1 /* 2 * Copyright (c) 2012 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 * Available space in Tx Vring 57 */ 58 static inline int wil_vring_avail_tx(struct vring *vring) 59 { 60 u32 swhead = vring->swhead; 61 u32 swtail = vring->swtail; 62 int used = (vring->size + swhead - swtail) % vring->size; 63 64 return vring->size - used - 1; 65 } 66 67 static int wil_vring_alloc(struct wil6210_priv *wil, struct vring *vring) 68 { 69 struct device *dev = wil_to_dev(wil); 70 size_t sz = vring->size * sizeof(vring->va[0]); 71 uint i; 72 73 BUILD_BUG_ON(sizeof(vring->va[0]) != 32); 74 75 vring->swhead = 0; 76 vring->swtail = 0; 77 vring->ctx = kcalloc(vring->size, sizeof(vring->ctx[0]), GFP_KERNEL); 78 if (!vring->ctx) { 79 vring->va = NULL; 80 return -ENOMEM; 81 } 82 /* 83 * vring->va should be aligned on its size rounded up to power of 2 84 * This is granted by the dma_alloc_coherent 85 */ 86 vring->va = dma_alloc_coherent(dev, sz, &vring->pa, GFP_KERNEL); 87 if (!vring->va) { 88 kfree(vring->ctx); 89 vring->ctx = NULL; 90 return -ENOMEM; 91 } 92 /* initially, all descriptors are SW owned 93 * For Tx and Rx, ownership bit is at the same location, thus 94 * we can use any 95 */ 96 for (i = 0; i < vring->size; i++) { 97 volatile struct vring_tx_desc *_d = &(vring->va[i].tx); 98 _d->dma.status = TX_DMA_STATUS_DU; 99 } 100 101 wil_dbg_misc(wil, "vring[%d] 0x%p:0x%016llx 0x%p\n", vring->size, 102 vring->va, (unsigned long long)vring->pa, vring->ctx); 103 104 return 0; 105 } 106 107 static void wil_txdesc_unmap(struct device *dev, struct vring_tx_desc *d, 108 struct wil_ctx *ctx) 109 { 110 dma_addr_t pa = wil_desc_addr(&d->dma.addr); 111 u16 dmalen = le16_to_cpu(d->dma.length); 112 switch (ctx->mapped_as) { 113 case wil_mapped_as_single: 114 dma_unmap_single(dev, pa, dmalen, DMA_TO_DEVICE); 115 break; 116 case wil_mapped_as_page: 117 dma_unmap_page(dev, pa, dmalen, DMA_TO_DEVICE); 118 break; 119 default: 120 break; 121 } 122 } 123 124 static void wil_vring_free(struct wil6210_priv *wil, struct vring *vring, 125 int tx) 126 { 127 struct device *dev = wil_to_dev(wil); 128 size_t sz = vring->size * sizeof(vring->va[0]); 129 130 while (!wil_vring_is_empty(vring)) { 131 dma_addr_t pa; 132 u16 dmalen; 133 struct wil_ctx *ctx; 134 135 if (tx) { 136 struct vring_tx_desc dd, *d = ⅆ 137 volatile struct vring_tx_desc *_d = 138 &vring->va[vring->swtail].tx; 139 140 ctx = &vring->ctx[vring->swtail]; 141 *d = *_d; 142 wil_txdesc_unmap(dev, d, ctx); 143 if (ctx->skb) 144 dev_kfree_skb_any(ctx->skb); 145 vring->swtail = wil_vring_next_tail(vring); 146 } else { /* rx */ 147 struct vring_rx_desc dd, *d = ⅆ 148 volatile struct vring_rx_desc *_d = 149 &vring->va[vring->swhead].rx; 150 151 ctx = &vring->ctx[vring->swhead]; 152 *d = *_d; 153 pa = wil_desc_addr(&d->dma.addr); 154 dmalen = le16_to_cpu(d->dma.length); 155 dma_unmap_single(dev, pa, dmalen, DMA_FROM_DEVICE); 156 kfree_skb(ctx->skb); 157 wil_vring_advance_head(vring, 1); 158 } 159 } 160 dma_free_coherent(dev, sz, (void *)vring->va, vring->pa); 161 kfree(vring->ctx); 162 vring->pa = 0; 163 vring->va = NULL; 164 vring->ctx = NULL; 165 } 166 167 /** 168 * Allocate one skb for Rx VRING 169 * 170 * Safe to call from IRQ 171 */ 172 static int wil_vring_alloc_skb(struct wil6210_priv *wil, struct vring *vring, 173 u32 i, int headroom) 174 { 175 struct device *dev = wil_to_dev(wil); 176 unsigned int sz = RX_BUF_LEN; 177 struct vring_rx_desc dd, *d = ⅆ 178 volatile struct vring_rx_desc *_d = &(vring->va[i].rx); 179 dma_addr_t pa; 180 181 /* TODO align */ 182 struct sk_buff *skb = dev_alloc_skb(sz + headroom); 183 if (unlikely(!skb)) 184 return -ENOMEM; 185 186 skb_reserve(skb, headroom); 187 skb_put(skb, sz); 188 189 pa = dma_map_single(dev, skb->data, skb->len, DMA_FROM_DEVICE); 190 if (unlikely(dma_mapping_error(dev, pa))) { 191 kfree_skb(skb); 192 return -ENOMEM; 193 } 194 195 d->dma.d0 = BIT(9) | RX_DMA_D0_CMD_DMA_IT; 196 wil_desc_addr_set(&d->dma.addr, pa); 197 /* ip_length don't care */ 198 /* b11 don't care */ 199 /* error don't care */ 200 d->dma.status = 0; /* BIT(0) should be 0 for HW_OWNED */ 201 d->dma.length = cpu_to_le16(sz); 202 *_d = *d; 203 vring->ctx[i].skb = skb; 204 205 return 0; 206 } 207 208 /** 209 * Adds radiotap header 210 * 211 * Any error indicated as "Bad FCS" 212 * 213 * Vendor data for 04:ce:14-1 (Wilocity-1) consists of: 214 * - Rx descriptor: 32 bytes 215 * - Phy info 216 */ 217 static void wil_rx_add_radiotap_header(struct wil6210_priv *wil, 218 struct sk_buff *skb) 219 { 220 struct wireless_dev *wdev = wil->wdev; 221 struct wil6210_rtap { 222 struct ieee80211_radiotap_header rthdr; 223 /* fields should be in the order of bits in rthdr.it_present */ 224 /* flags */ 225 u8 flags; 226 /* channel */ 227 __le16 chnl_freq __aligned(2); 228 __le16 chnl_flags; 229 /* MCS */ 230 u8 mcs_present; 231 u8 mcs_flags; 232 u8 mcs_index; 233 } __packed; 234 struct wil6210_rtap_vendor { 235 struct wil6210_rtap rtap; 236 /* vendor */ 237 u8 vendor_oui[3] __aligned(2); 238 u8 vendor_ns; 239 __le16 vendor_skip; 240 u8 vendor_data[0]; 241 } __packed; 242 struct vring_rx_desc *d = wil_skb_rxdesc(skb); 243 struct wil6210_rtap_vendor *rtap_vendor; 244 int rtap_len = sizeof(struct wil6210_rtap); 245 int phy_length = 0; /* phy info header size, bytes */ 246 static char phy_data[128]; 247 struct ieee80211_channel *ch = wdev->preset_chandef.chan; 248 249 if (rtap_include_phy_info) { 250 rtap_len = sizeof(*rtap_vendor) + sizeof(*d); 251 /* calculate additional length */ 252 if (d->dma.status & RX_DMA_STATUS_PHY_INFO) { 253 /** 254 * PHY info starts from 8-byte boundary 255 * there are 8-byte lines, last line may be partially 256 * written (HW bug), thus FW configures for last line 257 * to be excessive. Driver skips this last line. 258 */ 259 int len = min_t(int, 8 + sizeof(phy_data), 260 wil_rxdesc_phy_length(d)); 261 if (len > 8) { 262 void *p = skb_tail_pointer(skb); 263 void *pa = PTR_ALIGN(p, 8); 264 if (skb_tailroom(skb) >= len + (pa - p)) { 265 phy_length = len - 8; 266 memcpy(phy_data, pa, phy_length); 267 } 268 } 269 } 270 rtap_len += phy_length; 271 } 272 273 if (skb_headroom(skb) < rtap_len && 274 pskb_expand_head(skb, rtap_len, 0, GFP_ATOMIC)) { 275 wil_err(wil, "Unable to expand headrom to %d\n", rtap_len); 276 return; 277 } 278 279 rtap_vendor = (void *)skb_push(skb, rtap_len); 280 memset(rtap_vendor, 0, rtap_len); 281 282 rtap_vendor->rtap.rthdr.it_version = PKTHDR_RADIOTAP_VERSION; 283 rtap_vendor->rtap.rthdr.it_len = cpu_to_le16(rtap_len); 284 rtap_vendor->rtap.rthdr.it_present = cpu_to_le32( 285 (1 << IEEE80211_RADIOTAP_FLAGS) | 286 (1 << IEEE80211_RADIOTAP_CHANNEL) | 287 (1 << IEEE80211_RADIOTAP_MCS)); 288 if (d->dma.status & RX_DMA_STATUS_ERROR) 289 rtap_vendor->rtap.flags |= IEEE80211_RADIOTAP_F_BADFCS; 290 291 rtap_vendor->rtap.chnl_freq = cpu_to_le16(ch ? ch->center_freq : 58320); 292 rtap_vendor->rtap.chnl_flags = cpu_to_le16(0); 293 294 rtap_vendor->rtap.mcs_present = IEEE80211_RADIOTAP_MCS_HAVE_MCS; 295 rtap_vendor->rtap.mcs_flags = 0; 296 rtap_vendor->rtap.mcs_index = wil_rxdesc_mcs(d); 297 298 if (rtap_include_phy_info) { 299 rtap_vendor->rtap.rthdr.it_present |= cpu_to_le32(1 << 300 IEEE80211_RADIOTAP_VENDOR_NAMESPACE); 301 /* OUI for Wilocity 04:ce:14 */ 302 rtap_vendor->vendor_oui[0] = 0x04; 303 rtap_vendor->vendor_oui[1] = 0xce; 304 rtap_vendor->vendor_oui[2] = 0x14; 305 rtap_vendor->vendor_ns = 1; 306 /* Rx descriptor + PHY data */ 307 rtap_vendor->vendor_skip = cpu_to_le16(sizeof(*d) + 308 phy_length); 309 memcpy(rtap_vendor->vendor_data, (void *)d, sizeof(*d)); 310 memcpy(rtap_vendor->vendor_data + sizeof(*d), phy_data, 311 phy_length); 312 } 313 } 314 315 /* 316 * Fast swap in place between 2 registers 317 */ 318 static void wil_swap_u16(u16 *a, u16 *b) 319 { 320 *a ^= *b; 321 *b ^= *a; 322 *a ^= *b; 323 } 324 325 static void wil_swap_ethaddr(void *data) 326 { 327 struct ethhdr *eth = data; 328 u16 *s = (u16 *)eth->h_source; 329 u16 *d = (u16 *)eth->h_dest; 330 331 wil_swap_u16(s++, d++); 332 wil_swap_u16(s++, d++); 333 wil_swap_u16(s, d); 334 } 335 336 /** 337 * reap 1 frame from @swhead 338 * 339 * Rx descriptor copied to skb->cb 340 * 341 * Safe to call from IRQ 342 */ 343 static struct sk_buff *wil_vring_reap_rx(struct wil6210_priv *wil, 344 struct vring *vring) 345 { 346 struct device *dev = wil_to_dev(wil); 347 struct net_device *ndev = wil_to_ndev(wil); 348 volatile struct vring_rx_desc *_d; 349 struct vring_rx_desc *d; 350 struct sk_buff *skb; 351 dma_addr_t pa; 352 unsigned int sz = RX_BUF_LEN; 353 u16 dmalen; 354 u8 ftype; 355 u8 ds_bits; 356 int cid; 357 struct wil_net_stats *stats; 358 359 360 BUILD_BUG_ON(sizeof(struct vring_rx_desc) > sizeof(skb->cb)); 361 362 if (wil_vring_is_empty(vring)) 363 return NULL; 364 365 _d = &(vring->va[vring->swhead].rx); 366 if (!(_d->dma.status & RX_DMA_STATUS_DU)) { 367 /* it is not error, we just reached end of Rx done area */ 368 return NULL; 369 } 370 371 skb = vring->ctx[vring->swhead].skb; 372 d = wil_skb_rxdesc(skb); 373 *d = *_d; 374 pa = wil_desc_addr(&d->dma.addr); 375 vring->ctx[vring->swhead].skb = NULL; 376 wil_vring_advance_head(vring, 1); 377 378 dma_unmap_single(dev, pa, sz, DMA_FROM_DEVICE); 379 dmalen = le16_to_cpu(d->dma.length); 380 381 trace_wil6210_rx(vring->swhead, d); 382 wil_dbg_txrx(wil, "Rx[%3d] : %d bytes\n", vring->swhead, dmalen); 383 wil_hex_dump_txrx("Rx ", DUMP_PREFIX_NONE, 32, 4, 384 (const void *)d, sizeof(*d), false); 385 386 if (dmalen > sz) { 387 wil_err(wil, "Rx size too large: %d bytes!\n", dmalen); 388 kfree_skb(skb); 389 return NULL; 390 } 391 skb_trim(skb, dmalen); 392 393 prefetch(skb->data); 394 395 wil_hex_dump_txrx("Rx ", DUMP_PREFIX_OFFSET, 16, 1, 396 skb->data, skb_headlen(skb), false); 397 398 cid = wil_rxdesc_cid(d); 399 stats = &wil->sta[cid].stats; 400 stats->last_mcs_rx = wil_rxdesc_mcs(d); 401 wil->stats.last_mcs_rx = stats->last_mcs_rx; 402 403 /* use radiotap header only if required */ 404 if (ndev->type == ARPHRD_IEEE80211_RADIOTAP) 405 wil_rx_add_radiotap_header(wil, skb); 406 407 /* no extra checks if in sniffer mode */ 408 if (ndev->type != ARPHRD_ETHER) 409 return skb; 410 /* 411 * Non-data frames may be delivered through Rx DMA channel (ex: BAR) 412 * Driver should recognize it by frame type, that is found 413 * in Rx descriptor. If type is not data, it is 802.11 frame as is 414 */ 415 ftype = wil_rxdesc_ftype(d) << 2; 416 if (ftype != IEEE80211_FTYPE_DATA) { 417 wil_dbg_txrx(wil, "Non-data frame ftype 0x%08x\n", ftype); 418 /* TODO: process it */ 419 kfree_skb(skb); 420 return NULL; 421 } 422 423 if (skb->len < ETH_HLEN) { 424 wil_err(wil, "Short frame, len = %d\n", skb->len); 425 /* TODO: process it (i.e. BAR) */ 426 kfree_skb(skb); 427 return NULL; 428 } 429 430 /* L4 IDENT is on when HW calculated checksum, check status 431 * and in case of error drop the packet 432 * higher stack layers will handle retransmission (if required) 433 */ 434 if (d->dma.status & RX_DMA_STATUS_L4_IDENT) { 435 /* L4 protocol identified, csum calculated */ 436 if ((d->dma.error & RX_DMA_ERROR_L4_ERR) == 0) 437 skb->ip_summed = CHECKSUM_UNNECESSARY; 438 /* If HW reports bad checksum, let IP stack re-check it 439 * For example, HW don't understand Microsoft IP stack that 440 * mis-calculates TCP checksum - if it should be 0x0, 441 * it writes 0xffff in violation of RFC 1624 442 */ 443 } 444 445 ds_bits = wil_rxdesc_ds_bits(d); 446 if (ds_bits == 1) { 447 /* 448 * HW bug - in ToDS mode, i.e. Rx on AP side, 449 * addresses get swapped 450 */ 451 wil_swap_ethaddr(skb->data); 452 } 453 454 return skb; 455 } 456 457 /** 458 * allocate and fill up to @count buffers in rx ring 459 * buffers posted at @swtail 460 */ 461 static int wil_rx_refill(struct wil6210_priv *wil, int count) 462 { 463 struct net_device *ndev = wil_to_ndev(wil); 464 struct vring *v = &wil->vring_rx; 465 u32 next_tail; 466 int rc = 0; 467 int headroom = ndev->type == ARPHRD_IEEE80211_RADIOTAP ? 468 WIL6210_RTAP_SIZE : 0; 469 470 for (; next_tail = wil_vring_next_tail(v), 471 (next_tail != v->swhead) && (count-- > 0); 472 v->swtail = next_tail) { 473 rc = wil_vring_alloc_skb(wil, v, v->swtail, headroom); 474 if (rc) { 475 wil_err(wil, "Error %d in wil_rx_refill[%d]\n", 476 rc, v->swtail); 477 break; 478 } 479 } 480 iowrite32(v->swtail, wil->csr + HOSTADDR(v->hwtail)); 481 482 return rc; 483 } 484 485 /* 486 * Pass Rx packet to the netif. Update statistics. 487 * Called in softirq context (NAPI poll). 488 */ 489 void wil_netif_rx_any(struct sk_buff *skb, struct net_device *ndev) 490 { 491 gro_result_t rc; 492 struct wil6210_priv *wil = ndev_to_wil(ndev); 493 unsigned int len = skb->len; 494 struct vring_rx_desc *d = wil_skb_rxdesc(skb); 495 int cid = wil_rxdesc_cid(d); 496 struct wil_net_stats *stats = &wil->sta[cid].stats; 497 498 skb_orphan(skb); 499 500 rc = napi_gro_receive(&wil->napi_rx, skb); 501 502 if (unlikely(rc == GRO_DROP)) { 503 ndev->stats.rx_dropped++; 504 stats->rx_dropped++; 505 wil_dbg_txrx(wil, "Rx drop %d bytes\n", len); 506 } else { 507 ndev->stats.rx_packets++; 508 stats->rx_packets++; 509 ndev->stats.rx_bytes += len; 510 stats->rx_bytes += len; 511 } 512 } 513 514 /** 515 * Proceed all completed skb's from Rx VRING 516 * 517 * Safe to call from NAPI poll, i.e. softirq with interrupts enabled 518 */ 519 void wil_rx_handle(struct wil6210_priv *wil, int *quota) 520 { 521 struct net_device *ndev = wil_to_ndev(wil); 522 struct vring *v = &wil->vring_rx; 523 struct sk_buff *skb; 524 525 if (!v->va) { 526 wil_err(wil, "Rx IRQ while Rx not yet initialized\n"); 527 return; 528 } 529 wil_dbg_txrx(wil, "%s()\n", __func__); 530 while ((*quota > 0) && (NULL != (skb = wil_vring_reap_rx(wil, v)))) { 531 (*quota)--; 532 533 if (wil->wdev->iftype == NL80211_IFTYPE_MONITOR) { 534 skb->dev = ndev; 535 skb_reset_mac_header(skb); 536 skb->ip_summed = CHECKSUM_UNNECESSARY; 537 skb->pkt_type = PACKET_OTHERHOST; 538 skb->protocol = htons(ETH_P_802_2); 539 wil_netif_rx_any(skb, ndev); 540 } else { 541 struct ethhdr *eth = (void *)skb->data; 542 543 skb->protocol = eth_type_trans(skb, ndev); 544 545 if (is_unicast_ether_addr(eth->h_dest)) 546 wil_rx_reorder(wil, skb); 547 else 548 wil_netif_rx_any(skb, ndev); 549 } 550 551 } 552 wil_rx_refill(wil, v->size); 553 } 554 555 int wil_rx_init(struct wil6210_priv *wil) 556 { 557 struct vring *vring = &wil->vring_rx; 558 int rc; 559 560 if (vring->va) { 561 wil_err(wil, "Rx ring already allocated\n"); 562 return -EINVAL; 563 } 564 565 vring->size = WIL6210_RX_RING_SIZE; 566 rc = wil_vring_alloc(wil, vring); 567 if (rc) 568 return rc; 569 570 rc = wmi_rx_chain_add(wil, vring); 571 if (rc) 572 goto err_free; 573 574 rc = wil_rx_refill(wil, vring->size); 575 if (rc) 576 goto err_free; 577 578 return 0; 579 err_free: 580 wil_vring_free(wil, vring, 0); 581 582 return rc; 583 } 584 585 void wil_rx_fini(struct wil6210_priv *wil) 586 { 587 struct vring *vring = &wil->vring_rx; 588 589 if (vring->va) 590 wil_vring_free(wil, vring, 0); 591 } 592 593 int wil_vring_init_tx(struct wil6210_priv *wil, int id, int size, 594 int cid, int tid) 595 { 596 int rc; 597 struct wmi_vring_cfg_cmd cmd = { 598 .action = cpu_to_le32(WMI_VRING_CMD_ADD), 599 .vring_cfg = { 600 .tx_sw_ring = { 601 .max_mpdu_size = cpu_to_le16(TX_BUF_LEN), 602 .ring_size = cpu_to_le16(size), 603 }, 604 .ringid = id, 605 .cidxtid = mk_cidxtid(cid, tid), 606 .encap_trans_type = WMI_VRING_ENC_TYPE_802_3, 607 .mac_ctrl = 0, 608 .to_resolution = 0, 609 .agg_max_wsize = 16, 610 .schd_params = { 611 .priority = cpu_to_le16(0), 612 .timeslot_us = cpu_to_le16(0xfff), 613 }, 614 }, 615 }; 616 struct { 617 struct wil6210_mbox_hdr_wmi wmi; 618 struct wmi_vring_cfg_done_event cmd; 619 } __packed reply; 620 struct vring *vring = &wil->vring_tx[id]; 621 struct vring_tx_data *txdata = &wil->vring_tx_data[id]; 622 623 if (vring->va) { 624 wil_err(wil, "Tx ring [%d] already allocated\n", id); 625 rc = -EINVAL; 626 goto out; 627 } 628 629 memset(txdata, 0, sizeof(*txdata)); 630 vring->size = size; 631 rc = wil_vring_alloc(wil, vring); 632 if (rc) 633 goto out; 634 635 wil->vring2cid_tid[id][0] = cid; 636 wil->vring2cid_tid[id][1] = tid; 637 638 cmd.vring_cfg.tx_sw_ring.ring_mem_base = cpu_to_le64(vring->pa); 639 640 rc = wmi_call(wil, WMI_VRING_CFG_CMDID, &cmd, sizeof(cmd), 641 WMI_VRING_CFG_DONE_EVENTID, &reply, sizeof(reply), 100); 642 if (rc) 643 goto out_free; 644 645 if (reply.cmd.status != WMI_FW_STATUS_SUCCESS) { 646 wil_err(wil, "Tx config failed, status 0x%02x\n", 647 reply.cmd.status); 648 rc = -EINVAL; 649 goto out_free; 650 } 651 vring->hwtail = le32_to_cpu(reply.cmd.tx_vring_tail_ptr); 652 653 txdata->enabled = 1; 654 655 return 0; 656 out_free: 657 wil_vring_free(wil, vring, 1); 658 out: 659 660 return rc; 661 } 662 663 void wil_vring_fini_tx(struct wil6210_priv *wil, int id) 664 { 665 struct vring *vring = &wil->vring_tx[id]; 666 667 WARN_ON(!mutex_is_locked(&wil->mutex)); 668 669 if (!vring->va) 670 return; 671 672 /* make sure NAPI won't touch this vring */ 673 wil->vring_tx_data[id].enabled = 0; 674 if (test_bit(wil_status_napi_en, &wil->status)) 675 napi_synchronize(&wil->napi_tx); 676 677 wil_vring_free(wil, vring, 1); 678 } 679 680 static struct vring *wil_find_tx_vring(struct wil6210_priv *wil, 681 struct sk_buff *skb) 682 { 683 int i; 684 struct ethhdr *eth = (void *)skb->data; 685 int cid = wil_find_cid(wil, eth->h_dest); 686 687 if (cid < 0) 688 return NULL; 689 690 if (!wil->sta[cid].data_port_open && 691 (skb->protocol != cpu_to_be16(ETH_P_PAE))) 692 return NULL; 693 694 /* TODO: fix for multiple TID */ 695 for (i = 0; i < ARRAY_SIZE(wil->vring2cid_tid); i++) { 696 if (wil->vring2cid_tid[i][0] == cid) { 697 struct vring *v = &wil->vring_tx[i]; 698 wil_dbg_txrx(wil, "%s(%pM) -> [%d]\n", 699 __func__, eth->h_dest, i); 700 if (v->va) { 701 return v; 702 } else { 703 wil_dbg_txrx(wil, "vring[%d] not valid\n", i); 704 return NULL; 705 } 706 } 707 } 708 709 return NULL; 710 } 711 712 static void wil_set_da_for_vring(struct wil6210_priv *wil, 713 struct sk_buff *skb, int vring_index) 714 { 715 struct ethhdr *eth = (void *)skb->data; 716 int cid = wil->vring2cid_tid[vring_index][0]; 717 memcpy(eth->h_dest, wil->sta[cid].addr, ETH_ALEN); 718 } 719 720 static int wil_tx_vring(struct wil6210_priv *wil, struct vring *vring, 721 struct sk_buff *skb); 722 /* 723 * Find 1-st vring and return it; set dest address for this vring in skb 724 * duplicate skb and send it to other active vrings 725 */ 726 static struct vring *wil_tx_bcast(struct wil6210_priv *wil, 727 struct sk_buff *skb) 728 { 729 struct vring *v, *v2; 730 struct sk_buff *skb2; 731 int i; 732 u8 cid; 733 734 /* find 1-st vring eligible for data */ 735 for (i = 0; i < WIL6210_MAX_TX_RINGS; i++) { 736 v = &wil->vring_tx[i]; 737 if (!v->va) 738 continue; 739 740 cid = wil->vring2cid_tid[i][0]; 741 if (!wil->sta[cid].data_port_open) 742 continue; 743 744 goto found; 745 } 746 747 wil_err(wil, "Tx while no vrings active?\n"); 748 749 return NULL; 750 751 found: 752 wil_dbg_txrx(wil, "BCAST -> ring %d\n", i); 753 wil_set_da_for_vring(wil, skb, i); 754 755 /* find other active vrings and duplicate skb for each */ 756 for (i++; i < WIL6210_MAX_TX_RINGS; i++) { 757 v2 = &wil->vring_tx[i]; 758 if (!v2->va) 759 continue; 760 cid = wil->vring2cid_tid[i][0]; 761 if (!wil->sta[cid].data_port_open) 762 continue; 763 764 skb2 = skb_copy(skb, GFP_ATOMIC); 765 if (skb2) { 766 wil_dbg_txrx(wil, "BCAST DUP -> ring %d\n", i); 767 wil_set_da_for_vring(wil, skb2, i); 768 wil_tx_vring(wil, v2, skb2); 769 } else { 770 wil_err(wil, "skb_copy failed\n"); 771 } 772 } 773 774 return v; 775 } 776 777 static int wil_tx_desc_map(struct vring_tx_desc *d, dma_addr_t pa, u32 len, 778 int vring_index) 779 { 780 wil_desc_addr_set(&d->dma.addr, pa); 781 d->dma.ip_length = 0; 782 /* 0..6: mac_length; 7:ip_version 0-IP6 1-IP4*/ 783 d->dma.b11 = 0/*14 | BIT(7)*/; 784 d->dma.error = 0; 785 d->dma.status = 0; /* BIT(0) should be 0 for HW_OWNED */ 786 d->dma.length = cpu_to_le16((u16)len); 787 d->dma.d0 = (vring_index << DMA_CFG_DESC_TX_0_QID_POS); 788 d->mac.d[0] = 0; 789 d->mac.d[1] = 0; 790 d->mac.d[2] = 0; 791 d->mac.ucode_cmd = 0; 792 /* use dst index 0 */ 793 d->mac.d[1] |= BIT(MAC_CFG_DESC_TX_1_DST_INDEX_EN_POS) | 794 (0 << MAC_CFG_DESC_TX_1_DST_INDEX_POS); 795 /* translation type: 0 - bypass; 1 - 802.3; 2 - native wifi */ 796 d->mac.d[2] = BIT(MAC_CFG_DESC_TX_2_SNAP_HDR_INSERTION_EN_POS) | 797 (1 << MAC_CFG_DESC_TX_2_L2_TRANSLATION_TYPE_POS); 798 799 return 0; 800 } 801 802 static inline 803 void wil_tx_desc_set_nr_frags(struct vring_tx_desc *d, int nr_frags) 804 { 805 d->mac.d[2] |= ((nr_frags + 1) << 806 MAC_CFG_DESC_TX_2_NUM_OF_DESCRIPTORS_POS); 807 } 808 809 static int wil_tx_desc_offload_cksum_set(struct wil6210_priv *wil, 810 struct vring_tx_desc *d, 811 struct sk_buff *skb) 812 { 813 int protocol; 814 815 if (skb->ip_summed != CHECKSUM_PARTIAL) 816 return 0; 817 818 d->dma.b11 = ETH_HLEN; /* MAC header length */ 819 820 switch (skb->protocol) { 821 case cpu_to_be16(ETH_P_IP): 822 protocol = ip_hdr(skb)->protocol; 823 d->dma.b11 |= BIT(DMA_CFG_DESC_TX_OFFLOAD_CFG_L3T_IPV4_POS); 824 break; 825 case cpu_to_be16(ETH_P_IPV6): 826 protocol = ipv6_hdr(skb)->nexthdr; 827 break; 828 default: 829 return -EINVAL; 830 } 831 832 switch (protocol) { 833 case IPPROTO_TCP: 834 d->dma.d0 |= (2 << DMA_CFG_DESC_TX_0_L4_TYPE_POS); 835 /* L4 header len: TCP header length */ 836 d->dma.d0 |= 837 (tcp_hdrlen(skb) & DMA_CFG_DESC_TX_0_L4_LENGTH_MSK); 838 break; 839 case IPPROTO_UDP: 840 /* L4 header len: UDP header length */ 841 d->dma.d0 |= 842 (sizeof(struct udphdr) & DMA_CFG_DESC_TX_0_L4_LENGTH_MSK); 843 break; 844 default: 845 return -EINVAL; 846 } 847 848 d->dma.ip_length = skb_network_header_len(skb); 849 /* Enable TCP/UDP checksum */ 850 d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_TCP_UDP_CHECKSUM_EN_POS); 851 /* Calculate pseudo-header */ 852 d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_PSEUDO_HEADER_CALC_EN_POS); 853 854 return 0; 855 } 856 857 static int wil_tx_vring(struct wil6210_priv *wil, struct vring *vring, 858 struct sk_buff *skb) 859 { 860 struct device *dev = wil_to_dev(wil); 861 struct vring_tx_desc dd, *d = ⅆ 862 volatile struct vring_tx_desc *_d; 863 u32 swhead = vring->swhead; 864 int avail = wil_vring_avail_tx(vring); 865 int nr_frags = skb_shinfo(skb)->nr_frags; 866 uint f = 0; 867 int vring_index = vring - wil->vring_tx; 868 uint i = swhead; 869 dma_addr_t pa; 870 871 wil_dbg_txrx(wil, "%s()\n", __func__); 872 873 if (avail < 1 + nr_frags) { 874 wil_err(wil, "Tx ring full. No space for %d fragments\n", 875 1 + nr_frags); 876 return -ENOMEM; 877 } 878 _d = &(vring->va[i].tx); 879 880 pa = dma_map_single(dev, skb->data, 881 skb_headlen(skb), DMA_TO_DEVICE); 882 883 wil_dbg_txrx(wil, "Tx skb %d bytes %p -> %#08llx\n", skb_headlen(skb), 884 skb->data, (unsigned long long)pa); 885 wil_hex_dump_txrx("Tx ", DUMP_PREFIX_OFFSET, 16, 1, 886 skb->data, skb_headlen(skb), false); 887 888 if (unlikely(dma_mapping_error(dev, pa))) 889 return -EINVAL; 890 vring->ctx[i].mapped_as = wil_mapped_as_single; 891 /* 1-st segment */ 892 wil_tx_desc_map(d, pa, skb_headlen(skb), vring_index); 893 /* Process TCP/UDP checksum offloading */ 894 if (wil_tx_desc_offload_cksum_set(wil, d, skb)) { 895 wil_err(wil, "VRING #%d Failed to set cksum, drop packet\n", 896 vring_index); 897 goto dma_error; 898 } 899 900 vring->ctx[i].nr_frags = nr_frags; 901 wil_tx_desc_set_nr_frags(d, nr_frags); 902 if (nr_frags) 903 *_d = *d; 904 905 /* middle segments */ 906 for (; f < nr_frags; f++) { 907 const struct skb_frag_struct *frag = 908 &skb_shinfo(skb)->frags[f]; 909 int len = skb_frag_size(frag); 910 i = (swhead + f + 1) % vring->size; 911 _d = &(vring->va[i].tx); 912 pa = skb_frag_dma_map(dev, frag, 0, skb_frag_size(frag), 913 DMA_TO_DEVICE); 914 if (unlikely(dma_mapping_error(dev, pa))) 915 goto dma_error; 916 vring->ctx[i].mapped_as = wil_mapped_as_page; 917 wil_tx_desc_map(d, pa, len, vring_index); 918 /* no need to check return code - 919 * if it succeeded for 1-st descriptor, 920 * it will succeed here too 921 */ 922 wil_tx_desc_offload_cksum_set(wil, d, skb); 923 *_d = *d; 924 } 925 /* for the last seg only */ 926 d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_CMD_EOP_POS); 927 d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_CMD_MARK_WB_POS); 928 d->dma.d0 |= BIT(DMA_CFG_DESC_TX_0_CMD_DMA_IT_POS); 929 *_d = *d; 930 931 /* hold reference to skb 932 * to prevent skb release before accounting 933 * in case of immediate "tx done" 934 */ 935 vring->ctx[i].skb = skb_get(skb); 936 937 wil_hex_dump_txrx("Tx ", DUMP_PREFIX_NONE, 32, 4, 938 (const void *)d, sizeof(*d), false); 939 940 /* advance swhead */ 941 wil_vring_advance_head(vring, nr_frags + 1); 942 wil_dbg_txrx(wil, "Tx swhead %d -> %d\n", swhead, vring->swhead); 943 trace_wil6210_tx(vring_index, swhead, skb->len, nr_frags); 944 iowrite32(vring->swhead, wil->csr + HOSTADDR(vring->hwtail)); 945 946 return 0; 947 dma_error: 948 /* unmap what we have mapped */ 949 nr_frags = f + 1; /* frags mapped + one for skb head */ 950 for (f = 0; f < nr_frags; f++) { 951 struct wil_ctx *ctx; 952 953 i = (swhead + f) % vring->size; 954 ctx = &vring->ctx[i]; 955 _d = &(vring->va[i].tx); 956 *d = *_d; 957 _d->dma.status = TX_DMA_STATUS_DU; 958 wil_txdesc_unmap(dev, d, ctx); 959 960 if (ctx->skb) 961 dev_kfree_skb_any(ctx->skb); 962 963 memset(ctx, 0, sizeof(*ctx)); 964 } 965 966 return -EINVAL; 967 } 968 969 970 netdev_tx_t wil_start_xmit(struct sk_buff *skb, struct net_device *ndev) 971 { 972 struct wil6210_priv *wil = ndev_to_wil(ndev); 973 struct ethhdr *eth = (void *)skb->data; 974 struct vring *vring; 975 static bool pr_once_fw; 976 int rc; 977 978 wil_dbg_txrx(wil, "%s()\n", __func__); 979 if (!test_bit(wil_status_fwready, &wil->status)) { 980 if (!pr_once_fw) { 981 wil_err(wil, "FW not ready\n"); 982 pr_once_fw = true; 983 } 984 goto drop; 985 } 986 if (!test_bit(wil_status_fwconnected, &wil->status)) { 987 wil_err(wil, "FW not connected\n"); 988 goto drop; 989 } 990 if (wil->wdev->iftype == NL80211_IFTYPE_MONITOR) { 991 wil_err(wil, "Xmit in monitor mode not supported\n"); 992 goto drop; 993 } 994 pr_once_fw = false; 995 996 /* find vring */ 997 if (is_unicast_ether_addr(eth->h_dest)) { 998 vring = wil_find_tx_vring(wil, skb); 999 } else { 1000 vring = wil_tx_bcast(wil, skb); 1001 } 1002 if (!vring) { 1003 wil_err(wil, "No Tx VRING found for %pM\n", eth->h_dest); 1004 goto drop; 1005 } 1006 /* set up vring entry */ 1007 rc = wil_tx_vring(wil, vring, skb); 1008 1009 /* do we still have enough room in the vring? */ 1010 if (wil_vring_avail_tx(vring) < vring->size/8) 1011 netif_tx_stop_all_queues(wil_to_ndev(wil)); 1012 1013 switch (rc) { 1014 case 0: 1015 /* statistics will be updated on the tx_complete */ 1016 dev_kfree_skb_any(skb); 1017 return NETDEV_TX_OK; 1018 case -ENOMEM: 1019 return NETDEV_TX_BUSY; 1020 default: 1021 break; /* goto drop; */ 1022 } 1023 drop: 1024 ndev->stats.tx_dropped++; 1025 dev_kfree_skb_any(skb); 1026 1027 return NET_XMIT_DROP; 1028 } 1029 1030 /** 1031 * Clean up transmitted skb's from the Tx VRING 1032 * 1033 * Return number of descriptors cleared 1034 * 1035 * Safe to call from IRQ 1036 */ 1037 int wil_tx_complete(struct wil6210_priv *wil, int ringid) 1038 { 1039 struct net_device *ndev = wil_to_ndev(wil); 1040 struct device *dev = wil_to_dev(wil); 1041 struct vring *vring = &wil->vring_tx[ringid]; 1042 struct vring_tx_data *txdata = &wil->vring_tx_data[ringid]; 1043 int done = 0; 1044 int cid = wil->vring2cid_tid[ringid][0]; 1045 struct wil_net_stats *stats = &wil->sta[cid].stats; 1046 volatile struct vring_tx_desc *_d; 1047 1048 if (!vring->va) { 1049 wil_err(wil, "Tx irq[%d]: vring not initialized\n", ringid); 1050 return 0; 1051 } 1052 1053 if (!txdata->enabled) { 1054 wil_info(wil, "Tx irq[%d]: vring disabled\n", ringid); 1055 return 0; 1056 } 1057 1058 wil_dbg_txrx(wil, "%s(%d)\n", __func__, ringid); 1059 1060 while (!wil_vring_is_empty(vring)) { 1061 int new_swtail; 1062 struct wil_ctx *ctx = &vring->ctx[vring->swtail]; 1063 /** 1064 * For the fragmented skb, HW will set DU bit only for the 1065 * last fragment. look for it 1066 */ 1067 int lf = (vring->swtail + ctx->nr_frags) % vring->size; 1068 /* TODO: check we are not past head */ 1069 1070 _d = &vring->va[lf].tx; 1071 if (!(_d->dma.status & TX_DMA_STATUS_DU)) 1072 break; 1073 1074 new_swtail = (lf + 1) % vring->size; 1075 while (vring->swtail != new_swtail) { 1076 struct vring_tx_desc dd, *d = ⅆ 1077 u16 dmalen; 1078 struct wil_ctx *ctx = &vring->ctx[vring->swtail]; 1079 struct sk_buff *skb = ctx->skb; 1080 _d = &vring->va[vring->swtail].tx; 1081 1082 *d = *_d; 1083 1084 dmalen = le16_to_cpu(d->dma.length); 1085 trace_wil6210_tx_done(ringid, vring->swtail, dmalen, 1086 d->dma.error); 1087 wil_dbg_txrx(wil, 1088 "Tx[%3d] : %d bytes, status 0x%02x err 0x%02x\n", 1089 vring->swtail, dmalen, d->dma.status, 1090 d->dma.error); 1091 wil_hex_dump_txrx("TxC ", DUMP_PREFIX_NONE, 32, 4, 1092 (const void *)d, sizeof(*d), false); 1093 1094 wil_txdesc_unmap(dev, d, ctx); 1095 1096 if (skb) { 1097 if (d->dma.error == 0) { 1098 ndev->stats.tx_packets++; 1099 stats->tx_packets++; 1100 ndev->stats.tx_bytes += skb->len; 1101 stats->tx_bytes += skb->len; 1102 } else { 1103 ndev->stats.tx_errors++; 1104 stats->tx_errors++; 1105 } 1106 1107 dev_kfree_skb_any(skb); 1108 } 1109 memset(ctx, 0, sizeof(*ctx)); 1110 /* There is no need to touch HW descriptor: 1111 * - ststus bit TX_DMA_STATUS_DU is set by design, 1112 * so hardware will not try to process this desc., 1113 * - rest of descriptor will be initialized on Tx. 1114 */ 1115 vring->swtail = wil_vring_next_tail(vring); 1116 done++; 1117 } 1118 } 1119 if (wil_vring_avail_tx(vring) > vring->size/4) 1120 netif_tx_wake_all_queues(wil_to_ndev(wil)); 1121 1122 return done; 1123 } 1124