1 /*
2  * Copyright (c) 2012-2017 Qualcomm Atheros, Inc.
3  * Copyright (c) 2018, 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 
32 static bool rtap_include_phy_info;
33 module_param(rtap_include_phy_info, bool, 0444);
34 MODULE_PARM_DESC(rtap_include_phy_info,
35 		 " Include PHY info in the radiotap header, default - no");
36 
37 bool rx_align_2;
38 module_param(rx_align_2, bool, 0444);
39 MODULE_PARM_DESC(rx_align_2, " align Rx buffers on 4*n+2, default - no");
40 
41 bool rx_large_buf;
42 module_param(rx_large_buf, bool, 0444);
43 MODULE_PARM_DESC(rx_large_buf, " allocate 8KB RX buffers, default - no");
44 
45 static inline uint wil_rx_snaplen(void)
46 {
47 	return rx_align_2 ? 6 : 0;
48 }
49 
50 static inline int wil_vring_is_empty(struct vring *vring)
51 {
52 	return vring->swhead == vring->swtail;
53 }
54 
55 static inline u32 wil_vring_next_tail(struct vring *vring)
56 {
57 	return (vring->swtail + 1) % vring->size;
58 }
59 
60 static inline void wil_vring_advance_head(struct vring *vring, int n)
61 {
62 	vring->swhead = (vring->swhead + n) % vring->size;
63 }
64 
65 static inline int wil_vring_is_full(struct vring *vring)
66 {
67 	return wil_vring_next_tail(vring) == vring->swhead;
68 }
69 
70 /* Used space in Tx Vring */
71 static inline int wil_vring_used_tx(struct vring *vring)
72 {
73 	u32 swhead = vring->swhead;
74 	u32 swtail = vring->swtail;
75 	return (vring->size + swhead - swtail) % vring->size;
76 }
77 
78 /* Available space in Tx Vring */
79 static inline int wil_vring_avail_tx(struct vring *vring)
80 {
81 	return vring->size - wil_vring_used_tx(vring) - 1;
82 }
83 
84 /* wil_vring_wmark_low - low watermark for available descriptor space */
85 static inline int wil_vring_wmark_low(struct vring *vring)
86 {
87 	return vring->size/8;
88 }
89 
90 /* wil_vring_wmark_high - high watermark for available descriptor space */
91 static inline int wil_vring_wmark_high(struct vring *vring)
92 {
93 	return vring->size/4;
94 }
95 
96 /* returns true if num avail descriptors is lower than wmark_low */
97 static inline int wil_vring_avail_low(struct vring *vring)
98 {
99 	return wil_vring_avail_tx(vring) < wil_vring_wmark_low(vring);
100 }
101 
102 /* returns true if num avail descriptors is higher than wmark_high */
103 static inline int wil_vring_avail_high(struct vring *vring)
104 {
105 	return wil_vring_avail_tx(vring) > wil_vring_wmark_high(vring);
106 }
107 
108 /* returns true when all tx vrings are empty */
109 bool wil_is_tx_idle(struct wil6210_priv *wil)
110 {
111 	int i;
112 	unsigned long data_comp_to;
113 
114 	for (i = 0; i < WIL6210_MAX_TX_RINGS; i++) {
115 		struct vring *vring = &wil->vring_tx[i];
116 		int vring_index = vring - wil->vring_tx;
117 		struct vring_tx_data *txdata = &wil->vring_tx_data[vring_index];
118 
119 		spin_lock(&txdata->lock);
120 
121 		if (!vring->va || !txdata->enabled) {
122 			spin_unlock(&txdata->lock);
123 			continue;
124 		}
125 
126 		data_comp_to = jiffies + msecs_to_jiffies(
127 					WIL_DATA_COMPLETION_TO_MS);
128 		if (test_bit(wil_status_napi_en, wil->status)) {
129 			while (!wil_vring_is_empty(vring)) {
130 				if (time_after(jiffies, data_comp_to)) {
131 					wil_dbg_pm(wil,
132 						   "TO waiting for idle tx\n");
133 					spin_unlock(&txdata->lock);
134 					return false;
135 				}
136 				wil_dbg_ratelimited(wil,
137 						    "tx vring is not empty -> NAPI\n");
138 				spin_unlock(&txdata->lock);
139 				napi_synchronize(&wil->napi_tx);
140 				msleep(20);
141 				spin_lock(&txdata->lock);
142 				if (!vring->va || !txdata->enabled)
143 					break;
144 			}
145 		}
146 
147 		spin_unlock(&txdata->lock);
148 	}
149 
150 	return true;
151 }
152 
153 /* wil_val_in_range - check if value in [min,max) */
154 static inline bool wil_val_in_range(int val, int min, int max)
155 {
156 	return val >= min && val < max;
157 }
158 
159 static int wil_vring_alloc(struct wil6210_priv *wil, struct vring *vring)
160 {
161 	struct device *dev = wil_to_dev(wil);
162 	size_t sz = vring->size * sizeof(vring->va[0]);
163 	uint i;
164 
165 	wil_dbg_misc(wil, "vring_alloc:\n");
166 
167 	BUILD_BUG_ON(sizeof(vring->va[0]) != 32);
168 
169 	vring->swhead = 0;
170 	vring->swtail = 0;
171 	vring->ctx = kcalloc(vring->size, sizeof(vring->ctx[0]), GFP_KERNEL);
172 	if (!vring->ctx) {
173 		vring->va = NULL;
174 		return -ENOMEM;
175 	}
176 
177 	/* vring->va should be aligned on its size rounded up to power of 2
178 	 * This is granted by the dma_alloc_coherent.
179 	 *
180 	 * HW has limitation that all vrings addresses must share the same
181 	 * upper 16 msb bits part of 48 bits address. To workaround that,
182 	 * if we are using more than 32 bit addresses switch to 32 bit
183 	 * allocation before allocating vring memory.
184 	 *
185 	 * There's no check for the return value of dma_set_mask_and_coherent,
186 	 * since we assume if we were able to set the mask during
187 	 * initialization in this system it will not fail if we set it again
188 	 */
189 	if (wil->dma_addr_size > 32)
190 		dma_set_mask_and_coherent(dev, DMA_BIT_MASK(32));
191 
192 	vring->va = dma_alloc_coherent(dev, sz, &vring->pa, GFP_KERNEL);
193 	if (!vring->va) {
194 		kfree(vring->ctx);
195 		vring->ctx = NULL;
196 		return -ENOMEM;
197 	}
198 
199 	if (wil->dma_addr_size > 32)
200 		dma_set_mask_and_coherent(dev,
201 					  DMA_BIT_MASK(wil->dma_addr_size));
202 
203 	/* initially, all descriptors are SW owned
204 	 * For Tx and Rx, ownership bit is at the same location, thus
205 	 * we can use any
206 	 */
207 	for (i = 0; i < vring->size; i++) {
208 		volatile struct vring_tx_desc *_d = &vring->va[i].tx;
209 
210 		_d->dma.status = TX_DMA_STATUS_DU;
211 	}
212 
213 	wil_dbg_misc(wil, "vring[%d] 0x%p:%pad 0x%p\n", vring->size,
214 		     vring->va, &vring->pa, vring->ctx);
215 
216 	return 0;
217 }
218 
219 static void wil_txdesc_unmap(struct device *dev, struct vring_tx_desc *d,
220 			     struct wil_ctx *ctx)
221 {
222 	dma_addr_t pa = wil_desc_addr(&d->dma.addr);
223 	u16 dmalen = le16_to_cpu(d->dma.length);
224 
225 	switch (ctx->mapped_as) {
226 	case wil_mapped_as_single:
227 		dma_unmap_single(dev, pa, dmalen, DMA_TO_DEVICE);
228 		break;
229 	case wil_mapped_as_page:
230 		dma_unmap_page(dev, pa, dmalen, DMA_TO_DEVICE);
231 		break;
232 	default:
233 		break;
234 	}
235 }
236 
237 static void wil_vring_free(struct wil6210_priv *wil, struct vring *vring,
238 			   int tx)
239 {
240 	struct device *dev = wil_to_dev(wil);
241 	size_t sz = vring->size * sizeof(vring->va[0]);
242 
243 	lockdep_assert_held(&wil->mutex);
244 	if (tx) {
245 		int vring_index = vring - wil->vring_tx;
246 
247 		wil_dbg_misc(wil, "free Tx vring %d [%d] 0x%p:%pad 0x%p\n",
248 			     vring_index, vring->size, vring->va,
249 			     &vring->pa, vring->ctx);
250 	} else {
251 		wil_dbg_misc(wil, "free Rx vring [%d] 0x%p:%pad 0x%p\n",
252 			     vring->size, vring->va,
253 			     &vring->pa, vring->ctx);
254 	}
255 
256 	while (!wil_vring_is_empty(vring)) {
257 		dma_addr_t pa;
258 		u16 dmalen;
259 		struct wil_ctx *ctx;
260 
261 		if (tx) {
262 			struct vring_tx_desc dd, *d = &dd;
263 			volatile struct vring_tx_desc *_d =
264 					&vring->va[vring->swtail].tx;
265 
266 			ctx = &vring->ctx[vring->swtail];
267 			if (!ctx) {
268 				wil_dbg_txrx(wil,
269 					     "ctx(%d) was already completed\n",
270 					     vring->swtail);
271 				vring->swtail = wil_vring_next_tail(vring);
272 				continue;
273 			}
274 			*d = *_d;
275 			wil_txdesc_unmap(dev, d, ctx);
276 			if (ctx->skb)
277 				dev_kfree_skb_any(ctx->skb);
278 			vring->swtail = wil_vring_next_tail(vring);
279 		} else { /* rx */
280 			struct vring_rx_desc dd, *d = &dd;
281 			volatile struct vring_rx_desc *_d =
282 					&vring->va[vring->swhead].rx;
283 
284 			ctx = &vring->ctx[vring->swhead];
285 			*d = *_d;
286 			pa = wil_desc_addr(&d->dma.addr);
287 			dmalen = le16_to_cpu(d->dma.length);
288 			dma_unmap_single(dev, pa, dmalen, DMA_FROM_DEVICE);
289 			kfree_skb(ctx->skb);
290 			wil_vring_advance_head(vring, 1);
291 		}
292 	}
293 	dma_free_coherent(dev, sz, (void *)vring->va, vring->pa);
294 	kfree(vring->ctx);
295 	vring->pa = 0;
296 	vring->va = NULL;
297 	vring->ctx = NULL;
298 }
299 
300 /**
301  * Allocate one skb for Rx VRING
302  *
303  * Safe to call from IRQ
304  */
305 static int wil_vring_alloc_skb(struct wil6210_priv *wil, struct vring *vring,
306 			       u32 i, int headroom)
307 {
308 	struct device *dev = wil_to_dev(wil);
309 	unsigned int sz = wil->rx_buf_len + ETH_HLEN + wil_rx_snaplen();
310 	struct vring_rx_desc dd, *d = &dd;
311 	volatile struct vring_rx_desc *_d = &vring->va[i].rx;
312 	dma_addr_t pa;
313 	struct sk_buff *skb = dev_alloc_skb(sz + headroom);
314 
315 	if (unlikely(!skb))
316 		return -ENOMEM;
317 
318 	skb_reserve(skb, headroom);
319 	skb_put(skb, sz);
320 
321 	pa = dma_map_single(dev, skb->data, skb->len, DMA_FROM_DEVICE);
322 	if (unlikely(dma_mapping_error(dev, pa))) {
323 		kfree_skb(skb);
324 		return -ENOMEM;
325 	}
326 
327 	d->dma.d0 = RX_DMA_D0_CMD_DMA_RT | RX_DMA_D0_CMD_DMA_IT;
328 	wil_desc_addr_set(&d->dma.addr, pa);
329 	/* ip_length don't care */
330 	/* b11 don't care */
331 	/* error don't care */
332 	d->dma.status = 0; /* BIT(0) should be 0 for HW_OWNED */
333 	d->dma.length = cpu_to_le16(sz);
334 	*_d = *d;
335 	vring->ctx[i].skb = skb;
336 
337 	return 0;
338 }
339 
340 /**
341  * Adds radiotap header
342  *
343  * Any error indicated as "Bad FCS"
344  *
345  * Vendor data for 04:ce:14-1 (Wilocity-1) consists of:
346  *  - Rx descriptor: 32 bytes
347  *  - Phy info
348  */
349 static void wil_rx_add_radiotap_header(struct wil6210_priv *wil,
350 				       struct sk_buff *skb)
351 {
352 	struct wil6210_rtap {
353 		struct ieee80211_radiotap_header rthdr;
354 		/* fields should be in the order of bits in rthdr.it_present */
355 		/* flags */
356 		u8 flags;
357 		/* channel */
358 		__le16 chnl_freq __aligned(2);
359 		__le16 chnl_flags;
360 		/* MCS */
361 		u8 mcs_present;
362 		u8 mcs_flags;
363 		u8 mcs_index;
364 	} __packed;
365 	struct wil6210_rtap_vendor {
366 		struct wil6210_rtap rtap;
367 		/* vendor */
368 		u8 vendor_oui[3] __aligned(2);
369 		u8 vendor_ns;
370 		__le16 vendor_skip;
371 		u8 vendor_data[0];
372 	} __packed;
373 	struct vring_rx_desc *d = wil_skb_rxdesc(skb);
374 	struct wil6210_rtap_vendor *rtap_vendor;
375 	int rtap_len = sizeof(struct wil6210_rtap);
376 	int phy_length = 0; /* phy info header size, bytes */
377 	static char phy_data[128];
378 	struct ieee80211_channel *ch = wil->monitor_chandef.chan;
379 
380 	if (rtap_include_phy_info) {
381 		rtap_len = sizeof(*rtap_vendor) + sizeof(*d);
382 		/* calculate additional length */
383 		if (d->dma.status & RX_DMA_STATUS_PHY_INFO) {
384 			/**
385 			 * PHY info starts from 8-byte boundary
386 			 * there are 8-byte lines, last line may be partially
387 			 * written (HW bug), thus FW configures for last line
388 			 * to be excessive. Driver skips this last line.
389 			 */
390 			int len = min_t(int, 8 + sizeof(phy_data),
391 					wil_rxdesc_phy_length(d));
392 
393 			if (len > 8) {
394 				void *p = skb_tail_pointer(skb);
395 				void *pa = PTR_ALIGN(p, 8);
396 
397 				if (skb_tailroom(skb) >= len + (pa - p)) {
398 					phy_length = len - 8;
399 					memcpy(phy_data, pa, phy_length);
400 				}
401 			}
402 		}
403 		rtap_len += phy_length;
404 	}
405 
406 	if (skb_headroom(skb) < rtap_len &&
407 	    pskb_expand_head(skb, rtap_len, 0, GFP_ATOMIC)) {
408 		wil_err(wil, "Unable to expand headroom to %d\n", rtap_len);
409 		return;
410 	}
411 
412 	rtap_vendor = skb_push(skb, rtap_len);
413 	memset(rtap_vendor, 0, rtap_len);
414 
415 	rtap_vendor->rtap.rthdr.it_version = PKTHDR_RADIOTAP_VERSION;
416 	rtap_vendor->rtap.rthdr.it_len = cpu_to_le16(rtap_len);
417 	rtap_vendor->rtap.rthdr.it_present = cpu_to_le32(
418 			(1 << IEEE80211_RADIOTAP_FLAGS) |
419 			(1 << IEEE80211_RADIOTAP_CHANNEL) |
420 			(1 << IEEE80211_RADIOTAP_MCS));
421 	if (d->dma.status & RX_DMA_STATUS_ERROR)
422 		rtap_vendor->rtap.flags |= IEEE80211_RADIOTAP_F_BADFCS;
423 
424 	rtap_vendor->rtap.chnl_freq = cpu_to_le16(ch ? ch->center_freq : 58320);
425 	rtap_vendor->rtap.chnl_flags = cpu_to_le16(0);
426 
427 	rtap_vendor->rtap.mcs_present = IEEE80211_RADIOTAP_MCS_HAVE_MCS;
428 	rtap_vendor->rtap.mcs_flags = 0;
429 	rtap_vendor->rtap.mcs_index = wil_rxdesc_mcs(d);
430 
431 	if (rtap_include_phy_info) {
432 		rtap_vendor->rtap.rthdr.it_present |= cpu_to_le32(1 <<
433 				IEEE80211_RADIOTAP_VENDOR_NAMESPACE);
434 		/* OUI for Wilocity 04:ce:14 */
435 		rtap_vendor->vendor_oui[0] = 0x04;
436 		rtap_vendor->vendor_oui[1] = 0xce;
437 		rtap_vendor->vendor_oui[2] = 0x14;
438 		rtap_vendor->vendor_ns = 1;
439 		/* Rx descriptor + PHY data  */
440 		rtap_vendor->vendor_skip = cpu_to_le16(sizeof(*d) +
441 						       phy_length);
442 		memcpy(rtap_vendor->vendor_data, (void *)d, sizeof(*d));
443 		memcpy(rtap_vendor->vendor_data + sizeof(*d), phy_data,
444 		       phy_length);
445 	}
446 }
447 
448 /* similar to ieee80211_ version, but FC contain only 1-st byte */
449 static inline int wil_is_back_req(u8 fc)
450 {
451 	return (fc & (IEEE80211_FCTL_FTYPE | IEEE80211_FCTL_STYPE)) ==
452 	       (IEEE80211_FTYPE_CTL | IEEE80211_STYPE_BACK_REQ);
453 }
454 
455 bool wil_is_rx_idle(struct wil6210_priv *wil)
456 {
457 	struct vring_rx_desc *_d;
458 	struct vring *vring = &wil->vring_rx;
459 
460 	_d = (struct vring_rx_desc *)&vring->va[vring->swhead].rx;
461 	if (_d->dma.status & RX_DMA_STATUS_DU)
462 		return false;
463 
464 	return true;
465 }
466 
467 /**
468  * reap 1 frame from @swhead
469  *
470  * Rx descriptor copied to skb->cb
471  *
472  * Safe to call from IRQ
473  */
474 static struct sk_buff *wil_vring_reap_rx(struct wil6210_priv *wil,
475 					 struct vring *vring)
476 {
477 	struct device *dev = wil_to_dev(wil);
478 	struct wil6210_vif *vif;
479 	struct net_device *ndev;
480 	volatile struct vring_rx_desc *_d;
481 	struct vring_rx_desc *d;
482 	struct sk_buff *skb;
483 	dma_addr_t pa;
484 	unsigned int snaplen = wil_rx_snaplen();
485 	unsigned int sz = wil->rx_buf_len + ETH_HLEN + snaplen;
486 	u16 dmalen;
487 	u8 ftype;
488 	int cid, mid;
489 	int i;
490 	struct wil_net_stats *stats;
491 
492 	BUILD_BUG_ON(sizeof(struct vring_rx_desc) > sizeof(skb->cb));
493 
494 again:
495 	if (unlikely(wil_vring_is_empty(vring)))
496 		return NULL;
497 
498 	i = (int)vring->swhead;
499 	_d = &vring->va[i].rx;
500 	if (unlikely(!(_d->dma.status & RX_DMA_STATUS_DU))) {
501 		/* it is not error, we just reached end of Rx done area */
502 		return NULL;
503 	}
504 
505 	skb = vring->ctx[i].skb;
506 	vring->ctx[i].skb = NULL;
507 	wil_vring_advance_head(vring, 1);
508 	if (!skb) {
509 		wil_err(wil, "No Rx skb at [%d]\n", i);
510 		goto again;
511 	}
512 	d = wil_skb_rxdesc(skb);
513 	*d = *_d;
514 	pa = wil_desc_addr(&d->dma.addr);
515 
516 	dma_unmap_single(dev, pa, sz, DMA_FROM_DEVICE);
517 	dmalen = le16_to_cpu(d->dma.length);
518 
519 	trace_wil6210_rx(i, d);
520 	wil_dbg_txrx(wil, "Rx[%3d] : %d bytes\n", i, dmalen);
521 	wil_hex_dump_txrx("RxD ", DUMP_PREFIX_NONE, 32, 4,
522 			  (const void *)d, sizeof(*d), false);
523 
524 	cid = wil_rxdesc_cid(d);
525 	mid = wil_rxdesc_mid(d);
526 	vif = wil->vifs[mid];
527 
528 	if (unlikely(!vif)) {
529 		wil_dbg_txrx(wil, "skipped RX descriptor with invalid mid %d",
530 			     mid);
531 		kfree_skb(skb);
532 		goto again;
533 	}
534 	ndev = vif_to_ndev(vif);
535 	stats = &wil->sta[cid].stats;
536 
537 	if (unlikely(dmalen > sz)) {
538 		wil_err(wil, "Rx size too large: %d bytes!\n", dmalen);
539 		stats->rx_large_frame++;
540 		kfree_skb(skb);
541 		goto again;
542 	}
543 	skb_trim(skb, dmalen);
544 
545 	prefetch(skb->data);
546 
547 	wil_hex_dump_txrx("Rx ", DUMP_PREFIX_OFFSET, 16, 1,
548 			  skb->data, skb_headlen(skb), false);
549 
550 	stats->last_mcs_rx = wil_rxdesc_mcs(d);
551 	if (stats->last_mcs_rx < ARRAY_SIZE(stats->rx_per_mcs))
552 		stats->rx_per_mcs[stats->last_mcs_rx]++;
553 
554 	/* use radiotap header only if required */
555 	if (ndev->type == ARPHRD_IEEE80211_RADIOTAP)
556 		wil_rx_add_radiotap_header(wil, skb);
557 
558 	/* no extra checks if in sniffer mode */
559 	if (ndev->type != ARPHRD_ETHER)
560 		return skb;
561 	/* Non-data frames may be delivered through Rx DMA channel (ex: BAR)
562 	 * Driver should recognize it by frame type, that is found
563 	 * in Rx descriptor. If type is not data, it is 802.11 frame as is
564 	 */
565 	ftype = wil_rxdesc_ftype(d) << 2;
566 	if (unlikely(ftype != IEEE80211_FTYPE_DATA)) {
567 		u8 fc1 = wil_rxdesc_fc1(d);
568 		int tid = wil_rxdesc_tid(d);
569 		u16 seq = wil_rxdesc_seq(d);
570 
571 		wil_dbg_txrx(wil,
572 			     "Non-data frame FC[7:0] 0x%02x MID %d CID %d TID %d Seq 0x%03x\n",
573 			     fc1, mid, cid, tid, seq);
574 		stats->rx_non_data_frame++;
575 		if (wil_is_back_req(fc1)) {
576 			wil_dbg_txrx(wil,
577 				     "BAR: MID %d CID %d TID %d Seq 0x%03x\n",
578 				     mid, cid, tid, seq);
579 			wil_rx_bar(wil, vif, cid, tid, seq);
580 		} else {
581 			/* print again all info. One can enable only this
582 			 * without overhead for printing every Rx frame
583 			 */
584 			wil_dbg_txrx(wil,
585 				     "Unhandled non-data frame FC[7:0] 0x%02x MID %d CID %d TID %d Seq 0x%03x\n",
586 				     fc1, mid, cid, tid, seq);
587 			wil_hex_dump_txrx("RxD ", DUMP_PREFIX_NONE, 32, 4,
588 					  (const void *)d, sizeof(*d), false);
589 			wil_hex_dump_txrx("Rx ", DUMP_PREFIX_OFFSET, 16, 1,
590 					  skb->data, skb_headlen(skb), false);
591 		}
592 		kfree_skb(skb);
593 		goto again;
594 	}
595 
596 	if (unlikely(skb->len < ETH_HLEN + snaplen)) {
597 		wil_err(wil, "Short frame, len = %d\n", skb->len);
598 		stats->rx_short_frame++;
599 		kfree_skb(skb);
600 		goto again;
601 	}
602 
603 	/* L4 IDENT is on when HW calculated checksum, check status
604 	 * and in case of error drop the packet
605 	 * higher stack layers will handle retransmission (if required)
606 	 */
607 	if (likely(d->dma.status & RX_DMA_STATUS_L4I)) {
608 		/* L4 protocol identified, csum calculated */
609 		if (likely((d->dma.error & RX_DMA_ERROR_L4_ERR) == 0))
610 			skb->ip_summed = CHECKSUM_UNNECESSARY;
611 		/* If HW reports bad checksum, let IP stack re-check it
612 		 * For example, HW don't understand Microsoft IP stack that
613 		 * mis-calculates TCP checksum - if it should be 0x0,
614 		 * it writes 0xffff in violation of RFC 1624
615 		 */
616 	}
617 
618 	if (snaplen) {
619 		/* Packet layout
620 		 * +-------+-------+---------+------------+------+
621 		 * | SA(6) | DA(6) | SNAP(6) | ETHTYPE(2) | DATA |
622 		 * +-------+-------+---------+------------+------+
623 		 * Need to remove SNAP, shifting SA and DA forward
624 		 */
625 		memmove(skb->data + snaplen, skb->data, 2 * ETH_ALEN);
626 		skb_pull(skb, snaplen);
627 	}
628 
629 	return skb;
630 }
631 
632 /**
633  * allocate and fill up to @count buffers in rx ring
634  * buffers posted at @swtail
635  * Note: we have a single RX queue for servicing all VIFs, but we
636  * allocate skbs with headroom according to main interface only. This
637  * means it will not work with monitor interface together with other VIFs.
638  * Currently we only support monitor interface on its own without other VIFs,
639  * and we will need to fix this code once we add support.
640  */
641 static int wil_rx_refill(struct wil6210_priv *wil, int count)
642 {
643 	struct net_device *ndev = wil->main_ndev;
644 	struct vring *v = &wil->vring_rx;
645 	u32 next_tail;
646 	int rc = 0;
647 	int headroom = ndev->type == ARPHRD_IEEE80211_RADIOTAP ?
648 			WIL6210_RTAP_SIZE : 0;
649 
650 	for (; next_tail = wil_vring_next_tail(v),
651 			(next_tail != v->swhead) && (count-- > 0);
652 			v->swtail = next_tail) {
653 		rc = wil_vring_alloc_skb(wil, v, v->swtail, headroom);
654 		if (unlikely(rc)) {
655 			wil_err(wil, "Error %d in wil_rx_refill[%d]\n",
656 				rc, v->swtail);
657 			break;
658 		}
659 	}
660 
661 	/* make sure all writes to descriptors (shared memory) are done before
662 	 * committing them to HW
663 	 */
664 	wmb();
665 
666 	wil_w(wil, v->hwtail, v->swtail);
667 
668 	return rc;
669 }
670 
671 /**
672  * reverse_memcmp - Compare two areas of memory, in reverse order
673  * @cs: One area of memory
674  * @ct: Another area of memory
675  * @count: The size of the area.
676  *
677  * Cut'n'paste from original memcmp (see lib/string.c)
678  * with minimal modifications
679  */
680 static int reverse_memcmp(const void *cs, const void *ct, size_t count)
681 {
682 	const unsigned char *su1, *su2;
683 	int res = 0;
684 
685 	for (su1 = cs + count - 1, su2 = ct + count - 1; count > 0;
686 	     --su1, --su2, count--) {
687 		res = *su1 - *su2;
688 		if (res)
689 			break;
690 	}
691 	return res;
692 }
693 
694 static int wil_rx_crypto_check(struct wil6210_priv *wil, struct sk_buff *skb)
695 {
696 	struct vring_rx_desc *d = wil_skb_rxdesc(skb);
697 	int cid = wil_rxdesc_cid(d);
698 	int tid = wil_rxdesc_tid(d);
699 	int key_id = wil_rxdesc_key_id(d);
700 	int mc = wil_rxdesc_mcast(d);
701 	struct wil_sta_info *s = &wil->sta[cid];
702 	struct wil_tid_crypto_rx *c = mc ? &s->group_crypto_rx :
703 				      &s->tid_crypto_rx[tid];
704 	struct wil_tid_crypto_rx_single *cc = &c->key_id[key_id];
705 	const u8 *pn = (u8 *)&d->mac.pn_15_0;
706 
707 	if (!cc->key_set) {
708 		wil_err_ratelimited(wil,
709 				    "Key missing. CID %d TID %d MCast %d KEY_ID %d\n",
710 				    cid, tid, mc, key_id);
711 		return -EINVAL;
712 	}
713 
714 	if (reverse_memcmp(pn, cc->pn, IEEE80211_GCMP_PN_LEN) <= 0) {
715 		wil_err_ratelimited(wil,
716 				    "Replay attack. CID %d TID %d MCast %d KEY_ID %d PN %6phN last %6phN\n",
717 				    cid, tid, mc, key_id, pn, cc->pn);
718 		return -EINVAL;
719 	}
720 	memcpy(cc->pn, pn, IEEE80211_GCMP_PN_LEN);
721 
722 	return 0;
723 }
724 
725 /*
726  * Pass Rx packet to the netif. Update statistics.
727  * Called in softirq context (NAPI poll).
728  */
729 void wil_netif_rx_any(struct sk_buff *skb, struct net_device *ndev)
730 {
731 	gro_result_t rc = GRO_NORMAL;
732 	struct wil6210_vif *vif = ndev_to_vif(ndev);
733 	struct wil6210_priv *wil = ndev_to_wil(ndev);
734 	struct wireless_dev *wdev = vif_to_wdev(vif);
735 	unsigned int len = skb->len;
736 	struct vring_rx_desc *d = wil_skb_rxdesc(skb);
737 	int cid = wil_rxdesc_cid(d); /* always 0..7, no need to check */
738 	int security = wil_rxdesc_security(d);
739 	struct ethhdr *eth = (void *)skb->data;
740 	/* here looking for DA, not A1, thus Rxdesc's 'mcast' indication
741 	 * is not suitable, need to look at data
742 	 */
743 	int mcast = is_multicast_ether_addr(eth->h_dest);
744 	struct wil_net_stats *stats = &wil->sta[cid].stats;
745 	struct sk_buff *xmit_skb = NULL;
746 	static const char * const gro_res_str[] = {
747 		[GRO_MERGED]		= "GRO_MERGED",
748 		[GRO_MERGED_FREE]	= "GRO_MERGED_FREE",
749 		[GRO_HELD]		= "GRO_HELD",
750 		[GRO_NORMAL]		= "GRO_NORMAL",
751 		[GRO_DROP]		= "GRO_DROP",
752 	};
753 
754 	if (ndev->features & NETIF_F_RXHASH)
755 		/* fake L4 to ensure it won't be re-calculated later
756 		 * set hash to any non-zero value to activate rps
757 		 * mechanism, core will be chosen according
758 		 * to user-level rps configuration.
759 		 */
760 		skb_set_hash(skb, 1, PKT_HASH_TYPE_L4);
761 
762 	skb_orphan(skb);
763 
764 	if (security && (wil_rx_crypto_check(wil, skb) != 0)) {
765 		rc = GRO_DROP;
766 		dev_kfree_skb(skb);
767 		stats->rx_replay++;
768 		goto stats;
769 	}
770 
771 	if (wdev->iftype == NL80211_IFTYPE_AP && !vif->ap_isolate) {
772 		if (mcast) {
773 			/* send multicast frames both to higher layers in
774 			 * local net stack and back to the wireless medium
775 			 */
776 			xmit_skb = skb_copy(skb, GFP_ATOMIC);
777 		} else {
778 			int xmit_cid = wil_find_cid(wil, vif->mid,
779 						    eth->h_dest);
780 
781 			if (xmit_cid >= 0) {
782 				/* The destination station is associated to
783 				 * this AP (in this VLAN), so send the frame
784 				 * directly to it and do not pass it to local
785 				 * net stack.
786 				 */
787 				xmit_skb = skb;
788 				skb = NULL;
789 			}
790 		}
791 	}
792 	if (xmit_skb) {
793 		/* Send to wireless media and increase priority by 256 to
794 		 * keep the received priority instead of reclassifying
795 		 * the frame (see cfg80211_classify8021d).
796 		 */
797 		xmit_skb->dev = ndev;
798 		xmit_skb->priority += 256;
799 		xmit_skb->protocol = htons(ETH_P_802_3);
800 		skb_reset_network_header(xmit_skb);
801 		skb_reset_mac_header(xmit_skb);
802 		wil_dbg_txrx(wil, "Rx -> Tx %d bytes\n", len);
803 		dev_queue_xmit(xmit_skb);
804 	}
805 
806 	if (skb) { /* deliver to local stack */
807 		skb->protocol = eth_type_trans(skb, ndev);
808 		skb->dev = ndev;
809 		rc = napi_gro_receive(&wil->napi_rx, skb);
810 		wil_dbg_txrx(wil, "Rx complete %d bytes => %s\n",
811 			     len, gro_res_str[rc]);
812 	}
813 stats:
814 	/* statistics. rc set to GRO_NORMAL for AP bridging */
815 	if (unlikely(rc == GRO_DROP)) {
816 		ndev->stats.rx_dropped++;
817 		stats->rx_dropped++;
818 		wil_dbg_txrx(wil, "Rx drop %d bytes\n", len);
819 	} else {
820 		ndev->stats.rx_packets++;
821 		stats->rx_packets++;
822 		ndev->stats.rx_bytes += len;
823 		stats->rx_bytes += len;
824 		if (mcast)
825 			ndev->stats.multicast++;
826 	}
827 }
828 
829 /**
830  * Proceed all completed skb's from Rx VRING
831  *
832  * Safe to call from NAPI poll, i.e. softirq with interrupts enabled
833  */
834 void wil_rx_handle(struct wil6210_priv *wil, int *quota)
835 {
836 	struct net_device *ndev = wil->main_ndev;
837 	struct wireless_dev *wdev = ndev->ieee80211_ptr;
838 	struct vring *v = &wil->vring_rx;
839 	struct sk_buff *skb;
840 
841 	if (unlikely(!v->va)) {
842 		wil_err(wil, "Rx IRQ while Rx not yet initialized\n");
843 		return;
844 	}
845 	wil_dbg_txrx(wil, "rx_handle\n");
846 	while ((*quota > 0) && (NULL != (skb = wil_vring_reap_rx(wil, v)))) {
847 		(*quota)--;
848 
849 		/* monitor is currently supported on main interface only */
850 		if (wdev->iftype == NL80211_IFTYPE_MONITOR) {
851 			skb->dev = ndev;
852 			skb_reset_mac_header(skb);
853 			skb->ip_summed = CHECKSUM_UNNECESSARY;
854 			skb->pkt_type = PACKET_OTHERHOST;
855 			skb->protocol = htons(ETH_P_802_2);
856 			wil_netif_rx_any(skb, ndev);
857 		} else {
858 			wil_rx_reorder(wil, skb);
859 		}
860 	}
861 	wil_rx_refill(wil, v->size);
862 }
863 
864 static void wil_rx_buf_len_init(struct wil6210_priv *wil)
865 {
866 	wil->rx_buf_len = rx_large_buf ?
867 		WIL_MAX_ETH_MTU : TXRX_BUF_LEN_DEFAULT - WIL_MAX_MPDU_OVERHEAD;
868 	if (mtu_max > wil->rx_buf_len) {
869 		/* do not allow RX buffers to be smaller than mtu_max, for
870 		 * backward compatibility (mtu_max parameter was also used
871 		 * to support receiving large packets)
872 		 */
873 		wil_info(wil, "Override RX buffer to mtu_max(%d)\n", mtu_max);
874 		wil->rx_buf_len = mtu_max;
875 	}
876 }
877 
878 int wil_rx_init(struct wil6210_priv *wil, u16 size)
879 {
880 	struct vring *vring = &wil->vring_rx;
881 	int rc;
882 
883 	wil_dbg_misc(wil, "rx_init\n");
884 
885 	if (vring->va) {
886 		wil_err(wil, "Rx ring already allocated\n");
887 		return -EINVAL;
888 	}
889 
890 	wil_rx_buf_len_init(wil);
891 
892 	vring->size = size;
893 	rc = wil_vring_alloc(wil, vring);
894 	if (rc)
895 		return rc;
896 
897 	rc = wmi_rx_chain_add(wil, vring);
898 	if (rc)
899 		goto err_free;
900 
901 	rc = wil_rx_refill(wil, vring->size);
902 	if (rc)
903 		goto err_free;
904 
905 	return 0;
906  err_free:
907 	wil_vring_free(wil, vring, 0);
908 
909 	return rc;
910 }
911 
912 void wil_rx_fini(struct wil6210_priv *wil)
913 {
914 	struct vring *vring = &wil->vring_rx;
915 
916 	wil_dbg_misc(wil, "rx_fini\n");
917 
918 	if (vring->va)
919 		wil_vring_free(wil, vring, 0);
920 }
921 
922 static inline void wil_tx_data_init(struct vring_tx_data *txdata)
923 {
924 	spin_lock_bh(&txdata->lock);
925 	txdata->dot1x_open = 0;
926 	txdata->enabled = 0;
927 	txdata->idle = 0;
928 	txdata->last_idle = 0;
929 	txdata->begin = 0;
930 	txdata->agg_wsize = 0;
931 	txdata->agg_timeout = 0;
932 	txdata->agg_amsdu = 0;
933 	txdata->addba_in_progress = false;
934 	txdata->mid = U8_MAX;
935 	spin_unlock_bh(&txdata->lock);
936 }
937 
938 int wil_vring_init_tx(struct wil6210_vif *vif, int id, int size,
939 		      int cid, int tid)
940 {
941 	struct wil6210_priv *wil = vif_to_wil(vif);
942 	int rc;
943 	struct wmi_vring_cfg_cmd cmd = {
944 		.action = cpu_to_le32(WMI_VRING_CMD_ADD),
945 		.vring_cfg = {
946 			.tx_sw_ring = {
947 				.max_mpdu_size =
948 					cpu_to_le16(wil_mtu2macbuf(mtu_max)),
949 				.ring_size = cpu_to_le16(size),
950 			},
951 			.ringid = id,
952 			.cidxtid = mk_cidxtid(cid, tid),
953 			.encap_trans_type = WMI_VRING_ENC_TYPE_802_3,
954 			.mac_ctrl = 0,
955 			.to_resolution = 0,
956 			.agg_max_wsize = 0,
957 			.schd_params = {
958 				.priority = cpu_to_le16(0),
959 				.timeslot_us = cpu_to_le16(0xfff),
960 			},
961 		},
962 	};
963 	struct {
964 		struct wmi_cmd_hdr wmi;
965 		struct wmi_vring_cfg_done_event cmd;
966 	} __packed reply;
967 	struct vring *vring = &wil->vring_tx[id];
968 	struct vring_tx_data *txdata = &wil->vring_tx_data[id];
969 
970 	wil_dbg_misc(wil, "vring_init_tx: max_mpdu_size %d\n",
971 		     cmd.vring_cfg.tx_sw_ring.max_mpdu_size);
972 	lockdep_assert_held(&wil->mutex);
973 
974 	if (vring->va) {
975 		wil_err(wil, "Tx ring [%d] already allocated\n", id);
976 		rc = -EINVAL;
977 		goto out;
978 	}
979 
980 	wil_tx_data_init(txdata);
981 	vring->size = size;
982 	rc = wil_vring_alloc(wil, vring);
983 	if (rc)
984 		goto out;
985 
986 	wil->vring2cid_tid[id][0] = cid;
987 	wil->vring2cid_tid[id][1] = tid;
988 
989 	cmd.vring_cfg.tx_sw_ring.ring_mem_base = cpu_to_le64(vring->pa);
990 
991 	if (!vif->privacy)
992 		txdata->dot1x_open = true;
993 	rc = wmi_call(wil, WMI_VRING_CFG_CMDID, vif->mid, &cmd, sizeof(cmd),
994 		      WMI_VRING_CFG_DONE_EVENTID, &reply, sizeof(reply), 100);
995 	if (rc)
996 		goto out_free;
997 
998 	if (reply.cmd.status != WMI_FW_STATUS_SUCCESS) {
999 		wil_err(wil, "Tx config failed, status 0x%02x\n",
1000 			reply.cmd.status);
1001 		rc = -EINVAL;
1002 		goto out_free;
1003 	}
1004 
1005 	spin_lock_bh(&txdata->lock);
1006 	vring->hwtail = le32_to_cpu(reply.cmd.tx_vring_tail_ptr);
1007 	txdata->mid = vif->mid;
1008 	txdata->enabled = 1;
1009 	spin_unlock_bh(&txdata->lock);
1010 
1011 	if (txdata->dot1x_open && (agg_wsize >= 0))
1012 		wil_addba_tx_request(wil, id, agg_wsize);
1013 
1014 	return 0;
1015  out_free:
1016 	spin_lock_bh(&txdata->lock);
1017 	txdata->dot1x_open = false;
1018 	txdata->enabled = 0;
1019 	spin_unlock_bh(&txdata->lock);
1020 	wil_vring_free(wil, vring, 1);
1021 	wil->vring2cid_tid[id][0] = WIL6210_MAX_CID;
1022 	wil->vring2cid_tid[id][1] = 0;
1023 
1024  out:
1025 
1026 	return rc;
1027 }
1028 
1029 int wil_vring_init_bcast(struct wil6210_vif *vif, int id, int size)
1030 {
1031 	struct wil6210_priv *wil = vif_to_wil(vif);
1032 	int rc;
1033 	struct wmi_bcast_vring_cfg_cmd cmd = {
1034 		.action = cpu_to_le32(WMI_VRING_CMD_ADD),
1035 		.vring_cfg = {
1036 			.tx_sw_ring = {
1037 				.max_mpdu_size =
1038 					cpu_to_le16(wil_mtu2macbuf(mtu_max)),
1039 				.ring_size = cpu_to_le16(size),
1040 			},
1041 			.ringid = id,
1042 			.encap_trans_type = WMI_VRING_ENC_TYPE_802_3,
1043 		},
1044 	};
1045 	struct {
1046 		struct wmi_cmd_hdr wmi;
1047 		struct wmi_vring_cfg_done_event cmd;
1048 	} __packed reply;
1049 	struct vring *vring = &wil->vring_tx[id];
1050 	struct vring_tx_data *txdata = &wil->vring_tx_data[id];
1051 
1052 	wil_dbg_misc(wil, "vring_init_bcast: max_mpdu_size %d\n",
1053 		     cmd.vring_cfg.tx_sw_ring.max_mpdu_size);
1054 	lockdep_assert_held(&wil->mutex);
1055 
1056 	if (vring->va) {
1057 		wil_err(wil, "Tx ring [%d] already allocated\n", id);
1058 		rc = -EINVAL;
1059 		goto out;
1060 	}
1061 
1062 	wil_tx_data_init(txdata);
1063 	vring->size = size;
1064 	rc = wil_vring_alloc(wil, vring);
1065 	if (rc)
1066 		goto out;
1067 
1068 	wil->vring2cid_tid[id][0] = WIL6210_MAX_CID; /* CID */
1069 	wil->vring2cid_tid[id][1] = 0; /* TID */
1070 
1071 	cmd.vring_cfg.tx_sw_ring.ring_mem_base = cpu_to_le64(vring->pa);
1072 
1073 	if (!vif->privacy)
1074 		txdata->dot1x_open = true;
1075 	rc = wmi_call(wil, WMI_BCAST_VRING_CFG_CMDID, vif->mid,
1076 		      &cmd, sizeof(cmd),
1077 		      WMI_VRING_CFG_DONE_EVENTID, &reply, sizeof(reply), 100);
1078 	if (rc)
1079 		goto out_free;
1080 
1081 	if (reply.cmd.status != WMI_FW_STATUS_SUCCESS) {
1082 		wil_err(wil, "Tx config failed, status 0x%02x\n",
1083 			reply.cmd.status);
1084 		rc = -EINVAL;
1085 		goto out_free;
1086 	}
1087 
1088 	spin_lock_bh(&txdata->lock);
1089 	vring->hwtail = le32_to_cpu(reply.cmd.tx_vring_tail_ptr);
1090 	txdata->mid = vif->mid;
1091 	txdata->enabled = 1;
1092 	spin_unlock_bh(&txdata->lock);
1093 
1094 	return 0;
1095  out_free:
1096 	spin_lock_bh(&txdata->lock);
1097 	txdata->enabled = 0;
1098 	txdata->dot1x_open = false;
1099 	spin_unlock_bh(&txdata->lock);
1100 	wil_vring_free(wil, vring, 1);
1101  out:
1102 
1103 	return rc;
1104 }
1105 
1106 void wil_vring_fini_tx(struct wil6210_priv *wil, int id)
1107 {
1108 	struct vring *vring = &wil->vring_tx[id];
1109 	struct vring_tx_data *txdata = &wil->vring_tx_data[id];
1110 
1111 	lockdep_assert_held(&wil->mutex);
1112 
1113 	if (!vring->va)
1114 		return;
1115 
1116 	wil_dbg_misc(wil, "vring_fini_tx: id=%d\n", id);
1117 
1118 	spin_lock_bh(&txdata->lock);
1119 	txdata->dot1x_open = false;
1120 	txdata->mid = U8_MAX;
1121 	txdata->enabled = 0; /* no Tx can be in progress or start anew */
1122 	spin_unlock_bh(&txdata->lock);
1123 	/* napi_synchronize waits for completion of the current NAPI but will
1124 	 * not prevent the next NAPI run.
1125 	 * Add a memory barrier to guarantee that txdata->enabled is zeroed
1126 	 * before napi_synchronize so that the next scheduled NAPI will not
1127 	 * handle this vring
1128 	 */
1129 	wmb();
1130 	/* make sure NAPI won't touch this vring */
1131 	if (test_bit(wil_status_napi_en, wil->status))
1132 		napi_synchronize(&wil->napi_tx);
1133 
1134 	wil_vring_free(wil, vring, 1);
1135 }
1136 
1137 static struct vring *wil_find_tx_ucast(struct wil6210_priv *wil,
1138 				       struct wil6210_vif *vif,
1139 				       struct sk_buff *skb)
1140 {
1141 	int i;
1142 	struct ethhdr *eth = (void *)skb->data;
1143 	int cid = wil_find_cid(wil, vif->mid, eth->h_dest);
1144 
1145 	if (cid < 0)
1146 		return NULL;
1147 
1148 	/* TODO: fix for multiple TID */
1149 	for (i = 0; i < ARRAY_SIZE(wil->vring2cid_tid); i++) {
1150 		if (!wil->vring_tx_data[i].dot1x_open &&
1151 		    (skb->protocol != cpu_to_be16(ETH_P_PAE)))
1152 			continue;
1153 		if (wil->vring2cid_tid[i][0] == cid) {
1154 			struct vring *v = &wil->vring_tx[i];
1155 			struct vring_tx_data *txdata = &wil->vring_tx_data[i];
1156 
1157 			wil_dbg_txrx(wil, "find_tx_ucast: (%pM) -> [%d]\n",
1158 				     eth->h_dest, i);
1159 			if (v->va && txdata->enabled) {
1160 				return v;
1161 			} else {
1162 				wil_dbg_txrx(wil,
1163 					     "find_tx_ucast: vring[%d] not valid\n",
1164 					     i);
1165 				return NULL;
1166 			}
1167 		}
1168 	}
1169 
1170 	return NULL;
1171 }
1172 
1173 static int wil_tx_vring(struct wil6210_priv *wil, struct wil6210_vif *vif,
1174 			struct vring *vring, struct sk_buff *skb);
1175 
1176 static struct vring *wil_find_tx_vring_sta(struct wil6210_priv *wil,
1177 					   struct wil6210_vif *vif,
1178 					   struct sk_buff *skb)
1179 {
1180 	struct vring *v;
1181 	int i;
1182 	u8 cid;
1183 	struct vring_tx_data *txdata;
1184 
1185 	/* In the STA mode, it is expected to have only 1 VRING
1186 	 * for the AP we connected to.
1187 	 * find 1-st vring eligible for this skb and use it.
1188 	 */
1189 	for (i = 0; i < WIL6210_MAX_TX_RINGS; i++) {
1190 		v = &wil->vring_tx[i];
1191 		txdata = &wil->vring_tx_data[i];
1192 		if (!v->va || !txdata->enabled || txdata->mid != vif->mid)
1193 			continue;
1194 
1195 		cid = wil->vring2cid_tid[i][0];
1196 		if (cid >= WIL6210_MAX_CID) /* skip BCAST */
1197 			continue;
1198 
1199 		if (!wil->vring_tx_data[i].dot1x_open &&
1200 		    (skb->protocol != cpu_to_be16(ETH_P_PAE)))
1201 			continue;
1202 
1203 		wil_dbg_txrx(wil, "Tx -> ring %d\n", i);
1204 
1205 		return v;
1206 	}
1207 
1208 	wil_dbg_txrx(wil, "Tx while no vrings active?\n");
1209 
1210 	return NULL;
1211 }
1212 
1213 /* Use one of 2 strategies:
1214  *
1215  * 1. New (real broadcast):
1216  *    use dedicated broadcast vring
1217  * 2. Old (pseudo-DMS):
1218  *    Find 1-st vring and return it;
1219  *    duplicate skb and send it to other active vrings;
1220  *    in all cases override dest address to unicast peer's address
1221  * Use old strategy when new is not supported yet:
1222  *  - for PBSS
1223  */
1224 static struct vring *wil_find_tx_bcast_1(struct wil6210_priv *wil,
1225 					 struct wil6210_vif *vif,
1226 					 struct sk_buff *skb)
1227 {
1228 	struct vring *v;
1229 	struct vring_tx_data *txdata;
1230 	int i = vif->bcast_vring;
1231 
1232 	if (i < 0)
1233 		return NULL;
1234 	v = &wil->vring_tx[i];
1235 	txdata = &wil->vring_tx_data[i];
1236 	if (!v->va || !txdata->enabled)
1237 		return NULL;
1238 	if (!wil->vring_tx_data[i].dot1x_open &&
1239 	    (skb->protocol != cpu_to_be16(ETH_P_PAE)))
1240 		return NULL;
1241 
1242 	return v;
1243 }
1244 
1245 static void wil_set_da_for_vring(struct wil6210_priv *wil,
1246 				 struct sk_buff *skb, int vring_index)
1247 {
1248 	struct ethhdr *eth = (void *)skb->data;
1249 	int cid = wil->vring2cid_tid[vring_index][0];
1250 
1251 	ether_addr_copy(eth->h_dest, wil->sta[cid].addr);
1252 }
1253 
1254 static struct vring *wil_find_tx_bcast_2(struct wil6210_priv *wil,
1255 					 struct wil6210_vif *vif,
1256 					 struct sk_buff *skb)
1257 {
1258 	struct vring *v, *v2;
1259 	struct sk_buff *skb2;
1260 	int i;
1261 	u8 cid;
1262 	struct ethhdr *eth = (void *)skb->data;
1263 	char *src = eth->h_source;
1264 	struct vring_tx_data *txdata, *txdata2;
1265 
1266 	/* find 1-st vring eligible for data */
1267 	for (i = 0; i < WIL6210_MAX_TX_RINGS; i++) {
1268 		v = &wil->vring_tx[i];
1269 		txdata = &wil->vring_tx_data[i];
1270 		if (!v->va || !txdata->enabled || txdata->mid != vif->mid)
1271 			continue;
1272 
1273 		cid = wil->vring2cid_tid[i][0];
1274 		if (cid >= WIL6210_MAX_CID) /* skip BCAST */
1275 			continue;
1276 		if (!wil->vring_tx_data[i].dot1x_open &&
1277 		    (skb->protocol != cpu_to_be16(ETH_P_PAE)))
1278 			continue;
1279 
1280 		/* don't Tx back to source when re-routing Rx->Tx at the AP */
1281 		if (0 == memcmp(wil->sta[cid].addr, src, ETH_ALEN))
1282 			continue;
1283 
1284 		goto found;
1285 	}
1286 
1287 	wil_dbg_txrx(wil, "Tx while no vrings active?\n");
1288 
1289 	return NULL;
1290 
1291 found:
1292 	wil_dbg_txrx(wil, "BCAST -> ring %d\n", i);
1293 	wil_set_da_for_vring(wil, skb, i);
1294 
1295 	/* find other active vrings and duplicate skb for each */
1296 	for (i++; i < WIL6210_MAX_TX_RINGS; i++) {
1297 		v2 = &wil->vring_tx[i];
1298 		txdata2 = &wil->vring_tx_data[i];
1299 		if (!v2->va || txdata2->mid != vif->mid)
1300 			continue;
1301 		cid = wil->vring2cid_tid[i][0];
1302 		if (cid >= WIL6210_MAX_CID) /* skip BCAST */
1303 			continue;
1304 		if (!wil->vring_tx_data[i].dot1x_open &&
1305 		    (skb->protocol != cpu_to_be16(ETH_P_PAE)))
1306 			continue;
1307 
1308 		if (0 == memcmp(wil->sta[cid].addr, src, ETH_ALEN))
1309 			continue;
1310 
1311 		skb2 = skb_copy(skb, GFP_ATOMIC);
1312 		if (skb2) {
1313 			wil_dbg_txrx(wil, "BCAST DUP -> ring %d\n", i);
1314 			wil_set_da_for_vring(wil, skb2, i);
1315 			wil_tx_vring(wil, vif, v2, skb2);
1316 		} else {
1317 			wil_err(wil, "skb_copy failed\n");
1318 		}
1319 	}
1320 
1321 	return v;
1322 }
1323 
1324 static int wil_tx_desc_map(struct vring_tx_desc *d, dma_addr_t pa, u32 len,
1325 			   int vring_index)
1326 {
1327 	wil_desc_addr_set(&d->dma.addr, pa);
1328 	d->dma.ip_length = 0;
1329 	/* 0..6: mac_length; 7:ip_version 0-IP6 1-IP4*/
1330 	d->dma.b11 = 0/*14 | BIT(7)*/;
1331 	d->dma.error = 0;
1332 	d->dma.status = 0; /* BIT(0) should be 0 for HW_OWNED */
1333 	d->dma.length = cpu_to_le16((u16)len);
1334 	d->dma.d0 = (vring_index << DMA_CFG_DESC_TX_0_QID_POS);
1335 	d->mac.d[0] = 0;
1336 	d->mac.d[1] = 0;
1337 	d->mac.d[2] = 0;
1338 	d->mac.ucode_cmd = 0;
1339 	/* translation type:  0 - bypass; 1 - 802.3; 2 - native wifi */
1340 	d->mac.d[2] = BIT(MAC_CFG_DESC_TX_2_SNAP_HDR_INSERTION_EN_POS) |
1341 		      (1 << MAC_CFG_DESC_TX_2_L2_TRANSLATION_TYPE_POS);
1342 
1343 	return 0;
1344 }
1345 
1346 static inline
1347 void wil_tx_desc_set_nr_frags(struct vring_tx_desc *d, int nr_frags)
1348 {
1349 	d->mac.d[2] |= (nr_frags << MAC_CFG_DESC_TX_2_NUM_OF_DESCRIPTORS_POS);
1350 }
1351 
1352 /**
1353  * Sets the descriptor @d up for csum and/or TSO offloading. The corresponding
1354  * @skb is used to obtain the protocol and headers length.
1355  * @tso_desc_type is a descriptor type for TSO: 0 - a header, 1 - first data,
1356  * 2 - middle, 3 - last descriptor.
1357  */
1358 
1359 static void wil_tx_desc_offload_setup_tso(struct vring_tx_desc *d,
1360 					  struct sk_buff *skb,
1361 					  int tso_desc_type, bool is_ipv4,
1362 					  int tcp_hdr_len, int skb_net_hdr_len)
1363 {
1364 	d->dma.b11 = ETH_HLEN; /* MAC header length */
1365 	d->dma.b11 |= is_ipv4 << DMA_CFG_DESC_TX_OFFLOAD_CFG_L3T_IPV4_POS;
1366 
1367 	d->dma.d0 |= (2 << DMA_CFG_DESC_TX_0_L4_TYPE_POS);
1368 	/* L4 header len: TCP header length */
1369 	d->dma.d0 |= (tcp_hdr_len & DMA_CFG_DESC_TX_0_L4_LENGTH_MSK);
1370 
1371 	/* Setup TSO: bit and desc type */
1372 	d->dma.d0 |= (BIT(DMA_CFG_DESC_TX_0_TCP_SEG_EN_POS)) |
1373 		(tso_desc_type << DMA_CFG_DESC_TX_0_SEGMENT_BUF_DETAILS_POS);
1374 	d->dma.d0 |= (is_ipv4 << DMA_CFG_DESC_TX_0_IPV4_CHECKSUM_EN_POS);
1375 
1376 	d->dma.ip_length = skb_net_hdr_len;
1377 	/* Enable TCP/UDP checksum */
1378 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_TCP_UDP_CHECKSUM_EN_POS);
1379 	/* Calculate pseudo-header */
1380 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_PSEUDO_HEADER_CALC_EN_POS);
1381 }
1382 
1383 /**
1384  * Sets the descriptor @d up for csum. The corresponding
1385  * @skb is used to obtain the protocol and headers length.
1386  * Returns the protocol: 0 - not TCP, 1 - TCPv4, 2 - TCPv6.
1387  * Note, if d==NULL, the function only returns the protocol result.
1388  *
1389  * It is very similar to previous wil_tx_desc_offload_setup_tso. This
1390  * is "if unrolling" to optimize the critical path.
1391  */
1392 
1393 static int wil_tx_desc_offload_setup(struct vring_tx_desc *d,
1394 				     struct sk_buff *skb){
1395 	int protocol;
1396 
1397 	if (skb->ip_summed != CHECKSUM_PARTIAL)
1398 		return 0;
1399 
1400 	d->dma.b11 = ETH_HLEN; /* MAC header length */
1401 
1402 	switch (skb->protocol) {
1403 	case cpu_to_be16(ETH_P_IP):
1404 		protocol = ip_hdr(skb)->protocol;
1405 		d->dma.b11 |= BIT(DMA_CFG_DESC_TX_OFFLOAD_CFG_L3T_IPV4_POS);
1406 		break;
1407 	case cpu_to_be16(ETH_P_IPV6):
1408 		protocol = ipv6_hdr(skb)->nexthdr;
1409 		break;
1410 	default:
1411 		return -EINVAL;
1412 	}
1413 
1414 	switch (protocol) {
1415 	case IPPROTO_TCP:
1416 		d->dma.d0 |= (2 << DMA_CFG_DESC_TX_0_L4_TYPE_POS);
1417 		/* L4 header len: TCP header length */
1418 		d->dma.d0 |=
1419 		(tcp_hdrlen(skb) & DMA_CFG_DESC_TX_0_L4_LENGTH_MSK);
1420 		break;
1421 	case IPPROTO_UDP:
1422 		/* L4 header len: UDP header length */
1423 		d->dma.d0 |=
1424 		(sizeof(struct udphdr) & DMA_CFG_DESC_TX_0_L4_LENGTH_MSK);
1425 		break;
1426 	default:
1427 		return -EINVAL;
1428 	}
1429 
1430 	d->dma.ip_length = skb_network_header_len(skb);
1431 	/* Enable TCP/UDP checksum */
1432 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_TCP_UDP_CHECKSUM_EN_POS);
1433 	/* Calculate pseudo-header */
1434 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_PSEUDO_HEADER_CALC_EN_POS);
1435 
1436 	return 0;
1437 }
1438 
1439 static inline void wil_tx_last_desc(struct vring_tx_desc *d)
1440 {
1441 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_CMD_EOP_POS) |
1442 	      BIT(DMA_CFG_DESC_TX_0_CMD_MARK_WB_POS) |
1443 	      BIT(DMA_CFG_DESC_TX_0_CMD_DMA_IT_POS);
1444 }
1445 
1446 static inline void wil_set_tx_desc_last_tso(volatile struct vring_tx_desc *d)
1447 {
1448 	d->dma.d0 |= wil_tso_type_lst <<
1449 		  DMA_CFG_DESC_TX_0_SEGMENT_BUF_DETAILS_POS;
1450 }
1451 
1452 static int __wil_tx_vring_tso(struct wil6210_priv *wil, struct wil6210_vif *vif,
1453 			      struct vring *vring, struct sk_buff *skb)
1454 {
1455 	struct device *dev = wil_to_dev(wil);
1456 
1457 	/* point to descriptors in shared memory */
1458 	volatile struct vring_tx_desc *_desc = NULL, *_hdr_desc,
1459 				      *_first_desc = NULL;
1460 
1461 	/* pointers to shadow descriptors */
1462 	struct vring_tx_desc desc_mem, hdr_desc_mem, first_desc_mem,
1463 			     *d = &hdr_desc_mem, *hdr_desc = &hdr_desc_mem,
1464 			     *first_desc = &first_desc_mem;
1465 
1466 	/* pointer to shadow descriptors' context */
1467 	struct wil_ctx *hdr_ctx, *first_ctx = NULL;
1468 
1469 	int descs_used = 0; /* total number of used descriptors */
1470 	int sg_desc_cnt = 0; /* number of descriptors for current mss*/
1471 
1472 	u32 swhead = vring->swhead;
1473 	int used, avail = wil_vring_avail_tx(vring);
1474 	int nr_frags = skb_shinfo(skb)->nr_frags;
1475 	int min_desc_required = nr_frags + 1;
1476 	int mss = skb_shinfo(skb)->gso_size;	/* payload size w/o headers */
1477 	int f, len, hdrlen, headlen;
1478 	int vring_index = vring - wil->vring_tx;
1479 	struct vring_tx_data *txdata = &wil->vring_tx_data[vring_index];
1480 	uint i = swhead;
1481 	dma_addr_t pa;
1482 	const skb_frag_t *frag = NULL;
1483 	int rem_data = mss;
1484 	int lenmss;
1485 	int hdr_compensation_need = true;
1486 	int desc_tso_type = wil_tso_type_first;
1487 	bool is_ipv4;
1488 	int tcp_hdr_len;
1489 	int skb_net_hdr_len;
1490 	int gso_type;
1491 	int rc = -EINVAL;
1492 
1493 	wil_dbg_txrx(wil, "tx_vring_tso: %d bytes to vring %d\n", skb->len,
1494 		     vring_index);
1495 
1496 	if (unlikely(!txdata->enabled))
1497 		return -EINVAL;
1498 
1499 	/* A typical page 4K is 3-4 payloads, we assume each fragment
1500 	 * is a full payload, that's how min_desc_required has been
1501 	 * calculated. In real we might need more or less descriptors,
1502 	 * this is the initial check only.
1503 	 */
1504 	if (unlikely(avail < min_desc_required)) {
1505 		wil_err_ratelimited(wil,
1506 				    "TSO: Tx ring[%2d] full. No space for %d fragments\n",
1507 				    vring_index, min_desc_required);
1508 		return -ENOMEM;
1509 	}
1510 
1511 	/* Header Length = MAC header len + IP header len + TCP header len*/
1512 	hdrlen = ETH_HLEN +
1513 		(int)skb_network_header_len(skb) +
1514 		tcp_hdrlen(skb);
1515 
1516 	gso_type = skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV6 | SKB_GSO_TCPV4);
1517 	switch (gso_type) {
1518 	case SKB_GSO_TCPV4:
1519 		/* TCP v4, zero out the IP length and IPv4 checksum fields
1520 		 * as required by the offloading doc
1521 		 */
1522 		ip_hdr(skb)->tot_len = 0;
1523 		ip_hdr(skb)->check = 0;
1524 		is_ipv4 = true;
1525 		break;
1526 	case SKB_GSO_TCPV6:
1527 		/* TCP v6, zero out the payload length */
1528 		ipv6_hdr(skb)->payload_len = 0;
1529 		is_ipv4 = false;
1530 		break;
1531 	default:
1532 		/* other than TCPv4 or TCPv6 types are not supported for TSO.
1533 		 * It is also illegal for both to be set simultaneously
1534 		 */
1535 		return -EINVAL;
1536 	}
1537 
1538 	if (skb->ip_summed != CHECKSUM_PARTIAL)
1539 		return -EINVAL;
1540 
1541 	/* tcp header length and skb network header length are fixed for all
1542 	 * packet's descriptors - read then once here
1543 	 */
1544 	tcp_hdr_len = tcp_hdrlen(skb);
1545 	skb_net_hdr_len = skb_network_header_len(skb);
1546 
1547 	_hdr_desc = &vring->va[i].tx;
1548 
1549 	pa = dma_map_single(dev, skb->data, hdrlen, DMA_TO_DEVICE);
1550 	if (unlikely(dma_mapping_error(dev, pa))) {
1551 		wil_err(wil, "TSO: Skb head DMA map error\n");
1552 		goto err_exit;
1553 	}
1554 
1555 	wil_tx_desc_map(hdr_desc, pa, hdrlen, vring_index);
1556 	wil_tx_desc_offload_setup_tso(hdr_desc, skb, wil_tso_type_hdr, is_ipv4,
1557 				      tcp_hdr_len, skb_net_hdr_len);
1558 	wil_tx_last_desc(hdr_desc);
1559 
1560 	vring->ctx[i].mapped_as = wil_mapped_as_single;
1561 	hdr_ctx = &vring->ctx[i];
1562 
1563 	descs_used++;
1564 	headlen = skb_headlen(skb) - hdrlen;
1565 
1566 	for (f = headlen ? -1 : 0; f < nr_frags; f++)  {
1567 		if (headlen) {
1568 			len = headlen;
1569 			wil_dbg_txrx(wil, "TSO: process skb head, len %u\n",
1570 				     len);
1571 		} else {
1572 			frag = &skb_shinfo(skb)->frags[f];
1573 			len = frag->size;
1574 			wil_dbg_txrx(wil, "TSO: frag[%d]: len %u\n", f, len);
1575 		}
1576 
1577 		while (len) {
1578 			wil_dbg_txrx(wil,
1579 				     "TSO: len %d, rem_data %d, descs_used %d\n",
1580 				     len, rem_data, descs_used);
1581 
1582 			if (descs_used == avail)  {
1583 				wil_err_ratelimited(wil, "TSO: ring overflow\n");
1584 				rc = -ENOMEM;
1585 				goto mem_error;
1586 			}
1587 
1588 			lenmss = min_t(int, rem_data, len);
1589 			i = (swhead + descs_used) % vring->size;
1590 			wil_dbg_txrx(wil, "TSO: lenmss %d, i %d\n", lenmss, i);
1591 
1592 			if (!headlen) {
1593 				pa = skb_frag_dma_map(dev, frag,
1594 						      frag->size - len, lenmss,
1595 						      DMA_TO_DEVICE);
1596 				vring->ctx[i].mapped_as = wil_mapped_as_page;
1597 			} else {
1598 				pa = dma_map_single(dev,
1599 						    skb->data +
1600 						    skb_headlen(skb) - headlen,
1601 						    lenmss,
1602 						    DMA_TO_DEVICE);
1603 				vring->ctx[i].mapped_as = wil_mapped_as_single;
1604 				headlen -= lenmss;
1605 			}
1606 
1607 			if (unlikely(dma_mapping_error(dev, pa))) {
1608 				wil_err(wil, "TSO: DMA map page error\n");
1609 				goto mem_error;
1610 			}
1611 
1612 			_desc = &vring->va[i].tx;
1613 
1614 			if (!_first_desc) {
1615 				_first_desc = _desc;
1616 				first_ctx = &vring->ctx[i];
1617 				d = first_desc;
1618 			} else {
1619 				d = &desc_mem;
1620 			}
1621 
1622 			wil_tx_desc_map(d, pa, lenmss, vring_index);
1623 			wil_tx_desc_offload_setup_tso(d, skb, desc_tso_type,
1624 						      is_ipv4, tcp_hdr_len,
1625 						      skb_net_hdr_len);
1626 
1627 			/* use tso_type_first only once */
1628 			desc_tso_type = wil_tso_type_mid;
1629 
1630 			descs_used++;  /* desc used so far */
1631 			sg_desc_cnt++; /* desc used for this segment */
1632 			len -= lenmss;
1633 			rem_data -= lenmss;
1634 
1635 			wil_dbg_txrx(wil,
1636 				     "TSO: len %d, rem_data %d, descs_used %d, sg_desc_cnt %d,\n",
1637 				     len, rem_data, descs_used, sg_desc_cnt);
1638 
1639 			/* Close the segment if reached mss size or last frag*/
1640 			if (rem_data == 0 || (f == nr_frags - 1 && len == 0)) {
1641 				if (hdr_compensation_need) {
1642 					/* first segment include hdr desc for
1643 					 * release
1644 					 */
1645 					hdr_ctx->nr_frags = sg_desc_cnt;
1646 					wil_tx_desc_set_nr_frags(first_desc,
1647 								 sg_desc_cnt +
1648 								 1);
1649 					hdr_compensation_need = false;
1650 				} else {
1651 					wil_tx_desc_set_nr_frags(first_desc,
1652 								 sg_desc_cnt);
1653 				}
1654 				first_ctx->nr_frags = sg_desc_cnt - 1;
1655 
1656 				wil_tx_last_desc(d);
1657 
1658 				/* first descriptor may also be the last
1659 				 * for this mss - make sure not to copy
1660 				 * it twice
1661 				 */
1662 				if (first_desc != d)
1663 					*_first_desc = *first_desc;
1664 
1665 				/*last descriptor will be copied at the end
1666 				 * of this TS processing
1667 				 */
1668 				if (f < nr_frags - 1 || len > 0)
1669 					*_desc = *d;
1670 
1671 				rem_data = mss;
1672 				_first_desc = NULL;
1673 				sg_desc_cnt = 0;
1674 			} else if (first_desc != d) /* update mid descriptor */
1675 					*_desc = *d;
1676 		}
1677 	}
1678 
1679 	/* first descriptor may also be the last.
1680 	 * in this case d pointer is invalid
1681 	 */
1682 	if (_first_desc == _desc)
1683 		d = first_desc;
1684 
1685 	/* Last data descriptor */
1686 	wil_set_tx_desc_last_tso(d);
1687 	*_desc = *d;
1688 
1689 	/* Fill the total number of descriptors in first desc (hdr)*/
1690 	wil_tx_desc_set_nr_frags(hdr_desc, descs_used);
1691 	*_hdr_desc = *hdr_desc;
1692 
1693 	/* hold reference to skb
1694 	 * to prevent skb release before accounting
1695 	 * in case of immediate "tx done"
1696 	 */
1697 	vring->ctx[i].skb = skb_get(skb);
1698 
1699 	/* performance monitoring */
1700 	used = wil_vring_used_tx(vring);
1701 	if (wil_val_in_range(wil->vring_idle_trsh,
1702 			     used, used + descs_used)) {
1703 		txdata->idle += get_cycles() - txdata->last_idle;
1704 		wil_dbg_txrx(wil,  "Ring[%2d] not idle %d -> %d\n",
1705 			     vring_index, used, used + descs_used);
1706 	}
1707 
1708 	/* Make sure to advance the head only after descriptor update is done.
1709 	 * This will prevent a race condition where the completion thread
1710 	 * will see the DU bit set from previous run and will handle the
1711 	 * skb before it was completed.
1712 	 */
1713 	wmb();
1714 
1715 	/* advance swhead */
1716 	wil_vring_advance_head(vring, descs_used);
1717 	wil_dbg_txrx(wil, "TSO: Tx swhead %d -> %d\n", swhead, vring->swhead);
1718 
1719 	/* make sure all writes to descriptors (shared memory) are done before
1720 	 * committing them to HW
1721 	 */
1722 	wmb();
1723 
1724 	wil_w(wil, vring->hwtail, vring->swhead);
1725 	return 0;
1726 
1727 mem_error:
1728 	while (descs_used > 0) {
1729 		struct wil_ctx *ctx;
1730 
1731 		i = (swhead + descs_used - 1) % vring->size;
1732 		d = (struct vring_tx_desc *)&vring->va[i].tx;
1733 		_desc = &vring->va[i].tx;
1734 		*d = *_desc;
1735 		_desc->dma.status = TX_DMA_STATUS_DU;
1736 		ctx = &vring->ctx[i];
1737 		wil_txdesc_unmap(dev, d, ctx);
1738 		memset(ctx, 0, sizeof(*ctx));
1739 		descs_used--;
1740 	}
1741 err_exit:
1742 	return rc;
1743 }
1744 
1745 static int __wil_tx_vring(struct wil6210_priv *wil, struct wil6210_vif *vif,
1746 			  struct vring *vring, struct sk_buff *skb)
1747 {
1748 	struct device *dev = wil_to_dev(wil);
1749 	struct vring_tx_desc dd, *d = &dd;
1750 	volatile struct vring_tx_desc *_d;
1751 	u32 swhead = vring->swhead;
1752 	int avail = wil_vring_avail_tx(vring);
1753 	int nr_frags = skb_shinfo(skb)->nr_frags;
1754 	uint f = 0;
1755 	int vring_index = vring - wil->vring_tx;
1756 	struct vring_tx_data *txdata = &wil->vring_tx_data[vring_index];
1757 	uint i = swhead;
1758 	dma_addr_t pa;
1759 	int used;
1760 	bool mcast = (vring_index == vif->bcast_vring);
1761 	uint len = skb_headlen(skb);
1762 
1763 	wil_dbg_txrx(wil, "tx_vring: %d bytes to vring %d\n", skb->len,
1764 		     vring_index);
1765 
1766 	if (unlikely(!txdata->enabled))
1767 		return -EINVAL;
1768 
1769 	if (unlikely(avail < 1 + nr_frags)) {
1770 		wil_err_ratelimited(wil,
1771 				    "Tx ring[%2d] full. No space for %d fragments\n",
1772 				    vring_index, 1 + nr_frags);
1773 		return -ENOMEM;
1774 	}
1775 	_d = &vring->va[i].tx;
1776 
1777 	pa = dma_map_single(dev, skb->data, skb_headlen(skb), DMA_TO_DEVICE);
1778 
1779 	wil_dbg_txrx(wil, "Tx[%2d] skb %d bytes 0x%p -> %pad\n", vring_index,
1780 		     skb_headlen(skb), skb->data, &pa);
1781 	wil_hex_dump_txrx("Tx ", DUMP_PREFIX_OFFSET, 16, 1,
1782 			  skb->data, skb_headlen(skb), false);
1783 
1784 	if (unlikely(dma_mapping_error(dev, pa)))
1785 		return -EINVAL;
1786 	vring->ctx[i].mapped_as = wil_mapped_as_single;
1787 	/* 1-st segment */
1788 	wil_tx_desc_map(d, pa, len, vring_index);
1789 	if (unlikely(mcast)) {
1790 		d->mac.d[0] |= BIT(MAC_CFG_DESC_TX_0_MCS_EN_POS); /* MCS 0 */
1791 		if (unlikely(len > WIL_BCAST_MCS0_LIMIT)) /* set MCS 1 */
1792 			d->mac.d[0] |= (1 << MAC_CFG_DESC_TX_0_MCS_INDEX_POS);
1793 	}
1794 	/* Process TCP/UDP checksum offloading */
1795 	if (unlikely(wil_tx_desc_offload_setup(d, skb))) {
1796 		wil_err(wil, "Tx[%2d] Failed to set cksum, drop packet\n",
1797 			vring_index);
1798 		goto dma_error;
1799 	}
1800 
1801 	vring->ctx[i].nr_frags = nr_frags;
1802 	wil_tx_desc_set_nr_frags(d, nr_frags + 1);
1803 
1804 	/* middle segments */
1805 	for (; f < nr_frags; f++) {
1806 		const struct skb_frag_struct *frag =
1807 				&skb_shinfo(skb)->frags[f];
1808 		int len = skb_frag_size(frag);
1809 
1810 		*_d = *d;
1811 		wil_dbg_txrx(wil, "Tx[%2d] desc[%4d]\n", vring_index, i);
1812 		wil_hex_dump_txrx("TxD ", DUMP_PREFIX_NONE, 32, 4,
1813 				  (const void *)d, sizeof(*d), false);
1814 		i = (swhead + f + 1) % vring->size;
1815 		_d = &vring->va[i].tx;
1816 		pa = skb_frag_dma_map(dev, frag, 0, skb_frag_size(frag),
1817 				      DMA_TO_DEVICE);
1818 		if (unlikely(dma_mapping_error(dev, pa))) {
1819 			wil_err(wil, "Tx[%2d] failed to map fragment\n",
1820 				vring_index);
1821 			goto dma_error;
1822 		}
1823 		vring->ctx[i].mapped_as = wil_mapped_as_page;
1824 		wil_tx_desc_map(d, pa, len, vring_index);
1825 		/* no need to check return code -
1826 		 * if it succeeded for 1-st descriptor,
1827 		 * it will succeed here too
1828 		 */
1829 		wil_tx_desc_offload_setup(d, skb);
1830 	}
1831 	/* for the last seg only */
1832 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_CMD_EOP_POS);
1833 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_CMD_MARK_WB_POS);
1834 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_CMD_DMA_IT_POS);
1835 	*_d = *d;
1836 	wil_dbg_txrx(wil, "Tx[%2d] desc[%4d]\n", vring_index, i);
1837 	wil_hex_dump_txrx("TxD ", DUMP_PREFIX_NONE, 32, 4,
1838 			  (const void *)d, sizeof(*d), false);
1839 
1840 	/* hold reference to skb
1841 	 * to prevent skb release before accounting
1842 	 * in case of immediate "tx done"
1843 	 */
1844 	vring->ctx[i].skb = skb_get(skb);
1845 
1846 	/* performance monitoring */
1847 	used = wil_vring_used_tx(vring);
1848 	if (wil_val_in_range(wil->vring_idle_trsh,
1849 			     used, used + nr_frags + 1)) {
1850 		txdata->idle += get_cycles() - txdata->last_idle;
1851 		wil_dbg_txrx(wil,  "Ring[%2d] not idle %d -> %d\n",
1852 			     vring_index, used, used + nr_frags + 1);
1853 	}
1854 
1855 	/* Make sure to advance the head only after descriptor update is done.
1856 	 * This will prevent a race condition where the completion thread
1857 	 * will see the DU bit set from previous run and will handle the
1858 	 * skb before it was completed.
1859 	 */
1860 	wmb();
1861 
1862 	/* advance swhead */
1863 	wil_vring_advance_head(vring, nr_frags + 1);
1864 	wil_dbg_txrx(wil, "Tx[%2d] swhead %d -> %d\n", vring_index, swhead,
1865 		     vring->swhead);
1866 	trace_wil6210_tx(vring_index, swhead, skb->len, nr_frags);
1867 
1868 	/* make sure all writes to descriptors (shared memory) are done before
1869 	 * committing them to HW
1870 	 */
1871 	wmb();
1872 
1873 	wil_w(wil, vring->hwtail, vring->swhead);
1874 
1875 	return 0;
1876  dma_error:
1877 	/* unmap what we have mapped */
1878 	nr_frags = f + 1; /* frags mapped + one for skb head */
1879 	for (f = 0; f < nr_frags; f++) {
1880 		struct wil_ctx *ctx;
1881 
1882 		i = (swhead + f) % vring->size;
1883 		ctx = &vring->ctx[i];
1884 		_d = &vring->va[i].tx;
1885 		*d = *_d;
1886 		_d->dma.status = TX_DMA_STATUS_DU;
1887 		wil_txdesc_unmap(dev, d, ctx);
1888 
1889 		memset(ctx, 0, sizeof(*ctx));
1890 	}
1891 
1892 	return -EINVAL;
1893 }
1894 
1895 static int wil_tx_vring(struct wil6210_priv *wil, struct wil6210_vif *vif,
1896 			struct vring *vring, struct sk_buff *skb)
1897 {
1898 	int vring_index = vring - wil->vring_tx;
1899 	struct vring_tx_data *txdata = &wil->vring_tx_data[vring_index];
1900 	int rc;
1901 
1902 	spin_lock(&txdata->lock);
1903 
1904 	if (test_bit(wil_status_suspending, wil->status) ||
1905 	    test_bit(wil_status_suspended, wil->status) ||
1906 	    test_bit(wil_status_resuming, wil->status)) {
1907 		wil_dbg_txrx(wil,
1908 			     "suspend/resume in progress. drop packet\n");
1909 		spin_unlock(&txdata->lock);
1910 		return -EINVAL;
1911 	}
1912 
1913 	rc = (skb_is_gso(skb) ? __wil_tx_vring_tso : __wil_tx_vring)
1914 	     (wil, vif, vring, skb);
1915 
1916 	spin_unlock(&txdata->lock);
1917 
1918 	return rc;
1919 }
1920 
1921 /**
1922  * Check status of tx vrings and stop/wake net queues if needed
1923  * It will start/stop net queues of a specific VIF net_device.
1924  *
1925  * This function does one of two checks:
1926  * In case check_stop is true, will check if net queues need to be stopped. If
1927  * the conditions for stopping are met, netif_tx_stop_all_queues() is called.
1928  * In case check_stop is false, will check if net queues need to be waked. If
1929  * the conditions for waking are met, netif_tx_wake_all_queues() is called.
1930  * vring is the vring which is currently being modified by either adding
1931  * descriptors (tx) into it or removing descriptors (tx complete) from it. Can
1932  * be null when irrelevant (e.g. connect/disconnect events).
1933  *
1934  * The implementation is to stop net queues if modified vring has low
1935  * descriptor availability. Wake if all vrings are not in low descriptor
1936  * availability and modified vring has high descriptor availability.
1937  */
1938 static inline void __wil_update_net_queues(struct wil6210_priv *wil,
1939 					   struct wil6210_vif *vif,
1940 					   struct vring *vring,
1941 					   bool check_stop)
1942 {
1943 	int i;
1944 
1945 	if (unlikely(!vif))
1946 		return;
1947 
1948 	if (vring)
1949 		wil_dbg_txrx(wil, "vring %d, mid %d, check_stop=%d, stopped=%d",
1950 			     (int)(vring - wil->vring_tx), vif->mid, check_stop,
1951 			     vif->net_queue_stopped);
1952 	else
1953 		wil_dbg_txrx(wil, "check_stop=%d, mid=%d, stopped=%d",
1954 			     check_stop, vif->mid, vif->net_queue_stopped);
1955 
1956 	if (check_stop == vif->net_queue_stopped)
1957 		/* net queues already in desired state */
1958 		return;
1959 
1960 	if (check_stop) {
1961 		if (!vring || unlikely(wil_vring_avail_low(vring))) {
1962 			/* not enough room in the vring */
1963 			netif_tx_stop_all_queues(vif_to_ndev(vif));
1964 			vif->net_queue_stopped = true;
1965 			wil_dbg_txrx(wil, "netif_tx_stop called\n");
1966 		}
1967 		return;
1968 	}
1969 
1970 	/* Do not wake the queues in suspend flow */
1971 	if (test_bit(wil_status_suspending, wil->status) ||
1972 	    test_bit(wil_status_suspended, wil->status))
1973 		return;
1974 
1975 	/* check wake */
1976 	for (i = 0; i < WIL6210_MAX_TX_RINGS; i++) {
1977 		struct vring *cur_vring = &wil->vring_tx[i];
1978 		struct vring_tx_data *txdata = &wil->vring_tx_data[i];
1979 
1980 		if (txdata->mid != vif->mid || !cur_vring->va ||
1981 		    !txdata->enabled || cur_vring == vring)
1982 			continue;
1983 
1984 		if (wil_vring_avail_low(cur_vring)) {
1985 			wil_dbg_txrx(wil, "vring %d full, can't wake\n",
1986 				     (int)(cur_vring - wil->vring_tx));
1987 			return;
1988 		}
1989 	}
1990 
1991 	if (!vring || wil_vring_avail_high(vring)) {
1992 		/* enough room in the vring */
1993 		wil_dbg_txrx(wil, "calling netif_tx_wake\n");
1994 		netif_tx_wake_all_queues(vif_to_ndev(vif));
1995 		vif->net_queue_stopped = false;
1996 	}
1997 }
1998 
1999 void wil_update_net_queues(struct wil6210_priv *wil, struct wil6210_vif *vif,
2000 			   struct vring *vring, bool check_stop)
2001 {
2002 	spin_lock(&wil->net_queue_lock);
2003 	__wil_update_net_queues(wil, vif, vring, check_stop);
2004 	spin_unlock(&wil->net_queue_lock);
2005 }
2006 
2007 void wil_update_net_queues_bh(struct wil6210_priv *wil, struct wil6210_vif *vif,
2008 			      struct vring *vring, bool check_stop)
2009 {
2010 	spin_lock_bh(&wil->net_queue_lock);
2011 	__wil_update_net_queues(wil, vif, vring, check_stop);
2012 	spin_unlock_bh(&wil->net_queue_lock);
2013 }
2014 
2015 netdev_tx_t wil_start_xmit(struct sk_buff *skb, struct net_device *ndev)
2016 {
2017 	struct wil6210_vif *vif = ndev_to_vif(ndev);
2018 	struct wil6210_priv *wil = vif_to_wil(vif);
2019 	struct ethhdr *eth = (void *)skb->data;
2020 	bool bcast = is_multicast_ether_addr(eth->h_dest);
2021 	struct vring *vring;
2022 	static bool pr_once_fw;
2023 	int rc;
2024 
2025 	wil_dbg_txrx(wil, "start_xmit\n");
2026 	if (unlikely(!test_bit(wil_status_fwready, wil->status))) {
2027 		if (!pr_once_fw) {
2028 			wil_err(wil, "FW not ready\n");
2029 			pr_once_fw = true;
2030 		}
2031 		goto drop;
2032 	}
2033 	if (unlikely(!test_bit(wil_vif_fwconnected, vif->status))) {
2034 		wil_dbg_ratelimited(wil,
2035 				    "VIF not connected, packet dropped\n");
2036 		goto drop;
2037 	}
2038 	if (unlikely(vif->wdev.iftype == NL80211_IFTYPE_MONITOR)) {
2039 		wil_err(wil, "Xmit in monitor mode not supported\n");
2040 		goto drop;
2041 	}
2042 	pr_once_fw = false;
2043 
2044 	/* find vring */
2045 	if (vif->wdev.iftype == NL80211_IFTYPE_STATION && !vif->pbss) {
2046 		/* in STA mode (ESS), all to same VRING (to AP) */
2047 		vring = wil_find_tx_vring_sta(wil, vif, skb);
2048 	} else if (bcast) {
2049 		if (vif->pbss)
2050 			/* in pbss, no bcast VRING - duplicate skb in
2051 			 * all stations VRINGs
2052 			 */
2053 			vring = wil_find_tx_bcast_2(wil, vif, skb);
2054 		else if (vif->wdev.iftype == NL80211_IFTYPE_AP)
2055 			/* AP has a dedicated bcast VRING */
2056 			vring = wil_find_tx_bcast_1(wil, vif, skb);
2057 		else
2058 			/* unexpected combination, fallback to duplicating
2059 			 * the skb in all stations VRINGs
2060 			 */
2061 			vring = wil_find_tx_bcast_2(wil, vif, skb);
2062 	} else {
2063 		/* unicast, find specific VRING by dest. address */
2064 		vring = wil_find_tx_ucast(wil, vif, skb);
2065 	}
2066 	if (unlikely(!vring)) {
2067 		wil_dbg_txrx(wil, "No Tx VRING found for %pM\n", eth->h_dest);
2068 		goto drop;
2069 	}
2070 	/* set up vring entry */
2071 	rc = wil_tx_vring(wil, vif, vring, skb);
2072 
2073 	switch (rc) {
2074 	case 0:
2075 		/* shall we stop net queues? */
2076 		wil_update_net_queues_bh(wil, vif, vring, true);
2077 		/* statistics will be updated on the tx_complete */
2078 		dev_kfree_skb_any(skb);
2079 		return NETDEV_TX_OK;
2080 	case -ENOMEM:
2081 		return NETDEV_TX_BUSY;
2082 	default:
2083 		break; /* goto drop; */
2084 	}
2085  drop:
2086 	ndev->stats.tx_dropped++;
2087 	dev_kfree_skb_any(skb);
2088 
2089 	return NET_XMIT_DROP;
2090 }
2091 
2092 static inline bool wil_need_txstat(struct sk_buff *skb)
2093 {
2094 	struct ethhdr *eth = (void *)skb->data;
2095 
2096 	return is_unicast_ether_addr(eth->h_dest) && skb->sk &&
2097 	       (skb_shinfo(skb)->tx_flags & SKBTX_WIFI_STATUS);
2098 }
2099 
2100 static inline void wil_consume_skb(struct sk_buff *skb, bool acked)
2101 {
2102 	if (unlikely(wil_need_txstat(skb)))
2103 		skb_complete_wifi_ack(skb, acked);
2104 	else
2105 		acked ? dev_consume_skb_any(skb) : dev_kfree_skb_any(skb);
2106 }
2107 
2108 /**
2109  * Clean up transmitted skb's from the Tx VRING
2110  *
2111  * Return number of descriptors cleared
2112  *
2113  * Safe to call from IRQ
2114  */
2115 int wil_tx_complete(struct wil6210_vif *vif, int ringid)
2116 {
2117 	struct wil6210_priv *wil = vif_to_wil(vif);
2118 	struct net_device *ndev = vif_to_ndev(vif);
2119 	struct device *dev = wil_to_dev(wil);
2120 	struct vring *vring = &wil->vring_tx[ringid];
2121 	struct vring_tx_data *txdata = &wil->vring_tx_data[ringid];
2122 	int done = 0;
2123 	int cid = wil->vring2cid_tid[ringid][0];
2124 	struct wil_net_stats *stats = NULL;
2125 	volatile struct vring_tx_desc *_d;
2126 	int used_before_complete;
2127 	int used_new;
2128 
2129 	if (unlikely(!vring->va)) {
2130 		wil_err(wil, "Tx irq[%d]: vring not initialized\n", ringid);
2131 		return 0;
2132 	}
2133 
2134 	if (unlikely(!txdata->enabled)) {
2135 		wil_info(wil, "Tx irq[%d]: vring disabled\n", ringid);
2136 		return 0;
2137 	}
2138 
2139 	wil_dbg_txrx(wil, "tx_complete: (%d)\n", ringid);
2140 
2141 	used_before_complete = wil_vring_used_tx(vring);
2142 
2143 	if (cid < WIL6210_MAX_CID)
2144 		stats = &wil->sta[cid].stats;
2145 
2146 	while (!wil_vring_is_empty(vring)) {
2147 		int new_swtail;
2148 		struct wil_ctx *ctx = &vring->ctx[vring->swtail];
2149 		/**
2150 		 * For the fragmented skb, HW will set DU bit only for the
2151 		 * last fragment. look for it.
2152 		 * In TSO the first DU will include hdr desc
2153 		 */
2154 		int lf = (vring->swtail + ctx->nr_frags) % vring->size;
2155 		/* TODO: check we are not past head */
2156 
2157 		_d = &vring->va[lf].tx;
2158 		if (unlikely(!(_d->dma.status & TX_DMA_STATUS_DU)))
2159 			break;
2160 
2161 		new_swtail = (lf + 1) % vring->size;
2162 		while (vring->swtail != new_swtail) {
2163 			struct vring_tx_desc dd, *d = &dd;
2164 			u16 dmalen;
2165 			struct sk_buff *skb;
2166 
2167 			ctx = &vring->ctx[vring->swtail];
2168 			skb = ctx->skb;
2169 			_d = &vring->va[vring->swtail].tx;
2170 
2171 			*d = *_d;
2172 
2173 			dmalen = le16_to_cpu(d->dma.length);
2174 			trace_wil6210_tx_done(ringid, vring->swtail, dmalen,
2175 					      d->dma.error);
2176 			wil_dbg_txrx(wil,
2177 				     "TxC[%2d][%3d] : %d bytes, status 0x%02x err 0x%02x\n",
2178 				     ringid, vring->swtail, dmalen,
2179 				     d->dma.status, d->dma.error);
2180 			wil_hex_dump_txrx("TxCD ", DUMP_PREFIX_NONE, 32, 4,
2181 					  (const void *)d, sizeof(*d), false);
2182 
2183 			wil_txdesc_unmap(dev, d, ctx);
2184 
2185 			if (skb) {
2186 				if (likely(d->dma.error == 0)) {
2187 					ndev->stats.tx_packets++;
2188 					ndev->stats.tx_bytes += skb->len;
2189 					if (stats) {
2190 						stats->tx_packets++;
2191 						stats->tx_bytes += skb->len;
2192 					}
2193 				} else {
2194 					ndev->stats.tx_errors++;
2195 					if (stats)
2196 						stats->tx_errors++;
2197 				}
2198 				wil_consume_skb(skb, d->dma.error == 0);
2199 			}
2200 			memset(ctx, 0, sizeof(*ctx));
2201 			/* Make sure the ctx is zeroed before updating the tail
2202 			 * to prevent a case where wil_tx_vring will see
2203 			 * this descriptor as used and handle it before ctx zero
2204 			 * is completed.
2205 			 */
2206 			wmb();
2207 			/* There is no need to touch HW descriptor:
2208 			 * - ststus bit TX_DMA_STATUS_DU is set by design,
2209 			 *   so hardware will not try to process this desc.,
2210 			 * - rest of descriptor will be initialized on Tx.
2211 			 */
2212 			vring->swtail = wil_vring_next_tail(vring);
2213 			done++;
2214 		}
2215 	}
2216 
2217 	/* performance monitoring */
2218 	used_new = wil_vring_used_tx(vring);
2219 	if (wil_val_in_range(wil->vring_idle_trsh,
2220 			     used_new, used_before_complete)) {
2221 		wil_dbg_txrx(wil, "Ring[%2d] idle %d -> %d\n",
2222 			     ringid, used_before_complete, used_new);
2223 		txdata->last_idle = get_cycles();
2224 	}
2225 
2226 	/* shall we wake net queues? */
2227 	if (done)
2228 		wil_update_net_queues(wil, vif, vring, false);
2229 
2230 	return done;
2231 }
2232