1 /*
2  * Copyright (c) 2012-2014 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 static inline int wil_vring_is_empty(struct vring *vring)
37 {
38 	return vring->swhead == vring->swtail;
39 }
40 
41 static inline u32 wil_vring_next_tail(struct vring *vring)
42 {
43 	return (vring->swtail + 1) % vring->size;
44 }
45 
46 static inline void wil_vring_advance_head(struct vring *vring, int n)
47 {
48 	vring->swhead = (vring->swhead + n) % vring->size;
49 }
50 
51 static inline int wil_vring_is_full(struct vring *vring)
52 {
53 	return wil_vring_next_tail(vring) == vring->swhead;
54 }
55 
56 /*
57  * Available space in Tx Vring
58  */
59 static inline int wil_vring_avail_tx(struct vring *vring)
60 {
61 	u32 swhead = vring->swhead;
62 	u32 swtail = vring->swtail;
63 	int used = (vring->size + swhead - swtail) % vring->size;
64 
65 	return vring->size - used - 1;
66 }
67 
68 /**
69  * wil_vring_wmark_low - low watermark for available descriptor space
70  */
71 static inline int wil_vring_wmark_low(struct vring *vring)
72 {
73 	return vring->size/8;
74 }
75 
76 /**
77  * wil_vring_wmark_high - high watermark for available descriptor space
78  */
79 static inline int wil_vring_wmark_high(struct vring *vring)
80 {
81 	return vring->size/4;
82 }
83 
84 static int wil_vring_alloc(struct wil6210_priv *wil, struct vring *vring)
85 {
86 	struct device *dev = wil_to_dev(wil);
87 	size_t sz = vring->size * sizeof(vring->va[0]);
88 	uint i;
89 
90 	wil_dbg_misc(wil, "%s()\n", __func__);
91 
92 	BUILD_BUG_ON(sizeof(vring->va[0]) != 32);
93 
94 	vring->swhead = 0;
95 	vring->swtail = 0;
96 	vring->ctx = kcalloc(vring->size, sizeof(vring->ctx[0]), GFP_KERNEL);
97 	if (!vring->ctx) {
98 		vring->va = NULL;
99 		return -ENOMEM;
100 	}
101 	/*
102 	 * vring->va should be aligned on its size rounded up to power of 2
103 	 * This is granted by the dma_alloc_coherent
104 	 */
105 	vring->va = dma_alloc_coherent(dev, sz, &vring->pa, GFP_KERNEL);
106 	if (!vring->va) {
107 		kfree(vring->ctx);
108 		vring->ctx = NULL;
109 		return -ENOMEM;
110 	}
111 	/* initially, all descriptors are SW owned
112 	 * For Tx and Rx, ownership bit is at the same location, thus
113 	 * we can use any
114 	 */
115 	for (i = 0; i < vring->size; i++) {
116 		volatile struct vring_tx_desc *_d = &vring->va[i].tx;
117 
118 		_d->dma.status = TX_DMA_STATUS_DU;
119 	}
120 
121 	wil_dbg_misc(wil, "vring[%d] 0x%p:%pad 0x%p\n", vring->size,
122 		     vring->va, &vring->pa, vring->ctx);
123 
124 	return 0;
125 }
126 
127 static void wil_txdesc_unmap(struct device *dev, struct vring_tx_desc *d,
128 			     struct wil_ctx *ctx)
129 {
130 	dma_addr_t pa = wil_desc_addr(&d->dma.addr);
131 	u16 dmalen = le16_to_cpu(d->dma.length);
132 
133 	switch (ctx->mapped_as) {
134 	case wil_mapped_as_single:
135 		dma_unmap_single(dev, pa, dmalen, DMA_TO_DEVICE);
136 		break;
137 	case wil_mapped_as_page:
138 		dma_unmap_page(dev, pa, dmalen, DMA_TO_DEVICE);
139 		break;
140 	default:
141 		break;
142 	}
143 }
144 
145 static void wil_vring_free(struct wil6210_priv *wil, struct vring *vring,
146 			   int tx)
147 {
148 	struct device *dev = wil_to_dev(wil);
149 	size_t sz = vring->size * sizeof(vring->va[0]);
150 
151 	if (tx) {
152 		int vring_index = vring - wil->vring_tx;
153 
154 		wil_dbg_misc(wil, "free Tx vring %d [%d] 0x%p:%pad 0x%p\n",
155 			     vring_index, vring->size, vring->va,
156 			     &vring->pa, vring->ctx);
157 	} else {
158 		wil_dbg_misc(wil, "free Rx vring [%d] 0x%p:%pad 0x%p\n",
159 			     vring->size, vring->va,
160 			     &vring->pa, vring->ctx);
161 	}
162 
163 	while (!wil_vring_is_empty(vring)) {
164 		dma_addr_t pa;
165 		u16 dmalen;
166 		struct wil_ctx *ctx;
167 
168 		if (tx) {
169 			struct vring_tx_desc dd, *d = &dd;
170 			volatile struct vring_tx_desc *_d =
171 					&vring->va[vring->swtail].tx;
172 
173 			ctx = &vring->ctx[vring->swtail];
174 			*d = *_d;
175 			wil_txdesc_unmap(dev, d, ctx);
176 			if (ctx->skb)
177 				dev_kfree_skb_any(ctx->skb);
178 			vring->swtail = wil_vring_next_tail(vring);
179 		} else { /* rx */
180 			struct vring_rx_desc dd, *d = &dd;
181 			volatile struct vring_rx_desc *_d =
182 					&vring->va[vring->swhead].rx;
183 
184 			ctx = &vring->ctx[vring->swhead];
185 			*d = *_d;
186 			pa = wil_desc_addr(&d->dma.addr);
187 			dmalen = le16_to_cpu(d->dma.length);
188 			dma_unmap_single(dev, pa, dmalen, DMA_FROM_DEVICE);
189 			kfree_skb(ctx->skb);
190 			wil_vring_advance_head(vring, 1);
191 		}
192 	}
193 	dma_free_coherent(dev, sz, (void *)vring->va, vring->pa);
194 	kfree(vring->ctx);
195 	vring->pa = 0;
196 	vring->va = NULL;
197 	vring->ctx = NULL;
198 }
199 
200 /**
201  * Allocate one skb for Rx VRING
202  *
203  * Safe to call from IRQ
204  */
205 static int wil_vring_alloc_skb(struct wil6210_priv *wil, struct vring *vring,
206 			       u32 i, int headroom)
207 {
208 	struct device *dev = wil_to_dev(wil);
209 	unsigned int sz = mtu_max + ETH_HLEN;
210 	struct vring_rx_desc dd, *d = &dd;
211 	volatile struct vring_rx_desc *_d = &vring->va[i].rx;
212 	dma_addr_t pa;
213 	struct sk_buff *skb = dev_alloc_skb(sz + headroom);
214 
215 	if (unlikely(!skb))
216 		return -ENOMEM;
217 
218 	skb_reserve(skb, headroom);
219 	skb_put(skb, sz);
220 
221 	pa = dma_map_single(dev, skb->data, skb->len, DMA_FROM_DEVICE);
222 	if (unlikely(dma_mapping_error(dev, pa))) {
223 		kfree_skb(skb);
224 		return -ENOMEM;
225 	}
226 
227 	d->dma.d0 = BIT(9) | RX_DMA_D0_CMD_DMA_IT;
228 	wil_desc_addr_set(&d->dma.addr, pa);
229 	/* ip_length don't care */
230 	/* b11 don't care */
231 	/* error don't care */
232 	d->dma.status = 0; /* BIT(0) should be 0 for HW_OWNED */
233 	d->dma.length = cpu_to_le16(sz);
234 	*_d = *d;
235 	vring->ctx[i].skb = skb;
236 
237 	return 0;
238 }
239 
240 /**
241  * Adds radiotap header
242  *
243  * Any error indicated as "Bad FCS"
244  *
245  * Vendor data for 04:ce:14-1 (Wilocity-1) consists of:
246  *  - Rx descriptor: 32 bytes
247  *  - Phy info
248  */
249 static void wil_rx_add_radiotap_header(struct wil6210_priv *wil,
250 				       struct sk_buff *skb)
251 {
252 	struct wireless_dev *wdev = wil->wdev;
253 	struct wil6210_rtap {
254 		struct ieee80211_radiotap_header rthdr;
255 		/* fields should be in the order of bits in rthdr.it_present */
256 		/* flags */
257 		u8 flags;
258 		/* channel */
259 		__le16 chnl_freq __aligned(2);
260 		__le16 chnl_flags;
261 		/* MCS */
262 		u8 mcs_present;
263 		u8 mcs_flags;
264 		u8 mcs_index;
265 	} __packed;
266 	struct wil6210_rtap_vendor {
267 		struct wil6210_rtap rtap;
268 		/* vendor */
269 		u8 vendor_oui[3] __aligned(2);
270 		u8 vendor_ns;
271 		__le16 vendor_skip;
272 		u8 vendor_data[0];
273 	} __packed;
274 	struct vring_rx_desc *d = wil_skb_rxdesc(skb);
275 	struct wil6210_rtap_vendor *rtap_vendor;
276 	int rtap_len = sizeof(struct wil6210_rtap);
277 	int phy_length = 0; /* phy info header size, bytes */
278 	static char phy_data[128];
279 	struct ieee80211_channel *ch = wdev->preset_chandef.chan;
280 
281 	if (rtap_include_phy_info) {
282 		rtap_len = sizeof(*rtap_vendor) + sizeof(*d);
283 		/* calculate additional length */
284 		if (d->dma.status & RX_DMA_STATUS_PHY_INFO) {
285 			/**
286 			 * PHY info starts from 8-byte boundary
287 			 * there are 8-byte lines, last line may be partially
288 			 * written (HW bug), thus FW configures for last line
289 			 * to be excessive. Driver skips this last line.
290 			 */
291 			int len = min_t(int, 8 + sizeof(phy_data),
292 					wil_rxdesc_phy_length(d));
293 
294 			if (len > 8) {
295 				void *p = skb_tail_pointer(skb);
296 				void *pa = PTR_ALIGN(p, 8);
297 
298 				if (skb_tailroom(skb) >= len + (pa - p)) {
299 					phy_length = len - 8;
300 					memcpy(phy_data, pa, phy_length);
301 				}
302 			}
303 		}
304 		rtap_len += phy_length;
305 	}
306 
307 	if (skb_headroom(skb) < rtap_len &&
308 	    pskb_expand_head(skb, rtap_len, 0, GFP_ATOMIC)) {
309 		wil_err(wil, "Unable to expand headrom to %d\n", rtap_len);
310 		return;
311 	}
312 
313 	rtap_vendor = (void *)skb_push(skb, rtap_len);
314 	memset(rtap_vendor, 0, rtap_len);
315 
316 	rtap_vendor->rtap.rthdr.it_version = PKTHDR_RADIOTAP_VERSION;
317 	rtap_vendor->rtap.rthdr.it_len = cpu_to_le16(rtap_len);
318 	rtap_vendor->rtap.rthdr.it_present = cpu_to_le32(
319 			(1 << IEEE80211_RADIOTAP_FLAGS) |
320 			(1 << IEEE80211_RADIOTAP_CHANNEL) |
321 			(1 << IEEE80211_RADIOTAP_MCS));
322 	if (d->dma.status & RX_DMA_STATUS_ERROR)
323 		rtap_vendor->rtap.flags |= IEEE80211_RADIOTAP_F_BADFCS;
324 
325 	rtap_vendor->rtap.chnl_freq = cpu_to_le16(ch ? ch->center_freq : 58320);
326 	rtap_vendor->rtap.chnl_flags = cpu_to_le16(0);
327 
328 	rtap_vendor->rtap.mcs_present = IEEE80211_RADIOTAP_MCS_HAVE_MCS;
329 	rtap_vendor->rtap.mcs_flags = 0;
330 	rtap_vendor->rtap.mcs_index = wil_rxdesc_mcs(d);
331 
332 	if (rtap_include_phy_info) {
333 		rtap_vendor->rtap.rthdr.it_present |= cpu_to_le32(1 <<
334 				IEEE80211_RADIOTAP_VENDOR_NAMESPACE);
335 		/* OUI for Wilocity 04:ce:14 */
336 		rtap_vendor->vendor_oui[0] = 0x04;
337 		rtap_vendor->vendor_oui[1] = 0xce;
338 		rtap_vendor->vendor_oui[2] = 0x14;
339 		rtap_vendor->vendor_ns = 1;
340 		/* Rx descriptor + PHY data  */
341 		rtap_vendor->vendor_skip = cpu_to_le16(sizeof(*d) +
342 						       phy_length);
343 		memcpy(rtap_vendor->vendor_data, (void *)d, sizeof(*d));
344 		memcpy(rtap_vendor->vendor_data + sizeof(*d), phy_data,
345 		       phy_length);
346 	}
347 }
348 
349 /*
350  * Fast swap in place between 2 registers
351  */
352 static void wil_swap_u16(u16 *a, u16 *b)
353 {
354 	*a ^= *b;
355 	*b ^= *a;
356 	*a ^= *b;
357 }
358 
359 static void wil_swap_ethaddr(void *data)
360 {
361 	struct ethhdr *eth = data;
362 	u16 *s = (u16 *)eth->h_source;
363 	u16 *d = (u16 *)eth->h_dest;
364 
365 	wil_swap_u16(s++, d++);
366 	wil_swap_u16(s++, d++);
367 	wil_swap_u16(s, d);
368 }
369 
370 /**
371  * reap 1 frame from @swhead
372  *
373  * Rx descriptor copied to skb->cb
374  *
375  * Safe to call from IRQ
376  */
377 static struct sk_buff *wil_vring_reap_rx(struct wil6210_priv *wil,
378 					 struct vring *vring)
379 {
380 	struct device *dev = wil_to_dev(wil);
381 	struct net_device *ndev = wil_to_ndev(wil);
382 	volatile struct vring_rx_desc *_d;
383 	struct vring_rx_desc *d;
384 	struct sk_buff *skb;
385 	dma_addr_t pa;
386 	unsigned int sz = mtu_max + ETH_HLEN;
387 	u16 dmalen;
388 	u8 ftype;
389 	u8 ds_bits;
390 	int cid;
391 	struct wil_net_stats *stats;
392 
393 	BUILD_BUG_ON(sizeof(struct vring_rx_desc) > sizeof(skb->cb));
394 
395 	if (wil_vring_is_empty(vring))
396 		return NULL;
397 
398 	_d = &vring->va[vring->swhead].rx;
399 	if (!(_d->dma.status & RX_DMA_STATUS_DU)) {
400 		/* it is not error, we just reached end of Rx done area */
401 		return NULL;
402 	}
403 
404 	skb = vring->ctx[vring->swhead].skb;
405 	d = wil_skb_rxdesc(skb);
406 	*d = *_d;
407 	pa = wil_desc_addr(&d->dma.addr);
408 	vring->ctx[vring->swhead].skb = NULL;
409 	wil_vring_advance_head(vring, 1);
410 
411 	dma_unmap_single(dev, pa, sz, DMA_FROM_DEVICE);
412 	dmalen = le16_to_cpu(d->dma.length);
413 
414 	trace_wil6210_rx(vring->swhead, d);
415 	wil_dbg_txrx(wil, "Rx[%3d] : %d bytes\n", vring->swhead, dmalen);
416 	wil_hex_dump_txrx("Rx ", DUMP_PREFIX_NONE, 32, 4,
417 			  (const void *)d, sizeof(*d), false);
418 
419 	if (dmalen > sz) {
420 		wil_err(wil, "Rx size too large: %d bytes!\n", dmalen);
421 		kfree_skb(skb);
422 		return NULL;
423 	}
424 	skb_trim(skb, dmalen);
425 
426 	prefetch(skb->data);
427 
428 	wil_hex_dump_txrx("Rx ", DUMP_PREFIX_OFFSET, 16, 1,
429 			  skb->data, skb_headlen(skb), false);
430 
431 	cid = wil_rxdesc_cid(d);
432 	stats = &wil->sta[cid].stats;
433 	stats->last_mcs_rx = wil_rxdesc_mcs(d);
434 
435 	/* use radiotap header only if required */
436 	if (ndev->type == ARPHRD_IEEE80211_RADIOTAP)
437 		wil_rx_add_radiotap_header(wil, skb);
438 
439 	/* no extra checks if in sniffer mode */
440 	if (ndev->type != ARPHRD_ETHER)
441 		return skb;
442 	/*
443 	 * Non-data frames may be delivered through Rx DMA channel (ex: BAR)
444 	 * Driver should recognize it by frame type, that is found
445 	 * in Rx descriptor. If type is not data, it is 802.11 frame as is
446 	 */
447 	ftype = wil_rxdesc_ftype(d) << 2;
448 	if (ftype != IEEE80211_FTYPE_DATA) {
449 		wil_dbg_txrx(wil, "Non-data frame ftype 0x%08x\n", ftype);
450 		/* TODO: process it */
451 		kfree_skb(skb);
452 		return NULL;
453 	}
454 
455 	if (skb->len < ETH_HLEN) {
456 		wil_err(wil, "Short frame, len = %d\n", skb->len);
457 		/* TODO: process it (i.e. BAR) */
458 		kfree_skb(skb);
459 		return NULL;
460 	}
461 
462 	/* L4 IDENT is on when HW calculated checksum, check status
463 	 * and in case of error drop the packet
464 	 * higher stack layers will handle retransmission (if required)
465 	 */
466 	if (d->dma.status & RX_DMA_STATUS_L4_IDENT) {
467 		/* L4 protocol identified, csum calculated */
468 		if ((d->dma.error & RX_DMA_ERROR_L4_ERR) == 0)
469 			skb->ip_summed = CHECKSUM_UNNECESSARY;
470 		/* If HW reports bad checksum, let IP stack re-check it
471 		 * For example, HW don't understand Microsoft IP stack that
472 		 * mis-calculates TCP checksum - if it should be 0x0,
473 		 * it writes 0xffff in violation of RFC 1624
474 		 */
475 	}
476 
477 	ds_bits = wil_rxdesc_ds_bits(d);
478 	if (ds_bits == 1) {
479 		/*
480 		 * HW bug - in ToDS mode, i.e. Rx on AP side,
481 		 * addresses get swapped
482 		 */
483 		wil_swap_ethaddr(skb->data);
484 	}
485 
486 	return skb;
487 }
488 
489 /**
490  * allocate and fill up to @count buffers in rx ring
491  * buffers posted at @swtail
492  */
493 static int wil_rx_refill(struct wil6210_priv *wil, int count)
494 {
495 	struct net_device *ndev = wil_to_ndev(wil);
496 	struct vring *v = &wil->vring_rx;
497 	u32 next_tail;
498 	int rc = 0;
499 	int headroom = ndev->type == ARPHRD_IEEE80211_RADIOTAP ?
500 			WIL6210_RTAP_SIZE : 0;
501 
502 	for (; next_tail = wil_vring_next_tail(v),
503 			(next_tail != v->swhead) && (count-- > 0);
504 			v->swtail = next_tail) {
505 		rc = wil_vring_alloc_skb(wil, v, v->swtail, headroom);
506 		if (rc) {
507 			wil_err(wil, "Error %d in wil_rx_refill[%d]\n",
508 				rc, v->swtail);
509 			break;
510 		}
511 	}
512 	iowrite32(v->swtail, wil->csr + HOSTADDR(v->hwtail));
513 
514 	return rc;
515 }
516 
517 /*
518  * Pass Rx packet to the netif. Update statistics.
519  * Called in softirq context (NAPI poll).
520  */
521 void wil_netif_rx_any(struct sk_buff *skb, struct net_device *ndev)
522 {
523 	gro_result_t rc;
524 	struct wil6210_priv *wil = ndev_to_wil(ndev);
525 	unsigned int len = skb->len;
526 	struct vring_rx_desc *d = wil_skb_rxdesc(skb);
527 	int cid = wil_rxdesc_cid(d);
528 	struct wil_net_stats *stats = &wil->sta[cid].stats;
529 
530 	skb_orphan(skb);
531 
532 	rc = napi_gro_receive(&wil->napi_rx, skb);
533 
534 	if (unlikely(rc == GRO_DROP)) {
535 		ndev->stats.rx_dropped++;
536 		stats->rx_dropped++;
537 		wil_dbg_txrx(wil, "Rx drop %d bytes\n", len);
538 	} else {
539 		ndev->stats.rx_packets++;
540 		stats->rx_packets++;
541 		ndev->stats.rx_bytes += len;
542 		stats->rx_bytes += len;
543 	}
544 	{
545 		static const char * const gro_res_str[] = {
546 			[GRO_MERGED]		= "GRO_MERGED",
547 			[GRO_MERGED_FREE]	= "GRO_MERGED_FREE",
548 			[GRO_HELD]		= "GRO_HELD",
549 			[GRO_NORMAL]		= "GRO_NORMAL",
550 			[GRO_DROP]		= "GRO_DROP",
551 		};
552 		wil_dbg_txrx(wil, "Rx complete %d bytes => %s\n",
553 			     len, gro_res_str[rc]);
554 	}
555 }
556 
557 /**
558  * Proceed all completed skb's from Rx VRING
559  *
560  * Safe to call from NAPI poll, i.e. softirq with interrupts enabled
561  */
562 void wil_rx_handle(struct wil6210_priv *wil, int *quota)
563 {
564 	struct net_device *ndev = wil_to_ndev(wil);
565 	struct vring *v = &wil->vring_rx;
566 	struct sk_buff *skb;
567 
568 	if (!v->va) {
569 		wil_err(wil, "Rx IRQ while Rx not yet initialized\n");
570 		return;
571 	}
572 	wil_dbg_txrx(wil, "%s()\n", __func__);
573 	while ((*quota > 0) && (NULL != (skb = wil_vring_reap_rx(wil, v)))) {
574 		(*quota)--;
575 
576 		if (wil->wdev->iftype == NL80211_IFTYPE_MONITOR) {
577 			skb->dev = ndev;
578 			skb_reset_mac_header(skb);
579 			skb->ip_summed = CHECKSUM_UNNECESSARY;
580 			skb->pkt_type = PACKET_OTHERHOST;
581 			skb->protocol = htons(ETH_P_802_2);
582 			wil_netif_rx_any(skb, ndev);
583 		} else {
584 			struct ethhdr *eth = (void *)skb->data;
585 
586 			skb->protocol = eth_type_trans(skb, ndev);
587 
588 			if (is_unicast_ether_addr(eth->h_dest))
589 				wil_rx_reorder(wil, skb);
590 			else
591 				wil_netif_rx_any(skb, ndev);
592 		}
593 	}
594 	wil_rx_refill(wil, v->size);
595 }
596 
597 int wil_rx_init(struct wil6210_priv *wil, u16 size)
598 {
599 	struct vring *vring = &wil->vring_rx;
600 	int rc;
601 
602 	wil_dbg_misc(wil, "%s()\n", __func__);
603 
604 	if (vring->va) {
605 		wil_err(wil, "Rx ring already allocated\n");
606 		return -EINVAL;
607 	}
608 
609 	vring->size = size;
610 	rc = wil_vring_alloc(wil, vring);
611 	if (rc)
612 		return rc;
613 
614 	rc = wmi_rx_chain_add(wil, vring);
615 	if (rc)
616 		goto err_free;
617 
618 	rc = wil_rx_refill(wil, vring->size);
619 	if (rc)
620 		goto err_free;
621 
622 	return 0;
623  err_free:
624 	wil_vring_free(wil, vring, 0);
625 
626 	return rc;
627 }
628 
629 void wil_rx_fini(struct wil6210_priv *wil)
630 {
631 	struct vring *vring = &wil->vring_rx;
632 
633 	wil_dbg_misc(wil, "%s()\n", __func__);
634 
635 	if (vring->va)
636 		wil_vring_free(wil, vring, 0);
637 }
638 
639 int wil_vring_init_tx(struct wil6210_priv *wil, int id, int size,
640 		      int cid, int tid)
641 {
642 	int rc;
643 	struct wmi_vring_cfg_cmd cmd = {
644 		.action = cpu_to_le32(WMI_VRING_CMD_ADD),
645 		.vring_cfg = {
646 			.tx_sw_ring = {
647 				.max_mpdu_size =
648 					cpu_to_le16(mtu_max + ETH_HLEN),
649 				.ring_size = cpu_to_le16(size),
650 			},
651 			.ringid = id,
652 			.cidxtid = mk_cidxtid(cid, tid),
653 			.encap_trans_type = WMI_VRING_ENC_TYPE_802_3,
654 			.mac_ctrl = 0,
655 			.to_resolution = 0,
656 			.agg_max_wsize = 16,
657 			.schd_params = {
658 				.priority = cpu_to_le16(0),
659 				.timeslot_us = cpu_to_le16(0xfff),
660 			},
661 		},
662 	};
663 	struct {
664 		struct wil6210_mbox_hdr_wmi wmi;
665 		struct wmi_vring_cfg_done_event cmd;
666 	} __packed reply;
667 	struct vring *vring = &wil->vring_tx[id];
668 	struct vring_tx_data *txdata = &wil->vring_tx_data[id];
669 
670 	wil_dbg_misc(wil, "%s() max_mpdu_size %d\n", __func__,
671 		     cmd.vring_cfg.tx_sw_ring.max_mpdu_size);
672 
673 	if (vring->va) {
674 		wil_err(wil, "Tx ring [%d] already allocated\n", id);
675 		rc = -EINVAL;
676 		goto out;
677 	}
678 
679 	memset(txdata, 0, sizeof(*txdata));
680 	vring->size = size;
681 	rc = wil_vring_alloc(wil, vring);
682 	if (rc)
683 		goto out;
684 
685 	wil->vring2cid_tid[id][0] = cid;
686 	wil->vring2cid_tid[id][1] = tid;
687 
688 	cmd.vring_cfg.tx_sw_ring.ring_mem_base = cpu_to_le64(vring->pa);
689 
690 	rc = wmi_call(wil, WMI_VRING_CFG_CMDID, &cmd, sizeof(cmd),
691 		      WMI_VRING_CFG_DONE_EVENTID, &reply, sizeof(reply), 100);
692 	if (rc)
693 		goto out_free;
694 
695 	if (reply.cmd.status != WMI_FW_STATUS_SUCCESS) {
696 		wil_err(wil, "Tx config failed, status 0x%02x\n",
697 			reply.cmd.status);
698 		rc = -EINVAL;
699 		goto out_free;
700 	}
701 	vring->hwtail = le32_to_cpu(reply.cmd.tx_vring_tail_ptr);
702 
703 	txdata->enabled = 1;
704 
705 	return 0;
706  out_free:
707 	wil_vring_free(wil, vring, 1);
708  out:
709 
710 	return rc;
711 }
712 
713 void wil_vring_fini_tx(struct wil6210_priv *wil, int id)
714 {
715 	struct vring *vring = &wil->vring_tx[id];
716 
717 	WARN_ON(!mutex_is_locked(&wil->mutex));
718 
719 	if (!vring->va)
720 		return;
721 
722 	wil_dbg_misc(wil, "%s() id=%d\n", __func__, id);
723 
724 	/* make sure NAPI won't touch this vring */
725 	wil->vring_tx_data[id].enabled = 0;
726 	if (test_bit(wil_status_napi_en, &wil->status))
727 		napi_synchronize(&wil->napi_tx);
728 
729 	wil_vring_free(wil, vring, 1);
730 }
731 
732 static struct vring *wil_find_tx_vring(struct wil6210_priv *wil,
733 				       struct sk_buff *skb)
734 {
735 	int i;
736 	struct ethhdr *eth = (void *)skb->data;
737 	int cid = wil_find_cid(wil, eth->h_dest);
738 
739 	if (cid < 0)
740 		return NULL;
741 
742 	if (!wil->sta[cid].data_port_open &&
743 	    (skb->protocol != cpu_to_be16(ETH_P_PAE)))
744 		return NULL;
745 
746 	/* TODO: fix for multiple TID */
747 	for (i = 0; i < ARRAY_SIZE(wil->vring2cid_tid); i++) {
748 		if (wil->vring2cid_tid[i][0] == cid) {
749 			struct vring *v = &wil->vring_tx[i];
750 
751 			wil_dbg_txrx(wil, "%s(%pM) -> [%d]\n",
752 				     __func__, eth->h_dest, i);
753 			if (v->va) {
754 				return v;
755 			} else {
756 				wil_dbg_txrx(wil, "vring[%d] not valid\n", i);
757 				return NULL;
758 			}
759 		}
760 	}
761 
762 	return NULL;
763 }
764 
765 static void wil_set_da_for_vring(struct wil6210_priv *wil,
766 				 struct sk_buff *skb, int vring_index)
767 {
768 	struct ethhdr *eth = (void *)skb->data;
769 	int cid = wil->vring2cid_tid[vring_index][0];
770 
771 	memcpy(eth->h_dest, wil->sta[cid].addr, ETH_ALEN);
772 }
773 
774 static int wil_tx_vring(struct wil6210_priv *wil, struct vring *vring,
775 			struct sk_buff *skb);
776 /*
777  * Find 1-st vring and return it; set dest address for this vring in skb
778  * duplicate skb and send it to other active vrings
779  */
780 static struct vring *wil_tx_bcast(struct wil6210_priv *wil,
781 				  struct sk_buff *skb)
782 {
783 	struct vring *v, *v2;
784 	struct sk_buff *skb2;
785 	int i;
786 	u8 cid;
787 
788 	/* find 1-st vring eligible for data */
789 	for (i = 0; i < WIL6210_MAX_TX_RINGS; i++) {
790 		v = &wil->vring_tx[i];
791 		if (!v->va)
792 			continue;
793 
794 		cid = wil->vring2cid_tid[i][0];
795 		if (!wil->sta[cid].data_port_open)
796 			continue;
797 
798 		goto found;
799 	}
800 
801 	wil_dbg_txrx(wil, "Tx while no vrings active?\n");
802 
803 	return NULL;
804 
805 found:
806 	wil_dbg_txrx(wil, "BCAST -> ring %d\n", i);
807 	wil_set_da_for_vring(wil, skb, i);
808 
809 	/* find other active vrings and duplicate skb for each */
810 	for (i++; i < WIL6210_MAX_TX_RINGS; i++) {
811 		v2 = &wil->vring_tx[i];
812 		if (!v2->va)
813 			continue;
814 		cid = wil->vring2cid_tid[i][0];
815 		if (!wil->sta[cid].data_port_open)
816 			continue;
817 
818 		skb2 = skb_copy(skb, GFP_ATOMIC);
819 		if (skb2) {
820 			wil_dbg_txrx(wil, "BCAST DUP -> ring %d\n", i);
821 			wil_set_da_for_vring(wil, skb2, i);
822 			wil_tx_vring(wil, v2, skb2);
823 		} else {
824 			wil_err(wil, "skb_copy failed\n");
825 		}
826 	}
827 
828 	return v;
829 }
830 
831 static int wil_tx_desc_map(struct vring_tx_desc *d, dma_addr_t pa, u32 len,
832 			   int vring_index)
833 {
834 	wil_desc_addr_set(&d->dma.addr, pa);
835 	d->dma.ip_length = 0;
836 	/* 0..6: mac_length; 7:ip_version 0-IP6 1-IP4*/
837 	d->dma.b11 = 0/*14 | BIT(7)*/;
838 	d->dma.error = 0;
839 	d->dma.status = 0; /* BIT(0) should be 0 for HW_OWNED */
840 	d->dma.length = cpu_to_le16((u16)len);
841 	d->dma.d0 = (vring_index << DMA_CFG_DESC_TX_0_QID_POS);
842 	d->mac.d[0] = 0;
843 	d->mac.d[1] = 0;
844 	d->mac.d[2] = 0;
845 	d->mac.ucode_cmd = 0;
846 	/* use dst index 0 */
847 	d->mac.d[1] |= BIT(MAC_CFG_DESC_TX_1_DST_INDEX_EN_POS) |
848 		       (0 << MAC_CFG_DESC_TX_1_DST_INDEX_POS);
849 	/* translation type:  0 - bypass; 1 - 802.3; 2 - native wifi */
850 	d->mac.d[2] = BIT(MAC_CFG_DESC_TX_2_SNAP_HDR_INSERTION_EN_POS) |
851 		      (1 << MAC_CFG_DESC_TX_2_L2_TRANSLATION_TYPE_POS);
852 
853 	return 0;
854 }
855 
856 static inline
857 void wil_tx_desc_set_nr_frags(struct vring_tx_desc *d, int nr_frags)
858 {
859 	d->mac.d[2] |= ((nr_frags + 1) <<
860 		       MAC_CFG_DESC_TX_2_NUM_OF_DESCRIPTORS_POS);
861 }
862 
863 static int wil_tx_desc_offload_cksum_set(struct wil6210_priv *wil,
864 					 struct vring_tx_desc *d,
865 					 struct sk_buff *skb)
866 {
867 	int protocol;
868 
869 	if (skb->ip_summed != CHECKSUM_PARTIAL)
870 		return 0;
871 
872 	d->dma.b11 = ETH_HLEN; /* MAC header length */
873 
874 	switch (skb->protocol) {
875 	case cpu_to_be16(ETH_P_IP):
876 		protocol = ip_hdr(skb)->protocol;
877 		d->dma.b11 |= BIT(DMA_CFG_DESC_TX_OFFLOAD_CFG_L3T_IPV4_POS);
878 		break;
879 	case cpu_to_be16(ETH_P_IPV6):
880 		protocol = ipv6_hdr(skb)->nexthdr;
881 		break;
882 	default:
883 		return -EINVAL;
884 	}
885 
886 	switch (protocol) {
887 	case IPPROTO_TCP:
888 		d->dma.d0 |= (2 << DMA_CFG_DESC_TX_0_L4_TYPE_POS);
889 		/* L4 header len: TCP header length */
890 		d->dma.d0 |=
891 		(tcp_hdrlen(skb) & DMA_CFG_DESC_TX_0_L4_LENGTH_MSK);
892 		break;
893 	case IPPROTO_UDP:
894 		/* L4 header len: UDP header length */
895 		d->dma.d0 |=
896 		(sizeof(struct udphdr) & DMA_CFG_DESC_TX_0_L4_LENGTH_MSK);
897 		break;
898 	default:
899 		return -EINVAL;
900 	}
901 
902 	d->dma.ip_length = skb_network_header_len(skb);
903 	/* Enable TCP/UDP checksum */
904 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_TCP_UDP_CHECKSUM_EN_POS);
905 	/* Calculate pseudo-header */
906 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_PSEUDO_HEADER_CALC_EN_POS);
907 
908 	return 0;
909 }
910 
911 static int wil_tx_vring(struct wil6210_priv *wil, struct vring *vring,
912 			struct sk_buff *skb)
913 {
914 	struct device *dev = wil_to_dev(wil);
915 	struct vring_tx_desc dd, *d = &dd;
916 	volatile struct vring_tx_desc *_d;
917 	u32 swhead = vring->swhead;
918 	int avail = wil_vring_avail_tx(vring);
919 	int nr_frags = skb_shinfo(skb)->nr_frags;
920 	uint f = 0;
921 	int vring_index = vring - wil->vring_tx;
922 	struct vring_tx_data *txdata = &wil->vring_tx_data[vring_index];
923 	uint i = swhead;
924 	dma_addr_t pa;
925 
926 	wil_dbg_txrx(wil, "%s()\n", __func__);
927 
928 	if (avail < 1 + nr_frags) {
929 		wil_err_ratelimited(wil,
930 				    "Tx ring full. No space for %d fragments\n",
931 				    1 + nr_frags);
932 		return -ENOMEM;
933 	}
934 	_d = &vring->va[i].tx;
935 
936 	pa = dma_map_single(dev, skb->data, skb_headlen(skb), DMA_TO_DEVICE);
937 
938 	wil_dbg_txrx(wil, "Tx skb %d bytes 0x%p -> %pad\n", skb_headlen(skb),
939 		     skb->data, &pa);
940 	wil_hex_dump_txrx("Tx ", DUMP_PREFIX_OFFSET, 16, 1,
941 			  skb->data, skb_headlen(skb), false);
942 
943 	if (unlikely(dma_mapping_error(dev, pa)))
944 		return -EINVAL;
945 	vring->ctx[i].mapped_as = wil_mapped_as_single;
946 	/* 1-st segment */
947 	wil_tx_desc_map(d, pa, skb_headlen(skb), vring_index);
948 	/* Process TCP/UDP checksum offloading */
949 	if (wil_tx_desc_offload_cksum_set(wil, d, skb)) {
950 		wil_err(wil, "VRING #%d Failed to set cksum, drop packet\n",
951 			vring_index);
952 		goto dma_error;
953 	}
954 
955 	vring->ctx[i].nr_frags = nr_frags;
956 	wil_tx_desc_set_nr_frags(d, nr_frags);
957 	if (nr_frags)
958 		*_d = *d;
959 
960 	/* middle segments */
961 	for (; f < nr_frags; f++) {
962 		const struct skb_frag_struct *frag =
963 				&skb_shinfo(skb)->frags[f];
964 		int len = skb_frag_size(frag);
965 
966 		i = (swhead + f + 1) % vring->size;
967 		_d = &vring->va[i].tx;
968 		pa = skb_frag_dma_map(dev, frag, 0, skb_frag_size(frag),
969 				      DMA_TO_DEVICE);
970 		if (unlikely(dma_mapping_error(dev, pa)))
971 			goto dma_error;
972 		vring->ctx[i].mapped_as = wil_mapped_as_page;
973 		wil_tx_desc_map(d, pa, len, vring_index);
974 		/* no need to check return code -
975 		 * if it succeeded for 1-st descriptor,
976 		 * it will succeed here too
977 		 */
978 		wil_tx_desc_offload_cksum_set(wil, d, skb);
979 		*_d = *d;
980 	}
981 	/* for the last seg only */
982 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_CMD_EOP_POS);
983 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_CMD_MARK_WB_POS);
984 	d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_CMD_DMA_IT_POS);
985 	*_d = *d;
986 
987 	/* hold reference to skb
988 	 * to prevent skb release before accounting
989 	 * in case of immediate "tx done"
990 	 */
991 	vring->ctx[i].skb = skb_get(skb);
992 
993 	wil_hex_dump_txrx("Tx ", DUMP_PREFIX_NONE, 32, 4,
994 			  (const void *)d, sizeof(*d), false);
995 
996 	if (wil_vring_is_empty(vring)) /* performance monitoring */
997 		txdata->idle += get_cycles() - txdata->last_idle;
998 
999 	/* advance swhead */
1000 	wil_vring_advance_head(vring, nr_frags + 1);
1001 	wil_dbg_txrx(wil, "Tx swhead %d -> %d\n", swhead, vring->swhead);
1002 	trace_wil6210_tx(vring_index, swhead, skb->len, nr_frags);
1003 	iowrite32(vring->swhead, wil->csr + HOSTADDR(vring->hwtail));
1004 
1005 	return 0;
1006  dma_error:
1007 	/* unmap what we have mapped */
1008 	nr_frags = f + 1; /* frags mapped + one for skb head */
1009 	for (f = 0; f < nr_frags; f++) {
1010 		struct wil_ctx *ctx;
1011 
1012 		i = (swhead + f) % vring->size;
1013 		ctx = &vring->ctx[i];
1014 		_d = &vring->va[i].tx;
1015 		*d = *_d;
1016 		_d->dma.status = TX_DMA_STATUS_DU;
1017 		wil_txdesc_unmap(dev, d, ctx);
1018 
1019 		if (ctx->skb)
1020 			dev_kfree_skb_any(ctx->skb);
1021 
1022 		memset(ctx, 0, sizeof(*ctx));
1023 	}
1024 
1025 	return -EINVAL;
1026 }
1027 
1028 netdev_tx_t wil_start_xmit(struct sk_buff *skb, struct net_device *ndev)
1029 {
1030 	struct wil6210_priv *wil = ndev_to_wil(ndev);
1031 	struct ethhdr *eth = (void *)skb->data;
1032 	struct vring *vring;
1033 	static bool pr_once_fw;
1034 	int rc;
1035 
1036 	wil_dbg_txrx(wil, "%s()\n", __func__);
1037 	if (!test_bit(wil_status_fwready, &wil->status)) {
1038 		if (!pr_once_fw) {
1039 			wil_err(wil, "FW not ready\n");
1040 			pr_once_fw = true;
1041 		}
1042 		goto drop;
1043 	}
1044 	if (!test_bit(wil_status_fwconnected, &wil->status)) {
1045 		wil_err(wil, "FW not connected\n");
1046 		goto drop;
1047 	}
1048 	if (wil->wdev->iftype == NL80211_IFTYPE_MONITOR) {
1049 		wil_err(wil, "Xmit in monitor mode not supported\n");
1050 		goto drop;
1051 	}
1052 	pr_once_fw = false;
1053 
1054 	/* find vring */
1055 	if (is_unicast_ether_addr(eth->h_dest))
1056 		vring = wil_find_tx_vring(wil, skb);
1057 	else
1058 		vring = wil_tx_bcast(wil, skb);
1059 	if (!vring) {
1060 		wil_dbg_txrx(wil, "No Tx VRING found for %pM\n", eth->h_dest);
1061 		goto drop;
1062 	}
1063 
1064 	/* set up vring entry */
1065 	rc = wil_tx_vring(wil, vring, skb);
1066 
1067 	/* do we still have enough room in the vring? */
1068 	if (wil_vring_avail_tx(vring) < wil_vring_wmark_low(vring)) {
1069 		netif_tx_stop_all_queues(wil_to_ndev(wil));
1070 		wil_dbg_txrx(wil, "netif_tx_stop : ring full\n");
1071 	}
1072 
1073 	switch (rc) {
1074 	case 0:
1075 		/* statistics will be updated on the tx_complete */
1076 		dev_kfree_skb_any(skb);
1077 		return NETDEV_TX_OK;
1078 	case -ENOMEM:
1079 		return NETDEV_TX_BUSY;
1080 	default:
1081 		break; /* goto drop; */
1082 	}
1083  drop:
1084 	ndev->stats.tx_dropped++;
1085 	dev_kfree_skb_any(skb);
1086 
1087 	return NET_XMIT_DROP;
1088 }
1089 
1090 /**
1091  * Clean up transmitted skb's from the Tx VRING
1092  *
1093  * Return number of descriptors cleared
1094  *
1095  * Safe to call from IRQ
1096  */
1097 int wil_tx_complete(struct wil6210_priv *wil, int ringid)
1098 {
1099 	struct net_device *ndev = wil_to_ndev(wil);
1100 	struct device *dev = wil_to_dev(wil);
1101 	struct vring *vring = &wil->vring_tx[ringid];
1102 	struct vring_tx_data *txdata = &wil->vring_tx_data[ringid];
1103 	int done = 0;
1104 	int cid = wil->vring2cid_tid[ringid][0];
1105 	struct wil_net_stats *stats = &wil->sta[cid].stats;
1106 	volatile struct vring_tx_desc *_d;
1107 
1108 	if (!vring->va) {
1109 		wil_err(wil, "Tx irq[%d]: vring not initialized\n", ringid);
1110 		return 0;
1111 	}
1112 
1113 	if (!txdata->enabled) {
1114 		wil_info(wil, "Tx irq[%d]: vring disabled\n", ringid);
1115 		return 0;
1116 	}
1117 
1118 	wil_dbg_txrx(wil, "%s(%d)\n", __func__, ringid);
1119 
1120 	while (!wil_vring_is_empty(vring)) {
1121 		int new_swtail;
1122 		struct wil_ctx *ctx = &vring->ctx[vring->swtail];
1123 		/**
1124 		 * For the fragmented skb, HW will set DU bit only for the
1125 		 * last fragment. look for it
1126 		 */
1127 		int lf = (vring->swtail + ctx->nr_frags) % vring->size;
1128 		/* TODO: check we are not past head */
1129 
1130 		_d = &vring->va[lf].tx;
1131 		if (!(_d->dma.status & TX_DMA_STATUS_DU))
1132 			break;
1133 
1134 		new_swtail = (lf + 1) % vring->size;
1135 		while (vring->swtail != new_swtail) {
1136 			struct vring_tx_desc dd, *d = &dd;
1137 			u16 dmalen;
1138 			struct sk_buff *skb;
1139 
1140 			ctx = &vring->ctx[vring->swtail];
1141 			skb = ctx->skb;
1142 			_d = &vring->va[vring->swtail].tx;
1143 
1144 			*d = *_d;
1145 
1146 			dmalen = le16_to_cpu(d->dma.length);
1147 			trace_wil6210_tx_done(ringid, vring->swtail, dmalen,
1148 					      d->dma.error);
1149 			wil_dbg_txrx(wil,
1150 				     "Tx[%3d] : %d bytes, status 0x%02x err 0x%02x\n",
1151 				     vring->swtail, dmalen, d->dma.status,
1152 				     d->dma.error);
1153 			wil_hex_dump_txrx("TxC ", DUMP_PREFIX_NONE, 32, 4,
1154 					  (const void *)d, sizeof(*d), false);
1155 
1156 			wil_txdesc_unmap(dev, d, ctx);
1157 
1158 			if (skb) {
1159 				if (d->dma.error == 0) {
1160 					ndev->stats.tx_packets++;
1161 					stats->tx_packets++;
1162 					ndev->stats.tx_bytes += skb->len;
1163 					stats->tx_bytes += skb->len;
1164 				} else {
1165 					ndev->stats.tx_errors++;
1166 					stats->tx_errors++;
1167 				}
1168 
1169 				dev_kfree_skb_any(skb);
1170 			}
1171 			memset(ctx, 0, sizeof(*ctx));
1172 			/* There is no need to touch HW descriptor:
1173 			 * - ststus bit TX_DMA_STATUS_DU is set by design,
1174 			 *   so hardware will not try to process this desc.,
1175 			 * - rest of descriptor will be initialized on Tx.
1176 			 */
1177 			vring->swtail = wil_vring_next_tail(vring);
1178 			done++;
1179 		}
1180 	}
1181 
1182 	if (wil_vring_is_empty(vring)) { /* performance monitoring */
1183 		wil_dbg_txrx(wil, "Ring[%2d] empty\n", ringid);
1184 		txdata->last_idle = get_cycles();
1185 	}
1186 
1187 	if (wil_vring_avail_tx(vring) > wil_vring_wmark_high(vring)) {
1188 		wil_dbg_txrx(wil, "netif_tx_wake : ring not full\n");
1189 		netif_tx_wake_all_queues(wil_to_ndev(wil));
1190 	}
1191 
1192 	return done;
1193 }
1194