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 	lockdep_assert_held(&wil->mutex);
164 	if (tx) {
165 		int vring_index = vring - wil->vring_tx;
166 
167 		wil_dbg_misc(wil, "free Tx vring %d [%d] 0x%p:%pad 0x%p\n",
168 			     vring_index, vring->size, vring->va,
169 			     &vring->pa, vring->ctx);
170 	} else {
171 		wil_dbg_misc(wil, "free Rx vring [%d] 0x%p:%pad 0x%p\n",
172 			     vring->size, vring->va,
173 			     &vring->pa, vring->ctx);
174 	}
175 
176 	while (!wil_vring_is_empty(vring)) {
177 		dma_addr_t pa;
178 		u16 dmalen;
179 		struct wil_ctx *ctx;
180 
181 		if (tx) {
182 			struct vring_tx_desc dd, *d = &dd;
183 			volatile struct vring_tx_desc *_d =
184 					&vring->va[vring->swtail].tx;
185 
186 			ctx = &vring->ctx[vring->swtail];
187 			*d = *_d;
188 			wil_txdesc_unmap(dev, d, ctx);
189 			if (ctx->skb)
190 				dev_kfree_skb_any(ctx->skb);
191 			vring->swtail = wil_vring_next_tail(vring);
192 		} else { /* rx */
193 			struct vring_rx_desc dd, *d = &dd;
194 			volatile struct vring_rx_desc *_d =
195 					&vring->va[vring->swhead].rx;
196 
197 			ctx = &vring->ctx[vring->swhead];
198 			*d = *_d;
199 			pa = wil_desc_addr(&d->dma.addr);
200 			dmalen = le16_to_cpu(d->dma.length);
201 			dma_unmap_single(dev, pa, dmalen, DMA_FROM_DEVICE);
202 			kfree_skb(ctx->skb);
203 			wil_vring_advance_head(vring, 1);
204 		}
205 	}
206 	dma_free_coherent(dev, sz, (void *)vring->va, vring->pa);
207 	kfree(vring->ctx);
208 	vring->pa = 0;
209 	vring->va = NULL;
210 	vring->ctx = NULL;
211 }
212 
213 /**
214  * Allocate one skb for Rx VRING
215  *
216  * Safe to call from IRQ
217  */
218 static int wil_vring_alloc_skb(struct wil6210_priv *wil, struct vring *vring,
219 			       u32 i, int headroom)
220 {
221 	struct device *dev = wil_to_dev(wil);
222 	unsigned int sz = mtu_max + ETH_HLEN + wil_rx_snaplen();
223 	struct vring_rx_desc dd, *d = &dd;
224 	volatile struct vring_rx_desc *_d = &vring->va[i].rx;
225 	dma_addr_t pa;
226 	struct sk_buff *skb = dev_alloc_skb(sz + headroom);
227 
228 	if (unlikely(!skb))
229 		return -ENOMEM;
230 
231 	skb_reserve(skb, headroom);
232 	skb_put(skb, sz);
233 
234 	pa = dma_map_single(dev, skb->data, skb->len, DMA_FROM_DEVICE);
235 	if (unlikely(dma_mapping_error(dev, pa))) {
236 		kfree_skb(skb);
237 		return -ENOMEM;
238 	}
239 
240 	d->dma.d0 = RX_DMA_D0_CMD_DMA_RT | RX_DMA_D0_CMD_DMA_IT;
241 	wil_desc_addr_set(&d->dma.addr, pa);
242 	/* ip_length don't care */
243 	/* b11 don't care */
244 	/* error don't care */
245 	d->dma.status = 0; /* BIT(0) should be 0 for HW_OWNED */
246 	d->dma.length = cpu_to_le16(sz);
247 	*_d = *d;
248 	vring->ctx[i].skb = skb;
249 
250 	return 0;
251 }
252 
253 /**
254  * Adds radiotap header
255  *
256  * Any error indicated as "Bad FCS"
257  *
258  * Vendor data for 04:ce:14-1 (Wilocity-1) consists of:
259  *  - Rx descriptor: 32 bytes
260  *  - Phy info
261  */
262 static void wil_rx_add_radiotap_header(struct wil6210_priv *wil,
263 				       struct sk_buff *skb)
264 {
265 	struct wireless_dev *wdev = wil->wdev;
266 	struct wil6210_rtap {
267 		struct ieee80211_radiotap_header rthdr;
268 		/* fields should be in the order of bits in rthdr.it_present */
269 		/* flags */
270 		u8 flags;
271 		/* channel */
272 		__le16 chnl_freq __aligned(2);
273 		__le16 chnl_flags;
274 		/* MCS */
275 		u8 mcs_present;
276 		u8 mcs_flags;
277 		u8 mcs_index;
278 	} __packed;
279 	struct wil6210_rtap_vendor {
280 		struct wil6210_rtap rtap;
281 		/* vendor */
282 		u8 vendor_oui[3] __aligned(2);
283 		u8 vendor_ns;
284 		__le16 vendor_skip;
285 		u8 vendor_data[0];
286 	} __packed;
287 	struct vring_rx_desc *d = wil_skb_rxdesc(skb);
288 	struct wil6210_rtap_vendor *rtap_vendor;
289 	int rtap_len = sizeof(struct wil6210_rtap);
290 	int phy_length = 0; /* phy info header size, bytes */
291 	static char phy_data[128];
292 	struct ieee80211_channel *ch = wdev->preset_chandef.chan;
293 
294 	if (rtap_include_phy_info) {
295 		rtap_len = sizeof(*rtap_vendor) + sizeof(*d);
296 		/* calculate additional length */
297 		if (d->dma.status & RX_DMA_STATUS_PHY_INFO) {
298 			/**
299 			 * PHY info starts from 8-byte boundary
300 			 * there are 8-byte lines, last line may be partially
301 			 * written (HW bug), thus FW configures for last line
302 			 * to be excessive. Driver skips this last line.
303 			 */
304 			int len = min_t(int, 8 + sizeof(phy_data),
305 					wil_rxdesc_phy_length(d));
306 
307 			if (len > 8) {
308 				void *p = skb_tail_pointer(skb);
309 				void *pa = PTR_ALIGN(p, 8);
310 
311 				if (skb_tailroom(skb) >= len + (pa - p)) {
312 					phy_length = len - 8;
313 					memcpy(phy_data, pa, phy_length);
314 				}
315 			}
316 		}
317 		rtap_len += phy_length;
318 	}
319 
320 	if (skb_headroom(skb) < rtap_len &&
321 	    pskb_expand_head(skb, rtap_len, 0, GFP_ATOMIC)) {
322 		wil_err(wil, "Unable to expand headrom to %d\n", rtap_len);
323 		return;
324 	}
325 
326 	rtap_vendor = (void *)skb_push(skb, rtap_len);
327 	memset(rtap_vendor, 0, rtap_len);
328 
329 	rtap_vendor->rtap.rthdr.it_version = PKTHDR_RADIOTAP_VERSION;
330 	rtap_vendor->rtap.rthdr.it_len = cpu_to_le16(rtap_len);
331 	rtap_vendor->rtap.rthdr.it_present = cpu_to_le32(
332 			(1 << IEEE80211_RADIOTAP_FLAGS) |
333 			(1 << IEEE80211_RADIOTAP_CHANNEL) |
334 			(1 << IEEE80211_RADIOTAP_MCS));
335 	if (d->dma.status & RX_DMA_STATUS_ERROR)
336 		rtap_vendor->rtap.flags |= IEEE80211_RADIOTAP_F_BADFCS;
337 
338 	rtap_vendor->rtap.chnl_freq = cpu_to_le16(ch ? ch->center_freq : 58320);
339 	rtap_vendor->rtap.chnl_flags = cpu_to_le16(0);
340 
341 	rtap_vendor->rtap.mcs_present = IEEE80211_RADIOTAP_MCS_HAVE_MCS;
342 	rtap_vendor->rtap.mcs_flags = 0;
343 	rtap_vendor->rtap.mcs_index = wil_rxdesc_mcs(d);
344 
345 	if (rtap_include_phy_info) {
346 		rtap_vendor->rtap.rthdr.it_present |= cpu_to_le32(1 <<
347 				IEEE80211_RADIOTAP_VENDOR_NAMESPACE);
348 		/* OUI for Wilocity 04:ce:14 */
349 		rtap_vendor->vendor_oui[0] = 0x04;
350 		rtap_vendor->vendor_oui[1] = 0xce;
351 		rtap_vendor->vendor_oui[2] = 0x14;
352 		rtap_vendor->vendor_ns = 1;
353 		/* Rx descriptor + PHY data  */
354 		rtap_vendor->vendor_skip = cpu_to_le16(sizeof(*d) +
355 						       phy_length);
356 		memcpy(rtap_vendor->vendor_data, (void *)d, sizeof(*d));
357 		memcpy(rtap_vendor->vendor_data + sizeof(*d), phy_data,
358 		       phy_length);
359 	}
360 }
361 
362 /* similar to ieee80211_ version, but FC contain only 1-st byte */
363 static inline int wil_is_back_req(u8 fc)
364 {
365 	return (fc & (IEEE80211_FCTL_FTYPE | IEEE80211_FCTL_STYPE)) ==
366 	       (IEEE80211_FTYPE_CTL | IEEE80211_STYPE_BACK_REQ);
367 }
368 
369 /**
370  * reap 1 frame from @swhead
371  *
372  * Rx descriptor copied to skb->cb
373  *
374  * Safe to call from IRQ
375  */
376 static struct sk_buff *wil_vring_reap_rx(struct wil6210_priv *wil,
377 					 struct vring *vring)
378 {
379 	struct device *dev = wil_to_dev(wil);
380 	struct net_device *ndev = wil_to_ndev(wil);
381 	volatile struct vring_rx_desc *_d;
382 	struct vring_rx_desc *d;
383 	struct sk_buff *skb;
384 	dma_addr_t pa;
385 	unsigned int snaplen = wil_rx_snaplen();
386 	unsigned int sz = mtu_max + ETH_HLEN + snaplen;
387 	u16 dmalen;
388 	u8 ftype;
389 	int cid;
390 	int i;
391 	struct wil_net_stats *stats;
392 
393 	BUILD_BUG_ON(sizeof(struct vring_rx_desc) > sizeof(skb->cb));
394 
395 again:
396 	if (unlikely(wil_vring_is_empty(vring)))
397 		return NULL;
398 
399 	i = (int)vring->swhead;
400 	_d = &vring->va[i].rx;
401 	if (unlikely(!(_d->dma.status & RX_DMA_STATUS_DU))) {
402 		/* it is not error, we just reached end of Rx done area */
403 		return NULL;
404 	}
405 
406 	skb = vring->ctx[i].skb;
407 	vring->ctx[i].skb = NULL;
408 	wil_vring_advance_head(vring, 1);
409 	if (!skb) {
410 		wil_err(wil, "No Rx skb at [%d]\n", i);
411 		goto again;
412 	}
413 	d = wil_skb_rxdesc(skb);
414 	*d = *_d;
415 	pa = wil_desc_addr(&d->dma.addr);
416 
417 	dma_unmap_single(dev, pa, sz, DMA_FROM_DEVICE);
418 	dmalen = le16_to_cpu(d->dma.length);
419 
420 	trace_wil6210_rx(i, d);
421 	wil_dbg_txrx(wil, "Rx[%3d] : %d bytes\n", i, dmalen);
422 	wil_hex_dump_txrx("RxD ", DUMP_PREFIX_NONE, 32, 4,
423 			  (const void *)d, sizeof(*d), false);
424 
425 	cid = wil_rxdesc_cid(d);
426 	stats = &wil->sta[cid].stats;
427 
428 	if (unlikely(dmalen > sz)) {
429 		wil_err(wil, "Rx size too large: %d bytes!\n", dmalen);
430 		stats->rx_large_frame++;
431 		kfree_skb(skb);
432 		goto again;
433 	}
434 	skb_trim(skb, dmalen);
435 
436 	prefetch(skb->data);
437 
438 	wil_hex_dump_txrx("Rx ", DUMP_PREFIX_OFFSET, 16, 1,
439 			  skb->data, skb_headlen(skb), false);
440 
441 	stats->last_mcs_rx = wil_rxdesc_mcs(d);
442 	if (stats->last_mcs_rx < ARRAY_SIZE(stats->rx_per_mcs))
443 		stats->rx_per_mcs[stats->last_mcs_rx]++;
444 
445 	/* use radiotap header only if required */
446 	if (ndev->type == ARPHRD_IEEE80211_RADIOTAP)
447 		wil_rx_add_radiotap_header(wil, skb);
448 
449 	/* no extra checks if in sniffer mode */
450 	if (ndev->type != ARPHRD_ETHER)
451 		return skb;
452 	/* Non-data frames may be delivered through Rx DMA channel (ex: BAR)
453 	 * Driver should recognize it by frame type, that is found
454 	 * in Rx descriptor. If type is not data, it is 802.11 frame as is
455 	 */
456 	ftype = wil_rxdesc_ftype(d) << 2;
457 	if (unlikely(ftype != IEEE80211_FTYPE_DATA)) {
458 		u8 fc1 = wil_rxdesc_fc1(d);
459 		int mid = wil_rxdesc_mid(d);
460 		int tid = wil_rxdesc_tid(d);
461 		u16 seq = wil_rxdesc_seq(d);
462 
463 		wil_dbg_txrx(wil,
464 			     "Non-data frame FC[7:0] 0x%02x MID %d CID %d TID %d Seq 0x%03x\n",
465 			     fc1, mid, cid, tid, seq);
466 		stats->rx_non_data_frame++;
467 		if (wil_is_back_req(fc1)) {
468 			wil_dbg_txrx(wil,
469 				     "BAR: MID %d CID %d TID %d Seq 0x%03x\n",
470 				     mid, cid, tid, seq);
471 			wil_rx_bar(wil, cid, tid, seq);
472 		} else {
473 			/* print again all info. One can enable only this
474 			 * without overhead for printing every Rx frame
475 			 */
476 			wil_dbg_txrx(wil,
477 				     "Unhandled non-data frame FC[7:0] 0x%02x MID %d CID %d TID %d Seq 0x%03x\n",
478 				     fc1, mid, cid, tid, seq);
479 			wil_hex_dump_txrx("RxD ", DUMP_PREFIX_NONE, 32, 4,
480 					  (const void *)d, sizeof(*d), false);
481 			wil_hex_dump_txrx("Rx ", DUMP_PREFIX_OFFSET, 16, 1,
482 					  skb->data, skb_headlen(skb), false);
483 		}
484 		kfree_skb(skb);
485 		goto again;
486 	}
487 
488 	if (unlikely(skb->len < ETH_HLEN + snaplen)) {
489 		wil_err(wil, "Short frame, len = %d\n", skb->len);
490 		stats->rx_short_frame++;
491 		kfree_skb(skb);
492 		goto again;
493 	}
494 
495 	/* L4 IDENT is on when HW calculated checksum, check status
496 	 * and in case of error drop the packet
497 	 * higher stack layers will handle retransmission (if required)
498 	 */
499 	if (likely(d->dma.status & RX_DMA_STATUS_L4I)) {
500 		/* L4 protocol identified, csum calculated */
501 		if (likely((d->dma.error & RX_DMA_ERROR_L4_ERR) == 0))
502 			skb->ip_summed = CHECKSUM_UNNECESSARY;
503 		/* If HW reports bad checksum, let IP stack re-check it
504 		 * For example, HW don't understand Microsoft IP stack that
505 		 * mis-calculates TCP checksum - if it should be 0x0,
506 		 * it writes 0xffff in violation of RFC 1624
507 		 */
508 	}
509 
510 	if (snaplen) {
511 		/* Packet layout
512 		 * +-------+-------+---------+------------+------+
513 		 * | SA(6) | DA(6) | SNAP(6) | ETHTYPE(2) | DATA |
514 		 * +-------+-------+---------+------------+------+
515 		 * Need to remove SNAP, shifting SA and DA forward
516 		 */
517 		memmove(skb->data + snaplen, skb->data, 2 * ETH_ALEN);
518 		skb_pull(skb, snaplen);
519 	}
520 
521 	return skb;
522 }
523 
524 /**
525  * allocate and fill up to @count buffers in rx ring
526  * buffers posted at @swtail
527  */
528 static int wil_rx_refill(struct wil6210_priv *wil, int count)
529 {
530 	struct net_device *ndev = wil_to_ndev(wil);
531 	struct vring *v = &wil->vring_rx;
532 	u32 next_tail;
533 	int rc = 0;
534 	int headroom = ndev->type == ARPHRD_IEEE80211_RADIOTAP ?
535 			WIL6210_RTAP_SIZE : 0;
536 
537 	for (; next_tail = wil_vring_next_tail(v),
538 			(next_tail != v->swhead) && (count-- > 0);
539 			v->swtail = next_tail) {
540 		rc = wil_vring_alloc_skb(wil, v, v->swtail, headroom);
541 		if (unlikely(rc)) {
542 			wil_err(wil, "Error %d in wil_rx_refill[%d]\n",
543 				rc, v->swtail);
544 			break;
545 		}
546 	}
547 	wil_w(wil, v->hwtail, v->swtail);
548 
549 	return rc;
550 }
551 
552 /*
553  * Pass Rx packet to the netif. Update statistics.
554  * Called in softirq context (NAPI poll).
555  */
556 void wil_netif_rx_any(struct sk_buff *skb, struct net_device *ndev)
557 {
558 	gro_result_t rc = GRO_NORMAL;
559 	struct wil6210_priv *wil = ndev_to_wil(ndev);
560 	struct wireless_dev *wdev = wil_to_wdev(wil);
561 	unsigned int len = skb->len;
562 	struct vring_rx_desc *d = wil_skb_rxdesc(skb);
563 	int cid = wil_rxdesc_cid(d); /* always 0..7, no need to check */
564 	struct ethhdr *eth = (void *)skb->data;
565 	/* here looking for DA, not A1, thus Rxdesc's 'mcast' indication
566 	 * is not suitable, need to look at data
567 	 */
568 	int mcast = is_multicast_ether_addr(eth->h_dest);
569 	struct wil_net_stats *stats = &wil->sta[cid].stats;
570 	struct sk_buff *xmit_skb = NULL;
571 	static const char * const gro_res_str[] = {
572 		[GRO_MERGED]		= "GRO_MERGED",
573 		[GRO_MERGED_FREE]	= "GRO_MERGED_FREE",
574 		[GRO_HELD]		= "GRO_HELD",
575 		[GRO_NORMAL]		= "GRO_NORMAL",
576 		[GRO_DROP]		= "GRO_DROP",
577 	};
578 
579 	if (ndev->features & NETIF_F_RXHASH)
580 		/* fake L4 to ensure it won't be re-calculated later
581 		 * set hash to any non-zero value to activate rps
582 		 * mechanism, core will be chosen according
583 		 * to user-level rps configuration.
584 		 */
585 		skb_set_hash(skb, 1, PKT_HASH_TYPE_L4);
586 
587 	skb_orphan(skb);
588 
589 	if (wdev->iftype == NL80211_IFTYPE_AP && !wil->ap_isolate) {
590 		if (mcast) {
591 			/* send multicast frames both to higher layers in
592 			 * local net stack and back to the wireless medium
593 			 */
594 			xmit_skb = skb_copy(skb, GFP_ATOMIC);
595 		} else {
596 			int xmit_cid = wil_find_cid(wil, eth->h_dest);
597 
598 			if (xmit_cid >= 0) {
599 				/* The destination station is associated to
600 				 * this AP (in this VLAN), so send the frame
601 				 * directly to it and do not pass it to local
602 				 * net stack.
603 				 */
604 				xmit_skb = skb;
605 				skb = NULL;
606 			}
607 		}
608 	}
609 	if (xmit_skb) {
610 		/* Send to wireless media and increase priority by 256 to
611 		 * keep the received priority instead of reclassifying
612 		 * the frame (see cfg80211_classify8021d).
613 		 */
614 		xmit_skb->dev = ndev;
615 		xmit_skb->priority += 256;
616 		xmit_skb->protocol = htons(ETH_P_802_3);
617 		skb_reset_network_header(xmit_skb);
618 		skb_reset_mac_header(xmit_skb);
619 		wil_dbg_txrx(wil, "Rx -> Tx %d bytes\n", len);
620 		dev_queue_xmit(xmit_skb);
621 	}
622 
623 	if (skb) { /* deliver to local stack */
624 
625 		skb->protocol = eth_type_trans(skb, ndev);
626 		rc = napi_gro_receive(&wil->napi_rx, skb);
627 		wil_dbg_txrx(wil, "Rx complete %d bytes => %s\n",
628 			     len, gro_res_str[rc]);
629 	}
630 	/* statistics. rc set to GRO_NORMAL for AP bridging */
631 	if (unlikely(rc == GRO_DROP)) {
632 		ndev->stats.rx_dropped++;
633 		stats->rx_dropped++;
634 		wil_dbg_txrx(wil, "Rx drop %d bytes\n", len);
635 	} else {
636 		ndev->stats.rx_packets++;
637 		stats->rx_packets++;
638 		ndev->stats.rx_bytes += len;
639 		stats->rx_bytes += len;
640 		if (mcast)
641 			ndev->stats.multicast++;
642 	}
643 }
644 
645 /**
646  * Proceed all completed skb's from Rx VRING
647  *
648  * Safe to call from NAPI poll, i.e. softirq with interrupts enabled
649  */
650 void wil_rx_handle(struct wil6210_priv *wil, int *quota)
651 {
652 	struct net_device *ndev = wil_to_ndev(wil);
653 	struct vring *v = &wil->vring_rx;
654 	struct sk_buff *skb;
655 
656 	if (unlikely(!v->va)) {
657 		wil_err(wil, "Rx IRQ while Rx not yet initialized\n");
658 		return;
659 	}
660 	wil_dbg_txrx(wil, "%s()\n", __func__);
661 	while ((*quota > 0) && (NULL != (skb = wil_vring_reap_rx(wil, v)))) {
662 		(*quota)--;
663 
664 		if (wil->wdev->iftype == NL80211_IFTYPE_MONITOR) {
665 			skb->dev = ndev;
666 			skb_reset_mac_header(skb);
667 			skb->ip_summed = CHECKSUM_UNNECESSARY;
668 			skb->pkt_type = PACKET_OTHERHOST;
669 			skb->protocol = htons(ETH_P_802_2);
670 			wil_netif_rx_any(skb, ndev);
671 		} else {
672 			wil_rx_reorder(wil, skb);
673 		}
674 	}
675 	wil_rx_refill(wil, v->size);
676 }
677 
678 int wil_rx_init(struct wil6210_priv *wil, u16 size)
679 {
680 	struct vring *vring = &wil->vring_rx;
681 	int rc;
682 
683 	wil_dbg_misc(wil, "%s()\n", __func__);
684 
685 	if (vring->va) {
686 		wil_err(wil, "Rx ring already allocated\n");
687 		return -EINVAL;
688 	}
689 
690 	vring->size = size;
691 	rc = wil_vring_alloc(wil, vring);
692 	if (rc)
693 		return rc;
694 
695 	rc = wmi_rx_chain_add(wil, vring);
696 	if (rc)
697 		goto err_free;
698 
699 	rc = wil_rx_refill(wil, vring->size);
700 	if (rc)
701 		goto err_free;
702 
703 	return 0;
704  err_free:
705 	wil_vring_free(wil, vring, 0);
706 
707 	return rc;
708 }
709 
710 void wil_rx_fini(struct wil6210_priv *wil)
711 {
712 	struct vring *vring = &wil->vring_rx;
713 
714 	wil_dbg_misc(wil, "%s()\n", __func__);
715 
716 	if (vring->va)
717 		wil_vring_free(wil, vring, 0);
718 }
719 
720 int wil_vring_init_tx(struct wil6210_priv *wil, int id, int size,
721 		      int cid, int tid)
722 {
723 	int rc;
724 	struct wmi_vring_cfg_cmd cmd = {
725 		.action = cpu_to_le32(WMI_VRING_CMD_ADD),
726 		.vring_cfg = {
727 			.tx_sw_ring = {
728 				.max_mpdu_size =
729 					cpu_to_le16(wil_mtu2macbuf(mtu_max)),
730 				.ring_size = cpu_to_le16(size),
731 			},
732 			.ringid = id,
733 			.cidxtid = mk_cidxtid(cid, tid),
734 			.encap_trans_type = WMI_VRING_ENC_TYPE_802_3,
735 			.mac_ctrl = 0,
736 			.to_resolution = 0,
737 			.agg_max_wsize = 0,
738 			.schd_params = {
739 				.priority = cpu_to_le16(0),
740 				.timeslot_us = cpu_to_le16(0xfff),
741 			},
742 		},
743 	};
744 	struct {
745 		struct wil6210_mbox_hdr_wmi wmi;
746 		struct wmi_vring_cfg_done_event cmd;
747 	} __packed reply;
748 	struct vring *vring = &wil->vring_tx[id];
749 	struct vring_tx_data *txdata = &wil->vring_tx_data[id];
750 
751 	wil_dbg_misc(wil, "%s() max_mpdu_size %d\n", __func__,
752 		     cmd.vring_cfg.tx_sw_ring.max_mpdu_size);
753 	lockdep_assert_held(&wil->mutex);
754 
755 	if (vring->va) {
756 		wil_err(wil, "Tx ring [%d] already allocated\n", id);
757 		rc = -EINVAL;
758 		goto out;
759 	}
760 
761 	memset(txdata, 0, sizeof(*txdata));
762 	spin_lock_init(&txdata->lock);
763 	vring->size = size;
764 	rc = wil_vring_alloc(wil, vring);
765 	if (rc)
766 		goto out;
767 
768 	wil->vring2cid_tid[id][0] = cid;
769 	wil->vring2cid_tid[id][1] = tid;
770 
771 	cmd.vring_cfg.tx_sw_ring.ring_mem_base = cpu_to_le64(vring->pa);
772 
773 	if (!wil->privacy)
774 		txdata->dot1x_open = true;
775 	rc = wmi_call(wil, WMI_VRING_CFG_CMDID, &cmd, sizeof(cmd),
776 		      WMI_VRING_CFG_DONE_EVENTID, &reply, sizeof(reply), 100);
777 	if (rc)
778 		goto out_free;
779 
780 	if (reply.cmd.status != WMI_FW_STATUS_SUCCESS) {
781 		wil_err(wil, "Tx config failed, status 0x%02x\n",
782 			reply.cmd.status);
783 		rc = -EINVAL;
784 		goto out_free;
785 	}
786 	vring->hwtail = le32_to_cpu(reply.cmd.tx_vring_tail_ptr);
787 
788 	txdata->enabled = 1;
789 	if (txdata->dot1x_open && (agg_wsize >= 0))
790 		wil_addba_tx_request(wil, id, agg_wsize);
791 
792 	return 0;
793  out_free:
794 	txdata->dot1x_open = false;
795 	txdata->enabled = 0;
796 	wil_vring_free(wil, vring, 1);
797  out:
798 
799 	return rc;
800 }
801 
802 int wil_vring_init_bcast(struct wil6210_priv *wil, int id, int size)
803 {
804 	int rc;
805 	struct wmi_bcast_vring_cfg_cmd cmd = {
806 		.action = cpu_to_le32(WMI_VRING_CMD_ADD),
807 		.vring_cfg = {
808 			.tx_sw_ring = {
809 				.max_mpdu_size =
810 					cpu_to_le16(wil_mtu2macbuf(mtu_max)),
811 				.ring_size = cpu_to_le16(size),
812 			},
813 			.ringid = id,
814 			.encap_trans_type = WMI_VRING_ENC_TYPE_802_3,
815 		},
816 	};
817 	struct {
818 		struct wil6210_mbox_hdr_wmi wmi;
819 		struct wmi_vring_cfg_done_event cmd;
820 	} __packed reply;
821 	struct vring *vring = &wil->vring_tx[id];
822 	struct vring_tx_data *txdata = &wil->vring_tx_data[id];
823 
824 	wil_dbg_misc(wil, "%s() max_mpdu_size %d\n", __func__,
825 		     cmd.vring_cfg.tx_sw_ring.max_mpdu_size);
826 	lockdep_assert_held(&wil->mutex);
827 
828 	if (vring->va) {
829 		wil_err(wil, "Tx ring [%d] already allocated\n", id);
830 		rc = -EINVAL;
831 		goto out;
832 	}
833 
834 	memset(txdata, 0, sizeof(*txdata));
835 	spin_lock_init(&txdata->lock);
836 	vring->size = size;
837 	rc = wil_vring_alloc(wil, vring);
838 	if (rc)
839 		goto out;
840 
841 	wil->vring2cid_tid[id][0] = WIL6210_MAX_CID; /* CID */
842 	wil->vring2cid_tid[id][1] = 0; /* TID */
843 
844 	cmd.vring_cfg.tx_sw_ring.ring_mem_base = cpu_to_le64(vring->pa);
845 
846 	if (!wil->privacy)
847 		txdata->dot1x_open = true;
848 	rc = wmi_call(wil, WMI_BCAST_VRING_CFG_CMDID, &cmd, sizeof(cmd),
849 		      WMI_VRING_CFG_DONE_EVENTID, &reply, sizeof(reply), 100);
850 	if (rc)
851 		goto out_free;
852 
853 	if (reply.cmd.status != WMI_FW_STATUS_SUCCESS) {
854 		wil_err(wil, "Tx config failed, status 0x%02x\n",
855 			reply.cmd.status);
856 		rc = -EINVAL;
857 		goto out_free;
858 	}
859 	vring->hwtail = le32_to_cpu(reply.cmd.tx_vring_tail_ptr);
860 
861 	txdata->enabled = 1;
862 
863 	return 0;
864  out_free:
865 	txdata->enabled = 0;
866 	txdata->dot1x_open = false;
867 	wil_vring_free(wil, vring, 1);
868  out:
869 
870 	return rc;
871 }
872 
873 void wil_vring_fini_tx(struct wil6210_priv *wil, int id)
874 {
875 	struct vring *vring = &wil->vring_tx[id];
876 	struct vring_tx_data *txdata = &wil->vring_tx_data[id];
877 
878 	lockdep_assert_held(&wil->mutex);
879 
880 	if (!vring->va)
881 		return;
882 
883 	wil_dbg_misc(wil, "%s() id=%d\n", __func__, id);
884 
885 	spin_lock_bh(&txdata->lock);
886 	txdata->dot1x_open = false;
887 	txdata->enabled = 0; /* no Tx can be in progress or start anew */
888 	spin_unlock_bh(&txdata->lock);
889 	/* make sure NAPI won't touch this vring */
890 	if (test_bit(wil_status_napi_en, wil->status))
891 		napi_synchronize(&wil->napi_tx);
892 
893 	wil_vring_free(wil, vring, 1);
894 	memset(txdata, 0, sizeof(*txdata));
895 }
896 
897 static struct vring *wil_find_tx_ucast(struct wil6210_priv *wil,
898 				       struct sk_buff *skb)
899 {
900 	int i;
901 	struct ethhdr *eth = (void *)skb->data;
902 	int cid = wil_find_cid(wil, eth->h_dest);
903 
904 	if (cid < 0)
905 		return NULL;
906 
907 	/* TODO: fix for multiple TID */
908 	for (i = 0; i < ARRAY_SIZE(wil->vring2cid_tid); i++) {
909 		if (!wil->vring_tx_data[i].dot1x_open &&
910 		    (skb->protocol != cpu_to_be16(ETH_P_PAE)))
911 			continue;
912 		if (wil->vring2cid_tid[i][0] == cid) {
913 			struct vring *v = &wil->vring_tx[i];
914 
915 			wil_dbg_txrx(wil, "%s(%pM) -> [%d]\n",
916 				     __func__, eth->h_dest, i);
917 			if (v->va) {
918 				return v;
919 			} else {
920 				wil_dbg_txrx(wil, "vring[%d] not valid\n", i);
921 				return NULL;
922 			}
923 		}
924 	}
925 
926 	return NULL;
927 }
928 
929 static int wil_tx_vring(struct wil6210_priv *wil, struct vring *vring,
930 			struct sk_buff *skb);
931 
932 static struct vring *wil_find_tx_vring_sta(struct wil6210_priv *wil,
933 					   struct sk_buff *skb)
934 {
935 	struct vring *v;
936 	int i;
937 	u8 cid;
938 
939 	/* In the STA mode, it is expected to have only 1 VRING
940 	 * for the AP we connected to.
941 	 * find 1-st vring eligible for this skb and use it.
942 	 */
943 	for (i = 0; i < WIL6210_MAX_TX_RINGS; i++) {
944 		v = &wil->vring_tx[i];
945 		if (!v->va)
946 			continue;
947 
948 		cid = wil->vring2cid_tid[i][0];
949 		if (cid >= WIL6210_MAX_CID) /* skip BCAST */
950 			continue;
951 
952 		if (!wil->vring_tx_data[i].dot1x_open &&
953 		    (skb->protocol != cpu_to_be16(ETH_P_PAE)))
954 			continue;
955 
956 		wil_dbg_txrx(wil, "Tx -> ring %d\n", i);
957 
958 		return v;
959 	}
960 
961 	wil_dbg_txrx(wil, "Tx while no vrings active?\n");
962 
963 	return NULL;
964 }
965 
966 /* Use one of 2 strategies:
967  *
968  * 1. New (real broadcast):
969  *    use dedicated broadcast vring
970  * 2. Old (pseudo-DMS):
971  *    Find 1-st vring and return it;
972  *    duplicate skb and send it to other active vrings;
973  *    in all cases override dest address to unicast peer's address
974  * Use old strategy when new is not supported yet:
975  *  - for PBSS
976  */
977 static struct vring *wil_find_tx_bcast_1(struct wil6210_priv *wil,
978 					 struct sk_buff *skb)
979 {
980 	struct vring *v;
981 	int i = wil->bcast_vring;
982 
983 	if (i < 0)
984 		return NULL;
985 	v = &wil->vring_tx[i];
986 	if (!v->va)
987 		return NULL;
988 	if (!wil->vring_tx_data[i].dot1x_open &&
989 	    (skb->protocol != cpu_to_be16(ETH_P_PAE)))
990 		return NULL;
991 
992 	return v;
993 }
994 
995 static void wil_set_da_for_vring(struct wil6210_priv *wil,
996 				 struct sk_buff *skb, int vring_index)
997 {
998 	struct ethhdr *eth = (void *)skb->data;
999 	int cid = wil->vring2cid_tid[vring_index][0];
1000 
1001 	ether_addr_copy(eth->h_dest, wil->sta[cid].addr);
1002 }
1003 
1004 static struct vring *wil_find_tx_bcast_2(struct wil6210_priv *wil,
1005 					 struct sk_buff *skb)
1006 {
1007 	struct vring *v, *v2;
1008 	struct sk_buff *skb2;
1009 	int i;
1010 	u8 cid;
1011 	struct ethhdr *eth = (void *)skb->data;
1012 	char *src = eth->h_source;
1013 
1014 	/* find 1-st vring eligible for data */
1015 	for (i = 0; i < WIL6210_MAX_TX_RINGS; i++) {
1016 		v = &wil->vring_tx[i];
1017 		if (!v->va)
1018 			continue;
1019 
1020 		cid = wil->vring2cid_tid[i][0];
1021 		if (cid >= WIL6210_MAX_CID) /* skip BCAST */
1022 			continue;
1023 		if (!wil->vring_tx_data[i].dot1x_open &&
1024 		    (skb->protocol != cpu_to_be16(ETH_P_PAE)))
1025 			continue;
1026 
1027 		/* don't Tx back to source when re-routing Rx->Tx at the AP */
1028 		if (0 == memcmp(wil->sta[cid].addr, src, ETH_ALEN))
1029 			continue;
1030 
1031 		goto found;
1032 	}
1033 
1034 	wil_dbg_txrx(wil, "Tx while no vrings active?\n");
1035 
1036 	return NULL;
1037 
1038 found:
1039 	wil_dbg_txrx(wil, "BCAST -> ring %d\n", i);
1040 	wil_set_da_for_vring(wil, skb, i);
1041 
1042 	/* find other active vrings and duplicate skb for each */
1043 	for (i++; i < WIL6210_MAX_TX_RINGS; i++) {
1044 		v2 = &wil->vring_tx[i];
1045 		if (!v2->va)
1046 			continue;
1047 		cid = wil->vring2cid_tid[i][0];
1048 		if (cid >= WIL6210_MAX_CID) /* skip BCAST */
1049 			continue;
1050 		if (!wil->vring_tx_data[i].dot1x_open &&
1051 		    (skb->protocol != cpu_to_be16(ETH_P_PAE)))
1052 			continue;
1053 
1054 		if (0 == memcmp(wil->sta[cid].addr, src, ETH_ALEN))
1055 			continue;
1056 
1057 		skb2 = skb_copy(skb, GFP_ATOMIC);
1058 		if (skb2) {
1059 			wil_dbg_txrx(wil, "BCAST DUP -> ring %d\n", i);
1060 			wil_set_da_for_vring(wil, skb2, i);
1061 			wil_tx_vring(wil, v2, skb2);
1062 		} else {
1063 			wil_err(wil, "skb_copy failed\n");
1064 		}
1065 	}
1066 
1067 	return v;
1068 }
1069 
1070 static struct vring *wil_find_tx_bcast(struct wil6210_priv *wil,
1071 				       struct sk_buff *skb)
1072 {
1073 	struct wireless_dev *wdev = wil->wdev;
1074 
1075 	if (wdev->iftype != NL80211_IFTYPE_AP)
1076 		return wil_find_tx_bcast_2(wil, skb);
1077 
1078 	return wil_find_tx_bcast_1(wil, skb);
1079 }
1080 
1081 static int wil_tx_desc_map(struct vring_tx_desc *d, dma_addr_t pa, u32 len,
1082 			   int vring_index)
1083 {
1084 	wil_desc_addr_set(&d->dma.addr, pa);
1085 	d->dma.ip_length = 0;
1086 	/* 0..6: mac_length; 7:ip_version 0-IP6 1-IP4*/
1087 	d->dma.b11 = 0/*14 | BIT(7)*/;
1088 	d->dma.error = 0;
1089 	d->dma.status = 0; /* BIT(0) should be 0 for HW_OWNED */
1090 	d->dma.length = cpu_to_le16((u16)len);
1091 	d->dma.d0 = (vring_index << DMA_CFG_DESC_TX_0_QID_POS);
1092 	d->mac.d[0] = 0;
1093 	d->mac.d[1] = 0;
1094 	d->mac.d[2] = 0;
1095 	d->mac.ucode_cmd = 0;
1096 	/* translation type:  0 - bypass; 1 - 802.3; 2 - native wifi */
1097 	d->mac.d[2] = BIT(MAC_CFG_DESC_TX_2_SNAP_HDR_INSERTION_EN_POS) |
1098 		      (1 << MAC_CFG_DESC_TX_2_L2_TRANSLATION_TYPE_POS);
1099 
1100 	return 0;
1101 }
1102 
1103 static inline
1104 void wil_tx_desc_set_nr_frags(struct vring_tx_desc *d, int nr_frags)
1105 {
1106 	d->mac.d[2] |= (nr_frags << MAC_CFG_DESC_TX_2_NUM_OF_DESCRIPTORS_POS);
1107 }
1108 
1109 /**
1110  * Sets the descriptor @d up for csum and/or TSO offloading. The corresponding
1111  * @skb is used to obtain the protocol and headers length.
1112  * @tso_desc_type is a descriptor type for TSO: 0 - a header, 1 - first data,
1113  * 2 - middle, 3 - last descriptor.
1114  */
1115 
1116 static void wil_tx_desc_offload_setup_tso(struct vring_tx_desc *d,
1117 					  struct sk_buff *skb,
1118 					  int tso_desc_type, bool is_ipv4,
1119 					  int tcp_hdr_len, int skb_net_hdr_len)
1120 {
1121 	d->dma.b11 = ETH_HLEN; /* MAC header length */
1122 	d->dma.b11 |= is_ipv4 << DMA_CFG_DESC_TX_OFFLOAD_CFG_L3T_IPV4_POS;
1123 
1124 	d->dma.d0 |= (2 << DMA_CFG_DESC_TX_0_L4_TYPE_POS);
1125 	/* L4 header len: TCP header length */
1126 	d->dma.d0 |= (tcp_hdr_len & DMA_CFG_DESC_TX_0_L4_LENGTH_MSK);
1127 
1128 	/* Setup TSO: bit and desc type */
1129 	d->dma.d0 |= (BIT(DMA_CFG_DESC_TX_0_TCP_SEG_EN_POS)) |
1130 		(tso_desc_type << DMA_CFG_DESC_TX_0_SEGMENT_BUF_DETAILS_POS);
1131 	d->dma.d0 |= (is_ipv4 << DMA_CFG_DESC_TX_0_IPV4_CHECKSUM_EN_POS);
1132 
1133 	d->dma.ip_length = skb_net_hdr_len;
1134 	/* Enable TCP/UDP checksum */
1135 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_TCP_UDP_CHECKSUM_EN_POS);
1136 	/* Calculate pseudo-header */
1137 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_PSEUDO_HEADER_CALC_EN_POS);
1138 }
1139 
1140 /**
1141  * Sets the descriptor @d up for csum. The corresponding
1142  * @skb is used to obtain the protocol and headers length.
1143  * Returns the protocol: 0 - not TCP, 1 - TCPv4, 2 - TCPv6.
1144  * Note, if d==NULL, the function only returns the protocol result.
1145  *
1146  * It is very similar to previous wil_tx_desc_offload_setup_tso. This
1147  * is "if unrolling" to optimize the critical path.
1148  */
1149 
1150 static int wil_tx_desc_offload_setup(struct vring_tx_desc *d,
1151 				     struct sk_buff *skb){
1152 	int protocol;
1153 
1154 	if (skb->ip_summed != CHECKSUM_PARTIAL)
1155 		return 0;
1156 
1157 	d->dma.b11 = ETH_HLEN; /* MAC header length */
1158 
1159 	switch (skb->protocol) {
1160 	case cpu_to_be16(ETH_P_IP):
1161 		protocol = ip_hdr(skb)->protocol;
1162 		d->dma.b11 |= BIT(DMA_CFG_DESC_TX_OFFLOAD_CFG_L3T_IPV4_POS);
1163 		break;
1164 	case cpu_to_be16(ETH_P_IPV6):
1165 		protocol = ipv6_hdr(skb)->nexthdr;
1166 		break;
1167 	default:
1168 		return -EINVAL;
1169 	}
1170 
1171 	switch (protocol) {
1172 	case IPPROTO_TCP:
1173 		d->dma.d0 |= (2 << DMA_CFG_DESC_TX_0_L4_TYPE_POS);
1174 		/* L4 header len: TCP header length */
1175 		d->dma.d0 |=
1176 		(tcp_hdrlen(skb) & DMA_CFG_DESC_TX_0_L4_LENGTH_MSK);
1177 		break;
1178 	case IPPROTO_UDP:
1179 		/* L4 header len: UDP header length */
1180 		d->dma.d0 |=
1181 		(sizeof(struct udphdr) & DMA_CFG_DESC_TX_0_L4_LENGTH_MSK);
1182 		break;
1183 	default:
1184 		return -EINVAL;
1185 	}
1186 
1187 	d->dma.ip_length = skb_network_header_len(skb);
1188 	/* Enable TCP/UDP checksum */
1189 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_TCP_UDP_CHECKSUM_EN_POS);
1190 	/* Calculate pseudo-header */
1191 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_PSEUDO_HEADER_CALC_EN_POS);
1192 
1193 	return 0;
1194 }
1195 
1196 static inline void wil_tx_last_desc(struct vring_tx_desc *d)
1197 {
1198 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_CMD_EOP_POS) |
1199 	      BIT(DMA_CFG_DESC_TX_0_CMD_MARK_WB_POS) |
1200 	      BIT(DMA_CFG_DESC_TX_0_CMD_DMA_IT_POS);
1201 }
1202 
1203 static inline void wil_set_tx_desc_last_tso(volatile struct vring_tx_desc *d)
1204 {
1205 	d->dma.d0 |= wil_tso_type_lst <<
1206 		  DMA_CFG_DESC_TX_0_SEGMENT_BUF_DETAILS_POS;
1207 }
1208 
1209 static int __wil_tx_vring_tso(struct wil6210_priv *wil, struct vring *vring,
1210 			      struct sk_buff *skb)
1211 {
1212 	struct device *dev = wil_to_dev(wil);
1213 
1214 	/* point to descriptors in shared memory */
1215 	volatile struct vring_tx_desc *_desc = NULL, *_hdr_desc,
1216 				      *_first_desc = NULL;
1217 
1218 	/* pointers to shadow descriptors */
1219 	struct vring_tx_desc desc_mem, hdr_desc_mem, first_desc_mem,
1220 			     *d = &hdr_desc_mem, *hdr_desc = &hdr_desc_mem,
1221 			     *first_desc = &first_desc_mem;
1222 
1223 	/* pointer to shadow descriptors' context */
1224 	struct wil_ctx *hdr_ctx, *first_ctx = NULL;
1225 
1226 	int descs_used = 0; /* total number of used descriptors */
1227 	int sg_desc_cnt = 0; /* number of descriptors for current mss*/
1228 
1229 	u32 swhead = vring->swhead;
1230 	int used, avail = wil_vring_avail_tx(vring);
1231 	int nr_frags = skb_shinfo(skb)->nr_frags;
1232 	int min_desc_required = nr_frags + 1;
1233 	int mss = skb_shinfo(skb)->gso_size;	/* payload size w/o headers */
1234 	int f, len, hdrlen, headlen;
1235 	int vring_index = vring - wil->vring_tx;
1236 	struct vring_tx_data *txdata = &wil->vring_tx_data[vring_index];
1237 	uint i = swhead;
1238 	dma_addr_t pa;
1239 	const skb_frag_t *frag = NULL;
1240 	int rem_data = mss;
1241 	int lenmss;
1242 	int hdr_compensation_need = true;
1243 	int desc_tso_type = wil_tso_type_first;
1244 	bool is_ipv4;
1245 	int tcp_hdr_len;
1246 	int skb_net_hdr_len;
1247 	int gso_type;
1248 	int rc = -EINVAL;
1249 
1250 	wil_dbg_txrx(wil, "%s() %d bytes to vring %d\n",
1251 		     __func__, skb->len, vring_index);
1252 
1253 	if (unlikely(!txdata->enabled))
1254 		return -EINVAL;
1255 
1256 	/* A typical page 4K is 3-4 payloads, we assume each fragment
1257 	 * is a full payload, that's how min_desc_required has been
1258 	 * calculated. In real we might need more or less descriptors,
1259 	 * this is the initial check only.
1260 	 */
1261 	if (unlikely(avail < min_desc_required)) {
1262 		wil_err_ratelimited(wil,
1263 				    "TSO: Tx ring[%2d] full. No space for %d fragments\n",
1264 				    vring_index, min_desc_required);
1265 		return -ENOMEM;
1266 	}
1267 
1268 	/* Header Length = MAC header len + IP header len + TCP header len*/
1269 	hdrlen = ETH_HLEN +
1270 		(int)skb_network_header_len(skb) +
1271 		tcp_hdrlen(skb);
1272 
1273 	gso_type = skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV6 | SKB_GSO_TCPV4);
1274 	switch (gso_type) {
1275 	case SKB_GSO_TCPV4:
1276 		/* TCP v4, zero out the IP length and IPv4 checksum fields
1277 		 * as required by the offloading doc
1278 		 */
1279 		ip_hdr(skb)->tot_len = 0;
1280 		ip_hdr(skb)->check = 0;
1281 		is_ipv4 = true;
1282 		break;
1283 	case SKB_GSO_TCPV6:
1284 		/* TCP v6, zero out the payload length */
1285 		ipv6_hdr(skb)->payload_len = 0;
1286 		is_ipv4 = false;
1287 		break;
1288 	default:
1289 		/* other than TCPv4 or TCPv6 types are not supported for TSO.
1290 		 * It is also illegal for both to be set simultaneously
1291 		 */
1292 		return -EINVAL;
1293 	}
1294 
1295 	if (skb->ip_summed != CHECKSUM_PARTIAL)
1296 		return -EINVAL;
1297 
1298 	/* tcp header length and skb network header length are fixed for all
1299 	 * packet's descriptors - read then once here
1300 	 */
1301 	tcp_hdr_len = tcp_hdrlen(skb);
1302 	skb_net_hdr_len = skb_network_header_len(skb);
1303 
1304 	_hdr_desc = &vring->va[i].tx;
1305 
1306 	pa = dma_map_single(dev, skb->data, hdrlen, DMA_TO_DEVICE);
1307 	if (unlikely(dma_mapping_error(dev, pa))) {
1308 		wil_err(wil, "TSO: Skb head DMA map error\n");
1309 		goto err_exit;
1310 	}
1311 
1312 	wil_tx_desc_map(hdr_desc, pa, hdrlen, vring_index);
1313 	wil_tx_desc_offload_setup_tso(hdr_desc, skb, wil_tso_type_hdr, is_ipv4,
1314 				      tcp_hdr_len, skb_net_hdr_len);
1315 	wil_tx_last_desc(hdr_desc);
1316 
1317 	vring->ctx[i].mapped_as = wil_mapped_as_single;
1318 	hdr_ctx = &vring->ctx[i];
1319 
1320 	descs_used++;
1321 	headlen = skb_headlen(skb) - hdrlen;
1322 
1323 	for (f = headlen ? -1 : 0; f < nr_frags; f++)  {
1324 		if (headlen) {
1325 			len = headlen;
1326 			wil_dbg_txrx(wil, "TSO: process skb head, len %u\n",
1327 				     len);
1328 		} else {
1329 			frag = &skb_shinfo(skb)->frags[f];
1330 			len = frag->size;
1331 			wil_dbg_txrx(wil, "TSO: frag[%d]: len %u\n", f, len);
1332 		}
1333 
1334 		while (len) {
1335 			wil_dbg_txrx(wil,
1336 				     "TSO: len %d, rem_data %d, descs_used %d\n",
1337 				     len, rem_data, descs_used);
1338 
1339 			if (descs_used == avail)  {
1340 				wil_err_ratelimited(wil, "TSO: ring overflow\n");
1341 				rc = -ENOMEM;
1342 				goto mem_error;
1343 			}
1344 
1345 			lenmss = min_t(int, rem_data, len);
1346 			i = (swhead + descs_used) % vring->size;
1347 			wil_dbg_txrx(wil, "TSO: lenmss %d, i %d\n", lenmss, i);
1348 
1349 			if (!headlen) {
1350 				pa = skb_frag_dma_map(dev, frag,
1351 						      frag->size - len, lenmss,
1352 						      DMA_TO_DEVICE);
1353 				vring->ctx[i].mapped_as = wil_mapped_as_page;
1354 			} else {
1355 				pa = dma_map_single(dev,
1356 						    skb->data +
1357 						    skb_headlen(skb) - headlen,
1358 						    lenmss,
1359 						    DMA_TO_DEVICE);
1360 				vring->ctx[i].mapped_as = wil_mapped_as_single;
1361 				headlen -= lenmss;
1362 			}
1363 
1364 			if (unlikely(dma_mapping_error(dev, pa))) {
1365 				wil_err(wil, "TSO: DMA map page error\n");
1366 				goto mem_error;
1367 			}
1368 
1369 			_desc = &vring->va[i].tx;
1370 
1371 			if (!_first_desc) {
1372 				_first_desc = _desc;
1373 				first_ctx = &vring->ctx[i];
1374 				d = first_desc;
1375 			} else {
1376 				d = &desc_mem;
1377 			}
1378 
1379 			wil_tx_desc_map(d, pa, lenmss, vring_index);
1380 			wil_tx_desc_offload_setup_tso(d, skb, desc_tso_type,
1381 						      is_ipv4, tcp_hdr_len,
1382 						      skb_net_hdr_len);
1383 
1384 			/* use tso_type_first only once */
1385 			desc_tso_type = wil_tso_type_mid;
1386 
1387 			descs_used++;  /* desc used so far */
1388 			sg_desc_cnt++; /* desc used for this segment */
1389 			len -= lenmss;
1390 			rem_data -= lenmss;
1391 
1392 			wil_dbg_txrx(wil,
1393 				     "TSO: len %d, rem_data %d, descs_used %d, sg_desc_cnt %d,\n",
1394 				     len, rem_data, descs_used, sg_desc_cnt);
1395 
1396 			/* Close the segment if reached mss size or last frag*/
1397 			if (rem_data == 0 || (f == nr_frags - 1 && len == 0)) {
1398 				if (hdr_compensation_need) {
1399 					/* first segment include hdr desc for
1400 					 * release
1401 					 */
1402 					hdr_ctx->nr_frags = sg_desc_cnt;
1403 					wil_tx_desc_set_nr_frags(first_desc,
1404 								 sg_desc_cnt +
1405 								 1);
1406 					hdr_compensation_need = false;
1407 				} else {
1408 					wil_tx_desc_set_nr_frags(first_desc,
1409 								 sg_desc_cnt);
1410 				}
1411 				first_ctx->nr_frags = sg_desc_cnt - 1;
1412 
1413 				wil_tx_last_desc(d);
1414 
1415 				/* first descriptor may also be the last
1416 				 * for this mss - make sure not to copy
1417 				 * it twice
1418 				 */
1419 				if (first_desc != d)
1420 					*_first_desc = *first_desc;
1421 
1422 				/*last descriptor will be copied at the end
1423 				 * of this TS processing
1424 				 */
1425 				if (f < nr_frags - 1 || len > 0)
1426 					*_desc = *d;
1427 
1428 				rem_data = mss;
1429 				_first_desc = NULL;
1430 				sg_desc_cnt = 0;
1431 			} else if (first_desc != d) /* update mid descriptor */
1432 					*_desc = *d;
1433 		}
1434 	}
1435 
1436 	/* first descriptor may also be the last.
1437 	 * in this case d pointer is invalid
1438 	 */
1439 	if (_first_desc == _desc)
1440 		d = first_desc;
1441 
1442 	/* Last data descriptor */
1443 	wil_set_tx_desc_last_tso(d);
1444 	*_desc = *d;
1445 
1446 	/* Fill the total number of descriptors in first desc (hdr)*/
1447 	wil_tx_desc_set_nr_frags(hdr_desc, descs_used);
1448 	*_hdr_desc = *hdr_desc;
1449 
1450 	/* hold reference to skb
1451 	 * to prevent skb release before accounting
1452 	 * in case of immediate "tx done"
1453 	 */
1454 	vring->ctx[i].skb = skb_get(skb);
1455 
1456 	/* performance monitoring */
1457 	used = wil_vring_used_tx(vring);
1458 	if (wil_val_in_range(vring_idle_trsh,
1459 			     used, used + descs_used)) {
1460 		txdata->idle += get_cycles() - txdata->last_idle;
1461 		wil_dbg_txrx(wil,  "Ring[%2d] not idle %d -> %d\n",
1462 			     vring_index, used, used + descs_used);
1463 	}
1464 
1465 	/* advance swhead */
1466 	wil_vring_advance_head(vring, descs_used);
1467 	wil_dbg_txrx(wil, "TSO: Tx swhead %d -> %d\n", swhead, vring->swhead);
1468 
1469 	/* make sure all writes to descriptors (shared memory) are done before
1470 	 * committing them to HW
1471 	 */
1472 	wmb();
1473 
1474 	wil_w(wil, vring->hwtail, vring->swhead);
1475 	return 0;
1476 
1477 mem_error:
1478 	while (descs_used > 0) {
1479 		struct wil_ctx *ctx;
1480 
1481 		i = (swhead + descs_used) % vring->size;
1482 		d = (struct vring_tx_desc *)&vring->va[i].tx;
1483 		_desc = &vring->va[i].tx;
1484 		*d = *_desc;
1485 		_desc->dma.status = TX_DMA_STATUS_DU;
1486 		ctx = &vring->ctx[i];
1487 		wil_txdesc_unmap(dev, d, ctx);
1488 		memset(ctx, 0, sizeof(*ctx));
1489 		descs_used--;
1490 	}
1491 err_exit:
1492 	return rc;
1493 }
1494 
1495 static int __wil_tx_vring(struct wil6210_priv *wil, struct vring *vring,
1496 			  struct sk_buff *skb)
1497 {
1498 	struct device *dev = wil_to_dev(wil);
1499 	struct vring_tx_desc dd, *d = &dd;
1500 	volatile struct vring_tx_desc *_d;
1501 	u32 swhead = vring->swhead;
1502 	int avail = wil_vring_avail_tx(vring);
1503 	int nr_frags = skb_shinfo(skb)->nr_frags;
1504 	uint f = 0;
1505 	int vring_index = vring - wil->vring_tx;
1506 	struct vring_tx_data *txdata = &wil->vring_tx_data[vring_index];
1507 	uint i = swhead;
1508 	dma_addr_t pa;
1509 	int used;
1510 	bool mcast = (vring_index == wil->bcast_vring);
1511 	uint len = skb_headlen(skb);
1512 
1513 	wil_dbg_txrx(wil, "%s() %d bytes to vring %d\n",
1514 		     __func__, skb->len, vring_index);
1515 
1516 	if (unlikely(!txdata->enabled))
1517 		return -EINVAL;
1518 
1519 	if (unlikely(avail < 1 + nr_frags)) {
1520 		wil_err_ratelimited(wil,
1521 				    "Tx ring[%2d] full. No space for %d fragments\n",
1522 				    vring_index, 1 + nr_frags);
1523 		return -ENOMEM;
1524 	}
1525 	_d = &vring->va[i].tx;
1526 
1527 	pa = dma_map_single(dev, skb->data, skb_headlen(skb), DMA_TO_DEVICE);
1528 
1529 	wil_dbg_txrx(wil, "Tx[%2d] skb %d bytes 0x%p -> %pad\n", vring_index,
1530 		     skb_headlen(skb), skb->data, &pa);
1531 	wil_hex_dump_txrx("Tx ", DUMP_PREFIX_OFFSET, 16, 1,
1532 			  skb->data, skb_headlen(skb), false);
1533 
1534 	if (unlikely(dma_mapping_error(dev, pa)))
1535 		return -EINVAL;
1536 	vring->ctx[i].mapped_as = wil_mapped_as_single;
1537 	/* 1-st segment */
1538 	wil_tx_desc_map(d, pa, len, vring_index);
1539 	if (unlikely(mcast)) {
1540 		d->mac.d[0] |= BIT(MAC_CFG_DESC_TX_0_MCS_EN_POS); /* MCS 0 */
1541 		if (unlikely(len > WIL_BCAST_MCS0_LIMIT)) /* set MCS 1 */
1542 			d->mac.d[0] |= (1 << MAC_CFG_DESC_TX_0_MCS_INDEX_POS);
1543 	}
1544 	/* Process TCP/UDP checksum offloading */
1545 	if (unlikely(wil_tx_desc_offload_setup(d, skb))) {
1546 		wil_err(wil, "Tx[%2d] Failed to set cksum, drop packet\n",
1547 			vring_index);
1548 		goto dma_error;
1549 	}
1550 
1551 	vring->ctx[i].nr_frags = nr_frags;
1552 	wil_tx_desc_set_nr_frags(d, nr_frags + 1);
1553 
1554 	/* middle segments */
1555 	for (; f < nr_frags; f++) {
1556 		const struct skb_frag_struct *frag =
1557 				&skb_shinfo(skb)->frags[f];
1558 		int len = skb_frag_size(frag);
1559 
1560 		*_d = *d;
1561 		wil_dbg_txrx(wil, "Tx[%2d] desc[%4d]\n", vring_index, i);
1562 		wil_hex_dump_txrx("TxD ", DUMP_PREFIX_NONE, 32, 4,
1563 				  (const void *)d, sizeof(*d), false);
1564 		i = (swhead + f + 1) % vring->size;
1565 		_d = &vring->va[i].tx;
1566 		pa = skb_frag_dma_map(dev, frag, 0, skb_frag_size(frag),
1567 				      DMA_TO_DEVICE);
1568 		if (unlikely(dma_mapping_error(dev, pa))) {
1569 			wil_err(wil, "Tx[%2d] failed to map fragment\n",
1570 				vring_index);
1571 			goto dma_error;
1572 		}
1573 		vring->ctx[i].mapped_as = wil_mapped_as_page;
1574 		wil_tx_desc_map(d, pa, len, vring_index);
1575 		/* no need to check return code -
1576 		 * if it succeeded for 1-st descriptor,
1577 		 * it will succeed here too
1578 		 */
1579 		wil_tx_desc_offload_setup(d, skb);
1580 	}
1581 	/* for the last seg only */
1582 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_CMD_EOP_POS);
1583 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_CMD_MARK_WB_POS);
1584 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_CMD_DMA_IT_POS);
1585 	*_d = *d;
1586 	wil_dbg_txrx(wil, "Tx[%2d] desc[%4d]\n", vring_index, i);
1587 	wil_hex_dump_txrx("TxD ", DUMP_PREFIX_NONE, 32, 4,
1588 			  (const void *)d, sizeof(*d), false);
1589 
1590 	/* hold reference to skb
1591 	 * to prevent skb release before accounting
1592 	 * in case of immediate "tx done"
1593 	 */
1594 	vring->ctx[i].skb = skb_get(skb);
1595 
1596 	/* performance monitoring */
1597 	used = wil_vring_used_tx(vring);
1598 	if (wil_val_in_range(vring_idle_trsh,
1599 			     used, used + nr_frags + 1)) {
1600 		txdata->idle += get_cycles() - txdata->last_idle;
1601 		wil_dbg_txrx(wil,  "Ring[%2d] not idle %d -> %d\n",
1602 			     vring_index, used, used + nr_frags + 1);
1603 	}
1604 
1605 	/* advance swhead */
1606 	wil_vring_advance_head(vring, nr_frags + 1);
1607 	wil_dbg_txrx(wil, "Tx[%2d] swhead %d -> %d\n", vring_index, swhead,
1608 		     vring->swhead);
1609 	trace_wil6210_tx(vring_index, swhead, skb->len, nr_frags);
1610 
1611 	/* make sure all writes to descriptors (shared memory) are done before
1612 	 * committing them to HW
1613 	 */
1614 	wmb();
1615 
1616 	wil_w(wil, vring->hwtail, vring->swhead);
1617 
1618 	return 0;
1619  dma_error:
1620 	/* unmap what we have mapped */
1621 	nr_frags = f + 1; /* frags mapped + one for skb head */
1622 	for (f = 0; f < nr_frags; f++) {
1623 		struct wil_ctx *ctx;
1624 
1625 		i = (swhead + f) % vring->size;
1626 		ctx = &vring->ctx[i];
1627 		_d = &vring->va[i].tx;
1628 		*d = *_d;
1629 		_d->dma.status = TX_DMA_STATUS_DU;
1630 		wil_txdesc_unmap(dev, d, ctx);
1631 
1632 		memset(ctx, 0, sizeof(*ctx));
1633 	}
1634 
1635 	return -EINVAL;
1636 }
1637 
1638 static int wil_tx_vring(struct wil6210_priv *wil, struct vring *vring,
1639 			struct sk_buff *skb)
1640 {
1641 	int vring_index = vring - wil->vring_tx;
1642 	struct vring_tx_data *txdata = &wil->vring_tx_data[vring_index];
1643 	int rc;
1644 
1645 	spin_lock(&txdata->lock);
1646 
1647 	rc = (skb_is_gso(skb) ? __wil_tx_vring_tso : __wil_tx_vring)
1648 	     (wil, vring, skb);
1649 
1650 	spin_unlock(&txdata->lock);
1651 
1652 	return rc;
1653 }
1654 
1655 netdev_tx_t wil_start_xmit(struct sk_buff *skb, struct net_device *ndev)
1656 {
1657 	struct wil6210_priv *wil = ndev_to_wil(ndev);
1658 	struct ethhdr *eth = (void *)skb->data;
1659 	bool bcast = is_multicast_ether_addr(eth->h_dest);
1660 	struct vring *vring;
1661 	static bool pr_once_fw;
1662 	int rc;
1663 
1664 	wil_dbg_txrx(wil, "%s()\n", __func__);
1665 	if (unlikely(!test_bit(wil_status_fwready, wil->status))) {
1666 		if (!pr_once_fw) {
1667 			wil_err(wil, "FW not ready\n");
1668 			pr_once_fw = true;
1669 		}
1670 		goto drop;
1671 	}
1672 	if (unlikely(!test_bit(wil_status_fwconnected, wil->status))) {
1673 		wil_err_ratelimited(wil, "FW not connected\n");
1674 		goto drop;
1675 	}
1676 	if (unlikely(wil->wdev->iftype == NL80211_IFTYPE_MONITOR)) {
1677 		wil_err(wil, "Xmit in monitor mode not supported\n");
1678 		goto drop;
1679 	}
1680 	pr_once_fw = false;
1681 
1682 	/* find vring */
1683 	if (wil->wdev->iftype == NL80211_IFTYPE_STATION) {
1684 		/* in STA mode (ESS), all to same VRING */
1685 		vring = wil_find_tx_vring_sta(wil, skb);
1686 	} else { /* direct communication, find matching VRING */
1687 		vring = bcast ? wil_find_tx_bcast(wil, skb) :
1688 				wil_find_tx_ucast(wil, skb);
1689 	}
1690 	if (unlikely(!vring)) {
1691 		wil_dbg_txrx(wil, "No Tx VRING found for %pM\n", eth->h_dest);
1692 		goto drop;
1693 	}
1694 	/* set up vring entry */
1695 	rc = wil_tx_vring(wil, vring, skb);
1696 
1697 	/* do we still have enough room in the vring? */
1698 	if (unlikely(wil_vring_avail_tx(vring) < wil_vring_wmark_low(vring))) {
1699 		netif_tx_stop_all_queues(wil_to_ndev(wil));
1700 		wil_dbg_txrx(wil, "netif_tx_stop : ring full\n");
1701 	}
1702 
1703 	switch (rc) {
1704 	case 0:
1705 		/* statistics will be updated on the tx_complete */
1706 		dev_kfree_skb_any(skb);
1707 		return NETDEV_TX_OK;
1708 	case -ENOMEM:
1709 		return NETDEV_TX_BUSY;
1710 	default:
1711 		break; /* goto drop; */
1712 	}
1713  drop:
1714 	ndev->stats.tx_dropped++;
1715 	dev_kfree_skb_any(skb);
1716 
1717 	return NET_XMIT_DROP;
1718 }
1719 
1720 static inline bool wil_need_txstat(struct sk_buff *skb)
1721 {
1722 	struct ethhdr *eth = (void *)skb->data;
1723 
1724 	return is_unicast_ether_addr(eth->h_dest) && skb->sk &&
1725 	       (skb_shinfo(skb)->tx_flags & SKBTX_WIFI_STATUS);
1726 }
1727 
1728 static inline void wil_consume_skb(struct sk_buff *skb, bool acked)
1729 {
1730 	if (unlikely(wil_need_txstat(skb)))
1731 		skb_complete_wifi_ack(skb, acked);
1732 	else
1733 		acked ? dev_consume_skb_any(skb) : dev_kfree_skb_any(skb);
1734 }
1735 
1736 /**
1737  * Clean up transmitted skb's from the Tx VRING
1738  *
1739  * Return number of descriptors cleared
1740  *
1741  * Safe to call from IRQ
1742  */
1743 int wil_tx_complete(struct wil6210_priv *wil, int ringid)
1744 {
1745 	struct net_device *ndev = wil_to_ndev(wil);
1746 	struct device *dev = wil_to_dev(wil);
1747 	struct vring *vring = &wil->vring_tx[ringid];
1748 	struct vring_tx_data *txdata = &wil->vring_tx_data[ringid];
1749 	int done = 0;
1750 	int cid = wil->vring2cid_tid[ringid][0];
1751 	struct wil_net_stats *stats = NULL;
1752 	volatile struct vring_tx_desc *_d;
1753 	int used_before_complete;
1754 	int used_new;
1755 
1756 	if (unlikely(!vring->va)) {
1757 		wil_err(wil, "Tx irq[%d]: vring not initialized\n", ringid);
1758 		return 0;
1759 	}
1760 
1761 	if (unlikely(!txdata->enabled)) {
1762 		wil_info(wil, "Tx irq[%d]: vring disabled\n", ringid);
1763 		return 0;
1764 	}
1765 
1766 	wil_dbg_txrx(wil, "%s(%d)\n", __func__, ringid);
1767 
1768 	used_before_complete = wil_vring_used_tx(vring);
1769 
1770 	if (cid < WIL6210_MAX_CID)
1771 		stats = &wil->sta[cid].stats;
1772 
1773 	while (!wil_vring_is_empty(vring)) {
1774 		int new_swtail;
1775 		struct wil_ctx *ctx = &vring->ctx[vring->swtail];
1776 		/**
1777 		 * For the fragmented skb, HW will set DU bit only for the
1778 		 * last fragment. look for it.
1779 		 * In TSO the first DU will include hdr desc
1780 		 */
1781 		int lf = (vring->swtail + ctx->nr_frags) % vring->size;
1782 		/* TODO: check we are not past head */
1783 
1784 		_d = &vring->va[lf].tx;
1785 		if (unlikely(!(_d->dma.status & TX_DMA_STATUS_DU)))
1786 			break;
1787 
1788 		new_swtail = (lf + 1) % vring->size;
1789 		while (vring->swtail != new_swtail) {
1790 			struct vring_tx_desc dd, *d = &dd;
1791 			u16 dmalen;
1792 			struct sk_buff *skb;
1793 
1794 			ctx = &vring->ctx[vring->swtail];
1795 			skb = ctx->skb;
1796 			_d = &vring->va[vring->swtail].tx;
1797 
1798 			*d = *_d;
1799 
1800 			dmalen = le16_to_cpu(d->dma.length);
1801 			trace_wil6210_tx_done(ringid, vring->swtail, dmalen,
1802 					      d->dma.error);
1803 			wil_dbg_txrx(wil,
1804 				     "TxC[%2d][%3d] : %d bytes, status 0x%02x err 0x%02x\n",
1805 				     ringid, vring->swtail, dmalen,
1806 				     d->dma.status, d->dma.error);
1807 			wil_hex_dump_txrx("TxCD ", DUMP_PREFIX_NONE, 32, 4,
1808 					  (const void *)d, sizeof(*d), false);
1809 
1810 			wil_txdesc_unmap(dev, d, ctx);
1811 
1812 			if (skb) {
1813 				if (likely(d->dma.error == 0)) {
1814 					ndev->stats.tx_packets++;
1815 					ndev->stats.tx_bytes += skb->len;
1816 					if (stats) {
1817 						stats->tx_packets++;
1818 						stats->tx_bytes += skb->len;
1819 					}
1820 				} else {
1821 					ndev->stats.tx_errors++;
1822 					if (stats)
1823 						stats->tx_errors++;
1824 				}
1825 				wil_consume_skb(skb, d->dma.error == 0);
1826 			}
1827 			memset(ctx, 0, sizeof(*ctx));
1828 			/* There is no need to touch HW descriptor:
1829 			 * - ststus bit TX_DMA_STATUS_DU is set by design,
1830 			 *   so hardware will not try to process this desc.,
1831 			 * - rest of descriptor will be initialized on Tx.
1832 			 */
1833 			vring->swtail = wil_vring_next_tail(vring);
1834 			done++;
1835 		}
1836 	}
1837 
1838 	/* performance monitoring */
1839 	used_new = wil_vring_used_tx(vring);
1840 	if (wil_val_in_range(vring_idle_trsh,
1841 			     used_new, used_before_complete)) {
1842 		wil_dbg_txrx(wil, "Ring[%2d] idle %d -> %d\n",
1843 			     ringid, used_before_complete, used_new);
1844 		txdata->last_idle = get_cycles();
1845 	}
1846 
1847 	if (wil_vring_avail_tx(vring) > wil_vring_wmark_high(vring)) {
1848 		wil_dbg_txrx(wil, "netif_tx_wake : ring not full\n");
1849 		netif_tx_wake_all_queues(wil_to_ndev(wil));
1850 	}
1851 
1852 	return done;
1853 }
1854