2 * Copyright (c) 2012-2015 Qualcomm Atheros, Inc.
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.
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.
17 #include <linux/etherdevice.h>
18 #include <net/ieee80211_radiotap.h>
19 #include <linux/if_arp.h>
20 #include <linux/moduleparam.h>
22 #include <linux/ipv6.h>
24 #include <linux/prefetch.h>
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");
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");
40 static inline uint wil_rx_snaplen(void)
42 return rx_align_2 ? 6 : 0;
45 static inline int wil_vring_is_empty(struct vring *vring)
47 return vring->swhead == vring->swtail;
50 static inline u32 wil_vring_next_tail(struct vring *vring)
52 return (vring->swtail + 1) % vring->size;
55 static inline void wil_vring_advance_head(struct vring *vring, int n)
57 vring->swhead = (vring->swhead + n) % vring->size;
60 static inline int wil_vring_is_full(struct vring *vring)
62 return wil_vring_next_tail(vring) == vring->swhead;
65 /* Used space in Tx Vring */
66 static inline int wil_vring_used_tx(struct vring *vring)
68 u32 swhead = vring->swhead;
69 u32 swtail = vring->swtail;
70 return (vring->size + swhead - swtail) % vring->size;
73 /* Available space in Tx Vring */
74 static inline int wil_vring_avail_tx(struct vring *vring)
76 return vring->size - wil_vring_used_tx(vring) - 1;
79 /* wil_vring_wmark_low - low watermark for available descriptor space */
80 static inline int wil_vring_wmark_low(struct vring *vring)
85 /* wil_vring_wmark_high - high watermark for available descriptor space */
86 static inline int wil_vring_wmark_high(struct vring *vring)
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)
94 return val >= min && val < max;
97 static int wil_vring_alloc(struct wil6210_priv *wil, struct vring *vring)
99 struct device *dev = wil_to_dev(wil);
100 size_t sz = vring->size * sizeof(vring->va[0]);
103 wil_dbg_misc(wil, "%s()\n", __func__);
105 BUILD_BUG_ON(sizeof(vring->va[0]) != 32);
109 vring->ctx = kcalloc(vring->size, sizeof(vring->ctx[0]), GFP_KERNEL);
114 /* vring->va should be aligned on its size rounded up to power of 2
115 * This is granted by the dma_alloc_coherent
117 vring->va = dma_alloc_coherent(dev, sz, &vring->pa, GFP_KERNEL);
123 /* initially, all descriptors are SW owned
124 * For Tx and Rx, ownership bit is at the same location, thus
127 for (i = 0; i < vring->size; i++) {
128 volatile struct vring_tx_desc *_d = &vring->va[i].tx;
130 _d->dma.status = TX_DMA_STATUS_DU;
133 wil_dbg_misc(wil, "vring[%d] 0x%p:%pad 0x%p\n", vring->size,
134 vring->va, &vring->pa, vring->ctx);
139 static void wil_txdesc_unmap(struct device *dev, struct vring_tx_desc *d,
142 dma_addr_t pa = wil_desc_addr(&d->dma.addr);
143 u16 dmalen = le16_to_cpu(d->dma.length);
145 switch (ctx->mapped_as) {
146 case wil_mapped_as_single:
147 dma_unmap_single(dev, pa, dmalen, DMA_TO_DEVICE);
149 case wil_mapped_as_page:
150 dma_unmap_page(dev, pa, dmalen, DMA_TO_DEVICE);
157 static void wil_vring_free(struct wil6210_priv *wil, struct vring *vring,
160 struct device *dev = wil_to_dev(wil);
161 size_t sz = vring->size * sizeof(vring->va[0]);
164 int vring_index = vring - wil->vring_tx;
166 wil_dbg_misc(wil, "free Tx vring %d [%d] 0x%p:%pad 0x%p\n",
167 vring_index, vring->size, vring->va,
168 &vring->pa, vring->ctx);
170 wil_dbg_misc(wil, "free Rx vring [%d] 0x%p:%pad 0x%p\n",
171 vring->size, vring->va,
172 &vring->pa, vring->ctx);
175 while (!wil_vring_is_empty(vring)) {
181 struct vring_tx_desc dd, *d = ⅆ
182 volatile struct vring_tx_desc *_d =
183 &vring->va[vring->swtail].tx;
185 ctx = &vring->ctx[vring->swtail];
187 wil_txdesc_unmap(dev, d, ctx);
189 dev_kfree_skb_any(ctx->skb);
190 vring->swtail = wil_vring_next_tail(vring);
192 struct vring_rx_desc dd, *d = ⅆ
193 volatile struct vring_rx_desc *_d =
194 &vring->va[vring->swhead].rx;
196 ctx = &vring->ctx[vring->swhead];
198 pa = wil_desc_addr(&d->dma.addr);
199 dmalen = le16_to_cpu(d->dma.length);
200 dma_unmap_single(dev, pa, dmalen, DMA_FROM_DEVICE);
202 wil_vring_advance_head(vring, 1);
205 dma_free_coherent(dev, sz, (void *)vring->va, vring->pa);
213 * Allocate one skb for Rx VRING
215 * Safe to call from IRQ
217 static int wil_vring_alloc_skb(struct wil6210_priv *wil, struct vring *vring,
220 struct device *dev = wil_to_dev(wil);
221 unsigned int sz = mtu_max + ETH_HLEN + wil_rx_snaplen();
222 struct vring_rx_desc dd, *d = ⅆ
223 volatile struct vring_rx_desc *_d = &vring->va[i].rx;
225 struct sk_buff *skb = dev_alloc_skb(sz + headroom);
230 skb_reserve(skb, headroom);
233 pa = dma_map_single(dev, skb->data, skb->len, DMA_FROM_DEVICE);
234 if (unlikely(dma_mapping_error(dev, pa))) {
239 d->dma.d0 = RX_DMA_D0_CMD_DMA_RT | RX_DMA_D0_CMD_DMA_IT;
240 wil_desc_addr_set(&d->dma.addr, pa);
241 /* ip_length don't care */
243 /* error don't care */
244 d->dma.status = 0; /* BIT(0) should be 0 for HW_OWNED */
245 d->dma.length = cpu_to_le16(sz);
247 vring->ctx[i].skb = skb;
253 * Adds radiotap header
255 * Any error indicated as "Bad FCS"
257 * Vendor data for 04:ce:14-1 (Wilocity-1) consists of:
258 * - Rx descriptor: 32 bytes
261 static void wil_rx_add_radiotap_header(struct wil6210_priv *wil,
264 struct wireless_dev *wdev = wil->wdev;
265 struct wil6210_rtap {
266 struct ieee80211_radiotap_header rthdr;
267 /* fields should be in the order of bits in rthdr.it_present */
271 __le16 chnl_freq __aligned(2);
278 struct wil6210_rtap_vendor {
279 struct wil6210_rtap rtap;
281 u8 vendor_oui[3] __aligned(2);
286 struct vring_rx_desc *d = wil_skb_rxdesc(skb);
287 struct wil6210_rtap_vendor *rtap_vendor;
288 int rtap_len = sizeof(struct wil6210_rtap);
289 int phy_length = 0; /* phy info header size, bytes */
290 static char phy_data[128];
291 struct ieee80211_channel *ch = wdev->preset_chandef.chan;
293 if (rtap_include_phy_info) {
294 rtap_len = sizeof(*rtap_vendor) + sizeof(*d);
295 /* calculate additional length */
296 if (d->dma.status & RX_DMA_STATUS_PHY_INFO) {
298 * PHY info starts from 8-byte boundary
299 * there are 8-byte lines, last line may be partially
300 * written (HW bug), thus FW configures for last line
301 * to be excessive. Driver skips this last line.
303 int len = min_t(int, 8 + sizeof(phy_data),
304 wil_rxdesc_phy_length(d));
307 void *p = skb_tail_pointer(skb);
308 void *pa = PTR_ALIGN(p, 8);
310 if (skb_tailroom(skb) >= len + (pa - p)) {
311 phy_length = len - 8;
312 memcpy(phy_data, pa, phy_length);
316 rtap_len += phy_length;
319 if (skb_headroom(skb) < rtap_len &&
320 pskb_expand_head(skb, rtap_len, 0, GFP_ATOMIC)) {
321 wil_err(wil, "Unable to expand headrom to %d\n", rtap_len);
325 rtap_vendor = (void *)skb_push(skb, rtap_len);
326 memset(rtap_vendor, 0, rtap_len);
328 rtap_vendor->rtap.rthdr.it_version = PKTHDR_RADIOTAP_VERSION;
329 rtap_vendor->rtap.rthdr.it_len = cpu_to_le16(rtap_len);
330 rtap_vendor->rtap.rthdr.it_present = cpu_to_le32(
331 (1 << IEEE80211_RADIOTAP_FLAGS) |
332 (1 << IEEE80211_RADIOTAP_CHANNEL) |
333 (1 << IEEE80211_RADIOTAP_MCS));
334 if (d->dma.status & RX_DMA_STATUS_ERROR)
335 rtap_vendor->rtap.flags |= IEEE80211_RADIOTAP_F_BADFCS;
337 rtap_vendor->rtap.chnl_freq = cpu_to_le16(ch ? ch->center_freq : 58320);
338 rtap_vendor->rtap.chnl_flags = cpu_to_le16(0);
340 rtap_vendor->rtap.mcs_present = IEEE80211_RADIOTAP_MCS_HAVE_MCS;
341 rtap_vendor->rtap.mcs_flags = 0;
342 rtap_vendor->rtap.mcs_index = wil_rxdesc_mcs(d);
344 if (rtap_include_phy_info) {
345 rtap_vendor->rtap.rthdr.it_present |= cpu_to_le32(1 <<
346 IEEE80211_RADIOTAP_VENDOR_NAMESPACE);
347 /* OUI for Wilocity 04:ce:14 */
348 rtap_vendor->vendor_oui[0] = 0x04;
349 rtap_vendor->vendor_oui[1] = 0xce;
350 rtap_vendor->vendor_oui[2] = 0x14;
351 rtap_vendor->vendor_ns = 1;
352 /* Rx descriptor + PHY data */
353 rtap_vendor->vendor_skip = cpu_to_le16(sizeof(*d) +
355 memcpy(rtap_vendor->vendor_data, (void *)d, sizeof(*d));
356 memcpy(rtap_vendor->vendor_data + sizeof(*d), phy_data,
362 * reap 1 frame from @swhead
364 * Rx descriptor copied to skb->cb
366 * Safe to call from IRQ
368 static struct sk_buff *wil_vring_reap_rx(struct wil6210_priv *wil,
371 struct device *dev = wil_to_dev(wil);
372 struct net_device *ndev = wil_to_ndev(wil);
373 volatile struct vring_rx_desc *_d;
374 struct vring_rx_desc *d;
377 unsigned int snaplen = wil_rx_snaplen();
378 unsigned int sz = mtu_max + ETH_HLEN + snaplen;
382 int i = (int)vring->swhead;
383 struct wil_net_stats *stats;
385 BUILD_BUG_ON(sizeof(struct vring_rx_desc) > sizeof(skb->cb));
387 if (unlikely(wil_vring_is_empty(vring)))
390 _d = &vring->va[i].rx;
391 if (unlikely(!(_d->dma.status & RX_DMA_STATUS_DU))) {
392 /* it is not error, we just reached end of Rx done area */
396 skb = vring->ctx[i].skb;
397 vring->ctx[i].skb = NULL;
398 wil_vring_advance_head(vring, 1);
400 wil_err(wil, "No Rx skb at [%d]\n", i);
403 d = wil_skb_rxdesc(skb);
405 pa = wil_desc_addr(&d->dma.addr);
407 dma_unmap_single(dev, pa, sz, DMA_FROM_DEVICE);
408 dmalen = le16_to_cpu(d->dma.length);
410 trace_wil6210_rx(i, d);
411 wil_dbg_txrx(wil, "Rx[%3d] : %d bytes\n", i, dmalen);
412 wil_hex_dump_txrx("Rx ", DUMP_PREFIX_NONE, 32, 4,
413 (const void *)d, sizeof(*d), false);
415 if (unlikely(dmalen > sz)) {
416 wil_err(wil, "Rx size too large: %d bytes!\n", dmalen);
420 skb_trim(skb, dmalen);
424 wil_hex_dump_txrx("Rx ", DUMP_PREFIX_OFFSET, 16, 1,
425 skb->data, skb_headlen(skb), false);
427 cid = wil_rxdesc_cid(d);
428 stats = &wil->sta[cid].stats;
429 stats->last_mcs_rx = wil_rxdesc_mcs(d);
430 if (stats->last_mcs_rx < ARRAY_SIZE(stats->rx_per_mcs))
431 stats->rx_per_mcs[stats->last_mcs_rx]++;
433 /* use radiotap header only if required */
434 if (ndev->type == ARPHRD_IEEE80211_RADIOTAP)
435 wil_rx_add_radiotap_header(wil, skb);
437 /* no extra checks if in sniffer mode */
438 if (ndev->type != ARPHRD_ETHER)
441 * Non-data frames may be delivered through Rx DMA channel (ex: BAR)
442 * Driver should recognize it by frame type, that is found
443 * in Rx descriptor. If type is not data, it is 802.11 frame as is
445 ftype = wil_rxdesc_ftype(d) << 2;
446 if (unlikely(ftype != IEEE80211_FTYPE_DATA)) {
447 wil_dbg_txrx(wil, "Non-data frame ftype 0x%08x\n", ftype);
448 /* TODO: process it */
453 if (unlikely(skb->len < ETH_HLEN + snaplen)) {
454 wil_err(wil, "Short frame, len = %d\n", skb->len);
455 /* TODO: process it (i.e. BAR) */
460 /* L4 IDENT is on when HW calculated checksum, check status
461 * and in case of error drop the packet
462 * higher stack layers will handle retransmission (if required)
464 if (likely(d->dma.status & RX_DMA_STATUS_L4I)) {
465 /* L4 protocol identified, csum calculated */
466 if (likely((d->dma.error & RX_DMA_ERROR_L4_ERR) == 0))
467 skb->ip_summed = CHECKSUM_UNNECESSARY;
468 /* If HW reports bad checksum, let IP stack re-check it
469 * For example, HW don't understand Microsoft IP stack that
470 * mis-calculates TCP checksum - if it should be 0x0,
471 * it writes 0xffff in violation of RFC 1624
477 * +-------+-------+---------+------------+------+
478 * | SA(6) | DA(6) | SNAP(6) | ETHTYPE(2) | DATA |
479 * +-------+-------+---------+------------+------+
480 * Need to remove SNAP, shifting SA and DA forward
482 memmove(skb->data + snaplen, skb->data, 2 * ETH_ALEN);
483 skb_pull(skb, snaplen);
490 * allocate and fill up to @count buffers in rx ring
491 * buffers posted at @swtail
493 static int wil_rx_refill(struct wil6210_priv *wil, int count)
495 struct net_device *ndev = wil_to_ndev(wil);
496 struct vring *v = &wil->vring_rx;
499 int headroom = ndev->type == ARPHRD_IEEE80211_RADIOTAP ?
500 WIL6210_RTAP_SIZE : 0;
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);
507 wil_err(wil, "Error %d in wil_rx_refill[%d]\n",
512 iowrite32(v->swtail, wil->csr + HOSTADDR(v->hwtail));
518 * Pass Rx packet to the netif. Update statistics.
519 * Called in softirq context (NAPI poll).
521 void wil_netif_rx_any(struct sk_buff *skb, struct net_device *ndev)
523 gro_result_t rc = GRO_NORMAL;
524 struct wil6210_priv *wil = ndev_to_wil(ndev);
525 struct wireless_dev *wdev = wil_to_wdev(wil);
526 unsigned int len = skb->len;
527 struct vring_rx_desc *d = wil_skb_rxdesc(skb);
528 int cid = wil_rxdesc_cid(d); /* always 0..7, no need to check */
529 struct ethhdr *eth = (void *)skb->data;
530 /* here looking for DA, not A1, thus Rxdesc's 'mcast' indication
531 * is not suitable, need to look at data
533 int mcast = is_multicast_ether_addr(eth->h_dest);
534 struct wil_net_stats *stats = &wil->sta[cid].stats;
535 struct sk_buff *xmit_skb = NULL;
536 static const char * const gro_res_str[] = {
537 [GRO_MERGED] = "GRO_MERGED",
538 [GRO_MERGED_FREE] = "GRO_MERGED_FREE",
539 [GRO_HELD] = "GRO_HELD",
540 [GRO_NORMAL] = "GRO_NORMAL",
541 [GRO_DROP] = "GRO_DROP",
546 if (wdev->iftype == NL80211_IFTYPE_AP && !wil->ap_isolate) {
548 /* send multicast frames both to higher layers in
549 * local net stack and back to the wireless medium
551 xmit_skb = skb_copy(skb, GFP_ATOMIC);
553 int xmit_cid = wil_find_cid(wil, eth->h_dest);
556 /* The destination station is associated to
557 * this AP (in this VLAN), so send the frame
558 * directly to it and do not pass it to local
567 /* Send to wireless media and increase priority by 256 to
568 * keep the received priority instead of reclassifying
569 * the frame (see cfg80211_classify8021d).
571 xmit_skb->dev = ndev;
572 xmit_skb->priority += 256;
573 xmit_skb->protocol = htons(ETH_P_802_3);
574 skb_reset_network_header(xmit_skb);
575 skb_reset_mac_header(xmit_skb);
576 wil_dbg_txrx(wil, "Rx -> Tx %d bytes\n", len);
577 dev_queue_xmit(xmit_skb);
580 if (skb) { /* deliver to local stack */
582 skb->protocol = eth_type_trans(skb, ndev);
583 rc = napi_gro_receive(&wil->napi_rx, skb);
584 wil_dbg_txrx(wil, "Rx complete %d bytes => %s\n",
585 len, gro_res_str[rc]);
587 /* statistics. rc set to GRO_NORMAL for AP bridging */
588 if (unlikely(rc == GRO_DROP)) {
589 ndev->stats.rx_dropped++;
591 wil_dbg_txrx(wil, "Rx drop %d bytes\n", len);
593 ndev->stats.rx_packets++;
595 ndev->stats.rx_bytes += len;
596 stats->rx_bytes += len;
598 ndev->stats.multicast++;
603 * Proceed all completed skb's from Rx VRING
605 * Safe to call from NAPI poll, i.e. softirq with interrupts enabled
607 void wil_rx_handle(struct wil6210_priv *wil, int *quota)
609 struct net_device *ndev = wil_to_ndev(wil);
610 struct vring *v = &wil->vring_rx;
613 if (unlikely(!v->va)) {
614 wil_err(wil, "Rx IRQ while Rx not yet initialized\n");
617 wil_dbg_txrx(wil, "%s()\n", __func__);
618 while ((*quota > 0) && (NULL != (skb = wil_vring_reap_rx(wil, v)))) {
621 if (wil->wdev->iftype == NL80211_IFTYPE_MONITOR) {
623 skb_reset_mac_header(skb);
624 skb->ip_summed = CHECKSUM_UNNECESSARY;
625 skb->pkt_type = PACKET_OTHERHOST;
626 skb->protocol = htons(ETH_P_802_2);
627 wil_netif_rx_any(skb, ndev);
629 wil_rx_reorder(wil, skb);
632 wil_rx_refill(wil, v->size);
635 int wil_rx_init(struct wil6210_priv *wil, u16 size)
637 struct vring *vring = &wil->vring_rx;
640 wil_dbg_misc(wil, "%s()\n", __func__);
643 wil_err(wil, "Rx ring already allocated\n");
648 rc = wil_vring_alloc(wil, vring);
652 rc = wmi_rx_chain_add(wil, vring);
656 rc = wil_rx_refill(wil, vring->size);
662 wil_vring_free(wil, vring, 0);
667 void wil_rx_fini(struct wil6210_priv *wil)
669 struct vring *vring = &wil->vring_rx;
671 wil_dbg_misc(wil, "%s()\n", __func__);
674 wil_vring_free(wil, vring, 0);
677 int wil_vring_init_tx(struct wil6210_priv *wil, int id, int size,
681 struct wmi_vring_cfg_cmd cmd = {
682 .action = cpu_to_le32(WMI_VRING_CMD_ADD),
686 cpu_to_le16(wil_mtu2macbuf(mtu_max)),
687 .ring_size = cpu_to_le16(size),
690 .cidxtid = mk_cidxtid(cid, tid),
691 .encap_trans_type = WMI_VRING_ENC_TYPE_802_3,
696 .priority = cpu_to_le16(0),
697 .timeslot_us = cpu_to_le16(0xfff),
702 struct wil6210_mbox_hdr_wmi wmi;
703 struct wmi_vring_cfg_done_event cmd;
705 struct vring *vring = &wil->vring_tx[id];
706 struct vring_tx_data *txdata = &wil->vring_tx_data[id];
708 wil_dbg_misc(wil, "%s() max_mpdu_size %d\n", __func__,
709 cmd.vring_cfg.tx_sw_ring.max_mpdu_size);
712 wil_err(wil, "Tx ring [%d] already allocated\n", id);
717 memset(txdata, 0, sizeof(*txdata));
718 spin_lock_init(&txdata->lock);
720 rc = wil_vring_alloc(wil, vring);
724 wil->vring2cid_tid[id][0] = cid;
725 wil->vring2cid_tid[id][1] = tid;
727 cmd.vring_cfg.tx_sw_ring.ring_mem_base = cpu_to_le64(vring->pa);
730 txdata->dot1x_open = true;
731 rc = wmi_call(wil, WMI_VRING_CFG_CMDID, &cmd, sizeof(cmd),
732 WMI_VRING_CFG_DONE_EVENTID, &reply, sizeof(reply), 100);
736 if (reply.cmd.status != WMI_FW_STATUS_SUCCESS) {
737 wil_err(wil, "Tx config failed, status 0x%02x\n",
742 vring->hwtail = le32_to_cpu(reply.cmd.tx_vring_tail_ptr);
745 if (txdata->dot1x_open && (agg_wsize >= 0))
746 wil_addba_tx_request(wil, id, agg_wsize);
750 txdata->dot1x_open = false;
752 wil_vring_free(wil, vring, 1);
758 int wil_vring_init_bcast(struct wil6210_priv *wil, int id, int size)
761 struct wmi_bcast_vring_cfg_cmd cmd = {
762 .action = cpu_to_le32(WMI_VRING_CMD_ADD),
766 cpu_to_le16(wil_mtu2macbuf(mtu_max)),
767 .ring_size = cpu_to_le16(size),
770 .encap_trans_type = WMI_VRING_ENC_TYPE_802_3,
774 struct wil6210_mbox_hdr_wmi wmi;
775 struct wmi_vring_cfg_done_event cmd;
777 struct vring *vring = &wil->vring_tx[id];
778 struct vring_tx_data *txdata = &wil->vring_tx_data[id];
780 wil_dbg_misc(wil, "%s() max_mpdu_size %d\n", __func__,
781 cmd.vring_cfg.tx_sw_ring.max_mpdu_size);
784 wil_err(wil, "Tx ring [%d] already allocated\n", id);
789 memset(txdata, 0, sizeof(*txdata));
790 spin_lock_init(&txdata->lock);
792 rc = wil_vring_alloc(wil, vring);
796 wil->vring2cid_tid[id][0] = WIL6210_MAX_CID; /* CID */
797 wil->vring2cid_tid[id][1] = 0; /* TID */
799 cmd.vring_cfg.tx_sw_ring.ring_mem_base = cpu_to_le64(vring->pa);
802 txdata->dot1x_open = true;
803 rc = wmi_call(wil, WMI_BCAST_VRING_CFG_CMDID, &cmd, sizeof(cmd),
804 WMI_VRING_CFG_DONE_EVENTID, &reply, sizeof(reply), 100);
808 if (reply.cmd.status != WMI_FW_STATUS_SUCCESS) {
809 wil_err(wil, "Tx config failed, status 0x%02x\n",
814 vring->hwtail = le32_to_cpu(reply.cmd.tx_vring_tail_ptr);
821 txdata->dot1x_open = false;
822 wil_vring_free(wil, vring, 1);
828 void wil_vring_fini_tx(struct wil6210_priv *wil, int id)
830 struct vring *vring = &wil->vring_tx[id];
831 struct vring_tx_data *txdata = &wil->vring_tx_data[id];
833 WARN_ON(!mutex_is_locked(&wil->mutex));
838 wil_dbg_misc(wil, "%s() id=%d\n", __func__, id);
840 spin_lock_bh(&txdata->lock);
841 txdata->dot1x_open = false;
842 txdata->enabled = 0; /* no Tx can be in progress or start anew */
843 spin_unlock_bh(&txdata->lock);
844 /* make sure NAPI won't touch this vring */
845 if (test_bit(wil_status_napi_en, wil->status))
846 napi_synchronize(&wil->napi_tx);
848 wil_vring_free(wil, vring, 1);
849 memset(txdata, 0, sizeof(*txdata));
852 static struct vring *wil_find_tx_ucast(struct wil6210_priv *wil,
856 struct ethhdr *eth = (void *)skb->data;
857 int cid = wil_find_cid(wil, eth->h_dest);
862 /* TODO: fix for multiple TID */
863 for (i = 0; i < ARRAY_SIZE(wil->vring2cid_tid); i++) {
864 if (!wil->vring_tx_data[i].dot1x_open &&
865 (skb->protocol != cpu_to_be16(ETH_P_PAE)))
867 if (wil->vring2cid_tid[i][0] == cid) {
868 struct vring *v = &wil->vring_tx[i];
870 wil_dbg_txrx(wil, "%s(%pM) -> [%d]\n",
871 __func__, eth->h_dest, i);
875 wil_dbg_txrx(wil, "vring[%d] not valid\n", i);
884 static int wil_tx_vring(struct wil6210_priv *wil, struct vring *vring,
885 struct sk_buff *skb);
887 static struct vring *wil_find_tx_vring_sta(struct wil6210_priv *wil,
894 /* In the STA mode, it is expected to have only 1 VRING
895 * for the AP we connected to.
896 * find 1-st vring eligible for this skb and use it.
898 for (i = 0; i < WIL6210_MAX_TX_RINGS; i++) {
899 v = &wil->vring_tx[i];
903 cid = wil->vring2cid_tid[i][0];
904 if (cid >= WIL6210_MAX_CID) /* skip BCAST */
907 if (!wil->vring_tx_data[i].dot1x_open &&
908 (skb->protocol != cpu_to_be16(ETH_P_PAE)))
911 wil_dbg_txrx(wil, "Tx -> ring %d\n", i);
916 wil_dbg_txrx(wil, "Tx while no vrings active?\n");
921 /* Use one of 2 strategies:
923 * 1. New (real broadcast):
924 * use dedicated broadcast vring
925 * 2. Old (pseudo-DMS):
926 * Find 1-st vring and return it;
927 * duplicate skb and send it to other active vrings;
928 * in all cases override dest address to unicast peer's address
929 * Use old strategy when new is not supported yet:
932 static struct vring *wil_find_tx_bcast_1(struct wil6210_priv *wil,
936 int i = wil->bcast_vring;
940 v = &wil->vring_tx[i];
943 if (!wil->vring_tx_data[i].dot1x_open &&
944 (skb->protocol != cpu_to_be16(ETH_P_PAE)))
950 static void wil_set_da_for_vring(struct wil6210_priv *wil,
951 struct sk_buff *skb, int vring_index)
953 struct ethhdr *eth = (void *)skb->data;
954 int cid = wil->vring2cid_tid[vring_index][0];
956 ether_addr_copy(eth->h_dest, wil->sta[cid].addr);
959 static struct vring *wil_find_tx_bcast_2(struct wil6210_priv *wil,
962 struct vring *v, *v2;
963 struct sk_buff *skb2;
966 struct ethhdr *eth = (void *)skb->data;
967 char *src = eth->h_source;
969 /* find 1-st vring eligible for data */
970 for (i = 0; i < WIL6210_MAX_TX_RINGS; i++) {
971 v = &wil->vring_tx[i];
975 cid = wil->vring2cid_tid[i][0];
976 if (cid >= WIL6210_MAX_CID) /* skip BCAST */
978 if (!wil->vring_tx_data[i].dot1x_open &&
979 (skb->protocol != cpu_to_be16(ETH_P_PAE)))
982 /* don't Tx back to source when re-routing Rx->Tx at the AP */
983 if (0 == memcmp(wil->sta[cid].addr, src, ETH_ALEN))
989 wil_dbg_txrx(wil, "Tx while no vrings active?\n");
994 wil_dbg_txrx(wil, "BCAST -> ring %d\n", i);
995 wil_set_da_for_vring(wil, skb, i);
997 /* find other active vrings and duplicate skb for each */
998 for (i++; i < WIL6210_MAX_TX_RINGS; i++) {
999 v2 = &wil->vring_tx[i];
1002 cid = wil->vring2cid_tid[i][0];
1003 if (cid >= WIL6210_MAX_CID) /* skip BCAST */
1005 if (!wil->vring_tx_data[i].dot1x_open &&
1006 (skb->protocol != cpu_to_be16(ETH_P_PAE)))
1009 if (0 == memcmp(wil->sta[cid].addr, src, ETH_ALEN))
1012 skb2 = skb_copy(skb, GFP_ATOMIC);
1014 wil_dbg_txrx(wil, "BCAST DUP -> ring %d\n", i);
1015 wil_set_da_for_vring(wil, skb2, i);
1016 wil_tx_vring(wil, v2, skb2);
1018 wil_err(wil, "skb_copy failed\n");
1025 static struct vring *wil_find_tx_bcast(struct wil6210_priv *wil,
1026 struct sk_buff *skb)
1028 struct wireless_dev *wdev = wil->wdev;
1030 if (wdev->iftype != NL80211_IFTYPE_AP)
1031 return wil_find_tx_bcast_2(wil, skb);
1033 return wil_find_tx_bcast_1(wil, skb);
1036 static int wil_tx_desc_map(struct vring_tx_desc *d, dma_addr_t pa, u32 len,
1039 wil_desc_addr_set(&d->dma.addr, pa);
1040 d->dma.ip_length = 0;
1041 /* 0..6: mac_length; 7:ip_version 0-IP6 1-IP4*/
1042 d->dma.b11 = 0/*14 | BIT(7)*/;
1044 d->dma.status = 0; /* BIT(0) should be 0 for HW_OWNED */
1045 d->dma.length = cpu_to_le16((u16)len);
1046 d->dma.d0 = (vring_index << DMA_CFG_DESC_TX_0_QID_POS);
1050 d->mac.ucode_cmd = 0;
1051 /* translation type: 0 - bypass; 1 - 802.3; 2 - native wifi */
1052 d->mac.d[2] = BIT(MAC_CFG_DESC_TX_2_SNAP_HDR_INSERTION_EN_POS) |
1053 (1 << MAC_CFG_DESC_TX_2_L2_TRANSLATION_TYPE_POS);
1059 void wil_tx_desc_set_nr_frags(struct vring_tx_desc *d, int nr_frags)
1061 d->mac.d[2] |= ((nr_frags + 1) <<
1062 MAC_CFG_DESC_TX_2_NUM_OF_DESCRIPTORS_POS);
1065 static int wil_tx_desc_offload_cksum_set(struct wil6210_priv *wil,
1066 struct vring_tx_desc *d,
1067 struct sk_buff *skb)
1071 if (skb->ip_summed != CHECKSUM_PARTIAL)
1074 d->dma.b11 = ETH_HLEN; /* MAC header length */
1076 switch (skb->protocol) {
1077 case cpu_to_be16(ETH_P_IP):
1078 protocol = ip_hdr(skb)->protocol;
1079 d->dma.b11 |= BIT(DMA_CFG_DESC_TX_OFFLOAD_CFG_L3T_IPV4_POS);
1081 case cpu_to_be16(ETH_P_IPV6):
1082 protocol = ipv6_hdr(skb)->nexthdr;
1090 d->dma.d0 |= (2 << DMA_CFG_DESC_TX_0_L4_TYPE_POS);
1091 /* L4 header len: TCP header length */
1093 (tcp_hdrlen(skb) & DMA_CFG_DESC_TX_0_L4_LENGTH_MSK);
1096 /* L4 header len: UDP header length */
1098 (sizeof(struct udphdr) & DMA_CFG_DESC_TX_0_L4_LENGTH_MSK);
1104 d->dma.ip_length = skb_network_header_len(skb);
1105 /* Enable TCP/UDP checksum */
1106 d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_TCP_UDP_CHECKSUM_EN_POS);
1107 /* Calculate pseudo-header */
1108 d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_PSEUDO_HEADER_CALC_EN_POS);
1113 static int __wil_tx_vring(struct wil6210_priv *wil, struct vring *vring,
1114 struct sk_buff *skb)
1116 struct device *dev = wil_to_dev(wil);
1117 struct vring_tx_desc dd, *d = ⅆ
1118 volatile struct vring_tx_desc *_d;
1119 u32 swhead = vring->swhead;
1120 int avail = wil_vring_avail_tx(vring);
1121 int nr_frags = skb_shinfo(skb)->nr_frags;
1123 int vring_index = vring - wil->vring_tx;
1124 struct vring_tx_data *txdata = &wil->vring_tx_data[vring_index];
1128 bool mcast = (vring_index == wil->bcast_vring);
1129 uint len = skb_headlen(skb);
1131 wil_dbg_txrx(wil, "%s()\n", __func__);
1133 if (unlikely(!txdata->enabled))
1136 if (unlikely(avail < 1 + nr_frags)) {
1137 wil_err_ratelimited(wil,
1138 "Tx ring[%2d] full. No space for %d fragments\n",
1139 vring_index, 1 + nr_frags);
1142 _d = &vring->va[i].tx;
1144 pa = dma_map_single(dev, skb->data, skb_headlen(skb), DMA_TO_DEVICE);
1146 wil_dbg_txrx(wil, "Tx[%2d] skb %d bytes 0x%p -> %pad\n", vring_index,
1147 skb_headlen(skb), skb->data, &pa);
1148 wil_hex_dump_txrx("Tx ", DUMP_PREFIX_OFFSET, 16, 1,
1149 skb->data, skb_headlen(skb), false);
1151 if (unlikely(dma_mapping_error(dev, pa)))
1153 vring->ctx[i].mapped_as = wil_mapped_as_single;
1155 wil_tx_desc_map(d, pa, len, vring_index);
1156 if (unlikely(mcast)) {
1157 d->mac.d[0] |= BIT(MAC_CFG_DESC_TX_0_MCS_EN_POS); /* MCS 0 */
1158 if (unlikely(len > WIL_BCAST_MCS0_LIMIT)) /* set MCS 1 */
1159 d->mac.d[0] |= (1 << MAC_CFG_DESC_TX_0_MCS_INDEX_POS);
1161 /* Process TCP/UDP checksum offloading */
1162 if (unlikely(wil_tx_desc_offload_cksum_set(wil, d, skb))) {
1163 wil_err(wil, "Tx[%2d] Failed to set cksum, drop packet\n",
1168 vring->ctx[i].nr_frags = nr_frags;
1169 wil_tx_desc_set_nr_frags(d, nr_frags);
1171 /* middle segments */
1172 for (; f < nr_frags; f++) {
1173 const struct skb_frag_struct *frag =
1174 &skb_shinfo(skb)->frags[f];
1175 int len = skb_frag_size(frag);
1178 wil_dbg_txrx(wil, "Tx[%2d] desc[%4d]\n", vring_index, i);
1179 wil_hex_dump_txrx("TxD ", DUMP_PREFIX_NONE, 32, 4,
1180 (const void *)d, sizeof(*d), false);
1181 i = (swhead + f + 1) % vring->size;
1182 _d = &vring->va[i].tx;
1183 pa = skb_frag_dma_map(dev, frag, 0, skb_frag_size(frag),
1185 if (unlikely(dma_mapping_error(dev, pa)))
1187 vring->ctx[i].mapped_as = wil_mapped_as_page;
1188 wil_tx_desc_map(d, pa, len, vring_index);
1189 /* no need to check return code -
1190 * if it succeeded for 1-st descriptor,
1191 * it will succeed here too
1193 wil_tx_desc_offload_cksum_set(wil, d, skb);
1195 /* for the last seg only */
1196 d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_CMD_EOP_POS);
1197 d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_CMD_MARK_WB_POS);
1198 d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_CMD_DMA_IT_POS);
1200 wil_dbg_txrx(wil, "Tx[%2d] desc[%4d]\n", vring_index, i);
1201 wil_hex_dump_txrx("TxD ", DUMP_PREFIX_NONE, 32, 4,
1202 (const void *)d, sizeof(*d), false);
1204 /* hold reference to skb
1205 * to prevent skb release before accounting
1206 * in case of immediate "tx done"
1208 vring->ctx[i].skb = skb_get(skb);
1210 /* performance monitoring */
1211 used = wil_vring_used_tx(vring);
1212 if (wil_val_in_range(vring_idle_trsh,
1213 used, used + nr_frags + 1)) {
1214 txdata->idle += get_cycles() - txdata->last_idle;
1215 wil_dbg_txrx(wil, "Ring[%2d] not idle %d -> %d\n",
1216 vring_index, used, used + nr_frags + 1);
1219 /* advance swhead */
1220 wil_vring_advance_head(vring, nr_frags + 1);
1221 wil_dbg_txrx(wil, "Tx[%2d] swhead %d -> %d\n", vring_index, swhead,
1223 trace_wil6210_tx(vring_index, swhead, skb->len, nr_frags);
1224 iowrite32(vring->swhead, wil->csr + HOSTADDR(vring->hwtail));
1228 /* unmap what we have mapped */
1229 nr_frags = f + 1; /* frags mapped + one for skb head */
1230 for (f = 0; f < nr_frags; f++) {
1231 struct wil_ctx *ctx;
1233 i = (swhead + f) % vring->size;
1234 ctx = &vring->ctx[i];
1235 _d = &vring->va[i].tx;
1237 _d->dma.status = TX_DMA_STATUS_DU;
1238 wil_txdesc_unmap(dev, d, ctx);
1241 dev_kfree_skb_any(ctx->skb);
1243 memset(ctx, 0, sizeof(*ctx));
1249 static int wil_tx_vring(struct wil6210_priv *wil, struct vring *vring,
1250 struct sk_buff *skb)
1252 int vring_index = vring - wil->vring_tx;
1253 struct vring_tx_data *txdata = &wil->vring_tx_data[vring_index];
1256 spin_lock(&txdata->lock);
1257 rc = __wil_tx_vring(wil, vring, skb);
1258 spin_unlock(&txdata->lock);
1262 netdev_tx_t wil_start_xmit(struct sk_buff *skb, struct net_device *ndev)
1264 struct wil6210_priv *wil = ndev_to_wil(ndev);
1265 struct ethhdr *eth = (void *)skb->data;
1266 bool bcast = is_multicast_ether_addr(eth->h_dest);
1267 struct vring *vring;
1268 static bool pr_once_fw;
1271 wil_dbg_txrx(wil, "%s()\n", __func__);
1272 if (unlikely(!test_bit(wil_status_fwready, wil->status))) {
1274 wil_err(wil, "FW not ready\n");
1279 if (unlikely(!test_bit(wil_status_fwconnected, wil->status))) {
1280 wil_err(wil, "FW not connected\n");
1283 if (unlikely(wil->wdev->iftype == NL80211_IFTYPE_MONITOR)) {
1284 wil_err(wil, "Xmit in monitor mode not supported\n");
1290 if (wil->wdev->iftype == NL80211_IFTYPE_STATION) {
1291 /* in STA mode (ESS), all to same VRING */
1292 vring = wil_find_tx_vring_sta(wil, skb);
1293 } else { /* direct communication, find matching VRING */
1294 vring = bcast ? wil_find_tx_bcast(wil, skb) :
1295 wil_find_tx_ucast(wil, skb);
1297 if (unlikely(!vring)) {
1298 wil_dbg_txrx(wil, "No Tx VRING found for %pM\n", eth->h_dest);
1301 /* set up vring entry */
1302 rc = wil_tx_vring(wil, vring, skb);
1304 /* do we still have enough room in the vring? */
1305 if (unlikely(wil_vring_avail_tx(vring) < wil_vring_wmark_low(vring))) {
1306 netif_tx_stop_all_queues(wil_to_ndev(wil));
1307 wil_dbg_txrx(wil, "netif_tx_stop : ring full\n");
1312 /* statistics will be updated on the tx_complete */
1313 dev_kfree_skb_any(skb);
1314 return NETDEV_TX_OK;
1316 return NETDEV_TX_BUSY;
1318 break; /* goto drop; */
1321 ndev->stats.tx_dropped++;
1322 dev_kfree_skb_any(skb);
1324 return NET_XMIT_DROP;
1327 static inline bool wil_need_txstat(struct sk_buff *skb)
1329 struct ethhdr *eth = (void *)skb->data;
1331 return is_unicast_ether_addr(eth->h_dest) && skb->sk &&
1332 (skb_shinfo(skb)->tx_flags & SKBTX_WIFI_STATUS);
1335 static inline void wil_consume_skb(struct sk_buff *skb, bool acked)
1337 if (unlikely(wil_need_txstat(skb)))
1338 skb_complete_wifi_ack(skb, acked);
1340 acked ? dev_consume_skb_any(skb) : dev_kfree_skb_any(skb);
1344 * Clean up transmitted skb's from the Tx VRING
1346 * Return number of descriptors cleared
1348 * Safe to call from IRQ
1350 int wil_tx_complete(struct wil6210_priv *wil, int ringid)
1352 struct net_device *ndev = wil_to_ndev(wil);
1353 struct device *dev = wil_to_dev(wil);
1354 struct vring *vring = &wil->vring_tx[ringid];
1355 struct vring_tx_data *txdata = &wil->vring_tx_data[ringid];
1357 int cid = wil->vring2cid_tid[ringid][0];
1358 struct wil_net_stats *stats = NULL;
1359 volatile struct vring_tx_desc *_d;
1360 int used_before_complete;
1363 if (unlikely(!vring->va)) {
1364 wil_err(wil, "Tx irq[%d]: vring not initialized\n", ringid);
1368 if (unlikely(!txdata->enabled)) {
1369 wil_info(wil, "Tx irq[%d]: vring disabled\n", ringid);
1373 wil_dbg_txrx(wil, "%s(%d)\n", __func__, ringid);
1375 used_before_complete = wil_vring_used_tx(vring);
1377 if (cid < WIL6210_MAX_CID)
1378 stats = &wil->sta[cid].stats;
1380 while (!wil_vring_is_empty(vring)) {
1382 struct wil_ctx *ctx = &vring->ctx[vring->swtail];
1384 * For the fragmented skb, HW will set DU bit only for the
1385 * last fragment. look for it
1387 int lf = (vring->swtail + ctx->nr_frags) % vring->size;
1388 /* TODO: check we are not past head */
1390 _d = &vring->va[lf].tx;
1391 if (unlikely(!(_d->dma.status & TX_DMA_STATUS_DU)))
1394 new_swtail = (lf + 1) % vring->size;
1395 while (vring->swtail != new_swtail) {
1396 struct vring_tx_desc dd, *d = ⅆ
1398 struct sk_buff *skb;
1400 ctx = &vring->ctx[vring->swtail];
1402 _d = &vring->va[vring->swtail].tx;
1406 dmalen = le16_to_cpu(d->dma.length);
1407 trace_wil6210_tx_done(ringid, vring->swtail, dmalen,
1410 "TxC[%2d][%3d] : %d bytes, status 0x%02x err 0x%02x\n",
1411 ringid, vring->swtail, dmalen,
1412 d->dma.status, d->dma.error);
1413 wil_hex_dump_txrx("TxCD ", DUMP_PREFIX_NONE, 32, 4,
1414 (const void *)d, sizeof(*d), false);
1416 wil_txdesc_unmap(dev, d, ctx);
1419 if (likely(d->dma.error == 0)) {
1420 ndev->stats.tx_packets++;
1421 ndev->stats.tx_bytes += skb->len;
1423 stats->tx_packets++;
1424 stats->tx_bytes += skb->len;
1427 ndev->stats.tx_errors++;
1431 wil_consume_skb(skb, d->dma.error == 0);
1433 memset(ctx, 0, sizeof(*ctx));
1434 /* There is no need to touch HW descriptor:
1435 * - ststus bit TX_DMA_STATUS_DU is set by design,
1436 * so hardware will not try to process this desc.,
1437 * - rest of descriptor will be initialized on Tx.
1439 vring->swtail = wil_vring_next_tail(vring);
1444 /* performance monitoring */
1445 used_new = wil_vring_used_tx(vring);
1446 if (wil_val_in_range(vring_idle_trsh,
1447 used_new, used_before_complete)) {
1448 wil_dbg_txrx(wil, "Ring[%2d] idle %d -> %d\n",
1449 ringid, used_before_complete, used_new);
1450 txdata->last_idle = get_cycles();
1453 if (wil_vring_avail_tx(vring) > wil_vring_wmark_high(vring)) {
1454 wil_dbg_txrx(wil, "netif_tx_wake : ring not full\n");
1455 netif_tx_wake_all_queues(wil_to_ndev(wil));