xref: /linux/drivers/net/wireless/ath/wil6210/txrx.c (revision 93df8a1ed6231727c5db94a80b1a6bd5ee67cec3)
1 /*
2  * Copyright (c) 2012-2015 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 	if (tx) {
164 		int vring_index = vring - wil->vring_tx;
165 
166 		wil_dbg_misc(wil, "free Tx vring %d [%d] 0x%p:%pad 0x%p\n",
167 			     vring_index, vring->size, vring->va,
168 			     &vring->pa, vring->ctx);
169 	} else {
170 		wil_dbg_misc(wil, "free Rx vring [%d] 0x%p:%pad 0x%p\n",
171 			     vring->size, vring->va,
172 			     &vring->pa, vring->ctx);
173 	}
174 
175 	while (!wil_vring_is_empty(vring)) {
176 		dma_addr_t pa;
177 		u16 dmalen;
178 		struct wil_ctx *ctx;
179 
180 		if (tx) {
181 			struct vring_tx_desc dd, *d = &dd;
182 			volatile struct vring_tx_desc *_d =
183 					&vring->va[vring->swtail].tx;
184 
185 			ctx = &vring->ctx[vring->swtail];
186 			*d = *_d;
187 			wil_txdesc_unmap(dev, d, ctx);
188 			if (ctx->skb)
189 				dev_kfree_skb_any(ctx->skb);
190 			vring->swtail = wil_vring_next_tail(vring);
191 		} else { /* rx */
192 			struct vring_rx_desc dd, *d = &dd;
193 			volatile struct vring_rx_desc *_d =
194 					&vring->va[vring->swhead].rx;
195 
196 			ctx = &vring->ctx[vring->swhead];
197 			*d = *_d;
198 			pa = wil_desc_addr(&d->dma.addr);
199 			dmalen = le16_to_cpu(d->dma.length);
200 			dma_unmap_single(dev, pa, dmalen, DMA_FROM_DEVICE);
201 			kfree_skb(ctx->skb);
202 			wil_vring_advance_head(vring, 1);
203 		}
204 	}
205 	dma_free_coherent(dev, sz, (void *)vring->va, vring->pa);
206 	kfree(vring->ctx);
207 	vring->pa = 0;
208 	vring->va = NULL;
209 	vring->ctx = NULL;
210 }
211 
212 /**
213  * Allocate one skb for Rx VRING
214  *
215  * Safe to call from IRQ
216  */
217 static int wil_vring_alloc_skb(struct wil6210_priv *wil, struct vring *vring,
218 			       u32 i, int headroom)
219 {
220 	struct device *dev = wil_to_dev(wil);
221 	unsigned int sz = mtu_max + ETH_HLEN + wil_rx_snaplen();
222 	struct vring_rx_desc dd, *d = &dd;
223 	volatile struct vring_rx_desc *_d = &vring->va[i].rx;
224 	dma_addr_t pa;
225 	struct sk_buff *skb = dev_alloc_skb(sz + headroom);
226 
227 	if (unlikely(!skb))
228 		return -ENOMEM;
229 
230 	skb_reserve(skb, headroom);
231 	skb_put(skb, sz);
232 
233 	pa = dma_map_single(dev, skb->data, skb->len, DMA_FROM_DEVICE);
234 	if (unlikely(dma_mapping_error(dev, pa))) {
235 		kfree_skb(skb);
236 		return -ENOMEM;
237 	}
238 
239 	d->dma.d0 = RX_DMA_D0_CMD_DMA_RT | RX_DMA_D0_CMD_DMA_IT;
240 	wil_desc_addr_set(&d->dma.addr, pa);
241 	/* ip_length don't care */
242 	/* b11 don't care */
243 	/* error don't care */
244 	d->dma.status = 0; /* BIT(0) should be 0 for HW_OWNED */
245 	d->dma.length = cpu_to_le16(sz);
246 	*_d = *d;
247 	vring->ctx[i].skb = skb;
248 
249 	return 0;
250 }
251 
252 /**
253  * Adds radiotap header
254  *
255  * Any error indicated as "Bad FCS"
256  *
257  * Vendor data for 04:ce:14-1 (Wilocity-1) consists of:
258  *  - Rx descriptor: 32 bytes
259  *  - Phy info
260  */
261 static void wil_rx_add_radiotap_header(struct wil6210_priv *wil,
262 				       struct sk_buff *skb)
263 {
264 	struct wireless_dev *wdev = wil->wdev;
265 	struct wil6210_rtap {
266 		struct ieee80211_radiotap_header rthdr;
267 		/* fields should be in the order of bits in rthdr.it_present */
268 		/* flags */
269 		u8 flags;
270 		/* channel */
271 		__le16 chnl_freq __aligned(2);
272 		__le16 chnl_flags;
273 		/* MCS */
274 		u8 mcs_present;
275 		u8 mcs_flags;
276 		u8 mcs_index;
277 	} __packed;
278 	struct wil6210_rtap_vendor {
279 		struct wil6210_rtap rtap;
280 		/* vendor */
281 		u8 vendor_oui[3] __aligned(2);
282 		u8 vendor_ns;
283 		__le16 vendor_skip;
284 		u8 vendor_data[0];
285 	} __packed;
286 	struct vring_rx_desc *d = wil_skb_rxdesc(skb);
287 	struct wil6210_rtap_vendor *rtap_vendor;
288 	int rtap_len = sizeof(struct wil6210_rtap);
289 	int phy_length = 0; /* phy info header size, bytes */
290 	static char phy_data[128];
291 	struct ieee80211_channel *ch = wdev->preset_chandef.chan;
292 
293 	if (rtap_include_phy_info) {
294 		rtap_len = sizeof(*rtap_vendor) + sizeof(*d);
295 		/* calculate additional length */
296 		if (d->dma.status & RX_DMA_STATUS_PHY_INFO) {
297 			/**
298 			 * PHY info starts from 8-byte boundary
299 			 * there are 8-byte lines, last line may be partially
300 			 * written (HW bug), thus FW configures for last line
301 			 * to be excessive. Driver skips this last line.
302 			 */
303 			int len = min_t(int, 8 + sizeof(phy_data),
304 					wil_rxdesc_phy_length(d));
305 
306 			if (len > 8) {
307 				void *p = skb_tail_pointer(skb);
308 				void *pa = PTR_ALIGN(p, 8);
309 
310 				if (skb_tailroom(skb) >= len + (pa - p)) {
311 					phy_length = len - 8;
312 					memcpy(phy_data, pa, phy_length);
313 				}
314 			}
315 		}
316 		rtap_len += phy_length;
317 	}
318 
319 	if (skb_headroom(skb) < rtap_len &&
320 	    pskb_expand_head(skb, rtap_len, 0, GFP_ATOMIC)) {
321 		wil_err(wil, "Unable to expand headrom to %d\n", rtap_len);
322 		return;
323 	}
324 
325 	rtap_vendor = (void *)skb_push(skb, rtap_len);
326 	memset(rtap_vendor, 0, rtap_len);
327 
328 	rtap_vendor->rtap.rthdr.it_version = PKTHDR_RADIOTAP_VERSION;
329 	rtap_vendor->rtap.rthdr.it_len = cpu_to_le16(rtap_len);
330 	rtap_vendor->rtap.rthdr.it_present = cpu_to_le32(
331 			(1 << IEEE80211_RADIOTAP_FLAGS) |
332 			(1 << IEEE80211_RADIOTAP_CHANNEL) |
333 			(1 << IEEE80211_RADIOTAP_MCS));
334 	if (d->dma.status & RX_DMA_STATUS_ERROR)
335 		rtap_vendor->rtap.flags |= IEEE80211_RADIOTAP_F_BADFCS;
336 
337 	rtap_vendor->rtap.chnl_freq = cpu_to_le16(ch ? ch->center_freq : 58320);
338 	rtap_vendor->rtap.chnl_flags = cpu_to_le16(0);
339 
340 	rtap_vendor->rtap.mcs_present = IEEE80211_RADIOTAP_MCS_HAVE_MCS;
341 	rtap_vendor->rtap.mcs_flags = 0;
342 	rtap_vendor->rtap.mcs_index = wil_rxdesc_mcs(d);
343 
344 	if (rtap_include_phy_info) {
345 		rtap_vendor->rtap.rthdr.it_present |= cpu_to_le32(1 <<
346 				IEEE80211_RADIOTAP_VENDOR_NAMESPACE);
347 		/* OUI for Wilocity 04:ce:14 */
348 		rtap_vendor->vendor_oui[0] = 0x04;
349 		rtap_vendor->vendor_oui[1] = 0xce;
350 		rtap_vendor->vendor_oui[2] = 0x14;
351 		rtap_vendor->vendor_ns = 1;
352 		/* Rx descriptor + PHY data  */
353 		rtap_vendor->vendor_skip = cpu_to_le16(sizeof(*d) +
354 						       phy_length);
355 		memcpy(rtap_vendor->vendor_data, (void *)d, sizeof(*d));
356 		memcpy(rtap_vendor->vendor_data + sizeof(*d), phy_data,
357 		       phy_length);
358 	}
359 }
360 
361 /**
362  * reap 1 frame from @swhead
363  *
364  * Rx descriptor copied to skb->cb
365  *
366  * Safe to call from IRQ
367  */
368 static struct sk_buff *wil_vring_reap_rx(struct wil6210_priv *wil,
369 					 struct vring *vring)
370 {
371 	struct device *dev = wil_to_dev(wil);
372 	struct net_device *ndev = wil_to_ndev(wil);
373 	volatile struct vring_rx_desc *_d;
374 	struct vring_rx_desc *d;
375 	struct sk_buff *skb;
376 	dma_addr_t pa;
377 	unsigned int snaplen = wil_rx_snaplen();
378 	unsigned int sz = mtu_max + ETH_HLEN + snaplen;
379 	u16 dmalen;
380 	u8 ftype;
381 	int cid;
382 	int i = (int)vring->swhead;
383 	struct wil_net_stats *stats;
384 
385 	BUILD_BUG_ON(sizeof(struct vring_rx_desc) > sizeof(skb->cb));
386 
387 	if (unlikely(wil_vring_is_empty(vring)))
388 		return NULL;
389 
390 	_d = &vring->va[i].rx;
391 	if (unlikely(!(_d->dma.status & RX_DMA_STATUS_DU))) {
392 		/* it is not error, we just reached end of Rx done area */
393 		return NULL;
394 	}
395 
396 	skb = vring->ctx[i].skb;
397 	vring->ctx[i].skb = NULL;
398 	wil_vring_advance_head(vring, 1);
399 	if (!skb) {
400 		wil_err(wil, "No Rx skb at [%d]\n", i);
401 		return NULL;
402 	}
403 	d = wil_skb_rxdesc(skb);
404 	*d = *_d;
405 	pa = wil_desc_addr(&d->dma.addr);
406 
407 	dma_unmap_single(dev, pa, sz, DMA_FROM_DEVICE);
408 	dmalen = le16_to_cpu(d->dma.length);
409 
410 	trace_wil6210_rx(i, d);
411 	wil_dbg_txrx(wil, "Rx[%3d] : %d bytes\n", i, dmalen);
412 	wil_hex_dump_txrx("Rx ", DUMP_PREFIX_NONE, 32, 4,
413 			  (const void *)d, sizeof(*d), false);
414 
415 	if (unlikely(dmalen > sz)) {
416 		wil_err(wil, "Rx size too large: %d bytes!\n", dmalen);
417 		kfree_skb(skb);
418 		return NULL;
419 	}
420 	skb_trim(skb, dmalen);
421 
422 	prefetch(skb->data);
423 
424 	wil_hex_dump_txrx("Rx ", DUMP_PREFIX_OFFSET, 16, 1,
425 			  skb->data, skb_headlen(skb), false);
426 
427 	cid = wil_rxdesc_cid(d);
428 	stats = &wil->sta[cid].stats;
429 	stats->last_mcs_rx = wil_rxdesc_mcs(d);
430 	if (stats->last_mcs_rx < ARRAY_SIZE(stats->rx_per_mcs))
431 		stats->rx_per_mcs[stats->last_mcs_rx]++;
432 
433 	/* use radiotap header only if required */
434 	if (ndev->type == ARPHRD_IEEE80211_RADIOTAP)
435 		wil_rx_add_radiotap_header(wil, skb);
436 
437 	/* no extra checks if in sniffer mode */
438 	if (ndev->type != ARPHRD_ETHER)
439 		return skb;
440 	/*
441 	 * Non-data frames may be delivered through Rx DMA channel (ex: BAR)
442 	 * Driver should recognize it by frame type, that is found
443 	 * in Rx descriptor. If type is not data, it is 802.11 frame as is
444 	 */
445 	ftype = wil_rxdesc_ftype(d) << 2;
446 	if (unlikely(ftype != IEEE80211_FTYPE_DATA)) {
447 		wil_dbg_txrx(wil, "Non-data frame ftype 0x%08x\n", ftype);
448 		/* TODO: process it */
449 		kfree_skb(skb);
450 		return NULL;
451 	}
452 
453 	if (unlikely(skb->len < ETH_HLEN + snaplen)) {
454 		wil_err(wil, "Short frame, len = %d\n", skb->len);
455 		/* TODO: process it (i.e. BAR) */
456 		kfree_skb(skb);
457 		return NULL;
458 	}
459 
460 	/* L4 IDENT is on when HW calculated checksum, check status
461 	 * and in case of error drop the packet
462 	 * higher stack layers will handle retransmission (if required)
463 	 */
464 	if (likely(d->dma.status & RX_DMA_STATUS_L4I)) {
465 		/* L4 protocol identified, csum calculated */
466 		if (likely((d->dma.error & RX_DMA_ERROR_L4_ERR) == 0))
467 			skb->ip_summed = CHECKSUM_UNNECESSARY;
468 		/* If HW reports bad checksum, let IP stack re-check it
469 		 * For example, HW don't understand Microsoft IP stack that
470 		 * mis-calculates TCP checksum - if it should be 0x0,
471 		 * it writes 0xffff in violation of RFC 1624
472 		 */
473 	}
474 
475 	if (snaplen) {
476 		/* Packet layout
477 		 * +-------+-------+---------+------------+------+
478 		 * | SA(6) | DA(6) | SNAP(6) | ETHTYPE(2) | DATA |
479 		 * +-------+-------+---------+------------+------+
480 		 * Need to remove SNAP, shifting SA and DA forward
481 		 */
482 		memmove(skb->data + snaplen, skb->data, 2 * ETH_ALEN);
483 		skb_pull(skb, snaplen);
484 	}
485 
486 	return skb;
487 }
488 
489 /**
490  * allocate and fill up to @count buffers in rx ring
491  * buffers posted at @swtail
492  */
493 static int wil_rx_refill(struct wil6210_priv *wil, int count)
494 {
495 	struct net_device *ndev = wil_to_ndev(wil);
496 	struct vring *v = &wil->vring_rx;
497 	u32 next_tail;
498 	int rc = 0;
499 	int headroom = ndev->type == ARPHRD_IEEE80211_RADIOTAP ?
500 			WIL6210_RTAP_SIZE : 0;
501 
502 	for (; next_tail = wil_vring_next_tail(v),
503 			(next_tail != v->swhead) && (count-- > 0);
504 			v->swtail = next_tail) {
505 		rc = wil_vring_alloc_skb(wil, v, v->swtail, headroom);
506 		if (unlikely(rc)) {
507 			wil_err(wil, "Error %d in wil_rx_refill[%d]\n",
508 				rc, v->swtail);
509 			break;
510 		}
511 	}
512 	iowrite32(v->swtail, wil->csr + HOSTADDR(v->hwtail));
513 
514 	return rc;
515 }
516 
517 /*
518  * Pass Rx packet to the netif. Update statistics.
519  * Called in softirq context (NAPI poll).
520  */
521 void wil_netif_rx_any(struct sk_buff *skb, struct net_device *ndev)
522 {
523 	gro_result_t rc = GRO_NORMAL;
524 	struct wil6210_priv *wil = ndev_to_wil(ndev);
525 	struct wireless_dev *wdev = wil_to_wdev(wil);
526 	unsigned int len = skb->len;
527 	struct vring_rx_desc *d = wil_skb_rxdesc(skb);
528 	int cid = wil_rxdesc_cid(d); /* always 0..7, no need to check */
529 	struct ethhdr *eth = (void *)skb->data;
530 	/* here looking for DA, not A1, thus Rxdesc's 'mcast' indication
531 	 * is not suitable, need to look at data
532 	 */
533 	int mcast = is_multicast_ether_addr(eth->h_dest);
534 	struct wil_net_stats *stats = &wil->sta[cid].stats;
535 	struct sk_buff *xmit_skb = NULL;
536 	static const char * const gro_res_str[] = {
537 		[GRO_MERGED]		= "GRO_MERGED",
538 		[GRO_MERGED_FREE]	= "GRO_MERGED_FREE",
539 		[GRO_HELD]		= "GRO_HELD",
540 		[GRO_NORMAL]		= "GRO_NORMAL",
541 		[GRO_DROP]		= "GRO_DROP",
542 	};
543 
544 	skb_orphan(skb);
545 
546 	if (wdev->iftype == NL80211_IFTYPE_AP && !wil->ap_isolate) {
547 		if (mcast) {
548 			/* send multicast frames both to higher layers in
549 			 * local net stack and back to the wireless medium
550 			 */
551 			xmit_skb = skb_copy(skb, GFP_ATOMIC);
552 		} else {
553 			int xmit_cid = wil_find_cid(wil, eth->h_dest);
554 
555 			if (xmit_cid >= 0) {
556 				/* The destination station is associated to
557 				 * this AP (in this VLAN), so send the frame
558 				 * directly to it and do not pass it to local
559 				 * net stack.
560 				 */
561 				xmit_skb = skb;
562 				skb = NULL;
563 			}
564 		}
565 	}
566 	if (xmit_skb) {
567 		/* Send to wireless media and increase priority by 256 to
568 		 * keep the received priority instead of reclassifying
569 		 * the frame (see cfg80211_classify8021d).
570 		 */
571 		xmit_skb->dev = ndev;
572 		xmit_skb->priority += 256;
573 		xmit_skb->protocol = htons(ETH_P_802_3);
574 		skb_reset_network_header(xmit_skb);
575 		skb_reset_mac_header(xmit_skb);
576 		wil_dbg_txrx(wil, "Rx -> Tx %d bytes\n", len);
577 		dev_queue_xmit(xmit_skb);
578 	}
579 
580 	if (skb) { /* deliver to local stack */
581 
582 		skb->protocol = eth_type_trans(skb, ndev);
583 		rc = napi_gro_receive(&wil->napi_rx, skb);
584 		wil_dbg_txrx(wil, "Rx complete %d bytes => %s\n",
585 			     len, gro_res_str[rc]);
586 	}
587 	/* statistics. rc set to GRO_NORMAL for AP bridging */
588 	if (unlikely(rc == GRO_DROP)) {
589 		ndev->stats.rx_dropped++;
590 		stats->rx_dropped++;
591 		wil_dbg_txrx(wil, "Rx drop %d bytes\n", len);
592 	} else {
593 		ndev->stats.rx_packets++;
594 		stats->rx_packets++;
595 		ndev->stats.rx_bytes += len;
596 		stats->rx_bytes += len;
597 		if (mcast)
598 			ndev->stats.multicast++;
599 	}
600 }
601 
602 /**
603  * Proceed all completed skb's from Rx VRING
604  *
605  * Safe to call from NAPI poll, i.e. softirq with interrupts enabled
606  */
607 void wil_rx_handle(struct wil6210_priv *wil, int *quota)
608 {
609 	struct net_device *ndev = wil_to_ndev(wil);
610 	struct vring *v = &wil->vring_rx;
611 	struct sk_buff *skb;
612 
613 	if (unlikely(!v->va)) {
614 		wil_err(wil, "Rx IRQ while Rx not yet initialized\n");
615 		return;
616 	}
617 	wil_dbg_txrx(wil, "%s()\n", __func__);
618 	while ((*quota > 0) && (NULL != (skb = wil_vring_reap_rx(wil, v)))) {
619 		(*quota)--;
620 
621 		if (wil->wdev->iftype == NL80211_IFTYPE_MONITOR) {
622 			skb->dev = ndev;
623 			skb_reset_mac_header(skb);
624 			skb->ip_summed = CHECKSUM_UNNECESSARY;
625 			skb->pkt_type = PACKET_OTHERHOST;
626 			skb->protocol = htons(ETH_P_802_2);
627 			wil_netif_rx_any(skb, ndev);
628 		} else {
629 			wil_rx_reorder(wil, skb);
630 		}
631 	}
632 	wil_rx_refill(wil, v->size);
633 }
634 
635 int wil_rx_init(struct wil6210_priv *wil, u16 size)
636 {
637 	struct vring *vring = &wil->vring_rx;
638 	int rc;
639 
640 	wil_dbg_misc(wil, "%s()\n", __func__);
641 
642 	if (vring->va) {
643 		wil_err(wil, "Rx ring already allocated\n");
644 		return -EINVAL;
645 	}
646 
647 	vring->size = size;
648 	rc = wil_vring_alloc(wil, vring);
649 	if (rc)
650 		return rc;
651 
652 	rc = wmi_rx_chain_add(wil, vring);
653 	if (rc)
654 		goto err_free;
655 
656 	rc = wil_rx_refill(wil, vring->size);
657 	if (rc)
658 		goto err_free;
659 
660 	return 0;
661  err_free:
662 	wil_vring_free(wil, vring, 0);
663 
664 	return rc;
665 }
666 
667 void wil_rx_fini(struct wil6210_priv *wil)
668 {
669 	struct vring *vring = &wil->vring_rx;
670 
671 	wil_dbg_misc(wil, "%s()\n", __func__);
672 
673 	if (vring->va)
674 		wil_vring_free(wil, vring, 0);
675 }
676 
677 int wil_vring_init_tx(struct wil6210_priv *wil, int id, int size,
678 		      int cid, int tid)
679 {
680 	int rc;
681 	struct wmi_vring_cfg_cmd cmd = {
682 		.action = cpu_to_le32(WMI_VRING_CMD_ADD),
683 		.vring_cfg = {
684 			.tx_sw_ring = {
685 				.max_mpdu_size =
686 					cpu_to_le16(wil_mtu2macbuf(mtu_max)),
687 				.ring_size = cpu_to_le16(size),
688 			},
689 			.ringid = id,
690 			.cidxtid = mk_cidxtid(cid, tid),
691 			.encap_trans_type = WMI_VRING_ENC_TYPE_802_3,
692 			.mac_ctrl = 0,
693 			.to_resolution = 0,
694 			.agg_max_wsize = 0,
695 			.schd_params = {
696 				.priority = cpu_to_le16(0),
697 				.timeslot_us = cpu_to_le16(0xfff),
698 			},
699 		},
700 	};
701 	struct {
702 		struct wil6210_mbox_hdr_wmi wmi;
703 		struct wmi_vring_cfg_done_event cmd;
704 	} __packed reply;
705 	struct vring *vring = &wil->vring_tx[id];
706 	struct vring_tx_data *txdata = &wil->vring_tx_data[id];
707 
708 	wil_dbg_misc(wil, "%s() max_mpdu_size %d\n", __func__,
709 		     cmd.vring_cfg.tx_sw_ring.max_mpdu_size);
710 
711 	if (vring->va) {
712 		wil_err(wil, "Tx ring [%d] already allocated\n", id);
713 		rc = -EINVAL;
714 		goto out;
715 	}
716 
717 	memset(txdata, 0, sizeof(*txdata));
718 	spin_lock_init(&txdata->lock);
719 	vring->size = size;
720 	rc = wil_vring_alloc(wil, vring);
721 	if (rc)
722 		goto out;
723 
724 	wil->vring2cid_tid[id][0] = cid;
725 	wil->vring2cid_tid[id][1] = tid;
726 
727 	cmd.vring_cfg.tx_sw_ring.ring_mem_base = cpu_to_le64(vring->pa);
728 
729 	if (!wil->privacy)
730 		txdata->dot1x_open = true;
731 	rc = wmi_call(wil, WMI_VRING_CFG_CMDID, &cmd, sizeof(cmd),
732 		      WMI_VRING_CFG_DONE_EVENTID, &reply, sizeof(reply), 100);
733 	if (rc)
734 		goto out_free;
735 
736 	if (reply.cmd.status != WMI_FW_STATUS_SUCCESS) {
737 		wil_err(wil, "Tx config failed, status 0x%02x\n",
738 			reply.cmd.status);
739 		rc = -EINVAL;
740 		goto out_free;
741 	}
742 	vring->hwtail = le32_to_cpu(reply.cmd.tx_vring_tail_ptr);
743 
744 	txdata->enabled = 1;
745 	if (txdata->dot1x_open && (agg_wsize >= 0))
746 		wil_addba_tx_request(wil, id, agg_wsize);
747 
748 	return 0;
749  out_free:
750 	txdata->dot1x_open = false;
751 	txdata->enabled = 0;
752 	wil_vring_free(wil, vring, 1);
753  out:
754 
755 	return rc;
756 }
757 
758 int wil_vring_init_bcast(struct wil6210_priv *wil, int id, int size)
759 {
760 	int rc;
761 	struct wmi_bcast_vring_cfg_cmd cmd = {
762 		.action = cpu_to_le32(WMI_VRING_CMD_ADD),
763 		.vring_cfg = {
764 			.tx_sw_ring = {
765 				.max_mpdu_size =
766 					cpu_to_le16(wil_mtu2macbuf(mtu_max)),
767 				.ring_size = cpu_to_le16(size),
768 			},
769 			.ringid = id,
770 			.encap_trans_type = WMI_VRING_ENC_TYPE_802_3,
771 		},
772 	};
773 	struct {
774 		struct wil6210_mbox_hdr_wmi wmi;
775 		struct wmi_vring_cfg_done_event cmd;
776 	} __packed reply;
777 	struct vring *vring = &wil->vring_tx[id];
778 	struct vring_tx_data *txdata = &wil->vring_tx_data[id];
779 
780 	wil_dbg_misc(wil, "%s() max_mpdu_size %d\n", __func__,
781 		     cmd.vring_cfg.tx_sw_ring.max_mpdu_size);
782 
783 	if (vring->va) {
784 		wil_err(wil, "Tx ring [%d] already allocated\n", id);
785 		rc = -EINVAL;
786 		goto out;
787 	}
788 
789 	memset(txdata, 0, sizeof(*txdata));
790 	spin_lock_init(&txdata->lock);
791 	vring->size = size;
792 	rc = wil_vring_alloc(wil, vring);
793 	if (rc)
794 		goto out;
795 
796 	wil->vring2cid_tid[id][0] = WIL6210_MAX_CID; /* CID */
797 	wil->vring2cid_tid[id][1] = 0; /* TID */
798 
799 	cmd.vring_cfg.tx_sw_ring.ring_mem_base = cpu_to_le64(vring->pa);
800 
801 	if (!wil->privacy)
802 		txdata->dot1x_open = true;
803 	rc = wmi_call(wil, WMI_BCAST_VRING_CFG_CMDID, &cmd, sizeof(cmd),
804 		      WMI_VRING_CFG_DONE_EVENTID, &reply, sizeof(reply), 100);
805 	if (rc)
806 		goto out_free;
807 
808 	if (reply.cmd.status != WMI_FW_STATUS_SUCCESS) {
809 		wil_err(wil, "Tx config failed, status 0x%02x\n",
810 			reply.cmd.status);
811 		rc = -EINVAL;
812 		goto out_free;
813 	}
814 	vring->hwtail = le32_to_cpu(reply.cmd.tx_vring_tail_ptr);
815 
816 	txdata->enabled = 1;
817 
818 	return 0;
819  out_free:
820 	txdata->enabled = 0;
821 	txdata->dot1x_open = false;
822 	wil_vring_free(wil, vring, 1);
823  out:
824 
825 	return rc;
826 }
827 
828 void wil_vring_fini_tx(struct wil6210_priv *wil, int id)
829 {
830 	struct vring *vring = &wil->vring_tx[id];
831 	struct vring_tx_data *txdata = &wil->vring_tx_data[id];
832 
833 	WARN_ON(!mutex_is_locked(&wil->mutex));
834 
835 	if (!vring->va)
836 		return;
837 
838 	wil_dbg_misc(wil, "%s() id=%d\n", __func__, id);
839 
840 	spin_lock_bh(&txdata->lock);
841 	txdata->dot1x_open = false;
842 	txdata->enabled = 0; /* no Tx can be in progress or start anew */
843 	spin_unlock_bh(&txdata->lock);
844 	/* make sure NAPI won't touch this vring */
845 	if (test_bit(wil_status_napi_en, wil->status))
846 		napi_synchronize(&wil->napi_tx);
847 
848 	wil_vring_free(wil, vring, 1);
849 	memset(txdata, 0, sizeof(*txdata));
850 }
851 
852 static struct vring *wil_find_tx_ucast(struct wil6210_priv *wil,
853 				       struct sk_buff *skb)
854 {
855 	int i;
856 	struct ethhdr *eth = (void *)skb->data;
857 	int cid = wil_find_cid(wil, eth->h_dest);
858 
859 	if (cid < 0)
860 		return NULL;
861 
862 	/* TODO: fix for multiple TID */
863 	for (i = 0; i < ARRAY_SIZE(wil->vring2cid_tid); i++) {
864 		if (!wil->vring_tx_data[i].dot1x_open &&
865 		    (skb->protocol != cpu_to_be16(ETH_P_PAE)))
866 			continue;
867 		if (wil->vring2cid_tid[i][0] == cid) {
868 			struct vring *v = &wil->vring_tx[i];
869 
870 			wil_dbg_txrx(wil, "%s(%pM) -> [%d]\n",
871 				     __func__, eth->h_dest, i);
872 			if (v->va) {
873 				return v;
874 			} else {
875 				wil_dbg_txrx(wil, "vring[%d] not valid\n", i);
876 				return NULL;
877 			}
878 		}
879 	}
880 
881 	return NULL;
882 }
883 
884 static int wil_tx_vring(struct wil6210_priv *wil, struct vring *vring,
885 			struct sk_buff *skb);
886 
887 static struct vring *wil_find_tx_vring_sta(struct wil6210_priv *wil,
888 					   struct sk_buff *skb)
889 {
890 	struct vring *v;
891 	int i;
892 	u8 cid;
893 
894 	/* In the STA mode, it is expected to have only 1 VRING
895 	 * for the AP we connected to.
896 	 * find 1-st vring eligible for this skb and use it.
897 	 */
898 	for (i = 0; i < WIL6210_MAX_TX_RINGS; i++) {
899 		v = &wil->vring_tx[i];
900 		if (!v->va)
901 			continue;
902 
903 		cid = wil->vring2cid_tid[i][0];
904 		if (cid >= WIL6210_MAX_CID) /* skip BCAST */
905 			continue;
906 
907 		if (!wil->vring_tx_data[i].dot1x_open &&
908 		    (skb->protocol != cpu_to_be16(ETH_P_PAE)))
909 			continue;
910 
911 		wil_dbg_txrx(wil, "Tx -> ring %d\n", i);
912 
913 		return v;
914 	}
915 
916 	wil_dbg_txrx(wil, "Tx while no vrings active?\n");
917 
918 	return NULL;
919 }
920 
921 /* Use one of 2 strategies:
922  *
923  * 1. New (real broadcast):
924  *    use dedicated broadcast vring
925  * 2. Old (pseudo-DMS):
926  *    Find 1-st vring and return it;
927  *    duplicate skb and send it to other active vrings;
928  *    in all cases override dest address to unicast peer's address
929  * Use old strategy when new is not supported yet:
930  *  - for PBSS
931  */
932 static struct vring *wil_find_tx_bcast_1(struct wil6210_priv *wil,
933 					 struct sk_buff *skb)
934 {
935 	struct vring *v;
936 	int i = wil->bcast_vring;
937 
938 	if (i < 0)
939 		return NULL;
940 	v = &wil->vring_tx[i];
941 	if (!v->va)
942 		return NULL;
943 	if (!wil->vring_tx_data[i].dot1x_open &&
944 	    (skb->protocol != cpu_to_be16(ETH_P_PAE)))
945 		return NULL;
946 
947 	return v;
948 }
949 
950 static void wil_set_da_for_vring(struct wil6210_priv *wil,
951 				 struct sk_buff *skb, int vring_index)
952 {
953 	struct ethhdr *eth = (void *)skb->data;
954 	int cid = wil->vring2cid_tid[vring_index][0];
955 
956 	ether_addr_copy(eth->h_dest, wil->sta[cid].addr);
957 }
958 
959 static struct vring *wil_find_tx_bcast_2(struct wil6210_priv *wil,
960 					 struct sk_buff *skb)
961 {
962 	struct vring *v, *v2;
963 	struct sk_buff *skb2;
964 	int i;
965 	u8 cid;
966 	struct ethhdr *eth = (void *)skb->data;
967 	char *src = eth->h_source;
968 
969 	/* find 1-st vring eligible for data */
970 	for (i = 0; i < WIL6210_MAX_TX_RINGS; i++) {
971 		v = &wil->vring_tx[i];
972 		if (!v->va)
973 			continue;
974 
975 		cid = wil->vring2cid_tid[i][0];
976 		if (cid >= WIL6210_MAX_CID) /* skip BCAST */
977 			continue;
978 		if (!wil->vring_tx_data[i].dot1x_open &&
979 		    (skb->protocol != cpu_to_be16(ETH_P_PAE)))
980 			continue;
981 
982 		/* don't Tx back to source when re-routing Rx->Tx at the AP */
983 		if (0 == memcmp(wil->sta[cid].addr, src, ETH_ALEN))
984 			continue;
985 
986 		goto found;
987 	}
988 
989 	wil_dbg_txrx(wil, "Tx while no vrings active?\n");
990 
991 	return NULL;
992 
993 found:
994 	wil_dbg_txrx(wil, "BCAST -> ring %d\n", i);
995 	wil_set_da_for_vring(wil, skb, i);
996 
997 	/* find other active vrings and duplicate skb for each */
998 	for (i++; i < WIL6210_MAX_TX_RINGS; i++) {
999 		v2 = &wil->vring_tx[i];
1000 		if (!v2->va)
1001 			continue;
1002 		cid = wil->vring2cid_tid[i][0];
1003 		if (cid >= WIL6210_MAX_CID) /* skip BCAST */
1004 			continue;
1005 		if (!wil->vring_tx_data[i].dot1x_open &&
1006 		    (skb->protocol != cpu_to_be16(ETH_P_PAE)))
1007 			continue;
1008 
1009 		if (0 == memcmp(wil->sta[cid].addr, src, ETH_ALEN))
1010 			continue;
1011 
1012 		skb2 = skb_copy(skb, GFP_ATOMIC);
1013 		if (skb2) {
1014 			wil_dbg_txrx(wil, "BCAST DUP -> ring %d\n", i);
1015 			wil_set_da_for_vring(wil, skb2, i);
1016 			wil_tx_vring(wil, v2, skb2);
1017 		} else {
1018 			wil_err(wil, "skb_copy failed\n");
1019 		}
1020 	}
1021 
1022 	return v;
1023 }
1024 
1025 static struct vring *wil_find_tx_bcast(struct wil6210_priv *wil,
1026 				       struct sk_buff *skb)
1027 {
1028 	struct wireless_dev *wdev = wil->wdev;
1029 
1030 	if (wdev->iftype != NL80211_IFTYPE_AP)
1031 		return wil_find_tx_bcast_2(wil, skb);
1032 
1033 	return wil_find_tx_bcast_1(wil, skb);
1034 }
1035 
1036 static int wil_tx_desc_map(struct vring_tx_desc *d, dma_addr_t pa, u32 len,
1037 			   int vring_index)
1038 {
1039 	wil_desc_addr_set(&d->dma.addr, pa);
1040 	d->dma.ip_length = 0;
1041 	/* 0..6: mac_length; 7:ip_version 0-IP6 1-IP4*/
1042 	d->dma.b11 = 0/*14 | BIT(7)*/;
1043 	d->dma.error = 0;
1044 	d->dma.status = 0; /* BIT(0) should be 0 for HW_OWNED */
1045 	d->dma.length = cpu_to_le16((u16)len);
1046 	d->dma.d0 = (vring_index << DMA_CFG_DESC_TX_0_QID_POS);
1047 	d->mac.d[0] = 0;
1048 	d->mac.d[1] = 0;
1049 	d->mac.d[2] = 0;
1050 	d->mac.ucode_cmd = 0;
1051 	/* translation type:  0 - bypass; 1 - 802.3; 2 - native wifi */
1052 	d->mac.d[2] = BIT(MAC_CFG_DESC_TX_2_SNAP_HDR_INSERTION_EN_POS) |
1053 		      (1 << MAC_CFG_DESC_TX_2_L2_TRANSLATION_TYPE_POS);
1054 
1055 	return 0;
1056 }
1057 
1058 static inline
1059 void wil_tx_desc_set_nr_frags(struct vring_tx_desc *d, int nr_frags)
1060 {
1061 	d->mac.d[2] |= ((nr_frags + 1) <<
1062 		       MAC_CFG_DESC_TX_2_NUM_OF_DESCRIPTORS_POS);
1063 }
1064 
1065 static int wil_tx_desc_offload_cksum_set(struct wil6210_priv *wil,
1066 					 struct vring_tx_desc *d,
1067 					 struct sk_buff *skb)
1068 {
1069 	int protocol;
1070 
1071 	if (skb->ip_summed != CHECKSUM_PARTIAL)
1072 		return 0;
1073 
1074 	d->dma.b11 = ETH_HLEN; /* MAC header length */
1075 
1076 	switch (skb->protocol) {
1077 	case cpu_to_be16(ETH_P_IP):
1078 		protocol = ip_hdr(skb)->protocol;
1079 		d->dma.b11 |= BIT(DMA_CFG_DESC_TX_OFFLOAD_CFG_L3T_IPV4_POS);
1080 		break;
1081 	case cpu_to_be16(ETH_P_IPV6):
1082 		protocol = ipv6_hdr(skb)->nexthdr;
1083 		break;
1084 	default:
1085 		return -EINVAL;
1086 	}
1087 
1088 	switch (protocol) {
1089 	case IPPROTO_TCP:
1090 		d->dma.d0 |= (2 << DMA_CFG_DESC_TX_0_L4_TYPE_POS);
1091 		/* L4 header len: TCP header length */
1092 		d->dma.d0 |=
1093 		(tcp_hdrlen(skb) & DMA_CFG_DESC_TX_0_L4_LENGTH_MSK);
1094 		break;
1095 	case IPPROTO_UDP:
1096 		/* L4 header len: UDP header length */
1097 		d->dma.d0 |=
1098 		(sizeof(struct udphdr) & DMA_CFG_DESC_TX_0_L4_LENGTH_MSK);
1099 		break;
1100 	default:
1101 		return -EINVAL;
1102 	}
1103 
1104 	d->dma.ip_length = skb_network_header_len(skb);
1105 	/* Enable TCP/UDP checksum */
1106 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_TCP_UDP_CHECKSUM_EN_POS);
1107 	/* Calculate pseudo-header */
1108 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_PSEUDO_HEADER_CALC_EN_POS);
1109 
1110 	return 0;
1111 }
1112 
1113 static int __wil_tx_vring(struct wil6210_priv *wil, struct vring *vring,
1114 			  struct sk_buff *skb)
1115 {
1116 	struct device *dev = wil_to_dev(wil);
1117 	struct vring_tx_desc dd, *d = &dd;
1118 	volatile struct vring_tx_desc *_d;
1119 	u32 swhead = vring->swhead;
1120 	int avail = wil_vring_avail_tx(vring);
1121 	int nr_frags = skb_shinfo(skb)->nr_frags;
1122 	uint f = 0;
1123 	int vring_index = vring - wil->vring_tx;
1124 	struct vring_tx_data *txdata = &wil->vring_tx_data[vring_index];
1125 	uint i = swhead;
1126 	dma_addr_t pa;
1127 	int used;
1128 	bool mcast = (vring_index == wil->bcast_vring);
1129 	uint len = skb_headlen(skb);
1130 
1131 	wil_dbg_txrx(wil, "%s()\n", __func__);
1132 
1133 	if (unlikely(!txdata->enabled))
1134 		return -EINVAL;
1135 
1136 	if (unlikely(avail < 1 + nr_frags)) {
1137 		wil_err_ratelimited(wil,
1138 				    "Tx ring[%2d] full. No space for %d fragments\n",
1139 				    vring_index, 1 + nr_frags);
1140 		return -ENOMEM;
1141 	}
1142 	_d = &vring->va[i].tx;
1143 
1144 	pa = dma_map_single(dev, skb->data, skb_headlen(skb), DMA_TO_DEVICE);
1145 
1146 	wil_dbg_txrx(wil, "Tx[%2d] skb %d bytes 0x%p -> %pad\n", vring_index,
1147 		     skb_headlen(skb), skb->data, &pa);
1148 	wil_hex_dump_txrx("Tx ", DUMP_PREFIX_OFFSET, 16, 1,
1149 			  skb->data, skb_headlen(skb), false);
1150 
1151 	if (unlikely(dma_mapping_error(dev, pa)))
1152 		return -EINVAL;
1153 	vring->ctx[i].mapped_as = wil_mapped_as_single;
1154 	/* 1-st segment */
1155 	wil_tx_desc_map(d, pa, len, vring_index);
1156 	if (unlikely(mcast)) {
1157 		d->mac.d[0] |= BIT(MAC_CFG_DESC_TX_0_MCS_EN_POS); /* MCS 0 */
1158 		if (unlikely(len > WIL_BCAST_MCS0_LIMIT)) /* set MCS 1 */
1159 			d->mac.d[0] |= (1 << MAC_CFG_DESC_TX_0_MCS_INDEX_POS);
1160 	}
1161 	/* Process TCP/UDP checksum offloading */
1162 	if (unlikely(wil_tx_desc_offload_cksum_set(wil, d, skb))) {
1163 		wil_err(wil, "Tx[%2d] Failed to set cksum, drop packet\n",
1164 			vring_index);
1165 		goto dma_error;
1166 	}
1167 
1168 	vring->ctx[i].nr_frags = nr_frags;
1169 	wil_tx_desc_set_nr_frags(d, nr_frags);
1170 
1171 	/* middle segments */
1172 	for (; f < nr_frags; f++) {
1173 		const struct skb_frag_struct *frag =
1174 				&skb_shinfo(skb)->frags[f];
1175 		int len = skb_frag_size(frag);
1176 
1177 		*_d = *d;
1178 		wil_dbg_txrx(wil, "Tx[%2d] desc[%4d]\n", vring_index, i);
1179 		wil_hex_dump_txrx("TxD ", DUMP_PREFIX_NONE, 32, 4,
1180 				  (const void *)d, sizeof(*d), false);
1181 		i = (swhead + f + 1) % vring->size;
1182 		_d = &vring->va[i].tx;
1183 		pa = skb_frag_dma_map(dev, frag, 0, skb_frag_size(frag),
1184 				      DMA_TO_DEVICE);
1185 		if (unlikely(dma_mapping_error(dev, pa)))
1186 			goto dma_error;
1187 		vring->ctx[i].mapped_as = wil_mapped_as_page;
1188 		wil_tx_desc_map(d, pa, len, vring_index);
1189 		/* no need to check return code -
1190 		 * if it succeeded for 1-st descriptor,
1191 		 * it will succeed here too
1192 		 */
1193 		wil_tx_desc_offload_cksum_set(wil, d, skb);
1194 	}
1195 	/* for the last seg only */
1196 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_CMD_EOP_POS);
1197 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_CMD_MARK_WB_POS);
1198 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_CMD_DMA_IT_POS);
1199 	*_d = *d;
1200 	wil_dbg_txrx(wil, "Tx[%2d] desc[%4d]\n", vring_index, i);
1201 	wil_hex_dump_txrx("TxD ", DUMP_PREFIX_NONE, 32, 4,
1202 			  (const void *)d, sizeof(*d), false);
1203 
1204 	/* hold reference to skb
1205 	 * to prevent skb release before accounting
1206 	 * in case of immediate "tx done"
1207 	 */
1208 	vring->ctx[i].skb = skb_get(skb);
1209 
1210 	/* performance monitoring */
1211 	used = wil_vring_used_tx(vring);
1212 	if (wil_val_in_range(vring_idle_trsh,
1213 			     used, used + nr_frags + 1)) {
1214 		txdata->idle += get_cycles() - txdata->last_idle;
1215 		wil_dbg_txrx(wil,  "Ring[%2d] not idle %d -> %d\n",
1216 			     vring_index, used, used + nr_frags + 1);
1217 	}
1218 
1219 	/* advance swhead */
1220 	wil_vring_advance_head(vring, nr_frags + 1);
1221 	wil_dbg_txrx(wil, "Tx[%2d] swhead %d -> %d\n", vring_index, swhead,
1222 		     vring->swhead);
1223 	trace_wil6210_tx(vring_index, swhead, skb->len, nr_frags);
1224 	iowrite32(vring->swhead, wil->csr + HOSTADDR(vring->hwtail));
1225 
1226 	return 0;
1227  dma_error:
1228 	/* unmap what we have mapped */
1229 	nr_frags = f + 1; /* frags mapped + one for skb head */
1230 	for (f = 0; f < nr_frags; f++) {
1231 		struct wil_ctx *ctx;
1232 
1233 		i = (swhead + f) % vring->size;
1234 		ctx = &vring->ctx[i];
1235 		_d = &vring->va[i].tx;
1236 		*d = *_d;
1237 		_d->dma.status = TX_DMA_STATUS_DU;
1238 		wil_txdesc_unmap(dev, d, ctx);
1239 
1240 		if (ctx->skb)
1241 			dev_kfree_skb_any(ctx->skb);
1242 
1243 		memset(ctx, 0, sizeof(*ctx));
1244 	}
1245 
1246 	return -EINVAL;
1247 }
1248 
1249 static int wil_tx_vring(struct wil6210_priv *wil, struct vring *vring,
1250 			struct sk_buff *skb)
1251 {
1252 	int vring_index = vring - wil->vring_tx;
1253 	struct vring_tx_data *txdata = &wil->vring_tx_data[vring_index];
1254 	int rc;
1255 
1256 	spin_lock(&txdata->lock);
1257 	rc = __wil_tx_vring(wil, vring, skb);
1258 	spin_unlock(&txdata->lock);
1259 	return rc;
1260 }
1261 
1262 netdev_tx_t wil_start_xmit(struct sk_buff *skb, struct net_device *ndev)
1263 {
1264 	struct wil6210_priv *wil = ndev_to_wil(ndev);
1265 	struct ethhdr *eth = (void *)skb->data;
1266 	bool bcast = is_multicast_ether_addr(eth->h_dest);
1267 	struct vring *vring;
1268 	static bool pr_once_fw;
1269 	int rc;
1270 
1271 	wil_dbg_txrx(wil, "%s()\n", __func__);
1272 	if (unlikely(!test_bit(wil_status_fwready, wil->status))) {
1273 		if (!pr_once_fw) {
1274 			wil_err(wil, "FW not ready\n");
1275 			pr_once_fw = true;
1276 		}
1277 		goto drop;
1278 	}
1279 	if (unlikely(!test_bit(wil_status_fwconnected, wil->status))) {
1280 		wil_err(wil, "FW not connected\n");
1281 		goto drop;
1282 	}
1283 	if (unlikely(wil->wdev->iftype == NL80211_IFTYPE_MONITOR)) {
1284 		wil_err(wil, "Xmit in monitor mode not supported\n");
1285 		goto drop;
1286 	}
1287 	pr_once_fw = false;
1288 
1289 	/* find vring */
1290 	if (wil->wdev->iftype == NL80211_IFTYPE_STATION) {
1291 		/* in STA mode (ESS), all to same VRING */
1292 		vring = wil_find_tx_vring_sta(wil, skb);
1293 	} else { /* direct communication, find matching VRING */
1294 		vring = bcast ? wil_find_tx_bcast(wil, skb) :
1295 				wil_find_tx_ucast(wil, skb);
1296 	}
1297 	if (unlikely(!vring)) {
1298 		wil_dbg_txrx(wil, "No Tx VRING found for %pM\n", eth->h_dest);
1299 		goto drop;
1300 	}
1301 	/* set up vring entry */
1302 	rc = wil_tx_vring(wil, vring, skb);
1303 
1304 	/* do we still have enough room in the vring? */
1305 	if (unlikely(wil_vring_avail_tx(vring) < wil_vring_wmark_low(vring))) {
1306 		netif_tx_stop_all_queues(wil_to_ndev(wil));
1307 		wil_dbg_txrx(wil, "netif_tx_stop : ring full\n");
1308 	}
1309 
1310 	switch (rc) {
1311 	case 0:
1312 		/* statistics will be updated on the tx_complete */
1313 		dev_kfree_skb_any(skb);
1314 		return NETDEV_TX_OK;
1315 	case -ENOMEM:
1316 		return NETDEV_TX_BUSY;
1317 	default:
1318 		break; /* goto drop; */
1319 	}
1320  drop:
1321 	ndev->stats.tx_dropped++;
1322 	dev_kfree_skb_any(skb);
1323 
1324 	return NET_XMIT_DROP;
1325 }
1326 
1327 static inline bool wil_need_txstat(struct sk_buff *skb)
1328 {
1329 	struct ethhdr *eth = (void *)skb->data;
1330 
1331 	return is_unicast_ether_addr(eth->h_dest) && skb->sk &&
1332 	       (skb_shinfo(skb)->tx_flags & SKBTX_WIFI_STATUS);
1333 }
1334 
1335 static inline void wil_consume_skb(struct sk_buff *skb, bool acked)
1336 {
1337 	if (unlikely(wil_need_txstat(skb)))
1338 		skb_complete_wifi_ack(skb, acked);
1339 	else
1340 		acked ? dev_consume_skb_any(skb) : dev_kfree_skb_any(skb);
1341 }
1342 
1343 /**
1344  * Clean up transmitted skb's from the Tx VRING
1345  *
1346  * Return number of descriptors cleared
1347  *
1348  * Safe to call from IRQ
1349  */
1350 int wil_tx_complete(struct wil6210_priv *wil, int ringid)
1351 {
1352 	struct net_device *ndev = wil_to_ndev(wil);
1353 	struct device *dev = wil_to_dev(wil);
1354 	struct vring *vring = &wil->vring_tx[ringid];
1355 	struct vring_tx_data *txdata = &wil->vring_tx_data[ringid];
1356 	int done = 0;
1357 	int cid = wil->vring2cid_tid[ringid][0];
1358 	struct wil_net_stats *stats = NULL;
1359 	volatile struct vring_tx_desc *_d;
1360 	int used_before_complete;
1361 	int used_new;
1362 
1363 	if (unlikely(!vring->va)) {
1364 		wil_err(wil, "Tx irq[%d]: vring not initialized\n", ringid);
1365 		return 0;
1366 	}
1367 
1368 	if (unlikely(!txdata->enabled)) {
1369 		wil_info(wil, "Tx irq[%d]: vring disabled\n", ringid);
1370 		return 0;
1371 	}
1372 
1373 	wil_dbg_txrx(wil, "%s(%d)\n", __func__, ringid);
1374 
1375 	used_before_complete = wil_vring_used_tx(vring);
1376 
1377 	if (cid < WIL6210_MAX_CID)
1378 		stats = &wil->sta[cid].stats;
1379 
1380 	while (!wil_vring_is_empty(vring)) {
1381 		int new_swtail;
1382 		struct wil_ctx *ctx = &vring->ctx[vring->swtail];
1383 		/**
1384 		 * For the fragmented skb, HW will set DU bit only for the
1385 		 * last fragment. look for it
1386 		 */
1387 		int lf = (vring->swtail + ctx->nr_frags) % vring->size;
1388 		/* TODO: check we are not past head */
1389 
1390 		_d = &vring->va[lf].tx;
1391 		if (unlikely(!(_d->dma.status & TX_DMA_STATUS_DU)))
1392 			break;
1393 
1394 		new_swtail = (lf + 1) % vring->size;
1395 		while (vring->swtail != new_swtail) {
1396 			struct vring_tx_desc dd, *d = &dd;
1397 			u16 dmalen;
1398 			struct sk_buff *skb;
1399 
1400 			ctx = &vring->ctx[vring->swtail];
1401 			skb = ctx->skb;
1402 			_d = &vring->va[vring->swtail].tx;
1403 
1404 			*d = *_d;
1405 
1406 			dmalen = le16_to_cpu(d->dma.length);
1407 			trace_wil6210_tx_done(ringid, vring->swtail, dmalen,
1408 					      d->dma.error);
1409 			wil_dbg_txrx(wil,
1410 				     "TxC[%2d][%3d] : %d bytes, status 0x%02x err 0x%02x\n",
1411 				     ringid, vring->swtail, dmalen,
1412 				     d->dma.status, d->dma.error);
1413 			wil_hex_dump_txrx("TxCD ", DUMP_PREFIX_NONE, 32, 4,
1414 					  (const void *)d, sizeof(*d), false);
1415 
1416 			wil_txdesc_unmap(dev, d, ctx);
1417 
1418 			if (skb) {
1419 				if (likely(d->dma.error == 0)) {
1420 					ndev->stats.tx_packets++;
1421 					ndev->stats.tx_bytes += skb->len;
1422 					if (stats) {
1423 						stats->tx_packets++;
1424 						stats->tx_bytes += skb->len;
1425 					}
1426 				} else {
1427 					ndev->stats.tx_errors++;
1428 					if (stats)
1429 						stats->tx_errors++;
1430 				}
1431 				wil_consume_skb(skb, d->dma.error == 0);
1432 			}
1433 			memset(ctx, 0, sizeof(*ctx));
1434 			/* There is no need to touch HW descriptor:
1435 			 * - ststus bit TX_DMA_STATUS_DU is set by design,
1436 			 *   so hardware will not try to process this desc.,
1437 			 * - rest of descriptor will be initialized on Tx.
1438 			 */
1439 			vring->swtail = wil_vring_next_tail(vring);
1440 			done++;
1441 		}
1442 	}
1443 
1444 	/* performance monitoring */
1445 	used_new = wil_vring_used_tx(vring);
1446 	if (wil_val_in_range(vring_idle_trsh,
1447 			     used_new, used_before_complete)) {
1448 		wil_dbg_txrx(wil, "Ring[%2d] idle %d -> %d\n",
1449 			     ringid, used_before_complete, used_new);
1450 		txdata->last_idle = get_cycles();
1451 	}
1452 
1453 	if (wil_vring_avail_tx(vring) > wil_vring_wmark_high(vring)) {
1454 		wil_dbg_txrx(wil, "netif_tx_wake : ring not full\n");
1455 		netif_tx_wake_all_queues(wil_to_ndev(wil));
1456 	}
1457 
1458 	return done;
1459 }
1460