1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /**************************************************************************/ 3 /* */ 4 /* IBM System i and System p Virtual NIC Device Driver */ 5 /* Copyright (C) 2014 IBM Corp. */ 6 /* Santiago Leon (santi_leon@yahoo.com) */ 7 /* Thomas Falcon (tlfalcon@linux.vnet.ibm.com) */ 8 /* John Allen (jallen@linux.vnet.ibm.com) */ 9 /* */ 10 /* */ 11 /* This module contains the implementation of a virtual ethernet device */ 12 /* for use with IBM i/p Series LPAR Linux. It utilizes the logical LAN */ 13 /* option of the RS/6000 Platform Architecture to interface with virtual */ 14 /* ethernet NICs that are presented to the partition by the hypervisor. */ 15 /* */ 16 /* Messages are passed between the VNIC driver and the VNIC server using */ 17 /* Command/Response Queues (CRQs) and sub CRQs (sCRQs). CRQs are used to */ 18 /* issue and receive commands that initiate communication with the server */ 19 /* on driver initialization. Sub CRQs (sCRQs) are similar to CRQs, but */ 20 /* are used by the driver to notify the server that a packet is */ 21 /* ready for transmission or that a buffer has been added to receive a */ 22 /* packet. Subsequently, sCRQs are used by the server to notify the */ 23 /* driver that a packet transmission has been completed or that a packet */ 24 /* has been received and placed in a waiting buffer. */ 25 /* */ 26 /* In lieu of a more conventional "on-the-fly" DMA mapping strategy in */ 27 /* which skbs are DMA mapped and immediately unmapped when the transmit */ 28 /* or receive has been completed, the VNIC driver is required to use */ 29 /* "long term mapping". This entails that large, continuous DMA mapped */ 30 /* buffers are allocated on driver initialization and these buffers are */ 31 /* then continuously reused to pass skbs to and from the VNIC server. */ 32 /* */ 33 /**************************************************************************/ 34 35 #include <linux/module.h> 36 #include <linux/moduleparam.h> 37 #include <linux/types.h> 38 #include <linux/errno.h> 39 #include <linux/completion.h> 40 #include <linux/ioport.h> 41 #include <linux/dma-mapping.h> 42 #include <linux/kernel.h> 43 #include <linux/netdevice.h> 44 #include <linux/etherdevice.h> 45 #include <linux/skbuff.h> 46 #include <linux/init.h> 47 #include <linux/delay.h> 48 #include <linux/mm.h> 49 #include <linux/ethtool.h> 50 #include <linux/proc_fs.h> 51 #include <linux/if_arp.h> 52 #include <linux/in.h> 53 #include <linux/ip.h> 54 #include <linux/ipv6.h> 55 #include <linux/irq.h> 56 #include <linux/irqdomain.h> 57 #include <linux/kthread.h> 58 #include <linux/seq_file.h> 59 #include <linux/interrupt.h> 60 #include <net/net_namespace.h> 61 #include <asm/hvcall.h> 62 #include <linux/atomic.h> 63 #include <asm/vio.h> 64 #include <asm/xive.h> 65 #include <asm/iommu.h> 66 #include <linux/uaccess.h> 67 #include <asm/firmware.h> 68 #include <linux/workqueue.h> 69 #include <linux/if_vlan.h> 70 #include <linux/utsname.h> 71 #include <linux/cpu.h> 72 73 #include "ibmvnic.h" 74 75 static const char ibmvnic_driver_name[] = "ibmvnic"; 76 static const char ibmvnic_driver_string[] = "IBM System i/p Virtual NIC Driver"; 77 78 MODULE_AUTHOR("Santiago Leon"); 79 MODULE_DESCRIPTION("IBM System i/p Virtual NIC Driver"); 80 MODULE_LICENSE("GPL"); 81 MODULE_VERSION(IBMVNIC_DRIVER_VERSION); 82 83 static int ibmvnic_version = IBMVNIC_INITIAL_VERSION; 84 static void release_sub_crqs(struct ibmvnic_adapter *, bool); 85 static int ibmvnic_reset_crq(struct ibmvnic_adapter *); 86 static int ibmvnic_send_crq_init(struct ibmvnic_adapter *); 87 static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *); 88 static int ibmvnic_send_crq(struct ibmvnic_adapter *, union ibmvnic_crq *); 89 static int send_subcrq_indirect(struct ibmvnic_adapter *, u64, u64, u64); 90 static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance); 91 static int enable_scrq_irq(struct ibmvnic_adapter *, 92 struct ibmvnic_sub_crq_queue *); 93 static int disable_scrq_irq(struct ibmvnic_adapter *, 94 struct ibmvnic_sub_crq_queue *); 95 static int pending_scrq(struct ibmvnic_adapter *, 96 struct ibmvnic_sub_crq_queue *); 97 static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *, 98 struct ibmvnic_sub_crq_queue *); 99 static int ibmvnic_poll(struct napi_struct *napi, int data); 100 static int reset_sub_crq_queues(struct ibmvnic_adapter *adapter); 101 static inline void reinit_init_done(struct ibmvnic_adapter *adapter); 102 static void send_query_map(struct ibmvnic_adapter *adapter); 103 static int send_request_map(struct ibmvnic_adapter *, dma_addr_t, u32, u8); 104 static int send_request_unmap(struct ibmvnic_adapter *, u8); 105 static int send_login(struct ibmvnic_adapter *adapter); 106 static void send_query_cap(struct ibmvnic_adapter *adapter); 107 static int init_sub_crqs(struct ibmvnic_adapter *); 108 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter); 109 static int ibmvnic_reset_init(struct ibmvnic_adapter *, bool reset); 110 static void release_crq_queue(struct ibmvnic_adapter *); 111 static int __ibmvnic_set_mac(struct net_device *, u8 *); 112 static int init_crq_queue(struct ibmvnic_adapter *adapter); 113 static int send_query_phys_parms(struct ibmvnic_adapter *adapter); 114 static void ibmvnic_tx_scrq_clean_buffer(struct ibmvnic_adapter *adapter, 115 struct ibmvnic_sub_crq_queue *tx_scrq); 116 static void free_long_term_buff(struct ibmvnic_adapter *adapter, 117 struct ibmvnic_long_term_buff *ltb); 118 static void ibmvnic_disable_irqs(struct ibmvnic_adapter *adapter); 119 static void flush_reset_queue(struct ibmvnic_adapter *adapter); 120 static void print_subcrq_error(struct device *dev, int rc, const char *func); 121 122 struct ibmvnic_stat { 123 char name[ETH_GSTRING_LEN]; 124 int offset; 125 }; 126 127 #define IBMVNIC_STAT_OFF(stat) (offsetof(struct ibmvnic_adapter, stats) + \ 128 offsetof(struct ibmvnic_statistics, stat)) 129 #define IBMVNIC_GET_STAT(a, off) (*((u64 *)(((unsigned long)(a)) + (off)))) 130 131 static const struct ibmvnic_stat ibmvnic_stats[] = { 132 {"rx_packets", IBMVNIC_STAT_OFF(rx_packets)}, 133 {"rx_bytes", IBMVNIC_STAT_OFF(rx_bytes)}, 134 {"tx_packets", IBMVNIC_STAT_OFF(tx_packets)}, 135 {"tx_bytes", IBMVNIC_STAT_OFF(tx_bytes)}, 136 {"ucast_tx_packets", IBMVNIC_STAT_OFF(ucast_tx_packets)}, 137 {"ucast_rx_packets", IBMVNIC_STAT_OFF(ucast_rx_packets)}, 138 {"mcast_tx_packets", IBMVNIC_STAT_OFF(mcast_tx_packets)}, 139 {"mcast_rx_packets", IBMVNIC_STAT_OFF(mcast_rx_packets)}, 140 {"bcast_tx_packets", IBMVNIC_STAT_OFF(bcast_tx_packets)}, 141 {"bcast_rx_packets", IBMVNIC_STAT_OFF(bcast_rx_packets)}, 142 {"align_errors", IBMVNIC_STAT_OFF(align_errors)}, 143 {"fcs_errors", IBMVNIC_STAT_OFF(fcs_errors)}, 144 {"single_collision_frames", IBMVNIC_STAT_OFF(single_collision_frames)}, 145 {"multi_collision_frames", IBMVNIC_STAT_OFF(multi_collision_frames)}, 146 {"sqe_test_errors", IBMVNIC_STAT_OFF(sqe_test_errors)}, 147 {"deferred_tx", IBMVNIC_STAT_OFF(deferred_tx)}, 148 {"late_collisions", IBMVNIC_STAT_OFF(late_collisions)}, 149 {"excess_collisions", IBMVNIC_STAT_OFF(excess_collisions)}, 150 {"internal_mac_tx_errors", IBMVNIC_STAT_OFF(internal_mac_tx_errors)}, 151 {"carrier_sense", IBMVNIC_STAT_OFF(carrier_sense)}, 152 {"too_long_frames", IBMVNIC_STAT_OFF(too_long_frames)}, 153 {"internal_mac_rx_errors", IBMVNIC_STAT_OFF(internal_mac_rx_errors)}, 154 }; 155 156 static int send_crq_init_complete(struct ibmvnic_adapter *adapter) 157 { 158 union ibmvnic_crq crq; 159 160 memset(&crq, 0, sizeof(crq)); 161 crq.generic.first = IBMVNIC_CRQ_INIT_CMD; 162 crq.generic.cmd = IBMVNIC_CRQ_INIT_COMPLETE; 163 164 return ibmvnic_send_crq(adapter, &crq); 165 } 166 167 static int send_version_xchg(struct ibmvnic_adapter *adapter) 168 { 169 union ibmvnic_crq crq; 170 171 memset(&crq, 0, sizeof(crq)); 172 crq.version_exchange.first = IBMVNIC_CRQ_CMD; 173 crq.version_exchange.cmd = VERSION_EXCHANGE; 174 crq.version_exchange.version = cpu_to_be16(ibmvnic_version); 175 176 return ibmvnic_send_crq(adapter, &crq); 177 } 178 179 static void ibmvnic_clean_queue_affinity(struct ibmvnic_adapter *adapter, 180 struct ibmvnic_sub_crq_queue *queue) 181 { 182 if (!(queue && queue->irq)) 183 return; 184 185 cpumask_clear(queue->affinity_mask); 186 187 if (irq_set_affinity_and_hint(queue->irq, NULL)) 188 netdev_warn(adapter->netdev, 189 "%s: Clear affinity failed, queue addr = %p, IRQ = %d\n", 190 __func__, queue, queue->irq); 191 } 192 193 static void ibmvnic_clean_affinity(struct ibmvnic_adapter *adapter) 194 { 195 struct ibmvnic_sub_crq_queue **rxqs; 196 struct ibmvnic_sub_crq_queue **txqs; 197 int num_rxqs, num_txqs; 198 int i; 199 200 rxqs = adapter->rx_scrq; 201 txqs = adapter->tx_scrq; 202 num_txqs = adapter->num_active_tx_scrqs; 203 num_rxqs = adapter->num_active_rx_scrqs; 204 205 netdev_dbg(adapter->netdev, "%s: Cleaning irq affinity hints", __func__); 206 if (txqs) { 207 for (i = 0; i < num_txqs; i++) 208 ibmvnic_clean_queue_affinity(adapter, txqs[i]); 209 } 210 if (rxqs) { 211 for (i = 0; i < num_rxqs; i++) 212 ibmvnic_clean_queue_affinity(adapter, rxqs[i]); 213 } 214 } 215 216 static int ibmvnic_set_queue_affinity(struct ibmvnic_sub_crq_queue *queue, 217 unsigned int *cpu, int *stragglers, 218 int stride) 219 { 220 cpumask_var_t mask; 221 int i; 222 int rc = 0; 223 224 if (!(queue && queue->irq)) 225 return rc; 226 227 /* cpumask_var_t is either a pointer or array, allocation works here */ 228 if (!zalloc_cpumask_var(&mask, GFP_KERNEL)) 229 return -ENOMEM; 230 231 /* while we have extra cpu give one extra to this irq */ 232 if (*stragglers) { 233 stride++; 234 (*stragglers)--; 235 } 236 /* atomic write is safer than writing bit by bit directly */ 237 for_each_online_cpu_wrap(i, *cpu) { 238 if (!stride--) { 239 /* For the next queue we start from the first 240 * unused CPU in this queue 241 */ 242 *cpu = i; 243 break; 244 } 245 cpumask_set_cpu(i, mask); 246 } 247 248 /* set queue affinity mask */ 249 cpumask_copy(queue->affinity_mask, mask); 250 rc = irq_set_affinity_and_hint(queue->irq, queue->affinity_mask); 251 free_cpumask_var(mask); 252 253 return rc; 254 } 255 256 /* assumes cpu read lock is held */ 257 static void ibmvnic_set_affinity(struct ibmvnic_adapter *adapter) 258 { 259 struct ibmvnic_sub_crq_queue **rxqs = adapter->rx_scrq; 260 struct ibmvnic_sub_crq_queue **txqs = adapter->tx_scrq; 261 struct ibmvnic_sub_crq_queue *queue; 262 int num_rxqs = adapter->num_active_rx_scrqs, i_rxqs = 0; 263 int num_txqs = adapter->num_active_tx_scrqs, i_txqs = 0; 264 int total_queues, stride, stragglers, i; 265 unsigned int num_cpu, cpu = 0; 266 bool is_rx_queue; 267 int rc = 0; 268 269 netdev_dbg(adapter->netdev, "%s: Setting irq affinity hints", __func__); 270 if (!(adapter->rx_scrq && adapter->tx_scrq)) { 271 netdev_warn(adapter->netdev, 272 "%s: Set affinity failed, queues not allocated\n", 273 __func__); 274 return; 275 } 276 277 total_queues = num_rxqs + num_txqs; 278 num_cpu = num_online_cpus(); 279 /* number of cpu's assigned per irq */ 280 stride = max_t(int, num_cpu / total_queues, 1); 281 /* number of leftover cpu's */ 282 stragglers = num_cpu >= total_queues ? num_cpu % total_queues : 0; 283 284 for (i = 0; i < total_queues; i++) { 285 is_rx_queue = false; 286 /* balance core load by alternating rx and tx assignments 287 * ex: TX0 -> RX0 -> TX1 -> RX1 etc. 288 */ 289 if ((i % 2 == 1 && i_rxqs < num_rxqs) || i_txqs == num_txqs) { 290 queue = rxqs[i_rxqs++]; 291 is_rx_queue = true; 292 } else { 293 queue = txqs[i_txqs++]; 294 } 295 296 rc = ibmvnic_set_queue_affinity(queue, &cpu, &stragglers, 297 stride); 298 if (rc) 299 goto out; 300 301 if (!queue || is_rx_queue) 302 continue; 303 304 rc = __netif_set_xps_queue(adapter->netdev, 305 cpumask_bits(queue->affinity_mask), 306 i_txqs - 1, XPS_CPUS); 307 if (rc) 308 netdev_warn(adapter->netdev, "%s: Set XPS on queue %d failed, rc = %d.\n", 309 __func__, i_txqs - 1, rc); 310 } 311 312 out: 313 if (rc) { 314 netdev_warn(adapter->netdev, 315 "%s: Set affinity failed, queue addr = %p, IRQ = %d, rc = %d.\n", 316 __func__, queue, queue->irq, rc); 317 ibmvnic_clean_affinity(adapter); 318 } 319 } 320 321 static int ibmvnic_cpu_online(unsigned int cpu, struct hlist_node *node) 322 { 323 struct ibmvnic_adapter *adapter; 324 325 adapter = hlist_entry_safe(node, struct ibmvnic_adapter, node); 326 ibmvnic_set_affinity(adapter); 327 return 0; 328 } 329 330 static int ibmvnic_cpu_dead(unsigned int cpu, struct hlist_node *node) 331 { 332 struct ibmvnic_adapter *adapter; 333 334 adapter = hlist_entry_safe(node, struct ibmvnic_adapter, node_dead); 335 ibmvnic_set_affinity(adapter); 336 return 0; 337 } 338 339 static int ibmvnic_cpu_down_prep(unsigned int cpu, struct hlist_node *node) 340 { 341 struct ibmvnic_adapter *adapter; 342 343 adapter = hlist_entry_safe(node, struct ibmvnic_adapter, node); 344 ibmvnic_clean_affinity(adapter); 345 return 0; 346 } 347 348 static enum cpuhp_state ibmvnic_online; 349 350 static int ibmvnic_cpu_notif_add(struct ibmvnic_adapter *adapter) 351 { 352 int ret; 353 354 ret = cpuhp_state_add_instance_nocalls(ibmvnic_online, &adapter->node); 355 if (ret) 356 return ret; 357 ret = cpuhp_state_add_instance_nocalls(CPUHP_IBMVNIC_DEAD, 358 &adapter->node_dead); 359 if (!ret) 360 return ret; 361 cpuhp_state_remove_instance_nocalls(ibmvnic_online, &adapter->node); 362 return ret; 363 } 364 365 static void ibmvnic_cpu_notif_remove(struct ibmvnic_adapter *adapter) 366 { 367 cpuhp_state_remove_instance_nocalls(ibmvnic_online, &adapter->node); 368 cpuhp_state_remove_instance_nocalls(CPUHP_IBMVNIC_DEAD, 369 &adapter->node_dead); 370 } 371 372 static long h_reg_sub_crq(unsigned long unit_address, unsigned long token, 373 unsigned long length, unsigned long *number, 374 unsigned long *irq) 375 { 376 unsigned long retbuf[PLPAR_HCALL_BUFSIZE]; 377 long rc; 378 379 rc = plpar_hcall(H_REG_SUB_CRQ, retbuf, unit_address, token, length); 380 *number = retbuf[0]; 381 *irq = retbuf[1]; 382 383 return rc; 384 } 385 386 /** 387 * ibmvnic_wait_for_completion - Check device state and wait for completion 388 * @adapter: private device data 389 * @comp_done: completion structure to wait for 390 * @timeout: time to wait in milliseconds 391 * 392 * Wait for a completion signal or until the timeout limit is reached 393 * while checking that the device is still active. 394 */ 395 static int ibmvnic_wait_for_completion(struct ibmvnic_adapter *adapter, 396 struct completion *comp_done, 397 unsigned long timeout) 398 { 399 struct net_device *netdev; 400 unsigned long div_timeout; 401 u8 retry; 402 403 netdev = adapter->netdev; 404 retry = 5; 405 div_timeout = msecs_to_jiffies(timeout / retry); 406 while (true) { 407 if (!adapter->crq.active) { 408 netdev_err(netdev, "Device down!\n"); 409 return -ENODEV; 410 } 411 if (!retry--) 412 break; 413 if (wait_for_completion_timeout(comp_done, div_timeout)) 414 return 0; 415 } 416 netdev_err(netdev, "Operation timed out.\n"); 417 return -ETIMEDOUT; 418 } 419 420 /** 421 * reuse_ltb() - Check if a long term buffer can be reused 422 * @ltb: The long term buffer to be checked 423 * @size: The size of the long term buffer. 424 * 425 * An LTB can be reused unless its size has changed. 426 * 427 * Return: Return true if the LTB can be reused, false otherwise. 428 */ 429 static bool reuse_ltb(struct ibmvnic_long_term_buff *ltb, int size) 430 { 431 return (ltb->buff && ltb->size == size); 432 } 433 434 /** 435 * alloc_long_term_buff() - Allocate a long term buffer (LTB) 436 * 437 * @adapter: ibmvnic adapter associated to the LTB 438 * @ltb: container object for the LTB 439 * @size: size of the LTB 440 * 441 * Allocate an LTB of the specified size and notify VIOS. 442 * 443 * If the given @ltb already has the correct size, reuse it. Otherwise if 444 * its non-NULL, free it. Then allocate a new one of the correct size. 445 * Notify the VIOS either way since we may now be working with a new VIOS. 446 * 447 * Allocating larger chunks of memory during resets, specially LPM or under 448 * low memory situations can cause resets to fail/timeout and for LPAR to 449 * lose connectivity. So hold onto the LTB even if we fail to communicate 450 * with the VIOS and reuse it on next open. Free LTB when adapter is closed. 451 * 452 * Return: 0 if we were able to allocate the LTB and notify the VIOS and 453 * a negative value otherwise. 454 */ 455 static int alloc_long_term_buff(struct ibmvnic_adapter *adapter, 456 struct ibmvnic_long_term_buff *ltb, int size) 457 { 458 struct device *dev = &adapter->vdev->dev; 459 u64 prev = 0; 460 int rc; 461 462 if (!reuse_ltb(ltb, size)) { 463 dev_dbg(dev, 464 "LTB size changed from 0x%llx to 0x%x, reallocating\n", 465 ltb->size, size); 466 prev = ltb->size; 467 free_long_term_buff(adapter, ltb); 468 } 469 470 if (ltb->buff) { 471 dev_dbg(dev, "Reusing LTB [map %d, size 0x%llx]\n", 472 ltb->map_id, ltb->size); 473 } else { 474 ltb->buff = dma_alloc_coherent(dev, size, <b->addr, 475 GFP_KERNEL); 476 if (!ltb->buff) { 477 dev_err(dev, "Couldn't alloc long term buffer\n"); 478 return -ENOMEM; 479 } 480 ltb->size = size; 481 482 ltb->map_id = find_first_zero_bit(adapter->map_ids, 483 MAX_MAP_ID); 484 bitmap_set(adapter->map_ids, ltb->map_id, 1); 485 486 dev_dbg(dev, 487 "Allocated new LTB [map %d, size 0x%llx was 0x%llx]\n", 488 ltb->map_id, ltb->size, prev); 489 } 490 491 /* Ensure ltb is zeroed - specially when reusing it. */ 492 memset(ltb->buff, 0, ltb->size); 493 494 mutex_lock(&adapter->fw_lock); 495 adapter->fw_done_rc = 0; 496 reinit_completion(&adapter->fw_done); 497 498 rc = send_request_map(adapter, ltb->addr, ltb->size, ltb->map_id); 499 if (rc) { 500 dev_err(dev, "send_request_map failed, rc = %d\n", rc); 501 goto out; 502 } 503 504 rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000); 505 if (rc) { 506 dev_err(dev, "LTB map request aborted or timed out, rc = %d\n", 507 rc); 508 goto out; 509 } 510 511 if (adapter->fw_done_rc) { 512 dev_err(dev, "Couldn't map LTB, rc = %d\n", 513 adapter->fw_done_rc); 514 rc = -EIO; 515 goto out; 516 } 517 rc = 0; 518 out: 519 /* don't free LTB on communication error - see function header */ 520 mutex_unlock(&adapter->fw_lock); 521 return rc; 522 } 523 524 static void free_long_term_buff(struct ibmvnic_adapter *adapter, 525 struct ibmvnic_long_term_buff *ltb) 526 { 527 struct device *dev = &adapter->vdev->dev; 528 529 if (!ltb->buff) 530 return; 531 532 /* VIOS automatically unmaps the long term buffer at remote 533 * end for the following resets: 534 * FAILOVER, MOBILITY, TIMEOUT. 535 */ 536 if (adapter->reset_reason != VNIC_RESET_FAILOVER && 537 adapter->reset_reason != VNIC_RESET_MOBILITY && 538 adapter->reset_reason != VNIC_RESET_TIMEOUT) 539 send_request_unmap(adapter, ltb->map_id); 540 541 dma_free_coherent(dev, ltb->size, ltb->buff, ltb->addr); 542 543 ltb->buff = NULL; 544 /* mark this map_id free */ 545 bitmap_clear(adapter->map_ids, ltb->map_id, 1); 546 ltb->map_id = 0; 547 } 548 549 /** 550 * free_ltb_set - free the given set of long term buffers (LTBS) 551 * @adapter: The ibmvnic adapter containing this ltb set 552 * @ltb_set: The ltb_set to be freed 553 * 554 * Free the set of LTBs in the given set. 555 */ 556 557 static void free_ltb_set(struct ibmvnic_adapter *adapter, 558 struct ibmvnic_ltb_set *ltb_set) 559 { 560 int i; 561 562 for (i = 0; i < ltb_set->num_ltbs; i++) 563 free_long_term_buff(adapter, <b_set->ltbs[i]); 564 565 kfree(ltb_set->ltbs); 566 ltb_set->ltbs = NULL; 567 ltb_set->num_ltbs = 0; 568 } 569 570 /** 571 * alloc_ltb_set() - Allocate a set of long term buffers (LTBs) 572 * 573 * @adapter: ibmvnic adapter associated to the LTB 574 * @ltb_set: container object for the set of LTBs 575 * @num_buffs: Number of buffers in the LTB 576 * @buff_size: Size of each buffer in the LTB 577 * 578 * Allocate a set of LTBs to accommodate @num_buffs buffers of @buff_size 579 * each. We currently cap size each LTB to IBMVNIC_ONE_LTB_SIZE. If the 580 * new set of LTBs have fewer LTBs than the old set, free the excess LTBs. 581 * If new set needs more than in old set, allocate the remaining ones. 582 * Try and reuse as many LTBs as possible and avoid reallocation. 583 * 584 * Any changes to this allocation strategy must be reflected in 585 * map_rxpool_buff_to_ltb() and map_txpool_buff_to_ltb(). 586 */ 587 static int alloc_ltb_set(struct ibmvnic_adapter *adapter, 588 struct ibmvnic_ltb_set *ltb_set, int num_buffs, 589 int buff_size) 590 { 591 struct device *dev = &adapter->vdev->dev; 592 struct ibmvnic_ltb_set old_set; 593 struct ibmvnic_ltb_set new_set; 594 int rem_size; 595 int tot_size; /* size of all ltbs */ 596 int ltb_size; /* size of one ltb */ 597 int nltbs; 598 int rc; 599 int n; 600 int i; 601 602 dev_dbg(dev, "%s() num_buffs %d, buff_size %d\n", __func__, num_buffs, 603 buff_size); 604 605 ltb_size = rounddown(IBMVNIC_ONE_LTB_SIZE, buff_size); 606 tot_size = num_buffs * buff_size; 607 608 if (ltb_size > tot_size) 609 ltb_size = tot_size; 610 611 nltbs = tot_size / ltb_size; 612 if (tot_size % ltb_size) 613 nltbs++; 614 615 old_set = *ltb_set; 616 617 if (old_set.num_ltbs == nltbs) { 618 new_set = old_set; 619 } else { 620 int tmp = nltbs * sizeof(struct ibmvnic_long_term_buff); 621 622 new_set.ltbs = kzalloc(tmp, GFP_KERNEL); 623 if (!new_set.ltbs) 624 return -ENOMEM; 625 626 new_set.num_ltbs = nltbs; 627 628 /* Free any excess ltbs in old set */ 629 for (i = new_set.num_ltbs; i < old_set.num_ltbs; i++) 630 free_long_term_buff(adapter, &old_set.ltbs[i]); 631 632 /* Copy remaining ltbs to new set. All LTBs except the 633 * last one are of the same size. alloc_long_term_buff() 634 * will realloc if the size changes. 635 */ 636 n = min(old_set.num_ltbs, new_set.num_ltbs); 637 for (i = 0; i < n; i++) 638 new_set.ltbs[i] = old_set.ltbs[i]; 639 640 /* Any additional ltbs in new set will have NULL ltbs for 641 * now and will be allocated in alloc_long_term_buff(). 642 */ 643 644 /* We no longer need the old_set so free it. Note that we 645 * may have reused some ltbs from old set and freed excess 646 * ltbs above. So we only need to free the container now 647 * not the LTBs themselves. (i.e. dont free_ltb_set()!) 648 */ 649 kfree(old_set.ltbs); 650 old_set.ltbs = NULL; 651 old_set.num_ltbs = 0; 652 653 /* Install the new set. If allocations fail below, we will 654 * retry later and know what size LTBs we need. 655 */ 656 *ltb_set = new_set; 657 } 658 659 i = 0; 660 rem_size = tot_size; 661 while (rem_size) { 662 if (ltb_size > rem_size) 663 ltb_size = rem_size; 664 665 rem_size -= ltb_size; 666 667 rc = alloc_long_term_buff(adapter, &new_set.ltbs[i], ltb_size); 668 if (rc) 669 goto out; 670 i++; 671 } 672 673 WARN_ON(i != new_set.num_ltbs); 674 675 return 0; 676 out: 677 /* We may have allocated one/more LTBs before failing and we 678 * want to try and reuse on next reset. So don't free ltb set. 679 */ 680 return rc; 681 } 682 683 /** 684 * map_rxpool_buf_to_ltb - Map given rxpool buffer to offset in an LTB. 685 * @rxpool: The receive buffer pool containing buffer 686 * @bufidx: Index of buffer in rxpool 687 * @ltbp: (Output) pointer to the long term buffer containing the buffer 688 * @offset: (Output) offset of buffer in the LTB from @ltbp 689 * 690 * Map the given buffer identified by [rxpool, bufidx] to an LTB in the 691 * pool and its corresponding offset. Assume for now that each LTB is of 692 * different size but could possibly be optimized based on the allocation 693 * strategy in alloc_ltb_set(). 694 */ 695 static void map_rxpool_buf_to_ltb(struct ibmvnic_rx_pool *rxpool, 696 unsigned int bufidx, 697 struct ibmvnic_long_term_buff **ltbp, 698 unsigned int *offset) 699 { 700 struct ibmvnic_long_term_buff *ltb; 701 int nbufs; /* # of buffers in one ltb */ 702 int i; 703 704 WARN_ON(bufidx >= rxpool->size); 705 706 for (i = 0; i < rxpool->ltb_set.num_ltbs; i++) { 707 ltb = &rxpool->ltb_set.ltbs[i]; 708 nbufs = ltb->size / rxpool->buff_size; 709 if (bufidx < nbufs) 710 break; 711 bufidx -= nbufs; 712 } 713 714 *ltbp = ltb; 715 *offset = bufidx * rxpool->buff_size; 716 } 717 718 /** 719 * map_txpool_buf_to_ltb - Map given txpool buffer to offset in an LTB. 720 * @txpool: The transmit buffer pool containing buffer 721 * @bufidx: Index of buffer in txpool 722 * @ltbp: (Output) pointer to the long term buffer (LTB) containing the buffer 723 * @offset: (Output) offset of buffer in the LTB from @ltbp 724 * 725 * Map the given buffer identified by [txpool, bufidx] to an LTB in the 726 * pool and its corresponding offset. 727 */ 728 static void map_txpool_buf_to_ltb(struct ibmvnic_tx_pool *txpool, 729 unsigned int bufidx, 730 struct ibmvnic_long_term_buff **ltbp, 731 unsigned int *offset) 732 { 733 struct ibmvnic_long_term_buff *ltb; 734 int nbufs; /* # of buffers in one ltb */ 735 int i; 736 737 WARN_ON_ONCE(bufidx >= txpool->num_buffers); 738 739 for (i = 0; i < txpool->ltb_set.num_ltbs; i++) { 740 ltb = &txpool->ltb_set.ltbs[i]; 741 nbufs = ltb->size / txpool->buf_size; 742 if (bufidx < nbufs) 743 break; 744 bufidx -= nbufs; 745 } 746 747 *ltbp = ltb; 748 *offset = bufidx * txpool->buf_size; 749 } 750 751 static void deactivate_rx_pools(struct ibmvnic_adapter *adapter) 752 { 753 int i; 754 755 for (i = 0; i < adapter->num_active_rx_pools; i++) 756 adapter->rx_pool[i].active = 0; 757 } 758 759 static void replenish_rx_pool(struct ibmvnic_adapter *adapter, 760 struct ibmvnic_rx_pool *pool) 761 { 762 int count = pool->size - atomic_read(&pool->available); 763 u64 handle = adapter->rx_scrq[pool->index]->handle; 764 struct device *dev = &adapter->vdev->dev; 765 struct ibmvnic_ind_xmit_queue *ind_bufp; 766 struct ibmvnic_sub_crq_queue *rx_scrq; 767 struct ibmvnic_long_term_buff *ltb; 768 union sub_crq *sub_crq; 769 int buffers_added = 0; 770 unsigned long lpar_rc; 771 struct sk_buff *skb; 772 unsigned int offset; 773 dma_addr_t dma_addr; 774 unsigned char *dst; 775 int shift = 0; 776 int bufidx; 777 int i; 778 779 if (!pool->active) 780 return; 781 782 rx_scrq = adapter->rx_scrq[pool->index]; 783 ind_bufp = &rx_scrq->ind_buf; 784 785 /* netdev_skb_alloc() could have failed after we saved a few skbs 786 * in the indir_buf and we would not have sent them to VIOS yet. 787 * To account for them, start the loop at ind_bufp->index rather 788 * than 0. If we pushed all the skbs to VIOS, ind_bufp->index will 789 * be 0. 790 */ 791 for (i = ind_bufp->index; i < count; ++i) { 792 bufidx = pool->free_map[pool->next_free]; 793 794 /* We maybe reusing the skb from earlier resets. Allocate 795 * only if necessary. But since the LTB may have changed 796 * during reset (see init_rx_pools()), update LTB below 797 * even if reusing skb. 798 */ 799 skb = pool->rx_buff[bufidx].skb; 800 if (!skb) { 801 skb = netdev_alloc_skb(adapter->netdev, 802 pool->buff_size); 803 if (!skb) { 804 dev_err(dev, "Couldn't replenish rx buff\n"); 805 adapter->replenish_no_mem++; 806 break; 807 } 808 } 809 810 pool->free_map[pool->next_free] = IBMVNIC_INVALID_MAP; 811 pool->next_free = (pool->next_free + 1) % pool->size; 812 813 /* Copy the skb to the long term mapped DMA buffer */ 814 map_rxpool_buf_to_ltb(pool, bufidx, <b, &offset); 815 dst = ltb->buff + offset; 816 memset(dst, 0, pool->buff_size); 817 dma_addr = ltb->addr + offset; 818 819 /* add the skb to an rx_buff in the pool */ 820 pool->rx_buff[bufidx].data = dst; 821 pool->rx_buff[bufidx].dma = dma_addr; 822 pool->rx_buff[bufidx].skb = skb; 823 pool->rx_buff[bufidx].pool_index = pool->index; 824 pool->rx_buff[bufidx].size = pool->buff_size; 825 826 /* queue the rx_buff for the next send_subcrq_indirect */ 827 sub_crq = &ind_bufp->indir_arr[ind_bufp->index++]; 828 memset(sub_crq, 0, sizeof(*sub_crq)); 829 sub_crq->rx_add.first = IBMVNIC_CRQ_CMD; 830 sub_crq->rx_add.correlator = 831 cpu_to_be64((u64)&pool->rx_buff[bufidx]); 832 sub_crq->rx_add.ioba = cpu_to_be32(dma_addr); 833 sub_crq->rx_add.map_id = ltb->map_id; 834 835 /* The length field of the sCRQ is defined to be 24 bits so the 836 * buffer size needs to be left shifted by a byte before it is 837 * converted to big endian to prevent the last byte from being 838 * truncated. 839 */ 840 #ifdef __LITTLE_ENDIAN__ 841 shift = 8; 842 #endif 843 sub_crq->rx_add.len = cpu_to_be32(pool->buff_size << shift); 844 845 /* if send_subcrq_indirect queue is full, flush to VIOS */ 846 if (ind_bufp->index == IBMVNIC_MAX_IND_DESCS || 847 i == count - 1) { 848 lpar_rc = 849 send_subcrq_indirect(adapter, handle, 850 (u64)ind_bufp->indir_dma, 851 (u64)ind_bufp->index); 852 if (lpar_rc != H_SUCCESS) 853 goto failure; 854 buffers_added += ind_bufp->index; 855 adapter->replenish_add_buff_success += ind_bufp->index; 856 ind_bufp->index = 0; 857 } 858 } 859 atomic_add(buffers_added, &pool->available); 860 return; 861 862 failure: 863 if (lpar_rc != H_PARAMETER && lpar_rc != H_CLOSED) 864 dev_err_ratelimited(dev, "rx: replenish packet buffer failed\n"); 865 for (i = ind_bufp->index - 1; i >= 0; --i) { 866 struct ibmvnic_rx_buff *rx_buff; 867 868 pool->next_free = pool->next_free == 0 ? 869 pool->size - 1 : pool->next_free - 1; 870 sub_crq = &ind_bufp->indir_arr[i]; 871 rx_buff = (struct ibmvnic_rx_buff *) 872 be64_to_cpu(sub_crq->rx_add.correlator); 873 bufidx = (int)(rx_buff - pool->rx_buff); 874 pool->free_map[pool->next_free] = bufidx; 875 dev_kfree_skb_any(pool->rx_buff[bufidx].skb); 876 pool->rx_buff[bufidx].skb = NULL; 877 } 878 adapter->replenish_add_buff_failure += ind_bufp->index; 879 atomic_add(buffers_added, &pool->available); 880 ind_bufp->index = 0; 881 if (lpar_rc == H_CLOSED || adapter->failover_pending) { 882 /* Disable buffer pool replenishment and report carrier off if 883 * queue is closed or pending failover. 884 * Firmware guarantees that a signal will be sent to the 885 * driver, triggering a reset. 886 */ 887 deactivate_rx_pools(adapter); 888 netif_carrier_off(adapter->netdev); 889 } 890 } 891 892 static void replenish_pools(struct ibmvnic_adapter *adapter) 893 { 894 int i; 895 896 adapter->replenish_task_cycles++; 897 for (i = 0; i < adapter->num_active_rx_pools; i++) { 898 if (adapter->rx_pool[i].active) 899 replenish_rx_pool(adapter, &adapter->rx_pool[i]); 900 } 901 902 netdev_dbg(adapter->netdev, "Replenished %d pools\n", i); 903 } 904 905 static void release_stats_buffers(struct ibmvnic_adapter *adapter) 906 { 907 kfree(adapter->tx_stats_buffers); 908 kfree(adapter->rx_stats_buffers); 909 adapter->tx_stats_buffers = NULL; 910 adapter->rx_stats_buffers = NULL; 911 } 912 913 static int init_stats_buffers(struct ibmvnic_adapter *adapter) 914 { 915 adapter->tx_stats_buffers = 916 kcalloc(IBMVNIC_MAX_QUEUES, 917 sizeof(struct ibmvnic_tx_queue_stats), 918 GFP_KERNEL); 919 if (!adapter->tx_stats_buffers) 920 return -ENOMEM; 921 922 adapter->rx_stats_buffers = 923 kcalloc(IBMVNIC_MAX_QUEUES, 924 sizeof(struct ibmvnic_rx_queue_stats), 925 GFP_KERNEL); 926 if (!adapter->rx_stats_buffers) 927 return -ENOMEM; 928 929 return 0; 930 } 931 932 static void release_stats_token(struct ibmvnic_adapter *adapter) 933 { 934 struct device *dev = &adapter->vdev->dev; 935 936 if (!adapter->stats_token) 937 return; 938 939 dma_unmap_single(dev, adapter->stats_token, 940 sizeof(struct ibmvnic_statistics), 941 DMA_FROM_DEVICE); 942 adapter->stats_token = 0; 943 } 944 945 static int init_stats_token(struct ibmvnic_adapter *adapter) 946 { 947 struct device *dev = &adapter->vdev->dev; 948 dma_addr_t stok; 949 int rc; 950 951 stok = dma_map_single(dev, &adapter->stats, 952 sizeof(struct ibmvnic_statistics), 953 DMA_FROM_DEVICE); 954 rc = dma_mapping_error(dev, stok); 955 if (rc) { 956 dev_err(dev, "Couldn't map stats buffer, rc = %d\n", rc); 957 return rc; 958 } 959 960 adapter->stats_token = stok; 961 netdev_dbg(adapter->netdev, "Stats token initialized (%llx)\n", stok); 962 return 0; 963 } 964 965 /** 966 * release_rx_pools() - Release any rx pools attached to @adapter. 967 * @adapter: ibmvnic adapter 968 * 969 * Safe to call this multiple times - even if no pools are attached. 970 */ 971 static void release_rx_pools(struct ibmvnic_adapter *adapter) 972 { 973 struct ibmvnic_rx_pool *rx_pool; 974 int i, j; 975 976 if (!adapter->rx_pool) 977 return; 978 979 for (i = 0; i < adapter->num_active_rx_pools; i++) { 980 rx_pool = &adapter->rx_pool[i]; 981 982 netdev_dbg(adapter->netdev, "Releasing rx_pool[%d]\n", i); 983 984 kfree(rx_pool->free_map); 985 986 free_ltb_set(adapter, &rx_pool->ltb_set); 987 988 if (!rx_pool->rx_buff) 989 continue; 990 991 for (j = 0; j < rx_pool->size; j++) { 992 if (rx_pool->rx_buff[j].skb) { 993 dev_kfree_skb_any(rx_pool->rx_buff[j].skb); 994 rx_pool->rx_buff[j].skb = NULL; 995 } 996 } 997 998 kfree(rx_pool->rx_buff); 999 } 1000 1001 kfree(adapter->rx_pool); 1002 adapter->rx_pool = NULL; 1003 adapter->num_active_rx_pools = 0; 1004 adapter->prev_rx_pool_size = 0; 1005 } 1006 1007 /** 1008 * reuse_rx_pools() - Check if the existing rx pools can be reused. 1009 * @adapter: ibmvnic adapter 1010 * 1011 * Check if the existing rx pools in the adapter can be reused. The 1012 * pools can be reused if the pool parameters (number of pools, 1013 * number of buffers in the pool and size of each buffer) have not 1014 * changed. 1015 * 1016 * NOTE: This assumes that all pools have the same number of buffers 1017 * which is the case currently. If that changes, we must fix this. 1018 * 1019 * Return: true if the rx pools can be reused, false otherwise. 1020 */ 1021 static bool reuse_rx_pools(struct ibmvnic_adapter *adapter) 1022 { 1023 u64 old_num_pools, new_num_pools; 1024 u64 old_pool_size, new_pool_size; 1025 u64 old_buff_size, new_buff_size; 1026 1027 if (!adapter->rx_pool) 1028 return false; 1029 1030 old_num_pools = adapter->num_active_rx_pools; 1031 new_num_pools = adapter->req_rx_queues; 1032 1033 old_pool_size = adapter->prev_rx_pool_size; 1034 new_pool_size = adapter->req_rx_add_entries_per_subcrq; 1035 1036 old_buff_size = adapter->prev_rx_buf_sz; 1037 new_buff_size = adapter->cur_rx_buf_sz; 1038 1039 if (old_buff_size != new_buff_size || 1040 old_num_pools != new_num_pools || 1041 old_pool_size != new_pool_size) 1042 return false; 1043 1044 return true; 1045 } 1046 1047 /** 1048 * init_rx_pools(): Initialize the set of receiver pools in the adapter. 1049 * @netdev: net device associated with the vnic interface 1050 * 1051 * Initialize the set of receiver pools in the ibmvnic adapter associated 1052 * with the net_device @netdev. If possible, reuse the existing rx pools. 1053 * Otherwise free any existing pools and allocate a new set of pools 1054 * before initializing them. 1055 * 1056 * Return: 0 on success and negative value on error. 1057 */ 1058 static int init_rx_pools(struct net_device *netdev) 1059 { 1060 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 1061 struct device *dev = &adapter->vdev->dev; 1062 struct ibmvnic_rx_pool *rx_pool; 1063 u64 num_pools; 1064 u64 pool_size; /* # of buffers in one pool */ 1065 u64 buff_size; 1066 int i, j, rc; 1067 1068 pool_size = adapter->req_rx_add_entries_per_subcrq; 1069 num_pools = adapter->req_rx_queues; 1070 buff_size = adapter->cur_rx_buf_sz; 1071 1072 if (reuse_rx_pools(adapter)) { 1073 dev_dbg(dev, "Reusing rx pools\n"); 1074 goto update_ltb; 1075 } 1076 1077 /* Allocate/populate the pools. */ 1078 release_rx_pools(adapter); 1079 1080 adapter->rx_pool = kcalloc(num_pools, 1081 sizeof(struct ibmvnic_rx_pool), 1082 GFP_KERNEL); 1083 if (!adapter->rx_pool) { 1084 dev_err(dev, "Failed to allocate rx pools\n"); 1085 return -ENOMEM; 1086 } 1087 1088 /* Set num_active_rx_pools early. If we fail below after partial 1089 * allocation, release_rx_pools() will know how many to look for. 1090 */ 1091 adapter->num_active_rx_pools = num_pools; 1092 1093 for (i = 0; i < num_pools; i++) { 1094 rx_pool = &adapter->rx_pool[i]; 1095 1096 netdev_dbg(adapter->netdev, 1097 "Initializing rx_pool[%d], %lld buffs, %lld bytes each\n", 1098 i, pool_size, buff_size); 1099 1100 rx_pool->size = pool_size; 1101 rx_pool->index = i; 1102 rx_pool->buff_size = ALIGN(buff_size, L1_CACHE_BYTES); 1103 1104 rx_pool->free_map = kcalloc(rx_pool->size, sizeof(int), 1105 GFP_KERNEL); 1106 if (!rx_pool->free_map) { 1107 dev_err(dev, "Couldn't alloc free_map %d\n", i); 1108 rc = -ENOMEM; 1109 goto out_release; 1110 } 1111 1112 rx_pool->rx_buff = kcalloc(rx_pool->size, 1113 sizeof(struct ibmvnic_rx_buff), 1114 GFP_KERNEL); 1115 if (!rx_pool->rx_buff) { 1116 dev_err(dev, "Couldn't alloc rx buffers\n"); 1117 rc = -ENOMEM; 1118 goto out_release; 1119 } 1120 } 1121 1122 adapter->prev_rx_pool_size = pool_size; 1123 adapter->prev_rx_buf_sz = adapter->cur_rx_buf_sz; 1124 1125 update_ltb: 1126 for (i = 0; i < num_pools; i++) { 1127 rx_pool = &adapter->rx_pool[i]; 1128 dev_dbg(dev, "Updating LTB for rx pool %d [%d, %d]\n", 1129 i, rx_pool->size, rx_pool->buff_size); 1130 1131 rc = alloc_ltb_set(adapter, &rx_pool->ltb_set, 1132 rx_pool->size, rx_pool->buff_size); 1133 if (rc) 1134 goto out; 1135 1136 for (j = 0; j < rx_pool->size; ++j) { 1137 struct ibmvnic_rx_buff *rx_buff; 1138 1139 rx_pool->free_map[j] = j; 1140 1141 /* NOTE: Don't clear rx_buff->skb here - will leak 1142 * memory! replenish_rx_pool() will reuse skbs or 1143 * allocate as necessary. 1144 */ 1145 rx_buff = &rx_pool->rx_buff[j]; 1146 rx_buff->dma = 0; 1147 rx_buff->data = 0; 1148 rx_buff->size = 0; 1149 rx_buff->pool_index = 0; 1150 } 1151 1152 /* Mark pool "empty" so replenish_rx_pools() will 1153 * update the LTB info for each buffer 1154 */ 1155 atomic_set(&rx_pool->available, 0); 1156 rx_pool->next_alloc = 0; 1157 rx_pool->next_free = 0; 1158 /* replenish_rx_pool() may have called deactivate_rx_pools() 1159 * on failover. Ensure pool is active now. 1160 */ 1161 rx_pool->active = 1; 1162 } 1163 return 0; 1164 out_release: 1165 release_rx_pools(adapter); 1166 out: 1167 /* We failed to allocate one or more LTBs or map them on the VIOS. 1168 * Hold onto the pools and any LTBs that we did allocate/map. 1169 */ 1170 return rc; 1171 } 1172 1173 static void release_vpd_data(struct ibmvnic_adapter *adapter) 1174 { 1175 if (!adapter->vpd) 1176 return; 1177 1178 kfree(adapter->vpd->buff); 1179 kfree(adapter->vpd); 1180 1181 adapter->vpd = NULL; 1182 } 1183 1184 static void release_one_tx_pool(struct ibmvnic_adapter *adapter, 1185 struct ibmvnic_tx_pool *tx_pool) 1186 { 1187 kfree(tx_pool->tx_buff); 1188 kfree(tx_pool->free_map); 1189 free_ltb_set(adapter, &tx_pool->ltb_set); 1190 } 1191 1192 /** 1193 * release_tx_pools() - Release any tx pools attached to @adapter. 1194 * @adapter: ibmvnic adapter 1195 * 1196 * Safe to call this multiple times - even if no pools are attached. 1197 */ 1198 static void release_tx_pools(struct ibmvnic_adapter *adapter) 1199 { 1200 int i; 1201 1202 /* init_tx_pools() ensures that ->tx_pool and ->tso_pool are 1203 * both NULL or both non-NULL. So we only need to check one. 1204 */ 1205 if (!adapter->tx_pool) 1206 return; 1207 1208 for (i = 0; i < adapter->num_active_tx_pools; i++) { 1209 release_one_tx_pool(adapter, &adapter->tx_pool[i]); 1210 release_one_tx_pool(adapter, &adapter->tso_pool[i]); 1211 } 1212 1213 kfree(adapter->tx_pool); 1214 adapter->tx_pool = NULL; 1215 kfree(adapter->tso_pool); 1216 adapter->tso_pool = NULL; 1217 adapter->num_active_tx_pools = 0; 1218 adapter->prev_tx_pool_size = 0; 1219 } 1220 1221 static int init_one_tx_pool(struct net_device *netdev, 1222 struct ibmvnic_tx_pool *tx_pool, 1223 int pool_size, int buf_size) 1224 { 1225 int i; 1226 1227 tx_pool->tx_buff = kcalloc(pool_size, 1228 sizeof(struct ibmvnic_tx_buff), 1229 GFP_KERNEL); 1230 if (!tx_pool->tx_buff) 1231 return -ENOMEM; 1232 1233 tx_pool->free_map = kcalloc(pool_size, sizeof(int), GFP_KERNEL); 1234 if (!tx_pool->free_map) { 1235 kfree(tx_pool->tx_buff); 1236 tx_pool->tx_buff = NULL; 1237 return -ENOMEM; 1238 } 1239 1240 for (i = 0; i < pool_size; i++) 1241 tx_pool->free_map[i] = i; 1242 1243 tx_pool->consumer_index = 0; 1244 tx_pool->producer_index = 0; 1245 tx_pool->num_buffers = pool_size; 1246 tx_pool->buf_size = buf_size; 1247 1248 return 0; 1249 } 1250 1251 /** 1252 * reuse_tx_pools() - Check if the existing tx pools can be reused. 1253 * @adapter: ibmvnic adapter 1254 * 1255 * Check if the existing tx pools in the adapter can be reused. The 1256 * pools can be reused if the pool parameters (number of pools, 1257 * number of buffers in the pool and mtu) have not changed. 1258 * 1259 * NOTE: This assumes that all pools have the same number of buffers 1260 * which is the case currently. If that changes, we must fix this. 1261 * 1262 * Return: true if the tx pools can be reused, false otherwise. 1263 */ 1264 static bool reuse_tx_pools(struct ibmvnic_adapter *adapter) 1265 { 1266 u64 old_num_pools, new_num_pools; 1267 u64 old_pool_size, new_pool_size; 1268 u64 old_mtu, new_mtu; 1269 1270 if (!adapter->tx_pool) 1271 return false; 1272 1273 old_num_pools = adapter->num_active_tx_pools; 1274 new_num_pools = adapter->num_active_tx_scrqs; 1275 old_pool_size = adapter->prev_tx_pool_size; 1276 new_pool_size = adapter->req_tx_entries_per_subcrq; 1277 old_mtu = adapter->prev_mtu; 1278 new_mtu = adapter->req_mtu; 1279 1280 if (old_mtu != new_mtu || 1281 old_num_pools != new_num_pools || 1282 old_pool_size != new_pool_size) 1283 return false; 1284 1285 return true; 1286 } 1287 1288 /** 1289 * init_tx_pools(): Initialize the set of transmit pools in the adapter. 1290 * @netdev: net device associated with the vnic interface 1291 * 1292 * Initialize the set of transmit pools in the ibmvnic adapter associated 1293 * with the net_device @netdev. If possible, reuse the existing tx pools. 1294 * Otherwise free any existing pools and allocate a new set of pools 1295 * before initializing them. 1296 * 1297 * Return: 0 on success and negative value on error. 1298 */ 1299 static int init_tx_pools(struct net_device *netdev) 1300 { 1301 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 1302 struct device *dev = &adapter->vdev->dev; 1303 int num_pools; 1304 u64 pool_size; /* # of buffers in pool */ 1305 u64 buff_size; 1306 int i, j, rc; 1307 1308 num_pools = adapter->req_tx_queues; 1309 1310 /* We must notify the VIOS about the LTB on all resets - but we only 1311 * need to alloc/populate pools if either the number of buffers or 1312 * size of each buffer in the pool has changed. 1313 */ 1314 if (reuse_tx_pools(adapter)) { 1315 netdev_dbg(netdev, "Reusing tx pools\n"); 1316 goto update_ltb; 1317 } 1318 1319 /* Allocate/populate the pools. */ 1320 release_tx_pools(adapter); 1321 1322 pool_size = adapter->req_tx_entries_per_subcrq; 1323 num_pools = adapter->num_active_tx_scrqs; 1324 1325 adapter->tx_pool = kcalloc(num_pools, 1326 sizeof(struct ibmvnic_tx_pool), GFP_KERNEL); 1327 if (!adapter->tx_pool) 1328 return -ENOMEM; 1329 1330 adapter->tso_pool = kcalloc(num_pools, 1331 sizeof(struct ibmvnic_tx_pool), GFP_KERNEL); 1332 /* To simplify release_tx_pools() ensure that ->tx_pool and 1333 * ->tso_pool are either both NULL or both non-NULL. 1334 */ 1335 if (!adapter->tso_pool) { 1336 kfree(adapter->tx_pool); 1337 adapter->tx_pool = NULL; 1338 return -ENOMEM; 1339 } 1340 1341 /* Set num_active_tx_pools early. If we fail below after partial 1342 * allocation, release_tx_pools() will know how many to look for. 1343 */ 1344 adapter->num_active_tx_pools = num_pools; 1345 1346 buff_size = adapter->req_mtu + VLAN_HLEN; 1347 buff_size = ALIGN(buff_size, L1_CACHE_BYTES); 1348 1349 for (i = 0; i < num_pools; i++) { 1350 dev_dbg(dev, "Init tx pool %d [%llu, %llu]\n", 1351 i, adapter->req_tx_entries_per_subcrq, buff_size); 1352 1353 rc = init_one_tx_pool(netdev, &adapter->tx_pool[i], 1354 pool_size, buff_size); 1355 if (rc) 1356 goto out_release; 1357 1358 rc = init_one_tx_pool(netdev, &adapter->tso_pool[i], 1359 IBMVNIC_TSO_BUFS, 1360 IBMVNIC_TSO_BUF_SZ); 1361 if (rc) 1362 goto out_release; 1363 } 1364 1365 adapter->prev_tx_pool_size = pool_size; 1366 adapter->prev_mtu = adapter->req_mtu; 1367 1368 update_ltb: 1369 /* NOTE: All tx_pools have the same number of buffers (which is 1370 * same as pool_size). All tso_pools have IBMVNIC_TSO_BUFS 1371 * buffers (see calls init_one_tx_pool() for these). 1372 * For consistency, we use tx_pool->num_buffers and 1373 * tso_pool->num_buffers below. 1374 */ 1375 rc = -1; 1376 for (i = 0; i < num_pools; i++) { 1377 struct ibmvnic_tx_pool *tso_pool; 1378 struct ibmvnic_tx_pool *tx_pool; 1379 1380 tx_pool = &adapter->tx_pool[i]; 1381 1382 dev_dbg(dev, "Updating LTB for tx pool %d [%d, %d]\n", 1383 i, tx_pool->num_buffers, tx_pool->buf_size); 1384 1385 rc = alloc_ltb_set(adapter, &tx_pool->ltb_set, 1386 tx_pool->num_buffers, tx_pool->buf_size); 1387 if (rc) 1388 goto out; 1389 1390 tx_pool->consumer_index = 0; 1391 tx_pool->producer_index = 0; 1392 1393 for (j = 0; j < tx_pool->num_buffers; j++) 1394 tx_pool->free_map[j] = j; 1395 1396 tso_pool = &adapter->tso_pool[i]; 1397 1398 dev_dbg(dev, "Updating LTB for tso pool %d [%d, %d]\n", 1399 i, tso_pool->num_buffers, tso_pool->buf_size); 1400 1401 rc = alloc_ltb_set(adapter, &tso_pool->ltb_set, 1402 tso_pool->num_buffers, tso_pool->buf_size); 1403 if (rc) 1404 goto out; 1405 1406 tso_pool->consumer_index = 0; 1407 tso_pool->producer_index = 0; 1408 1409 for (j = 0; j < tso_pool->num_buffers; j++) 1410 tso_pool->free_map[j] = j; 1411 } 1412 1413 return 0; 1414 out_release: 1415 release_tx_pools(adapter); 1416 out: 1417 /* We failed to allocate one or more LTBs or map them on the VIOS. 1418 * Hold onto the pools and any LTBs that we did allocate/map. 1419 */ 1420 return rc; 1421 } 1422 1423 static void ibmvnic_napi_enable(struct ibmvnic_adapter *adapter) 1424 { 1425 int i; 1426 1427 if (adapter->napi_enabled) 1428 return; 1429 1430 for (i = 0; i < adapter->req_rx_queues; i++) 1431 napi_enable(&adapter->napi[i]); 1432 1433 adapter->napi_enabled = true; 1434 } 1435 1436 static void ibmvnic_napi_disable(struct ibmvnic_adapter *adapter) 1437 { 1438 int i; 1439 1440 if (!adapter->napi_enabled) 1441 return; 1442 1443 for (i = 0; i < adapter->req_rx_queues; i++) { 1444 netdev_dbg(adapter->netdev, "Disabling napi[%d]\n", i); 1445 napi_disable(&adapter->napi[i]); 1446 } 1447 1448 adapter->napi_enabled = false; 1449 } 1450 1451 static int init_napi(struct ibmvnic_adapter *adapter) 1452 { 1453 int i; 1454 1455 adapter->napi = kcalloc(adapter->req_rx_queues, 1456 sizeof(struct napi_struct), GFP_KERNEL); 1457 if (!adapter->napi) 1458 return -ENOMEM; 1459 1460 for (i = 0; i < adapter->req_rx_queues; i++) { 1461 netdev_dbg(adapter->netdev, "Adding napi[%d]\n", i); 1462 netif_napi_add(adapter->netdev, &adapter->napi[i], 1463 ibmvnic_poll); 1464 } 1465 1466 adapter->num_active_rx_napi = adapter->req_rx_queues; 1467 return 0; 1468 } 1469 1470 static void release_napi(struct ibmvnic_adapter *adapter) 1471 { 1472 int i; 1473 1474 if (!adapter->napi) 1475 return; 1476 1477 for (i = 0; i < adapter->num_active_rx_napi; i++) { 1478 netdev_dbg(adapter->netdev, "Releasing napi[%d]\n", i); 1479 netif_napi_del(&adapter->napi[i]); 1480 } 1481 1482 kfree(adapter->napi); 1483 adapter->napi = NULL; 1484 adapter->num_active_rx_napi = 0; 1485 adapter->napi_enabled = false; 1486 } 1487 1488 static const char *adapter_state_to_string(enum vnic_state state) 1489 { 1490 switch (state) { 1491 case VNIC_PROBING: 1492 return "PROBING"; 1493 case VNIC_PROBED: 1494 return "PROBED"; 1495 case VNIC_OPENING: 1496 return "OPENING"; 1497 case VNIC_OPEN: 1498 return "OPEN"; 1499 case VNIC_CLOSING: 1500 return "CLOSING"; 1501 case VNIC_CLOSED: 1502 return "CLOSED"; 1503 case VNIC_REMOVING: 1504 return "REMOVING"; 1505 case VNIC_REMOVED: 1506 return "REMOVED"; 1507 case VNIC_DOWN: 1508 return "DOWN"; 1509 } 1510 return "UNKNOWN"; 1511 } 1512 1513 static int ibmvnic_login(struct net_device *netdev) 1514 { 1515 unsigned long flags, timeout = msecs_to_jiffies(20000); 1516 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 1517 int retry_count = 0; 1518 int retries = 10; 1519 bool retry; 1520 int rc; 1521 1522 do { 1523 retry = false; 1524 if (retry_count > retries) { 1525 netdev_warn(netdev, "Login attempts exceeded\n"); 1526 return -EACCES; 1527 } 1528 1529 adapter->init_done_rc = 0; 1530 reinit_completion(&adapter->init_done); 1531 rc = send_login(adapter); 1532 if (rc) 1533 return rc; 1534 1535 if (!wait_for_completion_timeout(&adapter->init_done, 1536 timeout)) { 1537 netdev_warn(netdev, "Login timed out\n"); 1538 adapter->login_pending = false; 1539 goto partial_reset; 1540 } 1541 1542 if (adapter->init_done_rc == ABORTED) { 1543 netdev_warn(netdev, "Login aborted, retrying...\n"); 1544 retry = true; 1545 adapter->init_done_rc = 0; 1546 retry_count++; 1547 /* FW or device may be busy, so 1548 * wait a bit before retrying login 1549 */ 1550 msleep(500); 1551 } else if (adapter->init_done_rc == PARTIALSUCCESS) { 1552 retry_count++; 1553 release_sub_crqs(adapter, 1); 1554 1555 retry = true; 1556 netdev_dbg(netdev, 1557 "Received partial success, retrying...\n"); 1558 adapter->init_done_rc = 0; 1559 reinit_completion(&adapter->init_done); 1560 send_query_cap(adapter); 1561 if (!wait_for_completion_timeout(&adapter->init_done, 1562 timeout)) { 1563 netdev_warn(netdev, 1564 "Capabilities query timed out\n"); 1565 return -ETIMEDOUT; 1566 } 1567 1568 rc = init_sub_crqs(adapter); 1569 if (rc) { 1570 netdev_warn(netdev, 1571 "SCRQ initialization failed\n"); 1572 return rc; 1573 } 1574 1575 rc = init_sub_crq_irqs(adapter); 1576 if (rc) { 1577 netdev_warn(netdev, 1578 "SCRQ irq initialization failed\n"); 1579 return rc; 1580 } 1581 /* Default/timeout error handling, reset and start fresh */ 1582 } else if (adapter->init_done_rc) { 1583 netdev_warn(netdev, "Adapter login failed, init_done_rc = %d\n", 1584 adapter->init_done_rc); 1585 1586 partial_reset: 1587 /* adapter login failed, so free any CRQs or sub-CRQs 1588 * and register again before attempting to login again. 1589 * If we don't do this then the VIOS may think that 1590 * we are already logged in and reject any subsequent 1591 * attempts 1592 */ 1593 netdev_warn(netdev, 1594 "Freeing and re-registering CRQs before attempting to login again\n"); 1595 retry = true; 1596 adapter->init_done_rc = 0; 1597 release_sub_crqs(adapter, true); 1598 /* Much of this is similar logic as ibmvnic_probe(), 1599 * we are essentially re-initializing communication 1600 * with the server. We really should not run any 1601 * resets/failovers here because this is already a form 1602 * of reset and we do not want parallel resets occurring 1603 */ 1604 do { 1605 reinit_init_done(adapter); 1606 /* Clear any failovers we got in the previous 1607 * pass since we are re-initializing the CRQ 1608 */ 1609 adapter->failover_pending = false; 1610 release_crq_queue(adapter); 1611 /* If we don't sleep here then we risk an 1612 * unnecessary failover event from the VIOS. 1613 * This is a known VIOS issue caused by a vnic 1614 * device freeing and registering a CRQ too 1615 * quickly. 1616 */ 1617 msleep(1500); 1618 /* Avoid any resets, since we are currently 1619 * resetting. 1620 */ 1621 spin_lock_irqsave(&adapter->rwi_lock, flags); 1622 flush_reset_queue(adapter); 1623 spin_unlock_irqrestore(&adapter->rwi_lock, 1624 flags); 1625 1626 rc = init_crq_queue(adapter); 1627 if (rc) { 1628 netdev_err(netdev, "login recovery: init CRQ failed %d\n", 1629 rc); 1630 return -EIO; 1631 } 1632 1633 rc = ibmvnic_reset_init(adapter, false); 1634 if (rc) 1635 netdev_err(netdev, "login recovery: Reset init failed %d\n", 1636 rc); 1637 /* IBMVNIC_CRQ_INIT will return EAGAIN if it 1638 * fails, since ibmvnic_reset_init will free 1639 * irq's in failure, we won't be able to receive 1640 * new CRQs so we need to keep trying. probe() 1641 * handles this similarly. 1642 */ 1643 } while (rc == -EAGAIN && retry_count++ < retries); 1644 } 1645 } while (retry); 1646 1647 __ibmvnic_set_mac(netdev, adapter->mac_addr); 1648 1649 netdev_dbg(netdev, "[S:%s] Login succeeded\n", adapter_state_to_string(adapter->state)); 1650 return 0; 1651 } 1652 1653 static void release_login_buffer(struct ibmvnic_adapter *adapter) 1654 { 1655 if (!adapter->login_buf) 1656 return; 1657 1658 dma_unmap_single(&adapter->vdev->dev, adapter->login_buf_token, 1659 adapter->login_buf_sz, DMA_TO_DEVICE); 1660 kfree(adapter->login_buf); 1661 adapter->login_buf = NULL; 1662 } 1663 1664 static void release_login_rsp_buffer(struct ibmvnic_adapter *adapter) 1665 { 1666 if (!adapter->login_rsp_buf) 1667 return; 1668 1669 dma_unmap_single(&adapter->vdev->dev, adapter->login_rsp_buf_token, 1670 adapter->login_rsp_buf_sz, DMA_FROM_DEVICE); 1671 kfree(adapter->login_rsp_buf); 1672 adapter->login_rsp_buf = NULL; 1673 } 1674 1675 static void release_resources(struct ibmvnic_adapter *adapter) 1676 { 1677 release_vpd_data(adapter); 1678 1679 release_napi(adapter); 1680 release_login_buffer(adapter); 1681 release_login_rsp_buffer(adapter); 1682 } 1683 1684 static int set_link_state(struct ibmvnic_adapter *adapter, u8 link_state) 1685 { 1686 struct net_device *netdev = adapter->netdev; 1687 unsigned long timeout = msecs_to_jiffies(20000); 1688 union ibmvnic_crq crq; 1689 bool resend; 1690 int rc; 1691 1692 netdev_dbg(netdev, "setting link state %d\n", link_state); 1693 1694 memset(&crq, 0, sizeof(crq)); 1695 crq.logical_link_state.first = IBMVNIC_CRQ_CMD; 1696 crq.logical_link_state.cmd = LOGICAL_LINK_STATE; 1697 crq.logical_link_state.link_state = link_state; 1698 1699 do { 1700 resend = false; 1701 1702 reinit_completion(&adapter->init_done); 1703 rc = ibmvnic_send_crq(adapter, &crq); 1704 if (rc) { 1705 netdev_err(netdev, "Failed to set link state\n"); 1706 return rc; 1707 } 1708 1709 if (!wait_for_completion_timeout(&adapter->init_done, 1710 timeout)) { 1711 netdev_err(netdev, "timeout setting link state\n"); 1712 return -ETIMEDOUT; 1713 } 1714 1715 if (adapter->init_done_rc == PARTIALSUCCESS) { 1716 /* Partuial success, delay and re-send */ 1717 mdelay(1000); 1718 resend = true; 1719 } else if (adapter->init_done_rc) { 1720 netdev_warn(netdev, "Unable to set link state, rc=%d\n", 1721 adapter->init_done_rc); 1722 return adapter->init_done_rc; 1723 } 1724 } while (resend); 1725 1726 return 0; 1727 } 1728 1729 static int set_real_num_queues(struct net_device *netdev) 1730 { 1731 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 1732 int rc; 1733 1734 netdev_dbg(netdev, "Setting real tx/rx queues (%llx/%llx)\n", 1735 adapter->req_tx_queues, adapter->req_rx_queues); 1736 1737 rc = netif_set_real_num_tx_queues(netdev, adapter->req_tx_queues); 1738 if (rc) { 1739 netdev_err(netdev, "failed to set the number of tx queues\n"); 1740 return rc; 1741 } 1742 1743 rc = netif_set_real_num_rx_queues(netdev, adapter->req_rx_queues); 1744 if (rc) 1745 netdev_err(netdev, "failed to set the number of rx queues\n"); 1746 1747 return rc; 1748 } 1749 1750 static int ibmvnic_get_vpd(struct ibmvnic_adapter *adapter) 1751 { 1752 struct device *dev = &adapter->vdev->dev; 1753 union ibmvnic_crq crq; 1754 int len = 0; 1755 int rc; 1756 1757 if (adapter->vpd->buff) 1758 len = adapter->vpd->len; 1759 1760 mutex_lock(&adapter->fw_lock); 1761 adapter->fw_done_rc = 0; 1762 reinit_completion(&adapter->fw_done); 1763 1764 crq.get_vpd_size.first = IBMVNIC_CRQ_CMD; 1765 crq.get_vpd_size.cmd = GET_VPD_SIZE; 1766 rc = ibmvnic_send_crq(adapter, &crq); 1767 if (rc) { 1768 mutex_unlock(&adapter->fw_lock); 1769 return rc; 1770 } 1771 1772 rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000); 1773 if (rc) { 1774 dev_err(dev, "Could not retrieve VPD size, rc = %d\n", rc); 1775 mutex_unlock(&adapter->fw_lock); 1776 return rc; 1777 } 1778 mutex_unlock(&adapter->fw_lock); 1779 1780 if (!adapter->vpd->len) 1781 return -ENODATA; 1782 1783 if (!adapter->vpd->buff) 1784 adapter->vpd->buff = kzalloc(adapter->vpd->len, GFP_KERNEL); 1785 else if (adapter->vpd->len != len) 1786 adapter->vpd->buff = 1787 krealloc(adapter->vpd->buff, 1788 adapter->vpd->len, GFP_KERNEL); 1789 1790 if (!adapter->vpd->buff) { 1791 dev_err(dev, "Could allocate VPD buffer\n"); 1792 return -ENOMEM; 1793 } 1794 1795 adapter->vpd->dma_addr = 1796 dma_map_single(dev, adapter->vpd->buff, adapter->vpd->len, 1797 DMA_FROM_DEVICE); 1798 if (dma_mapping_error(dev, adapter->vpd->dma_addr)) { 1799 dev_err(dev, "Could not map VPD buffer\n"); 1800 kfree(adapter->vpd->buff); 1801 adapter->vpd->buff = NULL; 1802 return -ENOMEM; 1803 } 1804 1805 mutex_lock(&adapter->fw_lock); 1806 adapter->fw_done_rc = 0; 1807 reinit_completion(&adapter->fw_done); 1808 1809 crq.get_vpd.first = IBMVNIC_CRQ_CMD; 1810 crq.get_vpd.cmd = GET_VPD; 1811 crq.get_vpd.ioba = cpu_to_be32(adapter->vpd->dma_addr); 1812 crq.get_vpd.len = cpu_to_be32((u32)adapter->vpd->len); 1813 rc = ibmvnic_send_crq(adapter, &crq); 1814 if (rc) { 1815 kfree(adapter->vpd->buff); 1816 adapter->vpd->buff = NULL; 1817 mutex_unlock(&adapter->fw_lock); 1818 return rc; 1819 } 1820 1821 rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000); 1822 if (rc) { 1823 dev_err(dev, "Unable to retrieve VPD, rc = %d\n", rc); 1824 kfree(adapter->vpd->buff); 1825 adapter->vpd->buff = NULL; 1826 mutex_unlock(&adapter->fw_lock); 1827 return rc; 1828 } 1829 1830 mutex_unlock(&adapter->fw_lock); 1831 return 0; 1832 } 1833 1834 static int init_resources(struct ibmvnic_adapter *adapter) 1835 { 1836 struct net_device *netdev = adapter->netdev; 1837 int rc; 1838 1839 rc = set_real_num_queues(netdev); 1840 if (rc) 1841 return rc; 1842 1843 adapter->vpd = kzalloc(sizeof(*adapter->vpd), GFP_KERNEL); 1844 if (!adapter->vpd) 1845 return -ENOMEM; 1846 1847 /* Vital Product Data (VPD) */ 1848 rc = ibmvnic_get_vpd(adapter); 1849 if (rc) { 1850 netdev_err(netdev, "failed to initialize Vital Product Data (VPD)\n"); 1851 return rc; 1852 } 1853 1854 rc = init_napi(adapter); 1855 if (rc) 1856 return rc; 1857 1858 send_query_map(adapter); 1859 1860 rc = init_rx_pools(netdev); 1861 if (rc) 1862 return rc; 1863 1864 rc = init_tx_pools(netdev); 1865 return rc; 1866 } 1867 1868 static int __ibmvnic_open(struct net_device *netdev) 1869 { 1870 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 1871 enum vnic_state prev_state = adapter->state; 1872 int i, rc; 1873 1874 adapter->state = VNIC_OPENING; 1875 replenish_pools(adapter); 1876 ibmvnic_napi_enable(adapter); 1877 1878 /* We're ready to receive frames, enable the sub-crq interrupts and 1879 * set the logical link state to up 1880 */ 1881 for (i = 0; i < adapter->req_rx_queues; i++) { 1882 netdev_dbg(netdev, "Enabling rx_scrq[%d] irq\n", i); 1883 if (prev_state == VNIC_CLOSED) 1884 enable_irq(adapter->rx_scrq[i]->irq); 1885 enable_scrq_irq(adapter, adapter->rx_scrq[i]); 1886 } 1887 1888 for (i = 0; i < adapter->req_tx_queues; i++) { 1889 netdev_dbg(netdev, "Enabling tx_scrq[%d] irq\n", i); 1890 if (prev_state == VNIC_CLOSED) 1891 enable_irq(adapter->tx_scrq[i]->irq); 1892 enable_scrq_irq(adapter, adapter->tx_scrq[i]); 1893 /* netdev_tx_reset_queue will reset dql stats. During NON_FATAL 1894 * resets, don't reset the stats because there could be batched 1895 * skb's waiting to be sent. If we reset dql stats, we risk 1896 * num_completed being greater than num_queued. This will cause 1897 * a BUG_ON in dql_completed(). 1898 */ 1899 if (adapter->reset_reason != VNIC_RESET_NON_FATAL) 1900 netdev_tx_reset_queue(netdev_get_tx_queue(netdev, i)); 1901 } 1902 1903 rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_UP); 1904 if (rc) { 1905 ibmvnic_napi_disable(adapter); 1906 ibmvnic_disable_irqs(adapter); 1907 return rc; 1908 } 1909 1910 adapter->tx_queues_active = true; 1911 1912 /* Since queues were stopped until now, there shouldn't be any 1913 * one in ibmvnic_complete_tx() or ibmvnic_xmit() so maybe we 1914 * don't need the synchronize_rcu()? Leaving it for consistency 1915 * with setting ->tx_queues_active = false. 1916 */ 1917 synchronize_rcu(); 1918 1919 netif_tx_start_all_queues(netdev); 1920 1921 if (prev_state == VNIC_CLOSED) { 1922 for (i = 0; i < adapter->req_rx_queues; i++) 1923 napi_schedule(&adapter->napi[i]); 1924 } 1925 1926 adapter->state = VNIC_OPEN; 1927 return rc; 1928 } 1929 1930 static int ibmvnic_open(struct net_device *netdev) 1931 { 1932 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 1933 int rc; 1934 1935 ASSERT_RTNL(); 1936 1937 /* If device failover is pending or we are about to reset, just set 1938 * device state and return. Device operation will be handled by reset 1939 * routine. 1940 * 1941 * It should be safe to overwrite the adapter->state here. Since 1942 * we hold the rtnl, either the reset has not actually started or 1943 * the rtnl got dropped during the set_link_state() in do_reset(). 1944 * In the former case, no one else is changing the state (again we 1945 * have the rtnl) and in the latter case, do_reset() will detect and 1946 * honor our setting below. 1947 */ 1948 if (adapter->failover_pending || (test_bit(0, &adapter->resetting))) { 1949 netdev_dbg(netdev, "[S:%s FOP:%d] Resetting, deferring open\n", 1950 adapter_state_to_string(adapter->state), 1951 adapter->failover_pending); 1952 adapter->state = VNIC_OPEN; 1953 rc = 0; 1954 goto out; 1955 } 1956 1957 if (adapter->state != VNIC_CLOSED) { 1958 rc = ibmvnic_login(netdev); 1959 if (rc) 1960 goto out; 1961 1962 rc = init_resources(adapter); 1963 if (rc) { 1964 netdev_err(netdev, "failed to initialize resources\n"); 1965 goto out; 1966 } 1967 } 1968 1969 rc = __ibmvnic_open(netdev); 1970 1971 out: 1972 /* If open failed and there is a pending failover or in-progress reset, 1973 * set device state and return. Device operation will be handled by 1974 * reset routine. See also comments above regarding rtnl. 1975 */ 1976 if (rc && 1977 (adapter->failover_pending || (test_bit(0, &adapter->resetting)))) { 1978 adapter->state = VNIC_OPEN; 1979 rc = 0; 1980 } 1981 1982 if (rc) { 1983 release_resources(adapter); 1984 release_rx_pools(adapter); 1985 release_tx_pools(adapter); 1986 } 1987 1988 return rc; 1989 } 1990 1991 static void clean_rx_pools(struct ibmvnic_adapter *adapter) 1992 { 1993 struct ibmvnic_rx_pool *rx_pool; 1994 struct ibmvnic_rx_buff *rx_buff; 1995 u64 rx_entries; 1996 int rx_scrqs; 1997 int i, j; 1998 1999 if (!adapter->rx_pool) 2000 return; 2001 2002 rx_scrqs = adapter->num_active_rx_pools; 2003 rx_entries = adapter->req_rx_add_entries_per_subcrq; 2004 2005 /* Free any remaining skbs in the rx buffer pools */ 2006 for (i = 0; i < rx_scrqs; i++) { 2007 rx_pool = &adapter->rx_pool[i]; 2008 if (!rx_pool || !rx_pool->rx_buff) 2009 continue; 2010 2011 netdev_dbg(adapter->netdev, "Cleaning rx_pool[%d]\n", i); 2012 for (j = 0; j < rx_entries; j++) { 2013 rx_buff = &rx_pool->rx_buff[j]; 2014 if (rx_buff && rx_buff->skb) { 2015 dev_kfree_skb_any(rx_buff->skb); 2016 rx_buff->skb = NULL; 2017 } 2018 } 2019 } 2020 } 2021 2022 static void clean_one_tx_pool(struct ibmvnic_adapter *adapter, 2023 struct ibmvnic_tx_pool *tx_pool) 2024 { 2025 struct ibmvnic_tx_buff *tx_buff; 2026 u64 tx_entries; 2027 int i; 2028 2029 if (!tx_pool || !tx_pool->tx_buff) 2030 return; 2031 2032 tx_entries = tx_pool->num_buffers; 2033 2034 for (i = 0; i < tx_entries; i++) { 2035 tx_buff = &tx_pool->tx_buff[i]; 2036 if (tx_buff && tx_buff->skb) { 2037 dev_kfree_skb_any(tx_buff->skb); 2038 tx_buff->skb = NULL; 2039 } 2040 } 2041 } 2042 2043 static void clean_tx_pools(struct ibmvnic_adapter *adapter) 2044 { 2045 int tx_scrqs; 2046 int i; 2047 2048 if (!adapter->tx_pool || !adapter->tso_pool) 2049 return; 2050 2051 tx_scrqs = adapter->num_active_tx_pools; 2052 2053 /* Free any remaining skbs in the tx buffer pools */ 2054 for (i = 0; i < tx_scrqs; i++) { 2055 netdev_dbg(adapter->netdev, "Cleaning tx_pool[%d]\n", i); 2056 clean_one_tx_pool(adapter, &adapter->tx_pool[i]); 2057 clean_one_tx_pool(adapter, &adapter->tso_pool[i]); 2058 } 2059 } 2060 2061 static void ibmvnic_disable_irqs(struct ibmvnic_adapter *adapter) 2062 { 2063 struct net_device *netdev = adapter->netdev; 2064 int i; 2065 2066 if (adapter->tx_scrq) { 2067 for (i = 0; i < adapter->req_tx_queues; i++) 2068 if (adapter->tx_scrq[i]->irq) { 2069 netdev_dbg(netdev, 2070 "Disabling tx_scrq[%d] irq\n", i); 2071 disable_scrq_irq(adapter, adapter->tx_scrq[i]); 2072 disable_irq(adapter->tx_scrq[i]->irq); 2073 } 2074 } 2075 2076 if (adapter->rx_scrq) { 2077 for (i = 0; i < adapter->req_rx_queues; i++) { 2078 if (adapter->rx_scrq[i]->irq) { 2079 netdev_dbg(netdev, 2080 "Disabling rx_scrq[%d] irq\n", i); 2081 disable_scrq_irq(adapter, adapter->rx_scrq[i]); 2082 disable_irq(adapter->rx_scrq[i]->irq); 2083 } 2084 } 2085 } 2086 } 2087 2088 static void ibmvnic_cleanup(struct net_device *netdev) 2089 { 2090 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 2091 2092 /* ensure that transmissions are stopped if called by do_reset */ 2093 2094 adapter->tx_queues_active = false; 2095 2096 /* Ensure complete_tx() and ibmvnic_xmit() see ->tx_queues_active 2097 * update so they don't restart a queue after we stop it below. 2098 */ 2099 synchronize_rcu(); 2100 2101 if (test_bit(0, &adapter->resetting)) 2102 netif_tx_disable(netdev); 2103 else 2104 netif_tx_stop_all_queues(netdev); 2105 2106 ibmvnic_napi_disable(adapter); 2107 ibmvnic_disable_irqs(adapter); 2108 } 2109 2110 static int __ibmvnic_close(struct net_device *netdev) 2111 { 2112 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 2113 int rc = 0; 2114 2115 adapter->state = VNIC_CLOSING; 2116 rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN); 2117 adapter->state = VNIC_CLOSED; 2118 return rc; 2119 } 2120 2121 static int ibmvnic_close(struct net_device *netdev) 2122 { 2123 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 2124 int rc; 2125 2126 netdev_dbg(netdev, "[S:%s FOP:%d FRR:%d] Closing\n", 2127 adapter_state_to_string(adapter->state), 2128 adapter->failover_pending, 2129 adapter->force_reset_recovery); 2130 2131 /* If device failover is pending, just set device state and return. 2132 * Device operation will be handled by reset routine. 2133 */ 2134 if (adapter->failover_pending) { 2135 adapter->state = VNIC_CLOSED; 2136 return 0; 2137 } 2138 2139 rc = __ibmvnic_close(netdev); 2140 ibmvnic_cleanup(netdev); 2141 clean_rx_pools(adapter); 2142 clean_tx_pools(adapter); 2143 2144 return rc; 2145 } 2146 2147 /** 2148 * get_hdr_lens - fills list of L2/L3/L4 hdr lens 2149 * @hdr_field: bitfield determining needed headers 2150 * @skb: socket buffer 2151 * @hdr_len: array of header lengths to be filled 2152 * 2153 * Reads hdr_field to determine which headers are needed by firmware. 2154 * Builds a buffer containing these headers. Saves individual header 2155 * lengths and total buffer length to be used to build descriptors. 2156 * 2157 * Return: total len of all headers 2158 */ 2159 static int get_hdr_lens(u8 hdr_field, struct sk_buff *skb, 2160 int *hdr_len) 2161 { 2162 int len = 0; 2163 2164 2165 if ((hdr_field >> 6) & 1) { 2166 hdr_len[0] = skb_mac_header_len(skb); 2167 len += hdr_len[0]; 2168 } 2169 2170 if ((hdr_field >> 5) & 1) { 2171 hdr_len[1] = skb_network_header_len(skb); 2172 len += hdr_len[1]; 2173 } 2174 2175 if (!((hdr_field >> 4) & 1)) 2176 return len; 2177 2178 if (skb->protocol == htons(ETH_P_IP)) { 2179 if (ip_hdr(skb)->protocol == IPPROTO_TCP) 2180 hdr_len[2] = tcp_hdrlen(skb); 2181 else if (ip_hdr(skb)->protocol == IPPROTO_UDP) 2182 hdr_len[2] = sizeof(struct udphdr); 2183 } else if (skb->protocol == htons(ETH_P_IPV6)) { 2184 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP) 2185 hdr_len[2] = tcp_hdrlen(skb); 2186 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP) 2187 hdr_len[2] = sizeof(struct udphdr); 2188 } 2189 2190 return len + hdr_len[2]; 2191 } 2192 2193 /** 2194 * create_hdr_descs - create header and header extension descriptors 2195 * @hdr_field: bitfield determining needed headers 2196 * @hdr_data: buffer containing header data 2197 * @len: length of data buffer 2198 * @hdr_len: array of individual header lengths 2199 * @scrq_arr: descriptor array 2200 * 2201 * Creates header and, if needed, header extension descriptors and 2202 * places them in a descriptor array, scrq_arr 2203 * 2204 * Return: Number of header descs 2205 */ 2206 2207 static int create_hdr_descs(u8 hdr_field, u8 *hdr_data, int len, int *hdr_len, 2208 union sub_crq *scrq_arr) 2209 { 2210 union sub_crq *hdr_desc; 2211 int tmp_len = len; 2212 int num_descs = 0; 2213 u8 *data, *cur; 2214 int tmp; 2215 2216 while (tmp_len > 0) { 2217 cur = hdr_data + len - tmp_len; 2218 2219 hdr_desc = &scrq_arr[num_descs]; 2220 if (num_descs) { 2221 data = hdr_desc->hdr_ext.data; 2222 tmp = tmp_len > 29 ? 29 : tmp_len; 2223 hdr_desc->hdr_ext.first = IBMVNIC_CRQ_CMD; 2224 hdr_desc->hdr_ext.type = IBMVNIC_HDR_EXT_DESC; 2225 hdr_desc->hdr_ext.len = tmp; 2226 } else { 2227 data = hdr_desc->hdr.data; 2228 tmp = tmp_len > 24 ? 24 : tmp_len; 2229 hdr_desc->hdr.first = IBMVNIC_CRQ_CMD; 2230 hdr_desc->hdr.type = IBMVNIC_HDR_DESC; 2231 hdr_desc->hdr.len = tmp; 2232 hdr_desc->hdr.l2_len = (u8)hdr_len[0]; 2233 hdr_desc->hdr.l3_len = cpu_to_be16((u16)hdr_len[1]); 2234 hdr_desc->hdr.l4_len = (u8)hdr_len[2]; 2235 hdr_desc->hdr.flag = hdr_field << 1; 2236 } 2237 memcpy(data, cur, tmp); 2238 tmp_len -= tmp; 2239 num_descs++; 2240 } 2241 2242 return num_descs; 2243 } 2244 2245 /** 2246 * build_hdr_descs_arr - build a header descriptor array 2247 * @skb: tx socket buffer 2248 * @indir_arr: indirect array 2249 * @num_entries: number of descriptors to be sent 2250 * @hdr_field: bit field determining which headers will be sent 2251 * 2252 * This function will build a TX descriptor array with applicable 2253 * L2/L3/L4 packet header descriptors to be sent by send_subcrq_indirect. 2254 */ 2255 2256 static void build_hdr_descs_arr(struct sk_buff *skb, 2257 union sub_crq *indir_arr, 2258 int *num_entries, u8 hdr_field) 2259 { 2260 int hdr_len[3] = {0, 0, 0}; 2261 int tot_len; 2262 2263 tot_len = get_hdr_lens(hdr_field, skb, hdr_len); 2264 *num_entries += create_hdr_descs(hdr_field, skb_mac_header(skb), 2265 tot_len, hdr_len, indir_arr + 1); 2266 } 2267 2268 static int ibmvnic_xmit_workarounds(struct sk_buff *skb, 2269 struct net_device *netdev) 2270 { 2271 /* For some backing devices, mishandling of small packets 2272 * can result in a loss of connection or TX stall. Device 2273 * architects recommend that no packet should be smaller 2274 * than the minimum MTU value provided to the driver, so 2275 * pad any packets to that length 2276 */ 2277 if (skb->len < netdev->min_mtu) 2278 return skb_put_padto(skb, netdev->min_mtu); 2279 2280 return 0; 2281 } 2282 2283 static void ibmvnic_tx_scrq_clean_buffer(struct ibmvnic_adapter *adapter, 2284 struct ibmvnic_sub_crq_queue *tx_scrq) 2285 { 2286 struct ibmvnic_ind_xmit_queue *ind_bufp; 2287 struct ibmvnic_tx_buff *tx_buff; 2288 struct ibmvnic_tx_pool *tx_pool; 2289 union sub_crq tx_scrq_entry; 2290 int queue_num; 2291 int entries; 2292 int index; 2293 int i; 2294 2295 ind_bufp = &tx_scrq->ind_buf; 2296 entries = (u64)ind_bufp->index; 2297 queue_num = tx_scrq->pool_index; 2298 2299 for (i = entries - 1; i >= 0; --i) { 2300 tx_scrq_entry = ind_bufp->indir_arr[i]; 2301 if (tx_scrq_entry.v1.type != IBMVNIC_TX_DESC) 2302 continue; 2303 index = be32_to_cpu(tx_scrq_entry.v1.correlator); 2304 if (index & IBMVNIC_TSO_POOL_MASK) { 2305 tx_pool = &adapter->tso_pool[queue_num]; 2306 index &= ~IBMVNIC_TSO_POOL_MASK; 2307 } else { 2308 tx_pool = &adapter->tx_pool[queue_num]; 2309 } 2310 tx_pool->free_map[tx_pool->consumer_index] = index; 2311 tx_pool->consumer_index = tx_pool->consumer_index == 0 ? 2312 tx_pool->num_buffers - 1 : 2313 tx_pool->consumer_index - 1; 2314 tx_buff = &tx_pool->tx_buff[index]; 2315 adapter->netdev->stats.tx_packets--; 2316 adapter->netdev->stats.tx_bytes -= tx_buff->skb->len; 2317 adapter->tx_stats_buffers[queue_num].batched_packets--; 2318 adapter->tx_stats_buffers[queue_num].bytes -= 2319 tx_buff->skb->len; 2320 dev_kfree_skb_any(tx_buff->skb); 2321 tx_buff->skb = NULL; 2322 adapter->netdev->stats.tx_dropped++; 2323 } 2324 2325 ind_bufp->index = 0; 2326 2327 if (atomic_sub_return(entries, &tx_scrq->used) <= 2328 (adapter->req_tx_entries_per_subcrq / 2) && 2329 __netif_subqueue_stopped(adapter->netdev, queue_num)) { 2330 rcu_read_lock(); 2331 2332 if (adapter->tx_queues_active) { 2333 netif_wake_subqueue(adapter->netdev, queue_num); 2334 netdev_dbg(adapter->netdev, "Started queue %d\n", 2335 queue_num); 2336 } 2337 2338 rcu_read_unlock(); 2339 } 2340 } 2341 2342 static int send_subcrq_direct(struct ibmvnic_adapter *adapter, 2343 u64 remote_handle, u64 *entry) 2344 { 2345 unsigned int ua = adapter->vdev->unit_address; 2346 struct device *dev = &adapter->vdev->dev; 2347 int rc; 2348 2349 /* Make sure the hypervisor sees the complete request */ 2350 dma_wmb(); 2351 rc = plpar_hcall_norets(H_SEND_SUB_CRQ, ua, 2352 cpu_to_be64(remote_handle), 2353 cpu_to_be64(entry[0]), cpu_to_be64(entry[1]), 2354 cpu_to_be64(entry[2]), cpu_to_be64(entry[3])); 2355 2356 if (rc) 2357 print_subcrq_error(dev, rc, __func__); 2358 2359 return rc; 2360 } 2361 2362 static int ibmvnic_tx_scrq_flush(struct ibmvnic_adapter *adapter, 2363 struct ibmvnic_sub_crq_queue *tx_scrq, 2364 bool indirect) 2365 { 2366 struct ibmvnic_ind_xmit_queue *ind_bufp; 2367 u64 dma_addr; 2368 u64 entries; 2369 u64 handle; 2370 int rc; 2371 2372 ind_bufp = &tx_scrq->ind_buf; 2373 dma_addr = (u64)ind_bufp->indir_dma; 2374 entries = (u64)ind_bufp->index; 2375 handle = tx_scrq->handle; 2376 2377 if (!entries) 2378 return 0; 2379 2380 if (indirect) 2381 rc = send_subcrq_indirect(adapter, handle, dma_addr, entries); 2382 else 2383 rc = send_subcrq_direct(adapter, handle, 2384 (u64 *)ind_bufp->indir_arr); 2385 2386 if (rc) 2387 ibmvnic_tx_scrq_clean_buffer(adapter, tx_scrq); 2388 else 2389 ind_bufp->index = 0; 2390 return rc; 2391 } 2392 2393 static netdev_tx_t ibmvnic_xmit(struct sk_buff *skb, struct net_device *netdev) 2394 { 2395 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 2396 int queue_num = skb_get_queue_mapping(skb); 2397 u8 *hdrs = (u8 *)&adapter->tx_rx_desc_req; 2398 struct device *dev = &adapter->vdev->dev; 2399 struct ibmvnic_ind_xmit_queue *ind_bufp; 2400 struct ibmvnic_tx_buff *tx_buff = NULL; 2401 struct ibmvnic_sub_crq_queue *tx_scrq; 2402 struct ibmvnic_long_term_buff *ltb; 2403 struct ibmvnic_tx_pool *tx_pool; 2404 unsigned int tx_send_failed = 0; 2405 netdev_tx_t ret = NETDEV_TX_OK; 2406 unsigned int tx_map_failed = 0; 2407 union sub_crq indir_arr[16]; 2408 unsigned int tx_dropped = 0; 2409 unsigned int tx_dpackets = 0; 2410 unsigned int tx_bpackets = 0; 2411 unsigned int tx_bytes = 0; 2412 dma_addr_t data_dma_addr; 2413 struct netdev_queue *txq; 2414 unsigned long lpar_rc; 2415 union sub_crq tx_crq; 2416 unsigned int offset; 2417 bool use_scrq_send_direct = false; 2418 int num_entries = 1; 2419 unsigned char *dst; 2420 int bufidx = 0; 2421 u8 proto = 0; 2422 2423 /* If a reset is in progress, drop the packet since 2424 * the scrqs may get torn down. Otherwise use the 2425 * rcu to ensure reset waits for us to complete. 2426 */ 2427 rcu_read_lock(); 2428 if (!adapter->tx_queues_active) { 2429 dev_kfree_skb_any(skb); 2430 2431 tx_send_failed++; 2432 tx_dropped++; 2433 ret = NETDEV_TX_OK; 2434 goto out; 2435 } 2436 2437 tx_scrq = adapter->tx_scrq[queue_num]; 2438 txq = netdev_get_tx_queue(netdev, queue_num); 2439 ind_bufp = &tx_scrq->ind_buf; 2440 2441 if (ibmvnic_xmit_workarounds(skb, netdev)) { 2442 tx_dropped++; 2443 tx_send_failed++; 2444 ret = NETDEV_TX_OK; 2445 lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq, true); 2446 if (lpar_rc != H_SUCCESS) 2447 goto tx_err; 2448 goto out; 2449 } 2450 2451 if (skb_is_gso(skb)) 2452 tx_pool = &adapter->tso_pool[queue_num]; 2453 else 2454 tx_pool = &adapter->tx_pool[queue_num]; 2455 2456 bufidx = tx_pool->free_map[tx_pool->consumer_index]; 2457 2458 if (bufidx == IBMVNIC_INVALID_MAP) { 2459 dev_kfree_skb_any(skb); 2460 tx_send_failed++; 2461 tx_dropped++; 2462 ret = NETDEV_TX_OK; 2463 lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq, true); 2464 if (lpar_rc != H_SUCCESS) 2465 goto tx_err; 2466 goto out; 2467 } 2468 2469 tx_pool->free_map[tx_pool->consumer_index] = IBMVNIC_INVALID_MAP; 2470 2471 map_txpool_buf_to_ltb(tx_pool, bufidx, <b, &offset); 2472 2473 dst = ltb->buff + offset; 2474 memset(dst, 0, tx_pool->buf_size); 2475 data_dma_addr = ltb->addr + offset; 2476 2477 /* if we are going to send_subcrq_direct this then we need to 2478 * update the checksum before copying the data into ltb. Essentially 2479 * these packets force disable CSO so that we can guarantee that 2480 * FW does not need header info and we can send direct. Also, vnic 2481 * server must be able to xmit standard packets without header data 2482 */ 2483 if (*hdrs == 0 && !skb_is_gso(skb) && 2484 !ind_bufp->index && !netdev_xmit_more()) { 2485 use_scrq_send_direct = true; 2486 if (skb->ip_summed == CHECKSUM_PARTIAL && 2487 skb_checksum_help(skb)) 2488 use_scrq_send_direct = false; 2489 } 2490 2491 if (skb_shinfo(skb)->nr_frags) { 2492 int cur, i; 2493 2494 /* Copy the head */ 2495 skb_copy_from_linear_data(skb, dst, skb_headlen(skb)); 2496 cur = skb_headlen(skb); 2497 2498 /* Copy the frags */ 2499 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 2500 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 2501 2502 memcpy(dst + cur, skb_frag_address(frag), 2503 skb_frag_size(frag)); 2504 cur += skb_frag_size(frag); 2505 } 2506 } else { 2507 skb_copy_from_linear_data(skb, dst, skb->len); 2508 } 2509 2510 tx_pool->consumer_index = 2511 (tx_pool->consumer_index + 1) % tx_pool->num_buffers; 2512 2513 tx_buff = &tx_pool->tx_buff[bufidx]; 2514 2515 /* Sanity checks on our free map to make sure it points to an index 2516 * that is not being occupied by another skb. If skb memory is 2517 * not freed then we see congestion control kick in and halt tx. 2518 */ 2519 if (unlikely(tx_buff->skb)) { 2520 dev_warn_ratelimited(dev, "TX free map points to untracked skb (%s %d idx=%d)\n", 2521 skb_is_gso(skb) ? "tso_pool" : "tx_pool", 2522 queue_num, bufidx); 2523 dev_kfree_skb_any(tx_buff->skb); 2524 } 2525 2526 tx_buff->skb = skb; 2527 tx_buff->index = bufidx; 2528 tx_buff->pool_index = queue_num; 2529 2530 memset(&tx_crq, 0, sizeof(tx_crq)); 2531 tx_crq.v1.first = IBMVNIC_CRQ_CMD; 2532 tx_crq.v1.type = IBMVNIC_TX_DESC; 2533 tx_crq.v1.n_crq_elem = 1; 2534 tx_crq.v1.n_sge = 1; 2535 tx_crq.v1.flags1 = IBMVNIC_TX_COMP_NEEDED; 2536 2537 if (skb_is_gso(skb)) 2538 tx_crq.v1.correlator = 2539 cpu_to_be32(bufidx | IBMVNIC_TSO_POOL_MASK); 2540 else 2541 tx_crq.v1.correlator = cpu_to_be32(bufidx); 2542 tx_crq.v1.dma_reg = cpu_to_be16(ltb->map_id); 2543 tx_crq.v1.sge_len = cpu_to_be32(skb->len); 2544 tx_crq.v1.ioba = cpu_to_be64(data_dma_addr); 2545 2546 if (adapter->vlan_header_insertion && skb_vlan_tag_present(skb)) { 2547 tx_crq.v1.flags2 |= IBMVNIC_TX_VLAN_INSERT; 2548 tx_crq.v1.vlan_id = cpu_to_be16(skb->vlan_tci); 2549 } 2550 2551 if (skb->protocol == htons(ETH_P_IP)) { 2552 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV4; 2553 proto = ip_hdr(skb)->protocol; 2554 } else if (skb->protocol == htons(ETH_P_IPV6)) { 2555 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV6; 2556 proto = ipv6_hdr(skb)->nexthdr; 2557 } 2558 2559 if (proto == IPPROTO_TCP) 2560 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_TCP; 2561 else if (proto == IPPROTO_UDP) 2562 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_UDP; 2563 2564 if (skb->ip_summed == CHECKSUM_PARTIAL) { 2565 tx_crq.v1.flags1 |= IBMVNIC_TX_CHKSUM_OFFLOAD; 2566 hdrs += 2; 2567 } 2568 if (skb_is_gso(skb)) { 2569 tx_crq.v1.flags1 |= IBMVNIC_TX_LSO; 2570 tx_crq.v1.mss = cpu_to_be16(skb_shinfo(skb)->gso_size); 2571 hdrs += 2; 2572 } else if (use_scrq_send_direct) { 2573 /* See above comment, CSO disabled with direct xmit */ 2574 tx_crq.v1.flags1 &= ~(IBMVNIC_TX_CHKSUM_OFFLOAD); 2575 ind_bufp->index = 1; 2576 tx_buff->num_entries = 1; 2577 netdev_tx_sent_queue(txq, skb->len); 2578 ind_bufp->indir_arr[0] = tx_crq; 2579 lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq, false); 2580 if (lpar_rc != H_SUCCESS) 2581 goto tx_err; 2582 2583 tx_dpackets++; 2584 goto early_exit; 2585 } 2586 2587 if ((*hdrs >> 7) & 1) 2588 build_hdr_descs_arr(skb, indir_arr, &num_entries, *hdrs); 2589 2590 tx_crq.v1.n_crq_elem = num_entries; 2591 tx_buff->num_entries = num_entries; 2592 /* flush buffer if current entry can not fit */ 2593 if (num_entries + ind_bufp->index > IBMVNIC_MAX_IND_DESCS) { 2594 lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq, true); 2595 if (lpar_rc != H_SUCCESS) 2596 goto tx_flush_err; 2597 } 2598 2599 indir_arr[0] = tx_crq; 2600 memcpy(&ind_bufp->indir_arr[ind_bufp->index], &indir_arr[0], 2601 num_entries * sizeof(struct ibmvnic_generic_scrq)); 2602 2603 ind_bufp->index += num_entries; 2604 if (__netdev_tx_sent_queue(txq, skb->len, 2605 netdev_xmit_more() && 2606 ind_bufp->index < IBMVNIC_MAX_IND_DESCS)) { 2607 lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq, true); 2608 if (lpar_rc != H_SUCCESS) 2609 goto tx_err; 2610 } 2611 2612 tx_bpackets++; 2613 2614 early_exit: 2615 if (atomic_add_return(num_entries, &tx_scrq->used) 2616 >= adapter->req_tx_entries_per_subcrq) { 2617 netdev_dbg(netdev, "Stopping queue %d\n", queue_num); 2618 netif_stop_subqueue(netdev, queue_num); 2619 } 2620 2621 tx_bytes += skb->len; 2622 txq_trans_cond_update(txq); 2623 ret = NETDEV_TX_OK; 2624 goto out; 2625 2626 tx_flush_err: 2627 dev_kfree_skb_any(skb); 2628 tx_buff->skb = NULL; 2629 tx_pool->consumer_index = tx_pool->consumer_index == 0 ? 2630 tx_pool->num_buffers - 1 : 2631 tx_pool->consumer_index - 1; 2632 tx_dropped++; 2633 tx_err: 2634 if (lpar_rc != H_CLOSED && lpar_rc != H_PARAMETER) 2635 dev_err_ratelimited(dev, "tx: send failed\n"); 2636 2637 if (lpar_rc == H_CLOSED || adapter->failover_pending) { 2638 /* Disable TX and report carrier off if queue is closed 2639 * or pending failover. 2640 * Firmware guarantees that a signal will be sent to the 2641 * driver, triggering a reset or some other action. 2642 */ 2643 netif_tx_stop_all_queues(netdev); 2644 netif_carrier_off(netdev); 2645 } 2646 out: 2647 rcu_read_unlock(); 2648 netdev->stats.tx_dropped += tx_dropped; 2649 netdev->stats.tx_bytes += tx_bytes; 2650 netdev->stats.tx_packets += tx_bpackets + tx_dpackets; 2651 adapter->tx_send_failed += tx_send_failed; 2652 adapter->tx_map_failed += tx_map_failed; 2653 adapter->tx_stats_buffers[queue_num].batched_packets += tx_bpackets; 2654 adapter->tx_stats_buffers[queue_num].direct_packets += tx_dpackets; 2655 adapter->tx_stats_buffers[queue_num].bytes += tx_bytes; 2656 adapter->tx_stats_buffers[queue_num].dropped_packets += tx_dropped; 2657 2658 return ret; 2659 } 2660 2661 static void ibmvnic_set_multi(struct net_device *netdev) 2662 { 2663 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 2664 struct netdev_hw_addr *ha; 2665 union ibmvnic_crq crq; 2666 2667 memset(&crq, 0, sizeof(crq)); 2668 crq.request_capability.first = IBMVNIC_CRQ_CMD; 2669 crq.request_capability.cmd = REQUEST_CAPABILITY; 2670 2671 if (netdev->flags & IFF_PROMISC) { 2672 if (!adapter->promisc_supported) 2673 return; 2674 } else { 2675 if (netdev->flags & IFF_ALLMULTI) { 2676 /* Accept all multicast */ 2677 memset(&crq, 0, sizeof(crq)); 2678 crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD; 2679 crq.multicast_ctrl.cmd = MULTICAST_CTRL; 2680 crq.multicast_ctrl.flags = IBMVNIC_ENABLE_ALL; 2681 ibmvnic_send_crq(adapter, &crq); 2682 } else if (netdev_mc_empty(netdev)) { 2683 /* Reject all multicast */ 2684 memset(&crq, 0, sizeof(crq)); 2685 crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD; 2686 crq.multicast_ctrl.cmd = MULTICAST_CTRL; 2687 crq.multicast_ctrl.flags = IBMVNIC_DISABLE_ALL; 2688 ibmvnic_send_crq(adapter, &crq); 2689 } else { 2690 /* Accept one or more multicast(s) */ 2691 netdev_for_each_mc_addr(ha, netdev) { 2692 memset(&crq, 0, sizeof(crq)); 2693 crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD; 2694 crq.multicast_ctrl.cmd = MULTICAST_CTRL; 2695 crq.multicast_ctrl.flags = IBMVNIC_ENABLE_MC; 2696 ether_addr_copy(&crq.multicast_ctrl.mac_addr[0], 2697 ha->addr); 2698 ibmvnic_send_crq(adapter, &crq); 2699 } 2700 } 2701 } 2702 } 2703 2704 static int __ibmvnic_set_mac(struct net_device *netdev, u8 *dev_addr) 2705 { 2706 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 2707 union ibmvnic_crq crq; 2708 int rc; 2709 2710 if (!is_valid_ether_addr(dev_addr)) { 2711 rc = -EADDRNOTAVAIL; 2712 goto err; 2713 } 2714 2715 memset(&crq, 0, sizeof(crq)); 2716 crq.change_mac_addr.first = IBMVNIC_CRQ_CMD; 2717 crq.change_mac_addr.cmd = CHANGE_MAC_ADDR; 2718 ether_addr_copy(&crq.change_mac_addr.mac_addr[0], dev_addr); 2719 2720 mutex_lock(&adapter->fw_lock); 2721 adapter->fw_done_rc = 0; 2722 reinit_completion(&adapter->fw_done); 2723 2724 rc = ibmvnic_send_crq(adapter, &crq); 2725 if (rc) { 2726 rc = -EIO; 2727 mutex_unlock(&adapter->fw_lock); 2728 goto err; 2729 } 2730 2731 rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000); 2732 /* netdev->dev_addr is changed in handle_change_mac_rsp function */ 2733 if (rc || adapter->fw_done_rc) { 2734 rc = -EIO; 2735 mutex_unlock(&adapter->fw_lock); 2736 goto err; 2737 } 2738 mutex_unlock(&adapter->fw_lock); 2739 return 0; 2740 err: 2741 ether_addr_copy(adapter->mac_addr, netdev->dev_addr); 2742 return rc; 2743 } 2744 2745 static int ibmvnic_set_mac(struct net_device *netdev, void *p) 2746 { 2747 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 2748 struct sockaddr *addr = p; 2749 int rc; 2750 2751 rc = 0; 2752 if (!is_valid_ether_addr(addr->sa_data)) 2753 return -EADDRNOTAVAIL; 2754 2755 ether_addr_copy(adapter->mac_addr, addr->sa_data); 2756 if (adapter->state != VNIC_PROBED) 2757 rc = __ibmvnic_set_mac(netdev, addr->sa_data); 2758 2759 return rc; 2760 } 2761 2762 static const char *reset_reason_to_string(enum ibmvnic_reset_reason reason) 2763 { 2764 switch (reason) { 2765 case VNIC_RESET_FAILOVER: 2766 return "FAILOVER"; 2767 case VNIC_RESET_MOBILITY: 2768 return "MOBILITY"; 2769 case VNIC_RESET_FATAL: 2770 return "FATAL"; 2771 case VNIC_RESET_NON_FATAL: 2772 return "NON_FATAL"; 2773 case VNIC_RESET_TIMEOUT: 2774 return "TIMEOUT"; 2775 case VNIC_RESET_CHANGE_PARAM: 2776 return "CHANGE_PARAM"; 2777 case VNIC_RESET_PASSIVE_INIT: 2778 return "PASSIVE_INIT"; 2779 } 2780 return "UNKNOWN"; 2781 } 2782 2783 /* 2784 * Initialize the init_done completion and return code values. We 2785 * can get a transport event just after registering the CRQ and the 2786 * tasklet will use this to communicate the transport event. To ensure 2787 * we don't miss the notification/error, initialize these _before_ 2788 * regisering the CRQ. 2789 */ 2790 static inline void reinit_init_done(struct ibmvnic_adapter *adapter) 2791 { 2792 reinit_completion(&adapter->init_done); 2793 adapter->init_done_rc = 0; 2794 } 2795 2796 /* 2797 * do_reset returns zero if we are able to keep processing reset events, or 2798 * non-zero if we hit a fatal error and must halt. 2799 */ 2800 static int do_reset(struct ibmvnic_adapter *adapter, 2801 struct ibmvnic_rwi *rwi, u32 reset_state) 2802 { 2803 struct net_device *netdev = adapter->netdev; 2804 u64 old_num_rx_queues, old_num_tx_queues; 2805 u64 old_num_rx_slots, old_num_tx_slots; 2806 int rc; 2807 2808 netdev_dbg(adapter->netdev, 2809 "[S:%s FOP:%d] Reset reason: %s, reset_state: %s\n", 2810 adapter_state_to_string(adapter->state), 2811 adapter->failover_pending, 2812 reset_reason_to_string(rwi->reset_reason), 2813 adapter_state_to_string(reset_state)); 2814 2815 adapter->reset_reason = rwi->reset_reason; 2816 /* requestor of VNIC_RESET_CHANGE_PARAM already has the rtnl lock */ 2817 if (!(adapter->reset_reason == VNIC_RESET_CHANGE_PARAM)) 2818 rtnl_lock(); 2819 2820 /* Now that we have the rtnl lock, clear any pending failover. 2821 * This will ensure ibmvnic_open() has either completed or will 2822 * block until failover is complete. 2823 */ 2824 if (rwi->reset_reason == VNIC_RESET_FAILOVER) 2825 adapter->failover_pending = false; 2826 2827 /* read the state and check (again) after getting rtnl */ 2828 reset_state = adapter->state; 2829 2830 if (reset_state == VNIC_REMOVING || reset_state == VNIC_REMOVED) { 2831 rc = -EBUSY; 2832 goto out; 2833 } 2834 2835 netif_carrier_off(netdev); 2836 2837 old_num_rx_queues = adapter->req_rx_queues; 2838 old_num_tx_queues = adapter->req_tx_queues; 2839 old_num_rx_slots = adapter->req_rx_add_entries_per_subcrq; 2840 old_num_tx_slots = adapter->req_tx_entries_per_subcrq; 2841 2842 ibmvnic_cleanup(netdev); 2843 2844 if (reset_state == VNIC_OPEN && 2845 adapter->reset_reason != VNIC_RESET_MOBILITY && 2846 adapter->reset_reason != VNIC_RESET_FAILOVER) { 2847 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) { 2848 rc = __ibmvnic_close(netdev); 2849 if (rc) 2850 goto out; 2851 } else { 2852 adapter->state = VNIC_CLOSING; 2853 2854 /* Release the RTNL lock before link state change and 2855 * re-acquire after the link state change to allow 2856 * linkwatch_event to grab the RTNL lock and run during 2857 * a reset. 2858 */ 2859 rtnl_unlock(); 2860 rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN); 2861 rtnl_lock(); 2862 if (rc) 2863 goto out; 2864 2865 if (adapter->state == VNIC_OPEN) { 2866 /* When we dropped rtnl, ibmvnic_open() got 2867 * it and noticed that we are resetting and 2868 * set the adapter state to OPEN. Update our 2869 * new "target" state, and resume the reset 2870 * from VNIC_CLOSING state. 2871 */ 2872 netdev_dbg(netdev, 2873 "Open changed state from %s, updating.\n", 2874 adapter_state_to_string(reset_state)); 2875 reset_state = VNIC_OPEN; 2876 adapter->state = VNIC_CLOSING; 2877 } 2878 2879 if (adapter->state != VNIC_CLOSING) { 2880 /* If someone else changed the adapter state 2881 * when we dropped the rtnl, fail the reset 2882 */ 2883 rc = -EAGAIN; 2884 goto out; 2885 } 2886 adapter->state = VNIC_CLOSED; 2887 } 2888 } 2889 2890 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) { 2891 release_resources(adapter); 2892 release_sub_crqs(adapter, 1); 2893 release_crq_queue(adapter); 2894 } 2895 2896 if (adapter->reset_reason != VNIC_RESET_NON_FATAL) { 2897 /* remove the closed state so when we call open it appears 2898 * we are coming from the probed state. 2899 */ 2900 adapter->state = VNIC_PROBED; 2901 2902 reinit_init_done(adapter); 2903 2904 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) { 2905 rc = init_crq_queue(adapter); 2906 } else if (adapter->reset_reason == VNIC_RESET_MOBILITY) { 2907 rc = ibmvnic_reenable_crq_queue(adapter); 2908 release_sub_crqs(adapter, 1); 2909 } else { 2910 rc = ibmvnic_reset_crq(adapter); 2911 if (rc == H_CLOSED || rc == H_SUCCESS) { 2912 rc = vio_enable_interrupts(adapter->vdev); 2913 if (rc) 2914 netdev_err(adapter->netdev, 2915 "Reset failed to enable interrupts. rc=%d\n", 2916 rc); 2917 } 2918 } 2919 2920 if (rc) { 2921 netdev_err(adapter->netdev, 2922 "Reset couldn't initialize crq. rc=%d\n", rc); 2923 goto out; 2924 } 2925 2926 rc = ibmvnic_reset_init(adapter, true); 2927 if (rc) 2928 goto out; 2929 2930 /* If the adapter was in PROBE or DOWN state prior to the reset, 2931 * exit here. 2932 */ 2933 if (reset_state == VNIC_PROBED || reset_state == VNIC_DOWN) { 2934 rc = 0; 2935 goto out; 2936 } 2937 2938 rc = ibmvnic_login(netdev); 2939 if (rc) 2940 goto out; 2941 2942 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) { 2943 rc = init_resources(adapter); 2944 if (rc) 2945 goto out; 2946 } else if (adapter->req_rx_queues != old_num_rx_queues || 2947 adapter->req_tx_queues != old_num_tx_queues || 2948 adapter->req_rx_add_entries_per_subcrq != 2949 old_num_rx_slots || 2950 adapter->req_tx_entries_per_subcrq != 2951 old_num_tx_slots || 2952 !adapter->rx_pool || 2953 !adapter->tso_pool || 2954 !adapter->tx_pool) { 2955 release_napi(adapter); 2956 release_vpd_data(adapter); 2957 2958 rc = init_resources(adapter); 2959 if (rc) 2960 goto out; 2961 2962 } else { 2963 rc = init_tx_pools(netdev); 2964 if (rc) { 2965 netdev_dbg(netdev, 2966 "init tx pools failed (%d)\n", 2967 rc); 2968 goto out; 2969 } 2970 2971 rc = init_rx_pools(netdev); 2972 if (rc) { 2973 netdev_dbg(netdev, 2974 "init rx pools failed (%d)\n", 2975 rc); 2976 goto out; 2977 } 2978 } 2979 ibmvnic_disable_irqs(adapter); 2980 } 2981 adapter->state = VNIC_CLOSED; 2982 2983 if (reset_state == VNIC_CLOSED) { 2984 rc = 0; 2985 goto out; 2986 } 2987 2988 rc = __ibmvnic_open(netdev); 2989 if (rc) { 2990 rc = IBMVNIC_OPEN_FAILED; 2991 goto out; 2992 } 2993 2994 /* refresh device's multicast list */ 2995 ibmvnic_set_multi(netdev); 2996 2997 if (adapter->reset_reason == VNIC_RESET_FAILOVER || 2998 adapter->reset_reason == VNIC_RESET_MOBILITY) 2999 __netdev_notify_peers(netdev); 3000 3001 rc = 0; 3002 3003 out: 3004 /* restore the adapter state if reset failed */ 3005 if (rc) 3006 adapter->state = reset_state; 3007 /* requestor of VNIC_RESET_CHANGE_PARAM should still hold the rtnl lock */ 3008 if (!(adapter->reset_reason == VNIC_RESET_CHANGE_PARAM)) 3009 rtnl_unlock(); 3010 3011 netdev_dbg(adapter->netdev, "[S:%s FOP:%d] Reset done, rc %d\n", 3012 adapter_state_to_string(adapter->state), 3013 adapter->failover_pending, rc); 3014 return rc; 3015 } 3016 3017 static int do_hard_reset(struct ibmvnic_adapter *adapter, 3018 struct ibmvnic_rwi *rwi, u32 reset_state) 3019 { 3020 struct net_device *netdev = adapter->netdev; 3021 int rc; 3022 3023 netdev_dbg(adapter->netdev, "Hard resetting driver (%s)\n", 3024 reset_reason_to_string(rwi->reset_reason)); 3025 3026 /* read the state and check (again) after getting rtnl */ 3027 reset_state = adapter->state; 3028 3029 if (reset_state == VNIC_REMOVING || reset_state == VNIC_REMOVED) { 3030 rc = -EBUSY; 3031 goto out; 3032 } 3033 3034 netif_carrier_off(netdev); 3035 adapter->reset_reason = rwi->reset_reason; 3036 3037 ibmvnic_cleanup(netdev); 3038 release_resources(adapter); 3039 release_sub_crqs(adapter, 0); 3040 release_crq_queue(adapter); 3041 3042 /* remove the closed state so when we call open it appears 3043 * we are coming from the probed state. 3044 */ 3045 adapter->state = VNIC_PROBED; 3046 3047 reinit_init_done(adapter); 3048 3049 rc = init_crq_queue(adapter); 3050 if (rc) { 3051 netdev_err(adapter->netdev, 3052 "Couldn't initialize crq. rc=%d\n", rc); 3053 goto out; 3054 } 3055 3056 rc = ibmvnic_reset_init(adapter, false); 3057 if (rc) 3058 goto out; 3059 3060 /* If the adapter was in PROBE or DOWN state prior to the reset, 3061 * exit here. 3062 */ 3063 if (reset_state == VNIC_PROBED || reset_state == VNIC_DOWN) 3064 goto out; 3065 3066 rc = ibmvnic_login(netdev); 3067 if (rc) 3068 goto out; 3069 3070 rc = init_resources(adapter); 3071 if (rc) 3072 goto out; 3073 3074 ibmvnic_disable_irqs(adapter); 3075 adapter->state = VNIC_CLOSED; 3076 3077 if (reset_state == VNIC_CLOSED) 3078 goto out; 3079 3080 rc = __ibmvnic_open(netdev); 3081 if (rc) { 3082 rc = IBMVNIC_OPEN_FAILED; 3083 goto out; 3084 } 3085 3086 __netdev_notify_peers(netdev); 3087 out: 3088 /* restore adapter state if reset failed */ 3089 if (rc) 3090 adapter->state = reset_state; 3091 netdev_dbg(adapter->netdev, "[S:%s FOP:%d] Hard reset done, rc %d\n", 3092 adapter_state_to_string(adapter->state), 3093 adapter->failover_pending, rc); 3094 return rc; 3095 } 3096 3097 static struct ibmvnic_rwi *get_next_rwi(struct ibmvnic_adapter *adapter) 3098 { 3099 struct ibmvnic_rwi *rwi; 3100 unsigned long flags; 3101 3102 spin_lock_irqsave(&adapter->rwi_lock, flags); 3103 3104 if (!list_empty(&adapter->rwi_list)) { 3105 rwi = list_first_entry(&adapter->rwi_list, struct ibmvnic_rwi, 3106 list); 3107 list_del(&rwi->list); 3108 } else { 3109 rwi = NULL; 3110 } 3111 3112 spin_unlock_irqrestore(&adapter->rwi_lock, flags); 3113 return rwi; 3114 } 3115 3116 /** 3117 * do_passive_init - complete probing when partner device is detected. 3118 * @adapter: ibmvnic_adapter struct 3119 * 3120 * If the ibmvnic device does not have a partner device to communicate with at boot 3121 * and that partner device comes online at a later time, this function is called 3122 * to complete the initialization process of ibmvnic device. 3123 * Caller is expected to hold rtnl_lock(). 3124 * 3125 * Returns non-zero if sub-CRQs are not initialized properly leaving the device 3126 * in the down state. 3127 * Returns 0 upon success and the device is in PROBED state. 3128 */ 3129 3130 static int do_passive_init(struct ibmvnic_adapter *adapter) 3131 { 3132 unsigned long timeout = msecs_to_jiffies(30000); 3133 struct net_device *netdev = adapter->netdev; 3134 struct device *dev = &adapter->vdev->dev; 3135 int rc; 3136 3137 netdev_dbg(netdev, "Partner device found, probing.\n"); 3138 3139 adapter->state = VNIC_PROBING; 3140 reinit_completion(&adapter->init_done); 3141 adapter->init_done_rc = 0; 3142 adapter->crq.active = true; 3143 3144 rc = send_crq_init_complete(adapter); 3145 if (rc) 3146 goto out; 3147 3148 rc = send_version_xchg(adapter); 3149 if (rc) 3150 netdev_dbg(adapter->netdev, "send_version_xchg failed, rc=%d\n", rc); 3151 3152 if (!wait_for_completion_timeout(&adapter->init_done, timeout)) { 3153 dev_err(dev, "Initialization sequence timed out\n"); 3154 rc = -ETIMEDOUT; 3155 goto out; 3156 } 3157 3158 rc = init_sub_crqs(adapter); 3159 if (rc) { 3160 dev_err(dev, "Initialization of sub crqs failed, rc=%d\n", rc); 3161 goto out; 3162 } 3163 3164 rc = init_sub_crq_irqs(adapter); 3165 if (rc) { 3166 dev_err(dev, "Failed to initialize sub crq irqs\n, rc=%d", rc); 3167 goto init_failed; 3168 } 3169 3170 netdev->mtu = adapter->req_mtu - ETH_HLEN; 3171 netdev->min_mtu = adapter->min_mtu - ETH_HLEN; 3172 netdev->max_mtu = adapter->max_mtu - ETH_HLEN; 3173 3174 adapter->state = VNIC_PROBED; 3175 netdev_dbg(netdev, "Probed successfully. Waiting for signal from partner device.\n"); 3176 3177 return 0; 3178 3179 init_failed: 3180 release_sub_crqs(adapter, 1); 3181 out: 3182 adapter->state = VNIC_DOWN; 3183 return rc; 3184 } 3185 3186 static void __ibmvnic_reset(struct work_struct *work) 3187 { 3188 struct ibmvnic_adapter *adapter; 3189 unsigned int timeout = 5000; 3190 struct ibmvnic_rwi *tmprwi; 3191 bool saved_state = false; 3192 struct ibmvnic_rwi *rwi; 3193 unsigned long flags; 3194 struct device *dev; 3195 bool need_reset; 3196 int num_fails = 0; 3197 u32 reset_state; 3198 int rc = 0; 3199 3200 adapter = container_of(work, struct ibmvnic_adapter, ibmvnic_reset); 3201 dev = &adapter->vdev->dev; 3202 3203 /* Wait for ibmvnic_probe() to complete. If probe is taking too long 3204 * or if another reset is in progress, defer work for now. If probe 3205 * eventually fails it will flush and terminate our work. 3206 * 3207 * Three possibilities here: 3208 * 1. Adpater being removed - just return 3209 * 2. Timed out on probe or another reset in progress - delay the work 3210 * 3. Completed probe - perform any resets in queue 3211 */ 3212 if (adapter->state == VNIC_PROBING && 3213 !wait_for_completion_timeout(&adapter->probe_done, timeout)) { 3214 dev_err(dev, "Reset thread timed out on probe"); 3215 queue_delayed_work(system_long_wq, 3216 &adapter->ibmvnic_delayed_reset, 3217 IBMVNIC_RESET_DELAY); 3218 return; 3219 } 3220 3221 /* adapter is done with probe (i.e state is never VNIC_PROBING now) */ 3222 if (adapter->state == VNIC_REMOVING) 3223 return; 3224 3225 /* ->rwi_list is stable now (no one else is removing entries) */ 3226 3227 /* ibmvnic_probe() may have purged the reset queue after we were 3228 * scheduled to process a reset so there maybe no resets to process. 3229 * Before setting the ->resetting bit though, we have to make sure 3230 * that there is infact a reset to process. Otherwise we may race 3231 * with ibmvnic_open() and end up leaving the vnic down: 3232 * 3233 * __ibmvnic_reset() ibmvnic_open() 3234 * ----------------- -------------- 3235 * 3236 * set ->resetting bit 3237 * find ->resetting bit is set 3238 * set ->state to IBMVNIC_OPEN (i.e 3239 * assume reset will open device) 3240 * return 3241 * find reset queue empty 3242 * return 3243 * 3244 * Neither performed vnic login/open and vnic stays down 3245 * 3246 * If we hold the lock and conditionally set the bit, either we 3247 * or ibmvnic_open() will complete the open. 3248 */ 3249 need_reset = false; 3250 spin_lock(&adapter->rwi_lock); 3251 if (!list_empty(&adapter->rwi_list)) { 3252 if (test_and_set_bit_lock(0, &adapter->resetting)) { 3253 queue_delayed_work(system_long_wq, 3254 &adapter->ibmvnic_delayed_reset, 3255 IBMVNIC_RESET_DELAY); 3256 } else { 3257 need_reset = true; 3258 } 3259 } 3260 spin_unlock(&adapter->rwi_lock); 3261 3262 if (!need_reset) 3263 return; 3264 3265 rwi = get_next_rwi(adapter); 3266 while (rwi) { 3267 spin_lock_irqsave(&adapter->state_lock, flags); 3268 3269 if (adapter->state == VNIC_REMOVING || 3270 adapter->state == VNIC_REMOVED) { 3271 spin_unlock_irqrestore(&adapter->state_lock, flags); 3272 kfree(rwi); 3273 rc = EBUSY; 3274 break; 3275 } 3276 3277 if (!saved_state) { 3278 reset_state = adapter->state; 3279 saved_state = true; 3280 } 3281 spin_unlock_irqrestore(&adapter->state_lock, flags); 3282 3283 if (rwi->reset_reason == VNIC_RESET_PASSIVE_INIT) { 3284 rtnl_lock(); 3285 rc = do_passive_init(adapter); 3286 rtnl_unlock(); 3287 if (!rc) 3288 netif_carrier_on(adapter->netdev); 3289 } else if (adapter->force_reset_recovery) { 3290 /* Since we are doing a hard reset now, clear the 3291 * failover_pending flag so we don't ignore any 3292 * future MOBILITY or other resets. 3293 */ 3294 adapter->failover_pending = false; 3295 3296 /* Transport event occurred during previous reset */ 3297 if (adapter->wait_for_reset) { 3298 /* Previous was CHANGE_PARAM; caller locked */ 3299 adapter->force_reset_recovery = false; 3300 rc = do_hard_reset(adapter, rwi, reset_state); 3301 } else { 3302 rtnl_lock(); 3303 adapter->force_reset_recovery = false; 3304 rc = do_hard_reset(adapter, rwi, reset_state); 3305 rtnl_unlock(); 3306 } 3307 if (rc) 3308 num_fails++; 3309 else 3310 num_fails = 0; 3311 3312 /* If auto-priority-failover is enabled we can get 3313 * back to back failovers during resets, resulting 3314 * in at least two failed resets (from high-priority 3315 * backing device to low-priority one and then back) 3316 * If resets continue to fail beyond that, give the 3317 * adapter some time to settle down before retrying. 3318 */ 3319 if (num_fails >= 3) { 3320 netdev_dbg(adapter->netdev, 3321 "[S:%s] Hard reset failed %d times, waiting 60 secs\n", 3322 adapter_state_to_string(adapter->state), 3323 num_fails); 3324 set_current_state(TASK_UNINTERRUPTIBLE); 3325 schedule_timeout(60 * HZ); 3326 } 3327 } else { 3328 rc = do_reset(adapter, rwi, reset_state); 3329 } 3330 tmprwi = rwi; 3331 adapter->last_reset_time = jiffies; 3332 3333 if (rc) 3334 netdev_dbg(adapter->netdev, "Reset failed, rc=%d\n", rc); 3335 3336 rwi = get_next_rwi(adapter); 3337 3338 /* 3339 * If there are no resets queued and the previous reset failed, 3340 * the adapter would be in an undefined state. So retry the 3341 * previous reset as a hard reset. 3342 * 3343 * Else, free the previous rwi and, if there is another reset 3344 * queued, process the new reset even if previous reset failed 3345 * (the previous reset could have failed because of a fail 3346 * over for instance, so process the fail over). 3347 */ 3348 if (!rwi && rc) 3349 rwi = tmprwi; 3350 else 3351 kfree(tmprwi); 3352 3353 if (rwi && (rwi->reset_reason == VNIC_RESET_FAILOVER || 3354 rwi->reset_reason == VNIC_RESET_MOBILITY || rc)) 3355 adapter->force_reset_recovery = true; 3356 } 3357 3358 if (adapter->wait_for_reset) { 3359 adapter->reset_done_rc = rc; 3360 complete(&adapter->reset_done); 3361 } 3362 3363 clear_bit_unlock(0, &adapter->resetting); 3364 3365 netdev_dbg(adapter->netdev, 3366 "[S:%s FRR:%d WFR:%d] Done processing resets\n", 3367 adapter_state_to_string(adapter->state), 3368 adapter->force_reset_recovery, 3369 adapter->wait_for_reset); 3370 } 3371 3372 static void __ibmvnic_delayed_reset(struct work_struct *work) 3373 { 3374 struct ibmvnic_adapter *adapter; 3375 3376 adapter = container_of(work, struct ibmvnic_adapter, 3377 ibmvnic_delayed_reset.work); 3378 __ibmvnic_reset(&adapter->ibmvnic_reset); 3379 } 3380 3381 static void flush_reset_queue(struct ibmvnic_adapter *adapter) 3382 { 3383 struct list_head *entry, *tmp_entry; 3384 3385 if (!list_empty(&adapter->rwi_list)) { 3386 list_for_each_safe(entry, tmp_entry, &adapter->rwi_list) { 3387 list_del(entry); 3388 kfree(list_entry(entry, struct ibmvnic_rwi, list)); 3389 } 3390 } 3391 } 3392 3393 static int ibmvnic_reset(struct ibmvnic_adapter *adapter, 3394 enum ibmvnic_reset_reason reason) 3395 { 3396 struct net_device *netdev = adapter->netdev; 3397 struct ibmvnic_rwi *rwi, *tmp; 3398 unsigned long flags; 3399 int ret; 3400 3401 spin_lock_irqsave(&adapter->rwi_lock, flags); 3402 3403 /* If failover is pending don't schedule any other reset. 3404 * Instead let the failover complete. If there is already a 3405 * a failover reset scheduled, we will detect and drop the 3406 * duplicate reset when walking the ->rwi_list below. 3407 */ 3408 if (adapter->state == VNIC_REMOVING || 3409 adapter->state == VNIC_REMOVED || 3410 (adapter->failover_pending && reason != VNIC_RESET_FAILOVER)) { 3411 ret = EBUSY; 3412 netdev_dbg(netdev, "Adapter removing or pending failover, skipping reset\n"); 3413 goto err; 3414 } 3415 3416 list_for_each_entry(tmp, &adapter->rwi_list, list) { 3417 if (tmp->reset_reason == reason) { 3418 netdev_dbg(netdev, "Skipping matching reset, reason=%s\n", 3419 reset_reason_to_string(reason)); 3420 ret = EBUSY; 3421 goto err; 3422 } 3423 } 3424 3425 rwi = kzalloc(sizeof(*rwi), GFP_ATOMIC); 3426 if (!rwi) { 3427 ret = ENOMEM; 3428 goto err; 3429 } 3430 /* if we just received a transport event, 3431 * flush reset queue and process this reset 3432 */ 3433 if (adapter->force_reset_recovery) 3434 flush_reset_queue(adapter); 3435 3436 rwi->reset_reason = reason; 3437 list_add_tail(&rwi->list, &adapter->rwi_list); 3438 netdev_dbg(adapter->netdev, "Scheduling reset (reason %s)\n", 3439 reset_reason_to_string(reason)); 3440 queue_work(system_long_wq, &adapter->ibmvnic_reset); 3441 3442 ret = 0; 3443 err: 3444 /* ibmvnic_close() below can block, so drop the lock first */ 3445 spin_unlock_irqrestore(&adapter->rwi_lock, flags); 3446 3447 if (ret == ENOMEM) 3448 ibmvnic_close(netdev); 3449 3450 return -ret; 3451 } 3452 3453 static void ibmvnic_tx_timeout(struct net_device *dev, unsigned int txqueue) 3454 { 3455 struct ibmvnic_adapter *adapter = netdev_priv(dev); 3456 3457 if (test_bit(0, &adapter->resetting)) { 3458 netdev_err(adapter->netdev, 3459 "Adapter is resetting, skip timeout reset\n"); 3460 return; 3461 } 3462 /* No queuing up reset until at least 5 seconds (default watchdog val) 3463 * after last reset 3464 */ 3465 if (time_before(jiffies, (adapter->last_reset_time + dev->watchdog_timeo))) { 3466 netdev_dbg(dev, "Not yet time to tx timeout.\n"); 3467 return; 3468 } 3469 ibmvnic_reset(adapter, VNIC_RESET_TIMEOUT); 3470 } 3471 3472 static void remove_buff_from_pool(struct ibmvnic_adapter *adapter, 3473 struct ibmvnic_rx_buff *rx_buff) 3474 { 3475 struct ibmvnic_rx_pool *pool = &adapter->rx_pool[rx_buff->pool_index]; 3476 3477 rx_buff->skb = NULL; 3478 3479 pool->free_map[pool->next_alloc] = (int)(rx_buff - pool->rx_buff); 3480 pool->next_alloc = (pool->next_alloc + 1) % pool->size; 3481 3482 atomic_dec(&pool->available); 3483 } 3484 3485 static int ibmvnic_poll(struct napi_struct *napi, int budget) 3486 { 3487 struct ibmvnic_sub_crq_queue *rx_scrq; 3488 struct ibmvnic_adapter *adapter; 3489 struct net_device *netdev; 3490 int frames_processed; 3491 int scrq_num; 3492 3493 netdev = napi->dev; 3494 adapter = netdev_priv(netdev); 3495 scrq_num = (int)(napi - adapter->napi); 3496 frames_processed = 0; 3497 rx_scrq = adapter->rx_scrq[scrq_num]; 3498 3499 restart_poll: 3500 while (frames_processed < budget) { 3501 struct sk_buff *skb; 3502 struct ibmvnic_rx_buff *rx_buff; 3503 union sub_crq *next; 3504 u32 length; 3505 u16 offset; 3506 u8 flags = 0; 3507 3508 if (unlikely(test_bit(0, &adapter->resetting) && 3509 adapter->reset_reason != VNIC_RESET_NON_FATAL)) { 3510 enable_scrq_irq(adapter, rx_scrq); 3511 napi_complete_done(napi, frames_processed); 3512 return frames_processed; 3513 } 3514 3515 if (!pending_scrq(adapter, rx_scrq)) 3516 break; 3517 next = ibmvnic_next_scrq(adapter, rx_scrq); 3518 rx_buff = (struct ibmvnic_rx_buff *) 3519 be64_to_cpu(next->rx_comp.correlator); 3520 /* do error checking */ 3521 if (next->rx_comp.rc) { 3522 netdev_dbg(netdev, "rx buffer returned with rc %x\n", 3523 be16_to_cpu(next->rx_comp.rc)); 3524 /* free the entry */ 3525 next->rx_comp.first = 0; 3526 dev_kfree_skb_any(rx_buff->skb); 3527 remove_buff_from_pool(adapter, rx_buff); 3528 continue; 3529 } else if (!rx_buff->skb) { 3530 /* free the entry */ 3531 next->rx_comp.first = 0; 3532 remove_buff_from_pool(adapter, rx_buff); 3533 continue; 3534 } 3535 3536 length = be32_to_cpu(next->rx_comp.len); 3537 offset = be16_to_cpu(next->rx_comp.off_frame_data); 3538 flags = next->rx_comp.flags; 3539 skb = rx_buff->skb; 3540 /* load long_term_buff before copying to skb */ 3541 dma_rmb(); 3542 skb_copy_to_linear_data(skb, rx_buff->data + offset, 3543 length); 3544 3545 /* VLAN Header has been stripped by the system firmware and 3546 * needs to be inserted by the driver 3547 */ 3548 if (adapter->rx_vlan_header_insertion && 3549 (flags & IBMVNIC_VLAN_STRIPPED)) 3550 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), 3551 ntohs(next->rx_comp.vlan_tci)); 3552 3553 /* free the entry */ 3554 next->rx_comp.first = 0; 3555 remove_buff_from_pool(adapter, rx_buff); 3556 3557 skb_put(skb, length); 3558 skb->protocol = eth_type_trans(skb, netdev); 3559 skb_record_rx_queue(skb, scrq_num); 3560 3561 if (flags & IBMVNIC_IP_CHKSUM_GOOD && 3562 flags & IBMVNIC_TCP_UDP_CHKSUM_GOOD) { 3563 skb->ip_summed = CHECKSUM_UNNECESSARY; 3564 } 3565 3566 length = skb->len; 3567 napi_gro_receive(napi, skb); /* send it up */ 3568 netdev->stats.rx_packets++; 3569 netdev->stats.rx_bytes += length; 3570 adapter->rx_stats_buffers[scrq_num].packets++; 3571 adapter->rx_stats_buffers[scrq_num].bytes += length; 3572 frames_processed++; 3573 } 3574 3575 if (adapter->state != VNIC_CLOSING && 3576 (atomic_read(&adapter->rx_pool[scrq_num].available) < 3577 adapter->req_rx_add_entries_per_subcrq / 2)) 3578 replenish_rx_pool(adapter, &adapter->rx_pool[scrq_num]); 3579 if (frames_processed < budget) { 3580 if (napi_complete_done(napi, frames_processed)) { 3581 enable_scrq_irq(adapter, rx_scrq); 3582 if (pending_scrq(adapter, rx_scrq)) { 3583 if (napi_schedule(napi)) { 3584 disable_scrq_irq(adapter, rx_scrq); 3585 goto restart_poll; 3586 } 3587 } 3588 } 3589 } 3590 return frames_processed; 3591 } 3592 3593 static int wait_for_reset(struct ibmvnic_adapter *adapter) 3594 { 3595 int rc, ret; 3596 3597 adapter->fallback.mtu = adapter->req_mtu; 3598 adapter->fallback.rx_queues = adapter->req_rx_queues; 3599 adapter->fallback.tx_queues = adapter->req_tx_queues; 3600 adapter->fallback.rx_entries = adapter->req_rx_add_entries_per_subcrq; 3601 adapter->fallback.tx_entries = adapter->req_tx_entries_per_subcrq; 3602 3603 reinit_completion(&adapter->reset_done); 3604 adapter->wait_for_reset = true; 3605 rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM); 3606 3607 if (rc) { 3608 ret = rc; 3609 goto out; 3610 } 3611 rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done, 60000); 3612 if (rc) { 3613 ret = -ENODEV; 3614 goto out; 3615 } 3616 3617 ret = 0; 3618 if (adapter->reset_done_rc) { 3619 ret = -EIO; 3620 adapter->desired.mtu = adapter->fallback.mtu; 3621 adapter->desired.rx_queues = adapter->fallback.rx_queues; 3622 adapter->desired.tx_queues = adapter->fallback.tx_queues; 3623 adapter->desired.rx_entries = adapter->fallback.rx_entries; 3624 adapter->desired.tx_entries = adapter->fallback.tx_entries; 3625 3626 reinit_completion(&adapter->reset_done); 3627 adapter->wait_for_reset = true; 3628 rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM); 3629 if (rc) { 3630 ret = rc; 3631 goto out; 3632 } 3633 rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done, 3634 60000); 3635 if (rc) { 3636 ret = -ENODEV; 3637 goto out; 3638 } 3639 } 3640 out: 3641 adapter->wait_for_reset = false; 3642 3643 return ret; 3644 } 3645 3646 static int ibmvnic_change_mtu(struct net_device *netdev, int new_mtu) 3647 { 3648 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3649 3650 adapter->desired.mtu = new_mtu + ETH_HLEN; 3651 3652 return wait_for_reset(adapter); 3653 } 3654 3655 static netdev_features_t ibmvnic_features_check(struct sk_buff *skb, 3656 struct net_device *dev, 3657 netdev_features_t features) 3658 { 3659 /* Some backing hardware adapters can not 3660 * handle packets with a MSS less than 224 3661 * or with only one segment. 3662 */ 3663 if (skb_is_gso(skb)) { 3664 if (skb_shinfo(skb)->gso_size < 224 || 3665 skb_shinfo(skb)->gso_segs == 1) 3666 features &= ~NETIF_F_GSO_MASK; 3667 } 3668 3669 return features; 3670 } 3671 3672 static const struct net_device_ops ibmvnic_netdev_ops = { 3673 .ndo_open = ibmvnic_open, 3674 .ndo_stop = ibmvnic_close, 3675 .ndo_start_xmit = ibmvnic_xmit, 3676 .ndo_set_rx_mode = ibmvnic_set_multi, 3677 .ndo_set_mac_address = ibmvnic_set_mac, 3678 .ndo_validate_addr = eth_validate_addr, 3679 .ndo_tx_timeout = ibmvnic_tx_timeout, 3680 .ndo_change_mtu = ibmvnic_change_mtu, 3681 .ndo_features_check = ibmvnic_features_check, 3682 }; 3683 3684 /* ethtool functions */ 3685 3686 static int ibmvnic_get_link_ksettings(struct net_device *netdev, 3687 struct ethtool_link_ksettings *cmd) 3688 { 3689 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3690 int rc; 3691 3692 rc = send_query_phys_parms(adapter); 3693 if (rc) { 3694 adapter->speed = SPEED_UNKNOWN; 3695 adapter->duplex = DUPLEX_UNKNOWN; 3696 } 3697 cmd->base.speed = adapter->speed; 3698 cmd->base.duplex = adapter->duplex; 3699 cmd->base.port = PORT_FIBRE; 3700 cmd->base.phy_address = 0; 3701 cmd->base.autoneg = AUTONEG_ENABLE; 3702 3703 return 0; 3704 } 3705 3706 static void ibmvnic_get_drvinfo(struct net_device *netdev, 3707 struct ethtool_drvinfo *info) 3708 { 3709 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3710 3711 strscpy(info->driver, ibmvnic_driver_name, sizeof(info->driver)); 3712 strscpy(info->version, IBMVNIC_DRIVER_VERSION, sizeof(info->version)); 3713 strscpy(info->fw_version, adapter->fw_version, 3714 sizeof(info->fw_version)); 3715 } 3716 3717 static u32 ibmvnic_get_msglevel(struct net_device *netdev) 3718 { 3719 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3720 3721 return adapter->msg_enable; 3722 } 3723 3724 static void ibmvnic_set_msglevel(struct net_device *netdev, u32 data) 3725 { 3726 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3727 3728 adapter->msg_enable = data; 3729 } 3730 3731 static u32 ibmvnic_get_link(struct net_device *netdev) 3732 { 3733 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3734 3735 /* Don't need to send a query because we request a logical link up at 3736 * init and then we wait for link state indications 3737 */ 3738 return adapter->logical_link_state; 3739 } 3740 3741 static void ibmvnic_get_ringparam(struct net_device *netdev, 3742 struct ethtool_ringparam *ring, 3743 struct kernel_ethtool_ringparam *kernel_ring, 3744 struct netlink_ext_ack *extack) 3745 { 3746 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3747 3748 ring->rx_max_pending = adapter->max_rx_add_entries_per_subcrq; 3749 ring->tx_max_pending = adapter->max_tx_entries_per_subcrq; 3750 ring->rx_mini_max_pending = 0; 3751 ring->rx_jumbo_max_pending = 0; 3752 ring->rx_pending = adapter->req_rx_add_entries_per_subcrq; 3753 ring->tx_pending = adapter->req_tx_entries_per_subcrq; 3754 ring->rx_mini_pending = 0; 3755 ring->rx_jumbo_pending = 0; 3756 } 3757 3758 static int ibmvnic_set_ringparam(struct net_device *netdev, 3759 struct ethtool_ringparam *ring, 3760 struct kernel_ethtool_ringparam *kernel_ring, 3761 struct netlink_ext_ack *extack) 3762 { 3763 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3764 3765 if (ring->rx_pending > adapter->max_rx_add_entries_per_subcrq || 3766 ring->tx_pending > adapter->max_tx_entries_per_subcrq) { 3767 netdev_err(netdev, "Invalid request.\n"); 3768 netdev_err(netdev, "Max tx buffers = %llu\n", 3769 adapter->max_rx_add_entries_per_subcrq); 3770 netdev_err(netdev, "Max rx buffers = %llu\n", 3771 adapter->max_tx_entries_per_subcrq); 3772 return -EINVAL; 3773 } 3774 3775 adapter->desired.rx_entries = ring->rx_pending; 3776 adapter->desired.tx_entries = ring->tx_pending; 3777 3778 return wait_for_reset(adapter); 3779 } 3780 3781 static void ibmvnic_get_channels(struct net_device *netdev, 3782 struct ethtool_channels *channels) 3783 { 3784 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3785 3786 channels->max_rx = adapter->max_rx_queues; 3787 channels->max_tx = adapter->max_tx_queues; 3788 channels->max_other = 0; 3789 channels->max_combined = 0; 3790 channels->rx_count = adapter->req_rx_queues; 3791 channels->tx_count = adapter->req_tx_queues; 3792 channels->other_count = 0; 3793 channels->combined_count = 0; 3794 } 3795 3796 static int ibmvnic_set_channels(struct net_device *netdev, 3797 struct ethtool_channels *channels) 3798 { 3799 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3800 3801 adapter->desired.rx_queues = channels->rx_count; 3802 adapter->desired.tx_queues = channels->tx_count; 3803 3804 return wait_for_reset(adapter); 3805 } 3806 3807 static void ibmvnic_get_strings(struct net_device *dev, u32 stringset, u8 *data) 3808 { 3809 struct ibmvnic_adapter *adapter = netdev_priv(dev); 3810 int i; 3811 3812 if (stringset != ETH_SS_STATS) 3813 return; 3814 3815 for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); i++) 3816 ethtool_puts(&data, ibmvnic_stats[i].name); 3817 3818 for (i = 0; i < adapter->req_tx_queues; i++) { 3819 ethtool_sprintf(&data, "tx%d_batched_packets", i); 3820 ethtool_sprintf(&data, "tx%d_direct_packets", i); 3821 ethtool_sprintf(&data, "tx%d_bytes", i); 3822 ethtool_sprintf(&data, "tx%d_dropped_packets", i); 3823 } 3824 3825 for (i = 0; i < adapter->req_rx_queues; i++) { 3826 ethtool_sprintf(&data, "rx%d_packets", i); 3827 ethtool_sprintf(&data, "rx%d_bytes", i); 3828 ethtool_sprintf(&data, "rx%d_interrupts", i); 3829 } 3830 } 3831 3832 static int ibmvnic_get_sset_count(struct net_device *dev, int sset) 3833 { 3834 struct ibmvnic_adapter *adapter = netdev_priv(dev); 3835 3836 switch (sset) { 3837 case ETH_SS_STATS: 3838 return ARRAY_SIZE(ibmvnic_stats) + 3839 adapter->req_tx_queues * NUM_TX_STATS + 3840 adapter->req_rx_queues * NUM_RX_STATS; 3841 default: 3842 return -EOPNOTSUPP; 3843 } 3844 } 3845 3846 static void ibmvnic_get_ethtool_stats(struct net_device *dev, 3847 struct ethtool_stats *stats, u64 *data) 3848 { 3849 struct ibmvnic_adapter *adapter = netdev_priv(dev); 3850 union ibmvnic_crq crq; 3851 int i, j; 3852 int rc; 3853 3854 memset(&crq, 0, sizeof(crq)); 3855 crq.request_statistics.first = IBMVNIC_CRQ_CMD; 3856 crq.request_statistics.cmd = REQUEST_STATISTICS; 3857 crq.request_statistics.ioba = cpu_to_be32(adapter->stats_token); 3858 crq.request_statistics.len = 3859 cpu_to_be32(sizeof(struct ibmvnic_statistics)); 3860 3861 /* Wait for data to be written */ 3862 reinit_completion(&adapter->stats_done); 3863 rc = ibmvnic_send_crq(adapter, &crq); 3864 if (rc) 3865 return; 3866 rc = ibmvnic_wait_for_completion(adapter, &adapter->stats_done, 10000); 3867 if (rc) 3868 return; 3869 3870 for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); i++) 3871 data[i] = be64_to_cpu(IBMVNIC_GET_STAT 3872 (adapter, ibmvnic_stats[i].offset)); 3873 3874 for (j = 0; j < adapter->req_tx_queues; j++) { 3875 data[i] = adapter->tx_stats_buffers[j].batched_packets; 3876 i++; 3877 data[i] = adapter->tx_stats_buffers[j].direct_packets; 3878 i++; 3879 data[i] = adapter->tx_stats_buffers[j].bytes; 3880 i++; 3881 data[i] = adapter->tx_stats_buffers[j].dropped_packets; 3882 i++; 3883 } 3884 3885 for (j = 0; j < adapter->req_rx_queues; j++) { 3886 data[i] = adapter->rx_stats_buffers[j].packets; 3887 i++; 3888 data[i] = adapter->rx_stats_buffers[j].bytes; 3889 i++; 3890 data[i] = adapter->rx_stats_buffers[j].interrupts; 3891 i++; 3892 } 3893 } 3894 3895 static const struct ethtool_ops ibmvnic_ethtool_ops = { 3896 .get_drvinfo = ibmvnic_get_drvinfo, 3897 .get_msglevel = ibmvnic_get_msglevel, 3898 .set_msglevel = ibmvnic_set_msglevel, 3899 .get_link = ibmvnic_get_link, 3900 .get_ringparam = ibmvnic_get_ringparam, 3901 .set_ringparam = ibmvnic_set_ringparam, 3902 .get_channels = ibmvnic_get_channels, 3903 .set_channels = ibmvnic_set_channels, 3904 .get_strings = ibmvnic_get_strings, 3905 .get_sset_count = ibmvnic_get_sset_count, 3906 .get_ethtool_stats = ibmvnic_get_ethtool_stats, 3907 .get_link_ksettings = ibmvnic_get_link_ksettings, 3908 }; 3909 3910 /* Routines for managing CRQs/sCRQs */ 3911 3912 static int reset_one_sub_crq_queue(struct ibmvnic_adapter *adapter, 3913 struct ibmvnic_sub_crq_queue *scrq) 3914 { 3915 int rc; 3916 3917 if (!scrq) { 3918 netdev_dbg(adapter->netdev, "Invalid scrq reset.\n"); 3919 return -EINVAL; 3920 } 3921 3922 if (scrq->irq) { 3923 free_irq(scrq->irq, scrq); 3924 irq_dispose_mapping(scrq->irq); 3925 scrq->irq = 0; 3926 } 3927 3928 if (scrq->msgs) { 3929 memset(scrq->msgs, 0, 4 * PAGE_SIZE); 3930 atomic_set(&scrq->used, 0); 3931 scrq->cur = 0; 3932 scrq->ind_buf.index = 0; 3933 } else { 3934 netdev_dbg(adapter->netdev, "Invalid scrq reset\n"); 3935 return -EINVAL; 3936 } 3937 3938 rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token, 3939 4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq); 3940 return rc; 3941 } 3942 3943 static int reset_sub_crq_queues(struct ibmvnic_adapter *adapter) 3944 { 3945 int i, rc; 3946 3947 if (!adapter->tx_scrq || !adapter->rx_scrq) 3948 return -EINVAL; 3949 3950 ibmvnic_clean_affinity(adapter); 3951 3952 for (i = 0; i < adapter->req_tx_queues; i++) { 3953 netdev_dbg(adapter->netdev, "Re-setting tx_scrq[%d]\n", i); 3954 rc = reset_one_sub_crq_queue(adapter, adapter->tx_scrq[i]); 3955 if (rc) 3956 return rc; 3957 } 3958 3959 for (i = 0; i < adapter->req_rx_queues; i++) { 3960 netdev_dbg(adapter->netdev, "Re-setting rx_scrq[%d]\n", i); 3961 rc = reset_one_sub_crq_queue(adapter, adapter->rx_scrq[i]); 3962 if (rc) 3963 return rc; 3964 } 3965 3966 return rc; 3967 } 3968 3969 static void release_sub_crq_queue(struct ibmvnic_adapter *adapter, 3970 struct ibmvnic_sub_crq_queue *scrq, 3971 bool do_h_free) 3972 { 3973 struct device *dev = &adapter->vdev->dev; 3974 long rc; 3975 3976 netdev_dbg(adapter->netdev, "Releasing sub-CRQ\n"); 3977 3978 if (do_h_free) { 3979 /* Close the sub-crqs */ 3980 do { 3981 rc = plpar_hcall_norets(H_FREE_SUB_CRQ, 3982 adapter->vdev->unit_address, 3983 scrq->crq_num); 3984 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc)); 3985 3986 if (rc) { 3987 netdev_err(adapter->netdev, 3988 "Failed to release sub-CRQ %16lx, rc = %ld\n", 3989 scrq->crq_num, rc); 3990 } 3991 } 3992 3993 dma_free_coherent(dev, 3994 IBMVNIC_IND_ARR_SZ, 3995 scrq->ind_buf.indir_arr, 3996 scrq->ind_buf.indir_dma); 3997 3998 dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE, 3999 DMA_BIDIRECTIONAL); 4000 free_pages((unsigned long)scrq->msgs, 2); 4001 free_cpumask_var(scrq->affinity_mask); 4002 kfree(scrq); 4003 } 4004 4005 static struct ibmvnic_sub_crq_queue *init_sub_crq_queue(struct ibmvnic_adapter 4006 *adapter) 4007 { 4008 struct device *dev = &adapter->vdev->dev; 4009 struct ibmvnic_sub_crq_queue *scrq; 4010 int rc; 4011 4012 scrq = kzalloc(sizeof(*scrq), GFP_KERNEL); 4013 if (!scrq) 4014 return NULL; 4015 4016 scrq->msgs = 4017 (union sub_crq *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 2); 4018 if (!scrq->msgs) { 4019 dev_warn(dev, "Couldn't allocate crq queue messages page\n"); 4020 goto zero_page_failed; 4021 } 4022 if (!zalloc_cpumask_var(&scrq->affinity_mask, GFP_KERNEL)) 4023 goto cpumask_alloc_failed; 4024 4025 scrq->msg_token = dma_map_single(dev, scrq->msgs, 4 * PAGE_SIZE, 4026 DMA_BIDIRECTIONAL); 4027 if (dma_mapping_error(dev, scrq->msg_token)) { 4028 dev_warn(dev, "Couldn't map crq queue messages page\n"); 4029 goto map_failed; 4030 } 4031 4032 rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token, 4033 4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq); 4034 4035 if (rc == H_RESOURCE) 4036 rc = ibmvnic_reset_crq(adapter); 4037 4038 if (rc == H_CLOSED) { 4039 dev_warn(dev, "Partner adapter not ready, waiting.\n"); 4040 } else if (rc) { 4041 dev_warn(dev, "Error %d registering sub-crq\n", rc); 4042 goto reg_failed; 4043 } 4044 4045 scrq->adapter = adapter; 4046 scrq->size = 4 * PAGE_SIZE / sizeof(*scrq->msgs); 4047 scrq->ind_buf.index = 0; 4048 4049 scrq->ind_buf.indir_arr = 4050 dma_alloc_coherent(dev, 4051 IBMVNIC_IND_ARR_SZ, 4052 &scrq->ind_buf.indir_dma, 4053 GFP_KERNEL); 4054 4055 if (!scrq->ind_buf.indir_arr) 4056 goto indir_failed; 4057 4058 spin_lock_init(&scrq->lock); 4059 4060 netdev_dbg(adapter->netdev, 4061 "sub-crq initialized, num %lx, hw_irq=%lx, irq=%x\n", 4062 scrq->crq_num, scrq->hw_irq, scrq->irq); 4063 4064 return scrq; 4065 4066 indir_failed: 4067 do { 4068 rc = plpar_hcall_norets(H_FREE_SUB_CRQ, 4069 adapter->vdev->unit_address, 4070 scrq->crq_num); 4071 } while (rc == H_BUSY || rc == H_IS_LONG_BUSY(rc)); 4072 reg_failed: 4073 dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE, 4074 DMA_BIDIRECTIONAL); 4075 map_failed: 4076 free_cpumask_var(scrq->affinity_mask); 4077 cpumask_alloc_failed: 4078 free_pages((unsigned long)scrq->msgs, 2); 4079 zero_page_failed: 4080 kfree(scrq); 4081 4082 return NULL; 4083 } 4084 4085 static void release_sub_crqs(struct ibmvnic_adapter *adapter, bool do_h_free) 4086 { 4087 int i; 4088 4089 ibmvnic_clean_affinity(adapter); 4090 if (adapter->tx_scrq) { 4091 for (i = 0; i < adapter->num_active_tx_scrqs; i++) { 4092 if (!adapter->tx_scrq[i]) 4093 continue; 4094 4095 netdev_dbg(adapter->netdev, "Releasing tx_scrq[%d]\n", 4096 i); 4097 ibmvnic_tx_scrq_clean_buffer(adapter, adapter->tx_scrq[i]); 4098 if (adapter->tx_scrq[i]->irq) { 4099 free_irq(adapter->tx_scrq[i]->irq, 4100 adapter->tx_scrq[i]); 4101 irq_dispose_mapping(adapter->tx_scrq[i]->irq); 4102 adapter->tx_scrq[i]->irq = 0; 4103 } 4104 4105 release_sub_crq_queue(adapter, adapter->tx_scrq[i], 4106 do_h_free); 4107 } 4108 4109 kfree(adapter->tx_scrq); 4110 adapter->tx_scrq = NULL; 4111 adapter->num_active_tx_scrqs = 0; 4112 } 4113 4114 /* Clean any remaining outstanding SKBs 4115 * we freed the irq so we won't be hearing 4116 * from them 4117 */ 4118 clean_tx_pools(adapter); 4119 4120 if (adapter->rx_scrq) { 4121 for (i = 0; i < adapter->num_active_rx_scrqs; i++) { 4122 if (!adapter->rx_scrq[i]) 4123 continue; 4124 4125 netdev_dbg(adapter->netdev, "Releasing rx_scrq[%d]\n", 4126 i); 4127 if (adapter->rx_scrq[i]->irq) { 4128 free_irq(adapter->rx_scrq[i]->irq, 4129 adapter->rx_scrq[i]); 4130 irq_dispose_mapping(adapter->rx_scrq[i]->irq); 4131 adapter->rx_scrq[i]->irq = 0; 4132 } 4133 4134 release_sub_crq_queue(adapter, adapter->rx_scrq[i], 4135 do_h_free); 4136 } 4137 4138 kfree(adapter->rx_scrq); 4139 adapter->rx_scrq = NULL; 4140 adapter->num_active_rx_scrqs = 0; 4141 } 4142 } 4143 4144 static int disable_scrq_irq(struct ibmvnic_adapter *adapter, 4145 struct ibmvnic_sub_crq_queue *scrq) 4146 { 4147 struct device *dev = &adapter->vdev->dev; 4148 unsigned long rc; 4149 4150 rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address, 4151 H_DISABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0); 4152 if (rc) 4153 dev_err(dev, "Couldn't disable scrq irq 0x%lx. rc=%ld\n", 4154 scrq->hw_irq, rc); 4155 return rc; 4156 } 4157 4158 /* We can not use the IRQ chip EOI handler because that has the 4159 * unintended effect of changing the interrupt priority. 4160 */ 4161 static void ibmvnic_xics_eoi(struct device *dev, struct ibmvnic_sub_crq_queue *scrq) 4162 { 4163 u64 val = 0xff000000 | scrq->hw_irq; 4164 unsigned long rc; 4165 4166 rc = plpar_hcall_norets(H_EOI, val); 4167 if (rc) 4168 dev_err(dev, "H_EOI FAILED irq 0x%llx. rc=%ld\n", val, rc); 4169 } 4170 4171 /* Due to a firmware bug, the hypervisor can send an interrupt to a 4172 * transmit or receive queue just prior to a partition migration. 4173 * Force an EOI after migration. 4174 */ 4175 static void ibmvnic_clear_pending_interrupt(struct device *dev, 4176 struct ibmvnic_sub_crq_queue *scrq) 4177 { 4178 if (!xive_enabled()) 4179 ibmvnic_xics_eoi(dev, scrq); 4180 } 4181 4182 static int enable_scrq_irq(struct ibmvnic_adapter *adapter, 4183 struct ibmvnic_sub_crq_queue *scrq) 4184 { 4185 struct device *dev = &adapter->vdev->dev; 4186 unsigned long rc; 4187 4188 if (scrq->hw_irq > 0x100000000ULL) { 4189 dev_err(dev, "bad hw_irq = %lx\n", scrq->hw_irq); 4190 return 1; 4191 } 4192 4193 if (test_bit(0, &adapter->resetting) && 4194 adapter->reset_reason == VNIC_RESET_MOBILITY) { 4195 ibmvnic_clear_pending_interrupt(dev, scrq); 4196 } 4197 4198 rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address, 4199 H_ENABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0); 4200 if (rc) 4201 dev_err(dev, "Couldn't enable scrq irq 0x%lx. rc=%ld\n", 4202 scrq->hw_irq, rc); 4203 return rc; 4204 } 4205 4206 static int ibmvnic_complete_tx(struct ibmvnic_adapter *adapter, 4207 struct ibmvnic_sub_crq_queue *scrq) 4208 { 4209 struct device *dev = &adapter->vdev->dev; 4210 int num_packets = 0, total_bytes = 0; 4211 struct ibmvnic_tx_pool *tx_pool; 4212 struct ibmvnic_tx_buff *txbuff; 4213 struct netdev_queue *txq; 4214 union sub_crq *next; 4215 int index, i; 4216 4217 restart_loop: 4218 while (pending_scrq(adapter, scrq)) { 4219 unsigned int pool = scrq->pool_index; 4220 int num_entries = 0; 4221 next = ibmvnic_next_scrq(adapter, scrq); 4222 for (i = 0; i < next->tx_comp.num_comps; i++) { 4223 index = be32_to_cpu(next->tx_comp.correlators[i]); 4224 if (index & IBMVNIC_TSO_POOL_MASK) { 4225 tx_pool = &adapter->tso_pool[pool]; 4226 index &= ~IBMVNIC_TSO_POOL_MASK; 4227 } else { 4228 tx_pool = &adapter->tx_pool[pool]; 4229 } 4230 4231 txbuff = &tx_pool->tx_buff[index]; 4232 num_packets++; 4233 num_entries += txbuff->num_entries; 4234 if (txbuff->skb) { 4235 total_bytes += txbuff->skb->len; 4236 if (next->tx_comp.rcs[i]) { 4237 dev_err(dev, "tx error %x\n", 4238 next->tx_comp.rcs[i]); 4239 dev_kfree_skb_irq(txbuff->skb); 4240 } else { 4241 dev_consume_skb_irq(txbuff->skb); 4242 } 4243 txbuff->skb = NULL; 4244 } else { 4245 netdev_warn(adapter->netdev, 4246 "TX completion received with NULL socket buffer\n"); 4247 } 4248 tx_pool->free_map[tx_pool->producer_index] = index; 4249 tx_pool->producer_index = 4250 (tx_pool->producer_index + 1) % 4251 tx_pool->num_buffers; 4252 } 4253 /* remove tx_comp scrq*/ 4254 next->tx_comp.first = 0; 4255 4256 4257 if (atomic_sub_return(num_entries, &scrq->used) <= 4258 (adapter->req_tx_entries_per_subcrq / 2) && 4259 __netif_subqueue_stopped(adapter->netdev, 4260 scrq->pool_index)) { 4261 rcu_read_lock(); 4262 if (adapter->tx_queues_active) { 4263 netif_wake_subqueue(adapter->netdev, 4264 scrq->pool_index); 4265 netdev_dbg(adapter->netdev, 4266 "Started queue %d\n", 4267 scrq->pool_index); 4268 } 4269 rcu_read_unlock(); 4270 } 4271 } 4272 4273 enable_scrq_irq(adapter, scrq); 4274 4275 if (pending_scrq(adapter, scrq)) { 4276 disable_scrq_irq(adapter, scrq); 4277 goto restart_loop; 4278 } 4279 4280 txq = netdev_get_tx_queue(adapter->netdev, scrq->pool_index); 4281 netdev_tx_completed_queue(txq, num_packets, total_bytes); 4282 4283 return 0; 4284 } 4285 4286 static irqreturn_t ibmvnic_interrupt_tx(int irq, void *instance) 4287 { 4288 struct ibmvnic_sub_crq_queue *scrq = instance; 4289 struct ibmvnic_adapter *adapter = scrq->adapter; 4290 4291 disable_scrq_irq(adapter, scrq); 4292 ibmvnic_complete_tx(adapter, scrq); 4293 4294 return IRQ_HANDLED; 4295 } 4296 4297 static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance) 4298 { 4299 struct ibmvnic_sub_crq_queue *scrq = instance; 4300 struct ibmvnic_adapter *adapter = scrq->adapter; 4301 4302 /* When booting a kdump kernel we can hit pending interrupts 4303 * prior to completing driver initialization. 4304 */ 4305 if (unlikely(adapter->state != VNIC_OPEN)) 4306 return IRQ_NONE; 4307 4308 adapter->rx_stats_buffers[scrq->scrq_num].interrupts++; 4309 4310 if (napi_schedule_prep(&adapter->napi[scrq->scrq_num])) { 4311 disable_scrq_irq(adapter, scrq); 4312 __napi_schedule(&adapter->napi[scrq->scrq_num]); 4313 } 4314 4315 return IRQ_HANDLED; 4316 } 4317 4318 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter) 4319 { 4320 struct device *dev = &adapter->vdev->dev; 4321 struct ibmvnic_sub_crq_queue *scrq; 4322 int i = 0, j = 0; 4323 int rc = 0; 4324 4325 for (i = 0; i < adapter->req_tx_queues; i++) { 4326 netdev_dbg(adapter->netdev, "Initializing tx_scrq[%d] irq\n", 4327 i); 4328 scrq = adapter->tx_scrq[i]; 4329 scrq->irq = irq_create_mapping(NULL, scrq->hw_irq); 4330 4331 if (!scrq->irq) { 4332 rc = -EINVAL; 4333 dev_err(dev, "Error mapping irq\n"); 4334 goto req_tx_irq_failed; 4335 } 4336 4337 snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-tx%d", 4338 adapter->vdev->unit_address, i); 4339 rc = request_irq(scrq->irq, ibmvnic_interrupt_tx, 4340 0, scrq->name, scrq); 4341 4342 if (rc) { 4343 dev_err(dev, "Couldn't register tx irq 0x%x. rc=%d\n", 4344 scrq->irq, rc); 4345 irq_dispose_mapping(scrq->irq); 4346 goto req_tx_irq_failed; 4347 } 4348 } 4349 4350 for (i = 0; i < adapter->req_rx_queues; i++) { 4351 netdev_dbg(adapter->netdev, "Initializing rx_scrq[%d] irq\n", 4352 i); 4353 scrq = adapter->rx_scrq[i]; 4354 scrq->irq = irq_create_mapping(NULL, scrq->hw_irq); 4355 if (!scrq->irq) { 4356 rc = -EINVAL; 4357 dev_err(dev, "Error mapping irq\n"); 4358 goto req_rx_irq_failed; 4359 } 4360 snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-rx%d", 4361 adapter->vdev->unit_address, i); 4362 rc = request_irq(scrq->irq, ibmvnic_interrupt_rx, 4363 0, scrq->name, scrq); 4364 if (rc) { 4365 dev_err(dev, "Couldn't register rx irq 0x%x. rc=%d\n", 4366 scrq->irq, rc); 4367 irq_dispose_mapping(scrq->irq); 4368 goto req_rx_irq_failed; 4369 } 4370 } 4371 4372 cpus_read_lock(); 4373 ibmvnic_set_affinity(adapter); 4374 cpus_read_unlock(); 4375 4376 return rc; 4377 4378 req_rx_irq_failed: 4379 for (j = 0; j < i; j++) { 4380 free_irq(adapter->rx_scrq[j]->irq, adapter->rx_scrq[j]); 4381 irq_dispose_mapping(adapter->rx_scrq[j]->irq); 4382 } 4383 i = adapter->req_tx_queues; 4384 req_tx_irq_failed: 4385 for (j = 0; j < i; j++) { 4386 free_irq(adapter->tx_scrq[j]->irq, adapter->tx_scrq[j]); 4387 irq_dispose_mapping(adapter->tx_scrq[j]->irq); 4388 } 4389 release_sub_crqs(adapter, 1); 4390 return rc; 4391 } 4392 4393 static int init_sub_crqs(struct ibmvnic_adapter *adapter) 4394 { 4395 struct device *dev = &adapter->vdev->dev; 4396 struct ibmvnic_sub_crq_queue **allqueues; 4397 int registered_queues = 0; 4398 int total_queues; 4399 int more = 0; 4400 int i; 4401 4402 total_queues = adapter->req_tx_queues + adapter->req_rx_queues; 4403 4404 allqueues = kcalloc(total_queues, sizeof(*allqueues), GFP_KERNEL); 4405 if (!allqueues) 4406 return -ENOMEM; 4407 4408 for (i = 0; i < total_queues; i++) { 4409 allqueues[i] = init_sub_crq_queue(adapter); 4410 if (!allqueues[i]) { 4411 dev_warn(dev, "Couldn't allocate all sub-crqs\n"); 4412 break; 4413 } 4414 registered_queues++; 4415 } 4416 4417 /* Make sure we were able to register the minimum number of queues */ 4418 if (registered_queues < 4419 adapter->min_tx_queues + adapter->min_rx_queues) { 4420 dev_err(dev, "Fatal: Couldn't init min number of sub-crqs\n"); 4421 goto tx_failed; 4422 } 4423 4424 /* Distribute the failed allocated queues*/ 4425 for (i = 0; i < total_queues - registered_queues + more ; i++) { 4426 netdev_dbg(adapter->netdev, "Reducing number of queues\n"); 4427 switch (i % 3) { 4428 case 0: 4429 if (adapter->req_rx_queues > adapter->min_rx_queues) 4430 adapter->req_rx_queues--; 4431 else 4432 more++; 4433 break; 4434 case 1: 4435 if (adapter->req_tx_queues > adapter->min_tx_queues) 4436 adapter->req_tx_queues--; 4437 else 4438 more++; 4439 break; 4440 } 4441 } 4442 4443 adapter->tx_scrq = kcalloc(adapter->req_tx_queues, 4444 sizeof(*adapter->tx_scrq), GFP_KERNEL); 4445 if (!adapter->tx_scrq) 4446 goto tx_failed; 4447 4448 for (i = 0; i < adapter->req_tx_queues; i++) { 4449 adapter->tx_scrq[i] = allqueues[i]; 4450 adapter->tx_scrq[i]->pool_index = i; 4451 adapter->num_active_tx_scrqs++; 4452 } 4453 4454 adapter->rx_scrq = kcalloc(adapter->req_rx_queues, 4455 sizeof(*adapter->rx_scrq), GFP_KERNEL); 4456 if (!adapter->rx_scrq) 4457 goto rx_failed; 4458 4459 for (i = 0; i < adapter->req_rx_queues; i++) { 4460 adapter->rx_scrq[i] = allqueues[i + adapter->req_tx_queues]; 4461 adapter->rx_scrq[i]->scrq_num = i; 4462 adapter->num_active_rx_scrqs++; 4463 } 4464 4465 kfree(allqueues); 4466 return 0; 4467 4468 rx_failed: 4469 kfree(adapter->tx_scrq); 4470 adapter->tx_scrq = NULL; 4471 tx_failed: 4472 for (i = 0; i < registered_queues; i++) 4473 release_sub_crq_queue(adapter, allqueues[i], 1); 4474 kfree(allqueues); 4475 return -ENOMEM; 4476 } 4477 4478 static void send_request_cap(struct ibmvnic_adapter *adapter, int retry) 4479 { 4480 struct device *dev = &adapter->vdev->dev; 4481 union ibmvnic_crq crq; 4482 int max_entries; 4483 int cap_reqs; 4484 4485 /* We send out 6 or 7 REQUEST_CAPABILITY CRQs below (depending on 4486 * the PROMISC flag). Initialize this count upfront. When the tasklet 4487 * receives a response to all of these, it will send the next protocol 4488 * message (QUERY_IP_OFFLOAD). 4489 */ 4490 if (!(adapter->netdev->flags & IFF_PROMISC) || 4491 adapter->promisc_supported) 4492 cap_reqs = 7; 4493 else 4494 cap_reqs = 6; 4495 4496 if (!retry) { 4497 /* Sub-CRQ entries are 32 byte long */ 4498 int entries_page = 4 * PAGE_SIZE / (sizeof(u64) * 4); 4499 4500 atomic_set(&adapter->running_cap_crqs, cap_reqs); 4501 4502 if (adapter->min_tx_entries_per_subcrq > entries_page || 4503 adapter->min_rx_add_entries_per_subcrq > entries_page) { 4504 dev_err(dev, "Fatal, invalid entries per sub-crq\n"); 4505 return; 4506 } 4507 4508 if (adapter->desired.mtu) 4509 adapter->req_mtu = adapter->desired.mtu; 4510 else 4511 adapter->req_mtu = adapter->netdev->mtu + ETH_HLEN; 4512 4513 if (!adapter->desired.tx_entries) 4514 adapter->desired.tx_entries = 4515 adapter->max_tx_entries_per_subcrq; 4516 if (!adapter->desired.rx_entries) 4517 adapter->desired.rx_entries = 4518 adapter->max_rx_add_entries_per_subcrq; 4519 4520 max_entries = IBMVNIC_LTB_SET_SIZE / 4521 (adapter->req_mtu + IBMVNIC_BUFFER_HLEN); 4522 4523 if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) * 4524 adapter->desired.tx_entries > IBMVNIC_LTB_SET_SIZE) { 4525 adapter->desired.tx_entries = max_entries; 4526 } 4527 4528 if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) * 4529 adapter->desired.rx_entries > IBMVNIC_LTB_SET_SIZE) { 4530 adapter->desired.rx_entries = max_entries; 4531 } 4532 4533 if (adapter->desired.tx_entries) 4534 adapter->req_tx_entries_per_subcrq = 4535 adapter->desired.tx_entries; 4536 else 4537 adapter->req_tx_entries_per_subcrq = 4538 adapter->max_tx_entries_per_subcrq; 4539 4540 if (adapter->desired.rx_entries) 4541 adapter->req_rx_add_entries_per_subcrq = 4542 adapter->desired.rx_entries; 4543 else 4544 adapter->req_rx_add_entries_per_subcrq = 4545 adapter->max_rx_add_entries_per_subcrq; 4546 4547 if (adapter->desired.tx_queues) 4548 adapter->req_tx_queues = 4549 adapter->desired.tx_queues; 4550 else 4551 adapter->req_tx_queues = 4552 adapter->opt_tx_comp_sub_queues; 4553 4554 if (adapter->desired.rx_queues) 4555 adapter->req_rx_queues = 4556 adapter->desired.rx_queues; 4557 else 4558 adapter->req_rx_queues = 4559 adapter->opt_rx_comp_queues; 4560 4561 adapter->req_rx_add_queues = adapter->max_rx_add_queues; 4562 } else { 4563 atomic_add(cap_reqs, &adapter->running_cap_crqs); 4564 } 4565 memset(&crq, 0, sizeof(crq)); 4566 crq.request_capability.first = IBMVNIC_CRQ_CMD; 4567 crq.request_capability.cmd = REQUEST_CAPABILITY; 4568 4569 crq.request_capability.capability = cpu_to_be16(REQ_TX_QUEUES); 4570 crq.request_capability.number = cpu_to_be64(adapter->req_tx_queues); 4571 cap_reqs--; 4572 ibmvnic_send_crq(adapter, &crq); 4573 4574 crq.request_capability.capability = cpu_to_be16(REQ_RX_QUEUES); 4575 crq.request_capability.number = cpu_to_be64(adapter->req_rx_queues); 4576 cap_reqs--; 4577 ibmvnic_send_crq(adapter, &crq); 4578 4579 crq.request_capability.capability = cpu_to_be16(REQ_RX_ADD_QUEUES); 4580 crq.request_capability.number = cpu_to_be64(adapter->req_rx_add_queues); 4581 cap_reqs--; 4582 ibmvnic_send_crq(adapter, &crq); 4583 4584 crq.request_capability.capability = 4585 cpu_to_be16(REQ_TX_ENTRIES_PER_SUBCRQ); 4586 crq.request_capability.number = 4587 cpu_to_be64(adapter->req_tx_entries_per_subcrq); 4588 cap_reqs--; 4589 ibmvnic_send_crq(adapter, &crq); 4590 4591 crq.request_capability.capability = 4592 cpu_to_be16(REQ_RX_ADD_ENTRIES_PER_SUBCRQ); 4593 crq.request_capability.number = 4594 cpu_to_be64(adapter->req_rx_add_entries_per_subcrq); 4595 cap_reqs--; 4596 ibmvnic_send_crq(adapter, &crq); 4597 4598 crq.request_capability.capability = cpu_to_be16(REQ_MTU); 4599 crq.request_capability.number = cpu_to_be64(adapter->req_mtu); 4600 cap_reqs--; 4601 ibmvnic_send_crq(adapter, &crq); 4602 4603 if (adapter->netdev->flags & IFF_PROMISC) { 4604 if (adapter->promisc_supported) { 4605 crq.request_capability.capability = 4606 cpu_to_be16(PROMISC_REQUESTED); 4607 crq.request_capability.number = cpu_to_be64(1); 4608 cap_reqs--; 4609 ibmvnic_send_crq(adapter, &crq); 4610 } 4611 } else { 4612 crq.request_capability.capability = 4613 cpu_to_be16(PROMISC_REQUESTED); 4614 crq.request_capability.number = cpu_to_be64(0); 4615 cap_reqs--; 4616 ibmvnic_send_crq(adapter, &crq); 4617 } 4618 4619 /* Keep at end to catch any discrepancy between expected and actual 4620 * CRQs sent. 4621 */ 4622 WARN_ON(cap_reqs != 0); 4623 } 4624 4625 static int pending_scrq(struct ibmvnic_adapter *adapter, 4626 struct ibmvnic_sub_crq_queue *scrq) 4627 { 4628 union sub_crq *entry = &scrq->msgs[scrq->cur]; 4629 int rc; 4630 4631 rc = !!(entry->generic.first & IBMVNIC_CRQ_CMD_RSP); 4632 4633 /* Ensure that the SCRQ valid flag is loaded prior to loading the 4634 * contents of the SCRQ descriptor 4635 */ 4636 dma_rmb(); 4637 4638 return rc; 4639 } 4640 4641 static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *adapter, 4642 struct ibmvnic_sub_crq_queue *scrq) 4643 { 4644 union sub_crq *entry; 4645 unsigned long flags; 4646 4647 spin_lock_irqsave(&scrq->lock, flags); 4648 entry = &scrq->msgs[scrq->cur]; 4649 if (entry->generic.first & IBMVNIC_CRQ_CMD_RSP) { 4650 if (++scrq->cur == scrq->size) 4651 scrq->cur = 0; 4652 } else { 4653 entry = NULL; 4654 } 4655 spin_unlock_irqrestore(&scrq->lock, flags); 4656 4657 /* Ensure that the SCRQ valid flag is loaded prior to loading the 4658 * contents of the SCRQ descriptor 4659 */ 4660 dma_rmb(); 4661 4662 return entry; 4663 } 4664 4665 static union ibmvnic_crq *ibmvnic_next_crq(struct ibmvnic_adapter *adapter) 4666 { 4667 struct ibmvnic_crq_queue *queue = &adapter->crq; 4668 union ibmvnic_crq *crq; 4669 4670 crq = &queue->msgs[queue->cur]; 4671 if (crq->generic.first & IBMVNIC_CRQ_CMD_RSP) { 4672 if (++queue->cur == queue->size) 4673 queue->cur = 0; 4674 } else { 4675 crq = NULL; 4676 } 4677 4678 return crq; 4679 } 4680 4681 static void print_subcrq_error(struct device *dev, int rc, const char *func) 4682 { 4683 switch (rc) { 4684 case H_PARAMETER: 4685 dev_warn_ratelimited(dev, 4686 "%s failed: Send request is malformed or adapter failover pending. (rc=%d)\n", 4687 func, rc); 4688 break; 4689 case H_CLOSED: 4690 dev_warn_ratelimited(dev, 4691 "%s failed: Backing queue closed. Adapter is down or failover pending. (rc=%d)\n", 4692 func, rc); 4693 break; 4694 default: 4695 dev_err_ratelimited(dev, "%s failed: (rc=%d)\n", func, rc); 4696 break; 4697 } 4698 } 4699 4700 static int send_subcrq_indirect(struct ibmvnic_adapter *adapter, 4701 u64 remote_handle, u64 ioba, u64 num_entries) 4702 { 4703 unsigned int ua = adapter->vdev->unit_address; 4704 struct device *dev = &adapter->vdev->dev; 4705 int rc; 4706 4707 /* Make sure the hypervisor sees the complete request */ 4708 dma_wmb(); 4709 rc = plpar_hcall_norets(H_SEND_SUB_CRQ_INDIRECT, ua, 4710 cpu_to_be64(remote_handle), 4711 ioba, num_entries); 4712 4713 if (rc) 4714 print_subcrq_error(dev, rc, __func__); 4715 4716 return rc; 4717 } 4718 4719 static int ibmvnic_send_crq(struct ibmvnic_adapter *adapter, 4720 union ibmvnic_crq *crq) 4721 { 4722 unsigned int ua = adapter->vdev->unit_address; 4723 struct device *dev = &adapter->vdev->dev; 4724 u64 *u64_crq = (u64 *)crq; 4725 int rc; 4726 4727 netdev_dbg(adapter->netdev, "Sending CRQ: %016lx %016lx\n", 4728 (unsigned long)cpu_to_be64(u64_crq[0]), 4729 (unsigned long)cpu_to_be64(u64_crq[1])); 4730 4731 if (!adapter->crq.active && 4732 crq->generic.first != IBMVNIC_CRQ_INIT_CMD) { 4733 dev_warn(dev, "Invalid request detected while CRQ is inactive, possible device state change during reset\n"); 4734 return -EINVAL; 4735 } 4736 4737 /* Make sure the hypervisor sees the complete request */ 4738 dma_wmb(); 4739 4740 rc = plpar_hcall_norets(H_SEND_CRQ, ua, 4741 cpu_to_be64(u64_crq[0]), 4742 cpu_to_be64(u64_crq[1])); 4743 4744 if (rc) { 4745 if (rc == H_CLOSED) { 4746 dev_warn(dev, "CRQ Queue closed\n"); 4747 /* do not reset, report the fail, wait for passive init from server */ 4748 } 4749 4750 dev_warn(dev, "Send error (rc=%d)\n", rc); 4751 } 4752 4753 return rc; 4754 } 4755 4756 static int ibmvnic_send_crq_init(struct ibmvnic_adapter *adapter) 4757 { 4758 struct device *dev = &adapter->vdev->dev; 4759 union ibmvnic_crq crq; 4760 int retries = 100; 4761 int rc; 4762 4763 memset(&crq, 0, sizeof(crq)); 4764 crq.generic.first = IBMVNIC_CRQ_INIT_CMD; 4765 crq.generic.cmd = IBMVNIC_CRQ_INIT; 4766 netdev_dbg(adapter->netdev, "Sending CRQ init\n"); 4767 4768 do { 4769 rc = ibmvnic_send_crq(adapter, &crq); 4770 if (rc != H_CLOSED) 4771 break; 4772 retries--; 4773 msleep(50); 4774 4775 } while (retries > 0); 4776 4777 if (rc) { 4778 dev_err(dev, "Failed to send init request, rc = %d\n", rc); 4779 return rc; 4780 } 4781 4782 return 0; 4783 } 4784 4785 struct vnic_login_client_data { 4786 u8 type; 4787 __be16 len; 4788 char name[]; 4789 } __packed; 4790 4791 static int vnic_client_data_len(struct ibmvnic_adapter *adapter) 4792 { 4793 int len; 4794 4795 /* Calculate the amount of buffer space needed for the 4796 * vnic client data in the login buffer. There are four entries, 4797 * OS name, LPAR name, device name, and a null last entry. 4798 */ 4799 len = 4 * sizeof(struct vnic_login_client_data); 4800 len += 6; /* "Linux" plus NULL */ 4801 len += strlen(utsname()->nodename) + 1; 4802 len += strlen(adapter->netdev->name) + 1; 4803 4804 return len; 4805 } 4806 4807 static void vnic_add_client_data(struct ibmvnic_adapter *adapter, 4808 struct vnic_login_client_data *vlcd) 4809 { 4810 const char *os_name = "Linux"; 4811 int len; 4812 4813 /* Type 1 - LPAR OS */ 4814 vlcd->type = 1; 4815 len = strlen(os_name) + 1; 4816 vlcd->len = cpu_to_be16(len); 4817 strscpy(vlcd->name, os_name, len); 4818 vlcd = (struct vnic_login_client_data *)(vlcd->name + len); 4819 4820 /* Type 2 - LPAR name */ 4821 vlcd->type = 2; 4822 len = strlen(utsname()->nodename) + 1; 4823 vlcd->len = cpu_to_be16(len); 4824 strscpy(vlcd->name, utsname()->nodename, len); 4825 vlcd = (struct vnic_login_client_data *)(vlcd->name + len); 4826 4827 /* Type 3 - device name */ 4828 vlcd->type = 3; 4829 len = strlen(adapter->netdev->name) + 1; 4830 vlcd->len = cpu_to_be16(len); 4831 strscpy(vlcd->name, adapter->netdev->name, len); 4832 } 4833 4834 static int send_login(struct ibmvnic_adapter *adapter) 4835 { 4836 struct ibmvnic_login_rsp_buffer *login_rsp_buffer; 4837 struct ibmvnic_login_buffer *login_buffer; 4838 struct device *dev = &adapter->vdev->dev; 4839 struct vnic_login_client_data *vlcd; 4840 dma_addr_t rsp_buffer_token; 4841 dma_addr_t buffer_token; 4842 size_t rsp_buffer_size; 4843 union ibmvnic_crq crq; 4844 int client_data_len; 4845 size_t buffer_size; 4846 __be64 *tx_list_p; 4847 __be64 *rx_list_p; 4848 int rc; 4849 int i; 4850 4851 if (!adapter->tx_scrq || !adapter->rx_scrq) { 4852 netdev_err(adapter->netdev, 4853 "RX or TX queues are not allocated, device login failed\n"); 4854 return -ENOMEM; 4855 } 4856 4857 release_login_buffer(adapter); 4858 release_login_rsp_buffer(adapter); 4859 4860 client_data_len = vnic_client_data_len(adapter); 4861 4862 buffer_size = 4863 sizeof(struct ibmvnic_login_buffer) + 4864 sizeof(u64) * (adapter->req_tx_queues + adapter->req_rx_queues) + 4865 client_data_len; 4866 4867 login_buffer = kzalloc(buffer_size, GFP_ATOMIC); 4868 if (!login_buffer) 4869 goto buf_alloc_failed; 4870 4871 buffer_token = dma_map_single(dev, login_buffer, buffer_size, 4872 DMA_TO_DEVICE); 4873 if (dma_mapping_error(dev, buffer_token)) { 4874 dev_err(dev, "Couldn't map login buffer\n"); 4875 goto buf_map_failed; 4876 } 4877 4878 rsp_buffer_size = sizeof(struct ibmvnic_login_rsp_buffer) + 4879 sizeof(u64) * adapter->req_tx_queues + 4880 sizeof(u64) * adapter->req_rx_queues + 4881 sizeof(u64) * adapter->req_rx_queues + 4882 sizeof(u8) * IBMVNIC_TX_DESC_VERSIONS; 4883 4884 login_rsp_buffer = kmalloc(rsp_buffer_size, GFP_ATOMIC); 4885 if (!login_rsp_buffer) 4886 goto buf_rsp_alloc_failed; 4887 4888 rsp_buffer_token = dma_map_single(dev, login_rsp_buffer, 4889 rsp_buffer_size, DMA_FROM_DEVICE); 4890 if (dma_mapping_error(dev, rsp_buffer_token)) { 4891 dev_err(dev, "Couldn't map login rsp buffer\n"); 4892 goto buf_rsp_map_failed; 4893 } 4894 4895 adapter->login_buf = login_buffer; 4896 adapter->login_buf_token = buffer_token; 4897 adapter->login_buf_sz = buffer_size; 4898 adapter->login_rsp_buf = login_rsp_buffer; 4899 adapter->login_rsp_buf_token = rsp_buffer_token; 4900 adapter->login_rsp_buf_sz = rsp_buffer_size; 4901 4902 login_buffer->len = cpu_to_be32(buffer_size); 4903 login_buffer->version = cpu_to_be32(INITIAL_VERSION_LB); 4904 login_buffer->num_txcomp_subcrqs = cpu_to_be32(adapter->req_tx_queues); 4905 login_buffer->off_txcomp_subcrqs = 4906 cpu_to_be32(sizeof(struct ibmvnic_login_buffer)); 4907 login_buffer->num_rxcomp_subcrqs = cpu_to_be32(adapter->req_rx_queues); 4908 login_buffer->off_rxcomp_subcrqs = 4909 cpu_to_be32(sizeof(struct ibmvnic_login_buffer) + 4910 sizeof(u64) * adapter->req_tx_queues); 4911 login_buffer->login_rsp_ioba = cpu_to_be32(rsp_buffer_token); 4912 login_buffer->login_rsp_len = cpu_to_be32(rsp_buffer_size); 4913 4914 tx_list_p = (__be64 *)((char *)login_buffer + 4915 sizeof(struct ibmvnic_login_buffer)); 4916 rx_list_p = (__be64 *)((char *)login_buffer + 4917 sizeof(struct ibmvnic_login_buffer) + 4918 sizeof(u64) * adapter->req_tx_queues); 4919 4920 for (i = 0; i < adapter->req_tx_queues; i++) { 4921 if (adapter->tx_scrq[i]) { 4922 tx_list_p[i] = 4923 cpu_to_be64(adapter->tx_scrq[i]->crq_num); 4924 } 4925 } 4926 4927 for (i = 0; i < adapter->req_rx_queues; i++) { 4928 if (adapter->rx_scrq[i]) { 4929 rx_list_p[i] = 4930 cpu_to_be64(adapter->rx_scrq[i]->crq_num); 4931 } 4932 } 4933 4934 /* Insert vNIC login client data */ 4935 vlcd = (struct vnic_login_client_data *) 4936 ((char *)rx_list_p + (sizeof(u64) * adapter->req_rx_queues)); 4937 login_buffer->client_data_offset = 4938 cpu_to_be32((char *)vlcd - (char *)login_buffer); 4939 login_buffer->client_data_len = cpu_to_be32(client_data_len); 4940 4941 vnic_add_client_data(adapter, vlcd); 4942 4943 netdev_dbg(adapter->netdev, "Login Buffer:\n"); 4944 for (i = 0; i < (adapter->login_buf_sz - 1) / 8 + 1; i++) { 4945 netdev_dbg(adapter->netdev, "%016lx\n", 4946 ((unsigned long *)(adapter->login_buf))[i]); 4947 } 4948 4949 memset(&crq, 0, sizeof(crq)); 4950 crq.login.first = IBMVNIC_CRQ_CMD; 4951 crq.login.cmd = LOGIN; 4952 crq.login.ioba = cpu_to_be32(buffer_token); 4953 crq.login.len = cpu_to_be32(buffer_size); 4954 4955 adapter->login_pending = true; 4956 rc = ibmvnic_send_crq(adapter, &crq); 4957 if (rc) { 4958 adapter->login_pending = false; 4959 netdev_err(adapter->netdev, "Failed to send login, rc=%d\n", rc); 4960 goto buf_send_failed; 4961 } 4962 4963 return 0; 4964 4965 buf_send_failed: 4966 dma_unmap_single(dev, rsp_buffer_token, rsp_buffer_size, 4967 DMA_FROM_DEVICE); 4968 buf_rsp_map_failed: 4969 kfree(login_rsp_buffer); 4970 adapter->login_rsp_buf = NULL; 4971 buf_rsp_alloc_failed: 4972 dma_unmap_single(dev, buffer_token, buffer_size, DMA_TO_DEVICE); 4973 buf_map_failed: 4974 kfree(login_buffer); 4975 adapter->login_buf = NULL; 4976 buf_alloc_failed: 4977 return -ENOMEM; 4978 } 4979 4980 static int send_request_map(struct ibmvnic_adapter *adapter, dma_addr_t addr, 4981 u32 len, u8 map_id) 4982 { 4983 union ibmvnic_crq crq; 4984 4985 memset(&crq, 0, sizeof(crq)); 4986 crq.request_map.first = IBMVNIC_CRQ_CMD; 4987 crq.request_map.cmd = REQUEST_MAP; 4988 crq.request_map.map_id = map_id; 4989 crq.request_map.ioba = cpu_to_be32(addr); 4990 crq.request_map.len = cpu_to_be32(len); 4991 return ibmvnic_send_crq(adapter, &crq); 4992 } 4993 4994 static int send_request_unmap(struct ibmvnic_adapter *adapter, u8 map_id) 4995 { 4996 union ibmvnic_crq crq; 4997 4998 memset(&crq, 0, sizeof(crq)); 4999 crq.request_unmap.first = IBMVNIC_CRQ_CMD; 5000 crq.request_unmap.cmd = REQUEST_UNMAP; 5001 crq.request_unmap.map_id = map_id; 5002 return ibmvnic_send_crq(adapter, &crq); 5003 } 5004 5005 static void send_query_map(struct ibmvnic_adapter *adapter) 5006 { 5007 union ibmvnic_crq crq; 5008 5009 memset(&crq, 0, sizeof(crq)); 5010 crq.query_map.first = IBMVNIC_CRQ_CMD; 5011 crq.query_map.cmd = QUERY_MAP; 5012 ibmvnic_send_crq(adapter, &crq); 5013 } 5014 5015 /* Send a series of CRQs requesting various capabilities of the VNIC server */ 5016 static void send_query_cap(struct ibmvnic_adapter *adapter) 5017 { 5018 union ibmvnic_crq crq; 5019 int cap_reqs; 5020 5021 /* We send out 25 QUERY_CAPABILITY CRQs below. Initialize this count 5022 * upfront. When the tasklet receives a response to all of these, it 5023 * can send out the next protocol messaage (REQUEST_CAPABILITY). 5024 */ 5025 cap_reqs = 25; 5026 5027 atomic_set(&adapter->running_cap_crqs, cap_reqs); 5028 5029 memset(&crq, 0, sizeof(crq)); 5030 crq.query_capability.first = IBMVNIC_CRQ_CMD; 5031 crq.query_capability.cmd = QUERY_CAPABILITY; 5032 5033 crq.query_capability.capability = cpu_to_be16(MIN_TX_QUEUES); 5034 ibmvnic_send_crq(adapter, &crq); 5035 cap_reqs--; 5036 5037 crq.query_capability.capability = cpu_to_be16(MIN_RX_QUEUES); 5038 ibmvnic_send_crq(adapter, &crq); 5039 cap_reqs--; 5040 5041 crq.query_capability.capability = cpu_to_be16(MIN_RX_ADD_QUEUES); 5042 ibmvnic_send_crq(adapter, &crq); 5043 cap_reqs--; 5044 5045 crq.query_capability.capability = cpu_to_be16(MAX_TX_QUEUES); 5046 ibmvnic_send_crq(adapter, &crq); 5047 cap_reqs--; 5048 5049 crq.query_capability.capability = cpu_to_be16(MAX_RX_QUEUES); 5050 ibmvnic_send_crq(adapter, &crq); 5051 cap_reqs--; 5052 5053 crq.query_capability.capability = cpu_to_be16(MAX_RX_ADD_QUEUES); 5054 ibmvnic_send_crq(adapter, &crq); 5055 cap_reqs--; 5056 5057 crq.query_capability.capability = 5058 cpu_to_be16(MIN_TX_ENTRIES_PER_SUBCRQ); 5059 ibmvnic_send_crq(adapter, &crq); 5060 cap_reqs--; 5061 5062 crq.query_capability.capability = 5063 cpu_to_be16(MIN_RX_ADD_ENTRIES_PER_SUBCRQ); 5064 ibmvnic_send_crq(adapter, &crq); 5065 cap_reqs--; 5066 5067 crq.query_capability.capability = 5068 cpu_to_be16(MAX_TX_ENTRIES_PER_SUBCRQ); 5069 ibmvnic_send_crq(adapter, &crq); 5070 cap_reqs--; 5071 5072 crq.query_capability.capability = 5073 cpu_to_be16(MAX_RX_ADD_ENTRIES_PER_SUBCRQ); 5074 ibmvnic_send_crq(adapter, &crq); 5075 cap_reqs--; 5076 5077 crq.query_capability.capability = cpu_to_be16(TCP_IP_OFFLOAD); 5078 ibmvnic_send_crq(adapter, &crq); 5079 cap_reqs--; 5080 5081 crq.query_capability.capability = cpu_to_be16(PROMISC_SUPPORTED); 5082 ibmvnic_send_crq(adapter, &crq); 5083 cap_reqs--; 5084 5085 crq.query_capability.capability = cpu_to_be16(MIN_MTU); 5086 ibmvnic_send_crq(adapter, &crq); 5087 cap_reqs--; 5088 5089 crq.query_capability.capability = cpu_to_be16(MAX_MTU); 5090 ibmvnic_send_crq(adapter, &crq); 5091 cap_reqs--; 5092 5093 crq.query_capability.capability = cpu_to_be16(MAX_MULTICAST_FILTERS); 5094 ibmvnic_send_crq(adapter, &crq); 5095 cap_reqs--; 5096 5097 crq.query_capability.capability = cpu_to_be16(VLAN_HEADER_INSERTION); 5098 ibmvnic_send_crq(adapter, &crq); 5099 cap_reqs--; 5100 5101 crq.query_capability.capability = cpu_to_be16(RX_VLAN_HEADER_INSERTION); 5102 ibmvnic_send_crq(adapter, &crq); 5103 cap_reqs--; 5104 5105 crq.query_capability.capability = cpu_to_be16(MAX_TX_SG_ENTRIES); 5106 ibmvnic_send_crq(adapter, &crq); 5107 cap_reqs--; 5108 5109 crq.query_capability.capability = cpu_to_be16(RX_SG_SUPPORTED); 5110 ibmvnic_send_crq(adapter, &crq); 5111 cap_reqs--; 5112 5113 crq.query_capability.capability = cpu_to_be16(OPT_TX_COMP_SUB_QUEUES); 5114 ibmvnic_send_crq(adapter, &crq); 5115 cap_reqs--; 5116 5117 crq.query_capability.capability = cpu_to_be16(OPT_RX_COMP_QUEUES); 5118 ibmvnic_send_crq(adapter, &crq); 5119 cap_reqs--; 5120 5121 crq.query_capability.capability = 5122 cpu_to_be16(OPT_RX_BUFADD_Q_PER_RX_COMP_Q); 5123 ibmvnic_send_crq(adapter, &crq); 5124 cap_reqs--; 5125 5126 crq.query_capability.capability = 5127 cpu_to_be16(OPT_TX_ENTRIES_PER_SUBCRQ); 5128 ibmvnic_send_crq(adapter, &crq); 5129 cap_reqs--; 5130 5131 crq.query_capability.capability = 5132 cpu_to_be16(OPT_RXBA_ENTRIES_PER_SUBCRQ); 5133 ibmvnic_send_crq(adapter, &crq); 5134 cap_reqs--; 5135 5136 crq.query_capability.capability = cpu_to_be16(TX_RX_DESC_REQ); 5137 5138 ibmvnic_send_crq(adapter, &crq); 5139 cap_reqs--; 5140 5141 /* Keep at end to catch any discrepancy between expected and actual 5142 * CRQs sent. 5143 */ 5144 WARN_ON(cap_reqs != 0); 5145 } 5146 5147 static void send_query_ip_offload(struct ibmvnic_adapter *adapter) 5148 { 5149 int buf_sz = sizeof(struct ibmvnic_query_ip_offload_buffer); 5150 struct device *dev = &adapter->vdev->dev; 5151 union ibmvnic_crq crq; 5152 5153 adapter->ip_offload_tok = 5154 dma_map_single(dev, 5155 &adapter->ip_offload_buf, 5156 buf_sz, 5157 DMA_FROM_DEVICE); 5158 5159 if (dma_mapping_error(dev, adapter->ip_offload_tok)) { 5160 if (!firmware_has_feature(FW_FEATURE_CMO)) 5161 dev_err(dev, "Couldn't map offload buffer\n"); 5162 return; 5163 } 5164 5165 memset(&crq, 0, sizeof(crq)); 5166 crq.query_ip_offload.first = IBMVNIC_CRQ_CMD; 5167 crq.query_ip_offload.cmd = QUERY_IP_OFFLOAD; 5168 crq.query_ip_offload.len = cpu_to_be32(buf_sz); 5169 crq.query_ip_offload.ioba = 5170 cpu_to_be32(adapter->ip_offload_tok); 5171 5172 ibmvnic_send_crq(adapter, &crq); 5173 } 5174 5175 static void send_control_ip_offload(struct ibmvnic_adapter *adapter) 5176 { 5177 struct ibmvnic_control_ip_offload_buffer *ctrl_buf = &adapter->ip_offload_ctrl; 5178 struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf; 5179 struct device *dev = &adapter->vdev->dev; 5180 netdev_features_t old_hw_features = 0; 5181 union ibmvnic_crq crq; 5182 5183 adapter->ip_offload_ctrl_tok = 5184 dma_map_single(dev, 5185 ctrl_buf, 5186 sizeof(adapter->ip_offload_ctrl), 5187 DMA_TO_DEVICE); 5188 5189 if (dma_mapping_error(dev, adapter->ip_offload_ctrl_tok)) { 5190 dev_err(dev, "Couldn't map ip offload control buffer\n"); 5191 return; 5192 } 5193 5194 ctrl_buf->len = cpu_to_be32(sizeof(adapter->ip_offload_ctrl)); 5195 ctrl_buf->version = cpu_to_be32(INITIAL_VERSION_IOB); 5196 ctrl_buf->ipv4_chksum = buf->ipv4_chksum; 5197 ctrl_buf->ipv6_chksum = buf->ipv6_chksum; 5198 ctrl_buf->tcp_ipv4_chksum = buf->tcp_ipv4_chksum; 5199 ctrl_buf->udp_ipv4_chksum = buf->udp_ipv4_chksum; 5200 ctrl_buf->tcp_ipv6_chksum = buf->tcp_ipv6_chksum; 5201 ctrl_buf->udp_ipv6_chksum = buf->udp_ipv6_chksum; 5202 ctrl_buf->large_tx_ipv4 = buf->large_tx_ipv4; 5203 ctrl_buf->large_tx_ipv6 = buf->large_tx_ipv6; 5204 5205 /* large_rx disabled for now, additional features needed */ 5206 ctrl_buf->large_rx_ipv4 = 0; 5207 ctrl_buf->large_rx_ipv6 = 0; 5208 5209 if (adapter->state != VNIC_PROBING) { 5210 old_hw_features = adapter->netdev->hw_features; 5211 adapter->netdev->hw_features = 0; 5212 } 5213 5214 adapter->netdev->hw_features = NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO; 5215 5216 if (buf->tcp_ipv4_chksum || buf->udp_ipv4_chksum) 5217 adapter->netdev->hw_features |= NETIF_F_IP_CSUM; 5218 5219 if (buf->tcp_ipv6_chksum || buf->udp_ipv6_chksum) 5220 adapter->netdev->hw_features |= NETIF_F_IPV6_CSUM; 5221 5222 if ((adapter->netdev->features & 5223 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM))) 5224 adapter->netdev->hw_features |= NETIF_F_RXCSUM; 5225 5226 if (buf->large_tx_ipv4) 5227 adapter->netdev->hw_features |= NETIF_F_TSO; 5228 if (buf->large_tx_ipv6) 5229 adapter->netdev->hw_features |= NETIF_F_TSO6; 5230 5231 if (adapter->state == VNIC_PROBING) { 5232 adapter->netdev->features |= adapter->netdev->hw_features; 5233 } else if (old_hw_features != adapter->netdev->hw_features) { 5234 netdev_features_t tmp = 0; 5235 5236 /* disable features no longer supported */ 5237 adapter->netdev->features &= adapter->netdev->hw_features; 5238 /* turn on features now supported if previously enabled */ 5239 tmp = (old_hw_features ^ adapter->netdev->hw_features) & 5240 adapter->netdev->hw_features; 5241 adapter->netdev->features |= 5242 tmp & adapter->netdev->wanted_features; 5243 } 5244 5245 memset(&crq, 0, sizeof(crq)); 5246 crq.control_ip_offload.first = IBMVNIC_CRQ_CMD; 5247 crq.control_ip_offload.cmd = CONTROL_IP_OFFLOAD; 5248 crq.control_ip_offload.len = 5249 cpu_to_be32(sizeof(adapter->ip_offload_ctrl)); 5250 crq.control_ip_offload.ioba = cpu_to_be32(adapter->ip_offload_ctrl_tok); 5251 ibmvnic_send_crq(adapter, &crq); 5252 } 5253 5254 static void handle_vpd_size_rsp(union ibmvnic_crq *crq, 5255 struct ibmvnic_adapter *adapter) 5256 { 5257 struct device *dev = &adapter->vdev->dev; 5258 5259 if (crq->get_vpd_size_rsp.rc.code) { 5260 dev_err(dev, "Error retrieving VPD size, rc=%x\n", 5261 crq->get_vpd_size_rsp.rc.code); 5262 complete(&adapter->fw_done); 5263 return; 5264 } 5265 5266 adapter->vpd->len = be64_to_cpu(crq->get_vpd_size_rsp.len); 5267 complete(&adapter->fw_done); 5268 } 5269 5270 static void handle_vpd_rsp(union ibmvnic_crq *crq, 5271 struct ibmvnic_adapter *adapter) 5272 { 5273 struct device *dev = &adapter->vdev->dev; 5274 unsigned char *substr = NULL; 5275 u8 fw_level_len = 0; 5276 5277 memset(adapter->fw_version, 0, 32); 5278 5279 dma_unmap_single(dev, adapter->vpd->dma_addr, adapter->vpd->len, 5280 DMA_FROM_DEVICE); 5281 5282 if (crq->get_vpd_rsp.rc.code) { 5283 dev_err(dev, "Error retrieving VPD from device, rc=%x\n", 5284 crq->get_vpd_rsp.rc.code); 5285 goto complete; 5286 } 5287 5288 /* get the position of the firmware version info 5289 * located after the ASCII 'RM' substring in the buffer 5290 */ 5291 substr = strnstr(adapter->vpd->buff, "RM", adapter->vpd->len); 5292 if (!substr) { 5293 dev_info(dev, "Warning - No FW level has been provided in the VPD buffer by the VIOS Server\n"); 5294 goto complete; 5295 } 5296 5297 /* get length of firmware level ASCII substring */ 5298 if ((substr + 2) < (adapter->vpd->buff + adapter->vpd->len)) { 5299 fw_level_len = *(substr + 2); 5300 } else { 5301 dev_info(dev, "Length of FW substr extrapolated VDP buff\n"); 5302 goto complete; 5303 } 5304 5305 /* copy firmware version string from vpd into adapter */ 5306 if ((substr + 3 + fw_level_len) < 5307 (adapter->vpd->buff + adapter->vpd->len)) { 5308 strscpy(adapter->fw_version, substr + 3, 5309 sizeof(adapter->fw_version)); 5310 } else { 5311 dev_info(dev, "FW substr extrapolated VPD buff\n"); 5312 } 5313 5314 complete: 5315 if (adapter->fw_version[0] == '\0') 5316 strscpy((char *)adapter->fw_version, "N/A", sizeof(adapter->fw_version)); 5317 complete(&adapter->fw_done); 5318 } 5319 5320 static void handle_query_ip_offload_rsp(struct ibmvnic_adapter *adapter) 5321 { 5322 struct device *dev = &adapter->vdev->dev; 5323 struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf; 5324 int i; 5325 5326 dma_unmap_single(dev, adapter->ip_offload_tok, 5327 sizeof(adapter->ip_offload_buf), DMA_FROM_DEVICE); 5328 5329 netdev_dbg(adapter->netdev, "Query IP Offload Buffer:\n"); 5330 for (i = 0; i < (sizeof(adapter->ip_offload_buf) - 1) / 8 + 1; i++) 5331 netdev_dbg(adapter->netdev, "%016lx\n", 5332 ((unsigned long *)(buf))[i]); 5333 5334 netdev_dbg(adapter->netdev, "ipv4_chksum = %d\n", buf->ipv4_chksum); 5335 netdev_dbg(adapter->netdev, "ipv6_chksum = %d\n", buf->ipv6_chksum); 5336 netdev_dbg(adapter->netdev, "tcp_ipv4_chksum = %d\n", 5337 buf->tcp_ipv4_chksum); 5338 netdev_dbg(adapter->netdev, "tcp_ipv6_chksum = %d\n", 5339 buf->tcp_ipv6_chksum); 5340 netdev_dbg(adapter->netdev, "udp_ipv4_chksum = %d\n", 5341 buf->udp_ipv4_chksum); 5342 netdev_dbg(adapter->netdev, "udp_ipv6_chksum = %d\n", 5343 buf->udp_ipv6_chksum); 5344 netdev_dbg(adapter->netdev, "large_tx_ipv4 = %d\n", 5345 buf->large_tx_ipv4); 5346 netdev_dbg(adapter->netdev, "large_tx_ipv6 = %d\n", 5347 buf->large_tx_ipv6); 5348 netdev_dbg(adapter->netdev, "large_rx_ipv4 = %d\n", 5349 buf->large_rx_ipv4); 5350 netdev_dbg(adapter->netdev, "large_rx_ipv6 = %d\n", 5351 buf->large_rx_ipv6); 5352 netdev_dbg(adapter->netdev, "max_ipv4_hdr_sz = %d\n", 5353 buf->max_ipv4_header_size); 5354 netdev_dbg(adapter->netdev, "max_ipv6_hdr_sz = %d\n", 5355 buf->max_ipv6_header_size); 5356 netdev_dbg(adapter->netdev, "max_tcp_hdr_size = %d\n", 5357 buf->max_tcp_header_size); 5358 netdev_dbg(adapter->netdev, "max_udp_hdr_size = %d\n", 5359 buf->max_udp_header_size); 5360 netdev_dbg(adapter->netdev, "max_large_tx_size = %d\n", 5361 buf->max_large_tx_size); 5362 netdev_dbg(adapter->netdev, "max_large_rx_size = %d\n", 5363 buf->max_large_rx_size); 5364 netdev_dbg(adapter->netdev, "ipv6_ext_hdr = %d\n", 5365 buf->ipv6_extension_header); 5366 netdev_dbg(adapter->netdev, "tcp_pseudosum_req = %d\n", 5367 buf->tcp_pseudosum_req); 5368 netdev_dbg(adapter->netdev, "num_ipv6_ext_hd = %d\n", 5369 buf->num_ipv6_ext_headers); 5370 netdev_dbg(adapter->netdev, "off_ipv6_ext_hd = %d\n", 5371 buf->off_ipv6_ext_headers); 5372 5373 send_control_ip_offload(adapter); 5374 } 5375 5376 static const char *ibmvnic_fw_err_cause(u16 cause) 5377 { 5378 switch (cause) { 5379 case ADAPTER_PROBLEM: 5380 return "adapter problem"; 5381 case BUS_PROBLEM: 5382 return "bus problem"; 5383 case FW_PROBLEM: 5384 return "firmware problem"; 5385 case DD_PROBLEM: 5386 return "device driver problem"; 5387 case EEH_RECOVERY: 5388 return "EEH recovery"; 5389 case FW_UPDATED: 5390 return "firmware updated"; 5391 case LOW_MEMORY: 5392 return "low Memory"; 5393 default: 5394 return "unknown"; 5395 } 5396 } 5397 5398 static void handle_error_indication(union ibmvnic_crq *crq, 5399 struct ibmvnic_adapter *adapter) 5400 { 5401 struct device *dev = &adapter->vdev->dev; 5402 u16 cause; 5403 5404 cause = be16_to_cpu(crq->error_indication.error_cause); 5405 5406 dev_warn_ratelimited(dev, 5407 "Firmware reports %serror, cause: %s. Starting recovery...\n", 5408 crq->error_indication.flags 5409 & IBMVNIC_FATAL_ERROR ? "FATAL " : "", 5410 ibmvnic_fw_err_cause(cause)); 5411 5412 if (crq->error_indication.flags & IBMVNIC_FATAL_ERROR) 5413 ibmvnic_reset(adapter, VNIC_RESET_FATAL); 5414 else 5415 ibmvnic_reset(adapter, VNIC_RESET_NON_FATAL); 5416 } 5417 5418 static int handle_change_mac_rsp(union ibmvnic_crq *crq, 5419 struct ibmvnic_adapter *adapter) 5420 { 5421 struct net_device *netdev = adapter->netdev; 5422 struct device *dev = &adapter->vdev->dev; 5423 long rc; 5424 5425 rc = crq->change_mac_addr_rsp.rc.code; 5426 if (rc) { 5427 dev_err(dev, "Error %ld in CHANGE_MAC_ADDR_RSP\n", rc); 5428 goto out; 5429 } 5430 /* crq->change_mac_addr.mac_addr is the requested one 5431 * crq->change_mac_addr_rsp.mac_addr is the returned valid one. 5432 */ 5433 eth_hw_addr_set(netdev, &crq->change_mac_addr_rsp.mac_addr[0]); 5434 ether_addr_copy(adapter->mac_addr, 5435 &crq->change_mac_addr_rsp.mac_addr[0]); 5436 out: 5437 complete(&adapter->fw_done); 5438 return rc; 5439 } 5440 5441 static void handle_request_cap_rsp(union ibmvnic_crq *crq, 5442 struct ibmvnic_adapter *adapter) 5443 { 5444 struct device *dev = &adapter->vdev->dev; 5445 u64 *req_value; 5446 char *name; 5447 5448 atomic_dec(&adapter->running_cap_crqs); 5449 netdev_dbg(adapter->netdev, "Outstanding request-caps: %d\n", 5450 atomic_read(&adapter->running_cap_crqs)); 5451 switch (be16_to_cpu(crq->request_capability_rsp.capability)) { 5452 case REQ_TX_QUEUES: 5453 req_value = &adapter->req_tx_queues; 5454 name = "tx"; 5455 break; 5456 case REQ_RX_QUEUES: 5457 req_value = &adapter->req_rx_queues; 5458 name = "rx"; 5459 break; 5460 case REQ_RX_ADD_QUEUES: 5461 req_value = &adapter->req_rx_add_queues; 5462 name = "rx_add"; 5463 break; 5464 case REQ_TX_ENTRIES_PER_SUBCRQ: 5465 req_value = &adapter->req_tx_entries_per_subcrq; 5466 name = "tx_entries_per_subcrq"; 5467 break; 5468 case REQ_RX_ADD_ENTRIES_PER_SUBCRQ: 5469 req_value = &adapter->req_rx_add_entries_per_subcrq; 5470 name = "rx_add_entries_per_subcrq"; 5471 break; 5472 case REQ_MTU: 5473 req_value = &adapter->req_mtu; 5474 name = "mtu"; 5475 break; 5476 case PROMISC_REQUESTED: 5477 req_value = &adapter->promisc; 5478 name = "promisc"; 5479 break; 5480 default: 5481 dev_err(dev, "Got invalid cap request rsp %d\n", 5482 crq->request_capability.capability); 5483 return; 5484 } 5485 5486 switch (crq->request_capability_rsp.rc.code) { 5487 case SUCCESS: 5488 break; 5489 case PARTIALSUCCESS: 5490 dev_info(dev, "req=%lld, rsp=%ld in %s queue, retrying.\n", 5491 *req_value, 5492 (long)be64_to_cpu(crq->request_capability_rsp.number), 5493 name); 5494 5495 if (be16_to_cpu(crq->request_capability_rsp.capability) == 5496 REQ_MTU) { 5497 pr_err("mtu of %llu is not supported. Reverting.\n", 5498 *req_value); 5499 *req_value = adapter->fallback.mtu; 5500 } else { 5501 *req_value = 5502 be64_to_cpu(crq->request_capability_rsp.number); 5503 } 5504 5505 send_request_cap(adapter, 1); 5506 return; 5507 default: 5508 dev_err(dev, "Error %d in request cap rsp\n", 5509 crq->request_capability_rsp.rc.code); 5510 return; 5511 } 5512 5513 /* Done receiving requested capabilities, query IP offload support */ 5514 if (atomic_read(&adapter->running_cap_crqs) == 0) 5515 send_query_ip_offload(adapter); 5516 } 5517 5518 static int handle_login_rsp(union ibmvnic_crq *login_rsp_crq, 5519 struct ibmvnic_adapter *adapter) 5520 { 5521 struct device *dev = &adapter->vdev->dev; 5522 struct net_device *netdev = adapter->netdev; 5523 struct ibmvnic_login_rsp_buffer *login_rsp = adapter->login_rsp_buf; 5524 struct ibmvnic_login_buffer *login = adapter->login_buf; 5525 u64 *tx_handle_array; 5526 u64 *rx_handle_array; 5527 int num_tx_pools; 5528 int num_rx_pools; 5529 u64 *size_array; 5530 u32 rsp_len; 5531 int i; 5532 5533 /* CHECK: Test/set of login_pending does not need to be atomic 5534 * because only ibmvnic_tasklet tests/clears this. 5535 */ 5536 if (!adapter->login_pending) { 5537 netdev_warn(netdev, "Ignoring unexpected login response\n"); 5538 return 0; 5539 } 5540 adapter->login_pending = false; 5541 5542 /* If the number of queues requested can't be allocated by the 5543 * server, the login response will return with code 1. We will need 5544 * to resend the login buffer with fewer queues requested. 5545 */ 5546 if (login_rsp_crq->generic.rc.code) { 5547 adapter->init_done_rc = login_rsp_crq->generic.rc.code; 5548 complete(&adapter->init_done); 5549 return 0; 5550 } 5551 5552 if (adapter->failover_pending) { 5553 adapter->init_done_rc = -EAGAIN; 5554 netdev_dbg(netdev, "Failover pending, ignoring login response\n"); 5555 complete(&adapter->init_done); 5556 /* login response buffer will be released on reset */ 5557 return 0; 5558 } 5559 5560 netdev->mtu = adapter->req_mtu - ETH_HLEN; 5561 5562 netdev_dbg(adapter->netdev, "Login Response Buffer:\n"); 5563 for (i = 0; i < (adapter->login_rsp_buf_sz - 1) / 8 + 1; i++) { 5564 netdev_dbg(adapter->netdev, "%016lx\n", 5565 ((unsigned long *)(adapter->login_rsp_buf))[i]); 5566 } 5567 5568 /* Sanity checks */ 5569 if (login->num_txcomp_subcrqs != login_rsp->num_txsubm_subcrqs || 5570 (be32_to_cpu(login->num_rxcomp_subcrqs) * 5571 adapter->req_rx_add_queues != 5572 be32_to_cpu(login_rsp->num_rxadd_subcrqs))) { 5573 dev_err(dev, "FATAL: Inconsistent login and login rsp\n"); 5574 ibmvnic_reset(adapter, VNIC_RESET_FATAL); 5575 return -EIO; 5576 } 5577 5578 rsp_len = be32_to_cpu(login_rsp->len); 5579 if (be32_to_cpu(login->login_rsp_len) < rsp_len || 5580 rsp_len <= be32_to_cpu(login_rsp->off_txsubm_subcrqs) || 5581 rsp_len <= be32_to_cpu(login_rsp->off_rxadd_subcrqs) || 5582 rsp_len <= be32_to_cpu(login_rsp->off_rxadd_buff_size) || 5583 rsp_len <= be32_to_cpu(login_rsp->off_supp_tx_desc)) { 5584 /* This can happen if a login request times out and there are 5585 * 2 outstanding login requests sent, the LOGIN_RSP crq 5586 * could have been for the older login request. So we are 5587 * parsing the newer response buffer which may be incomplete 5588 */ 5589 dev_err(dev, "FATAL: Login rsp offsets/lengths invalid\n"); 5590 ibmvnic_reset(adapter, VNIC_RESET_FATAL); 5591 return -EIO; 5592 } 5593 5594 size_array = (u64 *)((u8 *)(adapter->login_rsp_buf) + 5595 be32_to_cpu(adapter->login_rsp_buf->off_rxadd_buff_size)); 5596 /* variable buffer sizes are not supported, so just read the 5597 * first entry. 5598 */ 5599 adapter->cur_rx_buf_sz = be64_to_cpu(size_array[0]); 5600 5601 num_tx_pools = be32_to_cpu(adapter->login_rsp_buf->num_txsubm_subcrqs); 5602 num_rx_pools = be32_to_cpu(adapter->login_rsp_buf->num_rxadd_subcrqs); 5603 5604 tx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) + 5605 be32_to_cpu(adapter->login_rsp_buf->off_txsubm_subcrqs)); 5606 rx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) + 5607 be32_to_cpu(adapter->login_rsp_buf->off_rxadd_subcrqs)); 5608 5609 for (i = 0; i < num_tx_pools; i++) 5610 adapter->tx_scrq[i]->handle = tx_handle_array[i]; 5611 5612 for (i = 0; i < num_rx_pools; i++) 5613 adapter->rx_scrq[i]->handle = rx_handle_array[i]; 5614 5615 adapter->num_active_tx_scrqs = num_tx_pools; 5616 adapter->num_active_rx_scrqs = num_rx_pools; 5617 release_login_rsp_buffer(adapter); 5618 release_login_buffer(adapter); 5619 complete(&adapter->init_done); 5620 5621 return 0; 5622 } 5623 5624 static void handle_request_unmap_rsp(union ibmvnic_crq *crq, 5625 struct ibmvnic_adapter *adapter) 5626 { 5627 struct device *dev = &adapter->vdev->dev; 5628 long rc; 5629 5630 rc = crq->request_unmap_rsp.rc.code; 5631 if (rc) 5632 dev_err(dev, "Error %ld in REQUEST_UNMAP_RSP\n", rc); 5633 } 5634 5635 static void handle_query_map_rsp(union ibmvnic_crq *crq, 5636 struct ibmvnic_adapter *adapter) 5637 { 5638 struct net_device *netdev = adapter->netdev; 5639 struct device *dev = &adapter->vdev->dev; 5640 long rc; 5641 5642 rc = crq->query_map_rsp.rc.code; 5643 if (rc) { 5644 dev_err(dev, "Error %ld in QUERY_MAP_RSP\n", rc); 5645 return; 5646 } 5647 netdev_dbg(netdev, "page_size = %d\ntot_pages = %u\nfree_pages = %u\n", 5648 crq->query_map_rsp.page_size, 5649 __be32_to_cpu(crq->query_map_rsp.tot_pages), 5650 __be32_to_cpu(crq->query_map_rsp.free_pages)); 5651 } 5652 5653 static void handle_query_cap_rsp(union ibmvnic_crq *crq, 5654 struct ibmvnic_adapter *adapter) 5655 { 5656 struct net_device *netdev = adapter->netdev; 5657 struct device *dev = &adapter->vdev->dev; 5658 long rc; 5659 5660 atomic_dec(&adapter->running_cap_crqs); 5661 netdev_dbg(netdev, "Outstanding queries: %d\n", 5662 atomic_read(&adapter->running_cap_crqs)); 5663 rc = crq->query_capability.rc.code; 5664 if (rc) { 5665 dev_err(dev, "Error %ld in QUERY_CAP_RSP\n", rc); 5666 goto out; 5667 } 5668 5669 switch (be16_to_cpu(crq->query_capability.capability)) { 5670 case MIN_TX_QUEUES: 5671 adapter->min_tx_queues = 5672 be64_to_cpu(crq->query_capability.number); 5673 netdev_dbg(netdev, "min_tx_queues = %lld\n", 5674 adapter->min_tx_queues); 5675 break; 5676 case MIN_RX_QUEUES: 5677 adapter->min_rx_queues = 5678 be64_to_cpu(crq->query_capability.number); 5679 netdev_dbg(netdev, "min_rx_queues = %lld\n", 5680 adapter->min_rx_queues); 5681 break; 5682 case MIN_RX_ADD_QUEUES: 5683 adapter->min_rx_add_queues = 5684 be64_to_cpu(crq->query_capability.number); 5685 netdev_dbg(netdev, "min_rx_add_queues = %lld\n", 5686 adapter->min_rx_add_queues); 5687 break; 5688 case MAX_TX_QUEUES: 5689 adapter->max_tx_queues = 5690 be64_to_cpu(crq->query_capability.number); 5691 netdev_dbg(netdev, "max_tx_queues = %lld\n", 5692 adapter->max_tx_queues); 5693 break; 5694 case MAX_RX_QUEUES: 5695 adapter->max_rx_queues = 5696 be64_to_cpu(crq->query_capability.number); 5697 netdev_dbg(netdev, "max_rx_queues = %lld\n", 5698 adapter->max_rx_queues); 5699 break; 5700 case MAX_RX_ADD_QUEUES: 5701 adapter->max_rx_add_queues = 5702 be64_to_cpu(crq->query_capability.number); 5703 netdev_dbg(netdev, "max_rx_add_queues = %lld\n", 5704 adapter->max_rx_add_queues); 5705 break; 5706 case MIN_TX_ENTRIES_PER_SUBCRQ: 5707 adapter->min_tx_entries_per_subcrq = 5708 be64_to_cpu(crq->query_capability.number); 5709 netdev_dbg(netdev, "min_tx_entries_per_subcrq = %lld\n", 5710 adapter->min_tx_entries_per_subcrq); 5711 break; 5712 case MIN_RX_ADD_ENTRIES_PER_SUBCRQ: 5713 adapter->min_rx_add_entries_per_subcrq = 5714 be64_to_cpu(crq->query_capability.number); 5715 netdev_dbg(netdev, "min_rx_add_entrs_per_subcrq = %lld\n", 5716 adapter->min_rx_add_entries_per_subcrq); 5717 break; 5718 case MAX_TX_ENTRIES_PER_SUBCRQ: 5719 adapter->max_tx_entries_per_subcrq = 5720 be64_to_cpu(crq->query_capability.number); 5721 netdev_dbg(netdev, "max_tx_entries_per_subcrq = %lld\n", 5722 adapter->max_tx_entries_per_subcrq); 5723 break; 5724 case MAX_RX_ADD_ENTRIES_PER_SUBCRQ: 5725 adapter->max_rx_add_entries_per_subcrq = 5726 be64_to_cpu(crq->query_capability.number); 5727 netdev_dbg(netdev, "max_rx_add_entrs_per_subcrq = %lld\n", 5728 adapter->max_rx_add_entries_per_subcrq); 5729 break; 5730 case TCP_IP_OFFLOAD: 5731 adapter->tcp_ip_offload = 5732 be64_to_cpu(crq->query_capability.number); 5733 netdev_dbg(netdev, "tcp_ip_offload = %lld\n", 5734 adapter->tcp_ip_offload); 5735 break; 5736 case PROMISC_SUPPORTED: 5737 adapter->promisc_supported = 5738 be64_to_cpu(crq->query_capability.number); 5739 netdev_dbg(netdev, "promisc_supported = %lld\n", 5740 adapter->promisc_supported); 5741 break; 5742 case MIN_MTU: 5743 adapter->min_mtu = be64_to_cpu(crq->query_capability.number); 5744 netdev->min_mtu = adapter->min_mtu - ETH_HLEN; 5745 netdev_dbg(netdev, "min_mtu = %lld\n", adapter->min_mtu); 5746 break; 5747 case MAX_MTU: 5748 adapter->max_mtu = be64_to_cpu(crq->query_capability.number); 5749 netdev->max_mtu = adapter->max_mtu - ETH_HLEN; 5750 netdev_dbg(netdev, "max_mtu = %lld\n", adapter->max_mtu); 5751 break; 5752 case MAX_MULTICAST_FILTERS: 5753 adapter->max_multicast_filters = 5754 be64_to_cpu(crq->query_capability.number); 5755 netdev_dbg(netdev, "max_multicast_filters = %lld\n", 5756 adapter->max_multicast_filters); 5757 break; 5758 case VLAN_HEADER_INSERTION: 5759 adapter->vlan_header_insertion = 5760 be64_to_cpu(crq->query_capability.number); 5761 if (adapter->vlan_header_insertion) 5762 netdev->features |= NETIF_F_HW_VLAN_STAG_TX; 5763 netdev_dbg(netdev, "vlan_header_insertion = %lld\n", 5764 adapter->vlan_header_insertion); 5765 break; 5766 case RX_VLAN_HEADER_INSERTION: 5767 adapter->rx_vlan_header_insertion = 5768 be64_to_cpu(crq->query_capability.number); 5769 netdev_dbg(netdev, "rx_vlan_header_insertion = %lld\n", 5770 adapter->rx_vlan_header_insertion); 5771 break; 5772 case MAX_TX_SG_ENTRIES: 5773 adapter->max_tx_sg_entries = 5774 be64_to_cpu(crq->query_capability.number); 5775 netdev_dbg(netdev, "max_tx_sg_entries = %lld\n", 5776 adapter->max_tx_sg_entries); 5777 break; 5778 case RX_SG_SUPPORTED: 5779 adapter->rx_sg_supported = 5780 be64_to_cpu(crq->query_capability.number); 5781 netdev_dbg(netdev, "rx_sg_supported = %lld\n", 5782 adapter->rx_sg_supported); 5783 break; 5784 case OPT_TX_COMP_SUB_QUEUES: 5785 adapter->opt_tx_comp_sub_queues = 5786 be64_to_cpu(crq->query_capability.number); 5787 netdev_dbg(netdev, "opt_tx_comp_sub_queues = %lld\n", 5788 adapter->opt_tx_comp_sub_queues); 5789 break; 5790 case OPT_RX_COMP_QUEUES: 5791 adapter->opt_rx_comp_queues = 5792 be64_to_cpu(crq->query_capability.number); 5793 netdev_dbg(netdev, "opt_rx_comp_queues = %lld\n", 5794 adapter->opt_rx_comp_queues); 5795 break; 5796 case OPT_RX_BUFADD_Q_PER_RX_COMP_Q: 5797 adapter->opt_rx_bufadd_q_per_rx_comp_q = 5798 be64_to_cpu(crq->query_capability.number); 5799 netdev_dbg(netdev, "opt_rx_bufadd_q_per_rx_comp_q = %lld\n", 5800 adapter->opt_rx_bufadd_q_per_rx_comp_q); 5801 break; 5802 case OPT_TX_ENTRIES_PER_SUBCRQ: 5803 adapter->opt_tx_entries_per_subcrq = 5804 be64_to_cpu(crq->query_capability.number); 5805 netdev_dbg(netdev, "opt_tx_entries_per_subcrq = %lld\n", 5806 adapter->opt_tx_entries_per_subcrq); 5807 break; 5808 case OPT_RXBA_ENTRIES_PER_SUBCRQ: 5809 adapter->opt_rxba_entries_per_subcrq = 5810 be64_to_cpu(crq->query_capability.number); 5811 netdev_dbg(netdev, "opt_rxba_entries_per_subcrq = %lld\n", 5812 adapter->opt_rxba_entries_per_subcrq); 5813 break; 5814 case TX_RX_DESC_REQ: 5815 adapter->tx_rx_desc_req = crq->query_capability.number; 5816 netdev_dbg(netdev, "tx_rx_desc_req = %llx\n", 5817 adapter->tx_rx_desc_req); 5818 break; 5819 5820 default: 5821 netdev_err(netdev, "Got invalid cap rsp %d\n", 5822 crq->query_capability.capability); 5823 } 5824 5825 out: 5826 if (atomic_read(&adapter->running_cap_crqs) == 0) 5827 send_request_cap(adapter, 0); 5828 } 5829 5830 static int send_query_phys_parms(struct ibmvnic_adapter *adapter) 5831 { 5832 union ibmvnic_crq crq; 5833 int rc; 5834 5835 memset(&crq, 0, sizeof(crq)); 5836 crq.query_phys_parms.first = IBMVNIC_CRQ_CMD; 5837 crq.query_phys_parms.cmd = QUERY_PHYS_PARMS; 5838 5839 mutex_lock(&adapter->fw_lock); 5840 adapter->fw_done_rc = 0; 5841 reinit_completion(&adapter->fw_done); 5842 5843 rc = ibmvnic_send_crq(adapter, &crq); 5844 if (rc) { 5845 mutex_unlock(&adapter->fw_lock); 5846 return rc; 5847 } 5848 5849 rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000); 5850 if (rc) { 5851 mutex_unlock(&adapter->fw_lock); 5852 return rc; 5853 } 5854 5855 mutex_unlock(&adapter->fw_lock); 5856 return adapter->fw_done_rc ? -EIO : 0; 5857 } 5858 5859 static int handle_query_phys_parms_rsp(union ibmvnic_crq *crq, 5860 struct ibmvnic_adapter *adapter) 5861 { 5862 struct net_device *netdev = adapter->netdev; 5863 int rc; 5864 __be32 rspeed = cpu_to_be32(crq->query_phys_parms_rsp.speed); 5865 5866 rc = crq->query_phys_parms_rsp.rc.code; 5867 if (rc) { 5868 netdev_err(netdev, "Error %d in QUERY_PHYS_PARMS\n", rc); 5869 return rc; 5870 } 5871 switch (rspeed) { 5872 case IBMVNIC_10MBPS: 5873 adapter->speed = SPEED_10; 5874 break; 5875 case IBMVNIC_100MBPS: 5876 adapter->speed = SPEED_100; 5877 break; 5878 case IBMVNIC_1GBPS: 5879 adapter->speed = SPEED_1000; 5880 break; 5881 case IBMVNIC_10GBPS: 5882 adapter->speed = SPEED_10000; 5883 break; 5884 case IBMVNIC_25GBPS: 5885 adapter->speed = SPEED_25000; 5886 break; 5887 case IBMVNIC_40GBPS: 5888 adapter->speed = SPEED_40000; 5889 break; 5890 case IBMVNIC_50GBPS: 5891 adapter->speed = SPEED_50000; 5892 break; 5893 case IBMVNIC_100GBPS: 5894 adapter->speed = SPEED_100000; 5895 break; 5896 case IBMVNIC_200GBPS: 5897 adapter->speed = SPEED_200000; 5898 break; 5899 default: 5900 if (netif_carrier_ok(netdev)) 5901 netdev_warn(netdev, "Unknown speed 0x%08x\n", rspeed); 5902 adapter->speed = SPEED_UNKNOWN; 5903 } 5904 if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_FULL_DUPLEX) 5905 adapter->duplex = DUPLEX_FULL; 5906 else if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_HALF_DUPLEX) 5907 adapter->duplex = DUPLEX_HALF; 5908 else 5909 adapter->duplex = DUPLEX_UNKNOWN; 5910 5911 return rc; 5912 } 5913 5914 static void ibmvnic_handle_crq(union ibmvnic_crq *crq, 5915 struct ibmvnic_adapter *adapter) 5916 { 5917 struct ibmvnic_generic_crq *gen_crq = &crq->generic; 5918 struct net_device *netdev = adapter->netdev; 5919 struct device *dev = &adapter->vdev->dev; 5920 u64 *u64_crq = (u64 *)crq; 5921 long rc; 5922 5923 netdev_dbg(netdev, "Handling CRQ: %016lx %016lx\n", 5924 (unsigned long)cpu_to_be64(u64_crq[0]), 5925 (unsigned long)cpu_to_be64(u64_crq[1])); 5926 switch (gen_crq->first) { 5927 case IBMVNIC_CRQ_INIT_RSP: 5928 switch (gen_crq->cmd) { 5929 case IBMVNIC_CRQ_INIT: 5930 dev_info(dev, "Partner initialized\n"); 5931 adapter->from_passive_init = true; 5932 /* Discard any stale login responses from prev reset. 5933 * CHECK: should we clear even on INIT_COMPLETE? 5934 */ 5935 adapter->login_pending = false; 5936 5937 if (adapter->state == VNIC_DOWN) 5938 rc = ibmvnic_reset(adapter, VNIC_RESET_PASSIVE_INIT); 5939 else 5940 rc = ibmvnic_reset(adapter, VNIC_RESET_FAILOVER); 5941 5942 if (rc && rc != -EBUSY) { 5943 /* We were unable to schedule the failover 5944 * reset either because the adapter was still 5945 * probing (eg: during kexec) or we could not 5946 * allocate memory. Clear the failover_pending 5947 * flag since no one else will. We ignore 5948 * EBUSY because it means either FAILOVER reset 5949 * is already scheduled or the adapter is 5950 * being removed. 5951 */ 5952 netdev_err(netdev, 5953 "Error %ld scheduling failover reset\n", 5954 rc); 5955 adapter->failover_pending = false; 5956 } 5957 5958 if (!completion_done(&adapter->init_done)) { 5959 if (!adapter->init_done_rc) 5960 adapter->init_done_rc = -EAGAIN; 5961 complete(&adapter->init_done); 5962 } 5963 5964 break; 5965 case IBMVNIC_CRQ_INIT_COMPLETE: 5966 dev_info(dev, "Partner initialization complete\n"); 5967 adapter->crq.active = true; 5968 send_version_xchg(adapter); 5969 break; 5970 default: 5971 dev_err(dev, "Unknown crq cmd: %d\n", gen_crq->cmd); 5972 } 5973 return; 5974 case IBMVNIC_CRQ_XPORT_EVENT: 5975 netif_carrier_off(netdev); 5976 adapter->crq.active = false; 5977 /* terminate any thread waiting for a response 5978 * from the device 5979 */ 5980 if (!completion_done(&adapter->fw_done)) { 5981 adapter->fw_done_rc = -EIO; 5982 complete(&adapter->fw_done); 5983 } 5984 5985 /* if we got here during crq-init, retry crq-init */ 5986 if (!completion_done(&adapter->init_done)) { 5987 adapter->init_done_rc = -EAGAIN; 5988 complete(&adapter->init_done); 5989 } 5990 5991 if (!completion_done(&adapter->stats_done)) 5992 complete(&adapter->stats_done); 5993 if (test_bit(0, &adapter->resetting)) 5994 adapter->force_reset_recovery = true; 5995 if (gen_crq->cmd == IBMVNIC_PARTITION_MIGRATED) { 5996 dev_info(dev, "Migrated, re-enabling adapter\n"); 5997 ibmvnic_reset(adapter, VNIC_RESET_MOBILITY); 5998 } else if (gen_crq->cmd == IBMVNIC_DEVICE_FAILOVER) { 5999 dev_info(dev, "Backing device failover detected\n"); 6000 adapter->failover_pending = true; 6001 } else { 6002 /* The adapter lost the connection */ 6003 dev_err(dev, "Virtual Adapter failed (rc=%d)\n", 6004 gen_crq->cmd); 6005 ibmvnic_reset(adapter, VNIC_RESET_FATAL); 6006 } 6007 return; 6008 case IBMVNIC_CRQ_CMD_RSP: 6009 break; 6010 default: 6011 dev_err(dev, "Got an invalid msg type 0x%02x\n", 6012 gen_crq->first); 6013 return; 6014 } 6015 6016 switch (gen_crq->cmd) { 6017 case VERSION_EXCHANGE_RSP: 6018 rc = crq->version_exchange_rsp.rc.code; 6019 if (rc) { 6020 dev_err(dev, "Error %ld in VERSION_EXCHG_RSP\n", rc); 6021 break; 6022 } 6023 ibmvnic_version = 6024 be16_to_cpu(crq->version_exchange_rsp.version); 6025 dev_info(dev, "Partner protocol version is %d\n", 6026 ibmvnic_version); 6027 send_query_cap(adapter); 6028 break; 6029 case QUERY_CAPABILITY_RSP: 6030 handle_query_cap_rsp(crq, adapter); 6031 break; 6032 case QUERY_MAP_RSP: 6033 handle_query_map_rsp(crq, adapter); 6034 break; 6035 case REQUEST_MAP_RSP: 6036 adapter->fw_done_rc = crq->request_map_rsp.rc.code; 6037 complete(&adapter->fw_done); 6038 break; 6039 case REQUEST_UNMAP_RSP: 6040 handle_request_unmap_rsp(crq, adapter); 6041 break; 6042 case REQUEST_CAPABILITY_RSP: 6043 handle_request_cap_rsp(crq, adapter); 6044 break; 6045 case LOGIN_RSP: 6046 netdev_dbg(netdev, "Got Login Response\n"); 6047 handle_login_rsp(crq, adapter); 6048 break; 6049 case LOGICAL_LINK_STATE_RSP: 6050 netdev_dbg(netdev, 6051 "Got Logical Link State Response, state: %d rc: %d\n", 6052 crq->logical_link_state_rsp.link_state, 6053 crq->logical_link_state_rsp.rc.code); 6054 adapter->logical_link_state = 6055 crq->logical_link_state_rsp.link_state; 6056 adapter->init_done_rc = crq->logical_link_state_rsp.rc.code; 6057 complete(&adapter->init_done); 6058 break; 6059 case LINK_STATE_INDICATION: 6060 netdev_dbg(netdev, "Got Logical Link State Indication\n"); 6061 adapter->phys_link_state = 6062 crq->link_state_indication.phys_link_state; 6063 adapter->logical_link_state = 6064 crq->link_state_indication.logical_link_state; 6065 if (adapter->phys_link_state && adapter->logical_link_state) 6066 netif_carrier_on(netdev); 6067 else 6068 netif_carrier_off(netdev); 6069 break; 6070 case CHANGE_MAC_ADDR_RSP: 6071 netdev_dbg(netdev, "Got MAC address change Response\n"); 6072 adapter->fw_done_rc = handle_change_mac_rsp(crq, adapter); 6073 break; 6074 case ERROR_INDICATION: 6075 netdev_dbg(netdev, "Got Error Indication\n"); 6076 handle_error_indication(crq, adapter); 6077 break; 6078 case REQUEST_STATISTICS_RSP: 6079 netdev_dbg(netdev, "Got Statistics Response\n"); 6080 complete(&adapter->stats_done); 6081 break; 6082 case QUERY_IP_OFFLOAD_RSP: 6083 netdev_dbg(netdev, "Got Query IP offload Response\n"); 6084 handle_query_ip_offload_rsp(adapter); 6085 break; 6086 case MULTICAST_CTRL_RSP: 6087 netdev_dbg(netdev, "Got multicast control Response\n"); 6088 break; 6089 case CONTROL_IP_OFFLOAD_RSP: 6090 netdev_dbg(netdev, "Got Control IP offload Response\n"); 6091 dma_unmap_single(dev, adapter->ip_offload_ctrl_tok, 6092 sizeof(adapter->ip_offload_ctrl), 6093 DMA_TO_DEVICE); 6094 complete(&adapter->init_done); 6095 break; 6096 case COLLECT_FW_TRACE_RSP: 6097 netdev_dbg(netdev, "Got Collect firmware trace Response\n"); 6098 complete(&adapter->fw_done); 6099 break; 6100 case GET_VPD_SIZE_RSP: 6101 handle_vpd_size_rsp(crq, adapter); 6102 break; 6103 case GET_VPD_RSP: 6104 handle_vpd_rsp(crq, adapter); 6105 break; 6106 case QUERY_PHYS_PARMS_RSP: 6107 adapter->fw_done_rc = handle_query_phys_parms_rsp(crq, adapter); 6108 complete(&adapter->fw_done); 6109 break; 6110 default: 6111 netdev_err(netdev, "Got an invalid cmd type 0x%02x\n", 6112 gen_crq->cmd); 6113 } 6114 } 6115 6116 static irqreturn_t ibmvnic_interrupt(int irq, void *instance) 6117 { 6118 struct ibmvnic_adapter *adapter = instance; 6119 6120 tasklet_schedule(&adapter->tasklet); 6121 return IRQ_HANDLED; 6122 } 6123 6124 static void ibmvnic_tasklet(struct tasklet_struct *t) 6125 { 6126 struct ibmvnic_adapter *adapter = from_tasklet(adapter, t, tasklet); 6127 struct ibmvnic_crq_queue *queue = &adapter->crq; 6128 union ibmvnic_crq *crq; 6129 unsigned long flags; 6130 6131 spin_lock_irqsave(&queue->lock, flags); 6132 6133 /* Pull all the valid messages off the CRQ */ 6134 while ((crq = ibmvnic_next_crq(adapter)) != NULL) { 6135 /* This barrier makes sure ibmvnic_next_crq()'s 6136 * crq->generic.first & IBMVNIC_CRQ_CMD_RSP is loaded 6137 * before ibmvnic_handle_crq()'s 6138 * switch(gen_crq->first) and switch(gen_crq->cmd). 6139 */ 6140 dma_rmb(); 6141 ibmvnic_handle_crq(crq, adapter); 6142 crq->generic.first = 0; 6143 } 6144 6145 spin_unlock_irqrestore(&queue->lock, flags); 6146 } 6147 6148 static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *adapter) 6149 { 6150 struct vio_dev *vdev = adapter->vdev; 6151 int rc; 6152 6153 do { 6154 rc = plpar_hcall_norets(H_ENABLE_CRQ, vdev->unit_address); 6155 } while (rc == H_IN_PROGRESS || rc == H_BUSY || H_IS_LONG_BUSY(rc)); 6156 6157 if (rc) 6158 dev_err(&vdev->dev, "Error enabling adapter (rc=%d)\n", rc); 6159 6160 return rc; 6161 } 6162 6163 static int ibmvnic_reset_crq(struct ibmvnic_adapter *adapter) 6164 { 6165 struct ibmvnic_crq_queue *crq = &adapter->crq; 6166 struct device *dev = &adapter->vdev->dev; 6167 struct vio_dev *vdev = adapter->vdev; 6168 int rc; 6169 6170 /* Close the CRQ */ 6171 do { 6172 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address); 6173 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc)); 6174 6175 /* Clean out the queue */ 6176 if (!crq->msgs) 6177 return -EINVAL; 6178 6179 memset(crq->msgs, 0, PAGE_SIZE); 6180 crq->cur = 0; 6181 crq->active = false; 6182 6183 /* And re-open it again */ 6184 rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address, 6185 crq->msg_token, PAGE_SIZE); 6186 6187 if (rc == H_CLOSED) 6188 /* Adapter is good, but other end is not ready */ 6189 dev_warn(dev, "Partner adapter not ready\n"); 6190 else if (rc != 0) 6191 dev_warn(dev, "Couldn't register crq (rc=%d)\n", rc); 6192 6193 return rc; 6194 } 6195 6196 static void release_crq_queue(struct ibmvnic_adapter *adapter) 6197 { 6198 struct ibmvnic_crq_queue *crq = &adapter->crq; 6199 struct vio_dev *vdev = adapter->vdev; 6200 long rc; 6201 6202 if (!crq->msgs) 6203 return; 6204 6205 netdev_dbg(adapter->netdev, "Releasing CRQ\n"); 6206 free_irq(vdev->irq, adapter); 6207 tasklet_kill(&adapter->tasklet); 6208 do { 6209 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address); 6210 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc)); 6211 6212 dma_unmap_single(&vdev->dev, crq->msg_token, PAGE_SIZE, 6213 DMA_BIDIRECTIONAL); 6214 free_page((unsigned long)crq->msgs); 6215 crq->msgs = NULL; 6216 crq->active = false; 6217 } 6218 6219 static int init_crq_queue(struct ibmvnic_adapter *adapter) 6220 { 6221 struct ibmvnic_crq_queue *crq = &adapter->crq; 6222 struct device *dev = &adapter->vdev->dev; 6223 struct vio_dev *vdev = adapter->vdev; 6224 int rc, retrc = -ENOMEM; 6225 6226 if (crq->msgs) 6227 return 0; 6228 6229 crq->msgs = (union ibmvnic_crq *)get_zeroed_page(GFP_KERNEL); 6230 /* Should we allocate more than one page? */ 6231 6232 if (!crq->msgs) 6233 return -ENOMEM; 6234 6235 crq->size = PAGE_SIZE / sizeof(*crq->msgs); 6236 crq->msg_token = dma_map_single(dev, crq->msgs, PAGE_SIZE, 6237 DMA_BIDIRECTIONAL); 6238 if (dma_mapping_error(dev, crq->msg_token)) 6239 goto map_failed; 6240 6241 rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address, 6242 crq->msg_token, PAGE_SIZE); 6243 6244 if (rc == H_RESOURCE) 6245 /* maybe kexecing and resource is busy. try a reset */ 6246 rc = ibmvnic_reset_crq(adapter); 6247 retrc = rc; 6248 6249 if (rc == H_CLOSED) { 6250 dev_warn(dev, "Partner adapter not ready\n"); 6251 } else if (rc) { 6252 dev_warn(dev, "Error %d opening adapter\n", rc); 6253 goto reg_crq_failed; 6254 } 6255 6256 retrc = 0; 6257 6258 tasklet_setup(&adapter->tasklet, (void *)ibmvnic_tasklet); 6259 6260 netdev_dbg(adapter->netdev, "registering irq 0x%x\n", vdev->irq); 6261 snprintf(crq->name, sizeof(crq->name), "ibmvnic-%x", 6262 adapter->vdev->unit_address); 6263 rc = request_irq(vdev->irq, ibmvnic_interrupt, 0, crq->name, adapter); 6264 if (rc) { 6265 dev_err(dev, "Couldn't register irq 0x%x. rc=%d\n", 6266 vdev->irq, rc); 6267 goto req_irq_failed; 6268 } 6269 6270 rc = vio_enable_interrupts(vdev); 6271 if (rc) { 6272 dev_err(dev, "Error %d enabling interrupts\n", rc); 6273 goto req_irq_failed; 6274 } 6275 6276 crq->cur = 0; 6277 spin_lock_init(&crq->lock); 6278 6279 /* process any CRQs that were queued before we enabled interrupts */ 6280 tasklet_schedule(&adapter->tasklet); 6281 6282 return retrc; 6283 6284 req_irq_failed: 6285 tasklet_kill(&adapter->tasklet); 6286 do { 6287 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address); 6288 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc)); 6289 reg_crq_failed: 6290 dma_unmap_single(dev, crq->msg_token, PAGE_SIZE, DMA_BIDIRECTIONAL); 6291 map_failed: 6292 free_page((unsigned long)crq->msgs); 6293 crq->msgs = NULL; 6294 return retrc; 6295 } 6296 6297 static int ibmvnic_reset_init(struct ibmvnic_adapter *adapter, bool reset) 6298 { 6299 struct device *dev = &adapter->vdev->dev; 6300 unsigned long timeout = msecs_to_jiffies(20000); 6301 u64 old_num_rx_queues = adapter->req_rx_queues; 6302 u64 old_num_tx_queues = adapter->req_tx_queues; 6303 int rc; 6304 6305 adapter->from_passive_init = false; 6306 6307 rc = ibmvnic_send_crq_init(adapter); 6308 if (rc) { 6309 dev_err(dev, "Send crq init failed with error %d\n", rc); 6310 return rc; 6311 } 6312 6313 if (!wait_for_completion_timeout(&adapter->init_done, timeout)) { 6314 dev_err(dev, "Initialization sequence timed out\n"); 6315 return -ETIMEDOUT; 6316 } 6317 6318 if (adapter->init_done_rc) { 6319 release_crq_queue(adapter); 6320 dev_err(dev, "CRQ-init failed, %d\n", adapter->init_done_rc); 6321 return adapter->init_done_rc; 6322 } 6323 6324 if (adapter->from_passive_init) { 6325 adapter->state = VNIC_OPEN; 6326 adapter->from_passive_init = false; 6327 dev_err(dev, "CRQ-init failed, passive-init\n"); 6328 return -EINVAL; 6329 } 6330 6331 if (reset && 6332 test_bit(0, &adapter->resetting) && !adapter->wait_for_reset && 6333 adapter->reset_reason != VNIC_RESET_MOBILITY) { 6334 if (adapter->req_rx_queues != old_num_rx_queues || 6335 adapter->req_tx_queues != old_num_tx_queues) { 6336 release_sub_crqs(adapter, 0); 6337 rc = init_sub_crqs(adapter); 6338 } else { 6339 /* no need to reinitialize completely, but we do 6340 * need to clean up transmits that were in flight 6341 * when we processed the reset. Failure to do so 6342 * will confound the upper layer, usually TCP, by 6343 * creating the illusion of transmits that are 6344 * awaiting completion. 6345 */ 6346 clean_tx_pools(adapter); 6347 6348 rc = reset_sub_crq_queues(adapter); 6349 } 6350 } else { 6351 rc = init_sub_crqs(adapter); 6352 } 6353 6354 if (rc) { 6355 dev_err(dev, "Initialization of sub crqs failed\n"); 6356 release_crq_queue(adapter); 6357 return rc; 6358 } 6359 6360 rc = init_sub_crq_irqs(adapter); 6361 if (rc) { 6362 dev_err(dev, "Failed to initialize sub crq irqs\n"); 6363 release_crq_queue(adapter); 6364 } 6365 6366 return rc; 6367 } 6368 6369 static struct device_attribute dev_attr_failover; 6370 6371 static int ibmvnic_probe(struct vio_dev *dev, const struct vio_device_id *id) 6372 { 6373 struct ibmvnic_adapter *adapter; 6374 struct net_device *netdev; 6375 unsigned char *mac_addr_p; 6376 unsigned long flags; 6377 bool init_success; 6378 int rc; 6379 6380 dev_dbg(&dev->dev, "entering ibmvnic_probe for UA 0x%x\n", 6381 dev->unit_address); 6382 6383 mac_addr_p = (unsigned char *)vio_get_attribute(dev, 6384 VETH_MAC_ADDR, NULL); 6385 if (!mac_addr_p) { 6386 dev_err(&dev->dev, 6387 "(%s:%3.3d) ERROR: Can't find MAC_ADDR attribute\n", 6388 __FILE__, __LINE__); 6389 return 0; 6390 } 6391 6392 netdev = alloc_etherdev_mq(sizeof(struct ibmvnic_adapter), 6393 IBMVNIC_MAX_QUEUES); 6394 if (!netdev) 6395 return -ENOMEM; 6396 6397 adapter = netdev_priv(netdev); 6398 adapter->state = VNIC_PROBING; 6399 dev_set_drvdata(&dev->dev, netdev); 6400 adapter->vdev = dev; 6401 adapter->netdev = netdev; 6402 adapter->login_pending = false; 6403 memset(&adapter->map_ids, 0, sizeof(adapter->map_ids)); 6404 /* map_ids start at 1, so ensure map_id 0 is always "in-use" */ 6405 bitmap_set(adapter->map_ids, 0, 1); 6406 6407 ether_addr_copy(adapter->mac_addr, mac_addr_p); 6408 eth_hw_addr_set(netdev, adapter->mac_addr); 6409 netdev->irq = dev->irq; 6410 netdev->netdev_ops = &ibmvnic_netdev_ops; 6411 netdev->ethtool_ops = &ibmvnic_ethtool_ops; 6412 SET_NETDEV_DEV(netdev, &dev->dev); 6413 6414 INIT_WORK(&adapter->ibmvnic_reset, __ibmvnic_reset); 6415 INIT_DELAYED_WORK(&adapter->ibmvnic_delayed_reset, 6416 __ibmvnic_delayed_reset); 6417 INIT_LIST_HEAD(&adapter->rwi_list); 6418 spin_lock_init(&adapter->rwi_lock); 6419 spin_lock_init(&adapter->state_lock); 6420 mutex_init(&adapter->fw_lock); 6421 init_completion(&adapter->probe_done); 6422 init_completion(&adapter->init_done); 6423 init_completion(&adapter->fw_done); 6424 init_completion(&adapter->reset_done); 6425 init_completion(&adapter->stats_done); 6426 clear_bit(0, &adapter->resetting); 6427 adapter->prev_rx_buf_sz = 0; 6428 adapter->prev_mtu = 0; 6429 6430 init_success = false; 6431 do { 6432 reinit_init_done(adapter); 6433 6434 /* clear any failovers we got in the previous pass 6435 * since we are reinitializing the CRQ 6436 */ 6437 adapter->failover_pending = false; 6438 6439 /* If we had already initialized CRQ, we may have one or 6440 * more resets queued already. Discard those and release 6441 * the CRQ before initializing the CRQ again. 6442 */ 6443 release_crq_queue(adapter); 6444 6445 /* Since we are still in PROBING state, __ibmvnic_reset() 6446 * will not access the ->rwi_list and since we released CRQ, 6447 * we won't get _new_ transport events. But there maybe an 6448 * ongoing ibmvnic_reset() call. So serialize access to 6449 * rwi_list. If we win the race, ibvmnic_reset() could add 6450 * a reset after we purged but thats ok - we just may end 6451 * up with an extra reset (i.e similar to having two or more 6452 * resets in the queue at once). 6453 * CHECK. 6454 */ 6455 spin_lock_irqsave(&adapter->rwi_lock, flags); 6456 flush_reset_queue(adapter); 6457 spin_unlock_irqrestore(&adapter->rwi_lock, flags); 6458 6459 rc = init_crq_queue(adapter); 6460 if (rc) { 6461 dev_err(&dev->dev, "Couldn't initialize crq. rc=%d\n", 6462 rc); 6463 goto ibmvnic_init_fail; 6464 } 6465 6466 rc = ibmvnic_reset_init(adapter, false); 6467 } while (rc == -EAGAIN); 6468 6469 /* We are ignoring the error from ibmvnic_reset_init() assuming that the 6470 * partner is not ready. CRQ is not active. When the partner becomes 6471 * ready, we will do the passive init reset. 6472 */ 6473 6474 if (!rc) 6475 init_success = true; 6476 6477 rc = init_stats_buffers(adapter); 6478 if (rc) 6479 goto ibmvnic_init_fail; 6480 6481 rc = init_stats_token(adapter); 6482 if (rc) 6483 goto ibmvnic_stats_fail; 6484 6485 rc = device_create_file(&dev->dev, &dev_attr_failover); 6486 if (rc) 6487 goto ibmvnic_dev_file_err; 6488 6489 netif_carrier_off(netdev); 6490 6491 if (init_success) { 6492 adapter->state = VNIC_PROBED; 6493 netdev->mtu = adapter->req_mtu - ETH_HLEN; 6494 netdev->min_mtu = adapter->min_mtu - ETH_HLEN; 6495 netdev->max_mtu = adapter->max_mtu - ETH_HLEN; 6496 } else { 6497 adapter->state = VNIC_DOWN; 6498 } 6499 6500 adapter->wait_for_reset = false; 6501 adapter->last_reset_time = jiffies; 6502 6503 rc = register_netdev(netdev); 6504 if (rc) { 6505 dev_err(&dev->dev, "failed to register netdev rc=%d\n", rc); 6506 goto ibmvnic_register_fail; 6507 } 6508 dev_info(&dev->dev, "ibmvnic registered\n"); 6509 6510 rc = ibmvnic_cpu_notif_add(adapter); 6511 if (rc) { 6512 netdev_err(netdev, "Registering cpu notifier failed\n"); 6513 goto cpu_notif_add_failed; 6514 } 6515 6516 complete(&adapter->probe_done); 6517 6518 return 0; 6519 6520 cpu_notif_add_failed: 6521 unregister_netdev(netdev); 6522 6523 ibmvnic_register_fail: 6524 device_remove_file(&dev->dev, &dev_attr_failover); 6525 6526 ibmvnic_dev_file_err: 6527 release_stats_token(adapter); 6528 6529 ibmvnic_stats_fail: 6530 release_stats_buffers(adapter); 6531 6532 ibmvnic_init_fail: 6533 release_sub_crqs(adapter, 1); 6534 release_crq_queue(adapter); 6535 6536 /* cleanup worker thread after releasing CRQ so we don't get 6537 * transport events (i.e new work items for the worker thread). 6538 */ 6539 adapter->state = VNIC_REMOVING; 6540 complete(&adapter->probe_done); 6541 flush_work(&adapter->ibmvnic_reset); 6542 flush_delayed_work(&adapter->ibmvnic_delayed_reset); 6543 6544 flush_reset_queue(adapter); 6545 6546 mutex_destroy(&adapter->fw_lock); 6547 free_netdev(netdev); 6548 6549 return rc; 6550 } 6551 6552 static void ibmvnic_remove(struct vio_dev *dev) 6553 { 6554 struct net_device *netdev = dev_get_drvdata(&dev->dev); 6555 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 6556 unsigned long flags; 6557 6558 spin_lock_irqsave(&adapter->state_lock, flags); 6559 6560 /* If ibmvnic_reset() is scheduling a reset, wait for it to 6561 * finish. Then, set the state to REMOVING to prevent it from 6562 * scheduling any more work and to have reset functions ignore 6563 * any resets that have already been scheduled. Drop the lock 6564 * after setting state, so __ibmvnic_reset() which is called 6565 * from the flush_work() below, can make progress. 6566 */ 6567 spin_lock(&adapter->rwi_lock); 6568 adapter->state = VNIC_REMOVING; 6569 spin_unlock(&adapter->rwi_lock); 6570 6571 spin_unlock_irqrestore(&adapter->state_lock, flags); 6572 6573 ibmvnic_cpu_notif_remove(adapter); 6574 6575 flush_work(&adapter->ibmvnic_reset); 6576 flush_delayed_work(&adapter->ibmvnic_delayed_reset); 6577 6578 rtnl_lock(); 6579 unregister_netdevice(netdev); 6580 6581 release_resources(adapter); 6582 release_rx_pools(adapter); 6583 release_tx_pools(adapter); 6584 release_sub_crqs(adapter, 1); 6585 release_crq_queue(adapter); 6586 6587 release_stats_token(adapter); 6588 release_stats_buffers(adapter); 6589 6590 adapter->state = VNIC_REMOVED; 6591 6592 rtnl_unlock(); 6593 mutex_destroy(&adapter->fw_lock); 6594 device_remove_file(&dev->dev, &dev_attr_failover); 6595 free_netdev(netdev); 6596 dev_set_drvdata(&dev->dev, NULL); 6597 } 6598 6599 static ssize_t failover_store(struct device *dev, struct device_attribute *attr, 6600 const char *buf, size_t count) 6601 { 6602 struct net_device *netdev = dev_get_drvdata(dev); 6603 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 6604 unsigned long retbuf[PLPAR_HCALL_BUFSIZE]; 6605 __be64 session_token; 6606 long rc; 6607 6608 if (!sysfs_streq(buf, "1")) 6609 return -EINVAL; 6610 6611 rc = plpar_hcall(H_VIOCTL, retbuf, adapter->vdev->unit_address, 6612 H_GET_SESSION_TOKEN, 0, 0, 0); 6613 if (rc) { 6614 netdev_err(netdev, "Couldn't retrieve session token, rc %ld\n", 6615 rc); 6616 goto last_resort; 6617 } 6618 6619 session_token = (__be64)retbuf[0]; 6620 netdev_dbg(netdev, "Initiating client failover, session id %llx\n", 6621 be64_to_cpu(session_token)); 6622 rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address, 6623 H_SESSION_ERR_DETECTED, session_token, 0, 0); 6624 if (rc) { 6625 netdev_err(netdev, 6626 "H_VIOCTL initiated failover failed, rc %ld\n", 6627 rc); 6628 goto last_resort; 6629 } 6630 6631 return count; 6632 6633 last_resort: 6634 netdev_dbg(netdev, "Trying to send CRQ_CMD, the last resort\n"); 6635 ibmvnic_reset(adapter, VNIC_RESET_FAILOVER); 6636 6637 return count; 6638 } 6639 static DEVICE_ATTR_WO(failover); 6640 6641 static unsigned long ibmvnic_get_desired_dma(struct vio_dev *vdev) 6642 { 6643 struct net_device *netdev = dev_get_drvdata(&vdev->dev); 6644 struct ibmvnic_adapter *adapter; 6645 struct iommu_table *tbl; 6646 unsigned long ret = 0; 6647 int i; 6648 6649 tbl = get_iommu_table_base(&vdev->dev); 6650 6651 /* netdev inits at probe time along with the structures we need below*/ 6652 if (!netdev) 6653 return IOMMU_PAGE_ALIGN(IBMVNIC_IO_ENTITLEMENT_DEFAULT, tbl); 6654 6655 adapter = netdev_priv(netdev); 6656 6657 ret += PAGE_SIZE; /* the crq message queue */ 6658 ret += IOMMU_PAGE_ALIGN(sizeof(struct ibmvnic_statistics), tbl); 6659 6660 for (i = 0; i < adapter->req_tx_queues + adapter->req_rx_queues; i++) 6661 ret += 4 * PAGE_SIZE; /* the scrq message queue */ 6662 6663 for (i = 0; i < adapter->num_active_rx_pools; i++) 6664 ret += adapter->rx_pool[i].size * 6665 IOMMU_PAGE_ALIGN(adapter->rx_pool[i].buff_size, tbl); 6666 6667 return ret; 6668 } 6669 6670 static int ibmvnic_resume(struct device *dev) 6671 { 6672 struct net_device *netdev = dev_get_drvdata(dev); 6673 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 6674 6675 if (adapter->state != VNIC_OPEN) 6676 return 0; 6677 6678 tasklet_schedule(&adapter->tasklet); 6679 6680 return 0; 6681 } 6682 6683 static const struct vio_device_id ibmvnic_device_table[] = { 6684 {"network", "IBM,vnic"}, 6685 {"", "" } 6686 }; 6687 MODULE_DEVICE_TABLE(vio, ibmvnic_device_table); 6688 6689 static const struct dev_pm_ops ibmvnic_pm_ops = { 6690 .resume = ibmvnic_resume 6691 }; 6692 6693 static struct vio_driver ibmvnic_driver = { 6694 .id_table = ibmvnic_device_table, 6695 .probe = ibmvnic_probe, 6696 .remove = ibmvnic_remove, 6697 .get_desired_dma = ibmvnic_get_desired_dma, 6698 .name = ibmvnic_driver_name, 6699 .pm = &ibmvnic_pm_ops, 6700 }; 6701 6702 /* module functions */ 6703 static int __init ibmvnic_module_init(void) 6704 { 6705 int ret; 6706 6707 ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN, "net/ibmvnic:online", 6708 ibmvnic_cpu_online, 6709 ibmvnic_cpu_down_prep); 6710 if (ret < 0) 6711 goto out; 6712 ibmvnic_online = ret; 6713 ret = cpuhp_setup_state_multi(CPUHP_IBMVNIC_DEAD, "net/ibmvnic:dead", 6714 NULL, ibmvnic_cpu_dead); 6715 if (ret) 6716 goto err_dead; 6717 6718 ret = vio_register_driver(&ibmvnic_driver); 6719 if (ret) 6720 goto err_vio_register; 6721 6722 pr_info("%s: %s %s\n", ibmvnic_driver_name, ibmvnic_driver_string, 6723 IBMVNIC_DRIVER_VERSION); 6724 6725 return 0; 6726 err_vio_register: 6727 cpuhp_remove_multi_state(CPUHP_IBMVNIC_DEAD); 6728 err_dead: 6729 cpuhp_remove_multi_state(ibmvnic_online); 6730 out: 6731 return ret; 6732 } 6733 6734 static void __exit ibmvnic_module_exit(void) 6735 { 6736 vio_unregister_driver(&ibmvnic_driver); 6737 cpuhp_remove_multi_state(CPUHP_IBMVNIC_DEAD); 6738 cpuhp_remove_multi_state(ibmvnic_online); 6739 } 6740 6741 module_init(ibmvnic_module_init); 6742 module_exit(ibmvnic_module_exit); 6743