2 * Copyright (c) 2004, 2005 Topspin Communications. All rights reserved.
3 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
4 * Copyright (c) 2005, 2006, 2007, 2008 Mellanox Technologies. All rights reserved.
5 * Copyright (c) 2006, 2007 Cisco Systems, Inc. All rights reserved.
7 * This software is available to you under a choice of one of two
8 * licenses. You may choose to be licensed under the terms of the GNU
9 * General Public License (GPL) Version 2, available from the file
10 * COPYING in the main directory of this source tree, or the
11 * OpenIB.org BSD license below:
13 * Redistribution and use in source and binary forms, with or
14 * without modification, are permitted provided that the following
17 * - Redistributions of source code must retain the above
18 * copyright notice, this list of conditions and the following
21 * - Redistributions in binary form must reproduce the above
22 * copyright notice, this list of conditions and the following
23 * disclaimer in the documentation and/or other materials
24 * provided with the distribution.
26 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
27 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
28 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
29 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
30 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
31 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
32 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
36 #include <linux/module.h>
37 #include <linux/kernel.h>
38 #include <linux/init.h>
39 #include <linux/errno.h>
40 #include <linux/pci.h>
41 #include <linux/dma-mapping.h>
42 #include <linux/slab.h>
43 #include <linux/io-mapping.h>
44 #include <linux/delay.h>
45 #include <linux/kmod.h>
46 #include <linux/etherdevice.h>
47 #include <net/devlink.h>
49 #include <uapi/rdma/mlx4-abi.h>
50 #include <linux/mlx4/device.h>
51 #include <linux/mlx4/doorbell.h>
57 MODULE_AUTHOR("Roland Dreier");
58 MODULE_DESCRIPTION("Mellanox ConnectX HCA low-level driver");
59 MODULE_LICENSE("Dual BSD/GPL");
60 MODULE_VERSION(DRV_VERSION);
62 struct workqueue_struct *mlx4_wq;
64 #ifdef CONFIG_MLX4_DEBUG
66 int mlx4_debug_level; /* 0 by default */
67 module_param_named(debug_level, mlx4_debug_level, int, 0644);
68 MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0");
70 #endif /* CONFIG_MLX4_DEBUG */
75 module_param(msi_x, int, 0444);
76 MODULE_PARM_DESC(msi_x, "0 - don't use MSI-X, 1 - use MSI-X, >1 - limit number of MSI-X irqs to msi_x");
78 #else /* CONFIG_PCI_MSI */
82 #endif /* CONFIG_PCI_MSI */
84 static uint8_t num_vfs[3] = {0, 0, 0};
85 static int num_vfs_argc;
86 module_param_array(num_vfs, byte, &num_vfs_argc, 0444);
87 MODULE_PARM_DESC(num_vfs, "enable #num_vfs functions if num_vfs > 0\n"
88 "num_vfs=port1,port2,port1+2");
90 static uint8_t probe_vf[3] = {0, 0, 0};
91 static int probe_vfs_argc;
92 module_param_array(probe_vf, byte, &probe_vfs_argc, 0444);
93 MODULE_PARM_DESC(probe_vf, "number of vfs to probe by pf driver (num_vfs > 0)\n"
94 "probe_vf=port1,port2,port1+2");
96 static int mlx4_log_num_mgm_entry_size = MLX4_DEFAULT_MGM_LOG_ENTRY_SIZE;
97 module_param_named(log_num_mgm_entry_size,
98 mlx4_log_num_mgm_entry_size, int, 0444);
99 MODULE_PARM_DESC(log_num_mgm_entry_size, "log mgm size, that defines the num"
100 " of qp per mcg, for example:"
101 " 10 gives 248.range: 7 <="
102 " log_num_mgm_entry_size <= 12."
103 " To activate device managed"
104 " flow steering when available, set to -1");
106 static bool enable_64b_cqe_eqe = true;
107 module_param(enable_64b_cqe_eqe, bool, 0444);
108 MODULE_PARM_DESC(enable_64b_cqe_eqe,
109 "Enable 64 byte CQEs/EQEs when the FW supports this (default: True)");
111 static bool enable_4k_uar;
112 module_param(enable_4k_uar, bool, 0444);
113 MODULE_PARM_DESC(enable_4k_uar,
114 "Enable using 4K UAR. Should not be enabled if have VFs which do not support 4K UARs (default: false)");
116 #define PF_CONTEXT_BEHAVIOUR_MASK (MLX4_FUNC_CAP_64B_EQE_CQE | \
117 MLX4_FUNC_CAP_EQE_CQE_STRIDE | \
118 MLX4_FUNC_CAP_DMFS_A0_STATIC)
120 #define RESET_PERSIST_MASK_FLAGS (MLX4_FLAG_SRIOV)
122 static char mlx4_version[] =
123 DRV_NAME ": Mellanox ConnectX core driver v"
126 static const struct mlx4_profile default_profile = {
129 .rdmarc_per_qp = 1 << 4,
133 .num_mtt = 1 << 20, /* It is really num mtt segements */
136 static const struct mlx4_profile low_mem_profile = {
139 .rdmarc_per_qp = 1 << 4,
146 static int log_num_mac = 7;
147 module_param_named(log_num_mac, log_num_mac, int, 0444);
148 MODULE_PARM_DESC(log_num_mac, "Log2 max number of MACs per ETH port (1-7)");
150 static int log_num_vlan;
151 module_param_named(log_num_vlan, log_num_vlan, int, 0444);
152 MODULE_PARM_DESC(log_num_vlan, "Log2 max number of VLANs per ETH port (0-7)");
153 /* Log2 max number of VLANs per ETH port (0-7) */
154 #define MLX4_LOG_NUM_VLANS 7
155 #define MLX4_MIN_LOG_NUM_VLANS 0
156 #define MLX4_MIN_LOG_NUM_MAC 1
158 static bool use_prio;
159 module_param_named(use_prio, use_prio, bool, 0444);
160 MODULE_PARM_DESC(use_prio, "Enable steering by VLAN priority on ETH ports (deprecated)");
162 int log_mtts_per_seg = ilog2(1);
163 module_param_named(log_mtts_per_seg, log_mtts_per_seg, int, 0444);
164 MODULE_PARM_DESC(log_mtts_per_seg, "Log2 number of MTT entries per segment "
165 "(0-7) (default: 0)");
167 static int port_type_array[2] = {MLX4_PORT_TYPE_NONE, MLX4_PORT_TYPE_NONE};
168 static int arr_argc = 2;
169 module_param_array(port_type_array, int, &arr_argc, 0444);
170 MODULE_PARM_DESC(port_type_array, "Array of port types: HW_DEFAULT (0) is default "
171 "1 for IB, 2 for Ethernet");
173 struct mlx4_port_config {
174 struct list_head list;
175 enum mlx4_port_type port_type[MLX4_MAX_PORTS + 1];
176 struct pci_dev *pdev;
179 static atomic_t pf_loading = ATOMIC_INIT(0);
181 static int mlx4_devlink_ierr_reset_get(struct devlink *devlink, u32 id,
182 struct devlink_param_gset_ctx *ctx)
184 ctx->val.vbool = !!mlx4_internal_err_reset;
188 static int mlx4_devlink_ierr_reset_set(struct devlink *devlink, u32 id,
189 struct devlink_param_gset_ctx *ctx)
191 mlx4_internal_err_reset = ctx->val.vbool;
195 static int mlx4_devlink_crdump_snapshot_get(struct devlink *devlink, u32 id,
196 struct devlink_param_gset_ctx *ctx)
198 struct mlx4_priv *priv = devlink_priv(devlink);
199 struct mlx4_dev *dev = &priv->dev;
201 ctx->val.vbool = dev->persist->crdump.snapshot_enable;
205 static int mlx4_devlink_crdump_snapshot_set(struct devlink *devlink, u32 id,
206 struct devlink_param_gset_ctx *ctx)
208 struct mlx4_priv *priv = devlink_priv(devlink);
209 struct mlx4_dev *dev = &priv->dev;
211 dev->persist->crdump.snapshot_enable = ctx->val.vbool;
216 mlx4_devlink_max_macs_validate(struct devlink *devlink, u32 id,
217 union devlink_param_value val,
218 struct netlink_ext_ack *extack)
220 u32 value = val.vu32;
222 if (value < 1 || value > 128)
225 if (!is_power_of_2(value)) {
226 NL_SET_ERR_MSG_MOD(extack, "max_macs supported must be power of 2");
233 enum mlx4_devlink_param_id {
234 MLX4_DEVLINK_PARAM_ID_BASE = DEVLINK_PARAM_GENERIC_ID_MAX,
235 MLX4_DEVLINK_PARAM_ID_ENABLE_64B_CQE_EQE,
236 MLX4_DEVLINK_PARAM_ID_ENABLE_4K_UAR,
239 static const struct devlink_param mlx4_devlink_params[] = {
240 DEVLINK_PARAM_GENERIC(INT_ERR_RESET,
241 BIT(DEVLINK_PARAM_CMODE_RUNTIME) |
242 BIT(DEVLINK_PARAM_CMODE_DRIVERINIT),
243 mlx4_devlink_ierr_reset_get,
244 mlx4_devlink_ierr_reset_set, NULL),
245 DEVLINK_PARAM_GENERIC(MAX_MACS,
246 BIT(DEVLINK_PARAM_CMODE_DRIVERINIT),
247 NULL, NULL, mlx4_devlink_max_macs_validate),
248 DEVLINK_PARAM_GENERIC(REGION_SNAPSHOT,
249 BIT(DEVLINK_PARAM_CMODE_RUNTIME) |
250 BIT(DEVLINK_PARAM_CMODE_DRIVERINIT),
251 mlx4_devlink_crdump_snapshot_get,
252 mlx4_devlink_crdump_snapshot_set, NULL),
253 DEVLINK_PARAM_DRIVER(MLX4_DEVLINK_PARAM_ID_ENABLE_64B_CQE_EQE,
254 "enable_64b_cqe_eqe", DEVLINK_PARAM_TYPE_BOOL,
255 BIT(DEVLINK_PARAM_CMODE_DRIVERINIT),
257 DEVLINK_PARAM_DRIVER(MLX4_DEVLINK_PARAM_ID_ENABLE_4K_UAR,
258 "enable_4k_uar", DEVLINK_PARAM_TYPE_BOOL,
259 BIT(DEVLINK_PARAM_CMODE_DRIVERINIT),
263 static void mlx4_devlink_set_params_init_values(struct devlink *devlink)
265 union devlink_param_value value;
267 value.vbool = !!mlx4_internal_err_reset;
268 devlink_param_driverinit_value_set(devlink,
269 DEVLINK_PARAM_GENERIC_ID_INT_ERR_RESET,
272 value.vu32 = 1UL << log_num_mac;
273 devlink_param_driverinit_value_set(devlink,
274 DEVLINK_PARAM_GENERIC_ID_MAX_MACS,
277 value.vbool = enable_64b_cqe_eqe;
278 devlink_param_driverinit_value_set(devlink,
279 MLX4_DEVLINK_PARAM_ID_ENABLE_64B_CQE_EQE,
282 value.vbool = enable_4k_uar;
283 devlink_param_driverinit_value_set(devlink,
284 MLX4_DEVLINK_PARAM_ID_ENABLE_4K_UAR,
288 devlink_param_driverinit_value_set(devlink,
289 DEVLINK_PARAM_GENERIC_ID_REGION_SNAPSHOT,
293 static inline void mlx4_set_num_reserved_uars(struct mlx4_dev *dev,
294 struct mlx4_dev_cap *dev_cap)
296 /* The reserved_uars is calculated by system page size unit.
297 * Therefore, adjustment is added when the uar page size is less
298 * than the system page size
300 dev->caps.reserved_uars =
302 mlx4_get_num_reserved_uar(dev),
303 dev_cap->reserved_uars /
304 (1 << (PAGE_SHIFT - dev->uar_page_shift)));
307 int mlx4_check_port_params(struct mlx4_dev *dev,
308 enum mlx4_port_type *port_type)
312 if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP)) {
313 for (i = 0; i < dev->caps.num_ports - 1; i++) {
314 if (port_type[i] != port_type[i + 1]) {
315 mlx4_err(dev, "Only same port types supported on this HCA, aborting\n");
321 for (i = 0; i < dev->caps.num_ports; i++) {
322 if (!(port_type[i] & dev->caps.supported_type[i+1])) {
323 mlx4_err(dev, "Requested port type for port %d is not supported on this HCA\n",
331 static void mlx4_set_port_mask(struct mlx4_dev *dev)
335 for (i = 1; i <= dev->caps.num_ports; ++i)
336 dev->caps.port_mask[i] = dev->caps.port_type[i];
340 MLX4_QUERY_FUNC_NUM_SYS_EQS = 1 << 0,
343 static int mlx4_query_func(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
346 struct mlx4_func func;
348 if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS) {
349 err = mlx4_QUERY_FUNC(dev, &func, 0);
351 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
354 dev_cap->max_eqs = func.max_eq;
355 dev_cap->reserved_eqs = func.rsvd_eqs;
356 dev_cap->reserved_uars = func.rsvd_uars;
357 err |= MLX4_QUERY_FUNC_NUM_SYS_EQS;
362 static void mlx4_enable_cqe_eqe_stride(struct mlx4_dev *dev)
364 struct mlx4_caps *dev_cap = &dev->caps;
366 /* FW not supporting or cancelled by user */
367 if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_EQE_STRIDE) ||
368 !(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_CQE_STRIDE))
371 /* Must have 64B CQE_EQE enabled by FW to use bigger stride
372 * When FW has NCSI it may decide not to report 64B CQE/EQEs
374 if (!(dev_cap->flags & MLX4_DEV_CAP_FLAG_64B_EQE) ||
375 !(dev_cap->flags & MLX4_DEV_CAP_FLAG_64B_CQE)) {
376 dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
377 dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
381 if (cache_line_size() == 128 || cache_line_size() == 256) {
382 mlx4_dbg(dev, "Enabling CQE stride cacheLine supported\n");
383 /* Changing the real data inside CQE size to 32B */
384 dev_cap->flags &= ~MLX4_DEV_CAP_FLAG_64B_CQE;
385 dev_cap->flags &= ~MLX4_DEV_CAP_FLAG_64B_EQE;
387 if (mlx4_is_master(dev))
388 dev_cap->function_caps |= MLX4_FUNC_CAP_EQE_CQE_STRIDE;
390 if (cache_line_size() != 32 && cache_line_size() != 64)
391 mlx4_dbg(dev, "Disabling CQE stride, cacheLine size unsupported\n");
392 dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
393 dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
397 static int _mlx4_dev_port(struct mlx4_dev *dev, int port,
398 struct mlx4_port_cap *port_cap)
400 dev->caps.vl_cap[port] = port_cap->max_vl;
401 dev->caps.ib_mtu_cap[port] = port_cap->ib_mtu;
402 dev->phys_caps.gid_phys_table_len[port] = port_cap->max_gids;
403 dev->phys_caps.pkey_phys_table_len[port] = port_cap->max_pkeys;
404 /* set gid and pkey table operating lengths by default
405 * to non-sriov values
407 dev->caps.gid_table_len[port] = port_cap->max_gids;
408 dev->caps.pkey_table_len[port] = port_cap->max_pkeys;
409 dev->caps.port_width_cap[port] = port_cap->max_port_width;
410 dev->caps.eth_mtu_cap[port] = port_cap->eth_mtu;
411 dev->caps.max_tc_eth = port_cap->max_tc_eth;
412 dev->caps.def_mac[port] = port_cap->def_mac;
413 dev->caps.supported_type[port] = port_cap->supported_port_types;
414 dev->caps.suggested_type[port] = port_cap->suggested_type;
415 dev->caps.default_sense[port] = port_cap->default_sense;
416 dev->caps.trans_type[port] = port_cap->trans_type;
417 dev->caps.vendor_oui[port] = port_cap->vendor_oui;
418 dev->caps.wavelength[port] = port_cap->wavelength;
419 dev->caps.trans_code[port] = port_cap->trans_code;
424 static int mlx4_dev_port(struct mlx4_dev *dev, int port,
425 struct mlx4_port_cap *port_cap)
429 err = mlx4_QUERY_PORT(dev, port, port_cap);
432 mlx4_err(dev, "QUERY_PORT command failed.\n");
437 static inline void mlx4_enable_ignore_fcs(struct mlx4_dev *dev)
439 if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_IGNORE_FCS))
442 if (mlx4_is_mfunc(dev)) {
443 mlx4_dbg(dev, "SRIOV mode - Disabling Ignore FCS");
444 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_IGNORE_FCS;
448 if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_FCS_KEEP)) {
450 "Keep FCS is not supported - Disabling Ignore FCS");
451 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_IGNORE_FCS;
456 #define MLX4_A0_STEERING_TABLE_SIZE 256
457 static int mlx4_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
462 err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
464 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
467 mlx4_dev_cap_dump(dev, dev_cap);
469 if (dev_cap->min_page_sz > PAGE_SIZE) {
470 mlx4_err(dev, "HCA minimum page size of %d bigger than kernel PAGE_SIZE of %ld, aborting\n",
471 dev_cap->min_page_sz, PAGE_SIZE);
474 if (dev_cap->num_ports > MLX4_MAX_PORTS) {
475 mlx4_err(dev, "HCA has %d ports, but we only support %d, aborting\n",
476 dev_cap->num_ports, MLX4_MAX_PORTS);
480 if (dev_cap->uar_size > pci_resource_len(dev->persist->pdev, 2)) {
481 mlx4_err(dev, "HCA reported UAR size of 0x%x bigger than PCI resource 2 size of 0x%llx, aborting\n",
484 pci_resource_len(dev->persist->pdev, 2));
488 dev->caps.num_ports = dev_cap->num_ports;
489 dev->caps.num_sys_eqs = dev_cap->num_sys_eqs;
490 dev->phys_caps.num_phys_eqs = dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS ?
491 dev->caps.num_sys_eqs :
493 for (i = 1; i <= dev->caps.num_ports; ++i) {
494 err = _mlx4_dev_port(dev, i, dev_cap->port_cap + i);
496 mlx4_err(dev, "QUERY_PORT command failed, aborting\n");
501 dev->caps.map_clock_to_user = dev_cap->map_clock_to_user;
502 dev->caps.uar_page_size = PAGE_SIZE;
503 dev->caps.num_uars = dev_cap->uar_size / PAGE_SIZE;
504 dev->caps.local_ca_ack_delay = dev_cap->local_ca_ack_delay;
505 dev->caps.bf_reg_size = dev_cap->bf_reg_size;
506 dev->caps.bf_regs_per_page = dev_cap->bf_regs_per_page;
507 dev->caps.max_sq_sg = dev_cap->max_sq_sg;
508 dev->caps.max_rq_sg = dev_cap->max_rq_sg;
509 dev->caps.max_wqes = dev_cap->max_qp_sz;
510 dev->caps.max_qp_init_rdma = dev_cap->max_requester_per_qp;
511 dev->caps.max_srq_wqes = dev_cap->max_srq_sz;
512 dev->caps.max_srq_sge = dev_cap->max_rq_sg - 1;
513 dev->caps.reserved_srqs = dev_cap->reserved_srqs;
514 dev->caps.max_sq_desc_sz = dev_cap->max_sq_desc_sz;
515 dev->caps.max_rq_desc_sz = dev_cap->max_rq_desc_sz;
517 * Subtract 1 from the limit because we need to allocate a
518 * spare CQE to enable resizing the CQ.
520 dev->caps.max_cqes = dev_cap->max_cq_sz - 1;
521 dev->caps.reserved_cqs = dev_cap->reserved_cqs;
522 dev->caps.reserved_eqs = dev_cap->reserved_eqs;
523 dev->caps.reserved_mtts = dev_cap->reserved_mtts;
524 dev->caps.reserved_mrws = dev_cap->reserved_mrws;
526 dev->caps.reserved_pds = dev_cap->reserved_pds;
527 dev->caps.reserved_xrcds = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
528 dev_cap->reserved_xrcds : 0;
529 dev->caps.max_xrcds = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
530 dev_cap->max_xrcds : 0;
531 dev->caps.mtt_entry_sz = dev_cap->mtt_entry_sz;
533 dev->caps.max_msg_sz = dev_cap->max_msg_sz;
534 dev->caps.page_size_cap = ~(u32) (dev_cap->min_page_sz - 1);
535 dev->caps.flags = dev_cap->flags;
536 dev->caps.flags2 = dev_cap->flags2;
537 dev->caps.bmme_flags = dev_cap->bmme_flags;
538 dev->caps.reserved_lkey = dev_cap->reserved_lkey;
539 dev->caps.stat_rate_support = dev_cap->stat_rate_support;
540 dev->caps.max_gso_sz = dev_cap->max_gso_sz;
541 dev->caps.max_rss_tbl_sz = dev_cap->max_rss_tbl_sz;
542 dev->caps.wol_port[1] = dev_cap->wol_port[1];
543 dev->caps.wol_port[2] = dev_cap->wol_port[2];
544 dev->caps.health_buffer_addrs = dev_cap->health_buffer_addrs;
546 /* Save uar page shift */
547 if (!mlx4_is_slave(dev)) {
548 /* Virtual PCI function needs to determine UAR page size from
549 * firmware. Only master PCI function can set the uar page size
551 if (enable_4k_uar || !dev->persist->num_vfs)
552 dev->uar_page_shift = DEFAULT_UAR_PAGE_SHIFT;
554 dev->uar_page_shift = PAGE_SHIFT;
556 mlx4_set_num_reserved_uars(dev, dev_cap);
559 if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_PHV_EN) {
560 struct mlx4_init_hca_param hca_param;
562 memset(&hca_param, 0, sizeof(hca_param));
563 err = mlx4_QUERY_HCA(dev, &hca_param);
564 /* Turn off PHV_EN flag in case phv_check_en is set.
565 * phv_check_en is a HW check that parse the packet and verify
566 * phv bit was reported correctly in the wqe. To allow QinQ
567 * PHV_EN flag should be set and phv_check_en must be cleared
568 * otherwise QinQ packets will be drop by the HW.
570 if (err || hca_param.phv_check_en)
571 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_PHV_EN;
574 /* Sense port always allowed on supported devices for ConnectX-1 and -2 */
575 if (mlx4_priv(dev)->pci_dev_data & MLX4_PCI_DEV_FORCE_SENSE_PORT)
576 dev->caps.flags |= MLX4_DEV_CAP_FLAG_SENSE_SUPPORT;
577 /* Don't do sense port on multifunction devices (for now at least) */
578 if (mlx4_is_mfunc(dev))
579 dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_SENSE_SUPPORT;
581 if (mlx4_low_memory_profile()) {
582 dev->caps.log_num_macs = MLX4_MIN_LOG_NUM_MAC;
583 dev->caps.log_num_vlans = MLX4_MIN_LOG_NUM_VLANS;
585 dev->caps.log_num_macs = log_num_mac;
586 dev->caps.log_num_vlans = MLX4_LOG_NUM_VLANS;
589 for (i = 1; i <= dev->caps.num_ports; ++i) {
590 dev->caps.port_type[i] = MLX4_PORT_TYPE_NONE;
591 if (dev->caps.supported_type[i]) {
592 /* if only ETH is supported - assign ETH */
593 if (dev->caps.supported_type[i] == MLX4_PORT_TYPE_ETH)
594 dev->caps.port_type[i] = MLX4_PORT_TYPE_ETH;
595 /* if only IB is supported, assign IB */
596 else if (dev->caps.supported_type[i] ==
598 dev->caps.port_type[i] = MLX4_PORT_TYPE_IB;
600 /* if IB and ETH are supported, we set the port
601 * type according to user selection of port type;
602 * if user selected none, take the FW hint */
603 if (port_type_array[i - 1] == MLX4_PORT_TYPE_NONE)
604 dev->caps.port_type[i] = dev->caps.suggested_type[i] ?
605 MLX4_PORT_TYPE_ETH : MLX4_PORT_TYPE_IB;
607 dev->caps.port_type[i] = port_type_array[i - 1];
611 * Link sensing is allowed on the port if 3 conditions are true:
612 * 1. Both protocols are supported on the port.
613 * 2. Different types are supported on the port
614 * 3. FW declared that it supports link sensing
616 mlx4_priv(dev)->sense.sense_allowed[i] =
617 ((dev->caps.supported_type[i] == MLX4_PORT_TYPE_AUTO) &&
618 (dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
619 (dev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT));
622 * If "default_sense" bit is set, we move the port to "AUTO" mode
623 * and perform sense_port FW command to try and set the correct
624 * port type from beginning
626 if (mlx4_priv(dev)->sense.sense_allowed[i] && dev->caps.default_sense[i]) {
627 enum mlx4_port_type sensed_port = MLX4_PORT_TYPE_NONE;
628 dev->caps.possible_type[i] = MLX4_PORT_TYPE_AUTO;
629 mlx4_SENSE_PORT(dev, i, &sensed_port);
630 if (sensed_port != MLX4_PORT_TYPE_NONE)
631 dev->caps.port_type[i] = sensed_port;
633 dev->caps.possible_type[i] = dev->caps.port_type[i];
636 if (dev->caps.log_num_macs > dev_cap->port_cap[i].log_max_macs) {
637 dev->caps.log_num_macs = dev_cap->port_cap[i].log_max_macs;
638 mlx4_warn(dev, "Requested number of MACs is too much for port %d, reducing to %d\n",
639 i, 1 << dev->caps.log_num_macs);
641 if (dev->caps.log_num_vlans > dev_cap->port_cap[i].log_max_vlans) {
642 dev->caps.log_num_vlans = dev_cap->port_cap[i].log_max_vlans;
643 mlx4_warn(dev, "Requested number of VLANs is too much for port %d, reducing to %d\n",
644 i, 1 << dev->caps.log_num_vlans);
648 if (mlx4_is_master(dev) && (dev->caps.num_ports == 2) &&
649 (port_type_array[0] == MLX4_PORT_TYPE_IB) &&
650 (port_type_array[1] == MLX4_PORT_TYPE_ETH)) {
652 "Granular QoS per VF not supported with IB/Eth configuration\n");
653 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_QOS_VPP;
656 dev->caps.max_counters = dev_cap->max_counters;
658 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] = dev_cap->reserved_qps;
659 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] =
660 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] =
661 (1 << dev->caps.log_num_macs) *
662 (1 << dev->caps.log_num_vlans) *
664 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH] = MLX4_NUM_FEXCH;
666 if (dev_cap->dmfs_high_rate_qpn_base > 0 &&
667 dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN)
668 dev->caps.dmfs_high_rate_qpn_base = dev_cap->dmfs_high_rate_qpn_base;
670 dev->caps.dmfs_high_rate_qpn_base =
671 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
673 if (dev_cap->dmfs_high_rate_qpn_range > 0 &&
674 dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN) {
675 dev->caps.dmfs_high_rate_qpn_range = dev_cap->dmfs_high_rate_qpn_range;
676 dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_DEFAULT;
677 dev->caps.flags2 |= MLX4_DEV_CAP_FLAG2_FS_A0;
679 dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_NOT_SUPPORTED;
680 dev->caps.dmfs_high_rate_qpn_base =
681 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
682 dev->caps.dmfs_high_rate_qpn_range = MLX4_A0_STEERING_TABLE_SIZE;
685 dev->caps.rl_caps = dev_cap->rl_caps;
687 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_RSS_RAW_ETH] =
688 dev->caps.dmfs_high_rate_qpn_range;
690 dev->caps.reserved_qps = dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] +
691 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] +
692 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] +
693 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH];
695 dev->caps.sqp_demux = (mlx4_is_master(dev)) ? MLX4_MAX_NUM_SLAVES : 0;
697 if (!enable_64b_cqe_eqe && !mlx4_is_slave(dev)) {
699 (MLX4_DEV_CAP_FLAG_64B_CQE | MLX4_DEV_CAP_FLAG_64B_EQE)) {
700 mlx4_warn(dev, "64B EQEs/CQEs supported by the device but not enabled\n");
701 dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_64B_CQE;
702 dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_64B_EQE;
705 if (dev_cap->flags2 &
706 (MLX4_DEV_CAP_FLAG2_CQE_STRIDE |
707 MLX4_DEV_CAP_FLAG2_EQE_STRIDE)) {
708 mlx4_warn(dev, "Disabling EQE/CQE stride per user request\n");
709 dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
710 dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
714 if ((dev->caps.flags &
715 (MLX4_DEV_CAP_FLAG_64B_CQE | MLX4_DEV_CAP_FLAG_64B_EQE)) &&
717 dev->caps.function_caps |= MLX4_FUNC_CAP_64B_EQE_CQE;
719 if (!mlx4_is_slave(dev)) {
720 mlx4_enable_cqe_eqe_stride(dev);
721 dev->caps.alloc_res_qp_mask =
722 (dev->caps.bf_reg_size ? MLX4_RESERVE_ETH_BF_QP : 0) |
725 if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ETS_CFG) &&
726 dev->caps.flags & MLX4_DEV_CAP_FLAG_SET_ETH_SCHED) {
727 mlx4_warn(dev, "Old device ETS support detected\n");
728 mlx4_warn(dev, "Consider upgrading device FW.\n");
729 dev->caps.flags2 |= MLX4_DEV_CAP_FLAG2_ETS_CFG;
733 dev->caps.alloc_res_qp_mask = 0;
736 mlx4_enable_ignore_fcs(dev);
741 /*The function checks if there are live vf, return the num of them*/
742 static int mlx4_how_many_lives_vf(struct mlx4_dev *dev)
744 struct mlx4_priv *priv = mlx4_priv(dev);
745 struct mlx4_slave_state *s_state;
749 for (i = 1/*the ppf is 0*/; i < dev->num_slaves; ++i) {
750 s_state = &priv->mfunc.master.slave_state[i];
751 if (s_state->active && s_state->last_cmd !=
752 MLX4_COMM_CMD_RESET) {
753 mlx4_warn(dev, "%s: slave: %d is still active\n",
761 int mlx4_get_parav_qkey(struct mlx4_dev *dev, u32 qpn, u32 *qkey)
763 u32 qk = MLX4_RESERVED_QKEY_BASE;
765 if (qpn >= dev->phys_caps.base_tunnel_sqpn + 8 * MLX4_MFUNC_MAX ||
766 qpn < dev->phys_caps.base_proxy_sqpn)
769 if (qpn >= dev->phys_caps.base_tunnel_sqpn)
771 qk += qpn - dev->phys_caps.base_tunnel_sqpn;
773 qk += qpn - dev->phys_caps.base_proxy_sqpn;
777 EXPORT_SYMBOL(mlx4_get_parav_qkey);
779 void mlx4_sync_pkey_table(struct mlx4_dev *dev, int slave, int port, int i, int val)
781 struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
783 if (!mlx4_is_master(dev))
786 priv->virt2phys_pkey[slave][port - 1][i] = val;
788 EXPORT_SYMBOL(mlx4_sync_pkey_table);
790 void mlx4_put_slave_node_guid(struct mlx4_dev *dev, int slave, __be64 guid)
792 struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
794 if (!mlx4_is_master(dev))
797 priv->slave_node_guids[slave] = guid;
799 EXPORT_SYMBOL(mlx4_put_slave_node_guid);
801 __be64 mlx4_get_slave_node_guid(struct mlx4_dev *dev, int slave)
803 struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
805 if (!mlx4_is_master(dev))
808 return priv->slave_node_guids[slave];
810 EXPORT_SYMBOL(mlx4_get_slave_node_guid);
812 int mlx4_is_slave_active(struct mlx4_dev *dev, int slave)
814 struct mlx4_priv *priv = mlx4_priv(dev);
815 struct mlx4_slave_state *s_slave;
817 if (!mlx4_is_master(dev))
820 s_slave = &priv->mfunc.master.slave_state[slave];
821 return !!s_slave->active;
823 EXPORT_SYMBOL(mlx4_is_slave_active);
825 void mlx4_handle_eth_header_mcast_prio(struct mlx4_net_trans_rule_hw_ctrl *ctrl,
826 struct _rule_hw *eth_header)
828 if (is_multicast_ether_addr(eth_header->eth.dst_mac) ||
829 is_broadcast_ether_addr(eth_header->eth.dst_mac)) {
830 struct mlx4_net_trans_rule_hw_eth *eth =
831 (struct mlx4_net_trans_rule_hw_eth *)eth_header;
832 struct _rule_hw *next_rule = (struct _rule_hw *)(eth + 1);
833 bool last_rule = next_rule->size == 0 && next_rule->id == 0 &&
834 next_rule->rsvd == 0;
837 ctrl->prio = cpu_to_be16(MLX4_DOMAIN_NIC);
840 EXPORT_SYMBOL(mlx4_handle_eth_header_mcast_prio);
842 static void slave_adjust_steering_mode(struct mlx4_dev *dev,
843 struct mlx4_dev_cap *dev_cap,
844 struct mlx4_init_hca_param *hca_param)
846 dev->caps.steering_mode = hca_param->steering_mode;
847 if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED) {
848 dev->caps.num_qp_per_mgm = dev_cap->fs_max_num_qp_per_entry;
849 dev->caps.fs_log_max_ucast_qp_range_size =
850 dev_cap->fs_log_max_ucast_qp_range_size;
852 dev->caps.num_qp_per_mgm =
853 4 * ((1 << hca_param->log_mc_entry_sz)/16 - 2);
855 mlx4_dbg(dev, "Steering mode is: %s\n",
856 mlx4_steering_mode_str(dev->caps.steering_mode));
859 static void mlx4_slave_destroy_special_qp_cap(struct mlx4_dev *dev)
861 kfree(dev->caps.spec_qps);
862 dev->caps.spec_qps = NULL;
865 static int mlx4_slave_special_qp_cap(struct mlx4_dev *dev)
867 struct mlx4_func_cap *func_cap = NULL;
868 struct mlx4_caps *caps = &dev->caps;
871 func_cap = kzalloc(sizeof(*func_cap), GFP_KERNEL);
872 caps->spec_qps = kcalloc(caps->num_ports, sizeof(*caps->spec_qps), GFP_KERNEL);
874 if (!func_cap || !caps->spec_qps) {
875 mlx4_err(dev, "Failed to allocate memory for special qps cap\n");
880 for (i = 1; i <= caps->num_ports; ++i) {
881 err = mlx4_QUERY_FUNC_CAP(dev, i, func_cap);
883 mlx4_err(dev, "QUERY_FUNC_CAP port command failed for port %d, aborting (%d)\n",
887 caps->spec_qps[i - 1] = func_cap->spec_qps;
888 caps->port_mask[i] = caps->port_type[i];
889 caps->phys_port_id[i] = func_cap->phys_port_id;
890 err = mlx4_get_slave_pkey_gid_tbl_len(dev, i,
891 &caps->gid_table_len[i],
892 &caps->pkey_table_len[i]);
894 mlx4_err(dev, "QUERY_PORT command failed for port %d, aborting (%d)\n",
902 mlx4_slave_destroy_special_qp_cap(dev);
907 static int mlx4_slave_cap(struct mlx4_dev *dev)
911 struct mlx4_dev_cap *dev_cap = NULL;
912 struct mlx4_func_cap *func_cap = NULL;
913 struct mlx4_init_hca_param *hca_param = NULL;
915 hca_param = kzalloc(sizeof(*hca_param), GFP_KERNEL);
916 func_cap = kzalloc(sizeof(*func_cap), GFP_KERNEL);
917 dev_cap = kzalloc(sizeof(*dev_cap), GFP_KERNEL);
918 if (!hca_param || !func_cap || !dev_cap) {
919 mlx4_err(dev, "Failed to allocate memory for slave_cap\n");
924 err = mlx4_QUERY_HCA(dev, hca_param);
926 mlx4_err(dev, "QUERY_HCA command failed, aborting\n");
930 /* fail if the hca has an unknown global capability
931 * at this time global_caps should be always zeroed
933 if (hca_param->global_caps) {
934 mlx4_err(dev, "Unknown hca global capabilities\n");
939 dev->caps.hca_core_clock = hca_param->hca_core_clock;
941 dev->caps.max_qp_dest_rdma = 1 << hca_param->log_rd_per_qp;
942 err = mlx4_dev_cap(dev, dev_cap);
944 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
948 err = mlx4_QUERY_FW(dev);
950 mlx4_err(dev, "QUERY_FW command failed: could not get FW version\n");
952 page_size = ~dev->caps.page_size_cap + 1;
953 mlx4_warn(dev, "HCA minimum page size:%d\n", page_size);
954 if (page_size > PAGE_SIZE) {
955 mlx4_err(dev, "HCA minimum page size of %d bigger than kernel PAGE_SIZE of %ld, aborting\n",
956 page_size, PAGE_SIZE);
961 /* Set uar_page_shift for VF */
962 dev->uar_page_shift = hca_param->uar_page_sz + 12;
964 /* Make sure the master uar page size is valid */
965 if (dev->uar_page_shift > PAGE_SHIFT) {
967 "Invalid configuration: uar page size is larger than system page size\n");
972 /* Set reserved_uars based on the uar_page_shift */
973 mlx4_set_num_reserved_uars(dev, dev_cap);
975 /* Although uar page size in FW differs from system page size,
976 * upper software layers (mlx4_ib, mlx4_en and part of mlx4_core)
977 * still works with assumption that uar page size == system page size
979 dev->caps.uar_page_size = PAGE_SIZE;
981 err = mlx4_QUERY_FUNC_CAP(dev, 0, func_cap);
983 mlx4_err(dev, "QUERY_FUNC_CAP general command failed, aborting (%d)\n",
988 if ((func_cap->pf_context_behaviour | PF_CONTEXT_BEHAVIOUR_MASK) !=
989 PF_CONTEXT_BEHAVIOUR_MASK) {
990 mlx4_err(dev, "Unknown pf context behaviour %x known flags %x\n",
991 func_cap->pf_context_behaviour,
992 PF_CONTEXT_BEHAVIOUR_MASK);
997 dev->caps.num_ports = func_cap->num_ports;
998 dev->quotas.qp = func_cap->qp_quota;
999 dev->quotas.srq = func_cap->srq_quota;
1000 dev->quotas.cq = func_cap->cq_quota;
1001 dev->quotas.mpt = func_cap->mpt_quota;
1002 dev->quotas.mtt = func_cap->mtt_quota;
1003 dev->caps.num_qps = 1 << hca_param->log_num_qps;
1004 dev->caps.num_srqs = 1 << hca_param->log_num_srqs;
1005 dev->caps.num_cqs = 1 << hca_param->log_num_cqs;
1006 dev->caps.num_mpts = 1 << hca_param->log_mpt_sz;
1007 dev->caps.num_eqs = func_cap->max_eq;
1008 dev->caps.reserved_eqs = func_cap->reserved_eq;
1009 dev->caps.reserved_lkey = func_cap->reserved_lkey;
1010 dev->caps.num_pds = MLX4_NUM_PDS;
1011 dev->caps.num_mgms = 0;
1012 dev->caps.num_amgms = 0;
1014 if (dev->caps.num_ports > MLX4_MAX_PORTS) {
1015 mlx4_err(dev, "HCA has %d ports, but we only support %d, aborting\n",
1016 dev->caps.num_ports, MLX4_MAX_PORTS);
1021 mlx4_replace_zero_macs(dev);
1023 err = mlx4_slave_special_qp_cap(dev);
1025 mlx4_err(dev, "Set special QP caps failed. aborting\n");
1029 if (dev->caps.uar_page_size * (dev->caps.num_uars -
1030 dev->caps.reserved_uars) >
1031 pci_resource_len(dev->persist->pdev,
1033 mlx4_err(dev, "HCA reported UAR region size of 0x%x bigger than PCI resource 2 size of 0x%llx, aborting\n",
1034 dev->caps.uar_page_size * dev->caps.num_uars,
1035 (unsigned long long)
1036 pci_resource_len(dev->persist->pdev, 2));
1041 if (hca_param->dev_cap_enabled & MLX4_DEV_CAP_64B_EQE_ENABLED) {
1042 dev->caps.eqe_size = 64;
1043 dev->caps.eqe_factor = 1;
1045 dev->caps.eqe_size = 32;
1046 dev->caps.eqe_factor = 0;
1049 if (hca_param->dev_cap_enabled & MLX4_DEV_CAP_64B_CQE_ENABLED) {
1050 dev->caps.cqe_size = 64;
1051 dev->caps.userspace_caps |= MLX4_USER_DEV_CAP_LARGE_CQE;
1053 dev->caps.cqe_size = 32;
1056 if (hca_param->dev_cap_enabled & MLX4_DEV_CAP_EQE_STRIDE_ENABLED) {
1057 dev->caps.eqe_size = hca_param->eqe_size;
1058 dev->caps.eqe_factor = 0;
1061 if (hca_param->dev_cap_enabled & MLX4_DEV_CAP_CQE_STRIDE_ENABLED) {
1062 dev->caps.cqe_size = hca_param->cqe_size;
1063 /* User still need to know when CQE > 32B */
1064 dev->caps.userspace_caps |= MLX4_USER_DEV_CAP_LARGE_CQE;
1067 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
1068 mlx4_warn(dev, "Timestamping is not supported in slave mode\n");
1070 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_USER_MAC_EN;
1071 mlx4_dbg(dev, "User MAC FW update is not supported in slave mode\n");
1073 slave_adjust_steering_mode(dev, dev_cap, hca_param);
1074 mlx4_dbg(dev, "RSS support for IP fragments is %s\n",
1075 hca_param->rss_ip_frags ? "on" : "off");
1077 if (func_cap->extra_flags & MLX4_QUERY_FUNC_FLAGS_BF_RES_QP &&
1078 dev->caps.bf_reg_size)
1079 dev->caps.alloc_res_qp_mask |= MLX4_RESERVE_ETH_BF_QP;
1081 if (func_cap->extra_flags & MLX4_QUERY_FUNC_FLAGS_A0_RES_QP)
1082 dev->caps.alloc_res_qp_mask |= MLX4_RESERVE_A0_QP;
1086 mlx4_slave_destroy_special_qp_cap(dev);
1094 static void mlx4_request_modules(struct mlx4_dev *dev)
1097 int has_ib_port = false;
1098 int has_eth_port = false;
1099 #define EN_DRV_NAME "mlx4_en"
1100 #define IB_DRV_NAME "mlx4_ib"
1102 for (port = 1; port <= dev->caps.num_ports; port++) {
1103 if (dev->caps.port_type[port] == MLX4_PORT_TYPE_IB)
1105 else if (dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH)
1106 has_eth_port = true;
1110 request_module_nowait(EN_DRV_NAME);
1111 if (has_ib_port || (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE))
1112 request_module_nowait(IB_DRV_NAME);
1116 * Change the port configuration of the device.
1117 * Every user of this function must hold the port mutex.
1119 int mlx4_change_port_types(struct mlx4_dev *dev,
1120 enum mlx4_port_type *port_types)
1126 for (port = 0; port < dev->caps.num_ports; port++) {
1127 /* Change the port type only if the new type is different
1128 * from the current, and not set to Auto */
1129 if (port_types[port] != dev->caps.port_type[port + 1])
1133 mlx4_unregister_device(dev);
1134 for (port = 1; port <= dev->caps.num_ports; port++) {
1135 mlx4_CLOSE_PORT(dev, port);
1136 dev->caps.port_type[port] = port_types[port - 1];
1137 err = mlx4_SET_PORT(dev, port, -1);
1139 mlx4_err(dev, "Failed to set port %d, aborting\n",
1144 mlx4_set_port_mask(dev);
1145 err = mlx4_register_device(dev);
1147 mlx4_err(dev, "Failed to register device\n");
1150 mlx4_request_modules(dev);
1157 static ssize_t show_port_type(struct device *dev,
1158 struct device_attribute *attr,
1161 struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1163 struct mlx4_dev *mdev = info->dev;
1167 (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_IB) ?
1169 if (mdev->caps.possible_type[info->port] == MLX4_PORT_TYPE_AUTO)
1170 sprintf(buf, "auto (%s)\n", type);
1172 sprintf(buf, "%s\n", type);
1177 static int __set_port_type(struct mlx4_port_info *info,
1178 enum mlx4_port_type port_type)
1180 struct mlx4_dev *mdev = info->dev;
1181 struct mlx4_priv *priv = mlx4_priv(mdev);
1182 enum mlx4_port_type types[MLX4_MAX_PORTS];
1183 enum mlx4_port_type new_types[MLX4_MAX_PORTS];
1187 if ((port_type & mdev->caps.supported_type[info->port]) != port_type) {
1189 "Requested port type for port %d is not supported on this HCA\n",
1194 mlx4_stop_sense(mdev);
1195 mutex_lock(&priv->port_mutex);
1196 info->tmp_type = port_type;
1198 /* Possible type is always the one that was delivered */
1199 mdev->caps.possible_type[info->port] = info->tmp_type;
1201 for (i = 0; i < mdev->caps.num_ports; i++) {
1202 types[i] = priv->port[i+1].tmp_type ? priv->port[i+1].tmp_type :
1203 mdev->caps.possible_type[i+1];
1204 if (types[i] == MLX4_PORT_TYPE_AUTO)
1205 types[i] = mdev->caps.port_type[i+1];
1208 if (!(mdev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
1209 !(mdev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT)) {
1210 for (i = 1; i <= mdev->caps.num_ports; i++) {
1211 if (mdev->caps.possible_type[i] == MLX4_PORT_TYPE_AUTO) {
1212 mdev->caps.possible_type[i] = mdev->caps.port_type[i];
1218 mlx4_err(mdev, "Auto sensing is not supported on this HCA. Set only 'eth' or 'ib' for both ports (should be the same)\n");
1222 mlx4_do_sense_ports(mdev, new_types, types);
1224 err = mlx4_check_port_params(mdev, new_types);
1228 /* We are about to apply the changes after the configuration
1229 * was verified, no need to remember the temporary types
1231 for (i = 0; i < mdev->caps.num_ports; i++)
1232 priv->port[i + 1].tmp_type = 0;
1234 err = mlx4_change_port_types(mdev, new_types);
1237 mlx4_start_sense(mdev);
1238 mutex_unlock(&priv->port_mutex);
1243 static ssize_t set_port_type(struct device *dev,
1244 struct device_attribute *attr,
1245 const char *buf, size_t count)
1247 struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1249 struct mlx4_dev *mdev = info->dev;
1250 enum mlx4_port_type port_type;
1251 static DEFINE_MUTEX(set_port_type_mutex);
1254 mutex_lock(&set_port_type_mutex);
1256 if (!strcmp(buf, "ib\n")) {
1257 port_type = MLX4_PORT_TYPE_IB;
1258 } else if (!strcmp(buf, "eth\n")) {
1259 port_type = MLX4_PORT_TYPE_ETH;
1260 } else if (!strcmp(buf, "auto\n")) {
1261 port_type = MLX4_PORT_TYPE_AUTO;
1263 mlx4_err(mdev, "%s is not supported port type\n", buf);
1268 err = __set_port_type(info, port_type);
1271 mutex_unlock(&set_port_type_mutex);
1273 return err ? err : count;
1284 static inline int int_to_ibta_mtu(int mtu)
1287 case 256: return IB_MTU_256;
1288 case 512: return IB_MTU_512;
1289 case 1024: return IB_MTU_1024;
1290 case 2048: return IB_MTU_2048;
1291 case 4096: return IB_MTU_4096;
1296 static inline int ibta_mtu_to_int(enum ibta_mtu mtu)
1299 case IB_MTU_256: return 256;
1300 case IB_MTU_512: return 512;
1301 case IB_MTU_1024: return 1024;
1302 case IB_MTU_2048: return 2048;
1303 case IB_MTU_4096: return 4096;
1308 static ssize_t show_port_ib_mtu(struct device *dev,
1309 struct device_attribute *attr,
1312 struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1314 struct mlx4_dev *mdev = info->dev;
1316 if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH)
1317 mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
1319 sprintf(buf, "%d\n",
1320 ibta_mtu_to_int(mdev->caps.port_ib_mtu[info->port]));
1324 static ssize_t set_port_ib_mtu(struct device *dev,
1325 struct device_attribute *attr,
1326 const char *buf, size_t count)
1328 struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1330 struct mlx4_dev *mdev = info->dev;
1331 struct mlx4_priv *priv = mlx4_priv(mdev);
1332 int err, port, mtu, ibta_mtu = -1;
1334 if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH) {
1335 mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
1339 err = kstrtoint(buf, 0, &mtu);
1341 ibta_mtu = int_to_ibta_mtu(mtu);
1343 if (err || ibta_mtu < 0) {
1344 mlx4_err(mdev, "%s is invalid IBTA mtu\n", buf);
1348 mdev->caps.port_ib_mtu[info->port] = ibta_mtu;
1350 mlx4_stop_sense(mdev);
1351 mutex_lock(&priv->port_mutex);
1352 mlx4_unregister_device(mdev);
1353 for (port = 1; port <= mdev->caps.num_ports; port++) {
1354 mlx4_CLOSE_PORT(mdev, port);
1355 err = mlx4_SET_PORT(mdev, port, -1);
1357 mlx4_err(mdev, "Failed to set port %d, aborting\n",
1362 err = mlx4_register_device(mdev);
1364 mutex_unlock(&priv->port_mutex);
1365 mlx4_start_sense(mdev);
1366 return err ? err : count;
1369 /* bond for multi-function device */
1370 #define MAX_MF_BOND_ALLOWED_SLAVES 63
1371 static int mlx4_mf_bond(struct mlx4_dev *dev)
1375 struct mlx4_slaves_pport slaves_port1;
1376 struct mlx4_slaves_pport slaves_port2;
1377 DECLARE_BITMAP(slaves_port_1_2, MLX4_MFUNC_MAX);
1379 slaves_port1 = mlx4_phys_to_slaves_pport(dev, 1);
1380 slaves_port2 = mlx4_phys_to_slaves_pport(dev, 2);
1381 bitmap_and(slaves_port_1_2,
1382 slaves_port1.slaves, slaves_port2.slaves,
1383 dev->persist->num_vfs + 1);
1385 /* only single port vfs are allowed */
1386 if (bitmap_weight(slaves_port_1_2, dev->persist->num_vfs + 1) > 1) {
1387 mlx4_warn(dev, "HA mode unsupported for dual ported VFs\n");
1391 /* number of virtual functions is number of total functions minus one
1392 * physical function for each port.
1394 nvfs = bitmap_weight(slaves_port1.slaves, dev->persist->num_vfs + 1) +
1395 bitmap_weight(slaves_port2.slaves, dev->persist->num_vfs + 1) - 2;
1397 /* limit on maximum allowed VFs */
1398 if (nvfs > MAX_MF_BOND_ALLOWED_SLAVES) {
1399 mlx4_warn(dev, "HA mode is not supported for %d VFs (max %d are allowed)\n",
1400 nvfs, MAX_MF_BOND_ALLOWED_SLAVES);
1404 if (dev->caps.steering_mode != MLX4_STEERING_MODE_DEVICE_MANAGED) {
1405 mlx4_warn(dev, "HA mode unsupported for NON DMFS steering\n");
1409 err = mlx4_bond_mac_table(dev);
1412 err = mlx4_bond_vlan_table(dev);
1415 err = mlx4_bond_fs_rules(dev);
1421 (void)mlx4_unbond_vlan_table(dev);
1423 (void)mlx4_unbond_mac_table(dev);
1427 static int mlx4_mf_unbond(struct mlx4_dev *dev)
1431 ret = mlx4_unbond_fs_rules(dev);
1433 mlx4_warn(dev, "multifunction unbond for flow rules failed (%d)\n", ret);
1434 ret1 = mlx4_unbond_mac_table(dev);
1436 mlx4_warn(dev, "multifunction unbond for MAC table failed (%d)\n", ret1);
1439 ret1 = mlx4_unbond_vlan_table(dev);
1441 mlx4_warn(dev, "multifunction unbond for VLAN table failed (%d)\n", ret1);
1447 int mlx4_bond(struct mlx4_dev *dev)
1450 struct mlx4_priv *priv = mlx4_priv(dev);
1452 mutex_lock(&priv->bond_mutex);
1454 if (!mlx4_is_bonded(dev)) {
1455 ret = mlx4_do_bond(dev, true);
1457 mlx4_err(dev, "Failed to bond device: %d\n", ret);
1458 if (!ret && mlx4_is_master(dev)) {
1459 ret = mlx4_mf_bond(dev);
1461 mlx4_err(dev, "bond for multifunction failed\n");
1462 mlx4_do_bond(dev, false);
1467 mutex_unlock(&priv->bond_mutex);
1469 mlx4_dbg(dev, "Device is bonded\n");
1473 EXPORT_SYMBOL_GPL(mlx4_bond);
1475 int mlx4_unbond(struct mlx4_dev *dev)
1478 struct mlx4_priv *priv = mlx4_priv(dev);
1480 mutex_lock(&priv->bond_mutex);
1482 if (mlx4_is_bonded(dev)) {
1485 ret = mlx4_do_bond(dev, false);
1487 mlx4_err(dev, "Failed to unbond device: %d\n", ret);
1488 if (mlx4_is_master(dev))
1489 ret2 = mlx4_mf_unbond(dev);
1491 mlx4_warn(dev, "Failed to unbond device for multifunction (%d)\n", ret2);
1496 mutex_unlock(&priv->bond_mutex);
1498 mlx4_dbg(dev, "Device is unbonded\n");
1502 EXPORT_SYMBOL_GPL(mlx4_unbond);
1505 int mlx4_port_map_set(struct mlx4_dev *dev, struct mlx4_port_map *v2p)
1507 u8 port1 = v2p->port1;
1508 u8 port2 = v2p->port2;
1509 struct mlx4_priv *priv = mlx4_priv(dev);
1512 if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_PORT_REMAP))
1515 mutex_lock(&priv->bond_mutex);
1517 /* zero means keep current mapping for this port */
1519 port1 = priv->v2p.port1;
1521 port2 = priv->v2p.port2;
1523 if ((port1 < 1) || (port1 > MLX4_MAX_PORTS) ||
1524 (port2 < 1) || (port2 > MLX4_MAX_PORTS) ||
1525 (port1 == 2 && port2 == 1)) {
1526 /* besides boundary checks cross mapping makes
1527 * no sense and therefore not allowed */
1529 } else if ((port1 == priv->v2p.port1) &&
1530 (port2 == priv->v2p.port2)) {
1533 err = mlx4_virt2phy_port_map(dev, port1, port2);
1535 mlx4_dbg(dev, "port map changed: [%d][%d]\n",
1537 priv->v2p.port1 = port1;
1538 priv->v2p.port2 = port2;
1540 mlx4_err(dev, "Failed to change port mape: %d\n", err);
1544 mutex_unlock(&priv->bond_mutex);
1547 EXPORT_SYMBOL_GPL(mlx4_port_map_set);
1549 static int mlx4_load_fw(struct mlx4_dev *dev)
1551 struct mlx4_priv *priv = mlx4_priv(dev);
1554 priv->fw.fw_icm = mlx4_alloc_icm(dev, priv->fw.fw_pages,
1555 GFP_HIGHUSER | __GFP_NOWARN, 0);
1556 if (!priv->fw.fw_icm) {
1557 mlx4_err(dev, "Couldn't allocate FW area, aborting\n");
1561 err = mlx4_MAP_FA(dev, priv->fw.fw_icm);
1563 mlx4_err(dev, "MAP_FA command failed, aborting\n");
1567 err = mlx4_RUN_FW(dev);
1569 mlx4_err(dev, "RUN_FW command failed, aborting\n");
1579 mlx4_free_icm(dev, priv->fw.fw_icm, 0);
1583 static int mlx4_init_cmpt_table(struct mlx4_dev *dev, u64 cmpt_base,
1586 struct mlx4_priv *priv = mlx4_priv(dev);
1590 err = mlx4_init_icm_table(dev, &priv->qp_table.cmpt_table,
1592 ((u64) (MLX4_CMPT_TYPE_QP *
1593 cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1594 cmpt_entry_sz, dev->caps.num_qps,
1595 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1600 err = mlx4_init_icm_table(dev, &priv->srq_table.cmpt_table,
1602 ((u64) (MLX4_CMPT_TYPE_SRQ *
1603 cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1604 cmpt_entry_sz, dev->caps.num_srqs,
1605 dev->caps.reserved_srqs, 0, 0);
1609 err = mlx4_init_icm_table(dev, &priv->cq_table.cmpt_table,
1611 ((u64) (MLX4_CMPT_TYPE_CQ *
1612 cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1613 cmpt_entry_sz, dev->caps.num_cqs,
1614 dev->caps.reserved_cqs, 0, 0);
1618 num_eqs = dev->phys_caps.num_phys_eqs;
1619 err = mlx4_init_icm_table(dev, &priv->eq_table.cmpt_table,
1621 ((u64) (MLX4_CMPT_TYPE_EQ *
1622 cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1623 cmpt_entry_sz, num_eqs, num_eqs, 0, 0);
1630 mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1633 mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1636 mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1642 static int mlx4_init_icm(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap,
1643 struct mlx4_init_hca_param *init_hca, u64 icm_size)
1645 struct mlx4_priv *priv = mlx4_priv(dev);
1650 err = mlx4_SET_ICM_SIZE(dev, icm_size, &aux_pages);
1652 mlx4_err(dev, "SET_ICM_SIZE command failed, aborting\n");
1656 mlx4_dbg(dev, "%lld KB of HCA context requires %lld KB aux memory\n",
1657 (unsigned long long) icm_size >> 10,
1658 (unsigned long long) aux_pages << 2);
1660 priv->fw.aux_icm = mlx4_alloc_icm(dev, aux_pages,
1661 GFP_HIGHUSER | __GFP_NOWARN, 0);
1662 if (!priv->fw.aux_icm) {
1663 mlx4_err(dev, "Couldn't allocate aux memory, aborting\n");
1667 err = mlx4_MAP_ICM_AUX(dev, priv->fw.aux_icm);
1669 mlx4_err(dev, "MAP_ICM_AUX command failed, aborting\n");
1673 err = mlx4_init_cmpt_table(dev, init_hca->cmpt_base, dev_cap->cmpt_entry_sz);
1675 mlx4_err(dev, "Failed to map cMPT context memory, aborting\n");
1680 num_eqs = dev->phys_caps.num_phys_eqs;
1681 err = mlx4_init_icm_table(dev, &priv->eq_table.table,
1682 init_hca->eqc_base, dev_cap->eqc_entry_sz,
1683 num_eqs, num_eqs, 0, 0);
1685 mlx4_err(dev, "Failed to map EQ context memory, aborting\n");
1686 goto err_unmap_cmpt;
1690 * Reserved MTT entries must be aligned up to a cacheline
1691 * boundary, since the FW will write to them, while the driver
1692 * writes to all other MTT entries. (The variable
1693 * dev->caps.mtt_entry_sz below is really the MTT segment
1694 * size, not the raw entry size)
1696 dev->caps.reserved_mtts =
1697 ALIGN(dev->caps.reserved_mtts * dev->caps.mtt_entry_sz,
1698 dma_get_cache_alignment()) / dev->caps.mtt_entry_sz;
1700 err = mlx4_init_icm_table(dev, &priv->mr_table.mtt_table,
1702 dev->caps.mtt_entry_sz,
1704 dev->caps.reserved_mtts, 1, 0);
1706 mlx4_err(dev, "Failed to map MTT context memory, aborting\n");
1710 err = mlx4_init_icm_table(dev, &priv->mr_table.dmpt_table,
1711 init_hca->dmpt_base,
1712 dev_cap->dmpt_entry_sz,
1714 dev->caps.reserved_mrws, 1, 1);
1716 mlx4_err(dev, "Failed to map dMPT context memory, aborting\n");
1720 err = mlx4_init_icm_table(dev, &priv->qp_table.qp_table,
1722 dev_cap->qpc_entry_sz,
1724 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1727 mlx4_err(dev, "Failed to map QP context memory, aborting\n");
1728 goto err_unmap_dmpt;
1731 err = mlx4_init_icm_table(dev, &priv->qp_table.auxc_table,
1732 init_hca->auxc_base,
1733 dev_cap->aux_entry_sz,
1735 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1738 mlx4_err(dev, "Failed to map AUXC context memory, aborting\n");
1742 err = mlx4_init_icm_table(dev, &priv->qp_table.altc_table,
1743 init_hca->altc_base,
1744 dev_cap->altc_entry_sz,
1746 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1749 mlx4_err(dev, "Failed to map ALTC context memory, aborting\n");
1750 goto err_unmap_auxc;
1753 err = mlx4_init_icm_table(dev, &priv->qp_table.rdmarc_table,
1754 init_hca->rdmarc_base,
1755 dev_cap->rdmarc_entry_sz << priv->qp_table.rdmarc_shift,
1757 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1760 mlx4_err(dev, "Failed to map RDMARC context memory, aborting\n");
1761 goto err_unmap_altc;
1764 err = mlx4_init_icm_table(dev, &priv->cq_table.table,
1766 dev_cap->cqc_entry_sz,
1768 dev->caps.reserved_cqs, 0, 0);
1770 mlx4_err(dev, "Failed to map CQ context memory, aborting\n");
1771 goto err_unmap_rdmarc;
1774 err = mlx4_init_icm_table(dev, &priv->srq_table.table,
1775 init_hca->srqc_base,
1776 dev_cap->srq_entry_sz,
1778 dev->caps.reserved_srqs, 0, 0);
1780 mlx4_err(dev, "Failed to map SRQ context memory, aborting\n");
1785 * For flow steering device managed mode it is required to use
1786 * mlx4_init_icm_table. For B0 steering mode it's not strictly
1787 * required, but for simplicity just map the whole multicast
1788 * group table now. The table isn't very big and it's a lot
1789 * easier than trying to track ref counts.
1791 err = mlx4_init_icm_table(dev, &priv->mcg_table.table,
1793 mlx4_get_mgm_entry_size(dev),
1794 dev->caps.num_mgms + dev->caps.num_amgms,
1795 dev->caps.num_mgms + dev->caps.num_amgms,
1798 mlx4_err(dev, "Failed to map MCG context memory, aborting\n");
1805 mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1808 mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1811 mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1814 mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1817 mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1820 mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1823 mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1826 mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
1829 mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
1832 mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1833 mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1834 mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1835 mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1838 mlx4_UNMAP_ICM_AUX(dev);
1841 mlx4_free_icm(dev, priv->fw.aux_icm, 0);
1846 static void mlx4_free_icms(struct mlx4_dev *dev)
1848 struct mlx4_priv *priv = mlx4_priv(dev);
1850 mlx4_cleanup_icm_table(dev, &priv->mcg_table.table);
1851 mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1852 mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1853 mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1854 mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1855 mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1856 mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1857 mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1858 mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
1859 mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
1860 mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1861 mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1862 mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1863 mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1865 mlx4_UNMAP_ICM_AUX(dev);
1866 mlx4_free_icm(dev, priv->fw.aux_icm, 0);
1869 static void mlx4_slave_exit(struct mlx4_dev *dev)
1871 struct mlx4_priv *priv = mlx4_priv(dev);
1873 mutex_lock(&priv->cmd.slave_cmd_mutex);
1874 if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, MLX4_COMM_CMD_NA_OP,
1876 mlx4_warn(dev, "Failed to close slave function\n");
1877 mutex_unlock(&priv->cmd.slave_cmd_mutex);
1880 static int map_bf_area(struct mlx4_dev *dev)
1882 struct mlx4_priv *priv = mlx4_priv(dev);
1883 resource_size_t bf_start;
1884 resource_size_t bf_len;
1887 if (!dev->caps.bf_reg_size)
1890 bf_start = pci_resource_start(dev->persist->pdev, 2) +
1891 (dev->caps.num_uars << PAGE_SHIFT);
1892 bf_len = pci_resource_len(dev->persist->pdev, 2) -
1893 (dev->caps.num_uars << PAGE_SHIFT);
1894 priv->bf_mapping = io_mapping_create_wc(bf_start, bf_len);
1895 if (!priv->bf_mapping)
1901 static void unmap_bf_area(struct mlx4_dev *dev)
1903 if (mlx4_priv(dev)->bf_mapping)
1904 io_mapping_free(mlx4_priv(dev)->bf_mapping);
1907 u64 mlx4_read_clock(struct mlx4_dev *dev)
1909 u32 clockhi, clocklo, clockhi1;
1912 struct mlx4_priv *priv = mlx4_priv(dev);
1914 for (i = 0; i < 10; i++) {
1915 clockhi = swab32(readl(priv->clock_mapping));
1916 clocklo = swab32(readl(priv->clock_mapping + 4));
1917 clockhi1 = swab32(readl(priv->clock_mapping));
1918 if (clockhi == clockhi1)
1922 cycles = (u64) clockhi << 32 | (u64) clocklo;
1926 EXPORT_SYMBOL_GPL(mlx4_read_clock);
1929 static int map_internal_clock(struct mlx4_dev *dev)
1931 struct mlx4_priv *priv = mlx4_priv(dev);
1933 priv->clock_mapping =
1934 ioremap(pci_resource_start(dev->persist->pdev,
1935 priv->fw.clock_bar) +
1936 priv->fw.clock_offset, MLX4_CLOCK_SIZE);
1938 if (!priv->clock_mapping)
1944 int mlx4_get_internal_clock_params(struct mlx4_dev *dev,
1945 struct mlx4_clock_params *params)
1947 struct mlx4_priv *priv = mlx4_priv(dev);
1949 if (mlx4_is_slave(dev))
1952 if (!dev->caps.map_clock_to_user) {
1953 mlx4_dbg(dev, "Map clock to user is not supported.\n");
1960 params->bar = priv->fw.clock_bar;
1961 params->offset = priv->fw.clock_offset;
1962 params->size = MLX4_CLOCK_SIZE;
1966 EXPORT_SYMBOL_GPL(mlx4_get_internal_clock_params);
1968 static void unmap_internal_clock(struct mlx4_dev *dev)
1970 struct mlx4_priv *priv = mlx4_priv(dev);
1972 if (priv->clock_mapping)
1973 iounmap(priv->clock_mapping);
1976 static void mlx4_close_hca(struct mlx4_dev *dev)
1978 unmap_internal_clock(dev);
1980 if (mlx4_is_slave(dev))
1981 mlx4_slave_exit(dev);
1983 mlx4_CLOSE_HCA(dev, 0);
1984 mlx4_free_icms(dev);
1988 static void mlx4_close_fw(struct mlx4_dev *dev)
1990 if (!mlx4_is_slave(dev)) {
1992 mlx4_free_icm(dev, mlx4_priv(dev)->fw.fw_icm, 0);
1996 static int mlx4_comm_check_offline(struct mlx4_dev *dev)
1998 #define COMM_CHAN_OFFLINE_OFFSET 0x09
2003 struct mlx4_priv *priv = mlx4_priv(dev);
2005 end = msecs_to_jiffies(MLX4_COMM_OFFLINE_TIME_OUT) + jiffies;
2006 while (time_before(jiffies, end)) {
2007 comm_flags = swab32(readl((__iomem char *)priv->mfunc.comm +
2008 MLX4_COMM_CHAN_FLAGS));
2009 offline_bit = (comm_flags &
2010 (u32)(1 << COMM_CHAN_OFFLINE_OFFSET));
2014 /* If device removal has been requested,
2015 * do not continue retrying.
2017 if (dev->persist->interface_state &
2018 MLX4_INTERFACE_STATE_NOWAIT)
2021 /* There are cases as part of AER/Reset flow that PF needs
2022 * around 100 msec to load. We therefore sleep for 100 msec
2023 * to allow other tasks to make use of that CPU during this
2028 mlx4_err(dev, "Communication channel is offline.\n");
2032 static void mlx4_reset_vf_support(struct mlx4_dev *dev)
2034 #define COMM_CHAN_RST_OFFSET 0x1e
2036 struct mlx4_priv *priv = mlx4_priv(dev);
2040 comm_caps = swab32(readl((__iomem char *)priv->mfunc.comm +
2041 MLX4_COMM_CHAN_CAPS));
2042 comm_rst = (comm_caps & (u32)(1 << COMM_CHAN_RST_OFFSET));
2045 dev->caps.vf_caps |= MLX4_VF_CAP_FLAG_RESET;
2048 static int mlx4_init_slave(struct mlx4_dev *dev)
2050 struct mlx4_priv *priv = mlx4_priv(dev);
2051 u64 dma = (u64) priv->mfunc.vhcr_dma;
2052 int ret_from_reset = 0;
2054 u32 cmd_channel_ver;
2056 if (atomic_read(&pf_loading)) {
2057 mlx4_warn(dev, "PF is not ready - Deferring probe\n");
2058 return -EPROBE_DEFER;
2061 mutex_lock(&priv->cmd.slave_cmd_mutex);
2062 priv->cmd.max_cmds = 1;
2063 if (mlx4_comm_check_offline(dev)) {
2064 mlx4_err(dev, "PF is not responsive, skipping initialization\n");
2068 mlx4_reset_vf_support(dev);
2069 mlx4_warn(dev, "Sending reset\n");
2070 ret_from_reset = mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0,
2071 MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME);
2072 /* if we are in the middle of flr the slave will try
2073 * NUM_OF_RESET_RETRIES times before leaving.*/
2074 if (ret_from_reset) {
2075 if (MLX4_DELAY_RESET_SLAVE == ret_from_reset) {
2076 mlx4_warn(dev, "slave is currently in the middle of FLR - Deferring probe\n");
2077 mutex_unlock(&priv->cmd.slave_cmd_mutex);
2078 return -EPROBE_DEFER;
2083 /* check the driver version - the slave I/F revision
2084 * must match the master's */
2085 slave_read = swab32(readl(&priv->mfunc.comm->slave_read));
2086 cmd_channel_ver = mlx4_comm_get_version();
2088 if (MLX4_COMM_GET_IF_REV(cmd_channel_ver) !=
2089 MLX4_COMM_GET_IF_REV(slave_read)) {
2090 mlx4_err(dev, "slave driver version is not supported by the master\n");
2094 mlx4_warn(dev, "Sending vhcr0\n");
2095 if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR0, dma >> 48,
2096 MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
2098 if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR1, dma >> 32,
2099 MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
2101 if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR2, dma >> 16,
2102 MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
2104 if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR_EN, dma,
2105 MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
2108 mutex_unlock(&priv->cmd.slave_cmd_mutex);
2112 mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, MLX4_COMM_CMD_NA_OP, 0);
2114 mutex_unlock(&priv->cmd.slave_cmd_mutex);
2118 static void mlx4_parav_master_pf_caps(struct mlx4_dev *dev)
2122 for (i = 1; i <= dev->caps.num_ports; i++) {
2123 if (dev->caps.port_type[i] == MLX4_PORT_TYPE_ETH)
2124 dev->caps.gid_table_len[i] =
2125 mlx4_get_slave_num_gids(dev, 0, i);
2127 dev->caps.gid_table_len[i] = 1;
2128 dev->caps.pkey_table_len[i] =
2129 dev->phys_caps.pkey_phys_table_len[i] - 1;
2133 static int choose_log_fs_mgm_entry_size(int qp_per_entry)
2135 int i = MLX4_MIN_MGM_LOG_ENTRY_SIZE;
2137 for (i = MLX4_MIN_MGM_LOG_ENTRY_SIZE; i <= MLX4_MAX_MGM_LOG_ENTRY_SIZE;
2139 if (qp_per_entry <= 4 * ((1 << i) / 16 - 2))
2143 return (i <= MLX4_MAX_MGM_LOG_ENTRY_SIZE) ? i : -1;
2146 static const char *dmfs_high_rate_steering_mode_str(int dmfs_high_steer_mode)
2148 switch (dmfs_high_steer_mode) {
2149 case MLX4_STEERING_DMFS_A0_DEFAULT:
2150 return "default performance";
2152 case MLX4_STEERING_DMFS_A0_DYNAMIC:
2153 return "dynamic hybrid mode";
2155 case MLX4_STEERING_DMFS_A0_STATIC:
2156 return "performance optimized for limited rule configuration (static)";
2158 case MLX4_STEERING_DMFS_A0_DISABLE:
2159 return "disabled performance optimized steering";
2161 case MLX4_STEERING_DMFS_A0_NOT_SUPPORTED:
2162 return "performance optimized steering not supported";
2165 return "Unrecognized mode";
2169 #define MLX4_DMFS_A0_STEERING (1UL << 2)
2171 static void choose_steering_mode(struct mlx4_dev *dev,
2172 struct mlx4_dev_cap *dev_cap)
2174 if (mlx4_log_num_mgm_entry_size <= 0) {
2175 if ((-mlx4_log_num_mgm_entry_size) & MLX4_DMFS_A0_STEERING) {
2176 if (dev->caps.dmfs_high_steer_mode ==
2177 MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
2178 mlx4_err(dev, "DMFS high rate mode not supported\n");
2180 dev->caps.dmfs_high_steer_mode =
2181 MLX4_STEERING_DMFS_A0_STATIC;
2185 if (mlx4_log_num_mgm_entry_size <= 0 &&
2186 dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_FS_EN &&
2187 (!mlx4_is_mfunc(dev) ||
2188 (dev_cap->fs_max_num_qp_per_entry >=
2189 (dev->persist->num_vfs + 1))) &&
2190 choose_log_fs_mgm_entry_size(dev_cap->fs_max_num_qp_per_entry) >=
2191 MLX4_MIN_MGM_LOG_ENTRY_SIZE) {
2192 dev->oper_log_mgm_entry_size =
2193 choose_log_fs_mgm_entry_size(dev_cap->fs_max_num_qp_per_entry);
2194 dev->caps.steering_mode = MLX4_STEERING_MODE_DEVICE_MANAGED;
2195 dev->caps.num_qp_per_mgm = dev_cap->fs_max_num_qp_per_entry;
2196 dev->caps.fs_log_max_ucast_qp_range_size =
2197 dev_cap->fs_log_max_ucast_qp_range_size;
2199 if (dev->caps.dmfs_high_steer_mode !=
2200 MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
2201 dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_DISABLE;
2202 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER &&
2203 dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER)
2204 dev->caps.steering_mode = MLX4_STEERING_MODE_B0;
2206 dev->caps.steering_mode = MLX4_STEERING_MODE_A0;
2208 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER ||
2209 dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER)
2210 mlx4_warn(dev, "Must have both UC_STEER and MC_STEER flags set to use B0 steering - falling back to A0 steering mode\n");
2212 dev->oper_log_mgm_entry_size =
2213 mlx4_log_num_mgm_entry_size > 0 ?
2214 mlx4_log_num_mgm_entry_size :
2215 MLX4_DEFAULT_MGM_LOG_ENTRY_SIZE;
2216 dev->caps.num_qp_per_mgm = mlx4_get_qp_per_mgm(dev);
2218 mlx4_dbg(dev, "Steering mode is: %s, oper_log_mgm_entry_size = %d, modparam log_num_mgm_entry_size = %d\n",
2219 mlx4_steering_mode_str(dev->caps.steering_mode),
2220 dev->oper_log_mgm_entry_size,
2221 mlx4_log_num_mgm_entry_size);
2224 static void choose_tunnel_offload_mode(struct mlx4_dev *dev,
2225 struct mlx4_dev_cap *dev_cap)
2227 if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2228 dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_VXLAN_OFFLOADS)
2229 dev->caps.tunnel_offload_mode = MLX4_TUNNEL_OFFLOAD_MODE_VXLAN;
2231 dev->caps.tunnel_offload_mode = MLX4_TUNNEL_OFFLOAD_MODE_NONE;
2233 mlx4_dbg(dev, "Tunneling offload mode is: %s\n", (dev->caps.tunnel_offload_mode
2234 == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) ? "vxlan" : "none");
2237 static int mlx4_validate_optimized_steering(struct mlx4_dev *dev)
2240 struct mlx4_port_cap port_cap;
2242 if (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
2245 for (i = 1; i <= dev->caps.num_ports; i++) {
2246 if (mlx4_dev_port(dev, i, &port_cap)) {
2248 "QUERY_DEV_CAP command failed, can't verify DMFS high rate steering.\n");
2249 } else if ((dev->caps.dmfs_high_steer_mode !=
2250 MLX4_STEERING_DMFS_A0_DEFAULT) &&
2251 (port_cap.dmfs_optimized_state ==
2252 !!(dev->caps.dmfs_high_steer_mode ==
2253 MLX4_STEERING_DMFS_A0_DISABLE))) {
2255 "DMFS high rate steer mode differ, driver requested %s but %s in FW.\n",
2256 dmfs_high_rate_steering_mode_str(
2257 dev->caps.dmfs_high_steer_mode),
2258 (port_cap.dmfs_optimized_state ?
2259 "enabled" : "disabled"));
2266 static int mlx4_init_fw(struct mlx4_dev *dev)
2268 struct mlx4_mod_stat_cfg mlx4_cfg;
2271 if (!mlx4_is_slave(dev)) {
2272 err = mlx4_QUERY_FW(dev);
2275 mlx4_info(dev, "non-primary physical function, skipping\n");
2277 mlx4_err(dev, "QUERY_FW command failed, aborting\n");
2281 err = mlx4_load_fw(dev);
2283 mlx4_err(dev, "Failed to start FW, aborting\n");
2287 mlx4_cfg.log_pg_sz_m = 1;
2288 mlx4_cfg.log_pg_sz = 0;
2289 err = mlx4_MOD_STAT_CFG(dev, &mlx4_cfg);
2291 mlx4_warn(dev, "Failed to override log_pg_sz parameter\n");
2297 static int mlx4_init_hca(struct mlx4_dev *dev)
2299 struct mlx4_priv *priv = mlx4_priv(dev);
2300 struct mlx4_init_hca_param *init_hca = NULL;
2301 struct mlx4_dev_cap *dev_cap = NULL;
2302 struct mlx4_adapter adapter;
2303 struct mlx4_profile profile;
2305 struct mlx4_config_dev_params params;
2308 if (!mlx4_is_slave(dev)) {
2309 dev_cap = kzalloc(sizeof(*dev_cap), GFP_KERNEL);
2310 init_hca = kzalloc(sizeof(*init_hca), GFP_KERNEL);
2312 if (!dev_cap || !init_hca) {
2317 err = mlx4_dev_cap(dev, dev_cap);
2319 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
2323 choose_steering_mode(dev, dev_cap);
2324 choose_tunnel_offload_mode(dev, dev_cap);
2326 if (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC &&
2327 mlx4_is_master(dev))
2328 dev->caps.function_caps |= MLX4_FUNC_CAP_DMFS_A0_STATIC;
2330 err = mlx4_get_phys_port_id(dev);
2332 mlx4_err(dev, "Fail to get physical port id\n");
2334 if (mlx4_is_master(dev))
2335 mlx4_parav_master_pf_caps(dev);
2337 if (mlx4_low_memory_profile()) {
2338 mlx4_info(dev, "Running from within kdump kernel. Using low memory profile\n");
2339 profile = low_mem_profile;
2341 profile = default_profile;
2343 if (dev->caps.steering_mode ==
2344 MLX4_STEERING_MODE_DEVICE_MANAGED)
2345 profile.num_mcg = MLX4_FS_NUM_MCG;
2347 icm_size = mlx4_make_profile(dev, &profile, dev_cap,
2349 if ((long long) icm_size < 0) {
2354 if (enable_4k_uar || !dev->persist->num_vfs) {
2355 init_hca->log_uar_sz = ilog2(dev->caps.num_uars) +
2356 PAGE_SHIFT - DEFAULT_UAR_PAGE_SHIFT;
2357 init_hca->uar_page_sz = DEFAULT_UAR_PAGE_SHIFT - 12;
2359 init_hca->log_uar_sz = ilog2(dev->caps.num_uars);
2360 init_hca->uar_page_sz = PAGE_SHIFT - 12;
2363 init_hca->mw_enabled = 0;
2364 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2365 dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN)
2366 init_hca->mw_enabled = INIT_HCA_TPT_MW_ENABLE;
2368 err = mlx4_init_icm(dev, dev_cap, init_hca, icm_size);
2372 err = mlx4_INIT_HCA(dev, init_hca);
2374 mlx4_err(dev, "INIT_HCA command failed, aborting\n");
2378 if (dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS) {
2379 err = mlx4_query_func(dev, dev_cap);
2381 mlx4_err(dev, "QUERY_FUNC command failed, aborting.\n");
2383 } else if (err & MLX4_QUERY_FUNC_NUM_SYS_EQS) {
2384 dev->caps.num_eqs = dev_cap->max_eqs;
2385 dev->caps.reserved_eqs = dev_cap->reserved_eqs;
2386 dev->caps.reserved_uars = dev_cap->reserved_uars;
2391 * If TS is supported by FW
2392 * read HCA frequency by QUERY_HCA command
2394 if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS) {
2395 err = mlx4_QUERY_HCA(dev, init_hca);
2397 mlx4_err(dev, "QUERY_HCA command failed, disable timestamp\n");
2398 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
2400 dev->caps.hca_core_clock =
2401 init_hca->hca_core_clock;
2404 /* In case we got HCA frequency 0 - disable timestamping
2405 * to avoid dividing by zero
2407 if (!dev->caps.hca_core_clock) {
2408 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
2410 "HCA frequency is 0 - timestamping is not supported\n");
2411 } else if (map_internal_clock(dev)) {
2413 * Map internal clock,
2414 * in case of failure disable timestamping
2416 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
2417 mlx4_err(dev, "Failed to map internal clock. Timestamping is not supported\n");
2421 if (dev->caps.dmfs_high_steer_mode !=
2422 MLX4_STEERING_DMFS_A0_NOT_SUPPORTED) {
2423 if (mlx4_validate_optimized_steering(dev))
2424 mlx4_warn(dev, "Optimized steering validation failed\n");
2426 if (dev->caps.dmfs_high_steer_mode ==
2427 MLX4_STEERING_DMFS_A0_DISABLE) {
2428 dev->caps.dmfs_high_rate_qpn_base =
2429 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
2430 dev->caps.dmfs_high_rate_qpn_range =
2431 MLX4_A0_STEERING_TABLE_SIZE;
2434 mlx4_info(dev, "DMFS high rate steer mode is: %s\n",
2435 dmfs_high_rate_steering_mode_str(
2436 dev->caps.dmfs_high_steer_mode));
2439 err = mlx4_init_slave(dev);
2441 if (err != -EPROBE_DEFER)
2442 mlx4_err(dev, "Failed to initialize slave\n");
2446 err = mlx4_slave_cap(dev);
2448 mlx4_err(dev, "Failed to obtain slave caps\n");
2453 if (map_bf_area(dev))
2454 mlx4_dbg(dev, "Failed to map blue flame area\n");
2456 /*Only the master set the ports, all the rest got it from it.*/
2457 if (!mlx4_is_slave(dev))
2458 mlx4_set_port_mask(dev);
2460 err = mlx4_QUERY_ADAPTER(dev, &adapter);
2462 mlx4_err(dev, "QUERY_ADAPTER command failed, aborting\n");
2466 /* Query CONFIG_DEV parameters */
2467 err = mlx4_config_dev_retrieval(dev, ¶ms);
2468 if (err && err != -EOPNOTSUPP) {
2469 mlx4_err(dev, "Failed to query CONFIG_DEV parameters\n");
2471 dev->caps.rx_checksum_flags_port[1] = params.rx_csum_flags_port_1;
2472 dev->caps.rx_checksum_flags_port[2] = params.rx_csum_flags_port_2;
2474 priv->eq_table.inta_pin = adapter.inta_pin;
2475 memcpy(dev->board_id, adapter.board_id, sizeof(dev->board_id));
2481 unmap_internal_clock(dev);
2484 if (mlx4_is_slave(dev))
2485 mlx4_slave_destroy_special_qp_cap(dev);
2488 if (mlx4_is_slave(dev))
2489 mlx4_slave_exit(dev);
2491 mlx4_CLOSE_HCA(dev, 0);
2494 if (!mlx4_is_slave(dev))
2495 mlx4_free_icms(dev);
2504 static int mlx4_init_counters_table(struct mlx4_dev *dev)
2506 struct mlx4_priv *priv = mlx4_priv(dev);
2509 if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2512 if (!dev->caps.max_counters)
2515 nent_pow2 = roundup_pow_of_two(dev->caps.max_counters);
2516 /* reserve last counter index for sink counter */
2517 return mlx4_bitmap_init(&priv->counters_bitmap, nent_pow2,
2519 nent_pow2 - dev->caps.max_counters + 1);
2522 static void mlx4_cleanup_counters_table(struct mlx4_dev *dev)
2524 if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2527 if (!dev->caps.max_counters)
2530 mlx4_bitmap_cleanup(&mlx4_priv(dev)->counters_bitmap);
2533 static void mlx4_cleanup_default_counters(struct mlx4_dev *dev)
2535 struct mlx4_priv *priv = mlx4_priv(dev);
2538 for (port = 0; port < dev->caps.num_ports; port++)
2539 if (priv->def_counter[port] != -1)
2540 mlx4_counter_free(dev, priv->def_counter[port]);
2543 static int mlx4_allocate_default_counters(struct mlx4_dev *dev)
2545 struct mlx4_priv *priv = mlx4_priv(dev);
2549 for (port = 0; port < dev->caps.num_ports; port++)
2550 priv->def_counter[port] = -1;
2552 for (port = 0; port < dev->caps.num_ports; port++) {
2553 err = mlx4_counter_alloc(dev, &idx, MLX4_RES_USAGE_DRIVER);
2555 if (!err || err == -ENOSPC) {
2556 priv->def_counter[port] = idx;
2558 } else if (err == -ENOENT) {
2561 } else if (mlx4_is_slave(dev) && err == -EINVAL) {
2562 priv->def_counter[port] = MLX4_SINK_COUNTER_INDEX(dev);
2563 mlx4_warn(dev, "can't allocate counter from old PF driver, using index %d\n",
2564 MLX4_SINK_COUNTER_INDEX(dev));
2567 mlx4_err(dev, "%s: failed to allocate default counter port %d err %d\n",
2568 __func__, port + 1, err);
2569 mlx4_cleanup_default_counters(dev);
2573 mlx4_dbg(dev, "%s: default counter index %d for port %d\n",
2574 __func__, priv->def_counter[port], port + 1);
2580 int __mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx)
2582 struct mlx4_priv *priv = mlx4_priv(dev);
2584 if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2587 *idx = mlx4_bitmap_alloc(&priv->counters_bitmap);
2589 *idx = MLX4_SINK_COUNTER_INDEX(dev);
2596 int mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx, u8 usage)
2598 u32 in_modifier = RES_COUNTER | (((u32)usage & 3) << 30);
2602 if (mlx4_is_mfunc(dev)) {
2603 err = mlx4_cmd_imm(dev, 0, &out_param, in_modifier,
2604 RES_OP_RESERVE, MLX4_CMD_ALLOC_RES,
2605 MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
2607 *idx = get_param_l(&out_param);
2608 if (WARN_ON(err == -ENOSPC))
2612 return __mlx4_counter_alloc(dev, idx);
2614 EXPORT_SYMBOL_GPL(mlx4_counter_alloc);
2616 static int __mlx4_clear_if_stat(struct mlx4_dev *dev,
2619 struct mlx4_cmd_mailbox *if_stat_mailbox;
2621 u32 if_stat_in_mod = (counter_index & 0xff) | MLX4_QUERY_IF_STAT_RESET;
2623 if_stat_mailbox = mlx4_alloc_cmd_mailbox(dev);
2624 if (IS_ERR(if_stat_mailbox))
2625 return PTR_ERR(if_stat_mailbox);
2627 err = mlx4_cmd_box(dev, 0, if_stat_mailbox->dma, if_stat_in_mod, 0,
2628 MLX4_CMD_QUERY_IF_STAT, MLX4_CMD_TIME_CLASS_C,
2631 mlx4_free_cmd_mailbox(dev, if_stat_mailbox);
2635 void __mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
2637 if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2640 if (idx == MLX4_SINK_COUNTER_INDEX(dev))
2643 __mlx4_clear_if_stat(dev, idx);
2645 mlx4_bitmap_free(&mlx4_priv(dev)->counters_bitmap, idx, MLX4_USE_RR);
2649 void mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
2653 if (mlx4_is_mfunc(dev)) {
2654 set_param_l(&in_param, idx);
2655 mlx4_cmd(dev, in_param, RES_COUNTER, RES_OP_RESERVE,
2656 MLX4_CMD_FREE_RES, MLX4_CMD_TIME_CLASS_A,
2660 __mlx4_counter_free(dev, idx);
2662 EXPORT_SYMBOL_GPL(mlx4_counter_free);
2664 int mlx4_get_default_counter_index(struct mlx4_dev *dev, int port)
2666 struct mlx4_priv *priv = mlx4_priv(dev);
2668 return priv->def_counter[port - 1];
2670 EXPORT_SYMBOL_GPL(mlx4_get_default_counter_index);
2672 void mlx4_set_admin_guid(struct mlx4_dev *dev, __be64 guid, int entry, int port)
2674 struct mlx4_priv *priv = mlx4_priv(dev);
2676 priv->mfunc.master.vf_admin[entry].vport[port].guid = guid;
2678 EXPORT_SYMBOL_GPL(mlx4_set_admin_guid);
2680 __be64 mlx4_get_admin_guid(struct mlx4_dev *dev, int entry, int port)
2682 struct mlx4_priv *priv = mlx4_priv(dev);
2684 return priv->mfunc.master.vf_admin[entry].vport[port].guid;
2686 EXPORT_SYMBOL_GPL(mlx4_get_admin_guid);
2688 void mlx4_set_random_admin_guid(struct mlx4_dev *dev, int entry, int port)
2690 struct mlx4_priv *priv = mlx4_priv(dev);
2697 get_random_bytes((char *)&guid, sizeof(guid));
2698 guid &= ~(cpu_to_be64(1ULL << 56));
2699 guid |= cpu_to_be64(1ULL << 57);
2700 priv->mfunc.master.vf_admin[entry].vport[port].guid = guid;
2703 static int mlx4_setup_hca(struct mlx4_dev *dev)
2705 struct mlx4_priv *priv = mlx4_priv(dev);
2708 __be32 ib_port_default_caps;
2710 err = mlx4_init_uar_table(dev);
2712 mlx4_err(dev, "Failed to initialize user access region table, aborting\n");
2716 err = mlx4_uar_alloc(dev, &priv->driver_uar);
2718 mlx4_err(dev, "Failed to allocate driver access region, aborting\n");
2719 goto err_uar_table_free;
2722 priv->kar = ioremap((phys_addr_t) priv->driver_uar.pfn << PAGE_SHIFT, PAGE_SIZE);
2724 mlx4_err(dev, "Couldn't map kernel access region, aborting\n");
2729 err = mlx4_init_pd_table(dev);
2731 mlx4_err(dev, "Failed to initialize protection domain table, aborting\n");
2735 err = mlx4_init_xrcd_table(dev);
2737 mlx4_err(dev, "Failed to initialize reliable connection domain table, aborting\n");
2738 goto err_pd_table_free;
2741 err = mlx4_init_mr_table(dev);
2743 mlx4_err(dev, "Failed to initialize memory region table, aborting\n");
2744 goto err_xrcd_table_free;
2747 if (!mlx4_is_slave(dev)) {
2748 err = mlx4_init_mcg_table(dev);
2750 mlx4_err(dev, "Failed to initialize multicast group table, aborting\n");
2751 goto err_mr_table_free;
2753 err = mlx4_config_mad_demux(dev);
2755 mlx4_err(dev, "Failed in config_mad_demux, aborting\n");
2756 goto err_mcg_table_free;
2760 err = mlx4_init_eq_table(dev);
2762 mlx4_err(dev, "Failed to initialize event queue table, aborting\n");
2763 goto err_mcg_table_free;
2766 err = mlx4_cmd_use_events(dev);
2768 mlx4_err(dev, "Failed to switch to event-driven firmware commands, aborting\n");
2769 goto err_eq_table_free;
2772 err = mlx4_NOP(dev);
2774 if (dev->flags & MLX4_FLAG_MSI_X) {
2775 mlx4_warn(dev, "NOP command failed to generate MSI-X interrupt IRQ %d)\n",
2776 priv->eq_table.eq[MLX4_EQ_ASYNC].irq);
2777 mlx4_warn(dev, "Trying again without MSI-X\n");
2779 mlx4_err(dev, "NOP command failed to generate interrupt (IRQ %d), aborting\n",
2780 priv->eq_table.eq[MLX4_EQ_ASYNC].irq);
2781 mlx4_err(dev, "BIOS or ACPI interrupt routing problem?\n");
2787 mlx4_dbg(dev, "NOP command IRQ test passed\n");
2789 err = mlx4_init_cq_table(dev);
2791 mlx4_err(dev, "Failed to initialize completion queue table, aborting\n");
2795 err = mlx4_init_srq_table(dev);
2797 mlx4_err(dev, "Failed to initialize shared receive queue table, aborting\n");
2798 goto err_cq_table_free;
2801 err = mlx4_init_qp_table(dev);
2803 mlx4_err(dev, "Failed to initialize queue pair table, aborting\n");
2804 goto err_srq_table_free;
2807 if (!mlx4_is_slave(dev)) {
2808 err = mlx4_init_counters_table(dev);
2809 if (err && err != -ENOENT) {
2810 mlx4_err(dev, "Failed to initialize counters table, aborting\n");
2811 goto err_qp_table_free;
2815 err = mlx4_allocate_default_counters(dev);
2817 mlx4_err(dev, "Failed to allocate default counters, aborting\n");
2818 goto err_counters_table_free;
2821 if (!mlx4_is_slave(dev)) {
2822 for (port = 1; port <= dev->caps.num_ports; port++) {
2823 ib_port_default_caps = 0;
2824 err = mlx4_get_port_ib_caps(dev, port,
2825 &ib_port_default_caps);
2827 mlx4_warn(dev, "failed to get port %d default ib capabilities (%d). Continuing with caps = 0\n",
2829 dev->caps.ib_port_def_cap[port] = ib_port_default_caps;
2831 /* initialize per-slave default ib port capabilities */
2832 if (mlx4_is_master(dev)) {
2834 for (i = 0; i < dev->num_slaves; i++) {
2835 if (i == mlx4_master_func_num(dev))
2837 priv->mfunc.master.slave_state[i].ib_cap_mask[port] =
2838 ib_port_default_caps;
2842 if (mlx4_is_mfunc(dev))
2843 dev->caps.port_ib_mtu[port] = IB_MTU_2048;
2845 dev->caps.port_ib_mtu[port] = IB_MTU_4096;
2847 err = mlx4_SET_PORT(dev, port, mlx4_is_master(dev) ?
2848 dev->caps.pkey_table_len[port] : -1);
2850 mlx4_err(dev, "Failed to set port %d, aborting\n",
2852 goto err_default_countes_free;
2859 err_default_countes_free:
2860 mlx4_cleanup_default_counters(dev);
2862 err_counters_table_free:
2863 if (!mlx4_is_slave(dev))
2864 mlx4_cleanup_counters_table(dev);
2867 mlx4_cleanup_qp_table(dev);
2870 mlx4_cleanup_srq_table(dev);
2873 mlx4_cleanup_cq_table(dev);
2876 mlx4_cmd_use_polling(dev);
2879 mlx4_cleanup_eq_table(dev);
2882 if (!mlx4_is_slave(dev))
2883 mlx4_cleanup_mcg_table(dev);
2886 mlx4_cleanup_mr_table(dev);
2888 err_xrcd_table_free:
2889 mlx4_cleanup_xrcd_table(dev);
2892 mlx4_cleanup_pd_table(dev);
2898 mlx4_uar_free(dev, &priv->driver_uar);
2901 mlx4_cleanup_uar_table(dev);
2905 static int mlx4_init_affinity_hint(struct mlx4_dev *dev, int port, int eqn)
2907 int requested_cpu = 0;
2908 struct mlx4_priv *priv = mlx4_priv(dev);
2913 if (eqn > dev->caps.num_comp_vectors)
2916 for (i = 1; i < port; i++)
2917 off += mlx4_get_eqs_per_port(dev, i);
2919 requested_cpu = eqn - off - !!(eqn > MLX4_EQ_ASYNC);
2921 /* Meaning EQs are shared, and this call comes from the second port */
2922 if (requested_cpu < 0)
2925 eq = &priv->eq_table.eq[eqn];
2927 if (!zalloc_cpumask_var(&eq->affinity_mask, GFP_KERNEL))
2930 cpumask_set_cpu(requested_cpu, eq->affinity_mask);
2935 static void mlx4_enable_msi_x(struct mlx4_dev *dev)
2937 struct mlx4_priv *priv = mlx4_priv(dev);
2938 struct msix_entry *entries;
2943 int nreq = min3(dev->caps.num_ports *
2944 (int)num_online_cpus() + 1,
2945 dev->caps.num_eqs - dev->caps.reserved_eqs,
2949 nreq = min_t(int, nreq, msi_x);
2951 entries = kcalloc(nreq, sizeof(*entries), GFP_KERNEL);
2955 for (i = 0; i < nreq; ++i)
2956 entries[i].entry = i;
2958 nreq = pci_enable_msix_range(dev->persist->pdev, entries, 2,
2961 if (nreq < 0 || nreq < MLX4_EQ_ASYNC) {
2965 /* 1 is reserved for events (asyncrounous EQ) */
2966 dev->caps.num_comp_vectors = nreq - 1;
2968 priv->eq_table.eq[MLX4_EQ_ASYNC].irq = entries[0].vector;
2969 bitmap_zero(priv->eq_table.eq[MLX4_EQ_ASYNC].actv_ports.ports,
2970 dev->caps.num_ports);
2972 for (i = 0; i < dev->caps.num_comp_vectors + 1; i++) {
2973 if (i == MLX4_EQ_ASYNC)
2976 priv->eq_table.eq[i].irq =
2977 entries[i + 1 - !!(i > MLX4_EQ_ASYNC)].vector;
2979 if (MLX4_IS_LEGACY_EQ_MODE(dev->caps)) {
2980 bitmap_fill(priv->eq_table.eq[i].actv_ports.ports,
2981 dev->caps.num_ports);
2982 /* We don't set affinity hint when there
2987 priv->eq_table.eq[i].actv_ports.ports);
2988 if (mlx4_init_affinity_hint(dev, port + 1, i))
2989 mlx4_warn(dev, "Couldn't init hint cpumask for EQ %d\n",
2992 /* We divide the Eqs evenly between the two ports.
2993 * (dev->caps.num_comp_vectors / dev->caps.num_ports)
2994 * refers to the number of Eqs per port
2995 * (i.e eqs_per_port). Theoretically, we would like to
2996 * write something like (i + 1) % eqs_per_port == 0.
2997 * However, since there's an asynchronous Eq, we have
2998 * to skip over it by comparing this condition to
2999 * !!((i + 1) > MLX4_EQ_ASYNC).
3001 if ((dev->caps.num_comp_vectors > dev->caps.num_ports) &&
3003 (dev->caps.num_comp_vectors / dev->caps.num_ports)) ==
3004 !!((i + 1) > MLX4_EQ_ASYNC))
3005 /* If dev->caps.num_comp_vectors < dev->caps.num_ports,
3006 * everything is shared anyway.
3011 dev->flags |= MLX4_FLAG_MSI_X;
3018 dev->caps.num_comp_vectors = 1;
3020 BUG_ON(MLX4_EQ_ASYNC >= 2);
3021 for (i = 0; i < 2; ++i) {
3022 priv->eq_table.eq[i].irq = dev->persist->pdev->irq;
3023 if (i != MLX4_EQ_ASYNC) {
3024 bitmap_fill(priv->eq_table.eq[i].actv_ports.ports,
3025 dev->caps.num_ports);
3030 static int mlx4_init_port_info(struct mlx4_dev *dev, int port)
3032 struct devlink *devlink = priv_to_devlink(mlx4_priv(dev));
3033 struct mlx4_port_info *info = &mlx4_priv(dev)->port[port];
3036 err = devlink_port_register(devlink, &info->devlink_port, port);
3040 /* Ethernet and IB drivers will normally set the port type,
3041 * but if they are not built set the type now to prevent
3042 * devlink_port_type_warn() from firing.
3044 if (!IS_ENABLED(CONFIG_MLX4_EN) &&
3045 dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH)
3046 devlink_port_type_eth_set(&info->devlink_port, NULL);
3047 else if (!IS_ENABLED(CONFIG_MLX4_INFINIBAND) &&
3048 dev->caps.port_type[port] == MLX4_PORT_TYPE_IB)
3049 devlink_port_type_ib_set(&info->devlink_port, NULL);
3053 if (!mlx4_is_slave(dev)) {
3054 mlx4_init_mac_table(dev, &info->mac_table);
3055 mlx4_init_vlan_table(dev, &info->vlan_table);
3056 mlx4_init_roce_gid_table(dev, &info->gid_table);
3057 info->base_qpn = mlx4_get_base_qpn(dev, port);
3060 sprintf(info->dev_name, "mlx4_port%d", port);
3061 info->port_attr.attr.name = info->dev_name;
3062 if (mlx4_is_mfunc(dev)) {
3063 info->port_attr.attr.mode = 0444;
3065 info->port_attr.attr.mode = 0644;
3066 info->port_attr.store = set_port_type;
3068 info->port_attr.show = show_port_type;
3069 sysfs_attr_init(&info->port_attr.attr);
3071 err = device_create_file(&dev->persist->pdev->dev, &info->port_attr);
3073 mlx4_err(dev, "Failed to create file for port %d\n", port);
3074 devlink_port_unregister(&info->devlink_port);
3079 sprintf(info->dev_mtu_name, "mlx4_port%d_mtu", port);
3080 info->port_mtu_attr.attr.name = info->dev_mtu_name;
3081 if (mlx4_is_mfunc(dev)) {
3082 info->port_mtu_attr.attr.mode = 0444;
3084 info->port_mtu_attr.attr.mode = 0644;
3085 info->port_mtu_attr.store = set_port_ib_mtu;
3087 info->port_mtu_attr.show = show_port_ib_mtu;
3088 sysfs_attr_init(&info->port_mtu_attr.attr);
3090 err = device_create_file(&dev->persist->pdev->dev,
3091 &info->port_mtu_attr);
3093 mlx4_err(dev, "Failed to create mtu file for port %d\n", port);
3094 device_remove_file(&info->dev->persist->pdev->dev,
3096 devlink_port_unregister(&info->devlink_port);
3104 static void mlx4_cleanup_port_info(struct mlx4_port_info *info)
3109 device_remove_file(&info->dev->persist->pdev->dev, &info->port_attr);
3110 device_remove_file(&info->dev->persist->pdev->dev,
3111 &info->port_mtu_attr);
3112 devlink_port_unregister(&info->devlink_port);
3114 #ifdef CONFIG_RFS_ACCEL
3115 free_irq_cpu_rmap(info->rmap);
3120 static int mlx4_init_steering(struct mlx4_dev *dev)
3122 struct mlx4_priv *priv = mlx4_priv(dev);
3123 int num_entries = dev->caps.num_ports;
3126 priv->steer = kcalloc(num_entries, sizeof(struct mlx4_steer),
3131 for (i = 0; i < num_entries; i++)
3132 for (j = 0; j < MLX4_NUM_STEERS; j++) {
3133 INIT_LIST_HEAD(&priv->steer[i].promisc_qps[j]);
3134 INIT_LIST_HEAD(&priv->steer[i].steer_entries[j]);
3139 static void mlx4_clear_steering(struct mlx4_dev *dev)
3141 struct mlx4_priv *priv = mlx4_priv(dev);
3142 struct mlx4_steer_index *entry, *tmp_entry;
3143 struct mlx4_promisc_qp *pqp, *tmp_pqp;
3144 int num_entries = dev->caps.num_ports;
3147 for (i = 0; i < num_entries; i++) {
3148 for (j = 0; j < MLX4_NUM_STEERS; j++) {
3149 list_for_each_entry_safe(pqp, tmp_pqp,
3150 &priv->steer[i].promisc_qps[j],
3152 list_del(&pqp->list);
3155 list_for_each_entry_safe(entry, tmp_entry,
3156 &priv->steer[i].steer_entries[j],
3158 list_del(&entry->list);
3159 list_for_each_entry_safe(pqp, tmp_pqp,
3162 list_del(&pqp->list);
3172 static int extended_func_num(struct pci_dev *pdev)
3174 return PCI_SLOT(pdev->devfn) * 8 + PCI_FUNC(pdev->devfn);
3177 #define MLX4_OWNER_BASE 0x8069c
3178 #define MLX4_OWNER_SIZE 4
3180 static int mlx4_get_ownership(struct mlx4_dev *dev)
3182 void __iomem *owner;
3185 if (pci_channel_offline(dev->persist->pdev))
3188 owner = ioremap(pci_resource_start(dev->persist->pdev, 0) +
3192 mlx4_err(dev, "Failed to obtain ownership bit\n");
3201 static void mlx4_free_ownership(struct mlx4_dev *dev)
3203 void __iomem *owner;
3205 if (pci_channel_offline(dev->persist->pdev))
3208 owner = ioremap(pci_resource_start(dev->persist->pdev, 0) +
3212 mlx4_err(dev, "Failed to obtain ownership bit\n");
3220 #define SRIOV_VALID_STATE(flags) (!!((flags) & MLX4_FLAG_SRIOV) ==\
3221 !!((flags) & MLX4_FLAG_MASTER))
3223 static u64 mlx4_enable_sriov(struct mlx4_dev *dev, struct pci_dev *pdev,
3224 u8 total_vfs, int existing_vfs, int reset_flow)
3226 u64 dev_flags = dev->flags;
3228 int fw_enabled_sriov_vfs = min(pci_sriov_get_totalvfs(pdev),
3232 dev->dev_vfs = kcalloc(total_vfs, sizeof(*dev->dev_vfs),
3239 atomic_inc(&pf_loading);
3240 if (dev->flags & MLX4_FLAG_SRIOV) {
3241 if (existing_vfs != total_vfs) {
3242 mlx4_err(dev, "SR-IOV was already enabled, but with num_vfs (%d) different than requested (%d)\n",
3243 existing_vfs, total_vfs);
3244 total_vfs = existing_vfs;
3248 dev->dev_vfs = kcalloc(total_vfs, sizeof(*dev->dev_vfs), GFP_KERNEL);
3249 if (NULL == dev->dev_vfs) {
3250 mlx4_err(dev, "Failed to allocate memory for VFs\n");
3254 if (!(dev->flags & MLX4_FLAG_SRIOV)) {
3255 if (total_vfs > fw_enabled_sriov_vfs) {
3256 mlx4_err(dev, "requested vfs (%d) > available vfs (%d). Continuing without SR_IOV\n",
3257 total_vfs, fw_enabled_sriov_vfs);
3261 mlx4_warn(dev, "Enabling SR-IOV with %d VFs\n", total_vfs);
3262 err = pci_enable_sriov(pdev, total_vfs);
3265 mlx4_err(dev, "Failed to enable SR-IOV, continuing without SR-IOV (err = %d)\n",
3269 mlx4_warn(dev, "Running in master mode\n");
3270 dev_flags |= MLX4_FLAG_SRIOV |
3272 dev_flags &= ~MLX4_FLAG_SLAVE;
3273 dev->persist->num_vfs = total_vfs;
3278 atomic_dec(&pf_loading);
3280 dev->persist->num_vfs = 0;
3281 kfree(dev->dev_vfs);
3282 dev->dev_vfs = NULL;
3283 return dev_flags & ~MLX4_FLAG_MASTER;
3287 MLX4_DEV_CAP_CHECK_NUM_VFS_ABOVE_64 = -1,
3290 static int mlx4_check_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap,
3293 int requested_vfs = nvfs[0] + nvfs[1] + nvfs[2];
3294 /* Checking for 64 VFs as a limitation of CX2 */
3295 if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_80_VFS) &&
3296 requested_vfs >= 64) {
3297 mlx4_err(dev, "Requested %d VFs, but FW does not support more than 64\n",
3299 return MLX4_DEV_CAP_CHECK_NUM_VFS_ABOVE_64;
3304 static int mlx4_pci_enable_device(struct mlx4_dev *dev)
3306 struct pci_dev *pdev = dev->persist->pdev;
3309 mutex_lock(&dev->persist->pci_status_mutex);
3310 if (dev->persist->pci_status == MLX4_PCI_STATUS_DISABLED) {
3311 err = pci_enable_device(pdev);
3313 dev->persist->pci_status = MLX4_PCI_STATUS_ENABLED;
3315 mutex_unlock(&dev->persist->pci_status_mutex);
3320 static void mlx4_pci_disable_device(struct mlx4_dev *dev)
3322 struct pci_dev *pdev = dev->persist->pdev;
3324 mutex_lock(&dev->persist->pci_status_mutex);
3325 if (dev->persist->pci_status == MLX4_PCI_STATUS_ENABLED) {
3326 pci_disable_device(pdev);
3327 dev->persist->pci_status = MLX4_PCI_STATUS_DISABLED;
3329 mutex_unlock(&dev->persist->pci_status_mutex);
3332 static int mlx4_load_one(struct pci_dev *pdev, int pci_dev_data,
3333 int total_vfs, int *nvfs, struct mlx4_priv *priv,
3336 struct mlx4_dev *dev;
3341 struct mlx4_dev_cap *dev_cap = NULL;
3342 int existing_vfs = 0;
3346 INIT_LIST_HEAD(&priv->ctx_list);
3347 spin_lock_init(&priv->ctx_lock);
3349 mutex_init(&priv->port_mutex);
3350 mutex_init(&priv->bond_mutex);
3352 INIT_LIST_HEAD(&priv->pgdir_list);
3353 mutex_init(&priv->pgdir_mutex);
3354 spin_lock_init(&priv->cmd.context_lock);
3356 INIT_LIST_HEAD(&priv->bf_list);
3357 mutex_init(&priv->bf_mutex);
3359 dev->rev_id = pdev->revision;
3360 dev->numa_node = dev_to_node(&pdev->dev);
3362 /* Detect if this device is a virtual function */
3363 if (pci_dev_data & MLX4_PCI_DEV_IS_VF) {
3364 mlx4_warn(dev, "Detected virtual function - running in slave mode\n");
3365 dev->flags |= MLX4_FLAG_SLAVE;
3367 /* We reset the device and enable SRIOV only for physical
3368 * devices. Try to claim ownership on the device;
3369 * if already taken, skip -- do not allow multiple PFs */
3370 err = mlx4_get_ownership(dev);
3375 mlx4_warn(dev, "Multiple PFs not yet supported - Skipping PF\n");
3380 atomic_set(&priv->opreq_count, 0);
3381 INIT_WORK(&priv->opreq_task, mlx4_opreq_action);
3384 * Now reset the HCA before we touch the PCI capabilities or
3385 * attempt a firmware command, since a boot ROM may have left
3386 * the HCA in an undefined state.
3388 err = mlx4_reset(dev);
3390 mlx4_err(dev, "Failed to reset HCA, aborting\n");
3395 dev->flags = MLX4_FLAG_MASTER;
3396 existing_vfs = pci_num_vf(pdev);
3398 dev->flags |= MLX4_FLAG_SRIOV;
3399 dev->persist->num_vfs = total_vfs;
3403 /* on load remove any previous indication of internal error,
3406 dev->persist->state = MLX4_DEVICE_STATE_UP;
3409 err = mlx4_cmd_init(dev);
3411 mlx4_err(dev, "Failed to init command interface, aborting\n");
3415 /* In slave functions, the communication channel must be initialized
3416 * before posting commands. Also, init num_slaves before calling
3418 if (mlx4_is_mfunc(dev)) {
3419 if (mlx4_is_master(dev)) {
3420 dev->num_slaves = MLX4_MAX_NUM_SLAVES;
3423 dev->num_slaves = 0;
3424 err = mlx4_multi_func_init(dev);
3426 mlx4_err(dev, "Failed to init slave mfunc interface, aborting\n");
3432 err = mlx4_init_fw(dev);
3434 mlx4_err(dev, "Failed to init fw, aborting.\n");
3438 if (mlx4_is_master(dev)) {
3439 /* when we hit the goto slave_start below, dev_cap already initialized */
3441 dev_cap = kzalloc(sizeof(*dev_cap), GFP_KERNEL);
3448 err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
3450 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
3454 if (mlx4_check_dev_cap(dev, dev_cap, nvfs))
3457 if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS)) {
3458 u64 dev_flags = mlx4_enable_sriov(dev, pdev,
3464 mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3465 dev->flags = dev_flags;
3466 if (!SRIOV_VALID_STATE(dev->flags)) {
3467 mlx4_err(dev, "Invalid SRIOV state\n");
3470 err = mlx4_reset(dev);
3472 mlx4_err(dev, "Failed to reset HCA, aborting.\n");
3478 /* Legacy mode FW requires SRIOV to be enabled before
3479 * doing QUERY_DEV_CAP, since max_eq's value is different if
3482 memset(dev_cap, 0, sizeof(*dev_cap));
3483 err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
3485 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
3489 if (mlx4_check_dev_cap(dev, dev_cap, nvfs))
3494 err = mlx4_init_hca(dev);
3496 if (err == -EACCES) {
3497 /* Not primary Physical function
3498 * Running in slave mode */
3499 mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3500 /* We're not a PF */
3501 if (dev->flags & MLX4_FLAG_SRIOV) {
3503 pci_disable_sriov(pdev);
3504 if (mlx4_is_master(dev) && !reset_flow)
3505 atomic_dec(&pf_loading);
3506 dev->flags &= ~MLX4_FLAG_SRIOV;
3508 if (!mlx4_is_slave(dev))
3509 mlx4_free_ownership(dev);
3510 dev->flags |= MLX4_FLAG_SLAVE;
3511 dev->flags &= ~MLX4_FLAG_MASTER;
3517 if (mlx4_is_master(dev) && (dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS)) {
3518 u64 dev_flags = mlx4_enable_sriov(dev, pdev, total_vfs,
3519 existing_vfs, reset_flow);
3521 if ((dev->flags ^ dev_flags) & (MLX4_FLAG_MASTER | MLX4_FLAG_SLAVE)) {
3522 mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_VHCR);
3523 dev->flags = dev_flags;
3524 err = mlx4_cmd_init(dev);
3526 /* Only VHCR is cleaned up, so could still
3529 mlx4_err(dev, "Failed to init VHCR command interface, aborting\n");
3533 dev->flags = dev_flags;
3536 if (!SRIOV_VALID_STATE(dev->flags)) {
3537 mlx4_err(dev, "Invalid SRIOV state\n");
3543 /* check if the device is functioning at its maximum possible speed.
3544 * No return code for this call, just warn the user in case of PCI
3545 * express device capabilities are under-satisfied by the bus.
3547 if (!mlx4_is_slave(dev))
3548 pcie_print_link_status(dev->persist->pdev);
3550 /* In master functions, the communication channel must be initialized
3551 * after obtaining its address from fw */
3552 if (mlx4_is_master(dev)) {
3553 if (dev->caps.num_ports < 2 &&
3557 "Error: Trying to configure VFs on port 2, but HCA has only %d physical ports\n",
3558 dev->caps.num_ports);
3561 memcpy(dev->persist->nvfs, nvfs, sizeof(dev->persist->nvfs));
3564 i < sizeof(dev->persist->nvfs)/
3565 sizeof(dev->persist->nvfs[0]); i++) {
3568 for (j = 0; j < dev->persist->nvfs[i]; ++sum, ++j) {
3569 dev->dev_vfs[sum].min_port = i < 2 ? i + 1 : 1;
3570 dev->dev_vfs[sum].n_ports = i < 2 ? 1 :
3571 dev->caps.num_ports;
3575 /* In master functions, the communication channel
3576 * must be initialized after obtaining its address from fw
3578 err = mlx4_multi_func_init(dev);
3580 mlx4_err(dev, "Failed to init master mfunc interface, aborting.\n");
3585 err = mlx4_alloc_eq_table(dev);
3587 goto err_master_mfunc;
3589 bitmap_zero(priv->msix_ctl.pool_bm, MAX_MSIX);
3590 mutex_init(&priv->msix_ctl.pool_lock);
3592 mlx4_enable_msi_x(dev);
3593 if ((mlx4_is_mfunc(dev)) &&
3594 !(dev->flags & MLX4_FLAG_MSI_X)) {
3596 mlx4_err(dev, "INTx is not supported in multi-function mode, aborting\n");
3600 if (!mlx4_is_slave(dev)) {
3601 err = mlx4_init_steering(dev);
3603 goto err_disable_msix;
3606 mlx4_init_quotas(dev);
3608 err = mlx4_setup_hca(dev);
3609 if (err == -EBUSY && (dev->flags & MLX4_FLAG_MSI_X) &&
3610 !mlx4_is_mfunc(dev)) {
3611 dev->flags &= ~MLX4_FLAG_MSI_X;
3612 dev->caps.num_comp_vectors = 1;
3613 pci_disable_msix(pdev);
3614 err = mlx4_setup_hca(dev);
3620 /* When PF resources are ready arm its comm channel to enable
3623 if (mlx4_is_master(dev)) {
3624 err = mlx4_ARM_COMM_CHANNEL(dev);
3626 mlx4_err(dev, " Failed to arm comm channel eq: %x\n",
3632 for (port = 1; port <= dev->caps.num_ports; port++) {
3633 err = mlx4_init_port_info(dev, port);
3638 priv->v2p.port1 = 1;
3639 priv->v2p.port2 = 2;
3641 err = mlx4_register_device(dev);
3645 mlx4_request_modules(dev);
3647 mlx4_sense_init(dev);
3648 mlx4_start_sense(dev);
3652 if (mlx4_is_master(dev) && dev->persist->num_vfs && !reset_flow)
3653 atomic_dec(&pf_loading);
3659 for (--port; port >= 1; --port)
3660 mlx4_cleanup_port_info(&priv->port[port]);
3662 mlx4_cleanup_default_counters(dev);
3663 if (!mlx4_is_slave(dev))
3664 mlx4_cleanup_counters_table(dev);
3665 mlx4_cleanup_qp_table(dev);
3666 mlx4_cleanup_srq_table(dev);
3667 mlx4_cleanup_cq_table(dev);
3668 mlx4_cmd_use_polling(dev);
3669 mlx4_cleanup_eq_table(dev);
3670 mlx4_cleanup_mcg_table(dev);
3671 mlx4_cleanup_mr_table(dev);
3672 mlx4_cleanup_xrcd_table(dev);
3673 mlx4_cleanup_pd_table(dev);
3674 mlx4_cleanup_uar_table(dev);
3677 if (!mlx4_is_slave(dev))
3678 mlx4_clear_steering(dev);
3681 if (dev->flags & MLX4_FLAG_MSI_X)
3682 pci_disable_msix(pdev);
3685 mlx4_free_eq_table(dev);
3688 if (mlx4_is_master(dev)) {
3689 mlx4_free_resource_tracker(dev, RES_TR_FREE_STRUCTS_ONLY);
3690 mlx4_multi_func_cleanup(dev);
3693 if (mlx4_is_slave(dev))
3694 mlx4_slave_destroy_special_qp_cap(dev);
3697 mlx4_close_hca(dev);
3703 if (mlx4_is_slave(dev))
3704 mlx4_multi_func_cleanup(dev);
3707 mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3710 if (dev->flags & MLX4_FLAG_SRIOV && !existing_vfs) {
3711 pci_disable_sriov(pdev);
3712 dev->flags &= ~MLX4_FLAG_SRIOV;
3715 if (mlx4_is_master(dev) && dev->persist->num_vfs && !reset_flow)
3716 atomic_dec(&pf_loading);
3718 kfree(priv->dev.dev_vfs);
3720 if (!mlx4_is_slave(dev))
3721 mlx4_free_ownership(dev);
3727 static int __mlx4_init_one(struct pci_dev *pdev, int pci_dev_data,
3728 struct mlx4_priv *priv)
3731 int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
3732 int prb_vf[MLX4_MAX_PORTS + 1] = {0, 0, 0};
3733 const int param_map[MLX4_MAX_PORTS + 1][MLX4_MAX_PORTS + 1] = {
3734 {2, 0, 0}, {0, 1, 2}, {0, 1, 2} };
3735 unsigned total_vfs = 0;
3738 pr_info(DRV_NAME ": Initializing %s\n", pci_name(pdev));
3740 err = mlx4_pci_enable_device(&priv->dev);
3742 dev_err(&pdev->dev, "Cannot enable PCI device, aborting\n");
3746 /* Due to requirement that all VFs and the PF are *guaranteed* 2 MACS
3747 * per port, we must limit the number of VFs to 63 (since their are
3750 for (i = 0; i < ARRAY_SIZE(nvfs) && i < num_vfs_argc;
3751 total_vfs += nvfs[param_map[num_vfs_argc - 1][i]], i++) {
3752 nvfs[param_map[num_vfs_argc - 1][i]] = num_vfs[i];
3754 dev_err(&pdev->dev, "num_vfs module parameter cannot be negative\n");
3756 goto err_disable_pdev;
3759 for (i = 0; i < ARRAY_SIZE(prb_vf) && i < probe_vfs_argc;
3761 prb_vf[param_map[probe_vfs_argc - 1][i]] = probe_vf[i];
3762 if (prb_vf[i] < 0 || prb_vf[i] > nvfs[i]) {
3763 dev_err(&pdev->dev, "probe_vf module parameter cannot be negative or greater than num_vfs\n");
3765 goto err_disable_pdev;
3768 if (total_vfs > MLX4_MAX_NUM_VF) {
3770 "Requested more VF's (%d) than allowed by hw (%d)\n",
3771 total_vfs, MLX4_MAX_NUM_VF);
3773 goto err_disable_pdev;
3776 for (i = 0; i < MLX4_MAX_PORTS; i++) {
3777 if (nvfs[i] + nvfs[2] > MLX4_MAX_NUM_VF_P_PORT) {
3779 "Requested more VF's (%d) for port (%d) than allowed by driver (%d)\n",
3780 nvfs[i] + nvfs[2], i + 1,
3781 MLX4_MAX_NUM_VF_P_PORT);
3783 goto err_disable_pdev;
3787 /* Check for BARs. */
3788 if (!(pci_dev_data & MLX4_PCI_DEV_IS_VF) &&
3789 !(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) {
3790 dev_err(&pdev->dev, "Missing DCS, aborting (driver_data: 0x%x, pci_resource_flags(pdev, 0):0x%lx)\n",
3791 pci_dev_data, pci_resource_flags(pdev, 0));
3793 goto err_disable_pdev;
3795 if (!(pci_resource_flags(pdev, 2) & IORESOURCE_MEM)) {
3796 dev_err(&pdev->dev, "Missing UAR, aborting\n");
3798 goto err_disable_pdev;
3801 err = pci_request_regions(pdev, DRV_NAME);
3803 dev_err(&pdev->dev, "Couldn't get PCI resources, aborting\n");
3804 goto err_disable_pdev;
3807 pci_set_master(pdev);
3809 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
3811 dev_warn(&pdev->dev, "Warning: couldn't set 64-bit PCI DMA mask\n");
3812 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
3814 dev_err(&pdev->dev, "Can't set PCI DMA mask, aborting\n");
3815 goto err_release_regions;
3818 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
3820 dev_warn(&pdev->dev, "Warning: couldn't set 64-bit consistent PCI DMA mask\n");
3821 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
3823 dev_err(&pdev->dev, "Can't set consistent PCI DMA mask, aborting\n");
3824 goto err_release_regions;
3828 /* Allow large DMA segments, up to the firmware limit of 1 GB */
3829 dma_set_max_seg_size(&pdev->dev, 1024 * 1024 * 1024);
3830 /* Detect if this device is a virtual function */
3831 if (pci_dev_data & MLX4_PCI_DEV_IS_VF) {
3832 /* When acting as pf, we normally skip vfs unless explicitly
3833 * requested to probe them.
3836 unsigned vfs_offset = 0;
3838 for (i = 0; i < ARRAY_SIZE(nvfs) &&
3839 vfs_offset + nvfs[i] < extended_func_num(pdev);
3840 vfs_offset += nvfs[i], i++)
3842 if (i == ARRAY_SIZE(nvfs)) {
3844 goto err_release_regions;
3846 if ((extended_func_num(pdev) - vfs_offset)
3848 dev_warn(&pdev->dev, "Skipping virtual function:%d\n",
3849 extended_func_num(pdev));
3851 goto err_release_regions;
3856 err = mlx4_crdump_init(&priv->dev);
3858 goto err_release_regions;
3860 err = mlx4_catas_init(&priv->dev);
3864 err = mlx4_load_one(pdev, pci_dev_data, total_vfs, nvfs, priv, 0);
3871 mlx4_catas_end(&priv->dev);
3874 mlx4_crdump_end(&priv->dev);
3876 err_release_regions:
3877 pci_release_regions(pdev);
3880 mlx4_pci_disable_device(&priv->dev);
3884 static int mlx4_devlink_port_type_set(struct devlink_port *devlink_port,
3885 enum devlink_port_type port_type)
3887 struct mlx4_port_info *info = container_of(devlink_port,
3888 struct mlx4_port_info,
3890 enum mlx4_port_type mlx4_port_type;
3892 switch (port_type) {
3893 case DEVLINK_PORT_TYPE_AUTO:
3894 mlx4_port_type = MLX4_PORT_TYPE_AUTO;
3896 case DEVLINK_PORT_TYPE_ETH:
3897 mlx4_port_type = MLX4_PORT_TYPE_ETH;
3899 case DEVLINK_PORT_TYPE_IB:
3900 mlx4_port_type = MLX4_PORT_TYPE_IB;
3906 return __set_port_type(info, mlx4_port_type);
3909 static void mlx4_devlink_param_load_driverinit_values(struct devlink *devlink)
3911 struct mlx4_priv *priv = devlink_priv(devlink);
3912 struct mlx4_dev *dev = &priv->dev;
3913 struct mlx4_fw_crdump *crdump = &dev->persist->crdump;
3914 union devlink_param_value saved_value;
3917 err = devlink_param_driverinit_value_get(devlink,
3918 DEVLINK_PARAM_GENERIC_ID_INT_ERR_RESET,
3920 if (!err && mlx4_internal_err_reset != saved_value.vbool) {
3921 mlx4_internal_err_reset = saved_value.vbool;
3922 /* Notify on value changed on runtime configuration mode */
3923 devlink_param_value_changed(devlink,
3924 DEVLINK_PARAM_GENERIC_ID_INT_ERR_RESET);
3926 err = devlink_param_driverinit_value_get(devlink,
3927 DEVLINK_PARAM_GENERIC_ID_MAX_MACS,
3930 log_num_mac = order_base_2(saved_value.vu32);
3931 err = devlink_param_driverinit_value_get(devlink,
3932 MLX4_DEVLINK_PARAM_ID_ENABLE_64B_CQE_EQE,
3935 enable_64b_cqe_eqe = saved_value.vbool;
3936 err = devlink_param_driverinit_value_get(devlink,
3937 MLX4_DEVLINK_PARAM_ID_ENABLE_4K_UAR,
3940 enable_4k_uar = saved_value.vbool;
3941 err = devlink_param_driverinit_value_get(devlink,
3942 DEVLINK_PARAM_GENERIC_ID_REGION_SNAPSHOT,
3944 if (!err && crdump->snapshot_enable != saved_value.vbool) {
3945 crdump->snapshot_enable = saved_value.vbool;
3946 devlink_param_value_changed(devlink,
3947 DEVLINK_PARAM_GENERIC_ID_REGION_SNAPSHOT);
3951 static void mlx4_restart_one_down(struct pci_dev *pdev);
3952 static int mlx4_restart_one_up(struct pci_dev *pdev, bool reload,
3953 struct devlink *devlink);
3955 static int mlx4_devlink_reload_down(struct devlink *devlink, bool netns_change,
3956 enum devlink_reload_action action,
3957 enum devlink_reload_limit limit,
3958 struct netlink_ext_ack *extack)
3960 struct mlx4_priv *priv = devlink_priv(devlink);
3961 struct mlx4_dev *dev = &priv->dev;
3962 struct mlx4_dev_persistent *persist = dev->persist;
3965 NL_SET_ERR_MSG_MOD(extack, "Namespace change is not supported");
3968 if (persist->num_vfs)
3969 mlx4_warn(persist->dev, "Reload performed on PF, will cause reset on operating Virtual Functions\n");
3970 mlx4_restart_one_down(persist->pdev);
3974 static int mlx4_devlink_reload_up(struct devlink *devlink, enum devlink_reload_action action,
3975 enum devlink_reload_limit limit, u32 *actions_performed,
3976 struct netlink_ext_ack *extack)
3978 struct mlx4_priv *priv = devlink_priv(devlink);
3979 struct mlx4_dev *dev = &priv->dev;
3980 struct mlx4_dev_persistent *persist = dev->persist;
3983 *actions_performed = BIT(DEVLINK_RELOAD_ACTION_DRIVER_REINIT);
3984 err = mlx4_restart_one_up(persist->pdev, true, devlink);
3986 mlx4_err(persist->dev, "mlx4_restart_one_up failed, ret=%d\n",
3992 static const struct devlink_ops mlx4_devlink_ops = {
3993 .port_type_set = mlx4_devlink_port_type_set,
3994 .reload_actions = BIT(DEVLINK_RELOAD_ACTION_DRIVER_REINIT),
3995 .reload_down = mlx4_devlink_reload_down,
3996 .reload_up = mlx4_devlink_reload_up,
3999 static int mlx4_init_one(struct pci_dev *pdev, const struct pci_device_id *id)
4001 struct devlink *devlink;
4002 struct mlx4_priv *priv;
4003 struct mlx4_dev *dev;
4006 printk_once(KERN_INFO "%s", mlx4_version);
4008 devlink = devlink_alloc(&mlx4_devlink_ops, sizeof(*priv), &pdev->dev);
4011 priv = devlink_priv(devlink);
4014 dev->persist = kzalloc(sizeof(*dev->persist), GFP_KERNEL);
4015 if (!dev->persist) {
4017 goto err_devlink_free;
4019 dev->persist->pdev = pdev;
4020 dev->persist->dev = dev;
4021 pci_set_drvdata(pdev, dev->persist);
4022 priv->pci_dev_data = id->driver_data;
4023 mutex_init(&dev->persist->device_state_mutex);
4024 mutex_init(&dev->persist->interface_state_mutex);
4025 mutex_init(&dev->persist->pci_status_mutex);
4027 ret = devlink_register(devlink);
4029 goto err_persist_free;
4030 ret = devlink_params_register(devlink, mlx4_devlink_params,
4031 ARRAY_SIZE(mlx4_devlink_params));
4033 goto err_devlink_unregister;
4034 mlx4_devlink_set_params_init_values(devlink);
4035 ret = __mlx4_init_one(pdev, id->driver_data, priv);
4037 goto err_params_unregister;
4039 devlink_params_publish(devlink);
4040 devlink_reload_enable(devlink);
4041 pci_save_state(pdev);
4044 err_params_unregister:
4045 devlink_params_unregister(devlink, mlx4_devlink_params,
4046 ARRAY_SIZE(mlx4_devlink_params));
4047 err_devlink_unregister:
4048 devlink_unregister(devlink);
4050 kfree(dev->persist);
4052 devlink_free(devlink);
4056 static void mlx4_clean_dev(struct mlx4_dev *dev)
4058 struct mlx4_dev_persistent *persist = dev->persist;
4059 struct mlx4_priv *priv = mlx4_priv(dev);
4060 unsigned long flags = (dev->flags & RESET_PERSIST_MASK_FLAGS);
4062 memset(priv, 0, sizeof(*priv));
4063 priv->dev.persist = persist;
4064 priv->dev.flags = flags;
4067 static void mlx4_unload_one(struct pci_dev *pdev)
4069 struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4070 struct mlx4_dev *dev = persist->dev;
4071 struct mlx4_priv *priv = mlx4_priv(dev);
4078 /* saving current ports type for further use */
4079 for (i = 0; i < dev->caps.num_ports; i++) {
4080 dev->persist->curr_port_type[i] = dev->caps.port_type[i + 1];
4081 dev->persist->curr_port_poss_type[i] = dev->caps.
4082 possible_type[i + 1];
4085 pci_dev_data = priv->pci_dev_data;
4087 mlx4_stop_sense(dev);
4088 mlx4_unregister_device(dev);
4090 for (p = 1; p <= dev->caps.num_ports; p++) {
4091 mlx4_cleanup_port_info(&priv->port[p]);
4092 mlx4_CLOSE_PORT(dev, p);
4095 if (mlx4_is_master(dev))
4096 mlx4_free_resource_tracker(dev,
4097 RES_TR_FREE_SLAVES_ONLY);
4099 mlx4_cleanup_default_counters(dev);
4100 if (!mlx4_is_slave(dev))
4101 mlx4_cleanup_counters_table(dev);
4102 mlx4_cleanup_qp_table(dev);
4103 mlx4_cleanup_srq_table(dev);
4104 mlx4_cleanup_cq_table(dev);
4105 mlx4_cmd_use_polling(dev);
4106 mlx4_cleanup_eq_table(dev);
4107 mlx4_cleanup_mcg_table(dev);
4108 mlx4_cleanup_mr_table(dev);
4109 mlx4_cleanup_xrcd_table(dev);
4110 mlx4_cleanup_pd_table(dev);
4112 if (mlx4_is_master(dev))
4113 mlx4_free_resource_tracker(dev,
4114 RES_TR_FREE_STRUCTS_ONLY);
4117 mlx4_uar_free(dev, &priv->driver_uar);
4118 mlx4_cleanup_uar_table(dev);
4119 if (!mlx4_is_slave(dev))
4120 mlx4_clear_steering(dev);
4121 mlx4_free_eq_table(dev);
4122 if (mlx4_is_master(dev))
4123 mlx4_multi_func_cleanup(dev);
4124 mlx4_close_hca(dev);
4126 if (mlx4_is_slave(dev))
4127 mlx4_multi_func_cleanup(dev);
4128 mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
4130 if (dev->flags & MLX4_FLAG_MSI_X)
4131 pci_disable_msix(pdev);
4133 if (!mlx4_is_slave(dev))
4134 mlx4_free_ownership(dev);
4136 mlx4_slave_destroy_special_qp_cap(dev);
4137 kfree(dev->dev_vfs);
4139 mlx4_clean_dev(dev);
4140 priv->pci_dev_data = pci_dev_data;
4144 static void mlx4_remove_one(struct pci_dev *pdev)
4146 struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4147 struct mlx4_dev *dev = persist->dev;
4148 struct mlx4_priv *priv = mlx4_priv(dev);
4149 struct devlink *devlink = priv_to_devlink(priv);
4152 devlink_reload_disable(devlink);
4154 if (mlx4_is_slave(dev))
4155 persist->interface_state |= MLX4_INTERFACE_STATE_NOWAIT;
4157 mutex_lock(&persist->interface_state_mutex);
4158 persist->interface_state |= MLX4_INTERFACE_STATE_DELETION;
4159 mutex_unlock(&persist->interface_state_mutex);
4161 /* Disabling SR-IOV is not allowed while there are active vf's */
4162 if (mlx4_is_master(dev) && dev->flags & MLX4_FLAG_SRIOV) {
4163 active_vfs = mlx4_how_many_lives_vf(dev);
4165 pr_warn("Removing PF when there are active VF's !!\n");
4166 pr_warn("Will not disable SR-IOV.\n");
4170 /* device marked to be under deletion running now without the lock
4171 * letting other tasks to be terminated
4173 if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
4174 mlx4_unload_one(pdev);
4176 mlx4_info(dev, "%s: interface is down\n", __func__);
4177 mlx4_catas_end(dev);
4178 mlx4_crdump_end(dev);
4179 if (dev->flags & MLX4_FLAG_SRIOV && !active_vfs) {
4180 mlx4_warn(dev, "Disabling SR-IOV\n");
4181 pci_disable_sriov(pdev);
4184 pci_release_regions(pdev);
4185 mlx4_pci_disable_device(dev);
4186 devlink_params_unregister(devlink, mlx4_devlink_params,
4187 ARRAY_SIZE(mlx4_devlink_params));
4188 devlink_unregister(devlink);
4189 kfree(dev->persist);
4190 devlink_free(devlink);
4193 static int restore_current_port_types(struct mlx4_dev *dev,
4194 enum mlx4_port_type *types,
4195 enum mlx4_port_type *poss_types)
4197 struct mlx4_priv *priv = mlx4_priv(dev);
4200 mlx4_stop_sense(dev);
4202 mutex_lock(&priv->port_mutex);
4203 for (i = 0; i < dev->caps.num_ports; i++)
4204 dev->caps.possible_type[i + 1] = poss_types[i];
4205 err = mlx4_change_port_types(dev, types);
4206 mlx4_start_sense(dev);
4207 mutex_unlock(&priv->port_mutex);
4212 static void mlx4_restart_one_down(struct pci_dev *pdev)
4214 mlx4_unload_one(pdev);
4217 static int mlx4_restart_one_up(struct pci_dev *pdev, bool reload,
4218 struct devlink *devlink)
4220 struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4221 struct mlx4_dev *dev = persist->dev;
4222 struct mlx4_priv *priv = mlx4_priv(dev);
4223 int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
4224 int pci_dev_data, err, total_vfs;
4226 pci_dev_data = priv->pci_dev_data;
4227 total_vfs = dev->persist->num_vfs;
4228 memcpy(nvfs, dev->persist->nvfs, sizeof(dev->persist->nvfs));
4231 mlx4_devlink_param_load_driverinit_values(devlink);
4232 err = mlx4_load_one(pdev, pci_dev_data, total_vfs, nvfs, priv, 1);
4234 mlx4_err(dev, "%s: ERROR: mlx4_load_one failed, pci_name=%s, err=%d\n",
4235 __func__, pci_name(pdev), err);
4239 err = restore_current_port_types(dev, dev->persist->curr_port_type,
4240 dev->persist->curr_port_poss_type);
4242 mlx4_err(dev, "could not restore original port types (%d)\n",
4248 int mlx4_restart_one(struct pci_dev *pdev)
4250 mlx4_restart_one_down(pdev);
4251 return mlx4_restart_one_up(pdev, false, NULL);
4254 #define MLX_SP(id) { PCI_VDEVICE(MELLANOX, id), MLX4_PCI_DEV_FORCE_SENSE_PORT }
4255 #define MLX_VF(id) { PCI_VDEVICE(MELLANOX, id), MLX4_PCI_DEV_IS_VF }
4256 #define MLX_GN(id) { PCI_VDEVICE(MELLANOX, id), 0 }
4258 static const struct pci_device_id mlx4_pci_table[] = {
4259 #ifdef CONFIG_MLX4_CORE_GEN2
4260 /* MT25408 "Hermon" */
4261 MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_SDR), /* SDR */
4262 MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_DDR), /* DDR */
4263 MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_QDR), /* QDR */
4264 MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_DDR_GEN2), /* DDR Gen2 */
4265 MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_QDR_GEN2), /* QDR Gen2 */
4266 MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_EN), /* EN 10GigE */
4267 MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_EN_GEN2), /* EN 10GigE Gen2 */
4268 /* MT25458 ConnectX EN 10GBASE-T */
4269 MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX_EN),
4270 MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX_EN_T_GEN2), /* Gen2 */
4271 /* MT26468 ConnectX EN 10GigE PCIe Gen2*/
4272 MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX_EN_GEN2),
4273 /* MT26438 ConnectX EN 40GigE PCIe Gen2 5GT/s */
4274 MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX_EN_5_GEN2),
4275 /* MT26478 ConnectX2 40GigE PCIe Gen2 */
4276 MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX2),
4277 /* MT25400 Family [ConnectX-2] */
4278 MLX_VF(0x1002), /* Virtual Function */
4279 #endif /* CONFIG_MLX4_CORE_GEN2 */
4280 /* MT27500 Family [ConnectX-3] */
4281 MLX_GN(PCI_DEVICE_ID_MELLANOX_CONNECTX3),
4282 MLX_VF(0x1004), /* Virtual Function */
4283 MLX_GN(0x1005), /* MT27510 Family */
4284 MLX_GN(0x1006), /* MT27511 Family */
4285 MLX_GN(PCI_DEVICE_ID_MELLANOX_CONNECTX3_PRO), /* MT27520 Family */
4286 MLX_GN(0x1008), /* MT27521 Family */
4287 MLX_GN(0x1009), /* MT27530 Family */
4288 MLX_GN(0x100a), /* MT27531 Family */
4289 MLX_GN(0x100b), /* MT27540 Family */
4290 MLX_GN(0x100c), /* MT27541 Family */
4291 MLX_GN(0x100d), /* MT27550 Family */
4292 MLX_GN(0x100e), /* MT27551 Family */
4293 MLX_GN(0x100f), /* MT27560 Family */
4294 MLX_GN(0x1010), /* MT27561 Family */
4297 * See the mellanox_check_broken_intx_masking() quirk when
4304 MODULE_DEVICE_TABLE(pci, mlx4_pci_table);
4306 static pci_ers_result_t mlx4_pci_err_detected(struct pci_dev *pdev,
4307 pci_channel_state_t state)
4309 struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4311 mlx4_err(persist->dev, "mlx4_pci_err_detected was called\n");
4312 mlx4_enter_error_state(persist);
4314 mutex_lock(&persist->interface_state_mutex);
4315 if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
4316 mlx4_unload_one(pdev);
4318 mutex_unlock(&persist->interface_state_mutex);
4319 if (state == pci_channel_io_perm_failure)
4320 return PCI_ERS_RESULT_DISCONNECT;
4322 mlx4_pci_disable_device(persist->dev);
4323 return PCI_ERS_RESULT_NEED_RESET;
4326 static pci_ers_result_t mlx4_pci_slot_reset(struct pci_dev *pdev)
4328 struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4329 struct mlx4_dev *dev = persist->dev;
4332 mlx4_err(dev, "mlx4_pci_slot_reset was called\n");
4333 err = mlx4_pci_enable_device(dev);
4335 mlx4_err(dev, "Can not re-enable device, err=%d\n", err);
4336 return PCI_ERS_RESULT_DISCONNECT;
4339 pci_set_master(pdev);
4340 pci_restore_state(pdev);
4341 pci_save_state(pdev);
4342 return PCI_ERS_RESULT_RECOVERED;
4345 static void mlx4_pci_resume(struct pci_dev *pdev)
4347 struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4348 struct mlx4_dev *dev = persist->dev;
4349 struct mlx4_priv *priv = mlx4_priv(dev);
4350 int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
4354 mlx4_err(dev, "%s was called\n", __func__);
4355 total_vfs = dev->persist->num_vfs;
4356 memcpy(nvfs, dev->persist->nvfs, sizeof(dev->persist->nvfs));
4358 mutex_lock(&persist->interface_state_mutex);
4359 if (!(persist->interface_state & MLX4_INTERFACE_STATE_UP)) {
4360 err = mlx4_load_one(pdev, priv->pci_dev_data, total_vfs, nvfs,
4363 mlx4_err(dev, "%s: mlx4_load_one failed, err=%d\n",
4368 err = restore_current_port_types(dev, dev->persist->
4369 curr_port_type, dev->persist->
4370 curr_port_poss_type);
4372 mlx4_err(dev, "could not restore original port types (%d)\n", err);
4375 mutex_unlock(&persist->interface_state_mutex);
4379 static void mlx4_shutdown(struct pci_dev *pdev)
4381 struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4382 struct mlx4_dev *dev = persist->dev;
4384 mlx4_info(persist->dev, "mlx4_shutdown was called\n");
4385 mutex_lock(&persist->interface_state_mutex);
4386 if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
4387 mlx4_unload_one(pdev);
4388 mutex_unlock(&persist->interface_state_mutex);
4389 mlx4_pci_disable_device(dev);
4392 static const struct pci_error_handlers mlx4_err_handler = {
4393 .error_detected = mlx4_pci_err_detected,
4394 .slot_reset = mlx4_pci_slot_reset,
4395 .resume = mlx4_pci_resume,
4398 static int __maybe_unused mlx4_suspend(struct device *dev_d)
4400 struct pci_dev *pdev = to_pci_dev(dev_d);
4401 struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4402 struct mlx4_dev *dev = persist->dev;
4404 mlx4_err(dev, "suspend was called\n");
4405 mutex_lock(&persist->interface_state_mutex);
4406 if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
4407 mlx4_unload_one(pdev);
4408 mutex_unlock(&persist->interface_state_mutex);
4413 static int __maybe_unused mlx4_resume(struct device *dev_d)
4415 struct pci_dev *pdev = to_pci_dev(dev_d);
4416 struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4417 struct mlx4_dev *dev = persist->dev;
4418 struct mlx4_priv *priv = mlx4_priv(dev);
4419 int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
4423 mlx4_err(dev, "resume was called\n");
4424 total_vfs = dev->persist->num_vfs;
4425 memcpy(nvfs, dev->persist->nvfs, sizeof(dev->persist->nvfs));
4427 mutex_lock(&persist->interface_state_mutex);
4428 if (!(persist->interface_state & MLX4_INTERFACE_STATE_UP)) {
4429 ret = mlx4_load_one(pdev, priv->pci_dev_data, total_vfs,
4432 ret = restore_current_port_types(dev,
4433 dev->persist->curr_port_type,
4434 dev->persist->curr_port_poss_type);
4436 mlx4_err(dev, "resume: could not restore original port types (%d)\n", ret);
4439 mutex_unlock(&persist->interface_state_mutex);
4444 static SIMPLE_DEV_PM_OPS(mlx4_pm_ops, mlx4_suspend, mlx4_resume);
4446 static struct pci_driver mlx4_driver = {
4448 .id_table = mlx4_pci_table,
4449 .probe = mlx4_init_one,
4450 .shutdown = mlx4_shutdown,
4451 .remove = mlx4_remove_one,
4452 .driver.pm = &mlx4_pm_ops,
4453 .err_handler = &mlx4_err_handler,
4456 static int __init mlx4_verify_params(void)
4459 pr_warn("mlx4_core: bad msi_x: %d\n", msi_x);
4463 if ((log_num_mac < 0) || (log_num_mac > 7)) {
4464 pr_warn("mlx4_core: bad num_mac: %d\n", log_num_mac);
4468 if (log_num_vlan != 0)
4469 pr_warn("mlx4_core: log_num_vlan - obsolete module param, using %d\n",
4470 MLX4_LOG_NUM_VLANS);
4473 pr_warn("mlx4_core: use_prio - obsolete module param, ignored\n");
4475 if ((log_mtts_per_seg < 0) || (log_mtts_per_seg > 7)) {
4476 pr_warn("mlx4_core: bad log_mtts_per_seg: %d\n",
4481 /* Check if module param for ports type has legal combination */
4482 if (port_type_array[0] == false && port_type_array[1] == true) {
4483 pr_warn("Module parameter configuration ETH/IB is not supported. Switching to default configuration IB/IB\n");
4484 port_type_array[0] = true;
4487 if (mlx4_log_num_mgm_entry_size < -7 ||
4488 (mlx4_log_num_mgm_entry_size > 0 &&
4489 (mlx4_log_num_mgm_entry_size < MLX4_MIN_MGM_LOG_ENTRY_SIZE ||
4490 mlx4_log_num_mgm_entry_size > MLX4_MAX_MGM_LOG_ENTRY_SIZE))) {
4491 pr_warn("mlx4_core: mlx4_log_num_mgm_entry_size (%d) not in legal range (-7..0 or %d..%d)\n",
4492 mlx4_log_num_mgm_entry_size,
4493 MLX4_MIN_MGM_LOG_ENTRY_SIZE,
4494 MLX4_MAX_MGM_LOG_ENTRY_SIZE);
4501 static int __init mlx4_init(void)
4505 if (mlx4_verify_params())
4509 mlx4_wq = create_singlethread_workqueue("mlx4");
4513 ret = pci_register_driver(&mlx4_driver);
4515 destroy_workqueue(mlx4_wq);
4516 return ret < 0 ? ret : 0;
4519 static void __exit mlx4_cleanup(void)
4521 pci_unregister_driver(&mlx4_driver);
4522 destroy_workqueue(mlx4_wq);
4525 module_init(mlx4_init);
4526 module_exit(mlx4_cleanup);