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 * $FreeBSD$ 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 <dev/mlx5/mlx5_fpga/core.h> 45 #include <dev/mlx5/mlx5_lib/mlx5.h> 46 #include "mlx5_core.h" 47 #include "fs_core.h" 48 49 static const char mlx5_version[] = "Mellanox Core driver " 50 DRIVER_VERSION " (" DRIVER_RELDATE ")"; 51 MODULE_AUTHOR("Eli Cohen <eli@mellanox.com>"); 52 MODULE_DESCRIPTION("Mellanox Connect-IB, ConnectX-4 core driver"); 53 MODULE_LICENSE("Dual BSD/GPL"); 54 #if (__FreeBSD_version >= 1100000) 55 MODULE_DEPEND(mlx5, linuxkpi, 1, 1, 1); 56 #endif 57 MODULE_VERSION(mlx5, 1); 58 59 int mlx5_core_debug_mask; 60 module_param_named(debug_mask, mlx5_core_debug_mask, int, 0644); 61 MODULE_PARM_DESC(debug_mask, "debug mask: 1 = dump cmd data, 2 = dump cmd exec time, 3 = both. Default=0"); 62 63 #define MLX5_DEFAULT_PROF 2 64 static int prof_sel = MLX5_DEFAULT_PROF; 65 module_param_named(prof_sel, prof_sel, int, 0444); 66 MODULE_PARM_DESC(prof_sel, "profile selector. Valid range 0 - 2"); 67 68 SYSCTL_NODE(_hw, OID_AUTO, mlx5, CTLFLAG_RW, 0, "mlx5 HW controls"); 69 70 #define NUMA_NO_NODE -1 71 72 static LIST_HEAD(intf_list); 73 static LIST_HEAD(dev_list); 74 static DEFINE_MUTEX(intf_mutex); 75 76 struct mlx5_device_context { 77 struct list_head list; 78 struct mlx5_interface *intf; 79 void *context; 80 }; 81 82 enum { 83 MLX5_ATOMIC_REQ_MODE_BE = 0x0, 84 MLX5_ATOMIC_REQ_MODE_HOST_ENDIANNESS = 0x1, 85 }; 86 87 static struct mlx5_profile profiles[] = { 88 [0] = { 89 .mask = 0, 90 }, 91 [1] = { 92 .mask = MLX5_PROF_MASK_QP_SIZE, 93 .log_max_qp = 12, 94 }, 95 [2] = { 96 .mask = MLX5_PROF_MASK_QP_SIZE | 97 MLX5_PROF_MASK_MR_CACHE, 98 .log_max_qp = 17, 99 .mr_cache[0] = { 100 .size = 500, 101 .limit = 250 102 }, 103 .mr_cache[1] = { 104 .size = 500, 105 .limit = 250 106 }, 107 .mr_cache[2] = { 108 .size = 500, 109 .limit = 250 110 }, 111 .mr_cache[3] = { 112 .size = 500, 113 .limit = 250 114 }, 115 .mr_cache[4] = { 116 .size = 500, 117 .limit = 250 118 }, 119 .mr_cache[5] = { 120 .size = 500, 121 .limit = 250 122 }, 123 .mr_cache[6] = { 124 .size = 500, 125 .limit = 250 126 }, 127 .mr_cache[7] = { 128 .size = 500, 129 .limit = 250 130 }, 131 .mr_cache[8] = { 132 .size = 500, 133 .limit = 250 134 }, 135 .mr_cache[9] = { 136 .size = 500, 137 .limit = 250 138 }, 139 .mr_cache[10] = { 140 .size = 500, 141 .limit = 250 142 }, 143 .mr_cache[11] = { 144 .size = 500, 145 .limit = 250 146 }, 147 .mr_cache[12] = { 148 .size = 64, 149 .limit = 32 150 }, 151 .mr_cache[13] = { 152 .size = 32, 153 .limit = 16 154 }, 155 .mr_cache[14] = { 156 .size = 16, 157 .limit = 8 158 }, 159 }, 160 [3] = { 161 .mask = MLX5_PROF_MASK_QP_SIZE, 162 .log_max_qp = 17, 163 }, 164 }; 165 166 static int set_dma_caps(struct pci_dev *pdev) 167 { 168 int err; 169 170 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64)); 171 if (err) { 172 device_printf((&pdev->dev)->bsddev, "WARN: ""Warning: couldn't set 64-bit PCI DMA mask\n"); 173 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32)); 174 if (err) { 175 device_printf((&pdev->dev)->bsddev, "ERR: ""Can't set PCI DMA mask, aborting\n"); 176 return err; 177 } 178 } 179 180 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64)); 181 if (err) { 182 device_printf((&pdev->dev)->bsddev, "WARN: ""Warning: couldn't set 64-bit consistent PCI DMA mask\n"); 183 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32)); 184 if (err) { 185 device_printf((&pdev->dev)->bsddev, "ERR: ""Can't set consistent PCI DMA mask, aborting\n"); 186 return err; 187 } 188 } 189 190 dma_set_max_seg_size(&pdev->dev, 2u * 1024 * 1024 * 1024); 191 return err; 192 } 193 194 static int mlx5_pci_enable_device(struct mlx5_core_dev *dev) 195 { 196 struct pci_dev *pdev = dev->pdev; 197 int err = 0; 198 199 mutex_lock(&dev->pci_status_mutex); 200 if (dev->pci_status == MLX5_PCI_STATUS_DISABLED) { 201 err = pci_enable_device(pdev); 202 if (!err) 203 dev->pci_status = MLX5_PCI_STATUS_ENABLED; 204 } 205 mutex_unlock(&dev->pci_status_mutex); 206 207 return err; 208 } 209 210 static void mlx5_pci_disable_device(struct mlx5_core_dev *dev) 211 { 212 struct pci_dev *pdev = dev->pdev; 213 214 mutex_lock(&dev->pci_status_mutex); 215 if (dev->pci_status == MLX5_PCI_STATUS_ENABLED) { 216 pci_disable_device(pdev); 217 dev->pci_status = MLX5_PCI_STATUS_DISABLED; 218 } 219 mutex_unlock(&dev->pci_status_mutex); 220 } 221 222 static int request_bar(struct pci_dev *pdev) 223 { 224 int err = 0; 225 226 if (!(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) { 227 device_printf((&pdev->dev)->bsddev, "ERR: ""Missing registers BAR, aborting\n"); 228 return -ENODEV; 229 } 230 231 err = pci_request_regions(pdev, DRIVER_NAME); 232 if (err) 233 device_printf((&pdev->dev)->bsddev, "ERR: ""Couldn't get PCI resources, aborting\n"); 234 235 return err; 236 } 237 238 static void release_bar(struct pci_dev *pdev) 239 { 240 pci_release_regions(pdev); 241 } 242 243 static int mlx5_enable_msix(struct mlx5_core_dev *dev) 244 { 245 struct mlx5_priv *priv = &dev->priv; 246 struct mlx5_eq_table *table = &priv->eq_table; 247 int num_eqs = 1 << MLX5_CAP_GEN(dev, log_max_eq); 248 int limit = dev->msix_eqvec; 249 int nvec = MLX5_EQ_VEC_COMP_BASE; 250 int i; 251 252 if (limit > 0) 253 nvec += limit; 254 else 255 nvec += MLX5_CAP_GEN(dev, num_ports) * num_online_cpus(); 256 257 nvec = min_t(int, nvec, num_eqs); 258 if (nvec <= MLX5_EQ_VEC_COMP_BASE) 259 return -ENOMEM; 260 261 priv->msix_arr = kzalloc(nvec * sizeof(*priv->msix_arr), GFP_KERNEL); 262 263 priv->irq_info = kzalloc(nvec * sizeof(*priv->irq_info), GFP_KERNEL); 264 265 for (i = 0; i < nvec; i++) 266 priv->msix_arr[i].entry = i; 267 268 nvec = pci_enable_msix_range(dev->pdev, priv->msix_arr, 269 MLX5_EQ_VEC_COMP_BASE + 1, nvec); 270 if (nvec < 0) 271 return nvec; 272 273 table->num_comp_vectors = nvec - MLX5_EQ_VEC_COMP_BASE; 274 275 return 0; 276 277 } 278 279 static void mlx5_disable_msix(struct mlx5_core_dev *dev) 280 { 281 struct mlx5_priv *priv = &dev->priv; 282 283 pci_disable_msix(dev->pdev); 284 kfree(priv->irq_info); 285 kfree(priv->msix_arr); 286 } 287 288 struct mlx5_reg_host_endianess { 289 u8 he; 290 u8 rsvd[15]; 291 }; 292 293 294 #define CAP_MASK(pos, size) ((u64)((1 << (size)) - 1) << (pos)) 295 296 enum { 297 MLX5_CAP_BITS_RW_MASK = CAP_MASK(MLX5_CAP_OFF_CMDIF_CSUM, 2) | 298 MLX5_DEV_CAP_FLAG_DCT | 299 MLX5_DEV_CAP_FLAG_DRAIN_SIGERR, 300 }; 301 302 static u16 to_fw_pkey_sz(u32 size) 303 { 304 switch (size) { 305 case 128: 306 return 0; 307 case 256: 308 return 1; 309 case 512: 310 return 2; 311 case 1024: 312 return 3; 313 case 2048: 314 return 4; 315 case 4096: 316 return 5; 317 default: 318 printf("mlx5_core: WARN: ""invalid pkey table size %d\n", size); 319 return 0; 320 } 321 } 322 323 static int mlx5_core_get_caps_mode(struct mlx5_core_dev *dev, 324 enum mlx5_cap_type cap_type, 325 enum mlx5_cap_mode cap_mode) 326 { 327 u8 in[MLX5_ST_SZ_BYTES(query_hca_cap_in)]; 328 int out_sz = MLX5_ST_SZ_BYTES(query_hca_cap_out); 329 void *out, *hca_caps; 330 u16 opmod = (cap_type << 1) | (cap_mode & 0x01); 331 int err; 332 333 memset(in, 0, sizeof(in)); 334 out = kzalloc(out_sz, GFP_KERNEL); 335 336 MLX5_SET(query_hca_cap_in, in, opcode, MLX5_CMD_OP_QUERY_HCA_CAP); 337 MLX5_SET(query_hca_cap_in, in, op_mod, opmod); 338 err = mlx5_cmd_exec(dev, in, sizeof(in), out, out_sz); 339 if (err) { 340 mlx5_core_warn(dev, 341 "QUERY_HCA_CAP : type(%x) opmode(%x) Failed(%d)\n", 342 cap_type, cap_mode, err); 343 goto query_ex; 344 } 345 346 hca_caps = MLX5_ADDR_OF(query_hca_cap_out, out, capability); 347 348 switch (cap_mode) { 349 case HCA_CAP_OPMOD_GET_MAX: 350 memcpy(dev->hca_caps_max[cap_type], hca_caps, 351 MLX5_UN_SZ_BYTES(hca_cap_union)); 352 break; 353 case HCA_CAP_OPMOD_GET_CUR: 354 memcpy(dev->hca_caps_cur[cap_type], hca_caps, 355 MLX5_UN_SZ_BYTES(hca_cap_union)); 356 break; 357 default: 358 mlx5_core_warn(dev, 359 "Tried to query dev cap type(%x) with wrong opmode(%x)\n", 360 cap_type, cap_mode); 361 err = -EINVAL; 362 break; 363 } 364 query_ex: 365 kfree(out); 366 return err; 367 } 368 369 int mlx5_core_get_caps(struct mlx5_core_dev *dev, enum mlx5_cap_type cap_type) 370 { 371 int ret; 372 373 ret = mlx5_core_get_caps_mode(dev, cap_type, HCA_CAP_OPMOD_GET_CUR); 374 if (ret) 375 return ret; 376 377 return mlx5_core_get_caps_mode(dev, cap_type, HCA_CAP_OPMOD_GET_MAX); 378 } 379 380 static int set_caps(struct mlx5_core_dev *dev, void *in, int in_sz) 381 { 382 u32 out[MLX5_ST_SZ_DW(set_hca_cap_out)] = {0}; 383 384 MLX5_SET(set_hca_cap_in, in, opcode, MLX5_CMD_OP_SET_HCA_CAP); 385 386 return mlx5_cmd_exec(dev, in, in_sz, out, sizeof(out)); 387 } 388 389 static int handle_hca_cap(struct mlx5_core_dev *dev) 390 { 391 void *set_ctx = NULL; 392 struct mlx5_profile *prof = dev->profile; 393 int err = -ENOMEM; 394 int set_sz = MLX5_ST_SZ_BYTES(set_hca_cap_in); 395 void *set_hca_cap; 396 397 set_ctx = kzalloc(set_sz, GFP_KERNEL); 398 399 err = mlx5_core_get_caps(dev, MLX5_CAP_GENERAL); 400 if (err) 401 goto query_ex; 402 403 set_hca_cap = MLX5_ADDR_OF(set_hca_cap_in, set_ctx, 404 capability); 405 memcpy(set_hca_cap, dev->hca_caps_cur[MLX5_CAP_GENERAL], 406 MLX5_ST_SZ_BYTES(cmd_hca_cap)); 407 408 mlx5_core_dbg(dev, "Current Pkey table size %d Setting new size %d\n", 409 mlx5_to_sw_pkey_sz(MLX5_CAP_GEN(dev, pkey_table_size)), 410 128); 411 /* we limit the size of the pkey table to 128 entries for now */ 412 MLX5_SET(cmd_hca_cap, set_hca_cap, pkey_table_size, 413 to_fw_pkey_sz(128)); 414 415 if (prof->mask & MLX5_PROF_MASK_QP_SIZE) 416 MLX5_SET(cmd_hca_cap, set_hca_cap, log_max_qp, 417 prof->log_max_qp); 418 419 /* disable cmdif checksum */ 420 MLX5_SET(cmd_hca_cap, set_hca_cap, cmdif_checksum, 0); 421 422 /* enable drain sigerr */ 423 MLX5_SET(cmd_hca_cap, set_hca_cap, drain_sigerr, 1); 424 425 MLX5_SET(cmd_hca_cap, set_hca_cap, log_uar_page_sz, PAGE_SHIFT - 12); 426 427 err = set_caps(dev, set_ctx, set_sz); 428 429 query_ex: 430 kfree(set_ctx); 431 return err; 432 } 433 434 static int handle_hca_cap_atomic(struct mlx5_core_dev *dev) 435 { 436 void *set_ctx; 437 void *set_hca_cap; 438 int set_sz = MLX5_ST_SZ_BYTES(set_hca_cap_in); 439 int req_endianness; 440 int err; 441 442 if (MLX5_CAP_GEN(dev, atomic)) { 443 err = mlx5_core_get_caps(dev, MLX5_CAP_ATOMIC); 444 if (err) 445 return err; 446 } else { 447 return 0; 448 } 449 450 req_endianness = 451 MLX5_CAP_ATOMIC(dev, 452 supported_atomic_req_8B_endianess_mode_1); 453 454 if (req_endianness != MLX5_ATOMIC_REQ_MODE_HOST_ENDIANNESS) 455 return 0; 456 457 set_ctx = kzalloc(set_sz, GFP_KERNEL); 458 if (!set_ctx) 459 return -ENOMEM; 460 461 MLX5_SET(set_hca_cap_in, set_ctx, op_mod, 462 MLX5_SET_HCA_CAP_OP_MOD_ATOMIC << 1); 463 set_hca_cap = MLX5_ADDR_OF(set_hca_cap_in, set_ctx, capability); 464 465 /* Set requestor to host endianness */ 466 MLX5_SET(atomic_caps, set_hca_cap, atomic_req_8B_endianess_mode, 467 MLX5_ATOMIC_REQ_MODE_HOST_ENDIANNESS); 468 469 err = set_caps(dev, set_ctx, set_sz); 470 471 kfree(set_ctx); 472 return err; 473 } 474 475 static int set_hca_ctrl(struct mlx5_core_dev *dev) 476 { 477 struct mlx5_reg_host_endianess he_in; 478 struct mlx5_reg_host_endianess he_out; 479 int err; 480 481 if (MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_ETH && 482 !MLX5_CAP_GEN(dev, roce)) 483 return 0; 484 485 memset(&he_in, 0, sizeof(he_in)); 486 he_in.he = MLX5_SET_HOST_ENDIANNESS; 487 err = mlx5_core_access_reg(dev, &he_in, sizeof(he_in), 488 &he_out, sizeof(he_out), 489 MLX5_REG_HOST_ENDIANNESS, 0, 1); 490 return err; 491 } 492 493 static int mlx5_core_enable_hca(struct mlx5_core_dev *dev) 494 { 495 u32 out[MLX5_ST_SZ_DW(enable_hca_out)] = {0}; 496 u32 in[MLX5_ST_SZ_DW(enable_hca_in)] = {0}; 497 498 MLX5_SET(enable_hca_in, in, opcode, MLX5_CMD_OP_ENABLE_HCA); 499 return mlx5_cmd_exec(dev, &in, sizeof(in), &out, sizeof(out)); 500 } 501 502 static int mlx5_core_disable_hca(struct mlx5_core_dev *dev) 503 { 504 u32 out[MLX5_ST_SZ_DW(disable_hca_out)] = {0}; 505 u32 in[MLX5_ST_SZ_DW(disable_hca_in)] = {0}; 506 507 MLX5_SET(disable_hca_in, in, opcode, MLX5_CMD_OP_DISABLE_HCA); 508 return mlx5_cmd_exec(dev, in, sizeof(in), out, sizeof(out)); 509 } 510 511 static int mlx5_core_set_issi(struct mlx5_core_dev *dev) 512 { 513 u32 query_in[MLX5_ST_SZ_DW(query_issi_in)] = {0}; 514 u32 query_out[MLX5_ST_SZ_DW(query_issi_out)] = {0}; 515 u32 sup_issi; 516 int err; 517 518 MLX5_SET(query_issi_in, query_in, opcode, MLX5_CMD_OP_QUERY_ISSI); 519 520 err = mlx5_cmd_exec(dev, query_in, sizeof(query_in), query_out, sizeof(query_out)); 521 if (err) { 522 u32 syndrome; 523 u8 status; 524 525 mlx5_cmd_mbox_status(query_out, &status, &syndrome); 526 if (status == MLX5_CMD_STAT_BAD_OP_ERR) { 527 pr_debug("Only ISSI 0 is supported\n"); 528 return 0; 529 } 530 531 printf("mlx5_core: ERR: ""failed to query ISSI\n"); 532 return err; 533 } 534 535 sup_issi = MLX5_GET(query_issi_out, query_out, supported_issi_dw0); 536 537 if (sup_issi & (1 << 1)) { 538 u32 set_in[MLX5_ST_SZ_DW(set_issi_in)] = {0}; 539 u32 set_out[MLX5_ST_SZ_DW(set_issi_out)] = {0}; 540 541 MLX5_SET(set_issi_in, set_in, opcode, MLX5_CMD_OP_SET_ISSI); 542 MLX5_SET(set_issi_in, set_in, current_issi, 1); 543 544 err = mlx5_cmd_exec(dev, set_in, sizeof(set_in), set_out, sizeof(set_out)); 545 if (err) { 546 printf("mlx5_core: ERR: ""failed to set ISSI=1 err(%d)\n", err); 547 return err; 548 } 549 550 dev->issi = 1; 551 552 return 0; 553 } else if (sup_issi & (1 << 0)) { 554 return 0; 555 } 556 557 return -ENOTSUPP; 558 } 559 560 561 int mlx5_vector2eqn(struct mlx5_core_dev *dev, int vector, int *eqn, int *irqn) 562 { 563 struct mlx5_eq_table *table = &dev->priv.eq_table; 564 struct mlx5_eq *eq; 565 int err = -ENOENT; 566 567 spin_lock(&table->lock); 568 list_for_each_entry(eq, &table->comp_eqs_list, list) { 569 if (eq->index == vector) { 570 *eqn = eq->eqn; 571 *irqn = eq->irqn; 572 err = 0; 573 break; 574 } 575 } 576 spin_unlock(&table->lock); 577 578 return err; 579 } 580 EXPORT_SYMBOL(mlx5_vector2eqn); 581 582 int mlx5_rename_eq(struct mlx5_core_dev *dev, int eq_ix, char *name) 583 { 584 struct mlx5_priv *priv = &dev->priv; 585 struct mlx5_eq_table *table = &priv->eq_table; 586 struct mlx5_eq *eq; 587 int err = -ENOENT; 588 589 spin_lock(&table->lock); 590 list_for_each_entry(eq, &table->comp_eqs_list, list) { 591 if (eq->index == eq_ix) { 592 int irq_ix = eq_ix + MLX5_EQ_VEC_COMP_BASE; 593 594 snprintf(priv->irq_info[irq_ix].name, MLX5_MAX_IRQ_NAME, 595 "%s-%d", name, eq_ix); 596 597 err = 0; 598 break; 599 } 600 } 601 spin_unlock(&table->lock); 602 603 return err; 604 } 605 606 static void free_comp_eqs(struct mlx5_core_dev *dev) 607 { 608 struct mlx5_eq_table *table = &dev->priv.eq_table; 609 struct mlx5_eq *eq, *n; 610 611 spin_lock(&table->lock); 612 list_for_each_entry_safe(eq, n, &table->comp_eqs_list, list) { 613 list_del(&eq->list); 614 spin_unlock(&table->lock); 615 if (mlx5_destroy_unmap_eq(dev, eq)) 616 mlx5_core_warn(dev, "failed to destroy EQ 0x%x\n", 617 eq->eqn); 618 kfree(eq); 619 spin_lock(&table->lock); 620 } 621 spin_unlock(&table->lock); 622 } 623 624 static int alloc_comp_eqs(struct mlx5_core_dev *dev) 625 { 626 struct mlx5_eq_table *table = &dev->priv.eq_table; 627 char name[MLX5_MAX_IRQ_NAME]; 628 struct mlx5_eq *eq; 629 int ncomp_vec; 630 int nent; 631 int err; 632 int i; 633 634 INIT_LIST_HEAD(&table->comp_eqs_list); 635 ncomp_vec = table->num_comp_vectors; 636 nent = MLX5_COMP_EQ_SIZE; 637 for (i = 0; i < ncomp_vec; i++) { 638 eq = kzalloc(sizeof(*eq), GFP_KERNEL); 639 640 snprintf(name, MLX5_MAX_IRQ_NAME, "mlx5_comp%d", i); 641 err = mlx5_create_map_eq(dev, eq, 642 i + MLX5_EQ_VEC_COMP_BASE, nent, 0, 643 name, &dev->priv.uuari.uars[0]); 644 if (err) { 645 kfree(eq); 646 goto clean; 647 } 648 mlx5_core_dbg(dev, "allocated completion EQN %d\n", eq->eqn); 649 eq->index = i; 650 spin_lock(&table->lock); 651 list_add_tail(&eq->list, &table->comp_eqs_list); 652 spin_unlock(&table->lock); 653 } 654 655 return 0; 656 657 clean: 658 free_comp_eqs(dev); 659 return err; 660 } 661 662 static int map_bf_area(struct mlx5_core_dev *dev) 663 { 664 resource_size_t bf_start = pci_resource_start(dev->pdev, 0); 665 resource_size_t bf_len = pci_resource_len(dev->pdev, 0); 666 667 dev->priv.bf_mapping = io_mapping_create_wc(bf_start, bf_len); 668 669 return dev->priv.bf_mapping ? 0 : -ENOMEM; 670 } 671 672 static void unmap_bf_area(struct mlx5_core_dev *dev) 673 { 674 if (dev->priv.bf_mapping) 675 io_mapping_free(dev->priv.bf_mapping); 676 } 677 678 static inline int fw_initializing(struct mlx5_core_dev *dev) 679 { 680 return ioread32be(&dev->iseg->initializing) >> 31; 681 } 682 683 static int wait_fw_init(struct mlx5_core_dev *dev, u32 max_wait_mili) 684 { 685 u64 end = jiffies + msecs_to_jiffies(max_wait_mili); 686 int err = 0; 687 688 while (fw_initializing(dev)) { 689 if (time_after(jiffies, end)) { 690 err = -EBUSY; 691 break; 692 } 693 msleep(FW_INIT_WAIT_MS); 694 } 695 696 return err; 697 } 698 699 static void mlx5_add_device(struct mlx5_interface *intf, struct mlx5_priv *priv) 700 { 701 struct mlx5_device_context *dev_ctx; 702 struct mlx5_core_dev *dev = container_of(priv, struct mlx5_core_dev, priv); 703 704 dev_ctx = kzalloc(sizeof(*dev_ctx), GFP_KERNEL); 705 if (!dev_ctx) 706 return; 707 708 dev_ctx->intf = intf; 709 CURVNET_SET_QUIET(vnet0); 710 dev_ctx->context = intf->add(dev); 711 CURVNET_RESTORE(); 712 713 if (dev_ctx->context) { 714 spin_lock_irq(&priv->ctx_lock); 715 list_add_tail(&dev_ctx->list, &priv->ctx_list); 716 spin_unlock_irq(&priv->ctx_lock); 717 } else { 718 kfree(dev_ctx); 719 } 720 } 721 722 static void mlx5_remove_device(struct mlx5_interface *intf, struct mlx5_priv *priv) 723 { 724 struct mlx5_device_context *dev_ctx; 725 struct mlx5_core_dev *dev = container_of(priv, struct mlx5_core_dev, priv); 726 727 list_for_each_entry(dev_ctx, &priv->ctx_list, list) 728 if (dev_ctx->intf == intf) { 729 spin_lock_irq(&priv->ctx_lock); 730 list_del(&dev_ctx->list); 731 spin_unlock_irq(&priv->ctx_lock); 732 733 intf->remove(dev, dev_ctx->context); 734 kfree(dev_ctx); 735 return; 736 } 737 } 738 739 int 740 mlx5_register_device(struct mlx5_core_dev *dev) 741 { 742 struct mlx5_priv *priv = &dev->priv; 743 struct mlx5_interface *intf; 744 745 mutex_lock(&intf_mutex); 746 list_add_tail(&priv->dev_list, &dev_list); 747 list_for_each_entry(intf, &intf_list, list) 748 mlx5_add_device(intf, priv); 749 mutex_unlock(&intf_mutex); 750 751 return 0; 752 } 753 754 void 755 mlx5_unregister_device(struct mlx5_core_dev *dev) 756 { 757 struct mlx5_priv *priv = &dev->priv; 758 struct mlx5_interface *intf; 759 760 mutex_lock(&intf_mutex); 761 list_for_each_entry(intf, &intf_list, list) 762 mlx5_remove_device(intf, priv); 763 list_del(&priv->dev_list); 764 mutex_unlock(&intf_mutex); 765 } 766 767 int mlx5_register_interface(struct mlx5_interface *intf) 768 { 769 struct mlx5_priv *priv; 770 771 if (!intf->add || !intf->remove) 772 return -EINVAL; 773 774 mutex_lock(&intf_mutex); 775 list_add_tail(&intf->list, &intf_list); 776 list_for_each_entry(priv, &dev_list, dev_list) 777 mlx5_add_device(intf, priv); 778 mutex_unlock(&intf_mutex); 779 780 return 0; 781 } 782 EXPORT_SYMBOL(mlx5_register_interface); 783 784 void mlx5_unregister_interface(struct mlx5_interface *intf) 785 { 786 struct mlx5_priv *priv; 787 788 mutex_lock(&intf_mutex); 789 list_for_each_entry(priv, &dev_list, dev_list) 790 mlx5_remove_device(intf, priv); 791 list_del(&intf->list); 792 mutex_unlock(&intf_mutex); 793 } 794 EXPORT_SYMBOL(mlx5_unregister_interface); 795 796 void *mlx5_get_protocol_dev(struct mlx5_core_dev *mdev, int protocol) 797 { 798 struct mlx5_priv *priv = &mdev->priv; 799 struct mlx5_device_context *dev_ctx; 800 unsigned long flags; 801 void *result = NULL; 802 803 spin_lock_irqsave(&priv->ctx_lock, flags); 804 805 list_for_each_entry(dev_ctx, &mdev->priv.ctx_list, list) 806 if ((dev_ctx->intf->protocol == protocol) && 807 dev_ctx->intf->get_dev) { 808 result = dev_ctx->intf->get_dev(dev_ctx->context); 809 break; 810 } 811 812 spin_unlock_irqrestore(&priv->ctx_lock, flags); 813 814 return result; 815 } 816 EXPORT_SYMBOL(mlx5_get_protocol_dev); 817 818 static int mlx5_pci_init(struct mlx5_core_dev *dev, struct mlx5_priv *priv) 819 { 820 struct pci_dev *pdev = dev->pdev; 821 int err = 0; 822 823 pci_set_drvdata(dev->pdev, dev); 824 strncpy(priv->name, dev_name(&pdev->dev), MLX5_MAX_NAME_LEN); 825 priv->name[MLX5_MAX_NAME_LEN - 1] = 0; 826 827 mutex_init(&priv->pgdir_mutex); 828 INIT_LIST_HEAD(&priv->pgdir_list); 829 spin_lock_init(&priv->mkey_lock); 830 831 priv->numa_node = NUMA_NO_NODE; 832 833 err = mlx5_pci_enable_device(dev); 834 if (err) { 835 device_printf((&pdev->dev)->bsddev, "ERR: ""Cannot enable PCI device, aborting\n"); 836 goto err_dbg; 837 } 838 839 err = request_bar(pdev); 840 if (err) { 841 device_printf((&pdev->dev)->bsddev, "ERR: ""error requesting BARs, aborting\n"); 842 goto err_disable; 843 } 844 845 pci_set_master(pdev); 846 847 err = set_dma_caps(pdev); 848 if (err) { 849 device_printf((&pdev->dev)->bsddev, "ERR: ""Failed setting DMA capabilities mask, aborting\n"); 850 goto err_clr_master; 851 } 852 853 dev->iseg_base = pci_resource_start(dev->pdev, 0); 854 dev->iseg = ioremap(dev->iseg_base, sizeof(*dev->iseg)); 855 if (!dev->iseg) { 856 err = -ENOMEM; 857 device_printf((&pdev->dev)->bsddev, "ERR: ""Failed mapping initialization segment, aborting\n"); 858 goto err_clr_master; 859 } 860 861 return 0; 862 863 err_clr_master: 864 pci_clear_master(dev->pdev); 865 release_bar(dev->pdev); 866 err_disable: 867 mlx5_pci_disable_device(dev); 868 err_dbg: 869 return err; 870 } 871 872 static void mlx5_pci_close(struct mlx5_core_dev *dev, struct mlx5_priv *priv) 873 { 874 iounmap(dev->iseg); 875 pci_clear_master(dev->pdev); 876 release_bar(dev->pdev); 877 mlx5_pci_disable_device(dev); 878 } 879 880 static int mlx5_init_once(struct mlx5_core_dev *dev, struct mlx5_priv *priv) 881 { 882 struct pci_dev *pdev = dev->pdev; 883 int err; 884 885 err = mlx5_vsc_find_cap(dev); 886 if (err) 887 dev_err(&pdev->dev, "Unable to find vendor specific capabilities\n"); 888 889 err = mlx5_query_hca_caps(dev); 890 if (err) { 891 dev_err(&pdev->dev, "query hca failed\n"); 892 goto out; 893 } 894 895 err = mlx5_query_board_id(dev); 896 if (err) { 897 dev_err(&pdev->dev, "query board id failed\n"); 898 goto out; 899 } 900 901 err = mlx5_eq_init(dev); 902 if (err) { 903 dev_err(&pdev->dev, "failed to initialize eq\n"); 904 goto out; 905 } 906 907 MLX5_INIT_DOORBELL_LOCK(&priv->cq_uar_lock); 908 909 err = mlx5_init_cq_table(dev); 910 if (err) { 911 dev_err(&pdev->dev, "failed to initialize cq table\n"); 912 goto err_eq_cleanup; 913 } 914 915 mlx5_init_qp_table(dev); 916 mlx5_init_srq_table(dev); 917 mlx5_init_mr_table(dev); 918 919 mlx5_init_reserved_gids(dev); 920 mlx5_fpga_init(dev); 921 922 #ifdef RATELIMIT 923 err = mlx5_init_rl_table(dev); 924 if (err) { 925 dev_err(&pdev->dev, "Failed to init rate limiting\n"); 926 goto err_tables_cleanup; 927 } 928 #endif 929 return 0; 930 931 #ifdef RATELIMIT 932 err_tables_cleanup: 933 mlx5_cleanup_mr_table(dev); 934 mlx5_cleanup_srq_table(dev); 935 mlx5_cleanup_qp_table(dev); 936 mlx5_cleanup_cq_table(dev); 937 #endif 938 939 err_eq_cleanup: 940 mlx5_eq_cleanup(dev); 941 942 out: 943 return err; 944 } 945 946 static void mlx5_cleanup_once(struct mlx5_core_dev *dev) 947 { 948 #ifdef RATELIMIT 949 mlx5_cleanup_rl_table(dev); 950 #endif 951 mlx5_fpga_cleanup(dev); 952 mlx5_cleanup_reserved_gids(dev); 953 mlx5_cleanup_mr_table(dev); 954 mlx5_cleanup_srq_table(dev); 955 mlx5_cleanup_qp_table(dev); 956 mlx5_cleanup_cq_table(dev); 957 mlx5_eq_cleanup(dev); 958 } 959 960 static int mlx5_load_one(struct mlx5_core_dev *dev, struct mlx5_priv *priv, 961 bool boot) 962 { 963 struct pci_dev *pdev = dev->pdev; 964 int err; 965 966 mutex_lock(&dev->intf_state_mutex); 967 if (test_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state)) { 968 dev_warn(&dev->pdev->dev, "%s: interface is up, NOP\n", 969 __func__); 970 goto out; 971 } 972 973 device_printf((&pdev->dev)->bsddev, "INFO: ""firmware version: %d.%d.%d\n", fw_rev_maj(dev), fw_rev_min(dev), fw_rev_sub(dev)); 974 975 /* 976 * On load removing any previous indication of internal error, 977 * device is up 978 */ 979 dev->state = MLX5_DEVICE_STATE_UP; 980 981 err = mlx5_cmd_init(dev); 982 if (err) { 983 device_printf((&pdev->dev)->bsddev, "ERR: ""Failed initializing command interface, aborting\n"); 984 goto out_err; 985 } 986 987 err = wait_fw_init(dev, FW_INIT_TIMEOUT_MILI); 988 if (err) { 989 device_printf((&dev->pdev->dev)->bsddev, "ERR: ""Firmware over %d MS in initializing state, aborting\n", FW_INIT_TIMEOUT_MILI); 990 goto err_cmd_cleanup; 991 } 992 993 err = mlx5_core_enable_hca(dev); 994 if (err) { 995 device_printf((&pdev->dev)->bsddev, "ERR: ""enable hca failed\n"); 996 goto err_cmd_cleanup; 997 } 998 999 err = mlx5_core_set_issi(dev); 1000 if (err) { 1001 device_printf((&pdev->dev)->bsddev, "ERR: ""failed to set issi\n"); 1002 goto err_disable_hca; 1003 } 1004 1005 err = mlx5_pagealloc_start(dev); 1006 if (err) { 1007 device_printf((&pdev->dev)->bsddev, "ERR: ""mlx5_pagealloc_start failed\n"); 1008 goto err_disable_hca; 1009 } 1010 1011 err = mlx5_satisfy_startup_pages(dev, 1); 1012 if (err) { 1013 device_printf((&pdev->dev)->bsddev, "ERR: ""failed to allocate boot pages\n"); 1014 goto err_pagealloc_stop; 1015 } 1016 1017 err = set_hca_ctrl(dev); 1018 if (err) { 1019 device_printf((&pdev->dev)->bsddev, "ERR: ""set_hca_ctrl failed\n"); 1020 goto reclaim_boot_pages; 1021 } 1022 1023 err = handle_hca_cap(dev); 1024 if (err) { 1025 device_printf((&pdev->dev)->bsddev, "ERR: ""handle_hca_cap failed\n"); 1026 goto reclaim_boot_pages; 1027 } 1028 1029 err = handle_hca_cap_atomic(dev); 1030 if (err) { 1031 device_printf((&pdev->dev)->bsddev, "ERR: ""handle_hca_cap_atomic failed\n"); 1032 goto reclaim_boot_pages; 1033 } 1034 1035 err = mlx5_satisfy_startup_pages(dev, 0); 1036 if (err) { 1037 device_printf((&pdev->dev)->bsddev, "ERR: ""failed to allocate init pages\n"); 1038 goto reclaim_boot_pages; 1039 } 1040 1041 err = mlx5_cmd_init_hca(dev); 1042 if (err) { 1043 device_printf((&pdev->dev)->bsddev, "ERR: ""init hca failed\n"); 1044 goto reclaim_boot_pages; 1045 } 1046 1047 mlx5_start_health_poll(dev); 1048 1049 if (boot && mlx5_init_once(dev, priv)) { 1050 dev_err(&pdev->dev, "sw objs init failed\n"); 1051 goto err_stop_poll; 1052 } 1053 1054 err = mlx5_enable_msix(dev); 1055 if (err) { 1056 device_printf((&pdev->dev)->bsddev, "ERR: ""enable msix failed\n"); 1057 goto err_cleanup_once; 1058 } 1059 1060 err = mlx5_alloc_uuars(dev, &priv->uuari); 1061 if (err) { 1062 device_printf((&pdev->dev)->bsddev, "ERR: ""Failed allocating uar, aborting\n"); 1063 goto err_disable_msix; 1064 } 1065 1066 err = mlx5_start_eqs(dev); 1067 if (err) { 1068 device_printf((&pdev->dev)->bsddev, "ERR: ""Failed to start pages and async EQs\n"); 1069 goto err_free_uar; 1070 } 1071 1072 err = alloc_comp_eqs(dev); 1073 if (err) { 1074 device_printf((&pdev->dev)->bsddev, "ERR: ""Failed to alloc completion EQs\n"); 1075 goto err_stop_eqs; 1076 } 1077 1078 if (map_bf_area(dev)) 1079 device_printf((&pdev->dev)->bsddev, "ERR: ""Failed to map blue flame area\n"); 1080 1081 err = mlx5_init_fs(dev); 1082 if (err) { 1083 mlx5_core_err(dev, "flow steering init %d\n", err); 1084 goto err_free_comp_eqs; 1085 } 1086 1087 err = mlx5_fpga_device_start(dev); 1088 if (err) { 1089 dev_err(&pdev->dev, "fpga device start failed %d\n", err); 1090 goto err_fpga_start; 1091 } 1092 1093 err = mlx5_register_device(dev); 1094 if (err) { 1095 dev_err(&pdev->dev, "mlx5_register_device failed %d\n", err); 1096 goto err_fs; 1097 } 1098 1099 clear_bit(MLX5_INTERFACE_STATE_DOWN, &dev->intf_state); 1100 set_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state); 1101 1102 out: 1103 mutex_unlock(&dev->intf_state_mutex); 1104 return 0; 1105 1106 err_fpga_start: 1107 err_fs: 1108 mlx5_cleanup_fs(dev); 1109 1110 err_free_comp_eqs: 1111 free_comp_eqs(dev); 1112 unmap_bf_area(dev); 1113 1114 err_stop_eqs: 1115 mlx5_stop_eqs(dev); 1116 1117 err_free_uar: 1118 mlx5_free_uuars(dev, &priv->uuari); 1119 1120 err_disable_msix: 1121 mlx5_disable_msix(dev); 1122 1123 err_cleanup_once: 1124 if (boot) 1125 mlx5_cleanup_once(dev); 1126 1127 err_stop_poll: 1128 mlx5_stop_health_poll(dev, boot); 1129 if (mlx5_cmd_teardown_hca(dev)) { 1130 device_printf((&dev->pdev->dev)->bsddev, "ERR: ""tear_down_hca failed, skip cleanup\n"); 1131 goto out_err; 1132 } 1133 1134 reclaim_boot_pages: 1135 mlx5_reclaim_startup_pages(dev); 1136 1137 err_pagealloc_stop: 1138 mlx5_pagealloc_stop(dev); 1139 1140 err_disable_hca: 1141 mlx5_core_disable_hca(dev); 1142 1143 err_cmd_cleanup: 1144 mlx5_cmd_cleanup(dev); 1145 1146 out_err: 1147 dev->state = MLX5_DEVICE_STATE_INTERNAL_ERROR; 1148 mutex_unlock(&dev->intf_state_mutex); 1149 1150 return err; 1151 } 1152 1153 static int mlx5_unload_one(struct mlx5_core_dev *dev, struct mlx5_priv *priv, 1154 bool cleanup) 1155 { 1156 int err = 0; 1157 1158 if (cleanup) 1159 mlx5_drain_health_recovery(dev); 1160 1161 mutex_lock(&dev->intf_state_mutex); 1162 if (test_bit(MLX5_INTERFACE_STATE_DOWN, &dev->intf_state)) { 1163 dev_warn(&dev->pdev->dev, "%s: interface is down, NOP\n", __func__); 1164 if (cleanup) 1165 mlx5_cleanup_once(dev); 1166 goto out; 1167 } 1168 1169 mlx5_unregister_device(dev); 1170 1171 mlx5_fpga_device_stop(dev); 1172 mlx5_cleanup_fs(dev); 1173 unmap_bf_area(dev); 1174 mlx5_wait_for_reclaim_vfs_pages(dev); 1175 free_comp_eqs(dev); 1176 mlx5_stop_eqs(dev); 1177 mlx5_free_uuars(dev, &priv->uuari); 1178 mlx5_disable_msix(dev); 1179 if (cleanup) 1180 mlx5_cleanup_once(dev); 1181 mlx5_stop_health_poll(dev, cleanup); 1182 err = mlx5_cmd_teardown_hca(dev); 1183 if (err) { 1184 device_printf((&dev->pdev->dev)->bsddev, "ERR: ""tear_down_hca failed, skip cleanup\n"); 1185 goto out; 1186 } 1187 mlx5_pagealloc_stop(dev); 1188 mlx5_reclaim_startup_pages(dev); 1189 mlx5_core_disable_hca(dev); 1190 mlx5_cmd_cleanup(dev); 1191 1192 out: 1193 clear_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state); 1194 set_bit(MLX5_INTERFACE_STATE_DOWN, &dev->intf_state); 1195 mutex_unlock(&dev->intf_state_mutex); 1196 return err; 1197 } 1198 1199 void mlx5_core_event(struct mlx5_core_dev *dev, enum mlx5_dev_event event, 1200 unsigned long param) 1201 { 1202 struct mlx5_priv *priv = &dev->priv; 1203 struct mlx5_device_context *dev_ctx; 1204 unsigned long flags; 1205 1206 spin_lock_irqsave(&priv->ctx_lock, flags); 1207 1208 list_for_each_entry(dev_ctx, &priv->ctx_list, list) 1209 if (dev_ctx->intf->event) 1210 dev_ctx->intf->event(dev, dev_ctx->context, event, param); 1211 1212 spin_unlock_irqrestore(&priv->ctx_lock, flags); 1213 } 1214 1215 struct mlx5_core_event_handler { 1216 void (*event)(struct mlx5_core_dev *dev, 1217 enum mlx5_dev_event event, 1218 void *data); 1219 }; 1220 1221 static int init_one(struct pci_dev *pdev, 1222 const struct pci_device_id *id) 1223 { 1224 struct mlx5_core_dev *dev; 1225 struct mlx5_priv *priv; 1226 device_t bsddev = pdev->dev.bsddev; 1227 int err; 1228 1229 dev = kzalloc(sizeof(*dev), GFP_KERNEL); 1230 priv = &dev->priv; 1231 if (id) 1232 priv->pci_dev_data = id->driver_data; 1233 1234 if (prof_sel < 0 || prof_sel >= ARRAY_SIZE(profiles)) { 1235 device_printf(bsddev, "WARN: selected profile out of range, selecting default (%d)\n", MLX5_DEFAULT_PROF); 1236 prof_sel = MLX5_DEFAULT_PROF; 1237 } 1238 dev->profile = &profiles[prof_sel]; 1239 dev->pdev = pdev; 1240 dev->event = mlx5_core_event; 1241 1242 /* Set desc */ 1243 device_set_desc(bsddev, mlx5_version); 1244 1245 sysctl_ctx_init(&dev->sysctl_ctx); 1246 SYSCTL_ADD_INT(&dev->sysctl_ctx, 1247 SYSCTL_CHILDREN(device_get_sysctl_tree(bsddev)), 1248 OID_AUTO, "msix_eqvec", CTLFLAG_RDTUN, &dev->msix_eqvec, 0, 1249 "Maximum number of MSIX event queue vectors, if set"); 1250 1251 INIT_LIST_HEAD(&priv->ctx_list); 1252 spin_lock_init(&priv->ctx_lock); 1253 mutex_init(&dev->pci_status_mutex); 1254 mutex_init(&dev->intf_state_mutex); 1255 err = mlx5_pci_init(dev, priv); 1256 if (err) { 1257 device_printf(bsddev, "ERR: mlx5_pci_init failed %d\n", err); 1258 goto clean_dev; 1259 } 1260 1261 err = mlx5_health_init(dev); 1262 if (err) { 1263 device_printf(bsddev, "ERR: mlx5_health_init failed %d\n", err); 1264 goto close_pci; 1265 } 1266 1267 mlx5_pagealloc_init(dev); 1268 1269 err = mlx5_load_one(dev, priv, true); 1270 if (err) { 1271 device_printf(bsddev, "ERR: mlx5_load_one failed %d\n", err); 1272 goto clean_health; 1273 } 1274 1275 mlx5_fwdump_prep(dev); 1276 1277 pci_save_state(bsddev); 1278 return 0; 1279 1280 clean_health: 1281 mlx5_pagealloc_cleanup(dev); 1282 mlx5_health_cleanup(dev); 1283 close_pci: 1284 mlx5_pci_close(dev, priv); 1285 clean_dev: 1286 sysctl_ctx_free(&dev->sysctl_ctx); 1287 kfree(dev); 1288 return err; 1289 } 1290 1291 static void remove_one(struct pci_dev *pdev) 1292 { 1293 struct mlx5_core_dev *dev = pci_get_drvdata(pdev); 1294 struct mlx5_priv *priv = &dev->priv; 1295 1296 if (mlx5_unload_one(dev, priv, true)) { 1297 dev_err(&dev->pdev->dev, "mlx5_unload_one failed\n"); 1298 mlx5_health_cleanup(dev); 1299 return; 1300 } 1301 1302 mlx5_fwdump_clean(dev); 1303 mlx5_pagealloc_cleanup(dev); 1304 mlx5_health_cleanup(dev); 1305 mlx5_pci_close(dev, priv); 1306 pci_set_drvdata(pdev, NULL); 1307 sysctl_ctx_free(&dev->sysctl_ctx); 1308 kfree(dev); 1309 } 1310 1311 static pci_ers_result_t mlx5_pci_err_detected(struct pci_dev *pdev, 1312 pci_channel_state_t state) 1313 { 1314 struct mlx5_core_dev *dev = pci_get_drvdata(pdev); 1315 struct mlx5_priv *priv = &dev->priv; 1316 1317 dev_info(&pdev->dev, "%s was called\n", __func__); 1318 mlx5_enter_error_state(dev, false); 1319 mlx5_unload_one(dev, priv, false); 1320 1321 if (state) { 1322 mlx5_drain_health_wq(dev); 1323 mlx5_pci_disable_device(dev); 1324 } 1325 1326 return state == pci_channel_io_perm_failure ? 1327 PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_NEED_RESET; 1328 } 1329 1330 static pci_ers_result_t mlx5_pci_slot_reset(struct pci_dev *pdev) 1331 { 1332 struct mlx5_core_dev *dev = pci_get_drvdata(pdev); 1333 int err = 0; 1334 1335 dev_info(&pdev->dev, "%s was called\n", __func__); 1336 1337 err = mlx5_pci_enable_device(dev); 1338 if (err) { 1339 dev_err(&pdev->dev, "%s: mlx5_pci_enable_device failed with error code: %d\n" 1340 , __func__, err); 1341 return PCI_ERS_RESULT_DISCONNECT; 1342 } 1343 pci_set_master(pdev); 1344 pci_set_powerstate(pdev->dev.bsddev, PCI_POWERSTATE_D0); 1345 pci_restore_state(pdev->dev.bsddev); 1346 pci_save_state(pdev->dev.bsddev); 1347 1348 return err ? PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_RECOVERED; 1349 } 1350 1351 /* wait for the device to show vital signs. For now we check 1352 * that we can read the device ID and that the health buffer 1353 * shows a non zero value which is different than 0xffffffff 1354 */ 1355 static void wait_vital(struct pci_dev *pdev) 1356 { 1357 struct mlx5_core_dev *dev = pci_get_drvdata(pdev); 1358 struct mlx5_core_health *health = &dev->priv.health; 1359 const int niter = 100; 1360 u32 count; 1361 u16 did; 1362 int i; 1363 1364 /* Wait for firmware to be ready after reset */ 1365 msleep(1000); 1366 for (i = 0; i < niter; i++) { 1367 if (pci_read_config_word(pdev, 2, &did)) { 1368 dev_warn(&pdev->dev, "failed reading config word\n"); 1369 break; 1370 } 1371 if (did == pdev->device) { 1372 dev_info(&pdev->dev, "device ID correctly read after %d iterations\n", i); 1373 break; 1374 } 1375 msleep(50); 1376 } 1377 if (i == niter) 1378 dev_warn(&pdev->dev, "%s-%d: could not read device ID\n", __func__, __LINE__); 1379 1380 for (i = 0; i < niter; i++) { 1381 count = ioread32be(health->health_counter); 1382 if (count && count != 0xffffffff) { 1383 dev_info(&pdev->dev, "Counter value 0x%x after %d iterations\n", count, i); 1384 break; 1385 } 1386 msleep(50); 1387 } 1388 1389 if (i == niter) 1390 dev_warn(&pdev->dev, "%s-%d: could not read device ID\n", __func__, __LINE__); 1391 } 1392 1393 static void mlx5_pci_resume(struct pci_dev *pdev) 1394 { 1395 struct mlx5_core_dev *dev = pci_get_drvdata(pdev); 1396 struct mlx5_priv *priv = &dev->priv; 1397 int err; 1398 1399 dev_info(&pdev->dev, "%s was called\n", __func__); 1400 1401 wait_vital(pdev); 1402 1403 err = mlx5_load_one(dev, priv, false); 1404 if (err) 1405 dev_err(&pdev->dev, "%s: mlx5_load_one failed with error code: %d\n" 1406 , __func__, err); 1407 else 1408 dev_info(&pdev->dev, "%s: device recovered\n", __func__); 1409 } 1410 1411 static const struct pci_error_handlers mlx5_err_handler = { 1412 .error_detected = mlx5_pci_err_detected, 1413 .slot_reset = mlx5_pci_slot_reset, 1414 .resume = mlx5_pci_resume 1415 }; 1416 1417 static int mlx5_try_fast_unload(struct mlx5_core_dev *dev) 1418 { 1419 int err; 1420 1421 if (!MLX5_CAP_GEN(dev, force_teardown)) { 1422 mlx5_core_dbg(dev, "force teardown is not supported in the firmware\n"); 1423 return -EOPNOTSUPP; 1424 } 1425 1426 if (dev->state == MLX5_DEVICE_STATE_INTERNAL_ERROR) { 1427 mlx5_core_dbg(dev, "Device in internal error state, giving up\n"); 1428 return -EAGAIN; 1429 } 1430 1431 /* Panic tear down fw command will stop the PCI bus communication 1432 * with the HCA, so the health polll is no longer needed. 1433 */ 1434 mlx5_drain_health_wq(dev); 1435 mlx5_stop_health_poll(dev, false); 1436 1437 err = mlx5_cmd_force_teardown_hca(dev); 1438 if (err) { 1439 mlx5_core_dbg(dev, "Firmware couldn't do fast unload error: %d\n", err); 1440 return err; 1441 } 1442 1443 mlx5_enter_error_state(dev, true); 1444 1445 return 0; 1446 } 1447 1448 static void shutdown_one(struct pci_dev *pdev) 1449 { 1450 struct mlx5_core_dev *dev = pci_get_drvdata(pdev); 1451 struct mlx5_priv *priv = &dev->priv; 1452 int err; 1453 1454 set_bit(MLX5_INTERFACE_STATE_SHUTDOWN, &dev->intf_state); 1455 err = mlx5_try_fast_unload(dev); 1456 if (err) 1457 mlx5_unload_one(dev, priv, false); 1458 mlx5_pci_disable_device(dev); 1459 } 1460 1461 static const struct pci_device_id mlx5_core_pci_table[] = { 1462 { PCI_VDEVICE(MELLANOX, 4113) }, /* Connect-IB */ 1463 { PCI_VDEVICE(MELLANOX, 4114) }, /* Connect-IB VF */ 1464 { PCI_VDEVICE(MELLANOX, 4115) }, /* ConnectX-4 */ 1465 { PCI_VDEVICE(MELLANOX, 4116) }, /* ConnectX-4 VF */ 1466 { PCI_VDEVICE(MELLANOX, 4117) }, /* ConnectX-4LX */ 1467 { PCI_VDEVICE(MELLANOX, 4118) }, /* ConnectX-4LX VF */ 1468 { PCI_VDEVICE(MELLANOX, 4119) }, /* ConnectX-5 */ 1469 { PCI_VDEVICE(MELLANOX, 4120) }, /* ConnectX-5 VF */ 1470 { PCI_VDEVICE(MELLANOX, 4121) }, 1471 { PCI_VDEVICE(MELLANOX, 4122) }, 1472 { PCI_VDEVICE(MELLANOX, 4123) }, 1473 { PCI_VDEVICE(MELLANOX, 4124) }, 1474 { PCI_VDEVICE(MELLANOX, 4125) }, 1475 { PCI_VDEVICE(MELLANOX, 4126) }, 1476 { PCI_VDEVICE(MELLANOX, 4127) }, 1477 { PCI_VDEVICE(MELLANOX, 4128) }, 1478 { PCI_VDEVICE(MELLANOX, 4129) }, 1479 { PCI_VDEVICE(MELLANOX, 4130) }, 1480 { PCI_VDEVICE(MELLANOX, 4131) }, 1481 { PCI_VDEVICE(MELLANOX, 4132) }, 1482 { PCI_VDEVICE(MELLANOX, 4133) }, 1483 { PCI_VDEVICE(MELLANOX, 4134) }, 1484 { PCI_VDEVICE(MELLANOX, 4135) }, 1485 { PCI_VDEVICE(MELLANOX, 4136) }, 1486 { PCI_VDEVICE(MELLANOX, 4137) }, 1487 { PCI_VDEVICE(MELLANOX, 4138) }, 1488 { PCI_VDEVICE(MELLANOX, 4139) }, 1489 { PCI_VDEVICE(MELLANOX, 4140) }, 1490 { PCI_VDEVICE(MELLANOX, 4141) }, 1491 { PCI_VDEVICE(MELLANOX, 4142) }, 1492 { PCI_VDEVICE(MELLANOX, 4143) }, 1493 { PCI_VDEVICE(MELLANOX, 4144) }, 1494 { 0, } 1495 }; 1496 1497 MODULE_DEVICE_TABLE(pci, mlx5_core_pci_table); 1498 1499 void mlx5_disable_device(struct mlx5_core_dev *dev) 1500 { 1501 mlx5_pci_err_detected(dev->pdev, 0); 1502 } 1503 1504 void mlx5_recover_device(struct mlx5_core_dev *dev) 1505 { 1506 mlx5_pci_disable_device(dev); 1507 if (mlx5_pci_slot_reset(dev->pdev) == PCI_ERS_RESULT_RECOVERED) 1508 mlx5_pci_resume(dev->pdev); 1509 } 1510 1511 struct pci_driver mlx5_core_driver = { 1512 .name = DRIVER_NAME, 1513 .id_table = mlx5_core_pci_table, 1514 .shutdown = shutdown_one, 1515 .probe = init_one, 1516 .remove = remove_one, 1517 .err_handler = &mlx5_err_handler 1518 }; 1519 1520 static int __init init(void) 1521 { 1522 int err; 1523 1524 err = pci_register_driver(&mlx5_core_driver); 1525 if (err) 1526 goto err_debug; 1527 1528 err = mlx5_fwdump_init(); 1529 if (err) 1530 goto err_fwdump; 1531 1532 return 0; 1533 1534 err_fwdump: 1535 pci_unregister_driver(&mlx5_core_driver); 1536 1537 err_debug: 1538 return err; 1539 } 1540 1541 static void __exit cleanup(void) 1542 { 1543 mlx5_fwdump_fini(); 1544 pci_unregister_driver(&mlx5_core_driver); 1545 } 1546 1547 module_init(init); 1548 module_exit(cleanup); 1549