2 * Copyright (c) 2013-2015, Mellanox Technologies. All rights reserved.
4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the
8 * OpenIB.org BSD license below:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
18 * - Redistributions in binary form must reproduce the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and/or other materials
21 * provided with the distribution.
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
33 #include <linux/highmem.h>
34 #include <linux/module.h>
35 #include <linux/init.h>
36 #include <linux/errno.h>
37 #include <linux/pci.h>
38 #include <linux/dma-mapping.h>
39 #include <linux/slab.h>
40 #include <linux/io-mapping.h>
41 #include <linux/interrupt.h>
42 #include <linux/delay.h>
43 #include <linux/mlx5/driver.h>
44 #include <linux/mlx5/cq.h>
45 #include <linux/mlx5/qp.h>
46 #include <linux/mlx5/srq.h>
47 #include <linux/debugfs.h>
48 #include <linux/kmod.h>
49 #include <linux/mlx5/mlx5_ifc.h>
50 #ifdef CONFIG_RFS_ACCEL
51 #include <linux/cpu_rmap.h>
53 #include <net/devlink.h>
54 #include "mlx5_core.h"
56 #ifdef CONFIG_MLX5_CORE_EN
60 MODULE_AUTHOR("Eli Cohen <eli@mellanox.com>");
61 MODULE_DESCRIPTION("Mellanox Connect-IB, ConnectX-4 core driver");
62 MODULE_LICENSE("Dual BSD/GPL");
63 MODULE_VERSION(DRIVER_VERSION);
65 unsigned int mlx5_core_debug_mask;
66 module_param_named(debug_mask, mlx5_core_debug_mask, uint, 0644);
67 MODULE_PARM_DESC(debug_mask, "debug mask: 1 = dump cmd data, 2 = dump cmd exec time, 3 = both. Default=0");
69 #define MLX5_DEFAULT_PROF 2
70 static unsigned int prof_sel = MLX5_DEFAULT_PROF;
71 module_param_named(prof_sel, prof_sel, uint, 0444);
72 MODULE_PARM_DESC(prof_sel, "profile selector. Valid range 0 - 2");
75 MLX5_ATOMIC_REQ_MODE_BE = 0x0,
76 MLX5_ATOMIC_REQ_MODE_HOST_ENDIANNESS = 0x1,
79 static struct mlx5_profile profile[] = {
84 .mask = MLX5_PROF_MASK_QP_SIZE,
88 .mask = MLX5_PROF_MASK_QP_SIZE |
89 MLX5_PROF_MASK_MR_CACHE,
158 #define FW_INIT_TIMEOUT_MILI 2000
159 #define FW_INIT_WAIT_MS 2
160 #define FW_PRE_INIT_TIMEOUT_MILI 10000
162 static int wait_fw_init(struct mlx5_core_dev *dev, u32 max_wait_mili)
164 unsigned long end = jiffies + msecs_to_jiffies(max_wait_mili);
167 while (fw_initializing(dev)) {
168 if (time_after(jiffies, end)) {
172 msleep(FW_INIT_WAIT_MS);
178 static int set_dma_caps(struct pci_dev *pdev)
182 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
184 dev_warn(&pdev->dev, "Warning: couldn't set 64-bit PCI DMA mask\n");
185 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
187 dev_err(&pdev->dev, "Can't set PCI DMA mask, aborting\n");
192 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
195 "Warning: couldn't set 64-bit consistent PCI DMA mask\n");
196 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
199 "Can't set consistent PCI DMA mask, aborting\n");
204 dma_set_max_seg_size(&pdev->dev, 2u * 1024 * 1024 * 1024);
208 static int mlx5_pci_enable_device(struct mlx5_core_dev *dev)
210 struct pci_dev *pdev = dev->pdev;
213 mutex_lock(&dev->pci_status_mutex);
214 if (dev->pci_status == MLX5_PCI_STATUS_DISABLED) {
215 err = pci_enable_device(pdev);
217 dev->pci_status = MLX5_PCI_STATUS_ENABLED;
219 mutex_unlock(&dev->pci_status_mutex);
224 static void mlx5_pci_disable_device(struct mlx5_core_dev *dev)
226 struct pci_dev *pdev = dev->pdev;
228 mutex_lock(&dev->pci_status_mutex);
229 if (dev->pci_status == MLX5_PCI_STATUS_ENABLED) {
230 pci_disable_device(pdev);
231 dev->pci_status = MLX5_PCI_STATUS_DISABLED;
233 mutex_unlock(&dev->pci_status_mutex);
236 static int request_bar(struct pci_dev *pdev)
240 if (!(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) {
241 dev_err(&pdev->dev, "Missing registers BAR, aborting\n");
245 err = pci_request_regions(pdev, DRIVER_NAME);
247 dev_err(&pdev->dev, "Couldn't get PCI resources, aborting\n");
252 static void release_bar(struct pci_dev *pdev)
254 pci_release_regions(pdev);
257 static int mlx5_enable_msix(struct mlx5_core_dev *dev)
259 struct mlx5_priv *priv = &dev->priv;
260 struct mlx5_eq_table *table = &priv->eq_table;
261 int num_eqs = 1 << MLX5_CAP_GEN(dev, log_max_eq);
265 nvec = MLX5_CAP_GEN(dev, num_ports) * num_online_cpus() +
266 MLX5_EQ_VEC_COMP_BASE;
267 nvec = min_t(int, nvec, num_eqs);
268 if (nvec <= MLX5_EQ_VEC_COMP_BASE)
271 priv->msix_arr = kcalloc(nvec, sizeof(*priv->msix_arr), GFP_KERNEL);
273 priv->irq_info = kcalloc(nvec, sizeof(*priv->irq_info), GFP_KERNEL);
274 if (!priv->msix_arr || !priv->irq_info)
277 for (i = 0; i < nvec; i++)
278 priv->msix_arr[i].entry = i;
280 nvec = pci_enable_msix_range(dev->pdev, priv->msix_arr,
281 MLX5_EQ_VEC_COMP_BASE + 1, nvec);
285 table->num_comp_vectors = nvec - MLX5_EQ_VEC_COMP_BASE;
290 kfree(priv->irq_info);
291 kfree(priv->msix_arr);
295 static void mlx5_disable_msix(struct mlx5_core_dev *dev)
297 struct mlx5_priv *priv = &dev->priv;
299 pci_disable_msix(dev->pdev);
300 kfree(priv->irq_info);
301 kfree(priv->msix_arr);
304 struct mlx5_reg_host_endianess {
310 #define CAP_MASK(pos, size) ((u64)((1 << (size)) - 1) << (pos))
313 MLX5_CAP_BITS_RW_MASK = CAP_MASK(MLX5_CAP_OFF_CMDIF_CSUM, 2) |
314 MLX5_DEV_CAP_FLAG_DCT,
317 static u16 to_fw_pkey_sz(struct mlx5_core_dev *dev, u32 size)
333 mlx5_core_warn(dev, "invalid pkey table size %d\n", size);
338 static int mlx5_core_get_caps_mode(struct mlx5_core_dev *dev,
339 enum mlx5_cap_type cap_type,
340 enum mlx5_cap_mode cap_mode)
342 u8 in[MLX5_ST_SZ_BYTES(query_hca_cap_in)];
343 int out_sz = MLX5_ST_SZ_BYTES(query_hca_cap_out);
344 void *out, *hca_caps;
345 u16 opmod = (cap_type << 1) | (cap_mode & 0x01);
348 memset(in, 0, sizeof(in));
349 out = kzalloc(out_sz, GFP_KERNEL);
353 MLX5_SET(query_hca_cap_in, in, opcode, MLX5_CMD_OP_QUERY_HCA_CAP);
354 MLX5_SET(query_hca_cap_in, in, op_mod, opmod);
355 err = mlx5_cmd_exec(dev, in, sizeof(in), out, out_sz);
358 "QUERY_HCA_CAP : type(%x) opmode(%x) Failed(%d)\n",
359 cap_type, cap_mode, err);
363 hca_caps = MLX5_ADDR_OF(query_hca_cap_out, out, capability);
366 case HCA_CAP_OPMOD_GET_MAX:
367 memcpy(dev->hca_caps_max[cap_type], hca_caps,
368 MLX5_UN_SZ_BYTES(hca_cap_union));
370 case HCA_CAP_OPMOD_GET_CUR:
371 memcpy(dev->hca_caps_cur[cap_type], hca_caps,
372 MLX5_UN_SZ_BYTES(hca_cap_union));
376 "Tried to query dev cap type(%x) with wrong opmode(%x)\n",
386 int mlx5_core_get_caps(struct mlx5_core_dev *dev, enum mlx5_cap_type cap_type)
390 ret = mlx5_core_get_caps_mode(dev, cap_type, HCA_CAP_OPMOD_GET_CUR);
393 return mlx5_core_get_caps_mode(dev, cap_type, HCA_CAP_OPMOD_GET_MAX);
396 static int set_caps(struct mlx5_core_dev *dev, void *in, int in_sz, int opmod)
398 u32 out[MLX5_ST_SZ_DW(set_hca_cap_out)] = {0};
400 MLX5_SET(set_hca_cap_in, in, opcode, MLX5_CMD_OP_SET_HCA_CAP);
401 MLX5_SET(set_hca_cap_in, in, op_mod, opmod << 1);
402 return mlx5_cmd_exec(dev, in, in_sz, out, sizeof(out));
405 static int handle_hca_cap_atomic(struct mlx5_core_dev *dev)
409 int set_sz = MLX5_ST_SZ_BYTES(set_hca_cap_in);
413 if (MLX5_CAP_GEN(dev, atomic)) {
414 err = mlx5_core_get_caps(dev, MLX5_CAP_ATOMIC);
423 supported_atomic_req_8B_endianess_mode_1);
425 if (req_endianness != MLX5_ATOMIC_REQ_MODE_HOST_ENDIANNESS)
428 set_ctx = kzalloc(set_sz, GFP_KERNEL);
432 set_hca_cap = MLX5_ADDR_OF(set_hca_cap_in, set_ctx, capability);
434 /* Set requestor to host endianness */
435 MLX5_SET(atomic_caps, set_hca_cap, atomic_req_8B_endianess_mode,
436 MLX5_ATOMIC_REQ_MODE_HOST_ENDIANNESS);
438 err = set_caps(dev, set_ctx, set_sz, MLX5_SET_HCA_CAP_OP_MOD_ATOMIC);
444 static int handle_hca_cap(struct mlx5_core_dev *dev)
446 void *set_ctx = NULL;
447 struct mlx5_profile *prof = dev->profile;
449 int set_sz = MLX5_ST_SZ_BYTES(set_hca_cap_in);
452 set_ctx = kzalloc(set_sz, GFP_KERNEL);
456 err = mlx5_core_get_caps(dev, MLX5_CAP_GENERAL);
460 set_hca_cap = MLX5_ADDR_OF(set_hca_cap_in, set_ctx,
462 memcpy(set_hca_cap, dev->hca_caps_cur[MLX5_CAP_GENERAL],
463 MLX5_ST_SZ_BYTES(cmd_hca_cap));
465 mlx5_core_dbg(dev, "Current Pkey table size %d Setting new size %d\n",
466 mlx5_to_sw_pkey_sz(MLX5_CAP_GEN(dev, pkey_table_size)),
468 /* we limit the size of the pkey table to 128 entries for now */
469 MLX5_SET(cmd_hca_cap, set_hca_cap, pkey_table_size,
470 to_fw_pkey_sz(dev, 128));
472 /* Check log_max_qp from HCA caps to set in current profile */
473 if (MLX5_CAP_GEN_MAX(dev, log_max_qp) < profile[prof_sel].log_max_qp) {
474 mlx5_core_warn(dev, "log_max_qp value in current profile is %d, changing it to HCA capability limit (%d)\n",
475 profile[prof_sel].log_max_qp,
476 MLX5_CAP_GEN_MAX(dev, log_max_qp));
477 profile[prof_sel].log_max_qp = MLX5_CAP_GEN_MAX(dev, log_max_qp);
479 if (prof->mask & MLX5_PROF_MASK_QP_SIZE)
480 MLX5_SET(cmd_hca_cap, set_hca_cap, log_max_qp,
483 /* disable cmdif checksum */
484 MLX5_SET(cmd_hca_cap, set_hca_cap, cmdif_checksum, 0);
486 MLX5_SET(cmd_hca_cap, set_hca_cap, log_uar_page_sz, PAGE_SHIFT - 12);
488 err = set_caps(dev, set_ctx, set_sz,
489 MLX5_SET_HCA_CAP_OP_MOD_GENERAL_DEVICE);
496 static int set_hca_ctrl(struct mlx5_core_dev *dev)
498 struct mlx5_reg_host_endianess he_in;
499 struct mlx5_reg_host_endianess he_out;
502 if (!mlx5_core_is_pf(dev))
505 memset(&he_in, 0, sizeof(he_in));
506 he_in.he = MLX5_SET_HOST_ENDIANNESS;
507 err = mlx5_core_access_reg(dev, &he_in, sizeof(he_in),
508 &he_out, sizeof(he_out),
509 MLX5_REG_HOST_ENDIANNESS, 0, 1);
513 int mlx5_core_enable_hca(struct mlx5_core_dev *dev, u16 func_id)
515 u32 out[MLX5_ST_SZ_DW(enable_hca_out)] = {0};
516 u32 in[MLX5_ST_SZ_DW(enable_hca_in)] = {0};
518 MLX5_SET(enable_hca_in, in, opcode, MLX5_CMD_OP_ENABLE_HCA);
519 MLX5_SET(enable_hca_in, in, function_id, func_id);
520 return mlx5_cmd_exec(dev, &in, sizeof(in), &out, sizeof(out));
523 int mlx5_core_disable_hca(struct mlx5_core_dev *dev, u16 func_id)
525 u32 out[MLX5_ST_SZ_DW(disable_hca_out)] = {0};
526 u32 in[MLX5_ST_SZ_DW(disable_hca_in)] = {0};
528 MLX5_SET(disable_hca_in, in, opcode, MLX5_CMD_OP_DISABLE_HCA);
529 MLX5_SET(disable_hca_in, in, function_id, func_id);
530 return mlx5_cmd_exec(dev, in, sizeof(in), out, sizeof(out));
533 cycle_t mlx5_read_internal_timer(struct mlx5_core_dev *dev)
535 u32 timer_h, timer_h1, timer_l;
537 timer_h = ioread32be(&dev->iseg->internal_timer_h);
538 timer_l = ioread32be(&dev->iseg->internal_timer_l);
539 timer_h1 = ioread32be(&dev->iseg->internal_timer_h);
540 if (timer_h != timer_h1) /* wrap around */
541 timer_l = ioread32be(&dev->iseg->internal_timer_l);
543 return (cycle_t)timer_l | (cycle_t)timer_h1 << 32;
546 static int mlx5_irq_set_affinity_hint(struct mlx5_core_dev *mdev, int i)
548 struct mlx5_priv *priv = &mdev->priv;
549 struct msix_entry *msix = priv->msix_arr;
550 int irq = msix[i + MLX5_EQ_VEC_COMP_BASE].vector;
552 if (!zalloc_cpumask_var(&priv->irq_info[i].mask, GFP_KERNEL)) {
553 mlx5_core_warn(mdev, "zalloc_cpumask_var failed");
557 cpumask_set_cpu(cpumask_local_spread(i, priv->numa_node),
558 priv->irq_info[i].mask);
560 if (IS_ENABLED(CONFIG_SMP) &&
561 irq_set_affinity_hint(irq, priv->irq_info[i].mask))
562 mlx5_core_warn(mdev, "irq_set_affinity_hint failed, irq 0x%.4x", irq);
567 static void mlx5_irq_clear_affinity_hint(struct mlx5_core_dev *mdev, int i)
569 struct mlx5_priv *priv = &mdev->priv;
570 struct msix_entry *msix = priv->msix_arr;
571 int irq = msix[i + MLX5_EQ_VEC_COMP_BASE].vector;
573 irq_set_affinity_hint(irq, NULL);
574 free_cpumask_var(priv->irq_info[i].mask);
577 static int mlx5_irq_set_affinity_hints(struct mlx5_core_dev *mdev)
582 for (i = 0; i < mdev->priv.eq_table.num_comp_vectors; i++) {
583 err = mlx5_irq_set_affinity_hint(mdev, i);
591 for (i--; i >= 0; i--)
592 mlx5_irq_clear_affinity_hint(mdev, i);
597 static void mlx5_irq_clear_affinity_hints(struct mlx5_core_dev *mdev)
601 for (i = 0; i < mdev->priv.eq_table.num_comp_vectors; i++)
602 mlx5_irq_clear_affinity_hint(mdev, i);
605 int mlx5_vector2eqn(struct mlx5_core_dev *dev, int vector, int *eqn,
608 struct mlx5_eq_table *table = &dev->priv.eq_table;
609 struct mlx5_eq *eq, *n;
612 spin_lock(&table->lock);
613 list_for_each_entry_safe(eq, n, &table->comp_eqs_list, list) {
614 if (eq->index == vector) {
621 spin_unlock(&table->lock);
625 EXPORT_SYMBOL(mlx5_vector2eqn);
627 struct mlx5_eq *mlx5_eqn2eq(struct mlx5_core_dev *dev, int eqn)
629 struct mlx5_eq_table *table = &dev->priv.eq_table;
632 spin_lock(&table->lock);
633 list_for_each_entry(eq, &table->comp_eqs_list, list)
634 if (eq->eqn == eqn) {
635 spin_unlock(&table->lock);
639 spin_unlock(&table->lock);
641 return ERR_PTR(-ENOENT);
644 static void free_comp_eqs(struct mlx5_core_dev *dev)
646 struct mlx5_eq_table *table = &dev->priv.eq_table;
647 struct mlx5_eq *eq, *n;
649 #ifdef CONFIG_RFS_ACCEL
651 free_irq_cpu_rmap(dev->rmap);
655 spin_lock(&table->lock);
656 list_for_each_entry_safe(eq, n, &table->comp_eqs_list, list) {
658 spin_unlock(&table->lock);
659 if (mlx5_destroy_unmap_eq(dev, eq))
660 mlx5_core_warn(dev, "failed to destroy EQ 0x%x\n",
663 spin_lock(&table->lock);
665 spin_unlock(&table->lock);
668 static int alloc_comp_eqs(struct mlx5_core_dev *dev)
670 struct mlx5_eq_table *table = &dev->priv.eq_table;
671 char name[MLX5_MAX_IRQ_NAME];
678 INIT_LIST_HEAD(&table->comp_eqs_list);
679 ncomp_vec = table->num_comp_vectors;
680 nent = MLX5_COMP_EQ_SIZE;
681 #ifdef CONFIG_RFS_ACCEL
682 dev->rmap = alloc_irq_cpu_rmap(ncomp_vec);
686 for (i = 0; i < ncomp_vec; i++) {
687 eq = kzalloc(sizeof(*eq), GFP_KERNEL);
693 #ifdef CONFIG_RFS_ACCEL
694 irq_cpu_rmap_add(dev->rmap,
695 dev->priv.msix_arr[i + MLX5_EQ_VEC_COMP_BASE].vector);
697 snprintf(name, MLX5_MAX_IRQ_NAME, "mlx5_comp%d", i);
698 err = mlx5_create_map_eq(dev, eq,
699 i + MLX5_EQ_VEC_COMP_BASE, nent, 0,
700 name, &dev->priv.uuari.uars[0]);
705 mlx5_core_dbg(dev, "allocated completion EQN %d\n", eq->eqn);
707 spin_lock(&table->lock);
708 list_add_tail(&eq->list, &table->comp_eqs_list);
709 spin_unlock(&table->lock);
719 static int mlx5_core_set_issi(struct mlx5_core_dev *dev)
721 u32 query_in[MLX5_ST_SZ_DW(query_issi_in)] = {0};
722 u32 query_out[MLX5_ST_SZ_DW(query_issi_out)] = {0};
726 MLX5_SET(query_issi_in, query_in, opcode, MLX5_CMD_OP_QUERY_ISSI);
727 err = mlx5_cmd_exec(dev, query_in, sizeof(query_in),
728 query_out, sizeof(query_out));
733 mlx5_cmd_mbox_status(query_out, &status, &syndrome);
734 if (!status || syndrome == MLX5_DRIVER_SYND) {
735 mlx5_core_err(dev, "Failed to query ISSI err(%d) status(%d) synd(%d)\n",
736 err, status, syndrome);
740 mlx5_core_warn(dev, "Query ISSI is not supported by FW, ISSI is 0\n");
745 sup_issi = MLX5_GET(query_issi_out, query_out, supported_issi_dw0);
747 if (sup_issi & (1 << 1)) {
748 u32 set_in[MLX5_ST_SZ_DW(set_issi_in)] = {0};
749 u32 set_out[MLX5_ST_SZ_DW(set_issi_out)] = {0};
751 MLX5_SET(set_issi_in, set_in, opcode, MLX5_CMD_OP_SET_ISSI);
752 MLX5_SET(set_issi_in, set_in, current_issi, 1);
753 err = mlx5_cmd_exec(dev, set_in, sizeof(set_in),
754 set_out, sizeof(set_out));
756 mlx5_core_err(dev, "Failed to set ISSI to 1 err(%d)\n",
764 } else if (sup_issi & (1 << 0) || !sup_issi) {
772 static int mlx5_pci_init(struct mlx5_core_dev *dev, struct mlx5_priv *priv)
774 struct pci_dev *pdev = dev->pdev;
777 pci_set_drvdata(dev->pdev, dev);
778 strncpy(priv->name, dev_name(&pdev->dev), MLX5_MAX_NAME_LEN);
779 priv->name[MLX5_MAX_NAME_LEN - 1] = 0;
781 mutex_init(&priv->pgdir_mutex);
782 INIT_LIST_HEAD(&priv->pgdir_list);
783 spin_lock_init(&priv->mkey_lock);
785 mutex_init(&priv->alloc_mutex);
787 priv->numa_node = dev_to_node(&dev->pdev->dev);
789 if (mlx5_debugfs_root)
791 debugfs_create_dir(pci_name(pdev), mlx5_debugfs_root);
793 err = mlx5_pci_enable_device(dev);
795 dev_err(&pdev->dev, "Cannot enable PCI device, aborting\n");
799 err = request_bar(pdev);
801 dev_err(&pdev->dev, "error requesting BARs, aborting\n");
805 pci_set_master(pdev);
807 err = set_dma_caps(pdev);
809 dev_err(&pdev->dev, "Failed setting DMA capabilities mask, aborting\n");
813 dev->iseg_base = pci_resource_start(dev->pdev, 0);
814 dev->iseg = ioremap(dev->iseg_base, sizeof(*dev->iseg));
817 dev_err(&pdev->dev, "Failed mapping initialization segment, aborting\n");
824 pci_clear_master(dev->pdev);
825 release_bar(dev->pdev);
827 mlx5_pci_disable_device(dev);
830 debugfs_remove(priv->dbg_root);
834 static void mlx5_pci_close(struct mlx5_core_dev *dev, struct mlx5_priv *priv)
837 pci_clear_master(dev->pdev);
838 release_bar(dev->pdev);
839 mlx5_pci_disable_device(dev);
840 debugfs_remove_recursive(priv->dbg_root);
843 static int mlx5_init_once(struct mlx5_core_dev *dev, struct mlx5_priv *priv)
845 struct pci_dev *pdev = dev->pdev;
848 err = mlx5_query_board_id(dev);
850 dev_err(&pdev->dev, "query board id failed\n");
854 err = mlx5_eq_init(dev);
856 dev_err(&pdev->dev, "failed to initialize eq\n");
860 MLX5_INIT_DOORBELL_LOCK(&priv->cq_uar_lock);
862 err = mlx5_init_cq_table(dev);
864 dev_err(&pdev->dev, "failed to initialize cq table\n");
868 mlx5_init_qp_table(dev);
870 mlx5_init_srq_table(dev);
872 mlx5_init_mkey_table(dev);
874 err = mlx5_init_rl_table(dev);
876 dev_err(&pdev->dev, "Failed to init rate limiting\n");
877 goto err_tables_cleanup;
880 #ifdef CONFIG_MLX5_CORE_EN
881 err = mlx5_eswitch_init(dev);
883 dev_err(&pdev->dev, "Failed to init eswitch %d\n", err);
888 err = mlx5_sriov_init(dev);
890 dev_err(&pdev->dev, "Failed to init sriov %d\n", err);
891 goto err_eswitch_cleanup;
897 #ifdef CONFIG_MLX5_CORE_EN
898 mlx5_eswitch_cleanup(dev->priv.eswitch);
902 mlx5_cleanup_rl_table(dev);
905 mlx5_cleanup_mkey_table(dev);
906 mlx5_cleanup_srq_table(dev);
907 mlx5_cleanup_qp_table(dev);
908 mlx5_cleanup_cq_table(dev);
911 mlx5_eq_cleanup(dev);
917 static void mlx5_cleanup_once(struct mlx5_core_dev *dev)
919 mlx5_sriov_cleanup(dev);
920 #ifdef CONFIG_MLX5_CORE_EN
921 mlx5_eswitch_cleanup(dev->priv.eswitch);
923 mlx5_cleanup_rl_table(dev);
924 mlx5_cleanup_mkey_table(dev);
925 mlx5_cleanup_srq_table(dev);
926 mlx5_cleanup_qp_table(dev);
927 mlx5_cleanup_cq_table(dev);
928 mlx5_eq_cleanup(dev);
931 static int mlx5_load_one(struct mlx5_core_dev *dev, struct mlx5_priv *priv,
934 struct pci_dev *pdev = dev->pdev;
937 mutex_lock(&dev->intf_state_mutex);
938 if (test_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state)) {
939 dev_warn(&dev->pdev->dev, "%s: interface is up, NOP\n",
944 dev_info(&pdev->dev, "firmware version: %d.%d.%d\n", fw_rev_maj(dev),
945 fw_rev_min(dev), fw_rev_sub(dev));
947 /* on load removing any previous indication of internal error, device is
950 dev->state = MLX5_DEVICE_STATE_UP;
952 /* wait for firmware to accept initialization segments configurations
954 err = wait_fw_init(dev, FW_PRE_INIT_TIMEOUT_MILI);
956 dev_err(&dev->pdev->dev, "Firmware over %d MS in pre-initializing state, aborting\n",
957 FW_PRE_INIT_TIMEOUT_MILI);
961 err = mlx5_cmd_init(dev);
963 dev_err(&pdev->dev, "Failed initializing command interface, aborting\n");
967 err = wait_fw_init(dev, FW_INIT_TIMEOUT_MILI);
969 dev_err(&dev->pdev->dev, "Firmware over %d MS in initializing state, aborting\n",
970 FW_INIT_TIMEOUT_MILI);
971 goto err_cmd_cleanup;
974 err = mlx5_core_enable_hca(dev, 0);
976 dev_err(&pdev->dev, "enable hca failed\n");
977 goto err_cmd_cleanup;
980 err = mlx5_core_set_issi(dev);
982 dev_err(&pdev->dev, "failed to set issi\n");
983 goto err_disable_hca;
986 err = mlx5_satisfy_startup_pages(dev, 1);
988 dev_err(&pdev->dev, "failed to allocate boot pages\n");
989 goto err_disable_hca;
992 err = set_hca_ctrl(dev);
994 dev_err(&pdev->dev, "set_hca_ctrl failed\n");
995 goto reclaim_boot_pages;
998 err = handle_hca_cap(dev);
1000 dev_err(&pdev->dev, "handle_hca_cap failed\n");
1001 goto reclaim_boot_pages;
1004 err = handle_hca_cap_atomic(dev);
1006 dev_err(&pdev->dev, "handle_hca_cap_atomic failed\n");
1007 goto reclaim_boot_pages;
1010 err = mlx5_satisfy_startup_pages(dev, 0);
1012 dev_err(&pdev->dev, "failed to allocate init pages\n");
1013 goto reclaim_boot_pages;
1016 err = mlx5_pagealloc_start(dev);
1018 dev_err(&pdev->dev, "mlx5_pagealloc_start failed\n");
1019 goto reclaim_boot_pages;
1022 err = mlx5_cmd_init_hca(dev);
1024 dev_err(&pdev->dev, "init hca failed\n");
1025 goto err_pagealloc_stop;
1028 mlx5_start_health_poll(dev);
1030 err = mlx5_query_hca_caps(dev);
1032 dev_err(&pdev->dev, "query hca failed\n");
1036 if (boot && mlx5_init_once(dev, priv)) {
1037 dev_err(&pdev->dev, "sw objs init failed\n");
1041 err = mlx5_enable_msix(dev);
1043 dev_err(&pdev->dev, "enable msix failed\n");
1044 goto err_cleanup_once;
1047 err = mlx5_alloc_uuars(dev, &priv->uuari);
1049 dev_err(&pdev->dev, "Failed allocating uar, aborting\n");
1050 goto err_disable_msix;
1053 err = mlx5_start_eqs(dev);
1055 dev_err(&pdev->dev, "Failed to start pages and async EQs\n");
1059 err = alloc_comp_eqs(dev);
1061 dev_err(&pdev->dev, "Failed to alloc completion EQs\n");
1065 err = mlx5_irq_set_affinity_hints(dev);
1067 dev_err(&pdev->dev, "Failed to alloc affinity hint cpumask\n");
1068 goto err_affinity_hints;
1071 err = mlx5_init_fs(dev);
1073 dev_err(&pdev->dev, "Failed to init flow steering\n");
1077 #ifdef CONFIG_MLX5_CORE_EN
1078 mlx5_eswitch_attach(dev->priv.eswitch);
1081 err = mlx5_sriov_attach(dev);
1083 dev_err(&pdev->dev, "sriov init failed %d\n", err);
1087 if (mlx5_device_registered(dev)) {
1088 mlx5_attach_device(dev);
1090 err = mlx5_register_device(dev);
1092 dev_err(&pdev->dev, "mlx5_register_device failed %d\n", err);
1097 clear_bit(MLX5_INTERFACE_STATE_DOWN, &dev->intf_state);
1098 set_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state);
1100 mutex_unlock(&dev->intf_state_mutex);
1105 mlx5_sriov_detach(dev);
1108 #ifdef CONFIG_MLX5_CORE_EN
1109 mlx5_eswitch_detach(dev->priv.eswitch);
1111 mlx5_cleanup_fs(dev);
1114 mlx5_irq_clear_affinity_hints(dev);
1123 mlx5_free_uuars(dev, &priv->uuari);
1126 mlx5_disable_msix(dev);
1130 mlx5_cleanup_once(dev);
1133 mlx5_stop_health_poll(dev, boot);
1134 if (mlx5_cmd_teardown_hca(dev)) {
1135 dev_err(&dev->pdev->dev, "tear_down_hca failed, skip cleanup\n");
1140 mlx5_pagealloc_stop(dev);
1143 mlx5_reclaim_startup_pages(dev);
1146 mlx5_core_disable_hca(dev, 0);
1149 mlx5_cmd_cleanup(dev);
1152 dev->state = MLX5_DEVICE_STATE_INTERNAL_ERROR;
1153 mutex_unlock(&dev->intf_state_mutex);
1158 static int mlx5_unload_one(struct mlx5_core_dev *dev, struct mlx5_priv *priv,
1164 mlx5_drain_health_recovery(dev);
1166 mutex_lock(&dev->intf_state_mutex);
1167 if (test_bit(MLX5_INTERFACE_STATE_DOWN, &dev->intf_state)) {
1168 dev_warn(&dev->pdev->dev, "%s: interface is down, NOP\n",
1171 mlx5_cleanup_once(dev);
1175 if (mlx5_device_registered(dev))
1176 mlx5_detach_device(dev);
1178 mlx5_sriov_detach(dev);
1179 #ifdef CONFIG_MLX5_CORE_EN
1180 mlx5_eswitch_detach(dev->priv.eswitch);
1182 mlx5_cleanup_fs(dev);
1183 mlx5_irq_clear_affinity_hints(dev);
1186 mlx5_free_uuars(dev, &priv->uuari);
1187 mlx5_disable_msix(dev);
1189 mlx5_cleanup_once(dev);
1190 mlx5_stop_health_poll(dev, cleanup);
1191 err = mlx5_cmd_teardown_hca(dev);
1193 dev_err(&dev->pdev->dev, "tear_down_hca failed, skip cleanup\n");
1196 mlx5_pagealloc_stop(dev);
1197 mlx5_reclaim_startup_pages(dev);
1198 mlx5_core_disable_hca(dev, 0);
1199 mlx5_cmd_cleanup(dev);
1202 clear_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state);
1203 set_bit(MLX5_INTERFACE_STATE_DOWN, &dev->intf_state);
1204 mutex_unlock(&dev->intf_state_mutex);
1208 struct mlx5_core_event_handler {
1209 void (*event)(struct mlx5_core_dev *dev,
1210 enum mlx5_dev_event event,
1214 static const struct devlink_ops mlx5_devlink_ops = {
1215 #ifdef CONFIG_MLX5_CORE_EN
1216 .eswitch_mode_set = mlx5_devlink_eswitch_mode_set,
1217 .eswitch_mode_get = mlx5_devlink_eswitch_mode_get,
1221 #define MLX5_IB_MOD "mlx5_ib"
1222 static int init_one(struct pci_dev *pdev,
1223 const struct pci_device_id *id)
1225 struct mlx5_core_dev *dev;
1226 struct devlink *devlink;
1227 struct mlx5_priv *priv;
1230 devlink = devlink_alloc(&mlx5_devlink_ops, sizeof(*dev));
1232 dev_err(&pdev->dev, "kzalloc failed\n");
1236 dev = devlink_priv(devlink);
1238 priv->pci_dev_data = id->driver_data;
1240 pci_set_drvdata(pdev, dev);
1243 dev->event = mlx5_core_event;
1244 dev->profile = &profile[prof_sel];
1246 INIT_LIST_HEAD(&priv->ctx_list);
1247 spin_lock_init(&priv->ctx_lock);
1248 mutex_init(&dev->pci_status_mutex);
1249 mutex_init(&dev->intf_state_mutex);
1250 err = mlx5_pci_init(dev, priv);
1252 dev_err(&pdev->dev, "mlx5_pci_init failed with error code %d\n", err);
1256 err = mlx5_health_init(dev);
1258 dev_err(&pdev->dev, "mlx5_health_init failed with error code %d\n", err);
1262 mlx5_pagealloc_init(dev);
1264 err = mlx5_load_one(dev, priv, true);
1266 dev_err(&pdev->dev, "mlx5_load_one failed with error code %d\n", err);
1270 err = request_module_nowait(MLX5_IB_MOD);
1272 pr_info("failed request module on %s\n", MLX5_IB_MOD);
1274 err = devlink_register(devlink, &pdev->dev);
1278 pci_save_state(pdev);
1282 mlx5_unload_one(dev, priv, true);
1284 mlx5_pagealloc_cleanup(dev);
1285 mlx5_health_cleanup(dev);
1287 mlx5_pci_close(dev, priv);
1289 pci_set_drvdata(pdev, NULL);
1290 devlink_free(devlink);
1295 static void remove_one(struct pci_dev *pdev)
1297 struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
1298 struct devlink *devlink = priv_to_devlink(dev);
1299 struct mlx5_priv *priv = &dev->priv;
1301 devlink_unregister(devlink);
1302 mlx5_unregister_device(dev);
1304 if (mlx5_unload_one(dev, priv, true)) {
1305 dev_err(&dev->pdev->dev, "mlx5_unload_one failed\n");
1306 mlx5_health_cleanup(dev);
1310 mlx5_pagealloc_cleanup(dev);
1311 mlx5_health_cleanup(dev);
1312 mlx5_pci_close(dev, priv);
1313 pci_set_drvdata(pdev, NULL);
1314 devlink_free(devlink);
1317 static pci_ers_result_t mlx5_pci_err_detected(struct pci_dev *pdev,
1318 pci_channel_state_t state)
1320 struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
1321 struct mlx5_priv *priv = &dev->priv;
1323 dev_info(&pdev->dev, "%s was called\n", __func__);
1325 mlx5_enter_error_state(dev);
1326 mlx5_unload_one(dev, priv, false);
1327 /* In case of kernel call drain the health wq */
1329 mlx5_drain_health_wq(dev);
1330 mlx5_pci_disable_device(dev);
1333 return state == pci_channel_io_perm_failure ?
1334 PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_NEED_RESET;
1337 /* wait for the device to show vital signs by waiting
1338 * for the health counter to start counting.
1340 static int wait_vital(struct pci_dev *pdev)
1342 struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
1343 struct mlx5_core_health *health = &dev->priv.health;
1344 const int niter = 100;
1349 for (i = 0; i < niter; i++) {
1350 count = ioread32be(health->health_counter);
1351 if (count && count != 0xffffffff) {
1352 if (last_count && last_count != count) {
1353 dev_info(&pdev->dev, "Counter value 0x%x after %d iterations\n", count, i);
1364 static pci_ers_result_t mlx5_pci_slot_reset(struct pci_dev *pdev)
1366 struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
1369 dev_info(&pdev->dev, "%s was called\n", __func__);
1371 err = mlx5_pci_enable_device(dev);
1373 dev_err(&pdev->dev, "%s: mlx5_pci_enable_device failed with error code: %d\n"
1375 return PCI_ERS_RESULT_DISCONNECT;
1378 pci_set_master(pdev);
1379 pci_restore_state(pdev);
1380 pci_save_state(pdev);
1382 if (wait_vital(pdev)) {
1383 dev_err(&pdev->dev, "%s: wait_vital timed out\n", __func__);
1384 return PCI_ERS_RESULT_DISCONNECT;
1387 return PCI_ERS_RESULT_RECOVERED;
1390 static void mlx5_pci_resume(struct pci_dev *pdev)
1392 struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
1393 struct mlx5_priv *priv = &dev->priv;
1396 dev_info(&pdev->dev, "%s was called\n", __func__);
1398 err = mlx5_load_one(dev, priv, false);
1400 dev_err(&pdev->dev, "%s: mlx5_load_one failed with error code: %d\n"
1403 dev_info(&pdev->dev, "%s: device recovered\n", __func__);
1406 static const struct pci_error_handlers mlx5_err_handler = {
1407 .error_detected = mlx5_pci_err_detected,
1408 .slot_reset = mlx5_pci_slot_reset,
1409 .resume = mlx5_pci_resume
1412 static void shutdown(struct pci_dev *pdev)
1414 struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
1415 struct mlx5_priv *priv = &dev->priv;
1417 dev_info(&pdev->dev, "Shutdown was called\n");
1418 /* Notify mlx5 clients that the kernel is being shut down */
1419 set_bit(MLX5_INTERFACE_STATE_SHUTDOWN, &dev->intf_state);
1420 mlx5_unload_one(dev, priv, false);
1421 mlx5_pci_disable_device(dev);
1424 static const struct pci_device_id mlx5_core_pci_table[] = {
1425 { PCI_VDEVICE(MELLANOX, 0x1011) }, /* Connect-IB */
1426 { PCI_VDEVICE(MELLANOX, 0x1012), MLX5_PCI_DEV_IS_VF}, /* Connect-IB VF */
1427 { PCI_VDEVICE(MELLANOX, 0x1013) }, /* ConnectX-4 */
1428 { PCI_VDEVICE(MELLANOX, 0x1014), MLX5_PCI_DEV_IS_VF}, /* ConnectX-4 VF */
1429 { PCI_VDEVICE(MELLANOX, 0x1015) }, /* ConnectX-4LX */
1430 { PCI_VDEVICE(MELLANOX, 0x1016), MLX5_PCI_DEV_IS_VF}, /* ConnectX-4LX VF */
1431 { PCI_VDEVICE(MELLANOX, 0x1017) }, /* ConnectX-5, PCIe 3.0 */
1432 { PCI_VDEVICE(MELLANOX, 0x1018), MLX5_PCI_DEV_IS_VF}, /* ConnectX-5 VF */
1433 { PCI_VDEVICE(MELLANOX, 0x1019) }, /* ConnectX-5, PCIe 4.0 */
1437 MODULE_DEVICE_TABLE(pci, mlx5_core_pci_table);
1439 void mlx5_disable_device(struct mlx5_core_dev *dev)
1441 mlx5_pci_err_detected(dev->pdev, 0);
1444 void mlx5_recover_device(struct mlx5_core_dev *dev)
1446 mlx5_pci_disable_device(dev);
1447 if (mlx5_pci_slot_reset(dev->pdev) == PCI_ERS_RESULT_RECOVERED)
1448 mlx5_pci_resume(dev->pdev);
1451 static struct pci_driver mlx5_core_driver = {
1452 .name = DRIVER_NAME,
1453 .id_table = mlx5_core_pci_table,
1455 .remove = remove_one,
1456 .shutdown = shutdown,
1457 .err_handler = &mlx5_err_handler,
1458 .sriov_configure = mlx5_core_sriov_configure,
1461 static void mlx5_core_verify_params(void)
1463 if (prof_sel >= ARRAY_SIZE(profile)) {
1464 pr_warn("mlx5_core: WARNING: Invalid module parameter prof_sel %d, valid range 0-%zu, changing back to default(%d)\n",
1466 ARRAY_SIZE(profile) - 1,
1468 prof_sel = MLX5_DEFAULT_PROF;
1472 static int __init init(void)
1476 mlx5_core_verify_params();
1477 mlx5_register_debugfs();
1479 err = pci_register_driver(&mlx5_core_driver);
1483 #ifdef CONFIG_MLX5_CORE_EN
1490 mlx5_unregister_debugfs();
1494 static void __exit cleanup(void)
1496 #ifdef CONFIG_MLX5_CORE_EN
1499 pci_unregister_driver(&mlx5_core_driver);
1500 mlx5_unregister_debugfs();
1504 module_exit(cleanup);