPPI5_TR_TYPE15, false, true,
2670 		      CPPI5_TR_EVENT_SIZE_COMPLETION, 0);
2671 	cppi5_tr_csf_set(&tr_req[0].flags, CPPI5_TR_CSF_SUPR_EVT);
2672 
2673 	tr_req[0].addr = src;
2674 	tr_req[0].icnt0 = tr0_cnt0;
2675 	tr_req[0].icnt1 = tr0_cnt1;
2676 	tr_req[0].icnt2 = 1;
2677 	tr_req[0].icnt3 = 1;
2678 	tr_req[0].dim1 = tr0_cnt0;
2679 
2680 	tr_req[0].daddr = dest;
2681 	tr_req[0].dicnt0 = tr0_cnt0;
2682 	tr_req[0].dicnt1 = tr0_cnt1;
2683 	tr_req[0].dicnt2 = 1;
2684 	tr_req[0].dicnt3 = 1;
2685 	tr_req[0].ddim1 = tr0_cnt0;
2686 
2687 	if (num_tr == 2) {
2688 		cppi5_tr_init(&tr_req[1].flags, CPPI5_TR_TYPE15, false, true,
2689 			      CPPI5_TR_EVENT_SIZE_COMPLETION, 0);
2690 		cppi5_tr_csf_set(&tr_req[1].flags, CPPI5_TR_CSF_SUPR_EVT);
2691 
2692 		tr_req[1].addr = src + tr0_cnt1 * tr0_cnt0;
2693 		tr_req[1].icnt0 = tr1_cnt0;
2694 		tr_req[1].icnt1 = 1;
2695 		tr_req[1].icnt2 = 1;
2696 		tr_req[1].icnt3 = 1;
2697 
2698 		tr_req[1].daddr = dest + tr0_cnt1 * tr0_cnt0;
2699 		tr_req[1].dicnt0 = tr1_cnt0;
2700 		tr_req[1].dicnt1 = 1;
2701 		tr_req[1].dicnt2 = 1;
2702 		tr_req[1].dicnt3 = 1;
2703 	}
2704 
2705 	cppi5_tr_csf_set(&tr_req[num_tr - 1].flags,
2706 			 CPPI5_TR_CSF_SUPR_EVT | CPPI5_TR_CSF_EOP);
2707 
2708 	if (uc->config.metadata_size)
2709 		d->vd.tx.metadata_ops = &metadata_ops;
2710 
2711 	return vchan_tx_prep(&uc->vc, &d->vd, tx_flags);
2712 }
2713 
2714 static void udma_issue_pending(struct dma_chan *chan)
2715 {
2716 	struct udma_chan *uc = to_udma_chan(chan);
2717 	unsigned long flags;
2718 
2719 	spin_lock_irqsave(&uc->vc.lock, flags);
2720 
2721 	/* If we have something pending and no active descriptor, then */
2722 	if (vchan_issue_pending(&uc->vc) && !uc->desc) {
2723 		/*
2724 		 * start a descriptor if the channel is NOT [marked as
2725 		 * terminating _and_ it is still running (teardown has not
2726 		 * completed yet)].
2727 		 */
2728 		if (!(uc->state == UDMA_CHAN_IS_TERMINATING &&
2729 		      udma_is_chan_running(uc)))
2730 			udma_start(uc);
2731 	}
2732 
2733 	spin_unlock_irqrestore(&uc->vc.lock, flags);
2734 }
2735 
2736 static enum dma_status udma_tx_status(struct dma_chan *chan,
2737 				      dma_cookie_t cookie,
2738 				      struct dma_tx_state *txstate)
2739 {
2740 	struct udma_chan *uc = to_udma_chan(chan);
2741 	enum dma_status ret;
2742 	unsigned long flags;
2743 
2744 	spin_lock_irqsave(&uc->vc.lock, flags);
2745 
2746 	ret = dma_cookie_status(chan, cookie, txstate);
2747 
2748 	if (!udma_is_chan_running(uc))
2749 		ret = DMA_COMPLETE;
2750 
2751 	if (ret == DMA_IN_PROGRESS && udma_is_chan_paused(uc))
2752 		ret = DMA_PAUSED;
2753 
2754 	if (ret == DMA_COMPLETE || !txstate)
2755 		goto out;
2756 
2757 	if (uc->desc && uc->desc->vd.tx.cookie == cookie) {
2758 		u32 peer_bcnt = 0;
2759 		u32 bcnt = 0;
2760 		u32 residue = uc->desc->residue;
2761 		u32 delay = 0;
2762 
2763 		if (uc->desc->dir == DMA_MEM_TO_DEV) {
2764 			bcnt = udma_tchanrt_read(uc->tchan,
2765 						 UDMA_TCHAN_RT_SBCNT_REG);
2766 
2767 			if (uc->config.ep_type != PSIL_EP_NATIVE) {
2768 				peer_bcnt = udma_tchanrt_read(uc->tchan,
2769 						UDMA_TCHAN_RT_PEER_BCNT_REG);
2770 
2771 				if (bcnt > peer_bcnt)
2772 					delay = bcnt - peer_bcnt;
2773 			}
2774 		} else if (uc->desc->dir == DMA_DEV_TO_MEM) {
2775 			bcnt = udma_rchanrt_read(uc->rchan,
2776 						 UDMA_RCHAN_RT_BCNT_REG);
2777 
2778 			if (uc->config.ep_type != PSIL_EP_NATIVE) {
2779 				peer_bcnt = udma_rchanrt_read(uc->rchan,
2780 						UDMA_RCHAN_RT_PEER_BCNT_REG);
2781 
2782 				if (peer_bcnt > bcnt)
2783 					delay = peer_bcnt - bcnt;
2784 			}
2785 		} else {
2786 			bcnt = udma_tchanrt_read(uc->tchan,
2787 						 UDMA_TCHAN_RT_BCNT_REG);
2788 		}
2789 
2790 		bcnt -= uc->bcnt;
2791 		if (bcnt && !(bcnt % uc->desc->residue))
2792 			residue = 0;
2793 		else
2794 			residue -= bcnt % uc->desc->residue;
2795 
2796 		if (!residue && (uc->config.dir == DMA_DEV_TO_MEM || !delay)) {
2797 			ret = DMA_COMPLETE;
2798 			delay = 0;
2799 		}
2800 
2801 		dma_set_residue(txstate, residue);
2802 		dma_set_in_flight_bytes(txstate, delay);
2803 
2804 	} else {
2805 		ret = DMA_COMPLETE;
2806 	}
2807 
2808 out:
2809 	spin_unlock_irqrestore(&uc->vc.lock, flags);
2810 	return ret;
2811 }
2812 
2813 static int udma_pause(struct dma_chan *chan)
2814 {
2815 	struct udma_chan *uc = to_udma_chan(chan);
2816 
2817 	/* pause the channel */
2818 	switch (uc->config.dir) {
2819 	case DMA_DEV_TO_MEM:
2820 		udma_rchanrt_update_bits(uc->rchan,
2821 					 UDMA_RCHAN_RT_PEER_RT_EN_REG,
2822 					 UDMA_PEER_RT_EN_PAUSE,
2823 					 UDMA_PEER_RT_EN_PAUSE);
2824 		break;
2825 	case DMA_MEM_TO_DEV:
2826 		udma_tchanrt_update_bits(uc->tchan,
2827 					 UDMA_TCHAN_RT_PEER_RT_EN_REG,
2828 					 UDMA_PEER_RT_EN_PAUSE,
2829 					 UDMA_PEER_RT_EN_PAUSE);
2830 		break;
2831 	case DMA_MEM_TO_MEM:
2832 		udma_tchanrt_update_bits(uc->tchan, UDMA_TCHAN_RT_CTL_REG,
2833 					 UDMA_CHAN_RT_CTL_PAUSE,
2834 					 UDMA_CHAN_RT_CTL_PAUSE);
2835 		break;
2836 	default:
2837 		return -EINVAL;
2838 	}
2839 
2840 	return 0;
2841 }
2842 
2843 static int udma_resume(struct dma_chan *chan)
2844 {
2845 	struct udma_chan *uc = to_udma_chan(chan);
2846 
2847 	/* resume the channel */
2848 	switch (uc->config.dir) {
2849 	case DMA_DEV_TO_MEM:
2850 		udma_rchanrt_update_bits(uc->rchan,
2851 					 UDMA_RCHAN_RT_PEER_RT_EN_REG,
2852 					 UDMA_PEER_RT_EN_PAUSE, 0);
2853 
2854 		break;
2855 	case DMA_MEM_TO_DEV:
2856 		udma_tchanrt_update_bits(uc->tchan,
2857 					 UDMA_TCHAN_RT_PEER_RT_EN_REG,
2858 					 UDMA_PEER_RT_EN_PAUSE, 0);
2859 		break;
2860 	case DMA_MEM_TO_MEM:
2861 		udma_tchanrt_update_bits(uc->tchan, UDMA_TCHAN_RT_CTL_REG,
2862 					 UDMA_CHAN_RT_CTL_PAUSE, 0);
2863 		break;
2864 	default:
2865 		return -EINVAL;
2866 	}
2867 
2868 	return 0;
2869 }
2870 
2871 static int udma_terminate_all(struct dma_chan *chan)
2872 {
2873 	struct udma_chan *uc = to_udma_chan(chan);
2874 	unsigned long flags;
2875 	LIST_HEAD(head);
2876 
2877 	spin_lock_irqsave(&uc->vc.lock, flags);
2878 
2879 	if (udma_is_chan_running(uc))
2880 		udma_stop(uc);
2881 
2882 	if (uc->desc) {
2883 		uc->terminated_desc = uc->desc;
2884 		uc->desc = NULL;
2885 		uc->terminated_desc->terminated = true;
2886 		cancel_delayed_work(&uc->tx_drain.work);
2887 	}
2888 
2889 	uc->paused = false;
2890 
2891 	vchan_get_all_descriptors(&uc->vc, &head);
2892 	spin_unlock_irqrestore(&uc->vc.lock, flags);
2893 	vchan_dma_desc_free_list(&uc->vc, &head);
2894 
2895 	return 0;
2896 }
2897 
2898 static void udma_synchronize(struct dma_chan *chan)
2899 {
2900 	struct udma_chan *uc = to_udma_chan(chan);
2901 	unsigned long timeout = msecs_to_jiffies(1000);
2902 
2903 	vchan_synchronize(&uc->vc);
2904 
2905 	if (uc->state == UDMA_CHAN_IS_TERMINATING) {
2906 		timeout = wait_for_completion_timeout(&uc->teardown_completed,
2907 						      timeout);
2908 		if (!timeout) {
2909 			dev_warn(uc->ud->dev, "chan%d teardown timeout!\n",
2910 				 uc->id);
2911 			udma_dump_chan_stdata(uc);
2912 			udma_reset_chan(uc, true);
2913 		}
2914 	}
2915 
2916 	udma_reset_chan(uc, false);
2917 	if (udma_is_chan_running(uc))
2918 		dev_warn(uc->ud->dev, "chan%d refused to stop!\n", uc->id);
2919 
2920 	cancel_delayed_work_sync(&uc->tx_drain.work);
2921 	udma_reset_rings(uc);
2922 }
2923 
2924 static void udma_desc_pre_callback(struct virt_dma_chan *vc,
2925 				   struct virt_dma_desc *vd,
2926 				   struct dmaengine_result *result)
2927 {
2928 	struct udma_chan *uc = to_udma_chan(&vc->chan);
2929 	struct udma_desc *d;
2930 
2931 	if (!vd)
2932 		return;
2933 
2934 	d = to_udma_desc(&vd->tx);
2935 
2936 	if (d->metadata_size)
2937 		udma_fetch_epib(uc, d);
2938 
2939 	/* Provide residue information for the client */
2940 	if (result) {
2941 		void *desc_vaddr = udma_curr_cppi5_desc_vaddr(d, d->desc_idx);
2942 
2943 		if (cppi5_desc_get_type(desc_vaddr) ==
2944 		    CPPI5_INFO0_DESC_TYPE_VAL_HOST) {
2945 			result->residue = d->residue -
2946 					  cppi5_hdesc_get_pktlen(desc_vaddr);
2947 			if (result->residue)
2948 				result->result = DMA_TRANS_ABORTED;
2949 			else
2950 				result->result = DMA_TRANS_NOERROR;
2951 		} else {
2952 			result->residue = 0;
2953 			result->result = DMA_TRANS_NOERROR;
2954 		}
2955 	}
2956 }
2957 
2958 /*
2959  * This tasklet handles the completion of a DMA descriptor by
2960  * calling its callback and freeing it.
2961  */
2962 static void udma_vchan_complete(unsigned long arg)
2963 {
2964 	struct virt_dma_chan *vc = (struct virt_dma_chan *)arg;
2965 	struct virt_dma_desc *vd, *_vd;
2966 	struct dmaengine_desc_callback cb;
2967 	LIST_HEAD(head);
2968 
2969 	spin_lock_irq(&vc->lock);
2970 	list_splice_tail_init(&vc->desc_completed, &head);
2971 	vd = vc->cyclic;
2972 	if (vd) {
2973 		vc->cyclic = NULL;
2974 		dmaengine_desc_get_callback(&vd->tx, &cb);
2975 	} else {
2976 		memset(&cb, 0, sizeof(cb));
2977 	}
2978 	spin_unlock_irq(&vc->lock);
2979 
2980 	udma_desc_pre_callback(vc, vd, NULL);
2981 	dmaengine_desc_callback_invoke(&cb, NULL);
2982 
2983 	list_for_each_entry_safe(vd, _vd, &head, node) {
2984 		struct dmaengine_result result;
2985 
2986 		dmaengine_desc_get_callback(&vd->tx, &cb);
2987 
2988 		list_del(&vd->node);
2989 
2990 		udma_desc_pre_callback(vc, vd, &result);
2991 		dmaengine_desc_callback_invoke(&cb, &result);
2992 
2993 		vchan_vdesc_fini(vd);
2994 	}
2995 }
2996 
2997 static void udma_free_chan_resources(struct dma_chan *chan)
2998 {
2999 	struct udma_chan *uc = to_udma_chan(chan);
3000 	struct udma_dev *ud = to_udma_dev(chan->device);
3001 
3002 	udma_terminate_all(chan);
3003 	if (uc->terminated_desc) {
3004 		udma_reset_chan(uc, false);
3005 		udma_reset_rings(uc);
3006 	}
3007 
3008 	cancel_delayed_work_sync(&uc->tx_drain.work);
3009 
3010 	if (uc->irq_num_ring > 0) {
3011 		free_irq(uc->irq_num_ring, uc);
3012 
3013 		uc->irq_num_ring = 0;
3014 	}
3015 	if (uc->irq_num_udma > 0) {
3016 		free_irq(uc->irq_num_udma, uc);
3017 
3018 		uc->irq_num_udma = 0;
3019 	}
3020 
3021 	/* Release PSI-L pairing */
3022 	if (uc->psil_paired) {
3023 		navss_psil_unpair(ud, uc->config.src_thread,
3024 				  uc->config.dst_thread);
3025 		uc->psil_paired = false;
3026 	}
3027 
3028 	vchan_free_chan_resources(&uc->vc);
3029 	tasklet_kill(&uc->vc.task);
3030 
3031 	udma_free_tx_resources(uc);
3032 	udma_free_rx_resources(uc);
3033 	udma_reset_uchan(uc);
3034 
3035 	if (uc->use_dma_pool) {
3036 		dma_pool_destroy(uc->hdesc_pool);
3037 		uc->use_dma_pool = false;
3038 	}
3039 }
3040 
3041 static struct platform_driver udma_driver;
3042 
3043 struct udma_filter_param {
3044 	int remote_thread_id;
3045 	u32 atype;
3046 };
3047 
3048 static bool udma_dma_filter_fn(struct dma_chan *chan, void *param)
3049 {
3050 	struct udma_chan_config *ucc;
3051 	struct psil_endpoint_config *ep_config;
3052 	struct udma_filter_param *filter_param;
3053 	struct udma_chan *uc;
3054 	struct udma_dev *ud;
3055 
3056 	if (chan->device->dev->driver != &udma_driver.driver)
3057 		return false;
3058 
3059 	uc = to_udma_chan(chan);
3060 	ucc = &uc->config;
3061 	ud = uc->ud;
3062 	filter_param = param;
3063 
3064 	if (filter_param->atype > 2) {
3065 		dev_err(ud->dev, "Invalid channel atype: %u\n",
3066 			filter_param->atype);
3067 		return false;
3068 	}
3069 
3070 	ucc->remote_thread_id = filter_param->remote_thread_id;
3071 	ucc->atype = filter_param->atype;
3072 
3073 	if (ucc->remote_thread_id & K3_PSIL_DST_THREAD_ID_OFFSET)
3074 		ucc->dir = DMA_MEM_TO_DEV;
3075 	else
3076 		ucc->dir = DMA_DEV_TO_MEM;
3077 
3078 	ep_config = psil_get_ep_config(ucc->remote_thread_id);
3079 	if (IS_ERR(ep_config)) {
3080 		dev_err(ud->dev, "No configuration for psi-l thread 0x%04x\n",
3081 			ucc->remote_thread_id);
3082 		ucc->dir = DMA_MEM_TO_MEM;
3083 		ucc->remote_thread_id = -1;
3084 		ucc->atype = 0;
3085 		return false;
3086 	}
3087 
3088 	ucc->pkt_mode = ep_config->pkt_mode;
3089 	ucc->channel_tpl = ep_config->channel_tpl;
3090 	ucc->notdpkt = ep_config->notdpkt;
3091 	ucc->ep_type = ep_config->ep_type;
3092 
3093 	if (ucc->ep_type != PSIL_EP_NATIVE) {
3094 		const struct udma_match_data *match_data = ud->match_data;
3095 
3096 		if (match_data->flags & UDMA_FLAG_PDMA_ACC32)
3097 			ucc->enable_acc32 = ep_config->pdma_acc32;
3098 		if (match_data->flags & UDMA_FLAG_PDMA_BURST)
3099 			ucc->enable_burst = ep_config->pdma_burst;
3100 	}
3101 
3102 	ucc->needs_epib = ep_config->needs_epib;
3103 	ucc->psd_size = ep_config->psd_size;
3104 	ucc->metadata_size =
3105 			(ucc->needs_epib ? CPPI5_INFO0_HDESC_EPIB_SIZE : 0) +
3106 			ucc->psd_size;
3107 
3108 	if (ucc->pkt_mode)
3109 		ucc->hdesc_size = ALIGN(sizeof(struct cppi5_host_desc_t) +
3110 				 ucc->metadata_size, ud->desc_align);
3111 
3112 	dev_dbg(ud->dev, "chan%d: Remote thread: 0x%04x (%s)\n", uc->id,
3113 		ucc->remote_thread_id, dmaengine_get_direction_text(ucc->dir));
3114 
3115 	return true;
3116 }
3117 
3118 static struct dma_chan *udma_of_xlate(struct of_phandle_args *dma_spec,
3119 				      struct of_dma *ofdma)
3120 {
3121 	struct udma_dev *ud = ofdma->of_dma_data;
3122 	dma_cap_mask_t mask = ud->ddev.cap_mask;
3123 	struct udma_filter_param filter_param;
3124 	struct dma_chan *chan;
3125 
3126 	if (dma_spec->args_count != 1 && dma_spec->args_count != 2)
3127 		return NULL;
3128 
3129 	filter_param.remote_thread_id = dma_spec->args[0];
3130 	if (dma_spec->args_count == 2)
3131 		filter_param.atype = dma_spec->args[1];
3132 	else
3133 		filter_param.atype = 0;
3134 
3135 	chan = __dma_request_channel(&mask, udma_dma_filter_fn, &filter_param,
3136 				     ofdma->of_node);
3137 	if (!chan) {
3138 		dev_err(ud->dev, "get channel fail in %s.\n", __func__);
3139 		return ERR_PTR(-EINVAL);
3140 	}
3141 
3142 	return chan;
3143 }
3144 
3145 static struct udma_match_data am654_main_data = {
3146 	.psil_base = 0x1000,
3147 	.enable_memcpy_support = true,
3148 	.statictr_z_mask = GENMASK(11, 0),
3149 	.rchan_oes_offset = 0x2000,
3150 	.tpl_levels = 2,
3151 	.level_start_idx = {
3152 		[0] = 8, /* Normal channels */
3153 		[1] = 0, /* High Throughput channels */
3154 	},
3155 };
3156 
3157 static struct udma_match_data am654_mcu_data = {
3158 	.psil_base = 0x6000,
3159 	.enable_memcpy_support = false,
3160 	.statictr_z_mask = GENMASK(11, 0),
3161 	.rchan_oes_offset = 0x2000,
3162 	.tpl_levels = 2,
3163 	.level_start_idx = {
3164 		[0] = 2, /* Normal channels */
3165 		[1] = 0, /* High Throughput channels */
3166 	},
3167 };
3168 
3169 static struct udma_match_data j721e_main_data = {
3170 	.psil_base = 0x1000,
3171 	.enable_memcpy_support = true,
3172 	.flags = UDMA_FLAG_PDMA_ACC32 | UDMA_FLAG_PDMA_BURST,
3173 	.statictr_z_mask = GENMASK(23, 0),
3174 	.rchan_oes_offset = 0x400,
3175 	.tpl_levels = 3,
3176 	.level_start_idx = {
3177 		[0] = 16, /* Normal channels */
3178 		[1] = 4, /* High Throughput channels */
3179 		[2] = 0, /* Ultra High Throughput channels */
3180 	},
3181 };
3182 
3183 static struct udma_match_data j721e_mcu_data = {
3184 	.psil_base = 0x6000,
3185 	.enable_memcpy_support = false, /* MEM_TO_MEM is slow via MCU UDMA */
3186 	.flags = UDMA_FLAG_PDMA_ACC32 | UDMA_FLAG_PDMA_BURST,
3187 	.statictr_z_mask = GENMASK(23, 0),
3188 	.rchan_oes_offset = 0x400,
3189 	.tpl_levels = 2,
3190 	.level_start_idx = {
3191 		[0] = 2, /* Normal channels */
3192 		[1] = 0, /* High Throughput channels */
3193 	},
3194 };
3195 
3196 static const struct of_device_id udma_of_match[] = {
3197 	{
3198 		.compatible = "ti,am654-navss-main-udmap",
3199 		.data = &am654_main_data,
3200 	},
3201 	{
3202 		.compatible = "ti,am654-navss-mcu-udmap",
3203 		.data = &am654_mcu_data,
3204 	}, {
3205 		.compatible = "ti,j721e-navss-main-udmap",
3206 		.data = &j721e_main_data,
3207 	}, {
3208 		.compatible = "ti,j721e-navss-mcu-udmap",
3209 		.data = &j721e_mcu_data,
3210 	},
3211 	{ /* Sentinel */ },
3212 };
3213 
3214 static int udma_get_mmrs(struct platform_device *pdev, struct udma_dev *ud)
3215 {
3216 	struct resource *res;
3217 	int i;
3218 
3219 	for (i = 0; i < MMR_LAST; i++) {
3220 		res = platform_get_resource_byname(pdev, IORESOURCE_MEM,
3221 						   mmr_names[i]);
3222 		ud->mmrs[i] = devm_ioremap_resource(&pdev->dev, res);
3223 		if (IS_ERR(ud->mmrs[i]))
3224 			return PTR_ERR(ud->mmrs[i]);
3225 	}
3226 
3227 	return 0;
3228 }
3229 
3230 static int udma_setup_resources(struct udma_dev *ud)
3231 {
3232 	struct device *dev = ud->dev;
3233 	int ch_count, ret, i, j;
3234 	u32 cap2, cap3;
3235 	struct ti_sci_resource_desc *rm_desc;
3236 	struct ti_sci_resource *rm_res, irq_res;
3237 	struct udma_tisci_rm *tisci_rm = &ud->tisci_rm;
3238 	static const char * const range_names[] = { "ti,sci-rm-range-tchan",
3239 						    "ti,sci-rm-range-rchan",
3240 						    "ti,sci-rm-range-rflow" };
3241 
3242 	cap2 = udma_read(ud->mmrs[MMR_GCFG], 0x28);
3243 	cap3 = udma_read(ud->mmrs[MMR_GCFG], 0x2c);
3244 
3245 	ud->rflow_cnt = cap3 & 0x3fff;
3246 	ud->tchan_cnt = cap2 & 0x1ff;
3247 	ud->echan_cnt = (cap2 >> 9) & 0x1ff;
3248 	ud->rchan_cnt = (cap2 >> 18) & 0x1ff;
3249 	ch_count  = ud->tchan_cnt + ud->rchan_cnt;
3250 
3251 	ud->tchan_map = devm_kmalloc_array(dev, BITS_TO_LONGS(ud->tchan_cnt),
3252 					   sizeof(unsigned long), GFP_KERNEL);
3253 	ud->tchans = devm_kcalloc(dev, ud->tchan_cnt, sizeof(*ud->tchans),
3254 				  GFP_KERNEL);
3255 	ud->rchan_map = devm_kmalloc_array(dev, BITS_TO_LONGS(ud->rchan_cnt),
3256 					   sizeof(unsigned long), GFP_KERNEL);
3257 	ud->rchans = devm_kcalloc(dev, ud->rchan_cnt, sizeof(*ud->rchans),
3258 				  GFP_KERNEL);
3259 	ud->rflow_gp_map = devm_kmalloc_array(dev, BITS_TO_LONGS(ud->rflow_cnt),
3260 					      sizeof(unsigned long),
3261 					      GFP_KERNEL);
3262 	ud->rflow_gp_map_allocated = devm_kcalloc(dev,
3263 						  BITS_TO_LONGS(ud->rflow_cnt),
3264 						  sizeof(unsigned long),
3265 						  GFP_KERNEL);
3266 	ud->rflow_in_use = devm_kcalloc(dev, BITS_TO_LONGS(ud->rflow_cnt),
3267 					sizeof(unsigned long),
3268 					GFP_KERNEL);
3269 	ud->rflows = devm_kcalloc(dev, ud->rflow_cnt, sizeof(*ud->rflows),
3270 				  GFP_KERNEL);
3271 
3272 	if (!ud->tchan_map || !ud->rchan_map || !ud->rflow_gp_map ||
3273 	    !ud->rflow_gp_map_allocated || !ud->tchans || !ud->rchans ||
3274 	    !ud->rflows || !ud->rflow_in_use)
3275 		return -ENOMEM;
3276 
3277 	/*
3278 	 * RX flows with the same Ids as RX channels are reserved to be used
3279 	 * as default flows if remote HW can't generate flow_ids. Those
3280 	 * RX flows can be requested only explicitly by id.
3281 	 */
3282 	bitmap_set(ud->rflow_gp_map_allocated, 0, ud->rchan_cnt);
3283 
3284 	/* by default no GP rflows are assigned to Linux */
3285 	bitmap_set(ud->rflow_gp_map, 0, ud->rflow_cnt);
3286 
3287 	/* Get resource ranges from tisci */
3288 	for (i = 0; i < RM_RANGE_LAST; i++)
3289 		tisci_rm->rm_ranges[i] =
3290 			devm_ti_sci_get_of_resource(tisci_rm->tisci, dev,
3291 						    tisci_rm->tisci_dev_id,
3292 						    (char *)range_names[i]);
3293 
3294 	/* tchan ranges */
3295 	rm_res = tisci_rm->rm_ranges[RM_RANGE_TCHAN];
3296 	if (IS_ERR(rm_res)) {
3297 		bitmap_zero(ud->tchan_map, ud->tchan_cnt);
3298 	} else {
3299 		bitmap_fill(ud->tchan_map, ud->tchan_cnt);
3300 		for (i = 0; i < rm_res->sets; i++) {
3301 			rm_desc = &rm_res->desc[i];
3302 			bitmap_clear(ud->tchan_map, rm_desc->start,
3303 				     rm_desc->num);
3304 			dev_dbg(dev, "ti-sci-res: tchan: %d:%d\n",
3305 				rm_desc->start, rm_desc->num);
3306 		}
3307 	}
3308 	irq_res.sets = rm_res->sets;
3309 
3310 	/* rchan and matching default flow ranges */
3311 	rm_res = tisci_rm->rm_ranges[RM_RANGE_RCHAN];
3312 	if (IS_ERR(rm_res)) {
3313 		bitmap_zero(ud->rchan_map, ud->rchan_cnt);
3314 	} else {
3315 		bitmap_fill(ud->rchan_map, ud->rchan_cnt);
3316 		for (i = 0; i < rm_res->sets; i++) {
3317 			rm_desc = &rm_res->desc[i];
3318 			bitmap_clear(ud->rchan_map, rm_desc->start,
3319 				     rm_desc->num);
3320 			dev_dbg(dev, "ti-sci-res: rchan: %d:%d\n",
3321 				rm_desc->start, rm_desc->num);
3322 		}
3323 	}
3324 
3325 	irq_res.sets += rm_res->sets;
3326 	irq_res.desc = kcalloc(irq_res.sets, sizeof(*irq_res.desc), GFP_KERNEL);
3327 	rm_res = tisci_rm->rm_ranges[RM_RANGE_TCHAN];
3328 	for (i = 0; i < rm_res->sets; i++) {
3329 		irq_res.desc[i].start = rm_res->desc[i].start;
3330 		irq_res.desc[i].num = rm_res->desc[i].num;
3331 	}
3332 	rm_res = tisci_rm->rm_ranges[RM_RANGE_RCHAN];
3333 	for (j = 0; j < rm_res->sets; j++, i++) {
3334 		irq_res.desc[i].start = rm_res->desc[j].start +
3335 					ud->match_data->rchan_oes_offset;
3336 		irq_res.desc[i].num = rm_res->desc[j].num;
3337 	}
3338 	ret = ti_sci_inta_msi_domain_alloc_irqs(ud->dev, &irq_res);
3339 	kfree(irq_res.desc);
3340 	if (ret) {
3341 		dev_err(ud->dev, "Failed to allocate MSI interrupts\n");
3342 		return ret;
3343 	}
3344 
3345 	/* GP rflow ranges */
3346 	rm_res = tisci_rm->rm_ranges[RM_RANGE_RFLOW];
3347 	if (IS_ERR(rm_res)) {
3348 		/* all gp flows are assigned exclusively to Linux */
3349 		bitmap_clear(ud->rflow_gp_map, ud->rchan_cnt,
3350 			     ud->rflow_cnt - ud->rchan_cnt);
3351 	} else {
3352 		for (i = 0; i < rm_res->sets; i++) {
3353 			rm_desc = &rm_res->desc[i];
3354 			bitmap_clear(ud->rflow_gp_map, rm_desc->start,
3355 				     rm_desc->num);
3356 			dev_dbg(dev, "ti-sci-res: rflow: %d:%d\n",
3357 				rm_desc->start, rm_desc->num);
3358 		}
3359 	}
3360 
3361 	ch_count -= bitmap_weight(ud->tchan_map, ud->tchan_cnt);
3362 	ch_count -= bitmap_weight(ud->rchan_map, ud->rchan_cnt);
3363 	if (!ch_count)
3364 		return -ENODEV;
3365 
3366 	ud->channels = devm_kcalloc(dev, ch_count, sizeof(*ud->channels),
3367 				    GFP_KERNEL);
3368 	if (!ud->channels)
3369 		return -ENOMEM;
3370 
3371 	dev_info(dev, "Channels: %d (tchan: %u, rchan: %u, gp-rflow: %u)\n",
3372 		 ch_count,
3373 		 ud->tchan_cnt - bitmap_weight(ud->tchan_map, ud->tchan_cnt),
3374 		 ud->rchan_cnt - bitmap_weight(ud->rchan_map, ud->rchan_cnt),
3375 		 ud->rflow_cnt - bitmap_weight(ud->rflow_gp_map,
3376 					       ud->rflow_cnt));
3377 
3378 	return ch_count;
3379 }
3380 
3381 static int udma_setup_rx_flush(struct udma_dev *ud)
3382 {
3383 	struct udma_rx_flush *rx_flush = &ud->rx_flush;
3384 	struct cppi5_desc_hdr_t *tr_desc;
3385 	struct cppi5_tr_type1_t *tr_req;
3386 	struct cppi5_host_desc_t *desc;
3387 	struct device *dev = ud->dev;
3388 	struct udma_hwdesc *hwdesc;
3389 	size_t tr_size;
3390 
3391 	/* Allocate 1K buffer for discarded data on RX channel teardown */
3392 	rx_flush->buffer_size = SZ_1K;
3393 	rx_flush->buffer_vaddr = devm_kzalloc(dev, rx_flush->buffer_size,
3394 					      GFP_KERNEL);
3395 	if (!rx_flush->buffer_vaddr)
3396 		return -ENOMEM;
3397 
3398 	rx_flush->buffer_paddr = dma_map_single(dev, rx_flush->buffer_vaddr,
3399 						rx_flush->buffer_size,
3400 						DMA_TO_DEVICE);
3401 	if (dma_mapping_error(dev, rx_flush->buffer_paddr))
3402 		return -ENOMEM;
3403 
3404 	/* Set up descriptor to be used for TR mode */
3405 	hwdesc = &rx_flush->hwdescs[0];
3406 	tr_size = sizeof(struct cppi5_tr_type1_t);
3407 	hwdesc->cppi5_desc_size = cppi5_trdesc_calc_size(tr_size, 1);
3408 	hwdesc->cppi5_desc_size = ALIGN(hwdesc->cppi5_desc_size,
3409 					ud->desc_align);
3410 
3411 	hwdesc->cppi5_desc_vaddr = devm_kzalloc(dev, hwdesc->cppi5_desc_size,
3412 						GFP_KERNEL);
3413 	if (!hwdesc->cppi5_desc_vaddr)
3414 		return -ENOMEM;
3415 
3416 	hwdesc->cppi5_desc_paddr = dma_map_single(dev, hwdesc->cppi5_desc_vaddr,
3417 						  hwdesc->cppi5_desc_size,
3418 						  DMA_TO_DEVICE);
3419 	if (dma_mapping_error(dev, hwdesc->cppi5_desc_paddr))
3420 		return -ENOMEM;
3421 
3422 	/* Start of the TR req records */
3423 	hwdesc->tr_req_base = hwdesc->cppi5_desc_vaddr + tr_size;
3424 	/* Start address of the TR response array */
3425 	hwdesc->tr_resp_base = hwdesc->tr_req_base + tr_size;
3426 
3427 	tr_desc = hwdesc->cppi5_desc_vaddr;
3428 	cppi5_trdesc_init(tr_desc, 1, tr_size, 0, 0);
3429 	cppi5_desc_set_pktids(tr_desc, 0, CPPI5_INFO1_DESC_FLOWID_DEFAULT);
3430 	cppi5_desc_set_retpolicy(tr_desc, 0, 0);
3431 
3432 	tr_req = hwdesc->tr_req_base;
3433 	cppi5_tr_init(&tr_req->flags, CPPI5_TR_TYPE1, false, false,
3434 		      CPPI5_TR_EVENT_SIZE_COMPLETION, 0);
3435 	cppi5_tr_csf_set(&tr_req->flags, CPPI5_TR_CSF_SUPR_EVT);
3436 
3437 	tr_req->addr = rx_flush->buffer_paddr;
3438 	tr_req->icnt0 = rx_flush->buffer_size;
3439 	tr_req->icnt1 = 1;
3440 
3441 	dma_sync_single_for_device(dev, hwdesc->cppi5_desc_paddr,
3442 				   hwdesc->cppi5_desc_size, DMA_TO_DEVICE);
3443 
3444 	/* Set up descriptor to be used for packet mode */
3445 	hwdesc = &rx_flush->hwdescs[1];
3446 	hwdesc->cppi5_desc_size = ALIGN(sizeof(struct cppi5_host_desc_t) +
3447 					CPPI5_INFO0_HDESC_EPIB_SIZE +
3448 					CPPI5_INFO0_HDESC_PSDATA_MAX_SIZE,
3449 					ud->desc_align);
3450 
3451 	hwdesc->cppi5_desc_vaddr = devm_kzalloc(dev, hwdesc->cppi5_desc_size,
3452 						GFP_KERNEL);
3453 	if (!hwdesc->cppi5_desc_vaddr)
3454 		return -ENOMEM;
3455 
3456 	hwdesc->cppi5_desc_paddr = dma_map_single(dev, hwdesc->cppi5_desc_vaddr,
3457 						  hwdesc->cppi5_desc_size,
3458 						  DMA_TO_DEVICE);
3459 	if (dma_mapping_error(dev, hwdesc->cppi5_desc_paddr))
3460 		return -ENOMEM;
3461 
3462 	desc = hwdesc->cppi5_desc_vaddr;
3463 	cppi5_hdesc_init(desc, 0, 0);
3464 	cppi5_desc_set_pktids(&desc->hdr, 0, CPPI5_INFO1_DESC_FLOWID_DEFAULT);
3465 	cppi5_desc_set_retpolicy(&desc->hdr, 0, 0);
3466 
3467 	cppi5_hdesc_attach_buf(desc,
3468 			       rx_flush->buffer_paddr, rx_flush->buffer_size,
3469 			       rx_flush->buffer_paddr, rx_flush->buffer_size);
3470 
3471 	dma_sync_single_for_device(dev, hwdesc->cppi5_desc_paddr,
3472 				   hwdesc->cppi5_desc_size, DMA_TO_DEVICE);
3473 	return 0;
3474 }
3475 
3476 #ifdef CONFIG_DEBUG_FS
3477 static void udma_dbg_summary_show_chan(struct seq_file *s,
3478 				       struct dma_chan *chan)
3479 {
3480 	struct udma_chan *uc = to_udma_chan(chan);
3481 	struct udma_chan_config *ucc = &uc->config;
3482 
3483 	seq_printf(s, " %-13s| %s", dma_chan_name(chan),
3484 		   chan->dbg_client_name ?: "in-use");
3485 	seq_printf(s, " (%s, ", dmaengine_get_direction_text(uc->config.dir));
3486 
3487 	switch (uc->config.dir) {
3488 	case DMA_MEM_TO_MEM:
3489 		seq_printf(s, "chan%d pair [0x%04x -> 0x%04x], ", uc->tchan->id,
3490 			   ucc->src_thread, ucc->dst_thread);
3491 		break;
3492 	case DMA_DEV_TO_MEM:
3493 		seq_printf(s, "rchan%d [0x%04x -> 0x%04x], ", uc->rchan->id,
3494 			   ucc->src_thread, ucc->dst_thread);
3495 		break;
3496 	case DMA_MEM_TO_DEV:
3497 		seq_printf(s, "tchan%d [0x%04x -> 0x%04x], ", uc->tchan->id,
3498 			   ucc->src_thread, ucc->dst_thread);
3499 		break;
3500 	default:
3501 		seq_printf(s, ")\n");
3502 		return;
3503 	}
3504 
3505 	if (ucc->ep_type == PSIL_EP_NATIVE) {
3506 		seq_printf(s, "PSI-L Native");
3507 		if (ucc->metadata_size) {
3508 			seq_printf(s, "[%s", ucc->needs_epib ? " EPIB" : "");
3509 			if (ucc->psd_size)
3510 				seq_printf(s, " PSDsize:%u", ucc->psd_size);
3511 			seq_printf(s, " ]");
3512 		}
3513 	} else {
3514 		seq_printf(s, "PDMA");
3515 		if (ucc->enable_acc32 || ucc->enable_burst)
3516 			seq_printf(s, "[%s%s ]",
3517 				   ucc->enable_acc32 ? " ACC32" : "",
3518 				   ucc->enable_burst ? " BURST" : "");
3519 	}
3520 
3521 	seq_printf(s, ", %s)\n", ucc->pkt_mode ? "Packet mode" : "TR mode");
3522 }
3523 
3524 static void udma_dbg_summary_show(struct seq_file *s,
3525 				  struct dma_device *dma_dev)
3526 {
3527 	struct dma_chan *chan;
3528 
3529 	list_for_each_entry(chan, &dma_dev->channels, device_node) {
3530 		if (chan->client_count)
3531 			udma_dbg_summary_show_chan(s, chan);
3532 	}
3533 }
3534 #endif /* CONFIG_DEBUG_FS */
3535 
3536 #define TI_UDMAC_BUSWIDTHS	(BIT(DMA_SLAVE_BUSWIDTH_1_BYTE) | \
3537 				 BIT(DMA_SLAVE_BUSWIDTH_2_BYTES) | \
3538 				 BIT(DMA_SLAVE_BUSWIDTH_3_BYTES) | \
3539 				 BIT(DMA_SLAVE_BUSWIDTH_4_BYTES) | \
3540 				 BIT(DMA_SLAVE_BUSWIDTH_8_BYTES))
3541 
3542 static int udma_probe(struct platform_device *pdev)
3543 {
3544 	struct device_node *navss_node = pdev->dev.parent->of_node;
3545 	struct device *dev = &pdev->dev;
3546 	struct udma_dev *ud;
3547 	const struct of_device_id *match;
3548 	int i, ret;
3549 	int ch_count;
3550 
3551 	ret = dma_coerce_mask_and_coherent(dev, DMA_BIT_MASK(48));
3552 	if (ret)
3553 		dev_err(dev, "failed to set dma mask stuff\n");
3554 
3555 	ud = devm_kzalloc(dev, sizeof(*ud), GFP_KERNEL);
3556 	if (!ud)
3557 		return -ENOMEM;
3558 
3559 	ret = udma_get_mmrs(pdev, ud);
3560 	if (ret)
3561 		return ret;
3562 
3563 	ud->tisci_rm.tisci = ti_sci_get_by_phandle(dev->of_node, "ti,sci");
3564 	if (IS_ERR(ud->tisci_rm.tisci))
3565 		return PTR_ERR(ud->tisci_rm.tisci);
3566 
3567 	ret = of_property_read_u32(dev->of_node, "ti,sci-dev-id",
3568 				   &ud->tisci_rm.tisci_dev_id);
3569 	if (ret) {
3570 		dev_err(dev, "ti,sci-dev-id read failure %d\n", ret);
3571 		return ret;
3572 	}
3573 	pdev->id = ud->tisci_rm.tisci_dev_id;
3574 
3575 	ret = of_property_read_u32(navss_node, "ti,sci-dev-id",
3576 				   &ud->tisci_rm.tisci_navss_dev_id);
3577 	if (ret) {
3578 		dev_err(dev, "NAVSS ti,sci-dev-id read failure %d\n", ret);
3579 		return ret;
3580 	}
3581 
3582 	ret = of_property_read_u32(dev->of_node, "ti,udma-atype", &ud->atype);
3583 	if (!ret && ud->atype > 2) {
3584 		dev_err(dev, "Invalid atype: %u\n", ud->atype);
3585 		return -EINVAL;
3586 	}
3587 
3588 	ud->tisci_rm.tisci_udmap_ops = &ud->tisci_rm.tisci->ops.rm_udmap_ops;
3589 	ud->tisci_rm.tisci_psil_ops = &ud->tisci_rm.tisci->ops.rm_psil_ops;
3590 
3591 	ud->ringacc = of_k3_ringacc_get_by_phandle(dev->of_node, "ti,ringacc");
3592 	if (IS_ERR(ud->ringacc))
3593 		return PTR_ERR(ud->ringacc);
3594 
3595 	dev->msi_domain = of_msi_get_domain(dev, dev->of_node,
3596 					    DOMAIN_BUS_TI_SCI_INTA_MSI);
3597 	if (!dev->msi_domain) {
3598 		dev_err(dev, "Failed to get MSI domain\n");
3599 		return -EPROBE_DEFER;
3600 	}
3601 
3602 	match = of_match_node(udma_of_match, dev->of_node);
3603 	if (!match) {
3604 		dev_err(dev, "No compatible match found\n");
3605 		return -ENODEV;
3606 	}
3607 	ud->match_data = match->data;
3608 
3609 	dma_cap_set(DMA_SLAVE, ud->ddev.cap_mask);
3610 	dma_cap_set(DMA_CYCLIC, ud->ddev.cap_mask);
3611 
3612 	ud->ddev.device_alloc_chan_resources = udma_alloc_chan_resources;
3613 	ud->ddev.device_config = udma_slave_config;
3614 	ud->ddev.device_prep_slave_sg = udma_prep_slave_sg;
3615 	ud->ddev.device_prep_dma_cyclic = udma_prep_dma_cyclic;
3616 	ud->ddev.device_issue_pending = udma_issue_pending;
3617 	ud->ddev.device_tx_status = udma_tx_status;
3618 	ud->ddev.device_pause = udma_pause;
3619 	ud->ddev.device_resume = udma_resume;
3620 	ud->ddev.device_terminate_all = udma_terminate_all;
3621 	ud->ddev.device_synchronize = udma_synchronize;
3622 #ifdef CONFIG_DEBUG_FS
3623 	ud->ddev.dbg_summary_show = udma_dbg_summary_show;
3624 #endif
3625 
3626 	ud->ddev.device_free_chan_resources = udma_free_chan_resources;
3627 	ud->ddev.src_addr_widths = TI_UDMAC_BUSWIDTHS;
3628 	ud->ddev.dst_addr_widths = TI_UDMAC_BUSWIDTHS;
3629 	ud->ddev.directions = BIT(DMA_DEV_TO_MEM) | BIT(DMA_MEM_TO_DEV);
3630 	ud->ddev.residue_granularity = DMA_RESIDUE_GRANULARITY_BURST;
3631 	ud->ddev.copy_align = DMAENGINE_ALIGN_8_BYTES;
3632 	ud->ddev.desc_metadata_modes = DESC_METADATA_CLIENT |
3633 				       DESC_METADATA_ENGINE;
3634 	if (ud->match_data->enable_memcpy_support) {
3635 		dma_cap_set(DMA_MEMCPY, ud->ddev.cap_mask);
3636 		ud->ddev.device_prep_dma_memcpy = udma_prep_dma_memcpy;
3637 		ud->ddev.directions |= BIT(DMA_MEM_TO_MEM);
3638 	}
3639 
3640 	ud->ddev.dev = dev;
3641 	ud->dev = dev;
3642 	ud->psil_base = ud->match_data->psil_base;
3643 
3644 	INIT_LIST_HEAD(&ud->ddev.channels);
3645 	INIT_LIST_HEAD(&ud->desc_to_purge);
3646 
3647 	ch_count = udma_setup_resources(ud);
3648 	if (ch_count <= 0)
3649 		return ch_count;
3650 
3651 	spin_lock_init(&ud->lock);
3652 	INIT_WORK(&ud->purge_work, udma_purge_desc_work);
3653 
3654 	ud->desc_align = 64;
3655 	if (ud->desc_align < dma_get_cache_alignment())
3656 		ud->desc_align = dma_get_cache_alignment();
3657 
3658 	ret = udma_setup_rx_flush(ud);
3659 	if (ret)
3660 		return ret;
3661 
3662 	for (i = 0; i < ud->tchan_cnt; i++) {
3663 		struct udma_tchan *tchan = &ud->tchans[i];
3664 
3665 		tchan->id = i;
3666 		tchan->reg_rt = ud->mmrs[MMR_TCHANRT] + i * 0x1000;
3667 	}
3668 
3669 	for (i = 0; i < ud->rchan_cnt; i++) {
3670 		struct udma_rchan *rchan = &ud->rchans[i];
3671 
3672 		rchan->id = i;
3673 		rchan->reg_rt = ud->mmrs[MMR_RCHANRT] + i * 0x1000;
3674 	}
3675 
3676 	for (i = 0; i < ud->rflow_cnt; i++) {
3677 		struct udma_rflow *rflow = &ud->rflows[i];
3678 
3679 		rflow->id = i;
3680 	}
3681 
3682 	for (i = 0; i < ch_count; i++) {
3683 		struct udma_chan *uc = &ud->channels[i];
3684 
3685 		uc->ud = ud;
3686 		uc->vc.desc_free = udma_desc_free;
3687 		uc->id = i;
3688 		uc->tchan = NULL;
3689 		uc->rchan = NULL;
3690 		uc->config.remote_thread_id = -1;
3691 		uc->config.dir = DMA_MEM_TO_MEM;
3692 		uc->name = devm_kasprintf(dev, GFP_KERNEL, "%s chan%d",
3693 					  dev_name(dev), i);
3694 
3695 		vchan_init(&uc->vc, &ud->ddev);
3696 		/* Use custom vchan completion handling */
3697 		tasklet_init(&uc->vc.task, udma_vchan_complete,
3698 			     (unsigned long)&uc->vc);
3699 		init_completion(&uc->teardown_completed);
3700 		INIT_DELAYED_WORK(&uc->tx_drain.work, udma_check_tx_completion);
3701 	}
3702 
3703 	ret = dma_async_device_register(&ud->ddev);
3704 	if (ret) {
3705 		dev_err(dev, "failed to register slave DMA engine: %d\n", ret);
3706 		return ret;
3707 	}
3708 
3709 	platform_set_drvdata(pdev, ud);
3710 
3711 	ret = of_dma_controller_register(dev->of_node, udma_of_xlate, ud);
3712 	if (ret) {
3713 		dev_err(dev, "failed to register of_dma controller\n");
3714 		dma_async_device_unregister(&ud->ddev);
3715 	}
3716 
3717 	return ret;
3718 }
3719 
3720 static struct platform_driver udma_driver = {
3721 	.driver = {
3722 		.name	= "ti-udma",
3723 		.of_match_table = udma_of_match,
3724 		.suppress_bind_attrs = true,
3725 	},
3726 	.probe		= udma_probe,
3727 };
3728 builtin_platform_driver(udma_driver);
3729 
3730 /* Private interfaces to UDMA */
3731 #include "k3-udma-private.c"
3732