1 /*
2  * Copyright (c) 2012-2017 Qualcomm Atheros, Inc.
3  * Copyright (c) 2018-2019, The Linux Foundation. All rights reserved.
4  *
5  * Permission to use, copy, modify, and/or distribute this software for any
6  * purpose with or without fee is hereby granted, provided that the above
7  * copyright notice and this permission notice appear in all copies.
8  *
9  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
10  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
11  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
12  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
13  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
14  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
15  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
16  */
17 
18 #include <linux/etherdevice.h>
19 #include <net/ieee80211_radiotap.h>
20 #include <linux/if_arp.h>
21 #include <linux/moduleparam.h>
22 #include <linux/ip.h>
23 #include <linux/ipv6.h>
24 #include <net/ipv6.h>
25 #include <linux/prefetch.h>
26 
27 #include "wil6210.h"
28 #include "wmi.h"
29 #include "txrx.h"
30 #include "trace.h"
31 #include "txrx_edma.h"
32 
33 bool rx_align_2;
34 module_param(rx_align_2, bool, 0444);
35 MODULE_PARM_DESC(rx_align_2, " align Rx buffers on 4*n+2, default - no");
36 
37 bool rx_large_buf;
38 module_param(rx_large_buf, bool, 0444);
39 MODULE_PARM_DESC(rx_large_buf, " allocate 8KB RX buffers, default - no");
40 
41 /* Drop Tx packets in case Tx ring is full */
42 bool drop_if_ring_full;
43 
44 static inline uint wil_rx_snaplen(void)
45 {
46 	return rx_align_2 ? 6 : 0;
47 }
48 
49 /* wil_ring_wmark_low - low watermark for available descriptor space */
50 static inline int wil_ring_wmark_low(struct wil_ring *ring)
51 {
52 	return ring->size / 8;
53 }
54 
55 /* wil_ring_wmark_high - high watermark for available descriptor space */
56 static inline int wil_ring_wmark_high(struct wil_ring *ring)
57 {
58 	return ring->size / 4;
59 }
60 
61 /* returns true if num avail descriptors is lower than wmark_low */
62 static inline int wil_ring_avail_low(struct wil_ring *ring)
63 {
64 	return wil_ring_avail_tx(ring) < wil_ring_wmark_low(ring);
65 }
66 
67 /* returns true if num avail descriptors is higher than wmark_high */
68 static inline int wil_ring_avail_high(struct wil_ring *ring)
69 {
70 	return wil_ring_avail_tx(ring) > wil_ring_wmark_high(ring);
71 }
72 
73 /* returns true when all tx vrings are empty */
74 bool wil_is_tx_idle(struct wil6210_priv *wil)
75 {
76 	int i;
77 	unsigned long data_comp_to;
78 	int min_ring_id = wil_get_min_tx_ring_id(wil);
79 
80 	for (i = min_ring_id; i < WIL6210_MAX_TX_RINGS; i++) {
81 		struct wil_ring *vring = &wil->ring_tx[i];
82 		int vring_index = vring - wil->ring_tx;
83 		struct wil_ring_tx_data *txdata =
84 			&wil->ring_tx_data[vring_index];
85 
86 		spin_lock(&txdata->lock);
87 
88 		if (!vring->va || !txdata->enabled) {
89 			spin_unlock(&txdata->lock);
90 			continue;
91 		}
92 
93 		data_comp_to = jiffies + msecs_to_jiffies(
94 					WIL_DATA_COMPLETION_TO_MS);
95 		if (test_bit(wil_status_napi_en, wil->status)) {
96 			while (!wil_ring_is_empty(vring)) {
97 				if (time_after(jiffies, data_comp_to)) {
98 					wil_dbg_pm(wil,
99 						   "TO waiting for idle tx\n");
100 					spin_unlock(&txdata->lock);
101 					return false;
102 				}
103 				wil_dbg_ratelimited(wil,
104 						    "tx vring is not empty -> NAPI\n");
105 				spin_unlock(&txdata->lock);
106 				napi_synchronize(&wil->napi_tx);
107 				msleep(20);
108 				spin_lock(&txdata->lock);
109 				if (!vring->va || !txdata->enabled)
110 					break;
111 			}
112 		}
113 
114 		spin_unlock(&txdata->lock);
115 	}
116 
117 	return true;
118 }
119 
120 static int wil_vring_alloc(struct wil6210_priv *wil, struct wil_ring *vring)
121 {
122 	struct device *dev = wil_to_dev(wil);
123 	size_t sz = vring->size * sizeof(vring->va[0]);
124 	uint i;
125 
126 	wil_dbg_misc(wil, "vring_alloc:\n");
127 
128 	BUILD_BUG_ON(sizeof(vring->va[0]) != 32);
129 
130 	vring->swhead = 0;
131 	vring->swtail = 0;
132 	vring->ctx = kcalloc(vring->size, sizeof(vring->ctx[0]), GFP_KERNEL);
133 	if (!vring->ctx) {
134 		vring->va = NULL;
135 		return -ENOMEM;
136 	}
137 
138 	/* vring->va should be aligned on its size rounded up to power of 2
139 	 * This is granted by the dma_alloc_coherent.
140 	 *
141 	 * HW has limitation that all vrings addresses must share the same
142 	 * upper 16 msb bits part of 48 bits address. To workaround that,
143 	 * if we are using more than 32 bit addresses switch to 32 bit
144 	 * allocation before allocating vring memory.
145 	 *
146 	 * There's no check for the return value of dma_set_mask_and_coherent,
147 	 * since we assume if we were able to set the mask during
148 	 * initialization in this system it will not fail if we set it again
149 	 */
150 	if (wil->dma_addr_size > 32)
151 		dma_set_mask_and_coherent(dev, DMA_BIT_MASK(32));
152 
153 	vring->va = dma_alloc_coherent(dev, sz, &vring->pa, GFP_KERNEL);
154 	if (!vring->va) {
155 		kfree(vring->ctx);
156 		vring->ctx = NULL;
157 		return -ENOMEM;
158 	}
159 
160 	if (wil->dma_addr_size > 32)
161 		dma_set_mask_and_coherent(dev,
162 					  DMA_BIT_MASK(wil->dma_addr_size));
163 
164 	/* initially, all descriptors are SW owned
165 	 * For Tx and Rx, ownership bit is at the same location, thus
166 	 * we can use any
167 	 */
168 	for (i = 0; i < vring->size; i++) {
169 		volatile struct vring_tx_desc *_d =
170 			&vring->va[i].tx.legacy;
171 
172 		_d->dma.status = TX_DMA_STATUS_DU;
173 	}
174 
175 	wil_dbg_misc(wil, "vring[%d] 0x%p:%pad 0x%p\n", vring->size,
176 		     vring->va, &vring->pa, vring->ctx);
177 
178 	return 0;
179 }
180 
181 static void wil_txdesc_unmap(struct device *dev, union wil_tx_desc *desc,
182 			     struct wil_ctx *ctx)
183 {
184 	struct vring_tx_desc *d = &desc->legacy;
185 	dma_addr_t pa = wil_desc_addr(&d->dma.addr);
186 	u16 dmalen = le16_to_cpu(d->dma.length);
187 
188 	switch (ctx->mapped_as) {
189 	case wil_mapped_as_single:
190 		dma_unmap_single(dev, pa, dmalen, DMA_TO_DEVICE);
191 		break;
192 	case wil_mapped_as_page:
193 		dma_unmap_page(dev, pa, dmalen, DMA_TO_DEVICE);
194 		break;
195 	default:
196 		break;
197 	}
198 }
199 
200 static void wil_vring_free(struct wil6210_priv *wil, struct wil_ring *vring)
201 {
202 	struct device *dev = wil_to_dev(wil);
203 	size_t sz = vring->size * sizeof(vring->va[0]);
204 
205 	lockdep_assert_held(&wil->mutex);
206 	if (!vring->is_rx) {
207 		int vring_index = vring - wil->ring_tx;
208 
209 		wil_dbg_misc(wil, "free Tx vring %d [%d] 0x%p:%pad 0x%p\n",
210 			     vring_index, vring->size, vring->va,
211 			     &vring->pa, vring->ctx);
212 	} else {
213 		wil_dbg_misc(wil, "free Rx vring [%d] 0x%p:%pad 0x%p\n",
214 			     vring->size, vring->va,
215 			     &vring->pa, vring->ctx);
216 	}
217 
218 	while (!wil_ring_is_empty(vring)) {
219 		dma_addr_t pa;
220 		u16 dmalen;
221 		struct wil_ctx *ctx;
222 
223 		if (!vring->is_rx) {
224 			struct vring_tx_desc dd, *d = &dd;
225 			volatile struct vring_tx_desc *_d =
226 					&vring->va[vring->swtail].tx.legacy;
227 
228 			ctx = &vring->ctx[vring->swtail];
229 			if (!ctx) {
230 				wil_dbg_txrx(wil,
231 					     "ctx(%d) was already completed\n",
232 					     vring->swtail);
233 				vring->swtail = wil_ring_next_tail(vring);
234 				continue;
235 			}
236 			*d = *_d;
237 			wil_txdesc_unmap(dev, (union wil_tx_desc *)d, ctx);
238 			if (ctx->skb)
239 				dev_kfree_skb_any(ctx->skb);
240 			vring->swtail = wil_ring_next_tail(vring);
241 		} else { /* rx */
242 			struct vring_rx_desc dd, *d = &dd;
243 			volatile struct vring_rx_desc *_d =
244 				&vring->va[vring->swhead].rx.legacy;
245 
246 			ctx = &vring->ctx[vring->swhead];
247 			*d = *_d;
248 			pa = wil_desc_addr(&d->dma.addr);
249 			dmalen = le16_to_cpu(d->dma.length);
250 			dma_unmap_single(dev, pa, dmalen, DMA_FROM_DEVICE);
251 			kfree_skb(ctx->skb);
252 			wil_ring_advance_head(vring, 1);
253 		}
254 	}
255 	dma_free_coherent(dev, sz, (void *)vring->va, vring->pa);
256 	kfree(vring->ctx);
257 	vring->pa = 0;
258 	vring->va = NULL;
259 	vring->ctx = NULL;
260 }
261 
262 /**
263  * Allocate one skb for Rx VRING
264  *
265  * Safe to call from IRQ
266  */
267 static int wil_vring_alloc_skb(struct wil6210_priv *wil, struct wil_ring *vring,
268 			       u32 i, int headroom)
269 {
270 	struct device *dev = wil_to_dev(wil);
271 	unsigned int sz = wil->rx_buf_len + ETH_HLEN + wil_rx_snaplen();
272 	struct vring_rx_desc dd, *d = &dd;
273 	volatile struct vring_rx_desc *_d = &vring->va[i].rx.legacy;
274 	dma_addr_t pa;
275 	struct sk_buff *skb = dev_alloc_skb(sz + headroom);
276 
277 	if (unlikely(!skb))
278 		return -ENOMEM;
279 
280 	skb_reserve(skb, headroom);
281 	skb_put(skb, sz);
282 
283 	/**
284 	 * Make sure that the network stack calculates checksum for packets
285 	 * which failed the HW checksum calculation
286 	 */
287 	skb->ip_summed = CHECKSUM_NONE;
288 
289 	pa = dma_map_single(dev, skb->data, skb->len, DMA_FROM_DEVICE);
290 	if (unlikely(dma_mapping_error(dev, pa))) {
291 		kfree_skb(skb);
292 		return -ENOMEM;
293 	}
294 
295 	d->dma.d0 = RX_DMA_D0_CMD_DMA_RT | RX_DMA_D0_CMD_DMA_IT;
296 	wil_desc_addr_set(&d->dma.addr, pa);
297 	/* ip_length don't care */
298 	/* b11 don't care */
299 	/* error don't care */
300 	d->dma.status = 0; /* BIT(0) should be 0 for HW_OWNED */
301 	d->dma.length = cpu_to_le16(sz);
302 	*_d = *d;
303 	vring->ctx[i].skb = skb;
304 
305 	return 0;
306 }
307 
308 /**
309  * Adds radiotap header
310  *
311  * Any error indicated as "Bad FCS"
312  *
313  * Vendor data for 04:ce:14-1 (Wilocity-1) consists of:
314  *  - Rx descriptor: 32 bytes
315  *  - Phy info
316  */
317 static void wil_rx_add_radiotap_header(struct wil6210_priv *wil,
318 				       struct sk_buff *skb)
319 {
320 	struct wil6210_rtap {
321 		struct ieee80211_radiotap_header rthdr;
322 		/* fields should be in the order of bits in rthdr.it_present */
323 		/* flags */
324 		u8 flags;
325 		/* channel */
326 		__le16 chnl_freq __aligned(2);
327 		__le16 chnl_flags;
328 		/* MCS */
329 		u8 mcs_present;
330 		u8 mcs_flags;
331 		u8 mcs_index;
332 	} __packed;
333 	struct vring_rx_desc *d = wil_skb_rxdesc(skb);
334 	struct wil6210_rtap *rtap;
335 	int rtap_len = sizeof(struct wil6210_rtap);
336 	struct ieee80211_channel *ch = wil->monitor_chandef.chan;
337 
338 	if (skb_headroom(skb) < rtap_len &&
339 	    pskb_expand_head(skb, rtap_len, 0, GFP_ATOMIC)) {
340 		wil_err(wil, "Unable to expand headroom to %d\n", rtap_len);
341 		return;
342 	}
343 
344 	rtap = skb_push(skb, rtap_len);
345 	memset(rtap, 0, rtap_len);
346 
347 	rtap->rthdr.it_version = PKTHDR_RADIOTAP_VERSION;
348 	rtap->rthdr.it_len = cpu_to_le16(rtap_len);
349 	rtap->rthdr.it_present = cpu_to_le32((1 << IEEE80211_RADIOTAP_FLAGS) |
350 			(1 << IEEE80211_RADIOTAP_CHANNEL) |
351 			(1 << IEEE80211_RADIOTAP_MCS));
352 	if (d->dma.status & RX_DMA_STATUS_ERROR)
353 		rtap->flags |= IEEE80211_RADIOTAP_F_BADFCS;
354 
355 	rtap->chnl_freq = cpu_to_le16(ch ? ch->center_freq : 58320);
356 	rtap->chnl_flags = cpu_to_le16(0);
357 
358 	rtap->mcs_present = IEEE80211_RADIOTAP_MCS_HAVE_MCS;
359 	rtap->mcs_flags = 0;
360 	rtap->mcs_index = wil_rxdesc_mcs(d);
361 }
362 
363 static bool wil_is_rx_idle(struct wil6210_priv *wil)
364 {
365 	struct vring_rx_desc *_d;
366 	struct wil_ring *ring = &wil->ring_rx;
367 
368 	_d = (struct vring_rx_desc *)&ring->va[ring->swhead].rx.legacy;
369 	if (_d->dma.status & RX_DMA_STATUS_DU)
370 		return false;
371 
372 	return true;
373 }
374 
375 static int wil_rx_get_cid_by_skb(struct wil6210_priv *wil, struct sk_buff *skb)
376 {
377 	struct vring_rx_desc *d = wil_skb_rxdesc(skb);
378 	int mid = wil_rxdesc_mid(d);
379 	struct wil6210_vif *vif = wil->vifs[mid];
380 	/* cid from DMA descriptor is limited to 3 bits.
381 	 * In case of cid>=8, the value would be cid modulo 8 and we need to
382 	 * find real cid by locating the transmitter (ta) inside sta array
383 	 */
384 	int cid = wil_rxdesc_cid(d);
385 	unsigned int snaplen = wil_rx_snaplen();
386 	struct ieee80211_hdr_3addr *hdr;
387 	int i;
388 	unsigned char *ta;
389 	u8 ftype;
390 
391 	/* in monitor mode there are no connections */
392 	if (vif->wdev.iftype == NL80211_IFTYPE_MONITOR)
393 		return cid;
394 
395 	ftype = wil_rxdesc_ftype(d) << 2;
396 	if (likely(ftype == IEEE80211_FTYPE_DATA)) {
397 		if (unlikely(skb->len < ETH_HLEN + snaplen)) {
398 			wil_err_ratelimited(wil,
399 					    "Short data frame, len = %d\n",
400 					    skb->len);
401 			return -ENOENT;
402 		}
403 		ta = wil_skb_get_sa(skb);
404 	} else {
405 		if (unlikely(skb->len < sizeof(struct ieee80211_hdr_3addr))) {
406 			wil_err_ratelimited(wil, "Short frame, len = %d\n",
407 					    skb->len);
408 			return -ENOENT;
409 		}
410 		hdr = (void *)skb->data;
411 		ta = hdr->addr2;
412 	}
413 
414 	if (wil->max_assoc_sta <= WIL6210_RX_DESC_MAX_CID)
415 		return cid;
416 
417 	/* assuming no concurrency between AP interfaces and STA interfaces.
418 	 * multista is used only in P2P_GO or AP mode. In other modes return
419 	 * cid from the rx descriptor
420 	 */
421 	if (vif->wdev.iftype != NL80211_IFTYPE_P2P_GO &&
422 	    vif->wdev.iftype != NL80211_IFTYPE_AP)
423 		return cid;
424 
425 	/* For Rx packets cid from rx descriptor is limited to 3 bits (0..7),
426 	 * to find the real cid, compare transmitter address with the stored
427 	 * stations mac address in the driver sta array
428 	 */
429 	for (i = cid; i < wil->max_assoc_sta; i += WIL6210_RX_DESC_MAX_CID) {
430 		if (wil->sta[i].status != wil_sta_unused &&
431 		    ether_addr_equal(wil->sta[i].addr, ta)) {
432 			cid = i;
433 			break;
434 		}
435 	}
436 	if (i >= wil->max_assoc_sta) {
437 		wil_err_ratelimited(wil, "Could not find cid for frame with transmit addr = %pM, iftype = %d, frametype = %d, len = %d\n",
438 				    ta, vif->wdev.iftype, ftype, skb->len);
439 		cid = -ENOENT;
440 	}
441 
442 	return cid;
443 }
444 
445 /**
446  * reap 1 frame from @swhead
447  *
448  * Rx descriptor copied to skb->cb
449  *
450  * Safe to call from IRQ
451  */
452 static struct sk_buff *wil_vring_reap_rx(struct wil6210_priv *wil,
453 					 struct wil_ring *vring)
454 {
455 	struct device *dev = wil_to_dev(wil);
456 	struct wil6210_vif *vif;
457 	struct net_device *ndev;
458 	volatile struct vring_rx_desc *_d;
459 	struct vring_rx_desc *d;
460 	struct sk_buff *skb;
461 	dma_addr_t pa;
462 	unsigned int snaplen = wil_rx_snaplen();
463 	unsigned int sz = wil->rx_buf_len + ETH_HLEN + snaplen;
464 	u16 dmalen;
465 	u8 ftype;
466 	int cid, mid;
467 	int i;
468 	struct wil_net_stats *stats;
469 
470 	BUILD_BUG_ON(sizeof(struct skb_rx_info) > sizeof(skb->cb));
471 
472 again:
473 	if (unlikely(wil_ring_is_empty(vring)))
474 		return NULL;
475 
476 	i = (int)vring->swhead;
477 	_d = &vring->va[i].rx.legacy;
478 	if (unlikely(!(_d->dma.status & RX_DMA_STATUS_DU))) {
479 		/* it is not error, we just reached end of Rx done area */
480 		return NULL;
481 	}
482 
483 	skb = vring->ctx[i].skb;
484 	vring->ctx[i].skb = NULL;
485 	wil_ring_advance_head(vring, 1);
486 	if (!skb) {
487 		wil_err(wil, "No Rx skb at [%d]\n", i);
488 		goto again;
489 	}
490 	d = wil_skb_rxdesc(skb);
491 	*d = *_d;
492 	pa = wil_desc_addr(&d->dma.addr);
493 
494 	dma_unmap_single(dev, pa, sz, DMA_FROM_DEVICE);
495 	dmalen = le16_to_cpu(d->dma.length);
496 
497 	trace_wil6210_rx(i, d);
498 	wil_dbg_txrx(wil, "Rx[%3d] : %d bytes\n", i, dmalen);
499 	wil_hex_dump_txrx("RxD ", DUMP_PREFIX_NONE, 32, 4,
500 			  (const void *)d, sizeof(*d), false);
501 
502 	mid = wil_rxdesc_mid(d);
503 	vif = wil->vifs[mid];
504 
505 	if (unlikely(!vif)) {
506 		wil_dbg_txrx(wil, "skipped RX descriptor with invalid mid %d",
507 			     mid);
508 		kfree_skb(skb);
509 		goto again;
510 	}
511 	ndev = vif_to_ndev(vif);
512 	if (unlikely(dmalen > sz)) {
513 		wil_err_ratelimited(wil, "Rx size too large: %d bytes!\n",
514 				    dmalen);
515 		kfree_skb(skb);
516 		goto again;
517 	}
518 	skb_trim(skb, dmalen);
519 
520 	prefetch(skb->data);
521 
522 	wil_hex_dump_txrx("Rx ", DUMP_PREFIX_OFFSET, 16, 1,
523 			  skb->data, skb_headlen(skb), false);
524 
525 	cid = wil_rx_get_cid_by_skb(wil, skb);
526 	if (cid == -ENOENT) {
527 		kfree_skb(skb);
528 		goto again;
529 	}
530 	wil_skb_set_cid(skb, (u8)cid);
531 	stats = &wil->sta[cid].stats;
532 
533 	stats->last_mcs_rx = wil_rxdesc_mcs(d);
534 	if (stats->last_mcs_rx < ARRAY_SIZE(stats->rx_per_mcs))
535 		stats->rx_per_mcs[stats->last_mcs_rx]++;
536 
537 	/* use radiotap header only if required */
538 	if (ndev->type == ARPHRD_IEEE80211_RADIOTAP)
539 		wil_rx_add_radiotap_header(wil, skb);
540 
541 	/* no extra checks if in sniffer mode */
542 	if (ndev->type != ARPHRD_ETHER)
543 		return skb;
544 	/* Non-data frames may be delivered through Rx DMA channel (ex: BAR)
545 	 * Driver should recognize it by frame type, that is found
546 	 * in Rx descriptor. If type is not data, it is 802.11 frame as is
547 	 */
548 	ftype = wil_rxdesc_ftype(d) << 2;
549 	if (unlikely(ftype != IEEE80211_FTYPE_DATA)) {
550 		u8 fc1 = wil_rxdesc_fc1(d);
551 		int tid = wil_rxdesc_tid(d);
552 		u16 seq = wil_rxdesc_seq(d);
553 
554 		wil_dbg_txrx(wil,
555 			     "Non-data frame FC[7:0] 0x%02x MID %d CID %d TID %d Seq 0x%03x\n",
556 			     fc1, mid, cid, tid, seq);
557 		stats->rx_non_data_frame++;
558 		if (wil_is_back_req(fc1)) {
559 			wil_dbg_txrx(wil,
560 				     "BAR: MID %d CID %d TID %d Seq 0x%03x\n",
561 				     mid, cid, tid, seq);
562 			wil_rx_bar(wil, vif, cid, tid, seq);
563 		} else {
564 			/* print again all info. One can enable only this
565 			 * without overhead for printing every Rx frame
566 			 */
567 			wil_dbg_txrx(wil,
568 				     "Unhandled non-data frame FC[7:0] 0x%02x MID %d CID %d TID %d Seq 0x%03x\n",
569 				     fc1, mid, cid, tid, seq);
570 			wil_hex_dump_txrx("RxD ", DUMP_PREFIX_NONE, 32, 4,
571 					  (const void *)d, sizeof(*d), false);
572 			wil_hex_dump_txrx("Rx ", DUMP_PREFIX_OFFSET, 16, 1,
573 					  skb->data, skb_headlen(skb), false);
574 		}
575 		kfree_skb(skb);
576 		goto again;
577 	}
578 
579 	/* L4 IDENT is on when HW calculated checksum, check status
580 	 * and in case of error drop the packet
581 	 * higher stack layers will handle retransmission (if required)
582 	 */
583 	if (likely(d->dma.status & RX_DMA_STATUS_L4I)) {
584 		/* L4 protocol identified, csum calculated */
585 		if (likely((d->dma.error & RX_DMA_ERROR_L4_ERR) == 0))
586 			skb->ip_summed = CHECKSUM_UNNECESSARY;
587 		/* If HW reports bad checksum, let IP stack re-check it
588 		 * For example, HW don't understand Microsoft IP stack that
589 		 * mis-calculates TCP checksum - if it should be 0x0,
590 		 * it writes 0xffff in violation of RFC 1624
591 		 */
592 		else
593 			stats->rx_csum_err++;
594 	}
595 
596 	if (snaplen) {
597 		/* Packet layout
598 		 * +-------+-------+---------+------------+------+
599 		 * | SA(6) | DA(6) | SNAP(6) | ETHTYPE(2) | DATA |
600 		 * +-------+-------+---------+------------+------+
601 		 * Need to remove SNAP, shifting SA and DA forward
602 		 */
603 		memmove(skb->data + snaplen, skb->data, 2 * ETH_ALEN);
604 		skb_pull(skb, snaplen);
605 	}
606 
607 	return skb;
608 }
609 
610 /**
611  * allocate and fill up to @count buffers in rx ring
612  * buffers posted at @swtail
613  * Note: we have a single RX queue for servicing all VIFs, but we
614  * allocate skbs with headroom according to main interface only. This
615  * means it will not work with monitor interface together with other VIFs.
616  * Currently we only support monitor interface on its own without other VIFs,
617  * and we will need to fix this code once we add support.
618  */
619 static int wil_rx_refill(struct wil6210_priv *wil, int count)
620 {
621 	struct net_device *ndev = wil->main_ndev;
622 	struct wil_ring *v = &wil->ring_rx;
623 	u32 next_tail;
624 	int rc = 0;
625 	int headroom = ndev->type == ARPHRD_IEEE80211_RADIOTAP ?
626 			WIL6210_RTAP_SIZE : 0;
627 
628 	for (; next_tail = wil_ring_next_tail(v),
629 	     (next_tail != v->swhead) && (count-- > 0);
630 	     v->swtail = next_tail) {
631 		rc = wil_vring_alloc_skb(wil, v, v->swtail, headroom);
632 		if (unlikely(rc)) {
633 			wil_err_ratelimited(wil, "Error %d in rx refill[%d]\n",
634 					    rc, v->swtail);
635 			break;
636 		}
637 	}
638 
639 	/* make sure all writes to descriptors (shared memory) are done before
640 	 * committing them to HW
641 	 */
642 	wmb();
643 
644 	wil_w(wil, v->hwtail, v->swtail);
645 
646 	return rc;
647 }
648 
649 /**
650  * reverse_memcmp - Compare two areas of memory, in reverse order
651  * @cs: One area of memory
652  * @ct: Another area of memory
653  * @count: The size of the area.
654  *
655  * Cut'n'paste from original memcmp (see lib/string.c)
656  * with minimal modifications
657  */
658 int reverse_memcmp(const void *cs, const void *ct, size_t count)
659 {
660 	const unsigned char *su1, *su2;
661 	int res = 0;
662 
663 	for (su1 = cs + count - 1, su2 = ct + count - 1; count > 0;
664 	     --su1, --su2, count--) {
665 		res = *su1 - *su2;
666 		if (res)
667 			break;
668 	}
669 	return res;
670 }
671 
672 static int wil_rx_crypto_check(struct wil6210_priv *wil, struct sk_buff *skb)
673 {
674 	struct vring_rx_desc *d = wil_skb_rxdesc(skb);
675 	int cid = wil_skb_get_cid(skb);
676 	int tid = wil_rxdesc_tid(d);
677 	int key_id = wil_rxdesc_key_id(d);
678 	int mc = wil_rxdesc_mcast(d);
679 	struct wil_sta_info *s = &wil->sta[cid];
680 	struct wil_tid_crypto_rx *c = mc ? &s->group_crypto_rx :
681 				      &s->tid_crypto_rx[tid];
682 	struct wil_tid_crypto_rx_single *cc = &c->key_id[key_id];
683 	const u8 *pn = (u8 *)&d->mac.pn_15_0;
684 
685 	if (!cc->key_set) {
686 		wil_err_ratelimited(wil,
687 				    "Key missing. CID %d TID %d MCast %d KEY_ID %d\n",
688 				    cid, tid, mc, key_id);
689 		return -EINVAL;
690 	}
691 
692 	if (reverse_memcmp(pn, cc->pn, IEEE80211_GCMP_PN_LEN) <= 0) {
693 		wil_err_ratelimited(wil,
694 				    "Replay attack. CID %d TID %d MCast %d KEY_ID %d PN %6phN last %6phN\n",
695 				    cid, tid, mc, key_id, pn, cc->pn);
696 		return -EINVAL;
697 	}
698 	memcpy(cc->pn, pn, IEEE80211_GCMP_PN_LEN);
699 
700 	return 0;
701 }
702 
703 static int wil_rx_error_check(struct wil6210_priv *wil, struct sk_buff *skb,
704 			      struct wil_net_stats *stats)
705 {
706 	struct vring_rx_desc *d = wil_skb_rxdesc(skb);
707 
708 	if ((d->dma.status & RX_DMA_STATUS_ERROR) &&
709 	    (d->dma.error & RX_DMA_ERROR_MIC)) {
710 		stats->rx_mic_error++;
711 		wil_dbg_txrx(wil, "MIC error, dropping packet\n");
712 		return -EFAULT;
713 	}
714 
715 	return 0;
716 }
717 
718 static void wil_get_netif_rx_params(struct sk_buff *skb, int *cid,
719 				    int *security)
720 {
721 	struct vring_rx_desc *d = wil_skb_rxdesc(skb);
722 
723 	*cid = wil_skb_get_cid(skb);
724 	*security = wil_rxdesc_security(d);
725 }
726 
727 /*
728  * Check if skb is ptk eapol key message
729  *
730  * returns a pointer to the start of the eapol key structure, NULL
731  * if frame is not PTK eapol key
732  */
733 static struct wil_eapol_key *wil_is_ptk_eapol_key(struct wil6210_priv *wil,
734 						  struct sk_buff *skb)
735 {
736 	u8 *buf;
737 	const struct wil_1x_hdr *hdr;
738 	struct wil_eapol_key *key;
739 	u16 key_info;
740 	int len = skb->len;
741 
742 	if (!skb_mac_header_was_set(skb)) {
743 		wil_err(wil, "mac header was not set\n");
744 		return NULL;
745 	}
746 
747 	len -= skb_mac_offset(skb);
748 
749 	if (len < sizeof(struct ethhdr) + sizeof(struct wil_1x_hdr) +
750 	    sizeof(struct wil_eapol_key))
751 		return NULL;
752 
753 	buf = skb_mac_header(skb) + sizeof(struct ethhdr);
754 
755 	hdr = (const struct wil_1x_hdr *)buf;
756 	if (hdr->type != WIL_1X_TYPE_EAPOL_KEY)
757 		return NULL;
758 
759 	key = (struct wil_eapol_key *)(buf + sizeof(struct wil_1x_hdr));
760 	if (key->type != WIL_EAPOL_KEY_TYPE_WPA &&
761 	    key->type != WIL_EAPOL_KEY_TYPE_RSN)
762 		return NULL;
763 
764 	key_info = be16_to_cpu(key->key_info);
765 	if (!(key_info & WIL_KEY_INFO_KEY_TYPE)) /* check if pairwise */
766 		return NULL;
767 
768 	return key;
769 }
770 
771 static bool wil_skb_is_eap_3(struct wil6210_priv *wil, struct sk_buff *skb)
772 {
773 	struct wil_eapol_key *key;
774 	u16 key_info;
775 
776 	key = wil_is_ptk_eapol_key(wil, skb);
777 	if (!key)
778 		return false;
779 
780 	key_info = be16_to_cpu(key->key_info);
781 	if (key_info & (WIL_KEY_INFO_MIC |
782 			WIL_KEY_INFO_ENCR_KEY_DATA)) {
783 		/* 3/4 of 4-Way Handshake */
784 		wil_dbg_misc(wil, "EAPOL key message 3\n");
785 		return true;
786 	}
787 	/* 1/4 of 4-Way Handshake */
788 	wil_dbg_misc(wil, "EAPOL key message 1\n");
789 
790 	return false;
791 }
792 
793 static bool wil_skb_is_eap_4(struct wil6210_priv *wil, struct sk_buff *skb)
794 {
795 	struct wil_eapol_key *key;
796 	u32 *nonce, i;
797 
798 	key = wil_is_ptk_eapol_key(wil, skb);
799 	if (!key)
800 		return false;
801 
802 	nonce = (u32 *)key->key_nonce;
803 	for (i = 0; i < WIL_EAP_NONCE_LEN / sizeof(u32); i++, nonce++) {
804 		if (*nonce != 0) {
805 			/* message 2/4 */
806 			wil_dbg_misc(wil, "EAPOL key message 2\n");
807 			return false;
808 		}
809 	}
810 	wil_dbg_misc(wil, "EAPOL key message 4\n");
811 
812 	return true;
813 }
814 
815 void wil_enable_tx_key_worker(struct work_struct *work)
816 {
817 	struct wil6210_vif *vif = container_of(work,
818 			struct wil6210_vif, enable_tx_key_worker);
819 	struct wil6210_priv *wil = vif_to_wil(vif);
820 	int rc, cid;
821 
822 	rtnl_lock();
823 	if (vif->ptk_rekey_state != WIL_REKEY_WAIT_M4_SENT) {
824 		wil_dbg_misc(wil, "Invalid rekey state = %d\n",
825 			     vif->ptk_rekey_state);
826 		rtnl_unlock();
827 		return;
828 	}
829 
830 	cid =  wil_find_cid_by_idx(wil, vif->mid, 0);
831 	if (!wil_cid_valid(wil, cid)) {
832 		wil_err(wil, "Invalid cid = %d\n", cid);
833 		rtnl_unlock();
834 		return;
835 	}
836 
837 	wil_dbg_misc(wil, "Apply PTK key after eapol was sent out\n");
838 	rc = wmi_add_cipher_key(vif, 0, wil->sta[cid].addr, 0, NULL,
839 				WMI_KEY_USE_APPLY_PTK);
840 
841 	vif->ptk_rekey_state = WIL_REKEY_IDLE;
842 	rtnl_unlock();
843 
844 	if (rc)
845 		wil_err(wil, "Apply PTK key failed %d\n", rc);
846 }
847 
848 void wil_tx_complete_handle_eapol(struct wil6210_vif *vif, struct sk_buff *skb)
849 {
850 	struct wil6210_priv *wil = vif_to_wil(vif);
851 	struct wireless_dev *wdev = vif_to_wdev(vif);
852 	bool q = false;
853 
854 	if (wdev->iftype != NL80211_IFTYPE_STATION ||
855 	    !test_bit(WMI_FW_CAPABILITY_SPLIT_REKEY, wil->fw_capabilities))
856 		return;
857 
858 	/* check if skb is an EAP message 4/4 */
859 	if (!wil_skb_is_eap_4(wil, skb))
860 		return;
861 
862 	spin_lock_bh(&wil->eap_lock);
863 	switch (vif->ptk_rekey_state) {
864 	case WIL_REKEY_IDLE:
865 		/* ignore idle state, can happen due to M4 retransmission */
866 		break;
867 	case WIL_REKEY_M3_RECEIVED:
868 		vif->ptk_rekey_state = WIL_REKEY_IDLE;
869 		break;
870 	case WIL_REKEY_WAIT_M4_SENT:
871 		q = true;
872 		break;
873 	default:
874 		wil_err(wil, "Unknown rekey state = %d",
875 			vif->ptk_rekey_state);
876 	}
877 	spin_unlock_bh(&wil->eap_lock);
878 
879 	if (q) {
880 		q = queue_work(wil->wmi_wq, &vif->enable_tx_key_worker);
881 		wil_dbg_misc(wil, "queue_work of enable_tx_key_worker -> %d\n",
882 			     q);
883 	}
884 }
885 
886 static void wil_rx_handle_eapol(struct wil6210_vif *vif, struct sk_buff *skb)
887 {
888 	struct wil6210_priv *wil = vif_to_wil(vif);
889 	struct wireless_dev *wdev = vif_to_wdev(vif);
890 
891 	if (wdev->iftype != NL80211_IFTYPE_STATION ||
892 	    !test_bit(WMI_FW_CAPABILITY_SPLIT_REKEY, wil->fw_capabilities))
893 		return;
894 
895 	/* check if skb is a EAP message 3/4 */
896 	if (!wil_skb_is_eap_3(wil, skb))
897 		return;
898 
899 	if (vif->ptk_rekey_state == WIL_REKEY_IDLE)
900 		vif->ptk_rekey_state = WIL_REKEY_M3_RECEIVED;
901 }
902 
903 /*
904  * Pass Rx packet to the netif. Update statistics.
905  * Called in softirq context (NAPI poll).
906  */
907 void wil_netif_rx(struct sk_buff *skb, struct net_device *ndev, int cid,
908 		  struct wil_net_stats *stats, bool gro)
909 {
910 	gro_result_t rc = GRO_NORMAL;
911 	struct wil6210_vif *vif = ndev_to_vif(ndev);
912 	struct wil6210_priv *wil = ndev_to_wil(ndev);
913 	struct wireless_dev *wdev = vif_to_wdev(vif);
914 	unsigned int len = skb->len;
915 	u8 *sa, *da = wil_skb_get_da(skb);
916 	/* here looking for DA, not A1, thus Rxdesc's 'mcast' indication
917 	 * is not suitable, need to look at data
918 	 */
919 	int mcast = is_multicast_ether_addr(da);
920 	struct sk_buff *xmit_skb = NULL;
921 	static const char * const gro_res_str[] = {
922 		[GRO_MERGED]		= "GRO_MERGED",
923 		[GRO_MERGED_FREE]	= "GRO_MERGED_FREE",
924 		[GRO_HELD]		= "GRO_HELD",
925 		[GRO_NORMAL]		= "GRO_NORMAL",
926 		[GRO_DROP]		= "GRO_DROP",
927 		[GRO_CONSUMED]		= "GRO_CONSUMED",
928 	};
929 
930 	if (wdev->iftype == NL80211_IFTYPE_STATION) {
931 		sa = wil_skb_get_sa(skb);
932 		if (mcast && ether_addr_equal(sa, ndev->dev_addr)) {
933 			/* mcast packet looped back to us */
934 			rc = GRO_DROP;
935 			dev_kfree_skb(skb);
936 			goto stats;
937 		}
938 	} else if (wdev->iftype == NL80211_IFTYPE_AP && !vif->ap_isolate) {
939 		if (mcast) {
940 			/* send multicast frames both to higher layers in
941 			 * local net stack and back to the wireless medium
942 			 */
943 			xmit_skb = skb_copy(skb, GFP_ATOMIC);
944 		} else {
945 			int xmit_cid = wil_find_cid(wil, vif->mid, da);
946 
947 			if (xmit_cid >= 0) {
948 				/* The destination station is associated to
949 				 * this AP (in this VLAN), so send the frame
950 				 * directly to it and do not pass it to local
951 				 * net stack.
952 				 */
953 				xmit_skb = skb;
954 				skb = NULL;
955 			}
956 		}
957 	}
958 	if (xmit_skb) {
959 		/* Send to wireless media and increase priority by 256 to
960 		 * keep the received priority instead of reclassifying
961 		 * the frame (see cfg80211_classify8021d).
962 		 */
963 		xmit_skb->dev = ndev;
964 		xmit_skb->priority += 256;
965 		xmit_skb->protocol = htons(ETH_P_802_3);
966 		skb_reset_network_header(xmit_skb);
967 		skb_reset_mac_header(xmit_skb);
968 		wil_dbg_txrx(wil, "Rx -> Tx %d bytes\n", len);
969 		dev_queue_xmit(xmit_skb);
970 	}
971 
972 	if (skb) { /* deliver to local stack */
973 		skb->protocol = eth_type_trans(skb, ndev);
974 		skb->dev = ndev;
975 
976 		if (skb->protocol == cpu_to_be16(ETH_P_PAE))
977 			wil_rx_handle_eapol(vif, skb);
978 
979 		if (gro)
980 			rc = napi_gro_receive(&wil->napi_rx, skb);
981 		else
982 			netif_rx_ni(skb);
983 		wil_dbg_txrx(wil, "Rx complete %d bytes => %s\n",
984 			     len, gro_res_str[rc]);
985 	}
986 stats:
987 	/* statistics. rc set to GRO_NORMAL for AP bridging */
988 	if (unlikely(rc == GRO_DROP)) {
989 		ndev->stats.rx_dropped++;
990 		stats->rx_dropped++;
991 		wil_dbg_txrx(wil, "Rx drop %d bytes\n", len);
992 	} else {
993 		ndev->stats.rx_packets++;
994 		stats->rx_packets++;
995 		ndev->stats.rx_bytes += len;
996 		stats->rx_bytes += len;
997 		if (mcast)
998 			ndev->stats.multicast++;
999 	}
1000 }
1001 
1002 void wil_netif_rx_any(struct sk_buff *skb, struct net_device *ndev)
1003 {
1004 	int cid, security;
1005 	struct wil6210_priv *wil = ndev_to_wil(ndev);
1006 	struct wil_net_stats *stats;
1007 
1008 	wil->txrx_ops.get_netif_rx_params(skb, &cid, &security);
1009 
1010 	stats = &wil->sta[cid].stats;
1011 
1012 	skb_orphan(skb);
1013 
1014 	if (security && (wil->txrx_ops.rx_crypto_check(wil, skb) != 0)) {
1015 		wil_dbg_txrx(wil, "Rx drop %d bytes\n", skb->len);
1016 		dev_kfree_skb(skb);
1017 		ndev->stats.rx_dropped++;
1018 		stats->rx_replay++;
1019 		stats->rx_dropped++;
1020 		return;
1021 	}
1022 
1023 	/* check errors reported by HW and update statistics */
1024 	if (unlikely(wil->txrx_ops.rx_error_check(wil, skb, stats))) {
1025 		dev_kfree_skb(skb);
1026 		return;
1027 	}
1028 
1029 	wil_netif_rx(skb, ndev, cid, stats, true);
1030 }
1031 
1032 /**
1033  * Proceed all completed skb's from Rx VRING
1034  *
1035  * Safe to call from NAPI poll, i.e. softirq with interrupts enabled
1036  */
1037 void wil_rx_handle(struct wil6210_priv *wil, int *quota)
1038 {
1039 	struct net_device *ndev = wil->main_ndev;
1040 	struct wireless_dev *wdev = ndev->ieee80211_ptr;
1041 	struct wil_ring *v = &wil->ring_rx;
1042 	struct sk_buff *skb;
1043 
1044 	if (unlikely(!v->va)) {
1045 		wil_err(wil, "Rx IRQ while Rx not yet initialized\n");
1046 		return;
1047 	}
1048 	wil_dbg_txrx(wil, "rx_handle\n");
1049 	while ((*quota > 0) && (NULL != (skb = wil_vring_reap_rx(wil, v)))) {
1050 		(*quota)--;
1051 
1052 		/* monitor is currently supported on main interface only */
1053 		if (wdev->iftype == NL80211_IFTYPE_MONITOR) {
1054 			skb->dev = ndev;
1055 			skb_reset_mac_header(skb);
1056 			skb->ip_summed = CHECKSUM_UNNECESSARY;
1057 			skb->pkt_type = PACKET_OTHERHOST;
1058 			skb->protocol = htons(ETH_P_802_2);
1059 			wil_netif_rx_any(skb, ndev);
1060 		} else {
1061 			wil_rx_reorder(wil, skb);
1062 		}
1063 	}
1064 	wil_rx_refill(wil, v->size);
1065 }
1066 
1067 static void wil_rx_buf_len_init(struct wil6210_priv *wil)
1068 {
1069 	wil->rx_buf_len = rx_large_buf ?
1070 		WIL_MAX_ETH_MTU : TXRX_BUF_LEN_DEFAULT - WIL_MAX_MPDU_OVERHEAD;
1071 	if (mtu_max > wil->rx_buf_len) {
1072 		/* do not allow RX buffers to be smaller than mtu_max, for
1073 		 * backward compatibility (mtu_max parameter was also used
1074 		 * to support receiving large packets)
1075 		 */
1076 		wil_info(wil, "Override RX buffer to mtu_max(%d)\n", mtu_max);
1077 		wil->rx_buf_len = mtu_max;
1078 	}
1079 }
1080 
1081 static int wil_rx_init(struct wil6210_priv *wil, uint order)
1082 {
1083 	struct wil_ring *vring = &wil->ring_rx;
1084 	int rc;
1085 
1086 	wil_dbg_misc(wil, "rx_init\n");
1087 
1088 	if (vring->va) {
1089 		wil_err(wil, "Rx ring already allocated\n");
1090 		return -EINVAL;
1091 	}
1092 
1093 	wil_rx_buf_len_init(wil);
1094 
1095 	vring->size = 1 << order;
1096 	vring->is_rx = true;
1097 	rc = wil_vring_alloc(wil, vring);
1098 	if (rc)
1099 		return rc;
1100 
1101 	rc = wmi_rx_chain_add(wil, vring);
1102 	if (rc)
1103 		goto err_free;
1104 
1105 	rc = wil_rx_refill(wil, vring->size);
1106 	if (rc)
1107 		goto err_free;
1108 
1109 	return 0;
1110  err_free:
1111 	wil_vring_free(wil, vring);
1112 
1113 	return rc;
1114 }
1115 
1116 static void wil_rx_fini(struct wil6210_priv *wil)
1117 {
1118 	struct wil_ring *vring = &wil->ring_rx;
1119 
1120 	wil_dbg_misc(wil, "rx_fini\n");
1121 
1122 	if (vring->va)
1123 		wil_vring_free(wil, vring);
1124 }
1125 
1126 static int wil_tx_desc_map(union wil_tx_desc *desc, dma_addr_t pa,
1127 			   u32 len, int vring_index)
1128 {
1129 	struct vring_tx_desc *d = &desc->legacy;
1130 
1131 	wil_desc_addr_set(&d->dma.addr, pa);
1132 	d->dma.ip_length = 0;
1133 	/* 0..6: mac_length; 7:ip_version 0-IP6 1-IP4*/
1134 	d->dma.b11 = 0/*14 | BIT(7)*/;
1135 	d->dma.error = 0;
1136 	d->dma.status = 0; /* BIT(0) should be 0 for HW_OWNED */
1137 	d->dma.length = cpu_to_le16((u16)len);
1138 	d->dma.d0 = (vring_index << DMA_CFG_DESC_TX_0_QID_POS);
1139 	d->mac.d[0] = 0;
1140 	d->mac.d[1] = 0;
1141 	d->mac.d[2] = 0;
1142 	d->mac.ucode_cmd = 0;
1143 	/* translation type:  0 - bypass; 1 - 802.3; 2 - native wifi */
1144 	d->mac.d[2] = BIT(MAC_CFG_DESC_TX_2_SNAP_HDR_INSERTION_EN_POS) |
1145 		      (1 << MAC_CFG_DESC_TX_2_L2_TRANSLATION_TYPE_POS);
1146 
1147 	return 0;
1148 }
1149 
1150 void wil_tx_data_init(struct wil_ring_tx_data *txdata)
1151 {
1152 	spin_lock_bh(&txdata->lock);
1153 	txdata->dot1x_open = 0;
1154 	txdata->enabled = 0;
1155 	txdata->idle = 0;
1156 	txdata->last_idle = 0;
1157 	txdata->begin = 0;
1158 	txdata->agg_wsize = 0;
1159 	txdata->agg_timeout = 0;
1160 	txdata->agg_amsdu = 0;
1161 	txdata->addba_in_progress = false;
1162 	txdata->mid = U8_MAX;
1163 	spin_unlock_bh(&txdata->lock);
1164 }
1165 
1166 static int wil_vring_init_tx(struct wil6210_vif *vif, int id, int size,
1167 			     int cid, int tid)
1168 {
1169 	struct wil6210_priv *wil = vif_to_wil(vif);
1170 	int rc;
1171 	struct wmi_vring_cfg_cmd cmd = {
1172 		.action = cpu_to_le32(WMI_VRING_CMD_ADD),
1173 		.vring_cfg = {
1174 			.tx_sw_ring = {
1175 				.max_mpdu_size =
1176 					cpu_to_le16(wil_mtu2macbuf(mtu_max)),
1177 				.ring_size = cpu_to_le16(size),
1178 			},
1179 			.ringid = id,
1180 			.encap_trans_type = WMI_VRING_ENC_TYPE_802_3,
1181 			.mac_ctrl = 0,
1182 			.to_resolution = 0,
1183 			.agg_max_wsize = 0,
1184 			.schd_params = {
1185 				.priority = cpu_to_le16(0),
1186 				.timeslot_us = cpu_to_le16(0xfff),
1187 			},
1188 		},
1189 	};
1190 	struct {
1191 		struct wmi_cmd_hdr wmi;
1192 		struct wmi_vring_cfg_done_event cmd;
1193 	} __packed reply = {
1194 		.cmd = {.status = WMI_FW_STATUS_FAILURE},
1195 	};
1196 	struct wil_ring *vring = &wil->ring_tx[id];
1197 	struct wil_ring_tx_data *txdata = &wil->ring_tx_data[id];
1198 
1199 	if (cid >= WIL6210_RX_DESC_MAX_CID) {
1200 		cmd.vring_cfg.cidxtid = CIDXTID_EXTENDED_CID_TID;
1201 		cmd.vring_cfg.cid = cid;
1202 		cmd.vring_cfg.tid = tid;
1203 	} else {
1204 		cmd.vring_cfg.cidxtid = mk_cidxtid(cid, tid);
1205 	}
1206 
1207 	wil_dbg_misc(wil, "vring_init_tx: max_mpdu_size %d\n",
1208 		     cmd.vring_cfg.tx_sw_ring.max_mpdu_size);
1209 	lockdep_assert_held(&wil->mutex);
1210 
1211 	if (vring->va) {
1212 		wil_err(wil, "Tx ring [%d] already allocated\n", id);
1213 		rc = -EINVAL;
1214 		goto out;
1215 	}
1216 
1217 	wil_tx_data_init(txdata);
1218 	vring->is_rx = false;
1219 	vring->size = size;
1220 	rc = wil_vring_alloc(wil, vring);
1221 	if (rc)
1222 		goto out;
1223 
1224 	wil->ring2cid_tid[id][0] = cid;
1225 	wil->ring2cid_tid[id][1] = tid;
1226 
1227 	cmd.vring_cfg.tx_sw_ring.ring_mem_base = cpu_to_le64(vring->pa);
1228 
1229 	if (!vif->privacy)
1230 		txdata->dot1x_open = true;
1231 	rc = wmi_call(wil, WMI_VRING_CFG_CMDID, vif->mid, &cmd, sizeof(cmd),
1232 		      WMI_VRING_CFG_DONE_EVENTID, &reply, sizeof(reply),
1233 		      WIL_WMI_CALL_GENERAL_TO_MS);
1234 	if (rc)
1235 		goto out_free;
1236 
1237 	if (reply.cmd.status != WMI_FW_STATUS_SUCCESS) {
1238 		wil_err(wil, "Tx config failed, status 0x%02x\n",
1239 			reply.cmd.status);
1240 		rc = -EINVAL;
1241 		goto out_free;
1242 	}
1243 
1244 	spin_lock_bh(&txdata->lock);
1245 	vring->hwtail = le32_to_cpu(reply.cmd.tx_vring_tail_ptr);
1246 	txdata->mid = vif->mid;
1247 	txdata->enabled = 1;
1248 	spin_unlock_bh(&txdata->lock);
1249 
1250 	if (txdata->dot1x_open && (agg_wsize >= 0))
1251 		wil_addba_tx_request(wil, id, agg_wsize);
1252 
1253 	return 0;
1254  out_free:
1255 	spin_lock_bh(&txdata->lock);
1256 	txdata->dot1x_open = false;
1257 	txdata->enabled = 0;
1258 	spin_unlock_bh(&txdata->lock);
1259 	wil_vring_free(wil, vring);
1260 	wil->ring2cid_tid[id][0] = wil->max_assoc_sta;
1261 	wil->ring2cid_tid[id][1] = 0;
1262 
1263  out:
1264 
1265 	return rc;
1266 }
1267 
1268 static int wil_tx_vring_modify(struct wil6210_vif *vif, int ring_id, int cid,
1269 			       int tid)
1270 {
1271 	struct wil6210_priv *wil = vif_to_wil(vif);
1272 	int rc;
1273 	struct wmi_vring_cfg_cmd cmd = {
1274 		.action = cpu_to_le32(WMI_VRING_CMD_MODIFY),
1275 		.vring_cfg = {
1276 			.tx_sw_ring = {
1277 				.max_mpdu_size =
1278 					cpu_to_le16(wil_mtu2macbuf(mtu_max)),
1279 				.ring_size = 0,
1280 			},
1281 			.ringid = ring_id,
1282 			.cidxtid = mk_cidxtid(cid, tid),
1283 			.encap_trans_type = WMI_VRING_ENC_TYPE_802_3,
1284 			.mac_ctrl = 0,
1285 			.to_resolution = 0,
1286 			.agg_max_wsize = 0,
1287 			.schd_params = {
1288 				.priority = cpu_to_le16(0),
1289 				.timeslot_us = cpu_to_le16(0xfff),
1290 			},
1291 		},
1292 	};
1293 	struct {
1294 		struct wmi_cmd_hdr wmi;
1295 		struct wmi_vring_cfg_done_event cmd;
1296 	} __packed reply = {
1297 		.cmd = {.status = WMI_FW_STATUS_FAILURE},
1298 	};
1299 	struct wil_ring *vring = &wil->ring_tx[ring_id];
1300 	struct wil_ring_tx_data *txdata = &wil->ring_tx_data[ring_id];
1301 
1302 	wil_dbg_misc(wil, "vring_modify: ring %d cid %d tid %d\n", ring_id,
1303 		     cid, tid);
1304 	lockdep_assert_held(&wil->mutex);
1305 
1306 	if (!vring->va) {
1307 		wil_err(wil, "Tx ring [%d] not allocated\n", ring_id);
1308 		return -EINVAL;
1309 	}
1310 
1311 	if (wil->ring2cid_tid[ring_id][0] != cid ||
1312 	    wil->ring2cid_tid[ring_id][1] != tid) {
1313 		wil_err(wil, "ring info does not match cid=%u tid=%u\n",
1314 			wil->ring2cid_tid[ring_id][0],
1315 			wil->ring2cid_tid[ring_id][1]);
1316 	}
1317 
1318 	cmd.vring_cfg.tx_sw_ring.ring_mem_base = cpu_to_le64(vring->pa);
1319 
1320 	rc = wmi_call(wil, WMI_VRING_CFG_CMDID, vif->mid, &cmd, sizeof(cmd),
1321 		      WMI_VRING_CFG_DONE_EVENTID, &reply, sizeof(reply),
1322 		      WIL_WMI_CALL_GENERAL_TO_MS);
1323 	if (rc)
1324 		goto fail;
1325 
1326 	if (reply.cmd.status != WMI_FW_STATUS_SUCCESS) {
1327 		wil_err(wil, "Tx modify failed, status 0x%02x\n",
1328 			reply.cmd.status);
1329 		rc = -EINVAL;
1330 		goto fail;
1331 	}
1332 
1333 	/* set BA aggregation window size to 0 to force a new BA with the
1334 	 * new AP
1335 	 */
1336 	txdata->agg_wsize = 0;
1337 	if (txdata->dot1x_open && agg_wsize >= 0)
1338 		wil_addba_tx_request(wil, ring_id, agg_wsize);
1339 
1340 	return 0;
1341 fail:
1342 	spin_lock_bh(&txdata->lock);
1343 	txdata->dot1x_open = false;
1344 	txdata->enabled = 0;
1345 	spin_unlock_bh(&txdata->lock);
1346 	wil->ring2cid_tid[ring_id][0] = wil->max_assoc_sta;
1347 	wil->ring2cid_tid[ring_id][1] = 0;
1348 	return rc;
1349 }
1350 
1351 int wil_vring_init_bcast(struct wil6210_vif *vif, int id, int size)
1352 {
1353 	struct wil6210_priv *wil = vif_to_wil(vif);
1354 	int rc;
1355 	struct wmi_bcast_vring_cfg_cmd cmd = {
1356 		.action = cpu_to_le32(WMI_VRING_CMD_ADD),
1357 		.vring_cfg = {
1358 			.tx_sw_ring = {
1359 				.max_mpdu_size =
1360 					cpu_to_le16(wil_mtu2macbuf(mtu_max)),
1361 				.ring_size = cpu_to_le16(size),
1362 			},
1363 			.ringid = id,
1364 			.encap_trans_type = WMI_VRING_ENC_TYPE_802_3,
1365 		},
1366 	};
1367 	struct {
1368 		struct wmi_cmd_hdr wmi;
1369 		struct wmi_vring_cfg_done_event cmd;
1370 	} __packed reply = {
1371 		.cmd = {.status = WMI_FW_STATUS_FAILURE},
1372 	};
1373 	struct wil_ring *vring = &wil->ring_tx[id];
1374 	struct wil_ring_tx_data *txdata = &wil->ring_tx_data[id];
1375 
1376 	wil_dbg_misc(wil, "vring_init_bcast: max_mpdu_size %d\n",
1377 		     cmd.vring_cfg.tx_sw_ring.max_mpdu_size);
1378 	lockdep_assert_held(&wil->mutex);
1379 
1380 	if (vring->va) {
1381 		wil_err(wil, "Tx ring [%d] already allocated\n", id);
1382 		rc = -EINVAL;
1383 		goto out;
1384 	}
1385 
1386 	wil_tx_data_init(txdata);
1387 	vring->is_rx = false;
1388 	vring->size = size;
1389 	rc = wil_vring_alloc(wil, vring);
1390 	if (rc)
1391 		goto out;
1392 
1393 	wil->ring2cid_tid[id][0] = wil->max_assoc_sta; /* CID */
1394 	wil->ring2cid_tid[id][1] = 0; /* TID */
1395 
1396 	cmd.vring_cfg.tx_sw_ring.ring_mem_base = cpu_to_le64(vring->pa);
1397 
1398 	if (!vif->privacy)
1399 		txdata->dot1x_open = true;
1400 	rc = wmi_call(wil, WMI_BCAST_VRING_CFG_CMDID, vif->mid,
1401 		      &cmd, sizeof(cmd),
1402 		      WMI_VRING_CFG_DONE_EVENTID, &reply, sizeof(reply),
1403 		      WIL_WMI_CALL_GENERAL_TO_MS);
1404 	if (rc)
1405 		goto out_free;
1406 
1407 	if (reply.cmd.status != WMI_FW_STATUS_SUCCESS) {
1408 		wil_err(wil, "Tx config failed, status 0x%02x\n",
1409 			reply.cmd.status);
1410 		rc = -EINVAL;
1411 		goto out_free;
1412 	}
1413 
1414 	spin_lock_bh(&txdata->lock);
1415 	vring->hwtail = le32_to_cpu(reply.cmd.tx_vring_tail_ptr);
1416 	txdata->mid = vif->mid;
1417 	txdata->enabled = 1;
1418 	spin_unlock_bh(&txdata->lock);
1419 
1420 	return 0;
1421  out_free:
1422 	spin_lock_bh(&txdata->lock);
1423 	txdata->enabled = 0;
1424 	txdata->dot1x_open = false;
1425 	spin_unlock_bh(&txdata->lock);
1426 	wil_vring_free(wil, vring);
1427  out:
1428 
1429 	return rc;
1430 }
1431 
1432 static struct wil_ring *wil_find_tx_ucast(struct wil6210_priv *wil,
1433 					  struct wil6210_vif *vif,
1434 					  struct sk_buff *skb)
1435 {
1436 	int i, cid;
1437 	const u8 *da = wil_skb_get_da(skb);
1438 	int min_ring_id = wil_get_min_tx_ring_id(wil);
1439 
1440 	cid = wil_find_cid(wil, vif->mid, da);
1441 
1442 	if (cid < 0 || cid >= wil->max_assoc_sta)
1443 		return NULL;
1444 
1445 	/* TODO: fix for multiple TID */
1446 	for (i = min_ring_id; i < ARRAY_SIZE(wil->ring2cid_tid); i++) {
1447 		if (!wil->ring_tx_data[i].dot1x_open &&
1448 		    skb->protocol != cpu_to_be16(ETH_P_PAE))
1449 			continue;
1450 		if (wil->ring2cid_tid[i][0] == cid) {
1451 			struct wil_ring *v = &wil->ring_tx[i];
1452 			struct wil_ring_tx_data *txdata = &wil->ring_tx_data[i];
1453 
1454 			wil_dbg_txrx(wil, "find_tx_ucast: (%pM) -> [%d]\n",
1455 				     da, i);
1456 			if (v->va && txdata->enabled) {
1457 				return v;
1458 			} else {
1459 				wil_dbg_txrx(wil,
1460 					     "find_tx_ucast: vring[%d] not valid\n",
1461 					     i);
1462 				return NULL;
1463 			}
1464 		}
1465 	}
1466 
1467 	return NULL;
1468 }
1469 
1470 static int wil_tx_ring(struct wil6210_priv *wil, struct wil6210_vif *vif,
1471 		       struct wil_ring *ring, struct sk_buff *skb);
1472 
1473 static struct wil_ring *wil_find_tx_ring_sta(struct wil6210_priv *wil,
1474 					     struct wil6210_vif *vif,
1475 					     struct sk_buff *skb)
1476 {
1477 	struct wil_ring *ring;
1478 	int i;
1479 	u8 cid;
1480 	struct wil_ring_tx_data  *txdata;
1481 	int min_ring_id = wil_get_min_tx_ring_id(wil);
1482 
1483 	/* In the STA mode, it is expected to have only 1 VRING
1484 	 * for the AP we connected to.
1485 	 * find 1-st vring eligible for this skb and use it.
1486 	 */
1487 	for (i = min_ring_id; i < WIL6210_MAX_TX_RINGS; i++) {
1488 		ring = &wil->ring_tx[i];
1489 		txdata = &wil->ring_tx_data[i];
1490 		if (!ring->va || !txdata->enabled || txdata->mid != vif->mid)
1491 			continue;
1492 
1493 		cid = wil->ring2cid_tid[i][0];
1494 		if (cid >= wil->max_assoc_sta) /* skip BCAST */
1495 			continue;
1496 
1497 		if (!wil->ring_tx_data[i].dot1x_open &&
1498 		    skb->protocol != cpu_to_be16(ETH_P_PAE))
1499 			continue;
1500 
1501 		wil_dbg_txrx(wil, "Tx -> ring %d\n", i);
1502 
1503 		return ring;
1504 	}
1505 
1506 	wil_dbg_txrx(wil, "Tx while no rings active?\n");
1507 
1508 	return NULL;
1509 }
1510 
1511 /* Use one of 2 strategies:
1512  *
1513  * 1. New (real broadcast):
1514  *    use dedicated broadcast vring
1515  * 2. Old (pseudo-DMS):
1516  *    Find 1-st vring and return it;
1517  *    duplicate skb and send it to other active vrings;
1518  *    in all cases override dest address to unicast peer's address
1519  * Use old strategy when new is not supported yet:
1520  *  - for PBSS
1521  */
1522 static struct wil_ring *wil_find_tx_bcast_1(struct wil6210_priv *wil,
1523 					    struct wil6210_vif *vif,
1524 					    struct sk_buff *skb)
1525 {
1526 	struct wil_ring *v;
1527 	struct wil_ring_tx_data *txdata;
1528 	int i = vif->bcast_ring;
1529 
1530 	if (i < 0)
1531 		return NULL;
1532 	v = &wil->ring_tx[i];
1533 	txdata = &wil->ring_tx_data[i];
1534 	if (!v->va || !txdata->enabled)
1535 		return NULL;
1536 	if (!wil->ring_tx_data[i].dot1x_open &&
1537 	    skb->protocol != cpu_to_be16(ETH_P_PAE))
1538 		return NULL;
1539 
1540 	return v;
1541 }
1542 
1543 static void wil_set_da_for_vring(struct wil6210_priv *wil,
1544 				 struct sk_buff *skb, int vring_index)
1545 {
1546 	u8 *da = wil_skb_get_da(skb);
1547 	int cid = wil->ring2cid_tid[vring_index][0];
1548 
1549 	ether_addr_copy(da, wil->sta[cid].addr);
1550 }
1551 
1552 static struct wil_ring *wil_find_tx_bcast_2(struct wil6210_priv *wil,
1553 					    struct wil6210_vif *vif,
1554 					    struct sk_buff *skb)
1555 {
1556 	struct wil_ring *v, *v2;
1557 	struct sk_buff *skb2;
1558 	int i;
1559 	u8 cid;
1560 	const u8 *src = wil_skb_get_sa(skb);
1561 	struct wil_ring_tx_data *txdata, *txdata2;
1562 	int min_ring_id = wil_get_min_tx_ring_id(wil);
1563 
1564 	/* find 1-st vring eligible for data */
1565 	for (i = min_ring_id; i < WIL6210_MAX_TX_RINGS; i++) {
1566 		v = &wil->ring_tx[i];
1567 		txdata = &wil->ring_tx_data[i];
1568 		if (!v->va || !txdata->enabled || txdata->mid != vif->mid)
1569 			continue;
1570 
1571 		cid = wil->ring2cid_tid[i][0];
1572 		if (cid >= wil->max_assoc_sta) /* skip BCAST */
1573 			continue;
1574 		if (!wil->ring_tx_data[i].dot1x_open &&
1575 		    skb->protocol != cpu_to_be16(ETH_P_PAE))
1576 			continue;
1577 
1578 		/* don't Tx back to source when re-routing Rx->Tx at the AP */
1579 		if (0 == memcmp(wil->sta[cid].addr, src, ETH_ALEN))
1580 			continue;
1581 
1582 		goto found;
1583 	}
1584 
1585 	wil_dbg_txrx(wil, "Tx while no vrings active?\n");
1586 
1587 	return NULL;
1588 
1589 found:
1590 	wil_dbg_txrx(wil, "BCAST -> ring %d\n", i);
1591 	wil_set_da_for_vring(wil, skb, i);
1592 
1593 	/* find other active vrings and duplicate skb for each */
1594 	for (i++; i < WIL6210_MAX_TX_RINGS; i++) {
1595 		v2 = &wil->ring_tx[i];
1596 		txdata2 = &wil->ring_tx_data[i];
1597 		if (!v2->va || txdata2->mid != vif->mid)
1598 			continue;
1599 		cid = wil->ring2cid_tid[i][0];
1600 		if (cid >= wil->max_assoc_sta) /* skip BCAST */
1601 			continue;
1602 		if (!wil->ring_tx_data[i].dot1x_open &&
1603 		    skb->protocol != cpu_to_be16(ETH_P_PAE))
1604 			continue;
1605 
1606 		if (0 == memcmp(wil->sta[cid].addr, src, ETH_ALEN))
1607 			continue;
1608 
1609 		skb2 = skb_copy(skb, GFP_ATOMIC);
1610 		if (skb2) {
1611 			wil_dbg_txrx(wil, "BCAST DUP -> ring %d\n", i);
1612 			wil_set_da_for_vring(wil, skb2, i);
1613 			wil_tx_ring(wil, vif, v2, skb2);
1614 			/* successful call to wil_tx_ring takes skb2 ref */
1615 			dev_kfree_skb_any(skb2);
1616 		} else {
1617 			wil_err(wil, "skb_copy failed\n");
1618 		}
1619 	}
1620 
1621 	return v;
1622 }
1623 
1624 static inline
1625 void wil_tx_desc_set_nr_frags(struct vring_tx_desc *d, int nr_frags)
1626 {
1627 	d->mac.d[2] |= (nr_frags << MAC_CFG_DESC_TX_2_NUM_OF_DESCRIPTORS_POS);
1628 }
1629 
1630 /**
1631  * Sets the descriptor @d up for csum and/or TSO offloading. The corresponding
1632  * @skb is used to obtain the protocol and headers length.
1633  * @tso_desc_type is a descriptor type for TSO: 0 - a header, 1 - first data,
1634  * 2 - middle, 3 - last descriptor.
1635  */
1636 
1637 static void wil_tx_desc_offload_setup_tso(struct vring_tx_desc *d,
1638 					  struct sk_buff *skb,
1639 					  int tso_desc_type, bool is_ipv4,
1640 					  int tcp_hdr_len, int skb_net_hdr_len)
1641 {
1642 	d->dma.b11 = ETH_HLEN; /* MAC header length */
1643 	d->dma.b11 |= is_ipv4 << DMA_CFG_DESC_TX_OFFLOAD_CFG_L3T_IPV4_POS;
1644 
1645 	d->dma.d0 |= (2 << DMA_CFG_DESC_TX_0_L4_TYPE_POS);
1646 	/* L4 header len: TCP header length */
1647 	d->dma.d0 |= (tcp_hdr_len & DMA_CFG_DESC_TX_0_L4_LENGTH_MSK);
1648 
1649 	/* Setup TSO: bit and desc type */
1650 	d->dma.d0 |= (BIT(DMA_CFG_DESC_TX_0_TCP_SEG_EN_POS)) |
1651 		(tso_desc_type << DMA_CFG_DESC_TX_0_SEGMENT_BUF_DETAILS_POS);
1652 	d->dma.d0 |= (is_ipv4 << DMA_CFG_DESC_TX_0_IPV4_CHECKSUM_EN_POS);
1653 
1654 	d->dma.ip_length = skb_net_hdr_len;
1655 	/* Enable TCP/UDP checksum */
1656 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_TCP_UDP_CHECKSUM_EN_POS);
1657 	/* Calculate pseudo-header */
1658 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_PSEUDO_HEADER_CALC_EN_POS);
1659 }
1660 
1661 /**
1662  * Sets the descriptor @d up for csum. The corresponding
1663  * @skb is used to obtain the protocol and headers length.
1664  * Returns the protocol: 0 - not TCP, 1 - TCPv4, 2 - TCPv6.
1665  * Note, if d==NULL, the function only returns the protocol result.
1666  *
1667  * It is very similar to previous wil_tx_desc_offload_setup_tso. This
1668  * is "if unrolling" to optimize the critical path.
1669  */
1670 
1671 static int wil_tx_desc_offload_setup(struct vring_tx_desc *d,
1672 				     struct sk_buff *skb){
1673 	int protocol;
1674 
1675 	if (skb->ip_summed != CHECKSUM_PARTIAL)
1676 		return 0;
1677 
1678 	d->dma.b11 = ETH_HLEN; /* MAC header length */
1679 
1680 	switch (skb->protocol) {
1681 	case cpu_to_be16(ETH_P_IP):
1682 		protocol = ip_hdr(skb)->protocol;
1683 		d->dma.b11 |= BIT(DMA_CFG_DESC_TX_OFFLOAD_CFG_L3T_IPV4_POS);
1684 		break;
1685 	case cpu_to_be16(ETH_P_IPV6):
1686 		protocol = ipv6_hdr(skb)->nexthdr;
1687 		break;
1688 	default:
1689 		return -EINVAL;
1690 	}
1691 
1692 	switch (protocol) {
1693 	case IPPROTO_TCP:
1694 		d->dma.d0 |= (2 << DMA_CFG_DESC_TX_0_L4_TYPE_POS);
1695 		/* L4 header len: TCP header length */
1696 		d->dma.d0 |=
1697 		(tcp_hdrlen(skb) & DMA_CFG_DESC_TX_0_L4_LENGTH_MSK);
1698 		break;
1699 	case IPPROTO_UDP:
1700 		/* L4 header len: UDP header length */
1701 		d->dma.d0 |=
1702 		(sizeof(struct udphdr) & DMA_CFG_DESC_TX_0_L4_LENGTH_MSK);
1703 		break;
1704 	default:
1705 		return -EINVAL;
1706 	}
1707 
1708 	d->dma.ip_length = skb_network_header_len(skb);
1709 	/* Enable TCP/UDP checksum */
1710 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_TCP_UDP_CHECKSUM_EN_POS);
1711 	/* Calculate pseudo-header */
1712 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_PSEUDO_HEADER_CALC_EN_POS);
1713 
1714 	return 0;
1715 }
1716 
1717 static inline void wil_tx_last_desc(struct vring_tx_desc *d)
1718 {
1719 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_CMD_EOP_POS) |
1720 	      BIT(DMA_CFG_DESC_TX_0_CMD_MARK_WB_POS) |
1721 	      BIT(DMA_CFG_DESC_TX_0_CMD_DMA_IT_POS);
1722 }
1723 
1724 static inline void wil_set_tx_desc_last_tso(volatile struct vring_tx_desc *d)
1725 {
1726 	d->dma.d0 |= wil_tso_type_lst <<
1727 		  DMA_CFG_DESC_TX_0_SEGMENT_BUF_DETAILS_POS;
1728 }
1729 
1730 static int __wil_tx_vring_tso(struct wil6210_priv *wil, struct wil6210_vif *vif,
1731 			      struct wil_ring *vring, struct sk_buff *skb)
1732 {
1733 	struct device *dev = wil_to_dev(wil);
1734 
1735 	/* point to descriptors in shared memory */
1736 	volatile struct vring_tx_desc *_desc = NULL, *_hdr_desc,
1737 				      *_first_desc = NULL;
1738 
1739 	/* pointers to shadow descriptors */
1740 	struct vring_tx_desc desc_mem, hdr_desc_mem, first_desc_mem,
1741 			     *d = &hdr_desc_mem, *hdr_desc = &hdr_desc_mem,
1742 			     *first_desc = &first_desc_mem;
1743 
1744 	/* pointer to shadow descriptors' context */
1745 	struct wil_ctx *hdr_ctx, *first_ctx = NULL;
1746 
1747 	int descs_used = 0; /* total number of used descriptors */
1748 	int sg_desc_cnt = 0; /* number of descriptors for current mss*/
1749 
1750 	u32 swhead = vring->swhead;
1751 	int used, avail = wil_ring_avail_tx(vring);
1752 	int nr_frags = skb_shinfo(skb)->nr_frags;
1753 	int min_desc_required = nr_frags + 1;
1754 	int mss = skb_shinfo(skb)->gso_size;	/* payload size w/o headers */
1755 	int f, len, hdrlen, headlen;
1756 	int vring_index = vring - wil->ring_tx;
1757 	struct wil_ring_tx_data *txdata = &wil->ring_tx_data[vring_index];
1758 	uint i = swhead;
1759 	dma_addr_t pa;
1760 	const skb_frag_t *frag = NULL;
1761 	int rem_data = mss;
1762 	int lenmss;
1763 	int hdr_compensation_need = true;
1764 	int desc_tso_type = wil_tso_type_first;
1765 	bool is_ipv4;
1766 	int tcp_hdr_len;
1767 	int skb_net_hdr_len;
1768 	int gso_type;
1769 	int rc = -EINVAL;
1770 
1771 	wil_dbg_txrx(wil, "tx_vring_tso: %d bytes to vring %d\n", skb->len,
1772 		     vring_index);
1773 
1774 	if (unlikely(!txdata->enabled))
1775 		return -EINVAL;
1776 
1777 	/* A typical page 4K is 3-4 payloads, we assume each fragment
1778 	 * is a full payload, that's how min_desc_required has been
1779 	 * calculated. In real we might need more or less descriptors,
1780 	 * this is the initial check only.
1781 	 */
1782 	if (unlikely(avail < min_desc_required)) {
1783 		wil_err_ratelimited(wil,
1784 				    "TSO: Tx ring[%2d] full. No space for %d fragments\n",
1785 				    vring_index, min_desc_required);
1786 		return -ENOMEM;
1787 	}
1788 
1789 	/* Header Length = MAC header len + IP header len + TCP header len*/
1790 	hdrlen = ETH_HLEN +
1791 		(int)skb_network_header_len(skb) +
1792 		tcp_hdrlen(skb);
1793 
1794 	gso_type = skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV6 | SKB_GSO_TCPV4);
1795 	switch (gso_type) {
1796 	case SKB_GSO_TCPV4:
1797 		/* TCP v4, zero out the IP length and IPv4 checksum fields
1798 		 * as required by the offloading doc
1799 		 */
1800 		ip_hdr(skb)->tot_len = 0;
1801 		ip_hdr(skb)->check = 0;
1802 		is_ipv4 = true;
1803 		break;
1804 	case SKB_GSO_TCPV6:
1805 		/* TCP v6, zero out the payload length */
1806 		ipv6_hdr(skb)->payload_len = 0;
1807 		is_ipv4 = false;
1808 		break;
1809 	default:
1810 		/* other than TCPv4 or TCPv6 types are not supported for TSO.
1811 		 * It is also illegal for both to be set simultaneously
1812 		 */
1813 		return -EINVAL;
1814 	}
1815 
1816 	if (skb->ip_summed != CHECKSUM_PARTIAL)
1817 		return -EINVAL;
1818 
1819 	/* tcp header length and skb network header length are fixed for all
1820 	 * packet's descriptors - read then once here
1821 	 */
1822 	tcp_hdr_len = tcp_hdrlen(skb);
1823 	skb_net_hdr_len = skb_network_header_len(skb);
1824 
1825 	_hdr_desc = &vring->va[i].tx.legacy;
1826 
1827 	pa = dma_map_single(dev, skb->data, hdrlen, DMA_TO_DEVICE);
1828 	if (unlikely(dma_mapping_error(dev, pa))) {
1829 		wil_err(wil, "TSO: Skb head DMA map error\n");
1830 		goto err_exit;
1831 	}
1832 
1833 	wil->txrx_ops.tx_desc_map((union wil_tx_desc *)hdr_desc, pa,
1834 				  hdrlen, vring_index);
1835 	wil_tx_desc_offload_setup_tso(hdr_desc, skb, wil_tso_type_hdr, is_ipv4,
1836 				      tcp_hdr_len, skb_net_hdr_len);
1837 	wil_tx_last_desc(hdr_desc);
1838 
1839 	vring->ctx[i].mapped_as = wil_mapped_as_single;
1840 	hdr_ctx = &vring->ctx[i];
1841 
1842 	descs_used++;
1843 	headlen = skb_headlen(skb) - hdrlen;
1844 
1845 	for (f = headlen ? -1 : 0; f < nr_frags; f++)  {
1846 		if (headlen) {
1847 			len = headlen;
1848 			wil_dbg_txrx(wil, "TSO: process skb head, len %u\n",
1849 				     len);
1850 		} else {
1851 			frag = &skb_shinfo(skb)->frags[f];
1852 			len = skb_frag_size(frag);
1853 			wil_dbg_txrx(wil, "TSO: frag[%d]: len %u\n", f, len);
1854 		}
1855 
1856 		while (len) {
1857 			wil_dbg_txrx(wil,
1858 				     "TSO: len %d, rem_data %d, descs_used %d\n",
1859 				     len, rem_data, descs_used);
1860 
1861 			if (descs_used == avail)  {
1862 				wil_err_ratelimited(wil, "TSO: ring overflow\n");
1863 				rc = -ENOMEM;
1864 				goto mem_error;
1865 			}
1866 
1867 			lenmss = min_t(int, rem_data, len);
1868 			i = (swhead + descs_used) % vring->size;
1869 			wil_dbg_txrx(wil, "TSO: lenmss %d, i %d\n", lenmss, i);
1870 
1871 			if (!headlen) {
1872 				pa = skb_frag_dma_map(dev, frag,
1873 						      skb_frag_size(frag) - len,
1874 						      lenmss, DMA_TO_DEVICE);
1875 				vring->ctx[i].mapped_as = wil_mapped_as_page;
1876 			} else {
1877 				pa = dma_map_single(dev,
1878 						    skb->data +
1879 						    skb_headlen(skb) - headlen,
1880 						    lenmss,
1881 						    DMA_TO_DEVICE);
1882 				vring->ctx[i].mapped_as = wil_mapped_as_single;
1883 				headlen -= lenmss;
1884 			}
1885 
1886 			if (unlikely(dma_mapping_error(dev, pa))) {
1887 				wil_err(wil, "TSO: DMA map page error\n");
1888 				goto mem_error;
1889 			}
1890 
1891 			_desc = &vring->va[i].tx.legacy;
1892 
1893 			if (!_first_desc) {
1894 				_first_desc = _desc;
1895 				first_ctx = &vring->ctx[i];
1896 				d = first_desc;
1897 			} else {
1898 				d = &desc_mem;
1899 			}
1900 
1901 			wil->txrx_ops.tx_desc_map((union wil_tx_desc *)d,
1902 						  pa, lenmss, vring_index);
1903 			wil_tx_desc_offload_setup_tso(d, skb, desc_tso_type,
1904 						      is_ipv4, tcp_hdr_len,
1905 						      skb_net_hdr_len);
1906 
1907 			/* use tso_type_first only once */
1908 			desc_tso_type = wil_tso_type_mid;
1909 
1910 			descs_used++;  /* desc used so far */
1911 			sg_desc_cnt++; /* desc used for this segment */
1912 			len -= lenmss;
1913 			rem_data -= lenmss;
1914 
1915 			wil_dbg_txrx(wil,
1916 				     "TSO: len %d, rem_data %d, descs_used %d, sg_desc_cnt %d,\n",
1917 				     len, rem_data, descs_used, sg_desc_cnt);
1918 
1919 			/* Close the segment if reached mss size or last frag*/
1920 			if (rem_data == 0 || (f == nr_frags - 1 && len == 0)) {
1921 				if (hdr_compensation_need) {
1922 					/* first segment include hdr desc for
1923 					 * release
1924 					 */
1925 					hdr_ctx->nr_frags = sg_desc_cnt;
1926 					wil_tx_desc_set_nr_frags(first_desc,
1927 								 sg_desc_cnt +
1928 								 1);
1929 					hdr_compensation_need = false;
1930 				} else {
1931 					wil_tx_desc_set_nr_frags(first_desc,
1932 								 sg_desc_cnt);
1933 				}
1934 				first_ctx->nr_frags = sg_desc_cnt - 1;
1935 
1936 				wil_tx_last_desc(d);
1937 
1938 				/* first descriptor may also be the last
1939 				 * for this mss - make sure not to copy
1940 				 * it twice
1941 				 */
1942 				if (first_desc != d)
1943 					*_first_desc = *first_desc;
1944 
1945 				/*last descriptor will be copied at the end
1946 				 * of this TS processing
1947 				 */
1948 				if (f < nr_frags - 1 || len > 0)
1949 					*_desc = *d;
1950 
1951 				rem_data = mss;
1952 				_first_desc = NULL;
1953 				sg_desc_cnt = 0;
1954 			} else if (first_desc != d) /* update mid descriptor */
1955 					*_desc = *d;
1956 		}
1957 	}
1958 
1959 	if (!_desc)
1960 		goto mem_error;
1961 
1962 	/* first descriptor may also be the last.
1963 	 * in this case d pointer is invalid
1964 	 */
1965 	if (_first_desc == _desc)
1966 		d = first_desc;
1967 
1968 	/* Last data descriptor */
1969 	wil_set_tx_desc_last_tso(d);
1970 	*_desc = *d;
1971 
1972 	/* Fill the total number of descriptors in first desc (hdr)*/
1973 	wil_tx_desc_set_nr_frags(hdr_desc, descs_used);
1974 	*_hdr_desc = *hdr_desc;
1975 
1976 	/* hold reference to skb
1977 	 * to prevent skb release before accounting
1978 	 * in case of immediate "tx done"
1979 	 */
1980 	vring->ctx[i].skb = skb_get(skb);
1981 
1982 	/* performance monitoring */
1983 	used = wil_ring_used_tx(vring);
1984 	if (wil_val_in_range(wil->ring_idle_trsh,
1985 			     used, used + descs_used)) {
1986 		txdata->idle += get_cycles() - txdata->last_idle;
1987 		wil_dbg_txrx(wil,  "Ring[%2d] not idle %d -> %d\n",
1988 			     vring_index, used, used + descs_used);
1989 	}
1990 
1991 	/* Make sure to advance the head only after descriptor update is done.
1992 	 * This will prevent a race condition where the completion thread
1993 	 * will see the DU bit set from previous run and will handle the
1994 	 * skb before it was completed.
1995 	 */
1996 	wmb();
1997 
1998 	/* advance swhead */
1999 	wil_ring_advance_head(vring, descs_used);
2000 	wil_dbg_txrx(wil, "TSO: Tx swhead %d -> %d\n", swhead, vring->swhead);
2001 
2002 	/* make sure all writes to descriptors (shared memory) are done before
2003 	 * committing them to HW
2004 	 */
2005 	wmb();
2006 
2007 	if (wil->tx_latency)
2008 		*(ktime_t *)&skb->cb = ktime_get();
2009 	else
2010 		memset(skb->cb, 0, sizeof(ktime_t));
2011 
2012 	wil_w(wil, vring->hwtail, vring->swhead);
2013 	return 0;
2014 
2015 mem_error:
2016 	while (descs_used > 0) {
2017 		struct wil_ctx *ctx;
2018 
2019 		i = (swhead + descs_used - 1) % vring->size;
2020 		d = (struct vring_tx_desc *)&vring->va[i].tx.legacy;
2021 		_desc = &vring->va[i].tx.legacy;
2022 		*d = *_desc;
2023 		_desc->dma.status = TX_DMA_STATUS_DU;
2024 		ctx = &vring->ctx[i];
2025 		wil_txdesc_unmap(dev, (union wil_tx_desc *)d, ctx);
2026 		memset(ctx, 0, sizeof(*ctx));
2027 		descs_used--;
2028 	}
2029 err_exit:
2030 	return rc;
2031 }
2032 
2033 static int __wil_tx_ring(struct wil6210_priv *wil, struct wil6210_vif *vif,
2034 			 struct wil_ring *ring, struct sk_buff *skb)
2035 {
2036 	struct device *dev = wil_to_dev(wil);
2037 	struct vring_tx_desc dd, *d = &dd;
2038 	volatile struct vring_tx_desc *_d;
2039 	u32 swhead = ring->swhead;
2040 	int avail = wil_ring_avail_tx(ring);
2041 	int nr_frags = skb_shinfo(skb)->nr_frags;
2042 	uint f = 0;
2043 	int ring_index = ring - wil->ring_tx;
2044 	struct wil_ring_tx_data  *txdata = &wil->ring_tx_data[ring_index];
2045 	uint i = swhead;
2046 	dma_addr_t pa;
2047 	int used;
2048 	bool mcast = (ring_index == vif->bcast_ring);
2049 	uint len = skb_headlen(skb);
2050 
2051 	wil_dbg_txrx(wil, "tx_ring: %d bytes to ring %d, nr_frags %d\n",
2052 		     skb->len, ring_index, nr_frags);
2053 
2054 	if (unlikely(!txdata->enabled))
2055 		return -EINVAL;
2056 
2057 	if (unlikely(avail < 1 + nr_frags)) {
2058 		wil_err_ratelimited(wil,
2059 				    "Tx ring[%2d] full. No space for %d fragments\n",
2060 				    ring_index, 1 + nr_frags);
2061 		return -ENOMEM;
2062 	}
2063 	_d = &ring->va[i].tx.legacy;
2064 
2065 	pa = dma_map_single(dev, skb->data, skb_headlen(skb), DMA_TO_DEVICE);
2066 
2067 	wil_dbg_txrx(wil, "Tx[%2d] skb %d bytes 0x%p -> %pad\n", ring_index,
2068 		     skb_headlen(skb), skb->data, &pa);
2069 	wil_hex_dump_txrx("Tx ", DUMP_PREFIX_OFFSET, 16, 1,
2070 			  skb->data, skb_headlen(skb), false);
2071 
2072 	if (unlikely(dma_mapping_error(dev, pa)))
2073 		return -EINVAL;
2074 	ring->ctx[i].mapped_as = wil_mapped_as_single;
2075 	/* 1-st segment */
2076 	wil->txrx_ops.tx_desc_map((union wil_tx_desc *)d, pa, len,
2077 				   ring_index);
2078 	if (unlikely(mcast)) {
2079 		d->mac.d[0] |= BIT(MAC_CFG_DESC_TX_0_MCS_EN_POS); /* MCS 0 */
2080 		if (unlikely(len > WIL_BCAST_MCS0_LIMIT)) /* set MCS 1 */
2081 			d->mac.d[0] |= (1 << MAC_CFG_DESC_TX_0_MCS_INDEX_POS);
2082 	}
2083 	/* Process TCP/UDP checksum offloading */
2084 	if (unlikely(wil_tx_desc_offload_setup(d, skb))) {
2085 		wil_err(wil, "Tx[%2d] Failed to set cksum, drop packet\n",
2086 			ring_index);
2087 		goto dma_error;
2088 	}
2089 
2090 	ring->ctx[i].nr_frags = nr_frags;
2091 	wil_tx_desc_set_nr_frags(d, nr_frags + 1);
2092 
2093 	/* middle segments */
2094 	for (; f < nr_frags; f++) {
2095 		const skb_frag_t *frag = &skb_shinfo(skb)->frags[f];
2096 		int len = skb_frag_size(frag);
2097 
2098 		*_d = *d;
2099 		wil_dbg_txrx(wil, "Tx[%2d] desc[%4d]\n", ring_index, i);
2100 		wil_hex_dump_txrx("TxD ", DUMP_PREFIX_NONE, 32, 4,
2101 				  (const void *)d, sizeof(*d), false);
2102 		i = (swhead + f + 1) % ring->size;
2103 		_d = &ring->va[i].tx.legacy;
2104 		pa = skb_frag_dma_map(dev, frag, 0, skb_frag_size(frag),
2105 				      DMA_TO_DEVICE);
2106 		if (unlikely(dma_mapping_error(dev, pa))) {
2107 			wil_err(wil, "Tx[%2d] failed to map fragment\n",
2108 				ring_index);
2109 			goto dma_error;
2110 		}
2111 		ring->ctx[i].mapped_as = wil_mapped_as_page;
2112 		wil->txrx_ops.tx_desc_map((union wil_tx_desc *)d,
2113 					   pa, len, ring_index);
2114 		/* no need to check return code -
2115 		 * if it succeeded for 1-st descriptor,
2116 		 * it will succeed here too
2117 		 */
2118 		wil_tx_desc_offload_setup(d, skb);
2119 	}
2120 	/* for the last seg only */
2121 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_CMD_EOP_POS);
2122 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_CMD_MARK_WB_POS);
2123 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_CMD_DMA_IT_POS);
2124 	*_d = *d;
2125 	wil_dbg_txrx(wil, "Tx[%2d] desc[%4d]\n", ring_index, i);
2126 	wil_hex_dump_txrx("TxD ", DUMP_PREFIX_NONE, 32, 4,
2127 			  (const void *)d, sizeof(*d), false);
2128 
2129 	/* hold reference to skb
2130 	 * to prevent skb release before accounting
2131 	 * in case of immediate "tx done"
2132 	 */
2133 	ring->ctx[i].skb = skb_get(skb);
2134 
2135 	/* performance monitoring */
2136 	used = wil_ring_used_tx(ring);
2137 	if (wil_val_in_range(wil->ring_idle_trsh,
2138 			     used, used + nr_frags + 1)) {
2139 		txdata->idle += get_cycles() - txdata->last_idle;
2140 		wil_dbg_txrx(wil,  "Ring[%2d] not idle %d -> %d\n",
2141 			     ring_index, used, used + nr_frags + 1);
2142 	}
2143 
2144 	/* Make sure to advance the head only after descriptor update is done.
2145 	 * This will prevent a race condition where the completion thread
2146 	 * will see the DU bit set from previous run and will handle the
2147 	 * skb before it was completed.
2148 	 */
2149 	wmb();
2150 
2151 	/* advance swhead */
2152 	wil_ring_advance_head(ring, nr_frags + 1);
2153 	wil_dbg_txrx(wil, "Tx[%2d] swhead %d -> %d\n", ring_index, swhead,
2154 		     ring->swhead);
2155 	trace_wil6210_tx(ring_index, swhead, skb->len, nr_frags);
2156 
2157 	/* make sure all writes to descriptors (shared memory) are done before
2158 	 * committing them to HW
2159 	 */
2160 	wmb();
2161 
2162 	if (wil->tx_latency)
2163 		*(ktime_t *)&skb->cb = ktime_get();
2164 	else
2165 		memset(skb->cb, 0, sizeof(ktime_t));
2166 
2167 	wil_w(wil, ring->hwtail, ring->swhead);
2168 
2169 	return 0;
2170  dma_error:
2171 	/* unmap what we have mapped */
2172 	nr_frags = f + 1; /* frags mapped + one for skb head */
2173 	for (f = 0; f < nr_frags; f++) {
2174 		struct wil_ctx *ctx;
2175 
2176 		i = (swhead + f) % ring->size;
2177 		ctx = &ring->ctx[i];
2178 		_d = &ring->va[i].tx.legacy;
2179 		*d = *_d;
2180 		_d->dma.status = TX_DMA_STATUS_DU;
2181 		wil->txrx_ops.tx_desc_unmap(dev,
2182 					    (union wil_tx_desc *)d,
2183 					    ctx);
2184 
2185 		memset(ctx, 0, sizeof(*ctx));
2186 	}
2187 
2188 	return -EINVAL;
2189 }
2190 
2191 static int wil_tx_ring(struct wil6210_priv *wil, struct wil6210_vif *vif,
2192 		       struct wil_ring *ring, struct sk_buff *skb)
2193 {
2194 	int ring_index = ring - wil->ring_tx;
2195 	struct wil_ring_tx_data *txdata = &wil->ring_tx_data[ring_index];
2196 	int rc;
2197 
2198 	spin_lock(&txdata->lock);
2199 
2200 	if (test_bit(wil_status_suspending, wil->status) ||
2201 	    test_bit(wil_status_suspended, wil->status) ||
2202 	    test_bit(wil_status_resuming, wil->status)) {
2203 		wil_dbg_txrx(wil,
2204 			     "suspend/resume in progress. drop packet\n");
2205 		spin_unlock(&txdata->lock);
2206 		return -EINVAL;
2207 	}
2208 
2209 	rc = (skb_is_gso(skb) ? wil->txrx_ops.tx_ring_tso : __wil_tx_ring)
2210 	     (wil, vif, ring, skb);
2211 
2212 	spin_unlock(&txdata->lock);
2213 
2214 	return rc;
2215 }
2216 
2217 /**
2218  * Check status of tx vrings and stop/wake net queues if needed
2219  * It will start/stop net queues of a specific VIF net_device.
2220  *
2221  * This function does one of two checks:
2222  * In case check_stop is true, will check if net queues need to be stopped. If
2223  * the conditions for stopping are met, netif_tx_stop_all_queues() is called.
2224  * In case check_stop is false, will check if net queues need to be waked. If
2225  * the conditions for waking are met, netif_tx_wake_all_queues() is called.
2226  * vring is the vring which is currently being modified by either adding
2227  * descriptors (tx) into it or removing descriptors (tx complete) from it. Can
2228  * be null when irrelevant (e.g. connect/disconnect events).
2229  *
2230  * The implementation is to stop net queues if modified vring has low
2231  * descriptor availability. Wake if all vrings are not in low descriptor
2232  * availability and modified vring has high descriptor availability.
2233  */
2234 static inline void __wil_update_net_queues(struct wil6210_priv *wil,
2235 					   struct wil6210_vif *vif,
2236 					   struct wil_ring *ring,
2237 					   bool check_stop)
2238 {
2239 	int i;
2240 	int min_ring_id = wil_get_min_tx_ring_id(wil);
2241 
2242 	if (unlikely(!vif))
2243 		return;
2244 
2245 	if (ring)
2246 		wil_dbg_txrx(wil, "vring %d, mid %d, check_stop=%d, stopped=%d",
2247 			     (int)(ring - wil->ring_tx), vif->mid, check_stop,
2248 			     vif->net_queue_stopped);
2249 	else
2250 		wil_dbg_txrx(wil, "check_stop=%d, mid=%d, stopped=%d",
2251 			     check_stop, vif->mid, vif->net_queue_stopped);
2252 
2253 	if (ring && drop_if_ring_full)
2254 		/* no need to stop/wake net queues */
2255 		return;
2256 
2257 	if (check_stop == vif->net_queue_stopped)
2258 		/* net queues already in desired state */
2259 		return;
2260 
2261 	if (check_stop) {
2262 		if (!ring || unlikely(wil_ring_avail_low(ring))) {
2263 			/* not enough room in the vring */
2264 			netif_tx_stop_all_queues(vif_to_ndev(vif));
2265 			vif->net_queue_stopped = true;
2266 			wil_dbg_txrx(wil, "netif_tx_stop called\n");
2267 		}
2268 		return;
2269 	}
2270 
2271 	/* Do not wake the queues in suspend flow */
2272 	if (test_bit(wil_status_suspending, wil->status) ||
2273 	    test_bit(wil_status_suspended, wil->status))
2274 		return;
2275 
2276 	/* check wake */
2277 	for (i = min_ring_id; i < WIL6210_MAX_TX_RINGS; i++) {
2278 		struct wil_ring *cur_ring = &wil->ring_tx[i];
2279 		struct wil_ring_tx_data  *txdata = &wil->ring_tx_data[i];
2280 
2281 		if (txdata->mid != vif->mid || !cur_ring->va ||
2282 		    !txdata->enabled || cur_ring == ring)
2283 			continue;
2284 
2285 		if (wil_ring_avail_low(cur_ring)) {
2286 			wil_dbg_txrx(wil, "ring %d full, can't wake\n",
2287 				     (int)(cur_ring - wil->ring_tx));
2288 			return;
2289 		}
2290 	}
2291 
2292 	if (!ring || wil_ring_avail_high(ring)) {
2293 		/* enough room in the ring */
2294 		wil_dbg_txrx(wil, "calling netif_tx_wake\n");
2295 		netif_tx_wake_all_queues(vif_to_ndev(vif));
2296 		vif->net_queue_stopped = false;
2297 	}
2298 }
2299 
2300 void wil_update_net_queues(struct wil6210_priv *wil, struct wil6210_vif *vif,
2301 			   struct wil_ring *ring, bool check_stop)
2302 {
2303 	spin_lock(&wil->net_queue_lock);
2304 	__wil_update_net_queues(wil, vif, ring, check_stop);
2305 	spin_unlock(&wil->net_queue_lock);
2306 }
2307 
2308 void wil_update_net_queues_bh(struct wil6210_priv *wil, struct wil6210_vif *vif,
2309 			      struct wil_ring *ring, bool check_stop)
2310 {
2311 	spin_lock_bh(&wil->net_queue_lock);
2312 	__wil_update_net_queues(wil, vif, ring, check_stop);
2313 	spin_unlock_bh(&wil->net_queue_lock);
2314 }
2315 
2316 netdev_tx_t wil_start_xmit(struct sk_buff *skb, struct net_device *ndev)
2317 {
2318 	struct wil6210_vif *vif = ndev_to_vif(ndev);
2319 	struct wil6210_priv *wil = vif_to_wil(vif);
2320 	const u8 *da = wil_skb_get_da(skb);
2321 	bool bcast = is_multicast_ether_addr(da);
2322 	struct wil_ring *ring;
2323 	static bool pr_once_fw;
2324 	int rc;
2325 
2326 	wil_dbg_txrx(wil, "start_xmit\n");
2327 	if (unlikely(!test_bit(wil_status_fwready, wil->status))) {
2328 		if (!pr_once_fw) {
2329 			wil_err(wil, "FW not ready\n");
2330 			pr_once_fw = true;
2331 		}
2332 		goto drop;
2333 	}
2334 	if (unlikely(!test_bit(wil_vif_fwconnected, vif->status))) {
2335 		wil_dbg_ratelimited(wil,
2336 				    "VIF not connected, packet dropped\n");
2337 		goto drop;
2338 	}
2339 	if (unlikely(vif->wdev.iftype == NL80211_IFTYPE_MONITOR)) {
2340 		wil_err(wil, "Xmit in monitor mode not supported\n");
2341 		goto drop;
2342 	}
2343 	pr_once_fw = false;
2344 
2345 	/* find vring */
2346 	if (vif->wdev.iftype == NL80211_IFTYPE_STATION && !vif->pbss) {
2347 		/* in STA mode (ESS), all to same VRING (to AP) */
2348 		ring = wil_find_tx_ring_sta(wil, vif, skb);
2349 	} else if (bcast) {
2350 		if (vif->pbss)
2351 			/* in pbss, no bcast VRING - duplicate skb in
2352 			 * all stations VRINGs
2353 			 */
2354 			ring = wil_find_tx_bcast_2(wil, vif, skb);
2355 		else if (vif->wdev.iftype == NL80211_IFTYPE_AP)
2356 			/* AP has a dedicated bcast VRING */
2357 			ring = wil_find_tx_bcast_1(wil, vif, skb);
2358 		else
2359 			/* unexpected combination, fallback to duplicating
2360 			 * the skb in all stations VRINGs
2361 			 */
2362 			ring = wil_find_tx_bcast_2(wil, vif, skb);
2363 	} else {
2364 		/* unicast, find specific VRING by dest. address */
2365 		ring = wil_find_tx_ucast(wil, vif, skb);
2366 	}
2367 	if (unlikely(!ring)) {
2368 		wil_dbg_txrx(wil, "No Tx RING found for %pM\n", da);
2369 		goto drop;
2370 	}
2371 	/* set up vring entry */
2372 	rc = wil_tx_ring(wil, vif, ring, skb);
2373 
2374 	switch (rc) {
2375 	case 0:
2376 		/* shall we stop net queues? */
2377 		wil_update_net_queues_bh(wil, vif, ring, true);
2378 		/* statistics will be updated on the tx_complete */
2379 		dev_kfree_skb_any(skb);
2380 		return NETDEV_TX_OK;
2381 	case -ENOMEM:
2382 		if (drop_if_ring_full)
2383 			goto drop;
2384 		return NETDEV_TX_BUSY;
2385 	default:
2386 		break; /* goto drop; */
2387 	}
2388  drop:
2389 	ndev->stats.tx_dropped++;
2390 	dev_kfree_skb_any(skb);
2391 
2392 	return NET_XMIT_DROP;
2393 }
2394 
2395 void wil_tx_latency_calc(struct wil6210_priv *wil, struct sk_buff *skb,
2396 			 struct wil_sta_info *sta)
2397 {
2398 	int skb_time_us;
2399 	int bin;
2400 
2401 	if (!wil->tx_latency)
2402 		return;
2403 
2404 	if (ktime_to_ms(*(ktime_t *)&skb->cb) == 0)
2405 		return;
2406 
2407 	skb_time_us = ktime_us_delta(ktime_get(), *(ktime_t *)&skb->cb);
2408 	bin = skb_time_us / wil->tx_latency_res;
2409 	bin = min_t(int, bin, WIL_NUM_LATENCY_BINS - 1);
2410 
2411 	wil_dbg_txrx(wil, "skb time %dus => bin %d\n", skb_time_us, bin);
2412 	sta->tx_latency_bins[bin]++;
2413 	sta->stats.tx_latency_total_us += skb_time_us;
2414 	if (skb_time_us < sta->stats.tx_latency_min_us)
2415 		sta->stats.tx_latency_min_us = skb_time_us;
2416 	if (skb_time_us > sta->stats.tx_latency_max_us)
2417 		sta->stats.tx_latency_max_us = skb_time_us;
2418 }
2419 
2420 /**
2421  * Clean up transmitted skb's from the Tx VRING
2422  *
2423  * Return number of descriptors cleared
2424  *
2425  * Safe to call from IRQ
2426  */
2427 int wil_tx_complete(struct wil6210_vif *vif, int ringid)
2428 {
2429 	struct wil6210_priv *wil = vif_to_wil(vif);
2430 	struct net_device *ndev = vif_to_ndev(vif);
2431 	struct device *dev = wil_to_dev(wil);
2432 	struct wil_ring *vring = &wil->ring_tx[ringid];
2433 	struct wil_ring_tx_data *txdata = &wil->ring_tx_data[ringid];
2434 	int done = 0;
2435 	int cid = wil->ring2cid_tid[ringid][0];
2436 	struct wil_net_stats *stats = NULL;
2437 	volatile struct vring_tx_desc *_d;
2438 	int used_before_complete;
2439 	int used_new;
2440 
2441 	if (unlikely(!vring->va)) {
2442 		wil_err(wil, "Tx irq[%d]: vring not initialized\n", ringid);
2443 		return 0;
2444 	}
2445 
2446 	if (unlikely(!txdata->enabled)) {
2447 		wil_info(wil, "Tx irq[%d]: vring disabled\n", ringid);
2448 		return 0;
2449 	}
2450 
2451 	wil_dbg_txrx(wil, "tx_complete: (%d)\n", ringid);
2452 
2453 	used_before_complete = wil_ring_used_tx(vring);
2454 
2455 	if (cid < wil->max_assoc_sta)
2456 		stats = &wil->sta[cid].stats;
2457 
2458 	while (!wil_ring_is_empty(vring)) {
2459 		int new_swtail;
2460 		struct wil_ctx *ctx = &vring->ctx[vring->swtail];
2461 		/**
2462 		 * For the fragmented skb, HW will set DU bit only for the
2463 		 * last fragment. look for it.
2464 		 * In TSO the first DU will include hdr desc
2465 		 */
2466 		int lf = (vring->swtail + ctx->nr_frags) % vring->size;
2467 		/* TODO: check we are not past head */
2468 
2469 		_d = &vring->va[lf].tx.legacy;
2470 		if (unlikely(!(_d->dma.status & TX_DMA_STATUS_DU)))
2471 			break;
2472 
2473 		new_swtail = (lf + 1) % vring->size;
2474 		while (vring->swtail != new_swtail) {
2475 			struct vring_tx_desc dd, *d = &dd;
2476 			u16 dmalen;
2477 			struct sk_buff *skb;
2478 
2479 			ctx = &vring->ctx[vring->swtail];
2480 			skb = ctx->skb;
2481 			_d = &vring->va[vring->swtail].tx.legacy;
2482 
2483 			*d = *_d;
2484 
2485 			dmalen = le16_to_cpu(d->dma.length);
2486 			trace_wil6210_tx_done(ringid, vring->swtail, dmalen,
2487 					      d->dma.error);
2488 			wil_dbg_txrx(wil,
2489 				     "TxC[%2d][%3d] : %d bytes, status 0x%02x err 0x%02x\n",
2490 				     ringid, vring->swtail, dmalen,
2491 				     d->dma.status, d->dma.error);
2492 			wil_hex_dump_txrx("TxCD ", DUMP_PREFIX_NONE, 32, 4,
2493 					  (const void *)d, sizeof(*d), false);
2494 
2495 			wil->txrx_ops.tx_desc_unmap(dev,
2496 						    (union wil_tx_desc *)d,
2497 						    ctx);
2498 
2499 			if (skb) {
2500 				if (likely(d->dma.error == 0)) {
2501 					ndev->stats.tx_packets++;
2502 					ndev->stats.tx_bytes += skb->len;
2503 					if (stats) {
2504 						stats->tx_packets++;
2505 						stats->tx_bytes += skb->len;
2506 
2507 						wil_tx_latency_calc(wil, skb,
2508 							&wil->sta[cid]);
2509 					}
2510 				} else {
2511 					ndev->stats.tx_errors++;
2512 					if (stats)
2513 						stats->tx_errors++;
2514 				}
2515 
2516 				if (skb->protocol == cpu_to_be16(ETH_P_PAE))
2517 					wil_tx_complete_handle_eapol(vif, skb);
2518 
2519 				wil_consume_skb(skb, d->dma.error == 0);
2520 			}
2521 			memset(ctx, 0, sizeof(*ctx));
2522 			/* Make sure the ctx is zeroed before updating the tail
2523 			 * to prevent a case where wil_tx_ring will see
2524 			 * this descriptor as used and handle it before ctx zero
2525 			 * is completed.
2526 			 */
2527 			wmb();
2528 			/* There is no need to touch HW descriptor:
2529 			 * - ststus bit TX_DMA_STATUS_DU is set by design,
2530 			 *   so hardware will not try to process this desc.,
2531 			 * - rest of descriptor will be initialized on Tx.
2532 			 */
2533 			vring->swtail = wil_ring_next_tail(vring);
2534 			done++;
2535 		}
2536 	}
2537 
2538 	/* performance monitoring */
2539 	used_new = wil_ring_used_tx(vring);
2540 	if (wil_val_in_range(wil->ring_idle_trsh,
2541 			     used_new, used_before_complete)) {
2542 		wil_dbg_txrx(wil, "Ring[%2d] idle %d -> %d\n",
2543 			     ringid, used_before_complete, used_new);
2544 		txdata->last_idle = get_cycles();
2545 	}
2546 
2547 	/* shall we wake net queues? */
2548 	if (done)
2549 		wil_update_net_queues(wil, vif, vring, false);
2550 
2551 	return done;
2552 }
2553 
2554 static inline int wil_tx_init(struct wil6210_priv *wil)
2555 {
2556 	return 0;
2557 }
2558 
2559 static inline void wil_tx_fini(struct wil6210_priv *wil) {}
2560 
2561 static void wil_get_reorder_params(struct wil6210_priv *wil,
2562 				   struct sk_buff *skb, int *tid, int *cid,
2563 				   int *mid, u16 *seq, int *mcast, int *retry)
2564 {
2565 	struct vring_rx_desc *d = wil_skb_rxdesc(skb);
2566 
2567 	*tid = wil_rxdesc_tid(d);
2568 	*cid = wil_skb_get_cid(skb);
2569 	*mid = wil_rxdesc_mid(d);
2570 	*seq = wil_rxdesc_seq(d);
2571 	*mcast = wil_rxdesc_mcast(d);
2572 	*retry = wil_rxdesc_retry(d);
2573 }
2574 
2575 void wil_init_txrx_ops_legacy_dma(struct wil6210_priv *wil)
2576 {
2577 	wil->txrx_ops.configure_interrupt_moderation =
2578 		wil_configure_interrupt_moderation;
2579 	/* TX ops */
2580 	wil->txrx_ops.tx_desc_map = wil_tx_desc_map;
2581 	wil->txrx_ops.tx_desc_unmap = wil_txdesc_unmap;
2582 	wil->txrx_ops.tx_ring_tso =  __wil_tx_vring_tso;
2583 	wil->txrx_ops.ring_init_tx = wil_vring_init_tx;
2584 	wil->txrx_ops.ring_fini_tx = wil_vring_free;
2585 	wil->txrx_ops.ring_init_bcast = wil_vring_init_bcast;
2586 	wil->txrx_ops.tx_init = wil_tx_init;
2587 	wil->txrx_ops.tx_fini = wil_tx_fini;
2588 	wil->txrx_ops.tx_ring_modify = wil_tx_vring_modify;
2589 	/* RX ops */
2590 	wil->txrx_ops.rx_init = wil_rx_init;
2591 	wil->txrx_ops.wmi_addba_rx_resp = wmi_addba_rx_resp;
2592 	wil->txrx_ops.get_reorder_params = wil_get_reorder_params;
2593 	wil->txrx_ops.get_netif_rx_params =
2594 		wil_get_netif_rx_params;
2595 	wil->txrx_ops.rx_crypto_check = wil_rx_crypto_check;
2596 	wil->txrx_ops.rx_error_check = wil_rx_error_check;
2597 	wil->txrx_ops.is_rx_idle = wil_is_rx_idle;
2598 	wil->txrx_ops.rx_fini = wil_rx_fini;
2599 }
2600