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 u64 prev = 0; 261 int rc; 262 263 if (!reuse_ltb(ltb, size)) { 264 dev_dbg(dev, 265 "LTB size changed from 0x%llx to 0x%x, reallocating\n", 266 ltb->size, size); 267 prev = ltb->size; 268 free_long_term_buff(adapter, ltb); 269 } 270 271 if (ltb->buff) { 272 dev_dbg(dev, "Reusing LTB [map %d, size 0x%llx]\n", 273 ltb->map_id, ltb->size); 274 } else { 275 ltb->buff = dma_alloc_coherent(dev, size, <b->addr, 276 GFP_KERNEL); 277 if (!ltb->buff) { 278 dev_err(dev, "Couldn't alloc long term buffer\n"); 279 return -ENOMEM; 280 } 281 ltb->size = size; 282 283 ltb->map_id = find_first_zero_bit(adapter->map_ids, 284 MAX_MAP_ID); 285 bitmap_set(adapter->map_ids, ltb->map_id, 1); 286 287 dev_dbg(dev, 288 "Allocated new LTB [map %d, size 0x%llx was 0x%llx]\n", 289 ltb->map_id, ltb->size, prev); 290 } 291 292 /* Ensure ltb is zeroed - specially when reusing it. */ 293 memset(ltb->buff, 0, ltb->size); 294 295 mutex_lock(&adapter->fw_lock); 296 adapter->fw_done_rc = 0; 297 reinit_completion(&adapter->fw_done); 298 299 rc = send_request_map(adapter, ltb->addr, ltb->size, ltb->map_id); 300 if (rc) { 301 dev_err(dev, "send_request_map failed, rc = %d\n", rc); 302 goto out; 303 } 304 305 rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000); 306 if (rc) { 307 dev_err(dev, "LTB map request aborted or timed out, rc = %d\n", 308 rc); 309 goto out; 310 } 311 312 if (adapter->fw_done_rc) { 313 dev_err(dev, "Couldn't map LTB, rc = %d\n", 314 adapter->fw_done_rc); 315 rc = -EIO; 316 goto out; 317 } 318 rc = 0; 319 out: 320 /* don't free LTB on communication error - see function header */ 321 mutex_unlock(&adapter->fw_lock); 322 return rc; 323 } 324 325 static void free_long_term_buff(struct ibmvnic_adapter *adapter, 326 struct ibmvnic_long_term_buff *ltb) 327 { 328 struct device *dev = &adapter->vdev->dev; 329 330 if (!ltb->buff) 331 return; 332 333 /* VIOS automatically unmaps the long term buffer at remote 334 * end for the following resets: 335 * FAILOVER, MOBILITY, TIMEOUT. 336 */ 337 if (adapter->reset_reason != VNIC_RESET_FAILOVER && 338 adapter->reset_reason != VNIC_RESET_MOBILITY && 339 adapter->reset_reason != VNIC_RESET_TIMEOUT) 340 send_request_unmap(adapter, ltb->map_id); 341 342 dma_free_coherent(dev, ltb->size, ltb->buff, ltb->addr); 343 344 ltb->buff = NULL; 345 /* mark this map_id free */ 346 bitmap_clear(adapter->map_ids, ltb->map_id, 1); 347 ltb->map_id = 0; 348 } 349 350 /** 351 * free_ltb_set - free the given set of long term buffers (LTBS) 352 * @adapter: The ibmvnic adapter containing this ltb set 353 * @ltb_set: The ltb_set to be freed 354 * 355 * Free the set of LTBs in the given set. 356 */ 357 358 static void free_ltb_set(struct ibmvnic_adapter *adapter, 359 struct ibmvnic_ltb_set *ltb_set) 360 { 361 int i; 362 363 for (i = 0; i < ltb_set->num_ltbs; i++) 364 free_long_term_buff(adapter, <b_set->ltbs[i]); 365 366 kfree(ltb_set->ltbs); 367 ltb_set->ltbs = NULL; 368 ltb_set->num_ltbs = 0; 369 } 370 371 /** 372 * alloc_ltb_set() - Allocate a set of long term buffers (LTBs) 373 * 374 * @adapter: ibmvnic adapter associated to the LTB 375 * @ltb_set: container object for the set of LTBs 376 * @num_buffs: Number of buffers in the LTB 377 * @buff_size: Size of each buffer in the LTB 378 * 379 * Allocate a set of LTBs to accommodate @num_buffs buffers of @buff_size 380 * each. We currently cap size each LTB to IBMVNIC_ONE_LTB_SIZE. If the 381 * new set of LTBs have fewer LTBs than the old set, free the excess LTBs. 382 * If new set needs more than in old set, allocate the remaining ones. 383 * Try and reuse as many LTBs as possible and avoid reallocation. 384 * 385 * Any changes to this allocation strategy must be reflected in 386 * map_rxpool_buff_to_ltb() and map_txpool_buff_to_ltb(). 387 */ 388 static int alloc_ltb_set(struct ibmvnic_adapter *adapter, 389 struct ibmvnic_ltb_set *ltb_set, int num_buffs, 390 int buff_size) 391 { 392 struct device *dev = &adapter->vdev->dev; 393 struct ibmvnic_ltb_set old_set; 394 struct ibmvnic_ltb_set new_set; 395 int rem_size; 396 int tot_size; /* size of all ltbs */ 397 int ltb_size; /* size of one ltb */ 398 int nltbs; 399 int rc; 400 int n; 401 int i; 402 403 dev_dbg(dev, "%s() num_buffs %d, buff_size %d\n", __func__, num_buffs, 404 buff_size); 405 406 ltb_size = rounddown(IBMVNIC_ONE_LTB_SIZE, buff_size); 407 tot_size = num_buffs * buff_size; 408 409 if (ltb_size > tot_size) 410 ltb_size = tot_size; 411 412 nltbs = tot_size / ltb_size; 413 if (tot_size % ltb_size) 414 nltbs++; 415 416 old_set = *ltb_set; 417 418 if (old_set.num_ltbs == nltbs) { 419 new_set = old_set; 420 } else { 421 int tmp = nltbs * sizeof(struct ibmvnic_long_term_buff); 422 423 new_set.ltbs = kzalloc(tmp, GFP_KERNEL); 424 if (!new_set.ltbs) 425 return -ENOMEM; 426 427 new_set.num_ltbs = nltbs; 428 429 /* Free any excess ltbs in old set */ 430 for (i = new_set.num_ltbs; i < old_set.num_ltbs; i++) 431 free_long_term_buff(adapter, &old_set.ltbs[i]); 432 433 /* Copy remaining ltbs to new set. All LTBs except the 434 * last one are of the same size. alloc_long_term_buff() 435 * will realloc if the size changes. 436 */ 437 n = min(old_set.num_ltbs, new_set.num_ltbs); 438 for (i = 0; i < n; i++) 439 new_set.ltbs[i] = old_set.ltbs[i]; 440 441 /* Any additional ltbs in new set will have NULL ltbs for 442 * now and will be allocated in alloc_long_term_buff(). 443 */ 444 445 /* We no longer need the old_set so free it. Note that we 446 * may have reused some ltbs from old set and freed excess 447 * ltbs above. So we only need to free the container now 448 * not the LTBs themselves. (i.e. dont free_ltb_set()!) 449 */ 450 kfree(old_set.ltbs); 451 old_set.ltbs = NULL; 452 old_set.num_ltbs = 0; 453 454 /* Install the new set. If allocations fail below, we will 455 * retry later and know what size LTBs we need. 456 */ 457 *ltb_set = new_set; 458 } 459 460 i = 0; 461 rem_size = tot_size; 462 while (rem_size) { 463 if (ltb_size > rem_size) 464 ltb_size = rem_size; 465 466 rem_size -= ltb_size; 467 468 rc = alloc_long_term_buff(adapter, &new_set.ltbs[i], ltb_size); 469 if (rc) 470 goto out; 471 i++; 472 } 473 474 WARN_ON(i != new_set.num_ltbs); 475 476 return 0; 477 out: 478 /* We may have allocated one/more LTBs before failing and we 479 * want to try and reuse on next reset. So don't free ltb set. 480 */ 481 return rc; 482 } 483 484 /** 485 * map_rxpool_buf_to_ltb - Map given rxpool buffer to offset in an LTB. 486 * @rxpool: The receive buffer pool containing buffer 487 * @bufidx: Index of buffer in rxpool 488 * @ltbp: (Output) pointer to the long term buffer containing the buffer 489 * @offset: (Output) offset of buffer in the LTB from @ltbp 490 * 491 * Map the given buffer identified by [rxpool, bufidx] to an LTB in the 492 * pool and its corresponding offset. Assume for now that each LTB is of 493 * different size but could possibly be optimized based on the allocation 494 * strategy in alloc_ltb_set(). 495 */ 496 static void map_rxpool_buf_to_ltb(struct ibmvnic_rx_pool *rxpool, 497 unsigned int bufidx, 498 struct ibmvnic_long_term_buff **ltbp, 499 unsigned int *offset) 500 { 501 struct ibmvnic_long_term_buff *ltb; 502 int nbufs; /* # of buffers in one ltb */ 503 int i; 504 505 WARN_ON(bufidx >= rxpool->size); 506 507 for (i = 0; i < rxpool->ltb_set.num_ltbs; i++) { 508 ltb = &rxpool->ltb_set.ltbs[i]; 509 nbufs = ltb->size / rxpool->buff_size; 510 if (bufidx < nbufs) 511 break; 512 bufidx -= nbufs; 513 } 514 515 *ltbp = ltb; 516 *offset = bufidx * rxpool->buff_size; 517 } 518 519 /** 520 * map_txpool_buf_to_ltb - Map given txpool buffer to offset in an LTB. 521 * @txpool: The transmit buffer pool containing buffer 522 * @bufidx: Index of buffer in txpool 523 * @ltbp: (Output) pointer to the long term buffer (LTB) containing the buffer 524 * @offset: (Output) offset of buffer in the LTB from @ltbp 525 * 526 * Map the given buffer identified by [txpool, bufidx] to an LTB in the 527 * pool and its corresponding offset. 528 */ 529 static void map_txpool_buf_to_ltb(struct ibmvnic_tx_pool *txpool, 530 unsigned int bufidx, 531 struct ibmvnic_long_term_buff **ltbp, 532 unsigned int *offset) 533 { 534 struct ibmvnic_long_term_buff *ltb; 535 int nbufs; /* # of buffers in one ltb */ 536 int i; 537 538 WARN_ON_ONCE(bufidx >= txpool->num_buffers); 539 540 for (i = 0; i < txpool->ltb_set.num_ltbs; i++) { 541 ltb = &txpool->ltb_set.ltbs[i]; 542 nbufs = ltb->size / txpool->buf_size; 543 if (bufidx < nbufs) 544 break; 545 bufidx -= nbufs; 546 } 547 548 *ltbp = ltb; 549 *offset = bufidx * txpool->buf_size; 550 } 551 552 static void deactivate_rx_pools(struct ibmvnic_adapter *adapter) 553 { 554 int i; 555 556 for (i = 0; i < adapter->num_active_rx_pools; i++) 557 adapter->rx_pool[i].active = 0; 558 } 559 560 static void replenish_rx_pool(struct ibmvnic_adapter *adapter, 561 struct ibmvnic_rx_pool *pool) 562 { 563 int count = pool->size - atomic_read(&pool->available); 564 u64 handle = adapter->rx_scrq[pool->index]->handle; 565 struct device *dev = &adapter->vdev->dev; 566 struct ibmvnic_ind_xmit_queue *ind_bufp; 567 struct ibmvnic_sub_crq_queue *rx_scrq; 568 struct ibmvnic_long_term_buff *ltb; 569 union sub_crq *sub_crq; 570 int buffers_added = 0; 571 unsigned long lpar_rc; 572 struct sk_buff *skb; 573 unsigned int offset; 574 dma_addr_t dma_addr; 575 unsigned char *dst; 576 int shift = 0; 577 int bufidx; 578 int i; 579 580 if (!pool->active) 581 return; 582 583 rx_scrq = adapter->rx_scrq[pool->index]; 584 ind_bufp = &rx_scrq->ind_buf; 585 586 /* netdev_skb_alloc() could have failed after we saved a few skbs 587 * in the indir_buf and we would not have sent them to VIOS yet. 588 * To account for them, start the loop at ind_bufp->index rather 589 * than 0. If we pushed all the skbs to VIOS, ind_bufp->index will 590 * be 0. 591 */ 592 for (i = ind_bufp->index; i < count; ++i) { 593 bufidx = pool->free_map[pool->next_free]; 594 595 /* We maybe reusing the skb from earlier resets. Allocate 596 * only if necessary. But since the LTB may have changed 597 * during reset (see init_rx_pools()), update LTB below 598 * even if reusing skb. 599 */ 600 skb = pool->rx_buff[bufidx].skb; 601 if (!skb) { 602 skb = netdev_alloc_skb(adapter->netdev, 603 pool->buff_size); 604 if (!skb) { 605 dev_err(dev, "Couldn't replenish rx buff\n"); 606 adapter->replenish_no_mem++; 607 break; 608 } 609 } 610 611 pool->free_map[pool->next_free] = IBMVNIC_INVALID_MAP; 612 pool->next_free = (pool->next_free + 1) % pool->size; 613 614 /* Copy the skb to the long term mapped DMA buffer */ 615 map_rxpool_buf_to_ltb(pool, bufidx, <b, &offset); 616 dst = ltb->buff + offset; 617 memset(dst, 0, pool->buff_size); 618 dma_addr = ltb->addr + offset; 619 620 /* add the skb to an rx_buff in the pool */ 621 pool->rx_buff[bufidx].data = dst; 622 pool->rx_buff[bufidx].dma = dma_addr; 623 pool->rx_buff[bufidx].skb = skb; 624 pool->rx_buff[bufidx].pool_index = pool->index; 625 pool->rx_buff[bufidx].size = pool->buff_size; 626 627 /* queue the rx_buff for the next send_subcrq_indirect */ 628 sub_crq = &ind_bufp->indir_arr[ind_bufp->index++]; 629 memset(sub_crq, 0, sizeof(*sub_crq)); 630 sub_crq->rx_add.first = IBMVNIC_CRQ_CMD; 631 sub_crq->rx_add.correlator = 632 cpu_to_be64((u64)&pool->rx_buff[bufidx]); 633 sub_crq->rx_add.ioba = cpu_to_be32(dma_addr); 634 sub_crq->rx_add.map_id = ltb->map_id; 635 636 /* The length field of the sCRQ is defined to be 24 bits so the 637 * buffer size needs to be left shifted by a byte before it is 638 * converted to big endian to prevent the last byte from being 639 * truncated. 640 */ 641 #ifdef __LITTLE_ENDIAN__ 642 shift = 8; 643 #endif 644 sub_crq->rx_add.len = cpu_to_be32(pool->buff_size << shift); 645 646 /* if send_subcrq_indirect queue is full, flush to VIOS */ 647 if (ind_bufp->index == IBMVNIC_MAX_IND_DESCS || 648 i == count - 1) { 649 lpar_rc = 650 send_subcrq_indirect(adapter, handle, 651 (u64)ind_bufp->indir_dma, 652 (u64)ind_bufp->index); 653 if (lpar_rc != H_SUCCESS) 654 goto failure; 655 buffers_added += ind_bufp->index; 656 adapter->replenish_add_buff_success += ind_bufp->index; 657 ind_bufp->index = 0; 658 } 659 } 660 atomic_add(buffers_added, &pool->available); 661 return; 662 663 failure: 664 if (lpar_rc != H_PARAMETER && lpar_rc != H_CLOSED) 665 dev_err_ratelimited(dev, "rx: replenish packet buffer failed\n"); 666 for (i = ind_bufp->index - 1; i >= 0; --i) { 667 struct ibmvnic_rx_buff *rx_buff; 668 669 pool->next_free = pool->next_free == 0 ? 670 pool->size - 1 : pool->next_free - 1; 671 sub_crq = &ind_bufp->indir_arr[i]; 672 rx_buff = (struct ibmvnic_rx_buff *) 673 be64_to_cpu(sub_crq->rx_add.correlator); 674 bufidx = (int)(rx_buff - pool->rx_buff); 675 pool->free_map[pool->next_free] = bufidx; 676 dev_kfree_skb_any(pool->rx_buff[bufidx].skb); 677 pool->rx_buff[bufidx].skb = NULL; 678 } 679 adapter->replenish_add_buff_failure += ind_bufp->index; 680 atomic_add(buffers_added, &pool->available); 681 ind_bufp->index = 0; 682 if (lpar_rc == H_CLOSED || adapter->failover_pending) { 683 /* Disable buffer pool replenishment and report carrier off if 684 * queue is closed or pending failover. 685 * Firmware guarantees that a signal will be sent to the 686 * driver, triggering a reset. 687 */ 688 deactivate_rx_pools(adapter); 689 netif_carrier_off(adapter->netdev); 690 } 691 } 692 693 static void replenish_pools(struct ibmvnic_adapter *adapter) 694 { 695 int i; 696 697 adapter->replenish_task_cycles++; 698 for (i = 0; i < adapter->num_active_rx_pools; i++) { 699 if (adapter->rx_pool[i].active) 700 replenish_rx_pool(adapter, &adapter->rx_pool[i]); 701 } 702 703 netdev_dbg(adapter->netdev, "Replenished %d pools\n", i); 704 } 705 706 static void release_stats_buffers(struct ibmvnic_adapter *adapter) 707 { 708 kfree(adapter->tx_stats_buffers); 709 kfree(adapter->rx_stats_buffers); 710 adapter->tx_stats_buffers = NULL; 711 adapter->rx_stats_buffers = NULL; 712 } 713 714 static int init_stats_buffers(struct ibmvnic_adapter *adapter) 715 { 716 adapter->tx_stats_buffers = 717 kcalloc(IBMVNIC_MAX_QUEUES, 718 sizeof(struct ibmvnic_tx_queue_stats), 719 GFP_KERNEL); 720 if (!adapter->tx_stats_buffers) 721 return -ENOMEM; 722 723 adapter->rx_stats_buffers = 724 kcalloc(IBMVNIC_MAX_QUEUES, 725 sizeof(struct ibmvnic_rx_queue_stats), 726 GFP_KERNEL); 727 if (!adapter->rx_stats_buffers) 728 return -ENOMEM; 729 730 return 0; 731 } 732 733 static void release_stats_token(struct ibmvnic_adapter *adapter) 734 { 735 struct device *dev = &adapter->vdev->dev; 736 737 if (!adapter->stats_token) 738 return; 739 740 dma_unmap_single(dev, adapter->stats_token, 741 sizeof(struct ibmvnic_statistics), 742 DMA_FROM_DEVICE); 743 adapter->stats_token = 0; 744 } 745 746 static int init_stats_token(struct ibmvnic_adapter *adapter) 747 { 748 struct device *dev = &adapter->vdev->dev; 749 dma_addr_t stok; 750 int rc; 751 752 stok = dma_map_single(dev, &adapter->stats, 753 sizeof(struct ibmvnic_statistics), 754 DMA_FROM_DEVICE); 755 rc = dma_mapping_error(dev, stok); 756 if (rc) { 757 dev_err(dev, "Couldn't map stats buffer, rc = %d\n", rc); 758 return rc; 759 } 760 761 adapter->stats_token = stok; 762 netdev_dbg(adapter->netdev, "Stats token initialized (%llx)\n", stok); 763 return 0; 764 } 765 766 /** 767 * release_rx_pools() - Release any rx pools attached to @adapter. 768 * @adapter: ibmvnic adapter 769 * 770 * Safe to call this multiple times - even if no pools are attached. 771 */ 772 static void release_rx_pools(struct ibmvnic_adapter *adapter) 773 { 774 struct ibmvnic_rx_pool *rx_pool; 775 int i, j; 776 777 if (!adapter->rx_pool) 778 return; 779 780 for (i = 0; i < adapter->num_active_rx_pools; i++) { 781 rx_pool = &adapter->rx_pool[i]; 782 783 netdev_dbg(adapter->netdev, "Releasing rx_pool[%d]\n", i); 784 785 kfree(rx_pool->free_map); 786 787 free_ltb_set(adapter, &rx_pool->ltb_set); 788 789 if (!rx_pool->rx_buff) 790 continue; 791 792 for (j = 0; j < rx_pool->size; j++) { 793 if (rx_pool->rx_buff[j].skb) { 794 dev_kfree_skb_any(rx_pool->rx_buff[j].skb); 795 rx_pool->rx_buff[j].skb = NULL; 796 } 797 } 798 799 kfree(rx_pool->rx_buff); 800 } 801 802 kfree(adapter->rx_pool); 803 adapter->rx_pool = NULL; 804 adapter->num_active_rx_pools = 0; 805 adapter->prev_rx_pool_size = 0; 806 } 807 808 /** 809 * reuse_rx_pools() - Check if the existing rx pools can be reused. 810 * @adapter: ibmvnic adapter 811 * 812 * Check if the existing rx pools in the adapter can be reused. The 813 * pools can be reused if the pool parameters (number of pools, 814 * number of buffers in the pool and size of each buffer) have not 815 * changed. 816 * 817 * NOTE: This assumes that all pools have the same number of buffers 818 * which is the case currently. If that changes, we must fix this. 819 * 820 * Return: true if the rx pools can be reused, false otherwise. 821 */ 822 static bool reuse_rx_pools(struct ibmvnic_adapter *adapter) 823 { 824 u64 old_num_pools, new_num_pools; 825 u64 old_pool_size, new_pool_size; 826 u64 old_buff_size, new_buff_size; 827 828 if (!adapter->rx_pool) 829 return false; 830 831 old_num_pools = adapter->num_active_rx_pools; 832 new_num_pools = adapter->req_rx_queues; 833 834 old_pool_size = adapter->prev_rx_pool_size; 835 new_pool_size = adapter->req_rx_add_entries_per_subcrq; 836 837 old_buff_size = adapter->prev_rx_buf_sz; 838 new_buff_size = adapter->cur_rx_buf_sz; 839 840 if (old_buff_size != new_buff_size || 841 old_num_pools != new_num_pools || 842 old_pool_size != new_pool_size) 843 return false; 844 845 return true; 846 } 847 848 /** 849 * init_rx_pools(): Initialize the set of receiver pools in the adapter. 850 * @netdev: net device associated with the vnic interface 851 * 852 * Initialize the set of receiver pools in the ibmvnic adapter associated 853 * with the net_device @netdev. If possible, reuse the existing rx pools. 854 * Otherwise free any existing pools and allocate a new set of pools 855 * before initializing them. 856 * 857 * Return: 0 on success and negative value on error. 858 */ 859 static int init_rx_pools(struct net_device *netdev) 860 { 861 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 862 struct device *dev = &adapter->vdev->dev; 863 struct ibmvnic_rx_pool *rx_pool; 864 u64 num_pools; 865 u64 pool_size; /* # of buffers in one pool */ 866 u64 buff_size; 867 int i, j, rc; 868 869 pool_size = adapter->req_rx_add_entries_per_subcrq; 870 num_pools = adapter->req_rx_queues; 871 buff_size = adapter->cur_rx_buf_sz; 872 873 if (reuse_rx_pools(adapter)) { 874 dev_dbg(dev, "Reusing rx pools\n"); 875 goto update_ltb; 876 } 877 878 /* Allocate/populate the pools. */ 879 release_rx_pools(adapter); 880 881 adapter->rx_pool = kcalloc(num_pools, 882 sizeof(struct ibmvnic_rx_pool), 883 GFP_KERNEL); 884 if (!adapter->rx_pool) { 885 dev_err(dev, "Failed to allocate rx pools\n"); 886 return -ENOMEM; 887 } 888 889 /* Set num_active_rx_pools early. If we fail below after partial 890 * allocation, release_rx_pools() will know how many to look for. 891 */ 892 adapter->num_active_rx_pools = num_pools; 893 894 for (i = 0; i < num_pools; i++) { 895 rx_pool = &adapter->rx_pool[i]; 896 897 netdev_dbg(adapter->netdev, 898 "Initializing rx_pool[%d], %lld buffs, %lld bytes each\n", 899 i, pool_size, buff_size); 900 901 rx_pool->size = pool_size; 902 rx_pool->index = i; 903 rx_pool->buff_size = ALIGN(buff_size, L1_CACHE_BYTES); 904 905 rx_pool->free_map = kcalloc(rx_pool->size, sizeof(int), 906 GFP_KERNEL); 907 if (!rx_pool->free_map) { 908 dev_err(dev, "Couldn't alloc free_map %d\n", i); 909 rc = -ENOMEM; 910 goto out_release; 911 } 912 913 rx_pool->rx_buff = kcalloc(rx_pool->size, 914 sizeof(struct ibmvnic_rx_buff), 915 GFP_KERNEL); 916 if (!rx_pool->rx_buff) { 917 dev_err(dev, "Couldn't alloc rx buffers\n"); 918 rc = -ENOMEM; 919 goto out_release; 920 } 921 } 922 923 adapter->prev_rx_pool_size = pool_size; 924 adapter->prev_rx_buf_sz = adapter->cur_rx_buf_sz; 925 926 update_ltb: 927 for (i = 0; i < num_pools; i++) { 928 rx_pool = &adapter->rx_pool[i]; 929 dev_dbg(dev, "Updating LTB for rx pool %d [%d, %d]\n", 930 i, rx_pool->size, rx_pool->buff_size); 931 932 rc = alloc_ltb_set(adapter, &rx_pool->ltb_set, 933 rx_pool->size, rx_pool->buff_size); 934 if (rc) 935 goto out; 936 937 for (j = 0; j < rx_pool->size; ++j) { 938 struct ibmvnic_rx_buff *rx_buff; 939 940 rx_pool->free_map[j] = j; 941 942 /* NOTE: Don't clear rx_buff->skb here - will leak 943 * memory! replenish_rx_pool() will reuse skbs or 944 * allocate as necessary. 945 */ 946 rx_buff = &rx_pool->rx_buff[j]; 947 rx_buff->dma = 0; 948 rx_buff->data = 0; 949 rx_buff->size = 0; 950 rx_buff->pool_index = 0; 951 } 952 953 /* Mark pool "empty" so replenish_rx_pools() will 954 * update the LTB info for each buffer 955 */ 956 atomic_set(&rx_pool->available, 0); 957 rx_pool->next_alloc = 0; 958 rx_pool->next_free = 0; 959 /* replenish_rx_pool() may have called deactivate_rx_pools() 960 * on failover. Ensure pool is active now. 961 */ 962 rx_pool->active = 1; 963 } 964 return 0; 965 out_release: 966 release_rx_pools(adapter); 967 out: 968 /* We failed to allocate one or more LTBs or map them on the VIOS. 969 * Hold onto the pools and any LTBs that we did allocate/map. 970 */ 971 return rc; 972 } 973 974 static void release_vpd_data(struct ibmvnic_adapter *adapter) 975 { 976 if (!adapter->vpd) 977 return; 978 979 kfree(adapter->vpd->buff); 980 kfree(adapter->vpd); 981 982 adapter->vpd = NULL; 983 } 984 985 static void release_one_tx_pool(struct ibmvnic_adapter *adapter, 986 struct ibmvnic_tx_pool *tx_pool) 987 { 988 kfree(tx_pool->tx_buff); 989 kfree(tx_pool->free_map); 990 free_ltb_set(adapter, &tx_pool->ltb_set); 991 } 992 993 /** 994 * release_tx_pools() - Release any tx pools attached to @adapter. 995 * @adapter: ibmvnic adapter 996 * 997 * Safe to call this multiple times - even if no pools are attached. 998 */ 999 static void release_tx_pools(struct ibmvnic_adapter *adapter) 1000 { 1001 int i; 1002 1003 /* init_tx_pools() ensures that ->tx_pool and ->tso_pool are 1004 * both NULL or both non-NULL. So we only need to check one. 1005 */ 1006 if (!adapter->tx_pool) 1007 return; 1008 1009 for (i = 0; i < adapter->num_active_tx_pools; i++) { 1010 release_one_tx_pool(adapter, &adapter->tx_pool[i]); 1011 release_one_tx_pool(adapter, &adapter->tso_pool[i]); 1012 } 1013 1014 kfree(adapter->tx_pool); 1015 adapter->tx_pool = NULL; 1016 kfree(adapter->tso_pool); 1017 adapter->tso_pool = NULL; 1018 adapter->num_active_tx_pools = 0; 1019 adapter->prev_tx_pool_size = 0; 1020 } 1021 1022 static int init_one_tx_pool(struct net_device *netdev, 1023 struct ibmvnic_tx_pool *tx_pool, 1024 int pool_size, int buf_size) 1025 { 1026 int i; 1027 1028 tx_pool->tx_buff = kcalloc(pool_size, 1029 sizeof(struct ibmvnic_tx_buff), 1030 GFP_KERNEL); 1031 if (!tx_pool->tx_buff) 1032 return -ENOMEM; 1033 1034 tx_pool->free_map = kcalloc(pool_size, sizeof(int), GFP_KERNEL); 1035 if (!tx_pool->free_map) { 1036 kfree(tx_pool->tx_buff); 1037 tx_pool->tx_buff = NULL; 1038 return -ENOMEM; 1039 } 1040 1041 for (i = 0; i < pool_size; i++) 1042 tx_pool->free_map[i] = i; 1043 1044 tx_pool->consumer_index = 0; 1045 tx_pool->producer_index = 0; 1046 tx_pool->num_buffers = pool_size; 1047 tx_pool->buf_size = buf_size; 1048 1049 return 0; 1050 } 1051 1052 /** 1053 * reuse_tx_pools() - Check if the existing tx pools can be reused. 1054 * @adapter: ibmvnic adapter 1055 * 1056 * Check if the existing tx pools in the adapter can be reused. The 1057 * pools can be reused if the pool parameters (number of pools, 1058 * number of buffers in the pool and mtu) have not changed. 1059 * 1060 * NOTE: This assumes that all pools have the same number of buffers 1061 * which is the case currently. If that changes, we must fix this. 1062 * 1063 * Return: true if the tx pools can be reused, false otherwise. 1064 */ 1065 static bool reuse_tx_pools(struct ibmvnic_adapter *adapter) 1066 { 1067 u64 old_num_pools, new_num_pools; 1068 u64 old_pool_size, new_pool_size; 1069 u64 old_mtu, new_mtu; 1070 1071 if (!adapter->tx_pool) 1072 return false; 1073 1074 old_num_pools = adapter->num_active_tx_pools; 1075 new_num_pools = adapter->num_active_tx_scrqs; 1076 old_pool_size = adapter->prev_tx_pool_size; 1077 new_pool_size = adapter->req_tx_entries_per_subcrq; 1078 old_mtu = adapter->prev_mtu; 1079 new_mtu = adapter->req_mtu; 1080 1081 if (old_mtu != new_mtu || 1082 old_num_pools != new_num_pools || 1083 old_pool_size != new_pool_size) 1084 return false; 1085 1086 return true; 1087 } 1088 1089 /** 1090 * init_tx_pools(): Initialize the set of transmit pools in the adapter. 1091 * @netdev: net device associated with the vnic interface 1092 * 1093 * Initialize the set of transmit pools in the ibmvnic adapter associated 1094 * with the net_device @netdev. If possible, reuse the existing tx pools. 1095 * Otherwise free any existing pools and allocate a new set of pools 1096 * before initializing them. 1097 * 1098 * Return: 0 on success and negative value on error. 1099 */ 1100 static int init_tx_pools(struct net_device *netdev) 1101 { 1102 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 1103 struct device *dev = &adapter->vdev->dev; 1104 int num_pools; 1105 u64 pool_size; /* # of buffers in pool */ 1106 u64 buff_size; 1107 int i, j, rc; 1108 1109 num_pools = adapter->req_tx_queues; 1110 1111 /* We must notify the VIOS about the LTB on all resets - but we only 1112 * need to alloc/populate pools if either the number of buffers or 1113 * size of each buffer in the pool has changed. 1114 */ 1115 if (reuse_tx_pools(adapter)) { 1116 netdev_dbg(netdev, "Reusing tx pools\n"); 1117 goto update_ltb; 1118 } 1119 1120 /* Allocate/populate the pools. */ 1121 release_tx_pools(adapter); 1122 1123 pool_size = adapter->req_tx_entries_per_subcrq; 1124 num_pools = adapter->num_active_tx_scrqs; 1125 1126 adapter->tx_pool = kcalloc(num_pools, 1127 sizeof(struct ibmvnic_tx_pool), GFP_KERNEL); 1128 if (!adapter->tx_pool) 1129 return -ENOMEM; 1130 1131 adapter->tso_pool = kcalloc(num_pools, 1132 sizeof(struct ibmvnic_tx_pool), GFP_KERNEL); 1133 /* To simplify release_tx_pools() ensure that ->tx_pool and 1134 * ->tso_pool are either both NULL or both non-NULL. 1135 */ 1136 if (!adapter->tso_pool) { 1137 kfree(adapter->tx_pool); 1138 adapter->tx_pool = NULL; 1139 return -ENOMEM; 1140 } 1141 1142 /* Set num_active_tx_pools early. If we fail below after partial 1143 * allocation, release_tx_pools() will know how many to look for. 1144 */ 1145 adapter->num_active_tx_pools = num_pools; 1146 1147 buff_size = adapter->req_mtu + VLAN_HLEN; 1148 buff_size = ALIGN(buff_size, L1_CACHE_BYTES); 1149 1150 for (i = 0; i < num_pools; i++) { 1151 dev_dbg(dev, "Init tx pool %d [%llu, %llu]\n", 1152 i, adapter->req_tx_entries_per_subcrq, buff_size); 1153 1154 rc = init_one_tx_pool(netdev, &adapter->tx_pool[i], 1155 pool_size, buff_size); 1156 if (rc) 1157 goto out_release; 1158 1159 rc = init_one_tx_pool(netdev, &adapter->tso_pool[i], 1160 IBMVNIC_TSO_BUFS, 1161 IBMVNIC_TSO_BUF_SZ); 1162 if (rc) 1163 goto out_release; 1164 } 1165 1166 adapter->prev_tx_pool_size = pool_size; 1167 adapter->prev_mtu = adapter->req_mtu; 1168 1169 update_ltb: 1170 /* NOTE: All tx_pools have the same number of buffers (which is 1171 * same as pool_size). All tso_pools have IBMVNIC_TSO_BUFS 1172 * buffers (see calls init_one_tx_pool() for these). 1173 * For consistency, we use tx_pool->num_buffers and 1174 * tso_pool->num_buffers below. 1175 */ 1176 rc = -1; 1177 for (i = 0; i < num_pools; i++) { 1178 struct ibmvnic_tx_pool *tso_pool; 1179 struct ibmvnic_tx_pool *tx_pool; 1180 1181 tx_pool = &adapter->tx_pool[i]; 1182 1183 dev_dbg(dev, "Updating LTB for tx pool %d [%d, %d]\n", 1184 i, tx_pool->num_buffers, tx_pool->buf_size); 1185 1186 rc = alloc_ltb_set(adapter, &tx_pool->ltb_set, 1187 tx_pool->num_buffers, tx_pool->buf_size); 1188 if (rc) 1189 goto out; 1190 1191 tx_pool->consumer_index = 0; 1192 tx_pool->producer_index = 0; 1193 1194 for (j = 0; j < tx_pool->num_buffers; j++) 1195 tx_pool->free_map[j] = j; 1196 1197 tso_pool = &adapter->tso_pool[i]; 1198 1199 dev_dbg(dev, "Updating LTB for tso pool %d [%d, %d]\n", 1200 i, tso_pool->num_buffers, tso_pool->buf_size); 1201 1202 rc = alloc_ltb_set(adapter, &tso_pool->ltb_set, 1203 tso_pool->num_buffers, tso_pool->buf_size); 1204 if (rc) 1205 goto out; 1206 1207 tso_pool->consumer_index = 0; 1208 tso_pool->producer_index = 0; 1209 1210 for (j = 0; j < tso_pool->num_buffers; j++) 1211 tso_pool->free_map[j] = j; 1212 } 1213 1214 return 0; 1215 out_release: 1216 release_tx_pools(adapter); 1217 out: 1218 /* We failed to allocate one or more LTBs or map them on the VIOS. 1219 * Hold onto the pools and any LTBs that we did allocate/map. 1220 */ 1221 return rc; 1222 } 1223 1224 static void ibmvnic_napi_enable(struct ibmvnic_adapter *adapter) 1225 { 1226 int i; 1227 1228 if (adapter->napi_enabled) 1229 return; 1230 1231 for (i = 0; i < adapter->req_rx_queues; i++) 1232 napi_enable(&adapter->napi[i]); 1233 1234 adapter->napi_enabled = true; 1235 } 1236 1237 static void ibmvnic_napi_disable(struct ibmvnic_adapter *adapter) 1238 { 1239 int i; 1240 1241 if (!adapter->napi_enabled) 1242 return; 1243 1244 for (i = 0; i < adapter->req_rx_queues; i++) { 1245 netdev_dbg(adapter->netdev, "Disabling napi[%d]\n", i); 1246 napi_disable(&adapter->napi[i]); 1247 } 1248 1249 adapter->napi_enabled = false; 1250 } 1251 1252 static int init_napi(struct ibmvnic_adapter *adapter) 1253 { 1254 int i; 1255 1256 adapter->napi = kcalloc(adapter->req_rx_queues, 1257 sizeof(struct napi_struct), GFP_KERNEL); 1258 if (!adapter->napi) 1259 return -ENOMEM; 1260 1261 for (i = 0; i < adapter->req_rx_queues; i++) { 1262 netdev_dbg(adapter->netdev, "Adding napi[%d]\n", i); 1263 netif_napi_add(adapter->netdev, &adapter->napi[i], 1264 ibmvnic_poll, NAPI_POLL_WEIGHT); 1265 } 1266 1267 adapter->num_active_rx_napi = adapter->req_rx_queues; 1268 return 0; 1269 } 1270 1271 static void release_napi(struct ibmvnic_adapter *adapter) 1272 { 1273 int i; 1274 1275 if (!adapter->napi) 1276 return; 1277 1278 for (i = 0; i < adapter->num_active_rx_napi; i++) { 1279 netdev_dbg(adapter->netdev, "Releasing napi[%d]\n", i); 1280 netif_napi_del(&adapter->napi[i]); 1281 } 1282 1283 kfree(adapter->napi); 1284 adapter->napi = NULL; 1285 adapter->num_active_rx_napi = 0; 1286 adapter->napi_enabled = false; 1287 } 1288 1289 static const char *adapter_state_to_string(enum vnic_state state) 1290 { 1291 switch (state) { 1292 case VNIC_PROBING: 1293 return "PROBING"; 1294 case VNIC_PROBED: 1295 return "PROBED"; 1296 case VNIC_OPENING: 1297 return "OPENING"; 1298 case VNIC_OPEN: 1299 return "OPEN"; 1300 case VNIC_CLOSING: 1301 return "CLOSING"; 1302 case VNIC_CLOSED: 1303 return "CLOSED"; 1304 case VNIC_REMOVING: 1305 return "REMOVING"; 1306 case VNIC_REMOVED: 1307 return "REMOVED"; 1308 case VNIC_DOWN: 1309 return "DOWN"; 1310 } 1311 return "UNKNOWN"; 1312 } 1313 1314 static int ibmvnic_login(struct net_device *netdev) 1315 { 1316 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 1317 unsigned long timeout = msecs_to_jiffies(20000); 1318 int retry_count = 0; 1319 int retries = 10; 1320 bool retry; 1321 int rc; 1322 1323 do { 1324 retry = false; 1325 if (retry_count > retries) { 1326 netdev_warn(netdev, "Login attempts exceeded\n"); 1327 return -EACCES; 1328 } 1329 1330 adapter->init_done_rc = 0; 1331 reinit_completion(&adapter->init_done); 1332 rc = send_login(adapter); 1333 if (rc) 1334 return rc; 1335 1336 if (!wait_for_completion_timeout(&adapter->init_done, 1337 timeout)) { 1338 netdev_warn(netdev, "Login timed out, retrying...\n"); 1339 retry = true; 1340 adapter->init_done_rc = 0; 1341 retry_count++; 1342 continue; 1343 } 1344 1345 if (adapter->init_done_rc == ABORTED) { 1346 netdev_warn(netdev, "Login aborted, retrying...\n"); 1347 retry = true; 1348 adapter->init_done_rc = 0; 1349 retry_count++; 1350 /* FW or device may be busy, so 1351 * wait a bit before retrying login 1352 */ 1353 msleep(500); 1354 } else if (adapter->init_done_rc == PARTIALSUCCESS) { 1355 retry_count++; 1356 release_sub_crqs(adapter, 1); 1357 1358 retry = true; 1359 netdev_dbg(netdev, 1360 "Received partial success, retrying...\n"); 1361 adapter->init_done_rc = 0; 1362 reinit_completion(&adapter->init_done); 1363 send_query_cap(adapter); 1364 if (!wait_for_completion_timeout(&adapter->init_done, 1365 timeout)) { 1366 netdev_warn(netdev, 1367 "Capabilities query timed out\n"); 1368 return -ETIMEDOUT; 1369 } 1370 1371 rc = init_sub_crqs(adapter); 1372 if (rc) { 1373 netdev_warn(netdev, 1374 "SCRQ initialization failed\n"); 1375 return rc; 1376 } 1377 1378 rc = init_sub_crq_irqs(adapter); 1379 if (rc) { 1380 netdev_warn(netdev, 1381 "SCRQ irq initialization failed\n"); 1382 return rc; 1383 } 1384 } else if (adapter->init_done_rc) { 1385 netdev_warn(netdev, "Adapter login failed, init_done_rc = %d\n", 1386 adapter->init_done_rc); 1387 return -EIO; 1388 } 1389 } while (retry); 1390 1391 __ibmvnic_set_mac(netdev, adapter->mac_addr); 1392 1393 netdev_dbg(netdev, "[S:%s] Login succeeded\n", adapter_state_to_string(adapter->state)); 1394 return 0; 1395 } 1396 1397 static void release_login_buffer(struct ibmvnic_adapter *adapter) 1398 { 1399 kfree(adapter->login_buf); 1400 adapter->login_buf = NULL; 1401 } 1402 1403 static void release_login_rsp_buffer(struct ibmvnic_adapter *adapter) 1404 { 1405 kfree(adapter->login_rsp_buf); 1406 adapter->login_rsp_buf = NULL; 1407 } 1408 1409 static void release_resources(struct ibmvnic_adapter *adapter) 1410 { 1411 release_vpd_data(adapter); 1412 1413 release_napi(adapter); 1414 release_login_buffer(adapter); 1415 release_login_rsp_buffer(adapter); 1416 } 1417 1418 static int set_link_state(struct ibmvnic_adapter *adapter, u8 link_state) 1419 { 1420 struct net_device *netdev = adapter->netdev; 1421 unsigned long timeout = msecs_to_jiffies(20000); 1422 union ibmvnic_crq crq; 1423 bool resend; 1424 int rc; 1425 1426 netdev_dbg(netdev, "setting link state %d\n", link_state); 1427 1428 memset(&crq, 0, sizeof(crq)); 1429 crq.logical_link_state.first = IBMVNIC_CRQ_CMD; 1430 crq.logical_link_state.cmd = LOGICAL_LINK_STATE; 1431 crq.logical_link_state.link_state = link_state; 1432 1433 do { 1434 resend = false; 1435 1436 reinit_completion(&adapter->init_done); 1437 rc = ibmvnic_send_crq(adapter, &crq); 1438 if (rc) { 1439 netdev_err(netdev, "Failed to set link state\n"); 1440 return rc; 1441 } 1442 1443 if (!wait_for_completion_timeout(&adapter->init_done, 1444 timeout)) { 1445 netdev_err(netdev, "timeout setting link state\n"); 1446 return -ETIMEDOUT; 1447 } 1448 1449 if (adapter->init_done_rc == PARTIALSUCCESS) { 1450 /* Partuial success, delay and re-send */ 1451 mdelay(1000); 1452 resend = true; 1453 } else if (adapter->init_done_rc) { 1454 netdev_warn(netdev, "Unable to set link state, rc=%d\n", 1455 adapter->init_done_rc); 1456 return adapter->init_done_rc; 1457 } 1458 } while (resend); 1459 1460 return 0; 1461 } 1462 1463 static int set_real_num_queues(struct net_device *netdev) 1464 { 1465 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 1466 int rc; 1467 1468 netdev_dbg(netdev, "Setting real tx/rx queues (%llx/%llx)\n", 1469 adapter->req_tx_queues, adapter->req_rx_queues); 1470 1471 rc = netif_set_real_num_tx_queues(netdev, adapter->req_tx_queues); 1472 if (rc) { 1473 netdev_err(netdev, "failed to set the number of tx queues\n"); 1474 return rc; 1475 } 1476 1477 rc = netif_set_real_num_rx_queues(netdev, adapter->req_rx_queues); 1478 if (rc) 1479 netdev_err(netdev, "failed to set the number of rx queues\n"); 1480 1481 return rc; 1482 } 1483 1484 static int ibmvnic_get_vpd(struct ibmvnic_adapter *adapter) 1485 { 1486 struct device *dev = &adapter->vdev->dev; 1487 union ibmvnic_crq crq; 1488 int len = 0; 1489 int rc; 1490 1491 if (adapter->vpd->buff) 1492 len = adapter->vpd->len; 1493 1494 mutex_lock(&adapter->fw_lock); 1495 adapter->fw_done_rc = 0; 1496 reinit_completion(&adapter->fw_done); 1497 1498 crq.get_vpd_size.first = IBMVNIC_CRQ_CMD; 1499 crq.get_vpd_size.cmd = GET_VPD_SIZE; 1500 rc = ibmvnic_send_crq(adapter, &crq); 1501 if (rc) { 1502 mutex_unlock(&adapter->fw_lock); 1503 return rc; 1504 } 1505 1506 rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000); 1507 if (rc) { 1508 dev_err(dev, "Could not retrieve VPD size, rc = %d\n", rc); 1509 mutex_unlock(&adapter->fw_lock); 1510 return rc; 1511 } 1512 mutex_unlock(&adapter->fw_lock); 1513 1514 if (!adapter->vpd->len) 1515 return -ENODATA; 1516 1517 if (!adapter->vpd->buff) 1518 adapter->vpd->buff = kzalloc(adapter->vpd->len, GFP_KERNEL); 1519 else if (adapter->vpd->len != len) 1520 adapter->vpd->buff = 1521 krealloc(adapter->vpd->buff, 1522 adapter->vpd->len, GFP_KERNEL); 1523 1524 if (!adapter->vpd->buff) { 1525 dev_err(dev, "Could allocate VPD buffer\n"); 1526 return -ENOMEM; 1527 } 1528 1529 adapter->vpd->dma_addr = 1530 dma_map_single(dev, adapter->vpd->buff, adapter->vpd->len, 1531 DMA_FROM_DEVICE); 1532 if (dma_mapping_error(dev, adapter->vpd->dma_addr)) { 1533 dev_err(dev, "Could not map VPD buffer\n"); 1534 kfree(adapter->vpd->buff); 1535 adapter->vpd->buff = NULL; 1536 return -ENOMEM; 1537 } 1538 1539 mutex_lock(&adapter->fw_lock); 1540 adapter->fw_done_rc = 0; 1541 reinit_completion(&adapter->fw_done); 1542 1543 crq.get_vpd.first = IBMVNIC_CRQ_CMD; 1544 crq.get_vpd.cmd = GET_VPD; 1545 crq.get_vpd.ioba = cpu_to_be32(adapter->vpd->dma_addr); 1546 crq.get_vpd.len = cpu_to_be32((u32)adapter->vpd->len); 1547 rc = ibmvnic_send_crq(adapter, &crq); 1548 if (rc) { 1549 kfree(adapter->vpd->buff); 1550 adapter->vpd->buff = NULL; 1551 mutex_unlock(&adapter->fw_lock); 1552 return rc; 1553 } 1554 1555 rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000); 1556 if (rc) { 1557 dev_err(dev, "Unable to retrieve VPD, rc = %d\n", rc); 1558 kfree(adapter->vpd->buff); 1559 adapter->vpd->buff = NULL; 1560 mutex_unlock(&adapter->fw_lock); 1561 return rc; 1562 } 1563 1564 mutex_unlock(&adapter->fw_lock); 1565 return 0; 1566 } 1567 1568 static int init_resources(struct ibmvnic_adapter *adapter) 1569 { 1570 struct net_device *netdev = adapter->netdev; 1571 int rc; 1572 1573 rc = set_real_num_queues(netdev); 1574 if (rc) 1575 return rc; 1576 1577 adapter->vpd = kzalloc(sizeof(*adapter->vpd), GFP_KERNEL); 1578 if (!adapter->vpd) 1579 return -ENOMEM; 1580 1581 /* Vital Product Data (VPD) */ 1582 rc = ibmvnic_get_vpd(adapter); 1583 if (rc) { 1584 netdev_err(netdev, "failed to initialize Vital Product Data (VPD)\n"); 1585 return rc; 1586 } 1587 1588 rc = init_napi(adapter); 1589 if (rc) 1590 return rc; 1591 1592 send_query_map(adapter); 1593 1594 rc = init_rx_pools(netdev); 1595 if (rc) 1596 return rc; 1597 1598 rc = init_tx_pools(netdev); 1599 return rc; 1600 } 1601 1602 static int __ibmvnic_open(struct net_device *netdev) 1603 { 1604 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 1605 enum vnic_state prev_state = adapter->state; 1606 int i, rc; 1607 1608 adapter->state = VNIC_OPENING; 1609 replenish_pools(adapter); 1610 ibmvnic_napi_enable(adapter); 1611 1612 /* We're ready to receive frames, enable the sub-crq interrupts and 1613 * set the logical link state to up 1614 */ 1615 for (i = 0; i < adapter->req_rx_queues; i++) { 1616 netdev_dbg(netdev, "Enabling rx_scrq[%d] irq\n", i); 1617 if (prev_state == VNIC_CLOSED) 1618 enable_irq(adapter->rx_scrq[i]->irq); 1619 enable_scrq_irq(adapter, adapter->rx_scrq[i]); 1620 } 1621 1622 for (i = 0; i < adapter->req_tx_queues; i++) { 1623 netdev_dbg(netdev, "Enabling tx_scrq[%d] irq\n", i); 1624 if (prev_state == VNIC_CLOSED) 1625 enable_irq(adapter->tx_scrq[i]->irq); 1626 enable_scrq_irq(adapter, adapter->tx_scrq[i]); 1627 netdev_tx_reset_queue(netdev_get_tx_queue(netdev, i)); 1628 } 1629 1630 rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_UP); 1631 if (rc) { 1632 ibmvnic_napi_disable(adapter); 1633 ibmvnic_disable_irqs(adapter); 1634 return rc; 1635 } 1636 1637 adapter->tx_queues_active = true; 1638 1639 /* Since queues were stopped until now, there shouldn't be any 1640 * one in ibmvnic_complete_tx() or ibmvnic_xmit() so maybe we 1641 * don't need the synchronize_rcu()? Leaving it for consistency 1642 * with setting ->tx_queues_active = false. 1643 */ 1644 synchronize_rcu(); 1645 1646 netif_tx_start_all_queues(netdev); 1647 1648 if (prev_state == VNIC_CLOSED) { 1649 for (i = 0; i < adapter->req_rx_queues; i++) 1650 napi_schedule(&adapter->napi[i]); 1651 } 1652 1653 adapter->state = VNIC_OPEN; 1654 return rc; 1655 } 1656 1657 static int ibmvnic_open(struct net_device *netdev) 1658 { 1659 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 1660 int rc; 1661 1662 ASSERT_RTNL(); 1663 1664 /* If device failover is pending or we are about to reset, just set 1665 * device state and return. Device operation will be handled by reset 1666 * routine. 1667 * 1668 * It should be safe to overwrite the adapter->state here. Since 1669 * we hold the rtnl, either the reset has not actually started or 1670 * the rtnl got dropped during the set_link_state() in do_reset(). 1671 * In the former case, no one else is changing the state (again we 1672 * have the rtnl) and in the latter case, do_reset() will detect and 1673 * honor our setting below. 1674 */ 1675 if (adapter->failover_pending || (test_bit(0, &adapter->resetting))) { 1676 netdev_dbg(netdev, "[S:%s FOP:%d] Resetting, deferring open\n", 1677 adapter_state_to_string(adapter->state), 1678 adapter->failover_pending); 1679 adapter->state = VNIC_OPEN; 1680 rc = 0; 1681 goto out; 1682 } 1683 1684 if (adapter->state != VNIC_CLOSED) { 1685 rc = ibmvnic_login(netdev); 1686 if (rc) 1687 goto out; 1688 1689 rc = init_resources(adapter); 1690 if (rc) { 1691 netdev_err(netdev, "failed to initialize resources\n"); 1692 goto out; 1693 } 1694 } 1695 1696 rc = __ibmvnic_open(netdev); 1697 1698 out: 1699 /* If open failed and there is a pending failover or in-progress reset, 1700 * set device state and return. Device operation will be handled by 1701 * reset routine. See also comments above regarding rtnl. 1702 */ 1703 if (rc && 1704 (adapter->failover_pending || (test_bit(0, &adapter->resetting)))) { 1705 adapter->state = VNIC_OPEN; 1706 rc = 0; 1707 } 1708 1709 if (rc) { 1710 release_resources(adapter); 1711 release_rx_pools(adapter); 1712 release_tx_pools(adapter); 1713 } 1714 1715 return rc; 1716 } 1717 1718 static void clean_rx_pools(struct ibmvnic_adapter *adapter) 1719 { 1720 struct ibmvnic_rx_pool *rx_pool; 1721 struct ibmvnic_rx_buff *rx_buff; 1722 u64 rx_entries; 1723 int rx_scrqs; 1724 int i, j; 1725 1726 if (!adapter->rx_pool) 1727 return; 1728 1729 rx_scrqs = adapter->num_active_rx_pools; 1730 rx_entries = adapter->req_rx_add_entries_per_subcrq; 1731 1732 /* Free any remaining skbs in the rx buffer pools */ 1733 for (i = 0; i < rx_scrqs; i++) { 1734 rx_pool = &adapter->rx_pool[i]; 1735 if (!rx_pool || !rx_pool->rx_buff) 1736 continue; 1737 1738 netdev_dbg(adapter->netdev, "Cleaning rx_pool[%d]\n", i); 1739 for (j = 0; j < rx_entries; j++) { 1740 rx_buff = &rx_pool->rx_buff[j]; 1741 if (rx_buff && rx_buff->skb) { 1742 dev_kfree_skb_any(rx_buff->skb); 1743 rx_buff->skb = NULL; 1744 } 1745 } 1746 } 1747 } 1748 1749 static void clean_one_tx_pool(struct ibmvnic_adapter *adapter, 1750 struct ibmvnic_tx_pool *tx_pool) 1751 { 1752 struct ibmvnic_tx_buff *tx_buff; 1753 u64 tx_entries; 1754 int i; 1755 1756 if (!tx_pool || !tx_pool->tx_buff) 1757 return; 1758 1759 tx_entries = tx_pool->num_buffers; 1760 1761 for (i = 0; i < tx_entries; i++) { 1762 tx_buff = &tx_pool->tx_buff[i]; 1763 if (tx_buff && tx_buff->skb) { 1764 dev_kfree_skb_any(tx_buff->skb); 1765 tx_buff->skb = NULL; 1766 } 1767 } 1768 } 1769 1770 static void clean_tx_pools(struct ibmvnic_adapter *adapter) 1771 { 1772 int tx_scrqs; 1773 int i; 1774 1775 if (!adapter->tx_pool || !adapter->tso_pool) 1776 return; 1777 1778 tx_scrqs = adapter->num_active_tx_pools; 1779 1780 /* Free any remaining skbs in the tx buffer pools */ 1781 for (i = 0; i < tx_scrqs; i++) { 1782 netdev_dbg(adapter->netdev, "Cleaning tx_pool[%d]\n", i); 1783 clean_one_tx_pool(adapter, &adapter->tx_pool[i]); 1784 clean_one_tx_pool(adapter, &adapter->tso_pool[i]); 1785 } 1786 } 1787 1788 static void ibmvnic_disable_irqs(struct ibmvnic_adapter *adapter) 1789 { 1790 struct net_device *netdev = adapter->netdev; 1791 int i; 1792 1793 if (adapter->tx_scrq) { 1794 for (i = 0; i < adapter->req_tx_queues; i++) 1795 if (adapter->tx_scrq[i]->irq) { 1796 netdev_dbg(netdev, 1797 "Disabling tx_scrq[%d] irq\n", i); 1798 disable_scrq_irq(adapter, adapter->tx_scrq[i]); 1799 disable_irq(adapter->tx_scrq[i]->irq); 1800 } 1801 } 1802 1803 if (adapter->rx_scrq) { 1804 for (i = 0; i < adapter->req_rx_queues; i++) { 1805 if (adapter->rx_scrq[i]->irq) { 1806 netdev_dbg(netdev, 1807 "Disabling rx_scrq[%d] irq\n", i); 1808 disable_scrq_irq(adapter, adapter->rx_scrq[i]); 1809 disable_irq(adapter->rx_scrq[i]->irq); 1810 } 1811 } 1812 } 1813 } 1814 1815 static void ibmvnic_cleanup(struct net_device *netdev) 1816 { 1817 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 1818 1819 /* ensure that transmissions are stopped if called by do_reset */ 1820 1821 adapter->tx_queues_active = false; 1822 1823 /* Ensure complete_tx() and ibmvnic_xmit() see ->tx_queues_active 1824 * update so they don't restart a queue after we stop it below. 1825 */ 1826 synchronize_rcu(); 1827 1828 if (test_bit(0, &adapter->resetting)) 1829 netif_tx_disable(netdev); 1830 else 1831 netif_tx_stop_all_queues(netdev); 1832 1833 ibmvnic_napi_disable(adapter); 1834 ibmvnic_disable_irqs(adapter); 1835 } 1836 1837 static int __ibmvnic_close(struct net_device *netdev) 1838 { 1839 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 1840 int rc = 0; 1841 1842 adapter->state = VNIC_CLOSING; 1843 rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN); 1844 adapter->state = VNIC_CLOSED; 1845 return rc; 1846 } 1847 1848 static int ibmvnic_close(struct net_device *netdev) 1849 { 1850 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 1851 int rc; 1852 1853 netdev_dbg(netdev, "[S:%s FOP:%d FRR:%d] Closing\n", 1854 adapter_state_to_string(adapter->state), 1855 adapter->failover_pending, 1856 adapter->force_reset_recovery); 1857 1858 /* If device failover is pending, just set device state and return. 1859 * Device operation will be handled by reset routine. 1860 */ 1861 if (adapter->failover_pending) { 1862 adapter->state = VNIC_CLOSED; 1863 return 0; 1864 } 1865 1866 rc = __ibmvnic_close(netdev); 1867 ibmvnic_cleanup(netdev); 1868 clean_rx_pools(adapter); 1869 clean_tx_pools(adapter); 1870 1871 return rc; 1872 } 1873 1874 /** 1875 * build_hdr_data - creates L2/L3/L4 header data buffer 1876 * @hdr_field: bitfield determining needed headers 1877 * @skb: socket buffer 1878 * @hdr_len: array of header lengths 1879 * @hdr_data: buffer to write the header to 1880 * 1881 * Reads hdr_field to determine which headers are needed by firmware. 1882 * Builds a buffer containing these headers. Saves individual header 1883 * lengths and total buffer length to be used to build descriptors. 1884 */ 1885 static int build_hdr_data(u8 hdr_field, struct sk_buff *skb, 1886 int *hdr_len, u8 *hdr_data) 1887 { 1888 int len = 0; 1889 u8 *hdr; 1890 1891 if (skb_vlan_tagged(skb) && !skb_vlan_tag_present(skb)) 1892 hdr_len[0] = sizeof(struct vlan_ethhdr); 1893 else 1894 hdr_len[0] = sizeof(struct ethhdr); 1895 1896 if (skb->protocol == htons(ETH_P_IP)) { 1897 hdr_len[1] = ip_hdr(skb)->ihl * 4; 1898 if (ip_hdr(skb)->protocol == IPPROTO_TCP) 1899 hdr_len[2] = tcp_hdrlen(skb); 1900 else if (ip_hdr(skb)->protocol == IPPROTO_UDP) 1901 hdr_len[2] = sizeof(struct udphdr); 1902 } else if (skb->protocol == htons(ETH_P_IPV6)) { 1903 hdr_len[1] = sizeof(struct ipv6hdr); 1904 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP) 1905 hdr_len[2] = tcp_hdrlen(skb); 1906 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP) 1907 hdr_len[2] = sizeof(struct udphdr); 1908 } else if (skb->protocol == htons(ETH_P_ARP)) { 1909 hdr_len[1] = arp_hdr_len(skb->dev); 1910 hdr_len[2] = 0; 1911 } 1912 1913 memset(hdr_data, 0, 120); 1914 if ((hdr_field >> 6) & 1) { 1915 hdr = skb_mac_header(skb); 1916 memcpy(hdr_data, hdr, hdr_len[0]); 1917 len += hdr_len[0]; 1918 } 1919 1920 if ((hdr_field >> 5) & 1) { 1921 hdr = skb_network_header(skb); 1922 memcpy(hdr_data + len, hdr, hdr_len[1]); 1923 len += hdr_len[1]; 1924 } 1925 1926 if ((hdr_field >> 4) & 1) { 1927 hdr = skb_transport_header(skb); 1928 memcpy(hdr_data + len, hdr, hdr_len[2]); 1929 len += hdr_len[2]; 1930 } 1931 return len; 1932 } 1933 1934 /** 1935 * create_hdr_descs - create header and header extension descriptors 1936 * @hdr_field: bitfield determining needed headers 1937 * @hdr_data: buffer containing header data 1938 * @len: length of data buffer 1939 * @hdr_len: array of individual header lengths 1940 * @scrq_arr: descriptor array 1941 * 1942 * Creates header and, if needed, header extension descriptors and 1943 * places them in a descriptor array, scrq_arr 1944 */ 1945 1946 static int create_hdr_descs(u8 hdr_field, u8 *hdr_data, int len, int *hdr_len, 1947 union sub_crq *scrq_arr) 1948 { 1949 union sub_crq hdr_desc; 1950 int tmp_len = len; 1951 int num_descs = 0; 1952 u8 *data, *cur; 1953 int tmp; 1954 1955 while (tmp_len > 0) { 1956 cur = hdr_data + len - tmp_len; 1957 1958 memset(&hdr_desc, 0, sizeof(hdr_desc)); 1959 if (cur != hdr_data) { 1960 data = hdr_desc.hdr_ext.data; 1961 tmp = tmp_len > 29 ? 29 : tmp_len; 1962 hdr_desc.hdr_ext.first = IBMVNIC_CRQ_CMD; 1963 hdr_desc.hdr_ext.type = IBMVNIC_HDR_EXT_DESC; 1964 hdr_desc.hdr_ext.len = tmp; 1965 } else { 1966 data = hdr_desc.hdr.data; 1967 tmp = tmp_len > 24 ? 24 : tmp_len; 1968 hdr_desc.hdr.first = IBMVNIC_CRQ_CMD; 1969 hdr_desc.hdr.type = IBMVNIC_HDR_DESC; 1970 hdr_desc.hdr.len = tmp; 1971 hdr_desc.hdr.l2_len = (u8)hdr_len[0]; 1972 hdr_desc.hdr.l3_len = cpu_to_be16((u16)hdr_len[1]); 1973 hdr_desc.hdr.l4_len = (u8)hdr_len[2]; 1974 hdr_desc.hdr.flag = hdr_field << 1; 1975 } 1976 memcpy(data, cur, tmp); 1977 tmp_len -= tmp; 1978 *scrq_arr = hdr_desc; 1979 scrq_arr++; 1980 num_descs++; 1981 } 1982 1983 return num_descs; 1984 } 1985 1986 /** 1987 * build_hdr_descs_arr - build a header descriptor array 1988 * @skb: tx socket buffer 1989 * @indir_arr: indirect array 1990 * @num_entries: number of descriptors to be sent 1991 * @hdr_field: bit field determining which headers will be sent 1992 * 1993 * This function will build a TX descriptor array with applicable 1994 * L2/L3/L4 packet header descriptors to be sent by send_subcrq_indirect. 1995 */ 1996 1997 static void build_hdr_descs_arr(struct sk_buff *skb, 1998 union sub_crq *indir_arr, 1999 int *num_entries, u8 hdr_field) 2000 { 2001 int hdr_len[3] = {0, 0, 0}; 2002 u8 hdr_data[140] = {0}; 2003 int tot_len; 2004 2005 tot_len = build_hdr_data(hdr_field, skb, hdr_len, 2006 hdr_data); 2007 *num_entries += create_hdr_descs(hdr_field, hdr_data, tot_len, hdr_len, 2008 indir_arr + 1); 2009 } 2010 2011 static int ibmvnic_xmit_workarounds(struct sk_buff *skb, 2012 struct net_device *netdev) 2013 { 2014 /* For some backing devices, mishandling of small packets 2015 * can result in a loss of connection or TX stall. Device 2016 * architects recommend that no packet should be smaller 2017 * than the minimum MTU value provided to the driver, so 2018 * pad any packets to that length 2019 */ 2020 if (skb->len < netdev->min_mtu) 2021 return skb_put_padto(skb, netdev->min_mtu); 2022 2023 return 0; 2024 } 2025 2026 static void ibmvnic_tx_scrq_clean_buffer(struct ibmvnic_adapter *adapter, 2027 struct ibmvnic_sub_crq_queue *tx_scrq) 2028 { 2029 struct ibmvnic_ind_xmit_queue *ind_bufp; 2030 struct ibmvnic_tx_buff *tx_buff; 2031 struct ibmvnic_tx_pool *tx_pool; 2032 union sub_crq tx_scrq_entry; 2033 int queue_num; 2034 int entries; 2035 int index; 2036 int i; 2037 2038 ind_bufp = &tx_scrq->ind_buf; 2039 entries = (u64)ind_bufp->index; 2040 queue_num = tx_scrq->pool_index; 2041 2042 for (i = entries - 1; i >= 0; --i) { 2043 tx_scrq_entry = ind_bufp->indir_arr[i]; 2044 if (tx_scrq_entry.v1.type != IBMVNIC_TX_DESC) 2045 continue; 2046 index = be32_to_cpu(tx_scrq_entry.v1.correlator); 2047 if (index & IBMVNIC_TSO_POOL_MASK) { 2048 tx_pool = &adapter->tso_pool[queue_num]; 2049 index &= ~IBMVNIC_TSO_POOL_MASK; 2050 } else { 2051 tx_pool = &adapter->tx_pool[queue_num]; 2052 } 2053 tx_pool->free_map[tx_pool->consumer_index] = index; 2054 tx_pool->consumer_index = tx_pool->consumer_index == 0 ? 2055 tx_pool->num_buffers - 1 : 2056 tx_pool->consumer_index - 1; 2057 tx_buff = &tx_pool->tx_buff[index]; 2058 adapter->netdev->stats.tx_packets--; 2059 adapter->netdev->stats.tx_bytes -= tx_buff->skb->len; 2060 adapter->tx_stats_buffers[queue_num].packets--; 2061 adapter->tx_stats_buffers[queue_num].bytes -= 2062 tx_buff->skb->len; 2063 dev_kfree_skb_any(tx_buff->skb); 2064 tx_buff->skb = NULL; 2065 adapter->netdev->stats.tx_dropped++; 2066 } 2067 2068 ind_bufp->index = 0; 2069 2070 if (atomic_sub_return(entries, &tx_scrq->used) <= 2071 (adapter->req_tx_entries_per_subcrq / 2) && 2072 __netif_subqueue_stopped(adapter->netdev, queue_num)) { 2073 rcu_read_lock(); 2074 2075 if (adapter->tx_queues_active) { 2076 netif_wake_subqueue(adapter->netdev, queue_num); 2077 netdev_dbg(adapter->netdev, "Started queue %d\n", 2078 queue_num); 2079 } 2080 2081 rcu_read_unlock(); 2082 } 2083 } 2084 2085 static int ibmvnic_tx_scrq_flush(struct ibmvnic_adapter *adapter, 2086 struct ibmvnic_sub_crq_queue *tx_scrq) 2087 { 2088 struct ibmvnic_ind_xmit_queue *ind_bufp; 2089 u64 dma_addr; 2090 u64 entries; 2091 u64 handle; 2092 int rc; 2093 2094 ind_bufp = &tx_scrq->ind_buf; 2095 dma_addr = (u64)ind_bufp->indir_dma; 2096 entries = (u64)ind_bufp->index; 2097 handle = tx_scrq->handle; 2098 2099 if (!entries) 2100 return 0; 2101 rc = send_subcrq_indirect(adapter, handle, dma_addr, entries); 2102 if (rc) 2103 ibmvnic_tx_scrq_clean_buffer(adapter, tx_scrq); 2104 else 2105 ind_bufp->index = 0; 2106 return 0; 2107 } 2108 2109 static netdev_tx_t ibmvnic_xmit(struct sk_buff *skb, struct net_device *netdev) 2110 { 2111 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 2112 int queue_num = skb_get_queue_mapping(skb); 2113 u8 *hdrs = (u8 *)&adapter->tx_rx_desc_req; 2114 struct device *dev = &adapter->vdev->dev; 2115 struct ibmvnic_ind_xmit_queue *ind_bufp; 2116 struct ibmvnic_tx_buff *tx_buff = NULL; 2117 struct ibmvnic_sub_crq_queue *tx_scrq; 2118 struct ibmvnic_long_term_buff *ltb; 2119 struct ibmvnic_tx_pool *tx_pool; 2120 unsigned int tx_send_failed = 0; 2121 netdev_tx_t ret = NETDEV_TX_OK; 2122 unsigned int tx_map_failed = 0; 2123 union sub_crq indir_arr[16]; 2124 unsigned int tx_dropped = 0; 2125 unsigned int tx_packets = 0; 2126 unsigned int tx_bytes = 0; 2127 dma_addr_t data_dma_addr; 2128 struct netdev_queue *txq; 2129 unsigned long lpar_rc; 2130 union sub_crq tx_crq; 2131 unsigned int offset; 2132 int num_entries = 1; 2133 unsigned char *dst; 2134 int bufidx = 0; 2135 u8 proto = 0; 2136 2137 /* If a reset is in progress, drop the packet since 2138 * the scrqs may get torn down. Otherwise use the 2139 * rcu to ensure reset waits for us to complete. 2140 */ 2141 rcu_read_lock(); 2142 if (!adapter->tx_queues_active) { 2143 dev_kfree_skb_any(skb); 2144 2145 tx_send_failed++; 2146 tx_dropped++; 2147 ret = NETDEV_TX_OK; 2148 goto out; 2149 } 2150 2151 tx_scrq = adapter->tx_scrq[queue_num]; 2152 txq = netdev_get_tx_queue(netdev, queue_num); 2153 ind_bufp = &tx_scrq->ind_buf; 2154 2155 if (ibmvnic_xmit_workarounds(skb, netdev)) { 2156 tx_dropped++; 2157 tx_send_failed++; 2158 ret = NETDEV_TX_OK; 2159 ibmvnic_tx_scrq_flush(adapter, tx_scrq); 2160 goto out; 2161 } 2162 2163 if (skb_is_gso(skb)) 2164 tx_pool = &adapter->tso_pool[queue_num]; 2165 else 2166 tx_pool = &adapter->tx_pool[queue_num]; 2167 2168 bufidx = tx_pool->free_map[tx_pool->consumer_index]; 2169 2170 if (bufidx == IBMVNIC_INVALID_MAP) { 2171 dev_kfree_skb_any(skb); 2172 tx_send_failed++; 2173 tx_dropped++; 2174 ibmvnic_tx_scrq_flush(adapter, tx_scrq); 2175 ret = NETDEV_TX_OK; 2176 goto out; 2177 } 2178 2179 tx_pool->free_map[tx_pool->consumer_index] = IBMVNIC_INVALID_MAP; 2180 2181 map_txpool_buf_to_ltb(tx_pool, bufidx, <b, &offset); 2182 2183 dst = ltb->buff + offset; 2184 memset(dst, 0, tx_pool->buf_size); 2185 data_dma_addr = ltb->addr + offset; 2186 2187 if (skb_shinfo(skb)->nr_frags) { 2188 int cur, i; 2189 2190 /* Copy the head */ 2191 skb_copy_from_linear_data(skb, dst, skb_headlen(skb)); 2192 cur = skb_headlen(skb); 2193 2194 /* Copy the frags */ 2195 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 2196 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 2197 2198 memcpy(dst + cur, skb_frag_address(frag), 2199 skb_frag_size(frag)); 2200 cur += skb_frag_size(frag); 2201 } 2202 } else { 2203 skb_copy_from_linear_data(skb, dst, skb->len); 2204 } 2205 2206 /* post changes to long_term_buff *dst before VIOS accessing it */ 2207 dma_wmb(); 2208 2209 tx_pool->consumer_index = 2210 (tx_pool->consumer_index + 1) % tx_pool->num_buffers; 2211 2212 tx_buff = &tx_pool->tx_buff[bufidx]; 2213 tx_buff->skb = skb; 2214 tx_buff->index = bufidx; 2215 tx_buff->pool_index = queue_num; 2216 2217 memset(&tx_crq, 0, sizeof(tx_crq)); 2218 tx_crq.v1.first = IBMVNIC_CRQ_CMD; 2219 tx_crq.v1.type = IBMVNIC_TX_DESC; 2220 tx_crq.v1.n_crq_elem = 1; 2221 tx_crq.v1.n_sge = 1; 2222 tx_crq.v1.flags1 = IBMVNIC_TX_COMP_NEEDED; 2223 2224 if (skb_is_gso(skb)) 2225 tx_crq.v1.correlator = 2226 cpu_to_be32(bufidx | IBMVNIC_TSO_POOL_MASK); 2227 else 2228 tx_crq.v1.correlator = cpu_to_be32(bufidx); 2229 tx_crq.v1.dma_reg = cpu_to_be16(ltb->map_id); 2230 tx_crq.v1.sge_len = cpu_to_be32(skb->len); 2231 tx_crq.v1.ioba = cpu_to_be64(data_dma_addr); 2232 2233 if (adapter->vlan_header_insertion && skb_vlan_tag_present(skb)) { 2234 tx_crq.v1.flags2 |= IBMVNIC_TX_VLAN_INSERT; 2235 tx_crq.v1.vlan_id = cpu_to_be16(skb->vlan_tci); 2236 } 2237 2238 if (skb->protocol == htons(ETH_P_IP)) { 2239 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV4; 2240 proto = ip_hdr(skb)->protocol; 2241 } else if (skb->protocol == htons(ETH_P_IPV6)) { 2242 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV6; 2243 proto = ipv6_hdr(skb)->nexthdr; 2244 } 2245 2246 if (proto == IPPROTO_TCP) 2247 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_TCP; 2248 else if (proto == IPPROTO_UDP) 2249 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_UDP; 2250 2251 if (skb->ip_summed == CHECKSUM_PARTIAL) { 2252 tx_crq.v1.flags1 |= IBMVNIC_TX_CHKSUM_OFFLOAD; 2253 hdrs += 2; 2254 } 2255 if (skb_is_gso(skb)) { 2256 tx_crq.v1.flags1 |= IBMVNIC_TX_LSO; 2257 tx_crq.v1.mss = cpu_to_be16(skb_shinfo(skb)->gso_size); 2258 hdrs += 2; 2259 } 2260 2261 if ((*hdrs >> 7) & 1) 2262 build_hdr_descs_arr(skb, indir_arr, &num_entries, *hdrs); 2263 2264 tx_crq.v1.n_crq_elem = num_entries; 2265 tx_buff->num_entries = num_entries; 2266 /* flush buffer if current entry can not fit */ 2267 if (num_entries + ind_bufp->index > IBMVNIC_MAX_IND_DESCS) { 2268 lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq); 2269 if (lpar_rc != H_SUCCESS) 2270 goto tx_flush_err; 2271 } 2272 2273 indir_arr[0] = tx_crq; 2274 memcpy(&ind_bufp->indir_arr[ind_bufp->index], &indir_arr[0], 2275 num_entries * sizeof(struct ibmvnic_generic_scrq)); 2276 ind_bufp->index += num_entries; 2277 if (__netdev_tx_sent_queue(txq, skb->len, 2278 netdev_xmit_more() && 2279 ind_bufp->index < IBMVNIC_MAX_IND_DESCS)) { 2280 lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq); 2281 if (lpar_rc != H_SUCCESS) 2282 goto tx_err; 2283 } 2284 2285 if (atomic_add_return(num_entries, &tx_scrq->used) 2286 >= adapter->req_tx_entries_per_subcrq) { 2287 netdev_dbg(netdev, "Stopping queue %d\n", queue_num); 2288 netif_stop_subqueue(netdev, queue_num); 2289 } 2290 2291 tx_packets++; 2292 tx_bytes += skb->len; 2293 txq_trans_cond_update(txq); 2294 ret = NETDEV_TX_OK; 2295 goto out; 2296 2297 tx_flush_err: 2298 dev_kfree_skb_any(skb); 2299 tx_buff->skb = NULL; 2300 tx_pool->consumer_index = tx_pool->consumer_index == 0 ? 2301 tx_pool->num_buffers - 1 : 2302 tx_pool->consumer_index - 1; 2303 tx_dropped++; 2304 tx_err: 2305 if (lpar_rc != H_CLOSED && lpar_rc != H_PARAMETER) 2306 dev_err_ratelimited(dev, "tx: send failed\n"); 2307 2308 if (lpar_rc == H_CLOSED || adapter->failover_pending) { 2309 /* Disable TX and report carrier off if queue is closed 2310 * or pending failover. 2311 * Firmware guarantees that a signal will be sent to the 2312 * driver, triggering a reset or some other action. 2313 */ 2314 netif_tx_stop_all_queues(netdev); 2315 netif_carrier_off(netdev); 2316 } 2317 out: 2318 rcu_read_unlock(); 2319 netdev->stats.tx_dropped += tx_dropped; 2320 netdev->stats.tx_bytes += tx_bytes; 2321 netdev->stats.tx_packets += tx_packets; 2322 adapter->tx_send_failed += tx_send_failed; 2323 adapter->tx_map_failed += tx_map_failed; 2324 adapter->tx_stats_buffers[queue_num].packets += tx_packets; 2325 adapter->tx_stats_buffers[queue_num].bytes += tx_bytes; 2326 adapter->tx_stats_buffers[queue_num].dropped_packets += tx_dropped; 2327 2328 return ret; 2329 } 2330 2331 static void ibmvnic_set_multi(struct net_device *netdev) 2332 { 2333 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 2334 struct netdev_hw_addr *ha; 2335 union ibmvnic_crq crq; 2336 2337 memset(&crq, 0, sizeof(crq)); 2338 crq.request_capability.first = IBMVNIC_CRQ_CMD; 2339 crq.request_capability.cmd = REQUEST_CAPABILITY; 2340 2341 if (netdev->flags & IFF_PROMISC) { 2342 if (!adapter->promisc_supported) 2343 return; 2344 } else { 2345 if (netdev->flags & IFF_ALLMULTI) { 2346 /* Accept all multicast */ 2347 memset(&crq, 0, sizeof(crq)); 2348 crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD; 2349 crq.multicast_ctrl.cmd = MULTICAST_CTRL; 2350 crq.multicast_ctrl.flags = IBMVNIC_ENABLE_ALL; 2351 ibmvnic_send_crq(adapter, &crq); 2352 } else if (netdev_mc_empty(netdev)) { 2353 /* Reject all multicast */ 2354 memset(&crq, 0, sizeof(crq)); 2355 crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD; 2356 crq.multicast_ctrl.cmd = MULTICAST_CTRL; 2357 crq.multicast_ctrl.flags = IBMVNIC_DISABLE_ALL; 2358 ibmvnic_send_crq(adapter, &crq); 2359 } else { 2360 /* Accept one or more multicast(s) */ 2361 netdev_for_each_mc_addr(ha, netdev) { 2362 memset(&crq, 0, sizeof(crq)); 2363 crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD; 2364 crq.multicast_ctrl.cmd = MULTICAST_CTRL; 2365 crq.multicast_ctrl.flags = IBMVNIC_ENABLE_MC; 2366 ether_addr_copy(&crq.multicast_ctrl.mac_addr[0], 2367 ha->addr); 2368 ibmvnic_send_crq(adapter, &crq); 2369 } 2370 } 2371 } 2372 } 2373 2374 static int __ibmvnic_set_mac(struct net_device *netdev, u8 *dev_addr) 2375 { 2376 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 2377 union ibmvnic_crq crq; 2378 int rc; 2379 2380 if (!is_valid_ether_addr(dev_addr)) { 2381 rc = -EADDRNOTAVAIL; 2382 goto err; 2383 } 2384 2385 memset(&crq, 0, sizeof(crq)); 2386 crq.change_mac_addr.first = IBMVNIC_CRQ_CMD; 2387 crq.change_mac_addr.cmd = CHANGE_MAC_ADDR; 2388 ether_addr_copy(&crq.change_mac_addr.mac_addr[0], dev_addr); 2389 2390 mutex_lock(&adapter->fw_lock); 2391 adapter->fw_done_rc = 0; 2392 reinit_completion(&adapter->fw_done); 2393 2394 rc = ibmvnic_send_crq(adapter, &crq); 2395 if (rc) { 2396 rc = -EIO; 2397 mutex_unlock(&adapter->fw_lock); 2398 goto err; 2399 } 2400 2401 rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000); 2402 /* netdev->dev_addr is changed in handle_change_mac_rsp function */ 2403 if (rc || adapter->fw_done_rc) { 2404 rc = -EIO; 2405 mutex_unlock(&adapter->fw_lock); 2406 goto err; 2407 } 2408 mutex_unlock(&adapter->fw_lock); 2409 return 0; 2410 err: 2411 ether_addr_copy(adapter->mac_addr, netdev->dev_addr); 2412 return rc; 2413 } 2414 2415 static int ibmvnic_set_mac(struct net_device *netdev, void *p) 2416 { 2417 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 2418 struct sockaddr *addr = p; 2419 int rc; 2420 2421 rc = 0; 2422 if (!is_valid_ether_addr(addr->sa_data)) 2423 return -EADDRNOTAVAIL; 2424 2425 ether_addr_copy(adapter->mac_addr, addr->sa_data); 2426 if (adapter->state != VNIC_PROBED) 2427 rc = __ibmvnic_set_mac(netdev, addr->sa_data); 2428 2429 return rc; 2430 } 2431 2432 static const char *reset_reason_to_string(enum ibmvnic_reset_reason reason) 2433 { 2434 switch (reason) { 2435 case VNIC_RESET_FAILOVER: 2436 return "FAILOVER"; 2437 case VNIC_RESET_MOBILITY: 2438 return "MOBILITY"; 2439 case VNIC_RESET_FATAL: 2440 return "FATAL"; 2441 case VNIC_RESET_NON_FATAL: 2442 return "NON_FATAL"; 2443 case VNIC_RESET_TIMEOUT: 2444 return "TIMEOUT"; 2445 case VNIC_RESET_CHANGE_PARAM: 2446 return "CHANGE_PARAM"; 2447 case VNIC_RESET_PASSIVE_INIT: 2448 return "PASSIVE_INIT"; 2449 } 2450 return "UNKNOWN"; 2451 } 2452 2453 /* 2454 * Initialize the init_done completion and return code values. We 2455 * can get a transport event just after registering the CRQ and the 2456 * tasklet will use this to communicate the transport event. To ensure 2457 * we don't miss the notification/error, initialize these _before_ 2458 * regisering the CRQ. 2459 */ 2460 static inline void reinit_init_done(struct ibmvnic_adapter *adapter) 2461 { 2462 reinit_completion(&adapter->init_done); 2463 adapter->init_done_rc = 0; 2464 } 2465 2466 /* 2467 * do_reset returns zero if we are able to keep processing reset events, or 2468 * non-zero if we hit a fatal error and must halt. 2469 */ 2470 static int do_reset(struct ibmvnic_adapter *adapter, 2471 struct ibmvnic_rwi *rwi, u32 reset_state) 2472 { 2473 struct net_device *netdev = adapter->netdev; 2474 u64 old_num_rx_queues, old_num_tx_queues; 2475 u64 old_num_rx_slots, old_num_tx_slots; 2476 int rc; 2477 2478 netdev_dbg(adapter->netdev, 2479 "[S:%s FOP:%d] Reset reason: %s, reset_state: %s\n", 2480 adapter_state_to_string(adapter->state), 2481 adapter->failover_pending, 2482 reset_reason_to_string(rwi->reset_reason), 2483 adapter_state_to_string(reset_state)); 2484 2485 adapter->reset_reason = rwi->reset_reason; 2486 /* requestor of VNIC_RESET_CHANGE_PARAM already has the rtnl lock */ 2487 if (!(adapter->reset_reason == VNIC_RESET_CHANGE_PARAM)) 2488 rtnl_lock(); 2489 2490 /* Now that we have the rtnl lock, clear any pending failover. 2491 * This will ensure ibmvnic_open() has either completed or will 2492 * block until failover is complete. 2493 */ 2494 if (rwi->reset_reason == VNIC_RESET_FAILOVER) 2495 adapter->failover_pending = false; 2496 2497 /* read the state and check (again) after getting rtnl */ 2498 reset_state = adapter->state; 2499 2500 if (reset_state == VNIC_REMOVING || reset_state == VNIC_REMOVED) { 2501 rc = -EBUSY; 2502 goto out; 2503 } 2504 2505 netif_carrier_off(netdev); 2506 2507 old_num_rx_queues = adapter->req_rx_queues; 2508 old_num_tx_queues = adapter->req_tx_queues; 2509 old_num_rx_slots = adapter->req_rx_add_entries_per_subcrq; 2510 old_num_tx_slots = adapter->req_tx_entries_per_subcrq; 2511 2512 ibmvnic_cleanup(netdev); 2513 2514 if (reset_state == VNIC_OPEN && 2515 adapter->reset_reason != VNIC_RESET_MOBILITY && 2516 adapter->reset_reason != VNIC_RESET_FAILOVER) { 2517 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) { 2518 rc = __ibmvnic_close(netdev); 2519 if (rc) 2520 goto out; 2521 } else { 2522 adapter->state = VNIC_CLOSING; 2523 2524 /* Release the RTNL lock before link state change and 2525 * re-acquire after the link state change to allow 2526 * linkwatch_event to grab the RTNL lock and run during 2527 * a reset. 2528 */ 2529 rtnl_unlock(); 2530 rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN); 2531 rtnl_lock(); 2532 if (rc) 2533 goto out; 2534 2535 if (adapter->state == VNIC_OPEN) { 2536 /* When we dropped rtnl, ibmvnic_open() got 2537 * it and noticed that we are resetting and 2538 * set the adapter state to OPEN. Update our 2539 * new "target" state, and resume the reset 2540 * from VNIC_CLOSING state. 2541 */ 2542 netdev_dbg(netdev, 2543 "Open changed state from %s, updating.\n", 2544 adapter_state_to_string(reset_state)); 2545 reset_state = VNIC_OPEN; 2546 adapter->state = VNIC_CLOSING; 2547 } 2548 2549 if (adapter->state != VNIC_CLOSING) { 2550 /* If someone else changed the adapter state 2551 * when we dropped the rtnl, fail the reset 2552 */ 2553 rc = -EAGAIN; 2554 goto out; 2555 } 2556 adapter->state = VNIC_CLOSED; 2557 } 2558 } 2559 2560 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) { 2561 release_resources(adapter); 2562 release_sub_crqs(adapter, 1); 2563 release_crq_queue(adapter); 2564 } 2565 2566 if (adapter->reset_reason != VNIC_RESET_NON_FATAL) { 2567 /* remove the closed state so when we call open it appears 2568 * we are coming from the probed state. 2569 */ 2570 adapter->state = VNIC_PROBED; 2571 2572 reinit_init_done(adapter); 2573 2574 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) { 2575 rc = init_crq_queue(adapter); 2576 } else if (adapter->reset_reason == VNIC_RESET_MOBILITY) { 2577 rc = ibmvnic_reenable_crq_queue(adapter); 2578 release_sub_crqs(adapter, 1); 2579 } else { 2580 rc = ibmvnic_reset_crq(adapter); 2581 if (rc == H_CLOSED || rc == H_SUCCESS) { 2582 rc = vio_enable_interrupts(adapter->vdev); 2583 if (rc) 2584 netdev_err(adapter->netdev, 2585 "Reset failed to enable interrupts. rc=%d\n", 2586 rc); 2587 } 2588 } 2589 2590 if (rc) { 2591 netdev_err(adapter->netdev, 2592 "Reset couldn't initialize crq. rc=%d\n", rc); 2593 goto out; 2594 } 2595 2596 rc = ibmvnic_reset_init(adapter, true); 2597 if (rc) 2598 goto out; 2599 2600 /* If the adapter was in PROBE or DOWN state prior to the reset, 2601 * exit here. 2602 */ 2603 if (reset_state == VNIC_PROBED || reset_state == VNIC_DOWN) { 2604 rc = 0; 2605 goto out; 2606 } 2607 2608 rc = ibmvnic_login(netdev); 2609 if (rc) 2610 goto out; 2611 2612 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) { 2613 rc = init_resources(adapter); 2614 if (rc) 2615 goto out; 2616 } else if (adapter->req_rx_queues != old_num_rx_queues || 2617 adapter->req_tx_queues != old_num_tx_queues || 2618 adapter->req_rx_add_entries_per_subcrq != 2619 old_num_rx_slots || 2620 adapter->req_tx_entries_per_subcrq != 2621 old_num_tx_slots || 2622 !adapter->rx_pool || 2623 !adapter->tso_pool || 2624 !adapter->tx_pool) { 2625 release_napi(adapter); 2626 release_vpd_data(adapter); 2627 2628 rc = init_resources(adapter); 2629 if (rc) 2630 goto out; 2631 2632 } else { 2633 rc = init_tx_pools(netdev); 2634 if (rc) { 2635 netdev_dbg(netdev, 2636 "init tx pools failed (%d)\n", 2637 rc); 2638 goto out; 2639 } 2640 2641 rc = init_rx_pools(netdev); 2642 if (rc) { 2643 netdev_dbg(netdev, 2644 "init rx pools failed (%d)\n", 2645 rc); 2646 goto out; 2647 } 2648 } 2649 ibmvnic_disable_irqs(adapter); 2650 } 2651 adapter->state = VNIC_CLOSED; 2652 2653 if (reset_state == VNIC_CLOSED) { 2654 rc = 0; 2655 goto out; 2656 } 2657 2658 rc = __ibmvnic_open(netdev); 2659 if (rc) { 2660 rc = IBMVNIC_OPEN_FAILED; 2661 goto out; 2662 } 2663 2664 /* refresh device's multicast list */ 2665 ibmvnic_set_multi(netdev); 2666 2667 if (adapter->reset_reason == VNIC_RESET_FAILOVER || 2668 adapter->reset_reason == VNIC_RESET_MOBILITY) 2669 __netdev_notify_peers(netdev); 2670 2671 rc = 0; 2672 2673 out: 2674 /* restore the adapter state if reset failed */ 2675 if (rc) 2676 adapter->state = reset_state; 2677 /* requestor of VNIC_RESET_CHANGE_PARAM should still hold the rtnl lock */ 2678 if (!(adapter->reset_reason == VNIC_RESET_CHANGE_PARAM)) 2679 rtnl_unlock(); 2680 2681 netdev_dbg(adapter->netdev, "[S:%s FOP:%d] Reset done, rc %d\n", 2682 adapter_state_to_string(adapter->state), 2683 adapter->failover_pending, rc); 2684 return rc; 2685 } 2686 2687 static int do_hard_reset(struct ibmvnic_adapter *adapter, 2688 struct ibmvnic_rwi *rwi, u32 reset_state) 2689 { 2690 struct net_device *netdev = adapter->netdev; 2691 int rc; 2692 2693 netdev_dbg(adapter->netdev, "Hard resetting driver (%s)\n", 2694 reset_reason_to_string(rwi->reset_reason)); 2695 2696 /* read the state and check (again) after getting rtnl */ 2697 reset_state = adapter->state; 2698 2699 if (reset_state == VNIC_REMOVING || reset_state == VNIC_REMOVED) { 2700 rc = -EBUSY; 2701 goto out; 2702 } 2703 2704 netif_carrier_off(netdev); 2705 adapter->reset_reason = rwi->reset_reason; 2706 2707 ibmvnic_cleanup(netdev); 2708 release_resources(adapter); 2709 release_sub_crqs(adapter, 0); 2710 release_crq_queue(adapter); 2711 2712 /* remove the closed state so when we call open it appears 2713 * we are coming from the probed state. 2714 */ 2715 adapter->state = VNIC_PROBED; 2716 2717 reinit_init_done(adapter); 2718 2719 rc = init_crq_queue(adapter); 2720 if (rc) { 2721 netdev_err(adapter->netdev, 2722 "Couldn't initialize crq. rc=%d\n", rc); 2723 goto out; 2724 } 2725 2726 rc = ibmvnic_reset_init(adapter, false); 2727 if (rc) 2728 goto out; 2729 2730 /* If the adapter was in PROBE or DOWN state prior to the reset, 2731 * exit here. 2732 */ 2733 if (reset_state == VNIC_PROBED || reset_state == VNIC_DOWN) 2734 goto out; 2735 2736 rc = ibmvnic_login(netdev); 2737 if (rc) 2738 goto out; 2739 2740 rc = init_resources(adapter); 2741 if (rc) 2742 goto out; 2743 2744 ibmvnic_disable_irqs(adapter); 2745 adapter->state = VNIC_CLOSED; 2746 2747 if (reset_state == VNIC_CLOSED) 2748 goto out; 2749 2750 rc = __ibmvnic_open(netdev); 2751 if (rc) { 2752 rc = IBMVNIC_OPEN_FAILED; 2753 goto out; 2754 } 2755 2756 __netdev_notify_peers(netdev); 2757 out: 2758 /* restore adapter state if reset failed */ 2759 if (rc) 2760 adapter->state = reset_state; 2761 netdev_dbg(adapter->netdev, "[S:%s FOP:%d] Hard reset done, rc %d\n", 2762 adapter_state_to_string(adapter->state), 2763 adapter->failover_pending, rc); 2764 return rc; 2765 } 2766 2767 static struct ibmvnic_rwi *get_next_rwi(struct ibmvnic_adapter *adapter) 2768 { 2769 struct ibmvnic_rwi *rwi; 2770 unsigned long flags; 2771 2772 spin_lock_irqsave(&adapter->rwi_lock, flags); 2773 2774 if (!list_empty(&adapter->rwi_list)) { 2775 rwi = list_first_entry(&adapter->rwi_list, struct ibmvnic_rwi, 2776 list); 2777 list_del(&rwi->list); 2778 } else { 2779 rwi = NULL; 2780 } 2781 2782 spin_unlock_irqrestore(&adapter->rwi_lock, flags); 2783 return rwi; 2784 } 2785 2786 /** 2787 * do_passive_init - complete probing when partner device is detected. 2788 * @adapter: ibmvnic_adapter struct 2789 * 2790 * If the ibmvnic device does not have a partner device to communicate with at boot 2791 * and that partner device comes online at a later time, this function is called 2792 * to complete the initialization process of ibmvnic device. 2793 * Caller is expected to hold rtnl_lock(). 2794 * 2795 * Returns non-zero if sub-CRQs are not initialized properly leaving the device 2796 * in the down state. 2797 * Returns 0 upon success and the device is in PROBED state. 2798 */ 2799 2800 static int do_passive_init(struct ibmvnic_adapter *adapter) 2801 { 2802 unsigned long timeout = msecs_to_jiffies(30000); 2803 struct net_device *netdev = adapter->netdev; 2804 struct device *dev = &adapter->vdev->dev; 2805 int rc; 2806 2807 netdev_dbg(netdev, "Partner device found, probing.\n"); 2808 2809 adapter->state = VNIC_PROBING; 2810 reinit_completion(&adapter->init_done); 2811 adapter->init_done_rc = 0; 2812 adapter->crq.active = true; 2813 2814 rc = send_crq_init_complete(adapter); 2815 if (rc) 2816 goto out; 2817 2818 rc = send_version_xchg(adapter); 2819 if (rc) 2820 netdev_dbg(adapter->netdev, "send_version_xchg failed, rc=%d\n", rc); 2821 2822 if (!wait_for_completion_timeout(&adapter->init_done, timeout)) { 2823 dev_err(dev, "Initialization sequence timed out\n"); 2824 rc = -ETIMEDOUT; 2825 goto out; 2826 } 2827 2828 rc = init_sub_crqs(adapter); 2829 if (rc) { 2830 dev_err(dev, "Initialization of sub crqs failed, rc=%d\n", rc); 2831 goto out; 2832 } 2833 2834 rc = init_sub_crq_irqs(adapter); 2835 if (rc) { 2836 dev_err(dev, "Failed to initialize sub crq irqs\n, rc=%d", rc); 2837 goto init_failed; 2838 } 2839 2840 netdev->mtu = adapter->req_mtu - ETH_HLEN; 2841 netdev->min_mtu = adapter->min_mtu - ETH_HLEN; 2842 netdev->max_mtu = adapter->max_mtu - ETH_HLEN; 2843 2844 adapter->state = VNIC_PROBED; 2845 netdev_dbg(netdev, "Probed successfully. Waiting for signal from partner device.\n"); 2846 2847 return 0; 2848 2849 init_failed: 2850 release_sub_crqs(adapter, 1); 2851 out: 2852 adapter->state = VNIC_DOWN; 2853 return rc; 2854 } 2855 2856 static void __ibmvnic_reset(struct work_struct *work) 2857 { 2858 struct ibmvnic_adapter *adapter; 2859 unsigned int timeout = 5000; 2860 struct ibmvnic_rwi *tmprwi; 2861 bool saved_state = false; 2862 struct ibmvnic_rwi *rwi; 2863 unsigned long flags; 2864 struct device *dev; 2865 bool need_reset; 2866 int num_fails = 0; 2867 u32 reset_state; 2868 int rc = 0; 2869 2870 adapter = container_of(work, struct ibmvnic_adapter, ibmvnic_reset); 2871 dev = &adapter->vdev->dev; 2872 2873 /* Wait for ibmvnic_probe() to complete. If probe is taking too long 2874 * or if another reset is in progress, defer work for now. If probe 2875 * eventually fails it will flush and terminate our work. 2876 * 2877 * Three possibilities here: 2878 * 1. Adpater being removed - just return 2879 * 2. Timed out on probe or another reset in progress - delay the work 2880 * 3. Completed probe - perform any resets in queue 2881 */ 2882 if (adapter->state == VNIC_PROBING && 2883 !wait_for_completion_timeout(&adapter->probe_done, timeout)) { 2884 dev_err(dev, "Reset thread timed out on probe"); 2885 queue_delayed_work(system_long_wq, 2886 &adapter->ibmvnic_delayed_reset, 2887 IBMVNIC_RESET_DELAY); 2888 return; 2889 } 2890 2891 /* adapter is done with probe (i.e state is never VNIC_PROBING now) */ 2892 if (adapter->state == VNIC_REMOVING) 2893 return; 2894 2895 /* ->rwi_list is stable now (no one else is removing entries) */ 2896 2897 /* ibmvnic_probe() may have purged the reset queue after we were 2898 * scheduled to process a reset so there maybe no resets to process. 2899 * Before setting the ->resetting bit though, we have to make sure 2900 * that there is infact a reset to process. Otherwise we may race 2901 * with ibmvnic_open() and end up leaving the vnic down: 2902 * 2903 * __ibmvnic_reset() ibmvnic_open() 2904 * ----------------- -------------- 2905 * 2906 * set ->resetting bit 2907 * find ->resetting bit is set 2908 * set ->state to IBMVNIC_OPEN (i.e 2909 * assume reset will open device) 2910 * return 2911 * find reset queue empty 2912 * return 2913 * 2914 * Neither performed vnic login/open and vnic stays down 2915 * 2916 * If we hold the lock and conditionally set the bit, either we 2917 * or ibmvnic_open() will complete the open. 2918 */ 2919 need_reset = false; 2920 spin_lock(&adapter->rwi_lock); 2921 if (!list_empty(&adapter->rwi_list)) { 2922 if (test_and_set_bit_lock(0, &adapter->resetting)) { 2923 queue_delayed_work(system_long_wq, 2924 &adapter->ibmvnic_delayed_reset, 2925 IBMVNIC_RESET_DELAY); 2926 } else { 2927 need_reset = true; 2928 } 2929 } 2930 spin_unlock(&adapter->rwi_lock); 2931 2932 if (!need_reset) 2933 return; 2934 2935 rwi = get_next_rwi(adapter); 2936 while (rwi) { 2937 spin_lock_irqsave(&adapter->state_lock, flags); 2938 2939 if (adapter->state == VNIC_REMOVING || 2940 adapter->state == VNIC_REMOVED) { 2941 spin_unlock_irqrestore(&adapter->state_lock, flags); 2942 kfree(rwi); 2943 rc = EBUSY; 2944 break; 2945 } 2946 2947 if (!saved_state) { 2948 reset_state = adapter->state; 2949 saved_state = true; 2950 } 2951 spin_unlock_irqrestore(&adapter->state_lock, flags); 2952 2953 if (rwi->reset_reason == VNIC_RESET_PASSIVE_INIT) { 2954 rtnl_lock(); 2955 rc = do_passive_init(adapter); 2956 rtnl_unlock(); 2957 if (!rc) 2958 netif_carrier_on(adapter->netdev); 2959 } else if (adapter->force_reset_recovery) { 2960 /* Since we are doing a hard reset now, clear the 2961 * failover_pending flag so we don't ignore any 2962 * future MOBILITY or other resets. 2963 */ 2964 adapter->failover_pending = false; 2965 2966 /* Transport event occurred during previous reset */ 2967 if (adapter->wait_for_reset) { 2968 /* Previous was CHANGE_PARAM; caller locked */ 2969 adapter->force_reset_recovery = false; 2970 rc = do_hard_reset(adapter, rwi, reset_state); 2971 } else { 2972 rtnl_lock(); 2973 adapter->force_reset_recovery = false; 2974 rc = do_hard_reset(adapter, rwi, reset_state); 2975 rtnl_unlock(); 2976 } 2977 if (rc) 2978 num_fails++; 2979 else 2980 num_fails = 0; 2981 2982 /* If auto-priority-failover is enabled we can get 2983 * back to back failovers during resets, resulting 2984 * in at least two failed resets (from high-priority 2985 * backing device to low-priority one and then back) 2986 * If resets continue to fail beyond that, give the 2987 * adapter some time to settle down before retrying. 2988 */ 2989 if (num_fails >= 3) { 2990 netdev_dbg(adapter->netdev, 2991 "[S:%s] Hard reset failed %d times, waiting 60 secs\n", 2992 adapter_state_to_string(adapter->state), 2993 num_fails); 2994 set_current_state(TASK_UNINTERRUPTIBLE); 2995 schedule_timeout(60 * HZ); 2996 } 2997 } else { 2998 rc = do_reset(adapter, rwi, reset_state); 2999 } 3000 tmprwi = rwi; 3001 adapter->last_reset_time = jiffies; 3002 3003 if (rc) 3004 netdev_dbg(adapter->netdev, "Reset failed, rc=%d\n", rc); 3005 3006 rwi = get_next_rwi(adapter); 3007 3008 /* 3009 * If there is another reset queued, free the previous rwi 3010 * and process the new reset even if previous reset failed 3011 * (the previous reset could have failed because of a fail 3012 * over for instance, so process the fail over). 3013 * 3014 * If there are no resets queued and the previous reset failed, 3015 * the adapter would be in an undefined state. So retry the 3016 * previous reset as a hard reset. 3017 */ 3018 if (rwi) 3019 kfree(tmprwi); 3020 else if (rc) 3021 rwi = tmprwi; 3022 3023 if (rwi && (rwi->reset_reason == VNIC_RESET_FAILOVER || 3024 rwi->reset_reason == VNIC_RESET_MOBILITY || rc)) 3025 adapter->force_reset_recovery = true; 3026 } 3027 3028 if (adapter->wait_for_reset) { 3029 adapter->reset_done_rc = rc; 3030 complete(&adapter->reset_done); 3031 } 3032 3033 clear_bit_unlock(0, &adapter->resetting); 3034 3035 netdev_dbg(adapter->netdev, 3036 "[S:%s FRR:%d WFR:%d] Done processing resets\n", 3037 adapter_state_to_string(adapter->state), 3038 adapter->force_reset_recovery, 3039 adapter->wait_for_reset); 3040 } 3041 3042 static void __ibmvnic_delayed_reset(struct work_struct *work) 3043 { 3044 struct ibmvnic_adapter *adapter; 3045 3046 adapter = container_of(work, struct ibmvnic_adapter, 3047 ibmvnic_delayed_reset.work); 3048 __ibmvnic_reset(&adapter->ibmvnic_reset); 3049 } 3050 3051 static void flush_reset_queue(struct ibmvnic_adapter *adapter) 3052 { 3053 struct list_head *entry, *tmp_entry; 3054 3055 if (!list_empty(&adapter->rwi_list)) { 3056 list_for_each_safe(entry, tmp_entry, &adapter->rwi_list) { 3057 list_del(entry); 3058 kfree(list_entry(entry, struct ibmvnic_rwi, list)); 3059 } 3060 } 3061 } 3062 3063 static int ibmvnic_reset(struct ibmvnic_adapter *adapter, 3064 enum ibmvnic_reset_reason reason) 3065 { 3066 struct net_device *netdev = adapter->netdev; 3067 struct ibmvnic_rwi *rwi, *tmp; 3068 unsigned long flags; 3069 int ret; 3070 3071 spin_lock_irqsave(&adapter->rwi_lock, flags); 3072 3073 /* If failover is pending don't schedule any other reset. 3074 * Instead let the failover complete. If there is already a 3075 * a failover reset scheduled, we will detect and drop the 3076 * duplicate reset when walking the ->rwi_list below. 3077 */ 3078 if (adapter->state == VNIC_REMOVING || 3079 adapter->state == VNIC_REMOVED || 3080 (adapter->failover_pending && reason != VNIC_RESET_FAILOVER)) { 3081 ret = EBUSY; 3082 netdev_dbg(netdev, "Adapter removing or pending failover, skipping reset\n"); 3083 goto err; 3084 } 3085 3086 list_for_each_entry(tmp, &adapter->rwi_list, list) { 3087 if (tmp->reset_reason == reason) { 3088 netdev_dbg(netdev, "Skipping matching reset, reason=%s\n", 3089 reset_reason_to_string(reason)); 3090 ret = EBUSY; 3091 goto err; 3092 } 3093 } 3094 3095 rwi = kzalloc(sizeof(*rwi), GFP_ATOMIC); 3096 if (!rwi) { 3097 ret = ENOMEM; 3098 goto err; 3099 } 3100 /* if we just received a transport event, 3101 * flush reset queue and process this reset 3102 */ 3103 if (adapter->force_reset_recovery) 3104 flush_reset_queue(adapter); 3105 3106 rwi->reset_reason = reason; 3107 list_add_tail(&rwi->list, &adapter->rwi_list); 3108 netdev_dbg(adapter->netdev, "Scheduling reset (reason %s)\n", 3109 reset_reason_to_string(reason)); 3110 queue_work(system_long_wq, &adapter->ibmvnic_reset); 3111 3112 ret = 0; 3113 err: 3114 /* ibmvnic_close() below can block, so drop the lock first */ 3115 spin_unlock_irqrestore(&adapter->rwi_lock, flags); 3116 3117 if (ret == ENOMEM) 3118 ibmvnic_close(netdev); 3119 3120 return -ret; 3121 } 3122 3123 static void ibmvnic_tx_timeout(struct net_device *dev, unsigned int txqueue) 3124 { 3125 struct ibmvnic_adapter *adapter = netdev_priv(dev); 3126 3127 if (test_bit(0, &adapter->resetting)) { 3128 netdev_err(adapter->netdev, 3129 "Adapter is resetting, skip timeout reset\n"); 3130 return; 3131 } 3132 /* No queuing up reset until at least 5 seconds (default watchdog val) 3133 * after last reset 3134 */ 3135 if (time_before(jiffies, (adapter->last_reset_time + dev->watchdog_timeo))) { 3136 netdev_dbg(dev, "Not yet time to tx timeout.\n"); 3137 return; 3138 } 3139 ibmvnic_reset(adapter, VNIC_RESET_TIMEOUT); 3140 } 3141 3142 static void remove_buff_from_pool(struct ibmvnic_adapter *adapter, 3143 struct ibmvnic_rx_buff *rx_buff) 3144 { 3145 struct ibmvnic_rx_pool *pool = &adapter->rx_pool[rx_buff->pool_index]; 3146 3147 rx_buff->skb = NULL; 3148 3149 pool->free_map[pool->next_alloc] = (int)(rx_buff - pool->rx_buff); 3150 pool->next_alloc = (pool->next_alloc + 1) % pool->size; 3151 3152 atomic_dec(&pool->available); 3153 } 3154 3155 static int ibmvnic_poll(struct napi_struct *napi, int budget) 3156 { 3157 struct ibmvnic_sub_crq_queue *rx_scrq; 3158 struct ibmvnic_adapter *adapter; 3159 struct net_device *netdev; 3160 int frames_processed; 3161 int scrq_num; 3162 3163 netdev = napi->dev; 3164 adapter = netdev_priv(netdev); 3165 scrq_num = (int)(napi - adapter->napi); 3166 frames_processed = 0; 3167 rx_scrq = adapter->rx_scrq[scrq_num]; 3168 3169 restart_poll: 3170 while (frames_processed < budget) { 3171 struct sk_buff *skb; 3172 struct ibmvnic_rx_buff *rx_buff; 3173 union sub_crq *next; 3174 u32 length; 3175 u16 offset; 3176 u8 flags = 0; 3177 3178 if (unlikely(test_bit(0, &adapter->resetting) && 3179 adapter->reset_reason != VNIC_RESET_NON_FATAL)) { 3180 enable_scrq_irq(adapter, rx_scrq); 3181 napi_complete_done(napi, frames_processed); 3182 return frames_processed; 3183 } 3184 3185 if (!pending_scrq(adapter, rx_scrq)) 3186 break; 3187 next = ibmvnic_next_scrq(adapter, rx_scrq); 3188 rx_buff = (struct ibmvnic_rx_buff *) 3189 be64_to_cpu(next->rx_comp.correlator); 3190 /* do error checking */ 3191 if (next->rx_comp.rc) { 3192 netdev_dbg(netdev, "rx buffer returned with rc %x\n", 3193 be16_to_cpu(next->rx_comp.rc)); 3194 /* free the entry */ 3195 next->rx_comp.first = 0; 3196 dev_kfree_skb_any(rx_buff->skb); 3197 remove_buff_from_pool(adapter, rx_buff); 3198 continue; 3199 } else if (!rx_buff->skb) { 3200 /* free the entry */ 3201 next->rx_comp.first = 0; 3202 remove_buff_from_pool(adapter, rx_buff); 3203 continue; 3204 } 3205 3206 length = be32_to_cpu(next->rx_comp.len); 3207 offset = be16_to_cpu(next->rx_comp.off_frame_data); 3208 flags = next->rx_comp.flags; 3209 skb = rx_buff->skb; 3210 /* load long_term_buff before copying to skb */ 3211 dma_rmb(); 3212 skb_copy_to_linear_data(skb, rx_buff->data + offset, 3213 length); 3214 3215 /* VLAN Header has been stripped by the system firmware and 3216 * needs to be inserted by the driver 3217 */ 3218 if (adapter->rx_vlan_header_insertion && 3219 (flags & IBMVNIC_VLAN_STRIPPED)) 3220 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), 3221 ntohs(next->rx_comp.vlan_tci)); 3222 3223 /* free the entry */ 3224 next->rx_comp.first = 0; 3225 remove_buff_from_pool(adapter, rx_buff); 3226 3227 skb_put(skb, length); 3228 skb->protocol = eth_type_trans(skb, netdev); 3229 skb_record_rx_queue(skb, scrq_num); 3230 3231 if (flags & IBMVNIC_IP_CHKSUM_GOOD && 3232 flags & IBMVNIC_TCP_UDP_CHKSUM_GOOD) { 3233 skb->ip_summed = CHECKSUM_UNNECESSARY; 3234 } 3235 3236 length = skb->len; 3237 napi_gro_receive(napi, skb); /* send it up */ 3238 netdev->stats.rx_packets++; 3239 netdev->stats.rx_bytes += length; 3240 adapter->rx_stats_buffers[scrq_num].packets++; 3241 adapter->rx_stats_buffers[scrq_num].bytes += length; 3242 frames_processed++; 3243 } 3244 3245 if (adapter->state != VNIC_CLOSING && 3246 ((atomic_read(&adapter->rx_pool[scrq_num].available) < 3247 adapter->req_rx_add_entries_per_subcrq / 2) || 3248 frames_processed < budget)) 3249 replenish_rx_pool(adapter, &adapter->rx_pool[scrq_num]); 3250 if (frames_processed < budget) { 3251 if (napi_complete_done(napi, frames_processed)) { 3252 enable_scrq_irq(adapter, rx_scrq); 3253 if (pending_scrq(adapter, rx_scrq)) { 3254 if (napi_reschedule(napi)) { 3255 disable_scrq_irq(adapter, rx_scrq); 3256 goto restart_poll; 3257 } 3258 } 3259 } 3260 } 3261 return frames_processed; 3262 } 3263 3264 static int wait_for_reset(struct ibmvnic_adapter *adapter) 3265 { 3266 int rc, ret; 3267 3268 adapter->fallback.mtu = adapter->req_mtu; 3269 adapter->fallback.rx_queues = adapter->req_rx_queues; 3270 adapter->fallback.tx_queues = adapter->req_tx_queues; 3271 adapter->fallback.rx_entries = adapter->req_rx_add_entries_per_subcrq; 3272 adapter->fallback.tx_entries = adapter->req_tx_entries_per_subcrq; 3273 3274 reinit_completion(&adapter->reset_done); 3275 adapter->wait_for_reset = true; 3276 rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM); 3277 3278 if (rc) { 3279 ret = rc; 3280 goto out; 3281 } 3282 rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done, 60000); 3283 if (rc) { 3284 ret = -ENODEV; 3285 goto out; 3286 } 3287 3288 ret = 0; 3289 if (adapter->reset_done_rc) { 3290 ret = -EIO; 3291 adapter->desired.mtu = adapter->fallback.mtu; 3292 adapter->desired.rx_queues = adapter->fallback.rx_queues; 3293 adapter->desired.tx_queues = adapter->fallback.tx_queues; 3294 adapter->desired.rx_entries = adapter->fallback.rx_entries; 3295 adapter->desired.tx_entries = adapter->fallback.tx_entries; 3296 3297 reinit_completion(&adapter->reset_done); 3298 adapter->wait_for_reset = true; 3299 rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM); 3300 if (rc) { 3301 ret = rc; 3302 goto out; 3303 } 3304 rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done, 3305 60000); 3306 if (rc) { 3307 ret = -ENODEV; 3308 goto out; 3309 } 3310 } 3311 out: 3312 adapter->wait_for_reset = false; 3313 3314 return ret; 3315 } 3316 3317 static int ibmvnic_change_mtu(struct net_device *netdev, int new_mtu) 3318 { 3319 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3320 3321 adapter->desired.mtu = new_mtu + ETH_HLEN; 3322 3323 return wait_for_reset(adapter); 3324 } 3325 3326 static netdev_features_t ibmvnic_features_check(struct sk_buff *skb, 3327 struct net_device *dev, 3328 netdev_features_t features) 3329 { 3330 /* Some backing hardware adapters can not 3331 * handle packets with a MSS less than 224 3332 * or with only one segment. 3333 */ 3334 if (skb_is_gso(skb)) { 3335 if (skb_shinfo(skb)->gso_size < 224 || 3336 skb_shinfo(skb)->gso_segs == 1) 3337 features &= ~NETIF_F_GSO_MASK; 3338 } 3339 3340 return features; 3341 } 3342 3343 static const struct net_device_ops ibmvnic_netdev_ops = { 3344 .ndo_open = ibmvnic_open, 3345 .ndo_stop = ibmvnic_close, 3346 .ndo_start_xmit = ibmvnic_xmit, 3347 .ndo_set_rx_mode = ibmvnic_set_multi, 3348 .ndo_set_mac_address = ibmvnic_set_mac, 3349 .ndo_validate_addr = eth_validate_addr, 3350 .ndo_tx_timeout = ibmvnic_tx_timeout, 3351 .ndo_change_mtu = ibmvnic_change_mtu, 3352 .ndo_features_check = ibmvnic_features_check, 3353 }; 3354 3355 /* ethtool functions */ 3356 3357 static int ibmvnic_get_link_ksettings(struct net_device *netdev, 3358 struct ethtool_link_ksettings *cmd) 3359 { 3360 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3361 int rc; 3362 3363 rc = send_query_phys_parms(adapter); 3364 if (rc) { 3365 adapter->speed = SPEED_UNKNOWN; 3366 adapter->duplex = DUPLEX_UNKNOWN; 3367 } 3368 cmd->base.speed = adapter->speed; 3369 cmd->base.duplex = adapter->duplex; 3370 cmd->base.port = PORT_FIBRE; 3371 cmd->base.phy_address = 0; 3372 cmd->base.autoneg = AUTONEG_ENABLE; 3373 3374 return 0; 3375 } 3376 3377 static void ibmvnic_get_drvinfo(struct net_device *netdev, 3378 struct ethtool_drvinfo *info) 3379 { 3380 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3381 3382 strscpy(info->driver, ibmvnic_driver_name, sizeof(info->driver)); 3383 strscpy(info->version, IBMVNIC_DRIVER_VERSION, sizeof(info->version)); 3384 strscpy(info->fw_version, adapter->fw_version, 3385 sizeof(info->fw_version)); 3386 } 3387 3388 static u32 ibmvnic_get_msglevel(struct net_device *netdev) 3389 { 3390 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3391 3392 return adapter->msg_enable; 3393 } 3394 3395 static void ibmvnic_set_msglevel(struct net_device *netdev, u32 data) 3396 { 3397 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3398 3399 adapter->msg_enable = data; 3400 } 3401 3402 static u32 ibmvnic_get_link(struct net_device *netdev) 3403 { 3404 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3405 3406 /* Don't need to send a query because we request a logical link up at 3407 * init and then we wait for link state indications 3408 */ 3409 return adapter->logical_link_state; 3410 } 3411 3412 static void ibmvnic_get_ringparam(struct net_device *netdev, 3413 struct ethtool_ringparam *ring, 3414 struct kernel_ethtool_ringparam *kernel_ring, 3415 struct netlink_ext_ack *extack) 3416 { 3417 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3418 3419 ring->rx_max_pending = adapter->max_rx_add_entries_per_subcrq; 3420 ring->tx_max_pending = adapter->max_tx_entries_per_subcrq; 3421 ring->rx_mini_max_pending = 0; 3422 ring->rx_jumbo_max_pending = 0; 3423 ring->rx_pending = adapter->req_rx_add_entries_per_subcrq; 3424 ring->tx_pending = adapter->req_tx_entries_per_subcrq; 3425 ring->rx_mini_pending = 0; 3426 ring->rx_jumbo_pending = 0; 3427 } 3428 3429 static int ibmvnic_set_ringparam(struct net_device *netdev, 3430 struct ethtool_ringparam *ring, 3431 struct kernel_ethtool_ringparam *kernel_ring, 3432 struct netlink_ext_ack *extack) 3433 { 3434 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3435 3436 if (ring->rx_pending > adapter->max_rx_add_entries_per_subcrq || 3437 ring->tx_pending > adapter->max_tx_entries_per_subcrq) { 3438 netdev_err(netdev, "Invalid request.\n"); 3439 netdev_err(netdev, "Max tx buffers = %llu\n", 3440 adapter->max_rx_add_entries_per_subcrq); 3441 netdev_err(netdev, "Max rx buffers = %llu\n", 3442 adapter->max_tx_entries_per_subcrq); 3443 return -EINVAL; 3444 } 3445 3446 adapter->desired.rx_entries = ring->rx_pending; 3447 adapter->desired.tx_entries = ring->tx_pending; 3448 3449 return wait_for_reset(adapter); 3450 } 3451 3452 static void ibmvnic_get_channels(struct net_device *netdev, 3453 struct ethtool_channels *channels) 3454 { 3455 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3456 3457 channels->max_rx = adapter->max_rx_queues; 3458 channels->max_tx = adapter->max_tx_queues; 3459 channels->max_other = 0; 3460 channels->max_combined = 0; 3461 channels->rx_count = adapter->req_rx_queues; 3462 channels->tx_count = adapter->req_tx_queues; 3463 channels->other_count = 0; 3464 channels->combined_count = 0; 3465 } 3466 3467 static int ibmvnic_set_channels(struct net_device *netdev, 3468 struct ethtool_channels *channels) 3469 { 3470 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 3471 3472 adapter->desired.rx_queues = channels->rx_count; 3473 adapter->desired.tx_queues = channels->tx_count; 3474 3475 return wait_for_reset(adapter); 3476 } 3477 3478 static void ibmvnic_get_strings(struct net_device *dev, u32 stringset, u8 *data) 3479 { 3480 struct ibmvnic_adapter *adapter = netdev_priv(dev); 3481 int i; 3482 3483 if (stringset != ETH_SS_STATS) 3484 return; 3485 3486 for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); i++, data += ETH_GSTRING_LEN) 3487 memcpy(data, ibmvnic_stats[i].name, ETH_GSTRING_LEN); 3488 3489 for (i = 0; i < adapter->req_tx_queues; i++) { 3490 snprintf(data, ETH_GSTRING_LEN, "tx%d_packets", i); 3491 data += ETH_GSTRING_LEN; 3492 3493 snprintf(data, ETH_GSTRING_LEN, "tx%d_bytes", i); 3494 data += ETH_GSTRING_LEN; 3495 3496 snprintf(data, ETH_GSTRING_LEN, "tx%d_dropped_packets", i); 3497 data += ETH_GSTRING_LEN; 3498 } 3499 3500 for (i = 0; i < adapter->req_rx_queues; i++) { 3501 snprintf(data, ETH_GSTRING_LEN, "rx%d_packets", i); 3502 data += ETH_GSTRING_LEN; 3503 3504 snprintf(data, ETH_GSTRING_LEN, "rx%d_bytes", i); 3505 data += ETH_GSTRING_LEN; 3506 3507 snprintf(data, ETH_GSTRING_LEN, "rx%d_interrupts", i); 3508 data += ETH_GSTRING_LEN; 3509 } 3510 } 3511 3512 static int ibmvnic_get_sset_count(struct net_device *dev, int sset) 3513 { 3514 struct ibmvnic_adapter *adapter = netdev_priv(dev); 3515 3516 switch (sset) { 3517 case ETH_SS_STATS: 3518 return ARRAY_SIZE(ibmvnic_stats) + 3519 adapter->req_tx_queues * NUM_TX_STATS + 3520 adapter->req_rx_queues * NUM_RX_STATS; 3521 default: 3522 return -EOPNOTSUPP; 3523 } 3524 } 3525 3526 static void ibmvnic_get_ethtool_stats(struct net_device *dev, 3527 struct ethtool_stats *stats, u64 *data) 3528 { 3529 struct ibmvnic_adapter *adapter = netdev_priv(dev); 3530 union ibmvnic_crq crq; 3531 int i, j; 3532 int rc; 3533 3534 memset(&crq, 0, sizeof(crq)); 3535 crq.request_statistics.first = IBMVNIC_CRQ_CMD; 3536 crq.request_statistics.cmd = REQUEST_STATISTICS; 3537 crq.request_statistics.ioba = cpu_to_be32(adapter->stats_token); 3538 crq.request_statistics.len = 3539 cpu_to_be32(sizeof(struct ibmvnic_statistics)); 3540 3541 /* Wait for data to be written */ 3542 reinit_completion(&adapter->stats_done); 3543 rc = ibmvnic_send_crq(adapter, &crq); 3544 if (rc) 3545 return; 3546 rc = ibmvnic_wait_for_completion(adapter, &adapter->stats_done, 10000); 3547 if (rc) 3548 return; 3549 3550 for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); i++) 3551 data[i] = be64_to_cpu(IBMVNIC_GET_STAT 3552 (adapter, ibmvnic_stats[i].offset)); 3553 3554 for (j = 0; j < adapter->req_tx_queues; j++) { 3555 data[i] = adapter->tx_stats_buffers[j].packets; 3556 i++; 3557 data[i] = adapter->tx_stats_buffers[j].bytes; 3558 i++; 3559 data[i] = adapter->tx_stats_buffers[j].dropped_packets; 3560 i++; 3561 } 3562 3563 for (j = 0; j < adapter->req_rx_queues; j++) { 3564 data[i] = adapter->rx_stats_buffers[j].packets; 3565 i++; 3566 data[i] = adapter->rx_stats_buffers[j].bytes; 3567 i++; 3568 data[i] = adapter->rx_stats_buffers[j].interrupts; 3569 i++; 3570 } 3571 } 3572 3573 static const struct ethtool_ops ibmvnic_ethtool_ops = { 3574 .get_drvinfo = ibmvnic_get_drvinfo, 3575 .get_msglevel = ibmvnic_get_msglevel, 3576 .set_msglevel = ibmvnic_set_msglevel, 3577 .get_link = ibmvnic_get_link, 3578 .get_ringparam = ibmvnic_get_ringparam, 3579 .set_ringparam = ibmvnic_set_ringparam, 3580 .get_channels = ibmvnic_get_channels, 3581 .set_channels = ibmvnic_set_channels, 3582 .get_strings = ibmvnic_get_strings, 3583 .get_sset_count = ibmvnic_get_sset_count, 3584 .get_ethtool_stats = ibmvnic_get_ethtool_stats, 3585 .get_link_ksettings = ibmvnic_get_link_ksettings, 3586 }; 3587 3588 /* Routines for managing CRQs/sCRQs */ 3589 3590 static int reset_one_sub_crq_queue(struct ibmvnic_adapter *adapter, 3591 struct ibmvnic_sub_crq_queue *scrq) 3592 { 3593 int rc; 3594 3595 if (!scrq) { 3596 netdev_dbg(adapter->netdev, "Invalid scrq reset.\n"); 3597 return -EINVAL; 3598 } 3599 3600 if (scrq->irq) { 3601 free_irq(scrq->irq, scrq); 3602 irq_dispose_mapping(scrq->irq); 3603 scrq->irq = 0; 3604 } 3605 3606 if (scrq->msgs) { 3607 memset(scrq->msgs, 0, 4 * PAGE_SIZE); 3608 atomic_set(&scrq->used, 0); 3609 scrq->cur = 0; 3610 scrq->ind_buf.index = 0; 3611 } else { 3612 netdev_dbg(adapter->netdev, "Invalid scrq reset\n"); 3613 return -EINVAL; 3614 } 3615 3616 rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token, 3617 4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq); 3618 return rc; 3619 } 3620 3621 static int reset_sub_crq_queues(struct ibmvnic_adapter *adapter) 3622 { 3623 int i, rc; 3624 3625 if (!adapter->tx_scrq || !adapter->rx_scrq) 3626 return -EINVAL; 3627 3628 for (i = 0; i < adapter->req_tx_queues; i++) { 3629 netdev_dbg(adapter->netdev, "Re-setting tx_scrq[%d]\n", i); 3630 rc = reset_one_sub_crq_queue(adapter, adapter->tx_scrq[i]); 3631 if (rc) 3632 return rc; 3633 } 3634 3635 for (i = 0; i < adapter->req_rx_queues; i++) { 3636 netdev_dbg(adapter->netdev, "Re-setting rx_scrq[%d]\n", i); 3637 rc = reset_one_sub_crq_queue(adapter, adapter->rx_scrq[i]); 3638 if (rc) 3639 return rc; 3640 } 3641 3642 return rc; 3643 } 3644 3645 static void release_sub_crq_queue(struct ibmvnic_adapter *adapter, 3646 struct ibmvnic_sub_crq_queue *scrq, 3647 bool do_h_free) 3648 { 3649 struct device *dev = &adapter->vdev->dev; 3650 long rc; 3651 3652 netdev_dbg(adapter->netdev, "Releasing sub-CRQ\n"); 3653 3654 if (do_h_free) { 3655 /* Close the sub-crqs */ 3656 do { 3657 rc = plpar_hcall_norets(H_FREE_SUB_CRQ, 3658 adapter->vdev->unit_address, 3659 scrq->crq_num); 3660 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc)); 3661 3662 if (rc) { 3663 netdev_err(adapter->netdev, 3664 "Failed to release sub-CRQ %16lx, rc = %ld\n", 3665 scrq->crq_num, rc); 3666 } 3667 } 3668 3669 dma_free_coherent(dev, 3670 IBMVNIC_IND_ARR_SZ, 3671 scrq->ind_buf.indir_arr, 3672 scrq->ind_buf.indir_dma); 3673 3674 dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE, 3675 DMA_BIDIRECTIONAL); 3676 free_pages((unsigned long)scrq->msgs, 2); 3677 kfree(scrq); 3678 } 3679 3680 static struct ibmvnic_sub_crq_queue *init_sub_crq_queue(struct ibmvnic_adapter 3681 *adapter) 3682 { 3683 struct device *dev = &adapter->vdev->dev; 3684 struct ibmvnic_sub_crq_queue *scrq; 3685 int rc; 3686 3687 scrq = kzalloc(sizeof(*scrq), GFP_KERNEL); 3688 if (!scrq) 3689 return NULL; 3690 3691 scrq->msgs = 3692 (union sub_crq *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 2); 3693 if (!scrq->msgs) { 3694 dev_warn(dev, "Couldn't allocate crq queue messages page\n"); 3695 goto zero_page_failed; 3696 } 3697 3698 scrq->msg_token = dma_map_single(dev, scrq->msgs, 4 * PAGE_SIZE, 3699 DMA_BIDIRECTIONAL); 3700 if (dma_mapping_error(dev, scrq->msg_token)) { 3701 dev_warn(dev, "Couldn't map crq queue messages page\n"); 3702 goto map_failed; 3703 } 3704 3705 rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token, 3706 4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq); 3707 3708 if (rc == H_RESOURCE) 3709 rc = ibmvnic_reset_crq(adapter); 3710 3711 if (rc == H_CLOSED) { 3712 dev_warn(dev, "Partner adapter not ready, waiting.\n"); 3713 } else if (rc) { 3714 dev_warn(dev, "Error %d registering sub-crq\n", rc); 3715 goto reg_failed; 3716 } 3717 3718 scrq->adapter = adapter; 3719 scrq->size = 4 * PAGE_SIZE / sizeof(*scrq->msgs); 3720 scrq->ind_buf.index = 0; 3721 3722 scrq->ind_buf.indir_arr = 3723 dma_alloc_coherent(dev, 3724 IBMVNIC_IND_ARR_SZ, 3725 &scrq->ind_buf.indir_dma, 3726 GFP_KERNEL); 3727 3728 if (!scrq->ind_buf.indir_arr) 3729 goto indir_failed; 3730 3731 spin_lock_init(&scrq->lock); 3732 3733 netdev_dbg(adapter->netdev, 3734 "sub-crq initialized, num %lx, hw_irq=%lx, irq=%x\n", 3735 scrq->crq_num, scrq->hw_irq, scrq->irq); 3736 3737 return scrq; 3738 3739 indir_failed: 3740 do { 3741 rc = plpar_hcall_norets(H_FREE_SUB_CRQ, 3742 adapter->vdev->unit_address, 3743 scrq->crq_num); 3744 } while (rc == H_BUSY || rc == H_IS_LONG_BUSY(rc)); 3745 reg_failed: 3746 dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE, 3747 DMA_BIDIRECTIONAL); 3748 map_failed: 3749 free_pages((unsigned long)scrq->msgs, 2); 3750 zero_page_failed: 3751 kfree(scrq); 3752 3753 return NULL; 3754 } 3755 3756 static void release_sub_crqs(struct ibmvnic_adapter *adapter, bool do_h_free) 3757 { 3758 int i; 3759 3760 if (adapter->tx_scrq) { 3761 for (i = 0; i < adapter->num_active_tx_scrqs; i++) { 3762 if (!adapter->tx_scrq[i]) 3763 continue; 3764 3765 netdev_dbg(adapter->netdev, "Releasing tx_scrq[%d]\n", 3766 i); 3767 ibmvnic_tx_scrq_clean_buffer(adapter, adapter->tx_scrq[i]); 3768 if (adapter->tx_scrq[i]->irq) { 3769 free_irq(adapter->tx_scrq[i]->irq, 3770 adapter->tx_scrq[i]); 3771 irq_dispose_mapping(adapter->tx_scrq[i]->irq); 3772 adapter->tx_scrq[i]->irq = 0; 3773 } 3774 3775 release_sub_crq_queue(adapter, adapter->tx_scrq[i], 3776 do_h_free); 3777 } 3778 3779 kfree(adapter->tx_scrq); 3780 adapter->tx_scrq = NULL; 3781 adapter->num_active_tx_scrqs = 0; 3782 } 3783 3784 if (adapter->rx_scrq) { 3785 for (i = 0; i < adapter->num_active_rx_scrqs; i++) { 3786 if (!adapter->rx_scrq[i]) 3787 continue; 3788 3789 netdev_dbg(adapter->netdev, "Releasing rx_scrq[%d]\n", 3790 i); 3791 if (adapter->rx_scrq[i]->irq) { 3792 free_irq(adapter->rx_scrq[i]->irq, 3793 adapter->rx_scrq[i]); 3794 irq_dispose_mapping(adapter->rx_scrq[i]->irq); 3795 adapter->rx_scrq[i]->irq = 0; 3796 } 3797 3798 release_sub_crq_queue(adapter, adapter->rx_scrq[i], 3799 do_h_free); 3800 } 3801 3802 kfree(adapter->rx_scrq); 3803 adapter->rx_scrq = NULL; 3804 adapter->num_active_rx_scrqs = 0; 3805 } 3806 } 3807 3808 static int disable_scrq_irq(struct ibmvnic_adapter *adapter, 3809 struct ibmvnic_sub_crq_queue *scrq) 3810 { 3811 struct device *dev = &adapter->vdev->dev; 3812 unsigned long rc; 3813 3814 rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address, 3815 H_DISABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0); 3816 if (rc) 3817 dev_err(dev, "Couldn't disable scrq irq 0x%lx. rc=%ld\n", 3818 scrq->hw_irq, rc); 3819 return rc; 3820 } 3821 3822 /* We can not use the IRQ chip EOI handler because that has the 3823 * unintended effect of changing the interrupt priority. 3824 */ 3825 static void ibmvnic_xics_eoi(struct device *dev, struct ibmvnic_sub_crq_queue *scrq) 3826 { 3827 u64 val = 0xff000000 | scrq->hw_irq; 3828 unsigned long rc; 3829 3830 rc = plpar_hcall_norets(H_EOI, val); 3831 if (rc) 3832 dev_err(dev, "H_EOI FAILED irq 0x%llx. rc=%ld\n", val, rc); 3833 } 3834 3835 /* Due to a firmware bug, the hypervisor can send an interrupt to a 3836 * transmit or receive queue just prior to a partition migration. 3837 * Force an EOI after migration. 3838 */ 3839 static void ibmvnic_clear_pending_interrupt(struct device *dev, 3840 struct ibmvnic_sub_crq_queue *scrq) 3841 { 3842 if (!xive_enabled()) 3843 ibmvnic_xics_eoi(dev, scrq); 3844 } 3845 3846 static int enable_scrq_irq(struct ibmvnic_adapter *adapter, 3847 struct ibmvnic_sub_crq_queue *scrq) 3848 { 3849 struct device *dev = &adapter->vdev->dev; 3850 unsigned long rc; 3851 3852 if (scrq->hw_irq > 0x100000000ULL) { 3853 dev_err(dev, "bad hw_irq = %lx\n", scrq->hw_irq); 3854 return 1; 3855 } 3856 3857 if (test_bit(0, &adapter->resetting) && 3858 adapter->reset_reason == VNIC_RESET_MOBILITY) { 3859 ibmvnic_clear_pending_interrupt(dev, scrq); 3860 } 3861 3862 rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address, 3863 H_ENABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0); 3864 if (rc) 3865 dev_err(dev, "Couldn't enable scrq irq 0x%lx. rc=%ld\n", 3866 scrq->hw_irq, rc); 3867 return rc; 3868 } 3869 3870 static int ibmvnic_complete_tx(struct ibmvnic_adapter *adapter, 3871 struct ibmvnic_sub_crq_queue *scrq) 3872 { 3873 struct device *dev = &adapter->vdev->dev; 3874 struct ibmvnic_tx_pool *tx_pool; 3875 struct ibmvnic_tx_buff *txbuff; 3876 struct netdev_queue *txq; 3877 union sub_crq *next; 3878 int index; 3879 int i; 3880 3881 restart_loop: 3882 while (pending_scrq(adapter, scrq)) { 3883 unsigned int pool = scrq->pool_index; 3884 int num_entries = 0; 3885 int total_bytes = 0; 3886 int num_packets = 0; 3887 3888 next = ibmvnic_next_scrq(adapter, scrq); 3889 for (i = 0; i < next->tx_comp.num_comps; i++) { 3890 index = be32_to_cpu(next->tx_comp.correlators[i]); 3891 if (index & IBMVNIC_TSO_POOL_MASK) { 3892 tx_pool = &adapter->tso_pool[pool]; 3893 index &= ~IBMVNIC_TSO_POOL_MASK; 3894 } else { 3895 tx_pool = &adapter->tx_pool[pool]; 3896 } 3897 3898 txbuff = &tx_pool->tx_buff[index]; 3899 num_packets++; 3900 num_entries += txbuff->num_entries; 3901 if (txbuff->skb) { 3902 total_bytes += txbuff->skb->len; 3903 if (next->tx_comp.rcs[i]) { 3904 dev_err(dev, "tx error %x\n", 3905 next->tx_comp.rcs[i]); 3906 dev_kfree_skb_irq(txbuff->skb); 3907 } else { 3908 dev_consume_skb_irq(txbuff->skb); 3909 } 3910 txbuff->skb = NULL; 3911 } else { 3912 netdev_warn(adapter->netdev, 3913 "TX completion received with NULL socket buffer\n"); 3914 } 3915 tx_pool->free_map[tx_pool->producer_index] = index; 3916 tx_pool->producer_index = 3917 (tx_pool->producer_index + 1) % 3918 tx_pool->num_buffers; 3919 } 3920 /* remove tx_comp scrq*/ 3921 next->tx_comp.first = 0; 3922 3923 txq = netdev_get_tx_queue(adapter->netdev, scrq->pool_index); 3924 netdev_tx_completed_queue(txq, num_packets, total_bytes); 3925 3926 if (atomic_sub_return(num_entries, &scrq->used) <= 3927 (adapter->req_tx_entries_per_subcrq / 2) && 3928 __netif_subqueue_stopped(adapter->netdev, 3929 scrq->pool_index)) { 3930 rcu_read_lock(); 3931 if (adapter->tx_queues_active) { 3932 netif_wake_subqueue(adapter->netdev, 3933 scrq->pool_index); 3934 netdev_dbg(adapter->netdev, 3935 "Started queue %d\n", 3936 scrq->pool_index); 3937 } 3938 rcu_read_unlock(); 3939 } 3940 } 3941 3942 enable_scrq_irq(adapter, scrq); 3943 3944 if (pending_scrq(adapter, scrq)) { 3945 disable_scrq_irq(adapter, scrq); 3946 goto restart_loop; 3947 } 3948 3949 return 0; 3950 } 3951 3952 static irqreturn_t ibmvnic_interrupt_tx(int irq, void *instance) 3953 { 3954 struct ibmvnic_sub_crq_queue *scrq = instance; 3955 struct ibmvnic_adapter *adapter = scrq->adapter; 3956 3957 disable_scrq_irq(adapter, scrq); 3958 ibmvnic_complete_tx(adapter, scrq); 3959 3960 return IRQ_HANDLED; 3961 } 3962 3963 static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance) 3964 { 3965 struct ibmvnic_sub_crq_queue *scrq = instance; 3966 struct ibmvnic_adapter *adapter = scrq->adapter; 3967 3968 /* When booting a kdump kernel we can hit pending interrupts 3969 * prior to completing driver initialization. 3970 */ 3971 if (unlikely(adapter->state != VNIC_OPEN)) 3972 return IRQ_NONE; 3973 3974 adapter->rx_stats_buffers[scrq->scrq_num].interrupts++; 3975 3976 if (napi_schedule_prep(&adapter->napi[scrq->scrq_num])) { 3977 disable_scrq_irq(adapter, scrq); 3978 __napi_schedule(&adapter->napi[scrq->scrq_num]); 3979 } 3980 3981 return IRQ_HANDLED; 3982 } 3983 3984 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter) 3985 { 3986 struct device *dev = &adapter->vdev->dev; 3987 struct ibmvnic_sub_crq_queue *scrq; 3988 int i = 0, j = 0; 3989 int rc = 0; 3990 3991 for (i = 0; i < adapter->req_tx_queues; i++) { 3992 netdev_dbg(adapter->netdev, "Initializing tx_scrq[%d] irq\n", 3993 i); 3994 scrq = adapter->tx_scrq[i]; 3995 scrq->irq = irq_create_mapping(NULL, scrq->hw_irq); 3996 3997 if (!scrq->irq) { 3998 rc = -EINVAL; 3999 dev_err(dev, "Error mapping irq\n"); 4000 goto req_tx_irq_failed; 4001 } 4002 4003 snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-tx%d", 4004 adapter->vdev->unit_address, i); 4005 rc = request_irq(scrq->irq, ibmvnic_interrupt_tx, 4006 0, scrq->name, scrq); 4007 4008 if (rc) { 4009 dev_err(dev, "Couldn't register tx irq 0x%x. rc=%d\n", 4010 scrq->irq, rc); 4011 irq_dispose_mapping(scrq->irq); 4012 goto req_tx_irq_failed; 4013 } 4014 } 4015 4016 for (i = 0; i < adapter->req_rx_queues; i++) { 4017 netdev_dbg(adapter->netdev, "Initializing rx_scrq[%d] irq\n", 4018 i); 4019 scrq = adapter->rx_scrq[i]; 4020 scrq->irq = irq_create_mapping(NULL, scrq->hw_irq); 4021 if (!scrq->irq) { 4022 rc = -EINVAL; 4023 dev_err(dev, "Error mapping irq\n"); 4024 goto req_rx_irq_failed; 4025 } 4026 snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-rx%d", 4027 adapter->vdev->unit_address, i); 4028 rc = request_irq(scrq->irq, ibmvnic_interrupt_rx, 4029 0, scrq->name, scrq); 4030 if (rc) { 4031 dev_err(dev, "Couldn't register rx irq 0x%x. rc=%d\n", 4032 scrq->irq, rc); 4033 irq_dispose_mapping(scrq->irq); 4034 goto req_rx_irq_failed; 4035 } 4036 } 4037 return rc; 4038 4039 req_rx_irq_failed: 4040 for (j = 0; j < i; j++) { 4041 free_irq(adapter->rx_scrq[j]->irq, adapter->rx_scrq[j]); 4042 irq_dispose_mapping(adapter->rx_scrq[j]->irq); 4043 } 4044 i = adapter->req_tx_queues; 4045 req_tx_irq_failed: 4046 for (j = 0; j < i; j++) { 4047 free_irq(adapter->tx_scrq[j]->irq, adapter->tx_scrq[j]); 4048 irq_dispose_mapping(adapter->tx_scrq[j]->irq); 4049 } 4050 release_sub_crqs(adapter, 1); 4051 return rc; 4052 } 4053 4054 static int init_sub_crqs(struct ibmvnic_adapter *adapter) 4055 { 4056 struct device *dev = &adapter->vdev->dev; 4057 struct ibmvnic_sub_crq_queue **allqueues; 4058 int registered_queues = 0; 4059 int total_queues; 4060 int more = 0; 4061 int i; 4062 4063 total_queues = adapter->req_tx_queues + adapter->req_rx_queues; 4064 4065 allqueues = kcalloc(total_queues, sizeof(*allqueues), GFP_KERNEL); 4066 if (!allqueues) 4067 return -ENOMEM; 4068 4069 for (i = 0; i < total_queues; i++) { 4070 allqueues[i] = init_sub_crq_queue(adapter); 4071 if (!allqueues[i]) { 4072 dev_warn(dev, "Couldn't allocate all sub-crqs\n"); 4073 break; 4074 } 4075 registered_queues++; 4076 } 4077 4078 /* Make sure we were able to register the minimum number of queues */ 4079 if (registered_queues < 4080 adapter->min_tx_queues + adapter->min_rx_queues) { 4081 dev_err(dev, "Fatal: Couldn't init min number of sub-crqs\n"); 4082 goto tx_failed; 4083 } 4084 4085 /* Distribute the failed allocated queues*/ 4086 for (i = 0; i < total_queues - registered_queues + more ; i++) { 4087 netdev_dbg(adapter->netdev, "Reducing number of queues\n"); 4088 switch (i % 3) { 4089 case 0: 4090 if (adapter->req_rx_queues > adapter->min_rx_queues) 4091 adapter->req_rx_queues--; 4092 else 4093 more++; 4094 break; 4095 case 1: 4096 if (adapter->req_tx_queues > adapter->min_tx_queues) 4097 adapter->req_tx_queues--; 4098 else 4099 more++; 4100 break; 4101 } 4102 } 4103 4104 adapter->tx_scrq = kcalloc(adapter->req_tx_queues, 4105 sizeof(*adapter->tx_scrq), GFP_KERNEL); 4106 if (!adapter->tx_scrq) 4107 goto tx_failed; 4108 4109 for (i = 0; i < adapter->req_tx_queues; i++) { 4110 adapter->tx_scrq[i] = allqueues[i]; 4111 adapter->tx_scrq[i]->pool_index = i; 4112 adapter->num_active_tx_scrqs++; 4113 } 4114 4115 adapter->rx_scrq = kcalloc(adapter->req_rx_queues, 4116 sizeof(*adapter->rx_scrq), GFP_KERNEL); 4117 if (!adapter->rx_scrq) 4118 goto rx_failed; 4119 4120 for (i = 0; i < adapter->req_rx_queues; i++) { 4121 adapter->rx_scrq[i] = allqueues[i + adapter->req_tx_queues]; 4122 adapter->rx_scrq[i]->scrq_num = i; 4123 adapter->num_active_rx_scrqs++; 4124 } 4125 4126 kfree(allqueues); 4127 return 0; 4128 4129 rx_failed: 4130 kfree(adapter->tx_scrq); 4131 adapter->tx_scrq = NULL; 4132 tx_failed: 4133 for (i = 0; i < registered_queues; i++) 4134 release_sub_crq_queue(adapter, allqueues[i], 1); 4135 kfree(allqueues); 4136 return -ENOMEM; 4137 } 4138 4139 static void send_request_cap(struct ibmvnic_adapter *adapter, int retry) 4140 { 4141 struct device *dev = &adapter->vdev->dev; 4142 union ibmvnic_crq crq; 4143 int max_entries; 4144 int cap_reqs; 4145 4146 /* We send out 6 or 7 REQUEST_CAPABILITY CRQs below (depending on 4147 * the PROMISC flag). Initialize this count upfront. When the tasklet 4148 * receives a response to all of these, it will send the next protocol 4149 * message (QUERY_IP_OFFLOAD). 4150 */ 4151 if (!(adapter->netdev->flags & IFF_PROMISC) || 4152 adapter->promisc_supported) 4153 cap_reqs = 7; 4154 else 4155 cap_reqs = 6; 4156 4157 if (!retry) { 4158 /* Sub-CRQ entries are 32 byte long */ 4159 int entries_page = 4 * PAGE_SIZE / (sizeof(u64) * 4); 4160 4161 atomic_set(&adapter->running_cap_crqs, cap_reqs); 4162 4163 if (adapter->min_tx_entries_per_subcrq > entries_page || 4164 adapter->min_rx_add_entries_per_subcrq > entries_page) { 4165 dev_err(dev, "Fatal, invalid entries per sub-crq\n"); 4166 return; 4167 } 4168 4169 if (adapter->desired.mtu) 4170 adapter->req_mtu = adapter->desired.mtu; 4171 else 4172 adapter->req_mtu = adapter->netdev->mtu + ETH_HLEN; 4173 4174 if (!adapter->desired.tx_entries) 4175 adapter->desired.tx_entries = 4176 adapter->max_tx_entries_per_subcrq; 4177 if (!adapter->desired.rx_entries) 4178 adapter->desired.rx_entries = 4179 adapter->max_rx_add_entries_per_subcrq; 4180 4181 max_entries = IBMVNIC_LTB_SET_SIZE / 4182 (adapter->req_mtu + IBMVNIC_BUFFER_HLEN); 4183 4184 if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) * 4185 adapter->desired.tx_entries > IBMVNIC_LTB_SET_SIZE) { 4186 adapter->desired.tx_entries = max_entries; 4187 } 4188 4189 if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) * 4190 adapter->desired.rx_entries > IBMVNIC_LTB_SET_SIZE) { 4191 adapter->desired.rx_entries = max_entries; 4192 } 4193 4194 if (adapter->desired.tx_entries) 4195 adapter->req_tx_entries_per_subcrq = 4196 adapter->desired.tx_entries; 4197 else 4198 adapter->req_tx_entries_per_subcrq = 4199 adapter->max_tx_entries_per_subcrq; 4200 4201 if (adapter->desired.rx_entries) 4202 adapter->req_rx_add_entries_per_subcrq = 4203 adapter->desired.rx_entries; 4204 else 4205 adapter->req_rx_add_entries_per_subcrq = 4206 adapter->max_rx_add_entries_per_subcrq; 4207 4208 if (adapter->desired.tx_queues) 4209 adapter->req_tx_queues = 4210 adapter->desired.tx_queues; 4211 else 4212 adapter->req_tx_queues = 4213 adapter->opt_tx_comp_sub_queues; 4214 4215 if (adapter->desired.rx_queues) 4216 adapter->req_rx_queues = 4217 adapter->desired.rx_queues; 4218 else 4219 adapter->req_rx_queues = 4220 adapter->opt_rx_comp_queues; 4221 4222 adapter->req_rx_add_queues = adapter->max_rx_add_queues; 4223 } else { 4224 atomic_add(cap_reqs, &adapter->running_cap_crqs); 4225 } 4226 memset(&crq, 0, sizeof(crq)); 4227 crq.request_capability.first = IBMVNIC_CRQ_CMD; 4228 crq.request_capability.cmd = REQUEST_CAPABILITY; 4229 4230 crq.request_capability.capability = cpu_to_be16(REQ_TX_QUEUES); 4231 crq.request_capability.number = cpu_to_be64(adapter->req_tx_queues); 4232 cap_reqs--; 4233 ibmvnic_send_crq(adapter, &crq); 4234 4235 crq.request_capability.capability = cpu_to_be16(REQ_RX_QUEUES); 4236 crq.request_capability.number = cpu_to_be64(adapter->req_rx_queues); 4237 cap_reqs--; 4238 ibmvnic_send_crq(adapter, &crq); 4239 4240 crq.request_capability.capability = cpu_to_be16(REQ_RX_ADD_QUEUES); 4241 crq.request_capability.number = cpu_to_be64(adapter->req_rx_add_queues); 4242 cap_reqs--; 4243 ibmvnic_send_crq(adapter, &crq); 4244 4245 crq.request_capability.capability = 4246 cpu_to_be16(REQ_TX_ENTRIES_PER_SUBCRQ); 4247 crq.request_capability.number = 4248 cpu_to_be64(adapter->req_tx_entries_per_subcrq); 4249 cap_reqs--; 4250 ibmvnic_send_crq(adapter, &crq); 4251 4252 crq.request_capability.capability = 4253 cpu_to_be16(REQ_RX_ADD_ENTRIES_PER_SUBCRQ); 4254 crq.request_capability.number = 4255 cpu_to_be64(adapter->req_rx_add_entries_per_subcrq); 4256 cap_reqs--; 4257 ibmvnic_send_crq(adapter, &crq); 4258 4259 crq.request_capability.capability = cpu_to_be16(REQ_MTU); 4260 crq.request_capability.number = cpu_to_be64(adapter->req_mtu); 4261 cap_reqs--; 4262 ibmvnic_send_crq(adapter, &crq); 4263 4264 if (adapter->netdev->flags & IFF_PROMISC) { 4265 if (adapter->promisc_supported) { 4266 crq.request_capability.capability = 4267 cpu_to_be16(PROMISC_REQUESTED); 4268 crq.request_capability.number = cpu_to_be64(1); 4269 cap_reqs--; 4270 ibmvnic_send_crq(adapter, &crq); 4271 } 4272 } else { 4273 crq.request_capability.capability = 4274 cpu_to_be16(PROMISC_REQUESTED); 4275 crq.request_capability.number = cpu_to_be64(0); 4276 cap_reqs--; 4277 ibmvnic_send_crq(adapter, &crq); 4278 } 4279 4280 /* Keep at end to catch any discrepancy between expected and actual 4281 * CRQs sent. 4282 */ 4283 WARN_ON(cap_reqs != 0); 4284 } 4285 4286 static int pending_scrq(struct ibmvnic_adapter *adapter, 4287 struct ibmvnic_sub_crq_queue *scrq) 4288 { 4289 union sub_crq *entry = &scrq->msgs[scrq->cur]; 4290 int rc; 4291 4292 rc = !!(entry->generic.first & IBMVNIC_CRQ_CMD_RSP); 4293 4294 /* Ensure that the SCRQ valid flag is loaded prior to loading the 4295 * contents of the SCRQ descriptor 4296 */ 4297 dma_rmb(); 4298 4299 return rc; 4300 } 4301 4302 static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *adapter, 4303 struct ibmvnic_sub_crq_queue *scrq) 4304 { 4305 union sub_crq *entry; 4306 unsigned long flags; 4307 4308 spin_lock_irqsave(&scrq->lock, flags); 4309 entry = &scrq->msgs[scrq->cur]; 4310 if (entry->generic.first & IBMVNIC_CRQ_CMD_RSP) { 4311 if (++scrq->cur == scrq->size) 4312 scrq->cur = 0; 4313 } else { 4314 entry = NULL; 4315 } 4316 spin_unlock_irqrestore(&scrq->lock, flags); 4317 4318 /* Ensure that the SCRQ valid flag is loaded prior to loading the 4319 * contents of the SCRQ descriptor 4320 */ 4321 dma_rmb(); 4322 4323 return entry; 4324 } 4325 4326 static union ibmvnic_crq *ibmvnic_next_crq(struct ibmvnic_adapter *adapter) 4327 { 4328 struct ibmvnic_crq_queue *queue = &adapter->crq; 4329 union ibmvnic_crq *crq; 4330 4331 crq = &queue->msgs[queue->cur]; 4332 if (crq->generic.first & IBMVNIC_CRQ_CMD_RSP) { 4333 if (++queue->cur == queue->size) 4334 queue->cur = 0; 4335 } else { 4336 crq = NULL; 4337 } 4338 4339 return crq; 4340 } 4341 4342 static void print_subcrq_error(struct device *dev, int rc, const char *func) 4343 { 4344 switch (rc) { 4345 case H_PARAMETER: 4346 dev_warn_ratelimited(dev, 4347 "%s failed: Send request is malformed or adapter failover pending. (rc=%d)\n", 4348 func, rc); 4349 break; 4350 case H_CLOSED: 4351 dev_warn_ratelimited(dev, 4352 "%s failed: Backing queue closed. Adapter is down or failover pending. (rc=%d)\n", 4353 func, rc); 4354 break; 4355 default: 4356 dev_err_ratelimited(dev, "%s failed: (rc=%d)\n", func, rc); 4357 break; 4358 } 4359 } 4360 4361 static int send_subcrq_indirect(struct ibmvnic_adapter *adapter, 4362 u64 remote_handle, u64 ioba, u64 num_entries) 4363 { 4364 unsigned int ua = adapter->vdev->unit_address; 4365 struct device *dev = &adapter->vdev->dev; 4366 int rc; 4367 4368 /* Make sure the hypervisor sees the complete request */ 4369 dma_wmb(); 4370 rc = plpar_hcall_norets(H_SEND_SUB_CRQ_INDIRECT, ua, 4371 cpu_to_be64(remote_handle), 4372 ioba, num_entries); 4373 4374 if (rc) 4375 print_subcrq_error(dev, rc, __func__); 4376 4377 return rc; 4378 } 4379 4380 static int ibmvnic_send_crq(struct ibmvnic_adapter *adapter, 4381 union ibmvnic_crq *crq) 4382 { 4383 unsigned int ua = adapter->vdev->unit_address; 4384 struct device *dev = &adapter->vdev->dev; 4385 u64 *u64_crq = (u64 *)crq; 4386 int rc; 4387 4388 netdev_dbg(adapter->netdev, "Sending CRQ: %016lx %016lx\n", 4389 (unsigned long)cpu_to_be64(u64_crq[0]), 4390 (unsigned long)cpu_to_be64(u64_crq[1])); 4391 4392 if (!adapter->crq.active && 4393 crq->generic.first != IBMVNIC_CRQ_INIT_CMD) { 4394 dev_warn(dev, "Invalid request detected while CRQ is inactive, possible device state change during reset\n"); 4395 return -EINVAL; 4396 } 4397 4398 /* Make sure the hypervisor sees the complete request */ 4399 dma_wmb(); 4400 4401 rc = plpar_hcall_norets(H_SEND_CRQ, ua, 4402 cpu_to_be64(u64_crq[0]), 4403 cpu_to_be64(u64_crq[1])); 4404 4405 if (rc) { 4406 if (rc == H_CLOSED) { 4407 dev_warn(dev, "CRQ Queue closed\n"); 4408 /* do not reset, report the fail, wait for passive init from server */ 4409 } 4410 4411 dev_warn(dev, "Send error (rc=%d)\n", rc); 4412 } 4413 4414 return rc; 4415 } 4416 4417 static int ibmvnic_send_crq_init(struct ibmvnic_adapter *adapter) 4418 { 4419 struct device *dev = &adapter->vdev->dev; 4420 union ibmvnic_crq crq; 4421 int retries = 100; 4422 int rc; 4423 4424 memset(&crq, 0, sizeof(crq)); 4425 crq.generic.first = IBMVNIC_CRQ_INIT_CMD; 4426 crq.generic.cmd = IBMVNIC_CRQ_INIT; 4427 netdev_dbg(adapter->netdev, "Sending CRQ init\n"); 4428 4429 do { 4430 rc = ibmvnic_send_crq(adapter, &crq); 4431 if (rc != H_CLOSED) 4432 break; 4433 retries--; 4434 msleep(50); 4435 4436 } while (retries > 0); 4437 4438 if (rc) { 4439 dev_err(dev, "Failed to send init request, rc = %d\n", rc); 4440 return rc; 4441 } 4442 4443 return 0; 4444 } 4445 4446 struct vnic_login_client_data { 4447 u8 type; 4448 __be16 len; 4449 char name[]; 4450 } __packed; 4451 4452 static int vnic_client_data_len(struct ibmvnic_adapter *adapter) 4453 { 4454 int len; 4455 4456 /* Calculate the amount of buffer space needed for the 4457 * vnic client data in the login buffer. There are four entries, 4458 * OS name, LPAR name, device name, and a null last entry. 4459 */ 4460 len = 4 * sizeof(struct vnic_login_client_data); 4461 len += 6; /* "Linux" plus NULL */ 4462 len += strlen(utsname()->nodename) + 1; 4463 len += strlen(adapter->netdev->name) + 1; 4464 4465 return len; 4466 } 4467 4468 static void vnic_add_client_data(struct ibmvnic_adapter *adapter, 4469 struct vnic_login_client_data *vlcd) 4470 { 4471 const char *os_name = "Linux"; 4472 int len; 4473 4474 /* Type 1 - LPAR OS */ 4475 vlcd->type = 1; 4476 len = strlen(os_name) + 1; 4477 vlcd->len = cpu_to_be16(len); 4478 strscpy(vlcd->name, os_name, len); 4479 vlcd = (struct vnic_login_client_data *)(vlcd->name + len); 4480 4481 /* Type 2 - LPAR name */ 4482 vlcd->type = 2; 4483 len = strlen(utsname()->nodename) + 1; 4484 vlcd->len = cpu_to_be16(len); 4485 strscpy(vlcd->name, utsname()->nodename, len); 4486 vlcd = (struct vnic_login_client_data *)(vlcd->name + len); 4487 4488 /* Type 3 - device name */ 4489 vlcd->type = 3; 4490 len = strlen(adapter->netdev->name) + 1; 4491 vlcd->len = cpu_to_be16(len); 4492 strscpy(vlcd->name, adapter->netdev->name, len); 4493 } 4494 4495 static int send_login(struct ibmvnic_adapter *adapter) 4496 { 4497 struct ibmvnic_login_rsp_buffer *login_rsp_buffer; 4498 struct ibmvnic_login_buffer *login_buffer; 4499 struct device *dev = &adapter->vdev->dev; 4500 struct vnic_login_client_data *vlcd; 4501 dma_addr_t rsp_buffer_token; 4502 dma_addr_t buffer_token; 4503 size_t rsp_buffer_size; 4504 union ibmvnic_crq crq; 4505 int client_data_len; 4506 size_t buffer_size; 4507 __be64 *tx_list_p; 4508 __be64 *rx_list_p; 4509 int rc; 4510 int i; 4511 4512 if (!adapter->tx_scrq || !adapter->rx_scrq) { 4513 netdev_err(adapter->netdev, 4514 "RX or TX queues are not allocated, device login failed\n"); 4515 return -ENOMEM; 4516 } 4517 4518 release_login_buffer(adapter); 4519 release_login_rsp_buffer(adapter); 4520 4521 client_data_len = vnic_client_data_len(adapter); 4522 4523 buffer_size = 4524 sizeof(struct ibmvnic_login_buffer) + 4525 sizeof(u64) * (adapter->req_tx_queues + adapter->req_rx_queues) + 4526 client_data_len; 4527 4528 login_buffer = kzalloc(buffer_size, GFP_ATOMIC); 4529 if (!login_buffer) 4530 goto buf_alloc_failed; 4531 4532 buffer_token = dma_map_single(dev, login_buffer, buffer_size, 4533 DMA_TO_DEVICE); 4534 if (dma_mapping_error(dev, buffer_token)) { 4535 dev_err(dev, "Couldn't map login buffer\n"); 4536 goto buf_map_failed; 4537 } 4538 4539 rsp_buffer_size = sizeof(struct ibmvnic_login_rsp_buffer) + 4540 sizeof(u64) * adapter->req_tx_queues + 4541 sizeof(u64) * adapter->req_rx_queues + 4542 sizeof(u64) * adapter->req_rx_queues + 4543 sizeof(u8) * IBMVNIC_TX_DESC_VERSIONS; 4544 4545 login_rsp_buffer = kmalloc(rsp_buffer_size, GFP_ATOMIC); 4546 if (!login_rsp_buffer) 4547 goto buf_rsp_alloc_failed; 4548 4549 rsp_buffer_token = dma_map_single(dev, login_rsp_buffer, 4550 rsp_buffer_size, DMA_FROM_DEVICE); 4551 if (dma_mapping_error(dev, rsp_buffer_token)) { 4552 dev_err(dev, "Couldn't map login rsp buffer\n"); 4553 goto buf_rsp_map_failed; 4554 } 4555 4556 adapter->login_buf = login_buffer; 4557 adapter->login_buf_token = buffer_token; 4558 adapter->login_buf_sz = buffer_size; 4559 adapter->login_rsp_buf = login_rsp_buffer; 4560 adapter->login_rsp_buf_token = rsp_buffer_token; 4561 adapter->login_rsp_buf_sz = rsp_buffer_size; 4562 4563 login_buffer->len = cpu_to_be32(buffer_size); 4564 login_buffer->version = cpu_to_be32(INITIAL_VERSION_LB); 4565 login_buffer->num_txcomp_subcrqs = cpu_to_be32(adapter->req_tx_queues); 4566 login_buffer->off_txcomp_subcrqs = 4567 cpu_to_be32(sizeof(struct ibmvnic_login_buffer)); 4568 login_buffer->num_rxcomp_subcrqs = cpu_to_be32(adapter->req_rx_queues); 4569 login_buffer->off_rxcomp_subcrqs = 4570 cpu_to_be32(sizeof(struct ibmvnic_login_buffer) + 4571 sizeof(u64) * adapter->req_tx_queues); 4572 login_buffer->login_rsp_ioba = cpu_to_be32(rsp_buffer_token); 4573 login_buffer->login_rsp_len = cpu_to_be32(rsp_buffer_size); 4574 4575 tx_list_p = (__be64 *)((char *)login_buffer + 4576 sizeof(struct ibmvnic_login_buffer)); 4577 rx_list_p = (__be64 *)((char *)login_buffer + 4578 sizeof(struct ibmvnic_login_buffer) + 4579 sizeof(u64) * adapter->req_tx_queues); 4580 4581 for (i = 0; i < adapter->req_tx_queues; i++) { 4582 if (adapter->tx_scrq[i]) { 4583 tx_list_p[i] = 4584 cpu_to_be64(adapter->tx_scrq[i]->crq_num); 4585 } 4586 } 4587 4588 for (i = 0; i < adapter->req_rx_queues; i++) { 4589 if (adapter->rx_scrq[i]) { 4590 rx_list_p[i] = 4591 cpu_to_be64(adapter->rx_scrq[i]->crq_num); 4592 } 4593 } 4594 4595 /* Insert vNIC login client data */ 4596 vlcd = (struct vnic_login_client_data *) 4597 ((char *)rx_list_p + (sizeof(u64) * adapter->req_rx_queues)); 4598 login_buffer->client_data_offset = 4599 cpu_to_be32((char *)vlcd - (char *)login_buffer); 4600 login_buffer->client_data_len = cpu_to_be32(client_data_len); 4601 4602 vnic_add_client_data(adapter, vlcd); 4603 4604 netdev_dbg(adapter->netdev, "Login Buffer:\n"); 4605 for (i = 0; i < (adapter->login_buf_sz - 1) / 8 + 1; i++) { 4606 netdev_dbg(adapter->netdev, "%016lx\n", 4607 ((unsigned long *)(adapter->login_buf))[i]); 4608 } 4609 4610 memset(&crq, 0, sizeof(crq)); 4611 crq.login.first = IBMVNIC_CRQ_CMD; 4612 crq.login.cmd = LOGIN; 4613 crq.login.ioba = cpu_to_be32(buffer_token); 4614 crq.login.len = cpu_to_be32(buffer_size); 4615 4616 adapter->login_pending = true; 4617 rc = ibmvnic_send_crq(adapter, &crq); 4618 if (rc) { 4619 adapter->login_pending = false; 4620 netdev_err(adapter->netdev, "Failed to send login, rc=%d\n", rc); 4621 goto buf_rsp_map_failed; 4622 } 4623 4624 return 0; 4625 4626 buf_rsp_map_failed: 4627 kfree(login_rsp_buffer); 4628 adapter->login_rsp_buf = NULL; 4629 buf_rsp_alloc_failed: 4630 dma_unmap_single(dev, buffer_token, buffer_size, DMA_TO_DEVICE); 4631 buf_map_failed: 4632 kfree(login_buffer); 4633 adapter->login_buf = NULL; 4634 buf_alloc_failed: 4635 return -ENOMEM; 4636 } 4637 4638 static int send_request_map(struct ibmvnic_adapter *adapter, dma_addr_t addr, 4639 u32 len, u8 map_id) 4640 { 4641 union ibmvnic_crq crq; 4642 4643 memset(&crq, 0, sizeof(crq)); 4644 crq.request_map.first = IBMVNIC_CRQ_CMD; 4645 crq.request_map.cmd = REQUEST_MAP; 4646 crq.request_map.map_id = map_id; 4647 crq.request_map.ioba = cpu_to_be32(addr); 4648 crq.request_map.len = cpu_to_be32(len); 4649 return ibmvnic_send_crq(adapter, &crq); 4650 } 4651 4652 static int send_request_unmap(struct ibmvnic_adapter *adapter, u8 map_id) 4653 { 4654 union ibmvnic_crq crq; 4655 4656 memset(&crq, 0, sizeof(crq)); 4657 crq.request_unmap.first = IBMVNIC_CRQ_CMD; 4658 crq.request_unmap.cmd = REQUEST_UNMAP; 4659 crq.request_unmap.map_id = map_id; 4660 return ibmvnic_send_crq(adapter, &crq); 4661 } 4662 4663 static void send_query_map(struct ibmvnic_adapter *adapter) 4664 { 4665 union ibmvnic_crq crq; 4666 4667 memset(&crq, 0, sizeof(crq)); 4668 crq.query_map.first = IBMVNIC_CRQ_CMD; 4669 crq.query_map.cmd = QUERY_MAP; 4670 ibmvnic_send_crq(adapter, &crq); 4671 } 4672 4673 /* Send a series of CRQs requesting various capabilities of the VNIC server */ 4674 static void send_query_cap(struct ibmvnic_adapter *adapter) 4675 { 4676 union ibmvnic_crq crq; 4677 int cap_reqs; 4678 4679 /* We send out 25 QUERY_CAPABILITY CRQs below. Initialize this count 4680 * upfront. When the tasklet receives a response to all of these, it 4681 * can send out the next protocol messaage (REQUEST_CAPABILITY). 4682 */ 4683 cap_reqs = 25; 4684 4685 atomic_set(&adapter->running_cap_crqs, cap_reqs); 4686 4687 memset(&crq, 0, sizeof(crq)); 4688 crq.query_capability.first = IBMVNIC_CRQ_CMD; 4689 crq.query_capability.cmd = QUERY_CAPABILITY; 4690 4691 crq.query_capability.capability = cpu_to_be16(MIN_TX_QUEUES); 4692 ibmvnic_send_crq(adapter, &crq); 4693 cap_reqs--; 4694 4695 crq.query_capability.capability = cpu_to_be16(MIN_RX_QUEUES); 4696 ibmvnic_send_crq(adapter, &crq); 4697 cap_reqs--; 4698 4699 crq.query_capability.capability = cpu_to_be16(MIN_RX_ADD_QUEUES); 4700 ibmvnic_send_crq(adapter, &crq); 4701 cap_reqs--; 4702 4703 crq.query_capability.capability = cpu_to_be16(MAX_TX_QUEUES); 4704 ibmvnic_send_crq(adapter, &crq); 4705 cap_reqs--; 4706 4707 crq.query_capability.capability = cpu_to_be16(MAX_RX_QUEUES); 4708 ibmvnic_send_crq(adapter, &crq); 4709 cap_reqs--; 4710 4711 crq.query_capability.capability = cpu_to_be16(MAX_RX_ADD_QUEUES); 4712 ibmvnic_send_crq(adapter, &crq); 4713 cap_reqs--; 4714 4715 crq.query_capability.capability = 4716 cpu_to_be16(MIN_TX_ENTRIES_PER_SUBCRQ); 4717 ibmvnic_send_crq(adapter, &crq); 4718 cap_reqs--; 4719 4720 crq.query_capability.capability = 4721 cpu_to_be16(MIN_RX_ADD_ENTRIES_PER_SUBCRQ); 4722 ibmvnic_send_crq(adapter, &crq); 4723 cap_reqs--; 4724 4725 crq.query_capability.capability = 4726 cpu_to_be16(MAX_TX_ENTRIES_PER_SUBCRQ); 4727 ibmvnic_send_crq(adapter, &crq); 4728 cap_reqs--; 4729 4730 crq.query_capability.capability = 4731 cpu_to_be16(MAX_RX_ADD_ENTRIES_PER_SUBCRQ); 4732 ibmvnic_send_crq(adapter, &crq); 4733 cap_reqs--; 4734 4735 crq.query_capability.capability = cpu_to_be16(TCP_IP_OFFLOAD); 4736 ibmvnic_send_crq(adapter, &crq); 4737 cap_reqs--; 4738 4739 crq.query_capability.capability = cpu_to_be16(PROMISC_SUPPORTED); 4740 ibmvnic_send_crq(adapter, &crq); 4741 cap_reqs--; 4742 4743 crq.query_capability.capability = cpu_to_be16(MIN_MTU); 4744 ibmvnic_send_crq(adapter, &crq); 4745 cap_reqs--; 4746 4747 crq.query_capability.capability = cpu_to_be16(MAX_MTU); 4748 ibmvnic_send_crq(adapter, &crq); 4749 cap_reqs--; 4750 4751 crq.query_capability.capability = cpu_to_be16(MAX_MULTICAST_FILTERS); 4752 ibmvnic_send_crq(adapter, &crq); 4753 cap_reqs--; 4754 4755 crq.query_capability.capability = cpu_to_be16(VLAN_HEADER_INSERTION); 4756 ibmvnic_send_crq(adapter, &crq); 4757 cap_reqs--; 4758 4759 crq.query_capability.capability = cpu_to_be16(RX_VLAN_HEADER_INSERTION); 4760 ibmvnic_send_crq(adapter, &crq); 4761 cap_reqs--; 4762 4763 crq.query_capability.capability = cpu_to_be16(MAX_TX_SG_ENTRIES); 4764 ibmvnic_send_crq(adapter, &crq); 4765 cap_reqs--; 4766 4767 crq.query_capability.capability = cpu_to_be16(RX_SG_SUPPORTED); 4768 ibmvnic_send_crq(adapter, &crq); 4769 cap_reqs--; 4770 4771 crq.query_capability.capability = cpu_to_be16(OPT_TX_COMP_SUB_QUEUES); 4772 ibmvnic_send_crq(adapter, &crq); 4773 cap_reqs--; 4774 4775 crq.query_capability.capability = cpu_to_be16(OPT_RX_COMP_QUEUES); 4776 ibmvnic_send_crq(adapter, &crq); 4777 cap_reqs--; 4778 4779 crq.query_capability.capability = 4780 cpu_to_be16(OPT_RX_BUFADD_Q_PER_RX_COMP_Q); 4781 ibmvnic_send_crq(adapter, &crq); 4782 cap_reqs--; 4783 4784 crq.query_capability.capability = 4785 cpu_to_be16(OPT_TX_ENTRIES_PER_SUBCRQ); 4786 ibmvnic_send_crq(adapter, &crq); 4787 cap_reqs--; 4788 4789 crq.query_capability.capability = 4790 cpu_to_be16(OPT_RXBA_ENTRIES_PER_SUBCRQ); 4791 ibmvnic_send_crq(adapter, &crq); 4792 cap_reqs--; 4793 4794 crq.query_capability.capability = cpu_to_be16(TX_RX_DESC_REQ); 4795 4796 ibmvnic_send_crq(adapter, &crq); 4797 cap_reqs--; 4798 4799 /* Keep at end to catch any discrepancy between expected and actual 4800 * CRQs sent. 4801 */ 4802 WARN_ON(cap_reqs != 0); 4803 } 4804 4805 static void send_query_ip_offload(struct ibmvnic_adapter *adapter) 4806 { 4807 int buf_sz = sizeof(struct ibmvnic_query_ip_offload_buffer); 4808 struct device *dev = &adapter->vdev->dev; 4809 union ibmvnic_crq crq; 4810 4811 adapter->ip_offload_tok = 4812 dma_map_single(dev, 4813 &adapter->ip_offload_buf, 4814 buf_sz, 4815 DMA_FROM_DEVICE); 4816 4817 if (dma_mapping_error(dev, adapter->ip_offload_tok)) { 4818 if (!firmware_has_feature(FW_FEATURE_CMO)) 4819 dev_err(dev, "Couldn't map offload buffer\n"); 4820 return; 4821 } 4822 4823 memset(&crq, 0, sizeof(crq)); 4824 crq.query_ip_offload.first = IBMVNIC_CRQ_CMD; 4825 crq.query_ip_offload.cmd = QUERY_IP_OFFLOAD; 4826 crq.query_ip_offload.len = cpu_to_be32(buf_sz); 4827 crq.query_ip_offload.ioba = 4828 cpu_to_be32(adapter->ip_offload_tok); 4829 4830 ibmvnic_send_crq(adapter, &crq); 4831 } 4832 4833 static void send_control_ip_offload(struct ibmvnic_adapter *adapter) 4834 { 4835 struct ibmvnic_control_ip_offload_buffer *ctrl_buf = &adapter->ip_offload_ctrl; 4836 struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf; 4837 struct device *dev = &adapter->vdev->dev; 4838 netdev_features_t old_hw_features = 0; 4839 union ibmvnic_crq crq; 4840 4841 adapter->ip_offload_ctrl_tok = 4842 dma_map_single(dev, 4843 ctrl_buf, 4844 sizeof(adapter->ip_offload_ctrl), 4845 DMA_TO_DEVICE); 4846 4847 if (dma_mapping_error(dev, adapter->ip_offload_ctrl_tok)) { 4848 dev_err(dev, "Couldn't map ip offload control buffer\n"); 4849 return; 4850 } 4851 4852 ctrl_buf->len = cpu_to_be32(sizeof(adapter->ip_offload_ctrl)); 4853 ctrl_buf->version = cpu_to_be32(INITIAL_VERSION_IOB); 4854 ctrl_buf->ipv4_chksum = buf->ipv4_chksum; 4855 ctrl_buf->ipv6_chksum = buf->ipv6_chksum; 4856 ctrl_buf->tcp_ipv4_chksum = buf->tcp_ipv4_chksum; 4857 ctrl_buf->udp_ipv4_chksum = buf->udp_ipv4_chksum; 4858 ctrl_buf->tcp_ipv6_chksum = buf->tcp_ipv6_chksum; 4859 ctrl_buf->udp_ipv6_chksum = buf->udp_ipv6_chksum; 4860 ctrl_buf->large_tx_ipv4 = buf->large_tx_ipv4; 4861 ctrl_buf->large_tx_ipv6 = buf->large_tx_ipv6; 4862 4863 /* large_rx disabled for now, additional features needed */ 4864 ctrl_buf->large_rx_ipv4 = 0; 4865 ctrl_buf->large_rx_ipv6 = 0; 4866 4867 if (adapter->state != VNIC_PROBING) { 4868 old_hw_features = adapter->netdev->hw_features; 4869 adapter->netdev->hw_features = 0; 4870 } 4871 4872 adapter->netdev->hw_features = NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO; 4873 4874 if (buf->tcp_ipv4_chksum || buf->udp_ipv4_chksum) 4875 adapter->netdev->hw_features |= NETIF_F_IP_CSUM; 4876 4877 if (buf->tcp_ipv6_chksum || buf->udp_ipv6_chksum) 4878 adapter->netdev->hw_features |= NETIF_F_IPV6_CSUM; 4879 4880 if ((adapter->netdev->features & 4881 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM))) 4882 adapter->netdev->hw_features |= NETIF_F_RXCSUM; 4883 4884 if (buf->large_tx_ipv4) 4885 adapter->netdev->hw_features |= NETIF_F_TSO; 4886 if (buf->large_tx_ipv6) 4887 adapter->netdev->hw_features |= NETIF_F_TSO6; 4888 4889 if (adapter->state == VNIC_PROBING) { 4890 adapter->netdev->features |= adapter->netdev->hw_features; 4891 } else if (old_hw_features != adapter->netdev->hw_features) { 4892 netdev_features_t tmp = 0; 4893 4894 /* disable features no longer supported */ 4895 adapter->netdev->features &= adapter->netdev->hw_features; 4896 /* turn on features now supported if previously enabled */ 4897 tmp = (old_hw_features ^ adapter->netdev->hw_features) & 4898 adapter->netdev->hw_features; 4899 adapter->netdev->features |= 4900 tmp & adapter->netdev->wanted_features; 4901 } 4902 4903 memset(&crq, 0, sizeof(crq)); 4904 crq.control_ip_offload.first = IBMVNIC_CRQ_CMD; 4905 crq.control_ip_offload.cmd = CONTROL_IP_OFFLOAD; 4906 crq.control_ip_offload.len = 4907 cpu_to_be32(sizeof(adapter->ip_offload_ctrl)); 4908 crq.control_ip_offload.ioba = cpu_to_be32(adapter->ip_offload_ctrl_tok); 4909 ibmvnic_send_crq(adapter, &crq); 4910 } 4911 4912 static void handle_vpd_size_rsp(union ibmvnic_crq *crq, 4913 struct ibmvnic_adapter *adapter) 4914 { 4915 struct device *dev = &adapter->vdev->dev; 4916 4917 if (crq->get_vpd_size_rsp.rc.code) { 4918 dev_err(dev, "Error retrieving VPD size, rc=%x\n", 4919 crq->get_vpd_size_rsp.rc.code); 4920 complete(&adapter->fw_done); 4921 return; 4922 } 4923 4924 adapter->vpd->len = be64_to_cpu(crq->get_vpd_size_rsp.len); 4925 complete(&adapter->fw_done); 4926 } 4927 4928 static void handle_vpd_rsp(union ibmvnic_crq *crq, 4929 struct ibmvnic_adapter *adapter) 4930 { 4931 struct device *dev = &adapter->vdev->dev; 4932 unsigned char *substr = NULL; 4933 u8 fw_level_len = 0; 4934 4935 memset(adapter->fw_version, 0, 32); 4936 4937 dma_unmap_single(dev, adapter->vpd->dma_addr, adapter->vpd->len, 4938 DMA_FROM_DEVICE); 4939 4940 if (crq->get_vpd_rsp.rc.code) { 4941 dev_err(dev, "Error retrieving VPD from device, rc=%x\n", 4942 crq->get_vpd_rsp.rc.code); 4943 goto complete; 4944 } 4945 4946 /* get the position of the firmware version info 4947 * located after the ASCII 'RM' substring in the buffer 4948 */ 4949 substr = strnstr(adapter->vpd->buff, "RM", adapter->vpd->len); 4950 if (!substr) { 4951 dev_info(dev, "Warning - No FW level has been provided in the VPD buffer by the VIOS Server\n"); 4952 goto complete; 4953 } 4954 4955 /* get length of firmware level ASCII substring */ 4956 if ((substr + 2) < (adapter->vpd->buff + adapter->vpd->len)) { 4957 fw_level_len = *(substr + 2); 4958 } else { 4959 dev_info(dev, "Length of FW substr extrapolated VDP buff\n"); 4960 goto complete; 4961 } 4962 4963 /* copy firmware version string from vpd into adapter */ 4964 if ((substr + 3 + fw_level_len) < 4965 (adapter->vpd->buff + adapter->vpd->len)) { 4966 strncpy((char *)adapter->fw_version, substr + 3, fw_level_len); 4967 } else { 4968 dev_info(dev, "FW substr extrapolated VPD buff\n"); 4969 } 4970 4971 complete: 4972 if (adapter->fw_version[0] == '\0') 4973 strscpy((char *)adapter->fw_version, "N/A", sizeof(adapter->fw_version)); 4974 complete(&adapter->fw_done); 4975 } 4976 4977 static void handle_query_ip_offload_rsp(struct ibmvnic_adapter *adapter) 4978 { 4979 struct device *dev = &adapter->vdev->dev; 4980 struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf; 4981 int i; 4982 4983 dma_unmap_single(dev, adapter->ip_offload_tok, 4984 sizeof(adapter->ip_offload_buf), DMA_FROM_DEVICE); 4985 4986 netdev_dbg(adapter->netdev, "Query IP Offload Buffer:\n"); 4987 for (i = 0; i < (sizeof(adapter->ip_offload_buf) - 1) / 8 + 1; i++) 4988 netdev_dbg(adapter->netdev, "%016lx\n", 4989 ((unsigned long *)(buf))[i]); 4990 4991 netdev_dbg(adapter->netdev, "ipv4_chksum = %d\n", buf->ipv4_chksum); 4992 netdev_dbg(adapter->netdev, "ipv6_chksum = %d\n", buf->ipv6_chksum); 4993 netdev_dbg(adapter->netdev, "tcp_ipv4_chksum = %d\n", 4994 buf->tcp_ipv4_chksum); 4995 netdev_dbg(adapter->netdev, "tcp_ipv6_chksum = %d\n", 4996 buf->tcp_ipv6_chksum); 4997 netdev_dbg(adapter->netdev, "udp_ipv4_chksum = %d\n", 4998 buf->udp_ipv4_chksum); 4999 netdev_dbg(adapter->netdev, "udp_ipv6_chksum = %d\n", 5000 buf->udp_ipv6_chksum); 5001 netdev_dbg(adapter->netdev, "large_tx_ipv4 = %d\n", 5002 buf->large_tx_ipv4); 5003 netdev_dbg(adapter->netdev, "large_tx_ipv6 = %d\n", 5004 buf->large_tx_ipv6); 5005 netdev_dbg(adapter->netdev, "large_rx_ipv4 = %d\n", 5006 buf->large_rx_ipv4); 5007 netdev_dbg(adapter->netdev, "large_rx_ipv6 = %d\n", 5008 buf->large_rx_ipv6); 5009 netdev_dbg(adapter->netdev, "max_ipv4_hdr_sz = %d\n", 5010 buf->max_ipv4_header_size); 5011 netdev_dbg(adapter->netdev, "max_ipv6_hdr_sz = %d\n", 5012 buf->max_ipv6_header_size); 5013 netdev_dbg(adapter->netdev, "max_tcp_hdr_size = %d\n", 5014 buf->max_tcp_header_size); 5015 netdev_dbg(adapter->netdev, "max_udp_hdr_size = %d\n", 5016 buf->max_udp_header_size); 5017 netdev_dbg(adapter->netdev, "max_large_tx_size = %d\n", 5018 buf->max_large_tx_size); 5019 netdev_dbg(adapter->netdev, "max_large_rx_size = %d\n", 5020 buf->max_large_rx_size); 5021 netdev_dbg(adapter->netdev, "ipv6_ext_hdr = %d\n", 5022 buf->ipv6_extension_header); 5023 netdev_dbg(adapter->netdev, "tcp_pseudosum_req = %d\n", 5024 buf->tcp_pseudosum_req); 5025 netdev_dbg(adapter->netdev, "num_ipv6_ext_hd = %d\n", 5026 buf->num_ipv6_ext_headers); 5027 netdev_dbg(adapter->netdev, "off_ipv6_ext_hd = %d\n", 5028 buf->off_ipv6_ext_headers); 5029 5030 send_control_ip_offload(adapter); 5031 } 5032 5033 static const char *ibmvnic_fw_err_cause(u16 cause) 5034 { 5035 switch (cause) { 5036 case ADAPTER_PROBLEM: 5037 return "adapter problem"; 5038 case BUS_PROBLEM: 5039 return "bus problem"; 5040 case FW_PROBLEM: 5041 return "firmware problem"; 5042 case DD_PROBLEM: 5043 return "device driver problem"; 5044 case EEH_RECOVERY: 5045 return "EEH recovery"; 5046 case FW_UPDATED: 5047 return "firmware updated"; 5048 case LOW_MEMORY: 5049 return "low Memory"; 5050 default: 5051 return "unknown"; 5052 } 5053 } 5054 5055 static void handle_error_indication(union ibmvnic_crq *crq, 5056 struct ibmvnic_adapter *adapter) 5057 { 5058 struct device *dev = &adapter->vdev->dev; 5059 u16 cause; 5060 5061 cause = be16_to_cpu(crq->error_indication.error_cause); 5062 5063 dev_warn_ratelimited(dev, 5064 "Firmware reports %serror, cause: %s. Starting recovery...\n", 5065 crq->error_indication.flags 5066 & IBMVNIC_FATAL_ERROR ? "FATAL " : "", 5067 ibmvnic_fw_err_cause(cause)); 5068 5069 if (crq->error_indication.flags & IBMVNIC_FATAL_ERROR) 5070 ibmvnic_reset(adapter, VNIC_RESET_FATAL); 5071 else 5072 ibmvnic_reset(adapter, VNIC_RESET_NON_FATAL); 5073 } 5074 5075 static int handle_change_mac_rsp(union ibmvnic_crq *crq, 5076 struct ibmvnic_adapter *adapter) 5077 { 5078 struct net_device *netdev = adapter->netdev; 5079 struct device *dev = &adapter->vdev->dev; 5080 long rc; 5081 5082 rc = crq->change_mac_addr_rsp.rc.code; 5083 if (rc) { 5084 dev_err(dev, "Error %ld in CHANGE_MAC_ADDR_RSP\n", rc); 5085 goto out; 5086 } 5087 /* crq->change_mac_addr.mac_addr is the requested one 5088 * crq->change_mac_addr_rsp.mac_addr is the returned valid one. 5089 */ 5090 eth_hw_addr_set(netdev, &crq->change_mac_addr_rsp.mac_addr[0]); 5091 ether_addr_copy(adapter->mac_addr, 5092 &crq->change_mac_addr_rsp.mac_addr[0]); 5093 out: 5094 complete(&adapter->fw_done); 5095 return rc; 5096 } 5097 5098 static void handle_request_cap_rsp(union ibmvnic_crq *crq, 5099 struct ibmvnic_adapter *adapter) 5100 { 5101 struct device *dev = &adapter->vdev->dev; 5102 u64 *req_value; 5103 char *name; 5104 5105 atomic_dec(&adapter->running_cap_crqs); 5106 netdev_dbg(adapter->netdev, "Outstanding request-caps: %d\n", 5107 atomic_read(&adapter->running_cap_crqs)); 5108 switch (be16_to_cpu(crq->request_capability_rsp.capability)) { 5109 case REQ_TX_QUEUES: 5110 req_value = &adapter->req_tx_queues; 5111 name = "tx"; 5112 break; 5113 case REQ_RX_QUEUES: 5114 req_value = &adapter->req_rx_queues; 5115 name = "rx"; 5116 break; 5117 case REQ_RX_ADD_QUEUES: 5118 req_value = &adapter->req_rx_add_queues; 5119 name = "rx_add"; 5120 break; 5121 case REQ_TX_ENTRIES_PER_SUBCRQ: 5122 req_value = &adapter->req_tx_entries_per_subcrq; 5123 name = "tx_entries_per_subcrq"; 5124 break; 5125 case REQ_RX_ADD_ENTRIES_PER_SUBCRQ: 5126 req_value = &adapter->req_rx_add_entries_per_subcrq; 5127 name = "rx_add_entries_per_subcrq"; 5128 break; 5129 case REQ_MTU: 5130 req_value = &adapter->req_mtu; 5131 name = "mtu"; 5132 break; 5133 case PROMISC_REQUESTED: 5134 req_value = &adapter->promisc; 5135 name = "promisc"; 5136 break; 5137 default: 5138 dev_err(dev, "Got invalid cap request rsp %d\n", 5139 crq->request_capability.capability); 5140 return; 5141 } 5142 5143 switch (crq->request_capability_rsp.rc.code) { 5144 case SUCCESS: 5145 break; 5146 case PARTIALSUCCESS: 5147 dev_info(dev, "req=%lld, rsp=%ld in %s queue, retrying.\n", 5148 *req_value, 5149 (long)be64_to_cpu(crq->request_capability_rsp.number), 5150 name); 5151 5152 if (be16_to_cpu(crq->request_capability_rsp.capability) == 5153 REQ_MTU) { 5154 pr_err("mtu of %llu is not supported. Reverting.\n", 5155 *req_value); 5156 *req_value = adapter->fallback.mtu; 5157 } else { 5158 *req_value = 5159 be64_to_cpu(crq->request_capability_rsp.number); 5160 } 5161 5162 send_request_cap(adapter, 1); 5163 return; 5164 default: 5165 dev_err(dev, "Error %d in request cap rsp\n", 5166 crq->request_capability_rsp.rc.code); 5167 return; 5168 } 5169 5170 /* Done receiving requested capabilities, query IP offload support */ 5171 if (atomic_read(&adapter->running_cap_crqs) == 0) 5172 send_query_ip_offload(adapter); 5173 } 5174 5175 static int handle_login_rsp(union ibmvnic_crq *login_rsp_crq, 5176 struct ibmvnic_adapter *adapter) 5177 { 5178 struct device *dev = &adapter->vdev->dev; 5179 struct net_device *netdev = adapter->netdev; 5180 struct ibmvnic_login_rsp_buffer *login_rsp = adapter->login_rsp_buf; 5181 struct ibmvnic_login_buffer *login = adapter->login_buf; 5182 u64 *tx_handle_array; 5183 u64 *rx_handle_array; 5184 int num_tx_pools; 5185 int num_rx_pools; 5186 u64 *size_array; 5187 int i; 5188 5189 /* CHECK: Test/set of login_pending does not need to be atomic 5190 * because only ibmvnic_tasklet tests/clears this. 5191 */ 5192 if (!adapter->login_pending) { 5193 netdev_warn(netdev, "Ignoring unexpected login response\n"); 5194 return 0; 5195 } 5196 adapter->login_pending = false; 5197 5198 dma_unmap_single(dev, adapter->login_buf_token, adapter->login_buf_sz, 5199 DMA_TO_DEVICE); 5200 dma_unmap_single(dev, adapter->login_rsp_buf_token, 5201 adapter->login_rsp_buf_sz, DMA_FROM_DEVICE); 5202 5203 /* If the number of queues requested can't be allocated by the 5204 * server, the login response will return with code 1. We will need 5205 * to resend the login buffer with fewer queues requested. 5206 */ 5207 if (login_rsp_crq->generic.rc.code) { 5208 adapter->init_done_rc = login_rsp_crq->generic.rc.code; 5209 complete(&adapter->init_done); 5210 return 0; 5211 } 5212 5213 if (adapter->failover_pending) { 5214 adapter->init_done_rc = -EAGAIN; 5215 netdev_dbg(netdev, "Failover pending, ignoring login response\n"); 5216 complete(&adapter->init_done); 5217 /* login response buffer will be released on reset */ 5218 return 0; 5219 } 5220 5221 netdev->mtu = adapter->req_mtu - ETH_HLEN; 5222 5223 netdev_dbg(adapter->netdev, "Login Response Buffer:\n"); 5224 for (i = 0; i < (adapter->login_rsp_buf_sz - 1) / 8 + 1; i++) { 5225 netdev_dbg(adapter->netdev, "%016lx\n", 5226 ((unsigned long *)(adapter->login_rsp_buf))[i]); 5227 } 5228 5229 /* Sanity checks */ 5230 if (login->num_txcomp_subcrqs != login_rsp->num_txsubm_subcrqs || 5231 (be32_to_cpu(login->num_rxcomp_subcrqs) * 5232 adapter->req_rx_add_queues != 5233 be32_to_cpu(login_rsp->num_rxadd_subcrqs))) { 5234 dev_err(dev, "FATAL: Inconsistent login and login rsp\n"); 5235 ibmvnic_reset(adapter, VNIC_RESET_FATAL); 5236 return -EIO; 5237 } 5238 size_array = (u64 *)((u8 *)(adapter->login_rsp_buf) + 5239 be32_to_cpu(adapter->login_rsp_buf->off_rxadd_buff_size)); 5240 /* variable buffer sizes are not supported, so just read the 5241 * first entry. 5242 */ 5243 adapter->cur_rx_buf_sz = be64_to_cpu(size_array[0]); 5244 5245 num_tx_pools = be32_to_cpu(adapter->login_rsp_buf->num_txsubm_subcrqs); 5246 num_rx_pools = be32_to_cpu(adapter->login_rsp_buf->num_rxadd_subcrqs); 5247 5248 tx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) + 5249 be32_to_cpu(adapter->login_rsp_buf->off_txsubm_subcrqs)); 5250 rx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) + 5251 be32_to_cpu(adapter->login_rsp_buf->off_rxadd_subcrqs)); 5252 5253 for (i = 0; i < num_tx_pools; i++) 5254 adapter->tx_scrq[i]->handle = tx_handle_array[i]; 5255 5256 for (i = 0; i < num_rx_pools; i++) 5257 adapter->rx_scrq[i]->handle = rx_handle_array[i]; 5258 5259 adapter->num_active_tx_scrqs = num_tx_pools; 5260 adapter->num_active_rx_scrqs = num_rx_pools; 5261 release_login_rsp_buffer(adapter); 5262 release_login_buffer(adapter); 5263 complete(&adapter->init_done); 5264 5265 return 0; 5266 } 5267 5268 static void handle_request_unmap_rsp(union ibmvnic_crq *crq, 5269 struct ibmvnic_adapter *adapter) 5270 { 5271 struct device *dev = &adapter->vdev->dev; 5272 long rc; 5273 5274 rc = crq->request_unmap_rsp.rc.code; 5275 if (rc) 5276 dev_err(dev, "Error %ld in REQUEST_UNMAP_RSP\n", rc); 5277 } 5278 5279 static void handle_query_map_rsp(union ibmvnic_crq *crq, 5280 struct ibmvnic_adapter *adapter) 5281 { 5282 struct net_device *netdev = adapter->netdev; 5283 struct device *dev = &adapter->vdev->dev; 5284 long rc; 5285 5286 rc = crq->query_map_rsp.rc.code; 5287 if (rc) { 5288 dev_err(dev, "Error %ld in QUERY_MAP_RSP\n", rc); 5289 return; 5290 } 5291 netdev_dbg(netdev, "page_size = %d\ntot_pages = %u\nfree_pages = %u\n", 5292 crq->query_map_rsp.page_size, 5293 __be32_to_cpu(crq->query_map_rsp.tot_pages), 5294 __be32_to_cpu(crq->query_map_rsp.free_pages)); 5295 } 5296 5297 static void handle_query_cap_rsp(union ibmvnic_crq *crq, 5298 struct ibmvnic_adapter *adapter) 5299 { 5300 struct net_device *netdev = adapter->netdev; 5301 struct device *dev = &adapter->vdev->dev; 5302 long rc; 5303 5304 atomic_dec(&adapter->running_cap_crqs); 5305 netdev_dbg(netdev, "Outstanding queries: %d\n", 5306 atomic_read(&adapter->running_cap_crqs)); 5307 rc = crq->query_capability.rc.code; 5308 if (rc) { 5309 dev_err(dev, "Error %ld in QUERY_CAP_RSP\n", rc); 5310 goto out; 5311 } 5312 5313 switch (be16_to_cpu(crq->query_capability.capability)) { 5314 case MIN_TX_QUEUES: 5315 adapter->min_tx_queues = 5316 be64_to_cpu(crq->query_capability.number); 5317 netdev_dbg(netdev, "min_tx_queues = %lld\n", 5318 adapter->min_tx_queues); 5319 break; 5320 case MIN_RX_QUEUES: 5321 adapter->min_rx_queues = 5322 be64_to_cpu(crq->query_capability.number); 5323 netdev_dbg(netdev, "min_rx_queues = %lld\n", 5324 adapter->min_rx_queues); 5325 break; 5326 case MIN_RX_ADD_QUEUES: 5327 adapter->min_rx_add_queues = 5328 be64_to_cpu(crq->query_capability.number); 5329 netdev_dbg(netdev, "min_rx_add_queues = %lld\n", 5330 adapter->min_rx_add_queues); 5331 break; 5332 case MAX_TX_QUEUES: 5333 adapter->max_tx_queues = 5334 be64_to_cpu(crq->query_capability.number); 5335 netdev_dbg(netdev, "max_tx_queues = %lld\n", 5336 adapter->max_tx_queues); 5337 break; 5338 case MAX_RX_QUEUES: 5339 adapter->max_rx_queues = 5340 be64_to_cpu(crq->query_capability.number); 5341 netdev_dbg(netdev, "max_rx_queues = %lld\n", 5342 adapter->max_rx_queues); 5343 break; 5344 case MAX_RX_ADD_QUEUES: 5345 adapter->max_rx_add_queues = 5346 be64_to_cpu(crq->query_capability.number); 5347 netdev_dbg(netdev, "max_rx_add_queues = %lld\n", 5348 adapter->max_rx_add_queues); 5349 break; 5350 case MIN_TX_ENTRIES_PER_SUBCRQ: 5351 adapter->min_tx_entries_per_subcrq = 5352 be64_to_cpu(crq->query_capability.number); 5353 netdev_dbg(netdev, "min_tx_entries_per_subcrq = %lld\n", 5354 adapter->min_tx_entries_per_subcrq); 5355 break; 5356 case MIN_RX_ADD_ENTRIES_PER_SUBCRQ: 5357 adapter->min_rx_add_entries_per_subcrq = 5358 be64_to_cpu(crq->query_capability.number); 5359 netdev_dbg(netdev, "min_rx_add_entrs_per_subcrq = %lld\n", 5360 adapter->min_rx_add_entries_per_subcrq); 5361 break; 5362 case MAX_TX_ENTRIES_PER_SUBCRQ: 5363 adapter->max_tx_entries_per_subcrq = 5364 be64_to_cpu(crq->query_capability.number); 5365 netdev_dbg(netdev, "max_tx_entries_per_subcrq = %lld\n", 5366 adapter->max_tx_entries_per_subcrq); 5367 break; 5368 case MAX_RX_ADD_ENTRIES_PER_SUBCRQ: 5369 adapter->max_rx_add_entries_per_subcrq = 5370 be64_to_cpu(crq->query_capability.number); 5371 netdev_dbg(netdev, "max_rx_add_entrs_per_subcrq = %lld\n", 5372 adapter->max_rx_add_entries_per_subcrq); 5373 break; 5374 case TCP_IP_OFFLOAD: 5375 adapter->tcp_ip_offload = 5376 be64_to_cpu(crq->query_capability.number); 5377 netdev_dbg(netdev, "tcp_ip_offload = %lld\n", 5378 adapter->tcp_ip_offload); 5379 break; 5380 case PROMISC_SUPPORTED: 5381 adapter->promisc_supported = 5382 be64_to_cpu(crq->query_capability.number); 5383 netdev_dbg(netdev, "promisc_supported = %lld\n", 5384 adapter->promisc_supported); 5385 break; 5386 case MIN_MTU: 5387 adapter->min_mtu = be64_to_cpu(crq->query_capability.number); 5388 netdev->min_mtu = adapter->min_mtu - ETH_HLEN; 5389 netdev_dbg(netdev, "min_mtu = %lld\n", adapter->min_mtu); 5390 break; 5391 case MAX_MTU: 5392 adapter->max_mtu = be64_to_cpu(crq->query_capability.number); 5393 netdev->max_mtu = adapter->max_mtu - ETH_HLEN; 5394 netdev_dbg(netdev, "max_mtu = %lld\n", adapter->max_mtu); 5395 break; 5396 case MAX_MULTICAST_FILTERS: 5397 adapter->max_multicast_filters = 5398 be64_to_cpu(crq->query_capability.number); 5399 netdev_dbg(netdev, "max_multicast_filters = %lld\n", 5400 adapter->max_multicast_filters); 5401 break; 5402 case VLAN_HEADER_INSERTION: 5403 adapter->vlan_header_insertion = 5404 be64_to_cpu(crq->query_capability.number); 5405 if (adapter->vlan_header_insertion) 5406 netdev->features |= NETIF_F_HW_VLAN_STAG_TX; 5407 netdev_dbg(netdev, "vlan_header_insertion = %lld\n", 5408 adapter->vlan_header_insertion); 5409 break; 5410 case RX_VLAN_HEADER_INSERTION: 5411 adapter->rx_vlan_header_insertion = 5412 be64_to_cpu(crq->query_capability.number); 5413 netdev_dbg(netdev, "rx_vlan_header_insertion = %lld\n", 5414 adapter->rx_vlan_header_insertion); 5415 break; 5416 case MAX_TX_SG_ENTRIES: 5417 adapter->max_tx_sg_entries = 5418 be64_to_cpu(crq->query_capability.number); 5419 netdev_dbg(netdev, "max_tx_sg_entries = %lld\n", 5420 adapter->max_tx_sg_entries); 5421 break; 5422 case RX_SG_SUPPORTED: 5423 adapter->rx_sg_supported = 5424 be64_to_cpu(crq->query_capability.number); 5425 netdev_dbg(netdev, "rx_sg_supported = %lld\n", 5426 adapter->rx_sg_supported); 5427 break; 5428 case OPT_TX_COMP_SUB_QUEUES: 5429 adapter->opt_tx_comp_sub_queues = 5430 be64_to_cpu(crq->query_capability.number); 5431 netdev_dbg(netdev, "opt_tx_comp_sub_queues = %lld\n", 5432 adapter->opt_tx_comp_sub_queues); 5433 break; 5434 case OPT_RX_COMP_QUEUES: 5435 adapter->opt_rx_comp_queues = 5436 be64_to_cpu(crq->query_capability.number); 5437 netdev_dbg(netdev, "opt_rx_comp_queues = %lld\n", 5438 adapter->opt_rx_comp_queues); 5439 break; 5440 case OPT_RX_BUFADD_Q_PER_RX_COMP_Q: 5441 adapter->opt_rx_bufadd_q_per_rx_comp_q = 5442 be64_to_cpu(crq->query_capability.number); 5443 netdev_dbg(netdev, "opt_rx_bufadd_q_per_rx_comp_q = %lld\n", 5444 adapter->opt_rx_bufadd_q_per_rx_comp_q); 5445 break; 5446 case OPT_TX_ENTRIES_PER_SUBCRQ: 5447 adapter->opt_tx_entries_per_subcrq = 5448 be64_to_cpu(crq->query_capability.number); 5449 netdev_dbg(netdev, "opt_tx_entries_per_subcrq = %lld\n", 5450 adapter->opt_tx_entries_per_subcrq); 5451 break; 5452 case OPT_RXBA_ENTRIES_PER_SUBCRQ: 5453 adapter->opt_rxba_entries_per_subcrq = 5454 be64_to_cpu(crq->query_capability.number); 5455 netdev_dbg(netdev, "opt_rxba_entries_per_subcrq = %lld\n", 5456 adapter->opt_rxba_entries_per_subcrq); 5457 break; 5458 case TX_RX_DESC_REQ: 5459 adapter->tx_rx_desc_req = crq->query_capability.number; 5460 netdev_dbg(netdev, "tx_rx_desc_req = %llx\n", 5461 adapter->tx_rx_desc_req); 5462 break; 5463 5464 default: 5465 netdev_err(netdev, "Got invalid cap rsp %d\n", 5466 crq->query_capability.capability); 5467 } 5468 5469 out: 5470 if (atomic_read(&adapter->running_cap_crqs) == 0) 5471 send_request_cap(adapter, 0); 5472 } 5473 5474 static int send_query_phys_parms(struct ibmvnic_adapter *adapter) 5475 { 5476 union ibmvnic_crq crq; 5477 int rc; 5478 5479 memset(&crq, 0, sizeof(crq)); 5480 crq.query_phys_parms.first = IBMVNIC_CRQ_CMD; 5481 crq.query_phys_parms.cmd = QUERY_PHYS_PARMS; 5482 5483 mutex_lock(&adapter->fw_lock); 5484 adapter->fw_done_rc = 0; 5485 reinit_completion(&adapter->fw_done); 5486 5487 rc = ibmvnic_send_crq(adapter, &crq); 5488 if (rc) { 5489 mutex_unlock(&adapter->fw_lock); 5490 return rc; 5491 } 5492 5493 rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000); 5494 if (rc) { 5495 mutex_unlock(&adapter->fw_lock); 5496 return rc; 5497 } 5498 5499 mutex_unlock(&adapter->fw_lock); 5500 return adapter->fw_done_rc ? -EIO : 0; 5501 } 5502 5503 static int handle_query_phys_parms_rsp(union ibmvnic_crq *crq, 5504 struct ibmvnic_adapter *adapter) 5505 { 5506 struct net_device *netdev = adapter->netdev; 5507 int rc; 5508 __be32 rspeed = cpu_to_be32(crq->query_phys_parms_rsp.speed); 5509 5510 rc = crq->query_phys_parms_rsp.rc.code; 5511 if (rc) { 5512 netdev_err(netdev, "Error %d in QUERY_PHYS_PARMS\n", rc); 5513 return rc; 5514 } 5515 switch (rspeed) { 5516 case IBMVNIC_10MBPS: 5517 adapter->speed = SPEED_10; 5518 break; 5519 case IBMVNIC_100MBPS: 5520 adapter->speed = SPEED_100; 5521 break; 5522 case IBMVNIC_1GBPS: 5523 adapter->speed = SPEED_1000; 5524 break; 5525 case IBMVNIC_10GBPS: 5526 adapter->speed = SPEED_10000; 5527 break; 5528 case IBMVNIC_25GBPS: 5529 adapter->speed = SPEED_25000; 5530 break; 5531 case IBMVNIC_40GBPS: 5532 adapter->speed = SPEED_40000; 5533 break; 5534 case IBMVNIC_50GBPS: 5535 adapter->speed = SPEED_50000; 5536 break; 5537 case IBMVNIC_100GBPS: 5538 adapter->speed = SPEED_100000; 5539 break; 5540 case IBMVNIC_200GBPS: 5541 adapter->speed = SPEED_200000; 5542 break; 5543 default: 5544 if (netif_carrier_ok(netdev)) 5545 netdev_warn(netdev, "Unknown speed 0x%08x\n", rspeed); 5546 adapter->speed = SPEED_UNKNOWN; 5547 } 5548 if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_FULL_DUPLEX) 5549 adapter->duplex = DUPLEX_FULL; 5550 else if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_HALF_DUPLEX) 5551 adapter->duplex = DUPLEX_HALF; 5552 else 5553 adapter->duplex = DUPLEX_UNKNOWN; 5554 5555 return rc; 5556 } 5557 5558 static void ibmvnic_handle_crq(union ibmvnic_crq *crq, 5559 struct ibmvnic_adapter *adapter) 5560 { 5561 struct ibmvnic_generic_crq *gen_crq = &crq->generic; 5562 struct net_device *netdev = adapter->netdev; 5563 struct device *dev = &adapter->vdev->dev; 5564 u64 *u64_crq = (u64 *)crq; 5565 long rc; 5566 5567 netdev_dbg(netdev, "Handling CRQ: %016lx %016lx\n", 5568 (unsigned long)cpu_to_be64(u64_crq[0]), 5569 (unsigned long)cpu_to_be64(u64_crq[1])); 5570 switch (gen_crq->first) { 5571 case IBMVNIC_CRQ_INIT_RSP: 5572 switch (gen_crq->cmd) { 5573 case IBMVNIC_CRQ_INIT: 5574 dev_info(dev, "Partner initialized\n"); 5575 adapter->from_passive_init = true; 5576 /* Discard any stale login responses from prev reset. 5577 * CHECK: should we clear even on INIT_COMPLETE? 5578 */ 5579 adapter->login_pending = false; 5580 5581 if (adapter->state == VNIC_DOWN) 5582 rc = ibmvnic_reset(adapter, VNIC_RESET_PASSIVE_INIT); 5583 else 5584 rc = ibmvnic_reset(adapter, VNIC_RESET_FAILOVER); 5585 5586 if (rc && rc != -EBUSY) { 5587 /* We were unable to schedule the failover 5588 * reset either because the adapter was still 5589 * probing (eg: during kexec) or we could not 5590 * allocate memory. Clear the failover_pending 5591 * flag since no one else will. We ignore 5592 * EBUSY because it means either FAILOVER reset 5593 * is already scheduled or the adapter is 5594 * being removed. 5595 */ 5596 netdev_err(netdev, 5597 "Error %ld scheduling failover reset\n", 5598 rc); 5599 adapter->failover_pending = false; 5600 } 5601 5602 if (!completion_done(&adapter->init_done)) { 5603 if (!adapter->init_done_rc) 5604 adapter->init_done_rc = -EAGAIN; 5605 complete(&adapter->init_done); 5606 } 5607 5608 break; 5609 case IBMVNIC_CRQ_INIT_COMPLETE: 5610 dev_info(dev, "Partner initialization complete\n"); 5611 adapter->crq.active = true; 5612 send_version_xchg(adapter); 5613 break; 5614 default: 5615 dev_err(dev, "Unknown crq cmd: %d\n", gen_crq->cmd); 5616 } 5617 return; 5618 case IBMVNIC_CRQ_XPORT_EVENT: 5619 netif_carrier_off(netdev); 5620 adapter->crq.active = false; 5621 /* terminate any thread waiting for a response 5622 * from the device 5623 */ 5624 if (!completion_done(&adapter->fw_done)) { 5625 adapter->fw_done_rc = -EIO; 5626 complete(&adapter->fw_done); 5627 } 5628 5629 /* if we got here during crq-init, retry crq-init */ 5630 if (!completion_done(&adapter->init_done)) { 5631 adapter->init_done_rc = -EAGAIN; 5632 complete(&adapter->init_done); 5633 } 5634 5635 if (!completion_done(&adapter->stats_done)) 5636 complete(&adapter->stats_done); 5637 if (test_bit(0, &adapter->resetting)) 5638 adapter->force_reset_recovery = true; 5639 if (gen_crq->cmd == IBMVNIC_PARTITION_MIGRATED) { 5640 dev_info(dev, "Migrated, re-enabling adapter\n"); 5641 ibmvnic_reset(adapter, VNIC_RESET_MOBILITY); 5642 } else if (gen_crq->cmd == IBMVNIC_DEVICE_FAILOVER) { 5643 dev_info(dev, "Backing device failover detected\n"); 5644 adapter->failover_pending = true; 5645 } else { 5646 /* The adapter lost the connection */ 5647 dev_err(dev, "Virtual Adapter failed (rc=%d)\n", 5648 gen_crq->cmd); 5649 ibmvnic_reset(adapter, VNIC_RESET_FATAL); 5650 } 5651 return; 5652 case IBMVNIC_CRQ_CMD_RSP: 5653 break; 5654 default: 5655 dev_err(dev, "Got an invalid msg type 0x%02x\n", 5656 gen_crq->first); 5657 return; 5658 } 5659 5660 switch (gen_crq->cmd) { 5661 case VERSION_EXCHANGE_RSP: 5662 rc = crq->version_exchange_rsp.rc.code; 5663 if (rc) { 5664 dev_err(dev, "Error %ld in VERSION_EXCHG_RSP\n", rc); 5665 break; 5666 } 5667 ibmvnic_version = 5668 be16_to_cpu(crq->version_exchange_rsp.version); 5669 dev_info(dev, "Partner protocol version is %d\n", 5670 ibmvnic_version); 5671 send_query_cap(adapter); 5672 break; 5673 case QUERY_CAPABILITY_RSP: 5674 handle_query_cap_rsp(crq, adapter); 5675 break; 5676 case QUERY_MAP_RSP: 5677 handle_query_map_rsp(crq, adapter); 5678 break; 5679 case REQUEST_MAP_RSP: 5680 adapter->fw_done_rc = crq->request_map_rsp.rc.code; 5681 complete(&adapter->fw_done); 5682 break; 5683 case REQUEST_UNMAP_RSP: 5684 handle_request_unmap_rsp(crq, adapter); 5685 break; 5686 case REQUEST_CAPABILITY_RSP: 5687 handle_request_cap_rsp(crq, adapter); 5688 break; 5689 case LOGIN_RSP: 5690 netdev_dbg(netdev, "Got Login Response\n"); 5691 handle_login_rsp(crq, adapter); 5692 break; 5693 case LOGICAL_LINK_STATE_RSP: 5694 netdev_dbg(netdev, 5695 "Got Logical Link State Response, state: %d rc: %d\n", 5696 crq->logical_link_state_rsp.link_state, 5697 crq->logical_link_state_rsp.rc.code); 5698 adapter->logical_link_state = 5699 crq->logical_link_state_rsp.link_state; 5700 adapter->init_done_rc = crq->logical_link_state_rsp.rc.code; 5701 complete(&adapter->init_done); 5702 break; 5703 case LINK_STATE_INDICATION: 5704 netdev_dbg(netdev, "Got Logical Link State Indication\n"); 5705 adapter->phys_link_state = 5706 crq->link_state_indication.phys_link_state; 5707 adapter->logical_link_state = 5708 crq->link_state_indication.logical_link_state; 5709 if (adapter->phys_link_state && adapter->logical_link_state) 5710 netif_carrier_on(netdev); 5711 else 5712 netif_carrier_off(netdev); 5713 break; 5714 case CHANGE_MAC_ADDR_RSP: 5715 netdev_dbg(netdev, "Got MAC address change Response\n"); 5716 adapter->fw_done_rc = handle_change_mac_rsp(crq, adapter); 5717 break; 5718 case ERROR_INDICATION: 5719 netdev_dbg(netdev, "Got Error Indication\n"); 5720 handle_error_indication(crq, adapter); 5721 break; 5722 case REQUEST_STATISTICS_RSP: 5723 netdev_dbg(netdev, "Got Statistics Response\n"); 5724 complete(&adapter->stats_done); 5725 break; 5726 case QUERY_IP_OFFLOAD_RSP: 5727 netdev_dbg(netdev, "Got Query IP offload Response\n"); 5728 handle_query_ip_offload_rsp(adapter); 5729 break; 5730 case MULTICAST_CTRL_RSP: 5731 netdev_dbg(netdev, "Got multicast control Response\n"); 5732 break; 5733 case CONTROL_IP_OFFLOAD_RSP: 5734 netdev_dbg(netdev, "Got Control IP offload Response\n"); 5735 dma_unmap_single(dev, adapter->ip_offload_ctrl_tok, 5736 sizeof(adapter->ip_offload_ctrl), 5737 DMA_TO_DEVICE); 5738 complete(&adapter->init_done); 5739 break; 5740 case COLLECT_FW_TRACE_RSP: 5741 netdev_dbg(netdev, "Got Collect firmware trace Response\n"); 5742 complete(&adapter->fw_done); 5743 break; 5744 case GET_VPD_SIZE_RSP: 5745 handle_vpd_size_rsp(crq, adapter); 5746 break; 5747 case GET_VPD_RSP: 5748 handle_vpd_rsp(crq, adapter); 5749 break; 5750 case QUERY_PHYS_PARMS_RSP: 5751 adapter->fw_done_rc = handle_query_phys_parms_rsp(crq, adapter); 5752 complete(&adapter->fw_done); 5753 break; 5754 default: 5755 netdev_err(netdev, "Got an invalid cmd type 0x%02x\n", 5756 gen_crq->cmd); 5757 } 5758 } 5759 5760 static irqreturn_t ibmvnic_interrupt(int irq, void *instance) 5761 { 5762 struct ibmvnic_adapter *adapter = instance; 5763 5764 tasklet_schedule(&adapter->tasklet); 5765 return IRQ_HANDLED; 5766 } 5767 5768 static void ibmvnic_tasklet(struct tasklet_struct *t) 5769 { 5770 struct ibmvnic_adapter *adapter = from_tasklet(adapter, t, tasklet); 5771 struct ibmvnic_crq_queue *queue = &adapter->crq; 5772 union ibmvnic_crq *crq; 5773 unsigned long flags; 5774 5775 spin_lock_irqsave(&queue->lock, flags); 5776 5777 /* Pull all the valid messages off the CRQ */ 5778 while ((crq = ibmvnic_next_crq(adapter)) != NULL) { 5779 /* This barrier makes sure ibmvnic_next_crq()'s 5780 * crq->generic.first & IBMVNIC_CRQ_CMD_RSP is loaded 5781 * before ibmvnic_handle_crq()'s 5782 * switch(gen_crq->first) and switch(gen_crq->cmd). 5783 */ 5784 dma_rmb(); 5785 ibmvnic_handle_crq(crq, adapter); 5786 crq->generic.first = 0; 5787 } 5788 5789 spin_unlock_irqrestore(&queue->lock, flags); 5790 } 5791 5792 static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *adapter) 5793 { 5794 struct vio_dev *vdev = adapter->vdev; 5795 int rc; 5796 5797 do { 5798 rc = plpar_hcall_norets(H_ENABLE_CRQ, vdev->unit_address); 5799 } while (rc == H_IN_PROGRESS || rc == H_BUSY || H_IS_LONG_BUSY(rc)); 5800 5801 if (rc) 5802 dev_err(&vdev->dev, "Error enabling adapter (rc=%d)\n", rc); 5803 5804 return rc; 5805 } 5806 5807 static int ibmvnic_reset_crq(struct ibmvnic_adapter *adapter) 5808 { 5809 struct ibmvnic_crq_queue *crq = &adapter->crq; 5810 struct device *dev = &adapter->vdev->dev; 5811 struct vio_dev *vdev = adapter->vdev; 5812 int rc; 5813 5814 /* Close the CRQ */ 5815 do { 5816 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address); 5817 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc)); 5818 5819 /* Clean out the queue */ 5820 if (!crq->msgs) 5821 return -EINVAL; 5822 5823 memset(crq->msgs, 0, PAGE_SIZE); 5824 crq->cur = 0; 5825 crq->active = false; 5826 5827 /* And re-open it again */ 5828 rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address, 5829 crq->msg_token, PAGE_SIZE); 5830 5831 if (rc == H_CLOSED) 5832 /* Adapter is good, but other end is not ready */ 5833 dev_warn(dev, "Partner adapter not ready\n"); 5834 else if (rc != 0) 5835 dev_warn(dev, "Couldn't register crq (rc=%d)\n", rc); 5836 5837 return rc; 5838 } 5839 5840 static void release_crq_queue(struct ibmvnic_adapter *adapter) 5841 { 5842 struct ibmvnic_crq_queue *crq = &adapter->crq; 5843 struct vio_dev *vdev = adapter->vdev; 5844 long rc; 5845 5846 if (!crq->msgs) 5847 return; 5848 5849 netdev_dbg(adapter->netdev, "Releasing CRQ\n"); 5850 free_irq(vdev->irq, adapter); 5851 tasklet_kill(&adapter->tasklet); 5852 do { 5853 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address); 5854 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc)); 5855 5856 dma_unmap_single(&vdev->dev, crq->msg_token, PAGE_SIZE, 5857 DMA_BIDIRECTIONAL); 5858 free_page((unsigned long)crq->msgs); 5859 crq->msgs = NULL; 5860 crq->active = false; 5861 } 5862 5863 static int init_crq_queue(struct ibmvnic_adapter *adapter) 5864 { 5865 struct ibmvnic_crq_queue *crq = &adapter->crq; 5866 struct device *dev = &adapter->vdev->dev; 5867 struct vio_dev *vdev = adapter->vdev; 5868 int rc, retrc = -ENOMEM; 5869 5870 if (crq->msgs) 5871 return 0; 5872 5873 crq->msgs = (union ibmvnic_crq *)get_zeroed_page(GFP_KERNEL); 5874 /* Should we allocate more than one page? */ 5875 5876 if (!crq->msgs) 5877 return -ENOMEM; 5878 5879 crq->size = PAGE_SIZE / sizeof(*crq->msgs); 5880 crq->msg_token = dma_map_single(dev, crq->msgs, PAGE_SIZE, 5881 DMA_BIDIRECTIONAL); 5882 if (dma_mapping_error(dev, crq->msg_token)) 5883 goto map_failed; 5884 5885 rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address, 5886 crq->msg_token, PAGE_SIZE); 5887 5888 if (rc == H_RESOURCE) 5889 /* maybe kexecing and resource is busy. try a reset */ 5890 rc = ibmvnic_reset_crq(adapter); 5891 retrc = rc; 5892 5893 if (rc == H_CLOSED) { 5894 dev_warn(dev, "Partner adapter not ready\n"); 5895 } else if (rc) { 5896 dev_warn(dev, "Error %d opening adapter\n", rc); 5897 goto reg_crq_failed; 5898 } 5899 5900 retrc = 0; 5901 5902 tasklet_setup(&adapter->tasklet, (void *)ibmvnic_tasklet); 5903 5904 netdev_dbg(adapter->netdev, "registering irq 0x%x\n", vdev->irq); 5905 snprintf(crq->name, sizeof(crq->name), "ibmvnic-%x", 5906 adapter->vdev->unit_address); 5907 rc = request_irq(vdev->irq, ibmvnic_interrupt, 0, crq->name, adapter); 5908 if (rc) { 5909 dev_err(dev, "Couldn't register irq 0x%x. rc=%d\n", 5910 vdev->irq, rc); 5911 goto req_irq_failed; 5912 } 5913 5914 rc = vio_enable_interrupts(vdev); 5915 if (rc) { 5916 dev_err(dev, "Error %d enabling interrupts\n", rc); 5917 goto req_irq_failed; 5918 } 5919 5920 crq->cur = 0; 5921 spin_lock_init(&crq->lock); 5922 5923 /* process any CRQs that were queued before we enabled interrupts */ 5924 tasklet_schedule(&adapter->tasklet); 5925 5926 return retrc; 5927 5928 req_irq_failed: 5929 tasklet_kill(&adapter->tasklet); 5930 do { 5931 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address); 5932 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc)); 5933 reg_crq_failed: 5934 dma_unmap_single(dev, crq->msg_token, PAGE_SIZE, DMA_BIDIRECTIONAL); 5935 map_failed: 5936 free_page((unsigned long)crq->msgs); 5937 crq->msgs = NULL; 5938 return retrc; 5939 } 5940 5941 static int ibmvnic_reset_init(struct ibmvnic_adapter *adapter, bool reset) 5942 { 5943 struct device *dev = &adapter->vdev->dev; 5944 unsigned long timeout = msecs_to_jiffies(20000); 5945 u64 old_num_rx_queues = adapter->req_rx_queues; 5946 u64 old_num_tx_queues = adapter->req_tx_queues; 5947 int rc; 5948 5949 adapter->from_passive_init = false; 5950 5951 rc = ibmvnic_send_crq_init(adapter); 5952 if (rc) { 5953 dev_err(dev, "Send crq init failed with error %d\n", rc); 5954 return rc; 5955 } 5956 5957 if (!wait_for_completion_timeout(&adapter->init_done, timeout)) { 5958 dev_err(dev, "Initialization sequence timed out\n"); 5959 return -ETIMEDOUT; 5960 } 5961 5962 if (adapter->init_done_rc) { 5963 release_crq_queue(adapter); 5964 dev_err(dev, "CRQ-init failed, %d\n", adapter->init_done_rc); 5965 return adapter->init_done_rc; 5966 } 5967 5968 if (adapter->from_passive_init) { 5969 adapter->state = VNIC_OPEN; 5970 adapter->from_passive_init = false; 5971 dev_err(dev, "CRQ-init failed, passive-init\n"); 5972 return -EINVAL; 5973 } 5974 5975 if (reset && 5976 test_bit(0, &adapter->resetting) && !adapter->wait_for_reset && 5977 adapter->reset_reason != VNIC_RESET_MOBILITY) { 5978 if (adapter->req_rx_queues != old_num_rx_queues || 5979 adapter->req_tx_queues != old_num_tx_queues) { 5980 release_sub_crqs(adapter, 0); 5981 rc = init_sub_crqs(adapter); 5982 } else { 5983 rc = reset_sub_crq_queues(adapter); 5984 } 5985 } else { 5986 rc = init_sub_crqs(adapter); 5987 } 5988 5989 if (rc) { 5990 dev_err(dev, "Initialization of sub crqs failed\n"); 5991 release_crq_queue(adapter); 5992 return rc; 5993 } 5994 5995 rc = init_sub_crq_irqs(adapter); 5996 if (rc) { 5997 dev_err(dev, "Failed to initialize sub crq irqs\n"); 5998 release_crq_queue(adapter); 5999 } 6000 6001 return rc; 6002 } 6003 6004 static struct device_attribute dev_attr_failover; 6005 6006 static int ibmvnic_probe(struct vio_dev *dev, const struct vio_device_id *id) 6007 { 6008 struct ibmvnic_adapter *adapter; 6009 struct net_device *netdev; 6010 unsigned char *mac_addr_p; 6011 unsigned long flags; 6012 bool init_success; 6013 int rc; 6014 6015 dev_dbg(&dev->dev, "entering ibmvnic_probe for UA 0x%x\n", 6016 dev->unit_address); 6017 6018 mac_addr_p = (unsigned char *)vio_get_attribute(dev, 6019 VETH_MAC_ADDR, NULL); 6020 if (!mac_addr_p) { 6021 dev_err(&dev->dev, 6022 "(%s:%3.3d) ERROR: Can't find MAC_ADDR attribute\n", 6023 __FILE__, __LINE__); 6024 return 0; 6025 } 6026 6027 netdev = alloc_etherdev_mq(sizeof(struct ibmvnic_adapter), 6028 IBMVNIC_MAX_QUEUES); 6029 if (!netdev) 6030 return -ENOMEM; 6031 6032 adapter = netdev_priv(netdev); 6033 adapter->state = VNIC_PROBING; 6034 dev_set_drvdata(&dev->dev, netdev); 6035 adapter->vdev = dev; 6036 adapter->netdev = netdev; 6037 adapter->login_pending = false; 6038 memset(&adapter->map_ids, 0, sizeof(adapter->map_ids)); 6039 /* map_ids start at 1, so ensure map_id 0 is always "in-use" */ 6040 bitmap_set(adapter->map_ids, 0, 1); 6041 6042 ether_addr_copy(adapter->mac_addr, mac_addr_p); 6043 eth_hw_addr_set(netdev, adapter->mac_addr); 6044 netdev->irq = dev->irq; 6045 netdev->netdev_ops = &ibmvnic_netdev_ops; 6046 netdev->ethtool_ops = &ibmvnic_ethtool_ops; 6047 SET_NETDEV_DEV(netdev, &dev->dev); 6048 6049 INIT_WORK(&adapter->ibmvnic_reset, __ibmvnic_reset); 6050 INIT_DELAYED_WORK(&adapter->ibmvnic_delayed_reset, 6051 __ibmvnic_delayed_reset); 6052 INIT_LIST_HEAD(&adapter->rwi_list); 6053 spin_lock_init(&adapter->rwi_lock); 6054 spin_lock_init(&adapter->state_lock); 6055 mutex_init(&adapter->fw_lock); 6056 init_completion(&adapter->probe_done); 6057 init_completion(&adapter->init_done); 6058 init_completion(&adapter->fw_done); 6059 init_completion(&adapter->reset_done); 6060 init_completion(&adapter->stats_done); 6061 clear_bit(0, &adapter->resetting); 6062 adapter->prev_rx_buf_sz = 0; 6063 adapter->prev_mtu = 0; 6064 6065 init_success = false; 6066 do { 6067 reinit_init_done(adapter); 6068 6069 /* clear any failovers we got in the previous pass 6070 * since we are reinitializing the CRQ 6071 */ 6072 adapter->failover_pending = false; 6073 6074 /* If we had already initialized CRQ, we may have one or 6075 * more resets queued already. Discard those and release 6076 * the CRQ before initializing the CRQ again. 6077 */ 6078 release_crq_queue(adapter); 6079 6080 /* Since we are still in PROBING state, __ibmvnic_reset() 6081 * will not access the ->rwi_list and since we released CRQ, 6082 * we won't get _new_ transport events. But there maybe an 6083 * ongoing ibmvnic_reset() call. So serialize access to 6084 * rwi_list. If we win the race, ibvmnic_reset() could add 6085 * a reset after we purged but thats ok - we just may end 6086 * up with an extra reset (i.e similar to having two or more 6087 * resets in the queue at once). 6088 * CHECK. 6089 */ 6090 spin_lock_irqsave(&adapter->rwi_lock, flags); 6091 flush_reset_queue(adapter); 6092 spin_unlock_irqrestore(&adapter->rwi_lock, flags); 6093 6094 rc = init_crq_queue(adapter); 6095 if (rc) { 6096 dev_err(&dev->dev, "Couldn't initialize crq. rc=%d\n", 6097 rc); 6098 goto ibmvnic_init_fail; 6099 } 6100 6101 rc = ibmvnic_reset_init(adapter, false); 6102 } while (rc == -EAGAIN); 6103 6104 /* We are ignoring the error from ibmvnic_reset_init() assuming that the 6105 * partner is not ready. CRQ is not active. When the partner becomes 6106 * ready, we will do the passive init reset. 6107 */ 6108 6109 if (!rc) 6110 init_success = true; 6111 6112 rc = init_stats_buffers(adapter); 6113 if (rc) 6114 goto ibmvnic_init_fail; 6115 6116 rc = init_stats_token(adapter); 6117 if (rc) 6118 goto ibmvnic_stats_fail; 6119 6120 rc = device_create_file(&dev->dev, &dev_attr_failover); 6121 if (rc) 6122 goto ibmvnic_dev_file_err; 6123 6124 netif_carrier_off(netdev); 6125 6126 if (init_success) { 6127 adapter->state = VNIC_PROBED; 6128 netdev->mtu = adapter->req_mtu - ETH_HLEN; 6129 netdev->min_mtu = adapter->min_mtu - ETH_HLEN; 6130 netdev->max_mtu = adapter->max_mtu - ETH_HLEN; 6131 } else { 6132 adapter->state = VNIC_DOWN; 6133 } 6134 6135 adapter->wait_for_reset = false; 6136 adapter->last_reset_time = jiffies; 6137 6138 rc = register_netdev(netdev); 6139 if (rc) { 6140 dev_err(&dev->dev, "failed to register netdev rc=%d\n", rc); 6141 goto ibmvnic_register_fail; 6142 } 6143 dev_info(&dev->dev, "ibmvnic registered\n"); 6144 6145 complete(&adapter->probe_done); 6146 6147 return 0; 6148 6149 ibmvnic_register_fail: 6150 device_remove_file(&dev->dev, &dev_attr_failover); 6151 6152 ibmvnic_dev_file_err: 6153 release_stats_token(adapter); 6154 6155 ibmvnic_stats_fail: 6156 release_stats_buffers(adapter); 6157 6158 ibmvnic_init_fail: 6159 release_sub_crqs(adapter, 1); 6160 release_crq_queue(adapter); 6161 6162 /* cleanup worker thread after releasing CRQ so we don't get 6163 * transport events (i.e new work items for the worker thread). 6164 */ 6165 adapter->state = VNIC_REMOVING; 6166 complete(&adapter->probe_done); 6167 flush_work(&adapter->ibmvnic_reset); 6168 flush_delayed_work(&adapter->ibmvnic_delayed_reset); 6169 6170 flush_reset_queue(adapter); 6171 6172 mutex_destroy(&adapter->fw_lock); 6173 free_netdev(netdev); 6174 6175 return rc; 6176 } 6177 6178 static void ibmvnic_remove(struct vio_dev *dev) 6179 { 6180 struct net_device *netdev = dev_get_drvdata(&dev->dev); 6181 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 6182 unsigned long flags; 6183 6184 spin_lock_irqsave(&adapter->state_lock, flags); 6185 6186 /* If ibmvnic_reset() is scheduling a reset, wait for it to 6187 * finish. Then, set the state to REMOVING to prevent it from 6188 * scheduling any more work and to have reset functions ignore 6189 * any resets that have already been scheduled. Drop the lock 6190 * after setting state, so __ibmvnic_reset() which is called 6191 * from the flush_work() below, can make progress. 6192 */ 6193 spin_lock(&adapter->rwi_lock); 6194 adapter->state = VNIC_REMOVING; 6195 spin_unlock(&adapter->rwi_lock); 6196 6197 spin_unlock_irqrestore(&adapter->state_lock, flags); 6198 6199 flush_work(&adapter->ibmvnic_reset); 6200 flush_delayed_work(&adapter->ibmvnic_delayed_reset); 6201 6202 rtnl_lock(); 6203 unregister_netdevice(netdev); 6204 6205 release_resources(adapter); 6206 release_rx_pools(adapter); 6207 release_tx_pools(adapter); 6208 release_sub_crqs(adapter, 1); 6209 release_crq_queue(adapter); 6210 6211 release_stats_token(adapter); 6212 release_stats_buffers(adapter); 6213 6214 adapter->state = VNIC_REMOVED; 6215 6216 rtnl_unlock(); 6217 mutex_destroy(&adapter->fw_lock); 6218 device_remove_file(&dev->dev, &dev_attr_failover); 6219 free_netdev(netdev); 6220 dev_set_drvdata(&dev->dev, NULL); 6221 } 6222 6223 static ssize_t failover_store(struct device *dev, struct device_attribute *attr, 6224 const char *buf, size_t count) 6225 { 6226 struct net_device *netdev = dev_get_drvdata(dev); 6227 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 6228 unsigned long retbuf[PLPAR_HCALL_BUFSIZE]; 6229 __be64 session_token; 6230 long rc; 6231 6232 if (!sysfs_streq(buf, "1")) 6233 return -EINVAL; 6234 6235 rc = plpar_hcall(H_VIOCTL, retbuf, adapter->vdev->unit_address, 6236 H_GET_SESSION_TOKEN, 0, 0, 0); 6237 if (rc) { 6238 netdev_err(netdev, "Couldn't retrieve session token, rc %ld\n", 6239 rc); 6240 goto last_resort; 6241 } 6242 6243 session_token = (__be64)retbuf[0]; 6244 netdev_dbg(netdev, "Initiating client failover, session id %llx\n", 6245 be64_to_cpu(session_token)); 6246 rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address, 6247 H_SESSION_ERR_DETECTED, session_token, 0, 0); 6248 if (rc) { 6249 netdev_err(netdev, 6250 "H_VIOCTL initiated failover failed, rc %ld\n", 6251 rc); 6252 goto last_resort; 6253 } 6254 6255 return count; 6256 6257 last_resort: 6258 netdev_dbg(netdev, "Trying to send CRQ_CMD, the last resort\n"); 6259 ibmvnic_reset(adapter, VNIC_RESET_FAILOVER); 6260 6261 return count; 6262 } 6263 static DEVICE_ATTR_WO(failover); 6264 6265 static unsigned long ibmvnic_get_desired_dma(struct vio_dev *vdev) 6266 { 6267 struct net_device *netdev = dev_get_drvdata(&vdev->dev); 6268 struct ibmvnic_adapter *adapter; 6269 struct iommu_table *tbl; 6270 unsigned long ret = 0; 6271 int i; 6272 6273 tbl = get_iommu_table_base(&vdev->dev); 6274 6275 /* netdev inits at probe time along with the structures we need below*/ 6276 if (!netdev) 6277 return IOMMU_PAGE_ALIGN(IBMVNIC_IO_ENTITLEMENT_DEFAULT, tbl); 6278 6279 adapter = netdev_priv(netdev); 6280 6281 ret += PAGE_SIZE; /* the crq message queue */ 6282 ret += IOMMU_PAGE_ALIGN(sizeof(struct ibmvnic_statistics), tbl); 6283 6284 for (i = 0; i < adapter->req_tx_queues + adapter->req_rx_queues; i++) 6285 ret += 4 * PAGE_SIZE; /* the scrq message queue */ 6286 6287 for (i = 0; i < adapter->num_active_rx_pools; i++) 6288 ret += adapter->rx_pool[i].size * 6289 IOMMU_PAGE_ALIGN(adapter->rx_pool[i].buff_size, tbl); 6290 6291 return ret; 6292 } 6293 6294 static int ibmvnic_resume(struct device *dev) 6295 { 6296 struct net_device *netdev = dev_get_drvdata(dev); 6297 struct ibmvnic_adapter *adapter = netdev_priv(netdev); 6298 6299 if (adapter->state != VNIC_OPEN) 6300 return 0; 6301 6302 tasklet_schedule(&adapter->tasklet); 6303 6304 return 0; 6305 } 6306 6307 static const struct vio_device_id ibmvnic_device_table[] = { 6308 {"network", "IBM,vnic"}, 6309 {"", "" } 6310 }; 6311 MODULE_DEVICE_TABLE(vio, ibmvnic_device_table); 6312 6313 static const struct dev_pm_ops ibmvnic_pm_ops = { 6314 .resume = ibmvnic_resume 6315 }; 6316 6317 static struct vio_driver ibmvnic_driver = { 6318 .id_table = ibmvnic_device_table, 6319 .probe = ibmvnic_probe, 6320 .remove = ibmvnic_remove, 6321 .get_desired_dma = ibmvnic_get_desired_dma, 6322 .name = ibmvnic_driver_name, 6323 .pm = &ibmvnic_pm_ops, 6324 }; 6325 6326 /* module functions */ 6327 static int __init ibmvnic_module_init(void) 6328 { 6329 pr_info("%s: %s %s\n", ibmvnic_driver_name, ibmvnic_driver_string, 6330 IBMVNIC_DRIVER_VERSION); 6331 6332 return vio_register_driver(&ibmvnic_driver); 6333 } 6334 6335 static void __exit ibmvnic_module_exit(void) 6336 { 6337 vio_unregister_driver(&ibmvnic_driver); 6338 } 6339 6340 module_init(ibmvnic_module_init); 6341 module_exit(ibmvnic_module_exit); 6342