1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /**************************************************************************/ 3 /* */ 4 /* IBM System i and System p Virtual NIC Device Driver */ 5 /* Copyright (C) 2014 IBM Corp. */ 6 /* Santiago Leon (santi_leon@yahoo.com) */ 7 /* Thomas Falcon (tlfalcon@linux.vnet.ibm.com) */ 8 /* John Allen (jallen@linux.vnet.ibm.com) */ 9 /* */ 10 /* */ 11 /* This module contains the implementation of a virtual ethernet device */ 12 /* for use with IBM i/p Series LPAR Linux. It utilizes the logical LAN */ 13 /* option of the RS/6000 Platform Architecture to interface with virtual */ 14 /* ethernet NICs that are presented to the partition by the hypervisor. */ 15 /* */ 16 /* Messages are passed between the VNIC driver and the VNIC server using */ 17 /* Command/Response Queues (CRQs) and sub CRQs (sCRQs). CRQs are used to */ 18 /* issue and receive commands that initiate communication with the server */ 19 /* on driver initialization. Sub CRQs (sCRQs) are similar to CRQs, but */ 20 /* are used by the driver to notify the server that a packet is */ 21 /* ready for transmission or that a buffer has been added to receive a */ 22 /* packet. Subsequently, sCRQs are used by the server to notify the */ 23 /* driver that a packet transmission has been completed or that a packet */ 24 /* has been received and placed in a waiting buffer. */ 25 /* */ 26 /* In lieu of a more conventional "on-the-fly" DMA mapping strategy in */ 27 /* which skbs are DMA mapped and immediately unmapped when the transmit */ 28 /* or receive has been completed, the VNIC driver is required to use */ 29 /* "long term mapping". This entails that large, continuous DMA mapped */ 30 /* buffers are allocated on driver initialization and these buffers are */ 31 /* then continuously reused to pass skbs to and from the VNIC server. */ 32 /* */ 33 /**************************************************************************/ 34 35 #include <linux/module.h> 36 #include <linux/moduleparam.h> 37 #include <linux/types.h> 38 #include <linux/errno.h> 39 #include <linux/completion.h> 40 #include <linux/ioport.h> 41 #include <linux/dma-mapping.h> 42 #include <linux/kernel.h> 43 #include <linux/netdevice.h> 44 #include <linux/etherdevice.h> 45 #include <linux/skbuff.h> 46 #include <linux/init.h> 47 #include <linux/delay.h> 48 #include <linux/mm.h> 49 #include <linux/ethtool.h> 50 #include <linux/proc_fs.h> 51 #include <linux/if_arp.h> 52 #include <linux/in.h> 53 #include <linux/ip.h> 54 #include <linux/ipv6.h> 55 #include <linux/irq.h> 56 #include <linux/irqdomain.h> 57 #include <linux/kthread.h> 58 #include <linux/seq_file.h> 59 #include <linux/interrupt.h> 60 #include <net/net_namespace.h> 61 #include <asm/hvcall.h> 62 #include <linux/atomic.h> 63 #include <asm/vio.h> 64 #include <asm/xive.h> 65 #include <asm/iommu.h> 66 #include <linux/uaccess.h> 67 #include <asm/firmware.h> 68 #include <linux/workqueue.h> 69 #include <linux/if_vlan.h> 70 #include <linux/utsname.h> 71 #include <linux/cpu.h> 72 73 #include "ibmvnic.h" 74 75 static const char ibmvnic_driver_name[] = "ibmvnic"; 76 static const char ibmvnic_driver_string[] = "IBM System i/p Virtual NIC Driver"; 77 78 MODULE_AUTHOR("Santiago Leon"); 79 MODULE_DESCRIPTION("IBM System i/p Virtual NIC Driver"); 80 MODULE_LICENSE("GPL"); 81 MODULE_VERSION(IBMVNIC_DRIVER_VERSION); 82 83 static int ibmvnic_version = IBMVNIC_INITIAL_VERSION; 84 static void release_sub_crqs(struct ibmvnic_adapter *, bool); 85 static int ibmvnic_reset_crq(struct ibmvnic_adapter *); 86 static int ibmvnic_send_crq_init(struct ibmvnic_adapter *); 87 static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *); 88 static int ibmvnic_send_crq(struct ibmvnic_adapter *, union ibmvnic_crq *); 89 static int send_subcrq_indirect(struct ibmvnic_adapter *, u64, u64, u64); 90 static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance); 91 static int enable_scrq_irq(struct ibmvnic_adapter *, 92 struct ibmvnic_sub_crq_queue *); 93 static int disable_scrq_irq(struct ibmvnic_adapter *, 94 struct ibmvnic_sub_crq_queue *); 95 static int pending_scrq(struct ibmvnic_adapter *, 96 struct ibmvnic_sub_crq_queue *); 97 static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *, 98 struct ibmvnic_sub_crq_queue *); 99 static int ibmvnic_poll(struct napi_struct *napi, int data); 100 static int reset_sub_crq_queues(struct ibmvnic_adapter *adapter); 101 static inline void reinit_init_done(struct ibmvnic_adapter *adapter); 102 static void send_query_map(struct ibmvnic_adapter *adapter); 103 static int send_request_map(struct ibmvnic_adapter *, dma_addr_t, u32, u8); 104 static int send_request_unmap(struct ibmvnic_adapter *, u8); 105 static int send_login(struct ibmvnic_adapter *adapter); 106 static void send_query_cap(struct ibmvnic_adapter *adapter); 107 static int init_sub_crqs(struct ibmvnic_adapter *); 108 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter); 109 static int ibmvnic_reset_init(struct ibmvnic_adapter *, bool reset); 110 static void release_crq_queue(struct ibmvnic_adapter *); 111 static int __ibmvnic_set_mac(struct net_device *, u8 *); 112 static int init_crq_queue(struct ibmvnic_adapter *adapter); 113 static int send_query_phys_parms(struct ibmvnic_adapter *adapter); 114 static void ibmvnic_tx_scrq_clean_buffer(struct ibmvnic_adapter *adapter, 115 struct ibmvnic_sub_crq_queue *tx_scrq); 116 static void free_long_term_buff(struct ibmvnic_adapter *adapter, 117 struct ibmvnic_long_term_buff *ltb); 118 static void ibmvnic_disable_irqs(struct ibmvnic_adapter *adapter); 119 static void flush_reset_queue(struct ibmvnic_adapter *adapter); 120 121 struct ibmvnic_stat { 122 char name[ETH_GSTRING_LEN]; 123 int offset; 124 }; 125 126 #define IBMVNIC_STAT_OFF(stat) (offsetof(struct ibmvnic_adapter, stats) + \ 127 offsetof(struct ibmvnic_statistics, stat)) 128 #define IBMVNIC_GET_STAT(a, off) (*((u64 *)(((unsigned long)(a)) + (off)))) 129 130 static const struct ibmvnic_stat ibmvnic_stats[] = { 131 {"rx_packets", IBMVNIC_STAT_OFF(rx_packets)}, 132 {"rx_bytes", IBMVNIC_STAT_OFF(rx_bytes)}, 133 {"tx_packets", IBMVNIC_STAT_OFF(tx_packets)}, 134 {"tx_bytes", IBMVNIC_STAT_OFF(tx_bytes)}, 135 {"ucast_tx_packets", IBMVNIC_STAT_OFF(ucast_tx_packets)}, 136 {"ucast_rx_packets", IBMVNIC_STAT_OFF(ucast_rx_packets)}, 137 {"mcast_tx_packets", IBMVNIC_STAT_OFF(mcast_tx_packets)}, 138 {"mcast_rx_packets", IBMVNIC_STAT_OFF(mcast_rx_packets)}, 139 {"bcast_tx_packets", IBMVNIC_STAT_OFF(bcast_tx_packets)}, 140 {"bcast_rx_packets", IBMVNIC_STAT_OFF(bcast_rx_packets)}, 141 {"align_errors", IBMVNIC_STAT_OFF(align_errors)}, 142 {"fcs_errors", IBMVNIC_STAT_OFF(fcs_errors)}, 143 {"single_collision_frames", IBMVNIC_STAT_OFF(single_collision_frames)}, 144 {"multi_collision_frames", IBMVNIC_STAT_OFF(multi_collision_frames)}, 145 {"sqe_test_errors", IBMVNIC_STAT_OFF(sqe_test_errors)}, 146 {"deferred_tx", IBMVNIC_STAT_OFF(deferred_tx)}, 147 {"late_collisions", IBMVNIC_STAT_OFF(late_collisions)}, 148 {"excess_collisions", IBMVNIC_STAT_OFF(excess_collisions)}, 149 {"internal_mac_tx_errors", IBMVNIC_STAT_OFF(internal_mac_tx_errors)}, 150 {"carrier_sense", IBMVNIC_STAT_OFF(carrier_sense)}, 151 {"too_long_frames", IBMVNIC_STAT_OFF(too_long_frames)}, 152 {"internal_mac_rx_errors", IBMVNIC_STAT_OFF(internal_mac_rx_errors)}, 153 }; 154 155 static int send_crq_init_complete(struct ibmvnic_adapter *adapter) 156 { 157 union ibmvnic_crq crq; 158 159 memset(&crq, 0, sizeof(crq)); 160 crq.generic.first = IBMVNIC_CRQ_INIT_CMD; 161 crq.generic.cmd = IBMVNIC_CRQ_INIT_COMPLETE; 162 163 return ibmvnic_send_crq(adapter, &crq); 164 } 165 166 static int send_version_xchg(struct ibmvnic_adapter *adapter) 167 { 168 union ibmvnic_crq crq; 169 170 memset(&crq, 0, sizeof(crq)); 171 crq.version_exchange.first = IBMVNIC_CRQ_CMD; 172 crq.version_exchange.cmd = VERSION_EXCHANGE; 173 crq.version_exchange.version = cpu_to_be16(ibmvnic_version); 174 175 return ibmvnic_send_crq(adapter, &crq); 176 } 177 178 static void ibmvnic_clean_queue_affinity(struct ibmvnic_adapter *adapter, 179 struct ibmvnic_sub_crq_queue *queue) 180 { 181 if (!(queue && queue->irq)) 182 return; 183 184 cpumask_clear(queue->affinity_mask); 185 186 if (irq_set_affinity_and_hint(queue->irq, NULL)) 187 netdev_warn(adapter->netdev, 188 "%s: Clear affinity failed, queue addr = %p, IRQ = %d\n", 189 __func__, queue, queue->irq); 190 } 191 192 static void ibmvnic_clean_affinity(struct ibmvnic_adapter *adapter) 193 { 194 struct ibmvnic_sub_crq_queue **rxqs; 195 struct ibmvnic_sub_crq_queue **txqs; 196 int num_rxqs, num_txqs; 197 int i; 198 199 rxqs = adapter->rx_scrq; 200 txqs = adapter->tx_scrq; 201 num_txqs = adapter->num_active_tx_scrqs; 202 num_rxqs = adapter->num_active_rx_scrqs; 203 204 netdev_dbg(adapter->netdev, "%s: Cleaning irq affinity hints", __func__); 205 if (txqs) { 206 for (i = 0; i < num_txqs; i++) 207 ibmvnic_clean_queue_affinity(adapter, txqs[i]); 208 } 209 if (rxqs) { 210 for (i = 0; i < num_rxqs; i++) 211 ibmvnic_clean_queue_affinity(adapter, rxqs[i]); 212 } 213 } 214 215 static int ibmvnic_set_queue_affinity(struct ibmvnic_sub_crq_queue *queue, 216 unsigned int *cpu, int *stragglers, 217 int stride) 218 { 219 cpumask_var_t mask; 220 int i; 221 int rc = 0; 222 223 if (!(queue && queue->irq)) 224 return rc; 225 226 /* cpumask_var_t is either a pointer or array, allocation works here */ 227 if (!zalloc_cpumask_var(&mask, GFP_KERNEL)) 228 return -ENOMEM; 229 230 /* while we have extra cpu give one extra to this irq */ 231 if (*stragglers) { 232 stride++; 233 (*stragglers)--; 234 } 235 /* atomic write is safer than writing bit by bit directly */ 236 for (i = 0; i < stride; i++) { 237 cpumask_set_cpu(*cpu, mask); 238 *cpu = cpumask_next_wrap(*cpu, cpu_online_mask, 239 nr_cpu_ids, false); 240 } 241 /* set queue affinity mask */ 242 cpumask_copy(queue->affinity_mask, mask); 243 rc = irq_set_affinity_and_hint(queue->irq, queue->affinity_mask); 244 free_cpumask_var(mask); 245 246 return rc; 247 } 248 249 /* assumes cpu read lock is held */ 250 static void ibmvnic_set_affinity(struct ibmvnic_adapter *adapter) 251 { 252 struct ibmvnic_sub_crq_queue **rxqs = adapter->rx_scrq; 253 struct ibmvnic_sub_crq_queue **txqs = adapter->tx_scrq; 254 struct ibmvnic_sub_crq_queue *queue; 255 int num_rxqs = adapter->num_active_rx_scrqs, i_rxqs = 0; 256 int num_txqs = adapter->num_active_tx_scrqs, i_txqs = 0; 257 int total_queues, stride, stragglers, i; 258 unsigned int num_cpu, cpu; 259 bool is_rx_queue; 260 int rc = 0; 261 262 netdev_dbg(adapter->netdev, "%s: Setting irq affinity hints", __func__); 263 if (!(adapter->rx_scrq && adapter->tx_scrq)) { 264 netdev_warn(adapter->netdev, 265 "%s: Set affinity failed, queues not allocated\n", 266 __func__); 267 return; 268 } 269 270 total_queues = num_rxqs + num_txqs; 271 num_cpu = num_online_cpus(); 272 /* number of cpu's assigned per irq */ 273 stride = max_t(int, num_cpu / total_queues, 1); 274 /* number of leftover cpu's */ 275 stragglers = num_cpu >= total_queues ? num_cpu % total_queues : 0; 276 /* next available cpu to assign irq to */ 277 cpu = cpumask_next(-1, cpu_online_mask); 278 279 for (i = 0; i < total_queues; i++) { 280 is_rx_queue = false; 281 /* balance core load by alternating rx and tx assignments 282 * ex: TX0 -> RX0 -> TX1 -> RX1 etc. 283 */ 284 if ((i % 2 == 1 && i_rxqs < num_rxqs) || i_txqs == num_txqs) { 285 queue = rxqs[i_rxqs++]; 286 is_rx_queue = true; 287 } else { 288 queue = txqs[i_txqs++]; 289 } 290 291 rc = ibmvnic_set_queue_affinity(queue, &cpu, &stragglers, 292 stride); 293 if (rc) 294 goto out; 295 296 if (!queue || is_rx_queue) 297 continue; 298 299 rc = __netif_set_xps_queue(adapter->netdev, 300 cpumask_bits(queue->affinity_mask), 301 i_txqs - 1, XPS_CPUS); 302 if (rc) 303 netdev_warn(adapter->netdev, "%s: Set XPS on queue %d failed, rc = %d.\n", 304 __func__, i_txqs - 1, rc); 305 } 306 307 out: 308 if (rc) { 309 netdev_warn(adapter->netdev, 310 "%s: Set affinity failed, queue addr = %p, IRQ = %d, rc = %d.\n", 311 __func__, queue, queue->irq, rc); 312 ibmvnic_clean_affinity(adapter); 313 } 314 } 315 316 static int ibmvnic_cpu_online(unsigned int cpu, struct hlist_node *node) 317 { 318 struct ibmvnic_adapter *adapter; 319 320 adapter = hlist_entry_safe(node, struct ibmvnic_adapter, node); 321 ibmvnic_set_affinity(adapter); 322 return 0; 323 } 324 325 static int ibmvnic_cpu_dead(unsigned int cpu, struct hlist_node *node) 326 { 327 struct ibmvnic_adapter *adapter; 328 329 adapter = hlist_entry_safe(node, struct ibmvnic_adapter, node_dead); 330 ibmvnic_set_affinity(adapter); 331 return 0; 332 } 333 334 static int ibmvnic_cpu_down_prep(unsigned int cpu, struct hlist_node *node) 335 { 336 struct ibmvnic_adapter *adapter; 337 338 adapter = hlist_entry_safe(node, struct ibmvnic_adapter, node); 339 ibmvnic_clean_affinity(adapter); 340 return 0; 341 } 342 343 static enum cpuhp_state ibmvnic_online; 344 345 static int ibmvnic_cpu_notif_add(struct ibmvnic_adapter *adapter) 346 { 347 int ret; 348 349 ret = cpuhp_state_add_instance_nocalls(ibmvnic_online, &adapter->node); 350 if (ret) 351 return ret; 352 ret = cpuhp_state_add_instance_nocalls(CPUHP_IBMVNIC_DEAD, 353 &adapter->node_dead); 354 if (!ret) 355 return ret; 356 cpuhp_state_remove_instance_nocalls(ibmvnic_online, &adapter->node); 357 return ret; 358 } 359 360 static void ibmvnic_cpu_notif_remove(struct ibmvnic_adapter *adapter) 361 { 362 cpuhp_state_remove_instance_nocalls(ibmvnic_online, &adapter->node); 363 cpuhp_state_remove_instance_nocalls(CPUHP_IBMVNIC_DEAD, 364 &adapter->node_dead); 365 } 366 367 static long h_reg_sub_crq(unsigned long unit_address, unsigned long token, 368 unsigned long length, unsigned long *number, 369 unsigned long *irq) 370 { 371 unsigned long retbuf[PLPAR_HCALL_BUFSIZE]; 372 long rc; 373 374 rc = plpar_hcall(H_REG_SUB_CRQ, retbuf, unit_address, token, length); 375 *number = retbuf[0]; 376 *irq = retbuf[1]; 377 378 return rc; 379 } 380 381 /** 382 * ibmvnic_wait_for_completion - Check device state and wait for completion 383 * @adapter: private device data 384 * @comp_done: completion structure to wait for 385 * @timeout: time to wait in milliseconds 386 * 387 * Wait for a completion signal or until the timeout limit is reached 388 * while checking that the device is still active. 389 */ 390 static int ibmvnic_wait_for_completion(struct ibmvnic_adapter *adapter, 391 struct completion *comp_done, 392 unsigned long timeout) 393 { 394 struct net_device *netdev; 395 unsigned long div_timeout; 396 u8 retry; 397 398 netdev = adapter->netdev; 399 retry = 5; 400 div_timeout = msecs_to_jiffies(timeout / retry); 401 while (true) { 402 if (!adapter->crq.active) { 403 netdev_err(netdev, "Device down!\n"); 404 return -ENODEV; 405 } 406 if (!retry--) 407 break; 408 if (wait_for_completion_timeout(comp_done, div_timeout)) 409 return 0; 410 } 411 netdev_err(netdev, "Operation timed out.\n"); 412 return -ETIMEDOUT; 413 } 414 415 /** 416 * reuse_ltb() - Check if a long term buffer can be reused 417 * @ltb: The long term buffer to be checked 418 * @size: The size of the long term buffer. 419 * 420 * An LTB can be reused unless its size has changed. 421 * 422 * Return: Return true if the LTB can be reused, false otherwise. 423 */ 424 static bool reuse_ltb(struct ibmvnic_long_term_buff *ltb, int size) 425 { 426 return (ltb->buff && ltb->size == size); 427 } 428 429 /** 430 * alloc_long_term_buff() - Allocate a long term buffer (LTB) 431 * 432 * @adapter: ibmvnic adapter associated to the LTB 433 * @ltb: container object for the LTB 434 * @size: size of the LTB 435 * 436 * Allocate an LTB of the specified size and notify VIOS. 437 * 438 * If the given @ltb already has the correct size, reuse it. Otherwise if 439 * its non-NULL, free it. Then allocate a new one of the correct size. 440 * Notify the VIOS either way since we may now be working with a new VIOS. 441 * 442 * Allocating larger chunks of memory during resets, specially LPM or under 443 * low memory situations can cause resets to fail/timeout and for LPAR to 444 * lose connectivity. So hold onto the LTB even if we fail to communicate 445 * with the VIOS and reuse it on next open. Free LTB when adapter is closed. 446 * 447 * Return: 0 if we were able to allocate the LTB and notify the VIOS and 448 * a negative value otherwise. 449 */ 450 static int alloc_long_term_buff(struct ibmvnic_adapter *adapter, 451 struct ibmvnic_long_term_buff *ltb, int size) 452 { 453 struct device *dev = &adapter->vdev->dev; 454 u64 prev = 0; 455 int rc; 456 457 if (!reuse_ltb(ltb, size)) { 458 dev_dbg(dev, 459 "LTB size changed from 0x%llx to 0x%x, reallocating\n", 460 ltb->size, size); 461 prev = ltb->size; 462 free_long_term_buff(adapter, ltb); 463 } 464 465 if (ltb->buff) { 466 dev_dbg(dev, "Reusing LTB [map %d, size 0x%llx]\n", 467 ltb->map_id, ltb->size); 468 } else { 469 ltb->buff = dma_alloc_coherent(dev, size, <b->addr, 470 GFP_KERNEL); 471 if (!ltb->buff) { 472 dev_err(dev, "Couldn't alloc long term buffer\n"); 473 return -ENOMEM; 474 } 475 ltb->size = size; 476 477 ltb->map_id = find_first_zero_bit(adapter->map_ids, 478 MAX_MAP_ID); 479 bitmap_set(adapter->map_ids, ltb->map_id, 1); 480 481 dev_dbg(dev, 482 "Allocated new LTB [map %d, size 0x%llx was 0x%llx]\n", 483 ltb->map_id, ltb->size, prev); 484 } 485 486 /* Ensure ltb is zeroed - specially when reusing it. */ 487 memset(ltb->buff, 0, ltb->size); 488 489 mutex_lock(&adapter->fw_lock); 490 adapter->fw_done_rc = 0; 491 reinit_completion(&adapter->fw_done); 492 493 rc = send_request_map(adapter, ltb->addr, ltb->size, ltb->map_id); 494 if (rc) { 495 dev_err(dev, "send_request_map failed, rc = %d\n", rc); 496 goto out; 497 } 498 499 rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000); 500 if (rc) { 501 dev_err(dev, "LTB map request aborted or timed out, rc = %d\n", 502 rc); 503 goto out; 504 } 505 506 if (adapter->fw_done_rc) { 507 dev_err(dev, "Couldn't map LTB, rc = %d\n", 508 adapter->fw_done_rc); 509 rc = -EIO; 510 goto out; 511 } 512 rc = 0; 513 out: 514 /* don't free LTB on communication error - see function header */ 515 mutex_unlock(&adapter->fw_lock); 516 return rc; 517 } 518 519 static void free_long_term_buff(struct ibmvnic_adapter *adapter, 520 struct ibmvnic_long_term_buff *ltb) 521 { 522 struct device *dev = &adapter->vdev->dev; 523 524 if (!ltb->buff) 525 return; 526 527 /* VIOS automatically unmaps the long term buffer at remote 528 * end for the following resets: 529 * FAILOVER, MOBILITY, TIMEOUT. 530 */ 531 if (adapter->reset_reason != VNIC_RESET_FAILOVER && 532 adapter->reset_reason != VNIC_RESET_MOBILITY && 533 adapter->reset_reason != VNIC_RESET_TIMEOUT) 534 send_request_unmap(adapter, ltb->map_id); 535 536 dma_free_coherent(dev, ltb->size, ltb->buff, ltb->addr); 537 538 ltb->buff = NULL; 539 /* mark this map_id free */ 540 bitmap_clear(adapter->map_ids, ltb->map_id, 1); 541 ltb->map_id = 0; 542 } 543 544 /** 545 * free_ltb_set - free the given set of long term buffers (LTBS) 546 * @adapter: The ibmvnic adapter containing this ltb set 547 * @ltb_set: The ltb_set to be freed 548 * 549 * Free the set of LTBs in the given set. 550 */ 551 552 static void free_ltb_set(struct ibmvnic_adapter *adapter, 553 struct ibmvnic_ltb_set *ltb_set) 554 { 555 int i; 556 557 for (i = 0; i < ltb_set->num_ltbs; i++) 558 free_long_term_buff(adapter, <b_set->ltbs[i]); 559 560 kfree(ltb_set->ltbs); 561 ltb_set->ltbs = NULL; 562 ltb_set->num_ltbs = 0; 563 } 564 565 /** 566 * alloc_ltb_set() - Allocate a set of long term buffers (LTBs) 567 * 568 * @adapter: ibmvnic adapter associated to the LTB 569 * @ltb_set: container object for the set of LTBs 570 * @num_buffs: Number of buffers in the LTB 571 * @buff_size: Size of each buffer in the LTB 572 * 573 * Allocate a set of LTBs to accommodate @num_buffs buffers of @buff_size 574 * each. We currently cap size each LTB to IBMVNIC_ONE_LTB_SIZE. If the 575 * new set of LTBs have fewer LTBs than the old set, free the excess LTBs. 576 * If new set needs more than in old set, allocate the remaining ones. 577 * Try and reuse as many LTBs as possible and avoid reallocation. 578 * 579 * Any changes to this allocation strategy must be reflected in 580 * map_rxpool_buff_to_ltb() and map_txpool_buff_to_ltb(). 581 */ 582 static int alloc_ltb_set(struct ibmvnic_adapter *adapter, 583 struct ibmvnic_ltb_set *ltb_set, int num_buffs, 584 int buff_size) 585 { 586 struct device *dev = &adapter->vdev->dev; 587 struct ibmvnic_ltb_set old_set; 588 struct ibmvnic_ltb_set new_set; 589 int rem_size; 590 int tot_size; /* size of all ltbs */ 591 int ltb_size; /* size of one ltb */ 592 int nltbs; 593 int rc; 594 int n; 595 int i; 596 597 dev_dbg(dev, "%s() num_buffs %d, buff_size %d\n", __func__, num_buffs, 598 buff_size); 599 600 ltb_size = rounddown(IBMVNIC_ONE_LTB_SIZE, buff_size); 601 tot_size = num_buffs * buff_size; 602 603 if (ltb_size > tot_size) 604 ltb_size = tot_size; 605 606 nltbs = tot_size / ltb_size; 607 if (tot_size % ltb_size) 608 nltbs++; 609 610 old_set = *ltb_set; 611 612 if (old_set.num_ltbs == nltbs) { 613 new_set = old_set; 614 } else { 615 int tmp = nltbs * sizeof(struct ibmvnic_long_term_buff); 616 617 new_set.ltbs = kzalloc(tmp, GFP_KERNEL); 618 if (!new_set.ltbs) 619 return -ENOMEM; 620 621 new_set.num_ltbs = nltbs; 622 623 /* Free any excess ltbs in old set */ 624 for (i = new_set.num_ltbs; i < old_set.num_ltbs; i++) 625 free_long_term_buff(adapter, &old_set.ltbs[i]); 626 627 /* Copy remaining ltbs to new set. All LTBs except the 628 * last one are of the same size. alloc_long_term_buff() 629 * will realloc if the size changes. 630 */ 631 n = min(old_set.num_ltbs, new_set.num_ltbs); 632 for (i = 0; i < n; i++) 633 new_set.ltbs[i] = old_set.ltbs[i]; 634 635 /* Any additional ltbs in new set will have NULL ltbs for 636 * now and will be allocated in alloc_long_term_buff(). 637 */ 638 639 /* We no longer need the old_set so free it. Note that we 640 * may have reused some ltbs from old set and freed excess 641 * ltbs above. So we only need to free the container now 642 * not the LTBs themselves. (i.e. dont free_ltb_set()!) 643 */ 644 kfree(old_set.ltbs); 645 old_set.ltbs = NULL; 646 old_set.num_ltbs = 0; 647 648 /* Install the new set. If allocations fail below, we will 649 * retry later and know what size LTBs we need. 650 */ 651 *ltb_set = new_set; 652 } 653 654 i = 0; 655 rem_size = tot_size; 656 while (rem_size) { 657 if (ltb_size > rem_size) 658 ltb_size = rem_size; 659 660 rem_size -= ltb_size; 661 662 rc = alloc_long_term_buff(adapter, &new_set.ltbs[i], ltb_size); 663 if (rc) 664 goto out; 665 i++; 666 } 667 668 WARN_ON(i != new_set.num_ltbs); 669 670 return 0; 671 out: 672 /* We may have allocated one/more LTBs before failing and we 673 * want to try and reuse on next reset. So don't free ltb set. 674 */ 675 return rc; 676 } 677 678 /** 679 * map_rxpool_buf_to_ltb - Map given rxpool buffer to offset in an LTB. 680 * @rxpool: The receive buffer pool containing buffer 681 * @bufidx: Index of buffer in rxpool 682 * @ltbp: (Output) pointer to the long term buffer containing the buffer 683 * @offset: (Output) offset of buffer in the LTB from @ltbp 684 * 685 * Map the given buffer identified by [rxpool, bufidx] to an LTB in the 686 * pool and its corresponding offset. Assume for now that each LTB is of 687 * different size but could possibly be optimized based on the allocation 688 * strategy in alloc_ltb_set(). 689 */ 690 static void map_rxpool_buf_to_ltb(struct ibmvnic_rx_pool *rxpool, 691 unsigned int bufidx, 692 struct ibmvnic_long_term_buff **ltbp, 693 unsigned int *offset) 694 { 695 struct ibmvnic_long_term_buff *ltb; 696 int nbufs; /* # of buffers in one ltb */ 697 int i; 698 699 WARN_ON(bufidx >= rxpool->size); 700 701 for (i = 0; i < rxpool->ltb_set.num_ltbs; i++) { 702 ltb = &rxpool->ltb_set.ltbs[i]; 703 nbufs = ltb->size / rxpool->buff_size; 704 if (bufidx < nbufs) 705 break; 706 bufidx -= nbufs; 707 } 708 709 *ltbp = ltb; 710 *offset = bufidx * rxpool->buff_size; 711 } 712 713 /** 714 * map_txpool_buf_to_ltb - Map given txpool buffer to offset in an LTB. 715 * @txpool: The transmit buffer pool containing buffer 716 * @bufidx: Index of buffer in txpool 717 * @ltbp: (Output) pointer to the long term buffer (LTB) containing the buffer 718 * @offset: (Output) offset of buffer in the LTB from @ltbp 719 * 720 * Map the given buffer identified by [txpool, bufidx] to an LTB in the 721 * pool and its corresponding offset. 722 */ 723 static void map_txpool_buf_to_ltb(struct ibmvnic_tx_pool *txpool, 724 unsigned int bufidx, 725 struct ibmvnic_long_term_buff **ltbp, 726 unsigned int *offset) 727 { 728 struct ibmvnic_long_term_buff *ltb; 729 int nbufs; /* # of buffers in one ltb */ 730 int i; 731 732 WARN_ON_ONCE(bufidx >= txpool->num_buffers); 733 734 for (i = 0; i < txpool->ltb_set.num_ltbs; i++) { 735 ltb = &txpool->ltb_set.ltbs[i]; 736 nbufs = ltb->size / txpool->buf_size; 737 if (bufidx < nbufs) 738 break; 739 bufidx -= nbufs; 740 } 741 742 *ltbp = ltb; 743 *offset = bufidx * txpool->buf_size; 744 } 745 746 static void deactivate_rx_pools(struct ibmvnic_adapter *adapter) 747 { 748 int i; 749 750 for (i = 0; i < adapter->num_active_rx_pools; i++) 751 adapter->rx_pool[i].active = 0; 752 } 753 754 static void replenish_rx_pool(struct ibmvnic_adapter *adapter, 755 struct ibmvnic_rx_pool *pool) 756 { 757 int count = pool->size - atomic_read(&pool->available); 758 u64 handle = adapter->rx_scrq[pool->index]->handle; 759 struct device *dev = &adapter->vdev->dev; 760 struct ibmvnic_ind_xmit_queue *ind_bufp; 761 struct ibmvnic_sub_crq_queue *rx_scrq; 762 struct ibmvnic_long_term_buff *ltb; 763 union sub_crq *sub_crq; 764 int buffers_added = 0; 765 unsigned long lpar_rc; 766 struct sk_buff *skb; 767 unsigned int offset; 768 dma_addr_t dma_addr; 769 unsigned char *dst; 770 int shift = 0; 771 int bufidx; 772 int i; 773 774 if (!pool->active) 775 return; 776 777 rx_scrq = adapter->rx_scrq[pool->index]; 778 ind_bufp = &rx_scrq->ind_buf; 779 780 /* netdev_skb_alloc() could have failed after we saved a few skbs 781 * in the indir_buf and we would not have sent them to VIOS yet. 782 * To account for them, start the loop at ind_bufp->index rather 783 * than 0. If we pushed all the skbs to VIOS, ind_bufp->index will 784 * be 0. 785 */ 786 for (i = ind_bufp->index; i < count; ++i) { 787 bufidx = pool->free_map[pool->next_free]; 788 789 /* We maybe reusing the skb from earlier resets. Allocate 790 * only if necessary. But since the LTB may have changed 791 * during reset (see init_rx_pools()), update LTB below 792 * even if reusing skb. 793 */ 794 skb = pool->rx_buff[bufidx].skb; 795 if (!skb) { 796 skb = netdev_alloc_skb(adapter->netdev, 797 pool->buff_size); 798 if (!skb) { 799 dev_err(dev, "Couldn't replenish rx buff\n"); 800 adapter->replenish_no_mem++; 801 break; 802 } 803 } 804 805 pool->free_map[pool->next_free] = IBMVNIC_INVALID_MAP; 806 pool->next_free = (pool->next_free + 1) % pool->size; 807 808 /* Copy the skb to the long term mapped DMA buffer */ 809 map_rxpool_buf_to_ltb(pool, bufidx, <b, &offset); 810 dst = ltb->buff + offset; 811 memset(dst, 0, pool->buff_size); 812 dma_addr = ltb->addr + offset; 813 814 /* add the skb to an rx_buff in the pool */ 815 pool->rx_buff[bufidx].data = dst; 816 pool->rx_buff[bufidx].dma = dma_addr; 817 pool->rx_buff[bufidx].skb = skb; 818 pool->rx_buff[bufidx].pool_index = pool->index; 819 pool->rx_buff[bufidx].size = pool->buff_size; 820 821 /* queue the rx_buff for the next send_subcrq_indirect */ 822 sub_crq = &ind_bufp->indir_arr[ind_bufp->index++]; 823 memset(sub_crq, 0, sizeof(*sub_crq)); 824 sub_crq->rx_add.first = IBMVNIC_CRQ_CMD; 825 sub_crq->rx_add.correlator = 826 cpu_to_be64((u64)&pool->rx_buff[bufidx]); 827 sub_crq->rx_add.ioba = cpu_to_be32(dma_addr); 828 sub_crq->rx_add.map_id = ltb->map_id; 829 830 /* The length field of the sCRQ is defined to be 24 bits so the 831 * buffer size needs to be left shifted by a byte before it is 832 * converted to big endian to prevent the last byte from being 833 * truncated. 834 */ 835 #ifdef __LITTLE_ENDIAN__ 836 shift = 8; 837 #endif 838 sub_crq->rx_add.len = cpu_to_be32(pool->buff_size << shift); 839 840 /* if send_subcrq_indirect queue is full, flush to VIOS */ 841 if (ind_bufp->index == IBMVNIC_MAX_IND_DESCS || 842 i == count - 1) { 843 lpar_rc = 844 send_subcrq_indirect(adapter, handle, 845 (u64)ind_bufp->indir_dma, 846 (u64)ind_bufp->index); 847 if (lpar_rc != H_SUCCESS) 848 goto failure; 849 buffers_added += ind_bufp->index; 850 adapter->replenish_add_buff_success += ind_bufp->index; 851 ind_bufp->index = 0; 852 } 853 } 854 atomic_add(buffers_added, &pool->available); 855 return; 856 857 failure: 858 if (lpar_rc != H_PARAMETER && lpar_rc != H_CLOSED) 859 dev_err_ratelimited(dev, "rx: replenish packet buffer failed\n"); 860 for (i = ind_bufp->index - 1; i >= 0; --i) { 861 struct ibmvnic_rx_buff *rx_buff; 862 863 pool->next_free = pool->next_free == 0 ? 864 pool->size - 1 : pool->next_free - 1; 865 sub_crq = &ind_bufp->indir_arr[i]; 866 rx_buff = (struct ibmvnic_rx_buff *) 867 be64_to_cpu(sub_crq->rx_add.correlator); 868 bufidx = (int)(rx_buff - pool->rx_buff); 869 pool->free_map[pool->next_free] = bufidx; 870 dev_kfree_skb_any(pool->rx_buff[bufidx].skb); 871 pool->rx_buff[bufidx].skb = NULL; 872 } 873 adapter->replenish_add_buff_failure += ind_bufp->index; 874 atomic_add(buffers_added, &pool->available); 875 ind_bufp->index = 0; 876 if (lpar_rc == H_CLOSED || adapter->failover_pending) { 877 /* Disable buffer pool replenishment and report carrier off if 878 * queue is closed or pending failover. 879 * Firmware guarantees that a signal will be sent to the 880 * driver, triggering a reset. 881 */ 882 deactivate_rx_pools(adapter); 883 netif_carrier_off(adapter->netdev); 884 } 885 } 886 887 static void replenish_pools(struct ibmvnic_adapter *adapter) 888 { 889 int i; 890 891 adapter->replenish_task_cycles++; 892 for (i = 0; i < adapter->num_active_rx_pools; i++) { 893 if (adapter->rx_pool[i].active) 894 replenish_rx_pool(adapter, &adapter->rx_pool[i]); 895 } 896 897 netdev_dbg(adapter->netdev, "Replenished %d pools\n", i); 898 } 899 900 static void release_stats_buffers(struct ibmvnic_adapter *adapter) 901 { 902 kfree(adapter->tx_stats_buffers); 903 kfree(adapter->rx_stats_buffers); 904 adapter->tx_stats_buffers = NULL; 905 adapter->rx_stats_buffers = NULL; 906 } 907 908 static int init_stats_buffers(struct ibmvnic_adapter *adapter) 909 { 910 adapter->tx_stats_buffers = 911 kcalloc(IBMVNIC_MAX_QUEUES, 912 sizeof(struct ibmvnic_tx_queue_stats), 913 GFP_KERNEL); 914 if (!adapter->tx_stats_buffers) 915 return -ENOMEM; 916 917 adapter->rx_stats_buffers = 918 kcalloc(IBMVNIC_MAX_QUEUES, 919 sizeof(struct ibmvnic_rx_queue_stats), 920 GFP_KERNEL); 921 if (!adapter->rx_stats_buffers) 922 return -ENOMEM; 923 924 return 0; 925 } 926 927 static void release_stats_token(struct ibmvnic_adapter *adapter) 928 { 929 struct device *dev = &adapter->vdev->dev; 930 931 if (!adapter->stats_token) 932 return; 933 934 dma_unmap_single(dev, adapter->stats_token, 935 sizeof(struct ibmvnic_statistics), 936 DMA_FROM_DEVICE); 937 adapter->stats_token = 0; 938 } 939 940 static int init_stats_token(struct ibmvnic_adapter *adapter) 941 { 942 struct device *dev = &adapter->vdev->dev; 943 dma_addr_t stok; 944 int rc; 945 946 stok = dma_map_single(dev, &adapter->stats, 947 sizeof(struct ibmvnic_statistics), 948 DMA_FROM_DEVICE); 949 rc = dma_mapping_error(dev, stok); 950 if (rc) { 951 dev_err(dev, "Couldn't map stats buffer, rc = %d\n", rc); 952 return rc; 953 } 954 955 adapter->stats_token = stok; 956 netdev_dbg(adapter->netdev, "Stats token initialized (%llx)\n", stok); 957 return 0; 958 } 959 960 /** 961 * release_rx_pools() - Release any rx pools attached to @adapter. 962 * @adapter: ibmvnic adapter 963 * 964 * Safe to call this multiple times - even if no pools are attached. 965 */ 966 static void release_rx_pools(struct ibmvnic_adapter *adapter) 967 { 968 struct ibmvnic_rx_pool *rx_pool; 969 int i, j; 970 971 if (!adapter->rx_pool) 972 return; 973 974 for (i = 0; i < adapter->num_active_rx_pools; i++) { 975 rx_pool = &adapter->rx_pool[i]; 976 977 netdev_dbg(adapter->netdev, "Releasing rx_pool[%d]\n", i); 978 979 kfree(rx_pool->free_map); 980 981 free_ltb_set(adapter, &rx_pool->ltb_set); 982 983 if (!rx_pool->rx_buff) 984 continue; 985 986 for (j = 0; j < rx_pool->size; j++) { 987 if (rx_pool->rx_buff[j].skb) { 988 dev_kfree_skb_any(rx_pool->rx_buff[j].skb); 989 rx_pool->rx_buff[j].skb = NULL; 990 } 991 } 992 993 kfree(rx_pool->rx_buff); 994 } 995 996 kfree(adapter->rx_pool); 997 adapter->rx_pool = NULL; 998 adapter->num_active_rx_pools = 0; 999 adapter->prev_rx_pool_size = 0; 1000 } 1001 1002 /** 1003 * reuse_rx_pools() - Check if the existing rx pools can be reused. 1004 * @adapter: ibmvnic adapter 1005 * 1006 * Check if the existing rx pools in the adapter can be reused. The 1007 * pools can be reused if the pool parameters (number of pools, 1008 * number of buffers in the pool and size of each buffer) have not 1009 * changed. 1010 * 1011 * NOTE: This assumes that all pools have the same number of buffers 1012 * which is the case currently. If that changes, we must fix this. 1013 * 1014 * Return: true if the rx pools can be reused, false otherwise. 1015 */ 1016 static bool reuse_rx_pools(struct ibmvnic_adapter *adapter) 1017 { 1018 u64 old_num_pools, new_num_pools; 1019 u64 old_pool_size, new_pool_size; 1020 u64 old_buff_size, new_buff_size; 1021 1022 if (!adapter->rx_pool) 1023 return false; 1024 1025 old_num_pools = adapter->num_active_rx_pools; 1026 new_num_pools = adapter->req_rx_queues; 1027 1028 old_pool_size = adapter->prev_rx_pool_size; 1029 new_pool_size = adapter->req_rx_add_entries_per_subcrq; 1030 1031 old_buff_size = adapter->prev_rx_buf_sz; 1032 new_buff_size = adapter->cur_rx_buf_sz; 1033 1034 if (old_buff_size != new_buff_size || 1035 old_num_pools != new_num_pools || 1036 old_pool_size != new_pool_size) 1037 return false; 1038 1039 return true; 1040 } 1041 1042 /** 1043 * init_rx_pools(): Initialize the set of receiver pools in the adapter. 1044 * @netdev: net device associated with the vnic interface 1045 * 1046 * Initialize the set of receiver pools in the ibmvnic adapter associated 1047 * with the net_device @netdev. If possible, reuse the existing rx pools. 1048 * Otherwise free any existing pools and allocate a new set of pools 1049 * before initializing them. 1050 * 1051 * Return: 0 on success and negative value on error. 1052 */ 1053 static int init_rx_pools(struct net_device *netdev) 1054 { 1055 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 1056 struct device *dev = &adapter->vdev->dev; 1057 struct ibmvnic_rx_pool *rx_pool; 1058 u64 num_pools; 1059 u64 pool_size; /* # of buffers in one pool */ 1060 u64 buff_size; 1061 int i, j, rc; 1062 1063 pool_size = adapter->req_rx_add_entries_per_subcrq; 1064 num_pools = adapter->req_rx_queues; 1065 buff_size = adapter->cur_rx_buf_sz; 1066 1067 if (reuse_rx_pools(adapter)) { 1068 dev_dbg(dev, "Reusing rx pools\n"); 1069 goto update_ltb; 1070 } 1071 1072 /* Allocate/populate the pools. */ 1073 release_rx_pools(adapter); 1074 1075 adapter->rx_pool = kcalloc(num_pools, 1076 sizeof(struct ibmvnic_rx_pool), 1077 GFP_KERNEL); 1078 if (!adapter->rx_pool) { 1079 dev_err(dev, "Failed to allocate rx pools\n"); 1080 return -ENOMEM; 1081 } 1082 1083 /* Set num_active_rx_pools early. If we fail below after partial 1084 * allocation, release_rx_pools() will know how many to look for. 1085 */ 1086 adapter->num_active_rx_pools = num_pools; 1087 1088 for (i = 0; i < num_pools; i++) { 1089 rx_pool = &adapter->rx_pool[i]; 1090 1091 netdev_dbg(adapter->netdev, 1092 "Initializing rx_pool[%d], %lld buffs, %lld bytes each\n", 1093 i, pool_size, buff_size); 1094 1095 rx_pool->size = pool_size; 1096 rx_pool->index = i; 1097 rx_pool->buff_size = ALIGN(buff_size, L1_CACHE_BYTES); 1098 1099 rx_pool->free_map = kcalloc(rx_pool->size, sizeof(int), 1100 GFP_KERNEL); 1101 if (!rx_pool->free_map) { 1102 dev_err(dev, "Couldn't alloc free_map %d\n", i); 1103 rc = -ENOMEM; 1104 goto out_release; 1105 } 1106 1107 rx_pool->rx_buff = kcalloc(rx_pool->size, 1108 sizeof(struct ibmvnic_rx_buff), 1109 GFP_KERNEL); 1110 if (!rx_pool->rx_buff) { 1111 dev_err(dev, "Couldn't alloc rx buffers\n"); 1112 rc = -ENOMEM; 1113 goto out_release; 1114 } 1115 } 1116 1117 adapter->prev_rx_pool_size = pool_size; 1118 adapter->prev_rx_buf_sz = adapter->cur_rx_buf_sz; 1119 1120 update_ltb: 1121 for (i = 0; i < num_pools; i++) { 1122 rx_pool = &adapter->rx_pool[i]; 1123 dev_dbg(dev, "Updating LTB for rx pool %d [%d, %d]\n", 1124 i, rx_pool->size, rx_pool->buff_size); 1125 1126 rc = alloc_ltb_set(adapter, &rx_pool->ltb_set, 1127 rx_pool->size, rx_pool->buff_size); 1128 if (rc) 1129 goto out; 1130 1131 for (j = 0; j < rx_pool->size; ++j) { 1132 struct ibmvnic_rx_buff *rx_buff; 1133 1134 rx_pool->free_map[j] = j; 1135 1136 /* NOTE: Don't clear rx_buff->skb here - will leak 1137 * memory! replenish_rx_pool() will reuse skbs or 1138 * allocate as necessary. 1139 */ 1140 rx_buff = &rx_pool->rx_buff[j]; 1141 rx_buff->dma = 0; 1142 rx_buff->data = 0; 1143 rx_buff->size = 0; 1144 rx_buff->pool_index = 0; 1145 } 1146 1147 /* Mark pool "empty" so replenish_rx_pools() will 1148 * update the LTB info for each buffer 1149 */ 1150 atomic_set(&rx_pool->available, 0); 1151 rx_pool->next_alloc = 0; 1152 rx_pool->next_free = 0; 1153 /* replenish_rx_pool() may have called deactivate_rx_pools() 1154 * on failover. Ensure pool is active now. 1155 */ 1156 rx_pool->active = 1; 1157 } 1158 return 0; 1159 out_release: 1160 release_rx_pools(adapter); 1161 out: 1162 /* We failed to allocate one or more LTBs or map them on the VIOS. 1163 * Hold onto the pools and any LTBs that we did allocate/map. 1164 */ 1165 return rc; 1166 } 1167 1168 static void release_vpd_data(struct ibmvnic_adapter *adapter) 1169 { 1170 if (!adapter->vpd) 1171 return; 1172 1173 kfree(adapter->vpd->buff); 1174 kfree(adapter->vpd); 1175 1176 adapter->vpd = NULL; 1177 } 1178 1179 static void release_one_tx_pool(struct ibmvnic_adapter *adapter, 1180 struct ibmvnic_tx_pool *tx_pool) 1181 { 1182 kfree(tx_pool->tx_buff); 1183 kfree(tx_pool->free_map); 1184 free_ltb_set(adapter, &tx_pool->ltb_set); 1185 } 1186 1187 /** 1188 * release_tx_pools() - Release any tx pools attached to @adapter. 1189 * @adapter: ibmvnic adapter 1190 * 1191 * Safe to call this multiple times - even if no pools are attached. 1192 */ 1193 static void release_tx_pools(struct ibmvnic_adapter *adapter) 1194 { 1195 int i; 1196 1197 /* init_tx_pools() ensures that ->tx_pool and ->tso_pool are 1198 * both NULL or both non-NULL. So we only need to check one. 1199 */ 1200 if (!adapter->tx_pool) 1201 return; 1202 1203 for (i = 0; i < adapter->num_active_tx_pools; i++) { 1204 release_one_tx_pool(adapter, &adapter->tx_pool[i]); 1205 release_one_tx_pool(adapter, &adapter->tso_pool[i]); 1206 } 1207 1208 kfree(adapter->tx_pool); 1209 adapter->tx_pool = NULL; 1210 kfree(adapter->tso_pool); 1211 adapter->tso_pool = NULL; 1212 adapter->num_active_tx_pools = 0; 1213 adapter->prev_tx_pool_size = 0; 1214 } 1215 1216 static int init_one_tx_pool(struct net_device *netdev, 1217 struct ibmvnic_tx_pool *tx_pool, 1218 int pool_size, int buf_size) 1219 { 1220 int i; 1221 1222 tx_pool->tx_buff = kcalloc(pool_size, 1223 sizeof(struct ibmvnic_tx_buff), 1224 GFP_KERNEL); 1225 if (!tx_pool->tx_buff) 1226 return -ENOMEM; 1227 1228 tx_pool->free_map = kcalloc(pool_size, sizeof(int), GFP_KERNEL); 1229 if (!tx_pool->free_map) { 1230 kfree(tx_pool->tx_buff); 1231 tx_pool->tx_buff = NULL; 1232 return -ENOMEM; 1233 } 1234 1235 for (i = 0; i < pool_size; i++) 1236 tx_pool->free_map[i] = i; 1237 1238 tx_pool->consumer_index = 0; 1239 tx_pool->producer_index = 0; 1240 tx_pool->num_buffers = pool_size; 1241 tx_pool->buf_size = buf_size; 1242 1243 return 0; 1244 } 1245 1246 /** 1247 * reuse_tx_pools() - Check if the existing tx pools can be reused. 1248 * @adapter: ibmvnic adapter 1249 * 1250 * Check if the existing tx pools in the adapter can be reused. The 1251 * pools can be reused if the pool parameters (number of pools, 1252 * number of buffers in the pool and mtu) have not changed. 1253 * 1254 * NOTE: This assumes that all pools have the same number of buffers 1255 * which is the case currently. If that changes, we must fix this. 1256 * 1257 * Return: true if the tx pools can be reused, false otherwise. 1258 */ 1259 static bool reuse_tx_pools(struct ibmvnic_adapter *adapter) 1260 { 1261 u64 old_num_pools, new_num_pools; 1262 u64 old_pool_size, new_pool_size; 1263 u64 old_mtu, new_mtu; 1264 1265 if (!adapter->tx_pool) 1266 return false; 1267 1268 old_num_pools = adapter->num_active_tx_pools; 1269 new_num_pools = adapter->num_active_tx_scrqs; 1270 old_pool_size = adapter->prev_tx_pool_size; 1271 new_pool_size = adapter->req_tx_entries_per_subcrq; 1272 old_mtu = adapter->prev_mtu; 1273 new_mtu = adapter->req_mtu; 1274 1275 if (old_mtu != new_mtu || 1276 old_num_pools != new_num_pools || 1277 old_pool_size != new_pool_size) 1278 return false; 1279 1280 return true; 1281 } 1282 1283 /** 1284 * init_tx_pools(): Initialize the set of transmit pools in the adapter. 1285 * @netdev: net device associated with the vnic interface 1286 * 1287 * Initialize the set of transmit pools in the ibmvnic adapter associated 1288 * with the net_device @netdev. If possible, reuse the existing tx pools. 1289 * Otherwise free any existing pools and allocate a new set of pools 1290 * before initializing them. 1291 * 1292 * Return: 0 on success and negative value on error. 1293 */ 1294 static int init_tx_pools(struct net_device *netdev) 1295 { 1296 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 1297 struct device *dev = &adapter->vdev->dev; 1298 int num_pools; 1299 u64 pool_size; /* # of buffers in pool */ 1300 u64 buff_size; 1301 int i, j, rc; 1302 1303 num_pools = adapter->req_tx_queues; 1304 1305 /* We must notify the VIOS about the LTB on all resets - but we only 1306 * need to alloc/populate pools if either the number of buffers or 1307 * size of each buffer in the pool has changed. 1308 */ 1309 if (reuse_tx_pools(adapter)) { 1310 netdev_dbg(netdev, "Reusing tx pools\n"); 1311 goto update_ltb; 1312 } 1313 1314 /* Allocate/populate the pools. */ 1315 release_tx_pools(adapter); 1316 1317 pool_size = adapter->req_tx_entries_per_subcrq; 1318 num_pools = adapter->num_active_tx_scrqs; 1319 1320 adapter->tx_pool = kcalloc(num_pools, 1321 sizeof(struct ibmvnic_tx_pool), GFP_KERNEL); 1322 if (!adapter->tx_pool) 1323 return -ENOMEM; 1324 1325 adapter->tso_pool = kcalloc(num_pools, 1326 sizeof(struct ibmvnic_tx_pool), GFP_KERNEL); 1327 /* To simplify release_tx_pools() ensure that ->tx_pool and 1328 * ->tso_pool are either both NULL or both non-NULL. 1329 */ 1330 if (!adapter->tso_pool) { 1331 kfree(adapter->tx_pool); 1332 adapter->tx_pool = NULL; 1333 return -ENOMEM; 1334 } 1335 1336 /* Set num_active_tx_pools early. If we fail below after partial 1337 * allocation, release_tx_pools() will know how many to look for. 1338 */ 1339 adapter->num_active_tx_pools = num_pools; 1340 1341 buff_size = adapter->req_mtu + VLAN_HLEN; 1342 buff_size = ALIGN(buff_size, L1_CACHE_BYTES); 1343 1344 for (i = 0; i < num_pools; i++) { 1345 dev_dbg(dev, "Init tx pool %d [%llu, %llu]\n", 1346 i, adapter->req_tx_entries_per_subcrq, buff_size); 1347 1348 rc = init_one_tx_pool(netdev, &adapter->tx_pool[i], 1349 pool_size, buff_size); 1350 if (rc) 1351 goto out_release; 1352 1353 rc = init_one_tx_pool(netdev, &adapter->tso_pool[i], 1354 IBMVNIC_TSO_BUFS, 1355 IBMVNIC_TSO_BUF_SZ); 1356 if (rc) 1357 goto out_release; 1358 } 1359 1360 adapter->prev_tx_pool_size = pool_size; 1361 adapter->prev_mtu = adapter->req_mtu; 1362 1363 update_ltb: 1364 /* NOTE: All tx_pools have the same number of buffers (which is 1365 * same as pool_size). All tso_pools have IBMVNIC_TSO_BUFS 1366 * buffers (see calls init_one_tx_pool() for these). 1367 * For consistency, we use tx_pool->num_buffers and 1368 * tso_pool->num_buffers below. 1369 */ 1370 rc = -1; 1371 for (i = 0; i < num_pools; i++) { 1372 struct ibmvnic_tx_pool *tso_pool; 1373 struct ibmvnic_tx_pool *tx_pool; 1374 1375 tx_pool = &adapter->tx_pool[i]; 1376 1377 dev_dbg(dev, "Updating LTB for tx pool %d [%d, %d]\n", 1378 i, tx_pool->num_buffers, tx_pool->buf_size); 1379 1380 rc = alloc_ltb_set(adapter, &tx_pool->ltb_set, 1381 tx_pool->num_buffers, tx_pool->buf_size); 1382 if (rc) 1383 goto out; 1384 1385 tx_pool->consumer_index = 0; 1386 tx_pool->producer_index = 0; 1387 1388 for (j = 0; j < tx_pool->num_buffers; j++) 1389 tx_pool->free_map[j] = j; 1390 1391 tso_pool = &adapter->tso_pool[i]; 1392 1393 dev_dbg(dev, "Updating LTB for tso pool %d [%d, %d]\n", 1394 i, tso_pool->num_buffers, tso_pool->buf_size); 1395 1396 rc = alloc_ltb_set(adapter, &tso_pool->ltb_set, 1397 tso_pool->num_buffers, tso_pool->buf_size); 1398 if (rc) 1399 goto out; 1400 1401 tso_pool->consumer_index = 0; 1402 tso_pool->producer_index = 0; 1403 1404 for (j = 0; j < tso_pool->num_buffers; j++) 1405 tso_pool->free_map[j] = j; 1406 } 1407 1408 return 0; 1409 out_release: 1410 release_tx_pools(adapter); 1411 out: 1412 /* We failed to allocate one or more LTBs or map them on the VIOS. 1413 * Hold onto the pools and any LTBs that we did allocate/map. 1414 */ 1415 return rc; 1416 } 1417 1418 static void ibmvnic_napi_enable(struct ibmvnic_adapter *adapter) 1419 { 1420 int i; 1421 1422 if (adapter->napi_enabled) 1423 return; 1424 1425 for (i = 0; i < adapter->req_rx_queues; i++) 1426 napi_enable(&adapter->napi[i]); 1427 1428 adapter->napi_enabled = true; 1429 } 1430 1431 static void ibmvnic_napi_disable(struct ibmvnic_adapter *adapter) 1432 { 1433 int i; 1434 1435 if (!adapter->napi_enabled) 1436 return; 1437 1438 for (i = 0; i < adapter->req_rx_queues; i++) { 1439 netdev_dbg(adapter->netdev, "Disabling napi[%d]\n", i); 1440 napi_disable(&adapter->napi[i]); 1441 } 1442 1443 adapter->napi_enabled = false; 1444 } 1445 1446 static int init_napi(struct ibmvnic_adapter *adapter) 1447 { 1448 int i; 1449 1450 adapter->napi = kcalloc(adapter->req_rx_queues, 1451 sizeof(struct napi_struct), GFP_KERNEL); 1452 if (!adapter->napi) 1453 return -ENOMEM; 1454 1455 for (i = 0; i < adapter->req_rx_queues; i++) { 1456 netdev_dbg(adapter->netdev, "Adding napi[%d]\n", i); 1457 netif_napi_add(adapter->netdev, &adapter->napi[i], 1458 ibmvnic_poll); 1459 } 1460 1461 adapter->num_active_rx_napi = adapter->req_rx_queues; 1462 return 0; 1463 } 1464 1465 static void release_napi(struct ibmvnic_adapter *adapter) 1466 { 1467 int i; 1468 1469 if (!adapter->napi) 1470 return; 1471 1472 for (i = 0; i < adapter->num_active_rx_napi; i++) { 1473 netdev_dbg(adapter->netdev, "Releasing napi[%d]\n", i); 1474 netif_napi_del(&adapter->napi[i]); 1475 } 1476 1477 kfree(adapter->napi); 1478 adapter->napi = NULL; 1479 adapter->num_active_rx_napi = 0; 1480 adapter->napi_enabled = false; 1481 } 1482 1483 static const char *adapter_state_to_string(enum vnic_state state) 1484 { 1485 switch (state) { 1486 case VNIC_PROBING: 1487 return "PROBING"; 1488 case VNIC_PROBED: 1489 return "PROBED"; 1490 case VNIC_OPENING: 1491 return "OPENING"; 1492 case VNIC_OPEN: 1493 return "OPEN"; 1494 case VNIC_CLOSING: 1495 return "CLOSING"; 1496 case VNIC_CLOSED: 1497 return "CLOSED"; 1498 case VNIC_REMOVING: 1499 return "REMOVING"; 1500 case VNIC_REMOVED: 1501 return "REMOVED"; 1502 case VNIC_DOWN: 1503 return "DOWN"; 1504 } 1505 return "UNKNOWN"; 1506 } 1507 1508 static int ibmvnic_login(struct net_device *netdev) 1509 { 1510 unsigned long flags, timeout = msecs_to_jiffies(20000); 1511 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 1512 int retry_count = 0; 1513 int retries = 10; 1514 bool retry; 1515 int rc; 1516 1517 do { 1518 retry = false; 1519 if (retry_count > retries) { 1520 netdev_warn(netdev, "Login attempts exceeded\n"); 1521 return -EACCES; 1522 } 1523 1524 adapter->init_done_rc = 0; 1525 reinit_completion(&adapter->init_done); 1526 rc = send_login(adapter); 1527 if (rc) 1528 return rc; 1529 1530 if (!wait_for_completion_timeout(&adapter->init_done, 1531 timeout)) { 1532 netdev_warn(netdev, "Login timed out\n"); 1533 adapter->login_pending = false; 1534 goto partial_reset; 1535 } 1536 1537 if (adapter->init_done_rc == ABORTED) { 1538 netdev_warn(netdev, "Login aborted, retrying...\n"); 1539 retry = true; 1540 adapter->init_done_rc = 0; 1541 retry_count++; 1542 /* FW or device may be busy, so 1543 * wait a bit before retrying login 1544 */ 1545 msleep(500); 1546 } else if (adapter->init_done_rc == PARTIALSUCCESS) { 1547 retry_count++; 1548 release_sub_crqs(adapter, 1); 1549 1550 retry = true; 1551 netdev_dbg(netdev, 1552 "Received partial success, retrying...\n"); 1553 adapter->init_done_rc = 0; 1554 reinit_completion(&adapter->init_done); 1555 send_query_cap(adapter); 1556 if (!wait_for_completion_timeout(&adapter->init_done, 1557 timeout)) { 1558 netdev_warn(netdev, 1559 "Capabilities query timed out\n"); 1560 return -ETIMEDOUT; 1561 } 1562 1563 rc = init_sub_crqs(adapter); 1564 if (rc) { 1565 netdev_warn(netdev, 1566 "SCRQ initialization failed\n"); 1567 return rc; 1568 } 1569 1570 rc = init_sub_crq_irqs(adapter); 1571 if (rc) { 1572 netdev_warn(netdev, 1573 "SCRQ irq initialization failed\n"); 1574 return rc; 1575 } 1576 /* Default/timeout error handling, reset and start fresh */ 1577 } else if (adapter->init_done_rc) { 1578 netdev_warn(netdev, "Adapter login failed, init_done_rc = %d\n", 1579 adapter->init_done_rc); 1580 1581 partial_reset: 1582 /* adapter login failed, so free any CRQs or sub-CRQs 1583 * and register again before attempting to login again. 1584 * If we don't do this then the VIOS may think that 1585 * we are already logged in and reject any subsequent 1586 * attempts 1587 */ 1588 netdev_warn(netdev, 1589 "Freeing and re-registering CRQs before attempting to login again\n"); 1590 retry = true; 1591 adapter->init_done_rc = 0; 1592 release_sub_crqs(adapter, true); 1593 /* Much of this is similar logic as ibmvnic_probe(), 1594 * we are essentially re-initializing communication 1595 * with the server. We really should not run any 1596 * resets/failovers here because this is already a form 1597 * of reset and we do not want parallel resets occurring 1598 */ 1599 do { 1600 reinit_init_done(adapter); 1601 /* Clear any failovers we got in the previous 1602 * pass since we are re-initializing the CRQ 1603 */ 1604 adapter->failover_pending = false; 1605 release_crq_queue(adapter); 1606 /* If we don't sleep here then we risk an 1607 * unnecessary failover event from the VIOS. 1608 * This is a known VIOS issue caused by a vnic 1609 * device freeing and registering a CRQ too 1610 * quickly. 1611 */ 1612 msleep(1500); 1613 /* Avoid any resets, since we are currently 1614 * resetting. 1615 */ 1616 spin_lock_irqsave(&adapter->rwi_lock, flags); 1617 flush_reset_queue(adapter); 1618 spin_unlock_irqrestore(&adapter->rwi_lock, 1619 flags); 1620 1621 rc = init_crq_queue(adapter); 1622 if (rc) { 1623 netdev_err(netdev, "login recovery: init CRQ failed %d\n", 1624 rc); 1625 return -EIO; 1626 } 1627 1628 rc = ibmvnic_reset_init(adapter, false); 1629 if (rc) 1630 netdev_err(netdev, "login recovery: Reset init failed %d\n", 1631 rc); 1632 /* IBMVNIC_CRQ_INIT will return EAGAIN if it 1633 * fails, since ibmvnic_reset_init will free 1634 * irq's in failure, we won't be able to receive 1635 * new CRQs so we need to keep trying. probe() 1636 * handles this similarly. 1637 */ 1638 } while (rc == -EAGAIN && retry_count++ < retries); 1639 } 1640 } while (retry); 1641 1642 __ibmvnic_set_mac(netdev, adapter->mac_addr); 1643 1644 netdev_dbg(netdev, "[S:%s] Login succeeded\n", adapter_state_to_string(adapter->state)); 1645 return 0; 1646 } 1647 1648 static void release_login_buffer(struct ibmvnic_adapter *adapter) 1649 { 1650 if (!adapter->login_buf) 1651 return; 1652 1653 dma_unmap_single(&adapter->vdev->dev, adapter->login_buf_token, 1654 adapter->login_buf_sz, DMA_TO_DEVICE); 1655 kfree(adapter->login_buf); 1656 adapter->login_buf = NULL; 1657 } 1658 1659 static void release_login_rsp_buffer(struct ibmvnic_adapter *adapter) 1660 { 1661 if (!adapter->login_rsp_buf) 1662 return; 1663 1664 dma_unmap_single(&adapter->vdev->dev, adapter->login_rsp_buf_token, 1665 adapter->login_rsp_buf_sz, DMA_FROM_DEVICE); 1666 kfree(adapter->login_rsp_buf); 1667 adapter->login_rsp_buf = NULL; 1668 } 1669 1670 static void release_resources(struct ibmvnic_adapter *adapter) 1671 { 1672 release_vpd_data(adapter); 1673 1674 release_napi(adapter); 1675 release_login_buffer(adapter); 1676 release_login_rsp_buffer(adapter); 1677 } 1678 1679 static int set_link_state(struct ibmvnic_adapter *adapter, u8 link_state) 1680 { 1681 struct net_device *netdev = adapter->netdev; 1682 unsigned long timeout = msecs_to_jiffies(20000); 1683 union ibmvnic_crq crq; 1684 bool resend; 1685 int rc; 1686 1687 netdev_dbg(netdev, "setting link state %d\n", link_state); 1688 1689 memset(&crq, 0, sizeof(crq)); 1690 crq.logical_link_state.first = IBMVNIC_CRQ_CMD; 1691 crq.logical_link_state.cmd = LOGICAL_LINK_STATE; 1692 crq.logical_link_state.link_state = link_state; 1693 1694 do { 1695 resend = false; 1696 1697 reinit_completion(&adapter->init_done); 1698 rc = ibmvnic_send_crq(adapter, &crq); 1699 if (rc) { 1700 netdev_err(netdev, "Failed to set link state\n"); 1701 return rc; 1702 } 1703 1704 if (!wait_for_completion_timeout(&adapter->init_done, 1705 timeout)) { 1706 netdev_err(netdev, "timeout setting link state\n"); 1707 return -ETIMEDOUT; 1708 } 1709 1710 if (adapter->init_done_rc == PARTIALSUCCESS) { 1711 /* Partuial success, delay and re-send */ 1712 mdelay(1000); 1713 resend = true; 1714 } else if (adapter->init_done_rc) { 1715 netdev_warn(netdev, "Unable to set link state, rc=%d\n", 1716 adapter->init_done_rc); 1717 return adapter->init_done_rc; 1718 } 1719 } while (resend); 1720 1721 return 0; 1722 } 1723 1724 static int set_real_num_queues(struct net_device *netdev) 1725 { 1726 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 1727 int rc; 1728 1729 netdev_dbg(netdev, "Setting real tx/rx queues (%llx/%llx)\n", 1730 adapter->req_tx_queues, adapter->req_rx_queues); 1731 1732 rc = netif_set_real_num_tx_queues(netdev, adapter->req_tx_queues); 1733 if (rc) { 1734 netdev_err(netdev, "failed to set the number of tx queues\n"); 1735 return rc; 1736 } 1737 1738 rc = netif_set_real_num_rx_queues(netdev, adapter->req_rx_queues); 1739 if (rc) 1740 netdev_err(netdev, "failed to set the number of rx queues\n"); 1741 1742 return rc; 1743 } 1744 1745 static int ibmvnic_get_vpd(struct ibmvnic_adapter *adapter) 1746 { 1747 struct device *dev = &adapter->vdev->dev; 1748 union ibmvnic_crq crq; 1749 int len = 0; 1750 int rc; 1751 1752 if (adapter->vpd->buff) 1753 len = adapter->vpd->len; 1754 1755 mutex_lock(&adapter->fw_lock); 1756 adapter->fw_done_rc = 0; 1757 reinit_completion(&adapter->fw_done); 1758 1759 crq.get_vpd_size.first = IBMVNIC_CRQ_CMD; 1760 crq.get_vpd_size.cmd = GET_VPD_SIZE; 1761 rc = ibmvnic_send_crq(adapter, &crq); 1762 if (rc) { 1763 mutex_unlock(&adapter->fw_lock); 1764 return rc; 1765 } 1766 1767 rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000); 1768 if (rc) { 1769 dev_err(dev, "Could not retrieve VPD size, rc = %d\n", rc); 1770 mutex_unlock(&adapter->fw_lock); 1771 return rc; 1772 } 1773 mutex_unlock(&adapter->fw_lock); 1774 1775 if (!adapter->vpd->len) 1776 return -ENODATA; 1777 1778 if (!adapter->vpd->buff) 1779 adapter->vpd->buff = kzalloc(adapter->vpd->len, GFP_KERNEL); 1780 else if (adapter->vpd->len != len) 1781 adapter->vpd->buff = 1782 krealloc(adapter->vpd->buff, 1783 adapter->vpd->len, GFP_KERNEL); 1784 1785 if (!adapter->vpd->buff) { 1786 dev_err(dev, "Could allocate VPD buffer\n"); 1787 return -ENOMEM; 1788 } 1789 1790 adapter->vpd->dma_addr = 1791 dma_map_single(dev, adapter->vpd->buff, adapter->vpd->len, 1792 DMA_FROM_DEVICE); 1793 if (dma_mapping_error(dev, adapter->vpd->dma_addr)) { 1794 dev_err(dev, "Could not map VPD buffer\n"); 1795 kfree(adapter->vpd->buff); 1796 adapter->vpd->buff = NULL; 1797 return -ENOMEM; 1798 } 1799 1800 mutex_lock(&adapter->fw_lock); 1801 adapter->fw_done_rc = 0; 1802 reinit_completion(&adapter->fw_done); 1803 1804 crq.get_vpd.first = IBMVNIC_CRQ_CMD; 1805 crq.get_vpd.cmd = GET_VPD; 1806 crq.get_vpd.ioba = cpu_to_be32(adapter->vpd->dma_addr); 1807 crq.get_vpd.len = cpu_to_be32((u32)adapter->vpd->len); 1808 rc = ibmvnic_send_crq(adapter, &crq); 1809 if (rc) { 1810 kfree(adapter->vpd->buff); 1811 adapter->vpd->buff = NULL; 1812 mutex_unlock(&adapter->fw_lock); 1813 return rc; 1814 } 1815 1816 rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000); 1817 if (rc) { 1818 dev_err(dev, "Unable to retrieve VPD, rc = %d\n", rc); 1819 kfree(adapter->vpd->buff); 1820 adapter->vpd->buff = NULL; 1821 mutex_unlock(&adapter->fw_lock); 1822 return rc; 1823 } 1824 1825 mutex_unlock(&adapter->fw_lock); 1826 return 0; 1827 } 1828 1829 static int init_resources(struct ibmvnic_adapter *adapter) 1830 { 1831 struct net_device *netdev = adapter->netdev; 1832 int rc; 1833 1834 rc = set_real_num_queues(netdev); 1835 if (rc) 1836 return rc; 1837 1838 adapter->vpd = kzalloc(sizeof(*adapter->vpd), GFP_KERNEL); 1839 if (!adapter->vpd) 1840 return -ENOMEM; 1841 1842 /* Vital Product Data (VPD) */ 1843 rc = ibmvnic_get_vpd(adapter); 1844 if (rc) { 1845 netdev_err(netdev, "failed to initialize Vital Product Data (VPD)\n"); 1846 return rc; 1847 } 1848 1849 rc = init_napi(adapter); 1850 if (rc) 1851 return rc; 1852 1853 send_query_map(adapter); 1854 1855 rc = init_rx_pools(netdev); 1856 if (rc) 1857 return rc; 1858 1859 rc = init_tx_pools(netdev); 1860 return rc; 1861 } 1862 1863 static int __ibmvnic_open(struct net_device *netdev) 1864 { 1865 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 1866 enum vnic_state prev_state = adapter->state; 1867 int i, rc; 1868 1869 adapter->state = VNIC_OPENING; 1870 replenish_pools(adapter); 1871 ibmvnic_napi_enable(adapter); 1872 1873 /* We're ready to receive frames, enable the sub-crq interrupts and 1874 * set the logical link state to up 1875 */ 1876 for (i = 0; i < adapter->req_rx_queues; i++) { 1877 netdev_dbg(netdev, "Enabling rx_scrq[%d] irq\n", i); 1878 if (prev_state == VNIC_CLOSED) 1879 enable_irq(adapter->rx_scrq[i]->irq); 1880 enable_scrq_irq(adapter, adapter->rx_scrq[i]); 1881 } 1882 1883 for (i = 0; i < adapter->req_tx_queues; i++) { 1884 netdev_dbg(netdev, "Enabling tx_scrq[%d] irq\n", i); 1885 if (prev_state == VNIC_CLOSED) 1886 enable_irq(adapter->tx_scrq[i]->irq); 1887 enable_scrq_irq(adapter, adapter->tx_scrq[i]); 1888 /* netdev_tx_reset_queue will reset dql stats. During NON_FATAL 1889 * resets, don't reset the stats because there could be batched 1890 * skb's waiting to be sent. If we reset dql stats, we risk 1891 * num_completed being greater than num_queued. This will cause 1892 * a BUG_ON in dql_completed(). 1893 */ 1894 if (adapter->reset_reason != VNIC_RESET_NON_FATAL) 1895 netdev_tx_reset_queue(netdev_get_tx_queue(netdev, i)); 1896 } 1897 1898 rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_UP); 1899 if (rc) { 1900 ibmvnic_napi_disable(adapter); 1901 ibmvnic_disable_irqs(adapter); 1902 return rc; 1903 } 1904 1905 adapter->tx_queues_active = true; 1906 1907 /* Since queues were stopped until now, there shouldn't be any 1908 * one in ibmvnic_complete_tx() or ibmvnic_xmit() so maybe we 1909 * don't need the synchronize_rcu()? Leaving it for consistency 1910 * with setting ->tx_queues_active = false. 1911 */ 1912 synchronize_rcu(); 1913 1914 netif_tx_start_all_queues(netdev); 1915 1916 if (prev_state == VNIC_CLOSED) { 1917 for (i = 0; i < adapter->req_rx_queues; i++) 1918 napi_schedule(&adapter->napi[i]); 1919 } 1920 1921 adapter->state = VNIC_OPEN; 1922 return rc; 1923 } 1924 1925 static int ibmvnic_open(struct net_device *netdev) 1926 { 1927 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 1928 int rc; 1929 1930 ASSERT_RTNL(); 1931 1932 /* If device failover is pending or we are about to reset, just set 1933 * device state and return. Device operation will be handled by reset 1934 * routine. 1935 * 1936 * It should be safe to overwrite the adapter->state here. Since 1937 * we hold the rtnl, either the reset has not actually started or 1938 * the rtnl got dropped during the set_link_state() in do_reset(). 1939 * In the former case, no one else is changing the state (again we 1940 * have the rtnl) and in the latter case, do_reset() will detect and 1941 * honor our setting below. 1942 */ 1943 if (adapter->failover_pending || (test_bit(0, &adapter->resetting))) { 1944 netdev_dbg(netdev, "[S:%s FOP:%d] Resetting, deferring open\n", 1945 adapter_state_to_string(adapter->state), 1946 adapter->failover_pending); 1947 adapter->state = VNIC_OPEN; 1948 rc = 0; 1949 goto out; 1950 } 1951 1952 if (adapter->state != VNIC_CLOSED) { 1953 rc = ibmvnic_login(netdev); 1954 if (rc) 1955 goto out; 1956 1957 rc = init_resources(adapter); 1958 if (rc) { 1959 netdev_err(netdev, "failed to initialize resources\n"); 1960 goto out; 1961 } 1962 } 1963 1964 rc = __ibmvnic_open(netdev); 1965 1966 out: 1967 /* If open failed and there is a pending failover or in-progress reset, 1968 * set device state and return. Device operation will be handled by 1969 * reset routine. See also comments above regarding rtnl. 1970 */ 1971 if (rc && 1972 (adapter->failover_pending || (test_bit(0, &adapter->resetting)))) { 1973 adapter->state = VNIC_OPEN; 1974 rc = 0; 1975 } 1976 1977 if (rc) { 1978 release_resources(adapter); 1979 release_rx_pools(adapter); 1980 release_tx_pools(adapter); 1981 } 1982 1983 return rc; 1984 } 1985 1986 static void clean_rx_pools(struct ibmvnic_adapter *adapter) 1987 { 1988 struct ibmvnic_rx_pool *rx_pool; 1989 struct ibmvnic_rx_buff *rx_buff; 1990 u64 rx_entries; 1991 int rx_scrqs; 1992 int i, j; 1993 1994 if (!adapter->rx_pool) 1995 return; 1996 1997 rx_scrqs = adapter->num_active_rx_pools; 1998 rx_entries = adapter->req_rx_add_entries_per_subcrq; 1999 2000 /* Free any remaining skbs in the rx buffer pools */ 2001 for (i = 0; i < rx_scrqs; i++) { 2002 rx_pool = &adapter->rx_pool[i]; 2003 if (!rx_pool || !rx_pool->rx_buff) 2004 continue; 2005 2006 netdev_dbg(adapter->netdev, "Cleaning rx_pool[%d]\n", i); 2007 for (j = 0; j < rx_entries; j++) { 2008 rx_buff = &rx_pool->rx_buff[j]; 2009 if (rx_buff && rx_buff->skb) { 2010 dev_kfree_skb_any(rx_buff->skb); 2011 rx_buff->skb = NULL; 2012 } 2013 } 2014 } 2015 } 2016 2017 static void clean_one_tx_pool(struct ibmvnic_adapter *adapter, 2018 struct ibmvnic_tx_pool *tx_pool) 2019 { 2020 struct ibmvnic_tx_buff *tx_buff; 2021 u64 tx_entries; 2022 int i; 2023 2024 if (!tx_pool || !tx_pool->tx_buff) 2025 return; 2026 2027 tx_entries = tx_pool->num_buffers; 2028 2029 for (i = 0; i < tx_entries; i++) { 2030 tx_buff = &tx_pool->tx_buff[i]; 2031 if (tx_buff && tx_buff->skb) { 2032 dev_kfree_skb_any(tx_buff->skb); 2033 tx_buff->skb = NULL; 2034 } 2035 } 2036 } 2037 2038 static void clean_tx_pools(struct ibmvnic_adapter *adapter) 2039 { 2040 int tx_scrqs; 2041 int i; 2042 2043 if (!adapter->tx_pool || !adapter->tso_pool) 2044 return; 2045 2046 tx_scrqs = adapter->num_active_tx_pools; 2047 2048 /* Free any remaining skbs in the tx buffer pools */ 2049 for (i = 0; i < tx_scrqs; i++) { 2050 netdev_dbg(adapter->netdev, "Cleaning tx_pool[%d]\n", i); 2051 clean_one_tx_pool(adapter, &adapter->tx_pool[i]); 2052 clean_one_tx_pool(adapter, &adapter->tso_pool[i]); 2053 } 2054 } 2055 2056 static void ibmvnic_disable_irqs(struct ibmvnic_adapter *adapter) 2057 { 2058 struct net_device *netdev = adapter->netdev; 2059 int i; 2060 2061 if (adapter->tx_scrq) { 2062 for (i = 0; i < adapter->req_tx_queues; i++) 2063 if (adapter->tx_scrq[i]->irq) { 2064 netdev_dbg(netdev, 2065 "Disabling tx_scrq[%d] irq\n", i); 2066 disable_scrq_irq(adapter, adapter->tx_scrq[i]); 2067 disable_irq(adapter->tx_scrq[i]->irq); 2068 } 2069 } 2070 2071 if (adapter->rx_scrq) { 2072 for (i = 0; i < adapter->req_rx_queues; i++) { 2073 if (adapter->rx_scrq[i]->irq) { 2074 netdev_dbg(netdev, 2075 "Disabling rx_scrq[%d] irq\n", i); 2076 disable_scrq_irq(adapter, adapter->rx_scrq[i]); 2077 disable_irq(adapter->rx_scrq[i]->irq); 2078 } 2079 } 2080 } 2081 } 2082 2083 static void ibmvnic_cleanup(struct net_device *netdev) 2084 { 2085 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 2086 2087 /* ensure that transmissions are stopped if called by do_reset */ 2088 2089 adapter->tx_queues_active = false; 2090 2091 /* Ensure complete_tx() and ibmvnic_xmit() see ->tx_queues_active 2092 * update so they don't restart a queue after we stop it below. 2093 */ 2094 synchronize_rcu(); 2095 2096 if (test_bit(0, &adapter->resetting)) 2097 netif_tx_disable(netdev); 2098 else 2099 netif_tx_stop_all_queues(netdev); 2100 2101 ibmvnic_napi_disable(adapter); 2102 ibmvnic_disable_irqs(adapter); 2103 } 2104 2105 static int __ibmvnic_close(struct net_device *netdev) 2106 { 2107 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 2108 int rc = 0; 2109 2110 adapter->state = VNIC_CLOSING; 2111 rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN); 2112 adapter->state = VNIC_CLOSED; 2113 return rc; 2114 } 2115 2116 static int ibmvnic_close(struct net_device *netdev) 2117 { 2118 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 2119 int rc; 2120 2121 netdev_dbg(netdev, "[S:%s FOP:%d FRR:%d] Closing\n", 2122 adapter_state_to_string(adapter->state), 2123 adapter->failover_pending, 2124 adapter->force_reset_recovery); 2125 2126 /* If device failover is pending, just set device state and return. 2127 * Device operation will be handled by reset routine. 2128 */ 2129 if (adapter->failover_pending) { 2130 adapter->state = VNIC_CLOSED; 2131 return 0; 2132 } 2133 2134 rc = __ibmvnic_close(netdev); 2135 ibmvnic_cleanup(netdev); 2136 clean_rx_pools(adapter); 2137 clean_tx_pools(adapter); 2138 2139 return rc; 2140 } 2141 2142 /** 2143 * build_hdr_data - creates L2/L3/L4 header data buffer 2144 * @hdr_field: bitfield determining needed headers 2145 * @skb: socket buffer 2146 * @hdr_len: array of header lengths 2147 * @hdr_data: buffer to write the header to 2148 * 2149 * Reads hdr_field to determine which headers are needed by firmware. 2150 * Builds a buffer containing these headers. Saves individual header 2151 * lengths and total buffer length to be used to build descriptors. 2152 */ 2153 static int build_hdr_data(u8 hdr_field, struct sk_buff *skb, 2154 int *hdr_len, u8 *hdr_data) 2155 { 2156 int len = 0; 2157 u8 *hdr; 2158 2159 if (skb_vlan_tagged(skb) && !skb_vlan_tag_present(skb)) 2160 hdr_len[0] = sizeof(struct vlan_ethhdr); 2161 else 2162 hdr_len[0] = sizeof(struct ethhdr); 2163 2164 if (skb->protocol == htons(ETH_P_IP)) { 2165 hdr_len[1] = ip_hdr(skb)->ihl * 4; 2166 if (ip_hdr(skb)->protocol == IPPROTO_TCP) 2167 hdr_len[2] = tcp_hdrlen(skb); 2168 else if (ip_hdr(skb)->protocol == IPPROTO_UDP) 2169 hdr_len[2] = sizeof(struct udphdr); 2170 } else if (skb->protocol == htons(ETH_P_IPV6)) { 2171 hdr_len[1] = sizeof(struct ipv6hdr); 2172 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP) 2173 hdr_len[2] = tcp_hdrlen(skb); 2174 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP) 2175 hdr_len[2] = sizeof(struct udphdr); 2176 } else if (skb->protocol == htons(ETH_P_ARP)) { 2177 hdr_len[1] = arp_hdr_len(skb->dev); 2178 hdr_len[2] = 0; 2179 } 2180 2181 memset(hdr_data, 0, 120); 2182 if ((hdr_field >> 6) & 1) { 2183 hdr = skb_mac_header(skb); 2184 memcpy(hdr_data, hdr, hdr_len[0]); 2185 len += hdr_len[0]; 2186 } 2187 2188 if ((hdr_field >> 5) & 1) { 2189 hdr = skb_network_header(skb); 2190 memcpy(hdr_data + len, hdr, hdr_len[1]); 2191 len += hdr_len[1]; 2192 } 2193 2194 if ((hdr_field >> 4) & 1) { 2195 hdr = skb_transport_header(skb); 2196 memcpy(hdr_data + len, hdr, hdr_len[2]); 2197 len += hdr_len[2]; 2198 } 2199 return len; 2200 } 2201 2202 /** 2203 * create_hdr_descs - create header and header extension descriptors 2204 * @hdr_field: bitfield determining needed headers 2205 * @hdr_data: buffer containing header data 2206 * @len: length of data buffer 2207 * @hdr_len: array of individual header lengths 2208 * @scrq_arr: descriptor array 2209 * 2210 * Creates header and, if needed, header extension descriptors and 2211 * places them in a descriptor array, scrq_arr 2212 */ 2213 2214 static int create_hdr_descs(u8 hdr_field, u8 *hdr_data, int len, int *hdr_len, 2215 union sub_crq *scrq_arr) 2216 { 2217 union sub_crq hdr_desc; 2218 int tmp_len = len; 2219 int num_descs = 0; 2220 u8 *data, *cur; 2221 int tmp; 2222 2223 while (tmp_len > 0) { 2224 cur = hdr_data + len - tmp_len; 2225 2226 memset(&hdr_desc, 0, sizeof(hdr_desc)); 2227 if (cur != hdr_data) { 2228 data = hdr_desc.hdr_ext.data; 2229 tmp = tmp_len > 29 ? 29 : tmp_len; 2230 hdr_desc.hdr_ext.first = IBMVNIC_CRQ_CMD; 2231 hdr_desc.hdr_ext.type = IBMVNIC_HDR_EXT_DESC; 2232 hdr_desc.hdr_ext.len = tmp; 2233 } else { 2234 data = hdr_desc.hdr.data; 2235 tmp = tmp_len > 24 ? 24 : tmp_len; 2236 hdr_desc.hdr.first = IBMVNIC_CRQ_CMD; 2237 hdr_desc.hdr.type = IBMVNIC_HDR_DESC; 2238 hdr_desc.hdr.len = tmp; 2239 hdr_desc.hdr.l2_len = (u8)hdr_len[0]; 2240 hdr_desc.hdr.l3_len = cpu_to_be16((u16)hdr_len[1]); 2241 hdr_desc.hdr.l4_len = (u8)hdr_len[2]; 2242 hdr_desc.hdr.flag = hdr_field << 1; 2243 } 2244 memcpy(data, cur, tmp); 2245 tmp_len -= tmp; 2246 *scrq_arr = hdr_desc; 2247 scrq_arr++; 2248 num_descs++; 2249 } 2250 2251 return num_descs; 2252 } 2253 2254 /** 2255 * build_hdr_descs_arr - build a header descriptor array 2256 * @skb: tx socket buffer 2257 * @indir_arr: indirect array 2258 * @num_entries: number of descriptors to be sent 2259 * @hdr_field: bit field determining which headers will be sent 2260 * 2261 * This function will build a TX descriptor array with applicable 2262 * L2/L3/L4 packet header descriptors to be sent by send_subcrq_indirect. 2263 */ 2264 2265 static void build_hdr_descs_arr(struct sk_buff *skb, 2266 union sub_crq *indir_arr, 2267 int *num_entries, u8 hdr_field) 2268 { 2269 int hdr_len[3] = {0, 0, 0}; 2270 u8 hdr_data[140] = {0}; 2271 int tot_len; 2272 2273 tot_len = build_hdr_data(hdr_field, skb, hdr_len, 2274 hdr_data); 2275 *num_entries += create_hdr_descs(hdr_field, hdr_data, tot_len, hdr_len, 2276 indir_arr + 1); 2277 } 2278 2279 static int ibmvnic_xmit_workarounds(struct sk_buff *skb, 2280 struct net_device *netdev) 2281 { 2282 /* For some backing devices, mishandling of small packets 2283 * can result in a loss of connection or TX stall. Device 2284 * architects recommend that no packet should be smaller 2285 * than the minimum MTU value provided to the driver, so 2286 * pad any packets to that length 2287 */ 2288 if (skb->len < netdev->min_mtu) 2289 return skb_put_padto(skb, netdev->min_mtu); 2290 2291 return 0; 2292 } 2293 2294 static void ibmvnic_tx_scrq_clean_buffer(struct ibmvnic_adapter *adapter, 2295 struct ibmvnic_sub_crq_queue *tx_scrq) 2296 { 2297 struct ibmvnic_ind_xmit_queue *ind_bufp; 2298 struct ibmvnic_tx_buff *tx_buff; 2299 struct ibmvnic_tx_pool *tx_pool; 2300 union sub_crq tx_scrq_entry; 2301 int queue_num; 2302 int entries; 2303 int index; 2304 int i; 2305 2306 ind_bufp = &tx_scrq->ind_buf; 2307 entries = (u64)ind_bufp->index; 2308 queue_num = tx_scrq->pool_index; 2309 2310 for (i = entries - 1; i >= 0; --i) { 2311 tx_scrq_entry = ind_bufp->indir_arr[i]; 2312 if (tx_scrq_entry.v1.type != IBMVNIC_TX_DESC) 2313 continue; 2314 index = be32_to_cpu(tx_scrq_entry.v1.correlator); 2315 if (index & IBMVNIC_TSO_POOL_MASK) { 2316 tx_pool = &adapter->tso_pool[queue_num]; 2317 index &= ~IBMVNIC_TSO_POOL_MASK; 2318 } else { 2319 tx_pool = &adapter->tx_pool[queue_num]; 2320 } 2321 tx_pool->free_map[tx_pool->consumer_index] = index; 2322 tx_pool->consumer_index = tx_pool->consumer_index == 0 ? 2323 tx_pool->num_buffers - 1 : 2324 tx_pool->consumer_index - 1; 2325 tx_buff = &tx_pool->tx_buff[index]; 2326 adapter->netdev->stats.tx_packets--; 2327 adapter->netdev->stats.tx_bytes -= tx_buff->skb->len; 2328 adapter->tx_stats_buffers[queue_num].packets--; 2329 adapter->tx_stats_buffers[queue_num].bytes -= 2330 tx_buff->skb->len; 2331 dev_kfree_skb_any(tx_buff->skb); 2332 tx_buff->skb = NULL; 2333 adapter->netdev->stats.tx_dropped++; 2334 } 2335 2336 ind_bufp->index = 0; 2337 2338 if (atomic_sub_return(entries, &tx_scrq->used) <= 2339 (adapter->req_tx_entries_per_subcrq / 2) && 2340 __netif_subqueue_stopped(adapter->netdev, queue_num)) { 2341 rcu_read_lock(); 2342 2343 if (adapter->tx_queues_active) { 2344 netif_wake_subqueue(adapter->netdev, queue_num); 2345 netdev_dbg(adapter->netdev, "Started queue %d\n", 2346 queue_num); 2347 } 2348 2349 rcu_read_unlock(); 2350 } 2351 } 2352 2353 static int ibmvnic_tx_scrq_flush(struct ibmvnic_adapter *adapter, 2354 struct ibmvnic_sub_crq_queue *tx_scrq) 2355 { 2356 struct ibmvnic_ind_xmit_queue *ind_bufp; 2357 u64 dma_addr; 2358 u64 entries; 2359 u64 handle; 2360 int rc; 2361 2362 ind_bufp = &tx_scrq->ind_buf; 2363 dma_addr = (u64)ind_bufp->indir_dma; 2364 entries = (u64)ind_bufp->index; 2365 handle = tx_scrq->handle; 2366 2367 if (!entries) 2368 return 0; 2369 rc = send_subcrq_indirect(adapter, handle, dma_addr, entries); 2370 if (rc) 2371 ibmvnic_tx_scrq_clean_buffer(adapter, tx_scrq); 2372 else 2373 ind_bufp->index = 0; 2374 return 0; 2375 } 2376 2377 static netdev_tx_t ibmvnic_xmit(struct sk_buff *skb, struct net_device *netdev) 2378 { 2379 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 2380 int queue_num = skb_get_queue_mapping(skb); 2381 u8 *hdrs = (u8 *)&adapter->tx_rx_desc_req; 2382 struct device *dev = &adapter->vdev->dev; 2383 struct ibmvnic_ind_xmit_queue *ind_bufp; 2384 struct ibmvnic_tx_buff *tx_buff = NULL; 2385 struct ibmvnic_sub_crq_queue *tx_scrq; 2386 struct ibmvnic_long_term_buff *ltb; 2387 struct ibmvnic_tx_pool *tx_pool; 2388 unsigned int tx_send_failed = 0; 2389 netdev_tx_t ret = NETDEV_TX_OK; 2390 unsigned int tx_map_failed = 0; 2391 union sub_crq indir_arr[16]; 2392 unsigned int tx_dropped = 0; 2393 unsigned int tx_packets = 0; 2394 unsigned int tx_bytes = 0; 2395 dma_addr_t data_dma_addr; 2396 struct netdev_queue *txq; 2397 unsigned long lpar_rc; 2398 union sub_crq tx_crq; 2399 unsigned int offset; 2400 int num_entries = 1; 2401 unsigned char *dst; 2402 int bufidx = 0; 2403 u8 proto = 0; 2404 2405 /* If a reset is in progress, drop the packet since 2406 * the scrqs may get torn down. Otherwise use the 2407 * rcu to ensure reset waits for us to complete. 2408 */ 2409 rcu_read_lock(); 2410 if (!adapter->tx_queues_active) { 2411 dev_kfree_skb_any(skb); 2412 2413 tx_send_failed++; 2414 tx_dropped++; 2415 ret = NETDEV_TX_OK; 2416 goto out; 2417 } 2418 2419 tx_scrq = adapter->tx_scrq[queue_num]; 2420 txq = netdev_get_tx_queue(netdev, queue_num); 2421 ind_bufp = &tx_scrq->ind_buf; 2422 2423 if (ibmvnic_xmit_workarounds(skb, netdev)) { 2424 tx_dropped++; 2425 tx_send_failed++; 2426 ret = NETDEV_TX_OK; 2427 ibmvnic_tx_scrq_flush(adapter, tx_scrq); 2428 goto out; 2429 } 2430 2431 if (skb_is_gso(skb)) 2432 tx_pool = &adapter->tso_pool[queue_num]; 2433 else 2434 tx_pool = &adapter->tx_pool[queue_num]; 2435 2436 bufidx = tx_pool->free_map[tx_pool->consumer_index]; 2437 2438 if (bufidx == IBMVNIC_INVALID_MAP) { 2439 dev_kfree_skb_any(skb); 2440 tx_send_failed++; 2441 tx_dropped++; 2442 ibmvnic_tx_scrq_flush(adapter, tx_scrq); 2443 ret = NETDEV_TX_OK; 2444 goto out; 2445 } 2446 2447 tx_pool->free_map[tx_pool->consumer_index] = IBMVNIC_INVALID_MAP; 2448 2449 map_txpool_buf_to_ltb(tx_pool, bufidx, <b, &offset); 2450 2451 dst = ltb->buff + offset; 2452 memset(dst, 0, tx_pool->buf_size); 2453 data_dma_addr = ltb->addr + offset; 2454 2455 if (skb_shinfo(skb)->nr_frags) { 2456 int cur, i; 2457 2458 /* Copy the head */ 2459 skb_copy_from_linear_data(skb, dst, skb_headlen(skb)); 2460 cur = skb_headlen(skb); 2461 2462 /* Copy the frags */ 2463 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 2464 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 2465 2466 memcpy(dst + cur, skb_frag_address(frag), 2467 skb_frag_size(frag)); 2468 cur += skb_frag_size(frag); 2469 } 2470 } else { 2471 skb_copy_from_linear_data(skb, dst, skb->len); 2472 } 2473 2474 /* post changes to long_term_buff *dst before VIOS accessing it */ 2475 dma_wmb(); 2476 2477 tx_pool->consumer_index = 2478 (tx_pool->consumer_index + 1) % tx_pool->num_buffers; 2479 2480 tx_buff = &tx_pool->tx_buff[bufidx]; 2481 2482 /* Sanity checks on our free map to make sure it points to an index 2483 * that is not being occupied by another skb. If skb memory is 2484 * not freed then we see congestion control kick in and halt tx. 2485 */ 2486 if (unlikely(tx_buff->skb)) { 2487 dev_warn_ratelimited(dev, "TX free map points to untracked skb (%s %d idx=%d)\n", 2488 skb_is_gso(skb) ? "tso_pool" : "tx_pool", 2489 queue_num, bufidx); 2490 dev_kfree_skb_any(tx_buff->skb); 2491 } 2492 2493 tx_buff->skb = skb; 2494 tx_buff->index = bufidx; 2495 tx_buff->pool_index = queue_num; 2496 2497 memset(&tx_crq, 0, sizeof(tx_crq)); 2498 tx_crq.v1.first = IBMVNIC_CRQ_CMD; 2499 tx_crq.v1.type = IBMVNIC_TX_DESC; 2500 tx_crq.v1.n_crq_elem = 1; 2501 tx_crq.v1.n_sge = 1; 2502 tx_crq.v1.flags1 = IBMVNIC_TX_COMP_NEEDED; 2503 2504 if (skb_is_gso(skb)) 2505 tx_crq.v1.correlator = 2506 cpu_to_be32(bufidx | IBMVNIC_TSO_POOL_MASK); 2507 else 2508 tx_crq.v1.correlator = cpu_to_be32(bufidx); 2509 tx_crq.v1.dma_reg = cpu_to_be16(ltb->map_id); 2510 tx_crq.v1.sge_len = cpu_to_be32(skb->len); 2511 tx_crq.v1.ioba = cpu_to_be64(data_dma_addr); 2512 2513 if (adapter->vlan_header_insertion && skb_vlan_tag_present(skb)) { 2514 tx_crq.v1.flags2 |= IBMVNIC_TX_VLAN_INSERT; 2515 tx_crq.v1.vlan_id = cpu_to_be16(skb->vlan_tci); 2516 } 2517 2518 if (skb->protocol == htons(ETH_P_IP)) { 2519 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV4; 2520 proto = ip_hdr(skb)->protocol; 2521 } else if (skb->protocol == htons(ETH_P_IPV6)) { 2522 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV6; 2523 proto = ipv6_hdr(skb)->nexthdr; 2524 } 2525 2526 if (proto == IPPROTO_TCP) 2527 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_TCP; 2528 else if (proto == IPPROTO_UDP) 2529 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_UDP; 2530 2531 if (skb->ip_summed == CHECKSUM_PARTIAL) { 2532 tx_crq.v1.flags1 |= IBMVNIC_TX_CHKSUM_OFFLOAD; 2533 hdrs += 2; 2534 } 2535 if (skb_is_gso(skb)) { 2536 tx_crq.v1.flags1 |= IBMVNIC_TX_LSO; 2537 tx_crq.v1.mss = cpu_to_be16(skb_shinfo(skb)->gso_size); 2538 hdrs += 2; 2539 } 2540 2541 if ((*hdrs >> 7) & 1) 2542 build_hdr_descs_arr(skb, indir_arr, &num_entries, *hdrs); 2543 2544 tx_crq.v1.n_crq_elem = num_entries; 2545 tx_buff->num_entries = num_entries; 2546 /* flush buffer if current entry can not fit */ 2547 if (num_entries + ind_bufp->index > IBMVNIC_MAX_IND_DESCS) { 2548 lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq); 2549 if (lpar_rc != H_SUCCESS) 2550 goto tx_flush_err; 2551 } 2552 2553 indir_arr[0] = tx_crq; 2554 memcpy(&ind_bufp->indir_arr[ind_bufp->index], &indir_arr[0], 2555 num_entries * sizeof(struct ibmvnic_generic_scrq)); 2556 ind_bufp->index += num_entries; 2557 if (__netdev_tx_sent_queue(txq, skb->len, 2558 netdev_xmit_more() && 2559 ind_bufp->index < IBMVNIC_MAX_IND_DESCS)) { 2560 lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq); 2561 if (lpar_rc != H_SUCCESS) 2562 goto tx_err; 2563 } 2564 2565 if (atomic_add_return(num_entries, &tx_scrq->used) 2566 >= adapter->req_tx_entries_per_subcrq) { 2567 netdev_dbg(netdev, "Stopping queue %d\n", queue_num); 2568 netif_stop_subqueue(netdev, queue_num); 2569 } 2570 2571 tx_packets++; 2572 tx_bytes += skb->len; 2573 txq_trans_cond_update(txq); 2574 ret = NETDEV_TX_OK; 2575 goto out; 2576 2577 tx_flush_err: 2578 dev_kfree_skb_any(skb); 2579 tx_buff->skb = NULL; 2580 tx_pool->consumer_index = tx_pool->consumer_index == 0 ? 2581 tx_pool->num_buffers - 1 : 2582 tx_pool->consumer_index - 1; 2583 tx_dropped++; 2584 tx_err: 2585 if (lpar_rc != H_CLOSED && lpar_rc != H_PARAMETER) 2586 dev_err_ratelimited(dev, "tx: send failed\n"); 2587 2588 if (lpar_rc == H_CLOSED || adapter->failover_pending) { 2589 /* Disable TX and report carrier off if queue is closed 2590 * or pending failover. 2591 * Firmware guarantees that a signal will be sent to the 2592 * driver, triggering a reset or some other action. 2593 */ 2594 netif_tx_stop_all_queues(netdev); 2595 netif_carrier_off(netdev); 2596 } 2597 out: 2598 rcu_read_unlock(); 2599 netdev->stats.tx_dropped += tx_dropped; 2600 netdev->stats.tx_bytes += tx_bytes; 2601 netdev->stats.tx_packets += tx_packets; 2602 adapter->tx_send_failed += tx_send_failed; 2603 adapter->tx_map_failed += tx_map_failed; 2604 adapter->tx_stats_buffers[queue_num].packets += tx_packets; 2605 adapter->tx_stats_buffers[queue_num].bytes += tx_bytes; 2606 adapter->tx_stats_buffers[queue_num].dropped_packets += tx_dropped; 2607 2608 return ret; 2609 } 2610 2611 static void ibmvnic_set_multi(struct net_device *netdev) 2612 { 2613 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 2614 struct netdev_hw_addr *ha; 2615 union ibmvnic_crq crq; 2616 2617 memset(&crq, 0, sizeof(crq)); 2618 crq.request_capability.first = IBMVNIC_CRQ_CMD; 2619 crq.request_capability.cmd = REQUEST_CAPABILITY; 2620 2621 if (netdev->flags & IFF_PROMISC) { 2622 if (!adapter->promisc_supported) 2623 return; 2624 } else { 2625 if (netdev->flags & IFF_ALLMULTI) { 2626 /* Accept all multicast */ 2627 memset(&crq, 0, sizeof(crq)); 2628 crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD; 2629 crq.multicast_ctrl.cmd = MULTICAST_CTRL; 2630 crq.multicast_ctrl.flags = IBMVNIC_ENABLE_ALL; 2631 ibmvnic_send_crq(adapter, &crq); 2632 } else if (netdev_mc_empty(netdev)) { 2633 /* Reject all multicast */ 2634 memset(&crq, 0, sizeof(crq)); 2635 crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD; 2636 crq.multicast_ctrl.cmd = MULTICAST_CTRL; 2637 crq.multicast_ctrl.flags = IBMVNIC_DISABLE_ALL; 2638 ibmvnic_send_crq(adapter, &crq); 2639 } else { 2640 /* Accept one or more multicast(s) */ 2641 netdev_for_each_mc_addr(ha, netdev) { 2642 memset(&crq, 0, sizeof(crq)); 2643 crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD; 2644 crq.multicast_ctrl.cmd = MULTICAST_CTRL; 2645 crq.multicast_ctrl.flags = IBMVNIC_ENABLE_MC; 2646 ether_addr_copy(&crq.multicast_ctrl.mac_addr[0], 2647 ha->addr); 2648 ibmvnic_send_crq(adapter, &crq); 2649 } 2650 } 2651 } 2652 } 2653 2654 static int __ibmvnic_set_mac(struct net_device *netdev, u8 *dev_addr) 2655 { 2656 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 2657 union ibmvnic_crq crq; 2658 int rc; 2659 2660 if (!is_valid_ether_addr(dev_addr)) { 2661 rc = -EADDRNOTAVAIL; 2662 goto err; 2663 } 2664 2665 memset(&crq, 0, sizeof(crq)); 2666 crq.change_mac_addr.first = IBMVNIC_CRQ_CMD; 2667 crq.change_mac_addr.cmd = CHANGE_MAC_ADDR; 2668 ether_addr_copy(&crq.change_mac_addr.mac_addr[0], dev_addr); 2669 2670 mutex_lock(&adapter->fw_lock); 2671 adapter->fw_done_rc = 0; 2672 reinit_completion(&adapter->fw_done); 2673 2674 rc = ibmvnic_send_crq(adapter, &crq); 2675 if (rc) { 2676 rc = -EIO; 2677 mutex_unlock(&adapter->fw_lock); 2678 goto err; 2679 } 2680 2681 rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000); 2682 /* netdev->dev_addr is changed in handle_change_mac_rsp function */ 2683 if (rc || adapter->fw_done_rc) { 2684 rc = -EIO; 2685 mutex_unlock(&adapter->fw_lock); 2686 goto err; 2687 } 2688 mutex_unlock(&adapter->fw_lock); 2689 return 0; 2690 err: 2691 ether_addr_copy(adapter->mac_addr, netdev->dev_addr); 2692 return rc; 2693 } 2694 2695 static int ibmvnic_set_mac(struct net_device *netdev, void *p) 2696 { 2697 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 2698 struct sockaddr *addr = p; 2699 int rc; 2700 2701 rc = 0; 2702 if (!is_valid_ether_addr(addr->sa_data)) 2703 return -EADDRNOTAVAIL; 2704 2705 ether_addr_copy(adapter->mac_addr, addr->sa_data); 2706 if (adapter->state != VNIC_PROBED) 2707 rc = __ibmvnic_set_mac(netdev, addr->sa_data); 2708 2709 return rc; 2710 } 2711 2712 static const char *reset_reason_to_string(enum ibmvnic_reset_reason reason) 2713 { 2714 switch (reason) { 2715 case VNIC_RESET_FAILOVER: 2716 return "FAILOVER"; 2717 case VNIC_RESET_MOBILITY: 2718 return "MOBILITY"; 2719 case VNIC_RESET_FATAL: 2720 return "FATAL"; 2721 case VNIC_RESET_NON_FATAL: 2722 return "NON_FATAL"; 2723 case VNIC_RESET_TIMEOUT: 2724 return "TIMEOUT"; 2725 case VNIC_RESET_CHANGE_PARAM: 2726 return "CHANGE_PARAM"; 2727 case VNIC_RESET_PASSIVE_INIT: 2728 return "PASSIVE_INIT"; 2729 } 2730 return "UNKNOWN"; 2731 } 2732 2733 /* 2734 * Initialize the init_done completion and return code values. We 2735 * can get a transport event just after registering the CRQ and the 2736 * tasklet will use this to communicate the transport event. To ensure 2737 * we don't miss the notification/error, initialize these _before_ 2738 * regisering the CRQ. 2739 */ 2740 static inline void reinit_init_done(struct ibmvnic_adapter *adapter) 2741 { 2742 reinit_completion(&adapter->init_done); 2743 adapter->init_done_rc = 0; 2744 } 2745 2746 /* 2747 * do_reset returns zero if we are able to keep processing reset events, or 2748 * non-zero if we hit a fatal error and must halt. 2749 */ 2750 static int do_reset(struct ibmvnic_adapter *adapter, 2751 struct ibmvnic_rwi *rwi, u32 reset_state) 2752 { 2753 struct net_device *netdev = adapter->netdev; 2754 u64 old_num_rx_queues, old_num_tx_queues; 2755 u64 old_num_rx_slots, old_num_tx_slots; 2756 int rc; 2757 2758 netdev_dbg(adapter->netdev, 2759 "[S:%s FOP:%d] Reset reason: %s, reset_state: %s\n", 2760 adapter_state_to_string(adapter->state), 2761 adapter->failover_pending, 2762 reset_reason_to_string(rwi->reset_reason), 2763 adapter_state_to_string(reset_state)); 2764 2765 adapter->reset_reason = rwi->reset_reason; 2766 /* requestor of VNIC_RESET_CHANGE_PARAM already has the rtnl lock */ 2767 if (!(adapter->reset_reason == VNIC_RESET_CHANGE_PARAM)) 2768 rtnl_lock(); 2769 2770 /* Now that we have the rtnl lock, clear any pending failover. 2771 * This will ensure ibmvnic_open() has either completed or will 2772 * block until failover is complete. 2773 */ 2774 if (rwi->reset_reason == VNIC_RESET_FAILOVER) 2775 adapter->failover_pending = false; 2776 2777 /* read the state and check (again) after getting rtnl */ 2778 reset_state = adapter->state; 2779 2780 if (reset_state == VNIC_REMOVING || reset_state == VNIC_REMOVED) { 2781 rc = -EBUSY; 2782 goto out; 2783 } 2784 2785 netif_carrier_off(netdev); 2786 2787 old_num_rx_queues = adapter->req_rx_queues; 2788 old_num_tx_queues = adapter->req_tx_queues; 2789 old_num_rx_slots = adapter->req_rx_add_entries_per_subcrq; 2790 old_num_tx_slots = adapter->req_tx_entries_per_subcrq; 2791 2792 ibmvnic_cleanup(netdev); 2793 2794 if (reset_state == VNIC_OPEN && 2795 adapter->reset_reason != VNIC_RESET_MOBILITY && 2796 adapter->reset_reason != VNIC_RESET_FAILOVER) { 2797 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) { 2798 rc = __ibmvnic_close(netdev); 2799 if (rc) 2800 goto out; 2801 } else { 2802 adapter->state = VNIC_CLOSING; 2803 2804 /* Release the RTNL lock before link state change and 2805 * re-acquire after the link state change to allow 2806 * linkwatch_event to grab the RTNL lock and run during 2807 * a reset. 2808 */ 2809 rtnl_unlock(); 2810 rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN); 2811 rtnl_lock(); 2812 if (rc) 2813 goto out; 2814 2815 if (adapter->state == VNIC_OPEN) { 2816 /* When we dropped rtnl, ibmvnic_open() got 2817 * it and noticed that we are resetting and 2818 * set the adapter state to OPEN. Update our 2819 * new "target" state, and resume the reset 2820 * from VNIC_CLOSING state. 2821 */ 2822 netdev_dbg(netdev, 2823 "Open changed state from %s, updating.\n", 2824 adapter_state_to_string(reset_state)); 2825 reset_state = VNIC_OPEN; 2826 adapter->state = VNIC_CLOSING; 2827 } 2828 2829 if (adapter->state != VNIC_CLOSING) { 2830 /* If someone else changed the adapter state 2831 * when we dropped the rtnl, fail the reset 2832 */ 2833 rc = -EAGAIN; 2834 goto out; 2835 } 2836 adapter->state = VNIC_CLOSED; 2837 } 2838 } 2839 2840 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) { 2841 release_resources(adapter); 2842 release_sub_crqs(adapter, 1); 2843 release_crq_queue(adapter); 2844 } 2845 2846 if (adapter->reset_reason != VNIC_RESET_NON_FATAL) { 2847 /* remove the closed state so when we call open it appears 2848 * we are coming from the probed state. 2849 */ 2850 adapter->state = VNIC_PROBED; 2851 2852 reinit_init_done(adapter); 2853 2854 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) { 2855 rc = init_crq_queue(adapter); 2856 } else if (adapter->reset_reason == VNIC_RESET_MOBILITY) { 2857 rc = ibmvnic_reenable_crq_queue(adapter); 2858 release_sub_crqs(adapter, 1); 2859 } else { 2860 rc = ibmvnic_reset_crq(adapter); 2861 if (rc == H_CLOSED || rc == H_SUCCESS) { 2862 rc = vio_enable_interrupts(adapter->vdev); 2863 if (rc) 2864 netdev_err(adapter->netdev, 2865 "Reset failed to enable interrupts. rc=%d\n", 2866 rc); 2867 } 2868 } 2869 2870 if (rc) { 2871 netdev_err(adapter->netdev, 2872 "Reset couldn't initialize crq. rc=%d\n", rc); 2873 goto out; 2874 } 2875 2876 rc = ibmvnic_reset_init(adapter, true); 2877 if (rc) 2878 goto out; 2879 2880 /* If the adapter was in PROBE or DOWN state prior to the reset, 2881 * exit here. 2882 */ 2883 if (reset_state == VNIC_PROBED || reset_state == VNIC_DOWN) { 2884 rc = 0; 2885 goto out; 2886 } 2887 2888 rc = ibmvnic_login(netdev); 2889 if (rc) 2890 goto out; 2891 2892 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) { 2893 rc = init_resources(adapter); 2894 if (rc) 2895 goto out; 2896 } else if (adapter->req_rx_queues != old_num_rx_queues || 2897 adapter->req_tx_queues != old_num_tx_queues || 2898 adapter->req_rx_add_entries_per_subcrq != 2899 old_num_rx_slots || 2900 adapter->req_tx_entries_per_subcrq != 2901 old_num_tx_slots || 2902 !adapter->rx_pool || 2903 !adapter->tso_pool || 2904 !adapter->tx_pool) { 2905 release_napi(adapter); 2906 release_vpd_data(adapter); 2907 2908 rc = init_resources(adapter); 2909 if (rc) 2910 goto out; 2911 2912 } else { 2913 rc = init_tx_pools(netdev); 2914 if (rc) { 2915 netdev_dbg(netdev, 2916 "init tx pools failed (%d)\n", 2917 rc); 2918 goto out; 2919 } 2920 2921 rc = init_rx_pools(netdev); 2922 if (rc) { 2923 netdev_dbg(netdev, 2924 "init rx pools failed (%d)\n", 2925 rc); 2926 goto out; 2927 } 2928 } 2929 ibmvnic_disable_irqs(adapter); 2930 } 2931 adapter->state = VNIC_CLOSED; 2932 2933 if (reset_state == VNIC_CLOSED) { 2934 rc = 0; 2935 goto out; 2936 } 2937 2938 rc = __ibmvnic_open(netdev); 2939 if (rc) { 2940 rc = IBMVNIC_OPEN_FAILED; 2941 goto out; 2942 } 2943 2944 /* refresh device's multicast list */ 2945 ibmvnic_set_multi(netdev); 2946 2947 if (adapter->reset_reason == VNIC_RESET_FAILOVER || 2948 adapter->reset_reason == VNIC_RESET_MOBILITY) 2949 __netdev_notify_peers(netdev); 2950 2951 rc = 0; 2952 2953 out: 2954 /* restore the adapter state if reset failed */ 2955 if (rc) 2956 adapter->state = reset_state; 2957 /* requestor of VNIC_RESET_CHANGE_PARAM should still hold the rtnl lock */ 2958 if (!(adapter->reset_reason == VNIC_RESET_CHANGE_PARAM)) 2959 rtnl_unlock(); 2960 2961 netdev_dbg(adapter->netdev, "[S:%s FOP:%d] Reset done, rc %d\n", 2962 adapter_state_to_string(adapter->state), 2963 adapter->failover_pending, rc); 2964 return rc; 2965 } 2966 2967 static int do_hard_reset(struct ibmvnic_adapter *adapter, 2968 struct ibmvnic_rwi *rwi, u32 reset_state) 2969 { 2970 struct net_device *netdev = adapter->netdev; 2971 int rc; 2972 2973 netdev_dbg(adapter->netdev, "Hard resetting driver (%s)\n", 2974 reset_reason_to_string(rwi->reset_reason)); 2975 2976 /* read the state and check (again) after getting rtnl */ 2977 reset_state = adapter->state; 2978 2979 if (reset_state == VNIC_REMOVING || reset_state == VNIC_REMOVED) { 2980 rc = -EBUSY; 2981 goto out; 2982 } 2983 2984 netif_carrier_off(netdev); 2985 adapter->reset_reason = rwi->reset_reason; 2986 2987 ibmvnic_cleanup(netdev); 2988 release_resources(adapter); 2989 release_sub_crqs(adapter, 0); 2990 release_crq_queue(adapter); 2991 2992 /* remove the closed state so when we call open it appears 2993 * we are coming from the probed state. 2994 */ 2995 adapter->state = VNIC_PROBED; 2996 2997 reinit_init_done(adapter); 2998 2999 rc = init_crq_queue(adapter); 3000 if (rc) { 3001 netdev_err(adapter->netdev, 3002 "Couldn't initialize crq. rc=%d\n", rc); 3003 goto out; 3004 } 3005 3006 rc = ibmvnic_reset_init(adapter, false); 3007 if (rc) 3008 goto out; 3009 3010 /* If the adapter was in PROBE or DOWN state prior to the reset, 3011 * exit here. 3012 */ 3013 if (reset_state == VNIC_PROBED || reset_state == VNIC_DOWN) 3014 goto out; 3015 3016 rc = ibmvnic_login(netdev); 3017 if (rc) 3018 goto out; 3019 3020 rc = init_resources(adapter); 3021 if (rc) 3022 goto out; 3023 3024 ibmvnic_disable_irqs(adapter); 3025 adapter->state = VNIC_CLOSED; 3026 3027 if (reset_state == VNIC_CLOSED) 3028 goto out; 3029 3030 rc = __ibmvnic_open(netdev); 3031 if (rc) { 3032 rc = IBMVNIC_OPEN_FAILED; 3033 goto out; 3034 } 3035 3036 __netdev_notify_peers(netdev); 3037 out: 3038 /* restore adapter state if reset failed */ 3039 if (rc) 3040 adapter->state = reset_state; 3041 netdev_dbg(adapter->netdev, "[S:%s FOP:%d] Hard reset done, rc %d\n", 3042 adapter_state_to_string(adapter->state), 3043 adapter->failover_pending, rc); 3044 return rc; 3045 } 3046 3047 static struct ibmvnic_rwi *get_next_rwi(struct ibmvnic_adapter *adapter) 3048 { 3049 struct ibmvnic_rwi *rwi; 3050 unsigned long flags; 3051 3052 spin_lock_irqsave(&adapter->rwi_lock, flags); 3053 3054 if (!list_empty(&adapter->rwi_list)) { 3055 rwi = list_first_entry(&adapter->rwi_list, struct ibmvnic_rwi, 3056 list); 3057 list_del(&rwi->list); 3058 } else { 3059 rwi = NULL; 3060 } 3061 3062 spin_unlock_irqrestore(&adapter->rwi_lock, flags); 3063 return rwi; 3064 } 3065 3066 /** 3067 * do_passive_init - complete probing when partner device is detected. 3068 * @adapter: ibmvnic_adapter struct 3069 * 3070 * If the ibmvnic device does not have a partner device to communicate with at boot 3071 * and that partner device comes online at a later time, this function is called 3072 * to complete the initialization process of ibmvnic device. 3073 * Caller is expected to hold rtnl_lock(). 3074 * 3075 * Returns non-zero if sub-CRQs are not initialized properly leaving the device 3076 * in the down state. 3077 * Returns 0 upon success and the device is in PROBED state. 3078 */ 3079 3080 static int do_passive_init(struct ibmvnic_adapter *adapter) 3081 { 3082 unsigned long timeout = msecs_to_jiffies(30000); 3083 struct net_device *netdev = adapter->netdev; 3084 struct device *dev = &adapter->vdev->dev; 3085 int rc; 3086 3087 netdev_dbg(netdev, "Partner device found, probing.\n"); 3088 3089 adapter->state = VNIC_PROBING; 3090 reinit_completion(&adapter->init_done); 3091 adapter->init_done_rc = 0; 3092 adapter->crq.active = true; 3093 3094 rc = send_crq_init_complete(adapter); 3095 if (rc) 3096 goto out; 3097 3098 rc = send_version_xchg(adapter); 3099 if (rc) 3100 netdev_dbg(adapter->netdev, "send_version_xchg failed, rc=%d\n", rc); 3101 3102 if (!wait_for_completion_timeout(&adapter->init_done, timeout)) { 3103 dev_err(dev, "Initialization sequence timed out\n"); 3104 rc = -ETIMEDOUT; 3105 goto out; 3106 } 3107 3108 rc = init_sub_crqs(adapter); 3109 if (rc) { 3110 dev_err(dev, "Initialization of sub crqs failed, rc=%d\n", rc); 3111 goto out; 3112 } 3113 3114 rc = init_sub_crq_irqs(adapter); 3115 if (rc) { 3116 dev_err(dev, "Failed to initialize sub crq irqs\n, rc=%d", rc); 3117 goto init_failed; 3118 } 3119 3120 netdev->mtu = adapter->req_mtu - ETH_HLEN; 3121 netdev->min_mtu = adapter->min_mtu - ETH_HLEN; 3122 netdev->max_mtu = adapter->max_mtu - ETH_HLEN; 3123 3124 adapter->state = VNIC_PROBED; 3125 netdev_dbg(netdev, "Probed successfully. Waiting for signal from partner device.\n"); 3126 3127 return 0; 3128 3129 init_failed: 3130 release_sub_crqs(adapter, 1); 3131 out: 3132 adapter->state = VNIC_DOWN; 3133 return rc; 3134 } 3135 3136 static void __ibmvnic_reset(struct work_struct *work) 3137 { 3138 struct ibmvnic_adapter *adapter; 3139 unsigned int timeout = 5000; 3140 struct ibmvnic_rwi *tmprwi; 3141 bool saved_state = false; 3142 struct ibmvnic_rwi *rwi; 3143 unsigned long flags; 3144 struct device *dev; 3145 bool need_reset; 3146 int num_fails = 0; 3147 u32 reset_state; 3148 int rc = 0; 3149 3150 adapter = container_of(work, struct ibmvnic_adapter, ibmvnic_reset); 3151 dev = &adapter->vdev->dev; 3152 3153 /* Wait for ibmvnic_probe() to complete. If probe is taking too long 3154 * or if another reset is in progress, defer work for now. If probe 3155 * eventually fails it will flush and terminate our work. 3156 * 3157 * Three possibilities here: 3158 * 1. Adpater being removed - just return 3159 * 2. Timed out on probe or another reset in progress - delay the work 3160 * 3. Completed probe - perform any resets in queue 3161 */ 3162 if (adapter->state == VNIC_PROBING && 3163 !wait_for_completion_timeout(&adapter->probe_done, timeout)) { 3164 dev_err(dev, "Reset thread timed out on probe"); 3165 queue_delayed_work(system_long_wq, 3166 &adapter->ibmvnic_delayed_reset, 3167 IBMVNIC_RESET_DELAY); 3168 return; 3169 } 3170 3171 /* adapter is done with probe (i.e state is never VNIC_PROBING now) */ 3172 if (adapter->state == VNIC_REMOVING) 3173 return; 3174 3175 /* ->rwi_list is stable now (no one else is removing entries) */ 3176 3177 /* ibmvnic_probe() may have purged the reset queue after we were 3178 * scheduled to process a reset so there maybe no resets to process. 3179 * Before setting the ->resetting bit though, we have to make sure 3180 * that there is infact a reset to process. Otherwise we may race 3181 * with ibmvnic_open() and end up leaving the vnic down: 3182 * 3183 * __ibmvnic_reset() ibmvnic_open() 3184 * ----------------- -------------- 3185 * 3186 * set ->resetting bit 3187 * find ->resetting bit is set 3188 * set ->state to IBMVNIC_OPEN (i.e 3189 * assume reset will open device) 3190 * return 3191 * find reset queue empty 3192 * return 3193 * 3194 * Neither performed vnic login/open and vnic stays down 3195 * 3196 * If we hold the lock and conditionally set the bit, either we 3197 * or ibmvnic_open() will complete the open. 3198 */ 3199 need_reset = false; 3200 spin_lock(&adapter->rwi_lock); 3201 if (!list_empty(&adapter->rwi_list)) { 3202 if (test_and_set_bit_lock(0, &adapter->resetting)) { 3203 queue_delayed_work(system_long_wq, 3204 &adapter->ibmvnic_delayed_reset, 3205 IBMVNIC_RESET_DELAY); 3206 } else { 3207 need_reset = true; 3208 } 3209 } 3210 spin_unlock(&adapter->rwi_lock); 3211 3212 if (!need_reset) 3213 return; 3214 3215 rwi = get_next_rwi(adapter); 3216 while (rwi) { 3217 spin_lock_irqsave(&adapter->state_lock, flags); 3218 3219 if (adapter->state == VNIC_REMOVING || 3220 adapter->state == VNIC_REMOVED) { 3221 spin_unlock_irqrestore(&adapter->state_lock, flags); 3222 kfree(rwi); 3223 rc = EBUSY; 3224 break; 3225 } 3226 3227 if (!saved_state) { 3228 reset_state = adapter->state; 3229 saved_state = true; 3230 } 3231 spin_unlock_irqrestore(&adapter->state_lock, flags); 3232 3233 if (rwi->reset_reason == VNIC_RESET_PASSIVE_INIT) { 3234 rtnl_lock(); 3235 rc = do_passive_init(adapter); 3236 rtnl_unlock(); 3237 if (!rc) 3238 netif_carrier_on(adapter->netdev); 3239 } else if (adapter->force_reset_recovery) { 3240 /* Since we are doing a hard reset now, clear the 3241 * failover_pending flag so we don't ignore any 3242 * future MOBILITY or other resets. 3243 */ 3244 adapter->failover_pending = false; 3245 3246 /* Transport event occurred during previous reset */ 3247 if (adapter->wait_for_reset) { 3248 /* Previous was CHANGE_PARAM; caller locked */ 3249 adapter->force_reset_recovery = false; 3250 rc = do_hard_reset(adapter, rwi, reset_state); 3251 } else { 3252 rtnl_lock(); 3253 adapter->force_reset_recovery = false; 3254 rc = do_hard_reset(adapter, rwi, reset_state); 3255 rtnl_unlock(); 3256 } 3257 if (rc) 3258 num_fails++; 3259 else 3260 num_fails = 0; 3261 3262 /* If auto-priority-failover is enabled we can get 3263 * back to back failovers during resets, resulting 3264 * in at least two failed resets (from high-priority 3265 * backing device to low-priority one and then back) 3266 * If resets continue to fail beyond that, give the 3267 * adapter some time to settle down before retrying. 3268 */ 3269 if (num_fails >= 3) { 3270 netdev_dbg(adapter->netdev, 3271 "[S:%s] Hard reset failed %d times, waiting 60 secs\n", 3272 adapter_state_to_string(adapter->state), 3273 num_fails); 3274 set_current_state(TASK_UNINTERRUPTIBLE); 3275 schedule_timeout(60 * HZ); 3276 } 3277 } else { 3278 rc = do_reset(adapter, rwi, reset_state); 3279 } 3280 tmprwi = rwi; 3281 adapter->last_reset_time = jiffies; 3282 3283 if (rc) 3284 netdev_dbg(adapter->netdev, "Reset failed, rc=%d\n", rc); 3285 3286 rwi = get_next_rwi(adapter); 3287 3288 /* 3289 * If there are no resets queued and the previous reset failed, 3290 * the adapter would be in an undefined state. So retry the 3291 * previous reset as a hard reset. 3292 * 3293 * Else, free the previous rwi and, if there is another reset 3294 * queued, process the new reset even if previous reset failed 3295 * (the previous reset could have failed because of a fail 3296 * over for instance, so process the fail over). 3297 */ 3298 if (!rwi && rc) 3299 rwi = tmprwi; 3300 else 3301 kfree(tmprwi); 3302 3303 if (rwi && (rwi->reset_reason == VNIC_RESET_FAILOVER || 3304 rwi->reset_reason == VNIC_RESET_MOBILITY || rc)) 3305 adapter->force_reset_recovery = true; 3306 } 3307 3308 if (adapter->wait_for_reset) { 3309 adapter->reset_done_rc = rc; 3310 complete(&adapter->reset_done); 3311 } 3312 3313 clear_bit_unlock(0, &adapter->resetting); 3314 3315 netdev_dbg(adapter->netdev, 3316 "[S:%s FRR:%d WFR:%d] Done processing resets\n", 3317 adapter_state_to_string(adapter->state), 3318 adapter->force_reset_recovery, 3319 adapter->wait_for_reset); 3320 } 3321 3322 static void __ibmvnic_delayed_reset(struct work_struct *work) 3323 { 3324 struct ibmvnic_adapter *adapter; 3325 3326 adapter = container_of(work, struct ibmvnic_adapter, 3327 ibmvnic_delayed_reset.work); 3328 __ibmvnic_reset(&adapter->ibmvnic_reset); 3329 } 3330 3331 static void flush_reset_queue(struct ibmvnic_adapter *adapter) 3332 { 3333 struct list_head *entry, *tmp_entry; 3334 3335 if (!list_empty(&adapter->rwi_list)) { 3336 list_for_each_safe(entry, tmp_entry, &adapter->rwi_list) { 3337 list_del(entry); 3338 kfree(list_entry(entry, struct ibmvnic_rwi, list)); 3339 } 3340 } 3341 } 3342 3343 static int ibmvnic_reset(struct ibmvnic_adapter *adapter, 3344 enum ibmvnic_reset_reason reason) 3345 { 3346 struct net_device *netdev = adapter->netdev; 3347 struct ibmvnic_rwi *rwi, *tmp; 3348 unsigned long flags; 3349 int ret; 3350 3351 spin_lock_irqsave(&adapter->rwi_lock, flags); 3352 3353 /* If failover is pending don't schedule any other reset. 3354 * Instead let the failover complete. If there is already a 3355 * a failover reset scheduled, we will detect and drop the 3356 * duplicate reset when walking the ->rwi_list below. 3357 */ 3358 if (adapter->state == VNIC_REMOVING || 3359 adapter->state == VNIC_REMOVED || 3360 (adapter->failover_pending && reason != VNIC_RESET_FAILOVER)) { 3361 ret = EBUSY; 3362 netdev_dbg(netdev, "Adapter removing or pending failover, skipping reset\n"); 3363 goto err; 3364 } 3365 3366 list_for_each_entry(tmp, &adapter->rwi_list, list) { 3367 if (tmp->reset_reason == reason) { 3368 netdev_dbg(netdev, "Skipping matching reset, reason=%s\n", 3369 reset_reason_to_string(reason)); 3370 ret = EBUSY; 3371 goto err; 3372 } 3373 } 3374 3375 rwi = kzalloc(sizeof(*rwi), GFP_ATOMIC); 3376 if (!rwi) { 3377 ret = ENOMEM; 3378 goto err; 3379 } 3380 /* if we just received a transport event, 3381 * flush reset queue and process this reset 3382 */ 3383 if (adapter->force_reset_recovery) 3384 flush_reset_queue(adapter); 3385 3386 rwi->reset_reason = reason; 3387 list_add_tail(&rwi->list, &adapter->rwi_list); 3388 netdev_dbg(adapter->netdev, "Scheduling reset (reason %s)\n", 3389 reset_reason_to_string(reason)); 3390 queue_work(system_long_wq, &adapter->ibmvnic_reset); 3391 3392 ret = 0; 3393 err: 3394 /* ibmvnic_close() below can block, so drop the lock first */ 3395 spin_unlock_irqrestore(&adapter->rwi_lock, flags); 3396 3397 if (ret == ENOMEM) 3398 ibmvnic_close(netdev); 3399 3400 return -ret; 3401 } 3402 3403 static void ibmvnic_tx_timeout(struct net_device *dev, unsigned int txqueue) 3404 { 3405 struct ibmvnic_adapter *adapter = netdev_priv(dev); 3406 3407 if (test_bit(0, &adapter->resetting)) { 3408 netdev_err(adapter->netdev, 3409 "Adapter is resetting, skip timeout reset\n"); 3410 return; 3411 } 3412 /* No queuing up reset until at least 5 seconds (default watchdog val) 3413 * after last reset 3414 */ 3415 if (time_before(jiffies, (adapter->last_reset_time + dev->watchdog_timeo))) { 3416 netdev_dbg(dev, "Not yet time to tx timeout.\n"); 3417 return; 3418 } 3419 ibmvnic_reset(adapter, VNIC_RESET_TIMEOUT); 3420 } 3421 3422 static void remove_buff_from_pool(struct ibmvnic_adapter *adapter, 3423 struct ibmvnic_rx_buff *rx_buff) 3424 { 3425 struct ibmvnic_rx_pool *pool = &adapter->rx_pool[rx_buff->pool_index]; 3426 3427 rx_buff->skb = NULL; 3428 3429 pool->free_map[pool->next_alloc] = (int)(rx_buff - pool->rx_buff); 3430 pool->next_alloc = (pool->next_alloc + 1) % pool->size; 3431 3432 atomic_dec(&pool->available); 3433 } 3434 3435 static int ibmvnic_poll(struct napi_struct *napi, int budget) 3436 { 3437 struct ibmvnic_sub_crq_queue *rx_scrq; 3438 struct ibmvnic_adapter *adapter; 3439 struct net_device *netdev; 3440 int frames_processed; 3441 int scrq_num; 3442 3443 netdev = napi->dev; 3444 adapter = netdev_priv(netdev); 3445 scrq_num = (int)(napi - adapter->napi); 3446 frames_processed = 0; 3447 rx_scrq = adapter->rx_scrq[scrq_num]; 3448 3449 restart_poll: 3450 while (frames_processed < budget) { 3451 struct sk_buff *skb; 3452 struct ibmvnic_rx_buff *rx_buff; 3453 union sub_crq *next; 3454 u32 length; 3455 u16 offset; 3456 u8 flags = 0; 3457 3458 if (unlikely(test_bit(0, &adapter->resetting) && 3459 adapter->reset_reason != VNIC_RESET_NON_FATAL)) { 3460 enable_scrq_irq(adapter, rx_scrq); 3461 napi_complete_done(napi, frames_processed); 3462 return frames_processed; 3463 } 3464 3465 if (!pending_scrq(adapter, rx_scrq)) 3466 break; 3467 next = ibmvnic_next_scrq(adapter, rx_scrq); 3468 rx_buff = (struct ibmvnic_rx_buff *) 3469 be64_to_cpu(next->rx_comp.correlator); 3470 /* do error checking */ 3471 if (next->rx_comp.rc) { 3472 netdev_dbg(netdev, "rx buffer returned with rc %x\n", 3473 be16_to_cpu(next->rx_comp.rc)); 3474 /* free the entry */ 3475 next->rx_comp.first = 0; 3476 dev_kfree_skb_any(rx_buff->skb); 3477 remove_buff_from_pool(adapter, rx_buff); 3478 continue; 3479 } else if (!rx_buff->skb) { 3480 /* free the entry */ 3481 next->rx_comp.first = 0; 3482 remove_buff_from_pool(adapter, rx_buff); 3483 continue; 3484 } 3485 3486 length = be32_to_cpu(next->rx_comp.len); 3487 offset = be16_to_cpu(next->rx_comp.off_frame_data); 3488 flags = next->rx_comp.flags; 3489 skb = rx_buff->skb; 3490 /* load long_term_buff before copying to skb */ 3491 dma_rmb(); 3492 skb_copy_to_linear_data(skb, rx_buff->data + offset, 3493 length); 3494 3495 /* VLAN Header has been stripped by the system firmware and 3496 * needs to be inserted by the driver 3497 */ 3498 if (adapter->rx_vlan_header_insertion && 3499 (flags & IBMVNIC_VLAN_STRIPPED)) 3500 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), 3501 ntohs(next->rx_comp.vlan_tci)); 3502 3503 /* free the entry */ 3504 next->rx_comp.first = 0; 3505 remove_buff_from_pool(adapter, rx_buff); 3506 3507 skb_put(skb, length); 3508 skb->protocol = eth_type_trans(skb, netdev); 3509 skb_record_rx_queue(skb, scrq_num); 3510 3511 if (flags & IBMVNIC_IP_CHKSUM_GOOD && 3512 flags & IBMVNIC_TCP_UDP_CHKSUM_GOOD) { 3513 skb->ip_summed = CHECKSUM_UNNECESSARY; 3514 } 3515 3516 length = skb->len; 3517 napi_gro_receive(napi, skb); /* send it up */ 3518 netdev->stats.rx_packets++; 3519 netdev->stats.rx_bytes += length; 3520 adapter->rx_stats_buffers[scrq_num].packets++; 3521 adapter->rx_stats_buffers[scrq_num].bytes += length; 3522 frames_processed++; 3523 } 3524 3525 if (adapter->state != VNIC_CLOSING && 3526 ((atomic_read(&adapter->rx_pool[scrq_num].available) < 3527 adapter->req_rx_add_entries_per_subcrq / 2) || 3528 frames_processed < budget)) 3529 replenish_rx_pool(adapter, &adapter->rx_pool[scrq_num]); 3530 if (frames_processed < budget) { 3531 if (napi_complete_done(napi, frames_processed)) { 3532 enable_scrq_irq(adapter, rx_scrq); 3533 if (pending_scrq(adapter, rx_scrq)) { 3534 if (napi_reschedule(napi)) { 3535 disable_scrq_irq(adapter, rx_scrq); 3536 goto restart_poll; 3537 } 3538 } 3539 } 3540 } 3541 return frames_processed; 3542 } 3543 3544 static int wait_for_reset(struct ibmvnic_adapter *adapter) 3545 { 3546 int rc, ret; 3547 3548 adapter->fallback.mtu = adapter->req_mtu; 3549 adapter->fallback.rx_queues = adapter->req_rx_queues; 3550 adapter->fallback.tx_queues = adapter->req_tx_queues; 3551 adapter->fallback.rx_entries = adapter->req_rx_add_entries_per_subcrq; 3552 adapter->fallback.tx_entries = adapter->req_tx_entries_per_subcrq; 3553 3554 reinit_completion(&adapter->reset_done); 3555 adapter->wait_for_reset = true; 3556 rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM); 3557 3558 if (rc) { 3559 ret = rc; 3560 goto out; 3561 } 3562 rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done, 60000); 3563 if (rc) { 3564 ret = -ENODEV; 3565 goto out; 3566 } 3567 3568 ret = 0; 3569 if (adapter->reset_done_rc) { 3570 ret = -EIO; 3571 adapter->desired.mtu = adapter->fallback.mtu; 3572 adapter->desired.rx_queues = adapter->fallback.rx_queues; 3573 adapter->desired.tx_queues = adapter->fallback.tx_queues; 3574 adapter->desired.rx_entries = adapter->fallback.rx_entries; 3575 adapter->desired.tx_entries = adapter->fallback.tx_entries; 3576 3577 reinit_completion(&adapter->reset_done); 3578 adapter->wait_for_reset = true; 3579 rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM); 3580 if (rc) { 3581 ret = rc; 3582 goto out; 3583 } 3584 rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done, 3585 60000); 3586 if (rc) { 3587 ret = -ENODEV; 3588 goto out; 3589 } 3590 } 3591 out: 3592 adapter->wait_for_reset = false; 3593 3594 return ret; 3595 } 3596 3597 static int ibmvnic_change_mtu(struct net_device *netdev, int new_mtu) 3598 { 3599 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3600 3601 adapter->desired.mtu = new_mtu + ETH_HLEN; 3602 3603 return wait_for_reset(adapter); 3604 } 3605 3606 static netdev_features_t ibmvnic_features_check(struct sk_buff *skb, 3607 struct net_device *dev, 3608 netdev_features_t features) 3609 { 3610 /* Some backing hardware adapters can not 3611 * handle packets with a MSS less than 224 3612 * or with only one segment. 3613 */ 3614 if (skb_is_gso(skb)) { 3615 if (skb_shinfo(skb)->gso_size < 224 || 3616 skb_shinfo(skb)->gso_segs == 1) 3617 features &= ~NETIF_F_GSO_MASK; 3618 } 3619 3620 return features; 3621 } 3622 3623 static const struct net_device_ops ibmvnic_netdev_ops = { 3624 .ndo_open = ibmvnic_open, 3625 .ndo_stop = ibmvnic_close, 3626 .ndo_start_xmit = ibmvnic_xmit, 3627 .ndo_set_rx_mode = ibmvnic_set_multi, 3628 .ndo_set_mac_address = ibmvnic_set_mac, 3629 .ndo_validate_addr = eth_validate_addr, 3630 .ndo_tx_timeout = ibmvnic_tx_timeout, 3631 .ndo_change_mtu = ibmvnic_change_mtu, 3632 .ndo_features_check = ibmvnic_features_check, 3633 }; 3634 3635 /* ethtool functions */ 3636 3637 static int ibmvnic_get_link_ksettings(struct net_device *netdev, 3638 struct ethtool_link_ksettings *cmd) 3639 { 3640 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3641 int rc; 3642 3643 rc = send_query_phys_parms(adapter); 3644 if (rc) { 3645 adapter->speed = SPEED_UNKNOWN; 3646 adapter->duplex = DUPLEX_UNKNOWN; 3647 } 3648 cmd->base.speed = adapter->speed; 3649 cmd->base.duplex = adapter->duplex; 3650 cmd->base.port = PORT_FIBRE; 3651 cmd->base.phy_address = 0; 3652 cmd->base.autoneg = AUTONEG_ENABLE; 3653 3654 return 0; 3655 } 3656 3657 static void ibmvnic_get_drvinfo(struct net_device *netdev, 3658 struct ethtool_drvinfo *info) 3659 { 3660 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3661 3662 strscpy(info->driver, ibmvnic_driver_name, sizeof(info->driver)); 3663 strscpy(info->version, IBMVNIC_DRIVER_VERSION, sizeof(info->version)); 3664 strscpy(info->fw_version, adapter->fw_version, 3665 sizeof(info->fw_version)); 3666 } 3667 3668 static u32 ibmvnic_get_msglevel(struct net_device *netdev) 3669 { 3670 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3671 3672 return adapter->msg_enable; 3673 } 3674 3675 static void ibmvnic_set_msglevel(struct net_device *netdev, u32 data) 3676 { 3677 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3678 3679 adapter->msg_enable = data; 3680 } 3681 3682 static u32 ibmvnic_get_link(struct net_device *netdev) 3683 { 3684 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3685 3686 /* Don't need to send a query because we request a logical link up at 3687 * init and then we wait for link state indications 3688 */ 3689 return adapter->logical_link_state; 3690 } 3691 3692 static void ibmvnic_get_ringparam(struct net_device *netdev, 3693 struct ethtool_ringparam *ring, 3694 struct kernel_ethtool_ringparam *kernel_ring, 3695 struct netlink_ext_ack *extack) 3696 { 3697 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3698 3699 ring->rx_max_pending = adapter->max_rx_add_entries_per_subcrq; 3700 ring->tx_max_pending = adapter->max_tx_entries_per_subcrq; 3701 ring->rx_mini_max_pending = 0; 3702 ring->rx_jumbo_max_pending = 0; 3703 ring->rx_pending = adapter->req_rx_add_entries_per_subcrq; 3704 ring->tx_pending = adapter->req_tx_entries_per_subcrq; 3705 ring->rx_mini_pending = 0; 3706 ring->rx_jumbo_pending = 0; 3707 } 3708 3709 static int ibmvnic_set_ringparam(struct net_device *netdev, 3710 struct ethtool_ringparam *ring, 3711 struct kernel_ethtool_ringparam *kernel_ring, 3712 struct netlink_ext_ack *extack) 3713 { 3714 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3715 3716 if (ring->rx_pending > adapter->max_rx_add_entries_per_subcrq || 3717 ring->tx_pending > adapter->max_tx_entries_per_subcrq) { 3718 netdev_err(netdev, "Invalid request.\n"); 3719 netdev_err(netdev, "Max tx buffers = %llu\n", 3720 adapter->max_rx_add_entries_per_subcrq); 3721 netdev_err(netdev, "Max rx buffers = %llu\n", 3722 adapter->max_tx_entries_per_subcrq); 3723 return -EINVAL; 3724 } 3725 3726 adapter->desired.rx_entries = ring->rx_pending; 3727 adapter->desired.tx_entries = ring->tx_pending; 3728 3729 return wait_for_reset(adapter); 3730 } 3731 3732 static void ibmvnic_get_channels(struct net_device *netdev, 3733 struct ethtool_channels *channels) 3734 { 3735 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3736 3737 channels->max_rx = adapter->max_rx_queues; 3738 channels->max_tx = adapter->max_tx_queues; 3739 channels->max_other = 0; 3740 channels->max_combined = 0; 3741 channels->rx_count = adapter->req_rx_queues; 3742 channels->tx_count = adapter->req_tx_queues; 3743 channels->other_count = 0; 3744 channels->combined_count = 0; 3745 } 3746 3747 static int ibmvnic_set_channels(struct net_device *netdev, 3748 struct ethtool_channels *channels) 3749 { 3750 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3751 3752 adapter->desired.rx_queues = channels->rx_count; 3753 adapter->desired.tx_queues = channels->tx_count; 3754 3755 return wait_for_reset(adapter); 3756 } 3757 3758 static void ibmvnic_get_strings(struct net_device *dev, u32 stringset, u8 *data) 3759 { 3760 struct ibmvnic_adapter *adapter = netdev_priv(dev); 3761 int i; 3762 3763 if (stringset != ETH_SS_STATS) 3764 return; 3765 3766 for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); i++, data += ETH_GSTRING_LEN) 3767 memcpy(data, ibmvnic_stats[i].name, ETH_GSTRING_LEN); 3768 3769 for (i = 0; i < adapter->req_tx_queues; i++) { 3770 snprintf(data, ETH_GSTRING_LEN, "tx%d_packets", i); 3771 data += ETH_GSTRING_LEN; 3772 3773 snprintf(data, ETH_GSTRING_LEN, "tx%d_bytes", i); 3774 data += ETH_GSTRING_LEN; 3775 3776 snprintf(data, ETH_GSTRING_LEN, "tx%d_dropped_packets", i); 3777 data += ETH_GSTRING_LEN; 3778 } 3779 3780 for (i = 0; i < adapter->req_rx_queues; i++) { 3781 snprintf(data, ETH_GSTRING_LEN, "rx%d_packets", i); 3782 data += ETH_GSTRING_LEN; 3783 3784 snprintf(data, ETH_GSTRING_LEN, "rx%d_bytes", i); 3785 data += ETH_GSTRING_LEN; 3786 3787 snprintf(data, ETH_GSTRING_LEN, "rx%d_interrupts", i); 3788 data += ETH_GSTRING_LEN; 3789 } 3790 } 3791 3792 static int ibmvnic_get_sset_count(struct net_device *dev, int sset) 3793 { 3794 struct ibmvnic_adapter *adapter = netdev_priv(dev); 3795 3796 switch (sset) { 3797 case ETH_SS_STATS: 3798 return ARRAY_SIZE(ibmvnic_stats) + 3799 adapter->req_tx_queues * NUM_TX_STATS + 3800 adapter->req_rx_queues * NUM_RX_STATS; 3801 default: 3802 return -EOPNOTSUPP; 3803 } 3804 } 3805 3806 static void ibmvnic_get_ethtool_stats(struct net_device *dev, 3807 struct ethtool_stats *stats, u64 *data) 3808 { 3809 struct ibmvnic_adapter *adapter = netdev_priv(dev); 3810 union ibmvnic_crq crq; 3811 int i, j; 3812 int rc; 3813 3814 memset(&crq, 0, sizeof(crq)); 3815 crq.request_statistics.first = IBMVNIC_CRQ_CMD; 3816 crq.request_statistics.cmd = REQUEST_STATISTICS; 3817 crq.request_statistics.ioba = cpu_to_be32(adapter->stats_token); 3818 crq.request_statistics.len = 3819 cpu_to_be32(sizeof(struct ibmvnic_statistics)); 3820 3821 /* Wait for data to be written */ 3822 reinit_completion(&adapter->stats_done); 3823 rc = ibmvnic_send_crq(adapter, &crq); 3824 if (rc) 3825 return; 3826 rc = ibmvnic_wait_for_completion(adapter, &adapter->stats_done, 10000); 3827 if (rc) 3828 return; 3829 3830 for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); i++) 3831 data[i] = be64_to_cpu(IBMVNIC_GET_STAT 3832 (adapter, ibmvnic_stats[i].offset)); 3833 3834 for (j = 0; j < adapter->req_tx_queues; j++) { 3835 data[i] = adapter->tx_stats_buffers[j].packets; 3836 i++; 3837 data[i] = adapter->tx_stats_buffers[j].bytes; 3838 i++; 3839 data[i] = adapter->tx_stats_buffers[j].dropped_packets; 3840 i++; 3841 } 3842 3843 for (j = 0; j < adapter->req_rx_queues; j++) { 3844 data[i] = adapter->rx_stats_buffers[j].packets; 3845 i++; 3846 data[i] = adapter->rx_stats_buffers[j].bytes; 3847 i++; 3848 data[i] = adapter->rx_stats_buffers[j].interrupts; 3849 i++; 3850 } 3851 } 3852 3853 static const struct ethtool_ops ibmvnic_ethtool_ops = { 3854 .get_drvinfo = ibmvnic_get_drvinfo, 3855 .get_msglevel = ibmvnic_get_msglevel, 3856 .set_msglevel = ibmvnic_set_msglevel, 3857 .get_link = ibmvnic_get_link, 3858 .get_ringparam = ibmvnic_get_ringparam, 3859 .set_ringparam = ibmvnic_set_ringparam, 3860 .get_channels = ibmvnic_get_channels, 3861 .set_channels = ibmvnic_set_channels, 3862 .get_strings = ibmvnic_get_strings, 3863 .get_sset_count = ibmvnic_get_sset_count, 3864 .get_ethtool_stats = ibmvnic_get_ethtool_stats, 3865 .get_link_ksettings = ibmvnic_get_link_ksettings, 3866 }; 3867 3868 /* Routines for managing CRQs/sCRQs */ 3869 3870 static int reset_one_sub_crq_queue(struct ibmvnic_adapter *adapter, 3871 struct ibmvnic_sub_crq_queue *scrq) 3872 { 3873 int rc; 3874 3875 if (!scrq) { 3876 netdev_dbg(adapter->netdev, "Invalid scrq reset.\n"); 3877 return -EINVAL; 3878 } 3879 3880 if (scrq->irq) { 3881 free_irq(scrq->irq, scrq); 3882 irq_dispose_mapping(scrq->irq); 3883 scrq->irq = 0; 3884 } 3885 3886 if (scrq->msgs) { 3887 memset(scrq->msgs, 0, 4 * PAGE_SIZE); 3888 atomic_set(&scrq->used, 0); 3889 scrq->cur = 0; 3890 scrq->ind_buf.index = 0; 3891 } else { 3892 netdev_dbg(adapter->netdev, "Invalid scrq reset\n"); 3893 return -EINVAL; 3894 } 3895 3896 rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token, 3897 4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq); 3898 return rc; 3899 } 3900 3901 static int reset_sub_crq_queues(struct ibmvnic_adapter *adapter) 3902 { 3903 int i, rc; 3904 3905 if (!adapter->tx_scrq || !adapter->rx_scrq) 3906 return -EINVAL; 3907 3908 ibmvnic_clean_affinity(adapter); 3909 3910 for (i = 0; i < adapter->req_tx_queues; i++) { 3911 netdev_dbg(adapter->netdev, "Re-setting tx_scrq[%d]\n", i); 3912 rc = reset_one_sub_crq_queue(adapter, adapter->tx_scrq[i]); 3913 if (rc) 3914 return rc; 3915 } 3916 3917 for (i = 0; i < adapter->req_rx_queues; i++) { 3918 netdev_dbg(adapter->netdev, "Re-setting rx_scrq[%d]\n", i); 3919 rc = reset_one_sub_crq_queue(adapter, adapter->rx_scrq[i]); 3920 if (rc) 3921 return rc; 3922 } 3923 3924 return rc; 3925 } 3926 3927 static void release_sub_crq_queue(struct ibmvnic_adapter *adapter, 3928 struct ibmvnic_sub_crq_queue *scrq, 3929 bool do_h_free) 3930 { 3931 struct device *dev = &adapter->vdev->dev; 3932 long rc; 3933 3934 netdev_dbg(adapter->netdev, "Releasing sub-CRQ\n"); 3935 3936 if (do_h_free) { 3937 /* Close the sub-crqs */ 3938 do { 3939 rc = plpar_hcall_norets(H_FREE_SUB_CRQ, 3940 adapter->vdev->unit_address, 3941 scrq->crq_num); 3942 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc)); 3943 3944 if (rc) { 3945 netdev_err(adapter->netdev, 3946 "Failed to release sub-CRQ %16lx, rc = %ld\n", 3947 scrq->crq_num, rc); 3948 } 3949 } 3950 3951 dma_free_coherent(dev, 3952 IBMVNIC_IND_ARR_SZ, 3953 scrq->ind_buf.indir_arr, 3954 scrq->ind_buf.indir_dma); 3955 3956 dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE, 3957 DMA_BIDIRECTIONAL); 3958 free_pages((unsigned long)scrq->msgs, 2); 3959 free_cpumask_var(scrq->affinity_mask); 3960 kfree(scrq); 3961 } 3962 3963 static struct ibmvnic_sub_crq_queue *init_sub_crq_queue(struct ibmvnic_adapter 3964 *adapter) 3965 { 3966 struct device *dev = &adapter->vdev->dev; 3967 struct ibmvnic_sub_crq_queue *scrq; 3968 int rc; 3969 3970 scrq = kzalloc(sizeof(*scrq), GFP_KERNEL); 3971 if (!scrq) 3972 return NULL; 3973 3974 scrq->msgs = 3975 (union sub_crq *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 2); 3976 if (!scrq->msgs) { 3977 dev_warn(dev, "Couldn't allocate crq queue messages page\n"); 3978 goto zero_page_failed; 3979 } 3980 if (!zalloc_cpumask_var(&scrq->affinity_mask, GFP_KERNEL)) 3981 goto cpumask_alloc_failed; 3982 3983 scrq->msg_token = dma_map_single(dev, scrq->msgs, 4 * PAGE_SIZE, 3984 DMA_BIDIRECTIONAL); 3985 if (dma_mapping_error(dev, scrq->msg_token)) { 3986 dev_warn(dev, "Couldn't map crq queue messages page\n"); 3987 goto map_failed; 3988 } 3989 3990 rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token, 3991 4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq); 3992 3993 if (rc == H_RESOURCE) 3994 rc = ibmvnic_reset_crq(adapter); 3995 3996 if (rc == H_CLOSED) { 3997 dev_warn(dev, "Partner adapter not ready, waiting.\n"); 3998 } else if (rc) { 3999 dev_warn(dev, "Error %d registering sub-crq\n", rc); 4000 goto reg_failed; 4001 } 4002 4003 scrq->adapter = adapter; 4004 scrq->size = 4 * PAGE_SIZE / sizeof(*scrq->msgs); 4005 scrq->ind_buf.index = 0; 4006 4007 scrq->ind_buf.indir_arr = 4008 dma_alloc_coherent(dev, 4009 IBMVNIC_IND_ARR_SZ, 4010 &scrq->ind_buf.indir_dma, 4011 GFP_KERNEL); 4012 4013 if (!scrq->ind_buf.indir_arr) 4014 goto indir_failed; 4015 4016 spin_lock_init(&scrq->lock); 4017 4018 netdev_dbg(adapter->netdev, 4019 "sub-crq initialized, num %lx, hw_irq=%lx, irq=%x\n", 4020 scrq->crq_num, scrq->hw_irq, scrq->irq); 4021 4022 return scrq; 4023 4024 indir_failed: 4025 do { 4026 rc = plpar_hcall_norets(H_FREE_SUB_CRQ, 4027 adapter->vdev->unit_address, 4028 scrq->crq_num); 4029 } while (rc == H_BUSY || rc == H_IS_LONG_BUSY(rc)); 4030 reg_failed: 4031 dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE, 4032 DMA_BIDIRECTIONAL); 4033 map_failed: 4034 free_cpumask_var(scrq->affinity_mask); 4035 cpumask_alloc_failed: 4036 free_pages((unsigned long)scrq->msgs, 2); 4037 zero_page_failed: 4038 kfree(scrq); 4039 4040 return NULL; 4041 } 4042 4043 static void release_sub_crqs(struct ibmvnic_adapter *adapter, bool do_h_free) 4044 { 4045 int i; 4046 4047 ibmvnic_clean_affinity(adapter); 4048 if (adapter->tx_scrq) { 4049 for (i = 0; i < adapter->num_active_tx_scrqs; i++) { 4050 if (!adapter->tx_scrq[i]) 4051 continue; 4052 4053 netdev_dbg(adapter->netdev, "Releasing tx_scrq[%d]\n", 4054 i); 4055 ibmvnic_tx_scrq_clean_buffer(adapter, adapter->tx_scrq[i]); 4056 if (adapter->tx_scrq[i]->irq) { 4057 free_irq(adapter->tx_scrq[i]->irq, 4058 adapter->tx_scrq[i]); 4059 irq_dispose_mapping(adapter->tx_scrq[i]->irq); 4060 adapter->tx_scrq[i]->irq = 0; 4061 } 4062 4063 release_sub_crq_queue(adapter, adapter->tx_scrq[i], 4064 do_h_free); 4065 } 4066 4067 kfree(adapter->tx_scrq); 4068 adapter->tx_scrq = NULL; 4069 adapter->num_active_tx_scrqs = 0; 4070 } 4071 4072 /* Clean any remaining outstanding SKBs 4073 * we freed the irq so we won't be hearing 4074 * from them 4075 */ 4076 clean_tx_pools(adapter); 4077 4078 if (adapter->rx_scrq) { 4079 for (i = 0; i < adapter->num_active_rx_scrqs; i++) { 4080 if (!adapter->rx_scrq[i]) 4081 continue; 4082 4083 netdev_dbg(adapter->netdev, "Releasing rx_scrq[%d]\n", 4084 i); 4085 if (adapter->rx_scrq[i]->irq) { 4086 free_irq(adapter->rx_scrq[i]->irq, 4087 adapter->rx_scrq[i]); 4088 irq_dispose_mapping(adapter->rx_scrq[i]->irq); 4089 adapter->rx_scrq[i]->irq = 0; 4090 } 4091 4092 release_sub_crq_queue(adapter, adapter->rx_scrq[i], 4093 do_h_free); 4094 } 4095 4096 kfree(adapter->rx_scrq); 4097 adapter->rx_scrq = NULL; 4098 adapter->num_active_rx_scrqs = 0; 4099 } 4100 } 4101 4102 static int disable_scrq_irq(struct ibmvnic_adapter *adapter, 4103 struct ibmvnic_sub_crq_queue *scrq) 4104 { 4105 struct device *dev = &adapter->vdev->dev; 4106 unsigned long rc; 4107 4108 rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address, 4109 H_DISABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0); 4110 if (rc) 4111 dev_err(dev, "Couldn't disable scrq irq 0x%lx. rc=%ld\n", 4112 scrq->hw_irq, rc); 4113 return rc; 4114 } 4115 4116 /* We can not use the IRQ chip EOI handler because that has the 4117 * unintended effect of changing the interrupt priority. 4118 */ 4119 static void ibmvnic_xics_eoi(struct device *dev, struct ibmvnic_sub_crq_queue *scrq) 4120 { 4121 u64 val = 0xff000000 | scrq->hw_irq; 4122 unsigned long rc; 4123 4124 rc = plpar_hcall_norets(H_EOI, val); 4125 if (rc) 4126 dev_err(dev, "H_EOI FAILED irq 0x%llx. rc=%ld\n", val, rc); 4127 } 4128 4129 /* Due to a firmware bug, the hypervisor can send an interrupt to a 4130 * transmit or receive queue just prior to a partition migration. 4131 * Force an EOI after migration. 4132 */ 4133 static void ibmvnic_clear_pending_interrupt(struct device *dev, 4134 struct ibmvnic_sub_crq_queue *scrq) 4135 { 4136 if (!xive_enabled()) 4137 ibmvnic_xics_eoi(dev, scrq); 4138 } 4139 4140 static int enable_scrq_irq(struct ibmvnic_adapter *adapter, 4141 struct ibmvnic_sub_crq_queue *scrq) 4142 { 4143 struct device *dev = &adapter->vdev->dev; 4144 unsigned long rc; 4145 4146 if (scrq->hw_irq > 0x100000000ULL) { 4147 dev_err(dev, "bad hw_irq = %lx\n", scrq->hw_irq); 4148 return 1; 4149 } 4150 4151 if (test_bit(0, &adapter->resetting) && 4152 adapter->reset_reason == VNIC_RESET_MOBILITY) { 4153 ibmvnic_clear_pending_interrupt(dev, scrq); 4154 } 4155 4156 rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address, 4157 H_ENABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0); 4158 if (rc) 4159 dev_err(dev, "Couldn't enable scrq irq 0x%lx. rc=%ld\n", 4160 scrq->hw_irq, rc); 4161 return rc; 4162 } 4163 4164 static int ibmvnic_complete_tx(struct ibmvnic_adapter *adapter, 4165 struct ibmvnic_sub_crq_queue *scrq) 4166 { 4167 struct device *dev = &adapter->vdev->dev; 4168 struct ibmvnic_tx_pool *tx_pool; 4169 struct ibmvnic_tx_buff *txbuff; 4170 struct netdev_queue *txq; 4171 union sub_crq *next; 4172 int index; 4173 int i; 4174 4175 restart_loop: 4176 while (pending_scrq(adapter, scrq)) { 4177 unsigned int pool = scrq->pool_index; 4178 int num_entries = 0; 4179 int total_bytes = 0; 4180 int num_packets = 0; 4181 4182 next = ibmvnic_next_scrq(adapter, scrq); 4183 for (i = 0; i < next->tx_comp.num_comps; i++) { 4184 index = be32_to_cpu(next->tx_comp.correlators[i]); 4185 if (index & IBMVNIC_TSO_POOL_MASK) { 4186 tx_pool = &adapter->tso_pool[pool]; 4187 index &= ~IBMVNIC_TSO_POOL_MASK; 4188 } else { 4189 tx_pool = &adapter->tx_pool[pool]; 4190 } 4191 4192 txbuff = &tx_pool->tx_buff[index]; 4193 num_packets++; 4194 num_entries += txbuff->num_entries; 4195 if (txbuff->skb) { 4196 total_bytes += txbuff->skb->len; 4197 if (next->tx_comp.rcs[i]) { 4198 dev_err(dev, "tx error %x\n", 4199 next->tx_comp.rcs[i]); 4200 dev_kfree_skb_irq(txbuff->skb); 4201 } else { 4202 dev_consume_skb_irq(txbuff->skb); 4203 } 4204 txbuff->skb = NULL; 4205 } else { 4206 netdev_warn(adapter->netdev, 4207 "TX completion received with NULL socket buffer\n"); 4208 } 4209 tx_pool->free_map[tx_pool->producer_index] = index; 4210 tx_pool->producer_index = 4211 (tx_pool->producer_index + 1) % 4212 tx_pool->num_buffers; 4213 } 4214 /* remove tx_comp scrq*/ 4215 next->tx_comp.first = 0; 4216 4217 txq = netdev_get_tx_queue(adapter->netdev, scrq->pool_index); 4218 netdev_tx_completed_queue(txq, num_packets, total_bytes); 4219 4220 if (atomic_sub_return(num_entries, &scrq->used) <= 4221 (adapter->req_tx_entries_per_subcrq / 2) && 4222 __netif_subqueue_stopped(adapter->netdev, 4223 scrq->pool_index)) { 4224 rcu_read_lock(); 4225 if (adapter->tx_queues_active) { 4226 netif_wake_subqueue(adapter->netdev, 4227 scrq->pool_index); 4228 netdev_dbg(adapter->netdev, 4229 "Started queue %d\n", 4230 scrq->pool_index); 4231 } 4232 rcu_read_unlock(); 4233 } 4234 } 4235 4236 enable_scrq_irq(adapter, scrq); 4237 4238 if (pending_scrq(adapter, scrq)) { 4239 disable_scrq_irq(adapter, scrq); 4240 goto restart_loop; 4241 } 4242 4243 return 0; 4244 } 4245 4246 static irqreturn_t ibmvnic_interrupt_tx(int irq, void *instance) 4247 { 4248 struct ibmvnic_sub_crq_queue *scrq = instance; 4249 struct ibmvnic_adapter *adapter = scrq->adapter; 4250 4251 disable_scrq_irq(adapter, scrq); 4252 ibmvnic_complete_tx(adapter, scrq); 4253 4254 return IRQ_HANDLED; 4255 } 4256 4257 static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance) 4258 { 4259 struct ibmvnic_sub_crq_queue *scrq = instance; 4260 struct ibmvnic_adapter *adapter = scrq->adapter; 4261 4262 /* When booting a kdump kernel we can hit pending interrupts 4263 * prior to completing driver initialization. 4264 */ 4265 if (unlikely(adapter->state != VNIC_OPEN)) 4266 return IRQ_NONE; 4267 4268 adapter->rx_stats_buffers[scrq->scrq_num].interrupts++; 4269 4270 if (napi_schedule_prep(&adapter->napi[scrq->scrq_num])) { 4271 disable_scrq_irq(adapter, scrq); 4272 __napi_schedule(&adapter->napi[scrq->scrq_num]); 4273 } 4274 4275 return IRQ_HANDLED; 4276 } 4277 4278 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter) 4279 { 4280 struct device *dev = &adapter->vdev->dev; 4281 struct ibmvnic_sub_crq_queue *scrq; 4282 int i = 0, j = 0; 4283 int rc = 0; 4284 4285 for (i = 0; i < adapter->req_tx_queues; i++) { 4286 netdev_dbg(adapter->netdev, "Initializing tx_scrq[%d] irq\n", 4287 i); 4288 scrq = adapter->tx_scrq[i]; 4289 scrq->irq = irq_create_mapping(NULL, scrq->hw_irq); 4290 4291 if (!scrq->irq) { 4292 rc = -EINVAL; 4293 dev_err(dev, "Error mapping irq\n"); 4294 goto req_tx_irq_failed; 4295 } 4296 4297 snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-tx%d", 4298 adapter->vdev->unit_address, i); 4299 rc = request_irq(scrq->irq, ibmvnic_interrupt_tx, 4300 0, scrq->name, scrq); 4301 4302 if (rc) { 4303 dev_err(dev, "Couldn't register tx irq 0x%x. rc=%d\n", 4304 scrq->irq, rc); 4305 irq_dispose_mapping(scrq->irq); 4306 goto req_tx_irq_failed; 4307 } 4308 } 4309 4310 for (i = 0; i < adapter->req_rx_queues; i++) { 4311 netdev_dbg(adapter->netdev, "Initializing rx_scrq[%d] irq\n", 4312 i); 4313 scrq = adapter->rx_scrq[i]; 4314 scrq->irq = irq_create_mapping(NULL, scrq->hw_irq); 4315 if (!scrq->irq) { 4316 rc = -EINVAL; 4317 dev_err(dev, "Error mapping irq\n"); 4318 goto req_rx_irq_failed; 4319 } 4320 snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-rx%d", 4321 adapter->vdev->unit_address, i); 4322 rc = request_irq(scrq->irq, ibmvnic_interrupt_rx, 4323 0, scrq->name, scrq); 4324 if (rc) { 4325 dev_err(dev, "Couldn't register rx irq 0x%x. rc=%d\n", 4326 scrq->irq, rc); 4327 irq_dispose_mapping(scrq->irq); 4328 goto req_rx_irq_failed; 4329 } 4330 } 4331 4332 cpus_read_lock(); 4333 ibmvnic_set_affinity(adapter); 4334 cpus_read_unlock(); 4335 4336 return rc; 4337 4338 req_rx_irq_failed: 4339 for (j = 0; j < i; j++) { 4340 free_irq(adapter->rx_scrq[j]->irq, adapter->rx_scrq[j]); 4341 irq_dispose_mapping(adapter->rx_scrq[j]->irq); 4342 } 4343 i = adapter->req_tx_queues; 4344 req_tx_irq_failed: 4345 for (j = 0; j < i; j++) { 4346 free_irq(adapter->tx_scrq[j]->irq, adapter->tx_scrq[j]); 4347 irq_dispose_mapping(adapter->tx_scrq[j]->irq); 4348 } 4349 release_sub_crqs(adapter, 1); 4350 return rc; 4351 } 4352 4353 static int init_sub_crqs(struct ibmvnic_adapter *adapter) 4354 { 4355 struct device *dev = &adapter->vdev->dev; 4356 struct ibmvnic_sub_crq_queue **allqueues; 4357 int registered_queues = 0; 4358 int total_queues; 4359 int more = 0; 4360 int i; 4361 4362 total_queues = adapter->req_tx_queues + adapter->req_rx_queues; 4363 4364 allqueues = kcalloc(total_queues, sizeof(*allqueues), GFP_KERNEL); 4365 if (!allqueues) 4366 return -ENOMEM; 4367 4368 for (i = 0; i < total_queues; i++) { 4369 allqueues[i] = init_sub_crq_queue(adapter); 4370 if (!allqueues[i]) { 4371 dev_warn(dev, "Couldn't allocate all sub-crqs\n"); 4372 break; 4373 } 4374 registered_queues++; 4375 } 4376 4377 /* Make sure we were able to register the minimum number of queues */ 4378 if (registered_queues < 4379 adapter->min_tx_queues + adapter->min_rx_queues) { 4380 dev_err(dev, "Fatal: Couldn't init min number of sub-crqs\n"); 4381 goto tx_failed; 4382 } 4383 4384 /* Distribute the failed allocated queues*/ 4385 for (i = 0; i < total_queues - registered_queues + more ; i++) { 4386 netdev_dbg(adapter->netdev, "Reducing number of queues\n"); 4387 switch (i % 3) { 4388 case 0: 4389 if (adapter->req_rx_queues > adapter->min_rx_queues) 4390 adapter->req_rx_queues--; 4391 else 4392 more++; 4393 break; 4394 case 1: 4395 if (adapter->req_tx_queues > adapter->min_tx_queues) 4396 adapter->req_tx_queues--; 4397 else 4398 more++; 4399 break; 4400 } 4401 } 4402 4403 adapter->tx_scrq = kcalloc(adapter->req_tx_queues, 4404 sizeof(*adapter->tx_scrq), GFP_KERNEL); 4405 if (!adapter->tx_scrq) 4406 goto tx_failed; 4407 4408 for (i = 0; i < adapter->req_tx_queues; i++) { 4409 adapter->tx_scrq[i] = allqueues[i]; 4410 adapter->tx_scrq[i]->pool_index = i; 4411 adapter->num_active_tx_scrqs++; 4412 } 4413 4414 adapter->rx_scrq = kcalloc(adapter->req_rx_queues, 4415 sizeof(*adapter->rx_scrq), GFP_KERNEL); 4416 if (!adapter->rx_scrq) 4417 goto rx_failed; 4418 4419 for (i = 0; i < adapter->req_rx_queues; i++) { 4420 adapter->rx_scrq[i] = allqueues[i + adapter->req_tx_queues]; 4421 adapter->rx_scrq[i]->scrq_num = i; 4422 adapter->num_active_rx_scrqs++; 4423 } 4424 4425 kfree(allqueues); 4426 return 0; 4427 4428 rx_failed: 4429 kfree(adapter->tx_scrq); 4430 adapter->tx_scrq = NULL; 4431 tx_failed: 4432 for (i = 0; i < registered_queues; i++) 4433 release_sub_crq_queue(adapter, allqueues[i], 1); 4434 kfree(allqueues); 4435 return -ENOMEM; 4436 } 4437 4438 static void send_request_cap(struct ibmvnic_adapter *adapter, int retry) 4439 { 4440 struct device *dev = &adapter->vdev->dev; 4441 union ibmvnic_crq crq; 4442 int max_entries; 4443 int cap_reqs; 4444 4445 /* We send out 6 or 7 REQUEST_CAPABILITY CRQs below (depending on 4446 * the PROMISC flag). Initialize this count upfront. When the tasklet 4447 * receives a response to all of these, it will send the next protocol 4448 * message (QUERY_IP_OFFLOAD). 4449 */ 4450 if (!(adapter->netdev->flags & IFF_PROMISC) || 4451 adapter->promisc_supported) 4452 cap_reqs = 7; 4453 else 4454 cap_reqs = 6; 4455 4456 if (!retry) { 4457 /* Sub-CRQ entries are 32 byte long */ 4458 int entries_page = 4 * PAGE_SIZE / (sizeof(u64) * 4); 4459 4460 atomic_set(&adapter->running_cap_crqs, cap_reqs); 4461 4462 if (adapter->min_tx_entries_per_subcrq > entries_page || 4463 adapter->min_rx_add_entries_per_subcrq > entries_page) { 4464 dev_err(dev, "Fatal, invalid entries per sub-crq\n"); 4465 return; 4466 } 4467 4468 if (adapter->desired.mtu) 4469 adapter->req_mtu = adapter->desired.mtu; 4470 else 4471 adapter->req_mtu = adapter->netdev->mtu + ETH_HLEN; 4472 4473 if (!adapter->desired.tx_entries) 4474 adapter->desired.tx_entries = 4475 adapter->max_tx_entries_per_subcrq; 4476 if (!adapter->desired.rx_entries) 4477 adapter->desired.rx_entries = 4478 adapter->max_rx_add_entries_per_subcrq; 4479 4480 max_entries = IBMVNIC_LTB_SET_SIZE / 4481 (adapter->req_mtu + IBMVNIC_BUFFER_HLEN); 4482 4483 if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) * 4484 adapter->desired.tx_entries > IBMVNIC_LTB_SET_SIZE) { 4485 adapter->desired.tx_entries = max_entries; 4486 } 4487 4488 if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) * 4489 adapter->desired.rx_entries > IBMVNIC_LTB_SET_SIZE) { 4490 adapter->desired.rx_entries = max_entries; 4491 } 4492 4493 if (adapter->desired.tx_entries) 4494 adapter->req_tx_entries_per_subcrq = 4495 adapter->desired.tx_entries; 4496 else 4497 adapter->req_tx_entries_per_subcrq = 4498 adapter->max_tx_entries_per_subcrq; 4499 4500 if (adapter->desired.rx_entries) 4501 adapter->req_rx_add_entries_per_subcrq = 4502 adapter->desired.rx_entries; 4503 else 4504 adapter->req_rx_add_entries_per_subcrq = 4505 adapter->max_rx_add_entries_per_subcrq; 4506 4507 if (adapter->desired.tx_queues) 4508 adapter->req_tx_queues = 4509 adapter->desired.tx_queues; 4510 else 4511 adapter->req_tx_queues = 4512 adapter->opt_tx_comp_sub_queues; 4513 4514 if (adapter->desired.rx_queues) 4515 adapter->req_rx_queues = 4516 adapter->desired.rx_queues; 4517 else 4518 adapter->req_rx_queues = 4519 adapter->opt_rx_comp_queues; 4520 4521 adapter->req_rx_add_queues = adapter->max_rx_add_queues; 4522 } else { 4523 atomic_add(cap_reqs, &adapter->running_cap_crqs); 4524 } 4525 memset(&crq, 0, sizeof(crq)); 4526 crq.request_capability.first = IBMVNIC_CRQ_CMD; 4527 crq.request_capability.cmd = REQUEST_CAPABILITY; 4528 4529 crq.request_capability.capability = cpu_to_be16(REQ_TX_QUEUES); 4530 crq.request_capability.number = cpu_to_be64(adapter->req_tx_queues); 4531 cap_reqs--; 4532 ibmvnic_send_crq(adapter, &crq); 4533 4534 crq.request_capability.capability = cpu_to_be16(REQ_RX_QUEUES); 4535 crq.request_capability.number = cpu_to_be64(adapter->req_rx_queues); 4536 cap_reqs--; 4537 ibmvnic_send_crq(adapter, &crq); 4538 4539 crq.request_capability.capability = cpu_to_be16(REQ_RX_ADD_QUEUES); 4540 crq.request_capability.number = cpu_to_be64(adapter->req_rx_add_queues); 4541 cap_reqs--; 4542 ibmvnic_send_crq(adapter, &crq); 4543 4544 crq.request_capability.capability = 4545 cpu_to_be16(REQ_TX_ENTRIES_PER_SUBCRQ); 4546 crq.request_capability.number = 4547 cpu_to_be64(adapter->req_tx_entries_per_subcrq); 4548 cap_reqs--; 4549 ibmvnic_send_crq(adapter, &crq); 4550 4551 crq.request_capability.capability = 4552 cpu_to_be16(REQ_RX_ADD_ENTRIES_PER_SUBCRQ); 4553 crq.request_capability.number = 4554 cpu_to_be64(adapter->req_rx_add_entries_per_subcrq); 4555 cap_reqs--; 4556 ibmvnic_send_crq(adapter, &crq); 4557 4558 crq.request_capability.capability = cpu_to_be16(REQ_MTU); 4559 crq.request_capability.number = cpu_to_be64(adapter->req_mtu); 4560 cap_reqs--; 4561 ibmvnic_send_crq(adapter, &crq); 4562 4563 if (adapter->netdev->flags & IFF_PROMISC) { 4564 if (adapter->promisc_supported) { 4565 crq.request_capability.capability = 4566 cpu_to_be16(PROMISC_REQUESTED); 4567 crq.request_capability.number = cpu_to_be64(1); 4568 cap_reqs--; 4569 ibmvnic_send_crq(adapter, &crq); 4570 } 4571 } else { 4572 crq.request_capability.capability = 4573 cpu_to_be16(PROMISC_REQUESTED); 4574 crq.request_capability.number = cpu_to_be64(0); 4575 cap_reqs--; 4576 ibmvnic_send_crq(adapter, &crq); 4577 } 4578 4579 /* Keep at end to catch any discrepancy between expected and actual 4580 * CRQs sent. 4581 */ 4582 WARN_ON(cap_reqs != 0); 4583 } 4584 4585 static int pending_scrq(struct ibmvnic_adapter *adapter, 4586 struct ibmvnic_sub_crq_queue *scrq) 4587 { 4588 union sub_crq *entry = &scrq->msgs[scrq->cur]; 4589 int rc; 4590 4591 rc = !!(entry->generic.first & IBMVNIC_CRQ_CMD_RSP); 4592 4593 /* Ensure that the SCRQ valid flag is loaded prior to loading the 4594 * contents of the SCRQ descriptor 4595 */ 4596 dma_rmb(); 4597 4598 return rc; 4599 } 4600 4601 static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *adapter, 4602 struct ibmvnic_sub_crq_queue *scrq) 4603 { 4604 union sub_crq *entry; 4605 unsigned long flags; 4606 4607 spin_lock_irqsave(&scrq->lock, flags); 4608 entry = &scrq->msgs[scrq->cur]; 4609 if (entry->generic.first & IBMVNIC_CRQ_CMD_RSP) { 4610 if (++scrq->cur == scrq->size) 4611 scrq->cur = 0; 4612 } else { 4613 entry = NULL; 4614 } 4615 spin_unlock_irqrestore(&scrq->lock, flags); 4616 4617 /* Ensure that the SCRQ valid flag is loaded prior to loading the 4618 * contents of the SCRQ descriptor 4619 */ 4620 dma_rmb(); 4621 4622 return entry; 4623 } 4624 4625 static union ibmvnic_crq *ibmvnic_next_crq(struct ibmvnic_adapter *adapter) 4626 { 4627 struct ibmvnic_crq_queue *queue = &adapter->crq; 4628 union ibmvnic_crq *crq; 4629 4630 crq = &queue->msgs[queue->cur]; 4631 if (crq->generic.first & IBMVNIC_CRQ_CMD_RSP) { 4632 if (++queue->cur == queue->size) 4633 queue->cur = 0; 4634 } else { 4635 crq = NULL; 4636 } 4637 4638 return crq; 4639 } 4640 4641 static void print_subcrq_error(struct device *dev, int rc, const char *func) 4642 { 4643 switch (rc) { 4644 case H_PARAMETER: 4645 dev_warn_ratelimited(dev, 4646 "%s failed: Send request is malformed or adapter failover pending. (rc=%d)\n", 4647 func, rc); 4648 break; 4649 case H_CLOSED: 4650 dev_warn_ratelimited(dev, 4651 "%s failed: Backing queue closed. Adapter is down or failover pending. (rc=%d)\n", 4652 func, rc); 4653 break; 4654 default: 4655 dev_err_ratelimited(dev, "%s failed: (rc=%d)\n", func, rc); 4656 break; 4657 } 4658 } 4659 4660 static int send_subcrq_indirect(struct ibmvnic_adapter *adapter, 4661 u64 remote_handle, u64 ioba, u64 num_entries) 4662 { 4663 unsigned int ua = adapter->vdev->unit_address; 4664 struct device *dev = &adapter->vdev->dev; 4665 int rc; 4666 4667 /* Make sure the hypervisor sees the complete request */ 4668 dma_wmb(); 4669 rc = plpar_hcall_norets(H_SEND_SUB_CRQ_INDIRECT, ua, 4670 cpu_to_be64(remote_handle), 4671 ioba, num_entries); 4672 4673 if (rc) 4674 print_subcrq_error(dev, rc, __func__); 4675 4676 return rc; 4677 } 4678 4679 static int ibmvnic_send_crq(struct ibmvnic_adapter *adapter, 4680 union ibmvnic_crq *crq) 4681 { 4682 unsigned int ua = adapter->vdev->unit_address; 4683 struct device *dev = &adapter->vdev->dev; 4684 u64 *u64_crq = (u64 *)crq; 4685 int rc; 4686 4687 netdev_dbg(adapter->netdev, "Sending CRQ: %016lx %016lx\n", 4688 (unsigned long)cpu_to_be64(u64_crq[0]), 4689 (unsigned long)cpu_to_be64(u64_crq[1])); 4690 4691 if (!adapter->crq.active && 4692 crq->generic.first != IBMVNIC_CRQ_INIT_CMD) { 4693 dev_warn(dev, "Invalid request detected while CRQ is inactive, possible device state change during reset\n"); 4694 return -EINVAL; 4695 } 4696 4697 /* Make sure the hypervisor sees the complete request */ 4698 dma_wmb(); 4699 4700 rc = plpar_hcall_norets(H_SEND_CRQ, ua, 4701 cpu_to_be64(u64_crq[0]), 4702 cpu_to_be64(u64_crq[1])); 4703 4704 if (rc) { 4705 if (rc == H_CLOSED) { 4706 dev_warn(dev, "CRQ Queue closed\n"); 4707 /* do not reset, report the fail, wait for passive init from server */ 4708 } 4709 4710 dev_warn(dev, "Send error (rc=%d)\n", rc); 4711 } 4712 4713 return rc; 4714 } 4715 4716 static int ibmvnic_send_crq_init(struct ibmvnic_adapter *adapter) 4717 { 4718 struct device *dev = &adapter->vdev->dev; 4719 union ibmvnic_crq crq; 4720 int retries = 100; 4721 int rc; 4722 4723 memset(&crq, 0, sizeof(crq)); 4724 crq.generic.first = IBMVNIC_CRQ_INIT_CMD; 4725 crq.generic.cmd = IBMVNIC_CRQ_INIT; 4726 netdev_dbg(adapter->netdev, "Sending CRQ init\n"); 4727 4728 do { 4729 rc = ibmvnic_send_crq(adapter, &crq); 4730 if (rc != H_CLOSED) 4731 break; 4732 retries--; 4733 msleep(50); 4734 4735 } while (retries > 0); 4736 4737 if (rc) { 4738 dev_err(dev, "Failed to send init request, rc = %d\n", rc); 4739 return rc; 4740 } 4741 4742 return 0; 4743 } 4744 4745 struct vnic_login_client_data { 4746 u8 type; 4747 __be16 len; 4748 char name[]; 4749 } __packed; 4750 4751 static int vnic_client_data_len(struct ibmvnic_adapter *adapter) 4752 { 4753 int len; 4754 4755 /* Calculate the amount of buffer space needed for the 4756 * vnic client data in the login buffer. There are four entries, 4757 * OS name, LPAR name, device name, and a null last entry. 4758 */ 4759 len = 4 * sizeof(struct vnic_login_client_data); 4760 len += 6; /* "Linux" plus NULL */ 4761 len += strlen(utsname()->nodename) + 1; 4762 len += strlen(adapter->netdev->name) + 1; 4763 4764 return len; 4765 } 4766 4767 static void vnic_add_client_data(struct ibmvnic_adapter *adapter, 4768 struct vnic_login_client_data *vlcd) 4769 { 4770 const char *os_name = "Linux"; 4771 int len; 4772 4773 /* Type 1 - LPAR OS */ 4774 vlcd->type = 1; 4775 len = strlen(os_name) + 1; 4776 vlcd->len = cpu_to_be16(len); 4777 strscpy(vlcd->name, os_name, len); 4778 vlcd = (struct vnic_login_client_data *)(vlcd->name + len); 4779 4780 /* Type 2 - LPAR name */ 4781 vlcd->type = 2; 4782 len = strlen(utsname()->nodename) + 1; 4783 vlcd->len = cpu_to_be16(len); 4784 strscpy(vlcd->name, utsname()->nodename, len); 4785 vlcd = (struct vnic_login_client_data *)(vlcd->name + len); 4786 4787 /* Type 3 - device name */ 4788 vlcd->type = 3; 4789 len = strlen(adapter->netdev->name) + 1; 4790 vlcd->len = cpu_to_be16(len); 4791 strscpy(vlcd->name, adapter->netdev->name, len); 4792 } 4793 4794 static int send_login(struct ibmvnic_adapter *adapter) 4795 { 4796 struct ibmvnic_login_rsp_buffer *login_rsp_buffer; 4797 struct ibmvnic_login_buffer *login_buffer; 4798 struct device *dev = &adapter->vdev->dev; 4799 struct vnic_login_client_data *vlcd; 4800 dma_addr_t rsp_buffer_token; 4801 dma_addr_t buffer_token; 4802 size_t rsp_buffer_size; 4803 union ibmvnic_crq crq; 4804 int client_data_len; 4805 size_t buffer_size; 4806 __be64 *tx_list_p; 4807 __be64 *rx_list_p; 4808 int rc; 4809 int i; 4810 4811 if (!adapter->tx_scrq || !adapter->rx_scrq) { 4812 netdev_err(adapter->netdev, 4813 "RX or TX queues are not allocated, device login failed\n"); 4814 return -ENOMEM; 4815 } 4816 4817 release_login_buffer(adapter); 4818 release_login_rsp_buffer(adapter); 4819 4820 client_data_len = vnic_client_data_len(adapter); 4821 4822 buffer_size = 4823 sizeof(struct ibmvnic_login_buffer) + 4824 sizeof(u64) * (adapter->req_tx_queues + adapter->req_rx_queues) + 4825 client_data_len; 4826 4827 login_buffer = kzalloc(buffer_size, GFP_ATOMIC); 4828 if (!login_buffer) 4829 goto buf_alloc_failed; 4830 4831 buffer_token = dma_map_single(dev, login_buffer, buffer_size, 4832 DMA_TO_DEVICE); 4833 if (dma_mapping_error(dev, buffer_token)) { 4834 dev_err(dev, "Couldn't map login buffer\n"); 4835 goto buf_map_failed; 4836 } 4837 4838 rsp_buffer_size = sizeof(struct ibmvnic_login_rsp_buffer) + 4839 sizeof(u64) * adapter->req_tx_queues + 4840 sizeof(u64) * adapter->req_rx_queues + 4841 sizeof(u64) * adapter->req_rx_queues + 4842 sizeof(u8) * IBMVNIC_TX_DESC_VERSIONS; 4843 4844 login_rsp_buffer = kmalloc(rsp_buffer_size, GFP_ATOMIC); 4845 if (!login_rsp_buffer) 4846 goto buf_rsp_alloc_failed; 4847 4848 rsp_buffer_token = dma_map_single(dev, login_rsp_buffer, 4849 rsp_buffer_size, DMA_FROM_DEVICE); 4850 if (dma_mapping_error(dev, rsp_buffer_token)) { 4851 dev_err(dev, "Couldn't map login rsp buffer\n"); 4852 goto buf_rsp_map_failed; 4853 } 4854 4855 adapter->login_buf = login_buffer; 4856 adapter->login_buf_token = buffer_token; 4857 adapter->login_buf_sz = buffer_size; 4858 adapter->login_rsp_buf = login_rsp_buffer; 4859 adapter->login_rsp_buf_token = rsp_buffer_token; 4860 adapter->login_rsp_buf_sz = rsp_buffer_size; 4861 4862 login_buffer->len = cpu_to_be32(buffer_size); 4863 login_buffer->version = cpu_to_be32(INITIAL_VERSION_LB); 4864 login_buffer->num_txcomp_subcrqs = cpu_to_be32(adapter->req_tx_queues); 4865 login_buffer->off_txcomp_subcrqs = 4866 cpu_to_be32(sizeof(struct ibmvnic_login_buffer)); 4867 login_buffer->num_rxcomp_subcrqs = cpu_to_be32(adapter->req_rx_queues); 4868 login_buffer->off_rxcomp_subcrqs = 4869 cpu_to_be32(sizeof(struct ibmvnic_login_buffer) + 4870 sizeof(u64) * adapter->req_tx_queues); 4871 login_buffer->login_rsp_ioba = cpu_to_be32(rsp_buffer_token); 4872 login_buffer->login_rsp_len = cpu_to_be32(rsp_buffer_size); 4873 4874 tx_list_p = (__be64 *)((char *)login_buffer + 4875 sizeof(struct ibmvnic_login_buffer)); 4876 rx_list_p = (__be64 *)((char *)login_buffer + 4877 sizeof(struct ibmvnic_login_buffer) + 4878 sizeof(u64) * adapter->req_tx_queues); 4879 4880 for (i = 0; i < adapter->req_tx_queues; i++) { 4881 if (adapter->tx_scrq[i]) { 4882 tx_list_p[i] = 4883 cpu_to_be64(adapter->tx_scrq[i]->crq_num); 4884 } 4885 } 4886 4887 for (i = 0; i < adapter->req_rx_queues; i++) { 4888 if (adapter->rx_scrq[i]) { 4889 rx_list_p[i] = 4890 cpu_to_be64(adapter->rx_scrq[i]->crq_num); 4891 } 4892 } 4893 4894 /* Insert vNIC login client data */ 4895 vlcd = (struct vnic_login_client_data *) 4896 ((char *)rx_list_p + (sizeof(u64) * adapter->req_rx_queues)); 4897 login_buffer->client_data_offset = 4898 cpu_to_be32((char *)vlcd - (char *)login_buffer); 4899 login_buffer->client_data_len = cpu_to_be32(client_data_len); 4900 4901 vnic_add_client_data(adapter, vlcd); 4902 4903 netdev_dbg(adapter->netdev, "Login Buffer:\n"); 4904 for (i = 0; i < (adapter->login_buf_sz - 1) / 8 + 1; i++) { 4905 netdev_dbg(adapter->netdev, "%016lx\n", 4906 ((unsigned long *)(adapter->login_buf))[i]); 4907 } 4908 4909 memset(&crq, 0, sizeof(crq)); 4910 crq.login.first = IBMVNIC_CRQ_CMD; 4911 crq.login.cmd = LOGIN; 4912 crq.login.ioba = cpu_to_be32(buffer_token); 4913 crq.login.len = cpu_to_be32(buffer_size); 4914 4915 adapter->login_pending = true; 4916 rc = ibmvnic_send_crq(adapter, &crq); 4917 if (rc) { 4918 adapter->login_pending = false; 4919 netdev_err(adapter->netdev, "Failed to send login, rc=%d\n", rc); 4920 goto buf_send_failed; 4921 } 4922 4923 return 0; 4924 4925 buf_send_failed: 4926 dma_unmap_single(dev, rsp_buffer_token, rsp_buffer_size, 4927 DMA_FROM_DEVICE); 4928 buf_rsp_map_failed: 4929 kfree(login_rsp_buffer); 4930 adapter->login_rsp_buf = NULL; 4931 buf_rsp_alloc_failed: 4932 dma_unmap_single(dev, buffer_token, buffer_size, DMA_TO_DEVICE); 4933 buf_map_failed: 4934 kfree(login_buffer); 4935 adapter->login_buf = NULL; 4936 buf_alloc_failed: 4937 return -ENOMEM; 4938 } 4939 4940 static int send_request_map(struct ibmvnic_adapter *adapter, dma_addr_t addr, 4941 u32 len, u8 map_id) 4942 { 4943 union ibmvnic_crq crq; 4944 4945 memset(&crq, 0, sizeof(crq)); 4946 crq.request_map.first = IBMVNIC_CRQ_CMD; 4947 crq.request_map.cmd = REQUEST_MAP; 4948 crq.request_map.map_id = map_id; 4949 crq.request_map.ioba = cpu_to_be32(addr); 4950 crq.request_map.len = cpu_to_be32(len); 4951 return ibmvnic_send_crq(adapter, &crq); 4952 } 4953 4954 static int send_request_unmap(struct ibmvnic_adapter *adapter, u8 map_id) 4955 { 4956 union ibmvnic_crq crq; 4957 4958 memset(&crq, 0, sizeof(crq)); 4959 crq.request_unmap.first = IBMVNIC_CRQ_CMD; 4960 crq.request_unmap.cmd = REQUEST_UNMAP; 4961 crq.request_unmap.map_id = map_id; 4962 return ibmvnic_send_crq(adapter, &crq); 4963 } 4964 4965 static void send_query_map(struct ibmvnic_adapter *adapter) 4966 { 4967 union ibmvnic_crq crq; 4968 4969 memset(&crq, 0, sizeof(crq)); 4970 crq.query_map.first = IBMVNIC_CRQ_CMD; 4971 crq.query_map.cmd = QUERY_MAP; 4972 ibmvnic_send_crq(adapter, &crq); 4973 } 4974 4975 /* Send a series of CRQs requesting various capabilities of the VNIC server */ 4976 static void send_query_cap(struct ibmvnic_adapter *adapter) 4977 { 4978 union ibmvnic_crq crq; 4979 int cap_reqs; 4980 4981 /* We send out 25 QUERY_CAPABILITY CRQs below. Initialize this count 4982 * upfront. When the tasklet receives a response to all of these, it 4983 * can send out the next protocol messaage (REQUEST_CAPABILITY). 4984 */ 4985 cap_reqs = 25; 4986 4987 atomic_set(&adapter->running_cap_crqs, cap_reqs); 4988 4989 memset(&crq, 0, sizeof(crq)); 4990 crq.query_capability.first = IBMVNIC_CRQ_CMD; 4991 crq.query_capability.cmd = QUERY_CAPABILITY; 4992 4993 crq.query_capability.capability = cpu_to_be16(MIN_TX_QUEUES); 4994 ibmvnic_send_crq(adapter, &crq); 4995 cap_reqs--; 4996 4997 crq.query_capability.capability = cpu_to_be16(MIN_RX_QUEUES); 4998 ibmvnic_send_crq(adapter, &crq); 4999 cap_reqs--; 5000 5001 crq.query_capability.capability = cpu_to_be16(MIN_RX_ADD_QUEUES); 5002 ibmvnic_send_crq(adapter, &crq); 5003 cap_reqs--; 5004 5005 crq.query_capability.capability = cpu_to_be16(MAX_TX_QUEUES); 5006 ibmvnic_send_crq(adapter, &crq); 5007 cap_reqs--; 5008 5009 crq.query_capability.capability = cpu_to_be16(MAX_RX_QUEUES); 5010 ibmvnic_send_crq(adapter, &crq); 5011 cap_reqs--; 5012 5013 crq.query_capability.capability = cpu_to_be16(MAX_RX_ADD_QUEUES); 5014 ibmvnic_send_crq(adapter, &crq); 5015 cap_reqs--; 5016 5017 crq.query_capability.capability = 5018 cpu_to_be16(MIN_TX_ENTRIES_PER_SUBCRQ); 5019 ibmvnic_send_crq(adapter, &crq); 5020 cap_reqs--; 5021 5022 crq.query_capability.capability = 5023 cpu_to_be16(MIN_RX_ADD_ENTRIES_PER_SUBCRQ); 5024 ibmvnic_send_crq(adapter, &crq); 5025 cap_reqs--; 5026 5027 crq.query_capability.capability = 5028 cpu_to_be16(MAX_TX_ENTRIES_PER_SUBCRQ); 5029 ibmvnic_send_crq(adapter, &crq); 5030 cap_reqs--; 5031 5032 crq.query_capability.capability = 5033 cpu_to_be16(MAX_RX_ADD_ENTRIES_PER_SUBCRQ); 5034 ibmvnic_send_crq(adapter, &crq); 5035 cap_reqs--; 5036 5037 crq.query_capability.capability = cpu_to_be16(TCP_IP_OFFLOAD); 5038 ibmvnic_send_crq(adapter, &crq); 5039 cap_reqs--; 5040 5041 crq.query_capability.capability = cpu_to_be16(PROMISC_SUPPORTED); 5042 ibmvnic_send_crq(adapter, &crq); 5043 cap_reqs--; 5044 5045 crq.query_capability.capability = cpu_to_be16(MIN_MTU); 5046 ibmvnic_send_crq(adapter, &crq); 5047 cap_reqs--; 5048 5049 crq.query_capability.capability = cpu_to_be16(MAX_MTU); 5050 ibmvnic_send_crq(adapter, &crq); 5051 cap_reqs--; 5052 5053 crq.query_capability.capability = cpu_to_be16(MAX_MULTICAST_FILTERS); 5054 ibmvnic_send_crq(adapter, &crq); 5055 cap_reqs--; 5056 5057 crq.query_capability.capability = cpu_to_be16(VLAN_HEADER_INSERTION); 5058 ibmvnic_send_crq(adapter, &crq); 5059 cap_reqs--; 5060 5061 crq.query_capability.capability = cpu_to_be16(RX_VLAN_HEADER_INSERTION); 5062 ibmvnic_send_crq(adapter, &crq); 5063 cap_reqs--; 5064 5065 crq.query_capability.capability = cpu_to_be16(MAX_TX_SG_ENTRIES); 5066 ibmvnic_send_crq(adapter, &crq); 5067 cap_reqs--; 5068 5069 crq.query_capability.capability = cpu_to_be16(RX_SG_SUPPORTED); 5070 ibmvnic_send_crq(adapter, &crq); 5071 cap_reqs--; 5072 5073 crq.query_capability.capability = cpu_to_be16(OPT_TX_COMP_SUB_QUEUES); 5074 ibmvnic_send_crq(adapter, &crq); 5075 cap_reqs--; 5076 5077 crq.query_capability.capability = cpu_to_be16(OPT_RX_COMP_QUEUES); 5078 ibmvnic_send_crq(adapter, &crq); 5079 cap_reqs--; 5080 5081 crq.query_capability.capability = 5082 cpu_to_be16(OPT_RX_BUFADD_Q_PER_RX_COMP_Q); 5083 ibmvnic_send_crq(adapter, &crq); 5084 cap_reqs--; 5085 5086 crq.query_capability.capability = 5087 cpu_to_be16(OPT_TX_ENTRIES_PER_SUBCRQ); 5088 ibmvnic_send_crq(adapter, &crq); 5089 cap_reqs--; 5090 5091 crq.query_capability.capability = 5092 cpu_to_be16(OPT_RXBA_ENTRIES_PER_SUBCRQ); 5093 ibmvnic_send_crq(adapter, &crq); 5094 cap_reqs--; 5095 5096 crq.query_capability.capability = cpu_to_be16(TX_RX_DESC_REQ); 5097 5098 ibmvnic_send_crq(adapter, &crq); 5099 cap_reqs--; 5100 5101 /* Keep at end to catch any discrepancy between expected and actual 5102 * CRQs sent. 5103 */ 5104 WARN_ON(cap_reqs != 0); 5105 } 5106 5107 static void send_query_ip_offload(struct ibmvnic_adapter *adapter) 5108 { 5109 int buf_sz = sizeof(struct ibmvnic_query_ip_offload_buffer); 5110 struct device *dev = &adapter->vdev->dev; 5111 union ibmvnic_crq crq; 5112 5113 adapter->ip_offload_tok = 5114 dma_map_single(dev, 5115 &adapter->ip_offload_buf, 5116 buf_sz, 5117 DMA_FROM_DEVICE); 5118 5119 if (dma_mapping_error(dev, adapter->ip_offload_tok)) { 5120 if (!firmware_has_feature(FW_FEATURE_CMO)) 5121 dev_err(dev, "Couldn't map offload buffer\n"); 5122 return; 5123 } 5124 5125 memset(&crq, 0, sizeof(crq)); 5126 crq.query_ip_offload.first = IBMVNIC_CRQ_CMD; 5127 crq.query_ip_offload.cmd = QUERY_IP_OFFLOAD; 5128 crq.query_ip_offload.len = cpu_to_be32(buf_sz); 5129 crq.query_ip_offload.ioba = 5130 cpu_to_be32(adapter->ip_offload_tok); 5131 5132 ibmvnic_send_crq(adapter, &crq); 5133 } 5134 5135 static void send_control_ip_offload(struct ibmvnic_adapter *adapter) 5136 { 5137 struct ibmvnic_control_ip_offload_buffer *ctrl_buf = &adapter->ip_offload_ctrl; 5138 struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf; 5139 struct device *dev = &adapter->vdev->dev; 5140 netdev_features_t old_hw_features = 0; 5141 union ibmvnic_crq crq; 5142 5143 adapter->ip_offload_ctrl_tok = 5144 dma_map_single(dev, 5145 ctrl_buf, 5146 sizeof(adapter->ip_offload_ctrl), 5147 DMA_TO_DEVICE); 5148 5149 if (dma_mapping_error(dev, adapter->ip_offload_ctrl_tok)) { 5150 dev_err(dev, "Couldn't map ip offload control buffer\n"); 5151 return; 5152 } 5153 5154 ctrl_buf->len = cpu_to_be32(sizeof(adapter->ip_offload_ctrl)); 5155 ctrl_buf->version = cpu_to_be32(INITIAL_VERSION_IOB); 5156 ctrl_buf->ipv4_chksum = buf->ipv4_chksum; 5157 ctrl_buf->ipv6_chksum = buf->ipv6_chksum; 5158 ctrl_buf->tcp_ipv4_chksum = buf->tcp_ipv4_chksum; 5159 ctrl_buf->udp_ipv4_chksum = buf->udp_ipv4_chksum; 5160 ctrl_buf->tcp_ipv6_chksum = buf->tcp_ipv6_chksum; 5161 ctrl_buf->udp_ipv6_chksum = buf->udp_ipv6_chksum; 5162 ctrl_buf->large_tx_ipv4 = buf->large_tx_ipv4; 5163 ctrl_buf->large_tx_ipv6 = buf->large_tx_ipv6; 5164 5165 /* large_rx disabled for now, additional features needed */ 5166 ctrl_buf->large_rx_ipv4 = 0; 5167 ctrl_buf->large_rx_ipv6 = 0; 5168 5169 if (adapter->state != VNIC_PROBING) { 5170 old_hw_features = adapter->netdev->hw_features; 5171 adapter->netdev->hw_features = 0; 5172 } 5173 5174 adapter->netdev->hw_features = NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO; 5175 5176 if (buf->tcp_ipv4_chksum || buf->udp_ipv4_chksum) 5177 adapter->netdev->hw_features |= NETIF_F_IP_CSUM; 5178 5179 if (buf->tcp_ipv6_chksum || buf->udp_ipv6_chksum) 5180 adapter->netdev->hw_features |= NETIF_F_IPV6_CSUM; 5181 5182 if ((adapter->netdev->features & 5183 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM))) 5184 adapter->netdev->hw_features |= NETIF_F_RXCSUM; 5185 5186 if (buf->large_tx_ipv4) 5187 adapter->netdev->hw_features |= NETIF_F_TSO; 5188 if (buf->large_tx_ipv6) 5189 adapter->netdev->hw_features |= NETIF_F_TSO6; 5190 5191 if (adapter->state == VNIC_PROBING) { 5192 adapter->netdev->features |= adapter->netdev->hw_features; 5193 } else if (old_hw_features != adapter->netdev->hw_features) { 5194 netdev_features_t tmp = 0; 5195 5196 /* disable features no longer supported */ 5197 adapter->netdev->features &= adapter->netdev->hw_features; 5198 /* turn on features now supported if previously enabled */ 5199 tmp = (old_hw_features ^ adapter->netdev->hw_features) & 5200 adapter->netdev->hw_features; 5201 adapter->netdev->features |= 5202 tmp & adapter->netdev->wanted_features; 5203 } 5204 5205 memset(&crq, 0, sizeof(crq)); 5206 crq.control_ip_offload.first = IBMVNIC_CRQ_CMD; 5207 crq.control_ip_offload.cmd = CONTROL_IP_OFFLOAD; 5208 crq.control_ip_offload.len = 5209 cpu_to_be32(sizeof(adapter->ip_offload_ctrl)); 5210 crq.control_ip_offload.ioba = cpu_to_be32(adapter->ip_offload_ctrl_tok); 5211 ibmvnic_send_crq(adapter, &crq); 5212 } 5213 5214 static void handle_vpd_size_rsp(union ibmvnic_crq *crq, 5215 struct ibmvnic_adapter *adapter) 5216 { 5217 struct device *dev = &adapter->vdev->dev; 5218 5219 if (crq->get_vpd_size_rsp.rc.code) { 5220 dev_err(dev, "Error retrieving VPD size, rc=%x\n", 5221 crq->get_vpd_size_rsp.rc.code); 5222 complete(&adapter->fw_done); 5223 return; 5224 } 5225 5226 adapter->vpd->len = be64_to_cpu(crq->get_vpd_size_rsp.len); 5227 complete(&adapter->fw_done); 5228 } 5229 5230 static void handle_vpd_rsp(union ibmvnic_crq *crq, 5231 struct ibmvnic_adapter *adapter) 5232 { 5233 struct device *dev = &adapter->vdev->dev; 5234 unsigned char *substr = NULL; 5235 u8 fw_level_len = 0; 5236 5237 memset(adapter->fw_version, 0, 32); 5238 5239 dma_unmap_single(dev, adapter->vpd->dma_addr, adapter->vpd->len, 5240 DMA_FROM_DEVICE); 5241 5242 if (crq->get_vpd_rsp.rc.code) { 5243 dev_err(dev, "Error retrieving VPD from device, rc=%x\n", 5244 crq->get_vpd_rsp.rc.code); 5245 goto complete; 5246 } 5247 5248 /* get the position of the firmware version info 5249 * located after the ASCII 'RM' substring in the buffer 5250 */ 5251 substr = strnstr(adapter->vpd->buff, "RM", adapter->vpd->len); 5252 if (!substr) { 5253 dev_info(dev, "Warning - No FW level has been provided in the VPD buffer by the VIOS Server\n"); 5254 goto complete; 5255 } 5256 5257 /* get length of firmware level ASCII substring */ 5258 if ((substr + 2) < (adapter->vpd->buff + adapter->vpd->len)) { 5259 fw_level_len = *(substr + 2); 5260 } else { 5261 dev_info(dev, "Length of FW substr extrapolated VDP buff\n"); 5262 goto complete; 5263 } 5264 5265 /* copy firmware version string from vpd into adapter */ 5266 if ((substr + 3 + fw_level_len) < 5267 (adapter->vpd->buff + adapter->vpd->len)) { 5268 strncpy((char *)adapter->fw_version, substr + 3, fw_level_len); 5269 } else { 5270 dev_info(dev, "FW substr extrapolated VPD buff\n"); 5271 } 5272 5273 complete: 5274 if (adapter->fw_version[0] == '\0') 5275 strscpy((char *)adapter->fw_version, "N/A", sizeof(adapter->fw_version)); 5276 complete(&adapter->fw_done); 5277 } 5278 5279 static void handle_query_ip_offload_rsp(struct ibmvnic_adapter *adapter) 5280 { 5281 struct device *dev = &adapter->vdev->dev; 5282 struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf; 5283 int i; 5284 5285 dma_unmap_single(dev, adapter->ip_offload_tok, 5286 sizeof(adapter->ip_offload_buf), DMA_FROM_DEVICE); 5287 5288 netdev_dbg(adapter->netdev, "Query IP Offload Buffer:\n"); 5289 for (i = 0; i < (sizeof(adapter->ip_offload_buf) - 1) / 8 + 1; i++) 5290 netdev_dbg(adapter->netdev, "%016lx\n", 5291 ((unsigned long *)(buf))[i]); 5292 5293 netdev_dbg(adapter->netdev, "ipv4_chksum = %d\n", buf->ipv4_chksum); 5294 netdev_dbg(adapter->netdev, "ipv6_chksum = %d\n", buf->ipv6_chksum); 5295 netdev_dbg(adapter->netdev, "tcp_ipv4_chksum = %d\n", 5296 buf->tcp_ipv4_chksum); 5297 netdev_dbg(adapter->netdev, "tcp_ipv6_chksum = %d\n", 5298 buf->tcp_ipv6_chksum); 5299 netdev_dbg(adapter->netdev, "udp_ipv4_chksum = %d\n", 5300 buf->udp_ipv4_chksum); 5301 netdev_dbg(adapter->netdev, "udp_ipv6_chksum = %d\n", 5302 buf->udp_ipv6_chksum); 5303 netdev_dbg(adapter->netdev, "large_tx_ipv4 = %d\n", 5304 buf->large_tx_ipv4); 5305 netdev_dbg(adapter->netdev, "large_tx_ipv6 = %d\n", 5306 buf->large_tx_ipv6); 5307 netdev_dbg(adapter->netdev, "large_rx_ipv4 = %d\n", 5308 buf->large_rx_ipv4); 5309 netdev_dbg(adapter->netdev, "large_rx_ipv6 = %d\n", 5310 buf->large_rx_ipv6); 5311 netdev_dbg(adapter->netdev, "max_ipv4_hdr_sz = %d\n", 5312 buf->max_ipv4_header_size); 5313 netdev_dbg(adapter->netdev, "max_ipv6_hdr_sz = %d\n", 5314 buf->max_ipv6_header_size); 5315 netdev_dbg(adapter->netdev, "max_tcp_hdr_size = %d\n", 5316 buf->max_tcp_header_size); 5317 netdev_dbg(adapter->netdev, "max_udp_hdr_size = %d\n", 5318 buf->max_udp_header_size); 5319 netdev_dbg(adapter->netdev, "max_large_tx_size = %d\n", 5320 buf->max_large_tx_size); 5321 netdev_dbg(adapter->netdev, "max_large_rx_size = %d\n", 5322 buf->max_large_rx_size); 5323 netdev_dbg(adapter->netdev, "ipv6_ext_hdr = %d\n", 5324 buf->ipv6_extension_header); 5325 netdev_dbg(adapter->netdev, "tcp_pseudosum_req = %d\n", 5326 buf->tcp_pseudosum_req); 5327 netdev_dbg(adapter->netdev, "num_ipv6_ext_hd = %d\n", 5328 buf->num_ipv6_ext_headers); 5329 netdev_dbg(adapter->netdev, "off_ipv6_ext_hd = %d\n", 5330 buf->off_ipv6_ext_headers); 5331 5332 send_control_ip_offload(adapter); 5333 } 5334 5335 static const char *ibmvnic_fw_err_cause(u16 cause) 5336 { 5337 switch (cause) { 5338 case ADAPTER_PROBLEM: 5339 return "adapter problem"; 5340 case BUS_PROBLEM: 5341 return "bus problem"; 5342 case FW_PROBLEM: 5343 return "firmware problem"; 5344 case DD_PROBLEM: 5345 return "device driver problem"; 5346 case EEH_RECOVERY: 5347 return "EEH recovery"; 5348 case FW_UPDATED: 5349 return "firmware updated"; 5350 case LOW_MEMORY: 5351 return "low Memory"; 5352 default: 5353 return "unknown"; 5354 } 5355 } 5356 5357 static void handle_error_indication(union ibmvnic_crq *crq, 5358 struct ibmvnic_adapter *adapter) 5359 { 5360 struct device *dev = &adapter->vdev->dev; 5361 u16 cause; 5362 5363 cause = be16_to_cpu(crq->error_indication.error_cause); 5364 5365 dev_warn_ratelimited(dev, 5366 "Firmware reports %serror, cause: %s. Starting recovery...\n", 5367 crq->error_indication.flags 5368 & IBMVNIC_FATAL_ERROR ? "FATAL " : "", 5369 ibmvnic_fw_err_cause(cause)); 5370 5371 if (crq->error_indication.flags & IBMVNIC_FATAL_ERROR) 5372 ibmvnic_reset(adapter, VNIC_RESET_FATAL); 5373 else 5374 ibmvnic_reset(adapter, VNIC_RESET_NON_FATAL); 5375 } 5376 5377 static int handle_change_mac_rsp(union ibmvnic_crq *crq, 5378 struct ibmvnic_adapter *adapter) 5379 { 5380 struct net_device *netdev = adapter->netdev; 5381 struct device *dev = &adapter->vdev->dev; 5382 long rc; 5383 5384 rc = crq->change_mac_addr_rsp.rc.code; 5385 if (rc) { 5386 dev_err(dev, "Error %ld in CHANGE_MAC_ADDR_RSP\n", rc); 5387 goto out; 5388 } 5389 /* crq->change_mac_addr.mac_addr is the requested one 5390 * crq->change_mac_addr_rsp.mac_addr is the returned valid one. 5391 */ 5392 eth_hw_addr_set(netdev, &crq->change_mac_addr_rsp.mac_addr[0]); 5393 ether_addr_copy(adapter->mac_addr, 5394 &crq->change_mac_addr_rsp.mac_addr[0]); 5395 out: 5396 complete(&adapter->fw_done); 5397 return rc; 5398 } 5399 5400 static void handle_request_cap_rsp(union ibmvnic_crq *crq, 5401 struct ibmvnic_adapter *adapter) 5402 { 5403 struct device *dev = &adapter->vdev->dev; 5404 u64 *req_value; 5405 char *name; 5406 5407 atomic_dec(&adapter->running_cap_crqs); 5408 netdev_dbg(adapter->netdev, "Outstanding request-caps: %d\n", 5409 atomic_read(&adapter->running_cap_crqs)); 5410 switch (be16_to_cpu(crq->request_capability_rsp.capability)) { 5411 case REQ_TX_QUEUES: 5412 req_value = &adapter->req_tx_queues; 5413 name = "tx"; 5414 break; 5415 case REQ_RX_QUEUES: 5416 req_value = &adapter->req_rx_queues; 5417 name = "rx"; 5418 break; 5419 case REQ_RX_ADD_QUEUES: 5420 req_value = &adapter->req_rx_add_queues; 5421 name = "rx_add"; 5422 break; 5423 case REQ_TX_ENTRIES_PER_SUBCRQ: 5424 req_value = &adapter->req_tx_entries_per_subcrq; 5425 name = "tx_entries_per_subcrq"; 5426 break; 5427 case REQ_RX_ADD_ENTRIES_PER_SUBCRQ: 5428 req_value = &adapter->req_rx_add_entries_per_subcrq; 5429 name = "rx_add_entries_per_subcrq"; 5430 break; 5431 case REQ_MTU: 5432 req_value = &adapter->req_mtu; 5433 name = "mtu"; 5434 break; 5435 case PROMISC_REQUESTED: 5436 req_value = &adapter->promisc; 5437 name = "promisc"; 5438 break; 5439 default: 5440 dev_err(dev, "Got invalid cap request rsp %d\n", 5441 crq->request_capability.capability); 5442 return; 5443 } 5444 5445 switch (crq->request_capability_rsp.rc.code) { 5446 case SUCCESS: 5447 break; 5448 case PARTIALSUCCESS: 5449 dev_info(dev, "req=%lld, rsp=%ld in %s queue, retrying.\n", 5450 *req_value, 5451 (long)be64_to_cpu(crq->request_capability_rsp.number), 5452 name); 5453 5454 if (be16_to_cpu(crq->request_capability_rsp.capability) == 5455 REQ_MTU) { 5456 pr_err("mtu of %llu is not supported. Reverting.\n", 5457 *req_value); 5458 *req_value = adapter->fallback.mtu; 5459 } else { 5460 *req_value = 5461 be64_to_cpu(crq->request_capability_rsp.number); 5462 } 5463 5464 send_request_cap(adapter, 1); 5465 return; 5466 default: 5467 dev_err(dev, "Error %d in request cap rsp\n", 5468 crq->request_capability_rsp.rc.code); 5469 return; 5470 } 5471 5472 /* Done receiving requested capabilities, query IP offload support */ 5473 if (atomic_read(&adapter->running_cap_crqs) == 0) 5474 send_query_ip_offload(adapter); 5475 } 5476 5477 static int handle_login_rsp(union ibmvnic_crq *login_rsp_crq, 5478 struct ibmvnic_adapter *adapter) 5479 { 5480 struct device *dev = &adapter->vdev->dev; 5481 struct net_device *netdev = adapter->netdev; 5482 struct ibmvnic_login_rsp_buffer *login_rsp = adapter->login_rsp_buf; 5483 struct ibmvnic_login_buffer *login = adapter->login_buf; 5484 u64 *tx_handle_array; 5485 u64 *rx_handle_array; 5486 int num_tx_pools; 5487 int num_rx_pools; 5488 u64 *size_array; 5489 u32 rsp_len; 5490 int i; 5491 5492 /* CHECK: Test/set of login_pending does not need to be atomic 5493 * because only ibmvnic_tasklet tests/clears this. 5494 */ 5495 if (!adapter->login_pending) { 5496 netdev_warn(netdev, "Ignoring unexpected login response\n"); 5497 return 0; 5498 } 5499 adapter->login_pending = false; 5500 5501 /* If the number of queues requested can't be allocated by the 5502 * server, the login response will return with code 1. We will need 5503 * to resend the login buffer with fewer queues requested. 5504 */ 5505 if (login_rsp_crq->generic.rc.code) { 5506 adapter->init_done_rc = login_rsp_crq->generic.rc.code; 5507 complete(&adapter->init_done); 5508 return 0; 5509 } 5510 5511 if (adapter->failover_pending) { 5512 adapter->init_done_rc = -EAGAIN; 5513 netdev_dbg(netdev, "Failover pending, ignoring login response\n"); 5514 complete(&adapter->init_done); 5515 /* login response buffer will be released on reset */ 5516 return 0; 5517 } 5518 5519 netdev->mtu = adapter->req_mtu - ETH_HLEN; 5520 5521 netdev_dbg(adapter->netdev, "Login Response Buffer:\n"); 5522 for (i = 0; i < (adapter->login_rsp_buf_sz - 1) / 8 + 1; i++) { 5523 netdev_dbg(adapter->netdev, "%016lx\n", 5524 ((unsigned long *)(adapter->login_rsp_buf))[i]); 5525 } 5526 5527 /* Sanity checks */ 5528 if (login->num_txcomp_subcrqs != login_rsp->num_txsubm_subcrqs || 5529 (be32_to_cpu(login->num_rxcomp_subcrqs) * 5530 adapter->req_rx_add_queues != 5531 be32_to_cpu(login_rsp->num_rxadd_subcrqs))) { 5532 dev_err(dev, "FATAL: Inconsistent login and login rsp\n"); 5533 ibmvnic_reset(adapter, VNIC_RESET_FATAL); 5534 return -EIO; 5535 } 5536 5537 rsp_len = be32_to_cpu(login_rsp->len); 5538 if (be32_to_cpu(login->login_rsp_len) < rsp_len || 5539 rsp_len <= be32_to_cpu(login_rsp->off_txsubm_subcrqs) || 5540 rsp_len <= be32_to_cpu(login_rsp->off_rxadd_subcrqs) || 5541 rsp_len <= be32_to_cpu(login_rsp->off_rxadd_buff_size) || 5542 rsp_len <= be32_to_cpu(login_rsp->off_supp_tx_desc)) { 5543 /* This can happen if a login request times out and there are 5544 * 2 outstanding login requests sent, the LOGIN_RSP crq 5545 * could have been for the older login request. So we are 5546 * parsing the newer response buffer which may be incomplete 5547 */ 5548 dev_err(dev, "FATAL: Login rsp offsets/lengths invalid\n"); 5549 ibmvnic_reset(adapter, VNIC_RESET_FATAL); 5550 return -EIO; 5551 } 5552 5553 size_array = (u64 *)((u8 *)(adapter->login_rsp_buf) + 5554 be32_to_cpu(adapter->login_rsp_buf->off_rxadd_buff_size)); 5555 /* variable buffer sizes are not supported, so just read the 5556 * first entry. 5557 */ 5558 adapter->cur_rx_buf_sz = be64_to_cpu(size_array[0]); 5559 5560 num_tx_pools = be32_to_cpu(adapter->login_rsp_buf->num_txsubm_subcrqs); 5561 num_rx_pools = be32_to_cpu(adapter->login_rsp_buf->num_rxadd_subcrqs); 5562 5563 tx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) + 5564 be32_to_cpu(adapter->login_rsp_buf->off_txsubm_subcrqs)); 5565 rx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) + 5566 be32_to_cpu(adapter->login_rsp_buf->off_rxadd_subcrqs)); 5567 5568 for (i = 0; i < num_tx_pools; i++) 5569 adapter->tx_scrq[i]->handle = tx_handle_array[i]; 5570 5571 for (i = 0; i < num_rx_pools; i++) 5572 adapter->rx_scrq[i]->handle = rx_handle_array[i]; 5573 5574 adapter->num_active_tx_scrqs = num_tx_pools; 5575 adapter->num_active_rx_scrqs = num_rx_pools; 5576 release_login_rsp_buffer(adapter); 5577 release_login_buffer(adapter); 5578 complete(&adapter->init_done); 5579 5580 return 0; 5581 } 5582 5583 static void handle_request_unmap_rsp(union ibmvnic_crq *crq, 5584 struct ibmvnic_adapter *adapter) 5585 { 5586 struct device *dev = &adapter->vdev->dev; 5587 long rc; 5588 5589 rc = crq->request_unmap_rsp.rc.code; 5590 if (rc) 5591 dev_err(dev, "Error %ld in REQUEST_UNMAP_RSP\n", rc); 5592 } 5593 5594 static void handle_query_map_rsp(union ibmvnic_crq *crq, 5595 struct ibmvnic_adapter *adapter) 5596 { 5597 struct net_device *netdev = adapter->netdev; 5598 struct device *dev = &adapter->vdev->dev; 5599 long rc; 5600 5601 rc = crq->query_map_rsp.rc.code; 5602 if (rc) { 5603 dev_err(dev, "Error %ld in QUERY_MAP_RSP\n", rc); 5604 return; 5605 } 5606 netdev_dbg(netdev, "page_size = %d\ntot_pages = %u\nfree_pages = %u\n", 5607 crq->query_map_rsp.page_size, 5608 __be32_to_cpu(crq->query_map_rsp.tot_pages), 5609 __be32_to_cpu(crq->query_map_rsp.free_pages)); 5610 } 5611 5612 static void handle_query_cap_rsp(union ibmvnic_crq *crq, 5613 struct ibmvnic_adapter *adapter) 5614 { 5615 struct net_device *netdev = adapter->netdev; 5616 struct device *dev = &adapter->vdev->dev; 5617 long rc; 5618 5619 atomic_dec(&adapter->running_cap_crqs); 5620 netdev_dbg(netdev, "Outstanding queries: %d\n", 5621 atomic_read(&adapter->running_cap_crqs)); 5622 rc = crq->query_capability.rc.code; 5623 if (rc) { 5624 dev_err(dev, "Error %ld in QUERY_CAP_RSP\n", rc); 5625 goto out; 5626 } 5627 5628 switch (be16_to_cpu(crq->query_capability.capability)) { 5629 case MIN_TX_QUEUES: 5630 adapter->min_tx_queues = 5631 be64_to_cpu(crq->query_capability.number); 5632 netdev_dbg(netdev, "min_tx_queues = %lld\n", 5633 adapter->min_tx_queues); 5634 break; 5635 case MIN_RX_QUEUES: 5636 adapter->min_rx_queues = 5637 be64_to_cpu(crq->query_capability.number); 5638 netdev_dbg(netdev, "min_rx_queues = %lld\n", 5639 adapter->min_rx_queues); 5640 break; 5641 case MIN_RX_ADD_QUEUES: 5642 adapter->min_rx_add_queues = 5643 be64_to_cpu(crq->query_capability.number); 5644 netdev_dbg(netdev, "min_rx_add_queues = %lld\n", 5645 adapter->min_rx_add_queues); 5646 break; 5647 case MAX_TX_QUEUES: 5648 adapter->max_tx_queues = 5649 be64_to_cpu(crq->query_capability.number); 5650 netdev_dbg(netdev, "max_tx_queues = %lld\n", 5651 adapter->max_tx_queues); 5652 break; 5653 case MAX_RX_QUEUES: 5654 adapter->max_rx_queues = 5655 be64_to_cpu(crq->query_capability.number); 5656 netdev_dbg(netdev, "max_rx_queues = %lld\n", 5657 adapter->max_rx_queues); 5658 break; 5659 case MAX_RX_ADD_QUEUES: 5660 adapter->max_rx_add_queues = 5661 be64_to_cpu(crq->query_capability.number); 5662 netdev_dbg(netdev, "max_rx_add_queues = %lld\n", 5663 adapter->max_rx_add_queues); 5664 break; 5665 case MIN_TX_ENTRIES_PER_SUBCRQ: 5666 adapter->min_tx_entries_per_subcrq = 5667 be64_to_cpu(crq->query_capability.number); 5668 netdev_dbg(netdev, "min_tx_entries_per_subcrq = %lld\n", 5669 adapter->min_tx_entries_per_subcrq); 5670 break; 5671 case MIN_RX_ADD_ENTRIES_PER_SUBCRQ: 5672 adapter->min_rx_add_entries_per_subcrq = 5673 be64_to_cpu(crq->query_capability.number); 5674 netdev_dbg(netdev, "min_rx_add_entrs_per_subcrq = %lld\n", 5675 adapter->min_rx_add_entries_per_subcrq); 5676 break; 5677 case MAX_TX_ENTRIES_PER_SUBCRQ: 5678 adapter->max_tx_entries_per_subcrq = 5679 be64_to_cpu(crq->query_capability.number); 5680 netdev_dbg(netdev, "max_tx_entries_per_subcrq = %lld\n", 5681 adapter->max_tx_entries_per_subcrq); 5682 break; 5683 case MAX_RX_ADD_ENTRIES_PER_SUBCRQ: 5684 adapter->max_rx_add_entries_per_subcrq = 5685 be64_to_cpu(crq->query_capability.number); 5686 netdev_dbg(netdev, "max_rx_add_entrs_per_subcrq = %lld\n", 5687 adapter->max_rx_add_entries_per_subcrq); 5688 break; 5689 case TCP_IP_OFFLOAD: 5690 adapter->tcp_ip_offload = 5691 be64_to_cpu(crq->query_capability.number); 5692 netdev_dbg(netdev, "tcp_ip_offload = %lld\n", 5693 adapter->tcp_ip_offload); 5694 break; 5695 case PROMISC_SUPPORTED: 5696 adapter->promisc_supported = 5697 be64_to_cpu(crq->query_capability.number); 5698 netdev_dbg(netdev, "promisc_supported = %lld\n", 5699 adapter->promisc_supported); 5700 break; 5701 case MIN_MTU: 5702 adapter->min_mtu = be64_to_cpu(crq->query_capability.number); 5703 netdev->min_mtu = adapter->min_mtu - ETH_HLEN; 5704 netdev_dbg(netdev, "min_mtu = %lld\n", adapter->min_mtu); 5705 break; 5706 case MAX_MTU: 5707 adapter->max_mtu = be64_to_cpu(crq->query_capability.number); 5708 netdev->max_mtu = adapter->max_mtu - ETH_HLEN; 5709 netdev_dbg(netdev, "max_mtu = %lld\n", adapter->max_mtu); 5710 break; 5711 case MAX_MULTICAST_FILTERS: 5712 adapter->max_multicast_filters = 5713 be64_to_cpu(crq->query_capability.number); 5714 netdev_dbg(netdev, "max_multicast_filters = %lld\n", 5715 adapter->max_multicast_filters); 5716 break; 5717 case VLAN_HEADER_INSERTION: 5718 adapter->vlan_header_insertion = 5719 be64_to_cpu(crq->query_capability.number); 5720 if (adapter->vlan_header_insertion) 5721 netdev->features |= NETIF_F_HW_VLAN_STAG_TX; 5722 netdev_dbg(netdev, "vlan_header_insertion = %lld\n", 5723 adapter->vlan_header_insertion); 5724 break; 5725 case RX_VLAN_HEADER_INSERTION: 5726 adapter->rx_vlan_header_insertion = 5727 be64_to_cpu(crq->query_capability.number); 5728 netdev_dbg(netdev, "rx_vlan_header_insertion = %lld\n", 5729 adapter->rx_vlan_header_insertion); 5730 break; 5731 case MAX_TX_SG_ENTRIES: 5732 adapter->max_tx_sg_entries = 5733 be64_to_cpu(crq->query_capability.number); 5734 netdev_dbg(netdev, "max_tx_sg_entries = %lld\n", 5735 adapter->max_tx_sg_entries); 5736 break; 5737 case RX_SG_SUPPORTED: 5738 adapter->rx_sg_supported = 5739 be64_to_cpu(crq->query_capability.number); 5740 netdev_dbg(netdev, "rx_sg_supported = %lld\n", 5741 adapter->rx_sg_supported); 5742 break; 5743 case OPT_TX_COMP_SUB_QUEUES: 5744 adapter->opt_tx_comp_sub_queues = 5745 be64_to_cpu(crq->query_capability.number); 5746 netdev_dbg(netdev, "opt_tx_comp_sub_queues = %lld\n", 5747 adapter->opt_tx_comp_sub_queues); 5748 break; 5749 case OPT_RX_COMP_QUEUES: 5750 adapter->opt_rx_comp_queues = 5751 be64_to_cpu(crq->query_capability.number); 5752 netdev_dbg(netdev, "opt_rx_comp_queues = %lld\n", 5753 adapter->opt_rx_comp_queues); 5754 break; 5755 case OPT_RX_BUFADD_Q_PER_RX_COMP_Q: 5756 adapter->opt_rx_bufadd_q_per_rx_comp_q = 5757 be64_to_cpu(crq->query_capability.number); 5758 netdev_dbg(netdev, "opt_rx_bufadd_q_per_rx_comp_q = %lld\n", 5759 adapter->opt_rx_bufadd_q_per_rx_comp_q); 5760 break; 5761 case OPT_TX_ENTRIES_PER_SUBCRQ: 5762 adapter->opt_tx_entries_per_subcrq = 5763 be64_to_cpu(crq->query_capability.number); 5764 netdev_dbg(netdev, "opt_tx_entries_per_subcrq = %lld\n", 5765 adapter->opt_tx_entries_per_subcrq); 5766 break; 5767 case OPT_RXBA_ENTRIES_PER_SUBCRQ: 5768 adapter->opt_rxba_entries_per_subcrq = 5769 be64_to_cpu(crq->query_capability.number); 5770 netdev_dbg(netdev, "opt_rxba_entries_per_subcrq = %lld\n", 5771 adapter->opt_rxba_entries_per_subcrq); 5772 break; 5773 case TX_RX_DESC_REQ: 5774 adapter->tx_rx_desc_req = crq->query_capability.number; 5775 netdev_dbg(netdev, "tx_rx_desc_req = %llx\n", 5776 adapter->tx_rx_desc_req); 5777 break; 5778 5779 default: 5780 netdev_err(netdev, "Got invalid cap rsp %d\n", 5781 crq->query_capability.capability); 5782 } 5783 5784 out: 5785 if (atomic_read(&adapter->running_cap_crqs) == 0) 5786 send_request_cap(adapter, 0); 5787 } 5788 5789 static int send_query_phys_parms(struct ibmvnic_adapter *adapter) 5790 { 5791 union ibmvnic_crq crq; 5792 int rc; 5793 5794 memset(&crq, 0, sizeof(crq)); 5795 crq.query_phys_parms.first = IBMVNIC_CRQ_CMD; 5796 crq.query_phys_parms.cmd = QUERY_PHYS_PARMS; 5797 5798 mutex_lock(&adapter->fw_lock); 5799 adapter->fw_done_rc = 0; 5800 reinit_completion(&adapter->fw_done); 5801 5802 rc = ibmvnic_send_crq(adapter, &crq); 5803 if (rc) { 5804 mutex_unlock(&adapter->fw_lock); 5805 return rc; 5806 } 5807 5808 rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000); 5809 if (rc) { 5810 mutex_unlock(&adapter->fw_lock); 5811 return rc; 5812 } 5813 5814 mutex_unlock(&adapter->fw_lock); 5815 return adapter->fw_done_rc ? -EIO : 0; 5816 } 5817 5818 static int handle_query_phys_parms_rsp(union ibmvnic_crq *crq, 5819 struct ibmvnic_adapter *adapter) 5820 { 5821 struct net_device *netdev = adapter->netdev; 5822 int rc; 5823 __be32 rspeed = cpu_to_be32(crq->query_phys_parms_rsp.speed); 5824 5825 rc = crq->query_phys_parms_rsp.rc.code; 5826 if (rc) { 5827 netdev_err(netdev, "Error %d in QUERY_PHYS_PARMS\n", rc); 5828 return rc; 5829 } 5830 switch (rspeed) { 5831 case IBMVNIC_10MBPS: 5832 adapter->speed = SPEED_10; 5833 break; 5834 case IBMVNIC_100MBPS: 5835 adapter->speed = SPEED_100; 5836 break; 5837 case IBMVNIC_1GBPS: 5838 adapter->speed = SPEED_1000; 5839 break; 5840 case IBMVNIC_10GBPS: 5841 adapter->speed = SPEED_10000; 5842 break; 5843 case IBMVNIC_25GBPS: 5844 adapter->speed = SPEED_25000; 5845 break; 5846 case IBMVNIC_40GBPS: 5847 adapter->speed = SPEED_40000; 5848 break; 5849 case IBMVNIC_50GBPS: 5850 adapter->speed = SPEED_50000; 5851 break; 5852 case IBMVNIC_100GBPS: 5853 adapter->speed = SPEED_100000; 5854 break; 5855 case IBMVNIC_200GBPS: 5856 adapter->speed = SPEED_200000; 5857 break; 5858 default: 5859 if (netif_carrier_ok(netdev)) 5860 netdev_warn(netdev, "Unknown speed 0x%08x\n", rspeed); 5861 adapter->speed = SPEED_UNKNOWN; 5862 } 5863 if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_FULL_DUPLEX) 5864 adapter->duplex = DUPLEX_FULL; 5865 else if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_HALF_DUPLEX) 5866 adapter->duplex = DUPLEX_HALF; 5867 else 5868 adapter->duplex = DUPLEX_UNKNOWN; 5869 5870 return rc; 5871 } 5872 5873 static void ibmvnic_handle_crq(union ibmvnic_crq *crq, 5874 struct ibmvnic_adapter *adapter) 5875 { 5876 struct ibmvnic_generic_crq *gen_crq = &crq->generic; 5877 struct net_device *netdev = adapter->netdev; 5878 struct device *dev = &adapter->vdev->dev; 5879 u64 *u64_crq = (u64 *)crq; 5880 long rc; 5881 5882 netdev_dbg(netdev, "Handling CRQ: %016lx %016lx\n", 5883 (unsigned long)cpu_to_be64(u64_crq[0]), 5884 (unsigned long)cpu_to_be64(u64_crq[1])); 5885 switch (gen_crq->first) { 5886 case IBMVNIC_CRQ_INIT_RSP: 5887 switch (gen_crq->cmd) { 5888 case IBMVNIC_CRQ_INIT: 5889 dev_info(dev, "Partner initialized\n"); 5890 adapter->from_passive_init = true; 5891 /* Discard any stale login responses from prev reset. 5892 * CHECK: should we clear even on INIT_COMPLETE? 5893 */ 5894 adapter->login_pending = false; 5895 5896 if (adapter->state == VNIC_DOWN) 5897 rc = ibmvnic_reset(adapter, VNIC_RESET_PASSIVE_INIT); 5898 else 5899 rc = ibmvnic_reset(adapter, VNIC_RESET_FAILOVER); 5900 5901 if (rc && rc != -EBUSY) { 5902 /* We were unable to schedule the failover 5903 * reset either because the adapter was still 5904 * probing (eg: during kexec) or we could not 5905 * allocate memory. Clear the failover_pending 5906 * flag since no one else will. We ignore 5907 * EBUSY because it means either FAILOVER reset 5908 * is already scheduled or the adapter is 5909 * being removed. 5910 */ 5911 netdev_err(netdev, 5912 "Error %ld scheduling failover reset\n", 5913 rc); 5914 adapter->failover_pending = false; 5915 } 5916 5917 if (!completion_done(&adapter->init_done)) { 5918 if (!adapter->init_done_rc) 5919 adapter->init_done_rc = -EAGAIN; 5920 complete(&adapter->init_done); 5921 } 5922 5923 break; 5924 case IBMVNIC_CRQ_INIT_COMPLETE: 5925 dev_info(dev, "Partner initialization complete\n"); 5926 adapter->crq.active = true; 5927 send_version_xchg(adapter); 5928 break; 5929 default: 5930 dev_err(dev, "Unknown crq cmd: %d\n", gen_crq->cmd); 5931 } 5932 return; 5933 case IBMVNIC_CRQ_XPORT_EVENT: 5934 netif_carrier_off(netdev); 5935 adapter->crq.active = false; 5936 /* terminate any thread waiting for a response 5937 * from the device 5938 */ 5939 if (!completion_done(&adapter->fw_done)) { 5940 adapter->fw_done_rc = -EIO; 5941 complete(&adapter->fw_done); 5942 } 5943 5944 /* if we got here during crq-init, retry crq-init */ 5945 if (!completion_done(&adapter->init_done)) { 5946 adapter->init_done_rc = -EAGAIN; 5947 complete(&adapter->init_done); 5948 } 5949 5950 if (!completion_done(&adapter->stats_done)) 5951 complete(&adapter->stats_done); 5952 if (test_bit(0, &adapter->resetting)) 5953 adapter->force_reset_recovery = true; 5954 if (gen_crq->cmd == IBMVNIC_PARTITION_MIGRATED) { 5955 dev_info(dev, "Migrated, re-enabling adapter\n"); 5956 ibmvnic_reset(adapter, VNIC_RESET_MOBILITY); 5957 } else if (gen_crq->cmd == IBMVNIC_DEVICE_FAILOVER) { 5958 dev_info(dev, "Backing device failover detected\n"); 5959 adapter->failover_pending = true; 5960 } else { 5961 /* The adapter lost the connection */ 5962 dev_err(dev, "Virtual Adapter failed (rc=%d)\n", 5963 gen_crq->cmd); 5964 ibmvnic_reset(adapter, VNIC_RESET_FATAL); 5965 } 5966 return; 5967 case IBMVNIC_CRQ_CMD_RSP: 5968 break; 5969 default: 5970 dev_err(dev, "Got an invalid msg type 0x%02x\n", 5971 gen_crq->first); 5972 return; 5973 } 5974 5975 switch (gen_crq->cmd) { 5976 case VERSION_EXCHANGE_RSP: 5977 rc = crq->version_exchange_rsp.rc.code; 5978 if (rc) { 5979 dev_err(dev, "Error %ld in VERSION_EXCHG_RSP\n", rc); 5980 break; 5981 } 5982 ibmvnic_version = 5983 be16_to_cpu(crq->version_exchange_rsp.version); 5984 dev_info(dev, "Partner protocol version is %d\n", 5985 ibmvnic_version); 5986 send_query_cap(adapter); 5987 break; 5988 case QUERY_CAPABILITY_RSP: 5989 handle_query_cap_rsp(crq, adapter); 5990 break; 5991 case QUERY_MAP_RSP: 5992 handle_query_map_rsp(crq, adapter); 5993 break; 5994 case REQUEST_MAP_RSP: 5995 adapter->fw_done_rc = crq->request_map_rsp.rc.code; 5996 complete(&adapter->fw_done); 5997 break; 5998 case REQUEST_UNMAP_RSP: 5999 handle_request_unmap_rsp(crq, adapter); 6000 break; 6001 case REQUEST_CAPABILITY_RSP: 6002 handle_request_cap_rsp(crq, adapter); 6003 break; 6004 case LOGIN_RSP: 6005 netdev_dbg(netdev, "Got Login Response\n"); 6006 handle_login_rsp(crq, adapter); 6007 break; 6008 case LOGICAL_LINK_STATE_RSP: 6009 netdev_dbg(netdev, 6010 "Got Logical Link State Response, state: %d rc: %d\n", 6011 crq->logical_link_state_rsp.link_state, 6012 crq->logical_link_state_rsp.rc.code); 6013 adapter->logical_link_state = 6014 crq->logical_link_state_rsp.link_state; 6015 adapter->init_done_rc = crq->logical_link_state_rsp.rc.code; 6016 complete(&adapter->init_done); 6017 break; 6018 case LINK_STATE_INDICATION: 6019 netdev_dbg(netdev, "Got Logical Link State Indication\n"); 6020 adapter->phys_link_state = 6021 crq->link_state_indication.phys_link_state; 6022 adapter->logical_link_state = 6023 crq->link_state_indication.logical_link_state; 6024 if (adapter->phys_link_state && adapter->logical_link_state) 6025 netif_carrier_on(netdev); 6026 else 6027 netif_carrier_off(netdev); 6028 break; 6029 case CHANGE_MAC_ADDR_RSP: 6030 netdev_dbg(netdev, "Got MAC address change Response\n"); 6031 adapter->fw_done_rc = handle_change_mac_rsp(crq, adapter); 6032 break; 6033 case ERROR_INDICATION: 6034 netdev_dbg(netdev, "Got Error Indication\n"); 6035 handle_error_indication(crq, adapter); 6036 break; 6037 case REQUEST_STATISTICS_RSP: 6038 netdev_dbg(netdev, "Got Statistics Response\n"); 6039 complete(&adapter->stats_done); 6040 break; 6041 case QUERY_IP_OFFLOAD_RSP: 6042 netdev_dbg(netdev, "Got Query IP offload Response\n"); 6043 handle_query_ip_offload_rsp(adapter); 6044 break; 6045 case MULTICAST_CTRL_RSP: 6046 netdev_dbg(netdev, "Got multicast control Response\n"); 6047 break; 6048 case CONTROL_IP_OFFLOAD_RSP: 6049 netdev_dbg(netdev, "Got Control IP offload Response\n"); 6050 dma_unmap_single(dev, adapter->ip_offload_ctrl_tok, 6051 sizeof(adapter->ip_offload_ctrl), 6052 DMA_TO_DEVICE); 6053 complete(&adapter->init_done); 6054 break; 6055 case COLLECT_FW_TRACE_RSP: 6056 netdev_dbg(netdev, "Got Collect firmware trace Response\n"); 6057 complete(&adapter->fw_done); 6058 break; 6059 case GET_VPD_SIZE_RSP: 6060 handle_vpd_size_rsp(crq, adapter); 6061 break; 6062 case GET_VPD_RSP: 6063 handle_vpd_rsp(crq, adapter); 6064 break; 6065 case QUERY_PHYS_PARMS_RSP: 6066 adapter->fw_done_rc = handle_query_phys_parms_rsp(crq, adapter); 6067 complete(&adapter->fw_done); 6068 break; 6069 default: 6070 netdev_err(netdev, "Got an invalid cmd type 0x%02x\n", 6071 gen_crq->cmd); 6072 } 6073 } 6074 6075 static irqreturn_t ibmvnic_interrupt(int irq, void *instance) 6076 { 6077 struct ibmvnic_adapter *adapter = instance; 6078 6079 tasklet_schedule(&adapter->tasklet); 6080 return IRQ_HANDLED; 6081 } 6082 6083 static void ibmvnic_tasklet(struct tasklet_struct *t) 6084 { 6085 struct ibmvnic_adapter *adapter = from_tasklet(adapter, t, tasklet); 6086 struct ibmvnic_crq_queue *queue = &adapter->crq; 6087 union ibmvnic_crq *crq; 6088 unsigned long flags; 6089 6090 spin_lock_irqsave(&queue->lock, flags); 6091 6092 /* Pull all the valid messages off the CRQ */ 6093 while ((crq = ibmvnic_next_crq(adapter)) != NULL) { 6094 /* This barrier makes sure ibmvnic_next_crq()'s 6095 * crq->generic.first & IBMVNIC_CRQ_CMD_RSP is loaded 6096 * before ibmvnic_handle_crq()'s 6097 * switch(gen_crq->first) and switch(gen_crq->cmd). 6098 */ 6099 dma_rmb(); 6100 ibmvnic_handle_crq(crq, adapter); 6101 crq->generic.first = 0; 6102 } 6103 6104 spin_unlock_irqrestore(&queue->lock, flags); 6105 } 6106 6107 static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *adapter) 6108 { 6109 struct vio_dev *vdev = adapter->vdev; 6110 int rc; 6111 6112 do { 6113 rc = plpar_hcall_norets(H_ENABLE_CRQ, vdev->unit_address); 6114 } while (rc == H_IN_PROGRESS || rc == H_BUSY || H_IS_LONG_BUSY(rc)); 6115 6116 if (rc) 6117 dev_err(&vdev->dev, "Error enabling adapter (rc=%d)\n", rc); 6118 6119 return rc; 6120 } 6121 6122 static int ibmvnic_reset_crq(struct ibmvnic_adapter *adapter) 6123 { 6124 struct ibmvnic_crq_queue *crq = &adapter->crq; 6125 struct device *dev = &adapter->vdev->dev; 6126 struct vio_dev *vdev = adapter->vdev; 6127 int rc; 6128 6129 /* Close the CRQ */ 6130 do { 6131 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address); 6132 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc)); 6133 6134 /* Clean out the queue */ 6135 if (!crq->msgs) 6136 return -EINVAL; 6137 6138 memset(crq->msgs, 0, PAGE_SIZE); 6139 crq->cur = 0; 6140 crq->active = false; 6141 6142 /* And re-open it again */ 6143 rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address, 6144 crq->msg_token, PAGE_SIZE); 6145 6146 if (rc == H_CLOSED) 6147 /* Adapter is good, but other end is not ready */ 6148 dev_warn(dev, "Partner adapter not ready\n"); 6149 else if (rc != 0) 6150 dev_warn(dev, "Couldn't register crq (rc=%d)\n", rc); 6151 6152 return rc; 6153 } 6154 6155 static void release_crq_queue(struct ibmvnic_adapter *adapter) 6156 { 6157 struct ibmvnic_crq_queue *crq = &adapter->crq; 6158 struct vio_dev *vdev = adapter->vdev; 6159 long rc; 6160 6161 if (!crq->msgs) 6162 return; 6163 6164 netdev_dbg(adapter->netdev, "Releasing CRQ\n"); 6165 free_irq(vdev->irq, adapter); 6166 tasklet_kill(&adapter->tasklet); 6167 do { 6168 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address); 6169 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc)); 6170 6171 dma_unmap_single(&vdev->dev, crq->msg_token, PAGE_SIZE, 6172 DMA_BIDIRECTIONAL); 6173 free_page((unsigned long)crq->msgs); 6174 crq->msgs = NULL; 6175 crq->active = false; 6176 } 6177 6178 static int init_crq_queue(struct ibmvnic_adapter *adapter) 6179 { 6180 struct ibmvnic_crq_queue *crq = &adapter->crq; 6181 struct device *dev = &adapter->vdev->dev; 6182 struct vio_dev *vdev = adapter->vdev; 6183 int rc, retrc = -ENOMEM; 6184 6185 if (crq->msgs) 6186 return 0; 6187 6188 crq->msgs = (union ibmvnic_crq *)get_zeroed_page(GFP_KERNEL); 6189 /* Should we allocate more than one page? */ 6190 6191 if (!crq->msgs) 6192 return -ENOMEM; 6193 6194 crq->size = PAGE_SIZE / sizeof(*crq->msgs); 6195 crq->msg_token = dma_map_single(dev, crq->msgs, PAGE_SIZE, 6196 DMA_BIDIRECTIONAL); 6197 if (dma_mapping_error(dev, crq->msg_token)) 6198 goto map_failed; 6199 6200 rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address, 6201 crq->msg_token, PAGE_SIZE); 6202 6203 if (rc == H_RESOURCE) 6204 /* maybe kexecing and resource is busy. try a reset */ 6205 rc = ibmvnic_reset_crq(adapter); 6206 retrc = rc; 6207 6208 if (rc == H_CLOSED) { 6209 dev_warn(dev, "Partner adapter not ready\n"); 6210 } else if (rc) { 6211 dev_warn(dev, "Error %d opening adapter\n", rc); 6212 goto reg_crq_failed; 6213 } 6214 6215 retrc = 0; 6216 6217 tasklet_setup(&adapter->tasklet, (void *)ibmvnic_tasklet); 6218 6219 netdev_dbg(adapter->netdev, "registering irq 0x%x\n", vdev->irq); 6220 snprintf(crq->name, sizeof(crq->name), "ibmvnic-%x", 6221 adapter->vdev->unit_address); 6222 rc = request_irq(vdev->irq, ibmvnic_interrupt, 0, crq->name, adapter); 6223 if (rc) { 6224 dev_err(dev, "Couldn't register irq 0x%x. rc=%d\n", 6225 vdev->irq, rc); 6226 goto req_irq_failed; 6227 } 6228 6229 rc = vio_enable_interrupts(vdev); 6230 if (rc) { 6231 dev_err(dev, "Error %d enabling interrupts\n", rc); 6232 goto req_irq_failed; 6233 } 6234 6235 crq->cur = 0; 6236 spin_lock_init(&crq->lock); 6237 6238 /* process any CRQs that were queued before we enabled interrupts */ 6239 tasklet_schedule(&adapter->tasklet); 6240 6241 return retrc; 6242 6243 req_irq_failed: 6244 tasklet_kill(&adapter->tasklet); 6245 do { 6246 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address); 6247 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc)); 6248 reg_crq_failed: 6249 dma_unmap_single(dev, crq->msg_token, PAGE_SIZE, DMA_BIDIRECTIONAL); 6250 map_failed: 6251 free_page((unsigned long)crq->msgs); 6252 crq->msgs = NULL; 6253 return retrc; 6254 } 6255 6256 static int ibmvnic_reset_init(struct ibmvnic_adapter *adapter, bool reset) 6257 { 6258 struct device *dev = &adapter->vdev->dev; 6259 unsigned long timeout = msecs_to_jiffies(20000); 6260 u64 old_num_rx_queues = adapter->req_rx_queues; 6261 u64 old_num_tx_queues = adapter->req_tx_queues; 6262 int rc; 6263 6264 adapter->from_passive_init = false; 6265 6266 rc = ibmvnic_send_crq_init(adapter); 6267 if (rc) { 6268 dev_err(dev, "Send crq init failed with error %d\n", rc); 6269 return rc; 6270 } 6271 6272 if (!wait_for_completion_timeout(&adapter->init_done, timeout)) { 6273 dev_err(dev, "Initialization sequence timed out\n"); 6274 return -ETIMEDOUT; 6275 } 6276 6277 if (adapter->init_done_rc) { 6278 release_crq_queue(adapter); 6279 dev_err(dev, "CRQ-init failed, %d\n", adapter->init_done_rc); 6280 return adapter->init_done_rc; 6281 } 6282 6283 if (adapter->from_passive_init) { 6284 adapter->state = VNIC_OPEN; 6285 adapter->from_passive_init = false; 6286 dev_err(dev, "CRQ-init failed, passive-init\n"); 6287 return -EINVAL; 6288 } 6289 6290 if (reset && 6291 test_bit(0, &adapter->resetting) && !adapter->wait_for_reset && 6292 adapter->reset_reason != VNIC_RESET_MOBILITY) { 6293 if (adapter->req_rx_queues != old_num_rx_queues || 6294 adapter->req_tx_queues != old_num_tx_queues) { 6295 release_sub_crqs(adapter, 0); 6296 rc = init_sub_crqs(adapter); 6297 } else { 6298 /* no need to reinitialize completely, but we do 6299 * need to clean up transmits that were in flight 6300 * when we processed the reset. Failure to do so 6301 * will confound the upper layer, usually TCP, by 6302 * creating the illusion of transmits that are 6303 * awaiting completion. 6304 */ 6305 clean_tx_pools(adapter); 6306 6307 rc = reset_sub_crq_queues(adapter); 6308 } 6309 } else { 6310 rc = init_sub_crqs(adapter); 6311 } 6312 6313 if (rc) { 6314 dev_err(dev, "Initialization of sub crqs failed\n"); 6315 release_crq_queue(adapter); 6316 return rc; 6317 } 6318 6319 rc = init_sub_crq_irqs(adapter); 6320 if (rc) { 6321 dev_err(dev, "Failed to initialize sub crq irqs\n"); 6322 release_crq_queue(adapter); 6323 } 6324 6325 return rc; 6326 } 6327 6328 static struct device_attribute dev_attr_failover; 6329 6330 static int ibmvnic_probe(struct vio_dev *dev, const struct vio_device_id *id) 6331 { 6332 struct ibmvnic_adapter *adapter; 6333 struct net_device *netdev; 6334 unsigned char *mac_addr_p; 6335 unsigned long flags; 6336 bool init_success; 6337 int rc; 6338 6339 dev_dbg(&dev->dev, "entering ibmvnic_probe for UA 0x%x\n", 6340 dev->unit_address); 6341 6342 mac_addr_p = (unsigned char *)vio_get_attribute(dev, 6343 VETH_MAC_ADDR, NULL); 6344 if (!mac_addr_p) { 6345 dev_err(&dev->dev, 6346 "(%s:%3.3d) ERROR: Can't find MAC_ADDR attribute\n", 6347 __FILE__, __LINE__); 6348 return 0; 6349 } 6350 6351 netdev = alloc_etherdev_mq(sizeof(struct ibmvnic_adapter), 6352 IBMVNIC_MAX_QUEUES); 6353 if (!netdev) 6354 return -ENOMEM; 6355 6356 adapter = netdev_priv(netdev); 6357 adapter->state = VNIC_PROBING; 6358 dev_set_drvdata(&dev->dev, netdev); 6359 adapter->vdev = dev; 6360 adapter->netdev = netdev; 6361 adapter->login_pending = false; 6362 memset(&adapter->map_ids, 0, sizeof(adapter->map_ids)); 6363 /* map_ids start at 1, so ensure map_id 0 is always "in-use" */ 6364 bitmap_set(adapter->map_ids, 0, 1); 6365 6366 ether_addr_copy(adapter->mac_addr, mac_addr_p); 6367 eth_hw_addr_set(netdev, adapter->mac_addr); 6368 netdev->irq = dev->irq; 6369 netdev->netdev_ops = &ibmvnic_netdev_ops; 6370 netdev->ethtool_ops = &ibmvnic_ethtool_ops; 6371 SET_NETDEV_DEV(netdev, &dev->dev); 6372 6373 INIT_WORK(&adapter->ibmvnic_reset, __ibmvnic_reset); 6374 INIT_DELAYED_WORK(&adapter->ibmvnic_delayed_reset, 6375 __ibmvnic_delayed_reset); 6376 INIT_LIST_HEAD(&adapter->rwi_list); 6377 spin_lock_init(&adapter->rwi_lock); 6378 spin_lock_init(&adapter->state_lock); 6379 mutex_init(&adapter->fw_lock); 6380 init_completion(&adapter->probe_done); 6381 init_completion(&adapter->init_done); 6382 init_completion(&adapter->fw_done); 6383 init_completion(&adapter->reset_done); 6384 init_completion(&adapter->stats_done); 6385 clear_bit(0, &adapter->resetting); 6386 adapter->prev_rx_buf_sz = 0; 6387 adapter->prev_mtu = 0; 6388 6389 init_success = false; 6390 do { 6391 reinit_init_done(adapter); 6392 6393 /* clear any failovers we got in the previous pass 6394 * since we are reinitializing the CRQ 6395 */ 6396 adapter->failover_pending = false; 6397 6398 /* If we had already initialized CRQ, we may have one or 6399 * more resets queued already. Discard those and release 6400 * the CRQ before initializing the CRQ again. 6401 */ 6402 release_crq_queue(adapter); 6403 6404 /* Since we are still in PROBING state, __ibmvnic_reset() 6405 * will not access the ->rwi_list and since we released CRQ, 6406 * we won't get _new_ transport events. But there maybe an 6407 * ongoing ibmvnic_reset() call. So serialize access to 6408 * rwi_list. If we win the race, ibvmnic_reset() could add 6409 * a reset after we purged but thats ok - we just may end 6410 * up with an extra reset (i.e similar to having two or more 6411 * resets in the queue at once). 6412 * CHECK. 6413 */ 6414 spin_lock_irqsave(&adapter->rwi_lock, flags); 6415 flush_reset_queue(adapter); 6416 spin_unlock_irqrestore(&adapter->rwi_lock, flags); 6417 6418 rc = init_crq_queue(adapter); 6419 if (rc) { 6420 dev_err(&dev->dev, "Couldn't initialize crq. rc=%d\n", 6421 rc); 6422 goto ibmvnic_init_fail; 6423 } 6424 6425 rc = ibmvnic_reset_init(adapter, false); 6426 } while (rc == -EAGAIN); 6427 6428 /* We are ignoring the error from ibmvnic_reset_init() assuming that the 6429 * partner is not ready. CRQ is not active. When the partner becomes 6430 * ready, we will do the passive init reset. 6431 */ 6432 6433 if (!rc) 6434 init_success = true; 6435 6436 rc = init_stats_buffers(adapter); 6437 if (rc) 6438 goto ibmvnic_init_fail; 6439 6440 rc = init_stats_token(adapter); 6441 if (rc) 6442 goto ibmvnic_stats_fail; 6443 6444 rc = device_create_file(&dev->dev, &dev_attr_failover); 6445 if (rc) 6446 goto ibmvnic_dev_file_err; 6447 6448 netif_carrier_off(netdev); 6449 6450 if (init_success) { 6451 adapter->state = VNIC_PROBED; 6452 netdev->mtu = adapter->req_mtu - ETH_HLEN; 6453 netdev->min_mtu = adapter->min_mtu - ETH_HLEN; 6454 netdev->max_mtu = adapter->max_mtu - ETH_HLEN; 6455 } else { 6456 adapter->state = VNIC_DOWN; 6457 } 6458 6459 adapter->wait_for_reset = false; 6460 adapter->last_reset_time = jiffies; 6461 6462 rc = register_netdev(netdev); 6463 if (rc) { 6464 dev_err(&dev->dev, "failed to register netdev rc=%d\n", rc); 6465 goto ibmvnic_register_fail; 6466 } 6467 dev_info(&dev->dev, "ibmvnic registered\n"); 6468 6469 rc = ibmvnic_cpu_notif_add(adapter); 6470 if (rc) { 6471 netdev_err(netdev, "Registering cpu notifier failed\n"); 6472 goto cpu_notif_add_failed; 6473 } 6474 6475 complete(&adapter->probe_done); 6476 6477 return 0; 6478 6479 cpu_notif_add_failed: 6480 unregister_netdev(netdev); 6481 6482 ibmvnic_register_fail: 6483 device_remove_file(&dev->dev, &dev_attr_failover); 6484 6485 ibmvnic_dev_file_err: 6486 release_stats_token(adapter); 6487 6488 ibmvnic_stats_fail: 6489 release_stats_buffers(adapter); 6490 6491 ibmvnic_init_fail: 6492 release_sub_crqs(adapter, 1); 6493 release_crq_queue(adapter); 6494 6495 /* cleanup worker thread after releasing CRQ so we don't get 6496 * transport events (i.e new work items for the worker thread). 6497 */ 6498 adapter->state = VNIC_REMOVING; 6499 complete(&adapter->probe_done); 6500 flush_work(&adapter->ibmvnic_reset); 6501 flush_delayed_work(&adapter->ibmvnic_delayed_reset); 6502 6503 flush_reset_queue(adapter); 6504 6505 mutex_destroy(&adapter->fw_lock); 6506 free_netdev(netdev); 6507 6508 return rc; 6509 } 6510 6511 static void ibmvnic_remove(struct vio_dev *dev) 6512 { 6513 struct net_device *netdev = dev_get_drvdata(&dev->dev); 6514 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 6515 unsigned long flags; 6516 6517 spin_lock_irqsave(&adapter->state_lock, flags); 6518 6519 /* If ibmvnic_reset() is scheduling a reset, wait for it to 6520 * finish. Then, set the state to REMOVING to prevent it from 6521 * scheduling any more work and to have reset functions ignore 6522 * any resets that have already been scheduled. Drop the lock 6523 * after setting state, so __ibmvnic_reset() which is called 6524 * from the flush_work() below, can make progress. 6525 */ 6526 spin_lock(&adapter->rwi_lock); 6527 adapter->state = VNIC_REMOVING; 6528 spin_unlock(&adapter->rwi_lock); 6529 6530 spin_unlock_irqrestore(&adapter->state_lock, flags); 6531 6532 ibmvnic_cpu_notif_remove(adapter); 6533 6534 flush_work(&adapter->ibmvnic_reset); 6535 flush_delayed_work(&adapter->ibmvnic_delayed_reset); 6536 6537 rtnl_lock(); 6538 unregister_netdevice(netdev); 6539 6540 release_resources(adapter); 6541 release_rx_pools(adapter); 6542 release_tx_pools(adapter); 6543 release_sub_crqs(adapter, 1); 6544 release_crq_queue(adapter); 6545 6546 release_stats_token(adapter); 6547 release_stats_buffers(adapter); 6548 6549 adapter->state = VNIC_REMOVED; 6550 6551 rtnl_unlock(); 6552 mutex_destroy(&adapter->fw_lock); 6553 device_remove_file(&dev->dev, &dev_attr_failover); 6554 free_netdev(netdev); 6555 dev_set_drvdata(&dev->dev, NULL); 6556 } 6557 6558 static ssize_t failover_store(struct device *dev, struct device_attribute *attr, 6559 const char *buf, size_t count) 6560 { 6561 struct net_device *netdev = dev_get_drvdata(dev); 6562 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 6563 unsigned long retbuf[PLPAR_HCALL_BUFSIZE]; 6564 __be64 session_token; 6565 long rc; 6566 6567 if (!sysfs_streq(buf, "1")) 6568 return -EINVAL; 6569 6570 rc = plpar_hcall(H_VIOCTL, retbuf, adapter->vdev->unit_address, 6571 H_GET_SESSION_TOKEN, 0, 0, 0); 6572 if (rc) { 6573 netdev_err(netdev, "Couldn't retrieve session token, rc %ld\n", 6574 rc); 6575 goto last_resort; 6576 } 6577 6578 session_token = (__be64)retbuf[0]; 6579 netdev_dbg(netdev, "Initiating client failover, session id %llx\n", 6580 be64_to_cpu(session_token)); 6581 rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address, 6582 H_SESSION_ERR_DETECTED, session_token, 0, 0); 6583 if (rc) { 6584 netdev_err(netdev, 6585 "H_VIOCTL initiated failover failed, rc %ld\n", 6586 rc); 6587 goto last_resort; 6588 } 6589 6590 return count; 6591 6592 last_resort: 6593 netdev_dbg(netdev, "Trying to send CRQ_CMD, the last resort\n"); 6594 ibmvnic_reset(adapter, VNIC_RESET_FAILOVER); 6595 6596 return count; 6597 } 6598 static DEVICE_ATTR_WO(failover); 6599 6600 static unsigned long ibmvnic_get_desired_dma(struct vio_dev *vdev) 6601 { 6602 struct net_device *netdev = dev_get_drvdata(&vdev->dev); 6603 struct ibmvnic_adapter *adapter; 6604 struct iommu_table *tbl; 6605 unsigned long ret = 0; 6606 int i; 6607 6608 tbl = get_iommu_table_base(&vdev->dev); 6609 6610 /* netdev inits at probe time along with the structures we need below*/ 6611 if (!netdev) 6612 return IOMMU_PAGE_ALIGN(IBMVNIC_IO_ENTITLEMENT_DEFAULT, tbl); 6613 6614 adapter = netdev_priv(netdev); 6615 6616 ret += PAGE_SIZE; /* the crq message queue */ 6617 ret += IOMMU_PAGE_ALIGN(sizeof(struct ibmvnic_statistics), tbl); 6618 6619 for (i = 0; i < adapter->req_tx_queues + adapter->req_rx_queues; i++) 6620 ret += 4 * PAGE_SIZE; /* the scrq message queue */ 6621 6622 for (i = 0; i < adapter->num_active_rx_pools; i++) 6623 ret += adapter->rx_pool[i].size * 6624 IOMMU_PAGE_ALIGN(adapter->rx_pool[i].buff_size, tbl); 6625 6626 return ret; 6627 } 6628 6629 static int ibmvnic_resume(struct device *dev) 6630 { 6631 struct net_device *netdev = dev_get_drvdata(dev); 6632 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 6633 6634 if (adapter->state != VNIC_OPEN) 6635 return 0; 6636 6637 tasklet_schedule(&adapter->tasklet); 6638 6639 return 0; 6640 } 6641 6642 static const struct vio_device_id ibmvnic_device_table[] = { 6643 {"network", "IBM,vnic"}, 6644 {"", "" } 6645 }; 6646 MODULE_DEVICE_TABLE(vio, ibmvnic_device_table); 6647 6648 static const struct dev_pm_ops ibmvnic_pm_ops = { 6649 .resume = ibmvnic_resume 6650 }; 6651 6652 static struct vio_driver ibmvnic_driver = { 6653 .id_table = ibmvnic_device_table, 6654 .probe = ibmvnic_probe, 6655 .remove = ibmvnic_remove, 6656 .get_desired_dma = ibmvnic_get_desired_dma, 6657 .name = ibmvnic_driver_name, 6658 .pm = &ibmvnic_pm_ops, 6659 }; 6660 6661 /* module functions */ 6662 static int __init ibmvnic_module_init(void) 6663 { 6664 int ret; 6665 6666 ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN, "net/ibmvnic:online", 6667 ibmvnic_cpu_online, 6668 ibmvnic_cpu_down_prep); 6669 if (ret < 0) 6670 goto out; 6671 ibmvnic_online = ret; 6672 ret = cpuhp_setup_state_multi(CPUHP_IBMVNIC_DEAD, "net/ibmvnic:dead", 6673 NULL, ibmvnic_cpu_dead); 6674 if (ret) 6675 goto err_dead; 6676 6677 ret = vio_register_driver(&ibmvnic_driver); 6678 if (ret) 6679 goto err_vio_register; 6680 6681 pr_info("%s: %s %s\n", ibmvnic_driver_name, ibmvnic_driver_string, 6682 IBMVNIC_DRIVER_VERSION); 6683 6684 return 0; 6685 err_vio_register: 6686 cpuhp_remove_multi_state(CPUHP_IBMVNIC_DEAD); 6687 err_dead: 6688 cpuhp_remove_multi_state(ibmvnic_online); 6689 out: 6690 return ret; 6691 } 6692 6693 static void __exit ibmvnic_module_exit(void) 6694 { 6695 vio_unregister_driver(&ibmvnic_driver); 6696 cpuhp_remove_multi_state(CPUHP_IBMVNIC_DEAD); 6697 cpuhp_remove_multi_state(ibmvnic_online); 6698 } 6699 6700 module_init(ibmvnic_module_init); 6701 module_exit(ibmvnic_module_exit); 6702