xref: /freebsd/sys/contrib/dev/mediatek/mt76/dma.c (revision 8ba4d145d351db26e07695b8e90697398c5dfec2)
1 // SPDX-License-Identifier: ISC
2 /*
3  * Copyright (C) 2016 Felix Fietkau <nbd@nbd.name>
4  */
5 
6 #include <linux/dma-mapping.h>
7 #if defined(__FreeBSD__)
8 #include <linux/cache.h>
9 #include <net/page_pool.h>
10 #endif
11 #include "mt76.h"
12 #include "dma.h"
13 
14 #if IS_ENABLED(CONFIG_NET_MEDIATEK_SOC_WED)
15 
16 #define Q_READ(_q, _field) ({						\
17 	u32 _offset = offsetof(struct mt76_queue_regs, _field);		\
18 	u32 _val;							\
19 	if ((_q)->flags & MT_QFLAG_WED)					\
20 		_val = mtk_wed_device_reg_read((_q)->wed,		\
21 					       ((_q)->wed_regs +	\
22 					        _offset));		\
23 	else								\
24 		_val = readl(&(_q)->regs->_field);			\
25 	_val;								\
26 })
27 
28 #define Q_WRITE(_q, _field, _val)	do {				\
29 	u32 _offset = offsetof(struct mt76_queue_regs, _field);		\
30 	if ((_q)->flags & MT_QFLAG_WED)					\
31 		mtk_wed_device_reg_write((_q)->wed,			\
32 					 ((_q)->wed_regs + _offset),	\
33 					 _val);				\
34 	else								\
35 		writel(_val, &(_q)->regs->_field);			\
36 } while (0)
37 
38 #else
39 
40 #define Q_READ(_q, _field)		readl(&(_q)->regs->_field)
41 #define Q_WRITE(_q, _field, _val)	writel(_val, &(_q)->regs->_field)
42 
43 #endif
44 
45 static struct mt76_txwi_cache *
mt76_alloc_txwi(struct mt76_dev * dev)46 mt76_alloc_txwi(struct mt76_dev *dev)
47 {
48 	struct mt76_txwi_cache *t;
49 	dma_addr_t addr;
50 	u8 *txwi;
51 	int size;
52 
53 	size = L1_CACHE_ALIGN(dev->drv->txwi_size + sizeof(*t));
54 	txwi = kzalloc(size, GFP_ATOMIC);
55 	if (!txwi)
56 		return NULL;
57 
58 	addr = dma_map_single(dev->dma_dev, txwi, dev->drv->txwi_size,
59 			      DMA_TO_DEVICE);
60 	if (unlikely(dma_mapping_error(dev->dma_dev, addr))) {
61 		kfree(txwi);
62 		return NULL;
63 	}
64 
65 	t = (struct mt76_txwi_cache *)(txwi + dev->drv->txwi_size);
66 	t->dma_addr = addr;
67 
68 	return t;
69 }
70 
71 static struct mt76_txwi_cache *
mt76_alloc_rxwi(struct mt76_dev * dev)72 mt76_alloc_rxwi(struct mt76_dev *dev)
73 {
74 	struct mt76_txwi_cache *t;
75 
76 	t = kzalloc(L1_CACHE_ALIGN(sizeof(*t)), GFP_ATOMIC);
77 	if (!t)
78 		return NULL;
79 
80 	t->ptr = NULL;
81 	return t;
82 }
83 
84 static struct mt76_txwi_cache *
__mt76_get_txwi(struct mt76_dev * dev)85 __mt76_get_txwi(struct mt76_dev *dev)
86 {
87 	struct mt76_txwi_cache *t = NULL;
88 
89 	spin_lock(&dev->lock);
90 	if (!list_empty(&dev->txwi_cache)) {
91 		t = list_first_entry(&dev->txwi_cache, struct mt76_txwi_cache,
92 				     list);
93 		list_del(&t->list);
94 	}
95 	spin_unlock(&dev->lock);
96 
97 	return t;
98 }
99 
100 static struct mt76_txwi_cache *
__mt76_get_rxwi(struct mt76_dev * dev)101 __mt76_get_rxwi(struct mt76_dev *dev)
102 {
103 	struct mt76_txwi_cache *t = NULL;
104 
105 	spin_lock_bh(&dev->wed_lock);
106 	if (!list_empty(&dev->rxwi_cache)) {
107 		t = list_first_entry(&dev->rxwi_cache, struct mt76_txwi_cache,
108 				     list);
109 		list_del(&t->list);
110 	}
111 	spin_unlock_bh(&dev->wed_lock);
112 
113 	return t;
114 }
115 
116 static struct mt76_txwi_cache *
mt76_get_txwi(struct mt76_dev * dev)117 mt76_get_txwi(struct mt76_dev *dev)
118 {
119 	struct mt76_txwi_cache *t = __mt76_get_txwi(dev);
120 
121 	if (t)
122 		return t;
123 
124 	return mt76_alloc_txwi(dev);
125 }
126 
127 struct mt76_txwi_cache *
mt76_get_rxwi(struct mt76_dev * dev)128 mt76_get_rxwi(struct mt76_dev *dev)
129 {
130 	struct mt76_txwi_cache *t = __mt76_get_rxwi(dev);
131 
132 	if (t)
133 		return t;
134 
135 	return mt76_alloc_rxwi(dev);
136 }
137 EXPORT_SYMBOL_GPL(mt76_get_rxwi);
138 
139 void
mt76_put_txwi(struct mt76_dev * dev,struct mt76_txwi_cache * t)140 mt76_put_txwi(struct mt76_dev *dev, struct mt76_txwi_cache *t)
141 {
142 	if (!t)
143 		return;
144 
145 	spin_lock(&dev->lock);
146 	list_add(&t->list, &dev->txwi_cache);
147 	spin_unlock(&dev->lock);
148 }
149 EXPORT_SYMBOL_GPL(mt76_put_txwi);
150 
151 void
mt76_put_rxwi(struct mt76_dev * dev,struct mt76_txwi_cache * t)152 mt76_put_rxwi(struct mt76_dev *dev, struct mt76_txwi_cache *t)
153 {
154 	if (!t)
155 		return;
156 
157 	spin_lock_bh(&dev->wed_lock);
158 	list_add(&t->list, &dev->rxwi_cache);
159 	spin_unlock_bh(&dev->wed_lock);
160 }
161 EXPORT_SYMBOL_GPL(mt76_put_rxwi);
162 
163 static void
mt76_free_pending_txwi(struct mt76_dev * dev)164 mt76_free_pending_txwi(struct mt76_dev *dev)
165 {
166 	struct mt76_txwi_cache *t;
167 
168 	local_bh_disable();
169 	while ((t = __mt76_get_txwi(dev)) != NULL) {
170 		dma_unmap_single(dev->dma_dev, t->dma_addr, dev->drv->txwi_size,
171 				 DMA_TO_DEVICE);
172 		kfree(mt76_get_txwi_ptr(dev, t));
173 	}
174 	local_bh_enable();
175 }
176 
177 void
mt76_free_pending_rxwi(struct mt76_dev * dev)178 mt76_free_pending_rxwi(struct mt76_dev *dev)
179 {
180 	struct mt76_txwi_cache *t;
181 
182 	local_bh_disable();
183 	while ((t = __mt76_get_rxwi(dev)) != NULL) {
184 		if (t->ptr)
185 			mt76_put_page_pool_buf(t->ptr, false);
186 		kfree(t);
187 	}
188 	local_bh_enable();
189 }
190 EXPORT_SYMBOL_GPL(mt76_free_pending_rxwi);
191 
192 static void
mt76_dma_sync_idx(struct mt76_dev * dev,struct mt76_queue * q)193 mt76_dma_sync_idx(struct mt76_dev *dev, struct mt76_queue *q)
194 {
195 	Q_WRITE(q, desc_base, q->desc_dma);
196 	if (q->flags & MT_QFLAG_WED_RRO_EN)
197 		Q_WRITE(q, ring_size, MT_DMA_RRO_EN | q->ndesc);
198 	else
199 		Q_WRITE(q, ring_size, q->ndesc);
200 	q->head = Q_READ(q, dma_idx);
201 	q->tail = q->head;
202 }
203 
__mt76_dma_queue_reset(struct mt76_dev * dev,struct mt76_queue * q,bool reset_idx)204 void __mt76_dma_queue_reset(struct mt76_dev *dev, struct mt76_queue *q,
205 			    bool reset_idx)
206 {
207 	if (!q || !q->ndesc)
208 		return;
209 
210 	if (!mt76_queue_is_wed_rro_ind(q)) {
211 		int i;
212 
213 		/* clear descriptors */
214 		for (i = 0; i < q->ndesc; i++)
215 			q->desc[i].ctrl = cpu_to_le32(MT_DMA_CTL_DMA_DONE);
216 	}
217 
218 	if (reset_idx) {
219 		Q_WRITE(q, cpu_idx, 0);
220 		Q_WRITE(q, dma_idx, 0);
221 	}
222 	mt76_dma_sync_idx(dev, q);
223 }
224 
mt76_dma_queue_reset(struct mt76_dev * dev,struct mt76_queue * q)225 void mt76_dma_queue_reset(struct mt76_dev *dev, struct mt76_queue *q)
226 {
227 	__mt76_dma_queue_reset(dev, q, true);
228 }
229 
230 static int
mt76_dma_add_rx_buf(struct mt76_dev * dev,struct mt76_queue * q,struct mt76_queue_buf * buf,void * data)231 mt76_dma_add_rx_buf(struct mt76_dev *dev, struct mt76_queue *q,
232 		    struct mt76_queue_buf *buf, void *data)
233 {
234 	struct mt76_queue_entry *entry = &q->entry[q->head];
235 	struct mt76_txwi_cache *txwi = NULL;
236 	struct mt76_desc *desc;
237 	int idx = q->head;
238 	u32 buf1 = 0, ctrl;
239 	int rx_token;
240 
241 	if (mt76_queue_is_wed_rro_ind(q)) {
242 		struct mt76_wed_rro_desc *rro_desc;
243 
244 		rro_desc = (struct mt76_wed_rro_desc *)q->desc;
245 		data = &rro_desc[q->head];
246 		goto done;
247 	}
248 
249 	desc = &q->desc[q->head];
250 	ctrl = FIELD_PREP(MT_DMA_CTL_SD_LEN0, buf[0].len);
251 #ifdef CONFIG_ARCH_DMA_ADDR_T_64BIT
252 	buf1 = FIELD_PREP(MT_DMA_CTL_SDP0_H, buf->addr >> 32);
253 #endif
254 
255 	if (mt76_queue_is_wed_rx(q)) {
256 		txwi = mt76_get_rxwi(dev);
257 		if (!txwi)
258 			return -ENOMEM;
259 
260 		rx_token = mt76_rx_token_consume(dev, data, txwi, buf->addr);
261 		if (rx_token < 0) {
262 			mt76_put_rxwi(dev, txwi);
263 			return -ENOMEM;
264 		}
265 
266 		buf1 |= FIELD_PREP(MT_DMA_CTL_TOKEN, rx_token);
267 		ctrl |= MT_DMA_CTL_TO_HOST;
268 	}
269 
270 	WRITE_ONCE(desc->buf0, cpu_to_le32(buf->addr));
271 	WRITE_ONCE(desc->buf1, cpu_to_le32(buf1));
272 	WRITE_ONCE(desc->ctrl, cpu_to_le32(ctrl));
273 	WRITE_ONCE(desc->info, 0);
274 
275 done:
276 	entry->dma_addr[0] = buf->addr;
277 	entry->dma_len[0] = buf->len;
278 	entry->txwi = txwi;
279 	entry->buf = data;
280 	entry->wcid = 0xffff;
281 	entry->skip_buf1 = true;
282 	q->head = (q->head + 1) % q->ndesc;
283 	q->queued++;
284 
285 	return idx;
286 }
287 
288 static int
mt76_dma_add_buf(struct mt76_dev * dev,struct mt76_queue * q,struct mt76_queue_buf * buf,int nbufs,u32 info,struct sk_buff * skb,void * txwi)289 mt76_dma_add_buf(struct mt76_dev *dev, struct mt76_queue *q,
290 		 struct mt76_queue_buf *buf, int nbufs, u32 info,
291 		 struct sk_buff *skb, void *txwi)
292 {
293 	struct mt76_queue_entry *entry;
294 	struct mt76_desc *desc;
295 	int i, idx = -1;
296 	u32 ctrl, next;
297 
298 	if (txwi) {
299 		q->entry[q->head].txwi = DMA_DUMMY_DATA;
300 		q->entry[q->head].skip_buf0 = true;
301 	}
302 
303 	for (i = 0; i < nbufs; i += 2, buf += 2) {
304 		u32 buf0 = buf[0].addr, buf1 = 0;
305 
306 		idx = q->head;
307 		next = (q->head + 1) % q->ndesc;
308 
309 		desc = &q->desc[idx];
310 		entry = &q->entry[idx];
311 
312 		if (buf[0].skip_unmap)
313 			entry->skip_buf0 = true;
314 		entry->skip_buf1 = i == nbufs - 1;
315 
316 		entry->dma_addr[0] = buf[0].addr;
317 		entry->dma_len[0] = buf[0].len;
318 
319 		ctrl = FIELD_PREP(MT_DMA_CTL_SD_LEN0, buf[0].len);
320 #ifdef CONFIG_ARCH_DMA_ADDR_T_64BIT
321 		info |= FIELD_PREP(MT_DMA_CTL_SDP0_H, buf[0].addr >> 32);
322 #endif
323 		if (i < nbufs - 1) {
324 			entry->dma_addr[1] = buf[1].addr;
325 			entry->dma_len[1] = buf[1].len;
326 			buf1 = buf[1].addr;
327 			ctrl |= FIELD_PREP(MT_DMA_CTL_SD_LEN1, buf[1].len);
328 #ifdef CONFIG_ARCH_DMA_ADDR_T_64BIT
329 			info |= FIELD_PREP(MT_DMA_CTL_SDP1_H,
330 					   buf[1].addr >> 32);
331 #endif
332 			if (buf[1].skip_unmap)
333 				entry->skip_buf1 = true;
334 		}
335 
336 		if (i == nbufs - 1)
337 			ctrl |= MT_DMA_CTL_LAST_SEC0;
338 		else if (i == nbufs - 2)
339 			ctrl |= MT_DMA_CTL_LAST_SEC1;
340 
341 		WRITE_ONCE(desc->buf0, cpu_to_le32(buf0));
342 		WRITE_ONCE(desc->buf1, cpu_to_le32(buf1));
343 		WRITE_ONCE(desc->info, cpu_to_le32(info));
344 		WRITE_ONCE(desc->ctrl, cpu_to_le32(ctrl));
345 
346 		q->head = next;
347 		q->queued++;
348 	}
349 
350 	q->entry[idx].txwi = txwi;
351 	q->entry[idx].skb = skb;
352 	q->entry[idx].wcid = 0xffff;
353 
354 	return idx;
355 }
356 
357 static void
mt76_dma_tx_cleanup_idx(struct mt76_dev * dev,struct mt76_queue * q,int idx,struct mt76_queue_entry * prev_e)358 mt76_dma_tx_cleanup_idx(struct mt76_dev *dev, struct mt76_queue *q, int idx,
359 			struct mt76_queue_entry *prev_e)
360 {
361 	struct mt76_queue_entry *e = &q->entry[idx];
362 
363 	if (!e->skip_buf0)
364 		dma_unmap_single(dev->dma_dev, e->dma_addr[0], e->dma_len[0],
365 				 DMA_TO_DEVICE);
366 
367 	if (!e->skip_buf1)
368 		dma_unmap_single(dev->dma_dev, e->dma_addr[1], e->dma_len[1],
369 				 DMA_TO_DEVICE);
370 
371 	if (e->txwi == DMA_DUMMY_DATA)
372 		e->txwi = NULL;
373 
374 	*prev_e = *e;
375 	memset(e, 0, sizeof(*e));
376 }
377 
378 static void
mt76_dma_kick_queue(struct mt76_dev * dev,struct mt76_queue * q)379 mt76_dma_kick_queue(struct mt76_dev *dev, struct mt76_queue *q)
380 {
381 	wmb();
382 	Q_WRITE(q, cpu_idx, q->head);
383 }
384 
385 static void
mt76_dma_tx_cleanup(struct mt76_dev * dev,struct mt76_queue * q,bool flush)386 mt76_dma_tx_cleanup(struct mt76_dev *dev, struct mt76_queue *q, bool flush)
387 {
388 	struct mt76_queue_entry entry;
389 	int last;
390 
391 	if (!q || !q->ndesc)
392 		return;
393 
394 	spin_lock_bh(&q->cleanup_lock);
395 	if (flush)
396 		last = -1;
397 	else
398 		last = Q_READ(q, dma_idx);
399 
400 	while (q->queued > 0 && q->tail != last) {
401 		mt76_dma_tx_cleanup_idx(dev, q, q->tail, &entry);
402 		mt76_queue_tx_complete(dev, q, &entry);
403 
404 		if (entry.txwi) {
405 			if (!(dev->drv->drv_flags & MT_DRV_TXWI_NO_FREE))
406 				mt76_put_txwi(dev, entry.txwi);
407 		}
408 
409 		if (!flush && q->tail == last)
410 			last = Q_READ(q, dma_idx);
411 	}
412 	spin_unlock_bh(&q->cleanup_lock);
413 
414 	if (flush) {
415 		spin_lock_bh(&q->lock);
416 		mt76_dma_sync_idx(dev, q);
417 		mt76_dma_kick_queue(dev, q);
418 		spin_unlock_bh(&q->lock);
419 	}
420 
421 	if (!q->queued)
422 		wake_up(&dev->tx_wait);
423 }
424 
425 static void *
mt76_dma_get_buf(struct mt76_dev * dev,struct mt76_queue * q,int idx,int * len,u32 * info,bool * more,bool * drop)426 mt76_dma_get_buf(struct mt76_dev *dev, struct mt76_queue *q, int idx,
427 		 int *len, u32 *info, bool *more, bool *drop)
428 {
429 	struct mt76_queue_entry *e = &q->entry[idx];
430 	struct mt76_desc *desc = &q->desc[idx];
431 	u32 ctrl, desc_info, buf1;
432 	void *buf = e->buf;
433 
434 	if (mt76_queue_is_wed_rro_ind(q))
435 		goto done;
436 
437 	ctrl = le32_to_cpu(READ_ONCE(desc->ctrl));
438 	if (len) {
439 		*len = FIELD_GET(MT_DMA_CTL_SD_LEN0, ctrl);
440 		*more = !(ctrl & MT_DMA_CTL_LAST_SEC0);
441 	}
442 
443 	desc_info = le32_to_cpu(desc->info);
444 	if (info)
445 		*info = desc_info;
446 
447 	buf1 = le32_to_cpu(desc->buf1);
448 	mt76_dma_should_drop_buf(drop, ctrl, buf1, desc_info);
449 
450 	if (mt76_queue_is_wed_rx(q)) {
451 		u32 token = FIELD_GET(MT_DMA_CTL_TOKEN, buf1);
452 		struct mt76_txwi_cache *t = mt76_rx_token_release(dev, token);
453 
454 		if (!t)
455 			return NULL;
456 
457 		dma_sync_single_for_cpu(dev->dma_dev, t->dma_addr,
458 				SKB_WITH_OVERHEAD(q->buf_size),
459 				page_pool_get_dma_dir(q->page_pool));
460 
461 		buf = t->ptr;
462 		t->dma_addr = 0;
463 		t->ptr = NULL;
464 
465 		mt76_put_rxwi(dev, t);
466 		if (drop)
467 			*drop |= !!(buf1 & MT_DMA_CTL_WO_DROP);
468 	} else {
469 		dma_sync_single_for_cpu(dev->dma_dev, e->dma_addr[0],
470 				SKB_WITH_OVERHEAD(q->buf_size),
471 				page_pool_get_dma_dir(q->page_pool));
472 	}
473 
474 done:
475 	e->buf = NULL;
476 	return buf;
477 }
478 
479 static void *
mt76_dma_dequeue(struct mt76_dev * dev,struct mt76_queue * q,bool flush,int * len,u32 * info,bool * more,bool * drop)480 mt76_dma_dequeue(struct mt76_dev *dev, struct mt76_queue *q, bool flush,
481 		 int *len, u32 *info, bool *more, bool *drop)
482 {
483 	int idx = q->tail;
484 
485 	*more = false;
486 	if (!q->queued)
487 		return NULL;
488 
489 	if (mt76_queue_is_wed_rro_data(q))
490 		return NULL;
491 
492 	if (!mt76_queue_is_wed_rro_ind(q)) {
493 		if (flush)
494 			q->desc[idx].ctrl |= cpu_to_le32(MT_DMA_CTL_DMA_DONE);
495 		else if (!(q->desc[idx].ctrl & cpu_to_le32(MT_DMA_CTL_DMA_DONE)))
496 			return NULL;
497 	}
498 
499 	q->tail = (q->tail + 1) % q->ndesc;
500 	q->queued--;
501 
502 	return mt76_dma_get_buf(dev, q, idx, len, info, more, drop);
503 }
504 
505 static int
mt76_dma_tx_queue_skb_raw(struct mt76_dev * dev,struct mt76_queue * q,struct sk_buff * skb,u32 tx_info)506 mt76_dma_tx_queue_skb_raw(struct mt76_dev *dev, struct mt76_queue *q,
507 			  struct sk_buff *skb, u32 tx_info)
508 {
509 	struct mt76_queue_buf buf = {};
510 	dma_addr_t addr;
511 
512 	if (test_bit(MT76_MCU_RESET, &dev->phy.state))
513 		goto error;
514 
515 	if (q->queued + 1 >= q->ndesc - 1)
516 		goto error;
517 
518 	addr = dma_map_single(dev->dma_dev, skb->data, skb->len,
519 			      DMA_TO_DEVICE);
520 	if (unlikely(dma_mapping_error(dev->dma_dev, addr)))
521 		goto error;
522 
523 	buf.addr = addr;
524 	buf.len = skb->len;
525 
526 	spin_lock_bh(&q->lock);
527 	mt76_dma_add_buf(dev, q, &buf, 1, tx_info, skb, NULL);
528 	mt76_dma_kick_queue(dev, q);
529 	spin_unlock_bh(&q->lock);
530 
531 	return 0;
532 
533 error:
534 	dev_kfree_skb(skb);
535 	return -ENOMEM;
536 }
537 
538 static int
mt76_dma_tx_queue_skb(struct mt76_phy * phy,struct mt76_queue * q,enum mt76_txq_id qid,struct sk_buff * skb,struct mt76_wcid * wcid,struct ieee80211_sta * sta)539 mt76_dma_tx_queue_skb(struct mt76_phy *phy, struct mt76_queue *q,
540 		      enum mt76_txq_id qid, struct sk_buff *skb,
541 		      struct mt76_wcid *wcid, struct ieee80211_sta *sta)
542 {
543 	struct ieee80211_tx_status status = {
544 		.sta = sta,
545 	};
546 	struct mt76_tx_info tx_info = {
547 		.skb = skb,
548 	};
549 	struct mt76_dev *dev = phy->dev;
550 	struct ieee80211_hw *hw;
551 	int len, n = 0, ret = -ENOMEM;
552 	struct mt76_txwi_cache *t;
553 	struct sk_buff *iter;
554 	dma_addr_t addr;
555 	u8 *txwi;
556 
557 	if (test_bit(MT76_RESET, &phy->state))
558 		goto free_skb;
559 
560 	t = mt76_get_txwi(dev);
561 	if (!t)
562 		goto free_skb;
563 
564 	txwi = mt76_get_txwi_ptr(dev, t);
565 
566 	skb->prev = skb->next = NULL;
567 	if (dev->drv->drv_flags & MT_DRV_TX_ALIGNED4_SKBS)
568 		mt76_insert_hdr_pad(skb);
569 
570 	len = skb_headlen(skb);
571 	addr = dma_map_single(dev->dma_dev, skb->data, len, DMA_TO_DEVICE);
572 	if (unlikely(dma_mapping_error(dev->dma_dev, addr)))
573 		goto free;
574 
575 	tx_info.buf[n].addr = t->dma_addr;
576 	tx_info.buf[n++].len = dev->drv->txwi_size;
577 	tx_info.buf[n].addr = addr;
578 	tx_info.buf[n++].len = len;
579 
580 	skb_walk_frags(skb, iter) {
581 		if (n == ARRAY_SIZE(tx_info.buf))
582 			goto unmap;
583 
584 		addr = dma_map_single(dev->dma_dev, iter->data, iter->len,
585 				      DMA_TO_DEVICE);
586 		if (unlikely(dma_mapping_error(dev->dma_dev, addr)))
587 			goto unmap;
588 
589 		tx_info.buf[n].addr = addr;
590 		tx_info.buf[n++].len = iter->len;
591 	}
592 	tx_info.nbuf = n;
593 
594 	if (q->queued + (tx_info.nbuf + 1) / 2 >= q->ndesc - 1) {
595 		ret = -ENOMEM;
596 		goto unmap;
597 	}
598 
599 	dma_sync_single_for_cpu(dev->dma_dev, t->dma_addr, dev->drv->txwi_size,
600 				DMA_TO_DEVICE);
601 	ret = dev->drv->tx_prepare_skb(dev, txwi, qid, wcid, sta, &tx_info);
602 	dma_sync_single_for_device(dev->dma_dev, t->dma_addr, dev->drv->txwi_size,
603 				   DMA_TO_DEVICE);
604 	if (ret < 0)
605 		goto unmap;
606 
607 	return mt76_dma_add_buf(dev, q, tx_info.buf, tx_info.nbuf,
608 				tx_info.info, tx_info.skb, t);
609 
610 unmap:
611 	for (n--; n > 0; n--)
612 		dma_unmap_single(dev->dma_dev, tx_info.buf[n].addr,
613 				 tx_info.buf[n].len, DMA_TO_DEVICE);
614 
615 free:
616 #ifdef CONFIG_NL80211_TESTMODE
617 	/* fix tx_done accounting on queue overflow */
618 	if (mt76_is_testmode_skb(dev, skb, &hw)) {
619 		struct mt76_phy *phy = hw->priv;
620 
621 		if (tx_info.skb == phy->test.tx_skb)
622 			phy->test.tx_done--;
623 	}
624 #endif
625 
626 	mt76_put_txwi(dev, t);
627 
628 free_skb:
629 	status.skb = tx_info.skb;
630 	hw = mt76_tx_status_get_hw(dev, tx_info.skb);
631 	spin_lock_bh(&dev->rx_lock);
632 	ieee80211_tx_status_ext(hw, &status);
633 	spin_unlock_bh(&dev->rx_lock);
634 
635 	return ret;
636 }
637 
638 static int
mt76_dma_rx_fill_buf(struct mt76_dev * dev,struct mt76_queue * q,bool allow_direct)639 mt76_dma_rx_fill_buf(struct mt76_dev *dev, struct mt76_queue *q,
640 		     bool allow_direct)
641 {
642 	int len = SKB_WITH_OVERHEAD(q->buf_size);
643 	int frames = 0;
644 
645 	if (!q->ndesc)
646 		return 0;
647 
648 	while (q->queued < q->ndesc - 1) {
649 		struct mt76_queue_buf qbuf = {};
650 		enum dma_data_direction dir;
651 		dma_addr_t addr;
652 		int offset;
653 		void *buf = NULL;
654 
655 		if (mt76_queue_is_wed_rro_ind(q))
656 			goto done;
657 
658 		buf = mt76_get_page_pool_buf(q, &offset, q->buf_size);
659 		if (!buf)
660 			break;
661 
662 		addr = page_pool_get_dma_addr(virt_to_head_page(buf)) + offset;
663 		dir = page_pool_get_dma_dir(q->page_pool);
664 		dma_sync_single_for_device(dev->dma_dev, addr, len, dir);
665 
666 		qbuf.addr = addr + q->buf_offset;
667 done:
668 		qbuf.len = len - q->buf_offset;
669 		qbuf.skip_unmap = false;
670 		if (mt76_dma_add_rx_buf(dev, q, &qbuf, buf) < 0) {
671 			mt76_put_page_pool_buf(buf, allow_direct);
672 			break;
673 		}
674 		frames++;
675 	}
676 
677 	if (frames || mt76_queue_is_wed_rx(q))
678 		mt76_dma_kick_queue(dev, q);
679 
680 	return frames;
681 }
682 
mt76_dma_rx_fill(struct mt76_dev * dev,struct mt76_queue * q,bool allow_direct)683 int mt76_dma_rx_fill(struct mt76_dev *dev, struct mt76_queue *q,
684 		     bool allow_direct)
685 {
686 	int frames;
687 
688 	if (!q->ndesc)
689 		return 0;
690 
691 	spin_lock_bh(&q->lock);
692 	frames = mt76_dma_rx_fill_buf(dev, q, allow_direct);
693 	spin_unlock_bh(&q->lock);
694 
695 	return frames;
696 }
697 
698 static int
mt76_dma_alloc_queue(struct mt76_dev * dev,struct mt76_queue * q,int idx,int n_desc,int bufsize,u32 ring_base)699 mt76_dma_alloc_queue(struct mt76_dev *dev, struct mt76_queue *q,
700 		     int idx, int n_desc, int bufsize,
701 		     u32 ring_base)
702 {
703 	int ret, size;
704 
705 	spin_lock_init(&q->lock);
706 	spin_lock_init(&q->cleanup_lock);
707 
708 #if defined(__linux__)
709 	q->regs = dev->mmio.regs + ring_base + idx * MT_RING_SIZE;
710 #elif defined(__FreeBSD__)
711 	q->regs = (void *)((u8 *)dev->mmio.regs + ring_base + idx * MT_RING_SIZE);
712 #endif
713 	q->ndesc = n_desc;
714 	q->buf_size = bufsize;
715 	q->hw_idx = idx;
716 
717 	size = mt76_queue_is_wed_rro_ind(q) ? sizeof(struct mt76_wed_rro_desc)
718 					    : sizeof(struct mt76_desc);
719 	q->desc = dmam_alloc_coherent(dev->dma_dev, q->ndesc * size,
720 				      &q->desc_dma, GFP_KERNEL);
721 	if (!q->desc)
722 		return -ENOMEM;
723 
724 	if (mt76_queue_is_wed_rro_ind(q)) {
725 		struct mt76_wed_rro_desc *rro_desc;
726 		int i;
727 
728 		rro_desc = (struct mt76_wed_rro_desc *)q->desc;
729 		for (i = 0; i < q->ndesc; i++) {
730 			struct mt76_wed_rro_ind *cmd;
731 
732 			cmd = (struct mt76_wed_rro_ind *)&rro_desc[i];
733 			cmd->magic_cnt = MT_DMA_WED_IND_CMD_CNT - 1;
734 		}
735 	}
736 
737 	size = q->ndesc * sizeof(*q->entry);
738 	q->entry = devm_kzalloc(dev->dev, size, GFP_KERNEL);
739 	if (!q->entry)
740 		return -ENOMEM;
741 
742 	ret = mt76_create_page_pool(dev, q);
743 	if (ret)
744 		return ret;
745 
746 	ret = mt76_wed_dma_setup(dev, q, false);
747 	if (ret)
748 		return ret;
749 
750 	if (mtk_wed_device_active(&dev->mmio.wed)) {
751 		if ((mtk_wed_get_rx_capa(&dev->mmio.wed) && mt76_queue_is_wed_rro(q)) ||
752 		    mt76_queue_is_wed_tx_free(q))
753 			return 0;
754 	}
755 
756 	mt76_dma_queue_reset(dev, q);
757 
758 	return 0;
759 }
760 
761 static void
mt76_dma_rx_cleanup(struct mt76_dev * dev,struct mt76_queue * q)762 mt76_dma_rx_cleanup(struct mt76_dev *dev, struct mt76_queue *q)
763 {
764 	void *buf;
765 	bool more;
766 
767 	if (!q->ndesc)
768 		return;
769 
770 	do {
771 		spin_lock_bh(&q->lock);
772 		buf = mt76_dma_dequeue(dev, q, true, NULL, NULL, &more, NULL);
773 		spin_unlock_bh(&q->lock);
774 
775 		if (!buf)
776 			break;
777 
778 		if (!mt76_queue_is_wed_rro(q))
779 			mt76_put_page_pool_buf(buf, false);
780 	} while (1);
781 
782 	spin_lock_bh(&q->lock);
783 	if (q->rx_head) {
784 		dev_kfree_skb(q->rx_head);
785 		q->rx_head = NULL;
786 	}
787 
788 	spin_unlock_bh(&q->lock);
789 }
790 
791 static void
mt76_dma_rx_reset(struct mt76_dev * dev,enum mt76_rxq_id qid)792 mt76_dma_rx_reset(struct mt76_dev *dev, enum mt76_rxq_id qid)
793 {
794 	struct mt76_queue *q = &dev->q_rx[qid];
795 
796 	if (!q->ndesc)
797 		return;
798 
799 	if (!mt76_queue_is_wed_rro_ind(q)) {
800 		int i;
801 
802 		for (i = 0; i < q->ndesc; i++)
803 			q->desc[i].ctrl = cpu_to_le32(MT_DMA_CTL_DMA_DONE);
804 	}
805 
806 	mt76_dma_rx_cleanup(dev, q);
807 
808 	/* reset WED rx queues */
809 	mt76_wed_dma_setup(dev, q, true);
810 
811 	if (mt76_queue_is_wed_tx_free(q))
812 		return;
813 
814 	if (mtk_wed_device_active(&dev->mmio.wed) &&
815 	    mt76_queue_is_wed_rro(q))
816 		return;
817 
818 	mt76_dma_sync_idx(dev, q);
819 	mt76_dma_rx_fill_buf(dev, q, false);
820 }
821 
822 static void
mt76_add_fragment(struct mt76_dev * dev,struct mt76_queue * q,void * data,int len,bool more,u32 info,bool allow_direct)823 mt76_add_fragment(struct mt76_dev *dev, struct mt76_queue *q, void *data,
824 		  int len, bool more, u32 info, bool allow_direct)
825 {
826 	struct sk_buff *skb = q->rx_head;
827 	struct skb_shared_info *shinfo = skb_shinfo(skb);
828 	int nr_frags = shinfo->nr_frags;
829 
830 	if (nr_frags < ARRAY_SIZE(shinfo->frags)) {
831 		struct page *page = virt_to_head_page(data);
832 #if defined(__linux__)
833 		int offset = data - page_address(page) + q->buf_offset;
834 #elif defined(__FreeBSD__)
835 		int offset = (u8 *)data - (u8 *)page_address(page) + q->buf_offset;
836 #endif
837 
838 		skb_add_rx_frag(skb, nr_frags, page, offset, len, q->buf_size);
839 	} else {
840 		mt76_put_page_pool_buf(data, allow_direct);
841 	}
842 
843 	if (more)
844 		return;
845 
846 	q->rx_head = NULL;
847 	if (nr_frags < ARRAY_SIZE(shinfo->frags))
848 		dev->drv->rx_skb(dev, q - dev->q_rx, skb, &info);
849 	else
850 		dev_kfree_skb(skb);
851 }
852 
853 static int
mt76_dma_rx_process(struct mt76_dev * dev,struct mt76_queue * q,int budget)854 mt76_dma_rx_process(struct mt76_dev *dev, struct mt76_queue *q, int budget)
855 {
856 	int len, data_len, done = 0, dma_idx;
857 	struct sk_buff *skb;
858 	unsigned char *data;
859 	bool check_ddone = false;
860 	bool allow_direct = !mt76_queue_is_wed_rx(q);
861 	bool more;
862 
863 	if (IS_ENABLED(CONFIG_NET_MEDIATEK_SOC_WED) &&
864 	    mt76_queue_is_wed_tx_free(q)) {
865 		dma_idx = Q_READ(q, dma_idx);
866 		check_ddone = true;
867 	}
868 
869 	while (done < budget) {
870 		bool drop = false;
871 		u32 info;
872 
873 		if (check_ddone) {
874 			if (q->tail == dma_idx)
875 				dma_idx = Q_READ(q, dma_idx);
876 
877 			if (q->tail == dma_idx)
878 				break;
879 		}
880 
881 		data = mt76_dma_dequeue(dev, q, false, &len, &info, &more,
882 					&drop);
883 		if (!data)
884 			break;
885 
886 		if (drop)
887 			goto free_frag;
888 
889 		if (q->rx_head)
890 			data_len = q->buf_size;
891 		else
892 			data_len = SKB_WITH_OVERHEAD(q->buf_size);
893 
894 		if (data_len < len + q->buf_offset) {
895 			dev_kfree_skb(q->rx_head);
896 			q->rx_head = NULL;
897 			goto free_frag;
898 		}
899 
900 		if (q->rx_head) {
901 			mt76_add_fragment(dev, q, data, len, more, info,
902 					  allow_direct);
903 			continue;
904 		}
905 
906 		if (!more && dev->drv->rx_check &&
907 		    !(dev->drv->rx_check(dev, data, len)))
908 			goto free_frag;
909 
910 		skb = napi_build_skb(data, q->buf_size);
911 		if (!skb)
912 			goto free_frag;
913 
914 		skb_reserve(skb, q->buf_offset);
915 		skb_mark_for_recycle(skb);
916 
917 		*(u32 *)skb->cb = info;
918 
919 		__skb_put(skb, len);
920 		done++;
921 
922 		if (more) {
923 			q->rx_head = skb;
924 			continue;
925 		}
926 
927 		dev->drv->rx_skb(dev, q - dev->q_rx, skb, &info);
928 		continue;
929 
930 free_frag:
931 		mt76_put_page_pool_buf(data, allow_direct);
932 	}
933 
934 	mt76_dma_rx_fill(dev, q, true);
935 	return done;
936 }
937 
mt76_dma_rx_poll(struct napi_struct * napi,int budget)938 int mt76_dma_rx_poll(struct napi_struct *napi, int budget)
939 {
940 	struct mt76_dev *dev;
941 	int qid, done = 0, cur;
942 
943 	dev = mt76_priv(napi->dev);
944 	qid = napi - dev->napi;
945 
946 	rcu_read_lock();
947 
948 	do {
949 		cur = mt76_dma_rx_process(dev, &dev->q_rx[qid], budget - done);
950 		mt76_rx_poll_complete(dev, qid, napi);
951 		done += cur;
952 	} while (cur && done < budget);
953 
954 	rcu_read_unlock();
955 
956 	if (done < budget && napi_complete(napi))
957 		dev->drv->rx_poll_complete(dev, qid);
958 
959 	return done;
960 }
961 EXPORT_SYMBOL_GPL(mt76_dma_rx_poll);
962 
963 static int
mt76_dma_init(struct mt76_dev * dev,int (* poll)(struct napi_struct * napi,int budget))964 mt76_dma_init(struct mt76_dev *dev,
965 	      int (*poll)(struct napi_struct *napi, int budget))
966 {
967 	struct mt76_dev **priv;
968 	int i;
969 
970 	dev->napi_dev = alloc_netdev_dummy(sizeof(struct mt76_dev *));
971 	if (!dev->napi_dev)
972 		return -ENOMEM;
973 
974 	/* napi_dev private data points to mt76_dev parent, so, mt76_dev
975 	 * can be retrieved given napi_dev
976 	 */
977 	priv = netdev_priv(dev->napi_dev);
978 	*priv = dev;
979 
980 	dev->tx_napi_dev = alloc_netdev_dummy(sizeof(struct mt76_dev *));
981 	if (!dev->tx_napi_dev) {
982 		free_netdev(dev->napi_dev);
983 		return -ENOMEM;
984 	}
985 	priv = netdev_priv(dev->tx_napi_dev);
986 	*priv = dev;
987 
988 	snprintf(dev->napi_dev->name, sizeof(dev->napi_dev->name), "%s",
989 		 wiphy_name(dev->hw->wiphy));
990 	dev->napi_dev->threaded = 1;
991 	init_completion(&dev->mmio.wed_reset);
992 	init_completion(&dev->mmio.wed_reset_complete);
993 
994 	mt76_for_each_q_rx(dev, i) {
995 		netif_napi_add(dev->napi_dev, &dev->napi[i], poll);
996 		mt76_dma_rx_fill_buf(dev, &dev->q_rx[i], false);
997 		napi_enable(&dev->napi[i]);
998 	}
999 
1000 	return 0;
1001 }
1002 
1003 static const struct mt76_queue_ops mt76_dma_ops = {
1004 	.init = mt76_dma_init,
1005 	.alloc = mt76_dma_alloc_queue,
1006 	.reset_q = mt76_dma_queue_reset,
1007 	.tx_queue_skb_raw = mt76_dma_tx_queue_skb_raw,
1008 	.tx_queue_skb = mt76_dma_tx_queue_skb,
1009 	.tx_cleanup = mt76_dma_tx_cleanup,
1010 	.rx_cleanup = mt76_dma_rx_cleanup,
1011 	.rx_reset = mt76_dma_rx_reset,
1012 	.kick = mt76_dma_kick_queue,
1013 };
1014 
mt76_dma_attach(struct mt76_dev * dev)1015 void mt76_dma_attach(struct mt76_dev *dev)
1016 {
1017 	dev->queue_ops = &mt76_dma_ops;
1018 }
1019 EXPORT_SYMBOL_GPL(mt76_dma_attach);
1020 
mt76_dma_cleanup(struct mt76_dev * dev)1021 void mt76_dma_cleanup(struct mt76_dev *dev)
1022 {
1023 	int i;
1024 
1025 	mt76_worker_disable(&dev->tx_worker);
1026 	netif_napi_del(&dev->tx_napi);
1027 
1028 	for (i = 0; i < ARRAY_SIZE(dev->phys); i++) {
1029 		struct mt76_phy *phy = dev->phys[i];
1030 		int j;
1031 
1032 		if (!phy)
1033 			continue;
1034 
1035 		for (j = 0; j < ARRAY_SIZE(phy->q_tx); j++)
1036 			mt76_dma_tx_cleanup(dev, phy->q_tx[j], true);
1037 	}
1038 
1039 	for (i = 0; i < ARRAY_SIZE(dev->q_mcu); i++)
1040 		mt76_dma_tx_cleanup(dev, dev->q_mcu[i], true);
1041 
1042 	mt76_for_each_q_rx(dev, i) {
1043 		struct mt76_queue *q = &dev->q_rx[i];
1044 
1045 		if (mtk_wed_device_active(&dev->mmio.wed) &&
1046 		    mt76_queue_is_wed_rro(q))
1047 			continue;
1048 
1049 		netif_napi_del(&dev->napi[i]);
1050 		mt76_dma_rx_cleanup(dev, q);
1051 
1052 		page_pool_destroy(q->page_pool);
1053 	}
1054 
1055 	if (mtk_wed_device_active(&dev->mmio.wed))
1056 		mtk_wed_device_detach(&dev->mmio.wed);
1057 
1058 	if (mtk_wed_device_active(&dev->mmio.wed_hif2))
1059 		mtk_wed_device_detach(&dev->mmio.wed_hif2);
1060 
1061 	mt76_free_pending_txwi(dev);
1062 	mt76_free_pending_rxwi(dev);
1063 	free_netdev(dev->napi_dev);
1064 	free_netdev(dev->tx_napi_dev);
1065 }
1066 EXPORT_SYMBOL_GPL(mt76_dma_cleanup);
1067