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