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