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