xref: /openbmc/linux/drivers/dma/ti/k3-udma.c (revision b58c6630)
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[0];
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 	u32 in_ring_cnt; /* number of descriptors in flight */
235 
236 	/* Channel configuration parameters */
237 	struct udma_chan_config config;
238 
239 	/* dmapool for packet mode descriptors */
240 	bool use_dma_pool;
241 	struct dma_pool *hdesc_pool;
242 
243 	u32 id;
244 };
245 
246 static inline struct udma_dev *to_udma_dev(struct dma_device *d)
247 {
248 	return container_of(d, struct udma_dev, ddev);
249 }
250 
251 static inline struct udma_chan *to_udma_chan(struct dma_chan *c)
252 {
253 	return container_of(c, struct udma_chan, vc.chan);
254 }
255 
256 static inline struct udma_desc *to_udma_desc(struct dma_async_tx_descriptor *t)
257 {
258 	return container_of(t, struct udma_desc, vd.tx);
259 }
260 
261 /* Generic register access functions */
262 static inline u32 udma_read(void __iomem *base, int reg)
263 {
264 	return readl(base + reg);
265 }
266 
267 static inline void udma_write(void __iomem *base, int reg, u32 val)
268 {
269 	writel(val, base + reg);
270 }
271 
272 static inline void udma_update_bits(void __iomem *base, int reg,
273 				    u32 mask, u32 val)
274 {
275 	u32 tmp, orig;
276 
277 	orig = readl(base + reg);
278 	tmp = orig & ~mask;
279 	tmp |= (val & mask);
280 
281 	if (tmp != orig)
282 		writel(tmp, base + reg);
283 }
284 
285 /* TCHANRT */
286 static inline u32 udma_tchanrt_read(struct udma_tchan *tchan, int reg)
287 {
288 	if (!tchan)
289 		return 0;
290 	return udma_read(tchan->reg_rt, reg);
291 }
292 
293 static inline void udma_tchanrt_write(struct udma_tchan *tchan, int reg,
294 				      u32 val)
295 {
296 	if (!tchan)
297 		return;
298 	udma_write(tchan->reg_rt, reg, val);
299 }
300 
301 static inline void udma_tchanrt_update_bits(struct udma_tchan *tchan, int reg,
302 					    u32 mask, u32 val)
303 {
304 	if (!tchan)
305 		return;
306 	udma_update_bits(tchan->reg_rt, reg, mask, val);
307 }
308 
309 /* RCHANRT */
310 static inline u32 udma_rchanrt_read(struct udma_rchan *rchan, int reg)
311 {
312 	if (!rchan)
313 		return 0;
314 	return udma_read(rchan->reg_rt, reg);
315 }
316 
317 static inline void udma_rchanrt_write(struct udma_rchan *rchan, int reg,
318 				      u32 val)
319 {
320 	if (!rchan)
321 		return;
322 	udma_write(rchan->reg_rt, reg, val);
323 }
324 
325 static inline void udma_rchanrt_update_bits(struct udma_rchan *rchan, int reg,
326 					    u32 mask, u32 val)
327 {
328 	if (!rchan)
329 		return;
330 	udma_update_bits(rchan->reg_rt, reg, mask, val);
331 }
332 
333 static int navss_psil_pair(struct udma_dev *ud, u32 src_thread, u32 dst_thread)
334 {
335 	struct udma_tisci_rm *tisci_rm = &ud->tisci_rm;
336 
337 	dst_thread |= K3_PSIL_DST_THREAD_ID_OFFSET;
338 	return tisci_rm->tisci_psil_ops->pair(tisci_rm->tisci,
339 					      tisci_rm->tisci_navss_dev_id,
340 					      src_thread, dst_thread);
341 }
342 
343 static int navss_psil_unpair(struct udma_dev *ud, u32 src_thread,
344 			     u32 dst_thread)
345 {
346 	struct udma_tisci_rm *tisci_rm = &ud->tisci_rm;
347 
348 	dst_thread |= K3_PSIL_DST_THREAD_ID_OFFSET;
349 	return tisci_rm->tisci_psil_ops->unpair(tisci_rm->tisci,
350 						tisci_rm->tisci_navss_dev_id,
351 						src_thread, dst_thread);
352 }
353 
354 static void udma_reset_uchan(struct udma_chan *uc)
355 {
356 	memset(&uc->config, 0, sizeof(uc->config));
357 	uc->config.remote_thread_id = -1;
358 	uc->state = UDMA_CHAN_IS_IDLE;
359 }
360 
361 static void udma_dump_chan_stdata(struct udma_chan *uc)
362 {
363 	struct device *dev = uc->ud->dev;
364 	u32 offset;
365 	int i;
366 
367 	if (uc->config.dir == DMA_MEM_TO_DEV || uc->config.dir == DMA_MEM_TO_MEM) {
368 		dev_dbg(dev, "TCHAN State data:\n");
369 		for (i = 0; i < 32; i++) {
370 			offset = UDMA_TCHAN_RT_STDATA_REG + i * 4;
371 			dev_dbg(dev, "TRT_STDATA[%02d]: 0x%08x\n", i,
372 				udma_tchanrt_read(uc->tchan, offset));
373 		}
374 	}
375 
376 	if (uc->config.dir == DMA_DEV_TO_MEM || uc->config.dir == DMA_MEM_TO_MEM) {
377 		dev_dbg(dev, "RCHAN State data:\n");
378 		for (i = 0; i < 32; i++) {
379 			offset = UDMA_RCHAN_RT_STDATA_REG + i * 4;
380 			dev_dbg(dev, "RRT_STDATA[%02d]: 0x%08x\n", i,
381 				udma_rchanrt_read(uc->rchan, offset));
382 		}
383 	}
384 }
385 
386 static inline dma_addr_t udma_curr_cppi5_desc_paddr(struct udma_desc *d,
387 						    int idx)
388 {
389 	return d->hwdesc[idx].cppi5_desc_paddr;
390 }
391 
392 static inline void *udma_curr_cppi5_desc_vaddr(struct udma_desc *d, int idx)
393 {
394 	return d->hwdesc[idx].cppi5_desc_vaddr;
395 }
396 
397 static struct udma_desc *udma_udma_desc_from_paddr(struct udma_chan *uc,
398 						   dma_addr_t paddr)
399 {
400 	struct udma_desc *d = uc->terminated_desc;
401 
402 	if (d) {
403 		dma_addr_t desc_paddr = udma_curr_cppi5_desc_paddr(d,
404 								   d->desc_idx);
405 
406 		if (desc_paddr != paddr)
407 			d = NULL;
408 	}
409 
410 	if (!d) {
411 		d = uc->desc;
412 		if (d) {
413 			dma_addr_t desc_paddr = udma_curr_cppi5_desc_paddr(d,
414 								d->desc_idx);
415 
416 			if (desc_paddr != paddr)
417 				d = NULL;
418 		}
419 	}
420 
421 	return d;
422 }
423 
424 static void udma_free_hwdesc(struct udma_chan *uc, struct udma_desc *d)
425 {
426 	if (uc->use_dma_pool) {
427 		int i;
428 
429 		for (i = 0; i < d->hwdesc_count; i++) {
430 			if (!d->hwdesc[i].cppi5_desc_vaddr)
431 				continue;
432 
433 			dma_pool_free(uc->hdesc_pool,
434 				      d->hwdesc[i].cppi5_desc_vaddr,
435 				      d->hwdesc[i].cppi5_desc_paddr);
436 
437 			d->hwdesc[i].cppi5_desc_vaddr = NULL;
438 		}
439 	} else if (d->hwdesc[0].cppi5_desc_vaddr) {
440 		struct udma_dev *ud = uc->ud;
441 
442 		dma_free_coherent(ud->dev, d->hwdesc[0].cppi5_desc_size,
443 				  d->hwdesc[0].cppi5_desc_vaddr,
444 				  d->hwdesc[0].cppi5_desc_paddr);
445 
446 		d->hwdesc[0].cppi5_desc_vaddr = NULL;
447 	}
448 }
449 
450 static void udma_purge_desc_work(struct work_struct *work)
451 {
452 	struct udma_dev *ud = container_of(work, typeof(*ud), purge_work);
453 	struct virt_dma_desc *vd, *_vd;
454 	unsigned long flags;
455 	LIST_HEAD(head);
456 
457 	spin_lock_irqsave(&ud->lock, flags);
458 	list_splice_tail_init(&ud->desc_to_purge, &head);
459 	spin_unlock_irqrestore(&ud->lock, flags);
460 
461 	list_for_each_entry_safe(vd, _vd, &head, node) {
462 		struct udma_chan *uc = to_udma_chan(vd->tx.chan);
463 		struct udma_desc *d = to_udma_desc(&vd->tx);
464 
465 		udma_free_hwdesc(uc, d);
466 		list_del(&vd->node);
467 		kfree(d);
468 	}
469 
470 	/* If more to purge, schedule the work again */
471 	if (!list_empty(&ud->desc_to_purge))
472 		schedule_work(&ud->purge_work);
473 }
474 
475 static void udma_desc_free(struct virt_dma_desc *vd)
476 {
477 	struct udma_dev *ud = to_udma_dev(vd->tx.chan->device);
478 	struct udma_chan *uc = to_udma_chan(vd->tx.chan);
479 	struct udma_desc *d = to_udma_desc(&vd->tx);
480 	unsigned long flags;
481 
482 	if (uc->terminated_desc == d)
483 		uc->terminated_desc = NULL;
484 
485 	if (uc->use_dma_pool) {
486 		udma_free_hwdesc(uc, d);
487 		kfree(d);
488 		return;
489 	}
490 
491 	spin_lock_irqsave(&ud->lock, flags);
492 	list_add_tail(&vd->node, &ud->desc_to_purge);
493 	spin_unlock_irqrestore(&ud->lock, flags);
494 
495 	schedule_work(&ud->purge_work);
496 }
497 
498 static bool udma_is_chan_running(struct udma_chan *uc)
499 {
500 	u32 trt_ctl = 0;
501 	u32 rrt_ctl = 0;
502 
503 	if (uc->tchan)
504 		trt_ctl = udma_tchanrt_read(uc->tchan, UDMA_TCHAN_RT_CTL_REG);
505 	if (uc->rchan)
506 		rrt_ctl = udma_rchanrt_read(uc->rchan, UDMA_RCHAN_RT_CTL_REG);
507 
508 	if (trt_ctl & UDMA_CHAN_RT_CTL_EN || rrt_ctl & UDMA_CHAN_RT_CTL_EN)
509 		return true;
510 
511 	return false;
512 }
513 
514 static bool udma_is_chan_paused(struct udma_chan *uc)
515 {
516 	u32 val, pause_mask;
517 
518 	switch (uc->config.dir) {
519 	case DMA_DEV_TO_MEM:
520 		val = udma_rchanrt_read(uc->rchan,
521 					UDMA_RCHAN_RT_PEER_RT_EN_REG);
522 		pause_mask = UDMA_PEER_RT_EN_PAUSE;
523 		break;
524 	case DMA_MEM_TO_DEV:
525 		val = udma_tchanrt_read(uc->tchan,
526 					UDMA_TCHAN_RT_PEER_RT_EN_REG);
527 		pause_mask = UDMA_PEER_RT_EN_PAUSE;
528 		break;
529 	case DMA_MEM_TO_MEM:
530 		val = udma_tchanrt_read(uc->tchan, UDMA_TCHAN_RT_CTL_REG);
531 		pause_mask = UDMA_CHAN_RT_CTL_PAUSE;
532 		break;
533 	default:
534 		return false;
535 	}
536 
537 	if (val & pause_mask)
538 		return true;
539 
540 	return false;
541 }
542 
543 static void udma_sync_for_device(struct udma_chan *uc, int idx)
544 {
545 	struct udma_desc *d = uc->desc;
546 
547 	if (uc->cyclic && uc->config.pkt_mode) {
548 		dma_sync_single_for_device(uc->ud->dev,
549 					   d->hwdesc[idx].cppi5_desc_paddr,
550 					   d->hwdesc[idx].cppi5_desc_size,
551 					   DMA_TO_DEVICE);
552 	} else {
553 		int i;
554 
555 		for (i = 0; i < d->hwdesc_count; i++) {
556 			if (!d->hwdesc[i].cppi5_desc_vaddr)
557 				continue;
558 
559 			dma_sync_single_for_device(uc->ud->dev,
560 						d->hwdesc[i].cppi5_desc_paddr,
561 						d->hwdesc[i].cppi5_desc_size,
562 						DMA_TO_DEVICE);
563 		}
564 	}
565 }
566 
567 static inline dma_addr_t udma_get_rx_flush_hwdesc_paddr(struct udma_chan *uc)
568 {
569 	return uc->ud->rx_flush.hwdescs[uc->config.pkt_mode].cppi5_desc_paddr;
570 }
571 
572 static int udma_push_to_ring(struct udma_chan *uc, int idx)
573 {
574 	struct udma_desc *d = uc->desc;
575 	struct k3_ring *ring = NULL;
576 	dma_addr_t paddr;
577 	int ret;
578 
579 	switch (uc->config.dir) {
580 	case DMA_DEV_TO_MEM:
581 		ring = uc->rflow->fd_ring;
582 		break;
583 	case DMA_MEM_TO_DEV:
584 	case DMA_MEM_TO_MEM:
585 		ring = uc->tchan->t_ring;
586 		break;
587 	default:
588 		return -EINVAL;
589 	}
590 
591 	/* RX flush packet: idx == -1 is only passed in case of DEV_TO_MEM */
592 	if (idx == -1) {
593 		paddr = udma_get_rx_flush_hwdesc_paddr(uc);
594 	} else {
595 		paddr = udma_curr_cppi5_desc_paddr(d, idx);
596 
597 		wmb(); /* Ensure that writes are not moved over this point */
598 		udma_sync_for_device(uc, idx);
599 	}
600 
601 	ret = k3_ringacc_ring_push(ring, &paddr);
602 	if (!ret)
603 		uc->in_ring_cnt++;
604 
605 	return ret;
606 }
607 
608 static bool udma_desc_is_rx_flush(struct udma_chan *uc, dma_addr_t addr)
609 {
610 	if (uc->config.dir != DMA_DEV_TO_MEM)
611 		return false;
612 
613 	if (addr == udma_get_rx_flush_hwdesc_paddr(uc))
614 		return true;
615 
616 	return false;
617 }
618 
619 static int udma_pop_from_ring(struct udma_chan *uc, dma_addr_t *addr)
620 {
621 	struct k3_ring *ring = NULL;
622 	int ret = -ENOENT;
623 
624 	switch (uc->config.dir) {
625 	case DMA_DEV_TO_MEM:
626 		ring = uc->rflow->r_ring;
627 		break;
628 	case DMA_MEM_TO_DEV:
629 	case DMA_MEM_TO_MEM:
630 		ring = uc->tchan->tc_ring;
631 		break;
632 	default:
633 		break;
634 	}
635 
636 	if (ring && k3_ringacc_ring_get_occ(ring)) {
637 		struct udma_desc *d = NULL;
638 
639 		ret = k3_ringacc_ring_pop(ring, addr);
640 		if (ret)
641 			return ret;
642 
643 		/* Teardown completion */
644 		if (cppi5_desc_is_tdcm(*addr))
645 			return ret;
646 
647 		/* Check for flush descriptor */
648 		if (udma_desc_is_rx_flush(uc, *addr))
649 			return -ENOENT;
650 
651 		d = udma_udma_desc_from_paddr(uc, *addr);
652 
653 		if (d)
654 			dma_sync_single_for_cpu(uc->ud->dev, *addr,
655 						d->hwdesc[0].cppi5_desc_size,
656 						DMA_FROM_DEVICE);
657 		rmb(); /* Ensure that reads are not moved before this point */
658 
659 		if (!ret)
660 			uc->in_ring_cnt--;
661 	}
662 
663 	return ret;
664 }
665 
666 static void udma_reset_rings(struct udma_chan *uc)
667 {
668 	struct k3_ring *ring1 = NULL;
669 	struct k3_ring *ring2 = NULL;
670 
671 	switch (uc->config.dir) {
672 	case DMA_DEV_TO_MEM:
673 		if (uc->rchan) {
674 			ring1 = uc->rflow->fd_ring;
675 			ring2 = uc->rflow->r_ring;
676 		}
677 		break;
678 	case DMA_MEM_TO_DEV:
679 	case DMA_MEM_TO_MEM:
680 		if (uc->tchan) {
681 			ring1 = uc->tchan->t_ring;
682 			ring2 = uc->tchan->tc_ring;
683 		}
684 		break;
685 	default:
686 		break;
687 	}
688 
689 	if (ring1)
690 		k3_ringacc_ring_reset_dma(ring1,
691 					  k3_ringacc_ring_get_occ(ring1));
692 	if (ring2)
693 		k3_ringacc_ring_reset(ring2);
694 
695 	/* make sure we are not leaking memory by stalled descriptor */
696 	if (uc->terminated_desc) {
697 		udma_desc_free(&uc->terminated_desc->vd);
698 		uc->terminated_desc = NULL;
699 	}
700 
701 	uc->in_ring_cnt = 0;
702 }
703 
704 static void udma_reset_counters(struct udma_chan *uc)
705 {
706 	u32 val;
707 
708 	if (uc->tchan) {
709 		val = udma_tchanrt_read(uc->tchan, UDMA_TCHAN_RT_BCNT_REG);
710 		udma_tchanrt_write(uc->tchan, UDMA_TCHAN_RT_BCNT_REG, val);
711 
712 		val = udma_tchanrt_read(uc->tchan, UDMA_TCHAN_RT_SBCNT_REG);
713 		udma_tchanrt_write(uc->tchan, UDMA_TCHAN_RT_SBCNT_REG, val);
714 
715 		val = udma_tchanrt_read(uc->tchan, UDMA_TCHAN_RT_PCNT_REG);
716 		udma_tchanrt_write(uc->tchan, UDMA_TCHAN_RT_PCNT_REG, val);
717 
718 		val = udma_tchanrt_read(uc->tchan, UDMA_TCHAN_RT_PEER_BCNT_REG);
719 		udma_tchanrt_write(uc->tchan, UDMA_TCHAN_RT_PEER_BCNT_REG, val);
720 	}
721 
722 	if (uc->rchan) {
723 		val = udma_rchanrt_read(uc->rchan, UDMA_RCHAN_RT_BCNT_REG);
724 		udma_rchanrt_write(uc->rchan, UDMA_RCHAN_RT_BCNT_REG, val);
725 
726 		val = udma_rchanrt_read(uc->rchan, UDMA_RCHAN_RT_SBCNT_REG);
727 		udma_rchanrt_write(uc->rchan, UDMA_RCHAN_RT_SBCNT_REG, val);
728 
729 		val = udma_rchanrt_read(uc->rchan, UDMA_RCHAN_RT_PCNT_REG);
730 		udma_rchanrt_write(uc->rchan, UDMA_RCHAN_RT_PCNT_REG, val);
731 
732 		val = udma_rchanrt_read(uc->rchan, UDMA_RCHAN_RT_PEER_BCNT_REG);
733 		udma_rchanrt_write(uc->rchan, UDMA_RCHAN_RT_PEER_BCNT_REG, val);
734 	}
735 
736 	uc->bcnt = 0;
737 }
738 
739 static int udma_reset_chan(struct udma_chan *uc, bool hard)
740 {
741 	switch (uc->config.dir) {
742 	case DMA_DEV_TO_MEM:
743 		udma_rchanrt_write(uc->rchan, UDMA_RCHAN_RT_PEER_RT_EN_REG, 0);
744 		udma_rchanrt_write(uc->rchan, UDMA_RCHAN_RT_CTL_REG, 0);
745 		break;
746 	case DMA_MEM_TO_DEV:
747 		udma_tchanrt_write(uc->tchan, UDMA_TCHAN_RT_CTL_REG, 0);
748 		udma_tchanrt_write(uc->tchan, UDMA_TCHAN_RT_PEER_RT_EN_REG, 0);
749 		break;
750 	case DMA_MEM_TO_MEM:
751 		udma_rchanrt_write(uc->rchan, UDMA_RCHAN_RT_CTL_REG, 0);
752 		udma_tchanrt_write(uc->tchan, UDMA_TCHAN_RT_CTL_REG, 0);
753 		break;
754 	default:
755 		return -EINVAL;
756 	}
757 
758 	/* Reset all counters */
759 	udma_reset_counters(uc);
760 
761 	/* Hard reset: re-initialize the channel to reset */
762 	if (hard) {
763 		struct udma_chan_config ucc_backup;
764 		int ret;
765 
766 		memcpy(&ucc_backup, &uc->config, sizeof(uc->config));
767 		uc->ud->ddev.device_free_chan_resources(&uc->vc.chan);
768 
769 		/* restore the channel configuration */
770 		memcpy(&uc->config, &ucc_backup, sizeof(uc->config));
771 		ret = uc->ud->ddev.device_alloc_chan_resources(&uc->vc.chan);
772 		if (ret)
773 			return ret;
774 
775 		/*
776 		 * Setting forced teardown after forced reset helps recovering
777 		 * the rchan.
778 		 */
779 		if (uc->config.dir == DMA_DEV_TO_MEM)
780 			udma_rchanrt_write(uc->rchan, UDMA_RCHAN_RT_CTL_REG,
781 					   UDMA_CHAN_RT_CTL_EN |
782 					   UDMA_CHAN_RT_CTL_TDOWN |
783 					   UDMA_CHAN_RT_CTL_FTDOWN);
784 	}
785 	uc->state = UDMA_CHAN_IS_IDLE;
786 
787 	return 0;
788 }
789 
790 static void udma_start_desc(struct udma_chan *uc)
791 {
792 	struct udma_chan_config *ucc = &uc->config;
793 
794 	if (ucc->pkt_mode && (uc->cyclic || ucc->dir == DMA_DEV_TO_MEM)) {
795 		int i;
796 
797 		/* Push all descriptors to ring for packet mode cyclic or RX */
798 		for (i = 0; i < uc->desc->sglen; i++)
799 			udma_push_to_ring(uc, i);
800 	} else {
801 		udma_push_to_ring(uc, 0);
802 	}
803 }
804 
805 static bool udma_chan_needs_reconfiguration(struct udma_chan *uc)
806 {
807 	/* Only PDMAs have staticTR */
808 	if (uc->config.ep_type == PSIL_EP_NATIVE)
809 		return false;
810 
811 	/* Check if the staticTR configuration has changed for TX */
812 	if (memcmp(&uc->static_tr, &uc->desc->static_tr, sizeof(uc->static_tr)))
813 		return true;
814 
815 	return false;
816 }
817 
818 static int udma_start(struct udma_chan *uc)
819 {
820 	struct virt_dma_desc *vd = vchan_next_desc(&uc->vc);
821 
822 	if (!vd) {
823 		uc->desc = NULL;
824 		return -ENOENT;
825 	}
826 
827 	list_del(&vd->node);
828 
829 	uc->desc = to_udma_desc(&vd->tx);
830 
831 	/* Channel is already running and does not need reconfiguration */
832 	if (udma_is_chan_running(uc) && !udma_chan_needs_reconfiguration(uc)) {
833 		udma_start_desc(uc);
834 		goto out;
835 	}
836 
837 	/* Make sure that we clear the teardown bit, if it is set */
838 	udma_reset_chan(uc, false);
839 
840 	/* Push descriptors before we start the channel */
841 	udma_start_desc(uc);
842 
843 	switch (uc->desc->dir) {
844 	case DMA_DEV_TO_MEM:
845 		/* Config remote TR */
846 		if (uc->config.ep_type == PSIL_EP_PDMA_XY) {
847 			u32 val = PDMA_STATIC_TR_Y(uc->desc->static_tr.elcnt) |
848 				  PDMA_STATIC_TR_X(uc->desc->static_tr.elsize);
849 			const struct udma_match_data *match_data =
850 							uc->ud->match_data;
851 
852 			if (uc->config.enable_acc32)
853 				val |= PDMA_STATIC_TR_XY_ACC32;
854 			if (uc->config.enable_burst)
855 				val |= PDMA_STATIC_TR_XY_BURST;
856 
857 			udma_rchanrt_write(uc->rchan,
858 				UDMA_RCHAN_RT_PEER_STATIC_TR_XY_REG, val);
859 
860 			udma_rchanrt_write(uc->rchan,
861 				UDMA_RCHAN_RT_PEER_STATIC_TR_Z_REG,
862 				PDMA_STATIC_TR_Z(uc->desc->static_tr.bstcnt,
863 						 match_data->statictr_z_mask));
864 
865 			/* save the current staticTR configuration */
866 			memcpy(&uc->static_tr, &uc->desc->static_tr,
867 			       sizeof(uc->static_tr));
868 		}
869 
870 		udma_rchanrt_write(uc->rchan, UDMA_RCHAN_RT_CTL_REG,
871 				   UDMA_CHAN_RT_CTL_EN);
872 
873 		/* Enable remote */
874 		udma_rchanrt_write(uc->rchan, UDMA_RCHAN_RT_PEER_RT_EN_REG,
875 				   UDMA_PEER_RT_EN_ENABLE);
876 
877 		break;
878 	case DMA_MEM_TO_DEV:
879 		/* Config remote TR */
880 		if (uc->config.ep_type == PSIL_EP_PDMA_XY) {
881 			u32 val = PDMA_STATIC_TR_Y(uc->desc->static_tr.elcnt) |
882 				  PDMA_STATIC_TR_X(uc->desc->static_tr.elsize);
883 
884 			if (uc->config.enable_acc32)
885 				val |= PDMA_STATIC_TR_XY_ACC32;
886 			if (uc->config.enable_burst)
887 				val |= PDMA_STATIC_TR_XY_BURST;
888 
889 			udma_tchanrt_write(uc->tchan,
890 				UDMA_TCHAN_RT_PEER_STATIC_TR_XY_REG, val);
891 
892 			/* save the current staticTR configuration */
893 			memcpy(&uc->static_tr, &uc->desc->static_tr,
894 			       sizeof(uc->static_tr));
895 		}
896 
897 		/* Enable remote */
898 		udma_tchanrt_write(uc->tchan, UDMA_TCHAN_RT_PEER_RT_EN_REG,
899 				   UDMA_PEER_RT_EN_ENABLE);
900 
901 		udma_tchanrt_write(uc->tchan, UDMA_TCHAN_RT_CTL_REG,
902 				   UDMA_CHAN_RT_CTL_EN);
903 
904 		break;
905 	case DMA_MEM_TO_MEM:
906 		udma_rchanrt_write(uc->rchan, UDMA_RCHAN_RT_CTL_REG,
907 				   UDMA_CHAN_RT_CTL_EN);
908 		udma_tchanrt_write(uc->tchan, UDMA_TCHAN_RT_CTL_REG,
909 				   UDMA_CHAN_RT_CTL_EN);
910 
911 		break;
912 	default:
913 		return -EINVAL;
914 	}
915 
916 	uc->state = UDMA_CHAN_IS_ACTIVE;
917 out:
918 
919 	return 0;
920 }
921 
922 static int udma_stop(struct udma_chan *uc)
923 {
924 	enum udma_chan_state old_state = uc->state;
925 
926 	uc->state = UDMA_CHAN_IS_TERMINATING;
927 	reinit_completion(&uc->teardown_completed);
928 
929 	switch (uc->config.dir) {
930 	case DMA_DEV_TO_MEM:
931 		if (!uc->cyclic && !uc->desc)
932 			udma_push_to_ring(uc, -1);
933 
934 		udma_rchanrt_write(uc->rchan, UDMA_RCHAN_RT_PEER_RT_EN_REG,
935 				   UDMA_PEER_RT_EN_ENABLE |
936 				   UDMA_PEER_RT_EN_TEARDOWN);
937 		break;
938 	case DMA_MEM_TO_DEV:
939 		udma_tchanrt_write(uc->tchan, UDMA_TCHAN_RT_PEER_RT_EN_REG,
940 				   UDMA_PEER_RT_EN_ENABLE |
941 				   UDMA_PEER_RT_EN_FLUSH);
942 		udma_tchanrt_write(uc->tchan, UDMA_TCHAN_RT_CTL_REG,
943 				   UDMA_CHAN_RT_CTL_EN |
944 				   UDMA_CHAN_RT_CTL_TDOWN);
945 		break;
946 	case DMA_MEM_TO_MEM:
947 		udma_tchanrt_write(uc->tchan, UDMA_TCHAN_RT_CTL_REG,
948 				   UDMA_CHAN_RT_CTL_EN |
949 				   UDMA_CHAN_RT_CTL_TDOWN);
950 		break;
951 	default:
952 		uc->state = old_state;
953 		complete_all(&uc->teardown_completed);
954 		return -EINVAL;
955 	}
956 
957 	return 0;
958 }
959 
960 static void udma_cyclic_packet_elapsed(struct udma_chan *uc)
961 {
962 	struct udma_desc *d = uc->desc;
963 	struct cppi5_host_desc_t *h_desc;
964 
965 	h_desc = d->hwdesc[d->desc_idx].cppi5_desc_vaddr;
966 	cppi5_hdesc_reset_to_original(h_desc);
967 	udma_push_to_ring(uc, d->desc_idx);
968 	d->desc_idx = (d->desc_idx + 1) % d->sglen;
969 }
970 
971 static inline void udma_fetch_epib(struct udma_chan *uc, struct udma_desc *d)
972 {
973 	struct cppi5_host_desc_t *h_desc = d->hwdesc[0].cppi5_desc_vaddr;
974 
975 	memcpy(d->metadata, h_desc->epib, d->metadata_size);
976 }
977 
978 static bool udma_is_desc_really_done(struct udma_chan *uc, struct udma_desc *d)
979 {
980 	u32 peer_bcnt, bcnt;
981 
982 	/* Only TX towards PDMA is affected */
983 	if (uc->config.ep_type == PSIL_EP_NATIVE ||
984 	    uc->config.dir != DMA_MEM_TO_DEV)
985 		return true;
986 
987 	peer_bcnt = udma_tchanrt_read(uc->tchan, UDMA_TCHAN_RT_PEER_BCNT_REG);
988 	bcnt = udma_tchanrt_read(uc->tchan, UDMA_TCHAN_RT_BCNT_REG);
989 
990 	/* Transfer is incomplete, store current residue and time stamp */
991 	if (peer_bcnt < bcnt) {
992 		uc->tx_drain.residue = bcnt - peer_bcnt;
993 		uc->tx_drain.tstamp = ktime_get();
994 		return false;
995 	}
996 
997 	return true;
998 }
999 
1000 static void udma_check_tx_completion(struct work_struct *work)
1001 {
1002 	struct udma_chan *uc = container_of(work, typeof(*uc),
1003 					    tx_drain.work.work);
1004 	bool desc_done = true;
1005 	u32 residue_diff;
1006 	ktime_t time_diff;
1007 	unsigned long delay;
1008 
1009 	while (1) {
1010 		if (uc->desc) {
1011 			/* Get previous residue and time stamp */
1012 			residue_diff = uc->tx_drain.residue;
1013 			time_diff = uc->tx_drain.tstamp;
1014 			/*
1015 			 * Get current residue and time stamp or see if
1016 			 * transfer is complete
1017 			 */
1018 			desc_done = udma_is_desc_really_done(uc, uc->desc);
1019 		}
1020 
1021 		if (!desc_done) {
1022 			/*
1023 			 * Find the time delta and residue delta w.r.t
1024 			 * previous poll
1025 			 */
1026 			time_diff = ktime_sub(uc->tx_drain.tstamp,
1027 					      time_diff) + 1;
1028 			residue_diff -= uc->tx_drain.residue;
1029 			if (residue_diff) {
1030 				/*
1031 				 * Try to guess when we should check
1032 				 * next time by calculating rate at
1033 				 * which data is being drained at the
1034 				 * peer device
1035 				 */
1036 				delay = (time_diff / residue_diff) *
1037 					uc->tx_drain.residue;
1038 			} else {
1039 				/* No progress, check again in 1 second  */
1040 				schedule_delayed_work(&uc->tx_drain.work, HZ);
1041 				break;
1042 			}
1043 
1044 			usleep_range(ktime_to_us(delay),
1045 				     ktime_to_us(delay) + 10);
1046 			continue;
1047 		}
1048 
1049 		if (uc->desc) {
1050 			struct udma_desc *d = uc->desc;
1051 
1052 			uc->bcnt += d->residue;
1053 			udma_start(uc);
1054 			vchan_cookie_complete(&d->vd);
1055 			break;
1056 		}
1057 
1058 		break;
1059 	}
1060 }
1061 
1062 static irqreturn_t udma_ring_irq_handler(int irq, void *data)
1063 {
1064 	struct udma_chan *uc = data;
1065 	struct udma_desc *d;
1066 	unsigned long flags;
1067 	dma_addr_t paddr = 0;
1068 
1069 	if (udma_pop_from_ring(uc, &paddr) || !paddr)
1070 		return IRQ_HANDLED;
1071 
1072 	spin_lock_irqsave(&uc->vc.lock, flags);
1073 
1074 	/* Teardown completion message */
1075 	if (cppi5_desc_is_tdcm(paddr)) {
1076 		/* Compensate our internal pop/push counter */
1077 		uc->in_ring_cnt++;
1078 
1079 		complete_all(&uc->teardown_completed);
1080 
1081 		if (uc->terminated_desc) {
1082 			udma_desc_free(&uc->terminated_desc->vd);
1083 			uc->terminated_desc = NULL;
1084 		}
1085 
1086 		if (!uc->desc)
1087 			udma_start(uc);
1088 
1089 		goto out;
1090 	}
1091 
1092 	d = udma_udma_desc_from_paddr(uc, paddr);
1093 
1094 	if (d) {
1095 		dma_addr_t desc_paddr = udma_curr_cppi5_desc_paddr(d,
1096 								   d->desc_idx);
1097 		if (desc_paddr != paddr) {
1098 			dev_err(uc->ud->dev, "not matching descriptors!\n");
1099 			goto out;
1100 		}
1101 
1102 		if (d == uc->desc) {
1103 			/* active descriptor */
1104 			if (uc->cyclic) {
1105 				udma_cyclic_packet_elapsed(uc);
1106 				vchan_cyclic_callback(&d->vd);
1107 			} else {
1108 				if (udma_is_desc_really_done(uc, d)) {
1109 					uc->bcnt += d->residue;
1110 					udma_start(uc);
1111 					vchan_cookie_complete(&d->vd);
1112 				} else {
1113 					schedule_delayed_work(&uc->tx_drain.work,
1114 							      0);
1115 				}
1116 			}
1117 		} else {
1118 			/*
1119 			 * terminated descriptor, mark the descriptor as
1120 			 * completed to update the channel's cookie marker
1121 			 */
1122 			dma_cookie_complete(&d->vd.tx);
1123 		}
1124 	}
1125 out:
1126 	spin_unlock_irqrestore(&uc->vc.lock, flags);
1127 
1128 	return IRQ_HANDLED;
1129 }
1130 
1131 static irqreturn_t udma_udma_irq_handler(int irq, void *data)
1132 {
1133 	struct udma_chan *uc = data;
1134 	struct udma_desc *d;
1135 	unsigned long flags;
1136 
1137 	spin_lock_irqsave(&uc->vc.lock, flags);
1138 	d = uc->desc;
1139 	if (d) {
1140 		d->tr_idx = (d->tr_idx + 1) % d->sglen;
1141 
1142 		if (uc->cyclic) {
1143 			vchan_cyclic_callback(&d->vd);
1144 		} else {
1145 			/* TODO: figure out the real amount of data */
1146 			uc->bcnt += d->residue;
1147 			udma_start(uc);
1148 			vchan_cookie_complete(&d->vd);
1149 		}
1150 	}
1151 
1152 	spin_unlock_irqrestore(&uc->vc.lock, flags);
1153 
1154 	return IRQ_HANDLED;
1155 }
1156 
1157 /**
1158  * __udma_alloc_gp_rflow_range - alloc range of GP RX flows
1159  * @ud: UDMA device
1160  * @from: Start the search from this flow id number
1161  * @cnt: Number of consecutive flow ids to allocate
1162  *
1163  * Allocate range of RX flow ids for future use, those flows can be requested
1164  * only using explicit flow id number. if @from is set to -1 it will try to find
1165  * first free range. if @from is positive value it will force allocation only
1166  * of the specified range of flows.
1167  *
1168  * Returns -ENOMEM if can't find free range.
1169  * -EEXIST if requested range is busy.
1170  * -EINVAL if wrong input values passed.
1171  * Returns flow id on success.
1172  */
1173 static int __udma_alloc_gp_rflow_range(struct udma_dev *ud, int from, int cnt)
1174 {
1175 	int start, tmp_from;
1176 	DECLARE_BITMAP(tmp, K3_UDMA_MAX_RFLOWS);
1177 
1178 	tmp_from = from;
1179 	if (tmp_from < 0)
1180 		tmp_from = ud->rchan_cnt;
1181 	/* default flows can't be allocated and accessible only by id */
1182 	if (tmp_from < ud->rchan_cnt)
1183 		return -EINVAL;
1184 
1185 	if (tmp_from + cnt > ud->rflow_cnt)
1186 		return -EINVAL;
1187 
1188 	bitmap_or(tmp, ud->rflow_gp_map, ud->rflow_gp_map_allocated,
1189 		  ud->rflow_cnt);
1190 
1191 	start = bitmap_find_next_zero_area(tmp,
1192 					   ud->rflow_cnt,
1193 					   tmp_from, cnt, 0);
1194 	if (start >= ud->rflow_cnt)
1195 		return -ENOMEM;
1196 
1197 	if (from >= 0 && start != from)
1198 		return -EEXIST;
1199 
1200 	bitmap_set(ud->rflow_gp_map_allocated, start, cnt);
1201 	return start;
1202 }
1203 
1204 static int __udma_free_gp_rflow_range(struct udma_dev *ud, int from, int cnt)
1205 {
1206 	if (from < ud->rchan_cnt)
1207 		return -EINVAL;
1208 	if (from + cnt > ud->rflow_cnt)
1209 		return -EINVAL;
1210 
1211 	bitmap_clear(ud->rflow_gp_map_allocated, from, cnt);
1212 	return 0;
1213 }
1214 
1215 static struct udma_rflow *__udma_get_rflow(struct udma_dev *ud, int id)
1216 {
1217 	/*
1218 	 * Attempt to request rflow by ID can be made for any rflow
1219 	 * if not in use with assumption that caller knows what's doing.
1220 	 * TI-SCI FW will perform additional permission check ant way, it's
1221 	 * safe
1222 	 */
1223 
1224 	if (id < 0 || id >= ud->rflow_cnt)
1225 		return ERR_PTR(-ENOENT);
1226 
1227 	if (test_bit(id, ud->rflow_in_use))
1228 		return ERR_PTR(-ENOENT);
1229 
1230 	/* GP rflow has to be allocated first */
1231 	if (!test_bit(id, ud->rflow_gp_map) &&
1232 	    !test_bit(id, ud->rflow_gp_map_allocated))
1233 		return ERR_PTR(-EINVAL);
1234 
1235 	dev_dbg(ud->dev, "get rflow%d\n", id);
1236 	set_bit(id, ud->rflow_in_use);
1237 	return &ud->rflows[id];
1238 }
1239 
1240 static void __udma_put_rflow(struct udma_dev *ud, struct udma_rflow *rflow)
1241 {
1242 	if (!test_bit(rflow->id, ud->rflow_in_use)) {
1243 		dev_err(ud->dev, "attempt to put unused rflow%d\n", rflow->id);
1244 		return;
1245 	}
1246 
1247 	dev_dbg(ud->dev, "put rflow%d\n", rflow->id);
1248 	clear_bit(rflow->id, ud->rflow_in_use);
1249 }
1250 
1251 #define UDMA_RESERVE_RESOURCE(res)					\
1252 static struct udma_##res *__udma_reserve_##res(struct udma_dev *ud,	\
1253 					       enum udma_tp_level tpl,	\
1254 					       int id)			\
1255 {									\
1256 	if (id >= 0) {							\
1257 		if (test_bit(id, ud->res##_map)) {			\
1258 			dev_err(ud->dev, "res##%d is in use\n", id);	\
1259 			return ERR_PTR(-ENOENT);			\
1260 		}							\
1261 	} else {							\
1262 		int start;						\
1263 									\
1264 		if (tpl >= ud->match_data->tpl_levels)			\
1265 			tpl = ud->match_data->tpl_levels - 1;		\
1266 									\
1267 		start = ud->match_data->level_start_idx[tpl];		\
1268 									\
1269 		id = find_next_zero_bit(ud->res##_map, ud->res##_cnt,	\
1270 					start);				\
1271 		if (id == ud->res##_cnt) {				\
1272 			return ERR_PTR(-ENOENT);			\
1273 		}							\
1274 	}								\
1275 									\
1276 	set_bit(id, ud->res##_map);					\
1277 	return &ud->res##s[id];						\
1278 }
1279 
1280 UDMA_RESERVE_RESOURCE(tchan);
1281 UDMA_RESERVE_RESOURCE(rchan);
1282 
1283 static int udma_get_tchan(struct udma_chan *uc)
1284 {
1285 	struct udma_dev *ud = uc->ud;
1286 
1287 	if (uc->tchan) {
1288 		dev_dbg(ud->dev, "chan%d: already have tchan%d allocated\n",
1289 			uc->id, uc->tchan->id);
1290 		return 0;
1291 	}
1292 
1293 	uc->tchan = __udma_reserve_tchan(ud, uc->config.channel_tpl, -1);
1294 	if (IS_ERR(uc->tchan))
1295 		return PTR_ERR(uc->tchan);
1296 
1297 	return 0;
1298 }
1299 
1300 static int udma_get_rchan(struct udma_chan *uc)
1301 {
1302 	struct udma_dev *ud = uc->ud;
1303 
1304 	if (uc->rchan) {
1305 		dev_dbg(ud->dev, "chan%d: already have rchan%d allocated\n",
1306 			uc->id, uc->rchan->id);
1307 		return 0;
1308 	}
1309 
1310 	uc->rchan = __udma_reserve_rchan(ud, uc->config.channel_tpl, -1);
1311 	if (IS_ERR(uc->rchan))
1312 		return PTR_ERR(uc->rchan);
1313 
1314 	return 0;
1315 }
1316 
1317 static int udma_get_chan_pair(struct udma_chan *uc)
1318 {
1319 	struct udma_dev *ud = uc->ud;
1320 	const struct udma_match_data *match_data = ud->match_data;
1321 	int chan_id, end;
1322 
1323 	if ((uc->tchan && uc->rchan) && uc->tchan->id == uc->rchan->id) {
1324 		dev_info(ud->dev, "chan%d: already have %d pair allocated\n",
1325 			 uc->id, uc->tchan->id);
1326 		return 0;
1327 	}
1328 
1329 	if (uc->tchan) {
1330 		dev_err(ud->dev, "chan%d: already have tchan%d allocated\n",
1331 			uc->id, uc->tchan->id);
1332 		return -EBUSY;
1333 	} else if (uc->rchan) {
1334 		dev_err(ud->dev, "chan%d: already have rchan%d allocated\n",
1335 			uc->id, uc->rchan->id);
1336 		return -EBUSY;
1337 	}
1338 
1339 	/* Can be optimized, but let's have it like this for now */
1340 	end = min(ud->tchan_cnt, ud->rchan_cnt);
1341 	/* Try to use the highest TPL channel pair for MEM_TO_MEM channels */
1342 	chan_id = match_data->level_start_idx[match_data->tpl_levels - 1];
1343 	for (; chan_id < end; chan_id++) {
1344 		if (!test_bit(chan_id, ud->tchan_map) &&
1345 		    !test_bit(chan_id, ud->rchan_map))
1346 			break;
1347 	}
1348 
1349 	if (chan_id == end)
1350 		return -ENOENT;
1351 
1352 	set_bit(chan_id, ud->tchan_map);
1353 	set_bit(chan_id, ud->rchan_map);
1354 	uc->tchan = &ud->tchans[chan_id];
1355 	uc->rchan = &ud->rchans[chan_id];
1356 
1357 	return 0;
1358 }
1359 
1360 static int udma_get_rflow(struct udma_chan *uc, int flow_id)
1361 {
1362 	struct udma_dev *ud = uc->ud;
1363 
1364 	if (!uc->rchan) {
1365 		dev_err(ud->dev, "chan%d: does not have rchan??\n", uc->id);
1366 		return -EINVAL;
1367 	}
1368 
1369 	if (uc->rflow) {
1370 		dev_dbg(ud->dev, "chan%d: already have rflow%d allocated\n",
1371 			uc->id, uc->rflow->id);
1372 		return 0;
1373 	}
1374 
1375 	uc->rflow = __udma_get_rflow(ud, flow_id);
1376 	if (IS_ERR(uc->rflow))
1377 		return PTR_ERR(uc->rflow);
1378 
1379 	return 0;
1380 }
1381 
1382 static void udma_put_rchan(struct udma_chan *uc)
1383 {
1384 	struct udma_dev *ud = uc->ud;
1385 
1386 	if (uc->rchan) {
1387 		dev_dbg(ud->dev, "chan%d: put rchan%d\n", uc->id,
1388 			uc->rchan->id);
1389 		clear_bit(uc->rchan->id, ud->rchan_map);
1390 		uc->rchan = NULL;
1391 	}
1392 }
1393 
1394 static void udma_put_tchan(struct udma_chan *uc)
1395 {
1396 	struct udma_dev *ud = uc->ud;
1397 
1398 	if (uc->tchan) {
1399 		dev_dbg(ud->dev, "chan%d: put tchan%d\n", uc->id,
1400 			uc->tchan->id);
1401 		clear_bit(uc->tchan->id, ud->tchan_map);
1402 		uc->tchan = NULL;
1403 	}
1404 }
1405 
1406 static void udma_put_rflow(struct udma_chan *uc)
1407 {
1408 	struct udma_dev *ud = uc->ud;
1409 
1410 	if (uc->rflow) {
1411 		dev_dbg(ud->dev, "chan%d: put rflow%d\n", uc->id,
1412 			uc->rflow->id);
1413 		__udma_put_rflow(ud, uc->rflow);
1414 		uc->rflow = NULL;
1415 	}
1416 }
1417 
1418 static void udma_free_tx_resources(struct udma_chan *uc)
1419 {
1420 	if (!uc->tchan)
1421 		return;
1422 
1423 	k3_ringacc_ring_free(uc->tchan->t_ring);
1424 	k3_ringacc_ring_free(uc->tchan->tc_ring);
1425 	uc->tchan->t_ring = NULL;
1426 	uc->tchan->tc_ring = NULL;
1427 
1428 	udma_put_tchan(uc);
1429 }
1430 
1431 static int udma_alloc_tx_resources(struct udma_chan *uc)
1432 {
1433 	struct k3_ring_cfg ring_cfg;
1434 	struct udma_dev *ud = uc->ud;
1435 	int ret;
1436 
1437 	ret = udma_get_tchan(uc);
1438 	if (ret)
1439 		return ret;
1440 
1441 	uc->tchan->t_ring = k3_ringacc_request_ring(ud->ringacc,
1442 						    uc->tchan->id, 0);
1443 	if (!uc->tchan->t_ring) {
1444 		ret = -EBUSY;
1445 		goto err_tx_ring;
1446 	}
1447 
1448 	uc->tchan->tc_ring = k3_ringacc_request_ring(ud->ringacc, -1, 0);
1449 	if (!uc->tchan->tc_ring) {
1450 		ret = -EBUSY;
1451 		goto err_txc_ring;
1452 	}
1453 
1454 	memset(&ring_cfg, 0, sizeof(ring_cfg));
1455 	ring_cfg.size = K3_UDMA_DEFAULT_RING_SIZE;
1456 	ring_cfg.elm_size = K3_RINGACC_RING_ELSIZE_8;
1457 	ring_cfg.mode = K3_RINGACC_RING_MODE_MESSAGE;
1458 
1459 	ret = k3_ringacc_ring_cfg(uc->tchan->t_ring, &ring_cfg);
1460 	ret |= k3_ringacc_ring_cfg(uc->tchan->tc_ring, &ring_cfg);
1461 
1462 	if (ret)
1463 		goto err_ringcfg;
1464 
1465 	return 0;
1466 
1467 err_ringcfg:
1468 	k3_ringacc_ring_free(uc->tchan->tc_ring);
1469 	uc->tchan->tc_ring = NULL;
1470 err_txc_ring:
1471 	k3_ringacc_ring_free(uc->tchan->t_ring);
1472 	uc->tchan->t_ring = NULL;
1473 err_tx_ring:
1474 	udma_put_tchan(uc);
1475 
1476 	return ret;
1477 }
1478 
1479 static void udma_free_rx_resources(struct udma_chan *uc)
1480 {
1481 	if (!uc->rchan)
1482 		return;
1483 
1484 	if (uc->rflow) {
1485 		struct udma_rflow *rflow = uc->rflow;
1486 
1487 		k3_ringacc_ring_free(rflow->fd_ring);
1488 		k3_ringacc_ring_free(rflow->r_ring);
1489 		rflow->fd_ring = NULL;
1490 		rflow->r_ring = NULL;
1491 
1492 		udma_put_rflow(uc);
1493 	}
1494 
1495 	udma_put_rchan(uc);
1496 }
1497 
1498 static int udma_alloc_rx_resources(struct udma_chan *uc)
1499 {
1500 	struct udma_dev *ud = uc->ud;
1501 	struct k3_ring_cfg ring_cfg;
1502 	struct udma_rflow *rflow;
1503 	int fd_ring_id;
1504 	int ret;
1505 
1506 	ret = udma_get_rchan(uc);
1507 	if (ret)
1508 		return ret;
1509 
1510 	/* For MEM_TO_MEM we don't need rflow or rings */
1511 	if (uc->config.dir == DMA_MEM_TO_MEM)
1512 		return 0;
1513 
1514 	ret = udma_get_rflow(uc, uc->rchan->id);
1515 	if (ret) {
1516 		ret = -EBUSY;
1517 		goto err_rflow;
1518 	}
1519 
1520 	rflow = uc->rflow;
1521 	fd_ring_id = ud->tchan_cnt + ud->echan_cnt + uc->rchan->id;
1522 	rflow->fd_ring = k3_ringacc_request_ring(ud->ringacc, fd_ring_id, 0);
1523 	if (!rflow->fd_ring) {
1524 		ret = -EBUSY;
1525 		goto err_rx_ring;
1526 	}
1527 
1528 	rflow->r_ring = k3_ringacc_request_ring(ud->ringacc, -1, 0);
1529 	if (!rflow->r_ring) {
1530 		ret = -EBUSY;
1531 		goto err_rxc_ring;
1532 	}
1533 
1534 	memset(&ring_cfg, 0, sizeof(ring_cfg));
1535 
1536 	if (uc->config.pkt_mode)
1537 		ring_cfg.size = SG_MAX_SEGMENTS;
1538 	else
1539 		ring_cfg.size = K3_UDMA_DEFAULT_RING_SIZE;
1540 
1541 	ring_cfg.elm_size = K3_RINGACC_RING_ELSIZE_8;
1542 	ring_cfg.mode = K3_RINGACC_RING_MODE_MESSAGE;
1543 
1544 	ret = k3_ringacc_ring_cfg(rflow->fd_ring, &ring_cfg);
1545 	ring_cfg.size = K3_UDMA_DEFAULT_RING_SIZE;
1546 	ret |= k3_ringacc_ring_cfg(rflow->r_ring, &ring_cfg);
1547 
1548 	if (ret)
1549 		goto err_ringcfg;
1550 
1551 	return 0;
1552 
1553 err_ringcfg:
1554 	k3_ringacc_ring_free(rflow->r_ring);
1555 	rflow->r_ring = NULL;
1556 err_rxc_ring:
1557 	k3_ringacc_ring_free(rflow->fd_ring);
1558 	rflow->fd_ring = NULL;
1559 err_rx_ring:
1560 	udma_put_rflow(uc);
1561 err_rflow:
1562 	udma_put_rchan(uc);
1563 
1564 	return ret;
1565 }
1566 
1567 #define TISCI_TCHAN_VALID_PARAMS (				\
1568 	TI_SCI_MSG_VALUE_RM_UDMAP_CH_PAUSE_ON_ERR_VALID |	\
1569 	TI_SCI_MSG_VALUE_RM_UDMAP_CH_TX_FILT_EINFO_VALID |	\
1570 	TI_SCI_MSG_VALUE_RM_UDMAP_CH_TX_FILT_PSWORDS_VALID |	\
1571 	TI_SCI_MSG_VALUE_RM_UDMAP_CH_CHAN_TYPE_VALID |		\
1572 	TI_SCI_MSG_VALUE_RM_UDMAP_CH_TX_SUPR_TDPKT_VALID |	\
1573 	TI_SCI_MSG_VALUE_RM_UDMAP_CH_FETCH_SIZE_VALID |		\
1574 	TI_SCI_MSG_VALUE_RM_UDMAP_CH_CQ_QNUM_VALID |		\
1575 	TI_SCI_MSG_VALUE_RM_UDMAP_CH_ATYPE_VALID)
1576 
1577 #define TISCI_RCHAN_VALID_PARAMS (				\
1578 	TI_SCI_MSG_VALUE_RM_UDMAP_CH_PAUSE_ON_ERR_VALID |	\
1579 	TI_SCI_MSG_VALUE_RM_UDMAP_CH_FETCH_SIZE_VALID |		\
1580 	TI_SCI_MSG_VALUE_RM_UDMAP_CH_CQ_QNUM_VALID |		\
1581 	TI_SCI_MSG_VALUE_RM_UDMAP_CH_CHAN_TYPE_VALID |		\
1582 	TI_SCI_MSG_VALUE_RM_UDMAP_CH_RX_IGNORE_SHORT_VALID |	\
1583 	TI_SCI_MSG_VALUE_RM_UDMAP_CH_RX_IGNORE_LONG_VALID |	\
1584 	TI_SCI_MSG_VALUE_RM_UDMAP_CH_RX_FLOWID_START_VALID |	\
1585 	TI_SCI_MSG_VALUE_RM_UDMAP_CH_RX_FLOWID_CNT_VALID |	\
1586 	TI_SCI_MSG_VALUE_RM_UDMAP_CH_ATYPE_VALID)
1587 
1588 static int udma_tisci_m2m_channel_config(struct udma_chan *uc)
1589 {
1590 	struct udma_dev *ud = uc->ud;
1591 	struct udma_tisci_rm *tisci_rm = &ud->tisci_rm;
1592 	const struct ti_sci_rm_udmap_ops *tisci_ops = tisci_rm->tisci_udmap_ops;
1593 	struct udma_tchan *tchan = uc->tchan;
1594 	struct udma_rchan *rchan = uc->rchan;
1595 	int ret = 0;
1596 
1597 	/* Non synchronized - mem to mem type of transfer */
1598 	int tc_ring = k3_ringacc_get_ring_id(tchan->tc_ring);
1599 	struct ti_sci_msg_rm_udmap_tx_ch_cfg req_tx = { 0 };
1600 	struct ti_sci_msg_rm_udmap_rx_ch_cfg req_rx = { 0 };
1601 
1602 	req_tx.valid_params = TISCI_TCHAN_VALID_PARAMS;
1603 	req_tx.nav_id = tisci_rm->tisci_dev_id;
1604 	req_tx.index = tchan->id;
1605 	req_tx.tx_chan_type = TI_SCI_RM_UDMAP_CHAN_TYPE_3RDP_BCOPY_PBRR;
1606 	req_tx.tx_fetch_size = sizeof(struct cppi5_desc_hdr_t) >> 2;
1607 	req_tx.txcq_qnum = tc_ring;
1608 	req_tx.tx_atype = ud->atype;
1609 
1610 	ret = tisci_ops->tx_ch_cfg(tisci_rm->tisci, &req_tx);
1611 	if (ret) {
1612 		dev_err(ud->dev, "tchan%d cfg failed %d\n", tchan->id, ret);
1613 		return ret;
1614 	}
1615 
1616 	req_rx.valid_params = TISCI_RCHAN_VALID_PARAMS;
1617 	req_rx.nav_id = tisci_rm->tisci_dev_id;
1618 	req_rx.index = rchan->id;
1619 	req_rx.rx_fetch_size = sizeof(struct cppi5_desc_hdr_t) >> 2;
1620 	req_rx.rxcq_qnum = tc_ring;
1621 	req_rx.rx_chan_type = TI_SCI_RM_UDMAP_CHAN_TYPE_3RDP_BCOPY_PBRR;
1622 	req_rx.rx_atype = ud->atype;
1623 
1624 	ret = tisci_ops->rx_ch_cfg(tisci_rm->tisci, &req_rx);
1625 	if (ret)
1626 		dev_err(ud->dev, "rchan%d alloc failed %d\n", rchan->id, ret);
1627 
1628 	return ret;
1629 }
1630 
1631 static int udma_tisci_tx_channel_config(struct udma_chan *uc)
1632 {
1633 	struct udma_dev *ud = uc->ud;
1634 	struct udma_tisci_rm *tisci_rm = &ud->tisci_rm;
1635 	const struct ti_sci_rm_udmap_ops *tisci_ops = tisci_rm->tisci_udmap_ops;
1636 	struct udma_tchan *tchan = uc->tchan;
1637 	int tc_ring = k3_ringacc_get_ring_id(tchan->tc_ring);
1638 	struct ti_sci_msg_rm_udmap_tx_ch_cfg req_tx = { 0 };
1639 	u32 mode, fetch_size;
1640 	int ret = 0;
1641 
1642 	if (uc->config.pkt_mode) {
1643 		mode = TI_SCI_RM_UDMAP_CHAN_TYPE_PKT_PBRR;
1644 		fetch_size = cppi5_hdesc_calc_size(uc->config.needs_epib,
1645 						   uc->config.psd_size, 0);
1646 	} else {
1647 		mode = TI_SCI_RM_UDMAP_CHAN_TYPE_3RDP_PBRR;
1648 		fetch_size = sizeof(struct cppi5_desc_hdr_t);
1649 	}
1650 
1651 	req_tx.valid_params = TISCI_TCHAN_VALID_PARAMS;
1652 	req_tx.nav_id = tisci_rm->tisci_dev_id;
1653 	req_tx.index = tchan->id;
1654 	req_tx.tx_chan_type = mode;
1655 	req_tx.tx_supr_tdpkt = uc->config.notdpkt;
1656 	req_tx.tx_fetch_size = fetch_size >> 2;
1657 	req_tx.txcq_qnum = tc_ring;
1658 	req_tx.tx_atype = uc->config.atype;
1659 
1660 	ret = tisci_ops->tx_ch_cfg(tisci_rm->tisci, &req_tx);
1661 	if (ret)
1662 		dev_err(ud->dev, "tchan%d cfg failed %d\n", tchan->id, ret);
1663 
1664 	return ret;
1665 }
1666 
1667 static int udma_tisci_rx_channel_config(struct udma_chan *uc)
1668 {
1669 	struct udma_dev *ud = uc->ud;
1670 	struct udma_tisci_rm *tisci_rm = &ud->tisci_rm;
1671 	const struct ti_sci_rm_udmap_ops *tisci_ops = tisci_rm->tisci_udmap_ops;
1672 	struct udma_rchan *rchan = uc->rchan;
1673 	int fd_ring = k3_ringacc_get_ring_id(uc->rflow->fd_ring);
1674 	int rx_ring = k3_ringacc_get_ring_id(uc->rflow->r_ring);
1675 	struct ti_sci_msg_rm_udmap_rx_ch_cfg req_rx = { 0 };
1676 	struct ti_sci_msg_rm_udmap_flow_cfg flow_req = { 0 };
1677 	u32 mode, fetch_size;
1678 	int ret = 0;
1679 
1680 	if (uc->config.pkt_mode) {
1681 		mode = TI_SCI_RM_UDMAP_CHAN_TYPE_PKT_PBRR;
1682 		fetch_size = cppi5_hdesc_calc_size(uc->config.needs_epib,
1683 						   uc->config.psd_size, 0);
1684 	} else {
1685 		mode = TI_SCI_RM_UDMAP_CHAN_TYPE_3RDP_PBRR;
1686 		fetch_size = sizeof(struct cppi5_desc_hdr_t);
1687 	}
1688 
1689 	req_rx.valid_params = TISCI_RCHAN_VALID_PARAMS;
1690 	req_rx.nav_id = tisci_rm->tisci_dev_id;
1691 	req_rx.index = rchan->id;
1692 	req_rx.rx_fetch_size =  fetch_size >> 2;
1693 	req_rx.rxcq_qnum = rx_ring;
1694 	req_rx.rx_chan_type = mode;
1695 	req_rx.rx_atype = uc->config.atype;
1696 
1697 	ret = tisci_ops->rx_ch_cfg(tisci_rm->tisci, &req_rx);
1698 	if (ret) {
1699 		dev_err(ud->dev, "rchan%d cfg failed %d\n", rchan->id, ret);
1700 		return ret;
1701 	}
1702 
1703 	flow_req.valid_params =
1704 		TI_SCI_MSG_VALUE_RM_UDMAP_FLOW_EINFO_PRESENT_VALID |
1705 		TI_SCI_MSG_VALUE_RM_UDMAP_FLOW_PSINFO_PRESENT_VALID |
1706 		TI_SCI_MSG_VALUE_RM_UDMAP_FLOW_ERROR_HANDLING_VALID |
1707 		TI_SCI_MSG_VALUE_RM_UDMAP_FLOW_DESC_TYPE_VALID |
1708 		TI_SCI_MSG_VALUE_RM_UDMAP_FLOW_DEST_QNUM_VALID |
1709 		TI_SCI_MSG_VALUE_RM_UDMAP_FLOW_SRC_TAG_HI_SEL_VALID |
1710 		TI_SCI_MSG_VALUE_RM_UDMAP_FLOW_SRC_TAG_LO_SEL_VALID |
1711 		TI_SCI_MSG_VALUE_RM_UDMAP_FLOW_DEST_TAG_HI_SEL_VALID |
1712 		TI_SCI_MSG_VALUE_RM_UDMAP_FLOW_DEST_TAG_LO_SEL_VALID |
1713 		TI_SCI_MSG_VALUE_RM_UDMAP_FLOW_FDQ0_SZ0_QNUM_VALID |
1714 		TI_SCI_MSG_VALUE_RM_UDMAP_FLOW_FDQ1_QNUM_VALID |
1715 		TI_SCI_MSG_VALUE_RM_UDMAP_FLOW_FDQ2_QNUM_VALID |
1716 		TI_SCI_MSG_VALUE_RM_UDMAP_FLOW_FDQ3_QNUM_VALID;
1717 
1718 	flow_req.nav_id = tisci_rm->tisci_dev_id;
1719 	flow_req.flow_index = rchan->id;
1720 
1721 	if (uc->config.needs_epib)
1722 		flow_req.rx_einfo_present = 1;
1723 	else
1724 		flow_req.rx_einfo_present = 0;
1725 	if (uc->config.psd_size)
1726 		flow_req.rx_psinfo_present = 1;
1727 	else
1728 		flow_req.rx_psinfo_present = 0;
1729 	flow_req.rx_error_handling = 1;
1730 	flow_req.rx_dest_qnum = rx_ring;
1731 	flow_req.rx_src_tag_hi_sel = UDMA_RFLOW_SRCTAG_NONE;
1732 	flow_req.rx_src_tag_lo_sel = UDMA_RFLOW_SRCTAG_SRC_TAG;
1733 	flow_req.rx_dest_tag_hi_sel = UDMA_RFLOW_DSTTAG_DST_TAG_HI;
1734 	flow_req.rx_dest_tag_lo_sel = UDMA_RFLOW_DSTTAG_DST_TAG_LO;
1735 	flow_req.rx_fdq0_sz0_qnum = fd_ring;
1736 	flow_req.rx_fdq1_qnum = fd_ring;
1737 	flow_req.rx_fdq2_qnum = fd_ring;
1738 	flow_req.rx_fdq3_qnum = fd_ring;
1739 
1740 	ret = tisci_ops->rx_flow_cfg(tisci_rm->tisci, &flow_req);
1741 
1742 	if (ret)
1743 		dev_err(ud->dev, "flow%d config failed: %d\n", rchan->id, ret);
1744 
1745 	return 0;
1746 }
1747 
1748 static int udma_alloc_chan_resources(struct dma_chan *chan)
1749 {
1750 	struct udma_chan *uc = to_udma_chan(chan);
1751 	struct udma_dev *ud = to_udma_dev(chan->device);
1752 	const struct udma_match_data *match_data = ud->match_data;
1753 	struct k3_ring *irq_ring;
1754 	u32 irq_udma_idx;
1755 	int ret;
1756 
1757 	if (uc->config.pkt_mode || uc->config.dir == DMA_MEM_TO_MEM) {
1758 		uc->use_dma_pool = true;
1759 		/* in case of MEM_TO_MEM we have maximum of two TRs */
1760 		if (uc->config.dir == DMA_MEM_TO_MEM) {
1761 			uc->config.hdesc_size = cppi5_trdesc_calc_size(
1762 					sizeof(struct cppi5_tr_type15_t), 2);
1763 			uc->config.pkt_mode = false;
1764 		}
1765 	}
1766 
1767 	if (uc->use_dma_pool) {
1768 		uc->hdesc_pool = dma_pool_create(uc->name, ud->ddev.dev,
1769 						 uc->config.hdesc_size,
1770 						 ud->desc_align,
1771 						 0);
1772 		if (!uc->hdesc_pool) {
1773 			dev_err(ud->ddev.dev,
1774 				"Descriptor pool allocation failed\n");
1775 			uc->use_dma_pool = false;
1776 			return -ENOMEM;
1777 		}
1778 	}
1779 
1780 	/*
1781 	 * Make sure that the completion is in a known state:
1782 	 * No teardown, the channel is idle
1783 	 */
1784 	reinit_completion(&uc->teardown_completed);
1785 	complete_all(&uc->teardown_completed);
1786 	uc->state = UDMA_CHAN_IS_IDLE;
1787 
1788 	switch (uc->config.dir) {
1789 	case DMA_MEM_TO_MEM:
1790 		/* Non synchronized - mem to mem type of transfer */
1791 		dev_dbg(uc->ud->dev, "%s: chan%d as MEM-to-MEM\n", __func__,
1792 			uc->id);
1793 
1794 		ret = udma_get_chan_pair(uc);
1795 		if (ret)
1796 			return ret;
1797 
1798 		ret = udma_alloc_tx_resources(uc);
1799 		if (ret)
1800 			return ret;
1801 
1802 		ret = udma_alloc_rx_resources(uc);
1803 		if (ret) {
1804 			udma_free_tx_resources(uc);
1805 			return ret;
1806 		}
1807 
1808 		uc->config.src_thread = ud->psil_base + uc->tchan->id;
1809 		uc->config.dst_thread = (ud->psil_base + uc->rchan->id) |
1810 					K3_PSIL_DST_THREAD_ID_OFFSET;
1811 
1812 		irq_ring = uc->tchan->tc_ring;
1813 		irq_udma_idx = uc->tchan->id;
1814 
1815 		ret = udma_tisci_m2m_channel_config(uc);
1816 		break;
1817 	case DMA_MEM_TO_DEV:
1818 		/* Slave transfer synchronized - mem to dev (TX) trasnfer */
1819 		dev_dbg(uc->ud->dev, "%s: chan%d as MEM-to-DEV\n", __func__,
1820 			uc->id);
1821 
1822 		ret = udma_alloc_tx_resources(uc);
1823 		if (ret) {
1824 			uc->config.remote_thread_id = -1;
1825 			return ret;
1826 		}
1827 
1828 		uc->config.src_thread = ud->psil_base + uc->tchan->id;
1829 		uc->config.dst_thread = uc->config.remote_thread_id;
1830 		uc->config.dst_thread |= K3_PSIL_DST_THREAD_ID_OFFSET;
1831 
1832 		irq_ring = uc->tchan->tc_ring;
1833 		irq_udma_idx = uc->tchan->id;
1834 
1835 		ret = udma_tisci_tx_channel_config(uc);
1836 		break;
1837 	case DMA_DEV_TO_MEM:
1838 		/* Slave transfer synchronized - dev to mem (RX) trasnfer */
1839 		dev_dbg(uc->ud->dev, "%s: chan%d as DEV-to-MEM\n", __func__,
1840 			uc->id);
1841 
1842 		ret = udma_alloc_rx_resources(uc);
1843 		if (ret) {
1844 			uc->config.remote_thread_id = -1;
1845 			return ret;
1846 		}
1847 
1848 		uc->config.src_thread = uc->config.remote_thread_id;
1849 		uc->config.dst_thread = (ud->psil_base + uc->rchan->id) |
1850 					K3_PSIL_DST_THREAD_ID_OFFSET;
1851 
1852 		irq_ring = uc->rflow->r_ring;
1853 		irq_udma_idx = match_data->rchan_oes_offset + uc->rchan->id;
1854 
1855 		ret = udma_tisci_rx_channel_config(uc);
1856 		break;
1857 	default:
1858 		/* Can not happen */
1859 		dev_err(uc->ud->dev, "%s: chan%d invalid direction (%u)\n",
1860 			__func__, uc->id, uc->config.dir);
1861 		return -EINVAL;
1862 	}
1863 
1864 	/* check if the channel configuration was successful */
1865 	if (ret)
1866 		goto err_res_free;
1867 
1868 	if (udma_is_chan_running(uc)) {
1869 		dev_warn(ud->dev, "chan%d: is running!\n", uc->id);
1870 		udma_stop(uc);
1871 		if (udma_is_chan_running(uc)) {
1872 			dev_err(ud->dev, "chan%d: won't stop!\n", uc->id);
1873 			goto err_res_free;
1874 		}
1875 	}
1876 
1877 	/* PSI-L pairing */
1878 	ret = navss_psil_pair(ud, uc->config.src_thread, uc->config.dst_thread);
1879 	if (ret) {
1880 		dev_err(ud->dev, "PSI-L pairing failed: 0x%04x -> 0x%04x\n",
1881 			uc->config.src_thread, uc->config.dst_thread);
1882 		goto err_res_free;
1883 	}
1884 
1885 	uc->psil_paired = true;
1886 
1887 	uc->irq_num_ring = k3_ringacc_get_ring_irq_num(irq_ring);
1888 	if (uc->irq_num_ring <= 0) {
1889 		dev_err(ud->dev, "Failed to get ring irq (index: %u)\n",
1890 			k3_ringacc_get_ring_id(irq_ring));
1891 		ret = -EINVAL;
1892 		goto err_psi_free;
1893 	}
1894 
1895 	ret = request_irq(uc->irq_num_ring, udma_ring_irq_handler,
1896 			  IRQF_TRIGGER_HIGH, uc->name, uc);
1897 	if (ret) {
1898 		dev_err(ud->dev, "chan%d: ring irq request failed\n", uc->id);
1899 		goto err_irq_free;
1900 	}
1901 
1902 	/* Event from UDMA (TR events) only needed for slave TR mode channels */
1903 	if (is_slave_direction(uc->config.dir) && !uc->config.pkt_mode) {
1904 		uc->irq_num_udma = ti_sci_inta_msi_get_virq(ud->dev,
1905 							    irq_udma_idx);
1906 		if (uc->irq_num_udma <= 0) {
1907 			dev_err(ud->dev, "Failed to get udma irq (index: %u)\n",
1908 				irq_udma_idx);
1909 			free_irq(uc->irq_num_ring, uc);
1910 			ret = -EINVAL;
1911 			goto err_irq_free;
1912 		}
1913 
1914 		ret = request_irq(uc->irq_num_udma, udma_udma_irq_handler, 0,
1915 				  uc->name, uc);
1916 		if (ret) {
1917 			dev_err(ud->dev, "chan%d: UDMA irq request failed\n",
1918 				uc->id);
1919 			free_irq(uc->irq_num_ring, uc);
1920 			goto err_irq_free;
1921 		}
1922 	} else {
1923 		uc->irq_num_udma = 0;
1924 	}
1925 
1926 	udma_reset_rings(uc);
1927 
1928 	INIT_DELAYED_WORK_ONSTACK(&uc->tx_drain.work,
1929 				  udma_check_tx_completion);
1930 	return 0;
1931 
1932 err_irq_free:
1933 	uc->irq_num_ring = 0;
1934 	uc->irq_num_udma = 0;
1935 err_psi_free:
1936 	navss_psil_unpair(ud, uc->config.src_thread, uc->config.dst_thread);
1937 	uc->psil_paired = false;
1938 err_res_free:
1939 	udma_free_tx_resources(uc);
1940 	udma_free_rx_resources(uc);
1941 
1942 	udma_reset_uchan(uc);
1943 
1944 	if (uc->use_dma_pool) {
1945 		dma_pool_destroy(uc->hdesc_pool);
1946 		uc->use_dma_pool = false;
1947 	}
1948 
1949 	return ret;
1950 }
1951 
1952 static int udma_slave_config(struct dma_chan *chan,
1953 			     struct dma_slave_config *cfg)
1954 {
1955 	struct udma_chan *uc = to_udma_chan(chan);
1956 
1957 	memcpy(&uc->cfg, cfg, sizeof(uc->cfg));
1958 
1959 	return 0;
1960 }
1961 
1962 static struct udma_desc *udma_alloc_tr_desc(struct udma_chan *uc,
1963 					    size_t tr_size, int tr_count,
1964 					    enum dma_transfer_direction dir)
1965 {
1966 	struct udma_hwdesc *hwdesc;
1967 	struct cppi5_desc_hdr_t *tr_desc;
1968 	struct udma_desc *d;
1969 	u32 reload_count = 0;
1970 	u32 ring_id;
1971 
1972 	switch (tr_size) {
1973 	case 16:
1974 	case 32:
1975 	case 64:
1976 	case 128:
1977 		break;
1978 	default:
1979 		dev_err(uc->ud->dev, "Unsupported TR size of %zu\n", tr_size);
1980 		return NULL;
1981 	}
1982 
1983 	/* We have only one descriptor containing multiple TRs */
1984 	d = kzalloc(sizeof(*d) + sizeof(d->hwdesc[0]), GFP_NOWAIT);
1985 	if (!d)
1986 		return NULL;
1987 
1988 	d->sglen = tr_count;
1989 
1990 	d->hwdesc_count = 1;
1991 	hwdesc = &d->hwdesc[0];
1992 
1993 	/* Allocate memory for DMA ring descriptor */
1994 	if (uc->use_dma_pool) {
1995 		hwdesc->cppi5_desc_size = uc->config.hdesc_size;
1996 		hwdesc->cppi5_desc_vaddr = dma_pool_zalloc(uc->hdesc_pool,
1997 						GFP_NOWAIT,
1998 						&hwdesc->cppi5_desc_paddr);
1999 	} else {
2000 		hwdesc->cppi5_desc_size = cppi5_trdesc_calc_size(tr_size,
2001 								 tr_count);
2002 		hwdesc->cppi5_desc_size = ALIGN(hwdesc->cppi5_desc_size,
2003 						uc->ud->desc_align);
2004 		hwdesc->cppi5_desc_vaddr = dma_alloc_coherent(uc->ud->dev,
2005 						hwdesc->cppi5_desc_size,
2006 						&hwdesc->cppi5_desc_paddr,
2007 						GFP_NOWAIT);
2008 	}
2009 
2010 	if (!hwdesc->cppi5_desc_vaddr) {
2011 		kfree(d);
2012 		return NULL;
2013 	}
2014 
2015 	/* Start of the TR req records */
2016 	hwdesc->tr_req_base = hwdesc->cppi5_desc_vaddr + tr_size;
2017 	/* Start address of the TR response array */
2018 	hwdesc->tr_resp_base = hwdesc->tr_req_base + tr_size * tr_count;
2019 
2020 	tr_desc = hwdesc->cppi5_desc_vaddr;
2021 
2022 	if (uc->cyclic)
2023 		reload_count = CPPI5_INFO0_TRDESC_RLDCNT_INFINITE;
2024 
2025 	if (dir == DMA_DEV_TO_MEM)
2026 		ring_id = k3_ringacc_get_ring_id(uc->rflow->r_ring);
2027 	else
2028 		ring_id = k3_ringacc_get_ring_id(uc->tchan->tc_ring);
2029 
2030 	cppi5_trdesc_init(tr_desc, tr_count, tr_size, 0, reload_count);
2031 	cppi5_desc_set_pktids(tr_desc, uc->id,
2032 			      CPPI5_INFO1_DESC_FLOWID_DEFAULT);
2033 	cppi5_desc_set_retpolicy(tr_desc, 0, ring_id);
2034 
2035 	return d;
2036 }
2037 
2038 /**
2039  * udma_get_tr_counters - calculate TR counters for a given length
2040  * @len: Length of the trasnfer
2041  * @align_to: Preferred alignment
2042  * @tr0_cnt0: First TR icnt0
2043  * @tr0_cnt1: First TR icnt1
2044  * @tr1_cnt0: Second (if used) TR icnt0
2045  *
2046  * For len < SZ_64K only one TR is enough, tr1_cnt0 is not updated
2047  * For len >= SZ_64K two TRs are used in a simple way:
2048  * First TR: SZ_64K-alignment blocks (tr0_cnt0, tr0_cnt1)
2049  * Second TR: the remaining length (tr1_cnt0)
2050  *
2051  * Returns the number of TRs the length needs (1 or 2)
2052  * -EINVAL if the length can not be supported
2053  */
2054 static int udma_get_tr_counters(size_t len, unsigned long align_to,
2055 				u16 *tr0_cnt0, u16 *tr0_cnt1, u16 *tr1_cnt0)
2056 {
2057 	if (len < SZ_64K) {
2058 		*tr0_cnt0 = len;
2059 		*tr0_cnt1 = 1;
2060 
2061 		return 1;
2062 	}
2063 
2064 	if (align_to > 3)
2065 		align_to = 3;
2066 
2067 realign:
2068 	*tr0_cnt0 = SZ_64K - BIT(align_to);
2069 	if (len / *tr0_cnt0 >= SZ_64K) {
2070 		if (align_to) {
2071 			align_to--;
2072 			goto realign;
2073 		}
2074 		return -EINVAL;
2075 	}
2076 
2077 	*tr0_cnt1 = len / *tr0_cnt0;
2078 	*tr1_cnt0 = len % *tr0_cnt0;
2079 
2080 	return 2;
2081 }
2082 
2083 static struct udma_desc *
2084 udma_prep_slave_sg_tr(struct udma_chan *uc, struct scatterlist *sgl,
2085 		      unsigned int sglen, enum dma_transfer_direction dir,
2086 		      unsigned long tx_flags, void *context)
2087 {
2088 	struct scatterlist *sgent;
2089 	struct udma_desc *d;
2090 	struct cppi5_tr_type1_t *tr_req = NULL;
2091 	u16 tr0_cnt0, tr0_cnt1, tr1_cnt0;
2092 	unsigned int i;
2093 	size_t tr_size;
2094 	int num_tr = 0;
2095 	int tr_idx = 0;
2096 
2097 	if (!is_slave_direction(dir)) {
2098 		dev_err(uc->ud->dev, "Only slave cyclic is supported\n");
2099 		return NULL;
2100 	}
2101 
2102 	/* estimate the number of TRs we will need */
2103 	for_each_sg(sgl, sgent, sglen, i) {
2104 		if (sg_dma_len(sgent) < SZ_64K)
2105 			num_tr++;
2106 		else
2107 			num_tr += 2;
2108 	}
2109 
2110 	/* Now allocate and setup the descriptor. */
2111 	tr_size = sizeof(struct cppi5_tr_type1_t);
2112 	d = udma_alloc_tr_desc(uc, tr_size, num_tr, dir);
2113 	if (!d)
2114 		return NULL;
2115 
2116 	d->sglen = sglen;
2117 
2118 	tr_req = d->hwdesc[0].tr_req_base;
2119 	for_each_sg(sgl, sgent, sglen, i) {
2120 		dma_addr_t sg_addr = sg_dma_address(sgent);
2121 
2122 		num_tr = udma_get_tr_counters(sg_dma_len(sgent), __ffs(sg_addr),
2123 					      &tr0_cnt0, &tr0_cnt1, &tr1_cnt0);
2124 		if (num_tr < 0) {
2125 			dev_err(uc->ud->dev, "size %u is not supported\n",
2126 				sg_dma_len(sgent));
2127 			udma_free_hwdesc(uc, d);
2128 			kfree(d);
2129 			return NULL;
2130 		}
2131 
2132 		cppi5_tr_init(&tr_req[i].flags, CPPI5_TR_TYPE1, false, false,
2133 			      CPPI5_TR_EVENT_SIZE_COMPLETION, 0);
2134 		cppi5_tr_csf_set(&tr_req[i].flags, CPPI5_TR_CSF_SUPR_EVT);
2135 
2136 		tr_req[tr_idx].addr = sg_addr;
2137 		tr_req[tr_idx].icnt0 = tr0_cnt0;
2138 		tr_req[tr_idx].icnt1 = tr0_cnt1;
2139 		tr_req[tr_idx].dim1 = tr0_cnt0;
2140 		tr_idx++;
2141 
2142 		if (num_tr == 2) {
2143 			cppi5_tr_init(&tr_req[tr_idx].flags, CPPI5_TR_TYPE1,
2144 				      false, false,
2145 				      CPPI5_TR_EVENT_SIZE_COMPLETION, 0);
2146 			cppi5_tr_csf_set(&tr_req[tr_idx].flags,
2147 					 CPPI5_TR_CSF_SUPR_EVT);
2148 
2149 			tr_req[tr_idx].addr = sg_addr + tr0_cnt1 * tr0_cnt0;
2150 			tr_req[tr_idx].icnt0 = tr1_cnt0;
2151 			tr_req[tr_idx].icnt1 = 1;
2152 			tr_req[tr_idx].dim1 = tr1_cnt0;
2153 			tr_idx++;
2154 		}
2155 
2156 		d->residue += sg_dma_len(sgent);
2157 	}
2158 
2159 	cppi5_tr_csf_set(&tr_req[tr_idx - 1].flags, CPPI5_TR_CSF_EOP);
2160 
2161 	return d;
2162 }
2163 
2164 static int udma_configure_statictr(struct udma_chan *uc, struct udma_desc *d,
2165 				   enum dma_slave_buswidth dev_width,
2166 				   u16 elcnt)
2167 {
2168 	if (uc->config.ep_type != PSIL_EP_PDMA_XY)
2169 		return 0;
2170 
2171 	/* Bus width translates to the element size (ES) */
2172 	switch (dev_width) {
2173 	case DMA_SLAVE_BUSWIDTH_1_BYTE:
2174 		d->static_tr.elsize = 0;
2175 		break;
2176 	case DMA_SLAVE_BUSWIDTH_2_BYTES:
2177 		d->static_tr.elsize = 1;
2178 		break;
2179 	case DMA_SLAVE_BUSWIDTH_3_BYTES:
2180 		d->static_tr.elsize = 2;
2181 		break;
2182 	case DMA_SLAVE_BUSWIDTH_4_BYTES:
2183 		d->static_tr.elsize = 3;
2184 		break;
2185 	case DMA_SLAVE_BUSWIDTH_8_BYTES:
2186 		d->static_tr.elsize = 4;
2187 		break;
2188 	default: /* not reached */
2189 		return -EINVAL;
2190 	}
2191 
2192 	d->static_tr.elcnt = elcnt;
2193 
2194 	/*
2195 	 * PDMA must to close the packet when the channel is in packet mode.
2196 	 * For TR mode when the channel is not cyclic we also need PDMA to close
2197 	 * the packet otherwise the transfer will stall because PDMA holds on
2198 	 * the data it has received from the peripheral.
2199 	 */
2200 	if (uc->config.pkt_mode || !uc->cyclic) {
2201 		unsigned int div = dev_width * elcnt;
2202 
2203 		if (uc->cyclic)
2204 			d->static_tr.bstcnt = d->residue / d->sglen / div;
2205 		else
2206 			d->static_tr.bstcnt = d->residue / div;
2207 
2208 		if (uc->config.dir == DMA_DEV_TO_MEM &&
2209 		    d->static_tr.bstcnt > uc->ud->match_data->statictr_z_mask)
2210 			return -EINVAL;
2211 	} else {
2212 		d->static_tr.bstcnt = 0;
2213 	}
2214 
2215 	return 0;
2216 }
2217 
2218 static struct udma_desc *
2219 udma_prep_slave_sg_pkt(struct udma_chan *uc, struct scatterlist *sgl,
2220 		       unsigned int sglen, enum dma_transfer_direction dir,
2221 		       unsigned long tx_flags, void *context)
2222 {
2223 	struct scatterlist *sgent;
2224 	struct cppi5_host_desc_t *h_desc = NULL;
2225 	struct udma_desc *d;
2226 	u32 ring_id;
2227 	unsigned int i;
2228 
2229 	d = kzalloc(sizeof(*d) + sglen * sizeof(d->hwdesc[0]), GFP_NOWAIT);
2230 	if (!d)
2231 		return NULL;
2232 
2233 	d->sglen = sglen;
2234 	d->hwdesc_count = sglen;
2235 
2236 	if (dir == DMA_DEV_TO_MEM)
2237 		ring_id = k3_ringacc_get_ring_id(uc->rflow->r_ring);
2238 	else
2239 		ring_id = k3_ringacc_get_ring_id(uc->tchan->tc_ring);
2240 
2241 	for_each_sg(sgl, sgent, sglen, i) {
2242 		struct udma_hwdesc *hwdesc = &d->hwdesc[i];
2243 		dma_addr_t sg_addr = sg_dma_address(sgent);
2244 		struct cppi5_host_desc_t *desc;
2245 		size_t sg_len = sg_dma_len(sgent);
2246 
2247 		hwdesc->cppi5_desc_vaddr = dma_pool_zalloc(uc->hdesc_pool,
2248 						GFP_NOWAIT,
2249 						&hwdesc->cppi5_desc_paddr);
2250 		if (!hwdesc->cppi5_desc_vaddr) {
2251 			dev_err(uc->ud->dev,
2252 				"descriptor%d allocation failed\n", i);
2253 
2254 			udma_free_hwdesc(uc, d);
2255 			kfree(d);
2256 			return NULL;
2257 		}
2258 
2259 		d->residue += sg_len;
2260 		hwdesc->cppi5_desc_size = uc->config.hdesc_size;
2261 		desc = hwdesc->cppi5_desc_vaddr;
2262 
2263 		if (i == 0) {
2264 			cppi5_hdesc_init(desc, 0, 0);
2265 			/* Flow and Packed ID */
2266 			cppi5_desc_set_pktids(&desc->hdr, uc->id,
2267 					      CPPI5_INFO1_DESC_FLOWID_DEFAULT);
2268 			cppi5_desc_set_retpolicy(&desc->hdr, 0, ring_id);
2269 		} else {
2270 			cppi5_hdesc_reset_hbdesc(desc);
2271 			cppi5_desc_set_retpolicy(&desc->hdr, 0, 0xffff);
2272 		}
2273 
2274 		/* attach the sg buffer to the descriptor */
2275 		cppi5_hdesc_attach_buf(desc, sg_addr, sg_len, sg_addr, sg_len);
2276 
2277 		/* Attach link as host buffer descriptor */
2278 		if (h_desc)
2279 			cppi5_hdesc_link_hbdesc(h_desc,
2280 						hwdesc->cppi5_desc_paddr);
2281 
2282 		if (dir == DMA_MEM_TO_DEV)
2283 			h_desc = desc;
2284 	}
2285 
2286 	if (d->residue >= SZ_4M) {
2287 		dev_err(uc->ud->dev,
2288 			"%s: Transfer size %u is over the supported 4M range\n",
2289 			__func__, d->residue);
2290 		udma_free_hwdesc(uc, d);
2291 		kfree(d);
2292 		return NULL;
2293 	}
2294 
2295 	h_desc = d->hwdesc[0].cppi5_desc_vaddr;
2296 	cppi5_hdesc_set_pktlen(h_desc, d->residue);
2297 
2298 	return d;
2299 }
2300 
2301 static int udma_attach_metadata(struct dma_async_tx_descriptor *desc,
2302 				void *data, size_t len)
2303 {
2304 	struct udma_desc *d = to_udma_desc(desc);
2305 	struct udma_chan *uc = to_udma_chan(desc->chan);
2306 	struct cppi5_host_desc_t *h_desc;
2307 	u32 psd_size = len;
2308 	u32 flags = 0;
2309 
2310 	if (!uc->config.pkt_mode || !uc->config.metadata_size)
2311 		return -ENOTSUPP;
2312 
2313 	if (!data || len > uc->config.metadata_size)
2314 		return -EINVAL;
2315 
2316 	if (uc->config.needs_epib && len < CPPI5_INFO0_HDESC_EPIB_SIZE)
2317 		return -EINVAL;
2318 
2319 	h_desc = d->hwdesc[0].cppi5_desc_vaddr;
2320 	if (d->dir == DMA_MEM_TO_DEV)
2321 		memcpy(h_desc->epib, data, len);
2322 
2323 	if (uc->config.needs_epib)
2324 		psd_size -= CPPI5_INFO0_HDESC_EPIB_SIZE;
2325 
2326 	d->metadata = data;
2327 	d->metadata_size = len;
2328 	if (uc->config.needs_epib)
2329 		flags |= CPPI5_INFO0_HDESC_EPIB_PRESENT;
2330 
2331 	cppi5_hdesc_update_flags(h_desc, flags);
2332 	cppi5_hdesc_update_psdata_size(h_desc, psd_size);
2333 
2334 	return 0;
2335 }
2336 
2337 static void *udma_get_metadata_ptr(struct dma_async_tx_descriptor *desc,
2338 				   size_t *payload_len, size_t *max_len)
2339 {
2340 	struct udma_desc *d = to_udma_desc(desc);
2341 	struct udma_chan *uc = to_udma_chan(desc->chan);
2342 	struct cppi5_host_desc_t *h_desc;
2343 
2344 	if (!uc->config.pkt_mode || !uc->config.metadata_size)
2345 		return ERR_PTR(-ENOTSUPP);
2346 
2347 	h_desc = d->hwdesc[0].cppi5_desc_vaddr;
2348 
2349 	*max_len = uc->config.metadata_size;
2350 
2351 	*payload_len = cppi5_hdesc_epib_present(&h_desc->hdr) ?
2352 		       CPPI5_INFO0_HDESC_EPIB_SIZE : 0;
2353 	*payload_len += cppi5_hdesc_get_psdata_size(h_desc);
2354 
2355 	return h_desc->epib;
2356 }
2357 
2358 static int udma_set_metadata_len(struct dma_async_tx_descriptor *desc,
2359 				 size_t payload_len)
2360 {
2361 	struct udma_desc *d = to_udma_desc(desc);
2362 	struct udma_chan *uc = to_udma_chan(desc->chan);
2363 	struct cppi5_host_desc_t *h_desc;
2364 	u32 psd_size = payload_len;
2365 	u32 flags = 0;
2366 
2367 	if (!uc->config.pkt_mode || !uc->config.metadata_size)
2368 		return -ENOTSUPP;
2369 
2370 	if (payload_len > uc->config.metadata_size)
2371 		return -EINVAL;
2372 
2373 	if (uc->config.needs_epib && payload_len < CPPI5_INFO0_HDESC_EPIB_SIZE)
2374 		return -EINVAL;
2375 
2376 	h_desc = d->hwdesc[0].cppi5_desc_vaddr;
2377 
2378 	if (uc->config.needs_epib) {
2379 		psd_size -= CPPI5_INFO0_HDESC_EPIB_SIZE;
2380 		flags |= CPPI5_INFO0_HDESC_EPIB_PRESENT;
2381 	}
2382 
2383 	cppi5_hdesc_update_flags(h_desc, flags);
2384 	cppi5_hdesc_update_psdata_size(h_desc, psd_size);
2385 
2386 	return 0;
2387 }
2388 
2389 static struct dma_descriptor_metadata_ops metadata_ops = {
2390 	.attach = udma_attach_metadata,
2391 	.get_ptr = udma_get_metadata_ptr,
2392 	.set_len = udma_set_metadata_len,
2393 };
2394 
2395 static struct dma_async_tx_descriptor *
2396 udma_prep_slave_sg(struct dma_chan *chan, struct scatterlist *sgl,
2397 		   unsigned int sglen, enum dma_transfer_direction dir,
2398 		   unsigned long tx_flags, void *context)
2399 {
2400 	struct udma_chan *uc = to_udma_chan(chan);
2401 	enum dma_slave_buswidth dev_width;
2402 	struct udma_desc *d;
2403 	u32 burst;
2404 
2405 	if (dir != uc->config.dir) {
2406 		dev_err(chan->device->dev,
2407 			"%s: chan%d is for %s, not supporting %s\n",
2408 			__func__, uc->id,
2409 			dmaengine_get_direction_text(uc->config.dir),
2410 			dmaengine_get_direction_text(dir));
2411 		return NULL;
2412 	}
2413 
2414 	if (dir == DMA_DEV_TO_MEM) {
2415 		dev_width = uc->cfg.src_addr_width;
2416 		burst = uc->cfg.src_maxburst;
2417 	} else if (dir == DMA_MEM_TO_DEV) {
2418 		dev_width = uc->cfg.dst_addr_width;
2419 		burst = uc->cfg.dst_maxburst;
2420 	} else {
2421 		dev_err(chan->device->dev, "%s: bad direction?\n", __func__);
2422 		return NULL;
2423 	}
2424 
2425 	if (!burst)
2426 		burst = 1;
2427 
2428 	if (uc->config.pkt_mode)
2429 		d = udma_prep_slave_sg_pkt(uc, sgl, sglen, dir, tx_flags,
2430 					   context);
2431 	else
2432 		d = udma_prep_slave_sg_tr(uc, sgl, sglen, dir, tx_flags,
2433 					  context);
2434 
2435 	if (!d)
2436 		return NULL;
2437 
2438 	d->dir = dir;
2439 	d->desc_idx = 0;
2440 	d->tr_idx = 0;
2441 
2442 	/* static TR for remote PDMA */
2443 	if (udma_configure_statictr(uc, d, dev_width, burst)) {
2444 		dev_err(uc->ud->dev,
2445 			"%s: StaticTR Z is limited to maximum 4095 (%u)\n",
2446 			__func__, d->static_tr.bstcnt);
2447 
2448 		udma_free_hwdesc(uc, d);
2449 		kfree(d);
2450 		return NULL;
2451 	}
2452 
2453 	if (uc->config.metadata_size)
2454 		d->vd.tx.metadata_ops = &metadata_ops;
2455 
2456 	return vchan_tx_prep(&uc->vc, &d->vd, tx_flags);
2457 }
2458 
2459 static struct udma_desc *
2460 udma_prep_dma_cyclic_tr(struct udma_chan *uc, dma_addr_t buf_addr,
2461 			size_t buf_len, size_t period_len,
2462 			enum dma_transfer_direction dir, unsigned long flags)
2463 {
2464 	struct udma_desc *d;
2465 	size_t tr_size, period_addr;
2466 	struct cppi5_tr_type1_t *tr_req;
2467 	unsigned int periods = buf_len / period_len;
2468 	u16 tr0_cnt0, tr0_cnt1, tr1_cnt0;
2469 	unsigned int i;
2470 	int num_tr;
2471 
2472 	if (!is_slave_direction(dir)) {
2473 		dev_err(uc->ud->dev, "Only slave cyclic is supported\n");
2474 		return NULL;
2475 	}
2476 
2477 	num_tr = udma_get_tr_counters(period_len, __ffs(buf_addr), &tr0_cnt0,
2478 				      &tr0_cnt1, &tr1_cnt0);
2479 	if (num_tr < 0) {
2480 		dev_err(uc->ud->dev, "size %zu is not supported\n",
2481 			period_len);
2482 		return NULL;
2483 	}
2484 
2485 	/* Now allocate and setup the descriptor. */
2486 	tr_size = sizeof(struct cppi5_tr_type1_t);
2487 	d = udma_alloc_tr_desc(uc, tr_size, periods * num_tr, dir);
2488 	if (!d)
2489 		return NULL;
2490 
2491 	tr_req = d->hwdesc[0].tr_req_base;
2492 	period_addr = buf_addr;
2493 	for (i = 0; i < periods; i++) {
2494 		int tr_idx = i * num_tr;
2495 
2496 		cppi5_tr_init(&tr_req[tr_idx].flags, CPPI5_TR_TYPE1, false,
2497 			      false, CPPI5_TR_EVENT_SIZE_COMPLETION, 0);
2498 
2499 		tr_req[tr_idx].addr = period_addr;
2500 		tr_req[tr_idx].icnt0 = tr0_cnt0;
2501 		tr_req[tr_idx].icnt1 = tr0_cnt1;
2502 		tr_req[tr_idx].dim1 = tr0_cnt0;
2503 
2504 		if (num_tr == 2) {
2505 			cppi5_tr_csf_set(&tr_req[tr_idx].flags,
2506 					 CPPI5_TR_CSF_SUPR_EVT);
2507 			tr_idx++;
2508 
2509 			cppi5_tr_init(&tr_req[tr_idx].flags, CPPI5_TR_TYPE1,
2510 				      false, false,
2511 				      CPPI5_TR_EVENT_SIZE_COMPLETION, 0);
2512 
2513 			tr_req[tr_idx].addr = period_addr + tr0_cnt1 * tr0_cnt0;
2514 			tr_req[tr_idx].icnt0 = tr1_cnt0;
2515 			tr_req[tr_idx].icnt1 = 1;
2516 			tr_req[tr_idx].dim1 = tr1_cnt0;
2517 		}
2518 
2519 		if (!(flags & DMA_PREP_INTERRUPT))
2520 			cppi5_tr_csf_set(&tr_req[tr_idx].flags,
2521 					 CPPI5_TR_CSF_SUPR_EVT);
2522 
2523 		period_addr += period_len;
2524 	}
2525 
2526 	return d;
2527 }
2528 
2529 static struct udma_desc *
2530 udma_prep_dma_cyclic_pkt(struct udma_chan *uc, dma_addr_t buf_addr,
2531 			 size_t buf_len, size_t period_len,
2532 			 enum dma_transfer_direction dir, unsigned long flags)
2533 {
2534 	struct udma_desc *d;
2535 	u32 ring_id;
2536 	int i;
2537 	int periods = buf_len / period_len;
2538 
2539 	if (periods > (K3_UDMA_DEFAULT_RING_SIZE - 1))
2540 		return NULL;
2541 
2542 	if (period_len >= SZ_4M)
2543 		return NULL;
2544 
2545 	d = kzalloc(sizeof(*d) + periods * sizeof(d->hwdesc[0]), GFP_NOWAIT);
2546 	if (!d)
2547 		return NULL;
2548 
2549 	d->hwdesc_count = periods;
2550 
2551 	/* TODO: re-check this... */
2552 	if (dir == DMA_DEV_TO_MEM)
2553 		ring_id = k3_ringacc_get_ring_id(uc->rflow->r_ring);
2554 	else
2555 		ring_id = k3_ringacc_get_ring_id(uc->tchan->tc_ring);
2556 
2557 	for (i = 0; i < periods; i++) {
2558 		struct udma_hwdesc *hwdesc = &d->hwdesc[i];
2559 		dma_addr_t period_addr = buf_addr + (period_len * i);
2560 		struct cppi5_host_desc_t *h_desc;
2561 
2562 		hwdesc->cppi5_desc_vaddr = dma_pool_zalloc(uc->hdesc_pool,
2563 						GFP_NOWAIT,
2564 						&hwdesc->cppi5_desc_paddr);
2565 		if (!hwdesc->cppi5_desc_vaddr) {
2566 			dev_err(uc->ud->dev,
2567 				"descriptor%d allocation failed\n", i);
2568 
2569 			udma_free_hwdesc(uc, d);
2570 			kfree(d);
2571 			return NULL;
2572 		}
2573 
2574 		hwdesc->cppi5_desc_size = uc->config.hdesc_size;
2575 		h_desc = hwdesc->cppi5_desc_vaddr;
2576 
2577 		cppi5_hdesc_init(h_desc, 0, 0);
2578 		cppi5_hdesc_set_pktlen(h_desc, period_len);
2579 
2580 		/* Flow and Packed ID */
2581 		cppi5_desc_set_pktids(&h_desc->hdr, uc->id,
2582 				      CPPI5_INFO1_DESC_FLOWID_DEFAULT);
2583 		cppi5_desc_set_retpolicy(&h_desc->hdr, 0, ring_id);
2584 
2585 		/* attach each period to a new descriptor */
2586 		cppi5_hdesc_attach_buf(h_desc,
2587 				       period_addr, period_len,
2588 				       period_addr, period_len);
2589 	}
2590 
2591 	return d;
2592 }
2593 
2594 static struct dma_async_tx_descriptor *
2595 udma_prep_dma_cyclic(struct dma_chan *chan, dma_addr_t buf_addr, size_t buf_len,
2596 		     size_t period_len, enum dma_transfer_direction dir,
2597 		     unsigned long flags)
2598 {
2599 	struct udma_chan *uc = to_udma_chan(chan);
2600 	enum dma_slave_buswidth dev_width;
2601 	struct udma_desc *d;
2602 	u32 burst;
2603 
2604 	if (dir != uc->config.dir) {
2605 		dev_err(chan->device->dev,
2606 			"%s: chan%d is for %s, not supporting %s\n",
2607 			__func__, uc->id,
2608 			dmaengine_get_direction_text(uc->config.dir),
2609 			dmaengine_get_direction_text(dir));
2610 		return NULL;
2611 	}
2612 
2613 	uc->cyclic = true;
2614 
2615 	if (dir == DMA_DEV_TO_MEM) {
2616 		dev_width = uc->cfg.src_addr_width;
2617 		burst = uc->cfg.src_maxburst;
2618 	} else if (dir == DMA_MEM_TO_DEV) {
2619 		dev_width = uc->cfg.dst_addr_width;
2620 		burst = uc->cfg.dst_maxburst;
2621 	} else {
2622 		dev_err(uc->ud->dev, "%s: bad direction?\n", __func__);
2623 		return NULL;
2624 	}
2625 
2626 	if (!burst)
2627 		burst = 1;
2628 
2629 	if (uc->config.pkt_mode)
2630 		d = udma_prep_dma_cyclic_pkt(uc, buf_addr, buf_len, period_len,
2631 					     dir, flags);
2632 	else
2633 		d = udma_prep_dma_cyclic_tr(uc, buf_addr, buf_len, period_len,
2634 					    dir, flags);
2635 
2636 	if (!d)
2637 		return NULL;
2638 
2639 	d->sglen = buf_len / period_len;
2640 
2641 	d->dir = dir;
2642 	d->residue = buf_len;
2643 
2644 	/* static TR for remote PDMA */
2645 	if (udma_configure_statictr(uc, d, dev_width, burst)) {
2646 		dev_err(uc->ud->dev,
2647 			"%s: StaticTR Z is limited to maximum 4095 (%u)\n",
2648 			__func__, d->static_tr.bstcnt);
2649 
2650 		udma_free_hwdesc(uc, d);
2651 		kfree(d);
2652 		return NULL;
2653 	}
2654 
2655 	if (uc->config.metadata_size)
2656 		d->vd.tx.metadata_ops = &metadata_ops;
2657 
2658 	return vchan_tx_prep(&uc->vc, &d->vd, flags);
2659 }
2660 
2661 static struct dma_async_tx_descriptor *
2662 udma_prep_dma_memcpy(struct dma_chan *chan, dma_addr_t dest, dma_addr_t src,
2663 		     size_t len, unsigned long tx_flags)
2664 {
2665 	struct udma_chan *uc = to_udma_chan(chan);
2666 	struct udma_desc *d;
2667 	struct cppi5_tr_type15_t *tr_req;
2668 	int num_tr;
2669 	size_t tr_size = sizeof(struct cppi5_tr_type15_t);
2670 	u16 tr0_cnt0, tr0_cnt1, tr1_cnt0;
2671 
2672 	if (uc->config.dir != DMA_MEM_TO_MEM) {
2673 		dev_err(chan->device->dev,
2674 			"%s: chan%d is for %s, not supporting %s\n",
2675 			__func__, uc->id,
2676 			dmaengine_get_direction_text(uc->config.dir),
2677 			dmaengine_get_direction_text(DMA_MEM_TO_MEM));
2678 		return NULL;
2679 	}
2680 
2681 	num_tr = udma_get_tr_counters(len, __ffs(src | dest), &tr0_cnt0,
2682 				      &tr0_cnt1, &tr1_cnt0);
2683 	if (num_tr < 0) {
2684 		dev_err(uc->ud->dev, "size %zu is not supported\n",
2685 			len);
2686 		return NULL;
2687 	}
2688 
2689 	d = udma_alloc_tr_desc(uc, tr_size, num_tr, DMA_MEM_TO_MEM);
2690 	if (!d)
2691 		return NULL;
2692 
2693 	d->dir = DMA_MEM_TO_MEM;
2694 	d->desc_idx = 0;
2695 	d->tr_idx = 0;
2696 	d->residue = len;
2697 
2698 	tr_req = d->hwdesc[0].tr_req_base;
2699 
2700 	cppi5_tr_init(&tr_req[0].flags, CPPI5_TR_TYPE15, false, true,
2701 		      CPPI5_TR_EVENT_SIZE_COMPLETION, 0);
2702 	cppi5_tr_csf_set(&tr_req[0].flags, CPPI5_TR_CSF_SUPR_EVT);
2703 
2704 	tr_req[0].addr = src;
2705 	tr_req[0].icnt0 = tr0_cnt0;
2706 	tr_req[0].icnt1 = tr0_cnt1;
2707 	tr_req[0].icnt2 = 1;
2708 	tr_req[0].icnt3 = 1;
2709 	tr_req[0].dim1 = tr0_cnt0;
2710 
2711 	tr_req[0].daddr = dest;
2712 	tr_req[0].dicnt0 = tr0_cnt0;
2713 	tr_req[0].dicnt1 = tr0_cnt1;
2714 	tr_req[0].dicnt2 = 1;
2715 	tr_req[0].dicnt3 = 1;
2716 	tr_req[0].ddim1 = tr0_cnt0;
2717 
2718 	if (num_tr == 2) {
2719 		cppi5_tr_init(&tr_req[1].flags, CPPI5_TR_TYPE15, false, true,
2720 			      CPPI5_TR_EVENT_SIZE_COMPLETION, 0);
2721 		cppi5_tr_csf_set(&tr_req[1].flags, CPPI5_TR_CSF_SUPR_EVT);
2722 
2723 		tr_req[1].addr = src + tr0_cnt1 * tr0_cnt0;
2724 		tr_req[1].icnt0 = tr1_cnt0;
2725 		tr_req[1].icnt1 = 1;
2726 		tr_req[1].icnt2 = 1;
2727 		tr_req[1].icnt3 = 1;
2728 
2729 		tr_req[1].daddr = dest + tr0_cnt1 * tr0_cnt0;
2730 		tr_req[1].dicnt0 = tr1_cnt0;
2731 		tr_req[1].dicnt1 = 1;
2732 		tr_req[1].dicnt2 = 1;
2733 		tr_req[1].dicnt3 = 1;
2734 	}
2735 
2736 	cppi5_tr_csf_set(&tr_req[num_tr - 1].flags, CPPI5_TR_CSF_EOP);
2737 
2738 	if (uc->config.metadata_size)
2739 		d->vd.tx.metadata_ops = &metadata_ops;
2740 
2741 	return vchan_tx_prep(&uc->vc, &d->vd, tx_flags);
2742 }
2743 
2744 static void udma_issue_pending(struct dma_chan *chan)
2745 {
2746 	struct udma_chan *uc = to_udma_chan(chan);
2747 	unsigned long flags;
2748 
2749 	spin_lock_irqsave(&uc->vc.lock, flags);
2750 
2751 	/* If we have something pending and no active descriptor, then */
2752 	if (vchan_issue_pending(&uc->vc) && !uc->desc) {
2753 		/*
2754 		 * start a descriptor if the channel is NOT [marked as
2755 		 * terminating _and_ it is still running (teardown has not
2756 		 * completed yet)].
2757 		 */
2758 		if (!(uc->state == UDMA_CHAN_IS_TERMINATING &&
2759 		      udma_is_chan_running(uc)))
2760 			udma_start(uc);
2761 	}
2762 
2763 	spin_unlock_irqrestore(&uc->vc.lock, flags);
2764 }
2765 
2766 static enum dma_status udma_tx_status(struct dma_chan *chan,
2767 				      dma_cookie_t cookie,
2768 				      struct dma_tx_state *txstate)
2769 {
2770 	struct udma_chan *uc = to_udma_chan(chan);
2771 	enum dma_status ret;
2772 	unsigned long flags;
2773 
2774 	spin_lock_irqsave(&uc->vc.lock, flags);
2775 
2776 	ret = dma_cookie_status(chan, cookie, txstate);
2777 
2778 	if (!udma_is_chan_running(uc))
2779 		ret = DMA_COMPLETE;
2780 
2781 	if (ret == DMA_IN_PROGRESS && udma_is_chan_paused(uc))
2782 		ret = DMA_PAUSED;
2783 
2784 	if (ret == DMA_COMPLETE || !txstate)
2785 		goto out;
2786 
2787 	if (uc->desc && uc->desc->vd.tx.cookie == cookie) {
2788 		u32 peer_bcnt = 0;
2789 		u32 bcnt = 0;
2790 		u32 residue = uc->desc->residue;
2791 		u32 delay = 0;
2792 
2793 		if (uc->desc->dir == DMA_MEM_TO_DEV) {
2794 			bcnt = udma_tchanrt_read(uc->tchan,
2795 						 UDMA_TCHAN_RT_SBCNT_REG);
2796 
2797 			if (uc->config.ep_type != PSIL_EP_NATIVE) {
2798 				peer_bcnt = udma_tchanrt_read(uc->tchan,
2799 						UDMA_TCHAN_RT_PEER_BCNT_REG);
2800 
2801 				if (bcnt > peer_bcnt)
2802 					delay = bcnt - peer_bcnt;
2803 			}
2804 		} else if (uc->desc->dir == DMA_DEV_TO_MEM) {
2805 			bcnt = udma_rchanrt_read(uc->rchan,
2806 						 UDMA_RCHAN_RT_BCNT_REG);
2807 
2808 			if (uc->config.ep_type != PSIL_EP_NATIVE) {
2809 				peer_bcnt = udma_rchanrt_read(uc->rchan,
2810 						UDMA_RCHAN_RT_PEER_BCNT_REG);
2811 
2812 				if (peer_bcnt > bcnt)
2813 					delay = peer_bcnt - bcnt;
2814 			}
2815 		} else {
2816 			bcnt = udma_tchanrt_read(uc->tchan,
2817 						 UDMA_TCHAN_RT_BCNT_REG);
2818 		}
2819 
2820 		bcnt -= uc->bcnt;
2821 		if (bcnt && !(bcnt % uc->desc->residue))
2822 			residue = 0;
2823 		else
2824 			residue -= bcnt % uc->desc->residue;
2825 
2826 		if (!residue && (uc->config.dir == DMA_DEV_TO_MEM || !delay)) {
2827 			ret = DMA_COMPLETE;
2828 			delay = 0;
2829 		}
2830 
2831 		dma_set_residue(txstate, residue);
2832 		dma_set_in_flight_bytes(txstate, delay);
2833 
2834 	} else {
2835 		ret = DMA_COMPLETE;
2836 	}
2837 
2838 out:
2839 	spin_unlock_irqrestore(&uc->vc.lock, flags);
2840 	return ret;
2841 }
2842 
2843 static int udma_pause(struct dma_chan *chan)
2844 {
2845 	struct udma_chan *uc = to_udma_chan(chan);
2846 
2847 	/* pause 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,
2853 					 UDMA_PEER_RT_EN_PAUSE);
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,
2859 					 UDMA_PEER_RT_EN_PAUSE);
2860 		break;
2861 	case DMA_MEM_TO_MEM:
2862 		udma_tchanrt_update_bits(uc->tchan, UDMA_TCHAN_RT_CTL_REG,
2863 					 UDMA_CHAN_RT_CTL_PAUSE,
2864 					 UDMA_CHAN_RT_CTL_PAUSE);
2865 		break;
2866 	default:
2867 		return -EINVAL;
2868 	}
2869 
2870 	return 0;
2871 }
2872 
2873 static int udma_resume(struct dma_chan *chan)
2874 {
2875 	struct udma_chan *uc = to_udma_chan(chan);
2876 
2877 	/* resume the channel */
2878 	switch (uc->config.dir) {
2879 	case DMA_DEV_TO_MEM:
2880 		udma_rchanrt_update_bits(uc->rchan,
2881 					 UDMA_RCHAN_RT_PEER_RT_EN_REG,
2882 					 UDMA_PEER_RT_EN_PAUSE, 0);
2883 
2884 		break;
2885 	case DMA_MEM_TO_DEV:
2886 		udma_tchanrt_update_bits(uc->tchan,
2887 					 UDMA_TCHAN_RT_PEER_RT_EN_REG,
2888 					 UDMA_PEER_RT_EN_PAUSE, 0);
2889 		break;
2890 	case DMA_MEM_TO_MEM:
2891 		udma_tchanrt_update_bits(uc->tchan, UDMA_TCHAN_RT_CTL_REG,
2892 					 UDMA_CHAN_RT_CTL_PAUSE, 0);
2893 		break;
2894 	default:
2895 		return -EINVAL;
2896 	}
2897 
2898 	return 0;
2899 }
2900 
2901 static int udma_terminate_all(struct dma_chan *chan)
2902 {
2903 	struct udma_chan *uc = to_udma_chan(chan);
2904 	unsigned long flags;
2905 	LIST_HEAD(head);
2906 
2907 	spin_lock_irqsave(&uc->vc.lock, flags);
2908 
2909 	if (udma_is_chan_running(uc))
2910 		udma_stop(uc);
2911 
2912 	if (uc->desc) {
2913 		uc->terminated_desc = uc->desc;
2914 		uc->desc = NULL;
2915 		uc->terminated_desc->terminated = true;
2916 		cancel_delayed_work(&uc->tx_drain.work);
2917 	}
2918 
2919 	uc->paused = false;
2920 
2921 	vchan_get_all_descriptors(&uc->vc, &head);
2922 	spin_unlock_irqrestore(&uc->vc.lock, flags);
2923 	vchan_dma_desc_free_list(&uc->vc, &head);
2924 
2925 	return 0;
2926 }
2927 
2928 static void udma_synchronize(struct dma_chan *chan)
2929 {
2930 	struct udma_chan *uc = to_udma_chan(chan);
2931 	unsigned long timeout = msecs_to_jiffies(1000);
2932 
2933 	vchan_synchronize(&uc->vc);
2934 
2935 	if (uc->state == UDMA_CHAN_IS_TERMINATING) {
2936 		timeout = wait_for_completion_timeout(&uc->teardown_completed,
2937 						      timeout);
2938 		if (!timeout) {
2939 			dev_warn(uc->ud->dev, "chan%d teardown timeout!\n",
2940 				 uc->id);
2941 			udma_dump_chan_stdata(uc);
2942 			udma_reset_chan(uc, true);
2943 		}
2944 	}
2945 
2946 	udma_reset_chan(uc, false);
2947 	if (udma_is_chan_running(uc))
2948 		dev_warn(uc->ud->dev, "chan%d refused to stop!\n", uc->id);
2949 
2950 	cancel_delayed_work_sync(&uc->tx_drain.work);
2951 	udma_reset_rings(uc);
2952 }
2953 
2954 static void udma_desc_pre_callback(struct virt_dma_chan *vc,
2955 				   struct virt_dma_desc *vd,
2956 				   struct dmaengine_result *result)
2957 {
2958 	struct udma_chan *uc = to_udma_chan(&vc->chan);
2959 	struct udma_desc *d;
2960 
2961 	if (!vd)
2962 		return;
2963 
2964 	d = to_udma_desc(&vd->tx);
2965 
2966 	if (d->metadata_size)
2967 		udma_fetch_epib(uc, d);
2968 
2969 	/* Provide residue information for the client */
2970 	if (result) {
2971 		void *desc_vaddr = udma_curr_cppi5_desc_vaddr(d, d->desc_idx);
2972 
2973 		if (cppi5_desc_get_type(desc_vaddr) ==
2974 		    CPPI5_INFO0_DESC_TYPE_VAL_HOST) {
2975 			result->residue = d->residue -
2976 					  cppi5_hdesc_get_pktlen(desc_vaddr);
2977 			if (result->residue)
2978 				result->result = DMA_TRANS_ABORTED;
2979 			else
2980 				result->result = DMA_TRANS_NOERROR;
2981 		} else {
2982 			result->residue = 0;
2983 			result->result = DMA_TRANS_NOERROR;
2984 		}
2985 	}
2986 }
2987 
2988 /*
2989  * This tasklet handles the completion of a DMA descriptor by
2990  * calling its callback and freeing it.
2991  */
2992 static void udma_vchan_complete(unsigned long arg)
2993 {
2994 	struct virt_dma_chan *vc = (struct virt_dma_chan *)arg;
2995 	struct virt_dma_desc *vd, *_vd;
2996 	struct dmaengine_desc_callback cb;
2997 	LIST_HEAD(head);
2998 
2999 	spin_lock_irq(&vc->lock);
3000 	list_splice_tail_init(&vc->desc_completed, &head);
3001 	vd = vc->cyclic;
3002 	if (vd) {
3003 		vc->cyclic = NULL;
3004 		dmaengine_desc_get_callback(&vd->tx, &cb);
3005 	} else {
3006 		memset(&cb, 0, sizeof(cb));
3007 	}
3008 	spin_unlock_irq(&vc->lock);
3009 
3010 	udma_desc_pre_callback(vc, vd, NULL);
3011 	dmaengine_desc_callback_invoke(&cb, NULL);
3012 
3013 	list_for_each_entry_safe(vd, _vd, &head, node) {
3014 		struct dmaengine_result result;
3015 
3016 		dmaengine_desc_get_callback(&vd->tx, &cb);
3017 
3018 		list_del(&vd->node);
3019 
3020 		udma_desc_pre_callback(vc, vd, &result);
3021 		dmaengine_desc_callback_invoke(&cb, &result);
3022 
3023 		vchan_vdesc_fini(vd);
3024 	}
3025 }
3026 
3027 static void udma_free_chan_resources(struct dma_chan *chan)
3028 {
3029 	struct udma_chan *uc = to_udma_chan(chan);
3030 	struct udma_dev *ud = to_udma_dev(chan->device);
3031 
3032 	udma_terminate_all(chan);
3033 	if (uc->terminated_desc) {
3034 		udma_reset_chan(uc, false);
3035 		udma_reset_rings(uc);
3036 	}
3037 
3038 	cancel_delayed_work_sync(&uc->tx_drain.work);
3039 	destroy_delayed_work_on_stack(&uc->tx_drain.work);
3040 
3041 	if (uc->irq_num_ring > 0) {
3042 		free_irq(uc->irq_num_ring, uc);
3043 
3044 		uc->irq_num_ring = 0;
3045 	}
3046 	if (uc->irq_num_udma > 0) {
3047 		free_irq(uc->irq_num_udma, uc);
3048 
3049 		uc->irq_num_udma = 0;
3050 	}
3051 
3052 	/* Release PSI-L pairing */
3053 	if (uc->psil_paired) {
3054 		navss_psil_unpair(ud, uc->config.src_thread,
3055 				  uc->config.dst_thread);
3056 		uc->psil_paired = false;
3057 	}
3058 
3059 	vchan_free_chan_resources(&uc->vc);
3060 	tasklet_kill(&uc->vc.task);
3061 
3062 	udma_free_tx_resources(uc);
3063 	udma_free_rx_resources(uc);
3064 	udma_reset_uchan(uc);
3065 
3066 	if (uc->use_dma_pool) {
3067 		dma_pool_destroy(uc->hdesc_pool);
3068 		uc->use_dma_pool = false;
3069 	}
3070 }
3071 
3072 static struct platform_driver udma_driver;
3073 
3074 struct udma_filter_param {
3075 	int remote_thread_id;
3076 	u32 atype;
3077 };
3078 
3079 static bool udma_dma_filter_fn(struct dma_chan *chan, void *param)
3080 {
3081 	struct udma_chan_config *ucc;
3082 	struct psil_endpoint_config *ep_config;
3083 	struct udma_filter_param *filter_param;
3084 	struct udma_chan *uc;
3085 	struct udma_dev *ud;
3086 
3087 	if (chan->device->dev->driver != &udma_driver.driver)
3088 		return false;
3089 
3090 	uc = to_udma_chan(chan);
3091 	ucc = &uc->config;
3092 	ud = uc->ud;
3093 	filter_param = param;
3094 
3095 	if (filter_param->atype > 2) {
3096 		dev_err(ud->dev, "Invalid channel atype: %u\n",
3097 			filter_param->atype);
3098 		return false;
3099 	}
3100 
3101 	ucc->remote_thread_id = filter_param->remote_thread_id;
3102 	ucc->atype = filter_param->atype;
3103 
3104 	if (ucc->remote_thread_id & K3_PSIL_DST_THREAD_ID_OFFSET)
3105 		ucc->dir = DMA_MEM_TO_DEV;
3106 	else
3107 		ucc->dir = DMA_DEV_TO_MEM;
3108 
3109 	ep_config = psil_get_ep_config(ucc->remote_thread_id);
3110 	if (IS_ERR(ep_config)) {
3111 		dev_err(ud->dev, "No configuration for psi-l thread 0x%04x\n",
3112 			ucc->remote_thread_id);
3113 		ucc->dir = DMA_MEM_TO_MEM;
3114 		ucc->remote_thread_id = -1;
3115 		ucc->atype = 0;
3116 		return false;
3117 	}
3118 
3119 	ucc->pkt_mode = ep_config->pkt_mode;
3120 	ucc->channel_tpl = ep_config->channel_tpl;
3121 	ucc->notdpkt = ep_config->notdpkt;
3122 	ucc->ep_type = ep_config->ep_type;
3123 
3124 	if (ucc->ep_type != PSIL_EP_NATIVE) {
3125 		const struct udma_match_data *match_data = ud->match_data;
3126 
3127 		if (match_data->flags & UDMA_FLAG_PDMA_ACC32)
3128 			ucc->enable_acc32 = ep_config->pdma_acc32;
3129 		if (match_data->flags & UDMA_FLAG_PDMA_BURST)
3130 			ucc->enable_burst = ep_config->pdma_burst;
3131 	}
3132 
3133 	ucc->needs_epib = ep_config->needs_epib;
3134 	ucc->psd_size = ep_config->psd_size;
3135 	ucc->metadata_size =
3136 			(ucc->needs_epib ? CPPI5_INFO0_HDESC_EPIB_SIZE : 0) +
3137 			ucc->psd_size;
3138 
3139 	if (ucc->pkt_mode)
3140 		ucc->hdesc_size = ALIGN(sizeof(struct cppi5_host_desc_t) +
3141 				 ucc->metadata_size, ud->desc_align);
3142 
3143 	dev_dbg(ud->dev, "chan%d: Remote thread: 0x%04x (%s)\n", uc->id,
3144 		ucc->remote_thread_id, dmaengine_get_direction_text(ucc->dir));
3145 
3146 	return true;
3147 }
3148 
3149 static struct dma_chan *udma_of_xlate(struct of_phandle_args *dma_spec,
3150 				      struct of_dma *ofdma)
3151 {
3152 	struct udma_dev *ud = ofdma->of_dma_data;
3153 	dma_cap_mask_t mask = ud->ddev.cap_mask;
3154 	struct udma_filter_param filter_param;
3155 	struct dma_chan *chan;
3156 
3157 	if (dma_spec->args_count != 1 && dma_spec->args_count != 2)
3158 		return NULL;
3159 
3160 	filter_param.remote_thread_id = dma_spec->args[0];
3161 	if (dma_spec->args_count == 2)
3162 		filter_param.atype = dma_spec->args[1];
3163 	else
3164 		filter_param.atype = 0;
3165 
3166 	chan = __dma_request_channel(&mask, udma_dma_filter_fn, &filter_param,
3167 				     ofdma->of_node);
3168 	if (!chan) {
3169 		dev_err(ud->dev, "get channel fail in %s.\n", __func__);
3170 		return ERR_PTR(-EINVAL);
3171 	}
3172 
3173 	return chan;
3174 }
3175 
3176 static struct udma_match_data am654_main_data = {
3177 	.psil_base = 0x1000,
3178 	.enable_memcpy_support = true,
3179 	.statictr_z_mask = GENMASK(11, 0),
3180 	.rchan_oes_offset = 0x2000,
3181 	.tpl_levels = 2,
3182 	.level_start_idx = {
3183 		[0] = 8, /* Normal channels */
3184 		[1] = 0, /* High Throughput channels */
3185 	},
3186 };
3187 
3188 static struct udma_match_data am654_mcu_data = {
3189 	.psil_base = 0x6000,
3190 	.enable_memcpy_support = true, /* TEST: DMA domains */
3191 	.statictr_z_mask = GENMASK(11, 0),
3192 	.rchan_oes_offset = 0x2000,
3193 	.tpl_levels = 2,
3194 	.level_start_idx = {
3195 		[0] = 2, /* Normal channels */
3196 		[1] = 0, /* High Throughput channels */
3197 	},
3198 };
3199 
3200 static struct udma_match_data j721e_main_data = {
3201 	.psil_base = 0x1000,
3202 	.enable_memcpy_support = true,
3203 	.flags = UDMA_FLAG_PDMA_ACC32 | UDMA_FLAG_PDMA_BURST,
3204 	.statictr_z_mask = GENMASK(23, 0),
3205 	.rchan_oes_offset = 0x400,
3206 	.tpl_levels = 3,
3207 	.level_start_idx = {
3208 		[0] = 16, /* Normal channels */
3209 		[1] = 4, /* High Throughput channels */
3210 		[2] = 0, /* Ultra High Throughput channels */
3211 	},
3212 };
3213 
3214 static struct udma_match_data j721e_mcu_data = {
3215 	.psil_base = 0x6000,
3216 	.enable_memcpy_support = false, /* MEM_TO_MEM is slow via MCU UDMA */
3217 	.flags = UDMA_FLAG_PDMA_ACC32 | UDMA_FLAG_PDMA_BURST,
3218 	.statictr_z_mask = GENMASK(23, 0),
3219 	.rchan_oes_offset = 0x400,
3220 	.tpl_levels = 2,
3221 	.level_start_idx = {
3222 		[0] = 2, /* Normal channels */
3223 		[1] = 0, /* High Throughput channels */
3224 	},
3225 };
3226 
3227 static const struct of_device_id udma_of_match[] = {
3228 	{
3229 		.compatible = "ti,am654-navss-main-udmap",
3230 		.data = &am654_main_data,
3231 	},
3232 	{
3233 		.compatible = "ti,am654-navss-mcu-udmap",
3234 		.data = &am654_mcu_data,
3235 	}, {
3236 		.compatible = "ti,j721e-navss-main-udmap",
3237 		.data = &j721e_main_data,
3238 	}, {
3239 		.compatible = "ti,j721e-navss-mcu-udmap",
3240 		.data = &j721e_mcu_data,
3241 	},
3242 	{ /* Sentinel */ },
3243 };
3244 
3245 static int udma_get_mmrs(struct platform_device *pdev, struct udma_dev *ud)
3246 {
3247 	struct resource *res;
3248 	int i;
3249 
3250 	for (i = 0; i < MMR_LAST; i++) {
3251 		res = platform_get_resource_byname(pdev, IORESOURCE_MEM,
3252 						   mmr_names[i]);
3253 		ud->mmrs[i] = devm_ioremap_resource(&pdev->dev, res);
3254 		if (IS_ERR(ud->mmrs[i]))
3255 			return PTR_ERR(ud->mmrs[i]);
3256 	}
3257 
3258 	return 0;
3259 }
3260 
3261 static int udma_setup_resources(struct udma_dev *ud)
3262 {
3263 	struct device *dev = ud->dev;
3264 	int ch_count, ret, i, j;
3265 	u32 cap2, cap3;
3266 	struct ti_sci_resource_desc *rm_desc;
3267 	struct ti_sci_resource *rm_res, irq_res;
3268 	struct udma_tisci_rm *tisci_rm = &ud->tisci_rm;
3269 	static const char * const range_names[] = { "ti,sci-rm-range-tchan",
3270 						    "ti,sci-rm-range-rchan",
3271 						    "ti,sci-rm-range-rflow" };
3272 
3273 	cap2 = udma_read(ud->mmrs[MMR_GCFG], 0x28);
3274 	cap3 = udma_read(ud->mmrs[MMR_GCFG], 0x2c);
3275 
3276 	ud->rflow_cnt = cap3 & 0x3fff;
3277 	ud->tchan_cnt = cap2 & 0x1ff;
3278 	ud->echan_cnt = (cap2 >> 9) & 0x1ff;
3279 	ud->rchan_cnt = (cap2 >> 18) & 0x1ff;
3280 	ch_count  = ud->tchan_cnt + ud->rchan_cnt;
3281 
3282 	ud->tchan_map = devm_kmalloc_array(dev, BITS_TO_LONGS(ud->tchan_cnt),
3283 					   sizeof(unsigned long), GFP_KERNEL);
3284 	ud->tchans = devm_kcalloc(dev, ud->tchan_cnt, sizeof(*ud->tchans),
3285 				  GFP_KERNEL);
3286 	ud->rchan_map = devm_kmalloc_array(dev, BITS_TO_LONGS(ud->rchan_cnt),
3287 					   sizeof(unsigned long), GFP_KERNEL);
3288 	ud->rchans = devm_kcalloc(dev, ud->rchan_cnt, sizeof(*ud->rchans),
3289 				  GFP_KERNEL);
3290 	ud->rflow_gp_map = devm_kmalloc_array(dev, BITS_TO_LONGS(ud->rflow_cnt),
3291 					      sizeof(unsigned long),
3292 					      GFP_KERNEL);
3293 	ud->rflow_gp_map_allocated = devm_kcalloc(dev,
3294 						  BITS_TO_LONGS(ud->rflow_cnt),
3295 						  sizeof(unsigned long),
3296 						  GFP_KERNEL);
3297 	ud->rflow_in_use = devm_kcalloc(dev, BITS_TO_LONGS(ud->rflow_cnt),
3298 					sizeof(unsigned long),
3299 					GFP_KERNEL);
3300 	ud->rflows = devm_kcalloc(dev, ud->rflow_cnt, sizeof(*ud->rflows),
3301 				  GFP_KERNEL);
3302 
3303 	if (!ud->tchan_map || !ud->rchan_map || !ud->rflow_gp_map ||
3304 	    !ud->rflow_gp_map_allocated || !ud->tchans || !ud->rchans ||
3305 	    !ud->rflows || !ud->rflow_in_use)
3306 		return -ENOMEM;
3307 
3308 	/*
3309 	 * RX flows with the same Ids as RX channels are reserved to be used
3310 	 * as default flows if remote HW can't generate flow_ids. Those
3311 	 * RX flows can be requested only explicitly by id.
3312 	 */
3313 	bitmap_set(ud->rflow_gp_map_allocated, 0, ud->rchan_cnt);
3314 
3315 	/* by default no GP rflows are assigned to Linux */
3316 	bitmap_set(ud->rflow_gp_map, 0, ud->rflow_cnt);
3317 
3318 	/* Get resource ranges from tisci */
3319 	for (i = 0; i < RM_RANGE_LAST; i++)
3320 		tisci_rm->rm_ranges[i] =
3321 			devm_ti_sci_get_of_resource(tisci_rm->tisci, dev,
3322 						    tisci_rm->tisci_dev_id,
3323 						    (char *)range_names[i]);
3324 
3325 	/* tchan ranges */
3326 	rm_res = tisci_rm->rm_ranges[RM_RANGE_TCHAN];
3327 	if (IS_ERR(rm_res)) {
3328 		bitmap_zero(ud->tchan_map, ud->tchan_cnt);
3329 	} else {
3330 		bitmap_fill(ud->tchan_map, ud->tchan_cnt);
3331 		for (i = 0; i < rm_res->sets; i++) {
3332 			rm_desc = &rm_res->desc[i];
3333 			bitmap_clear(ud->tchan_map, rm_desc->start,
3334 				     rm_desc->num);
3335 			dev_dbg(dev, "ti-sci-res: tchan: %d:%d\n",
3336 				rm_desc->start, rm_desc->num);
3337 		}
3338 	}
3339 	irq_res.sets = rm_res->sets;
3340 
3341 	/* rchan and matching default flow ranges */
3342 	rm_res = tisci_rm->rm_ranges[RM_RANGE_RCHAN];
3343 	if (IS_ERR(rm_res)) {
3344 		bitmap_zero(ud->rchan_map, ud->rchan_cnt);
3345 	} else {
3346 		bitmap_fill(ud->rchan_map, ud->rchan_cnt);
3347 		for (i = 0; i < rm_res->sets; i++) {
3348 			rm_desc = &rm_res->desc[i];
3349 			bitmap_clear(ud->rchan_map, rm_desc->start,
3350 				     rm_desc->num);
3351 			dev_dbg(dev, "ti-sci-res: rchan: %d:%d\n",
3352 				rm_desc->start, rm_desc->num);
3353 		}
3354 	}
3355 
3356 	irq_res.sets += rm_res->sets;
3357 	irq_res.desc = kcalloc(irq_res.sets, sizeof(*irq_res.desc), GFP_KERNEL);
3358 	rm_res = tisci_rm->rm_ranges[RM_RANGE_TCHAN];
3359 	for (i = 0; i < rm_res->sets; i++) {
3360 		irq_res.desc[i].start = rm_res->desc[i].start;
3361 		irq_res.desc[i].num = rm_res->desc[i].num;
3362 	}
3363 	rm_res = tisci_rm->rm_ranges[RM_RANGE_RCHAN];
3364 	for (j = 0; j < rm_res->sets; j++, i++) {
3365 		irq_res.desc[i].start = rm_res->desc[j].start +
3366 					ud->match_data->rchan_oes_offset;
3367 		irq_res.desc[i].num = rm_res->desc[j].num;
3368 	}
3369 	ret = ti_sci_inta_msi_domain_alloc_irqs(ud->dev, &irq_res);
3370 	kfree(irq_res.desc);
3371 	if (ret) {
3372 		dev_err(ud->dev, "Failed to allocate MSI interrupts\n");
3373 		return ret;
3374 	}
3375 
3376 	/* GP rflow ranges */
3377 	rm_res = tisci_rm->rm_ranges[RM_RANGE_RFLOW];
3378 	if (IS_ERR(rm_res)) {
3379 		/* all gp flows are assigned exclusively to Linux */
3380 		bitmap_clear(ud->rflow_gp_map, ud->rchan_cnt,
3381 			     ud->rflow_cnt - ud->rchan_cnt);
3382 	} else {
3383 		for (i = 0; i < rm_res->sets; i++) {
3384 			rm_desc = &rm_res->desc[i];
3385 			bitmap_clear(ud->rflow_gp_map, rm_desc->start,
3386 				     rm_desc->num);
3387 			dev_dbg(dev, "ti-sci-res: rflow: %d:%d\n",
3388 				rm_desc->start, rm_desc->num);
3389 		}
3390 	}
3391 
3392 	ch_count -= bitmap_weight(ud->tchan_map, ud->tchan_cnt);
3393 	ch_count -= bitmap_weight(ud->rchan_map, ud->rchan_cnt);
3394 	if (!ch_count)
3395 		return -ENODEV;
3396 
3397 	ud->channels = devm_kcalloc(dev, ch_count, sizeof(*ud->channels),
3398 				    GFP_KERNEL);
3399 	if (!ud->channels)
3400 		return -ENOMEM;
3401 
3402 	dev_info(dev, "Channels: %d (tchan: %u, rchan: %u, gp-rflow: %u)\n",
3403 		 ch_count,
3404 		 ud->tchan_cnt - bitmap_weight(ud->tchan_map, ud->tchan_cnt),
3405 		 ud->rchan_cnt - bitmap_weight(ud->rchan_map, ud->rchan_cnt),
3406 		 ud->rflow_cnt - bitmap_weight(ud->rflow_gp_map,
3407 					       ud->rflow_cnt));
3408 
3409 	return ch_count;
3410 }
3411 
3412 static int udma_setup_rx_flush(struct udma_dev *ud)
3413 {
3414 	struct udma_rx_flush *rx_flush = &ud->rx_flush;
3415 	struct cppi5_desc_hdr_t *tr_desc;
3416 	struct cppi5_tr_type1_t *tr_req;
3417 	struct cppi5_host_desc_t *desc;
3418 	struct device *dev = ud->dev;
3419 	struct udma_hwdesc *hwdesc;
3420 	size_t tr_size;
3421 
3422 	/* Allocate 1K buffer for discarded data on RX channel teardown */
3423 	rx_flush->buffer_size = SZ_1K;
3424 	rx_flush->buffer_vaddr = devm_kzalloc(dev, rx_flush->buffer_size,
3425 					      GFP_KERNEL);
3426 	if (!rx_flush->buffer_vaddr)
3427 		return -ENOMEM;
3428 
3429 	rx_flush->buffer_paddr = dma_map_single(dev, rx_flush->buffer_vaddr,
3430 						rx_flush->buffer_size,
3431 						DMA_TO_DEVICE);
3432 	if (dma_mapping_error(dev, rx_flush->buffer_paddr))
3433 		return -ENOMEM;
3434 
3435 	/* Set up descriptor to be used for TR mode */
3436 	hwdesc = &rx_flush->hwdescs[0];
3437 	tr_size = sizeof(struct cppi5_tr_type1_t);
3438 	hwdesc->cppi5_desc_size = cppi5_trdesc_calc_size(tr_size, 1);
3439 	hwdesc->cppi5_desc_size = ALIGN(hwdesc->cppi5_desc_size,
3440 					ud->desc_align);
3441 
3442 	hwdesc->cppi5_desc_vaddr = devm_kzalloc(dev, hwdesc->cppi5_desc_size,
3443 						GFP_KERNEL);
3444 	if (!hwdesc->cppi5_desc_vaddr)
3445 		return -ENOMEM;
3446 
3447 	hwdesc->cppi5_desc_paddr = dma_map_single(dev, hwdesc->cppi5_desc_vaddr,
3448 						  hwdesc->cppi5_desc_size,
3449 						  DMA_TO_DEVICE);
3450 	if (dma_mapping_error(dev, hwdesc->cppi5_desc_paddr))
3451 		return -ENOMEM;
3452 
3453 	/* Start of the TR req records */
3454 	hwdesc->tr_req_base = hwdesc->cppi5_desc_vaddr + tr_size;
3455 	/* Start address of the TR response array */
3456 	hwdesc->tr_resp_base = hwdesc->tr_req_base + tr_size;
3457 
3458 	tr_desc = hwdesc->cppi5_desc_vaddr;
3459 	cppi5_trdesc_init(tr_desc, 1, tr_size, 0, 0);
3460 	cppi5_desc_set_pktids(tr_desc, 0, CPPI5_INFO1_DESC_FLOWID_DEFAULT);
3461 	cppi5_desc_set_retpolicy(tr_desc, 0, 0);
3462 
3463 	tr_req = hwdesc->tr_req_base;
3464 	cppi5_tr_init(&tr_req->flags, CPPI5_TR_TYPE1, false, false,
3465 		      CPPI5_TR_EVENT_SIZE_COMPLETION, 0);
3466 	cppi5_tr_csf_set(&tr_req->flags, CPPI5_TR_CSF_SUPR_EVT);
3467 
3468 	tr_req->addr = rx_flush->buffer_paddr;
3469 	tr_req->icnt0 = rx_flush->buffer_size;
3470 	tr_req->icnt1 = 1;
3471 
3472 	/* Set up descriptor to be used for packet mode */
3473 	hwdesc = &rx_flush->hwdescs[1];
3474 	hwdesc->cppi5_desc_size = ALIGN(sizeof(struct cppi5_host_desc_t) +
3475 					CPPI5_INFO0_HDESC_EPIB_SIZE +
3476 					CPPI5_INFO0_HDESC_PSDATA_MAX_SIZE,
3477 					ud->desc_align);
3478 
3479 	hwdesc->cppi5_desc_vaddr = devm_kzalloc(dev, hwdesc->cppi5_desc_size,
3480 						GFP_KERNEL);
3481 	if (!hwdesc->cppi5_desc_vaddr)
3482 		return -ENOMEM;
3483 
3484 	hwdesc->cppi5_desc_paddr = dma_map_single(dev, hwdesc->cppi5_desc_vaddr,
3485 						  hwdesc->cppi5_desc_size,
3486 						  DMA_TO_DEVICE);
3487 	if (dma_mapping_error(dev, hwdesc->cppi5_desc_paddr))
3488 		return -ENOMEM;
3489 
3490 	desc = hwdesc->cppi5_desc_vaddr;
3491 	cppi5_hdesc_init(desc, 0, 0);
3492 	cppi5_desc_set_pktids(&desc->hdr, 0, CPPI5_INFO1_DESC_FLOWID_DEFAULT);
3493 	cppi5_desc_set_retpolicy(&desc->hdr, 0, 0);
3494 
3495 	cppi5_hdesc_attach_buf(desc,
3496 			       rx_flush->buffer_paddr, rx_flush->buffer_size,
3497 			       rx_flush->buffer_paddr, rx_flush->buffer_size);
3498 
3499 	dma_sync_single_for_device(dev, hwdesc->cppi5_desc_paddr,
3500 				   hwdesc->cppi5_desc_size, DMA_TO_DEVICE);
3501 	return 0;
3502 }
3503 
3504 #ifdef CONFIG_DEBUG_FS
3505 static void udma_dbg_summary_show_chan(struct seq_file *s,
3506 				       struct dma_chan *chan)
3507 {
3508 	struct udma_chan *uc = to_udma_chan(chan);
3509 	struct udma_chan_config *ucc = &uc->config;
3510 
3511 	seq_printf(s, " %-13s| %s", dma_chan_name(chan),
3512 		   chan->dbg_client_name ?: "in-use");
3513 	seq_printf(s, " (%s, ", dmaengine_get_direction_text(uc->config.dir));
3514 
3515 	switch (uc->config.dir) {
3516 	case DMA_MEM_TO_MEM:
3517 		seq_printf(s, "chan%d pair [0x%04x -> 0x%04x], ", uc->tchan->id,
3518 			   ucc->src_thread, ucc->dst_thread);
3519 		break;
3520 	case DMA_DEV_TO_MEM:
3521 		seq_printf(s, "rchan%d [0x%04x -> 0x%04x], ", uc->rchan->id,
3522 			   ucc->src_thread, ucc->dst_thread);
3523 		break;
3524 	case DMA_MEM_TO_DEV:
3525 		seq_printf(s, "tchan%d [0x%04x -> 0x%04x], ", uc->tchan->id,
3526 			   ucc->src_thread, ucc->dst_thread);
3527 		break;
3528 	default:
3529 		seq_printf(s, ")\n");
3530 		return;
3531 	}
3532 
3533 	if (ucc->ep_type == PSIL_EP_NATIVE) {
3534 		seq_printf(s, "PSI-L Native");
3535 		if (ucc->metadata_size) {
3536 			seq_printf(s, "[%s", ucc->needs_epib ? " EPIB" : "");
3537 			if (ucc->psd_size)
3538 				seq_printf(s, " PSDsize:%u", ucc->psd_size);
3539 			seq_printf(s, " ]");
3540 		}
3541 	} else {
3542 		seq_printf(s, "PDMA");
3543 		if (ucc->enable_acc32 || ucc->enable_burst)
3544 			seq_printf(s, "[%s%s ]",
3545 				   ucc->enable_acc32 ? " ACC32" : "",
3546 				   ucc->enable_burst ? " BURST" : "");
3547 	}
3548 
3549 	seq_printf(s, ", %s)\n", ucc->pkt_mode ? "Packet mode" : "TR mode");
3550 }
3551 
3552 static void udma_dbg_summary_show(struct seq_file *s,
3553 				  struct dma_device *dma_dev)
3554 {
3555 	struct dma_chan *chan;
3556 
3557 	list_for_each_entry(chan, &dma_dev->channels, device_node) {
3558 		if (chan->client_count)
3559 			udma_dbg_summary_show_chan(s, chan);
3560 	}
3561 }
3562 #endif /* CONFIG_DEBUG_FS */
3563 
3564 #define TI_UDMAC_BUSWIDTHS	(BIT(DMA_SLAVE_BUSWIDTH_1_BYTE) | \
3565 				 BIT(DMA_SLAVE_BUSWIDTH_2_BYTES) | \
3566 				 BIT(DMA_SLAVE_BUSWIDTH_3_BYTES) | \
3567 				 BIT(DMA_SLAVE_BUSWIDTH_4_BYTES) | \
3568 				 BIT(DMA_SLAVE_BUSWIDTH_8_BYTES))
3569 
3570 static int udma_probe(struct platform_device *pdev)
3571 {
3572 	struct device_node *navss_node = pdev->dev.parent->of_node;
3573 	struct device *dev = &pdev->dev;
3574 	struct udma_dev *ud;
3575 	const struct of_device_id *match;
3576 	int i, ret;
3577 	int ch_count;
3578 
3579 	ret = dma_coerce_mask_and_coherent(dev, DMA_BIT_MASK(48));
3580 	if (ret)
3581 		dev_err(dev, "failed to set dma mask stuff\n");
3582 
3583 	ud = devm_kzalloc(dev, sizeof(*ud), GFP_KERNEL);
3584 	if (!ud)
3585 		return -ENOMEM;
3586 
3587 	ret = udma_get_mmrs(pdev, ud);
3588 	if (ret)
3589 		return ret;
3590 
3591 	ud->tisci_rm.tisci = ti_sci_get_by_phandle(dev->of_node, "ti,sci");
3592 	if (IS_ERR(ud->tisci_rm.tisci))
3593 		return PTR_ERR(ud->tisci_rm.tisci);
3594 
3595 	ret = of_property_read_u32(dev->of_node, "ti,sci-dev-id",
3596 				   &ud->tisci_rm.tisci_dev_id);
3597 	if (ret) {
3598 		dev_err(dev, "ti,sci-dev-id read failure %d\n", ret);
3599 		return ret;
3600 	}
3601 	pdev->id = ud->tisci_rm.tisci_dev_id;
3602 
3603 	ret = of_property_read_u32(navss_node, "ti,sci-dev-id",
3604 				   &ud->tisci_rm.tisci_navss_dev_id);
3605 	if (ret) {
3606 		dev_err(dev, "NAVSS ti,sci-dev-id read failure %d\n", ret);
3607 		return ret;
3608 	}
3609 
3610 	ret = of_property_read_u32(navss_node, "ti,udma-atype", &ud->atype);
3611 	if (!ret && ud->atype > 2) {
3612 		dev_err(dev, "Invalid atype: %u\n", ud->atype);
3613 		return -EINVAL;
3614 	}
3615 
3616 	ud->tisci_rm.tisci_udmap_ops = &ud->tisci_rm.tisci->ops.rm_udmap_ops;
3617 	ud->tisci_rm.tisci_psil_ops = &ud->tisci_rm.tisci->ops.rm_psil_ops;
3618 
3619 	ud->ringacc = of_k3_ringacc_get_by_phandle(dev->of_node, "ti,ringacc");
3620 	if (IS_ERR(ud->ringacc))
3621 		return PTR_ERR(ud->ringacc);
3622 
3623 	dev->msi_domain = of_msi_get_domain(dev, dev->of_node,
3624 					    DOMAIN_BUS_TI_SCI_INTA_MSI);
3625 	if (!dev->msi_domain) {
3626 		dev_err(dev, "Failed to get MSI domain\n");
3627 		return -EPROBE_DEFER;
3628 	}
3629 
3630 	match = of_match_node(udma_of_match, dev->of_node);
3631 	if (!match) {
3632 		dev_err(dev, "No compatible match found\n");
3633 		return -ENODEV;
3634 	}
3635 	ud->match_data = match->data;
3636 
3637 	dma_cap_set(DMA_SLAVE, ud->ddev.cap_mask);
3638 	dma_cap_set(DMA_CYCLIC, ud->ddev.cap_mask);
3639 
3640 	ud->ddev.device_alloc_chan_resources = udma_alloc_chan_resources;
3641 	ud->ddev.device_config = udma_slave_config;
3642 	ud->ddev.device_prep_slave_sg = udma_prep_slave_sg;
3643 	ud->ddev.device_prep_dma_cyclic = udma_prep_dma_cyclic;
3644 	ud->ddev.device_issue_pending = udma_issue_pending;
3645 	ud->ddev.device_tx_status = udma_tx_status;
3646 	ud->ddev.device_pause = udma_pause;
3647 	ud->ddev.device_resume = udma_resume;
3648 	ud->ddev.device_terminate_all = udma_terminate_all;
3649 	ud->ddev.device_synchronize = udma_synchronize;
3650 #ifdef CONFIG_DEBUG_FS
3651 	ud->ddev.dbg_summary_show = udma_dbg_summary_show;
3652 #endif
3653 
3654 	ud->ddev.device_free_chan_resources = udma_free_chan_resources;
3655 	ud->ddev.src_addr_widths = TI_UDMAC_BUSWIDTHS;
3656 	ud->ddev.dst_addr_widths = TI_UDMAC_BUSWIDTHS;
3657 	ud->ddev.directions = BIT(DMA_DEV_TO_MEM) | BIT(DMA_MEM_TO_DEV);
3658 	ud->ddev.residue_granularity = DMA_RESIDUE_GRANULARITY_BURST;
3659 	ud->ddev.copy_align = DMAENGINE_ALIGN_8_BYTES;
3660 	ud->ddev.desc_metadata_modes = DESC_METADATA_CLIENT |
3661 				       DESC_METADATA_ENGINE;
3662 	if (ud->match_data->enable_memcpy_support) {
3663 		dma_cap_set(DMA_MEMCPY, ud->ddev.cap_mask);
3664 		ud->ddev.device_prep_dma_memcpy = udma_prep_dma_memcpy;
3665 		ud->ddev.directions |= BIT(DMA_MEM_TO_MEM);
3666 	}
3667 
3668 	ud->ddev.dev = dev;
3669 	ud->dev = dev;
3670 	ud->psil_base = ud->match_data->psil_base;
3671 
3672 	INIT_LIST_HEAD(&ud->ddev.channels);
3673 	INIT_LIST_HEAD(&ud->desc_to_purge);
3674 
3675 	ch_count = udma_setup_resources(ud);
3676 	if (ch_count <= 0)
3677 		return ch_count;
3678 
3679 	spin_lock_init(&ud->lock);
3680 	INIT_WORK(&ud->purge_work, udma_purge_desc_work);
3681 
3682 	ud->desc_align = 64;
3683 	if (ud->desc_align < dma_get_cache_alignment())
3684 		ud->desc_align = dma_get_cache_alignment();
3685 
3686 	ret = udma_setup_rx_flush(ud);
3687 	if (ret)
3688 		return ret;
3689 
3690 	for (i = 0; i < ud->tchan_cnt; i++) {
3691 		struct udma_tchan *tchan = &ud->tchans[i];
3692 
3693 		tchan->id = i;
3694 		tchan->reg_rt = ud->mmrs[MMR_TCHANRT] + i * 0x1000;
3695 	}
3696 
3697 	for (i = 0; i < ud->rchan_cnt; i++) {
3698 		struct udma_rchan *rchan = &ud->rchans[i];
3699 
3700 		rchan->id = i;
3701 		rchan->reg_rt = ud->mmrs[MMR_RCHANRT] + i * 0x1000;
3702 	}
3703 
3704 	for (i = 0; i < ud->rflow_cnt; i++) {
3705 		struct udma_rflow *rflow = &ud->rflows[i];
3706 
3707 		rflow->id = i;
3708 	}
3709 
3710 	for (i = 0; i < ch_count; i++) {
3711 		struct udma_chan *uc = &ud->channels[i];
3712 
3713 		uc->ud = ud;
3714 		uc->vc.desc_free = udma_desc_free;
3715 		uc->id = i;
3716 		uc->tchan = NULL;
3717 		uc->rchan = NULL;
3718 		uc->config.remote_thread_id = -1;
3719 		uc->config.dir = DMA_MEM_TO_MEM;
3720 		uc->name = devm_kasprintf(dev, GFP_KERNEL, "%s chan%d",
3721 					  dev_name(dev), i);
3722 
3723 		vchan_init(&uc->vc, &ud->ddev);
3724 		/* Use custom vchan completion handling */
3725 		tasklet_init(&uc->vc.task, udma_vchan_complete,
3726 			     (unsigned long)&uc->vc);
3727 		init_completion(&uc->teardown_completed);
3728 	}
3729 
3730 	ret = dma_async_device_register(&ud->ddev);
3731 	if (ret) {
3732 		dev_err(dev, "failed to register slave DMA engine: %d\n", ret);
3733 		return ret;
3734 	}
3735 
3736 	platform_set_drvdata(pdev, ud);
3737 
3738 	ret = of_dma_controller_register(dev->of_node, udma_of_xlate, ud);
3739 	if (ret) {
3740 		dev_err(dev, "failed to register of_dma controller\n");
3741 		dma_async_device_unregister(&ud->ddev);
3742 	}
3743 
3744 	return ret;
3745 }
3746 
3747 static struct platform_driver udma_driver = {
3748 	.driver = {
3749 		.name	= "ti-udma",
3750 		.of_match_table = udma_of_match,
3751 		.suppress_bind_attrs = true,
3752 	},
3753 	.probe		= udma_probe,
3754 };
3755 builtin_platform_driver(udma_driver);
3756 
3757 /* Private interfaces to UDMA */
3758 #include "k3-udma-private.c"
3759