1/* 2 * Copyright (c) 2005 Ammasso, Inc. All rights reserved. 3 * Copyright (c) 2005 Open Grid Computing, Inc. All rights reserved. 4 * 5 * This software is available to you under a choice of one of two 6 * licenses. You may choose to be licensed under the terms of the GNU 7 * General Public License (GPL) Version 2, available from the file 8 * COPYING in the main directory of this source tree, or the 9 * OpenIB.org BSD license below: 10 * 11 * Redistribution and use in source and binary forms, with or 12 * without modification, are permitted provided that the following 13 * conditions are met: 14 * 15 * - Redistributions of source code must retain the above 16 * copyright notice, this list of conditions and the following 17 * disclaimer. 18 * 19 * - Redistributions in binary form must reproduce the above 20 * copyright notice, this list of conditions and the following 21 * disclaimer in the documentation and/or other materials 22 * provided with the distribution. 23 * 24 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 25 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 26 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 27 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS 28 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN 29 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN 30 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 31 * SOFTWARE. 32 */ 33#include <linux/module.h> 34#include <linux/moduleparam.h> 35#include <linux/pci.h> 36#include <linux/netdevice.h> 37#include <linux/etherdevice.h> 38#include <linux/inetdevice.h> 39#include <linux/interrupt.h> 40#include <linux/delay.h> 41#include <linux/ethtool.h> 42#include <linux/mii.h> 43#include <linux/if_vlan.h> 44#include <linux/crc32.h> 45#include <linux/in.h> 46#include <linux/ip.h> 47#include <linux/tcp.h> 48#include <linux/init.h> 49#include <linux/dma-mapping.h> 50#include <linux/slab.h> 51#include <linux/prefetch.h> 52 53#include <asm/io.h> 54#include <asm/irq.h> 55#include <asm/byteorder.h> 56 57#include <rdma/ib_smi.h> 58#include "c2.h" 59#include "c2_provider.h" 60 61MODULE_AUTHOR("Tom Tucker <tom@opengridcomputing.com>"); 62MODULE_DESCRIPTION("Ammasso AMSO1100 Low-level iWARP Driver"); 63MODULE_LICENSE("Dual BSD/GPL"); 64MODULE_VERSION(DRV_VERSION); 65 66static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_LINK 67 | NETIF_MSG_IFUP | NETIF_MSG_IFDOWN; 68 69static int debug = -1; /* defaults above */ 70module_param(debug, int, 0); 71MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)"); 72 73static int c2_up(struct net_device *netdev); 74static int c2_down(struct net_device *netdev); 75static int c2_xmit_frame(struct sk_buff *skb, struct net_device *netdev); 76static void c2_tx_interrupt(struct net_device *netdev); 77static void c2_rx_interrupt(struct net_device *netdev); 78static irqreturn_t c2_interrupt(int irq, void *dev_id); 79static void c2_tx_timeout(struct net_device *netdev); 80static int c2_change_mtu(struct net_device *netdev, int new_mtu); 81static void c2_reset(struct c2_port *c2_port); 82 83static struct pci_device_id c2_pci_table[] = { 84 { PCI_DEVICE(0x18b8, 0xb001) }, 85 { 0 } 86}; 87 88MODULE_DEVICE_TABLE(pci, c2_pci_table); 89 90static void c2_print_macaddr(struct net_device *netdev) 91{ 92 pr_debug("%s: MAC %pM, IRQ %u\n", netdev->name, netdev->dev_addr, netdev->irq); 93} 94 95static void c2_set_rxbufsize(struct c2_port *c2_port) 96{ 97 struct net_device *netdev = c2_port->netdev; 98 99 if (netdev->mtu > RX_BUF_SIZE) 100 c2_port->rx_buf_size = 101 netdev->mtu + ETH_HLEN + sizeof(struct c2_rxp_hdr) + 102 NET_IP_ALIGN; 103 else 104 c2_port->rx_buf_size = sizeof(struct c2_rxp_hdr) + RX_BUF_SIZE; 105} 106 107/* 108 * Allocate TX ring elements and chain them together. 109 * One-to-one association of adapter descriptors with ring elements. 110 */ 111static int c2_tx_ring_alloc(struct c2_ring *tx_ring, void *vaddr, 112 dma_addr_t base, void __iomem * mmio_txp_ring) 113{ 114 struct c2_tx_desc *tx_desc; 115 struct c2_txp_desc __iomem *txp_desc; 116 struct c2_element *elem; 117 int i; 118 119 tx_ring->start = kmalloc(sizeof(*elem) * tx_ring->count, GFP_KERNEL); 120 if (!tx_ring->start) 121 return -ENOMEM; 122 123 elem = tx_ring->start; 124 tx_desc = vaddr; 125 txp_desc = mmio_txp_ring; 126 for (i = 0; i < tx_ring->count; i++, elem++, tx_desc++, txp_desc++) { 127 tx_desc->len = 0; 128 tx_desc->status = 0; 129 130 /* Set TXP_HTXD_UNINIT */ 131 __raw_writeq((__force u64) cpu_to_be64(0x1122334455667788ULL), 132 (void __iomem *) txp_desc + C2_TXP_ADDR); 133 __raw_writew(0, (void __iomem *) txp_desc + C2_TXP_LEN); 134 __raw_writew((__force u16) cpu_to_be16(TXP_HTXD_UNINIT), 135 (void __iomem *) txp_desc + C2_TXP_FLAGS); 136 137 elem->skb = NULL; 138 elem->ht_desc = tx_desc; 139 elem->hw_desc = txp_desc; 140 141 if (i == tx_ring->count - 1) { 142 elem->next = tx_ring->start; 143 tx_desc->next_offset = base; 144 } else { 145 elem->next = elem + 1; 146 tx_desc->next_offset = 147 base + (i + 1) * sizeof(*tx_desc); 148 } 149 } 150 151 tx_ring->to_use = tx_ring->to_clean = tx_ring->start; 152 153 return 0; 154} 155 156/* 157 * Allocate RX ring elements and chain them together. 158 * One-to-one association of adapter descriptors with ring elements. 159 */ 160static int c2_rx_ring_alloc(struct c2_ring *rx_ring, void *vaddr, 161 dma_addr_t base, void __iomem * mmio_rxp_ring) 162{ 163 struct c2_rx_desc *rx_desc; 164 struct c2_rxp_desc __iomem *rxp_desc; 165 struct c2_element *elem; 166 int i; 167 168 rx_ring->start = kmalloc(sizeof(*elem) * rx_ring->count, GFP_KERNEL); 169 if (!rx_ring->start) 170 return -ENOMEM; 171 172 elem = rx_ring->start; 173 rx_desc = vaddr; 174 rxp_desc = mmio_rxp_ring; 175 for (i = 0; i < rx_ring->count; i++, elem++, rx_desc++, rxp_desc++) { 176 rx_desc->len = 0; 177 rx_desc->status = 0; 178 179 /* Set RXP_HRXD_UNINIT */ 180 __raw_writew((__force u16) cpu_to_be16(RXP_HRXD_OK), 181 (void __iomem *) rxp_desc + C2_RXP_STATUS); 182 __raw_writew(0, (void __iomem *) rxp_desc + C2_RXP_COUNT); 183 __raw_writew(0, (void __iomem *) rxp_desc + C2_RXP_LEN); 184 __raw_writeq((__force u64) cpu_to_be64(0x99aabbccddeeffULL), 185 (void __iomem *) rxp_desc + C2_RXP_ADDR); 186 __raw_writew((__force u16) cpu_to_be16(RXP_HRXD_UNINIT), 187 (void __iomem *) rxp_desc + C2_RXP_FLAGS); 188 189 elem->skb = NULL; 190 elem->ht_desc = rx_desc; 191 elem->hw_desc = rxp_desc; 192 193 if (i == rx_ring->count - 1) { 194 elem->next = rx_ring->start; 195 rx_desc->next_offset = base; 196 } else { 197 elem->next = elem + 1; 198 rx_desc->next_offset = 199 base + (i + 1) * sizeof(*rx_desc); 200 } 201 } 202 203 rx_ring->to_use = rx_ring->to_clean = rx_ring->start; 204 205 return 0; 206} 207 208/* Setup buffer for receiving */ 209static inline int c2_rx_alloc(struct c2_port *c2_port, struct c2_element *elem) 210{ 211 struct c2_dev *c2dev = c2_port->c2dev; 212 struct c2_rx_desc *rx_desc = elem->ht_desc; 213 struct sk_buff *skb; 214 dma_addr_t mapaddr; 215 u32 maplen; 216 struct c2_rxp_hdr *rxp_hdr; 217 218 skb = dev_alloc_skb(c2_port->rx_buf_size); 219 if (unlikely(!skb)) { 220 pr_debug("%s: out of memory for receive\n", 221 c2_port->netdev->name); 222 return -ENOMEM; 223 } 224 225 /* Zero out the rxp hdr in the sk_buff */ 226 memset(skb->data, 0, sizeof(*rxp_hdr)); 227 228 skb->dev = c2_port->netdev; 229 230 maplen = c2_port->rx_buf_size; 231 mapaddr = 232 pci_map_single(c2dev->pcidev, skb->data, maplen, 233 PCI_DMA_FROMDEVICE); 234 235 /* Set the sk_buff RXP_header to RXP_HRXD_READY */ 236 rxp_hdr = (struct c2_rxp_hdr *) skb->data; 237 rxp_hdr->flags = RXP_HRXD_READY; 238 239 __raw_writew(0, elem->hw_desc + C2_RXP_STATUS); 240 __raw_writew((__force u16) cpu_to_be16((u16) maplen - sizeof(*rxp_hdr)), 241 elem->hw_desc + C2_RXP_LEN); 242 __raw_writeq((__force u64) cpu_to_be64(mapaddr), elem->hw_desc + C2_RXP_ADDR); 243 __raw_writew((__force u16) cpu_to_be16(RXP_HRXD_READY), 244 elem->hw_desc + C2_RXP_FLAGS); 245 246 elem->skb = skb; 247 elem->mapaddr = mapaddr; 248 elem->maplen = maplen; 249 rx_desc->len = maplen; 250 251 return 0; 252} 253 254/* 255 * Allocate buffers for the Rx ring 256 * For receive: rx_ring.to_clean is next received frame 257 */ 258static int c2_rx_fill(struct c2_port *c2_port) 259{ 260 struct c2_ring *rx_ring = &c2_port->rx_ring; 261 struct c2_element *elem; 262 int ret = 0; 263 264 elem = rx_ring->start; 265 do { 266 if (c2_rx_alloc(c2_port, elem)) { 267 ret = 1; 268 break; 269 } 270 } while ((elem = elem->next) != rx_ring->start); 271 272 rx_ring->to_clean = rx_ring->start; 273 return ret; 274} 275 276/* Free all buffers in RX ring, assumes receiver stopped */ 277static void c2_rx_clean(struct c2_port *c2_port) 278{ 279 struct c2_dev *c2dev = c2_port->c2dev; 280 struct c2_ring *rx_ring = &c2_port->rx_ring; 281 struct c2_element *elem; 282 struct c2_rx_desc *rx_desc; 283 284 elem = rx_ring->start; 285 do { 286 rx_desc = elem->ht_desc; 287 rx_desc->len = 0; 288 289 __raw_writew(0, elem->hw_desc + C2_RXP_STATUS); 290 __raw_writew(0, elem->hw_desc + C2_RXP_COUNT); 291 __raw_writew(0, elem->hw_desc + C2_RXP_LEN); 292 __raw_writeq((__force u64) cpu_to_be64(0x99aabbccddeeffULL), 293 elem->hw_desc + C2_RXP_ADDR); 294 __raw_writew((__force u16) cpu_to_be16(RXP_HRXD_UNINIT), 295 elem->hw_desc + C2_RXP_FLAGS); 296 297 if (elem->skb) { 298 pci_unmap_single(c2dev->pcidev, elem->mapaddr, 299 elem->maplen, PCI_DMA_FROMDEVICE); 300 dev_kfree_skb(elem->skb); 301 elem->skb = NULL; 302 } 303 } while ((elem = elem->next) != rx_ring->start); 304} 305 306static inline int c2_tx_free(struct c2_dev *c2dev, struct c2_element *elem) 307{ 308 struct c2_tx_desc *tx_desc = elem->ht_desc; 309 310 tx_desc->len = 0; 311 312 pci_unmap_single(c2dev->pcidev, elem->mapaddr, elem->maplen, 313 PCI_DMA_TODEVICE); 314 315 if (elem->skb) { 316 dev_kfree_skb_any(elem->skb); 317 elem->skb = NULL; 318 } 319 320 return 0; 321} 322 323/* Free all buffers in TX ring, assumes transmitter stopped */ 324static void c2_tx_clean(struct c2_port *c2_port) 325{ 326 struct c2_ring *tx_ring = &c2_port->tx_ring; 327 struct c2_element *elem; 328 struct c2_txp_desc txp_htxd; 329 int retry; 330 unsigned long flags; 331 332 spin_lock_irqsave(&c2_port->tx_lock, flags); 333 334 elem = tx_ring->start; 335 336 do { 337 retry = 0; 338 do { 339 txp_htxd.flags = 340 readw(elem->hw_desc + C2_TXP_FLAGS); 341 342 if (txp_htxd.flags == TXP_HTXD_READY) { 343 retry = 1; 344 __raw_writew(0, 345 elem->hw_desc + C2_TXP_LEN); 346 __raw_writeq(0, 347 elem->hw_desc + C2_TXP_ADDR); 348 __raw_writew((__force u16) cpu_to_be16(TXP_HTXD_DONE), 349 elem->hw_desc + C2_TXP_FLAGS); 350 c2_port->netdev->stats.tx_dropped++; 351 break; 352 } else { 353 __raw_writew(0, 354 elem->hw_desc + C2_TXP_LEN); 355 __raw_writeq((__force u64) cpu_to_be64(0x1122334455667788ULL), 356 elem->hw_desc + C2_TXP_ADDR); 357 __raw_writew((__force u16) cpu_to_be16(TXP_HTXD_UNINIT), 358 elem->hw_desc + C2_TXP_FLAGS); 359 } 360 361 c2_tx_free(c2_port->c2dev, elem); 362 363 } while ((elem = elem->next) != tx_ring->start); 364 } while (retry); 365 366 c2_port->tx_avail = c2_port->tx_ring.count - 1; 367 c2_port->c2dev->cur_tx = tx_ring->to_use - tx_ring->start; 368 369 if (c2_port->tx_avail > MAX_SKB_FRAGS + 1) 370 netif_wake_queue(c2_port->netdev); 371 372 spin_unlock_irqrestore(&c2_port->tx_lock, flags); 373} 374 375/* 376 * Process transmit descriptors marked 'DONE' by the firmware, 377 * freeing up their unneeded sk_buffs. 378 */ 379static void c2_tx_interrupt(struct net_device *netdev) 380{ 381 struct c2_port *c2_port = netdev_priv(netdev); 382 struct c2_dev *c2dev = c2_port->c2dev; 383 struct c2_ring *tx_ring = &c2_port->tx_ring; 384 struct c2_element *elem; 385 struct c2_txp_desc txp_htxd; 386 387 spin_lock(&c2_port->tx_lock); 388 389 for (elem = tx_ring->to_clean; elem != tx_ring->to_use; 390 elem = elem->next) { 391 txp_htxd.flags = 392 be16_to_cpu((__force __be16) readw(elem->hw_desc + C2_TXP_FLAGS)); 393 394 if (txp_htxd.flags != TXP_HTXD_DONE) 395 break; 396 397 if (netif_msg_tx_done(c2_port)) { 398 /* PCI reads are expensive in fast path */ 399 txp_htxd.len = 400 be16_to_cpu((__force __be16) readw(elem->hw_desc + C2_TXP_LEN)); 401 pr_debug("%s: tx done slot %3Zu status 0x%x len " 402 "%5u bytes\n", 403 netdev->name, elem - tx_ring->start, 404 txp_htxd.flags, txp_htxd.len); 405 } 406 407 c2_tx_free(c2dev, elem); 408 ++(c2_port->tx_avail); 409 } 410 411 tx_ring->to_clean = elem; 412 413 if (netif_queue_stopped(netdev) 414 && c2_port->tx_avail > MAX_SKB_FRAGS + 1) 415 netif_wake_queue(netdev); 416 417 spin_unlock(&c2_port->tx_lock); 418} 419 420static void c2_rx_error(struct c2_port *c2_port, struct c2_element *elem) 421{ 422 struct c2_rx_desc *rx_desc = elem->ht_desc; 423 struct c2_rxp_hdr *rxp_hdr = (struct c2_rxp_hdr *) elem->skb->data; 424 425 if (rxp_hdr->status != RXP_HRXD_OK || 426 rxp_hdr->len > (rx_desc->len - sizeof(*rxp_hdr))) { 427 pr_debug("BAD RXP_HRXD\n"); 428 pr_debug(" rx_desc : %p\n", rx_desc); 429 pr_debug(" index : %Zu\n", 430 elem - c2_port->rx_ring.start); 431 pr_debug(" len : %u\n", rx_desc->len); 432 pr_debug(" rxp_hdr : %p [PA %p]\n", rxp_hdr, 433 (void *) __pa((unsigned long) rxp_hdr)); 434 pr_debug(" flags : 0x%x\n", rxp_hdr->flags); 435 pr_debug(" status: 0x%x\n", rxp_hdr->status); 436 pr_debug(" len : %u\n", rxp_hdr->len); 437 pr_debug(" rsvd : 0x%x\n", rxp_hdr->rsvd); 438 } 439 440 /* Setup the skb for reuse since we're dropping this pkt */ 441 elem->skb->data = elem->skb->head; 442 skb_reset_tail_pointer(elem->skb); 443 444 /* Zero out the rxp hdr in the sk_buff */ 445 memset(elem->skb->data, 0, sizeof(*rxp_hdr)); 446 447 /* Write the descriptor to the adapter's rx ring */ 448 __raw_writew(0, elem->hw_desc + C2_RXP_STATUS); 449 __raw_writew(0, elem->hw_desc + C2_RXP_COUNT); 450 __raw_writew((__force u16) cpu_to_be16((u16) elem->maplen - sizeof(*rxp_hdr)), 451 elem->hw_desc + C2_RXP_LEN); 452 __raw_writeq((__force u64) cpu_to_be64(elem->mapaddr), 453 elem->hw_desc + C2_RXP_ADDR); 454 __raw_writew((__force u16) cpu_to_be16(RXP_HRXD_READY), 455 elem->hw_desc + C2_RXP_FLAGS); 456 457 pr_debug("packet dropped\n"); 458 c2_port->netdev->stats.rx_dropped++; 459} 460 461static void c2_rx_interrupt(struct net_device *netdev) 462{ 463 struct c2_port *c2_port = netdev_priv(netdev); 464 struct c2_dev *c2dev = c2_port->c2dev; 465 struct c2_ring *rx_ring = &c2_port->rx_ring; 466 struct c2_element *elem; 467 struct c2_rx_desc *rx_desc; 468 struct c2_rxp_hdr *rxp_hdr; 469 struct sk_buff *skb; 470 dma_addr_t mapaddr; 471 u32 maplen, buflen; 472 unsigned long flags; 473 474 spin_lock_irqsave(&c2dev->lock, flags); 475 476 /* Begin where we left off */ 477 rx_ring->to_clean = rx_ring->start + c2dev->cur_rx; 478 479 for (elem = rx_ring->to_clean; elem->next != rx_ring->to_clean; 480 elem = elem->next) { 481 rx_desc = elem->ht_desc; 482 mapaddr = elem->mapaddr; 483 maplen = elem->maplen; 484 skb = elem->skb; 485 rxp_hdr = (struct c2_rxp_hdr *) skb->data; 486 487 if (rxp_hdr->flags != RXP_HRXD_DONE) 488 break; 489 buflen = rxp_hdr->len; 490 491 /* Sanity check the RXP header */ 492 if (rxp_hdr->status != RXP_HRXD_OK || 493 buflen > (rx_desc->len - sizeof(*rxp_hdr))) { 494 c2_rx_error(c2_port, elem); 495 continue; 496 } 497 498 /* 499 * Allocate and map a new skb for replenishing the host 500 * RX desc 501 */ 502 if (c2_rx_alloc(c2_port, elem)) { 503 c2_rx_error(c2_port, elem); 504 continue; 505 } 506 507 /* Unmap the old skb */ 508 pci_unmap_single(c2dev->pcidev, mapaddr, maplen, 509 PCI_DMA_FROMDEVICE); 510 511 prefetch(skb->data); 512 513 /* 514 * Skip past the leading 8 bytes comprising of the 515 * "struct c2_rxp_hdr", prepended by the adapter 516 * to the usual Ethernet header ("struct ethhdr"), 517 * to the start of the raw Ethernet packet. 518 * 519 * Fix up the various fields in the sk_buff before 520 * passing it up to netif_rx(). The transfer size 521 * (in bytes) specified by the adapter len field of 522 * the "struct rxp_hdr_t" does NOT include the 523 * "sizeof(struct c2_rxp_hdr)". 524 */ 525 skb->data += sizeof(*rxp_hdr); 526 skb_set_tail_pointer(skb, buflen); 527 skb->len = buflen; 528 skb->protocol = eth_type_trans(skb, netdev); 529 530 netif_rx(skb); 531 532 netdev->stats.rx_packets++; 533 netdev->stats.rx_bytes += buflen; 534 } 535 536 /* Save where we left off */ 537 rx_ring->to_clean = elem; 538 c2dev->cur_rx = elem - rx_ring->start; 539 C2_SET_CUR_RX(c2dev, c2dev->cur_rx); 540 541 spin_unlock_irqrestore(&c2dev->lock, flags); 542} 543 544/* 545 * Handle netisr0 TX & RX interrupts. 546 */ 547static irqreturn_t c2_interrupt(int irq, void *dev_id) 548{ 549 unsigned int netisr0, dmaisr; 550 int handled = 0; 551 struct c2_dev *c2dev = (struct c2_dev *) dev_id; 552 553 /* Process CCILNET interrupts */ 554 netisr0 = readl(c2dev->regs + C2_NISR0); 555 if (netisr0) { 556 557 /* 558 * There is an issue with the firmware that always 559 * provides the status of RX for both TX & RX 560 * interrupts. So process both queues here. 561 */ 562 c2_rx_interrupt(c2dev->netdev); 563 c2_tx_interrupt(c2dev->netdev); 564 565 /* Clear the interrupt */ 566 writel(netisr0, c2dev->regs + C2_NISR0); 567 handled++; 568 } 569 570 /* Process RNIC interrupts */ 571 dmaisr = readl(c2dev->regs + C2_DISR); 572 if (dmaisr) { 573 writel(dmaisr, c2dev->regs + C2_DISR); 574 c2_rnic_interrupt(c2dev); 575 handled++; 576 } 577 578 if (handled) { 579 return IRQ_HANDLED; 580 } else { 581 return IRQ_NONE; 582 } 583} 584 585static int c2_up(struct net_device *netdev) 586{ 587 struct c2_port *c2_port = netdev_priv(netdev); 588 struct c2_dev *c2dev = c2_port->c2dev; 589 struct c2_element *elem; 590 struct c2_rxp_hdr *rxp_hdr; 591 struct in_device *in_dev; 592 size_t rx_size, tx_size; 593 int ret, i; 594 unsigned int netimr0; 595 596 if (netif_msg_ifup(c2_port)) 597 pr_debug("%s: enabling interface\n", netdev->name); 598 599 /* Set the Rx buffer size based on MTU */ 600 c2_set_rxbufsize(c2_port); 601 602 /* Allocate DMA'able memory for Tx/Rx host descriptor rings */ 603 rx_size = c2_port->rx_ring.count * sizeof(struct c2_rx_desc); 604 tx_size = c2_port->tx_ring.count * sizeof(struct c2_tx_desc); 605 606 c2_port->mem_size = tx_size + rx_size; 607 c2_port->mem = pci_zalloc_consistent(c2dev->pcidev, c2_port->mem_size, 608 &c2_port->dma); 609 if (c2_port->mem == NULL) { 610 pr_debug("Unable to allocate memory for " 611 "host descriptor rings\n"); 612 return -ENOMEM; 613 } 614 615 /* Create the Rx host descriptor ring */ 616 if ((ret = 617 c2_rx_ring_alloc(&c2_port->rx_ring, c2_port->mem, c2_port->dma, 618 c2dev->mmio_rxp_ring))) { 619 pr_debug("Unable to create RX ring\n"); 620 goto bail0; 621 } 622 623 /* Allocate Rx buffers for the host descriptor ring */ 624 if (c2_rx_fill(c2_port)) { 625 pr_debug("Unable to fill RX ring\n"); 626 goto bail1; 627 } 628 629 /* Create the Tx host descriptor ring */ 630 if ((ret = c2_tx_ring_alloc(&c2_port->tx_ring, c2_port->mem + rx_size, 631 c2_port->dma + rx_size, 632 c2dev->mmio_txp_ring))) { 633 pr_debug("Unable to create TX ring\n"); 634 goto bail1; 635 } 636 637 /* Set the TX pointer to where we left off */ 638 c2_port->tx_avail = c2_port->tx_ring.count - 1; 639 c2_port->tx_ring.to_use = c2_port->tx_ring.to_clean = 640 c2_port->tx_ring.start + c2dev->cur_tx; 641 642 /* missing: Initialize MAC */ 643 644 BUG_ON(c2_port->tx_ring.to_use != c2_port->tx_ring.to_clean); 645 646 /* Reset the adapter, ensures the driver is in sync with the RXP */ 647 c2_reset(c2_port); 648 649 /* Reset the READY bit in the sk_buff RXP headers & adapter HRXDQ */ 650 for (i = 0, elem = c2_port->rx_ring.start; i < c2_port->rx_ring.count; 651 i++, elem++) { 652 rxp_hdr = (struct c2_rxp_hdr *) elem->skb->data; 653 rxp_hdr->flags = 0; 654 __raw_writew((__force u16) cpu_to_be16(RXP_HRXD_READY), 655 elem->hw_desc + C2_RXP_FLAGS); 656 } 657 658 /* Enable network packets */ 659 netif_start_queue(netdev); 660 661 /* Enable IRQ */ 662 writel(0, c2dev->regs + C2_IDIS); 663 netimr0 = readl(c2dev->regs + C2_NIMR0); 664 netimr0 &= ~(C2_PCI_HTX_INT | C2_PCI_HRX_INT); 665 writel(netimr0, c2dev->regs + C2_NIMR0); 666 667 /* Tell the stack to ignore arp requests for ipaddrs bound to 668 * other interfaces. This is needed to prevent the host stack 669 * from responding to arp requests to the ipaddr bound on the 670 * rdma interface. 671 */ 672 in_dev = in_dev_get(netdev); 673 IN_DEV_CONF_SET(in_dev, ARP_IGNORE, 1); 674 in_dev_put(in_dev); 675 676 return 0; 677 678 bail1: 679 c2_rx_clean(c2_port); 680 kfree(c2_port->rx_ring.start); 681 682 bail0: 683 pci_free_consistent(c2dev->pcidev, c2_port->mem_size, c2_port->mem, 684 c2_port->dma); 685 686 return ret; 687} 688 689static int c2_down(struct net_device *netdev) 690{ 691 struct c2_port *c2_port = netdev_priv(netdev); 692 struct c2_dev *c2dev = c2_port->c2dev; 693 694 if (netif_msg_ifdown(c2_port)) 695 pr_debug("%s: disabling interface\n", 696 netdev->name); 697 698 /* Wait for all the queued packets to get sent */ 699 c2_tx_interrupt(netdev); 700 701 /* Disable network packets */ 702 netif_stop_queue(netdev); 703 704 /* Disable IRQs by clearing the interrupt mask */ 705 writel(1, c2dev->regs + C2_IDIS); 706 writel(0, c2dev->regs + C2_NIMR0); 707 708 /* missing: Stop transmitter */ 709 710 /* missing: Stop receiver */ 711 712 /* Reset the adapter, ensures the driver is in sync with the RXP */ 713 c2_reset(c2_port); 714 715 /* missing: Turn off LEDs here */ 716 717 /* Free all buffers in the host descriptor rings */ 718 c2_tx_clean(c2_port); 719 c2_rx_clean(c2_port); 720 721 /* Free the host descriptor rings */ 722 kfree(c2_port->rx_ring.start); 723 kfree(c2_port->tx_ring.start); 724 pci_free_consistent(c2dev->pcidev, c2_port->mem_size, c2_port->mem, 725 c2_port->dma); 726 727 return 0; 728} 729 730static void c2_reset(struct c2_port *c2_port) 731{ 732 struct c2_dev *c2dev = c2_port->c2dev; 733 unsigned int cur_rx = c2dev->cur_rx; 734 735 /* Tell the hardware to quiesce */ 736 C2_SET_CUR_RX(c2dev, cur_rx | C2_PCI_HRX_QUI); 737 738 /* 739 * The hardware will reset the C2_PCI_HRX_QUI bit once 740 * the RXP is quiesced. Wait 2 seconds for this. 741 */ 742 ssleep(2); 743 744 cur_rx = C2_GET_CUR_RX(c2dev); 745 746 if (cur_rx & C2_PCI_HRX_QUI) 747 pr_debug("c2_reset: failed to quiesce the hardware!\n"); 748 749 cur_rx &= ~C2_PCI_HRX_QUI; 750 751 c2dev->cur_rx = cur_rx; 752 753 pr_debug("Current RX: %u\n", c2dev->cur_rx); 754} 755 756static int c2_xmit_frame(struct sk_buff *skb, struct net_device *netdev) 757{ 758 struct c2_port *c2_port = netdev_priv(netdev); 759 struct c2_dev *c2dev = c2_port->c2dev; 760 struct c2_ring *tx_ring = &c2_port->tx_ring; 761 struct c2_element *elem; 762 dma_addr_t mapaddr; 763 u32 maplen; 764 unsigned long flags; 765 unsigned int i; 766 767 spin_lock_irqsave(&c2_port->tx_lock, flags); 768 769 if (unlikely(c2_port->tx_avail < (skb_shinfo(skb)->nr_frags + 1))) { 770 netif_stop_queue(netdev); 771 spin_unlock_irqrestore(&c2_port->tx_lock, flags); 772 773 pr_debug("%s: Tx ring full when queue awake!\n", 774 netdev->name); 775 return NETDEV_TX_BUSY; 776 } 777 778 maplen = skb_headlen(skb); 779 mapaddr = 780 pci_map_single(c2dev->pcidev, skb->data, maplen, PCI_DMA_TODEVICE); 781 782 elem = tx_ring->to_use; 783 elem->skb = skb; 784 elem->mapaddr = mapaddr; 785 elem->maplen = maplen; 786 787 /* Tell HW to xmit */ 788 __raw_writeq((__force u64) cpu_to_be64(mapaddr), 789 elem->hw_desc + C2_TXP_ADDR); 790 __raw_writew((__force u16) cpu_to_be16(maplen), 791 elem->hw_desc + C2_TXP_LEN); 792 __raw_writew((__force u16) cpu_to_be16(TXP_HTXD_READY), 793 elem->hw_desc + C2_TXP_FLAGS); 794 795 netdev->stats.tx_packets++; 796 netdev->stats.tx_bytes += maplen; 797 798 /* Loop thru additional data fragments and queue them */ 799 if (skb_shinfo(skb)->nr_frags) { 800 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 801 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 802 maplen = skb_frag_size(frag); 803 mapaddr = skb_frag_dma_map(&c2dev->pcidev->dev, frag, 804 0, maplen, DMA_TO_DEVICE); 805 elem = elem->next; 806 elem->skb = NULL; 807 elem->mapaddr = mapaddr; 808 elem->maplen = maplen; 809 810 /* Tell HW to xmit */ 811 __raw_writeq((__force u64) cpu_to_be64(mapaddr), 812 elem->hw_desc + C2_TXP_ADDR); 813 __raw_writew((__force u16) cpu_to_be16(maplen), 814 elem->hw_desc + C2_TXP_LEN); 815 __raw_writew((__force u16) cpu_to_be16(TXP_HTXD_READY), 816 elem->hw_desc + C2_TXP_FLAGS); 817 818 netdev->stats.tx_packets++; 819 netdev->stats.tx_bytes += maplen; 820 } 821 } 822 823 tx_ring->to_use = elem->next; 824 c2_port->tx_avail -= (skb_shinfo(skb)->nr_frags + 1); 825 826 if (c2_port->tx_avail <= MAX_SKB_FRAGS + 1) { 827 netif_stop_queue(netdev); 828 if (netif_msg_tx_queued(c2_port)) 829 pr_debug("%s: transmit queue full\n", 830 netdev->name); 831 } 832 833 spin_unlock_irqrestore(&c2_port->tx_lock, flags); 834 835 netdev->trans_start = jiffies; 836 837 return NETDEV_TX_OK; 838} 839 840static void c2_tx_timeout(struct net_device *netdev) 841{ 842 struct c2_port *c2_port = netdev_priv(netdev); 843 844 if (netif_msg_timer(c2_port)) 845 pr_debug("%s: tx timeout\n", netdev->name); 846 847 c2_tx_clean(c2_port); 848} 849 850static int c2_change_mtu(struct net_device *netdev, int new_mtu) 851{ 852 int ret = 0; 853 854 if (new_mtu < ETH_ZLEN || new_mtu > ETH_JUMBO_MTU) 855 return -EINVAL; 856 857 netdev->mtu = new_mtu; 858 859 if (netif_running(netdev)) { 860 c2_down(netdev); 861 862 c2_up(netdev); 863 } 864 865 return ret; 866} 867 868static const struct net_device_ops c2_netdev = { 869 .ndo_open = c2_up, 870 .ndo_stop = c2_down, 871 .ndo_start_xmit = c2_xmit_frame, 872 .ndo_tx_timeout = c2_tx_timeout, 873 .ndo_change_mtu = c2_change_mtu, 874 .ndo_set_mac_address = eth_mac_addr, 875 .ndo_validate_addr = eth_validate_addr, 876}; 877 878/* Initialize network device */ 879static struct net_device *c2_devinit(struct c2_dev *c2dev, 880 void __iomem * mmio_addr) 881{ 882 struct c2_port *c2_port = NULL; 883 struct net_device *netdev = alloc_etherdev(sizeof(*c2_port)); 884 885 if (!netdev) { 886 pr_debug("c2_port etherdev alloc failed"); 887 return NULL; 888 } 889 890 SET_NETDEV_DEV(netdev, &c2dev->pcidev->dev); 891 892 netdev->netdev_ops = &c2_netdev; 893 netdev->watchdog_timeo = C2_TX_TIMEOUT; 894 netdev->irq = c2dev->pcidev->irq; 895 896 c2_port = netdev_priv(netdev); 897 c2_port->netdev = netdev; 898 c2_port->c2dev = c2dev; 899 c2_port->msg_enable = netif_msg_init(debug, default_msg); 900 c2_port->tx_ring.count = C2_NUM_TX_DESC; 901 c2_port->rx_ring.count = C2_NUM_RX_DESC; 902 903 spin_lock_init(&c2_port->tx_lock); 904 905 /* Copy our 48-bit ethernet hardware address */ 906 memcpy_fromio(netdev->dev_addr, mmio_addr + C2_REGS_ENADDR, 6); 907 908 /* Validate the MAC address */ 909 if (!is_valid_ether_addr(netdev->dev_addr)) { 910 pr_debug("Invalid MAC Address\n"); 911 c2_print_macaddr(netdev); 912 free_netdev(netdev); 913 return NULL; 914 } 915 916 c2dev->netdev = netdev; 917 918 return netdev; 919} 920 921static int c2_probe(struct pci_dev *pcidev, const struct pci_device_id *ent) 922{ 923 int ret = 0, i; 924 unsigned long reg0_start, reg0_flags, reg0_len; 925 unsigned long reg2_start, reg2_flags, reg2_len; 926 unsigned long reg4_start, reg4_flags, reg4_len; 927 unsigned kva_map_size; 928 struct net_device *netdev = NULL; 929 struct c2_dev *c2dev = NULL; 930 void __iomem *mmio_regs = NULL; 931 932 printk(KERN_INFO PFX "AMSO1100 Gigabit Ethernet driver v%s loaded\n", 933 DRV_VERSION); 934 935 /* Enable PCI device */ 936 ret = pci_enable_device(pcidev); 937 if (ret) { 938 printk(KERN_ERR PFX "%s: Unable to enable PCI device\n", 939 pci_name(pcidev)); 940 goto bail0; 941 } 942 943 reg0_start = pci_resource_start(pcidev, BAR_0); 944 reg0_len = pci_resource_len(pcidev, BAR_0); 945 reg0_flags = pci_resource_flags(pcidev, BAR_0); 946 947 reg2_start = pci_resource_start(pcidev, BAR_2); 948 reg2_len = pci_resource_len(pcidev, BAR_2); 949 reg2_flags = pci_resource_flags(pcidev, BAR_2); 950 951 reg4_start = pci_resource_start(pcidev, BAR_4); 952 reg4_len = pci_resource_len(pcidev, BAR_4); 953 reg4_flags = pci_resource_flags(pcidev, BAR_4); 954 955 pr_debug("BAR0 size = 0x%lX bytes\n", reg0_len); 956 pr_debug("BAR2 size = 0x%lX bytes\n", reg2_len); 957 pr_debug("BAR4 size = 0x%lX bytes\n", reg4_len); 958 959 /* Make sure PCI base addr are MMIO */ 960 if (!(reg0_flags & IORESOURCE_MEM) || 961 !(reg2_flags & IORESOURCE_MEM) || !(reg4_flags & IORESOURCE_MEM)) { 962 printk(KERN_ERR PFX "PCI regions not an MMIO resource\n"); 963 ret = -ENODEV; 964 goto bail1; 965 } 966 967 /* Check for weird/broken PCI region reporting */ 968 if ((reg0_len < C2_REG0_SIZE) || 969 (reg2_len < C2_REG2_SIZE) || (reg4_len < C2_REG4_SIZE)) { 970 printk(KERN_ERR PFX "Invalid PCI region sizes\n"); 971 ret = -ENODEV; 972 goto bail1; 973 } 974 975 /* Reserve PCI I/O and memory resources */ 976 ret = pci_request_regions(pcidev, DRV_NAME); 977 if (ret) { 978 printk(KERN_ERR PFX "%s: Unable to request regions\n", 979 pci_name(pcidev)); 980 goto bail1; 981 } 982 983 if ((sizeof(dma_addr_t) > 4)) { 984 ret = pci_set_dma_mask(pcidev, DMA_BIT_MASK(64)); 985 if (ret < 0) { 986 printk(KERN_ERR PFX "64b DMA configuration failed\n"); 987 goto bail2; 988 } 989 } else { 990 ret = pci_set_dma_mask(pcidev, DMA_BIT_MASK(32)); 991 if (ret < 0) { 992 printk(KERN_ERR PFX "32b DMA configuration failed\n"); 993 goto bail2; 994 } 995 } 996 997 /* Enables bus-mastering on the device */ 998 pci_set_master(pcidev); 999 1000 /* Remap the adapter PCI registers in BAR4 */ 1001 mmio_regs = ioremap_nocache(reg4_start + C2_PCI_REGS_OFFSET, 1002 sizeof(struct c2_adapter_pci_regs)); 1003 if (!mmio_regs) { 1004 printk(KERN_ERR PFX 1005 "Unable to remap adapter PCI registers in BAR4\n"); 1006 ret = -EIO; 1007 goto bail2; 1008 } 1009 1010 /* Validate PCI regs magic */ 1011 for (i = 0; i < sizeof(c2_magic); i++) { 1012 if (c2_magic[i] != readb(mmio_regs + C2_REGS_MAGIC + i)) { 1013 printk(KERN_ERR PFX "Downlevel Firmware boot loader " 1014 "[%d/%Zd: got 0x%x, exp 0x%x]. Use the cc_flash " 1015 "utility to update your boot loader\n", 1016 i + 1, sizeof(c2_magic), 1017 readb(mmio_regs + C2_REGS_MAGIC + i), 1018 c2_magic[i]); 1019 printk(KERN_ERR PFX "Adapter not claimed\n"); 1020 iounmap(mmio_regs); 1021 ret = -EIO; 1022 goto bail2; 1023 } 1024 } 1025 1026 /* Validate the adapter version */ 1027 if (be32_to_cpu((__force __be32) readl(mmio_regs + C2_REGS_VERS)) != C2_VERSION) { 1028 printk(KERN_ERR PFX "Version mismatch " 1029 "[fw=%u, c2=%u], Adapter not claimed\n", 1030 be32_to_cpu((__force __be32) readl(mmio_regs + C2_REGS_VERS)), 1031 C2_VERSION); 1032 ret = -EINVAL; 1033 iounmap(mmio_regs); 1034 goto bail2; 1035 } 1036 1037 /* Validate the adapter IVN */ 1038 if (be32_to_cpu((__force __be32) readl(mmio_regs + C2_REGS_IVN)) != C2_IVN) { 1039 printk(KERN_ERR PFX "Downlevel FIrmware level. You should be using " 1040 "the OpenIB device support kit. " 1041 "[fw=0x%x, c2=0x%x], Adapter not claimed\n", 1042 be32_to_cpu((__force __be32) readl(mmio_regs + C2_REGS_IVN)), 1043 C2_IVN); 1044 ret = -EINVAL; 1045 iounmap(mmio_regs); 1046 goto bail2; 1047 } 1048 1049 /* Allocate hardware structure */ 1050 c2dev = (struct c2_dev *) ib_alloc_device(sizeof(*c2dev)); 1051 if (!c2dev) { 1052 printk(KERN_ERR PFX "%s: Unable to alloc hardware struct\n", 1053 pci_name(pcidev)); 1054 ret = -ENOMEM; 1055 iounmap(mmio_regs); 1056 goto bail2; 1057 } 1058 1059 memset(c2dev, 0, sizeof(*c2dev)); 1060 spin_lock_init(&c2dev->lock); 1061 c2dev->pcidev = pcidev; 1062 c2dev->cur_tx = 0; 1063 1064 /* Get the last RX index */ 1065 c2dev->cur_rx = 1066 (be32_to_cpu((__force __be32) readl(mmio_regs + C2_REGS_HRX_CUR)) - 1067 0xffffc000) / sizeof(struct c2_rxp_desc); 1068 1069 /* Request an interrupt line for the driver */ 1070 ret = request_irq(pcidev->irq, c2_interrupt, IRQF_SHARED, DRV_NAME, c2dev); 1071 if (ret) { 1072 printk(KERN_ERR PFX "%s: requested IRQ %u is busy\n", 1073 pci_name(pcidev), pcidev->irq); 1074 iounmap(mmio_regs); 1075 goto bail3; 1076 } 1077 1078 /* Set driver specific data */ 1079 pci_set_drvdata(pcidev, c2dev); 1080 1081 /* Initialize network device */ 1082 if ((netdev = c2_devinit(c2dev, mmio_regs)) == NULL) { 1083 ret = -ENOMEM; 1084 iounmap(mmio_regs); 1085 goto bail4; 1086 } 1087 1088 /* Save off the actual size prior to unmapping mmio_regs */ 1089 kva_map_size = be32_to_cpu((__force __be32) readl(mmio_regs + C2_REGS_PCI_WINSIZE)); 1090 1091 /* Unmap the adapter PCI registers in BAR4 */ 1092 iounmap(mmio_regs); 1093 1094 /* Register network device */ 1095 ret = register_netdev(netdev); 1096 if (ret) { 1097 printk(KERN_ERR PFX "Unable to register netdev, ret = %d\n", 1098 ret); 1099 goto bail5; 1100 } 1101 1102 /* Disable network packets */ 1103 netif_stop_queue(netdev); 1104 1105 /* Remap the adapter HRXDQ PA space to kernel VA space */ 1106 c2dev->mmio_rxp_ring = ioremap_nocache(reg4_start + C2_RXP_HRXDQ_OFFSET, 1107 C2_RXP_HRXDQ_SIZE); 1108 if (!c2dev->mmio_rxp_ring) { 1109 printk(KERN_ERR PFX "Unable to remap MMIO HRXDQ region\n"); 1110 ret = -EIO; 1111 goto bail6; 1112 } 1113 1114 /* Remap the adapter HTXDQ PA space to kernel VA space */ 1115 c2dev->mmio_txp_ring = ioremap_nocache(reg4_start + C2_TXP_HTXDQ_OFFSET, 1116 C2_TXP_HTXDQ_SIZE); 1117 if (!c2dev->mmio_txp_ring) { 1118 printk(KERN_ERR PFX "Unable to remap MMIO HTXDQ region\n"); 1119 ret = -EIO; 1120 goto bail7; 1121 } 1122 1123 /* Save off the current RX index in the last 4 bytes of the TXP Ring */ 1124 C2_SET_CUR_RX(c2dev, c2dev->cur_rx); 1125 1126 /* Remap the PCI registers in adapter BAR0 to kernel VA space */ 1127 c2dev->regs = ioremap_nocache(reg0_start, reg0_len); 1128 if (!c2dev->regs) { 1129 printk(KERN_ERR PFX "Unable to remap BAR0\n"); 1130 ret = -EIO; 1131 goto bail8; 1132 } 1133 1134 /* Remap the PCI registers in adapter BAR4 to kernel VA space */ 1135 c2dev->pa = reg4_start + C2_PCI_REGS_OFFSET; 1136 c2dev->kva = ioremap_nocache(reg4_start + C2_PCI_REGS_OFFSET, 1137 kva_map_size); 1138 if (!c2dev->kva) { 1139 printk(KERN_ERR PFX "Unable to remap BAR4\n"); 1140 ret = -EIO; 1141 goto bail9; 1142 } 1143 1144 /* Print out the MAC address */ 1145 c2_print_macaddr(netdev); 1146 1147 ret = c2_rnic_init(c2dev); 1148 if (ret) { 1149 printk(KERN_ERR PFX "c2_rnic_init failed: %d\n", ret); 1150 goto bail10; 1151 } 1152 1153 ret = c2_register_device(c2dev); 1154 if (ret) 1155 goto bail10; 1156 1157 return 0; 1158 1159 bail10: 1160 iounmap(c2dev->kva); 1161 1162 bail9: 1163 iounmap(c2dev->regs); 1164 1165 bail8: 1166 iounmap(c2dev->mmio_txp_ring); 1167 1168 bail7: 1169 iounmap(c2dev->mmio_rxp_ring); 1170 1171 bail6: 1172 unregister_netdev(netdev); 1173 1174 bail5: 1175 free_netdev(netdev); 1176 1177 bail4: 1178 free_irq(pcidev->irq, c2dev); 1179 1180 bail3: 1181 ib_dealloc_device(&c2dev->ibdev); 1182 1183 bail2: 1184 pci_release_regions(pcidev); 1185 1186 bail1: 1187 pci_disable_device(pcidev); 1188 1189 bail0: 1190 return ret; 1191} 1192 1193static void c2_remove(struct pci_dev *pcidev) 1194{ 1195 struct c2_dev *c2dev = pci_get_drvdata(pcidev); 1196 struct net_device *netdev = c2dev->netdev; 1197 1198 /* Unregister with OpenIB */ 1199 c2_unregister_device(c2dev); 1200 1201 /* Clean up the RNIC resources */ 1202 c2_rnic_term(c2dev); 1203 1204 /* Remove network device from the kernel */ 1205 unregister_netdev(netdev); 1206 1207 /* Free network device */ 1208 free_netdev(netdev); 1209 1210 /* Free the interrupt line */ 1211 free_irq(pcidev->irq, c2dev); 1212 1213 /* missing: Turn LEDs off here */ 1214 1215 /* Unmap adapter PA space */ 1216 iounmap(c2dev->kva); 1217 iounmap(c2dev->regs); 1218 iounmap(c2dev->mmio_txp_ring); 1219 iounmap(c2dev->mmio_rxp_ring); 1220 1221 /* Free the hardware structure */ 1222 ib_dealloc_device(&c2dev->ibdev); 1223 1224 /* Release reserved PCI I/O and memory resources */ 1225 pci_release_regions(pcidev); 1226 1227 /* Disable PCI device */ 1228 pci_disable_device(pcidev); 1229 1230 /* Clear driver specific data */ 1231 pci_set_drvdata(pcidev, NULL); 1232} 1233 1234static struct pci_driver c2_pci_driver = { 1235 .name = DRV_NAME, 1236 .id_table = c2_pci_table, 1237 .probe = c2_probe, 1238 .remove = c2_remove, 1239}; 1240 1241module_pci_driver(c2_pci_driver); 1242