1 /*
2  * Copyright (c) 2012-2017 Qualcomm Atheros, Inc.
3  *
4  * Permission to use, copy, modify, and/or distribute this software for any
5  * purpose with or without fee is hereby granted, provided that the above
6  * copyright notice and this permission notice appear in all copies.
7  *
8  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
9  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
10  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
11  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
12  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
13  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
14  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
15  */
16 
17 #include <linux/etherdevice.h>
18 #include <net/ieee80211_radiotap.h>
19 #include <linux/if_arp.h>
20 #include <linux/moduleparam.h>
21 #include <linux/ip.h>
22 #include <linux/ipv6.h>
23 #include <net/ipv6.h>
24 #include <linux/prefetch.h>
25 
26 #include "wil6210.h"
27 #include "wmi.h"
28 #include "txrx.h"
29 #include "trace.h"
30 
31 static bool rtap_include_phy_info;
32 module_param(rtap_include_phy_info, bool, 0444);
33 MODULE_PARM_DESC(rtap_include_phy_info,
34 		 " Include PHY info in the radiotap header, default - no");
35 
36 bool rx_align_2;
37 module_param(rx_align_2, bool, 0444);
38 MODULE_PARM_DESC(rx_align_2, " align Rx buffers on 4*n+2, default - no");
39 
40 static inline uint wil_rx_snaplen(void)
41 {
42 	return rx_align_2 ? 6 : 0;
43 }
44 
45 static inline int wil_vring_is_empty(struct vring *vring)
46 {
47 	return vring->swhead == vring->swtail;
48 }
49 
50 static inline u32 wil_vring_next_tail(struct vring *vring)
51 {
52 	return (vring->swtail + 1) % vring->size;
53 }
54 
55 static inline void wil_vring_advance_head(struct vring *vring, int n)
56 {
57 	vring->swhead = (vring->swhead + n) % vring->size;
58 }
59 
60 static inline int wil_vring_is_full(struct vring *vring)
61 {
62 	return wil_vring_next_tail(vring) == vring->swhead;
63 }
64 
65 /* Used space in Tx Vring */
66 static inline int wil_vring_used_tx(struct vring *vring)
67 {
68 	u32 swhead = vring->swhead;
69 	u32 swtail = vring->swtail;
70 	return (vring->size + swhead - swtail) % vring->size;
71 }
72 
73 /* Available space in Tx Vring */
74 static inline int wil_vring_avail_tx(struct vring *vring)
75 {
76 	return vring->size - wil_vring_used_tx(vring) - 1;
77 }
78 
79 /* wil_vring_wmark_low - low watermark for available descriptor space */
80 static inline int wil_vring_wmark_low(struct vring *vring)
81 {
82 	return vring->size/8;
83 }
84 
85 /* wil_vring_wmark_high - high watermark for available descriptor space */
86 static inline int wil_vring_wmark_high(struct vring *vring)
87 {
88 	return vring->size/4;
89 }
90 
91 /* returns true if num avail descriptors is lower than wmark_low */
92 static inline int wil_vring_avail_low(struct vring *vring)
93 {
94 	return wil_vring_avail_tx(vring) < wil_vring_wmark_low(vring);
95 }
96 
97 /* returns true if num avail descriptors is higher than wmark_high */
98 static inline int wil_vring_avail_high(struct vring *vring)
99 {
100 	return wil_vring_avail_tx(vring) > wil_vring_wmark_high(vring);
101 }
102 
103 /* wil_val_in_range - check if value in [min,max) */
104 static inline bool wil_val_in_range(int val, int min, int max)
105 {
106 	return val >= min && val < max;
107 }
108 
109 static int wil_vring_alloc(struct wil6210_priv *wil, struct vring *vring)
110 {
111 	struct device *dev = wil_to_dev(wil);
112 	size_t sz = vring->size * sizeof(vring->va[0]);
113 	uint i;
114 
115 	wil_dbg_misc(wil, "vring_alloc:\n");
116 
117 	BUILD_BUG_ON(sizeof(vring->va[0]) != 32);
118 
119 	vring->swhead = 0;
120 	vring->swtail = 0;
121 	vring->ctx = kcalloc(vring->size, sizeof(vring->ctx[0]), GFP_KERNEL);
122 	if (!vring->ctx) {
123 		vring->va = NULL;
124 		return -ENOMEM;
125 	}
126 	/* vring->va should be aligned on its size rounded up to power of 2
127 	 * This is granted by the dma_alloc_coherent
128 	 */
129 	vring->va = dma_alloc_coherent(dev, sz, &vring->pa, GFP_KERNEL);
130 	if (!vring->va) {
131 		kfree(vring->ctx);
132 		vring->ctx = NULL;
133 		return -ENOMEM;
134 	}
135 	/* initially, all descriptors are SW owned
136 	 * For Tx and Rx, ownership bit is at the same location, thus
137 	 * we can use any
138 	 */
139 	for (i = 0; i < vring->size; i++) {
140 		volatile struct vring_tx_desc *_d = &vring->va[i].tx;
141 
142 		_d->dma.status = TX_DMA_STATUS_DU;
143 	}
144 
145 	wil_dbg_misc(wil, "vring[%d] 0x%p:%pad 0x%p\n", vring->size,
146 		     vring->va, &vring->pa, vring->ctx);
147 
148 	return 0;
149 }
150 
151 static void wil_txdesc_unmap(struct device *dev, struct vring_tx_desc *d,
152 			     struct wil_ctx *ctx)
153 {
154 	dma_addr_t pa = wil_desc_addr(&d->dma.addr);
155 	u16 dmalen = le16_to_cpu(d->dma.length);
156 
157 	switch (ctx->mapped_as) {
158 	case wil_mapped_as_single:
159 		dma_unmap_single(dev, pa, dmalen, DMA_TO_DEVICE);
160 		break;
161 	case wil_mapped_as_page:
162 		dma_unmap_page(dev, pa, dmalen, DMA_TO_DEVICE);
163 		break;
164 	default:
165 		break;
166 	}
167 }
168 
169 static void wil_vring_free(struct wil6210_priv *wil, struct vring *vring,
170 			   int tx)
171 {
172 	struct device *dev = wil_to_dev(wil);
173 	size_t sz = vring->size * sizeof(vring->va[0]);
174 
175 	lockdep_assert_held(&wil->mutex);
176 	if (tx) {
177 		int vring_index = vring - wil->vring_tx;
178 
179 		wil_dbg_misc(wil, "free Tx vring %d [%d] 0x%p:%pad 0x%p\n",
180 			     vring_index, vring->size, vring->va,
181 			     &vring->pa, vring->ctx);
182 	} else {
183 		wil_dbg_misc(wil, "free Rx vring [%d] 0x%p:%pad 0x%p\n",
184 			     vring->size, vring->va,
185 			     &vring->pa, vring->ctx);
186 	}
187 
188 	while (!wil_vring_is_empty(vring)) {
189 		dma_addr_t pa;
190 		u16 dmalen;
191 		struct wil_ctx *ctx;
192 
193 		if (tx) {
194 			struct vring_tx_desc dd, *d = &dd;
195 			volatile struct vring_tx_desc *_d =
196 					&vring->va[vring->swtail].tx;
197 
198 			ctx = &vring->ctx[vring->swtail];
199 			if (!ctx) {
200 				wil_dbg_txrx(wil,
201 					     "ctx(%d) was already completed\n",
202 					     vring->swtail);
203 				vring->swtail = wil_vring_next_tail(vring);
204 				continue;
205 			}
206 			*d = *_d;
207 			wil_txdesc_unmap(dev, d, ctx);
208 			if (ctx->skb)
209 				dev_kfree_skb_any(ctx->skb);
210 			vring->swtail = wil_vring_next_tail(vring);
211 		} else { /* rx */
212 			struct vring_rx_desc dd, *d = &dd;
213 			volatile struct vring_rx_desc *_d =
214 					&vring->va[vring->swhead].rx;
215 
216 			ctx = &vring->ctx[vring->swhead];
217 			*d = *_d;
218 			pa = wil_desc_addr(&d->dma.addr);
219 			dmalen = le16_to_cpu(d->dma.length);
220 			dma_unmap_single(dev, pa, dmalen, DMA_FROM_DEVICE);
221 			kfree_skb(ctx->skb);
222 			wil_vring_advance_head(vring, 1);
223 		}
224 	}
225 	dma_free_coherent(dev, sz, (void *)vring->va, vring->pa);
226 	kfree(vring->ctx);
227 	vring->pa = 0;
228 	vring->va = NULL;
229 	vring->ctx = NULL;
230 }
231 
232 /**
233  * Allocate one skb for Rx VRING
234  *
235  * Safe to call from IRQ
236  */
237 static int wil_vring_alloc_skb(struct wil6210_priv *wil, struct vring *vring,
238 			       u32 i, int headroom)
239 {
240 	struct device *dev = wil_to_dev(wil);
241 	unsigned int sz = mtu_max + ETH_HLEN + wil_rx_snaplen();
242 	struct vring_rx_desc dd, *d = &dd;
243 	volatile struct vring_rx_desc *_d = &vring->va[i].rx;
244 	dma_addr_t pa;
245 	struct sk_buff *skb = dev_alloc_skb(sz + headroom);
246 
247 	if (unlikely(!skb))
248 		return -ENOMEM;
249 
250 	skb_reserve(skb, headroom);
251 	skb_put(skb, sz);
252 
253 	pa = dma_map_single(dev, skb->data, skb->len, DMA_FROM_DEVICE);
254 	if (unlikely(dma_mapping_error(dev, pa))) {
255 		kfree_skb(skb);
256 		return -ENOMEM;
257 	}
258 
259 	d->dma.d0 = RX_DMA_D0_CMD_DMA_RT | RX_DMA_D0_CMD_DMA_IT;
260 	wil_desc_addr_set(&d->dma.addr, pa);
261 	/* ip_length don't care */
262 	/* b11 don't care */
263 	/* error don't care */
264 	d->dma.status = 0; /* BIT(0) should be 0 for HW_OWNED */
265 	d->dma.length = cpu_to_le16(sz);
266 	*_d = *d;
267 	vring->ctx[i].skb = skb;
268 
269 	return 0;
270 }
271 
272 /**
273  * Adds radiotap header
274  *
275  * Any error indicated as "Bad FCS"
276  *
277  * Vendor data for 04:ce:14-1 (Wilocity-1) consists of:
278  *  - Rx descriptor: 32 bytes
279  *  - Phy info
280  */
281 static void wil_rx_add_radiotap_header(struct wil6210_priv *wil,
282 				       struct sk_buff *skb)
283 {
284 	struct wireless_dev *wdev = wil->wdev;
285 	struct wil6210_rtap {
286 		struct ieee80211_radiotap_header rthdr;
287 		/* fields should be in the order of bits in rthdr.it_present */
288 		/* flags */
289 		u8 flags;
290 		/* channel */
291 		__le16 chnl_freq __aligned(2);
292 		__le16 chnl_flags;
293 		/* MCS */
294 		u8 mcs_present;
295 		u8 mcs_flags;
296 		u8 mcs_index;
297 	} __packed;
298 	struct wil6210_rtap_vendor {
299 		struct wil6210_rtap rtap;
300 		/* vendor */
301 		u8 vendor_oui[3] __aligned(2);
302 		u8 vendor_ns;
303 		__le16 vendor_skip;
304 		u8 vendor_data[0];
305 	} __packed;
306 	struct vring_rx_desc *d = wil_skb_rxdesc(skb);
307 	struct wil6210_rtap_vendor *rtap_vendor;
308 	int rtap_len = sizeof(struct wil6210_rtap);
309 	int phy_length = 0; /* phy info header size, bytes */
310 	static char phy_data[128];
311 	struct ieee80211_channel *ch = wdev->preset_chandef.chan;
312 
313 	if (rtap_include_phy_info) {
314 		rtap_len = sizeof(*rtap_vendor) + sizeof(*d);
315 		/* calculate additional length */
316 		if (d->dma.status & RX_DMA_STATUS_PHY_INFO) {
317 			/**
318 			 * PHY info starts from 8-byte boundary
319 			 * there are 8-byte lines, last line may be partially
320 			 * written (HW bug), thus FW configures for last line
321 			 * to be excessive. Driver skips this last line.
322 			 */
323 			int len = min_t(int, 8 + sizeof(phy_data),
324 					wil_rxdesc_phy_length(d));
325 
326 			if (len > 8) {
327 				void *p = skb_tail_pointer(skb);
328 				void *pa = PTR_ALIGN(p, 8);
329 
330 				if (skb_tailroom(skb) >= len + (pa - p)) {
331 					phy_length = len - 8;
332 					memcpy(phy_data, pa, phy_length);
333 				}
334 			}
335 		}
336 		rtap_len += phy_length;
337 	}
338 
339 	if (skb_headroom(skb) < rtap_len &&
340 	    pskb_expand_head(skb, rtap_len, 0, GFP_ATOMIC)) {
341 		wil_err(wil, "Unable to expand headroom to %d\n", rtap_len);
342 		return;
343 	}
344 
345 	rtap_vendor = (void *)skb_push(skb, rtap_len);
346 	memset(rtap_vendor, 0, rtap_len);
347 
348 	rtap_vendor->rtap.rthdr.it_version = PKTHDR_RADIOTAP_VERSION;
349 	rtap_vendor->rtap.rthdr.it_len = cpu_to_le16(rtap_len);
350 	rtap_vendor->rtap.rthdr.it_present = cpu_to_le32(
351 			(1 << IEEE80211_RADIOTAP_FLAGS) |
352 			(1 << IEEE80211_RADIOTAP_CHANNEL) |
353 			(1 << IEEE80211_RADIOTAP_MCS));
354 	if (d->dma.status & RX_DMA_STATUS_ERROR)
355 		rtap_vendor->rtap.flags |= IEEE80211_RADIOTAP_F_BADFCS;
356 
357 	rtap_vendor->rtap.chnl_freq = cpu_to_le16(ch ? ch->center_freq : 58320);
358 	rtap_vendor->rtap.chnl_flags = cpu_to_le16(0);
359 
360 	rtap_vendor->rtap.mcs_present = IEEE80211_RADIOTAP_MCS_HAVE_MCS;
361 	rtap_vendor->rtap.mcs_flags = 0;
362 	rtap_vendor->rtap.mcs_index = wil_rxdesc_mcs(d);
363 
364 	if (rtap_include_phy_info) {
365 		rtap_vendor->rtap.rthdr.it_present |= cpu_to_le32(1 <<
366 				IEEE80211_RADIOTAP_VENDOR_NAMESPACE);
367 		/* OUI for Wilocity 04:ce:14 */
368 		rtap_vendor->vendor_oui[0] = 0x04;
369 		rtap_vendor->vendor_oui[1] = 0xce;
370 		rtap_vendor->vendor_oui[2] = 0x14;
371 		rtap_vendor->vendor_ns = 1;
372 		/* Rx descriptor + PHY data  */
373 		rtap_vendor->vendor_skip = cpu_to_le16(sizeof(*d) +
374 						       phy_length);
375 		memcpy(rtap_vendor->vendor_data, (void *)d, sizeof(*d));
376 		memcpy(rtap_vendor->vendor_data + sizeof(*d), phy_data,
377 		       phy_length);
378 	}
379 }
380 
381 /* similar to ieee80211_ version, but FC contain only 1-st byte */
382 static inline int wil_is_back_req(u8 fc)
383 {
384 	return (fc & (IEEE80211_FCTL_FTYPE | IEEE80211_FCTL_STYPE)) ==
385 	       (IEEE80211_FTYPE_CTL | IEEE80211_STYPE_BACK_REQ);
386 }
387 
388 /**
389  * reap 1 frame from @swhead
390  *
391  * Rx descriptor copied to skb->cb
392  *
393  * Safe to call from IRQ
394  */
395 static struct sk_buff *wil_vring_reap_rx(struct wil6210_priv *wil,
396 					 struct vring *vring)
397 {
398 	struct device *dev = wil_to_dev(wil);
399 	struct net_device *ndev = wil_to_ndev(wil);
400 	volatile struct vring_rx_desc *_d;
401 	struct vring_rx_desc *d;
402 	struct sk_buff *skb;
403 	dma_addr_t pa;
404 	unsigned int snaplen = wil_rx_snaplen();
405 	unsigned int sz = mtu_max + ETH_HLEN + snaplen;
406 	u16 dmalen;
407 	u8 ftype;
408 	int cid;
409 	int i;
410 	struct wil_net_stats *stats;
411 
412 	BUILD_BUG_ON(sizeof(struct vring_rx_desc) > sizeof(skb->cb));
413 
414 again:
415 	if (unlikely(wil_vring_is_empty(vring)))
416 		return NULL;
417 
418 	i = (int)vring->swhead;
419 	_d = &vring->va[i].rx;
420 	if (unlikely(!(_d->dma.status & RX_DMA_STATUS_DU))) {
421 		/* it is not error, we just reached end of Rx done area */
422 		return NULL;
423 	}
424 
425 	skb = vring->ctx[i].skb;
426 	vring->ctx[i].skb = NULL;
427 	wil_vring_advance_head(vring, 1);
428 	if (!skb) {
429 		wil_err(wil, "No Rx skb at [%d]\n", i);
430 		goto again;
431 	}
432 	d = wil_skb_rxdesc(skb);
433 	*d = *_d;
434 	pa = wil_desc_addr(&d->dma.addr);
435 
436 	dma_unmap_single(dev, pa, sz, DMA_FROM_DEVICE);
437 	dmalen = le16_to_cpu(d->dma.length);
438 
439 	trace_wil6210_rx(i, d);
440 	wil_dbg_txrx(wil, "Rx[%3d] : %d bytes\n", i, dmalen);
441 	wil_hex_dump_txrx("RxD ", DUMP_PREFIX_NONE, 32, 4,
442 			  (const void *)d, sizeof(*d), false);
443 
444 	cid = wil_rxdesc_cid(d);
445 	stats = &wil->sta[cid].stats;
446 
447 	if (unlikely(dmalen > sz)) {
448 		wil_err(wil, "Rx size too large: %d bytes!\n", dmalen);
449 		stats->rx_large_frame++;
450 		kfree_skb(skb);
451 		goto again;
452 	}
453 	skb_trim(skb, dmalen);
454 
455 	prefetch(skb->data);
456 
457 	wil_hex_dump_txrx("Rx ", DUMP_PREFIX_OFFSET, 16, 1,
458 			  skb->data, skb_headlen(skb), false);
459 
460 	stats->last_mcs_rx = wil_rxdesc_mcs(d);
461 	if (stats->last_mcs_rx < ARRAY_SIZE(stats->rx_per_mcs))
462 		stats->rx_per_mcs[stats->last_mcs_rx]++;
463 
464 	/* use radiotap header only if required */
465 	if (ndev->type == ARPHRD_IEEE80211_RADIOTAP)
466 		wil_rx_add_radiotap_header(wil, skb);
467 
468 	/* no extra checks if in sniffer mode */
469 	if (ndev->type != ARPHRD_ETHER)
470 		return skb;
471 	/* Non-data frames may be delivered through Rx DMA channel (ex: BAR)
472 	 * Driver should recognize it by frame type, that is found
473 	 * in Rx descriptor. If type is not data, it is 802.11 frame as is
474 	 */
475 	ftype = wil_rxdesc_ftype(d) << 2;
476 	if (unlikely(ftype != IEEE80211_FTYPE_DATA)) {
477 		u8 fc1 = wil_rxdesc_fc1(d);
478 		int mid = wil_rxdesc_mid(d);
479 		int tid = wil_rxdesc_tid(d);
480 		u16 seq = wil_rxdesc_seq(d);
481 
482 		wil_dbg_txrx(wil,
483 			     "Non-data frame FC[7:0] 0x%02x MID %d CID %d TID %d Seq 0x%03x\n",
484 			     fc1, mid, cid, tid, seq);
485 		stats->rx_non_data_frame++;
486 		if (wil_is_back_req(fc1)) {
487 			wil_dbg_txrx(wil,
488 				     "BAR: MID %d CID %d TID %d Seq 0x%03x\n",
489 				     mid, cid, tid, seq);
490 			wil_rx_bar(wil, cid, tid, seq);
491 		} else {
492 			/* print again all info. One can enable only this
493 			 * without overhead for printing every Rx frame
494 			 */
495 			wil_dbg_txrx(wil,
496 				     "Unhandled non-data frame FC[7:0] 0x%02x MID %d CID %d TID %d Seq 0x%03x\n",
497 				     fc1, mid, cid, tid, seq);
498 			wil_hex_dump_txrx("RxD ", DUMP_PREFIX_NONE, 32, 4,
499 					  (const void *)d, sizeof(*d), false);
500 			wil_hex_dump_txrx("Rx ", DUMP_PREFIX_OFFSET, 16, 1,
501 					  skb->data, skb_headlen(skb), false);
502 		}
503 		kfree_skb(skb);
504 		goto again;
505 	}
506 
507 	if (unlikely(skb->len < ETH_HLEN + snaplen)) {
508 		wil_err(wil, "Short frame, len = %d\n", skb->len);
509 		stats->rx_short_frame++;
510 		kfree_skb(skb);
511 		goto again;
512 	}
513 
514 	/* L4 IDENT is on when HW calculated checksum, check status
515 	 * and in case of error drop the packet
516 	 * higher stack layers will handle retransmission (if required)
517 	 */
518 	if (likely(d->dma.status & RX_DMA_STATUS_L4I)) {
519 		/* L4 protocol identified, csum calculated */
520 		if (likely((d->dma.error & RX_DMA_ERROR_L4_ERR) == 0))
521 			skb->ip_summed = CHECKSUM_UNNECESSARY;
522 		/* If HW reports bad checksum, let IP stack re-check it
523 		 * For example, HW don't understand Microsoft IP stack that
524 		 * mis-calculates TCP checksum - if it should be 0x0,
525 		 * it writes 0xffff in violation of RFC 1624
526 		 */
527 	}
528 
529 	if (snaplen) {
530 		/* Packet layout
531 		 * +-------+-------+---------+------------+------+
532 		 * | SA(6) | DA(6) | SNAP(6) | ETHTYPE(2) | DATA |
533 		 * +-------+-------+---------+------------+------+
534 		 * Need to remove SNAP, shifting SA and DA forward
535 		 */
536 		memmove(skb->data + snaplen, skb->data, 2 * ETH_ALEN);
537 		skb_pull(skb, snaplen);
538 	}
539 
540 	return skb;
541 }
542 
543 /**
544  * allocate and fill up to @count buffers in rx ring
545  * buffers posted at @swtail
546  */
547 static int wil_rx_refill(struct wil6210_priv *wil, int count)
548 {
549 	struct net_device *ndev = wil_to_ndev(wil);
550 	struct vring *v = &wil->vring_rx;
551 	u32 next_tail;
552 	int rc = 0;
553 	int headroom = ndev->type == ARPHRD_IEEE80211_RADIOTAP ?
554 			WIL6210_RTAP_SIZE : 0;
555 
556 	for (; next_tail = wil_vring_next_tail(v),
557 			(next_tail != v->swhead) && (count-- > 0);
558 			v->swtail = next_tail) {
559 		rc = wil_vring_alloc_skb(wil, v, v->swtail, headroom);
560 		if (unlikely(rc)) {
561 			wil_err(wil, "Error %d in wil_rx_refill[%d]\n",
562 				rc, v->swtail);
563 			break;
564 		}
565 	}
566 
567 	/* make sure all writes to descriptors (shared memory) are done before
568 	 * committing them to HW
569 	 */
570 	wmb();
571 
572 	wil_w(wil, v->hwtail, v->swtail);
573 
574 	return rc;
575 }
576 
577 /**
578  * reverse_memcmp - Compare two areas of memory, in reverse order
579  * @cs: One area of memory
580  * @ct: Another area of memory
581  * @count: The size of the area.
582  *
583  * Cut'n'paste from original memcmp (see lib/string.c)
584  * with minimal modifications
585  */
586 static int reverse_memcmp(const void *cs, const void *ct, size_t count)
587 {
588 	const unsigned char *su1, *su2;
589 	int res = 0;
590 
591 	for (su1 = cs + count - 1, su2 = ct + count - 1; count > 0;
592 	     --su1, --su2, count--) {
593 		res = *su1 - *su2;
594 		if (res)
595 			break;
596 	}
597 	return res;
598 }
599 
600 static int wil_rx_crypto_check(struct wil6210_priv *wil, struct sk_buff *skb)
601 {
602 	struct vring_rx_desc *d = wil_skb_rxdesc(skb);
603 	int cid = wil_rxdesc_cid(d);
604 	int tid = wil_rxdesc_tid(d);
605 	int key_id = wil_rxdesc_key_id(d);
606 	int mc = wil_rxdesc_mcast(d);
607 	struct wil_sta_info *s = &wil->sta[cid];
608 	struct wil_tid_crypto_rx *c = mc ? &s->group_crypto_rx :
609 				      &s->tid_crypto_rx[tid];
610 	struct wil_tid_crypto_rx_single *cc = &c->key_id[key_id];
611 	const u8 *pn = (u8 *)&d->mac.pn_15_0;
612 
613 	if (!cc->key_set) {
614 		wil_err_ratelimited(wil,
615 				    "Key missing. CID %d TID %d MCast %d KEY_ID %d\n",
616 				    cid, tid, mc, key_id);
617 		return -EINVAL;
618 	}
619 
620 	if (reverse_memcmp(pn, cc->pn, IEEE80211_GCMP_PN_LEN) <= 0) {
621 		wil_err_ratelimited(wil,
622 				    "Replay attack. CID %d TID %d MCast %d KEY_ID %d PN %6phN last %6phN\n",
623 				    cid, tid, mc, key_id, pn, cc->pn);
624 		return -EINVAL;
625 	}
626 	memcpy(cc->pn, pn, IEEE80211_GCMP_PN_LEN);
627 
628 	return 0;
629 }
630 
631 /*
632  * Pass Rx packet to the netif. Update statistics.
633  * Called in softirq context (NAPI poll).
634  */
635 void wil_netif_rx_any(struct sk_buff *skb, struct net_device *ndev)
636 {
637 	gro_result_t rc = GRO_NORMAL;
638 	struct wil6210_priv *wil = ndev_to_wil(ndev);
639 	struct wireless_dev *wdev = wil_to_wdev(wil);
640 	unsigned int len = skb->len;
641 	struct vring_rx_desc *d = wil_skb_rxdesc(skb);
642 	int cid = wil_rxdesc_cid(d); /* always 0..7, no need to check */
643 	int security = wil_rxdesc_security(d);
644 	struct ethhdr *eth = (void *)skb->data;
645 	/* here looking for DA, not A1, thus Rxdesc's 'mcast' indication
646 	 * is not suitable, need to look at data
647 	 */
648 	int mcast = is_multicast_ether_addr(eth->h_dest);
649 	struct wil_net_stats *stats = &wil->sta[cid].stats;
650 	struct sk_buff *xmit_skb = NULL;
651 	static const char * const gro_res_str[] = {
652 		[GRO_MERGED]		= "GRO_MERGED",
653 		[GRO_MERGED_FREE]	= "GRO_MERGED_FREE",
654 		[GRO_HELD]		= "GRO_HELD",
655 		[GRO_NORMAL]		= "GRO_NORMAL",
656 		[GRO_DROP]		= "GRO_DROP",
657 	};
658 
659 	if (ndev->features & NETIF_F_RXHASH)
660 		/* fake L4 to ensure it won't be re-calculated later
661 		 * set hash to any non-zero value to activate rps
662 		 * mechanism, core will be chosen according
663 		 * to user-level rps configuration.
664 		 */
665 		skb_set_hash(skb, 1, PKT_HASH_TYPE_L4);
666 
667 	skb_orphan(skb);
668 
669 	if (security && (wil_rx_crypto_check(wil, skb) != 0)) {
670 		rc = GRO_DROP;
671 		dev_kfree_skb(skb);
672 		stats->rx_replay++;
673 		goto stats;
674 	}
675 
676 	if (wdev->iftype == NL80211_IFTYPE_AP && !wil->ap_isolate) {
677 		if (mcast) {
678 			/* send multicast frames both to higher layers in
679 			 * local net stack and back to the wireless medium
680 			 */
681 			xmit_skb = skb_copy(skb, GFP_ATOMIC);
682 		} else {
683 			int xmit_cid = wil_find_cid(wil, eth->h_dest);
684 
685 			if (xmit_cid >= 0) {
686 				/* The destination station is associated to
687 				 * this AP (in this VLAN), so send the frame
688 				 * directly to it and do not pass it to local
689 				 * net stack.
690 				 */
691 				xmit_skb = skb;
692 				skb = NULL;
693 			}
694 		}
695 	}
696 	if (xmit_skb) {
697 		/* Send to wireless media and increase priority by 256 to
698 		 * keep the received priority instead of reclassifying
699 		 * the frame (see cfg80211_classify8021d).
700 		 */
701 		xmit_skb->dev = ndev;
702 		xmit_skb->priority += 256;
703 		xmit_skb->protocol = htons(ETH_P_802_3);
704 		skb_reset_network_header(xmit_skb);
705 		skb_reset_mac_header(xmit_skb);
706 		wil_dbg_txrx(wil, "Rx -> Tx %d bytes\n", len);
707 		dev_queue_xmit(xmit_skb);
708 	}
709 
710 	if (skb) { /* deliver to local stack */
711 
712 		skb->protocol = eth_type_trans(skb, ndev);
713 		rc = napi_gro_receive(&wil->napi_rx, skb);
714 		wil_dbg_txrx(wil, "Rx complete %d bytes => %s\n",
715 			     len, gro_res_str[rc]);
716 	}
717 stats:
718 	/* statistics. rc set to GRO_NORMAL for AP bridging */
719 	if (unlikely(rc == GRO_DROP)) {
720 		ndev->stats.rx_dropped++;
721 		stats->rx_dropped++;
722 		wil_dbg_txrx(wil, "Rx drop %d bytes\n", len);
723 	} else {
724 		ndev->stats.rx_packets++;
725 		stats->rx_packets++;
726 		ndev->stats.rx_bytes += len;
727 		stats->rx_bytes += len;
728 		if (mcast)
729 			ndev->stats.multicast++;
730 	}
731 }
732 
733 /**
734  * Proceed all completed skb's from Rx VRING
735  *
736  * Safe to call from NAPI poll, i.e. softirq with interrupts enabled
737  */
738 void wil_rx_handle(struct wil6210_priv *wil, int *quota)
739 {
740 	struct net_device *ndev = wil_to_ndev(wil);
741 	struct vring *v = &wil->vring_rx;
742 	struct sk_buff *skb;
743 
744 	if (unlikely(!v->va)) {
745 		wil_err(wil, "Rx IRQ while Rx not yet initialized\n");
746 		return;
747 	}
748 	wil_dbg_txrx(wil, "rx_handle\n");
749 	while ((*quota > 0) && (NULL != (skb = wil_vring_reap_rx(wil, v)))) {
750 		(*quota)--;
751 
752 		if (wil->wdev->iftype == NL80211_IFTYPE_MONITOR) {
753 			skb->dev = ndev;
754 			skb_reset_mac_header(skb);
755 			skb->ip_summed = CHECKSUM_UNNECESSARY;
756 			skb->pkt_type = PACKET_OTHERHOST;
757 			skb->protocol = htons(ETH_P_802_2);
758 			wil_netif_rx_any(skb, ndev);
759 		} else {
760 			wil_rx_reorder(wil, skb);
761 		}
762 	}
763 	wil_rx_refill(wil, v->size);
764 }
765 
766 int wil_rx_init(struct wil6210_priv *wil, u16 size)
767 {
768 	struct vring *vring = &wil->vring_rx;
769 	int rc;
770 
771 	wil_dbg_misc(wil, "rx_init\n");
772 
773 	if (vring->va) {
774 		wil_err(wil, "Rx ring already allocated\n");
775 		return -EINVAL;
776 	}
777 
778 	vring->size = size;
779 	rc = wil_vring_alloc(wil, vring);
780 	if (rc)
781 		return rc;
782 
783 	rc = wmi_rx_chain_add(wil, vring);
784 	if (rc)
785 		goto err_free;
786 
787 	rc = wil_rx_refill(wil, vring->size);
788 	if (rc)
789 		goto err_free;
790 
791 	return 0;
792  err_free:
793 	wil_vring_free(wil, vring, 0);
794 
795 	return rc;
796 }
797 
798 void wil_rx_fini(struct wil6210_priv *wil)
799 {
800 	struct vring *vring = &wil->vring_rx;
801 
802 	wil_dbg_misc(wil, "rx_fini\n");
803 
804 	if (vring->va)
805 		wil_vring_free(wil, vring, 0);
806 }
807 
808 static inline void wil_tx_data_init(struct vring_tx_data *txdata)
809 {
810 	spin_lock_bh(&txdata->lock);
811 	txdata->dot1x_open = 0;
812 	txdata->enabled = 0;
813 	txdata->idle = 0;
814 	txdata->last_idle = 0;
815 	txdata->begin = 0;
816 	txdata->agg_wsize = 0;
817 	txdata->agg_timeout = 0;
818 	txdata->agg_amsdu = 0;
819 	txdata->addba_in_progress = false;
820 	spin_unlock_bh(&txdata->lock);
821 }
822 
823 int wil_vring_init_tx(struct wil6210_priv *wil, int id, int size,
824 		      int cid, int tid)
825 {
826 	int rc;
827 	struct wmi_vring_cfg_cmd cmd = {
828 		.action = cpu_to_le32(WMI_VRING_CMD_ADD),
829 		.vring_cfg = {
830 			.tx_sw_ring = {
831 				.max_mpdu_size =
832 					cpu_to_le16(wil_mtu2macbuf(mtu_max)),
833 				.ring_size = cpu_to_le16(size),
834 			},
835 			.ringid = id,
836 			.cidxtid = mk_cidxtid(cid, tid),
837 			.encap_trans_type = WMI_VRING_ENC_TYPE_802_3,
838 			.mac_ctrl = 0,
839 			.to_resolution = 0,
840 			.agg_max_wsize = 0,
841 			.schd_params = {
842 				.priority = cpu_to_le16(0),
843 				.timeslot_us = cpu_to_le16(0xfff),
844 			},
845 		},
846 	};
847 	struct {
848 		struct wmi_cmd_hdr wmi;
849 		struct wmi_vring_cfg_done_event cmd;
850 	} __packed reply;
851 	struct vring *vring = &wil->vring_tx[id];
852 	struct vring_tx_data *txdata = &wil->vring_tx_data[id];
853 
854 	wil_dbg_misc(wil, "vring_init_tx: max_mpdu_size %d\n",
855 		     cmd.vring_cfg.tx_sw_ring.max_mpdu_size);
856 	lockdep_assert_held(&wil->mutex);
857 
858 	if (vring->va) {
859 		wil_err(wil, "Tx ring [%d] already allocated\n", id);
860 		rc = -EINVAL;
861 		goto out;
862 	}
863 
864 	wil_tx_data_init(txdata);
865 	vring->size = size;
866 	rc = wil_vring_alloc(wil, vring);
867 	if (rc)
868 		goto out;
869 
870 	wil->vring2cid_tid[id][0] = cid;
871 	wil->vring2cid_tid[id][1] = tid;
872 
873 	cmd.vring_cfg.tx_sw_ring.ring_mem_base = cpu_to_le64(vring->pa);
874 
875 	if (!wil->privacy)
876 		txdata->dot1x_open = true;
877 	rc = wmi_call(wil, WMI_VRING_CFG_CMDID, &cmd, sizeof(cmd),
878 		      WMI_VRING_CFG_DONE_EVENTID, &reply, sizeof(reply), 100);
879 	if (rc)
880 		goto out_free;
881 
882 	if (reply.cmd.status != WMI_FW_STATUS_SUCCESS) {
883 		wil_err(wil, "Tx config failed, status 0x%02x\n",
884 			reply.cmd.status);
885 		rc = -EINVAL;
886 		goto out_free;
887 	}
888 
889 	spin_lock_bh(&txdata->lock);
890 	vring->hwtail = le32_to_cpu(reply.cmd.tx_vring_tail_ptr);
891 	txdata->enabled = 1;
892 	spin_unlock_bh(&txdata->lock);
893 
894 	if (txdata->dot1x_open && (agg_wsize >= 0))
895 		wil_addba_tx_request(wil, id, agg_wsize);
896 
897 	return 0;
898  out_free:
899 	spin_lock_bh(&txdata->lock);
900 	txdata->dot1x_open = false;
901 	txdata->enabled = 0;
902 	spin_unlock_bh(&txdata->lock);
903 	wil_vring_free(wil, vring, 1);
904 	wil->vring2cid_tid[id][0] = WIL6210_MAX_CID;
905 	wil->vring2cid_tid[id][1] = 0;
906 
907  out:
908 
909 	return rc;
910 }
911 
912 int wil_vring_init_bcast(struct wil6210_priv *wil, int id, int size)
913 {
914 	int rc;
915 	struct wmi_bcast_vring_cfg_cmd cmd = {
916 		.action = cpu_to_le32(WMI_VRING_CMD_ADD),
917 		.vring_cfg = {
918 			.tx_sw_ring = {
919 				.max_mpdu_size =
920 					cpu_to_le16(wil_mtu2macbuf(mtu_max)),
921 				.ring_size = cpu_to_le16(size),
922 			},
923 			.ringid = id,
924 			.encap_trans_type = WMI_VRING_ENC_TYPE_802_3,
925 		},
926 	};
927 	struct {
928 		struct wmi_cmd_hdr wmi;
929 		struct wmi_vring_cfg_done_event cmd;
930 	} __packed reply;
931 	struct vring *vring = &wil->vring_tx[id];
932 	struct vring_tx_data *txdata = &wil->vring_tx_data[id];
933 
934 	wil_dbg_misc(wil, "vring_init_bcast: max_mpdu_size %d\n",
935 		     cmd.vring_cfg.tx_sw_ring.max_mpdu_size);
936 	lockdep_assert_held(&wil->mutex);
937 
938 	if (vring->va) {
939 		wil_err(wil, "Tx ring [%d] already allocated\n", id);
940 		rc = -EINVAL;
941 		goto out;
942 	}
943 
944 	wil_tx_data_init(txdata);
945 	vring->size = size;
946 	rc = wil_vring_alloc(wil, vring);
947 	if (rc)
948 		goto out;
949 
950 	wil->vring2cid_tid[id][0] = WIL6210_MAX_CID; /* CID */
951 	wil->vring2cid_tid[id][1] = 0; /* TID */
952 
953 	cmd.vring_cfg.tx_sw_ring.ring_mem_base = cpu_to_le64(vring->pa);
954 
955 	if (!wil->privacy)
956 		txdata->dot1x_open = true;
957 	rc = wmi_call(wil, WMI_BCAST_VRING_CFG_CMDID, &cmd, sizeof(cmd),
958 		      WMI_VRING_CFG_DONE_EVENTID, &reply, sizeof(reply), 100);
959 	if (rc)
960 		goto out_free;
961 
962 	if (reply.cmd.status != WMI_FW_STATUS_SUCCESS) {
963 		wil_err(wil, "Tx config failed, status 0x%02x\n",
964 			reply.cmd.status);
965 		rc = -EINVAL;
966 		goto out_free;
967 	}
968 
969 	spin_lock_bh(&txdata->lock);
970 	vring->hwtail = le32_to_cpu(reply.cmd.tx_vring_tail_ptr);
971 	txdata->enabled = 1;
972 	spin_unlock_bh(&txdata->lock);
973 
974 	return 0;
975  out_free:
976 	spin_lock_bh(&txdata->lock);
977 	txdata->enabled = 0;
978 	txdata->dot1x_open = false;
979 	spin_unlock_bh(&txdata->lock);
980 	wil_vring_free(wil, vring, 1);
981  out:
982 
983 	return rc;
984 }
985 
986 void wil_vring_fini_tx(struct wil6210_priv *wil, int id)
987 {
988 	struct vring *vring = &wil->vring_tx[id];
989 	struct vring_tx_data *txdata = &wil->vring_tx_data[id];
990 
991 	lockdep_assert_held(&wil->mutex);
992 
993 	if (!vring->va)
994 		return;
995 
996 	wil_dbg_misc(wil, "vring_fini_tx: id=%d\n", id);
997 
998 	spin_lock_bh(&txdata->lock);
999 	txdata->dot1x_open = false;
1000 	txdata->enabled = 0; /* no Tx can be in progress or start anew */
1001 	spin_unlock_bh(&txdata->lock);
1002 	/* napi_synchronize waits for completion of the current NAPI but will
1003 	 * not prevent the next NAPI run.
1004 	 * Add a memory barrier to guarantee that txdata->enabled is zeroed
1005 	 * before napi_synchronize so that the next scheduled NAPI will not
1006 	 * handle this vring
1007 	 */
1008 	wmb();
1009 	/* make sure NAPI won't touch this vring */
1010 	if (test_bit(wil_status_napi_en, wil->status))
1011 		napi_synchronize(&wil->napi_tx);
1012 
1013 	wil_vring_free(wil, vring, 1);
1014 }
1015 
1016 static struct vring *wil_find_tx_ucast(struct wil6210_priv *wil,
1017 				       struct sk_buff *skb)
1018 {
1019 	int i;
1020 	struct ethhdr *eth = (void *)skb->data;
1021 	int cid = wil_find_cid(wil, eth->h_dest);
1022 
1023 	if (cid < 0)
1024 		return NULL;
1025 
1026 	/* TODO: fix for multiple TID */
1027 	for (i = 0; i < ARRAY_SIZE(wil->vring2cid_tid); i++) {
1028 		if (!wil->vring_tx_data[i].dot1x_open &&
1029 		    (skb->protocol != cpu_to_be16(ETH_P_PAE)))
1030 			continue;
1031 		if (wil->vring2cid_tid[i][0] == cid) {
1032 			struct vring *v = &wil->vring_tx[i];
1033 			struct vring_tx_data *txdata = &wil->vring_tx_data[i];
1034 
1035 			wil_dbg_txrx(wil, "find_tx_ucast: (%pM) -> [%d]\n",
1036 				     eth->h_dest, i);
1037 			if (v->va && txdata->enabled) {
1038 				return v;
1039 			} else {
1040 				wil_dbg_txrx(wil,
1041 					     "find_tx_ucast: vring[%d] not valid\n",
1042 					     i);
1043 				return NULL;
1044 			}
1045 		}
1046 	}
1047 
1048 	return NULL;
1049 }
1050 
1051 static int wil_tx_vring(struct wil6210_priv *wil, struct vring *vring,
1052 			struct sk_buff *skb);
1053 
1054 static struct vring *wil_find_tx_vring_sta(struct wil6210_priv *wil,
1055 					   struct sk_buff *skb)
1056 {
1057 	struct vring *v;
1058 	int i;
1059 	u8 cid;
1060 	struct vring_tx_data *txdata;
1061 
1062 	/* In the STA mode, it is expected to have only 1 VRING
1063 	 * for the AP we connected to.
1064 	 * find 1-st vring eligible for this skb and use it.
1065 	 */
1066 	for (i = 0; i < WIL6210_MAX_TX_RINGS; i++) {
1067 		v = &wil->vring_tx[i];
1068 		txdata = &wil->vring_tx_data[i];
1069 		if (!v->va || !txdata->enabled)
1070 			continue;
1071 
1072 		cid = wil->vring2cid_tid[i][0];
1073 		if (cid >= WIL6210_MAX_CID) /* skip BCAST */
1074 			continue;
1075 
1076 		if (!wil->vring_tx_data[i].dot1x_open &&
1077 		    (skb->protocol != cpu_to_be16(ETH_P_PAE)))
1078 			continue;
1079 
1080 		wil_dbg_txrx(wil, "Tx -> ring %d\n", i);
1081 
1082 		return v;
1083 	}
1084 
1085 	wil_dbg_txrx(wil, "Tx while no vrings active?\n");
1086 
1087 	return NULL;
1088 }
1089 
1090 /* Use one of 2 strategies:
1091  *
1092  * 1. New (real broadcast):
1093  *    use dedicated broadcast vring
1094  * 2. Old (pseudo-DMS):
1095  *    Find 1-st vring and return it;
1096  *    duplicate skb and send it to other active vrings;
1097  *    in all cases override dest address to unicast peer's address
1098  * Use old strategy when new is not supported yet:
1099  *  - for PBSS
1100  */
1101 static struct vring *wil_find_tx_bcast_1(struct wil6210_priv *wil,
1102 					 struct sk_buff *skb)
1103 {
1104 	struct vring *v;
1105 	struct vring_tx_data *txdata;
1106 	int i = wil->bcast_vring;
1107 
1108 	if (i < 0)
1109 		return NULL;
1110 	v = &wil->vring_tx[i];
1111 	txdata = &wil->vring_tx_data[i];
1112 	if (!v->va || !txdata->enabled)
1113 		return NULL;
1114 	if (!wil->vring_tx_data[i].dot1x_open &&
1115 	    (skb->protocol != cpu_to_be16(ETH_P_PAE)))
1116 		return NULL;
1117 
1118 	return v;
1119 }
1120 
1121 static void wil_set_da_for_vring(struct wil6210_priv *wil,
1122 				 struct sk_buff *skb, int vring_index)
1123 {
1124 	struct ethhdr *eth = (void *)skb->data;
1125 	int cid = wil->vring2cid_tid[vring_index][0];
1126 
1127 	ether_addr_copy(eth->h_dest, wil->sta[cid].addr);
1128 }
1129 
1130 static struct vring *wil_find_tx_bcast_2(struct wil6210_priv *wil,
1131 					 struct sk_buff *skb)
1132 {
1133 	struct vring *v, *v2;
1134 	struct sk_buff *skb2;
1135 	int i;
1136 	u8 cid;
1137 	struct ethhdr *eth = (void *)skb->data;
1138 	char *src = eth->h_source;
1139 	struct vring_tx_data *txdata;
1140 
1141 	/* find 1-st vring eligible for data */
1142 	for (i = 0; i < WIL6210_MAX_TX_RINGS; i++) {
1143 		v = &wil->vring_tx[i];
1144 		txdata = &wil->vring_tx_data[i];
1145 		if (!v->va || !txdata->enabled)
1146 			continue;
1147 
1148 		cid = wil->vring2cid_tid[i][0];
1149 		if (cid >= WIL6210_MAX_CID) /* skip BCAST */
1150 			continue;
1151 		if (!wil->vring_tx_data[i].dot1x_open &&
1152 		    (skb->protocol != cpu_to_be16(ETH_P_PAE)))
1153 			continue;
1154 
1155 		/* don't Tx back to source when re-routing Rx->Tx at the AP */
1156 		if (0 == memcmp(wil->sta[cid].addr, src, ETH_ALEN))
1157 			continue;
1158 
1159 		goto found;
1160 	}
1161 
1162 	wil_dbg_txrx(wil, "Tx while no vrings active?\n");
1163 
1164 	return NULL;
1165 
1166 found:
1167 	wil_dbg_txrx(wil, "BCAST -> ring %d\n", i);
1168 	wil_set_da_for_vring(wil, skb, i);
1169 
1170 	/* find other active vrings and duplicate skb for each */
1171 	for (i++; i < WIL6210_MAX_TX_RINGS; i++) {
1172 		v2 = &wil->vring_tx[i];
1173 		if (!v2->va)
1174 			continue;
1175 		cid = wil->vring2cid_tid[i][0];
1176 		if (cid >= WIL6210_MAX_CID) /* skip BCAST */
1177 			continue;
1178 		if (!wil->vring_tx_data[i].dot1x_open &&
1179 		    (skb->protocol != cpu_to_be16(ETH_P_PAE)))
1180 			continue;
1181 
1182 		if (0 == memcmp(wil->sta[cid].addr, src, ETH_ALEN))
1183 			continue;
1184 
1185 		skb2 = skb_copy(skb, GFP_ATOMIC);
1186 		if (skb2) {
1187 			wil_dbg_txrx(wil, "BCAST DUP -> ring %d\n", i);
1188 			wil_set_da_for_vring(wil, skb2, i);
1189 			wil_tx_vring(wil, v2, skb2);
1190 		} else {
1191 			wil_err(wil, "skb_copy failed\n");
1192 		}
1193 	}
1194 
1195 	return v;
1196 }
1197 
1198 static int wil_tx_desc_map(struct vring_tx_desc *d, dma_addr_t pa, u32 len,
1199 			   int vring_index)
1200 {
1201 	wil_desc_addr_set(&d->dma.addr, pa);
1202 	d->dma.ip_length = 0;
1203 	/* 0..6: mac_length; 7:ip_version 0-IP6 1-IP4*/
1204 	d->dma.b11 = 0/*14 | BIT(7)*/;
1205 	d->dma.error = 0;
1206 	d->dma.status = 0; /* BIT(0) should be 0 for HW_OWNED */
1207 	d->dma.length = cpu_to_le16((u16)len);
1208 	d->dma.d0 = (vring_index << DMA_CFG_DESC_TX_0_QID_POS);
1209 	d->mac.d[0] = 0;
1210 	d->mac.d[1] = 0;
1211 	d->mac.d[2] = 0;
1212 	d->mac.ucode_cmd = 0;
1213 	/* translation type:  0 - bypass; 1 - 802.3; 2 - native wifi */
1214 	d->mac.d[2] = BIT(MAC_CFG_DESC_TX_2_SNAP_HDR_INSERTION_EN_POS) |
1215 		      (1 << MAC_CFG_DESC_TX_2_L2_TRANSLATION_TYPE_POS);
1216 
1217 	return 0;
1218 }
1219 
1220 static inline
1221 void wil_tx_desc_set_nr_frags(struct vring_tx_desc *d, int nr_frags)
1222 {
1223 	d->mac.d[2] |= (nr_frags << MAC_CFG_DESC_TX_2_NUM_OF_DESCRIPTORS_POS);
1224 }
1225 
1226 /**
1227  * Sets the descriptor @d up for csum and/or TSO offloading. The corresponding
1228  * @skb is used to obtain the protocol and headers length.
1229  * @tso_desc_type is a descriptor type for TSO: 0 - a header, 1 - first data,
1230  * 2 - middle, 3 - last descriptor.
1231  */
1232 
1233 static void wil_tx_desc_offload_setup_tso(struct vring_tx_desc *d,
1234 					  struct sk_buff *skb,
1235 					  int tso_desc_type, bool is_ipv4,
1236 					  int tcp_hdr_len, int skb_net_hdr_len)
1237 {
1238 	d->dma.b11 = ETH_HLEN; /* MAC header length */
1239 	d->dma.b11 |= is_ipv4 << DMA_CFG_DESC_TX_OFFLOAD_CFG_L3T_IPV4_POS;
1240 
1241 	d->dma.d0 |= (2 << DMA_CFG_DESC_TX_0_L4_TYPE_POS);
1242 	/* L4 header len: TCP header length */
1243 	d->dma.d0 |= (tcp_hdr_len & DMA_CFG_DESC_TX_0_L4_LENGTH_MSK);
1244 
1245 	/* Setup TSO: bit and desc type */
1246 	d->dma.d0 |= (BIT(DMA_CFG_DESC_TX_0_TCP_SEG_EN_POS)) |
1247 		(tso_desc_type << DMA_CFG_DESC_TX_0_SEGMENT_BUF_DETAILS_POS);
1248 	d->dma.d0 |= (is_ipv4 << DMA_CFG_DESC_TX_0_IPV4_CHECKSUM_EN_POS);
1249 
1250 	d->dma.ip_length = skb_net_hdr_len;
1251 	/* Enable TCP/UDP checksum */
1252 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_TCP_UDP_CHECKSUM_EN_POS);
1253 	/* Calculate pseudo-header */
1254 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_PSEUDO_HEADER_CALC_EN_POS);
1255 }
1256 
1257 /**
1258  * Sets the descriptor @d up for csum. The corresponding
1259  * @skb is used to obtain the protocol and headers length.
1260  * Returns the protocol: 0 - not TCP, 1 - TCPv4, 2 - TCPv6.
1261  * Note, if d==NULL, the function only returns the protocol result.
1262  *
1263  * It is very similar to previous wil_tx_desc_offload_setup_tso. This
1264  * is "if unrolling" to optimize the critical path.
1265  */
1266 
1267 static int wil_tx_desc_offload_setup(struct vring_tx_desc *d,
1268 				     struct sk_buff *skb){
1269 	int protocol;
1270 
1271 	if (skb->ip_summed != CHECKSUM_PARTIAL)
1272 		return 0;
1273 
1274 	d->dma.b11 = ETH_HLEN; /* MAC header length */
1275 
1276 	switch (skb->protocol) {
1277 	case cpu_to_be16(ETH_P_IP):
1278 		protocol = ip_hdr(skb)->protocol;
1279 		d->dma.b11 |= BIT(DMA_CFG_DESC_TX_OFFLOAD_CFG_L3T_IPV4_POS);
1280 		break;
1281 	case cpu_to_be16(ETH_P_IPV6):
1282 		protocol = ipv6_hdr(skb)->nexthdr;
1283 		break;
1284 	default:
1285 		return -EINVAL;
1286 	}
1287 
1288 	switch (protocol) {
1289 	case IPPROTO_TCP:
1290 		d->dma.d0 |= (2 << DMA_CFG_DESC_TX_0_L4_TYPE_POS);
1291 		/* L4 header len: TCP header length */
1292 		d->dma.d0 |=
1293 		(tcp_hdrlen(skb) & DMA_CFG_DESC_TX_0_L4_LENGTH_MSK);
1294 		break;
1295 	case IPPROTO_UDP:
1296 		/* L4 header len: UDP header length */
1297 		d->dma.d0 |=
1298 		(sizeof(struct udphdr) & DMA_CFG_DESC_TX_0_L4_LENGTH_MSK);
1299 		break;
1300 	default:
1301 		return -EINVAL;
1302 	}
1303 
1304 	d->dma.ip_length = skb_network_header_len(skb);
1305 	/* Enable TCP/UDP checksum */
1306 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_TCP_UDP_CHECKSUM_EN_POS);
1307 	/* Calculate pseudo-header */
1308 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_PSEUDO_HEADER_CALC_EN_POS);
1309 
1310 	return 0;
1311 }
1312 
1313 static inline void wil_tx_last_desc(struct vring_tx_desc *d)
1314 {
1315 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_CMD_EOP_POS) |
1316 	      BIT(DMA_CFG_DESC_TX_0_CMD_MARK_WB_POS) |
1317 	      BIT(DMA_CFG_DESC_TX_0_CMD_DMA_IT_POS);
1318 }
1319 
1320 static inline void wil_set_tx_desc_last_tso(volatile struct vring_tx_desc *d)
1321 {
1322 	d->dma.d0 |= wil_tso_type_lst <<
1323 		  DMA_CFG_DESC_TX_0_SEGMENT_BUF_DETAILS_POS;
1324 }
1325 
1326 static int __wil_tx_vring_tso(struct wil6210_priv *wil, struct vring *vring,
1327 			      struct sk_buff *skb)
1328 {
1329 	struct device *dev = wil_to_dev(wil);
1330 
1331 	/* point to descriptors in shared memory */
1332 	volatile struct vring_tx_desc *_desc = NULL, *_hdr_desc,
1333 				      *_first_desc = NULL;
1334 
1335 	/* pointers to shadow descriptors */
1336 	struct vring_tx_desc desc_mem, hdr_desc_mem, first_desc_mem,
1337 			     *d = &hdr_desc_mem, *hdr_desc = &hdr_desc_mem,
1338 			     *first_desc = &first_desc_mem;
1339 
1340 	/* pointer to shadow descriptors' context */
1341 	struct wil_ctx *hdr_ctx, *first_ctx = NULL;
1342 
1343 	int descs_used = 0; /* total number of used descriptors */
1344 	int sg_desc_cnt = 0; /* number of descriptors for current mss*/
1345 
1346 	u32 swhead = vring->swhead;
1347 	int used, avail = wil_vring_avail_tx(vring);
1348 	int nr_frags = skb_shinfo(skb)->nr_frags;
1349 	int min_desc_required = nr_frags + 1;
1350 	int mss = skb_shinfo(skb)->gso_size;	/* payload size w/o headers */
1351 	int f, len, hdrlen, headlen;
1352 	int vring_index = vring - wil->vring_tx;
1353 	struct vring_tx_data *txdata = &wil->vring_tx_data[vring_index];
1354 	uint i = swhead;
1355 	dma_addr_t pa;
1356 	const skb_frag_t *frag = NULL;
1357 	int rem_data = mss;
1358 	int lenmss;
1359 	int hdr_compensation_need = true;
1360 	int desc_tso_type = wil_tso_type_first;
1361 	bool is_ipv4;
1362 	int tcp_hdr_len;
1363 	int skb_net_hdr_len;
1364 	int gso_type;
1365 	int rc = -EINVAL;
1366 
1367 	wil_dbg_txrx(wil, "tx_vring_tso: %d bytes to vring %d\n", skb->len,
1368 		     vring_index);
1369 
1370 	if (unlikely(!txdata->enabled))
1371 		return -EINVAL;
1372 
1373 	/* A typical page 4K is 3-4 payloads, we assume each fragment
1374 	 * is a full payload, that's how min_desc_required has been
1375 	 * calculated. In real we might need more or less descriptors,
1376 	 * this is the initial check only.
1377 	 */
1378 	if (unlikely(avail < min_desc_required)) {
1379 		wil_err_ratelimited(wil,
1380 				    "TSO: Tx ring[%2d] full. No space for %d fragments\n",
1381 				    vring_index, min_desc_required);
1382 		return -ENOMEM;
1383 	}
1384 
1385 	/* Header Length = MAC header len + IP header len + TCP header len*/
1386 	hdrlen = ETH_HLEN +
1387 		(int)skb_network_header_len(skb) +
1388 		tcp_hdrlen(skb);
1389 
1390 	gso_type = skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV6 | SKB_GSO_TCPV4);
1391 	switch (gso_type) {
1392 	case SKB_GSO_TCPV4:
1393 		/* TCP v4, zero out the IP length and IPv4 checksum fields
1394 		 * as required by the offloading doc
1395 		 */
1396 		ip_hdr(skb)->tot_len = 0;
1397 		ip_hdr(skb)->check = 0;
1398 		is_ipv4 = true;
1399 		break;
1400 	case SKB_GSO_TCPV6:
1401 		/* TCP v6, zero out the payload length */
1402 		ipv6_hdr(skb)->payload_len = 0;
1403 		is_ipv4 = false;
1404 		break;
1405 	default:
1406 		/* other than TCPv4 or TCPv6 types are not supported for TSO.
1407 		 * It is also illegal for both to be set simultaneously
1408 		 */
1409 		return -EINVAL;
1410 	}
1411 
1412 	if (skb->ip_summed != CHECKSUM_PARTIAL)
1413 		return -EINVAL;
1414 
1415 	/* tcp header length and skb network header length are fixed for all
1416 	 * packet's descriptors - read then once here
1417 	 */
1418 	tcp_hdr_len = tcp_hdrlen(skb);
1419 	skb_net_hdr_len = skb_network_header_len(skb);
1420 
1421 	_hdr_desc = &vring->va[i].tx;
1422 
1423 	pa = dma_map_single(dev, skb->data, hdrlen, DMA_TO_DEVICE);
1424 	if (unlikely(dma_mapping_error(dev, pa))) {
1425 		wil_err(wil, "TSO: Skb head DMA map error\n");
1426 		goto err_exit;
1427 	}
1428 
1429 	wil_tx_desc_map(hdr_desc, pa, hdrlen, vring_index);
1430 	wil_tx_desc_offload_setup_tso(hdr_desc, skb, wil_tso_type_hdr, is_ipv4,
1431 				      tcp_hdr_len, skb_net_hdr_len);
1432 	wil_tx_last_desc(hdr_desc);
1433 
1434 	vring->ctx[i].mapped_as = wil_mapped_as_single;
1435 	hdr_ctx = &vring->ctx[i];
1436 
1437 	descs_used++;
1438 	headlen = skb_headlen(skb) - hdrlen;
1439 
1440 	for (f = headlen ? -1 : 0; f < nr_frags; f++)  {
1441 		if (headlen) {
1442 			len = headlen;
1443 			wil_dbg_txrx(wil, "TSO: process skb head, len %u\n",
1444 				     len);
1445 		} else {
1446 			frag = &skb_shinfo(skb)->frags[f];
1447 			len = frag->size;
1448 			wil_dbg_txrx(wil, "TSO: frag[%d]: len %u\n", f, len);
1449 		}
1450 
1451 		while (len) {
1452 			wil_dbg_txrx(wil,
1453 				     "TSO: len %d, rem_data %d, descs_used %d\n",
1454 				     len, rem_data, descs_used);
1455 
1456 			if (descs_used == avail)  {
1457 				wil_err_ratelimited(wil, "TSO: ring overflow\n");
1458 				rc = -ENOMEM;
1459 				goto mem_error;
1460 			}
1461 
1462 			lenmss = min_t(int, rem_data, len);
1463 			i = (swhead + descs_used) % vring->size;
1464 			wil_dbg_txrx(wil, "TSO: lenmss %d, i %d\n", lenmss, i);
1465 
1466 			if (!headlen) {
1467 				pa = skb_frag_dma_map(dev, frag,
1468 						      frag->size - len, lenmss,
1469 						      DMA_TO_DEVICE);
1470 				vring->ctx[i].mapped_as = wil_mapped_as_page;
1471 			} else {
1472 				pa = dma_map_single(dev,
1473 						    skb->data +
1474 						    skb_headlen(skb) - headlen,
1475 						    lenmss,
1476 						    DMA_TO_DEVICE);
1477 				vring->ctx[i].mapped_as = wil_mapped_as_single;
1478 				headlen -= lenmss;
1479 			}
1480 
1481 			if (unlikely(dma_mapping_error(dev, pa))) {
1482 				wil_err(wil, "TSO: DMA map page error\n");
1483 				goto mem_error;
1484 			}
1485 
1486 			_desc = &vring->va[i].tx;
1487 
1488 			if (!_first_desc) {
1489 				_first_desc = _desc;
1490 				first_ctx = &vring->ctx[i];
1491 				d = first_desc;
1492 			} else {
1493 				d = &desc_mem;
1494 			}
1495 
1496 			wil_tx_desc_map(d, pa, lenmss, vring_index);
1497 			wil_tx_desc_offload_setup_tso(d, skb, desc_tso_type,
1498 						      is_ipv4, tcp_hdr_len,
1499 						      skb_net_hdr_len);
1500 
1501 			/* use tso_type_first only once */
1502 			desc_tso_type = wil_tso_type_mid;
1503 
1504 			descs_used++;  /* desc used so far */
1505 			sg_desc_cnt++; /* desc used for this segment */
1506 			len -= lenmss;
1507 			rem_data -= lenmss;
1508 
1509 			wil_dbg_txrx(wil,
1510 				     "TSO: len %d, rem_data %d, descs_used %d, sg_desc_cnt %d,\n",
1511 				     len, rem_data, descs_used, sg_desc_cnt);
1512 
1513 			/* Close the segment if reached mss size or last frag*/
1514 			if (rem_data == 0 || (f == nr_frags - 1 && len == 0)) {
1515 				if (hdr_compensation_need) {
1516 					/* first segment include hdr desc for
1517 					 * release
1518 					 */
1519 					hdr_ctx->nr_frags = sg_desc_cnt;
1520 					wil_tx_desc_set_nr_frags(first_desc,
1521 								 sg_desc_cnt +
1522 								 1);
1523 					hdr_compensation_need = false;
1524 				} else {
1525 					wil_tx_desc_set_nr_frags(first_desc,
1526 								 sg_desc_cnt);
1527 				}
1528 				first_ctx->nr_frags = sg_desc_cnt - 1;
1529 
1530 				wil_tx_last_desc(d);
1531 
1532 				/* first descriptor may also be the last
1533 				 * for this mss - make sure not to copy
1534 				 * it twice
1535 				 */
1536 				if (first_desc != d)
1537 					*_first_desc = *first_desc;
1538 
1539 				/*last descriptor will be copied at the end
1540 				 * of this TS processing
1541 				 */
1542 				if (f < nr_frags - 1 || len > 0)
1543 					*_desc = *d;
1544 
1545 				rem_data = mss;
1546 				_first_desc = NULL;
1547 				sg_desc_cnt = 0;
1548 			} else if (first_desc != d) /* update mid descriptor */
1549 					*_desc = *d;
1550 		}
1551 	}
1552 
1553 	/* first descriptor may also be the last.
1554 	 * in this case d pointer is invalid
1555 	 */
1556 	if (_first_desc == _desc)
1557 		d = first_desc;
1558 
1559 	/* Last data descriptor */
1560 	wil_set_tx_desc_last_tso(d);
1561 	*_desc = *d;
1562 
1563 	/* Fill the total number of descriptors in first desc (hdr)*/
1564 	wil_tx_desc_set_nr_frags(hdr_desc, descs_used);
1565 	*_hdr_desc = *hdr_desc;
1566 
1567 	/* hold reference to skb
1568 	 * to prevent skb release before accounting
1569 	 * in case of immediate "tx done"
1570 	 */
1571 	vring->ctx[i].skb = skb_get(skb);
1572 
1573 	/* performance monitoring */
1574 	used = wil_vring_used_tx(vring);
1575 	if (wil_val_in_range(vring_idle_trsh,
1576 			     used, used + descs_used)) {
1577 		txdata->idle += get_cycles() - txdata->last_idle;
1578 		wil_dbg_txrx(wil,  "Ring[%2d] not idle %d -> %d\n",
1579 			     vring_index, used, used + descs_used);
1580 	}
1581 
1582 	/* Make sure to advance the head only after descriptor update is done.
1583 	 * This will prevent a race condition where the completion thread
1584 	 * will see the DU bit set from previous run and will handle the
1585 	 * skb before it was completed.
1586 	 */
1587 	wmb();
1588 
1589 	/* advance swhead */
1590 	wil_vring_advance_head(vring, descs_used);
1591 	wil_dbg_txrx(wil, "TSO: Tx swhead %d -> %d\n", swhead, vring->swhead);
1592 
1593 	/* make sure all writes to descriptors (shared memory) are done before
1594 	 * committing them to HW
1595 	 */
1596 	wmb();
1597 
1598 	wil_w(wil, vring->hwtail, vring->swhead);
1599 	return 0;
1600 
1601 mem_error:
1602 	while (descs_used > 0) {
1603 		struct wil_ctx *ctx;
1604 
1605 		i = (swhead + descs_used - 1) % vring->size;
1606 		d = (struct vring_tx_desc *)&vring->va[i].tx;
1607 		_desc = &vring->va[i].tx;
1608 		*d = *_desc;
1609 		_desc->dma.status = TX_DMA_STATUS_DU;
1610 		ctx = &vring->ctx[i];
1611 		wil_txdesc_unmap(dev, d, ctx);
1612 		memset(ctx, 0, sizeof(*ctx));
1613 		descs_used--;
1614 	}
1615 err_exit:
1616 	return rc;
1617 }
1618 
1619 static int __wil_tx_vring(struct wil6210_priv *wil, struct vring *vring,
1620 			  struct sk_buff *skb)
1621 {
1622 	struct device *dev = wil_to_dev(wil);
1623 	struct vring_tx_desc dd, *d = &dd;
1624 	volatile struct vring_tx_desc *_d;
1625 	u32 swhead = vring->swhead;
1626 	int avail = wil_vring_avail_tx(vring);
1627 	int nr_frags = skb_shinfo(skb)->nr_frags;
1628 	uint f = 0;
1629 	int vring_index = vring - wil->vring_tx;
1630 	struct vring_tx_data *txdata = &wil->vring_tx_data[vring_index];
1631 	uint i = swhead;
1632 	dma_addr_t pa;
1633 	int used;
1634 	bool mcast = (vring_index == wil->bcast_vring);
1635 	uint len = skb_headlen(skb);
1636 
1637 	wil_dbg_txrx(wil, "tx_vring: %d bytes to vring %d\n", skb->len,
1638 		     vring_index);
1639 
1640 	if (unlikely(!txdata->enabled))
1641 		return -EINVAL;
1642 
1643 	if (unlikely(avail < 1 + nr_frags)) {
1644 		wil_err_ratelimited(wil,
1645 				    "Tx ring[%2d] full. No space for %d fragments\n",
1646 				    vring_index, 1 + nr_frags);
1647 		return -ENOMEM;
1648 	}
1649 	_d = &vring->va[i].tx;
1650 
1651 	pa = dma_map_single(dev, skb->data, skb_headlen(skb), DMA_TO_DEVICE);
1652 
1653 	wil_dbg_txrx(wil, "Tx[%2d] skb %d bytes 0x%p -> %pad\n", vring_index,
1654 		     skb_headlen(skb), skb->data, &pa);
1655 	wil_hex_dump_txrx("Tx ", DUMP_PREFIX_OFFSET, 16, 1,
1656 			  skb->data, skb_headlen(skb), false);
1657 
1658 	if (unlikely(dma_mapping_error(dev, pa)))
1659 		return -EINVAL;
1660 	vring->ctx[i].mapped_as = wil_mapped_as_single;
1661 	/* 1-st segment */
1662 	wil_tx_desc_map(d, pa, len, vring_index);
1663 	if (unlikely(mcast)) {
1664 		d->mac.d[0] |= BIT(MAC_CFG_DESC_TX_0_MCS_EN_POS); /* MCS 0 */
1665 		if (unlikely(len > WIL_BCAST_MCS0_LIMIT)) /* set MCS 1 */
1666 			d->mac.d[0] |= (1 << MAC_CFG_DESC_TX_0_MCS_INDEX_POS);
1667 	}
1668 	/* Process TCP/UDP checksum offloading */
1669 	if (unlikely(wil_tx_desc_offload_setup(d, skb))) {
1670 		wil_err(wil, "Tx[%2d] Failed to set cksum, drop packet\n",
1671 			vring_index);
1672 		goto dma_error;
1673 	}
1674 
1675 	vring->ctx[i].nr_frags = nr_frags;
1676 	wil_tx_desc_set_nr_frags(d, nr_frags + 1);
1677 
1678 	/* middle segments */
1679 	for (; f < nr_frags; f++) {
1680 		const struct skb_frag_struct *frag =
1681 				&skb_shinfo(skb)->frags[f];
1682 		int len = skb_frag_size(frag);
1683 
1684 		*_d = *d;
1685 		wil_dbg_txrx(wil, "Tx[%2d] desc[%4d]\n", vring_index, i);
1686 		wil_hex_dump_txrx("TxD ", DUMP_PREFIX_NONE, 32, 4,
1687 				  (const void *)d, sizeof(*d), false);
1688 		i = (swhead + f + 1) % vring->size;
1689 		_d = &vring->va[i].tx;
1690 		pa = skb_frag_dma_map(dev, frag, 0, skb_frag_size(frag),
1691 				      DMA_TO_DEVICE);
1692 		if (unlikely(dma_mapping_error(dev, pa))) {
1693 			wil_err(wil, "Tx[%2d] failed to map fragment\n",
1694 				vring_index);
1695 			goto dma_error;
1696 		}
1697 		vring->ctx[i].mapped_as = wil_mapped_as_page;
1698 		wil_tx_desc_map(d, pa, len, vring_index);
1699 		/* no need to check return code -
1700 		 * if it succeeded for 1-st descriptor,
1701 		 * it will succeed here too
1702 		 */
1703 		wil_tx_desc_offload_setup(d, skb);
1704 	}
1705 	/* for the last seg only */
1706 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_CMD_EOP_POS);
1707 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_CMD_MARK_WB_POS);
1708 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_CMD_DMA_IT_POS);
1709 	*_d = *d;
1710 	wil_dbg_txrx(wil, "Tx[%2d] desc[%4d]\n", vring_index, i);
1711 	wil_hex_dump_txrx("TxD ", DUMP_PREFIX_NONE, 32, 4,
1712 			  (const void *)d, sizeof(*d), false);
1713 
1714 	/* hold reference to skb
1715 	 * to prevent skb release before accounting
1716 	 * in case of immediate "tx done"
1717 	 */
1718 	vring->ctx[i].skb = skb_get(skb);
1719 
1720 	/* performance monitoring */
1721 	used = wil_vring_used_tx(vring);
1722 	if (wil_val_in_range(vring_idle_trsh,
1723 			     used, used + nr_frags + 1)) {
1724 		txdata->idle += get_cycles() - txdata->last_idle;
1725 		wil_dbg_txrx(wil,  "Ring[%2d] not idle %d -> %d\n",
1726 			     vring_index, used, used + nr_frags + 1);
1727 	}
1728 
1729 	/* Make sure to advance the head only after descriptor update is done.
1730 	 * This will prevent a race condition where the completion thread
1731 	 * will see the DU bit set from previous run and will handle the
1732 	 * skb before it was completed.
1733 	 */
1734 	wmb();
1735 
1736 	/* advance swhead */
1737 	wil_vring_advance_head(vring, nr_frags + 1);
1738 	wil_dbg_txrx(wil, "Tx[%2d] swhead %d -> %d\n", vring_index, swhead,
1739 		     vring->swhead);
1740 	trace_wil6210_tx(vring_index, swhead, skb->len, nr_frags);
1741 
1742 	/* make sure all writes to descriptors (shared memory) are done before
1743 	 * committing them to HW
1744 	 */
1745 	wmb();
1746 
1747 	wil_w(wil, vring->hwtail, vring->swhead);
1748 
1749 	return 0;
1750  dma_error:
1751 	/* unmap what we have mapped */
1752 	nr_frags = f + 1; /* frags mapped + one for skb head */
1753 	for (f = 0; f < nr_frags; f++) {
1754 		struct wil_ctx *ctx;
1755 
1756 		i = (swhead + f) % vring->size;
1757 		ctx = &vring->ctx[i];
1758 		_d = &vring->va[i].tx;
1759 		*d = *_d;
1760 		_d->dma.status = TX_DMA_STATUS_DU;
1761 		wil_txdesc_unmap(dev, d, ctx);
1762 
1763 		memset(ctx, 0, sizeof(*ctx));
1764 	}
1765 
1766 	return -EINVAL;
1767 }
1768 
1769 static int wil_tx_vring(struct wil6210_priv *wil, struct vring *vring,
1770 			struct sk_buff *skb)
1771 {
1772 	int vring_index = vring - wil->vring_tx;
1773 	struct vring_tx_data *txdata = &wil->vring_tx_data[vring_index];
1774 	int rc;
1775 
1776 	spin_lock(&txdata->lock);
1777 
1778 	rc = (skb_is_gso(skb) ? __wil_tx_vring_tso : __wil_tx_vring)
1779 	     (wil, vring, skb);
1780 
1781 	spin_unlock(&txdata->lock);
1782 
1783 	return rc;
1784 }
1785 
1786 /**
1787  * Check status of tx vrings and stop/wake net queues if needed
1788  *
1789  * This function does one of two checks:
1790  * In case check_stop is true, will check if net queues need to be stopped. If
1791  * the conditions for stopping are met, netif_tx_stop_all_queues() is called.
1792  * In case check_stop is false, will check if net queues need to be waked. If
1793  * the conditions for waking are met, netif_tx_wake_all_queues() is called.
1794  * vring is the vring which is currently being modified by either adding
1795  * descriptors (tx) into it or removing descriptors (tx complete) from it. Can
1796  * be null when irrelevant (e.g. connect/disconnect events).
1797  *
1798  * The implementation is to stop net queues if modified vring has low
1799  * descriptor availability. Wake if all vrings are not in low descriptor
1800  * availability and modified vring has high descriptor availability.
1801  */
1802 static inline void __wil_update_net_queues(struct wil6210_priv *wil,
1803 					   struct vring *vring,
1804 					   bool check_stop)
1805 {
1806 	int i;
1807 
1808 	if (vring)
1809 		wil_dbg_txrx(wil, "vring %d, check_stop=%d, stopped=%d",
1810 			     (int)(vring - wil->vring_tx), check_stop,
1811 			     wil->net_queue_stopped);
1812 	else
1813 		wil_dbg_txrx(wil, "check_stop=%d, stopped=%d",
1814 			     check_stop, wil->net_queue_stopped);
1815 
1816 	if (check_stop == wil->net_queue_stopped)
1817 		/* net queues already in desired state */
1818 		return;
1819 
1820 	if (check_stop) {
1821 		if (!vring || unlikely(wil_vring_avail_low(vring))) {
1822 			/* not enough room in the vring */
1823 			netif_tx_stop_all_queues(wil_to_ndev(wil));
1824 			wil->net_queue_stopped = true;
1825 			wil_dbg_txrx(wil, "netif_tx_stop called\n");
1826 		}
1827 		return;
1828 	}
1829 
1830 	/* check wake */
1831 	for (i = 0; i < WIL6210_MAX_TX_RINGS; i++) {
1832 		struct vring *cur_vring = &wil->vring_tx[i];
1833 		struct vring_tx_data *txdata = &wil->vring_tx_data[i];
1834 
1835 		if (!cur_vring->va || !txdata->enabled || cur_vring == vring)
1836 			continue;
1837 
1838 		if (wil_vring_avail_low(cur_vring)) {
1839 			wil_dbg_txrx(wil, "vring %d full, can't wake\n",
1840 				     (int)(cur_vring - wil->vring_tx));
1841 			return;
1842 		}
1843 	}
1844 
1845 	if (!vring || wil_vring_avail_high(vring)) {
1846 		/* enough room in the vring */
1847 		wil_dbg_txrx(wil, "calling netif_tx_wake\n");
1848 		netif_tx_wake_all_queues(wil_to_ndev(wil));
1849 		wil->net_queue_stopped = false;
1850 	}
1851 }
1852 
1853 void wil_update_net_queues(struct wil6210_priv *wil, struct vring *vring,
1854 			   bool check_stop)
1855 {
1856 	spin_lock(&wil->net_queue_lock);
1857 	__wil_update_net_queues(wil, vring, check_stop);
1858 	spin_unlock(&wil->net_queue_lock);
1859 }
1860 
1861 void wil_update_net_queues_bh(struct wil6210_priv *wil, struct vring *vring,
1862 			      bool check_stop)
1863 {
1864 	spin_lock_bh(&wil->net_queue_lock);
1865 	__wil_update_net_queues(wil, vring, check_stop);
1866 	spin_unlock_bh(&wil->net_queue_lock);
1867 }
1868 
1869 netdev_tx_t wil_start_xmit(struct sk_buff *skb, struct net_device *ndev)
1870 {
1871 	struct wil6210_priv *wil = ndev_to_wil(ndev);
1872 	struct ethhdr *eth = (void *)skb->data;
1873 	bool bcast = is_multicast_ether_addr(eth->h_dest);
1874 	struct vring *vring;
1875 	static bool pr_once_fw;
1876 	int rc;
1877 
1878 	wil_dbg_txrx(wil, "start_xmit\n");
1879 	if (unlikely(!test_bit(wil_status_fwready, wil->status))) {
1880 		if (!pr_once_fw) {
1881 			wil_err(wil, "FW not ready\n");
1882 			pr_once_fw = true;
1883 		}
1884 		goto drop;
1885 	}
1886 	if (unlikely(!test_bit(wil_status_fwconnected, wil->status))) {
1887 		wil_dbg_ratelimited(wil, "FW not connected, packet dropped\n");
1888 		goto drop;
1889 	}
1890 	if (unlikely(wil->wdev->iftype == NL80211_IFTYPE_MONITOR)) {
1891 		wil_err(wil, "Xmit in monitor mode not supported\n");
1892 		goto drop;
1893 	}
1894 	pr_once_fw = false;
1895 
1896 	/* find vring */
1897 	if (wil->wdev->iftype == NL80211_IFTYPE_STATION && !wil->pbss) {
1898 		/* in STA mode (ESS), all to same VRING (to AP) */
1899 		vring = wil_find_tx_vring_sta(wil, skb);
1900 	} else if (bcast) {
1901 		if (wil->pbss)
1902 			/* in pbss, no bcast VRING - duplicate skb in
1903 			 * all stations VRINGs
1904 			 */
1905 			vring = wil_find_tx_bcast_2(wil, skb);
1906 		else if (wil->wdev->iftype == NL80211_IFTYPE_AP)
1907 			/* AP has a dedicated bcast VRING */
1908 			vring = wil_find_tx_bcast_1(wil, skb);
1909 		else
1910 			/* unexpected combination, fallback to duplicating
1911 			 * the skb in all stations VRINGs
1912 			 */
1913 			vring = wil_find_tx_bcast_2(wil, skb);
1914 	} else {
1915 		/* unicast, find specific VRING by dest. address */
1916 		vring = wil_find_tx_ucast(wil, skb);
1917 	}
1918 	if (unlikely(!vring)) {
1919 		wil_dbg_txrx(wil, "No Tx VRING found for %pM\n", eth->h_dest);
1920 		goto drop;
1921 	}
1922 	/* set up vring entry */
1923 	rc = wil_tx_vring(wil, vring, skb);
1924 
1925 	switch (rc) {
1926 	case 0:
1927 		/* shall we stop net queues? */
1928 		wil_update_net_queues_bh(wil, vring, true);
1929 		/* statistics will be updated on the tx_complete */
1930 		dev_kfree_skb_any(skb);
1931 		return NETDEV_TX_OK;
1932 	case -ENOMEM:
1933 		return NETDEV_TX_BUSY;
1934 	default:
1935 		break; /* goto drop; */
1936 	}
1937  drop:
1938 	ndev->stats.tx_dropped++;
1939 	dev_kfree_skb_any(skb);
1940 
1941 	return NET_XMIT_DROP;
1942 }
1943 
1944 static inline bool wil_need_txstat(struct sk_buff *skb)
1945 {
1946 	struct ethhdr *eth = (void *)skb->data;
1947 
1948 	return is_unicast_ether_addr(eth->h_dest) && skb->sk &&
1949 	       (skb_shinfo(skb)->tx_flags & SKBTX_WIFI_STATUS);
1950 }
1951 
1952 static inline void wil_consume_skb(struct sk_buff *skb, bool acked)
1953 {
1954 	if (unlikely(wil_need_txstat(skb)))
1955 		skb_complete_wifi_ack(skb, acked);
1956 	else
1957 		acked ? dev_consume_skb_any(skb) : dev_kfree_skb_any(skb);
1958 }
1959 
1960 /**
1961  * Clean up transmitted skb's from the Tx VRING
1962  *
1963  * Return number of descriptors cleared
1964  *
1965  * Safe to call from IRQ
1966  */
1967 int wil_tx_complete(struct wil6210_priv *wil, int ringid)
1968 {
1969 	struct net_device *ndev = wil_to_ndev(wil);
1970 	struct device *dev = wil_to_dev(wil);
1971 	struct vring *vring = &wil->vring_tx[ringid];
1972 	struct vring_tx_data *txdata = &wil->vring_tx_data[ringid];
1973 	int done = 0;
1974 	int cid = wil->vring2cid_tid[ringid][0];
1975 	struct wil_net_stats *stats = NULL;
1976 	volatile struct vring_tx_desc *_d;
1977 	int used_before_complete;
1978 	int used_new;
1979 
1980 	if (unlikely(!vring->va)) {
1981 		wil_err(wil, "Tx irq[%d]: vring not initialized\n", ringid);
1982 		return 0;
1983 	}
1984 
1985 	if (unlikely(!txdata->enabled)) {
1986 		wil_info(wil, "Tx irq[%d]: vring disabled\n", ringid);
1987 		return 0;
1988 	}
1989 
1990 	wil_dbg_txrx(wil, "tx_complete: (%d)\n", ringid);
1991 
1992 	used_before_complete = wil_vring_used_tx(vring);
1993 
1994 	if (cid < WIL6210_MAX_CID)
1995 		stats = &wil->sta[cid].stats;
1996 
1997 	while (!wil_vring_is_empty(vring)) {
1998 		int new_swtail;
1999 		struct wil_ctx *ctx = &vring->ctx[vring->swtail];
2000 		/**
2001 		 * For the fragmented skb, HW will set DU bit only for the
2002 		 * last fragment. look for it.
2003 		 * In TSO the first DU will include hdr desc
2004 		 */
2005 		int lf = (vring->swtail + ctx->nr_frags) % vring->size;
2006 		/* TODO: check we are not past head */
2007 
2008 		_d = &vring->va[lf].tx;
2009 		if (unlikely(!(_d->dma.status & TX_DMA_STATUS_DU)))
2010 			break;
2011 
2012 		new_swtail = (lf + 1) % vring->size;
2013 		while (vring->swtail != new_swtail) {
2014 			struct vring_tx_desc dd, *d = &dd;
2015 			u16 dmalen;
2016 			struct sk_buff *skb;
2017 
2018 			ctx = &vring->ctx[vring->swtail];
2019 			skb = ctx->skb;
2020 			_d = &vring->va[vring->swtail].tx;
2021 
2022 			*d = *_d;
2023 
2024 			dmalen = le16_to_cpu(d->dma.length);
2025 			trace_wil6210_tx_done(ringid, vring->swtail, dmalen,
2026 					      d->dma.error);
2027 			wil_dbg_txrx(wil,
2028 				     "TxC[%2d][%3d] : %d bytes, status 0x%02x err 0x%02x\n",
2029 				     ringid, vring->swtail, dmalen,
2030 				     d->dma.status, d->dma.error);
2031 			wil_hex_dump_txrx("TxCD ", DUMP_PREFIX_NONE, 32, 4,
2032 					  (const void *)d, sizeof(*d), false);
2033 
2034 			wil_txdesc_unmap(dev, d, ctx);
2035 
2036 			if (skb) {
2037 				if (likely(d->dma.error == 0)) {
2038 					ndev->stats.tx_packets++;
2039 					ndev->stats.tx_bytes += skb->len;
2040 					if (stats) {
2041 						stats->tx_packets++;
2042 						stats->tx_bytes += skb->len;
2043 					}
2044 				} else {
2045 					ndev->stats.tx_errors++;
2046 					if (stats)
2047 						stats->tx_errors++;
2048 				}
2049 				wil_consume_skb(skb, d->dma.error == 0);
2050 			}
2051 			memset(ctx, 0, sizeof(*ctx));
2052 			/* Make sure the ctx is zeroed before updating the tail
2053 			 * to prevent a case where wil_tx_vring will see
2054 			 * this descriptor as used and handle it before ctx zero
2055 			 * is completed.
2056 			 */
2057 			wmb();
2058 			/* There is no need to touch HW descriptor:
2059 			 * - ststus bit TX_DMA_STATUS_DU is set by design,
2060 			 *   so hardware will not try to process this desc.,
2061 			 * - rest of descriptor will be initialized on Tx.
2062 			 */
2063 			vring->swtail = wil_vring_next_tail(vring);
2064 			done++;
2065 		}
2066 	}
2067 
2068 	/* performance monitoring */
2069 	used_new = wil_vring_used_tx(vring);
2070 	if (wil_val_in_range(vring_idle_trsh,
2071 			     used_new, used_before_complete)) {
2072 		wil_dbg_txrx(wil, "Ring[%2d] idle %d -> %d\n",
2073 			     ringid, used_before_complete, used_new);
2074 		txdata->last_idle = get_cycles();
2075 	}
2076 
2077 	/* shall we wake net queues? */
2078 	if (done)
2079 		wil_update_net_queues(wil, vring, false);
2080 
2081 	return done;
2082 }
2083