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, u32, 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 static void ibmvnic_tx_scrq_clean_buffer(struct ibmvnic_adapter *adapter, 110 struct ibmvnic_sub_crq_queue *tx_scrq); 111 static void free_long_term_buff(struct ibmvnic_adapter *adapter, 112 struct ibmvnic_long_term_buff *ltb); 113 114 struct ibmvnic_stat { 115 char name[ETH_GSTRING_LEN]; 116 int offset; 117 }; 118 119 #define IBMVNIC_STAT_OFF(stat) (offsetof(struct ibmvnic_adapter, stats) + \ 120 offsetof(struct ibmvnic_statistics, stat)) 121 #define IBMVNIC_GET_STAT(a, off) (*((u64 *)(((unsigned long)(a)) + (off)))) 122 123 static const struct ibmvnic_stat ibmvnic_stats[] = { 124 {"rx_packets", IBMVNIC_STAT_OFF(rx_packets)}, 125 {"rx_bytes", IBMVNIC_STAT_OFF(rx_bytes)}, 126 {"tx_packets", IBMVNIC_STAT_OFF(tx_packets)}, 127 {"tx_bytes", IBMVNIC_STAT_OFF(tx_bytes)}, 128 {"ucast_tx_packets", IBMVNIC_STAT_OFF(ucast_tx_packets)}, 129 {"ucast_rx_packets", IBMVNIC_STAT_OFF(ucast_rx_packets)}, 130 {"mcast_tx_packets", IBMVNIC_STAT_OFF(mcast_tx_packets)}, 131 {"mcast_rx_packets", IBMVNIC_STAT_OFF(mcast_rx_packets)}, 132 {"bcast_tx_packets", IBMVNIC_STAT_OFF(bcast_tx_packets)}, 133 {"bcast_rx_packets", IBMVNIC_STAT_OFF(bcast_rx_packets)}, 134 {"align_errors", IBMVNIC_STAT_OFF(align_errors)}, 135 {"fcs_errors", IBMVNIC_STAT_OFF(fcs_errors)}, 136 {"single_collision_frames", IBMVNIC_STAT_OFF(single_collision_frames)}, 137 {"multi_collision_frames", IBMVNIC_STAT_OFF(multi_collision_frames)}, 138 {"sqe_test_errors", IBMVNIC_STAT_OFF(sqe_test_errors)}, 139 {"deferred_tx", IBMVNIC_STAT_OFF(deferred_tx)}, 140 {"late_collisions", IBMVNIC_STAT_OFF(late_collisions)}, 141 {"excess_collisions", IBMVNIC_STAT_OFF(excess_collisions)}, 142 {"internal_mac_tx_errors", IBMVNIC_STAT_OFF(internal_mac_tx_errors)}, 143 {"carrier_sense", IBMVNIC_STAT_OFF(carrier_sense)}, 144 {"too_long_frames", IBMVNIC_STAT_OFF(too_long_frames)}, 145 {"internal_mac_rx_errors", IBMVNIC_STAT_OFF(internal_mac_rx_errors)}, 146 }; 147 148 static int send_crq_init_complete(struct ibmvnic_adapter *adapter) 149 { 150 union ibmvnic_crq crq; 151 152 memset(&crq, 0, sizeof(crq)); 153 crq.generic.first = IBMVNIC_CRQ_INIT_CMD; 154 crq.generic.cmd = IBMVNIC_CRQ_INIT_COMPLETE; 155 156 return ibmvnic_send_crq(adapter, &crq); 157 } 158 159 static int send_version_xchg(struct ibmvnic_adapter *adapter) 160 { 161 union ibmvnic_crq crq; 162 163 memset(&crq, 0, sizeof(crq)); 164 crq.version_exchange.first = IBMVNIC_CRQ_CMD; 165 crq.version_exchange.cmd = VERSION_EXCHANGE; 166 crq.version_exchange.version = cpu_to_be16(ibmvnic_version); 167 168 return ibmvnic_send_crq(adapter, &crq); 169 } 170 171 static long h_reg_sub_crq(unsigned long unit_address, unsigned long token, 172 unsigned long length, unsigned long *number, 173 unsigned long *irq) 174 { 175 unsigned long retbuf[PLPAR_HCALL_BUFSIZE]; 176 long rc; 177 178 rc = plpar_hcall(H_REG_SUB_CRQ, retbuf, unit_address, token, length); 179 *number = retbuf[0]; 180 *irq = retbuf[1]; 181 182 return rc; 183 } 184 185 /** 186 * ibmvnic_wait_for_completion - Check device state and wait for completion 187 * @adapter: private device data 188 * @comp_done: completion structure to wait for 189 * @timeout: time to wait in milliseconds 190 * 191 * Wait for a completion signal or until the timeout limit is reached 192 * while checking that the device is still active. 193 */ 194 static int ibmvnic_wait_for_completion(struct ibmvnic_adapter *adapter, 195 struct completion *comp_done, 196 unsigned long timeout) 197 { 198 struct net_device *netdev; 199 unsigned long div_timeout; 200 u8 retry; 201 202 netdev = adapter->netdev; 203 retry = 5; 204 div_timeout = msecs_to_jiffies(timeout / retry); 205 while (true) { 206 if (!adapter->crq.active) { 207 netdev_err(netdev, "Device down!\n"); 208 return -ENODEV; 209 } 210 if (!retry--) 211 break; 212 if (wait_for_completion_timeout(comp_done, div_timeout)) 213 return 0; 214 } 215 netdev_err(netdev, "Operation timed out.\n"); 216 return -ETIMEDOUT; 217 } 218 219 /** 220 * reuse_ltb() - Check if a long term buffer can be reused 221 * @ltb: The long term buffer to be checked 222 * @size: The size of the long term buffer. 223 * 224 * An LTB can be reused unless its size has changed. 225 * 226 * Return: Return true if the LTB can be reused, false otherwise. 227 */ 228 static bool reuse_ltb(struct ibmvnic_long_term_buff *ltb, int size) 229 { 230 return (ltb->buff && ltb->size == size); 231 } 232 233 /** 234 * alloc_long_term_buff() - Allocate a long term buffer (LTB) 235 * 236 * @adapter: ibmvnic adapter associated to the LTB 237 * @ltb: container object for the LTB 238 * @size: size of the LTB 239 * 240 * Allocate an LTB of the specified size and notify VIOS. 241 * 242 * If the given @ltb already has the correct size, reuse it. Otherwise if 243 * its non-NULL, free it. Then allocate a new one of the correct size. 244 * Notify the VIOS either way since we may now be working with a new VIOS. 245 * 246 * Allocating larger chunks of memory during resets, specially LPM or under 247 * low memory situations can cause resets to fail/timeout and for LPAR to 248 * lose connectivity. So hold onto the LTB even if we fail to communicate 249 * with the VIOS and reuse it on next open. Free LTB when adapter is closed. 250 * 251 * Return: 0 if we were able to allocate the LTB and notify the VIOS and 252 * a negative value otherwise. 253 */ 254 static int alloc_long_term_buff(struct ibmvnic_adapter *adapter, 255 struct ibmvnic_long_term_buff *ltb, int size) 256 { 257 struct device *dev = &adapter->vdev->dev; 258 int rc; 259 260 if (!reuse_ltb(ltb, size)) { 261 dev_dbg(dev, 262 "LTB size changed from 0x%llx to 0x%x, reallocating\n", 263 ltb->size, size); 264 free_long_term_buff(adapter, ltb); 265 } 266 267 if (ltb->buff) { 268 dev_dbg(dev, "Reusing LTB [map %d, size 0x%llx]\n", 269 ltb->map_id, ltb->size); 270 } else { 271 ltb->buff = dma_alloc_coherent(dev, size, <b->addr, 272 GFP_KERNEL); 273 if (!ltb->buff) { 274 dev_err(dev, "Couldn't alloc long term buffer\n"); 275 return -ENOMEM; 276 } 277 ltb->size = size; 278 279 ltb->map_id = find_first_zero_bit(adapter->map_ids, 280 MAX_MAP_ID); 281 bitmap_set(adapter->map_ids, ltb->map_id, 1); 282 283 dev_dbg(dev, 284 "Allocated new LTB [map %d, size 0x%llx]\n", 285 ltb->map_id, ltb->size); 286 } 287 288 /* Ensure ltb is zeroed - specially when reusing it. */ 289 memset(ltb->buff, 0, ltb->size); 290 291 mutex_lock(&adapter->fw_lock); 292 adapter->fw_done_rc = 0; 293 reinit_completion(&adapter->fw_done); 294 295 rc = send_request_map(adapter, ltb->addr, ltb->size, ltb->map_id); 296 if (rc) { 297 dev_err(dev, "send_request_map failed, rc = %d\n", rc); 298 goto out; 299 } 300 301 rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000); 302 if (rc) { 303 dev_err(dev, "LTB map request aborted or timed out, rc = %d\n", 304 rc); 305 goto out; 306 } 307 308 if (adapter->fw_done_rc) { 309 dev_err(dev, "Couldn't map LTB, rc = %d\n", 310 adapter->fw_done_rc); 311 rc = -1; 312 goto out; 313 } 314 rc = 0; 315 out: 316 /* don't free LTB on communication error - see function header */ 317 mutex_unlock(&adapter->fw_lock); 318 return rc; 319 } 320 321 static void free_long_term_buff(struct ibmvnic_adapter *adapter, 322 struct ibmvnic_long_term_buff *ltb) 323 { 324 struct device *dev = &adapter->vdev->dev; 325 326 if (!ltb->buff) 327 return; 328 329 /* VIOS automatically unmaps the long term buffer at remote 330 * end for the following resets: 331 * FAILOVER, MOBILITY, TIMEOUT. 332 */ 333 if (adapter->reset_reason != VNIC_RESET_FAILOVER && 334 adapter->reset_reason != VNIC_RESET_MOBILITY && 335 adapter->reset_reason != VNIC_RESET_TIMEOUT) 336 send_request_unmap(adapter, ltb->map_id); 337 338 dma_free_coherent(dev, ltb->size, ltb->buff, ltb->addr); 339 340 ltb->buff = NULL; 341 /* mark this map_id free */ 342 bitmap_clear(adapter->map_ids, ltb->map_id, 1); 343 ltb->map_id = 0; 344 } 345 346 static void deactivate_rx_pools(struct ibmvnic_adapter *adapter) 347 { 348 int i; 349 350 for (i = 0; i < adapter->num_active_rx_pools; i++) 351 adapter->rx_pool[i].active = 0; 352 } 353 354 static void replenish_rx_pool(struct ibmvnic_adapter *adapter, 355 struct ibmvnic_rx_pool *pool) 356 { 357 int count = pool->size - atomic_read(&pool->available); 358 u64 handle = adapter->rx_scrq[pool->index]->handle; 359 struct device *dev = &adapter->vdev->dev; 360 struct ibmvnic_ind_xmit_queue *ind_bufp; 361 struct ibmvnic_sub_crq_queue *rx_scrq; 362 union sub_crq *sub_crq; 363 int buffers_added = 0; 364 unsigned long lpar_rc; 365 struct sk_buff *skb; 366 unsigned int offset; 367 dma_addr_t dma_addr; 368 unsigned char *dst; 369 int shift = 0; 370 int index; 371 int i; 372 373 if (!pool->active) 374 return; 375 376 rx_scrq = adapter->rx_scrq[pool->index]; 377 ind_bufp = &rx_scrq->ind_buf; 378 379 /* netdev_skb_alloc() could have failed after we saved a few skbs 380 * in the indir_buf and we would not have sent them to VIOS yet. 381 * To account for them, start the loop at ind_bufp->index rather 382 * than 0. If we pushed all the skbs to VIOS, ind_bufp->index will 383 * be 0. 384 */ 385 for (i = ind_bufp->index; i < count; ++i) { 386 index = pool->free_map[pool->next_free]; 387 388 /* We maybe reusing the skb from earlier resets. Allocate 389 * only if necessary. But since the LTB may have changed 390 * during reset (see init_rx_pools()), update LTB below 391 * even if reusing skb. 392 */ 393 skb = pool->rx_buff[index].skb; 394 if (!skb) { 395 skb = netdev_alloc_skb(adapter->netdev, 396 pool->buff_size); 397 if (!skb) { 398 dev_err(dev, "Couldn't replenish rx buff\n"); 399 adapter->replenish_no_mem++; 400 break; 401 } 402 } 403 404 pool->free_map[pool->next_free] = IBMVNIC_INVALID_MAP; 405 pool->next_free = (pool->next_free + 1) % pool->size; 406 407 /* Copy the skb to the long term mapped DMA buffer */ 408 offset = index * pool->buff_size; 409 dst = pool->long_term_buff.buff + offset; 410 memset(dst, 0, pool->buff_size); 411 dma_addr = pool->long_term_buff.addr + offset; 412 413 /* add the skb to an rx_buff in the pool */ 414 pool->rx_buff[index].data = dst; 415 pool->rx_buff[index].dma = dma_addr; 416 pool->rx_buff[index].skb = skb; 417 pool->rx_buff[index].pool_index = pool->index; 418 pool->rx_buff[index].size = pool->buff_size; 419 420 /* queue the rx_buff for the next send_subcrq_indirect */ 421 sub_crq = &ind_bufp->indir_arr[ind_bufp->index++]; 422 memset(sub_crq, 0, sizeof(*sub_crq)); 423 sub_crq->rx_add.first = IBMVNIC_CRQ_CMD; 424 sub_crq->rx_add.correlator = 425 cpu_to_be64((u64)&pool->rx_buff[index]); 426 sub_crq->rx_add.ioba = cpu_to_be32(dma_addr); 427 sub_crq->rx_add.map_id = pool->long_term_buff.map_id; 428 429 /* The length field of the sCRQ is defined to be 24 bits so the 430 * buffer size needs to be left shifted by a byte before it is 431 * converted to big endian to prevent the last byte from being 432 * truncated. 433 */ 434 #ifdef __LITTLE_ENDIAN__ 435 shift = 8; 436 #endif 437 sub_crq->rx_add.len = cpu_to_be32(pool->buff_size << shift); 438 439 /* if send_subcrq_indirect queue is full, flush to VIOS */ 440 if (ind_bufp->index == IBMVNIC_MAX_IND_DESCS || 441 i == count - 1) { 442 lpar_rc = 443 send_subcrq_indirect(adapter, handle, 444 (u64)ind_bufp->indir_dma, 445 (u64)ind_bufp->index); 446 if (lpar_rc != H_SUCCESS) 447 goto failure; 448 buffers_added += ind_bufp->index; 449 adapter->replenish_add_buff_success += ind_bufp->index; 450 ind_bufp->index = 0; 451 } 452 } 453 atomic_add(buffers_added, &pool->available); 454 return; 455 456 failure: 457 if (lpar_rc != H_PARAMETER && lpar_rc != H_CLOSED) 458 dev_err_ratelimited(dev, "rx: replenish packet buffer failed\n"); 459 for (i = ind_bufp->index - 1; i >= 0; --i) { 460 struct ibmvnic_rx_buff *rx_buff; 461 462 pool->next_free = pool->next_free == 0 ? 463 pool->size - 1 : pool->next_free - 1; 464 sub_crq = &ind_bufp->indir_arr[i]; 465 rx_buff = (struct ibmvnic_rx_buff *) 466 be64_to_cpu(sub_crq->rx_add.correlator); 467 index = (int)(rx_buff - pool->rx_buff); 468 pool->free_map[pool->next_free] = index; 469 dev_kfree_skb_any(pool->rx_buff[index].skb); 470 pool->rx_buff[index].skb = NULL; 471 } 472 adapter->replenish_add_buff_failure += ind_bufp->index; 473 atomic_add(buffers_added, &pool->available); 474 ind_bufp->index = 0; 475 if (lpar_rc == H_CLOSED || adapter->failover_pending) { 476 /* Disable buffer pool replenishment and report carrier off if 477 * queue is closed or pending failover. 478 * Firmware guarantees that a signal will be sent to the 479 * driver, triggering a reset. 480 */ 481 deactivate_rx_pools(adapter); 482 netif_carrier_off(adapter->netdev); 483 } 484 } 485 486 static void replenish_pools(struct ibmvnic_adapter *adapter) 487 { 488 int i; 489 490 adapter->replenish_task_cycles++; 491 for (i = 0; i < adapter->num_active_rx_pools; i++) { 492 if (adapter->rx_pool[i].active) 493 replenish_rx_pool(adapter, &adapter->rx_pool[i]); 494 } 495 496 netdev_dbg(adapter->netdev, "Replenished %d pools\n", i); 497 } 498 499 static void release_stats_buffers(struct ibmvnic_adapter *adapter) 500 { 501 kfree(adapter->tx_stats_buffers); 502 kfree(adapter->rx_stats_buffers); 503 adapter->tx_stats_buffers = NULL; 504 adapter->rx_stats_buffers = NULL; 505 } 506 507 static int init_stats_buffers(struct ibmvnic_adapter *adapter) 508 { 509 adapter->tx_stats_buffers = 510 kcalloc(IBMVNIC_MAX_QUEUES, 511 sizeof(struct ibmvnic_tx_queue_stats), 512 GFP_KERNEL); 513 if (!adapter->tx_stats_buffers) 514 return -ENOMEM; 515 516 adapter->rx_stats_buffers = 517 kcalloc(IBMVNIC_MAX_QUEUES, 518 sizeof(struct ibmvnic_rx_queue_stats), 519 GFP_KERNEL); 520 if (!adapter->rx_stats_buffers) 521 return -ENOMEM; 522 523 return 0; 524 } 525 526 static void release_stats_token(struct ibmvnic_adapter *adapter) 527 { 528 struct device *dev = &adapter->vdev->dev; 529 530 if (!adapter->stats_token) 531 return; 532 533 dma_unmap_single(dev, adapter->stats_token, 534 sizeof(struct ibmvnic_statistics), 535 DMA_FROM_DEVICE); 536 adapter->stats_token = 0; 537 } 538 539 static int init_stats_token(struct ibmvnic_adapter *adapter) 540 { 541 struct device *dev = &adapter->vdev->dev; 542 dma_addr_t stok; 543 544 stok = dma_map_single(dev, &adapter->stats, 545 sizeof(struct ibmvnic_statistics), 546 DMA_FROM_DEVICE); 547 if (dma_mapping_error(dev, stok)) { 548 dev_err(dev, "Couldn't map stats buffer\n"); 549 return -1; 550 } 551 552 adapter->stats_token = stok; 553 netdev_dbg(adapter->netdev, "Stats token initialized (%llx)\n", stok); 554 return 0; 555 } 556 557 /** 558 * release_rx_pools() - Release any rx pools attached to @adapter. 559 * @adapter: ibmvnic adapter 560 * 561 * Safe to call this multiple times - even if no pools are attached. 562 */ 563 static void release_rx_pools(struct ibmvnic_adapter *adapter) 564 { 565 struct ibmvnic_rx_pool *rx_pool; 566 int i, j; 567 568 if (!adapter->rx_pool) 569 return; 570 571 for (i = 0; i < adapter->num_active_rx_pools; i++) { 572 rx_pool = &adapter->rx_pool[i]; 573 574 netdev_dbg(adapter->netdev, "Releasing rx_pool[%d]\n", i); 575 576 kfree(rx_pool->free_map); 577 578 free_long_term_buff(adapter, &rx_pool->long_term_buff); 579 580 if (!rx_pool->rx_buff) 581 continue; 582 583 for (j = 0; j < rx_pool->size; j++) { 584 if (rx_pool->rx_buff[j].skb) { 585 dev_kfree_skb_any(rx_pool->rx_buff[j].skb); 586 rx_pool->rx_buff[j].skb = NULL; 587 } 588 } 589 590 kfree(rx_pool->rx_buff); 591 } 592 593 kfree(adapter->rx_pool); 594 adapter->rx_pool = NULL; 595 adapter->num_active_rx_pools = 0; 596 adapter->prev_rx_pool_size = 0; 597 } 598 599 /** 600 * reuse_rx_pools() - Check if the existing rx pools can be reused. 601 * @adapter: ibmvnic adapter 602 * 603 * Check if the existing rx pools in the adapter can be reused. The 604 * pools can be reused if the pool parameters (number of pools, 605 * number of buffers in the pool and size of each buffer) have not 606 * changed. 607 * 608 * NOTE: This assumes that all pools have the same number of buffers 609 * which is the case currently. If that changes, we must fix this. 610 * 611 * Return: true if the rx pools can be reused, false otherwise. 612 */ 613 static bool reuse_rx_pools(struct ibmvnic_adapter *adapter) 614 { 615 u64 old_num_pools, new_num_pools; 616 u64 old_pool_size, new_pool_size; 617 u64 old_buff_size, new_buff_size; 618 619 if (!adapter->rx_pool) 620 return false; 621 622 old_num_pools = adapter->num_active_rx_pools; 623 new_num_pools = adapter->req_rx_queues; 624 625 old_pool_size = adapter->prev_rx_pool_size; 626 new_pool_size = adapter->req_rx_add_entries_per_subcrq; 627 628 old_buff_size = adapter->prev_rx_buf_sz; 629 new_buff_size = adapter->cur_rx_buf_sz; 630 631 /* Require buff size to be exactly same for now */ 632 if (old_buff_size != new_buff_size) 633 return false; 634 635 if (old_num_pools == new_num_pools && old_pool_size == new_pool_size) 636 return true; 637 638 if (old_num_pools < adapter->min_rx_queues || 639 old_num_pools > adapter->max_rx_queues || 640 old_pool_size < adapter->min_rx_add_entries_per_subcrq || 641 old_pool_size > adapter->max_rx_add_entries_per_subcrq) 642 return false; 643 644 return true; 645 } 646 647 /** 648 * init_rx_pools(): Initialize the set of receiver pools in the adapter. 649 * @netdev: net device associated with the vnic interface 650 * 651 * Initialize the set of receiver pools in the ibmvnic adapter associated 652 * with the net_device @netdev. If possible, reuse the existing rx pools. 653 * Otherwise free any existing pools and allocate a new set of pools 654 * before initializing them. 655 * 656 * Return: 0 on success and negative value on error. 657 */ 658 static int init_rx_pools(struct net_device *netdev) 659 { 660 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 661 struct device *dev = &adapter->vdev->dev; 662 struct ibmvnic_rx_pool *rx_pool; 663 u64 num_pools; 664 u64 pool_size; /* # of buffers in one pool */ 665 u64 buff_size; 666 int i, j; 667 668 pool_size = adapter->req_rx_add_entries_per_subcrq; 669 num_pools = adapter->req_rx_queues; 670 buff_size = adapter->cur_rx_buf_sz; 671 672 if (reuse_rx_pools(adapter)) { 673 dev_dbg(dev, "Reusing rx pools\n"); 674 goto update_ltb; 675 } 676 677 /* Allocate/populate the pools. */ 678 release_rx_pools(adapter); 679 680 adapter->rx_pool = kcalloc(num_pools, 681 sizeof(struct ibmvnic_rx_pool), 682 GFP_KERNEL); 683 if (!adapter->rx_pool) { 684 dev_err(dev, "Failed to allocate rx pools\n"); 685 return -1; 686 } 687 688 /* Set num_active_rx_pools early. If we fail below after partial 689 * allocation, release_rx_pools() will know how many to look for. 690 */ 691 adapter->num_active_rx_pools = num_pools; 692 693 for (i = 0; i < num_pools; i++) { 694 rx_pool = &adapter->rx_pool[i]; 695 696 netdev_dbg(adapter->netdev, 697 "Initializing rx_pool[%d], %lld buffs, %lld bytes each\n", 698 i, pool_size, buff_size); 699 700 rx_pool->size = pool_size; 701 rx_pool->index = i; 702 rx_pool->buff_size = ALIGN(buff_size, L1_CACHE_BYTES); 703 704 rx_pool->free_map = kcalloc(rx_pool->size, sizeof(int), 705 GFP_KERNEL); 706 if (!rx_pool->free_map) { 707 dev_err(dev, "Couldn't alloc free_map %d\n", i); 708 goto out_release; 709 } 710 711 rx_pool->rx_buff = kcalloc(rx_pool->size, 712 sizeof(struct ibmvnic_rx_buff), 713 GFP_KERNEL); 714 if (!rx_pool->rx_buff) { 715 dev_err(dev, "Couldn't alloc rx buffers\n"); 716 goto out_release; 717 } 718 } 719 720 adapter->prev_rx_pool_size = pool_size; 721 adapter->prev_rx_buf_sz = adapter->cur_rx_buf_sz; 722 723 update_ltb: 724 for (i = 0; i < num_pools; i++) { 725 rx_pool = &adapter->rx_pool[i]; 726 dev_dbg(dev, "Updating LTB for rx pool %d [%d, %d]\n", 727 i, rx_pool->size, rx_pool->buff_size); 728 729 if (alloc_long_term_buff(adapter, &rx_pool->long_term_buff, 730 rx_pool->size * rx_pool->buff_size)) 731 goto out; 732 733 for (j = 0; j < rx_pool->size; ++j) { 734 struct ibmvnic_rx_buff *rx_buff; 735 736 rx_pool->free_map[j] = j; 737 738 /* NOTE: Don't clear rx_buff->skb here - will leak 739 * memory! replenish_rx_pool() will reuse skbs or 740 * allocate as necessary. 741 */ 742 rx_buff = &rx_pool->rx_buff[j]; 743 rx_buff->dma = 0; 744 rx_buff->data = 0; 745 rx_buff->size = 0; 746 rx_buff->pool_index = 0; 747 } 748 749 /* Mark pool "empty" so replenish_rx_pools() will 750 * update the LTB info for each buffer 751 */ 752 atomic_set(&rx_pool->available, 0); 753 rx_pool->next_alloc = 0; 754 rx_pool->next_free = 0; 755 /* replenish_rx_pool() may have called deactivate_rx_pools() 756 * on failover. Ensure pool is active now. 757 */ 758 rx_pool->active = 1; 759 } 760 return 0; 761 out_release: 762 release_rx_pools(adapter); 763 out: 764 /* We failed to allocate one or more LTBs or map them on the VIOS. 765 * Hold onto the pools and any LTBs that we did allocate/map. 766 */ 767 return -1; 768 } 769 770 static void release_vpd_data(struct ibmvnic_adapter *adapter) 771 { 772 if (!adapter->vpd) 773 return; 774 775 kfree(adapter->vpd->buff); 776 kfree(adapter->vpd); 777 778 adapter->vpd = NULL; 779 } 780 781 static void release_one_tx_pool(struct ibmvnic_adapter *adapter, 782 struct ibmvnic_tx_pool *tx_pool) 783 { 784 kfree(tx_pool->tx_buff); 785 kfree(tx_pool->free_map); 786 free_long_term_buff(adapter, &tx_pool->long_term_buff); 787 } 788 789 /** 790 * release_tx_pools() - Release any tx pools attached to @adapter. 791 * @adapter: ibmvnic adapter 792 * 793 * Safe to call this multiple times - even if no pools are attached. 794 */ 795 static void release_tx_pools(struct ibmvnic_adapter *adapter) 796 { 797 int i; 798 799 /* init_tx_pools() ensures that ->tx_pool and ->tso_pool are 800 * both NULL or both non-NULL. So we only need to check one. 801 */ 802 if (!adapter->tx_pool) 803 return; 804 805 for (i = 0; i < adapter->num_active_tx_pools; i++) { 806 release_one_tx_pool(adapter, &adapter->tx_pool[i]); 807 release_one_tx_pool(adapter, &adapter->tso_pool[i]); 808 } 809 810 kfree(adapter->tx_pool); 811 adapter->tx_pool = NULL; 812 kfree(adapter->tso_pool); 813 adapter->tso_pool = NULL; 814 adapter->num_active_tx_pools = 0; 815 adapter->prev_tx_pool_size = 0; 816 } 817 818 static int init_one_tx_pool(struct net_device *netdev, 819 struct ibmvnic_tx_pool *tx_pool, 820 int pool_size, int buf_size) 821 { 822 int i; 823 824 tx_pool->tx_buff = kcalloc(pool_size, 825 sizeof(struct ibmvnic_tx_buff), 826 GFP_KERNEL); 827 if (!tx_pool->tx_buff) 828 return -1; 829 830 tx_pool->free_map = kcalloc(pool_size, sizeof(int), GFP_KERNEL); 831 if (!tx_pool->free_map) { 832 kfree(tx_pool->tx_buff); 833 tx_pool->tx_buff = NULL; 834 return -1; 835 } 836 837 for (i = 0; i < pool_size; i++) 838 tx_pool->free_map[i] = i; 839 840 tx_pool->consumer_index = 0; 841 tx_pool->producer_index = 0; 842 tx_pool->num_buffers = pool_size; 843 tx_pool->buf_size = buf_size; 844 845 return 0; 846 } 847 848 /** 849 * reuse_tx_pools() - Check if the existing tx pools can be reused. 850 * @adapter: ibmvnic adapter 851 * 852 * Check if the existing tx pools in the adapter can be reused. The 853 * pools can be reused if the pool parameters (number of pools, 854 * number of buffers in the pool and mtu) have not changed. 855 * 856 * NOTE: This assumes that all pools have the same number of buffers 857 * which is the case currently. If that changes, we must fix this. 858 * 859 * Return: true if the tx pools can be reused, false otherwise. 860 */ 861 static bool reuse_tx_pools(struct ibmvnic_adapter *adapter) 862 { 863 u64 old_num_pools, new_num_pools; 864 u64 old_pool_size, new_pool_size; 865 u64 old_mtu, new_mtu; 866 867 if (!adapter->tx_pool) 868 return false; 869 870 old_num_pools = adapter->num_active_tx_pools; 871 new_num_pools = adapter->num_active_tx_scrqs; 872 old_pool_size = adapter->prev_tx_pool_size; 873 new_pool_size = adapter->req_tx_entries_per_subcrq; 874 old_mtu = adapter->prev_mtu; 875 new_mtu = adapter->req_mtu; 876 877 /* Require MTU to be exactly same to reuse pools for now */ 878 if (old_mtu != new_mtu) 879 return false; 880 881 if (old_num_pools == new_num_pools && old_pool_size == new_pool_size) 882 return true; 883 884 if (old_num_pools < adapter->min_tx_queues || 885 old_num_pools > adapter->max_tx_queues || 886 old_pool_size < adapter->min_tx_entries_per_subcrq || 887 old_pool_size > adapter->max_tx_entries_per_subcrq) 888 return false; 889 890 return true; 891 } 892 893 /** 894 * init_tx_pools(): Initialize the set of transmit pools in the adapter. 895 * @netdev: net device associated with the vnic interface 896 * 897 * Initialize the set of transmit pools in the ibmvnic adapter associated 898 * with the net_device @netdev. If possible, reuse the existing tx pools. 899 * Otherwise free any existing pools and allocate a new set of pools 900 * before initializing them. 901 * 902 * Return: 0 on success and negative value on error. 903 */ 904 static int init_tx_pools(struct net_device *netdev) 905 { 906 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 907 struct device *dev = &adapter->vdev->dev; 908 int num_pools; 909 u64 pool_size; /* # of buffers in pool */ 910 u64 buff_size; 911 int i, j, rc; 912 913 num_pools = adapter->req_tx_queues; 914 915 /* We must notify the VIOS about the LTB on all resets - but we only 916 * need to alloc/populate pools if either the number of buffers or 917 * size of each buffer in the pool has changed. 918 */ 919 if (reuse_tx_pools(adapter)) { 920 netdev_dbg(netdev, "Reusing tx pools\n"); 921 goto update_ltb; 922 } 923 924 /* Allocate/populate the pools. */ 925 release_tx_pools(adapter); 926 927 pool_size = adapter->req_tx_entries_per_subcrq; 928 num_pools = adapter->num_active_tx_scrqs; 929 930 adapter->tx_pool = kcalloc(num_pools, 931 sizeof(struct ibmvnic_tx_pool), GFP_KERNEL); 932 if (!adapter->tx_pool) 933 return -1; 934 935 adapter->tso_pool = kcalloc(num_pools, 936 sizeof(struct ibmvnic_tx_pool), GFP_KERNEL); 937 /* To simplify release_tx_pools() ensure that ->tx_pool and 938 * ->tso_pool are either both NULL or both non-NULL. 939 */ 940 if (!adapter->tso_pool) { 941 kfree(adapter->tx_pool); 942 adapter->tx_pool = NULL; 943 return -1; 944 } 945 946 /* Set num_active_tx_pools early. If we fail below after partial 947 * allocation, release_tx_pools() will know how many to look for. 948 */ 949 adapter->num_active_tx_pools = num_pools; 950 951 buff_size = adapter->req_mtu + VLAN_HLEN; 952 buff_size = ALIGN(buff_size, L1_CACHE_BYTES); 953 954 for (i = 0; i < num_pools; i++) { 955 dev_dbg(dev, "Init tx pool %d [%llu, %llu]\n", 956 i, adapter->req_tx_entries_per_subcrq, buff_size); 957 958 rc = init_one_tx_pool(netdev, &adapter->tx_pool[i], 959 pool_size, buff_size); 960 if (rc) 961 goto out_release; 962 963 rc = init_one_tx_pool(netdev, &adapter->tso_pool[i], 964 IBMVNIC_TSO_BUFS, 965 IBMVNIC_TSO_BUF_SZ); 966 if (rc) 967 goto out_release; 968 } 969 970 adapter->prev_tx_pool_size = pool_size; 971 adapter->prev_mtu = adapter->req_mtu; 972 973 update_ltb: 974 /* NOTE: All tx_pools have the same number of buffers (which is 975 * same as pool_size). All tso_pools have IBMVNIC_TSO_BUFS 976 * buffers (see calls init_one_tx_pool() for these). 977 * For consistency, we use tx_pool->num_buffers and 978 * tso_pool->num_buffers below. 979 */ 980 rc = -1; 981 for (i = 0; i < num_pools; i++) { 982 struct ibmvnic_tx_pool *tso_pool; 983 struct ibmvnic_tx_pool *tx_pool; 984 u32 ltb_size; 985 986 tx_pool = &adapter->tx_pool[i]; 987 ltb_size = tx_pool->num_buffers * tx_pool->buf_size; 988 if (alloc_long_term_buff(adapter, &tx_pool->long_term_buff, 989 ltb_size)) 990 goto out; 991 992 dev_dbg(dev, "Updated LTB for tx pool %d [%p, %d, %d]\n", 993 i, tx_pool->long_term_buff.buff, 994 tx_pool->num_buffers, tx_pool->buf_size); 995 996 tx_pool->consumer_index = 0; 997 tx_pool->producer_index = 0; 998 999 for (j = 0; j < tx_pool->num_buffers; j++) 1000 tx_pool->free_map[j] = j; 1001 1002 tso_pool = &adapter->tso_pool[i]; 1003 ltb_size = tso_pool->num_buffers * tso_pool->buf_size; 1004 if (alloc_long_term_buff(adapter, &tso_pool->long_term_buff, 1005 ltb_size)) 1006 goto out; 1007 1008 dev_dbg(dev, "Updated LTB for tso pool %d [%p, %d, %d]\n", 1009 i, tso_pool->long_term_buff.buff, 1010 tso_pool->num_buffers, tso_pool->buf_size); 1011 1012 tso_pool->consumer_index = 0; 1013 tso_pool->producer_index = 0; 1014 1015 for (j = 0; j < tso_pool->num_buffers; j++) 1016 tso_pool->free_map[j] = j; 1017 } 1018 1019 return 0; 1020 out_release: 1021 release_tx_pools(adapter); 1022 out: 1023 /* We failed to allocate one or more LTBs or map them on the VIOS. 1024 * Hold onto the pools and any LTBs that we did allocate/map. 1025 */ 1026 return rc; 1027 } 1028 1029 static void ibmvnic_napi_enable(struct ibmvnic_adapter *adapter) 1030 { 1031 int i; 1032 1033 if (adapter->napi_enabled) 1034 return; 1035 1036 for (i = 0; i < adapter->req_rx_queues; i++) 1037 napi_enable(&adapter->napi[i]); 1038 1039 adapter->napi_enabled = true; 1040 } 1041 1042 static void ibmvnic_napi_disable(struct ibmvnic_adapter *adapter) 1043 { 1044 int i; 1045 1046 if (!adapter->napi_enabled) 1047 return; 1048 1049 for (i = 0; i < adapter->req_rx_queues; i++) { 1050 netdev_dbg(adapter->netdev, "Disabling napi[%d]\n", i); 1051 napi_disable(&adapter->napi[i]); 1052 } 1053 1054 adapter->napi_enabled = false; 1055 } 1056 1057 static int init_napi(struct ibmvnic_adapter *adapter) 1058 { 1059 int i; 1060 1061 adapter->napi = kcalloc(adapter->req_rx_queues, 1062 sizeof(struct napi_struct), GFP_KERNEL); 1063 if (!adapter->napi) 1064 return -ENOMEM; 1065 1066 for (i = 0; i < adapter->req_rx_queues; i++) { 1067 netdev_dbg(adapter->netdev, "Adding napi[%d]\n", i); 1068 netif_napi_add(adapter->netdev, &adapter->napi[i], 1069 ibmvnic_poll, NAPI_POLL_WEIGHT); 1070 } 1071 1072 adapter->num_active_rx_napi = adapter->req_rx_queues; 1073 return 0; 1074 } 1075 1076 static void release_napi(struct ibmvnic_adapter *adapter) 1077 { 1078 int i; 1079 1080 if (!adapter->napi) 1081 return; 1082 1083 for (i = 0; i < adapter->num_active_rx_napi; i++) { 1084 netdev_dbg(adapter->netdev, "Releasing napi[%d]\n", i); 1085 netif_napi_del(&adapter->napi[i]); 1086 } 1087 1088 kfree(adapter->napi); 1089 adapter->napi = NULL; 1090 adapter->num_active_rx_napi = 0; 1091 adapter->napi_enabled = false; 1092 } 1093 1094 static const char *adapter_state_to_string(enum vnic_state state) 1095 { 1096 switch (state) { 1097 case VNIC_PROBING: 1098 return "PROBING"; 1099 case VNIC_PROBED: 1100 return "PROBED"; 1101 case VNIC_OPENING: 1102 return "OPENING"; 1103 case VNIC_OPEN: 1104 return "OPEN"; 1105 case VNIC_CLOSING: 1106 return "CLOSING"; 1107 case VNIC_CLOSED: 1108 return "CLOSED"; 1109 case VNIC_REMOVING: 1110 return "REMOVING"; 1111 case VNIC_REMOVED: 1112 return "REMOVED"; 1113 case VNIC_DOWN: 1114 return "DOWN"; 1115 } 1116 return "UNKNOWN"; 1117 } 1118 1119 static int ibmvnic_login(struct net_device *netdev) 1120 { 1121 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 1122 unsigned long timeout = msecs_to_jiffies(20000); 1123 int retry_count = 0; 1124 int retries = 10; 1125 bool retry; 1126 int rc; 1127 1128 do { 1129 retry = false; 1130 if (retry_count > retries) { 1131 netdev_warn(netdev, "Login attempts exceeded\n"); 1132 return -1; 1133 } 1134 1135 adapter->init_done_rc = 0; 1136 reinit_completion(&adapter->init_done); 1137 rc = send_login(adapter); 1138 if (rc) 1139 return rc; 1140 1141 if (!wait_for_completion_timeout(&adapter->init_done, 1142 timeout)) { 1143 netdev_warn(netdev, "Login timed out, retrying...\n"); 1144 retry = true; 1145 adapter->init_done_rc = 0; 1146 retry_count++; 1147 continue; 1148 } 1149 1150 if (adapter->init_done_rc == ABORTED) { 1151 netdev_warn(netdev, "Login aborted, retrying...\n"); 1152 retry = true; 1153 adapter->init_done_rc = 0; 1154 retry_count++; 1155 /* FW or device may be busy, so 1156 * wait a bit before retrying login 1157 */ 1158 msleep(500); 1159 } else if (adapter->init_done_rc == PARTIALSUCCESS) { 1160 retry_count++; 1161 release_sub_crqs(adapter, 1); 1162 1163 retry = true; 1164 netdev_dbg(netdev, 1165 "Received partial success, retrying...\n"); 1166 adapter->init_done_rc = 0; 1167 reinit_completion(&adapter->init_done); 1168 send_query_cap(adapter); 1169 if (!wait_for_completion_timeout(&adapter->init_done, 1170 timeout)) { 1171 netdev_warn(netdev, 1172 "Capabilities query timed out\n"); 1173 return -1; 1174 } 1175 1176 rc = init_sub_crqs(adapter); 1177 if (rc) { 1178 netdev_warn(netdev, 1179 "SCRQ initialization failed\n"); 1180 return -1; 1181 } 1182 1183 rc = init_sub_crq_irqs(adapter); 1184 if (rc) { 1185 netdev_warn(netdev, 1186 "SCRQ irq initialization failed\n"); 1187 return -1; 1188 } 1189 } else if (adapter->init_done_rc) { 1190 netdev_warn(netdev, "Adapter login failed\n"); 1191 return -1; 1192 } 1193 } while (retry); 1194 1195 __ibmvnic_set_mac(netdev, adapter->mac_addr); 1196 1197 netdev_dbg(netdev, "[S:%s] Login succeeded\n", adapter_state_to_string(adapter->state)); 1198 return 0; 1199 } 1200 1201 static void release_login_buffer(struct ibmvnic_adapter *adapter) 1202 { 1203 kfree(adapter->login_buf); 1204 adapter->login_buf = NULL; 1205 } 1206 1207 static void release_login_rsp_buffer(struct ibmvnic_adapter *adapter) 1208 { 1209 kfree(adapter->login_rsp_buf); 1210 adapter->login_rsp_buf = NULL; 1211 } 1212 1213 static void release_resources(struct ibmvnic_adapter *adapter) 1214 { 1215 release_vpd_data(adapter); 1216 1217 release_napi(adapter); 1218 release_login_buffer(adapter); 1219 release_login_rsp_buffer(adapter); 1220 } 1221 1222 static int set_link_state(struct ibmvnic_adapter *adapter, u8 link_state) 1223 { 1224 struct net_device *netdev = adapter->netdev; 1225 unsigned long timeout = msecs_to_jiffies(20000); 1226 union ibmvnic_crq crq; 1227 bool resend; 1228 int rc; 1229 1230 netdev_dbg(netdev, "setting link state %d\n", link_state); 1231 1232 memset(&crq, 0, sizeof(crq)); 1233 crq.logical_link_state.first = IBMVNIC_CRQ_CMD; 1234 crq.logical_link_state.cmd = LOGICAL_LINK_STATE; 1235 crq.logical_link_state.link_state = link_state; 1236 1237 do { 1238 resend = false; 1239 1240 reinit_completion(&adapter->init_done); 1241 rc = ibmvnic_send_crq(adapter, &crq); 1242 if (rc) { 1243 netdev_err(netdev, "Failed to set link state\n"); 1244 return rc; 1245 } 1246 1247 if (!wait_for_completion_timeout(&adapter->init_done, 1248 timeout)) { 1249 netdev_err(netdev, "timeout setting link state\n"); 1250 return -1; 1251 } 1252 1253 if (adapter->init_done_rc == PARTIALSUCCESS) { 1254 /* Partuial success, delay and re-send */ 1255 mdelay(1000); 1256 resend = true; 1257 } else if (adapter->init_done_rc) { 1258 netdev_warn(netdev, "Unable to set link state, rc=%d\n", 1259 adapter->init_done_rc); 1260 return adapter->init_done_rc; 1261 } 1262 } while (resend); 1263 1264 return 0; 1265 } 1266 1267 static int set_real_num_queues(struct net_device *netdev) 1268 { 1269 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 1270 int rc; 1271 1272 netdev_dbg(netdev, "Setting real tx/rx queues (%llx/%llx)\n", 1273 adapter->req_tx_queues, adapter->req_rx_queues); 1274 1275 rc = netif_set_real_num_tx_queues(netdev, adapter->req_tx_queues); 1276 if (rc) { 1277 netdev_err(netdev, "failed to set the number of tx queues\n"); 1278 return rc; 1279 } 1280 1281 rc = netif_set_real_num_rx_queues(netdev, adapter->req_rx_queues); 1282 if (rc) 1283 netdev_err(netdev, "failed to set the number of rx queues\n"); 1284 1285 return rc; 1286 } 1287 1288 static int ibmvnic_get_vpd(struct ibmvnic_adapter *adapter) 1289 { 1290 struct device *dev = &adapter->vdev->dev; 1291 union ibmvnic_crq crq; 1292 int len = 0; 1293 int rc; 1294 1295 if (adapter->vpd->buff) 1296 len = adapter->vpd->len; 1297 1298 mutex_lock(&adapter->fw_lock); 1299 adapter->fw_done_rc = 0; 1300 reinit_completion(&adapter->fw_done); 1301 1302 crq.get_vpd_size.first = IBMVNIC_CRQ_CMD; 1303 crq.get_vpd_size.cmd = GET_VPD_SIZE; 1304 rc = ibmvnic_send_crq(adapter, &crq); 1305 if (rc) { 1306 mutex_unlock(&adapter->fw_lock); 1307 return rc; 1308 } 1309 1310 rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000); 1311 if (rc) { 1312 dev_err(dev, "Could not retrieve VPD size, rc = %d\n", rc); 1313 mutex_unlock(&adapter->fw_lock); 1314 return rc; 1315 } 1316 mutex_unlock(&adapter->fw_lock); 1317 1318 if (!adapter->vpd->len) 1319 return -ENODATA; 1320 1321 if (!adapter->vpd->buff) 1322 adapter->vpd->buff = kzalloc(adapter->vpd->len, GFP_KERNEL); 1323 else if (adapter->vpd->len != len) 1324 adapter->vpd->buff = 1325 krealloc(adapter->vpd->buff, 1326 adapter->vpd->len, GFP_KERNEL); 1327 1328 if (!adapter->vpd->buff) { 1329 dev_err(dev, "Could allocate VPD buffer\n"); 1330 return -ENOMEM; 1331 } 1332 1333 adapter->vpd->dma_addr = 1334 dma_map_single(dev, adapter->vpd->buff, adapter->vpd->len, 1335 DMA_FROM_DEVICE); 1336 if (dma_mapping_error(dev, adapter->vpd->dma_addr)) { 1337 dev_err(dev, "Could not map VPD buffer\n"); 1338 kfree(adapter->vpd->buff); 1339 adapter->vpd->buff = NULL; 1340 return -ENOMEM; 1341 } 1342 1343 mutex_lock(&adapter->fw_lock); 1344 adapter->fw_done_rc = 0; 1345 reinit_completion(&adapter->fw_done); 1346 1347 crq.get_vpd.first = IBMVNIC_CRQ_CMD; 1348 crq.get_vpd.cmd = GET_VPD; 1349 crq.get_vpd.ioba = cpu_to_be32(adapter->vpd->dma_addr); 1350 crq.get_vpd.len = cpu_to_be32((u32)adapter->vpd->len); 1351 rc = ibmvnic_send_crq(adapter, &crq); 1352 if (rc) { 1353 kfree(adapter->vpd->buff); 1354 adapter->vpd->buff = NULL; 1355 mutex_unlock(&adapter->fw_lock); 1356 return rc; 1357 } 1358 1359 rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000); 1360 if (rc) { 1361 dev_err(dev, "Unable to retrieve VPD, rc = %d\n", rc); 1362 kfree(adapter->vpd->buff); 1363 adapter->vpd->buff = NULL; 1364 mutex_unlock(&adapter->fw_lock); 1365 return rc; 1366 } 1367 1368 mutex_unlock(&adapter->fw_lock); 1369 return 0; 1370 } 1371 1372 static int init_resources(struct ibmvnic_adapter *adapter) 1373 { 1374 struct net_device *netdev = adapter->netdev; 1375 int rc; 1376 1377 rc = set_real_num_queues(netdev); 1378 if (rc) 1379 return rc; 1380 1381 adapter->vpd = kzalloc(sizeof(*adapter->vpd), GFP_KERNEL); 1382 if (!adapter->vpd) 1383 return -ENOMEM; 1384 1385 /* Vital Product Data (VPD) */ 1386 rc = ibmvnic_get_vpd(adapter); 1387 if (rc) { 1388 netdev_err(netdev, "failed to initialize Vital Product Data (VPD)\n"); 1389 return rc; 1390 } 1391 1392 rc = init_napi(adapter); 1393 if (rc) 1394 return rc; 1395 1396 send_query_map(adapter); 1397 1398 rc = init_rx_pools(netdev); 1399 if (rc) 1400 return rc; 1401 1402 rc = init_tx_pools(netdev); 1403 return rc; 1404 } 1405 1406 static int __ibmvnic_open(struct net_device *netdev) 1407 { 1408 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 1409 enum vnic_state prev_state = adapter->state; 1410 int i, rc; 1411 1412 adapter->state = VNIC_OPENING; 1413 replenish_pools(adapter); 1414 ibmvnic_napi_enable(adapter); 1415 1416 /* We're ready to receive frames, enable the sub-crq interrupts and 1417 * set the logical link state to up 1418 */ 1419 for (i = 0; i < adapter->req_rx_queues; i++) { 1420 netdev_dbg(netdev, "Enabling rx_scrq[%d] irq\n", i); 1421 if (prev_state == VNIC_CLOSED) 1422 enable_irq(adapter->rx_scrq[i]->irq); 1423 enable_scrq_irq(adapter, adapter->rx_scrq[i]); 1424 } 1425 1426 for (i = 0; i < adapter->req_tx_queues; i++) { 1427 netdev_dbg(netdev, "Enabling tx_scrq[%d] irq\n", i); 1428 if (prev_state == VNIC_CLOSED) 1429 enable_irq(adapter->tx_scrq[i]->irq); 1430 enable_scrq_irq(adapter, adapter->tx_scrq[i]); 1431 netdev_tx_reset_queue(netdev_get_tx_queue(netdev, i)); 1432 } 1433 1434 rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_UP); 1435 if (rc) { 1436 ibmvnic_napi_disable(adapter); 1437 release_resources(adapter); 1438 return rc; 1439 } 1440 1441 netif_tx_start_all_queues(netdev); 1442 1443 if (prev_state == VNIC_CLOSED) { 1444 for (i = 0; i < adapter->req_rx_queues; i++) 1445 napi_schedule(&adapter->napi[i]); 1446 } 1447 1448 adapter->state = VNIC_OPEN; 1449 return rc; 1450 } 1451 1452 static int ibmvnic_open(struct net_device *netdev) 1453 { 1454 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 1455 int rc; 1456 1457 ASSERT_RTNL(); 1458 1459 /* If device failover is pending or we are about to reset, just set 1460 * device state and return. Device operation will be handled by reset 1461 * routine. 1462 * 1463 * It should be safe to overwrite the adapter->state here. Since 1464 * we hold the rtnl, either the reset has not actually started or 1465 * the rtnl got dropped during the set_link_state() in do_reset(). 1466 * In the former case, no one else is changing the state (again we 1467 * have the rtnl) and in the latter case, do_reset() will detect and 1468 * honor our setting below. 1469 */ 1470 if (adapter->failover_pending || (test_bit(0, &adapter->resetting))) { 1471 netdev_dbg(netdev, "[S:%s FOP:%d] Resetting, deferring open\n", 1472 adapter_state_to_string(adapter->state), 1473 adapter->failover_pending); 1474 adapter->state = VNIC_OPEN; 1475 rc = 0; 1476 goto out; 1477 } 1478 1479 if (adapter->state != VNIC_CLOSED) { 1480 rc = ibmvnic_login(netdev); 1481 if (rc) 1482 goto out; 1483 1484 rc = init_resources(adapter); 1485 if (rc) { 1486 netdev_err(netdev, "failed to initialize resources\n"); 1487 release_resources(adapter); 1488 release_rx_pools(adapter); 1489 release_tx_pools(adapter); 1490 goto out; 1491 } 1492 } 1493 1494 rc = __ibmvnic_open(netdev); 1495 1496 out: 1497 /* If open failed and there is a pending failover or in-progress reset, 1498 * set device state and return. Device operation will be handled by 1499 * reset routine. See also comments above regarding rtnl. 1500 */ 1501 if (rc && 1502 (adapter->failover_pending || (test_bit(0, &adapter->resetting)))) { 1503 adapter->state = VNIC_OPEN; 1504 rc = 0; 1505 } 1506 return rc; 1507 } 1508 1509 static void clean_rx_pools(struct ibmvnic_adapter *adapter) 1510 { 1511 struct ibmvnic_rx_pool *rx_pool; 1512 struct ibmvnic_rx_buff *rx_buff; 1513 u64 rx_entries; 1514 int rx_scrqs; 1515 int i, j; 1516 1517 if (!adapter->rx_pool) 1518 return; 1519 1520 rx_scrqs = adapter->num_active_rx_pools; 1521 rx_entries = adapter->req_rx_add_entries_per_subcrq; 1522 1523 /* Free any remaining skbs in the rx buffer pools */ 1524 for (i = 0; i < rx_scrqs; i++) { 1525 rx_pool = &adapter->rx_pool[i]; 1526 if (!rx_pool || !rx_pool->rx_buff) 1527 continue; 1528 1529 netdev_dbg(adapter->netdev, "Cleaning rx_pool[%d]\n", i); 1530 for (j = 0; j < rx_entries; j++) { 1531 rx_buff = &rx_pool->rx_buff[j]; 1532 if (rx_buff && rx_buff->skb) { 1533 dev_kfree_skb_any(rx_buff->skb); 1534 rx_buff->skb = NULL; 1535 } 1536 } 1537 } 1538 } 1539 1540 static void clean_one_tx_pool(struct ibmvnic_adapter *adapter, 1541 struct ibmvnic_tx_pool *tx_pool) 1542 { 1543 struct ibmvnic_tx_buff *tx_buff; 1544 u64 tx_entries; 1545 int i; 1546 1547 if (!tx_pool || !tx_pool->tx_buff) 1548 return; 1549 1550 tx_entries = tx_pool->num_buffers; 1551 1552 for (i = 0; i < tx_entries; i++) { 1553 tx_buff = &tx_pool->tx_buff[i]; 1554 if (tx_buff && tx_buff->skb) { 1555 dev_kfree_skb_any(tx_buff->skb); 1556 tx_buff->skb = NULL; 1557 } 1558 } 1559 } 1560 1561 static void clean_tx_pools(struct ibmvnic_adapter *adapter) 1562 { 1563 int tx_scrqs; 1564 int i; 1565 1566 if (!adapter->tx_pool || !adapter->tso_pool) 1567 return; 1568 1569 tx_scrqs = adapter->num_active_tx_pools; 1570 1571 /* Free any remaining skbs in the tx buffer pools */ 1572 for (i = 0; i < tx_scrqs; i++) { 1573 netdev_dbg(adapter->netdev, "Cleaning tx_pool[%d]\n", i); 1574 clean_one_tx_pool(adapter, &adapter->tx_pool[i]); 1575 clean_one_tx_pool(adapter, &adapter->tso_pool[i]); 1576 } 1577 } 1578 1579 static void ibmvnic_disable_irqs(struct ibmvnic_adapter *adapter) 1580 { 1581 struct net_device *netdev = adapter->netdev; 1582 int i; 1583 1584 if (adapter->tx_scrq) { 1585 for (i = 0; i < adapter->req_tx_queues; i++) 1586 if (adapter->tx_scrq[i]->irq) { 1587 netdev_dbg(netdev, 1588 "Disabling tx_scrq[%d] irq\n", i); 1589 disable_scrq_irq(adapter, adapter->tx_scrq[i]); 1590 disable_irq(adapter->tx_scrq[i]->irq); 1591 } 1592 } 1593 1594 if (adapter->rx_scrq) { 1595 for (i = 0; i < adapter->req_rx_queues; i++) { 1596 if (adapter->rx_scrq[i]->irq) { 1597 netdev_dbg(netdev, 1598 "Disabling rx_scrq[%d] irq\n", i); 1599 disable_scrq_irq(adapter, adapter->rx_scrq[i]); 1600 disable_irq(adapter->rx_scrq[i]->irq); 1601 } 1602 } 1603 } 1604 } 1605 1606 static void ibmvnic_cleanup(struct net_device *netdev) 1607 { 1608 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 1609 1610 /* ensure that transmissions are stopped if called by do_reset */ 1611 if (test_bit(0, &adapter->resetting)) 1612 netif_tx_disable(netdev); 1613 else 1614 netif_tx_stop_all_queues(netdev); 1615 1616 ibmvnic_napi_disable(adapter); 1617 ibmvnic_disable_irqs(adapter); 1618 } 1619 1620 static int __ibmvnic_close(struct net_device *netdev) 1621 { 1622 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 1623 int rc = 0; 1624 1625 adapter->state = VNIC_CLOSING; 1626 rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN); 1627 adapter->state = VNIC_CLOSED; 1628 return rc; 1629 } 1630 1631 static int ibmvnic_close(struct net_device *netdev) 1632 { 1633 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 1634 int rc; 1635 1636 netdev_dbg(netdev, "[S:%s FOP:%d FRR:%d] Closing\n", 1637 adapter_state_to_string(adapter->state), 1638 adapter->failover_pending, 1639 adapter->force_reset_recovery); 1640 1641 /* If device failover is pending, just set device state and return. 1642 * Device operation will be handled by reset routine. 1643 */ 1644 if (adapter->failover_pending) { 1645 adapter->state = VNIC_CLOSED; 1646 return 0; 1647 } 1648 1649 rc = __ibmvnic_close(netdev); 1650 ibmvnic_cleanup(netdev); 1651 clean_rx_pools(adapter); 1652 clean_tx_pools(adapter); 1653 1654 return rc; 1655 } 1656 1657 /** 1658 * build_hdr_data - creates L2/L3/L4 header data buffer 1659 * @hdr_field: bitfield determining needed headers 1660 * @skb: socket buffer 1661 * @hdr_len: array of header lengths 1662 * @hdr_data: buffer to write the header to 1663 * 1664 * Reads hdr_field to determine which headers are needed by firmware. 1665 * Builds a buffer containing these headers. Saves individual header 1666 * lengths and total buffer length to be used to build descriptors. 1667 */ 1668 static int build_hdr_data(u8 hdr_field, struct sk_buff *skb, 1669 int *hdr_len, u8 *hdr_data) 1670 { 1671 int len = 0; 1672 u8 *hdr; 1673 1674 if (skb_vlan_tagged(skb) && !skb_vlan_tag_present(skb)) 1675 hdr_len[0] = sizeof(struct vlan_ethhdr); 1676 else 1677 hdr_len[0] = sizeof(struct ethhdr); 1678 1679 if (skb->protocol == htons(ETH_P_IP)) { 1680 hdr_len[1] = ip_hdr(skb)->ihl * 4; 1681 if (ip_hdr(skb)->protocol == IPPROTO_TCP) 1682 hdr_len[2] = tcp_hdrlen(skb); 1683 else if (ip_hdr(skb)->protocol == IPPROTO_UDP) 1684 hdr_len[2] = sizeof(struct udphdr); 1685 } else if (skb->protocol == htons(ETH_P_IPV6)) { 1686 hdr_len[1] = sizeof(struct ipv6hdr); 1687 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP) 1688 hdr_len[2] = tcp_hdrlen(skb); 1689 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP) 1690 hdr_len[2] = sizeof(struct udphdr); 1691 } else if (skb->protocol == htons(ETH_P_ARP)) { 1692 hdr_len[1] = arp_hdr_len(skb->dev); 1693 hdr_len[2] = 0; 1694 } 1695 1696 memset(hdr_data, 0, 120); 1697 if ((hdr_field >> 6) & 1) { 1698 hdr = skb_mac_header(skb); 1699 memcpy(hdr_data, hdr, hdr_len[0]); 1700 len += hdr_len[0]; 1701 } 1702 1703 if ((hdr_field >> 5) & 1) { 1704 hdr = skb_network_header(skb); 1705 memcpy(hdr_data + len, hdr, hdr_len[1]); 1706 len += hdr_len[1]; 1707 } 1708 1709 if ((hdr_field >> 4) & 1) { 1710 hdr = skb_transport_header(skb); 1711 memcpy(hdr_data + len, hdr, hdr_len[2]); 1712 len += hdr_len[2]; 1713 } 1714 return len; 1715 } 1716 1717 /** 1718 * create_hdr_descs - create header and header extension descriptors 1719 * @hdr_field: bitfield determining needed headers 1720 * @hdr_data: buffer containing header data 1721 * @len: length of data buffer 1722 * @hdr_len: array of individual header lengths 1723 * @scrq_arr: descriptor array 1724 * 1725 * Creates header and, if needed, header extension descriptors and 1726 * places them in a descriptor array, scrq_arr 1727 */ 1728 1729 static int create_hdr_descs(u8 hdr_field, u8 *hdr_data, int len, int *hdr_len, 1730 union sub_crq *scrq_arr) 1731 { 1732 union sub_crq hdr_desc; 1733 int tmp_len = len; 1734 int num_descs = 0; 1735 u8 *data, *cur; 1736 int tmp; 1737 1738 while (tmp_len > 0) { 1739 cur = hdr_data + len - tmp_len; 1740 1741 memset(&hdr_desc, 0, sizeof(hdr_desc)); 1742 if (cur != hdr_data) { 1743 data = hdr_desc.hdr_ext.data; 1744 tmp = tmp_len > 29 ? 29 : tmp_len; 1745 hdr_desc.hdr_ext.first = IBMVNIC_CRQ_CMD; 1746 hdr_desc.hdr_ext.type = IBMVNIC_HDR_EXT_DESC; 1747 hdr_desc.hdr_ext.len = tmp; 1748 } else { 1749 data = hdr_desc.hdr.data; 1750 tmp = tmp_len > 24 ? 24 : tmp_len; 1751 hdr_desc.hdr.first = IBMVNIC_CRQ_CMD; 1752 hdr_desc.hdr.type = IBMVNIC_HDR_DESC; 1753 hdr_desc.hdr.len = tmp; 1754 hdr_desc.hdr.l2_len = (u8)hdr_len[0]; 1755 hdr_desc.hdr.l3_len = cpu_to_be16((u16)hdr_len[1]); 1756 hdr_desc.hdr.l4_len = (u8)hdr_len[2]; 1757 hdr_desc.hdr.flag = hdr_field << 1; 1758 } 1759 memcpy(data, cur, tmp); 1760 tmp_len -= tmp; 1761 *scrq_arr = hdr_desc; 1762 scrq_arr++; 1763 num_descs++; 1764 } 1765 1766 return num_descs; 1767 } 1768 1769 /** 1770 * build_hdr_descs_arr - build a header descriptor array 1771 * @skb: tx socket buffer 1772 * @indir_arr: indirect array 1773 * @num_entries: number of descriptors to be sent 1774 * @hdr_field: bit field determining which headers will be sent 1775 * 1776 * This function will build a TX descriptor array with applicable 1777 * L2/L3/L4 packet header descriptors to be sent by send_subcrq_indirect. 1778 */ 1779 1780 static void build_hdr_descs_arr(struct sk_buff *skb, 1781 union sub_crq *indir_arr, 1782 int *num_entries, u8 hdr_field) 1783 { 1784 int hdr_len[3] = {0, 0, 0}; 1785 u8 hdr_data[140] = {0}; 1786 int tot_len; 1787 1788 tot_len = build_hdr_data(hdr_field, skb, hdr_len, 1789 hdr_data); 1790 *num_entries += create_hdr_descs(hdr_field, hdr_data, tot_len, hdr_len, 1791 indir_arr + 1); 1792 } 1793 1794 static int ibmvnic_xmit_workarounds(struct sk_buff *skb, 1795 struct net_device *netdev) 1796 { 1797 /* For some backing devices, mishandling of small packets 1798 * can result in a loss of connection or TX stall. Device 1799 * architects recommend that no packet should be smaller 1800 * than the minimum MTU value provided to the driver, so 1801 * pad any packets to that length 1802 */ 1803 if (skb->len < netdev->min_mtu) 1804 return skb_put_padto(skb, netdev->min_mtu); 1805 1806 return 0; 1807 } 1808 1809 static void ibmvnic_tx_scrq_clean_buffer(struct ibmvnic_adapter *adapter, 1810 struct ibmvnic_sub_crq_queue *tx_scrq) 1811 { 1812 struct ibmvnic_ind_xmit_queue *ind_bufp; 1813 struct ibmvnic_tx_buff *tx_buff; 1814 struct ibmvnic_tx_pool *tx_pool; 1815 union sub_crq tx_scrq_entry; 1816 int queue_num; 1817 int entries; 1818 int index; 1819 int i; 1820 1821 ind_bufp = &tx_scrq->ind_buf; 1822 entries = (u64)ind_bufp->index; 1823 queue_num = tx_scrq->pool_index; 1824 1825 for (i = entries - 1; i >= 0; --i) { 1826 tx_scrq_entry = ind_bufp->indir_arr[i]; 1827 if (tx_scrq_entry.v1.type != IBMVNIC_TX_DESC) 1828 continue; 1829 index = be32_to_cpu(tx_scrq_entry.v1.correlator); 1830 if (index & IBMVNIC_TSO_POOL_MASK) { 1831 tx_pool = &adapter->tso_pool[queue_num]; 1832 index &= ~IBMVNIC_TSO_POOL_MASK; 1833 } else { 1834 tx_pool = &adapter->tx_pool[queue_num]; 1835 } 1836 tx_pool->free_map[tx_pool->consumer_index] = index; 1837 tx_pool->consumer_index = tx_pool->consumer_index == 0 ? 1838 tx_pool->num_buffers - 1 : 1839 tx_pool->consumer_index - 1; 1840 tx_buff = &tx_pool->tx_buff[index]; 1841 adapter->netdev->stats.tx_packets--; 1842 adapter->netdev->stats.tx_bytes -= tx_buff->skb->len; 1843 adapter->tx_stats_buffers[queue_num].packets--; 1844 adapter->tx_stats_buffers[queue_num].bytes -= 1845 tx_buff->skb->len; 1846 dev_kfree_skb_any(tx_buff->skb); 1847 tx_buff->skb = NULL; 1848 adapter->netdev->stats.tx_dropped++; 1849 } 1850 ind_bufp->index = 0; 1851 if (atomic_sub_return(entries, &tx_scrq->used) <= 1852 (adapter->req_tx_entries_per_subcrq / 2) && 1853 __netif_subqueue_stopped(adapter->netdev, queue_num) && 1854 !test_bit(0, &adapter->resetting)) { 1855 netif_wake_subqueue(adapter->netdev, queue_num); 1856 netdev_dbg(adapter->netdev, "Started queue %d\n", 1857 queue_num); 1858 } 1859 } 1860 1861 static int ibmvnic_tx_scrq_flush(struct ibmvnic_adapter *adapter, 1862 struct ibmvnic_sub_crq_queue *tx_scrq) 1863 { 1864 struct ibmvnic_ind_xmit_queue *ind_bufp; 1865 u64 dma_addr; 1866 u64 entries; 1867 u64 handle; 1868 int rc; 1869 1870 ind_bufp = &tx_scrq->ind_buf; 1871 dma_addr = (u64)ind_bufp->indir_dma; 1872 entries = (u64)ind_bufp->index; 1873 handle = tx_scrq->handle; 1874 1875 if (!entries) 1876 return 0; 1877 rc = send_subcrq_indirect(adapter, handle, dma_addr, entries); 1878 if (rc) 1879 ibmvnic_tx_scrq_clean_buffer(adapter, tx_scrq); 1880 else 1881 ind_bufp->index = 0; 1882 return 0; 1883 } 1884 1885 static netdev_tx_t ibmvnic_xmit(struct sk_buff *skb, struct net_device *netdev) 1886 { 1887 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 1888 int queue_num = skb_get_queue_mapping(skb); 1889 u8 *hdrs = (u8 *)&adapter->tx_rx_desc_req; 1890 struct device *dev = &adapter->vdev->dev; 1891 struct ibmvnic_ind_xmit_queue *ind_bufp; 1892 struct ibmvnic_tx_buff *tx_buff = NULL; 1893 struct ibmvnic_sub_crq_queue *tx_scrq; 1894 struct ibmvnic_tx_pool *tx_pool; 1895 unsigned int tx_send_failed = 0; 1896 netdev_tx_t ret = NETDEV_TX_OK; 1897 unsigned int tx_map_failed = 0; 1898 union sub_crq indir_arr[16]; 1899 unsigned int tx_dropped = 0; 1900 unsigned int tx_packets = 0; 1901 unsigned int tx_bytes = 0; 1902 dma_addr_t data_dma_addr; 1903 struct netdev_queue *txq; 1904 unsigned long lpar_rc; 1905 union sub_crq tx_crq; 1906 unsigned int offset; 1907 int num_entries = 1; 1908 unsigned char *dst; 1909 int index = 0; 1910 u8 proto = 0; 1911 1912 tx_scrq = adapter->tx_scrq[queue_num]; 1913 txq = netdev_get_tx_queue(netdev, queue_num); 1914 ind_bufp = &tx_scrq->ind_buf; 1915 1916 if (test_bit(0, &adapter->resetting)) { 1917 if (!netif_subqueue_stopped(netdev, skb)) 1918 netif_stop_subqueue(netdev, queue_num); 1919 dev_kfree_skb_any(skb); 1920 1921 tx_send_failed++; 1922 tx_dropped++; 1923 ret = NETDEV_TX_OK; 1924 goto out; 1925 } 1926 1927 if (ibmvnic_xmit_workarounds(skb, netdev)) { 1928 tx_dropped++; 1929 tx_send_failed++; 1930 ret = NETDEV_TX_OK; 1931 ibmvnic_tx_scrq_flush(adapter, tx_scrq); 1932 goto out; 1933 } 1934 if (skb_is_gso(skb)) 1935 tx_pool = &adapter->tso_pool[queue_num]; 1936 else 1937 tx_pool = &adapter->tx_pool[queue_num]; 1938 1939 index = tx_pool->free_map[tx_pool->consumer_index]; 1940 1941 if (index == IBMVNIC_INVALID_MAP) { 1942 dev_kfree_skb_any(skb); 1943 tx_send_failed++; 1944 tx_dropped++; 1945 ibmvnic_tx_scrq_flush(adapter, tx_scrq); 1946 ret = NETDEV_TX_OK; 1947 goto out; 1948 } 1949 1950 tx_pool->free_map[tx_pool->consumer_index] = IBMVNIC_INVALID_MAP; 1951 1952 offset = index * tx_pool->buf_size; 1953 dst = tx_pool->long_term_buff.buff + offset; 1954 memset(dst, 0, tx_pool->buf_size); 1955 data_dma_addr = tx_pool->long_term_buff.addr + offset; 1956 1957 if (skb_shinfo(skb)->nr_frags) { 1958 int cur, i; 1959 1960 /* Copy the head */ 1961 skb_copy_from_linear_data(skb, dst, skb_headlen(skb)); 1962 cur = skb_headlen(skb); 1963 1964 /* Copy the frags */ 1965 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 1966 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 1967 1968 memcpy(dst + cur, skb_frag_address(frag), 1969 skb_frag_size(frag)); 1970 cur += skb_frag_size(frag); 1971 } 1972 } else { 1973 skb_copy_from_linear_data(skb, dst, skb->len); 1974 } 1975 1976 /* post changes to long_term_buff *dst before VIOS accessing it */ 1977 dma_wmb(); 1978 1979 tx_pool->consumer_index = 1980 (tx_pool->consumer_index + 1) % tx_pool->num_buffers; 1981 1982 tx_buff = &tx_pool->tx_buff[index]; 1983 tx_buff->skb = skb; 1984 tx_buff->index = index; 1985 tx_buff->pool_index = queue_num; 1986 1987 memset(&tx_crq, 0, sizeof(tx_crq)); 1988 tx_crq.v1.first = IBMVNIC_CRQ_CMD; 1989 tx_crq.v1.type = IBMVNIC_TX_DESC; 1990 tx_crq.v1.n_crq_elem = 1; 1991 tx_crq.v1.n_sge = 1; 1992 tx_crq.v1.flags1 = IBMVNIC_TX_COMP_NEEDED; 1993 1994 if (skb_is_gso(skb)) 1995 tx_crq.v1.correlator = 1996 cpu_to_be32(index | IBMVNIC_TSO_POOL_MASK); 1997 else 1998 tx_crq.v1.correlator = cpu_to_be32(index); 1999 tx_crq.v1.dma_reg = cpu_to_be16(tx_pool->long_term_buff.map_id); 2000 tx_crq.v1.sge_len = cpu_to_be32(skb->len); 2001 tx_crq.v1.ioba = cpu_to_be64(data_dma_addr); 2002 2003 if (adapter->vlan_header_insertion && skb_vlan_tag_present(skb)) { 2004 tx_crq.v1.flags2 |= IBMVNIC_TX_VLAN_INSERT; 2005 tx_crq.v1.vlan_id = cpu_to_be16(skb->vlan_tci); 2006 } 2007 2008 if (skb->protocol == htons(ETH_P_IP)) { 2009 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV4; 2010 proto = ip_hdr(skb)->protocol; 2011 } else if (skb->protocol == htons(ETH_P_IPV6)) { 2012 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV6; 2013 proto = ipv6_hdr(skb)->nexthdr; 2014 } 2015 2016 if (proto == IPPROTO_TCP) 2017 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_TCP; 2018 else if (proto == IPPROTO_UDP) 2019 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_UDP; 2020 2021 if (skb->ip_summed == CHECKSUM_PARTIAL) { 2022 tx_crq.v1.flags1 |= IBMVNIC_TX_CHKSUM_OFFLOAD; 2023 hdrs += 2; 2024 } 2025 if (skb_is_gso(skb)) { 2026 tx_crq.v1.flags1 |= IBMVNIC_TX_LSO; 2027 tx_crq.v1.mss = cpu_to_be16(skb_shinfo(skb)->gso_size); 2028 hdrs += 2; 2029 } 2030 2031 if ((*hdrs >> 7) & 1) 2032 build_hdr_descs_arr(skb, indir_arr, &num_entries, *hdrs); 2033 2034 tx_crq.v1.n_crq_elem = num_entries; 2035 tx_buff->num_entries = num_entries; 2036 /* flush buffer if current entry can not fit */ 2037 if (num_entries + ind_bufp->index > IBMVNIC_MAX_IND_DESCS) { 2038 lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq); 2039 if (lpar_rc != H_SUCCESS) 2040 goto tx_flush_err; 2041 } 2042 2043 indir_arr[0] = tx_crq; 2044 memcpy(&ind_bufp->indir_arr[ind_bufp->index], &indir_arr[0], 2045 num_entries * sizeof(struct ibmvnic_generic_scrq)); 2046 ind_bufp->index += num_entries; 2047 if (__netdev_tx_sent_queue(txq, skb->len, 2048 netdev_xmit_more() && 2049 ind_bufp->index < IBMVNIC_MAX_IND_DESCS)) { 2050 lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq); 2051 if (lpar_rc != H_SUCCESS) 2052 goto tx_err; 2053 } 2054 2055 if (atomic_add_return(num_entries, &tx_scrq->used) 2056 >= adapter->req_tx_entries_per_subcrq) { 2057 netdev_dbg(netdev, "Stopping queue %d\n", queue_num); 2058 netif_stop_subqueue(netdev, queue_num); 2059 } 2060 2061 tx_packets++; 2062 tx_bytes += skb->len; 2063 txq->trans_start = jiffies; 2064 ret = NETDEV_TX_OK; 2065 goto out; 2066 2067 tx_flush_err: 2068 dev_kfree_skb_any(skb); 2069 tx_buff->skb = NULL; 2070 tx_pool->consumer_index = tx_pool->consumer_index == 0 ? 2071 tx_pool->num_buffers - 1 : 2072 tx_pool->consumer_index - 1; 2073 tx_dropped++; 2074 tx_err: 2075 if (lpar_rc != H_CLOSED && lpar_rc != H_PARAMETER) 2076 dev_err_ratelimited(dev, "tx: send failed\n"); 2077 2078 if (lpar_rc == H_CLOSED || adapter->failover_pending) { 2079 /* Disable TX and report carrier off if queue is closed 2080 * or pending failover. 2081 * Firmware guarantees that a signal will be sent to the 2082 * driver, triggering a reset or some other action. 2083 */ 2084 netif_tx_stop_all_queues(netdev); 2085 netif_carrier_off(netdev); 2086 } 2087 out: 2088 netdev->stats.tx_dropped += tx_dropped; 2089 netdev->stats.tx_bytes += tx_bytes; 2090 netdev->stats.tx_packets += tx_packets; 2091 adapter->tx_send_failed += tx_send_failed; 2092 adapter->tx_map_failed += tx_map_failed; 2093 adapter->tx_stats_buffers[queue_num].packets += tx_packets; 2094 adapter->tx_stats_buffers[queue_num].bytes += tx_bytes; 2095 adapter->tx_stats_buffers[queue_num].dropped_packets += tx_dropped; 2096 2097 return ret; 2098 } 2099 2100 static void ibmvnic_set_multi(struct net_device *netdev) 2101 { 2102 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 2103 struct netdev_hw_addr *ha; 2104 union ibmvnic_crq crq; 2105 2106 memset(&crq, 0, sizeof(crq)); 2107 crq.request_capability.first = IBMVNIC_CRQ_CMD; 2108 crq.request_capability.cmd = REQUEST_CAPABILITY; 2109 2110 if (netdev->flags & IFF_PROMISC) { 2111 if (!adapter->promisc_supported) 2112 return; 2113 } else { 2114 if (netdev->flags & IFF_ALLMULTI) { 2115 /* Accept all multicast */ 2116 memset(&crq, 0, sizeof(crq)); 2117 crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD; 2118 crq.multicast_ctrl.cmd = MULTICAST_CTRL; 2119 crq.multicast_ctrl.flags = IBMVNIC_ENABLE_ALL; 2120 ibmvnic_send_crq(adapter, &crq); 2121 } else if (netdev_mc_empty(netdev)) { 2122 /* Reject all multicast */ 2123 memset(&crq, 0, sizeof(crq)); 2124 crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD; 2125 crq.multicast_ctrl.cmd = MULTICAST_CTRL; 2126 crq.multicast_ctrl.flags = IBMVNIC_DISABLE_ALL; 2127 ibmvnic_send_crq(adapter, &crq); 2128 } else { 2129 /* Accept one or more multicast(s) */ 2130 netdev_for_each_mc_addr(ha, netdev) { 2131 memset(&crq, 0, sizeof(crq)); 2132 crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD; 2133 crq.multicast_ctrl.cmd = MULTICAST_CTRL; 2134 crq.multicast_ctrl.flags = IBMVNIC_ENABLE_MC; 2135 ether_addr_copy(&crq.multicast_ctrl.mac_addr[0], 2136 ha->addr); 2137 ibmvnic_send_crq(adapter, &crq); 2138 } 2139 } 2140 } 2141 } 2142 2143 static int __ibmvnic_set_mac(struct net_device *netdev, u8 *dev_addr) 2144 { 2145 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 2146 union ibmvnic_crq crq; 2147 int rc; 2148 2149 if (!is_valid_ether_addr(dev_addr)) { 2150 rc = -EADDRNOTAVAIL; 2151 goto err; 2152 } 2153 2154 memset(&crq, 0, sizeof(crq)); 2155 crq.change_mac_addr.first = IBMVNIC_CRQ_CMD; 2156 crq.change_mac_addr.cmd = CHANGE_MAC_ADDR; 2157 ether_addr_copy(&crq.change_mac_addr.mac_addr[0], dev_addr); 2158 2159 mutex_lock(&adapter->fw_lock); 2160 adapter->fw_done_rc = 0; 2161 reinit_completion(&adapter->fw_done); 2162 2163 rc = ibmvnic_send_crq(adapter, &crq); 2164 if (rc) { 2165 rc = -EIO; 2166 mutex_unlock(&adapter->fw_lock); 2167 goto err; 2168 } 2169 2170 rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000); 2171 /* netdev->dev_addr is changed in handle_change_mac_rsp function */ 2172 if (rc || adapter->fw_done_rc) { 2173 rc = -EIO; 2174 mutex_unlock(&adapter->fw_lock); 2175 goto err; 2176 } 2177 mutex_unlock(&adapter->fw_lock); 2178 return 0; 2179 err: 2180 ether_addr_copy(adapter->mac_addr, netdev->dev_addr); 2181 return rc; 2182 } 2183 2184 static int ibmvnic_set_mac(struct net_device *netdev, void *p) 2185 { 2186 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 2187 struct sockaddr *addr = p; 2188 int rc; 2189 2190 rc = 0; 2191 if (!is_valid_ether_addr(addr->sa_data)) 2192 return -EADDRNOTAVAIL; 2193 2194 ether_addr_copy(adapter->mac_addr, addr->sa_data); 2195 if (adapter->state != VNIC_PROBED) 2196 rc = __ibmvnic_set_mac(netdev, addr->sa_data); 2197 2198 return rc; 2199 } 2200 2201 static const char *reset_reason_to_string(enum ibmvnic_reset_reason reason) 2202 { 2203 switch (reason) { 2204 case VNIC_RESET_FAILOVER: 2205 return "FAILOVER"; 2206 case VNIC_RESET_MOBILITY: 2207 return "MOBILITY"; 2208 case VNIC_RESET_FATAL: 2209 return "FATAL"; 2210 case VNIC_RESET_NON_FATAL: 2211 return "NON_FATAL"; 2212 case VNIC_RESET_TIMEOUT: 2213 return "TIMEOUT"; 2214 case VNIC_RESET_CHANGE_PARAM: 2215 return "CHANGE_PARAM"; 2216 case VNIC_RESET_PASSIVE_INIT: 2217 return "PASSIVE_INIT"; 2218 } 2219 return "UNKNOWN"; 2220 } 2221 2222 /* 2223 * do_reset returns zero if we are able to keep processing reset events, or 2224 * non-zero if we hit a fatal error and must halt. 2225 */ 2226 static int do_reset(struct ibmvnic_adapter *adapter, 2227 struct ibmvnic_rwi *rwi, u32 reset_state) 2228 { 2229 struct net_device *netdev = adapter->netdev; 2230 u64 old_num_rx_queues, old_num_tx_queues; 2231 u64 old_num_rx_slots, old_num_tx_slots; 2232 int rc; 2233 2234 netdev_dbg(adapter->netdev, 2235 "[S:%s FOP:%d] Reset reason: %s, reset_state: %s\n", 2236 adapter_state_to_string(adapter->state), 2237 adapter->failover_pending, 2238 reset_reason_to_string(rwi->reset_reason), 2239 adapter_state_to_string(reset_state)); 2240 2241 adapter->reset_reason = rwi->reset_reason; 2242 /* requestor of VNIC_RESET_CHANGE_PARAM already has the rtnl lock */ 2243 if (!(adapter->reset_reason == VNIC_RESET_CHANGE_PARAM)) 2244 rtnl_lock(); 2245 2246 /* Now that we have the rtnl lock, clear any pending failover. 2247 * This will ensure ibmvnic_open() has either completed or will 2248 * block until failover is complete. 2249 */ 2250 if (rwi->reset_reason == VNIC_RESET_FAILOVER) 2251 adapter->failover_pending = false; 2252 2253 /* read the state and check (again) after getting rtnl */ 2254 reset_state = adapter->state; 2255 2256 if (reset_state == VNIC_REMOVING || reset_state == VNIC_REMOVED) { 2257 rc = -EBUSY; 2258 goto out; 2259 } 2260 2261 netif_carrier_off(netdev); 2262 2263 old_num_rx_queues = adapter->req_rx_queues; 2264 old_num_tx_queues = adapter->req_tx_queues; 2265 old_num_rx_slots = adapter->req_rx_add_entries_per_subcrq; 2266 old_num_tx_slots = adapter->req_tx_entries_per_subcrq; 2267 2268 ibmvnic_cleanup(netdev); 2269 2270 if (reset_state == VNIC_OPEN && 2271 adapter->reset_reason != VNIC_RESET_MOBILITY && 2272 adapter->reset_reason != VNIC_RESET_FAILOVER) { 2273 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) { 2274 rc = __ibmvnic_close(netdev); 2275 if (rc) 2276 goto out; 2277 } else { 2278 adapter->state = VNIC_CLOSING; 2279 2280 /* Release the RTNL lock before link state change and 2281 * re-acquire after the link state change to allow 2282 * linkwatch_event to grab the RTNL lock and run during 2283 * a reset. 2284 */ 2285 rtnl_unlock(); 2286 rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN); 2287 rtnl_lock(); 2288 if (rc) 2289 goto out; 2290 2291 if (adapter->state == VNIC_OPEN) { 2292 /* When we dropped rtnl, ibmvnic_open() got 2293 * it and noticed that we are resetting and 2294 * set the adapter state to OPEN. Update our 2295 * new "target" state, and resume the reset 2296 * from VNIC_CLOSING state. 2297 */ 2298 netdev_dbg(netdev, 2299 "Open changed state from %s, updating.\n", 2300 adapter_state_to_string(reset_state)); 2301 reset_state = VNIC_OPEN; 2302 adapter->state = VNIC_CLOSING; 2303 } 2304 2305 if (adapter->state != VNIC_CLOSING) { 2306 /* If someone else changed the adapter state 2307 * when we dropped the rtnl, fail the reset 2308 */ 2309 rc = -1; 2310 goto out; 2311 } 2312 adapter->state = VNIC_CLOSED; 2313 } 2314 } 2315 2316 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) { 2317 release_resources(adapter); 2318 release_sub_crqs(adapter, 1); 2319 release_crq_queue(adapter); 2320 } 2321 2322 if (adapter->reset_reason != VNIC_RESET_NON_FATAL) { 2323 /* remove the closed state so when we call open it appears 2324 * we are coming from the probed state. 2325 */ 2326 adapter->state = VNIC_PROBED; 2327 2328 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) { 2329 rc = init_crq_queue(adapter); 2330 } else if (adapter->reset_reason == VNIC_RESET_MOBILITY) { 2331 rc = ibmvnic_reenable_crq_queue(adapter); 2332 release_sub_crqs(adapter, 1); 2333 } else { 2334 rc = ibmvnic_reset_crq(adapter); 2335 if (rc == H_CLOSED || rc == H_SUCCESS) { 2336 rc = vio_enable_interrupts(adapter->vdev); 2337 if (rc) 2338 netdev_err(adapter->netdev, 2339 "Reset failed to enable interrupts. rc=%d\n", 2340 rc); 2341 } 2342 } 2343 2344 if (rc) { 2345 netdev_err(adapter->netdev, 2346 "Reset couldn't initialize crq. rc=%d\n", rc); 2347 goto out; 2348 } 2349 2350 rc = ibmvnic_reset_init(adapter, true); 2351 if (rc) { 2352 rc = IBMVNIC_INIT_FAILED; 2353 goto out; 2354 } 2355 2356 /* If the adapter was in PROBE or DOWN state prior to the reset, 2357 * exit here. 2358 */ 2359 if (reset_state == VNIC_PROBED || reset_state == VNIC_DOWN) { 2360 rc = 0; 2361 goto out; 2362 } 2363 2364 rc = ibmvnic_login(netdev); 2365 if (rc) 2366 goto out; 2367 2368 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) { 2369 rc = init_resources(adapter); 2370 if (rc) 2371 goto out; 2372 } else if (adapter->req_rx_queues != old_num_rx_queues || 2373 adapter->req_tx_queues != old_num_tx_queues || 2374 adapter->req_rx_add_entries_per_subcrq != 2375 old_num_rx_slots || 2376 adapter->req_tx_entries_per_subcrq != 2377 old_num_tx_slots || 2378 !adapter->rx_pool || 2379 !adapter->tso_pool || 2380 !adapter->tx_pool) { 2381 release_napi(adapter); 2382 release_vpd_data(adapter); 2383 2384 rc = init_resources(adapter); 2385 if (rc) 2386 goto out; 2387 2388 } else { 2389 rc = init_tx_pools(netdev); 2390 if (rc) { 2391 netdev_dbg(netdev, 2392 "init tx pools failed (%d)\n", 2393 rc); 2394 goto out; 2395 } 2396 2397 rc = init_rx_pools(netdev); 2398 if (rc) { 2399 netdev_dbg(netdev, 2400 "init rx pools failed (%d)\n", 2401 rc); 2402 goto out; 2403 } 2404 } 2405 ibmvnic_disable_irqs(adapter); 2406 } 2407 adapter->state = VNIC_CLOSED; 2408 2409 if (reset_state == VNIC_CLOSED) { 2410 rc = 0; 2411 goto out; 2412 } 2413 2414 rc = __ibmvnic_open(netdev); 2415 if (rc) { 2416 rc = IBMVNIC_OPEN_FAILED; 2417 goto out; 2418 } 2419 2420 /* refresh device's multicast list */ 2421 ibmvnic_set_multi(netdev); 2422 2423 if (adapter->reset_reason == VNIC_RESET_FAILOVER || 2424 adapter->reset_reason == VNIC_RESET_MOBILITY) 2425 __netdev_notify_peers(netdev); 2426 2427 rc = 0; 2428 2429 out: 2430 /* restore the adapter state if reset failed */ 2431 if (rc) 2432 adapter->state = reset_state; 2433 /* requestor of VNIC_RESET_CHANGE_PARAM should still hold the rtnl lock */ 2434 if (!(adapter->reset_reason == VNIC_RESET_CHANGE_PARAM)) 2435 rtnl_unlock(); 2436 2437 netdev_dbg(adapter->netdev, "[S:%s FOP:%d] Reset done, rc %d\n", 2438 adapter_state_to_string(adapter->state), 2439 adapter->failover_pending, rc); 2440 return rc; 2441 } 2442 2443 static int do_hard_reset(struct ibmvnic_adapter *adapter, 2444 struct ibmvnic_rwi *rwi, u32 reset_state) 2445 { 2446 struct net_device *netdev = adapter->netdev; 2447 int rc; 2448 2449 netdev_dbg(adapter->netdev, "Hard resetting driver (%s)\n", 2450 reset_reason_to_string(rwi->reset_reason)); 2451 2452 /* read the state and check (again) after getting rtnl */ 2453 reset_state = adapter->state; 2454 2455 if (reset_state == VNIC_REMOVING || reset_state == VNIC_REMOVED) { 2456 rc = -EBUSY; 2457 goto out; 2458 } 2459 2460 netif_carrier_off(netdev); 2461 adapter->reset_reason = rwi->reset_reason; 2462 2463 ibmvnic_cleanup(netdev); 2464 release_resources(adapter); 2465 release_sub_crqs(adapter, 0); 2466 release_crq_queue(adapter); 2467 2468 /* remove the closed state so when we call open it appears 2469 * we are coming from the probed state. 2470 */ 2471 adapter->state = VNIC_PROBED; 2472 2473 reinit_completion(&adapter->init_done); 2474 rc = init_crq_queue(adapter); 2475 if (rc) { 2476 netdev_err(adapter->netdev, 2477 "Couldn't initialize crq. rc=%d\n", rc); 2478 goto out; 2479 } 2480 2481 rc = ibmvnic_reset_init(adapter, false); 2482 if (rc) 2483 goto out; 2484 2485 /* If the adapter was in PROBE or DOWN state prior to the reset, 2486 * exit here. 2487 */ 2488 if (reset_state == VNIC_PROBED || reset_state == VNIC_DOWN) 2489 goto out; 2490 2491 rc = ibmvnic_login(netdev); 2492 if (rc) 2493 goto out; 2494 2495 rc = init_resources(adapter); 2496 if (rc) 2497 goto out; 2498 2499 ibmvnic_disable_irqs(adapter); 2500 adapter->state = VNIC_CLOSED; 2501 2502 if (reset_state == VNIC_CLOSED) 2503 goto out; 2504 2505 rc = __ibmvnic_open(netdev); 2506 if (rc) { 2507 rc = IBMVNIC_OPEN_FAILED; 2508 goto out; 2509 } 2510 2511 __netdev_notify_peers(netdev); 2512 out: 2513 /* restore adapter state if reset failed */ 2514 if (rc) 2515 adapter->state = reset_state; 2516 netdev_dbg(adapter->netdev, "[S:%s FOP:%d] Hard reset done, rc %d\n", 2517 adapter_state_to_string(adapter->state), 2518 adapter->failover_pending, rc); 2519 return rc; 2520 } 2521 2522 static struct ibmvnic_rwi *get_next_rwi(struct ibmvnic_adapter *adapter) 2523 { 2524 struct ibmvnic_rwi *rwi; 2525 unsigned long flags; 2526 2527 spin_lock_irqsave(&adapter->rwi_lock, flags); 2528 2529 if (!list_empty(&adapter->rwi_list)) { 2530 rwi = list_first_entry(&adapter->rwi_list, struct ibmvnic_rwi, 2531 list); 2532 list_del(&rwi->list); 2533 } else { 2534 rwi = NULL; 2535 } 2536 2537 spin_unlock_irqrestore(&adapter->rwi_lock, flags); 2538 return rwi; 2539 } 2540 2541 /** 2542 * do_passive_init - complete probing when partner device is detected. 2543 * @adapter: ibmvnic_adapter struct 2544 * 2545 * If the ibmvnic device does not have a partner device to communicate with at boot 2546 * and that partner device comes online at a later time, this function is called 2547 * to complete the initialization process of ibmvnic device. 2548 * Caller is expected to hold rtnl_lock(). 2549 * 2550 * Returns non-zero if sub-CRQs are not initialized properly leaving the device 2551 * in the down state. 2552 * Returns 0 upon success and the device is in PROBED state. 2553 */ 2554 2555 static int do_passive_init(struct ibmvnic_adapter *adapter) 2556 { 2557 unsigned long timeout = msecs_to_jiffies(30000); 2558 struct net_device *netdev = adapter->netdev; 2559 struct device *dev = &adapter->vdev->dev; 2560 int rc; 2561 2562 netdev_dbg(netdev, "Partner device found, probing.\n"); 2563 2564 adapter->state = VNIC_PROBING; 2565 reinit_completion(&adapter->init_done); 2566 adapter->init_done_rc = 0; 2567 adapter->crq.active = true; 2568 2569 rc = send_crq_init_complete(adapter); 2570 if (rc) 2571 goto out; 2572 2573 rc = send_version_xchg(adapter); 2574 if (rc) 2575 netdev_dbg(adapter->netdev, "send_version_xchg failed, rc=%d\n", rc); 2576 2577 if (!wait_for_completion_timeout(&adapter->init_done, timeout)) { 2578 dev_err(dev, "Initialization sequence timed out\n"); 2579 rc = -ETIMEDOUT; 2580 goto out; 2581 } 2582 2583 rc = init_sub_crqs(adapter); 2584 if (rc) { 2585 dev_err(dev, "Initialization of sub crqs failed, rc=%d\n", rc); 2586 goto out; 2587 } 2588 2589 rc = init_sub_crq_irqs(adapter); 2590 if (rc) { 2591 dev_err(dev, "Failed to initialize sub crq irqs\n, rc=%d", rc); 2592 goto init_failed; 2593 } 2594 2595 netdev->mtu = adapter->req_mtu - ETH_HLEN; 2596 netdev->min_mtu = adapter->min_mtu - ETH_HLEN; 2597 netdev->max_mtu = adapter->max_mtu - ETH_HLEN; 2598 2599 adapter->state = VNIC_PROBED; 2600 netdev_dbg(netdev, "Probed successfully. Waiting for signal from partner device.\n"); 2601 2602 return 0; 2603 2604 init_failed: 2605 release_sub_crqs(adapter, 1); 2606 out: 2607 adapter->state = VNIC_DOWN; 2608 return rc; 2609 } 2610 2611 static void __ibmvnic_reset(struct work_struct *work) 2612 { 2613 struct ibmvnic_adapter *adapter; 2614 bool saved_state = false; 2615 struct ibmvnic_rwi *tmprwi; 2616 struct ibmvnic_rwi *rwi; 2617 unsigned long flags; 2618 u32 reset_state; 2619 int rc = 0; 2620 2621 adapter = container_of(work, struct ibmvnic_adapter, ibmvnic_reset); 2622 2623 if (test_and_set_bit_lock(0, &adapter->resetting)) { 2624 queue_delayed_work(system_long_wq, 2625 &adapter->ibmvnic_delayed_reset, 2626 IBMVNIC_RESET_DELAY); 2627 return; 2628 } 2629 2630 rwi = get_next_rwi(adapter); 2631 while (rwi) { 2632 spin_lock_irqsave(&adapter->state_lock, flags); 2633 2634 if (adapter->state == VNIC_REMOVING || 2635 adapter->state == VNIC_REMOVED) { 2636 spin_unlock_irqrestore(&adapter->state_lock, flags); 2637 kfree(rwi); 2638 rc = EBUSY; 2639 break; 2640 } 2641 2642 if (!saved_state) { 2643 reset_state = adapter->state; 2644 saved_state = true; 2645 } 2646 spin_unlock_irqrestore(&adapter->state_lock, flags); 2647 2648 if (rwi->reset_reason == VNIC_RESET_PASSIVE_INIT) { 2649 rtnl_lock(); 2650 rc = do_passive_init(adapter); 2651 rtnl_unlock(); 2652 if (!rc) 2653 netif_carrier_on(adapter->netdev); 2654 } else if (adapter->force_reset_recovery) { 2655 /* Since we are doing a hard reset now, clear the 2656 * failover_pending flag so we don't ignore any 2657 * future MOBILITY or other resets. 2658 */ 2659 adapter->failover_pending = false; 2660 2661 /* Transport event occurred during previous reset */ 2662 if (adapter->wait_for_reset) { 2663 /* Previous was CHANGE_PARAM; caller locked */ 2664 adapter->force_reset_recovery = false; 2665 rc = do_hard_reset(adapter, rwi, reset_state); 2666 } else { 2667 rtnl_lock(); 2668 adapter->force_reset_recovery = false; 2669 rc = do_hard_reset(adapter, rwi, reset_state); 2670 rtnl_unlock(); 2671 } 2672 if (rc) { 2673 /* give backing device time to settle down */ 2674 netdev_dbg(adapter->netdev, 2675 "[S:%s] Hard reset failed, waiting 60 secs\n", 2676 adapter_state_to_string(adapter->state)); 2677 set_current_state(TASK_UNINTERRUPTIBLE); 2678 schedule_timeout(60 * HZ); 2679 } 2680 } else { 2681 rc = do_reset(adapter, rwi, reset_state); 2682 } 2683 tmprwi = rwi; 2684 adapter->last_reset_time = jiffies; 2685 2686 if (rc) 2687 netdev_dbg(adapter->netdev, "Reset failed, rc=%d\n", rc); 2688 2689 rwi = get_next_rwi(adapter); 2690 2691 /* 2692 * If there is another reset queued, free the previous rwi 2693 * and process the new reset even if previous reset failed 2694 * (the previous reset could have failed because of a fail 2695 * over for instance, so process the fail over). 2696 * 2697 * If there are no resets queued and the previous reset failed, 2698 * the adapter would be in an undefined state. So retry the 2699 * previous reset as a hard reset. 2700 */ 2701 if (rwi) 2702 kfree(tmprwi); 2703 else if (rc) 2704 rwi = tmprwi; 2705 2706 if (rwi && (rwi->reset_reason == VNIC_RESET_FAILOVER || 2707 rwi->reset_reason == VNIC_RESET_MOBILITY || rc)) 2708 adapter->force_reset_recovery = true; 2709 } 2710 2711 if (adapter->wait_for_reset) { 2712 adapter->reset_done_rc = rc; 2713 complete(&adapter->reset_done); 2714 } 2715 2716 clear_bit_unlock(0, &adapter->resetting); 2717 2718 netdev_dbg(adapter->netdev, 2719 "[S:%s FRR:%d WFR:%d] Done processing resets\n", 2720 adapter_state_to_string(adapter->state), 2721 adapter->force_reset_recovery, 2722 adapter->wait_for_reset); 2723 } 2724 2725 static void __ibmvnic_delayed_reset(struct work_struct *work) 2726 { 2727 struct ibmvnic_adapter *adapter; 2728 2729 adapter = container_of(work, struct ibmvnic_adapter, 2730 ibmvnic_delayed_reset.work); 2731 __ibmvnic_reset(&adapter->ibmvnic_reset); 2732 } 2733 2734 static int ibmvnic_reset(struct ibmvnic_adapter *adapter, 2735 enum ibmvnic_reset_reason reason) 2736 { 2737 struct list_head *entry, *tmp_entry; 2738 struct ibmvnic_rwi *rwi, *tmp; 2739 struct net_device *netdev = adapter->netdev; 2740 unsigned long flags; 2741 int ret; 2742 2743 spin_lock_irqsave(&adapter->rwi_lock, flags); 2744 2745 /* If failover is pending don't schedule any other reset. 2746 * Instead let the failover complete. If there is already a 2747 * a failover reset scheduled, we will detect and drop the 2748 * duplicate reset when walking the ->rwi_list below. 2749 */ 2750 if (adapter->state == VNIC_REMOVING || 2751 adapter->state == VNIC_REMOVED || 2752 (adapter->failover_pending && reason != VNIC_RESET_FAILOVER)) { 2753 ret = EBUSY; 2754 netdev_dbg(netdev, "Adapter removing or pending failover, skipping reset\n"); 2755 goto err; 2756 } 2757 2758 if (adapter->state == VNIC_PROBING) { 2759 netdev_warn(netdev, "Adapter reset during probe\n"); 2760 adapter->init_done_rc = EAGAIN; 2761 ret = EAGAIN; 2762 goto err; 2763 } 2764 2765 list_for_each_entry(tmp, &adapter->rwi_list, list) { 2766 if (tmp->reset_reason == reason) { 2767 netdev_dbg(netdev, "Skipping matching reset, reason=%s\n", 2768 reset_reason_to_string(reason)); 2769 ret = EBUSY; 2770 goto err; 2771 } 2772 } 2773 2774 rwi = kzalloc(sizeof(*rwi), GFP_ATOMIC); 2775 if (!rwi) { 2776 ret = ENOMEM; 2777 goto err; 2778 } 2779 /* if we just received a transport event, 2780 * flush reset queue and process this reset 2781 */ 2782 if (adapter->force_reset_recovery && !list_empty(&adapter->rwi_list)) { 2783 list_for_each_safe(entry, tmp_entry, &adapter->rwi_list) 2784 list_del(entry); 2785 } 2786 rwi->reset_reason = reason; 2787 list_add_tail(&rwi->list, &adapter->rwi_list); 2788 netdev_dbg(adapter->netdev, "Scheduling reset (reason %s)\n", 2789 reset_reason_to_string(reason)); 2790 queue_work(system_long_wq, &adapter->ibmvnic_reset); 2791 2792 ret = 0; 2793 err: 2794 /* ibmvnic_close() below can block, so drop the lock first */ 2795 spin_unlock_irqrestore(&adapter->rwi_lock, flags); 2796 2797 if (ret == ENOMEM) 2798 ibmvnic_close(netdev); 2799 2800 return -ret; 2801 } 2802 2803 static void ibmvnic_tx_timeout(struct net_device *dev, unsigned int txqueue) 2804 { 2805 struct ibmvnic_adapter *adapter = netdev_priv(dev); 2806 2807 if (test_bit(0, &adapter->resetting)) { 2808 netdev_err(adapter->netdev, 2809 "Adapter is resetting, skip timeout reset\n"); 2810 return; 2811 } 2812 /* No queuing up reset until at least 5 seconds (default watchdog val) 2813 * after last reset 2814 */ 2815 if (time_before(jiffies, (adapter->last_reset_time + dev->watchdog_timeo))) { 2816 netdev_dbg(dev, "Not yet time to tx timeout.\n"); 2817 return; 2818 } 2819 ibmvnic_reset(adapter, VNIC_RESET_TIMEOUT); 2820 } 2821 2822 static void remove_buff_from_pool(struct ibmvnic_adapter *adapter, 2823 struct ibmvnic_rx_buff *rx_buff) 2824 { 2825 struct ibmvnic_rx_pool *pool = &adapter->rx_pool[rx_buff->pool_index]; 2826 2827 rx_buff->skb = NULL; 2828 2829 pool->free_map[pool->next_alloc] = (int)(rx_buff - pool->rx_buff); 2830 pool->next_alloc = (pool->next_alloc + 1) % pool->size; 2831 2832 atomic_dec(&pool->available); 2833 } 2834 2835 static int ibmvnic_poll(struct napi_struct *napi, int budget) 2836 { 2837 struct ibmvnic_sub_crq_queue *rx_scrq; 2838 struct ibmvnic_adapter *adapter; 2839 struct net_device *netdev; 2840 int frames_processed; 2841 int scrq_num; 2842 2843 netdev = napi->dev; 2844 adapter = netdev_priv(netdev); 2845 scrq_num = (int)(napi - adapter->napi); 2846 frames_processed = 0; 2847 rx_scrq = adapter->rx_scrq[scrq_num]; 2848 2849 restart_poll: 2850 while (frames_processed < budget) { 2851 struct sk_buff *skb; 2852 struct ibmvnic_rx_buff *rx_buff; 2853 union sub_crq *next; 2854 u32 length; 2855 u16 offset; 2856 u8 flags = 0; 2857 2858 if (unlikely(test_bit(0, &adapter->resetting) && 2859 adapter->reset_reason != VNIC_RESET_NON_FATAL)) { 2860 enable_scrq_irq(adapter, rx_scrq); 2861 napi_complete_done(napi, frames_processed); 2862 return frames_processed; 2863 } 2864 2865 if (!pending_scrq(adapter, rx_scrq)) 2866 break; 2867 next = ibmvnic_next_scrq(adapter, rx_scrq); 2868 rx_buff = (struct ibmvnic_rx_buff *) 2869 be64_to_cpu(next->rx_comp.correlator); 2870 /* do error checking */ 2871 if (next->rx_comp.rc) { 2872 netdev_dbg(netdev, "rx buffer returned with rc %x\n", 2873 be16_to_cpu(next->rx_comp.rc)); 2874 /* free the entry */ 2875 next->rx_comp.first = 0; 2876 dev_kfree_skb_any(rx_buff->skb); 2877 remove_buff_from_pool(adapter, rx_buff); 2878 continue; 2879 } else if (!rx_buff->skb) { 2880 /* free the entry */ 2881 next->rx_comp.first = 0; 2882 remove_buff_from_pool(adapter, rx_buff); 2883 continue; 2884 } 2885 2886 length = be32_to_cpu(next->rx_comp.len); 2887 offset = be16_to_cpu(next->rx_comp.off_frame_data); 2888 flags = next->rx_comp.flags; 2889 skb = rx_buff->skb; 2890 /* load long_term_buff before copying to skb */ 2891 dma_rmb(); 2892 skb_copy_to_linear_data(skb, rx_buff->data + offset, 2893 length); 2894 2895 /* VLAN Header has been stripped by the system firmware and 2896 * needs to be inserted by the driver 2897 */ 2898 if (adapter->rx_vlan_header_insertion && 2899 (flags & IBMVNIC_VLAN_STRIPPED)) 2900 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), 2901 ntohs(next->rx_comp.vlan_tci)); 2902 2903 /* free the entry */ 2904 next->rx_comp.first = 0; 2905 remove_buff_from_pool(adapter, rx_buff); 2906 2907 skb_put(skb, length); 2908 skb->protocol = eth_type_trans(skb, netdev); 2909 skb_record_rx_queue(skb, scrq_num); 2910 2911 if (flags & IBMVNIC_IP_CHKSUM_GOOD && 2912 flags & IBMVNIC_TCP_UDP_CHKSUM_GOOD) { 2913 skb->ip_summed = CHECKSUM_UNNECESSARY; 2914 } 2915 2916 length = skb->len; 2917 napi_gro_receive(napi, skb); /* send it up */ 2918 netdev->stats.rx_packets++; 2919 netdev->stats.rx_bytes += length; 2920 adapter->rx_stats_buffers[scrq_num].packets++; 2921 adapter->rx_stats_buffers[scrq_num].bytes += length; 2922 frames_processed++; 2923 } 2924 2925 if (adapter->state != VNIC_CLOSING && 2926 ((atomic_read(&adapter->rx_pool[scrq_num].available) < 2927 adapter->req_rx_add_entries_per_subcrq / 2) || 2928 frames_processed < budget)) 2929 replenish_rx_pool(adapter, &adapter->rx_pool[scrq_num]); 2930 if (frames_processed < budget) { 2931 if (napi_complete_done(napi, frames_processed)) { 2932 enable_scrq_irq(adapter, rx_scrq); 2933 if (pending_scrq(adapter, rx_scrq)) { 2934 if (napi_reschedule(napi)) { 2935 disable_scrq_irq(adapter, rx_scrq); 2936 goto restart_poll; 2937 } 2938 } 2939 } 2940 } 2941 return frames_processed; 2942 } 2943 2944 static int wait_for_reset(struct ibmvnic_adapter *adapter) 2945 { 2946 int rc, ret; 2947 2948 adapter->fallback.mtu = adapter->req_mtu; 2949 adapter->fallback.rx_queues = adapter->req_rx_queues; 2950 adapter->fallback.tx_queues = adapter->req_tx_queues; 2951 adapter->fallback.rx_entries = adapter->req_rx_add_entries_per_subcrq; 2952 adapter->fallback.tx_entries = adapter->req_tx_entries_per_subcrq; 2953 2954 reinit_completion(&adapter->reset_done); 2955 adapter->wait_for_reset = true; 2956 rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM); 2957 2958 if (rc) { 2959 ret = rc; 2960 goto out; 2961 } 2962 rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done, 60000); 2963 if (rc) { 2964 ret = -ENODEV; 2965 goto out; 2966 } 2967 2968 ret = 0; 2969 if (adapter->reset_done_rc) { 2970 ret = -EIO; 2971 adapter->desired.mtu = adapter->fallback.mtu; 2972 adapter->desired.rx_queues = adapter->fallback.rx_queues; 2973 adapter->desired.tx_queues = adapter->fallback.tx_queues; 2974 adapter->desired.rx_entries = adapter->fallback.rx_entries; 2975 adapter->desired.tx_entries = adapter->fallback.tx_entries; 2976 2977 reinit_completion(&adapter->reset_done); 2978 adapter->wait_for_reset = true; 2979 rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM); 2980 if (rc) { 2981 ret = rc; 2982 goto out; 2983 } 2984 rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done, 2985 60000); 2986 if (rc) { 2987 ret = -ENODEV; 2988 goto out; 2989 } 2990 } 2991 out: 2992 adapter->wait_for_reset = false; 2993 2994 return ret; 2995 } 2996 2997 static int ibmvnic_change_mtu(struct net_device *netdev, int new_mtu) 2998 { 2999 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3000 3001 adapter->desired.mtu = new_mtu + ETH_HLEN; 3002 3003 return wait_for_reset(adapter); 3004 } 3005 3006 static netdev_features_t ibmvnic_features_check(struct sk_buff *skb, 3007 struct net_device *dev, 3008 netdev_features_t features) 3009 { 3010 /* Some backing hardware adapters can not 3011 * handle packets with a MSS less than 224 3012 * or with only one segment. 3013 */ 3014 if (skb_is_gso(skb)) { 3015 if (skb_shinfo(skb)->gso_size < 224 || 3016 skb_shinfo(skb)->gso_segs == 1) 3017 features &= ~NETIF_F_GSO_MASK; 3018 } 3019 3020 return features; 3021 } 3022 3023 static const struct net_device_ops ibmvnic_netdev_ops = { 3024 .ndo_open = ibmvnic_open, 3025 .ndo_stop = ibmvnic_close, 3026 .ndo_start_xmit = ibmvnic_xmit, 3027 .ndo_set_rx_mode = ibmvnic_set_multi, 3028 .ndo_set_mac_address = ibmvnic_set_mac, 3029 .ndo_validate_addr = eth_validate_addr, 3030 .ndo_tx_timeout = ibmvnic_tx_timeout, 3031 .ndo_change_mtu = ibmvnic_change_mtu, 3032 .ndo_features_check = ibmvnic_features_check, 3033 }; 3034 3035 /* ethtool functions */ 3036 3037 static int ibmvnic_get_link_ksettings(struct net_device *netdev, 3038 struct ethtool_link_ksettings *cmd) 3039 { 3040 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3041 int rc; 3042 3043 rc = send_query_phys_parms(adapter); 3044 if (rc) { 3045 adapter->speed = SPEED_UNKNOWN; 3046 adapter->duplex = DUPLEX_UNKNOWN; 3047 } 3048 cmd->base.speed = adapter->speed; 3049 cmd->base.duplex = adapter->duplex; 3050 cmd->base.port = PORT_FIBRE; 3051 cmd->base.phy_address = 0; 3052 cmd->base.autoneg = AUTONEG_ENABLE; 3053 3054 return 0; 3055 } 3056 3057 static void ibmvnic_get_drvinfo(struct net_device *netdev, 3058 struct ethtool_drvinfo *info) 3059 { 3060 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3061 3062 strscpy(info->driver, ibmvnic_driver_name, sizeof(info->driver)); 3063 strscpy(info->version, IBMVNIC_DRIVER_VERSION, sizeof(info->version)); 3064 strscpy(info->fw_version, adapter->fw_version, 3065 sizeof(info->fw_version)); 3066 } 3067 3068 static u32 ibmvnic_get_msglevel(struct net_device *netdev) 3069 { 3070 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3071 3072 return adapter->msg_enable; 3073 } 3074 3075 static void ibmvnic_set_msglevel(struct net_device *netdev, u32 data) 3076 { 3077 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3078 3079 adapter->msg_enable = data; 3080 } 3081 3082 static u32 ibmvnic_get_link(struct net_device *netdev) 3083 { 3084 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3085 3086 /* Don't need to send a query because we request a logical link up at 3087 * init and then we wait for link state indications 3088 */ 3089 return adapter->logical_link_state; 3090 } 3091 3092 static void ibmvnic_get_ringparam(struct net_device *netdev, 3093 struct ethtool_ringparam *ring) 3094 { 3095 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3096 3097 if (adapter->priv_flags & IBMVNIC_USE_SERVER_MAXES) { 3098 ring->rx_max_pending = adapter->max_rx_add_entries_per_subcrq; 3099 ring->tx_max_pending = adapter->max_tx_entries_per_subcrq; 3100 } else { 3101 ring->rx_max_pending = IBMVNIC_MAX_QUEUE_SZ; 3102 ring->tx_max_pending = IBMVNIC_MAX_QUEUE_SZ; 3103 } 3104 ring->rx_mini_max_pending = 0; 3105 ring->rx_jumbo_max_pending = 0; 3106 ring->rx_pending = adapter->req_rx_add_entries_per_subcrq; 3107 ring->tx_pending = adapter->req_tx_entries_per_subcrq; 3108 ring->rx_mini_pending = 0; 3109 ring->rx_jumbo_pending = 0; 3110 } 3111 3112 static int ibmvnic_set_ringparam(struct net_device *netdev, 3113 struct ethtool_ringparam *ring) 3114 { 3115 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3116 int ret; 3117 3118 ret = 0; 3119 adapter->desired.rx_entries = ring->rx_pending; 3120 adapter->desired.tx_entries = ring->tx_pending; 3121 3122 ret = wait_for_reset(adapter); 3123 3124 if (!ret && 3125 (adapter->req_rx_add_entries_per_subcrq != ring->rx_pending || 3126 adapter->req_tx_entries_per_subcrq != ring->tx_pending)) 3127 netdev_info(netdev, 3128 "Could not match full ringsize request. Requested: RX %d, TX %d; Allowed: RX %llu, TX %llu\n", 3129 ring->rx_pending, ring->tx_pending, 3130 adapter->req_rx_add_entries_per_subcrq, 3131 adapter->req_tx_entries_per_subcrq); 3132 return ret; 3133 } 3134 3135 static void ibmvnic_get_channels(struct net_device *netdev, 3136 struct ethtool_channels *channels) 3137 { 3138 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3139 3140 if (adapter->priv_flags & IBMVNIC_USE_SERVER_MAXES) { 3141 channels->max_rx = adapter->max_rx_queues; 3142 channels->max_tx = adapter->max_tx_queues; 3143 } else { 3144 channels->max_rx = IBMVNIC_MAX_QUEUES; 3145 channels->max_tx = IBMVNIC_MAX_QUEUES; 3146 } 3147 3148 channels->max_other = 0; 3149 channels->max_combined = 0; 3150 channels->rx_count = adapter->req_rx_queues; 3151 channels->tx_count = adapter->req_tx_queues; 3152 channels->other_count = 0; 3153 channels->combined_count = 0; 3154 } 3155 3156 static int ibmvnic_set_channels(struct net_device *netdev, 3157 struct ethtool_channels *channels) 3158 { 3159 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3160 int ret; 3161 3162 ret = 0; 3163 adapter->desired.rx_queues = channels->rx_count; 3164 adapter->desired.tx_queues = channels->tx_count; 3165 3166 ret = wait_for_reset(adapter); 3167 3168 if (!ret && 3169 (adapter->req_rx_queues != channels->rx_count || 3170 adapter->req_tx_queues != channels->tx_count)) 3171 netdev_info(netdev, 3172 "Could not match full channels request. Requested: RX %d, TX %d; Allowed: RX %llu, TX %llu\n", 3173 channels->rx_count, channels->tx_count, 3174 adapter->req_rx_queues, adapter->req_tx_queues); 3175 return ret; 3176 } 3177 3178 static void ibmvnic_get_strings(struct net_device *dev, u32 stringset, u8 *data) 3179 { 3180 struct ibmvnic_adapter *adapter = netdev_priv(dev); 3181 int i; 3182 3183 switch (stringset) { 3184 case ETH_SS_STATS: 3185 for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); 3186 i++, data += ETH_GSTRING_LEN) 3187 memcpy(data, ibmvnic_stats[i].name, ETH_GSTRING_LEN); 3188 3189 for (i = 0; i < adapter->req_tx_queues; i++) { 3190 snprintf(data, ETH_GSTRING_LEN, "tx%d_packets", i); 3191 data += ETH_GSTRING_LEN; 3192 3193 snprintf(data, ETH_GSTRING_LEN, "tx%d_bytes", i); 3194 data += ETH_GSTRING_LEN; 3195 3196 snprintf(data, ETH_GSTRING_LEN, 3197 "tx%d_dropped_packets", i); 3198 data += ETH_GSTRING_LEN; 3199 } 3200 3201 for (i = 0; i < adapter->req_rx_queues; i++) { 3202 snprintf(data, ETH_GSTRING_LEN, "rx%d_packets", i); 3203 data += ETH_GSTRING_LEN; 3204 3205 snprintf(data, ETH_GSTRING_LEN, "rx%d_bytes", i); 3206 data += ETH_GSTRING_LEN; 3207 3208 snprintf(data, ETH_GSTRING_LEN, "rx%d_interrupts", i); 3209 data += ETH_GSTRING_LEN; 3210 } 3211 break; 3212 3213 case ETH_SS_PRIV_FLAGS: 3214 for (i = 0; i < ARRAY_SIZE(ibmvnic_priv_flags); i++) 3215 strcpy(data + i * ETH_GSTRING_LEN, 3216 ibmvnic_priv_flags[i]); 3217 break; 3218 default: 3219 return; 3220 } 3221 } 3222 3223 static int ibmvnic_get_sset_count(struct net_device *dev, int sset) 3224 { 3225 struct ibmvnic_adapter *adapter = netdev_priv(dev); 3226 3227 switch (sset) { 3228 case ETH_SS_STATS: 3229 return ARRAY_SIZE(ibmvnic_stats) + 3230 adapter->req_tx_queues * NUM_TX_STATS + 3231 adapter->req_rx_queues * NUM_RX_STATS; 3232 case ETH_SS_PRIV_FLAGS: 3233 return ARRAY_SIZE(ibmvnic_priv_flags); 3234 default: 3235 return -EOPNOTSUPP; 3236 } 3237 } 3238 3239 static void ibmvnic_get_ethtool_stats(struct net_device *dev, 3240 struct ethtool_stats *stats, u64 *data) 3241 { 3242 struct ibmvnic_adapter *adapter = netdev_priv(dev); 3243 union ibmvnic_crq crq; 3244 int i, j; 3245 int rc; 3246 3247 memset(&crq, 0, sizeof(crq)); 3248 crq.request_statistics.first = IBMVNIC_CRQ_CMD; 3249 crq.request_statistics.cmd = REQUEST_STATISTICS; 3250 crq.request_statistics.ioba = cpu_to_be32(adapter->stats_token); 3251 crq.request_statistics.len = 3252 cpu_to_be32(sizeof(struct ibmvnic_statistics)); 3253 3254 /* Wait for data to be written */ 3255 reinit_completion(&adapter->stats_done); 3256 rc = ibmvnic_send_crq(adapter, &crq); 3257 if (rc) 3258 return; 3259 rc = ibmvnic_wait_for_completion(adapter, &adapter->stats_done, 10000); 3260 if (rc) 3261 return; 3262 3263 for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); i++) 3264 data[i] = be64_to_cpu(IBMVNIC_GET_STAT 3265 (adapter, ibmvnic_stats[i].offset)); 3266 3267 for (j = 0; j < adapter->req_tx_queues; j++) { 3268 data[i] = adapter->tx_stats_buffers[j].packets; 3269 i++; 3270 data[i] = adapter->tx_stats_buffers[j].bytes; 3271 i++; 3272 data[i] = adapter->tx_stats_buffers[j].dropped_packets; 3273 i++; 3274 } 3275 3276 for (j = 0; j < adapter->req_rx_queues; j++) { 3277 data[i] = adapter->rx_stats_buffers[j].packets; 3278 i++; 3279 data[i] = adapter->rx_stats_buffers[j].bytes; 3280 i++; 3281 data[i] = adapter->rx_stats_buffers[j].interrupts; 3282 i++; 3283 } 3284 } 3285 3286 static u32 ibmvnic_get_priv_flags(struct net_device *netdev) 3287 { 3288 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3289 3290 return adapter->priv_flags; 3291 } 3292 3293 static int ibmvnic_set_priv_flags(struct net_device *netdev, u32 flags) 3294 { 3295 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3296 bool which_maxes = !!(flags & IBMVNIC_USE_SERVER_MAXES); 3297 3298 if (which_maxes) 3299 adapter->priv_flags |= IBMVNIC_USE_SERVER_MAXES; 3300 else 3301 adapter->priv_flags &= ~IBMVNIC_USE_SERVER_MAXES; 3302 3303 return 0; 3304 } 3305 3306 static const struct ethtool_ops ibmvnic_ethtool_ops = { 3307 .get_drvinfo = ibmvnic_get_drvinfo, 3308 .get_msglevel = ibmvnic_get_msglevel, 3309 .set_msglevel = ibmvnic_set_msglevel, 3310 .get_link = ibmvnic_get_link, 3311 .get_ringparam = ibmvnic_get_ringparam, 3312 .set_ringparam = ibmvnic_set_ringparam, 3313 .get_channels = ibmvnic_get_channels, 3314 .set_channels = ibmvnic_set_channels, 3315 .get_strings = ibmvnic_get_strings, 3316 .get_sset_count = ibmvnic_get_sset_count, 3317 .get_ethtool_stats = ibmvnic_get_ethtool_stats, 3318 .get_link_ksettings = ibmvnic_get_link_ksettings, 3319 .get_priv_flags = ibmvnic_get_priv_flags, 3320 .set_priv_flags = ibmvnic_set_priv_flags, 3321 }; 3322 3323 /* Routines for managing CRQs/sCRQs */ 3324 3325 static int reset_one_sub_crq_queue(struct ibmvnic_adapter *adapter, 3326 struct ibmvnic_sub_crq_queue *scrq) 3327 { 3328 int rc; 3329 3330 if (!scrq) { 3331 netdev_dbg(adapter->netdev, "Invalid scrq reset.\n"); 3332 return -EINVAL; 3333 } 3334 3335 if (scrq->irq) { 3336 free_irq(scrq->irq, scrq); 3337 irq_dispose_mapping(scrq->irq); 3338 scrq->irq = 0; 3339 } 3340 3341 if (scrq->msgs) { 3342 memset(scrq->msgs, 0, 4 * PAGE_SIZE); 3343 atomic_set(&scrq->used, 0); 3344 scrq->cur = 0; 3345 scrq->ind_buf.index = 0; 3346 } else { 3347 netdev_dbg(adapter->netdev, "Invalid scrq reset\n"); 3348 return -EINVAL; 3349 } 3350 3351 rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token, 3352 4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq); 3353 return rc; 3354 } 3355 3356 static int reset_sub_crq_queues(struct ibmvnic_adapter *adapter) 3357 { 3358 int i, rc; 3359 3360 if (!adapter->tx_scrq || !adapter->rx_scrq) 3361 return -EINVAL; 3362 3363 for (i = 0; i < adapter->req_tx_queues; i++) { 3364 netdev_dbg(adapter->netdev, "Re-setting tx_scrq[%d]\n", i); 3365 rc = reset_one_sub_crq_queue(adapter, adapter->tx_scrq[i]); 3366 if (rc) 3367 return rc; 3368 } 3369 3370 for (i = 0; i < adapter->req_rx_queues; i++) { 3371 netdev_dbg(adapter->netdev, "Re-setting rx_scrq[%d]\n", i); 3372 rc = reset_one_sub_crq_queue(adapter, adapter->rx_scrq[i]); 3373 if (rc) 3374 return rc; 3375 } 3376 3377 return rc; 3378 } 3379 3380 static void release_sub_crq_queue(struct ibmvnic_adapter *adapter, 3381 struct ibmvnic_sub_crq_queue *scrq, 3382 bool do_h_free) 3383 { 3384 struct device *dev = &adapter->vdev->dev; 3385 long rc; 3386 3387 netdev_dbg(adapter->netdev, "Releasing sub-CRQ\n"); 3388 3389 if (do_h_free) { 3390 /* Close the sub-crqs */ 3391 do { 3392 rc = plpar_hcall_norets(H_FREE_SUB_CRQ, 3393 adapter->vdev->unit_address, 3394 scrq->crq_num); 3395 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc)); 3396 3397 if (rc) { 3398 netdev_err(adapter->netdev, 3399 "Failed to release sub-CRQ %16lx, rc = %ld\n", 3400 scrq->crq_num, rc); 3401 } 3402 } 3403 3404 dma_free_coherent(dev, 3405 IBMVNIC_IND_ARR_SZ, 3406 scrq->ind_buf.indir_arr, 3407 scrq->ind_buf.indir_dma); 3408 3409 dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE, 3410 DMA_BIDIRECTIONAL); 3411 free_pages((unsigned long)scrq->msgs, 2); 3412 kfree(scrq); 3413 } 3414 3415 static struct ibmvnic_sub_crq_queue *init_sub_crq_queue(struct ibmvnic_adapter 3416 *adapter) 3417 { 3418 struct device *dev = &adapter->vdev->dev; 3419 struct ibmvnic_sub_crq_queue *scrq; 3420 int rc; 3421 3422 scrq = kzalloc(sizeof(*scrq), GFP_KERNEL); 3423 if (!scrq) 3424 return NULL; 3425 3426 scrq->msgs = 3427 (union sub_crq *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 2); 3428 if (!scrq->msgs) { 3429 dev_warn(dev, "Couldn't allocate crq queue messages page\n"); 3430 goto zero_page_failed; 3431 } 3432 3433 scrq->msg_token = dma_map_single(dev, scrq->msgs, 4 * PAGE_SIZE, 3434 DMA_BIDIRECTIONAL); 3435 if (dma_mapping_error(dev, scrq->msg_token)) { 3436 dev_warn(dev, "Couldn't map crq queue messages page\n"); 3437 goto map_failed; 3438 } 3439 3440 rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token, 3441 4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq); 3442 3443 if (rc == H_RESOURCE) 3444 rc = ibmvnic_reset_crq(adapter); 3445 3446 if (rc == H_CLOSED) { 3447 dev_warn(dev, "Partner adapter not ready, waiting.\n"); 3448 } else if (rc) { 3449 dev_warn(dev, "Error %d registering sub-crq\n", rc); 3450 goto reg_failed; 3451 } 3452 3453 scrq->adapter = adapter; 3454 scrq->size = 4 * PAGE_SIZE / sizeof(*scrq->msgs); 3455 scrq->ind_buf.index = 0; 3456 3457 scrq->ind_buf.indir_arr = 3458 dma_alloc_coherent(dev, 3459 IBMVNIC_IND_ARR_SZ, 3460 &scrq->ind_buf.indir_dma, 3461 GFP_KERNEL); 3462 3463 if (!scrq->ind_buf.indir_arr) 3464 goto indir_failed; 3465 3466 spin_lock_init(&scrq->lock); 3467 3468 netdev_dbg(adapter->netdev, 3469 "sub-crq initialized, num %lx, hw_irq=%lx, irq=%x\n", 3470 scrq->crq_num, scrq->hw_irq, scrq->irq); 3471 3472 return scrq; 3473 3474 indir_failed: 3475 do { 3476 rc = plpar_hcall_norets(H_FREE_SUB_CRQ, 3477 adapter->vdev->unit_address, 3478 scrq->crq_num); 3479 } while (rc == H_BUSY || rc == H_IS_LONG_BUSY(rc)); 3480 reg_failed: 3481 dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE, 3482 DMA_BIDIRECTIONAL); 3483 map_failed: 3484 free_pages((unsigned long)scrq->msgs, 2); 3485 zero_page_failed: 3486 kfree(scrq); 3487 3488 return NULL; 3489 } 3490 3491 static void release_sub_crqs(struct ibmvnic_adapter *adapter, bool do_h_free) 3492 { 3493 int i; 3494 3495 if (adapter->tx_scrq) { 3496 for (i = 0; i < adapter->num_active_tx_scrqs; i++) { 3497 if (!adapter->tx_scrq[i]) 3498 continue; 3499 3500 netdev_dbg(adapter->netdev, "Releasing tx_scrq[%d]\n", 3501 i); 3502 ibmvnic_tx_scrq_clean_buffer(adapter, adapter->tx_scrq[i]); 3503 if (adapter->tx_scrq[i]->irq) { 3504 free_irq(adapter->tx_scrq[i]->irq, 3505 adapter->tx_scrq[i]); 3506 irq_dispose_mapping(adapter->tx_scrq[i]->irq); 3507 adapter->tx_scrq[i]->irq = 0; 3508 } 3509 3510 release_sub_crq_queue(adapter, adapter->tx_scrq[i], 3511 do_h_free); 3512 } 3513 3514 kfree(adapter->tx_scrq); 3515 adapter->tx_scrq = NULL; 3516 adapter->num_active_tx_scrqs = 0; 3517 } 3518 3519 if (adapter->rx_scrq) { 3520 for (i = 0; i < adapter->num_active_rx_scrqs; i++) { 3521 if (!adapter->rx_scrq[i]) 3522 continue; 3523 3524 netdev_dbg(adapter->netdev, "Releasing rx_scrq[%d]\n", 3525 i); 3526 if (adapter->rx_scrq[i]->irq) { 3527 free_irq(adapter->rx_scrq[i]->irq, 3528 adapter->rx_scrq[i]); 3529 irq_dispose_mapping(adapter->rx_scrq[i]->irq); 3530 adapter->rx_scrq[i]->irq = 0; 3531 } 3532 3533 release_sub_crq_queue(adapter, adapter->rx_scrq[i], 3534 do_h_free); 3535 } 3536 3537 kfree(adapter->rx_scrq); 3538 adapter->rx_scrq = NULL; 3539 adapter->num_active_rx_scrqs = 0; 3540 } 3541 } 3542 3543 static int disable_scrq_irq(struct ibmvnic_adapter *adapter, 3544 struct ibmvnic_sub_crq_queue *scrq) 3545 { 3546 struct device *dev = &adapter->vdev->dev; 3547 unsigned long rc; 3548 3549 rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address, 3550 H_DISABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0); 3551 if (rc) 3552 dev_err(dev, "Couldn't disable scrq irq 0x%lx. rc=%ld\n", 3553 scrq->hw_irq, rc); 3554 return rc; 3555 } 3556 3557 static int enable_scrq_irq(struct ibmvnic_adapter *adapter, 3558 struct ibmvnic_sub_crq_queue *scrq) 3559 { 3560 struct device *dev = &adapter->vdev->dev; 3561 unsigned long rc; 3562 3563 if (scrq->hw_irq > 0x100000000ULL) { 3564 dev_err(dev, "bad hw_irq = %lx\n", scrq->hw_irq); 3565 return 1; 3566 } 3567 3568 if (test_bit(0, &adapter->resetting) && 3569 adapter->reset_reason == VNIC_RESET_MOBILITY) { 3570 u64 val = (0xff000000) | scrq->hw_irq; 3571 3572 rc = plpar_hcall_norets(H_EOI, val); 3573 /* H_EOI would fail with rc = H_FUNCTION when running 3574 * in XIVE mode which is expected, but not an error. 3575 */ 3576 if (rc && (rc != H_FUNCTION)) 3577 dev_err(dev, "H_EOI FAILED irq 0x%llx. rc=%ld\n", 3578 val, rc); 3579 } 3580 3581 rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address, 3582 H_ENABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0); 3583 if (rc) 3584 dev_err(dev, "Couldn't enable scrq irq 0x%lx. rc=%ld\n", 3585 scrq->hw_irq, rc); 3586 return rc; 3587 } 3588 3589 static int ibmvnic_complete_tx(struct ibmvnic_adapter *adapter, 3590 struct ibmvnic_sub_crq_queue *scrq) 3591 { 3592 struct device *dev = &adapter->vdev->dev; 3593 struct ibmvnic_tx_pool *tx_pool; 3594 struct ibmvnic_tx_buff *txbuff; 3595 struct netdev_queue *txq; 3596 union sub_crq *next; 3597 int index; 3598 int i; 3599 3600 restart_loop: 3601 while (pending_scrq(adapter, scrq)) { 3602 unsigned int pool = scrq->pool_index; 3603 int num_entries = 0; 3604 int total_bytes = 0; 3605 int num_packets = 0; 3606 3607 next = ibmvnic_next_scrq(adapter, scrq); 3608 for (i = 0; i < next->tx_comp.num_comps; i++) { 3609 index = be32_to_cpu(next->tx_comp.correlators[i]); 3610 if (index & IBMVNIC_TSO_POOL_MASK) { 3611 tx_pool = &adapter->tso_pool[pool]; 3612 index &= ~IBMVNIC_TSO_POOL_MASK; 3613 } else { 3614 tx_pool = &adapter->tx_pool[pool]; 3615 } 3616 3617 txbuff = &tx_pool->tx_buff[index]; 3618 num_packets++; 3619 num_entries += txbuff->num_entries; 3620 if (txbuff->skb) { 3621 total_bytes += txbuff->skb->len; 3622 if (next->tx_comp.rcs[i]) { 3623 dev_err(dev, "tx error %x\n", 3624 next->tx_comp.rcs[i]); 3625 dev_kfree_skb_irq(txbuff->skb); 3626 } else { 3627 dev_consume_skb_irq(txbuff->skb); 3628 } 3629 txbuff->skb = NULL; 3630 } else { 3631 netdev_warn(adapter->netdev, 3632 "TX completion received with NULL socket buffer\n"); 3633 } 3634 tx_pool->free_map[tx_pool->producer_index] = index; 3635 tx_pool->producer_index = 3636 (tx_pool->producer_index + 1) % 3637 tx_pool->num_buffers; 3638 } 3639 /* remove tx_comp scrq*/ 3640 next->tx_comp.first = 0; 3641 3642 txq = netdev_get_tx_queue(adapter->netdev, scrq->pool_index); 3643 netdev_tx_completed_queue(txq, num_packets, total_bytes); 3644 3645 if (atomic_sub_return(num_entries, &scrq->used) <= 3646 (adapter->req_tx_entries_per_subcrq / 2) && 3647 __netif_subqueue_stopped(adapter->netdev, 3648 scrq->pool_index)) { 3649 netif_wake_subqueue(adapter->netdev, scrq->pool_index); 3650 netdev_dbg(adapter->netdev, "Started queue %d\n", 3651 scrq->pool_index); 3652 } 3653 } 3654 3655 enable_scrq_irq(adapter, scrq); 3656 3657 if (pending_scrq(adapter, scrq)) { 3658 disable_scrq_irq(adapter, scrq); 3659 goto restart_loop; 3660 } 3661 3662 return 0; 3663 } 3664 3665 static irqreturn_t ibmvnic_interrupt_tx(int irq, void *instance) 3666 { 3667 struct ibmvnic_sub_crq_queue *scrq = instance; 3668 struct ibmvnic_adapter *adapter = scrq->adapter; 3669 3670 disable_scrq_irq(adapter, scrq); 3671 ibmvnic_complete_tx(adapter, scrq); 3672 3673 return IRQ_HANDLED; 3674 } 3675 3676 static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance) 3677 { 3678 struct ibmvnic_sub_crq_queue *scrq = instance; 3679 struct ibmvnic_adapter *adapter = scrq->adapter; 3680 3681 /* When booting a kdump kernel we can hit pending interrupts 3682 * prior to completing driver initialization. 3683 */ 3684 if (unlikely(adapter->state != VNIC_OPEN)) 3685 return IRQ_NONE; 3686 3687 adapter->rx_stats_buffers[scrq->scrq_num].interrupts++; 3688 3689 if (napi_schedule_prep(&adapter->napi[scrq->scrq_num])) { 3690 disable_scrq_irq(adapter, scrq); 3691 __napi_schedule(&adapter->napi[scrq->scrq_num]); 3692 } 3693 3694 return IRQ_HANDLED; 3695 } 3696 3697 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter) 3698 { 3699 struct device *dev = &adapter->vdev->dev; 3700 struct ibmvnic_sub_crq_queue *scrq; 3701 int i = 0, j = 0; 3702 int rc = 0; 3703 3704 for (i = 0; i < adapter->req_tx_queues; i++) { 3705 netdev_dbg(adapter->netdev, "Initializing tx_scrq[%d] irq\n", 3706 i); 3707 scrq = adapter->tx_scrq[i]; 3708 scrq->irq = irq_create_mapping(NULL, scrq->hw_irq); 3709 3710 if (!scrq->irq) { 3711 rc = -EINVAL; 3712 dev_err(dev, "Error mapping irq\n"); 3713 goto req_tx_irq_failed; 3714 } 3715 3716 snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-tx%d", 3717 adapter->vdev->unit_address, i); 3718 rc = request_irq(scrq->irq, ibmvnic_interrupt_tx, 3719 0, scrq->name, scrq); 3720 3721 if (rc) { 3722 dev_err(dev, "Couldn't register tx irq 0x%x. rc=%d\n", 3723 scrq->irq, rc); 3724 irq_dispose_mapping(scrq->irq); 3725 goto req_tx_irq_failed; 3726 } 3727 } 3728 3729 for (i = 0; i < adapter->req_rx_queues; i++) { 3730 netdev_dbg(adapter->netdev, "Initializing rx_scrq[%d] irq\n", 3731 i); 3732 scrq = adapter->rx_scrq[i]; 3733 scrq->irq = irq_create_mapping(NULL, scrq->hw_irq); 3734 if (!scrq->irq) { 3735 rc = -EINVAL; 3736 dev_err(dev, "Error mapping irq\n"); 3737 goto req_rx_irq_failed; 3738 } 3739 snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-rx%d", 3740 adapter->vdev->unit_address, i); 3741 rc = request_irq(scrq->irq, ibmvnic_interrupt_rx, 3742 0, scrq->name, scrq); 3743 if (rc) { 3744 dev_err(dev, "Couldn't register rx irq 0x%x. rc=%d\n", 3745 scrq->irq, rc); 3746 irq_dispose_mapping(scrq->irq); 3747 goto req_rx_irq_failed; 3748 } 3749 } 3750 return rc; 3751 3752 req_rx_irq_failed: 3753 for (j = 0; j < i; j++) { 3754 free_irq(adapter->rx_scrq[j]->irq, adapter->rx_scrq[j]); 3755 irq_dispose_mapping(adapter->rx_scrq[j]->irq); 3756 } 3757 i = adapter->req_tx_queues; 3758 req_tx_irq_failed: 3759 for (j = 0; j < i; j++) { 3760 free_irq(adapter->tx_scrq[j]->irq, adapter->tx_scrq[j]); 3761 irq_dispose_mapping(adapter->tx_scrq[j]->irq); 3762 } 3763 release_sub_crqs(adapter, 1); 3764 return rc; 3765 } 3766 3767 static int init_sub_crqs(struct ibmvnic_adapter *adapter) 3768 { 3769 struct device *dev = &adapter->vdev->dev; 3770 struct ibmvnic_sub_crq_queue **allqueues; 3771 int registered_queues = 0; 3772 int total_queues; 3773 int more = 0; 3774 int i; 3775 3776 total_queues = adapter->req_tx_queues + adapter->req_rx_queues; 3777 3778 allqueues = kcalloc(total_queues, sizeof(*allqueues), GFP_KERNEL); 3779 if (!allqueues) 3780 return -1; 3781 3782 for (i = 0; i < total_queues; i++) { 3783 allqueues[i] = init_sub_crq_queue(adapter); 3784 if (!allqueues[i]) { 3785 dev_warn(dev, "Couldn't allocate all sub-crqs\n"); 3786 break; 3787 } 3788 registered_queues++; 3789 } 3790 3791 /* Make sure we were able to register the minimum number of queues */ 3792 if (registered_queues < 3793 adapter->min_tx_queues + adapter->min_rx_queues) { 3794 dev_err(dev, "Fatal: Couldn't init min number of sub-crqs\n"); 3795 goto tx_failed; 3796 } 3797 3798 /* Distribute the failed allocated queues*/ 3799 for (i = 0; i < total_queues - registered_queues + more ; i++) { 3800 netdev_dbg(adapter->netdev, "Reducing number of queues\n"); 3801 switch (i % 3) { 3802 case 0: 3803 if (adapter->req_rx_queues > adapter->min_rx_queues) 3804 adapter->req_rx_queues--; 3805 else 3806 more++; 3807 break; 3808 case 1: 3809 if (adapter->req_tx_queues > adapter->min_tx_queues) 3810 adapter->req_tx_queues--; 3811 else 3812 more++; 3813 break; 3814 } 3815 } 3816 3817 adapter->tx_scrq = kcalloc(adapter->req_tx_queues, 3818 sizeof(*adapter->tx_scrq), GFP_KERNEL); 3819 if (!adapter->tx_scrq) 3820 goto tx_failed; 3821 3822 for (i = 0; i < adapter->req_tx_queues; i++) { 3823 adapter->tx_scrq[i] = allqueues[i]; 3824 adapter->tx_scrq[i]->pool_index = i; 3825 adapter->num_active_tx_scrqs++; 3826 } 3827 3828 adapter->rx_scrq = kcalloc(adapter->req_rx_queues, 3829 sizeof(*adapter->rx_scrq), GFP_KERNEL); 3830 if (!adapter->rx_scrq) 3831 goto rx_failed; 3832 3833 for (i = 0; i < adapter->req_rx_queues; i++) { 3834 adapter->rx_scrq[i] = allqueues[i + adapter->req_tx_queues]; 3835 adapter->rx_scrq[i]->scrq_num = i; 3836 adapter->num_active_rx_scrqs++; 3837 } 3838 3839 kfree(allqueues); 3840 return 0; 3841 3842 rx_failed: 3843 kfree(adapter->tx_scrq); 3844 adapter->tx_scrq = NULL; 3845 tx_failed: 3846 for (i = 0; i < registered_queues; i++) 3847 release_sub_crq_queue(adapter, allqueues[i], 1); 3848 kfree(allqueues); 3849 return -1; 3850 } 3851 3852 static void send_request_cap(struct ibmvnic_adapter *adapter, int retry) 3853 { 3854 struct device *dev = &adapter->vdev->dev; 3855 union ibmvnic_crq crq; 3856 int max_entries; 3857 3858 if (!retry) { 3859 /* Sub-CRQ entries are 32 byte long */ 3860 int entries_page = 4 * PAGE_SIZE / (sizeof(u64) * 4); 3861 3862 if (adapter->min_tx_entries_per_subcrq > entries_page || 3863 adapter->min_rx_add_entries_per_subcrq > entries_page) { 3864 dev_err(dev, "Fatal, invalid entries per sub-crq\n"); 3865 return; 3866 } 3867 3868 if (adapter->desired.mtu) 3869 adapter->req_mtu = adapter->desired.mtu; 3870 else 3871 adapter->req_mtu = adapter->netdev->mtu + ETH_HLEN; 3872 3873 if (!adapter->desired.tx_entries) 3874 adapter->desired.tx_entries = 3875 adapter->max_tx_entries_per_subcrq; 3876 if (!adapter->desired.rx_entries) 3877 adapter->desired.rx_entries = 3878 adapter->max_rx_add_entries_per_subcrq; 3879 3880 max_entries = IBMVNIC_MAX_LTB_SIZE / 3881 (adapter->req_mtu + IBMVNIC_BUFFER_HLEN); 3882 3883 if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) * 3884 adapter->desired.tx_entries > IBMVNIC_MAX_LTB_SIZE) { 3885 adapter->desired.tx_entries = max_entries; 3886 } 3887 3888 if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) * 3889 adapter->desired.rx_entries > IBMVNIC_MAX_LTB_SIZE) { 3890 adapter->desired.rx_entries = max_entries; 3891 } 3892 3893 if (adapter->desired.tx_entries) 3894 adapter->req_tx_entries_per_subcrq = 3895 adapter->desired.tx_entries; 3896 else 3897 adapter->req_tx_entries_per_subcrq = 3898 adapter->max_tx_entries_per_subcrq; 3899 3900 if (adapter->desired.rx_entries) 3901 adapter->req_rx_add_entries_per_subcrq = 3902 adapter->desired.rx_entries; 3903 else 3904 adapter->req_rx_add_entries_per_subcrq = 3905 adapter->max_rx_add_entries_per_subcrq; 3906 3907 if (adapter->desired.tx_queues) 3908 adapter->req_tx_queues = 3909 adapter->desired.tx_queues; 3910 else 3911 adapter->req_tx_queues = 3912 adapter->opt_tx_comp_sub_queues; 3913 3914 if (adapter->desired.rx_queues) 3915 adapter->req_rx_queues = 3916 adapter->desired.rx_queues; 3917 else 3918 adapter->req_rx_queues = 3919 adapter->opt_rx_comp_queues; 3920 3921 adapter->req_rx_add_queues = adapter->max_rx_add_queues; 3922 } 3923 3924 memset(&crq, 0, sizeof(crq)); 3925 crq.request_capability.first = IBMVNIC_CRQ_CMD; 3926 crq.request_capability.cmd = REQUEST_CAPABILITY; 3927 3928 crq.request_capability.capability = cpu_to_be16(REQ_TX_QUEUES); 3929 crq.request_capability.number = cpu_to_be64(adapter->req_tx_queues); 3930 atomic_inc(&adapter->running_cap_crqs); 3931 ibmvnic_send_crq(adapter, &crq); 3932 3933 crq.request_capability.capability = cpu_to_be16(REQ_RX_QUEUES); 3934 crq.request_capability.number = cpu_to_be64(adapter->req_rx_queues); 3935 atomic_inc(&adapter->running_cap_crqs); 3936 ibmvnic_send_crq(adapter, &crq); 3937 3938 crq.request_capability.capability = cpu_to_be16(REQ_RX_ADD_QUEUES); 3939 crq.request_capability.number = cpu_to_be64(adapter->req_rx_add_queues); 3940 atomic_inc(&adapter->running_cap_crqs); 3941 ibmvnic_send_crq(adapter, &crq); 3942 3943 crq.request_capability.capability = 3944 cpu_to_be16(REQ_TX_ENTRIES_PER_SUBCRQ); 3945 crq.request_capability.number = 3946 cpu_to_be64(adapter->req_tx_entries_per_subcrq); 3947 atomic_inc(&adapter->running_cap_crqs); 3948 ibmvnic_send_crq(adapter, &crq); 3949 3950 crq.request_capability.capability = 3951 cpu_to_be16(REQ_RX_ADD_ENTRIES_PER_SUBCRQ); 3952 crq.request_capability.number = 3953 cpu_to_be64(adapter->req_rx_add_entries_per_subcrq); 3954 atomic_inc(&adapter->running_cap_crqs); 3955 ibmvnic_send_crq(adapter, &crq); 3956 3957 crq.request_capability.capability = cpu_to_be16(REQ_MTU); 3958 crq.request_capability.number = cpu_to_be64(adapter->req_mtu); 3959 atomic_inc(&adapter->running_cap_crqs); 3960 ibmvnic_send_crq(adapter, &crq); 3961 3962 if (adapter->netdev->flags & IFF_PROMISC) { 3963 if (adapter->promisc_supported) { 3964 crq.request_capability.capability = 3965 cpu_to_be16(PROMISC_REQUESTED); 3966 crq.request_capability.number = cpu_to_be64(1); 3967 atomic_inc(&adapter->running_cap_crqs); 3968 ibmvnic_send_crq(adapter, &crq); 3969 } 3970 } else { 3971 crq.request_capability.capability = 3972 cpu_to_be16(PROMISC_REQUESTED); 3973 crq.request_capability.number = cpu_to_be64(0); 3974 atomic_inc(&adapter->running_cap_crqs); 3975 ibmvnic_send_crq(adapter, &crq); 3976 } 3977 } 3978 3979 static int pending_scrq(struct ibmvnic_adapter *adapter, 3980 struct ibmvnic_sub_crq_queue *scrq) 3981 { 3982 union sub_crq *entry = &scrq->msgs[scrq->cur]; 3983 int rc; 3984 3985 rc = !!(entry->generic.first & IBMVNIC_CRQ_CMD_RSP); 3986 3987 /* Ensure that the SCRQ valid flag is loaded prior to loading the 3988 * contents of the SCRQ descriptor 3989 */ 3990 dma_rmb(); 3991 3992 return rc; 3993 } 3994 3995 static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *adapter, 3996 struct ibmvnic_sub_crq_queue *scrq) 3997 { 3998 union sub_crq *entry; 3999 unsigned long flags; 4000 4001 spin_lock_irqsave(&scrq->lock, flags); 4002 entry = &scrq->msgs[scrq->cur]; 4003 if (entry->generic.first & IBMVNIC_CRQ_CMD_RSP) { 4004 if (++scrq->cur == scrq->size) 4005 scrq->cur = 0; 4006 } else { 4007 entry = NULL; 4008 } 4009 spin_unlock_irqrestore(&scrq->lock, flags); 4010 4011 /* Ensure that the SCRQ valid flag is loaded prior to loading the 4012 * contents of the SCRQ descriptor 4013 */ 4014 dma_rmb(); 4015 4016 return entry; 4017 } 4018 4019 static union ibmvnic_crq *ibmvnic_next_crq(struct ibmvnic_adapter *adapter) 4020 { 4021 struct ibmvnic_crq_queue *queue = &adapter->crq; 4022 union ibmvnic_crq *crq; 4023 4024 crq = &queue->msgs[queue->cur]; 4025 if (crq->generic.first & IBMVNIC_CRQ_CMD_RSP) { 4026 if (++queue->cur == queue->size) 4027 queue->cur = 0; 4028 } else { 4029 crq = NULL; 4030 } 4031 4032 return crq; 4033 } 4034 4035 static void print_subcrq_error(struct device *dev, int rc, const char *func) 4036 { 4037 switch (rc) { 4038 case H_PARAMETER: 4039 dev_warn_ratelimited(dev, 4040 "%s failed: Send request is malformed or adapter failover pending. (rc=%d)\n", 4041 func, rc); 4042 break; 4043 case H_CLOSED: 4044 dev_warn_ratelimited(dev, 4045 "%s failed: Backing queue closed. Adapter is down or failover pending. (rc=%d)\n", 4046 func, rc); 4047 break; 4048 default: 4049 dev_err_ratelimited(dev, "%s failed: (rc=%d)\n", func, rc); 4050 break; 4051 } 4052 } 4053 4054 static int send_subcrq_indirect(struct ibmvnic_adapter *adapter, 4055 u64 remote_handle, u64 ioba, u64 num_entries) 4056 { 4057 unsigned int ua = adapter->vdev->unit_address; 4058 struct device *dev = &adapter->vdev->dev; 4059 int rc; 4060 4061 /* Make sure the hypervisor sees the complete request */ 4062 dma_wmb(); 4063 rc = plpar_hcall_norets(H_SEND_SUB_CRQ_INDIRECT, ua, 4064 cpu_to_be64(remote_handle), 4065 ioba, num_entries); 4066 4067 if (rc) 4068 print_subcrq_error(dev, rc, __func__); 4069 4070 return rc; 4071 } 4072 4073 static int ibmvnic_send_crq(struct ibmvnic_adapter *adapter, 4074 union ibmvnic_crq *crq) 4075 { 4076 unsigned int ua = adapter->vdev->unit_address; 4077 struct device *dev = &adapter->vdev->dev; 4078 u64 *u64_crq = (u64 *)crq; 4079 int rc; 4080 4081 netdev_dbg(adapter->netdev, "Sending CRQ: %016lx %016lx\n", 4082 (unsigned long)cpu_to_be64(u64_crq[0]), 4083 (unsigned long)cpu_to_be64(u64_crq[1])); 4084 4085 if (!adapter->crq.active && 4086 crq->generic.first != IBMVNIC_CRQ_INIT_CMD) { 4087 dev_warn(dev, "Invalid request detected while CRQ is inactive, possible device state change during reset\n"); 4088 return -EINVAL; 4089 } 4090 4091 /* Make sure the hypervisor sees the complete request */ 4092 dma_wmb(); 4093 4094 rc = plpar_hcall_norets(H_SEND_CRQ, ua, 4095 cpu_to_be64(u64_crq[0]), 4096 cpu_to_be64(u64_crq[1])); 4097 4098 if (rc) { 4099 if (rc == H_CLOSED) { 4100 dev_warn(dev, "CRQ Queue closed\n"); 4101 /* do not reset, report the fail, wait for passive init from server */ 4102 } 4103 4104 dev_warn(dev, "Send error (rc=%d)\n", rc); 4105 } 4106 4107 return rc; 4108 } 4109 4110 static int ibmvnic_send_crq_init(struct ibmvnic_adapter *adapter) 4111 { 4112 struct device *dev = &adapter->vdev->dev; 4113 union ibmvnic_crq crq; 4114 int retries = 100; 4115 int rc; 4116 4117 memset(&crq, 0, sizeof(crq)); 4118 crq.generic.first = IBMVNIC_CRQ_INIT_CMD; 4119 crq.generic.cmd = IBMVNIC_CRQ_INIT; 4120 netdev_dbg(adapter->netdev, "Sending CRQ init\n"); 4121 4122 do { 4123 rc = ibmvnic_send_crq(adapter, &crq); 4124 if (rc != H_CLOSED) 4125 break; 4126 retries--; 4127 msleep(50); 4128 4129 } while (retries > 0); 4130 4131 if (rc) { 4132 dev_err(dev, "Failed to send init request, rc = %d\n", rc); 4133 return rc; 4134 } 4135 4136 return 0; 4137 } 4138 4139 struct vnic_login_client_data { 4140 u8 type; 4141 __be16 len; 4142 char name[]; 4143 } __packed; 4144 4145 static int vnic_client_data_len(struct ibmvnic_adapter *adapter) 4146 { 4147 int len; 4148 4149 /* Calculate the amount of buffer space needed for the 4150 * vnic client data in the login buffer. There are four entries, 4151 * OS name, LPAR name, device name, and a null last entry. 4152 */ 4153 len = 4 * sizeof(struct vnic_login_client_data); 4154 len += 6; /* "Linux" plus NULL */ 4155 len += strlen(utsname()->nodename) + 1; 4156 len += strlen(adapter->netdev->name) + 1; 4157 4158 return len; 4159 } 4160 4161 static void vnic_add_client_data(struct ibmvnic_adapter *adapter, 4162 struct vnic_login_client_data *vlcd) 4163 { 4164 const char *os_name = "Linux"; 4165 int len; 4166 4167 /* Type 1 - LPAR OS */ 4168 vlcd->type = 1; 4169 len = strlen(os_name) + 1; 4170 vlcd->len = cpu_to_be16(len); 4171 strscpy(vlcd->name, os_name, len); 4172 vlcd = (struct vnic_login_client_data *)(vlcd->name + len); 4173 4174 /* Type 2 - LPAR name */ 4175 vlcd->type = 2; 4176 len = strlen(utsname()->nodename) + 1; 4177 vlcd->len = cpu_to_be16(len); 4178 strscpy(vlcd->name, utsname()->nodename, len); 4179 vlcd = (struct vnic_login_client_data *)(vlcd->name + len); 4180 4181 /* Type 3 - device name */ 4182 vlcd->type = 3; 4183 len = strlen(adapter->netdev->name) + 1; 4184 vlcd->len = cpu_to_be16(len); 4185 strscpy(vlcd->name, adapter->netdev->name, len); 4186 } 4187 4188 static int send_login(struct ibmvnic_adapter *adapter) 4189 { 4190 struct ibmvnic_login_rsp_buffer *login_rsp_buffer; 4191 struct ibmvnic_login_buffer *login_buffer; 4192 struct device *dev = &adapter->vdev->dev; 4193 struct vnic_login_client_data *vlcd; 4194 dma_addr_t rsp_buffer_token; 4195 dma_addr_t buffer_token; 4196 size_t rsp_buffer_size; 4197 union ibmvnic_crq crq; 4198 int client_data_len; 4199 size_t buffer_size; 4200 __be64 *tx_list_p; 4201 __be64 *rx_list_p; 4202 int rc; 4203 int i; 4204 4205 if (!adapter->tx_scrq || !adapter->rx_scrq) { 4206 netdev_err(adapter->netdev, 4207 "RX or TX queues are not allocated, device login failed\n"); 4208 return -1; 4209 } 4210 4211 release_login_buffer(adapter); 4212 release_login_rsp_buffer(adapter); 4213 4214 client_data_len = vnic_client_data_len(adapter); 4215 4216 buffer_size = 4217 sizeof(struct ibmvnic_login_buffer) + 4218 sizeof(u64) * (adapter->req_tx_queues + adapter->req_rx_queues) + 4219 client_data_len; 4220 4221 login_buffer = kzalloc(buffer_size, GFP_ATOMIC); 4222 if (!login_buffer) 4223 goto buf_alloc_failed; 4224 4225 buffer_token = dma_map_single(dev, login_buffer, buffer_size, 4226 DMA_TO_DEVICE); 4227 if (dma_mapping_error(dev, buffer_token)) { 4228 dev_err(dev, "Couldn't map login buffer\n"); 4229 goto buf_map_failed; 4230 } 4231 4232 rsp_buffer_size = sizeof(struct ibmvnic_login_rsp_buffer) + 4233 sizeof(u64) * adapter->req_tx_queues + 4234 sizeof(u64) * adapter->req_rx_queues + 4235 sizeof(u64) * adapter->req_rx_queues + 4236 sizeof(u8) * IBMVNIC_TX_DESC_VERSIONS; 4237 4238 login_rsp_buffer = kmalloc(rsp_buffer_size, GFP_ATOMIC); 4239 if (!login_rsp_buffer) 4240 goto buf_rsp_alloc_failed; 4241 4242 rsp_buffer_token = dma_map_single(dev, login_rsp_buffer, 4243 rsp_buffer_size, DMA_FROM_DEVICE); 4244 if (dma_mapping_error(dev, rsp_buffer_token)) { 4245 dev_err(dev, "Couldn't map login rsp buffer\n"); 4246 goto buf_rsp_map_failed; 4247 } 4248 4249 adapter->login_buf = login_buffer; 4250 adapter->login_buf_token = buffer_token; 4251 adapter->login_buf_sz = buffer_size; 4252 adapter->login_rsp_buf = login_rsp_buffer; 4253 adapter->login_rsp_buf_token = rsp_buffer_token; 4254 adapter->login_rsp_buf_sz = rsp_buffer_size; 4255 4256 login_buffer->len = cpu_to_be32(buffer_size); 4257 login_buffer->version = cpu_to_be32(INITIAL_VERSION_LB); 4258 login_buffer->num_txcomp_subcrqs = cpu_to_be32(adapter->req_tx_queues); 4259 login_buffer->off_txcomp_subcrqs = 4260 cpu_to_be32(sizeof(struct ibmvnic_login_buffer)); 4261 login_buffer->num_rxcomp_subcrqs = cpu_to_be32(adapter->req_rx_queues); 4262 login_buffer->off_rxcomp_subcrqs = 4263 cpu_to_be32(sizeof(struct ibmvnic_login_buffer) + 4264 sizeof(u64) * adapter->req_tx_queues); 4265 login_buffer->login_rsp_ioba = cpu_to_be32(rsp_buffer_token); 4266 login_buffer->login_rsp_len = cpu_to_be32(rsp_buffer_size); 4267 4268 tx_list_p = (__be64 *)((char *)login_buffer + 4269 sizeof(struct ibmvnic_login_buffer)); 4270 rx_list_p = (__be64 *)((char *)login_buffer + 4271 sizeof(struct ibmvnic_login_buffer) + 4272 sizeof(u64) * adapter->req_tx_queues); 4273 4274 for (i = 0; i < adapter->req_tx_queues; i++) { 4275 if (adapter->tx_scrq[i]) { 4276 tx_list_p[i] = 4277 cpu_to_be64(adapter->tx_scrq[i]->crq_num); 4278 } 4279 } 4280 4281 for (i = 0; i < adapter->req_rx_queues; i++) { 4282 if (adapter->rx_scrq[i]) { 4283 rx_list_p[i] = 4284 cpu_to_be64(adapter->rx_scrq[i]->crq_num); 4285 } 4286 } 4287 4288 /* Insert vNIC login client data */ 4289 vlcd = (struct vnic_login_client_data *) 4290 ((char *)rx_list_p + (sizeof(u64) * adapter->req_rx_queues)); 4291 login_buffer->client_data_offset = 4292 cpu_to_be32((char *)vlcd - (char *)login_buffer); 4293 login_buffer->client_data_len = cpu_to_be32(client_data_len); 4294 4295 vnic_add_client_data(adapter, vlcd); 4296 4297 netdev_dbg(adapter->netdev, "Login Buffer:\n"); 4298 for (i = 0; i < (adapter->login_buf_sz - 1) / 8 + 1; i++) { 4299 netdev_dbg(adapter->netdev, "%016lx\n", 4300 ((unsigned long *)(adapter->login_buf))[i]); 4301 } 4302 4303 memset(&crq, 0, sizeof(crq)); 4304 crq.login.first = IBMVNIC_CRQ_CMD; 4305 crq.login.cmd = LOGIN; 4306 crq.login.ioba = cpu_to_be32(buffer_token); 4307 crq.login.len = cpu_to_be32(buffer_size); 4308 4309 adapter->login_pending = true; 4310 rc = ibmvnic_send_crq(adapter, &crq); 4311 if (rc) { 4312 adapter->login_pending = false; 4313 netdev_err(adapter->netdev, "Failed to send login, rc=%d\n", rc); 4314 goto buf_rsp_map_failed; 4315 } 4316 4317 return 0; 4318 4319 buf_rsp_map_failed: 4320 kfree(login_rsp_buffer); 4321 adapter->login_rsp_buf = NULL; 4322 buf_rsp_alloc_failed: 4323 dma_unmap_single(dev, buffer_token, buffer_size, DMA_TO_DEVICE); 4324 buf_map_failed: 4325 kfree(login_buffer); 4326 adapter->login_buf = NULL; 4327 buf_alloc_failed: 4328 return -1; 4329 } 4330 4331 static int send_request_map(struct ibmvnic_adapter *adapter, dma_addr_t addr, 4332 u32 len, u8 map_id) 4333 { 4334 union ibmvnic_crq crq; 4335 4336 memset(&crq, 0, sizeof(crq)); 4337 crq.request_map.first = IBMVNIC_CRQ_CMD; 4338 crq.request_map.cmd = REQUEST_MAP; 4339 crq.request_map.map_id = map_id; 4340 crq.request_map.ioba = cpu_to_be32(addr); 4341 crq.request_map.len = cpu_to_be32(len); 4342 return ibmvnic_send_crq(adapter, &crq); 4343 } 4344 4345 static int send_request_unmap(struct ibmvnic_adapter *adapter, u8 map_id) 4346 { 4347 union ibmvnic_crq crq; 4348 4349 memset(&crq, 0, sizeof(crq)); 4350 crq.request_unmap.first = IBMVNIC_CRQ_CMD; 4351 crq.request_unmap.cmd = REQUEST_UNMAP; 4352 crq.request_unmap.map_id = map_id; 4353 return ibmvnic_send_crq(adapter, &crq); 4354 } 4355 4356 static void send_query_map(struct ibmvnic_adapter *adapter) 4357 { 4358 union ibmvnic_crq crq; 4359 4360 memset(&crq, 0, sizeof(crq)); 4361 crq.query_map.first = IBMVNIC_CRQ_CMD; 4362 crq.query_map.cmd = QUERY_MAP; 4363 ibmvnic_send_crq(adapter, &crq); 4364 } 4365 4366 /* Send a series of CRQs requesting various capabilities of the VNIC server */ 4367 static void send_query_cap(struct ibmvnic_adapter *adapter) 4368 { 4369 union ibmvnic_crq crq; 4370 4371 atomic_set(&adapter->running_cap_crqs, 0); 4372 memset(&crq, 0, sizeof(crq)); 4373 crq.query_capability.first = IBMVNIC_CRQ_CMD; 4374 crq.query_capability.cmd = QUERY_CAPABILITY; 4375 4376 crq.query_capability.capability = cpu_to_be16(MIN_TX_QUEUES); 4377 atomic_inc(&adapter->running_cap_crqs); 4378 ibmvnic_send_crq(adapter, &crq); 4379 4380 crq.query_capability.capability = cpu_to_be16(MIN_RX_QUEUES); 4381 atomic_inc(&adapter->running_cap_crqs); 4382 ibmvnic_send_crq(adapter, &crq); 4383 4384 crq.query_capability.capability = cpu_to_be16(MIN_RX_ADD_QUEUES); 4385 atomic_inc(&adapter->running_cap_crqs); 4386 ibmvnic_send_crq(adapter, &crq); 4387 4388 crq.query_capability.capability = cpu_to_be16(MAX_TX_QUEUES); 4389 atomic_inc(&adapter->running_cap_crqs); 4390 ibmvnic_send_crq(adapter, &crq); 4391 4392 crq.query_capability.capability = cpu_to_be16(MAX_RX_QUEUES); 4393 atomic_inc(&adapter->running_cap_crqs); 4394 ibmvnic_send_crq(adapter, &crq); 4395 4396 crq.query_capability.capability = cpu_to_be16(MAX_RX_ADD_QUEUES); 4397 atomic_inc(&adapter->running_cap_crqs); 4398 ibmvnic_send_crq(adapter, &crq); 4399 4400 crq.query_capability.capability = 4401 cpu_to_be16(MIN_TX_ENTRIES_PER_SUBCRQ); 4402 atomic_inc(&adapter->running_cap_crqs); 4403 ibmvnic_send_crq(adapter, &crq); 4404 4405 crq.query_capability.capability = 4406 cpu_to_be16(MIN_RX_ADD_ENTRIES_PER_SUBCRQ); 4407 atomic_inc(&adapter->running_cap_crqs); 4408 ibmvnic_send_crq(adapter, &crq); 4409 4410 crq.query_capability.capability = 4411 cpu_to_be16(MAX_TX_ENTRIES_PER_SUBCRQ); 4412 atomic_inc(&adapter->running_cap_crqs); 4413 ibmvnic_send_crq(adapter, &crq); 4414 4415 crq.query_capability.capability = 4416 cpu_to_be16(MAX_RX_ADD_ENTRIES_PER_SUBCRQ); 4417 atomic_inc(&adapter->running_cap_crqs); 4418 ibmvnic_send_crq(adapter, &crq); 4419 4420 crq.query_capability.capability = cpu_to_be16(TCP_IP_OFFLOAD); 4421 atomic_inc(&adapter->running_cap_crqs); 4422 ibmvnic_send_crq(adapter, &crq); 4423 4424 crq.query_capability.capability = cpu_to_be16(PROMISC_SUPPORTED); 4425 atomic_inc(&adapter->running_cap_crqs); 4426 ibmvnic_send_crq(adapter, &crq); 4427 4428 crq.query_capability.capability = cpu_to_be16(MIN_MTU); 4429 atomic_inc(&adapter->running_cap_crqs); 4430 ibmvnic_send_crq(adapter, &crq); 4431 4432 crq.query_capability.capability = cpu_to_be16(MAX_MTU); 4433 atomic_inc(&adapter->running_cap_crqs); 4434 ibmvnic_send_crq(adapter, &crq); 4435 4436 crq.query_capability.capability = cpu_to_be16(MAX_MULTICAST_FILTERS); 4437 atomic_inc(&adapter->running_cap_crqs); 4438 ibmvnic_send_crq(adapter, &crq); 4439 4440 crq.query_capability.capability = cpu_to_be16(VLAN_HEADER_INSERTION); 4441 atomic_inc(&adapter->running_cap_crqs); 4442 ibmvnic_send_crq(adapter, &crq); 4443 4444 crq.query_capability.capability = cpu_to_be16(RX_VLAN_HEADER_INSERTION); 4445 atomic_inc(&adapter->running_cap_crqs); 4446 ibmvnic_send_crq(adapter, &crq); 4447 4448 crq.query_capability.capability = cpu_to_be16(MAX_TX_SG_ENTRIES); 4449 atomic_inc(&adapter->running_cap_crqs); 4450 ibmvnic_send_crq(adapter, &crq); 4451 4452 crq.query_capability.capability = cpu_to_be16(RX_SG_SUPPORTED); 4453 atomic_inc(&adapter->running_cap_crqs); 4454 ibmvnic_send_crq(adapter, &crq); 4455 4456 crq.query_capability.capability = cpu_to_be16(OPT_TX_COMP_SUB_QUEUES); 4457 atomic_inc(&adapter->running_cap_crqs); 4458 ibmvnic_send_crq(adapter, &crq); 4459 4460 crq.query_capability.capability = cpu_to_be16(OPT_RX_COMP_QUEUES); 4461 atomic_inc(&adapter->running_cap_crqs); 4462 ibmvnic_send_crq(adapter, &crq); 4463 4464 crq.query_capability.capability = 4465 cpu_to_be16(OPT_RX_BUFADD_Q_PER_RX_COMP_Q); 4466 atomic_inc(&adapter->running_cap_crqs); 4467 ibmvnic_send_crq(adapter, &crq); 4468 4469 crq.query_capability.capability = 4470 cpu_to_be16(OPT_TX_ENTRIES_PER_SUBCRQ); 4471 atomic_inc(&adapter->running_cap_crqs); 4472 ibmvnic_send_crq(adapter, &crq); 4473 4474 crq.query_capability.capability = 4475 cpu_to_be16(OPT_RXBA_ENTRIES_PER_SUBCRQ); 4476 atomic_inc(&adapter->running_cap_crqs); 4477 ibmvnic_send_crq(adapter, &crq); 4478 4479 crq.query_capability.capability = cpu_to_be16(TX_RX_DESC_REQ); 4480 atomic_inc(&adapter->running_cap_crqs); 4481 ibmvnic_send_crq(adapter, &crq); 4482 } 4483 4484 static void send_query_ip_offload(struct ibmvnic_adapter *adapter) 4485 { 4486 int buf_sz = sizeof(struct ibmvnic_query_ip_offload_buffer); 4487 struct device *dev = &adapter->vdev->dev; 4488 union ibmvnic_crq crq; 4489 4490 adapter->ip_offload_tok = 4491 dma_map_single(dev, 4492 &adapter->ip_offload_buf, 4493 buf_sz, 4494 DMA_FROM_DEVICE); 4495 4496 if (dma_mapping_error(dev, adapter->ip_offload_tok)) { 4497 if (!firmware_has_feature(FW_FEATURE_CMO)) 4498 dev_err(dev, "Couldn't map offload buffer\n"); 4499 return; 4500 } 4501 4502 memset(&crq, 0, sizeof(crq)); 4503 crq.query_ip_offload.first = IBMVNIC_CRQ_CMD; 4504 crq.query_ip_offload.cmd = QUERY_IP_OFFLOAD; 4505 crq.query_ip_offload.len = cpu_to_be32(buf_sz); 4506 crq.query_ip_offload.ioba = 4507 cpu_to_be32(adapter->ip_offload_tok); 4508 4509 ibmvnic_send_crq(adapter, &crq); 4510 } 4511 4512 static void send_control_ip_offload(struct ibmvnic_adapter *adapter) 4513 { 4514 struct ibmvnic_control_ip_offload_buffer *ctrl_buf = &adapter->ip_offload_ctrl; 4515 struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf; 4516 struct device *dev = &adapter->vdev->dev; 4517 netdev_features_t old_hw_features = 0; 4518 union ibmvnic_crq crq; 4519 4520 adapter->ip_offload_ctrl_tok = 4521 dma_map_single(dev, 4522 ctrl_buf, 4523 sizeof(adapter->ip_offload_ctrl), 4524 DMA_TO_DEVICE); 4525 4526 if (dma_mapping_error(dev, adapter->ip_offload_ctrl_tok)) { 4527 dev_err(dev, "Couldn't map ip offload control buffer\n"); 4528 return; 4529 } 4530 4531 ctrl_buf->len = cpu_to_be32(sizeof(adapter->ip_offload_ctrl)); 4532 ctrl_buf->version = cpu_to_be32(INITIAL_VERSION_IOB); 4533 ctrl_buf->ipv4_chksum = buf->ipv4_chksum; 4534 ctrl_buf->ipv6_chksum = buf->ipv6_chksum; 4535 ctrl_buf->tcp_ipv4_chksum = buf->tcp_ipv4_chksum; 4536 ctrl_buf->udp_ipv4_chksum = buf->udp_ipv4_chksum; 4537 ctrl_buf->tcp_ipv6_chksum = buf->tcp_ipv6_chksum; 4538 ctrl_buf->udp_ipv6_chksum = buf->udp_ipv6_chksum; 4539 ctrl_buf->large_tx_ipv4 = buf->large_tx_ipv4; 4540 ctrl_buf->large_tx_ipv6 = buf->large_tx_ipv6; 4541 4542 /* large_rx disabled for now, additional features needed */ 4543 ctrl_buf->large_rx_ipv4 = 0; 4544 ctrl_buf->large_rx_ipv6 = 0; 4545 4546 if (adapter->state != VNIC_PROBING) { 4547 old_hw_features = adapter->netdev->hw_features; 4548 adapter->netdev->hw_features = 0; 4549 } 4550 4551 adapter->netdev->hw_features = NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO; 4552 4553 if (buf->tcp_ipv4_chksum || buf->udp_ipv4_chksum) 4554 adapter->netdev->hw_features |= NETIF_F_IP_CSUM; 4555 4556 if (buf->tcp_ipv6_chksum || buf->udp_ipv6_chksum) 4557 adapter->netdev->hw_features |= NETIF_F_IPV6_CSUM; 4558 4559 if ((adapter->netdev->features & 4560 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM))) 4561 adapter->netdev->hw_features |= NETIF_F_RXCSUM; 4562 4563 if (buf->large_tx_ipv4) 4564 adapter->netdev->hw_features |= NETIF_F_TSO; 4565 if (buf->large_tx_ipv6) 4566 adapter->netdev->hw_features |= NETIF_F_TSO6; 4567 4568 if (adapter->state == VNIC_PROBING) { 4569 adapter->netdev->features |= adapter->netdev->hw_features; 4570 } else if (old_hw_features != adapter->netdev->hw_features) { 4571 netdev_features_t tmp = 0; 4572 4573 /* disable features no longer supported */ 4574 adapter->netdev->features &= adapter->netdev->hw_features; 4575 /* turn on features now supported if previously enabled */ 4576 tmp = (old_hw_features ^ adapter->netdev->hw_features) & 4577 adapter->netdev->hw_features; 4578 adapter->netdev->features |= 4579 tmp & adapter->netdev->wanted_features; 4580 } 4581 4582 memset(&crq, 0, sizeof(crq)); 4583 crq.control_ip_offload.first = IBMVNIC_CRQ_CMD; 4584 crq.control_ip_offload.cmd = CONTROL_IP_OFFLOAD; 4585 crq.control_ip_offload.len = 4586 cpu_to_be32(sizeof(adapter->ip_offload_ctrl)); 4587 crq.control_ip_offload.ioba = cpu_to_be32(adapter->ip_offload_ctrl_tok); 4588 ibmvnic_send_crq(adapter, &crq); 4589 } 4590 4591 static void handle_vpd_size_rsp(union ibmvnic_crq *crq, 4592 struct ibmvnic_adapter *adapter) 4593 { 4594 struct device *dev = &adapter->vdev->dev; 4595 4596 if (crq->get_vpd_size_rsp.rc.code) { 4597 dev_err(dev, "Error retrieving VPD size, rc=%x\n", 4598 crq->get_vpd_size_rsp.rc.code); 4599 complete(&adapter->fw_done); 4600 return; 4601 } 4602 4603 adapter->vpd->len = be64_to_cpu(crq->get_vpd_size_rsp.len); 4604 complete(&adapter->fw_done); 4605 } 4606 4607 static void handle_vpd_rsp(union ibmvnic_crq *crq, 4608 struct ibmvnic_adapter *adapter) 4609 { 4610 struct device *dev = &adapter->vdev->dev; 4611 unsigned char *substr = NULL; 4612 u8 fw_level_len = 0; 4613 4614 memset(adapter->fw_version, 0, 32); 4615 4616 dma_unmap_single(dev, adapter->vpd->dma_addr, adapter->vpd->len, 4617 DMA_FROM_DEVICE); 4618 4619 if (crq->get_vpd_rsp.rc.code) { 4620 dev_err(dev, "Error retrieving VPD from device, rc=%x\n", 4621 crq->get_vpd_rsp.rc.code); 4622 goto complete; 4623 } 4624 4625 /* get the position of the firmware version info 4626 * located after the ASCII 'RM' substring in the buffer 4627 */ 4628 substr = strnstr(adapter->vpd->buff, "RM", adapter->vpd->len); 4629 if (!substr) { 4630 dev_info(dev, "Warning - No FW level has been provided in the VPD buffer by the VIOS Server\n"); 4631 goto complete; 4632 } 4633 4634 /* get length of firmware level ASCII substring */ 4635 if ((substr + 2) < (adapter->vpd->buff + adapter->vpd->len)) { 4636 fw_level_len = *(substr + 2); 4637 } else { 4638 dev_info(dev, "Length of FW substr extrapolated VDP buff\n"); 4639 goto complete; 4640 } 4641 4642 /* copy firmware version string from vpd into adapter */ 4643 if ((substr + 3 + fw_level_len) < 4644 (adapter->vpd->buff + adapter->vpd->len)) { 4645 strncpy((char *)adapter->fw_version, substr + 3, fw_level_len); 4646 } else { 4647 dev_info(dev, "FW substr extrapolated VPD buff\n"); 4648 } 4649 4650 complete: 4651 if (adapter->fw_version[0] == '\0') 4652 strscpy((char *)adapter->fw_version, "N/A", sizeof(adapter->fw_version)); 4653 complete(&adapter->fw_done); 4654 } 4655 4656 static void handle_query_ip_offload_rsp(struct ibmvnic_adapter *adapter) 4657 { 4658 struct device *dev = &adapter->vdev->dev; 4659 struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf; 4660 int i; 4661 4662 dma_unmap_single(dev, adapter->ip_offload_tok, 4663 sizeof(adapter->ip_offload_buf), DMA_FROM_DEVICE); 4664 4665 netdev_dbg(adapter->netdev, "Query IP Offload Buffer:\n"); 4666 for (i = 0; i < (sizeof(adapter->ip_offload_buf) - 1) / 8 + 1; i++) 4667 netdev_dbg(adapter->netdev, "%016lx\n", 4668 ((unsigned long *)(buf))[i]); 4669 4670 netdev_dbg(adapter->netdev, "ipv4_chksum = %d\n", buf->ipv4_chksum); 4671 netdev_dbg(adapter->netdev, "ipv6_chksum = %d\n", buf->ipv6_chksum); 4672 netdev_dbg(adapter->netdev, "tcp_ipv4_chksum = %d\n", 4673 buf->tcp_ipv4_chksum); 4674 netdev_dbg(adapter->netdev, "tcp_ipv6_chksum = %d\n", 4675 buf->tcp_ipv6_chksum); 4676 netdev_dbg(adapter->netdev, "udp_ipv4_chksum = %d\n", 4677 buf->udp_ipv4_chksum); 4678 netdev_dbg(adapter->netdev, "udp_ipv6_chksum = %d\n", 4679 buf->udp_ipv6_chksum); 4680 netdev_dbg(adapter->netdev, "large_tx_ipv4 = %d\n", 4681 buf->large_tx_ipv4); 4682 netdev_dbg(adapter->netdev, "large_tx_ipv6 = %d\n", 4683 buf->large_tx_ipv6); 4684 netdev_dbg(adapter->netdev, "large_rx_ipv4 = %d\n", 4685 buf->large_rx_ipv4); 4686 netdev_dbg(adapter->netdev, "large_rx_ipv6 = %d\n", 4687 buf->large_rx_ipv6); 4688 netdev_dbg(adapter->netdev, "max_ipv4_hdr_sz = %d\n", 4689 buf->max_ipv4_header_size); 4690 netdev_dbg(adapter->netdev, "max_ipv6_hdr_sz = %d\n", 4691 buf->max_ipv6_header_size); 4692 netdev_dbg(adapter->netdev, "max_tcp_hdr_size = %d\n", 4693 buf->max_tcp_header_size); 4694 netdev_dbg(adapter->netdev, "max_udp_hdr_size = %d\n", 4695 buf->max_udp_header_size); 4696 netdev_dbg(adapter->netdev, "max_large_tx_size = %d\n", 4697 buf->max_large_tx_size); 4698 netdev_dbg(adapter->netdev, "max_large_rx_size = %d\n", 4699 buf->max_large_rx_size); 4700 netdev_dbg(adapter->netdev, "ipv6_ext_hdr = %d\n", 4701 buf->ipv6_extension_header); 4702 netdev_dbg(adapter->netdev, "tcp_pseudosum_req = %d\n", 4703 buf->tcp_pseudosum_req); 4704 netdev_dbg(adapter->netdev, "num_ipv6_ext_hd = %d\n", 4705 buf->num_ipv6_ext_headers); 4706 netdev_dbg(adapter->netdev, "off_ipv6_ext_hd = %d\n", 4707 buf->off_ipv6_ext_headers); 4708 4709 send_control_ip_offload(adapter); 4710 } 4711 4712 static const char *ibmvnic_fw_err_cause(u16 cause) 4713 { 4714 switch (cause) { 4715 case ADAPTER_PROBLEM: 4716 return "adapter problem"; 4717 case BUS_PROBLEM: 4718 return "bus problem"; 4719 case FW_PROBLEM: 4720 return "firmware problem"; 4721 case DD_PROBLEM: 4722 return "device driver problem"; 4723 case EEH_RECOVERY: 4724 return "EEH recovery"; 4725 case FW_UPDATED: 4726 return "firmware updated"; 4727 case LOW_MEMORY: 4728 return "low Memory"; 4729 default: 4730 return "unknown"; 4731 } 4732 } 4733 4734 static void handle_error_indication(union ibmvnic_crq *crq, 4735 struct ibmvnic_adapter *adapter) 4736 { 4737 struct device *dev = &adapter->vdev->dev; 4738 u16 cause; 4739 4740 cause = be16_to_cpu(crq->error_indication.error_cause); 4741 4742 dev_warn_ratelimited(dev, 4743 "Firmware reports %serror, cause: %s. Starting recovery...\n", 4744 crq->error_indication.flags 4745 & IBMVNIC_FATAL_ERROR ? "FATAL " : "", 4746 ibmvnic_fw_err_cause(cause)); 4747 4748 if (crq->error_indication.flags & IBMVNIC_FATAL_ERROR) 4749 ibmvnic_reset(adapter, VNIC_RESET_FATAL); 4750 else 4751 ibmvnic_reset(adapter, VNIC_RESET_NON_FATAL); 4752 } 4753 4754 static int handle_change_mac_rsp(union ibmvnic_crq *crq, 4755 struct ibmvnic_adapter *adapter) 4756 { 4757 struct net_device *netdev = adapter->netdev; 4758 struct device *dev = &adapter->vdev->dev; 4759 long rc; 4760 4761 rc = crq->change_mac_addr_rsp.rc.code; 4762 if (rc) { 4763 dev_err(dev, "Error %ld in CHANGE_MAC_ADDR_RSP\n", rc); 4764 goto out; 4765 } 4766 /* crq->change_mac_addr.mac_addr is the requested one 4767 * crq->change_mac_addr_rsp.mac_addr is the returned valid one. 4768 */ 4769 eth_hw_addr_set(netdev, &crq->change_mac_addr_rsp.mac_addr[0]); 4770 ether_addr_copy(adapter->mac_addr, 4771 &crq->change_mac_addr_rsp.mac_addr[0]); 4772 out: 4773 complete(&adapter->fw_done); 4774 return rc; 4775 } 4776 4777 static void handle_request_cap_rsp(union ibmvnic_crq *crq, 4778 struct ibmvnic_adapter *adapter) 4779 { 4780 struct device *dev = &adapter->vdev->dev; 4781 u64 *req_value; 4782 char *name; 4783 4784 atomic_dec(&adapter->running_cap_crqs); 4785 switch (be16_to_cpu(crq->request_capability_rsp.capability)) { 4786 case REQ_TX_QUEUES: 4787 req_value = &adapter->req_tx_queues; 4788 name = "tx"; 4789 break; 4790 case REQ_RX_QUEUES: 4791 req_value = &adapter->req_rx_queues; 4792 name = "rx"; 4793 break; 4794 case REQ_RX_ADD_QUEUES: 4795 req_value = &adapter->req_rx_add_queues; 4796 name = "rx_add"; 4797 break; 4798 case REQ_TX_ENTRIES_PER_SUBCRQ: 4799 req_value = &adapter->req_tx_entries_per_subcrq; 4800 name = "tx_entries_per_subcrq"; 4801 break; 4802 case REQ_RX_ADD_ENTRIES_PER_SUBCRQ: 4803 req_value = &adapter->req_rx_add_entries_per_subcrq; 4804 name = "rx_add_entries_per_subcrq"; 4805 break; 4806 case REQ_MTU: 4807 req_value = &adapter->req_mtu; 4808 name = "mtu"; 4809 break; 4810 case PROMISC_REQUESTED: 4811 req_value = &adapter->promisc; 4812 name = "promisc"; 4813 break; 4814 default: 4815 dev_err(dev, "Got invalid cap request rsp %d\n", 4816 crq->request_capability.capability); 4817 return; 4818 } 4819 4820 switch (crq->request_capability_rsp.rc.code) { 4821 case SUCCESS: 4822 break; 4823 case PARTIALSUCCESS: 4824 dev_info(dev, "req=%lld, rsp=%ld in %s queue, retrying.\n", 4825 *req_value, 4826 (long)be64_to_cpu(crq->request_capability_rsp.number), 4827 name); 4828 4829 if (be16_to_cpu(crq->request_capability_rsp.capability) == 4830 REQ_MTU) { 4831 pr_err("mtu of %llu is not supported. Reverting.\n", 4832 *req_value); 4833 *req_value = adapter->fallback.mtu; 4834 } else { 4835 *req_value = 4836 be64_to_cpu(crq->request_capability_rsp.number); 4837 } 4838 4839 send_request_cap(adapter, 1); 4840 return; 4841 default: 4842 dev_err(dev, "Error %d in request cap rsp\n", 4843 crq->request_capability_rsp.rc.code); 4844 return; 4845 } 4846 4847 /* Done receiving requested capabilities, query IP offload support */ 4848 if (atomic_read(&adapter->running_cap_crqs) == 0) { 4849 adapter->wait_capability = false; 4850 send_query_ip_offload(adapter); 4851 } 4852 } 4853 4854 static int handle_login_rsp(union ibmvnic_crq *login_rsp_crq, 4855 struct ibmvnic_adapter *adapter) 4856 { 4857 struct device *dev = &adapter->vdev->dev; 4858 struct net_device *netdev = adapter->netdev; 4859 struct ibmvnic_login_rsp_buffer *login_rsp = adapter->login_rsp_buf; 4860 struct ibmvnic_login_buffer *login = adapter->login_buf; 4861 u64 *tx_handle_array; 4862 u64 *rx_handle_array; 4863 int num_tx_pools; 4864 int num_rx_pools; 4865 u64 *size_array; 4866 int i; 4867 4868 /* CHECK: Test/set of login_pending does not need to be atomic 4869 * because only ibmvnic_tasklet tests/clears this. 4870 */ 4871 if (!adapter->login_pending) { 4872 netdev_warn(netdev, "Ignoring unexpected login response\n"); 4873 return 0; 4874 } 4875 adapter->login_pending = false; 4876 4877 dma_unmap_single(dev, adapter->login_buf_token, adapter->login_buf_sz, 4878 DMA_TO_DEVICE); 4879 dma_unmap_single(dev, adapter->login_rsp_buf_token, 4880 adapter->login_rsp_buf_sz, DMA_FROM_DEVICE); 4881 4882 /* If the number of queues requested can't be allocated by the 4883 * server, the login response will return with code 1. We will need 4884 * to resend the login buffer with fewer queues requested. 4885 */ 4886 if (login_rsp_crq->generic.rc.code) { 4887 adapter->init_done_rc = login_rsp_crq->generic.rc.code; 4888 complete(&adapter->init_done); 4889 return 0; 4890 } 4891 4892 if (adapter->failover_pending) { 4893 adapter->init_done_rc = -EAGAIN; 4894 netdev_dbg(netdev, "Failover pending, ignoring login response\n"); 4895 complete(&adapter->init_done); 4896 /* login response buffer will be released on reset */ 4897 return 0; 4898 } 4899 4900 netdev->mtu = adapter->req_mtu - ETH_HLEN; 4901 4902 netdev_dbg(adapter->netdev, "Login Response Buffer:\n"); 4903 for (i = 0; i < (adapter->login_rsp_buf_sz - 1) / 8 + 1; i++) { 4904 netdev_dbg(adapter->netdev, "%016lx\n", 4905 ((unsigned long *)(adapter->login_rsp_buf))[i]); 4906 } 4907 4908 /* Sanity checks */ 4909 if (login->num_txcomp_subcrqs != login_rsp->num_txsubm_subcrqs || 4910 (be32_to_cpu(login->num_rxcomp_subcrqs) * 4911 adapter->req_rx_add_queues != 4912 be32_to_cpu(login_rsp->num_rxadd_subcrqs))) { 4913 dev_err(dev, "FATAL: Inconsistent login and login rsp\n"); 4914 ibmvnic_reset(adapter, VNIC_RESET_FATAL); 4915 return -EIO; 4916 } 4917 size_array = (u64 *)((u8 *)(adapter->login_rsp_buf) + 4918 be32_to_cpu(adapter->login_rsp_buf->off_rxadd_buff_size)); 4919 /* variable buffer sizes are not supported, so just read the 4920 * first entry. 4921 */ 4922 adapter->cur_rx_buf_sz = be64_to_cpu(size_array[0]); 4923 4924 num_tx_pools = be32_to_cpu(adapter->login_rsp_buf->num_txsubm_subcrqs); 4925 num_rx_pools = be32_to_cpu(adapter->login_rsp_buf->num_rxadd_subcrqs); 4926 4927 tx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) + 4928 be32_to_cpu(adapter->login_rsp_buf->off_txsubm_subcrqs)); 4929 rx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) + 4930 be32_to_cpu(adapter->login_rsp_buf->off_rxadd_subcrqs)); 4931 4932 for (i = 0; i < num_tx_pools; i++) 4933 adapter->tx_scrq[i]->handle = tx_handle_array[i]; 4934 4935 for (i = 0; i < num_rx_pools; i++) 4936 adapter->rx_scrq[i]->handle = rx_handle_array[i]; 4937 4938 adapter->num_active_tx_scrqs = num_tx_pools; 4939 adapter->num_active_rx_scrqs = num_rx_pools; 4940 release_login_rsp_buffer(adapter); 4941 release_login_buffer(adapter); 4942 complete(&adapter->init_done); 4943 4944 return 0; 4945 } 4946 4947 static void handle_request_unmap_rsp(union ibmvnic_crq *crq, 4948 struct ibmvnic_adapter *adapter) 4949 { 4950 struct device *dev = &adapter->vdev->dev; 4951 long rc; 4952 4953 rc = crq->request_unmap_rsp.rc.code; 4954 if (rc) 4955 dev_err(dev, "Error %ld in REQUEST_UNMAP_RSP\n", rc); 4956 } 4957 4958 static void handle_query_map_rsp(union ibmvnic_crq *crq, 4959 struct ibmvnic_adapter *adapter) 4960 { 4961 struct net_device *netdev = adapter->netdev; 4962 struct device *dev = &adapter->vdev->dev; 4963 long rc; 4964 4965 rc = crq->query_map_rsp.rc.code; 4966 if (rc) { 4967 dev_err(dev, "Error %ld in QUERY_MAP_RSP\n", rc); 4968 return; 4969 } 4970 netdev_dbg(netdev, "page_size = %d\ntot_pages = %u\nfree_pages = %u\n", 4971 crq->query_map_rsp.page_size, 4972 __be32_to_cpu(crq->query_map_rsp.tot_pages), 4973 __be32_to_cpu(crq->query_map_rsp.free_pages)); 4974 } 4975 4976 static void handle_query_cap_rsp(union ibmvnic_crq *crq, 4977 struct ibmvnic_adapter *adapter) 4978 { 4979 struct net_device *netdev = adapter->netdev; 4980 struct device *dev = &adapter->vdev->dev; 4981 long rc; 4982 4983 atomic_dec(&adapter->running_cap_crqs); 4984 netdev_dbg(netdev, "Outstanding queries: %d\n", 4985 atomic_read(&adapter->running_cap_crqs)); 4986 rc = crq->query_capability.rc.code; 4987 if (rc) { 4988 dev_err(dev, "Error %ld in QUERY_CAP_RSP\n", rc); 4989 goto out; 4990 } 4991 4992 switch (be16_to_cpu(crq->query_capability.capability)) { 4993 case MIN_TX_QUEUES: 4994 adapter->min_tx_queues = 4995 be64_to_cpu(crq->query_capability.number); 4996 netdev_dbg(netdev, "min_tx_queues = %lld\n", 4997 adapter->min_tx_queues); 4998 break; 4999 case MIN_RX_QUEUES: 5000 adapter->min_rx_queues = 5001 be64_to_cpu(crq->query_capability.number); 5002 netdev_dbg(netdev, "min_rx_queues = %lld\n", 5003 adapter->min_rx_queues); 5004 break; 5005 case MIN_RX_ADD_QUEUES: 5006 adapter->min_rx_add_queues = 5007 be64_to_cpu(crq->query_capability.number); 5008 netdev_dbg(netdev, "min_rx_add_queues = %lld\n", 5009 adapter->min_rx_add_queues); 5010 break; 5011 case MAX_TX_QUEUES: 5012 adapter->max_tx_queues = 5013 be64_to_cpu(crq->query_capability.number); 5014 netdev_dbg(netdev, "max_tx_queues = %lld\n", 5015 adapter->max_tx_queues); 5016 break; 5017 case MAX_RX_QUEUES: 5018 adapter->max_rx_queues = 5019 be64_to_cpu(crq->query_capability.number); 5020 netdev_dbg(netdev, "max_rx_queues = %lld\n", 5021 adapter->max_rx_queues); 5022 break; 5023 case MAX_RX_ADD_QUEUES: 5024 adapter->max_rx_add_queues = 5025 be64_to_cpu(crq->query_capability.number); 5026 netdev_dbg(netdev, "max_rx_add_queues = %lld\n", 5027 adapter->max_rx_add_queues); 5028 break; 5029 case MIN_TX_ENTRIES_PER_SUBCRQ: 5030 adapter->min_tx_entries_per_subcrq = 5031 be64_to_cpu(crq->query_capability.number); 5032 netdev_dbg(netdev, "min_tx_entries_per_subcrq = %lld\n", 5033 adapter->min_tx_entries_per_subcrq); 5034 break; 5035 case MIN_RX_ADD_ENTRIES_PER_SUBCRQ: 5036 adapter->min_rx_add_entries_per_subcrq = 5037 be64_to_cpu(crq->query_capability.number); 5038 netdev_dbg(netdev, "min_rx_add_entrs_per_subcrq = %lld\n", 5039 adapter->min_rx_add_entries_per_subcrq); 5040 break; 5041 case MAX_TX_ENTRIES_PER_SUBCRQ: 5042 adapter->max_tx_entries_per_subcrq = 5043 be64_to_cpu(crq->query_capability.number); 5044 netdev_dbg(netdev, "max_tx_entries_per_subcrq = %lld\n", 5045 adapter->max_tx_entries_per_subcrq); 5046 break; 5047 case MAX_RX_ADD_ENTRIES_PER_SUBCRQ: 5048 adapter->max_rx_add_entries_per_subcrq = 5049 be64_to_cpu(crq->query_capability.number); 5050 netdev_dbg(netdev, "max_rx_add_entrs_per_subcrq = %lld\n", 5051 adapter->max_rx_add_entries_per_subcrq); 5052 break; 5053 case TCP_IP_OFFLOAD: 5054 adapter->tcp_ip_offload = 5055 be64_to_cpu(crq->query_capability.number); 5056 netdev_dbg(netdev, "tcp_ip_offload = %lld\n", 5057 adapter->tcp_ip_offload); 5058 break; 5059 case PROMISC_SUPPORTED: 5060 adapter->promisc_supported = 5061 be64_to_cpu(crq->query_capability.number); 5062 netdev_dbg(netdev, "promisc_supported = %lld\n", 5063 adapter->promisc_supported); 5064 break; 5065 case MIN_MTU: 5066 adapter->min_mtu = be64_to_cpu(crq->query_capability.number); 5067 netdev->min_mtu = adapter->min_mtu - ETH_HLEN; 5068 netdev_dbg(netdev, "min_mtu = %lld\n", adapter->min_mtu); 5069 break; 5070 case MAX_MTU: 5071 adapter->max_mtu = be64_to_cpu(crq->query_capability.number); 5072 netdev->max_mtu = adapter->max_mtu - ETH_HLEN; 5073 netdev_dbg(netdev, "max_mtu = %lld\n", adapter->max_mtu); 5074 break; 5075 case MAX_MULTICAST_FILTERS: 5076 adapter->max_multicast_filters = 5077 be64_to_cpu(crq->query_capability.number); 5078 netdev_dbg(netdev, "max_multicast_filters = %lld\n", 5079 adapter->max_multicast_filters); 5080 break; 5081 case VLAN_HEADER_INSERTION: 5082 adapter->vlan_header_insertion = 5083 be64_to_cpu(crq->query_capability.number); 5084 if (adapter->vlan_header_insertion) 5085 netdev->features |= NETIF_F_HW_VLAN_STAG_TX; 5086 netdev_dbg(netdev, "vlan_header_insertion = %lld\n", 5087 adapter->vlan_header_insertion); 5088 break; 5089 case RX_VLAN_HEADER_INSERTION: 5090 adapter->rx_vlan_header_insertion = 5091 be64_to_cpu(crq->query_capability.number); 5092 netdev_dbg(netdev, "rx_vlan_header_insertion = %lld\n", 5093 adapter->rx_vlan_header_insertion); 5094 break; 5095 case MAX_TX_SG_ENTRIES: 5096 adapter->max_tx_sg_entries = 5097 be64_to_cpu(crq->query_capability.number); 5098 netdev_dbg(netdev, "max_tx_sg_entries = %lld\n", 5099 adapter->max_tx_sg_entries); 5100 break; 5101 case RX_SG_SUPPORTED: 5102 adapter->rx_sg_supported = 5103 be64_to_cpu(crq->query_capability.number); 5104 netdev_dbg(netdev, "rx_sg_supported = %lld\n", 5105 adapter->rx_sg_supported); 5106 break; 5107 case OPT_TX_COMP_SUB_QUEUES: 5108 adapter->opt_tx_comp_sub_queues = 5109 be64_to_cpu(crq->query_capability.number); 5110 netdev_dbg(netdev, "opt_tx_comp_sub_queues = %lld\n", 5111 adapter->opt_tx_comp_sub_queues); 5112 break; 5113 case OPT_RX_COMP_QUEUES: 5114 adapter->opt_rx_comp_queues = 5115 be64_to_cpu(crq->query_capability.number); 5116 netdev_dbg(netdev, "opt_rx_comp_queues = %lld\n", 5117 adapter->opt_rx_comp_queues); 5118 break; 5119 case OPT_RX_BUFADD_Q_PER_RX_COMP_Q: 5120 adapter->opt_rx_bufadd_q_per_rx_comp_q = 5121 be64_to_cpu(crq->query_capability.number); 5122 netdev_dbg(netdev, "opt_rx_bufadd_q_per_rx_comp_q = %lld\n", 5123 adapter->opt_rx_bufadd_q_per_rx_comp_q); 5124 break; 5125 case OPT_TX_ENTRIES_PER_SUBCRQ: 5126 adapter->opt_tx_entries_per_subcrq = 5127 be64_to_cpu(crq->query_capability.number); 5128 netdev_dbg(netdev, "opt_tx_entries_per_subcrq = %lld\n", 5129 adapter->opt_tx_entries_per_subcrq); 5130 break; 5131 case OPT_RXBA_ENTRIES_PER_SUBCRQ: 5132 adapter->opt_rxba_entries_per_subcrq = 5133 be64_to_cpu(crq->query_capability.number); 5134 netdev_dbg(netdev, "opt_rxba_entries_per_subcrq = %lld\n", 5135 adapter->opt_rxba_entries_per_subcrq); 5136 break; 5137 case TX_RX_DESC_REQ: 5138 adapter->tx_rx_desc_req = crq->query_capability.number; 5139 netdev_dbg(netdev, "tx_rx_desc_req = %llx\n", 5140 adapter->tx_rx_desc_req); 5141 break; 5142 5143 default: 5144 netdev_err(netdev, "Got invalid cap rsp %d\n", 5145 crq->query_capability.capability); 5146 } 5147 5148 out: 5149 if (atomic_read(&adapter->running_cap_crqs) == 0) { 5150 adapter->wait_capability = false; 5151 send_request_cap(adapter, 0); 5152 } 5153 } 5154 5155 static int send_query_phys_parms(struct ibmvnic_adapter *adapter) 5156 { 5157 union ibmvnic_crq crq; 5158 int rc; 5159 5160 memset(&crq, 0, sizeof(crq)); 5161 crq.query_phys_parms.first = IBMVNIC_CRQ_CMD; 5162 crq.query_phys_parms.cmd = QUERY_PHYS_PARMS; 5163 5164 mutex_lock(&adapter->fw_lock); 5165 adapter->fw_done_rc = 0; 5166 reinit_completion(&adapter->fw_done); 5167 5168 rc = ibmvnic_send_crq(adapter, &crq); 5169 if (rc) { 5170 mutex_unlock(&adapter->fw_lock); 5171 return rc; 5172 } 5173 5174 rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000); 5175 if (rc) { 5176 mutex_unlock(&adapter->fw_lock); 5177 return rc; 5178 } 5179 5180 mutex_unlock(&adapter->fw_lock); 5181 return adapter->fw_done_rc ? -EIO : 0; 5182 } 5183 5184 static int handle_query_phys_parms_rsp(union ibmvnic_crq *crq, 5185 struct ibmvnic_adapter *adapter) 5186 { 5187 struct net_device *netdev = adapter->netdev; 5188 int rc; 5189 __be32 rspeed = cpu_to_be32(crq->query_phys_parms_rsp.speed); 5190 5191 rc = crq->query_phys_parms_rsp.rc.code; 5192 if (rc) { 5193 netdev_err(netdev, "Error %d in QUERY_PHYS_PARMS\n", rc); 5194 return rc; 5195 } 5196 switch (rspeed) { 5197 case IBMVNIC_10MBPS: 5198 adapter->speed = SPEED_10; 5199 break; 5200 case IBMVNIC_100MBPS: 5201 adapter->speed = SPEED_100; 5202 break; 5203 case IBMVNIC_1GBPS: 5204 adapter->speed = SPEED_1000; 5205 break; 5206 case IBMVNIC_10GBPS: 5207 adapter->speed = SPEED_10000; 5208 break; 5209 case IBMVNIC_25GBPS: 5210 adapter->speed = SPEED_25000; 5211 break; 5212 case IBMVNIC_40GBPS: 5213 adapter->speed = SPEED_40000; 5214 break; 5215 case IBMVNIC_50GBPS: 5216 adapter->speed = SPEED_50000; 5217 break; 5218 case IBMVNIC_100GBPS: 5219 adapter->speed = SPEED_100000; 5220 break; 5221 case IBMVNIC_200GBPS: 5222 adapter->speed = SPEED_200000; 5223 break; 5224 default: 5225 if (netif_carrier_ok(netdev)) 5226 netdev_warn(netdev, "Unknown speed 0x%08x\n", rspeed); 5227 adapter->speed = SPEED_UNKNOWN; 5228 } 5229 if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_FULL_DUPLEX) 5230 adapter->duplex = DUPLEX_FULL; 5231 else if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_HALF_DUPLEX) 5232 adapter->duplex = DUPLEX_HALF; 5233 else 5234 adapter->duplex = DUPLEX_UNKNOWN; 5235 5236 return rc; 5237 } 5238 5239 static void ibmvnic_handle_crq(union ibmvnic_crq *crq, 5240 struct ibmvnic_adapter *adapter) 5241 { 5242 struct ibmvnic_generic_crq *gen_crq = &crq->generic; 5243 struct net_device *netdev = adapter->netdev; 5244 struct device *dev = &adapter->vdev->dev; 5245 u64 *u64_crq = (u64 *)crq; 5246 long rc; 5247 5248 netdev_dbg(netdev, "Handling CRQ: %016lx %016lx\n", 5249 (unsigned long)cpu_to_be64(u64_crq[0]), 5250 (unsigned long)cpu_to_be64(u64_crq[1])); 5251 switch (gen_crq->first) { 5252 case IBMVNIC_CRQ_INIT_RSP: 5253 switch (gen_crq->cmd) { 5254 case IBMVNIC_CRQ_INIT: 5255 dev_info(dev, "Partner initialized\n"); 5256 adapter->from_passive_init = true; 5257 /* Discard any stale login responses from prev reset. 5258 * CHECK: should we clear even on INIT_COMPLETE? 5259 */ 5260 adapter->login_pending = false; 5261 5262 if (!completion_done(&adapter->init_done)) { 5263 complete(&adapter->init_done); 5264 adapter->init_done_rc = -EIO; 5265 } 5266 5267 if (adapter->state == VNIC_DOWN) 5268 rc = ibmvnic_reset(adapter, VNIC_RESET_PASSIVE_INIT); 5269 else 5270 rc = ibmvnic_reset(adapter, VNIC_RESET_FAILOVER); 5271 5272 if (rc && rc != -EBUSY) { 5273 /* We were unable to schedule the failover 5274 * reset either because the adapter was still 5275 * probing (eg: during kexec) or we could not 5276 * allocate memory. Clear the failover_pending 5277 * flag since no one else will. We ignore 5278 * EBUSY because it means either FAILOVER reset 5279 * is already scheduled or the adapter is 5280 * being removed. 5281 */ 5282 netdev_err(netdev, 5283 "Error %ld scheduling failover reset\n", 5284 rc); 5285 adapter->failover_pending = false; 5286 } 5287 break; 5288 case IBMVNIC_CRQ_INIT_COMPLETE: 5289 dev_info(dev, "Partner initialization complete\n"); 5290 adapter->crq.active = true; 5291 send_version_xchg(adapter); 5292 break; 5293 default: 5294 dev_err(dev, "Unknown crq cmd: %d\n", gen_crq->cmd); 5295 } 5296 return; 5297 case IBMVNIC_CRQ_XPORT_EVENT: 5298 netif_carrier_off(netdev); 5299 adapter->crq.active = false; 5300 /* terminate any thread waiting for a response 5301 * from the device 5302 */ 5303 if (!completion_done(&adapter->fw_done)) { 5304 adapter->fw_done_rc = -EIO; 5305 complete(&adapter->fw_done); 5306 } 5307 if (!completion_done(&adapter->stats_done)) 5308 complete(&adapter->stats_done); 5309 if (test_bit(0, &adapter->resetting)) 5310 adapter->force_reset_recovery = true; 5311 if (gen_crq->cmd == IBMVNIC_PARTITION_MIGRATED) { 5312 dev_info(dev, "Migrated, re-enabling adapter\n"); 5313 ibmvnic_reset(adapter, VNIC_RESET_MOBILITY); 5314 } else if (gen_crq->cmd == IBMVNIC_DEVICE_FAILOVER) { 5315 dev_info(dev, "Backing device failover detected\n"); 5316 adapter->failover_pending = true; 5317 } else { 5318 /* The adapter lost the connection */ 5319 dev_err(dev, "Virtual Adapter failed (rc=%d)\n", 5320 gen_crq->cmd); 5321 ibmvnic_reset(adapter, VNIC_RESET_FATAL); 5322 } 5323 return; 5324 case IBMVNIC_CRQ_CMD_RSP: 5325 break; 5326 default: 5327 dev_err(dev, "Got an invalid msg type 0x%02x\n", 5328 gen_crq->first); 5329 return; 5330 } 5331 5332 switch (gen_crq->cmd) { 5333 case VERSION_EXCHANGE_RSP: 5334 rc = crq->version_exchange_rsp.rc.code; 5335 if (rc) { 5336 dev_err(dev, "Error %ld in VERSION_EXCHG_RSP\n", rc); 5337 break; 5338 } 5339 ibmvnic_version = 5340 be16_to_cpu(crq->version_exchange_rsp.version); 5341 dev_info(dev, "Partner protocol version is %d\n", 5342 ibmvnic_version); 5343 send_query_cap(adapter); 5344 break; 5345 case QUERY_CAPABILITY_RSP: 5346 handle_query_cap_rsp(crq, adapter); 5347 break; 5348 case QUERY_MAP_RSP: 5349 handle_query_map_rsp(crq, adapter); 5350 break; 5351 case REQUEST_MAP_RSP: 5352 adapter->fw_done_rc = crq->request_map_rsp.rc.code; 5353 complete(&adapter->fw_done); 5354 break; 5355 case REQUEST_UNMAP_RSP: 5356 handle_request_unmap_rsp(crq, adapter); 5357 break; 5358 case REQUEST_CAPABILITY_RSP: 5359 handle_request_cap_rsp(crq, adapter); 5360 break; 5361 case LOGIN_RSP: 5362 netdev_dbg(netdev, "Got Login Response\n"); 5363 handle_login_rsp(crq, adapter); 5364 break; 5365 case LOGICAL_LINK_STATE_RSP: 5366 netdev_dbg(netdev, 5367 "Got Logical Link State Response, state: %d rc: %d\n", 5368 crq->logical_link_state_rsp.link_state, 5369 crq->logical_link_state_rsp.rc.code); 5370 adapter->logical_link_state = 5371 crq->logical_link_state_rsp.link_state; 5372 adapter->init_done_rc = crq->logical_link_state_rsp.rc.code; 5373 complete(&adapter->init_done); 5374 break; 5375 case LINK_STATE_INDICATION: 5376 netdev_dbg(netdev, "Got Logical Link State Indication\n"); 5377 adapter->phys_link_state = 5378 crq->link_state_indication.phys_link_state; 5379 adapter->logical_link_state = 5380 crq->link_state_indication.logical_link_state; 5381 if (adapter->phys_link_state && adapter->logical_link_state) 5382 netif_carrier_on(netdev); 5383 else 5384 netif_carrier_off(netdev); 5385 break; 5386 case CHANGE_MAC_ADDR_RSP: 5387 netdev_dbg(netdev, "Got MAC address change Response\n"); 5388 adapter->fw_done_rc = handle_change_mac_rsp(crq, adapter); 5389 break; 5390 case ERROR_INDICATION: 5391 netdev_dbg(netdev, "Got Error Indication\n"); 5392 handle_error_indication(crq, adapter); 5393 break; 5394 case REQUEST_STATISTICS_RSP: 5395 netdev_dbg(netdev, "Got Statistics Response\n"); 5396 complete(&adapter->stats_done); 5397 break; 5398 case QUERY_IP_OFFLOAD_RSP: 5399 netdev_dbg(netdev, "Got Query IP offload Response\n"); 5400 handle_query_ip_offload_rsp(adapter); 5401 break; 5402 case MULTICAST_CTRL_RSP: 5403 netdev_dbg(netdev, "Got multicast control Response\n"); 5404 break; 5405 case CONTROL_IP_OFFLOAD_RSP: 5406 netdev_dbg(netdev, "Got Control IP offload Response\n"); 5407 dma_unmap_single(dev, adapter->ip_offload_ctrl_tok, 5408 sizeof(adapter->ip_offload_ctrl), 5409 DMA_TO_DEVICE); 5410 complete(&adapter->init_done); 5411 break; 5412 case COLLECT_FW_TRACE_RSP: 5413 netdev_dbg(netdev, "Got Collect firmware trace Response\n"); 5414 complete(&adapter->fw_done); 5415 break; 5416 case GET_VPD_SIZE_RSP: 5417 handle_vpd_size_rsp(crq, adapter); 5418 break; 5419 case GET_VPD_RSP: 5420 handle_vpd_rsp(crq, adapter); 5421 break; 5422 case QUERY_PHYS_PARMS_RSP: 5423 adapter->fw_done_rc = handle_query_phys_parms_rsp(crq, adapter); 5424 complete(&adapter->fw_done); 5425 break; 5426 default: 5427 netdev_err(netdev, "Got an invalid cmd type 0x%02x\n", 5428 gen_crq->cmd); 5429 } 5430 } 5431 5432 static irqreturn_t ibmvnic_interrupt(int irq, void *instance) 5433 { 5434 struct ibmvnic_adapter *adapter = instance; 5435 5436 tasklet_schedule(&adapter->tasklet); 5437 return IRQ_HANDLED; 5438 } 5439 5440 static void ibmvnic_tasklet(struct tasklet_struct *t) 5441 { 5442 struct ibmvnic_adapter *adapter = from_tasklet(adapter, t, tasklet); 5443 struct ibmvnic_crq_queue *queue = &adapter->crq; 5444 union ibmvnic_crq *crq; 5445 unsigned long flags; 5446 bool done = false; 5447 5448 spin_lock_irqsave(&queue->lock, flags); 5449 while (!done) { 5450 /* Pull all the valid messages off the CRQ */ 5451 while ((crq = ibmvnic_next_crq(adapter)) != NULL) { 5452 /* This barrier makes sure ibmvnic_next_crq()'s 5453 * crq->generic.first & IBMVNIC_CRQ_CMD_RSP is loaded 5454 * before ibmvnic_handle_crq()'s 5455 * switch(gen_crq->first) and switch(gen_crq->cmd). 5456 */ 5457 dma_rmb(); 5458 ibmvnic_handle_crq(crq, adapter); 5459 crq->generic.first = 0; 5460 } 5461 5462 /* remain in tasklet until all 5463 * capabilities responses are received 5464 */ 5465 if (!adapter->wait_capability) 5466 done = true; 5467 } 5468 /* if capabilities CRQ's were sent in this tasklet, the following 5469 * tasklet must wait until all responses are received 5470 */ 5471 if (atomic_read(&adapter->running_cap_crqs) != 0) 5472 adapter->wait_capability = true; 5473 spin_unlock_irqrestore(&queue->lock, flags); 5474 } 5475 5476 static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *adapter) 5477 { 5478 struct vio_dev *vdev = adapter->vdev; 5479 int rc; 5480 5481 do { 5482 rc = plpar_hcall_norets(H_ENABLE_CRQ, vdev->unit_address); 5483 } while (rc == H_IN_PROGRESS || rc == H_BUSY || H_IS_LONG_BUSY(rc)); 5484 5485 if (rc) 5486 dev_err(&vdev->dev, "Error enabling adapter (rc=%d)\n", rc); 5487 5488 return rc; 5489 } 5490 5491 static int ibmvnic_reset_crq(struct ibmvnic_adapter *adapter) 5492 { 5493 struct ibmvnic_crq_queue *crq = &adapter->crq; 5494 struct device *dev = &adapter->vdev->dev; 5495 struct vio_dev *vdev = adapter->vdev; 5496 int rc; 5497 5498 /* Close the CRQ */ 5499 do { 5500 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address); 5501 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc)); 5502 5503 /* Clean out the queue */ 5504 if (!crq->msgs) 5505 return -EINVAL; 5506 5507 memset(crq->msgs, 0, PAGE_SIZE); 5508 crq->cur = 0; 5509 crq->active = false; 5510 5511 /* And re-open it again */ 5512 rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address, 5513 crq->msg_token, PAGE_SIZE); 5514 5515 if (rc == H_CLOSED) 5516 /* Adapter is good, but other end is not ready */ 5517 dev_warn(dev, "Partner adapter not ready\n"); 5518 else if (rc != 0) 5519 dev_warn(dev, "Couldn't register crq (rc=%d)\n", rc); 5520 5521 return rc; 5522 } 5523 5524 static void release_crq_queue(struct ibmvnic_adapter *adapter) 5525 { 5526 struct ibmvnic_crq_queue *crq = &adapter->crq; 5527 struct vio_dev *vdev = adapter->vdev; 5528 long rc; 5529 5530 if (!crq->msgs) 5531 return; 5532 5533 netdev_dbg(adapter->netdev, "Releasing CRQ\n"); 5534 free_irq(vdev->irq, adapter); 5535 tasklet_kill(&adapter->tasklet); 5536 do { 5537 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address); 5538 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc)); 5539 5540 dma_unmap_single(&vdev->dev, crq->msg_token, PAGE_SIZE, 5541 DMA_BIDIRECTIONAL); 5542 free_page((unsigned long)crq->msgs); 5543 crq->msgs = NULL; 5544 crq->active = false; 5545 } 5546 5547 static int init_crq_queue(struct ibmvnic_adapter *adapter) 5548 { 5549 struct ibmvnic_crq_queue *crq = &adapter->crq; 5550 struct device *dev = &adapter->vdev->dev; 5551 struct vio_dev *vdev = adapter->vdev; 5552 int rc, retrc = -ENOMEM; 5553 5554 if (crq->msgs) 5555 return 0; 5556 5557 crq->msgs = (union ibmvnic_crq *)get_zeroed_page(GFP_KERNEL); 5558 /* Should we allocate more than one page? */ 5559 5560 if (!crq->msgs) 5561 return -ENOMEM; 5562 5563 crq->size = PAGE_SIZE / sizeof(*crq->msgs); 5564 crq->msg_token = dma_map_single(dev, crq->msgs, PAGE_SIZE, 5565 DMA_BIDIRECTIONAL); 5566 if (dma_mapping_error(dev, crq->msg_token)) 5567 goto map_failed; 5568 5569 rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address, 5570 crq->msg_token, PAGE_SIZE); 5571 5572 if (rc == H_RESOURCE) 5573 /* maybe kexecing and resource is busy. try a reset */ 5574 rc = ibmvnic_reset_crq(adapter); 5575 retrc = rc; 5576 5577 if (rc == H_CLOSED) { 5578 dev_warn(dev, "Partner adapter not ready\n"); 5579 } else if (rc) { 5580 dev_warn(dev, "Error %d opening adapter\n", rc); 5581 goto reg_crq_failed; 5582 } 5583 5584 retrc = 0; 5585 5586 tasklet_setup(&adapter->tasklet, (void *)ibmvnic_tasklet); 5587 5588 netdev_dbg(adapter->netdev, "registering irq 0x%x\n", vdev->irq); 5589 snprintf(crq->name, sizeof(crq->name), "ibmvnic-%x", 5590 adapter->vdev->unit_address); 5591 rc = request_irq(vdev->irq, ibmvnic_interrupt, 0, crq->name, adapter); 5592 if (rc) { 5593 dev_err(dev, "Couldn't register irq 0x%x. rc=%d\n", 5594 vdev->irq, rc); 5595 goto req_irq_failed; 5596 } 5597 5598 rc = vio_enable_interrupts(vdev); 5599 if (rc) { 5600 dev_err(dev, "Error %d enabling interrupts\n", rc); 5601 goto req_irq_failed; 5602 } 5603 5604 crq->cur = 0; 5605 spin_lock_init(&crq->lock); 5606 5607 return retrc; 5608 5609 req_irq_failed: 5610 tasklet_kill(&adapter->tasklet); 5611 do { 5612 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address); 5613 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc)); 5614 reg_crq_failed: 5615 dma_unmap_single(dev, crq->msg_token, PAGE_SIZE, DMA_BIDIRECTIONAL); 5616 map_failed: 5617 free_page((unsigned long)crq->msgs); 5618 crq->msgs = NULL; 5619 return retrc; 5620 } 5621 5622 static int ibmvnic_reset_init(struct ibmvnic_adapter *adapter, bool reset) 5623 { 5624 struct device *dev = &adapter->vdev->dev; 5625 unsigned long timeout = msecs_to_jiffies(20000); 5626 u64 old_num_rx_queues = adapter->req_rx_queues; 5627 u64 old_num_tx_queues = adapter->req_tx_queues; 5628 int rc; 5629 5630 adapter->from_passive_init = false; 5631 5632 if (reset) 5633 reinit_completion(&adapter->init_done); 5634 5635 adapter->init_done_rc = 0; 5636 rc = ibmvnic_send_crq_init(adapter); 5637 if (rc) { 5638 dev_err(dev, "Send crq init failed with error %d\n", rc); 5639 return rc; 5640 } 5641 5642 if (!wait_for_completion_timeout(&adapter->init_done, timeout)) { 5643 dev_err(dev, "Initialization sequence timed out\n"); 5644 return -1; 5645 } 5646 5647 if (adapter->init_done_rc) { 5648 release_crq_queue(adapter); 5649 return adapter->init_done_rc; 5650 } 5651 5652 if (adapter->from_passive_init) { 5653 adapter->state = VNIC_OPEN; 5654 adapter->from_passive_init = false; 5655 return -1; 5656 } 5657 5658 if (reset && 5659 test_bit(0, &adapter->resetting) && !adapter->wait_for_reset && 5660 adapter->reset_reason != VNIC_RESET_MOBILITY) { 5661 if (adapter->req_rx_queues != old_num_rx_queues || 5662 adapter->req_tx_queues != old_num_tx_queues) { 5663 release_sub_crqs(adapter, 0); 5664 rc = init_sub_crqs(adapter); 5665 } else { 5666 rc = reset_sub_crq_queues(adapter); 5667 } 5668 } else { 5669 rc = init_sub_crqs(adapter); 5670 } 5671 5672 if (rc) { 5673 dev_err(dev, "Initialization of sub crqs failed\n"); 5674 release_crq_queue(adapter); 5675 return rc; 5676 } 5677 5678 rc = init_sub_crq_irqs(adapter); 5679 if (rc) { 5680 dev_err(dev, "Failed to initialize sub crq irqs\n"); 5681 release_crq_queue(adapter); 5682 } 5683 5684 return rc; 5685 } 5686 5687 static struct device_attribute dev_attr_failover; 5688 5689 static int ibmvnic_probe(struct vio_dev *dev, const struct vio_device_id *id) 5690 { 5691 struct ibmvnic_adapter *adapter; 5692 struct net_device *netdev; 5693 unsigned char *mac_addr_p; 5694 bool init_success; 5695 int rc; 5696 5697 dev_dbg(&dev->dev, "entering ibmvnic_probe for UA 0x%x\n", 5698 dev->unit_address); 5699 5700 mac_addr_p = (unsigned char *)vio_get_attribute(dev, 5701 VETH_MAC_ADDR, NULL); 5702 if (!mac_addr_p) { 5703 dev_err(&dev->dev, 5704 "(%s:%3.3d) ERROR: Can't find MAC_ADDR attribute\n", 5705 __FILE__, __LINE__); 5706 return 0; 5707 } 5708 5709 netdev = alloc_etherdev_mq(sizeof(struct ibmvnic_adapter), 5710 IBMVNIC_MAX_QUEUES); 5711 if (!netdev) 5712 return -ENOMEM; 5713 5714 adapter = netdev_priv(netdev); 5715 adapter->state = VNIC_PROBING; 5716 dev_set_drvdata(&dev->dev, netdev); 5717 adapter->vdev = dev; 5718 adapter->netdev = netdev; 5719 adapter->login_pending = false; 5720 memset(&adapter->map_ids, 0, sizeof(adapter->map_ids)); 5721 /* map_ids start at 1, so ensure map_id 0 is always "in-use" */ 5722 bitmap_set(adapter->map_ids, 0, 1); 5723 5724 ether_addr_copy(adapter->mac_addr, mac_addr_p); 5725 eth_hw_addr_set(netdev, adapter->mac_addr); 5726 netdev->irq = dev->irq; 5727 netdev->netdev_ops = &ibmvnic_netdev_ops; 5728 netdev->ethtool_ops = &ibmvnic_ethtool_ops; 5729 SET_NETDEV_DEV(netdev, &dev->dev); 5730 5731 INIT_WORK(&adapter->ibmvnic_reset, __ibmvnic_reset); 5732 INIT_DELAYED_WORK(&adapter->ibmvnic_delayed_reset, 5733 __ibmvnic_delayed_reset); 5734 INIT_LIST_HEAD(&adapter->rwi_list); 5735 spin_lock_init(&adapter->rwi_lock); 5736 spin_lock_init(&adapter->state_lock); 5737 mutex_init(&adapter->fw_lock); 5738 init_completion(&adapter->init_done); 5739 init_completion(&adapter->fw_done); 5740 init_completion(&adapter->reset_done); 5741 init_completion(&adapter->stats_done); 5742 clear_bit(0, &adapter->resetting); 5743 adapter->prev_rx_buf_sz = 0; 5744 adapter->prev_mtu = 0; 5745 5746 init_success = false; 5747 do { 5748 rc = init_crq_queue(adapter); 5749 if (rc) { 5750 dev_err(&dev->dev, "Couldn't initialize crq. rc=%d\n", 5751 rc); 5752 goto ibmvnic_init_fail; 5753 } 5754 5755 rc = ibmvnic_reset_init(adapter, false); 5756 } while (rc == EAGAIN); 5757 5758 /* We are ignoring the error from ibmvnic_reset_init() assuming that the 5759 * partner is not ready. CRQ is not active. When the partner becomes 5760 * ready, we will do the passive init reset. 5761 */ 5762 5763 if (!rc) 5764 init_success = true; 5765 5766 rc = init_stats_buffers(adapter); 5767 if (rc) 5768 goto ibmvnic_init_fail; 5769 5770 rc = init_stats_token(adapter); 5771 if (rc) 5772 goto ibmvnic_stats_fail; 5773 5774 rc = device_create_file(&dev->dev, &dev_attr_failover); 5775 if (rc) 5776 goto ibmvnic_dev_file_err; 5777 5778 netif_carrier_off(netdev); 5779 rc = register_netdev(netdev); 5780 if (rc) { 5781 dev_err(&dev->dev, "failed to register netdev rc=%d\n", rc); 5782 goto ibmvnic_register_fail; 5783 } 5784 dev_info(&dev->dev, "ibmvnic registered\n"); 5785 5786 if (init_success) { 5787 adapter->state = VNIC_PROBED; 5788 netdev->mtu = adapter->req_mtu - ETH_HLEN; 5789 netdev->min_mtu = adapter->min_mtu - ETH_HLEN; 5790 netdev->max_mtu = adapter->max_mtu - ETH_HLEN; 5791 } else { 5792 adapter->state = VNIC_DOWN; 5793 } 5794 5795 adapter->wait_for_reset = false; 5796 adapter->last_reset_time = jiffies; 5797 return 0; 5798 5799 ibmvnic_register_fail: 5800 device_remove_file(&dev->dev, &dev_attr_failover); 5801 5802 ibmvnic_dev_file_err: 5803 release_stats_token(adapter); 5804 5805 ibmvnic_stats_fail: 5806 release_stats_buffers(adapter); 5807 5808 ibmvnic_init_fail: 5809 release_sub_crqs(adapter, 1); 5810 release_crq_queue(adapter); 5811 mutex_destroy(&adapter->fw_lock); 5812 free_netdev(netdev); 5813 5814 return rc; 5815 } 5816 5817 static void ibmvnic_remove(struct vio_dev *dev) 5818 { 5819 struct net_device *netdev = dev_get_drvdata(&dev->dev); 5820 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 5821 unsigned long flags; 5822 5823 spin_lock_irqsave(&adapter->state_lock, flags); 5824 5825 /* If ibmvnic_reset() is scheduling a reset, wait for it to 5826 * finish. Then, set the state to REMOVING to prevent it from 5827 * scheduling any more work and to have reset functions ignore 5828 * any resets that have already been scheduled. Drop the lock 5829 * after setting state, so __ibmvnic_reset() which is called 5830 * from the flush_work() below, can make progress. 5831 */ 5832 spin_lock(&adapter->rwi_lock); 5833 adapter->state = VNIC_REMOVING; 5834 spin_unlock(&adapter->rwi_lock); 5835 5836 spin_unlock_irqrestore(&adapter->state_lock, flags); 5837 5838 flush_work(&adapter->ibmvnic_reset); 5839 flush_delayed_work(&adapter->ibmvnic_delayed_reset); 5840 5841 rtnl_lock(); 5842 unregister_netdevice(netdev); 5843 5844 release_resources(adapter); 5845 release_rx_pools(adapter); 5846 release_tx_pools(adapter); 5847 release_sub_crqs(adapter, 1); 5848 release_crq_queue(adapter); 5849 5850 release_stats_token(adapter); 5851 release_stats_buffers(adapter); 5852 5853 adapter->state = VNIC_REMOVED; 5854 5855 rtnl_unlock(); 5856 mutex_destroy(&adapter->fw_lock); 5857 device_remove_file(&dev->dev, &dev_attr_failover); 5858 free_netdev(netdev); 5859 dev_set_drvdata(&dev->dev, NULL); 5860 } 5861 5862 static ssize_t failover_store(struct device *dev, struct device_attribute *attr, 5863 const char *buf, size_t count) 5864 { 5865 struct net_device *netdev = dev_get_drvdata(dev); 5866 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 5867 unsigned long retbuf[PLPAR_HCALL_BUFSIZE]; 5868 __be64 session_token; 5869 long rc; 5870 5871 if (!sysfs_streq(buf, "1")) 5872 return -EINVAL; 5873 5874 rc = plpar_hcall(H_VIOCTL, retbuf, adapter->vdev->unit_address, 5875 H_GET_SESSION_TOKEN, 0, 0, 0); 5876 if (rc) { 5877 netdev_err(netdev, "Couldn't retrieve session token, rc %ld\n", 5878 rc); 5879 goto last_resort; 5880 } 5881 5882 session_token = (__be64)retbuf[0]; 5883 netdev_dbg(netdev, "Initiating client failover, session id %llx\n", 5884 be64_to_cpu(session_token)); 5885 rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address, 5886 H_SESSION_ERR_DETECTED, session_token, 0, 0); 5887 if (rc) 5888 netdev_err(netdev, 5889 "H_VIOCTL initiated failover failed, rc %ld\n", 5890 rc); 5891 5892 last_resort: 5893 netdev_dbg(netdev, "Trying to send CRQ_CMD, the last resort\n"); 5894 ibmvnic_reset(adapter, VNIC_RESET_FAILOVER); 5895 5896 return count; 5897 } 5898 static DEVICE_ATTR_WO(failover); 5899 5900 static unsigned long ibmvnic_get_desired_dma(struct vio_dev *vdev) 5901 { 5902 struct net_device *netdev = dev_get_drvdata(&vdev->dev); 5903 struct ibmvnic_adapter *adapter; 5904 struct iommu_table *tbl; 5905 unsigned long ret = 0; 5906 int i; 5907 5908 tbl = get_iommu_table_base(&vdev->dev); 5909 5910 /* netdev inits at probe time along with the structures we need below*/ 5911 if (!netdev) 5912 return IOMMU_PAGE_ALIGN(IBMVNIC_IO_ENTITLEMENT_DEFAULT, tbl); 5913 5914 adapter = netdev_priv(netdev); 5915 5916 ret += PAGE_SIZE; /* the crq message queue */ 5917 ret += IOMMU_PAGE_ALIGN(sizeof(struct ibmvnic_statistics), tbl); 5918 5919 for (i = 0; i < adapter->req_tx_queues + adapter->req_rx_queues; i++) 5920 ret += 4 * PAGE_SIZE; /* the scrq message queue */ 5921 5922 for (i = 0; i < adapter->num_active_rx_pools; i++) 5923 ret += adapter->rx_pool[i].size * 5924 IOMMU_PAGE_ALIGN(adapter->rx_pool[i].buff_size, tbl); 5925 5926 return ret; 5927 } 5928 5929 static int ibmvnic_resume(struct device *dev) 5930 { 5931 struct net_device *netdev = dev_get_drvdata(dev); 5932 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 5933 5934 if (adapter->state != VNIC_OPEN) 5935 return 0; 5936 5937 tasklet_schedule(&adapter->tasklet); 5938 5939 return 0; 5940 } 5941 5942 static const struct vio_device_id ibmvnic_device_table[] = { 5943 {"network", "IBM,vnic"}, 5944 {"", "" } 5945 }; 5946 MODULE_DEVICE_TABLE(vio, ibmvnic_device_table); 5947 5948 static const struct dev_pm_ops ibmvnic_pm_ops = { 5949 .resume = ibmvnic_resume 5950 }; 5951 5952 static struct vio_driver ibmvnic_driver = { 5953 .id_table = ibmvnic_device_table, 5954 .probe = ibmvnic_probe, 5955 .remove = ibmvnic_remove, 5956 .get_desired_dma = ibmvnic_get_desired_dma, 5957 .name = ibmvnic_driver_name, 5958 .pm = &ibmvnic_pm_ops, 5959 }; 5960 5961 /* module functions */ 5962 static int __init ibmvnic_module_init(void) 5963 { 5964 pr_info("%s: %s %s\n", ibmvnic_driver_name, ibmvnic_driver_string, 5965 IBMVNIC_DRIVER_VERSION); 5966 5967 return vio_register_driver(&ibmvnic_driver); 5968 } 5969 5970 static void __exit ibmvnic_module_exit(void) 5971 { 5972 vio_unregister_driver(&ibmvnic_driver); 5973 } 5974 5975 module_init(ibmvnic_module_init); 5976 module_exit(ibmvnic_module_exit); 5977