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