vmxnet3_drv.c revision 72e85c45b9541e1f3d233e775da1dc6f68b85867
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#define VMXNET3_MAX_DEVICES 10 48static int enable_mq = 1; 49static int irq_share_mode; 50 51static void 52vmxnet3_write_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac); 53 54/* 55 * Enable/Disable the given intr 56 */ 57static void 58vmxnet3_enable_intr(struct vmxnet3_adapter *adapter, unsigned intr_idx) 59{ 60 VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_IMR + intr_idx * 8, 0); 61} 62 63 64static void 65vmxnet3_disable_intr(struct vmxnet3_adapter *adapter, unsigned intr_idx) 66{ 67 VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_IMR + intr_idx * 8, 1); 68} 69 70 71/* 72 * Enable/Disable all intrs used by the device 73 */ 74static void 75vmxnet3_enable_all_intrs(struct vmxnet3_adapter *adapter) 76{ 77 int i; 78 79 for (i = 0; i < adapter->intr.num_intrs; i++) 80 vmxnet3_enable_intr(adapter, i); 81 adapter->shared->devRead.intrConf.intrCtrl &= 82 cpu_to_le32(~VMXNET3_IC_DISABLE_ALL); 83} 84 85 86static void 87vmxnet3_disable_all_intrs(struct vmxnet3_adapter *adapter) 88{ 89 int i; 90 91 adapter->shared->devRead.intrConf.intrCtrl |= 92 cpu_to_le32(VMXNET3_IC_DISABLE_ALL); 93 for (i = 0; i < adapter->intr.num_intrs; i++) 94 vmxnet3_disable_intr(adapter, i); 95} 96 97 98static void 99vmxnet3_ack_events(struct vmxnet3_adapter *adapter, u32 events) 100{ 101 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_ECR, events); 102} 103 104 105static bool 106vmxnet3_tq_stopped(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter) 107{ 108 return tq->stopped; 109} 110 111 112static void 113vmxnet3_tq_start(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter) 114{ 115 tq->stopped = false; 116 netif_start_subqueue(adapter->netdev, tq - adapter->tx_queue); 117} 118 119 120static void 121vmxnet3_tq_wake(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter) 122{ 123 tq->stopped = false; 124 netif_wake_subqueue(adapter->netdev, (tq - adapter->tx_queue)); 125} 126 127 128static void 129vmxnet3_tq_stop(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter) 130{ 131 tq->stopped = true; 132 tq->num_stop++; 133 netif_stop_subqueue(adapter->netdev, (tq - adapter->tx_queue)); 134} 135 136 137/* 138 * Check the link state. This may start or stop the tx queue. 139 */ 140static void 141vmxnet3_check_link(struct vmxnet3_adapter *adapter, bool affectTxQueue) 142{ 143 u32 ret; 144 int i; 145 unsigned long flags; 146 147 spin_lock_irqsave(&adapter->cmd_lock, flags); 148 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, VMXNET3_CMD_GET_LINK); 149 ret = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD); 150 spin_unlock_irqrestore(&adapter->cmd_lock, flags); 151 152 adapter->link_speed = ret >> 16; 153 if (ret & 1) { /* Link is up. */ 154 printk(KERN_INFO "%s: NIC Link is Up %d Mbps\n", 155 adapter->netdev->name, adapter->link_speed); 156 if (!netif_carrier_ok(adapter->netdev)) 157 netif_carrier_on(adapter->netdev); 158 159 if (affectTxQueue) { 160 for (i = 0; i < adapter->num_tx_queues; i++) 161 vmxnet3_tq_start(&adapter->tx_queue[i], 162 adapter); 163 } 164 } else { 165 printk(KERN_INFO "%s: NIC Link is Down\n", 166 adapter->netdev->name); 167 if (netif_carrier_ok(adapter->netdev)) 168 netif_carrier_off(adapter->netdev); 169 170 if (affectTxQueue) { 171 for (i = 0; i < adapter->num_tx_queues; i++) 172 vmxnet3_tq_stop(&adapter->tx_queue[i], adapter); 173 } 174 } 175} 176 177static void 178vmxnet3_process_events(struct vmxnet3_adapter *adapter) 179{ 180 int i; 181 unsigned long flags; 182 u32 events = le32_to_cpu(adapter->shared->ecr); 183 if (!events) 184 return; 185 186 vmxnet3_ack_events(adapter, events); 187 188 /* Check if link state has changed */ 189 if (events & VMXNET3_ECR_LINK) 190 vmxnet3_check_link(adapter, true); 191 192 /* Check if there is an error on xmit/recv queues */ 193 if (events & (VMXNET3_ECR_TQERR | VMXNET3_ECR_RQERR)) { 194 spin_lock_irqsave(&adapter->cmd_lock, flags); 195 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, 196 VMXNET3_CMD_GET_QUEUE_STATUS); 197 spin_unlock_irqrestore(&adapter->cmd_lock, flags); 198 199 for (i = 0; i < adapter->num_tx_queues; i++) 200 if (adapter->tqd_start[i].status.stopped) 201 dev_err(&adapter->netdev->dev, 202 "%s: tq[%d] error 0x%x\n", 203 adapter->netdev->name, i, le32_to_cpu( 204 adapter->tqd_start[i].status.error)); 205 for (i = 0; i < adapter->num_rx_queues; i++) 206 if (adapter->rqd_start[i].status.stopped) 207 dev_err(&adapter->netdev->dev, 208 "%s: rq[%d] error 0x%x\n", 209 adapter->netdev->name, i, 210 adapter->rqd_start[i].status.error); 211 212 schedule_work(&adapter->work); 213 } 214} 215 216#ifdef __BIG_ENDIAN_BITFIELD 217/* 218 * The device expects the bitfields in shared structures to be written in 219 * little endian. When CPU is big endian, the following routines are used to 220 * correctly read and write into ABI. 221 * The general technique used here is : double word bitfields are defined in 222 * opposite order for big endian architecture. Then before reading them in 223 * driver the complete double word is translated using le32_to_cpu. Similarly 224 * After the driver writes into bitfields, cpu_to_le32 is used to translate the 225 * double words into required format. 226 * In order to avoid touching bits in shared structure more than once, temporary 227 * descriptors are used. These are passed as srcDesc to following functions. 228 */ 229static void vmxnet3_RxDescToCPU(const struct Vmxnet3_RxDesc *srcDesc, 230 struct Vmxnet3_RxDesc *dstDesc) 231{ 232 u32 *src = (u32 *)srcDesc + 2; 233 u32 *dst = (u32 *)dstDesc + 2; 234 dstDesc->addr = le64_to_cpu(srcDesc->addr); 235 *dst = le32_to_cpu(*src); 236 dstDesc->ext1 = le32_to_cpu(srcDesc->ext1); 237} 238 239static void vmxnet3_TxDescToLe(const struct Vmxnet3_TxDesc *srcDesc, 240 struct Vmxnet3_TxDesc *dstDesc) 241{ 242 int i; 243 u32 *src = (u32 *)(srcDesc + 1); 244 u32 *dst = (u32 *)(dstDesc + 1); 245 246 /* Working backwards so that the gen bit is set at the end. */ 247 for (i = 2; i > 0; i--) { 248 src--; 249 dst--; 250 *dst = cpu_to_le32(*src); 251 } 252} 253 254 255static void vmxnet3_RxCompToCPU(const struct Vmxnet3_RxCompDesc *srcDesc, 256 struct Vmxnet3_RxCompDesc *dstDesc) 257{ 258 int i = 0; 259 u32 *src = (u32 *)srcDesc; 260 u32 *dst = (u32 *)dstDesc; 261 for (i = 0; i < sizeof(struct Vmxnet3_RxCompDesc) / sizeof(u32); i++) { 262 *dst = le32_to_cpu(*src); 263 src++; 264 dst++; 265 } 266} 267 268 269/* Used to read bitfield values from double words. */ 270static u32 get_bitfield32(const __le32 *bitfield, u32 pos, u32 size) 271{ 272 u32 temp = le32_to_cpu(*bitfield); 273 u32 mask = ((1 << size) - 1) << pos; 274 temp &= mask; 275 temp >>= pos; 276 return temp; 277} 278 279 280 281#endif /* __BIG_ENDIAN_BITFIELD */ 282 283#ifdef __BIG_ENDIAN_BITFIELD 284 285# define VMXNET3_TXDESC_GET_GEN(txdesc) get_bitfield32(((const __le32 *) \ 286 txdesc) + VMXNET3_TXD_GEN_DWORD_SHIFT, \ 287 VMXNET3_TXD_GEN_SHIFT, VMXNET3_TXD_GEN_SIZE) 288# define VMXNET3_TXDESC_GET_EOP(txdesc) get_bitfield32(((const __le32 *) \ 289 txdesc) + VMXNET3_TXD_EOP_DWORD_SHIFT, \ 290 VMXNET3_TXD_EOP_SHIFT, VMXNET3_TXD_EOP_SIZE) 291# define VMXNET3_TCD_GET_GEN(tcd) get_bitfield32(((const __le32 *)tcd) + \ 292 VMXNET3_TCD_GEN_DWORD_SHIFT, VMXNET3_TCD_GEN_SHIFT, \ 293 VMXNET3_TCD_GEN_SIZE) 294# define VMXNET3_TCD_GET_TXIDX(tcd) get_bitfield32((const __le32 *)tcd, \ 295 VMXNET3_TCD_TXIDX_SHIFT, VMXNET3_TCD_TXIDX_SIZE) 296# define vmxnet3_getRxComp(dstrcd, rcd, tmp) do { \ 297 (dstrcd) = (tmp); \ 298 vmxnet3_RxCompToCPU((rcd), (tmp)); \ 299 } while (0) 300# define vmxnet3_getRxDesc(dstrxd, rxd, tmp) do { \ 301 (dstrxd) = (tmp); \ 302 vmxnet3_RxDescToCPU((rxd), (tmp)); \ 303 } while (0) 304 305#else 306 307# define VMXNET3_TXDESC_GET_GEN(txdesc) ((txdesc)->gen) 308# define VMXNET3_TXDESC_GET_EOP(txdesc) ((txdesc)->eop) 309# define VMXNET3_TCD_GET_GEN(tcd) ((tcd)->gen) 310# define VMXNET3_TCD_GET_TXIDX(tcd) ((tcd)->txdIdx) 311# define vmxnet3_getRxComp(dstrcd, rcd, tmp) (dstrcd) = (rcd) 312# define vmxnet3_getRxDesc(dstrxd, rxd, tmp) (dstrxd) = (rxd) 313 314#endif /* __BIG_ENDIAN_BITFIELD */ 315 316 317static void 318vmxnet3_unmap_tx_buf(struct vmxnet3_tx_buf_info *tbi, 319 struct pci_dev *pdev) 320{ 321 if (tbi->map_type == VMXNET3_MAP_SINGLE) 322 pci_unmap_single(pdev, tbi->dma_addr, tbi->len, 323 PCI_DMA_TODEVICE); 324 else if (tbi->map_type == VMXNET3_MAP_PAGE) 325 pci_unmap_page(pdev, tbi->dma_addr, tbi->len, 326 PCI_DMA_TODEVICE); 327 else 328 BUG_ON(tbi->map_type != VMXNET3_MAP_NONE); 329 330 tbi->map_type = VMXNET3_MAP_NONE; /* to help debugging */ 331} 332 333 334static int 335vmxnet3_unmap_pkt(u32 eop_idx, struct vmxnet3_tx_queue *tq, 336 struct pci_dev *pdev, struct vmxnet3_adapter *adapter) 337{ 338 struct sk_buff *skb; 339 int entries = 0; 340 341 /* no out of order completion */ 342 BUG_ON(tq->buf_info[eop_idx].sop_idx != tq->tx_ring.next2comp); 343 BUG_ON(VMXNET3_TXDESC_GET_EOP(&(tq->tx_ring.base[eop_idx].txd)) != 1); 344 345 skb = tq->buf_info[eop_idx].skb; 346 BUG_ON(skb == NULL); 347 tq->buf_info[eop_idx].skb = NULL; 348 349 VMXNET3_INC_RING_IDX_ONLY(eop_idx, tq->tx_ring.size); 350 351 while (tq->tx_ring.next2comp != eop_idx) { 352 vmxnet3_unmap_tx_buf(tq->buf_info + tq->tx_ring.next2comp, 353 pdev); 354 355 /* update next2comp w/o tx_lock. Since we are marking more, 356 * instead of less, tx ring entries avail, the worst case is 357 * that the tx routine incorrectly re-queues a pkt due to 358 * insufficient tx ring entries. 359 */ 360 vmxnet3_cmd_ring_adv_next2comp(&tq->tx_ring); 361 entries++; 362 } 363 364 dev_kfree_skb_any(skb); 365 return entries; 366} 367 368 369static int 370vmxnet3_tq_tx_complete(struct vmxnet3_tx_queue *tq, 371 struct vmxnet3_adapter *adapter) 372{ 373 int completed = 0; 374 union Vmxnet3_GenericDesc *gdesc; 375 376 gdesc = tq->comp_ring.base + tq->comp_ring.next2proc; 377 while (VMXNET3_TCD_GET_GEN(&gdesc->tcd) == tq->comp_ring.gen) { 378 completed += vmxnet3_unmap_pkt(VMXNET3_TCD_GET_TXIDX( 379 &gdesc->tcd), tq, adapter->pdev, 380 adapter); 381 382 vmxnet3_comp_ring_adv_next2proc(&tq->comp_ring); 383 gdesc = tq->comp_ring.base + tq->comp_ring.next2proc; 384 } 385 386 if (completed) { 387 spin_lock(&tq->tx_lock); 388 if (unlikely(vmxnet3_tq_stopped(tq, adapter) && 389 vmxnet3_cmd_ring_desc_avail(&tq->tx_ring) > 390 VMXNET3_WAKE_QUEUE_THRESHOLD(tq) && 391 netif_carrier_ok(adapter->netdev))) { 392 vmxnet3_tq_wake(tq, adapter); 393 } 394 spin_unlock(&tq->tx_lock); 395 } 396 return completed; 397} 398 399 400static void 401vmxnet3_tq_cleanup(struct vmxnet3_tx_queue *tq, 402 struct vmxnet3_adapter *adapter) 403{ 404 int i; 405 406 while (tq->tx_ring.next2comp != tq->tx_ring.next2fill) { 407 struct vmxnet3_tx_buf_info *tbi; 408 409 tbi = tq->buf_info + tq->tx_ring.next2comp; 410 411 vmxnet3_unmap_tx_buf(tbi, adapter->pdev); 412 if (tbi->skb) { 413 dev_kfree_skb_any(tbi->skb); 414 tbi->skb = NULL; 415 } 416 vmxnet3_cmd_ring_adv_next2comp(&tq->tx_ring); 417 } 418 419 /* sanity check, verify all buffers are indeed unmapped and freed */ 420 for (i = 0; i < tq->tx_ring.size; i++) { 421 BUG_ON(tq->buf_info[i].skb != NULL || 422 tq->buf_info[i].map_type != VMXNET3_MAP_NONE); 423 } 424 425 tq->tx_ring.gen = VMXNET3_INIT_GEN; 426 tq->tx_ring.next2fill = tq->tx_ring.next2comp = 0; 427 428 tq->comp_ring.gen = VMXNET3_INIT_GEN; 429 tq->comp_ring.next2proc = 0; 430} 431 432 433static void 434vmxnet3_tq_destroy(struct vmxnet3_tx_queue *tq, 435 struct vmxnet3_adapter *adapter) 436{ 437 if (tq->tx_ring.base) { 438 pci_free_consistent(adapter->pdev, tq->tx_ring.size * 439 sizeof(struct Vmxnet3_TxDesc), 440 tq->tx_ring.base, tq->tx_ring.basePA); 441 tq->tx_ring.base = NULL; 442 } 443 if (tq->data_ring.base) { 444 pci_free_consistent(adapter->pdev, tq->data_ring.size * 445 sizeof(struct Vmxnet3_TxDataDesc), 446 tq->data_ring.base, tq->data_ring.basePA); 447 tq->data_ring.base = NULL; 448 } 449 if (tq->comp_ring.base) { 450 pci_free_consistent(adapter->pdev, tq->comp_ring.size * 451 sizeof(struct Vmxnet3_TxCompDesc), 452 tq->comp_ring.base, tq->comp_ring.basePA); 453 tq->comp_ring.base = NULL; 454 } 455 kfree(tq->buf_info); 456 tq->buf_info = NULL; 457} 458 459 460/* Destroy all tx queues */ 461void 462vmxnet3_tq_destroy_all(struct vmxnet3_adapter *adapter) 463{ 464 int i; 465 466 for (i = 0; i < adapter->num_tx_queues; i++) 467 vmxnet3_tq_destroy(&adapter->tx_queue[i], adapter); 468} 469 470 471static void 472vmxnet3_tq_init(struct vmxnet3_tx_queue *tq, 473 struct vmxnet3_adapter *adapter) 474{ 475 int i; 476 477 /* reset the tx ring contents to 0 and reset the tx ring states */ 478 memset(tq->tx_ring.base, 0, tq->tx_ring.size * 479 sizeof(struct Vmxnet3_TxDesc)); 480 tq->tx_ring.next2fill = tq->tx_ring.next2comp = 0; 481 tq->tx_ring.gen = VMXNET3_INIT_GEN; 482 483 memset(tq->data_ring.base, 0, tq->data_ring.size * 484 sizeof(struct Vmxnet3_TxDataDesc)); 485 486 /* reset the tx comp ring contents to 0 and reset comp ring states */ 487 memset(tq->comp_ring.base, 0, tq->comp_ring.size * 488 sizeof(struct Vmxnet3_TxCompDesc)); 489 tq->comp_ring.next2proc = 0; 490 tq->comp_ring.gen = VMXNET3_INIT_GEN; 491 492 /* reset the bookkeeping data */ 493 memset(tq->buf_info, 0, sizeof(tq->buf_info[0]) * tq->tx_ring.size); 494 for (i = 0; i < tq->tx_ring.size; i++) 495 tq->buf_info[i].map_type = VMXNET3_MAP_NONE; 496 497 /* stats are not reset */ 498} 499 500 501static int 502vmxnet3_tq_create(struct vmxnet3_tx_queue *tq, 503 struct vmxnet3_adapter *adapter) 504{ 505 BUG_ON(tq->tx_ring.base || tq->data_ring.base || 506 tq->comp_ring.base || tq->buf_info); 507 508 tq->tx_ring.base = pci_alloc_consistent(adapter->pdev, tq->tx_ring.size 509 * sizeof(struct Vmxnet3_TxDesc), 510 &tq->tx_ring.basePA); 511 if (!tq->tx_ring.base) { 512 printk(KERN_ERR "%s: failed to allocate tx ring\n", 513 adapter->netdev->name); 514 goto err; 515 } 516 517 tq->data_ring.base = pci_alloc_consistent(adapter->pdev, 518 tq->data_ring.size * 519 sizeof(struct Vmxnet3_TxDataDesc), 520 &tq->data_ring.basePA); 521 if (!tq->data_ring.base) { 522 printk(KERN_ERR "%s: failed to allocate data ring\n", 523 adapter->netdev->name); 524 goto err; 525 } 526 527 tq->comp_ring.base = pci_alloc_consistent(adapter->pdev, 528 tq->comp_ring.size * 529 sizeof(struct Vmxnet3_TxCompDesc), 530 &tq->comp_ring.basePA); 531 if (!tq->comp_ring.base) { 532 printk(KERN_ERR "%s: failed to allocate tx comp ring\n", 533 adapter->netdev->name); 534 goto err; 535 } 536 537 tq->buf_info = kcalloc(tq->tx_ring.size, sizeof(tq->buf_info[0]), 538 GFP_KERNEL); 539 if (!tq->buf_info) { 540 printk(KERN_ERR "%s: failed to allocate tx bufinfo\n", 541 adapter->netdev->name); 542 goto err; 543 } 544 545 return 0; 546 547err: 548 vmxnet3_tq_destroy(tq, adapter); 549 return -ENOMEM; 550} 551 552static void 553vmxnet3_tq_cleanup_all(struct vmxnet3_adapter *adapter) 554{ 555 int i; 556 557 for (i = 0; i < adapter->num_tx_queues; i++) 558 vmxnet3_tq_cleanup(&adapter->tx_queue[i], adapter); 559} 560 561/* 562 * starting from ring->next2fill, allocate rx buffers for the given ring 563 * of the rx queue and update the rx desc. stop after @num_to_alloc buffers 564 * are allocated or allocation fails 565 */ 566 567static int 568vmxnet3_rq_alloc_rx_buf(struct vmxnet3_rx_queue *rq, u32 ring_idx, 569 int num_to_alloc, struct vmxnet3_adapter *adapter) 570{ 571 int num_allocated = 0; 572 struct vmxnet3_rx_buf_info *rbi_base = rq->buf_info[ring_idx]; 573 struct vmxnet3_cmd_ring *ring = &rq->rx_ring[ring_idx]; 574 u32 val; 575 576 while (num_allocated < num_to_alloc) { 577 struct vmxnet3_rx_buf_info *rbi; 578 union Vmxnet3_GenericDesc *gd; 579 580 rbi = rbi_base + ring->next2fill; 581 gd = ring->base + ring->next2fill; 582 583 if (rbi->buf_type == VMXNET3_RX_BUF_SKB) { 584 if (rbi->skb == NULL) { 585 rbi->skb = dev_alloc_skb(rbi->len + 586 NET_IP_ALIGN); 587 if (unlikely(rbi->skb == NULL)) { 588 rq->stats.rx_buf_alloc_failure++; 589 break; 590 } 591 rbi->skb->dev = adapter->netdev; 592 593 skb_reserve(rbi->skb, NET_IP_ALIGN); 594 rbi->dma_addr = pci_map_single(adapter->pdev, 595 rbi->skb->data, rbi->len, 596 PCI_DMA_FROMDEVICE); 597 } else { 598 /* rx buffer skipped by the device */ 599 } 600 val = VMXNET3_RXD_BTYPE_HEAD << VMXNET3_RXD_BTYPE_SHIFT; 601 } else { 602 BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_PAGE || 603 rbi->len != PAGE_SIZE); 604 605 if (rbi->page == NULL) { 606 rbi->page = alloc_page(GFP_ATOMIC); 607 if (unlikely(rbi->page == NULL)) { 608 rq->stats.rx_buf_alloc_failure++; 609 break; 610 } 611 rbi->dma_addr = pci_map_page(adapter->pdev, 612 rbi->page, 0, PAGE_SIZE, 613 PCI_DMA_FROMDEVICE); 614 } else { 615 /* rx buffers skipped by the device */ 616 } 617 val = VMXNET3_RXD_BTYPE_BODY << VMXNET3_RXD_BTYPE_SHIFT; 618 } 619 620 BUG_ON(rbi->dma_addr == 0); 621 gd->rxd.addr = cpu_to_le64(rbi->dma_addr); 622 gd->dword[2] = cpu_to_le32((ring->gen << VMXNET3_RXD_GEN_SHIFT) 623 | val | rbi->len); 624 625 num_allocated++; 626 vmxnet3_cmd_ring_adv_next2fill(ring); 627 } 628 rq->uncommitted[ring_idx] += num_allocated; 629 630 dev_dbg(&adapter->netdev->dev, 631 "alloc_rx_buf: %d allocated, next2fill %u, next2comp " 632 "%u, uncommited %u\n", num_allocated, ring->next2fill, 633 ring->next2comp, rq->uncommitted[ring_idx]); 634 635 /* so that the device can distinguish a full ring and an empty ring */ 636 BUG_ON(num_allocated != 0 && ring->next2fill == ring->next2comp); 637 638 return num_allocated; 639} 640 641 642static void 643vmxnet3_append_frag(struct sk_buff *skb, struct Vmxnet3_RxCompDesc *rcd, 644 struct vmxnet3_rx_buf_info *rbi) 645{ 646 struct skb_frag_struct *frag = skb_shinfo(skb)->frags + 647 skb_shinfo(skb)->nr_frags; 648 649 BUG_ON(skb_shinfo(skb)->nr_frags >= MAX_SKB_FRAGS); 650 651 frag->page = rbi->page; 652 frag->page_offset = 0; 653 frag->size = rcd->len; 654 skb->data_len += frag->size; 655 skb_shinfo(skb)->nr_frags++; 656} 657 658 659static void 660vmxnet3_map_pkt(struct sk_buff *skb, struct vmxnet3_tx_ctx *ctx, 661 struct vmxnet3_tx_queue *tq, struct pci_dev *pdev, 662 struct vmxnet3_adapter *adapter) 663{ 664 u32 dw2, len; 665 unsigned long buf_offset; 666 int i; 667 union Vmxnet3_GenericDesc *gdesc; 668 struct vmxnet3_tx_buf_info *tbi = NULL; 669 670 BUG_ON(ctx->copy_size > skb_headlen(skb)); 671 672 /* use the previous gen bit for the SOP desc */ 673 dw2 = (tq->tx_ring.gen ^ 0x1) << VMXNET3_TXD_GEN_SHIFT; 674 675 ctx->sop_txd = tq->tx_ring.base + tq->tx_ring.next2fill; 676 gdesc = ctx->sop_txd; /* both loops below can be skipped */ 677 678 /* no need to map the buffer if headers are copied */ 679 if (ctx->copy_size) { 680 ctx->sop_txd->txd.addr = cpu_to_le64(tq->data_ring.basePA + 681 tq->tx_ring.next2fill * 682 sizeof(struct Vmxnet3_TxDataDesc)); 683 ctx->sop_txd->dword[2] = cpu_to_le32(dw2 | ctx->copy_size); 684 ctx->sop_txd->dword[3] = 0; 685 686 tbi = tq->buf_info + tq->tx_ring.next2fill; 687 tbi->map_type = VMXNET3_MAP_NONE; 688 689 dev_dbg(&adapter->netdev->dev, 690 "txd[%u]: 0x%Lx 0x%x 0x%x\n", 691 tq->tx_ring.next2fill, 692 le64_to_cpu(ctx->sop_txd->txd.addr), 693 ctx->sop_txd->dword[2], ctx->sop_txd->dword[3]); 694 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring); 695 696 /* use the right gen for non-SOP desc */ 697 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT; 698 } 699 700 /* linear part can use multiple tx desc if it's big */ 701 len = skb_headlen(skb) - ctx->copy_size; 702 buf_offset = ctx->copy_size; 703 while (len) { 704 u32 buf_size; 705 706 if (len < VMXNET3_MAX_TX_BUF_SIZE) { 707 buf_size = len; 708 dw2 |= len; 709 } else { 710 buf_size = VMXNET3_MAX_TX_BUF_SIZE; 711 /* spec says that for TxDesc.len, 0 == 2^14 */ 712 } 713 714 tbi = tq->buf_info + tq->tx_ring.next2fill; 715 tbi->map_type = VMXNET3_MAP_SINGLE; 716 tbi->dma_addr = pci_map_single(adapter->pdev, 717 skb->data + buf_offset, buf_size, 718 PCI_DMA_TODEVICE); 719 720 tbi->len = buf_size; 721 722 gdesc = tq->tx_ring.base + tq->tx_ring.next2fill; 723 BUG_ON(gdesc->txd.gen == tq->tx_ring.gen); 724 725 gdesc->txd.addr = cpu_to_le64(tbi->dma_addr); 726 gdesc->dword[2] = cpu_to_le32(dw2); 727 gdesc->dword[3] = 0; 728 729 dev_dbg(&adapter->netdev->dev, 730 "txd[%u]: 0x%Lx 0x%x 0x%x\n", 731 tq->tx_ring.next2fill, le64_to_cpu(gdesc->txd.addr), 732 le32_to_cpu(gdesc->dword[2]), gdesc->dword[3]); 733 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring); 734 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT; 735 736 len -= buf_size; 737 buf_offset += buf_size; 738 } 739 740 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 741 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i]; 742 743 tbi = tq->buf_info + tq->tx_ring.next2fill; 744 tbi->map_type = VMXNET3_MAP_PAGE; 745 tbi->dma_addr = pci_map_page(adapter->pdev, frag->page, 746 frag->page_offset, frag->size, 747 PCI_DMA_TODEVICE); 748 749 tbi->len = frag->size; 750 751 gdesc = tq->tx_ring.base + tq->tx_ring.next2fill; 752 BUG_ON(gdesc->txd.gen == tq->tx_ring.gen); 753 754 gdesc->txd.addr = cpu_to_le64(tbi->dma_addr); 755 gdesc->dword[2] = cpu_to_le32(dw2 | frag->size); 756 gdesc->dword[3] = 0; 757 758 dev_dbg(&adapter->netdev->dev, 759 "txd[%u]: 0x%llu %u %u\n", 760 tq->tx_ring.next2fill, le64_to_cpu(gdesc->txd.addr), 761 le32_to_cpu(gdesc->dword[2]), gdesc->dword[3]); 762 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring); 763 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT; 764 } 765 766 ctx->eop_txd = gdesc; 767 768 /* set the last buf_info for the pkt */ 769 tbi->skb = skb; 770 tbi->sop_idx = ctx->sop_txd - tq->tx_ring.base; 771} 772 773 774/* Init all tx queues */ 775static void 776vmxnet3_tq_init_all(struct vmxnet3_adapter *adapter) 777{ 778 int i; 779 780 for (i = 0; i < adapter->num_tx_queues; i++) 781 vmxnet3_tq_init(&adapter->tx_queue[i], adapter); 782} 783 784 785/* 786 * parse and copy relevant protocol headers: 787 * For a tso pkt, relevant headers are L2/3/4 including options 788 * For a pkt requesting csum offloading, they are L2/3 and may include L4 789 * if it's a TCP/UDP pkt 790 * 791 * Returns: 792 * -1: error happens during parsing 793 * 0: protocol headers parsed, but too big to be copied 794 * 1: protocol headers parsed and copied 795 * 796 * Other effects: 797 * 1. related *ctx fields are updated. 798 * 2. ctx->copy_size is # of bytes copied 799 * 3. the portion copied is guaranteed to be in the linear part 800 * 801 */ 802static int 803vmxnet3_parse_and_copy_hdr(struct sk_buff *skb, struct vmxnet3_tx_queue *tq, 804 struct vmxnet3_tx_ctx *ctx, 805 struct vmxnet3_adapter *adapter) 806{ 807 struct Vmxnet3_TxDataDesc *tdd; 808 809 if (ctx->mss) { /* TSO */ 810 ctx->eth_ip_hdr_size = skb_transport_offset(skb); 811 ctx->l4_hdr_size = ((struct tcphdr *) 812 skb_transport_header(skb))->doff * 4; 813 ctx->copy_size = ctx->eth_ip_hdr_size + ctx->l4_hdr_size; 814 } else { 815 if (skb->ip_summed == CHECKSUM_PARTIAL) { 816 ctx->eth_ip_hdr_size = skb_checksum_start_offset(skb); 817 818 if (ctx->ipv4) { 819 struct iphdr *iph = (struct iphdr *) 820 skb_network_header(skb); 821 if (iph->protocol == IPPROTO_TCP) 822 ctx->l4_hdr_size = ((struct tcphdr *) 823 skb_transport_header(skb))->doff * 4; 824 else if (iph->protocol == IPPROTO_UDP) 825 /* 826 * Use tcp header size so that bytes to 827 * be copied are more than required by 828 * the device. 829 */ 830 ctx->l4_hdr_size = 831 sizeof(struct tcphdr); 832 else 833 ctx->l4_hdr_size = 0; 834 } else { 835 /* for simplicity, don't copy L4 headers */ 836 ctx->l4_hdr_size = 0; 837 } 838 ctx->copy_size = ctx->eth_ip_hdr_size + 839 ctx->l4_hdr_size; 840 } else { 841 ctx->eth_ip_hdr_size = 0; 842 ctx->l4_hdr_size = 0; 843 /* copy as much as allowed */ 844 ctx->copy_size = min((unsigned int)VMXNET3_HDR_COPY_SIZE 845 , skb_headlen(skb)); 846 } 847 848 /* make sure headers are accessible directly */ 849 if (unlikely(!pskb_may_pull(skb, ctx->copy_size))) 850 goto err; 851 } 852 853 if (unlikely(ctx->copy_size > VMXNET3_HDR_COPY_SIZE)) { 854 tq->stats.oversized_hdr++; 855 ctx->copy_size = 0; 856 return 0; 857 } 858 859 tdd = tq->data_ring.base + tq->tx_ring.next2fill; 860 861 memcpy(tdd->data, skb->data, ctx->copy_size); 862 dev_dbg(&adapter->netdev->dev, 863 "copy %u bytes to dataRing[%u]\n", 864 ctx->copy_size, tq->tx_ring.next2fill); 865 return 1; 866 867err: 868 return -1; 869} 870 871 872static void 873vmxnet3_prepare_tso(struct sk_buff *skb, 874 struct vmxnet3_tx_ctx *ctx) 875{ 876 struct tcphdr *tcph = (struct tcphdr *)skb_transport_header(skb); 877 if (ctx->ipv4) { 878 struct iphdr *iph = (struct iphdr *)skb_network_header(skb); 879 iph->check = 0; 880 tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, 0, 881 IPPROTO_TCP, 0); 882 } else { 883 struct ipv6hdr *iph = (struct ipv6hdr *)skb_network_header(skb); 884 tcph->check = ~csum_ipv6_magic(&iph->saddr, &iph->daddr, 0, 885 IPPROTO_TCP, 0); 886 } 887} 888 889 890/* 891 * Transmits a pkt thru a given tq 892 * Returns: 893 * NETDEV_TX_OK: descriptors are setup successfully 894 * NETDEV_TX_OK: error occurred, the pkt is dropped 895 * NETDEV_TX_BUSY: tx ring is full, queue is stopped 896 * 897 * Side-effects: 898 * 1. tx ring may be changed 899 * 2. tq stats may be updated accordingly 900 * 3. shared->txNumDeferred may be updated 901 */ 902 903static int 904vmxnet3_tq_xmit(struct sk_buff *skb, struct vmxnet3_tx_queue *tq, 905 struct vmxnet3_adapter *adapter, struct net_device *netdev) 906{ 907 int ret; 908 u32 count; 909 unsigned long flags; 910 struct vmxnet3_tx_ctx ctx; 911 union Vmxnet3_GenericDesc *gdesc; 912#ifdef __BIG_ENDIAN_BITFIELD 913 /* Use temporary descriptor to avoid touching bits multiple times */ 914 union Vmxnet3_GenericDesc tempTxDesc; 915#endif 916 917 /* conservatively estimate # of descriptors to use */ 918 count = VMXNET3_TXD_NEEDED(skb_headlen(skb)) + 919 skb_shinfo(skb)->nr_frags + 1; 920 921 ctx.ipv4 = (vlan_get_protocol(skb) == cpu_to_be16(ETH_P_IP)); 922 923 ctx.mss = skb_shinfo(skb)->gso_size; 924 if (ctx.mss) { 925 if (skb_header_cloned(skb)) { 926 if (unlikely(pskb_expand_head(skb, 0, 0, 927 GFP_ATOMIC) != 0)) { 928 tq->stats.drop_tso++; 929 goto drop_pkt; 930 } 931 tq->stats.copy_skb_header++; 932 } 933 vmxnet3_prepare_tso(skb, &ctx); 934 } else { 935 if (unlikely(count > VMXNET3_MAX_TXD_PER_PKT)) { 936 937 /* non-tso pkts must not use more than 938 * VMXNET3_MAX_TXD_PER_PKT entries 939 */ 940 if (skb_linearize(skb) != 0) { 941 tq->stats.drop_too_many_frags++; 942 goto drop_pkt; 943 } 944 tq->stats.linearized++; 945 946 /* recalculate the # of descriptors to use */ 947 count = VMXNET3_TXD_NEEDED(skb_headlen(skb)) + 1; 948 } 949 } 950 951 spin_lock_irqsave(&tq->tx_lock, flags); 952 953 if (count > vmxnet3_cmd_ring_desc_avail(&tq->tx_ring)) { 954 tq->stats.tx_ring_full++; 955 dev_dbg(&adapter->netdev->dev, 956 "tx queue stopped on %s, next2comp %u" 957 " next2fill %u\n", adapter->netdev->name, 958 tq->tx_ring.next2comp, tq->tx_ring.next2fill); 959 960 vmxnet3_tq_stop(tq, adapter); 961 spin_unlock_irqrestore(&tq->tx_lock, flags); 962 return NETDEV_TX_BUSY; 963 } 964 965 966 ret = vmxnet3_parse_and_copy_hdr(skb, tq, &ctx, adapter); 967 if (ret >= 0) { 968 BUG_ON(ret <= 0 && ctx.copy_size != 0); 969 /* hdrs parsed, check against other limits */ 970 if (ctx.mss) { 971 if (unlikely(ctx.eth_ip_hdr_size + ctx.l4_hdr_size > 972 VMXNET3_MAX_TX_BUF_SIZE)) { 973 goto hdr_too_big; 974 } 975 } else { 976 if (skb->ip_summed == CHECKSUM_PARTIAL) { 977 if (unlikely(ctx.eth_ip_hdr_size + 978 skb->csum_offset > 979 VMXNET3_MAX_CSUM_OFFSET)) { 980 goto hdr_too_big; 981 } 982 } 983 } 984 } else { 985 tq->stats.drop_hdr_inspect_err++; 986 goto unlock_drop_pkt; 987 } 988 989 /* fill tx descs related to addr & len */ 990 vmxnet3_map_pkt(skb, &ctx, tq, adapter->pdev, adapter); 991 992 /* setup the EOP desc */ 993 ctx.eop_txd->dword[3] = cpu_to_le32(VMXNET3_TXD_CQ | VMXNET3_TXD_EOP); 994 995 /* setup the SOP desc */ 996#ifdef __BIG_ENDIAN_BITFIELD 997 gdesc = &tempTxDesc; 998 gdesc->dword[2] = ctx.sop_txd->dword[2]; 999 gdesc->dword[3] = ctx.sop_txd->dword[3]; 1000#else 1001 gdesc = ctx.sop_txd; 1002#endif 1003 if (ctx.mss) { 1004 gdesc->txd.hlen = ctx.eth_ip_hdr_size + ctx.l4_hdr_size; 1005 gdesc->txd.om = VMXNET3_OM_TSO; 1006 gdesc->txd.msscof = ctx.mss; 1007 le32_add_cpu(&tq->shared->txNumDeferred, (skb->len - 1008 gdesc->txd.hlen + ctx.mss - 1) / ctx.mss); 1009 } else { 1010 if (skb->ip_summed == CHECKSUM_PARTIAL) { 1011 gdesc->txd.hlen = ctx.eth_ip_hdr_size; 1012 gdesc->txd.om = VMXNET3_OM_CSUM; 1013 gdesc->txd.msscof = ctx.eth_ip_hdr_size + 1014 skb->csum_offset; 1015 } else { 1016 gdesc->txd.om = 0; 1017 gdesc->txd.msscof = 0; 1018 } 1019 le32_add_cpu(&tq->shared->txNumDeferred, 1); 1020 } 1021 1022 if (vlan_tx_tag_present(skb)) { 1023 gdesc->txd.ti = 1; 1024 gdesc->txd.tci = vlan_tx_tag_get(skb); 1025 } 1026 1027 /* finally flips the GEN bit of the SOP desc. */ 1028 gdesc->dword[2] = cpu_to_le32(le32_to_cpu(gdesc->dword[2]) ^ 1029 VMXNET3_TXD_GEN); 1030#ifdef __BIG_ENDIAN_BITFIELD 1031 /* Finished updating in bitfields of Tx Desc, so write them in original 1032 * place. 1033 */ 1034 vmxnet3_TxDescToLe((struct Vmxnet3_TxDesc *)gdesc, 1035 (struct Vmxnet3_TxDesc *)ctx.sop_txd); 1036 gdesc = ctx.sop_txd; 1037#endif 1038 dev_dbg(&adapter->netdev->dev, 1039 "txd[%u]: SOP 0x%Lx 0x%x 0x%x\n", 1040 (u32)((union Vmxnet3_GenericDesc *)ctx.sop_txd - 1041 tq->tx_ring.base), le64_to_cpu(gdesc->txd.addr), 1042 le32_to_cpu(gdesc->dword[2]), le32_to_cpu(gdesc->dword[3])); 1043 1044 spin_unlock_irqrestore(&tq->tx_lock, flags); 1045 1046 if (le32_to_cpu(tq->shared->txNumDeferred) >= 1047 le32_to_cpu(tq->shared->txThreshold)) { 1048 tq->shared->txNumDeferred = 0; 1049 VMXNET3_WRITE_BAR0_REG(adapter, 1050 VMXNET3_REG_TXPROD + tq->qid * 8, 1051 tq->tx_ring.next2fill); 1052 } 1053 1054 return NETDEV_TX_OK; 1055 1056hdr_too_big: 1057 tq->stats.drop_oversized_hdr++; 1058unlock_drop_pkt: 1059 spin_unlock_irqrestore(&tq->tx_lock, flags); 1060drop_pkt: 1061 tq->stats.drop_total++; 1062 dev_kfree_skb(skb); 1063 return NETDEV_TX_OK; 1064} 1065 1066 1067static netdev_tx_t 1068vmxnet3_xmit_frame(struct sk_buff *skb, struct net_device *netdev) 1069{ 1070 struct vmxnet3_adapter *adapter = netdev_priv(netdev); 1071 1072 BUG_ON(skb->queue_mapping > adapter->num_tx_queues); 1073 return vmxnet3_tq_xmit(skb, 1074 &adapter->tx_queue[skb->queue_mapping], 1075 adapter, netdev); 1076} 1077 1078 1079static void 1080vmxnet3_rx_csum(struct vmxnet3_adapter *adapter, 1081 struct sk_buff *skb, 1082 union Vmxnet3_GenericDesc *gdesc) 1083{ 1084 if (!gdesc->rcd.cnc && adapter->netdev->features & NETIF_F_RXCSUM) { 1085 /* typical case: TCP/UDP over IP and both csums are correct */ 1086 if ((le32_to_cpu(gdesc->dword[3]) & VMXNET3_RCD_CSUM_OK) == 1087 VMXNET3_RCD_CSUM_OK) { 1088 skb->ip_summed = CHECKSUM_UNNECESSARY; 1089 BUG_ON(!(gdesc->rcd.tcp || gdesc->rcd.udp)); 1090 BUG_ON(!(gdesc->rcd.v4 || gdesc->rcd.v6)); 1091 BUG_ON(gdesc->rcd.frg); 1092 } else { 1093 if (gdesc->rcd.csum) { 1094 skb->csum = htons(gdesc->rcd.csum); 1095 skb->ip_summed = CHECKSUM_PARTIAL; 1096 } else { 1097 skb_checksum_none_assert(skb); 1098 } 1099 } 1100 } else { 1101 skb_checksum_none_assert(skb); 1102 } 1103} 1104 1105 1106static void 1107vmxnet3_rx_error(struct vmxnet3_rx_queue *rq, struct Vmxnet3_RxCompDesc *rcd, 1108 struct vmxnet3_rx_ctx *ctx, struct vmxnet3_adapter *adapter) 1109{ 1110 rq->stats.drop_err++; 1111 if (!rcd->fcs) 1112 rq->stats.drop_fcs++; 1113 1114 rq->stats.drop_total++; 1115 1116 /* 1117 * We do not unmap and chain the rx buffer to the skb. 1118 * We basically pretend this buffer is not used and will be recycled 1119 * by vmxnet3_rq_alloc_rx_buf() 1120 */ 1121 1122 /* 1123 * ctx->skb may be NULL if this is the first and the only one 1124 * desc for the pkt 1125 */ 1126 if (ctx->skb) 1127 dev_kfree_skb_irq(ctx->skb); 1128 1129 ctx->skb = NULL; 1130} 1131 1132 1133static int 1134vmxnet3_rq_rx_complete(struct vmxnet3_rx_queue *rq, 1135 struct vmxnet3_adapter *adapter, int quota) 1136{ 1137 static const u32 rxprod_reg[2] = { 1138 VMXNET3_REG_RXPROD, VMXNET3_REG_RXPROD2 1139 }; 1140 u32 num_rxd = 0; 1141 struct Vmxnet3_RxCompDesc *rcd; 1142 struct vmxnet3_rx_ctx *ctx = &rq->rx_ctx; 1143#ifdef __BIG_ENDIAN_BITFIELD 1144 struct Vmxnet3_RxDesc rxCmdDesc; 1145 struct Vmxnet3_RxCompDesc rxComp; 1146#endif 1147 vmxnet3_getRxComp(rcd, &rq->comp_ring.base[rq->comp_ring.next2proc].rcd, 1148 &rxComp); 1149 while (rcd->gen == rq->comp_ring.gen) { 1150 struct vmxnet3_rx_buf_info *rbi; 1151 struct sk_buff *skb; 1152 int num_to_alloc; 1153 struct Vmxnet3_RxDesc *rxd; 1154 u32 idx, ring_idx; 1155 1156 if (num_rxd >= quota) { 1157 /* we may stop even before we see the EOP desc of 1158 * the current pkt 1159 */ 1160 break; 1161 } 1162 num_rxd++; 1163 BUG_ON(rcd->rqID != rq->qid && rcd->rqID != rq->qid2); 1164 idx = rcd->rxdIdx; 1165 ring_idx = rcd->rqID < adapter->num_rx_queues ? 0 : 1; 1166 vmxnet3_getRxDesc(rxd, &rq->rx_ring[ring_idx].base[idx].rxd, 1167 &rxCmdDesc); 1168 rbi = rq->buf_info[ring_idx] + idx; 1169 1170 BUG_ON(rxd->addr != rbi->dma_addr || 1171 rxd->len != rbi->len); 1172 1173 if (unlikely(rcd->eop && rcd->err)) { 1174 vmxnet3_rx_error(rq, rcd, ctx, adapter); 1175 goto rcd_done; 1176 } 1177 1178 if (rcd->sop) { /* first buf of the pkt */ 1179 BUG_ON(rxd->btype != VMXNET3_RXD_BTYPE_HEAD || 1180 rcd->rqID != rq->qid); 1181 1182 BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_SKB); 1183 BUG_ON(ctx->skb != NULL || rbi->skb == NULL); 1184 1185 if (unlikely(rcd->len == 0)) { 1186 /* Pretend the rx buffer is skipped. */ 1187 BUG_ON(!(rcd->sop && rcd->eop)); 1188 dev_dbg(&adapter->netdev->dev, 1189 "rxRing[%u][%u] 0 length\n", 1190 ring_idx, idx); 1191 goto rcd_done; 1192 } 1193 1194 ctx->skb = rbi->skb; 1195 rbi->skb = NULL; 1196 1197 pci_unmap_single(adapter->pdev, rbi->dma_addr, rbi->len, 1198 PCI_DMA_FROMDEVICE); 1199 1200 skb_put(ctx->skb, rcd->len); 1201 } else { 1202 BUG_ON(ctx->skb == NULL); 1203 /* non SOP buffer must be type 1 in most cases */ 1204 if (rbi->buf_type == VMXNET3_RX_BUF_PAGE) { 1205 BUG_ON(rxd->btype != VMXNET3_RXD_BTYPE_BODY); 1206 1207 if (rcd->len) { 1208 pci_unmap_page(adapter->pdev, 1209 rbi->dma_addr, rbi->len, 1210 PCI_DMA_FROMDEVICE); 1211 1212 vmxnet3_append_frag(ctx->skb, rcd, rbi); 1213 rbi->page = NULL; 1214 } 1215 } else { 1216 /* 1217 * The only time a non-SOP buffer is type 0 is 1218 * when it's EOP and error flag is raised, which 1219 * has already been handled. 1220 */ 1221 BUG_ON(true); 1222 } 1223 } 1224 1225 skb = ctx->skb; 1226 if (rcd->eop) { 1227 skb->len += skb->data_len; 1228 skb->truesize += skb->data_len; 1229 1230 vmxnet3_rx_csum(adapter, skb, 1231 (union Vmxnet3_GenericDesc *)rcd); 1232 skb->protocol = eth_type_trans(skb, adapter->netdev); 1233 1234 if (unlikely(rcd->ts)) 1235 __vlan_hwaccel_put_tag(skb, rcd->tci); 1236 1237 netif_receive_skb(skb); 1238 1239 ctx->skb = NULL; 1240 } 1241 1242rcd_done: 1243 /* device may skip some rx descs */ 1244 rq->rx_ring[ring_idx].next2comp = idx; 1245 VMXNET3_INC_RING_IDX_ONLY(rq->rx_ring[ring_idx].next2comp, 1246 rq->rx_ring[ring_idx].size); 1247 1248 /* refill rx buffers frequently to avoid starving the h/w */ 1249 num_to_alloc = vmxnet3_cmd_ring_desc_avail(rq->rx_ring + 1250 ring_idx); 1251 if (unlikely(num_to_alloc > VMXNET3_RX_ALLOC_THRESHOLD(rq, 1252 ring_idx, adapter))) { 1253 vmxnet3_rq_alloc_rx_buf(rq, ring_idx, num_to_alloc, 1254 adapter); 1255 1256 /* if needed, update the register */ 1257 if (unlikely(rq->shared->updateRxProd)) { 1258 VMXNET3_WRITE_BAR0_REG(adapter, 1259 rxprod_reg[ring_idx] + rq->qid * 8, 1260 rq->rx_ring[ring_idx].next2fill); 1261 rq->uncommitted[ring_idx] = 0; 1262 } 1263 } 1264 1265 vmxnet3_comp_ring_adv_next2proc(&rq->comp_ring); 1266 vmxnet3_getRxComp(rcd, 1267 &rq->comp_ring.base[rq->comp_ring.next2proc].rcd, &rxComp); 1268 } 1269 1270 return num_rxd; 1271} 1272 1273 1274static void 1275vmxnet3_rq_cleanup(struct vmxnet3_rx_queue *rq, 1276 struct vmxnet3_adapter *adapter) 1277{ 1278 u32 i, ring_idx; 1279 struct Vmxnet3_RxDesc *rxd; 1280 1281 for (ring_idx = 0; ring_idx < 2; ring_idx++) { 1282 for (i = 0; i < rq->rx_ring[ring_idx].size; i++) { 1283#ifdef __BIG_ENDIAN_BITFIELD 1284 struct Vmxnet3_RxDesc rxDesc; 1285#endif 1286 vmxnet3_getRxDesc(rxd, 1287 &rq->rx_ring[ring_idx].base[i].rxd, &rxDesc); 1288 1289 if (rxd->btype == VMXNET3_RXD_BTYPE_HEAD && 1290 rq->buf_info[ring_idx][i].skb) { 1291 pci_unmap_single(adapter->pdev, rxd->addr, 1292 rxd->len, PCI_DMA_FROMDEVICE); 1293 dev_kfree_skb(rq->buf_info[ring_idx][i].skb); 1294 rq->buf_info[ring_idx][i].skb = NULL; 1295 } else if (rxd->btype == VMXNET3_RXD_BTYPE_BODY && 1296 rq->buf_info[ring_idx][i].page) { 1297 pci_unmap_page(adapter->pdev, rxd->addr, 1298 rxd->len, PCI_DMA_FROMDEVICE); 1299 put_page(rq->buf_info[ring_idx][i].page); 1300 rq->buf_info[ring_idx][i].page = NULL; 1301 } 1302 } 1303 1304 rq->rx_ring[ring_idx].gen = VMXNET3_INIT_GEN; 1305 rq->rx_ring[ring_idx].next2fill = 1306 rq->rx_ring[ring_idx].next2comp = 0; 1307 rq->uncommitted[ring_idx] = 0; 1308 } 1309 1310 rq->comp_ring.gen = VMXNET3_INIT_GEN; 1311 rq->comp_ring.next2proc = 0; 1312} 1313 1314 1315static void 1316vmxnet3_rq_cleanup_all(struct vmxnet3_adapter *adapter) 1317{ 1318 int i; 1319 1320 for (i = 0; i < adapter->num_rx_queues; i++) 1321 vmxnet3_rq_cleanup(&adapter->rx_queue[i], adapter); 1322} 1323 1324 1325void vmxnet3_rq_destroy(struct vmxnet3_rx_queue *rq, 1326 struct vmxnet3_adapter *adapter) 1327{ 1328 int i; 1329 int j; 1330 1331 /* all rx buffers must have already been freed */ 1332 for (i = 0; i < 2; i++) { 1333 if (rq->buf_info[i]) { 1334 for (j = 0; j < rq->rx_ring[i].size; j++) 1335 BUG_ON(rq->buf_info[i][j].page != NULL); 1336 } 1337 } 1338 1339 1340 kfree(rq->buf_info[0]); 1341 1342 for (i = 0; i < 2; i++) { 1343 if (rq->rx_ring[i].base) { 1344 pci_free_consistent(adapter->pdev, rq->rx_ring[i].size 1345 * sizeof(struct Vmxnet3_RxDesc), 1346 rq->rx_ring[i].base, 1347 rq->rx_ring[i].basePA); 1348 rq->rx_ring[i].base = NULL; 1349 } 1350 rq->buf_info[i] = NULL; 1351 } 1352 1353 if (rq->comp_ring.base) { 1354 pci_free_consistent(adapter->pdev, rq->comp_ring.size * 1355 sizeof(struct Vmxnet3_RxCompDesc), 1356 rq->comp_ring.base, rq->comp_ring.basePA); 1357 rq->comp_ring.base = NULL; 1358 } 1359} 1360 1361 1362static int 1363vmxnet3_rq_init(struct vmxnet3_rx_queue *rq, 1364 struct vmxnet3_adapter *adapter) 1365{ 1366 int i; 1367 1368 /* initialize buf_info */ 1369 for (i = 0; i < rq->rx_ring[0].size; i++) { 1370 1371 /* 1st buf for a pkt is skbuff */ 1372 if (i % adapter->rx_buf_per_pkt == 0) { 1373 rq->buf_info[0][i].buf_type = VMXNET3_RX_BUF_SKB; 1374 rq->buf_info[0][i].len = adapter->skb_buf_size; 1375 } else { /* subsequent bufs for a pkt is frag */ 1376 rq->buf_info[0][i].buf_type = VMXNET3_RX_BUF_PAGE; 1377 rq->buf_info[0][i].len = PAGE_SIZE; 1378 } 1379 } 1380 for (i = 0; i < rq->rx_ring[1].size; i++) { 1381 rq->buf_info[1][i].buf_type = VMXNET3_RX_BUF_PAGE; 1382 rq->buf_info[1][i].len = PAGE_SIZE; 1383 } 1384 1385 /* reset internal state and allocate buffers for both rings */ 1386 for (i = 0; i < 2; i++) { 1387 rq->rx_ring[i].next2fill = rq->rx_ring[i].next2comp = 0; 1388 rq->uncommitted[i] = 0; 1389 1390 memset(rq->rx_ring[i].base, 0, rq->rx_ring[i].size * 1391 sizeof(struct Vmxnet3_RxDesc)); 1392 rq->rx_ring[i].gen = VMXNET3_INIT_GEN; 1393 } 1394 if (vmxnet3_rq_alloc_rx_buf(rq, 0, rq->rx_ring[0].size - 1, 1395 adapter) == 0) { 1396 /* at least has 1 rx buffer for the 1st ring */ 1397 return -ENOMEM; 1398 } 1399 vmxnet3_rq_alloc_rx_buf(rq, 1, rq->rx_ring[1].size - 1, adapter); 1400 1401 /* reset the comp ring */ 1402 rq->comp_ring.next2proc = 0; 1403 memset(rq->comp_ring.base, 0, rq->comp_ring.size * 1404 sizeof(struct Vmxnet3_RxCompDesc)); 1405 rq->comp_ring.gen = VMXNET3_INIT_GEN; 1406 1407 /* reset rxctx */ 1408 rq->rx_ctx.skb = NULL; 1409 1410 /* stats are not reset */ 1411 return 0; 1412} 1413 1414 1415static int 1416vmxnet3_rq_init_all(struct vmxnet3_adapter *adapter) 1417{ 1418 int i, err = 0; 1419 1420 for (i = 0; i < adapter->num_rx_queues; i++) { 1421 err = vmxnet3_rq_init(&adapter->rx_queue[i], adapter); 1422 if (unlikely(err)) { 1423 dev_err(&adapter->netdev->dev, "%s: failed to " 1424 "initialize rx queue%i\n", 1425 adapter->netdev->name, i); 1426 break; 1427 } 1428 } 1429 return err; 1430 1431} 1432 1433 1434static int 1435vmxnet3_rq_create(struct vmxnet3_rx_queue *rq, struct vmxnet3_adapter *adapter) 1436{ 1437 int i; 1438 size_t sz; 1439 struct vmxnet3_rx_buf_info *bi; 1440 1441 for (i = 0; i < 2; i++) { 1442 1443 sz = rq->rx_ring[i].size * sizeof(struct Vmxnet3_RxDesc); 1444 rq->rx_ring[i].base = pci_alloc_consistent(adapter->pdev, sz, 1445 &rq->rx_ring[i].basePA); 1446 if (!rq->rx_ring[i].base) { 1447 printk(KERN_ERR "%s: failed to allocate rx ring %d\n", 1448 adapter->netdev->name, i); 1449 goto err; 1450 } 1451 } 1452 1453 sz = rq->comp_ring.size * sizeof(struct Vmxnet3_RxCompDesc); 1454 rq->comp_ring.base = pci_alloc_consistent(adapter->pdev, sz, 1455 &rq->comp_ring.basePA); 1456 if (!rq->comp_ring.base) { 1457 printk(KERN_ERR "%s: failed to allocate rx comp ring\n", 1458 adapter->netdev->name); 1459 goto err; 1460 } 1461 1462 sz = sizeof(struct vmxnet3_rx_buf_info) * (rq->rx_ring[0].size + 1463 rq->rx_ring[1].size); 1464 bi = kzalloc(sz, GFP_KERNEL); 1465 if (!bi) { 1466 printk(KERN_ERR "%s: failed to allocate rx bufinfo\n", 1467 adapter->netdev->name); 1468 goto err; 1469 } 1470 rq->buf_info[0] = bi; 1471 rq->buf_info[1] = bi + rq->rx_ring[0].size; 1472 1473 return 0; 1474 1475err: 1476 vmxnet3_rq_destroy(rq, adapter); 1477 return -ENOMEM; 1478} 1479 1480 1481static int 1482vmxnet3_rq_create_all(struct vmxnet3_adapter *adapter) 1483{ 1484 int i, err = 0; 1485 1486 for (i = 0; i < adapter->num_rx_queues; i++) { 1487 err = vmxnet3_rq_create(&adapter->rx_queue[i], adapter); 1488 if (unlikely(err)) { 1489 dev_err(&adapter->netdev->dev, 1490 "%s: failed to create rx queue%i\n", 1491 adapter->netdev->name, i); 1492 goto err_out; 1493 } 1494 } 1495 return err; 1496err_out: 1497 vmxnet3_rq_destroy_all(adapter); 1498 return err; 1499 1500} 1501 1502/* Multiple queue aware polling function for tx and rx */ 1503 1504static int 1505vmxnet3_do_poll(struct vmxnet3_adapter *adapter, int budget) 1506{ 1507 int rcd_done = 0, i; 1508 if (unlikely(adapter->shared->ecr)) 1509 vmxnet3_process_events(adapter); 1510 for (i = 0; i < adapter->num_tx_queues; i++) 1511 vmxnet3_tq_tx_complete(&adapter->tx_queue[i], adapter); 1512 1513 for (i = 0; i < adapter->num_rx_queues; i++) 1514 rcd_done += vmxnet3_rq_rx_complete(&adapter->rx_queue[i], 1515 adapter, budget); 1516 return rcd_done; 1517} 1518 1519 1520static int 1521vmxnet3_poll(struct napi_struct *napi, int budget) 1522{ 1523 struct vmxnet3_rx_queue *rx_queue = container_of(napi, 1524 struct vmxnet3_rx_queue, napi); 1525 int rxd_done; 1526 1527 rxd_done = vmxnet3_do_poll(rx_queue->adapter, budget); 1528 1529 if (rxd_done < budget) { 1530 napi_complete(napi); 1531 vmxnet3_enable_all_intrs(rx_queue->adapter); 1532 } 1533 return rxd_done; 1534} 1535 1536/* 1537 * NAPI polling function for MSI-X mode with multiple Rx queues 1538 * Returns the # of the NAPI credit consumed (# of rx descriptors processed) 1539 */ 1540 1541static int 1542vmxnet3_poll_rx_only(struct napi_struct *napi, int budget) 1543{ 1544 struct vmxnet3_rx_queue *rq = container_of(napi, 1545 struct vmxnet3_rx_queue, napi); 1546 struct vmxnet3_adapter *adapter = rq->adapter; 1547 int rxd_done; 1548 1549 /* When sharing interrupt with corresponding tx queue, process 1550 * tx completions in that queue as well 1551 */ 1552 if (adapter->share_intr == VMXNET3_INTR_BUDDYSHARE) { 1553 struct vmxnet3_tx_queue *tq = 1554 &adapter->tx_queue[rq - adapter->rx_queue]; 1555 vmxnet3_tq_tx_complete(tq, adapter); 1556 } 1557 1558 rxd_done = vmxnet3_rq_rx_complete(rq, adapter, budget); 1559 1560 if (rxd_done < budget) { 1561 napi_complete(napi); 1562 vmxnet3_enable_intr(adapter, rq->comp_ring.intr_idx); 1563 } 1564 return rxd_done; 1565} 1566 1567 1568#ifdef CONFIG_PCI_MSI 1569 1570/* 1571 * Handle completion interrupts on tx queues 1572 * Returns whether or not the intr is handled 1573 */ 1574 1575static irqreturn_t 1576vmxnet3_msix_tx(int irq, void *data) 1577{ 1578 struct vmxnet3_tx_queue *tq = data; 1579 struct vmxnet3_adapter *adapter = tq->adapter; 1580 1581 if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE) 1582 vmxnet3_disable_intr(adapter, tq->comp_ring.intr_idx); 1583 1584 /* Handle the case where only one irq is allocate for all tx queues */ 1585 if (adapter->share_intr == VMXNET3_INTR_TXSHARE) { 1586 int i; 1587 for (i = 0; i < adapter->num_tx_queues; i++) { 1588 struct vmxnet3_tx_queue *txq = &adapter->tx_queue[i]; 1589 vmxnet3_tq_tx_complete(txq, adapter); 1590 } 1591 } else { 1592 vmxnet3_tq_tx_complete(tq, adapter); 1593 } 1594 vmxnet3_enable_intr(adapter, tq->comp_ring.intr_idx); 1595 1596 return IRQ_HANDLED; 1597} 1598 1599 1600/* 1601 * Handle completion interrupts on rx queues. Returns whether or not the 1602 * intr is handled 1603 */ 1604 1605static irqreturn_t 1606vmxnet3_msix_rx(int irq, void *data) 1607{ 1608 struct vmxnet3_rx_queue *rq = data; 1609 struct vmxnet3_adapter *adapter = rq->adapter; 1610 1611 /* disable intr if needed */ 1612 if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE) 1613 vmxnet3_disable_intr(adapter, rq->comp_ring.intr_idx); 1614 napi_schedule(&rq->napi); 1615 1616 return IRQ_HANDLED; 1617} 1618 1619/* 1620 *---------------------------------------------------------------------------- 1621 * 1622 * vmxnet3_msix_event -- 1623 * 1624 * vmxnet3 msix event intr handler 1625 * 1626 * Result: 1627 * whether or not the intr is handled 1628 * 1629 *---------------------------------------------------------------------------- 1630 */ 1631 1632static irqreturn_t 1633vmxnet3_msix_event(int irq, void *data) 1634{ 1635 struct net_device *dev = data; 1636 struct vmxnet3_adapter *adapter = netdev_priv(dev); 1637 1638 /* disable intr if needed */ 1639 if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE) 1640 vmxnet3_disable_intr(adapter, adapter->intr.event_intr_idx); 1641 1642 if (adapter->shared->ecr) 1643 vmxnet3_process_events(adapter); 1644 1645 vmxnet3_enable_intr(adapter, adapter->intr.event_intr_idx); 1646 1647 return IRQ_HANDLED; 1648} 1649 1650#endif /* CONFIG_PCI_MSI */ 1651 1652 1653/* Interrupt handler for vmxnet3 */ 1654static irqreturn_t 1655vmxnet3_intr(int irq, void *dev_id) 1656{ 1657 struct net_device *dev = dev_id; 1658 struct vmxnet3_adapter *adapter = netdev_priv(dev); 1659 1660 if (adapter->intr.type == VMXNET3_IT_INTX) { 1661 u32 icr = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_ICR); 1662 if (unlikely(icr == 0)) 1663 /* not ours */ 1664 return IRQ_NONE; 1665 } 1666 1667 1668 /* disable intr if needed */ 1669 if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE) 1670 vmxnet3_disable_all_intrs(adapter); 1671 1672 napi_schedule(&adapter->rx_queue[0].napi); 1673 1674 return IRQ_HANDLED; 1675} 1676 1677#ifdef CONFIG_NET_POLL_CONTROLLER 1678 1679/* netpoll callback. */ 1680static void 1681vmxnet3_netpoll(struct net_device *netdev) 1682{ 1683 struct vmxnet3_adapter *adapter = netdev_priv(netdev); 1684 1685 if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE) 1686 vmxnet3_disable_all_intrs(adapter); 1687 1688 vmxnet3_do_poll(adapter, adapter->rx_queue[0].rx_ring[0].size); 1689 vmxnet3_enable_all_intrs(adapter); 1690 1691} 1692#endif /* CONFIG_NET_POLL_CONTROLLER */ 1693 1694static int 1695vmxnet3_request_irqs(struct vmxnet3_adapter *adapter) 1696{ 1697 struct vmxnet3_intr *intr = &adapter->intr; 1698 int err = 0, i; 1699 int vector = 0; 1700 1701#ifdef CONFIG_PCI_MSI 1702 if (adapter->intr.type == VMXNET3_IT_MSIX) { 1703 for (i = 0; i < adapter->num_tx_queues; i++) { 1704 if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE) { 1705 sprintf(adapter->tx_queue[i].name, "%s-tx-%d", 1706 adapter->netdev->name, vector); 1707 err = request_irq( 1708 intr->msix_entries[vector].vector, 1709 vmxnet3_msix_tx, 0, 1710 adapter->tx_queue[i].name, 1711 &adapter->tx_queue[i]); 1712 } else { 1713 sprintf(adapter->tx_queue[i].name, "%s-rxtx-%d", 1714 adapter->netdev->name, vector); 1715 } 1716 if (err) { 1717 dev_err(&adapter->netdev->dev, 1718 "Failed to request irq for MSIX, %s, " 1719 "error %d\n", 1720 adapter->tx_queue[i].name, err); 1721 return err; 1722 } 1723 1724 /* Handle the case where only 1 MSIx was allocated for 1725 * all tx queues */ 1726 if (adapter->share_intr == VMXNET3_INTR_TXSHARE) { 1727 for (; i < adapter->num_tx_queues; i++) 1728 adapter->tx_queue[i].comp_ring.intr_idx 1729 = vector; 1730 vector++; 1731 break; 1732 } else { 1733 adapter->tx_queue[i].comp_ring.intr_idx 1734 = vector++; 1735 } 1736 } 1737 if (adapter->share_intr == VMXNET3_INTR_BUDDYSHARE) 1738 vector = 0; 1739 1740 for (i = 0; i < adapter->num_rx_queues; i++) { 1741 if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE) 1742 sprintf(adapter->rx_queue[i].name, "%s-rx-%d", 1743 adapter->netdev->name, vector); 1744 else 1745 sprintf(adapter->rx_queue[i].name, "%s-rxtx-%d", 1746 adapter->netdev->name, vector); 1747 err = request_irq(intr->msix_entries[vector].vector, 1748 vmxnet3_msix_rx, 0, 1749 adapter->rx_queue[i].name, 1750 &(adapter->rx_queue[i])); 1751 if (err) { 1752 printk(KERN_ERR "Failed to request irq for MSIX" 1753 ", %s, error %d\n", 1754 adapter->rx_queue[i].name, err); 1755 return err; 1756 } 1757 1758 adapter->rx_queue[i].comp_ring.intr_idx = vector++; 1759 } 1760 1761 sprintf(intr->event_msi_vector_name, "%s-event-%d", 1762 adapter->netdev->name, vector); 1763 err = request_irq(intr->msix_entries[vector].vector, 1764 vmxnet3_msix_event, 0, 1765 intr->event_msi_vector_name, adapter->netdev); 1766 intr->event_intr_idx = vector; 1767 1768 } else if (intr->type == VMXNET3_IT_MSI) { 1769 adapter->num_rx_queues = 1; 1770 err = request_irq(adapter->pdev->irq, vmxnet3_intr, 0, 1771 adapter->netdev->name, adapter->netdev); 1772 } else { 1773#endif 1774 adapter->num_rx_queues = 1; 1775 err = request_irq(adapter->pdev->irq, vmxnet3_intr, 1776 IRQF_SHARED, adapter->netdev->name, 1777 adapter->netdev); 1778#ifdef CONFIG_PCI_MSI 1779 } 1780#endif 1781 intr->num_intrs = vector + 1; 1782 if (err) { 1783 printk(KERN_ERR "Failed to request irq %s (intr type:%d), error" 1784 ":%d\n", adapter->netdev->name, intr->type, err); 1785 } else { 1786 /* Number of rx queues will not change after this */ 1787 for (i = 0; i < adapter->num_rx_queues; i++) { 1788 struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i]; 1789 rq->qid = i; 1790 rq->qid2 = i + adapter->num_rx_queues; 1791 } 1792 1793 1794 1795 /* init our intr settings */ 1796 for (i = 0; i < intr->num_intrs; i++) 1797 intr->mod_levels[i] = UPT1_IML_ADAPTIVE; 1798 if (adapter->intr.type != VMXNET3_IT_MSIX) { 1799 adapter->intr.event_intr_idx = 0; 1800 for (i = 0; i < adapter->num_tx_queues; i++) 1801 adapter->tx_queue[i].comp_ring.intr_idx = 0; 1802 adapter->rx_queue[0].comp_ring.intr_idx = 0; 1803 } 1804 1805 printk(KERN_INFO "%s: intr type %u, mode %u, %u vectors " 1806 "allocated\n", adapter->netdev->name, intr->type, 1807 intr->mask_mode, intr->num_intrs); 1808 } 1809 1810 return err; 1811} 1812 1813 1814static void 1815vmxnet3_free_irqs(struct vmxnet3_adapter *adapter) 1816{ 1817 struct vmxnet3_intr *intr = &adapter->intr; 1818 BUG_ON(intr->type == VMXNET3_IT_AUTO || intr->num_intrs <= 0); 1819 1820 switch (intr->type) { 1821#ifdef CONFIG_PCI_MSI 1822 case VMXNET3_IT_MSIX: 1823 { 1824 int i, vector = 0; 1825 1826 if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE) { 1827 for (i = 0; i < adapter->num_tx_queues; i++) { 1828 free_irq(intr->msix_entries[vector++].vector, 1829 &(adapter->tx_queue[i])); 1830 if (adapter->share_intr == VMXNET3_INTR_TXSHARE) 1831 break; 1832 } 1833 } 1834 1835 for (i = 0; i < adapter->num_rx_queues; i++) { 1836 free_irq(intr->msix_entries[vector++].vector, 1837 &(adapter->rx_queue[i])); 1838 } 1839 1840 free_irq(intr->msix_entries[vector].vector, 1841 adapter->netdev); 1842 BUG_ON(vector >= intr->num_intrs); 1843 break; 1844 } 1845#endif 1846 case VMXNET3_IT_MSI: 1847 free_irq(adapter->pdev->irq, adapter->netdev); 1848 break; 1849 case VMXNET3_IT_INTX: 1850 free_irq(adapter->pdev->irq, adapter->netdev); 1851 break; 1852 default: 1853 BUG_ON(true); 1854 } 1855} 1856 1857 1858static void 1859vmxnet3_restore_vlan(struct vmxnet3_adapter *adapter) 1860{ 1861 u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable; 1862 u16 vid; 1863 1864 /* allow untagged pkts */ 1865 VMXNET3_SET_VFTABLE_ENTRY(vfTable, 0); 1866 1867 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID) 1868 VMXNET3_SET_VFTABLE_ENTRY(vfTable, vid); 1869} 1870 1871 1872static void 1873vmxnet3_vlan_rx_add_vid(struct net_device *netdev, u16 vid) 1874{ 1875 struct vmxnet3_adapter *adapter = netdev_priv(netdev); 1876 u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable; 1877 unsigned long flags; 1878 1879 VMXNET3_SET_VFTABLE_ENTRY(vfTable, vid); 1880 spin_lock_irqsave(&adapter->cmd_lock, flags); 1881 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, 1882 VMXNET3_CMD_UPDATE_VLAN_FILTERS); 1883 spin_unlock_irqrestore(&adapter->cmd_lock, flags); 1884 1885 set_bit(vid, adapter->active_vlans); 1886} 1887 1888 1889static void 1890vmxnet3_vlan_rx_kill_vid(struct net_device *netdev, u16 vid) 1891{ 1892 struct vmxnet3_adapter *adapter = netdev_priv(netdev); 1893 u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable; 1894 unsigned long flags; 1895 1896 VMXNET3_CLEAR_VFTABLE_ENTRY(vfTable, vid); 1897 spin_lock_irqsave(&adapter->cmd_lock, flags); 1898 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, 1899 VMXNET3_CMD_UPDATE_VLAN_FILTERS); 1900 spin_unlock_irqrestore(&adapter->cmd_lock, flags); 1901 1902 clear_bit(vid, adapter->active_vlans); 1903} 1904 1905 1906static u8 * 1907vmxnet3_copy_mc(struct net_device *netdev) 1908{ 1909 u8 *buf = NULL; 1910 u32 sz = netdev_mc_count(netdev) * ETH_ALEN; 1911 1912 /* struct Vmxnet3_RxFilterConf.mfTableLen is u16. */ 1913 if (sz <= 0xffff) { 1914 /* We may be called with BH disabled */ 1915 buf = kmalloc(sz, GFP_ATOMIC); 1916 if (buf) { 1917 struct netdev_hw_addr *ha; 1918 int i = 0; 1919 1920 netdev_for_each_mc_addr(ha, netdev) 1921 memcpy(buf + i++ * ETH_ALEN, ha->addr, 1922 ETH_ALEN); 1923 } 1924 } 1925 return buf; 1926} 1927 1928 1929static void 1930vmxnet3_set_mc(struct net_device *netdev) 1931{ 1932 struct vmxnet3_adapter *adapter = netdev_priv(netdev); 1933 unsigned long flags; 1934 struct Vmxnet3_RxFilterConf *rxConf = 1935 &adapter->shared->devRead.rxFilterConf; 1936 u8 *new_table = NULL; 1937 u32 new_mode = VMXNET3_RXM_UCAST; 1938 1939 if (netdev->flags & IFF_PROMISC) { 1940 u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable; 1941 memset(vfTable, 0, VMXNET3_VFT_SIZE * sizeof(*vfTable)); 1942 1943 new_mode |= VMXNET3_RXM_PROMISC; 1944 } else { 1945 vmxnet3_restore_vlan(adapter); 1946 } 1947 1948 if (netdev->flags & IFF_BROADCAST) 1949 new_mode |= VMXNET3_RXM_BCAST; 1950 1951 if (netdev->flags & IFF_ALLMULTI) 1952 new_mode |= VMXNET3_RXM_ALL_MULTI; 1953 else 1954 if (!netdev_mc_empty(netdev)) { 1955 new_table = vmxnet3_copy_mc(netdev); 1956 if (new_table) { 1957 new_mode |= VMXNET3_RXM_MCAST; 1958 rxConf->mfTableLen = cpu_to_le16( 1959 netdev_mc_count(netdev) * ETH_ALEN); 1960 rxConf->mfTablePA = cpu_to_le64(virt_to_phys( 1961 new_table)); 1962 } else { 1963 printk(KERN_INFO "%s: failed to copy mcast list" 1964 ", setting ALL_MULTI\n", netdev->name); 1965 new_mode |= VMXNET3_RXM_ALL_MULTI; 1966 } 1967 } 1968 1969 1970 if (!(new_mode & VMXNET3_RXM_MCAST)) { 1971 rxConf->mfTableLen = 0; 1972 rxConf->mfTablePA = 0; 1973 } 1974 1975 spin_lock_irqsave(&adapter->cmd_lock, flags); 1976 if (new_mode != rxConf->rxMode) { 1977 rxConf->rxMode = cpu_to_le32(new_mode); 1978 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, 1979 VMXNET3_CMD_UPDATE_RX_MODE); 1980 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, 1981 VMXNET3_CMD_UPDATE_VLAN_FILTERS); 1982 } 1983 1984 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, 1985 VMXNET3_CMD_UPDATE_MAC_FILTERS); 1986 spin_unlock_irqrestore(&adapter->cmd_lock, flags); 1987 1988 kfree(new_table); 1989} 1990 1991void 1992vmxnet3_rq_destroy_all(struct vmxnet3_adapter *adapter) 1993{ 1994 int i; 1995 1996 for (i = 0; i < adapter->num_rx_queues; i++) 1997 vmxnet3_rq_destroy(&adapter->rx_queue[i], adapter); 1998} 1999 2000 2001/* 2002 * Set up driver_shared based on settings in adapter. 2003 */ 2004 2005static void 2006vmxnet3_setup_driver_shared(struct vmxnet3_adapter *adapter) 2007{ 2008 struct Vmxnet3_DriverShared *shared = adapter->shared; 2009 struct Vmxnet3_DSDevRead *devRead = &shared->devRead; 2010 struct Vmxnet3_TxQueueConf *tqc; 2011 struct Vmxnet3_RxQueueConf *rqc; 2012 int i; 2013 2014 memset(shared, 0, sizeof(*shared)); 2015 2016 /* driver settings */ 2017 shared->magic = cpu_to_le32(VMXNET3_REV1_MAGIC); 2018 devRead->misc.driverInfo.version = cpu_to_le32( 2019 VMXNET3_DRIVER_VERSION_NUM); 2020 devRead->misc.driverInfo.gos.gosBits = (sizeof(void *) == 4 ? 2021 VMXNET3_GOS_BITS_32 : VMXNET3_GOS_BITS_64); 2022 devRead->misc.driverInfo.gos.gosType = VMXNET3_GOS_TYPE_LINUX; 2023 *((u32 *)&devRead->misc.driverInfo.gos) = cpu_to_le32( 2024 *((u32 *)&devRead->misc.driverInfo.gos)); 2025 devRead->misc.driverInfo.vmxnet3RevSpt = cpu_to_le32(1); 2026 devRead->misc.driverInfo.uptVerSpt = cpu_to_le32(1); 2027 2028 devRead->misc.ddPA = cpu_to_le64(virt_to_phys(adapter)); 2029 devRead->misc.ddLen = cpu_to_le32(sizeof(struct vmxnet3_adapter)); 2030 2031 /* set up feature flags */ 2032 if (adapter->netdev->features & NETIF_F_RXCSUM) 2033 devRead->misc.uptFeatures |= UPT1_F_RXCSUM; 2034 2035 if (adapter->netdev->features & NETIF_F_LRO) { 2036 devRead->misc.uptFeatures |= UPT1_F_LRO; 2037 devRead->misc.maxNumRxSG = cpu_to_le16(1 + MAX_SKB_FRAGS); 2038 } 2039 if (adapter->netdev->features & NETIF_F_HW_VLAN_RX) 2040 devRead->misc.uptFeatures |= UPT1_F_RXVLAN; 2041 2042 devRead->misc.mtu = cpu_to_le32(adapter->netdev->mtu); 2043 devRead->misc.queueDescPA = cpu_to_le64(adapter->queue_desc_pa); 2044 devRead->misc.queueDescLen = cpu_to_le32( 2045 adapter->num_tx_queues * sizeof(struct Vmxnet3_TxQueueDesc) + 2046 adapter->num_rx_queues * sizeof(struct Vmxnet3_RxQueueDesc)); 2047 2048 /* tx queue settings */ 2049 devRead->misc.numTxQueues = adapter->num_tx_queues; 2050 for (i = 0; i < adapter->num_tx_queues; i++) { 2051 struct vmxnet3_tx_queue *tq = &adapter->tx_queue[i]; 2052 BUG_ON(adapter->tx_queue[i].tx_ring.base == NULL); 2053 tqc = &adapter->tqd_start[i].conf; 2054 tqc->txRingBasePA = cpu_to_le64(tq->tx_ring.basePA); 2055 tqc->dataRingBasePA = cpu_to_le64(tq->data_ring.basePA); 2056 tqc->compRingBasePA = cpu_to_le64(tq->comp_ring.basePA); 2057 tqc->ddPA = cpu_to_le64(virt_to_phys(tq->buf_info)); 2058 tqc->txRingSize = cpu_to_le32(tq->tx_ring.size); 2059 tqc->dataRingSize = cpu_to_le32(tq->data_ring.size); 2060 tqc->compRingSize = cpu_to_le32(tq->comp_ring.size); 2061 tqc->ddLen = cpu_to_le32( 2062 sizeof(struct vmxnet3_tx_buf_info) * 2063 tqc->txRingSize); 2064 tqc->intrIdx = tq->comp_ring.intr_idx; 2065 } 2066 2067 /* rx queue settings */ 2068 devRead->misc.numRxQueues = adapter->num_rx_queues; 2069 for (i = 0; i < adapter->num_rx_queues; i++) { 2070 struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i]; 2071 rqc = &adapter->rqd_start[i].conf; 2072 rqc->rxRingBasePA[0] = cpu_to_le64(rq->rx_ring[0].basePA); 2073 rqc->rxRingBasePA[1] = cpu_to_le64(rq->rx_ring[1].basePA); 2074 rqc->compRingBasePA = cpu_to_le64(rq->comp_ring.basePA); 2075 rqc->ddPA = cpu_to_le64(virt_to_phys( 2076 rq->buf_info)); 2077 rqc->rxRingSize[0] = cpu_to_le32(rq->rx_ring[0].size); 2078 rqc->rxRingSize[1] = cpu_to_le32(rq->rx_ring[1].size); 2079 rqc->compRingSize = cpu_to_le32(rq->comp_ring.size); 2080 rqc->ddLen = cpu_to_le32( 2081 sizeof(struct vmxnet3_rx_buf_info) * 2082 (rqc->rxRingSize[0] + 2083 rqc->rxRingSize[1])); 2084 rqc->intrIdx = rq->comp_ring.intr_idx; 2085 } 2086 2087#ifdef VMXNET3_RSS 2088 memset(adapter->rss_conf, 0, sizeof(*adapter->rss_conf)); 2089 2090 if (adapter->rss) { 2091 struct UPT1_RSSConf *rssConf = adapter->rss_conf; 2092 devRead->misc.uptFeatures |= UPT1_F_RSS; 2093 devRead->misc.numRxQueues = adapter->num_rx_queues; 2094 rssConf->hashType = UPT1_RSS_HASH_TYPE_TCP_IPV4 | 2095 UPT1_RSS_HASH_TYPE_IPV4 | 2096 UPT1_RSS_HASH_TYPE_TCP_IPV6 | 2097 UPT1_RSS_HASH_TYPE_IPV6; 2098 rssConf->hashFunc = UPT1_RSS_HASH_FUNC_TOEPLITZ; 2099 rssConf->hashKeySize = UPT1_RSS_MAX_KEY_SIZE; 2100 rssConf->indTableSize = VMXNET3_RSS_IND_TABLE_SIZE; 2101 get_random_bytes(&rssConf->hashKey[0], rssConf->hashKeySize); 2102 for (i = 0; i < rssConf->indTableSize; i++) 2103 rssConf->indTable[i] = i % adapter->num_rx_queues; 2104 2105 devRead->rssConfDesc.confVer = 1; 2106 devRead->rssConfDesc.confLen = sizeof(*rssConf); 2107 devRead->rssConfDesc.confPA = virt_to_phys(rssConf); 2108 } 2109 2110#endif /* VMXNET3_RSS */ 2111 2112 /* intr settings */ 2113 devRead->intrConf.autoMask = adapter->intr.mask_mode == 2114 VMXNET3_IMM_AUTO; 2115 devRead->intrConf.numIntrs = adapter->intr.num_intrs; 2116 for (i = 0; i < adapter->intr.num_intrs; i++) 2117 devRead->intrConf.modLevels[i] = adapter->intr.mod_levels[i]; 2118 2119 devRead->intrConf.eventIntrIdx = adapter->intr.event_intr_idx; 2120 devRead->intrConf.intrCtrl |= cpu_to_le32(VMXNET3_IC_DISABLE_ALL); 2121 2122 /* rx filter settings */ 2123 devRead->rxFilterConf.rxMode = 0; 2124 vmxnet3_restore_vlan(adapter); 2125 vmxnet3_write_mac_addr(adapter, adapter->netdev->dev_addr); 2126 2127 /* the rest are already zeroed */ 2128} 2129 2130 2131int 2132vmxnet3_activate_dev(struct vmxnet3_adapter *adapter) 2133{ 2134 int err, i; 2135 u32 ret; 2136 unsigned long flags; 2137 2138 dev_dbg(&adapter->netdev->dev, "%s: skb_buf_size %d, rx_buf_per_pkt %d," 2139 " ring sizes %u %u %u\n", adapter->netdev->name, 2140 adapter->skb_buf_size, adapter->rx_buf_per_pkt, 2141 adapter->tx_queue[0].tx_ring.size, 2142 adapter->rx_queue[0].rx_ring[0].size, 2143 adapter->rx_queue[0].rx_ring[1].size); 2144 2145 vmxnet3_tq_init_all(adapter); 2146 err = vmxnet3_rq_init_all(adapter); 2147 if (err) { 2148 printk(KERN_ERR "Failed to init rx queue for %s: error %d\n", 2149 adapter->netdev->name, err); 2150 goto rq_err; 2151 } 2152 2153 err = vmxnet3_request_irqs(adapter); 2154 if (err) { 2155 printk(KERN_ERR "Failed to setup irq for %s: error %d\n", 2156 adapter->netdev->name, err); 2157 goto irq_err; 2158 } 2159 2160 vmxnet3_setup_driver_shared(adapter); 2161 2162 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL, VMXNET3_GET_ADDR_LO( 2163 adapter->shared_pa)); 2164 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH, VMXNET3_GET_ADDR_HI( 2165 adapter->shared_pa)); 2166 spin_lock_irqsave(&adapter->cmd_lock, flags); 2167 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, 2168 VMXNET3_CMD_ACTIVATE_DEV); 2169 ret = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD); 2170 spin_unlock_irqrestore(&adapter->cmd_lock, flags); 2171 2172 if (ret != 0) { 2173 printk(KERN_ERR "Failed to activate dev %s: error %u\n", 2174 adapter->netdev->name, ret); 2175 err = -EINVAL; 2176 goto activate_err; 2177 } 2178 2179 for (i = 0; i < adapter->num_rx_queues; i++) { 2180 VMXNET3_WRITE_BAR0_REG(adapter, 2181 VMXNET3_REG_RXPROD + i * VMXNET3_REG_ALIGN, 2182 adapter->rx_queue[i].rx_ring[0].next2fill); 2183 VMXNET3_WRITE_BAR0_REG(adapter, (VMXNET3_REG_RXPROD2 + 2184 (i * VMXNET3_REG_ALIGN)), 2185 adapter->rx_queue[i].rx_ring[1].next2fill); 2186 } 2187 2188 /* Apply the rx filter settins last. */ 2189 vmxnet3_set_mc(adapter->netdev); 2190 2191 /* 2192 * Check link state when first activating device. It will start the 2193 * tx queue if the link is up. 2194 */ 2195 vmxnet3_check_link(adapter, true); 2196 for (i = 0; i < adapter->num_rx_queues; i++) 2197 napi_enable(&adapter->rx_queue[i].napi); 2198 vmxnet3_enable_all_intrs(adapter); 2199 clear_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state); 2200 return 0; 2201 2202activate_err: 2203 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL, 0); 2204 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH, 0); 2205 vmxnet3_free_irqs(adapter); 2206irq_err: 2207rq_err: 2208 /* free up buffers we allocated */ 2209 vmxnet3_rq_cleanup_all(adapter); 2210 return err; 2211} 2212 2213 2214void 2215vmxnet3_reset_dev(struct vmxnet3_adapter *adapter) 2216{ 2217 unsigned long flags; 2218 spin_lock_irqsave(&adapter->cmd_lock, flags); 2219 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, VMXNET3_CMD_RESET_DEV); 2220 spin_unlock_irqrestore(&adapter->cmd_lock, flags); 2221} 2222 2223 2224int 2225vmxnet3_quiesce_dev(struct vmxnet3_adapter *adapter) 2226{ 2227 int i; 2228 unsigned long flags; 2229 if (test_and_set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state)) 2230 return 0; 2231 2232 2233 spin_lock_irqsave(&adapter->cmd_lock, flags); 2234 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, 2235 VMXNET3_CMD_QUIESCE_DEV); 2236 spin_unlock_irqrestore(&adapter->cmd_lock, flags); 2237 vmxnet3_disable_all_intrs(adapter); 2238 2239 for (i = 0; i < adapter->num_rx_queues; i++) 2240 napi_disable(&adapter->rx_queue[i].napi); 2241 netif_tx_disable(adapter->netdev); 2242 adapter->link_speed = 0; 2243 netif_carrier_off(adapter->netdev); 2244 2245 vmxnet3_tq_cleanup_all(adapter); 2246 vmxnet3_rq_cleanup_all(adapter); 2247 vmxnet3_free_irqs(adapter); 2248 return 0; 2249} 2250 2251 2252static void 2253vmxnet3_write_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac) 2254{ 2255 u32 tmp; 2256 2257 tmp = *(u32 *)mac; 2258 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACL, tmp); 2259 2260 tmp = (mac[5] << 8) | mac[4]; 2261 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACH, tmp); 2262} 2263 2264 2265static int 2266vmxnet3_set_mac_addr(struct net_device *netdev, void *p) 2267{ 2268 struct sockaddr *addr = p; 2269 struct vmxnet3_adapter *adapter = netdev_priv(netdev); 2270 2271 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len); 2272 vmxnet3_write_mac_addr(adapter, addr->sa_data); 2273 2274 return 0; 2275} 2276 2277 2278/* ==================== initialization and cleanup routines ============ */ 2279 2280static int 2281vmxnet3_alloc_pci_resources(struct vmxnet3_adapter *adapter, bool *dma64) 2282{ 2283 int err; 2284 unsigned long mmio_start, mmio_len; 2285 struct pci_dev *pdev = adapter->pdev; 2286 2287 err = pci_enable_device(pdev); 2288 if (err) { 2289 printk(KERN_ERR "Failed to enable adapter %s: error %d\n", 2290 pci_name(pdev), err); 2291 return err; 2292 } 2293 2294 if (pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) == 0) { 2295 if (pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64)) != 0) { 2296 printk(KERN_ERR "pci_set_consistent_dma_mask failed " 2297 "for adapter %s\n", pci_name(pdev)); 2298 err = -EIO; 2299 goto err_set_mask; 2300 } 2301 *dma64 = true; 2302 } else { 2303 if (pci_set_dma_mask(pdev, DMA_BIT_MASK(32)) != 0) { 2304 printk(KERN_ERR "pci_set_dma_mask failed for adapter " 2305 "%s\n", pci_name(pdev)); 2306 err = -EIO; 2307 goto err_set_mask; 2308 } 2309 *dma64 = false; 2310 } 2311 2312 err = pci_request_selected_regions(pdev, (1 << 2) - 1, 2313 vmxnet3_driver_name); 2314 if (err) { 2315 printk(KERN_ERR "Failed to request region for adapter %s: " 2316 "error %d\n", pci_name(pdev), err); 2317 goto err_set_mask; 2318 } 2319 2320 pci_set_master(pdev); 2321 2322 mmio_start = pci_resource_start(pdev, 0); 2323 mmio_len = pci_resource_len(pdev, 0); 2324 adapter->hw_addr0 = ioremap(mmio_start, mmio_len); 2325 if (!adapter->hw_addr0) { 2326 printk(KERN_ERR "Failed to map bar0 for adapter %s\n", 2327 pci_name(pdev)); 2328 err = -EIO; 2329 goto err_ioremap; 2330 } 2331 2332 mmio_start = pci_resource_start(pdev, 1); 2333 mmio_len = pci_resource_len(pdev, 1); 2334 adapter->hw_addr1 = ioremap(mmio_start, mmio_len); 2335 if (!adapter->hw_addr1) { 2336 printk(KERN_ERR "Failed to map bar1 for adapter %s\n", 2337 pci_name(pdev)); 2338 err = -EIO; 2339 goto err_bar1; 2340 } 2341 return 0; 2342 2343err_bar1: 2344 iounmap(adapter->hw_addr0); 2345err_ioremap: 2346 pci_release_selected_regions(pdev, (1 << 2) - 1); 2347err_set_mask: 2348 pci_disable_device(pdev); 2349 return err; 2350} 2351 2352 2353static void 2354vmxnet3_free_pci_resources(struct vmxnet3_adapter *adapter) 2355{ 2356 BUG_ON(!adapter->pdev); 2357 2358 iounmap(adapter->hw_addr0); 2359 iounmap(adapter->hw_addr1); 2360 pci_release_selected_regions(adapter->pdev, (1 << 2) - 1); 2361 pci_disable_device(adapter->pdev); 2362} 2363 2364 2365static void 2366vmxnet3_adjust_rx_ring_size(struct vmxnet3_adapter *adapter) 2367{ 2368 size_t sz, i, ring0_size, ring1_size, comp_size; 2369 struct vmxnet3_rx_queue *rq = &adapter->rx_queue[0]; 2370 2371 2372 if (adapter->netdev->mtu <= VMXNET3_MAX_SKB_BUF_SIZE - 2373 VMXNET3_MAX_ETH_HDR_SIZE) { 2374 adapter->skb_buf_size = adapter->netdev->mtu + 2375 VMXNET3_MAX_ETH_HDR_SIZE; 2376 if (adapter->skb_buf_size < VMXNET3_MIN_T0_BUF_SIZE) 2377 adapter->skb_buf_size = VMXNET3_MIN_T0_BUF_SIZE; 2378 2379 adapter->rx_buf_per_pkt = 1; 2380 } else { 2381 adapter->skb_buf_size = VMXNET3_MAX_SKB_BUF_SIZE; 2382 sz = adapter->netdev->mtu - VMXNET3_MAX_SKB_BUF_SIZE + 2383 VMXNET3_MAX_ETH_HDR_SIZE; 2384 adapter->rx_buf_per_pkt = 1 + (sz + PAGE_SIZE - 1) / PAGE_SIZE; 2385 } 2386 2387 /* 2388 * for simplicity, force the ring0 size to be a multiple of 2389 * rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN 2390 */ 2391 sz = adapter->rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN; 2392 ring0_size = adapter->rx_queue[0].rx_ring[0].size; 2393 ring0_size = (ring0_size + sz - 1) / sz * sz; 2394 ring0_size = min_t(u32, ring0_size, VMXNET3_RX_RING_MAX_SIZE / 2395 sz * sz); 2396 ring1_size = adapter->rx_queue[0].rx_ring[1].size; 2397 comp_size = ring0_size + ring1_size; 2398 2399 for (i = 0; i < adapter->num_rx_queues; i++) { 2400 rq = &adapter->rx_queue[i]; 2401 rq->rx_ring[0].size = ring0_size; 2402 rq->rx_ring[1].size = ring1_size; 2403 rq->comp_ring.size = comp_size; 2404 } 2405} 2406 2407 2408int 2409vmxnet3_create_queues(struct vmxnet3_adapter *adapter, u32 tx_ring_size, 2410 u32 rx_ring_size, u32 rx_ring2_size) 2411{ 2412 int err = 0, i; 2413 2414 for (i = 0; i < adapter->num_tx_queues; i++) { 2415 struct vmxnet3_tx_queue *tq = &adapter->tx_queue[i]; 2416 tq->tx_ring.size = tx_ring_size; 2417 tq->data_ring.size = tx_ring_size; 2418 tq->comp_ring.size = tx_ring_size; 2419 tq->shared = &adapter->tqd_start[i].ctrl; 2420 tq->stopped = true; 2421 tq->adapter = adapter; 2422 tq->qid = i; 2423 err = vmxnet3_tq_create(tq, adapter); 2424 /* 2425 * Too late to change num_tx_queues. We cannot do away with 2426 * lesser number of queues than what we asked for 2427 */ 2428 if (err) 2429 goto queue_err; 2430 } 2431 2432 adapter->rx_queue[0].rx_ring[0].size = rx_ring_size; 2433 adapter->rx_queue[0].rx_ring[1].size = rx_ring2_size; 2434 vmxnet3_adjust_rx_ring_size(adapter); 2435 for (i = 0; i < adapter->num_rx_queues; i++) { 2436 struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i]; 2437 /* qid and qid2 for rx queues will be assigned later when num 2438 * of rx queues is finalized after allocating intrs */ 2439 rq->shared = &adapter->rqd_start[i].ctrl; 2440 rq->adapter = adapter; 2441 err = vmxnet3_rq_create(rq, adapter); 2442 if (err) { 2443 if (i == 0) { 2444 printk(KERN_ERR "Could not allocate any rx" 2445 "queues. Aborting.\n"); 2446 goto queue_err; 2447 } else { 2448 printk(KERN_INFO "Number of rx queues changed " 2449 "to : %d.\n", i); 2450 adapter->num_rx_queues = i; 2451 err = 0; 2452 break; 2453 } 2454 } 2455 } 2456 return err; 2457queue_err: 2458 vmxnet3_tq_destroy_all(adapter); 2459 return err; 2460} 2461 2462static int 2463vmxnet3_open(struct net_device *netdev) 2464{ 2465 struct vmxnet3_adapter *adapter; 2466 int err, i; 2467 2468 adapter = netdev_priv(netdev); 2469 2470 for (i = 0; i < adapter->num_tx_queues; i++) 2471 spin_lock_init(&adapter->tx_queue[i].tx_lock); 2472 2473 err = vmxnet3_create_queues(adapter, VMXNET3_DEF_TX_RING_SIZE, 2474 VMXNET3_DEF_RX_RING_SIZE, 2475 VMXNET3_DEF_RX_RING_SIZE); 2476 if (err) 2477 goto queue_err; 2478 2479 err = vmxnet3_activate_dev(adapter); 2480 if (err) 2481 goto activate_err; 2482 2483 return 0; 2484 2485activate_err: 2486 vmxnet3_rq_destroy_all(adapter); 2487 vmxnet3_tq_destroy_all(adapter); 2488queue_err: 2489 return err; 2490} 2491 2492 2493static int 2494vmxnet3_close(struct net_device *netdev) 2495{ 2496 struct vmxnet3_adapter *adapter = netdev_priv(netdev); 2497 2498 /* 2499 * Reset_work may be in the middle of resetting the device, wait for its 2500 * completion. 2501 */ 2502 while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state)) 2503 msleep(1); 2504 2505 vmxnet3_quiesce_dev(adapter); 2506 2507 vmxnet3_rq_destroy_all(adapter); 2508 vmxnet3_tq_destroy_all(adapter); 2509 2510 clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state); 2511 2512 2513 return 0; 2514} 2515 2516 2517void 2518vmxnet3_force_close(struct vmxnet3_adapter *adapter) 2519{ 2520 int i; 2521 2522 /* 2523 * we must clear VMXNET3_STATE_BIT_RESETTING, otherwise 2524 * vmxnet3_close() will deadlock. 2525 */ 2526 BUG_ON(test_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state)); 2527 2528 /* we need to enable NAPI, otherwise dev_close will deadlock */ 2529 for (i = 0; i < adapter->num_rx_queues; i++) 2530 napi_enable(&adapter->rx_queue[i].napi); 2531 dev_close(adapter->netdev); 2532} 2533 2534 2535static int 2536vmxnet3_change_mtu(struct net_device *netdev, int new_mtu) 2537{ 2538 struct vmxnet3_adapter *adapter = netdev_priv(netdev); 2539 int err = 0; 2540 2541 if (new_mtu < VMXNET3_MIN_MTU || new_mtu > VMXNET3_MAX_MTU) 2542 return -EINVAL; 2543 2544 netdev->mtu = new_mtu; 2545 2546 /* 2547 * Reset_work may be in the middle of resetting the device, wait for its 2548 * completion. 2549 */ 2550 while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state)) 2551 msleep(1); 2552 2553 if (netif_running(netdev)) { 2554 vmxnet3_quiesce_dev(adapter); 2555 vmxnet3_reset_dev(adapter); 2556 2557 /* we need to re-create the rx queue based on the new mtu */ 2558 vmxnet3_rq_destroy_all(adapter); 2559 vmxnet3_adjust_rx_ring_size(adapter); 2560 err = vmxnet3_rq_create_all(adapter); 2561 if (err) { 2562 printk(KERN_ERR "%s: failed to re-create rx queues," 2563 " error %d. Closing it.\n", netdev->name, err); 2564 goto out; 2565 } 2566 2567 err = vmxnet3_activate_dev(adapter); 2568 if (err) { 2569 printk(KERN_ERR "%s: failed to re-activate, error %d. " 2570 "Closing it\n", netdev->name, err); 2571 goto out; 2572 } 2573 } 2574 2575out: 2576 clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state); 2577 if (err) 2578 vmxnet3_force_close(adapter); 2579 2580 return err; 2581} 2582 2583 2584static void 2585vmxnet3_declare_features(struct vmxnet3_adapter *adapter, bool dma64) 2586{ 2587 struct net_device *netdev = adapter->netdev; 2588 2589 netdev->hw_features = NETIF_F_SG | NETIF_F_RXCSUM | 2590 NETIF_F_HW_CSUM | NETIF_F_HW_VLAN_TX | 2591 NETIF_F_HW_VLAN_RX | NETIF_F_TSO | NETIF_F_TSO6 | 2592 NETIF_F_LRO; 2593 if (dma64) 2594 netdev->features |= NETIF_F_HIGHDMA; 2595 netdev->vlan_features = netdev->hw_features & 2596 ~(NETIF_F_HW_VLAN_TX | NETIF_F_HW_VLAN_RX); 2597 netdev->features = netdev->hw_features | NETIF_F_HW_VLAN_FILTER; 2598 2599 netdev_info(adapter->netdev, 2600 "features: sg csum vlan jf tso tsoIPv6 lro%s\n", 2601 dma64 ? " highDMA" : ""); 2602} 2603 2604 2605static void 2606vmxnet3_read_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac) 2607{ 2608 u32 tmp; 2609 2610 tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACL); 2611 *(u32 *)mac = tmp; 2612 2613 tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACH); 2614 mac[4] = tmp & 0xff; 2615 mac[5] = (tmp >> 8) & 0xff; 2616} 2617 2618#ifdef CONFIG_PCI_MSI 2619 2620/* 2621 * Enable MSIx vectors. 2622 * Returns : 2623 * 0 on successful enabling of required vectors, 2624 * VMXNET3_LINUX_MIN_MSIX_VECT when only minimum number of vectors required 2625 * could be enabled. 2626 * number of vectors which can be enabled otherwise (this number is smaller 2627 * than VMXNET3_LINUX_MIN_MSIX_VECT) 2628 */ 2629 2630static int 2631vmxnet3_acquire_msix_vectors(struct vmxnet3_adapter *adapter, 2632 int vectors) 2633{ 2634 int err = 0, vector_threshold; 2635 vector_threshold = VMXNET3_LINUX_MIN_MSIX_VECT; 2636 2637 while (vectors >= vector_threshold) { 2638 err = pci_enable_msix(adapter->pdev, adapter->intr.msix_entries, 2639 vectors); 2640 if (!err) { 2641 adapter->intr.num_intrs = vectors; 2642 return 0; 2643 } else if (err < 0) { 2644 printk(KERN_ERR "Failed to enable MSI-X for %s, error" 2645 " %d\n", adapter->netdev->name, err); 2646 vectors = 0; 2647 } else if (err < vector_threshold) { 2648 break; 2649 } else { 2650 /* If fails to enable required number of MSI-x vectors 2651 * try enabling minimum number of vectors required. 2652 */ 2653 vectors = vector_threshold; 2654 printk(KERN_ERR "Failed to enable %d MSI-X for %s, try" 2655 " %d instead\n", vectors, adapter->netdev->name, 2656 vector_threshold); 2657 } 2658 } 2659 2660 printk(KERN_INFO "Number of MSI-X interrupts which can be allocatedi" 2661 " are lower than min threshold required.\n"); 2662 return err; 2663} 2664 2665 2666#endif /* CONFIG_PCI_MSI */ 2667 2668static void 2669vmxnet3_alloc_intr_resources(struct vmxnet3_adapter *adapter) 2670{ 2671 u32 cfg; 2672 unsigned long flags; 2673 2674 /* intr settings */ 2675 spin_lock_irqsave(&adapter->cmd_lock, flags); 2676 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, 2677 VMXNET3_CMD_GET_CONF_INTR); 2678 cfg = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD); 2679 spin_unlock_irqrestore(&adapter->cmd_lock, flags); 2680 adapter->intr.type = cfg & 0x3; 2681 adapter->intr.mask_mode = (cfg >> 2) & 0x3; 2682 2683 if (adapter->intr.type == VMXNET3_IT_AUTO) { 2684 adapter->intr.type = VMXNET3_IT_MSIX; 2685 } 2686 2687#ifdef CONFIG_PCI_MSI 2688 if (adapter->intr.type == VMXNET3_IT_MSIX) { 2689 int vector, err = 0; 2690 2691 adapter->intr.num_intrs = (adapter->share_intr == 2692 VMXNET3_INTR_TXSHARE) ? 1 : 2693 adapter->num_tx_queues; 2694 adapter->intr.num_intrs += (adapter->share_intr == 2695 VMXNET3_INTR_BUDDYSHARE) ? 0 : 2696 adapter->num_rx_queues; 2697 adapter->intr.num_intrs += 1; /* for link event */ 2698 2699 adapter->intr.num_intrs = (adapter->intr.num_intrs > 2700 VMXNET3_LINUX_MIN_MSIX_VECT 2701 ? adapter->intr.num_intrs : 2702 VMXNET3_LINUX_MIN_MSIX_VECT); 2703 2704 for (vector = 0; vector < adapter->intr.num_intrs; vector++) 2705 adapter->intr.msix_entries[vector].entry = vector; 2706 2707 err = vmxnet3_acquire_msix_vectors(adapter, 2708 adapter->intr.num_intrs); 2709 /* If we cannot allocate one MSIx vector per queue 2710 * then limit the number of rx queues to 1 2711 */ 2712 if (err == VMXNET3_LINUX_MIN_MSIX_VECT) { 2713 if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE 2714 || adapter->num_rx_queues != 1) { 2715 adapter->share_intr = VMXNET3_INTR_TXSHARE; 2716 printk(KERN_ERR "Number of rx queues : 1\n"); 2717 adapter->num_rx_queues = 1; 2718 adapter->intr.num_intrs = 2719 VMXNET3_LINUX_MIN_MSIX_VECT; 2720 } 2721 return; 2722 } 2723 if (!err) 2724 return; 2725 2726 /* If we cannot allocate MSIx vectors use only one rx queue */ 2727 printk(KERN_INFO "Failed to enable MSI-X for %s, error %d." 2728 "#rx queues : 1, try MSI\n", adapter->netdev->name, err); 2729 2730 adapter->intr.type = VMXNET3_IT_MSI; 2731 } 2732 2733 if (adapter->intr.type == VMXNET3_IT_MSI) { 2734 int err; 2735 err = pci_enable_msi(adapter->pdev); 2736 if (!err) { 2737 adapter->num_rx_queues = 1; 2738 adapter->intr.num_intrs = 1; 2739 return; 2740 } 2741 } 2742#endif /* CONFIG_PCI_MSI */ 2743 2744 adapter->num_rx_queues = 1; 2745 printk(KERN_INFO "Using INTx interrupt, #Rx queues: 1.\n"); 2746 adapter->intr.type = VMXNET3_IT_INTX; 2747 2748 /* INT-X related setting */ 2749 adapter->intr.num_intrs = 1; 2750} 2751 2752 2753static void 2754vmxnet3_free_intr_resources(struct vmxnet3_adapter *adapter) 2755{ 2756 if (adapter->intr.type == VMXNET3_IT_MSIX) 2757 pci_disable_msix(adapter->pdev); 2758 else if (adapter->intr.type == VMXNET3_IT_MSI) 2759 pci_disable_msi(adapter->pdev); 2760 else 2761 BUG_ON(adapter->intr.type != VMXNET3_IT_INTX); 2762} 2763 2764 2765static void 2766vmxnet3_tx_timeout(struct net_device *netdev) 2767{ 2768 struct vmxnet3_adapter *adapter = netdev_priv(netdev); 2769 adapter->tx_timeout_count++; 2770 2771 printk(KERN_ERR "%s: tx hang\n", adapter->netdev->name); 2772 schedule_work(&adapter->work); 2773 netif_wake_queue(adapter->netdev); 2774} 2775 2776 2777static void 2778vmxnet3_reset_work(struct work_struct *data) 2779{ 2780 struct vmxnet3_adapter *adapter; 2781 2782 adapter = container_of(data, struct vmxnet3_adapter, work); 2783 2784 /* if another thread is resetting the device, no need to proceed */ 2785 if (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state)) 2786 return; 2787 2788 /* if the device is closed, we must leave it alone */ 2789 rtnl_lock(); 2790 if (netif_running(adapter->netdev)) { 2791 printk(KERN_INFO "%s: resetting\n", adapter->netdev->name); 2792 vmxnet3_quiesce_dev(adapter); 2793 vmxnet3_reset_dev(adapter); 2794 vmxnet3_activate_dev(adapter); 2795 } else { 2796 printk(KERN_INFO "%s: already closed\n", adapter->netdev->name); 2797 } 2798 rtnl_unlock(); 2799 2800 clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state); 2801} 2802 2803 2804static int __devinit 2805vmxnet3_probe_device(struct pci_dev *pdev, 2806 const struct pci_device_id *id) 2807{ 2808 static const struct net_device_ops vmxnet3_netdev_ops = { 2809 .ndo_open = vmxnet3_open, 2810 .ndo_stop = vmxnet3_close, 2811 .ndo_start_xmit = vmxnet3_xmit_frame, 2812 .ndo_set_mac_address = vmxnet3_set_mac_addr, 2813 .ndo_change_mtu = vmxnet3_change_mtu, 2814 .ndo_set_features = vmxnet3_set_features, 2815 .ndo_get_stats64 = vmxnet3_get_stats64, 2816 .ndo_tx_timeout = vmxnet3_tx_timeout, 2817 .ndo_set_multicast_list = vmxnet3_set_mc, 2818 .ndo_vlan_rx_add_vid = vmxnet3_vlan_rx_add_vid, 2819 .ndo_vlan_rx_kill_vid = vmxnet3_vlan_rx_kill_vid, 2820#ifdef CONFIG_NET_POLL_CONTROLLER 2821 .ndo_poll_controller = vmxnet3_netpoll, 2822#endif 2823 }; 2824 int err; 2825 bool dma64 = false; /* stupid gcc */ 2826 u32 ver; 2827 struct net_device *netdev; 2828 struct vmxnet3_adapter *adapter; 2829 u8 mac[ETH_ALEN]; 2830 int size; 2831 int num_tx_queues; 2832 int num_rx_queues; 2833 2834 if (!pci_msi_enabled()) 2835 enable_mq = 0; 2836 2837#ifdef VMXNET3_RSS 2838 if (enable_mq) 2839 num_rx_queues = min(VMXNET3_DEVICE_MAX_RX_QUEUES, 2840 (int)num_online_cpus()); 2841 else 2842#endif 2843 num_rx_queues = 1; 2844 2845 if (enable_mq) 2846 num_tx_queues = min(VMXNET3_DEVICE_MAX_TX_QUEUES, 2847 (int)num_online_cpus()); 2848 else 2849 num_tx_queues = 1; 2850 2851 netdev = alloc_etherdev_mq(sizeof(struct vmxnet3_adapter), 2852 max(num_tx_queues, num_rx_queues)); 2853 printk(KERN_INFO "# of Tx queues : %d, # of Rx queues : %d\n", 2854 num_tx_queues, num_rx_queues); 2855 2856 if (!netdev) { 2857 printk(KERN_ERR "Failed to alloc ethernet device for adapter " 2858 "%s\n", pci_name(pdev)); 2859 return -ENOMEM; 2860 } 2861 2862 pci_set_drvdata(pdev, netdev); 2863 adapter = netdev_priv(netdev); 2864 adapter->netdev = netdev; 2865 adapter->pdev = pdev; 2866 2867 spin_lock_init(&adapter->cmd_lock); 2868 adapter->shared = pci_alloc_consistent(adapter->pdev, 2869 sizeof(struct Vmxnet3_DriverShared), 2870 &adapter->shared_pa); 2871 if (!adapter->shared) { 2872 printk(KERN_ERR "Failed to allocate memory for %s\n", 2873 pci_name(pdev)); 2874 err = -ENOMEM; 2875 goto err_alloc_shared; 2876 } 2877 2878 adapter->num_rx_queues = num_rx_queues; 2879 adapter->num_tx_queues = num_tx_queues; 2880 2881 size = sizeof(struct Vmxnet3_TxQueueDesc) * adapter->num_tx_queues; 2882 size += sizeof(struct Vmxnet3_RxQueueDesc) * adapter->num_rx_queues; 2883 adapter->tqd_start = pci_alloc_consistent(adapter->pdev, size, 2884 &adapter->queue_desc_pa); 2885 2886 if (!adapter->tqd_start) { 2887 printk(KERN_ERR "Failed to allocate memory for %s\n", 2888 pci_name(pdev)); 2889 err = -ENOMEM; 2890 goto err_alloc_queue_desc; 2891 } 2892 adapter->rqd_start = (struct Vmxnet3_RxQueueDesc *)(adapter->tqd_start + 2893 adapter->num_tx_queues); 2894 2895 adapter->pm_conf = kmalloc(sizeof(struct Vmxnet3_PMConf), GFP_KERNEL); 2896 if (adapter->pm_conf == NULL) { 2897 printk(KERN_ERR "Failed to allocate memory for %s\n", 2898 pci_name(pdev)); 2899 err = -ENOMEM; 2900 goto err_alloc_pm; 2901 } 2902 2903#ifdef VMXNET3_RSS 2904 2905 adapter->rss_conf = kmalloc(sizeof(struct UPT1_RSSConf), GFP_KERNEL); 2906 if (adapter->rss_conf == NULL) { 2907 printk(KERN_ERR "Failed to allocate memory for %s\n", 2908 pci_name(pdev)); 2909 err = -ENOMEM; 2910 goto err_alloc_rss; 2911 } 2912#endif /* VMXNET3_RSS */ 2913 2914 err = vmxnet3_alloc_pci_resources(adapter, &dma64); 2915 if (err < 0) 2916 goto err_alloc_pci; 2917 2918 ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_VRRS); 2919 if (ver & 1) { 2920 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_VRRS, 1); 2921 } else { 2922 printk(KERN_ERR "Incompatible h/w version (0x%x) for adapter" 2923 " %s\n", ver, pci_name(pdev)); 2924 err = -EBUSY; 2925 goto err_ver; 2926 } 2927 2928 ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_UVRS); 2929 if (ver & 1) { 2930 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_UVRS, 1); 2931 } else { 2932 printk(KERN_ERR "Incompatible upt version (0x%x) for " 2933 "adapter %s\n", ver, pci_name(pdev)); 2934 err = -EBUSY; 2935 goto err_ver; 2936 } 2937 2938 vmxnet3_declare_features(adapter, dma64); 2939 2940 adapter->dev_number = atomic_read(&devices_found); 2941 2942 adapter->share_intr = irq_share_mode; 2943 if (adapter->share_intr == VMXNET3_INTR_BUDDYSHARE && 2944 adapter->num_tx_queues != adapter->num_rx_queues) 2945 adapter->share_intr = VMXNET3_INTR_DONTSHARE; 2946 2947 vmxnet3_alloc_intr_resources(adapter); 2948 2949#ifdef VMXNET3_RSS 2950 if (adapter->num_rx_queues > 1 && 2951 adapter->intr.type == VMXNET3_IT_MSIX) { 2952 adapter->rss = true; 2953 printk(KERN_INFO "RSS is enabled.\n"); 2954 } else { 2955 adapter->rss = false; 2956 } 2957#endif 2958 2959 vmxnet3_read_mac_addr(adapter, mac); 2960 memcpy(netdev->dev_addr, mac, netdev->addr_len); 2961 2962 netdev->netdev_ops = &vmxnet3_netdev_ops; 2963 vmxnet3_set_ethtool_ops(netdev); 2964 netdev->watchdog_timeo = 5 * HZ; 2965 2966 INIT_WORK(&adapter->work, vmxnet3_reset_work); 2967 2968 if (adapter->intr.type == VMXNET3_IT_MSIX) { 2969 int i; 2970 for (i = 0; i < adapter->num_rx_queues; i++) { 2971 netif_napi_add(adapter->netdev, 2972 &adapter->rx_queue[i].napi, 2973 vmxnet3_poll_rx_only, 64); 2974 } 2975 } else { 2976 netif_napi_add(adapter->netdev, &adapter->rx_queue[0].napi, 2977 vmxnet3_poll, 64); 2978 } 2979 2980 netif_set_real_num_tx_queues(adapter->netdev, adapter->num_tx_queues); 2981 netif_set_real_num_rx_queues(adapter->netdev, adapter->num_rx_queues); 2982 2983 SET_NETDEV_DEV(netdev, &pdev->dev); 2984 err = register_netdev(netdev); 2985 2986 if (err) { 2987 printk(KERN_ERR "Failed to register adapter %s\n", 2988 pci_name(pdev)); 2989 goto err_register; 2990 } 2991 2992 set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state); 2993 vmxnet3_check_link(adapter, false); 2994 atomic_inc(&devices_found); 2995 return 0; 2996 2997err_register: 2998 vmxnet3_free_intr_resources(adapter); 2999err_ver: 3000 vmxnet3_free_pci_resources(adapter); 3001err_alloc_pci: 3002#ifdef VMXNET3_RSS 3003 kfree(adapter->rss_conf); 3004err_alloc_rss: 3005#endif 3006 kfree(adapter->pm_conf); 3007err_alloc_pm: 3008 pci_free_consistent(adapter->pdev, size, adapter->tqd_start, 3009 adapter->queue_desc_pa); 3010err_alloc_queue_desc: 3011 pci_free_consistent(adapter->pdev, sizeof(struct Vmxnet3_DriverShared), 3012 adapter->shared, adapter->shared_pa); 3013err_alloc_shared: 3014 pci_set_drvdata(pdev, NULL); 3015 free_netdev(netdev); 3016 return err; 3017} 3018 3019 3020static void __devexit 3021vmxnet3_remove_device(struct pci_dev *pdev) 3022{ 3023 struct net_device *netdev = pci_get_drvdata(pdev); 3024 struct vmxnet3_adapter *adapter = netdev_priv(netdev); 3025 int size = 0; 3026 int num_rx_queues; 3027 3028#ifdef VMXNET3_RSS 3029 if (enable_mq) 3030 num_rx_queues = min(VMXNET3_DEVICE_MAX_RX_QUEUES, 3031 (int)num_online_cpus()); 3032 else 3033#endif 3034 num_rx_queues = 1; 3035 3036 cancel_work_sync(&adapter->work); 3037 3038 unregister_netdev(netdev); 3039 3040 vmxnet3_free_intr_resources(adapter); 3041 vmxnet3_free_pci_resources(adapter); 3042#ifdef VMXNET3_RSS 3043 kfree(adapter->rss_conf); 3044#endif 3045 kfree(adapter->pm_conf); 3046 3047 size = sizeof(struct Vmxnet3_TxQueueDesc) * adapter->num_tx_queues; 3048 size += sizeof(struct Vmxnet3_RxQueueDesc) * num_rx_queues; 3049 pci_free_consistent(adapter->pdev, size, adapter->tqd_start, 3050 adapter->queue_desc_pa); 3051 pci_free_consistent(adapter->pdev, sizeof(struct Vmxnet3_DriverShared), 3052 adapter->shared, adapter->shared_pa); 3053 free_netdev(netdev); 3054} 3055 3056 3057#ifdef CONFIG_PM 3058 3059static int 3060vmxnet3_suspend(struct device *device) 3061{ 3062 struct pci_dev *pdev = to_pci_dev(device); 3063 struct net_device *netdev = pci_get_drvdata(pdev); 3064 struct vmxnet3_adapter *adapter = netdev_priv(netdev); 3065 struct Vmxnet3_PMConf *pmConf; 3066 struct ethhdr *ehdr; 3067 struct arphdr *ahdr; 3068 u8 *arpreq; 3069 struct in_device *in_dev; 3070 struct in_ifaddr *ifa; 3071 unsigned long flags; 3072 int i = 0; 3073 3074 if (!netif_running(netdev)) 3075 return 0; 3076 3077 for (i = 0; i < adapter->num_rx_queues; i++) 3078 napi_disable(&adapter->rx_queue[i].napi); 3079 3080 vmxnet3_disable_all_intrs(adapter); 3081 vmxnet3_free_irqs(adapter); 3082 vmxnet3_free_intr_resources(adapter); 3083 3084 netif_device_detach(netdev); 3085 netif_tx_stop_all_queues(netdev); 3086 3087 /* Create wake-up filters. */ 3088 pmConf = adapter->pm_conf; 3089 memset(pmConf, 0, sizeof(*pmConf)); 3090 3091 if (adapter->wol & WAKE_UCAST) { 3092 pmConf->filters[i].patternSize = ETH_ALEN; 3093 pmConf->filters[i].maskSize = 1; 3094 memcpy(pmConf->filters[i].pattern, netdev->dev_addr, ETH_ALEN); 3095 pmConf->filters[i].mask[0] = 0x3F; /* LSB ETH_ALEN bits */ 3096 3097 pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_FILTER; 3098 i++; 3099 } 3100 3101 if (adapter->wol & WAKE_ARP) { 3102 in_dev = in_dev_get(netdev); 3103 if (!in_dev) 3104 goto skip_arp; 3105 3106 ifa = (struct in_ifaddr *)in_dev->ifa_list; 3107 if (!ifa) 3108 goto skip_arp; 3109 3110 pmConf->filters[i].patternSize = ETH_HLEN + /* Ethernet header*/ 3111 sizeof(struct arphdr) + /* ARP header */ 3112 2 * ETH_ALEN + /* 2 Ethernet addresses*/ 3113 2 * sizeof(u32); /*2 IPv4 addresses */ 3114 pmConf->filters[i].maskSize = 3115 (pmConf->filters[i].patternSize - 1) / 8 + 1; 3116 3117 /* ETH_P_ARP in Ethernet header. */ 3118 ehdr = (struct ethhdr *)pmConf->filters[i].pattern; 3119 ehdr->h_proto = htons(ETH_P_ARP); 3120 3121 /* ARPOP_REQUEST in ARP header. */ 3122 ahdr = (struct arphdr *)&pmConf->filters[i].pattern[ETH_HLEN]; 3123 ahdr->ar_op = htons(ARPOP_REQUEST); 3124 arpreq = (u8 *)(ahdr + 1); 3125 3126 /* The Unicast IPv4 address in 'tip' field. */ 3127 arpreq += 2 * ETH_ALEN + sizeof(u32); 3128 *(u32 *)arpreq = ifa->ifa_address; 3129 3130 /* The mask for the relevant bits. */ 3131 pmConf->filters[i].mask[0] = 0x00; 3132 pmConf->filters[i].mask[1] = 0x30; /* ETH_P_ARP */ 3133 pmConf->filters[i].mask[2] = 0x30; /* ARPOP_REQUEST */ 3134 pmConf->filters[i].mask[3] = 0x00; 3135 pmConf->filters[i].mask[4] = 0xC0; /* IPv4 TIP */ 3136 pmConf->filters[i].mask[5] = 0x03; /* IPv4 TIP */ 3137 in_dev_put(in_dev); 3138 3139 pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_FILTER; 3140 i++; 3141 } 3142 3143skip_arp: 3144 if (adapter->wol & WAKE_MAGIC) 3145 pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_MAGIC; 3146 3147 pmConf->numFilters = i; 3148 3149 adapter->shared->devRead.pmConfDesc.confVer = cpu_to_le32(1); 3150 adapter->shared->devRead.pmConfDesc.confLen = cpu_to_le32(sizeof( 3151 *pmConf)); 3152 adapter->shared->devRead.pmConfDesc.confPA = cpu_to_le64(virt_to_phys( 3153 pmConf)); 3154 3155 spin_lock_irqsave(&adapter->cmd_lock, flags); 3156 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, 3157 VMXNET3_CMD_UPDATE_PMCFG); 3158 spin_unlock_irqrestore(&adapter->cmd_lock, flags); 3159 3160 pci_save_state(pdev); 3161 pci_enable_wake(pdev, pci_choose_state(pdev, PMSG_SUSPEND), 3162 adapter->wol); 3163 pci_disable_device(pdev); 3164 pci_set_power_state(pdev, pci_choose_state(pdev, PMSG_SUSPEND)); 3165 3166 return 0; 3167} 3168 3169 3170static int 3171vmxnet3_resume(struct device *device) 3172{ 3173 int err, i = 0; 3174 unsigned long flags; 3175 struct pci_dev *pdev = to_pci_dev(device); 3176 struct net_device *netdev = pci_get_drvdata(pdev); 3177 struct vmxnet3_adapter *adapter = netdev_priv(netdev); 3178 struct Vmxnet3_PMConf *pmConf; 3179 3180 if (!netif_running(netdev)) 3181 return 0; 3182 3183 /* Destroy wake-up filters. */ 3184 pmConf = adapter->pm_conf; 3185 memset(pmConf, 0, sizeof(*pmConf)); 3186 3187 adapter->shared->devRead.pmConfDesc.confVer = cpu_to_le32(1); 3188 adapter->shared->devRead.pmConfDesc.confLen = cpu_to_le32(sizeof( 3189 *pmConf)); 3190 adapter->shared->devRead.pmConfDesc.confPA = cpu_to_le64(virt_to_phys( 3191 pmConf)); 3192 3193 netif_device_attach(netdev); 3194 pci_set_power_state(pdev, PCI_D0); 3195 pci_restore_state(pdev); 3196 err = pci_enable_device_mem(pdev); 3197 if (err != 0) 3198 return err; 3199 3200 pci_enable_wake(pdev, PCI_D0, 0); 3201 3202 spin_lock_irqsave(&adapter->cmd_lock, flags); 3203 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, 3204 VMXNET3_CMD_UPDATE_PMCFG); 3205 spin_unlock_irqrestore(&adapter->cmd_lock, flags); 3206 vmxnet3_alloc_intr_resources(adapter); 3207 vmxnet3_request_irqs(adapter); 3208 for (i = 0; i < adapter->num_rx_queues; i++) 3209 napi_enable(&adapter->rx_queue[i].napi); 3210 vmxnet3_enable_all_intrs(adapter); 3211 3212 return 0; 3213} 3214 3215static const struct dev_pm_ops vmxnet3_pm_ops = { 3216 .suspend = vmxnet3_suspend, 3217 .resume = vmxnet3_resume, 3218}; 3219#endif 3220 3221static struct pci_driver vmxnet3_driver = { 3222 .name = vmxnet3_driver_name, 3223 .id_table = vmxnet3_pciid_table, 3224 .probe = vmxnet3_probe_device, 3225 .remove = __devexit_p(vmxnet3_remove_device), 3226#ifdef CONFIG_PM 3227 .driver.pm = &vmxnet3_pm_ops, 3228#endif 3229}; 3230 3231 3232static int __init 3233vmxnet3_init_module(void) 3234{ 3235 printk(KERN_INFO "%s - version %s\n", VMXNET3_DRIVER_DESC, 3236 VMXNET3_DRIVER_VERSION_REPORT); 3237 return pci_register_driver(&vmxnet3_driver); 3238} 3239 3240module_init(vmxnet3_init_module); 3241 3242 3243static void 3244vmxnet3_exit_module(void) 3245{ 3246 pci_unregister_driver(&vmxnet3_driver); 3247} 3248 3249module_exit(vmxnet3_exit_module); 3250 3251MODULE_AUTHOR("VMware, Inc."); 3252MODULE_DESCRIPTION(VMXNET3_DRIVER_DESC); 3253MODULE_LICENSE("GPL v2"); 3254MODULE_VERSION(VMXNET3_DRIVER_VERSION_STRING); 3255