xref: /linux/drivers/net/wireless/ath/wil6210/txrx.c (revision 0883c2c06fb5bcf5b9e008270827e63c09a88c1e)
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 			*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  * reverse_memcmp - Compare two areas of memory, in reverse order
554  * @cs: One area of memory
555  * @ct: Another area of memory
556  * @count: The size of the area.
557  *
558  * Cut'n'paste from original memcmp (see lib/string.c)
559  * with minimal modifications
560  */
561 static int reverse_memcmp(const void *cs, const void *ct, size_t count)
562 {
563 	const unsigned char *su1, *su2;
564 	int res = 0;
565 
566 	for (su1 = cs + count - 1, su2 = ct + count - 1; count > 0;
567 	     --su1, --su2, count--) {
568 		res = *su1 - *su2;
569 		if (res)
570 			break;
571 	}
572 	return res;
573 }
574 
575 static int wil_rx_crypto_check(struct wil6210_priv *wil, struct sk_buff *skb)
576 {
577 	struct vring_rx_desc *d = wil_skb_rxdesc(skb);
578 	int cid = wil_rxdesc_cid(d);
579 	int tid = wil_rxdesc_tid(d);
580 	int key_id = wil_rxdesc_key_id(d);
581 	int mc = wil_rxdesc_mcast(d);
582 	struct wil_sta_info *s = &wil->sta[cid];
583 	struct wil_tid_crypto_rx *c = mc ? &s->group_crypto_rx :
584 				      &s->tid_crypto_rx[tid];
585 	struct wil_tid_crypto_rx_single *cc = &c->key_id[key_id];
586 	const u8 *pn = (u8 *)&d->mac.pn_15_0;
587 
588 	if (!cc->key_set) {
589 		wil_err_ratelimited(wil,
590 				    "Key missing. CID %d TID %d MCast %d KEY_ID %d\n",
591 				    cid, tid, mc, key_id);
592 		return -EINVAL;
593 	}
594 
595 	if (reverse_memcmp(pn, cc->pn, IEEE80211_GCMP_PN_LEN) <= 0) {
596 		wil_err_ratelimited(wil,
597 				    "Replay attack. CID %d TID %d MCast %d KEY_ID %d PN %6phN last %6phN\n",
598 				    cid, tid, mc, key_id, pn, cc->pn);
599 		return -EINVAL;
600 	}
601 	memcpy(cc->pn, pn, IEEE80211_GCMP_PN_LEN);
602 
603 	return 0;
604 }
605 
606 /*
607  * Pass Rx packet to the netif. Update statistics.
608  * Called in softirq context (NAPI poll).
609  */
610 void wil_netif_rx_any(struct sk_buff *skb, struct net_device *ndev)
611 {
612 	gro_result_t rc = GRO_NORMAL;
613 	struct wil6210_priv *wil = ndev_to_wil(ndev);
614 	struct wireless_dev *wdev = wil_to_wdev(wil);
615 	unsigned int len = skb->len;
616 	struct vring_rx_desc *d = wil_skb_rxdesc(skb);
617 	int cid = wil_rxdesc_cid(d); /* always 0..7, no need to check */
618 	int security = wil_rxdesc_security(d);
619 	struct ethhdr *eth = (void *)skb->data;
620 	/* here looking for DA, not A1, thus Rxdesc's 'mcast' indication
621 	 * is not suitable, need to look at data
622 	 */
623 	int mcast = is_multicast_ether_addr(eth->h_dest);
624 	struct wil_net_stats *stats = &wil->sta[cid].stats;
625 	struct sk_buff *xmit_skb = NULL;
626 	static const char * const gro_res_str[] = {
627 		[GRO_MERGED]		= "GRO_MERGED",
628 		[GRO_MERGED_FREE]	= "GRO_MERGED_FREE",
629 		[GRO_HELD]		= "GRO_HELD",
630 		[GRO_NORMAL]		= "GRO_NORMAL",
631 		[GRO_DROP]		= "GRO_DROP",
632 	};
633 
634 	if (ndev->features & NETIF_F_RXHASH)
635 		/* fake L4 to ensure it won't be re-calculated later
636 		 * set hash to any non-zero value to activate rps
637 		 * mechanism, core will be chosen according
638 		 * to user-level rps configuration.
639 		 */
640 		skb_set_hash(skb, 1, PKT_HASH_TYPE_L4);
641 
642 	skb_orphan(skb);
643 
644 	if (security && (wil_rx_crypto_check(wil, skb) != 0)) {
645 		rc = GRO_DROP;
646 		dev_kfree_skb(skb);
647 		stats->rx_replay++;
648 		goto stats;
649 	}
650 
651 	if (wdev->iftype == NL80211_IFTYPE_AP && !wil->ap_isolate) {
652 		if (mcast) {
653 			/* send multicast frames both to higher layers in
654 			 * local net stack and back to the wireless medium
655 			 */
656 			xmit_skb = skb_copy(skb, GFP_ATOMIC);
657 		} else {
658 			int xmit_cid = wil_find_cid(wil, eth->h_dest);
659 
660 			if (xmit_cid >= 0) {
661 				/* The destination station is associated to
662 				 * this AP (in this VLAN), so send the frame
663 				 * directly to it and do not pass it to local
664 				 * net stack.
665 				 */
666 				xmit_skb = skb;
667 				skb = NULL;
668 			}
669 		}
670 	}
671 	if (xmit_skb) {
672 		/* Send to wireless media and increase priority by 256 to
673 		 * keep the received priority instead of reclassifying
674 		 * the frame (see cfg80211_classify8021d).
675 		 */
676 		xmit_skb->dev = ndev;
677 		xmit_skb->priority += 256;
678 		xmit_skb->protocol = htons(ETH_P_802_3);
679 		skb_reset_network_header(xmit_skb);
680 		skb_reset_mac_header(xmit_skb);
681 		wil_dbg_txrx(wil, "Rx -> Tx %d bytes\n", len);
682 		dev_queue_xmit(xmit_skb);
683 	}
684 
685 	if (skb) { /* deliver to local stack */
686 
687 		skb->protocol = eth_type_trans(skb, ndev);
688 		rc = napi_gro_receive(&wil->napi_rx, skb);
689 		wil_dbg_txrx(wil, "Rx complete %d bytes => %s\n",
690 			     len, gro_res_str[rc]);
691 	}
692 stats:
693 	/* statistics. rc set to GRO_NORMAL for AP bridging */
694 	if (unlikely(rc == GRO_DROP)) {
695 		ndev->stats.rx_dropped++;
696 		stats->rx_dropped++;
697 		wil_dbg_txrx(wil, "Rx drop %d bytes\n", len);
698 	} else {
699 		ndev->stats.rx_packets++;
700 		stats->rx_packets++;
701 		ndev->stats.rx_bytes += len;
702 		stats->rx_bytes += len;
703 		if (mcast)
704 			ndev->stats.multicast++;
705 	}
706 }
707 
708 /**
709  * Proceed all completed skb's from Rx VRING
710  *
711  * Safe to call from NAPI poll, i.e. softirq with interrupts enabled
712  */
713 void wil_rx_handle(struct wil6210_priv *wil, int *quota)
714 {
715 	struct net_device *ndev = wil_to_ndev(wil);
716 	struct vring *v = &wil->vring_rx;
717 	struct sk_buff *skb;
718 
719 	if (unlikely(!v->va)) {
720 		wil_err(wil, "Rx IRQ while Rx not yet initialized\n");
721 		return;
722 	}
723 	wil_dbg_txrx(wil, "%s()\n", __func__);
724 	while ((*quota > 0) && (NULL != (skb = wil_vring_reap_rx(wil, v)))) {
725 		(*quota)--;
726 
727 		if (wil->wdev->iftype == NL80211_IFTYPE_MONITOR) {
728 			skb->dev = ndev;
729 			skb_reset_mac_header(skb);
730 			skb->ip_summed = CHECKSUM_UNNECESSARY;
731 			skb->pkt_type = PACKET_OTHERHOST;
732 			skb->protocol = htons(ETH_P_802_2);
733 			wil_netif_rx_any(skb, ndev);
734 		} else {
735 			wil_rx_reorder(wil, skb);
736 		}
737 	}
738 	wil_rx_refill(wil, v->size);
739 }
740 
741 int wil_rx_init(struct wil6210_priv *wil, u16 size)
742 {
743 	struct vring *vring = &wil->vring_rx;
744 	int rc;
745 
746 	wil_dbg_misc(wil, "%s()\n", __func__);
747 
748 	if (vring->va) {
749 		wil_err(wil, "Rx ring already allocated\n");
750 		return -EINVAL;
751 	}
752 
753 	vring->size = size;
754 	rc = wil_vring_alloc(wil, vring);
755 	if (rc)
756 		return rc;
757 
758 	rc = wmi_rx_chain_add(wil, vring);
759 	if (rc)
760 		goto err_free;
761 
762 	rc = wil_rx_refill(wil, vring->size);
763 	if (rc)
764 		goto err_free;
765 
766 	return 0;
767  err_free:
768 	wil_vring_free(wil, vring, 0);
769 
770 	return rc;
771 }
772 
773 void wil_rx_fini(struct wil6210_priv *wil)
774 {
775 	struct vring *vring = &wil->vring_rx;
776 
777 	wil_dbg_misc(wil, "%s()\n", __func__);
778 
779 	if (vring->va)
780 		wil_vring_free(wil, vring, 0);
781 }
782 
783 static inline void wil_tx_data_init(struct vring_tx_data *txdata)
784 {
785 	spin_lock_bh(&txdata->lock);
786 	txdata->dot1x_open = 0;
787 	txdata->enabled = 0;
788 	txdata->idle = 0;
789 	txdata->last_idle = 0;
790 	txdata->begin = 0;
791 	txdata->agg_wsize = 0;
792 	txdata->agg_timeout = 0;
793 	txdata->agg_amsdu = 0;
794 	txdata->addba_in_progress = false;
795 	spin_unlock_bh(&txdata->lock);
796 }
797 
798 int wil_vring_init_tx(struct wil6210_priv *wil, int id, int size,
799 		      int cid, int tid)
800 {
801 	int rc;
802 	struct wmi_vring_cfg_cmd cmd = {
803 		.action = cpu_to_le32(WMI_VRING_CMD_ADD),
804 		.vring_cfg = {
805 			.tx_sw_ring = {
806 				.max_mpdu_size =
807 					cpu_to_le16(wil_mtu2macbuf(mtu_max)),
808 				.ring_size = cpu_to_le16(size),
809 			},
810 			.ringid = id,
811 			.cidxtid = mk_cidxtid(cid, tid),
812 			.encap_trans_type = WMI_VRING_ENC_TYPE_802_3,
813 			.mac_ctrl = 0,
814 			.to_resolution = 0,
815 			.agg_max_wsize = 0,
816 			.schd_params = {
817 				.priority = cpu_to_le16(0),
818 				.timeslot_us = cpu_to_le16(0xfff),
819 			},
820 		},
821 	};
822 	struct {
823 		struct wmi_cmd_hdr wmi;
824 		struct wmi_vring_cfg_done_event cmd;
825 	} __packed reply;
826 	struct vring *vring = &wil->vring_tx[id];
827 	struct vring_tx_data *txdata = &wil->vring_tx_data[id];
828 
829 	wil_dbg_misc(wil, "%s() max_mpdu_size %d\n", __func__,
830 		     cmd.vring_cfg.tx_sw_ring.max_mpdu_size);
831 	lockdep_assert_held(&wil->mutex);
832 
833 	if (vring->va) {
834 		wil_err(wil, "Tx ring [%d] already allocated\n", id);
835 		rc = -EINVAL;
836 		goto out;
837 	}
838 
839 	wil_tx_data_init(txdata);
840 	vring->size = size;
841 	rc = wil_vring_alloc(wil, vring);
842 	if (rc)
843 		goto out;
844 
845 	wil->vring2cid_tid[id][0] = cid;
846 	wil->vring2cid_tid[id][1] = tid;
847 
848 	cmd.vring_cfg.tx_sw_ring.ring_mem_base = cpu_to_le64(vring->pa);
849 
850 	if (!wil->privacy)
851 		txdata->dot1x_open = true;
852 	rc = wmi_call(wil, WMI_VRING_CFG_CMDID, &cmd, sizeof(cmd),
853 		      WMI_VRING_CFG_DONE_EVENTID, &reply, sizeof(reply), 100);
854 	if (rc)
855 		goto out_free;
856 
857 	if (reply.cmd.status != WMI_FW_STATUS_SUCCESS) {
858 		wil_err(wil, "Tx config failed, status 0x%02x\n",
859 			reply.cmd.status);
860 		rc = -EINVAL;
861 		goto out_free;
862 	}
863 	vring->hwtail = le32_to_cpu(reply.cmd.tx_vring_tail_ptr);
864 
865 	txdata->enabled = 1;
866 	if (txdata->dot1x_open && (agg_wsize >= 0))
867 		wil_addba_tx_request(wil, id, agg_wsize);
868 
869 	return 0;
870  out_free:
871 	spin_lock_bh(&txdata->lock);
872 	txdata->dot1x_open = false;
873 	txdata->enabled = 0;
874 	spin_unlock_bh(&txdata->lock);
875 	wil_vring_free(wil, vring, 1);
876 	wil->vring2cid_tid[id][0] = WIL6210_MAX_CID;
877 	wil->vring2cid_tid[id][1] = 0;
878 
879  out:
880 
881 	return rc;
882 }
883 
884 int wil_vring_init_bcast(struct wil6210_priv *wil, int id, int size)
885 {
886 	int rc;
887 	struct wmi_bcast_vring_cfg_cmd cmd = {
888 		.action = cpu_to_le32(WMI_VRING_CMD_ADD),
889 		.vring_cfg = {
890 			.tx_sw_ring = {
891 				.max_mpdu_size =
892 					cpu_to_le16(wil_mtu2macbuf(mtu_max)),
893 				.ring_size = cpu_to_le16(size),
894 			},
895 			.ringid = id,
896 			.encap_trans_type = WMI_VRING_ENC_TYPE_802_3,
897 		},
898 	};
899 	struct {
900 		struct wmi_cmd_hdr wmi;
901 		struct wmi_vring_cfg_done_event cmd;
902 	} __packed reply;
903 	struct vring *vring = &wil->vring_tx[id];
904 	struct vring_tx_data *txdata = &wil->vring_tx_data[id];
905 
906 	wil_dbg_misc(wil, "%s() max_mpdu_size %d\n", __func__,
907 		     cmd.vring_cfg.tx_sw_ring.max_mpdu_size);
908 	lockdep_assert_held(&wil->mutex);
909 
910 	if (vring->va) {
911 		wil_err(wil, "Tx ring [%d] already allocated\n", id);
912 		rc = -EINVAL;
913 		goto out;
914 	}
915 
916 	wil_tx_data_init(txdata);
917 	vring->size = size;
918 	rc = wil_vring_alloc(wil, vring);
919 	if (rc)
920 		goto out;
921 
922 	wil->vring2cid_tid[id][0] = WIL6210_MAX_CID; /* CID */
923 	wil->vring2cid_tid[id][1] = 0; /* TID */
924 
925 	cmd.vring_cfg.tx_sw_ring.ring_mem_base = cpu_to_le64(vring->pa);
926 
927 	if (!wil->privacy)
928 		txdata->dot1x_open = true;
929 	rc = wmi_call(wil, WMI_BCAST_VRING_CFG_CMDID, &cmd, sizeof(cmd),
930 		      WMI_VRING_CFG_DONE_EVENTID, &reply, sizeof(reply), 100);
931 	if (rc)
932 		goto out_free;
933 
934 	if (reply.cmd.status != WMI_FW_STATUS_SUCCESS) {
935 		wil_err(wil, "Tx config failed, status 0x%02x\n",
936 			reply.cmd.status);
937 		rc = -EINVAL;
938 		goto out_free;
939 	}
940 	vring->hwtail = le32_to_cpu(reply.cmd.tx_vring_tail_ptr);
941 
942 	txdata->enabled = 1;
943 
944 	return 0;
945  out_free:
946 	spin_lock_bh(&txdata->lock);
947 	txdata->enabled = 0;
948 	txdata->dot1x_open = false;
949 	spin_unlock_bh(&txdata->lock);
950 	wil_vring_free(wil, vring, 1);
951  out:
952 
953 	return rc;
954 }
955 
956 void wil_vring_fini_tx(struct wil6210_priv *wil, int id)
957 {
958 	struct vring *vring = &wil->vring_tx[id];
959 	struct vring_tx_data *txdata = &wil->vring_tx_data[id];
960 
961 	lockdep_assert_held(&wil->mutex);
962 
963 	if (!vring->va)
964 		return;
965 
966 	wil_dbg_misc(wil, "%s() id=%d\n", __func__, id);
967 
968 	spin_lock_bh(&txdata->lock);
969 	txdata->dot1x_open = false;
970 	txdata->enabled = 0; /* no Tx can be in progress or start anew */
971 	spin_unlock_bh(&txdata->lock);
972 	/* make sure NAPI won't touch this vring */
973 	if (test_bit(wil_status_napi_en, wil->status))
974 		napi_synchronize(&wil->napi_tx);
975 
976 	wil_vring_free(wil, vring, 1);
977 }
978 
979 static struct vring *wil_find_tx_ucast(struct wil6210_priv *wil,
980 				       struct sk_buff *skb)
981 {
982 	int i;
983 	struct ethhdr *eth = (void *)skb->data;
984 	int cid = wil_find_cid(wil, eth->h_dest);
985 
986 	if (cid < 0)
987 		return NULL;
988 
989 	/* TODO: fix for multiple TID */
990 	for (i = 0; i < ARRAY_SIZE(wil->vring2cid_tid); i++) {
991 		if (!wil->vring_tx_data[i].dot1x_open &&
992 		    (skb->protocol != cpu_to_be16(ETH_P_PAE)))
993 			continue;
994 		if (wil->vring2cid_tid[i][0] == cid) {
995 			struct vring *v = &wil->vring_tx[i];
996 			struct vring_tx_data *txdata = &wil->vring_tx_data[i];
997 
998 			wil_dbg_txrx(wil, "%s(%pM) -> [%d]\n",
999 				     __func__, eth->h_dest, i);
1000 			if (v->va && txdata->enabled) {
1001 				return v;
1002 			} else {
1003 				wil_dbg_txrx(wil, "vring[%d] not valid\n", i);
1004 				return NULL;
1005 			}
1006 		}
1007 	}
1008 
1009 	return NULL;
1010 }
1011 
1012 static int wil_tx_vring(struct wil6210_priv *wil, struct vring *vring,
1013 			struct sk_buff *skb);
1014 
1015 static struct vring *wil_find_tx_vring_sta(struct wil6210_priv *wil,
1016 					   struct sk_buff *skb)
1017 {
1018 	struct vring *v;
1019 	int i;
1020 	u8 cid;
1021 	struct vring_tx_data *txdata;
1022 
1023 	/* In the STA mode, it is expected to have only 1 VRING
1024 	 * for the AP we connected to.
1025 	 * find 1-st vring eligible for this skb and use it.
1026 	 */
1027 	for (i = 0; i < WIL6210_MAX_TX_RINGS; i++) {
1028 		v = &wil->vring_tx[i];
1029 		txdata = &wil->vring_tx_data[i];
1030 		if (!v->va || !txdata->enabled)
1031 			continue;
1032 
1033 		cid = wil->vring2cid_tid[i][0];
1034 		if (cid >= WIL6210_MAX_CID) /* skip BCAST */
1035 			continue;
1036 
1037 		if (!wil->vring_tx_data[i].dot1x_open &&
1038 		    (skb->protocol != cpu_to_be16(ETH_P_PAE)))
1039 			continue;
1040 
1041 		wil_dbg_txrx(wil, "Tx -> ring %d\n", i);
1042 
1043 		return v;
1044 	}
1045 
1046 	wil_dbg_txrx(wil, "Tx while no vrings active?\n");
1047 
1048 	return NULL;
1049 }
1050 
1051 /* Use one of 2 strategies:
1052  *
1053  * 1. New (real broadcast):
1054  *    use dedicated broadcast vring
1055  * 2. Old (pseudo-DMS):
1056  *    Find 1-st vring and return it;
1057  *    duplicate skb and send it to other active vrings;
1058  *    in all cases override dest address to unicast peer's address
1059  * Use old strategy when new is not supported yet:
1060  *  - for PBSS
1061  */
1062 static struct vring *wil_find_tx_bcast_1(struct wil6210_priv *wil,
1063 					 struct sk_buff *skb)
1064 {
1065 	struct vring *v;
1066 	struct vring_tx_data *txdata;
1067 	int i = wil->bcast_vring;
1068 
1069 	if (i < 0)
1070 		return NULL;
1071 	v = &wil->vring_tx[i];
1072 	txdata = &wil->vring_tx_data[i];
1073 	if (!v->va || !txdata->enabled)
1074 		return NULL;
1075 	if (!wil->vring_tx_data[i].dot1x_open &&
1076 	    (skb->protocol != cpu_to_be16(ETH_P_PAE)))
1077 		return NULL;
1078 
1079 	return v;
1080 }
1081 
1082 static void wil_set_da_for_vring(struct wil6210_priv *wil,
1083 				 struct sk_buff *skb, int vring_index)
1084 {
1085 	struct ethhdr *eth = (void *)skb->data;
1086 	int cid = wil->vring2cid_tid[vring_index][0];
1087 
1088 	ether_addr_copy(eth->h_dest, wil->sta[cid].addr);
1089 }
1090 
1091 static struct vring *wil_find_tx_bcast_2(struct wil6210_priv *wil,
1092 					 struct sk_buff *skb)
1093 {
1094 	struct vring *v, *v2;
1095 	struct sk_buff *skb2;
1096 	int i;
1097 	u8 cid;
1098 	struct ethhdr *eth = (void *)skb->data;
1099 	char *src = eth->h_source;
1100 	struct vring_tx_data *txdata;
1101 
1102 	/* find 1-st vring eligible for data */
1103 	for (i = 0; i < WIL6210_MAX_TX_RINGS; i++) {
1104 		v = &wil->vring_tx[i];
1105 		txdata = &wil->vring_tx_data[i];
1106 		if (!v->va || !txdata->enabled)
1107 			continue;
1108 
1109 		cid = wil->vring2cid_tid[i][0];
1110 		if (cid >= WIL6210_MAX_CID) /* skip BCAST */
1111 			continue;
1112 		if (!wil->vring_tx_data[i].dot1x_open &&
1113 		    (skb->protocol != cpu_to_be16(ETH_P_PAE)))
1114 			continue;
1115 
1116 		/* don't Tx back to source when re-routing Rx->Tx at the AP */
1117 		if (0 == memcmp(wil->sta[cid].addr, src, ETH_ALEN))
1118 			continue;
1119 
1120 		goto found;
1121 	}
1122 
1123 	wil_dbg_txrx(wil, "Tx while no vrings active?\n");
1124 
1125 	return NULL;
1126 
1127 found:
1128 	wil_dbg_txrx(wil, "BCAST -> ring %d\n", i);
1129 	wil_set_da_for_vring(wil, skb, i);
1130 
1131 	/* find other active vrings and duplicate skb for each */
1132 	for (i++; i < WIL6210_MAX_TX_RINGS; i++) {
1133 		v2 = &wil->vring_tx[i];
1134 		if (!v2->va)
1135 			continue;
1136 		cid = wil->vring2cid_tid[i][0];
1137 		if (cid >= WIL6210_MAX_CID) /* skip BCAST */
1138 			continue;
1139 		if (!wil->vring_tx_data[i].dot1x_open &&
1140 		    (skb->protocol != cpu_to_be16(ETH_P_PAE)))
1141 			continue;
1142 
1143 		if (0 == memcmp(wil->sta[cid].addr, src, ETH_ALEN))
1144 			continue;
1145 
1146 		skb2 = skb_copy(skb, GFP_ATOMIC);
1147 		if (skb2) {
1148 			wil_dbg_txrx(wil, "BCAST DUP -> ring %d\n", i);
1149 			wil_set_da_for_vring(wil, skb2, i);
1150 			wil_tx_vring(wil, v2, skb2);
1151 		} else {
1152 			wil_err(wil, "skb_copy failed\n");
1153 		}
1154 	}
1155 
1156 	return v;
1157 }
1158 
1159 static struct vring *wil_find_tx_bcast(struct wil6210_priv *wil,
1160 				       struct sk_buff *skb)
1161 {
1162 	struct wireless_dev *wdev = wil->wdev;
1163 
1164 	if (wdev->iftype != NL80211_IFTYPE_AP)
1165 		return wil_find_tx_bcast_2(wil, skb);
1166 
1167 	return wil_find_tx_bcast_1(wil, skb);
1168 }
1169 
1170 static int wil_tx_desc_map(struct vring_tx_desc *d, dma_addr_t pa, u32 len,
1171 			   int vring_index)
1172 {
1173 	wil_desc_addr_set(&d->dma.addr, pa);
1174 	d->dma.ip_length = 0;
1175 	/* 0..6: mac_length; 7:ip_version 0-IP6 1-IP4*/
1176 	d->dma.b11 = 0/*14 | BIT(7)*/;
1177 	d->dma.error = 0;
1178 	d->dma.status = 0; /* BIT(0) should be 0 for HW_OWNED */
1179 	d->dma.length = cpu_to_le16((u16)len);
1180 	d->dma.d0 = (vring_index << DMA_CFG_DESC_TX_0_QID_POS);
1181 	d->mac.d[0] = 0;
1182 	d->mac.d[1] = 0;
1183 	d->mac.d[2] = 0;
1184 	d->mac.ucode_cmd = 0;
1185 	/* translation type:  0 - bypass; 1 - 802.3; 2 - native wifi */
1186 	d->mac.d[2] = BIT(MAC_CFG_DESC_TX_2_SNAP_HDR_INSERTION_EN_POS) |
1187 		      (1 << MAC_CFG_DESC_TX_2_L2_TRANSLATION_TYPE_POS);
1188 
1189 	return 0;
1190 }
1191 
1192 static inline
1193 void wil_tx_desc_set_nr_frags(struct vring_tx_desc *d, int nr_frags)
1194 {
1195 	d->mac.d[2] |= (nr_frags << MAC_CFG_DESC_TX_2_NUM_OF_DESCRIPTORS_POS);
1196 }
1197 
1198 /**
1199  * Sets the descriptor @d up for csum and/or TSO offloading. The corresponding
1200  * @skb is used to obtain the protocol and headers length.
1201  * @tso_desc_type is a descriptor type for TSO: 0 - a header, 1 - first data,
1202  * 2 - middle, 3 - last descriptor.
1203  */
1204 
1205 static void wil_tx_desc_offload_setup_tso(struct vring_tx_desc *d,
1206 					  struct sk_buff *skb,
1207 					  int tso_desc_type, bool is_ipv4,
1208 					  int tcp_hdr_len, int skb_net_hdr_len)
1209 {
1210 	d->dma.b11 = ETH_HLEN; /* MAC header length */
1211 	d->dma.b11 |= is_ipv4 << DMA_CFG_DESC_TX_OFFLOAD_CFG_L3T_IPV4_POS;
1212 
1213 	d->dma.d0 |= (2 << DMA_CFG_DESC_TX_0_L4_TYPE_POS);
1214 	/* L4 header len: TCP header length */
1215 	d->dma.d0 |= (tcp_hdr_len & DMA_CFG_DESC_TX_0_L4_LENGTH_MSK);
1216 
1217 	/* Setup TSO: bit and desc type */
1218 	d->dma.d0 |= (BIT(DMA_CFG_DESC_TX_0_TCP_SEG_EN_POS)) |
1219 		(tso_desc_type << DMA_CFG_DESC_TX_0_SEGMENT_BUF_DETAILS_POS);
1220 	d->dma.d0 |= (is_ipv4 << DMA_CFG_DESC_TX_0_IPV4_CHECKSUM_EN_POS);
1221 
1222 	d->dma.ip_length = skb_net_hdr_len;
1223 	/* Enable TCP/UDP checksum */
1224 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_TCP_UDP_CHECKSUM_EN_POS);
1225 	/* Calculate pseudo-header */
1226 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_PSEUDO_HEADER_CALC_EN_POS);
1227 }
1228 
1229 /**
1230  * Sets the descriptor @d up for csum. The corresponding
1231  * @skb is used to obtain the protocol and headers length.
1232  * Returns the protocol: 0 - not TCP, 1 - TCPv4, 2 - TCPv6.
1233  * Note, if d==NULL, the function only returns the protocol result.
1234  *
1235  * It is very similar to previous wil_tx_desc_offload_setup_tso. This
1236  * is "if unrolling" to optimize the critical path.
1237  */
1238 
1239 static int wil_tx_desc_offload_setup(struct vring_tx_desc *d,
1240 				     struct sk_buff *skb){
1241 	int protocol;
1242 
1243 	if (skb->ip_summed != CHECKSUM_PARTIAL)
1244 		return 0;
1245 
1246 	d->dma.b11 = ETH_HLEN; /* MAC header length */
1247 
1248 	switch (skb->protocol) {
1249 	case cpu_to_be16(ETH_P_IP):
1250 		protocol = ip_hdr(skb)->protocol;
1251 		d->dma.b11 |= BIT(DMA_CFG_DESC_TX_OFFLOAD_CFG_L3T_IPV4_POS);
1252 		break;
1253 	case cpu_to_be16(ETH_P_IPV6):
1254 		protocol = ipv6_hdr(skb)->nexthdr;
1255 		break;
1256 	default:
1257 		return -EINVAL;
1258 	}
1259 
1260 	switch (protocol) {
1261 	case IPPROTO_TCP:
1262 		d->dma.d0 |= (2 << DMA_CFG_DESC_TX_0_L4_TYPE_POS);
1263 		/* L4 header len: TCP header length */
1264 		d->dma.d0 |=
1265 		(tcp_hdrlen(skb) & DMA_CFG_DESC_TX_0_L4_LENGTH_MSK);
1266 		break;
1267 	case IPPROTO_UDP:
1268 		/* L4 header len: UDP header length */
1269 		d->dma.d0 |=
1270 		(sizeof(struct udphdr) & DMA_CFG_DESC_TX_0_L4_LENGTH_MSK);
1271 		break;
1272 	default:
1273 		return -EINVAL;
1274 	}
1275 
1276 	d->dma.ip_length = skb_network_header_len(skb);
1277 	/* Enable TCP/UDP checksum */
1278 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_TCP_UDP_CHECKSUM_EN_POS);
1279 	/* Calculate pseudo-header */
1280 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_PSEUDO_HEADER_CALC_EN_POS);
1281 
1282 	return 0;
1283 }
1284 
1285 static inline void wil_tx_last_desc(struct vring_tx_desc *d)
1286 {
1287 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_CMD_EOP_POS) |
1288 	      BIT(DMA_CFG_DESC_TX_0_CMD_MARK_WB_POS) |
1289 	      BIT(DMA_CFG_DESC_TX_0_CMD_DMA_IT_POS);
1290 }
1291 
1292 static inline void wil_set_tx_desc_last_tso(volatile struct vring_tx_desc *d)
1293 {
1294 	d->dma.d0 |= wil_tso_type_lst <<
1295 		  DMA_CFG_DESC_TX_0_SEGMENT_BUF_DETAILS_POS;
1296 }
1297 
1298 static int __wil_tx_vring_tso(struct wil6210_priv *wil, struct vring *vring,
1299 			      struct sk_buff *skb)
1300 {
1301 	struct device *dev = wil_to_dev(wil);
1302 
1303 	/* point to descriptors in shared memory */
1304 	volatile struct vring_tx_desc *_desc = NULL, *_hdr_desc,
1305 				      *_first_desc = NULL;
1306 
1307 	/* pointers to shadow descriptors */
1308 	struct vring_tx_desc desc_mem, hdr_desc_mem, first_desc_mem,
1309 			     *d = &hdr_desc_mem, *hdr_desc = &hdr_desc_mem,
1310 			     *first_desc = &first_desc_mem;
1311 
1312 	/* pointer to shadow descriptors' context */
1313 	struct wil_ctx *hdr_ctx, *first_ctx = NULL;
1314 
1315 	int descs_used = 0; /* total number of used descriptors */
1316 	int sg_desc_cnt = 0; /* number of descriptors for current mss*/
1317 
1318 	u32 swhead = vring->swhead;
1319 	int used, avail = wil_vring_avail_tx(vring);
1320 	int nr_frags = skb_shinfo(skb)->nr_frags;
1321 	int min_desc_required = nr_frags + 1;
1322 	int mss = skb_shinfo(skb)->gso_size;	/* payload size w/o headers */
1323 	int f, len, hdrlen, headlen;
1324 	int vring_index = vring - wil->vring_tx;
1325 	struct vring_tx_data *txdata = &wil->vring_tx_data[vring_index];
1326 	uint i = swhead;
1327 	dma_addr_t pa;
1328 	const skb_frag_t *frag = NULL;
1329 	int rem_data = mss;
1330 	int lenmss;
1331 	int hdr_compensation_need = true;
1332 	int desc_tso_type = wil_tso_type_first;
1333 	bool is_ipv4;
1334 	int tcp_hdr_len;
1335 	int skb_net_hdr_len;
1336 	int gso_type;
1337 	int rc = -EINVAL;
1338 
1339 	wil_dbg_txrx(wil, "%s() %d bytes to vring %d\n",
1340 		     __func__, skb->len, vring_index);
1341 
1342 	if (unlikely(!txdata->enabled))
1343 		return -EINVAL;
1344 
1345 	/* A typical page 4K is 3-4 payloads, we assume each fragment
1346 	 * is a full payload, that's how min_desc_required has been
1347 	 * calculated. In real we might need more or less descriptors,
1348 	 * this is the initial check only.
1349 	 */
1350 	if (unlikely(avail < min_desc_required)) {
1351 		wil_err_ratelimited(wil,
1352 				    "TSO: Tx ring[%2d] full. No space for %d fragments\n",
1353 				    vring_index, min_desc_required);
1354 		return -ENOMEM;
1355 	}
1356 
1357 	/* Header Length = MAC header len + IP header len + TCP header len*/
1358 	hdrlen = ETH_HLEN +
1359 		(int)skb_network_header_len(skb) +
1360 		tcp_hdrlen(skb);
1361 
1362 	gso_type = skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV6 | SKB_GSO_TCPV4);
1363 	switch (gso_type) {
1364 	case SKB_GSO_TCPV4:
1365 		/* TCP v4, zero out the IP length and IPv4 checksum fields
1366 		 * as required by the offloading doc
1367 		 */
1368 		ip_hdr(skb)->tot_len = 0;
1369 		ip_hdr(skb)->check = 0;
1370 		is_ipv4 = true;
1371 		break;
1372 	case SKB_GSO_TCPV6:
1373 		/* TCP v6, zero out the payload length */
1374 		ipv6_hdr(skb)->payload_len = 0;
1375 		is_ipv4 = false;
1376 		break;
1377 	default:
1378 		/* other than TCPv4 or TCPv6 types are not supported for TSO.
1379 		 * It is also illegal for both to be set simultaneously
1380 		 */
1381 		return -EINVAL;
1382 	}
1383 
1384 	if (skb->ip_summed != CHECKSUM_PARTIAL)
1385 		return -EINVAL;
1386 
1387 	/* tcp header length and skb network header length are fixed for all
1388 	 * packet's descriptors - read then once here
1389 	 */
1390 	tcp_hdr_len = tcp_hdrlen(skb);
1391 	skb_net_hdr_len = skb_network_header_len(skb);
1392 
1393 	_hdr_desc = &vring->va[i].tx;
1394 
1395 	pa = dma_map_single(dev, skb->data, hdrlen, DMA_TO_DEVICE);
1396 	if (unlikely(dma_mapping_error(dev, pa))) {
1397 		wil_err(wil, "TSO: Skb head DMA map error\n");
1398 		goto err_exit;
1399 	}
1400 
1401 	wil_tx_desc_map(hdr_desc, pa, hdrlen, vring_index);
1402 	wil_tx_desc_offload_setup_tso(hdr_desc, skb, wil_tso_type_hdr, is_ipv4,
1403 				      tcp_hdr_len, skb_net_hdr_len);
1404 	wil_tx_last_desc(hdr_desc);
1405 
1406 	vring->ctx[i].mapped_as = wil_mapped_as_single;
1407 	hdr_ctx = &vring->ctx[i];
1408 
1409 	descs_used++;
1410 	headlen = skb_headlen(skb) - hdrlen;
1411 
1412 	for (f = headlen ? -1 : 0; f < nr_frags; f++)  {
1413 		if (headlen) {
1414 			len = headlen;
1415 			wil_dbg_txrx(wil, "TSO: process skb head, len %u\n",
1416 				     len);
1417 		} else {
1418 			frag = &skb_shinfo(skb)->frags[f];
1419 			len = frag->size;
1420 			wil_dbg_txrx(wil, "TSO: frag[%d]: len %u\n", f, len);
1421 		}
1422 
1423 		while (len) {
1424 			wil_dbg_txrx(wil,
1425 				     "TSO: len %d, rem_data %d, descs_used %d\n",
1426 				     len, rem_data, descs_used);
1427 
1428 			if (descs_used == avail)  {
1429 				wil_err_ratelimited(wil, "TSO: ring overflow\n");
1430 				rc = -ENOMEM;
1431 				goto mem_error;
1432 			}
1433 
1434 			lenmss = min_t(int, rem_data, len);
1435 			i = (swhead + descs_used) % vring->size;
1436 			wil_dbg_txrx(wil, "TSO: lenmss %d, i %d\n", lenmss, i);
1437 
1438 			if (!headlen) {
1439 				pa = skb_frag_dma_map(dev, frag,
1440 						      frag->size - len, lenmss,
1441 						      DMA_TO_DEVICE);
1442 				vring->ctx[i].mapped_as = wil_mapped_as_page;
1443 			} else {
1444 				pa = dma_map_single(dev,
1445 						    skb->data +
1446 						    skb_headlen(skb) - headlen,
1447 						    lenmss,
1448 						    DMA_TO_DEVICE);
1449 				vring->ctx[i].mapped_as = wil_mapped_as_single;
1450 				headlen -= lenmss;
1451 			}
1452 
1453 			if (unlikely(dma_mapping_error(dev, pa))) {
1454 				wil_err(wil, "TSO: DMA map page error\n");
1455 				goto mem_error;
1456 			}
1457 
1458 			_desc = &vring->va[i].tx;
1459 
1460 			if (!_first_desc) {
1461 				_first_desc = _desc;
1462 				first_ctx = &vring->ctx[i];
1463 				d = first_desc;
1464 			} else {
1465 				d = &desc_mem;
1466 			}
1467 
1468 			wil_tx_desc_map(d, pa, lenmss, vring_index);
1469 			wil_tx_desc_offload_setup_tso(d, skb, desc_tso_type,
1470 						      is_ipv4, tcp_hdr_len,
1471 						      skb_net_hdr_len);
1472 
1473 			/* use tso_type_first only once */
1474 			desc_tso_type = wil_tso_type_mid;
1475 
1476 			descs_used++;  /* desc used so far */
1477 			sg_desc_cnt++; /* desc used for this segment */
1478 			len -= lenmss;
1479 			rem_data -= lenmss;
1480 
1481 			wil_dbg_txrx(wil,
1482 				     "TSO: len %d, rem_data %d, descs_used %d, sg_desc_cnt %d,\n",
1483 				     len, rem_data, descs_used, sg_desc_cnt);
1484 
1485 			/* Close the segment if reached mss size or last frag*/
1486 			if (rem_data == 0 || (f == nr_frags - 1 && len == 0)) {
1487 				if (hdr_compensation_need) {
1488 					/* first segment include hdr desc for
1489 					 * release
1490 					 */
1491 					hdr_ctx->nr_frags = sg_desc_cnt;
1492 					wil_tx_desc_set_nr_frags(first_desc,
1493 								 sg_desc_cnt +
1494 								 1);
1495 					hdr_compensation_need = false;
1496 				} else {
1497 					wil_tx_desc_set_nr_frags(first_desc,
1498 								 sg_desc_cnt);
1499 				}
1500 				first_ctx->nr_frags = sg_desc_cnt - 1;
1501 
1502 				wil_tx_last_desc(d);
1503 
1504 				/* first descriptor may also be the last
1505 				 * for this mss - make sure not to copy
1506 				 * it twice
1507 				 */
1508 				if (first_desc != d)
1509 					*_first_desc = *first_desc;
1510 
1511 				/*last descriptor will be copied at the end
1512 				 * of this TS processing
1513 				 */
1514 				if (f < nr_frags - 1 || len > 0)
1515 					*_desc = *d;
1516 
1517 				rem_data = mss;
1518 				_first_desc = NULL;
1519 				sg_desc_cnt = 0;
1520 			} else if (first_desc != d) /* update mid descriptor */
1521 					*_desc = *d;
1522 		}
1523 	}
1524 
1525 	/* first descriptor may also be the last.
1526 	 * in this case d pointer is invalid
1527 	 */
1528 	if (_first_desc == _desc)
1529 		d = first_desc;
1530 
1531 	/* Last data descriptor */
1532 	wil_set_tx_desc_last_tso(d);
1533 	*_desc = *d;
1534 
1535 	/* Fill the total number of descriptors in first desc (hdr)*/
1536 	wil_tx_desc_set_nr_frags(hdr_desc, descs_used);
1537 	*_hdr_desc = *hdr_desc;
1538 
1539 	/* hold reference to skb
1540 	 * to prevent skb release before accounting
1541 	 * in case of immediate "tx done"
1542 	 */
1543 	vring->ctx[i].skb = skb_get(skb);
1544 
1545 	/* performance monitoring */
1546 	used = wil_vring_used_tx(vring);
1547 	if (wil_val_in_range(vring_idle_trsh,
1548 			     used, used + descs_used)) {
1549 		txdata->idle += get_cycles() - txdata->last_idle;
1550 		wil_dbg_txrx(wil,  "Ring[%2d] not idle %d -> %d\n",
1551 			     vring_index, used, used + descs_used);
1552 	}
1553 
1554 	/* advance swhead */
1555 	wil_vring_advance_head(vring, descs_used);
1556 	wil_dbg_txrx(wil, "TSO: Tx swhead %d -> %d\n", swhead, vring->swhead);
1557 
1558 	/* make sure all writes to descriptors (shared memory) are done before
1559 	 * committing them to HW
1560 	 */
1561 	wmb();
1562 
1563 	wil_w(wil, vring->hwtail, vring->swhead);
1564 	return 0;
1565 
1566 mem_error:
1567 	while (descs_used > 0) {
1568 		struct wil_ctx *ctx;
1569 
1570 		i = (swhead + descs_used) % vring->size;
1571 		d = (struct vring_tx_desc *)&vring->va[i].tx;
1572 		_desc = &vring->va[i].tx;
1573 		*d = *_desc;
1574 		_desc->dma.status = TX_DMA_STATUS_DU;
1575 		ctx = &vring->ctx[i];
1576 		wil_txdesc_unmap(dev, d, ctx);
1577 		memset(ctx, 0, sizeof(*ctx));
1578 		descs_used--;
1579 	}
1580 err_exit:
1581 	return rc;
1582 }
1583 
1584 static int __wil_tx_vring(struct wil6210_priv *wil, struct vring *vring,
1585 			  struct sk_buff *skb)
1586 {
1587 	struct device *dev = wil_to_dev(wil);
1588 	struct vring_tx_desc dd, *d = &dd;
1589 	volatile struct vring_tx_desc *_d;
1590 	u32 swhead = vring->swhead;
1591 	int avail = wil_vring_avail_tx(vring);
1592 	int nr_frags = skb_shinfo(skb)->nr_frags;
1593 	uint f = 0;
1594 	int vring_index = vring - wil->vring_tx;
1595 	struct vring_tx_data *txdata = &wil->vring_tx_data[vring_index];
1596 	uint i = swhead;
1597 	dma_addr_t pa;
1598 	int used;
1599 	bool mcast = (vring_index == wil->bcast_vring);
1600 	uint len = skb_headlen(skb);
1601 
1602 	wil_dbg_txrx(wil, "%s() %d bytes to vring %d\n",
1603 		     __func__, skb->len, vring_index);
1604 
1605 	if (unlikely(!txdata->enabled))
1606 		return -EINVAL;
1607 
1608 	if (unlikely(avail < 1 + nr_frags)) {
1609 		wil_err_ratelimited(wil,
1610 				    "Tx ring[%2d] full. No space for %d fragments\n",
1611 				    vring_index, 1 + nr_frags);
1612 		return -ENOMEM;
1613 	}
1614 	_d = &vring->va[i].tx;
1615 
1616 	pa = dma_map_single(dev, skb->data, skb_headlen(skb), DMA_TO_DEVICE);
1617 
1618 	wil_dbg_txrx(wil, "Tx[%2d] skb %d bytes 0x%p -> %pad\n", vring_index,
1619 		     skb_headlen(skb), skb->data, &pa);
1620 	wil_hex_dump_txrx("Tx ", DUMP_PREFIX_OFFSET, 16, 1,
1621 			  skb->data, skb_headlen(skb), false);
1622 
1623 	if (unlikely(dma_mapping_error(dev, pa)))
1624 		return -EINVAL;
1625 	vring->ctx[i].mapped_as = wil_mapped_as_single;
1626 	/* 1-st segment */
1627 	wil_tx_desc_map(d, pa, len, vring_index);
1628 	if (unlikely(mcast)) {
1629 		d->mac.d[0] |= BIT(MAC_CFG_DESC_TX_0_MCS_EN_POS); /* MCS 0 */
1630 		if (unlikely(len > WIL_BCAST_MCS0_LIMIT)) /* set MCS 1 */
1631 			d->mac.d[0] |= (1 << MAC_CFG_DESC_TX_0_MCS_INDEX_POS);
1632 	}
1633 	/* Process TCP/UDP checksum offloading */
1634 	if (unlikely(wil_tx_desc_offload_setup(d, skb))) {
1635 		wil_err(wil, "Tx[%2d] Failed to set cksum, drop packet\n",
1636 			vring_index);
1637 		goto dma_error;
1638 	}
1639 
1640 	vring->ctx[i].nr_frags = nr_frags;
1641 	wil_tx_desc_set_nr_frags(d, nr_frags + 1);
1642 
1643 	/* middle segments */
1644 	for (; f < nr_frags; f++) {
1645 		const struct skb_frag_struct *frag =
1646 				&skb_shinfo(skb)->frags[f];
1647 		int len = skb_frag_size(frag);
1648 
1649 		*_d = *d;
1650 		wil_dbg_txrx(wil, "Tx[%2d] desc[%4d]\n", vring_index, i);
1651 		wil_hex_dump_txrx("TxD ", DUMP_PREFIX_NONE, 32, 4,
1652 				  (const void *)d, sizeof(*d), false);
1653 		i = (swhead + f + 1) % vring->size;
1654 		_d = &vring->va[i].tx;
1655 		pa = skb_frag_dma_map(dev, frag, 0, skb_frag_size(frag),
1656 				      DMA_TO_DEVICE);
1657 		if (unlikely(dma_mapping_error(dev, pa))) {
1658 			wil_err(wil, "Tx[%2d] failed to map fragment\n",
1659 				vring_index);
1660 			goto dma_error;
1661 		}
1662 		vring->ctx[i].mapped_as = wil_mapped_as_page;
1663 		wil_tx_desc_map(d, pa, len, vring_index);
1664 		/* no need to check return code -
1665 		 * if it succeeded for 1-st descriptor,
1666 		 * it will succeed here too
1667 		 */
1668 		wil_tx_desc_offload_setup(d, skb);
1669 	}
1670 	/* for the last seg only */
1671 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_CMD_EOP_POS);
1672 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_CMD_MARK_WB_POS);
1673 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_CMD_DMA_IT_POS);
1674 	*_d = *d;
1675 	wil_dbg_txrx(wil, "Tx[%2d] desc[%4d]\n", vring_index, i);
1676 	wil_hex_dump_txrx("TxD ", DUMP_PREFIX_NONE, 32, 4,
1677 			  (const void *)d, sizeof(*d), false);
1678 
1679 	/* hold reference to skb
1680 	 * to prevent skb release before accounting
1681 	 * in case of immediate "tx done"
1682 	 */
1683 	vring->ctx[i].skb = skb_get(skb);
1684 
1685 	/* performance monitoring */
1686 	used = wil_vring_used_tx(vring);
1687 	if (wil_val_in_range(vring_idle_trsh,
1688 			     used, used + nr_frags + 1)) {
1689 		txdata->idle += get_cycles() - txdata->last_idle;
1690 		wil_dbg_txrx(wil,  "Ring[%2d] not idle %d -> %d\n",
1691 			     vring_index, used, used + nr_frags + 1);
1692 	}
1693 
1694 	/* advance swhead */
1695 	wil_vring_advance_head(vring, nr_frags + 1);
1696 	wil_dbg_txrx(wil, "Tx[%2d] swhead %d -> %d\n", vring_index, swhead,
1697 		     vring->swhead);
1698 	trace_wil6210_tx(vring_index, swhead, skb->len, nr_frags);
1699 
1700 	/* make sure all writes to descriptors (shared memory) are done before
1701 	 * committing them to HW
1702 	 */
1703 	wmb();
1704 
1705 	wil_w(wil, vring->hwtail, vring->swhead);
1706 
1707 	return 0;
1708  dma_error:
1709 	/* unmap what we have mapped */
1710 	nr_frags = f + 1; /* frags mapped + one for skb head */
1711 	for (f = 0; f < nr_frags; f++) {
1712 		struct wil_ctx *ctx;
1713 
1714 		i = (swhead + f) % vring->size;
1715 		ctx = &vring->ctx[i];
1716 		_d = &vring->va[i].tx;
1717 		*d = *_d;
1718 		_d->dma.status = TX_DMA_STATUS_DU;
1719 		wil_txdesc_unmap(dev, d, ctx);
1720 
1721 		memset(ctx, 0, sizeof(*ctx));
1722 	}
1723 
1724 	return -EINVAL;
1725 }
1726 
1727 static int wil_tx_vring(struct wil6210_priv *wil, struct vring *vring,
1728 			struct sk_buff *skb)
1729 {
1730 	int vring_index = vring - wil->vring_tx;
1731 	struct vring_tx_data *txdata = &wil->vring_tx_data[vring_index];
1732 	int rc;
1733 
1734 	spin_lock(&txdata->lock);
1735 
1736 	rc = (skb_is_gso(skb) ? __wil_tx_vring_tso : __wil_tx_vring)
1737 	     (wil, vring, skb);
1738 
1739 	spin_unlock(&txdata->lock);
1740 
1741 	return rc;
1742 }
1743 
1744 netdev_tx_t wil_start_xmit(struct sk_buff *skb, struct net_device *ndev)
1745 {
1746 	struct wil6210_priv *wil = ndev_to_wil(ndev);
1747 	struct ethhdr *eth = (void *)skb->data;
1748 	bool bcast = is_multicast_ether_addr(eth->h_dest);
1749 	struct vring *vring;
1750 	static bool pr_once_fw;
1751 	int rc;
1752 
1753 	wil_dbg_txrx(wil, "%s()\n", __func__);
1754 	if (unlikely(!test_bit(wil_status_fwready, wil->status))) {
1755 		if (!pr_once_fw) {
1756 			wil_err(wil, "FW not ready\n");
1757 			pr_once_fw = true;
1758 		}
1759 		goto drop;
1760 	}
1761 	if (unlikely(!test_bit(wil_status_fwconnected, wil->status))) {
1762 		wil_dbg_ratelimited(wil, "FW not connected, packet dropped\n");
1763 		goto drop;
1764 	}
1765 	if (unlikely(wil->wdev->iftype == NL80211_IFTYPE_MONITOR)) {
1766 		wil_err(wil, "Xmit in monitor mode not supported\n");
1767 		goto drop;
1768 	}
1769 	pr_once_fw = false;
1770 
1771 	/* find vring */
1772 	if (wil->wdev->iftype == NL80211_IFTYPE_STATION) {
1773 		/* in STA mode (ESS), all to same VRING */
1774 		vring = wil_find_tx_vring_sta(wil, skb);
1775 	} else { /* direct communication, find matching VRING */
1776 		vring = bcast ? wil_find_tx_bcast(wil, skb) :
1777 				wil_find_tx_ucast(wil, skb);
1778 	}
1779 	if (unlikely(!vring)) {
1780 		wil_dbg_txrx(wil, "No Tx VRING found for %pM\n", eth->h_dest);
1781 		goto drop;
1782 	}
1783 	/* set up vring entry */
1784 	rc = wil_tx_vring(wil, vring, skb);
1785 
1786 	/* do we still have enough room in the vring? */
1787 	if (unlikely(wil_vring_avail_tx(vring) < wil_vring_wmark_low(vring))) {
1788 		netif_tx_stop_all_queues(wil_to_ndev(wil));
1789 		wil_dbg_txrx(wil, "netif_tx_stop : ring full\n");
1790 	}
1791 
1792 	switch (rc) {
1793 	case 0:
1794 		/* statistics will be updated on the tx_complete */
1795 		dev_kfree_skb_any(skb);
1796 		return NETDEV_TX_OK;
1797 	case -ENOMEM:
1798 		return NETDEV_TX_BUSY;
1799 	default:
1800 		break; /* goto drop; */
1801 	}
1802  drop:
1803 	ndev->stats.tx_dropped++;
1804 	dev_kfree_skb_any(skb);
1805 
1806 	return NET_XMIT_DROP;
1807 }
1808 
1809 static inline bool wil_need_txstat(struct sk_buff *skb)
1810 {
1811 	struct ethhdr *eth = (void *)skb->data;
1812 
1813 	return is_unicast_ether_addr(eth->h_dest) && skb->sk &&
1814 	       (skb_shinfo(skb)->tx_flags & SKBTX_WIFI_STATUS);
1815 }
1816 
1817 static inline void wil_consume_skb(struct sk_buff *skb, bool acked)
1818 {
1819 	if (unlikely(wil_need_txstat(skb)))
1820 		skb_complete_wifi_ack(skb, acked);
1821 	else
1822 		acked ? dev_consume_skb_any(skb) : dev_kfree_skb_any(skb);
1823 }
1824 
1825 /**
1826  * Clean up transmitted skb's from the Tx VRING
1827  *
1828  * Return number of descriptors cleared
1829  *
1830  * Safe to call from IRQ
1831  */
1832 int wil_tx_complete(struct wil6210_priv *wil, int ringid)
1833 {
1834 	struct net_device *ndev = wil_to_ndev(wil);
1835 	struct device *dev = wil_to_dev(wil);
1836 	struct vring *vring = &wil->vring_tx[ringid];
1837 	struct vring_tx_data *txdata = &wil->vring_tx_data[ringid];
1838 	int done = 0;
1839 	int cid = wil->vring2cid_tid[ringid][0];
1840 	struct wil_net_stats *stats = NULL;
1841 	volatile struct vring_tx_desc *_d;
1842 	int used_before_complete;
1843 	int used_new;
1844 
1845 	if (unlikely(!vring->va)) {
1846 		wil_err(wil, "Tx irq[%d]: vring not initialized\n", ringid);
1847 		return 0;
1848 	}
1849 
1850 	if (unlikely(!txdata->enabled)) {
1851 		wil_info(wil, "Tx irq[%d]: vring disabled\n", ringid);
1852 		return 0;
1853 	}
1854 
1855 	wil_dbg_txrx(wil, "%s(%d)\n", __func__, ringid);
1856 
1857 	used_before_complete = wil_vring_used_tx(vring);
1858 
1859 	if (cid < WIL6210_MAX_CID)
1860 		stats = &wil->sta[cid].stats;
1861 
1862 	while (!wil_vring_is_empty(vring)) {
1863 		int new_swtail;
1864 		struct wil_ctx *ctx = &vring->ctx[vring->swtail];
1865 		/**
1866 		 * For the fragmented skb, HW will set DU bit only for the
1867 		 * last fragment. look for it.
1868 		 * In TSO the first DU will include hdr desc
1869 		 */
1870 		int lf = (vring->swtail + ctx->nr_frags) % vring->size;
1871 		/* TODO: check we are not past head */
1872 
1873 		_d = &vring->va[lf].tx;
1874 		if (unlikely(!(_d->dma.status & TX_DMA_STATUS_DU)))
1875 			break;
1876 
1877 		new_swtail = (lf + 1) % vring->size;
1878 		while (vring->swtail != new_swtail) {
1879 			struct vring_tx_desc dd, *d = &dd;
1880 			u16 dmalen;
1881 			struct sk_buff *skb;
1882 
1883 			ctx = &vring->ctx[vring->swtail];
1884 			skb = ctx->skb;
1885 			_d = &vring->va[vring->swtail].tx;
1886 
1887 			*d = *_d;
1888 
1889 			dmalen = le16_to_cpu(d->dma.length);
1890 			trace_wil6210_tx_done(ringid, vring->swtail, dmalen,
1891 					      d->dma.error);
1892 			wil_dbg_txrx(wil,
1893 				     "TxC[%2d][%3d] : %d bytes, status 0x%02x err 0x%02x\n",
1894 				     ringid, vring->swtail, dmalen,
1895 				     d->dma.status, d->dma.error);
1896 			wil_hex_dump_txrx("TxCD ", DUMP_PREFIX_NONE, 32, 4,
1897 					  (const void *)d, sizeof(*d), false);
1898 
1899 			wil_txdesc_unmap(dev, d, ctx);
1900 
1901 			if (skb) {
1902 				if (likely(d->dma.error == 0)) {
1903 					ndev->stats.tx_packets++;
1904 					ndev->stats.tx_bytes += skb->len;
1905 					if (stats) {
1906 						stats->tx_packets++;
1907 						stats->tx_bytes += skb->len;
1908 					}
1909 				} else {
1910 					ndev->stats.tx_errors++;
1911 					if (stats)
1912 						stats->tx_errors++;
1913 				}
1914 				wil_consume_skb(skb, d->dma.error == 0);
1915 			}
1916 			memset(ctx, 0, sizeof(*ctx));
1917 			/* There is no need to touch HW descriptor:
1918 			 * - ststus bit TX_DMA_STATUS_DU is set by design,
1919 			 *   so hardware will not try to process this desc.,
1920 			 * - rest of descriptor will be initialized on Tx.
1921 			 */
1922 			vring->swtail = wil_vring_next_tail(vring);
1923 			done++;
1924 		}
1925 	}
1926 
1927 	/* performance monitoring */
1928 	used_new = wil_vring_used_tx(vring);
1929 	if (wil_val_in_range(vring_idle_trsh,
1930 			     used_new, used_before_complete)) {
1931 		wil_dbg_txrx(wil, "Ring[%2d] idle %d -> %d\n",
1932 			     ringid, used_before_complete, used_new);
1933 		txdata->last_idle = get_cycles();
1934 	}
1935 
1936 	if (wil_vring_avail_tx(vring) > wil_vring_wmark_high(vring)) {
1937 		wil_dbg_txrx(wil, "netif_tx_wake : ring not full\n");
1938 		netif_tx_wake_all_queues(wil_to_ndev(wil));
1939 	}
1940 
1941 	return done;
1942 }
1943