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/xive.h> 64 #include <asm/iommu.h> 65 #include <linux/uaccess.h> 66 #include <asm/firmware.h> 67 #include <linux/workqueue.h> 68 #include <linux/if_vlan.h> 69 #include <linux/utsname.h> 70 71 #include "ibmvnic.h" 72 73 static const char ibmvnic_driver_name[] = "ibmvnic"; 74 static const char ibmvnic_driver_string[] = "IBM System i/p Virtual NIC Driver"; 75 76 MODULE_AUTHOR("Santiago Leon"); 77 MODULE_DESCRIPTION("IBM System i/p Virtual NIC Driver"); 78 MODULE_LICENSE("GPL"); 79 MODULE_VERSION(IBMVNIC_DRIVER_VERSION); 80 81 static int ibmvnic_version = IBMVNIC_INITIAL_VERSION; 82 static void release_sub_crqs(struct ibmvnic_adapter *, bool); 83 static int ibmvnic_reset_crq(struct ibmvnic_adapter *); 84 static int ibmvnic_send_crq_init(struct ibmvnic_adapter *); 85 static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *); 86 static int ibmvnic_send_crq(struct ibmvnic_adapter *, union ibmvnic_crq *); 87 static int send_subcrq_indirect(struct ibmvnic_adapter *, u64, u64, u64); 88 static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance); 89 static int enable_scrq_irq(struct ibmvnic_adapter *, 90 struct ibmvnic_sub_crq_queue *); 91 static int disable_scrq_irq(struct ibmvnic_adapter *, 92 struct ibmvnic_sub_crq_queue *); 93 static int pending_scrq(struct ibmvnic_adapter *, 94 struct ibmvnic_sub_crq_queue *); 95 static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *, 96 struct ibmvnic_sub_crq_queue *); 97 static int ibmvnic_poll(struct napi_struct *napi, int data); 98 static void send_query_map(struct ibmvnic_adapter *adapter); 99 static int send_request_map(struct ibmvnic_adapter *, dma_addr_t, u32, u8); 100 static int send_request_unmap(struct ibmvnic_adapter *, u8); 101 static int send_login(struct ibmvnic_adapter *adapter); 102 static void send_query_cap(struct ibmvnic_adapter *adapter); 103 static int init_sub_crqs(struct ibmvnic_adapter *); 104 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter); 105 static int ibmvnic_reset_init(struct ibmvnic_adapter *, bool reset); 106 static void release_crq_queue(struct ibmvnic_adapter *); 107 static int __ibmvnic_set_mac(struct net_device *, u8 *); 108 static int init_crq_queue(struct ibmvnic_adapter *adapter); 109 static int send_query_phys_parms(struct ibmvnic_adapter *adapter); 110 static void ibmvnic_tx_scrq_clean_buffer(struct ibmvnic_adapter *adapter, 111 struct ibmvnic_sub_crq_queue *tx_scrq); 112 static void free_long_term_buff(struct ibmvnic_adapter *adapter, 113 struct ibmvnic_long_term_buff *ltb); 114 static void ibmvnic_disable_irqs(struct ibmvnic_adapter *adapter); 115 116 struct ibmvnic_stat { 117 char name[ETH_GSTRING_LEN]; 118 int offset; 119 }; 120 121 #define IBMVNIC_STAT_OFF(stat) (offsetof(struct ibmvnic_adapter, stats) + \ 122 offsetof(struct ibmvnic_statistics, stat)) 123 #define IBMVNIC_GET_STAT(a, off) (*((u64 *)(((unsigned long)(a)) + (off)))) 124 125 static const struct ibmvnic_stat ibmvnic_stats[] = { 126 {"rx_packets", IBMVNIC_STAT_OFF(rx_packets)}, 127 {"rx_bytes", IBMVNIC_STAT_OFF(rx_bytes)}, 128 {"tx_packets", IBMVNIC_STAT_OFF(tx_packets)}, 129 {"tx_bytes", IBMVNIC_STAT_OFF(tx_bytes)}, 130 {"ucast_tx_packets", IBMVNIC_STAT_OFF(ucast_tx_packets)}, 131 {"ucast_rx_packets", IBMVNIC_STAT_OFF(ucast_rx_packets)}, 132 {"mcast_tx_packets", IBMVNIC_STAT_OFF(mcast_tx_packets)}, 133 {"mcast_rx_packets", IBMVNIC_STAT_OFF(mcast_rx_packets)}, 134 {"bcast_tx_packets", IBMVNIC_STAT_OFF(bcast_tx_packets)}, 135 {"bcast_rx_packets", IBMVNIC_STAT_OFF(bcast_rx_packets)}, 136 {"align_errors", IBMVNIC_STAT_OFF(align_errors)}, 137 {"fcs_errors", IBMVNIC_STAT_OFF(fcs_errors)}, 138 {"single_collision_frames", IBMVNIC_STAT_OFF(single_collision_frames)}, 139 {"multi_collision_frames", IBMVNIC_STAT_OFF(multi_collision_frames)}, 140 {"sqe_test_errors", IBMVNIC_STAT_OFF(sqe_test_errors)}, 141 {"deferred_tx", IBMVNIC_STAT_OFF(deferred_tx)}, 142 {"late_collisions", IBMVNIC_STAT_OFF(late_collisions)}, 143 {"excess_collisions", IBMVNIC_STAT_OFF(excess_collisions)}, 144 {"internal_mac_tx_errors", IBMVNIC_STAT_OFF(internal_mac_tx_errors)}, 145 {"carrier_sense", IBMVNIC_STAT_OFF(carrier_sense)}, 146 {"too_long_frames", IBMVNIC_STAT_OFF(too_long_frames)}, 147 {"internal_mac_rx_errors", IBMVNIC_STAT_OFF(internal_mac_rx_errors)}, 148 }; 149 150 static int send_crq_init_complete(struct ibmvnic_adapter *adapter) 151 { 152 union ibmvnic_crq crq; 153 154 memset(&crq, 0, sizeof(crq)); 155 crq.generic.first = IBMVNIC_CRQ_INIT_CMD; 156 crq.generic.cmd = IBMVNIC_CRQ_INIT_COMPLETE; 157 158 return ibmvnic_send_crq(adapter, &crq); 159 } 160 161 static int send_version_xchg(struct ibmvnic_adapter *adapter) 162 { 163 union ibmvnic_crq crq; 164 165 memset(&crq, 0, sizeof(crq)); 166 crq.version_exchange.first = IBMVNIC_CRQ_CMD; 167 crq.version_exchange.cmd = VERSION_EXCHANGE; 168 crq.version_exchange.version = cpu_to_be16(ibmvnic_version); 169 170 return ibmvnic_send_crq(adapter, &crq); 171 } 172 173 static long h_reg_sub_crq(unsigned long unit_address, unsigned long token, 174 unsigned long length, unsigned long *number, 175 unsigned long *irq) 176 { 177 unsigned long retbuf[PLPAR_HCALL_BUFSIZE]; 178 long rc; 179 180 rc = plpar_hcall(H_REG_SUB_CRQ, retbuf, unit_address, token, length); 181 *number = retbuf[0]; 182 *irq = retbuf[1]; 183 184 return rc; 185 } 186 187 /** 188 * ibmvnic_wait_for_completion - Check device state and wait for completion 189 * @adapter: private device data 190 * @comp_done: completion structure to wait for 191 * @timeout: time to wait in milliseconds 192 * 193 * Wait for a completion signal or until the timeout limit is reached 194 * while checking that the device is still active. 195 */ 196 static int ibmvnic_wait_for_completion(struct ibmvnic_adapter *adapter, 197 struct completion *comp_done, 198 unsigned long timeout) 199 { 200 struct net_device *netdev; 201 unsigned long div_timeout; 202 u8 retry; 203 204 netdev = adapter->netdev; 205 retry = 5; 206 div_timeout = msecs_to_jiffies(timeout / retry); 207 while (true) { 208 if (!adapter->crq.active) { 209 netdev_err(netdev, "Device down!\n"); 210 return -ENODEV; 211 } 212 if (!retry--) 213 break; 214 if (wait_for_completion_timeout(comp_done, div_timeout)) 215 return 0; 216 } 217 netdev_err(netdev, "Operation timed out.\n"); 218 return -ETIMEDOUT; 219 } 220 221 /** 222 * reuse_ltb() - Check if a long term buffer can be reused 223 * @ltb: The long term buffer to be checked 224 * @size: The size of the long term buffer. 225 * 226 * An LTB can be reused unless its size has changed. 227 * 228 * Return: Return true if the LTB can be reused, false otherwise. 229 */ 230 static bool reuse_ltb(struct ibmvnic_long_term_buff *ltb, int size) 231 { 232 return (ltb->buff && ltb->size == size); 233 } 234 235 /** 236 * alloc_long_term_buff() - Allocate a long term buffer (LTB) 237 * 238 * @adapter: ibmvnic adapter associated to the LTB 239 * @ltb: container object for the LTB 240 * @size: size of the LTB 241 * 242 * Allocate an LTB of the specified size and notify VIOS. 243 * 244 * If the given @ltb already has the correct size, reuse it. Otherwise if 245 * its non-NULL, free it. Then allocate a new one of the correct size. 246 * Notify the VIOS either way since we may now be working with a new VIOS. 247 * 248 * Allocating larger chunks of memory during resets, specially LPM or under 249 * low memory situations can cause resets to fail/timeout and for LPAR to 250 * lose connectivity. So hold onto the LTB even if we fail to communicate 251 * with the VIOS and reuse it on next open. Free LTB when adapter is closed. 252 * 253 * Return: 0 if we were able to allocate the LTB and notify the VIOS and 254 * a negative value otherwise. 255 */ 256 static int alloc_long_term_buff(struct ibmvnic_adapter *adapter, 257 struct ibmvnic_long_term_buff *ltb, int size) 258 { 259 struct device *dev = &adapter->vdev->dev; 260 int rc; 261 262 if (!reuse_ltb(ltb, size)) { 263 dev_dbg(dev, 264 "LTB size changed from 0x%llx to 0x%x, reallocating\n", 265 ltb->size, size); 266 free_long_term_buff(adapter, ltb); 267 } 268 269 if (ltb->buff) { 270 dev_dbg(dev, "Reusing LTB [map %d, size 0x%llx]\n", 271 ltb->map_id, ltb->size); 272 } else { 273 ltb->buff = dma_alloc_coherent(dev, size, <b->addr, 274 GFP_KERNEL); 275 if (!ltb->buff) { 276 dev_err(dev, "Couldn't alloc long term buffer\n"); 277 return -ENOMEM; 278 } 279 ltb->size = size; 280 281 ltb->map_id = find_first_zero_bit(adapter->map_ids, 282 MAX_MAP_ID); 283 bitmap_set(adapter->map_ids, ltb->map_id, 1); 284 285 dev_dbg(dev, 286 "Allocated new LTB [map %d, size 0x%llx]\n", 287 ltb->map_id, ltb->size); 288 } 289 290 /* Ensure ltb is zeroed - specially when reusing it. */ 291 memset(ltb->buff, 0, ltb->size); 292 293 mutex_lock(&adapter->fw_lock); 294 adapter->fw_done_rc = 0; 295 reinit_completion(&adapter->fw_done); 296 297 rc = send_request_map(adapter, ltb->addr, ltb->size, ltb->map_id); 298 if (rc) { 299 dev_err(dev, "send_request_map failed, rc = %d\n", rc); 300 goto out; 301 } 302 303 rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000); 304 if (rc) { 305 dev_err(dev, "LTB map request aborted or timed out, rc = %d\n", 306 rc); 307 goto out; 308 } 309 310 if (adapter->fw_done_rc) { 311 dev_err(dev, "Couldn't map LTB, rc = %d\n", 312 adapter->fw_done_rc); 313 rc = -EIO; 314 goto out; 315 } 316 rc = 0; 317 out: 318 /* don't free LTB on communication error - see function header */ 319 mutex_unlock(&adapter->fw_lock); 320 return rc; 321 } 322 323 static void free_long_term_buff(struct ibmvnic_adapter *adapter, 324 struct ibmvnic_long_term_buff *ltb) 325 { 326 struct device *dev = &adapter->vdev->dev; 327 328 if (!ltb->buff) 329 return; 330 331 /* VIOS automatically unmaps the long term buffer at remote 332 * end for the following resets: 333 * FAILOVER, MOBILITY, TIMEOUT. 334 */ 335 if (adapter->reset_reason != VNIC_RESET_FAILOVER && 336 adapter->reset_reason != VNIC_RESET_MOBILITY && 337 adapter->reset_reason != VNIC_RESET_TIMEOUT) 338 send_request_unmap(adapter, ltb->map_id); 339 340 dma_free_coherent(dev, ltb->size, ltb->buff, ltb->addr); 341 342 ltb->buff = NULL; 343 /* mark this map_id free */ 344 bitmap_clear(adapter->map_ids, ltb->map_id, 1); 345 ltb->map_id = 0; 346 } 347 348 static void deactivate_rx_pools(struct ibmvnic_adapter *adapter) 349 { 350 int i; 351 352 for (i = 0; i < adapter->num_active_rx_pools; i++) 353 adapter->rx_pool[i].active = 0; 354 } 355 356 static void replenish_rx_pool(struct ibmvnic_adapter *adapter, 357 struct ibmvnic_rx_pool *pool) 358 { 359 int count = pool->size - atomic_read(&pool->available); 360 u64 handle = adapter->rx_scrq[pool->index]->handle; 361 struct device *dev = &adapter->vdev->dev; 362 struct ibmvnic_ind_xmit_queue *ind_bufp; 363 struct ibmvnic_sub_crq_queue *rx_scrq; 364 union sub_crq *sub_crq; 365 int buffers_added = 0; 366 unsigned long lpar_rc; 367 struct sk_buff *skb; 368 unsigned int offset; 369 dma_addr_t dma_addr; 370 unsigned char *dst; 371 int shift = 0; 372 int index; 373 int i; 374 375 if (!pool->active) 376 return; 377 378 rx_scrq = adapter->rx_scrq[pool->index]; 379 ind_bufp = &rx_scrq->ind_buf; 380 381 /* netdev_skb_alloc() could have failed after we saved a few skbs 382 * in the indir_buf and we would not have sent them to VIOS yet. 383 * To account for them, start the loop at ind_bufp->index rather 384 * than 0. If we pushed all the skbs to VIOS, ind_bufp->index will 385 * be 0. 386 */ 387 for (i = ind_bufp->index; i < count; ++i) { 388 index = pool->free_map[pool->next_free]; 389 390 /* We maybe reusing the skb from earlier resets. Allocate 391 * only if necessary. But since the LTB may have changed 392 * during reset (see init_rx_pools()), update LTB below 393 * even if reusing skb. 394 */ 395 skb = pool->rx_buff[index].skb; 396 if (!skb) { 397 skb = netdev_alloc_skb(adapter->netdev, 398 pool->buff_size); 399 if (!skb) { 400 dev_err(dev, "Couldn't replenish rx buff\n"); 401 adapter->replenish_no_mem++; 402 break; 403 } 404 } 405 406 pool->free_map[pool->next_free] = IBMVNIC_INVALID_MAP; 407 pool->next_free = (pool->next_free + 1) % pool->size; 408 409 /* Copy the skb to the long term mapped DMA buffer */ 410 offset = index * pool->buff_size; 411 dst = pool->long_term_buff.buff + offset; 412 memset(dst, 0, pool->buff_size); 413 dma_addr = pool->long_term_buff.addr + offset; 414 415 /* add the skb to an rx_buff in the pool */ 416 pool->rx_buff[index].data = dst; 417 pool->rx_buff[index].dma = dma_addr; 418 pool->rx_buff[index].skb = skb; 419 pool->rx_buff[index].pool_index = pool->index; 420 pool->rx_buff[index].size = pool->buff_size; 421 422 /* queue the rx_buff for the next send_subcrq_indirect */ 423 sub_crq = &ind_bufp->indir_arr[ind_bufp->index++]; 424 memset(sub_crq, 0, sizeof(*sub_crq)); 425 sub_crq->rx_add.first = IBMVNIC_CRQ_CMD; 426 sub_crq->rx_add.correlator = 427 cpu_to_be64((u64)&pool->rx_buff[index]); 428 sub_crq->rx_add.ioba = cpu_to_be32(dma_addr); 429 sub_crq->rx_add.map_id = pool->long_term_buff.map_id; 430 431 /* The length field of the sCRQ is defined to be 24 bits so the 432 * buffer size needs to be left shifted by a byte before it is 433 * converted to big endian to prevent the last byte from being 434 * truncated. 435 */ 436 #ifdef __LITTLE_ENDIAN__ 437 shift = 8; 438 #endif 439 sub_crq->rx_add.len = cpu_to_be32(pool->buff_size << shift); 440 441 /* if send_subcrq_indirect queue is full, flush to VIOS */ 442 if (ind_bufp->index == IBMVNIC_MAX_IND_DESCS || 443 i == count - 1) { 444 lpar_rc = 445 send_subcrq_indirect(adapter, handle, 446 (u64)ind_bufp->indir_dma, 447 (u64)ind_bufp->index); 448 if (lpar_rc != H_SUCCESS) 449 goto failure; 450 buffers_added += ind_bufp->index; 451 adapter->replenish_add_buff_success += ind_bufp->index; 452 ind_bufp->index = 0; 453 } 454 } 455 atomic_add(buffers_added, &pool->available); 456 return; 457 458 failure: 459 if (lpar_rc != H_PARAMETER && lpar_rc != H_CLOSED) 460 dev_err_ratelimited(dev, "rx: replenish packet buffer failed\n"); 461 for (i = ind_bufp->index - 1; i >= 0; --i) { 462 struct ibmvnic_rx_buff *rx_buff; 463 464 pool->next_free = pool->next_free == 0 ? 465 pool->size - 1 : pool->next_free - 1; 466 sub_crq = &ind_bufp->indir_arr[i]; 467 rx_buff = (struct ibmvnic_rx_buff *) 468 be64_to_cpu(sub_crq->rx_add.correlator); 469 index = (int)(rx_buff - pool->rx_buff); 470 pool->free_map[pool->next_free] = index; 471 dev_kfree_skb_any(pool->rx_buff[index].skb); 472 pool->rx_buff[index].skb = NULL; 473 } 474 adapter->replenish_add_buff_failure += ind_bufp->index; 475 atomic_add(buffers_added, &pool->available); 476 ind_bufp->index = 0; 477 if (lpar_rc == H_CLOSED || adapter->failover_pending) { 478 /* Disable buffer pool replenishment and report carrier off if 479 * queue is closed or pending failover. 480 * Firmware guarantees that a signal will be sent to the 481 * driver, triggering a reset. 482 */ 483 deactivate_rx_pools(adapter); 484 netif_carrier_off(adapter->netdev); 485 } 486 } 487 488 static void replenish_pools(struct ibmvnic_adapter *adapter) 489 { 490 int i; 491 492 adapter->replenish_task_cycles++; 493 for (i = 0; i < adapter->num_active_rx_pools; i++) { 494 if (adapter->rx_pool[i].active) 495 replenish_rx_pool(adapter, &adapter->rx_pool[i]); 496 } 497 498 netdev_dbg(adapter->netdev, "Replenished %d pools\n", i); 499 } 500 501 static void release_stats_buffers(struct ibmvnic_adapter *adapter) 502 { 503 kfree(adapter->tx_stats_buffers); 504 kfree(adapter->rx_stats_buffers); 505 adapter->tx_stats_buffers = NULL; 506 adapter->rx_stats_buffers = NULL; 507 } 508 509 static int init_stats_buffers(struct ibmvnic_adapter *adapter) 510 { 511 adapter->tx_stats_buffers = 512 kcalloc(IBMVNIC_MAX_QUEUES, 513 sizeof(struct ibmvnic_tx_queue_stats), 514 GFP_KERNEL); 515 if (!adapter->tx_stats_buffers) 516 return -ENOMEM; 517 518 adapter->rx_stats_buffers = 519 kcalloc(IBMVNIC_MAX_QUEUES, 520 sizeof(struct ibmvnic_rx_queue_stats), 521 GFP_KERNEL); 522 if (!adapter->rx_stats_buffers) 523 return -ENOMEM; 524 525 return 0; 526 } 527 528 static void release_stats_token(struct ibmvnic_adapter *adapter) 529 { 530 struct device *dev = &adapter->vdev->dev; 531 532 if (!adapter->stats_token) 533 return; 534 535 dma_unmap_single(dev, adapter->stats_token, 536 sizeof(struct ibmvnic_statistics), 537 DMA_FROM_DEVICE); 538 adapter->stats_token = 0; 539 } 540 541 static int init_stats_token(struct ibmvnic_adapter *adapter) 542 { 543 struct device *dev = &adapter->vdev->dev; 544 dma_addr_t stok; 545 int rc; 546 547 stok = dma_map_single(dev, &adapter->stats, 548 sizeof(struct ibmvnic_statistics), 549 DMA_FROM_DEVICE); 550 rc = dma_mapping_error(dev, stok); 551 if (rc) { 552 dev_err(dev, "Couldn't map stats buffer, rc = %d\n", rc); 553 return rc; 554 } 555 556 adapter->stats_token = stok; 557 netdev_dbg(adapter->netdev, "Stats token initialized (%llx)\n", stok); 558 return 0; 559 } 560 561 /** 562 * release_rx_pools() - Release any rx pools attached to @adapter. 563 * @adapter: ibmvnic adapter 564 * 565 * Safe to call this multiple times - even if no pools are attached. 566 */ 567 static void release_rx_pools(struct ibmvnic_adapter *adapter) 568 { 569 struct ibmvnic_rx_pool *rx_pool; 570 int i, j; 571 572 if (!adapter->rx_pool) 573 return; 574 575 for (i = 0; i < adapter->num_active_rx_pools; i++) { 576 rx_pool = &adapter->rx_pool[i]; 577 578 netdev_dbg(adapter->netdev, "Releasing rx_pool[%d]\n", i); 579 580 kfree(rx_pool->free_map); 581 582 free_long_term_buff(adapter, &rx_pool->long_term_buff); 583 584 if (!rx_pool->rx_buff) 585 continue; 586 587 for (j = 0; j < rx_pool->size; j++) { 588 if (rx_pool->rx_buff[j].skb) { 589 dev_kfree_skb_any(rx_pool->rx_buff[j].skb); 590 rx_pool->rx_buff[j].skb = NULL; 591 } 592 } 593 594 kfree(rx_pool->rx_buff); 595 } 596 597 kfree(adapter->rx_pool); 598 adapter->rx_pool = NULL; 599 adapter->num_active_rx_pools = 0; 600 adapter->prev_rx_pool_size = 0; 601 } 602 603 /** 604 * reuse_rx_pools() - Check if the existing rx pools can be reused. 605 * @adapter: ibmvnic adapter 606 * 607 * Check if the existing rx pools in the adapter can be reused. The 608 * pools can be reused if the pool parameters (number of pools, 609 * number of buffers in the pool and size of each buffer) have not 610 * changed. 611 * 612 * NOTE: This assumes that all pools have the same number of buffers 613 * which is the case currently. If that changes, we must fix this. 614 * 615 * Return: true if the rx pools can be reused, false otherwise. 616 */ 617 static bool reuse_rx_pools(struct ibmvnic_adapter *adapter) 618 { 619 u64 old_num_pools, new_num_pools; 620 u64 old_pool_size, new_pool_size; 621 u64 old_buff_size, new_buff_size; 622 623 if (!adapter->rx_pool) 624 return false; 625 626 old_num_pools = adapter->num_active_rx_pools; 627 new_num_pools = adapter->req_rx_queues; 628 629 old_pool_size = adapter->prev_rx_pool_size; 630 new_pool_size = adapter->req_rx_add_entries_per_subcrq; 631 632 old_buff_size = adapter->prev_rx_buf_sz; 633 new_buff_size = adapter->cur_rx_buf_sz; 634 635 if (old_buff_size != new_buff_size || 636 old_num_pools != new_num_pools || 637 old_pool_size != new_pool_size) 638 return false; 639 640 return true; 641 } 642 643 /** 644 * init_rx_pools(): Initialize the set of receiver pools in the adapter. 645 * @netdev: net device associated with the vnic interface 646 * 647 * Initialize the set of receiver pools in the ibmvnic adapter associated 648 * with the net_device @netdev. If possible, reuse the existing rx pools. 649 * Otherwise free any existing pools and allocate a new set of pools 650 * before initializing them. 651 * 652 * Return: 0 on success and negative value on error. 653 */ 654 static int init_rx_pools(struct net_device *netdev) 655 { 656 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 657 struct device *dev = &adapter->vdev->dev; 658 struct ibmvnic_rx_pool *rx_pool; 659 u64 num_pools; 660 u64 pool_size; /* # of buffers in one pool */ 661 u64 buff_size; 662 int i, j, rc; 663 664 pool_size = adapter->req_rx_add_entries_per_subcrq; 665 num_pools = adapter->req_rx_queues; 666 buff_size = adapter->cur_rx_buf_sz; 667 668 if (reuse_rx_pools(adapter)) { 669 dev_dbg(dev, "Reusing rx pools\n"); 670 goto update_ltb; 671 } 672 673 /* Allocate/populate the pools. */ 674 release_rx_pools(adapter); 675 676 adapter->rx_pool = kcalloc(num_pools, 677 sizeof(struct ibmvnic_rx_pool), 678 GFP_KERNEL); 679 if (!adapter->rx_pool) { 680 dev_err(dev, "Failed to allocate rx pools\n"); 681 return -ENOMEM; 682 } 683 684 /* Set num_active_rx_pools early. If we fail below after partial 685 * allocation, release_rx_pools() will know how many to look for. 686 */ 687 adapter->num_active_rx_pools = num_pools; 688 689 for (i = 0; i < num_pools; i++) { 690 rx_pool = &adapter->rx_pool[i]; 691 692 netdev_dbg(adapter->netdev, 693 "Initializing rx_pool[%d], %lld buffs, %lld bytes each\n", 694 i, pool_size, buff_size); 695 696 rx_pool->size = pool_size; 697 rx_pool->index = i; 698 rx_pool->buff_size = ALIGN(buff_size, L1_CACHE_BYTES); 699 700 rx_pool->free_map = kcalloc(rx_pool->size, sizeof(int), 701 GFP_KERNEL); 702 if (!rx_pool->free_map) { 703 dev_err(dev, "Couldn't alloc free_map %d\n", i); 704 rc = -ENOMEM; 705 goto out_release; 706 } 707 708 rx_pool->rx_buff = kcalloc(rx_pool->size, 709 sizeof(struct ibmvnic_rx_buff), 710 GFP_KERNEL); 711 if (!rx_pool->rx_buff) { 712 dev_err(dev, "Couldn't alloc rx buffers\n"); 713 rc = -ENOMEM; 714 goto out_release; 715 } 716 } 717 718 adapter->prev_rx_pool_size = pool_size; 719 adapter->prev_rx_buf_sz = adapter->cur_rx_buf_sz; 720 721 update_ltb: 722 for (i = 0; i < num_pools; i++) { 723 rx_pool = &adapter->rx_pool[i]; 724 dev_dbg(dev, "Updating LTB for rx pool %d [%d, %d]\n", 725 i, rx_pool->size, rx_pool->buff_size); 726 727 rc = alloc_long_term_buff(adapter, &rx_pool->long_term_buff, 728 rx_pool->size * rx_pool->buff_size); 729 if (rc) 730 goto out; 731 732 for (j = 0; j < rx_pool->size; ++j) { 733 struct ibmvnic_rx_buff *rx_buff; 734 735 rx_pool->free_map[j] = j; 736 737 /* NOTE: Don't clear rx_buff->skb here - will leak 738 * memory! replenish_rx_pool() will reuse skbs or 739 * allocate as necessary. 740 */ 741 rx_buff = &rx_pool->rx_buff[j]; 742 rx_buff->dma = 0; 743 rx_buff->data = 0; 744 rx_buff->size = 0; 745 rx_buff->pool_index = 0; 746 } 747 748 /* Mark pool "empty" so replenish_rx_pools() will 749 * update the LTB info for each buffer 750 */ 751 atomic_set(&rx_pool->available, 0); 752 rx_pool->next_alloc = 0; 753 rx_pool->next_free = 0; 754 /* replenish_rx_pool() may have called deactivate_rx_pools() 755 * on failover. Ensure pool is active now. 756 */ 757 rx_pool->active = 1; 758 } 759 return 0; 760 out_release: 761 release_rx_pools(adapter); 762 out: 763 /* We failed to allocate one or more LTBs or map them on the VIOS. 764 * Hold onto the pools and any LTBs that we did allocate/map. 765 */ 766 return rc; 767 } 768 769 static void release_vpd_data(struct ibmvnic_adapter *adapter) 770 { 771 if (!adapter->vpd) 772 return; 773 774 kfree(adapter->vpd->buff); 775 kfree(adapter->vpd); 776 777 adapter->vpd = NULL; 778 } 779 780 static void release_one_tx_pool(struct ibmvnic_adapter *adapter, 781 struct ibmvnic_tx_pool *tx_pool) 782 { 783 kfree(tx_pool->tx_buff); 784 kfree(tx_pool->free_map); 785 free_long_term_buff(adapter, &tx_pool->long_term_buff); 786 } 787 788 /** 789 * release_tx_pools() - Release any tx pools attached to @adapter. 790 * @adapter: ibmvnic adapter 791 * 792 * Safe to call this multiple times - even if no pools are attached. 793 */ 794 static void release_tx_pools(struct ibmvnic_adapter *adapter) 795 { 796 int i; 797 798 /* init_tx_pools() ensures that ->tx_pool and ->tso_pool are 799 * both NULL or both non-NULL. So we only need to check one. 800 */ 801 if (!adapter->tx_pool) 802 return; 803 804 for (i = 0; i < adapter->num_active_tx_pools; i++) { 805 release_one_tx_pool(adapter, &adapter->tx_pool[i]); 806 release_one_tx_pool(adapter, &adapter->tso_pool[i]); 807 } 808 809 kfree(adapter->tx_pool); 810 adapter->tx_pool = NULL; 811 kfree(adapter->tso_pool); 812 adapter->tso_pool = NULL; 813 adapter->num_active_tx_pools = 0; 814 adapter->prev_tx_pool_size = 0; 815 } 816 817 static int init_one_tx_pool(struct net_device *netdev, 818 struct ibmvnic_tx_pool *tx_pool, 819 int pool_size, int buf_size) 820 { 821 int i; 822 823 tx_pool->tx_buff = kcalloc(pool_size, 824 sizeof(struct ibmvnic_tx_buff), 825 GFP_KERNEL); 826 if (!tx_pool->tx_buff) 827 return -ENOMEM; 828 829 tx_pool->free_map = kcalloc(pool_size, sizeof(int), GFP_KERNEL); 830 if (!tx_pool->free_map) { 831 kfree(tx_pool->tx_buff); 832 tx_pool->tx_buff = NULL; 833 return -ENOMEM; 834 } 835 836 for (i = 0; i < pool_size; i++) 837 tx_pool->free_map[i] = i; 838 839 tx_pool->consumer_index = 0; 840 tx_pool->producer_index = 0; 841 tx_pool->num_buffers = pool_size; 842 tx_pool->buf_size = buf_size; 843 844 return 0; 845 } 846 847 /** 848 * reuse_tx_pools() - Check if the existing tx pools can be reused. 849 * @adapter: ibmvnic adapter 850 * 851 * Check if the existing tx pools in the adapter can be reused. The 852 * pools can be reused if the pool parameters (number of pools, 853 * number of buffers in the pool and mtu) have not changed. 854 * 855 * NOTE: This assumes that all pools have the same number of buffers 856 * which is the case currently. If that changes, we must fix this. 857 * 858 * Return: true if the tx pools can be reused, false otherwise. 859 */ 860 static bool reuse_tx_pools(struct ibmvnic_adapter *adapter) 861 { 862 u64 old_num_pools, new_num_pools; 863 u64 old_pool_size, new_pool_size; 864 u64 old_mtu, new_mtu; 865 866 if (!adapter->tx_pool) 867 return false; 868 869 old_num_pools = adapter->num_active_tx_pools; 870 new_num_pools = adapter->num_active_tx_scrqs; 871 old_pool_size = adapter->prev_tx_pool_size; 872 new_pool_size = adapter->req_tx_entries_per_subcrq; 873 old_mtu = adapter->prev_mtu; 874 new_mtu = adapter->req_mtu; 875 876 if (old_mtu != new_mtu || 877 old_num_pools != new_num_pools || 878 old_pool_size != new_pool_size) 879 return false; 880 881 return true; 882 } 883 884 /** 885 * init_tx_pools(): Initialize the set of transmit pools in the adapter. 886 * @netdev: net device associated with the vnic interface 887 * 888 * Initialize the set of transmit pools in the ibmvnic adapter associated 889 * with the net_device @netdev. If possible, reuse the existing tx pools. 890 * Otherwise free any existing pools and allocate a new set of pools 891 * before initializing them. 892 * 893 * Return: 0 on success and negative value on error. 894 */ 895 static int init_tx_pools(struct net_device *netdev) 896 { 897 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 898 struct device *dev = &adapter->vdev->dev; 899 int num_pools; 900 u64 pool_size; /* # of buffers in pool */ 901 u64 buff_size; 902 int i, j, rc; 903 904 num_pools = adapter->req_tx_queues; 905 906 /* We must notify the VIOS about the LTB on all resets - but we only 907 * need to alloc/populate pools if either the number of buffers or 908 * size of each buffer in the pool has changed. 909 */ 910 if (reuse_tx_pools(adapter)) { 911 netdev_dbg(netdev, "Reusing tx pools\n"); 912 goto update_ltb; 913 } 914 915 /* Allocate/populate the pools. */ 916 release_tx_pools(adapter); 917 918 pool_size = adapter->req_tx_entries_per_subcrq; 919 num_pools = adapter->num_active_tx_scrqs; 920 921 adapter->tx_pool = kcalloc(num_pools, 922 sizeof(struct ibmvnic_tx_pool), GFP_KERNEL); 923 if (!adapter->tx_pool) 924 return -ENOMEM; 925 926 adapter->tso_pool = kcalloc(num_pools, 927 sizeof(struct ibmvnic_tx_pool), GFP_KERNEL); 928 /* To simplify release_tx_pools() ensure that ->tx_pool and 929 * ->tso_pool are either both NULL or both non-NULL. 930 */ 931 if (!adapter->tso_pool) { 932 kfree(adapter->tx_pool); 933 adapter->tx_pool = NULL; 934 return -ENOMEM; 935 } 936 937 /* Set num_active_tx_pools early. If we fail below after partial 938 * allocation, release_tx_pools() will know how many to look for. 939 */ 940 adapter->num_active_tx_pools = num_pools; 941 942 buff_size = adapter->req_mtu + VLAN_HLEN; 943 buff_size = ALIGN(buff_size, L1_CACHE_BYTES); 944 945 for (i = 0; i < num_pools; i++) { 946 dev_dbg(dev, "Init tx pool %d [%llu, %llu]\n", 947 i, adapter->req_tx_entries_per_subcrq, buff_size); 948 949 rc = init_one_tx_pool(netdev, &adapter->tx_pool[i], 950 pool_size, buff_size); 951 if (rc) 952 goto out_release; 953 954 rc = init_one_tx_pool(netdev, &adapter->tso_pool[i], 955 IBMVNIC_TSO_BUFS, 956 IBMVNIC_TSO_BUF_SZ); 957 if (rc) 958 goto out_release; 959 } 960 961 adapter->prev_tx_pool_size = pool_size; 962 adapter->prev_mtu = adapter->req_mtu; 963 964 update_ltb: 965 /* NOTE: All tx_pools have the same number of buffers (which is 966 * same as pool_size). All tso_pools have IBMVNIC_TSO_BUFS 967 * buffers (see calls init_one_tx_pool() for these). 968 * For consistency, we use tx_pool->num_buffers and 969 * tso_pool->num_buffers below. 970 */ 971 rc = -1; 972 for (i = 0; i < num_pools; i++) { 973 struct ibmvnic_tx_pool *tso_pool; 974 struct ibmvnic_tx_pool *tx_pool; 975 u32 ltb_size; 976 977 tx_pool = &adapter->tx_pool[i]; 978 ltb_size = tx_pool->num_buffers * tx_pool->buf_size; 979 if (alloc_long_term_buff(adapter, &tx_pool->long_term_buff, 980 ltb_size)) 981 goto out; 982 983 dev_dbg(dev, "Updated LTB for tx pool %d [%p, %d, %d]\n", 984 i, tx_pool->long_term_buff.buff, 985 tx_pool->num_buffers, tx_pool->buf_size); 986 987 tx_pool->consumer_index = 0; 988 tx_pool->producer_index = 0; 989 990 for (j = 0; j < tx_pool->num_buffers; j++) 991 tx_pool->free_map[j] = j; 992 993 tso_pool = &adapter->tso_pool[i]; 994 ltb_size = tso_pool->num_buffers * tso_pool->buf_size; 995 if (alloc_long_term_buff(adapter, &tso_pool->long_term_buff, 996 ltb_size)) 997 goto out; 998 999 dev_dbg(dev, "Updated LTB for tso pool %d [%p, %d, %d]\n", 1000 i, tso_pool->long_term_buff.buff, 1001 tso_pool->num_buffers, tso_pool->buf_size); 1002 1003 tso_pool->consumer_index = 0; 1004 tso_pool->producer_index = 0; 1005 1006 for (j = 0; j < tso_pool->num_buffers; j++) 1007 tso_pool->free_map[j] = j; 1008 } 1009 1010 return 0; 1011 out_release: 1012 release_tx_pools(adapter); 1013 out: 1014 /* We failed to allocate one or more LTBs or map them on the VIOS. 1015 * Hold onto the pools and any LTBs that we did allocate/map. 1016 */ 1017 return rc; 1018 } 1019 1020 static void ibmvnic_napi_enable(struct ibmvnic_adapter *adapter) 1021 { 1022 int i; 1023 1024 if (adapter->napi_enabled) 1025 return; 1026 1027 for (i = 0; i < adapter->req_rx_queues; i++) 1028 napi_enable(&adapter->napi[i]); 1029 1030 adapter->napi_enabled = true; 1031 } 1032 1033 static void ibmvnic_napi_disable(struct ibmvnic_adapter *adapter) 1034 { 1035 int i; 1036 1037 if (!adapter->napi_enabled) 1038 return; 1039 1040 for (i = 0; i < adapter->req_rx_queues; i++) { 1041 netdev_dbg(adapter->netdev, "Disabling napi[%d]\n", i); 1042 napi_disable(&adapter->napi[i]); 1043 } 1044 1045 adapter->napi_enabled = false; 1046 } 1047 1048 static int init_napi(struct ibmvnic_adapter *adapter) 1049 { 1050 int i; 1051 1052 adapter->napi = kcalloc(adapter->req_rx_queues, 1053 sizeof(struct napi_struct), GFP_KERNEL); 1054 if (!adapter->napi) 1055 return -ENOMEM; 1056 1057 for (i = 0; i < adapter->req_rx_queues; i++) { 1058 netdev_dbg(adapter->netdev, "Adding napi[%d]\n", i); 1059 netif_napi_add(adapter->netdev, &adapter->napi[i], 1060 ibmvnic_poll, NAPI_POLL_WEIGHT); 1061 } 1062 1063 adapter->num_active_rx_napi = adapter->req_rx_queues; 1064 return 0; 1065 } 1066 1067 static void release_napi(struct ibmvnic_adapter *adapter) 1068 { 1069 int i; 1070 1071 if (!adapter->napi) 1072 return; 1073 1074 for (i = 0; i < adapter->num_active_rx_napi; i++) { 1075 netdev_dbg(adapter->netdev, "Releasing napi[%d]\n", i); 1076 netif_napi_del(&adapter->napi[i]); 1077 } 1078 1079 kfree(adapter->napi); 1080 adapter->napi = NULL; 1081 adapter->num_active_rx_napi = 0; 1082 adapter->napi_enabled = false; 1083 } 1084 1085 static const char *adapter_state_to_string(enum vnic_state state) 1086 { 1087 switch (state) { 1088 case VNIC_PROBING: 1089 return "PROBING"; 1090 case VNIC_PROBED: 1091 return "PROBED"; 1092 case VNIC_OPENING: 1093 return "OPENING"; 1094 case VNIC_OPEN: 1095 return "OPEN"; 1096 case VNIC_CLOSING: 1097 return "CLOSING"; 1098 case VNIC_CLOSED: 1099 return "CLOSED"; 1100 case VNIC_REMOVING: 1101 return "REMOVING"; 1102 case VNIC_REMOVED: 1103 return "REMOVED"; 1104 case VNIC_DOWN: 1105 return "DOWN"; 1106 } 1107 return "UNKNOWN"; 1108 } 1109 1110 static int ibmvnic_login(struct net_device *netdev) 1111 { 1112 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 1113 unsigned long timeout = msecs_to_jiffies(20000); 1114 int retry_count = 0; 1115 int retries = 10; 1116 bool retry; 1117 int rc; 1118 1119 do { 1120 retry = false; 1121 if (retry_count > retries) { 1122 netdev_warn(netdev, "Login attempts exceeded\n"); 1123 return -EACCES; 1124 } 1125 1126 adapter->init_done_rc = 0; 1127 reinit_completion(&adapter->init_done); 1128 rc = send_login(adapter); 1129 if (rc) 1130 return rc; 1131 1132 if (!wait_for_completion_timeout(&adapter->init_done, 1133 timeout)) { 1134 netdev_warn(netdev, "Login timed out, retrying...\n"); 1135 retry = true; 1136 adapter->init_done_rc = 0; 1137 retry_count++; 1138 continue; 1139 } 1140 1141 if (adapter->init_done_rc == ABORTED) { 1142 netdev_warn(netdev, "Login aborted, retrying...\n"); 1143 retry = true; 1144 adapter->init_done_rc = 0; 1145 retry_count++; 1146 /* FW or device may be busy, so 1147 * wait a bit before retrying login 1148 */ 1149 msleep(500); 1150 } else if (adapter->init_done_rc == PARTIALSUCCESS) { 1151 retry_count++; 1152 release_sub_crqs(adapter, 1); 1153 1154 retry = true; 1155 netdev_dbg(netdev, 1156 "Received partial success, retrying...\n"); 1157 adapter->init_done_rc = 0; 1158 reinit_completion(&adapter->init_done); 1159 send_query_cap(adapter); 1160 if (!wait_for_completion_timeout(&adapter->init_done, 1161 timeout)) { 1162 netdev_warn(netdev, 1163 "Capabilities query timed out\n"); 1164 return -ETIMEDOUT; 1165 } 1166 1167 rc = init_sub_crqs(adapter); 1168 if (rc) { 1169 netdev_warn(netdev, 1170 "SCRQ initialization failed\n"); 1171 return rc; 1172 } 1173 1174 rc = init_sub_crq_irqs(adapter); 1175 if (rc) { 1176 netdev_warn(netdev, 1177 "SCRQ irq initialization failed\n"); 1178 return rc; 1179 } 1180 } else if (adapter->init_done_rc) { 1181 netdev_warn(netdev, "Adapter login failed, init_done_rc = %d\n", 1182 adapter->init_done_rc); 1183 return -EIO; 1184 } 1185 } while (retry); 1186 1187 __ibmvnic_set_mac(netdev, adapter->mac_addr); 1188 1189 netdev_dbg(netdev, "[S:%s] Login succeeded\n", adapter_state_to_string(adapter->state)); 1190 return 0; 1191 } 1192 1193 static void release_login_buffer(struct ibmvnic_adapter *adapter) 1194 { 1195 kfree(adapter->login_buf); 1196 adapter->login_buf = NULL; 1197 } 1198 1199 static void release_login_rsp_buffer(struct ibmvnic_adapter *adapter) 1200 { 1201 kfree(adapter->login_rsp_buf); 1202 adapter->login_rsp_buf = NULL; 1203 } 1204 1205 static void release_resources(struct ibmvnic_adapter *adapter) 1206 { 1207 release_vpd_data(adapter); 1208 1209 release_napi(adapter); 1210 release_login_buffer(adapter); 1211 release_login_rsp_buffer(adapter); 1212 } 1213 1214 static int set_link_state(struct ibmvnic_adapter *adapter, u8 link_state) 1215 { 1216 struct net_device *netdev = adapter->netdev; 1217 unsigned long timeout = msecs_to_jiffies(20000); 1218 union ibmvnic_crq crq; 1219 bool resend; 1220 int rc; 1221 1222 netdev_dbg(netdev, "setting link state %d\n", link_state); 1223 1224 memset(&crq, 0, sizeof(crq)); 1225 crq.logical_link_state.first = IBMVNIC_CRQ_CMD; 1226 crq.logical_link_state.cmd = LOGICAL_LINK_STATE; 1227 crq.logical_link_state.link_state = link_state; 1228 1229 do { 1230 resend = false; 1231 1232 reinit_completion(&adapter->init_done); 1233 rc = ibmvnic_send_crq(adapter, &crq); 1234 if (rc) { 1235 netdev_err(netdev, "Failed to set link state\n"); 1236 return rc; 1237 } 1238 1239 if (!wait_for_completion_timeout(&adapter->init_done, 1240 timeout)) { 1241 netdev_err(netdev, "timeout setting link state\n"); 1242 return -ETIMEDOUT; 1243 } 1244 1245 if (adapter->init_done_rc == PARTIALSUCCESS) { 1246 /* Partuial success, delay and re-send */ 1247 mdelay(1000); 1248 resend = true; 1249 } else if (adapter->init_done_rc) { 1250 netdev_warn(netdev, "Unable to set link state, rc=%d\n", 1251 adapter->init_done_rc); 1252 return adapter->init_done_rc; 1253 } 1254 } while (resend); 1255 1256 return 0; 1257 } 1258 1259 static int set_real_num_queues(struct net_device *netdev) 1260 { 1261 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 1262 int rc; 1263 1264 netdev_dbg(netdev, "Setting real tx/rx queues (%llx/%llx)\n", 1265 adapter->req_tx_queues, adapter->req_rx_queues); 1266 1267 rc = netif_set_real_num_tx_queues(netdev, adapter->req_tx_queues); 1268 if (rc) { 1269 netdev_err(netdev, "failed to set the number of tx queues\n"); 1270 return rc; 1271 } 1272 1273 rc = netif_set_real_num_rx_queues(netdev, adapter->req_rx_queues); 1274 if (rc) 1275 netdev_err(netdev, "failed to set the number of rx queues\n"); 1276 1277 return rc; 1278 } 1279 1280 static int ibmvnic_get_vpd(struct ibmvnic_adapter *adapter) 1281 { 1282 struct device *dev = &adapter->vdev->dev; 1283 union ibmvnic_crq crq; 1284 int len = 0; 1285 int rc; 1286 1287 if (adapter->vpd->buff) 1288 len = adapter->vpd->len; 1289 1290 mutex_lock(&adapter->fw_lock); 1291 adapter->fw_done_rc = 0; 1292 reinit_completion(&adapter->fw_done); 1293 1294 crq.get_vpd_size.first = IBMVNIC_CRQ_CMD; 1295 crq.get_vpd_size.cmd = GET_VPD_SIZE; 1296 rc = ibmvnic_send_crq(adapter, &crq); 1297 if (rc) { 1298 mutex_unlock(&adapter->fw_lock); 1299 return rc; 1300 } 1301 1302 rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000); 1303 if (rc) { 1304 dev_err(dev, "Could not retrieve VPD size, rc = %d\n", rc); 1305 mutex_unlock(&adapter->fw_lock); 1306 return rc; 1307 } 1308 mutex_unlock(&adapter->fw_lock); 1309 1310 if (!adapter->vpd->len) 1311 return -ENODATA; 1312 1313 if (!adapter->vpd->buff) 1314 adapter->vpd->buff = kzalloc(adapter->vpd->len, GFP_KERNEL); 1315 else if (adapter->vpd->len != len) 1316 adapter->vpd->buff = 1317 krealloc(adapter->vpd->buff, 1318 adapter->vpd->len, GFP_KERNEL); 1319 1320 if (!adapter->vpd->buff) { 1321 dev_err(dev, "Could allocate VPD buffer\n"); 1322 return -ENOMEM; 1323 } 1324 1325 adapter->vpd->dma_addr = 1326 dma_map_single(dev, adapter->vpd->buff, adapter->vpd->len, 1327 DMA_FROM_DEVICE); 1328 if (dma_mapping_error(dev, adapter->vpd->dma_addr)) { 1329 dev_err(dev, "Could not map VPD buffer\n"); 1330 kfree(adapter->vpd->buff); 1331 adapter->vpd->buff = NULL; 1332 return -ENOMEM; 1333 } 1334 1335 mutex_lock(&adapter->fw_lock); 1336 adapter->fw_done_rc = 0; 1337 reinit_completion(&adapter->fw_done); 1338 1339 crq.get_vpd.first = IBMVNIC_CRQ_CMD; 1340 crq.get_vpd.cmd = GET_VPD; 1341 crq.get_vpd.ioba = cpu_to_be32(adapter->vpd->dma_addr); 1342 crq.get_vpd.len = cpu_to_be32((u32)adapter->vpd->len); 1343 rc = ibmvnic_send_crq(adapter, &crq); 1344 if (rc) { 1345 kfree(adapter->vpd->buff); 1346 adapter->vpd->buff = NULL; 1347 mutex_unlock(&adapter->fw_lock); 1348 return rc; 1349 } 1350 1351 rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000); 1352 if (rc) { 1353 dev_err(dev, "Unable to retrieve VPD, rc = %d\n", rc); 1354 kfree(adapter->vpd->buff); 1355 adapter->vpd->buff = NULL; 1356 mutex_unlock(&adapter->fw_lock); 1357 return rc; 1358 } 1359 1360 mutex_unlock(&adapter->fw_lock); 1361 return 0; 1362 } 1363 1364 static int init_resources(struct ibmvnic_adapter *adapter) 1365 { 1366 struct net_device *netdev = adapter->netdev; 1367 int rc; 1368 1369 rc = set_real_num_queues(netdev); 1370 if (rc) 1371 return rc; 1372 1373 adapter->vpd = kzalloc(sizeof(*adapter->vpd), GFP_KERNEL); 1374 if (!adapter->vpd) 1375 return -ENOMEM; 1376 1377 /* Vital Product Data (VPD) */ 1378 rc = ibmvnic_get_vpd(adapter); 1379 if (rc) { 1380 netdev_err(netdev, "failed to initialize Vital Product Data (VPD)\n"); 1381 return rc; 1382 } 1383 1384 rc = init_napi(adapter); 1385 if (rc) 1386 return rc; 1387 1388 send_query_map(adapter); 1389 1390 rc = init_rx_pools(netdev); 1391 if (rc) 1392 return rc; 1393 1394 rc = init_tx_pools(netdev); 1395 return rc; 1396 } 1397 1398 static int __ibmvnic_open(struct net_device *netdev) 1399 { 1400 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 1401 enum vnic_state prev_state = adapter->state; 1402 int i, rc; 1403 1404 adapter->state = VNIC_OPENING; 1405 replenish_pools(adapter); 1406 ibmvnic_napi_enable(adapter); 1407 1408 /* We're ready to receive frames, enable the sub-crq interrupts and 1409 * set the logical link state to up 1410 */ 1411 for (i = 0; i < adapter->req_rx_queues; i++) { 1412 netdev_dbg(netdev, "Enabling rx_scrq[%d] irq\n", i); 1413 if (prev_state == VNIC_CLOSED) 1414 enable_irq(adapter->rx_scrq[i]->irq); 1415 enable_scrq_irq(adapter, adapter->rx_scrq[i]); 1416 } 1417 1418 for (i = 0; i < adapter->req_tx_queues; i++) { 1419 netdev_dbg(netdev, "Enabling tx_scrq[%d] irq\n", i); 1420 if (prev_state == VNIC_CLOSED) 1421 enable_irq(adapter->tx_scrq[i]->irq); 1422 enable_scrq_irq(adapter, adapter->tx_scrq[i]); 1423 netdev_tx_reset_queue(netdev_get_tx_queue(netdev, i)); 1424 } 1425 1426 rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_UP); 1427 if (rc) { 1428 ibmvnic_napi_disable(adapter); 1429 ibmvnic_disable_irqs(adapter); 1430 return rc; 1431 } 1432 1433 netif_tx_start_all_queues(netdev); 1434 1435 if (prev_state == VNIC_CLOSED) { 1436 for (i = 0; i < adapter->req_rx_queues; i++) 1437 napi_schedule(&adapter->napi[i]); 1438 } 1439 1440 adapter->state = VNIC_OPEN; 1441 return rc; 1442 } 1443 1444 static int ibmvnic_open(struct net_device *netdev) 1445 { 1446 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 1447 int rc; 1448 1449 ASSERT_RTNL(); 1450 1451 /* If device failover is pending or we are about to reset, just set 1452 * device state and return. Device operation will be handled by reset 1453 * routine. 1454 * 1455 * It should be safe to overwrite the adapter->state here. Since 1456 * we hold the rtnl, either the reset has not actually started or 1457 * the rtnl got dropped during the set_link_state() in do_reset(). 1458 * In the former case, no one else is changing the state (again we 1459 * have the rtnl) and in the latter case, do_reset() will detect and 1460 * honor our setting below. 1461 */ 1462 if (adapter->failover_pending || (test_bit(0, &adapter->resetting))) { 1463 netdev_dbg(netdev, "[S:%s FOP:%d] Resetting, deferring open\n", 1464 adapter_state_to_string(adapter->state), 1465 adapter->failover_pending); 1466 adapter->state = VNIC_OPEN; 1467 rc = 0; 1468 goto out; 1469 } 1470 1471 if (adapter->state != VNIC_CLOSED) { 1472 rc = ibmvnic_login(netdev); 1473 if (rc) 1474 goto out; 1475 1476 rc = init_resources(adapter); 1477 if (rc) { 1478 netdev_err(netdev, "failed to initialize resources\n"); 1479 goto out; 1480 } 1481 } 1482 1483 rc = __ibmvnic_open(netdev); 1484 1485 out: 1486 /* If open failed and there is a pending failover or in-progress reset, 1487 * set device state and return. Device operation will be handled by 1488 * reset routine. See also comments above regarding rtnl. 1489 */ 1490 if (rc && 1491 (adapter->failover_pending || (test_bit(0, &adapter->resetting)))) { 1492 adapter->state = VNIC_OPEN; 1493 rc = 0; 1494 } 1495 1496 if (rc) { 1497 release_resources(adapter); 1498 release_rx_pools(adapter); 1499 release_tx_pools(adapter); 1500 } 1501 1502 return rc; 1503 } 1504 1505 static void clean_rx_pools(struct ibmvnic_adapter *adapter) 1506 { 1507 struct ibmvnic_rx_pool *rx_pool; 1508 struct ibmvnic_rx_buff *rx_buff; 1509 u64 rx_entries; 1510 int rx_scrqs; 1511 int i, j; 1512 1513 if (!adapter->rx_pool) 1514 return; 1515 1516 rx_scrqs = adapter->num_active_rx_pools; 1517 rx_entries = adapter->req_rx_add_entries_per_subcrq; 1518 1519 /* Free any remaining skbs in the rx buffer pools */ 1520 for (i = 0; i < rx_scrqs; i++) { 1521 rx_pool = &adapter->rx_pool[i]; 1522 if (!rx_pool || !rx_pool->rx_buff) 1523 continue; 1524 1525 netdev_dbg(adapter->netdev, "Cleaning rx_pool[%d]\n", i); 1526 for (j = 0; j < rx_entries; j++) { 1527 rx_buff = &rx_pool->rx_buff[j]; 1528 if (rx_buff && rx_buff->skb) { 1529 dev_kfree_skb_any(rx_buff->skb); 1530 rx_buff->skb = NULL; 1531 } 1532 } 1533 } 1534 } 1535 1536 static void clean_one_tx_pool(struct ibmvnic_adapter *adapter, 1537 struct ibmvnic_tx_pool *tx_pool) 1538 { 1539 struct ibmvnic_tx_buff *tx_buff; 1540 u64 tx_entries; 1541 int i; 1542 1543 if (!tx_pool || !tx_pool->tx_buff) 1544 return; 1545 1546 tx_entries = tx_pool->num_buffers; 1547 1548 for (i = 0; i < tx_entries; i++) { 1549 tx_buff = &tx_pool->tx_buff[i]; 1550 if (tx_buff && tx_buff->skb) { 1551 dev_kfree_skb_any(tx_buff->skb); 1552 tx_buff->skb = NULL; 1553 } 1554 } 1555 } 1556 1557 static void clean_tx_pools(struct ibmvnic_adapter *adapter) 1558 { 1559 int tx_scrqs; 1560 int i; 1561 1562 if (!adapter->tx_pool || !adapter->tso_pool) 1563 return; 1564 1565 tx_scrqs = adapter->num_active_tx_pools; 1566 1567 /* Free any remaining skbs in the tx buffer pools */ 1568 for (i = 0; i < tx_scrqs; i++) { 1569 netdev_dbg(adapter->netdev, "Cleaning tx_pool[%d]\n", i); 1570 clean_one_tx_pool(adapter, &adapter->tx_pool[i]); 1571 clean_one_tx_pool(adapter, &adapter->tso_pool[i]); 1572 } 1573 } 1574 1575 static void ibmvnic_disable_irqs(struct ibmvnic_adapter *adapter) 1576 { 1577 struct net_device *netdev = adapter->netdev; 1578 int i; 1579 1580 if (adapter->tx_scrq) { 1581 for (i = 0; i < adapter->req_tx_queues; i++) 1582 if (adapter->tx_scrq[i]->irq) { 1583 netdev_dbg(netdev, 1584 "Disabling tx_scrq[%d] irq\n", i); 1585 disable_scrq_irq(adapter, adapter->tx_scrq[i]); 1586 disable_irq(adapter->tx_scrq[i]->irq); 1587 } 1588 } 1589 1590 if (adapter->rx_scrq) { 1591 for (i = 0; i < adapter->req_rx_queues; i++) { 1592 if (adapter->rx_scrq[i]->irq) { 1593 netdev_dbg(netdev, 1594 "Disabling rx_scrq[%d] irq\n", i); 1595 disable_scrq_irq(adapter, adapter->rx_scrq[i]); 1596 disable_irq(adapter->rx_scrq[i]->irq); 1597 } 1598 } 1599 } 1600 } 1601 1602 static void ibmvnic_cleanup(struct net_device *netdev) 1603 { 1604 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 1605 1606 /* ensure that transmissions are stopped if called by do_reset */ 1607 if (test_bit(0, &adapter->resetting)) 1608 netif_tx_disable(netdev); 1609 else 1610 netif_tx_stop_all_queues(netdev); 1611 1612 ibmvnic_napi_disable(adapter); 1613 ibmvnic_disable_irqs(adapter); 1614 } 1615 1616 static int __ibmvnic_close(struct net_device *netdev) 1617 { 1618 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 1619 int rc = 0; 1620 1621 adapter->state = VNIC_CLOSING; 1622 rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN); 1623 adapter->state = VNIC_CLOSED; 1624 return rc; 1625 } 1626 1627 static int ibmvnic_close(struct net_device *netdev) 1628 { 1629 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 1630 int rc; 1631 1632 netdev_dbg(netdev, "[S:%s FOP:%d FRR:%d] Closing\n", 1633 adapter_state_to_string(adapter->state), 1634 adapter->failover_pending, 1635 adapter->force_reset_recovery); 1636 1637 /* If device failover is pending, just set device state and return. 1638 * Device operation will be handled by reset routine. 1639 */ 1640 if (adapter->failover_pending) { 1641 adapter->state = VNIC_CLOSED; 1642 return 0; 1643 } 1644 1645 rc = __ibmvnic_close(netdev); 1646 ibmvnic_cleanup(netdev); 1647 clean_rx_pools(adapter); 1648 clean_tx_pools(adapter); 1649 1650 return rc; 1651 } 1652 1653 /** 1654 * build_hdr_data - creates L2/L3/L4 header data buffer 1655 * @hdr_field: bitfield determining needed headers 1656 * @skb: socket buffer 1657 * @hdr_len: array of header lengths 1658 * @hdr_data: buffer to write the header to 1659 * 1660 * Reads hdr_field to determine which headers are needed by firmware. 1661 * Builds a buffer containing these headers. Saves individual header 1662 * lengths and total buffer length to be used to build descriptors. 1663 */ 1664 static int build_hdr_data(u8 hdr_field, struct sk_buff *skb, 1665 int *hdr_len, u8 *hdr_data) 1666 { 1667 int len = 0; 1668 u8 *hdr; 1669 1670 if (skb_vlan_tagged(skb) && !skb_vlan_tag_present(skb)) 1671 hdr_len[0] = sizeof(struct vlan_ethhdr); 1672 else 1673 hdr_len[0] = sizeof(struct ethhdr); 1674 1675 if (skb->protocol == htons(ETH_P_IP)) { 1676 hdr_len[1] = ip_hdr(skb)->ihl * 4; 1677 if (ip_hdr(skb)->protocol == IPPROTO_TCP) 1678 hdr_len[2] = tcp_hdrlen(skb); 1679 else if (ip_hdr(skb)->protocol == IPPROTO_UDP) 1680 hdr_len[2] = sizeof(struct udphdr); 1681 } else if (skb->protocol == htons(ETH_P_IPV6)) { 1682 hdr_len[1] = sizeof(struct ipv6hdr); 1683 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP) 1684 hdr_len[2] = tcp_hdrlen(skb); 1685 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP) 1686 hdr_len[2] = sizeof(struct udphdr); 1687 } else if (skb->protocol == htons(ETH_P_ARP)) { 1688 hdr_len[1] = arp_hdr_len(skb->dev); 1689 hdr_len[2] = 0; 1690 } 1691 1692 memset(hdr_data, 0, 120); 1693 if ((hdr_field >> 6) & 1) { 1694 hdr = skb_mac_header(skb); 1695 memcpy(hdr_data, hdr, hdr_len[0]); 1696 len += hdr_len[0]; 1697 } 1698 1699 if ((hdr_field >> 5) & 1) { 1700 hdr = skb_network_header(skb); 1701 memcpy(hdr_data + len, hdr, hdr_len[1]); 1702 len += hdr_len[1]; 1703 } 1704 1705 if ((hdr_field >> 4) & 1) { 1706 hdr = skb_transport_header(skb); 1707 memcpy(hdr_data + len, hdr, hdr_len[2]); 1708 len += hdr_len[2]; 1709 } 1710 return len; 1711 } 1712 1713 /** 1714 * create_hdr_descs - create header and header extension descriptors 1715 * @hdr_field: bitfield determining needed headers 1716 * @hdr_data: buffer containing header data 1717 * @len: length of data buffer 1718 * @hdr_len: array of individual header lengths 1719 * @scrq_arr: descriptor array 1720 * 1721 * Creates header and, if needed, header extension descriptors and 1722 * places them in a descriptor array, scrq_arr 1723 */ 1724 1725 static int create_hdr_descs(u8 hdr_field, u8 *hdr_data, int len, int *hdr_len, 1726 union sub_crq *scrq_arr) 1727 { 1728 union sub_crq hdr_desc; 1729 int tmp_len = len; 1730 int num_descs = 0; 1731 u8 *data, *cur; 1732 int tmp; 1733 1734 while (tmp_len > 0) { 1735 cur = hdr_data + len - tmp_len; 1736 1737 memset(&hdr_desc, 0, sizeof(hdr_desc)); 1738 if (cur != hdr_data) { 1739 data = hdr_desc.hdr_ext.data; 1740 tmp = tmp_len > 29 ? 29 : tmp_len; 1741 hdr_desc.hdr_ext.first = IBMVNIC_CRQ_CMD; 1742 hdr_desc.hdr_ext.type = IBMVNIC_HDR_EXT_DESC; 1743 hdr_desc.hdr_ext.len = tmp; 1744 } else { 1745 data = hdr_desc.hdr.data; 1746 tmp = tmp_len > 24 ? 24 : tmp_len; 1747 hdr_desc.hdr.first = IBMVNIC_CRQ_CMD; 1748 hdr_desc.hdr.type = IBMVNIC_HDR_DESC; 1749 hdr_desc.hdr.len = tmp; 1750 hdr_desc.hdr.l2_len = (u8)hdr_len[0]; 1751 hdr_desc.hdr.l3_len = cpu_to_be16((u16)hdr_len[1]); 1752 hdr_desc.hdr.l4_len = (u8)hdr_len[2]; 1753 hdr_desc.hdr.flag = hdr_field << 1; 1754 } 1755 memcpy(data, cur, tmp); 1756 tmp_len -= tmp; 1757 *scrq_arr = hdr_desc; 1758 scrq_arr++; 1759 num_descs++; 1760 } 1761 1762 return num_descs; 1763 } 1764 1765 /** 1766 * build_hdr_descs_arr - build a header descriptor array 1767 * @skb: tx socket buffer 1768 * @indir_arr: indirect array 1769 * @num_entries: number of descriptors to be sent 1770 * @hdr_field: bit field determining which headers will be sent 1771 * 1772 * This function will build a TX descriptor array with applicable 1773 * L2/L3/L4 packet header descriptors to be sent by send_subcrq_indirect. 1774 */ 1775 1776 static void build_hdr_descs_arr(struct sk_buff *skb, 1777 union sub_crq *indir_arr, 1778 int *num_entries, u8 hdr_field) 1779 { 1780 int hdr_len[3] = {0, 0, 0}; 1781 u8 hdr_data[140] = {0}; 1782 int tot_len; 1783 1784 tot_len = build_hdr_data(hdr_field, skb, hdr_len, 1785 hdr_data); 1786 *num_entries += create_hdr_descs(hdr_field, hdr_data, tot_len, hdr_len, 1787 indir_arr + 1); 1788 } 1789 1790 static int ibmvnic_xmit_workarounds(struct sk_buff *skb, 1791 struct net_device *netdev) 1792 { 1793 /* For some backing devices, mishandling of small packets 1794 * can result in a loss of connection or TX stall. Device 1795 * architects recommend that no packet should be smaller 1796 * than the minimum MTU value provided to the driver, so 1797 * pad any packets to that length 1798 */ 1799 if (skb->len < netdev->min_mtu) 1800 return skb_put_padto(skb, netdev->min_mtu); 1801 1802 return 0; 1803 } 1804 1805 static void ibmvnic_tx_scrq_clean_buffer(struct ibmvnic_adapter *adapter, 1806 struct ibmvnic_sub_crq_queue *tx_scrq) 1807 { 1808 struct ibmvnic_ind_xmit_queue *ind_bufp; 1809 struct ibmvnic_tx_buff *tx_buff; 1810 struct ibmvnic_tx_pool *tx_pool; 1811 union sub_crq tx_scrq_entry; 1812 int queue_num; 1813 int entries; 1814 int index; 1815 int i; 1816 1817 ind_bufp = &tx_scrq->ind_buf; 1818 entries = (u64)ind_bufp->index; 1819 queue_num = tx_scrq->pool_index; 1820 1821 for (i = entries - 1; i >= 0; --i) { 1822 tx_scrq_entry = ind_bufp->indir_arr[i]; 1823 if (tx_scrq_entry.v1.type != IBMVNIC_TX_DESC) 1824 continue; 1825 index = be32_to_cpu(tx_scrq_entry.v1.correlator); 1826 if (index & IBMVNIC_TSO_POOL_MASK) { 1827 tx_pool = &adapter->tso_pool[queue_num]; 1828 index &= ~IBMVNIC_TSO_POOL_MASK; 1829 } else { 1830 tx_pool = &adapter->tx_pool[queue_num]; 1831 } 1832 tx_pool->free_map[tx_pool->consumer_index] = index; 1833 tx_pool->consumer_index = tx_pool->consumer_index == 0 ? 1834 tx_pool->num_buffers - 1 : 1835 tx_pool->consumer_index - 1; 1836 tx_buff = &tx_pool->tx_buff[index]; 1837 adapter->netdev->stats.tx_packets--; 1838 adapter->netdev->stats.tx_bytes -= tx_buff->skb->len; 1839 adapter->tx_stats_buffers[queue_num].packets--; 1840 adapter->tx_stats_buffers[queue_num].bytes -= 1841 tx_buff->skb->len; 1842 dev_kfree_skb_any(tx_buff->skb); 1843 tx_buff->skb = NULL; 1844 adapter->netdev->stats.tx_dropped++; 1845 } 1846 ind_bufp->index = 0; 1847 if (atomic_sub_return(entries, &tx_scrq->used) <= 1848 (adapter->req_tx_entries_per_subcrq / 2) && 1849 __netif_subqueue_stopped(adapter->netdev, queue_num) && 1850 !test_bit(0, &adapter->resetting)) { 1851 netif_wake_subqueue(adapter->netdev, queue_num); 1852 netdev_dbg(adapter->netdev, "Started queue %d\n", 1853 queue_num); 1854 } 1855 } 1856 1857 static int ibmvnic_tx_scrq_flush(struct ibmvnic_adapter *adapter, 1858 struct ibmvnic_sub_crq_queue *tx_scrq) 1859 { 1860 struct ibmvnic_ind_xmit_queue *ind_bufp; 1861 u64 dma_addr; 1862 u64 entries; 1863 u64 handle; 1864 int rc; 1865 1866 ind_bufp = &tx_scrq->ind_buf; 1867 dma_addr = (u64)ind_bufp->indir_dma; 1868 entries = (u64)ind_bufp->index; 1869 handle = tx_scrq->handle; 1870 1871 if (!entries) 1872 return 0; 1873 rc = send_subcrq_indirect(adapter, handle, dma_addr, entries); 1874 if (rc) 1875 ibmvnic_tx_scrq_clean_buffer(adapter, tx_scrq); 1876 else 1877 ind_bufp->index = 0; 1878 return 0; 1879 } 1880 1881 static netdev_tx_t ibmvnic_xmit(struct sk_buff *skb, struct net_device *netdev) 1882 { 1883 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 1884 int queue_num = skb_get_queue_mapping(skb); 1885 u8 *hdrs = (u8 *)&adapter->tx_rx_desc_req; 1886 struct device *dev = &adapter->vdev->dev; 1887 struct ibmvnic_ind_xmit_queue *ind_bufp; 1888 struct ibmvnic_tx_buff *tx_buff = NULL; 1889 struct ibmvnic_sub_crq_queue *tx_scrq; 1890 struct ibmvnic_tx_pool *tx_pool; 1891 unsigned int tx_send_failed = 0; 1892 netdev_tx_t ret = NETDEV_TX_OK; 1893 unsigned int tx_map_failed = 0; 1894 union sub_crq indir_arr[16]; 1895 unsigned int tx_dropped = 0; 1896 unsigned int tx_packets = 0; 1897 unsigned int tx_bytes = 0; 1898 dma_addr_t data_dma_addr; 1899 struct netdev_queue *txq; 1900 unsigned long lpar_rc; 1901 union sub_crq tx_crq; 1902 unsigned int offset; 1903 int num_entries = 1; 1904 unsigned char *dst; 1905 int index = 0; 1906 u8 proto = 0; 1907 1908 tx_scrq = adapter->tx_scrq[queue_num]; 1909 txq = netdev_get_tx_queue(netdev, queue_num); 1910 ind_bufp = &tx_scrq->ind_buf; 1911 1912 if (test_bit(0, &adapter->resetting)) { 1913 dev_kfree_skb_any(skb); 1914 1915 tx_send_failed++; 1916 tx_dropped++; 1917 ret = NETDEV_TX_OK; 1918 goto out; 1919 } 1920 1921 if (ibmvnic_xmit_workarounds(skb, netdev)) { 1922 tx_dropped++; 1923 tx_send_failed++; 1924 ret = NETDEV_TX_OK; 1925 ibmvnic_tx_scrq_flush(adapter, tx_scrq); 1926 goto out; 1927 } 1928 if (skb_is_gso(skb)) 1929 tx_pool = &adapter->tso_pool[queue_num]; 1930 else 1931 tx_pool = &adapter->tx_pool[queue_num]; 1932 1933 index = tx_pool->free_map[tx_pool->consumer_index]; 1934 1935 if (index == IBMVNIC_INVALID_MAP) { 1936 dev_kfree_skb_any(skb); 1937 tx_send_failed++; 1938 tx_dropped++; 1939 ibmvnic_tx_scrq_flush(adapter, tx_scrq); 1940 ret = NETDEV_TX_OK; 1941 goto out; 1942 } 1943 1944 tx_pool->free_map[tx_pool->consumer_index] = IBMVNIC_INVALID_MAP; 1945 1946 offset = index * tx_pool->buf_size; 1947 dst = tx_pool->long_term_buff.buff + offset; 1948 memset(dst, 0, tx_pool->buf_size); 1949 data_dma_addr = tx_pool->long_term_buff.addr + offset; 1950 1951 if (skb_shinfo(skb)->nr_frags) { 1952 int cur, i; 1953 1954 /* Copy the head */ 1955 skb_copy_from_linear_data(skb, dst, skb_headlen(skb)); 1956 cur = skb_headlen(skb); 1957 1958 /* Copy the frags */ 1959 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 1960 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 1961 1962 memcpy(dst + cur, skb_frag_address(frag), 1963 skb_frag_size(frag)); 1964 cur += skb_frag_size(frag); 1965 } 1966 } else { 1967 skb_copy_from_linear_data(skb, dst, skb->len); 1968 } 1969 1970 /* post changes to long_term_buff *dst before VIOS accessing it */ 1971 dma_wmb(); 1972 1973 tx_pool->consumer_index = 1974 (tx_pool->consumer_index + 1) % tx_pool->num_buffers; 1975 1976 tx_buff = &tx_pool->tx_buff[index]; 1977 tx_buff->skb = skb; 1978 tx_buff->index = index; 1979 tx_buff->pool_index = queue_num; 1980 1981 memset(&tx_crq, 0, sizeof(tx_crq)); 1982 tx_crq.v1.first = IBMVNIC_CRQ_CMD; 1983 tx_crq.v1.type = IBMVNIC_TX_DESC; 1984 tx_crq.v1.n_crq_elem = 1; 1985 tx_crq.v1.n_sge = 1; 1986 tx_crq.v1.flags1 = IBMVNIC_TX_COMP_NEEDED; 1987 1988 if (skb_is_gso(skb)) 1989 tx_crq.v1.correlator = 1990 cpu_to_be32(index | IBMVNIC_TSO_POOL_MASK); 1991 else 1992 tx_crq.v1.correlator = cpu_to_be32(index); 1993 tx_crq.v1.dma_reg = cpu_to_be16(tx_pool->long_term_buff.map_id); 1994 tx_crq.v1.sge_len = cpu_to_be32(skb->len); 1995 tx_crq.v1.ioba = cpu_to_be64(data_dma_addr); 1996 1997 if (adapter->vlan_header_insertion && skb_vlan_tag_present(skb)) { 1998 tx_crq.v1.flags2 |= IBMVNIC_TX_VLAN_INSERT; 1999 tx_crq.v1.vlan_id = cpu_to_be16(skb->vlan_tci); 2000 } 2001 2002 if (skb->protocol == htons(ETH_P_IP)) { 2003 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV4; 2004 proto = ip_hdr(skb)->protocol; 2005 } else if (skb->protocol == htons(ETH_P_IPV6)) { 2006 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV6; 2007 proto = ipv6_hdr(skb)->nexthdr; 2008 } 2009 2010 if (proto == IPPROTO_TCP) 2011 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_TCP; 2012 else if (proto == IPPROTO_UDP) 2013 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_UDP; 2014 2015 if (skb->ip_summed == CHECKSUM_PARTIAL) { 2016 tx_crq.v1.flags1 |= IBMVNIC_TX_CHKSUM_OFFLOAD; 2017 hdrs += 2; 2018 } 2019 if (skb_is_gso(skb)) { 2020 tx_crq.v1.flags1 |= IBMVNIC_TX_LSO; 2021 tx_crq.v1.mss = cpu_to_be16(skb_shinfo(skb)->gso_size); 2022 hdrs += 2; 2023 } 2024 2025 if ((*hdrs >> 7) & 1) 2026 build_hdr_descs_arr(skb, indir_arr, &num_entries, *hdrs); 2027 2028 tx_crq.v1.n_crq_elem = num_entries; 2029 tx_buff->num_entries = num_entries; 2030 /* flush buffer if current entry can not fit */ 2031 if (num_entries + ind_bufp->index > IBMVNIC_MAX_IND_DESCS) { 2032 lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq); 2033 if (lpar_rc != H_SUCCESS) 2034 goto tx_flush_err; 2035 } 2036 2037 indir_arr[0] = tx_crq; 2038 memcpy(&ind_bufp->indir_arr[ind_bufp->index], &indir_arr[0], 2039 num_entries * sizeof(struct ibmvnic_generic_scrq)); 2040 ind_bufp->index += num_entries; 2041 if (__netdev_tx_sent_queue(txq, skb->len, 2042 netdev_xmit_more() && 2043 ind_bufp->index < IBMVNIC_MAX_IND_DESCS)) { 2044 lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq); 2045 if (lpar_rc != H_SUCCESS) 2046 goto tx_err; 2047 } 2048 2049 if (atomic_add_return(num_entries, &tx_scrq->used) 2050 >= adapter->req_tx_entries_per_subcrq) { 2051 netdev_dbg(netdev, "Stopping queue %d\n", queue_num); 2052 netif_stop_subqueue(netdev, queue_num); 2053 } 2054 2055 tx_packets++; 2056 tx_bytes += skb->len; 2057 txq_trans_cond_update(txq); 2058 ret = NETDEV_TX_OK; 2059 goto out; 2060 2061 tx_flush_err: 2062 dev_kfree_skb_any(skb); 2063 tx_buff->skb = NULL; 2064 tx_pool->consumer_index = tx_pool->consumer_index == 0 ? 2065 tx_pool->num_buffers - 1 : 2066 tx_pool->consumer_index - 1; 2067 tx_dropped++; 2068 tx_err: 2069 if (lpar_rc != H_CLOSED && lpar_rc != H_PARAMETER) 2070 dev_err_ratelimited(dev, "tx: send failed\n"); 2071 2072 if (lpar_rc == H_CLOSED || adapter->failover_pending) { 2073 /* Disable TX and report carrier off if queue is closed 2074 * or pending failover. 2075 * Firmware guarantees that a signal will be sent to the 2076 * driver, triggering a reset or some other action. 2077 */ 2078 netif_tx_stop_all_queues(netdev); 2079 netif_carrier_off(netdev); 2080 } 2081 out: 2082 netdev->stats.tx_dropped += tx_dropped; 2083 netdev->stats.tx_bytes += tx_bytes; 2084 netdev->stats.tx_packets += tx_packets; 2085 adapter->tx_send_failed += tx_send_failed; 2086 adapter->tx_map_failed += tx_map_failed; 2087 adapter->tx_stats_buffers[queue_num].packets += tx_packets; 2088 adapter->tx_stats_buffers[queue_num].bytes += tx_bytes; 2089 adapter->tx_stats_buffers[queue_num].dropped_packets += tx_dropped; 2090 2091 return ret; 2092 } 2093 2094 static void ibmvnic_set_multi(struct net_device *netdev) 2095 { 2096 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 2097 struct netdev_hw_addr *ha; 2098 union ibmvnic_crq crq; 2099 2100 memset(&crq, 0, sizeof(crq)); 2101 crq.request_capability.first = IBMVNIC_CRQ_CMD; 2102 crq.request_capability.cmd = REQUEST_CAPABILITY; 2103 2104 if (netdev->flags & IFF_PROMISC) { 2105 if (!adapter->promisc_supported) 2106 return; 2107 } else { 2108 if (netdev->flags & IFF_ALLMULTI) { 2109 /* Accept all multicast */ 2110 memset(&crq, 0, sizeof(crq)); 2111 crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD; 2112 crq.multicast_ctrl.cmd = MULTICAST_CTRL; 2113 crq.multicast_ctrl.flags = IBMVNIC_ENABLE_ALL; 2114 ibmvnic_send_crq(adapter, &crq); 2115 } else if (netdev_mc_empty(netdev)) { 2116 /* Reject all multicast */ 2117 memset(&crq, 0, sizeof(crq)); 2118 crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD; 2119 crq.multicast_ctrl.cmd = MULTICAST_CTRL; 2120 crq.multicast_ctrl.flags = IBMVNIC_DISABLE_ALL; 2121 ibmvnic_send_crq(adapter, &crq); 2122 } else { 2123 /* Accept one or more multicast(s) */ 2124 netdev_for_each_mc_addr(ha, netdev) { 2125 memset(&crq, 0, sizeof(crq)); 2126 crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD; 2127 crq.multicast_ctrl.cmd = MULTICAST_CTRL; 2128 crq.multicast_ctrl.flags = IBMVNIC_ENABLE_MC; 2129 ether_addr_copy(&crq.multicast_ctrl.mac_addr[0], 2130 ha->addr); 2131 ibmvnic_send_crq(adapter, &crq); 2132 } 2133 } 2134 } 2135 } 2136 2137 static int __ibmvnic_set_mac(struct net_device *netdev, u8 *dev_addr) 2138 { 2139 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 2140 union ibmvnic_crq crq; 2141 int rc; 2142 2143 if (!is_valid_ether_addr(dev_addr)) { 2144 rc = -EADDRNOTAVAIL; 2145 goto err; 2146 } 2147 2148 memset(&crq, 0, sizeof(crq)); 2149 crq.change_mac_addr.first = IBMVNIC_CRQ_CMD; 2150 crq.change_mac_addr.cmd = CHANGE_MAC_ADDR; 2151 ether_addr_copy(&crq.change_mac_addr.mac_addr[0], dev_addr); 2152 2153 mutex_lock(&adapter->fw_lock); 2154 adapter->fw_done_rc = 0; 2155 reinit_completion(&adapter->fw_done); 2156 2157 rc = ibmvnic_send_crq(adapter, &crq); 2158 if (rc) { 2159 rc = -EIO; 2160 mutex_unlock(&adapter->fw_lock); 2161 goto err; 2162 } 2163 2164 rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000); 2165 /* netdev->dev_addr is changed in handle_change_mac_rsp function */ 2166 if (rc || adapter->fw_done_rc) { 2167 rc = -EIO; 2168 mutex_unlock(&adapter->fw_lock); 2169 goto err; 2170 } 2171 mutex_unlock(&adapter->fw_lock); 2172 return 0; 2173 err: 2174 ether_addr_copy(adapter->mac_addr, netdev->dev_addr); 2175 return rc; 2176 } 2177 2178 static int ibmvnic_set_mac(struct net_device *netdev, void *p) 2179 { 2180 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 2181 struct sockaddr *addr = p; 2182 int rc; 2183 2184 rc = 0; 2185 if (!is_valid_ether_addr(addr->sa_data)) 2186 return -EADDRNOTAVAIL; 2187 2188 ether_addr_copy(adapter->mac_addr, addr->sa_data); 2189 if (adapter->state != VNIC_PROBED) 2190 rc = __ibmvnic_set_mac(netdev, addr->sa_data); 2191 2192 return rc; 2193 } 2194 2195 static const char *reset_reason_to_string(enum ibmvnic_reset_reason reason) 2196 { 2197 switch (reason) { 2198 case VNIC_RESET_FAILOVER: 2199 return "FAILOVER"; 2200 case VNIC_RESET_MOBILITY: 2201 return "MOBILITY"; 2202 case VNIC_RESET_FATAL: 2203 return "FATAL"; 2204 case VNIC_RESET_NON_FATAL: 2205 return "NON_FATAL"; 2206 case VNIC_RESET_TIMEOUT: 2207 return "TIMEOUT"; 2208 case VNIC_RESET_CHANGE_PARAM: 2209 return "CHANGE_PARAM"; 2210 case VNIC_RESET_PASSIVE_INIT: 2211 return "PASSIVE_INIT"; 2212 } 2213 return "UNKNOWN"; 2214 } 2215 2216 /* 2217 * Initialize the init_done completion and return code values. We 2218 * can get a transport event just after registering the CRQ and the 2219 * tasklet will use this to communicate the transport event. To ensure 2220 * we don't miss the notification/error, initialize these _before_ 2221 * regisering the CRQ. 2222 */ 2223 static inline void reinit_init_done(struct ibmvnic_adapter *adapter) 2224 { 2225 reinit_completion(&adapter->init_done); 2226 adapter->init_done_rc = 0; 2227 } 2228 2229 /* 2230 * do_reset returns zero if we are able to keep processing reset events, or 2231 * non-zero if we hit a fatal error and must halt. 2232 */ 2233 static int do_reset(struct ibmvnic_adapter *adapter, 2234 struct ibmvnic_rwi *rwi, u32 reset_state) 2235 { 2236 struct net_device *netdev = adapter->netdev; 2237 u64 old_num_rx_queues, old_num_tx_queues; 2238 u64 old_num_rx_slots, old_num_tx_slots; 2239 int rc; 2240 2241 netdev_dbg(adapter->netdev, 2242 "[S:%s FOP:%d] Reset reason: %s, reset_state: %s\n", 2243 adapter_state_to_string(adapter->state), 2244 adapter->failover_pending, 2245 reset_reason_to_string(rwi->reset_reason), 2246 adapter_state_to_string(reset_state)); 2247 2248 adapter->reset_reason = rwi->reset_reason; 2249 /* requestor of VNIC_RESET_CHANGE_PARAM already has the rtnl lock */ 2250 if (!(adapter->reset_reason == VNIC_RESET_CHANGE_PARAM)) 2251 rtnl_lock(); 2252 2253 /* Now that we have the rtnl lock, clear any pending failover. 2254 * This will ensure ibmvnic_open() has either completed or will 2255 * block until failover is complete. 2256 */ 2257 if (rwi->reset_reason == VNIC_RESET_FAILOVER) 2258 adapter->failover_pending = false; 2259 2260 /* read the state and check (again) after getting rtnl */ 2261 reset_state = adapter->state; 2262 2263 if (reset_state == VNIC_REMOVING || reset_state == VNIC_REMOVED) { 2264 rc = -EBUSY; 2265 goto out; 2266 } 2267 2268 netif_carrier_off(netdev); 2269 2270 old_num_rx_queues = adapter->req_rx_queues; 2271 old_num_tx_queues = adapter->req_tx_queues; 2272 old_num_rx_slots = adapter->req_rx_add_entries_per_subcrq; 2273 old_num_tx_slots = adapter->req_tx_entries_per_subcrq; 2274 2275 ibmvnic_cleanup(netdev); 2276 2277 if (reset_state == VNIC_OPEN && 2278 adapter->reset_reason != VNIC_RESET_MOBILITY && 2279 adapter->reset_reason != VNIC_RESET_FAILOVER) { 2280 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) { 2281 rc = __ibmvnic_close(netdev); 2282 if (rc) 2283 goto out; 2284 } else { 2285 adapter->state = VNIC_CLOSING; 2286 2287 /* Release the RTNL lock before link state change and 2288 * re-acquire after the link state change to allow 2289 * linkwatch_event to grab the RTNL lock and run during 2290 * a reset. 2291 */ 2292 rtnl_unlock(); 2293 rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN); 2294 rtnl_lock(); 2295 if (rc) 2296 goto out; 2297 2298 if (adapter->state == VNIC_OPEN) { 2299 /* When we dropped rtnl, ibmvnic_open() got 2300 * it and noticed that we are resetting and 2301 * set the adapter state to OPEN. Update our 2302 * new "target" state, and resume the reset 2303 * from VNIC_CLOSING state. 2304 */ 2305 netdev_dbg(netdev, 2306 "Open changed state from %s, updating.\n", 2307 adapter_state_to_string(reset_state)); 2308 reset_state = VNIC_OPEN; 2309 adapter->state = VNIC_CLOSING; 2310 } 2311 2312 if (adapter->state != VNIC_CLOSING) { 2313 /* If someone else changed the adapter state 2314 * when we dropped the rtnl, fail the reset 2315 */ 2316 rc = -EAGAIN; 2317 goto out; 2318 } 2319 adapter->state = VNIC_CLOSED; 2320 } 2321 } 2322 2323 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) { 2324 release_resources(adapter); 2325 release_sub_crqs(adapter, 1); 2326 release_crq_queue(adapter); 2327 } 2328 2329 if (adapter->reset_reason != VNIC_RESET_NON_FATAL) { 2330 /* remove the closed state so when we call open it appears 2331 * we are coming from the probed state. 2332 */ 2333 adapter->state = VNIC_PROBED; 2334 2335 reinit_init_done(adapter); 2336 2337 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) { 2338 rc = init_crq_queue(adapter); 2339 } else if (adapter->reset_reason == VNIC_RESET_MOBILITY) { 2340 rc = ibmvnic_reenable_crq_queue(adapter); 2341 release_sub_crqs(adapter, 1); 2342 } else { 2343 rc = ibmvnic_reset_crq(adapter); 2344 if (rc == H_CLOSED || rc == H_SUCCESS) { 2345 rc = vio_enable_interrupts(adapter->vdev); 2346 if (rc) 2347 netdev_err(adapter->netdev, 2348 "Reset failed to enable interrupts. rc=%d\n", 2349 rc); 2350 } 2351 } 2352 2353 if (rc) { 2354 netdev_err(adapter->netdev, 2355 "Reset couldn't initialize crq. rc=%d\n", rc); 2356 goto out; 2357 } 2358 2359 rc = ibmvnic_reset_init(adapter, true); 2360 if (rc) 2361 goto out; 2362 2363 /* If the adapter was in PROBE or DOWN state prior to the reset, 2364 * exit here. 2365 */ 2366 if (reset_state == VNIC_PROBED || reset_state == VNIC_DOWN) { 2367 rc = 0; 2368 goto out; 2369 } 2370 2371 rc = ibmvnic_login(netdev); 2372 if (rc) 2373 goto out; 2374 2375 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) { 2376 rc = init_resources(adapter); 2377 if (rc) 2378 goto out; 2379 } else if (adapter->req_rx_queues != old_num_rx_queues || 2380 adapter->req_tx_queues != old_num_tx_queues || 2381 adapter->req_rx_add_entries_per_subcrq != 2382 old_num_rx_slots || 2383 adapter->req_tx_entries_per_subcrq != 2384 old_num_tx_slots || 2385 !adapter->rx_pool || 2386 !adapter->tso_pool || 2387 !adapter->tx_pool) { 2388 release_napi(adapter); 2389 release_vpd_data(adapter); 2390 2391 rc = init_resources(adapter); 2392 if (rc) 2393 goto out; 2394 2395 } else { 2396 rc = init_tx_pools(netdev); 2397 if (rc) { 2398 netdev_dbg(netdev, 2399 "init tx pools failed (%d)\n", 2400 rc); 2401 goto out; 2402 } 2403 2404 rc = init_rx_pools(netdev); 2405 if (rc) { 2406 netdev_dbg(netdev, 2407 "init rx pools failed (%d)\n", 2408 rc); 2409 goto out; 2410 } 2411 } 2412 ibmvnic_disable_irqs(adapter); 2413 } 2414 adapter->state = VNIC_CLOSED; 2415 2416 if (reset_state == VNIC_CLOSED) { 2417 rc = 0; 2418 goto out; 2419 } 2420 2421 rc = __ibmvnic_open(netdev); 2422 if (rc) { 2423 rc = IBMVNIC_OPEN_FAILED; 2424 goto out; 2425 } 2426 2427 /* refresh device's multicast list */ 2428 ibmvnic_set_multi(netdev); 2429 2430 if (adapter->reset_reason == VNIC_RESET_FAILOVER || 2431 adapter->reset_reason == VNIC_RESET_MOBILITY) 2432 __netdev_notify_peers(netdev); 2433 2434 rc = 0; 2435 2436 out: 2437 /* restore the adapter state if reset failed */ 2438 if (rc) 2439 adapter->state = reset_state; 2440 /* requestor of VNIC_RESET_CHANGE_PARAM should still hold the rtnl lock */ 2441 if (!(adapter->reset_reason == VNIC_RESET_CHANGE_PARAM)) 2442 rtnl_unlock(); 2443 2444 netdev_dbg(adapter->netdev, "[S:%s FOP:%d] Reset done, rc %d\n", 2445 adapter_state_to_string(adapter->state), 2446 adapter->failover_pending, rc); 2447 return rc; 2448 } 2449 2450 static int do_hard_reset(struct ibmvnic_adapter *adapter, 2451 struct ibmvnic_rwi *rwi, u32 reset_state) 2452 { 2453 struct net_device *netdev = adapter->netdev; 2454 int rc; 2455 2456 netdev_dbg(adapter->netdev, "Hard resetting driver (%s)\n", 2457 reset_reason_to_string(rwi->reset_reason)); 2458 2459 /* read the state and check (again) after getting rtnl */ 2460 reset_state = adapter->state; 2461 2462 if (reset_state == VNIC_REMOVING || reset_state == VNIC_REMOVED) { 2463 rc = -EBUSY; 2464 goto out; 2465 } 2466 2467 netif_carrier_off(netdev); 2468 adapter->reset_reason = rwi->reset_reason; 2469 2470 ibmvnic_cleanup(netdev); 2471 release_resources(adapter); 2472 release_sub_crqs(adapter, 0); 2473 release_crq_queue(adapter); 2474 2475 /* remove the closed state so when we call open it appears 2476 * we are coming from the probed state. 2477 */ 2478 adapter->state = VNIC_PROBED; 2479 2480 reinit_init_done(adapter); 2481 2482 rc = init_crq_queue(adapter); 2483 if (rc) { 2484 netdev_err(adapter->netdev, 2485 "Couldn't initialize crq. rc=%d\n", rc); 2486 goto out; 2487 } 2488 2489 rc = ibmvnic_reset_init(adapter, false); 2490 if (rc) 2491 goto out; 2492 2493 /* If the adapter was in PROBE or DOWN state prior to the reset, 2494 * exit here. 2495 */ 2496 if (reset_state == VNIC_PROBED || reset_state == VNIC_DOWN) 2497 goto out; 2498 2499 rc = ibmvnic_login(netdev); 2500 if (rc) 2501 goto out; 2502 2503 rc = init_resources(adapter); 2504 if (rc) 2505 goto out; 2506 2507 ibmvnic_disable_irqs(adapter); 2508 adapter->state = VNIC_CLOSED; 2509 2510 if (reset_state == VNIC_CLOSED) 2511 goto out; 2512 2513 rc = __ibmvnic_open(netdev); 2514 if (rc) { 2515 rc = IBMVNIC_OPEN_FAILED; 2516 goto out; 2517 } 2518 2519 __netdev_notify_peers(netdev); 2520 out: 2521 /* restore adapter state if reset failed */ 2522 if (rc) 2523 adapter->state = reset_state; 2524 netdev_dbg(adapter->netdev, "[S:%s FOP:%d] Hard reset done, rc %d\n", 2525 adapter_state_to_string(adapter->state), 2526 adapter->failover_pending, rc); 2527 return rc; 2528 } 2529 2530 static struct ibmvnic_rwi *get_next_rwi(struct ibmvnic_adapter *adapter) 2531 { 2532 struct ibmvnic_rwi *rwi; 2533 unsigned long flags; 2534 2535 spin_lock_irqsave(&adapter->rwi_lock, flags); 2536 2537 if (!list_empty(&adapter->rwi_list)) { 2538 rwi = list_first_entry(&adapter->rwi_list, struct ibmvnic_rwi, 2539 list); 2540 list_del(&rwi->list); 2541 } else { 2542 rwi = NULL; 2543 } 2544 2545 spin_unlock_irqrestore(&adapter->rwi_lock, flags); 2546 return rwi; 2547 } 2548 2549 /** 2550 * do_passive_init - complete probing when partner device is detected. 2551 * @adapter: ibmvnic_adapter struct 2552 * 2553 * If the ibmvnic device does not have a partner device to communicate with at boot 2554 * and that partner device comes online at a later time, this function is called 2555 * to complete the initialization process of ibmvnic device. 2556 * Caller is expected to hold rtnl_lock(). 2557 * 2558 * Returns non-zero if sub-CRQs are not initialized properly leaving the device 2559 * in the down state. 2560 * Returns 0 upon success and the device is in PROBED state. 2561 */ 2562 2563 static int do_passive_init(struct ibmvnic_adapter *adapter) 2564 { 2565 unsigned long timeout = msecs_to_jiffies(30000); 2566 struct net_device *netdev = adapter->netdev; 2567 struct device *dev = &adapter->vdev->dev; 2568 int rc; 2569 2570 netdev_dbg(netdev, "Partner device found, probing.\n"); 2571 2572 adapter->state = VNIC_PROBING; 2573 reinit_completion(&adapter->init_done); 2574 adapter->init_done_rc = 0; 2575 adapter->crq.active = true; 2576 2577 rc = send_crq_init_complete(adapter); 2578 if (rc) 2579 goto out; 2580 2581 rc = send_version_xchg(adapter); 2582 if (rc) 2583 netdev_dbg(adapter->netdev, "send_version_xchg failed, rc=%d\n", rc); 2584 2585 if (!wait_for_completion_timeout(&adapter->init_done, timeout)) { 2586 dev_err(dev, "Initialization sequence timed out\n"); 2587 rc = -ETIMEDOUT; 2588 goto out; 2589 } 2590 2591 rc = init_sub_crqs(adapter); 2592 if (rc) { 2593 dev_err(dev, "Initialization of sub crqs failed, rc=%d\n", rc); 2594 goto out; 2595 } 2596 2597 rc = init_sub_crq_irqs(adapter); 2598 if (rc) { 2599 dev_err(dev, "Failed to initialize sub crq irqs\n, rc=%d", rc); 2600 goto init_failed; 2601 } 2602 2603 netdev->mtu = adapter->req_mtu - ETH_HLEN; 2604 netdev->min_mtu = adapter->min_mtu - ETH_HLEN; 2605 netdev->max_mtu = adapter->max_mtu - ETH_HLEN; 2606 2607 adapter->state = VNIC_PROBED; 2608 netdev_dbg(netdev, "Probed successfully. Waiting for signal from partner device.\n"); 2609 2610 return 0; 2611 2612 init_failed: 2613 release_sub_crqs(adapter, 1); 2614 out: 2615 adapter->state = VNIC_DOWN; 2616 return rc; 2617 } 2618 2619 static void __ibmvnic_reset(struct work_struct *work) 2620 { 2621 struct ibmvnic_adapter *adapter; 2622 unsigned int timeout = 5000; 2623 struct ibmvnic_rwi *tmprwi; 2624 bool saved_state = false; 2625 struct ibmvnic_rwi *rwi; 2626 unsigned long flags; 2627 struct device *dev; 2628 bool need_reset; 2629 int num_fails = 0; 2630 u32 reset_state; 2631 int rc = 0; 2632 2633 adapter = container_of(work, struct ibmvnic_adapter, ibmvnic_reset); 2634 dev = &adapter->vdev->dev; 2635 2636 /* Wait for ibmvnic_probe() to complete. If probe is taking too long 2637 * or if another reset is in progress, defer work for now. If probe 2638 * eventually fails it will flush and terminate our work. 2639 * 2640 * Three possibilities here: 2641 * 1. Adpater being removed - just return 2642 * 2. Timed out on probe or another reset in progress - delay the work 2643 * 3. Completed probe - perform any resets in queue 2644 */ 2645 if (adapter->state == VNIC_PROBING && 2646 !wait_for_completion_timeout(&adapter->probe_done, timeout)) { 2647 dev_err(dev, "Reset thread timed out on probe"); 2648 queue_delayed_work(system_long_wq, 2649 &adapter->ibmvnic_delayed_reset, 2650 IBMVNIC_RESET_DELAY); 2651 return; 2652 } 2653 2654 /* adapter is done with probe (i.e state is never VNIC_PROBING now) */ 2655 if (adapter->state == VNIC_REMOVING) 2656 return; 2657 2658 /* ->rwi_list is stable now (no one else is removing entries) */ 2659 2660 /* ibmvnic_probe() may have purged the reset queue after we were 2661 * scheduled to process a reset so there maybe no resets to process. 2662 * Before setting the ->resetting bit though, we have to make sure 2663 * that there is infact a reset to process. Otherwise we may race 2664 * with ibmvnic_open() and end up leaving the vnic down: 2665 * 2666 * __ibmvnic_reset() ibmvnic_open() 2667 * ----------------- -------------- 2668 * 2669 * set ->resetting bit 2670 * find ->resetting bit is set 2671 * set ->state to IBMVNIC_OPEN (i.e 2672 * assume reset will open device) 2673 * return 2674 * find reset queue empty 2675 * return 2676 * 2677 * Neither performed vnic login/open and vnic stays down 2678 * 2679 * If we hold the lock and conditionally set the bit, either we 2680 * or ibmvnic_open() will complete the open. 2681 */ 2682 need_reset = false; 2683 spin_lock(&adapter->rwi_lock); 2684 if (!list_empty(&adapter->rwi_list)) { 2685 if (test_and_set_bit_lock(0, &adapter->resetting)) { 2686 queue_delayed_work(system_long_wq, 2687 &adapter->ibmvnic_delayed_reset, 2688 IBMVNIC_RESET_DELAY); 2689 } else { 2690 need_reset = true; 2691 } 2692 } 2693 spin_unlock(&adapter->rwi_lock); 2694 2695 if (!need_reset) 2696 return; 2697 2698 rwi = get_next_rwi(adapter); 2699 while (rwi) { 2700 spin_lock_irqsave(&adapter->state_lock, flags); 2701 2702 if (adapter->state == VNIC_REMOVING || 2703 adapter->state == VNIC_REMOVED) { 2704 spin_unlock_irqrestore(&adapter->state_lock, flags); 2705 kfree(rwi); 2706 rc = EBUSY; 2707 break; 2708 } 2709 2710 if (!saved_state) { 2711 reset_state = adapter->state; 2712 saved_state = true; 2713 } 2714 spin_unlock_irqrestore(&adapter->state_lock, flags); 2715 2716 if (rwi->reset_reason == VNIC_RESET_PASSIVE_INIT) { 2717 rtnl_lock(); 2718 rc = do_passive_init(adapter); 2719 rtnl_unlock(); 2720 if (!rc) 2721 netif_carrier_on(adapter->netdev); 2722 } else if (adapter->force_reset_recovery) { 2723 /* Since we are doing a hard reset now, clear the 2724 * failover_pending flag so we don't ignore any 2725 * future MOBILITY or other resets. 2726 */ 2727 adapter->failover_pending = false; 2728 2729 /* Transport event occurred during previous reset */ 2730 if (adapter->wait_for_reset) { 2731 /* Previous was CHANGE_PARAM; caller locked */ 2732 adapter->force_reset_recovery = false; 2733 rc = do_hard_reset(adapter, rwi, reset_state); 2734 } else { 2735 rtnl_lock(); 2736 adapter->force_reset_recovery = false; 2737 rc = do_hard_reset(adapter, rwi, reset_state); 2738 rtnl_unlock(); 2739 } 2740 if (rc) 2741 num_fails++; 2742 else 2743 num_fails = 0; 2744 2745 /* If auto-priority-failover is enabled we can get 2746 * back to back failovers during resets, resulting 2747 * in at least two failed resets (from high-priority 2748 * backing device to low-priority one and then back) 2749 * If resets continue to fail beyond that, give the 2750 * adapter some time to settle down before retrying. 2751 */ 2752 if (num_fails >= 3) { 2753 netdev_dbg(adapter->netdev, 2754 "[S:%s] Hard reset failed %d times, waiting 60 secs\n", 2755 adapter_state_to_string(adapter->state), 2756 num_fails); 2757 set_current_state(TASK_UNINTERRUPTIBLE); 2758 schedule_timeout(60 * HZ); 2759 } 2760 } else { 2761 rc = do_reset(adapter, rwi, reset_state); 2762 } 2763 tmprwi = rwi; 2764 adapter->last_reset_time = jiffies; 2765 2766 if (rc) 2767 netdev_dbg(adapter->netdev, "Reset failed, rc=%d\n", rc); 2768 2769 rwi = get_next_rwi(adapter); 2770 2771 /* 2772 * If there is another reset queued, free the previous rwi 2773 * and process the new reset even if previous reset failed 2774 * (the previous reset could have failed because of a fail 2775 * over for instance, so process the fail over). 2776 * 2777 * If there are no resets queued and the previous reset failed, 2778 * the adapter would be in an undefined state. So retry the 2779 * previous reset as a hard reset. 2780 */ 2781 if (rwi) 2782 kfree(tmprwi); 2783 else if (rc) 2784 rwi = tmprwi; 2785 2786 if (rwi && (rwi->reset_reason == VNIC_RESET_FAILOVER || 2787 rwi->reset_reason == VNIC_RESET_MOBILITY || rc)) 2788 adapter->force_reset_recovery = true; 2789 } 2790 2791 if (adapter->wait_for_reset) { 2792 adapter->reset_done_rc = rc; 2793 complete(&adapter->reset_done); 2794 } 2795 2796 clear_bit_unlock(0, &adapter->resetting); 2797 2798 netdev_dbg(adapter->netdev, 2799 "[S:%s FRR:%d WFR:%d] Done processing resets\n", 2800 adapter_state_to_string(adapter->state), 2801 adapter->force_reset_recovery, 2802 adapter->wait_for_reset); 2803 } 2804 2805 static void __ibmvnic_delayed_reset(struct work_struct *work) 2806 { 2807 struct ibmvnic_adapter *adapter; 2808 2809 adapter = container_of(work, struct ibmvnic_adapter, 2810 ibmvnic_delayed_reset.work); 2811 __ibmvnic_reset(&adapter->ibmvnic_reset); 2812 } 2813 2814 static void flush_reset_queue(struct ibmvnic_adapter *adapter) 2815 { 2816 struct list_head *entry, *tmp_entry; 2817 2818 if (!list_empty(&adapter->rwi_list)) { 2819 list_for_each_safe(entry, tmp_entry, &adapter->rwi_list) { 2820 list_del(entry); 2821 kfree(list_entry(entry, struct ibmvnic_rwi, list)); 2822 } 2823 } 2824 } 2825 2826 static int ibmvnic_reset(struct ibmvnic_adapter *adapter, 2827 enum ibmvnic_reset_reason reason) 2828 { 2829 struct net_device *netdev = adapter->netdev; 2830 struct ibmvnic_rwi *rwi, *tmp; 2831 unsigned long flags; 2832 int ret; 2833 2834 spin_lock_irqsave(&adapter->rwi_lock, flags); 2835 2836 /* If failover is pending don't schedule any other reset. 2837 * Instead let the failover complete. If there is already a 2838 * a failover reset scheduled, we will detect and drop the 2839 * duplicate reset when walking the ->rwi_list below. 2840 */ 2841 if (adapter->state == VNIC_REMOVING || 2842 adapter->state == VNIC_REMOVED || 2843 (adapter->failover_pending && reason != VNIC_RESET_FAILOVER)) { 2844 ret = EBUSY; 2845 netdev_dbg(netdev, "Adapter removing or pending failover, skipping reset\n"); 2846 goto err; 2847 } 2848 2849 list_for_each_entry(tmp, &adapter->rwi_list, list) { 2850 if (tmp->reset_reason == reason) { 2851 netdev_dbg(netdev, "Skipping matching reset, reason=%s\n", 2852 reset_reason_to_string(reason)); 2853 ret = EBUSY; 2854 goto err; 2855 } 2856 } 2857 2858 rwi = kzalloc(sizeof(*rwi), GFP_ATOMIC); 2859 if (!rwi) { 2860 ret = ENOMEM; 2861 goto err; 2862 } 2863 /* if we just received a transport event, 2864 * flush reset queue and process this reset 2865 */ 2866 if (adapter->force_reset_recovery) 2867 flush_reset_queue(adapter); 2868 2869 rwi->reset_reason = reason; 2870 list_add_tail(&rwi->list, &adapter->rwi_list); 2871 netdev_dbg(adapter->netdev, "Scheduling reset (reason %s)\n", 2872 reset_reason_to_string(reason)); 2873 queue_work(system_long_wq, &adapter->ibmvnic_reset); 2874 2875 ret = 0; 2876 err: 2877 /* ibmvnic_close() below can block, so drop the lock first */ 2878 spin_unlock_irqrestore(&adapter->rwi_lock, flags); 2879 2880 if (ret == ENOMEM) 2881 ibmvnic_close(netdev); 2882 2883 return -ret; 2884 } 2885 2886 static void ibmvnic_tx_timeout(struct net_device *dev, unsigned int txqueue) 2887 { 2888 struct ibmvnic_adapter *adapter = netdev_priv(dev); 2889 2890 if (test_bit(0, &adapter->resetting)) { 2891 netdev_err(adapter->netdev, 2892 "Adapter is resetting, skip timeout reset\n"); 2893 return; 2894 } 2895 /* No queuing up reset until at least 5 seconds (default watchdog val) 2896 * after last reset 2897 */ 2898 if (time_before(jiffies, (adapter->last_reset_time + dev->watchdog_timeo))) { 2899 netdev_dbg(dev, "Not yet time to tx timeout.\n"); 2900 return; 2901 } 2902 ibmvnic_reset(adapter, VNIC_RESET_TIMEOUT); 2903 } 2904 2905 static void remove_buff_from_pool(struct ibmvnic_adapter *adapter, 2906 struct ibmvnic_rx_buff *rx_buff) 2907 { 2908 struct ibmvnic_rx_pool *pool = &adapter->rx_pool[rx_buff->pool_index]; 2909 2910 rx_buff->skb = NULL; 2911 2912 pool->free_map[pool->next_alloc] = (int)(rx_buff - pool->rx_buff); 2913 pool->next_alloc = (pool->next_alloc + 1) % pool->size; 2914 2915 atomic_dec(&pool->available); 2916 } 2917 2918 static int ibmvnic_poll(struct napi_struct *napi, int budget) 2919 { 2920 struct ibmvnic_sub_crq_queue *rx_scrq; 2921 struct ibmvnic_adapter *adapter; 2922 struct net_device *netdev; 2923 int frames_processed; 2924 int scrq_num; 2925 2926 netdev = napi->dev; 2927 adapter = netdev_priv(netdev); 2928 scrq_num = (int)(napi - adapter->napi); 2929 frames_processed = 0; 2930 rx_scrq = adapter->rx_scrq[scrq_num]; 2931 2932 restart_poll: 2933 while (frames_processed < budget) { 2934 struct sk_buff *skb; 2935 struct ibmvnic_rx_buff *rx_buff; 2936 union sub_crq *next; 2937 u32 length; 2938 u16 offset; 2939 u8 flags = 0; 2940 2941 if (unlikely(test_bit(0, &adapter->resetting) && 2942 adapter->reset_reason != VNIC_RESET_NON_FATAL)) { 2943 enable_scrq_irq(adapter, rx_scrq); 2944 napi_complete_done(napi, frames_processed); 2945 return frames_processed; 2946 } 2947 2948 if (!pending_scrq(adapter, rx_scrq)) 2949 break; 2950 next = ibmvnic_next_scrq(adapter, rx_scrq); 2951 rx_buff = (struct ibmvnic_rx_buff *) 2952 be64_to_cpu(next->rx_comp.correlator); 2953 /* do error checking */ 2954 if (next->rx_comp.rc) { 2955 netdev_dbg(netdev, "rx buffer returned with rc %x\n", 2956 be16_to_cpu(next->rx_comp.rc)); 2957 /* free the entry */ 2958 next->rx_comp.first = 0; 2959 dev_kfree_skb_any(rx_buff->skb); 2960 remove_buff_from_pool(adapter, rx_buff); 2961 continue; 2962 } else if (!rx_buff->skb) { 2963 /* free the entry */ 2964 next->rx_comp.first = 0; 2965 remove_buff_from_pool(adapter, rx_buff); 2966 continue; 2967 } 2968 2969 length = be32_to_cpu(next->rx_comp.len); 2970 offset = be16_to_cpu(next->rx_comp.off_frame_data); 2971 flags = next->rx_comp.flags; 2972 skb = rx_buff->skb; 2973 /* load long_term_buff before copying to skb */ 2974 dma_rmb(); 2975 skb_copy_to_linear_data(skb, rx_buff->data + offset, 2976 length); 2977 2978 /* VLAN Header has been stripped by the system firmware and 2979 * needs to be inserted by the driver 2980 */ 2981 if (adapter->rx_vlan_header_insertion && 2982 (flags & IBMVNIC_VLAN_STRIPPED)) 2983 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), 2984 ntohs(next->rx_comp.vlan_tci)); 2985 2986 /* free the entry */ 2987 next->rx_comp.first = 0; 2988 remove_buff_from_pool(adapter, rx_buff); 2989 2990 skb_put(skb, length); 2991 skb->protocol = eth_type_trans(skb, netdev); 2992 skb_record_rx_queue(skb, scrq_num); 2993 2994 if (flags & IBMVNIC_IP_CHKSUM_GOOD && 2995 flags & IBMVNIC_TCP_UDP_CHKSUM_GOOD) { 2996 skb->ip_summed = CHECKSUM_UNNECESSARY; 2997 } 2998 2999 length = skb->len; 3000 napi_gro_receive(napi, skb); /* send it up */ 3001 netdev->stats.rx_packets++; 3002 netdev->stats.rx_bytes += length; 3003 adapter->rx_stats_buffers[scrq_num].packets++; 3004 adapter->rx_stats_buffers[scrq_num].bytes += length; 3005 frames_processed++; 3006 } 3007 3008 if (adapter->state != VNIC_CLOSING && 3009 ((atomic_read(&adapter->rx_pool[scrq_num].available) < 3010 adapter->req_rx_add_entries_per_subcrq / 2) || 3011 frames_processed < budget)) 3012 replenish_rx_pool(adapter, &adapter->rx_pool[scrq_num]); 3013 if (frames_processed < budget) { 3014 if (napi_complete_done(napi, frames_processed)) { 3015 enable_scrq_irq(adapter, rx_scrq); 3016 if (pending_scrq(adapter, rx_scrq)) { 3017 if (napi_reschedule(napi)) { 3018 disable_scrq_irq(adapter, rx_scrq); 3019 goto restart_poll; 3020 } 3021 } 3022 } 3023 } 3024 return frames_processed; 3025 } 3026 3027 static int wait_for_reset(struct ibmvnic_adapter *adapter) 3028 { 3029 int rc, ret; 3030 3031 adapter->fallback.mtu = adapter->req_mtu; 3032 adapter->fallback.rx_queues = adapter->req_rx_queues; 3033 adapter->fallback.tx_queues = adapter->req_tx_queues; 3034 adapter->fallback.rx_entries = adapter->req_rx_add_entries_per_subcrq; 3035 adapter->fallback.tx_entries = adapter->req_tx_entries_per_subcrq; 3036 3037 reinit_completion(&adapter->reset_done); 3038 adapter->wait_for_reset = true; 3039 rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM); 3040 3041 if (rc) { 3042 ret = rc; 3043 goto out; 3044 } 3045 rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done, 60000); 3046 if (rc) { 3047 ret = -ENODEV; 3048 goto out; 3049 } 3050 3051 ret = 0; 3052 if (adapter->reset_done_rc) { 3053 ret = -EIO; 3054 adapter->desired.mtu = adapter->fallback.mtu; 3055 adapter->desired.rx_queues = adapter->fallback.rx_queues; 3056 adapter->desired.tx_queues = adapter->fallback.tx_queues; 3057 adapter->desired.rx_entries = adapter->fallback.rx_entries; 3058 adapter->desired.tx_entries = adapter->fallback.tx_entries; 3059 3060 reinit_completion(&adapter->reset_done); 3061 adapter->wait_for_reset = true; 3062 rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM); 3063 if (rc) { 3064 ret = rc; 3065 goto out; 3066 } 3067 rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done, 3068 60000); 3069 if (rc) { 3070 ret = -ENODEV; 3071 goto out; 3072 } 3073 } 3074 out: 3075 adapter->wait_for_reset = false; 3076 3077 return ret; 3078 } 3079 3080 static int ibmvnic_change_mtu(struct net_device *netdev, int new_mtu) 3081 { 3082 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3083 3084 adapter->desired.mtu = new_mtu + ETH_HLEN; 3085 3086 return wait_for_reset(adapter); 3087 } 3088 3089 static netdev_features_t ibmvnic_features_check(struct sk_buff *skb, 3090 struct net_device *dev, 3091 netdev_features_t features) 3092 { 3093 /* Some backing hardware adapters can not 3094 * handle packets with a MSS less than 224 3095 * or with only one segment. 3096 */ 3097 if (skb_is_gso(skb)) { 3098 if (skb_shinfo(skb)->gso_size < 224 || 3099 skb_shinfo(skb)->gso_segs == 1) 3100 features &= ~NETIF_F_GSO_MASK; 3101 } 3102 3103 return features; 3104 } 3105 3106 static const struct net_device_ops ibmvnic_netdev_ops = { 3107 .ndo_open = ibmvnic_open, 3108 .ndo_stop = ibmvnic_close, 3109 .ndo_start_xmit = ibmvnic_xmit, 3110 .ndo_set_rx_mode = ibmvnic_set_multi, 3111 .ndo_set_mac_address = ibmvnic_set_mac, 3112 .ndo_validate_addr = eth_validate_addr, 3113 .ndo_tx_timeout = ibmvnic_tx_timeout, 3114 .ndo_change_mtu = ibmvnic_change_mtu, 3115 .ndo_features_check = ibmvnic_features_check, 3116 }; 3117 3118 /* ethtool functions */ 3119 3120 static int ibmvnic_get_link_ksettings(struct net_device *netdev, 3121 struct ethtool_link_ksettings *cmd) 3122 { 3123 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3124 int rc; 3125 3126 rc = send_query_phys_parms(adapter); 3127 if (rc) { 3128 adapter->speed = SPEED_UNKNOWN; 3129 adapter->duplex = DUPLEX_UNKNOWN; 3130 } 3131 cmd->base.speed = adapter->speed; 3132 cmd->base.duplex = adapter->duplex; 3133 cmd->base.port = PORT_FIBRE; 3134 cmd->base.phy_address = 0; 3135 cmd->base.autoneg = AUTONEG_ENABLE; 3136 3137 return 0; 3138 } 3139 3140 static void ibmvnic_get_drvinfo(struct net_device *netdev, 3141 struct ethtool_drvinfo *info) 3142 { 3143 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3144 3145 strscpy(info->driver, ibmvnic_driver_name, sizeof(info->driver)); 3146 strscpy(info->version, IBMVNIC_DRIVER_VERSION, sizeof(info->version)); 3147 strscpy(info->fw_version, adapter->fw_version, 3148 sizeof(info->fw_version)); 3149 } 3150 3151 static u32 ibmvnic_get_msglevel(struct net_device *netdev) 3152 { 3153 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3154 3155 return adapter->msg_enable; 3156 } 3157 3158 static void ibmvnic_set_msglevel(struct net_device *netdev, u32 data) 3159 { 3160 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3161 3162 adapter->msg_enable = data; 3163 } 3164 3165 static u32 ibmvnic_get_link(struct net_device *netdev) 3166 { 3167 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3168 3169 /* Don't need to send a query because we request a logical link up at 3170 * init and then we wait for link state indications 3171 */ 3172 return adapter->logical_link_state; 3173 } 3174 3175 static void ibmvnic_get_ringparam(struct net_device *netdev, 3176 struct ethtool_ringparam *ring, 3177 struct kernel_ethtool_ringparam *kernel_ring, 3178 struct netlink_ext_ack *extack) 3179 { 3180 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3181 3182 if (adapter->priv_flags & IBMVNIC_USE_SERVER_MAXES) { 3183 ring->rx_max_pending = adapter->max_rx_add_entries_per_subcrq; 3184 ring->tx_max_pending = adapter->max_tx_entries_per_subcrq; 3185 } else { 3186 ring->rx_max_pending = IBMVNIC_MAX_QUEUE_SZ; 3187 ring->tx_max_pending = IBMVNIC_MAX_QUEUE_SZ; 3188 } 3189 ring->rx_mini_max_pending = 0; 3190 ring->rx_jumbo_max_pending = 0; 3191 ring->rx_pending = adapter->req_rx_add_entries_per_subcrq; 3192 ring->tx_pending = adapter->req_tx_entries_per_subcrq; 3193 ring->rx_mini_pending = 0; 3194 ring->rx_jumbo_pending = 0; 3195 } 3196 3197 static int ibmvnic_set_ringparam(struct net_device *netdev, 3198 struct ethtool_ringparam *ring, 3199 struct kernel_ethtool_ringparam *kernel_ring, 3200 struct netlink_ext_ack *extack) 3201 { 3202 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3203 int ret; 3204 3205 ret = 0; 3206 adapter->desired.rx_entries = ring->rx_pending; 3207 adapter->desired.tx_entries = ring->tx_pending; 3208 3209 ret = wait_for_reset(adapter); 3210 3211 if (!ret && 3212 (adapter->req_rx_add_entries_per_subcrq != ring->rx_pending || 3213 adapter->req_tx_entries_per_subcrq != ring->tx_pending)) 3214 netdev_info(netdev, 3215 "Could not match full ringsize request. Requested: RX %d, TX %d; Allowed: RX %llu, TX %llu\n", 3216 ring->rx_pending, ring->tx_pending, 3217 adapter->req_rx_add_entries_per_subcrq, 3218 adapter->req_tx_entries_per_subcrq); 3219 return ret; 3220 } 3221 3222 static void ibmvnic_get_channels(struct net_device *netdev, 3223 struct ethtool_channels *channels) 3224 { 3225 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3226 3227 if (adapter->priv_flags & IBMVNIC_USE_SERVER_MAXES) { 3228 channels->max_rx = adapter->max_rx_queues; 3229 channels->max_tx = adapter->max_tx_queues; 3230 } else { 3231 channels->max_rx = IBMVNIC_MAX_QUEUES; 3232 channels->max_tx = IBMVNIC_MAX_QUEUES; 3233 } 3234 3235 channels->max_other = 0; 3236 channels->max_combined = 0; 3237 channels->rx_count = adapter->req_rx_queues; 3238 channels->tx_count = adapter->req_tx_queues; 3239 channels->other_count = 0; 3240 channels->combined_count = 0; 3241 } 3242 3243 static int ibmvnic_set_channels(struct net_device *netdev, 3244 struct ethtool_channels *channels) 3245 { 3246 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3247 int ret; 3248 3249 ret = 0; 3250 adapter->desired.rx_queues = channels->rx_count; 3251 adapter->desired.tx_queues = channels->tx_count; 3252 3253 ret = wait_for_reset(adapter); 3254 3255 if (!ret && 3256 (adapter->req_rx_queues != channels->rx_count || 3257 adapter->req_tx_queues != channels->tx_count)) 3258 netdev_info(netdev, 3259 "Could not match full channels request. Requested: RX %d, TX %d; Allowed: RX %llu, TX %llu\n", 3260 channels->rx_count, channels->tx_count, 3261 adapter->req_rx_queues, adapter->req_tx_queues); 3262 return ret; 3263 } 3264 3265 static void ibmvnic_get_strings(struct net_device *dev, u32 stringset, u8 *data) 3266 { 3267 struct ibmvnic_adapter *adapter = netdev_priv(dev); 3268 int i; 3269 3270 switch (stringset) { 3271 case ETH_SS_STATS: 3272 for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); 3273 i++, data += ETH_GSTRING_LEN) 3274 memcpy(data, ibmvnic_stats[i].name, ETH_GSTRING_LEN); 3275 3276 for (i = 0; i < adapter->req_tx_queues; i++) { 3277 snprintf(data, ETH_GSTRING_LEN, "tx%d_packets", i); 3278 data += ETH_GSTRING_LEN; 3279 3280 snprintf(data, ETH_GSTRING_LEN, "tx%d_bytes", i); 3281 data += ETH_GSTRING_LEN; 3282 3283 snprintf(data, ETH_GSTRING_LEN, 3284 "tx%d_dropped_packets", i); 3285 data += ETH_GSTRING_LEN; 3286 } 3287 3288 for (i = 0; i < adapter->req_rx_queues; i++) { 3289 snprintf(data, ETH_GSTRING_LEN, "rx%d_packets", i); 3290 data += ETH_GSTRING_LEN; 3291 3292 snprintf(data, ETH_GSTRING_LEN, "rx%d_bytes", i); 3293 data += ETH_GSTRING_LEN; 3294 3295 snprintf(data, ETH_GSTRING_LEN, "rx%d_interrupts", i); 3296 data += ETH_GSTRING_LEN; 3297 } 3298 break; 3299 3300 case ETH_SS_PRIV_FLAGS: 3301 for (i = 0; i < ARRAY_SIZE(ibmvnic_priv_flags); i++) 3302 strcpy(data + i * ETH_GSTRING_LEN, 3303 ibmvnic_priv_flags[i]); 3304 break; 3305 default: 3306 return; 3307 } 3308 } 3309 3310 static int ibmvnic_get_sset_count(struct net_device *dev, int sset) 3311 { 3312 struct ibmvnic_adapter *adapter = netdev_priv(dev); 3313 3314 switch (sset) { 3315 case ETH_SS_STATS: 3316 return ARRAY_SIZE(ibmvnic_stats) + 3317 adapter->req_tx_queues * NUM_TX_STATS + 3318 adapter->req_rx_queues * NUM_RX_STATS; 3319 case ETH_SS_PRIV_FLAGS: 3320 return ARRAY_SIZE(ibmvnic_priv_flags); 3321 default: 3322 return -EOPNOTSUPP; 3323 } 3324 } 3325 3326 static void ibmvnic_get_ethtool_stats(struct net_device *dev, 3327 struct ethtool_stats *stats, u64 *data) 3328 { 3329 struct ibmvnic_adapter *adapter = netdev_priv(dev); 3330 union ibmvnic_crq crq; 3331 int i, j; 3332 int rc; 3333 3334 memset(&crq, 0, sizeof(crq)); 3335 crq.request_statistics.first = IBMVNIC_CRQ_CMD; 3336 crq.request_statistics.cmd = REQUEST_STATISTICS; 3337 crq.request_statistics.ioba = cpu_to_be32(adapter->stats_token); 3338 crq.request_statistics.len = 3339 cpu_to_be32(sizeof(struct ibmvnic_statistics)); 3340 3341 /* Wait for data to be written */ 3342 reinit_completion(&adapter->stats_done); 3343 rc = ibmvnic_send_crq(adapter, &crq); 3344 if (rc) 3345 return; 3346 rc = ibmvnic_wait_for_completion(adapter, &adapter->stats_done, 10000); 3347 if (rc) 3348 return; 3349 3350 for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); i++) 3351 data[i] = be64_to_cpu(IBMVNIC_GET_STAT 3352 (adapter, ibmvnic_stats[i].offset)); 3353 3354 for (j = 0; j < adapter->req_tx_queues; j++) { 3355 data[i] = adapter->tx_stats_buffers[j].packets; 3356 i++; 3357 data[i] = adapter->tx_stats_buffers[j].bytes; 3358 i++; 3359 data[i] = adapter->tx_stats_buffers[j].dropped_packets; 3360 i++; 3361 } 3362 3363 for (j = 0; j < adapter->req_rx_queues; j++) { 3364 data[i] = adapter->rx_stats_buffers[j].packets; 3365 i++; 3366 data[i] = adapter->rx_stats_buffers[j].bytes; 3367 i++; 3368 data[i] = adapter->rx_stats_buffers[j].interrupts; 3369 i++; 3370 } 3371 } 3372 3373 static u32 ibmvnic_get_priv_flags(struct net_device *netdev) 3374 { 3375 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3376 3377 return adapter->priv_flags; 3378 } 3379 3380 static int ibmvnic_set_priv_flags(struct net_device *netdev, u32 flags) 3381 { 3382 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3383 bool which_maxes = !!(flags & IBMVNIC_USE_SERVER_MAXES); 3384 3385 if (which_maxes) 3386 adapter->priv_flags |= IBMVNIC_USE_SERVER_MAXES; 3387 else 3388 adapter->priv_flags &= ~IBMVNIC_USE_SERVER_MAXES; 3389 3390 return 0; 3391 } 3392 3393 static const struct ethtool_ops ibmvnic_ethtool_ops = { 3394 .get_drvinfo = ibmvnic_get_drvinfo, 3395 .get_msglevel = ibmvnic_get_msglevel, 3396 .set_msglevel = ibmvnic_set_msglevel, 3397 .get_link = ibmvnic_get_link, 3398 .get_ringparam = ibmvnic_get_ringparam, 3399 .set_ringparam = ibmvnic_set_ringparam, 3400 .get_channels = ibmvnic_get_channels, 3401 .set_channels = ibmvnic_set_channels, 3402 .get_strings = ibmvnic_get_strings, 3403 .get_sset_count = ibmvnic_get_sset_count, 3404 .get_ethtool_stats = ibmvnic_get_ethtool_stats, 3405 .get_link_ksettings = ibmvnic_get_link_ksettings, 3406 .get_priv_flags = ibmvnic_get_priv_flags, 3407 .set_priv_flags = ibmvnic_set_priv_flags, 3408 }; 3409 3410 /* Routines for managing CRQs/sCRQs */ 3411 3412 static int reset_one_sub_crq_queue(struct ibmvnic_adapter *adapter, 3413 struct ibmvnic_sub_crq_queue *scrq) 3414 { 3415 int rc; 3416 3417 if (!scrq) { 3418 netdev_dbg(adapter->netdev, "Invalid scrq reset.\n"); 3419 return -EINVAL; 3420 } 3421 3422 if (scrq->irq) { 3423 free_irq(scrq->irq, scrq); 3424 irq_dispose_mapping(scrq->irq); 3425 scrq->irq = 0; 3426 } 3427 3428 if (scrq->msgs) { 3429 memset(scrq->msgs, 0, 4 * PAGE_SIZE); 3430 atomic_set(&scrq->used, 0); 3431 scrq->cur = 0; 3432 scrq->ind_buf.index = 0; 3433 } else { 3434 netdev_dbg(adapter->netdev, "Invalid scrq reset\n"); 3435 return -EINVAL; 3436 } 3437 3438 rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token, 3439 4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq); 3440 return rc; 3441 } 3442 3443 static int reset_sub_crq_queues(struct ibmvnic_adapter *adapter) 3444 { 3445 int i, rc; 3446 3447 if (!adapter->tx_scrq || !adapter->rx_scrq) 3448 return -EINVAL; 3449 3450 for (i = 0; i < adapter->req_tx_queues; i++) { 3451 netdev_dbg(adapter->netdev, "Re-setting tx_scrq[%d]\n", i); 3452 rc = reset_one_sub_crq_queue(adapter, adapter->tx_scrq[i]); 3453 if (rc) 3454 return rc; 3455 } 3456 3457 for (i = 0; i < adapter->req_rx_queues; i++) { 3458 netdev_dbg(adapter->netdev, "Re-setting rx_scrq[%d]\n", i); 3459 rc = reset_one_sub_crq_queue(adapter, adapter->rx_scrq[i]); 3460 if (rc) 3461 return rc; 3462 } 3463 3464 return rc; 3465 } 3466 3467 static void release_sub_crq_queue(struct ibmvnic_adapter *adapter, 3468 struct ibmvnic_sub_crq_queue *scrq, 3469 bool do_h_free) 3470 { 3471 struct device *dev = &adapter->vdev->dev; 3472 long rc; 3473 3474 netdev_dbg(adapter->netdev, "Releasing sub-CRQ\n"); 3475 3476 if (do_h_free) { 3477 /* Close the sub-crqs */ 3478 do { 3479 rc = plpar_hcall_norets(H_FREE_SUB_CRQ, 3480 adapter->vdev->unit_address, 3481 scrq->crq_num); 3482 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc)); 3483 3484 if (rc) { 3485 netdev_err(adapter->netdev, 3486 "Failed to release sub-CRQ %16lx, rc = %ld\n", 3487 scrq->crq_num, rc); 3488 } 3489 } 3490 3491 dma_free_coherent(dev, 3492 IBMVNIC_IND_ARR_SZ, 3493 scrq->ind_buf.indir_arr, 3494 scrq->ind_buf.indir_dma); 3495 3496 dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE, 3497 DMA_BIDIRECTIONAL); 3498 free_pages((unsigned long)scrq->msgs, 2); 3499 kfree(scrq); 3500 } 3501 3502 static struct ibmvnic_sub_crq_queue *init_sub_crq_queue(struct ibmvnic_adapter 3503 *adapter) 3504 { 3505 struct device *dev = &adapter->vdev->dev; 3506 struct ibmvnic_sub_crq_queue *scrq; 3507 int rc; 3508 3509 scrq = kzalloc(sizeof(*scrq), GFP_KERNEL); 3510 if (!scrq) 3511 return NULL; 3512 3513 scrq->msgs = 3514 (union sub_crq *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 2); 3515 if (!scrq->msgs) { 3516 dev_warn(dev, "Couldn't allocate crq queue messages page\n"); 3517 goto zero_page_failed; 3518 } 3519 3520 scrq->msg_token = dma_map_single(dev, scrq->msgs, 4 * PAGE_SIZE, 3521 DMA_BIDIRECTIONAL); 3522 if (dma_mapping_error(dev, scrq->msg_token)) { 3523 dev_warn(dev, "Couldn't map crq queue messages page\n"); 3524 goto map_failed; 3525 } 3526 3527 rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token, 3528 4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq); 3529 3530 if (rc == H_RESOURCE) 3531 rc = ibmvnic_reset_crq(adapter); 3532 3533 if (rc == H_CLOSED) { 3534 dev_warn(dev, "Partner adapter not ready, waiting.\n"); 3535 } else if (rc) { 3536 dev_warn(dev, "Error %d registering sub-crq\n", rc); 3537 goto reg_failed; 3538 } 3539 3540 scrq->adapter = adapter; 3541 scrq->size = 4 * PAGE_SIZE / sizeof(*scrq->msgs); 3542 scrq->ind_buf.index = 0; 3543 3544 scrq->ind_buf.indir_arr = 3545 dma_alloc_coherent(dev, 3546 IBMVNIC_IND_ARR_SZ, 3547 &scrq->ind_buf.indir_dma, 3548 GFP_KERNEL); 3549 3550 if (!scrq->ind_buf.indir_arr) 3551 goto indir_failed; 3552 3553 spin_lock_init(&scrq->lock); 3554 3555 netdev_dbg(adapter->netdev, 3556 "sub-crq initialized, num %lx, hw_irq=%lx, irq=%x\n", 3557 scrq->crq_num, scrq->hw_irq, scrq->irq); 3558 3559 return scrq; 3560 3561 indir_failed: 3562 do { 3563 rc = plpar_hcall_norets(H_FREE_SUB_CRQ, 3564 adapter->vdev->unit_address, 3565 scrq->crq_num); 3566 } while (rc == H_BUSY || rc == H_IS_LONG_BUSY(rc)); 3567 reg_failed: 3568 dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE, 3569 DMA_BIDIRECTIONAL); 3570 map_failed: 3571 free_pages((unsigned long)scrq->msgs, 2); 3572 zero_page_failed: 3573 kfree(scrq); 3574 3575 return NULL; 3576 } 3577 3578 static void release_sub_crqs(struct ibmvnic_adapter *adapter, bool do_h_free) 3579 { 3580 int i; 3581 3582 if (adapter->tx_scrq) { 3583 for (i = 0; i < adapter->num_active_tx_scrqs; i++) { 3584 if (!adapter->tx_scrq[i]) 3585 continue; 3586 3587 netdev_dbg(adapter->netdev, "Releasing tx_scrq[%d]\n", 3588 i); 3589 ibmvnic_tx_scrq_clean_buffer(adapter, adapter->tx_scrq[i]); 3590 if (adapter->tx_scrq[i]->irq) { 3591 free_irq(adapter->tx_scrq[i]->irq, 3592 adapter->tx_scrq[i]); 3593 irq_dispose_mapping(adapter->tx_scrq[i]->irq); 3594 adapter->tx_scrq[i]->irq = 0; 3595 } 3596 3597 release_sub_crq_queue(adapter, adapter->tx_scrq[i], 3598 do_h_free); 3599 } 3600 3601 kfree(adapter->tx_scrq); 3602 adapter->tx_scrq = NULL; 3603 adapter->num_active_tx_scrqs = 0; 3604 } 3605 3606 if (adapter->rx_scrq) { 3607 for (i = 0; i < adapter->num_active_rx_scrqs; i++) { 3608 if (!adapter->rx_scrq[i]) 3609 continue; 3610 3611 netdev_dbg(adapter->netdev, "Releasing rx_scrq[%d]\n", 3612 i); 3613 if (adapter->rx_scrq[i]->irq) { 3614 free_irq(adapter->rx_scrq[i]->irq, 3615 adapter->rx_scrq[i]); 3616 irq_dispose_mapping(adapter->rx_scrq[i]->irq); 3617 adapter->rx_scrq[i]->irq = 0; 3618 } 3619 3620 release_sub_crq_queue(adapter, adapter->rx_scrq[i], 3621 do_h_free); 3622 } 3623 3624 kfree(adapter->rx_scrq); 3625 adapter->rx_scrq = NULL; 3626 adapter->num_active_rx_scrqs = 0; 3627 } 3628 } 3629 3630 static int disable_scrq_irq(struct ibmvnic_adapter *adapter, 3631 struct ibmvnic_sub_crq_queue *scrq) 3632 { 3633 struct device *dev = &adapter->vdev->dev; 3634 unsigned long rc; 3635 3636 rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address, 3637 H_DISABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0); 3638 if (rc) 3639 dev_err(dev, "Couldn't disable scrq irq 0x%lx. rc=%ld\n", 3640 scrq->hw_irq, rc); 3641 return rc; 3642 } 3643 3644 /* We can not use the IRQ chip EOI handler because that has the 3645 * unintended effect of changing the interrupt priority. 3646 */ 3647 static void ibmvnic_xics_eoi(struct device *dev, struct ibmvnic_sub_crq_queue *scrq) 3648 { 3649 u64 val = 0xff000000 | scrq->hw_irq; 3650 unsigned long rc; 3651 3652 rc = plpar_hcall_norets(H_EOI, val); 3653 if (rc) 3654 dev_err(dev, "H_EOI FAILED irq 0x%llx. rc=%ld\n", val, rc); 3655 } 3656 3657 /* Due to a firmware bug, the hypervisor can send an interrupt to a 3658 * transmit or receive queue just prior to a partition migration. 3659 * Force an EOI after migration. 3660 */ 3661 static void ibmvnic_clear_pending_interrupt(struct device *dev, 3662 struct ibmvnic_sub_crq_queue *scrq) 3663 { 3664 if (!xive_enabled()) 3665 ibmvnic_xics_eoi(dev, scrq); 3666 } 3667 3668 static int enable_scrq_irq(struct ibmvnic_adapter *adapter, 3669 struct ibmvnic_sub_crq_queue *scrq) 3670 { 3671 struct device *dev = &adapter->vdev->dev; 3672 unsigned long rc; 3673 3674 if (scrq->hw_irq > 0x100000000ULL) { 3675 dev_err(dev, "bad hw_irq = %lx\n", scrq->hw_irq); 3676 return 1; 3677 } 3678 3679 if (test_bit(0, &adapter->resetting) && 3680 adapter->reset_reason == VNIC_RESET_MOBILITY) { 3681 ibmvnic_clear_pending_interrupt(dev, scrq); 3682 } 3683 3684 rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address, 3685 H_ENABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0); 3686 if (rc) 3687 dev_err(dev, "Couldn't enable scrq irq 0x%lx. rc=%ld\n", 3688 scrq->hw_irq, rc); 3689 return rc; 3690 } 3691 3692 static int ibmvnic_complete_tx(struct ibmvnic_adapter *adapter, 3693 struct ibmvnic_sub_crq_queue *scrq) 3694 { 3695 struct device *dev = &adapter->vdev->dev; 3696 struct ibmvnic_tx_pool *tx_pool; 3697 struct ibmvnic_tx_buff *txbuff; 3698 struct netdev_queue *txq; 3699 union sub_crq *next; 3700 int index; 3701 int i; 3702 3703 restart_loop: 3704 while (pending_scrq(adapter, scrq)) { 3705 unsigned int pool = scrq->pool_index; 3706 int num_entries = 0; 3707 int total_bytes = 0; 3708 int num_packets = 0; 3709 3710 next = ibmvnic_next_scrq(adapter, scrq); 3711 for (i = 0; i < next->tx_comp.num_comps; i++) { 3712 index = be32_to_cpu(next->tx_comp.correlators[i]); 3713 if (index & IBMVNIC_TSO_POOL_MASK) { 3714 tx_pool = &adapter->tso_pool[pool]; 3715 index &= ~IBMVNIC_TSO_POOL_MASK; 3716 } else { 3717 tx_pool = &adapter->tx_pool[pool]; 3718 } 3719 3720 txbuff = &tx_pool->tx_buff[index]; 3721 num_packets++; 3722 num_entries += txbuff->num_entries; 3723 if (txbuff->skb) { 3724 total_bytes += txbuff->skb->len; 3725 if (next->tx_comp.rcs[i]) { 3726 dev_err(dev, "tx error %x\n", 3727 next->tx_comp.rcs[i]); 3728 dev_kfree_skb_irq(txbuff->skb); 3729 } else { 3730 dev_consume_skb_irq(txbuff->skb); 3731 } 3732 txbuff->skb = NULL; 3733 } else { 3734 netdev_warn(adapter->netdev, 3735 "TX completion received with NULL socket buffer\n"); 3736 } 3737 tx_pool->free_map[tx_pool->producer_index] = index; 3738 tx_pool->producer_index = 3739 (tx_pool->producer_index + 1) % 3740 tx_pool->num_buffers; 3741 } 3742 /* remove tx_comp scrq*/ 3743 next->tx_comp.first = 0; 3744 3745 txq = netdev_get_tx_queue(adapter->netdev, scrq->pool_index); 3746 netdev_tx_completed_queue(txq, num_packets, total_bytes); 3747 3748 if (atomic_sub_return(num_entries, &scrq->used) <= 3749 (adapter->req_tx_entries_per_subcrq / 2) && 3750 __netif_subqueue_stopped(adapter->netdev, 3751 scrq->pool_index)) { 3752 netif_wake_subqueue(adapter->netdev, scrq->pool_index); 3753 netdev_dbg(adapter->netdev, "Started queue %d\n", 3754 scrq->pool_index); 3755 } 3756 } 3757 3758 enable_scrq_irq(adapter, scrq); 3759 3760 if (pending_scrq(adapter, scrq)) { 3761 disable_scrq_irq(adapter, scrq); 3762 goto restart_loop; 3763 } 3764 3765 return 0; 3766 } 3767 3768 static irqreturn_t ibmvnic_interrupt_tx(int irq, void *instance) 3769 { 3770 struct ibmvnic_sub_crq_queue *scrq = instance; 3771 struct ibmvnic_adapter *adapter = scrq->adapter; 3772 3773 disable_scrq_irq(adapter, scrq); 3774 ibmvnic_complete_tx(adapter, scrq); 3775 3776 return IRQ_HANDLED; 3777 } 3778 3779 static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance) 3780 { 3781 struct ibmvnic_sub_crq_queue *scrq = instance; 3782 struct ibmvnic_adapter *adapter = scrq->adapter; 3783 3784 /* When booting a kdump kernel we can hit pending interrupts 3785 * prior to completing driver initialization. 3786 */ 3787 if (unlikely(adapter->state != VNIC_OPEN)) 3788 return IRQ_NONE; 3789 3790 adapter->rx_stats_buffers[scrq->scrq_num].interrupts++; 3791 3792 if (napi_schedule_prep(&adapter->napi[scrq->scrq_num])) { 3793 disable_scrq_irq(adapter, scrq); 3794 __napi_schedule(&adapter->napi[scrq->scrq_num]); 3795 } 3796 3797 return IRQ_HANDLED; 3798 } 3799 3800 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter) 3801 { 3802 struct device *dev = &adapter->vdev->dev; 3803 struct ibmvnic_sub_crq_queue *scrq; 3804 int i = 0, j = 0; 3805 int rc = 0; 3806 3807 for (i = 0; i < adapter->req_tx_queues; i++) { 3808 netdev_dbg(adapter->netdev, "Initializing tx_scrq[%d] irq\n", 3809 i); 3810 scrq = adapter->tx_scrq[i]; 3811 scrq->irq = irq_create_mapping(NULL, scrq->hw_irq); 3812 3813 if (!scrq->irq) { 3814 rc = -EINVAL; 3815 dev_err(dev, "Error mapping irq\n"); 3816 goto req_tx_irq_failed; 3817 } 3818 3819 snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-tx%d", 3820 adapter->vdev->unit_address, i); 3821 rc = request_irq(scrq->irq, ibmvnic_interrupt_tx, 3822 0, scrq->name, scrq); 3823 3824 if (rc) { 3825 dev_err(dev, "Couldn't register tx irq 0x%x. rc=%d\n", 3826 scrq->irq, rc); 3827 irq_dispose_mapping(scrq->irq); 3828 goto req_tx_irq_failed; 3829 } 3830 } 3831 3832 for (i = 0; i < adapter->req_rx_queues; i++) { 3833 netdev_dbg(adapter->netdev, "Initializing rx_scrq[%d] irq\n", 3834 i); 3835 scrq = adapter->rx_scrq[i]; 3836 scrq->irq = irq_create_mapping(NULL, scrq->hw_irq); 3837 if (!scrq->irq) { 3838 rc = -EINVAL; 3839 dev_err(dev, "Error mapping irq\n"); 3840 goto req_rx_irq_failed; 3841 } 3842 snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-rx%d", 3843 adapter->vdev->unit_address, i); 3844 rc = request_irq(scrq->irq, ibmvnic_interrupt_rx, 3845 0, scrq->name, scrq); 3846 if (rc) { 3847 dev_err(dev, "Couldn't register rx irq 0x%x. rc=%d\n", 3848 scrq->irq, rc); 3849 irq_dispose_mapping(scrq->irq); 3850 goto req_rx_irq_failed; 3851 } 3852 } 3853 return rc; 3854 3855 req_rx_irq_failed: 3856 for (j = 0; j < i; j++) { 3857 free_irq(adapter->rx_scrq[j]->irq, adapter->rx_scrq[j]); 3858 irq_dispose_mapping(adapter->rx_scrq[j]->irq); 3859 } 3860 i = adapter->req_tx_queues; 3861 req_tx_irq_failed: 3862 for (j = 0; j < i; j++) { 3863 free_irq(adapter->tx_scrq[j]->irq, adapter->tx_scrq[j]); 3864 irq_dispose_mapping(adapter->tx_scrq[j]->irq); 3865 } 3866 release_sub_crqs(adapter, 1); 3867 return rc; 3868 } 3869 3870 static int init_sub_crqs(struct ibmvnic_adapter *adapter) 3871 { 3872 struct device *dev = &adapter->vdev->dev; 3873 struct ibmvnic_sub_crq_queue **allqueues; 3874 int registered_queues = 0; 3875 int total_queues; 3876 int more = 0; 3877 int i; 3878 3879 total_queues = adapter->req_tx_queues + adapter->req_rx_queues; 3880 3881 allqueues = kcalloc(total_queues, sizeof(*allqueues), GFP_KERNEL); 3882 if (!allqueues) 3883 return -ENOMEM; 3884 3885 for (i = 0; i < total_queues; i++) { 3886 allqueues[i] = init_sub_crq_queue(adapter); 3887 if (!allqueues[i]) { 3888 dev_warn(dev, "Couldn't allocate all sub-crqs\n"); 3889 break; 3890 } 3891 registered_queues++; 3892 } 3893 3894 /* Make sure we were able to register the minimum number of queues */ 3895 if (registered_queues < 3896 adapter->min_tx_queues + adapter->min_rx_queues) { 3897 dev_err(dev, "Fatal: Couldn't init min number of sub-crqs\n"); 3898 goto tx_failed; 3899 } 3900 3901 /* Distribute the failed allocated queues*/ 3902 for (i = 0; i < total_queues - registered_queues + more ; i++) { 3903 netdev_dbg(adapter->netdev, "Reducing number of queues\n"); 3904 switch (i % 3) { 3905 case 0: 3906 if (adapter->req_rx_queues > adapter->min_rx_queues) 3907 adapter->req_rx_queues--; 3908 else 3909 more++; 3910 break; 3911 case 1: 3912 if (adapter->req_tx_queues > adapter->min_tx_queues) 3913 adapter->req_tx_queues--; 3914 else 3915 more++; 3916 break; 3917 } 3918 } 3919 3920 adapter->tx_scrq = kcalloc(adapter->req_tx_queues, 3921 sizeof(*adapter->tx_scrq), GFP_KERNEL); 3922 if (!adapter->tx_scrq) 3923 goto tx_failed; 3924 3925 for (i = 0; i < adapter->req_tx_queues; i++) { 3926 adapter->tx_scrq[i] = allqueues[i]; 3927 adapter->tx_scrq[i]->pool_index = i; 3928 adapter->num_active_tx_scrqs++; 3929 } 3930 3931 adapter->rx_scrq = kcalloc(adapter->req_rx_queues, 3932 sizeof(*adapter->rx_scrq), GFP_KERNEL); 3933 if (!adapter->rx_scrq) 3934 goto rx_failed; 3935 3936 for (i = 0; i < adapter->req_rx_queues; i++) { 3937 adapter->rx_scrq[i] = allqueues[i + adapter->req_tx_queues]; 3938 adapter->rx_scrq[i]->scrq_num = i; 3939 adapter->num_active_rx_scrqs++; 3940 } 3941 3942 kfree(allqueues); 3943 return 0; 3944 3945 rx_failed: 3946 kfree(adapter->tx_scrq); 3947 adapter->tx_scrq = NULL; 3948 tx_failed: 3949 for (i = 0; i < registered_queues; i++) 3950 release_sub_crq_queue(adapter, allqueues[i], 1); 3951 kfree(allqueues); 3952 return -ENOMEM; 3953 } 3954 3955 static void send_request_cap(struct ibmvnic_adapter *adapter, int retry) 3956 { 3957 struct device *dev = &adapter->vdev->dev; 3958 union ibmvnic_crq crq; 3959 int max_entries; 3960 int cap_reqs; 3961 3962 /* We send out 6 or 7 REQUEST_CAPABILITY CRQs below (depending on 3963 * the PROMISC flag). Initialize this count upfront. When the tasklet 3964 * receives a response to all of these, it will send the next protocol 3965 * message (QUERY_IP_OFFLOAD). 3966 */ 3967 if (!(adapter->netdev->flags & IFF_PROMISC) || 3968 adapter->promisc_supported) 3969 cap_reqs = 7; 3970 else 3971 cap_reqs = 6; 3972 3973 if (!retry) { 3974 /* Sub-CRQ entries are 32 byte long */ 3975 int entries_page = 4 * PAGE_SIZE / (sizeof(u64) * 4); 3976 3977 atomic_set(&adapter->running_cap_crqs, cap_reqs); 3978 3979 if (adapter->min_tx_entries_per_subcrq > entries_page || 3980 adapter->min_rx_add_entries_per_subcrq > entries_page) { 3981 dev_err(dev, "Fatal, invalid entries per sub-crq\n"); 3982 return; 3983 } 3984 3985 if (adapter->desired.mtu) 3986 adapter->req_mtu = adapter->desired.mtu; 3987 else 3988 adapter->req_mtu = adapter->netdev->mtu + ETH_HLEN; 3989 3990 if (!adapter->desired.tx_entries) 3991 adapter->desired.tx_entries = 3992 adapter->max_tx_entries_per_subcrq; 3993 if (!adapter->desired.rx_entries) 3994 adapter->desired.rx_entries = 3995 adapter->max_rx_add_entries_per_subcrq; 3996 3997 max_entries = IBMVNIC_MAX_LTB_SIZE / 3998 (adapter->req_mtu + IBMVNIC_BUFFER_HLEN); 3999 4000 if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) * 4001 adapter->desired.tx_entries > IBMVNIC_MAX_LTB_SIZE) { 4002 adapter->desired.tx_entries = max_entries; 4003 } 4004 4005 if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) * 4006 adapter->desired.rx_entries > IBMVNIC_MAX_LTB_SIZE) { 4007 adapter->desired.rx_entries = max_entries; 4008 } 4009 4010 if (adapter->desired.tx_entries) 4011 adapter->req_tx_entries_per_subcrq = 4012 adapter->desired.tx_entries; 4013 else 4014 adapter->req_tx_entries_per_subcrq = 4015 adapter->max_tx_entries_per_subcrq; 4016 4017 if (adapter->desired.rx_entries) 4018 adapter->req_rx_add_entries_per_subcrq = 4019 adapter->desired.rx_entries; 4020 else 4021 adapter->req_rx_add_entries_per_subcrq = 4022 adapter->max_rx_add_entries_per_subcrq; 4023 4024 if (adapter->desired.tx_queues) 4025 adapter->req_tx_queues = 4026 adapter->desired.tx_queues; 4027 else 4028 adapter->req_tx_queues = 4029 adapter->opt_tx_comp_sub_queues; 4030 4031 if (adapter->desired.rx_queues) 4032 adapter->req_rx_queues = 4033 adapter->desired.rx_queues; 4034 else 4035 adapter->req_rx_queues = 4036 adapter->opt_rx_comp_queues; 4037 4038 adapter->req_rx_add_queues = adapter->max_rx_add_queues; 4039 } else { 4040 atomic_add(cap_reqs, &adapter->running_cap_crqs); 4041 } 4042 memset(&crq, 0, sizeof(crq)); 4043 crq.request_capability.first = IBMVNIC_CRQ_CMD; 4044 crq.request_capability.cmd = REQUEST_CAPABILITY; 4045 4046 crq.request_capability.capability = cpu_to_be16(REQ_TX_QUEUES); 4047 crq.request_capability.number = cpu_to_be64(adapter->req_tx_queues); 4048 cap_reqs--; 4049 ibmvnic_send_crq(adapter, &crq); 4050 4051 crq.request_capability.capability = cpu_to_be16(REQ_RX_QUEUES); 4052 crq.request_capability.number = cpu_to_be64(adapter->req_rx_queues); 4053 cap_reqs--; 4054 ibmvnic_send_crq(adapter, &crq); 4055 4056 crq.request_capability.capability = cpu_to_be16(REQ_RX_ADD_QUEUES); 4057 crq.request_capability.number = cpu_to_be64(adapter->req_rx_add_queues); 4058 cap_reqs--; 4059 ibmvnic_send_crq(adapter, &crq); 4060 4061 crq.request_capability.capability = 4062 cpu_to_be16(REQ_TX_ENTRIES_PER_SUBCRQ); 4063 crq.request_capability.number = 4064 cpu_to_be64(adapter->req_tx_entries_per_subcrq); 4065 cap_reqs--; 4066 ibmvnic_send_crq(adapter, &crq); 4067 4068 crq.request_capability.capability = 4069 cpu_to_be16(REQ_RX_ADD_ENTRIES_PER_SUBCRQ); 4070 crq.request_capability.number = 4071 cpu_to_be64(adapter->req_rx_add_entries_per_subcrq); 4072 cap_reqs--; 4073 ibmvnic_send_crq(adapter, &crq); 4074 4075 crq.request_capability.capability = cpu_to_be16(REQ_MTU); 4076 crq.request_capability.number = cpu_to_be64(adapter->req_mtu); 4077 cap_reqs--; 4078 ibmvnic_send_crq(adapter, &crq); 4079 4080 if (adapter->netdev->flags & IFF_PROMISC) { 4081 if (adapter->promisc_supported) { 4082 crq.request_capability.capability = 4083 cpu_to_be16(PROMISC_REQUESTED); 4084 crq.request_capability.number = cpu_to_be64(1); 4085 cap_reqs--; 4086 ibmvnic_send_crq(adapter, &crq); 4087 } 4088 } else { 4089 crq.request_capability.capability = 4090 cpu_to_be16(PROMISC_REQUESTED); 4091 crq.request_capability.number = cpu_to_be64(0); 4092 cap_reqs--; 4093 ibmvnic_send_crq(adapter, &crq); 4094 } 4095 4096 /* Keep at end to catch any discrepancy between expected and actual 4097 * CRQs sent. 4098 */ 4099 WARN_ON(cap_reqs != 0); 4100 } 4101 4102 static int pending_scrq(struct ibmvnic_adapter *adapter, 4103 struct ibmvnic_sub_crq_queue *scrq) 4104 { 4105 union sub_crq *entry = &scrq->msgs[scrq->cur]; 4106 int rc; 4107 4108 rc = !!(entry->generic.first & IBMVNIC_CRQ_CMD_RSP); 4109 4110 /* Ensure that the SCRQ valid flag is loaded prior to loading the 4111 * contents of the SCRQ descriptor 4112 */ 4113 dma_rmb(); 4114 4115 return rc; 4116 } 4117 4118 static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *adapter, 4119 struct ibmvnic_sub_crq_queue *scrq) 4120 { 4121 union sub_crq *entry; 4122 unsigned long flags; 4123 4124 spin_lock_irqsave(&scrq->lock, flags); 4125 entry = &scrq->msgs[scrq->cur]; 4126 if (entry->generic.first & IBMVNIC_CRQ_CMD_RSP) { 4127 if (++scrq->cur == scrq->size) 4128 scrq->cur = 0; 4129 } else { 4130 entry = NULL; 4131 } 4132 spin_unlock_irqrestore(&scrq->lock, flags); 4133 4134 /* Ensure that the SCRQ valid flag is loaded prior to loading the 4135 * contents of the SCRQ descriptor 4136 */ 4137 dma_rmb(); 4138 4139 return entry; 4140 } 4141 4142 static union ibmvnic_crq *ibmvnic_next_crq(struct ibmvnic_adapter *adapter) 4143 { 4144 struct ibmvnic_crq_queue *queue = &adapter->crq; 4145 union ibmvnic_crq *crq; 4146 4147 crq = &queue->msgs[queue->cur]; 4148 if (crq->generic.first & IBMVNIC_CRQ_CMD_RSP) { 4149 if (++queue->cur == queue->size) 4150 queue->cur = 0; 4151 } else { 4152 crq = NULL; 4153 } 4154 4155 return crq; 4156 } 4157 4158 static void print_subcrq_error(struct device *dev, int rc, const char *func) 4159 { 4160 switch (rc) { 4161 case H_PARAMETER: 4162 dev_warn_ratelimited(dev, 4163 "%s failed: Send request is malformed or adapter failover pending. (rc=%d)\n", 4164 func, rc); 4165 break; 4166 case H_CLOSED: 4167 dev_warn_ratelimited(dev, 4168 "%s failed: Backing queue closed. Adapter is down or failover pending. (rc=%d)\n", 4169 func, rc); 4170 break; 4171 default: 4172 dev_err_ratelimited(dev, "%s failed: (rc=%d)\n", func, rc); 4173 break; 4174 } 4175 } 4176 4177 static int send_subcrq_indirect(struct ibmvnic_adapter *adapter, 4178 u64 remote_handle, u64 ioba, u64 num_entries) 4179 { 4180 unsigned int ua = adapter->vdev->unit_address; 4181 struct device *dev = &adapter->vdev->dev; 4182 int rc; 4183 4184 /* Make sure the hypervisor sees the complete request */ 4185 dma_wmb(); 4186 rc = plpar_hcall_norets(H_SEND_SUB_CRQ_INDIRECT, ua, 4187 cpu_to_be64(remote_handle), 4188 ioba, num_entries); 4189 4190 if (rc) 4191 print_subcrq_error(dev, rc, __func__); 4192 4193 return rc; 4194 } 4195 4196 static int ibmvnic_send_crq(struct ibmvnic_adapter *adapter, 4197 union ibmvnic_crq *crq) 4198 { 4199 unsigned int ua = adapter->vdev->unit_address; 4200 struct device *dev = &adapter->vdev->dev; 4201 u64 *u64_crq = (u64 *)crq; 4202 int rc; 4203 4204 netdev_dbg(adapter->netdev, "Sending CRQ: %016lx %016lx\n", 4205 (unsigned long)cpu_to_be64(u64_crq[0]), 4206 (unsigned long)cpu_to_be64(u64_crq[1])); 4207 4208 if (!adapter->crq.active && 4209 crq->generic.first != IBMVNIC_CRQ_INIT_CMD) { 4210 dev_warn(dev, "Invalid request detected while CRQ is inactive, possible device state change during reset\n"); 4211 return -EINVAL; 4212 } 4213 4214 /* Make sure the hypervisor sees the complete request */ 4215 dma_wmb(); 4216 4217 rc = plpar_hcall_norets(H_SEND_CRQ, ua, 4218 cpu_to_be64(u64_crq[0]), 4219 cpu_to_be64(u64_crq[1])); 4220 4221 if (rc) { 4222 if (rc == H_CLOSED) { 4223 dev_warn(dev, "CRQ Queue closed\n"); 4224 /* do not reset, report the fail, wait for passive init from server */ 4225 } 4226 4227 dev_warn(dev, "Send error (rc=%d)\n", rc); 4228 } 4229 4230 return rc; 4231 } 4232 4233 static int ibmvnic_send_crq_init(struct ibmvnic_adapter *adapter) 4234 { 4235 struct device *dev = &adapter->vdev->dev; 4236 union ibmvnic_crq crq; 4237 int retries = 100; 4238 int rc; 4239 4240 memset(&crq, 0, sizeof(crq)); 4241 crq.generic.first = IBMVNIC_CRQ_INIT_CMD; 4242 crq.generic.cmd = IBMVNIC_CRQ_INIT; 4243 netdev_dbg(adapter->netdev, "Sending CRQ init\n"); 4244 4245 do { 4246 rc = ibmvnic_send_crq(adapter, &crq); 4247 if (rc != H_CLOSED) 4248 break; 4249 retries--; 4250 msleep(50); 4251 4252 } while (retries > 0); 4253 4254 if (rc) { 4255 dev_err(dev, "Failed to send init request, rc = %d\n", rc); 4256 return rc; 4257 } 4258 4259 return 0; 4260 } 4261 4262 struct vnic_login_client_data { 4263 u8 type; 4264 __be16 len; 4265 char name[]; 4266 } __packed; 4267 4268 static int vnic_client_data_len(struct ibmvnic_adapter *adapter) 4269 { 4270 int len; 4271 4272 /* Calculate the amount of buffer space needed for the 4273 * vnic client data in the login buffer. There are four entries, 4274 * OS name, LPAR name, device name, and a null last entry. 4275 */ 4276 len = 4 * sizeof(struct vnic_login_client_data); 4277 len += 6; /* "Linux" plus NULL */ 4278 len += strlen(utsname()->nodename) + 1; 4279 len += strlen(adapter->netdev->name) + 1; 4280 4281 return len; 4282 } 4283 4284 static void vnic_add_client_data(struct ibmvnic_adapter *adapter, 4285 struct vnic_login_client_data *vlcd) 4286 { 4287 const char *os_name = "Linux"; 4288 int len; 4289 4290 /* Type 1 - LPAR OS */ 4291 vlcd->type = 1; 4292 len = strlen(os_name) + 1; 4293 vlcd->len = cpu_to_be16(len); 4294 strscpy(vlcd->name, os_name, len); 4295 vlcd = (struct vnic_login_client_data *)(vlcd->name + len); 4296 4297 /* Type 2 - LPAR name */ 4298 vlcd->type = 2; 4299 len = strlen(utsname()->nodename) + 1; 4300 vlcd->len = cpu_to_be16(len); 4301 strscpy(vlcd->name, utsname()->nodename, len); 4302 vlcd = (struct vnic_login_client_data *)(vlcd->name + len); 4303 4304 /* Type 3 - device name */ 4305 vlcd->type = 3; 4306 len = strlen(adapter->netdev->name) + 1; 4307 vlcd->len = cpu_to_be16(len); 4308 strscpy(vlcd->name, adapter->netdev->name, len); 4309 } 4310 4311 static int send_login(struct ibmvnic_adapter *adapter) 4312 { 4313 struct ibmvnic_login_rsp_buffer *login_rsp_buffer; 4314 struct ibmvnic_login_buffer *login_buffer; 4315 struct device *dev = &adapter->vdev->dev; 4316 struct vnic_login_client_data *vlcd; 4317 dma_addr_t rsp_buffer_token; 4318 dma_addr_t buffer_token; 4319 size_t rsp_buffer_size; 4320 union ibmvnic_crq crq; 4321 int client_data_len; 4322 size_t buffer_size; 4323 __be64 *tx_list_p; 4324 __be64 *rx_list_p; 4325 int rc; 4326 int i; 4327 4328 if (!adapter->tx_scrq || !adapter->rx_scrq) { 4329 netdev_err(adapter->netdev, 4330 "RX or TX queues are not allocated, device login failed\n"); 4331 return -ENOMEM; 4332 } 4333 4334 release_login_buffer(adapter); 4335 release_login_rsp_buffer(adapter); 4336 4337 client_data_len = vnic_client_data_len(adapter); 4338 4339 buffer_size = 4340 sizeof(struct ibmvnic_login_buffer) + 4341 sizeof(u64) * (adapter->req_tx_queues + adapter->req_rx_queues) + 4342 client_data_len; 4343 4344 login_buffer = kzalloc(buffer_size, GFP_ATOMIC); 4345 if (!login_buffer) 4346 goto buf_alloc_failed; 4347 4348 buffer_token = dma_map_single(dev, login_buffer, buffer_size, 4349 DMA_TO_DEVICE); 4350 if (dma_mapping_error(dev, buffer_token)) { 4351 dev_err(dev, "Couldn't map login buffer\n"); 4352 goto buf_map_failed; 4353 } 4354 4355 rsp_buffer_size = sizeof(struct ibmvnic_login_rsp_buffer) + 4356 sizeof(u64) * adapter->req_tx_queues + 4357 sizeof(u64) * adapter->req_rx_queues + 4358 sizeof(u64) * adapter->req_rx_queues + 4359 sizeof(u8) * IBMVNIC_TX_DESC_VERSIONS; 4360 4361 login_rsp_buffer = kmalloc(rsp_buffer_size, GFP_ATOMIC); 4362 if (!login_rsp_buffer) 4363 goto buf_rsp_alloc_failed; 4364 4365 rsp_buffer_token = dma_map_single(dev, login_rsp_buffer, 4366 rsp_buffer_size, DMA_FROM_DEVICE); 4367 if (dma_mapping_error(dev, rsp_buffer_token)) { 4368 dev_err(dev, "Couldn't map login rsp buffer\n"); 4369 goto buf_rsp_map_failed; 4370 } 4371 4372 adapter->login_buf = login_buffer; 4373 adapter->login_buf_token = buffer_token; 4374 adapter->login_buf_sz = buffer_size; 4375 adapter->login_rsp_buf = login_rsp_buffer; 4376 adapter->login_rsp_buf_token = rsp_buffer_token; 4377 adapter->login_rsp_buf_sz = rsp_buffer_size; 4378 4379 login_buffer->len = cpu_to_be32(buffer_size); 4380 login_buffer->version = cpu_to_be32(INITIAL_VERSION_LB); 4381 login_buffer->num_txcomp_subcrqs = cpu_to_be32(adapter->req_tx_queues); 4382 login_buffer->off_txcomp_subcrqs = 4383 cpu_to_be32(sizeof(struct ibmvnic_login_buffer)); 4384 login_buffer->num_rxcomp_subcrqs = cpu_to_be32(adapter->req_rx_queues); 4385 login_buffer->off_rxcomp_subcrqs = 4386 cpu_to_be32(sizeof(struct ibmvnic_login_buffer) + 4387 sizeof(u64) * adapter->req_tx_queues); 4388 login_buffer->login_rsp_ioba = cpu_to_be32(rsp_buffer_token); 4389 login_buffer->login_rsp_len = cpu_to_be32(rsp_buffer_size); 4390 4391 tx_list_p = (__be64 *)((char *)login_buffer + 4392 sizeof(struct ibmvnic_login_buffer)); 4393 rx_list_p = (__be64 *)((char *)login_buffer + 4394 sizeof(struct ibmvnic_login_buffer) + 4395 sizeof(u64) * adapter->req_tx_queues); 4396 4397 for (i = 0; i < adapter->req_tx_queues; i++) { 4398 if (adapter->tx_scrq[i]) { 4399 tx_list_p[i] = 4400 cpu_to_be64(adapter->tx_scrq[i]->crq_num); 4401 } 4402 } 4403 4404 for (i = 0; i < adapter->req_rx_queues; i++) { 4405 if (adapter->rx_scrq[i]) { 4406 rx_list_p[i] = 4407 cpu_to_be64(adapter->rx_scrq[i]->crq_num); 4408 } 4409 } 4410 4411 /* Insert vNIC login client data */ 4412 vlcd = (struct vnic_login_client_data *) 4413 ((char *)rx_list_p + (sizeof(u64) * adapter->req_rx_queues)); 4414 login_buffer->client_data_offset = 4415 cpu_to_be32((char *)vlcd - (char *)login_buffer); 4416 login_buffer->client_data_len = cpu_to_be32(client_data_len); 4417 4418 vnic_add_client_data(adapter, vlcd); 4419 4420 netdev_dbg(adapter->netdev, "Login Buffer:\n"); 4421 for (i = 0; i < (adapter->login_buf_sz - 1) / 8 + 1; i++) { 4422 netdev_dbg(adapter->netdev, "%016lx\n", 4423 ((unsigned long *)(adapter->login_buf))[i]); 4424 } 4425 4426 memset(&crq, 0, sizeof(crq)); 4427 crq.login.first = IBMVNIC_CRQ_CMD; 4428 crq.login.cmd = LOGIN; 4429 crq.login.ioba = cpu_to_be32(buffer_token); 4430 crq.login.len = cpu_to_be32(buffer_size); 4431 4432 adapter->login_pending = true; 4433 rc = ibmvnic_send_crq(adapter, &crq); 4434 if (rc) { 4435 adapter->login_pending = false; 4436 netdev_err(adapter->netdev, "Failed to send login, rc=%d\n", rc); 4437 goto buf_rsp_map_failed; 4438 } 4439 4440 return 0; 4441 4442 buf_rsp_map_failed: 4443 kfree(login_rsp_buffer); 4444 adapter->login_rsp_buf = NULL; 4445 buf_rsp_alloc_failed: 4446 dma_unmap_single(dev, buffer_token, buffer_size, DMA_TO_DEVICE); 4447 buf_map_failed: 4448 kfree(login_buffer); 4449 adapter->login_buf = NULL; 4450 buf_alloc_failed: 4451 return -ENOMEM; 4452 } 4453 4454 static int send_request_map(struct ibmvnic_adapter *adapter, dma_addr_t addr, 4455 u32 len, u8 map_id) 4456 { 4457 union ibmvnic_crq crq; 4458 4459 memset(&crq, 0, sizeof(crq)); 4460 crq.request_map.first = IBMVNIC_CRQ_CMD; 4461 crq.request_map.cmd = REQUEST_MAP; 4462 crq.request_map.map_id = map_id; 4463 crq.request_map.ioba = cpu_to_be32(addr); 4464 crq.request_map.len = cpu_to_be32(len); 4465 return ibmvnic_send_crq(adapter, &crq); 4466 } 4467 4468 static int send_request_unmap(struct ibmvnic_adapter *adapter, u8 map_id) 4469 { 4470 union ibmvnic_crq crq; 4471 4472 memset(&crq, 0, sizeof(crq)); 4473 crq.request_unmap.first = IBMVNIC_CRQ_CMD; 4474 crq.request_unmap.cmd = REQUEST_UNMAP; 4475 crq.request_unmap.map_id = map_id; 4476 return ibmvnic_send_crq(adapter, &crq); 4477 } 4478 4479 static void send_query_map(struct ibmvnic_adapter *adapter) 4480 { 4481 union ibmvnic_crq crq; 4482 4483 memset(&crq, 0, sizeof(crq)); 4484 crq.query_map.first = IBMVNIC_CRQ_CMD; 4485 crq.query_map.cmd = QUERY_MAP; 4486 ibmvnic_send_crq(adapter, &crq); 4487 } 4488 4489 /* Send a series of CRQs requesting various capabilities of the VNIC server */ 4490 static void send_query_cap(struct ibmvnic_adapter *adapter) 4491 { 4492 union ibmvnic_crq crq; 4493 int cap_reqs; 4494 4495 /* We send out 25 QUERY_CAPABILITY CRQs below. Initialize this count 4496 * upfront. When the tasklet receives a response to all of these, it 4497 * can send out the next protocol messaage (REQUEST_CAPABILITY). 4498 */ 4499 cap_reqs = 25; 4500 4501 atomic_set(&adapter->running_cap_crqs, cap_reqs); 4502 4503 memset(&crq, 0, sizeof(crq)); 4504 crq.query_capability.first = IBMVNIC_CRQ_CMD; 4505 crq.query_capability.cmd = QUERY_CAPABILITY; 4506 4507 crq.query_capability.capability = cpu_to_be16(MIN_TX_QUEUES); 4508 ibmvnic_send_crq(adapter, &crq); 4509 cap_reqs--; 4510 4511 crq.query_capability.capability = cpu_to_be16(MIN_RX_QUEUES); 4512 ibmvnic_send_crq(adapter, &crq); 4513 cap_reqs--; 4514 4515 crq.query_capability.capability = cpu_to_be16(MIN_RX_ADD_QUEUES); 4516 ibmvnic_send_crq(adapter, &crq); 4517 cap_reqs--; 4518 4519 crq.query_capability.capability = cpu_to_be16(MAX_TX_QUEUES); 4520 ibmvnic_send_crq(adapter, &crq); 4521 cap_reqs--; 4522 4523 crq.query_capability.capability = cpu_to_be16(MAX_RX_QUEUES); 4524 ibmvnic_send_crq(adapter, &crq); 4525 cap_reqs--; 4526 4527 crq.query_capability.capability = cpu_to_be16(MAX_RX_ADD_QUEUES); 4528 ibmvnic_send_crq(adapter, &crq); 4529 cap_reqs--; 4530 4531 crq.query_capability.capability = 4532 cpu_to_be16(MIN_TX_ENTRIES_PER_SUBCRQ); 4533 ibmvnic_send_crq(adapter, &crq); 4534 cap_reqs--; 4535 4536 crq.query_capability.capability = 4537 cpu_to_be16(MIN_RX_ADD_ENTRIES_PER_SUBCRQ); 4538 ibmvnic_send_crq(adapter, &crq); 4539 cap_reqs--; 4540 4541 crq.query_capability.capability = 4542 cpu_to_be16(MAX_TX_ENTRIES_PER_SUBCRQ); 4543 ibmvnic_send_crq(adapter, &crq); 4544 cap_reqs--; 4545 4546 crq.query_capability.capability = 4547 cpu_to_be16(MAX_RX_ADD_ENTRIES_PER_SUBCRQ); 4548 ibmvnic_send_crq(adapter, &crq); 4549 cap_reqs--; 4550 4551 crq.query_capability.capability = cpu_to_be16(TCP_IP_OFFLOAD); 4552 ibmvnic_send_crq(adapter, &crq); 4553 cap_reqs--; 4554 4555 crq.query_capability.capability = cpu_to_be16(PROMISC_SUPPORTED); 4556 ibmvnic_send_crq(adapter, &crq); 4557 cap_reqs--; 4558 4559 crq.query_capability.capability = cpu_to_be16(MIN_MTU); 4560 ibmvnic_send_crq(adapter, &crq); 4561 cap_reqs--; 4562 4563 crq.query_capability.capability = cpu_to_be16(MAX_MTU); 4564 ibmvnic_send_crq(adapter, &crq); 4565 cap_reqs--; 4566 4567 crq.query_capability.capability = cpu_to_be16(MAX_MULTICAST_FILTERS); 4568 ibmvnic_send_crq(adapter, &crq); 4569 cap_reqs--; 4570 4571 crq.query_capability.capability = cpu_to_be16(VLAN_HEADER_INSERTION); 4572 ibmvnic_send_crq(adapter, &crq); 4573 cap_reqs--; 4574 4575 crq.query_capability.capability = cpu_to_be16(RX_VLAN_HEADER_INSERTION); 4576 ibmvnic_send_crq(adapter, &crq); 4577 cap_reqs--; 4578 4579 crq.query_capability.capability = cpu_to_be16(MAX_TX_SG_ENTRIES); 4580 ibmvnic_send_crq(adapter, &crq); 4581 cap_reqs--; 4582 4583 crq.query_capability.capability = cpu_to_be16(RX_SG_SUPPORTED); 4584 ibmvnic_send_crq(adapter, &crq); 4585 cap_reqs--; 4586 4587 crq.query_capability.capability = cpu_to_be16(OPT_TX_COMP_SUB_QUEUES); 4588 ibmvnic_send_crq(adapter, &crq); 4589 cap_reqs--; 4590 4591 crq.query_capability.capability = cpu_to_be16(OPT_RX_COMP_QUEUES); 4592 ibmvnic_send_crq(adapter, &crq); 4593 cap_reqs--; 4594 4595 crq.query_capability.capability = 4596 cpu_to_be16(OPT_RX_BUFADD_Q_PER_RX_COMP_Q); 4597 ibmvnic_send_crq(adapter, &crq); 4598 cap_reqs--; 4599 4600 crq.query_capability.capability = 4601 cpu_to_be16(OPT_TX_ENTRIES_PER_SUBCRQ); 4602 ibmvnic_send_crq(adapter, &crq); 4603 cap_reqs--; 4604 4605 crq.query_capability.capability = 4606 cpu_to_be16(OPT_RXBA_ENTRIES_PER_SUBCRQ); 4607 ibmvnic_send_crq(adapter, &crq); 4608 cap_reqs--; 4609 4610 crq.query_capability.capability = cpu_to_be16(TX_RX_DESC_REQ); 4611 4612 ibmvnic_send_crq(adapter, &crq); 4613 cap_reqs--; 4614 4615 /* Keep at end to catch any discrepancy between expected and actual 4616 * CRQs sent. 4617 */ 4618 WARN_ON(cap_reqs != 0); 4619 } 4620 4621 static void send_query_ip_offload(struct ibmvnic_adapter *adapter) 4622 { 4623 int buf_sz = sizeof(struct ibmvnic_query_ip_offload_buffer); 4624 struct device *dev = &adapter->vdev->dev; 4625 union ibmvnic_crq crq; 4626 4627 adapter->ip_offload_tok = 4628 dma_map_single(dev, 4629 &adapter->ip_offload_buf, 4630 buf_sz, 4631 DMA_FROM_DEVICE); 4632 4633 if (dma_mapping_error(dev, adapter->ip_offload_tok)) { 4634 if (!firmware_has_feature(FW_FEATURE_CMO)) 4635 dev_err(dev, "Couldn't map offload buffer\n"); 4636 return; 4637 } 4638 4639 memset(&crq, 0, sizeof(crq)); 4640 crq.query_ip_offload.first = IBMVNIC_CRQ_CMD; 4641 crq.query_ip_offload.cmd = QUERY_IP_OFFLOAD; 4642 crq.query_ip_offload.len = cpu_to_be32(buf_sz); 4643 crq.query_ip_offload.ioba = 4644 cpu_to_be32(adapter->ip_offload_tok); 4645 4646 ibmvnic_send_crq(adapter, &crq); 4647 } 4648 4649 static void send_control_ip_offload(struct ibmvnic_adapter *adapter) 4650 { 4651 struct ibmvnic_control_ip_offload_buffer *ctrl_buf = &adapter->ip_offload_ctrl; 4652 struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf; 4653 struct device *dev = &adapter->vdev->dev; 4654 netdev_features_t old_hw_features = 0; 4655 union ibmvnic_crq crq; 4656 4657 adapter->ip_offload_ctrl_tok = 4658 dma_map_single(dev, 4659 ctrl_buf, 4660 sizeof(adapter->ip_offload_ctrl), 4661 DMA_TO_DEVICE); 4662 4663 if (dma_mapping_error(dev, adapter->ip_offload_ctrl_tok)) { 4664 dev_err(dev, "Couldn't map ip offload control buffer\n"); 4665 return; 4666 } 4667 4668 ctrl_buf->len = cpu_to_be32(sizeof(adapter->ip_offload_ctrl)); 4669 ctrl_buf->version = cpu_to_be32(INITIAL_VERSION_IOB); 4670 ctrl_buf->ipv4_chksum = buf->ipv4_chksum; 4671 ctrl_buf->ipv6_chksum = buf->ipv6_chksum; 4672 ctrl_buf->tcp_ipv4_chksum = buf->tcp_ipv4_chksum; 4673 ctrl_buf->udp_ipv4_chksum = buf->udp_ipv4_chksum; 4674 ctrl_buf->tcp_ipv6_chksum = buf->tcp_ipv6_chksum; 4675 ctrl_buf->udp_ipv6_chksum = buf->udp_ipv6_chksum; 4676 ctrl_buf->large_tx_ipv4 = buf->large_tx_ipv4; 4677 ctrl_buf->large_tx_ipv6 = buf->large_tx_ipv6; 4678 4679 /* large_rx disabled for now, additional features needed */ 4680 ctrl_buf->large_rx_ipv4 = 0; 4681 ctrl_buf->large_rx_ipv6 = 0; 4682 4683 if (adapter->state != VNIC_PROBING) { 4684 old_hw_features = adapter->netdev->hw_features; 4685 adapter->netdev->hw_features = 0; 4686 } 4687 4688 adapter->netdev->hw_features = NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO; 4689 4690 if (buf->tcp_ipv4_chksum || buf->udp_ipv4_chksum) 4691 adapter->netdev->hw_features |= NETIF_F_IP_CSUM; 4692 4693 if (buf->tcp_ipv6_chksum || buf->udp_ipv6_chksum) 4694 adapter->netdev->hw_features |= NETIF_F_IPV6_CSUM; 4695 4696 if ((adapter->netdev->features & 4697 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM))) 4698 adapter->netdev->hw_features |= NETIF_F_RXCSUM; 4699 4700 if (buf->large_tx_ipv4) 4701 adapter->netdev->hw_features |= NETIF_F_TSO; 4702 if (buf->large_tx_ipv6) 4703 adapter->netdev->hw_features |= NETIF_F_TSO6; 4704 4705 if (adapter->state == VNIC_PROBING) { 4706 adapter->netdev->features |= adapter->netdev->hw_features; 4707 } else if (old_hw_features != adapter->netdev->hw_features) { 4708 netdev_features_t tmp = 0; 4709 4710 /* disable features no longer supported */ 4711 adapter->netdev->features &= adapter->netdev->hw_features; 4712 /* turn on features now supported if previously enabled */ 4713 tmp = (old_hw_features ^ adapter->netdev->hw_features) & 4714 adapter->netdev->hw_features; 4715 adapter->netdev->features |= 4716 tmp & adapter->netdev->wanted_features; 4717 } 4718 4719 memset(&crq, 0, sizeof(crq)); 4720 crq.control_ip_offload.first = IBMVNIC_CRQ_CMD; 4721 crq.control_ip_offload.cmd = CONTROL_IP_OFFLOAD; 4722 crq.control_ip_offload.len = 4723 cpu_to_be32(sizeof(adapter->ip_offload_ctrl)); 4724 crq.control_ip_offload.ioba = cpu_to_be32(adapter->ip_offload_ctrl_tok); 4725 ibmvnic_send_crq(adapter, &crq); 4726 } 4727 4728 static void handle_vpd_size_rsp(union ibmvnic_crq *crq, 4729 struct ibmvnic_adapter *adapter) 4730 { 4731 struct device *dev = &adapter->vdev->dev; 4732 4733 if (crq->get_vpd_size_rsp.rc.code) { 4734 dev_err(dev, "Error retrieving VPD size, rc=%x\n", 4735 crq->get_vpd_size_rsp.rc.code); 4736 complete(&adapter->fw_done); 4737 return; 4738 } 4739 4740 adapter->vpd->len = be64_to_cpu(crq->get_vpd_size_rsp.len); 4741 complete(&adapter->fw_done); 4742 } 4743 4744 static void handle_vpd_rsp(union ibmvnic_crq *crq, 4745 struct ibmvnic_adapter *adapter) 4746 { 4747 struct device *dev = &adapter->vdev->dev; 4748 unsigned char *substr = NULL; 4749 u8 fw_level_len = 0; 4750 4751 memset(adapter->fw_version, 0, 32); 4752 4753 dma_unmap_single(dev, adapter->vpd->dma_addr, adapter->vpd->len, 4754 DMA_FROM_DEVICE); 4755 4756 if (crq->get_vpd_rsp.rc.code) { 4757 dev_err(dev, "Error retrieving VPD from device, rc=%x\n", 4758 crq->get_vpd_rsp.rc.code); 4759 goto complete; 4760 } 4761 4762 /* get the position of the firmware version info 4763 * located after the ASCII 'RM' substring in the buffer 4764 */ 4765 substr = strnstr(adapter->vpd->buff, "RM", adapter->vpd->len); 4766 if (!substr) { 4767 dev_info(dev, "Warning - No FW level has been provided in the VPD buffer by the VIOS Server\n"); 4768 goto complete; 4769 } 4770 4771 /* get length of firmware level ASCII substring */ 4772 if ((substr + 2) < (adapter->vpd->buff + adapter->vpd->len)) { 4773 fw_level_len = *(substr + 2); 4774 } else { 4775 dev_info(dev, "Length of FW substr extrapolated VDP buff\n"); 4776 goto complete; 4777 } 4778 4779 /* copy firmware version string from vpd into adapter */ 4780 if ((substr + 3 + fw_level_len) < 4781 (adapter->vpd->buff + adapter->vpd->len)) { 4782 strncpy((char *)adapter->fw_version, substr + 3, fw_level_len); 4783 } else { 4784 dev_info(dev, "FW substr extrapolated VPD buff\n"); 4785 } 4786 4787 complete: 4788 if (adapter->fw_version[0] == '\0') 4789 strscpy((char *)adapter->fw_version, "N/A", sizeof(adapter->fw_version)); 4790 complete(&adapter->fw_done); 4791 } 4792 4793 static void handle_query_ip_offload_rsp(struct ibmvnic_adapter *adapter) 4794 { 4795 struct device *dev = &adapter->vdev->dev; 4796 struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf; 4797 int i; 4798 4799 dma_unmap_single(dev, adapter->ip_offload_tok, 4800 sizeof(adapter->ip_offload_buf), DMA_FROM_DEVICE); 4801 4802 netdev_dbg(adapter->netdev, "Query IP Offload Buffer:\n"); 4803 for (i = 0; i < (sizeof(adapter->ip_offload_buf) - 1) / 8 + 1; i++) 4804 netdev_dbg(adapter->netdev, "%016lx\n", 4805 ((unsigned long *)(buf))[i]); 4806 4807 netdev_dbg(adapter->netdev, "ipv4_chksum = %d\n", buf->ipv4_chksum); 4808 netdev_dbg(adapter->netdev, "ipv6_chksum = %d\n", buf->ipv6_chksum); 4809 netdev_dbg(adapter->netdev, "tcp_ipv4_chksum = %d\n", 4810 buf->tcp_ipv4_chksum); 4811 netdev_dbg(adapter->netdev, "tcp_ipv6_chksum = %d\n", 4812 buf->tcp_ipv6_chksum); 4813 netdev_dbg(adapter->netdev, "udp_ipv4_chksum = %d\n", 4814 buf->udp_ipv4_chksum); 4815 netdev_dbg(adapter->netdev, "udp_ipv6_chksum = %d\n", 4816 buf->udp_ipv6_chksum); 4817 netdev_dbg(adapter->netdev, "large_tx_ipv4 = %d\n", 4818 buf->large_tx_ipv4); 4819 netdev_dbg(adapter->netdev, "large_tx_ipv6 = %d\n", 4820 buf->large_tx_ipv6); 4821 netdev_dbg(adapter->netdev, "large_rx_ipv4 = %d\n", 4822 buf->large_rx_ipv4); 4823 netdev_dbg(adapter->netdev, "large_rx_ipv6 = %d\n", 4824 buf->large_rx_ipv6); 4825 netdev_dbg(adapter->netdev, "max_ipv4_hdr_sz = %d\n", 4826 buf->max_ipv4_header_size); 4827 netdev_dbg(adapter->netdev, "max_ipv6_hdr_sz = %d\n", 4828 buf->max_ipv6_header_size); 4829 netdev_dbg(adapter->netdev, "max_tcp_hdr_size = %d\n", 4830 buf->max_tcp_header_size); 4831 netdev_dbg(adapter->netdev, "max_udp_hdr_size = %d\n", 4832 buf->max_udp_header_size); 4833 netdev_dbg(adapter->netdev, "max_large_tx_size = %d\n", 4834 buf->max_large_tx_size); 4835 netdev_dbg(adapter->netdev, "max_large_rx_size = %d\n", 4836 buf->max_large_rx_size); 4837 netdev_dbg(adapter->netdev, "ipv6_ext_hdr = %d\n", 4838 buf->ipv6_extension_header); 4839 netdev_dbg(adapter->netdev, "tcp_pseudosum_req = %d\n", 4840 buf->tcp_pseudosum_req); 4841 netdev_dbg(adapter->netdev, "num_ipv6_ext_hd = %d\n", 4842 buf->num_ipv6_ext_headers); 4843 netdev_dbg(adapter->netdev, "off_ipv6_ext_hd = %d\n", 4844 buf->off_ipv6_ext_headers); 4845 4846 send_control_ip_offload(adapter); 4847 } 4848 4849 static const char *ibmvnic_fw_err_cause(u16 cause) 4850 { 4851 switch (cause) { 4852 case ADAPTER_PROBLEM: 4853 return "adapter problem"; 4854 case BUS_PROBLEM: 4855 return "bus problem"; 4856 case FW_PROBLEM: 4857 return "firmware problem"; 4858 case DD_PROBLEM: 4859 return "device driver problem"; 4860 case EEH_RECOVERY: 4861 return "EEH recovery"; 4862 case FW_UPDATED: 4863 return "firmware updated"; 4864 case LOW_MEMORY: 4865 return "low Memory"; 4866 default: 4867 return "unknown"; 4868 } 4869 } 4870 4871 static void handle_error_indication(union ibmvnic_crq *crq, 4872 struct ibmvnic_adapter *adapter) 4873 { 4874 struct device *dev = &adapter->vdev->dev; 4875 u16 cause; 4876 4877 cause = be16_to_cpu(crq->error_indication.error_cause); 4878 4879 dev_warn_ratelimited(dev, 4880 "Firmware reports %serror, cause: %s. Starting recovery...\n", 4881 crq->error_indication.flags 4882 & IBMVNIC_FATAL_ERROR ? "FATAL " : "", 4883 ibmvnic_fw_err_cause(cause)); 4884 4885 if (crq->error_indication.flags & IBMVNIC_FATAL_ERROR) 4886 ibmvnic_reset(adapter, VNIC_RESET_FATAL); 4887 else 4888 ibmvnic_reset(adapter, VNIC_RESET_NON_FATAL); 4889 } 4890 4891 static int handle_change_mac_rsp(union ibmvnic_crq *crq, 4892 struct ibmvnic_adapter *adapter) 4893 { 4894 struct net_device *netdev = adapter->netdev; 4895 struct device *dev = &adapter->vdev->dev; 4896 long rc; 4897 4898 rc = crq->change_mac_addr_rsp.rc.code; 4899 if (rc) { 4900 dev_err(dev, "Error %ld in CHANGE_MAC_ADDR_RSP\n", rc); 4901 goto out; 4902 } 4903 /* crq->change_mac_addr.mac_addr is the requested one 4904 * crq->change_mac_addr_rsp.mac_addr is the returned valid one. 4905 */ 4906 eth_hw_addr_set(netdev, &crq->change_mac_addr_rsp.mac_addr[0]); 4907 ether_addr_copy(adapter->mac_addr, 4908 &crq->change_mac_addr_rsp.mac_addr[0]); 4909 out: 4910 complete(&adapter->fw_done); 4911 return rc; 4912 } 4913 4914 static void handle_request_cap_rsp(union ibmvnic_crq *crq, 4915 struct ibmvnic_adapter *adapter) 4916 { 4917 struct device *dev = &adapter->vdev->dev; 4918 u64 *req_value; 4919 char *name; 4920 4921 atomic_dec(&adapter->running_cap_crqs); 4922 netdev_dbg(adapter->netdev, "Outstanding request-caps: %d\n", 4923 atomic_read(&adapter->running_cap_crqs)); 4924 switch (be16_to_cpu(crq->request_capability_rsp.capability)) { 4925 case REQ_TX_QUEUES: 4926 req_value = &adapter->req_tx_queues; 4927 name = "tx"; 4928 break; 4929 case REQ_RX_QUEUES: 4930 req_value = &adapter->req_rx_queues; 4931 name = "rx"; 4932 break; 4933 case REQ_RX_ADD_QUEUES: 4934 req_value = &adapter->req_rx_add_queues; 4935 name = "rx_add"; 4936 break; 4937 case REQ_TX_ENTRIES_PER_SUBCRQ: 4938 req_value = &adapter->req_tx_entries_per_subcrq; 4939 name = "tx_entries_per_subcrq"; 4940 break; 4941 case REQ_RX_ADD_ENTRIES_PER_SUBCRQ: 4942 req_value = &adapter->req_rx_add_entries_per_subcrq; 4943 name = "rx_add_entries_per_subcrq"; 4944 break; 4945 case REQ_MTU: 4946 req_value = &adapter->req_mtu; 4947 name = "mtu"; 4948 break; 4949 case PROMISC_REQUESTED: 4950 req_value = &adapter->promisc; 4951 name = "promisc"; 4952 break; 4953 default: 4954 dev_err(dev, "Got invalid cap request rsp %d\n", 4955 crq->request_capability.capability); 4956 return; 4957 } 4958 4959 switch (crq->request_capability_rsp.rc.code) { 4960 case SUCCESS: 4961 break; 4962 case PARTIALSUCCESS: 4963 dev_info(dev, "req=%lld, rsp=%ld in %s queue, retrying.\n", 4964 *req_value, 4965 (long)be64_to_cpu(crq->request_capability_rsp.number), 4966 name); 4967 4968 if (be16_to_cpu(crq->request_capability_rsp.capability) == 4969 REQ_MTU) { 4970 pr_err("mtu of %llu is not supported. Reverting.\n", 4971 *req_value); 4972 *req_value = adapter->fallback.mtu; 4973 } else { 4974 *req_value = 4975 be64_to_cpu(crq->request_capability_rsp.number); 4976 } 4977 4978 send_request_cap(adapter, 1); 4979 return; 4980 default: 4981 dev_err(dev, "Error %d in request cap rsp\n", 4982 crq->request_capability_rsp.rc.code); 4983 return; 4984 } 4985 4986 /* Done receiving requested capabilities, query IP offload support */ 4987 if (atomic_read(&adapter->running_cap_crqs) == 0) 4988 send_query_ip_offload(adapter); 4989 } 4990 4991 static int handle_login_rsp(union ibmvnic_crq *login_rsp_crq, 4992 struct ibmvnic_adapter *adapter) 4993 { 4994 struct device *dev = &adapter->vdev->dev; 4995 struct net_device *netdev = adapter->netdev; 4996 struct ibmvnic_login_rsp_buffer *login_rsp = adapter->login_rsp_buf; 4997 struct ibmvnic_login_buffer *login = adapter->login_buf; 4998 u64 *tx_handle_array; 4999 u64 *rx_handle_array; 5000 int num_tx_pools; 5001 int num_rx_pools; 5002 u64 *size_array; 5003 int i; 5004 5005 /* CHECK: Test/set of login_pending does not need to be atomic 5006 * because only ibmvnic_tasklet tests/clears this. 5007 */ 5008 if (!adapter->login_pending) { 5009 netdev_warn(netdev, "Ignoring unexpected login response\n"); 5010 return 0; 5011 } 5012 adapter->login_pending = false; 5013 5014 dma_unmap_single(dev, adapter->login_buf_token, adapter->login_buf_sz, 5015 DMA_TO_DEVICE); 5016 dma_unmap_single(dev, adapter->login_rsp_buf_token, 5017 adapter->login_rsp_buf_sz, DMA_FROM_DEVICE); 5018 5019 /* If the number of queues requested can't be allocated by the 5020 * server, the login response will return with code 1. We will need 5021 * to resend the login buffer with fewer queues requested. 5022 */ 5023 if (login_rsp_crq->generic.rc.code) { 5024 adapter->init_done_rc = login_rsp_crq->generic.rc.code; 5025 complete(&adapter->init_done); 5026 return 0; 5027 } 5028 5029 if (adapter->failover_pending) { 5030 adapter->init_done_rc = -EAGAIN; 5031 netdev_dbg(netdev, "Failover pending, ignoring login response\n"); 5032 complete(&adapter->init_done); 5033 /* login response buffer will be released on reset */ 5034 return 0; 5035 } 5036 5037 netdev->mtu = adapter->req_mtu - ETH_HLEN; 5038 5039 netdev_dbg(adapter->netdev, "Login Response Buffer:\n"); 5040 for (i = 0; i < (adapter->login_rsp_buf_sz - 1) / 8 + 1; i++) { 5041 netdev_dbg(adapter->netdev, "%016lx\n", 5042 ((unsigned long *)(adapter->login_rsp_buf))[i]); 5043 } 5044 5045 /* Sanity checks */ 5046 if (login->num_txcomp_subcrqs != login_rsp->num_txsubm_subcrqs || 5047 (be32_to_cpu(login->num_rxcomp_subcrqs) * 5048 adapter->req_rx_add_queues != 5049 be32_to_cpu(login_rsp->num_rxadd_subcrqs))) { 5050 dev_err(dev, "FATAL: Inconsistent login and login rsp\n"); 5051 ibmvnic_reset(adapter, VNIC_RESET_FATAL); 5052 return -EIO; 5053 } 5054 size_array = (u64 *)((u8 *)(adapter->login_rsp_buf) + 5055 be32_to_cpu(adapter->login_rsp_buf->off_rxadd_buff_size)); 5056 /* variable buffer sizes are not supported, so just read the 5057 * first entry. 5058 */ 5059 adapter->cur_rx_buf_sz = be64_to_cpu(size_array[0]); 5060 5061 num_tx_pools = be32_to_cpu(adapter->login_rsp_buf->num_txsubm_subcrqs); 5062 num_rx_pools = be32_to_cpu(adapter->login_rsp_buf->num_rxadd_subcrqs); 5063 5064 tx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) + 5065 be32_to_cpu(adapter->login_rsp_buf->off_txsubm_subcrqs)); 5066 rx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) + 5067 be32_to_cpu(adapter->login_rsp_buf->off_rxadd_subcrqs)); 5068 5069 for (i = 0; i < num_tx_pools; i++) 5070 adapter->tx_scrq[i]->handle = tx_handle_array[i]; 5071 5072 for (i = 0; i < num_rx_pools; i++) 5073 adapter->rx_scrq[i]->handle = rx_handle_array[i]; 5074 5075 adapter->num_active_tx_scrqs = num_tx_pools; 5076 adapter->num_active_rx_scrqs = num_rx_pools; 5077 release_login_rsp_buffer(adapter); 5078 release_login_buffer(adapter); 5079 complete(&adapter->init_done); 5080 5081 return 0; 5082 } 5083 5084 static void handle_request_unmap_rsp(union ibmvnic_crq *crq, 5085 struct ibmvnic_adapter *adapter) 5086 { 5087 struct device *dev = &adapter->vdev->dev; 5088 long rc; 5089 5090 rc = crq->request_unmap_rsp.rc.code; 5091 if (rc) 5092 dev_err(dev, "Error %ld in REQUEST_UNMAP_RSP\n", rc); 5093 } 5094 5095 static void handle_query_map_rsp(union ibmvnic_crq *crq, 5096 struct ibmvnic_adapter *adapter) 5097 { 5098 struct net_device *netdev = adapter->netdev; 5099 struct device *dev = &adapter->vdev->dev; 5100 long rc; 5101 5102 rc = crq->query_map_rsp.rc.code; 5103 if (rc) { 5104 dev_err(dev, "Error %ld in QUERY_MAP_RSP\n", rc); 5105 return; 5106 } 5107 netdev_dbg(netdev, "page_size = %d\ntot_pages = %u\nfree_pages = %u\n", 5108 crq->query_map_rsp.page_size, 5109 __be32_to_cpu(crq->query_map_rsp.tot_pages), 5110 __be32_to_cpu(crq->query_map_rsp.free_pages)); 5111 } 5112 5113 static void handle_query_cap_rsp(union ibmvnic_crq *crq, 5114 struct ibmvnic_adapter *adapter) 5115 { 5116 struct net_device *netdev = adapter->netdev; 5117 struct device *dev = &adapter->vdev->dev; 5118 long rc; 5119 5120 atomic_dec(&adapter->running_cap_crqs); 5121 netdev_dbg(netdev, "Outstanding queries: %d\n", 5122 atomic_read(&adapter->running_cap_crqs)); 5123 rc = crq->query_capability.rc.code; 5124 if (rc) { 5125 dev_err(dev, "Error %ld in QUERY_CAP_RSP\n", rc); 5126 goto out; 5127 } 5128 5129 switch (be16_to_cpu(crq->query_capability.capability)) { 5130 case MIN_TX_QUEUES: 5131 adapter->min_tx_queues = 5132 be64_to_cpu(crq->query_capability.number); 5133 netdev_dbg(netdev, "min_tx_queues = %lld\n", 5134 adapter->min_tx_queues); 5135 break; 5136 case MIN_RX_QUEUES: 5137 adapter->min_rx_queues = 5138 be64_to_cpu(crq->query_capability.number); 5139 netdev_dbg(netdev, "min_rx_queues = %lld\n", 5140 adapter->min_rx_queues); 5141 break; 5142 case MIN_RX_ADD_QUEUES: 5143 adapter->min_rx_add_queues = 5144 be64_to_cpu(crq->query_capability.number); 5145 netdev_dbg(netdev, "min_rx_add_queues = %lld\n", 5146 adapter->min_rx_add_queues); 5147 break; 5148 case MAX_TX_QUEUES: 5149 adapter->max_tx_queues = 5150 be64_to_cpu(crq->query_capability.number); 5151 netdev_dbg(netdev, "max_tx_queues = %lld\n", 5152 adapter->max_tx_queues); 5153 break; 5154 case MAX_RX_QUEUES: 5155 adapter->max_rx_queues = 5156 be64_to_cpu(crq->query_capability.number); 5157 netdev_dbg(netdev, "max_rx_queues = %lld\n", 5158 adapter->max_rx_queues); 5159 break; 5160 case MAX_RX_ADD_QUEUES: 5161 adapter->max_rx_add_queues = 5162 be64_to_cpu(crq->query_capability.number); 5163 netdev_dbg(netdev, "max_rx_add_queues = %lld\n", 5164 adapter->max_rx_add_queues); 5165 break; 5166 case MIN_TX_ENTRIES_PER_SUBCRQ: 5167 adapter->min_tx_entries_per_subcrq = 5168 be64_to_cpu(crq->query_capability.number); 5169 netdev_dbg(netdev, "min_tx_entries_per_subcrq = %lld\n", 5170 adapter->min_tx_entries_per_subcrq); 5171 break; 5172 case MIN_RX_ADD_ENTRIES_PER_SUBCRQ: 5173 adapter->min_rx_add_entries_per_subcrq = 5174 be64_to_cpu(crq->query_capability.number); 5175 netdev_dbg(netdev, "min_rx_add_entrs_per_subcrq = %lld\n", 5176 adapter->min_rx_add_entries_per_subcrq); 5177 break; 5178 case MAX_TX_ENTRIES_PER_SUBCRQ: 5179 adapter->max_tx_entries_per_subcrq = 5180 be64_to_cpu(crq->query_capability.number); 5181 netdev_dbg(netdev, "max_tx_entries_per_subcrq = %lld\n", 5182 adapter->max_tx_entries_per_subcrq); 5183 break; 5184 case MAX_RX_ADD_ENTRIES_PER_SUBCRQ: 5185 adapter->max_rx_add_entries_per_subcrq = 5186 be64_to_cpu(crq->query_capability.number); 5187 netdev_dbg(netdev, "max_rx_add_entrs_per_subcrq = %lld\n", 5188 adapter->max_rx_add_entries_per_subcrq); 5189 break; 5190 case TCP_IP_OFFLOAD: 5191 adapter->tcp_ip_offload = 5192 be64_to_cpu(crq->query_capability.number); 5193 netdev_dbg(netdev, "tcp_ip_offload = %lld\n", 5194 adapter->tcp_ip_offload); 5195 break; 5196 case PROMISC_SUPPORTED: 5197 adapter->promisc_supported = 5198 be64_to_cpu(crq->query_capability.number); 5199 netdev_dbg(netdev, "promisc_supported = %lld\n", 5200 adapter->promisc_supported); 5201 break; 5202 case MIN_MTU: 5203 adapter->min_mtu = be64_to_cpu(crq->query_capability.number); 5204 netdev->min_mtu = adapter->min_mtu - ETH_HLEN; 5205 netdev_dbg(netdev, "min_mtu = %lld\n", adapter->min_mtu); 5206 break; 5207 case MAX_MTU: 5208 adapter->max_mtu = be64_to_cpu(crq->query_capability.number); 5209 netdev->max_mtu = adapter->max_mtu - ETH_HLEN; 5210 netdev_dbg(netdev, "max_mtu = %lld\n", adapter->max_mtu); 5211 break; 5212 case MAX_MULTICAST_FILTERS: 5213 adapter->max_multicast_filters = 5214 be64_to_cpu(crq->query_capability.number); 5215 netdev_dbg(netdev, "max_multicast_filters = %lld\n", 5216 adapter->max_multicast_filters); 5217 break; 5218 case VLAN_HEADER_INSERTION: 5219 adapter->vlan_header_insertion = 5220 be64_to_cpu(crq->query_capability.number); 5221 if (adapter->vlan_header_insertion) 5222 netdev->features |= NETIF_F_HW_VLAN_STAG_TX; 5223 netdev_dbg(netdev, "vlan_header_insertion = %lld\n", 5224 adapter->vlan_header_insertion); 5225 break; 5226 case RX_VLAN_HEADER_INSERTION: 5227 adapter->rx_vlan_header_insertion = 5228 be64_to_cpu(crq->query_capability.number); 5229 netdev_dbg(netdev, "rx_vlan_header_insertion = %lld\n", 5230 adapter->rx_vlan_header_insertion); 5231 break; 5232 case MAX_TX_SG_ENTRIES: 5233 adapter->max_tx_sg_entries = 5234 be64_to_cpu(crq->query_capability.number); 5235 netdev_dbg(netdev, "max_tx_sg_entries = %lld\n", 5236 adapter->max_tx_sg_entries); 5237 break; 5238 case RX_SG_SUPPORTED: 5239 adapter->rx_sg_supported = 5240 be64_to_cpu(crq->query_capability.number); 5241 netdev_dbg(netdev, "rx_sg_supported = %lld\n", 5242 adapter->rx_sg_supported); 5243 break; 5244 case OPT_TX_COMP_SUB_QUEUES: 5245 adapter->opt_tx_comp_sub_queues = 5246 be64_to_cpu(crq->query_capability.number); 5247 netdev_dbg(netdev, "opt_tx_comp_sub_queues = %lld\n", 5248 adapter->opt_tx_comp_sub_queues); 5249 break; 5250 case OPT_RX_COMP_QUEUES: 5251 adapter->opt_rx_comp_queues = 5252 be64_to_cpu(crq->query_capability.number); 5253 netdev_dbg(netdev, "opt_rx_comp_queues = %lld\n", 5254 adapter->opt_rx_comp_queues); 5255 break; 5256 case OPT_RX_BUFADD_Q_PER_RX_COMP_Q: 5257 adapter->opt_rx_bufadd_q_per_rx_comp_q = 5258 be64_to_cpu(crq->query_capability.number); 5259 netdev_dbg(netdev, "opt_rx_bufadd_q_per_rx_comp_q = %lld\n", 5260 adapter->opt_rx_bufadd_q_per_rx_comp_q); 5261 break; 5262 case OPT_TX_ENTRIES_PER_SUBCRQ: 5263 adapter->opt_tx_entries_per_subcrq = 5264 be64_to_cpu(crq->query_capability.number); 5265 netdev_dbg(netdev, "opt_tx_entries_per_subcrq = %lld\n", 5266 adapter->opt_tx_entries_per_subcrq); 5267 break; 5268 case OPT_RXBA_ENTRIES_PER_SUBCRQ: 5269 adapter->opt_rxba_entries_per_subcrq = 5270 be64_to_cpu(crq->query_capability.number); 5271 netdev_dbg(netdev, "opt_rxba_entries_per_subcrq = %lld\n", 5272 adapter->opt_rxba_entries_per_subcrq); 5273 break; 5274 case TX_RX_DESC_REQ: 5275 adapter->tx_rx_desc_req = crq->query_capability.number; 5276 netdev_dbg(netdev, "tx_rx_desc_req = %llx\n", 5277 adapter->tx_rx_desc_req); 5278 break; 5279 5280 default: 5281 netdev_err(netdev, "Got invalid cap rsp %d\n", 5282 crq->query_capability.capability); 5283 } 5284 5285 out: 5286 if (atomic_read(&adapter->running_cap_crqs) == 0) 5287 send_request_cap(adapter, 0); 5288 } 5289 5290 static int send_query_phys_parms(struct ibmvnic_adapter *adapter) 5291 { 5292 union ibmvnic_crq crq; 5293 int rc; 5294 5295 memset(&crq, 0, sizeof(crq)); 5296 crq.query_phys_parms.first = IBMVNIC_CRQ_CMD; 5297 crq.query_phys_parms.cmd = QUERY_PHYS_PARMS; 5298 5299 mutex_lock(&adapter->fw_lock); 5300 adapter->fw_done_rc = 0; 5301 reinit_completion(&adapter->fw_done); 5302 5303 rc = ibmvnic_send_crq(adapter, &crq); 5304 if (rc) { 5305 mutex_unlock(&adapter->fw_lock); 5306 return rc; 5307 } 5308 5309 rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000); 5310 if (rc) { 5311 mutex_unlock(&adapter->fw_lock); 5312 return rc; 5313 } 5314 5315 mutex_unlock(&adapter->fw_lock); 5316 return adapter->fw_done_rc ? -EIO : 0; 5317 } 5318 5319 static int handle_query_phys_parms_rsp(union ibmvnic_crq *crq, 5320 struct ibmvnic_adapter *adapter) 5321 { 5322 struct net_device *netdev = adapter->netdev; 5323 int rc; 5324 __be32 rspeed = cpu_to_be32(crq->query_phys_parms_rsp.speed); 5325 5326 rc = crq->query_phys_parms_rsp.rc.code; 5327 if (rc) { 5328 netdev_err(netdev, "Error %d in QUERY_PHYS_PARMS\n", rc); 5329 return rc; 5330 } 5331 switch (rspeed) { 5332 case IBMVNIC_10MBPS: 5333 adapter->speed = SPEED_10; 5334 break; 5335 case IBMVNIC_100MBPS: 5336 adapter->speed = SPEED_100; 5337 break; 5338 case IBMVNIC_1GBPS: 5339 adapter->speed = SPEED_1000; 5340 break; 5341 case IBMVNIC_10GBPS: 5342 adapter->speed = SPEED_10000; 5343 break; 5344 case IBMVNIC_25GBPS: 5345 adapter->speed = SPEED_25000; 5346 break; 5347 case IBMVNIC_40GBPS: 5348 adapter->speed = SPEED_40000; 5349 break; 5350 case IBMVNIC_50GBPS: 5351 adapter->speed = SPEED_50000; 5352 break; 5353 case IBMVNIC_100GBPS: 5354 adapter->speed = SPEED_100000; 5355 break; 5356 case IBMVNIC_200GBPS: 5357 adapter->speed = SPEED_200000; 5358 break; 5359 default: 5360 if (netif_carrier_ok(netdev)) 5361 netdev_warn(netdev, "Unknown speed 0x%08x\n", rspeed); 5362 adapter->speed = SPEED_UNKNOWN; 5363 } 5364 if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_FULL_DUPLEX) 5365 adapter->duplex = DUPLEX_FULL; 5366 else if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_HALF_DUPLEX) 5367 adapter->duplex = DUPLEX_HALF; 5368 else 5369 adapter->duplex = DUPLEX_UNKNOWN; 5370 5371 return rc; 5372 } 5373 5374 static void ibmvnic_handle_crq(union ibmvnic_crq *crq, 5375 struct ibmvnic_adapter *adapter) 5376 { 5377 struct ibmvnic_generic_crq *gen_crq = &crq->generic; 5378 struct net_device *netdev = adapter->netdev; 5379 struct device *dev = &adapter->vdev->dev; 5380 u64 *u64_crq = (u64 *)crq; 5381 long rc; 5382 5383 netdev_dbg(netdev, "Handling CRQ: %016lx %016lx\n", 5384 (unsigned long)cpu_to_be64(u64_crq[0]), 5385 (unsigned long)cpu_to_be64(u64_crq[1])); 5386 switch (gen_crq->first) { 5387 case IBMVNIC_CRQ_INIT_RSP: 5388 switch (gen_crq->cmd) { 5389 case IBMVNIC_CRQ_INIT: 5390 dev_info(dev, "Partner initialized\n"); 5391 adapter->from_passive_init = true; 5392 /* Discard any stale login responses from prev reset. 5393 * CHECK: should we clear even on INIT_COMPLETE? 5394 */ 5395 adapter->login_pending = false; 5396 5397 if (adapter->state == VNIC_DOWN) 5398 rc = ibmvnic_reset(adapter, VNIC_RESET_PASSIVE_INIT); 5399 else 5400 rc = ibmvnic_reset(adapter, VNIC_RESET_FAILOVER); 5401 5402 if (rc && rc != -EBUSY) { 5403 /* We were unable to schedule the failover 5404 * reset either because the adapter was still 5405 * probing (eg: during kexec) or we could not 5406 * allocate memory. Clear the failover_pending 5407 * flag since no one else will. We ignore 5408 * EBUSY because it means either FAILOVER reset 5409 * is already scheduled or the adapter is 5410 * being removed. 5411 */ 5412 netdev_err(netdev, 5413 "Error %ld scheduling failover reset\n", 5414 rc); 5415 adapter->failover_pending = false; 5416 } 5417 5418 if (!completion_done(&adapter->init_done)) { 5419 if (!adapter->init_done_rc) 5420 adapter->init_done_rc = -EAGAIN; 5421 complete(&adapter->init_done); 5422 } 5423 5424 break; 5425 case IBMVNIC_CRQ_INIT_COMPLETE: 5426 dev_info(dev, "Partner initialization complete\n"); 5427 adapter->crq.active = true; 5428 send_version_xchg(adapter); 5429 break; 5430 default: 5431 dev_err(dev, "Unknown crq cmd: %d\n", gen_crq->cmd); 5432 } 5433 return; 5434 case IBMVNIC_CRQ_XPORT_EVENT: 5435 netif_carrier_off(netdev); 5436 adapter->crq.active = false; 5437 /* terminate any thread waiting for a response 5438 * from the device 5439 */ 5440 if (!completion_done(&adapter->fw_done)) { 5441 adapter->fw_done_rc = -EIO; 5442 complete(&adapter->fw_done); 5443 } 5444 5445 /* if we got here during crq-init, retry crq-init */ 5446 if (!completion_done(&adapter->init_done)) { 5447 adapter->init_done_rc = -EAGAIN; 5448 complete(&adapter->init_done); 5449 } 5450 5451 if (!completion_done(&adapter->stats_done)) 5452 complete(&adapter->stats_done); 5453 if (test_bit(0, &adapter->resetting)) 5454 adapter->force_reset_recovery = true; 5455 if (gen_crq->cmd == IBMVNIC_PARTITION_MIGRATED) { 5456 dev_info(dev, "Migrated, re-enabling adapter\n"); 5457 ibmvnic_reset(adapter, VNIC_RESET_MOBILITY); 5458 } else if (gen_crq->cmd == IBMVNIC_DEVICE_FAILOVER) { 5459 dev_info(dev, "Backing device failover detected\n"); 5460 adapter->failover_pending = true; 5461 } else { 5462 /* The adapter lost the connection */ 5463 dev_err(dev, "Virtual Adapter failed (rc=%d)\n", 5464 gen_crq->cmd); 5465 ibmvnic_reset(adapter, VNIC_RESET_FATAL); 5466 } 5467 return; 5468 case IBMVNIC_CRQ_CMD_RSP: 5469 break; 5470 default: 5471 dev_err(dev, "Got an invalid msg type 0x%02x\n", 5472 gen_crq->first); 5473 return; 5474 } 5475 5476 switch (gen_crq->cmd) { 5477 case VERSION_EXCHANGE_RSP: 5478 rc = crq->version_exchange_rsp.rc.code; 5479 if (rc) { 5480 dev_err(dev, "Error %ld in VERSION_EXCHG_RSP\n", rc); 5481 break; 5482 } 5483 ibmvnic_version = 5484 be16_to_cpu(crq->version_exchange_rsp.version); 5485 dev_info(dev, "Partner protocol version is %d\n", 5486 ibmvnic_version); 5487 send_query_cap(adapter); 5488 break; 5489 case QUERY_CAPABILITY_RSP: 5490 handle_query_cap_rsp(crq, adapter); 5491 break; 5492 case QUERY_MAP_RSP: 5493 handle_query_map_rsp(crq, adapter); 5494 break; 5495 case REQUEST_MAP_RSP: 5496 adapter->fw_done_rc = crq->request_map_rsp.rc.code; 5497 complete(&adapter->fw_done); 5498 break; 5499 case REQUEST_UNMAP_RSP: 5500 handle_request_unmap_rsp(crq, adapter); 5501 break; 5502 case REQUEST_CAPABILITY_RSP: 5503 handle_request_cap_rsp(crq, adapter); 5504 break; 5505 case LOGIN_RSP: 5506 netdev_dbg(netdev, "Got Login Response\n"); 5507 handle_login_rsp(crq, adapter); 5508 break; 5509 case LOGICAL_LINK_STATE_RSP: 5510 netdev_dbg(netdev, 5511 "Got Logical Link State Response, state: %d rc: %d\n", 5512 crq->logical_link_state_rsp.link_state, 5513 crq->logical_link_state_rsp.rc.code); 5514 adapter->logical_link_state = 5515 crq->logical_link_state_rsp.link_state; 5516 adapter->init_done_rc = crq->logical_link_state_rsp.rc.code; 5517 complete(&adapter->init_done); 5518 break; 5519 case LINK_STATE_INDICATION: 5520 netdev_dbg(netdev, "Got Logical Link State Indication\n"); 5521 adapter->phys_link_state = 5522 crq->link_state_indication.phys_link_state; 5523 adapter->logical_link_state = 5524 crq->link_state_indication.logical_link_state; 5525 if (adapter->phys_link_state && adapter->logical_link_state) 5526 netif_carrier_on(netdev); 5527 else 5528 netif_carrier_off(netdev); 5529 break; 5530 case CHANGE_MAC_ADDR_RSP: 5531 netdev_dbg(netdev, "Got MAC address change Response\n"); 5532 adapter->fw_done_rc = handle_change_mac_rsp(crq, adapter); 5533 break; 5534 case ERROR_INDICATION: 5535 netdev_dbg(netdev, "Got Error Indication\n"); 5536 handle_error_indication(crq, adapter); 5537 break; 5538 case REQUEST_STATISTICS_RSP: 5539 netdev_dbg(netdev, "Got Statistics Response\n"); 5540 complete(&adapter->stats_done); 5541 break; 5542 case QUERY_IP_OFFLOAD_RSP: 5543 netdev_dbg(netdev, "Got Query IP offload Response\n"); 5544 handle_query_ip_offload_rsp(adapter); 5545 break; 5546 case MULTICAST_CTRL_RSP: 5547 netdev_dbg(netdev, "Got multicast control Response\n"); 5548 break; 5549 case CONTROL_IP_OFFLOAD_RSP: 5550 netdev_dbg(netdev, "Got Control IP offload Response\n"); 5551 dma_unmap_single(dev, adapter->ip_offload_ctrl_tok, 5552 sizeof(adapter->ip_offload_ctrl), 5553 DMA_TO_DEVICE); 5554 complete(&adapter->init_done); 5555 break; 5556 case COLLECT_FW_TRACE_RSP: 5557 netdev_dbg(netdev, "Got Collect firmware trace Response\n"); 5558 complete(&adapter->fw_done); 5559 break; 5560 case GET_VPD_SIZE_RSP: 5561 handle_vpd_size_rsp(crq, adapter); 5562 break; 5563 case GET_VPD_RSP: 5564 handle_vpd_rsp(crq, adapter); 5565 break; 5566 case QUERY_PHYS_PARMS_RSP: 5567 adapter->fw_done_rc = handle_query_phys_parms_rsp(crq, adapter); 5568 complete(&adapter->fw_done); 5569 break; 5570 default: 5571 netdev_err(netdev, "Got an invalid cmd type 0x%02x\n", 5572 gen_crq->cmd); 5573 } 5574 } 5575 5576 static irqreturn_t ibmvnic_interrupt(int irq, void *instance) 5577 { 5578 struct ibmvnic_adapter *adapter = instance; 5579 5580 tasklet_schedule(&adapter->tasklet); 5581 return IRQ_HANDLED; 5582 } 5583 5584 static void ibmvnic_tasklet(struct tasklet_struct *t) 5585 { 5586 struct ibmvnic_adapter *adapter = from_tasklet(adapter, t, tasklet); 5587 struct ibmvnic_crq_queue *queue = &adapter->crq; 5588 union ibmvnic_crq *crq; 5589 unsigned long flags; 5590 5591 spin_lock_irqsave(&queue->lock, flags); 5592 5593 /* Pull all the valid messages off the CRQ */ 5594 while ((crq = ibmvnic_next_crq(adapter)) != NULL) { 5595 /* This barrier makes sure ibmvnic_next_crq()'s 5596 * crq->generic.first & IBMVNIC_CRQ_CMD_RSP is loaded 5597 * before ibmvnic_handle_crq()'s 5598 * switch(gen_crq->first) and switch(gen_crq->cmd). 5599 */ 5600 dma_rmb(); 5601 ibmvnic_handle_crq(crq, adapter); 5602 crq->generic.first = 0; 5603 } 5604 5605 spin_unlock_irqrestore(&queue->lock, flags); 5606 } 5607 5608 static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *adapter) 5609 { 5610 struct vio_dev *vdev = adapter->vdev; 5611 int rc; 5612 5613 do { 5614 rc = plpar_hcall_norets(H_ENABLE_CRQ, vdev->unit_address); 5615 } while (rc == H_IN_PROGRESS || rc == H_BUSY || H_IS_LONG_BUSY(rc)); 5616 5617 if (rc) 5618 dev_err(&vdev->dev, "Error enabling adapter (rc=%d)\n", rc); 5619 5620 return rc; 5621 } 5622 5623 static int ibmvnic_reset_crq(struct ibmvnic_adapter *adapter) 5624 { 5625 struct ibmvnic_crq_queue *crq = &adapter->crq; 5626 struct device *dev = &adapter->vdev->dev; 5627 struct vio_dev *vdev = adapter->vdev; 5628 int rc; 5629 5630 /* Close the CRQ */ 5631 do { 5632 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address); 5633 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc)); 5634 5635 /* Clean out the queue */ 5636 if (!crq->msgs) 5637 return -EINVAL; 5638 5639 memset(crq->msgs, 0, PAGE_SIZE); 5640 crq->cur = 0; 5641 crq->active = false; 5642 5643 /* And re-open it again */ 5644 rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address, 5645 crq->msg_token, PAGE_SIZE); 5646 5647 if (rc == H_CLOSED) 5648 /* Adapter is good, but other end is not ready */ 5649 dev_warn(dev, "Partner adapter not ready\n"); 5650 else if (rc != 0) 5651 dev_warn(dev, "Couldn't register crq (rc=%d)\n", rc); 5652 5653 return rc; 5654 } 5655 5656 static void release_crq_queue(struct ibmvnic_adapter *adapter) 5657 { 5658 struct ibmvnic_crq_queue *crq = &adapter->crq; 5659 struct vio_dev *vdev = adapter->vdev; 5660 long rc; 5661 5662 if (!crq->msgs) 5663 return; 5664 5665 netdev_dbg(adapter->netdev, "Releasing CRQ\n"); 5666 free_irq(vdev->irq, adapter); 5667 tasklet_kill(&adapter->tasklet); 5668 do { 5669 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address); 5670 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc)); 5671 5672 dma_unmap_single(&vdev->dev, crq->msg_token, PAGE_SIZE, 5673 DMA_BIDIRECTIONAL); 5674 free_page((unsigned long)crq->msgs); 5675 crq->msgs = NULL; 5676 crq->active = false; 5677 } 5678 5679 static int init_crq_queue(struct ibmvnic_adapter *adapter) 5680 { 5681 struct ibmvnic_crq_queue *crq = &adapter->crq; 5682 struct device *dev = &adapter->vdev->dev; 5683 struct vio_dev *vdev = adapter->vdev; 5684 int rc, retrc = -ENOMEM; 5685 5686 if (crq->msgs) 5687 return 0; 5688 5689 crq->msgs = (union ibmvnic_crq *)get_zeroed_page(GFP_KERNEL); 5690 /* Should we allocate more than one page? */ 5691 5692 if (!crq->msgs) 5693 return -ENOMEM; 5694 5695 crq->size = PAGE_SIZE / sizeof(*crq->msgs); 5696 crq->msg_token = dma_map_single(dev, crq->msgs, PAGE_SIZE, 5697 DMA_BIDIRECTIONAL); 5698 if (dma_mapping_error(dev, crq->msg_token)) 5699 goto map_failed; 5700 5701 rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address, 5702 crq->msg_token, PAGE_SIZE); 5703 5704 if (rc == H_RESOURCE) 5705 /* maybe kexecing and resource is busy. try a reset */ 5706 rc = ibmvnic_reset_crq(adapter); 5707 retrc = rc; 5708 5709 if (rc == H_CLOSED) { 5710 dev_warn(dev, "Partner adapter not ready\n"); 5711 } else if (rc) { 5712 dev_warn(dev, "Error %d opening adapter\n", rc); 5713 goto reg_crq_failed; 5714 } 5715 5716 retrc = 0; 5717 5718 tasklet_setup(&adapter->tasklet, (void *)ibmvnic_tasklet); 5719 5720 netdev_dbg(adapter->netdev, "registering irq 0x%x\n", vdev->irq); 5721 snprintf(crq->name, sizeof(crq->name), "ibmvnic-%x", 5722 adapter->vdev->unit_address); 5723 rc = request_irq(vdev->irq, ibmvnic_interrupt, 0, crq->name, adapter); 5724 if (rc) { 5725 dev_err(dev, "Couldn't register irq 0x%x. rc=%d\n", 5726 vdev->irq, rc); 5727 goto req_irq_failed; 5728 } 5729 5730 rc = vio_enable_interrupts(vdev); 5731 if (rc) { 5732 dev_err(dev, "Error %d enabling interrupts\n", rc); 5733 goto req_irq_failed; 5734 } 5735 5736 crq->cur = 0; 5737 spin_lock_init(&crq->lock); 5738 5739 /* process any CRQs that were queued before we enabled interrupts */ 5740 tasklet_schedule(&adapter->tasklet); 5741 5742 return retrc; 5743 5744 req_irq_failed: 5745 tasklet_kill(&adapter->tasklet); 5746 do { 5747 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address); 5748 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc)); 5749 reg_crq_failed: 5750 dma_unmap_single(dev, crq->msg_token, PAGE_SIZE, DMA_BIDIRECTIONAL); 5751 map_failed: 5752 free_page((unsigned long)crq->msgs); 5753 crq->msgs = NULL; 5754 return retrc; 5755 } 5756 5757 static int ibmvnic_reset_init(struct ibmvnic_adapter *adapter, bool reset) 5758 { 5759 struct device *dev = &adapter->vdev->dev; 5760 unsigned long timeout = msecs_to_jiffies(20000); 5761 u64 old_num_rx_queues = adapter->req_rx_queues; 5762 u64 old_num_tx_queues = adapter->req_tx_queues; 5763 int rc; 5764 5765 adapter->from_passive_init = false; 5766 5767 rc = ibmvnic_send_crq_init(adapter); 5768 if (rc) { 5769 dev_err(dev, "Send crq init failed with error %d\n", rc); 5770 return rc; 5771 } 5772 5773 if (!wait_for_completion_timeout(&adapter->init_done, timeout)) { 5774 dev_err(dev, "Initialization sequence timed out\n"); 5775 return -ETIMEDOUT; 5776 } 5777 5778 if (adapter->init_done_rc) { 5779 release_crq_queue(adapter); 5780 dev_err(dev, "CRQ-init failed, %d\n", adapter->init_done_rc); 5781 return adapter->init_done_rc; 5782 } 5783 5784 if (adapter->from_passive_init) { 5785 adapter->state = VNIC_OPEN; 5786 adapter->from_passive_init = false; 5787 dev_err(dev, "CRQ-init failed, passive-init\n"); 5788 return -EINVAL; 5789 } 5790 5791 if (reset && 5792 test_bit(0, &adapter->resetting) && !adapter->wait_for_reset && 5793 adapter->reset_reason != VNIC_RESET_MOBILITY) { 5794 if (adapter->req_rx_queues != old_num_rx_queues || 5795 adapter->req_tx_queues != old_num_tx_queues) { 5796 release_sub_crqs(adapter, 0); 5797 rc = init_sub_crqs(adapter); 5798 } else { 5799 rc = reset_sub_crq_queues(adapter); 5800 } 5801 } else { 5802 rc = init_sub_crqs(adapter); 5803 } 5804 5805 if (rc) { 5806 dev_err(dev, "Initialization of sub crqs failed\n"); 5807 release_crq_queue(adapter); 5808 return rc; 5809 } 5810 5811 rc = init_sub_crq_irqs(adapter); 5812 if (rc) { 5813 dev_err(dev, "Failed to initialize sub crq irqs\n"); 5814 release_crq_queue(adapter); 5815 } 5816 5817 return rc; 5818 } 5819 5820 static struct device_attribute dev_attr_failover; 5821 5822 static int ibmvnic_probe(struct vio_dev *dev, const struct vio_device_id *id) 5823 { 5824 struct ibmvnic_adapter *adapter; 5825 struct net_device *netdev; 5826 unsigned char *mac_addr_p; 5827 unsigned long flags; 5828 bool init_success; 5829 int rc; 5830 5831 dev_dbg(&dev->dev, "entering ibmvnic_probe for UA 0x%x\n", 5832 dev->unit_address); 5833 5834 mac_addr_p = (unsigned char *)vio_get_attribute(dev, 5835 VETH_MAC_ADDR, NULL); 5836 if (!mac_addr_p) { 5837 dev_err(&dev->dev, 5838 "(%s:%3.3d) ERROR: Can't find MAC_ADDR attribute\n", 5839 __FILE__, __LINE__); 5840 return 0; 5841 } 5842 5843 netdev = alloc_etherdev_mq(sizeof(struct ibmvnic_adapter), 5844 IBMVNIC_MAX_QUEUES); 5845 if (!netdev) 5846 return -ENOMEM; 5847 5848 adapter = netdev_priv(netdev); 5849 adapter->state = VNIC_PROBING; 5850 dev_set_drvdata(&dev->dev, netdev); 5851 adapter->vdev = dev; 5852 adapter->netdev = netdev; 5853 adapter->login_pending = false; 5854 memset(&adapter->map_ids, 0, sizeof(adapter->map_ids)); 5855 /* map_ids start at 1, so ensure map_id 0 is always "in-use" */ 5856 bitmap_set(adapter->map_ids, 0, 1); 5857 5858 ether_addr_copy(adapter->mac_addr, mac_addr_p); 5859 eth_hw_addr_set(netdev, adapter->mac_addr); 5860 netdev->irq = dev->irq; 5861 netdev->netdev_ops = &ibmvnic_netdev_ops; 5862 netdev->ethtool_ops = &ibmvnic_ethtool_ops; 5863 SET_NETDEV_DEV(netdev, &dev->dev); 5864 5865 INIT_WORK(&adapter->ibmvnic_reset, __ibmvnic_reset); 5866 INIT_DELAYED_WORK(&adapter->ibmvnic_delayed_reset, 5867 __ibmvnic_delayed_reset); 5868 INIT_LIST_HEAD(&adapter->rwi_list); 5869 spin_lock_init(&adapter->rwi_lock); 5870 spin_lock_init(&adapter->state_lock); 5871 mutex_init(&adapter->fw_lock); 5872 init_completion(&adapter->probe_done); 5873 init_completion(&adapter->init_done); 5874 init_completion(&adapter->fw_done); 5875 init_completion(&adapter->reset_done); 5876 init_completion(&adapter->stats_done); 5877 clear_bit(0, &adapter->resetting); 5878 adapter->prev_rx_buf_sz = 0; 5879 adapter->prev_mtu = 0; 5880 5881 init_success = false; 5882 do { 5883 reinit_init_done(adapter); 5884 5885 /* clear any failovers we got in the previous pass 5886 * since we are reinitializing the CRQ 5887 */ 5888 adapter->failover_pending = false; 5889 5890 /* If we had already initialized CRQ, we may have one or 5891 * more resets queued already. Discard those and release 5892 * the CRQ before initializing the CRQ again. 5893 */ 5894 release_crq_queue(adapter); 5895 5896 /* Since we are still in PROBING state, __ibmvnic_reset() 5897 * will not access the ->rwi_list and since we released CRQ, 5898 * we won't get _new_ transport events. But there maybe an 5899 * ongoing ibmvnic_reset() call. So serialize access to 5900 * rwi_list. If we win the race, ibvmnic_reset() could add 5901 * a reset after we purged but thats ok - we just may end 5902 * up with an extra reset (i.e similar to having two or more 5903 * resets in the queue at once). 5904 * CHECK. 5905 */ 5906 spin_lock_irqsave(&adapter->rwi_lock, flags); 5907 flush_reset_queue(adapter); 5908 spin_unlock_irqrestore(&adapter->rwi_lock, flags); 5909 5910 rc = init_crq_queue(adapter); 5911 if (rc) { 5912 dev_err(&dev->dev, "Couldn't initialize crq. rc=%d\n", 5913 rc); 5914 goto ibmvnic_init_fail; 5915 } 5916 5917 rc = ibmvnic_reset_init(adapter, false); 5918 } while (rc == -EAGAIN); 5919 5920 /* We are ignoring the error from ibmvnic_reset_init() assuming that the 5921 * partner is not ready. CRQ is not active. When the partner becomes 5922 * ready, we will do the passive init reset. 5923 */ 5924 5925 if (!rc) 5926 init_success = true; 5927 5928 rc = init_stats_buffers(adapter); 5929 if (rc) 5930 goto ibmvnic_init_fail; 5931 5932 rc = init_stats_token(adapter); 5933 if (rc) 5934 goto ibmvnic_stats_fail; 5935 5936 rc = device_create_file(&dev->dev, &dev_attr_failover); 5937 if (rc) 5938 goto ibmvnic_dev_file_err; 5939 5940 netif_carrier_off(netdev); 5941 5942 if (init_success) { 5943 adapter->state = VNIC_PROBED; 5944 netdev->mtu = adapter->req_mtu - ETH_HLEN; 5945 netdev->min_mtu = adapter->min_mtu - ETH_HLEN; 5946 netdev->max_mtu = adapter->max_mtu - ETH_HLEN; 5947 } else { 5948 adapter->state = VNIC_DOWN; 5949 } 5950 5951 adapter->wait_for_reset = false; 5952 adapter->last_reset_time = jiffies; 5953 5954 rc = register_netdev(netdev); 5955 if (rc) { 5956 dev_err(&dev->dev, "failed to register netdev rc=%d\n", rc); 5957 goto ibmvnic_register_fail; 5958 } 5959 dev_info(&dev->dev, "ibmvnic registered\n"); 5960 5961 complete(&adapter->probe_done); 5962 5963 return 0; 5964 5965 ibmvnic_register_fail: 5966 device_remove_file(&dev->dev, &dev_attr_failover); 5967 5968 ibmvnic_dev_file_err: 5969 release_stats_token(adapter); 5970 5971 ibmvnic_stats_fail: 5972 release_stats_buffers(adapter); 5973 5974 ibmvnic_init_fail: 5975 release_sub_crqs(adapter, 1); 5976 release_crq_queue(adapter); 5977 5978 /* cleanup worker thread after releasing CRQ so we don't get 5979 * transport events (i.e new work items for the worker thread). 5980 */ 5981 adapter->state = VNIC_REMOVING; 5982 complete(&adapter->probe_done); 5983 flush_work(&adapter->ibmvnic_reset); 5984 flush_delayed_work(&adapter->ibmvnic_delayed_reset); 5985 5986 flush_reset_queue(adapter); 5987 5988 mutex_destroy(&adapter->fw_lock); 5989 free_netdev(netdev); 5990 5991 return rc; 5992 } 5993 5994 static void ibmvnic_remove(struct vio_dev *dev) 5995 { 5996 struct net_device *netdev = dev_get_drvdata(&dev->dev); 5997 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 5998 unsigned long flags; 5999 6000 spin_lock_irqsave(&adapter->state_lock, flags); 6001 6002 /* If ibmvnic_reset() is scheduling a reset, wait for it to 6003 * finish. Then, set the state to REMOVING to prevent it from 6004 * scheduling any more work and to have reset functions ignore 6005 * any resets that have already been scheduled. Drop the lock 6006 * after setting state, so __ibmvnic_reset() which is called 6007 * from the flush_work() below, can make progress. 6008 */ 6009 spin_lock(&adapter->rwi_lock); 6010 adapter->state = VNIC_REMOVING; 6011 spin_unlock(&adapter->rwi_lock); 6012 6013 spin_unlock_irqrestore(&adapter->state_lock, flags); 6014 6015 flush_work(&adapter->ibmvnic_reset); 6016 flush_delayed_work(&adapter->ibmvnic_delayed_reset); 6017 6018 rtnl_lock(); 6019 unregister_netdevice(netdev); 6020 6021 release_resources(adapter); 6022 release_rx_pools(adapter); 6023 release_tx_pools(adapter); 6024 release_sub_crqs(adapter, 1); 6025 release_crq_queue(adapter); 6026 6027 release_stats_token(adapter); 6028 release_stats_buffers(adapter); 6029 6030 adapter->state = VNIC_REMOVED; 6031 6032 rtnl_unlock(); 6033 mutex_destroy(&adapter->fw_lock); 6034 device_remove_file(&dev->dev, &dev_attr_failover); 6035 free_netdev(netdev); 6036 dev_set_drvdata(&dev->dev, NULL); 6037 } 6038 6039 static ssize_t failover_store(struct device *dev, struct device_attribute *attr, 6040 const char *buf, size_t count) 6041 { 6042 struct net_device *netdev = dev_get_drvdata(dev); 6043 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 6044 unsigned long retbuf[PLPAR_HCALL_BUFSIZE]; 6045 __be64 session_token; 6046 long rc; 6047 6048 if (!sysfs_streq(buf, "1")) 6049 return -EINVAL; 6050 6051 rc = plpar_hcall(H_VIOCTL, retbuf, adapter->vdev->unit_address, 6052 H_GET_SESSION_TOKEN, 0, 0, 0); 6053 if (rc) { 6054 netdev_err(netdev, "Couldn't retrieve session token, rc %ld\n", 6055 rc); 6056 goto last_resort; 6057 } 6058 6059 session_token = (__be64)retbuf[0]; 6060 netdev_dbg(netdev, "Initiating client failover, session id %llx\n", 6061 be64_to_cpu(session_token)); 6062 rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address, 6063 H_SESSION_ERR_DETECTED, session_token, 0, 0); 6064 if (rc) { 6065 netdev_err(netdev, 6066 "H_VIOCTL initiated failover failed, rc %ld\n", 6067 rc); 6068 goto last_resort; 6069 } 6070 6071 return count; 6072 6073 last_resort: 6074 netdev_dbg(netdev, "Trying to send CRQ_CMD, the last resort\n"); 6075 ibmvnic_reset(adapter, VNIC_RESET_FAILOVER); 6076 6077 return count; 6078 } 6079 static DEVICE_ATTR_WO(failover); 6080 6081 static unsigned long ibmvnic_get_desired_dma(struct vio_dev *vdev) 6082 { 6083 struct net_device *netdev = dev_get_drvdata(&vdev->dev); 6084 struct ibmvnic_adapter *adapter; 6085 struct iommu_table *tbl; 6086 unsigned long ret = 0; 6087 int i; 6088 6089 tbl = get_iommu_table_base(&vdev->dev); 6090 6091 /* netdev inits at probe time along with the structures we need below*/ 6092 if (!netdev) 6093 return IOMMU_PAGE_ALIGN(IBMVNIC_IO_ENTITLEMENT_DEFAULT, tbl); 6094 6095 adapter = netdev_priv(netdev); 6096 6097 ret += PAGE_SIZE; /* the crq message queue */ 6098 ret += IOMMU_PAGE_ALIGN(sizeof(struct ibmvnic_statistics), tbl); 6099 6100 for (i = 0; i < adapter->req_tx_queues + adapter->req_rx_queues; i++) 6101 ret += 4 * PAGE_SIZE; /* the scrq message queue */ 6102 6103 for (i = 0; i < adapter->num_active_rx_pools; i++) 6104 ret += adapter->rx_pool[i].size * 6105 IOMMU_PAGE_ALIGN(adapter->rx_pool[i].buff_size, tbl); 6106 6107 return ret; 6108 } 6109 6110 static int ibmvnic_resume(struct device *dev) 6111 { 6112 struct net_device *netdev = dev_get_drvdata(dev); 6113 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 6114 6115 if (adapter->state != VNIC_OPEN) 6116 return 0; 6117 6118 tasklet_schedule(&adapter->tasklet); 6119 6120 return 0; 6121 } 6122 6123 static const struct vio_device_id ibmvnic_device_table[] = { 6124 {"network", "IBM,vnic"}, 6125 {"", "" } 6126 }; 6127 MODULE_DEVICE_TABLE(vio, ibmvnic_device_table); 6128 6129 static const struct dev_pm_ops ibmvnic_pm_ops = { 6130 .resume = ibmvnic_resume 6131 }; 6132 6133 static struct vio_driver ibmvnic_driver = { 6134 .id_table = ibmvnic_device_table, 6135 .probe = ibmvnic_probe, 6136 .remove = ibmvnic_remove, 6137 .get_desired_dma = ibmvnic_get_desired_dma, 6138 .name = ibmvnic_driver_name, 6139 .pm = &ibmvnic_pm_ops, 6140 }; 6141 6142 /* module functions */ 6143 static int __init ibmvnic_module_init(void) 6144 { 6145 pr_info("%s: %s %s\n", ibmvnic_driver_name, ibmvnic_driver_string, 6146 IBMVNIC_DRIVER_VERSION); 6147 6148 return vio_register_driver(&ibmvnic_driver); 6149 } 6150 6151 static void __exit ibmvnic_module_exit(void) 6152 { 6153 vio_unregister_driver(&ibmvnic_driver); 6154 } 6155 6156 module_init(ibmvnic_module_init); 6157 module_exit(ibmvnic_module_exit); 6158