vmxnet3_drv.c revision b738127dfb469bb9f595cdace30e7f881e8146b2
1/* 2 * Linux driver for VMware's vmxnet3 ethernet NIC. 3 * 4 * Copyright (C) 2008-2009, VMware, Inc. All Rights Reserved. 5 * 6 * This program is free software; you can redistribute it and/or modify it 7 * under the terms of the GNU General Public License as published by the 8 * Free Software Foundation; version 2 of the License and no later version. 9 * 10 * This program is distributed in the hope that it will be useful, but 11 * WITHOUT ANY WARRANTY; without even the implied warranty of 12 * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or 13 * NON INFRINGEMENT. See the GNU General Public License for more 14 * details. 15 * 16 * You should have received a copy of the GNU General Public License 17 * along with this program; if not, write to the Free Software 18 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 19 * 20 * The full GNU General Public License is included in this distribution in 21 * the file called "COPYING". 22 * 23 * Maintained by: Shreyas Bhatewara <pv-drivers@vmware.com> 24 * 25 */ 26 27#include <net/ip6_checksum.h> 28 29#include "vmxnet3_int.h" 30 31char vmxnet3_driver_name[] = "vmxnet3"; 32#define VMXNET3_DRIVER_DESC "VMware vmxnet3 virtual NIC driver" 33 34/* 35 * PCI Device ID Table 36 * Last entry must be all 0s 37 */ 38static DEFINE_PCI_DEVICE_TABLE(vmxnet3_pciid_table) = { 39 {PCI_VDEVICE(VMWARE, PCI_DEVICE_ID_VMWARE_VMXNET3)}, 40 {0} 41}; 42 43MODULE_DEVICE_TABLE(pci, vmxnet3_pciid_table); 44 45static atomic_t devices_found; 46 47 48/* 49 * Enable/Disable the given intr 50 */ 51static void 52vmxnet3_enable_intr(struct vmxnet3_adapter *adapter, unsigned intr_idx) 53{ 54 VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_IMR + intr_idx * 8, 0); 55} 56 57 58static void 59vmxnet3_disable_intr(struct vmxnet3_adapter *adapter, unsigned intr_idx) 60{ 61 VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_IMR + intr_idx * 8, 1); 62} 63 64 65/* 66 * Enable/Disable all intrs used by the device 67 */ 68static void 69vmxnet3_enable_all_intrs(struct vmxnet3_adapter *adapter) 70{ 71 int i; 72 73 for (i = 0; i < adapter->intr.num_intrs; i++) 74 vmxnet3_enable_intr(adapter, i); 75 adapter->shared->devRead.intrConf.intrCtrl &= 76 cpu_to_le32(~VMXNET3_IC_DISABLE_ALL); 77} 78 79 80static void 81vmxnet3_disable_all_intrs(struct vmxnet3_adapter *adapter) 82{ 83 int i; 84 85 adapter->shared->devRead.intrConf.intrCtrl |= 86 cpu_to_le32(VMXNET3_IC_DISABLE_ALL); 87 for (i = 0; i < adapter->intr.num_intrs; i++) 88 vmxnet3_disable_intr(adapter, i); 89} 90 91 92static void 93vmxnet3_ack_events(struct vmxnet3_adapter *adapter, u32 events) 94{ 95 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_ECR, events); 96} 97 98 99static bool 100vmxnet3_tq_stopped(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter) 101{ 102 return netif_queue_stopped(adapter->netdev); 103} 104 105 106static void 107vmxnet3_tq_start(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter) 108{ 109 tq->stopped = false; 110 netif_start_queue(adapter->netdev); 111} 112 113 114static void 115vmxnet3_tq_wake(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter) 116{ 117 tq->stopped = false; 118 netif_wake_queue(adapter->netdev); 119} 120 121 122static void 123vmxnet3_tq_stop(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter) 124{ 125 tq->stopped = true; 126 tq->num_stop++; 127 netif_stop_queue(adapter->netdev); 128} 129 130 131/* 132 * Check the link state. This may start or stop the tx queue. 133 */ 134static void 135vmxnet3_check_link(struct vmxnet3_adapter *adapter, bool affectTxQueue) 136{ 137 u32 ret; 138 139 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, VMXNET3_CMD_GET_LINK); 140 ret = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD); 141 adapter->link_speed = ret >> 16; 142 if (ret & 1) { /* Link is up. */ 143 printk(KERN_INFO "%s: NIC Link is Up %d Mbps\n", 144 adapter->netdev->name, adapter->link_speed); 145 if (!netif_carrier_ok(adapter->netdev)) 146 netif_carrier_on(adapter->netdev); 147 148 if (affectTxQueue) 149 vmxnet3_tq_start(&adapter->tx_queue, adapter); 150 } else { 151 printk(KERN_INFO "%s: NIC Link is Down\n", 152 adapter->netdev->name); 153 if (netif_carrier_ok(adapter->netdev)) 154 netif_carrier_off(adapter->netdev); 155 156 if (affectTxQueue) 157 vmxnet3_tq_stop(&adapter->tx_queue, adapter); 158 } 159} 160 161static void 162vmxnet3_process_events(struct vmxnet3_adapter *adapter) 163{ 164 u32 events = le32_to_cpu(adapter->shared->ecr); 165 if (!events) 166 return; 167 168 vmxnet3_ack_events(adapter, events); 169 170 /* Check if link state has changed */ 171 if (events & VMXNET3_ECR_LINK) 172 vmxnet3_check_link(adapter, true); 173 174 /* Check if there is an error on xmit/recv queues */ 175 if (events & (VMXNET3_ECR_TQERR | VMXNET3_ECR_RQERR)) { 176 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, 177 VMXNET3_CMD_GET_QUEUE_STATUS); 178 179 if (adapter->tqd_start->status.stopped) { 180 printk(KERN_ERR "%s: tq error 0x%x\n", 181 adapter->netdev->name, 182 le32_to_cpu(adapter->tqd_start->status.error)); 183 } 184 if (adapter->rqd_start->status.stopped) { 185 printk(KERN_ERR "%s: rq error 0x%x\n", 186 adapter->netdev->name, 187 adapter->rqd_start->status.error); 188 } 189 190 schedule_work(&adapter->work); 191 } 192} 193 194#ifdef __BIG_ENDIAN_BITFIELD 195/* 196 * The device expects the bitfields in shared structures to be written in 197 * little endian. When CPU is big endian, the following routines are used to 198 * correctly read and write into ABI. 199 * The general technique used here is : double word bitfields are defined in 200 * opposite order for big endian architecture. Then before reading them in 201 * driver the complete double word is translated using le32_to_cpu. Similarly 202 * After the driver writes into bitfields, cpu_to_le32 is used to translate the 203 * double words into required format. 204 * In order to avoid touching bits in shared structure more than once, temporary 205 * descriptors are used. These are passed as srcDesc to following functions. 206 */ 207static void vmxnet3_RxDescToCPU(const struct Vmxnet3_RxDesc *srcDesc, 208 struct Vmxnet3_RxDesc *dstDesc) 209{ 210 u32 *src = (u32 *)srcDesc + 2; 211 u32 *dst = (u32 *)dstDesc + 2; 212 dstDesc->addr = le64_to_cpu(srcDesc->addr); 213 *dst = le32_to_cpu(*src); 214 dstDesc->ext1 = le32_to_cpu(srcDesc->ext1); 215} 216 217static void vmxnet3_TxDescToLe(const struct Vmxnet3_TxDesc *srcDesc, 218 struct Vmxnet3_TxDesc *dstDesc) 219{ 220 int i; 221 u32 *src = (u32 *)(srcDesc + 1); 222 u32 *dst = (u32 *)(dstDesc + 1); 223 224 /* Working backwards so that the gen bit is set at the end. */ 225 for (i = 2; i > 0; i--) { 226 src--; 227 dst--; 228 *dst = cpu_to_le32(*src); 229 } 230} 231 232 233static void vmxnet3_RxCompToCPU(const struct Vmxnet3_RxCompDesc *srcDesc, 234 struct Vmxnet3_RxCompDesc *dstDesc) 235{ 236 int i = 0; 237 u32 *src = (u32 *)srcDesc; 238 u32 *dst = (u32 *)dstDesc; 239 for (i = 0; i < sizeof(struct Vmxnet3_RxCompDesc) / sizeof(u32); i++) { 240 *dst = le32_to_cpu(*src); 241 src++; 242 dst++; 243 } 244} 245 246 247/* Used to read bitfield values from double words. */ 248static u32 get_bitfield32(const __le32 *bitfield, u32 pos, u32 size) 249{ 250 u32 temp = le32_to_cpu(*bitfield); 251 u32 mask = ((1 << size) - 1) << pos; 252 temp &= mask; 253 temp >>= pos; 254 return temp; 255} 256 257 258 259#endif /* __BIG_ENDIAN_BITFIELD */ 260 261#ifdef __BIG_ENDIAN_BITFIELD 262 263# define VMXNET3_TXDESC_GET_GEN(txdesc) get_bitfield32(((const __le32 *) \ 264 txdesc) + VMXNET3_TXD_GEN_DWORD_SHIFT, \ 265 VMXNET3_TXD_GEN_SHIFT, VMXNET3_TXD_GEN_SIZE) 266# define VMXNET3_TXDESC_GET_EOP(txdesc) get_bitfield32(((const __le32 *) \ 267 txdesc) + VMXNET3_TXD_EOP_DWORD_SHIFT, \ 268 VMXNET3_TXD_EOP_SHIFT, VMXNET3_TXD_EOP_SIZE) 269# define VMXNET3_TCD_GET_GEN(tcd) get_bitfield32(((const __le32 *)tcd) + \ 270 VMXNET3_TCD_GEN_DWORD_SHIFT, VMXNET3_TCD_GEN_SHIFT, \ 271 VMXNET3_TCD_GEN_SIZE) 272# define VMXNET3_TCD_GET_TXIDX(tcd) get_bitfield32((const __le32 *)tcd, \ 273 VMXNET3_TCD_TXIDX_SHIFT, VMXNET3_TCD_TXIDX_SIZE) 274# define vmxnet3_getRxComp(dstrcd, rcd, tmp) do { \ 275 (dstrcd) = (tmp); \ 276 vmxnet3_RxCompToCPU((rcd), (tmp)); \ 277 } while (0) 278# define vmxnet3_getRxDesc(dstrxd, rxd, tmp) do { \ 279 (dstrxd) = (tmp); \ 280 vmxnet3_RxDescToCPU((rxd), (tmp)); \ 281 } while (0) 282 283#else 284 285# define VMXNET3_TXDESC_GET_GEN(txdesc) ((txdesc)->gen) 286# define VMXNET3_TXDESC_GET_EOP(txdesc) ((txdesc)->eop) 287# define VMXNET3_TCD_GET_GEN(tcd) ((tcd)->gen) 288# define VMXNET3_TCD_GET_TXIDX(tcd) ((tcd)->txdIdx) 289# define vmxnet3_getRxComp(dstrcd, rcd, tmp) (dstrcd) = (rcd) 290# define vmxnet3_getRxDesc(dstrxd, rxd, tmp) (dstrxd) = (rxd) 291 292#endif /* __BIG_ENDIAN_BITFIELD */ 293 294 295static void 296vmxnet3_unmap_tx_buf(struct vmxnet3_tx_buf_info *tbi, 297 struct pci_dev *pdev) 298{ 299 if (tbi->map_type == VMXNET3_MAP_SINGLE) 300 pci_unmap_single(pdev, tbi->dma_addr, tbi->len, 301 PCI_DMA_TODEVICE); 302 else if (tbi->map_type == VMXNET3_MAP_PAGE) 303 pci_unmap_page(pdev, tbi->dma_addr, tbi->len, 304 PCI_DMA_TODEVICE); 305 else 306 BUG_ON(tbi->map_type != VMXNET3_MAP_NONE); 307 308 tbi->map_type = VMXNET3_MAP_NONE; /* to help debugging */ 309} 310 311 312static int 313vmxnet3_unmap_pkt(u32 eop_idx, struct vmxnet3_tx_queue *tq, 314 struct pci_dev *pdev, struct vmxnet3_adapter *adapter) 315{ 316 struct sk_buff *skb; 317 int entries = 0; 318 319 /* no out of order completion */ 320 BUG_ON(tq->buf_info[eop_idx].sop_idx != tq->tx_ring.next2comp); 321 BUG_ON(VMXNET3_TXDESC_GET_EOP(&(tq->tx_ring.base[eop_idx].txd)) != 1); 322 323 skb = tq->buf_info[eop_idx].skb; 324 BUG_ON(skb == NULL); 325 tq->buf_info[eop_idx].skb = NULL; 326 327 VMXNET3_INC_RING_IDX_ONLY(eop_idx, tq->tx_ring.size); 328 329 while (tq->tx_ring.next2comp != eop_idx) { 330 vmxnet3_unmap_tx_buf(tq->buf_info + tq->tx_ring.next2comp, 331 pdev); 332 333 /* update next2comp w/o tx_lock. Since we are marking more, 334 * instead of less, tx ring entries avail, the worst case is 335 * that the tx routine incorrectly re-queues a pkt due to 336 * insufficient tx ring entries. 337 */ 338 vmxnet3_cmd_ring_adv_next2comp(&tq->tx_ring); 339 entries++; 340 } 341 342 dev_kfree_skb_any(skb); 343 return entries; 344} 345 346 347static int 348vmxnet3_tq_tx_complete(struct vmxnet3_tx_queue *tq, 349 struct vmxnet3_adapter *adapter) 350{ 351 int completed = 0; 352 union Vmxnet3_GenericDesc *gdesc; 353 354 gdesc = tq->comp_ring.base + tq->comp_ring.next2proc; 355 while (VMXNET3_TCD_GET_GEN(&gdesc->tcd) == tq->comp_ring.gen) { 356 completed += vmxnet3_unmap_pkt(VMXNET3_TCD_GET_TXIDX( 357 &gdesc->tcd), tq, adapter->pdev, 358 adapter); 359 360 vmxnet3_comp_ring_adv_next2proc(&tq->comp_ring); 361 gdesc = tq->comp_ring.base + tq->comp_ring.next2proc; 362 } 363 364 if (completed) { 365 spin_lock(&tq->tx_lock); 366 if (unlikely(vmxnet3_tq_stopped(tq, adapter) && 367 vmxnet3_cmd_ring_desc_avail(&tq->tx_ring) > 368 VMXNET3_WAKE_QUEUE_THRESHOLD(tq) && 369 netif_carrier_ok(adapter->netdev))) { 370 vmxnet3_tq_wake(tq, adapter); 371 } 372 spin_unlock(&tq->tx_lock); 373 } 374 return completed; 375} 376 377 378static void 379vmxnet3_tq_cleanup(struct vmxnet3_tx_queue *tq, 380 struct vmxnet3_adapter *adapter) 381{ 382 int i; 383 384 while (tq->tx_ring.next2comp != tq->tx_ring.next2fill) { 385 struct vmxnet3_tx_buf_info *tbi; 386 union Vmxnet3_GenericDesc *gdesc; 387 388 tbi = tq->buf_info + tq->tx_ring.next2comp; 389 gdesc = tq->tx_ring.base + tq->tx_ring.next2comp; 390 391 vmxnet3_unmap_tx_buf(tbi, adapter->pdev); 392 if (tbi->skb) { 393 dev_kfree_skb_any(tbi->skb); 394 tbi->skb = NULL; 395 } 396 vmxnet3_cmd_ring_adv_next2comp(&tq->tx_ring); 397 } 398 399 /* sanity check, verify all buffers are indeed unmapped and freed */ 400 for (i = 0; i < tq->tx_ring.size; i++) { 401 BUG_ON(tq->buf_info[i].skb != NULL || 402 tq->buf_info[i].map_type != VMXNET3_MAP_NONE); 403 } 404 405 tq->tx_ring.gen = VMXNET3_INIT_GEN; 406 tq->tx_ring.next2fill = tq->tx_ring.next2comp = 0; 407 408 tq->comp_ring.gen = VMXNET3_INIT_GEN; 409 tq->comp_ring.next2proc = 0; 410} 411 412 413void 414vmxnet3_tq_destroy(struct vmxnet3_tx_queue *tq, 415 struct vmxnet3_adapter *adapter) 416{ 417 if (tq->tx_ring.base) { 418 pci_free_consistent(adapter->pdev, tq->tx_ring.size * 419 sizeof(struct Vmxnet3_TxDesc), 420 tq->tx_ring.base, tq->tx_ring.basePA); 421 tq->tx_ring.base = NULL; 422 } 423 if (tq->data_ring.base) { 424 pci_free_consistent(adapter->pdev, tq->data_ring.size * 425 sizeof(struct Vmxnet3_TxDataDesc), 426 tq->data_ring.base, tq->data_ring.basePA); 427 tq->data_ring.base = NULL; 428 } 429 if (tq->comp_ring.base) { 430 pci_free_consistent(adapter->pdev, tq->comp_ring.size * 431 sizeof(struct Vmxnet3_TxCompDesc), 432 tq->comp_ring.base, tq->comp_ring.basePA); 433 tq->comp_ring.base = NULL; 434 } 435 kfree(tq->buf_info); 436 tq->buf_info = NULL; 437} 438 439 440static void 441vmxnet3_tq_init(struct vmxnet3_tx_queue *tq, 442 struct vmxnet3_adapter *adapter) 443{ 444 int i; 445 446 /* reset the tx ring contents to 0 and reset the tx ring states */ 447 memset(tq->tx_ring.base, 0, tq->tx_ring.size * 448 sizeof(struct Vmxnet3_TxDesc)); 449 tq->tx_ring.next2fill = tq->tx_ring.next2comp = 0; 450 tq->tx_ring.gen = VMXNET3_INIT_GEN; 451 452 memset(tq->data_ring.base, 0, tq->data_ring.size * 453 sizeof(struct Vmxnet3_TxDataDesc)); 454 455 /* reset the tx comp ring contents to 0 and reset comp ring states */ 456 memset(tq->comp_ring.base, 0, tq->comp_ring.size * 457 sizeof(struct Vmxnet3_TxCompDesc)); 458 tq->comp_ring.next2proc = 0; 459 tq->comp_ring.gen = VMXNET3_INIT_GEN; 460 461 /* reset the bookkeeping data */ 462 memset(tq->buf_info, 0, sizeof(tq->buf_info[0]) * tq->tx_ring.size); 463 for (i = 0; i < tq->tx_ring.size; i++) 464 tq->buf_info[i].map_type = VMXNET3_MAP_NONE; 465 466 /* stats are not reset */ 467} 468 469 470static int 471vmxnet3_tq_create(struct vmxnet3_tx_queue *tq, 472 struct vmxnet3_adapter *adapter) 473{ 474 BUG_ON(tq->tx_ring.base || tq->data_ring.base || 475 tq->comp_ring.base || tq->buf_info); 476 477 tq->tx_ring.base = pci_alloc_consistent(adapter->pdev, tq->tx_ring.size 478 * sizeof(struct Vmxnet3_TxDesc), 479 &tq->tx_ring.basePA); 480 if (!tq->tx_ring.base) { 481 printk(KERN_ERR "%s: failed to allocate tx ring\n", 482 adapter->netdev->name); 483 goto err; 484 } 485 486 tq->data_ring.base = pci_alloc_consistent(adapter->pdev, 487 tq->data_ring.size * 488 sizeof(struct Vmxnet3_TxDataDesc), 489 &tq->data_ring.basePA); 490 if (!tq->data_ring.base) { 491 printk(KERN_ERR "%s: failed to allocate data ring\n", 492 adapter->netdev->name); 493 goto err; 494 } 495 496 tq->comp_ring.base = pci_alloc_consistent(adapter->pdev, 497 tq->comp_ring.size * 498 sizeof(struct Vmxnet3_TxCompDesc), 499 &tq->comp_ring.basePA); 500 if (!tq->comp_ring.base) { 501 printk(KERN_ERR "%s: failed to allocate tx comp ring\n", 502 adapter->netdev->name); 503 goto err; 504 } 505 506 tq->buf_info = kcalloc(tq->tx_ring.size, sizeof(tq->buf_info[0]), 507 GFP_KERNEL); 508 if (!tq->buf_info) { 509 printk(KERN_ERR "%s: failed to allocate tx bufinfo\n", 510 adapter->netdev->name); 511 goto err; 512 } 513 514 return 0; 515 516err: 517 vmxnet3_tq_destroy(tq, adapter); 518 return -ENOMEM; 519} 520 521 522/* 523 * starting from ring->next2fill, allocate rx buffers for the given ring 524 * of the rx queue and update the rx desc. stop after @num_to_alloc buffers 525 * are allocated or allocation fails 526 */ 527 528static int 529vmxnet3_rq_alloc_rx_buf(struct vmxnet3_rx_queue *rq, u32 ring_idx, 530 int num_to_alloc, struct vmxnet3_adapter *adapter) 531{ 532 int num_allocated = 0; 533 struct vmxnet3_rx_buf_info *rbi_base = rq->buf_info[ring_idx]; 534 struct vmxnet3_cmd_ring *ring = &rq->rx_ring[ring_idx]; 535 u32 val; 536 537 while (num_allocated < num_to_alloc) { 538 struct vmxnet3_rx_buf_info *rbi; 539 union Vmxnet3_GenericDesc *gd; 540 541 rbi = rbi_base + ring->next2fill; 542 gd = ring->base + ring->next2fill; 543 544 if (rbi->buf_type == VMXNET3_RX_BUF_SKB) { 545 if (rbi->skb == NULL) { 546 rbi->skb = dev_alloc_skb(rbi->len + 547 NET_IP_ALIGN); 548 if (unlikely(rbi->skb == NULL)) { 549 rq->stats.rx_buf_alloc_failure++; 550 break; 551 } 552 rbi->skb->dev = adapter->netdev; 553 554 skb_reserve(rbi->skb, NET_IP_ALIGN); 555 rbi->dma_addr = pci_map_single(adapter->pdev, 556 rbi->skb->data, rbi->len, 557 PCI_DMA_FROMDEVICE); 558 } else { 559 /* rx buffer skipped by the device */ 560 } 561 val = VMXNET3_RXD_BTYPE_HEAD << VMXNET3_RXD_BTYPE_SHIFT; 562 } else { 563 BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_PAGE || 564 rbi->len != PAGE_SIZE); 565 566 if (rbi->page == NULL) { 567 rbi->page = alloc_page(GFP_ATOMIC); 568 if (unlikely(rbi->page == NULL)) { 569 rq->stats.rx_buf_alloc_failure++; 570 break; 571 } 572 rbi->dma_addr = pci_map_page(adapter->pdev, 573 rbi->page, 0, PAGE_SIZE, 574 PCI_DMA_FROMDEVICE); 575 } else { 576 /* rx buffers skipped by the device */ 577 } 578 val = VMXNET3_RXD_BTYPE_BODY << VMXNET3_RXD_BTYPE_SHIFT; 579 } 580 581 BUG_ON(rbi->dma_addr == 0); 582 gd->rxd.addr = cpu_to_le64(rbi->dma_addr); 583 gd->dword[2] = cpu_to_le32((ring->gen << VMXNET3_RXD_GEN_SHIFT) 584 | val | rbi->len); 585 586 num_allocated++; 587 vmxnet3_cmd_ring_adv_next2fill(ring); 588 } 589 rq->uncommitted[ring_idx] += num_allocated; 590 591 dev_dbg(&adapter->netdev->dev, 592 "alloc_rx_buf: %d allocated, next2fill %u, next2comp " 593 "%u, uncommited %u\n", num_allocated, ring->next2fill, 594 ring->next2comp, rq->uncommitted[ring_idx]); 595 596 /* so that the device can distinguish a full ring and an empty ring */ 597 BUG_ON(num_allocated != 0 && ring->next2fill == ring->next2comp); 598 599 return num_allocated; 600} 601 602 603static void 604vmxnet3_append_frag(struct sk_buff *skb, struct Vmxnet3_RxCompDesc *rcd, 605 struct vmxnet3_rx_buf_info *rbi) 606{ 607 struct skb_frag_struct *frag = skb_shinfo(skb)->frags + 608 skb_shinfo(skb)->nr_frags; 609 610 BUG_ON(skb_shinfo(skb)->nr_frags >= MAX_SKB_FRAGS); 611 612 frag->page = rbi->page; 613 frag->page_offset = 0; 614 frag->size = rcd->len; 615 skb->data_len += frag->size; 616 skb_shinfo(skb)->nr_frags++; 617} 618 619 620static void 621vmxnet3_map_pkt(struct sk_buff *skb, struct vmxnet3_tx_ctx *ctx, 622 struct vmxnet3_tx_queue *tq, struct pci_dev *pdev, 623 struct vmxnet3_adapter *adapter) 624{ 625 u32 dw2, len; 626 unsigned long buf_offset; 627 int i; 628 union Vmxnet3_GenericDesc *gdesc; 629 struct vmxnet3_tx_buf_info *tbi = NULL; 630 631 BUG_ON(ctx->copy_size > skb_headlen(skb)); 632 633 /* use the previous gen bit for the SOP desc */ 634 dw2 = (tq->tx_ring.gen ^ 0x1) << VMXNET3_TXD_GEN_SHIFT; 635 636 ctx->sop_txd = tq->tx_ring.base + tq->tx_ring.next2fill; 637 gdesc = ctx->sop_txd; /* both loops below can be skipped */ 638 639 /* no need to map the buffer if headers are copied */ 640 if (ctx->copy_size) { 641 ctx->sop_txd->txd.addr = cpu_to_le64(tq->data_ring.basePA + 642 tq->tx_ring.next2fill * 643 sizeof(struct Vmxnet3_TxDataDesc)); 644 ctx->sop_txd->dword[2] = cpu_to_le32(dw2 | ctx->copy_size); 645 ctx->sop_txd->dword[3] = 0; 646 647 tbi = tq->buf_info + tq->tx_ring.next2fill; 648 tbi->map_type = VMXNET3_MAP_NONE; 649 650 dev_dbg(&adapter->netdev->dev, 651 "txd[%u]: 0x%Lx 0x%x 0x%x\n", 652 tq->tx_ring.next2fill, 653 le64_to_cpu(ctx->sop_txd->txd.addr), 654 ctx->sop_txd->dword[2], ctx->sop_txd->dword[3]); 655 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring); 656 657 /* use the right gen for non-SOP desc */ 658 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT; 659 } 660 661 /* linear part can use multiple tx desc if it's big */ 662 len = skb_headlen(skb) - ctx->copy_size; 663 buf_offset = ctx->copy_size; 664 while (len) { 665 u32 buf_size; 666 667 if (len < VMXNET3_MAX_TX_BUF_SIZE) { 668 buf_size = len; 669 dw2 |= len; 670 } else { 671 buf_size = VMXNET3_MAX_TX_BUF_SIZE; 672 /* spec says that for TxDesc.len, 0 == 2^14 */ 673 } 674 675 tbi = tq->buf_info + tq->tx_ring.next2fill; 676 tbi->map_type = VMXNET3_MAP_SINGLE; 677 tbi->dma_addr = pci_map_single(adapter->pdev, 678 skb->data + buf_offset, buf_size, 679 PCI_DMA_TODEVICE); 680 681 tbi->len = buf_size; 682 683 gdesc = tq->tx_ring.base + tq->tx_ring.next2fill; 684 BUG_ON(gdesc->txd.gen == tq->tx_ring.gen); 685 686 gdesc->txd.addr = cpu_to_le64(tbi->dma_addr); 687 gdesc->dword[2] = cpu_to_le32(dw2); 688 gdesc->dword[3] = 0; 689 690 dev_dbg(&adapter->netdev->dev, 691 "txd[%u]: 0x%Lx 0x%x 0x%x\n", 692 tq->tx_ring.next2fill, le64_to_cpu(gdesc->txd.addr), 693 le32_to_cpu(gdesc->dword[2]), gdesc->dword[3]); 694 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring); 695 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT; 696 697 len -= buf_size; 698 buf_offset += buf_size; 699 } 700 701 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 702 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i]; 703 704 tbi = tq->buf_info + tq->tx_ring.next2fill; 705 tbi->map_type = VMXNET3_MAP_PAGE; 706 tbi->dma_addr = pci_map_page(adapter->pdev, frag->page, 707 frag->page_offset, frag->size, 708 PCI_DMA_TODEVICE); 709 710 tbi->len = frag->size; 711 712 gdesc = tq->tx_ring.base + tq->tx_ring.next2fill; 713 BUG_ON(gdesc->txd.gen == tq->tx_ring.gen); 714 715 gdesc->txd.addr = cpu_to_le64(tbi->dma_addr); 716 gdesc->dword[2] = cpu_to_le32(dw2 | frag->size); 717 gdesc->dword[3] = 0; 718 719 dev_dbg(&adapter->netdev->dev, 720 "txd[%u]: 0x%llu %u %u\n", 721 tq->tx_ring.next2fill, le64_to_cpu(gdesc->txd.addr), 722 le32_to_cpu(gdesc->dword[2]), gdesc->dword[3]); 723 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring); 724 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT; 725 } 726 727 ctx->eop_txd = gdesc; 728 729 /* set the last buf_info for the pkt */ 730 tbi->skb = skb; 731 tbi->sop_idx = ctx->sop_txd - tq->tx_ring.base; 732} 733 734 735/* 736 * parse and copy relevant protocol headers: 737 * For a tso pkt, relevant headers are L2/3/4 including options 738 * For a pkt requesting csum offloading, they are L2/3 and may include L4 739 * if it's a TCP/UDP pkt 740 * 741 * Returns: 742 * -1: error happens during parsing 743 * 0: protocol headers parsed, but too big to be copied 744 * 1: protocol headers parsed and copied 745 * 746 * Other effects: 747 * 1. related *ctx fields are updated. 748 * 2. ctx->copy_size is # of bytes copied 749 * 3. the portion copied is guaranteed to be in the linear part 750 * 751 */ 752static int 753vmxnet3_parse_and_copy_hdr(struct sk_buff *skb, struct vmxnet3_tx_queue *tq, 754 struct vmxnet3_tx_ctx *ctx, 755 struct vmxnet3_adapter *adapter) 756{ 757 struct Vmxnet3_TxDataDesc *tdd; 758 759 if (ctx->mss) { 760 ctx->eth_ip_hdr_size = skb_transport_offset(skb); 761 ctx->l4_hdr_size = ((struct tcphdr *) 762 skb_transport_header(skb))->doff * 4; 763 ctx->copy_size = ctx->eth_ip_hdr_size + ctx->l4_hdr_size; 764 } else { 765 unsigned int pull_size; 766 767 if (skb->ip_summed == CHECKSUM_PARTIAL) { 768 ctx->eth_ip_hdr_size = skb_transport_offset(skb); 769 770 if (ctx->ipv4) { 771 struct iphdr *iph = (struct iphdr *) 772 skb_network_header(skb); 773 if (iph->protocol == IPPROTO_TCP) { 774 pull_size = ctx->eth_ip_hdr_size + 775 sizeof(struct tcphdr); 776 777 if (unlikely(!pskb_may_pull(skb, 778 pull_size))) { 779 goto err; 780 } 781 ctx->l4_hdr_size = ((struct tcphdr *) 782 skb_transport_header(skb))->doff * 4; 783 } else if (iph->protocol == IPPROTO_UDP) { 784 ctx->l4_hdr_size = 785 sizeof(struct udphdr); 786 } else { 787 ctx->l4_hdr_size = 0; 788 } 789 } else { 790 /* for simplicity, don't copy L4 headers */ 791 ctx->l4_hdr_size = 0; 792 } 793 ctx->copy_size = ctx->eth_ip_hdr_size + 794 ctx->l4_hdr_size; 795 } else { 796 ctx->eth_ip_hdr_size = 0; 797 ctx->l4_hdr_size = 0; 798 /* copy as much as allowed */ 799 ctx->copy_size = min((unsigned int)VMXNET3_HDR_COPY_SIZE 800 , skb_headlen(skb)); 801 } 802 803 /* make sure headers are accessible directly */ 804 if (unlikely(!pskb_may_pull(skb, ctx->copy_size))) 805 goto err; 806 } 807 808 if (unlikely(ctx->copy_size > VMXNET3_HDR_COPY_SIZE)) { 809 tq->stats.oversized_hdr++; 810 ctx->copy_size = 0; 811 return 0; 812 } 813 814 tdd = tq->data_ring.base + tq->tx_ring.next2fill; 815 816 memcpy(tdd->data, skb->data, ctx->copy_size); 817 dev_dbg(&adapter->netdev->dev, 818 "copy %u bytes to dataRing[%u]\n", 819 ctx->copy_size, tq->tx_ring.next2fill); 820 return 1; 821 822err: 823 return -1; 824} 825 826 827static void 828vmxnet3_prepare_tso(struct sk_buff *skb, 829 struct vmxnet3_tx_ctx *ctx) 830{ 831 struct tcphdr *tcph = (struct tcphdr *)skb_transport_header(skb); 832 if (ctx->ipv4) { 833 struct iphdr *iph = (struct iphdr *)skb_network_header(skb); 834 iph->check = 0; 835 tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, 0, 836 IPPROTO_TCP, 0); 837 } else { 838 struct ipv6hdr *iph = (struct ipv6hdr *)skb_network_header(skb); 839 tcph->check = ~csum_ipv6_magic(&iph->saddr, &iph->daddr, 0, 840 IPPROTO_TCP, 0); 841 } 842} 843 844 845/* 846 * Transmits a pkt thru a given tq 847 * Returns: 848 * NETDEV_TX_OK: descriptors are setup successfully 849 * NETDEV_TX_OK: error occured, the pkt is dropped 850 * NETDEV_TX_BUSY: tx ring is full, queue is stopped 851 * 852 * Side-effects: 853 * 1. tx ring may be changed 854 * 2. tq stats may be updated accordingly 855 * 3. shared->txNumDeferred may be updated 856 */ 857 858static int 859vmxnet3_tq_xmit(struct sk_buff *skb, struct vmxnet3_tx_queue *tq, 860 struct vmxnet3_adapter *adapter, struct net_device *netdev) 861{ 862 int ret; 863 u32 count; 864 unsigned long flags; 865 struct vmxnet3_tx_ctx ctx; 866 union Vmxnet3_GenericDesc *gdesc; 867#ifdef __BIG_ENDIAN_BITFIELD 868 /* Use temporary descriptor to avoid touching bits multiple times */ 869 union Vmxnet3_GenericDesc tempTxDesc; 870#endif 871 872 /* conservatively estimate # of descriptors to use */ 873 count = VMXNET3_TXD_NEEDED(skb_headlen(skb)) + 874 skb_shinfo(skb)->nr_frags + 1; 875 876 ctx.ipv4 = (skb->protocol == __constant_ntohs(ETH_P_IP)); 877 878 ctx.mss = skb_shinfo(skb)->gso_size; 879 if (ctx.mss) { 880 if (skb_header_cloned(skb)) { 881 if (unlikely(pskb_expand_head(skb, 0, 0, 882 GFP_ATOMIC) != 0)) { 883 tq->stats.drop_tso++; 884 goto drop_pkt; 885 } 886 tq->stats.copy_skb_header++; 887 } 888 vmxnet3_prepare_tso(skb, &ctx); 889 } else { 890 if (unlikely(count > VMXNET3_MAX_TXD_PER_PKT)) { 891 892 /* non-tso pkts must not use more than 893 * VMXNET3_MAX_TXD_PER_PKT entries 894 */ 895 if (skb_linearize(skb) != 0) { 896 tq->stats.drop_too_many_frags++; 897 goto drop_pkt; 898 } 899 tq->stats.linearized++; 900 901 /* recalculate the # of descriptors to use */ 902 count = VMXNET3_TXD_NEEDED(skb_headlen(skb)) + 1; 903 } 904 } 905 906 ret = vmxnet3_parse_and_copy_hdr(skb, tq, &ctx, adapter); 907 if (ret >= 0) { 908 BUG_ON(ret <= 0 && ctx.copy_size != 0); 909 /* hdrs parsed, check against other limits */ 910 if (ctx.mss) { 911 if (unlikely(ctx.eth_ip_hdr_size + ctx.l4_hdr_size > 912 VMXNET3_MAX_TX_BUF_SIZE)) { 913 goto hdr_too_big; 914 } 915 } else { 916 if (skb->ip_summed == CHECKSUM_PARTIAL) { 917 if (unlikely(ctx.eth_ip_hdr_size + 918 skb->csum_offset > 919 VMXNET3_MAX_CSUM_OFFSET)) { 920 goto hdr_too_big; 921 } 922 } 923 } 924 } else { 925 tq->stats.drop_hdr_inspect_err++; 926 goto drop_pkt; 927 } 928 929 spin_lock_irqsave(&tq->tx_lock, flags); 930 931 if (count > vmxnet3_cmd_ring_desc_avail(&tq->tx_ring)) { 932 tq->stats.tx_ring_full++; 933 dev_dbg(&adapter->netdev->dev, 934 "tx queue stopped on %s, next2comp %u" 935 " next2fill %u\n", adapter->netdev->name, 936 tq->tx_ring.next2comp, tq->tx_ring.next2fill); 937 938 vmxnet3_tq_stop(tq, adapter); 939 spin_unlock_irqrestore(&tq->tx_lock, flags); 940 return NETDEV_TX_BUSY; 941 } 942 943 /* fill tx descs related to addr & len */ 944 vmxnet3_map_pkt(skb, &ctx, tq, adapter->pdev, adapter); 945 946 /* setup the EOP desc */ 947 ctx.eop_txd->dword[3] = cpu_to_le32(VMXNET3_TXD_CQ | VMXNET3_TXD_EOP); 948 949 /* setup the SOP desc */ 950#ifdef __BIG_ENDIAN_BITFIELD 951 gdesc = &tempTxDesc; 952 gdesc->dword[2] = ctx.sop_txd->dword[2]; 953 gdesc->dword[3] = ctx.sop_txd->dword[3]; 954#else 955 gdesc = ctx.sop_txd; 956#endif 957 if (ctx.mss) { 958 gdesc->txd.hlen = ctx.eth_ip_hdr_size + ctx.l4_hdr_size; 959 gdesc->txd.om = VMXNET3_OM_TSO; 960 gdesc->txd.msscof = ctx.mss; 961 le32_add_cpu(&tq->shared->txNumDeferred, (skb->len - 962 gdesc->txd.hlen + ctx.mss - 1) / ctx.mss); 963 } else { 964 if (skb->ip_summed == CHECKSUM_PARTIAL) { 965 gdesc->txd.hlen = ctx.eth_ip_hdr_size; 966 gdesc->txd.om = VMXNET3_OM_CSUM; 967 gdesc->txd.msscof = ctx.eth_ip_hdr_size + 968 skb->csum_offset; 969 } else { 970 gdesc->txd.om = 0; 971 gdesc->txd.msscof = 0; 972 } 973 le32_add_cpu(&tq->shared->txNumDeferred, 1); 974 } 975 976 if (vlan_tx_tag_present(skb)) { 977 gdesc->txd.ti = 1; 978 gdesc->txd.tci = vlan_tx_tag_get(skb); 979 } 980 981 /* finally flips the GEN bit of the SOP desc. */ 982 gdesc->dword[2] = cpu_to_le32(le32_to_cpu(gdesc->dword[2]) ^ 983 VMXNET3_TXD_GEN); 984#ifdef __BIG_ENDIAN_BITFIELD 985 /* Finished updating in bitfields of Tx Desc, so write them in original 986 * place. 987 */ 988 vmxnet3_TxDescToLe((struct Vmxnet3_TxDesc *)gdesc, 989 (struct Vmxnet3_TxDesc *)ctx.sop_txd); 990 gdesc = ctx.sop_txd; 991#endif 992 dev_dbg(&adapter->netdev->dev, 993 "txd[%u]: SOP 0x%Lx 0x%x 0x%x\n", 994 (u32)((union Vmxnet3_GenericDesc *)ctx.sop_txd - 995 tq->tx_ring.base), le64_to_cpu(gdesc->txd.addr), 996 le32_to_cpu(gdesc->dword[2]), le32_to_cpu(gdesc->dword[3])); 997 998 spin_unlock_irqrestore(&tq->tx_lock, flags); 999 1000 if (le32_to_cpu(tq->shared->txNumDeferred) >= 1001 le32_to_cpu(tq->shared->txThreshold)) { 1002 tq->shared->txNumDeferred = 0; 1003 VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_TXPROD, 1004 tq->tx_ring.next2fill); 1005 } 1006 1007 return NETDEV_TX_OK; 1008 1009hdr_too_big: 1010 tq->stats.drop_oversized_hdr++; 1011drop_pkt: 1012 tq->stats.drop_total++; 1013 dev_kfree_skb(skb); 1014 return NETDEV_TX_OK; 1015} 1016 1017 1018static netdev_tx_t 1019vmxnet3_xmit_frame(struct sk_buff *skb, struct net_device *netdev) 1020{ 1021 struct vmxnet3_adapter *adapter = netdev_priv(netdev); 1022 1023 return vmxnet3_tq_xmit(skb, &adapter->tx_queue, adapter, netdev); 1024} 1025 1026 1027static void 1028vmxnet3_rx_csum(struct vmxnet3_adapter *adapter, 1029 struct sk_buff *skb, 1030 union Vmxnet3_GenericDesc *gdesc) 1031{ 1032 if (!gdesc->rcd.cnc && adapter->rxcsum) { 1033 /* typical case: TCP/UDP over IP and both csums are correct */ 1034 if ((le32_to_cpu(gdesc->dword[3]) & VMXNET3_RCD_CSUM_OK) == 1035 VMXNET3_RCD_CSUM_OK) { 1036 skb->ip_summed = CHECKSUM_UNNECESSARY; 1037 BUG_ON(!(gdesc->rcd.tcp || gdesc->rcd.udp)); 1038 BUG_ON(!(gdesc->rcd.v4 || gdesc->rcd.v6)); 1039 BUG_ON(gdesc->rcd.frg); 1040 } else { 1041 if (gdesc->rcd.csum) { 1042 skb->csum = htons(gdesc->rcd.csum); 1043 skb->ip_summed = CHECKSUM_PARTIAL; 1044 } else { 1045 skb_checksum_none_assert(skb); 1046 } 1047 } 1048 } else { 1049 skb_checksum_none_assert(skb); 1050 } 1051} 1052 1053 1054static void 1055vmxnet3_rx_error(struct vmxnet3_rx_queue *rq, struct Vmxnet3_RxCompDesc *rcd, 1056 struct vmxnet3_rx_ctx *ctx, struct vmxnet3_adapter *adapter) 1057{ 1058 rq->stats.drop_err++; 1059 if (!rcd->fcs) 1060 rq->stats.drop_fcs++; 1061 1062 rq->stats.drop_total++; 1063 1064 /* 1065 * We do not unmap and chain the rx buffer to the skb. 1066 * We basically pretend this buffer is not used and will be recycled 1067 * by vmxnet3_rq_alloc_rx_buf() 1068 */ 1069 1070 /* 1071 * ctx->skb may be NULL if this is the first and the only one 1072 * desc for the pkt 1073 */ 1074 if (ctx->skb) 1075 dev_kfree_skb_irq(ctx->skb); 1076 1077 ctx->skb = NULL; 1078} 1079 1080 1081static int 1082vmxnet3_rq_rx_complete(struct vmxnet3_rx_queue *rq, 1083 struct vmxnet3_adapter *adapter, int quota) 1084{ 1085 static u32 rxprod_reg[2] = {VMXNET3_REG_RXPROD, VMXNET3_REG_RXPROD2}; 1086 u32 num_rxd = 0; 1087 struct Vmxnet3_RxCompDesc *rcd; 1088 struct vmxnet3_rx_ctx *ctx = &rq->rx_ctx; 1089#ifdef __BIG_ENDIAN_BITFIELD 1090 struct Vmxnet3_RxDesc rxCmdDesc; 1091 struct Vmxnet3_RxCompDesc rxComp; 1092#endif 1093 vmxnet3_getRxComp(rcd, &rq->comp_ring.base[rq->comp_ring.next2proc].rcd, 1094 &rxComp); 1095 while (rcd->gen == rq->comp_ring.gen) { 1096 struct vmxnet3_rx_buf_info *rbi; 1097 struct sk_buff *skb; 1098 int num_to_alloc; 1099 struct Vmxnet3_RxDesc *rxd; 1100 u32 idx, ring_idx; 1101 1102 if (num_rxd >= quota) { 1103 /* we may stop even before we see the EOP desc of 1104 * the current pkt 1105 */ 1106 break; 1107 } 1108 num_rxd++; 1109 1110 idx = rcd->rxdIdx; 1111 ring_idx = rcd->rqID == rq->qid ? 0 : 1; 1112 vmxnet3_getRxDesc(rxd, &rq->rx_ring[ring_idx].base[idx].rxd, 1113 &rxCmdDesc); 1114 rbi = rq->buf_info[ring_idx] + idx; 1115 1116 BUG_ON(rxd->addr != rbi->dma_addr || 1117 rxd->len != rbi->len); 1118 1119 if (unlikely(rcd->eop && rcd->err)) { 1120 vmxnet3_rx_error(rq, rcd, ctx, adapter); 1121 goto rcd_done; 1122 } 1123 1124 if (rcd->sop) { /* first buf of the pkt */ 1125 BUG_ON(rxd->btype != VMXNET3_RXD_BTYPE_HEAD || 1126 rcd->rqID != rq->qid); 1127 1128 BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_SKB); 1129 BUG_ON(ctx->skb != NULL || rbi->skb == NULL); 1130 1131 if (unlikely(rcd->len == 0)) { 1132 /* Pretend the rx buffer is skipped. */ 1133 BUG_ON(!(rcd->sop && rcd->eop)); 1134 dev_dbg(&adapter->netdev->dev, 1135 "rxRing[%u][%u] 0 length\n", 1136 ring_idx, idx); 1137 goto rcd_done; 1138 } 1139 1140 ctx->skb = rbi->skb; 1141 rbi->skb = NULL; 1142 1143 pci_unmap_single(adapter->pdev, rbi->dma_addr, rbi->len, 1144 PCI_DMA_FROMDEVICE); 1145 1146 skb_put(ctx->skb, rcd->len); 1147 } else { 1148 BUG_ON(ctx->skb == NULL); 1149 /* non SOP buffer must be type 1 in most cases */ 1150 if (rbi->buf_type == VMXNET3_RX_BUF_PAGE) { 1151 BUG_ON(rxd->btype != VMXNET3_RXD_BTYPE_BODY); 1152 1153 if (rcd->len) { 1154 pci_unmap_page(adapter->pdev, 1155 rbi->dma_addr, rbi->len, 1156 PCI_DMA_FROMDEVICE); 1157 1158 vmxnet3_append_frag(ctx->skb, rcd, rbi); 1159 rbi->page = NULL; 1160 } 1161 } else { 1162 /* 1163 * The only time a non-SOP buffer is type 0 is 1164 * when it's EOP and error flag is raised, which 1165 * has already been handled. 1166 */ 1167 BUG_ON(true); 1168 } 1169 } 1170 1171 skb = ctx->skb; 1172 if (rcd->eop) { 1173 skb->len += skb->data_len; 1174 skb->truesize += skb->data_len; 1175 1176 vmxnet3_rx_csum(adapter, skb, 1177 (union Vmxnet3_GenericDesc *)rcd); 1178 skb->protocol = eth_type_trans(skb, adapter->netdev); 1179 1180 if (unlikely(adapter->vlan_grp && rcd->ts)) { 1181 vlan_hwaccel_receive_skb(skb, 1182 adapter->vlan_grp, rcd->tci); 1183 } else { 1184 netif_receive_skb(skb); 1185 } 1186 1187 ctx->skb = NULL; 1188 } 1189 1190rcd_done: 1191 /* device may skip some rx descs */ 1192 rq->rx_ring[ring_idx].next2comp = idx; 1193 VMXNET3_INC_RING_IDX_ONLY(rq->rx_ring[ring_idx].next2comp, 1194 rq->rx_ring[ring_idx].size); 1195 1196 /* refill rx buffers frequently to avoid starving the h/w */ 1197 num_to_alloc = vmxnet3_cmd_ring_desc_avail(rq->rx_ring + 1198 ring_idx); 1199 if (unlikely(num_to_alloc > VMXNET3_RX_ALLOC_THRESHOLD(rq, 1200 ring_idx, adapter))) { 1201 vmxnet3_rq_alloc_rx_buf(rq, ring_idx, num_to_alloc, 1202 adapter); 1203 1204 /* if needed, update the register */ 1205 if (unlikely(rq->shared->updateRxProd)) { 1206 VMXNET3_WRITE_BAR0_REG(adapter, 1207 rxprod_reg[ring_idx] + rq->qid * 8, 1208 rq->rx_ring[ring_idx].next2fill); 1209 rq->uncommitted[ring_idx] = 0; 1210 } 1211 } 1212 1213 vmxnet3_comp_ring_adv_next2proc(&rq->comp_ring); 1214 vmxnet3_getRxComp(rcd, 1215 &rq->comp_ring.base[rq->comp_ring.next2proc].rcd, &rxComp); 1216 } 1217 1218 return num_rxd; 1219} 1220 1221 1222static void 1223vmxnet3_rq_cleanup(struct vmxnet3_rx_queue *rq, 1224 struct vmxnet3_adapter *adapter) 1225{ 1226 u32 i, ring_idx; 1227 struct Vmxnet3_RxDesc *rxd; 1228 1229 for (ring_idx = 0; ring_idx < 2; ring_idx++) { 1230 for (i = 0; i < rq->rx_ring[ring_idx].size; i++) { 1231#ifdef __BIG_ENDIAN_BITFIELD 1232 struct Vmxnet3_RxDesc rxDesc; 1233#endif 1234 vmxnet3_getRxDesc(rxd, 1235 &rq->rx_ring[ring_idx].base[i].rxd, &rxDesc); 1236 1237 if (rxd->btype == VMXNET3_RXD_BTYPE_HEAD && 1238 rq->buf_info[ring_idx][i].skb) { 1239 pci_unmap_single(adapter->pdev, rxd->addr, 1240 rxd->len, PCI_DMA_FROMDEVICE); 1241 dev_kfree_skb(rq->buf_info[ring_idx][i].skb); 1242 rq->buf_info[ring_idx][i].skb = NULL; 1243 } else if (rxd->btype == VMXNET3_RXD_BTYPE_BODY && 1244 rq->buf_info[ring_idx][i].page) { 1245 pci_unmap_page(adapter->pdev, rxd->addr, 1246 rxd->len, PCI_DMA_FROMDEVICE); 1247 put_page(rq->buf_info[ring_idx][i].page); 1248 rq->buf_info[ring_idx][i].page = NULL; 1249 } 1250 } 1251 1252 rq->rx_ring[ring_idx].gen = VMXNET3_INIT_GEN; 1253 rq->rx_ring[ring_idx].next2fill = 1254 rq->rx_ring[ring_idx].next2comp = 0; 1255 rq->uncommitted[ring_idx] = 0; 1256 } 1257 1258 rq->comp_ring.gen = VMXNET3_INIT_GEN; 1259 rq->comp_ring.next2proc = 0; 1260} 1261 1262 1263void vmxnet3_rq_destroy(struct vmxnet3_rx_queue *rq, 1264 struct vmxnet3_adapter *adapter) 1265{ 1266 int i; 1267 int j; 1268 1269 /* all rx buffers must have already been freed */ 1270 for (i = 0; i < 2; i++) { 1271 if (rq->buf_info[i]) { 1272 for (j = 0; j < rq->rx_ring[i].size; j++) 1273 BUG_ON(rq->buf_info[i][j].page != NULL); 1274 } 1275 } 1276 1277 1278 kfree(rq->buf_info[0]); 1279 1280 for (i = 0; i < 2; i++) { 1281 if (rq->rx_ring[i].base) { 1282 pci_free_consistent(adapter->pdev, rq->rx_ring[i].size 1283 * sizeof(struct Vmxnet3_RxDesc), 1284 rq->rx_ring[i].base, 1285 rq->rx_ring[i].basePA); 1286 rq->rx_ring[i].base = NULL; 1287 } 1288 rq->buf_info[i] = NULL; 1289 } 1290 1291 if (rq->comp_ring.base) { 1292 pci_free_consistent(adapter->pdev, rq->comp_ring.size * 1293 sizeof(struct Vmxnet3_RxCompDesc), 1294 rq->comp_ring.base, rq->comp_ring.basePA); 1295 rq->comp_ring.base = NULL; 1296 } 1297} 1298 1299 1300static int 1301vmxnet3_rq_init(struct vmxnet3_rx_queue *rq, 1302 struct vmxnet3_adapter *adapter) 1303{ 1304 int i; 1305 1306 /* initialize buf_info */ 1307 for (i = 0; i < rq->rx_ring[0].size; i++) { 1308 1309 /* 1st buf for a pkt is skbuff */ 1310 if (i % adapter->rx_buf_per_pkt == 0) { 1311 rq->buf_info[0][i].buf_type = VMXNET3_RX_BUF_SKB; 1312 rq->buf_info[0][i].len = adapter->skb_buf_size; 1313 } else { /* subsequent bufs for a pkt is frag */ 1314 rq->buf_info[0][i].buf_type = VMXNET3_RX_BUF_PAGE; 1315 rq->buf_info[0][i].len = PAGE_SIZE; 1316 } 1317 } 1318 for (i = 0; i < rq->rx_ring[1].size; i++) { 1319 rq->buf_info[1][i].buf_type = VMXNET3_RX_BUF_PAGE; 1320 rq->buf_info[1][i].len = PAGE_SIZE; 1321 } 1322 1323 /* reset internal state and allocate buffers for both rings */ 1324 for (i = 0; i < 2; i++) { 1325 rq->rx_ring[i].next2fill = rq->rx_ring[i].next2comp = 0; 1326 rq->uncommitted[i] = 0; 1327 1328 memset(rq->rx_ring[i].base, 0, rq->rx_ring[i].size * 1329 sizeof(struct Vmxnet3_RxDesc)); 1330 rq->rx_ring[i].gen = VMXNET3_INIT_GEN; 1331 } 1332 if (vmxnet3_rq_alloc_rx_buf(rq, 0, rq->rx_ring[0].size - 1, 1333 adapter) == 0) { 1334 /* at least has 1 rx buffer for the 1st ring */ 1335 return -ENOMEM; 1336 } 1337 vmxnet3_rq_alloc_rx_buf(rq, 1, rq->rx_ring[1].size - 1, adapter); 1338 1339 /* reset the comp ring */ 1340 rq->comp_ring.next2proc = 0; 1341 memset(rq->comp_ring.base, 0, rq->comp_ring.size * 1342 sizeof(struct Vmxnet3_RxCompDesc)); 1343 rq->comp_ring.gen = VMXNET3_INIT_GEN; 1344 1345 /* reset rxctx */ 1346 rq->rx_ctx.skb = NULL; 1347 1348 /* stats are not reset */ 1349 return 0; 1350} 1351 1352 1353static int 1354vmxnet3_rq_create(struct vmxnet3_rx_queue *rq, struct vmxnet3_adapter *adapter) 1355{ 1356 int i; 1357 size_t sz; 1358 struct vmxnet3_rx_buf_info *bi; 1359 1360 for (i = 0; i < 2; i++) { 1361 1362 sz = rq->rx_ring[i].size * sizeof(struct Vmxnet3_RxDesc); 1363 rq->rx_ring[i].base = pci_alloc_consistent(adapter->pdev, sz, 1364 &rq->rx_ring[i].basePA); 1365 if (!rq->rx_ring[i].base) { 1366 printk(KERN_ERR "%s: failed to allocate rx ring %d\n", 1367 adapter->netdev->name, i); 1368 goto err; 1369 } 1370 } 1371 1372 sz = rq->comp_ring.size * sizeof(struct Vmxnet3_RxCompDesc); 1373 rq->comp_ring.base = pci_alloc_consistent(adapter->pdev, sz, 1374 &rq->comp_ring.basePA); 1375 if (!rq->comp_ring.base) { 1376 printk(KERN_ERR "%s: failed to allocate rx comp ring\n", 1377 adapter->netdev->name); 1378 goto err; 1379 } 1380 1381 sz = sizeof(struct vmxnet3_rx_buf_info) * (rq->rx_ring[0].size + 1382 rq->rx_ring[1].size); 1383 bi = kzalloc(sz, GFP_KERNEL); 1384 if (!bi) { 1385 printk(KERN_ERR "%s: failed to allocate rx bufinfo\n", 1386 adapter->netdev->name); 1387 goto err; 1388 } 1389 rq->buf_info[0] = bi; 1390 rq->buf_info[1] = bi + rq->rx_ring[0].size; 1391 1392 return 0; 1393 1394err: 1395 vmxnet3_rq_destroy(rq, adapter); 1396 return -ENOMEM; 1397} 1398 1399 1400static int 1401vmxnet3_do_poll(struct vmxnet3_adapter *adapter, int budget) 1402{ 1403 if (unlikely(adapter->shared->ecr)) 1404 vmxnet3_process_events(adapter); 1405 1406 vmxnet3_tq_tx_complete(&adapter->tx_queue, adapter); 1407 return vmxnet3_rq_rx_complete(&adapter->rx_queue, adapter, budget); 1408} 1409 1410 1411static int 1412vmxnet3_poll(struct napi_struct *napi, int budget) 1413{ 1414 struct vmxnet3_adapter *adapter = container_of(napi, 1415 struct vmxnet3_adapter, napi); 1416 int rxd_done; 1417 1418 rxd_done = vmxnet3_do_poll(adapter, budget); 1419 1420 if (rxd_done < budget) { 1421 napi_complete(napi); 1422 vmxnet3_enable_intr(adapter, 0); 1423 } 1424 return rxd_done; 1425} 1426 1427 1428/* Interrupt handler for vmxnet3 */ 1429static irqreturn_t 1430vmxnet3_intr(int irq, void *dev_id) 1431{ 1432 struct net_device *dev = dev_id; 1433 struct vmxnet3_adapter *adapter = netdev_priv(dev); 1434 1435 if (unlikely(adapter->intr.type == VMXNET3_IT_INTX)) { 1436 u32 icr = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_ICR); 1437 if (unlikely(icr == 0)) 1438 /* not ours */ 1439 return IRQ_NONE; 1440 } 1441 1442 1443 /* disable intr if needed */ 1444 if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE) 1445 vmxnet3_disable_intr(adapter, 0); 1446 1447 napi_schedule(&adapter->napi); 1448 1449 return IRQ_HANDLED; 1450} 1451 1452#ifdef CONFIG_NET_POLL_CONTROLLER 1453 1454 1455/* netpoll callback. */ 1456static void 1457vmxnet3_netpoll(struct net_device *netdev) 1458{ 1459 struct vmxnet3_adapter *adapter = netdev_priv(netdev); 1460 int irq; 1461 1462#ifdef CONFIG_PCI_MSI 1463 if (adapter->intr.type == VMXNET3_IT_MSIX) 1464 irq = adapter->intr.msix_entries[0].vector; 1465 else 1466#endif 1467 irq = adapter->pdev->irq; 1468 1469 disable_irq(irq); 1470 vmxnet3_intr(irq, netdev); 1471 enable_irq(irq); 1472} 1473#endif 1474 1475static int 1476vmxnet3_request_irqs(struct vmxnet3_adapter *adapter) 1477{ 1478 int err; 1479 1480#ifdef CONFIG_PCI_MSI 1481 if (adapter->intr.type == VMXNET3_IT_MSIX) { 1482 /* we only use 1 MSI-X vector */ 1483 err = request_irq(adapter->intr.msix_entries[0].vector, 1484 vmxnet3_intr, 0, adapter->netdev->name, 1485 adapter->netdev); 1486 } else if (adapter->intr.type == VMXNET3_IT_MSI) { 1487 err = request_irq(adapter->pdev->irq, vmxnet3_intr, 0, 1488 adapter->netdev->name, adapter->netdev); 1489 } else 1490#endif 1491 { 1492 err = request_irq(adapter->pdev->irq, vmxnet3_intr, 1493 IRQF_SHARED, adapter->netdev->name, 1494 adapter->netdev); 1495 } 1496 1497 if (err) 1498 printk(KERN_ERR "Failed to request irq %s (intr type:%d), error" 1499 ":%d\n", adapter->netdev->name, adapter->intr.type, err); 1500 1501 1502 if (!err) { 1503 int i; 1504 /* init our intr settings */ 1505 for (i = 0; i < adapter->intr.num_intrs; i++) 1506 adapter->intr.mod_levels[i] = UPT1_IML_ADAPTIVE; 1507 1508 /* next setup intr index for all intr sources */ 1509 adapter->tx_queue.comp_ring.intr_idx = 0; 1510 adapter->rx_queue.comp_ring.intr_idx = 0; 1511 adapter->intr.event_intr_idx = 0; 1512 1513 printk(KERN_INFO "%s: intr type %u, mode %u, %u vectors " 1514 "allocated\n", adapter->netdev->name, adapter->intr.type, 1515 adapter->intr.mask_mode, adapter->intr.num_intrs); 1516 } 1517 1518 return err; 1519} 1520 1521 1522static void 1523vmxnet3_free_irqs(struct vmxnet3_adapter *adapter) 1524{ 1525 BUG_ON(adapter->intr.type == VMXNET3_IT_AUTO || 1526 adapter->intr.num_intrs <= 0); 1527 1528 switch (adapter->intr.type) { 1529#ifdef CONFIG_PCI_MSI 1530 case VMXNET3_IT_MSIX: 1531 { 1532 int i; 1533 1534 for (i = 0; i < adapter->intr.num_intrs; i++) 1535 free_irq(adapter->intr.msix_entries[i].vector, 1536 adapter->netdev); 1537 break; 1538 } 1539#endif 1540 case VMXNET3_IT_MSI: 1541 free_irq(adapter->pdev->irq, adapter->netdev); 1542 break; 1543 case VMXNET3_IT_INTX: 1544 free_irq(adapter->pdev->irq, adapter->netdev); 1545 break; 1546 default: 1547 BUG_ON(true); 1548 } 1549} 1550 1551 1552inline void set_flag_le16(__le16 *data, u16 flag) 1553{ 1554 *data = cpu_to_le16(le16_to_cpu(*data) | flag); 1555} 1556 1557inline void set_flag_le64(__le64 *data, u64 flag) 1558{ 1559 *data = cpu_to_le64(le64_to_cpu(*data) | flag); 1560} 1561 1562inline void reset_flag_le64(__le64 *data, u64 flag) 1563{ 1564 *data = cpu_to_le64(le64_to_cpu(*data) & ~flag); 1565} 1566 1567 1568static void 1569vmxnet3_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp) 1570{ 1571 struct vmxnet3_adapter *adapter = netdev_priv(netdev); 1572 struct Vmxnet3_DriverShared *shared = adapter->shared; 1573 u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable; 1574 1575 if (grp) { 1576 /* add vlan rx stripping. */ 1577 if (adapter->netdev->features & NETIF_F_HW_VLAN_RX) { 1578 int i; 1579 struct Vmxnet3_DSDevRead *devRead = &shared->devRead; 1580 adapter->vlan_grp = grp; 1581 1582 /* update FEATURES to device */ 1583 set_flag_le64(&devRead->misc.uptFeatures, 1584 UPT1_F_RXVLAN); 1585 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, 1586 VMXNET3_CMD_UPDATE_FEATURE); 1587 /* 1588 * Clear entire vfTable; then enable untagged pkts. 1589 * Note: setting one entry in vfTable to non-zero turns 1590 * on VLAN rx filtering. 1591 */ 1592 for (i = 0; i < VMXNET3_VFT_SIZE; i++) 1593 vfTable[i] = 0; 1594 1595 VMXNET3_SET_VFTABLE_ENTRY(vfTable, 0); 1596 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, 1597 VMXNET3_CMD_UPDATE_VLAN_FILTERS); 1598 } else { 1599 printk(KERN_ERR "%s: vlan_rx_register when device has " 1600 "no NETIF_F_HW_VLAN_RX\n", netdev->name); 1601 } 1602 } else { 1603 /* remove vlan rx stripping. */ 1604 struct Vmxnet3_DSDevRead *devRead = &shared->devRead; 1605 adapter->vlan_grp = NULL; 1606 1607 if (le64_to_cpu(devRead->misc.uptFeatures) & UPT1_F_RXVLAN) { 1608 int i; 1609 1610 for (i = 0; i < VMXNET3_VFT_SIZE; i++) { 1611 /* clear entire vfTable; this also disables 1612 * VLAN rx filtering 1613 */ 1614 vfTable[i] = 0; 1615 } 1616 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, 1617 VMXNET3_CMD_UPDATE_VLAN_FILTERS); 1618 1619 /* update FEATURES to device */ 1620 reset_flag_le64(&devRead->misc.uptFeatures, 1621 UPT1_F_RXVLAN); 1622 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, 1623 VMXNET3_CMD_UPDATE_FEATURE); 1624 } 1625 } 1626} 1627 1628 1629static void 1630vmxnet3_restore_vlan(struct vmxnet3_adapter *adapter) 1631{ 1632 if (adapter->vlan_grp) { 1633 u16 vid; 1634 u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable; 1635 bool activeVlan = false; 1636 1637 for (vid = 0; vid < VLAN_N_VID; vid++) { 1638 if (vlan_group_get_device(adapter->vlan_grp, vid)) { 1639 VMXNET3_SET_VFTABLE_ENTRY(vfTable, vid); 1640 activeVlan = true; 1641 } 1642 } 1643 if (activeVlan) { 1644 /* continue to allow untagged pkts */ 1645 VMXNET3_SET_VFTABLE_ENTRY(vfTable, 0); 1646 } 1647 } 1648} 1649 1650 1651static void 1652vmxnet3_vlan_rx_add_vid(struct net_device *netdev, u16 vid) 1653{ 1654 struct vmxnet3_adapter *adapter = netdev_priv(netdev); 1655 u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable; 1656 1657 VMXNET3_SET_VFTABLE_ENTRY(vfTable, vid); 1658 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, 1659 VMXNET3_CMD_UPDATE_VLAN_FILTERS); 1660} 1661 1662 1663static void 1664vmxnet3_vlan_rx_kill_vid(struct net_device *netdev, u16 vid) 1665{ 1666 struct vmxnet3_adapter *adapter = netdev_priv(netdev); 1667 u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable; 1668 1669 VMXNET3_CLEAR_VFTABLE_ENTRY(vfTable, vid); 1670 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, 1671 VMXNET3_CMD_UPDATE_VLAN_FILTERS); 1672} 1673 1674 1675static u8 * 1676vmxnet3_copy_mc(struct net_device *netdev) 1677{ 1678 u8 *buf = NULL; 1679 u32 sz = netdev_mc_count(netdev) * ETH_ALEN; 1680 1681 /* struct Vmxnet3_RxFilterConf.mfTableLen is u16. */ 1682 if (sz <= 0xffff) { 1683 /* We may be called with BH disabled */ 1684 buf = kmalloc(sz, GFP_ATOMIC); 1685 if (buf) { 1686 struct netdev_hw_addr *ha; 1687 int i = 0; 1688 1689 netdev_for_each_mc_addr(ha, netdev) 1690 memcpy(buf + i++ * ETH_ALEN, ha->addr, 1691 ETH_ALEN); 1692 } 1693 } 1694 return buf; 1695} 1696 1697 1698static void 1699vmxnet3_set_mc(struct net_device *netdev) 1700{ 1701 struct vmxnet3_adapter *adapter = netdev_priv(netdev); 1702 struct Vmxnet3_RxFilterConf *rxConf = 1703 &adapter->shared->devRead.rxFilterConf; 1704 u8 *new_table = NULL; 1705 u32 new_mode = VMXNET3_RXM_UCAST; 1706 1707 if (netdev->flags & IFF_PROMISC) 1708 new_mode |= VMXNET3_RXM_PROMISC; 1709 1710 if (netdev->flags & IFF_BROADCAST) 1711 new_mode |= VMXNET3_RXM_BCAST; 1712 1713 if (netdev->flags & IFF_ALLMULTI) 1714 new_mode |= VMXNET3_RXM_ALL_MULTI; 1715 else 1716 if (!netdev_mc_empty(netdev)) { 1717 new_table = vmxnet3_copy_mc(netdev); 1718 if (new_table) { 1719 new_mode |= VMXNET3_RXM_MCAST; 1720 rxConf->mfTableLen = cpu_to_le16( 1721 netdev_mc_count(netdev) * ETH_ALEN); 1722 rxConf->mfTablePA = cpu_to_le64(virt_to_phys( 1723 new_table)); 1724 } else { 1725 printk(KERN_INFO "%s: failed to copy mcast list" 1726 ", setting ALL_MULTI\n", netdev->name); 1727 new_mode |= VMXNET3_RXM_ALL_MULTI; 1728 } 1729 } 1730 1731 1732 if (!(new_mode & VMXNET3_RXM_MCAST)) { 1733 rxConf->mfTableLen = 0; 1734 rxConf->mfTablePA = 0; 1735 } 1736 1737 if (new_mode != rxConf->rxMode) { 1738 rxConf->rxMode = cpu_to_le32(new_mode); 1739 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, 1740 VMXNET3_CMD_UPDATE_RX_MODE); 1741 } 1742 1743 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, 1744 VMXNET3_CMD_UPDATE_MAC_FILTERS); 1745 1746 kfree(new_table); 1747} 1748 1749 1750/* 1751 * Set up driver_shared based on settings in adapter. 1752 */ 1753 1754static void 1755vmxnet3_setup_driver_shared(struct vmxnet3_adapter *adapter) 1756{ 1757 struct Vmxnet3_DriverShared *shared = adapter->shared; 1758 struct Vmxnet3_DSDevRead *devRead = &shared->devRead; 1759 struct Vmxnet3_TxQueueConf *tqc; 1760 struct Vmxnet3_RxQueueConf *rqc; 1761 int i; 1762 1763 memset(shared, 0, sizeof(*shared)); 1764 1765 /* driver settings */ 1766 shared->magic = cpu_to_le32(VMXNET3_REV1_MAGIC); 1767 devRead->misc.driverInfo.version = cpu_to_le32( 1768 VMXNET3_DRIVER_VERSION_NUM); 1769 devRead->misc.driverInfo.gos.gosBits = (sizeof(void *) == 4 ? 1770 VMXNET3_GOS_BITS_32 : VMXNET3_GOS_BITS_64); 1771 devRead->misc.driverInfo.gos.gosType = VMXNET3_GOS_TYPE_LINUX; 1772 *((u32 *)&devRead->misc.driverInfo.gos) = cpu_to_le32( 1773 *((u32 *)&devRead->misc.driverInfo.gos)); 1774 devRead->misc.driverInfo.vmxnet3RevSpt = cpu_to_le32(1); 1775 devRead->misc.driverInfo.uptVerSpt = cpu_to_le32(1); 1776 1777 devRead->misc.ddPA = cpu_to_le64(virt_to_phys(adapter)); 1778 devRead->misc.ddLen = cpu_to_le32(sizeof(struct vmxnet3_adapter)); 1779 1780 /* set up feature flags */ 1781 if (adapter->rxcsum) 1782 set_flag_le64(&devRead->misc.uptFeatures, UPT1_F_RXCSUM); 1783 1784 if (adapter->lro) { 1785 set_flag_le64(&devRead->misc.uptFeatures, UPT1_F_LRO); 1786 devRead->misc.maxNumRxSG = cpu_to_le16(1 + MAX_SKB_FRAGS); 1787 } 1788 if ((adapter->netdev->features & NETIF_F_HW_VLAN_RX) && 1789 adapter->vlan_grp) { 1790 set_flag_le64(&devRead->misc.uptFeatures, UPT1_F_RXVLAN); 1791 } 1792 1793 devRead->misc.mtu = cpu_to_le32(adapter->netdev->mtu); 1794 devRead->misc.queueDescPA = cpu_to_le64(adapter->queue_desc_pa); 1795 devRead->misc.queueDescLen = cpu_to_le32( 1796 sizeof(struct Vmxnet3_TxQueueDesc) + 1797 sizeof(struct Vmxnet3_RxQueueDesc)); 1798 1799 /* tx queue settings */ 1800 BUG_ON(adapter->tx_queue.tx_ring.base == NULL); 1801 1802 devRead->misc.numTxQueues = 1; 1803 tqc = &adapter->tqd_start->conf; 1804 tqc->txRingBasePA = cpu_to_le64(adapter->tx_queue.tx_ring.basePA); 1805 tqc->dataRingBasePA = cpu_to_le64(adapter->tx_queue.data_ring.basePA); 1806 tqc->compRingBasePA = cpu_to_le64(adapter->tx_queue.comp_ring.basePA); 1807 tqc->ddPA = cpu_to_le64(virt_to_phys( 1808 adapter->tx_queue.buf_info)); 1809 tqc->txRingSize = cpu_to_le32(adapter->tx_queue.tx_ring.size); 1810 tqc->dataRingSize = cpu_to_le32(adapter->tx_queue.data_ring.size); 1811 tqc->compRingSize = cpu_to_le32(adapter->tx_queue.comp_ring.size); 1812 tqc->ddLen = cpu_to_le32(sizeof(struct vmxnet3_tx_buf_info) * 1813 tqc->txRingSize); 1814 tqc->intrIdx = adapter->tx_queue.comp_ring.intr_idx; 1815 1816 /* rx queue settings */ 1817 devRead->misc.numRxQueues = 1; 1818 rqc = &adapter->rqd_start->conf; 1819 rqc->rxRingBasePA[0] = cpu_to_le64(adapter->rx_queue.rx_ring[0].basePA); 1820 rqc->rxRingBasePA[1] = cpu_to_le64(adapter->rx_queue.rx_ring[1].basePA); 1821 rqc->compRingBasePA = cpu_to_le64(adapter->rx_queue.comp_ring.basePA); 1822 rqc->ddPA = cpu_to_le64(virt_to_phys( 1823 adapter->rx_queue.buf_info)); 1824 rqc->rxRingSize[0] = cpu_to_le32(adapter->rx_queue.rx_ring[0].size); 1825 rqc->rxRingSize[1] = cpu_to_le32(adapter->rx_queue.rx_ring[1].size); 1826 rqc->compRingSize = cpu_to_le32(adapter->rx_queue.comp_ring.size); 1827 rqc->ddLen = cpu_to_le32(sizeof(struct vmxnet3_rx_buf_info) * 1828 (rqc->rxRingSize[0] + rqc->rxRingSize[1])); 1829 rqc->intrIdx = adapter->rx_queue.comp_ring.intr_idx; 1830 1831 /* intr settings */ 1832 devRead->intrConf.autoMask = adapter->intr.mask_mode == 1833 VMXNET3_IMM_AUTO; 1834 devRead->intrConf.numIntrs = adapter->intr.num_intrs; 1835 for (i = 0; i < adapter->intr.num_intrs; i++) 1836 devRead->intrConf.modLevels[i] = adapter->intr.mod_levels[i]; 1837 1838 devRead->intrConf.eventIntrIdx = adapter->intr.event_intr_idx; 1839 devRead->intrConf.intrCtrl |= cpu_to_le32(VMXNET3_IC_DISABLE_ALL); 1840 1841 /* rx filter settings */ 1842 devRead->rxFilterConf.rxMode = 0; 1843 vmxnet3_restore_vlan(adapter); 1844 /* the rest are already zeroed */ 1845} 1846 1847 1848int 1849vmxnet3_activate_dev(struct vmxnet3_adapter *adapter) 1850{ 1851 int err; 1852 u32 ret; 1853 1854 dev_dbg(&adapter->netdev->dev, 1855 "%s: skb_buf_size %d, rx_buf_per_pkt %d, ring sizes" 1856 " %u %u %u\n", adapter->netdev->name, adapter->skb_buf_size, 1857 adapter->rx_buf_per_pkt, adapter->tx_queue.tx_ring.size, 1858 adapter->rx_queue.rx_ring[0].size, 1859 adapter->rx_queue.rx_ring[1].size); 1860 1861 vmxnet3_tq_init(&adapter->tx_queue, adapter); 1862 err = vmxnet3_rq_init(&adapter->rx_queue, adapter); 1863 if (err) { 1864 printk(KERN_ERR "Failed to init rx queue for %s: error %d\n", 1865 adapter->netdev->name, err); 1866 goto rq_err; 1867 } 1868 1869 err = vmxnet3_request_irqs(adapter); 1870 if (err) { 1871 printk(KERN_ERR "Failed to setup irq for %s: error %d\n", 1872 adapter->netdev->name, err); 1873 goto irq_err; 1874 } 1875 1876 vmxnet3_setup_driver_shared(adapter); 1877 1878 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL, VMXNET3_GET_ADDR_LO( 1879 adapter->shared_pa)); 1880 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH, VMXNET3_GET_ADDR_HI( 1881 adapter->shared_pa)); 1882 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, 1883 VMXNET3_CMD_ACTIVATE_DEV); 1884 ret = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD); 1885 1886 if (ret != 0) { 1887 printk(KERN_ERR "Failed to activate dev %s: error %u\n", 1888 adapter->netdev->name, ret); 1889 err = -EINVAL; 1890 goto activate_err; 1891 } 1892 VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_RXPROD, 1893 adapter->rx_queue.rx_ring[0].next2fill); 1894 VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_RXPROD2, 1895 adapter->rx_queue.rx_ring[1].next2fill); 1896 1897 /* Apply the rx filter settins last. */ 1898 vmxnet3_set_mc(adapter->netdev); 1899 1900 /* 1901 * Check link state when first activating device. It will start the 1902 * tx queue if the link is up. 1903 */ 1904 vmxnet3_check_link(adapter, true); 1905 1906 napi_enable(&adapter->napi); 1907 vmxnet3_enable_all_intrs(adapter); 1908 clear_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state); 1909 return 0; 1910 1911activate_err: 1912 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL, 0); 1913 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH, 0); 1914 vmxnet3_free_irqs(adapter); 1915irq_err: 1916rq_err: 1917 /* free up buffers we allocated */ 1918 vmxnet3_rq_cleanup(&adapter->rx_queue, adapter); 1919 return err; 1920} 1921 1922 1923void 1924vmxnet3_reset_dev(struct vmxnet3_adapter *adapter) 1925{ 1926 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, VMXNET3_CMD_RESET_DEV); 1927} 1928 1929 1930int 1931vmxnet3_quiesce_dev(struct vmxnet3_adapter *adapter) 1932{ 1933 if (test_and_set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state)) 1934 return 0; 1935 1936 1937 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, 1938 VMXNET3_CMD_QUIESCE_DEV); 1939 vmxnet3_disable_all_intrs(adapter); 1940 1941 napi_disable(&adapter->napi); 1942 netif_tx_disable(adapter->netdev); 1943 adapter->link_speed = 0; 1944 netif_carrier_off(adapter->netdev); 1945 1946 vmxnet3_tq_cleanup(&adapter->tx_queue, adapter); 1947 vmxnet3_rq_cleanup(&adapter->rx_queue, adapter); 1948 vmxnet3_free_irqs(adapter); 1949 return 0; 1950} 1951 1952 1953static void 1954vmxnet3_write_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac) 1955{ 1956 u32 tmp; 1957 1958 tmp = *(u32 *)mac; 1959 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACL, tmp); 1960 1961 tmp = (mac[5] << 8) | mac[4]; 1962 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACH, tmp); 1963} 1964 1965 1966static int 1967vmxnet3_set_mac_addr(struct net_device *netdev, void *p) 1968{ 1969 struct sockaddr *addr = p; 1970 struct vmxnet3_adapter *adapter = netdev_priv(netdev); 1971 1972 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len); 1973 vmxnet3_write_mac_addr(adapter, addr->sa_data); 1974 1975 return 0; 1976} 1977 1978 1979/* ==================== initialization and cleanup routines ============ */ 1980 1981static int 1982vmxnet3_alloc_pci_resources(struct vmxnet3_adapter *adapter, bool *dma64) 1983{ 1984 int err; 1985 unsigned long mmio_start, mmio_len; 1986 struct pci_dev *pdev = adapter->pdev; 1987 1988 err = pci_enable_device(pdev); 1989 if (err) { 1990 printk(KERN_ERR "Failed to enable adapter %s: error %d\n", 1991 pci_name(pdev), err); 1992 return err; 1993 } 1994 1995 if (pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) == 0) { 1996 if (pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64)) != 0) { 1997 printk(KERN_ERR "pci_set_consistent_dma_mask failed " 1998 "for adapter %s\n", pci_name(pdev)); 1999 err = -EIO; 2000 goto err_set_mask; 2001 } 2002 *dma64 = true; 2003 } else { 2004 if (pci_set_dma_mask(pdev, DMA_BIT_MASK(32)) != 0) { 2005 printk(KERN_ERR "pci_set_dma_mask failed for adapter " 2006 "%s\n", pci_name(pdev)); 2007 err = -EIO; 2008 goto err_set_mask; 2009 } 2010 *dma64 = false; 2011 } 2012 2013 err = pci_request_selected_regions(pdev, (1 << 2) - 1, 2014 vmxnet3_driver_name); 2015 if (err) { 2016 printk(KERN_ERR "Failed to request region for adapter %s: " 2017 "error %d\n", pci_name(pdev), err); 2018 goto err_set_mask; 2019 } 2020 2021 pci_set_master(pdev); 2022 2023 mmio_start = pci_resource_start(pdev, 0); 2024 mmio_len = pci_resource_len(pdev, 0); 2025 adapter->hw_addr0 = ioremap(mmio_start, mmio_len); 2026 if (!adapter->hw_addr0) { 2027 printk(KERN_ERR "Failed to map bar0 for adapter %s\n", 2028 pci_name(pdev)); 2029 err = -EIO; 2030 goto err_ioremap; 2031 } 2032 2033 mmio_start = pci_resource_start(pdev, 1); 2034 mmio_len = pci_resource_len(pdev, 1); 2035 adapter->hw_addr1 = ioremap(mmio_start, mmio_len); 2036 if (!adapter->hw_addr1) { 2037 printk(KERN_ERR "Failed to map bar1 for adapter %s\n", 2038 pci_name(pdev)); 2039 err = -EIO; 2040 goto err_bar1; 2041 } 2042 return 0; 2043 2044err_bar1: 2045 iounmap(adapter->hw_addr0); 2046err_ioremap: 2047 pci_release_selected_regions(pdev, (1 << 2) - 1); 2048err_set_mask: 2049 pci_disable_device(pdev); 2050 return err; 2051} 2052 2053 2054static void 2055vmxnet3_free_pci_resources(struct vmxnet3_adapter *adapter) 2056{ 2057 BUG_ON(!adapter->pdev); 2058 2059 iounmap(adapter->hw_addr0); 2060 iounmap(adapter->hw_addr1); 2061 pci_release_selected_regions(adapter->pdev, (1 << 2) - 1); 2062 pci_disable_device(adapter->pdev); 2063} 2064 2065 2066static void 2067vmxnet3_adjust_rx_ring_size(struct vmxnet3_adapter *adapter) 2068{ 2069 size_t sz; 2070 2071 if (adapter->netdev->mtu <= VMXNET3_MAX_SKB_BUF_SIZE - 2072 VMXNET3_MAX_ETH_HDR_SIZE) { 2073 adapter->skb_buf_size = adapter->netdev->mtu + 2074 VMXNET3_MAX_ETH_HDR_SIZE; 2075 if (adapter->skb_buf_size < VMXNET3_MIN_T0_BUF_SIZE) 2076 adapter->skb_buf_size = VMXNET3_MIN_T0_BUF_SIZE; 2077 2078 adapter->rx_buf_per_pkt = 1; 2079 } else { 2080 adapter->skb_buf_size = VMXNET3_MAX_SKB_BUF_SIZE; 2081 sz = adapter->netdev->mtu - VMXNET3_MAX_SKB_BUF_SIZE + 2082 VMXNET3_MAX_ETH_HDR_SIZE; 2083 adapter->rx_buf_per_pkt = 1 + (sz + PAGE_SIZE - 1) / PAGE_SIZE; 2084 } 2085 2086 /* 2087 * for simplicity, force the ring0 size to be a multiple of 2088 * rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN 2089 */ 2090 sz = adapter->rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN; 2091 adapter->rx_queue.rx_ring[0].size = (adapter->rx_queue.rx_ring[0].size + 2092 sz - 1) / sz * sz; 2093 adapter->rx_queue.rx_ring[0].size = min_t(u32, 2094 adapter->rx_queue.rx_ring[0].size, 2095 VMXNET3_RX_RING_MAX_SIZE / sz * sz); 2096} 2097 2098 2099int 2100vmxnet3_create_queues(struct vmxnet3_adapter *adapter, u32 tx_ring_size, 2101 u32 rx_ring_size, u32 rx_ring2_size) 2102{ 2103 int err; 2104 2105 adapter->tx_queue.tx_ring.size = tx_ring_size; 2106 adapter->tx_queue.data_ring.size = tx_ring_size; 2107 adapter->tx_queue.comp_ring.size = tx_ring_size; 2108 adapter->tx_queue.shared = &adapter->tqd_start->ctrl; 2109 adapter->tx_queue.stopped = true; 2110 err = vmxnet3_tq_create(&adapter->tx_queue, adapter); 2111 if (err) 2112 return err; 2113 2114 adapter->rx_queue.rx_ring[0].size = rx_ring_size; 2115 adapter->rx_queue.rx_ring[1].size = rx_ring2_size; 2116 vmxnet3_adjust_rx_ring_size(adapter); 2117 adapter->rx_queue.comp_ring.size = adapter->rx_queue.rx_ring[0].size + 2118 adapter->rx_queue.rx_ring[1].size; 2119 adapter->rx_queue.qid = 0; 2120 adapter->rx_queue.qid2 = 1; 2121 adapter->rx_queue.shared = &adapter->rqd_start->ctrl; 2122 err = vmxnet3_rq_create(&adapter->rx_queue, adapter); 2123 if (err) 2124 vmxnet3_tq_destroy(&adapter->tx_queue, adapter); 2125 2126 return err; 2127} 2128 2129static int 2130vmxnet3_open(struct net_device *netdev) 2131{ 2132 struct vmxnet3_adapter *adapter; 2133 int err; 2134 2135 adapter = netdev_priv(netdev); 2136 2137 spin_lock_init(&adapter->tx_queue.tx_lock); 2138 2139 err = vmxnet3_create_queues(adapter, VMXNET3_DEF_TX_RING_SIZE, 2140 VMXNET3_DEF_RX_RING_SIZE, 2141 VMXNET3_DEF_RX_RING_SIZE); 2142 if (err) 2143 goto queue_err; 2144 2145 err = vmxnet3_activate_dev(adapter); 2146 if (err) 2147 goto activate_err; 2148 2149 return 0; 2150 2151activate_err: 2152 vmxnet3_rq_destroy(&adapter->rx_queue, adapter); 2153 vmxnet3_tq_destroy(&adapter->tx_queue, adapter); 2154queue_err: 2155 return err; 2156} 2157 2158 2159static int 2160vmxnet3_close(struct net_device *netdev) 2161{ 2162 struct vmxnet3_adapter *adapter = netdev_priv(netdev); 2163 2164 /* 2165 * Reset_work may be in the middle of resetting the device, wait for its 2166 * completion. 2167 */ 2168 while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state)) 2169 msleep(1); 2170 2171 vmxnet3_quiesce_dev(adapter); 2172 2173 vmxnet3_rq_destroy(&adapter->rx_queue, adapter); 2174 vmxnet3_tq_destroy(&adapter->tx_queue, adapter); 2175 2176 clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state); 2177 2178 2179 return 0; 2180} 2181 2182 2183void 2184vmxnet3_force_close(struct vmxnet3_adapter *adapter) 2185{ 2186 /* 2187 * we must clear VMXNET3_STATE_BIT_RESETTING, otherwise 2188 * vmxnet3_close() will deadlock. 2189 */ 2190 BUG_ON(test_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state)); 2191 2192 /* we need to enable NAPI, otherwise dev_close will deadlock */ 2193 napi_enable(&adapter->napi); 2194 dev_close(adapter->netdev); 2195} 2196 2197 2198static int 2199vmxnet3_change_mtu(struct net_device *netdev, int new_mtu) 2200{ 2201 struct vmxnet3_adapter *adapter = netdev_priv(netdev); 2202 int err = 0; 2203 2204 if (new_mtu < VMXNET3_MIN_MTU || new_mtu > VMXNET3_MAX_MTU) 2205 return -EINVAL; 2206 2207 if (new_mtu > 1500 && !adapter->jumbo_frame) 2208 return -EINVAL; 2209 2210 netdev->mtu = new_mtu; 2211 2212 /* 2213 * Reset_work may be in the middle of resetting the device, wait for its 2214 * completion. 2215 */ 2216 while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state)) 2217 msleep(1); 2218 2219 if (netif_running(netdev)) { 2220 vmxnet3_quiesce_dev(adapter); 2221 vmxnet3_reset_dev(adapter); 2222 2223 /* we need to re-create the rx queue based on the new mtu */ 2224 vmxnet3_rq_destroy(&adapter->rx_queue, adapter); 2225 vmxnet3_adjust_rx_ring_size(adapter); 2226 adapter->rx_queue.comp_ring.size = 2227 adapter->rx_queue.rx_ring[0].size + 2228 adapter->rx_queue.rx_ring[1].size; 2229 err = vmxnet3_rq_create(&adapter->rx_queue, adapter); 2230 if (err) { 2231 printk(KERN_ERR "%s: failed to re-create rx queue," 2232 " error %d. Closing it.\n", netdev->name, err); 2233 goto out; 2234 } 2235 2236 err = vmxnet3_activate_dev(adapter); 2237 if (err) { 2238 printk(KERN_ERR "%s: failed to re-activate, error %d. " 2239 "Closing it\n", netdev->name, err); 2240 goto out; 2241 } 2242 } 2243 2244out: 2245 clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state); 2246 if (err) 2247 vmxnet3_force_close(adapter); 2248 2249 return err; 2250} 2251 2252 2253static void 2254vmxnet3_declare_features(struct vmxnet3_adapter *adapter, bool dma64) 2255{ 2256 struct net_device *netdev = adapter->netdev; 2257 2258 netdev->features = NETIF_F_SG | 2259 NETIF_F_HW_CSUM | 2260 NETIF_F_HW_VLAN_TX | 2261 NETIF_F_HW_VLAN_RX | 2262 NETIF_F_HW_VLAN_FILTER | 2263 NETIF_F_TSO | 2264 NETIF_F_TSO6 | 2265 NETIF_F_LRO; 2266 2267 printk(KERN_INFO "features: sg csum vlan jf tso tsoIPv6 lro"); 2268 2269 adapter->rxcsum = true; 2270 adapter->jumbo_frame = true; 2271 adapter->lro = true; 2272 2273 if (dma64) { 2274 netdev->features |= NETIF_F_HIGHDMA; 2275 printk(" highDMA"); 2276 } 2277 2278 netdev->vlan_features = netdev->features; 2279 printk("\n"); 2280} 2281 2282 2283static void 2284vmxnet3_read_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac) 2285{ 2286 u32 tmp; 2287 2288 tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACL); 2289 *(u32 *)mac = tmp; 2290 2291 tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACH); 2292 mac[4] = tmp & 0xff; 2293 mac[5] = (tmp >> 8) & 0xff; 2294} 2295 2296 2297static void 2298vmxnet3_alloc_intr_resources(struct vmxnet3_adapter *adapter) 2299{ 2300 u32 cfg; 2301 2302 /* intr settings */ 2303 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, 2304 VMXNET3_CMD_GET_CONF_INTR); 2305 cfg = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD); 2306 adapter->intr.type = cfg & 0x3; 2307 adapter->intr.mask_mode = (cfg >> 2) & 0x3; 2308 2309 if (adapter->intr.type == VMXNET3_IT_AUTO) { 2310 adapter->intr.type = VMXNET3_IT_MSIX; 2311 } 2312 2313#ifdef CONFIG_PCI_MSI 2314 if (adapter->intr.type == VMXNET3_IT_MSIX) { 2315 int err; 2316 2317 adapter->intr.msix_entries[0].entry = 0; 2318 err = pci_enable_msix(adapter->pdev, adapter->intr.msix_entries, 2319 VMXNET3_LINUX_MAX_MSIX_VECT); 2320 if (!err) { 2321 adapter->intr.num_intrs = 1; 2322 adapter->intr.type = VMXNET3_IT_MSIX; 2323 return; 2324 } 2325 adapter->intr.type = VMXNET3_IT_MSI; 2326 } 2327 2328 if (adapter->intr.type == VMXNET3_IT_MSI) { 2329 int err; 2330 err = pci_enable_msi(adapter->pdev); 2331 if (!err) { 2332 adapter->intr.num_intrs = 1; 2333 return; 2334 } 2335 } 2336#endif /* CONFIG_PCI_MSI */ 2337 2338 adapter->intr.type = VMXNET3_IT_INTX; 2339 2340 /* INT-X related setting */ 2341 adapter->intr.num_intrs = 1; 2342} 2343 2344 2345static void 2346vmxnet3_free_intr_resources(struct vmxnet3_adapter *adapter) 2347{ 2348 if (adapter->intr.type == VMXNET3_IT_MSIX) 2349 pci_disable_msix(adapter->pdev); 2350 else if (adapter->intr.type == VMXNET3_IT_MSI) 2351 pci_disable_msi(adapter->pdev); 2352 else 2353 BUG_ON(adapter->intr.type != VMXNET3_IT_INTX); 2354} 2355 2356 2357static void 2358vmxnet3_tx_timeout(struct net_device *netdev) 2359{ 2360 struct vmxnet3_adapter *adapter = netdev_priv(netdev); 2361 adapter->tx_timeout_count++; 2362 2363 printk(KERN_ERR "%s: tx hang\n", adapter->netdev->name); 2364 schedule_work(&adapter->work); 2365} 2366 2367 2368static void 2369vmxnet3_reset_work(struct work_struct *data) 2370{ 2371 struct vmxnet3_adapter *adapter; 2372 2373 adapter = container_of(data, struct vmxnet3_adapter, work); 2374 2375 /* if another thread is resetting the device, no need to proceed */ 2376 if (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state)) 2377 return; 2378 2379 /* if the device is closed, we must leave it alone */ 2380 rtnl_lock(); 2381 if (netif_running(adapter->netdev)) { 2382 printk(KERN_INFO "%s: resetting\n", adapter->netdev->name); 2383 vmxnet3_quiesce_dev(adapter); 2384 vmxnet3_reset_dev(adapter); 2385 vmxnet3_activate_dev(adapter); 2386 } else { 2387 printk(KERN_INFO "%s: already closed\n", adapter->netdev->name); 2388 } 2389 rtnl_unlock(); 2390 2391 clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state); 2392} 2393 2394 2395static int __devinit 2396vmxnet3_probe_device(struct pci_dev *pdev, 2397 const struct pci_device_id *id) 2398{ 2399 static const struct net_device_ops vmxnet3_netdev_ops = { 2400 .ndo_open = vmxnet3_open, 2401 .ndo_stop = vmxnet3_close, 2402 .ndo_start_xmit = vmxnet3_xmit_frame, 2403 .ndo_set_mac_address = vmxnet3_set_mac_addr, 2404 .ndo_change_mtu = vmxnet3_change_mtu, 2405 .ndo_get_stats = vmxnet3_get_stats, 2406 .ndo_tx_timeout = vmxnet3_tx_timeout, 2407 .ndo_set_multicast_list = vmxnet3_set_mc, 2408 .ndo_vlan_rx_register = vmxnet3_vlan_rx_register, 2409 .ndo_vlan_rx_add_vid = vmxnet3_vlan_rx_add_vid, 2410 .ndo_vlan_rx_kill_vid = vmxnet3_vlan_rx_kill_vid, 2411#ifdef CONFIG_NET_POLL_CONTROLLER 2412 .ndo_poll_controller = vmxnet3_netpoll, 2413#endif 2414 }; 2415 int err; 2416 bool dma64 = false; /* stupid gcc */ 2417 u32 ver; 2418 struct net_device *netdev; 2419 struct vmxnet3_adapter *adapter; 2420 u8 mac[ETH_ALEN]; 2421 2422 netdev = alloc_etherdev(sizeof(struct vmxnet3_adapter)); 2423 if (!netdev) { 2424 printk(KERN_ERR "Failed to alloc ethernet device for adapter " 2425 "%s\n", pci_name(pdev)); 2426 return -ENOMEM; 2427 } 2428 2429 pci_set_drvdata(pdev, netdev); 2430 adapter = netdev_priv(netdev); 2431 adapter->netdev = netdev; 2432 adapter->pdev = pdev; 2433 2434 adapter->shared = pci_alloc_consistent(adapter->pdev, 2435 sizeof(struct Vmxnet3_DriverShared), 2436 &adapter->shared_pa); 2437 if (!adapter->shared) { 2438 printk(KERN_ERR "Failed to allocate memory for %s\n", 2439 pci_name(pdev)); 2440 err = -ENOMEM; 2441 goto err_alloc_shared; 2442 } 2443 2444 adapter->tqd_start = pci_alloc_consistent(adapter->pdev, 2445 sizeof(struct Vmxnet3_TxQueueDesc) + 2446 sizeof(struct Vmxnet3_RxQueueDesc), 2447 &adapter->queue_desc_pa); 2448 2449 if (!adapter->tqd_start) { 2450 printk(KERN_ERR "Failed to allocate memory for %s\n", 2451 pci_name(pdev)); 2452 err = -ENOMEM; 2453 goto err_alloc_queue_desc; 2454 } 2455 adapter->rqd_start = (struct Vmxnet3_RxQueueDesc *)(adapter->tqd_start 2456 + 1); 2457 2458 adapter->pm_conf = kmalloc(sizeof(struct Vmxnet3_PMConf), GFP_KERNEL); 2459 if (adapter->pm_conf == NULL) { 2460 printk(KERN_ERR "Failed to allocate memory for %s\n", 2461 pci_name(pdev)); 2462 err = -ENOMEM; 2463 goto err_alloc_pm; 2464 } 2465 2466 err = vmxnet3_alloc_pci_resources(adapter, &dma64); 2467 if (err < 0) 2468 goto err_alloc_pci; 2469 2470 ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_VRRS); 2471 if (ver & 1) { 2472 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_VRRS, 1); 2473 } else { 2474 printk(KERN_ERR "Incompatible h/w version (0x%x) for adapter" 2475 " %s\n", ver, pci_name(pdev)); 2476 err = -EBUSY; 2477 goto err_ver; 2478 } 2479 2480 ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_UVRS); 2481 if (ver & 1) { 2482 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_UVRS, 1); 2483 } else { 2484 printk(KERN_ERR "Incompatible upt version (0x%x) for " 2485 "adapter %s\n", ver, pci_name(pdev)); 2486 err = -EBUSY; 2487 goto err_ver; 2488 } 2489 2490 vmxnet3_declare_features(adapter, dma64); 2491 2492 adapter->dev_number = atomic_read(&devices_found); 2493 vmxnet3_alloc_intr_resources(adapter); 2494 2495 vmxnet3_read_mac_addr(adapter, mac); 2496 memcpy(netdev->dev_addr, mac, netdev->addr_len); 2497 2498 netdev->netdev_ops = &vmxnet3_netdev_ops; 2499 netdev->watchdog_timeo = 5 * HZ; 2500 vmxnet3_set_ethtool_ops(netdev); 2501 2502 INIT_WORK(&adapter->work, vmxnet3_reset_work); 2503 2504 netif_napi_add(netdev, &adapter->napi, vmxnet3_poll, 64); 2505 SET_NETDEV_DEV(netdev, &pdev->dev); 2506 err = register_netdev(netdev); 2507 2508 if (err) { 2509 printk(KERN_ERR "Failed to register adapter %s\n", 2510 pci_name(pdev)); 2511 goto err_register; 2512 } 2513 2514 set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state); 2515 vmxnet3_check_link(adapter, false); 2516 atomic_inc(&devices_found); 2517 return 0; 2518 2519err_register: 2520 vmxnet3_free_intr_resources(adapter); 2521err_ver: 2522 vmxnet3_free_pci_resources(adapter); 2523err_alloc_pci: 2524 kfree(adapter->pm_conf); 2525err_alloc_pm: 2526 pci_free_consistent(adapter->pdev, sizeof(struct Vmxnet3_TxQueueDesc) + 2527 sizeof(struct Vmxnet3_RxQueueDesc), 2528 adapter->tqd_start, adapter->queue_desc_pa); 2529err_alloc_queue_desc: 2530 pci_free_consistent(adapter->pdev, sizeof(struct Vmxnet3_DriverShared), 2531 adapter->shared, adapter->shared_pa); 2532err_alloc_shared: 2533 pci_set_drvdata(pdev, NULL); 2534 free_netdev(netdev); 2535 return err; 2536} 2537 2538 2539static void __devexit 2540vmxnet3_remove_device(struct pci_dev *pdev) 2541{ 2542 struct net_device *netdev = pci_get_drvdata(pdev); 2543 struct vmxnet3_adapter *adapter = netdev_priv(netdev); 2544 2545 flush_scheduled_work(); 2546 2547 unregister_netdev(netdev); 2548 2549 vmxnet3_free_intr_resources(adapter); 2550 vmxnet3_free_pci_resources(adapter); 2551 kfree(adapter->pm_conf); 2552 pci_free_consistent(adapter->pdev, sizeof(struct Vmxnet3_TxQueueDesc) + 2553 sizeof(struct Vmxnet3_RxQueueDesc), 2554 adapter->tqd_start, adapter->queue_desc_pa); 2555 pci_free_consistent(adapter->pdev, sizeof(struct Vmxnet3_DriverShared), 2556 adapter->shared, adapter->shared_pa); 2557 free_netdev(netdev); 2558} 2559 2560 2561#ifdef CONFIG_PM 2562 2563static int 2564vmxnet3_suspend(struct device *device) 2565{ 2566 struct pci_dev *pdev = to_pci_dev(device); 2567 struct net_device *netdev = pci_get_drvdata(pdev); 2568 struct vmxnet3_adapter *adapter = netdev_priv(netdev); 2569 struct Vmxnet3_PMConf *pmConf; 2570 struct ethhdr *ehdr; 2571 struct arphdr *ahdr; 2572 u8 *arpreq; 2573 struct in_device *in_dev; 2574 struct in_ifaddr *ifa; 2575 int i = 0; 2576 2577 if (!netif_running(netdev)) 2578 return 0; 2579 2580 vmxnet3_disable_all_intrs(adapter); 2581 vmxnet3_free_irqs(adapter); 2582 vmxnet3_free_intr_resources(adapter); 2583 2584 netif_device_detach(netdev); 2585 netif_stop_queue(netdev); 2586 2587 /* Create wake-up filters. */ 2588 pmConf = adapter->pm_conf; 2589 memset(pmConf, 0, sizeof(*pmConf)); 2590 2591 if (adapter->wol & WAKE_UCAST) { 2592 pmConf->filters[i].patternSize = ETH_ALEN; 2593 pmConf->filters[i].maskSize = 1; 2594 memcpy(pmConf->filters[i].pattern, netdev->dev_addr, ETH_ALEN); 2595 pmConf->filters[i].mask[0] = 0x3F; /* LSB ETH_ALEN bits */ 2596 2597 set_flag_le16(&pmConf->wakeUpEvents, VMXNET3_PM_WAKEUP_FILTER); 2598 i++; 2599 } 2600 2601 if (adapter->wol & WAKE_ARP) { 2602 in_dev = in_dev_get(netdev); 2603 if (!in_dev) 2604 goto skip_arp; 2605 2606 ifa = (struct in_ifaddr *)in_dev->ifa_list; 2607 if (!ifa) 2608 goto skip_arp; 2609 2610 pmConf->filters[i].patternSize = ETH_HLEN + /* Ethernet header*/ 2611 sizeof(struct arphdr) + /* ARP header */ 2612 2 * ETH_ALEN + /* 2 Ethernet addresses*/ 2613 2 * sizeof(u32); /*2 IPv4 addresses */ 2614 pmConf->filters[i].maskSize = 2615 (pmConf->filters[i].patternSize - 1) / 8 + 1; 2616 2617 /* ETH_P_ARP in Ethernet header. */ 2618 ehdr = (struct ethhdr *)pmConf->filters[i].pattern; 2619 ehdr->h_proto = htons(ETH_P_ARP); 2620 2621 /* ARPOP_REQUEST in ARP header. */ 2622 ahdr = (struct arphdr *)&pmConf->filters[i].pattern[ETH_HLEN]; 2623 ahdr->ar_op = htons(ARPOP_REQUEST); 2624 arpreq = (u8 *)(ahdr + 1); 2625 2626 /* The Unicast IPv4 address in 'tip' field. */ 2627 arpreq += 2 * ETH_ALEN + sizeof(u32); 2628 *(u32 *)arpreq = ifa->ifa_address; 2629 2630 /* The mask for the relevant bits. */ 2631 pmConf->filters[i].mask[0] = 0x00; 2632 pmConf->filters[i].mask[1] = 0x30; /* ETH_P_ARP */ 2633 pmConf->filters[i].mask[2] = 0x30; /* ARPOP_REQUEST */ 2634 pmConf->filters[i].mask[3] = 0x00; 2635 pmConf->filters[i].mask[4] = 0xC0; /* IPv4 TIP */ 2636 pmConf->filters[i].mask[5] = 0x03; /* IPv4 TIP */ 2637 in_dev_put(in_dev); 2638 2639 set_flag_le16(&pmConf->wakeUpEvents, VMXNET3_PM_WAKEUP_FILTER); 2640 i++; 2641 } 2642 2643skip_arp: 2644 if (adapter->wol & WAKE_MAGIC) 2645 set_flag_le16(&pmConf->wakeUpEvents, VMXNET3_PM_WAKEUP_MAGIC); 2646 2647 pmConf->numFilters = i; 2648 2649 adapter->shared->devRead.pmConfDesc.confVer = cpu_to_le32(1); 2650 adapter->shared->devRead.pmConfDesc.confLen = cpu_to_le32(sizeof( 2651 *pmConf)); 2652 adapter->shared->devRead.pmConfDesc.confPA = cpu_to_le64(virt_to_phys( 2653 pmConf)); 2654 2655 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, 2656 VMXNET3_CMD_UPDATE_PMCFG); 2657 2658 pci_save_state(pdev); 2659 pci_enable_wake(pdev, pci_choose_state(pdev, PMSG_SUSPEND), 2660 adapter->wol); 2661 pci_disable_device(pdev); 2662 pci_set_power_state(pdev, pci_choose_state(pdev, PMSG_SUSPEND)); 2663 2664 return 0; 2665} 2666 2667 2668static int 2669vmxnet3_resume(struct device *device) 2670{ 2671 int err; 2672 struct pci_dev *pdev = to_pci_dev(device); 2673 struct net_device *netdev = pci_get_drvdata(pdev); 2674 struct vmxnet3_adapter *adapter = netdev_priv(netdev); 2675 struct Vmxnet3_PMConf *pmConf; 2676 2677 if (!netif_running(netdev)) 2678 return 0; 2679 2680 /* Destroy wake-up filters. */ 2681 pmConf = adapter->pm_conf; 2682 memset(pmConf, 0, sizeof(*pmConf)); 2683 2684 adapter->shared->devRead.pmConfDesc.confVer = cpu_to_le32(1); 2685 adapter->shared->devRead.pmConfDesc.confLen = cpu_to_le32(sizeof( 2686 *pmConf)); 2687 adapter->shared->devRead.pmConfDesc.confPA = cpu_to_le32(virt_to_phys( 2688 pmConf)); 2689 2690 netif_device_attach(netdev); 2691 pci_set_power_state(pdev, PCI_D0); 2692 pci_restore_state(pdev); 2693 err = pci_enable_device_mem(pdev); 2694 if (err != 0) 2695 return err; 2696 2697 pci_enable_wake(pdev, PCI_D0, 0); 2698 2699 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, 2700 VMXNET3_CMD_UPDATE_PMCFG); 2701 vmxnet3_alloc_intr_resources(adapter); 2702 vmxnet3_request_irqs(adapter); 2703 vmxnet3_enable_all_intrs(adapter); 2704 2705 return 0; 2706} 2707 2708static const struct dev_pm_ops vmxnet3_pm_ops = { 2709 .suspend = vmxnet3_suspend, 2710 .resume = vmxnet3_resume, 2711}; 2712#endif 2713 2714static struct pci_driver vmxnet3_driver = { 2715 .name = vmxnet3_driver_name, 2716 .id_table = vmxnet3_pciid_table, 2717 .probe = vmxnet3_probe_device, 2718 .remove = __devexit_p(vmxnet3_remove_device), 2719#ifdef CONFIG_PM 2720 .driver.pm = &vmxnet3_pm_ops, 2721#endif 2722}; 2723 2724 2725static int __init 2726vmxnet3_init_module(void) 2727{ 2728 printk(KERN_INFO "%s - version %s\n", VMXNET3_DRIVER_DESC, 2729 VMXNET3_DRIVER_VERSION_REPORT); 2730 return pci_register_driver(&vmxnet3_driver); 2731} 2732 2733module_init(vmxnet3_init_module); 2734 2735 2736static void 2737vmxnet3_exit_module(void) 2738{ 2739 pci_unregister_driver(&vmxnet3_driver); 2740} 2741 2742module_exit(vmxnet3_exit_module); 2743 2744MODULE_AUTHOR("VMware, Inc."); 2745MODULE_DESCRIPTION(VMXNET3_DRIVER_DESC); 2746MODULE_LICENSE("GPL v2"); 2747MODULE_VERSION(VMXNET3_DRIVER_VERSION_STRING); 2748