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/kthread.h> 57 #include <linux/seq_file.h> 58 #include <linux/interrupt.h> 59 #include <net/net_namespace.h> 60 #include <asm/hvcall.h> 61 #include <linux/atomic.h> 62 #include <asm/vio.h> 63 #include <asm/iommu.h> 64 #include <linux/uaccess.h> 65 #include <asm/firmware.h> 66 #include <linux/workqueue.h> 67 #include <linux/if_vlan.h> 68 #include <linux/utsname.h> 69 70 #include "ibmvnic.h" 71 72 static const char ibmvnic_driver_name[] = "ibmvnic"; 73 static const char ibmvnic_driver_string[] = "IBM System i/p Virtual NIC Driver"; 74 75 MODULE_AUTHOR("Santiago Leon"); 76 MODULE_DESCRIPTION("IBM System i/p Virtual NIC Driver"); 77 MODULE_LICENSE("GPL"); 78 MODULE_VERSION(IBMVNIC_DRIVER_VERSION); 79 80 static int ibmvnic_version = IBMVNIC_INITIAL_VERSION; 81 static void release_sub_crqs(struct ibmvnic_adapter *, bool); 82 static int ibmvnic_reset_crq(struct ibmvnic_adapter *); 83 static int ibmvnic_send_crq_init(struct ibmvnic_adapter *); 84 static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *); 85 static int ibmvnic_send_crq(struct ibmvnic_adapter *, union ibmvnic_crq *); 86 static int send_subcrq_indirect(struct ibmvnic_adapter *, u64, u64, u64); 87 static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance); 88 static int enable_scrq_irq(struct ibmvnic_adapter *, 89 struct ibmvnic_sub_crq_queue *); 90 static int disable_scrq_irq(struct ibmvnic_adapter *, 91 struct ibmvnic_sub_crq_queue *); 92 static int pending_scrq(struct ibmvnic_adapter *, 93 struct ibmvnic_sub_crq_queue *); 94 static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *, 95 struct ibmvnic_sub_crq_queue *); 96 static int ibmvnic_poll(struct napi_struct *napi, int data); 97 static void send_query_map(struct ibmvnic_adapter *adapter); 98 static int send_request_map(struct ibmvnic_adapter *, dma_addr_t, __be32, u8); 99 static int send_request_unmap(struct ibmvnic_adapter *, u8); 100 static int send_login(struct ibmvnic_adapter *adapter); 101 static void send_query_cap(struct ibmvnic_adapter *adapter); 102 static int init_sub_crqs(struct ibmvnic_adapter *); 103 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter); 104 static int ibmvnic_reset_init(struct ibmvnic_adapter *, bool reset); 105 static void release_crq_queue(struct ibmvnic_adapter *); 106 static int __ibmvnic_set_mac(struct net_device *, u8 *); 107 static int init_crq_queue(struct ibmvnic_adapter *adapter); 108 static int send_query_phys_parms(struct ibmvnic_adapter *adapter); 109 110 struct ibmvnic_stat { 111 char name[ETH_GSTRING_LEN]; 112 int offset; 113 }; 114 115 #define IBMVNIC_STAT_OFF(stat) (offsetof(struct ibmvnic_adapter, stats) + \ 116 offsetof(struct ibmvnic_statistics, stat)) 117 #define IBMVNIC_GET_STAT(a, off) (*((u64 *)(((unsigned long)(a)) + (off)))) 118 119 static const struct ibmvnic_stat ibmvnic_stats[] = { 120 {"rx_packets", IBMVNIC_STAT_OFF(rx_packets)}, 121 {"rx_bytes", IBMVNIC_STAT_OFF(rx_bytes)}, 122 {"tx_packets", IBMVNIC_STAT_OFF(tx_packets)}, 123 {"tx_bytes", IBMVNIC_STAT_OFF(tx_bytes)}, 124 {"ucast_tx_packets", IBMVNIC_STAT_OFF(ucast_tx_packets)}, 125 {"ucast_rx_packets", IBMVNIC_STAT_OFF(ucast_rx_packets)}, 126 {"mcast_tx_packets", IBMVNIC_STAT_OFF(mcast_tx_packets)}, 127 {"mcast_rx_packets", IBMVNIC_STAT_OFF(mcast_rx_packets)}, 128 {"bcast_tx_packets", IBMVNIC_STAT_OFF(bcast_tx_packets)}, 129 {"bcast_rx_packets", IBMVNIC_STAT_OFF(bcast_rx_packets)}, 130 {"align_errors", IBMVNIC_STAT_OFF(align_errors)}, 131 {"fcs_errors", IBMVNIC_STAT_OFF(fcs_errors)}, 132 {"single_collision_frames", IBMVNIC_STAT_OFF(single_collision_frames)}, 133 {"multi_collision_frames", IBMVNIC_STAT_OFF(multi_collision_frames)}, 134 {"sqe_test_errors", IBMVNIC_STAT_OFF(sqe_test_errors)}, 135 {"deferred_tx", IBMVNIC_STAT_OFF(deferred_tx)}, 136 {"late_collisions", IBMVNIC_STAT_OFF(late_collisions)}, 137 {"excess_collisions", IBMVNIC_STAT_OFF(excess_collisions)}, 138 {"internal_mac_tx_errors", IBMVNIC_STAT_OFF(internal_mac_tx_errors)}, 139 {"carrier_sense", IBMVNIC_STAT_OFF(carrier_sense)}, 140 {"too_long_frames", IBMVNIC_STAT_OFF(too_long_frames)}, 141 {"internal_mac_rx_errors", IBMVNIC_STAT_OFF(internal_mac_rx_errors)}, 142 }; 143 144 static long h_reg_sub_crq(unsigned long unit_address, unsigned long token, 145 unsigned long length, unsigned long *number, 146 unsigned long *irq) 147 { 148 unsigned long retbuf[PLPAR_HCALL_BUFSIZE]; 149 long rc; 150 151 rc = plpar_hcall(H_REG_SUB_CRQ, retbuf, unit_address, token, length); 152 *number = retbuf[0]; 153 *irq = retbuf[1]; 154 155 return rc; 156 } 157 158 /** 159 * ibmvnic_wait_for_completion - Check device state and wait for completion 160 * @adapter: private device data 161 * @comp_done: completion structure to wait for 162 * @timeout: time to wait in milliseconds 163 * 164 * Wait for a completion signal or until the timeout limit is reached 165 * while checking that the device is still active. 166 */ 167 static int ibmvnic_wait_for_completion(struct ibmvnic_adapter *adapter, 168 struct completion *comp_done, 169 unsigned long timeout) 170 { 171 struct net_device *netdev; 172 unsigned long div_timeout; 173 u8 retry; 174 175 netdev = adapter->netdev; 176 retry = 5; 177 div_timeout = msecs_to_jiffies(timeout / retry); 178 while (true) { 179 if (!adapter->crq.active) { 180 netdev_err(netdev, "Device down!\n"); 181 return -ENODEV; 182 } 183 if (!retry--) 184 break; 185 if (wait_for_completion_timeout(comp_done, div_timeout)) 186 return 0; 187 } 188 netdev_err(netdev, "Operation timed out.\n"); 189 return -ETIMEDOUT; 190 } 191 192 static int alloc_long_term_buff(struct ibmvnic_adapter *adapter, 193 struct ibmvnic_long_term_buff *ltb, int size) 194 { 195 struct device *dev = &adapter->vdev->dev; 196 int rc; 197 198 ltb->size = size; 199 ltb->buff = dma_alloc_coherent(dev, ltb->size, <b->addr, 200 GFP_KERNEL); 201 202 if (!ltb->buff) { 203 dev_err(dev, "Couldn't alloc long term buffer\n"); 204 return -ENOMEM; 205 } 206 ltb->map_id = adapter->map_id; 207 adapter->map_id++; 208 209 mutex_lock(&adapter->fw_lock); 210 adapter->fw_done_rc = 0; 211 reinit_completion(&adapter->fw_done); 212 rc = send_request_map(adapter, ltb->addr, 213 ltb->size, ltb->map_id); 214 if (rc) { 215 dma_free_coherent(dev, ltb->size, ltb->buff, ltb->addr); 216 mutex_unlock(&adapter->fw_lock); 217 return rc; 218 } 219 220 rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000); 221 if (rc) { 222 dev_err(dev, 223 "Long term map request aborted or timed out,rc = %d\n", 224 rc); 225 dma_free_coherent(dev, ltb->size, ltb->buff, ltb->addr); 226 mutex_unlock(&adapter->fw_lock); 227 return rc; 228 } 229 230 if (adapter->fw_done_rc) { 231 dev_err(dev, "Couldn't map long term buffer,rc = %d\n", 232 adapter->fw_done_rc); 233 dma_free_coherent(dev, ltb->size, ltb->buff, ltb->addr); 234 mutex_unlock(&adapter->fw_lock); 235 return -1; 236 } 237 mutex_unlock(&adapter->fw_lock); 238 return 0; 239 } 240 241 static void free_long_term_buff(struct ibmvnic_adapter *adapter, 242 struct ibmvnic_long_term_buff *ltb) 243 { 244 struct device *dev = &adapter->vdev->dev; 245 246 if (!ltb->buff) 247 return; 248 249 /* VIOS automatically unmaps the long term buffer at remote 250 * end for the following resets: 251 * FAILOVER, MOBILITY, TIMEOUT. 252 */ 253 if (adapter->reset_reason != VNIC_RESET_FAILOVER && 254 adapter->reset_reason != VNIC_RESET_MOBILITY && 255 adapter->reset_reason != VNIC_RESET_TIMEOUT) 256 send_request_unmap(adapter, ltb->map_id); 257 dma_free_coherent(dev, ltb->size, ltb->buff, ltb->addr); 258 } 259 260 static int reset_long_term_buff(struct ibmvnic_long_term_buff *ltb) 261 { 262 if (!ltb->buff) 263 return -EINVAL; 264 265 memset(ltb->buff, 0, ltb->size); 266 return 0; 267 } 268 269 static void deactivate_rx_pools(struct ibmvnic_adapter *adapter) 270 { 271 int i; 272 273 for (i = 0; i < adapter->num_active_rx_pools; i++) 274 adapter->rx_pool[i].active = 0; 275 } 276 277 static void replenish_rx_pool(struct ibmvnic_adapter *adapter, 278 struct ibmvnic_rx_pool *pool) 279 { 280 int count = pool->size - atomic_read(&pool->available); 281 u64 handle = adapter->rx_scrq[pool->index]->handle; 282 struct device *dev = &adapter->vdev->dev; 283 struct ibmvnic_ind_xmit_queue *ind_bufp; 284 struct ibmvnic_sub_crq_queue *rx_scrq; 285 union sub_crq *sub_crq; 286 int buffers_added = 0; 287 unsigned long lpar_rc; 288 struct sk_buff *skb; 289 unsigned int offset; 290 dma_addr_t dma_addr; 291 unsigned char *dst; 292 int shift = 0; 293 int index; 294 int i; 295 296 if (!pool->active) 297 return; 298 299 rx_scrq = adapter->rx_scrq[pool->index]; 300 ind_bufp = &rx_scrq->ind_buf; 301 for (i = 0; i < count; ++i) { 302 skb = netdev_alloc_skb(adapter->netdev, pool->buff_size); 303 if (!skb) { 304 dev_err(dev, "Couldn't replenish rx buff\n"); 305 adapter->replenish_no_mem++; 306 break; 307 } 308 309 index = pool->free_map[pool->next_free]; 310 311 if (pool->rx_buff[index].skb) 312 dev_err(dev, "Inconsistent free_map!\n"); 313 314 /* Copy the skb to the long term mapped DMA buffer */ 315 offset = index * pool->buff_size; 316 dst = pool->long_term_buff.buff + offset; 317 memset(dst, 0, pool->buff_size); 318 dma_addr = pool->long_term_buff.addr + offset; 319 pool->rx_buff[index].data = dst; 320 321 pool->free_map[pool->next_free] = IBMVNIC_INVALID_MAP; 322 pool->rx_buff[index].dma = dma_addr; 323 pool->rx_buff[index].skb = skb; 324 pool->rx_buff[index].pool_index = pool->index; 325 pool->rx_buff[index].size = pool->buff_size; 326 327 sub_crq = &ind_bufp->indir_arr[ind_bufp->index++]; 328 memset(sub_crq, 0, sizeof(*sub_crq)); 329 sub_crq->rx_add.first = IBMVNIC_CRQ_CMD; 330 sub_crq->rx_add.correlator = 331 cpu_to_be64((u64)&pool->rx_buff[index]); 332 sub_crq->rx_add.ioba = cpu_to_be32(dma_addr); 333 sub_crq->rx_add.map_id = pool->long_term_buff.map_id; 334 335 /* The length field of the sCRQ is defined to be 24 bits so the 336 * buffer size needs to be left shifted by a byte before it is 337 * converted to big endian to prevent the last byte from being 338 * truncated. 339 */ 340 #ifdef __LITTLE_ENDIAN__ 341 shift = 8; 342 #endif 343 sub_crq->rx_add.len = cpu_to_be32(pool->buff_size << shift); 344 pool->next_free = (pool->next_free + 1) % pool->size; 345 if (ind_bufp->index == IBMVNIC_MAX_IND_DESCS || 346 i == count - 1) { 347 lpar_rc = 348 send_subcrq_indirect(adapter, handle, 349 (u64)ind_bufp->indir_dma, 350 (u64)ind_bufp->index); 351 if (lpar_rc != H_SUCCESS) 352 goto failure; 353 buffers_added += ind_bufp->index; 354 adapter->replenish_add_buff_success += ind_bufp->index; 355 ind_bufp->index = 0; 356 } 357 } 358 atomic_add(buffers_added, &pool->available); 359 return; 360 361 failure: 362 if (lpar_rc != H_PARAMETER && lpar_rc != H_CLOSED) 363 dev_err_ratelimited(dev, "rx: replenish packet buffer failed\n"); 364 for (i = ind_bufp->index - 1; i >= 0; --i) { 365 struct ibmvnic_rx_buff *rx_buff; 366 367 pool->next_free = pool->next_free == 0 ? 368 pool->size - 1 : pool->next_free - 1; 369 sub_crq = &ind_bufp->indir_arr[i]; 370 rx_buff = (struct ibmvnic_rx_buff *) 371 be64_to_cpu(sub_crq->rx_add.correlator); 372 index = (int)(rx_buff - pool->rx_buff); 373 pool->free_map[pool->next_free] = index; 374 dev_kfree_skb_any(pool->rx_buff[index].skb); 375 pool->rx_buff[index].skb = NULL; 376 } 377 adapter->replenish_add_buff_failure += ind_bufp->index; 378 atomic_add(buffers_added, &pool->available); 379 ind_bufp->index = 0; 380 if (lpar_rc == H_CLOSED || adapter->failover_pending) { 381 /* Disable buffer pool replenishment and report carrier off if 382 * queue is closed or pending failover. 383 * Firmware guarantees that a signal will be sent to the 384 * driver, triggering a reset. 385 */ 386 deactivate_rx_pools(adapter); 387 netif_carrier_off(adapter->netdev); 388 } 389 } 390 391 static void replenish_pools(struct ibmvnic_adapter *adapter) 392 { 393 int i; 394 395 adapter->replenish_task_cycles++; 396 for (i = 0; i < adapter->num_active_rx_pools; i++) { 397 if (adapter->rx_pool[i].active) 398 replenish_rx_pool(adapter, &adapter->rx_pool[i]); 399 } 400 401 netdev_dbg(adapter->netdev, "Replenished %d pools\n", i); 402 } 403 404 static void release_stats_buffers(struct ibmvnic_adapter *adapter) 405 { 406 kfree(adapter->tx_stats_buffers); 407 kfree(adapter->rx_stats_buffers); 408 adapter->tx_stats_buffers = NULL; 409 adapter->rx_stats_buffers = NULL; 410 } 411 412 static int init_stats_buffers(struct ibmvnic_adapter *adapter) 413 { 414 adapter->tx_stats_buffers = 415 kcalloc(IBMVNIC_MAX_QUEUES, 416 sizeof(struct ibmvnic_tx_queue_stats), 417 GFP_KERNEL); 418 if (!adapter->tx_stats_buffers) 419 return -ENOMEM; 420 421 adapter->rx_stats_buffers = 422 kcalloc(IBMVNIC_MAX_QUEUES, 423 sizeof(struct ibmvnic_rx_queue_stats), 424 GFP_KERNEL); 425 if (!adapter->rx_stats_buffers) 426 return -ENOMEM; 427 428 return 0; 429 } 430 431 static void release_stats_token(struct ibmvnic_adapter *adapter) 432 { 433 struct device *dev = &adapter->vdev->dev; 434 435 if (!adapter->stats_token) 436 return; 437 438 dma_unmap_single(dev, adapter->stats_token, 439 sizeof(struct ibmvnic_statistics), 440 DMA_FROM_DEVICE); 441 adapter->stats_token = 0; 442 } 443 444 static int init_stats_token(struct ibmvnic_adapter *adapter) 445 { 446 struct device *dev = &adapter->vdev->dev; 447 dma_addr_t stok; 448 449 stok = dma_map_single(dev, &adapter->stats, 450 sizeof(struct ibmvnic_statistics), 451 DMA_FROM_DEVICE); 452 if (dma_mapping_error(dev, stok)) { 453 dev_err(dev, "Couldn't map stats buffer\n"); 454 return -1; 455 } 456 457 adapter->stats_token = stok; 458 netdev_dbg(adapter->netdev, "Stats token initialized (%llx)\n", stok); 459 return 0; 460 } 461 462 static int reset_rx_pools(struct ibmvnic_adapter *adapter) 463 { 464 struct ibmvnic_rx_pool *rx_pool; 465 u64 buff_size; 466 int rx_scrqs; 467 int i, j, rc; 468 469 if (!adapter->rx_pool) 470 return -1; 471 472 buff_size = adapter->cur_rx_buf_sz; 473 rx_scrqs = adapter->num_active_rx_pools; 474 for (i = 0; i < rx_scrqs; i++) { 475 rx_pool = &adapter->rx_pool[i]; 476 477 netdev_dbg(adapter->netdev, "Re-setting rx_pool[%d]\n", i); 478 479 if (rx_pool->buff_size != buff_size) { 480 free_long_term_buff(adapter, &rx_pool->long_term_buff); 481 rx_pool->buff_size = ALIGN(buff_size, L1_CACHE_BYTES); 482 rc = alloc_long_term_buff(adapter, 483 &rx_pool->long_term_buff, 484 rx_pool->size * 485 rx_pool->buff_size); 486 } else { 487 rc = reset_long_term_buff(&rx_pool->long_term_buff); 488 } 489 490 if (rc) 491 return rc; 492 493 for (j = 0; j < rx_pool->size; j++) 494 rx_pool->free_map[j] = j; 495 496 memset(rx_pool->rx_buff, 0, 497 rx_pool->size * sizeof(struct ibmvnic_rx_buff)); 498 499 atomic_set(&rx_pool->available, 0); 500 rx_pool->next_alloc = 0; 501 rx_pool->next_free = 0; 502 rx_pool->active = 1; 503 } 504 505 return 0; 506 } 507 508 static void release_rx_pools(struct ibmvnic_adapter *adapter) 509 { 510 struct ibmvnic_rx_pool *rx_pool; 511 int i, j; 512 513 if (!adapter->rx_pool) 514 return; 515 516 for (i = 0; i < adapter->num_active_rx_pools; i++) { 517 rx_pool = &adapter->rx_pool[i]; 518 519 netdev_dbg(adapter->netdev, "Releasing rx_pool[%d]\n", i); 520 521 kfree(rx_pool->free_map); 522 free_long_term_buff(adapter, &rx_pool->long_term_buff); 523 524 if (!rx_pool->rx_buff) 525 continue; 526 527 for (j = 0; j < rx_pool->size; j++) { 528 if (rx_pool->rx_buff[j].skb) { 529 dev_kfree_skb_any(rx_pool->rx_buff[j].skb); 530 rx_pool->rx_buff[j].skb = NULL; 531 } 532 } 533 534 kfree(rx_pool->rx_buff); 535 } 536 537 kfree(adapter->rx_pool); 538 adapter->rx_pool = NULL; 539 adapter->num_active_rx_pools = 0; 540 } 541 542 static int init_rx_pools(struct net_device *netdev) 543 { 544 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 545 struct device *dev = &adapter->vdev->dev; 546 struct ibmvnic_rx_pool *rx_pool; 547 int rxadd_subcrqs; 548 u64 buff_size; 549 int i, j; 550 551 rxadd_subcrqs = adapter->num_active_rx_scrqs; 552 buff_size = adapter->cur_rx_buf_sz; 553 554 adapter->rx_pool = kcalloc(rxadd_subcrqs, 555 sizeof(struct ibmvnic_rx_pool), 556 GFP_KERNEL); 557 if (!adapter->rx_pool) { 558 dev_err(dev, "Failed to allocate rx pools\n"); 559 return -1; 560 } 561 562 adapter->num_active_rx_pools = rxadd_subcrqs; 563 564 for (i = 0; i < rxadd_subcrqs; i++) { 565 rx_pool = &adapter->rx_pool[i]; 566 567 netdev_dbg(adapter->netdev, 568 "Initializing rx_pool[%d], %lld buffs, %lld bytes each\n", 569 i, adapter->req_rx_add_entries_per_subcrq, 570 buff_size); 571 572 rx_pool->size = adapter->req_rx_add_entries_per_subcrq; 573 rx_pool->index = i; 574 rx_pool->buff_size = ALIGN(buff_size, L1_CACHE_BYTES); 575 rx_pool->active = 1; 576 577 rx_pool->free_map = kcalloc(rx_pool->size, sizeof(int), 578 GFP_KERNEL); 579 if (!rx_pool->free_map) { 580 release_rx_pools(adapter); 581 return -1; 582 } 583 584 rx_pool->rx_buff = kcalloc(rx_pool->size, 585 sizeof(struct ibmvnic_rx_buff), 586 GFP_KERNEL); 587 if (!rx_pool->rx_buff) { 588 dev_err(dev, "Couldn't alloc rx buffers\n"); 589 release_rx_pools(adapter); 590 return -1; 591 } 592 593 if (alloc_long_term_buff(adapter, &rx_pool->long_term_buff, 594 rx_pool->size * rx_pool->buff_size)) { 595 release_rx_pools(adapter); 596 return -1; 597 } 598 599 for (j = 0; j < rx_pool->size; ++j) 600 rx_pool->free_map[j] = j; 601 602 atomic_set(&rx_pool->available, 0); 603 rx_pool->next_alloc = 0; 604 rx_pool->next_free = 0; 605 } 606 607 return 0; 608 } 609 610 static int reset_one_tx_pool(struct ibmvnic_tx_pool *tx_pool) 611 { 612 int rc, i; 613 614 rc = reset_long_term_buff(&tx_pool->long_term_buff); 615 if (rc) 616 return rc; 617 618 memset(tx_pool->tx_buff, 0, 619 tx_pool->num_buffers * 620 sizeof(struct ibmvnic_tx_buff)); 621 622 for (i = 0; i < tx_pool->num_buffers; i++) 623 tx_pool->free_map[i] = i; 624 625 tx_pool->consumer_index = 0; 626 tx_pool->producer_index = 0; 627 628 return 0; 629 } 630 631 static int reset_tx_pools(struct ibmvnic_adapter *adapter) 632 { 633 int tx_scrqs; 634 int i, rc; 635 636 if (!adapter->tx_pool) 637 return -1; 638 639 tx_scrqs = adapter->num_active_tx_pools; 640 for (i = 0; i < tx_scrqs; i++) { 641 rc = reset_one_tx_pool(&adapter->tso_pool[i]); 642 if (rc) 643 return rc; 644 rc = reset_one_tx_pool(&adapter->tx_pool[i]); 645 if (rc) 646 return rc; 647 } 648 649 return 0; 650 } 651 652 static void release_vpd_data(struct ibmvnic_adapter *adapter) 653 { 654 if (!adapter->vpd) 655 return; 656 657 kfree(adapter->vpd->buff); 658 kfree(adapter->vpd); 659 660 adapter->vpd = NULL; 661 } 662 663 static void release_one_tx_pool(struct ibmvnic_adapter *adapter, 664 struct ibmvnic_tx_pool *tx_pool) 665 { 666 kfree(tx_pool->tx_buff); 667 kfree(tx_pool->free_map); 668 free_long_term_buff(adapter, &tx_pool->long_term_buff); 669 } 670 671 static void release_tx_pools(struct ibmvnic_adapter *adapter) 672 { 673 int i; 674 675 if (!adapter->tx_pool) 676 return; 677 678 for (i = 0; i < adapter->num_active_tx_pools; i++) { 679 release_one_tx_pool(adapter, &adapter->tx_pool[i]); 680 release_one_tx_pool(adapter, &adapter->tso_pool[i]); 681 } 682 683 kfree(adapter->tx_pool); 684 adapter->tx_pool = NULL; 685 kfree(adapter->tso_pool); 686 adapter->tso_pool = NULL; 687 adapter->num_active_tx_pools = 0; 688 } 689 690 static int init_one_tx_pool(struct net_device *netdev, 691 struct ibmvnic_tx_pool *tx_pool, 692 int num_entries, int buf_size) 693 { 694 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 695 int i; 696 697 tx_pool->tx_buff = kcalloc(num_entries, 698 sizeof(struct ibmvnic_tx_buff), 699 GFP_KERNEL); 700 if (!tx_pool->tx_buff) 701 return -1; 702 703 if (alloc_long_term_buff(adapter, &tx_pool->long_term_buff, 704 num_entries * buf_size)) 705 return -1; 706 707 tx_pool->free_map = kcalloc(num_entries, sizeof(int), GFP_KERNEL); 708 if (!tx_pool->free_map) 709 return -1; 710 711 for (i = 0; i < num_entries; i++) 712 tx_pool->free_map[i] = i; 713 714 tx_pool->consumer_index = 0; 715 tx_pool->producer_index = 0; 716 tx_pool->num_buffers = num_entries; 717 tx_pool->buf_size = buf_size; 718 719 return 0; 720 } 721 722 static int init_tx_pools(struct net_device *netdev) 723 { 724 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 725 int tx_subcrqs; 726 u64 buff_size; 727 int i, rc; 728 729 tx_subcrqs = adapter->num_active_tx_scrqs; 730 adapter->tx_pool = kcalloc(tx_subcrqs, 731 sizeof(struct ibmvnic_tx_pool), GFP_KERNEL); 732 if (!adapter->tx_pool) 733 return -1; 734 735 adapter->tso_pool = kcalloc(tx_subcrqs, 736 sizeof(struct ibmvnic_tx_pool), GFP_KERNEL); 737 if (!adapter->tso_pool) 738 return -1; 739 740 adapter->num_active_tx_pools = tx_subcrqs; 741 742 for (i = 0; i < tx_subcrqs; i++) { 743 buff_size = adapter->req_mtu + VLAN_HLEN; 744 buff_size = ALIGN(buff_size, L1_CACHE_BYTES); 745 rc = init_one_tx_pool(netdev, &adapter->tx_pool[i], 746 adapter->req_tx_entries_per_subcrq, 747 buff_size); 748 if (rc) { 749 release_tx_pools(adapter); 750 return rc; 751 } 752 753 rc = init_one_tx_pool(netdev, &adapter->tso_pool[i], 754 IBMVNIC_TSO_BUFS, 755 IBMVNIC_TSO_BUF_SZ); 756 if (rc) { 757 release_tx_pools(adapter); 758 return rc; 759 } 760 } 761 762 return 0; 763 } 764 765 static void ibmvnic_napi_enable(struct ibmvnic_adapter *adapter) 766 { 767 int i; 768 769 if (adapter->napi_enabled) 770 return; 771 772 for (i = 0; i < adapter->req_rx_queues; i++) 773 napi_enable(&adapter->napi[i]); 774 775 adapter->napi_enabled = true; 776 } 777 778 static void ibmvnic_napi_disable(struct ibmvnic_adapter *adapter) 779 { 780 int i; 781 782 if (!adapter->napi_enabled) 783 return; 784 785 for (i = 0; i < adapter->req_rx_queues; i++) { 786 netdev_dbg(adapter->netdev, "Disabling napi[%d]\n", i); 787 napi_disable(&adapter->napi[i]); 788 } 789 790 adapter->napi_enabled = false; 791 } 792 793 static int init_napi(struct ibmvnic_adapter *adapter) 794 { 795 int i; 796 797 adapter->napi = kcalloc(adapter->req_rx_queues, 798 sizeof(struct napi_struct), GFP_KERNEL); 799 if (!adapter->napi) 800 return -ENOMEM; 801 802 for (i = 0; i < adapter->req_rx_queues; i++) { 803 netdev_dbg(adapter->netdev, "Adding napi[%d]\n", i); 804 netif_napi_add(adapter->netdev, &adapter->napi[i], 805 ibmvnic_poll, NAPI_POLL_WEIGHT); 806 } 807 808 adapter->num_active_rx_napi = adapter->req_rx_queues; 809 return 0; 810 } 811 812 static void release_napi(struct ibmvnic_adapter *adapter) 813 { 814 int i; 815 816 if (!adapter->napi) 817 return; 818 819 for (i = 0; i < adapter->num_active_rx_napi; i++) { 820 netdev_dbg(adapter->netdev, "Releasing napi[%d]\n", i); 821 netif_napi_del(&adapter->napi[i]); 822 } 823 824 kfree(adapter->napi); 825 adapter->napi = NULL; 826 adapter->num_active_rx_napi = 0; 827 adapter->napi_enabled = false; 828 } 829 830 static const char *adapter_state_to_string(enum vnic_state state) 831 { 832 switch (state) { 833 case VNIC_PROBING: 834 return "PROBING"; 835 case VNIC_PROBED: 836 return "PROBED"; 837 case VNIC_OPENING: 838 return "OPENING"; 839 case VNIC_OPEN: 840 return "OPEN"; 841 case VNIC_CLOSING: 842 return "CLOSING"; 843 case VNIC_CLOSED: 844 return "CLOSED"; 845 case VNIC_REMOVING: 846 return "REMOVING"; 847 case VNIC_REMOVED: 848 return "REMOVED"; 849 } 850 return "UNKNOWN"; 851 } 852 853 static int ibmvnic_login(struct net_device *netdev) 854 { 855 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 856 unsigned long timeout = msecs_to_jiffies(20000); 857 int retry_count = 0; 858 int retries = 10; 859 bool retry; 860 int rc; 861 862 do { 863 retry = false; 864 if (retry_count > retries) { 865 netdev_warn(netdev, "Login attempts exceeded\n"); 866 return -1; 867 } 868 869 adapter->init_done_rc = 0; 870 reinit_completion(&adapter->init_done); 871 rc = send_login(adapter); 872 if (rc) 873 return rc; 874 875 if (!wait_for_completion_timeout(&adapter->init_done, 876 timeout)) { 877 netdev_warn(netdev, "Login timed out, retrying...\n"); 878 retry = true; 879 adapter->init_done_rc = 0; 880 retry_count++; 881 continue; 882 } 883 884 if (adapter->init_done_rc == ABORTED) { 885 netdev_warn(netdev, "Login aborted, retrying...\n"); 886 retry = true; 887 adapter->init_done_rc = 0; 888 retry_count++; 889 /* FW or device may be busy, so 890 * wait a bit before retrying login 891 */ 892 msleep(500); 893 } else if (adapter->init_done_rc == PARTIALSUCCESS) { 894 retry_count++; 895 release_sub_crqs(adapter, 1); 896 897 retry = true; 898 netdev_dbg(netdev, 899 "Received partial success, retrying...\n"); 900 adapter->init_done_rc = 0; 901 reinit_completion(&adapter->init_done); 902 send_query_cap(adapter); 903 if (!wait_for_completion_timeout(&adapter->init_done, 904 timeout)) { 905 netdev_warn(netdev, 906 "Capabilities query timed out\n"); 907 return -1; 908 } 909 910 rc = init_sub_crqs(adapter); 911 if (rc) { 912 netdev_warn(netdev, 913 "SCRQ initialization failed\n"); 914 return -1; 915 } 916 917 rc = init_sub_crq_irqs(adapter); 918 if (rc) { 919 netdev_warn(netdev, 920 "SCRQ irq initialization failed\n"); 921 return -1; 922 } 923 } else if (adapter->init_done_rc) { 924 netdev_warn(netdev, "Adapter login failed\n"); 925 return -1; 926 } 927 } while (retry); 928 929 __ibmvnic_set_mac(netdev, adapter->mac_addr); 930 931 netdev_dbg(netdev, "[S:%s] Login succeeded\n", adapter_state_to_string(adapter->state)); 932 return 0; 933 } 934 935 static void release_login_buffer(struct ibmvnic_adapter *adapter) 936 { 937 kfree(adapter->login_buf); 938 adapter->login_buf = NULL; 939 } 940 941 static void release_login_rsp_buffer(struct ibmvnic_adapter *adapter) 942 { 943 kfree(adapter->login_rsp_buf); 944 adapter->login_rsp_buf = NULL; 945 } 946 947 static void release_resources(struct ibmvnic_adapter *adapter) 948 { 949 release_vpd_data(adapter); 950 951 release_tx_pools(adapter); 952 release_rx_pools(adapter); 953 954 release_napi(adapter); 955 release_login_buffer(adapter); 956 release_login_rsp_buffer(adapter); 957 } 958 959 static int set_link_state(struct ibmvnic_adapter *adapter, u8 link_state) 960 { 961 struct net_device *netdev = adapter->netdev; 962 unsigned long timeout = msecs_to_jiffies(20000); 963 union ibmvnic_crq crq; 964 bool resend; 965 int rc; 966 967 netdev_dbg(netdev, "setting link state %d\n", link_state); 968 969 memset(&crq, 0, sizeof(crq)); 970 crq.logical_link_state.first = IBMVNIC_CRQ_CMD; 971 crq.logical_link_state.cmd = LOGICAL_LINK_STATE; 972 crq.logical_link_state.link_state = link_state; 973 974 do { 975 resend = false; 976 977 reinit_completion(&adapter->init_done); 978 rc = ibmvnic_send_crq(adapter, &crq); 979 if (rc) { 980 netdev_err(netdev, "Failed to set link state\n"); 981 return rc; 982 } 983 984 if (!wait_for_completion_timeout(&adapter->init_done, 985 timeout)) { 986 netdev_err(netdev, "timeout setting link state\n"); 987 return -1; 988 } 989 990 if (adapter->init_done_rc == PARTIALSUCCESS) { 991 /* Partuial success, delay and re-send */ 992 mdelay(1000); 993 resend = true; 994 } else if (adapter->init_done_rc) { 995 netdev_warn(netdev, "Unable to set link state, rc=%d\n", 996 adapter->init_done_rc); 997 return adapter->init_done_rc; 998 } 999 } while (resend); 1000 1001 return 0; 1002 } 1003 1004 static int set_real_num_queues(struct net_device *netdev) 1005 { 1006 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 1007 int rc; 1008 1009 netdev_dbg(netdev, "Setting real tx/rx queues (%llx/%llx)\n", 1010 adapter->req_tx_queues, adapter->req_rx_queues); 1011 1012 rc = netif_set_real_num_tx_queues(netdev, adapter->req_tx_queues); 1013 if (rc) { 1014 netdev_err(netdev, "failed to set the number of tx queues\n"); 1015 return rc; 1016 } 1017 1018 rc = netif_set_real_num_rx_queues(netdev, adapter->req_rx_queues); 1019 if (rc) 1020 netdev_err(netdev, "failed to set the number of rx queues\n"); 1021 1022 return rc; 1023 } 1024 1025 static int ibmvnic_get_vpd(struct ibmvnic_adapter *adapter) 1026 { 1027 struct device *dev = &adapter->vdev->dev; 1028 union ibmvnic_crq crq; 1029 int len = 0; 1030 int rc; 1031 1032 if (adapter->vpd->buff) 1033 len = adapter->vpd->len; 1034 1035 mutex_lock(&adapter->fw_lock); 1036 adapter->fw_done_rc = 0; 1037 reinit_completion(&adapter->fw_done); 1038 1039 crq.get_vpd_size.first = IBMVNIC_CRQ_CMD; 1040 crq.get_vpd_size.cmd = GET_VPD_SIZE; 1041 rc = ibmvnic_send_crq(adapter, &crq); 1042 if (rc) { 1043 mutex_unlock(&adapter->fw_lock); 1044 return rc; 1045 } 1046 1047 rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000); 1048 if (rc) { 1049 dev_err(dev, "Could not retrieve VPD size, rc = %d\n", rc); 1050 mutex_unlock(&adapter->fw_lock); 1051 return rc; 1052 } 1053 mutex_unlock(&adapter->fw_lock); 1054 1055 if (!adapter->vpd->len) 1056 return -ENODATA; 1057 1058 if (!adapter->vpd->buff) 1059 adapter->vpd->buff = kzalloc(adapter->vpd->len, GFP_KERNEL); 1060 else if (adapter->vpd->len != len) 1061 adapter->vpd->buff = 1062 krealloc(adapter->vpd->buff, 1063 adapter->vpd->len, GFP_KERNEL); 1064 1065 if (!adapter->vpd->buff) { 1066 dev_err(dev, "Could allocate VPD buffer\n"); 1067 return -ENOMEM; 1068 } 1069 1070 adapter->vpd->dma_addr = 1071 dma_map_single(dev, adapter->vpd->buff, adapter->vpd->len, 1072 DMA_FROM_DEVICE); 1073 if (dma_mapping_error(dev, adapter->vpd->dma_addr)) { 1074 dev_err(dev, "Could not map VPD buffer\n"); 1075 kfree(adapter->vpd->buff); 1076 adapter->vpd->buff = NULL; 1077 return -ENOMEM; 1078 } 1079 1080 mutex_lock(&adapter->fw_lock); 1081 adapter->fw_done_rc = 0; 1082 reinit_completion(&adapter->fw_done); 1083 1084 crq.get_vpd.first = IBMVNIC_CRQ_CMD; 1085 crq.get_vpd.cmd = GET_VPD; 1086 crq.get_vpd.ioba = cpu_to_be32(adapter->vpd->dma_addr); 1087 crq.get_vpd.len = cpu_to_be32((u32)adapter->vpd->len); 1088 rc = ibmvnic_send_crq(adapter, &crq); 1089 if (rc) { 1090 kfree(adapter->vpd->buff); 1091 adapter->vpd->buff = NULL; 1092 mutex_unlock(&adapter->fw_lock); 1093 return rc; 1094 } 1095 1096 rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000); 1097 if (rc) { 1098 dev_err(dev, "Unable to retrieve VPD, rc = %d\n", rc); 1099 kfree(adapter->vpd->buff); 1100 adapter->vpd->buff = NULL; 1101 mutex_unlock(&adapter->fw_lock); 1102 return rc; 1103 } 1104 1105 mutex_unlock(&adapter->fw_lock); 1106 return 0; 1107 } 1108 1109 static int init_resources(struct ibmvnic_adapter *adapter) 1110 { 1111 struct net_device *netdev = adapter->netdev; 1112 int rc; 1113 1114 rc = set_real_num_queues(netdev); 1115 if (rc) 1116 return rc; 1117 1118 adapter->vpd = kzalloc(sizeof(*adapter->vpd), GFP_KERNEL); 1119 if (!adapter->vpd) 1120 return -ENOMEM; 1121 1122 /* Vital Product Data (VPD) */ 1123 rc = ibmvnic_get_vpd(adapter); 1124 if (rc) { 1125 netdev_err(netdev, "failed to initialize Vital Product Data (VPD)\n"); 1126 return rc; 1127 } 1128 1129 adapter->map_id = 1; 1130 1131 rc = init_napi(adapter); 1132 if (rc) 1133 return rc; 1134 1135 send_query_map(adapter); 1136 1137 rc = init_rx_pools(netdev); 1138 if (rc) 1139 return rc; 1140 1141 rc = init_tx_pools(netdev); 1142 return rc; 1143 } 1144 1145 static int __ibmvnic_open(struct net_device *netdev) 1146 { 1147 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 1148 enum vnic_state prev_state = adapter->state; 1149 int i, rc; 1150 1151 adapter->state = VNIC_OPENING; 1152 replenish_pools(adapter); 1153 ibmvnic_napi_enable(adapter); 1154 1155 /* We're ready to receive frames, enable the sub-crq interrupts and 1156 * set the logical link state to up 1157 */ 1158 for (i = 0; i < adapter->req_rx_queues; i++) { 1159 netdev_dbg(netdev, "Enabling rx_scrq[%d] irq\n", i); 1160 if (prev_state == VNIC_CLOSED) 1161 enable_irq(adapter->rx_scrq[i]->irq); 1162 enable_scrq_irq(adapter, adapter->rx_scrq[i]); 1163 } 1164 1165 for (i = 0; i < adapter->req_tx_queues; i++) { 1166 netdev_dbg(netdev, "Enabling tx_scrq[%d] irq\n", i); 1167 if (prev_state == VNIC_CLOSED) 1168 enable_irq(adapter->tx_scrq[i]->irq); 1169 enable_scrq_irq(adapter, adapter->tx_scrq[i]); 1170 netdev_tx_reset_queue(netdev_get_tx_queue(netdev, i)); 1171 } 1172 1173 rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_UP); 1174 if (rc) { 1175 ibmvnic_napi_disable(adapter); 1176 release_resources(adapter); 1177 return rc; 1178 } 1179 1180 netif_tx_start_all_queues(netdev); 1181 1182 adapter->state = VNIC_OPEN; 1183 return rc; 1184 } 1185 1186 static int ibmvnic_open(struct net_device *netdev) 1187 { 1188 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 1189 int rc; 1190 1191 ASSERT_RTNL(); 1192 1193 /* If device failover is pending or we are about to reset, just set 1194 * device state and return. Device operation will be handled by reset 1195 * routine. 1196 * 1197 * It should be safe to overwrite the adapter->state here. Since 1198 * we hold the rtnl, either the reset has not actually started or 1199 * the rtnl got dropped during the set_link_state() in do_reset(). 1200 * In the former case, no one else is changing the state (again we 1201 * have the rtnl) and in the latter case, do_reset() will detect and 1202 * honor our setting below. 1203 */ 1204 if (adapter->failover_pending || (test_bit(0, &adapter->resetting))) { 1205 netdev_dbg(netdev, "[S:%s FOP:%d] Resetting, deferring open\n", 1206 adapter_state_to_string(adapter->state), 1207 adapter->failover_pending); 1208 adapter->state = VNIC_OPEN; 1209 rc = 0; 1210 goto out; 1211 } 1212 1213 if (adapter->state != VNIC_CLOSED) { 1214 rc = ibmvnic_login(netdev); 1215 if (rc) 1216 goto out; 1217 1218 rc = init_resources(adapter); 1219 if (rc) { 1220 netdev_err(netdev, "failed to initialize resources\n"); 1221 release_resources(adapter); 1222 goto out; 1223 } 1224 } 1225 1226 rc = __ibmvnic_open(netdev); 1227 1228 out: 1229 /* If open failed and there is a pending failover or in-progress reset, 1230 * set device state and return. Device operation will be handled by 1231 * reset routine. See also comments above regarding rtnl. 1232 */ 1233 if (rc && 1234 (adapter->failover_pending || (test_bit(0, &adapter->resetting)))) { 1235 adapter->state = VNIC_OPEN; 1236 rc = 0; 1237 } 1238 return rc; 1239 } 1240 1241 static void clean_rx_pools(struct ibmvnic_adapter *adapter) 1242 { 1243 struct ibmvnic_rx_pool *rx_pool; 1244 struct ibmvnic_rx_buff *rx_buff; 1245 u64 rx_entries; 1246 int rx_scrqs; 1247 int i, j; 1248 1249 if (!adapter->rx_pool) 1250 return; 1251 1252 rx_scrqs = adapter->num_active_rx_pools; 1253 rx_entries = adapter->req_rx_add_entries_per_subcrq; 1254 1255 /* Free any remaining skbs in the rx buffer pools */ 1256 for (i = 0; i < rx_scrqs; i++) { 1257 rx_pool = &adapter->rx_pool[i]; 1258 if (!rx_pool || !rx_pool->rx_buff) 1259 continue; 1260 1261 netdev_dbg(adapter->netdev, "Cleaning rx_pool[%d]\n", i); 1262 for (j = 0; j < rx_entries; j++) { 1263 rx_buff = &rx_pool->rx_buff[j]; 1264 if (rx_buff && rx_buff->skb) { 1265 dev_kfree_skb_any(rx_buff->skb); 1266 rx_buff->skb = NULL; 1267 } 1268 } 1269 } 1270 } 1271 1272 static void clean_one_tx_pool(struct ibmvnic_adapter *adapter, 1273 struct ibmvnic_tx_pool *tx_pool) 1274 { 1275 struct ibmvnic_tx_buff *tx_buff; 1276 u64 tx_entries; 1277 int i; 1278 1279 if (!tx_pool || !tx_pool->tx_buff) 1280 return; 1281 1282 tx_entries = tx_pool->num_buffers; 1283 1284 for (i = 0; i < tx_entries; i++) { 1285 tx_buff = &tx_pool->tx_buff[i]; 1286 if (tx_buff && tx_buff->skb) { 1287 dev_kfree_skb_any(tx_buff->skb); 1288 tx_buff->skb = NULL; 1289 } 1290 } 1291 } 1292 1293 static void clean_tx_pools(struct ibmvnic_adapter *adapter) 1294 { 1295 int tx_scrqs; 1296 int i; 1297 1298 if (!adapter->tx_pool || !adapter->tso_pool) 1299 return; 1300 1301 tx_scrqs = adapter->num_active_tx_pools; 1302 1303 /* Free any remaining skbs in the tx buffer pools */ 1304 for (i = 0; i < tx_scrqs; i++) { 1305 netdev_dbg(adapter->netdev, "Cleaning tx_pool[%d]\n", i); 1306 clean_one_tx_pool(adapter, &adapter->tx_pool[i]); 1307 clean_one_tx_pool(adapter, &adapter->tso_pool[i]); 1308 } 1309 } 1310 1311 static void ibmvnic_disable_irqs(struct ibmvnic_adapter *adapter) 1312 { 1313 struct net_device *netdev = adapter->netdev; 1314 int i; 1315 1316 if (adapter->tx_scrq) { 1317 for (i = 0; i < adapter->req_tx_queues; i++) 1318 if (adapter->tx_scrq[i]->irq) { 1319 netdev_dbg(netdev, 1320 "Disabling tx_scrq[%d] irq\n", i); 1321 disable_scrq_irq(adapter, adapter->tx_scrq[i]); 1322 disable_irq(adapter->tx_scrq[i]->irq); 1323 } 1324 } 1325 1326 if (adapter->rx_scrq) { 1327 for (i = 0; i < adapter->req_rx_queues; i++) { 1328 if (adapter->rx_scrq[i]->irq) { 1329 netdev_dbg(netdev, 1330 "Disabling rx_scrq[%d] irq\n", i); 1331 disable_scrq_irq(adapter, adapter->rx_scrq[i]); 1332 disable_irq(adapter->rx_scrq[i]->irq); 1333 } 1334 } 1335 } 1336 } 1337 1338 static void ibmvnic_cleanup(struct net_device *netdev) 1339 { 1340 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 1341 1342 /* ensure that transmissions are stopped if called by do_reset */ 1343 if (test_bit(0, &adapter->resetting)) 1344 netif_tx_disable(netdev); 1345 else 1346 netif_tx_stop_all_queues(netdev); 1347 1348 ibmvnic_napi_disable(adapter); 1349 ibmvnic_disable_irqs(adapter); 1350 1351 clean_rx_pools(adapter); 1352 clean_tx_pools(adapter); 1353 } 1354 1355 static int __ibmvnic_close(struct net_device *netdev) 1356 { 1357 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 1358 int rc = 0; 1359 1360 adapter->state = VNIC_CLOSING; 1361 rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN); 1362 adapter->state = VNIC_CLOSED; 1363 return rc; 1364 } 1365 1366 static int ibmvnic_close(struct net_device *netdev) 1367 { 1368 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 1369 int rc; 1370 1371 netdev_dbg(netdev, "[S:%s FOP:%d FRR:%d] Closing\n", 1372 adapter_state_to_string(adapter->state), 1373 adapter->failover_pending, 1374 adapter->force_reset_recovery); 1375 1376 /* If device failover is pending, just set device state and return. 1377 * Device operation will be handled by reset routine. 1378 */ 1379 if (adapter->failover_pending) { 1380 adapter->state = VNIC_CLOSED; 1381 return 0; 1382 } 1383 1384 rc = __ibmvnic_close(netdev); 1385 ibmvnic_cleanup(netdev); 1386 1387 return rc; 1388 } 1389 1390 /** 1391 * build_hdr_data - creates L2/L3/L4 header data buffer 1392 * @hdr_field: bitfield determining needed headers 1393 * @skb: socket buffer 1394 * @hdr_len: array of header lengths 1395 * @hdr_data: buffer to write the header to 1396 * 1397 * Reads hdr_field to determine which headers are needed by firmware. 1398 * Builds a buffer containing these headers. Saves individual header 1399 * lengths and total buffer length to be used to build descriptors. 1400 */ 1401 static int build_hdr_data(u8 hdr_field, struct sk_buff *skb, 1402 int *hdr_len, u8 *hdr_data) 1403 { 1404 int len = 0; 1405 u8 *hdr; 1406 1407 if (skb_vlan_tagged(skb) && !skb_vlan_tag_present(skb)) 1408 hdr_len[0] = sizeof(struct vlan_ethhdr); 1409 else 1410 hdr_len[0] = sizeof(struct ethhdr); 1411 1412 if (skb->protocol == htons(ETH_P_IP)) { 1413 hdr_len[1] = ip_hdr(skb)->ihl * 4; 1414 if (ip_hdr(skb)->protocol == IPPROTO_TCP) 1415 hdr_len[2] = tcp_hdrlen(skb); 1416 else if (ip_hdr(skb)->protocol == IPPROTO_UDP) 1417 hdr_len[2] = sizeof(struct udphdr); 1418 } else if (skb->protocol == htons(ETH_P_IPV6)) { 1419 hdr_len[1] = sizeof(struct ipv6hdr); 1420 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP) 1421 hdr_len[2] = tcp_hdrlen(skb); 1422 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP) 1423 hdr_len[2] = sizeof(struct udphdr); 1424 } else if (skb->protocol == htons(ETH_P_ARP)) { 1425 hdr_len[1] = arp_hdr_len(skb->dev); 1426 hdr_len[2] = 0; 1427 } 1428 1429 memset(hdr_data, 0, 120); 1430 if ((hdr_field >> 6) & 1) { 1431 hdr = skb_mac_header(skb); 1432 memcpy(hdr_data, hdr, hdr_len[0]); 1433 len += hdr_len[0]; 1434 } 1435 1436 if ((hdr_field >> 5) & 1) { 1437 hdr = skb_network_header(skb); 1438 memcpy(hdr_data + len, hdr, hdr_len[1]); 1439 len += hdr_len[1]; 1440 } 1441 1442 if ((hdr_field >> 4) & 1) { 1443 hdr = skb_transport_header(skb); 1444 memcpy(hdr_data + len, hdr, hdr_len[2]); 1445 len += hdr_len[2]; 1446 } 1447 return len; 1448 } 1449 1450 /** 1451 * create_hdr_descs - create header and header extension descriptors 1452 * @hdr_field: bitfield determining needed headers 1453 * @hdr_data: buffer containing header data 1454 * @len: length of data buffer 1455 * @hdr_len: array of individual header lengths 1456 * @scrq_arr: descriptor array 1457 * 1458 * Creates header and, if needed, header extension descriptors and 1459 * places them in a descriptor array, scrq_arr 1460 */ 1461 1462 static int create_hdr_descs(u8 hdr_field, u8 *hdr_data, int len, int *hdr_len, 1463 union sub_crq *scrq_arr) 1464 { 1465 union sub_crq hdr_desc; 1466 int tmp_len = len; 1467 int num_descs = 0; 1468 u8 *data, *cur; 1469 int tmp; 1470 1471 while (tmp_len > 0) { 1472 cur = hdr_data + len - tmp_len; 1473 1474 memset(&hdr_desc, 0, sizeof(hdr_desc)); 1475 if (cur != hdr_data) { 1476 data = hdr_desc.hdr_ext.data; 1477 tmp = tmp_len > 29 ? 29 : tmp_len; 1478 hdr_desc.hdr_ext.first = IBMVNIC_CRQ_CMD; 1479 hdr_desc.hdr_ext.type = IBMVNIC_HDR_EXT_DESC; 1480 hdr_desc.hdr_ext.len = tmp; 1481 } else { 1482 data = hdr_desc.hdr.data; 1483 tmp = tmp_len > 24 ? 24 : tmp_len; 1484 hdr_desc.hdr.first = IBMVNIC_CRQ_CMD; 1485 hdr_desc.hdr.type = IBMVNIC_HDR_DESC; 1486 hdr_desc.hdr.len = tmp; 1487 hdr_desc.hdr.l2_len = (u8)hdr_len[0]; 1488 hdr_desc.hdr.l3_len = cpu_to_be16((u16)hdr_len[1]); 1489 hdr_desc.hdr.l4_len = (u8)hdr_len[2]; 1490 hdr_desc.hdr.flag = hdr_field << 1; 1491 } 1492 memcpy(data, cur, tmp); 1493 tmp_len -= tmp; 1494 *scrq_arr = hdr_desc; 1495 scrq_arr++; 1496 num_descs++; 1497 } 1498 1499 return num_descs; 1500 } 1501 1502 /** 1503 * build_hdr_descs_arr - build a header descriptor array 1504 * @txbuff: tx buffer 1505 * @num_entries: number of descriptors to be sent 1506 * @hdr_field: bit field determining which headers will be sent 1507 * 1508 * This function will build a TX descriptor array with applicable 1509 * L2/L3/L4 packet header descriptors to be sent by send_subcrq_indirect. 1510 */ 1511 1512 static void build_hdr_descs_arr(struct sk_buff *skb, 1513 union sub_crq *indir_arr, 1514 int *num_entries, u8 hdr_field) 1515 { 1516 int hdr_len[3] = {0, 0, 0}; 1517 u8 hdr_data[140] = {0}; 1518 int tot_len; 1519 1520 tot_len = build_hdr_data(hdr_field, skb, hdr_len, 1521 hdr_data); 1522 *num_entries += create_hdr_descs(hdr_field, hdr_data, tot_len, hdr_len, 1523 indir_arr + 1); 1524 } 1525 1526 static int ibmvnic_xmit_workarounds(struct sk_buff *skb, 1527 struct net_device *netdev) 1528 { 1529 /* For some backing devices, mishandling of small packets 1530 * can result in a loss of connection or TX stall. Device 1531 * architects recommend that no packet should be smaller 1532 * than the minimum MTU value provided to the driver, so 1533 * pad any packets to that length 1534 */ 1535 if (skb->len < netdev->min_mtu) 1536 return skb_put_padto(skb, netdev->min_mtu); 1537 1538 return 0; 1539 } 1540 1541 static void ibmvnic_tx_scrq_clean_buffer(struct ibmvnic_adapter *adapter, 1542 struct ibmvnic_sub_crq_queue *tx_scrq) 1543 { 1544 struct ibmvnic_ind_xmit_queue *ind_bufp; 1545 struct ibmvnic_tx_buff *tx_buff; 1546 struct ibmvnic_tx_pool *tx_pool; 1547 union sub_crq tx_scrq_entry; 1548 int queue_num; 1549 int entries; 1550 int index; 1551 int i; 1552 1553 ind_bufp = &tx_scrq->ind_buf; 1554 entries = (u64)ind_bufp->index; 1555 queue_num = tx_scrq->pool_index; 1556 1557 for (i = entries - 1; i >= 0; --i) { 1558 tx_scrq_entry = ind_bufp->indir_arr[i]; 1559 if (tx_scrq_entry.v1.type != IBMVNIC_TX_DESC) 1560 continue; 1561 index = be32_to_cpu(tx_scrq_entry.v1.correlator); 1562 if (index & IBMVNIC_TSO_POOL_MASK) { 1563 tx_pool = &adapter->tso_pool[queue_num]; 1564 index &= ~IBMVNIC_TSO_POOL_MASK; 1565 } else { 1566 tx_pool = &adapter->tx_pool[queue_num]; 1567 } 1568 tx_pool->free_map[tx_pool->consumer_index] = index; 1569 tx_pool->consumer_index = tx_pool->consumer_index == 0 ? 1570 tx_pool->num_buffers - 1 : 1571 tx_pool->consumer_index - 1; 1572 tx_buff = &tx_pool->tx_buff[index]; 1573 adapter->netdev->stats.tx_packets--; 1574 adapter->netdev->stats.tx_bytes -= tx_buff->skb->len; 1575 adapter->tx_stats_buffers[queue_num].packets--; 1576 adapter->tx_stats_buffers[queue_num].bytes -= 1577 tx_buff->skb->len; 1578 dev_kfree_skb_any(tx_buff->skb); 1579 tx_buff->skb = NULL; 1580 adapter->netdev->stats.tx_dropped++; 1581 } 1582 ind_bufp->index = 0; 1583 if (atomic_sub_return(entries, &tx_scrq->used) <= 1584 (adapter->req_tx_entries_per_subcrq / 2) && 1585 __netif_subqueue_stopped(adapter->netdev, queue_num)) { 1586 netif_wake_subqueue(adapter->netdev, queue_num); 1587 netdev_dbg(adapter->netdev, "Started queue %d\n", 1588 queue_num); 1589 } 1590 } 1591 1592 static int ibmvnic_tx_scrq_flush(struct ibmvnic_adapter *adapter, 1593 struct ibmvnic_sub_crq_queue *tx_scrq) 1594 { 1595 struct ibmvnic_ind_xmit_queue *ind_bufp; 1596 u64 dma_addr; 1597 u64 entries; 1598 u64 handle; 1599 int rc; 1600 1601 ind_bufp = &tx_scrq->ind_buf; 1602 dma_addr = (u64)ind_bufp->indir_dma; 1603 entries = (u64)ind_bufp->index; 1604 handle = tx_scrq->handle; 1605 1606 if (!entries) 1607 return 0; 1608 rc = send_subcrq_indirect(adapter, handle, dma_addr, entries); 1609 if (rc) 1610 ibmvnic_tx_scrq_clean_buffer(adapter, tx_scrq); 1611 else 1612 ind_bufp->index = 0; 1613 return 0; 1614 } 1615 1616 static netdev_tx_t ibmvnic_xmit(struct sk_buff *skb, struct net_device *netdev) 1617 { 1618 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 1619 int queue_num = skb_get_queue_mapping(skb); 1620 u8 *hdrs = (u8 *)&adapter->tx_rx_desc_req; 1621 struct device *dev = &adapter->vdev->dev; 1622 struct ibmvnic_ind_xmit_queue *ind_bufp; 1623 struct ibmvnic_tx_buff *tx_buff = NULL; 1624 struct ibmvnic_sub_crq_queue *tx_scrq; 1625 struct ibmvnic_tx_pool *tx_pool; 1626 unsigned int tx_send_failed = 0; 1627 netdev_tx_t ret = NETDEV_TX_OK; 1628 unsigned int tx_map_failed = 0; 1629 union sub_crq indir_arr[16]; 1630 unsigned int tx_dropped = 0; 1631 unsigned int tx_packets = 0; 1632 unsigned int tx_bytes = 0; 1633 dma_addr_t data_dma_addr; 1634 struct netdev_queue *txq; 1635 unsigned long lpar_rc; 1636 union sub_crq tx_crq; 1637 unsigned int offset; 1638 int num_entries = 1; 1639 unsigned char *dst; 1640 int index = 0; 1641 u8 proto = 0; 1642 1643 tx_scrq = adapter->tx_scrq[queue_num]; 1644 txq = netdev_get_tx_queue(netdev, queue_num); 1645 ind_bufp = &tx_scrq->ind_buf; 1646 1647 if (test_bit(0, &adapter->resetting)) { 1648 if (!netif_subqueue_stopped(netdev, skb)) 1649 netif_stop_subqueue(netdev, queue_num); 1650 dev_kfree_skb_any(skb); 1651 1652 tx_send_failed++; 1653 tx_dropped++; 1654 ret = NETDEV_TX_OK; 1655 ibmvnic_tx_scrq_flush(adapter, tx_scrq); 1656 goto out; 1657 } 1658 1659 if (ibmvnic_xmit_workarounds(skb, netdev)) { 1660 tx_dropped++; 1661 tx_send_failed++; 1662 ret = NETDEV_TX_OK; 1663 ibmvnic_tx_scrq_flush(adapter, tx_scrq); 1664 goto out; 1665 } 1666 if (skb_is_gso(skb)) 1667 tx_pool = &adapter->tso_pool[queue_num]; 1668 else 1669 tx_pool = &adapter->tx_pool[queue_num]; 1670 1671 index = tx_pool->free_map[tx_pool->consumer_index]; 1672 1673 if (index == IBMVNIC_INVALID_MAP) { 1674 dev_kfree_skb_any(skb); 1675 tx_send_failed++; 1676 tx_dropped++; 1677 ret = NETDEV_TX_OK; 1678 ibmvnic_tx_scrq_flush(adapter, tx_scrq); 1679 goto out; 1680 } 1681 1682 tx_pool->free_map[tx_pool->consumer_index] = IBMVNIC_INVALID_MAP; 1683 1684 offset = index * tx_pool->buf_size; 1685 dst = tx_pool->long_term_buff.buff + offset; 1686 memset(dst, 0, tx_pool->buf_size); 1687 data_dma_addr = tx_pool->long_term_buff.addr + offset; 1688 1689 if (skb_shinfo(skb)->nr_frags) { 1690 int cur, i; 1691 1692 /* Copy the head */ 1693 skb_copy_from_linear_data(skb, dst, skb_headlen(skb)); 1694 cur = skb_headlen(skb); 1695 1696 /* Copy the frags */ 1697 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 1698 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 1699 1700 memcpy(dst + cur, skb_frag_address(frag), 1701 skb_frag_size(frag)); 1702 cur += skb_frag_size(frag); 1703 } 1704 } else { 1705 skb_copy_from_linear_data(skb, dst, skb->len); 1706 } 1707 1708 /* post changes to long_term_buff *dst before VIOS accessing it */ 1709 dma_wmb(); 1710 1711 tx_pool->consumer_index = 1712 (tx_pool->consumer_index + 1) % tx_pool->num_buffers; 1713 1714 tx_buff = &tx_pool->tx_buff[index]; 1715 tx_buff->skb = skb; 1716 tx_buff->index = index; 1717 tx_buff->pool_index = queue_num; 1718 1719 memset(&tx_crq, 0, sizeof(tx_crq)); 1720 tx_crq.v1.first = IBMVNIC_CRQ_CMD; 1721 tx_crq.v1.type = IBMVNIC_TX_DESC; 1722 tx_crq.v1.n_crq_elem = 1; 1723 tx_crq.v1.n_sge = 1; 1724 tx_crq.v1.flags1 = IBMVNIC_TX_COMP_NEEDED; 1725 1726 if (skb_is_gso(skb)) 1727 tx_crq.v1.correlator = 1728 cpu_to_be32(index | IBMVNIC_TSO_POOL_MASK); 1729 else 1730 tx_crq.v1.correlator = cpu_to_be32(index); 1731 tx_crq.v1.dma_reg = cpu_to_be16(tx_pool->long_term_buff.map_id); 1732 tx_crq.v1.sge_len = cpu_to_be32(skb->len); 1733 tx_crq.v1.ioba = cpu_to_be64(data_dma_addr); 1734 1735 if (adapter->vlan_header_insertion && skb_vlan_tag_present(skb)) { 1736 tx_crq.v1.flags2 |= IBMVNIC_TX_VLAN_INSERT; 1737 tx_crq.v1.vlan_id = cpu_to_be16(skb->vlan_tci); 1738 } 1739 1740 if (skb->protocol == htons(ETH_P_IP)) { 1741 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV4; 1742 proto = ip_hdr(skb)->protocol; 1743 } else if (skb->protocol == htons(ETH_P_IPV6)) { 1744 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV6; 1745 proto = ipv6_hdr(skb)->nexthdr; 1746 } 1747 1748 if (proto == IPPROTO_TCP) 1749 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_TCP; 1750 else if (proto == IPPROTO_UDP) 1751 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_UDP; 1752 1753 if (skb->ip_summed == CHECKSUM_PARTIAL) { 1754 tx_crq.v1.flags1 |= IBMVNIC_TX_CHKSUM_OFFLOAD; 1755 hdrs += 2; 1756 } 1757 if (skb_is_gso(skb)) { 1758 tx_crq.v1.flags1 |= IBMVNIC_TX_LSO; 1759 tx_crq.v1.mss = cpu_to_be16(skb_shinfo(skb)->gso_size); 1760 hdrs += 2; 1761 } 1762 1763 if ((*hdrs >> 7) & 1) 1764 build_hdr_descs_arr(skb, indir_arr, &num_entries, *hdrs); 1765 1766 tx_crq.v1.n_crq_elem = num_entries; 1767 tx_buff->num_entries = num_entries; 1768 /* flush buffer if current entry can not fit */ 1769 if (num_entries + ind_bufp->index > IBMVNIC_MAX_IND_DESCS) { 1770 lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq); 1771 if (lpar_rc != H_SUCCESS) 1772 goto tx_flush_err; 1773 } 1774 1775 indir_arr[0] = tx_crq; 1776 memcpy(&ind_bufp->indir_arr[ind_bufp->index], &indir_arr[0], 1777 num_entries * sizeof(struct ibmvnic_generic_scrq)); 1778 ind_bufp->index += num_entries; 1779 if (__netdev_tx_sent_queue(txq, skb->len, 1780 netdev_xmit_more() && 1781 ind_bufp->index < IBMVNIC_MAX_IND_DESCS)) { 1782 lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq); 1783 if (lpar_rc != H_SUCCESS) 1784 goto tx_err; 1785 } 1786 1787 if (atomic_add_return(num_entries, &tx_scrq->used) 1788 >= adapter->req_tx_entries_per_subcrq) { 1789 netdev_dbg(netdev, "Stopping queue %d\n", queue_num); 1790 netif_stop_subqueue(netdev, queue_num); 1791 } 1792 1793 tx_packets++; 1794 tx_bytes += skb->len; 1795 txq->trans_start = jiffies; 1796 ret = NETDEV_TX_OK; 1797 goto out; 1798 1799 tx_flush_err: 1800 dev_kfree_skb_any(skb); 1801 tx_buff->skb = NULL; 1802 tx_pool->consumer_index = tx_pool->consumer_index == 0 ? 1803 tx_pool->num_buffers - 1 : 1804 tx_pool->consumer_index - 1; 1805 tx_dropped++; 1806 tx_err: 1807 if (lpar_rc != H_CLOSED && lpar_rc != H_PARAMETER) 1808 dev_err_ratelimited(dev, "tx: send failed\n"); 1809 1810 if (lpar_rc == H_CLOSED || adapter->failover_pending) { 1811 /* Disable TX and report carrier off if queue is closed 1812 * or pending failover. 1813 * Firmware guarantees that a signal will be sent to the 1814 * driver, triggering a reset or some other action. 1815 */ 1816 netif_tx_stop_all_queues(netdev); 1817 netif_carrier_off(netdev); 1818 } 1819 out: 1820 netdev->stats.tx_dropped += tx_dropped; 1821 netdev->stats.tx_bytes += tx_bytes; 1822 netdev->stats.tx_packets += tx_packets; 1823 adapter->tx_send_failed += tx_send_failed; 1824 adapter->tx_map_failed += tx_map_failed; 1825 adapter->tx_stats_buffers[queue_num].packets += tx_packets; 1826 adapter->tx_stats_buffers[queue_num].bytes += tx_bytes; 1827 adapter->tx_stats_buffers[queue_num].dropped_packets += tx_dropped; 1828 1829 return ret; 1830 } 1831 1832 static void ibmvnic_set_multi(struct net_device *netdev) 1833 { 1834 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 1835 struct netdev_hw_addr *ha; 1836 union ibmvnic_crq crq; 1837 1838 memset(&crq, 0, sizeof(crq)); 1839 crq.request_capability.first = IBMVNIC_CRQ_CMD; 1840 crq.request_capability.cmd = REQUEST_CAPABILITY; 1841 1842 if (netdev->flags & IFF_PROMISC) { 1843 if (!adapter->promisc_supported) 1844 return; 1845 } else { 1846 if (netdev->flags & IFF_ALLMULTI) { 1847 /* Accept all multicast */ 1848 memset(&crq, 0, sizeof(crq)); 1849 crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD; 1850 crq.multicast_ctrl.cmd = MULTICAST_CTRL; 1851 crq.multicast_ctrl.flags = IBMVNIC_ENABLE_ALL; 1852 ibmvnic_send_crq(adapter, &crq); 1853 } else if (netdev_mc_empty(netdev)) { 1854 /* Reject all multicast */ 1855 memset(&crq, 0, sizeof(crq)); 1856 crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD; 1857 crq.multicast_ctrl.cmd = MULTICAST_CTRL; 1858 crq.multicast_ctrl.flags = IBMVNIC_DISABLE_ALL; 1859 ibmvnic_send_crq(adapter, &crq); 1860 } else { 1861 /* Accept one or more multicast(s) */ 1862 netdev_for_each_mc_addr(ha, netdev) { 1863 memset(&crq, 0, sizeof(crq)); 1864 crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD; 1865 crq.multicast_ctrl.cmd = MULTICAST_CTRL; 1866 crq.multicast_ctrl.flags = IBMVNIC_ENABLE_MC; 1867 ether_addr_copy(&crq.multicast_ctrl.mac_addr[0], 1868 ha->addr); 1869 ibmvnic_send_crq(adapter, &crq); 1870 } 1871 } 1872 } 1873 } 1874 1875 static int __ibmvnic_set_mac(struct net_device *netdev, u8 *dev_addr) 1876 { 1877 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 1878 union ibmvnic_crq crq; 1879 int rc; 1880 1881 if (!is_valid_ether_addr(dev_addr)) { 1882 rc = -EADDRNOTAVAIL; 1883 goto err; 1884 } 1885 1886 memset(&crq, 0, sizeof(crq)); 1887 crq.change_mac_addr.first = IBMVNIC_CRQ_CMD; 1888 crq.change_mac_addr.cmd = CHANGE_MAC_ADDR; 1889 ether_addr_copy(&crq.change_mac_addr.mac_addr[0], dev_addr); 1890 1891 mutex_lock(&adapter->fw_lock); 1892 adapter->fw_done_rc = 0; 1893 reinit_completion(&adapter->fw_done); 1894 1895 rc = ibmvnic_send_crq(adapter, &crq); 1896 if (rc) { 1897 rc = -EIO; 1898 mutex_unlock(&adapter->fw_lock); 1899 goto err; 1900 } 1901 1902 rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000); 1903 /* netdev->dev_addr is changed in handle_change_mac_rsp function */ 1904 if (rc || adapter->fw_done_rc) { 1905 rc = -EIO; 1906 mutex_unlock(&adapter->fw_lock); 1907 goto err; 1908 } 1909 mutex_unlock(&adapter->fw_lock); 1910 return 0; 1911 err: 1912 ether_addr_copy(adapter->mac_addr, netdev->dev_addr); 1913 return rc; 1914 } 1915 1916 static int ibmvnic_set_mac(struct net_device *netdev, void *p) 1917 { 1918 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 1919 struct sockaddr *addr = p; 1920 int rc; 1921 1922 rc = 0; 1923 if (!is_valid_ether_addr(addr->sa_data)) 1924 return -EADDRNOTAVAIL; 1925 1926 ether_addr_copy(adapter->mac_addr, addr->sa_data); 1927 if (adapter->state != VNIC_PROBED) 1928 rc = __ibmvnic_set_mac(netdev, addr->sa_data); 1929 1930 return rc; 1931 } 1932 1933 static const char *reset_reason_to_string(enum ibmvnic_reset_reason reason) 1934 { 1935 switch (reason) { 1936 case VNIC_RESET_FAILOVER: 1937 return "FAILOVER"; 1938 case VNIC_RESET_MOBILITY: 1939 return "MOBILITY"; 1940 case VNIC_RESET_FATAL: 1941 return "FATAL"; 1942 case VNIC_RESET_NON_FATAL: 1943 return "NON_FATAL"; 1944 case VNIC_RESET_TIMEOUT: 1945 return "TIMEOUT"; 1946 case VNIC_RESET_CHANGE_PARAM: 1947 return "CHANGE_PARAM"; 1948 } 1949 return "UNKNOWN"; 1950 } 1951 1952 /* 1953 * do_reset returns zero if we are able to keep processing reset events, or 1954 * non-zero if we hit a fatal error and must halt. 1955 */ 1956 static int do_reset(struct ibmvnic_adapter *adapter, 1957 struct ibmvnic_rwi *rwi, u32 reset_state) 1958 { 1959 u64 old_num_rx_queues, old_num_tx_queues; 1960 u64 old_num_rx_slots, old_num_tx_slots; 1961 struct net_device *netdev = adapter->netdev; 1962 int rc; 1963 1964 netdev_dbg(adapter->netdev, 1965 "[S:%s FOP:%d] Reset reason: %s, reset_state: %s\n", 1966 adapter_state_to_string(adapter->state), 1967 adapter->failover_pending, 1968 reset_reason_to_string(rwi->reset_reason), 1969 adapter_state_to_string(reset_state)); 1970 1971 adapter->reset_reason = rwi->reset_reason; 1972 /* requestor of VNIC_RESET_CHANGE_PARAM already has the rtnl lock */ 1973 if (!(adapter->reset_reason == VNIC_RESET_CHANGE_PARAM)) 1974 rtnl_lock(); 1975 1976 /* Now that we have the rtnl lock, clear any pending failover. 1977 * This will ensure ibmvnic_open() has either completed or will 1978 * block until failover is complete. 1979 */ 1980 if (rwi->reset_reason == VNIC_RESET_FAILOVER) 1981 adapter->failover_pending = false; 1982 1983 /* read the state and check (again) after getting rtnl */ 1984 reset_state = adapter->state; 1985 1986 if (reset_state == VNIC_REMOVING || reset_state == VNIC_REMOVED) { 1987 rc = -EBUSY; 1988 goto out; 1989 } 1990 1991 netif_carrier_off(netdev); 1992 1993 old_num_rx_queues = adapter->req_rx_queues; 1994 old_num_tx_queues = adapter->req_tx_queues; 1995 old_num_rx_slots = adapter->req_rx_add_entries_per_subcrq; 1996 old_num_tx_slots = adapter->req_tx_entries_per_subcrq; 1997 1998 ibmvnic_cleanup(netdev); 1999 2000 if (reset_state == VNIC_OPEN && 2001 adapter->reset_reason != VNIC_RESET_MOBILITY && 2002 adapter->reset_reason != VNIC_RESET_FAILOVER) { 2003 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) { 2004 rc = __ibmvnic_close(netdev); 2005 if (rc) 2006 goto out; 2007 } else { 2008 adapter->state = VNIC_CLOSING; 2009 2010 /* Release the RTNL lock before link state change and 2011 * re-acquire after the link state change to allow 2012 * linkwatch_event to grab the RTNL lock and run during 2013 * a reset. 2014 */ 2015 rtnl_unlock(); 2016 rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN); 2017 rtnl_lock(); 2018 if (rc) 2019 goto out; 2020 2021 if (adapter->state == VNIC_OPEN) { 2022 /* When we dropped rtnl, ibmvnic_open() got 2023 * it and noticed that we are resetting and 2024 * set the adapter state to OPEN. Update our 2025 * new "target" state, and resume the reset 2026 * from VNIC_CLOSING state. 2027 */ 2028 netdev_dbg(netdev, 2029 "Open changed state from %s, updating.\n", 2030 adapter_state_to_string(reset_state)); 2031 reset_state = VNIC_OPEN; 2032 adapter->state = VNIC_CLOSING; 2033 } 2034 2035 if (adapter->state != VNIC_CLOSING) { 2036 /* If someone else changed the adapter state 2037 * when we dropped the rtnl, fail the reset 2038 */ 2039 rc = -1; 2040 goto out; 2041 } 2042 adapter->state = VNIC_CLOSED; 2043 } 2044 } 2045 2046 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) { 2047 release_resources(adapter); 2048 release_sub_crqs(adapter, 1); 2049 release_crq_queue(adapter); 2050 } 2051 2052 if (adapter->reset_reason != VNIC_RESET_NON_FATAL) { 2053 /* remove the closed state so when we call open it appears 2054 * we are coming from the probed state. 2055 */ 2056 adapter->state = VNIC_PROBED; 2057 2058 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) { 2059 rc = init_crq_queue(adapter); 2060 } else if (adapter->reset_reason == VNIC_RESET_MOBILITY) { 2061 rc = ibmvnic_reenable_crq_queue(adapter); 2062 release_sub_crqs(adapter, 1); 2063 } else { 2064 rc = ibmvnic_reset_crq(adapter); 2065 if (rc == H_CLOSED || rc == H_SUCCESS) { 2066 rc = vio_enable_interrupts(adapter->vdev); 2067 if (rc) 2068 netdev_err(adapter->netdev, 2069 "Reset failed to enable interrupts. rc=%d\n", 2070 rc); 2071 } 2072 } 2073 2074 if (rc) { 2075 netdev_err(adapter->netdev, 2076 "Reset couldn't initialize crq. rc=%d\n", rc); 2077 goto out; 2078 } 2079 2080 rc = ibmvnic_reset_init(adapter, true); 2081 if (rc) { 2082 rc = IBMVNIC_INIT_FAILED; 2083 goto out; 2084 } 2085 2086 /* If the adapter was in PROBE state prior to the reset, 2087 * exit here. 2088 */ 2089 if (reset_state == VNIC_PROBED) { 2090 rc = 0; 2091 goto out; 2092 } 2093 2094 rc = ibmvnic_login(netdev); 2095 if (rc) 2096 goto out; 2097 2098 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) { 2099 rc = init_resources(adapter); 2100 if (rc) 2101 goto out; 2102 } else if (adapter->req_rx_queues != old_num_rx_queues || 2103 adapter->req_tx_queues != old_num_tx_queues || 2104 adapter->req_rx_add_entries_per_subcrq != 2105 old_num_rx_slots || 2106 adapter->req_tx_entries_per_subcrq != 2107 old_num_tx_slots || 2108 !adapter->rx_pool || 2109 !adapter->tso_pool || 2110 !adapter->tx_pool) { 2111 release_rx_pools(adapter); 2112 release_tx_pools(adapter); 2113 release_napi(adapter); 2114 release_vpd_data(adapter); 2115 2116 rc = init_resources(adapter); 2117 if (rc) 2118 goto out; 2119 2120 } else { 2121 rc = reset_tx_pools(adapter); 2122 if (rc) { 2123 netdev_dbg(adapter->netdev, "reset tx pools failed (%d)\n", 2124 rc); 2125 goto out; 2126 } 2127 2128 rc = reset_rx_pools(adapter); 2129 if (rc) { 2130 netdev_dbg(adapter->netdev, "reset rx pools failed (%d)\n", 2131 rc); 2132 goto out; 2133 } 2134 } 2135 ibmvnic_disable_irqs(adapter); 2136 } 2137 adapter->state = VNIC_CLOSED; 2138 2139 if (reset_state == VNIC_CLOSED) { 2140 rc = 0; 2141 goto out; 2142 } 2143 2144 rc = __ibmvnic_open(netdev); 2145 if (rc) { 2146 rc = IBMVNIC_OPEN_FAILED; 2147 goto out; 2148 } 2149 2150 /* refresh device's multicast list */ 2151 ibmvnic_set_multi(netdev); 2152 2153 if (adapter->reset_reason == VNIC_RESET_FAILOVER || 2154 adapter->reset_reason == VNIC_RESET_MOBILITY) 2155 __netdev_notify_peers(netdev); 2156 2157 rc = 0; 2158 2159 out: 2160 /* restore the adapter state if reset failed */ 2161 if (rc) 2162 adapter->state = reset_state; 2163 /* requestor of VNIC_RESET_CHANGE_PARAM should still hold the rtnl lock */ 2164 if (!(adapter->reset_reason == VNIC_RESET_CHANGE_PARAM)) 2165 rtnl_unlock(); 2166 2167 netdev_dbg(adapter->netdev, "[S:%s FOP:%d] Reset done, rc %d\n", 2168 adapter_state_to_string(adapter->state), 2169 adapter->failover_pending, rc); 2170 return rc; 2171 } 2172 2173 static int do_hard_reset(struct ibmvnic_adapter *adapter, 2174 struct ibmvnic_rwi *rwi, u32 reset_state) 2175 { 2176 struct net_device *netdev = adapter->netdev; 2177 int rc; 2178 2179 netdev_dbg(adapter->netdev, "Hard resetting driver (%s)\n", 2180 reset_reason_to_string(rwi->reset_reason)); 2181 2182 /* read the state and check (again) after getting rtnl */ 2183 reset_state = adapter->state; 2184 2185 if (reset_state == VNIC_REMOVING || reset_state == VNIC_REMOVED) { 2186 rc = -EBUSY; 2187 goto out; 2188 } 2189 2190 netif_carrier_off(netdev); 2191 adapter->reset_reason = rwi->reset_reason; 2192 2193 ibmvnic_cleanup(netdev); 2194 release_resources(adapter); 2195 release_sub_crqs(adapter, 0); 2196 release_crq_queue(adapter); 2197 2198 /* remove the closed state so when we call open it appears 2199 * we are coming from the probed state. 2200 */ 2201 adapter->state = VNIC_PROBED; 2202 2203 reinit_completion(&adapter->init_done); 2204 rc = init_crq_queue(adapter); 2205 if (rc) { 2206 netdev_err(adapter->netdev, 2207 "Couldn't initialize crq. rc=%d\n", rc); 2208 goto out; 2209 } 2210 2211 rc = ibmvnic_reset_init(adapter, false); 2212 if (rc) 2213 goto out; 2214 2215 /* If the adapter was in PROBE state prior to the reset, 2216 * exit here. 2217 */ 2218 if (reset_state == VNIC_PROBED) 2219 goto out; 2220 2221 rc = ibmvnic_login(netdev); 2222 if (rc) 2223 goto out; 2224 2225 rc = init_resources(adapter); 2226 if (rc) 2227 goto out; 2228 2229 ibmvnic_disable_irqs(adapter); 2230 adapter->state = VNIC_CLOSED; 2231 2232 if (reset_state == VNIC_CLOSED) 2233 goto out; 2234 2235 rc = __ibmvnic_open(netdev); 2236 if (rc) { 2237 rc = IBMVNIC_OPEN_FAILED; 2238 goto out; 2239 } 2240 2241 __netdev_notify_peers(netdev); 2242 out: 2243 /* restore adapter state if reset failed */ 2244 if (rc) 2245 adapter->state = reset_state; 2246 netdev_dbg(adapter->netdev, "[S:%s FOP:%d] Hard reset done, rc %d\n", 2247 adapter_state_to_string(adapter->state), 2248 adapter->failover_pending, rc); 2249 return rc; 2250 } 2251 2252 static struct ibmvnic_rwi *get_next_rwi(struct ibmvnic_adapter *adapter) 2253 { 2254 struct ibmvnic_rwi *rwi; 2255 unsigned long flags; 2256 2257 spin_lock_irqsave(&adapter->rwi_lock, flags); 2258 2259 if (!list_empty(&adapter->rwi_list)) { 2260 rwi = list_first_entry(&adapter->rwi_list, struct ibmvnic_rwi, 2261 list); 2262 list_del(&rwi->list); 2263 } else { 2264 rwi = NULL; 2265 } 2266 2267 spin_unlock_irqrestore(&adapter->rwi_lock, flags); 2268 return rwi; 2269 } 2270 2271 static void __ibmvnic_reset(struct work_struct *work) 2272 { 2273 struct ibmvnic_rwi *rwi; 2274 struct ibmvnic_adapter *adapter; 2275 bool saved_state = false; 2276 unsigned long flags; 2277 u32 reset_state; 2278 int rc = 0; 2279 2280 adapter = container_of(work, struct ibmvnic_adapter, ibmvnic_reset); 2281 2282 if (test_and_set_bit_lock(0, &adapter->resetting)) { 2283 queue_delayed_work(system_long_wq, 2284 &adapter->ibmvnic_delayed_reset, 2285 IBMVNIC_RESET_DELAY); 2286 return; 2287 } 2288 2289 rwi = get_next_rwi(adapter); 2290 while (rwi) { 2291 spin_lock_irqsave(&adapter->state_lock, flags); 2292 2293 if (adapter->state == VNIC_REMOVING || 2294 adapter->state == VNIC_REMOVED) { 2295 spin_unlock_irqrestore(&adapter->state_lock, flags); 2296 kfree(rwi); 2297 rc = EBUSY; 2298 break; 2299 } 2300 2301 if (!saved_state) { 2302 reset_state = adapter->state; 2303 saved_state = true; 2304 } 2305 spin_unlock_irqrestore(&adapter->state_lock, flags); 2306 2307 if (adapter->force_reset_recovery) { 2308 /* Since we are doing a hard reset now, clear the 2309 * failover_pending flag so we don't ignore any 2310 * future MOBILITY or other resets. 2311 */ 2312 adapter->failover_pending = false; 2313 2314 /* Transport event occurred during previous reset */ 2315 if (adapter->wait_for_reset) { 2316 /* Previous was CHANGE_PARAM; caller locked */ 2317 adapter->force_reset_recovery = false; 2318 rc = do_hard_reset(adapter, rwi, reset_state); 2319 } else { 2320 rtnl_lock(); 2321 adapter->force_reset_recovery = false; 2322 rc = do_hard_reset(adapter, rwi, reset_state); 2323 rtnl_unlock(); 2324 } 2325 if (rc) { 2326 /* give backing device time to settle down */ 2327 netdev_dbg(adapter->netdev, 2328 "[S:%s] Hard reset failed, waiting 60 secs\n", 2329 adapter_state_to_string(adapter->state)); 2330 set_current_state(TASK_UNINTERRUPTIBLE); 2331 schedule_timeout(60 * HZ); 2332 } 2333 } else { 2334 rc = do_reset(adapter, rwi, reset_state); 2335 } 2336 kfree(rwi); 2337 adapter->last_reset_time = jiffies; 2338 2339 if (rc) 2340 netdev_dbg(adapter->netdev, "Reset failed, rc=%d\n", rc); 2341 2342 rwi = get_next_rwi(adapter); 2343 2344 if (rwi && (rwi->reset_reason == VNIC_RESET_FAILOVER || 2345 rwi->reset_reason == VNIC_RESET_MOBILITY)) 2346 adapter->force_reset_recovery = true; 2347 } 2348 2349 if (adapter->wait_for_reset) { 2350 adapter->reset_done_rc = rc; 2351 complete(&adapter->reset_done); 2352 } 2353 2354 clear_bit_unlock(0, &adapter->resetting); 2355 2356 netdev_dbg(adapter->netdev, 2357 "[S:%s FRR:%d WFR:%d] Done processing resets\n", 2358 adapter_state_to_string(adapter->state), 2359 adapter->force_reset_recovery, 2360 adapter->wait_for_reset); 2361 } 2362 2363 static void __ibmvnic_delayed_reset(struct work_struct *work) 2364 { 2365 struct ibmvnic_adapter *adapter; 2366 2367 adapter = container_of(work, struct ibmvnic_adapter, 2368 ibmvnic_delayed_reset.work); 2369 __ibmvnic_reset(&adapter->ibmvnic_reset); 2370 } 2371 2372 static int ibmvnic_reset(struct ibmvnic_adapter *adapter, 2373 enum ibmvnic_reset_reason reason) 2374 { 2375 struct list_head *entry, *tmp_entry; 2376 struct ibmvnic_rwi *rwi, *tmp; 2377 struct net_device *netdev = adapter->netdev; 2378 unsigned long flags; 2379 int ret; 2380 2381 spin_lock_irqsave(&adapter->rwi_lock, flags); 2382 2383 /* If failover is pending don't schedule any other reset. 2384 * Instead let the failover complete. If there is already a 2385 * a failover reset scheduled, we will detect and drop the 2386 * duplicate reset when walking the ->rwi_list below. 2387 */ 2388 if (adapter->state == VNIC_REMOVING || 2389 adapter->state == VNIC_REMOVED || 2390 (adapter->failover_pending && reason != VNIC_RESET_FAILOVER)) { 2391 ret = EBUSY; 2392 netdev_dbg(netdev, "Adapter removing or pending failover, skipping reset\n"); 2393 goto err; 2394 } 2395 2396 if (adapter->state == VNIC_PROBING) { 2397 netdev_warn(netdev, "Adapter reset during probe\n"); 2398 adapter->init_done_rc = EAGAIN; 2399 ret = EAGAIN; 2400 goto err; 2401 } 2402 2403 list_for_each(entry, &adapter->rwi_list) { 2404 tmp = list_entry(entry, struct ibmvnic_rwi, list); 2405 if (tmp->reset_reason == reason) { 2406 netdev_dbg(netdev, "Skipping matching reset, reason=%s\n", 2407 reset_reason_to_string(reason)); 2408 ret = EBUSY; 2409 goto err; 2410 } 2411 } 2412 2413 rwi = kzalloc(sizeof(*rwi), GFP_ATOMIC); 2414 if (!rwi) { 2415 ret = ENOMEM; 2416 goto err; 2417 } 2418 /* if we just received a transport event, 2419 * flush reset queue and process this reset 2420 */ 2421 if (adapter->force_reset_recovery && !list_empty(&adapter->rwi_list)) { 2422 list_for_each_safe(entry, tmp_entry, &adapter->rwi_list) 2423 list_del(entry); 2424 } 2425 rwi->reset_reason = reason; 2426 list_add_tail(&rwi->list, &adapter->rwi_list); 2427 netdev_dbg(adapter->netdev, "Scheduling reset (reason %s)\n", 2428 reset_reason_to_string(reason)); 2429 queue_work(system_long_wq, &adapter->ibmvnic_reset); 2430 2431 ret = 0; 2432 err: 2433 /* ibmvnic_close() below can block, so drop the lock first */ 2434 spin_unlock_irqrestore(&adapter->rwi_lock, flags); 2435 2436 if (ret == ENOMEM) 2437 ibmvnic_close(netdev); 2438 2439 return -ret; 2440 } 2441 2442 static void ibmvnic_tx_timeout(struct net_device *dev, unsigned int txqueue) 2443 { 2444 struct ibmvnic_adapter *adapter = netdev_priv(dev); 2445 2446 if (test_bit(0, &adapter->resetting)) { 2447 netdev_err(adapter->netdev, 2448 "Adapter is resetting, skip timeout reset\n"); 2449 return; 2450 } 2451 /* No queuing up reset until at least 5 seconds (default watchdog val) 2452 * after last reset 2453 */ 2454 if (time_before(jiffies, (adapter->last_reset_time + dev->watchdog_timeo))) { 2455 netdev_dbg(dev, "Not yet time to tx timeout.\n"); 2456 return; 2457 } 2458 ibmvnic_reset(adapter, VNIC_RESET_TIMEOUT); 2459 } 2460 2461 static void remove_buff_from_pool(struct ibmvnic_adapter *adapter, 2462 struct ibmvnic_rx_buff *rx_buff) 2463 { 2464 struct ibmvnic_rx_pool *pool = &adapter->rx_pool[rx_buff->pool_index]; 2465 2466 rx_buff->skb = NULL; 2467 2468 pool->free_map[pool->next_alloc] = (int)(rx_buff - pool->rx_buff); 2469 pool->next_alloc = (pool->next_alloc + 1) % pool->size; 2470 2471 atomic_dec(&pool->available); 2472 } 2473 2474 static int ibmvnic_poll(struct napi_struct *napi, int budget) 2475 { 2476 struct ibmvnic_sub_crq_queue *rx_scrq; 2477 struct ibmvnic_adapter *adapter; 2478 struct net_device *netdev; 2479 int frames_processed; 2480 int scrq_num; 2481 2482 netdev = napi->dev; 2483 adapter = netdev_priv(netdev); 2484 scrq_num = (int)(napi - adapter->napi); 2485 frames_processed = 0; 2486 rx_scrq = adapter->rx_scrq[scrq_num]; 2487 2488 restart_poll: 2489 while (frames_processed < budget) { 2490 struct sk_buff *skb; 2491 struct ibmvnic_rx_buff *rx_buff; 2492 union sub_crq *next; 2493 u32 length; 2494 u16 offset; 2495 u8 flags = 0; 2496 2497 if (unlikely(test_bit(0, &adapter->resetting) && 2498 adapter->reset_reason != VNIC_RESET_NON_FATAL)) { 2499 enable_scrq_irq(adapter, rx_scrq); 2500 napi_complete_done(napi, frames_processed); 2501 return frames_processed; 2502 } 2503 2504 if (!pending_scrq(adapter, rx_scrq)) 2505 break; 2506 next = ibmvnic_next_scrq(adapter, rx_scrq); 2507 rx_buff = (struct ibmvnic_rx_buff *) 2508 be64_to_cpu(next->rx_comp.correlator); 2509 /* do error checking */ 2510 if (next->rx_comp.rc) { 2511 netdev_dbg(netdev, "rx buffer returned with rc %x\n", 2512 be16_to_cpu(next->rx_comp.rc)); 2513 /* free the entry */ 2514 next->rx_comp.first = 0; 2515 dev_kfree_skb_any(rx_buff->skb); 2516 remove_buff_from_pool(adapter, rx_buff); 2517 continue; 2518 } else if (!rx_buff->skb) { 2519 /* free the entry */ 2520 next->rx_comp.first = 0; 2521 remove_buff_from_pool(adapter, rx_buff); 2522 continue; 2523 } 2524 2525 length = be32_to_cpu(next->rx_comp.len); 2526 offset = be16_to_cpu(next->rx_comp.off_frame_data); 2527 flags = next->rx_comp.flags; 2528 skb = rx_buff->skb; 2529 /* load long_term_buff before copying to skb */ 2530 dma_rmb(); 2531 skb_copy_to_linear_data(skb, rx_buff->data + offset, 2532 length); 2533 2534 /* VLAN Header has been stripped by the system firmware and 2535 * needs to be inserted by the driver 2536 */ 2537 if (adapter->rx_vlan_header_insertion && 2538 (flags & IBMVNIC_VLAN_STRIPPED)) 2539 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), 2540 ntohs(next->rx_comp.vlan_tci)); 2541 2542 /* free the entry */ 2543 next->rx_comp.first = 0; 2544 remove_buff_from_pool(adapter, rx_buff); 2545 2546 skb_put(skb, length); 2547 skb->protocol = eth_type_trans(skb, netdev); 2548 skb_record_rx_queue(skb, scrq_num); 2549 2550 if (flags & IBMVNIC_IP_CHKSUM_GOOD && 2551 flags & IBMVNIC_TCP_UDP_CHKSUM_GOOD) { 2552 skb->ip_summed = CHECKSUM_UNNECESSARY; 2553 } 2554 2555 length = skb->len; 2556 napi_gro_receive(napi, skb); /* send it up */ 2557 netdev->stats.rx_packets++; 2558 netdev->stats.rx_bytes += length; 2559 adapter->rx_stats_buffers[scrq_num].packets++; 2560 adapter->rx_stats_buffers[scrq_num].bytes += length; 2561 frames_processed++; 2562 } 2563 2564 if (adapter->state != VNIC_CLOSING && 2565 ((atomic_read(&adapter->rx_pool[scrq_num].available) < 2566 adapter->req_rx_add_entries_per_subcrq / 2) || 2567 frames_processed < budget)) 2568 replenish_rx_pool(adapter, &adapter->rx_pool[scrq_num]); 2569 if (frames_processed < budget) { 2570 if (napi_complete_done(napi, frames_processed)) { 2571 enable_scrq_irq(adapter, rx_scrq); 2572 if (pending_scrq(adapter, rx_scrq)) { 2573 if (napi_reschedule(napi)) { 2574 disable_scrq_irq(adapter, rx_scrq); 2575 goto restart_poll; 2576 } 2577 } 2578 } 2579 } 2580 return frames_processed; 2581 } 2582 2583 static int wait_for_reset(struct ibmvnic_adapter *adapter) 2584 { 2585 int rc, ret; 2586 2587 adapter->fallback.mtu = adapter->req_mtu; 2588 adapter->fallback.rx_queues = adapter->req_rx_queues; 2589 adapter->fallback.tx_queues = adapter->req_tx_queues; 2590 adapter->fallback.rx_entries = adapter->req_rx_add_entries_per_subcrq; 2591 adapter->fallback.tx_entries = adapter->req_tx_entries_per_subcrq; 2592 2593 reinit_completion(&adapter->reset_done); 2594 adapter->wait_for_reset = true; 2595 rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM); 2596 2597 if (rc) { 2598 ret = rc; 2599 goto out; 2600 } 2601 rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done, 60000); 2602 if (rc) { 2603 ret = -ENODEV; 2604 goto out; 2605 } 2606 2607 ret = 0; 2608 if (adapter->reset_done_rc) { 2609 ret = -EIO; 2610 adapter->desired.mtu = adapter->fallback.mtu; 2611 adapter->desired.rx_queues = adapter->fallback.rx_queues; 2612 adapter->desired.tx_queues = adapter->fallback.tx_queues; 2613 adapter->desired.rx_entries = adapter->fallback.rx_entries; 2614 adapter->desired.tx_entries = adapter->fallback.tx_entries; 2615 2616 reinit_completion(&adapter->reset_done); 2617 adapter->wait_for_reset = true; 2618 rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM); 2619 if (rc) { 2620 ret = rc; 2621 goto out; 2622 } 2623 rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done, 2624 60000); 2625 if (rc) { 2626 ret = -ENODEV; 2627 goto out; 2628 } 2629 } 2630 out: 2631 adapter->wait_for_reset = false; 2632 2633 return ret; 2634 } 2635 2636 static int ibmvnic_change_mtu(struct net_device *netdev, int new_mtu) 2637 { 2638 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 2639 2640 adapter->desired.mtu = new_mtu + ETH_HLEN; 2641 2642 return wait_for_reset(adapter); 2643 } 2644 2645 static netdev_features_t ibmvnic_features_check(struct sk_buff *skb, 2646 struct net_device *dev, 2647 netdev_features_t features) 2648 { 2649 /* Some backing hardware adapters can not 2650 * handle packets with a MSS less than 224 2651 * or with only one segment. 2652 */ 2653 if (skb_is_gso(skb)) { 2654 if (skb_shinfo(skb)->gso_size < 224 || 2655 skb_shinfo(skb)->gso_segs == 1) 2656 features &= ~NETIF_F_GSO_MASK; 2657 } 2658 2659 return features; 2660 } 2661 2662 static const struct net_device_ops ibmvnic_netdev_ops = { 2663 .ndo_open = ibmvnic_open, 2664 .ndo_stop = ibmvnic_close, 2665 .ndo_start_xmit = ibmvnic_xmit, 2666 .ndo_set_rx_mode = ibmvnic_set_multi, 2667 .ndo_set_mac_address = ibmvnic_set_mac, 2668 .ndo_validate_addr = eth_validate_addr, 2669 .ndo_tx_timeout = ibmvnic_tx_timeout, 2670 .ndo_change_mtu = ibmvnic_change_mtu, 2671 .ndo_features_check = ibmvnic_features_check, 2672 }; 2673 2674 /* ethtool functions */ 2675 2676 static int ibmvnic_get_link_ksettings(struct net_device *netdev, 2677 struct ethtool_link_ksettings *cmd) 2678 { 2679 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 2680 int rc; 2681 2682 rc = send_query_phys_parms(adapter); 2683 if (rc) { 2684 adapter->speed = SPEED_UNKNOWN; 2685 adapter->duplex = DUPLEX_UNKNOWN; 2686 } 2687 cmd->base.speed = adapter->speed; 2688 cmd->base.duplex = adapter->duplex; 2689 cmd->base.port = PORT_FIBRE; 2690 cmd->base.phy_address = 0; 2691 cmd->base.autoneg = AUTONEG_ENABLE; 2692 2693 return 0; 2694 } 2695 2696 static void ibmvnic_get_drvinfo(struct net_device *netdev, 2697 struct ethtool_drvinfo *info) 2698 { 2699 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 2700 2701 strscpy(info->driver, ibmvnic_driver_name, sizeof(info->driver)); 2702 strscpy(info->version, IBMVNIC_DRIVER_VERSION, sizeof(info->version)); 2703 strscpy(info->fw_version, adapter->fw_version, 2704 sizeof(info->fw_version)); 2705 } 2706 2707 static u32 ibmvnic_get_msglevel(struct net_device *netdev) 2708 { 2709 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 2710 2711 return adapter->msg_enable; 2712 } 2713 2714 static void ibmvnic_set_msglevel(struct net_device *netdev, u32 data) 2715 { 2716 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 2717 2718 adapter->msg_enable = data; 2719 } 2720 2721 static u32 ibmvnic_get_link(struct net_device *netdev) 2722 { 2723 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 2724 2725 /* Don't need to send a query because we request a logical link up at 2726 * init and then we wait for link state indications 2727 */ 2728 return adapter->logical_link_state; 2729 } 2730 2731 static void ibmvnic_get_ringparam(struct net_device *netdev, 2732 struct ethtool_ringparam *ring) 2733 { 2734 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 2735 2736 if (adapter->priv_flags & IBMVNIC_USE_SERVER_MAXES) { 2737 ring->rx_max_pending = adapter->max_rx_add_entries_per_subcrq; 2738 ring->tx_max_pending = adapter->max_tx_entries_per_subcrq; 2739 } else { 2740 ring->rx_max_pending = IBMVNIC_MAX_QUEUE_SZ; 2741 ring->tx_max_pending = IBMVNIC_MAX_QUEUE_SZ; 2742 } 2743 ring->rx_mini_max_pending = 0; 2744 ring->rx_jumbo_max_pending = 0; 2745 ring->rx_pending = adapter->req_rx_add_entries_per_subcrq; 2746 ring->tx_pending = adapter->req_tx_entries_per_subcrq; 2747 ring->rx_mini_pending = 0; 2748 ring->rx_jumbo_pending = 0; 2749 } 2750 2751 static int ibmvnic_set_ringparam(struct net_device *netdev, 2752 struct ethtool_ringparam *ring) 2753 { 2754 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 2755 int ret; 2756 2757 ret = 0; 2758 adapter->desired.rx_entries = ring->rx_pending; 2759 adapter->desired.tx_entries = ring->tx_pending; 2760 2761 ret = wait_for_reset(adapter); 2762 2763 if (!ret && 2764 (adapter->req_rx_add_entries_per_subcrq != ring->rx_pending || 2765 adapter->req_tx_entries_per_subcrq != ring->tx_pending)) 2766 netdev_info(netdev, 2767 "Could not match full ringsize request. Requested: RX %d, TX %d; Allowed: RX %llu, TX %llu\n", 2768 ring->rx_pending, ring->tx_pending, 2769 adapter->req_rx_add_entries_per_subcrq, 2770 adapter->req_tx_entries_per_subcrq); 2771 return ret; 2772 } 2773 2774 static void ibmvnic_get_channels(struct net_device *netdev, 2775 struct ethtool_channels *channels) 2776 { 2777 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 2778 2779 if (adapter->priv_flags & IBMVNIC_USE_SERVER_MAXES) { 2780 channels->max_rx = adapter->max_rx_queues; 2781 channels->max_tx = adapter->max_tx_queues; 2782 } else { 2783 channels->max_rx = IBMVNIC_MAX_QUEUES; 2784 channels->max_tx = IBMVNIC_MAX_QUEUES; 2785 } 2786 2787 channels->max_other = 0; 2788 channels->max_combined = 0; 2789 channels->rx_count = adapter->req_rx_queues; 2790 channels->tx_count = adapter->req_tx_queues; 2791 channels->other_count = 0; 2792 channels->combined_count = 0; 2793 } 2794 2795 static int ibmvnic_set_channels(struct net_device *netdev, 2796 struct ethtool_channels *channels) 2797 { 2798 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 2799 int ret; 2800 2801 ret = 0; 2802 adapter->desired.rx_queues = channels->rx_count; 2803 adapter->desired.tx_queues = channels->tx_count; 2804 2805 ret = wait_for_reset(adapter); 2806 2807 if (!ret && 2808 (adapter->req_rx_queues != channels->rx_count || 2809 adapter->req_tx_queues != channels->tx_count)) 2810 netdev_info(netdev, 2811 "Could not match full channels request. Requested: RX %d, TX %d; Allowed: RX %llu, TX %llu\n", 2812 channels->rx_count, channels->tx_count, 2813 adapter->req_rx_queues, adapter->req_tx_queues); 2814 return ret; 2815 } 2816 2817 static void ibmvnic_get_strings(struct net_device *dev, u32 stringset, u8 *data) 2818 { 2819 struct ibmvnic_adapter *adapter = netdev_priv(dev); 2820 int i; 2821 2822 switch (stringset) { 2823 case ETH_SS_STATS: 2824 for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); 2825 i++, data += ETH_GSTRING_LEN) 2826 memcpy(data, ibmvnic_stats[i].name, ETH_GSTRING_LEN); 2827 2828 for (i = 0; i < adapter->req_tx_queues; i++) { 2829 snprintf(data, ETH_GSTRING_LEN, "tx%d_packets", i); 2830 data += ETH_GSTRING_LEN; 2831 2832 snprintf(data, ETH_GSTRING_LEN, "tx%d_bytes", i); 2833 data += ETH_GSTRING_LEN; 2834 2835 snprintf(data, ETH_GSTRING_LEN, 2836 "tx%d_dropped_packets", i); 2837 data += ETH_GSTRING_LEN; 2838 } 2839 2840 for (i = 0; i < adapter->req_rx_queues; i++) { 2841 snprintf(data, ETH_GSTRING_LEN, "rx%d_packets", i); 2842 data += ETH_GSTRING_LEN; 2843 2844 snprintf(data, ETH_GSTRING_LEN, "rx%d_bytes", i); 2845 data += ETH_GSTRING_LEN; 2846 2847 snprintf(data, ETH_GSTRING_LEN, "rx%d_interrupts", i); 2848 data += ETH_GSTRING_LEN; 2849 } 2850 break; 2851 2852 case ETH_SS_PRIV_FLAGS: 2853 for (i = 0; i < ARRAY_SIZE(ibmvnic_priv_flags); i++) 2854 strcpy(data + i * ETH_GSTRING_LEN, 2855 ibmvnic_priv_flags[i]); 2856 break; 2857 default: 2858 return; 2859 } 2860 } 2861 2862 static int ibmvnic_get_sset_count(struct net_device *dev, int sset) 2863 { 2864 struct ibmvnic_adapter *adapter = netdev_priv(dev); 2865 2866 switch (sset) { 2867 case ETH_SS_STATS: 2868 return ARRAY_SIZE(ibmvnic_stats) + 2869 adapter->req_tx_queues * NUM_TX_STATS + 2870 adapter->req_rx_queues * NUM_RX_STATS; 2871 case ETH_SS_PRIV_FLAGS: 2872 return ARRAY_SIZE(ibmvnic_priv_flags); 2873 default: 2874 return -EOPNOTSUPP; 2875 } 2876 } 2877 2878 static void ibmvnic_get_ethtool_stats(struct net_device *dev, 2879 struct ethtool_stats *stats, u64 *data) 2880 { 2881 struct ibmvnic_adapter *adapter = netdev_priv(dev); 2882 union ibmvnic_crq crq; 2883 int i, j; 2884 int rc; 2885 2886 memset(&crq, 0, sizeof(crq)); 2887 crq.request_statistics.first = IBMVNIC_CRQ_CMD; 2888 crq.request_statistics.cmd = REQUEST_STATISTICS; 2889 crq.request_statistics.ioba = cpu_to_be32(adapter->stats_token); 2890 crq.request_statistics.len = 2891 cpu_to_be32(sizeof(struct ibmvnic_statistics)); 2892 2893 /* Wait for data to be written */ 2894 reinit_completion(&adapter->stats_done); 2895 rc = ibmvnic_send_crq(adapter, &crq); 2896 if (rc) 2897 return; 2898 rc = ibmvnic_wait_for_completion(adapter, &adapter->stats_done, 10000); 2899 if (rc) 2900 return; 2901 2902 for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); i++) 2903 data[i] = be64_to_cpu(IBMVNIC_GET_STAT 2904 (adapter, ibmvnic_stats[i].offset)); 2905 2906 for (j = 0; j < adapter->req_tx_queues; j++) { 2907 data[i] = adapter->tx_stats_buffers[j].packets; 2908 i++; 2909 data[i] = adapter->tx_stats_buffers[j].bytes; 2910 i++; 2911 data[i] = adapter->tx_stats_buffers[j].dropped_packets; 2912 i++; 2913 } 2914 2915 for (j = 0; j < adapter->req_rx_queues; j++) { 2916 data[i] = adapter->rx_stats_buffers[j].packets; 2917 i++; 2918 data[i] = adapter->rx_stats_buffers[j].bytes; 2919 i++; 2920 data[i] = adapter->rx_stats_buffers[j].interrupts; 2921 i++; 2922 } 2923 } 2924 2925 static u32 ibmvnic_get_priv_flags(struct net_device *netdev) 2926 { 2927 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 2928 2929 return adapter->priv_flags; 2930 } 2931 2932 static int ibmvnic_set_priv_flags(struct net_device *netdev, u32 flags) 2933 { 2934 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 2935 bool which_maxes = !!(flags & IBMVNIC_USE_SERVER_MAXES); 2936 2937 if (which_maxes) 2938 adapter->priv_flags |= IBMVNIC_USE_SERVER_MAXES; 2939 else 2940 adapter->priv_flags &= ~IBMVNIC_USE_SERVER_MAXES; 2941 2942 return 0; 2943 } 2944 2945 static const struct ethtool_ops ibmvnic_ethtool_ops = { 2946 .get_drvinfo = ibmvnic_get_drvinfo, 2947 .get_msglevel = ibmvnic_get_msglevel, 2948 .set_msglevel = ibmvnic_set_msglevel, 2949 .get_link = ibmvnic_get_link, 2950 .get_ringparam = ibmvnic_get_ringparam, 2951 .set_ringparam = ibmvnic_set_ringparam, 2952 .get_channels = ibmvnic_get_channels, 2953 .set_channels = ibmvnic_set_channels, 2954 .get_strings = ibmvnic_get_strings, 2955 .get_sset_count = ibmvnic_get_sset_count, 2956 .get_ethtool_stats = ibmvnic_get_ethtool_stats, 2957 .get_link_ksettings = ibmvnic_get_link_ksettings, 2958 .get_priv_flags = ibmvnic_get_priv_flags, 2959 .set_priv_flags = ibmvnic_set_priv_flags, 2960 }; 2961 2962 /* Routines for managing CRQs/sCRQs */ 2963 2964 static int reset_one_sub_crq_queue(struct ibmvnic_adapter *adapter, 2965 struct ibmvnic_sub_crq_queue *scrq) 2966 { 2967 int rc; 2968 2969 if (!scrq) { 2970 netdev_dbg(adapter->netdev, "Invalid scrq reset.\n"); 2971 return -EINVAL; 2972 } 2973 2974 if (scrq->irq) { 2975 free_irq(scrq->irq, scrq); 2976 irq_dispose_mapping(scrq->irq); 2977 scrq->irq = 0; 2978 } 2979 2980 if (scrq->msgs) { 2981 memset(scrq->msgs, 0, 4 * PAGE_SIZE); 2982 atomic_set(&scrq->used, 0); 2983 scrq->cur = 0; 2984 scrq->ind_buf.index = 0; 2985 } else { 2986 netdev_dbg(adapter->netdev, "Invalid scrq reset\n"); 2987 return -EINVAL; 2988 } 2989 2990 rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token, 2991 4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq); 2992 return rc; 2993 } 2994 2995 static int reset_sub_crq_queues(struct ibmvnic_adapter *adapter) 2996 { 2997 int i, rc; 2998 2999 if (!adapter->tx_scrq || !adapter->rx_scrq) 3000 return -EINVAL; 3001 3002 for (i = 0; i < adapter->req_tx_queues; i++) { 3003 netdev_dbg(adapter->netdev, "Re-setting tx_scrq[%d]\n", i); 3004 rc = reset_one_sub_crq_queue(adapter, adapter->tx_scrq[i]); 3005 if (rc) 3006 return rc; 3007 } 3008 3009 for (i = 0; i < adapter->req_rx_queues; i++) { 3010 netdev_dbg(adapter->netdev, "Re-setting rx_scrq[%d]\n", i); 3011 rc = reset_one_sub_crq_queue(adapter, adapter->rx_scrq[i]); 3012 if (rc) 3013 return rc; 3014 } 3015 3016 return rc; 3017 } 3018 3019 static void release_sub_crq_queue(struct ibmvnic_adapter *adapter, 3020 struct ibmvnic_sub_crq_queue *scrq, 3021 bool do_h_free) 3022 { 3023 struct device *dev = &adapter->vdev->dev; 3024 long rc; 3025 3026 netdev_dbg(adapter->netdev, "Releasing sub-CRQ\n"); 3027 3028 if (do_h_free) { 3029 /* Close the sub-crqs */ 3030 do { 3031 rc = plpar_hcall_norets(H_FREE_SUB_CRQ, 3032 adapter->vdev->unit_address, 3033 scrq->crq_num); 3034 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc)); 3035 3036 if (rc) { 3037 netdev_err(adapter->netdev, 3038 "Failed to release sub-CRQ %16lx, rc = %ld\n", 3039 scrq->crq_num, rc); 3040 } 3041 } 3042 3043 dma_free_coherent(dev, 3044 IBMVNIC_IND_ARR_SZ, 3045 scrq->ind_buf.indir_arr, 3046 scrq->ind_buf.indir_dma); 3047 3048 dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE, 3049 DMA_BIDIRECTIONAL); 3050 free_pages((unsigned long)scrq->msgs, 2); 3051 kfree(scrq); 3052 } 3053 3054 static struct ibmvnic_sub_crq_queue *init_sub_crq_queue(struct ibmvnic_adapter 3055 *adapter) 3056 { 3057 struct device *dev = &adapter->vdev->dev; 3058 struct ibmvnic_sub_crq_queue *scrq; 3059 int rc; 3060 3061 scrq = kzalloc(sizeof(*scrq), GFP_KERNEL); 3062 if (!scrq) 3063 return NULL; 3064 3065 scrq->msgs = 3066 (union sub_crq *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 2); 3067 if (!scrq->msgs) { 3068 dev_warn(dev, "Couldn't allocate crq queue messages page\n"); 3069 goto zero_page_failed; 3070 } 3071 3072 scrq->msg_token = dma_map_single(dev, scrq->msgs, 4 * PAGE_SIZE, 3073 DMA_BIDIRECTIONAL); 3074 if (dma_mapping_error(dev, scrq->msg_token)) { 3075 dev_warn(dev, "Couldn't map crq queue messages page\n"); 3076 goto map_failed; 3077 } 3078 3079 rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token, 3080 4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq); 3081 3082 if (rc == H_RESOURCE) 3083 rc = ibmvnic_reset_crq(adapter); 3084 3085 if (rc == H_CLOSED) { 3086 dev_warn(dev, "Partner adapter not ready, waiting.\n"); 3087 } else if (rc) { 3088 dev_warn(dev, "Error %d registering sub-crq\n", rc); 3089 goto reg_failed; 3090 } 3091 3092 scrq->adapter = adapter; 3093 scrq->size = 4 * PAGE_SIZE / sizeof(*scrq->msgs); 3094 scrq->ind_buf.index = 0; 3095 3096 scrq->ind_buf.indir_arr = 3097 dma_alloc_coherent(dev, 3098 IBMVNIC_IND_ARR_SZ, 3099 &scrq->ind_buf.indir_dma, 3100 GFP_KERNEL); 3101 3102 if (!scrq->ind_buf.indir_arr) 3103 goto indir_failed; 3104 3105 spin_lock_init(&scrq->lock); 3106 3107 netdev_dbg(adapter->netdev, 3108 "sub-crq initialized, num %lx, hw_irq=%lx, irq=%x\n", 3109 scrq->crq_num, scrq->hw_irq, scrq->irq); 3110 3111 return scrq; 3112 3113 indir_failed: 3114 do { 3115 rc = plpar_hcall_norets(H_FREE_SUB_CRQ, 3116 adapter->vdev->unit_address, 3117 scrq->crq_num); 3118 } while (rc == H_BUSY || rc == H_IS_LONG_BUSY(rc)); 3119 reg_failed: 3120 dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE, 3121 DMA_BIDIRECTIONAL); 3122 map_failed: 3123 free_pages((unsigned long)scrq->msgs, 2); 3124 zero_page_failed: 3125 kfree(scrq); 3126 3127 return NULL; 3128 } 3129 3130 static void release_sub_crqs(struct ibmvnic_adapter *adapter, bool do_h_free) 3131 { 3132 int i; 3133 3134 if (adapter->tx_scrq) { 3135 for (i = 0; i < adapter->num_active_tx_scrqs; i++) { 3136 if (!adapter->tx_scrq[i]) 3137 continue; 3138 3139 netdev_dbg(adapter->netdev, "Releasing tx_scrq[%d]\n", 3140 i); 3141 if (adapter->tx_scrq[i]->irq) { 3142 free_irq(adapter->tx_scrq[i]->irq, 3143 adapter->tx_scrq[i]); 3144 irq_dispose_mapping(adapter->tx_scrq[i]->irq); 3145 adapter->tx_scrq[i]->irq = 0; 3146 } 3147 3148 release_sub_crq_queue(adapter, adapter->tx_scrq[i], 3149 do_h_free); 3150 } 3151 3152 kfree(adapter->tx_scrq); 3153 adapter->tx_scrq = NULL; 3154 adapter->num_active_tx_scrqs = 0; 3155 } 3156 3157 if (adapter->rx_scrq) { 3158 for (i = 0; i < adapter->num_active_rx_scrqs; i++) { 3159 if (!adapter->rx_scrq[i]) 3160 continue; 3161 3162 netdev_dbg(adapter->netdev, "Releasing rx_scrq[%d]\n", 3163 i); 3164 if (adapter->rx_scrq[i]->irq) { 3165 free_irq(adapter->rx_scrq[i]->irq, 3166 adapter->rx_scrq[i]); 3167 irq_dispose_mapping(adapter->rx_scrq[i]->irq); 3168 adapter->rx_scrq[i]->irq = 0; 3169 } 3170 3171 release_sub_crq_queue(adapter, adapter->rx_scrq[i], 3172 do_h_free); 3173 } 3174 3175 kfree(adapter->rx_scrq); 3176 adapter->rx_scrq = NULL; 3177 adapter->num_active_rx_scrqs = 0; 3178 } 3179 } 3180 3181 static int disable_scrq_irq(struct ibmvnic_adapter *adapter, 3182 struct ibmvnic_sub_crq_queue *scrq) 3183 { 3184 struct device *dev = &adapter->vdev->dev; 3185 unsigned long rc; 3186 3187 rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address, 3188 H_DISABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0); 3189 if (rc) 3190 dev_err(dev, "Couldn't disable scrq irq 0x%lx. rc=%ld\n", 3191 scrq->hw_irq, rc); 3192 return rc; 3193 } 3194 3195 static int enable_scrq_irq(struct ibmvnic_adapter *adapter, 3196 struct ibmvnic_sub_crq_queue *scrq) 3197 { 3198 struct device *dev = &adapter->vdev->dev; 3199 unsigned long rc; 3200 3201 if (scrq->hw_irq > 0x100000000ULL) { 3202 dev_err(dev, "bad hw_irq = %lx\n", scrq->hw_irq); 3203 return 1; 3204 } 3205 3206 if (test_bit(0, &adapter->resetting) && 3207 adapter->reset_reason == VNIC_RESET_MOBILITY) { 3208 u64 val = (0xff000000) | scrq->hw_irq; 3209 3210 rc = plpar_hcall_norets(H_EOI, val); 3211 /* H_EOI would fail with rc = H_FUNCTION when running 3212 * in XIVE mode which is expected, but not an error. 3213 */ 3214 if (rc && rc != H_FUNCTION) 3215 dev_err(dev, "H_EOI FAILED irq 0x%llx. rc=%ld\n", 3216 val, rc); 3217 } 3218 3219 rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address, 3220 H_ENABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0); 3221 if (rc) 3222 dev_err(dev, "Couldn't enable scrq irq 0x%lx. rc=%ld\n", 3223 scrq->hw_irq, rc); 3224 return rc; 3225 } 3226 3227 static int ibmvnic_complete_tx(struct ibmvnic_adapter *adapter, 3228 struct ibmvnic_sub_crq_queue *scrq) 3229 { 3230 struct device *dev = &adapter->vdev->dev; 3231 struct ibmvnic_tx_pool *tx_pool; 3232 struct ibmvnic_tx_buff *txbuff; 3233 struct netdev_queue *txq; 3234 union sub_crq *next; 3235 int index; 3236 int i; 3237 3238 restart_loop: 3239 while (pending_scrq(adapter, scrq)) { 3240 unsigned int pool = scrq->pool_index; 3241 int num_entries = 0; 3242 int total_bytes = 0; 3243 int num_packets = 0; 3244 3245 next = ibmvnic_next_scrq(adapter, scrq); 3246 for (i = 0; i < next->tx_comp.num_comps; i++) { 3247 index = be32_to_cpu(next->tx_comp.correlators[i]); 3248 if (index & IBMVNIC_TSO_POOL_MASK) { 3249 tx_pool = &adapter->tso_pool[pool]; 3250 index &= ~IBMVNIC_TSO_POOL_MASK; 3251 } else { 3252 tx_pool = &adapter->tx_pool[pool]; 3253 } 3254 3255 txbuff = &tx_pool->tx_buff[index]; 3256 num_packets++; 3257 num_entries += txbuff->num_entries; 3258 if (txbuff->skb) { 3259 total_bytes += txbuff->skb->len; 3260 if (next->tx_comp.rcs[i]) { 3261 dev_err(dev, "tx error %x\n", 3262 next->tx_comp.rcs[i]); 3263 dev_kfree_skb_irq(txbuff->skb); 3264 } else { 3265 dev_consume_skb_irq(txbuff->skb); 3266 } 3267 txbuff->skb = NULL; 3268 } else { 3269 netdev_warn(adapter->netdev, 3270 "TX completion received with NULL socket buffer\n"); 3271 } 3272 tx_pool->free_map[tx_pool->producer_index] = index; 3273 tx_pool->producer_index = 3274 (tx_pool->producer_index + 1) % 3275 tx_pool->num_buffers; 3276 } 3277 /* remove tx_comp scrq*/ 3278 next->tx_comp.first = 0; 3279 3280 txq = netdev_get_tx_queue(adapter->netdev, scrq->pool_index); 3281 netdev_tx_completed_queue(txq, num_packets, total_bytes); 3282 3283 if (atomic_sub_return(num_entries, &scrq->used) <= 3284 (adapter->req_tx_entries_per_subcrq / 2) && 3285 __netif_subqueue_stopped(adapter->netdev, 3286 scrq->pool_index)) { 3287 netif_wake_subqueue(adapter->netdev, scrq->pool_index); 3288 netdev_dbg(adapter->netdev, "Started queue %d\n", 3289 scrq->pool_index); 3290 } 3291 } 3292 3293 enable_scrq_irq(adapter, scrq); 3294 3295 if (pending_scrq(adapter, scrq)) { 3296 disable_scrq_irq(adapter, scrq); 3297 goto restart_loop; 3298 } 3299 3300 return 0; 3301 } 3302 3303 static irqreturn_t ibmvnic_interrupt_tx(int irq, void *instance) 3304 { 3305 struct ibmvnic_sub_crq_queue *scrq = instance; 3306 struct ibmvnic_adapter *adapter = scrq->adapter; 3307 3308 disable_scrq_irq(adapter, scrq); 3309 ibmvnic_complete_tx(adapter, scrq); 3310 3311 return IRQ_HANDLED; 3312 } 3313 3314 static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance) 3315 { 3316 struct ibmvnic_sub_crq_queue *scrq = instance; 3317 struct ibmvnic_adapter *adapter = scrq->adapter; 3318 3319 /* When booting a kdump kernel we can hit pending interrupts 3320 * prior to completing driver initialization. 3321 */ 3322 if (unlikely(adapter->state != VNIC_OPEN)) 3323 return IRQ_NONE; 3324 3325 adapter->rx_stats_buffers[scrq->scrq_num].interrupts++; 3326 3327 if (napi_schedule_prep(&adapter->napi[scrq->scrq_num])) { 3328 disable_scrq_irq(adapter, scrq); 3329 __napi_schedule(&adapter->napi[scrq->scrq_num]); 3330 } 3331 3332 return IRQ_HANDLED; 3333 } 3334 3335 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter) 3336 { 3337 struct device *dev = &adapter->vdev->dev; 3338 struct ibmvnic_sub_crq_queue *scrq; 3339 int i = 0, j = 0; 3340 int rc = 0; 3341 3342 for (i = 0; i < adapter->req_tx_queues; i++) { 3343 netdev_dbg(adapter->netdev, "Initializing tx_scrq[%d] irq\n", 3344 i); 3345 scrq = adapter->tx_scrq[i]; 3346 scrq->irq = irq_create_mapping(NULL, scrq->hw_irq); 3347 3348 if (!scrq->irq) { 3349 rc = -EINVAL; 3350 dev_err(dev, "Error mapping irq\n"); 3351 goto req_tx_irq_failed; 3352 } 3353 3354 snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-tx%d", 3355 adapter->vdev->unit_address, i); 3356 rc = request_irq(scrq->irq, ibmvnic_interrupt_tx, 3357 0, scrq->name, scrq); 3358 3359 if (rc) { 3360 dev_err(dev, "Couldn't register tx irq 0x%x. rc=%d\n", 3361 scrq->irq, rc); 3362 irq_dispose_mapping(scrq->irq); 3363 goto req_tx_irq_failed; 3364 } 3365 } 3366 3367 for (i = 0; i < adapter->req_rx_queues; i++) { 3368 netdev_dbg(adapter->netdev, "Initializing rx_scrq[%d] irq\n", 3369 i); 3370 scrq = adapter->rx_scrq[i]; 3371 scrq->irq = irq_create_mapping(NULL, scrq->hw_irq); 3372 if (!scrq->irq) { 3373 rc = -EINVAL; 3374 dev_err(dev, "Error mapping irq\n"); 3375 goto req_rx_irq_failed; 3376 } 3377 snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-rx%d", 3378 adapter->vdev->unit_address, i); 3379 rc = request_irq(scrq->irq, ibmvnic_interrupt_rx, 3380 0, scrq->name, scrq); 3381 if (rc) { 3382 dev_err(dev, "Couldn't register rx irq 0x%x. rc=%d\n", 3383 scrq->irq, rc); 3384 irq_dispose_mapping(scrq->irq); 3385 goto req_rx_irq_failed; 3386 } 3387 } 3388 return rc; 3389 3390 req_rx_irq_failed: 3391 for (j = 0; j < i; j++) { 3392 free_irq(adapter->rx_scrq[j]->irq, adapter->rx_scrq[j]); 3393 irq_dispose_mapping(adapter->rx_scrq[j]->irq); 3394 } 3395 i = adapter->req_tx_queues; 3396 req_tx_irq_failed: 3397 for (j = 0; j < i; j++) { 3398 free_irq(adapter->tx_scrq[j]->irq, adapter->tx_scrq[j]); 3399 irq_dispose_mapping(adapter->tx_scrq[j]->irq); 3400 } 3401 release_sub_crqs(adapter, 1); 3402 return rc; 3403 } 3404 3405 static int init_sub_crqs(struct ibmvnic_adapter *adapter) 3406 { 3407 struct device *dev = &adapter->vdev->dev; 3408 struct ibmvnic_sub_crq_queue **allqueues; 3409 int registered_queues = 0; 3410 int total_queues; 3411 int more = 0; 3412 int i; 3413 3414 total_queues = adapter->req_tx_queues + adapter->req_rx_queues; 3415 3416 allqueues = kcalloc(total_queues, sizeof(*allqueues), GFP_KERNEL); 3417 if (!allqueues) 3418 return -1; 3419 3420 for (i = 0; i < total_queues; i++) { 3421 allqueues[i] = init_sub_crq_queue(adapter); 3422 if (!allqueues[i]) { 3423 dev_warn(dev, "Couldn't allocate all sub-crqs\n"); 3424 break; 3425 } 3426 registered_queues++; 3427 } 3428 3429 /* Make sure we were able to register the minimum number of queues */ 3430 if (registered_queues < 3431 adapter->min_tx_queues + adapter->min_rx_queues) { 3432 dev_err(dev, "Fatal: Couldn't init min number of sub-crqs\n"); 3433 goto tx_failed; 3434 } 3435 3436 /* Distribute the failed allocated queues*/ 3437 for (i = 0; i < total_queues - registered_queues + more ; i++) { 3438 netdev_dbg(adapter->netdev, "Reducing number of queues\n"); 3439 switch (i % 3) { 3440 case 0: 3441 if (adapter->req_rx_queues > adapter->min_rx_queues) 3442 adapter->req_rx_queues--; 3443 else 3444 more++; 3445 break; 3446 case 1: 3447 if (adapter->req_tx_queues > adapter->min_tx_queues) 3448 adapter->req_tx_queues--; 3449 else 3450 more++; 3451 break; 3452 } 3453 } 3454 3455 adapter->tx_scrq = kcalloc(adapter->req_tx_queues, 3456 sizeof(*adapter->tx_scrq), GFP_KERNEL); 3457 if (!adapter->tx_scrq) 3458 goto tx_failed; 3459 3460 for (i = 0; i < adapter->req_tx_queues; i++) { 3461 adapter->tx_scrq[i] = allqueues[i]; 3462 adapter->tx_scrq[i]->pool_index = i; 3463 adapter->num_active_tx_scrqs++; 3464 } 3465 3466 adapter->rx_scrq = kcalloc(adapter->req_rx_queues, 3467 sizeof(*adapter->rx_scrq), GFP_KERNEL); 3468 if (!adapter->rx_scrq) 3469 goto rx_failed; 3470 3471 for (i = 0; i < adapter->req_rx_queues; i++) { 3472 adapter->rx_scrq[i] = allqueues[i + adapter->req_tx_queues]; 3473 adapter->rx_scrq[i]->scrq_num = i; 3474 adapter->num_active_rx_scrqs++; 3475 } 3476 3477 kfree(allqueues); 3478 return 0; 3479 3480 rx_failed: 3481 kfree(adapter->tx_scrq); 3482 adapter->tx_scrq = NULL; 3483 tx_failed: 3484 for (i = 0; i < registered_queues; i++) 3485 release_sub_crq_queue(adapter, allqueues[i], 1); 3486 kfree(allqueues); 3487 return -1; 3488 } 3489 3490 static void send_request_cap(struct ibmvnic_adapter *adapter, int retry) 3491 { 3492 struct device *dev = &adapter->vdev->dev; 3493 union ibmvnic_crq crq; 3494 int max_entries; 3495 3496 if (!retry) { 3497 /* Sub-CRQ entries are 32 byte long */ 3498 int entries_page = 4 * PAGE_SIZE / (sizeof(u64) * 4); 3499 3500 if (adapter->min_tx_entries_per_subcrq > entries_page || 3501 adapter->min_rx_add_entries_per_subcrq > entries_page) { 3502 dev_err(dev, "Fatal, invalid entries per sub-crq\n"); 3503 return; 3504 } 3505 3506 if (adapter->desired.mtu) 3507 adapter->req_mtu = adapter->desired.mtu; 3508 else 3509 adapter->req_mtu = adapter->netdev->mtu + ETH_HLEN; 3510 3511 if (!adapter->desired.tx_entries) 3512 adapter->desired.tx_entries = 3513 adapter->max_tx_entries_per_subcrq; 3514 if (!adapter->desired.rx_entries) 3515 adapter->desired.rx_entries = 3516 adapter->max_rx_add_entries_per_subcrq; 3517 3518 max_entries = IBMVNIC_MAX_LTB_SIZE / 3519 (adapter->req_mtu + IBMVNIC_BUFFER_HLEN); 3520 3521 if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) * 3522 adapter->desired.tx_entries > IBMVNIC_MAX_LTB_SIZE) { 3523 adapter->desired.tx_entries = max_entries; 3524 } 3525 3526 if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) * 3527 adapter->desired.rx_entries > IBMVNIC_MAX_LTB_SIZE) { 3528 adapter->desired.rx_entries = max_entries; 3529 } 3530 3531 if (adapter->desired.tx_entries) 3532 adapter->req_tx_entries_per_subcrq = 3533 adapter->desired.tx_entries; 3534 else 3535 adapter->req_tx_entries_per_subcrq = 3536 adapter->max_tx_entries_per_subcrq; 3537 3538 if (adapter->desired.rx_entries) 3539 adapter->req_rx_add_entries_per_subcrq = 3540 adapter->desired.rx_entries; 3541 else 3542 adapter->req_rx_add_entries_per_subcrq = 3543 adapter->max_rx_add_entries_per_subcrq; 3544 3545 if (adapter->desired.tx_queues) 3546 adapter->req_tx_queues = 3547 adapter->desired.tx_queues; 3548 else 3549 adapter->req_tx_queues = 3550 adapter->opt_tx_comp_sub_queues; 3551 3552 if (adapter->desired.rx_queues) 3553 adapter->req_rx_queues = 3554 adapter->desired.rx_queues; 3555 else 3556 adapter->req_rx_queues = 3557 adapter->opt_rx_comp_queues; 3558 3559 adapter->req_rx_add_queues = adapter->max_rx_add_queues; 3560 } 3561 3562 memset(&crq, 0, sizeof(crq)); 3563 crq.request_capability.first = IBMVNIC_CRQ_CMD; 3564 crq.request_capability.cmd = REQUEST_CAPABILITY; 3565 3566 crq.request_capability.capability = cpu_to_be16(REQ_TX_QUEUES); 3567 crq.request_capability.number = cpu_to_be64(adapter->req_tx_queues); 3568 atomic_inc(&adapter->running_cap_crqs); 3569 ibmvnic_send_crq(adapter, &crq); 3570 3571 crq.request_capability.capability = cpu_to_be16(REQ_RX_QUEUES); 3572 crq.request_capability.number = cpu_to_be64(adapter->req_rx_queues); 3573 atomic_inc(&adapter->running_cap_crqs); 3574 ibmvnic_send_crq(adapter, &crq); 3575 3576 crq.request_capability.capability = cpu_to_be16(REQ_RX_ADD_QUEUES); 3577 crq.request_capability.number = cpu_to_be64(adapter->req_rx_add_queues); 3578 atomic_inc(&adapter->running_cap_crqs); 3579 ibmvnic_send_crq(adapter, &crq); 3580 3581 crq.request_capability.capability = 3582 cpu_to_be16(REQ_TX_ENTRIES_PER_SUBCRQ); 3583 crq.request_capability.number = 3584 cpu_to_be64(adapter->req_tx_entries_per_subcrq); 3585 atomic_inc(&adapter->running_cap_crqs); 3586 ibmvnic_send_crq(adapter, &crq); 3587 3588 crq.request_capability.capability = 3589 cpu_to_be16(REQ_RX_ADD_ENTRIES_PER_SUBCRQ); 3590 crq.request_capability.number = 3591 cpu_to_be64(adapter->req_rx_add_entries_per_subcrq); 3592 atomic_inc(&adapter->running_cap_crqs); 3593 ibmvnic_send_crq(adapter, &crq); 3594 3595 crq.request_capability.capability = cpu_to_be16(REQ_MTU); 3596 crq.request_capability.number = cpu_to_be64(adapter->req_mtu); 3597 atomic_inc(&adapter->running_cap_crqs); 3598 ibmvnic_send_crq(adapter, &crq); 3599 3600 if (adapter->netdev->flags & IFF_PROMISC) { 3601 if (adapter->promisc_supported) { 3602 crq.request_capability.capability = 3603 cpu_to_be16(PROMISC_REQUESTED); 3604 crq.request_capability.number = cpu_to_be64(1); 3605 atomic_inc(&adapter->running_cap_crqs); 3606 ibmvnic_send_crq(adapter, &crq); 3607 } 3608 } else { 3609 crq.request_capability.capability = 3610 cpu_to_be16(PROMISC_REQUESTED); 3611 crq.request_capability.number = cpu_to_be64(0); 3612 atomic_inc(&adapter->running_cap_crqs); 3613 ibmvnic_send_crq(adapter, &crq); 3614 } 3615 } 3616 3617 static int pending_scrq(struct ibmvnic_adapter *adapter, 3618 struct ibmvnic_sub_crq_queue *scrq) 3619 { 3620 union sub_crq *entry = &scrq->msgs[scrq->cur]; 3621 int rc; 3622 3623 rc = !!(entry->generic.first & IBMVNIC_CRQ_CMD_RSP); 3624 3625 /* Ensure that the SCRQ valid flag is loaded prior to loading the 3626 * contents of the SCRQ descriptor 3627 */ 3628 dma_rmb(); 3629 3630 return rc; 3631 } 3632 3633 static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *adapter, 3634 struct ibmvnic_sub_crq_queue *scrq) 3635 { 3636 union sub_crq *entry; 3637 unsigned long flags; 3638 3639 spin_lock_irqsave(&scrq->lock, flags); 3640 entry = &scrq->msgs[scrq->cur]; 3641 if (entry->generic.first & IBMVNIC_CRQ_CMD_RSP) { 3642 if (++scrq->cur == scrq->size) 3643 scrq->cur = 0; 3644 } else { 3645 entry = NULL; 3646 } 3647 spin_unlock_irqrestore(&scrq->lock, flags); 3648 3649 /* Ensure that the SCRQ valid flag is loaded prior to loading the 3650 * contents of the SCRQ descriptor 3651 */ 3652 dma_rmb(); 3653 3654 return entry; 3655 } 3656 3657 static union ibmvnic_crq *ibmvnic_next_crq(struct ibmvnic_adapter *adapter) 3658 { 3659 struct ibmvnic_crq_queue *queue = &adapter->crq; 3660 union ibmvnic_crq *crq; 3661 3662 crq = &queue->msgs[queue->cur]; 3663 if (crq->generic.first & IBMVNIC_CRQ_CMD_RSP) { 3664 if (++queue->cur == queue->size) 3665 queue->cur = 0; 3666 } else { 3667 crq = NULL; 3668 } 3669 3670 return crq; 3671 } 3672 3673 static void print_subcrq_error(struct device *dev, int rc, const char *func) 3674 { 3675 switch (rc) { 3676 case H_PARAMETER: 3677 dev_warn_ratelimited(dev, 3678 "%s failed: Send request is malformed or adapter failover pending. (rc=%d)\n", 3679 func, rc); 3680 break; 3681 case H_CLOSED: 3682 dev_warn_ratelimited(dev, 3683 "%s failed: Backing queue closed. Adapter is down or failover pending. (rc=%d)\n", 3684 func, rc); 3685 break; 3686 default: 3687 dev_err_ratelimited(dev, "%s failed: (rc=%d)\n", func, rc); 3688 break; 3689 } 3690 } 3691 3692 static int send_subcrq_indirect(struct ibmvnic_adapter *adapter, 3693 u64 remote_handle, u64 ioba, u64 num_entries) 3694 { 3695 unsigned int ua = adapter->vdev->unit_address; 3696 struct device *dev = &adapter->vdev->dev; 3697 int rc; 3698 3699 /* Make sure the hypervisor sees the complete request */ 3700 dma_wmb(); 3701 rc = plpar_hcall_norets(H_SEND_SUB_CRQ_INDIRECT, ua, 3702 cpu_to_be64(remote_handle), 3703 ioba, num_entries); 3704 3705 if (rc) 3706 print_subcrq_error(dev, rc, __func__); 3707 3708 return rc; 3709 } 3710 3711 static int ibmvnic_send_crq(struct ibmvnic_adapter *adapter, 3712 union ibmvnic_crq *crq) 3713 { 3714 unsigned int ua = adapter->vdev->unit_address; 3715 struct device *dev = &adapter->vdev->dev; 3716 u64 *u64_crq = (u64 *)crq; 3717 int rc; 3718 3719 netdev_dbg(adapter->netdev, "Sending CRQ: %016lx %016lx\n", 3720 (unsigned long)cpu_to_be64(u64_crq[0]), 3721 (unsigned long)cpu_to_be64(u64_crq[1])); 3722 3723 if (!adapter->crq.active && 3724 crq->generic.first != IBMVNIC_CRQ_INIT_CMD) { 3725 dev_warn(dev, "Invalid request detected while CRQ is inactive, possible device state change during reset\n"); 3726 return -EINVAL; 3727 } 3728 3729 /* Make sure the hypervisor sees the complete request */ 3730 dma_wmb(); 3731 3732 rc = plpar_hcall_norets(H_SEND_CRQ, ua, 3733 cpu_to_be64(u64_crq[0]), 3734 cpu_to_be64(u64_crq[1])); 3735 3736 if (rc) { 3737 if (rc == H_CLOSED) { 3738 dev_warn(dev, "CRQ Queue closed\n"); 3739 /* do not reset, report the fail, wait for passive init from server */ 3740 } 3741 3742 dev_warn(dev, "Send error (rc=%d)\n", rc); 3743 } 3744 3745 return rc; 3746 } 3747 3748 static int ibmvnic_send_crq_init(struct ibmvnic_adapter *adapter) 3749 { 3750 struct device *dev = &adapter->vdev->dev; 3751 union ibmvnic_crq crq; 3752 int retries = 100; 3753 int rc; 3754 3755 memset(&crq, 0, sizeof(crq)); 3756 crq.generic.first = IBMVNIC_CRQ_INIT_CMD; 3757 crq.generic.cmd = IBMVNIC_CRQ_INIT; 3758 netdev_dbg(adapter->netdev, "Sending CRQ init\n"); 3759 3760 do { 3761 rc = ibmvnic_send_crq(adapter, &crq); 3762 if (rc != H_CLOSED) 3763 break; 3764 retries--; 3765 msleep(50); 3766 3767 } while (retries > 0); 3768 3769 if (rc) { 3770 dev_err(dev, "Failed to send init request, rc = %d\n", rc); 3771 return rc; 3772 } 3773 3774 return 0; 3775 } 3776 3777 static int send_version_xchg(struct ibmvnic_adapter *adapter) 3778 { 3779 union ibmvnic_crq crq; 3780 3781 memset(&crq, 0, sizeof(crq)); 3782 crq.version_exchange.first = IBMVNIC_CRQ_CMD; 3783 crq.version_exchange.cmd = VERSION_EXCHANGE; 3784 crq.version_exchange.version = cpu_to_be16(ibmvnic_version); 3785 3786 return ibmvnic_send_crq(adapter, &crq); 3787 } 3788 3789 struct vnic_login_client_data { 3790 u8 type; 3791 __be16 len; 3792 char name[]; 3793 } __packed; 3794 3795 static int vnic_client_data_len(struct ibmvnic_adapter *adapter) 3796 { 3797 int len; 3798 3799 /* Calculate the amount of buffer space needed for the 3800 * vnic client data in the login buffer. There are four entries, 3801 * OS name, LPAR name, device name, and a null last entry. 3802 */ 3803 len = 4 * sizeof(struct vnic_login_client_data); 3804 len += 6; /* "Linux" plus NULL */ 3805 len += strlen(utsname()->nodename) + 1; 3806 len += strlen(adapter->netdev->name) + 1; 3807 3808 return len; 3809 } 3810 3811 static void vnic_add_client_data(struct ibmvnic_adapter *adapter, 3812 struct vnic_login_client_data *vlcd) 3813 { 3814 const char *os_name = "Linux"; 3815 int len; 3816 3817 /* Type 1 - LPAR OS */ 3818 vlcd->type = 1; 3819 len = strlen(os_name) + 1; 3820 vlcd->len = cpu_to_be16(len); 3821 strncpy(vlcd->name, os_name, len); 3822 vlcd = (struct vnic_login_client_data *)(vlcd->name + len); 3823 3824 /* Type 2 - LPAR name */ 3825 vlcd->type = 2; 3826 len = strlen(utsname()->nodename) + 1; 3827 vlcd->len = cpu_to_be16(len); 3828 strncpy(vlcd->name, utsname()->nodename, len); 3829 vlcd = (struct vnic_login_client_data *)(vlcd->name + len); 3830 3831 /* Type 3 - device name */ 3832 vlcd->type = 3; 3833 len = strlen(adapter->netdev->name) + 1; 3834 vlcd->len = cpu_to_be16(len); 3835 strncpy(vlcd->name, adapter->netdev->name, len); 3836 } 3837 3838 static int send_login(struct ibmvnic_adapter *adapter) 3839 { 3840 struct ibmvnic_login_rsp_buffer *login_rsp_buffer; 3841 struct ibmvnic_login_buffer *login_buffer; 3842 struct device *dev = &adapter->vdev->dev; 3843 struct vnic_login_client_data *vlcd; 3844 dma_addr_t rsp_buffer_token; 3845 dma_addr_t buffer_token; 3846 size_t rsp_buffer_size; 3847 union ibmvnic_crq crq; 3848 int client_data_len; 3849 size_t buffer_size; 3850 __be64 *tx_list_p; 3851 __be64 *rx_list_p; 3852 int rc; 3853 int i; 3854 3855 if (!adapter->tx_scrq || !adapter->rx_scrq) { 3856 netdev_err(adapter->netdev, 3857 "RX or TX queues are not allocated, device login failed\n"); 3858 return -1; 3859 } 3860 3861 release_login_buffer(adapter); 3862 release_login_rsp_buffer(adapter); 3863 3864 client_data_len = vnic_client_data_len(adapter); 3865 3866 buffer_size = 3867 sizeof(struct ibmvnic_login_buffer) + 3868 sizeof(u64) * (adapter->req_tx_queues + adapter->req_rx_queues) + 3869 client_data_len; 3870 3871 login_buffer = kzalloc(buffer_size, GFP_ATOMIC); 3872 if (!login_buffer) 3873 goto buf_alloc_failed; 3874 3875 buffer_token = dma_map_single(dev, login_buffer, buffer_size, 3876 DMA_TO_DEVICE); 3877 if (dma_mapping_error(dev, buffer_token)) { 3878 dev_err(dev, "Couldn't map login buffer\n"); 3879 goto buf_map_failed; 3880 } 3881 3882 rsp_buffer_size = sizeof(struct ibmvnic_login_rsp_buffer) + 3883 sizeof(u64) * adapter->req_tx_queues + 3884 sizeof(u64) * adapter->req_rx_queues + 3885 sizeof(u64) * adapter->req_rx_queues + 3886 sizeof(u8) * IBMVNIC_TX_DESC_VERSIONS; 3887 3888 login_rsp_buffer = kmalloc(rsp_buffer_size, GFP_ATOMIC); 3889 if (!login_rsp_buffer) 3890 goto buf_rsp_alloc_failed; 3891 3892 rsp_buffer_token = dma_map_single(dev, login_rsp_buffer, 3893 rsp_buffer_size, DMA_FROM_DEVICE); 3894 if (dma_mapping_error(dev, rsp_buffer_token)) { 3895 dev_err(dev, "Couldn't map login rsp buffer\n"); 3896 goto buf_rsp_map_failed; 3897 } 3898 3899 adapter->login_buf = login_buffer; 3900 adapter->login_buf_token = buffer_token; 3901 adapter->login_buf_sz = buffer_size; 3902 adapter->login_rsp_buf = login_rsp_buffer; 3903 adapter->login_rsp_buf_token = rsp_buffer_token; 3904 adapter->login_rsp_buf_sz = rsp_buffer_size; 3905 3906 login_buffer->len = cpu_to_be32(buffer_size); 3907 login_buffer->version = cpu_to_be32(INITIAL_VERSION_LB); 3908 login_buffer->num_txcomp_subcrqs = cpu_to_be32(adapter->req_tx_queues); 3909 login_buffer->off_txcomp_subcrqs = 3910 cpu_to_be32(sizeof(struct ibmvnic_login_buffer)); 3911 login_buffer->num_rxcomp_subcrqs = cpu_to_be32(adapter->req_rx_queues); 3912 login_buffer->off_rxcomp_subcrqs = 3913 cpu_to_be32(sizeof(struct ibmvnic_login_buffer) + 3914 sizeof(u64) * adapter->req_tx_queues); 3915 login_buffer->login_rsp_ioba = cpu_to_be32(rsp_buffer_token); 3916 login_buffer->login_rsp_len = cpu_to_be32(rsp_buffer_size); 3917 3918 tx_list_p = (__be64 *)((char *)login_buffer + 3919 sizeof(struct ibmvnic_login_buffer)); 3920 rx_list_p = (__be64 *)((char *)login_buffer + 3921 sizeof(struct ibmvnic_login_buffer) + 3922 sizeof(u64) * adapter->req_tx_queues); 3923 3924 for (i = 0; i < adapter->req_tx_queues; i++) { 3925 if (adapter->tx_scrq[i]) { 3926 tx_list_p[i] = 3927 cpu_to_be64(adapter->tx_scrq[i]->crq_num); 3928 } 3929 } 3930 3931 for (i = 0; i < adapter->req_rx_queues; i++) { 3932 if (adapter->rx_scrq[i]) { 3933 rx_list_p[i] = 3934 cpu_to_be64(adapter->rx_scrq[i]->crq_num); 3935 } 3936 } 3937 3938 /* Insert vNIC login client data */ 3939 vlcd = (struct vnic_login_client_data *) 3940 ((char *)rx_list_p + (sizeof(u64) * adapter->req_rx_queues)); 3941 login_buffer->client_data_offset = 3942 cpu_to_be32((char *)vlcd - (char *)login_buffer); 3943 login_buffer->client_data_len = cpu_to_be32(client_data_len); 3944 3945 vnic_add_client_data(adapter, vlcd); 3946 3947 netdev_dbg(adapter->netdev, "Login Buffer:\n"); 3948 for (i = 0; i < (adapter->login_buf_sz - 1) / 8 + 1; i++) { 3949 netdev_dbg(adapter->netdev, "%016lx\n", 3950 ((unsigned long *)(adapter->login_buf))[i]); 3951 } 3952 3953 memset(&crq, 0, sizeof(crq)); 3954 crq.login.first = IBMVNIC_CRQ_CMD; 3955 crq.login.cmd = LOGIN; 3956 crq.login.ioba = cpu_to_be32(buffer_token); 3957 crq.login.len = cpu_to_be32(buffer_size); 3958 3959 adapter->login_pending = true; 3960 rc = ibmvnic_send_crq(adapter, &crq); 3961 if (rc) { 3962 adapter->login_pending = false; 3963 netdev_err(adapter->netdev, "Failed to send login, rc=%d\n", rc); 3964 goto buf_rsp_map_failed; 3965 } 3966 3967 return 0; 3968 3969 buf_rsp_map_failed: 3970 kfree(login_rsp_buffer); 3971 adapter->login_rsp_buf = NULL; 3972 buf_rsp_alloc_failed: 3973 dma_unmap_single(dev, buffer_token, buffer_size, DMA_TO_DEVICE); 3974 buf_map_failed: 3975 kfree(login_buffer); 3976 adapter->login_buf = NULL; 3977 buf_alloc_failed: 3978 return -1; 3979 } 3980 3981 static int send_request_map(struct ibmvnic_adapter *adapter, dma_addr_t addr, 3982 u32 len, u8 map_id) 3983 { 3984 union ibmvnic_crq crq; 3985 3986 memset(&crq, 0, sizeof(crq)); 3987 crq.request_map.first = IBMVNIC_CRQ_CMD; 3988 crq.request_map.cmd = REQUEST_MAP; 3989 crq.request_map.map_id = map_id; 3990 crq.request_map.ioba = cpu_to_be32(addr); 3991 crq.request_map.len = cpu_to_be32(len); 3992 return ibmvnic_send_crq(adapter, &crq); 3993 } 3994 3995 static int send_request_unmap(struct ibmvnic_adapter *adapter, u8 map_id) 3996 { 3997 union ibmvnic_crq crq; 3998 3999 memset(&crq, 0, sizeof(crq)); 4000 crq.request_unmap.first = IBMVNIC_CRQ_CMD; 4001 crq.request_unmap.cmd = REQUEST_UNMAP; 4002 crq.request_unmap.map_id = map_id; 4003 return ibmvnic_send_crq(adapter, &crq); 4004 } 4005 4006 static void send_query_map(struct ibmvnic_adapter *adapter) 4007 { 4008 union ibmvnic_crq crq; 4009 4010 memset(&crq, 0, sizeof(crq)); 4011 crq.query_map.first = IBMVNIC_CRQ_CMD; 4012 crq.query_map.cmd = QUERY_MAP; 4013 ibmvnic_send_crq(adapter, &crq); 4014 } 4015 4016 /* Send a series of CRQs requesting various capabilities of the VNIC server */ 4017 static void send_query_cap(struct ibmvnic_adapter *adapter) 4018 { 4019 union ibmvnic_crq crq; 4020 4021 atomic_set(&adapter->running_cap_crqs, 0); 4022 memset(&crq, 0, sizeof(crq)); 4023 crq.query_capability.first = IBMVNIC_CRQ_CMD; 4024 crq.query_capability.cmd = QUERY_CAPABILITY; 4025 4026 crq.query_capability.capability = cpu_to_be16(MIN_TX_QUEUES); 4027 atomic_inc(&adapter->running_cap_crqs); 4028 ibmvnic_send_crq(adapter, &crq); 4029 4030 crq.query_capability.capability = cpu_to_be16(MIN_RX_QUEUES); 4031 atomic_inc(&adapter->running_cap_crqs); 4032 ibmvnic_send_crq(adapter, &crq); 4033 4034 crq.query_capability.capability = cpu_to_be16(MIN_RX_ADD_QUEUES); 4035 atomic_inc(&adapter->running_cap_crqs); 4036 ibmvnic_send_crq(adapter, &crq); 4037 4038 crq.query_capability.capability = cpu_to_be16(MAX_TX_QUEUES); 4039 atomic_inc(&adapter->running_cap_crqs); 4040 ibmvnic_send_crq(adapter, &crq); 4041 4042 crq.query_capability.capability = cpu_to_be16(MAX_RX_QUEUES); 4043 atomic_inc(&adapter->running_cap_crqs); 4044 ibmvnic_send_crq(adapter, &crq); 4045 4046 crq.query_capability.capability = cpu_to_be16(MAX_RX_ADD_QUEUES); 4047 atomic_inc(&adapter->running_cap_crqs); 4048 ibmvnic_send_crq(adapter, &crq); 4049 4050 crq.query_capability.capability = 4051 cpu_to_be16(MIN_TX_ENTRIES_PER_SUBCRQ); 4052 atomic_inc(&adapter->running_cap_crqs); 4053 ibmvnic_send_crq(adapter, &crq); 4054 4055 crq.query_capability.capability = 4056 cpu_to_be16(MIN_RX_ADD_ENTRIES_PER_SUBCRQ); 4057 atomic_inc(&adapter->running_cap_crqs); 4058 ibmvnic_send_crq(adapter, &crq); 4059 4060 crq.query_capability.capability = 4061 cpu_to_be16(MAX_TX_ENTRIES_PER_SUBCRQ); 4062 atomic_inc(&adapter->running_cap_crqs); 4063 ibmvnic_send_crq(adapter, &crq); 4064 4065 crq.query_capability.capability = 4066 cpu_to_be16(MAX_RX_ADD_ENTRIES_PER_SUBCRQ); 4067 atomic_inc(&adapter->running_cap_crqs); 4068 ibmvnic_send_crq(adapter, &crq); 4069 4070 crq.query_capability.capability = cpu_to_be16(TCP_IP_OFFLOAD); 4071 atomic_inc(&adapter->running_cap_crqs); 4072 ibmvnic_send_crq(adapter, &crq); 4073 4074 crq.query_capability.capability = cpu_to_be16(PROMISC_SUPPORTED); 4075 atomic_inc(&adapter->running_cap_crqs); 4076 ibmvnic_send_crq(adapter, &crq); 4077 4078 crq.query_capability.capability = cpu_to_be16(MIN_MTU); 4079 atomic_inc(&adapter->running_cap_crqs); 4080 ibmvnic_send_crq(adapter, &crq); 4081 4082 crq.query_capability.capability = cpu_to_be16(MAX_MTU); 4083 atomic_inc(&adapter->running_cap_crqs); 4084 ibmvnic_send_crq(adapter, &crq); 4085 4086 crq.query_capability.capability = cpu_to_be16(MAX_MULTICAST_FILTERS); 4087 atomic_inc(&adapter->running_cap_crqs); 4088 ibmvnic_send_crq(adapter, &crq); 4089 4090 crq.query_capability.capability = cpu_to_be16(VLAN_HEADER_INSERTION); 4091 atomic_inc(&adapter->running_cap_crqs); 4092 ibmvnic_send_crq(adapter, &crq); 4093 4094 crq.query_capability.capability = cpu_to_be16(RX_VLAN_HEADER_INSERTION); 4095 atomic_inc(&adapter->running_cap_crqs); 4096 ibmvnic_send_crq(adapter, &crq); 4097 4098 crq.query_capability.capability = cpu_to_be16(MAX_TX_SG_ENTRIES); 4099 atomic_inc(&adapter->running_cap_crqs); 4100 ibmvnic_send_crq(adapter, &crq); 4101 4102 crq.query_capability.capability = cpu_to_be16(RX_SG_SUPPORTED); 4103 atomic_inc(&adapter->running_cap_crqs); 4104 ibmvnic_send_crq(adapter, &crq); 4105 4106 crq.query_capability.capability = cpu_to_be16(OPT_TX_COMP_SUB_QUEUES); 4107 atomic_inc(&adapter->running_cap_crqs); 4108 ibmvnic_send_crq(adapter, &crq); 4109 4110 crq.query_capability.capability = cpu_to_be16(OPT_RX_COMP_QUEUES); 4111 atomic_inc(&adapter->running_cap_crqs); 4112 ibmvnic_send_crq(adapter, &crq); 4113 4114 crq.query_capability.capability = 4115 cpu_to_be16(OPT_RX_BUFADD_Q_PER_RX_COMP_Q); 4116 atomic_inc(&adapter->running_cap_crqs); 4117 ibmvnic_send_crq(adapter, &crq); 4118 4119 crq.query_capability.capability = 4120 cpu_to_be16(OPT_TX_ENTRIES_PER_SUBCRQ); 4121 atomic_inc(&adapter->running_cap_crqs); 4122 ibmvnic_send_crq(adapter, &crq); 4123 4124 crq.query_capability.capability = 4125 cpu_to_be16(OPT_RXBA_ENTRIES_PER_SUBCRQ); 4126 atomic_inc(&adapter->running_cap_crqs); 4127 ibmvnic_send_crq(adapter, &crq); 4128 4129 crq.query_capability.capability = cpu_to_be16(TX_RX_DESC_REQ); 4130 atomic_inc(&adapter->running_cap_crqs); 4131 ibmvnic_send_crq(adapter, &crq); 4132 } 4133 4134 static void send_query_ip_offload(struct ibmvnic_adapter *adapter) 4135 { 4136 int buf_sz = sizeof(struct ibmvnic_query_ip_offload_buffer); 4137 struct device *dev = &adapter->vdev->dev; 4138 union ibmvnic_crq crq; 4139 4140 adapter->ip_offload_tok = 4141 dma_map_single(dev, 4142 &adapter->ip_offload_buf, 4143 buf_sz, 4144 DMA_FROM_DEVICE); 4145 4146 if (dma_mapping_error(dev, adapter->ip_offload_tok)) { 4147 if (!firmware_has_feature(FW_FEATURE_CMO)) 4148 dev_err(dev, "Couldn't map offload buffer\n"); 4149 return; 4150 } 4151 4152 memset(&crq, 0, sizeof(crq)); 4153 crq.query_ip_offload.first = IBMVNIC_CRQ_CMD; 4154 crq.query_ip_offload.cmd = QUERY_IP_OFFLOAD; 4155 crq.query_ip_offload.len = cpu_to_be32(buf_sz); 4156 crq.query_ip_offload.ioba = 4157 cpu_to_be32(adapter->ip_offload_tok); 4158 4159 ibmvnic_send_crq(adapter, &crq); 4160 } 4161 4162 static void send_control_ip_offload(struct ibmvnic_adapter *adapter) 4163 { 4164 struct ibmvnic_control_ip_offload_buffer *ctrl_buf = &adapter->ip_offload_ctrl; 4165 struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf; 4166 struct device *dev = &adapter->vdev->dev; 4167 netdev_features_t old_hw_features = 0; 4168 union ibmvnic_crq crq; 4169 4170 adapter->ip_offload_ctrl_tok = 4171 dma_map_single(dev, 4172 ctrl_buf, 4173 sizeof(adapter->ip_offload_ctrl), 4174 DMA_TO_DEVICE); 4175 4176 if (dma_mapping_error(dev, adapter->ip_offload_ctrl_tok)) { 4177 dev_err(dev, "Couldn't map ip offload control buffer\n"); 4178 return; 4179 } 4180 4181 ctrl_buf->len = cpu_to_be32(sizeof(adapter->ip_offload_ctrl)); 4182 ctrl_buf->version = cpu_to_be32(INITIAL_VERSION_IOB); 4183 ctrl_buf->ipv4_chksum = buf->ipv4_chksum; 4184 ctrl_buf->ipv6_chksum = buf->ipv6_chksum; 4185 ctrl_buf->tcp_ipv4_chksum = buf->tcp_ipv4_chksum; 4186 ctrl_buf->udp_ipv4_chksum = buf->udp_ipv4_chksum; 4187 ctrl_buf->tcp_ipv6_chksum = buf->tcp_ipv6_chksum; 4188 ctrl_buf->udp_ipv6_chksum = buf->udp_ipv6_chksum; 4189 ctrl_buf->large_tx_ipv4 = buf->large_tx_ipv4; 4190 ctrl_buf->large_tx_ipv6 = buf->large_tx_ipv6; 4191 4192 /* large_rx disabled for now, additional features needed */ 4193 ctrl_buf->large_rx_ipv4 = 0; 4194 ctrl_buf->large_rx_ipv6 = 0; 4195 4196 if (adapter->state != VNIC_PROBING) { 4197 old_hw_features = adapter->netdev->hw_features; 4198 adapter->netdev->hw_features = 0; 4199 } 4200 4201 adapter->netdev->hw_features = NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO; 4202 4203 if (buf->tcp_ipv4_chksum || buf->udp_ipv4_chksum) 4204 adapter->netdev->hw_features |= NETIF_F_IP_CSUM; 4205 4206 if (buf->tcp_ipv6_chksum || buf->udp_ipv6_chksum) 4207 adapter->netdev->hw_features |= NETIF_F_IPV6_CSUM; 4208 4209 if ((adapter->netdev->features & 4210 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM))) 4211 adapter->netdev->hw_features |= NETIF_F_RXCSUM; 4212 4213 if (buf->large_tx_ipv4) 4214 adapter->netdev->hw_features |= NETIF_F_TSO; 4215 if (buf->large_tx_ipv6) 4216 adapter->netdev->hw_features |= NETIF_F_TSO6; 4217 4218 if (adapter->state == VNIC_PROBING) { 4219 adapter->netdev->features |= adapter->netdev->hw_features; 4220 } else if (old_hw_features != adapter->netdev->hw_features) { 4221 netdev_features_t tmp = 0; 4222 4223 /* disable features no longer supported */ 4224 adapter->netdev->features &= adapter->netdev->hw_features; 4225 /* turn on features now supported if previously enabled */ 4226 tmp = (old_hw_features ^ adapter->netdev->hw_features) & 4227 adapter->netdev->hw_features; 4228 adapter->netdev->features |= 4229 tmp & adapter->netdev->wanted_features; 4230 } 4231 4232 memset(&crq, 0, sizeof(crq)); 4233 crq.control_ip_offload.first = IBMVNIC_CRQ_CMD; 4234 crq.control_ip_offload.cmd = CONTROL_IP_OFFLOAD; 4235 crq.control_ip_offload.len = 4236 cpu_to_be32(sizeof(adapter->ip_offload_ctrl)); 4237 crq.control_ip_offload.ioba = cpu_to_be32(adapter->ip_offload_ctrl_tok); 4238 ibmvnic_send_crq(adapter, &crq); 4239 } 4240 4241 static void handle_vpd_size_rsp(union ibmvnic_crq *crq, 4242 struct ibmvnic_adapter *adapter) 4243 { 4244 struct device *dev = &adapter->vdev->dev; 4245 4246 if (crq->get_vpd_size_rsp.rc.code) { 4247 dev_err(dev, "Error retrieving VPD size, rc=%x\n", 4248 crq->get_vpd_size_rsp.rc.code); 4249 complete(&adapter->fw_done); 4250 return; 4251 } 4252 4253 adapter->vpd->len = be64_to_cpu(crq->get_vpd_size_rsp.len); 4254 complete(&adapter->fw_done); 4255 } 4256 4257 static void handle_vpd_rsp(union ibmvnic_crq *crq, 4258 struct ibmvnic_adapter *adapter) 4259 { 4260 struct device *dev = &adapter->vdev->dev; 4261 unsigned char *substr = NULL; 4262 u8 fw_level_len = 0; 4263 4264 memset(adapter->fw_version, 0, 32); 4265 4266 dma_unmap_single(dev, adapter->vpd->dma_addr, adapter->vpd->len, 4267 DMA_FROM_DEVICE); 4268 4269 if (crq->get_vpd_rsp.rc.code) { 4270 dev_err(dev, "Error retrieving VPD from device, rc=%x\n", 4271 crq->get_vpd_rsp.rc.code); 4272 goto complete; 4273 } 4274 4275 /* get the position of the firmware version info 4276 * located after the ASCII 'RM' substring in the buffer 4277 */ 4278 substr = strnstr(adapter->vpd->buff, "RM", adapter->vpd->len); 4279 if (!substr) { 4280 dev_info(dev, "Warning - No FW level has been provided in the VPD buffer by the VIOS Server\n"); 4281 goto complete; 4282 } 4283 4284 /* get length of firmware level ASCII substring */ 4285 if ((substr + 2) < (adapter->vpd->buff + adapter->vpd->len)) { 4286 fw_level_len = *(substr + 2); 4287 } else { 4288 dev_info(dev, "Length of FW substr extrapolated VDP buff\n"); 4289 goto complete; 4290 } 4291 4292 /* copy firmware version string from vpd into adapter */ 4293 if ((substr + 3 + fw_level_len) < 4294 (adapter->vpd->buff + adapter->vpd->len)) { 4295 strncpy((char *)adapter->fw_version, substr + 3, fw_level_len); 4296 } else { 4297 dev_info(dev, "FW substr extrapolated VPD buff\n"); 4298 } 4299 4300 complete: 4301 if (adapter->fw_version[0] == '\0') 4302 strncpy((char *)adapter->fw_version, "N/A", 3 * sizeof(char)); 4303 complete(&adapter->fw_done); 4304 } 4305 4306 static void handle_query_ip_offload_rsp(struct ibmvnic_adapter *adapter) 4307 { 4308 struct device *dev = &adapter->vdev->dev; 4309 struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf; 4310 int i; 4311 4312 dma_unmap_single(dev, adapter->ip_offload_tok, 4313 sizeof(adapter->ip_offload_buf), DMA_FROM_DEVICE); 4314 4315 netdev_dbg(adapter->netdev, "Query IP Offload Buffer:\n"); 4316 for (i = 0; i < (sizeof(adapter->ip_offload_buf) - 1) / 8 + 1; i++) 4317 netdev_dbg(adapter->netdev, "%016lx\n", 4318 ((unsigned long *)(buf))[i]); 4319 4320 netdev_dbg(adapter->netdev, "ipv4_chksum = %d\n", buf->ipv4_chksum); 4321 netdev_dbg(adapter->netdev, "ipv6_chksum = %d\n", buf->ipv6_chksum); 4322 netdev_dbg(adapter->netdev, "tcp_ipv4_chksum = %d\n", 4323 buf->tcp_ipv4_chksum); 4324 netdev_dbg(adapter->netdev, "tcp_ipv6_chksum = %d\n", 4325 buf->tcp_ipv6_chksum); 4326 netdev_dbg(adapter->netdev, "udp_ipv4_chksum = %d\n", 4327 buf->udp_ipv4_chksum); 4328 netdev_dbg(adapter->netdev, "udp_ipv6_chksum = %d\n", 4329 buf->udp_ipv6_chksum); 4330 netdev_dbg(adapter->netdev, "large_tx_ipv4 = %d\n", 4331 buf->large_tx_ipv4); 4332 netdev_dbg(adapter->netdev, "large_tx_ipv6 = %d\n", 4333 buf->large_tx_ipv6); 4334 netdev_dbg(adapter->netdev, "large_rx_ipv4 = %d\n", 4335 buf->large_rx_ipv4); 4336 netdev_dbg(adapter->netdev, "large_rx_ipv6 = %d\n", 4337 buf->large_rx_ipv6); 4338 netdev_dbg(adapter->netdev, "max_ipv4_hdr_sz = %d\n", 4339 buf->max_ipv4_header_size); 4340 netdev_dbg(adapter->netdev, "max_ipv6_hdr_sz = %d\n", 4341 buf->max_ipv6_header_size); 4342 netdev_dbg(adapter->netdev, "max_tcp_hdr_size = %d\n", 4343 buf->max_tcp_header_size); 4344 netdev_dbg(adapter->netdev, "max_udp_hdr_size = %d\n", 4345 buf->max_udp_header_size); 4346 netdev_dbg(adapter->netdev, "max_large_tx_size = %d\n", 4347 buf->max_large_tx_size); 4348 netdev_dbg(adapter->netdev, "max_large_rx_size = %d\n", 4349 buf->max_large_rx_size); 4350 netdev_dbg(adapter->netdev, "ipv6_ext_hdr = %d\n", 4351 buf->ipv6_extension_header); 4352 netdev_dbg(adapter->netdev, "tcp_pseudosum_req = %d\n", 4353 buf->tcp_pseudosum_req); 4354 netdev_dbg(adapter->netdev, "num_ipv6_ext_hd = %d\n", 4355 buf->num_ipv6_ext_headers); 4356 netdev_dbg(adapter->netdev, "off_ipv6_ext_hd = %d\n", 4357 buf->off_ipv6_ext_headers); 4358 4359 send_control_ip_offload(adapter); 4360 } 4361 4362 static const char *ibmvnic_fw_err_cause(u16 cause) 4363 { 4364 switch (cause) { 4365 case ADAPTER_PROBLEM: 4366 return "adapter problem"; 4367 case BUS_PROBLEM: 4368 return "bus problem"; 4369 case FW_PROBLEM: 4370 return "firmware problem"; 4371 case DD_PROBLEM: 4372 return "device driver problem"; 4373 case EEH_RECOVERY: 4374 return "EEH recovery"; 4375 case FW_UPDATED: 4376 return "firmware updated"; 4377 case LOW_MEMORY: 4378 return "low Memory"; 4379 default: 4380 return "unknown"; 4381 } 4382 } 4383 4384 static void handle_error_indication(union ibmvnic_crq *crq, 4385 struct ibmvnic_adapter *adapter) 4386 { 4387 struct device *dev = &adapter->vdev->dev; 4388 u16 cause; 4389 4390 cause = be16_to_cpu(crq->error_indication.error_cause); 4391 4392 dev_warn_ratelimited(dev, 4393 "Firmware reports %serror, cause: %s. Starting recovery...\n", 4394 crq->error_indication.flags 4395 & IBMVNIC_FATAL_ERROR ? "FATAL " : "", 4396 ibmvnic_fw_err_cause(cause)); 4397 4398 if (crq->error_indication.flags & IBMVNIC_FATAL_ERROR) 4399 ibmvnic_reset(adapter, VNIC_RESET_FATAL); 4400 else 4401 ibmvnic_reset(adapter, VNIC_RESET_NON_FATAL); 4402 } 4403 4404 static int handle_change_mac_rsp(union ibmvnic_crq *crq, 4405 struct ibmvnic_adapter *adapter) 4406 { 4407 struct net_device *netdev = adapter->netdev; 4408 struct device *dev = &adapter->vdev->dev; 4409 long rc; 4410 4411 rc = crq->change_mac_addr_rsp.rc.code; 4412 if (rc) { 4413 dev_err(dev, "Error %ld in CHANGE_MAC_ADDR_RSP\n", rc); 4414 goto out; 4415 } 4416 /* crq->change_mac_addr.mac_addr is the requested one 4417 * crq->change_mac_addr_rsp.mac_addr is the returned valid one. 4418 */ 4419 ether_addr_copy(netdev->dev_addr, 4420 &crq->change_mac_addr_rsp.mac_addr[0]); 4421 ether_addr_copy(adapter->mac_addr, 4422 &crq->change_mac_addr_rsp.mac_addr[0]); 4423 out: 4424 complete(&adapter->fw_done); 4425 return rc; 4426 } 4427 4428 static void handle_request_cap_rsp(union ibmvnic_crq *crq, 4429 struct ibmvnic_adapter *adapter) 4430 { 4431 struct device *dev = &adapter->vdev->dev; 4432 u64 *req_value; 4433 char *name; 4434 4435 atomic_dec(&adapter->running_cap_crqs); 4436 switch (be16_to_cpu(crq->request_capability_rsp.capability)) { 4437 case REQ_TX_QUEUES: 4438 req_value = &adapter->req_tx_queues; 4439 name = "tx"; 4440 break; 4441 case REQ_RX_QUEUES: 4442 req_value = &adapter->req_rx_queues; 4443 name = "rx"; 4444 break; 4445 case REQ_RX_ADD_QUEUES: 4446 req_value = &adapter->req_rx_add_queues; 4447 name = "rx_add"; 4448 break; 4449 case REQ_TX_ENTRIES_PER_SUBCRQ: 4450 req_value = &adapter->req_tx_entries_per_subcrq; 4451 name = "tx_entries_per_subcrq"; 4452 break; 4453 case REQ_RX_ADD_ENTRIES_PER_SUBCRQ: 4454 req_value = &adapter->req_rx_add_entries_per_subcrq; 4455 name = "rx_add_entries_per_subcrq"; 4456 break; 4457 case REQ_MTU: 4458 req_value = &adapter->req_mtu; 4459 name = "mtu"; 4460 break; 4461 case PROMISC_REQUESTED: 4462 req_value = &adapter->promisc; 4463 name = "promisc"; 4464 break; 4465 default: 4466 dev_err(dev, "Got invalid cap request rsp %d\n", 4467 crq->request_capability.capability); 4468 return; 4469 } 4470 4471 switch (crq->request_capability_rsp.rc.code) { 4472 case SUCCESS: 4473 break; 4474 case PARTIALSUCCESS: 4475 dev_info(dev, "req=%lld, rsp=%ld in %s queue, retrying.\n", 4476 *req_value, 4477 (long)be64_to_cpu(crq->request_capability_rsp.number), 4478 name); 4479 4480 if (be16_to_cpu(crq->request_capability_rsp.capability) == 4481 REQ_MTU) { 4482 pr_err("mtu of %llu is not supported. Reverting.\n", 4483 *req_value); 4484 *req_value = adapter->fallback.mtu; 4485 } else { 4486 *req_value = 4487 be64_to_cpu(crq->request_capability_rsp.number); 4488 } 4489 4490 send_request_cap(adapter, 1); 4491 return; 4492 default: 4493 dev_err(dev, "Error %d in request cap rsp\n", 4494 crq->request_capability_rsp.rc.code); 4495 return; 4496 } 4497 4498 /* Done receiving requested capabilities, query IP offload support */ 4499 if (atomic_read(&adapter->running_cap_crqs) == 0) { 4500 adapter->wait_capability = false; 4501 send_query_ip_offload(adapter); 4502 } 4503 } 4504 4505 static int handle_login_rsp(union ibmvnic_crq *login_rsp_crq, 4506 struct ibmvnic_adapter *adapter) 4507 { 4508 struct device *dev = &adapter->vdev->dev; 4509 struct net_device *netdev = adapter->netdev; 4510 struct ibmvnic_login_rsp_buffer *login_rsp = adapter->login_rsp_buf; 4511 struct ibmvnic_login_buffer *login = adapter->login_buf; 4512 u64 *tx_handle_array; 4513 u64 *rx_handle_array; 4514 int num_tx_pools; 4515 int num_rx_pools; 4516 u64 *size_array; 4517 int i; 4518 4519 /* CHECK: Test/set of login_pending does not need to be atomic 4520 * because only ibmvnic_tasklet tests/clears this. 4521 */ 4522 if (!adapter->login_pending) { 4523 netdev_warn(netdev, "Ignoring unexpected login response\n"); 4524 return 0; 4525 } 4526 adapter->login_pending = false; 4527 4528 dma_unmap_single(dev, adapter->login_buf_token, adapter->login_buf_sz, 4529 DMA_TO_DEVICE); 4530 dma_unmap_single(dev, adapter->login_rsp_buf_token, 4531 adapter->login_rsp_buf_sz, DMA_FROM_DEVICE); 4532 4533 /* If the number of queues requested can't be allocated by the 4534 * server, the login response will return with code 1. We will need 4535 * to resend the login buffer with fewer queues requested. 4536 */ 4537 if (login_rsp_crq->generic.rc.code) { 4538 adapter->init_done_rc = login_rsp_crq->generic.rc.code; 4539 complete(&adapter->init_done); 4540 return 0; 4541 } 4542 4543 netdev->mtu = adapter->req_mtu - ETH_HLEN; 4544 4545 netdev_dbg(adapter->netdev, "Login Response Buffer:\n"); 4546 for (i = 0; i < (adapter->login_rsp_buf_sz - 1) / 8 + 1; i++) { 4547 netdev_dbg(adapter->netdev, "%016lx\n", 4548 ((unsigned long *)(adapter->login_rsp_buf))[i]); 4549 } 4550 4551 /* Sanity checks */ 4552 if (login->num_txcomp_subcrqs != login_rsp->num_txsubm_subcrqs || 4553 (be32_to_cpu(login->num_rxcomp_subcrqs) * 4554 adapter->req_rx_add_queues != 4555 be32_to_cpu(login_rsp->num_rxadd_subcrqs))) { 4556 dev_err(dev, "FATAL: Inconsistent login and login rsp\n"); 4557 ibmvnic_reset(adapter, VNIC_RESET_FATAL); 4558 return -EIO; 4559 } 4560 size_array = (u64 *)((u8 *)(adapter->login_rsp_buf) + 4561 be32_to_cpu(adapter->login_rsp_buf->off_rxadd_buff_size)); 4562 /* variable buffer sizes are not supported, so just read the 4563 * first entry. 4564 */ 4565 adapter->cur_rx_buf_sz = be64_to_cpu(size_array[0]); 4566 4567 num_tx_pools = be32_to_cpu(adapter->login_rsp_buf->num_txsubm_subcrqs); 4568 num_rx_pools = be32_to_cpu(adapter->login_rsp_buf->num_rxadd_subcrqs); 4569 4570 tx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) + 4571 be32_to_cpu(adapter->login_rsp_buf->off_txsubm_subcrqs)); 4572 rx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) + 4573 be32_to_cpu(adapter->login_rsp_buf->off_rxadd_subcrqs)); 4574 4575 for (i = 0; i < num_tx_pools; i++) 4576 adapter->tx_scrq[i]->handle = tx_handle_array[i]; 4577 4578 for (i = 0; i < num_rx_pools; i++) 4579 adapter->rx_scrq[i]->handle = rx_handle_array[i]; 4580 4581 adapter->num_active_tx_scrqs = num_tx_pools; 4582 adapter->num_active_rx_scrqs = num_rx_pools; 4583 release_login_rsp_buffer(adapter); 4584 release_login_buffer(adapter); 4585 complete(&adapter->init_done); 4586 4587 return 0; 4588 } 4589 4590 static void handle_request_unmap_rsp(union ibmvnic_crq *crq, 4591 struct ibmvnic_adapter *adapter) 4592 { 4593 struct device *dev = &adapter->vdev->dev; 4594 long rc; 4595 4596 rc = crq->request_unmap_rsp.rc.code; 4597 if (rc) 4598 dev_err(dev, "Error %ld in REQUEST_UNMAP_RSP\n", rc); 4599 } 4600 4601 static void handle_query_map_rsp(union ibmvnic_crq *crq, 4602 struct ibmvnic_adapter *adapter) 4603 { 4604 struct net_device *netdev = adapter->netdev; 4605 struct device *dev = &adapter->vdev->dev; 4606 long rc; 4607 4608 rc = crq->query_map_rsp.rc.code; 4609 if (rc) { 4610 dev_err(dev, "Error %ld in QUERY_MAP_RSP\n", rc); 4611 return; 4612 } 4613 netdev_dbg(netdev, "page_size = %d\ntot_pages = %d\nfree_pages = %d\n", 4614 crq->query_map_rsp.page_size, crq->query_map_rsp.tot_pages, 4615 crq->query_map_rsp.free_pages); 4616 } 4617 4618 static void handle_query_cap_rsp(union ibmvnic_crq *crq, 4619 struct ibmvnic_adapter *adapter) 4620 { 4621 struct net_device *netdev = adapter->netdev; 4622 struct device *dev = &adapter->vdev->dev; 4623 long rc; 4624 4625 atomic_dec(&adapter->running_cap_crqs); 4626 netdev_dbg(netdev, "Outstanding queries: %d\n", 4627 atomic_read(&adapter->running_cap_crqs)); 4628 rc = crq->query_capability.rc.code; 4629 if (rc) { 4630 dev_err(dev, "Error %ld in QUERY_CAP_RSP\n", rc); 4631 goto out; 4632 } 4633 4634 switch (be16_to_cpu(crq->query_capability.capability)) { 4635 case MIN_TX_QUEUES: 4636 adapter->min_tx_queues = 4637 be64_to_cpu(crq->query_capability.number); 4638 netdev_dbg(netdev, "min_tx_queues = %lld\n", 4639 adapter->min_tx_queues); 4640 break; 4641 case MIN_RX_QUEUES: 4642 adapter->min_rx_queues = 4643 be64_to_cpu(crq->query_capability.number); 4644 netdev_dbg(netdev, "min_rx_queues = %lld\n", 4645 adapter->min_rx_queues); 4646 break; 4647 case MIN_RX_ADD_QUEUES: 4648 adapter->min_rx_add_queues = 4649 be64_to_cpu(crq->query_capability.number); 4650 netdev_dbg(netdev, "min_rx_add_queues = %lld\n", 4651 adapter->min_rx_add_queues); 4652 break; 4653 case MAX_TX_QUEUES: 4654 adapter->max_tx_queues = 4655 be64_to_cpu(crq->query_capability.number); 4656 netdev_dbg(netdev, "max_tx_queues = %lld\n", 4657 adapter->max_tx_queues); 4658 break; 4659 case MAX_RX_QUEUES: 4660 adapter->max_rx_queues = 4661 be64_to_cpu(crq->query_capability.number); 4662 netdev_dbg(netdev, "max_rx_queues = %lld\n", 4663 adapter->max_rx_queues); 4664 break; 4665 case MAX_RX_ADD_QUEUES: 4666 adapter->max_rx_add_queues = 4667 be64_to_cpu(crq->query_capability.number); 4668 netdev_dbg(netdev, "max_rx_add_queues = %lld\n", 4669 adapter->max_rx_add_queues); 4670 break; 4671 case MIN_TX_ENTRIES_PER_SUBCRQ: 4672 adapter->min_tx_entries_per_subcrq = 4673 be64_to_cpu(crq->query_capability.number); 4674 netdev_dbg(netdev, "min_tx_entries_per_subcrq = %lld\n", 4675 adapter->min_tx_entries_per_subcrq); 4676 break; 4677 case MIN_RX_ADD_ENTRIES_PER_SUBCRQ: 4678 adapter->min_rx_add_entries_per_subcrq = 4679 be64_to_cpu(crq->query_capability.number); 4680 netdev_dbg(netdev, "min_rx_add_entrs_per_subcrq = %lld\n", 4681 adapter->min_rx_add_entries_per_subcrq); 4682 break; 4683 case MAX_TX_ENTRIES_PER_SUBCRQ: 4684 adapter->max_tx_entries_per_subcrq = 4685 be64_to_cpu(crq->query_capability.number); 4686 netdev_dbg(netdev, "max_tx_entries_per_subcrq = %lld\n", 4687 adapter->max_tx_entries_per_subcrq); 4688 break; 4689 case MAX_RX_ADD_ENTRIES_PER_SUBCRQ: 4690 adapter->max_rx_add_entries_per_subcrq = 4691 be64_to_cpu(crq->query_capability.number); 4692 netdev_dbg(netdev, "max_rx_add_entrs_per_subcrq = %lld\n", 4693 adapter->max_rx_add_entries_per_subcrq); 4694 break; 4695 case TCP_IP_OFFLOAD: 4696 adapter->tcp_ip_offload = 4697 be64_to_cpu(crq->query_capability.number); 4698 netdev_dbg(netdev, "tcp_ip_offload = %lld\n", 4699 adapter->tcp_ip_offload); 4700 break; 4701 case PROMISC_SUPPORTED: 4702 adapter->promisc_supported = 4703 be64_to_cpu(crq->query_capability.number); 4704 netdev_dbg(netdev, "promisc_supported = %lld\n", 4705 adapter->promisc_supported); 4706 break; 4707 case MIN_MTU: 4708 adapter->min_mtu = be64_to_cpu(crq->query_capability.number); 4709 netdev->min_mtu = adapter->min_mtu - ETH_HLEN; 4710 netdev_dbg(netdev, "min_mtu = %lld\n", adapter->min_mtu); 4711 break; 4712 case MAX_MTU: 4713 adapter->max_mtu = be64_to_cpu(crq->query_capability.number); 4714 netdev->max_mtu = adapter->max_mtu - ETH_HLEN; 4715 netdev_dbg(netdev, "max_mtu = %lld\n", adapter->max_mtu); 4716 break; 4717 case MAX_MULTICAST_FILTERS: 4718 adapter->max_multicast_filters = 4719 be64_to_cpu(crq->query_capability.number); 4720 netdev_dbg(netdev, "max_multicast_filters = %lld\n", 4721 adapter->max_multicast_filters); 4722 break; 4723 case VLAN_HEADER_INSERTION: 4724 adapter->vlan_header_insertion = 4725 be64_to_cpu(crq->query_capability.number); 4726 if (adapter->vlan_header_insertion) 4727 netdev->features |= NETIF_F_HW_VLAN_STAG_TX; 4728 netdev_dbg(netdev, "vlan_header_insertion = %lld\n", 4729 adapter->vlan_header_insertion); 4730 break; 4731 case RX_VLAN_HEADER_INSERTION: 4732 adapter->rx_vlan_header_insertion = 4733 be64_to_cpu(crq->query_capability.number); 4734 netdev_dbg(netdev, "rx_vlan_header_insertion = %lld\n", 4735 adapter->rx_vlan_header_insertion); 4736 break; 4737 case MAX_TX_SG_ENTRIES: 4738 adapter->max_tx_sg_entries = 4739 be64_to_cpu(crq->query_capability.number); 4740 netdev_dbg(netdev, "max_tx_sg_entries = %lld\n", 4741 adapter->max_tx_sg_entries); 4742 break; 4743 case RX_SG_SUPPORTED: 4744 adapter->rx_sg_supported = 4745 be64_to_cpu(crq->query_capability.number); 4746 netdev_dbg(netdev, "rx_sg_supported = %lld\n", 4747 adapter->rx_sg_supported); 4748 break; 4749 case OPT_TX_COMP_SUB_QUEUES: 4750 adapter->opt_tx_comp_sub_queues = 4751 be64_to_cpu(crq->query_capability.number); 4752 netdev_dbg(netdev, "opt_tx_comp_sub_queues = %lld\n", 4753 adapter->opt_tx_comp_sub_queues); 4754 break; 4755 case OPT_RX_COMP_QUEUES: 4756 adapter->opt_rx_comp_queues = 4757 be64_to_cpu(crq->query_capability.number); 4758 netdev_dbg(netdev, "opt_rx_comp_queues = %lld\n", 4759 adapter->opt_rx_comp_queues); 4760 break; 4761 case OPT_RX_BUFADD_Q_PER_RX_COMP_Q: 4762 adapter->opt_rx_bufadd_q_per_rx_comp_q = 4763 be64_to_cpu(crq->query_capability.number); 4764 netdev_dbg(netdev, "opt_rx_bufadd_q_per_rx_comp_q = %lld\n", 4765 adapter->opt_rx_bufadd_q_per_rx_comp_q); 4766 break; 4767 case OPT_TX_ENTRIES_PER_SUBCRQ: 4768 adapter->opt_tx_entries_per_subcrq = 4769 be64_to_cpu(crq->query_capability.number); 4770 netdev_dbg(netdev, "opt_tx_entries_per_subcrq = %lld\n", 4771 adapter->opt_tx_entries_per_subcrq); 4772 break; 4773 case OPT_RXBA_ENTRIES_PER_SUBCRQ: 4774 adapter->opt_rxba_entries_per_subcrq = 4775 be64_to_cpu(crq->query_capability.number); 4776 netdev_dbg(netdev, "opt_rxba_entries_per_subcrq = %lld\n", 4777 adapter->opt_rxba_entries_per_subcrq); 4778 break; 4779 case TX_RX_DESC_REQ: 4780 adapter->tx_rx_desc_req = crq->query_capability.number; 4781 netdev_dbg(netdev, "tx_rx_desc_req = %llx\n", 4782 adapter->tx_rx_desc_req); 4783 break; 4784 4785 default: 4786 netdev_err(netdev, "Got invalid cap rsp %d\n", 4787 crq->query_capability.capability); 4788 } 4789 4790 out: 4791 if (atomic_read(&adapter->running_cap_crqs) == 0) { 4792 adapter->wait_capability = false; 4793 send_request_cap(adapter, 0); 4794 } 4795 } 4796 4797 static int send_query_phys_parms(struct ibmvnic_adapter *adapter) 4798 { 4799 union ibmvnic_crq crq; 4800 int rc; 4801 4802 memset(&crq, 0, sizeof(crq)); 4803 crq.query_phys_parms.first = IBMVNIC_CRQ_CMD; 4804 crq.query_phys_parms.cmd = QUERY_PHYS_PARMS; 4805 4806 mutex_lock(&adapter->fw_lock); 4807 adapter->fw_done_rc = 0; 4808 reinit_completion(&adapter->fw_done); 4809 4810 rc = ibmvnic_send_crq(adapter, &crq); 4811 if (rc) { 4812 mutex_unlock(&adapter->fw_lock); 4813 return rc; 4814 } 4815 4816 rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000); 4817 if (rc) { 4818 mutex_unlock(&adapter->fw_lock); 4819 return rc; 4820 } 4821 4822 mutex_unlock(&adapter->fw_lock); 4823 return adapter->fw_done_rc ? -EIO : 0; 4824 } 4825 4826 static int handle_query_phys_parms_rsp(union ibmvnic_crq *crq, 4827 struct ibmvnic_adapter *adapter) 4828 { 4829 struct net_device *netdev = adapter->netdev; 4830 int rc; 4831 __be32 rspeed = cpu_to_be32(crq->query_phys_parms_rsp.speed); 4832 4833 rc = crq->query_phys_parms_rsp.rc.code; 4834 if (rc) { 4835 netdev_err(netdev, "Error %d in QUERY_PHYS_PARMS\n", rc); 4836 return rc; 4837 } 4838 switch (rspeed) { 4839 case IBMVNIC_10MBPS: 4840 adapter->speed = SPEED_10; 4841 break; 4842 case IBMVNIC_100MBPS: 4843 adapter->speed = SPEED_100; 4844 break; 4845 case IBMVNIC_1GBPS: 4846 adapter->speed = SPEED_1000; 4847 break; 4848 case IBMVNIC_10GBPS: 4849 adapter->speed = SPEED_10000; 4850 break; 4851 case IBMVNIC_25GBPS: 4852 adapter->speed = SPEED_25000; 4853 break; 4854 case IBMVNIC_40GBPS: 4855 adapter->speed = SPEED_40000; 4856 break; 4857 case IBMVNIC_50GBPS: 4858 adapter->speed = SPEED_50000; 4859 break; 4860 case IBMVNIC_100GBPS: 4861 adapter->speed = SPEED_100000; 4862 break; 4863 case IBMVNIC_200GBPS: 4864 adapter->speed = SPEED_200000; 4865 break; 4866 default: 4867 if (netif_carrier_ok(netdev)) 4868 netdev_warn(netdev, "Unknown speed 0x%08x\n", rspeed); 4869 adapter->speed = SPEED_UNKNOWN; 4870 } 4871 if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_FULL_DUPLEX) 4872 adapter->duplex = DUPLEX_FULL; 4873 else if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_HALF_DUPLEX) 4874 adapter->duplex = DUPLEX_HALF; 4875 else 4876 adapter->duplex = DUPLEX_UNKNOWN; 4877 4878 return rc; 4879 } 4880 4881 static void ibmvnic_handle_crq(union ibmvnic_crq *crq, 4882 struct ibmvnic_adapter *adapter) 4883 { 4884 struct ibmvnic_generic_crq *gen_crq = &crq->generic; 4885 struct net_device *netdev = adapter->netdev; 4886 struct device *dev = &adapter->vdev->dev; 4887 u64 *u64_crq = (u64 *)crq; 4888 long rc; 4889 4890 netdev_dbg(netdev, "Handling CRQ: %016lx %016lx\n", 4891 (unsigned long)cpu_to_be64(u64_crq[0]), 4892 (unsigned long)cpu_to_be64(u64_crq[1])); 4893 switch (gen_crq->first) { 4894 case IBMVNIC_CRQ_INIT_RSP: 4895 switch (gen_crq->cmd) { 4896 case IBMVNIC_CRQ_INIT: 4897 dev_info(dev, "Partner initialized\n"); 4898 adapter->from_passive_init = true; 4899 /* Discard any stale login responses from prev reset. 4900 * CHECK: should we clear even on INIT_COMPLETE? 4901 */ 4902 adapter->login_pending = false; 4903 4904 if (!completion_done(&adapter->init_done)) { 4905 complete(&adapter->init_done); 4906 adapter->init_done_rc = -EIO; 4907 } 4908 rc = ibmvnic_reset(adapter, VNIC_RESET_FAILOVER); 4909 if (rc && rc != -EBUSY) { 4910 /* We were unable to schedule the failover 4911 * reset either because the adapter was still 4912 * probing (eg: during kexec) or we could not 4913 * allocate memory. Clear the failover_pending 4914 * flag since no one else will. We ignore 4915 * EBUSY because it means either FAILOVER reset 4916 * is already scheduled or the adapter is 4917 * being removed. 4918 */ 4919 netdev_err(netdev, 4920 "Error %ld scheduling failover reset\n", 4921 rc); 4922 adapter->failover_pending = false; 4923 } 4924 break; 4925 case IBMVNIC_CRQ_INIT_COMPLETE: 4926 dev_info(dev, "Partner initialization complete\n"); 4927 adapter->crq.active = true; 4928 send_version_xchg(adapter); 4929 break; 4930 default: 4931 dev_err(dev, "Unknown crq cmd: %d\n", gen_crq->cmd); 4932 } 4933 return; 4934 case IBMVNIC_CRQ_XPORT_EVENT: 4935 netif_carrier_off(netdev); 4936 adapter->crq.active = false; 4937 /* terminate any thread waiting for a response 4938 * from the device 4939 */ 4940 if (!completion_done(&adapter->fw_done)) { 4941 adapter->fw_done_rc = -EIO; 4942 complete(&adapter->fw_done); 4943 } 4944 if (!completion_done(&adapter->stats_done)) 4945 complete(&adapter->stats_done); 4946 if (test_bit(0, &adapter->resetting)) 4947 adapter->force_reset_recovery = true; 4948 if (gen_crq->cmd == IBMVNIC_PARTITION_MIGRATED) { 4949 dev_info(dev, "Migrated, re-enabling adapter\n"); 4950 ibmvnic_reset(adapter, VNIC_RESET_MOBILITY); 4951 } else if (gen_crq->cmd == IBMVNIC_DEVICE_FAILOVER) { 4952 dev_info(dev, "Backing device failover detected\n"); 4953 adapter->failover_pending = true; 4954 } else { 4955 /* The adapter lost the connection */ 4956 dev_err(dev, "Virtual Adapter failed (rc=%d)\n", 4957 gen_crq->cmd); 4958 ibmvnic_reset(adapter, VNIC_RESET_FATAL); 4959 } 4960 return; 4961 case IBMVNIC_CRQ_CMD_RSP: 4962 break; 4963 default: 4964 dev_err(dev, "Got an invalid msg type 0x%02x\n", 4965 gen_crq->first); 4966 return; 4967 } 4968 4969 switch (gen_crq->cmd) { 4970 case VERSION_EXCHANGE_RSP: 4971 rc = crq->version_exchange_rsp.rc.code; 4972 if (rc) { 4973 dev_err(dev, "Error %ld in VERSION_EXCHG_RSP\n", rc); 4974 break; 4975 } 4976 ibmvnic_version = 4977 be16_to_cpu(crq->version_exchange_rsp.version); 4978 dev_info(dev, "Partner protocol version is %d\n", 4979 ibmvnic_version); 4980 send_query_cap(adapter); 4981 break; 4982 case QUERY_CAPABILITY_RSP: 4983 handle_query_cap_rsp(crq, adapter); 4984 break; 4985 case QUERY_MAP_RSP: 4986 handle_query_map_rsp(crq, adapter); 4987 break; 4988 case REQUEST_MAP_RSP: 4989 adapter->fw_done_rc = crq->request_map_rsp.rc.code; 4990 complete(&adapter->fw_done); 4991 break; 4992 case REQUEST_UNMAP_RSP: 4993 handle_request_unmap_rsp(crq, adapter); 4994 break; 4995 case REQUEST_CAPABILITY_RSP: 4996 handle_request_cap_rsp(crq, adapter); 4997 break; 4998 case LOGIN_RSP: 4999 netdev_dbg(netdev, "Got Login Response\n"); 5000 handle_login_rsp(crq, adapter); 5001 break; 5002 case LOGICAL_LINK_STATE_RSP: 5003 netdev_dbg(netdev, 5004 "Got Logical Link State Response, state: %d rc: %d\n", 5005 crq->logical_link_state_rsp.link_state, 5006 crq->logical_link_state_rsp.rc.code); 5007 adapter->logical_link_state = 5008 crq->logical_link_state_rsp.link_state; 5009 adapter->init_done_rc = crq->logical_link_state_rsp.rc.code; 5010 complete(&adapter->init_done); 5011 break; 5012 case LINK_STATE_INDICATION: 5013 netdev_dbg(netdev, "Got Logical Link State Indication\n"); 5014 adapter->phys_link_state = 5015 crq->link_state_indication.phys_link_state; 5016 adapter->logical_link_state = 5017 crq->link_state_indication.logical_link_state; 5018 if (adapter->phys_link_state && adapter->logical_link_state) 5019 netif_carrier_on(netdev); 5020 else 5021 netif_carrier_off(netdev); 5022 break; 5023 case CHANGE_MAC_ADDR_RSP: 5024 netdev_dbg(netdev, "Got MAC address change Response\n"); 5025 adapter->fw_done_rc = handle_change_mac_rsp(crq, adapter); 5026 break; 5027 case ERROR_INDICATION: 5028 netdev_dbg(netdev, "Got Error Indication\n"); 5029 handle_error_indication(crq, adapter); 5030 break; 5031 case REQUEST_STATISTICS_RSP: 5032 netdev_dbg(netdev, "Got Statistics Response\n"); 5033 complete(&adapter->stats_done); 5034 break; 5035 case QUERY_IP_OFFLOAD_RSP: 5036 netdev_dbg(netdev, "Got Query IP offload Response\n"); 5037 handle_query_ip_offload_rsp(adapter); 5038 break; 5039 case MULTICAST_CTRL_RSP: 5040 netdev_dbg(netdev, "Got multicast control Response\n"); 5041 break; 5042 case CONTROL_IP_OFFLOAD_RSP: 5043 netdev_dbg(netdev, "Got Control IP offload Response\n"); 5044 dma_unmap_single(dev, adapter->ip_offload_ctrl_tok, 5045 sizeof(adapter->ip_offload_ctrl), 5046 DMA_TO_DEVICE); 5047 complete(&adapter->init_done); 5048 break; 5049 case COLLECT_FW_TRACE_RSP: 5050 netdev_dbg(netdev, "Got Collect firmware trace Response\n"); 5051 complete(&adapter->fw_done); 5052 break; 5053 case GET_VPD_SIZE_RSP: 5054 handle_vpd_size_rsp(crq, adapter); 5055 break; 5056 case GET_VPD_RSP: 5057 handle_vpd_rsp(crq, adapter); 5058 break; 5059 case QUERY_PHYS_PARMS_RSP: 5060 adapter->fw_done_rc = handle_query_phys_parms_rsp(crq, adapter); 5061 complete(&adapter->fw_done); 5062 break; 5063 default: 5064 netdev_err(netdev, "Got an invalid cmd type 0x%02x\n", 5065 gen_crq->cmd); 5066 } 5067 } 5068 5069 static irqreturn_t ibmvnic_interrupt(int irq, void *instance) 5070 { 5071 struct ibmvnic_adapter *adapter = instance; 5072 5073 tasklet_schedule(&adapter->tasklet); 5074 return IRQ_HANDLED; 5075 } 5076 5077 static void ibmvnic_tasklet(struct tasklet_struct *t) 5078 { 5079 struct ibmvnic_adapter *adapter = from_tasklet(adapter, t, tasklet); 5080 struct ibmvnic_crq_queue *queue = &adapter->crq; 5081 union ibmvnic_crq *crq; 5082 unsigned long flags; 5083 bool done = false; 5084 5085 spin_lock_irqsave(&queue->lock, flags); 5086 while (!done) { 5087 /* Pull all the valid messages off the CRQ */ 5088 while ((crq = ibmvnic_next_crq(adapter)) != NULL) { 5089 /* This barrier makes sure ibmvnic_next_crq()'s 5090 * crq->generic.first & IBMVNIC_CRQ_CMD_RSP is loaded 5091 * before ibmvnic_handle_crq()'s 5092 * switch(gen_crq->first) and switch(gen_crq->cmd). 5093 */ 5094 dma_rmb(); 5095 ibmvnic_handle_crq(crq, adapter); 5096 crq->generic.first = 0; 5097 } 5098 5099 /* remain in tasklet until all 5100 * capabilities responses are received 5101 */ 5102 if (!adapter->wait_capability) 5103 done = true; 5104 } 5105 /* if capabilities CRQ's were sent in this tasklet, the following 5106 * tasklet must wait until all responses are received 5107 */ 5108 if (atomic_read(&adapter->running_cap_crqs) != 0) 5109 adapter->wait_capability = true; 5110 spin_unlock_irqrestore(&queue->lock, flags); 5111 } 5112 5113 static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *adapter) 5114 { 5115 struct vio_dev *vdev = adapter->vdev; 5116 int rc; 5117 5118 do { 5119 rc = plpar_hcall_norets(H_ENABLE_CRQ, vdev->unit_address); 5120 } while (rc == H_IN_PROGRESS || rc == H_BUSY || H_IS_LONG_BUSY(rc)); 5121 5122 if (rc) 5123 dev_err(&vdev->dev, "Error enabling adapter (rc=%d)\n", rc); 5124 5125 return rc; 5126 } 5127 5128 static int ibmvnic_reset_crq(struct ibmvnic_adapter *adapter) 5129 { 5130 struct ibmvnic_crq_queue *crq = &adapter->crq; 5131 struct device *dev = &adapter->vdev->dev; 5132 struct vio_dev *vdev = adapter->vdev; 5133 int rc; 5134 5135 /* Close the CRQ */ 5136 do { 5137 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address); 5138 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc)); 5139 5140 /* Clean out the queue */ 5141 if (!crq->msgs) 5142 return -EINVAL; 5143 5144 memset(crq->msgs, 0, PAGE_SIZE); 5145 crq->cur = 0; 5146 crq->active = false; 5147 5148 /* And re-open it again */ 5149 rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address, 5150 crq->msg_token, PAGE_SIZE); 5151 5152 if (rc == H_CLOSED) 5153 /* Adapter is good, but other end is not ready */ 5154 dev_warn(dev, "Partner adapter not ready\n"); 5155 else if (rc != 0) 5156 dev_warn(dev, "Couldn't register crq (rc=%d)\n", rc); 5157 5158 return rc; 5159 } 5160 5161 static void release_crq_queue(struct ibmvnic_adapter *adapter) 5162 { 5163 struct ibmvnic_crq_queue *crq = &adapter->crq; 5164 struct vio_dev *vdev = adapter->vdev; 5165 long rc; 5166 5167 if (!crq->msgs) 5168 return; 5169 5170 netdev_dbg(adapter->netdev, "Releasing CRQ\n"); 5171 free_irq(vdev->irq, adapter); 5172 tasklet_kill(&adapter->tasklet); 5173 do { 5174 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address); 5175 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc)); 5176 5177 dma_unmap_single(&vdev->dev, crq->msg_token, PAGE_SIZE, 5178 DMA_BIDIRECTIONAL); 5179 free_page((unsigned long)crq->msgs); 5180 crq->msgs = NULL; 5181 crq->active = false; 5182 } 5183 5184 static int init_crq_queue(struct ibmvnic_adapter *adapter) 5185 { 5186 struct ibmvnic_crq_queue *crq = &adapter->crq; 5187 struct device *dev = &adapter->vdev->dev; 5188 struct vio_dev *vdev = adapter->vdev; 5189 int rc, retrc = -ENOMEM; 5190 5191 if (crq->msgs) 5192 return 0; 5193 5194 crq->msgs = (union ibmvnic_crq *)get_zeroed_page(GFP_KERNEL); 5195 /* Should we allocate more than one page? */ 5196 5197 if (!crq->msgs) 5198 return -ENOMEM; 5199 5200 crq->size = PAGE_SIZE / sizeof(*crq->msgs); 5201 crq->msg_token = dma_map_single(dev, crq->msgs, PAGE_SIZE, 5202 DMA_BIDIRECTIONAL); 5203 if (dma_mapping_error(dev, crq->msg_token)) 5204 goto map_failed; 5205 5206 rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address, 5207 crq->msg_token, PAGE_SIZE); 5208 5209 if (rc == H_RESOURCE) 5210 /* maybe kexecing and resource is busy. try a reset */ 5211 rc = ibmvnic_reset_crq(adapter); 5212 retrc = rc; 5213 5214 if (rc == H_CLOSED) { 5215 dev_warn(dev, "Partner adapter not ready\n"); 5216 } else if (rc) { 5217 dev_warn(dev, "Error %d opening adapter\n", rc); 5218 goto reg_crq_failed; 5219 } 5220 5221 retrc = 0; 5222 5223 tasklet_setup(&adapter->tasklet, (void *)ibmvnic_tasklet); 5224 5225 netdev_dbg(adapter->netdev, "registering irq 0x%x\n", vdev->irq); 5226 snprintf(crq->name, sizeof(crq->name), "ibmvnic-%x", 5227 adapter->vdev->unit_address); 5228 rc = request_irq(vdev->irq, ibmvnic_interrupt, 0, crq->name, adapter); 5229 if (rc) { 5230 dev_err(dev, "Couldn't register irq 0x%x. rc=%d\n", 5231 vdev->irq, rc); 5232 goto req_irq_failed; 5233 } 5234 5235 rc = vio_enable_interrupts(vdev); 5236 if (rc) { 5237 dev_err(dev, "Error %d enabling interrupts\n", rc); 5238 goto req_irq_failed; 5239 } 5240 5241 crq->cur = 0; 5242 spin_lock_init(&crq->lock); 5243 5244 return retrc; 5245 5246 req_irq_failed: 5247 tasklet_kill(&adapter->tasklet); 5248 do { 5249 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address); 5250 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc)); 5251 reg_crq_failed: 5252 dma_unmap_single(dev, crq->msg_token, PAGE_SIZE, DMA_BIDIRECTIONAL); 5253 map_failed: 5254 free_page((unsigned long)crq->msgs); 5255 crq->msgs = NULL; 5256 return retrc; 5257 } 5258 5259 static int ibmvnic_reset_init(struct ibmvnic_adapter *adapter, bool reset) 5260 { 5261 struct device *dev = &adapter->vdev->dev; 5262 unsigned long timeout = msecs_to_jiffies(20000); 5263 u64 old_num_rx_queues = adapter->req_rx_queues; 5264 u64 old_num_tx_queues = adapter->req_tx_queues; 5265 int rc; 5266 5267 adapter->from_passive_init = false; 5268 5269 if (reset) 5270 reinit_completion(&adapter->init_done); 5271 5272 adapter->init_done_rc = 0; 5273 rc = ibmvnic_send_crq_init(adapter); 5274 if (rc) { 5275 dev_err(dev, "Send crq init failed with error %d\n", rc); 5276 return rc; 5277 } 5278 5279 if (!wait_for_completion_timeout(&adapter->init_done, timeout)) { 5280 dev_err(dev, "Initialization sequence timed out\n"); 5281 return -1; 5282 } 5283 5284 if (adapter->init_done_rc) { 5285 release_crq_queue(adapter); 5286 return adapter->init_done_rc; 5287 } 5288 5289 if (adapter->from_passive_init) { 5290 adapter->state = VNIC_OPEN; 5291 adapter->from_passive_init = false; 5292 return -1; 5293 } 5294 5295 if (reset && 5296 test_bit(0, &adapter->resetting) && !adapter->wait_for_reset && 5297 adapter->reset_reason != VNIC_RESET_MOBILITY) { 5298 if (adapter->req_rx_queues != old_num_rx_queues || 5299 adapter->req_tx_queues != old_num_tx_queues) { 5300 release_sub_crqs(adapter, 0); 5301 rc = init_sub_crqs(adapter); 5302 } else { 5303 rc = reset_sub_crq_queues(adapter); 5304 } 5305 } else { 5306 rc = init_sub_crqs(adapter); 5307 } 5308 5309 if (rc) { 5310 dev_err(dev, "Initialization of sub crqs failed\n"); 5311 release_crq_queue(adapter); 5312 return rc; 5313 } 5314 5315 rc = init_sub_crq_irqs(adapter); 5316 if (rc) { 5317 dev_err(dev, "Failed to initialize sub crq irqs\n"); 5318 release_crq_queue(adapter); 5319 } 5320 5321 return rc; 5322 } 5323 5324 static struct device_attribute dev_attr_failover; 5325 5326 static int ibmvnic_probe(struct vio_dev *dev, const struct vio_device_id *id) 5327 { 5328 struct ibmvnic_adapter *adapter; 5329 struct net_device *netdev; 5330 unsigned char *mac_addr_p; 5331 int rc; 5332 5333 dev_dbg(&dev->dev, "entering ibmvnic_probe for UA 0x%x\n", 5334 dev->unit_address); 5335 5336 mac_addr_p = (unsigned char *)vio_get_attribute(dev, 5337 VETH_MAC_ADDR, NULL); 5338 if (!mac_addr_p) { 5339 dev_err(&dev->dev, 5340 "(%s:%3.3d) ERROR: Can't find MAC_ADDR attribute\n", 5341 __FILE__, __LINE__); 5342 return 0; 5343 } 5344 5345 netdev = alloc_etherdev_mq(sizeof(struct ibmvnic_adapter), 5346 IBMVNIC_MAX_QUEUES); 5347 if (!netdev) 5348 return -ENOMEM; 5349 5350 adapter = netdev_priv(netdev); 5351 adapter->state = VNIC_PROBING; 5352 dev_set_drvdata(&dev->dev, netdev); 5353 adapter->vdev = dev; 5354 adapter->netdev = netdev; 5355 adapter->login_pending = false; 5356 5357 ether_addr_copy(adapter->mac_addr, mac_addr_p); 5358 ether_addr_copy(netdev->dev_addr, adapter->mac_addr); 5359 netdev->irq = dev->irq; 5360 netdev->netdev_ops = &ibmvnic_netdev_ops; 5361 netdev->ethtool_ops = &ibmvnic_ethtool_ops; 5362 SET_NETDEV_DEV(netdev, &dev->dev); 5363 5364 INIT_WORK(&adapter->ibmvnic_reset, __ibmvnic_reset); 5365 INIT_DELAYED_WORK(&adapter->ibmvnic_delayed_reset, 5366 __ibmvnic_delayed_reset); 5367 INIT_LIST_HEAD(&adapter->rwi_list); 5368 spin_lock_init(&adapter->rwi_lock); 5369 spin_lock_init(&adapter->state_lock); 5370 mutex_init(&adapter->fw_lock); 5371 init_completion(&adapter->init_done); 5372 init_completion(&adapter->fw_done); 5373 init_completion(&adapter->reset_done); 5374 init_completion(&adapter->stats_done); 5375 clear_bit(0, &adapter->resetting); 5376 5377 do { 5378 rc = init_crq_queue(adapter); 5379 if (rc) { 5380 dev_err(&dev->dev, "Couldn't initialize crq. rc=%d\n", 5381 rc); 5382 goto ibmvnic_init_fail; 5383 } 5384 5385 rc = ibmvnic_reset_init(adapter, false); 5386 if (rc && rc != EAGAIN) 5387 goto ibmvnic_init_fail; 5388 } while (rc == EAGAIN); 5389 5390 rc = init_stats_buffers(adapter); 5391 if (rc) 5392 goto ibmvnic_init_fail; 5393 5394 rc = init_stats_token(adapter); 5395 if (rc) 5396 goto ibmvnic_stats_fail; 5397 5398 netdev->mtu = adapter->req_mtu - ETH_HLEN; 5399 netdev->min_mtu = adapter->min_mtu - ETH_HLEN; 5400 netdev->max_mtu = adapter->max_mtu - ETH_HLEN; 5401 5402 rc = device_create_file(&dev->dev, &dev_attr_failover); 5403 if (rc) 5404 goto ibmvnic_dev_file_err; 5405 5406 netif_carrier_off(netdev); 5407 rc = register_netdev(netdev); 5408 if (rc) { 5409 dev_err(&dev->dev, "failed to register netdev rc=%d\n", rc); 5410 goto ibmvnic_register_fail; 5411 } 5412 dev_info(&dev->dev, "ibmvnic registered\n"); 5413 5414 adapter->state = VNIC_PROBED; 5415 5416 adapter->wait_for_reset = false; 5417 adapter->last_reset_time = jiffies; 5418 return 0; 5419 5420 ibmvnic_register_fail: 5421 device_remove_file(&dev->dev, &dev_attr_failover); 5422 5423 ibmvnic_dev_file_err: 5424 release_stats_token(adapter); 5425 5426 ibmvnic_stats_fail: 5427 release_stats_buffers(adapter); 5428 5429 ibmvnic_init_fail: 5430 release_sub_crqs(adapter, 1); 5431 release_crq_queue(adapter); 5432 mutex_destroy(&adapter->fw_lock); 5433 free_netdev(netdev); 5434 5435 return rc; 5436 } 5437 5438 static void ibmvnic_remove(struct vio_dev *dev) 5439 { 5440 struct net_device *netdev = dev_get_drvdata(&dev->dev); 5441 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 5442 unsigned long flags; 5443 5444 spin_lock_irqsave(&adapter->state_lock, flags); 5445 5446 /* If ibmvnic_reset() is scheduling a reset, wait for it to 5447 * finish. Then, set the state to REMOVING to prevent it from 5448 * scheduling any more work and to have reset functions ignore 5449 * any resets that have already been scheduled. Drop the lock 5450 * after setting state, so __ibmvnic_reset() which is called 5451 * from the flush_work() below, can make progress. 5452 */ 5453 spin_lock(&adapter->rwi_lock); 5454 adapter->state = VNIC_REMOVING; 5455 spin_unlock(&adapter->rwi_lock); 5456 5457 spin_unlock_irqrestore(&adapter->state_lock, flags); 5458 5459 flush_work(&adapter->ibmvnic_reset); 5460 flush_delayed_work(&adapter->ibmvnic_delayed_reset); 5461 5462 rtnl_lock(); 5463 unregister_netdevice(netdev); 5464 5465 release_resources(adapter); 5466 release_sub_crqs(adapter, 1); 5467 release_crq_queue(adapter); 5468 5469 release_stats_token(adapter); 5470 release_stats_buffers(adapter); 5471 5472 adapter->state = VNIC_REMOVED; 5473 5474 rtnl_unlock(); 5475 mutex_destroy(&adapter->fw_lock); 5476 device_remove_file(&dev->dev, &dev_attr_failover); 5477 free_netdev(netdev); 5478 dev_set_drvdata(&dev->dev, NULL); 5479 } 5480 5481 static ssize_t failover_store(struct device *dev, struct device_attribute *attr, 5482 const char *buf, size_t count) 5483 { 5484 struct net_device *netdev = dev_get_drvdata(dev); 5485 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 5486 unsigned long retbuf[PLPAR_HCALL_BUFSIZE]; 5487 __be64 session_token; 5488 long rc; 5489 5490 if (!sysfs_streq(buf, "1")) 5491 return -EINVAL; 5492 5493 rc = plpar_hcall(H_VIOCTL, retbuf, adapter->vdev->unit_address, 5494 H_GET_SESSION_TOKEN, 0, 0, 0); 5495 if (rc) { 5496 netdev_err(netdev, "Couldn't retrieve session token, rc %ld\n", 5497 rc); 5498 goto last_resort; 5499 } 5500 5501 session_token = (__be64)retbuf[0]; 5502 netdev_dbg(netdev, "Initiating client failover, session id %llx\n", 5503 be64_to_cpu(session_token)); 5504 rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address, 5505 H_SESSION_ERR_DETECTED, session_token, 0, 0); 5506 if (rc) 5507 netdev_err(netdev, 5508 "H_VIOCTL initiated failover failed, rc %ld\n", 5509 rc); 5510 5511 last_resort: 5512 netdev_dbg(netdev, "Trying to send CRQ_CMD, the last resort\n"); 5513 ibmvnic_reset(adapter, VNIC_RESET_FAILOVER); 5514 5515 return count; 5516 } 5517 static DEVICE_ATTR_WO(failover); 5518 5519 static unsigned long ibmvnic_get_desired_dma(struct vio_dev *vdev) 5520 { 5521 struct net_device *netdev = dev_get_drvdata(&vdev->dev); 5522 struct ibmvnic_adapter *adapter; 5523 struct iommu_table *tbl; 5524 unsigned long ret = 0; 5525 int i; 5526 5527 tbl = get_iommu_table_base(&vdev->dev); 5528 5529 /* netdev inits at probe time along with the structures we need below*/ 5530 if (!netdev) 5531 return IOMMU_PAGE_ALIGN(IBMVNIC_IO_ENTITLEMENT_DEFAULT, tbl); 5532 5533 adapter = netdev_priv(netdev); 5534 5535 ret += PAGE_SIZE; /* the crq message queue */ 5536 ret += IOMMU_PAGE_ALIGN(sizeof(struct ibmvnic_statistics), tbl); 5537 5538 for (i = 0; i < adapter->req_tx_queues + adapter->req_rx_queues; i++) 5539 ret += 4 * PAGE_SIZE; /* the scrq message queue */ 5540 5541 for (i = 0; i < adapter->num_active_rx_pools; i++) 5542 ret += adapter->rx_pool[i].size * 5543 IOMMU_PAGE_ALIGN(adapter->rx_pool[i].buff_size, tbl); 5544 5545 return ret; 5546 } 5547 5548 static int ibmvnic_resume(struct device *dev) 5549 { 5550 struct net_device *netdev = dev_get_drvdata(dev); 5551 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 5552 5553 if (adapter->state != VNIC_OPEN) 5554 return 0; 5555 5556 tasklet_schedule(&adapter->tasklet); 5557 5558 return 0; 5559 } 5560 5561 static const struct vio_device_id ibmvnic_device_table[] = { 5562 {"network", "IBM,vnic"}, 5563 {"", "" } 5564 }; 5565 MODULE_DEVICE_TABLE(vio, ibmvnic_device_table); 5566 5567 static const struct dev_pm_ops ibmvnic_pm_ops = { 5568 .resume = ibmvnic_resume 5569 }; 5570 5571 static struct vio_driver ibmvnic_driver = { 5572 .id_table = ibmvnic_device_table, 5573 .probe = ibmvnic_probe, 5574 .remove = ibmvnic_remove, 5575 .get_desired_dma = ibmvnic_get_desired_dma, 5576 .name = ibmvnic_driver_name, 5577 .pm = &ibmvnic_pm_ops, 5578 }; 5579 5580 /* module functions */ 5581 static int __init ibmvnic_module_init(void) 5582 { 5583 pr_info("%s: %s %s\n", ibmvnic_driver_name, ibmvnic_driver_string, 5584 IBMVNIC_DRIVER_VERSION); 5585 5586 return vio_register_driver(&ibmvnic_driver); 5587 } 5588 5589 static void __exit ibmvnic_module_exit(void) 5590 { 5591 vio_unregister_driver(&ibmvnic_driver); 5592 } 5593 5594 module_init(ibmvnic_module_init); 5595 module_exit(ibmvnic_module_exit); 5596