2 * Copyright (c) 2012-2014 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 = BIT(9) | 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);
431 /* use radiotap header only if required */
432 if (ndev->type == ARPHRD_IEEE80211_RADIOTAP)
433 wil_rx_add_radiotap_header(wil, skb);
435 /* no extra checks if in sniffer mode */
436 if (ndev->type != ARPHRD_ETHER)
439 * Non-data frames may be delivered through Rx DMA channel (ex: BAR)
440 * Driver should recognize it by frame type, that is found
441 * in Rx descriptor. If type is not data, it is 802.11 frame as is
443 ftype = wil_rxdesc_ftype(d) << 2;
444 if (unlikely(ftype != IEEE80211_FTYPE_DATA)) {
445 wil_dbg_txrx(wil, "Non-data frame ftype 0x%08x\n", ftype);
446 /* TODO: process it */
451 if (unlikely(skb->len < ETH_HLEN + snaplen)) {
452 wil_err(wil, "Short frame, len = %d\n", skb->len);
453 /* TODO: process it (i.e. BAR) */
458 /* L4 IDENT is on when HW calculated checksum, check status
459 * and in case of error drop the packet
460 * higher stack layers will handle retransmission (if required)
462 if (likely(d->dma.status & RX_DMA_STATUS_L4I)) {
463 /* L4 protocol identified, csum calculated */
464 if (likely((d->dma.error & RX_DMA_ERROR_L4_ERR) == 0))
465 skb->ip_summed = CHECKSUM_UNNECESSARY;
466 /* If HW reports bad checksum, let IP stack re-check it
467 * For example, HW don't understand Microsoft IP stack that
468 * mis-calculates TCP checksum - if it should be 0x0,
469 * it writes 0xffff in violation of RFC 1624
475 * +-------+-------+---------+------------+------+
476 * | SA(6) | DA(6) | SNAP(6) | ETHTYPE(2) | DATA |
477 * +-------+-------+---------+------------+------+
478 * Need to remove SNAP, shifting SA and DA forward
480 memmove(skb->data + snaplen, skb->data, 2 * ETH_ALEN);
481 skb_pull(skb, snaplen);
488 * allocate and fill up to @count buffers in rx ring
489 * buffers posted at @swtail
491 static int wil_rx_refill(struct wil6210_priv *wil, int count)
493 struct net_device *ndev = wil_to_ndev(wil);
494 struct vring *v = &wil->vring_rx;
497 int headroom = ndev->type == ARPHRD_IEEE80211_RADIOTAP ?
498 WIL6210_RTAP_SIZE : 0;
500 for (; next_tail = wil_vring_next_tail(v),
501 (next_tail != v->swhead) && (count-- > 0);
502 v->swtail = next_tail) {
503 rc = wil_vring_alloc_skb(wil, v, v->swtail, headroom);
505 wil_err(wil, "Error %d in wil_rx_refill[%d]\n",
510 iowrite32(v->swtail, wil->csr + HOSTADDR(v->hwtail));
516 * Pass Rx packet to the netif. Update statistics.
517 * Called in softirq context (NAPI poll).
519 void wil_netif_rx_any(struct sk_buff *skb, struct net_device *ndev)
521 gro_result_t rc = GRO_NORMAL;
522 struct wil6210_priv *wil = ndev_to_wil(ndev);
523 struct wireless_dev *wdev = wil_to_wdev(wil);
524 unsigned int len = skb->len;
525 struct vring_rx_desc *d = wil_skb_rxdesc(skb);
526 int cid = wil_rxdesc_cid(d); /* always 0..7, no need to check */
527 struct ethhdr *eth = (void *)skb->data;
528 /* here looking for DA, not A1, thus Rxdesc's 'mcast' indication
529 * is not suitable, need to look at data
531 int mcast = is_multicast_ether_addr(eth->h_dest);
532 struct wil_net_stats *stats = &wil->sta[cid].stats;
533 struct sk_buff *xmit_skb = NULL;
534 static const char * const gro_res_str[] = {
535 [GRO_MERGED] = "GRO_MERGED",
536 [GRO_MERGED_FREE] = "GRO_MERGED_FREE",
537 [GRO_HELD] = "GRO_HELD",
538 [GRO_NORMAL] = "GRO_NORMAL",
539 [GRO_DROP] = "GRO_DROP",
544 if (wdev->iftype == NL80211_IFTYPE_AP && !wil->ap_isolate) {
546 /* send multicast frames both to higher layers in
547 * local net stack and back to the wireless medium
549 xmit_skb = skb_copy(skb, GFP_ATOMIC);
551 int xmit_cid = wil_find_cid(wil, eth->h_dest);
554 /* The destination station is associated to
555 * this AP (in this VLAN), so send the frame
556 * directly to it and do not pass it to local
565 /* Send to wireless media and increase priority by 256 to
566 * keep the received priority instead of reclassifying
567 * the frame (see cfg80211_classify8021d).
569 xmit_skb->dev = ndev;
570 xmit_skb->priority += 256;
571 xmit_skb->protocol = htons(ETH_P_802_3);
572 skb_reset_network_header(xmit_skb);
573 skb_reset_mac_header(xmit_skb);
574 wil_dbg_txrx(wil, "Rx -> Tx %d bytes\n", len);
575 dev_queue_xmit(xmit_skb);
578 if (skb) { /* deliver to local stack */
580 skb->protocol = eth_type_trans(skb, ndev);
581 rc = napi_gro_receive(&wil->napi_rx, skb);
582 wil_dbg_txrx(wil, "Rx complete %d bytes => %s\n",
583 len, gro_res_str[rc]);
585 /* statistics. rc set to GRO_NORMAL for AP bridging */
586 if (unlikely(rc == GRO_DROP)) {
587 ndev->stats.rx_dropped++;
589 wil_dbg_txrx(wil, "Rx drop %d bytes\n", len);
591 ndev->stats.rx_packets++;
593 ndev->stats.rx_bytes += len;
594 stats->rx_bytes += len;
596 ndev->stats.multicast++;
601 * Proceed all completed skb's from Rx VRING
603 * Safe to call from NAPI poll, i.e. softirq with interrupts enabled
605 void wil_rx_handle(struct wil6210_priv *wil, int *quota)
607 struct net_device *ndev = wil_to_ndev(wil);
608 struct vring *v = &wil->vring_rx;
611 if (unlikely(!v->va)) {
612 wil_err(wil, "Rx IRQ while Rx not yet initialized\n");
615 wil_dbg_txrx(wil, "%s()\n", __func__);
616 while ((*quota > 0) && (NULL != (skb = wil_vring_reap_rx(wil, v)))) {
619 if (wil->wdev->iftype == NL80211_IFTYPE_MONITOR) {
621 skb_reset_mac_header(skb);
622 skb->ip_summed = CHECKSUM_UNNECESSARY;
623 skb->pkt_type = PACKET_OTHERHOST;
624 skb->protocol = htons(ETH_P_802_2);
625 wil_netif_rx_any(skb, ndev);
627 wil_rx_reorder(wil, skb);
630 wil_rx_refill(wil, v->size);
633 int wil_rx_init(struct wil6210_priv *wil, u16 size)
635 struct vring *vring = &wil->vring_rx;
638 wil_dbg_misc(wil, "%s()\n", __func__);
641 wil_err(wil, "Rx ring already allocated\n");
646 rc = wil_vring_alloc(wil, vring);
650 rc = wmi_rx_chain_add(wil, vring);
654 rc = wil_rx_refill(wil, vring->size);
660 wil_vring_free(wil, vring, 0);
665 void wil_rx_fini(struct wil6210_priv *wil)
667 struct vring *vring = &wil->vring_rx;
669 wil_dbg_misc(wil, "%s()\n", __func__);
672 wil_vring_free(wil, vring, 0);
675 int wil_vring_init_tx(struct wil6210_priv *wil, int id, int size,
679 struct wmi_vring_cfg_cmd cmd = {
680 .action = cpu_to_le32(WMI_VRING_CMD_ADD),
684 cpu_to_le16(wil_mtu2macbuf(mtu_max)),
685 .ring_size = cpu_to_le16(size),
688 .cidxtid = mk_cidxtid(cid, tid),
689 .encap_trans_type = WMI_VRING_ENC_TYPE_802_3,
694 .priority = cpu_to_le16(0),
695 .timeslot_us = cpu_to_le16(0xfff),
700 struct wil6210_mbox_hdr_wmi wmi;
701 struct wmi_vring_cfg_done_event cmd;
703 struct vring *vring = &wil->vring_tx[id];
704 struct vring_tx_data *txdata = &wil->vring_tx_data[id];
706 wil_dbg_misc(wil, "%s() max_mpdu_size %d\n", __func__,
707 cmd.vring_cfg.tx_sw_ring.max_mpdu_size);
710 wil_err(wil, "Tx ring [%d] already allocated\n", id);
715 memset(txdata, 0, sizeof(*txdata));
716 spin_lock_init(&txdata->lock);
718 rc = wil_vring_alloc(wil, vring);
722 wil->vring2cid_tid[id][0] = cid;
723 wil->vring2cid_tid[id][1] = tid;
725 cmd.vring_cfg.tx_sw_ring.ring_mem_base = cpu_to_le64(vring->pa);
727 rc = wmi_call(wil, WMI_VRING_CFG_CMDID, &cmd, sizeof(cmd),
728 WMI_VRING_CFG_DONE_EVENTID, &reply, sizeof(reply), 100);
732 if (reply.cmd.status != WMI_FW_STATUS_SUCCESS) {
733 wil_err(wil, "Tx config failed, status 0x%02x\n",
738 vring->hwtail = le32_to_cpu(reply.cmd.tx_vring_tail_ptr);
741 if (wil->sta[cid].data_port_open && (agg_wsize >= 0))
742 wil_addba_tx_request(wil, id, agg_wsize);
746 wil_vring_free(wil, vring, 1);
752 int wil_vring_init_bcast(struct wil6210_priv *wil, int id, int size)
755 struct wmi_bcast_vring_cfg_cmd cmd = {
756 .action = cpu_to_le32(WMI_VRING_CMD_ADD),
760 cpu_to_le16(wil_mtu2macbuf(mtu_max)),
761 .ring_size = cpu_to_le16(size),
764 .encap_trans_type = WMI_VRING_ENC_TYPE_802_3,
768 struct wil6210_mbox_hdr_wmi wmi;
769 struct wmi_vring_cfg_done_event cmd;
771 struct vring *vring = &wil->vring_tx[id];
772 struct vring_tx_data *txdata = &wil->vring_tx_data[id];
774 wil_dbg_misc(wil, "%s() max_mpdu_size %d\n", __func__,
775 cmd.vring_cfg.tx_sw_ring.max_mpdu_size);
778 wil_err(wil, "Tx ring [%d] already allocated\n", id);
783 memset(txdata, 0, sizeof(*txdata));
784 spin_lock_init(&txdata->lock);
786 rc = wil_vring_alloc(wil, vring);
790 wil->vring2cid_tid[id][0] = WIL6210_MAX_CID; /* CID */
791 wil->vring2cid_tid[id][1] = 0; /* TID */
793 cmd.vring_cfg.tx_sw_ring.ring_mem_base = cpu_to_le64(vring->pa);
795 rc = wmi_call(wil, WMI_BCAST_VRING_CFG_CMDID, &cmd, sizeof(cmd),
796 WMI_VRING_CFG_DONE_EVENTID, &reply, sizeof(reply), 100);
800 if (reply.cmd.status != WMI_FW_STATUS_SUCCESS) {
801 wil_err(wil, "Tx config failed, status 0x%02x\n",
806 vring->hwtail = le32_to_cpu(reply.cmd.tx_vring_tail_ptr);
812 wil_vring_free(wil, vring, 1);
818 void wil_vring_fini_tx(struct wil6210_priv *wil, int id)
820 struct vring *vring = &wil->vring_tx[id];
821 struct vring_tx_data *txdata = &wil->vring_tx_data[id];
823 WARN_ON(!mutex_is_locked(&wil->mutex));
828 wil_dbg_misc(wil, "%s() id=%d\n", __func__, id);
830 spin_lock_bh(&txdata->lock);
831 txdata->enabled = 0; /* no Tx can be in progress or start anew */
832 spin_unlock_bh(&txdata->lock);
833 /* make sure NAPI won't touch this vring */
834 if (test_bit(wil_status_napi_en, wil->status))
835 napi_synchronize(&wil->napi_tx);
837 wil_vring_free(wil, vring, 1);
838 memset(txdata, 0, sizeof(*txdata));
841 static struct vring *wil_find_tx_ucast(struct wil6210_priv *wil,
845 struct ethhdr *eth = (void *)skb->data;
846 int cid = wil_find_cid(wil, eth->h_dest);
851 if (!wil->sta[cid].data_port_open &&
852 (skb->protocol != cpu_to_be16(ETH_P_PAE)))
855 /* TODO: fix for multiple TID */
856 for (i = 0; i < ARRAY_SIZE(wil->vring2cid_tid); i++) {
857 if (wil->vring2cid_tid[i][0] == cid) {
858 struct vring *v = &wil->vring_tx[i];
860 wil_dbg_txrx(wil, "%s(%pM) -> [%d]\n",
861 __func__, eth->h_dest, i);
865 wil_dbg_txrx(wil, "vring[%d] not valid\n", i);
874 static int wil_tx_vring(struct wil6210_priv *wil, struct vring *vring,
875 struct sk_buff *skb);
877 static struct vring *wil_find_tx_vring_sta(struct wil6210_priv *wil,
884 /* In the STA mode, it is expected to have only 1 VRING
885 * for the AP we connected to.
886 * find 1-st vring and see whether it is eligible for data
888 for (i = 0; i < WIL6210_MAX_TX_RINGS; i++) {
889 v = &wil->vring_tx[i];
893 cid = wil->vring2cid_tid[i][0];
894 if (cid >= WIL6210_MAX_CID) /* skip BCAST */
897 if (!wil->sta[cid].data_port_open &&
898 (skb->protocol != cpu_to_be16(ETH_P_PAE)))
901 wil_dbg_txrx(wil, "Tx -> ring %d\n", i);
906 wil_dbg_txrx(wil, "Tx while no vrings active?\n");
911 /* Use one of 2 strategies:
913 * 1. New (real broadcast):
914 * use dedicated broadcast vring
915 * 2. Old (pseudo-DMS):
916 * Find 1-st vring and return it;
917 * duplicate skb and send it to other active vrings;
918 * in all cases override dest address to unicast peer's address
919 * Use old strategy when new is not supported yet:
923 static struct vring *wil_find_tx_bcast_1(struct wil6210_priv *wil,
927 int i = wil->bcast_vring;
931 v = &wil->vring_tx[i];
938 static void wil_set_da_for_vring(struct wil6210_priv *wil,
939 struct sk_buff *skb, int vring_index)
941 struct ethhdr *eth = (void *)skb->data;
942 int cid = wil->vring2cid_tid[vring_index][0];
944 ether_addr_copy(eth->h_dest, wil->sta[cid].addr);
947 static struct vring *wil_find_tx_bcast_2(struct wil6210_priv *wil,
950 struct vring *v, *v2;
951 struct sk_buff *skb2;
954 struct ethhdr *eth = (void *)skb->data;
955 char *src = eth->h_source;
957 /* find 1-st vring eligible for data */
958 for (i = 0; i < WIL6210_MAX_TX_RINGS; i++) {
959 v = &wil->vring_tx[i];
963 cid = wil->vring2cid_tid[i][0];
964 if (cid >= WIL6210_MAX_CID) /* skip BCAST */
966 if (!wil->sta[cid].data_port_open)
969 /* don't Tx back to source when re-routing Rx->Tx at the AP */
970 if (0 == memcmp(wil->sta[cid].addr, src, ETH_ALEN))
976 wil_dbg_txrx(wil, "Tx while no vrings active?\n");
981 wil_dbg_txrx(wil, "BCAST -> ring %d\n", i);
982 wil_set_da_for_vring(wil, skb, i);
984 /* find other active vrings and duplicate skb for each */
985 for (i++; i < WIL6210_MAX_TX_RINGS; i++) {
986 v2 = &wil->vring_tx[i];
989 cid = wil->vring2cid_tid[i][0];
990 if (cid >= WIL6210_MAX_CID) /* skip BCAST */
992 if (!wil->sta[cid].data_port_open)
995 if (0 == memcmp(wil->sta[cid].addr, src, ETH_ALEN))
998 skb2 = skb_copy(skb, GFP_ATOMIC);
1000 wil_dbg_txrx(wil, "BCAST DUP -> ring %d\n", i);
1001 wil_set_da_for_vring(wil, skb2, i);
1002 wil_tx_vring(wil, v2, skb2);
1004 wil_err(wil, "skb_copy failed\n");
1011 static struct vring *wil_find_tx_bcast(struct wil6210_priv *wil,
1012 struct sk_buff *skb)
1014 struct wireless_dev *wdev = wil->wdev;
1016 if (wdev->iftype != NL80211_IFTYPE_AP)
1017 return wil_find_tx_bcast_2(wil, skb);
1020 return wil_find_tx_bcast_2(wil, skb);
1022 return wil_find_tx_bcast_1(wil, skb);
1025 static int wil_tx_desc_map(struct vring_tx_desc *d, dma_addr_t pa, u32 len,
1028 wil_desc_addr_set(&d->dma.addr, pa);
1029 d->dma.ip_length = 0;
1030 /* 0..6: mac_length; 7:ip_version 0-IP6 1-IP4*/
1031 d->dma.b11 = 0/*14 | BIT(7)*/;
1033 d->dma.status = 0; /* BIT(0) should be 0 for HW_OWNED */
1034 d->dma.length = cpu_to_le16((u16)len);
1035 d->dma.d0 = (vring_index << DMA_CFG_DESC_TX_0_QID_POS);
1039 d->mac.ucode_cmd = 0;
1040 /* translation type: 0 - bypass; 1 - 802.3; 2 - native wifi */
1041 d->mac.d[2] = BIT(MAC_CFG_DESC_TX_2_SNAP_HDR_INSERTION_EN_POS) |
1042 (1 << MAC_CFG_DESC_TX_2_L2_TRANSLATION_TYPE_POS);
1048 void wil_tx_desc_set_nr_frags(struct vring_tx_desc *d, int nr_frags)
1050 d->mac.d[2] |= ((nr_frags + 1) <<
1051 MAC_CFG_DESC_TX_2_NUM_OF_DESCRIPTORS_POS);
1054 static int wil_tx_desc_offload_cksum_set(struct wil6210_priv *wil,
1055 struct vring_tx_desc *d,
1056 struct sk_buff *skb)
1060 if (skb->ip_summed != CHECKSUM_PARTIAL)
1063 d->dma.b11 = ETH_HLEN; /* MAC header length */
1065 switch (skb->protocol) {
1066 case cpu_to_be16(ETH_P_IP):
1067 protocol = ip_hdr(skb)->protocol;
1068 d->dma.b11 |= BIT(DMA_CFG_DESC_TX_OFFLOAD_CFG_L3T_IPV4_POS);
1070 case cpu_to_be16(ETH_P_IPV6):
1071 protocol = ipv6_hdr(skb)->nexthdr;
1079 d->dma.d0 |= (2 << DMA_CFG_DESC_TX_0_L4_TYPE_POS);
1080 /* L4 header len: TCP header length */
1082 (tcp_hdrlen(skb) & DMA_CFG_DESC_TX_0_L4_LENGTH_MSK);
1085 /* L4 header len: UDP header length */
1087 (sizeof(struct udphdr) & DMA_CFG_DESC_TX_0_L4_LENGTH_MSK);
1093 d->dma.ip_length = skb_network_header_len(skb);
1094 /* Enable TCP/UDP checksum */
1095 d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_TCP_UDP_CHECKSUM_EN_POS);
1096 /* Calculate pseudo-header */
1097 d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_PSEUDO_HEADER_CALC_EN_POS);
1102 static int __wil_tx_vring(struct wil6210_priv *wil, struct vring *vring,
1103 struct sk_buff *skb)
1105 struct device *dev = wil_to_dev(wil);
1106 struct vring_tx_desc dd, *d = ⅆ
1107 volatile struct vring_tx_desc *_d;
1108 u32 swhead = vring->swhead;
1109 int avail = wil_vring_avail_tx(vring);
1110 int nr_frags = skb_shinfo(skb)->nr_frags;
1112 int vring_index = vring - wil->vring_tx;
1113 struct vring_tx_data *txdata = &wil->vring_tx_data[vring_index];
1117 bool mcast = (vring_index == wil->bcast_vring);
1118 uint len = skb_headlen(skb);
1120 wil_dbg_txrx(wil, "%s()\n", __func__);
1122 if (unlikely(!txdata->enabled))
1125 if (unlikely(avail < 1 + nr_frags)) {
1126 wil_err_ratelimited(wil,
1127 "Tx ring[%2d] full. No space for %d fragments\n",
1128 vring_index, 1 + nr_frags);
1131 _d = &vring->va[i].tx;
1133 pa = dma_map_single(dev, skb->data, skb_headlen(skb), DMA_TO_DEVICE);
1135 wil_dbg_txrx(wil, "Tx[%2d] skb %d bytes 0x%p -> %pad\n", vring_index,
1136 skb_headlen(skb), skb->data, &pa);
1137 wil_hex_dump_txrx("Tx ", DUMP_PREFIX_OFFSET, 16, 1,
1138 skb->data, skb_headlen(skb), false);
1140 if (unlikely(dma_mapping_error(dev, pa)))
1142 vring->ctx[i].mapped_as = wil_mapped_as_single;
1144 wil_tx_desc_map(d, pa, len, vring_index);
1145 if (unlikely(mcast)) {
1146 d->mac.d[0] |= BIT(MAC_CFG_DESC_TX_0_MCS_EN_POS); /* MCS 0 */
1147 if (unlikely(len > WIL_BCAST_MCS0_LIMIT)) {
1149 d->mac.d[0] |= (1 << MAC_CFG_DESC_TX_0_MCS_INDEX_POS);
1151 d->mac.d[1] |= BIT(MAC_CFG_DESC_TX_1_PKT_MODE_EN_POS) |
1152 (2 << MAC_CFG_DESC_TX_1_PKT_MODE_POS);
1155 /* Process TCP/UDP checksum offloading */
1156 if (unlikely(wil_tx_desc_offload_cksum_set(wil, d, skb))) {
1157 wil_err(wil, "Tx[%2d] Failed to set cksum, drop packet\n",
1162 vring->ctx[i].nr_frags = nr_frags;
1163 wil_tx_desc_set_nr_frags(d, nr_frags);
1165 /* middle segments */
1166 for (; f < nr_frags; f++) {
1167 const struct skb_frag_struct *frag =
1168 &skb_shinfo(skb)->frags[f];
1169 int len = skb_frag_size(frag);
1172 wil_dbg_txrx(wil, "Tx[%2d] desc[%4d]\n", vring_index, i);
1173 wil_hex_dump_txrx("TxD ", DUMP_PREFIX_NONE, 32, 4,
1174 (const void *)d, sizeof(*d), false);
1175 i = (swhead + f + 1) % vring->size;
1176 _d = &vring->va[i].tx;
1177 pa = skb_frag_dma_map(dev, frag, 0, skb_frag_size(frag),
1179 if (unlikely(dma_mapping_error(dev, pa)))
1181 vring->ctx[i].mapped_as = wil_mapped_as_page;
1182 wil_tx_desc_map(d, pa, len, vring_index);
1183 /* no need to check return code -
1184 * if it succeeded for 1-st descriptor,
1185 * it will succeed here too
1187 wil_tx_desc_offload_cksum_set(wil, d, skb);
1189 /* for the last seg only */
1190 d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_CMD_EOP_POS);
1191 d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_CMD_MARK_WB_POS);
1192 d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_CMD_DMA_IT_POS);
1194 wil_dbg_txrx(wil, "Tx[%2d] desc[%4d]\n", vring_index, i);
1195 wil_hex_dump_txrx("TxD ", DUMP_PREFIX_NONE, 32, 4,
1196 (const void *)d, sizeof(*d), false);
1198 /* hold reference to skb
1199 * to prevent skb release before accounting
1200 * in case of immediate "tx done"
1202 vring->ctx[i].skb = skb_get(skb);
1204 /* performance monitoring */
1205 used = wil_vring_used_tx(vring);
1206 if (wil_val_in_range(vring_idle_trsh,
1207 used, used + nr_frags + 1)) {
1208 txdata->idle += get_cycles() - txdata->last_idle;
1209 wil_dbg_txrx(wil, "Ring[%2d] not idle %d -> %d\n",
1210 vring_index, used, used + nr_frags + 1);
1213 /* advance swhead */
1214 wil_vring_advance_head(vring, nr_frags + 1);
1215 wil_dbg_txrx(wil, "Tx[%2d] swhead %d -> %d\n", vring_index, swhead,
1217 trace_wil6210_tx(vring_index, swhead, skb->len, nr_frags);
1218 iowrite32(vring->swhead, wil->csr + HOSTADDR(vring->hwtail));
1222 /* unmap what we have mapped */
1223 nr_frags = f + 1; /* frags mapped + one for skb head */
1224 for (f = 0; f < nr_frags; f++) {
1225 struct wil_ctx *ctx;
1227 i = (swhead + f) % vring->size;
1228 ctx = &vring->ctx[i];
1229 _d = &vring->va[i].tx;
1231 _d->dma.status = TX_DMA_STATUS_DU;
1232 wil_txdesc_unmap(dev, d, ctx);
1235 dev_kfree_skb_any(ctx->skb);
1237 memset(ctx, 0, sizeof(*ctx));
1243 static int wil_tx_vring(struct wil6210_priv *wil, struct vring *vring,
1244 struct sk_buff *skb)
1246 int vring_index = vring - wil->vring_tx;
1247 struct vring_tx_data *txdata = &wil->vring_tx_data[vring_index];
1250 spin_lock(&txdata->lock);
1251 rc = __wil_tx_vring(wil, vring, skb);
1252 spin_unlock(&txdata->lock);
1256 netdev_tx_t wil_start_xmit(struct sk_buff *skb, struct net_device *ndev)
1258 struct wil6210_priv *wil = ndev_to_wil(ndev);
1259 struct ethhdr *eth = (void *)skb->data;
1260 bool bcast = is_multicast_ether_addr(eth->h_dest);
1261 struct vring *vring;
1262 static bool pr_once_fw;
1265 wil_dbg_txrx(wil, "%s()\n", __func__);
1266 if (unlikely(!test_bit(wil_status_fwready, wil->status))) {
1268 wil_err(wil, "FW not ready\n");
1273 if (unlikely(!test_bit(wil_status_fwconnected, wil->status))) {
1274 wil_err(wil, "FW not connected\n");
1277 if (unlikely(wil->wdev->iftype == NL80211_IFTYPE_MONITOR)) {
1278 wil_err(wil, "Xmit in monitor mode not supported\n");
1284 if (wil->wdev->iftype == NL80211_IFTYPE_STATION) {
1285 /* in STA mode (ESS), all to same VRING */
1286 vring = wil_find_tx_vring_sta(wil, skb);
1287 } else { /* direct communication, find matching VRING */
1288 vring = bcast ? wil_find_tx_bcast(wil, skb) :
1289 wil_find_tx_ucast(wil, skb);
1291 if (unlikely(!vring)) {
1292 wil_dbg_txrx(wil, "No Tx VRING found for %pM\n", eth->h_dest);
1295 /* set up vring entry */
1296 rc = wil_tx_vring(wil, vring, skb);
1298 /* do we still have enough room in the vring? */
1299 if (unlikely(wil_vring_avail_tx(vring) < wil_vring_wmark_low(vring))) {
1300 netif_tx_stop_all_queues(wil_to_ndev(wil));
1301 wil_dbg_txrx(wil, "netif_tx_stop : ring full\n");
1306 /* statistics will be updated on the tx_complete */
1307 dev_kfree_skb_any(skb);
1308 return NETDEV_TX_OK;
1310 return NETDEV_TX_BUSY;
1312 break; /* goto drop; */
1315 ndev->stats.tx_dropped++;
1316 dev_kfree_skb_any(skb);
1318 return NET_XMIT_DROP;
1321 static inline bool wil_need_txstat(struct sk_buff *skb)
1323 struct ethhdr *eth = (void *)skb->data;
1325 return is_unicast_ether_addr(eth->h_dest) && skb->sk &&
1326 (skb_shinfo(skb)->tx_flags & SKBTX_WIFI_STATUS);
1329 static inline void wil_consume_skb(struct sk_buff *skb, bool acked)
1331 if (unlikely(wil_need_txstat(skb)))
1332 skb_complete_wifi_ack(skb, acked);
1334 acked ? dev_consume_skb_any(skb) : dev_kfree_skb_any(skb);
1338 * Clean up transmitted skb's from the Tx VRING
1340 * Return number of descriptors cleared
1342 * Safe to call from IRQ
1344 int wil_tx_complete(struct wil6210_priv *wil, int ringid)
1346 struct net_device *ndev = wil_to_ndev(wil);
1347 struct device *dev = wil_to_dev(wil);
1348 struct vring *vring = &wil->vring_tx[ringid];
1349 struct vring_tx_data *txdata = &wil->vring_tx_data[ringid];
1351 int cid = wil->vring2cid_tid[ringid][0];
1352 struct wil_net_stats *stats = NULL;
1353 volatile struct vring_tx_desc *_d;
1354 int used_before_complete;
1357 if (unlikely(!vring->va)) {
1358 wil_err(wil, "Tx irq[%d]: vring not initialized\n", ringid);
1362 if (unlikely(!txdata->enabled)) {
1363 wil_info(wil, "Tx irq[%d]: vring disabled\n", ringid);
1367 wil_dbg_txrx(wil, "%s(%d)\n", __func__, ringid);
1369 used_before_complete = wil_vring_used_tx(vring);
1371 if (cid < WIL6210_MAX_CID)
1372 stats = &wil->sta[cid].stats;
1374 while (!wil_vring_is_empty(vring)) {
1376 struct wil_ctx *ctx = &vring->ctx[vring->swtail];
1378 * For the fragmented skb, HW will set DU bit only for the
1379 * last fragment. look for it
1381 int lf = (vring->swtail + ctx->nr_frags) % vring->size;
1382 /* TODO: check we are not past head */
1384 _d = &vring->va[lf].tx;
1385 if (unlikely(!(_d->dma.status & TX_DMA_STATUS_DU)))
1388 new_swtail = (lf + 1) % vring->size;
1389 while (vring->swtail != new_swtail) {
1390 struct vring_tx_desc dd, *d = ⅆ
1392 struct sk_buff *skb;
1394 ctx = &vring->ctx[vring->swtail];
1396 _d = &vring->va[vring->swtail].tx;
1400 dmalen = le16_to_cpu(d->dma.length);
1401 trace_wil6210_tx_done(ringid, vring->swtail, dmalen,
1404 "TxC[%2d][%3d] : %d bytes, status 0x%02x err 0x%02x\n",
1405 ringid, vring->swtail, dmalen,
1406 d->dma.status, d->dma.error);
1407 wil_hex_dump_txrx("TxCD ", DUMP_PREFIX_NONE, 32, 4,
1408 (const void *)d, sizeof(*d), false);
1410 wil_txdesc_unmap(dev, d, ctx);
1413 if (likely(d->dma.error == 0)) {
1414 ndev->stats.tx_packets++;
1415 ndev->stats.tx_bytes += skb->len;
1417 stats->tx_packets++;
1418 stats->tx_bytes += skb->len;
1421 ndev->stats.tx_errors++;
1425 wil_consume_skb(skb, d->dma.error == 0);
1427 memset(ctx, 0, sizeof(*ctx));
1428 /* There is no need to touch HW descriptor:
1429 * - ststus bit TX_DMA_STATUS_DU is set by design,
1430 * so hardware will not try to process this desc.,
1431 * - rest of descriptor will be initialized on Tx.
1433 vring->swtail = wil_vring_next_tail(vring);
1438 /* performance monitoring */
1439 used_new = wil_vring_used_tx(vring);
1440 if (wil_val_in_range(vring_idle_trsh,
1441 used_new, used_before_complete)) {
1442 wil_dbg_txrx(wil, "Ring[%2d] idle %d -> %d\n",
1443 ringid, used_before_complete, used_new);
1444 txdata->last_idle = get_cycles();
1447 if (wil_vring_avail_tx(vring) > wil_vring_wmark_high(vring)) {
1448 wil_dbg_txrx(wil, "netif_tx_wake : ring not full\n");
1449 netif_tx_wake_all_queues(wil_to_ndev(wil));