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