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 static void print_subcrq_error(struct device *dev, int rc, const char *func); 121 122 struct ibmvnic_stat { 123 char name[ETH_GSTRING_LEN]; 124 int offset; 125 }; 126 127 #define IBMVNIC_STAT_OFF(stat) (offsetof(struct ibmvnic_adapter, stats) + \ 128 offsetof(struct ibmvnic_statistics, stat)) 129 #define IBMVNIC_GET_STAT(a, off) (*((u64 *)(((unsigned long)(a)) + (off)))) 130 131 static const struct ibmvnic_stat ibmvnic_stats[] = { 132 {"rx_packets", IBMVNIC_STAT_OFF(rx_packets)}, 133 {"rx_bytes", IBMVNIC_STAT_OFF(rx_bytes)}, 134 {"tx_packets", IBMVNIC_STAT_OFF(tx_packets)}, 135 {"tx_bytes", IBMVNIC_STAT_OFF(tx_bytes)}, 136 {"ucast_tx_packets", IBMVNIC_STAT_OFF(ucast_tx_packets)}, 137 {"ucast_rx_packets", IBMVNIC_STAT_OFF(ucast_rx_packets)}, 138 {"mcast_tx_packets", IBMVNIC_STAT_OFF(mcast_tx_packets)}, 139 {"mcast_rx_packets", IBMVNIC_STAT_OFF(mcast_rx_packets)}, 140 {"bcast_tx_packets", IBMVNIC_STAT_OFF(bcast_tx_packets)}, 141 {"bcast_rx_packets", IBMVNIC_STAT_OFF(bcast_rx_packets)}, 142 {"align_errors", IBMVNIC_STAT_OFF(align_errors)}, 143 {"fcs_errors", IBMVNIC_STAT_OFF(fcs_errors)}, 144 {"single_collision_frames", IBMVNIC_STAT_OFF(single_collision_frames)}, 145 {"multi_collision_frames", IBMVNIC_STAT_OFF(multi_collision_frames)}, 146 {"sqe_test_errors", IBMVNIC_STAT_OFF(sqe_test_errors)}, 147 {"deferred_tx", IBMVNIC_STAT_OFF(deferred_tx)}, 148 {"late_collisions", IBMVNIC_STAT_OFF(late_collisions)}, 149 {"excess_collisions", IBMVNIC_STAT_OFF(excess_collisions)}, 150 {"internal_mac_tx_errors", IBMVNIC_STAT_OFF(internal_mac_tx_errors)}, 151 {"carrier_sense", IBMVNIC_STAT_OFF(carrier_sense)}, 152 {"too_long_frames", IBMVNIC_STAT_OFF(too_long_frames)}, 153 {"internal_mac_rx_errors", IBMVNIC_STAT_OFF(internal_mac_rx_errors)}, 154 }; 155 156 static int send_crq_init_complete(struct ibmvnic_adapter *adapter) 157 { 158 union ibmvnic_crq crq; 159 160 memset(&crq, 0, sizeof(crq)); 161 crq.generic.first = IBMVNIC_CRQ_INIT_CMD; 162 crq.generic.cmd = IBMVNIC_CRQ_INIT_COMPLETE; 163 164 return ibmvnic_send_crq(adapter, &crq); 165 } 166 167 static int send_version_xchg(struct ibmvnic_adapter *adapter) 168 { 169 union ibmvnic_crq crq; 170 171 memset(&crq, 0, sizeof(crq)); 172 crq.version_exchange.first = IBMVNIC_CRQ_CMD; 173 crq.version_exchange.cmd = VERSION_EXCHANGE; 174 crq.version_exchange.version = cpu_to_be16(ibmvnic_version); 175 176 return ibmvnic_send_crq(adapter, &crq); 177 } 178 179 static void ibmvnic_clean_queue_affinity(struct ibmvnic_adapter *adapter, 180 struct ibmvnic_sub_crq_queue *queue) 181 { 182 if (!(queue && queue->irq)) 183 return; 184 185 cpumask_clear(queue->affinity_mask); 186 187 if (irq_set_affinity_and_hint(queue->irq, NULL)) 188 netdev_warn(adapter->netdev, 189 "%s: Clear affinity failed, queue addr = %p, IRQ = %d\n", 190 __func__, queue, queue->irq); 191 } 192 193 static void ibmvnic_clean_affinity(struct ibmvnic_adapter *adapter) 194 { 195 struct ibmvnic_sub_crq_queue **rxqs; 196 struct ibmvnic_sub_crq_queue **txqs; 197 int num_rxqs, num_txqs; 198 int i; 199 200 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_each_online_cpu_wrap(i, *cpu) { 238 if (!stride--) { 239 /* For the next queue we start from the first 240 * unused CPU in this queue 241 */ 242 *cpu = i; 243 break; 244 } 245 cpumask_set_cpu(i, mask); 246 } 247 248 /* set queue affinity mask */ 249 cpumask_copy(queue->affinity_mask, mask); 250 rc = irq_set_affinity_and_hint(queue->irq, queue->affinity_mask); 251 free_cpumask_var(mask); 252 253 return rc; 254 } 255 256 /* assumes cpu read lock is held */ 257 static void ibmvnic_set_affinity(struct ibmvnic_adapter *adapter) 258 { 259 struct ibmvnic_sub_crq_queue **rxqs = adapter->rx_scrq; 260 struct ibmvnic_sub_crq_queue **txqs = adapter->tx_scrq; 261 struct ibmvnic_sub_crq_queue *queue; 262 int num_rxqs = adapter->num_active_rx_scrqs, i_rxqs = 0; 263 int num_txqs = adapter->num_active_tx_scrqs, i_txqs = 0; 264 int total_queues, stride, stragglers, i; 265 unsigned int num_cpu, cpu = 0; 266 bool is_rx_queue; 267 int rc = 0; 268 269 netdev_dbg(adapter->netdev, "%s: Setting irq affinity hints", __func__); 270 if (!(adapter->rx_scrq && adapter->tx_scrq)) { 271 netdev_warn(adapter->netdev, 272 "%s: Set affinity failed, queues not allocated\n", 273 __func__); 274 return; 275 } 276 277 total_queues = num_rxqs + num_txqs; 278 num_cpu = num_online_cpus(); 279 /* number of cpu's assigned per irq */ 280 stride = max_t(int, num_cpu / total_queues, 1); 281 /* number of leftover cpu's */ 282 stragglers = num_cpu >= total_queues ? num_cpu % total_queues : 0; 283 284 for (i = 0; i < total_queues; i++) { 285 is_rx_queue = false; 286 /* balance core load by alternating rx and tx assignments 287 * ex: TX0 -> RX0 -> TX1 -> RX1 etc. 288 */ 289 if ((i % 2 == 1 && i_rxqs < num_rxqs) || i_txqs == num_txqs) { 290 queue = rxqs[i_rxqs++]; 291 is_rx_queue = true; 292 } else { 293 queue = txqs[i_txqs++]; 294 } 295 296 rc = ibmvnic_set_queue_affinity(queue, &cpu, &stragglers, 297 stride); 298 if (rc) 299 goto out; 300 301 if (!queue || is_rx_queue) 302 continue; 303 304 rc = __netif_set_xps_queue(adapter->netdev, 305 cpumask_bits(queue->affinity_mask), 306 i_txqs - 1, XPS_CPUS); 307 if (rc) 308 netdev_warn(adapter->netdev, "%s: Set XPS on queue %d failed, rc = %d.\n", 309 __func__, i_txqs - 1, rc); 310 } 311 312 out: 313 if (rc) { 314 netdev_warn(adapter->netdev, 315 "%s: Set affinity failed, queue addr = %p, IRQ = %d, rc = %d.\n", 316 __func__, queue, queue->irq, rc); 317 ibmvnic_clean_affinity(adapter); 318 } 319 } 320 321 static int ibmvnic_cpu_online(unsigned int cpu, struct hlist_node *node) 322 { 323 struct ibmvnic_adapter *adapter; 324 325 adapter = hlist_entry_safe(node, struct ibmvnic_adapter, node); 326 ibmvnic_set_affinity(adapter); 327 return 0; 328 } 329 330 static int ibmvnic_cpu_dead(unsigned int cpu, struct hlist_node *node) 331 { 332 struct ibmvnic_adapter *adapter; 333 334 adapter = hlist_entry_safe(node, struct ibmvnic_adapter, node_dead); 335 ibmvnic_set_affinity(adapter); 336 return 0; 337 } 338 339 static int ibmvnic_cpu_down_prep(unsigned int cpu, struct hlist_node *node) 340 { 341 struct ibmvnic_adapter *adapter; 342 343 adapter = hlist_entry_safe(node, struct ibmvnic_adapter, node); 344 ibmvnic_clean_affinity(adapter); 345 return 0; 346 } 347 348 static enum cpuhp_state ibmvnic_online; 349 350 static int ibmvnic_cpu_notif_add(struct ibmvnic_adapter *adapter) 351 { 352 int ret; 353 354 ret = cpuhp_state_add_instance_nocalls(ibmvnic_online, &adapter->node); 355 if (ret) 356 return ret; 357 ret = cpuhp_state_add_instance_nocalls(CPUHP_IBMVNIC_DEAD, 358 &adapter->node_dead); 359 if (!ret) 360 return ret; 361 cpuhp_state_remove_instance_nocalls(ibmvnic_online, &adapter->node); 362 return ret; 363 } 364 365 static void ibmvnic_cpu_notif_remove(struct ibmvnic_adapter *adapter) 366 { 367 cpuhp_state_remove_instance_nocalls(ibmvnic_online, &adapter->node); 368 cpuhp_state_remove_instance_nocalls(CPUHP_IBMVNIC_DEAD, 369 &adapter->node_dead); 370 } 371 372 static long h_reg_sub_crq(unsigned long unit_address, unsigned long token, 373 unsigned long length, unsigned long *number, 374 unsigned long *irq) 375 { 376 unsigned long retbuf[PLPAR_HCALL_BUFSIZE]; 377 long rc; 378 379 rc = plpar_hcall(H_REG_SUB_CRQ, retbuf, unit_address, token, length); 380 *number = retbuf[0]; 381 *irq = retbuf[1]; 382 383 return rc; 384 } 385 386 /** 387 * ibmvnic_wait_for_completion - Check device state and wait for completion 388 * @adapter: private device data 389 * @comp_done: completion structure to wait for 390 * @timeout: time to wait in milliseconds 391 * 392 * Wait for a completion signal or until the timeout limit is reached 393 * while checking that the device is still active. 394 */ 395 static int ibmvnic_wait_for_completion(struct ibmvnic_adapter *adapter, 396 struct completion *comp_done, 397 unsigned long timeout) 398 { 399 struct net_device *netdev; 400 unsigned long div_timeout; 401 u8 retry; 402 403 netdev = adapter->netdev; 404 retry = 5; 405 div_timeout = msecs_to_jiffies(timeout / retry); 406 while (true) { 407 if (!adapter->crq.active) { 408 netdev_err(netdev, "Device down!\n"); 409 return -ENODEV; 410 } 411 if (!retry--) 412 break; 413 if (wait_for_completion_timeout(comp_done, div_timeout)) 414 return 0; 415 } 416 netdev_err(netdev, "Operation timed out.\n"); 417 return -ETIMEDOUT; 418 } 419 420 /** 421 * reuse_ltb() - Check if a long term buffer can be reused 422 * @ltb: The long term buffer to be checked 423 * @size: The size of the long term buffer. 424 * 425 * An LTB can be reused unless its size has changed. 426 * 427 * Return: Return true if the LTB can be reused, false otherwise. 428 */ 429 static bool reuse_ltb(struct ibmvnic_long_term_buff *ltb, int size) 430 { 431 return (ltb->buff && ltb->size == size); 432 } 433 434 /** 435 * alloc_long_term_buff() - Allocate a long term buffer (LTB) 436 * 437 * @adapter: ibmvnic adapter associated to the LTB 438 * @ltb: container object for the LTB 439 * @size: size of the LTB 440 * 441 * Allocate an LTB of the specified size and notify VIOS. 442 * 443 * If the given @ltb already has the correct size, reuse it. Otherwise if 444 * its non-NULL, free it. Then allocate a new one of the correct size. 445 * Notify the VIOS either way since we may now be working with a new VIOS. 446 * 447 * Allocating larger chunks of memory during resets, specially LPM or under 448 * low memory situations can cause resets to fail/timeout and for LPAR to 449 * lose connectivity. So hold onto the LTB even if we fail to communicate 450 * with the VIOS and reuse it on next open. Free LTB when adapter is closed. 451 * 452 * Return: 0 if we were able to allocate the LTB and notify the VIOS and 453 * a negative value otherwise. 454 */ 455 static int alloc_long_term_buff(struct ibmvnic_adapter *adapter, 456 struct ibmvnic_long_term_buff *ltb, int size) 457 { 458 struct device *dev = &adapter->vdev->dev; 459 u64 prev = 0; 460 int rc; 461 462 if (!reuse_ltb(ltb, size)) { 463 dev_dbg(dev, 464 "LTB size changed from 0x%llx to 0x%x, reallocating\n", 465 ltb->size, size); 466 prev = ltb->size; 467 free_long_term_buff(adapter, ltb); 468 } 469 470 if (ltb->buff) { 471 dev_dbg(dev, "Reusing LTB [map %d, size 0x%llx]\n", 472 ltb->map_id, ltb->size); 473 } else { 474 ltb->buff = dma_alloc_coherent(dev, size, <b->addr, 475 GFP_KERNEL); 476 if (!ltb->buff) { 477 dev_err(dev, "Couldn't alloc long term buffer\n"); 478 return -ENOMEM; 479 } 480 ltb->size = size; 481 482 ltb->map_id = find_first_zero_bit(adapter->map_ids, 483 MAX_MAP_ID); 484 bitmap_set(adapter->map_ids, ltb->map_id, 1); 485 486 dev_dbg(dev, 487 "Allocated new LTB [map %d, size 0x%llx was 0x%llx]\n", 488 ltb->map_id, ltb->size, prev); 489 } 490 491 /* Ensure ltb is zeroed - specially when reusing it. */ 492 memset(ltb->buff, 0, ltb->size); 493 494 mutex_lock(&adapter->fw_lock); 495 adapter->fw_done_rc = 0; 496 reinit_completion(&adapter->fw_done); 497 498 rc = send_request_map(adapter, ltb->addr, ltb->size, ltb->map_id); 499 if (rc) { 500 dev_err(dev, "send_request_map failed, rc = %d\n", rc); 501 goto out; 502 } 503 504 rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000); 505 if (rc) { 506 dev_err(dev, "LTB map request aborted or timed out, rc = %d\n", 507 rc); 508 goto out; 509 } 510 511 if (adapter->fw_done_rc) { 512 dev_err(dev, "Couldn't map LTB, rc = %d\n", 513 adapter->fw_done_rc); 514 rc = -EIO; 515 goto out; 516 } 517 rc = 0; 518 out: 519 /* don't free LTB on communication error - see function header */ 520 mutex_unlock(&adapter->fw_lock); 521 return rc; 522 } 523 524 static void free_long_term_buff(struct ibmvnic_adapter *adapter, 525 struct ibmvnic_long_term_buff *ltb) 526 { 527 struct device *dev = &adapter->vdev->dev; 528 529 if (!ltb->buff) 530 return; 531 532 /* VIOS automatically unmaps the long term buffer at remote 533 * end for the following resets: 534 * FAILOVER, MOBILITY, TIMEOUT. 535 */ 536 if (adapter->reset_reason != VNIC_RESET_FAILOVER && 537 adapter->reset_reason != VNIC_RESET_MOBILITY && 538 adapter->reset_reason != VNIC_RESET_TIMEOUT) 539 send_request_unmap(adapter, ltb->map_id); 540 541 dma_free_coherent(dev, ltb->size, ltb->buff, ltb->addr); 542 543 ltb->buff = NULL; 544 /* mark this map_id free */ 545 bitmap_clear(adapter->map_ids, ltb->map_id, 1); 546 ltb->map_id = 0; 547 } 548 549 /** 550 * free_ltb_set - free the given set of long term buffers (LTBS) 551 * @adapter: The ibmvnic adapter containing this ltb set 552 * @ltb_set: The ltb_set to be freed 553 * 554 * Free the set of LTBs in the given set. 555 */ 556 557 static void free_ltb_set(struct ibmvnic_adapter *adapter, 558 struct ibmvnic_ltb_set *ltb_set) 559 { 560 int i; 561 562 for (i = 0; i < ltb_set->num_ltbs; i++) 563 free_long_term_buff(adapter, <b_set->ltbs[i]); 564 565 kfree(ltb_set->ltbs); 566 ltb_set->ltbs = NULL; 567 ltb_set->num_ltbs = 0; 568 } 569 570 /** 571 * alloc_ltb_set() - Allocate a set of long term buffers (LTBs) 572 * 573 * @adapter: ibmvnic adapter associated to the LTB 574 * @ltb_set: container object for the set of LTBs 575 * @num_buffs: Number of buffers in the LTB 576 * @buff_size: Size of each buffer in the LTB 577 * 578 * Allocate a set of LTBs to accommodate @num_buffs buffers of @buff_size 579 * each. We currently cap size each LTB to IBMVNIC_ONE_LTB_SIZE. If the 580 * new set of LTBs have fewer LTBs than the old set, free the excess LTBs. 581 * If new set needs more than in old set, allocate the remaining ones. 582 * Try and reuse as many LTBs as possible and avoid reallocation. 583 * 584 * Any changes to this allocation strategy must be reflected in 585 * map_rxpool_buff_to_ltb() and map_txpool_buff_to_ltb(). 586 */ 587 static int alloc_ltb_set(struct ibmvnic_adapter *adapter, 588 struct ibmvnic_ltb_set *ltb_set, int num_buffs, 589 int buff_size) 590 { 591 struct device *dev = &adapter->vdev->dev; 592 struct ibmvnic_ltb_set old_set; 593 struct ibmvnic_ltb_set new_set; 594 int rem_size; 595 int tot_size; /* size of all ltbs */ 596 int ltb_size; /* size of one ltb */ 597 int nltbs; 598 int rc; 599 int n; 600 int i; 601 602 dev_dbg(dev, "%s() num_buffs %d, buff_size %d\n", __func__, num_buffs, 603 buff_size); 604 605 ltb_size = rounddown(IBMVNIC_ONE_LTB_SIZE, buff_size); 606 tot_size = num_buffs * buff_size; 607 608 if (ltb_size > tot_size) 609 ltb_size = tot_size; 610 611 nltbs = tot_size / ltb_size; 612 if (tot_size % ltb_size) 613 nltbs++; 614 615 old_set = *ltb_set; 616 617 if (old_set.num_ltbs == nltbs) { 618 new_set = old_set; 619 } else { 620 int tmp = nltbs * sizeof(struct ibmvnic_long_term_buff); 621 622 new_set.ltbs = kzalloc(tmp, GFP_KERNEL); 623 if (!new_set.ltbs) 624 return -ENOMEM; 625 626 new_set.num_ltbs = nltbs; 627 628 /* Free any excess ltbs in old set */ 629 for (i = new_set.num_ltbs; i < old_set.num_ltbs; i++) 630 free_long_term_buff(adapter, &old_set.ltbs[i]); 631 632 /* Copy remaining ltbs to new set. All LTBs except the 633 * last one are of the same size. alloc_long_term_buff() 634 * will realloc if the size changes. 635 */ 636 n = min(old_set.num_ltbs, new_set.num_ltbs); 637 for (i = 0; i < n; i++) 638 new_set.ltbs[i] = old_set.ltbs[i]; 639 640 /* Any additional ltbs in new set will have NULL ltbs for 641 * now and will be allocated in alloc_long_term_buff(). 642 */ 643 644 /* We no longer need the old_set so free it. Note that we 645 * may have reused some ltbs from old set and freed excess 646 * ltbs above. So we only need to free the container now 647 * not the LTBs themselves. (i.e. dont free_ltb_set()!) 648 */ 649 kfree(old_set.ltbs); 650 old_set.ltbs = NULL; 651 old_set.num_ltbs = 0; 652 653 /* Install the new set. If allocations fail below, we will 654 * retry later and know what size LTBs we need. 655 */ 656 *ltb_set = new_set; 657 } 658 659 i = 0; 660 rem_size = tot_size; 661 while (rem_size) { 662 if (ltb_size > rem_size) 663 ltb_size = rem_size; 664 665 rem_size -= ltb_size; 666 667 rc = alloc_long_term_buff(adapter, &new_set.ltbs[i], ltb_size); 668 if (rc) 669 goto out; 670 i++; 671 } 672 673 WARN_ON(i != new_set.num_ltbs); 674 675 return 0; 676 out: 677 /* We may have allocated one/more LTBs before failing and we 678 * want to try and reuse on next reset. So don't free ltb set. 679 */ 680 return rc; 681 } 682 683 /** 684 * map_rxpool_buf_to_ltb - Map given rxpool buffer to offset in an LTB. 685 * @rxpool: The receive buffer pool containing buffer 686 * @bufidx: Index of buffer in rxpool 687 * @ltbp: (Output) pointer to the long term buffer containing the buffer 688 * @offset: (Output) offset of buffer in the LTB from @ltbp 689 * 690 * Map the given buffer identified by [rxpool, bufidx] to an LTB in the 691 * pool and its corresponding offset. Assume for now that each LTB is of 692 * different size but could possibly be optimized based on the allocation 693 * strategy in alloc_ltb_set(). 694 */ 695 static void map_rxpool_buf_to_ltb(struct ibmvnic_rx_pool *rxpool, 696 unsigned int bufidx, 697 struct ibmvnic_long_term_buff **ltbp, 698 unsigned int *offset) 699 { 700 struct ibmvnic_long_term_buff *ltb; 701 int nbufs; /* # of buffers in one ltb */ 702 int i; 703 704 WARN_ON(bufidx >= rxpool->size); 705 706 for (i = 0; i < rxpool->ltb_set.num_ltbs; i++) { 707 ltb = &rxpool->ltb_set.ltbs[i]; 708 nbufs = ltb->size / rxpool->buff_size; 709 if (bufidx < nbufs) 710 break; 711 bufidx -= nbufs; 712 } 713 714 *ltbp = ltb; 715 *offset = bufidx * rxpool->buff_size; 716 } 717 718 /** 719 * map_txpool_buf_to_ltb - Map given txpool buffer to offset in an LTB. 720 * @txpool: The transmit buffer pool containing buffer 721 * @bufidx: Index of buffer in txpool 722 * @ltbp: (Output) pointer to the long term buffer (LTB) containing the buffer 723 * @offset: (Output) offset of buffer in the LTB from @ltbp 724 * 725 * Map the given buffer identified by [txpool, bufidx] to an LTB in the 726 * pool and its corresponding offset. 727 */ 728 static void map_txpool_buf_to_ltb(struct ibmvnic_tx_pool *txpool, 729 unsigned int bufidx, 730 struct ibmvnic_long_term_buff **ltbp, 731 unsigned int *offset) 732 { 733 struct ibmvnic_long_term_buff *ltb; 734 int nbufs; /* # of buffers in one ltb */ 735 int i; 736 737 WARN_ON_ONCE(bufidx >= txpool->num_buffers); 738 739 for (i = 0; i < txpool->ltb_set.num_ltbs; i++) { 740 ltb = &txpool->ltb_set.ltbs[i]; 741 nbufs = ltb->size / txpool->buf_size; 742 if (bufidx < nbufs) 743 break; 744 bufidx -= nbufs; 745 } 746 747 *ltbp = ltb; 748 *offset = bufidx * txpool->buf_size; 749 } 750 751 static void deactivate_rx_pools(struct ibmvnic_adapter *adapter) 752 { 753 int i; 754 755 for (i = 0; i < adapter->num_active_rx_pools; i++) 756 adapter->rx_pool[i].active = 0; 757 } 758 759 static void ibmvnic_set_safe_max_ind_descs(struct ibmvnic_adapter *adapter) 760 { 761 if (adapter->cur_max_ind_descs > IBMVNIC_SAFE_IND_DESC) { 762 netdev_info(adapter->netdev, 763 "set max ind descs from %u to safe limit %u\n", 764 adapter->cur_max_ind_descs, 765 IBMVNIC_SAFE_IND_DESC); 766 adapter->cur_max_ind_descs = IBMVNIC_SAFE_IND_DESC; 767 } 768 } 769 770 static void replenish_rx_pool(struct ibmvnic_adapter *adapter, 771 struct ibmvnic_rx_pool *pool) 772 { 773 int count = pool->size - atomic_read(&pool->available); 774 u64 handle = adapter->rx_scrq[pool->index]->handle; 775 struct device *dev = &adapter->vdev->dev; 776 struct ibmvnic_ind_xmit_queue *ind_bufp; 777 struct ibmvnic_sub_crq_queue *rx_scrq; 778 struct ibmvnic_long_term_buff *ltb; 779 union sub_crq *sub_crq; 780 int buffers_added = 0; 781 unsigned long lpar_rc; 782 struct sk_buff *skb; 783 unsigned int offset; 784 dma_addr_t dma_addr; 785 unsigned char *dst; 786 int shift = 0; 787 int bufidx; 788 int i; 789 790 if (!pool->active) 791 return; 792 793 rx_scrq = adapter->rx_scrq[pool->index]; 794 ind_bufp = &rx_scrq->ind_buf; 795 796 /* netdev_skb_alloc() could have failed after we saved a few skbs 797 * in the indir_buf and we would not have sent them to VIOS yet. 798 * To account for them, start the loop at ind_bufp->index rather 799 * than 0. If we pushed all the skbs to VIOS, ind_bufp->index will 800 * be 0. 801 */ 802 for (i = ind_bufp->index; i < count; ++i) { 803 bufidx = pool->free_map[pool->next_free]; 804 805 /* We maybe reusing the skb from earlier resets. Allocate 806 * only if necessary. But since the LTB may have changed 807 * during reset (see init_rx_pools()), update LTB below 808 * even if reusing skb. 809 */ 810 skb = pool->rx_buff[bufidx].skb; 811 if (!skb) { 812 skb = netdev_alloc_skb(adapter->netdev, 813 pool->buff_size); 814 if (!skb) { 815 dev_err(dev, "Couldn't replenish rx buff\n"); 816 adapter->replenish_no_mem++; 817 break; 818 } 819 } 820 821 pool->free_map[pool->next_free] = IBMVNIC_INVALID_MAP; 822 pool->next_free = (pool->next_free + 1) % pool->size; 823 824 /* Copy the skb to the long term mapped DMA buffer */ 825 map_rxpool_buf_to_ltb(pool, bufidx, <b, &offset); 826 dst = ltb->buff + offset; 827 memset(dst, 0, pool->buff_size); 828 dma_addr = ltb->addr + offset; 829 830 /* add the skb to an rx_buff in the pool */ 831 pool->rx_buff[bufidx].data = dst; 832 pool->rx_buff[bufidx].dma = dma_addr; 833 pool->rx_buff[bufidx].skb = skb; 834 pool->rx_buff[bufidx].pool_index = pool->index; 835 pool->rx_buff[bufidx].size = pool->buff_size; 836 837 /* queue the rx_buff for the next send_subcrq_indirect */ 838 sub_crq = &ind_bufp->indir_arr[ind_bufp->index++]; 839 memset(sub_crq, 0, sizeof(*sub_crq)); 840 sub_crq->rx_add.first = IBMVNIC_CRQ_CMD; 841 sub_crq->rx_add.correlator = 842 cpu_to_be64((u64)&pool->rx_buff[bufidx]); 843 sub_crq->rx_add.ioba = cpu_to_be32(dma_addr); 844 sub_crq->rx_add.map_id = ltb->map_id; 845 846 /* The length field of the sCRQ is defined to be 24 bits so the 847 * buffer size needs to be left shifted by a byte before it is 848 * converted to big endian to prevent the last byte from being 849 * truncated. 850 */ 851 #ifdef __LITTLE_ENDIAN__ 852 shift = 8; 853 #endif 854 sub_crq->rx_add.len = cpu_to_be32(pool->buff_size << shift); 855 856 /* if send_subcrq_indirect queue is full, flush to VIOS */ 857 if (ind_bufp->index == adapter->cur_max_ind_descs || 858 i == count - 1) { 859 lpar_rc = 860 send_subcrq_indirect(adapter, handle, 861 (u64)ind_bufp->indir_dma, 862 (u64)ind_bufp->index); 863 if (lpar_rc != H_SUCCESS) 864 goto failure; 865 buffers_added += ind_bufp->index; 866 adapter->replenish_add_buff_success += ind_bufp->index; 867 ind_bufp->index = 0; 868 } 869 } 870 atomic_add(buffers_added, &pool->available); 871 return; 872 873 failure: 874 if (lpar_rc != H_PARAMETER && lpar_rc != H_CLOSED) 875 dev_err_ratelimited(dev, "rx: replenish packet buffer failed\n"); 876 877 /* Detect platform limit H_PARAMETER */ 878 if (lpar_rc == H_PARAMETER) 879 ibmvnic_set_safe_max_ind_descs(adapter); 880 881 /* For all error case, temporarily drop only this batch 882 * Rely on TCP/IP retransmissions to retry and recover 883 */ 884 for (i = ind_bufp->index - 1; i >= 0; --i) { 885 struct ibmvnic_rx_buff *rx_buff; 886 887 pool->next_free = pool->next_free == 0 ? 888 pool->size - 1 : pool->next_free - 1; 889 sub_crq = &ind_bufp->indir_arr[i]; 890 rx_buff = (struct ibmvnic_rx_buff *) 891 be64_to_cpu(sub_crq->rx_add.correlator); 892 bufidx = (int)(rx_buff - pool->rx_buff); 893 pool->free_map[pool->next_free] = bufidx; 894 dev_kfree_skb_any(pool->rx_buff[bufidx].skb); 895 pool->rx_buff[bufidx].skb = NULL; 896 } 897 adapter->replenish_add_buff_failure += ind_bufp->index; 898 atomic_add(buffers_added, &pool->available); 899 ind_bufp->index = 0; 900 if (lpar_rc == H_CLOSED || adapter->failover_pending) { 901 /* Disable buffer pool replenishment and report carrier off if 902 * queue is closed or pending failover. 903 * Firmware guarantees that a signal will be sent to the 904 * driver, triggering a reset. 905 */ 906 deactivate_rx_pools(adapter); 907 netif_carrier_off(adapter->netdev); 908 } 909 } 910 911 static void replenish_pools(struct ibmvnic_adapter *adapter) 912 { 913 int i; 914 915 adapter->replenish_task_cycles++; 916 for (i = 0; i < adapter->num_active_rx_pools; i++) { 917 if (adapter->rx_pool[i].active) 918 replenish_rx_pool(adapter, &adapter->rx_pool[i]); 919 } 920 921 netdev_dbg(adapter->netdev, "Replenished %d pools\n", i); 922 } 923 924 static void release_stats_buffers(struct ibmvnic_adapter *adapter) 925 { 926 kfree(adapter->tx_stats_buffers); 927 kfree(adapter->rx_stats_buffers); 928 adapter->tx_stats_buffers = NULL; 929 adapter->rx_stats_buffers = NULL; 930 } 931 932 static int init_stats_buffers(struct ibmvnic_adapter *adapter) 933 { 934 adapter->tx_stats_buffers = 935 kcalloc(IBMVNIC_MAX_QUEUES, 936 sizeof(struct ibmvnic_tx_queue_stats), 937 GFP_KERNEL); 938 if (!adapter->tx_stats_buffers) 939 return -ENOMEM; 940 941 adapter->rx_stats_buffers = 942 kcalloc(IBMVNIC_MAX_QUEUES, 943 sizeof(struct ibmvnic_rx_queue_stats), 944 GFP_KERNEL); 945 if (!adapter->rx_stats_buffers) 946 return -ENOMEM; 947 948 return 0; 949 } 950 951 static void release_stats_token(struct ibmvnic_adapter *adapter) 952 { 953 struct device *dev = &adapter->vdev->dev; 954 955 if (!adapter->stats_token) 956 return; 957 958 dma_unmap_single(dev, adapter->stats_token, 959 sizeof(struct ibmvnic_statistics), 960 DMA_FROM_DEVICE); 961 adapter->stats_token = 0; 962 } 963 964 static int init_stats_token(struct ibmvnic_adapter *adapter) 965 { 966 struct device *dev = &adapter->vdev->dev; 967 dma_addr_t stok; 968 int rc; 969 970 stok = dma_map_single(dev, &adapter->stats, 971 sizeof(struct ibmvnic_statistics), 972 DMA_FROM_DEVICE); 973 rc = dma_mapping_error(dev, stok); 974 if (rc) { 975 dev_err(dev, "Couldn't map stats buffer, rc = %d\n", rc); 976 return rc; 977 } 978 979 adapter->stats_token = stok; 980 netdev_dbg(adapter->netdev, "Stats token initialized (%llx)\n", stok); 981 return 0; 982 } 983 984 /** 985 * release_rx_pools() - Release any rx pools attached to @adapter. 986 * @adapter: ibmvnic adapter 987 * 988 * Safe to call this multiple times - even if no pools are attached. 989 */ 990 static void release_rx_pools(struct ibmvnic_adapter *adapter) 991 { 992 struct ibmvnic_rx_pool *rx_pool; 993 int i, j; 994 995 if (!adapter->rx_pool) 996 return; 997 998 for (i = 0; i < adapter->num_active_rx_pools; i++) { 999 rx_pool = &adapter->rx_pool[i]; 1000 1001 netdev_dbg(adapter->netdev, "Releasing rx_pool[%d]\n", i); 1002 1003 kfree(rx_pool->free_map); 1004 1005 free_ltb_set(adapter, &rx_pool->ltb_set); 1006 1007 if (!rx_pool->rx_buff) 1008 continue; 1009 1010 for (j = 0; j < rx_pool->size; j++) { 1011 if (rx_pool->rx_buff[j].skb) { 1012 dev_kfree_skb_any(rx_pool->rx_buff[j].skb); 1013 rx_pool->rx_buff[j].skb = NULL; 1014 } 1015 } 1016 1017 kfree(rx_pool->rx_buff); 1018 } 1019 1020 kfree(adapter->rx_pool); 1021 adapter->rx_pool = NULL; 1022 adapter->num_active_rx_pools = 0; 1023 adapter->prev_rx_pool_size = 0; 1024 } 1025 1026 /** 1027 * reuse_rx_pools() - Check if the existing rx pools can be reused. 1028 * @adapter: ibmvnic adapter 1029 * 1030 * Check if the existing rx pools in the adapter can be reused. The 1031 * pools can be reused if the pool parameters (number of pools, 1032 * number of buffers in the pool and size of each buffer) have not 1033 * changed. 1034 * 1035 * NOTE: This assumes that all pools have the same number of buffers 1036 * which is the case currently. If that changes, we must fix this. 1037 * 1038 * Return: true if the rx pools can be reused, false otherwise. 1039 */ 1040 static bool reuse_rx_pools(struct ibmvnic_adapter *adapter) 1041 { 1042 u64 old_num_pools, new_num_pools; 1043 u64 old_pool_size, new_pool_size; 1044 u64 old_buff_size, new_buff_size; 1045 1046 if (!adapter->rx_pool) 1047 return false; 1048 1049 old_num_pools = adapter->num_active_rx_pools; 1050 new_num_pools = adapter->req_rx_queues; 1051 1052 old_pool_size = adapter->prev_rx_pool_size; 1053 new_pool_size = adapter->req_rx_add_entries_per_subcrq; 1054 1055 old_buff_size = adapter->prev_rx_buf_sz; 1056 new_buff_size = adapter->cur_rx_buf_sz; 1057 1058 if (old_buff_size != new_buff_size || 1059 old_num_pools != new_num_pools || 1060 old_pool_size != new_pool_size) 1061 return false; 1062 1063 return true; 1064 } 1065 1066 /** 1067 * init_rx_pools(): Initialize the set of receiver pools in the adapter. 1068 * @netdev: net device associated with the vnic interface 1069 * 1070 * Initialize the set of receiver pools in the ibmvnic adapter associated 1071 * with the net_device @netdev. If possible, reuse the existing rx pools. 1072 * Otherwise free any existing pools and allocate a new set of pools 1073 * before initializing them. 1074 * 1075 * Return: 0 on success and negative value on error. 1076 */ 1077 static int init_rx_pools(struct net_device *netdev) 1078 { 1079 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 1080 struct device *dev = &adapter->vdev->dev; 1081 struct ibmvnic_rx_pool *rx_pool; 1082 u64 num_pools; 1083 u64 pool_size; /* # of buffers in one pool */ 1084 u64 buff_size; 1085 int i, j, rc; 1086 1087 pool_size = adapter->req_rx_add_entries_per_subcrq; 1088 num_pools = adapter->req_rx_queues; 1089 buff_size = adapter->cur_rx_buf_sz; 1090 1091 if (reuse_rx_pools(adapter)) { 1092 dev_dbg(dev, "Reusing rx pools\n"); 1093 goto update_ltb; 1094 } 1095 1096 /* Allocate/populate the pools. */ 1097 release_rx_pools(adapter); 1098 1099 adapter->rx_pool = kcalloc(num_pools, 1100 sizeof(struct ibmvnic_rx_pool), 1101 GFP_KERNEL); 1102 if (!adapter->rx_pool) { 1103 dev_err(dev, "Failed to allocate rx pools\n"); 1104 return -ENOMEM; 1105 } 1106 1107 /* Set num_active_rx_pools early. If we fail below after partial 1108 * allocation, release_rx_pools() will know how many to look for. 1109 */ 1110 adapter->num_active_rx_pools = num_pools; 1111 1112 for (i = 0; i < num_pools; i++) { 1113 rx_pool = &adapter->rx_pool[i]; 1114 1115 netdev_dbg(adapter->netdev, 1116 "Initializing rx_pool[%d], %lld buffs, %lld bytes each\n", 1117 i, pool_size, buff_size); 1118 1119 rx_pool->size = pool_size; 1120 rx_pool->index = i; 1121 rx_pool->buff_size = ALIGN(buff_size, L1_CACHE_BYTES); 1122 1123 rx_pool->free_map = kcalloc(rx_pool->size, sizeof(int), 1124 GFP_KERNEL); 1125 if (!rx_pool->free_map) { 1126 dev_err(dev, "Couldn't alloc free_map %d\n", i); 1127 rc = -ENOMEM; 1128 goto out_release; 1129 } 1130 1131 rx_pool->rx_buff = kcalloc(rx_pool->size, 1132 sizeof(struct ibmvnic_rx_buff), 1133 GFP_KERNEL); 1134 if (!rx_pool->rx_buff) { 1135 dev_err(dev, "Couldn't alloc rx buffers\n"); 1136 rc = -ENOMEM; 1137 goto out_release; 1138 } 1139 } 1140 1141 adapter->prev_rx_pool_size = pool_size; 1142 adapter->prev_rx_buf_sz = adapter->cur_rx_buf_sz; 1143 1144 update_ltb: 1145 for (i = 0; i < num_pools; i++) { 1146 rx_pool = &adapter->rx_pool[i]; 1147 dev_dbg(dev, "Updating LTB for rx pool %d [%d, %d]\n", 1148 i, rx_pool->size, rx_pool->buff_size); 1149 1150 rc = alloc_ltb_set(adapter, &rx_pool->ltb_set, 1151 rx_pool->size, rx_pool->buff_size); 1152 if (rc) 1153 goto out; 1154 1155 for (j = 0; j < rx_pool->size; ++j) { 1156 struct ibmvnic_rx_buff *rx_buff; 1157 1158 rx_pool->free_map[j] = j; 1159 1160 /* NOTE: Don't clear rx_buff->skb here - will leak 1161 * memory! replenish_rx_pool() will reuse skbs or 1162 * allocate as necessary. 1163 */ 1164 rx_buff = &rx_pool->rx_buff[j]; 1165 rx_buff->dma = 0; 1166 rx_buff->data = 0; 1167 rx_buff->size = 0; 1168 rx_buff->pool_index = 0; 1169 } 1170 1171 /* Mark pool "empty" so replenish_rx_pools() will 1172 * update the LTB info for each buffer 1173 */ 1174 atomic_set(&rx_pool->available, 0); 1175 rx_pool->next_alloc = 0; 1176 rx_pool->next_free = 0; 1177 /* replenish_rx_pool() may have called deactivate_rx_pools() 1178 * on failover. Ensure pool is active now. 1179 */ 1180 rx_pool->active = 1; 1181 } 1182 return 0; 1183 out_release: 1184 release_rx_pools(adapter); 1185 out: 1186 /* We failed to allocate one or more LTBs or map them on the VIOS. 1187 * Hold onto the pools and any LTBs that we did allocate/map. 1188 */ 1189 return rc; 1190 } 1191 1192 static void release_vpd_data(struct ibmvnic_adapter *adapter) 1193 { 1194 if (!adapter->vpd) 1195 return; 1196 1197 kfree(adapter->vpd->buff); 1198 kfree(adapter->vpd); 1199 1200 adapter->vpd = NULL; 1201 } 1202 1203 static void release_one_tx_pool(struct ibmvnic_adapter *adapter, 1204 struct ibmvnic_tx_pool *tx_pool) 1205 { 1206 kfree(tx_pool->tx_buff); 1207 kfree(tx_pool->free_map); 1208 free_ltb_set(adapter, &tx_pool->ltb_set); 1209 } 1210 1211 /** 1212 * release_tx_pools() - Release any tx pools attached to @adapter. 1213 * @adapter: ibmvnic adapter 1214 * 1215 * Safe to call this multiple times - even if no pools are attached. 1216 */ 1217 static void release_tx_pools(struct ibmvnic_adapter *adapter) 1218 { 1219 int i; 1220 1221 /* init_tx_pools() ensures that ->tx_pool and ->tso_pool are 1222 * both NULL or both non-NULL. So we only need to check one. 1223 */ 1224 if (!adapter->tx_pool) 1225 return; 1226 1227 for (i = 0; i < adapter->num_active_tx_pools; i++) { 1228 release_one_tx_pool(adapter, &adapter->tx_pool[i]); 1229 release_one_tx_pool(adapter, &adapter->tso_pool[i]); 1230 } 1231 1232 kfree(adapter->tx_pool); 1233 adapter->tx_pool = NULL; 1234 kfree(adapter->tso_pool); 1235 adapter->tso_pool = NULL; 1236 adapter->num_active_tx_pools = 0; 1237 adapter->prev_tx_pool_size = 0; 1238 } 1239 1240 static int init_one_tx_pool(struct net_device *netdev, 1241 struct ibmvnic_tx_pool *tx_pool, 1242 int pool_size, int buf_size) 1243 { 1244 int i; 1245 1246 tx_pool->tx_buff = kcalloc(pool_size, 1247 sizeof(struct ibmvnic_tx_buff), 1248 GFP_KERNEL); 1249 if (!tx_pool->tx_buff) 1250 return -ENOMEM; 1251 1252 tx_pool->free_map = kcalloc(pool_size, sizeof(int), GFP_KERNEL); 1253 if (!tx_pool->free_map) { 1254 kfree(tx_pool->tx_buff); 1255 tx_pool->tx_buff = NULL; 1256 return -ENOMEM; 1257 } 1258 1259 for (i = 0; i < pool_size; i++) 1260 tx_pool->free_map[i] = i; 1261 1262 tx_pool->consumer_index = 0; 1263 tx_pool->producer_index = 0; 1264 tx_pool->num_buffers = pool_size; 1265 tx_pool->buf_size = buf_size; 1266 1267 return 0; 1268 } 1269 1270 /** 1271 * reuse_tx_pools() - Check if the existing tx pools can be reused. 1272 * @adapter: ibmvnic adapter 1273 * 1274 * Check if the existing tx pools in the adapter can be reused. The 1275 * pools can be reused if the pool parameters (number of pools, 1276 * number of buffers in the pool and mtu) have not changed. 1277 * 1278 * NOTE: This assumes that all pools have the same number of buffers 1279 * which is the case currently. If that changes, we must fix this. 1280 * 1281 * Return: true if the tx pools can be reused, false otherwise. 1282 */ 1283 static bool reuse_tx_pools(struct ibmvnic_adapter *adapter) 1284 { 1285 u64 old_num_pools, new_num_pools; 1286 u64 old_pool_size, new_pool_size; 1287 u64 old_mtu, new_mtu; 1288 1289 if (!adapter->tx_pool) 1290 return false; 1291 1292 old_num_pools = adapter->num_active_tx_pools; 1293 new_num_pools = adapter->num_active_tx_scrqs; 1294 old_pool_size = adapter->prev_tx_pool_size; 1295 new_pool_size = adapter->req_tx_entries_per_subcrq; 1296 old_mtu = adapter->prev_mtu; 1297 new_mtu = adapter->req_mtu; 1298 1299 if (old_mtu != new_mtu || 1300 old_num_pools != new_num_pools || 1301 old_pool_size != new_pool_size) 1302 return false; 1303 1304 return true; 1305 } 1306 1307 /** 1308 * init_tx_pools(): Initialize the set of transmit pools in the adapter. 1309 * @netdev: net device associated with the vnic interface 1310 * 1311 * Initialize the set of transmit pools in the ibmvnic adapter associated 1312 * with the net_device @netdev. If possible, reuse the existing tx pools. 1313 * Otherwise free any existing pools and allocate a new set of pools 1314 * before initializing them. 1315 * 1316 * Return: 0 on success and negative value on error. 1317 */ 1318 static int init_tx_pools(struct net_device *netdev) 1319 { 1320 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 1321 struct device *dev = &adapter->vdev->dev; 1322 int num_pools; 1323 u64 pool_size; /* # of buffers in pool */ 1324 u64 buff_size; 1325 int i, j, rc; 1326 1327 num_pools = adapter->req_tx_queues; 1328 1329 /* We must notify the VIOS about the LTB on all resets - but we only 1330 * need to alloc/populate pools if either the number of buffers or 1331 * size of each buffer in the pool has changed. 1332 */ 1333 if (reuse_tx_pools(adapter)) { 1334 netdev_dbg(netdev, "Reusing tx pools\n"); 1335 goto update_ltb; 1336 } 1337 1338 /* Allocate/populate the pools. */ 1339 release_tx_pools(adapter); 1340 1341 pool_size = adapter->req_tx_entries_per_subcrq; 1342 num_pools = adapter->num_active_tx_scrqs; 1343 1344 adapter->tx_pool = kcalloc(num_pools, 1345 sizeof(struct ibmvnic_tx_pool), GFP_KERNEL); 1346 if (!adapter->tx_pool) 1347 return -ENOMEM; 1348 1349 adapter->tso_pool = kcalloc(num_pools, 1350 sizeof(struct ibmvnic_tx_pool), GFP_KERNEL); 1351 /* To simplify release_tx_pools() ensure that ->tx_pool and 1352 * ->tso_pool are either both NULL or both non-NULL. 1353 */ 1354 if (!adapter->tso_pool) { 1355 kfree(adapter->tx_pool); 1356 adapter->tx_pool = NULL; 1357 return -ENOMEM; 1358 } 1359 1360 /* Set num_active_tx_pools early. If we fail below after partial 1361 * allocation, release_tx_pools() will know how many to look for. 1362 */ 1363 adapter->num_active_tx_pools = num_pools; 1364 1365 buff_size = adapter->req_mtu + VLAN_HLEN; 1366 buff_size = ALIGN(buff_size, L1_CACHE_BYTES); 1367 1368 for (i = 0; i < num_pools; i++) { 1369 dev_dbg(dev, "Init tx pool %d [%llu, %llu]\n", 1370 i, adapter->req_tx_entries_per_subcrq, buff_size); 1371 1372 rc = init_one_tx_pool(netdev, &adapter->tx_pool[i], 1373 pool_size, buff_size); 1374 if (rc) 1375 goto out_release; 1376 1377 rc = init_one_tx_pool(netdev, &adapter->tso_pool[i], 1378 IBMVNIC_TSO_BUFS, 1379 IBMVNIC_TSO_BUF_SZ); 1380 if (rc) 1381 goto out_release; 1382 } 1383 1384 adapter->prev_tx_pool_size = pool_size; 1385 adapter->prev_mtu = adapter->req_mtu; 1386 1387 update_ltb: 1388 /* NOTE: All tx_pools have the same number of buffers (which is 1389 * same as pool_size). All tso_pools have IBMVNIC_TSO_BUFS 1390 * buffers (see calls init_one_tx_pool() for these). 1391 * For consistency, we use tx_pool->num_buffers and 1392 * tso_pool->num_buffers below. 1393 */ 1394 rc = -1; 1395 for (i = 0; i < num_pools; i++) { 1396 struct ibmvnic_tx_pool *tso_pool; 1397 struct ibmvnic_tx_pool *tx_pool; 1398 1399 tx_pool = &adapter->tx_pool[i]; 1400 1401 dev_dbg(dev, "Updating LTB for tx pool %d [%d, %d]\n", 1402 i, tx_pool->num_buffers, tx_pool->buf_size); 1403 1404 rc = alloc_ltb_set(adapter, &tx_pool->ltb_set, 1405 tx_pool->num_buffers, tx_pool->buf_size); 1406 if (rc) 1407 goto out; 1408 1409 tx_pool->consumer_index = 0; 1410 tx_pool->producer_index = 0; 1411 1412 for (j = 0; j < tx_pool->num_buffers; j++) 1413 tx_pool->free_map[j] = j; 1414 1415 tso_pool = &adapter->tso_pool[i]; 1416 1417 dev_dbg(dev, "Updating LTB for tso pool %d [%d, %d]\n", 1418 i, tso_pool->num_buffers, tso_pool->buf_size); 1419 1420 rc = alloc_ltb_set(adapter, &tso_pool->ltb_set, 1421 tso_pool->num_buffers, tso_pool->buf_size); 1422 if (rc) 1423 goto out; 1424 1425 tso_pool->consumer_index = 0; 1426 tso_pool->producer_index = 0; 1427 1428 for (j = 0; j < tso_pool->num_buffers; j++) 1429 tso_pool->free_map[j] = j; 1430 } 1431 1432 return 0; 1433 out_release: 1434 release_tx_pools(adapter); 1435 out: 1436 /* We failed to allocate one or more LTBs or map them on the VIOS. 1437 * Hold onto the pools and any LTBs that we did allocate/map. 1438 */ 1439 return rc; 1440 } 1441 1442 static void ibmvnic_napi_enable(struct ibmvnic_adapter *adapter) 1443 { 1444 int i; 1445 1446 if (adapter->napi_enabled) 1447 return; 1448 1449 for (i = 0; i < adapter->req_rx_queues; i++) 1450 napi_enable(&adapter->napi[i]); 1451 1452 adapter->napi_enabled = true; 1453 } 1454 1455 static void ibmvnic_napi_disable(struct ibmvnic_adapter *adapter) 1456 { 1457 int i; 1458 1459 if (!adapter->napi_enabled) 1460 return; 1461 1462 for (i = 0; i < adapter->req_rx_queues; i++) { 1463 netdev_dbg(adapter->netdev, "Disabling napi[%d]\n", i); 1464 napi_disable(&adapter->napi[i]); 1465 } 1466 1467 adapter->napi_enabled = false; 1468 } 1469 1470 static int init_napi(struct ibmvnic_adapter *adapter) 1471 { 1472 int i; 1473 1474 adapter->napi = kcalloc(adapter->req_rx_queues, 1475 sizeof(struct napi_struct), GFP_KERNEL); 1476 if (!adapter->napi) 1477 return -ENOMEM; 1478 1479 for (i = 0; i < adapter->req_rx_queues; i++) { 1480 netdev_dbg(adapter->netdev, "Adding napi[%d]\n", i); 1481 netif_napi_add(adapter->netdev, &adapter->napi[i], 1482 ibmvnic_poll); 1483 } 1484 1485 adapter->num_active_rx_napi = adapter->req_rx_queues; 1486 return 0; 1487 } 1488 1489 static void release_napi(struct ibmvnic_adapter *adapter) 1490 { 1491 int i; 1492 1493 if (!adapter->napi) 1494 return; 1495 1496 for (i = 0; i < adapter->num_active_rx_napi; i++) { 1497 netdev_dbg(adapter->netdev, "Releasing napi[%d]\n", i); 1498 netif_napi_del(&adapter->napi[i]); 1499 } 1500 1501 kfree(adapter->napi); 1502 adapter->napi = NULL; 1503 adapter->num_active_rx_napi = 0; 1504 adapter->napi_enabled = false; 1505 } 1506 1507 static const char *adapter_state_to_string(enum vnic_state state) 1508 { 1509 switch (state) { 1510 case VNIC_PROBING: 1511 return "PROBING"; 1512 case VNIC_PROBED: 1513 return "PROBED"; 1514 case VNIC_OPENING: 1515 return "OPENING"; 1516 case VNIC_OPEN: 1517 return "OPEN"; 1518 case VNIC_CLOSING: 1519 return "CLOSING"; 1520 case VNIC_CLOSED: 1521 return "CLOSED"; 1522 case VNIC_REMOVING: 1523 return "REMOVING"; 1524 case VNIC_REMOVED: 1525 return "REMOVED"; 1526 case VNIC_DOWN: 1527 return "DOWN"; 1528 } 1529 return "UNKNOWN"; 1530 } 1531 1532 static int ibmvnic_login(struct net_device *netdev) 1533 { 1534 unsigned long flags, timeout = msecs_to_jiffies(20000); 1535 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 1536 int retry_count = 0; 1537 int retries = 10; 1538 bool retry; 1539 int rc; 1540 1541 do { 1542 retry = false; 1543 if (retry_count > retries) { 1544 netdev_warn(netdev, "Login attempts exceeded\n"); 1545 return -EACCES; 1546 } 1547 1548 adapter->init_done_rc = 0; 1549 reinit_completion(&adapter->init_done); 1550 rc = send_login(adapter); 1551 if (rc) 1552 return rc; 1553 1554 if (!wait_for_completion_timeout(&adapter->init_done, 1555 timeout)) { 1556 netdev_warn(netdev, "Login timed out\n"); 1557 adapter->login_pending = false; 1558 goto partial_reset; 1559 } 1560 1561 if (adapter->init_done_rc == ABORTED) { 1562 netdev_warn(netdev, "Login aborted, retrying...\n"); 1563 retry = true; 1564 adapter->init_done_rc = 0; 1565 retry_count++; 1566 /* FW or device may be busy, so 1567 * wait a bit before retrying login 1568 */ 1569 msleep(500); 1570 } else if (adapter->init_done_rc == PARTIALSUCCESS) { 1571 retry_count++; 1572 release_sub_crqs(adapter, 1); 1573 1574 retry = true; 1575 netdev_dbg(netdev, 1576 "Received partial success, retrying...\n"); 1577 adapter->init_done_rc = 0; 1578 reinit_completion(&adapter->init_done); 1579 send_query_cap(adapter); 1580 if (!wait_for_completion_timeout(&adapter->init_done, 1581 timeout)) { 1582 netdev_warn(netdev, 1583 "Capabilities query timed out\n"); 1584 return -ETIMEDOUT; 1585 } 1586 1587 rc = init_sub_crqs(adapter); 1588 if (rc) { 1589 netdev_warn(netdev, 1590 "SCRQ initialization failed\n"); 1591 return rc; 1592 } 1593 1594 rc = init_sub_crq_irqs(adapter); 1595 if (rc) { 1596 netdev_warn(netdev, 1597 "SCRQ irq initialization failed\n"); 1598 return rc; 1599 } 1600 /* Default/timeout error handling, reset and start fresh */ 1601 } else if (adapter->init_done_rc) { 1602 netdev_warn(netdev, "Adapter login failed, init_done_rc = %d\n", 1603 adapter->init_done_rc); 1604 1605 partial_reset: 1606 /* adapter login failed, so free any CRQs or sub-CRQs 1607 * and register again before attempting to login again. 1608 * If we don't do this then the VIOS may think that 1609 * we are already logged in and reject any subsequent 1610 * attempts 1611 */ 1612 netdev_warn(netdev, 1613 "Freeing and re-registering CRQs before attempting to login again\n"); 1614 retry = true; 1615 adapter->init_done_rc = 0; 1616 release_sub_crqs(adapter, true); 1617 /* Much of this is similar logic as ibmvnic_probe(), 1618 * we are essentially re-initializing communication 1619 * with the server. We really should not run any 1620 * resets/failovers here because this is already a form 1621 * of reset and we do not want parallel resets occurring 1622 */ 1623 do { 1624 reinit_init_done(adapter); 1625 /* Clear any failovers we got in the previous 1626 * pass since we are re-initializing the CRQ 1627 */ 1628 adapter->failover_pending = false; 1629 release_crq_queue(adapter); 1630 /* If we don't sleep here then we risk an 1631 * unnecessary failover event from the VIOS. 1632 * This is a known VIOS issue caused by a vnic 1633 * device freeing and registering a CRQ too 1634 * quickly. 1635 */ 1636 msleep(1500); 1637 /* Avoid any resets, since we are currently 1638 * resetting. 1639 */ 1640 spin_lock_irqsave(&adapter->rwi_lock, flags); 1641 flush_reset_queue(adapter); 1642 spin_unlock_irqrestore(&adapter->rwi_lock, 1643 flags); 1644 1645 rc = init_crq_queue(adapter); 1646 if (rc) { 1647 netdev_err(netdev, "login recovery: init CRQ failed %d\n", 1648 rc); 1649 return -EIO; 1650 } 1651 1652 rc = ibmvnic_reset_init(adapter, false); 1653 if (rc) 1654 netdev_err(netdev, "login recovery: Reset init failed %d\n", 1655 rc); 1656 /* IBMVNIC_CRQ_INIT will return EAGAIN if it 1657 * fails, since ibmvnic_reset_init will free 1658 * irq's in failure, we won't be able to receive 1659 * new CRQs so we need to keep trying. probe() 1660 * handles this similarly. 1661 */ 1662 } while (rc == -EAGAIN && retry_count++ < retries); 1663 } 1664 } while (retry); 1665 1666 __ibmvnic_set_mac(netdev, adapter->mac_addr); 1667 1668 netdev_dbg(netdev, "[S:%s] Login succeeded\n", adapter_state_to_string(adapter->state)); 1669 return 0; 1670 } 1671 1672 static void release_login_buffer(struct ibmvnic_adapter *adapter) 1673 { 1674 if (!adapter->login_buf) 1675 return; 1676 1677 dma_unmap_single(&adapter->vdev->dev, adapter->login_buf_token, 1678 adapter->login_buf_sz, DMA_TO_DEVICE); 1679 kfree(adapter->login_buf); 1680 adapter->login_buf = NULL; 1681 } 1682 1683 static void release_login_rsp_buffer(struct ibmvnic_adapter *adapter) 1684 { 1685 if (!adapter->login_rsp_buf) 1686 return; 1687 1688 dma_unmap_single(&adapter->vdev->dev, adapter->login_rsp_buf_token, 1689 adapter->login_rsp_buf_sz, DMA_FROM_DEVICE); 1690 kfree(adapter->login_rsp_buf); 1691 adapter->login_rsp_buf = NULL; 1692 } 1693 1694 static void release_resources(struct ibmvnic_adapter *adapter) 1695 { 1696 release_vpd_data(adapter); 1697 1698 release_napi(adapter); 1699 release_login_buffer(adapter); 1700 release_login_rsp_buffer(adapter); 1701 } 1702 1703 static int set_link_state(struct ibmvnic_adapter *adapter, u8 link_state) 1704 { 1705 struct net_device *netdev = adapter->netdev; 1706 unsigned long timeout = msecs_to_jiffies(20000); 1707 union ibmvnic_crq crq; 1708 bool resend; 1709 int rc; 1710 1711 netdev_dbg(netdev, "setting link state %d\n", link_state); 1712 1713 memset(&crq, 0, sizeof(crq)); 1714 crq.logical_link_state.first = IBMVNIC_CRQ_CMD; 1715 crq.logical_link_state.cmd = LOGICAL_LINK_STATE; 1716 crq.logical_link_state.link_state = link_state; 1717 1718 do { 1719 resend = false; 1720 1721 reinit_completion(&adapter->init_done); 1722 rc = ibmvnic_send_crq(adapter, &crq); 1723 if (rc) { 1724 netdev_err(netdev, "Failed to set link state\n"); 1725 return rc; 1726 } 1727 1728 if (!wait_for_completion_timeout(&adapter->init_done, 1729 timeout)) { 1730 netdev_err(netdev, "timeout setting link state\n"); 1731 return -ETIMEDOUT; 1732 } 1733 1734 if (adapter->init_done_rc == PARTIALSUCCESS) { 1735 /* Partuial success, delay and re-send */ 1736 mdelay(1000); 1737 resend = true; 1738 } else if (adapter->init_done_rc) { 1739 netdev_warn(netdev, "Unable to set link state, rc=%d\n", 1740 adapter->init_done_rc); 1741 return adapter->init_done_rc; 1742 } 1743 } while (resend); 1744 1745 return 0; 1746 } 1747 1748 static int set_real_num_queues(struct net_device *netdev) 1749 { 1750 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 1751 int rc; 1752 1753 netdev_dbg(netdev, "Setting real tx/rx queues (%llx/%llx)\n", 1754 adapter->req_tx_queues, adapter->req_rx_queues); 1755 1756 rc = netif_set_real_num_tx_queues(netdev, adapter->req_tx_queues); 1757 if (rc) { 1758 netdev_err(netdev, "failed to set the number of tx queues\n"); 1759 return rc; 1760 } 1761 1762 rc = netif_set_real_num_rx_queues(netdev, adapter->req_rx_queues); 1763 if (rc) 1764 netdev_err(netdev, "failed to set the number of rx queues\n"); 1765 1766 return rc; 1767 } 1768 1769 static int ibmvnic_get_vpd(struct ibmvnic_adapter *adapter) 1770 { 1771 struct device *dev = &adapter->vdev->dev; 1772 union ibmvnic_crq crq; 1773 int len = 0; 1774 int rc; 1775 1776 if (adapter->vpd->buff) 1777 len = adapter->vpd->len; 1778 1779 mutex_lock(&adapter->fw_lock); 1780 adapter->fw_done_rc = 0; 1781 reinit_completion(&adapter->fw_done); 1782 1783 crq.get_vpd_size.first = IBMVNIC_CRQ_CMD; 1784 crq.get_vpd_size.cmd = GET_VPD_SIZE; 1785 rc = ibmvnic_send_crq(adapter, &crq); 1786 if (rc) { 1787 mutex_unlock(&adapter->fw_lock); 1788 return rc; 1789 } 1790 1791 rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000); 1792 if (rc) { 1793 dev_err(dev, "Could not retrieve VPD size, rc = %d\n", rc); 1794 mutex_unlock(&adapter->fw_lock); 1795 return rc; 1796 } 1797 mutex_unlock(&adapter->fw_lock); 1798 1799 if (!adapter->vpd->len) 1800 return -ENODATA; 1801 1802 if (!adapter->vpd->buff) 1803 adapter->vpd->buff = kzalloc(adapter->vpd->len, GFP_KERNEL); 1804 else if (adapter->vpd->len != len) 1805 adapter->vpd->buff = 1806 krealloc(adapter->vpd->buff, 1807 adapter->vpd->len, GFP_KERNEL); 1808 1809 if (!adapter->vpd->buff) { 1810 dev_err(dev, "Could allocate VPD buffer\n"); 1811 return -ENOMEM; 1812 } 1813 1814 adapter->vpd->dma_addr = 1815 dma_map_single(dev, adapter->vpd->buff, adapter->vpd->len, 1816 DMA_FROM_DEVICE); 1817 if (dma_mapping_error(dev, adapter->vpd->dma_addr)) { 1818 dev_err(dev, "Could not map VPD buffer\n"); 1819 kfree(adapter->vpd->buff); 1820 adapter->vpd->buff = NULL; 1821 return -ENOMEM; 1822 } 1823 1824 mutex_lock(&adapter->fw_lock); 1825 adapter->fw_done_rc = 0; 1826 reinit_completion(&adapter->fw_done); 1827 1828 crq.get_vpd.first = IBMVNIC_CRQ_CMD; 1829 crq.get_vpd.cmd = GET_VPD; 1830 crq.get_vpd.ioba = cpu_to_be32(adapter->vpd->dma_addr); 1831 crq.get_vpd.len = cpu_to_be32((u32)adapter->vpd->len); 1832 rc = ibmvnic_send_crq(adapter, &crq); 1833 if (rc) { 1834 kfree(adapter->vpd->buff); 1835 adapter->vpd->buff = NULL; 1836 mutex_unlock(&adapter->fw_lock); 1837 return rc; 1838 } 1839 1840 rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000); 1841 if (rc) { 1842 dev_err(dev, "Unable to retrieve VPD, rc = %d\n", rc); 1843 kfree(adapter->vpd->buff); 1844 adapter->vpd->buff = NULL; 1845 mutex_unlock(&adapter->fw_lock); 1846 return rc; 1847 } 1848 1849 mutex_unlock(&adapter->fw_lock); 1850 return 0; 1851 } 1852 1853 static int init_resources(struct ibmvnic_adapter *adapter) 1854 { 1855 struct net_device *netdev = adapter->netdev; 1856 int rc; 1857 1858 rc = set_real_num_queues(netdev); 1859 if (rc) 1860 return rc; 1861 1862 adapter->vpd = kzalloc(sizeof(*adapter->vpd), GFP_KERNEL); 1863 if (!adapter->vpd) 1864 return -ENOMEM; 1865 1866 /* Vital Product Data (VPD) */ 1867 rc = ibmvnic_get_vpd(adapter); 1868 if (rc) { 1869 netdev_err(netdev, "failed to initialize Vital Product Data (VPD)\n"); 1870 return rc; 1871 } 1872 1873 rc = init_napi(adapter); 1874 if (rc) 1875 return rc; 1876 1877 send_query_map(adapter); 1878 1879 rc = init_rx_pools(netdev); 1880 if (rc) 1881 return rc; 1882 1883 rc = init_tx_pools(netdev); 1884 return rc; 1885 } 1886 1887 static int __ibmvnic_open(struct net_device *netdev) 1888 { 1889 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 1890 enum vnic_state prev_state = adapter->state; 1891 int i, rc; 1892 1893 adapter->state = VNIC_OPENING; 1894 replenish_pools(adapter); 1895 ibmvnic_napi_enable(adapter); 1896 1897 /* We're ready to receive frames, enable the sub-crq interrupts and 1898 * set the logical link state to up 1899 */ 1900 for (i = 0; i < adapter->req_rx_queues; i++) { 1901 netdev_dbg(netdev, "Enabling rx_scrq[%d] irq\n", i); 1902 if (prev_state == VNIC_CLOSED) 1903 enable_irq(adapter->rx_scrq[i]->irq); 1904 enable_scrq_irq(adapter, adapter->rx_scrq[i]); 1905 } 1906 1907 for (i = 0; i < adapter->req_tx_queues; i++) { 1908 netdev_dbg(netdev, "Enabling tx_scrq[%d] irq\n", i); 1909 if (prev_state == VNIC_CLOSED) 1910 enable_irq(adapter->tx_scrq[i]->irq); 1911 enable_scrq_irq(adapter, adapter->tx_scrq[i]); 1912 /* netdev_tx_reset_queue will reset dql stats. During NON_FATAL 1913 * resets, don't reset the stats because there could be batched 1914 * skb's waiting to be sent. If we reset dql stats, we risk 1915 * num_completed being greater than num_queued. This will cause 1916 * a BUG_ON in dql_completed(). 1917 */ 1918 if (adapter->reset_reason != VNIC_RESET_NON_FATAL) 1919 netdev_tx_reset_queue(netdev_get_tx_queue(netdev, i)); 1920 } 1921 1922 rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_UP); 1923 if (rc) { 1924 ibmvnic_napi_disable(adapter); 1925 ibmvnic_disable_irqs(adapter); 1926 return rc; 1927 } 1928 1929 adapter->tx_queues_active = true; 1930 1931 /* Since queues were stopped until now, there shouldn't be any 1932 * one in ibmvnic_complete_tx() or ibmvnic_xmit() so maybe we 1933 * don't need the synchronize_rcu()? Leaving it for consistency 1934 * with setting ->tx_queues_active = false. 1935 */ 1936 synchronize_rcu(); 1937 1938 netif_tx_start_all_queues(netdev); 1939 1940 if (prev_state == VNIC_CLOSED) { 1941 for (i = 0; i < adapter->req_rx_queues; i++) 1942 napi_schedule(&adapter->napi[i]); 1943 } 1944 1945 adapter->state = VNIC_OPEN; 1946 return rc; 1947 } 1948 1949 static int ibmvnic_open(struct net_device *netdev) 1950 { 1951 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 1952 int rc; 1953 1954 ASSERT_RTNL(); 1955 1956 /* If device failover is pending or we are about to reset, just set 1957 * device state and return. Device operation will be handled by reset 1958 * routine. 1959 * 1960 * It should be safe to overwrite the adapter->state here. Since 1961 * we hold the rtnl, either the reset has not actually started or 1962 * the rtnl got dropped during the set_link_state() in do_reset(). 1963 * In the former case, no one else is changing the state (again we 1964 * have the rtnl) and in the latter case, do_reset() will detect and 1965 * honor our setting below. 1966 */ 1967 if (adapter->failover_pending || (test_bit(0, &adapter->resetting))) { 1968 netdev_dbg(netdev, "[S:%s FOP:%d] Resetting, deferring open\n", 1969 adapter_state_to_string(adapter->state), 1970 adapter->failover_pending); 1971 adapter->state = VNIC_OPEN; 1972 rc = 0; 1973 goto out; 1974 } 1975 1976 if (adapter->state != VNIC_CLOSED) { 1977 rc = ibmvnic_login(netdev); 1978 if (rc) 1979 goto out; 1980 1981 rc = init_resources(adapter); 1982 if (rc) { 1983 netdev_err(netdev, "failed to initialize resources\n"); 1984 goto out; 1985 } 1986 } 1987 1988 rc = __ibmvnic_open(netdev); 1989 1990 out: 1991 /* If open failed and there is a pending failover or in-progress reset, 1992 * set device state and return. Device operation will be handled by 1993 * reset routine. See also comments above regarding rtnl. 1994 */ 1995 if (rc && 1996 (adapter->failover_pending || (test_bit(0, &adapter->resetting)))) { 1997 adapter->state = VNIC_OPEN; 1998 rc = 0; 1999 } 2000 2001 if (rc) { 2002 release_resources(adapter); 2003 release_rx_pools(adapter); 2004 release_tx_pools(adapter); 2005 } 2006 2007 return rc; 2008 } 2009 2010 static void clean_rx_pools(struct ibmvnic_adapter *adapter) 2011 { 2012 struct ibmvnic_rx_pool *rx_pool; 2013 struct ibmvnic_rx_buff *rx_buff; 2014 u64 rx_entries; 2015 int rx_scrqs; 2016 int i, j; 2017 2018 if (!adapter->rx_pool) 2019 return; 2020 2021 rx_scrqs = adapter->num_active_rx_pools; 2022 rx_entries = adapter->req_rx_add_entries_per_subcrq; 2023 2024 /* Free any remaining skbs in the rx buffer pools */ 2025 for (i = 0; i < rx_scrqs; i++) { 2026 rx_pool = &adapter->rx_pool[i]; 2027 if (!rx_pool || !rx_pool->rx_buff) 2028 continue; 2029 2030 netdev_dbg(adapter->netdev, "Cleaning rx_pool[%d]\n", i); 2031 for (j = 0; j < rx_entries; j++) { 2032 rx_buff = &rx_pool->rx_buff[j]; 2033 if (rx_buff && rx_buff->skb) { 2034 dev_kfree_skb_any(rx_buff->skb); 2035 rx_buff->skb = NULL; 2036 } 2037 } 2038 } 2039 } 2040 2041 static void clean_one_tx_pool(struct ibmvnic_adapter *adapter, 2042 struct ibmvnic_tx_pool *tx_pool) 2043 { 2044 struct ibmvnic_tx_buff *tx_buff; 2045 u64 tx_entries; 2046 int i; 2047 2048 if (!tx_pool || !tx_pool->tx_buff) 2049 return; 2050 2051 tx_entries = tx_pool->num_buffers; 2052 2053 for (i = 0; i < tx_entries; i++) { 2054 tx_buff = &tx_pool->tx_buff[i]; 2055 if (tx_buff && tx_buff->skb) { 2056 dev_kfree_skb_any(tx_buff->skb); 2057 tx_buff->skb = NULL; 2058 } 2059 } 2060 } 2061 2062 static void clean_tx_pools(struct ibmvnic_adapter *adapter) 2063 { 2064 int tx_scrqs; 2065 int i; 2066 2067 if (!adapter->tx_pool || !adapter->tso_pool) 2068 return; 2069 2070 tx_scrqs = adapter->num_active_tx_pools; 2071 2072 /* Free any remaining skbs in the tx buffer pools */ 2073 for (i = 0; i < tx_scrqs; i++) { 2074 netdev_dbg(adapter->netdev, "Cleaning tx_pool[%d]\n", i); 2075 clean_one_tx_pool(adapter, &adapter->tx_pool[i]); 2076 clean_one_tx_pool(adapter, &adapter->tso_pool[i]); 2077 } 2078 } 2079 2080 static void ibmvnic_disable_irqs(struct ibmvnic_adapter *adapter) 2081 { 2082 struct net_device *netdev = adapter->netdev; 2083 int i; 2084 2085 if (adapter->tx_scrq) { 2086 for (i = 0; i < adapter->req_tx_queues; i++) 2087 if (adapter->tx_scrq[i]->irq) { 2088 netdev_dbg(netdev, 2089 "Disabling tx_scrq[%d] irq\n", i); 2090 disable_scrq_irq(adapter, adapter->tx_scrq[i]); 2091 disable_irq(adapter->tx_scrq[i]->irq); 2092 } 2093 } 2094 2095 if (adapter->rx_scrq) { 2096 for (i = 0; i < adapter->req_rx_queues; i++) { 2097 if (adapter->rx_scrq[i]->irq) { 2098 netdev_dbg(netdev, 2099 "Disabling rx_scrq[%d] irq\n", i); 2100 disable_scrq_irq(adapter, adapter->rx_scrq[i]); 2101 disable_irq(adapter->rx_scrq[i]->irq); 2102 } 2103 } 2104 } 2105 } 2106 2107 static void ibmvnic_cleanup(struct net_device *netdev) 2108 { 2109 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 2110 2111 /* ensure that transmissions are stopped if called by do_reset */ 2112 2113 adapter->tx_queues_active = false; 2114 2115 /* Ensure complete_tx() and ibmvnic_xmit() see ->tx_queues_active 2116 * update so they don't restart a queue after we stop it below. 2117 */ 2118 synchronize_rcu(); 2119 2120 if (test_bit(0, &adapter->resetting)) 2121 netif_tx_disable(netdev); 2122 else 2123 netif_tx_stop_all_queues(netdev); 2124 2125 ibmvnic_napi_disable(adapter); 2126 ibmvnic_disable_irqs(adapter); 2127 } 2128 2129 static int __ibmvnic_close(struct net_device *netdev) 2130 { 2131 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 2132 int rc = 0; 2133 2134 adapter->state = VNIC_CLOSING; 2135 rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN); 2136 adapter->state = VNIC_CLOSED; 2137 return rc; 2138 } 2139 2140 static int ibmvnic_close(struct net_device *netdev) 2141 { 2142 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 2143 int rc; 2144 2145 netdev_dbg(netdev, "[S:%s FOP:%d FRR:%d] Closing\n", 2146 adapter_state_to_string(adapter->state), 2147 adapter->failover_pending, 2148 adapter->force_reset_recovery); 2149 2150 /* If device failover is pending, just set device state and return. 2151 * Device operation will be handled by reset routine. 2152 */ 2153 if (adapter->failover_pending) { 2154 adapter->state = VNIC_CLOSED; 2155 return 0; 2156 } 2157 2158 rc = __ibmvnic_close(netdev); 2159 ibmvnic_cleanup(netdev); 2160 clean_rx_pools(adapter); 2161 clean_tx_pools(adapter); 2162 2163 return rc; 2164 } 2165 2166 /** 2167 * get_hdr_lens - fills list of L2/L3/L4 hdr lens 2168 * @hdr_field: bitfield determining needed headers 2169 * @skb: socket buffer 2170 * @hdr_len: array of header lengths to be filled 2171 * 2172 * Reads hdr_field to determine which headers are needed by firmware. 2173 * Builds a buffer containing these headers. Saves individual header 2174 * lengths and total buffer length to be used to build descriptors. 2175 * 2176 * Return: total len of all headers 2177 */ 2178 static int get_hdr_lens(u8 hdr_field, struct sk_buff *skb, 2179 int *hdr_len) 2180 { 2181 int len = 0; 2182 2183 2184 if ((hdr_field >> 6) & 1) { 2185 hdr_len[0] = skb_mac_header_len(skb); 2186 len += hdr_len[0]; 2187 } 2188 2189 if ((hdr_field >> 5) & 1) { 2190 hdr_len[1] = skb_network_header_len(skb); 2191 len += hdr_len[1]; 2192 } 2193 2194 if (!((hdr_field >> 4) & 1)) 2195 return len; 2196 2197 if (skb->protocol == htons(ETH_P_IP)) { 2198 if (ip_hdr(skb)->protocol == IPPROTO_TCP) 2199 hdr_len[2] = tcp_hdrlen(skb); 2200 else if (ip_hdr(skb)->protocol == IPPROTO_UDP) 2201 hdr_len[2] = sizeof(struct udphdr); 2202 } else if (skb->protocol == htons(ETH_P_IPV6)) { 2203 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP) 2204 hdr_len[2] = tcp_hdrlen(skb); 2205 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP) 2206 hdr_len[2] = sizeof(struct udphdr); 2207 } 2208 2209 return len + hdr_len[2]; 2210 } 2211 2212 /** 2213 * create_hdr_descs - create header and header extension descriptors 2214 * @hdr_field: bitfield determining needed headers 2215 * @hdr_data: buffer containing header data 2216 * @len: length of data buffer 2217 * @hdr_len: array of individual header lengths 2218 * @scrq_arr: descriptor array 2219 * 2220 * Creates header and, if needed, header extension descriptors and 2221 * places them in a descriptor array, scrq_arr 2222 * 2223 * Return: Number of header descs 2224 */ 2225 2226 static int create_hdr_descs(u8 hdr_field, u8 *hdr_data, int len, int *hdr_len, 2227 union sub_crq *scrq_arr) 2228 { 2229 union sub_crq *hdr_desc; 2230 int tmp_len = len; 2231 int num_descs = 0; 2232 u8 *data, *cur; 2233 int tmp; 2234 2235 while (tmp_len > 0) { 2236 cur = hdr_data + len - tmp_len; 2237 2238 hdr_desc = &scrq_arr[num_descs]; 2239 if (num_descs) { 2240 data = hdr_desc->hdr_ext.data; 2241 tmp = tmp_len > 29 ? 29 : tmp_len; 2242 hdr_desc->hdr_ext.first = IBMVNIC_CRQ_CMD; 2243 hdr_desc->hdr_ext.type = IBMVNIC_HDR_EXT_DESC; 2244 hdr_desc->hdr_ext.len = tmp; 2245 } else { 2246 data = hdr_desc->hdr.data; 2247 tmp = tmp_len > 24 ? 24 : tmp_len; 2248 hdr_desc->hdr.first = IBMVNIC_CRQ_CMD; 2249 hdr_desc->hdr.type = IBMVNIC_HDR_DESC; 2250 hdr_desc->hdr.len = tmp; 2251 hdr_desc->hdr.l2_len = (u8)hdr_len[0]; 2252 hdr_desc->hdr.l3_len = cpu_to_be16((u16)hdr_len[1]); 2253 hdr_desc->hdr.l4_len = (u8)hdr_len[2]; 2254 hdr_desc->hdr.flag = hdr_field << 1; 2255 } 2256 memcpy(data, cur, tmp); 2257 tmp_len -= tmp; 2258 num_descs++; 2259 } 2260 2261 return num_descs; 2262 } 2263 2264 /** 2265 * build_hdr_descs_arr - build a header descriptor array 2266 * @skb: tx socket buffer 2267 * @indir_arr: indirect array 2268 * @num_entries: number of descriptors to be sent 2269 * @hdr_field: bit field determining which headers will be sent 2270 * 2271 * This function will build a TX descriptor array with applicable 2272 * L2/L3/L4 packet header descriptors to be sent by send_subcrq_indirect. 2273 */ 2274 2275 static void build_hdr_descs_arr(struct sk_buff *skb, 2276 union sub_crq *indir_arr, 2277 int *num_entries, u8 hdr_field) 2278 { 2279 int hdr_len[3] = {0, 0, 0}; 2280 int tot_len; 2281 2282 tot_len = get_hdr_lens(hdr_field, skb, hdr_len); 2283 *num_entries += create_hdr_descs(hdr_field, skb_mac_header(skb), 2284 tot_len, hdr_len, indir_arr + 1); 2285 } 2286 2287 static int ibmvnic_xmit_workarounds(struct sk_buff *skb, 2288 struct net_device *netdev) 2289 { 2290 /* For some backing devices, mishandling of small packets 2291 * can result in a loss of connection or TX stall. Device 2292 * architects recommend that no packet should be smaller 2293 * than the minimum MTU value provided to the driver, so 2294 * pad any packets to that length 2295 */ 2296 if (skb->len < netdev->min_mtu) 2297 return skb_put_padto(skb, netdev->min_mtu); 2298 2299 return 0; 2300 } 2301 2302 static void ibmvnic_tx_scrq_clean_buffer(struct ibmvnic_adapter *adapter, 2303 struct ibmvnic_sub_crq_queue *tx_scrq) 2304 { 2305 struct ibmvnic_ind_xmit_queue *ind_bufp; 2306 struct ibmvnic_tx_buff *tx_buff; 2307 struct ibmvnic_tx_pool *tx_pool; 2308 union sub_crq tx_scrq_entry; 2309 int queue_num; 2310 int entries; 2311 int index; 2312 int i; 2313 2314 ind_bufp = &tx_scrq->ind_buf; 2315 entries = (u64)ind_bufp->index; 2316 queue_num = tx_scrq->pool_index; 2317 2318 for (i = entries - 1; i >= 0; --i) { 2319 tx_scrq_entry = ind_bufp->indir_arr[i]; 2320 if (tx_scrq_entry.v1.type != IBMVNIC_TX_DESC) 2321 continue; 2322 index = be32_to_cpu(tx_scrq_entry.v1.correlator); 2323 if (index & IBMVNIC_TSO_POOL_MASK) { 2324 tx_pool = &adapter->tso_pool[queue_num]; 2325 index &= ~IBMVNIC_TSO_POOL_MASK; 2326 } else { 2327 tx_pool = &adapter->tx_pool[queue_num]; 2328 } 2329 tx_pool->free_map[tx_pool->consumer_index] = index; 2330 tx_pool->consumer_index = tx_pool->consumer_index == 0 ? 2331 tx_pool->num_buffers - 1 : 2332 tx_pool->consumer_index - 1; 2333 tx_buff = &tx_pool->tx_buff[index]; 2334 adapter->tx_stats_buffers[queue_num].batched_packets--; 2335 adapter->tx_stats_buffers[queue_num].bytes -= 2336 tx_buff->skb->len; 2337 dev_kfree_skb_any(tx_buff->skb); 2338 tx_buff->skb = NULL; 2339 adapter->netdev->stats.tx_dropped++; 2340 } 2341 2342 ind_bufp->index = 0; 2343 2344 if (atomic_sub_return(entries, &tx_scrq->used) <= 2345 (adapter->req_tx_entries_per_subcrq / 2) && 2346 __netif_subqueue_stopped(adapter->netdev, queue_num)) { 2347 rcu_read_lock(); 2348 2349 if (adapter->tx_queues_active) { 2350 netif_wake_subqueue(adapter->netdev, queue_num); 2351 netdev_dbg(adapter->netdev, "Started queue %d\n", 2352 queue_num); 2353 } 2354 2355 rcu_read_unlock(); 2356 } 2357 } 2358 2359 static int send_subcrq_direct(struct ibmvnic_adapter *adapter, 2360 u64 remote_handle, u64 *entry) 2361 { 2362 unsigned int ua = adapter->vdev->unit_address; 2363 struct device *dev = &adapter->vdev->dev; 2364 int rc; 2365 2366 /* Make sure the hypervisor sees the complete request */ 2367 dma_wmb(); 2368 rc = plpar_hcall_norets(H_SEND_SUB_CRQ, ua, 2369 cpu_to_be64(remote_handle), 2370 cpu_to_be64(entry[0]), cpu_to_be64(entry[1]), 2371 cpu_to_be64(entry[2]), cpu_to_be64(entry[3])); 2372 2373 if (rc) 2374 print_subcrq_error(dev, rc, __func__); 2375 2376 return rc; 2377 } 2378 2379 static int ibmvnic_tx_scrq_flush(struct ibmvnic_adapter *adapter, 2380 struct ibmvnic_sub_crq_queue *tx_scrq, 2381 bool indirect) 2382 { 2383 struct ibmvnic_ind_xmit_queue *ind_bufp; 2384 u64 dma_addr; 2385 u64 entries; 2386 u64 handle; 2387 int rc; 2388 2389 ind_bufp = &tx_scrq->ind_buf; 2390 dma_addr = (u64)ind_bufp->indir_dma; 2391 entries = (u64)ind_bufp->index; 2392 handle = tx_scrq->handle; 2393 2394 if (!entries) 2395 return 0; 2396 2397 if (indirect) 2398 rc = send_subcrq_indirect(adapter, handle, dma_addr, entries); 2399 else 2400 rc = send_subcrq_direct(adapter, handle, 2401 (u64 *)ind_bufp->indir_arr); 2402 2403 if (rc) { 2404 dev_err_ratelimited(&adapter->vdev->dev, 2405 "tx_flush failed, rc=%u (%llu entries dma=%pad handle=%llx)\n", 2406 rc, entries, &dma_addr, handle); 2407 /* Detect platform limit H_PARAMETER */ 2408 if (rc == H_PARAMETER) 2409 ibmvnic_set_safe_max_ind_descs(adapter); 2410 2411 /* For all error case, temporarily drop only this batch 2412 * Rely on TCP/IP retransmissions to retry and recover 2413 */ 2414 ibmvnic_tx_scrq_clean_buffer(adapter, tx_scrq); 2415 } else { 2416 ind_bufp->index = 0; 2417 } 2418 return rc; 2419 } 2420 2421 static netdev_tx_t ibmvnic_xmit(struct sk_buff *skb, struct net_device *netdev) 2422 { 2423 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 2424 u32 cur_max_ind_descs = adapter->cur_max_ind_descs; 2425 int queue_num = skb_get_queue_mapping(skb); 2426 u8 *hdrs = (u8 *)&adapter->tx_rx_desc_req; 2427 struct device *dev = &adapter->vdev->dev; 2428 struct ibmvnic_ind_xmit_queue *ind_bufp; 2429 struct ibmvnic_tx_buff *tx_buff = NULL; 2430 struct ibmvnic_sub_crq_queue *tx_scrq; 2431 struct ibmvnic_long_term_buff *ltb; 2432 struct ibmvnic_tx_pool *tx_pool; 2433 unsigned int tx_send_failed = 0; 2434 netdev_tx_t ret = NETDEV_TX_OK; 2435 unsigned int tx_map_failed = 0; 2436 union sub_crq indir_arr[16]; 2437 unsigned int tx_dropped = 0; 2438 unsigned int tx_dpackets = 0; 2439 unsigned int tx_bpackets = 0; 2440 unsigned int tx_bytes = 0; 2441 dma_addr_t data_dma_addr; 2442 struct netdev_queue *txq; 2443 unsigned long lpar_rc; 2444 unsigned int skblen; 2445 union sub_crq tx_crq; 2446 unsigned int offset; 2447 bool use_scrq_send_direct = false; 2448 int num_entries = 1; 2449 unsigned char *dst; 2450 int bufidx = 0; 2451 u8 proto = 0; 2452 2453 /* If a reset is in progress, drop the packet since 2454 * the scrqs may get torn down. Otherwise use the 2455 * rcu to ensure reset waits for us to complete. 2456 */ 2457 rcu_read_lock(); 2458 if (!adapter->tx_queues_active) { 2459 dev_kfree_skb_any(skb); 2460 2461 tx_send_failed++; 2462 tx_dropped++; 2463 ret = NETDEV_TX_OK; 2464 goto out; 2465 } 2466 2467 tx_scrq = adapter->tx_scrq[queue_num]; 2468 txq = netdev_get_tx_queue(netdev, queue_num); 2469 ind_bufp = &tx_scrq->ind_buf; 2470 2471 if (ibmvnic_xmit_workarounds(skb, netdev)) { 2472 tx_dropped++; 2473 tx_send_failed++; 2474 ret = NETDEV_TX_OK; 2475 lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq, true); 2476 if (lpar_rc != H_SUCCESS) 2477 goto tx_err; 2478 goto out; 2479 } 2480 2481 if (skb_is_gso(skb)) 2482 tx_pool = &adapter->tso_pool[queue_num]; 2483 else 2484 tx_pool = &adapter->tx_pool[queue_num]; 2485 2486 bufidx = tx_pool->free_map[tx_pool->consumer_index]; 2487 2488 if (bufidx == IBMVNIC_INVALID_MAP) { 2489 dev_kfree_skb_any(skb); 2490 tx_send_failed++; 2491 tx_dropped++; 2492 ret = NETDEV_TX_OK; 2493 lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq, true); 2494 if (lpar_rc != H_SUCCESS) 2495 goto tx_err; 2496 goto out; 2497 } 2498 2499 tx_pool->free_map[tx_pool->consumer_index] = IBMVNIC_INVALID_MAP; 2500 2501 map_txpool_buf_to_ltb(tx_pool, bufidx, <b, &offset); 2502 2503 dst = ltb->buff + offset; 2504 memset(dst, 0, tx_pool->buf_size); 2505 data_dma_addr = ltb->addr + offset; 2506 2507 /* if we are going to send_subcrq_direct this then we need to 2508 * update the checksum before copying the data into ltb. Essentially 2509 * these packets force disable CSO so that we can guarantee that 2510 * FW does not need header info and we can send direct. Also, vnic 2511 * server must be able to xmit standard packets without header data 2512 */ 2513 if (*hdrs == 0 && !skb_is_gso(skb) && 2514 !ind_bufp->index && !netdev_xmit_more()) { 2515 use_scrq_send_direct = true; 2516 if (skb->ip_summed == CHECKSUM_PARTIAL && 2517 skb_checksum_help(skb)) 2518 use_scrq_send_direct = false; 2519 } 2520 2521 if (skb_shinfo(skb)->nr_frags) { 2522 int cur, i; 2523 2524 /* Copy the head */ 2525 skb_copy_from_linear_data(skb, dst, skb_headlen(skb)); 2526 cur = skb_headlen(skb); 2527 2528 /* Copy the frags */ 2529 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 2530 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 2531 2532 memcpy(dst + cur, skb_frag_address(frag), 2533 skb_frag_size(frag)); 2534 cur += skb_frag_size(frag); 2535 } 2536 } else { 2537 skb_copy_from_linear_data(skb, dst, skb->len); 2538 } 2539 2540 tx_pool->consumer_index = 2541 (tx_pool->consumer_index + 1) % tx_pool->num_buffers; 2542 2543 tx_buff = &tx_pool->tx_buff[bufidx]; 2544 2545 /* Sanity checks on our free map to make sure it points to an index 2546 * that is not being occupied by another skb. If skb memory is 2547 * not freed then we see congestion control kick in and halt tx. 2548 */ 2549 if (unlikely(tx_buff->skb)) { 2550 dev_warn_ratelimited(dev, "TX free map points to untracked skb (%s %d idx=%d)\n", 2551 skb_is_gso(skb) ? "tso_pool" : "tx_pool", 2552 queue_num, bufidx); 2553 dev_kfree_skb_any(tx_buff->skb); 2554 } 2555 2556 tx_buff->skb = skb; 2557 tx_buff->index = bufidx; 2558 tx_buff->pool_index = queue_num; 2559 skblen = skb->len; 2560 2561 memset(&tx_crq, 0, sizeof(tx_crq)); 2562 tx_crq.v1.first = IBMVNIC_CRQ_CMD; 2563 tx_crq.v1.type = IBMVNIC_TX_DESC; 2564 tx_crq.v1.n_crq_elem = 1; 2565 tx_crq.v1.n_sge = 1; 2566 tx_crq.v1.flags1 = IBMVNIC_TX_COMP_NEEDED; 2567 2568 if (skb_is_gso(skb)) 2569 tx_crq.v1.correlator = 2570 cpu_to_be32(bufidx | IBMVNIC_TSO_POOL_MASK); 2571 else 2572 tx_crq.v1.correlator = cpu_to_be32(bufidx); 2573 tx_crq.v1.dma_reg = cpu_to_be16(ltb->map_id); 2574 tx_crq.v1.sge_len = cpu_to_be32(skb->len); 2575 tx_crq.v1.ioba = cpu_to_be64(data_dma_addr); 2576 2577 if (adapter->vlan_header_insertion && skb_vlan_tag_present(skb)) { 2578 tx_crq.v1.flags2 |= IBMVNIC_TX_VLAN_INSERT; 2579 tx_crq.v1.vlan_id = cpu_to_be16(skb->vlan_tci); 2580 } 2581 2582 if (skb->protocol == htons(ETH_P_IP)) { 2583 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV4; 2584 proto = ip_hdr(skb)->protocol; 2585 } else if (skb->protocol == htons(ETH_P_IPV6)) { 2586 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV6; 2587 proto = ipv6_hdr(skb)->nexthdr; 2588 } 2589 2590 if (proto == IPPROTO_TCP) 2591 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_TCP; 2592 else if (proto == IPPROTO_UDP) 2593 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_UDP; 2594 2595 if (skb->ip_summed == CHECKSUM_PARTIAL) { 2596 tx_crq.v1.flags1 |= IBMVNIC_TX_CHKSUM_OFFLOAD; 2597 hdrs += 2; 2598 } 2599 if (skb_is_gso(skb)) { 2600 tx_crq.v1.flags1 |= IBMVNIC_TX_LSO; 2601 tx_crq.v1.mss = cpu_to_be16(skb_shinfo(skb)->gso_size); 2602 hdrs += 2; 2603 } else if (use_scrq_send_direct) { 2604 /* See above comment, CSO disabled with direct xmit */ 2605 tx_crq.v1.flags1 &= ~(IBMVNIC_TX_CHKSUM_OFFLOAD); 2606 ind_bufp->index = 1; 2607 tx_buff->num_entries = 1; 2608 netdev_tx_sent_queue(txq, skb->len); 2609 ind_bufp->indir_arr[0] = tx_crq; 2610 lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq, false); 2611 if (lpar_rc != H_SUCCESS) 2612 goto tx_err; 2613 2614 tx_dpackets++; 2615 goto early_exit; 2616 } 2617 2618 if ((*hdrs >> 7) & 1) 2619 build_hdr_descs_arr(skb, indir_arr, &num_entries, *hdrs); 2620 2621 tx_crq.v1.n_crq_elem = num_entries; 2622 tx_buff->num_entries = num_entries; 2623 /* flush buffer if current entry can not fit */ 2624 if (num_entries + ind_bufp->index > cur_max_ind_descs) { 2625 lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq, true); 2626 if (lpar_rc != H_SUCCESS) 2627 goto tx_flush_err; 2628 } 2629 2630 indir_arr[0] = tx_crq; 2631 memcpy(&ind_bufp->indir_arr[ind_bufp->index], &indir_arr[0], 2632 num_entries * sizeof(struct ibmvnic_generic_scrq)); 2633 2634 ind_bufp->index += num_entries; 2635 if (__netdev_tx_sent_queue(txq, skb->len, 2636 netdev_xmit_more() && 2637 ind_bufp->index < cur_max_ind_descs)) { 2638 lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq, true); 2639 if (lpar_rc != H_SUCCESS) 2640 goto tx_err; 2641 } 2642 2643 tx_bpackets++; 2644 2645 early_exit: 2646 if (atomic_add_return(num_entries, &tx_scrq->used) 2647 >= adapter->req_tx_entries_per_subcrq) { 2648 netdev_dbg(netdev, "Stopping queue %d\n", queue_num); 2649 netif_stop_subqueue(netdev, queue_num); 2650 } 2651 2652 tx_bytes += skblen; 2653 txq_trans_cond_update(txq); 2654 ret = NETDEV_TX_OK; 2655 goto out; 2656 2657 tx_flush_err: 2658 dev_kfree_skb_any(skb); 2659 tx_buff->skb = NULL; 2660 tx_pool->consumer_index = tx_pool->consumer_index == 0 ? 2661 tx_pool->num_buffers - 1 : 2662 tx_pool->consumer_index - 1; 2663 tx_dropped++; 2664 tx_err: 2665 if (lpar_rc != H_CLOSED && lpar_rc != H_PARAMETER) 2666 dev_err_ratelimited(dev, "tx: send failed\n"); 2667 2668 if (lpar_rc == H_CLOSED || adapter->failover_pending) { 2669 /* Disable TX and report carrier off if queue is closed 2670 * or pending failover. 2671 * Firmware guarantees that a signal will be sent to the 2672 * driver, triggering a reset or some other action. 2673 */ 2674 netif_tx_stop_all_queues(netdev); 2675 netif_carrier_off(netdev); 2676 } 2677 out: 2678 rcu_read_unlock(); 2679 adapter->tx_send_failed += tx_send_failed; 2680 adapter->tx_map_failed += tx_map_failed; 2681 adapter->tx_stats_buffers[queue_num].batched_packets += tx_bpackets; 2682 adapter->tx_stats_buffers[queue_num].direct_packets += tx_dpackets; 2683 adapter->tx_stats_buffers[queue_num].bytes += tx_bytes; 2684 adapter->tx_stats_buffers[queue_num].dropped_packets += tx_dropped; 2685 2686 return ret; 2687 } 2688 2689 static void ibmvnic_set_multi(struct net_device *netdev) 2690 { 2691 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 2692 struct netdev_hw_addr *ha; 2693 union ibmvnic_crq crq; 2694 2695 memset(&crq, 0, sizeof(crq)); 2696 crq.request_capability.first = IBMVNIC_CRQ_CMD; 2697 crq.request_capability.cmd = REQUEST_CAPABILITY; 2698 2699 if (netdev->flags & IFF_PROMISC) { 2700 if (!adapter->promisc_supported) 2701 return; 2702 } else { 2703 if (netdev->flags & IFF_ALLMULTI) { 2704 /* Accept all multicast */ 2705 memset(&crq, 0, sizeof(crq)); 2706 crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD; 2707 crq.multicast_ctrl.cmd = MULTICAST_CTRL; 2708 crq.multicast_ctrl.flags = IBMVNIC_ENABLE_ALL; 2709 ibmvnic_send_crq(adapter, &crq); 2710 } else if (netdev_mc_empty(netdev)) { 2711 /* Reject all multicast */ 2712 memset(&crq, 0, sizeof(crq)); 2713 crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD; 2714 crq.multicast_ctrl.cmd = MULTICAST_CTRL; 2715 crq.multicast_ctrl.flags = IBMVNIC_DISABLE_ALL; 2716 ibmvnic_send_crq(adapter, &crq); 2717 } else { 2718 /* Accept one or more multicast(s) */ 2719 netdev_for_each_mc_addr(ha, netdev) { 2720 memset(&crq, 0, sizeof(crq)); 2721 crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD; 2722 crq.multicast_ctrl.cmd = MULTICAST_CTRL; 2723 crq.multicast_ctrl.flags = IBMVNIC_ENABLE_MC; 2724 ether_addr_copy(&crq.multicast_ctrl.mac_addr[0], 2725 ha->addr); 2726 ibmvnic_send_crq(adapter, &crq); 2727 } 2728 } 2729 } 2730 } 2731 2732 static int __ibmvnic_set_mac(struct net_device *netdev, u8 *dev_addr) 2733 { 2734 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 2735 union ibmvnic_crq crq; 2736 int rc; 2737 2738 if (!is_valid_ether_addr(dev_addr)) { 2739 rc = -EADDRNOTAVAIL; 2740 goto err; 2741 } 2742 2743 memset(&crq, 0, sizeof(crq)); 2744 crq.change_mac_addr.first = IBMVNIC_CRQ_CMD; 2745 crq.change_mac_addr.cmd = CHANGE_MAC_ADDR; 2746 ether_addr_copy(&crq.change_mac_addr.mac_addr[0], dev_addr); 2747 2748 mutex_lock(&adapter->fw_lock); 2749 adapter->fw_done_rc = 0; 2750 reinit_completion(&adapter->fw_done); 2751 2752 rc = ibmvnic_send_crq(adapter, &crq); 2753 if (rc) { 2754 rc = -EIO; 2755 mutex_unlock(&adapter->fw_lock); 2756 goto err; 2757 } 2758 2759 rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000); 2760 /* netdev->dev_addr is changed in handle_change_mac_rsp function */ 2761 if (rc || adapter->fw_done_rc) { 2762 rc = -EIO; 2763 mutex_unlock(&adapter->fw_lock); 2764 goto err; 2765 } 2766 mutex_unlock(&adapter->fw_lock); 2767 return 0; 2768 err: 2769 ether_addr_copy(adapter->mac_addr, netdev->dev_addr); 2770 return rc; 2771 } 2772 2773 static int ibmvnic_set_mac(struct net_device *netdev, void *p) 2774 { 2775 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 2776 struct sockaddr *addr = p; 2777 int rc; 2778 2779 rc = 0; 2780 if (!is_valid_ether_addr(addr->sa_data)) 2781 return -EADDRNOTAVAIL; 2782 2783 ether_addr_copy(adapter->mac_addr, addr->sa_data); 2784 if (adapter->state != VNIC_PROBED) 2785 rc = __ibmvnic_set_mac(netdev, addr->sa_data); 2786 2787 return rc; 2788 } 2789 2790 static const char *reset_reason_to_string(enum ibmvnic_reset_reason reason) 2791 { 2792 switch (reason) { 2793 case VNIC_RESET_FAILOVER: 2794 return "FAILOVER"; 2795 case VNIC_RESET_MOBILITY: 2796 return "MOBILITY"; 2797 case VNIC_RESET_FATAL: 2798 return "FATAL"; 2799 case VNIC_RESET_NON_FATAL: 2800 return "NON_FATAL"; 2801 case VNIC_RESET_TIMEOUT: 2802 return "TIMEOUT"; 2803 case VNIC_RESET_CHANGE_PARAM: 2804 return "CHANGE_PARAM"; 2805 case VNIC_RESET_PASSIVE_INIT: 2806 return "PASSIVE_INIT"; 2807 } 2808 return "UNKNOWN"; 2809 } 2810 2811 /* 2812 * Initialize the init_done completion and return code values. We 2813 * can get a transport event just after registering the CRQ and the 2814 * tasklet will use this to communicate the transport event. To ensure 2815 * we don't miss the notification/error, initialize these _before_ 2816 * regisering the CRQ. 2817 */ 2818 static inline void reinit_init_done(struct ibmvnic_adapter *adapter) 2819 { 2820 reinit_completion(&adapter->init_done); 2821 adapter->init_done_rc = 0; 2822 } 2823 2824 /* 2825 * do_reset returns zero if we are able to keep processing reset events, or 2826 * non-zero if we hit a fatal error and must halt. 2827 */ 2828 static int do_reset(struct ibmvnic_adapter *adapter, 2829 struct ibmvnic_rwi *rwi, u32 reset_state) 2830 { 2831 struct net_device *netdev = adapter->netdev; 2832 u64 old_num_rx_queues, old_num_tx_queues; 2833 u64 old_num_rx_slots, old_num_tx_slots; 2834 int rc; 2835 2836 netdev_dbg(adapter->netdev, 2837 "[S:%s FOP:%d] Reset reason: %s, reset_state: %s\n", 2838 adapter_state_to_string(adapter->state), 2839 adapter->failover_pending, 2840 reset_reason_to_string(rwi->reset_reason), 2841 adapter_state_to_string(reset_state)); 2842 2843 adapter->reset_reason = rwi->reset_reason; 2844 /* requestor of VNIC_RESET_CHANGE_PARAM already has the rtnl lock */ 2845 if (!(adapter->reset_reason == VNIC_RESET_CHANGE_PARAM)) 2846 rtnl_lock(); 2847 2848 /* Now that we have the rtnl lock, clear any pending failover. 2849 * This will ensure ibmvnic_open() has either completed or will 2850 * block until failover is complete. 2851 */ 2852 if (rwi->reset_reason == VNIC_RESET_FAILOVER) 2853 adapter->failover_pending = false; 2854 2855 /* read the state and check (again) after getting rtnl */ 2856 reset_state = adapter->state; 2857 2858 if (reset_state == VNIC_REMOVING || reset_state == VNIC_REMOVED) { 2859 rc = -EBUSY; 2860 goto out; 2861 } 2862 2863 netif_carrier_off(netdev); 2864 2865 old_num_rx_queues = adapter->req_rx_queues; 2866 old_num_tx_queues = adapter->req_tx_queues; 2867 old_num_rx_slots = adapter->req_rx_add_entries_per_subcrq; 2868 old_num_tx_slots = adapter->req_tx_entries_per_subcrq; 2869 2870 ibmvnic_cleanup(netdev); 2871 2872 if (reset_state == VNIC_OPEN && 2873 adapter->reset_reason != VNIC_RESET_MOBILITY && 2874 adapter->reset_reason != VNIC_RESET_FAILOVER) { 2875 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) { 2876 rc = __ibmvnic_close(netdev); 2877 if (rc) 2878 goto out; 2879 } else { 2880 adapter->state = VNIC_CLOSING; 2881 2882 /* Release the RTNL lock before link state change and 2883 * re-acquire after the link state change to allow 2884 * linkwatch_event to grab the RTNL lock and run during 2885 * a reset. 2886 */ 2887 rtnl_unlock(); 2888 rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN); 2889 rtnl_lock(); 2890 if (rc) 2891 goto out; 2892 2893 if (adapter->state == VNIC_OPEN) { 2894 /* When we dropped rtnl, ibmvnic_open() got 2895 * it and noticed that we are resetting and 2896 * set the adapter state to OPEN. Update our 2897 * new "target" state, and resume the reset 2898 * from VNIC_CLOSING state. 2899 */ 2900 netdev_dbg(netdev, 2901 "Open changed state from %s, updating.\n", 2902 adapter_state_to_string(reset_state)); 2903 reset_state = VNIC_OPEN; 2904 adapter->state = VNIC_CLOSING; 2905 } 2906 2907 if (adapter->state != VNIC_CLOSING) { 2908 /* If someone else changed the adapter state 2909 * when we dropped the rtnl, fail the reset 2910 */ 2911 rc = -EAGAIN; 2912 goto out; 2913 } 2914 adapter->state = VNIC_CLOSED; 2915 } 2916 } 2917 2918 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) { 2919 release_resources(adapter); 2920 release_sub_crqs(adapter, 1); 2921 release_crq_queue(adapter); 2922 } 2923 2924 if (adapter->reset_reason != VNIC_RESET_NON_FATAL) { 2925 /* remove the closed state so when we call open it appears 2926 * we are coming from the probed state. 2927 */ 2928 adapter->state = VNIC_PROBED; 2929 2930 reinit_init_done(adapter); 2931 2932 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) { 2933 rc = init_crq_queue(adapter); 2934 } else if (adapter->reset_reason == VNIC_RESET_MOBILITY) { 2935 rc = ibmvnic_reenable_crq_queue(adapter); 2936 release_sub_crqs(adapter, 1); 2937 } else { 2938 rc = ibmvnic_reset_crq(adapter); 2939 if (rc == H_CLOSED || rc == H_SUCCESS) { 2940 rc = vio_enable_interrupts(adapter->vdev); 2941 if (rc) 2942 netdev_err(adapter->netdev, 2943 "Reset failed to enable interrupts. rc=%d\n", 2944 rc); 2945 } 2946 } 2947 2948 if (rc) { 2949 netdev_err(adapter->netdev, 2950 "Reset couldn't initialize crq. rc=%d\n", rc); 2951 goto out; 2952 } 2953 2954 rc = ibmvnic_reset_init(adapter, true); 2955 if (rc) 2956 goto out; 2957 2958 /* If the adapter was in PROBE or DOWN state prior to the reset, 2959 * exit here. 2960 */ 2961 if (reset_state == VNIC_PROBED || reset_state == VNIC_DOWN) { 2962 rc = 0; 2963 goto out; 2964 } 2965 2966 rc = ibmvnic_login(netdev); 2967 if (rc) 2968 goto out; 2969 2970 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) { 2971 rc = init_resources(adapter); 2972 if (rc) 2973 goto out; 2974 } else if (adapter->req_rx_queues != old_num_rx_queues || 2975 adapter->req_tx_queues != old_num_tx_queues || 2976 adapter->req_rx_add_entries_per_subcrq != 2977 old_num_rx_slots || 2978 adapter->req_tx_entries_per_subcrq != 2979 old_num_tx_slots || 2980 !adapter->rx_pool || 2981 !adapter->tso_pool || 2982 !adapter->tx_pool) { 2983 release_napi(adapter); 2984 release_vpd_data(adapter); 2985 2986 rc = init_resources(adapter); 2987 if (rc) 2988 goto out; 2989 2990 } else { 2991 rc = init_tx_pools(netdev); 2992 if (rc) { 2993 netdev_dbg(netdev, 2994 "init tx pools failed (%d)\n", 2995 rc); 2996 goto out; 2997 } 2998 2999 rc = init_rx_pools(netdev); 3000 if (rc) { 3001 netdev_dbg(netdev, 3002 "init rx pools failed (%d)\n", 3003 rc); 3004 goto out; 3005 } 3006 } 3007 ibmvnic_disable_irqs(adapter); 3008 } 3009 adapter->state = VNIC_CLOSED; 3010 3011 if (reset_state == VNIC_CLOSED) { 3012 rc = 0; 3013 goto out; 3014 } 3015 3016 rc = __ibmvnic_open(netdev); 3017 if (rc) { 3018 rc = IBMVNIC_OPEN_FAILED; 3019 goto out; 3020 } 3021 3022 /* refresh device's multicast list */ 3023 ibmvnic_set_multi(netdev); 3024 3025 if (adapter->reset_reason == VNIC_RESET_FAILOVER || 3026 adapter->reset_reason == VNIC_RESET_MOBILITY) 3027 __netdev_notify_peers(netdev); 3028 3029 rc = 0; 3030 3031 out: 3032 /* restore the adapter state if reset failed */ 3033 if (rc) 3034 adapter->state = reset_state; 3035 /* requestor of VNIC_RESET_CHANGE_PARAM should still hold the rtnl lock */ 3036 if (!(adapter->reset_reason == VNIC_RESET_CHANGE_PARAM)) 3037 rtnl_unlock(); 3038 3039 netdev_dbg(adapter->netdev, "[S:%s FOP:%d] Reset done, rc %d\n", 3040 adapter_state_to_string(adapter->state), 3041 adapter->failover_pending, rc); 3042 return rc; 3043 } 3044 3045 static int do_hard_reset(struct ibmvnic_adapter *adapter, 3046 struct ibmvnic_rwi *rwi, u32 reset_state) 3047 { 3048 struct net_device *netdev = adapter->netdev; 3049 int rc; 3050 3051 netdev_dbg(adapter->netdev, "Hard resetting driver (%s)\n", 3052 reset_reason_to_string(rwi->reset_reason)); 3053 3054 /* read the state and check (again) after getting rtnl */ 3055 reset_state = adapter->state; 3056 3057 if (reset_state == VNIC_REMOVING || reset_state == VNIC_REMOVED) { 3058 rc = -EBUSY; 3059 goto out; 3060 } 3061 3062 netif_carrier_off(netdev); 3063 adapter->reset_reason = rwi->reset_reason; 3064 3065 ibmvnic_cleanup(netdev); 3066 release_resources(adapter); 3067 release_sub_crqs(adapter, 0); 3068 release_crq_queue(adapter); 3069 3070 /* remove the closed state so when we call open it appears 3071 * we are coming from the probed state. 3072 */ 3073 adapter->state = VNIC_PROBED; 3074 3075 reinit_init_done(adapter); 3076 3077 rc = init_crq_queue(adapter); 3078 if (rc) { 3079 netdev_err(adapter->netdev, 3080 "Couldn't initialize crq. rc=%d\n", rc); 3081 goto out; 3082 } 3083 3084 rc = ibmvnic_reset_init(adapter, false); 3085 if (rc) 3086 goto out; 3087 3088 /* If the adapter was in PROBE or DOWN state prior to the reset, 3089 * exit here. 3090 */ 3091 if (reset_state == VNIC_PROBED || reset_state == VNIC_DOWN) 3092 goto out; 3093 3094 rc = ibmvnic_login(netdev); 3095 if (rc) 3096 goto out; 3097 3098 rc = init_resources(adapter); 3099 if (rc) 3100 goto out; 3101 3102 ibmvnic_disable_irqs(adapter); 3103 adapter->state = VNIC_CLOSED; 3104 3105 if (reset_state == VNIC_CLOSED) 3106 goto out; 3107 3108 rc = __ibmvnic_open(netdev); 3109 if (rc) { 3110 rc = IBMVNIC_OPEN_FAILED; 3111 goto out; 3112 } 3113 3114 __netdev_notify_peers(netdev); 3115 out: 3116 /* restore adapter state if reset failed */ 3117 if (rc) 3118 adapter->state = reset_state; 3119 netdev_dbg(adapter->netdev, "[S:%s FOP:%d] Hard reset done, rc %d\n", 3120 adapter_state_to_string(adapter->state), 3121 adapter->failover_pending, rc); 3122 return rc; 3123 } 3124 3125 static struct ibmvnic_rwi *get_next_rwi(struct ibmvnic_adapter *adapter) 3126 { 3127 struct ibmvnic_rwi *rwi; 3128 unsigned long flags; 3129 3130 spin_lock_irqsave(&adapter->rwi_lock, flags); 3131 3132 if (!list_empty(&adapter->rwi_list)) { 3133 rwi = list_first_entry(&adapter->rwi_list, struct ibmvnic_rwi, 3134 list); 3135 list_del(&rwi->list); 3136 } else { 3137 rwi = NULL; 3138 } 3139 3140 spin_unlock_irqrestore(&adapter->rwi_lock, flags); 3141 return rwi; 3142 } 3143 3144 /** 3145 * do_passive_init - complete probing when partner device is detected. 3146 * @adapter: ibmvnic_adapter struct 3147 * 3148 * If the ibmvnic device does not have a partner device to communicate with at boot 3149 * and that partner device comes online at a later time, this function is called 3150 * to complete the initialization process of ibmvnic device. 3151 * Caller is expected to hold rtnl_lock(). 3152 * 3153 * Returns non-zero if sub-CRQs are not initialized properly leaving the device 3154 * in the down state. 3155 * Returns 0 upon success and the device is in PROBED state. 3156 */ 3157 3158 static int do_passive_init(struct ibmvnic_adapter *adapter) 3159 { 3160 unsigned long timeout = msecs_to_jiffies(30000); 3161 struct net_device *netdev = adapter->netdev; 3162 struct device *dev = &adapter->vdev->dev; 3163 int rc; 3164 3165 netdev_dbg(netdev, "Partner device found, probing.\n"); 3166 3167 adapter->state = VNIC_PROBING; 3168 reinit_completion(&adapter->init_done); 3169 adapter->init_done_rc = 0; 3170 adapter->crq.active = true; 3171 3172 rc = send_crq_init_complete(adapter); 3173 if (rc) 3174 goto out; 3175 3176 rc = send_version_xchg(adapter); 3177 if (rc) 3178 netdev_dbg(adapter->netdev, "send_version_xchg failed, rc=%d\n", rc); 3179 3180 if (!wait_for_completion_timeout(&adapter->init_done, timeout)) { 3181 dev_err(dev, "Initialization sequence timed out\n"); 3182 rc = -ETIMEDOUT; 3183 goto out; 3184 } 3185 3186 rc = init_sub_crqs(adapter); 3187 if (rc) { 3188 dev_err(dev, "Initialization of sub crqs failed, rc=%d\n", rc); 3189 goto out; 3190 } 3191 3192 rc = init_sub_crq_irqs(adapter); 3193 if (rc) { 3194 dev_err(dev, "Failed to initialize sub crq irqs\n, rc=%d", rc); 3195 goto init_failed; 3196 } 3197 3198 netdev->mtu = adapter->req_mtu - ETH_HLEN; 3199 netdev->min_mtu = adapter->min_mtu - ETH_HLEN; 3200 netdev->max_mtu = adapter->max_mtu - ETH_HLEN; 3201 3202 adapter->state = VNIC_PROBED; 3203 netdev_dbg(netdev, "Probed successfully. Waiting for signal from partner device.\n"); 3204 3205 return 0; 3206 3207 init_failed: 3208 release_sub_crqs(adapter, 1); 3209 out: 3210 adapter->state = VNIC_DOWN; 3211 return rc; 3212 } 3213 3214 static void __ibmvnic_reset(struct work_struct *work) 3215 { 3216 struct ibmvnic_adapter *adapter; 3217 unsigned int timeout = 5000; 3218 struct ibmvnic_rwi *tmprwi; 3219 bool saved_state = false; 3220 struct ibmvnic_rwi *rwi; 3221 unsigned long flags; 3222 struct device *dev; 3223 bool need_reset; 3224 int num_fails = 0; 3225 u32 reset_state; 3226 int rc = 0; 3227 3228 adapter = container_of(work, struct ibmvnic_adapter, ibmvnic_reset); 3229 dev = &adapter->vdev->dev; 3230 3231 /* Wait for ibmvnic_probe() to complete. If probe is taking too long 3232 * or if another reset is in progress, defer work for now. If probe 3233 * eventually fails it will flush and terminate our work. 3234 * 3235 * Three possibilities here: 3236 * 1. Adpater being removed - just return 3237 * 2. Timed out on probe or another reset in progress - delay the work 3238 * 3. Completed probe - perform any resets in queue 3239 */ 3240 if (adapter->state == VNIC_PROBING && 3241 !wait_for_completion_timeout(&adapter->probe_done, timeout)) { 3242 dev_err(dev, "Reset thread timed out on probe"); 3243 queue_delayed_work(system_long_wq, 3244 &adapter->ibmvnic_delayed_reset, 3245 IBMVNIC_RESET_DELAY); 3246 return; 3247 } 3248 3249 /* adapter is done with probe (i.e state is never VNIC_PROBING now) */ 3250 if (adapter->state == VNIC_REMOVING) 3251 return; 3252 3253 /* ->rwi_list is stable now (no one else is removing entries) */ 3254 3255 /* ibmvnic_probe() may have purged the reset queue after we were 3256 * scheduled to process a reset so there maybe no resets to process. 3257 * Before setting the ->resetting bit though, we have to make sure 3258 * that there is infact a reset to process. Otherwise we may race 3259 * with ibmvnic_open() and end up leaving the vnic down: 3260 * 3261 * __ibmvnic_reset() ibmvnic_open() 3262 * ----------------- -------------- 3263 * 3264 * set ->resetting bit 3265 * find ->resetting bit is set 3266 * set ->state to IBMVNIC_OPEN (i.e 3267 * assume reset will open device) 3268 * return 3269 * find reset queue empty 3270 * return 3271 * 3272 * Neither performed vnic login/open and vnic stays down 3273 * 3274 * If we hold the lock and conditionally set the bit, either we 3275 * or ibmvnic_open() will complete the open. 3276 */ 3277 need_reset = false; 3278 spin_lock(&adapter->rwi_lock); 3279 if (!list_empty(&adapter->rwi_list)) { 3280 if (test_and_set_bit_lock(0, &adapter->resetting)) { 3281 queue_delayed_work(system_long_wq, 3282 &adapter->ibmvnic_delayed_reset, 3283 IBMVNIC_RESET_DELAY); 3284 } else { 3285 need_reset = true; 3286 } 3287 } 3288 spin_unlock(&adapter->rwi_lock); 3289 3290 if (!need_reset) 3291 return; 3292 3293 rwi = get_next_rwi(adapter); 3294 while (rwi) { 3295 spin_lock_irqsave(&adapter->state_lock, flags); 3296 3297 if (adapter->state == VNIC_REMOVING || 3298 adapter->state == VNIC_REMOVED) { 3299 spin_unlock_irqrestore(&adapter->state_lock, flags); 3300 kfree(rwi); 3301 rc = EBUSY; 3302 break; 3303 } 3304 3305 if (!saved_state) { 3306 reset_state = adapter->state; 3307 saved_state = true; 3308 } 3309 spin_unlock_irqrestore(&adapter->state_lock, flags); 3310 3311 if (rwi->reset_reason == VNIC_RESET_PASSIVE_INIT) { 3312 rtnl_lock(); 3313 rc = do_passive_init(adapter); 3314 rtnl_unlock(); 3315 if (!rc) 3316 netif_carrier_on(adapter->netdev); 3317 } else if (adapter->force_reset_recovery) { 3318 /* Since we are doing a hard reset now, clear the 3319 * failover_pending flag so we don't ignore any 3320 * future MOBILITY or other resets. 3321 */ 3322 adapter->failover_pending = false; 3323 3324 /* Transport event occurred during previous reset */ 3325 if (adapter->wait_for_reset) { 3326 /* Previous was CHANGE_PARAM; caller locked */ 3327 adapter->force_reset_recovery = false; 3328 rc = do_hard_reset(adapter, rwi, reset_state); 3329 } else { 3330 rtnl_lock(); 3331 adapter->force_reset_recovery = false; 3332 rc = do_hard_reset(adapter, rwi, reset_state); 3333 rtnl_unlock(); 3334 } 3335 if (rc) 3336 num_fails++; 3337 else 3338 num_fails = 0; 3339 3340 /* If auto-priority-failover is enabled we can get 3341 * back to back failovers during resets, resulting 3342 * in at least two failed resets (from high-priority 3343 * backing device to low-priority one and then back) 3344 * If resets continue to fail beyond that, give the 3345 * adapter some time to settle down before retrying. 3346 */ 3347 if (num_fails >= 3) { 3348 netdev_dbg(adapter->netdev, 3349 "[S:%s] Hard reset failed %d times, waiting 60 secs\n", 3350 adapter_state_to_string(adapter->state), 3351 num_fails); 3352 set_current_state(TASK_UNINTERRUPTIBLE); 3353 schedule_timeout(60 * HZ); 3354 } 3355 } else { 3356 rc = do_reset(adapter, rwi, reset_state); 3357 } 3358 tmprwi = rwi; 3359 adapter->last_reset_time = jiffies; 3360 3361 if (rc) 3362 netdev_dbg(adapter->netdev, "Reset failed, rc=%d\n", rc); 3363 3364 rwi = get_next_rwi(adapter); 3365 3366 /* 3367 * If there are no resets queued and the previous reset failed, 3368 * the adapter would be in an undefined state. So retry the 3369 * previous reset as a hard reset. 3370 * 3371 * Else, free the previous rwi and, if there is another reset 3372 * queued, process the new reset even if previous reset failed 3373 * (the previous reset could have failed because of a fail 3374 * over for instance, so process the fail over). 3375 */ 3376 if (!rwi && rc) 3377 rwi = tmprwi; 3378 else 3379 kfree(tmprwi); 3380 3381 if (rwi && (rwi->reset_reason == VNIC_RESET_FAILOVER || 3382 rwi->reset_reason == VNIC_RESET_MOBILITY || rc)) 3383 adapter->force_reset_recovery = true; 3384 } 3385 3386 if (adapter->wait_for_reset) { 3387 adapter->reset_done_rc = rc; 3388 complete(&adapter->reset_done); 3389 } 3390 3391 clear_bit_unlock(0, &adapter->resetting); 3392 3393 netdev_dbg(adapter->netdev, 3394 "[S:%s FRR:%d WFR:%d] Done processing resets\n", 3395 adapter_state_to_string(adapter->state), 3396 adapter->force_reset_recovery, 3397 adapter->wait_for_reset); 3398 } 3399 3400 static void __ibmvnic_delayed_reset(struct work_struct *work) 3401 { 3402 struct ibmvnic_adapter *adapter; 3403 3404 adapter = container_of(work, struct ibmvnic_adapter, 3405 ibmvnic_delayed_reset.work); 3406 __ibmvnic_reset(&adapter->ibmvnic_reset); 3407 } 3408 3409 static void flush_reset_queue(struct ibmvnic_adapter *adapter) 3410 { 3411 struct list_head *entry, *tmp_entry; 3412 3413 if (!list_empty(&adapter->rwi_list)) { 3414 list_for_each_safe(entry, tmp_entry, &adapter->rwi_list) { 3415 list_del(entry); 3416 kfree(list_entry(entry, struct ibmvnic_rwi, list)); 3417 } 3418 } 3419 } 3420 3421 static int ibmvnic_reset(struct ibmvnic_adapter *adapter, 3422 enum ibmvnic_reset_reason reason) 3423 { 3424 struct net_device *netdev = adapter->netdev; 3425 struct ibmvnic_rwi *rwi, *tmp; 3426 unsigned long flags; 3427 int ret; 3428 3429 spin_lock_irqsave(&adapter->rwi_lock, flags); 3430 3431 /* If failover is pending don't schedule any other reset. 3432 * Instead let the failover complete. If there is already a 3433 * a failover reset scheduled, we will detect and drop the 3434 * duplicate reset when walking the ->rwi_list below. 3435 */ 3436 if (adapter->state == VNIC_REMOVING || 3437 adapter->state == VNIC_REMOVED || 3438 (adapter->failover_pending && reason != VNIC_RESET_FAILOVER)) { 3439 ret = EBUSY; 3440 netdev_dbg(netdev, "Adapter removing or pending failover, skipping reset\n"); 3441 goto err; 3442 } 3443 3444 list_for_each_entry(tmp, &adapter->rwi_list, list) { 3445 if (tmp->reset_reason == reason) { 3446 netdev_dbg(netdev, "Skipping matching reset, reason=%s\n", 3447 reset_reason_to_string(reason)); 3448 ret = EBUSY; 3449 goto err; 3450 } 3451 } 3452 3453 rwi = kzalloc(sizeof(*rwi), GFP_ATOMIC); 3454 if (!rwi) { 3455 ret = ENOMEM; 3456 goto err; 3457 } 3458 /* if we just received a transport event, 3459 * flush reset queue and process this reset 3460 */ 3461 if (adapter->force_reset_recovery) 3462 flush_reset_queue(adapter); 3463 3464 rwi->reset_reason = reason; 3465 list_add_tail(&rwi->list, &adapter->rwi_list); 3466 netdev_dbg(adapter->netdev, "Scheduling reset (reason %s)\n", 3467 reset_reason_to_string(reason)); 3468 queue_work(system_long_wq, &adapter->ibmvnic_reset); 3469 3470 ret = 0; 3471 err: 3472 /* ibmvnic_close() below can block, so drop the lock first */ 3473 spin_unlock_irqrestore(&adapter->rwi_lock, flags); 3474 3475 if (ret == ENOMEM) 3476 ibmvnic_close(netdev); 3477 3478 return -ret; 3479 } 3480 3481 static void ibmvnic_get_stats64(struct net_device *netdev, 3482 struct rtnl_link_stats64 *stats) 3483 { 3484 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3485 int i; 3486 3487 for (i = 0; i < adapter->req_rx_queues; i++) { 3488 stats->rx_packets += adapter->rx_stats_buffers[i].packets; 3489 stats->rx_bytes += adapter->rx_stats_buffers[i].bytes; 3490 } 3491 3492 for (i = 0; i < adapter->req_tx_queues; i++) { 3493 stats->tx_packets += adapter->tx_stats_buffers[i].batched_packets; 3494 stats->tx_packets += adapter->tx_stats_buffers[i].direct_packets; 3495 stats->tx_bytes += adapter->tx_stats_buffers[i].bytes; 3496 stats->tx_dropped += adapter->tx_stats_buffers[i].dropped_packets; 3497 } 3498 } 3499 3500 static void ibmvnic_tx_timeout(struct net_device *dev, unsigned int txqueue) 3501 { 3502 struct ibmvnic_adapter *adapter = netdev_priv(dev); 3503 3504 if (test_bit(0, &adapter->resetting)) { 3505 netdev_err(adapter->netdev, 3506 "Adapter is resetting, skip timeout reset\n"); 3507 return; 3508 } 3509 /* No queuing up reset until at least 5 seconds (default watchdog val) 3510 * after last reset 3511 */ 3512 if (time_before(jiffies, (adapter->last_reset_time + dev->watchdog_timeo))) { 3513 netdev_dbg(dev, "Not yet time to tx timeout.\n"); 3514 return; 3515 } 3516 ibmvnic_reset(adapter, VNIC_RESET_TIMEOUT); 3517 } 3518 3519 static void remove_buff_from_pool(struct ibmvnic_adapter *adapter, 3520 struct ibmvnic_rx_buff *rx_buff) 3521 { 3522 struct ibmvnic_rx_pool *pool = &adapter->rx_pool[rx_buff->pool_index]; 3523 3524 rx_buff->skb = NULL; 3525 3526 pool->free_map[pool->next_alloc] = (int)(rx_buff - pool->rx_buff); 3527 pool->next_alloc = (pool->next_alloc + 1) % pool->size; 3528 3529 atomic_dec(&pool->available); 3530 } 3531 3532 static int ibmvnic_poll(struct napi_struct *napi, int budget) 3533 { 3534 struct ibmvnic_sub_crq_queue *rx_scrq; 3535 struct ibmvnic_adapter *adapter; 3536 struct net_device *netdev; 3537 int frames_processed; 3538 int scrq_num; 3539 3540 netdev = napi->dev; 3541 adapter = netdev_priv(netdev); 3542 scrq_num = (int)(napi - adapter->napi); 3543 frames_processed = 0; 3544 rx_scrq = adapter->rx_scrq[scrq_num]; 3545 3546 restart_poll: 3547 while (frames_processed < budget) { 3548 struct sk_buff *skb; 3549 struct ibmvnic_rx_buff *rx_buff; 3550 union sub_crq *next; 3551 u32 length; 3552 u16 offset; 3553 u8 flags = 0; 3554 3555 if (unlikely(test_bit(0, &adapter->resetting) && 3556 adapter->reset_reason != VNIC_RESET_NON_FATAL)) { 3557 enable_scrq_irq(adapter, rx_scrq); 3558 napi_complete_done(napi, frames_processed); 3559 return frames_processed; 3560 } 3561 3562 if (!pending_scrq(adapter, rx_scrq)) 3563 break; 3564 next = ibmvnic_next_scrq(adapter, rx_scrq); 3565 rx_buff = (struct ibmvnic_rx_buff *) 3566 be64_to_cpu(next->rx_comp.correlator); 3567 /* do error checking */ 3568 if (next->rx_comp.rc) { 3569 netdev_dbg(netdev, "rx buffer returned with rc %x\n", 3570 be16_to_cpu(next->rx_comp.rc)); 3571 /* free the entry */ 3572 next->rx_comp.first = 0; 3573 dev_kfree_skb_any(rx_buff->skb); 3574 remove_buff_from_pool(adapter, rx_buff); 3575 continue; 3576 } else if (!rx_buff->skb) { 3577 /* free the entry */ 3578 next->rx_comp.first = 0; 3579 remove_buff_from_pool(adapter, rx_buff); 3580 continue; 3581 } 3582 3583 length = be32_to_cpu(next->rx_comp.len); 3584 offset = be16_to_cpu(next->rx_comp.off_frame_data); 3585 flags = next->rx_comp.flags; 3586 skb = rx_buff->skb; 3587 /* load long_term_buff before copying to skb */ 3588 dma_rmb(); 3589 skb_copy_to_linear_data(skb, rx_buff->data + offset, 3590 length); 3591 3592 /* VLAN Header has been stripped by the system firmware and 3593 * needs to be inserted by the driver 3594 */ 3595 if (adapter->rx_vlan_header_insertion && 3596 (flags & IBMVNIC_VLAN_STRIPPED)) 3597 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), 3598 ntohs(next->rx_comp.vlan_tci)); 3599 3600 /* free the entry */ 3601 next->rx_comp.first = 0; 3602 remove_buff_from_pool(adapter, rx_buff); 3603 3604 skb_put(skb, length); 3605 skb->protocol = eth_type_trans(skb, netdev); 3606 skb_record_rx_queue(skb, scrq_num); 3607 3608 if (flags & IBMVNIC_IP_CHKSUM_GOOD && 3609 flags & IBMVNIC_TCP_UDP_CHKSUM_GOOD) { 3610 skb->ip_summed = CHECKSUM_UNNECESSARY; 3611 } 3612 3613 length = skb->len; 3614 napi_gro_receive(napi, skb); /* send it up */ 3615 adapter->rx_stats_buffers[scrq_num].packets++; 3616 adapter->rx_stats_buffers[scrq_num].bytes += length; 3617 frames_processed++; 3618 } 3619 3620 if (adapter->state != VNIC_CLOSING && 3621 (atomic_read(&adapter->rx_pool[scrq_num].available) < 3622 adapter->req_rx_add_entries_per_subcrq / 2)) 3623 replenish_rx_pool(adapter, &adapter->rx_pool[scrq_num]); 3624 if (frames_processed < budget) { 3625 if (napi_complete_done(napi, frames_processed)) { 3626 enable_scrq_irq(adapter, rx_scrq); 3627 if (pending_scrq(adapter, rx_scrq)) { 3628 if (napi_schedule(napi)) { 3629 disable_scrq_irq(adapter, rx_scrq); 3630 goto restart_poll; 3631 } 3632 } 3633 } 3634 } 3635 return frames_processed; 3636 } 3637 3638 static int wait_for_reset(struct ibmvnic_adapter *adapter) 3639 { 3640 int rc, ret; 3641 3642 adapter->fallback.mtu = adapter->req_mtu; 3643 adapter->fallback.rx_queues = adapter->req_rx_queues; 3644 adapter->fallback.tx_queues = adapter->req_tx_queues; 3645 adapter->fallback.rx_entries = adapter->req_rx_add_entries_per_subcrq; 3646 adapter->fallback.tx_entries = adapter->req_tx_entries_per_subcrq; 3647 3648 reinit_completion(&adapter->reset_done); 3649 adapter->wait_for_reset = true; 3650 rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM); 3651 3652 if (rc) { 3653 ret = rc; 3654 goto out; 3655 } 3656 rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done, 60000); 3657 if (rc) { 3658 ret = -ENODEV; 3659 goto out; 3660 } 3661 3662 ret = 0; 3663 if (adapter->reset_done_rc) { 3664 ret = -EIO; 3665 adapter->desired.mtu = adapter->fallback.mtu; 3666 adapter->desired.rx_queues = adapter->fallback.rx_queues; 3667 adapter->desired.tx_queues = adapter->fallback.tx_queues; 3668 adapter->desired.rx_entries = adapter->fallback.rx_entries; 3669 adapter->desired.tx_entries = adapter->fallback.tx_entries; 3670 3671 reinit_completion(&adapter->reset_done); 3672 adapter->wait_for_reset = true; 3673 rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM); 3674 if (rc) { 3675 ret = rc; 3676 goto out; 3677 } 3678 rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done, 3679 60000); 3680 if (rc) { 3681 ret = -ENODEV; 3682 goto out; 3683 } 3684 } 3685 out: 3686 adapter->wait_for_reset = false; 3687 3688 return ret; 3689 } 3690 3691 static int ibmvnic_change_mtu(struct net_device *netdev, int new_mtu) 3692 { 3693 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3694 3695 adapter->desired.mtu = new_mtu + ETH_HLEN; 3696 3697 return wait_for_reset(adapter); 3698 } 3699 3700 static netdev_features_t ibmvnic_features_check(struct sk_buff *skb, 3701 struct net_device *dev, 3702 netdev_features_t features) 3703 { 3704 /* Some backing hardware adapters can not 3705 * handle packets with a MSS less than 224 3706 * or with only one segment. 3707 */ 3708 if (skb_is_gso(skb)) { 3709 if (skb_shinfo(skb)->gso_size < 224 || 3710 skb_shinfo(skb)->gso_segs == 1) 3711 features &= ~NETIF_F_GSO_MASK; 3712 } 3713 3714 return features; 3715 } 3716 3717 static const struct net_device_ops ibmvnic_netdev_ops = { 3718 .ndo_open = ibmvnic_open, 3719 .ndo_stop = ibmvnic_close, 3720 .ndo_start_xmit = ibmvnic_xmit, 3721 .ndo_set_rx_mode = ibmvnic_set_multi, 3722 .ndo_set_mac_address = ibmvnic_set_mac, 3723 .ndo_validate_addr = eth_validate_addr, 3724 .ndo_get_stats64 = ibmvnic_get_stats64, 3725 .ndo_tx_timeout = ibmvnic_tx_timeout, 3726 .ndo_change_mtu = ibmvnic_change_mtu, 3727 .ndo_features_check = ibmvnic_features_check, 3728 }; 3729 3730 /* ethtool functions */ 3731 3732 static int ibmvnic_get_link_ksettings(struct net_device *netdev, 3733 struct ethtool_link_ksettings *cmd) 3734 { 3735 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3736 int rc; 3737 3738 rc = send_query_phys_parms(adapter); 3739 if (rc) { 3740 adapter->speed = SPEED_UNKNOWN; 3741 adapter->duplex = DUPLEX_UNKNOWN; 3742 } 3743 cmd->base.speed = adapter->speed; 3744 cmd->base.duplex = adapter->duplex; 3745 cmd->base.port = PORT_FIBRE; 3746 cmd->base.phy_address = 0; 3747 cmd->base.autoneg = AUTONEG_ENABLE; 3748 3749 return 0; 3750 } 3751 3752 static void ibmvnic_get_drvinfo(struct net_device *netdev, 3753 struct ethtool_drvinfo *info) 3754 { 3755 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3756 3757 strscpy(info->driver, ibmvnic_driver_name, sizeof(info->driver)); 3758 strscpy(info->version, IBMVNIC_DRIVER_VERSION, sizeof(info->version)); 3759 strscpy(info->fw_version, adapter->fw_version, 3760 sizeof(info->fw_version)); 3761 } 3762 3763 static u32 ibmvnic_get_msglevel(struct net_device *netdev) 3764 { 3765 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3766 3767 return adapter->msg_enable; 3768 } 3769 3770 static void ibmvnic_set_msglevel(struct net_device *netdev, u32 data) 3771 { 3772 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3773 3774 adapter->msg_enable = data; 3775 } 3776 3777 static u32 ibmvnic_get_link(struct net_device *netdev) 3778 { 3779 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3780 3781 /* Don't need to send a query because we request a logical link up at 3782 * init and then we wait for link state indications 3783 */ 3784 return adapter->logical_link_state; 3785 } 3786 3787 static void ibmvnic_get_ringparam(struct net_device *netdev, 3788 struct ethtool_ringparam *ring, 3789 struct kernel_ethtool_ringparam *kernel_ring, 3790 struct netlink_ext_ack *extack) 3791 { 3792 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3793 3794 ring->rx_max_pending = adapter->max_rx_add_entries_per_subcrq; 3795 ring->tx_max_pending = adapter->max_tx_entries_per_subcrq; 3796 ring->rx_mini_max_pending = 0; 3797 ring->rx_jumbo_max_pending = 0; 3798 ring->rx_pending = adapter->req_rx_add_entries_per_subcrq; 3799 ring->tx_pending = adapter->req_tx_entries_per_subcrq; 3800 ring->rx_mini_pending = 0; 3801 ring->rx_jumbo_pending = 0; 3802 } 3803 3804 static int ibmvnic_set_ringparam(struct net_device *netdev, 3805 struct ethtool_ringparam *ring, 3806 struct kernel_ethtool_ringparam *kernel_ring, 3807 struct netlink_ext_ack *extack) 3808 { 3809 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3810 3811 if (ring->rx_pending > adapter->max_rx_add_entries_per_subcrq || 3812 ring->tx_pending > adapter->max_tx_entries_per_subcrq) { 3813 netdev_err(netdev, "Invalid request.\n"); 3814 netdev_err(netdev, "Max tx buffers = %llu\n", 3815 adapter->max_rx_add_entries_per_subcrq); 3816 netdev_err(netdev, "Max rx buffers = %llu\n", 3817 adapter->max_tx_entries_per_subcrq); 3818 return -EINVAL; 3819 } 3820 3821 adapter->desired.rx_entries = ring->rx_pending; 3822 adapter->desired.tx_entries = ring->tx_pending; 3823 3824 return wait_for_reset(adapter); 3825 } 3826 3827 static void ibmvnic_get_channels(struct net_device *netdev, 3828 struct ethtool_channels *channels) 3829 { 3830 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3831 3832 channels->max_rx = adapter->max_rx_queues; 3833 channels->max_tx = adapter->max_tx_queues; 3834 channels->max_other = 0; 3835 channels->max_combined = 0; 3836 channels->rx_count = adapter->req_rx_queues; 3837 channels->tx_count = adapter->req_tx_queues; 3838 channels->other_count = 0; 3839 channels->combined_count = 0; 3840 } 3841 3842 static int ibmvnic_set_channels(struct net_device *netdev, 3843 struct ethtool_channels *channels) 3844 { 3845 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3846 3847 adapter->desired.rx_queues = channels->rx_count; 3848 adapter->desired.tx_queues = channels->tx_count; 3849 3850 return wait_for_reset(adapter); 3851 } 3852 3853 static void ibmvnic_get_strings(struct net_device *dev, u32 stringset, u8 *data) 3854 { 3855 struct ibmvnic_adapter *adapter = netdev_priv(dev); 3856 int i; 3857 3858 if (stringset != ETH_SS_STATS) 3859 return; 3860 3861 for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); i++) 3862 ethtool_puts(&data, ibmvnic_stats[i].name); 3863 3864 for (i = 0; i < adapter->req_tx_queues; i++) { 3865 ethtool_sprintf(&data, "tx%d_batched_packets", i); 3866 ethtool_sprintf(&data, "tx%d_direct_packets", i); 3867 ethtool_sprintf(&data, "tx%d_bytes", i); 3868 ethtool_sprintf(&data, "tx%d_dropped_packets", i); 3869 } 3870 3871 for (i = 0; i < adapter->req_rx_queues; i++) { 3872 ethtool_sprintf(&data, "rx%d_packets", i); 3873 ethtool_sprintf(&data, "rx%d_bytes", i); 3874 ethtool_sprintf(&data, "rx%d_interrupts", i); 3875 } 3876 } 3877 3878 static int ibmvnic_get_sset_count(struct net_device *dev, int sset) 3879 { 3880 struct ibmvnic_adapter *adapter = netdev_priv(dev); 3881 3882 switch (sset) { 3883 case ETH_SS_STATS: 3884 return ARRAY_SIZE(ibmvnic_stats) + 3885 adapter->req_tx_queues * NUM_TX_STATS + 3886 adapter->req_rx_queues * NUM_RX_STATS; 3887 default: 3888 return -EOPNOTSUPP; 3889 } 3890 } 3891 3892 static void ibmvnic_get_ethtool_stats(struct net_device *dev, 3893 struct ethtool_stats *stats, u64 *data) 3894 { 3895 struct ibmvnic_adapter *adapter = netdev_priv(dev); 3896 union ibmvnic_crq crq; 3897 int i, j; 3898 int rc; 3899 3900 memset(&crq, 0, sizeof(crq)); 3901 crq.request_statistics.first = IBMVNIC_CRQ_CMD; 3902 crq.request_statistics.cmd = REQUEST_STATISTICS; 3903 crq.request_statistics.ioba = cpu_to_be32(adapter->stats_token); 3904 crq.request_statistics.len = 3905 cpu_to_be32(sizeof(struct ibmvnic_statistics)); 3906 3907 /* Wait for data to be written */ 3908 reinit_completion(&adapter->stats_done); 3909 rc = ibmvnic_send_crq(adapter, &crq); 3910 if (rc) 3911 return; 3912 rc = ibmvnic_wait_for_completion(adapter, &adapter->stats_done, 10000); 3913 if (rc) 3914 return; 3915 3916 for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); i++) 3917 data[i] = be64_to_cpu(IBMVNIC_GET_STAT 3918 (adapter, ibmvnic_stats[i].offset)); 3919 3920 for (j = 0; j < adapter->req_tx_queues; j++) { 3921 data[i] = adapter->tx_stats_buffers[j].batched_packets; 3922 i++; 3923 data[i] = adapter->tx_stats_buffers[j].direct_packets; 3924 i++; 3925 data[i] = adapter->tx_stats_buffers[j].bytes; 3926 i++; 3927 data[i] = adapter->tx_stats_buffers[j].dropped_packets; 3928 i++; 3929 } 3930 3931 for (j = 0; j < adapter->req_rx_queues; j++) { 3932 data[i] = adapter->rx_stats_buffers[j].packets; 3933 i++; 3934 data[i] = adapter->rx_stats_buffers[j].bytes; 3935 i++; 3936 data[i] = adapter->rx_stats_buffers[j].interrupts; 3937 i++; 3938 } 3939 } 3940 3941 static const struct ethtool_ops ibmvnic_ethtool_ops = { 3942 .get_drvinfo = ibmvnic_get_drvinfo, 3943 .get_msglevel = ibmvnic_get_msglevel, 3944 .set_msglevel = ibmvnic_set_msglevel, 3945 .get_link = ibmvnic_get_link, 3946 .get_ringparam = ibmvnic_get_ringparam, 3947 .set_ringparam = ibmvnic_set_ringparam, 3948 .get_channels = ibmvnic_get_channels, 3949 .set_channels = ibmvnic_set_channels, 3950 .get_strings = ibmvnic_get_strings, 3951 .get_sset_count = ibmvnic_get_sset_count, 3952 .get_ethtool_stats = ibmvnic_get_ethtool_stats, 3953 .get_link_ksettings = ibmvnic_get_link_ksettings, 3954 }; 3955 3956 /* Routines for managing CRQs/sCRQs */ 3957 3958 static int reset_one_sub_crq_queue(struct ibmvnic_adapter *adapter, 3959 struct ibmvnic_sub_crq_queue *scrq) 3960 { 3961 int rc; 3962 3963 if (!scrq) { 3964 netdev_dbg(adapter->netdev, "Invalid scrq reset.\n"); 3965 return -EINVAL; 3966 } 3967 3968 if (scrq->irq) { 3969 free_irq(scrq->irq, scrq); 3970 irq_dispose_mapping(scrq->irq); 3971 scrq->irq = 0; 3972 } 3973 3974 if (scrq->msgs) { 3975 memset(scrq->msgs, 0, 4 * PAGE_SIZE); 3976 atomic_set(&scrq->used, 0); 3977 scrq->cur = 0; 3978 scrq->ind_buf.index = 0; 3979 } else { 3980 netdev_dbg(adapter->netdev, "Invalid scrq reset\n"); 3981 return -EINVAL; 3982 } 3983 3984 rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token, 3985 4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq); 3986 return rc; 3987 } 3988 3989 static int reset_sub_crq_queues(struct ibmvnic_adapter *adapter) 3990 { 3991 int i, rc; 3992 3993 if (!adapter->tx_scrq || !adapter->rx_scrq) 3994 return -EINVAL; 3995 3996 ibmvnic_clean_affinity(adapter); 3997 3998 for (i = 0; i < adapter->req_tx_queues; i++) { 3999 netdev_dbg(adapter->netdev, "Re-setting tx_scrq[%d]\n", i); 4000 rc = reset_one_sub_crq_queue(adapter, adapter->tx_scrq[i]); 4001 if (rc) 4002 return rc; 4003 } 4004 4005 for (i = 0; i < adapter->req_rx_queues; i++) { 4006 netdev_dbg(adapter->netdev, "Re-setting rx_scrq[%d]\n", i); 4007 rc = reset_one_sub_crq_queue(adapter, adapter->rx_scrq[i]); 4008 if (rc) 4009 return rc; 4010 } 4011 4012 return rc; 4013 } 4014 4015 static void release_sub_crq_queue(struct ibmvnic_adapter *adapter, 4016 struct ibmvnic_sub_crq_queue *scrq, 4017 bool do_h_free) 4018 { 4019 struct device *dev = &adapter->vdev->dev; 4020 long rc; 4021 4022 netdev_dbg(adapter->netdev, "Releasing sub-CRQ\n"); 4023 4024 if (do_h_free) { 4025 /* Close the sub-crqs */ 4026 do { 4027 rc = plpar_hcall_norets(H_FREE_SUB_CRQ, 4028 adapter->vdev->unit_address, 4029 scrq->crq_num); 4030 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc)); 4031 4032 if (rc) { 4033 netdev_err(adapter->netdev, 4034 "Failed to release sub-CRQ %16lx, rc = %ld\n", 4035 scrq->crq_num, rc); 4036 } 4037 } 4038 4039 dma_free_coherent(dev, 4040 IBMVNIC_IND_MAX_ARR_SZ, 4041 scrq->ind_buf.indir_arr, 4042 scrq->ind_buf.indir_dma); 4043 4044 dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE, 4045 DMA_BIDIRECTIONAL); 4046 free_pages((unsigned long)scrq->msgs, 2); 4047 free_cpumask_var(scrq->affinity_mask); 4048 kfree(scrq); 4049 } 4050 4051 static struct ibmvnic_sub_crq_queue *init_sub_crq_queue(struct ibmvnic_adapter 4052 *adapter) 4053 { 4054 struct device *dev = &adapter->vdev->dev; 4055 struct ibmvnic_sub_crq_queue *scrq; 4056 int rc; 4057 4058 scrq = kzalloc(sizeof(*scrq), GFP_KERNEL); 4059 if (!scrq) 4060 return NULL; 4061 4062 scrq->msgs = 4063 (union sub_crq *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 2); 4064 if (!scrq->msgs) { 4065 dev_warn(dev, "Couldn't allocate crq queue messages page\n"); 4066 goto zero_page_failed; 4067 } 4068 if (!zalloc_cpumask_var(&scrq->affinity_mask, GFP_KERNEL)) 4069 goto cpumask_alloc_failed; 4070 4071 scrq->msg_token = dma_map_single(dev, scrq->msgs, 4 * PAGE_SIZE, 4072 DMA_BIDIRECTIONAL); 4073 if (dma_mapping_error(dev, scrq->msg_token)) { 4074 dev_warn(dev, "Couldn't map crq queue messages page\n"); 4075 goto map_failed; 4076 } 4077 4078 rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token, 4079 4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq); 4080 4081 if (rc == H_RESOURCE) 4082 rc = ibmvnic_reset_crq(adapter); 4083 4084 if (rc == H_CLOSED) { 4085 dev_warn(dev, "Partner adapter not ready, waiting.\n"); 4086 } else if (rc) { 4087 dev_warn(dev, "Error %d registering sub-crq\n", rc); 4088 goto reg_failed; 4089 } 4090 4091 scrq->adapter = adapter; 4092 scrq->size = 4 * PAGE_SIZE / sizeof(*scrq->msgs); 4093 scrq->ind_buf.index = 0; 4094 4095 scrq->ind_buf.indir_arr = 4096 dma_alloc_coherent(dev, 4097 IBMVNIC_IND_MAX_ARR_SZ, 4098 &scrq->ind_buf.indir_dma, 4099 GFP_KERNEL); 4100 4101 if (!scrq->ind_buf.indir_arr) 4102 goto indir_failed; 4103 4104 spin_lock_init(&scrq->lock); 4105 4106 netdev_dbg(adapter->netdev, 4107 "sub-crq initialized, num %lx, hw_irq=%lx, irq=%x\n", 4108 scrq->crq_num, scrq->hw_irq, scrq->irq); 4109 4110 return scrq; 4111 4112 indir_failed: 4113 do { 4114 rc = plpar_hcall_norets(H_FREE_SUB_CRQ, 4115 adapter->vdev->unit_address, 4116 scrq->crq_num); 4117 } while (rc == H_BUSY || rc == H_IS_LONG_BUSY(rc)); 4118 reg_failed: 4119 dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE, 4120 DMA_BIDIRECTIONAL); 4121 map_failed: 4122 free_cpumask_var(scrq->affinity_mask); 4123 cpumask_alloc_failed: 4124 free_pages((unsigned long)scrq->msgs, 2); 4125 zero_page_failed: 4126 kfree(scrq); 4127 4128 return NULL; 4129 } 4130 4131 static void release_sub_crqs(struct ibmvnic_adapter *adapter, bool do_h_free) 4132 { 4133 int i; 4134 4135 ibmvnic_clean_affinity(adapter); 4136 if (adapter->tx_scrq) { 4137 for (i = 0; i < adapter->num_active_tx_scrqs; i++) { 4138 if (!adapter->tx_scrq[i]) 4139 continue; 4140 4141 netdev_dbg(adapter->netdev, "Releasing tx_scrq[%d]\n", 4142 i); 4143 ibmvnic_tx_scrq_clean_buffer(adapter, adapter->tx_scrq[i]); 4144 if (adapter->tx_scrq[i]->irq) { 4145 free_irq(adapter->tx_scrq[i]->irq, 4146 adapter->tx_scrq[i]); 4147 irq_dispose_mapping(adapter->tx_scrq[i]->irq); 4148 adapter->tx_scrq[i]->irq = 0; 4149 } 4150 4151 release_sub_crq_queue(adapter, adapter->tx_scrq[i], 4152 do_h_free); 4153 } 4154 4155 kfree(adapter->tx_scrq); 4156 adapter->tx_scrq = NULL; 4157 adapter->num_active_tx_scrqs = 0; 4158 } 4159 4160 /* Clean any remaining outstanding SKBs 4161 * we freed the irq so we won't be hearing 4162 * from them 4163 */ 4164 clean_tx_pools(adapter); 4165 4166 if (adapter->rx_scrq) { 4167 for (i = 0; i < adapter->num_active_rx_scrqs; i++) { 4168 if (!adapter->rx_scrq[i]) 4169 continue; 4170 4171 netdev_dbg(adapter->netdev, "Releasing rx_scrq[%d]\n", 4172 i); 4173 if (adapter->rx_scrq[i]->irq) { 4174 free_irq(adapter->rx_scrq[i]->irq, 4175 adapter->rx_scrq[i]); 4176 irq_dispose_mapping(adapter->rx_scrq[i]->irq); 4177 adapter->rx_scrq[i]->irq = 0; 4178 } 4179 4180 release_sub_crq_queue(adapter, adapter->rx_scrq[i], 4181 do_h_free); 4182 } 4183 4184 kfree(adapter->rx_scrq); 4185 adapter->rx_scrq = NULL; 4186 adapter->num_active_rx_scrqs = 0; 4187 } 4188 } 4189 4190 static int disable_scrq_irq(struct ibmvnic_adapter *adapter, 4191 struct ibmvnic_sub_crq_queue *scrq) 4192 { 4193 struct device *dev = &adapter->vdev->dev; 4194 unsigned long rc; 4195 4196 rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address, 4197 H_DISABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0); 4198 if (rc) 4199 dev_err(dev, "Couldn't disable scrq irq 0x%lx. rc=%ld\n", 4200 scrq->hw_irq, rc); 4201 return rc; 4202 } 4203 4204 /* We can not use the IRQ chip EOI handler because that has the 4205 * unintended effect of changing the interrupt priority. 4206 */ 4207 static void ibmvnic_xics_eoi(struct device *dev, struct ibmvnic_sub_crq_queue *scrq) 4208 { 4209 u64 val = 0xff000000 | scrq->hw_irq; 4210 unsigned long rc; 4211 4212 rc = plpar_hcall_norets(H_EOI, val); 4213 if (rc) 4214 dev_err(dev, "H_EOI FAILED irq 0x%llx. rc=%ld\n", val, rc); 4215 } 4216 4217 /* Due to a firmware bug, the hypervisor can send an interrupt to a 4218 * transmit or receive queue just prior to a partition migration. 4219 * Force an EOI after migration. 4220 */ 4221 static void ibmvnic_clear_pending_interrupt(struct device *dev, 4222 struct ibmvnic_sub_crq_queue *scrq) 4223 { 4224 if (!xive_enabled()) 4225 ibmvnic_xics_eoi(dev, scrq); 4226 } 4227 4228 static int enable_scrq_irq(struct ibmvnic_adapter *adapter, 4229 struct ibmvnic_sub_crq_queue *scrq) 4230 { 4231 struct device *dev = &adapter->vdev->dev; 4232 unsigned long rc; 4233 4234 if (scrq->hw_irq > 0x100000000ULL) { 4235 dev_err(dev, "bad hw_irq = %lx\n", scrq->hw_irq); 4236 return 1; 4237 } 4238 4239 if (test_bit(0, &adapter->resetting) && 4240 adapter->reset_reason == VNIC_RESET_MOBILITY) { 4241 ibmvnic_clear_pending_interrupt(dev, scrq); 4242 } 4243 4244 rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address, 4245 H_ENABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0); 4246 if (rc) 4247 dev_err(dev, "Couldn't enable scrq irq 0x%lx. rc=%ld\n", 4248 scrq->hw_irq, rc); 4249 return rc; 4250 } 4251 4252 static int ibmvnic_complete_tx(struct ibmvnic_adapter *adapter, 4253 struct ibmvnic_sub_crq_queue *scrq) 4254 { 4255 struct device *dev = &adapter->vdev->dev; 4256 int num_packets = 0, total_bytes = 0; 4257 struct ibmvnic_tx_pool *tx_pool; 4258 struct ibmvnic_tx_buff *txbuff; 4259 struct netdev_queue *txq; 4260 union sub_crq *next; 4261 int index, i; 4262 4263 restart_loop: 4264 while (pending_scrq(adapter, scrq)) { 4265 unsigned int pool = scrq->pool_index; 4266 int num_entries = 0; 4267 next = ibmvnic_next_scrq(adapter, scrq); 4268 for (i = 0; i < next->tx_comp.num_comps; i++) { 4269 index = be32_to_cpu(next->tx_comp.correlators[i]); 4270 if (index & IBMVNIC_TSO_POOL_MASK) { 4271 tx_pool = &adapter->tso_pool[pool]; 4272 index &= ~IBMVNIC_TSO_POOL_MASK; 4273 } else { 4274 tx_pool = &adapter->tx_pool[pool]; 4275 } 4276 4277 txbuff = &tx_pool->tx_buff[index]; 4278 num_packets++; 4279 num_entries += txbuff->num_entries; 4280 if (txbuff->skb) { 4281 total_bytes += txbuff->skb->len; 4282 if (next->tx_comp.rcs[i]) { 4283 dev_err(dev, "tx error %x\n", 4284 next->tx_comp.rcs[i]); 4285 dev_kfree_skb_irq(txbuff->skb); 4286 } else { 4287 dev_consume_skb_irq(txbuff->skb); 4288 } 4289 txbuff->skb = NULL; 4290 } else { 4291 netdev_warn(adapter->netdev, 4292 "TX completion received with NULL socket buffer\n"); 4293 } 4294 tx_pool->free_map[tx_pool->producer_index] = index; 4295 tx_pool->producer_index = 4296 (tx_pool->producer_index + 1) % 4297 tx_pool->num_buffers; 4298 } 4299 /* remove tx_comp scrq*/ 4300 next->tx_comp.first = 0; 4301 4302 4303 if (atomic_sub_return(num_entries, &scrq->used) <= 4304 (adapter->req_tx_entries_per_subcrq / 2) && 4305 __netif_subqueue_stopped(adapter->netdev, 4306 scrq->pool_index)) { 4307 rcu_read_lock(); 4308 if (adapter->tx_queues_active) { 4309 netif_wake_subqueue(adapter->netdev, 4310 scrq->pool_index); 4311 netdev_dbg(adapter->netdev, 4312 "Started queue %d\n", 4313 scrq->pool_index); 4314 } 4315 rcu_read_unlock(); 4316 } 4317 } 4318 4319 enable_scrq_irq(adapter, scrq); 4320 4321 if (pending_scrq(adapter, scrq)) { 4322 disable_scrq_irq(adapter, scrq); 4323 goto restart_loop; 4324 } 4325 4326 txq = netdev_get_tx_queue(adapter->netdev, scrq->pool_index); 4327 netdev_tx_completed_queue(txq, num_packets, total_bytes); 4328 4329 return 0; 4330 } 4331 4332 static irqreturn_t ibmvnic_interrupt_tx(int irq, void *instance) 4333 { 4334 struct ibmvnic_sub_crq_queue *scrq = instance; 4335 struct ibmvnic_adapter *adapter = scrq->adapter; 4336 4337 disable_scrq_irq(adapter, scrq); 4338 ibmvnic_complete_tx(adapter, scrq); 4339 4340 return IRQ_HANDLED; 4341 } 4342 4343 static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance) 4344 { 4345 struct ibmvnic_sub_crq_queue *scrq = instance; 4346 struct ibmvnic_adapter *adapter = scrq->adapter; 4347 4348 /* When booting a kdump kernel we can hit pending interrupts 4349 * prior to completing driver initialization. 4350 */ 4351 if (unlikely(adapter->state != VNIC_OPEN)) 4352 return IRQ_NONE; 4353 4354 adapter->rx_stats_buffers[scrq->scrq_num].interrupts++; 4355 4356 if (napi_schedule_prep(&adapter->napi[scrq->scrq_num])) { 4357 disable_scrq_irq(adapter, scrq); 4358 __napi_schedule(&adapter->napi[scrq->scrq_num]); 4359 } 4360 4361 return IRQ_HANDLED; 4362 } 4363 4364 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter) 4365 { 4366 struct device *dev = &adapter->vdev->dev; 4367 struct ibmvnic_sub_crq_queue *scrq; 4368 int i = 0, j = 0; 4369 int rc = 0; 4370 4371 for (i = 0; i < adapter->req_tx_queues; i++) { 4372 netdev_dbg(adapter->netdev, "Initializing tx_scrq[%d] irq\n", 4373 i); 4374 scrq = adapter->tx_scrq[i]; 4375 scrq->irq = irq_create_mapping(NULL, scrq->hw_irq); 4376 4377 if (!scrq->irq) { 4378 rc = -EINVAL; 4379 dev_err(dev, "Error mapping irq\n"); 4380 goto req_tx_irq_failed; 4381 } 4382 4383 snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-tx%d", 4384 adapter->vdev->unit_address, i); 4385 rc = request_irq(scrq->irq, ibmvnic_interrupt_tx, 4386 0, scrq->name, scrq); 4387 4388 if (rc) { 4389 dev_err(dev, "Couldn't register tx irq 0x%x. rc=%d\n", 4390 scrq->irq, rc); 4391 irq_dispose_mapping(scrq->irq); 4392 goto req_tx_irq_failed; 4393 } 4394 } 4395 4396 for (i = 0; i < adapter->req_rx_queues; i++) { 4397 netdev_dbg(adapter->netdev, "Initializing rx_scrq[%d] irq\n", 4398 i); 4399 scrq = adapter->rx_scrq[i]; 4400 scrq->irq = irq_create_mapping(NULL, scrq->hw_irq); 4401 if (!scrq->irq) { 4402 rc = -EINVAL; 4403 dev_err(dev, "Error mapping irq\n"); 4404 goto req_rx_irq_failed; 4405 } 4406 snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-rx%d", 4407 adapter->vdev->unit_address, i); 4408 rc = request_irq(scrq->irq, ibmvnic_interrupt_rx, 4409 0, scrq->name, scrq); 4410 if (rc) { 4411 dev_err(dev, "Couldn't register rx irq 0x%x. rc=%d\n", 4412 scrq->irq, rc); 4413 irq_dispose_mapping(scrq->irq); 4414 goto req_rx_irq_failed; 4415 } 4416 } 4417 4418 cpus_read_lock(); 4419 ibmvnic_set_affinity(adapter); 4420 cpus_read_unlock(); 4421 4422 return rc; 4423 4424 req_rx_irq_failed: 4425 for (j = 0; j < i; j++) { 4426 free_irq(adapter->rx_scrq[j]->irq, adapter->rx_scrq[j]); 4427 irq_dispose_mapping(adapter->rx_scrq[j]->irq); 4428 } 4429 i = adapter->req_tx_queues; 4430 req_tx_irq_failed: 4431 for (j = 0; j < i; j++) { 4432 free_irq(adapter->tx_scrq[j]->irq, adapter->tx_scrq[j]); 4433 irq_dispose_mapping(adapter->tx_scrq[j]->irq); 4434 } 4435 release_sub_crqs(adapter, 1); 4436 return rc; 4437 } 4438 4439 static int init_sub_crqs(struct ibmvnic_adapter *adapter) 4440 { 4441 struct device *dev = &adapter->vdev->dev; 4442 struct ibmvnic_sub_crq_queue **allqueues; 4443 int registered_queues = 0; 4444 int total_queues; 4445 int more = 0; 4446 int i; 4447 4448 total_queues = adapter->req_tx_queues + adapter->req_rx_queues; 4449 4450 allqueues = kcalloc(total_queues, sizeof(*allqueues), GFP_KERNEL); 4451 if (!allqueues) 4452 return -ENOMEM; 4453 4454 for (i = 0; i < total_queues; i++) { 4455 allqueues[i] = init_sub_crq_queue(adapter); 4456 if (!allqueues[i]) { 4457 dev_warn(dev, "Couldn't allocate all sub-crqs\n"); 4458 break; 4459 } 4460 registered_queues++; 4461 } 4462 4463 /* Make sure we were able to register the minimum number of queues */ 4464 if (registered_queues < 4465 adapter->min_tx_queues + adapter->min_rx_queues) { 4466 dev_err(dev, "Fatal: Couldn't init min number of sub-crqs\n"); 4467 goto tx_failed; 4468 } 4469 4470 /* Distribute the failed allocated queues*/ 4471 for (i = 0; i < total_queues - registered_queues + more ; i++) { 4472 netdev_dbg(adapter->netdev, "Reducing number of queues\n"); 4473 switch (i % 3) { 4474 case 0: 4475 if (adapter->req_rx_queues > adapter->min_rx_queues) 4476 adapter->req_rx_queues--; 4477 else 4478 more++; 4479 break; 4480 case 1: 4481 if (adapter->req_tx_queues > adapter->min_tx_queues) 4482 adapter->req_tx_queues--; 4483 else 4484 more++; 4485 break; 4486 } 4487 } 4488 4489 adapter->tx_scrq = kcalloc(adapter->req_tx_queues, 4490 sizeof(*adapter->tx_scrq), GFP_KERNEL); 4491 if (!adapter->tx_scrq) 4492 goto tx_failed; 4493 4494 for (i = 0; i < adapter->req_tx_queues; i++) { 4495 adapter->tx_scrq[i] = allqueues[i]; 4496 adapter->tx_scrq[i]->pool_index = i; 4497 adapter->num_active_tx_scrqs++; 4498 } 4499 4500 adapter->rx_scrq = kcalloc(adapter->req_rx_queues, 4501 sizeof(*adapter->rx_scrq), GFP_KERNEL); 4502 if (!adapter->rx_scrq) 4503 goto rx_failed; 4504 4505 for (i = 0; i < adapter->req_rx_queues; i++) { 4506 adapter->rx_scrq[i] = allqueues[i + adapter->req_tx_queues]; 4507 adapter->rx_scrq[i]->scrq_num = i; 4508 adapter->num_active_rx_scrqs++; 4509 } 4510 4511 kfree(allqueues); 4512 return 0; 4513 4514 rx_failed: 4515 kfree(adapter->tx_scrq); 4516 adapter->tx_scrq = NULL; 4517 tx_failed: 4518 for (i = 0; i < registered_queues; i++) 4519 release_sub_crq_queue(adapter, allqueues[i], 1); 4520 kfree(allqueues); 4521 return -ENOMEM; 4522 } 4523 4524 static void send_request_cap(struct ibmvnic_adapter *adapter, int retry) 4525 { 4526 struct device *dev = &adapter->vdev->dev; 4527 union ibmvnic_crq crq; 4528 int max_entries; 4529 int cap_reqs; 4530 4531 /* We send out 6 or 7 REQUEST_CAPABILITY CRQs below (depending on 4532 * the PROMISC flag). Initialize this count upfront. When the tasklet 4533 * receives a response to all of these, it will send the next protocol 4534 * message (QUERY_IP_OFFLOAD). 4535 */ 4536 if (!(adapter->netdev->flags & IFF_PROMISC) || 4537 adapter->promisc_supported) 4538 cap_reqs = 7; 4539 else 4540 cap_reqs = 6; 4541 4542 if (!retry) { 4543 /* Sub-CRQ entries are 32 byte long */ 4544 int entries_page = 4 * PAGE_SIZE / (sizeof(u64) * 4); 4545 4546 atomic_set(&adapter->running_cap_crqs, cap_reqs); 4547 4548 if (adapter->min_tx_entries_per_subcrq > entries_page || 4549 adapter->min_rx_add_entries_per_subcrq > entries_page) { 4550 dev_err(dev, "Fatal, invalid entries per sub-crq\n"); 4551 return; 4552 } 4553 4554 if (adapter->desired.mtu) 4555 adapter->req_mtu = adapter->desired.mtu; 4556 else 4557 adapter->req_mtu = adapter->netdev->mtu + ETH_HLEN; 4558 4559 if (!adapter->desired.tx_entries) 4560 adapter->desired.tx_entries = 4561 adapter->max_tx_entries_per_subcrq; 4562 if (!adapter->desired.rx_entries) 4563 adapter->desired.rx_entries = 4564 adapter->max_rx_add_entries_per_subcrq; 4565 4566 max_entries = IBMVNIC_LTB_SET_SIZE / 4567 (adapter->req_mtu + IBMVNIC_BUFFER_HLEN); 4568 4569 if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) * 4570 adapter->desired.tx_entries > IBMVNIC_LTB_SET_SIZE) { 4571 adapter->desired.tx_entries = max_entries; 4572 } 4573 4574 if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) * 4575 adapter->desired.rx_entries > IBMVNIC_LTB_SET_SIZE) { 4576 adapter->desired.rx_entries = max_entries; 4577 } 4578 4579 if (adapter->desired.tx_entries) 4580 adapter->req_tx_entries_per_subcrq = 4581 adapter->desired.tx_entries; 4582 else 4583 adapter->req_tx_entries_per_subcrq = 4584 adapter->max_tx_entries_per_subcrq; 4585 4586 if (adapter->desired.rx_entries) 4587 adapter->req_rx_add_entries_per_subcrq = 4588 adapter->desired.rx_entries; 4589 else 4590 adapter->req_rx_add_entries_per_subcrq = 4591 adapter->max_rx_add_entries_per_subcrq; 4592 4593 if (adapter->desired.tx_queues) 4594 adapter->req_tx_queues = 4595 adapter->desired.tx_queues; 4596 else 4597 adapter->req_tx_queues = 4598 adapter->opt_tx_comp_sub_queues; 4599 4600 if (adapter->desired.rx_queues) 4601 adapter->req_rx_queues = 4602 adapter->desired.rx_queues; 4603 else 4604 adapter->req_rx_queues = 4605 adapter->opt_rx_comp_queues; 4606 4607 adapter->req_rx_add_queues = adapter->max_rx_add_queues; 4608 } else { 4609 atomic_add(cap_reqs, &adapter->running_cap_crqs); 4610 } 4611 memset(&crq, 0, sizeof(crq)); 4612 crq.request_capability.first = IBMVNIC_CRQ_CMD; 4613 crq.request_capability.cmd = REQUEST_CAPABILITY; 4614 4615 crq.request_capability.capability = cpu_to_be16(REQ_TX_QUEUES); 4616 crq.request_capability.number = cpu_to_be64(adapter->req_tx_queues); 4617 cap_reqs--; 4618 ibmvnic_send_crq(adapter, &crq); 4619 4620 crq.request_capability.capability = cpu_to_be16(REQ_RX_QUEUES); 4621 crq.request_capability.number = cpu_to_be64(adapter->req_rx_queues); 4622 cap_reqs--; 4623 ibmvnic_send_crq(adapter, &crq); 4624 4625 crq.request_capability.capability = cpu_to_be16(REQ_RX_ADD_QUEUES); 4626 crq.request_capability.number = cpu_to_be64(adapter->req_rx_add_queues); 4627 cap_reqs--; 4628 ibmvnic_send_crq(adapter, &crq); 4629 4630 crq.request_capability.capability = 4631 cpu_to_be16(REQ_TX_ENTRIES_PER_SUBCRQ); 4632 crq.request_capability.number = 4633 cpu_to_be64(adapter->req_tx_entries_per_subcrq); 4634 cap_reqs--; 4635 ibmvnic_send_crq(adapter, &crq); 4636 4637 crq.request_capability.capability = 4638 cpu_to_be16(REQ_RX_ADD_ENTRIES_PER_SUBCRQ); 4639 crq.request_capability.number = 4640 cpu_to_be64(adapter->req_rx_add_entries_per_subcrq); 4641 cap_reqs--; 4642 ibmvnic_send_crq(adapter, &crq); 4643 4644 crq.request_capability.capability = cpu_to_be16(REQ_MTU); 4645 crq.request_capability.number = cpu_to_be64(adapter->req_mtu); 4646 cap_reqs--; 4647 ibmvnic_send_crq(adapter, &crq); 4648 4649 if (adapter->netdev->flags & IFF_PROMISC) { 4650 if (adapter->promisc_supported) { 4651 crq.request_capability.capability = 4652 cpu_to_be16(PROMISC_REQUESTED); 4653 crq.request_capability.number = cpu_to_be64(1); 4654 cap_reqs--; 4655 ibmvnic_send_crq(adapter, &crq); 4656 } 4657 } else { 4658 crq.request_capability.capability = 4659 cpu_to_be16(PROMISC_REQUESTED); 4660 crq.request_capability.number = cpu_to_be64(0); 4661 cap_reqs--; 4662 ibmvnic_send_crq(adapter, &crq); 4663 } 4664 4665 /* Keep at end to catch any discrepancy between expected and actual 4666 * CRQs sent. 4667 */ 4668 WARN_ON(cap_reqs != 0); 4669 } 4670 4671 static int pending_scrq(struct ibmvnic_adapter *adapter, 4672 struct ibmvnic_sub_crq_queue *scrq) 4673 { 4674 union sub_crq *entry = &scrq->msgs[scrq->cur]; 4675 int rc; 4676 4677 rc = !!(entry->generic.first & IBMVNIC_CRQ_CMD_RSP); 4678 4679 /* Ensure that the SCRQ valid flag is loaded prior to loading the 4680 * contents of the SCRQ descriptor 4681 */ 4682 dma_rmb(); 4683 4684 return rc; 4685 } 4686 4687 static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *adapter, 4688 struct ibmvnic_sub_crq_queue *scrq) 4689 { 4690 union sub_crq *entry; 4691 unsigned long flags; 4692 4693 spin_lock_irqsave(&scrq->lock, flags); 4694 entry = &scrq->msgs[scrq->cur]; 4695 if (entry->generic.first & IBMVNIC_CRQ_CMD_RSP) { 4696 if (++scrq->cur == scrq->size) 4697 scrq->cur = 0; 4698 } else { 4699 entry = NULL; 4700 } 4701 spin_unlock_irqrestore(&scrq->lock, flags); 4702 4703 /* Ensure that the SCRQ valid flag is loaded prior to loading the 4704 * contents of the SCRQ descriptor 4705 */ 4706 dma_rmb(); 4707 4708 return entry; 4709 } 4710 4711 static union ibmvnic_crq *ibmvnic_next_crq(struct ibmvnic_adapter *adapter) 4712 { 4713 struct ibmvnic_crq_queue *queue = &adapter->crq; 4714 union ibmvnic_crq *crq; 4715 4716 crq = &queue->msgs[queue->cur]; 4717 if (crq->generic.first & IBMVNIC_CRQ_CMD_RSP) { 4718 if (++queue->cur == queue->size) 4719 queue->cur = 0; 4720 } else { 4721 crq = NULL; 4722 } 4723 4724 return crq; 4725 } 4726 4727 static void print_subcrq_error(struct device *dev, int rc, const char *func) 4728 { 4729 switch (rc) { 4730 case H_PARAMETER: 4731 dev_warn_ratelimited(dev, 4732 "%s failed: Send request is malformed or adapter failover pending. (rc=%d)\n", 4733 func, rc); 4734 break; 4735 case H_CLOSED: 4736 dev_warn_ratelimited(dev, 4737 "%s failed: Backing queue closed. Adapter is down or failover pending. (rc=%d)\n", 4738 func, rc); 4739 break; 4740 default: 4741 dev_err_ratelimited(dev, "%s failed: (rc=%d)\n", func, rc); 4742 break; 4743 } 4744 } 4745 4746 static int send_subcrq_indirect(struct ibmvnic_adapter *adapter, 4747 u64 remote_handle, u64 ioba, u64 num_entries) 4748 { 4749 unsigned int ua = adapter->vdev->unit_address; 4750 struct device *dev = &adapter->vdev->dev; 4751 int rc; 4752 4753 /* Make sure the hypervisor sees the complete request */ 4754 dma_wmb(); 4755 rc = plpar_hcall_norets(H_SEND_SUB_CRQ_INDIRECT, ua, 4756 cpu_to_be64(remote_handle), 4757 ioba, num_entries); 4758 4759 if (rc) 4760 print_subcrq_error(dev, rc, __func__); 4761 4762 return rc; 4763 } 4764 4765 static int ibmvnic_send_crq(struct ibmvnic_adapter *adapter, 4766 union ibmvnic_crq *crq) 4767 { 4768 unsigned int ua = adapter->vdev->unit_address; 4769 struct device *dev = &adapter->vdev->dev; 4770 u64 *u64_crq = (u64 *)crq; 4771 int rc; 4772 4773 netdev_dbg(adapter->netdev, "Sending CRQ: %016lx %016lx\n", 4774 (unsigned long)cpu_to_be64(u64_crq[0]), 4775 (unsigned long)cpu_to_be64(u64_crq[1])); 4776 4777 if (!adapter->crq.active && 4778 crq->generic.first != IBMVNIC_CRQ_INIT_CMD) { 4779 dev_warn(dev, "Invalid request detected while CRQ is inactive, possible device state change during reset\n"); 4780 return -EINVAL; 4781 } 4782 4783 /* Make sure the hypervisor sees the complete request */ 4784 dma_wmb(); 4785 4786 rc = plpar_hcall_norets(H_SEND_CRQ, ua, 4787 cpu_to_be64(u64_crq[0]), 4788 cpu_to_be64(u64_crq[1])); 4789 4790 if (rc) { 4791 if (rc == H_CLOSED) { 4792 dev_warn(dev, "CRQ Queue closed\n"); 4793 /* do not reset, report the fail, wait for passive init from server */ 4794 } 4795 4796 dev_warn(dev, "Send error (rc=%d)\n", rc); 4797 } 4798 4799 return rc; 4800 } 4801 4802 static int ibmvnic_send_crq_init(struct ibmvnic_adapter *adapter) 4803 { 4804 struct device *dev = &adapter->vdev->dev; 4805 union ibmvnic_crq crq; 4806 int retries = 100; 4807 int rc; 4808 4809 memset(&crq, 0, sizeof(crq)); 4810 crq.generic.first = IBMVNIC_CRQ_INIT_CMD; 4811 crq.generic.cmd = IBMVNIC_CRQ_INIT; 4812 netdev_dbg(adapter->netdev, "Sending CRQ init\n"); 4813 4814 do { 4815 rc = ibmvnic_send_crq(adapter, &crq); 4816 if (rc != H_CLOSED) 4817 break; 4818 retries--; 4819 msleep(50); 4820 4821 } while (retries > 0); 4822 4823 if (rc) { 4824 dev_err(dev, "Failed to send init request, rc = %d\n", rc); 4825 return rc; 4826 } 4827 4828 return 0; 4829 } 4830 4831 struct vnic_login_client_data { 4832 u8 type; 4833 __be16 len; 4834 char name[]; 4835 } __packed; 4836 4837 static int vnic_client_data_len(struct ibmvnic_adapter *adapter) 4838 { 4839 int len; 4840 4841 /* Calculate the amount of buffer space needed for the 4842 * vnic client data in the login buffer. There are four entries, 4843 * OS name, LPAR name, device name, and a null last entry. 4844 */ 4845 len = 4 * sizeof(struct vnic_login_client_data); 4846 len += 6; /* "Linux" plus NULL */ 4847 len += strlen(utsname()->nodename) + 1; 4848 len += strlen(adapter->netdev->name) + 1; 4849 4850 return len; 4851 } 4852 4853 static void vnic_add_client_data(struct ibmvnic_adapter *adapter, 4854 struct vnic_login_client_data *vlcd) 4855 { 4856 const char *os_name = "Linux"; 4857 int len; 4858 4859 /* Type 1 - LPAR OS */ 4860 vlcd->type = 1; 4861 len = strlen(os_name) + 1; 4862 vlcd->len = cpu_to_be16(len); 4863 strscpy(vlcd->name, os_name, len); 4864 vlcd = (struct vnic_login_client_data *)(vlcd->name + len); 4865 4866 /* Type 2 - LPAR name */ 4867 vlcd->type = 2; 4868 len = strlen(utsname()->nodename) + 1; 4869 vlcd->len = cpu_to_be16(len); 4870 strscpy(vlcd->name, utsname()->nodename, len); 4871 vlcd = (struct vnic_login_client_data *)(vlcd->name + len); 4872 4873 /* Type 3 - device name */ 4874 vlcd->type = 3; 4875 len = strlen(adapter->netdev->name) + 1; 4876 vlcd->len = cpu_to_be16(len); 4877 strscpy(vlcd->name, adapter->netdev->name, len); 4878 } 4879 4880 static void ibmvnic_print_hex_dump(struct net_device *dev, void *buf, 4881 size_t len) 4882 { 4883 unsigned char hex_str[16 * 3]; 4884 4885 for (size_t i = 0; i < len; i += 16) { 4886 hex_dump_to_buffer((unsigned char *)buf + i, len - i, 16, 8, 4887 hex_str, sizeof(hex_str), false); 4888 netdev_dbg(dev, "%s\n", hex_str); 4889 } 4890 } 4891 4892 static int send_login(struct ibmvnic_adapter *adapter) 4893 { 4894 struct ibmvnic_login_rsp_buffer *login_rsp_buffer; 4895 struct ibmvnic_login_buffer *login_buffer; 4896 struct device *dev = &adapter->vdev->dev; 4897 struct vnic_login_client_data *vlcd; 4898 dma_addr_t rsp_buffer_token; 4899 dma_addr_t buffer_token; 4900 size_t rsp_buffer_size; 4901 union ibmvnic_crq crq; 4902 int client_data_len; 4903 size_t buffer_size; 4904 __be64 *tx_list_p; 4905 __be64 *rx_list_p; 4906 int rc; 4907 int i; 4908 4909 if (!adapter->tx_scrq || !adapter->rx_scrq) { 4910 netdev_err(adapter->netdev, 4911 "RX or TX queues are not allocated, device login failed\n"); 4912 return -ENOMEM; 4913 } 4914 4915 release_login_buffer(adapter); 4916 release_login_rsp_buffer(adapter); 4917 4918 client_data_len = vnic_client_data_len(adapter); 4919 4920 buffer_size = 4921 sizeof(struct ibmvnic_login_buffer) + 4922 sizeof(u64) * (adapter->req_tx_queues + adapter->req_rx_queues) + 4923 client_data_len; 4924 4925 login_buffer = kzalloc(buffer_size, GFP_ATOMIC); 4926 if (!login_buffer) 4927 goto buf_alloc_failed; 4928 4929 buffer_token = dma_map_single(dev, login_buffer, buffer_size, 4930 DMA_TO_DEVICE); 4931 if (dma_mapping_error(dev, buffer_token)) { 4932 dev_err(dev, "Couldn't map login buffer\n"); 4933 goto buf_map_failed; 4934 } 4935 4936 rsp_buffer_size = sizeof(struct ibmvnic_login_rsp_buffer) + 4937 sizeof(u64) * adapter->req_tx_queues + 4938 sizeof(u64) * adapter->req_rx_queues + 4939 sizeof(u64) * adapter->req_rx_queues + 4940 sizeof(u8) * IBMVNIC_TX_DESC_VERSIONS; 4941 4942 login_rsp_buffer = kmalloc(rsp_buffer_size, GFP_ATOMIC); 4943 if (!login_rsp_buffer) 4944 goto buf_rsp_alloc_failed; 4945 4946 rsp_buffer_token = dma_map_single(dev, login_rsp_buffer, 4947 rsp_buffer_size, DMA_FROM_DEVICE); 4948 if (dma_mapping_error(dev, rsp_buffer_token)) { 4949 dev_err(dev, "Couldn't map login rsp buffer\n"); 4950 goto buf_rsp_map_failed; 4951 } 4952 4953 adapter->login_buf = login_buffer; 4954 adapter->login_buf_token = buffer_token; 4955 adapter->login_buf_sz = buffer_size; 4956 adapter->login_rsp_buf = login_rsp_buffer; 4957 adapter->login_rsp_buf_token = rsp_buffer_token; 4958 adapter->login_rsp_buf_sz = rsp_buffer_size; 4959 4960 login_buffer->len = cpu_to_be32(buffer_size); 4961 login_buffer->version = cpu_to_be32(INITIAL_VERSION_LB); 4962 login_buffer->num_txcomp_subcrqs = cpu_to_be32(adapter->req_tx_queues); 4963 login_buffer->off_txcomp_subcrqs = 4964 cpu_to_be32(sizeof(struct ibmvnic_login_buffer)); 4965 login_buffer->num_rxcomp_subcrqs = cpu_to_be32(adapter->req_rx_queues); 4966 login_buffer->off_rxcomp_subcrqs = 4967 cpu_to_be32(sizeof(struct ibmvnic_login_buffer) + 4968 sizeof(u64) * adapter->req_tx_queues); 4969 login_buffer->login_rsp_ioba = cpu_to_be32(rsp_buffer_token); 4970 login_buffer->login_rsp_len = cpu_to_be32(rsp_buffer_size); 4971 4972 tx_list_p = (__be64 *)((char *)login_buffer + 4973 sizeof(struct ibmvnic_login_buffer)); 4974 rx_list_p = (__be64 *)((char *)login_buffer + 4975 sizeof(struct ibmvnic_login_buffer) + 4976 sizeof(u64) * adapter->req_tx_queues); 4977 4978 for (i = 0; i < adapter->req_tx_queues; i++) { 4979 if (adapter->tx_scrq[i]) { 4980 tx_list_p[i] = 4981 cpu_to_be64(adapter->tx_scrq[i]->crq_num); 4982 } 4983 } 4984 4985 for (i = 0; i < adapter->req_rx_queues; i++) { 4986 if (adapter->rx_scrq[i]) { 4987 rx_list_p[i] = 4988 cpu_to_be64(adapter->rx_scrq[i]->crq_num); 4989 } 4990 } 4991 4992 /* Insert vNIC login client data */ 4993 vlcd = (struct vnic_login_client_data *) 4994 ((char *)rx_list_p + (sizeof(u64) * adapter->req_rx_queues)); 4995 login_buffer->client_data_offset = 4996 cpu_to_be32((char *)vlcd - (char *)login_buffer); 4997 login_buffer->client_data_len = cpu_to_be32(client_data_len); 4998 4999 vnic_add_client_data(adapter, vlcd); 5000 5001 netdev_dbg(adapter->netdev, "Login Buffer:\n"); 5002 ibmvnic_print_hex_dump(adapter->netdev, adapter->login_buf, 5003 adapter->login_buf_sz); 5004 5005 memset(&crq, 0, sizeof(crq)); 5006 crq.login.first = IBMVNIC_CRQ_CMD; 5007 crq.login.cmd = LOGIN; 5008 crq.login.ioba = cpu_to_be32(buffer_token); 5009 crq.login.len = cpu_to_be32(buffer_size); 5010 5011 adapter->login_pending = true; 5012 rc = ibmvnic_send_crq(adapter, &crq); 5013 if (rc) { 5014 adapter->login_pending = false; 5015 netdev_err(adapter->netdev, "Failed to send login, rc=%d\n", rc); 5016 goto buf_send_failed; 5017 } 5018 5019 return 0; 5020 5021 buf_send_failed: 5022 dma_unmap_single(dev, rsp_buffer_token, rsp_buffer_size, 5023 DMA_FROM_DEVICE); 5024 buf_rsp_map_failed: 5025 kfree(login_rsp_buffer); 5026 adapter->login_rsp_buf = NULL; 5027 buf_rsp_alloc_failed: 5028 dma_unmap_single(dev, buffer_token, buffer_size, DMA_TO_DEVICE); 5029 buf_map_failed: 5030 kfree(login_buffer); 5031 adapter->login_buf = NULL; 5032 buf_alloc_failed: 5033 return -ENOMEM; 5034 } 5035 5036 static int send_request_map(struct ibmvnic_adapter *adapter, dma_addr_t addr, 5037 u32 len, u8 map_id) 5038 { 5039 union ibmvnic_crq crq; 5040 5041 memset(&crq, 0, sizeof(crq)); 5042 crq.request_map.first = IBMVNIC_CRQ_CMD; 5043 crq.request_map.cmd = REQUEST_MAP; 5044 crq.request_map.map_id = map_id; 5045 crq.request_map.ioba = cpu_to_be32(addr); 5046 crq.request_map.len = cpu_to_be32(len); 5047 return ibmvnic_send_crq(adapter, &crq); 5048 } 5049 5050 static int send_request_unmap(struct ibmvnic_adapter *adapter, u8 map_id) 5051 { 5052 union ibmvnic_crq crq; 5053 5054 memset(&crq, 0, sizeof(crq)); 5055 crq.request_unmap.first = IBMVNIC_CRQ_CMD; 5056 crq.request_unmap.cmd = REQUEST_UNMAP; 5057 crq.request_unmap.map_id = map_id; 5058 return ibmvnic_send_crq(adapter, &crq); 5059 } 5060 5061 static void send_query_map(struct ibmvnic_adapter *adapter) 5062 { 5063 union ibmvnic_crq crq; 5064 5065 memset(&crq, 0, sizeof(crq)); 5066 crq.query_map.first = IBMVNIC_CRQ_CMD; 5067 crq.query_map.cmd = QUERY_MAP; 5068 ibmvnic_send_crq(adapter, &crq); 5069 } 5070 5071 /* Send a series of CRQs requesting various capabilities of the VNIC server */ 5072 static void send_query_cap(struct ibmvnic_adapter *adapter) 5073 { 5074 union ibmvnic_crq crq; 5075 int cap_reqs; 5076 5077 /* We send out 25 QUERY_CAPABILITY CRQs below. Initialize this count 5078 * upfront. When the tasklet receives a response to all of these, it 5079 * can send out the next protocol messaage (REQUEST_CAPABILITY). 5080 */ 5081 cap_reqs = 25; 5082 5083 atomic_set(&adapter->running_cap_crqs, cap_reqs); 5084 5085 memset(&crq, 0, sizeof(crq)); 5086 crq.query_capability.first = IBMVNIC_CRQ_CMD; 5087 crq.query_capability.cmd = QUERY_CAPABILITY; 5088 5089 crq.query_capability.capability = cpu_to_be16(MIN_TX_QUEUES); 5090 ibmvnic_send_crq(adapter, &crq); 5091 cap_reqs--; 5092 5093 crq.query_capability.capability = cpu_to_be16(MIN_RX_QUEUES); 5094 ibmvnic_send_crq(adapter, &crq); 5095 cap_reqs--; 5096 5097 crq.query_capability.capability = cpu_to_be16(MIN_RX_ADD_QUEUES); 5098 ibmvnic_send_crq(adapter, &crq); 5099 cap_reqs--; 5100 5101 crq.query_capability.capability = cpu_to_be16(MAX_TX_QUEUES); 5102 ibmvnic_send_crq(adapter, &crq); 5103 cap_reqs--; 5104 5105 crq.query_capability.capability = cpu_to_be16(MAX_RX_QUEUES); 5106 ibmvnic_send_crq(adapter, &crq); 5107 cap_reqs--; 5108 5109 crq.query_capability.capability = cpu_to_be16(MAX_RX_ADD_QUEUES); 5110 ibmvnic_send_crq(adapter, &crq); 5111 cap_reqs--; 5112 5113 crq.query_capability.capability = 5114 cpu_to_be16(MIN_TX_ENTRIES_PER_SUBCRQ); 5115 ibmvnic_send_crq(adapter, &crq); 5116 cap_reqs--; 5117 5118 crq.query_capability.capability = 5119 cpu_to_be16(MIN_RX_ADD_ENTRIES_PER_SUBCRQ); 5120 ibmvnic_send_crq(adapter, &crq); 5121 cap_reqs--; 5122 5123 crq.query_capability.capability = 5124 cpu_to_be16(MAX_TX_ENTRIES_PER_SUBCRQ); 5125 ibmvnic_send_crq(adapter, &crq); 5126 cap_reqs--; 5127 5128 crq.query_capability.capability = 5129 cpu_to_be16(MAX_RX_ADD_ENTRIES_PER_SUBCRQ); 5130 ibmvnic_send_crq(adapter, &crq); 5131 cap_reqs--; 5132 5133 crq.query_capability.capability = cpu_to_be16(TCP_IP_OFFLOAD); 5134 ibmvnic_send_crq(adapter, &crq); 5135 cap_reqs--; 5136 5137 crq.query_capability.capability = cpu_to_be16(PROMISC_SUPPORTED); 5138 ibmvnic_send_crq(adapter, &crq); 5139 cap_reqs--; 5140 5141 crq.query_capability.capability = cpu_to_be16(MIN_MTU); 5142 ibmvnic_send_crq(adapter, &crq); 5143 cap_reqs--; 5144 5145 crq.query_capability.capability = cpu_to_be16(MAX_MTU); 5146 ibmvnic_send_crq(adapter, &crq); 5147 cap_reqs--; 5148 5149 crq.query_capability.capability = cpu_to_be16(MAX_MULTICAST_FILTERS); 5150 ibmvnic_send_crq(adapter, &crq); 5151 cap_reqs--; 5152 5153 crq.query_capability.capability = cpu_to_be16(VLAN_HEADER_INSERTION); 5154 ibmvnic_send_crq(adapter, &crq); 5155 cap_reqs--; 5156 5157 crq.query_capability.capability = cpu_to_be16(RX_VLAN_HEADER_INSERTION); 5158 ibmvnic_send_crq(adapter, &crq); 5159 cap_reqs--; 5160 5161 crq.query_capability.capability = cpu_to_be16(MAX_TX_SG_ENTRIES); 5162 ibmvnic_send_crq(adapter, &crq); 5163 cap_reqs--; 5164 5165 crq.query_capability.capability = cpu_to_be16(RX_SG_SUPPORTED); 5166 ibmvnic_send_crq(adapter, &crq); 5167 cap_reqs--; 5168 5169 crq.query_capability.capability = cpu_to_be16(OPT_TX_COMP_SUB_QUEUES); 5170 ibmvnic_send_crq(adapter, &crq); 5171 cap_reqs--; 5172 5173 crq.query_capability.capability = cpu_to_be16(OPT_RX_COMP_QUEUES); 5174 ibmvnic_send_crq(adapter, &crq); 5175 cap_reqs--; 5176 5177 crq.query_capability.capability = 5178 cpu_to_be16(OPT_RX_BUFADD_Q_PER_RX_COMP_Q); 5179 ibmvnic_send_crq(adapter, &crq); 5180 cap_reqs--; 5181 5182 crq.query_capability.capability = 5183 cpu_to_be16(OPT_TX_ENTRIES_PER_SUBCRQ); 5184 ibmvnic_send_crq(adapter, &crq); 5185 cap_reqs--; 5186 5187 crq.query_capability.capability = 5188 cpu_to_be16(OPT_RXBA_ENTRIES_PER_SUBCRQ); 5189 ibmvnic_send_crq(adapter, &crq); 5190 cap_reqs--; 5191 5192 crq.query_capability.capability = cpu_to_be16(TX_RX_DESC_REQ); 5193 5194 ibmvnic_send_crq(adapter, &crq); 5195 cap_reqs--; 5196 5197 /* Keep at end to catch any discrepancy between expected and actual 5198 * CRQs sent. 5199 */ 5200 WARN_ON(cap_reqs != 0); 5201 } 5202 5203 static void send_query_ip_offload(struct ibmvnic_adapter *adapter) 5204 { 5205 int buf_sz = sizeof(struct ibmvnic_query_ip_offload_buffer); 5206 struct device *dev = &adapter->vdev->dev; 5207 union ibmvnic_crq crq; 5208 5209 adapter->ip_offload_tok = 5210 dma_map_single(dev, 5211 &adapter->ip_offload_buf, 5212 buf_sz, 5213 DMA_FROM_DEVICE); 5214 5215 if (dma_mapping_error(dev, adapter->ip_offload_tok)) { 5216 if (!firmware_has_feature(FW_FEATURE_CMO)) 5217 dev_err(dev, "Couldn't map offload buffer\n"); 5218 return; 5219 } 5220 5221 memset(&crq, 0, sizeof(crq)); 5222 crq.query_ip_offload.first = IBMVNIC_CRQ_CMD; 5223 crq.query_ip_offload.cmd = QUERY_IP_OFFLOAD; 5224 crq.query_ip_offload.len = cpu_to_be32(buf_sz); 5225 crq.query_ip_offload.ioba = 5226 cpu_to_be32(adapter->ip_offload_tok); 5227 5228 ibmvnic_send_crq(adapter, &crq); 5229 } 5230 5231 static void send_control_ip_offload(struct ibmvnic_adapter *adapter) 5232 { 5233 struct ibmvnic_control_ip_offload_buffer *ctrl_buf = &adapter->ip_offload_ctrl; 5234 struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf; 5235 struct device *dev = &adapter->vdev->dev; 5236 netdev_features_t old_hw_features = 0; 5237 union ibmvnic_crq crq; 5238 5239 adapter->ip_offload_ctrl_tok = 5240 dma_map_single(dev, 5241 ctrl_buf, 5242 sizeof(adapter->ip_offload_ctrl), 5243 DMA_TO_DEVICE); 5244 5245 if (dma_mapping_error(dev, adapter->ip_offload_ctrl_tok)) { 5246 dev_err(dev, "Couldn't map ip offload control buffer\n"); 5247 return; 5248 } 5249 5250 ctrl_buf->len = cpu_to_be32(sizeof(adapter->ip_offload_ctrl)); 5251 ctrl_buf->version = cpu_to_be32(INITIAL_VERSION_IOB); 5252 ctrl_buf->ipv4_chksum = buf->ipv4_chksum; 5253 ctrl_buf->ipv6_chksum = buf->ipv6_chksum; 5254 ctrl_buf->tcp_ipv4_chksum = buf->tcp_ipv4_chksum; 5255 ctrl_buf->udp_ipv4_chksum = buf->udp_ipv4_chksum; 5256 ctrl_buf->tcp_ipv6_chksum = buf->tcp_ipv6_chksum; 5257 ctrl_buf->udp_ipv6_chksum = buf->udp_ipv6_chksum; 5258 ctrl_buf->large_tx_ipv4 = buf->large_tx_ipv4; 5259 ctrl_buf->large_tx_ipv6 = buf->large_tx_ipv6; 5260 5261 /* large_rx disabled for now, additional features needed */ 5262 ctrl_buf->large_rx_ipv4 = 0; 5263 ctrl_buf->large_rx_ipv6 = 0; 5264 5265 if (adapter->state != VNIC_PROBING) { 5266 old_hw_features = adapter->netdev->hw_features; 5267 adapter->netdev->hw_features = 0; 5268 } 5269 5270 adapter->netdev->hw_features = NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO; 5271 5272 if (buf->tcp_ipv4_chksum || buf->udp_ipv4_chksum) 5273 adapter->netdev->hw_features |= NETIF_F_IP_CSUM; 5274 5275 if (buf->tcp_ipv6_chksum || buf->udp_ipv6_chksum) 5276 adapter->netdev->hw_features |= NETIF_F_IPV6_CSUM; 5277 5278 if ((adapter->netdev->features & 5279 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM))) 5280 adapter->netdev->hw_features |= NETIF_F_RXCSUM; 5281 5282 if (buf->large_tx_ipv4) 5283 adapter->netdev->hw_features |= NETIF_F_TSO; 5284 if (buf->large_tx_ipv6) 5285 adapter->netdev->hw_features |= NETIF_F_TSO6; 5286 5287 if (adapter->state == VNIC_PROBING) { 5288 adapter->netdev->features |= adapter->netdev->hw_features; 5289 } else if (old_hw_features != adapter->netdev->hw_features) { 5290 netdev_features_t tmp = 0; 5291 5292 /* disable features no longer supported */ 5293 adapter->netdev->features &= adapter->netdev->hw_features; 5294 /* turn on features now supported if previously enabled */ 5295 tmp = (old_hw_features ^ adapter->netdev->hw_features) & 5296 adapter->netdev->hw_features; 5297 adapter->netdev->features |= 5298 tmp & adapter->netdev->wanted_features; 5299 } 5300 5301 memset(&crq, 0, sizeof(crq)); 5302 crq.control_ip_offload.first = IBMVNIC_CRQ_CMD; 5303 crq.control_ip_offload.cmd = CONTROL_IP_OFFLOAD; 5304 crq.control_ip_offload.len = 5305 cpu_to_be32(sizeof(adapter->ip_offload_ctrl)); 5306 crq.control_ip_offload.ioba = cpu_to_be32(adapter->ip_offload_ctrl_tok); 5307 ibmvnic_send_crq(adapter, &crq); 5308 } 5309 5310 static void handle_vpd_size_rsp(union ibmvnic_crq *crq, 5311 struct ibmvnic_adapter *adapter) 5312 { 5313 struct device *dev = &adapter->vdev->dev; 5314 5315 if (crq->get_vpd_size_rsp.rc.code) { 5316 dev_err(dev, "Error retrieving VPD size, rc=%x\n", 5317 crq->get_vpd_size_rsp.rc.code); 5318 complete(&adapter->fw_done); 5319 return; 5320 } 5321 5322 adapter->vpd->len = be64_to_cpu(crq->get_vpd_size_rsp.len); 5323 complete(&adapter->fw_done); 5324 } 5325 5326 static void handle_vpd_rsp(union ibmvnic_crq *crq, 5327 struct ibmvnic_adapter *adapter) 5328 { 5329 struct device *dev = &adapter->vdev->dev; 5330 unsigned char *substr = NULL; 5331 u8 fw_level_len = 0; 5332 5333 memset(adapter->fw_version, 0, 32); 5334 5335 dma_unmap_single(dev, adapter->vpd->dma_addr, adapter->vpd->len, 5336 DMA_FROM_DEVICE); 5337 5338 if (crq->get_vpd_rsp.rc.code) { 5339 dev_err(dev, "Error retrieving VPD from device, rc=%x\n", 5340 crq->get_vpd_rsp.rc.code); 5341 goto complete; 5342 } 5343 5344 /* get the position of the firmware version info 5345 * located after the ASCII 'RM' substring in the buffer 5346 */ 5347 substr = strnstr(adapter->vpd->buff, "RM", adapter->vpd->len); 5348 if (!substr) { 5349 dev_info(dev, "Warning - No FW level has been provided in the VPD buffer by the VIOS Server\n"); 5350 goto complete; 5351 } 5352 5353 /* get length of firmware level ASCII substring */ 5354 if ((substr + 2) < (adapter->vpd->buff + adapter->vpd->len)) { 5355 fw_level_len = *(substr + 2); 5356 } else { 5357 dev_info(dev, "Length of FW substr extrapolated VDP buff\n"); 5358 goto complete; 5359 } 5360 5361 /* copy firmware version string from vpd into adapter */ 5362 if ((substr + 3 + fw_level_len) < 5363 (adapter->vpd->buff + adapter->vpd->len)) { 5364 strscpy(adapter->fw_version, substr + 3, 5365 sizeof(adapter->fw_version)); 5366 } else { 5367 dev_info(dev, "FW substr extrapolated VPD buff\n"); 5368 } 5369 5370 complete: 5371 if (adapter->fw_version[0] == '\0') 5372 strscpy((char *)adapter->fw_version, "N/A", sizeof(adapter->fw_version)); 5373 complete(&adapter->fw_done); 5374 } 5375 5376 static void handle_query_ip_offload_rsp(struct ibmvnic_adapter *adapter) 5377 { 5378 struct device *dev = &adapter->vdev->dev; 5379 struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf; 5380 5381 dma_unmap_single(dev, adapter->ip_offload_tok, 5382 sizeof(adapter->ip_offload_buf), DMA_FROM_DEVICE); 5383 5384 netdev_dbg(adapter->netdev, "Query IP Offload Buffer:\n"); 5385 ibmvnic_print_hex_dump(adapter->netdev, buf, 5386 sizeof(adapter->ip_offload_buf)); 5387 5388 netdev_dbg(adapter->netdev, "ipv4_chksum = %d\n", buf->ipv4_chksum); 5389 netdev_dbg(adapter->netdev, "ipv6_chksum = %d\n", buf->ipv6_chksum); 5390 netdev_dbg(adapter->netdev, "tcp_ipv4_chksum = %d\n", 5391 buf->tcp_ipv4_chksum); 5392 netdev_dbg(adapter->netdev, "tcp_ipv6_chksum = %d\n", 5393 buf->tcp_ipv6_chksum); 5394 netdev_dbg(adapter->netdev, "udp_ipv4_chksum = %d\n", 5395 buf->udp_ipv4_chksum); 5396 netdev_dbg(adapter->netdev, "udp_ipv6_chksum = %d\n", 5397 buf->udp_ipv6_chksum); 5398 netdev_dbg(adapter->netdev, "large_tx_ipv4 = %d\n", 5399 buf->large_tx_ipv4); 5400 netdev_dbg(adapter->netdev, "large_tx_ipv6 = %d\n", 5401 buf->large_tx_ipv6); 5402 netdev_dbg(adapter->netdev, "large_rx_ipv4 = %d\n", 5403 buf->large_rx_ipv4); 5404 netdev_dbg(adapter->netdev, "large_rx_ipv6 = %d\n", 5405 buf->large_rx_ipv6); 5406 netdev_dbg(adapter->netdev, "max_ipv4_hdr_sz = %d\n", 5407 buf->max_ipv4_header_size); 5408 netdev_dbg(adapter->netdev, "max_ipv6_hdr_sz = %d\n", 5409 buf->max_ipv6_header_size); 5410 netdev_dbg(adapter->netdev, "max_tcp_hdr_size = %d\n", 5411 buf->max_tcp_header_size); 5412 netdev_dbg(adapter->netdev, "max_udp_hdr_size = %d\n", 5413 buf->max_udp_header_size); 5414 netdev_dbg(adapter->netdev, "max_large_tx_size = %d\n", 5415 buf->max_large_tx_size); 5416 netdev_dbg(adapter->netdev, "max_large_rx_size = %d\n", 5417 buf->max_large_rx_size); 5418 netdev_dbg(adapter->netdev, "ipv6_ext_hdr = %d\n", 5419 buf->ipv6_extension_header); 5420 netdev_dbg(adapter->netdev, "tcp_pseudosum_req = %d\n", 5421 buf->tcp_pseudosum_req); 5422 netdev_dbg(adapter->netdev, "num_ipv6_ext_hd = %d\n", 5423 buf->num_ipv6_ext_headers); 5424 netdev_dbg(adapter->netdev, "off_ipv6_ext_hd = %d\n", 5425 buf->off_ipv6_ext_headers); 5426 5427 send_control_ip_offload(adapter); 5428 } 5429 5430 static const char *ibmvnic_fw_err_cause(u16 cause) 5431 { 5432 switch (cause) { 5433 case ADAPTER_PROBLEM: 5434 return "adapter problem"; 5435 case BUS_PROBLEM: 5436 return "bus problem"; 5437 case FW_PROBLEM: 5438 return "firmware problem"; 5439 case DD_PROBLEM: 5440 return "device driver problem"; 5441 case EEH_RECOVERY: 5442 return "EEH recovery"; 5443 case FW_UPDATED: 5444 return "firmware updated"; 5445 case LOW_MEMORY: 5446 return "low Memory"; 5447 default: 5448 return "unknown"; 5449 } 5450 } 5451 5452 static void handle_error_indication(union ibmvnic_crq *crq, 5453 struct ibmvnic_adapter *adapter) 5454 { 5455 struct device *dev = &adapter->vdev->dev; 5456 u16 cause; 5457 5458 cause = be16_to_cpu(crq->error_indication.error_cause); 5459 5460 dev_warn_ratelimited(dev, 5461 "Firmware reports %serror, cause: %s. Starting recovery...\n", 5462 crq->error_indication.flags 5463 & IBMVNIC_FATAL_ERROR ? "FATAL " : "", 5464 ibmvnic_fw_err_cause(cause)); 5465 5466 if (crq->error_indication.flags & IBMVNIC_FATAL_ERROR) 5467 ibmvnic_reset(adapter, VNIC_RESET_FATAL); 5468 else 5469 ibmvnic_reset(adapter, VNIC_RESET_NON_FATAL); 5470 } 5471 5472 static int handle_change_mac_rsp(union ibmvnic_crq *crq, 5473 struct ibmvnic_adapter *adapter) 5474 { 5475 struct net_device *netdev = adapter->netdev; 5476 struct device *dev = &adapter->vdev->dev; 5477 long rc; 5478 5479 rc = crq->change_mac_addr_rsp.rc.code; 5480 if (rc) { 5481 dev_err(dev, "Error %ld in CHANGE_MAC_ADDR_RSP\n", rc); 5482 goto out; 5483 } 5484 /* crq->change_mac_addr.mac_addr is the requested one 5485 * crq->change_mac_addr_rsp.mac_addr is the returned valid one. 5486 */ 5487 eth_hw_addr_set(netdev, &crq->change_mac_addr_rsp.mac_addr[0]); 5488 ether_addr_copy(adapter->mac_addr, 5489 &crq->change_mac_addr_rsp.mac_addr[0]); 5490 out: 5491 complete(&adapter->fw_done); 5492 return rc; 5493 } 5494 5495 static void handle_request_cap_rsp(union ibmvnic_crq *crq, 5496 struct ibmvnic_adapter *adapter) 5497 { 5498 struct device *dev = &adapter->vdev->dev; 5499 u64 *req_value; 5500 char *name; 5501 5502 atomic_dec(&adapter->running_cap_crqs); 5503 netdev_dbg(adapter->netdev, "Outstanding request-caps: %d\n", 5504 atomic_read(&adapter->running_cap_crqs)); 5505 switch (be16_to_cpu(crq->request_capability_rsp.capability)) { 5506 case REQ_TX_QUEUES: 5507 req_value = &adapter->req_tx_queues; 5508 name = "tx"; 5509 break; 5510 case REQ_RX_QUEUES: 5511 req_value = &adapter->req_rx_queues; 5512 name = "rx"; 5513 break; 5514 case REQ_RX_ADD_QUEUES: 5515 req_value = &adapter->req_rx_add_queues; 5516 name = "rx_add"; 5517 break; 5518 case REQ_TX_ENTRIES_PER_SUBCRQ: 5519 req_value = &adapter->req_tx_entries_per_subcrq; 5520 name = "tx_entries_per_subcrq"; 5521 break; 5522 case REQ_RX_ADD_ENTRIES_PER_SUBCRQ: 5523 req_value = &adapter->req_rx_add_entries_per_subcrq; 5524 name = "rx_add_entries_per_subcrq"; 5525 break; 5526 case REQ_MTU: 5527 req_value = &adapter->req_mtu; 5528 name = "mtu"; 5529 break; 5530 case PROMISC_REQUESTED: 5531 req_value = &adapter->promisc; 5532 name = "promisc"; 5533 break; 5534 default: 5535 dev_err(dev, "Got invalid cap request rsp %d\n", 5536 crq->request_capability.capability); 5537 return; 5538 } 5539 5540 switch (crq->request_capability_rsp.rc.code) { 5541 case SUCCESS: 5542 break; 5543 case PARTIALSUCCESS: 5544 dev_info(dev, "req=%lld, rsp=%ld in %s queue, retrying.\n", 5545 *req_value, 5546 (long)be64_to_cpu(crq->request_capability_rsp.number), 5547 name); 5548 5549 if (be16_to_cpu(crq->request_capability_rsp.capability) == 5550 REQ_MTU) { 5551 pr_err("mtu of %llu is not supported. Reverting.\n", 5552 *req_value); 5553 *req_value = adapter->fallback.mtu; 5554 } else { 5555 *req_value = 5556 be64_to_cpu(crq->request_capability_rsp.number); 5557 } 5558 5559 send_request_cap(adapter, 1); 5560 return; 5561 default: 5562 dev_err(dev, "Error %d in request cap rsp\n", 5563 crq->request_capability_rsp.rc.code); 5564 return; 5565 } 5566 5567 /* Done receiving requested capabilities, query IP offload support */ 5568 if (atomic_read(&adapter->running_cap_crqs) == 0) 5569 send_query_ip_offload(adapter); 5570 } 5571 5572 static int handle_login_rsp(union ibmvnic_crq *login_rsp_crq, 5573 struct ibmvnic_adapter *adapter) 5574 { 5575 struct device *dev = &adapter->vdev->dev; 5576 struct net_device *netdev = adapter->netdev; 5577 struct ibmvnic_login_rsp_buffer *login_rsp = adapter->login_rsp_buf; 5578 struct ibmvnic_login_buffer *login = adapter->login_buf; 5579 u64 *tx_handle_array; 5580 u64 *rx_handle_array; 5581 int num_tx_pools; 5582 int num_rx_pools; 5583 u64 *size_array; 5584 u32 rsp_len; 5585 int i; 5586 5587 /* CHECK: Test/set of login_pending does not need to be atomic 5588 * because only ibmvnic_tasklet tests/clears this. 5589 */ 5590 if (!adapter->login_pending) { 5591 netdev_warn(netdev, "Ignoring unexpected login response\n"); 5592 return 0; 5593 } 5594 adapter->login_pending = false; 5595 5596 /* If the number of queues requested can't be allocated by the 5597 * server, the login response will return with code 1. We will need 5598 * to resend the login buffer with fewer queues requested. 5599 */ 5600 if (login_rsp_crq->generic.rc.code) { 5601 adapter->init_done_rc = login_rsp_crq->generic.rc.code; 5602 complete(&adapter->init_done); 5603 return 0; 5604 } 5605 5606 if (adapter->failover_pending) { 5607 adapter->init_done_rc = -EAGAIN; 5608 netdev_dbg(netdev, "Failover pending, ignoring login response\n"); 5609 complete(&adapter->init_done); 5610 /* login response buffer will be released on reset */ 5611 return 0; 5612 } 5613 5614 netdev->mtu = adapter->req_mtu - ETH_HLEN; 5615 5616 netdev_dbg(adapter->netdev, "Login Response Buffer:\n"); 5617 ibmvnic_print_hex_dump(netdev, adapter->login_rsp_buf, 5618 adapter->login_rsp_buf_sz); 5619 5620 /* Sanity checks */ 5621 if (login->num_txcomp_subcrqs != login_rsp->num_txsubm_subcrqs || 5622 (be32_to_cpu(login->num_rxcomp_subcrqs) * 5623 adapter->req_rx_add_queues != 5624 be32_to_cpu(login_rsp->num_rxadd_subcrqs))) { 5625 dev_err(dev, "FATAL: Inconsistent login and login rsp\n"); 5626 ibmvnic_reset(adapter, VNIC_RESET_FATAL); 5627 return -EIO; 5628 } 5629 5630 rsp_len = be32_to_cpu(login_rsp->len); 5631 if (be32_to_cpu(login->login_rsp_len) < rsp_len || 5632 rsp_len <= be32_to_cpu(login_rsp->off_txsubm_subcrqs) || 5633 rsp_len <= be32_to_cpu(login_rsp->off_rxadd_subcrqs) || 5634 rsp_len <= be32_to_cpu(login_rsp->off_rxadd_buff_size) || 5635 rsp_len <= be32_to_cpu(login_rsp->off_supp_tx_desc)) { 5636 /* This can happen if a login request times out and there are 5637 * 2 outstanding login requests sent, the LOGIN_RSP crq 5638 * could have been for the older login request. So we are 5639 * parsing the newer response buffer which may be incomplete 5640 */ 5641 dev_err(dev, "FATAL: Login rsp offsets/lengths invalid\n"); 5642 ibmvnic_reset(adapter, VNIC_RESET_FATAL); 5643 return -EIO; 5644 } 5645 5646 size_array = (u64 *)((u8 *)(adapter->login_rsp_buf) + 5647 be32_to_cpu(adapter->login_rsp_buf->off_rxadd_buff_size)); 5648 /* variable buffer sizes are not supported, so just read the 5649 * first entry. 5650 */ 5651 adapter->cur_rx_buf_sz = be64_to_cpu(size_array[0]); 5652 5653 num_tx_pools = be32_to_cpu(adapter->login_rsp_buf->num_txsubm_subcrqs); 5654 num_rx_pools = be32_to_cpu(adapter->login_rsp_buf->num_rxadd_subcrqs); 5655 5656 tx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) + 5657 be32_to_cpu(adapter->login_rsp_buf->off_txsubm_subcrqs)); 5658 rx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) + 5659 be32_to_cpu(adapter->login_rsp_buf->off_rxadd_subcrqs)); 5660 5661 for (i = 0; i < num_tx_pools; i++) 5662 adapter->tx_scrq[i]->handle = tx_handle_array[i]; 5663 5664 for (i = 0; i < num_rx_pools; i++) 5665 adapter->rx_scrq[i]->handle = rx_handle_array[i]; 5666 5667 adapter->num_active_tx_scrqs = num_tx_pools; 5668 adapter->num_active_rx_scrqs = num_rx_pools; 5669 release_login_rsp_buffer(adapter); 5670 release_login_buffer(adapter); 5671 complete(&adapter->init_done); 5672 5673 return 0; 5674 } 5675 5676 static void handle_request_unmap_rsp(union ibmvnic_crq *crq, 5677 struct ibmvnic_adapter *adapter) 5678 { 5679 struct device *dev = &adapter->vdev->dev; 5680 long rc; 5681 5682 rc = crq->request_unmap_rsp.rc.code; 5683 if (rc) 5684 dev_err(dev, "Error %ld in REQUEST_UNMAP_RSP\n", rc); 5685 } 5686 5687 static void handle_query_map_rsp(union ibmvnic_crq *crq, 5688 struct ibmvnic_adapter *adapter) 5689 { 5690 struct net_device *netdev = adapter->netdev; 5691 struct device *dev = &adapter->vdev->dev; 5692 long rc; 5693 5694 rc = crq->query_map_rsp.rc.code; 5695 if (rc) { 5696 dev_err(dev, "Error %ld in QUERY_MAP_RSP\n", rc); 5697 return; 5698 } 5699 netdev_dbg(netdev, "page_size = %d\ntot_pages = %u\nfree_pages = %u\n", 5700 crq->query_map_rsp.page_size, 5701 __be32_to_cpu(crq->query_map_rsp.tot_pages), 5702 __be32_to_cpu(crq->query_map_rsp.free_pages)); 5703 } 5704 5705 static void handle_query_cap_rsp(union ibmvnic_crq *crq, 5706 struct ibmvnic_adapter *adapter) 5707 { 5708 struct net_device *netdev = adapter->netdev; 5709 struct device *dev = &adapter->vdev->dev; 5710 long rc; 5711 5712 atomic_dec(&adapter->running_cap_crqs); 5713 netdev_dbg(netdev, "Outstanding queries: %d\n", 5714 atomic_read(&adapter->running_cap_crqs)); 5715 rc = crq->query_capability.rc.code; 5716 if (rc) { 5717 dev_err(dev, "Error %ld in QUERY_CAP_RSP\n", rc); 5718 goto out; 5719 } 5720 5721 switch (be16_to_cpu(crq->query_capability.capability)) { 5722 case MIN_TX_QUEUES: 5723 adapter->min_tx_queues = 5724 be64_to_cpu(crq->query_capability.number); 5725 netdev_dbg(netdev, "min_tx_queues = %lld\n", 5726 adapter->min_tx_queues); 5727 break; 5728 case MIN_RX_QUEUES: 5729 adapter->min_rx_queues = 5730 be64_to_cpu(crq->query_capability.number); 5731 netdev_dbg(netdev, "min_rx_queues = %lld\n", 5732 adapter->min_rx_queues); 5733 break; 5734 case MIN_RX_ADD_QUEUES: 5735 adapter->min_rx_add_queues = 5736 be64_to_cpu(crq->query_capability.number); 5737 netdev_dbg(netdev, "min_rx_add_queues = %lld\n", 5738 adapter->min_rx_add_queues); 5739 break; 5740 case MAX_TX_QUEUES: 5741 adapter->max_tx_queues = 5742 be64_to_cpu(crq->query_capability.number); 5743 netdev_dbg(netdev, "max_tx_queues = %lld\n", 5744 adapter->max_tx_queues); 5745 break; 5746 case MAX_RX_QUEUES: 5747 adapter->max_rx_queues = 5748 be64_to_cpu(crq->query_capability.number); 5749 netdev_dbg(netdev, "max_rx_queues = %lld\n", 5750 adapter->max_rx_queues); 5751 break; 5752 case MAX_RX_ADD_QUEUES: 5753 adapter->max_rx_add_queues = 5754 be64_to_cpu(crq->query_capability.number); 5755 netdev_dbg(netdev, "max_rx_add_queues = %lld\n", 5756 adapter->max_rx_add_queues); 5757 break; 5758 case MIN_TX_ENTRIES_PER_SUBCRQ: 5759 adapter->min_tx_entries_per_subcrq = 5760 be64_to_cpu(crq->query_capability.number); 5761 netdev_dbg(netdev, "min_tx_entries_per_subcrq = %lld\n", 5762 adapter->min_tx_entries_per_subcrq); 5763 break; 5764 case MIN_RX_ADD_ENTRIES_PER_SUBCRQ: 5765 adapter->min_rx_add_entries_per_subcrq = 5766 be64_to_cpu(crq->query_capability.number); 5767 netdev_dbg(netdev, "min_rx_add_entrs_per_subcrq = %lld\n", 5768 adapter->min_rx_add_entries_per_subcrq); 5769 break; 5770 case MAX_TX_ENTRIES_PER_SUBCRQ: 5771 adapter->max_tx_entries_per_subcrq = 5772 be64_to_cpu(crq->query_capability.number); 5773 netdev_dbg(netdev, "max_tx_entries_per_subcrq = %lld\n", 5774 adapter->max_tx_entries_per_subcrq); 5775 break; 5776 case MAX_RX_ADD_ENTRIES_PER_SUBCRQ: 5777 adapter->max_rx_add_entries_per_subcrq = 5778 be64_to_cpu(crq->query_capability.number); 5779 netdev_dbg(netdev, "max_rx_add_entrs_per_subcrq = %lld\n", 5780 adapter->max_rx_add_entries_per_subcrq); 5781 break; 5782 case TCP_IP_OFFLOAD: 5783 adapter->tcp_ip_offload = 5784 be64_to_cpu(crq->query_capability.number); 5785 netdev_dbg(netdev, "tcp_ip_offload = %lld\n", 5786 adapter->tcp_ip_offload); 5787 break; 5788 case PROMISC_SUPPORTED: 5789 adapter->promisc_supported = 5790 be64_to_cpu(crq->query_capability.number); 5791 netdev_dbg(netdev, "promisc_supported = %lld\n", 5792 adapter->promisc_supported); 5793 break; 5794 case MIN_MTU: 5795 adapter->min_mtu = be64_to_cpu(crq->query_capability.number); 5796 netdev->min_mtu = adapter->min_mtu - ETH_HLEN; 5797 netdev_dbg(netdev, "min_mtu = %lld\n", adapter->min_mtu); 5798 break; 5799 case MAX_MTU: 5800 adapter->max_mtu = be64_to_cpu(crq->query_capability.number); 5801 netdev->max_mtu = adapter->max_mtu - ETH_HLEN; 5802 netdev_dbg(netdev, "max_mtu = %lld\n", adapter->max_mtu); 5803 break; 5804 case MAX_MULTICAST_FILTERS: 5805 adapter->max_multicast_filters = 5806 be64_to_cpu(crq->query_capability.number); 5807 netdev_dbg(netdev, "max_multicast_filters = %lld\n", 5808 adapter->max_multicast_filters); 5809 break; 5810 case VLAN_HEADER_INSERTION: 5811 adapter->vlan_header_insertion = 5812 be64_to_cpu(crq->query_capability.number); 5813 if (adapter->vlan_header_insertion) 5814 netdev->features |= NETIF_F_HW_VLAN_STAG_TX; 5815 netdev_dbg(netdev, "vlan_header_insertion = %lld\n", 5816 adapter->vlan_header_insertion); 5817 break; 5818 case RX_VLAN_HEADER_INSERTION: 5819 adapter->rx_vlan_header_insertion = 5820 be64_to_cpu(crq->query_capability.number); 5821 netdev_dbg(netdev, "rx_vlan_header_insertion = %lld\n", 5822 adapter->rx_vlan_header_insertion); 5823 break; 5824 case MAX_TX_SG_ENTRIES: 5825 adapter->max_tx_sg_entries = 5826 be64_to_cpu(crq->query_capability.number); 5827 netdev_dbg(netdev, "max_tx_sg_entries = %lld\n", 5828 adapter->max_tx_sg_entries); 5829 break; 5830 case RX_SG_SUPPORTED: 5831 adapter->rx_sg_supported = 5832 be64_to_cpu(crq->query_capability.number); 5833 netdev_dbg(netdev, "rx_sg_supported = %lld\n", 5834 adapter->rx_sg_supported); 5835 break; 5836 case OPT_TX_COMP_SUB_QUEUES: 5837 adapter->opt_tx_comp_sub_queues = 5838 be64_to_cpu(crq->query_capability.number); 5839 netdev_dbg(netdev, "opt_tx_comp_sub_queues = %lld\n", 5840 adapter->opt_tx_comp_sub_queues); 5841 break; 5842 case OPT_RX_COMP_QUEUES: 5843 adapter->opt_rx_comp_queues = 5844 be64_to_cpu(crq->query_capability.number); 5845 netdev_dbg(netdev, "opt_rx_comp_queues = %lld\n", 5846 adapter->opt_rx_comp_queues); 5847 break; 5848 case OPT_RX_BUFADD_Q_PER_RX_COMP_Q: 5849 adapter->opt_rx_bufadd_q_per_rx_comp_q = 5850 be64_to_cpu(crq->query_capability.number); 5851 netdev_dbg(netdev, "opt_rx_bufadd_q_per_rx_comp_q = %lld\n", 5852 adapter->opt_rx_bufadd_q_per_rx_comp_q); 5853 break; 5854 case OPT_TX_ENTRIES_PER_SUBCRQ: 5855 adapter->opt_tx_entries_per_subcrq = 5856 be64_to_cpu(crq->query_capability.number); 5857 netdev_dbg(netdev, "opt_tx_entries_per_subcrq = %lld\n", 5858 adapter->opt_tx_entries_per_subcrq); 5859 break; 5860 case OPT_RXBA_ENTRIES_PER_SUBCRQ: 5861 adapter->opt_rxba_entries_per_subcrq = 5862 be64_to_cpu(crq->query_capability.number); 5863 netdev_dbg(netdev, "opt_rxba_entries_per_subcrq = %lld\n", 5864 adapter->opt_rxba_entries_per_subcrq); 5865 break; 5866 case TX_RX_DESC_REQ: 5867 adapter->tx_rx_desc_req = crq->query_capability.number; 5868 netdev_dbg(netdev, "tx_rx_desc_req = %llx\n", 5869 adapter->tx_rx_desc_req); 5870 break; 5871 5872 default: 5873 netdev_err(netdev, "Got invalid cap rsp %d\n", 5874 crq->query_capability.capability); 5875 } 5876 5877 out: 5878 if (atomic_read(&adapter->running_cap_crqs) == 0) 5879 send_request_cap(adapter, 0); 5880 } 5881 5882 static int send_query_phys_parms(struct ibmvnic_adapter *adapter) 5883 { 5884 union ibmvnic_crq crq; 5885 int rc; 5886 5887 memset(&crq, 0, sizeof(crq)); 5888 crq.query_phys_parms.first = IBMVNIC_CRQ_CMD; 5889 crq.query_phys_parms.cmd = QUERY_PHYS_PARMS; 5890 5891 mutex_lock(&adapter->fw_lock); 5892 adapter->fw_done_rc = 0; 5893 reinit_completion(&adapter->fw_done); 5894 5895 rc = ibmvnic_send_crq(adapter, &crq); 5896 if (rc) { 5897 mutex_unlock(&adapter->fw_lock); 5898 return rc; 5899 } 5900 5901 rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000); 5902 if (rc) { 5903 mutex_unlock(&adapter->fw_lock); 5904 return rc; 5905 } 5906 5907 mutex_unlock(&adapter->fw_lock); 5908 return adapter->fw_done_rc ? -EIO : 0; 5909 } 5910 5911 static int handle_query_phys_parms_rsp(union ibmvnic_crq *crq, 5912 struct ibmvnic_adapter *adapter) 5913 { 5914 struct net_device *netdev = adapter->netdev; 5915 int rc; 5916 __be32 rspeed = cpu_to_be32(crq->query_phys_parms_rsp.speed); 5917 5918 rc = crq->query_phys_parms_rsp.rc.code; 5919 if (rc) { 5920 netdev_err(netdev, "Error %d in QUERY_PHYS_PARMS\n", rc); 5921 return rc; 5922 } 5923 switch (rspeed) { 5924 case IBMVNIC_10MBPS: 5925 adapter->speed = SPEED_10; 5926 break; 5927 case IBMVNIC_100MBPS: 5928 adapter->speed = SPEED_100; 5929 break; 5930 case IBMVNIC_1GBPS: 5931 adapter->speed = SPEED_1000; 5932 break; 5933 case IBMVNIC_10GBPS: 5934 adapter->speed = SPEED_10000; 5935 break; 5936 case IBMVNIC_25GBPS: 5937 adapter->speed = SPEED_25000; 5938 break; 5939 case IBMVNIC_40GBPS: 5940 adapter->speed = SPEED_40000; 5941 break; 5942 case IBMVNIC_50GBPS: 5943 adapter->speed = SPEED_50000; 5944 break; 5945 case IBMVNIC_100GBPS: 5946 adapter->speed = SPEED_100000; 5947 break; 5948 case IBMVNIC_200GBPS: 5949 adapter->speed = SPEED_200000; 5950 break; 5951 default: 5952 if (netif_carrier_ok(netdev)) 5953 netdev_warn(netdev, "Unknown speed 0x%08x\n", rspeed); 5954 adapter->speed = SPEED_UNKNOWN; 5955 } 5956 if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_FULL_DUPLEX) 5957 adapter->duplex = DUPLEX_FULL; 5958 else if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_HALF_DUPLEX) 5959 adapter->duplex = DUPLEX_HALF; 5960 else 5961 adapter->duplex = DUPLEX_UNKNOWN; 5962 5963 return rc; 5964 } 5965 5966 static void ibmvnic_handle_crq(union ibmvnic_crq *crq, 5967 struct ibmvnic_adapter *adapter) 5968 { 5969 struct ibmvnic_generic_crq *gen_crq = &crq->generic; 5970 struct net_device *netdev = adapter->netdev; 5971 struct device *dev = &adapter->vdev->dev; 5972 u64 *u64_crq = (u64 *)crq; 5973 long rc; 5974 5975 netdev_dbg(netdev, "Handling CRQ: %016lx %016lx\n", 5976 (unsigned long)cpu_to_be64(u64_crq[0]), 5977 (unsigned long)cpu_to_be64(u64_crq[1])); 5978 switch (gen_crq->first) { 5979 case IBMVNIC_CRQ_INIT_RSP: 5980 switch (gen_crq->cmd) { 5981 case IBMVNIC_CRQ_INIT: 5982 dev_info(dev, "Partner initialized\n"); 5983 adapter->from_passive_init = true; 5984 /* Discard any stale login responses from prev reset. 5985 * CHECK: should we clear even on INIT_COMPLETE? 5986 */ 5987 adapter->login_pending = false; 5988 5989 if (adapter->state == VNIC_DOWN) 5990 rc = ibmvnic_reset(adapter, VNIC_RESET_PASSIVE_INIT); 5991 else 5992 rc = ibmvnic_reset(adapter, VNIC_RESET_FAILOVER); 5993 5994 if (rc && rc != -EBUSY) { 5995 /* We were unable to schedule the failover 5996 * reset either because the adapter was still 5997 * probing (eg: during kexec) or we could not 5998 * allocate memory. Clear the failover_pending 5999 * flag since no one else will. We ignore 6000 * EBUSY because it means either FAILOVER reset 6001 * is already scheduled or the adapter is 6002 * being removed. 6003 */ 6004 netdev_err(netdev, 6005 "Error %ld scheduling failover reset\n", 6006 rc); 6007 adapter->failover_pending = false; 6008 } 6009 6010 if (!completion_done(&adapter->init_done)) { 6011 if (!adapter->init_done_rc) 6012 adapter->init_done_rc = -EAGAIN; 6013 complete(&adapter->init_done); 6014 } 6015 6016 break; 6017 case IBMVNIC_CRQ_INIT_COMPLETE: 6018 dev_info(dev, "Partner initialization complete\n"); 6019 adapter->crq.active = true; 6020 send_version_xchg(adapter); 6021 break; 6022 default: 6023 dev_err(dev, "Unknown crq cmd: %d\n", gen_crq->cmd); 6024 } 6025 return; 6026 case IBMVNIC_CRQ_XPORT_EVENT: 6027 netif_carrier_off(netdev); 6028 adapter->crq.active = false; 6029 /* terminate any thread waiting for a response 6030 * from the device 6031 */ 6032 if (!completion_done(&adapter->fw_done)) { 6033 adapter->fw_done_rc = -EIO; 6034 complete(&adapter->fw_done); 6035 } 6036 6037 /* if we got here during crq-init, retry crq-init */ 6038 if (!completion_done(&adapter->init_done)) { 6039 adapter->init_done_rc = -EAGAIN; 6040 complete(&adapter->init_done); 6041 } 6042 6043 if (!completion_done(&adapter->stats_done)) 6044 complete(&adapter->stats_done); 6045 if (test_bit(0, &adapter->resetting)) 6046 adapter->force_reset_recovery = true; 6047 if (gen_crq->cmd == IBMVNIC_PARTITION_MIGRATED) { 6048 dev_info(dev, "Migrated, re-enabling adapter\n"); 6049 ibmvnic_reset(adapter, VNIC_RESET_MOBILITY); 6050 } else if (gen_crq->cmd == IBMVNIC_DEVICE_FAILOVER) { 6051 dev_info(dev, "Backing device failover detected\n"); 6052 adapter->failover_pending = true; 6053 } else { 6054 /* The adapter lost the connection */ 6055 dev_err(dev, "Virtual Adapter failed (rc=%d)\n", 6056 gen_crq->cmd); 6057 ibmvnic_reset(adapter, VNIC_RESET_FATAL); 6058 } 6059 return; 6060 case IBMVNIC_CRQ_CMD_RSP: 6061 break; 6062 default: 6063 dev_err(dev, "Got an invalid msg type 0x%02x\n", 6064 gen_crq->first); 6065 return; 6066 } 6067 6068 switch (gen_crq->cmd) { 6069 case VERSION_EXCHANGE_RSP: 6070 rc = crq->version_exchange_rsp.rc.code; 6071 if (rc) { 6072 dev_err(dev, "Error %ld in VERSION_EXCHG_RSP\n", rc); 6073 break; 6074 } 6075 ibmvnic_version = 6076 be16_to_cpu(crq->version_exchange_rsp.version); 6077 dev_info(dev, "Partner protocol version is %d\n", 6078 ibmvnic_version); 6079 send_query_cap(adapter); 6080 break; 6081 case QUERY_CAPABILITY_RSP: 6082 handle_query_cap_rsp(crq, adapter); 6083 break; 6084 case QUERY_MAP_RSP: 6085 handle_query_map_rsp(crq, adapter); 6086 break; 6087 case REQUEST_MAP_RSP: 6088 adapter->fw_done_rc = crq->request_map_rsp.rc.code; 6089 complete(&adapter->fw_done); 6090 break; 6091 case REQUEST_UNMAP_RSP: 6092 handle_request_unmap_rsp(crq, adapter); 6093 break; 6094 case REQUEST_CAPABILITY_RSP: 6095 handle_request_cap_rsp(crq, adapter); 6096 break; 6097 case LOGIN_RSP: 6098 netdev_dbg(netdev, "Got Login Response\n"); 6099 handle_login_rsp(crq, adapter); 6100 break; 6101 case LOGICAL_LINK_STATE_RSP: 6102 netdev_dbg(netdev, 6103 "Got Logical Link State Response, state: %d rc: %d\n", 6104 crq->logical_link_state_rsp.link_state, 6105 crq->logical_link_state_rsp.rc.code); 6106 adapter->logical_link_state = 6107 crq->logical_link_state_rsp.link_state; 6108 adapter->init_done_rc = crq->logical_link_state_rsp.rc.code; 6109 complete(&adapter->init_done); 6110 break; 6111 case LINK_STATE_INDICATION: 6112 netdev_dbg(netdev, "Got Logical Link State Indication\n"); 6113 adapter->phys_link_state = 6114 crq->link_state_indication.phys_link_state; 6115 adapter->logical_link_state = 6116 crq->link_state_indication.logical_link_state; 6117 if (adapter->phys_link_state && adapter->logical_link_state) 6118 netif_carrier_on(netdev); 6119 else 6120 netif_carrier_off(netdev); 6121 break; 6122 case CHANGE_MAC_ADDR_RSP: 6123 netdev_dbg(netdev, "Got MAC address change Response\n"); 6124 adapter->fw_done_rc = handle_change_mac_rsp(crq, adapter); 6125 break; 6126 case ERROR_INDICATION: 6127 netdev_dbg(netdev, "Got Error Indication\n"); 6128 handle_error_indication(crq, adapter); 6129 break; 6130 case REQUEST_STATISTICS_RSP: 6131 netdev_dbg(netdev, "Got Statistics Response\n"); 6132 complete(&adapter->stats_done); 6133 break; 6134 case QUERY_IP_OFFLOAD_RSP: 6135 netdev_dbg(netdev, "Got Query IP offload Response\n"); 6136 handle_query_ip_offload_rsp(adapter); 6137 break; 6138 case MULTICAST_CTRL_RSP: 6139 netdev_dbg(netdev, "Got multicast control Response\n"); 6140 break; 6141 case CONTROL_IP_OFFLOAD_RSP: 6142 netdev_dbg(netdev, "Got Control IP offload Response\n"); 6143 dma_unmap_single(dev, adapter->ip_offload_ctrl_tok, 6144 sizeof(adapter->ip_offload_ctrl), 6145 DMA_TO_DEVICE); 6146 complete(&adapter->init_done); 6147 break; 6148 case COLLECT_FW_TRACE_RSP: 6149 netdev_dbg(netdev, "Got Collect firmware trace Response\n"); 6150 complete(&adapter->fw_done); 6151 break; 6152 case GET_VPD_SIZE_RSP: 6153 handle_vpd_size_rsp(crq, adapter); 6154 break; 6155 case GET_VPD_RSP: 6156 handle_vpd_rsp(crq, adapter); 6157 break; 6158 case QUERY_PHYS_PARMS_RSP: 6159 adapter->fw_done_rc = handle_query_phys_parms_rsp(crq, adapter); 6160 complete(&adapter->fw_done); 6161 break; 6162 default: 6163 netdev_err(netdev, "Got an invalid cmd type 0x%02x\n", 6164 gen_crq->cmd); 6165 } 6166 } 6167 6168 static irqreturn_t ibmvnic_interrupt(int irq, void *instance) 6169 { 6170 struct ibmvnic_adapter *adapter = instance; 6171 6172 tasklet_schedule(&adapter->tasklet); 6173 return IRQ_HANDLED; 6174 } 6175 6176 static void ibmvnic_tasklet(struct tasklet_struct *t) 6177 { 6178 struct ibmvnic_adapter *adapter = from_tasklet(adapter, t, tasklet); 6179 struct ibmvnic_crq_queue *queue = &adapter->crq; 6180 union ibmvnic_crq *crq; 6181 unsigned long flags; 6182 6183 spin_lock_irqsave(&queue->lock, flags); 6184 6185 /* Pull all the valid messages off the CRQ */ 6186 while ((crq = ibmvnic_next_crq(adapter)) != NULL) { 6187 /* This barrier makes sure ibmvnic_next_crq()'s 6188 * crq->generic.first & IBMVNIC_CRQ_CMD_RSP is loaded 6189 * before ibmvnic_handle_crq()'s 6190 * switch(gen_crq->first) and switch(gen_crq->cmd). 6191 */ 6192 dma_rmb(); 6193 ibmvnic_handle_crq(crq, adapter); 6194 crq->generic.first = 0; 6195 } 6196 6197 spin_unlock_irqrestore(&queue->lock, flags); 6198 } 6199 6200 static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *adapter) 6201 { 6202 struct vio_dev *vdev = adapter->vdev; 6203 int rc; 6204 6205 do { 6206 rc = plpar_hcall_norets(H_ENABLE_CRQ, vdev->unit_address); 6207 } while (rc == H_IN_PROGRESS || rc == H_BUSY || H_IS_LONG_BUSY(rc)); 6208 6209 if (rc) 6210 dev_err(&vdev->dev, "Error enabling adapter (rc=%d)\n", rc); 6211 6212 return rc; 6213 } 6214 6215 static int ibmvnic_reset_crq(struct ibmvnic_adapter *adapter) 6216 { 6217 struct ibmvnic_crq_queue *crq = &adapter->crq; 6218 struct device *dev = &adapter->vdev->dev; 6219 struct vio_dev *vdev = adapter->vdev; 6220 int rc; 6221 6222 /* Close the CRQ */ 6223 do { 6224 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address); 6225 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc)); 6226 6227 /* Clean out the queue */ 6228 if (!crq->msgs) 6229 return -EINVAL; 6230 6231 memset(crq->msgs, 0, PAGE_SIZE); 6232 crq->cur = 0; 6233 crq->active = false; 6234 6235 /* And re-open it again */ 6236 rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address, 6237 crq->msg_token, PAGE_SIZE); 6238 6239 if (rc == H_CLOSED) 6240 /* Adapter is good, but other end is not ready */ 6241 dev_warn(dev, "Partner adapter not ready\n"); 6242 else if (rc != 0) 6243 dev_warn(dev, "Couldn't register crq (rc=%d)\n", rc); 6244 6245 return rc; 6246 } 6247 6248 static void release_crq_queue(struct ibmvnic_adapter *adapter) 6249 { 6250 struct ibmvnic_crq_queue *crq = &adapter->crq; 6251 struct vio_dev *vdev = adapter->vdev; 6252 long rc; 6253 6254 if (!crq->msgs) 6255 return; 6256 6257 netdev_dbg(adapter->netdev, "Releasing CRQ\n"); 6258 free_irq(vdev->irq, adapter); 6259 tasklet_kill(&adapter->tasklet); 6260 do { 6261 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address); 6262 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc)); 6263 6264 dma_unmap_single(&vdev->dev, crq->msg_token, PAGE_SIZE, 6265 DMA_BIDIRECTIONAL); 6266 free_page((unsigned long)crq->msgs); 6267 crq->msgs = NULL; 6268 crq->active = false; 6269 } 6270 6271 static int init_crq_queue(struct ibmvnic_adapter *adapter) 6272 { 6273 struct ibmvnic_crq_queue *crq = &adapter->crq; 6274 struct device *dev = &adapter->vdev->dev; 6275 struct vio_dev *vdev = adapter->vdev; 6276 int rc, retrc = -ENOMEM; 6277 6278 if (crq->msgs) 6279 return 0; 6280 6281 crq->msgs = (union ibmvnic_crq *)get_zeroed_page(GFP_KERNEL); 6282 /* Should we allocate more than one page? */ 6283 6284 if (!crq->msgs) 6285 return -ENOMEM; 6286 6287 crq->size = PAGE_SIZE / sizeof(*crq->msgs); 6288 crq->msg_token = dma_map_single(dev, crq->msgs, PAGE_SIZE, 6289 DMA_BIDIRECTIONAL); 6290 if (dma_mapping_error(dev, crq->msg_token)) 6291 goto map_failed; 6292 6293 rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address, 6294 crq->msg_token, PAGE_SIZE); 6295 6296 if (rc == H_RESOURCE) 6297 /* maybe kexecing and resource is busy. try a reset */ 6298 rc = ibmvnic_reset_crq(adapter); 6299 retrc = rc; 6300 6301 if (rc == H_CLOSED) { 6302 dev_warn(dev, "Partner adapter not ready\n"); 6303 } else if (rc) { 6304 dev_warn(dev, "Error %d opening adapter\n", rc); 6305 goto reg_crq_failed; 6306 } 6307 6308 retrc = 0; 6309 6310 tasklet_setup(&adapter->tasklet, (void *)ibmvnic_tasklet); 6311 6312 netdev_dbg(adapter->netdev, "registering irq 0x%x\n", vdev->irq); 6313 snprintf(crq->name, sizeof(crq->name), "ibmvnic-%x", 6314 adapter->vdev->unit_address); 6315 rc = request_irq(vdev->irq, ibmvnic_interrupt, 0, crq->name, adapter); 6316 if (rc) { 6317 dev_err(dev, "Couldn't register irq 0x%x. rc=%d\n", 6318 vdev->irq, rc); 6319 goto req_irq_failed; 6320 } 6321 6322 rc = vio_enable_interrupts(vdev); 6323 if (rc) { 6324 dev_err(dev, "Error %d enabling interrupts\n", rc); 6325 goto req_irq_failed; 6326 } 6327 6328 crq->cur = 0; 6329 spin_lock_init(&crq->lock); 6330 6331 /* process any CRQs that were queued before we enabled interrupts */ 6332 tasklet_schedule(&adapter->tasklet); 6333 6334 return retrc; 6335 6336 req_irq_failed: 6337 tasklet_kill(&adapter->tasklet); 6338 do { 6339 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address); 6340 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc)); 6341 reg_crq_failed: 6342 dma_unmap_single(dev, crq->msg_token, PAGE_SIZE, DMA_BIDIRECTIONAL); 6343 map_failed: 6344 free_page((unsigned long)crq->msgs); 6345 crq->msgs = NULL; 6346 return retrc; 6347 } 6348 6349 static int ibmvnic_reset_init(struct ibmvnic_adapter *adapter, bool reset) 6350 { 6351 struct device *dev = &adapter->vdev->dev; 6352 unsigned long timeout = msecs_to_jiffies(20000); 6353 u64 old_num_rx_queues = adapter->req_rx_queues; 6354 u64 old_num_tx_queues = adapter->req_tx_queues; 6355 int rc; 6356 6357 adapter->from_passive_init = false; 6358 6359 rc = ibmvnic_send_crq_init(adapter); 6360 if (rc) { 6361 dev_err(dev, "Send crq init failed with error %d\n", rc); 6362 return rc; 6363 } 6364 6365 if (!wait_for_completion_timeout(&adapter->init_done, timeout)) { 6366 dev_err(dev, "Initialization sequence timed out\n"); 6367 return -ETIMEDOUT; 6368 } 6369 6370 if (adapter->init_done_rc) { 6371 release_crq_queue(adapter); 6372 dev_err(dev, "CRQ-init failed, %d\n", adapter->init_done_rc); 6373 return adapter->init_done_rc; 6374 } 6375 6376 if (adapter->from_passive_init) { 6377 adapter->state = VNIC_OPEN; 6378 adapter->from_passive_init = false; 6379 dev_err(dev, "CRQ-init failed, passive-init\n"); 6380 return -EINVAL; 6381 } 6382 6383 if (reset && 6384 test_bit(0, &adapter->resetting) && !adapter->wait_for_reset && 6385 adapter->reset_reason != VNIC_RESET_MOBILITY) { 6386 if (adapter->req_rx_queues != old_num_rx_queues || 6387 adapter->req_tx_queues != old_num_tx_queues) { 6388 release_sub_crqs(adapter, 0); 6389 rc = init_sub_crqs(adapter); 6390 } else { 6391 /* no need to reinitialize completely, but we do 6392 * need to clean up transmits that were in flight 6393 * when we processed the reset. Failure to do so 6394 * will confound the upper layer, usually TCP, by 6395 * creating the illusion of transmits that are 6396 * awaiting completion. 6397 */ 6398 clean_tx_pools(adapter); 6399 6400 rc = reset_sub_crq_queues(adapter); 6401 } 6402 } else { 6403 if (adapter->reset_reason == VNIC_RESET_MOBILITY) { 6404 /* After an LPM, reset the max number of indirect 6405 * subcrq descriptors per H_SEND_SUB_CRQ_INDIRECT 6406 * hcall to the default max (e.g POWER8 -> POWER10) 6407 * 6408 * If the new destination platform does not support 6409 * the higher limit max (e.g. POWER10-> POWER8 LPM) 6410 * H_PARAMETER will trigger automatic fallback to the 6411 * safe minimum limit. 6412 */ 6413 adapter->cur_max_ind_descs = IBMVNIC_MAX_IND_DESCS; 6414 } 6415 6416 rc = init_sub_crqs(adapter); 6417 } 6418 6419 if (rc) { 6420 dev_err(dev, "Initialization of sub crqs failed\n"); 6421 release_crq_queue(adapter); 6422 return rc; 6423 } 6424 6425 rc = init_sub_crq_irqs(adapter); 6426 if (rc) { 6427 dev_err(dev, "Failed to initialize sub crq irqs\n"); 6428 release_crq_queue(adapter); 6429 } 6430 6431 return rc; 6432 } 6433 6434 static struct device_attribute dev_attr_failover; 6435 6436 static int ibmvnic_probe(struct vio_dev *dev, const struct vio_device_id *id) 6437 { 6438 struct ibmvnic_adapter *adapter; 6439 struct net_device *netdev; 6440 unsigned char *mac_addr_p; 6441 unsigned long flags; 6442 bool init_success; 6443 int rc; 6444 6445 dev_dbg(&dev->dev, "entering ibmvnic_probe for UA 0x%x\n", 6446 dev->unit_address); 6447 6448 mac_addr_p = (unsigned char *)vio_get_attribute(dev, 6449 VETH_MAC_ADDR, NULL); 6450 if (!mac_addr_p) { 6451 dev_err(&dev->dev, 6452 "(%s:%3.3d) ERROR: Can't find MAC_ADDR attribute\n", 6453 __FILE__, __LINE__); 6454 return 0; 6455 } 6456 6457 netdev = alloc_etherdev_mq(sizeof(struct ibmvnic_adapter), 6458 IBMVNIC_MAX_QUEUES); 6459 if (!netdev) 6460 return -ENOMEM; 6461 6462 adapter = netdev_priv(netdev); 6463 adapter->state = VNIC_PROBING; 6464 dev_set_drvdata(&dev->dev, netdev); 6465 adapter->vdev = dev; 6466 adapter->netdev = netdev; 6467 adapter->login_pending = false; 6468 memset(&adapter->map_ids, 0, sizeof(adapter->map_ids)); 6469 /* map_ids start at 1, so ensure map_id 0 is always "in-use" */ 6470 bitmap_set(adapter->map_ids, 0, 1); 6471 6472 ether_addr_copy(adapter->mac_addr, mac_addr_p); 6473 eth_hw_addr_set(netdev, adapter->mac_addr); 6474 netdev->irq = dev->irq; 6475 netdev->netdev_ops = &ibmvnic_netdev_ops; 6476 netdev->ethtool_ops = &ibmvnic_ethtool_ops; 6477 SET_NETDEV_DEV(netdev, &dev->dev); 6478 6479 INIT_WORK(&adapter->ibmvnic_reset, __ibmvnic_reset); 6480 INIT_DELAYED_WORK(&adapter->ibmvnic_delayed_reset, 6481 __ibmvnic_delayed_reset); 6482 INIT_LIST_HEAD(&adapter->rwi_list); 6483 spin_lock_init(&adapter->rwi_lock); 6484 spin_lock_init(&adapter->state_lock); 6485 mutex_init(&adapter->fw_lock); 6486 init_completion(&adapter->probe_done); 6487 init_completion(&adapter->init_done); 6488 init_completion(&adapter->fw_done); 6489 init_completion(&adapter->reset_done); 6490 init_completion(&adapter->stats_done); 6491 clear_bit(0, &adapter->resetting); 6492 adapter->prev_rx_buf_sz = 0; 6493 adapter->prev_mtu = 0; 6494 6495 init_success = false; 6496 do { 6497 reinit_init_done(adapter); 6498 6499 /* clear any failovers we got in the previous pass 6500 * since we are reinitializing the CRQ 6501 */ 6502 adapter->failover_pending = false; 6503 6504 /* If we had already initialized CRQ, we may have one or 6505 * more resets queued already. Discard those and release 6506 * the CRQ before initializing the CRQ again. 6507 */ 6508 release_crq_queue(adapter); 6509 6510 /* Since we are still in PROBING state, __ibmvnic_reset() 6511 * will not access the ->rwi_list and since we released CRQ, 6512 * we won't get _new_ transport events. But there maybe an 6513 * ongoing ibmvnic_reset() call. So serialize access to 6514 * rwi_list. If we win the race, ibvmnic_reset() could add 6515 * a reset after we purged but thats ok - we just may end 6516 * up with an extra reset (i.e similar to having two or more 6517 * resets in the queue at once). 6518 * CHECK. 6519 */ 6520 spin_lock_irqsave(&adapter->rwi_lock, flags); 6521 flush_reset_queue(adapter); 6522 spin_unlock_irqrestore(&adapter->rwi_lock, flags); 6523 6524 rc = init_crq_queue(adapter); 6525 if (rc) { 6526 dev_err(&dev->dev, "Couldn't initialize crq. rc=%d\n", 6527 rc); 6528 goto ibmvnic_init_fail; 6529 } 6530 6531 rc = ibmvnic_reset_init(adapter, false); 6532 } while (rc == -EAGAIN); 6533 6534 /* We are ignoring the error from ibmvnic_reset_init() assuming that the 6535 * partner is not ready. CRQ is not active. When the partner becomes 6536 * ready, we will do the passive init reset. 6537 */ 6538 6539 if (!rc) 6540 init_success = true; 6541 6542 rc = init_stats_buffers(adapter); 6543 if (rc) 6544 goto ibmvnic_init_fail; 6545 6546 rc = init_stats_token(adapter); 6547 if (rc) 6548 goto ibmvnic_stats_fail; 6549 6550 rc = device_create_file(&dev->dev, &dev_attr_failover); 6551 if (rc) 6552 goto ibmvnic_dev_file_err; 6553 6554 netif_carrier_off(netdev); 6555 6556 if (init_success) { 6557 adapter->state = VNIC_PROBED; 6558 netdev->mtu = adapter->req_mtu - ETH_HLEN; 6559 netdev->min_mtu = adapter->min_mtu - ETH_HLEN; 6560 netdev->max_mtu = adapter->max_mtu - ETH_HLEN; 6561 } else { 6562 adapter->state = VNIC_DOWN; 6563 } 6564 6565 adapter->wait_for_reset = false; 6566 adapter->last_reset_time = jiffies; 6567 adapter->cur_max_ind_descs = IBMVNIC_MAX_IND_DESCS; 6568 6569 rc = register_netdev(netdev); 6570 if (rc) { 6571 dev_err(&dev->dev, "failed to register netdev rc=%d\n", rc); 6572 goto ibmvnic_register_fail; 6573 } 6574 dev_info(&dev->dev, "ibmvnic registered\n"); 6575 6576 rc = ibmvnic_cpu_notif_add(adapter); 6577 if (rc) { 6578 netdev_err(netdev, "Registering cpu notifier failed\n"); 6579 goto cpu_notif_add_failed; 6580 } 6581 6582 complete(&adapter->probe_done); 6583 6584 return 0; 6585 6586 cpu_notif_add_failed: 6587 unregister_netdev(netdev); 6588 6589 ibmvnic_register_fail: 6590 device_remove_file(&dev->dev, &dev_attr_failover); 6591 6592 ibmvnic_dev_file_err: 6593 release_stats_token(adapter); 6594 6595 ibmvnic_stats_fail: 6596 release_stats_buffers(adapter); 6597 6598 ibmvnic_init_fail: 6599 release_sub_crqs(adapter, 1); 6600 release_crq_queue(adapter); 6601 6602 /* cleanup worker thread after releasing CRQ so we don't get 6603 * transport events (i.e new work items for the worker thread). 6604 */ 6605 adapter->state = VNIC_REMOVING; 6606 complete(&adapter->probe_done); 6607 flush_work(&adapter->ibmvnic_reset); 6608 flush_delayed_work(&adapter->ibmvnic_delayed_reset); 6609 6610 flush_reset_queue(adapter); 6611 6612 mutex_destroy(&adapter->fw_lock); 6613 free_netdev(netdev); 6614 6615 return rc; 6616 } 6617 6618 static void ibmvnic_remove(struct vio_dev *dev) 6619 { 6620 struct net_device *netdev = dev_get_drvdata(&dev->dev); 6621 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 6622 unsigned long flags; 6623 6624 spin_lock_irqsave(&adapter->state_lock, flags); 6625 6626 /* If ibmvnic_reset() is scheduling a reset, wait for it to 6627 * finish. Then, set the state to REMOVING to prevent it from 6628 * scheduling any more work and to have reset functions ignore 6629 * any resets that have already been scheduled. Drop the lock 6630 * after setting state, so __ibmvnic_reset() which is called 6631 * from the flush_work() below, can make progress. 6632 */ 6633 spin_lock(&adapter->rwi_lock); 6634 adapter->state = VNIC_REMOVING; 6635 spin_unlock(&adapter->rwi_lock); 6636 6637 spin_unlock_irqrestore(&adapter->state_lock, flags); 6638 6639 ibmvnic_cpu_notif_remove(adapter); 6640 6641 flush_work(&adapter->ibmvnic_reset); 6642 flush_delayed_work(&adapter->ibmvnic_delayed_reset); 6643 6644 rtnl_lock(); 6645 unregister_netdevice(netdev); 6646 6647 release_resources(adapter); 6648 release_rx_pools(adapter); 6649 release_tx_pools(adapter); 6650 release_sub_crqs(adapter, 1); 6651 release_crq_queue(adapter); 6652 6653 release_stats_token(adapter); 6654 release_stats_buffers(adapter); 6655 6656 adapter->state = VNIC_REMOVED; 6657 6658 rtnl_unlock(); 6659 mutex_destroy(&adapter->fw_lock); 6660 device_remove_file(&dev->dev, &dev_attr_failover); 6661 free_netdev(netdev); 6662 dev_set_drvdata(&dev->dev, NULL); 6663 } 6664 6665 static ssize_t failover_store(struct device *dev, struct device_attribute *attr, 6666 const char *buf, size_t count) 6667 { 6668 struct net_device *netdev = dev_get_drvdata(dev); 6669 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 6670 unsigned long retbuf[PLPAR_HCALL_BUFSIZE]; 6671 __be64 session_token; 6672 long rc; 6673 6674 if (!sysfs_streq(buf, "1")) 6675 return -EINVAL; 6676 6677 rc = plpar_hcall(H_VIOCTL, retbuf, adapter->vdev->unit_address, 6678 H_GET_SESSION_TOKEN, 0, 0, 0); 6679 if (rc) { 6680 netdev_err(netdev, "Couldn't retrieve session token, rc %ld\n", 6681 rc); 6682 goto last_resort; 6683 } 6684 6685 session_token = (__be64)retbuf[0]; 6686 netdev_dbg(netdev, "Initiating client failover, session id %llx\n", 6687 be64_to_cpu(session_token)); 6688 rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address, 6689 H_SESSION_ERR_DETECTED, session_token, 0, 0); 6690 if (rc) { 6691 netdev_err(netdev, 6692 "H_VIOCTL initiated failover failed, rc %ld\n", 6693 rc); 6694 goto last_resort; 6695 } 6696 6697 return count; 6698 6699 last_resort: 6700 netdev_dbg(netdev, "Trying to send CRQ_CMD, the last resort\n"); 6701 ibmvnic_reset(adapter, VNIC_RESET_FAILOVER); 6702 6703 return count; 6704 } 6705 static DEVICE_ATTR_WO(failover); 6706 6707 static unsigned long ibmvnic_get_desired_dma(struct vio_dev *vdev) 6708 { 6709 struct net_device *netdev = dev_get_drvdata(&vdev->dev); 6710 struct ibmvnic_adapter *adapter; 6711 struct iommu_table *tbl; 6712 unsigned long ret = 0; 6713 int i; 6714 6715 tbl = get_iommu_table_base(&vdev->dev); 6716 6717 /* netdev inits at probe time along with the structures we need below*/ 6718 if (!netdev) 6719 return IOMMU_PAGE_ALIGN(IBMVNIC_IO_ENTITLEMENT_DEFAULT, tbl); 6720 6721 adapter = netdev_priv(netdev); 6722 6723 ret += PAGE_SIZE; /* the crq message queue */ 6724 ret += IOMMU_PAGE_ALIGN(sizeof(struct ibmvnic_statistics), tbl); 6725 6726 for (i = 0; i < adapter->req_tx_queues + adapter->req_rx_queues; i++) 6727 ret += 4 * PAGE_SIZE; /* the scrq message queue */ 6728 6729 for (i = 0; i < adapter->num_active_rx_pools; i++) 6730 ret += adapter->rx_pool[i].size * 6731 IOMMU_PAGE_ALIGN(adapter->rx_pool[i].buff_size, tbl); 6732 6733 return ret; 6734 } 6735 6736 static int ibmvnic_resume(struct device *dev) 6737 { 6738 struct net_device *netdev = dev_get_drvdata(dev); 6739 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 6740 6741 if (adapter->state != VNIC_OPEN) 6742 return 0; 6743 6744 tasklet_schedule(&adapter->tasklet); 6745 6746 return 0; 6747 } 6748 6749 static const struct vio_device_id ibmvnic_device_table[] = { 6750 {"network", "IBM,vnic"}, 6751 {"", "" } 6752 }; 6753 MODULE_DEVICE_TABLE(vio, ibmvnic_device_table); 6754 6755 static const struct dev_pm_ops ibmvnic_pm_ops = { 6756 .resume = ibmvnic_resume 6757 }; 6758 6759 static struct vio_driver ibmvnic_driver = { 6760 .id_table = ibmvnic_device_table, 6761 .probe = ibmvnic_probe, 6762 .remove = ibmvnic_remove, 6763 .get_desired_dma = ibmvnic_get_desired_dma, 6764 .name = ibmvnic_driver_name, 6765 .pm = &ibmvnic_pm_ops, 6766 }; 6767 6768 /* module functions */ 6769 static int __init ibmvnic_module_init(void) 6770 { 6771 int ret; 6772 6773 ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN, "net/ibmvnic:online", 6774 ibmvnic_cpu_online, 6775 ibmvnic_cpu_down_prep); 6776 if (ret < 0) 6777 goto out; 6778 ibmvnic_online = ret; 6779 ret = cpuhp_setup_state_multi(CPUHP_IBMVNIC_DEAD, "net/ibmvnic:dead", 6780 NULL, ibmvnic_cpu_dead); 6781 if (ret) 6782 goto err_dead; 6783 6784 ret = vio_register_driver(&ibmvnic_driver); 6785 if (ret) 6786 goto err_vio_register; 6787 6788 pr_info("%s: %s %s\n", ibmvnic_driver_name, ibmvnic_driver_string, 6789 IBMVNIC_DRIVER_VERSION); 6790 6791 return 0; 6792 err_vio_register: 6793 cpuhp_remove_multi_state(CPUHP_IBMVNIC_DEAD); 6794 err_dead: 6795 cpuhp_remove_multi_state(ibmvnic_online); 6796 out: 6797 return ret; 6798 } 6799 6800 static void __exit ibmvnic_module_exit(void) 6801 { 6802 vio_unregister_driver(&ibmvnic_driver); 6803 cpuhp_remove_multi_state(CPUHP_IBMVNIC_DEAD); 6804 cpuhp_remove_multi_state(ibmvnic_online); 6805 } 6806 6807 module_init(ibmvnic_module_init); 6808 module_exit(ibmvnic_module_exit); 6809