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/delay.h>
50 #include <linux/mlx5/mlx5_ifc.h>
51 #ifdef CONFIG_RFS_ACCEL
52 #include <linux/cpu_rmap.h>
54 #include <net/devlink.h>
55 #include "mlx5_core.h"
57 #ifdef CONFIG_MLX5_CORE_EN
61 MODULE_AUTHOR("Eli Cohen <eli@mellanox.com>");
62 MODULE_DESCRIPTION("Mellanox Connect-IB, ConnectX-4 core driver");
63 MODULE_LICENSE("Dual BSD/GPL");
64 MODULE_VERSION(DRIVER_VERSION);
66 int mlx5_core_debug_mask;
67 module_param_named(debug_mask, mlx5_core_debug_mask, int, 0644);
68 MODULE_PARM_DESC(debug_mask, "debug mask: 1 = dump cmd data, 2 = dump cmd exec time, 3 = both. Default=0");
70 #define MLX5_DEFAULT_PROF 2
71 static int prof_sel = MLX5_DEFAULT_PROF;
72 module_param_named(prof_sel, prof_sel, int, 0444);
73 MODULE_PARM_DESC(prof_sel, "profile selector. Valid range 0 - 2");
75 static LIST_HEAD(intf_list);
76 static LIST_HEAD(dev_list);
77 static DEFINE_MUTEX(intf_mutex);
79 struct mlx5_device_context {
80 struct list_head list;
81 struct mlx5_interface *intf;
86 MLX5_ATOMIC_REQ_MODE_BE = 0x0,
87 MLX5_ATOMIC_REQ_MODE_HOST_ENDIANNESS = 0x1,
90 static struct mlx5_profile profile[] = {
95 .mask = MLX5_PROF_MASK_QP_SIZE,
99 .mask = MLX5_PROF_MASK_QP_SIZE |
100 MLX5_PROF_MASK_MR_CACHE,
169 #define FW_INIT_TIMEOUT_MILI 2000
170 #define FW_INIT_WAIT_MS 2
172 static int wait_fw_init(struct mlx5_core_dev *dev, u32 max_wait_mili)
174 unsigned long end = jiffies + msecs_to_jiffies(max_wait_mili);
177 while (fw_initializing(dev)) {
178 if (time_after(jiffies, end)) {
182 msleep(FW_INIT_WAIT_MS);
188 static int set_dma_caps(struct pci_dev *pdev)
192 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
194 dev_warn(&pdev->dev, "Warning: couldn't set 64-bit PCI DMA mask\n");
195 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
197 dev_err(&pdev->dev, "Can't set PCI DMA mask, aborting\n");
202 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
205 "Warning: couldn't set 64-bit consistent PCI DMA mask\n");
206 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
209 "Can't set consistent PCI DMA mask, aborting\n");
214 dma_set_max_seg_size(&pdev->dev, 2u * 1024 * 1024 * 1024);
218 static int mlx5_pci_enable_device(struct mlx5_core_dev *dev)
220 struct pci_dev *pdev = dev->pdev;
223 mutex_lock(&dev->pci_status_mutex);
224 if (dev->pci_status == MLX5_PCI_STATUS_DISABLED) {
225 err = pci_enable_device(pdev);
227 dev->pci_status = MLX5_PCI_STATUS_ENABLED;
229 mutex_unlock(&dev->pci_status_mutex);
234 static void mlx5_pci_disable_device(struct mlx5_core_dev *dev)
236 struct pci_dev *pdev = dev->pdev;
238 mutex_lock(&dev->pci_status_mutex);
239 if (dev->pci_status == MLX5_PCI_STATUS_ENABLED) {
240 pci_disable_device(pdev);
241 dev->pci_status = MLX5_PCI_STATUS_DISABLED;
243 mutex_unlock(&dev->pci_status_mutex);
246 static int request_bar(struct pci_dev *pdev)
250 if (!(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) {
251 dev_err(&pdev->dev, "Missing registers BAR, aborting\n");
255 err = pci_request_regions(pdev, DRIVER_NAME);
257 dev_err(&pdev->dev, "Couldn't get PCI resources, aborting\n");
262 static void release_bar(struct pci_dev *pdev)
264 pci_release_regions(pdev);
267 static int mlx5_enable_msix(struct mlx5_core_dev *dev)
269 struct mlx5_priv *priv = &dev->priv;
270 struct mlx5_eq_table *table = &priv->eq_table;
271 int num_eqs = 1 << MLX5_CAP_GEN(dev, log_max_eq);
275 nvec = MLX5_CAP_GEN(dev, num_ports) * num_online_cpus() +
276 MLX5_EQ_VEC_COMP_BASE;
277 nvec = min_t(int, nvec, num_eqs);
278 if (nvec <= MLX5_EQ_VEC_COMP_BASE)
281 priv->msix_arr = kcalloc(nvec, sizeof(*priv->msix_arr), GFP_KERNEL);
283 priv->irq_info = kcalloc(nvec, sizeof(*priv->irq_info), GFP_KERNEL);
284 if (!priv->msix_arr || !priv->irq_info)
287 for (i = 0; i < nvec; i++)
288 priv->msix_arr[i].entry = i;
290 nvec = pci_enable_msix_range(dev->pdev, priv->msix_arr,
291 MLX5_EQ_VEC_COMP_BASE + 1, nvec);
295 table->num_comp_vectors = nvec - MLX5_EQ_VEC_COMP_BASE;
300 kfree(priv->irq_info);
301 kfree(priv->msix_arr);
305 static void mlx5_disable_msix(struct mlx5_core_dev *dev)
307 struct mlx5_priv *priv = &dev->priv;
309 pci_disable_msix(dev->pdev);
310 kfree(priv->irq_info);
311 kfree(priv->msix_arr);
314 struct mlx5_reg_host_endianess {
320 #define CAP_MASK(pos, size) ((u64)((1 << (size)) - 1) << (pos))
323 MLX5_CAP_BITS_RW_MASK = CAP_MASK(MLX5_CAP_OFF_CMDIF_CSUM, 2) |
324 MLX5_DEV_CAP_FLAG_DCT,
327 static u16 to_fw_pkey_sz(u32 size)
343 pr_warn("invalid pkey table size %d\n", size);
348 static int mlx5_core_get_caps_mode(struct mlx5_core_dev *dev,
349 enum mlx5_cap_type cap_type,
350 enum mlx5_cap_mode cap_mode)
352 u8 in[MLX5_ST_SZ_BYTES(query_hca_cap_in)];
353 int out_sz = MLX5_ST_SZ_BYTES(query_hca_cap_out);
354 void *out, *hca_caps;
355 u16 opmod = (cap_type << 1) | (cap_mode & 0x01);
358 memset(in, 0, sizeof(in));
359 out = kzalloc(out_sz, GFP_KERNEL);
363 MLX5_SET(query_hca_cap_in, in, opcode, MLX5_CMD_OP_QUERY_HCA_CAP);
364 MLX5_SET(query_hca_cap_in, in, op_mod, opmod);
365 err = mlx5_cmd_exec(dev, in, sizeof(in), out, out_sz);
369 err = mlx5_cmd_status_to_err_v2(out);
372 "QUERY_HCA_CAP : type(%x) opmode(%x) Failed(%d)\n",
373 cap_type, cap_mode, err);
377 hca_caps = MLX5_ADDR_OF(query_hca_cap_out, out, capability);
380 case HCA_CAP_OPMOD_GET_MAX:
381 memcpy(dev->hca_caps_max[cap_type], hca_caps,
382 MLX5_UN_SZ_BYTES(hca_cap_union));
384 case HCA_CAP_OPMOD_GET_CUR:
385 memcpy(dev->hca_caps_cur[cap_type], hca_caps,
386 MLX5_UN_SZ_BYTES(hca_cap_union));
390 "Tried to query dev cap type(%x) with wrong opmode(%x)\n",
400 int mlx5_core_get_caps(struct mlx5_core_dev *dev, enum mlx5_cap_type cap_type)
404 ret = mlx5_core_get_caps_mode(dev, cap_type, HCA_CAP_OPMOD_GET_CUR);
407 return mlx5_core_get_caps_mode(dev, cap_type, HCA_CAP_OPMOD_GET_MAX);
410 static int set_caps(struct mlx5_core_dev *dev, void *in, int in_sz, int opmod)
412 u32 out[MLX5_ST_SZ_DW(set_hca_cap_out)];
415 memset(out, 0, sizeof(out));
417 MLX5_SET(set_hca_cap_in, in, opcode, MLX5_CMD_OP_SET_HCA_CAP);
418 MLX5_SET(set_hca_cap_in, in, op_mod, opmod << 1);
419 err = mlx5_cmd_exec(dev, in, in_sz, out, sizeof(out));
423 err = mlx5_cmd_status_to_err_v2(out);
428 static int handle_hca_cap_atomic(struct mlx5_core_dev *dev)
432 int set_sz = MLX5_ST_SZ_BYTES(set_hca_cap_in);
436 if (MLX5_CAP_GEN(dev, atomic)) {
437 err = mlx5_core_get_caps(dev, MLX5_CAP_ATOMIC);
446 supported_atomic_req_8B_endianess_mode_1);
448 if (req_endianness != MLX5_ATOMIC_REQ_MODE_HOST_ENDIANNESS)
451 set_ctx = kzalloc(set_sz, GFP_KERNEL);
455 set_hca_cap = MLX5_ADDR_OF(set_hca_cap_in, set_ctx, capability);
457 /* Set requestor to host endianness */
458 MLX5_SET(atomic_caps, set_hca_cap, atomic_req_8B_endianess_mode,
459 MLX5_ATOMIC_REQ_MODE_HOST_ENDIANNESS);
461 err = set_caps(dev, set_ctx, set_sz, MLX5_SET_HCA_CAP_OP_MOD_ATOMIC);
467 static int handle_hca_cap(struct mlx5_core_dev *dev)
469 void *set_ctx = NULL;
470 struct mlx5_profile *prof = dev->profile;
472 int set_sz = MLX5_ST_SZ_BYTES(set_hca_cap_in);
475 set_ctx = kzalloc(set_sz, GFP_KERNEL);
479 err = mlx5_core_get_caps(dev, MLX5_CAP_GENERAL);
483 set_hca_cap = MLX5_ADDR_OF(set_hca_cap_in, set_ctx,
485 memcpy(set_hca_cap, dev->hca_caps_cur[MLX5_CAP_GENERAL],
486 MLX5_ST_SZ_BYTES(cmd_hca_cap));
488 mlx5_core_dbg(dev, "Current Pkey table size %d Setting new size %d\n",
489 mlx5_to_sw_pkey_sz(MLX5_CAP_GEN(dev, pkey_table_size)),
491 /* we limit the size of the pkey table to 128 entries for now */
492 MLX5_SET(cmd_hca_cap, set_hca_cap, pkey_table_size,
495 if (prof->mask & MLX5_PROF_MASK_QP_SIZE)
496 MLX5_SET(cmd_hca_cap, set_hca_cap, log_max_qp,
499 /* disable cmdif checksum */
500 MLX5_SET(cmd_hca_cap, set_hca_cap, cmdif_checksum, 0);
502 MLX5_SET(cmd_hca_cap, set_hca_cap, log_uar_page_sz, PAGE_SHIFT - 12);
504 err = set_caps(dev, set_ctx, set_sz,
505 MLX5_SET_HCA_CAP_OP_MOD_GENERAL_DEVICE);
512 static int set_hca_ctrl(struct mlx5_core_dev *dev)
514 struct mlx5_reg_host_endianess he_in;
515 struct mlx5_reg_host_endianess he_out;
518 if (!mlx5_core_is_pf(dev))
521 memset(&he_in, 0, sizeof(he_in));
522 he_in.he = MLX5_SET_HOST_ENDIANNESS;
523 err = mlx5_core_access_reg(dev, &he_in, sizeof(he_in),
524 &he_out, sizeof(he_out),
525 MLX5_REG_HOST_ENDIANNESS, 0, 1);
529 int mlx5_core_enable_hca(struct mlx5_core_dev *dev, u16 func_id)
531 u32 out[MLX5_ST_SZ_DW(enable_hca_out)];
532 u32 in[MLX5_ST_SZ_DW(enable_hca_in)];
535 memset(in, 0, sizeof(in));
536 MLX5_SET(enable_hca_in, in, opcode, MLX5_CMD_OP_ENABLE_HCA);
537 MLX5_SET(enable_hca_in, in, function_id, func_id);
538 memset(out, 0, sizeof(out));
540 err = mlx5_cmd_exec(dev, &in, sizeof(in), &out, sizeof(out));
544 return mlx5_cmd_status_to_err_v2(out);
547 int mlx5_core_disable_hca(struct mlx5_core_dev *dev, u16 func_id)
549 u32 out[MLX5_ST_SZ_DW(disable_hca_out)];
550 u32 in[MLX5_ST_SZ_DW(disable_hca_in)];
553 memset(in, 0, sizeof(in));
554 MLX5_SET(disable_hca_in, in, opcode, MLX5_CMD_OP_DISABLE_HCA);
555 MLX5_SET(disable_hca_in, in, function_id, func_id);
556 memset(out, 0, sizeof(out));
557 err = mlx5_cmd_exec(dev, in, sizeof(in), out, sizeof(out));
561 return mlx5_cmd_status_to_err_v2(out);
564 cycle_t mlx5_read_internal_timer(struct mlx5_core_dev *dev)
566 u32 timer_h, timer_h1, timer_l;
568 timer_h = ioread32be(&dev->iseg->internal_timer_h);
569 timer_l = ioread32be(&dev->iseg->internal_timer_l);
570 timer_h1 = ioread32be(&dev->iseg->internal_timer_h);
571 if (timer_h != timer_h1) /* wrap around */
572 timer_l = ioread32be(&dev->iseg->internal_timer_l);
574 return (cycle_t)timer_l | (cycle_t)timer_h1 << 32;
577 static int mlx5_irq_set_affinity_hint(struct mlx5_core_dev *mdev, int i)
579 struct mlx5_priv *priv = &mdev->priv;
580 struct msix_entry *msix = priv->msix_arr;
581 int irq = msix[i + MLX5_EQ_VEC_COMP_BASE].vector;
582 int numa_node = priv->numa_node;
585 if (!zalloc_cpumask_var(&priv->irq_info[i].mask, GFP_KERNEL)) {
586 mlx5_core_warn(mdev, "zalloc_cpumask_var failed");
590 cpumask_set_cpu(cpumask_local_spread(i, numa_node),
591 priv->irq_info[i].mask);
593 err = irq_set_affinity_hint(irq, priv->irq_info[i].mask);
595 mlx5_core_warn(mdev, "irq_set_affinity_hint failed,irq 0x%.4x",
603 free_cpumask_var(priv->irq_info[i].mask);
607 static void mlx5_irq_clear_affinity_hint(struct mlx5_core_dev *mdev, int i)
609 struct mlx5_priv *priv = &mdev->priv;
610 struct msix_entry *msix = priv->msix_arr;
611 int irq = msix[i + MLX5_EQ_VEC_COMP_BASE].vector;
613 irq_set_affinity_hint(irq, NULL);
614 free_cpumask_var(priv->irq_info[i].mask);
617 static int mlx5_irq_set_affinity_hints(struct mlx5_core_dev *mdev)
622 for (i = 0; i < mdev->priv.eq_table.num_comp_vectors; i++) {
623 err = mlx5_irq_set_affinity_hint(mdev, i);
631 for (i--; i >= 0; i--)
632 mlx5_irq_clear_affinity_hint(mdev, i);
637 static void mlx5_irq_clear_affinity_hints(struct mlx5_core_dev *mdev)
641 for (i = 0; i < mdev->priv.eq_table.num_comp_vectors; i++)
642 mlx5_irq_clear_affinity_hint(mdev, i);
645 int mlx5_vector2eqn(struct mlx5_core_dev *dev, int vector, int *eqn,
648 struct mlx5_eq_table *table = &dev->priv.eq_table;
649 struct mlx5_eq *eq, *n;
652 spin_lock(&table->lock);
653 list_for_each_entry_safe(eq, n, &table->comp_eqs_list, list) {
654 if (eq->index == vector) {
661 spin_unlock(&table->lock);
665 EXPORT_SYMBOL(mlx5_vector2eqn);
667 struct mlx5_eq *mlx5_eqn2eq(struct mlx5_core_dev *dev, int eqn)
669 struct mlx5_eq_table *table = &dev->priv.eq_table;
672 spin_lock(&table->lock);
673 list_for_each_entry(eq, &table->comp_eqs_list, list)
674 if (eq->eqn == eqn) {
675 spin_unlock(&table->lock);
679 spin_unlock(&table->lock);
681 return ERR_PTR(-ENOENT);
684 static void free_comp_eqs(struct mlx5_core_dev *dev)
686 struct mlx5_eq_table *table = &dev->priv.eq_table;
687 struct mlx5_eq *eq, *n;
689 #ifdef CONFIG_RFS_ACCEL
691 free_irq_cpu_rmap(dev->rmap);
695 spin_lock(&table->lock);
696 list_for_each_entry_safe(eq, n, &table->comp_eqs_list, list) {
698 spin_unlock(&table->lock);
699 if (mlx5_destroy_unmap_eq(dev, eq))
700 mlx5_core_warn(dev, "failed to destroy EQ 0x%x\n",
703 spin_lock(&table->lock);
705 spin_unlock(&table->lock);
708 static int alloc_comp_eqs(struct mlx5_core_dev *dev)
710 struct mlx5_eq_table *table = &dev->priv.eq_table;
711 char name[MLX5_MAX_IRQ_NAME];
718 INIT_LIST_HEAD(&table->comp_eqs_list);
719 ncomp_vec = table->num_comp_vectors;
720 nent = MLX5_COMP_EQ_SIZE;
721 #ifdef CONFIG_RFS_ACCEL
722 dev->rmap = alloc_irq_cpu_rmap(ncomp_vec);
726 for (i = 0; i < ncomp_vec; i++) {
727 eq = kzalloc(sizeof(*eq), GFP_KERNEL);
733 #ifdef CONFIG_RFS_ACCEL
734 irq_cpu_rmap_add(dev->rmap,
735 dev->priv.msix_arr[i + MLX5_EQ_VEC_COMP_BASE].vector);
737 snprintf(name, MLX5_MAX_IRQ_NAME, "mlx5_comp%d", i);
738 err = mlx5_create_map_eq(dev, eq,
739 i + MLX5_EQ_VEC_COMP_BASE, nent, 0,
740 name, &dev->priv.uuari.uars[0]);
745 mlx5_core_dbg(dev, "allocated completion EQN %d\n", eq->eqn);
747 spin_lock(&table->lock);
748 list_add_tail(&eq->list, &table->comp_eqs_list);
749 spin_unlock(&table->lock);
759 static int mlx5_core_set_issi(struct mlx5_core_dev *dev)
761 u32 query_in[MLX5_ST_SZ_DW(query_issi_in)];
762 u32 query_out[MLX5_ST_SZ_DW(query_issi_out)];
763 u32 set_in[MLX5_ST_SZ_DW(set_issi_in)];
764 u32 set_out[MLX5_ST_SZ_DW(set_issi_out)];
768 memset(query_in, 0, sizeof(query_in));
769 memset(query_out, 0, sizeof(query_out));
771 MLX5_SET(query_issi_in, query_in, opcode, MLX5_CMD_OP_QUERY_ISSI);
773 err = mlx5_cmd_exec_check_status(dev, query_in, sizeof(query_in),
774 query_out, sizeof(query_out));
776 if (((struct mlx5_outbox_hdr *)query_out)->status ==
777 MLX5_CMD_STAT_BAD_OP_ERR) {
778 pr_debug("Only ISSI 0 is supported\n");
782 pr_err("failed to query ISSI\n");
786 sup_issi = MLX5_GET(query_issi_out, query_out, supported_issi_dw0);
788 if (sup_issi & (1 << 1)) {
789 memset(set_in, 0, sizeof(set_in));
790 memset(set_out, 0, sizeof(set_out));
792 MLX5_SET(set_issi_in, set_in, opcode, MLX5_CMD_OP_SET_ISSI);
793 MLX5_SET(set_issi_in, set_in, current_issi, 1);
795 err = mlx5_cmd_exec_check_status(dev, set_in, sizeof(set_in),
796 set_out, sizeof(set_out));
798 pr_err("failed to set ISSI=1\n");
805 } else if (sup_issi & (1 << 0) || !sup_issi) {
812 static void mlx5_add_device(struct mlx5_interface *intf, struct mlx5_priv *priv)
814 struct mlx5_device_context *dev_ctx;
815 struct mlx5_core_dev *dev = container_of(priv, struct mlx5_core_dev, priv);
817 dev_ctx = kmalloc(sizeof(*dev_ctx), GFP_KERNEL);
821 dev_ctx->intf = intf;
822 dev_ctx->context = intf->add(dev);
824 if (dev_ctx->context) {
825 spin_lock_irq(&priv->ctx_lock);
826 list_add_tail(&dev_ctx->list, &priv->ctx_list);
827 spin_unlock_irq(&priv->ctx_lock);
833 static void mlx5_remove_device(struct mlx5_interface *intf, struct mlx5_priv *priv)
835 struct mlx5_device_context *dev_ctx;
836 struct mlx5_core_dev *dev = container_of(priv, struct mlx5_core_dev, priv);
838 list_for_each_entry(dev_ctx, &priv->ctx_list, list)
839 if (dev_ctx->intf == intf) {
840 spin_lock_irq(&priv->ctx_lock);
841 list_del(&dev_ctx->list);
842 spin_unlock_irq(&priv->ctx_lock);
844 intf->remove(dev, dev_ctx->context);
850 static int mlx5_register_device(struct mlx5_core_dev *dev)
852 struct mlx5_priv *priv = &dev->priv;
853 struct mlx5_interface *intf;
855 mutex_lock(&intf_mutex);
856 list_add_tail(&priv->dev_list, &dev_list);
857 list_for_each_entry(intf, &intf_list, list)
858 mlx5_add_device(intf, priv);
859 mutex_unlock(&intf_mutex);
864 static void mlx5_unregister_device(struct mlx5_core_dev *dev)
866 struct mlx5_priv *priv = &dev->priv;
867 struct mlx5_interface *intf;
869 mutex_lock(&intf_mutex);
870 list_for_each_entry(intf, &intf_list, list)
871 mlx5_remove_device(intf, priv);
872 list_del(&priv->dev_list);
873 mutex_unlock(&intf_mutex);
876 int mlx5_register_interface(struct mlx5_interface *intf)
878 struct mlx5_priv *priv;
880 if (!intf->add || !intf->remove)
883 mutex_lock(&intf_mutex);
884 list_add_tail(&intf->list, &intf_list);
885 list_for_each_entry(priv, &dev_list, dev_list)
886 mlx5_add_device(intf, priv);
887 mutex_unlock(&intf_mutex);
891 EXPORT_SYMBOL(mlx5_register_interface);
893 void mlx5_unregister_interface(struct mlx5_interface *intf)
895 struct mlx5_priv *priv;
897 mutex_lock(&intf_mutex);
898 list_for_each_entry(priv, &dev_list, dev_list)
899 mlx5_remove_device(intf, priv);
900 list_del(&intf->list);
901 mutex_unlock(&intf_mutex);
903 EXPORT_SYMBOL(mlx5_unregister_interface);
905 void *mlx5_get_protocol_dev(struct mlx5_core_dev *mdev, int protocol)
907 struct mlx5_priv *priv = &mdev->priv;
908 struct mlx5_device_context *dev_ctx;
912 spin_lock_irqsave(&priv->ctx_lock, flags);
914 list_for_each_entry(dev_ctx, &mdev->priv.ctx_list, list)
915 if ((dev_ctx->intf->protocol == protocol) &&
916 dev_ctx->intf->get_dev) {
917 result = dev_ctx->intf->get_dev(dev_ctx->context);
921 spin_unlock_irqrestore(&priv->ctx_lock, flags);
925 EXPORT_SYMBOL(mlx5_get_protocol_dev);
927 static int mlx5_pci_init(struct mlx5_core_dev *dev, struct mlx5_priv *priv)
929 struct pci_dev *pdev = dev->pdev;
932 pci_set_drvdata(dev->pdev, dev);
933 strncpy(priv->name, dev_name(&pdev->dev), MLX5_MAX_NAME_LEN);
934 priv->name[MLX5_MAX_NAME_LEN - 1] = 0;
936 mutex_init(&priv->pgdir_mutex);
937 INIT_LIST_HEAD(&priv->pgdir_list);
938 spin_lock_init(&priv->mkey_lock);
940 mutex_init(&priv->alloc_mutex);
942 priv->numa_node = dev_to_node(&dev->pdev->dev);
944 priv->dbg_root = debugfs_create_dir(dev_name(&pdev->dev), mlx5_debugfs_root);
948 err = mlx5_pci_enable_device(dev);
950 dev_err(&pdev->dev, "Cannot enable PCI device, aborting\n");
954 err = request_bar(pdev);
956 dev_err(&pdev->dev, "error requesting BARs, aborting\n");
960 pci_set_master(pdev);
962 err = set_dma_caps(pdev);
964 dev_err(&pdev->dev, "Failed setting DMA capabilities mask, aborting\n");
968 dev->iseg_base = pci_resource_start(dev->pdev, 0);
969 dev->iseg = ioremap(dev->iseg_base, sizeof(*dev->iseg));
972 dev_err(&pdev->dev, "Failed mapping initialization segment, aborting\n");
979 pci_clear_master(dev->pdev);
980 release_bar(dev->pdev);
982 mlx5_pci_disable_device(dev);
985 debugfs_remove(priv->dbg_root);
989 static void mlx5_pci_close(struct mlx5_core_dev *dev, struct mlx5_priv *priv)
992 pci_clear_master(dev->pdev);
993 release_bar(dev->pdev);
994 mlx5_pci_disable_device(dev);
995 debugfs_remove(priv->dbg_root);
998 #define MLX5_IB_MOD "mlx5_ib"
999 static int mlx5_load_one(struct mlx5_core_dev *dev, struct mlx5_priv *priv)
1001 struct pci_dev *pdev = dev->pdev;
1004 mutex_lock(&dev->intf_state_mutex);
1005 if (test_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state)) {
1006 dev_warn(&dev->pdev->dev, "%s: interface is up, NOP\n",
1011 dev_info(&pdev->dev, "firmware version: %d.%d.%d\n", fw_rev_maj(dev),
1012 fw_rev_min(dev), fw_rev_sub(dev));
1014 /* on load removing any previous indication of internal error, device is
1017 dev->state = MLX5_DEVICE_STATE_UP;
1019 err = mlx5_cmd_init(dev);
1021 dev_err(&pdev->dev, "Failed initializing command interface, aborting\n");
1025 err = wait_fw_init(dev, FW_INIT_TIMEOUT_MILI);
1027 dev_err(&dev->pdev->dev, "Firmware over %d MS in initializing state, aborting\n",
1028 FW_INIT_TIMEOUT_MILI);
1032 mlx5_pagealloc_init(dev);
1034 err = mlx5_core_enable_hca(dev, 0);
1036 dev_err(&pdev->dev, "enable hca failed\n");
1037 goto err_pagealloc_cleanup;
1040 err = mlx5_core_set_issi(dev);
1042 dev_err(&pdev->dev, "failed to set issi\n");
1043 goto err_disable_hca;
1046 err = mlx5_satisfy_startup_pages(dev, 1);
1048 dev_err(&pdev->dev, "failed to allocate boot pages\n");
1049 goto err_disable_hca;
1052 err = set_hca_ctrl(dev);
1054 dev_err(&pdev->dev, "set_hca_ctrl failed\n");
1055 goto reclaim_boot_pages;
1058 err = handle_hca_cap(dev);
1060 dev_err(&pdev->dev, "handle_hca_cap failed\n");
1061 goto reclaim_boot_pages;
1064 err = handle_hca_cap_atomic(dev);
1066 dev_err(&pdev->dev, "handle_hca_cap_atomic failed\n");
1067 goto reclaim_boot_pages;
1070 err = mlx5_satisfy_startup_pages(dev, 0);
1072 dev_err(&pdev->dev, "failed to allocate init pages\n");
1073 goto reclaim_boot_pages;
1076 err = mlx5_pagealloc_start(dev);
1078 dev_err(&pdev->dev, "mlx5_pagealloc_start failed\n");
1079 goto reclaim_boot_pages;
1082 err = mlx5_cmd_init_hca(dev);
1084 dev_err(&pdev->dev, "init hca failed\n");
1085 goto err_pagealloc_stop;
1088 mlx5_start_health_poll(dev);
1090 err = mlx5_query_hca_caps(dev);
1092 dev_err(&pdev->dev, "query hca failed\n");
1096 err = mlx5_query_board_id(dev);
1098 dev_err(&pdev->dev, "query board id failed\n");
1102 err = mlx5_enable_msix(dev);
1104 dev_err(&pdev->dev, "enable msix failed\n");
1108 err = mlx5_eq_init(dev);
1110 dev_err(&pdev->dev, "failed to initialize eq\n");
1114 err = mlx5_alloc_uuars(dev, &priv->uuari);
1116 dev_err(&pdev->dev, "Failed allocating uar, aborting\n");
1117 goto err_eq_cleanup;
1120 err = mlx5_start_eqs(dev);
1122 dev_err(&pdev->dev, "Failed to start pages and async EQs\n");
1126 err = alloc_comp_eqs(dev);
1128 dev_err(&pdev->dev, "Failed to alloc completion EQs\n");
1132 err = mlx5_irq_set_affinity_hints(dev);
1134 dev_err(&pdev->dev, "Failed to alloc affinity hint cpumask\n");
1136 MLX5_INIT_DOORBELL_LOCK(&priv->cq_uar_lock);
1138 mlx5_init_cq_table(dev);
1139 mlx5_init_qp_table(dev);
1140 mlx5_init_srq_table(dev);
1141 mlx5_init_mkey_table(dev);
1143 err = mlx5_init_fs(dev);
1145 dev_err(&pdev->dev, "Failed to init flow steering\n");
1149 err = mlx5_init_rl_table(dev);
1151 dev_err(&pdev->dev, "Failed to init rate limiting\n");
1155 #ifdef CONFIG_MLX5_CORE_EN
1156 err = mlx5_eswitch_init(dev);
1158 dev_err(&pdev->dev, "eswitch init failed %d\n", err);
1163 err = mlx5_sriov_init(dev);
1165 dev_err(&pdev->dev, "sriov init failed %d\n", err);
1169 err = mlx5_register_device(dev);
1171 dev_err(&pdev->dev, "mlx5_register_device failed %d\n", err);
1175 err = request_module_nowait(MLX5_IB_MOD);
1177 pr_info("failed request module on %s\n", MLX5_IB_MOD);
1179 clear_bit(MLX5_INTERFACE_STATE_DOWN, &dev->intf_state);
1180 set_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state);
1182 mutex_unlock(&dev->intf_state_mutex);
1187 if (mlx5_sriov_cleanup(dev))
1188 dev_err(&dev->pdev->dev, "sriov cleanup failed\n");
1190 #ifdef CONFIG_MLX5_CORE_EN
1191 mlx5_eswitch_cleanup(dev->priv.eswitch);
1194 mlx5_cleanup_rl_table(dev);
1196 mlx5_cleanup_fs(dev);
1198 mlx5_cleanup_mkey_table(dev);
1199 mlx5_cleanup_srq_table(dev);
1200 mlx5_cleanup_qp_table(dev);
1201 mlx5_cleanup_cq_table(dev);
1202 mlx5_irq_clear_affinity_hints(dev);
1209 mlx5_free_uuars(dev, &priv->uuari);
1212 mlx5_eq_cleanup(dev);
1215 mlx5_disable_msix(dev);
1218 mlx5_stop_health_poll(dev);
1219 if (mlx5_cmd_teardown_hca(dev)) {
1220 dev_err(&dev->pdev->dev, "tear_down_hca failed, skip cleanup\n");
1225 mlx5_pagealloc_stop(dev);
1228 mlx5_reclaim_startup_pages(dev);
1231 mlx5_core_disable_hca(dev, 0);
1233 err_pagealloc_cleanup:
1234 mlx5_pagealloc_cleanup(dev);
1235 mlx5_cmd_cleanup(dev);
1238 dev->state = MLX5_DEVICE_STATE_INTERNAL_ERROR;
1239 mutex_unlock(&dev->intf_state_mutex);
1244 static int mlx5_unload_one(struct mlx5_core_dev *dev, struct mlx5_priv *priv)
1248 err = mlx5_sriov_cleanup(dev);
1250 dev_warn(&dev->pdev->dev, "%s: sriov cleanup failed - abort\n",
1255 mutex_lock(&dev->intf_state_mutex);
1256 if (test_bit(MLX5_INTERFACE_STATE_DOWN, &dev->intf_state)) {
1257 dev_warn(&dev->pdev->dev, "%s: interface is down, NOP\n",
1261 mlx5_unregister_device(dev);
1262 #ifdef CONFIG_MLX5_CORE_EN
1263 mlx5_eswitch_cleanup(dev->priv.eswitch);
1266 mlx5_cleanup_rl_table(dev);
1267 mlx5_cleanup_fs(dev);
1268 mlx5_cleanup_mkey_table(dev);
1269 mlx5_cleanup_srq_table(dev);
1270 mlx5_cleanup_qp_table(dev);
1271 mlx5_cleanup_cq_table(dev);
1272 mlx5_irq_clear_affinity_hints(dev);
1275 mlx5_free_uuars(dev, &priv->uuari);
1276 mlx5_eq_cleanup(dev);
1277 mlx5_disable_msix(dev);
1278 mlx5_stop_health_poll(dev);
1279 err = mlx5_cmd_teardown_hca(dev);
1281 dev_err(&dev->pdev->dev, "tear_down_hca failed, skip cleanup\n");
1284 mlx5_pagealloc_stop(dev);
1285 mlx5_reclaim_startup_pages(dev);
1286 mlx5_core_disable_hca(dev, 0);
1287 mlx5_pagealloc_cleanup(dev);
1288 mlx5_cmd_cleanup(dev);
1291 clear_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state);
1292 set_bit(MLX5_INTERFACE_STATE_DOWN, &dev->intf_state);
1293 mutex_unlock(&dev->intf_state_mutex);
1297 void mlx5_core_event(struct mlx5_core_dev *dev, enum mlx5_dev_event event,
1298 unsigned long param)
1300 struct mlx5_priv *priv = &dev->priv;
1301 struct mlx5_device_context *dev_ctx;
1302 unsigned long flags;
1304 spin_lock_irqsave(&priv->ctx_lock, flags);
1306 list_for_each_entry(dev_ctx, &priv->ctx_list, list)
1307 if (dev_ctx->intf->event)
1308 dev_ctx->intf->event(dev, dev_ctx->context, event, param);
1310 spin_unlock_irqrestore(&priv->ctx_lock, flags);
1313 struct mlx5_core_event_handler {
1314 void (*event)(struct mlx5_core_dev *dev,
1315 enum mlx5_dev_event event,
1319 static const struct devlink_ops mlx5_devlink_ops = {
1320 #ifdef CONFIG_MLX5_CORE_EN
1321 .eswitch_mode_set = mlx5_devlink_eswitch_mode_set,
1322 .eswitch_mode_get = mlx5_devlink_eswitch_mode_get,
1326 static int init_one(struct pci_dev *pdev,
1327 const struct pci_device_id *id)
1329 struct mlx5_core_dev *dev;
1330 struct devlink *devlink;
1331 struct mlx5_priv *priv;
1334 devlink = devlink_alloc(&mlx5_devlink_ops, sizeof(*dev));
1336 dev_err(&pdev->dev, "kzalloc failed\n");
1340 dev = devlink_priv(devlink);
1342 priv->pci_dev_data = id->driver_data;
1344 pci_set_drvdata(pdev, dev);
1346 if (prof_sel < 0 || prof_sel >= ARRAY_SIZE(profile)) {
1347 pr_warn("selected profile out of range, selecting default (%d)\n",
1349 prof_sel = MLX5_DEFAULT_PROF;
1351 dev->profile = &profile[prof_sel];
1353 dev->event = mlx5_core_event;
1355 INIT_LIST_HEAD(&priv->ctx_list);
1356 spin_lock_init(&priv->ctx_lock);
1357 mutex_init(&dev->pci_status_mutex);
1358 mutex_init(&dev->intf_state_mutex);
1359 err = mlx5_pci_init(dev, priv);
1361 dev_err(&pdev->dev, "mlx5_pci_init failed with error code %d\n", err);
1365 err = mlx5_health_init(dev);
1367 dev_err(&pdev->dev, "mlx5_health_init failed with error code %d\n", err);
1371 err = mlx5_load_one(dev, priv);
1373 dev_err(&pdev->dev, "mlx5_load_one failed with error code %d\n", err);
1377 err = devlink_register(devlink, &pdev->dev);
1384 mlx5_unload_one(dev, priv);
1386 mlx5_health_cleanup(dev);
1388 mlx5_pci_close(dev, priv);
1390 pci_set_drvdata(pdev, NULL);
1391 devlink_free(devlink);
1396 static void remove_one(struct pci_dev *pdev)
1398 struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
1399 struct devlink *devlink = priv_to_devlink(dev);
1400 struct mlx5_priv *priv = &dev->priv;
1402 devlink_unregister(devlink);
1403 if (mlx5_unload_one(dev, priv)) {
1404 dev_err(&dev->pdev->dev, "mlx5_unload_one failed\n");
1405 mlx5_health_cleanup(dev);
1408 mlx5_health_cleanup(dev);
1409 mlx5_pci_close(dev, priv);
1410 pci_set_drvdata(pdev, NULL);
1411 devlink_free(devlink);
1414 static pci_ers_result_t mlx5_pci_err_detected(struct pci_dev *pdev,
1415 pci_channel_state_t state)
1417 struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
1418 struct mlx5_priv *priv = &dev->priv;
1420 dev_info(&pdev->dev, "%s was called\n", __func__);
1421 mlx5_enter_error_state(dev);
1422 mlx5_unload_one(dev, priv);
1423 mlx5_pci_disable_device(dev);
1424 return state == pci_channel_io_perm_failure ?
1425 PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_NEED_RESET;
1428 static pci_ers_result_t mlx5_pci_slot_reset(struct pci_dev *pdev)
1430 struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
1433 dev_info(&pdev->dev, "%s was called\n", __func__);
1435 err = mlx5_pci_enable_device(dev);
1437 dev_err(&pdev->dev, "%s: mlx5_pci_enable_device failed with error code: %d\n"
1439 return PCI_ERS_RESULT_DISCONNECT;
1441 pci_set_master(pdev);
1442 pci_set_power_state(pdev, PCI_D0);
1443 pci_restore_state(pdev);
1445 return err ? PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_RECOVERED;
1448 void mlx5_disable_device(struct mlx5_core_dev *dev)
1450 mlx5_pci_err_detected(dev->pdev, 0);
1453 /* wait for the device to show vital signs by waiting
1454 * for the health counter to start counting.
1456 static int wait_vital(struct pci_dev *pdev)
1458 struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
1459 struct mlx5_core_health *health = &dev->priv.health;
1460 const int niter = 100;
1465 for (i = 0; i < niter; i++) {
1466 count = ioread32be(health->health_counter);
1467 if (count && count != 0xffffffff) {
1468 if (last_count && last_count != count) {
1469 dev_info(&pdev->dev, "Counter value 0x%x after %d iterations\n", count, i);
1480 static void mlx5_pci_resume(struct pci_dev *pdev)
1482 struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
1483 struct mlx5_priv *priv = &dev->priv;
1486 dev_info(&pdev->dev, "%s was called\n", __func__);
1488 pci_save_state(pdev);
1489 err = wait_vital(pdev);
1491 dev_err(&pdev->dev, "%s: wait_vital timed out\n", __func__);
1495 err = mlx5_load_one(dev, priv);
1497 dev_err(&pdev->dev, "%s: mlx5_load_one failed with error code: %d\n"
1500 dev_info(&pdev->dev, "%s: device recovered\n", __func__);
1503 static const struct pci_error_handlers mlx5_err_handler = {
1504 .error_detected = mlx5_pci_err_detected,
1505 .slot_reset = mlx5_pci_slot_reset,
1506 .resume = mlx5_pci_resume
1509 static void shutdown(struct pci_dev *pdev)
1511 struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
1512 struct mlx5_priv *priv = &dev->priv;
1514 dev_info(&pdev->dev, "Shutdown was called\n");
1515 /* Notify mlx5 clients that the kernel is being shut down */
1516 set_bit(MLX5_INTERFACE_STATE_SHUTDOWN, &dev->intf_state);
1517 mlx5_unload_one(dev, priv);
1518 mlx5_pci_disable_device(dev);
1521 static const struct pci_device_id mlx5_core_pci_table[] = {
1522 { PCI_VDEVICE(MELLANOX, 0x1011) }, /* Connect-IB */
1523 { PCI_VDEVICE(MELLANOX, 0x1012), MLX5_PCI_DEV_IS_VF}, /* Connect-IB VF */
1524 { PCI_VDEVICE(MELLANOX, 0x1013) }, /* ConnectX-4 */
1525 { PCI_VDEVICE(MELLANOX, 0x1014), MLX5_PCI_DEV_IS_VF}, /* ConnectX-4 VF */
1526 { PCI_VDEVICE(MELLANOX, 0x1015) }, /* ConnectX-4LX */
1527 { PCI_VDEVICE(MELLANOX, 0x1016), MLX5_PCI_DEV_IS_VF}, /* ConnectX-4LX VF */
1528 { PCI_VDEVICE(MELLANOX, 0x1017) }, /* ConnectX-5, PCIe 3.0 */
1529 { PCI_VDEVICE(MELLANOX, 0x1018), MLX5_PCI_DEV_IS_VF}, /* ConnectX-5 VF */
1530 { PCI_VDEVICE(MELLANOX, 0x1019) }, /* ConnectX-5, PCIe 4.0 */
1534 MODULE_DEVICE_TABLE(pci, mlx5_core_pci_table);
1536 static struct pci_driver mlx5_core_driver = {
1537 .name = DRIVER_NAME,
1538 .id_table = mlx5_core_pci_table,
1540 .remove = remove_one,
1541 .shutdown = shutdown,
1542 .err_handler = &mlx5_err_handler,
1543 .sriov_configure = mlx5_core_sriov_configure,
1546 static int __init init(void)
1550 mlx5_register_debugfs();
1552 err = pci_register_driver(&mlx5_core_driver);
1556 #ifdef CONFIG_MLX5_CORE_EN
1563 mlx5_unregister_debugfs();
1567 static void __exit cleanup(void)
1569 #ifdef CONFIG_MLX5_CORE_EN
1572 pci_unregister_driver(&mlx5_core_driver);
1573 mlx5_unregister_debugfs();
1577 module_exit(cleanup);