mlx5_main.c revision 331914
1/*- 2 * Copyright (c) 2013-2017, Mellanox Technologies, Ltd. All rights reserved. 3 * 4 * Redistribution and use in source and binary forms, with or without 5 * modification, are permitted provided that the following conditions 6 * are met: 7 * 1. Redistributions of source code must retain the above copyright 8 * notice, this list of conditions and the following disclaimer. 9 * 2. Redistributions in binary form must reproduce the above copyright 10 * notice, this list of conditions and the following disclaimer in the 11 * documentation and/or other materials provided with the distribution. 12 * 13 * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS `AS IS' AND 14 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 15 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 16 * ARE DISCLAIMED. IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE 17 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 18 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 19 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 20 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 21 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 22 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 23 * SUCH DAMAGE. 24 * 25 * $FreeBSD: stable/11/sys/dev/mlx5/mlx5_core/mlx5_main.c 331914 2018-04-03 09:27:05Z hselasky $ 26 */ 27 28#define LINUXKPI_PARAM_PREFIX mlx5_ 29 30#include <linux/kmod.h> 31#include <linux/module.h> 32#include <linux/errno.h> 33#include <linux/pci.h> 34#include <linux/dma-mapping.h> 35#include <linux/slab.h> 36#include <linux/io-mapping.h> 37#include <linux/interrupt.h> 38#include <dev/mlx5/driver.h> 39#include <dev/mlx5/cq.h> 40#include <dev/mlx5/qp.h> 41#include <dev/mlx5/srq.h> 42#include <linux/delay.h> 43#include <dev/mlx5/mlx5_ifc.h> 44#include "mlx5_core.h" 45#include "fs_core.h" 46 47MODULE_AUTHOR("Eli Cohen <eli@mellanox.com>"); 48MODULE_DESCRIPTION("Mellanox Connect-IB, ConnectX-4 core driver"); 49MODULE_LICENSE("Dual BSD/GPL"); 50#if (__FreeBSD_version >= 1100000) 51MODULE_DEPEND(mlx5, linuxkpi, 1, 1, 1); 52#endif 53MODULE_VERSION(mlx5, 1); 54 55int mlx5_core_debug_mask; 56module_param_named(debug_mask, mlx5_core_debug_mask, int, 0644); 57MODULE_PARM_DESC(debug_mask, "debug mask: 1 = dump cmd data, 2 = dump cmd exec time, 3 = both. Default=0"); 58 59#define MLX5_DEFAULT_PROF 2 60static int prof_sel = MLX5_DEFAULT_PROF; 61module_param_named(prof_sel, prof_sel, int, 0444); 62MODULE_PARM_DESC(prof_sel, "profile selector. Valid range 0 - 2"); 63 64#define NUMA_NO_NODE -1 65 66static LIST_HEAD(intf_list); 67static LIST_HEAD(dev_list); 68static DEFINE_MUTEX(intf_mutex); 69 70struct mlx5_device_context { 71 struct list_head list; 72 struct mlx5_interface *intf; 73 void *context; 74}; 75 76enum { 77 MLX5_ATOMIC_REQ_MODE_BE = 0x0, 78 MLX5_ATOMIC_REQ_MODE_HOST_ENDIANNESS = 0x1, 79}; 80 81static struct mlx5_profile profiles[] = { 82 [0] = { 83 .mask = 0, 84 }, 85 [1] = { 86 .mask = MLX5_PROF_MASK_QP_SIZE, 87 .log_max_qp = 12, 88 }, 89 [2] = { 90 .mask = MLX5_PROF_MASK_QP_SIZE | 91 MLX5_PROF_MASK_MR_CACHE, 92 .log_max_qp = 17, 93 .mr_cache[0] = { 94 .size = 500, 95 .limit = 250 96 }, 97 .mr_cache[1] = { 98 .size = 500, 99 .limit = 250 100 }, 101 .mr_cache[2] = { 102 .size = 500, 103 .limit = 250 104 }, 105 .mr_cache[3] = { 106 .size = 500, 107 .limit = 250 108 }, 109 .mr_cache[4] = { 110 .size = 500, 111 .limit = 250 112 }, 113 .mr_cache[5] = { 114 .size = 500, 115 .limit = 250 116 }, 117 .mr_cache[6] = { 118 .size = 500, 119 .limit = 250 120 }, 121 .mr_cache[7] = { 122 .size = 500, 123 .limit = 250 124 }, 125 .mr_cache[8] = { 126 .size = 500, 127 .limit = 250 128 }, 129 .mr_cache[9] = { 130 .size = 500, 131 .limit = 250 132 }, 133 .mr_cache[10] = { 134 .size = 500, 135 .limit = 250 136 }, 137 .mr_cache[11] = { 138 .size = 500, 139 .limit = 250 140 }, 141 .mr_cache[12] = { 142 .size = 64, 143 .limit = 32 144 }, 145 .mr_cache[13] = { 146 .size = 32, 147 .limit = 16 148 }, 149 .mr_cache[14] = { 150 .size = 16, 151 .limit = 8 152 }, 153 }, 154 [3] = { 155 .mask = MLX5_PROF_MASK_QP_SIZE, 156 .log_max_qp = 17, 157 }, 158}; 159 160static int set_dma_caps(struct pci_dev *pdev) 161{ 162 int err; 163 164 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64)); 165 if (err) { 166 device_printf((&pdev->dev)->bsddev, "WARN: ""Warning: couldn't set 64-bit PCI DMA mask\n"); 167 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32)); 168 if (err) { 169 device_printf((&pdev->dev)->bsddev, "ERR: ""Can't set PCI DMA mask, aborting\n"); 170 return err; 171 } 172 } 173 174 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64)); 175 if (err) { 176 device_printf((&pdev->dev)->bsddev, "WARN: ""Warning: couldn't set 64-bit consistent PCI DMA mask\n"); 177 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32)); 178 if (err) { 179 device_printf((&pdev->dev)->bsddev, "ERR: ""Can't set consistent PCI DMA mask, aborting\n"); 180 return err; 181 } 182 } 183 184 dma_set_max_seg_size(&pdev->dev, 2u * 1024 * 1024 * 1024); 185 return err; 186} 187 188static int mlx5_pci_enable_device(struct mlx5_core_dev *dev) 189{ 190 struct pci_dev *pdev = dev->pdev; 191 int err = 0; 192 193 mutex_lock(&dev->pci_status_mutex); 194 if (dev->pci_status == MLX5_PCI_STATUS_DISABLED) { 195 err = pci_enable_device(pdev); 196 if (!err) 197 dev->pci_status = MLX5_PCI_STATUS_ENABLED; 198 } 199 mutex_unlock(&dev->pci_status_mutex); 200 201 return err; 202} 203 204static void mlx5_pci_disable_device(struct mlx5_core_dev *dev) 205{ 206 struct pci_dev *pdev = dev->pdev; 207 208 mutex_lock(&dev->pci_status_mutex); 209 if (dev->pci_status == MLX5_PCI_STATUS_ENABLED) { 210 pci_disable_device(pdev); 211 dev->pci_status = MLX5_PCI_STATUS_DISABLED; 212 } 213 mutex_unlock(&dev->pci_status_mutex); 214} 215 216static int request_bar(struct pci_dev *pdev) 217{ 218 int err = 0; 219 220 if (!(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) { 221 device_printf((&pdev->dev)->bsddev, "ERR: ""Missing registers BAR, aborting\n"); 222 return -ENODEV; 223 } 224 225 err = pci_request_regions(pdev, DRIVER_NAME); 226 if (err) 227 device_printf((&pdev->dev)->bsddev, "ERR: ""Couldn't get PCI resources, aborting\n"); 228 229 return err; 230} 231 232static void release_bar(struct pci_dev *pdev) 233{ 234 pci_release_regions(pdev); 235} 236 237static int mlx5_enable_msix(struct mlx5_core_dev *dev) 238{ 239 struct mlx5_priv *priv = &dev->priv; 240 struct mlx5_eq_table *table = &priv->eq_table; 241 int num_eqs = 1 << MLX5_CAP_GEN(dev, log_max_eq); 242 int nvec; 243 int i; 244 245 nvec = MLX5_CAP_GEN(dev, num_ports) * num_online_cpus() + 246 MLX5_EQ_VEC_COMP_BASE; 247 nvec = min_t(int, nvec, num_eqs); 248 if (nvec <= MLX5_EQ_VEC_COMP_BASE) 249 return -ENOMEM; 250 251 priv->msix_arr = kzalloc(nvec * sizeof(*priv->msix_arr), GFP_KERNEL); 252 253 priv->irq_info = kzalloc(nvec * sizeof(*priv->irq_info), GFP_KERNEL); 254 255 for (i = 0; i < nvec; i++) 256 priv->msix_arr[i].entry = i; 257 258 nvec = pci_enable_msix_range(dev->pdev, priv->msix_arr, 259 MLX5_EQ_VEC_COMP_BASE + 1, nvec); 260 if (nvec < 0) 261 return nvec; 262 263 table->num_comp_vectors = nvec - MLX5_EQ_VEC_COMP_BASE; 264 265 return 0; 266 267} 268 269static void mlx5_disable_msix(struct mlx5_core_dev *dev) 270{ 271 struct mlx5_priv *priv = &dev->priv; 272 273 pci_disable_msix(dev->pdev); 274 kfree(priv->irq_info); 275 kfree(priv->msix_arr); 276} 277 278struct mlx5_reg_host_endianess { 279 u8 he; 280 u8 rsvd[15]; 281}; 282 283 284#define CAP_MASK(pos, size) ((u64)((1 << (size)) - 1) << (pos)) 285 286enum { 287 MLX5_CAP_BITS_RW_MASK = CAP_MASK(MLX5_CAP_OFF_CMDIF_CSUM, 2) | 288 MLX5_DEV_CAP_FLAG_DCT | 289 MLX5_DEV_CAP_FLAG_DRAIN_SIGERR, 290}; 291 292static u16 to_fw_pkey_sz(u32 size) 293{ 294 switch (size) { 295 case 128: 296 return 0; 297 case 256: 298 return 1; 299 case 512: 300 return 2; 301 case 1024: 302 return 3; 303 case 2048: 304 return 4; 305 case 4096: 306 return 5; 307 default: 308 printf("mlx5_core: WARN: ""invalid pkey table size %d\n", size); 309 return 0; 310 } 311} 312 313static int mlx5_core_get_caps_mode(struct mlx5_core_dev *dev, 314 enum mlx5_cap_type cap_type, 315 enum mlx5_cap_mode cap_mode) 316{ 317 u8 in[MLX5_ST_SZ_BYTES(query_hca_cap_in)]; 318 int out_sz = MLX5_ST_SZ_BYTES(query_hca_cap_out); 319 void *out, *hca_caps; 320 u16 opmod = (cap_type << 1) | (cap_mode & 0x01); 321 int err; 322 323 memset(in, 0, sizeof(in)); 324 out = kzalloc(out_sz, GFP_KERNEL); 325 326 MLX5_SET(query_hca_cap_in, in, opcode, MLX5_CMD_OP_QUERY_HCA_CAP); 327 MLX5_SET(query_hca_cap_in, in, op_mod, opmod); 328 err = mlx5_cmd_exec(dev, in, sizeof(in), out, out_sz); 329 if (err) { 330 mlx5_core_warn(dev, 331 "QUERY_HCA_CAP : type(%x) opmode(%x) Failed(%d)\n", 332 cap_type, cap_mode, err); 333 goto query_ex; 334 } 335 336 hca_caps = MLX5_ADDR_OF(query_hca_cap_out, out, capability); 337 338 switch (cap_mode) { 339 case HCA_CAP_OPMOD_GET_MAX: 340 memcpy(dev->hca_caps_max[cap_type], hca_caps, 341 MLX5_UN_SZ_BYTES(hca_cap_union)); 342 break; 343 case HCA_CAP_OPMOD_GET_CUR: 344 memcpy(dev->hca_caps_cur[cap_type], hca_caps, 345 MLX5_UN_SZ_BYTES(hca_cap_union)); 346 break; 347 default: 348 mlx5_core_warn(dev, 349 "Tried to query dev cap type(%x) with wrong opmode(%x)\n", 350 cap_type, cap_mode); 351 err = -EINVAL; 352 break; 353 } 354query_ex: 355 kfree(out); 356 return err; 357} 358 359int mlx5_core_get_caps(struct mlx5_core_dev *dev, enum mlx5_cap_type cap_type) 360{ 361 int ret; 362 363 ret = mlx5_core_get_caps_mode(dev, cap_type, HCA_CAP_OPMOD_GET_CUR); 364 if (ret) 365 return ret; 366 367 return mlx5_core_get_caps_mode(dev, cap_type, HCA_CAP_OPMOD_GET_MAX); 368} 369 370static int set_caps(struct mlx5_core_dev *dev, void *in, int in_sz) 371{ 372 u32 out[MLX5_ST_SZ_DW(set_hca_cap_out)] = {0}; 373 374 MLX5_SET(set_hca_cap_in, in, opcode, MLX5_CMD_OP_SET_HCA_CAP); 375 376 return mlx5_cmd_exec(dev, in, in_sz, out, sizeof(out)); 377} 378 379static int handle_hca_cap(struct mlx5_core_dev *dev) 380{ 381 void *set_ctx = NULL; 382 struct mlx5_profile *prof = dev->profile; 383 int err = -ENOMEM; 384 int set_sz = MLX5_ST_SZ_BYTES(set_hca_cap_in); 385 void *set_hca_cap; 386 387 set_ctx = kzalloc(set_sz, GFP_KERNEL); 388 389 err = mlx5_core_get_caps(dev, MLX5_CAP_GENERAL); 390 if (err) 391 goto query_ex; 392 393 set_hca_cap = MLX5_ADDR_OF(set_hca_cap_in, set_ctx, 394 capability); 395 memcpy(set_hca_cap, dev->hca_caps_cur[MLX5_CAP_GENERAL], 396 MLX5_ST_SZ_BYTES(cmd_hca_cap)); 397 398 mlx5_core_dbg(dev, "Current Pkey table size %d Setting new size %d\n", 399 mlx5_to_sw_pkey_sz(MLX5_CAP_GEN(dev, pkey_table_size)), 400 128); 401 /* we limit the size of the pkey table to 128 entries for now */ 402 MLX5_SET(cmd_hca_cap, set_hca_cap, pkey_table_size, 403 to_fw_pkey_sz(128)); 404 405 if (prof->mask & MLX5_PROF_MASK_QP_SIZE) 406 MLX5_SET(cmd_hca_cap, set_hca_cap, log_max_qp, 407 prof->log_max_qp); 408 409 /* disable cmdif checksum */ 410 MLX5_SET(cmd_hca_cap, set_hca_cap, cmdif_checksum, 0); 411 412 /* enable drain sigerr */ 413 MLX5_SET(cmd_hca_cap, set_hca_cap, drain_sigerr, 1); 414 415 MLX5_SET(cmd_hca_cap, set_hca_cap, log_uar_page_sz, PAGE_SHIFT - 12); 416 417 err = set_caps(dev, set_ctx, set_sz); 418 419query_ex: 420 kfree(set_ctx); 421 return err; 422} 423 424static int handle_hca_cap_atomic(struct mlx5_core_dev *dev) 425{ 426 void *set_ctx; 427 void *set_hca_cap; 428 int set_sz = MLX5_ST_SZ_BYTES(set_hca_cap_in); 429 int req_endianness; 430 int err; 431 432 if (MLX5_CAP_GEN(dev, atomic)) { 433 err = mlx5_core_get_caps(dev, MLX5_CAP_ATOMIC); 434 if (err) 435 return err; 436 } else { 437 return 0; 438 } 439 440 req_endianness = 441 MLX5_CAP_ATOMIC(dev, 442 supported_atomic_req_8B_endianess_mode_1); 443 444 if (req_endianness != MLX5_ATOMIC_REQ_MODE_HOST_ENDIANNESS) 445 return 0; 446 447 set_ctx = kzalloc(set_sz, GFP_KERNEL); 448 if (!set_ctx) 449 return -ENOMEM; 450 451 MLX5_SET(set_hca_cap_in, set_ctx, op_mod, 452 MLX5_SET_HCA_CAP_OP_MOD_ATOMIC << 1); 453 set_hca_cap = MLX5_ADDR_OF(set_hca_cap_in, set_ctx, capability); 454 455 /* Set requestor to host endianness */ 456 MLX5_SET(atomic_caps, set_hca_cap, atomic_req_8B_endianess_mode, 457 MLX5_ATOMIC_REQ_MODE_HOST_ENDIANNESS); 458 459 err = set_caps(dev, set_ctx, set_sz); 460 461 kfree(set_ctx); 462 return err; 463} 464 465static int set_hca_ctrl(struct mlx5_core_dev *dev) 466{ 467 struct mlx5_reg_host_endianess he_in; 468 struct mlx5_reg_host_endianess he_out; 469 int err; 470 471 if (MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_ETH && 472 !MLX5_CAP_GEN(dev, roce)) 473 return 0; 474 475 memset(&he_in, 0, sizeof(he_in)); 476 he_in.he = MLX5_SET_HOST_ENDIANNESS; 477 err = mlx5_core_access_reg(dev, &he_in, sizeof(he_in), 478 &he_out, sizeof(he_out), 479 MLX5_REG_HOST_ENDIANNESS, 0, 1); 480 return err; 481} 482 483static int mlx5_core_enable_hca(struct mlx5_core_dev *dev) 484{ 485 u32 out[MLX5_ST_SZ_DW(enable_hca_out)] = {0}; 486 u32 in[MLX5_ST_SZ_DW(enable_hca_in)] = {0}; 487 488 MLX5_SET(enable_hca_in, in, opcode, MLX5_CMD_OP_ENABLE_HCA); 489 return mlx5_cmd_exec(dev, &in, sizeof(in), &out, sizeof(out)); 490} 491 492static int mlx5_core_disable_hca(struct mlx5_core_dev *dev) 493{ 494 u32 out[MLX5_ST_SZ_DW(disable_hca_out)] = {0}; 495 u32 in[MLX5_ST_SZ_DW(disable_hca_in)] = {0}; 496 497 MLX5_SET(disable_hca_in, in, opcode, MLX5_CMD_OP_DISABLE_HCA); 498 return mlx5_cmd_exec(dev, in, sizeof(in), out, sizeof(out)); 499} 500 501static int mlx5_core_set_issi(struct mlx5_core_dev *dev) 502{ 503 u32 query_in[MLX5_ST_SZ_DW(query_issi_in)] = {0}; 504 u32 query_out[MLX5_ST_SZ_DW(query_issi_out)] = {0}; 505 u32 sup_issi; 506 int err; 507 508 MLX5_SET(query_issi_in, query_in, opcode, MLX5_CMD_OP_QUERY_ISSI); 509 510 err = mlx5_cmd_exec(dev, query_in, sizeof(query_in), query_out, sizeof(query_out)); 511 if (err) { 512 u32 syndrome; 513 u8 status; 514 515 mlx5_cmd_mbox_status(query_out, &status, &syndrome); 516 if (status == MLX5_CMD_STAT_BAD_OP_ERR) { 517 pr_debug("Only ISSI 0 is supported\n"); 518 return 0; 519 } 520 521 printf("mlx5_core: ERR: ""failed to query ISSI\n"); 522 return err; 523 } 524 525 sup_issi = MLX5_GET(query_issi_out, query_out, supported_issi_dw0); 526 527 if (sup_issi & (1 << 1)) { 528 u32 set_in[MLX5_ST_SZ_DW(set_issi_in)] = {0}; 529 u32 set_out[MLX5_ST_SZ_DW(set_issi_out)] = {0}; 530 531 MLX5_SET(set_issi_in, set_in, opcode, MLX5_CMD_OP_SET_ISSI); 532 MLX5_SET(set_issi_in, set_in, current_issi, 1); 533 534 err = mlx5_cmd_exec(dev, set_in, sizeof(set_in), set_out, sizeof(set_out)); 535 if (err) { 536 printf("mlx5_core: ERR: ""failed to set ISSI=1 err(%d)\n", err); 537 return err; 538 } 539 540 dev->issi = 1; 541 542 return 0; 543 } else if (sup_issi & (1 << 0)) { 544 return 0; 545 } 546 547 return -ENOTSUPP; 548} 549 550 551int mlx5_vector2eqn(struct mlx5_core_dev *dev, int vector, int *eqn, int *irqn) 552{ 553 struct mlx5_eq_table *table = &dev->priv.eq_table; 554 struct mlx5_eq *eq; 555 int err = -ENOENT; 556 557 spin_lock(&table->lock); 558 list_for_each_entry(eq, &table->comp_eqs_list, list) { 559 if (eq->index == vector) { 560 *eqn = eq->eqn; 561 *irqn = eq->irqn; 562 err = 0; 563 break; 564 } 565 } 566 spin_unlock(&table->lock); 567 568 return err; 569} 570EXPORT_SYMBOL(mlx5_vector2eqn); 571 572int mlx5_rename_eq(struct mlx5_core_dev *dev, int eq_ix, char *name) 573{ 574 struct mlx5_priv *priv = &dev->priv; 575 struct mlx5_eq_table *table = &priv->eq_table; 576 struct mlx5_eq *eq; 577 int err = -ENOENT; 578 579 spin_lock(&table->lock); 580 list_for_each_entry(eq, &table->comp_eqs_list, list) { 581 if (eq->index == eq_ix) { 582 int irq_ix = eq_ix + MLX5_EQ_VEC_COMP_BASE; 583 584 snprintf(priv->irq_info[irq_ix].name, MLX5_MAX_IRQ_NAME, 585 "%s-%d", name, eq_ix); 586 587 err = 0; 588 break; 589 } 590 } 591 spin_unlock(&table->lock); 592 593 return err; 594} 595 596static void free_comp_eqs(struct mlx5_core_dev *dev) 597{ 598 struct mlx5_eq_table *table = &dev->priv.eq_table; 599 struct mlx5_eq *eq, *n; 600 601 spin_lock(&table->lock); 602 list_for_each_entry_safe(eq, n, &table->comp_eqs_list, list) { 603 list_del(&eq->list); 604 spin_unlock(&table->lock); 605 if (mlx5_destroy_unmap_eq(dev, eq)) 606 mlx5_core_warn(dev, "failed to destroy EQ 0x%x\n", 607 eq->eqn); 608 kfree(eq); 609 spin_lock(&table->lock); 610 } 611 spin_unlock(&table->lock); 612} 613 614static int alloc_comp_eqs(struct mlx5_core_dev *dev) 615{ 616 struct mlx5_eq_table *table = &dev->priv.eq_table; 617 char name[MLX5_MAX_IRQ_NAME]; 618 struct mlx5_eq *eq; 619 int ncomp_vec; 620 int nent; 621 int err; 622 int i; 623 624 INIT_LIST_HEAD(&table->comp_eqs_list); 625 ncomp_vec = table->num_comp_vectors; 626 nent = MLX5_COMP_EQ_SIZE; 627 for (i = 0; i < ncomp_vec; i++) { 628 eq = kzalloc(sizeof(*eq), GFP_KERNEL); 629 630 snprintf(name, MLX5_MAX_IRQ_NAME, "mlx5_comp%d", i); 631 err = mlx5_create_map_eq(dev, eq, 632 i + MLX5_EQ_VEC_COMP_BASE, nent, 0, 633 name, &dev->priv.uuari.uars[0]); 634 if (err) { 635 kfree(eq); 636 goto clean; 637 } 638 mlx5_core_dbg(dev, "allocated completion EQN %d\n", eq->eqn); 639 eq->index = i; 640 spin_lock(&table->lock); 641 list_add_tail(&eq->list, &table->comp_eqs_list); 642 spin_unlock(&table->lock); 643 } 644 645 return 0; 646 647clean: 648 free_comp_eqs(dev); 649 return err; 650} 651 652static int map_bf_area(struct mlx5_core_dev *dev) 653{ 654 resource_size_t bf_start = pci_resource_start(dev->pdev, 0); 655 resource_size_t bf_len = pci_resource_len(dev->pdev, 0); 656 657 dev->priv.bf_mapping = io_mapping_create_wc(bf_start, bf_len); 658 659 return dev->priv.bf_mapping ? 0 : -ENOMEM; 660} 661 662static void unmap_bf_area(struct mlx5_core_dev *dev) 663{ 664 if (dev->priv.bf_mapping) 665 io_mapping_free(dev->priv.bf_mapping); 666} 667 668static inline int fw_initializing(struct mlx5_core_dev *dev) 669{ 670 return ioread32be(&dev->iseg->initializing) >> 31; 671} 672 673static int wait_fw_init(struct mlx5_core_dev *dev, u32 max_wait_mili) 674{ 675 u64 end = jiffies + msecs_to_jiffies(max_wait_mili); 676 int err = 0; 677 678 while (fw_initializing(dev)) { 679 if (time_after(jiffies, end)) { 680 err = -EBUSY; 681 break; 682 } 683 msleep(FW_INIT_WAIT_MS); 684 } 685 686 return err; 687} 688 689static void mlx5_add_device(struct mlx5_interface *intf, struct mlx5_priv *priv) 690{ 691 struct mlx5_device_context *dev_ctx; 692 struct mlx5_core_dev *dev = container_of(priv, struct mlx5_core_dev, priv); 693 694 dev_ctx = kzalloc(sizeof(*dev_ctx), GFP_KERNEL); 695 if (!dev_ctx) 696 return; 697 698 dev_ctx->intf = intf; 699 CURVNET_SET_QUIET(vnet0); 700 dev_ctx->context = intf->add(dev); 701 CURVNET_RESTORE(); 702 703 if (dev_ctx->context) { 704 spin_lock_irq(&priv->ctx_lock); 705 list_add_tail(&dev_ctx->list, &priv->ctx_list); 706 spin_unlock_irq(&priv->ctx_lock); 707 } else { 708 kfree(dev_ctx); 709 } 710} 711 712static void mlx5_remove_device(struct mlx5_interface *intf, struct mlx5_priv *priv) 713{ 714 struct mlx5_device_context *dev_ctx; 715 struct mlx5_core_dev *dev = container_of(priv, struct mlx5_core_dev, priv); 716 717 list_for_each_entry(dev_ctx, &priv->ctx_list, list) 718 if (dev_ctx->intf == intf) { 719 spin_lock_irq(&priv->ctx_lock); 720 list_del(&dev_ctx->list); 721 spin_unlock_irq(&priv->ctx_lock); 722 723 intf->remove(dev, dev_ctx->context); 724 kfree(dev_ctx); 725 return; 726 } 727} 728 729static int mlx5_register_device(struct mlx5_core_dev *dev) 730{ 731 struct mlx5_priv *priv = &dev->priv; 732 struct mlx5_interface *intf; 733 734 mutex_lock(&intf_mutex); 735 list_add_tail(&priv->dev_list, &dev_list); 736 list_for_each_entry(intf, &intf_list, list) 737 mlx5_add_device(intf, priv); 738 mutex_unlock(&intf_mutex); 739 740 return 0; 741} 742 743static void mlx5_unregister_device(struct mlx5_core_dev *dev) 744{ 745 struct mlx5_priv *priv = &dev->priv; 746 struct mlx5_interface *intf; 747 748 mutex_lock(&intf_mutex); 749 list_for_each_entry(intf, &intf_list, list) 750 mlx5_remove_device(intf, priv); 751 list_del(&priv->dev_list); 752 mutex_unlock(&intf_mutex); 753} 754 755int mlx5_register_interface(struct mlx5_interface *intf) 756{ 757 struct mlx5_priv *priv; 758 759 if (!intf->add || !intf->remove) 760 return -EINVAL; 761 762 mutex_lock(&intf_mutex); 763 list_add_tail(&intf->list, &intf_list); 764 list_for_each_entry(priv, &dev_list, dev_list) 765 mlx5_add_device(intf, priv); 766 mutex_unlock(&intf_mutex); 767 768 return 0; 769} 770EXPORT_SYMBOL(mlx5_register_interface); 771 772void mlx5_unregister_interface(struct mlx5_interface *intf) 773{ 774 struct mlx5_priv *priv; 775 776 mutex_lock(&intf_mutex); 777 list_for_each_entry(priv, &dev_list, dev_list) 778 mlx5_remove_device(intf, priv); 779 list_del(&intf->list); 780 mutex_unlock(&intf_mutex); 781} 782EXPORT_SYMBOL(mlx5_unregister_interface); 783 784void *mlx5_get_protocol_dev(struct mlx5_core_dev *mdev, int protocol) 785{ 786 struct mlx5_priv *priv = &mdev->priv; 787 struct mlx5_device_context *dev_ctx; 788 unsigned long flags; 789 void *result = NULL; 790 791 spin_lock_irqsave(&priv->ctx_lock, flags); 792 793 list_for_each_entry(dev_ctx, &mdev->priv.ctx_list, list) 794 if ((dev_ctx->intf->protocol == protocol) && 795 dev_ctx->intf->get_dev) { 796 result = dev_ctx->intf->get_dev(dev_ctx->context); 797 break; 798 } 799 800 spin_unlock_irqrestore(&priv->ctx_lock, flags); 801 802 return result; 803} 804EXPORT_SYMBOL(mlx5_get_protocol_dev); 805 806static int mlx5_pci_init(struct mlx5_core_dev *dev, struct mlx5_priv *priv) 807{ 808 struct pci_dev *pdev = dev->pdev; 809 int err = 0; 810 811 pci_set_drvdata(dev->pdev, dev); 812 strncpy(priv->name, dev_name(&pdev->dev), MLX5_MAX_NAME_LEN); 813 priv->name[MLX5_MAX_NAME_LEN - 1] = 0; 814 815 mutex_init(&priv->pgdir_mutex); 816 INIT_LIST_HEAD(&priv->pgdir_list); 817 spin_lock_init(&priv->mkey_lock); 818 819 priv->numa_node = NUMA_NO_NODE; 820 821 err = mlx5_pci_enable_device(dev); 822 if (err) { 823 device_printf((&pdev->dev)->bsddev, "ERR: ""Cannot enable PCI device, aborting\n"); 824 goto err_dbg; 825 } 826 827 err = request_bar(pdev); 828 if (err) { 829 device_printf((&pdev->dev)->bsddev, "ERR: ""error requesting BARs, aborting\n"); 830 goto err_disable; 831 } 832 833 pci_set_master(pdev); 834 835 err = set_dma_caps(pdev); 836 if (err) { 837 device_printf((&pdev->dev)->bsddev, "ERR: ""Failed setting DMA capabilities mask, aborting\n"); 838 goto err_clr_master; 839 } 840 841 dev->iseg_base = pci_resource_start(dev->pdev, 0); 842 dev->iseg = ioremap(dev->iseg_base, sizeof(*dev->iseg)); 843 if (!dev->iseg) { 844 err = -ENOMEM; 845 device_printf((&pdev->dev)->bsddev, "ERR: ""Failed mapping initialization segment, aborting\n"); 846 goto err_clr_master; 847 } 848 849 if (mlx5_vsc_find_cap(dev)) 850 dev_err(&pdev->dev, "Unable to find vendor specific capabilities\n"); 851 852 return 0; 853 854err_clr_master: 855 pci_clear_master(dev->pdev); 856 release_bar(dev->pdev); 857err_disable: 858 mlx5_pci_disable_device(dev); 859err_dbg: 860 return err; 861} 862 863static void mlx5_pci_close(struct mlx5_core_dev *dev, struct mlx5_priv *priv) 864{ 865 iounmap(dev->iseg); 866 pci_clear_master(dev->pdev); 867 release_bar(dev->pdev); 868 mlx5_pci_disable_device(dev); 869} 870 871static int mlx5_init_once(struct mlx5_core_dev *dev, struct mlx5_priv *priv) 872{ 873 struct pci_dev *pdev = dev->pdev; 874 int err; 875 876 mlx5_vsec_init(dev); 877 878 err = mlx5_query_hca_caps(dev); 879 if (err) { 880 dev_err(&pdev->dev, "query hca failed\n"); 881 goto out; 882 } 883 884 err = mlx5_query_board_id(dev); 885 if (err) { 886 dev_err(&pdev->dev, "query board id failed\n"); 887 goto out; 888 } 889 890 err = mlx5_eq_init(dev); 891 if (err) { 892 dev_err(&pdev->dev, "failed to initialize eq\n"); 893 goto out; 894 } 895 896 MLX5_INIT_DOORBELL_LOCK(&priv->cq_uar_lock); 897 898 err = mlx5_init_cq_table(dev); 899 if (err) { 900 dev_err(&pdev->dev, "failed to initialize cq table\n"); 901 goto err_eq_cleanup; 902 } 903 904 mlx5_init_qp_table(dev); 905 mlx5_init_srq_table(dev); 906 mlx5_init_mr_table(dev); 907 908 return 0; 909 910err_eq_cleanup: 911 mlx5_eq_cleanup(dev); 912 913out: 914 return err; 915} 916 917static void mlx5_cleanup_once(struct mlx5_core_dev *dev) 918{ 919 mlx5_cleanup_mr_table(dev); 920 mlx5_cleanup_srq_table(dev); 921 mlx5_cleanup_qp_table(dev); 922 mlx5_cleanup_cq_table(dev); 923 mlx5_eq_cleanup(dev); 924} 925 926static int mlx5_load_one(struct mlx5_core_dev *dev, struct mlx5_priv *priv, 927 bool boot) 928{ 929 struct pci_dev *pdev = dev->pdev; 930 int err; 931 932 mutex_lock(&dev->intf_state_mutex); 933 if (test_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state)) { 934 dev_warn(&dev->pdev->dev, "%s: interface is up, NOP\n", 935 __func__); 936 goto out; 937 } 938 939 device_printf((&pdev->dev)->bsddev, "INFO: ""firmware version: %d.%d.%d\n", fw_rev_maj(dev), fw_rev_min(dev), fw_rev_sub(dev)); 940 941 /* 942 * On load removing any previous indication of internal error, 943 * device is up 944 */ 945 dev->state = MLX5_DEVICE_STATE_UP; 946 947 err = mlx5_cmd_init(dev); 948 if (err) { 949 device_printf((&pdev->dev)->bsddev, "ERR: ""Failed initializing command interface, aborting\n"); 950 goto out_err; 951 } 952 953 err = wait_fw_init(dev, FW_INIT_TIMEOUT_MILI); 954 if (err) { 955 device_printf((&dev->pdev->dev)->bsddev, "ERR: ""Firmware over %d MS in initializing state, aborting\n", FW_INIT_TIMEOUT_MILI); 956 goto err_cmd_cleanup; 957 } 958 959 err = mlx5_core_enable_hca(dev); 960 if (err) { 961 device_printf((&pdev->dev)->bsddev, "ERR: ""enable hca failed\n"); 962 goto err_cmd_cleanup; 963 } 964 965 err = mlx5_core_set_issi(dev); 966 if (err) { 967 device_printf((&pdev->dev)->bsddev, "ERR: ""failed to set issi\n"); 968 goto err_disable_hca; 969 } 970 971 err = mlx5_pagealloc_start(dev); 972 if (err) { 973 device_printf((&pdev->dev)->bsddev, "ERR: ""mlx5_pagealloc_start failed\n"); 974 goto err_disable_hca; 975 } 976 977 err = mlx5_satisfy_startup_pages(dev, 1); 978 if (err) { 979 device_printf((&pdev->dev)->bsddev, "ERR: ""failed to allocate boot pages\n"); 980 goto err_pagealloc_stop; 981 } 982 983 err = set_hca_ctrl(dev); 984 if (err) { 985 device_printf((&pdev->dev)->bsddev, "ERR: ""set_hca_ctrl failed\n"); 986 goto reclaim_boot_pages; 987 } 988 989 err = handle_hca_cap(dev); 990 if (err) { 991 device_printf((&pdev->dev)->bsddev, "ERR: ""handle_hca_cap failed\n"); 992 goto reclaim_boot_pages; 993 } 994 995 err = handle_hca_cap_atomic(dev); 996 if (err) { 997 device_printf((&pdev->dev)->bsddev, "ERR: ""handle_hca_cap_atomic failed\n"); 998 goto reclaim_boot_pages; 999 } 1000 1001 err = mlx5_satisfy_startup_pages(dev, 0); 1002 if (err) { 1003 device_printf((&pdev->dev)->bsddev, "ERR: ""failed to allocate init pages\n"); 1004 goto reclaim_boot_pages; 1005 } 1006 1007 err = mlx5_cmd_init_hca(dev); 1008 if (err) { 1009 device_printf((&pdev->dev)->bsddev, "ERR: ""init hca failed\n"); 1010 goto reclaim_boot_pages; 1011 } 1012 1013 mlx5_start_health_poll(dev); 1014 1015 if (boot && mlx5_init_once(dev, priv)) { 1016 dev_err(&pdev->dev, "sw objs init failed\n"); 1017 goto err_stop_poll; 1018 } 1019 1020 err = mlx5_enable_msix(dev); 1021 if (err) { 1022 device_printf((&pdev->dev)->bsddev, "ERR: ""enable msix failed\n"); 1023 goto err_cleanup_once; 1024 } 1025 1026 err = mlx5_alloc_uuars(dev, &priv->uuari); 1027 if (err) { 1028 device_printf((&pdev->dev)->bsddev, "ERR: ""Failed allocating uar, aborting\n"); 1029 goto err_disable_msix; 1030 } 1031 1032 err = mlx5_start_eqs(dev); 1033 if (err) { 1034 device_printf((&pdev->dev)->bsddev, "ERR: ""Failed to start pages and async EQs\n"); 1035 goto err_free_uar; 1036 } 1037 1038 err = alloc_comp_eqs(dev); 1039 if (err) { 1040 device_printf((&pdev->dev)->bsddev, "ERR: ""Failed to alloc completion EQs\n"); 1041 goto err_stop_eqs; 1042 } 1043 1044 if (map_bf_area(dev)) 1045 device_printf((&pdev->dev)->bsddev, "ERR: ""Failed to map blue flame area\n"); 1046 1047 err = mlx5_init_fs(dev); 1048 if (err) { 1049 mlx5_core_err(dev, "flow steering init %d\n", err); 1050 goto err_free_comp_eqs; 1051 } 1052 1053 err = mlx5_register_device(dev); 1054 if (err) { 1055 dev_err(&pdev->dev, "mlx5_register_device failed %d\n", err); 1056 goto err_fs; 1057 } 1058 1059 clear_bit(MLX5_INTERFACE_STATE_DOWN, &dev->intf_state); 1060 set_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state); 1061 1062out: 1063 mutex_unlock(&dev->intf_state_mutex); 1064 return 0; 1065 1066err_fs: 1067 mlx5_cleanup_fs(dev); 1068 1069err_free_comp_eqs: 1070 free_comp_eqs(dev); 1071 unmap_bf_area(dev); 1072 1073err_stop_eqs: 1074 mlx5_stop_eqs(dev); 1075 1076err_free_uar: 1077 mlx5_free_uuars(dev, &priv->uuari); 1078 1079err_disable_msix: 1080 mlx5_disable_msix(dev); 1081 1082err_cleanup_once: 1083 if (boot) 1084 mlx5_cleanup_once(dev); 1085 1086err_stop_poll: 1087 mlx5_stop_health_poll(dev); 1088 if (mlx5_cmd_teardown_hca(dev)) { 1089 device_printf((&dev->pdev->dev)->bsddev, "ERR: ""tear_down_hca failed, skip cleanup\n"); 1090 goto out_err; 1091 } 1092 1093reclaim_boot_pages: 1094 mlx5_reclaim_startup_pages(dev); 1095 1096err_pagealloc_stop: 1097 mlx5_pagealloc_stop(dev); 1098 1099err_disable_hca: 1100 mlx5_core_disable_hca(dev); 1101 1102err_cmd_cleanup: 1103 mlx5_cmd_cleanup(dev); 1104 1105out_err: 1106 dev->state = MLX5_DEVICE_STATE_INTERNAL_ERROR; 1107 mutex_unlock(&dev->intf_state_mutex); 1108 1109 return err; 1110} 1111 1112static int mlx5_unload_one(struct mlx5_core_dev *dev, struct mlx5_priv *priv, 1113 bool cleanup) 1114{ 1115 int err = 0; 1116 1117 if (cleanup) 1118 mlx5_drain_health_recovery(dev); 1119 1120 mutex_lock(&dev->intf_state_mutex); 1121 if (test_bit(MLX5_INTERFACE_STATE_DOWN, &dev->intf_state)) { 1122 dev_warn(&dev->pdev->dev, "%s: interface is down, NOP\n", __func__); 1123 if (cleanup) 1124 mlx5_cleanup_once(dev); 1125 goto out; 1126 } 1127 1128 mlx5_unregister_device(dev); 1129 1130 mlx5_cleanup_fs(dev); 1131 unmap_bf_area(dev); 1132 mlx5_wait_for_reclaim_vfs_pages(dev); 1133 free_comp_eqs(dev); 1134 mlx5_stop_eqs(dev); 1135 mlx5_free_uuars(dev, &priv->uuari); 1136 mlx5_disable_msix(dev); 1137 if (cleanup) 1138 mlx5_cleanup_once(dev); 1139 mlx5_stop_health_poll(dev); 1140 err = mlx5_cmd_teardown_hca(dev); 1141 if (err) { 1142 device_printf((&dev->pdev->dev)->bsddev, "ERR: ""tear_down_hca failed, skip cleanup\n"); 1143 goto out; 1144 } 1145 mlx5_pagealloc_stop(dev); 1146 mlx5_reclaim_startup_pages(dev); 1147 mlx5_core_disable_hca(dev); 1148 mlx5_cmd_cleanup(dev); 1149 1150out: 1151 clear_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state); 1152 set_bit(MLX5_INTERFACE_STATE_DOWN, &dev->intf_state); 1153 mutex_unlock(&dev->intf_state_mutex); 1154 return err; 1155} 1156 1157void mlx5_core_event(struct mlx5_core_dev *dev, enum mlx5_dev_event event, 1158 unsigned long param) 1159{ 1160 struct mlx5_priv *priv = &dev->priv; 1161 struct mlx5_device_context *dev_ctx; 1162 unsigned long flags; 1163 1164 spin_lock_irqsave(&priv->ctx_lock, flags); 1165 1166 list_for_each_entry(dev_ctx, &priv->ctx_list, list) 1167 if (dev_ctx->intf->event) 1168 dev_ctx->intf->event(dev, dev_ctx->context, event, param); 1169 1170 spin_unlock_irqrestore(&priv->ctx_lock, flags); 1171} 1172 1173struct mlx5_core_event_handler { 1174 void (*event)(struct mlx5_core_dev *dev, 1175 enum mlx5_dev_event event, 1176 void *data); 1177}; 1178 1179static int init_one(struct pci_dev *pdev, 1180 const struct pci_device_id *id) 1181{ 1182 struct mlx5_core_dev *dev; 1183 struct mlx5_priv *priv; 1184 int err; 1185 1186 dev = kzalloc(sizeof(*dev), GFP_KERNEL); 1187 priv = &dev->priv; 1188 if (id) 1189 priv->pci_dev_data = id->driver_data; 1190 1191 if (prof_sel < 0 || prof_sel >= ARRAY_SIZE(profiles)) { 1192 printf("mlx5_core: WARN: ""selected profile out of range, selecting default (%d)\n", MLX5_DEFAULT_PROF); 1193 prof_sel = MLX5_DEFAULT_PROF; 1194 } 1195 dev->profile = &profiles[prof_sel]; 1196 dev->pdev = pdev; 1197 dev->event = mlx5_core_event; 1198 1199 INIT_LIST_HEAD(&priv->ctx_list); 1200 spin_lock_init(&priv->ctx_lock); 1201 mutex_init(&dev->pci_status_mutex); 1202 mutex_init(&dev->intf_state_mutex); 1203 err = mlx5_pci_init(dev, priv); 1204 if (err) { 1205 device_printf((&pdev->dev)->bsddev, "ERR: ""mlx5_pci_init failed %d\n", err); 1206 goto clean_dev; 1207 } 1208 1209 err = mlx5_health_init(dev); 1210 if (err) { 1211 device_printf((&pdev->dev)->bsddev, "ERR: ""mlx5_health_init failed %d\n", err); 1212 goto close_pci; 1213 } 1214 1215 mlx5_pagealloc_init(dev); 1216 1217 err = mlx5_load_one(dev, priv, true); 1218 if (err) { 1219 device_printf((&pdev->dev)->bsddev, "ERR: ""mlx5_register_device failed %d\n", err); 1220 goto clean_health; 1221 } 1222 1223 mlx5_fwdump_prep(dev); 1224 1225 pci_save_state(pdev->dev.bsddev); 1226 return 0; 1227 1228clean_health: 1229 mlx5_pagealloc_cleanup(dev); 1230 mlx5_health_cleanup(dev); 1231close_pci: 1232 mlx5_pci_close(dev, priv); 1233clean_dev: 1234 kfree(dev); 1235 return err; 1236} 1237 1238static void remove_one(struct pci_dev *pdev) 1239{ 1240 struct mlx5_core_dev *dev = pci_get_drvdata(pdev); 1241 struct mlx5_priv *priv = &dev->priv; 1242 1243 if (mlx5_unload_one(dev, priv, true)) { 1244 dev_err(&dev->pdev->dev, "mlx5_unload_one failed\n"); 1245 mlx5_health_cleanup(dev); 1246 return; 1247 } 1248 1249 mlx5_fwdump_clean(dev); 1250 mlx5_pagealloc_cleanup(dev); 1251 mlx5_health_cleanup(dev); 1252 mlx5_pci_close(dev, priv); 1253 pci_set_drvdata(pdev, NULL); 1254 kfree(dev); 1255} 1256 1257static pci_ers_result_t mlx5_pci_err_detected(struct pci_dev *pdev, 1258 pci_channel_state_t state) 1259{ 1260 struct mlx5_core_dev *dev = pci_get_drvdata(pdev); 1261 struct mlx5_priv *priv = &dev->priv; 1262 1263 dev_info(&pdev->dev, "%s was called\n", __func__); 1264 mlx5_enter_error_state(dev, false); 1265 mlx5_unload_one(dev, priv, false); 1266 1267 if (state) { 1268 mlx5_drain_health_wq(dev); 1269 mlx5_pci_disable_device(dev); 1270 } 1271 1272 return state == pci_channel_io_perm_failure ? 1273 PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_NEED_RESET; 1274} 1275 1276static pci_ers_result_t mlx5_pci_slot_reset(struct pci_dev *pdev) 1277{ 1278 struct mlx5_core_dev *dev = pci_get_drvdata(pdev); 1279 int err = 0; 1280 1281 dev_info(&pdev->dev, "%s was called\n", __func__); 1282 1283 err = mlx5_pci_enable_device(dev); 1284 if (err) { 1285 dev_err(&pdev->dev, "%s: mlx5_pci_enable_device failed with error code: %d\n" 1286 , __func__, err); 1287 return PCI_ERS_RESULT_DISCONNECT; 1288 } 1289 pci_set_master(pdev); 1290 pci_set_powerstate(pdev->dev.bsddev, PCI_POWERSTATE_D0); 1291 pci_restore_state(pdev->dev.bsddev); 1292 pci_save_state(pdev->dev.bsddev); 1293 1294 return err ? PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_RECOVERED; 1295} 1296 1297/* wait for the device to show vital signs. For now we check 1298 * that we can read the device ID and that the health buffer 1299 * shows a non zero value which is different than 0xffffffff 1300 */ 1301static void wait_vital(struct pci_dev *pdev) 1302{ 1303 struct mlx5_core_dev *dev = pci_get_drvdata(pdev); 1304 struct mlx5_core_health *health = &dev->priv.health; 1305 const int niter = 100; 1306 u32 count; 1307 u16 did; 1308 int i; 1309 1310 /* Wait for firmware to be ready after reset */ 1311 msleep(1000); 1312 for (i = 0; i < niter; i++) { 1313 if (pci_read_config_word(pdev, 2, &did)) { 1314 dev_warn(&pdev->dev, "failed reading config word\n"); 1315 break; 1316 } 1317 if (did == pdev->device) { 1318 dev_info(&pdev->dev, "device ID correctly read after %d iterations\n", i); 1319 break; 1320 } 1321 msleep(50); 1322 } 1323 if (i == niter) 1324 dev_warn(&pdev->dev, "%s-%d: could not read device ID\n", __func__, __LINE__); 1325 1326 for (i = 0; i < niter; i++) { 1327 count = ioread32be(health->health_counter); 1328 if (count && count != 0xffffffff) { 1329 dev_info(&pdev->dev, "Counter value 0x%x after %d iterations\n", count, i); 1330 break; 1331 } 1332 msleep(50); 1333 } 1334 1335 if (i == niter) 1336 dev_warn(&pdev->dev, "%s-%d: could not read device ID\n", __func__, __LINE__); 1337} 1338 1339static void mlx5_pci_resume(struct pci_dev *pdev) 1340{ 1341 struct mlx5_core_dev *dev = pci_get_drvdata(pdev); 1342 struct mlx5_priv *priv = &dev->priv; 1343 int err; 1344 1345 dev_info(&pdev->dev, "%s was called\n", __func__); 1346 1347 wait_vital(pdev); 1348 1349 err = mlx5_load_one(dev, priv, false); 1350 if (err) 1351 dev_err(&pdev->dev, "%s: mlx5_load_one failed with error code: %d\n" 1352 , __func__, err); 1353 else 1354 dev_info(&pdev->dev, "%s: device recovered\n", __func__); 1355} 1356 1357static const struct pci_error_handlers mlx5_err_handler = { 1358 .error_detected = mlx5_pci_err_detected, 1359 .slot_reset = mlx5_pci_slot_reset, 1360 .resume = mlx5_pci_resume 1361}; 1362 1363static int mlx5_try_fast_unload(struct mlx5_core_dev *dev) 1364{ 1365 int err; 1366 1367 if (!MLX5_CAP_GEN(dev, force_teardown)) { 1368 mlx5_core_dbg(dev, "force teardown is not supported in the firmware\n"); 1369 return -EOPNOTSUPP; 1370 } 1371 1372 if (dev->state == MLX5_DEVICE_STATE_INTERNAL_ERROR) { 1373 mlx5_core_dbg(dev, "Device in internal error state, giving up\n"); 1374 return -EAGAIN; 1375 } 1376 1377 err = mlx5_cmd_force_teardown_hca(dev); 1378 if (err) { 1379 mlx5_core_dbg(dev, "Firmware couldn't do fast unload error: %d\n", err); 1380 return err; 1381 } 1382 1383 mlx5_enter_error_state(dev, true); 1384 1385 return 0; 1386} 1387 1388static void shutdown_one(struct pci_dev *pdev) 1389{ 1390 struct mlx5_core_dev *dev = pci_get_drvdata(pdev); 1391 struct mlx5_priv *priv = &dev->priv; 1392 int err; 1393 1394 set_bit(MLX5_INTERFACE_STATE_SHUTDOWN, &dev->intf_state); 1395 err = mlx5_try_fast_unload(dev); 1396 if (err) 1397 mlx5_unload_one(dev, priv, false); 1398 mlx5_pci_disable_device(dev); 1399} 1400 1401static const struct pci_device_id mlx5_core_pci_table[] = { 1402 { PCI_VDEVICE(MELLANOX, 4113) }, /* Connect-IB */ 1403 { PCI_VDEVICE(MELLANOX, 4114) }, /* Connect-IB VF */ 1404 { PCI_VDEVICE(MELLANOX, 4115) }, /* ConnectX-4 */ 1405 { PCI_VDEVICE(MELLANOX, 4116) }, /* ConnectX-4 VF */ 1406 { PCI_VDEVICE(MELLANOX, 4117) }, /* ConnectX-4LX */ 1407 { PCI_VDEVICE(MELLANOX, 4118) }, /* ConnectX-4LX VF */ 1408 { PCI_VDEVICE(MELLANOX, 4119) }, /* ConnectX-5 */ 1409 { PCI_VDEVICE(MELLANOX, 4120) }, /* ConnectX-5 VF */ 1410 { PCI_VDEVICE(MELLANOX, 4121) }, 1411 { PCI_VDEVICE(MELLANOX, 4122) }, 1412 { PCI_VDEVICE(MELLANOX, 4123) }, 1413 { PCI_VDEVICE(MELLANOX, 4124) }, 1414 { PCI_VDEVICE(MELLANOX, 4125) }, 1415 { PCI_VDEVICE(MELLANOX, 4126) }, 1416 { PCI_VDEVICE(MELLANOX, 4127) }, 1417 { PCI_VDEVICE(MELLANOX, 4128) }, 1418 { PCI_VDEVICE(MELLANOX, 4129) }, 1419 { PCI_VDEVICE(MELLANOX, 4130) }, 1420 { PCI_VDEVICE(MELLANOX, 4131) }, 1421 { PCI_VDEVICE(MELLANOX, 4132) }, 1422 { PCI_VDEVICE(MELLANOX, 4133) }, 1423 { PCI_VDEVICE(MELLANOX, 4134) }, 1424 { PCI_VDEVICE(MELLANOX, 4135) }, 1425 { PCI_VDEVICE(MELLANOX, 4136) }, 1426 { PCI_VDEVICE(MELLANOX, 4137) }, 1427 { PCI_VDEVICE(MELLANOX, 4138) }, 1428 { PCI_VDEVICE(MELLANOX, 4139) }, 1429 { PCI_VDEVICE(MELLANOX, 4140) }, 1430 { PCI_VDEVICE(MELLANOX, 4141) }, 1431 { PCI_VDEVICE(MELLANOX, 4142) }, 1432 { PCI_VDEVICE(MELLANOX, 4143) }, 1433 { PCI_VDEVICE(MELLANOX, 4144) }, 1434 { 0, } 1435}; 1436 1437MODULE_DEVICE_TABLE(pci, mlx5_core_pci_table); 1438 1439void mlx5_disable_device(struct mlx5_core_dev *dev) 1440{ 1441 mlx5_pci_err_detected(dev->pdev, 0); 1442} 1443 1444void mlx5_recover_device(struct mlx5_core_dev *dev) 1445{ 1446 mlx5_pci_disable_device(dev); 1447 if (mlx5_pci_slot_reset(dev->pdev) == PCI_ERS_RESULT_RECOVERED) 1448 mlx5_pci_resume(dev->pdev); 1449} 1450 1451struct pci_driver mlx5_core_driver = { 1452 .name = DRIVER_NAME, 1453 .id_table = mlx5_core_pci_table, 1454 .shutdown = shutdown_one, 1455 .probe = init_one, 1456 .remove = remove_one, 1457 .err_handler = &mlx5_err_handler 1458}; 1459 1460static int __init init(void) 1461{ 1462 int err; 1463 1464 err = pci_register_driver(&mlx5_core_driver); 1465 if (err) 1466 goto err_debug; 1467 1468 err = mlx5_fwdump_init(); 1469 if (err) 1470 goto err_fwdump; 1471 1472 return 0; 1473 1474err_fwdump: 1475 pci_unregister_driver(&mlx5_core_driver); 1476 1477err_debug: 1478 return err; 1479} 1480 1481static void __exit cleanup(void) 1482{ 1483 mlx5_fwdump_fini(); 1484 pci_unregister_driver(&mlx5_core_driver); 1485} 1486 1487module_init(init); 1488module_exit(cleanup); 1489