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 /* similar to ieee80211_ version, but FC contain only 1-st byte */
362 static inline int wil_is_back_req(u8 fc)
363 {
364 	return (fc & (IEEE80211_FCTL_FTYPE | IEEE80211_FCTL_STYPE)) ==
365 	       (IEEE80211_FTYPE_CTL | IEEE80211_STYPE_BACK_REQ);
366 }
367 
368 /**
369  * reap 1 frame from @swhead
370  *
371  * Rx descriptor copied to skb->cb
372  *
373  * Safe to call from IRQ
374  */
375 static struct sk_buff *wil_vring_reap_rx(struct wil6210_priv *wil,
376 					 struct vring *vring)
377 {
378 	struct device *dev = wil_to_dev(wil);
379 	struct net_device *ndev = wil_to_ndev(wil);
380 	volatile struct vring_rx_desc *_d;
381 	struct vring_rx_desc *d;
382 	struct sk_buff *skb;
383 	dma_addr_t pa;
384 	unsigned int snaplen = wil_rx_snaplen();
385 	unsigned int sz = mtu_max + ETH_HLEN + snaplen;
386 	u16 dmalen;
387 	u8 ftype;
388 	int cid;
389 	int i;
390 	struct wil_net_stats *stats;
391 
392 	BUILD_BUG_ON(sizeof(struct vring_rx_desc) > sizeof(skb->cb));
393 
394 again:
395 	if (unlikely(wil_vring_is_empty(vring)))
396 		return NULL;
397 
398 	i = (int)vring->swhead;
399 	_d = &vring->va[i].rx;
400 	if (unlikely(!(_d->dma.status & RX_DMA_STATUS_DU))) {
401 		/* it is not error, we just reached end of Rx done area */
402 		return NULL;
403 	}
404 
405 	skb = vring->ctx[i].skb;
406 	vring->ctx[i].skb = NULL;
407 	wil_vring_advance_head(vring, 1);
408 	if (!skb) {
409 		wil_err(wil, "No Rx skb at [%d]\n", i);
410 		goto again;
411 	}
412 	d = wil_skb_rxdesc(skb);
413 	*d = *_d;
414 	pa = wil_desc_addr(&d->dma.addr);
415 
416 	dma_unmap_single(dev, pa, sz, DMA_FROM_DEVICE);
417 	dmalen = le16_to_cpu(d->dma.length);
418 
419 	trace_wil6210_rx(i, d);
420 	wil_dbg_txrx(wil, "Rx[%3d] : %d bytes\n", i, dmalen);
421 	wil_hex_dump_txrx("RxD ", DUMP_PREFIX_NONE, 32, 4,
422 			  (const void *)d, sizeof(*d), false);
423 
424 	cid = wil_rxdesc_cid(d);
425 	stats = &wil->sta[cid].stats;
426 
427 	if (unlikely(dmalen > sz)) {
428 		wil_err(wil, "Rx size too large: %d bytes!\n", dmalen);
429 		stats->rx_large_frame++;
430 		kfree_skb(skb);
431 		goto again;
432 	}
433 	skb_trim(skb, dmalen);
434 
435 	prefetch(skb->data);
436 
437 	wil_hex_dump_txrx("Rx ", DUMP_PREFIX_OFFSET, 16, 1,
438 			  skb->data, skb_headlen(skb), false);
439 
440 	stats->last_mcs_rx = wil_rxdesc_mcs(d);
441 	if (stats->last_mcs_rx < ARRAY_SIZE(stats->rx_per_mcs))
442 		stats->rx_per_mcs[stats->last_mcs_rx]++;
443 
444 	/* use radiotap header only if required */
445 	if (ndev->type == ARPHRD_IEEE80211_RADIOTAP)
446 		wil_rx_add_radiotap_header(wil, skb);
447 
448 	/* no extra checks if in sniffer mode */
449 	if (ndev->type != ARPHRD_ETHER)
450 		return skb;
451 	/* Non-data frames may be delivered through Rx DMA channel (ex: BAR)
452 	 * Driver should recognize it by frame type, that is found
453 	 * in Rx descriptor. If type is not data, it is 802.11 frame as is
454 	 */
455 	ftype = wil_rxdesc_ftype(d) << 2;
456 	if (unlikely(ftype != IEEE80211_FTYPE_DATA)) {
457 		u8 fc1 = wil_rxdesc_fc1(d);
458 		int mid = wil_rxdesc_mid(d);
459 		int tid = wil_rxdesc_tid(d);
460 		u16 seq = wil_rxdesc_seq(d);
461 
462 		wil_dbg_txrx(wil,
463 			     "Non-data frame FC[7:0] 0x%02x MID %d CID %d TID %d Seq 0x%03x\n",
464 			     fc1, mid, cid, tid, seq);
465 		stats->rx_non_data_frame++;
466 		if (wil_is_back_req(fc1)) {
467 			wil_dbg_txrx(wil,
468 				     "BAR: MID %d CID %d TID %d Seq 0x%03x\n",
469 				     mid, cid, tid, seq);
470 			wil_rx_bar(wil, cid, tid, seq);
471 		} else {
472 			/* print again all info. One can enable only this
473 			 * without overhead for printing every Rx frame
474 			 */
475 			wil_dbg_txrx(wil,
476 				     "Unhandled non-data frame FC[7:0] 0x%02x MID %d CID %d TID %d Seq 0x%03x\n",
477 				     fc1, mid, cid, tid, seq);
478 			wil_hex_dump_txrx("RxD ", DUMP_PREFIX_NONE, 32, 4,
479 					  (const void *)d, sizeof(*d), false);
480 			wil_hex_dump_txrx("Rx ", DUMP_PREFIX_OFFSET, 16, 1,
481 					  skb->data, skb_headlen(skb), false);
482 		}
483 		kfree_skb(skb);
484 		goto again;
485 	}
486 
487 	if (unlikely(skb->len < ETH_HLEN + snaplen)) {
488 		wil_err(wil, "Short frame, len = %d\n", skb->len);
489 		stats->rx_short_frame++;
490 		kfree_skb(skb);
491 		goto again;
492 	}
493 
494 	/* L4 IDENT is on when HW calculated checksum, check status
495 	 * and in case of error drop the packet
496 	 * higher stack layers will handle retransmission (if required)
497 	 */
498 	if (likely(d->dma.status & RX_DMA_STATUS_L4I)) {
499 		/* L4 protocol identified, csum calculated */
500 		if (likely((d->dma.error & RX_DMA_ERROR_L4_ERR) == 0))
501 			skb->ip_summed = CHECKSUM_UNNECESSARY;
502 		/* If HW reports bad checksum, let IP stack re-check it
503 		 * For example, HW don't understand Microsoft IP stack that
504 		 * mis-calculates TCP checksum - if it should be 0x0,
505 		 * it writes 0xffff in violation of RFC 1624
506 		 */
507 	}
508 
509 	if (snaplen) {
510 		/* Packet layout
511 		 * +-------+-------+---------+------------+------+
512 		 * | SA(6) | DA(6) | SNAP(6) | ETHTYPE(2) | DATA |
513 		 * +-------+-------+---------+------------+------+
514 		 * Need to remove SNAP, shifting SA and DA forward
515 		 */
516 		memmove(skb->data + snaplen, skb->data, 2 * ETH_ALEN);
517 		skb_pull(skb, snaplen);
518 	}
519 
520 	return skb;
521 }
522 
523 /**
524  * allocate and fill up to @count buffers in rx ring
525  * buffers posted at @swtail
526  */
527 static int wil_rx_refill(struct wil6210_priv *wil, int count)
528 {
529 	struct net_device *ndev = wil_to_ndev(wil);
530 	struct vring *v = &wil->vring_rx;
531 	u32 next_tail;
532 	int rc = 0;
533 	int headroom = ndev->type == ARPHRD_IEEE80211_RADIOTAP ?
534 			WIL6210_RTAP_SIZE : 0;
535 
536 	for (; next_tail = wil_vring_next_tail(v),
537 			(next_tail != v->swhead) && (count-- > 0);
538 			v->swtail = next_tail) {
539 		rc = wil_vring_alloc_skb(wil, v, v->swtail, headroom);
540 		if (unlikely(rc)) {
541 			wil_err(wil, "Error %d in wil_rx_refill[%d]\n",
542 				rc, v->swtail);
543 			break;
544 		}
545 	}
546 	wil_w(wil, v->hwtail, v->swtail);
547 
548 	return rc;
549 }
550 
551 /*
552  * Pass Rx packet to the netif. Update statistics.
553  * Called in softirq context (NAPI poll).
554  */
555 void wil_netif_rx_any(struct sk_buff *skb, struct net_device *ndev)
556 {
557 	gro_result_t rc = GRO_NORMAL;
558 	struct wil6210_priv *wil = ndev_to_wil(ndev);
559 	struct wireless_dev *wdev = wil_to_wdev(wil);
560 	unsigned int len = skb->len;
561 	struct vring_rx_desc *d = wil_skb_rxdesc(skb);
562 	int cid = wil_rxdesc_cid(d); /* always 0..7, no need to check */
563 	struct ethhdr *eth = (void *)skb->data;
564 	/* here looking for DA, not A1, thus Rxdesc's 'mcast' indication
565 	 * is not suitable, need to look at data
566 	 */
567 	int mcast = is_multicast_ether_addr(eth->h_dest);
568 	struct wil_net_stats *stats = &wil->sta[cid].stats;
569 	struct sk_buff *xmit_skb = NULL;
570 	static const char * const gro_res_str[] = {
571 		[GRO_MERGED]		= "GRO_MERGED",
572 		[GRO_MERGED_FREE]	= "GRO_MERGED_FREE",
573 		[GRO_HELD]		= "GRO_HELD",
574 		[GRO_NORMAL]		= "GRO_NORMAL",
575 		[GRO_DROP]		= "GRO_DROP",
576 	};
577 
578 	if (ndev->features & NETIF_F_RXHASH)
579 		/* fake L4 to ensure it won't be re-calculated later
580 		 * set hash to any non-zero value to activate rps
581 		 * mechanism, core will be chosen according
582 		 * to user-level rps configuration.
583 		 */
584 		skb_set_hash(skb, 1, PKT_HASH_TYPE_L4);
585 
586 	skb_orphan(skb);
587 
588 	if (wdev->iftype == NL80211_IFTYPE_AP && !wil->ap_isolate) {
589 		if (mcast) {
590 			/* send multicast frames both to higher layers in
591 			 * local net stack and back to the wireless medium
592 			 */
593 			xmit_skb = skb_copy(skb, GFP_ATOMIC);
594 		} else {
595 			int xmit_cid = wil_find_cid(wil, eth->h_dest);
596 
597 			if (xmit_cid >= 0) {
598 				/* The destination station is associated to
599 				 * this AP (in this VLAN), so send the frame
600 				 * directly to it and do not pass it to local
601 				 * net stack.
602 				 */
603 				xmit_skb = skb;
604 				skb = NULL;
605 			}
606 		}
607 	}
608 	if (xmit_skb) {
609 		/* Send to wireless media and increase priority by 256 to
610 		 * keep the received priority instead of reclassifying
611 		 * the frame (see cfg80211_classify8021d).
612 		 */
613 		xmit_skb->dev = ndev;
614 		xmit_skb->priority += 256;
615 		xmit_skb->protocol = htons(ETH_P_802_3);
616 		skb_reset_network_header(xmit_skb);
617 		skb_reset_mac_header(xmit_skb);
618 		wil_dbg_txrx(wil, "Rx -> Tx %d bytes\n", len);
619 		dev_queue_xmit(xmit_skb);
620 	}
621 
622 	if (skb) { /* deliver to local stack */
623 
624 		skb->protocol = eth_type_trans(skb, ndev);
625 		rc = napi_gro_receive(&wil->napi_rx, skb);
626 		wil_dbg_txrx(wil, "Rx complete %d bytes => %s\n",
627 			     len, gro_res_str[rc]);
628 	}
629 	/* statistics. rc set to GRO_NORMAL for AP bridging */
630 	if (unlikely(rc == GRO_DROP)) {
631 		ndev->stats.rx_dropped++;
632 		stats->rx_dropped++;
633 		wil_dbg_txrx(wil, "Rx drop %d bytes\n", len);
634 	} else {
635 		ndev->stats.rx_packets++;
636 		stats->rx_packets++;
637 		ndev->stats.rx_bytes += len;
638 		stats->rx_bytes += len;
639 		if (mcast)
640 			ndev->stats.multicast++;
641 	}
642 }
643 
644 /**
645  * Proceed all completed skb's from Rx VRING
646  *
647  * Safe to call from NAPI poll, i.e. softirq with interrupts enabled
648  */
649 void wil_rx_handle(struct wil6210_priv *wil, int *quota)
650 {
651 	struct net_device *ndev = wil_to_ndev(wil);
652 	struct vring *v = &wil->vring_rx;
653 	struct sk_buff *skb;
654 
655 	if (unlikely(!v->va)) {
656 		wil_err(wil, "Rx IRQ while Rx not yet initialized\n");
657 		return;
658 	}
659 	wil_dbg_txrx(wil, "%s()\n", __func__);
660 	while ((*quota > 0) && (NULL != (skb = wil_vring_reap_rx(wil, v)))) {
661 		(*quota)--;
662 
663 		if (wil->wdev->iftype == NL80211_IFTYPE_MONITOR) {
664 			skb->dev = ndev;
665 			skb_reset_mac_header(skb);
666 			skb->ip_summed = CHECKSUM_UNNECESSARY;
667 			skb->pkt_type = PACKET_OTHERHOST;
668 			skb->protocol = htons(ETH_P_802_2);
669 			wil_netif_rx_any(skb, ndev);
670 		} else {
671 			wil_rx_reorder(wil, skb);
672 		}
673 	}
674 	wil_rx_refill(wil, v->size);
675 }
676 
677 int wil_rx_init(struct wil6210_priv *wil, u16 size)
678 {
679 	struct vring *vring = &wil->vring_rx;
680 	int rc;
681 
682 	wil_dbg_misc(wil, "%s()\n", __func__);
683 
684 	if (vring->va) {
685 		wil_err(wil, "Rx ring already allocated\n");
686 		return -EINVAL;
687 	}
688 
689 	vring->size = size;
690 	rc = wil_vring_alloc(wil, vring);
691 	if (rc)
692 		return rc;
693 
694 	rc = wmi_rx_chain_add(wil, vring);
695 	if (rc)
696 		goto err_free;
697 
698 	rc = wil_rx_refill(wil, vring->size);
699 	if (rc)
700 		goto err_free;
701 
702 	return 0;
703  err_free:
704 	wil_vring_free(wil, vring, 0);
705 
706 	return rc;
707 }
708 
709 void wil_rx_fini(struct wil6210_priv *wil)
710 {
711 	struct vring *vring = &wil->vring_rx;
712 
713 	wil_dbg_misc(wil, "%s()\n", __func__);
714 
715 	if (vring->va)
716 		wil_vring_free(wil, vring, 0);
717 }
718 
719 int wil_vring_init_tx(struct wil6210_priv *wil, int id, int size,
720 		      int cid, int tid)
721 {
722 	int rc;
723 	struct wmi_vring_cfg_cmd cmd = {
724 		.action = cpu_to_le32(WMI_VRING_CMD_ADD),
725 		.vring_cfg = {
726 			.tx_sw_ring = {
727 				.max_mpdu_size =
728 					cpu_to_le16(wil_mtu2macbuf(mtu_max)),
729 				.ring_size = cpu_to_le16(size),
730 			},
731 			.ringid = id,
732 			.cidxtid = mk_cidxtid(cid, tid),
733 			.encap_trans_type = WMI_VRING_ENC_TYPE_802_3,
734 			.mac_ctrl = 0,
735 			.to_resolution = 0,
736 			.agg_max_wsize = 0,
737 			.schd_params = {
738 				.priority = cpu_to_le16(0),
739 				.timeslot_us = cpu_to_le16(0xfff),
740 			},
741 		},
742 	};
743 	struct {
744 		struct wil6210_mbox_hdr_wmi wmi;
745 		struct wmi_vring_cfg_done_event cmd;
746 	} __packed reply;
747 	struct vring *vring = &wil->vring_tx[id];
748 	struct vring_tx_data *txdata = &wil->vring_tx_data[id];
749 
750 	wil_dbg_misc(wil, "%s() max_mpdu_size %d\n", __func__,
751 		     cmd.vring_cfg.tx_sw_ring.max_mpdu_size);
752 
753 	if (vring->va) {
754 		wil_err(wil, "Tx ring [%d] already allocated\n", id);
755 		rc = -EINVAL;
756 		goto out;
757 	}
758 
759 	memset(txdata, 0, sizeof(*txdata));
760 	spin_lock_init(&txdata->lock);
761 	vring->size = size;
762 	rc = wil_vring_alloc(wil, vring);
763 	if (rc)
764 		goto out;
765 
766 	wil->vring2cid_tid[id][0] = cid;
767 	wil->vring2cid_tid[id][1] = tid;
768 
769 	cmd.vring_cfg.tx_sw_ring.ring_mem_base = cpu_to_le64(vring->pa);
770 
771 	if (!wil->privacy)
772 		txdata->dot1x_open = true;
773 	rc = wmi_call(wil, WMI_VRING_CFG_CMDID, &cmd, sizeof(cmd),
774 		      WMI_VRING_CFG_DONE_EVENTID, &reply, sizeof(reply), 100);
775 	if (rc)
776 		goto out_free;
777 
778 	if (reply.cmd.status != WMI_FW_STATUS_SUCCESS) {
779 		wil_err(wil, "Tx config failed, status 0x%02x\n",
780 			reply.cmd.status);
781 		rc = -EINVAL;
782 		goto out_free;
783 	}
784 	vring->hwtail = le32_to_cpu(reply.cmd.tx_vring_tail_ptr);
785 
786 	txdata->enabled = 1;
787 	if (txdata->dot1x_open && (agg_wsize >= 0))
788 		wil_addba_tx_request(wil, id, agg_wsize);
789 
790 	return 0;
791  out_free:
792 	txdata->dot1x_open = false;
793 	txdata->enabled = 0;
794 	wil_vring_free(wil, vring, 1);
795  out:
796 
797 	return rc;
798 }
799 
800 int wil_vring_init_bcast(struct wil6210_priv *wil, int id, int size)
801 {
802 	int rc;
803 	struct wmi_bcast_vring_cfg_cmd cmd = {
804 		.action = cpu_to_le32(WMI_VRING_CMD_ADD),
805 		.vring_cfg = {
806 			.tx_sw_ring = {
807 				.max_mpdu_size =
808 					cpu_to_le16(wil_mtu2macbuf(mtu_max)),
809 				.ring_size = cpu_to_le16(size),
810 			},
811 			.ringid = id,
812 			.encap_trans_type = WMI_VRING_ENC_TYPE_802_3,
813 		},
814 	};
815 	struct {
816 		struct wil6210_mbox_hdr_wmi wmi;
817 		struct wmi_vring_cfg_done_event cmd;
818 	} __packed reply;
819 	struct vring *vring = &wil->vring_tx[id];
820 	struct vring_tx_data *txdata = &wil->vring_tx_data[id];
821 
822 	wil_dbg_misc(wil, "%s() max_mpdu_size %d\n", __func__,
823 		     cmd.vring_cfg.tx_sw_ring.max_mpdu_size);
824 
825 	if (vring->va) {
826 		wil_err(wil, "Tx ring [%d] already allocated\n", id);
827 		rc = -EINVAL;
828 		goto out;
829 	}
830 
831 	memset(txdata, 0, sizeof(*txdata));
832 	spin_lock_init(&txdata->lock);
833 	vring->size = size;
834 	rc = wil_vring_alloc(wil, vring);
835 	if (rc)
836 		goto out;
837 
838 	wil->vring2cid_tid[id][0] = WIL6210_MAX_CID; /* CID */
839 	wil->vring2cid_tid[id][1] = 0; /* TID */
840 
841 	cmd.vring_cfg.tx_sw_ring.ring_mem_base = cpu_to_le64(vring->pa);
842 
843 	if (!wil->privacy)
844 		txdata->dot1x_open = true;
845 	rc = wmi_call(wil, WMI_BCAST_VRING_CFG_CMDID, &cmd, sizeof(cmd),
846 		      WMI_VRING_CFG_DONE_EVENTID, &reply, sizeof(reply), 100);
847 	if (rc)
848 		goto out_free;
849 
850 	if (reply.cmd.status != WMI_FW_STATUS_SUCCESS) {
851 		wil_err(wil, "Tx config failed, status 0x%02x\n",
852 			reply.cmd.status);
853 		rc = -EINVAL;
854 		goto out_free;
855 	}
856 	vring->hwtail = le32_to_cpu(reply.cmd.tx_vring_tail_ptr);
857 
858 	txdata->enabled = 1;
859 
860 	return 0;
861  out_free:
862 	txdata->enabled = 0;
863 	txdata->dot1x_open = false;
864 	wil_vring_free(wil, vring, 1);
865  out:
866 
867 	return rc;
868 }
869 
870 void wil_vring_fini_tx(struct wil6210_priv *wil, int id)
871 {
872 	struct vring *vring = &wil->vring_tx[id];
873 	struct vring_tx_data *txdata = &wil->vring_tx_data[id];
874 
875 	WARN_ON(!mutex_is_locked(&wil->mutex));
876 
877 	if (!vring->va)
878 		return;
879 
880 	wil_dbg_misc(wil, "%s() id=%d\n", __func__, id);
881 
882 	spin_lock_bh(&txdata->lock);
883 	txdata->dot1x_open = false;
884 	txdata->enabled = 0; /* no Tx can be in progress or start anew */
885 	spin_unlock_bh(&txdata->lock);
886 	/* make sure NAPI won't touch this vring */
887 	if (test_bit(wil_status_napi_en, wil->status))
888 		napi_synchronize(&wil->napi_tx);
889 
890 	wil_vring_free(wil, vring, 1);
891 	memset(txdata, 0, sizeof(*txdata));
892 }
893 
894 static struct vring *wil_find_tx_ucast(struct wil6210_priv *wil,
895 				       struct sk_buff *skb)
896 {
897 	int i;
898 	struct ethhdr *eth = (void *)skb->data;
899 	int cid = wil_find_cid(wil, eth->h_dest);
900 
901 	if (cid < 0)
902 		return NULL;
903 
904 	/* TODO: fix for multiple TID */
905 	for (i = 0; i < ARRAY_SIZE(wil->vring2cid_tid); i++) {
906 		if (!wil->vring_tx_data[i].dot1x_open &&
907 		    (skb->protocol != cpu_to_be16(ETH_P_PAE)))
908 			continue;
909 		if (wil->vring2cid_tid[i][0] == cid) {
910 			struct vring *v = &wil->vring_tx[i];
911 
912 			wil_dbg_txrx(wil, "%s(%pM) -> [%d]\n",
913 				     __func__, eth->h_dest, i);
914 			if (v->va) {
915 				return v;
916 			} else {
917 				wil_dbg_txrx(wil, "vring[%d] not valid\n", i);
918 				return NULL;
919 			}
920 		}
921 	}
922 
923 	return NULL;
924 }
925 
926 static int wil_tx_vring(struct wil6210_priv *wil, struct vring *vring,
927 			struct sk_buff *skb);
928 
929 static struct vring *wil_find_tx_vring_sta(struct wil6210_priv *wil,
930 					   struct sk_buff *skb)
931 {
932 	struct vring *v;
933 	int i;
934 	u8 cid;
935 
936 	/* In the STA mode, it is expected to have only 1 VRING
937 	 * for the AP we connected to.
938 	 * find 1-st vring eligible for this skb and use it.
939 	 */
940 	for (i = 0; i < WIL6210_MAX_TX_RINGS; i++) {
941 		v = &wil->vring_tx[i];
942 		if (!v->va)
943 			continue;
944 
945 		cid = wil->vring2cid_tid[i][0];
946 		if (cid >= WIL6210_MAX_CID) /* skip BCAST */
947 			continue;
948 
949 		if (!wil->vring_tx_data[i].dot1x_open &&
950 		    (skb->protocol != cpu_to_be16(ETH_P_PAE)))
951 			continue;
952 
953 		wil_dbg_txrx(wil, "Tx -> ring %d\n", i);
954 
955 		return v;
956 	}
957 
958 	wil_dbg_txrx(wil, "Tx while no vrings active?\n");
959 
960 	return NULL;
961 }
962 
963 /* Use one of 2 strategies:
964  *
965  * 1. New (real broadcast):
966  *    use dedicated broadcast vring
967  * 2. Old (pseudo-DMS):
968  *    Find 1-st vring and return it;
969  *    duplicate skb and send it to other active vrings;
970  *    in all cases override dest address to unicast peer's address
971  * Use old strategy when new is not supported yet:
972  *  - for PBSS
973  */
974 static struct vring *wil_find_tx_bcast_1(struct wil6210_priv *wil,
975 					 struct sk_buff *skb)
976 {
977 	struct vring *v;
978 	int i = wil->bcast_vring;
979 
980 	if (i < 0)
981 		return NULL;
982 	v = &wil->vring_tx[i];
983 	if (!v->va)
984 		return NULL;
985 	if (!wil->vring_tx_data[i].dot1x_open &&
986 	    (skb->protocol != cpu_to_be16(ETH_P_PAE)))
987 		return NULL;
988 
989 	return v;
990 }
991 
992 static void wil_set_da_for_vring(struct wil6210_priv *wil,
993 				 struct sk_buff *skb, int vring_index)
994 {
995 	struct ethhdr *eth = (void *)skb->data;
996 	int cid = wil->vring2cid_tid[vring_index][0];
997 
998 	ether_addr_copy(eth->h_dest, wil->sta[cid].addr);
999 }
1000 
1001 static struct vring *wil_find_tx_bcast_2(struct wil6210_priv *wil,
1002 					 struct sk_buff *skb)
1003 {
1004 	struct vring *v, *v2;
1005 	struct sk_buff *skb2;
1006 	int i;
1007 	u8 cid;
1008 	struct ethhdr *eth = (void *)skb->data;
1009 	char *src = eth->h_source;
1010 
1011 	/* find 1-st vring eligible for data */
1012 	for (i = 0; i < WIL6210_MAX_TX_RINGS; i++) {
1013 		v = &wil->vring_tx[i];
1014 		if (!v->va)
1015 			continue;
1016 
1017 		cid = wil->vring2cid_tid[i][0];
1018 		if (cid >= WIL6210_MAX_CID) /* skip BCAST */
1019 			continue;
1020 		if (!wil->vring_tx_data[i].dot1x_open &&
1021 		    (skb->protocol != cpu_to_be16(ETH_P_PAE)))
1022 			continue;
1023 
1024 		/* don't Tx back to source when re-routing Rx->Tx at the AP */
1025 		if (0 == memcmp(wil->sta[cid].addr, src, ETH_ALEN))
1026 			continue;
1027 
1028 		goto found;
1029 	}
1030 
1031 	wil_dbg_txrx(wil, "Tx while no vrings active?\n");
1032 
1033 	return NULL;
1034 
1035 found:
1036 	wil_dbg_txrx(wil, "BCAST -> ring %d\n", i);
1037 	wil_set_da_for_vring(wil, skb, i);
1038 
1039 	/* find other active vrings and duplicate skb for each */
1040 	for (i++; i < WIL6210_MAX_TX_RINGS; i++) {
1041 		v2 = &wil->vring_tx[i];
1042 		if (!v2->va)
1043 			continue;
1044 		cid = wil->vring2cid_tid[i][0];
1045 		if (cid >= WIL6210_MAX_CID) /* skip BCAST */
1046 			continue;
1047 		if (!wil->vring_tx_data[i].dot1x_open &&
1048 		    (skb->protocol != cpu_to_be16(ETH_P_PAE)))
1049 			continue;
1050 
1051 		if (0 == memcmp(wil->sta[cid].addr, src, ETH_ALEN))
1052 			continue;
1053 
1054 		skb2 = skb_copy(skb, GFP_ATOMIC);
1055 		if (skb2) {
1056 			wil_dbg_txrx(wil, "BCAST DUP -> ring %d\n", i);
1057 			wil_set_da_for_vring(wil, skb2, i);
1058 			wil_tx_vring(wil, v2, skb2);
1059 		} else {
1060 			wil_err(wil, "skb_copy failed\n");
1061 		}
1062 	}
1063 
1064 	return v;
1065 }
1066 
1067 static struct vring *wil_find_tx_bcast(struct wil6210_priv *wil,
1068 				       struct sk_buff *skb)
1069 {
1070 	struct wireless_dev *wdev = wil->wdev;
1071 
1072 	if (wdev->iftype != NL80211_IFTYPE_AP)
1073 		return wil_find_tx_bcast_2(wil, skb);
1074 
1075 	return wil_find_tx_bcast_1(wil, skb);
1076 }
1077 
1078 static int wil_tx_desc_map(struct vring_tx_desc *d, dma_addr_t pa, u32 len,
1079 			   int vring_index)
1080 {
1081 	wil_desc_addr_set(&d->dma.addr, pa);
1082 	d->dma.ip_length = 0;
1083 	/* 0..6: mac_length; 7:ip_version 0-IP6 1-IP4*/
1084 	d->dma.b11 = 0/*14 | BIT(7)*/;
1085 	d->dma.error = 0;
1086 	d->dma.status = 0; /* BIT(0) should be 0 for HW_OWNED */
1087 	d->dma.length = cpu_to_le16((u16)len);
1088 	d->dma.d0 = (vring_index << DMA_CFG_DESC_TX_0_QID_POS);
1089 	d->mac.d[0] = 0;
1090 	d->mac.d[1] = 0;
1091 	d->mac.d[2] = 0;
1092 	d->mac.ucode_cmd = 0;
1093 	/* translation type:  0 - bypass; 1 - 802.3; 2 - native wifi */
1094 	d->mac.d[2] = BIT(MAC_CFG_DESC_TX_2_SNAP_HDR_INSERTION_EN_POS) |
1095 		      (1 << MAC_CFG_DESC_TX_2_L2_TRANSLATION_TYPE_POS);
1096 
1097 	return 0;
1098 }
1099 
1100 static inline
1101 void wil_tx_desc_set_nr_frags(struct vring_tx_desc *d, int nr_frags)
1102 {
1103 	d->mac.d[2] |= (nr_frags << MAC_CFG_DESC_TX_2_NUM_OF_DESCRIPTORS_POS);
1104 }
1105 
1106 /**
1107  * Sets the descriptor @d up for csum and/or TSO offloading. The corresponding
1108  * @skb is used to obtain the protocol and headers length.
1109  * @tso_desc_type is a descriptor type for TSO: 0 - a header, 1 - first data,
1110  * 2 - middle, 3 - last descriptor.
1111  */
1112 
1113 static void wil_tx_desc_offload_setup_tso(struct vring_tx_desc *d,
1114 					  struct sk_buff *skb,
1115 					  int tso_desc_type, bool is_ipv4,
1116 					  int tcp_hdr_len, int skb_net_hdr_len)
1117 {
1118 	d->dma.b11 = ETH_HLEN; /* MAC header length */
1119 	d->dma.b11 |= is_ipv4 << DMA_CFG_DESC_TX_OFFLOAD_CFG_L3T_IPV4_POS;
1120 
1121 	d->dma.d0 |= (2 << DMA_CFG_DESC_TX_0_L4_TYPE_POS);
1122 	/* L4 header len: TCP header length */
1123 	d->dma.d0 |= (tcp_hdr_len & DMA_CFG_DESC_TX_0_L4_LENGTH_MSK);
1124 
1125 	/* Setup TSO: bit and desc type */
1126 	d->dma.d0 |= (BIT(DMA_CFG_DESC_TX_0_TCP_SEG_EN_POS)) |
1127 		(tso_desc_type << DMA_CFG_DESC_TX_0_SEGMENT_BUF_DETAILS_POS);
1128 	d->dma.d0 |= (is_ipv4 << DMA_CFG_DESC_TX_0_IPV4_CHECKSUM_EN_POS);
1129 
1130 	d->dma.ip_length = skb_net_hdr_len;
1131 	/* Enable TCP/UDP checksum */
1132 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_TCP_UDP_CHECKSUM_EN_POS);
1133 	/* Calculate pseudo-header */
1134 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_PSEUDO_HEADER_CALC_EN_POS);
1135 }
1136 
1137 /**
1138  * Sets the descriptor @d up for csum. The corresponding
1139  * @skb is used to obtain the protocol and headers length.
1140  * Returns the protocol: 0 - not TCP, 1 - TCPv4, 2 - TCPv6.
1141  * Note, if d==NULL, the function only returns the protocol result.
1142  *
1143  * It is very similar to previous wil_tx_desc_offload_setup_tso. This
1144  * is "if unrolling" to optimize the critical path.
1145  */
1146 
1147 static int wil_tx_desc_offload_setup(struct vring_tx_desc *d,
1148 				     struct sk_buff *skb){
1149 	int protocol;
1150 
1151 	if (skb->ip_summed != CHECKSUM_PARTIAL)
1152 		return 0;
1153 
1154 	d->dma.b11 = ETH_HLEN; /* MAC header length */
1155 
1156 	switch (skb->protocol) {
1157 	case cpu_to_be16(ETH_P_IP):
1158 		protocol = ip_hdr(skb)->protocol;
1159 		d->dma.b11 |= BIT(DMA_CFG_DESC_TX_OFFLOAD_CFG_L3T_IPV4_POS);
1160 		break;
1161 	case cpu_to_be16(ETH_P_IPV6):
1162 		protocol = ipv6_hdr(skb)->nexthdr;
1163 		break;
1164 	default:
1165 		return -EINVAL;
1166 	}
1167 
1168 	switch (protocol) {
1169 	case IPPROTO_TCP:
1170 		d->dma.d0 |= (2 << DMA_CFG_DESC_TX_0_L4_TYPE_POS);
1171 		/* L4 header len: TCP header length */
1172 		d->dma.d0 |=
1173 		(tcp_hdrlen(skb) & DMA_CFG_DESC_TX_0_L4_LENGTH_MSK);
1174 		break;
1175 	case IPPROTO_UDP:
1176 		/* L4 header len: UDP header length */
1177 		d->dma.d0 |=
1178 		(sizeof(struct udphdr) & DMA_CFG_DESC_TX_0_L4_LENGTH_MSK);
1179 		break;
1180 	default:
1181 		return -EINVAL;
1182 	}
1183 
1184 	d->dma.ip_length = skb_network_header_len(skb);
1185 	/* Enable TCP/UDP checksum */
1186 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_TCP_UDP_CHECKSUM_EN_POS);
1187 	/* Calculate pseudo-header */
1188 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_PSEUDO_HEADER_CALC_EN_POS);
1189 
1190 	return 0;
1191 }
1192 
1193 static inline void wil_tx_last_desc(struct vring_tx_desc *d)
1194 {
1195 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_CMD_EOP_POS) |
1196 	      BIT(DMA_CFG_DESC_TX_0_CMD_MARK_WB_POS) |
1197 	      BIT(DMA_CFG_DESC_TX_0_CMD_DMA_IT_POS);
1198 }
1199 
1200 static inline void wil_set_tx_desc_last_tso(volatile struct vring_tx_desc *d)
1201 {
1202 	d->dma.d0 |= wil_tso_type_lst <<
1203 		  DMA_CFG_DESC_TX_0_SEGMENT_BUF_DETAILS_POS;
1204 }
1205 
1206 static int __wil_tx_vring_tso(struct wil6210_priv *wil, struct vring *vring,
1207 			      struct sk_buff *skb)
1208 {
1209 	struct device *dev = wil_to_dev(wil);
1210 
1211 	/* point to descriptors in shared memory */
1212 	volatile struct vring_tx_desc *_desc = NULL, *_hdr_desc,
1213 				      *_first_desc = NULL;
1214 
1215 	/* pointers to shadow descriptors */
1216 	struct vring_tx_desc desc_mem, hdr_desc_mem, first_desc_mem,
1217 			     *d = &hdr_desc_mem, *hdr_desc = &hdr_desc_mem,
1218 			     *first_desc = &first_desc_mem;
1219 
1220 	/* pointer to shadow descriptors' context */
1221 	struct wil_ctx *hdr_ctx, *first_ctx = NULL;
1222 
1223 	int descs_used = 0; /* total number of used descriptors */
1224 	int sg_desc_cnt = 0; /* number of descriptors for current mss*/
1225 
1226 	u32 swhead = vring->swhead;
1227 	int used, avail = wil_vring_avail_tx(vring);
1228 	int nr_frags = skb_shinfo(skb)->nr_frags;
1229 	int min_desc_required = nr_frags + 1;
1230 	int mss = skb_shinfo(skb)->gso_size;	/* payload size w/o headers */
1231 	int f, len, hdrlen, headlen;
1232 	int vring_index = vring - wil->vring_tx;
1233 	struct vring_tx_data *txdata = &wil->vring_tx_data[vring_index];
1234 	uint i = swhead;
1235 	dma_addr_t pa;
1236 	const skb_frag_t *frag = NULL;
1237 	int rem_data = mss;
1238 	int lenmss;
1239 	int hdr_compensation_need = true;
1240 	int desc_tso_type = wil_tso_type_first;
1241 	bool is_ipv4;
1242 	int tcp_hdr_len;
1243 	int skb_net_hdr_len;
1244 	int gso_type;
1245 
1246 	wil_dbg_txrx(wil, "%s() %d bytes to vring %d\n",
1247 		     __func__, skb->len, vring_index);
1248 
1249 	if (unlikely(!txdata->enabled))
1250 		return -EINVAL;
1251 
1252 	/* A typical page 4K is 3-4 payloads, we assume each fragment
1253 	 * is a full payload, that's how min_desc_required has been
1254 	 * calculated. In real we might need more or less descriptors,
1255 	 * this is the initial check only.
1256 	 */
1257 	if (unlikely(avail < min_desc_required)) {
1258 		wil_err_ratelimited(wil,
1259 				    "TSO: Tx ring[%2d] full. No space for %d fragments\n",
1260 				    vring_index, min_desc_required);
1261 		return -ENOMEM;
1262 	}
1263 
1264 	/* Header Length = MAC header len + IP header len + TCP header len*/
1265 	hdrlen = ETH_HLEN +
1266 		(int)skb_network_header_len(skb) +
1267 		tcp_hdrlen(skb);
1268 
1269 	gso_type = skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV6 | SKB_GSO_TCPV4);
1270 	switch (gso_type) {
1271 	case SKB_GSO_TCPV4:
1272 		/* TCP v4, zero out the IP length and IPv4 checksum fields
1273 		 * as required by the offloading doc
1274 		 */
1275 		ip_hdr(skb)->tot_len = 0;
1276 		ip_hdr(skb)->check = 0;
1277 		is_ipv4 = true;
1278 		break;
1279 	case SKB_GSO_TCPV6:
1280 		/* TCP v6, zero out the payload length */
1281 		ipv6_hdr(skb)->payload_len = 0;
1282 		is_ipv4 = false;
1283 		break;
1284 	default:
1285 		/* other than TCPv4 or TCPv6 types are not supported for TSO.
1286 		 * It is also illegal for both to be set simultaneously
1287 		 */
1288 		return -EINVAL;
1289 	}
1290 
1291 	if (skb->ip_summed != CHECKSUM_PARTIAL)
1292 		return -EINVAL;
1293 
1294 	/* tcp header length and skb network header length are fixed for all
1295 	 * packet's descriptors - read then once here
1296 	 */
1297 	tcp_hdr_len = tcp_hdrlen(skb);
1298 	skb_net_hdr_len = skb_network_header_len(skb);
1299 
1300 	_hdr_desc = &vring->va[i].tx;
1301 
1302 	pa = dma_map_single(dev, skb->data, hdrlen, DMA_TO_DEVICE);
1303 	if (unlikely(dma_mapping_error(dev, pa))) {
1304 		wil_err(wil, "TSO: Skb head DMA map error\n");
1305 		goto err_exit;
1306 	}
1307 
1308 	wil_tx_desc_map(hdr_desc, pa, hdrlen, vring_index);
1309 	wil_tx_desc_offload_setup_tso(hdr_desc, skb, wil_tso_type_hdr, is_ipv4,
1310 				      tcp_hdr_len, skb_net_hdr_len);
1311 	wil_tx_last_desc(hdr_desc);
1312 
1313 	vring->ctx[i].mapped_as = wil_mapped_as_single;
1314 	hdr_ctx = &vring->ctx[i];
1315 
1316 	descs_used++;
1317 	headlen = skb_headlen(skb) - hdrlen;
1318 
1319 	for (f = headlen ? -1 : 0; f < nr_frags; f++)  {
1320 		if (headlen) {
1321 			len = headlen;
1322 			wil_dbg_txrx(wil, "TSO: process skb head, len %u\n",
1323 				     len);
1324 		} else {
1325 			frag = &skb_shinfo(skb)->frags[f];
1326 			len = frag->size;
1327 			wil_dbg_txrx(wil, "TSO: frag[%d]: len %u\n", f, len);
1328 		}
1329 
1330 		while (len) {
1331 			wil_dbg_txrx(wil,
1332 				     "TSO: len %d, rem_data %d, descs_used %d\n",
1333 				     len, rem_data, descs_used);
1334 
1335 			if (descs_used == avail)  {
1336 				wil_err(wil, "TSO: ring overflow\n");
1337 				goto dma_error;
1338 			}
1339 
1340 			lenmss = min_t(int, rem_data, len);
1341 			i = (swhead + descs_used) % vring->size;
1342 			wil_dbg_txrx(wil, "TSO: lenmss %d, i %d\n", lenmss, i);
1343 
1344 			if (!headlen) {
1345 				pa = skb_frag_dma_map(dev, frag,
1346 						      frag->size - len, lenmss,
1347 						      DMA_TO_DEVICE);
1348 				vring->ctx[i].mapped_as = wil_mapped_as_page;
1349 			} else {
1350 				pa = dma_map_single(dev,
1351 						    skb->data +
1352 						    skb_headlen(skb) - headlen,
1353 						    lenmss,
1354 						    DMA_TO_DEVICE);
1355 				vring->ctx[i].mapped_as = wil_mapped_as_single;
1356 				headlen -= lenmss;
1357 			}
1358 
1359 			if (unlikely(dma_mapping_error(dev, pa)))
1360 				goto dma_error;
1361 
1362 			_desc = &vring->va[i].tx;
1363 
1364 			if (!_first_desc) {
1365 				_first_desc = _desc;
1366 				first_ctx = &vring->ctx[i];
1367 				d = first_desc;
1368 			} else {
1369 				d = &desc_mem;
1370 			}
1371 
1372 			wil_tx_desc_map(d, pa, lenmss, vring_index);
1373 			wil_tx_desc_offload_setup_tso(d, skb, desc_tso_type,
1374 						      is_ipv4, tcp_hdr_len,
1375 						      skb_net_hdr_len);
1376 
1377 			/* use tso_type_first only once */
1378 			desc_tso_type = wil_tso_type_mid;
1379 
1380 			descs_used++;  /* desc used so far */
1381 			sg_desc_cnt++; /* desc used for this segment */
1382 			len -= lenmss;
1383 			rem_data -= lenmss;
1384 
1385 			wil_dbg_txrx(wil,
1386 				     "TSO: len %d, rem_data %d, descs_used %d, sg_desc_cnt %d,\n",
1387 				     len, rem_data, descs_used, sg_desc_cnt);
1388 
1389 			/* Close the segment if reached mss size or last frag*/
1390 			if (rem_data == 0 || (f == nr_frags - 1 && len == 0)) {
1391 				if (hdr_compensation_need) {
1392 					/* first segment include hdr desc for
1393 					 * release
1394 					 */
1395 					hdr_ctx->nr_frags = sg_desc_cnt;
1396 					wil_tx_desc_set_nr_frags(first_desc,
1397 								 sg_desc_cnt +
1398 								 1);
1399 					hdr_compensation_need = false;
1400 				} else {
1401 					wil_tx_desc_set_nr_frags(first_desc,
1402 								 sg_desc_cnt);
1403 				}
1404 				first_ctx->nr_frags = sg_desc_cnt - 1;
1405 
1406 				wil_tx_last_desc(d);
1407 
1408 				/* first descriptor may also be the last
1409 				 * for this mss - make sure not to copy
1410 				 * it twice
1411 				 */
1412 				if (first_desc != d)
1413 					*_first_desc = *first_desc;
1414 
1415 				/*last descriptor will be copied at the end
1416 				 * of this TS processing
1417 				 */
1418 				if (f < nr_frags - 1 || len > 0)
1419 					*_desc = *d;
1420 
1421 				rem_data = mss;
1422 				_first_desc = NULL;
1423 				sg_desc_cnt = 0;
1424 			} else if (first_desc != d) /* update mid descriptor */
1425 					*_desc = *d;
1426 		}
1427 	}
1428 
1429 	/* first descriptor may also be the last.
1430 	 * in this case d pointer is invalid
1431 	 */
1432 	if (_first_desc == _desc)
1433 		d = first_desc;
1434 
1435 	/* Last data descriptor */
1436 	wil_set_tx_desc_last_tso(d);
1437 	*_desc = *d;
1438 
1439 	/* Fill the total number of descriptors in first desc (hdr)*/
1440 	wil_tx_desc_set_nr_frags(hdr_desc, descs_used);
1441 	*_hdr_desc = *hdr_desc;
1442 
1443 	/* hold reference to skb
1444 	 * to prevent skb release before accounting
1445 	 * in case of immediate "tx done"
1446 	 */
1447 	vring->ctx[i].skb = skb_get(skb);
1448 
1449 	/* performance monitoring */
1450 	used = wil_vring_used_tx(vring);
1451 	if (wil_val_in_range(vring_idle_trsh,
1452 			     used, used + descs_used)) {
1453 		txdata->idle += get_cycles() - txdata->last_idle;
1454 		wil_dbg_txrx(wil,  "Ring[%2d] not idle %d -> %d\n",
1455 			     vring_index, used, used + descs_used);
1456 	}
1457 
1458 	/* advance swhead */
1459 	wil_dbg_txrx(wil, "TSO: Tx swhead %d -> %d\n", swhead, vring->swhead);
1460 	wil_vring_advance_head(vring, descs_used);
1461 
1462 	/* make sure all writes to descriptors (shared memory) are done before
1463 	 * committing them to HW
1464 	 */
1465 	wmb();
1466 
1467 	wil_w(wil, vring->hwtail, vring->swhead);
1468 	return 0;
1469 
1470 dma_error:
1471 	wil_err(wil, "TSO: DMA map page error\n");
1472 	while (descs_used > 0) {
1473 		struct wil_ctx *ctx;
1474 
1475 		i = (swhead + descs_used) % vring->size;
1476 		d = (struct vring_tx_desc *)&vring->va[i].tx;
1477 		_desc = &vring->va[i].tx;
1478 		*d = *_desc;
1479 		_desc->dma.status = TX_DMA_STATUS_DU;
1480 		ctx = &vring->ctx[i];
1481 		wil_txdesc_unmap(dev, d, ctx);
1482 		if (ctx->skb)
1483 			dev_kfree_skb_any(ctx->skb);
1484 		memset(ctx, 0, sizeof(*ctx));
1485 		descs_used--;
1486 	}
1487 
1488 err_exit:
1489 	return -EINVAL;
1490 }
1491 
1492 static int __wil_tx_vring(struct wil6210_priv *wil, struct vring *vring,
1493 			  struct sk_buff *skb)
1494 {
1495 	struct device *dev = wil_to_dev(wil);
1496 	struct vring_tx_desc dd, *d = &dd;
1497 	volatile struct vring_tx_desc *_d;
1498 	u32 swhead = vring->swhead;
1499 	int avail = wil_vring_avail_tx(vring);
1500 	int nr_frags = skb_shinfo(skb)->nr_frags;
1501 	uint f = 0;
1502 	int vring_index = vring - wil->vring_tx;
1503 	struct vring_tx_data *txdata = &wil->vring_tx_data[vring_index];
1504 	uint i = swhead;
1505 	dma_addr_t pa;
1506 	int used;
1507 	bool mcast = (vring_index == wil->bcast_vring);
1508 	uint len = skb_headlen(skb);
1509 
1510 	wil_dbg_txrx(wil, "%s() %d bytes to vring %d\n",
1511 		     __func__, skb->len, vring_index);
1512 
1513 	if (unlikely(!txdata->enabled))
1514 		return -EINVAL;
1515 
1516 	if (unlikely(avail < 1 + nr_frags)) {
1517 		wil_err_ratelimited(wil,
1518 				    "Tx ring[%2d] full. No space for %d fragments\n",
1519 				    vring_index, 1 + nr_frags);
1520 		return -ENOMEM;
1521 	}
1522 	_d = &vring->va[i].tx;
1523 
1524 	pa = dma_map_single(dev, skb->data, skb_headlen(skb), DMA_TO_DEVICE);
1525 
1526 	wil_dbg_txrx(wil, "Tx[%2d] skb %d bytes 0x%p -> %pad\n", vring_index,
1527 		     skb_headlen(skb), skb->data, &pa);
1528 	wil_hex_dump_txrx("Tx ", DUMP_PREFIX_OFFSET, 16, 1,
1529 			  skb->data, skb_headlen(skb), false);
1530 
1531 	if (unlikely(dma_mapping_error(dev, pa)))
1532 		return -EINVAL;
1533 	vring->ctx[i].mapped_as = wil_mapped_as_single;
1534 	/* 1-st segment */
1535 	wil_tx_desc_map(d, pa, len, vring_index);
1536 	if (unlikely(mcast)) {
1537 		d->mac.d[0] |= BIT(MAC_CFG_DESC_TX_0_MCS_EN_POS); /* MCS 0 */
1538 		if (unlikely(len > WIL_BCAST_MCS0_LIMIT)) /* set MCS 1 */
1539 			d->mac.d[0] |= (1 << MAC_CFG_DESC_TX_0_MCS_INDEX_POS);
1540 	}
1541 	/* Process TCP/UDP checksum offloading */
1542 	if (unlikely(wil_tx_desc_offload_setup(d, skb))) {
1543 		wil_err(wil, "Tx[%2d] Failed to set cksum, drop packet\n",
1544 			vring_index);
1545 		goto dma_error;
1546 	}
1547 
1548 	vring->ctx[i].nr_frags = nr_frags;
1549 	wil_tx_desc_set_nr_frags(d, nr_frags + 1);
1550 
1551 	/* middle segments */
1552 	for (; f < nr_frags; f++) {
1553 		const struct skb_frag_struct *frag =
1554 				&skb_shinfo(skb)->frags[f];
1555 		int len = skb_frag_size(frag);
1556 
1557 		*_d = *d;
1558 		wil_dbg_txrx(wil, "Tx[%2d] desc[%4d]\n", vring_index, i);
1559 		wil_hex_dump_txrx("TxD ", DUMP_PREFIX_NONE, 32, 4,
1560 				  (const void *)d, sizeof(*d), false);
1561 		i = (swhead + f + 1) % vring->size;
1562 		_d = &vring->va[i].tx;
1563 		pa = skb_frag_dma_map(dev, frag, 0, skb_frag_size(frag),
1564 				      DMA_TO_DEVICE);
1565 		if (unlikely(dma_mapping_error(dev, pa)))
1566 			goto dma_error;
1567 		vring->ctx[i].mapped_as = wil_mapped_as_page;
1568 		wil_tx_desc_map(d, pa, len, vring_index);
1569 		/* no need to check return code -
1570 		 * if it succeeded for 1-st descriptor,
1571 		 * it will succeed here too
1572 		 */
1573 		wil_tx_desc_offload_setup(d, skb);
1574 	}
1575 	/* for the last seg only */
1576 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_CMD_EOP_POS);
1577 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_CMD_MARK_WB_POS);
1578 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_CMD_DMA_IT_POS);
1579 	*_d = *d;
1580 	wil_dbg_txrx(wil, "Tx[%2d] desc[%4d]\n", vring_index, i);
1581 	wil_hex_dump_txrx("TxD ", DUMP_PREFIX_NONE, 32, 4,
1582 			  (const void *)d, sizeof(*d), false);
1583 
1584 	/* hold reference to skb
1585 	 * to prevent skb release before accounting
1586 	 * in case of immediate "tx done"
1587 	 */
1588 	vring->ctx[i].skb = skb_get(skb);
1589 
1590 	/* performance monitoring */
1591 	used = wil_vring_used_tx(vring);
1592 	if (wil_val_in_range(vring_idle_trsh,
1593 			     used, used + nr_frags + 1)) {
1594 		txdata->idle += get_cycles() - txdata->last_idle;
1595 		wil_dbg_txrx(wil,  "Ring[%2d] not idle %d -> %d\n",
1596 			     vring_index, used, used + nr_frags + 1);
1597 	}
1598 
1599 	/* advance swhead */
1600 	wil_vring_advance_head(vring, nr_frags + 1);
1601 	wil_dbg_txrx(wil, "Tx[%2d] swhead %d -> %d\n", vring_index, swhead,
1602 		     vring->swhead);
1603 	trace_wil6210_tx(vring_index, swhead, skb->len, nr_frags);
1604 
1605 	/* make sure all writes to descriptors (shared memory) are done before
1606 	 * committing them to HW
1607 	 */
1608 	wmb();
1609 
1610 	wil_w(wil, vring->hwtail, vring->swhead);
1611 
1612 	return 0;
1613  dma_error:
1614 	/* unmap what we have mapped */
1615 	nr_frags = f + 1; /* frags mapped + one for skb head */
1616 	for (f = 0; f < nr_frags; f++) {
1617 		struct wil_ctx *ctx;
1618 
1619 		i = (swhead + f) % vring->size;
1620 		ctx = &vring->ctx[i];
1621 		_d = &vring->va[i].tx;
1622 		*d = *_d;
1623 		_d->dma.status = TX_DMA_STATUS_DU;
1624 		wil_txdesc_unmap(dev, d, ctx);
1625 
1626 		if (ctx->skb)
1627 			dev_kfree_skb_any(ctx->skb);
1628 
1629 		memset(ctx, 0, sizeof(*ctx));
1630 	}
1631 
1632 	return -EINVAL;
1633 }
1634 
1635 static int wil_tx_vring(struct wil6210_priv *wil, struct vring *vring,
1636 			struct sk_buff *skb)
1637 {
1638 	int vring_index = vring - wil->vring_tx;
1639 	struct vring_tx_data *txdata = &wil->vring_tx_data[vring_index];
1640 	int rc;
1641 
1642 	spin_lock(&txdata->lock);
1643 
1644 	rc = (skb_is_gso(skb) ? __wil_tx_vring_tso : __wil_tx_vring)
1645 	     (wil, vring, skb);
1646 
1647 	spin_unlock(&txdata->lock);
1648 
1649 	return rc;
1650 }
1651 
1652 netdev_tx_t wil_start_xmit(struct sk_buff *skb, struct net_device *ndev)
1653 {
1654 	struct wil6210_priv *wil = ndev_to_wil(ndev);
1655 	struct ethhdr *eth = (void *)skb->data;
1656 	bool bcast = is_multicast_ether_addr(eth->h_dest);
1657 	struct vring *vring;
1658 	static bool pr_once_fw;
1659 	int rc;
1660 
1661 	wil_dbg_txrx(wil, "%s()\n", __func__);
1662 	if (unlikely(!test_bit(wil_status_fwready, wil->status))) {
1663 		if (!pr_once_fw) {
1664 			wil_err(wil, "FW not ready\n");
1665 			pr_once_fw = true;
1666 		}
1667 		goto drop;
1668 	}
1669 	if (unlikely(!test_bit(wil_status_fwconnected, wil->status))) {
1670 		wil_err(wil, "FW not connected\n");
1671 		goto drop;
1672 	}
1673 	if (unlikely(wil->wdev->iftype == NL80211_IFTYPE_MONITOR)) {
1674 		wil_err(wil, "Xmit in monitor mode not supported\n");
1675 		goto drop;
1676 	}
1677 	pr_once_fw = false;
1678 
1679 	/* find vring */
1680 	if (wil->wdev->iftype == NL80211_IFTYPE_STATION) {
1681 		/* in STA mode (ESS), all to same VRING */
1682 		vring = wil_find_tx_vring_sta(wil, skb);
1683 	} else { /* direct communication, find matching VRING */
1684 		vring = bcast ? wil_find_tx_bcast(wil, skb) :
1685 				wil_find_tx_ucast(wil, skb);
1686 	}
1687 	if (unlikely(!vring)) {
1688 		wil_dbg_txrx(wil, "No Tx VRING found for %pM\n", eth->h_dest);
1689 		goto drop;
1690 	}
1691 	/* set up vring entry */
1692 	rc = wil_tx_vring(wil, vring, skb);
1693 
1694 	/* do we still have enough room in the vring? */
1695 	if (unlikely(wil_vring_avail_tx(vring) < wil_vring_wmark_low(vring))) {
1696 		netif_tx_stop_all_queues(wil_to_ndev(wil));
1697 		wil_dbg_txrx(wil, "netif_tx_stop : ring full\n");
1698 	}
1699 
1700 	switch (rc) {
1701 	case 0:
1702 		/* statistics will be updated on the tx_complete */
1703 		dev_kfree_skb_any(skb);
1704 		return NETDEV_TX_OK;
1705 	case -ENOMEM:
1706 		return NETDEV_TX_BUSY;
1707 	default:
1708 		break; /* goto drop; */
1709 	}
1710  drop:
1711 	ndev->stats.tx_dropped++;
1712 	dev_kfree_skb_any(skb);
1713 
1714 	return NET_XMIT_DROP;
1715 }
1716 
1717 static inline bool wil_need_txstat(struct sk_buff *skb)
1718 {
1719 	struct ethhdr *eth = (void *)skb->data;
1720 
1721 	return is_unicast_ether_addr(eth->h_dest) && skb->sk &&
1722 	       (skb_shinfo(skb)->tx_flags & SKBTX_WIFI_STATUS);
1723 }
1724 
1725 static inline void wil_consume_skb(struct sk_buff *skb, bool acked)
1726 {
1727 	if (unlikely(wil_need_txstat(skb)))
1728 		skb_complete_wifi_ack(skb, acked);
1729 	else
1730 		acked ? dev_consume_skb_any(skb) : dev_kfree_skb_any(skb);
1731 }
1732 
1733 /**
1734  * Clean up transmitted skb's from the Tx VRING
1735  *
1736  * Return number of descriptors cleared
1737  *
1738  * Safe to call from IRQ
1739  */
1740 int wil_tx_complete(struct wil6210_priv *wil, int ringid)
1741 {
1742 	struct net_device *ndev = wil_to_ndev(wil);
1743 	struct device *dev = wil_to_dev(wil);
1744 	struct vring *vring = &wil->vring_tx[ringid];
1745 	struct vring_tx_data *txdata = &wil->vring_tx_data[ringid];
1746 	int done = 0;
1747 	int cid = wil->vring2cid_tid[ringid][0];
1748 	struct wil_net_stats *stats = NULL;
1749 	volatile struct vring_tx_desc *_d;
1750 	int used_before_complete;
1751 	int used_new;
1752 
1753 	if (unlikely(!vring->va)) {
1754 		wil_err(wil, "Tx irq[%d]: vring not initialized\n", ringid);
1755 		return 0;
1756 	}
1757 
1758 	if (unlikely(!txdata->enabled)) {
1759 		wil_info(wil, "Tx irq[%d]: vring disabled\n", ringid);
1760 		return 0;
1761 	}
1762 
1763 	wil_dbg_txrx(wil, "%s(%d)\n", __func__, ringid);
1764 
1765 	used_before_complete = wil_vring_used_tx(vring);
1766 
1767 	if (cid < WIL6210_MAX_CID)
1768 		stats = &wil->sta[cid].stats;
1769 
1770 	while (!wil_vring_is_empty(vring)) {
1771 		int new_swtail;
1772 		struct wil_ctx *ctx = &vring->ctx[vring->swtail];
1773 		/**
1774 		 * For the fragmented skb, HW will set DU bit only for the
1775 		 * last fragment. look for it.
1776 		 * In TSO the first DU will include hdr desc
1777 		 */
1778 		int lf = (vring->swtail + ctx->nr_frags) % vring->size;
1779 		/* TODO: check we are not past head */
1780 
1781 		_d = &vring->va[lf].tx;
1782 		if (unlikely(!(_d->dma.status & TX_DMA_STATUS_DU)))
1783 			break;
1784 
1785 		new_swtail = (lf + 1) % vring->size;
1786 		while (vring->swtail != new_swtail) {
1787 			struct vring_tx_desc dd, *d = &dd;
1788 			u16 dmalen;
1789 			struct sk_buff *skb;
1790 
1791 			ctx = &vring->ctx[vring->swtail];
1792 			skb = ctx->skb;
1793 			_d = &vring->va[vring->swtail].tx;
1794 
1795 			*d = *_d;
1796 
1797 			dmalen = le16_to_cpu(d->dma.length);
1798 			trace_wil6210_tx_done(ringid, vring->swtail, dmalen,
1799 					      d->dma.error);
1800 			wil_dbg_txrx(wil,
1801 				     "TxC[%2d][%3d] : %d bytes, status 0x%02x err 0x%02x\n",
1802 				     ringid, vring->swtail, dmalen,
1803 				     d->dma.status, d->dma.error);
1804 			wil_hex_dump_txrx("TxCD ", DUMP_PREFIX_NONE, 32, 4,
1805 					  (const void *)d, sizeof(*d), false);
1806 
1807 			wil_txdesc_unmap(dev, d, ctx);
1808 
1809 			if (skb) {
1810 				if (likely(d->dma.error == 0)) {
1811 					ndev->stats.tx_packets++;
1812 					ndev->stats.tx_bytes += skb->len;
1813 					if (stats) {
1814 						stats->tx_packets++;
1815 						stats->tx_bytes += skb->len;
1816 					}
1817 				} else {
1818 					ndev->stats.tx_errors++;
1819 					if (stats)
1820 						stats->tx_errors++;
1821 				}
1822 				wil_consume_skb(skb, d->dma.error == 0);
1823 			}
1824 			memset(ctx, 0, sizeof(*ctx));
1825 			/* There is no need to touch HW descriptor:
1826 			 * - ststus bit TX_DMA_STATUS_DU is set by design,
1827 			 *   so hardware will not try to process this desc.,
1828 			 * - rest of descriptor will be initialized on Tx.
1829 			 */
1830 			vring->swtail = wil_vring_next_tail(vring);
1831 			done++;
1832 		}
1833 	}
1834 
1835 	/* performance monitoring */
1836 	used_new = wil_vring_used_tx(vring);
1837 	if (wil_val_in_range(vring_idle_trsh,
1838 			     used_new, used_before_complete)) {
1839 		wil_dbg_txrx(wil, "Ring[%2d] idle %d -> %d\n",
1840 			     ringid, used_before_complete, used_new);
1841 		txdata->last_idle = get_cycles();
1842 	}
1843 
1844 	if (wil_vring_avail_tx(vring) > wil_vring_wmark_high(vring)) {
1845 		wil_dbg_txrx(wil, "netif_tx_wake : ring not full\n");
1846 		netif_tx_wake_all_queues(wil_to_ndev(wil));
1847 	}
1848 
1849 	return done;
1850 }
1851