1 /* bnx2i.c: Broadcom NetXtreme II iSCSI driver. 2 * 3 * Copyright (c) 2006 - 2013 Broadcom Corporation 4 * Copyright (c) 2007, 2008 Red Hat, Inc. All rights reserved. 5 * Copyright (c) 2007, 2008 Mike Christie 6 * 7 * This program is free software; you can redistribute it and/or modify 8 * it under the terms of the GNU General Public License as published by 9 * the Free Software Foundation. 10 * 11 * Written by: Anil Veerabhadrappa (anilgv@broadcom.com) 12 * Maintained by: Eddie Wai (eddie.wai@broadcom.com) 13 */ 14 15 #include "bnx2i.h" 16 17 static struct list_head adapter_list = LIST_HEAD_INIT(adapter_list); 18 static u32 adapter_count; 19 20 #define DRV_MODULE_NAME "bnx2i" 21 #define DRV_MODULE_VERSION "2.7.6.2" 22 #define DRV_MODULE_RELDATE "Jun 06, 2013" 23 24 static char version[] = 25 "Broadcom NetXtreme II iSCSI Driver " DRV_MODULE_NAME \ 26 " v" DRV_MODULE_VERSION " (" DRV_MODULE_RELDATE ")\n"; 27 28 29 MODULE_AUTHOR("Anil Veerabhadrappa <anilgv@broadcom.com> and " 30 "Eddie Wai <eddie.wai@broadcom.com>"); 31 32 MODULE_DESCRIPTION("Broadcom NetXtreme II BCM5706/5708/5709/57710/57711/57712" 33 "/57800/57810/57840 iSCSI Driver"); 34 MODULE_LICENSE("GPL"); 35 MODULE_VERSION(DRV_MODULE_VERSION); 36 37 static DEFINE_MUTEX(bnx2i_dev_lock); 38 39 unsigned int event_coal_min = 24; 40 module_param(event_coal_min, int, 0664); 41 MODULE_PARM_DESC(event_coal_min, "Event Coalescing Minimum Commands"); 42 43 unsigned int event_coal_div = 2; 44 module_param(event_coal_div, int, 0664); 45 MODULE_PARM_DESC(event_coal_div, "Event Coalescing Divide Factor"); 46 47 unsigned int en_tcp_dack = 1; 48 module_param(en_tcp_dack, int, 0664); 49 MODULE_PARM_DESC(en_tcp_dack, "Enable TCP Delayed ACK"); 50 51 unsigned int error_mask1 = 0x00; 52 module_param(error_mask1, uint, 0664); 53 MODULE_PARM_DESC(error_mask1, "Config FW iSCSI Error Mask #1"); 54 55 unsigned int error_mask2 = 0x00; 56 module_param(error_mask2, uint, 0664); 57 MODULE_PARM_DESC(error_mask2, "Config FW iSCSI Error Mask #2"); 58 59 unsigned int sq_size; 60 module_param(sq_size, int, 0664); 61 MODULE_PARM_DESC(sq_size, "Configure SQ size"); 62 63 unsigned int rq_size = BNX2I_RQ_WQES_DEFAULT; 64 module_param(rq_size, int, 0664); 65 MODULE_PARM_DESC(rq_size, "Configure RQ size"); 66 67 u64 iscsi_error_mask = 0x00; 68 69 DEFINE_PER_CPU(struct bnx2i_percpu_s, bnx2i_percpu); 70 71 static int bnx2i_cpu_callback(struct notifier_block *nfb, 72 unsigned long action, void *hcpu); 73 /* notification function for CPU hotplug events */ 74 static struct notifier_block bnx2i_cpu_notifier = { 75 .notifier_call = bnx2i_cpu_callback, 76 }; 77 78 79 /** 80 * bnx2i_identify_device - identifies NetXtreme II device type 81 * @hba: Adapter structure pointer 82 * @cnic: Corresponding cnic device 83 * 84 * This function identifies the NX2 device type and sets appropriate 85 * queue mailbox register access method, 5709 requires driver to 86 * access MBOX regs using *bin* mode 87 */ 88 void bnx2i_identify_device(struct bnx2i_hba *hba, struct cnic_dev *dev) 89 { 90 hba->cnic_dev_type = 0; 91 if (test_bit(CNIC_F_BNX2_CLASS, &dev->flags)) { 92 if (hba->pci_did == PCI_DEVICE_ID_NX2_5706 || 93 hba->pci_did == PCI_DEVICE_ID_NX2_5706S) { 94 set_bit(BNX2I_NX2_DEV_5706, &hba->cnic_dev_type); 95 } else if (hba->pci_did == PCI_DEVICE_ID_NX2_5708 || 96 hba->pci_did == PCI_DEVICE_ID_NX2_5708S) { 97 set_bit(BNX2I_NX2_DEV_5708, &hba->cnic_dev_type); 98 } else if (hba->pci_did == PCI_DEVICE_ID_NX2_5709 || 99 hba->pci_did == PCI_DEVICE_ID_NX2_5709S) { 100 set_bit(BNX2I_NX2_DEV_5709, &hba->cnic_dev_type); 101 hba->mail_queue_access = BNX2I_MQ_BIN_MODE; 102 } 103 } else if (test_bit(CNIC_F_BNX2X_CLASS, &dev->flags)) { 104 set_bit(BNX2I_NX2_DEV_57710, &hba->cnic_dev_type); 105 } else { 106 printk(KERN_ALERT "bnx2i: unknown device, 0x%x\n", 107 hba->pci_did); 108 } 109 } 110 111 112 /** 113 * get_adapter_list_head - returns head of adapter list 114 */ 115 struct bnx2i_hba *get_adapter_list_head(void) 116 { 117 struct bnx2i_hba *hba = NULL; 118 struct bnx2i_hba *tmp_hba; 119 120 if (!adapter_count) 121 goto hba_not_found; 122 123 mutex_lock(&bnx2i_dev_lock); 124 list_for_each_entry(tmp_hba, &adapter_list, link) { 125 if (tmp_hba->cnic && tmp_hba->cnic->cm_select_dev) { 126 hba = tmp_hba; 127 break; 128 } 129 } 130 mutex_unlock(&bnx2i_dev_lock); 131 hba_not_found: 132 return hba; 133 } 134 135 136 /** 137 * bnx2i_find_hba_for_cnic - maps cnic device instance to bnx2i adapter instance 138 * @cnic: pointer to cnic device instance 139 * 140 */ 141 struct bnx2i_hba *bnx2i_find_hba_for_cnic(struct cnic_dev *cnic) 142 { 143 struct bnx2i_hba *hba, *temp; 144 145 mutex_lock(&bnx2i_dev_lock); 146 list_for_each_entry_safe(hba, temp, &adapter_list, link) { 147 if (hba->cnic == cnic) { 148 mutex_unlock(&bnx2i_dev_lock); 149 return hba; 150 } 151 } 152 mutex_unlock(&bnx2i_dev_lock); 153 return NULL; 154 } 155 156 157 /** 158 * bnx2i_start - cnic callback to initialize & start adapter instance 159 * @handle: transparent handle pointing to adapter structure 160 * 161 * This function maps adapter structure to pcidev structure and initiates 162 * firmware handshake to enable/initialize on chip iscsi components 163 * This bnx2i - cnic interface api callback is issued after following 164 * 2 conditions are met - 165 * a) underlying network interface is up (marked by event 'NETDEV_UP' 166 * from netdev 167 * b) bnx2i adapter instance is registered 168 */ 169 void bnx2i_start(void *handle) 170 { 171 #define BNX2I_INIT_POLL_TIME (1000 / HZ) 172 struct bnx2i_hba *hba = handle; 173 int i = HZ; 174 175 /* On some bnx2x devices, it is possible that iSCSI is no 176 * longer supported after firmware is downloaded. In that 177 * case, the iscsi_init_msg will return failure. 178 */ 179 180 bnx2i_send_fw_iscsi_init_msg(hba); 181 while (!test_bit(ADAPTER_STATE_UP, &hba->adapter_state) && 182 !test_bit(ADAPTER_STATE_INIT_FAILED, &hba->adapter_state) && i--) 183 msleep(BNX2I_INIT_POLL_TIME); 184 } 185 186 187 /** 188 * bnx2i_chip_cleanup - local routine to handle chip cleanup 189 * @hba: Adapter instance to register 190 * 191 * Driver checks if adapter still has any active connections before 192 * executing the cleanup process 193 */ 194 static void bnx2i_chip_cleanup(struct bnx2i_hba *hba) 195 { 196 struct bnx2i_endpoint *bnx2i_ep; 197 struct list_head *pos, *tmp; 198 199 if (hba->ofld_conns_active) { 200 /* Stage to force the disconnection 201 * This is the case where the daemon is either slow or 202 * not present 203 */ 204 printk(KERN_ALERT "bnx2i: (%s) chip cleanup for %d active " 205 "connections\n", hba->netdev->name, 206 hba->ofld_conns_active); 207 mutex_lock(&hba->net_dev_lock); 208 list_for_each_safe(pos, tmp, &hba->ep_active_list) { 209 bnx2i_ep = list_entry(pos, struct bnx2i_endpoint, link); 210 /* Clean up the chip only */ 211 bnx2i_hw_ep_disconnect(bnx2i_ep); 212 bnx2i_ep->cm_sk = NULL; 213 } 214 mutex_unlock(&hba->net_dev_lock); 215 } 216 } 217 218 219 /** 220 * bnx2i_stop - cnic callback to shutdown adapter instance 221 * @handle: transparent handle pointing to adapter structure 222 * 223 * driver checks if adapter is already in shutdown mode, if not start 224 * the shutdown process 225 */ 226 void bnx2i_stop(void *handle) 227 { 228 struct bnx2i_hba *hba = handle; 229 int conns_active; 230 int wait_delay = 1 * HZ; 231 232 /* check if cleanup happened in GOING_DOWN context */ 233 if (!test_and_set_bit(ADAPTER_STATE_GOING_DOWN, 234 &hba->adapter_state)) { 235 iscsi_host_for_each_session(hba->shost, 236 bnx2i_drop_session); 237 wait_delay = hba->hba_shutdown_tmo; 238 } 239 /* Wait for inflight offload connection tasks to complete before 240 * proceeding. Forcefully terminate all connection recovery in 241 * progress at the earliest, either in bind(), send_pdu(LOGIN), 242 * or conn_start() 243 */ 244 wait_event_interruptible_timeout(hba->eh_wait, 245 (list_empty(&hba->ep_ofld_list) && 246 list_empty(&hba->ep_destroy_list)), 247 2 * HZ); 248 /* Wait for all endpoints to be torn down, Chip will be reset once 249 * control returns to network driver. So it is required to cleanup and 250 * release all connection resources before returning from this routine. 251 */ 252 while (hba->ofld_conns_active) { 253 conns_active = hba->ofld_conns_active; 254 wait_event_interruptible_timeout(hba->eh_wait, 255 (hba->ofld_conns_active != conns_active), 256 wait_delay); 257 if (hba->ofld_conns_active == conns_active) 258 break; 259 } 260 bnx2i_chip_cleanup(hba); 261 262 /* This flag should be cleared last so that ep_disconnect() gracefully 263 * cleans up connection context 264 */ 265 clear_bit(ADAPTER_STATE_GOING_DOWN, &hba->adapter_state); 266 clear_bit(ADAPTER_STATE_UP, &hba->adapter_state); 267 } 268 269 270 /** 271 * bnx2i_init_one - initialize an adapter instance and allocate memory resources 272 * @hba: bnx2i adapter instance 273 * @cnic: cnic device handle 274 * 275 * Global resource lock is held during critical sections below. This routine is 276 * called from either cnic_register_driver() or device hot plug context and 277 * and does majority of device specific initialization 278 */ 279 static int bnx2i_init_one(struct bnx2i_hba *hba, struct cnic_dev *cnic) 280 { 281 int rc; 282 283 mutex_lock(&bnx2i_dev_lock); 284 if (!cnic->max_iscsi_conn) { 285 printk(KERN_ALERT "bnx2i: dev %s does not support " 286 "iSCSI\n", hba->netdev->name); 287 rc = -EOPNOTSUPP; 288 goto out; 289 } 290 291 hba->cnic = cnic; 292 rc = cnic->register_device(cnic, CNIC_ULP_ISCSI, hba); 293 if (!rc) { 294 hba->age++; 295 set_bit(BNX2I_CNIC_REGISTERED, &hba->reg_with_cnic); 296 list_add_tail(&hba->link, &adapter_list); 297 adapter_count++; 298 } else if (rc == -EBUSY) /* duplicate registration */ 299 printk(KERN_ALERT "bnx2i, duplicate registration" 300 "hba=%p, cnic=%p\n", hba, cnic); 301 else if (rc == -EAGAIN) 302 printk(KERN_ERR "bnx2i, driver not registered\n"); 303 else if (rc == -EINVAL) 304 printk(KERN_ERR "bnx2i, invalid type %d\n", CNIC_ULP_ISCSI); 305 else 306 printk(KERN_ERR "bnx2i dev reg, unknown error, %d\n", rc); 307 308 out: 309 mutex_unlock(&bnx2i_dev_lock); 310 311 return rc; 312 } 313 314 315 /** 316 * bnx2i_ulp_init - initialize an adapter instance 317 * @dev: cnic device handle 318 * 319 * Called from cnic_register_driver() context to initialize all enumerated 320 * cnic devices. This routine allocate adapter structure and other 321 * device specific resources. 322 */ 323 void bnx2i_ulp_init(struct cnic_dev *dev) 324 { 325 struct bnx2i_hba *hba; 326 327 /* Allocate a HBA structure for this device */ 328 hba = bnx2i_alloc_hba(dev); 329 if (!hba) { 330 printk(KERN_ERR "bnx2i init: hba initialization failed\n"); 331 return; 332 } 333 334 /* Get PCI related information and update hba struct members */ 335 clear_bit(BNX2I_CNIC_REGISTERED, &hba->reg_with_cnic); 336 if (bnx2i_init_one(hba, dev)) { 337 printk(KERN_ERR "bnx2i - hba %p init failed\n", hba); 338 bnx2i_free_hba(hba); 339 } 340 } 341 342 343 /** 344 * bnx2i_ulp_exit - shuts down adapter instance and frees all resources 345 * @dev: cnic device handle 346 * 347 */ 348 void bnx2i_ulp_exit(struct cnic_dev *dev) 349 { 350 struct bnx2i_hba *hba; 351 352 hba = bnx2i_find_hba_for_cnic(dev); 353 if (!hba) { 354 printk(KERN_INFO "bnx2i_ulp_exit: hba not " 355 "found, dev 0x%p\n", dev); 356 return; 357 } 358 mutex_lock(&bnx2i_dev_lock); 359 list_del_init(&hba->link); 360 adapter_count--; 361 362 if (test_bit(BNX2I_CNIC_REGISTERED, &hba->reg_with_cnic)) { 363 hba->cnic->unregister_device(hba->cnic, CNIC_ULP_ISCSI); 364 clear_bit(BNX2I_CNIC_REGISTERED, &hba->reg_with_cnic); 365 } 366 mutex_unlock(&bnx2i_dev_lock); 367 368 bnx2i_free_hba(hba); 369 } 370 371 372 /** 373 * bnx2i_get_stats - Retrieve various statistic from iSCSI offload 374 * @handle: bnx2i_hba 375 * 376 * function callback exported via bnx2i - cnic driver interface to 377 * retrieve various iSCSI offload related statistics. 378 */ 379 int bnx2i_get_stats(void *handle) 380 { 381 struct bnx2i_hba *hba = handle; 382 struct iscsi_stats_info *stats; 383 384 if (!hba) 385 return -EINVAL; 386 387 stats = (struct iscsi_stats_info *)hba->cnic->stats_addr; 388 389 if (!stats) 390 return -ENOMEM; 391 392 strlcpy(stats->version, DRV_MODULE_VERSION, sizeof(stats->version)); 393 memcpy(stats->mac_add1 + 2, hba->cnic->mac_addr, ETH_ALEN); 394 395 stats->max_frame_size = hba->netdev->mtu; 396 stats->txq_size = hba->max_sqes; 397 stats->rxq_size = hba->max_cqes; 398 399 stats->txq_avg_depth = 0; 400 stats->rxq_avg_depth = 0; 401 402 GET_STATS_64(hba, stats, rx_pdus); 403 GET_STATS_64(hba, stats, rx_bytes); 404 405 GET_STATS_64(hba, stats, tx_pdus); 406 GET_STATS_64(hba, stats, tx_bytes); 407 408 return 0; 409 } 410 411 412 /** 413 * bnx2i_percpu_thread_create - Create a receive thread for an 414 * online CPU 415 * 416 * @cpu: cpu index for the online cpu 417 */ 418 static void bnx2i_percpu_thread_create(unsigned int cpu) 419 { 420 struct bnx2i_percpu_s *p; 421 struct task_struct *thread; 422 423 p = &per_cpu(bnx2i_percpu, cpu); 424 425 thread = kthread_create_on_node(bnx2i_percpu_io_thread, (void *)p, 426 cpu_to_node(cpu), 427 "bnx2i_thread/%d", cpu); 428 /* bind thread to the cpu */ 429 if (likely(!IS_ERR(thread))) { 430 kthread_bind(thread, cpu); 431 p->iothread = thread; 432 wake_up_process(thread); 433 } 434 } 435 436 437 static void bnx2i_percpu_thread_destroy(unsigned int cpu) 438 { 439 struct bnx2i_percpu_s *p; 440 struct task_struct *thread; 441 struct bnx2i_work *work, *tmp; 442 443 /* Prevent any new work from being queued for this CPU */ 444 p = &per_cpu(bnx2i_percpu, cpu); 445 spin_lock_bh(&p->p_work_lock); 446 thread = p->iothread; 447 p->iothread = NULL; 448 449 /* Free all work in the list */ 450 list_for_each_entry_safe(work, tmp, &p->work_list, list) { 451 list_del_init(&work->list); 452 bnx2i_process_scsi_cmd_resp(work->session, 453 work->bnx2i_conn, &work->cqe); 454 kfree(work); 455 } 456 457 spin_unlock_bh(&p->p_work_lock); 458 if (thread) 459 kthread_stop(thread); 460 } 461 462 463 /** 464 * bnx2i_cpu_callback - Handler for CPU hotplug events 465 * 466 * @nfb: The callback data block 467 * @action: The event triggering the callback 468 * @hcpu: The index of the CPU that the event is for 469 * 470 * This creates or destroys per-CPU data for iSCSI 471 * 472 * Returns NOTIFY_OK always. 473 */ 474 static int bnx2i_cpu_callback(struct notifier_block *nfb, 475 unsigned long action, void *hcpu) 476 { 477 unsigned cpu = (unsigned long)hcpu; 478 479 switch (action) { 480 case CPU_ONLINE: 481 case CPU_ONLINE_FROZEN: 482 printk(KERN_INFO "bnx2i: CPU %x online: Create Rx thread\n", 483 cpu); 484 bnx2i_percpu_thread_create(cpu); 485 break; 486 case CPU_DEAD: 487 case CPU_DEAD_FROZEN: 488 printk(KERN_INFO "CPU %x offline: Remove Rx thread\n", cpu); 489 bnx2i_percpu_thread_destroy(cpu); 490 break; 491 default: 492 break; 493 } 494 return NOTIFY_OK; 495 } 496 497 498 /** 499 * bnx2i_mod_init - module init entry point 500 * 501 * initialize any driver wide global data structures such as endpoint pool, 502 * tcp port manager/queue, sysfs. finally driver will register itself 503 * with the cnic module 504 */ 505 static int __init bnx2i_mod_init(void) 506 { 507 int err; 508 unsigned cpu = 0; 509 struct bnx2i_percpu_s *p; 510 511 printk(KERN_INFO "%s", version); 512 513 if (sq_size && !is_power_of_2(sq_size)) 514 sq_size = roundup_pow_of_two(sq_size); 515 516 mutex_init(&bnx2i_dev_lock); 517 518 bnx2i_scsi_xport_template = 519 iscsi_register_transport(&bnx2i_iscsi_transport); 520 if (!bnx2i_scsi_xport_template) { 521 printk(KERN_ERR "Could not register bnx2i transport.\n"); 522 err = -ENOMEM; 523 goto out; 524 } 525 526 err = cnic_register_driver(CNIC_ULP_ISCSI, &bnx2i_cnic_cb); 527 if (err) { 528 printk(KERN_ERR "Could not register bnx2i cnic driver.\n"); 529 goto unreg_xport; 530 } 531 532 /* Create percpu kernel threads to handle iSCSI I/O completions */ 533 for_each_possible_cpu(cpu) { 534 p = &per_cpu(bnx2i_percpu, cpu); 535 INIT_LIST_HEAD(&p->work_list); 536 spin_lock_init(&p->p_work_lock); 537 p->iothread = NULL; 538 } 539 540 cpu_notifier_register_begin(); 541 542 for_each_online_cpu(cpu) 543 bnx2i_percpu_thread_create(cpu); 544 545 /* Initialize per CPU interrupt thread */ 546 __register_hotcpu_notifier(&bnx2i_cpu_notifier); 547 548 cpu_notifier_register_done(); 549 550 return 0; 551 552 unreg_xport: 553 iscsi_unregister_transport(&bnx2i_iscsi_transport); 554 out: 555 return err; 556 } 557 558 559 /** 560 * bnx2i_mod_exit - module cleanup/exit entry point 561 * 562 * Global resource lock and host adapter lock is held during critical sections 563 * in this function. Driver will browse through the adapter list, cleans-up 564 * each instance, unregisters iscsi transport name and finally driver will 565 * unregister itself with the cnic module 566 */ 567 static void __exit bnx2i_mod_exit(void) 568 { 569 struct bnx2i_hba *hba; 570 unsigned cpu = 0; 571 572 mutex_lock(&bnx2i_dev_lock); 573 while (!list_empty(&adapter_list)) { 574 hba = list_entry(adapter_list.next, struct bnx2i_hba, link); 575 list_del(&hba->link); 576 adapter_count--; 577 578 if (test_bit(BNX2I_CNIC_REGISTERED, &hba->reg_with_cnic)) { 579 bnx2i_chip_cleanup(hba); 580 hba->cnic->unregister_device(hba->cnic, CNIC_ULP_ISCSI); 581 clear_bit(BNX2I_CNIC_REGISTERED, &hba->reg_with_cnic); 582 } 583 584 bnx2i_free_hba(hba); 585 } 586 mutex_unlock(&bnx2i_dev_lock); 587 588 cpu_notifier_register_begin(); 589 590 for_each_online_cpu(cpu) 591 bnx2i_percpu_thread_destroy(cpu); 592 593 __unregister_hotcpu_notifier(&bnx2i_cpu_notifier); 594 595 cpu_notifier_register_done(); 596 597 iscsi_unregister_transport(&bnx2i_iscsi_transport); 598 cnic_unregister_driver(CNIC_ULP_ISCSI); 599 } 600 601 module_init(bnx2i_mod_init); 602 module_exit(bnx2i_mod_exit); 603