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/etherdevice.h>
46 #include <net/devlink.h>
48 #include <uapi/rdma/mlx4-abi.h>
49 #include <linux/mlx4/device.h>
50 #include <linux/mlx4/doorbell.h>
56 MODULE_AUTHOR("Roland Dreier");
57 MODULE_DESCRIPTION("Mellanox ConnectX HCA low-level driver");
58 MODULE_LICENSE("Dual BSD/GPL");
59 MODULE_VERSION(DRV_VERSION);
61 struct workqueue_struct *mlx4_wq;
63 #ifdef CONFIG_MLX4_DEBUG
65 int mlx4_debug_level; /* 0 by default */
66 module_param_named(debug_level, mlx4_debug_level, int, 0644);
67 MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0");
69 #endif /* CONFIG_MLX4_DEBUG */
74 module_param(msi_x, int, 0444);
75 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");
77 #else /* CONFIG_PCI_MSI */
81 #endif /* CONFIG_PCI_MSI */
83 static uint8_t num_vfs[3] = {0, 0, 0};
84 static int num_vfs_argc;
85 module_param_array(num_vfs, byte, &num_vfs_argc, 0444);
86 MODULE_PARM_DESC(num_vfs, "enable #num_vfs functions if num_vfs > 0\n"
87 "num_vfs=port1,port2,port1+2");
89 static uint8_t probe_vf[3] = {0, 0, 0};
90 static int probe_vfs_argc;
91 module_param_array(probe_vf, byte, &probe_vfs_argc, 0444);
92 MODULE_PARM_DESC(probe_vf, "number of vfs to probe by pf driver (num_vfs > 0)\n"
93 "probe_vf=port1,port2,port1+2");
95 static int mlx4_log_num_mgm_entry_size = MLX4_DEFAULT_MGM_LOG_ENTRY_SIZE;
96 module_param_named(log_num_mgm_entry_size,
97 mlx4_log_num_mgm_entry_size, int, 0444);
98 MODULE_PARM_DESC(log_num_mgm_entry_size, "log mgm size, that defines the num"
99 " of qp per mcg, for example:"
100 " 10 gives 248.range: 7 <="
101 " log_num_mgm_entry_size <= 12."
102 " To activate device managed"
103 " flow steering when available, set to -1");
105 static bool enable_64b_cqe_eqe = true;
106 module_param(enable_64b_cqe_eqe, bool, 0444);
107 MODULE_PARM_DESC(enable_64b_cqe_eqe,
108 "Enable 64 byte CQEs/EQEs when the FW supports this (default: True)");
110 static bool enable_4k_uar;
111 module_param(enable_4k_uar, bool, 0444);
112 MODULE_PARM_DESC(enable_4k_uar,
113 "Enable using 4K UAR. Should not be enabled if have VFs which do not support 4K UARs (default: false)");
115 #define PF_CONTEXT_BEHAVIOUR_MASK (MLX4_FUNC_CAP_64B_EQE_CQE | \
116 MLX4_FUNC_CAP_EQE_CQE_STRIDE | \
117 MLX4_FUNC_CAP_DMFS_A0_STATIC)
119 #define RESET_PERSIST_MASK_FLAGS (MLX4_FLAG_SRIOV)
121 static char mlx4_version[] =
122 DRV_NAME ": Mellanox ConnectX core driver v"
125 static const struct mlx4_profile default_profile = {
128 .rdmarc_per_qp = 1 << 4,
132 .num_mtt = 1 << 20, /* It is really num mtt segements */
135 static const struct mlx4_profile low_mem_profile = {
138 .rdmarc_per_qp = 1 << 4,
145 static int log_num_mac = 7;
146 module_param_named(log_num_mac, log_num_mac, int, 0444);
147 MODULE_PARM_DESC(log_num_mac, "Log2 max number of MACs per ETH port (1-7)");
149 static int log_num_vlan;
150 module_param_named(log_num_vlan, log_num_vlan, int, 0444);
151 MODULE_PARM_DESC(log_num_vlan, "Log2 max number of VLANs per ETH port (0-7)");
152 /* Log2 max number of VLANs per ETH port (0-7) */
153 #define MLX4_LOG_NUM_VLANS 7
154 #define MLX4_MIN_LOG_NUM_VLANS 0
155 #define MLX4_MIN_LOG_NUM_MAC 1
157 static bool use_prio;
158 module_param_named(use_prio, use_prio, bool, 0444);
159 MODULE_PARM_DESC(use_prio, "Enable steering by VLAN priority on ETH ports (deprecated)");
161 int log_mtts_per_seg = ilog2(1);
162 module_param_named(log_mtts_per_seg, log_mtts_per_seg, int, 0444);
163 MODULE_PARM_DESC(log_mtts_per_seg, "Log2 number of MTT entries per segment "
164 "(0-7) (default: 0)");
166 static int port_type_array[2] = {MLX4_PORT_TYPE_NONE, MLX4_PORT_TYPE_NONE};
167 static int arr_argc = 2;
168 module_param_array(port_type_array, int, &arr_argc, 0444);
169 MODULE_PARM_DESC(port_type_array, "Array of port types: HW_DEFAULT (0) is default "
170 "1 for IB, 2 for Ethernet");
172 struct mlx4_port_config {
173 struct list_head list;
174 enum mlx4_port_type port_type[MLX4_MAX_PORTS + 1];
175 struct pci_dev *pdev;
178 static atomic_t pf_loading = ATOMIC_INIT(0);
180 static int mlx4_devlink_ierr_reset_get(struct devlink *devlink, u32 id,
181 struct devlink_param_gset_ctx *ctx)
183 ctx->val.vbool = !!mlx4_internal_err_reset;
187 static int mlx4_devlink_ierr_reset_set(struct devlink *devlink, u32 id,
188 struct devlink_param_gset_ctx *ctx)
190 mlx4_internal_err_reset = ctx->val.vbool;
194 static int mlx4_devlink_crdump_snapshot_get(struct devlink *devlink, u32 id,
195 struct devlink_param_gset_ctx *ctx)
197 struct mlx4_priv *priv = devlink_priv(devlink);
198 struct mlx4_dev *dev = &priv->dev;
200 ctx->val.vbool = dev->persist->crdump.snapshot_enable;
204 static int mlx4_devlink_crdump_snapshot_set(struct devlink *devlink, u32 id,
205 struct devlink_param_gset_ctx *ctx)
207 struct mlx4_priv *priv = devlink_priv(devlink);
208 struct mlx4_dev *dev = &priv->dev;
210 dev->persist->crdump.snapshot_enable = ctx->val.vbool;
215 mlx4_devlink_max_macs_validate(struct devlink *devlink, u32 id,
216 union devlink_param_value val,
217 struct netlink_ext_ack *extack)
219 u32 value = val.vu32;
221 if (value < 1 || value > 128)
224 if (!is_power_of_2(value)) {
225 NL_SET_ERR_MSG_MOD(extack, "max_macs supported must be power of 2");
232 enum mlx4_devlink_param_id {
233 MLX4_DEVLINK_PARAM_ID_BASE = DEVLINK_PARAM_GENERIC_ID_MAX,
234 MLX4_DEVLINK_PARAM_ID_ENABLE_64B_CQE_EQE,
235 MLX4_DEVLINK_PARAM_ID_ENABLE_4K_UAR,
238 static const struct devlink_param mlx4_devlink_params[] = {
239 DEVLINK_PARAM_GENERIC(INT_ERR_RESET,
240 BIT(DEVLINK_PARAM_CMODE_RUNTIME) |
241 BIT(DEVLINK_PARAM_CMODE_DRIVERINIT),
242 mlx4_devlink_ierr_reset_get,
243 mlx4_devlink_ierr_reset_set, NULL),
244 DEVLINK_PARAM_GENERIC(MAX_MACS,
245 BIT(DEVLINK_PARAM_CMODE_DRIVERINIT),
246 NULL, NULL, mlx4_devlink_max_macs_validate),
247 DEVLINK_PARAM_GENERIC(REGION_SNAPSHOT,
248 BIT(DEVLINK_PARAM_CMODE_RUNTIME) |
249 BIT(DEVLINK_PARAM_CMODE_DRIVERINIT),
250 mlx4_devlink_crdump_snapshot_get,
251 mlx4_devlink_crdump_snapshot_set, NULL),
252 DEVLINK_PARAM_DRIVER(MLX4_DEVLINK_PARAM_ID_ENABLE_64B_CQE_EQE,
253 "enable_64b_cqe_eqe", DEVLINK_PARAM_TYPE_BOOL,
254 BIT(DEVLINK_PARAM_CMODE_DRIVERINIT),
256 DEVLINK_PARAM_DRIVER(MLX4_DEVLINK_PARAM_ID_ENABLE_4K_UAR,
257 "enable_4k_uar", DEVLINK_PARAM_TYPE_BOOL,
258 BIT(DEVLINK_PARAM_CMODE_DRIVERINIT),
262 static void mlx4_devlink_set_params_init_values(struct devlink *devlink)
264 union devlink_param_value value;
266 value.vbool = !!mlx4_internal_err_reset;
267 devl_param_driverinit_value_set(devlink,
268 DEVLINK_PARAM_GENERIC_ID_INT_ERR_RESET,
271 value.vu32 = 1UL << log_num_mac;
272 devl_param_driverinit_value_set(devlink,
273 DEVLINK_PARAM_GENERIC_ID_MAX_MACS,
276 value.vbool = enable_64b_cqe_eqe;
277 devl_param_driverinit_value_set(devlink,
278 MLX4_DEVLINK_PARAM_ID_ENABLE_64B_CQE_EQE,
281 value.vbool = enable_4k_uar;
282 devl_param_driverinit_value_set(devlink,
283 MLX4_DEVLINK_PARAM_ID_ENABLE_4K_UAR,
287 devl_param_driverinit_value_set(devlink,
288 DEVLINK_PARAM_GENERIC_ID_REGION_SNAPSHOT,
292 static inline void mlx4_set_num_reserved_uars(struct mlx4_dev *dev,
293 struct mlx4_dev_cap *dev_cap)
295 /* The reserved_uars is calculated by system page size unit.
296 * Therefore, adjustment is added when the uar page size is less
297 * than the system page size
299 dev->caps.reserved_uars =
301 mlx4_get_num_reserved_uar(dev),
302 dev_cap->reserved_uars /
303 (1 << (PAGE_SHIFT - dev->uar_page_shift)));
306 int mlx4_check_port_params(struct mlx4_dev *dev,
307 enum mlx4_port_type *port_type)
311 if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP)) {
312 for (i = 0; i < dev->caps.num_ports - 1; i++) {
313 if (port_type[i] != port_type[i + 1]) {
314 mlx4_err(dev, "Only same port types supported on this HCA, aborting\n");
320 for (i = 0; i < dev->caps.num_ports; i++) {
321 if (!(port_type[i] & dev->caps.supported_type[i+1])) {
322 mlx4_err(dev, "Requested port type for port %d is not supported on this HCA\n",
330 static void mlx4_set_port_mask(struct mlx4_dev *dev)
334 for (i = 1; i <= dev->caps.num_ports; ++i)
335 dev->caps.port_mask[i] = dev->caps.port_type[i];
339 MLX4_QUERY_FUNC_NUM_SYS_EQS = 1 << 0,
342 static int mlx4_query_func(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
345 struct mlx4_func func;
347 if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS) {
348 err = mlx4_QUERY_FUNC(dev, &func, 0);
350 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
353 dev_cap->max_eqs = func.max_eq;
354 dev_cap->reserved_eqs = func.rsvd_eqs;
355 dev_cap->reserved_uars = func.rsvd_uars;
356 err |= MLX4_QUERY_FUNC_NUM_SYS_EQS;
361 static void mlx4_enable_cqe_eqe_stride(struct mlx4_dev *dev)
363 struct mlx4_caps *dev_cap = &dev->caps;
365 /* FW not supporting or cancelled by user */
366 if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_EQE_STRIDE) ||
367 !(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_CQE_STRIDE))
370 /* Must have 64B CQE_EQE enabled by FW to use bigger stride
371 * When FW has NCSI it may decide not to report 64B CQE/EQEs
373 if (!(dev_cap->flags & MLX4_DEV_CAP_FLAG_64B_EQE) ||
374 !(dev_cap->flags & MLX4_DEV_CAP_FLAG_64B_CQE)) {
375 dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
376 dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
380 if (cache_line_size() == 128 || cache_line_size() == 256) {
381 mlx4_dbg(dev, "Enabling CQE stride cacheLine supported\n");
382 /* Changing the real data inside CQE size to 32B */
383 dev_cap->flags &= ~MLX4_DEV_CAP_FLAG_64B_CQE;
384 dev_cap->flags &= ~MLX4_DEV_CAP_FLAG_64B_EQE;
386 if (mlx4_is_master(dev))
387 dev_cap->function_caps |= MLX4_FUNC_CAP_EQE_CQE_STRIDE;
389 if (cache_line_size() != 32 && cache_line_size() != 64)
390 mlx4_dbg(dev, "Disabling CQE stride, cacheLine size unsupported\n");
391 dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
392 dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
396 static int _mlx4_dev_port(struct mlx4_dev *dev, int port,
397 struct mlx4_port_cap *port_cap)
399 dev->caps.vl_cap[port] = port_cap->max_vl;
400 dev->caps.ib_mtu_cap[port] = port_cap->ib_mtu;
401 dev->phys_caps.gid_phys_table_len[port] = port_cap->max_gids;
402 dev->phys_caps.pkey_phys_table_len[port] = port_cap->max_pkeys;
403 /* set gid and pkey table operating lengths by default
404 * to non-sriov values
406 dev->caps.gid_table_len[port] = port_cap->max_gids;
407 dev->caps.pkey_table_len[port] = port_cap->max_pkeys;
408 dev->caps.port_width_cap[port] = port_cap->max_port_width;
409 dev->caps.eth_mtu_cap[port] = port_cap->eth_mtu;
410 dev->caps.max_tc_eth = port_cap->max_tc_eth;
411 dev->caps.def_mac[port] = port_cap->def_mac;
412 dev->caps.supported_type[port] = port_cap->supported_port_types;
413 dev->caps.suggested_type[port] = port_cap->suggested_type;
414 dev->caps.default_sense[port] = port_cap->default_sense;
415 dev->caps.trans_type[port] = port_cap->trans_type;
416 dev->caps.vendor_oui[port] = port_cap->vendor_oui;
417 dev->caps.wavelength[port] = port_cap->wavelength;
418 dev->caps.trans_code[port] = port_cap->trans_code;
423 static int mlx4_dev_port(struct mlx4_dev *dev, int port,
424 struct mlx4_port_cap *port_cap)
428 err = mlx4_QUERY_PORT(dev, port, port_cap);
431 mlx4_err(dev, "QUERY_PORT command failed.\n");
436 static inline void mlx4_enable_ignore_fcs(struct mlx4_dev *dev)
438 if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_IGNORE_FCS))
441 if (mlx4_is_mfunc(dev)) {
442 mlx4_dbg(dev, "SRIOV mode - Disabling Ignore FCS");
443 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_IGNORE_FCS;
447 if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_FCS_KEEP)) {
449 "Keep FCS is not supported - Disabling Ignore FCS");
450 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_IGNORE_FCS;
455 #define MLX4_A0_STEERING_TABLE_SIZE 256
456 static int mlx4_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
461 err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
463 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
466 mlx4_dev_cap_dump(dev, dev_cap);
468 if (dev_cap->min_page_sz > PAGE_SIZE) {
469 mlx4_err(dev, "HCA minimum page size of %d bigger than kernel PAGE_SIZE of %ld, aborting\n",
470 dev_cap->min_page_sz, PAGE_SIZE);
473 if (dev_cap->num_ports > MLX4_MAX_PORTS) {
474 mlx4_err(dev, "HCA has %d ports, but we only support %d, aborting\n",
475 dev_cap->num_ports, MLX4_MAX_PORTS);
479 if (dev_cap->uar_size > pci_resource_len(dev->persist->pdev, 2)) {
480 mlx4_err(dev, "HCA reported UAR size of 0x%x bigger than PCI resource 2 size of 0x%llx, aborting\n",
483 pci_resource_len(dev->persist->pdev, 2));
487 dev->caps.num_ports = dev_cap->num_ports;
488 dev->caps.num_sys_eqs = dev_cap->num_sys_eqs;
489 dev->phys_caps.num_phys_eqs = dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS ?
490 dev->caps.num_sys_eqs :
492 for (i = 1; i <= dev->caps.num_ports; ++i) {
493 err = _mlx4_dev_port(dev, i, dev_cap->port_cap + i);
495 mlx4_err(dev, "QUERY_PORT command failed, aborting\n");
500 dev->caps.map_clock_to_user = dev_cap->map_clock_to_user;
501 dev->caps.uar_page_size = PAGE_SIZE;
502 dev->caps.num_uars = dev_cap->uar_size / PAGE_SIZE;
503 dev->caps.local_ca_ack_delay = dev_cap->local_ca_ack_delay;
504 dev->caps.bf_reg_size = dev_cap->bf_reg_size;
505 dev->caps.bf_regs_per_page = dev_cap->bf_regs_per_page;
506 dev->caps.max_sq_sg = dev_cap->max_sq_sg;
507 dev->caps.max_rq_sg = dev_cap->max_rq_sg;
508 dev->caps.max_wqes = dev_cap->max_qp_sz;
509 dev->caps.max_qp_init_rdma = dev_cap->max_requester_per_qp;
510 dev->caps.max_srq_wqes = dev_cap->max_srq_sz;
511 dev->caps.max_srq_sge = dev_cap->max_rq_sg - 1;
512 dev->caps.reserved_srqs = dev_cap->reserved_srqs;
513 dev->caps.max_sq_desc_sz = dev_cap->max_sq_desc_sz;
514 dev->caps.max_rq_desc_sz = dev_cap->max_rq_desc_sz;
516 * Subtract 1 from the limit because we need to allocate a
517 * spare CQE to enable resizing the CQ.
519 dev->caps.max_cqes = dev_cap->max_cq_sz - 1;
520 dev->caps.reserved_cqs = dev_cap->reserved_cqs;
521 dev->caps.reserved_eqs = dev_cap->reserved_eqs;
522 dev->caps.reserved_mtts = dev_cap->reserved_mtts;
523 dev->caps.reserved_mrws = dev_cap->reserved_mrws;
525 dev->caps.reserved_pds = dev_cap->reserved_pds;
526 dev->caps.reserved_xrcds = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
527 dev_cap->reserved_xrcds : 0;
528 dev->caps.max_xrcds = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
529 dev_cap->max_xrcds : 0;
530 dev->caps.mtt_entry_sz = dev_cap->mtt_entry_sz;
532 dev->caps.max_msg_sz = dev_cap->max_msg_sz;
533 dev->caps.page_size_cap = ~(u32) (dev_cap->min_page_sz - 1);
534 dev->caps.flags = dev_cap->flags;
535 dev->caps.flags2 = dev_cap->flags2;
536 dev->caps.bmme_flags = dev_cap->bmme_flags;
537 dev->caps.reserved_lkey = dev_cap->reserved_lkey;
538 dev->caps.stat_rate_support = dev_cap->stat_rate_support;
539 dev->caps.max_gso_sz = dev_cap->max_gso_sz;
540 dev->caps.max_rss_tbl_sz = dev_cap->max_rss_tbl_sz;
541 dev->caps.wol_port[1] = dev_cap->wol_port[1];
542 dev->caps.wol_port[2] = dev_cap->wol_port[2];
543 dev->caps.health_buffer_addrs = dev_cap->health_buffer_addrs;
545 /* Save uar page shift */
546 if (!mlx4_is_slave(dev)) {
547 /* Virtual PCI function needs to determine UAR page size from
548 * firmware. Only master PCI function can set the uar page size
550 if (enable_4k_uar || !dev->persist->num_vfs)
551 dev->uar_page_shift = DEFAULT_UAR_PAGE_SHIFT;
553 dev->uar_page_shift = PAGE_SHIFT;
555 mlx4_set_num_reserved_uars(dev, dev_cap);
558 if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_PHV_EN) {
559 struct mlx4_init_hca_param hca_param;
561 memset(&hca_param, 0, sizeof(hca_param));
562 err = mlx4_QUERY_HCA(dev, &hca_param);
563 /* Turn off PHV_EN flag in case phv_check_en is set.
564 * phv_check_en is a HW check that parse the packet and verify
565 * phv bit was reported correctly in the wqe. To allow QinQ
566 * PHV_EN flag should be set and phv_check_en must be cleared
567 * otherwise QinQ packets will be drop by the HW.
569 if (err || hca_param.phv_check_en)
570 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_PHV_EN;
573 /* Sense port always allowed on supported devices for ConnectX-1 and -2 */
574 if (mlx4_priv(dev)->pci_dev_data & MLX4_PCI_DEV_FORCE_SENSE_PORT)
575 dev->caps.flags |= MLX4_DEV_CAP_FLAG_SENSE_SUPPORT;
576 /* Don't do sense port on multifunction devices (for now at least) */
577 if (mlx4_is_mfunc(dev))
578 dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_SENSE_SUPPORT;
580 if (mlx4_low_memory_profile()) {
581 dev->caps.log_num_macs = MLX4_MIN_LOG_NUM_MAC;
582 dev->caps.log_num_vlans = MLX4_MIN_LOG_NUM_VLANS;
584 dev->caps.log_num_macs = log_num_mac;
585 dev->caps.log_num_vlans = MLX4_LOG_NUM_VLANS;
588 for (i = 1; i <= dev->caps.num_ports; ++i) {
589 dev->caps.port_type[i] = MLX4_PORT_TYPE_NONE;
590 if (dev->caps.supported_type[i]) {
591 /* if only ETH is supported - assign ETH */
592 if (dev->caps.supported_type[i] == MLX4_PORT_TYPE_ETH)
593 dev->caps.port_type[i] = MLX4_PORT_TYPE_ETH;
594 /* if only IB is supported, assign IB */
595 else if (dev->caps.supported_type[i] ==
597 dev->caps.port_type[i] = MLX4_PORT_TYPE_IB;
599 /* if IB and ETH are supported, we set the port
600 * type according to user selection of port type;
601 * if user selected none, take the FW hint */
602 if (port_type_array[i - 1] == MLX4_PORT_TYPE_NONE)
603 dev->caps.port_type[i] = dev->caps.suggested_type[i] ?
604 MLX4_PORT_TYPE_ETH : MLX4_PORT_TYPE_IB;
606 dev->caps.port_type[i] = port_type_array[i - 1];
610 * Link sensing is allowed on the port if 3 conditions are true:
611 * 1. Both protocols are supported on the port.
612 * 2. Different types are supported on the port
613 * 3. FW declared that it supports link sensing
615 mlx4_priv(dev)->sense.sense_allowed[i] =
616 ((dev->caps.supported_type[i] == MLX4_PORT_TYPE_AUTO) &&
617 (dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
618 (dev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT));
621 * If "default_sense" bit is set, we move the port to "AUTO" mode
622 * and perform sense_port FW command to try and set the correct
623 * port type from beginning
625 if (mlx4_priv(dev)->sense.sense_allowed[i] && dev->caps.default_sense[i]) {
626 enum mlx4_port_type sensed_port = MLX4_PORT_TYPE_NONE;
627 dev->caps.possible_type[i] = MLX4_PORT_TYPE_AUTO;
628 mlx4_SENSE_PORT(dev, i, &sensed_port);
629 if (sensed_port != MLX4_PORT_TYPE_NONE)
630 dev->caps.port_type[i] = sensed_port;
632 dev->caps.possible_type[i] = dev->caps.port_type[i];
635 if (dev->caps.log_num_macs > dev_cap->port_cap[i].log_max_macs) {
636 dev->caps.log_num_macs = dev_cap->port_cap[i].log_max_macs;
637 mlx4_warn(dev, "Requested number of MACs is too much for port %d, reducing to %d\n",
638 i, 1 << dev->caps.log_num_macs);
640 if (dev->caps.log_num_vlans > dev_cap->port_cap[i].log_max_vlans) {
641 dev->caps.log_num_vlans = dev_cap->port_cap[i].log_max_vlans;
642 mlx4_warn(dev, "Requested number of VLANs is too much for port %d, reducing to %d\n",
643 i, 1 << dev->caps.log_num_vlans);
647 if (mlx4_is_master(dev) && (dev->caps.num_ports == 2) &&
648 (port_type_array[0] == MLX4_PORT_TYPE_IB) &&
649 (port_type_array[1] == MLX4_PORT_TYPE_ETH)) {
651 "Granular QoS per VF not supported with IB/Eth configuration\n");
652 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_QOS_VPP;
655 dev->caps.max_counters = dev_cap->max_counters;
657 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] = dev_cap->reserved_qps;
658 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] =
659 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] =
660 (1 << dev->caps.log_num_macs) *
661 (1 << dev->caps.log_num_vlans) *
663 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH] = MLX4_NUM_FEXCH;
665 if (dev_cap->dmfs_high_rate_qpn_base > 0 &&
666 dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN)
667 dev->caps.dmfs_high_rate_qpn_base = dev_cap->dmfs_high_rate_qpn_base;
669 dev->caps.dmfs_high_rate_qpn_base =
670 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
672 if (dev_cap->dmfs_high_rate_qpn_range > 0 &&
673 dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN) {
674 dev->caps.dmfs_high_rate_qpn_range = dev_cap->dmfs_high_rate_qpn_range;
675 dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_DEFAULT;
676 dev->caps.flags2 |= MLX4_DEV_CAP_FLAG2_FS_A0;
678 dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_NOT_SUPPORTED;
679 dev->caps.dmfs_high_rate_qpn_base =
680 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
681 dev->caps.dmfs_high_rate_qpn_range = MLX4_A0_STEERING_TABLE_SIZE;
684 dev->caps.rl_caps = dev_cap->rl_caps;
686 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_RSS_RAW_ETH] =
687 dev->caps.dmfs_high_rate_qpn_range;
689 dev->caps.reserved_qps = dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] +
690 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] +
691 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] +
692 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH];
694 dev->caps.sqp_demux = (mlx4_is_master(dev)) ? MLX4_MAX_NUM_SLAVES : 0;
696 if (!enable_64b_cqe_eqe && !mlx4_is_slave(dev)) {
698 (MLX4_DEV_CAP_FLAG_64B_CQE | MLX4_DEV_CAP_FLAG_64B_EQE)) {
699 mlx4_warn(dev, "64B EQEs/CQEs supported by the device but not enabled\n");
700 dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_64B_CQE;
701 dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_64B_EQE;
704 if (dev_cap->flags2 &
705 (MLX4_DEV_CAP_FLAG2_CQE_STRIDE |
706 MLX4_DEV_CAP_FLAG2_EQE_STRIDE)) {
707 mlx4_warn(dev, "Disabling EQE/CQE stride per user request\n");
708 dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
709 dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
713 if ((dev->caps.flags &
714 (MLX4_DEV_CAP_FLAG_64B_CQE | MLX4_DEV_CAP_FLAG_64B_EQE)) &&
716 dev->caps.function_caps |= MLX4_FUNC_CAP_64B_EQE_CQE;
718 if (!mlx4_is_slave(dev)) {
719 mlx4_enable_cqe_eqe_stride(dev);
720 dev->caps.alloc_res_qp_mask =
721 (dev->caps.bf_reg_size ? MLX4_RESERVE_ETH_BF_QP : 0) |
724 if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ETS_CFG) &&
725 dev->caps.flags & MLX4_DEV_CAP_FLAG_SET_ETH_SCHED) {
726 mlx4_warn(dev, "Old device ETS support detected\n");
727 mlx4_warn(dev, "Consider upgrading device FW.\n");
728 dev->caps.flags2 |= MLX4_DEV_CAP_FLAG2_ETS_CFG;
732 dev->caps.alloc_res_qp_mask = 0;
735 mlx4_enable_ignore_fcs(dev);
740 /*The function checks if there are live vf, return the num of them*/
741 static int mlx4_how_many_lives_vf(struct mlx4_dev *dev)
743 struct mlx4_priv *priv = mlx4_priv(dev);
744 struct mlx4_slave_state *s_state;
748 for (i = 1/*the ppf is 0*/; i < dev->num_slaves; ++i) {
749 s_state = &priv->mfunc.master.slave_state[i];
750 if (s_state->active && s_state->last_cmd !=
751 MLX4_COMM_CMD_RESET) {
752 mlx4_warn(dev, "%s: slave: %d is still active\n",
760 int mlx4_get_parav_qkey(struct mlx4_dev *dev, u32 qpn, u32 *qkey)
762 u32 qk = MLX4_RESERVED_QKEY_BASE;
764 if (qpn >= dev->phys_caps.base_tunnel_sqpn + 8 * MLX4_MFUNC_MAX ||
765 qpn < dev->phys_caps.base_proxy_sqpn)
768 if (qpn >= dev->phys_caps.base_tunnel_sqpn)
770 qk += qpn - dev->phys_caps.base_tunnel_sqpn;
772 qk += qpn - dev->phys_caps.base_proxy_sqpn;
776 EXPORT_SYMBOL(mlx4_get_parav_qkey);
778 void mlx4_sync_pkey_table(struct mlx4_dev *dev, int slave, int port, int i, int val)
780 struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
782 if (!mlx4_is_master(dev))
785 priv->virt2phys_pkey[slave][port - 1][i] = val;
787 EXPORT_SYMBOL(mlx4_sync_pkey_table);
789 void mlx4_put_slave_node_guid(struct mlx4_dev *dev, int slave, __be64 guid)
791 struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
793 if (!mlx4_is_master(dev))
796 priv->slave_node_guids[slave] = guid;
798 EXPORT_SYMBOL(mlx4_put_slave_node_guid);
800 __be64 mlx4_get_slave_node_guid(struct mlx4_dev *dev, int slave)
802 struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
804 if (!mlx4_is_master(dev))
807 return priv->slave_node_guids[slave];
809 EXPORT_SYMBOL(mlx4_get_slave_node_guid);
811 int mlx4_is_slave_active(struct mlx4_dev *dev, int slave)
813 struct mlx4_priv *priv = mlx4_priv(dev);
814 struct mlx4_slave_state *s_slave;
816 if (!mlx4_is_master(dev))
819 s_slave = &priv->mfunc.master.slave_state[slave];
820 return !!s_slave->active;
822 EXPORT_SYMBOL(mlx4_is_slave_active);
824 void mlx4_handle_eth_header_mcast_prio(struct mlx4_net_trans_rule_hw_ctrl *ctrl,
825 struct _rule_hw *eth_header)
827 if (is_multicast_ether_addr(eth_header->eth.dst_mac) ||
828 is_broadcast_ether_addr(eth_header->eth.dst_mac)) {
829 struct mlx4_net_trans_rule_hw_eth *eth =
830 (struct mlx4_net_trans_rule_hw_eth *)eth_header;
831 struct _rule_hw *next_rule = (struct _rule_hw *)(eth + 1);
832 bool last_rule = next_rule->size == 0 && next_rule->id == 0 &&
833 next_rule->rsvd == 0;
836 ctrl->prio = cpu_to_be16(MLX4_DOMAIN_NIC);
839 EXPORT_SYMBOL(mlx4_handle_eth_header_mcast_prio);
841 static void slave_adjust_steering_mode(struct mlx4_dev *dev,
842 struct mlx4_dev_cap *dev_cap,
843 struct mlx4_init_hca_param *hca_param)
845 dev->caps.steering_mode = hca_param->steering_mode;
846 if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED) {
847 dev->caps.num_qp_per_mgm = dev_cap->fs_max_num_qp_per_entry;
848 dev->caps.fs_log_max_ucast_qp_range_size =
849 dev_cap->fs_log_max_ucast_qp_range_size;
851 dev->caps.num_qp_per_mgm =
852 4 * ((1 << hca_param->log_mc_entry_sz)/16 - 2);
854 mlx4_dbg(dev, "Steering mode is: %s\n",
855 mlx4_steering_mode_str(dev->caps.steering_mode));
858 static void mlx4_slave_destroy_special_qp_cap(struct mlx4_dev *dev)
860 kfree(dev->caps.spec_qps);
861 dev->caps.spec_qps = NULL;
864 static int mlx4_slave_special_qp_cap(struct mlx4_dev *dev)
866 struct mlx4_func_cap *func_cap;
867 struct mlx4_caps *caps = &dev->caps;
870 func_cap = kzalloc(sizeof(*func_cap), GFP_KERNEL);
871 caps->spec_qps = kcalloc(caps->num_ports, sizeof(*caps->spec_qps), GFP_KERNEL);
873 if (!func_cap || !caps->spec_qps) {
874 mlx4_err(dev, "Failed to allocate memory for special qps cap\n");
879 for (i = 1; i <= caps->num_ports; ++i) {
880 err = mlx4_QUERY_FUNC_CAP(dev, i, func_cap);
882 mlx4_err(dev, "QUERY_FUNC_CAP port command failed for port %d, aborting (%d)\n",
886 caps->spec_qps[i - 1] = func_cap->spec_qps;
887 caps->port_mask[i] = caps->port_type[i];
888 caps->phys_port_id[i] = func_cap->phys_port_id;
889 err = mlx4_get_slave_pkey_gid_tbl_len(dev, i,
890 &caps->gid_table_len[i],
891 &caps->pkey_table_len[i]);
893 mlx4_err(dev, "QUERY_PORT command failed for port %d, aborting (%d)\n",
901 mlx4_slave_destroy_special_qp_cap(dev);
906 static int mlx4_slave_cap(struct mlx4_dev *dev)
910 struct mlx4_dev_cap *dev_cap;
911 struct mlx4_func_cap *func_cap;
912 struct mlx4_init_hca_param *hca_param;
914 hca_param = kzalloc(sizeof(*hca_param), GFP_KERNEL);
915 func_cap = kzalloc(sizeof(*func_cap), GFP_KERNEL);
916 dev_cap = kzalloc(sizeof(*dev_cap), GFP_KERNEL);
917 if (!hca_param || !func_cap || !dev_cap) {
918 mlx4_err(dev, "Failed to allocate memory for slave_cap\n");
923 err = mlx4_QUERY_HCA(dev, hca_param);
925 mlx4_err(dev, "QUERY_HCA command failed, aborting\n");
929 /* fail if the hca has an unknown global capability
930 * at this time global_caps should be always zeroed
932 if (hca_param->global_caps) {
933 mlx4_err(dev, "Unknown hca global capabilities\n");
938 dev->caps.hca_core_clock = hca_param->hca_core_clock;
940 dev->caps.max_qp_dest_rdma = 1 << hca_param->log_rd_per_qp;
941 err = mlx4_dev_cap(dev, dev_cap);
943 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
947 err = mlx4_QUERY_FW(dev);
949 mlx4_err(dev, "QUERY_FW command failed: could not get FW version\n");
951 page_size = ~dev->caps.page_size_cap + 1;
952 mlx4_warn(dev, "HCA minimum page size:%d\n", page_size);
953 if (page_size > PAGE_SIZE) {
954 mlx4_err(dev, "HCA minimum page size of %d bigger than kernel PAGE_SIZE of %ld, aborting\n",
955 page_size, PAGE_SIZE);
960 /* Set uar_page_shift for VF */
961 dev->uar_page_shift = hca_param->uar_page_sz + 12;
963 /* Make sure the master uar page size is valid */
964 if (dev->uar_page_shift > PAGE_SHIFT) {
966 "Invalid configuration: uar page size is larger than system page size\n");
971 /* Set reserved_uars based on the uar_page_shift */
972 mlx4_set_num_reserved_uars(dev, dev_cap);
974 /* Although uar page size in FW differs from system page size,
975 * upper software layers (mlx4_ib, mlx4_en and part of mlx4_core)
976 * still works with assumption that uar page size == system page size
978 dev->caps.uar_page_size = PAGE_SIZE;
980 err = mlx4_QUERY_FUNC_CAP(dev, 0, func_cap);
982 mlx4_err(dev, "QUERY_FUNC_CAP general command failed, aborting (%d)\n",
987 if ((func_cap->pf_context_behaviour | PF_CONTEXT_BEHAVIOUR_MASK) !=
988 PF_CONTEXT_BEHAVIOUR_MASK) {
989 mlx4_err(dev, "Unknown pf context behaviour %x known flags %x\n",
990 func_cap->pf_context_behaviour,
991 PF_CONTEXT_BEHAVIOUR_MASK);
996 dev->caps.num_ports = func_cap->num_ports;
997 dev->quotas.qp = func_cap->qp_quota;
998 dev->quotas.srq = func_cap->srq_quota;
999 dev->quotas.cq = func_cap->cq_quota;
1000 dev->quotas.mpt = func_cap->mpt_quota;
1001 dev->quotas.mtt = func_cap->mtt_quota;
1002 dev->caps.num_qps = 1 << hca_param->log_num_qps;
1003 dev->caps.num_srqs = 1 << hca_param->log_num_srqs;
1004 dev->caps.num_cqs = 1 << hca_param->log_num_cqs;
1005 dev->caps.num_mpts = 1 << hca_param->log_mpt_sz;
1006 dev->caps.num_eqs = func_cap->max_eq;
1007 dev->caps.reserved_eqs = func_cap->reserved_eq;
1008 dev->caps.reserved_lkey = func_cap->reserved_lkey;
1009 dev->caps.num_pds = MLX4_NUM_PDS;
1010 dev->caps.num_mgms = 0;
1011 dev->caps.num_amgms = 0;
1013 if (dev->caps.num_ports > MLX4_MAX_PORTS) {
1014 mlx4_err(dev, "HCA has %d ports, but we only support %d, aborting\n",
1015 dev->caps.num_ports, MLX4_MAX_PORTS);
1020 mlx4_replace_zero_macs(dev);
1022 err = mlx4_slave_special_qp_cap(dev);
1024 mlx4_err(dev, "Set special QP caps failed. aborting\n");
1028 if (dev->caps.uar_page_size * (dev->caps.num_uars -
1029 dev->caps.reserved_uars) >
1030 pci_resource_len(dev->persist->pdev,
1032 mlx4_err(dev, "HCA reported UAR region size of 0x%x bigger than PCI resource 2 size of 0x%llx, aborting\n",
1033 dev->caps.uar_page_size * dev->caps.num_uars,
1034 (unsigned long long)
1035 pci_resource_len(dev->persist->pdev, 2));
1040 if (hca_param->dev_cap_enabled & MLX4_DEV_CAP_64B_EQE_ENABLED) {
1041 dev->caps.eqe_size = 64;
1042 dev->caps.eqe_factor = 1;
1044 dev->caps.eqe_size = 32;
1045 dev->caps.eqe_factor = 0;
1048 if (hca_param->dev_cap_enabled & MLX4_DEV_CAP_64B_CQE_ENABLED) {
1049 dev->caps.cqe_size = 64;
1050 dev->caps.userspace_caps |= MLX4_USER_DEV_CAP_LARGE_CQE;
1052 dev->caps.cqe_size = 32;
1055 if (hca_param->dev_cap_enabled & MLX4_DEV_CAP_EQE_STRIDE_ENABLED) {
1056 dev->caps.eqe_size = hca_param->eqe_size;
1057 dev->caps.eqe_factor = 0;
1060 if (hca_param->dev_cap_enabled & MLX4_DEV_CAP_CQE_STRIDE_ENABLED) {
1061 dev->caps.cqe_size = hca_param->cqe_size;
1062 /* User still need to know when CQE > 32B */
1063 dev->caps.userspace_caps |= MLX4_USER_DEV_CAP_LARGE_CQE;
1066 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
1067 mlx4_warn(dev, "Timestamping is not supported in slave mode\n");
1069 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_USER_MAC_EN;
1070 mlx4_dbg(dev, "User MAC FW update is not supported in slave mode\n");
1072 slave_adjust_steering_mode(dev, dev_cap, hca_param);
1073 mlx4_dbg(dev, "RSS support for IP fragments is %s\n",
1074 hca_param->rss_ip_frags ? "on" : "off");
1076 if (func_cap->extra_flags & MLX4_QUERY_FUNC_FLAGS_BF_RES_QP &&
1077 dev->caps.bf_reg_size)
1078 dev->caps.alloc_res_qp_mask |= MLX4_RESERVE_ETH_BF_QP;
1080 if (func_cap->extra_flags & MLX4_QUERY_FUNC_FLAGS_A0_RES_QP)
1081 dev->caps.alloc_res_qp_mask |= MLX4_RESERVE_A0_QP;
1085 mlx4_slave_destroy_special_qp_cap(dev);
1094 * Change the port configuration of the device.
1095 * Every user of this function must hold the port mutex.
1097 int mlx4_change_port_types(struct mlx4_dev *dev,
1098 enum mlx4_port_type *port_types)
1104 for (port = 0; port < dev->caps.num_ports; port++) {
1105 /* Change the port type only if the new type is different
1106 * from the current, and not set to Auto */
1107 if (port_types[port] != dev->caps.port_type[port + 1])
1111 mlx4_unregister_device(dev);
1112 for (port = 1; port <= dev->caps.num_ports; port++) {
1113 mlx4_CLOSE_PORT(dev, port);
1114 dev->caps.port_type[port] = port_types[port - 1];
1115 err = mlx4_SET_PORT(dev, port, -1);
1117 mlx4_err(dev, "Failed to set port %d, aborting\n",
1122 mlx4_set_port_mask(dev);
1123 err = mlx4_register_device(dev);
1125 mlx4_err(dev, "Failed to register device\n");
1134 static ssize_t show_port_type(struct device *dev,
1135 struct device_attribute *attr,
1138 struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1140 struct mlx4_dev *mdev = info->dev;
1144 (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_IB) ?
1146 if (mdev->caps.possible_type[info->port] == MLX4_PORT_TYPE_AUTO)
1147 sprintf(buf, "auto (%s)\n", type);
1149 sprintf(buf, "%s\n", type);
1154 static int __set_port_type(struct mlx4_port_info *info,
1155 enum mlx4_port_type port_type)
1157 struct mlx4_dev *mdev = info->dev;
1158 struct mlx4_priv *priv = mlx4_priv(mdev);
1159 enum mlx4_port_type types[MLX4_MAX_PORTS];
1160 enum mlx4_port_type new_types[MLX4_MAX_PORTS];
1164 if ((port_type & mdev->caps.supported_type[info->port]) != port_type) {
1166 "Requested port type for port %d is not supported on this HCA\n",
1171 mlx4_stop_sense(mdev);
1172 mutex_lock(&priv->port_mutex);
1173 info->tmp_type = port_type;
1175 /* Possible type is always the one that was delivered */
1176 mdev->caps.possible_type[info->port] = info->tmp_type;
1178 for (i = 0; i < mdev->caps.num_ports; i++) {
1179 types[i] = priv->port[i+1].tmp_type ? priv->port[i+1].tmp_type :
1180 mdev->caps.possible_type[i+1];
1181 if (types[i] == MLX4_PORT_TYPE_AUTO)
1182 types[i] = mdev->caps.port_type[i+1];
1185 if (!(mdev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
1186 !(mdev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT)) {
1187 for (i = 1; i <= mdev->caps.num_ports; i++) {
1188 if (mdev->caps.possible_type[i] == MLX4_PORT_TYPE_AUTO) {
1189 mdev->caps.possible_type[i] = mdev->caps.port_type[i];
1195 mlx4_err(mdev, "Auto sensing is not supported on this HCA. Set only 'eth' or 'ib' for both ports (should be the same)\n");
1199 mlx4_do_sense_ports(mdev, new_types, types);
1201 err = mlx4_check_port_params(mdev, new_types);
1205 /* We are about to apply the changes after the configuration
1206 * was verified, no need to remember the temporary types
1208 for (i = 0; i < mdev->caps.num_ports; i++)
1209 priv->port[i + 1].tmp_type = 0;
1211 err = mlx4_change_port_types(mdev, new_types);
1214 mlx4_start_sense(mdev);
1215 mutex_unlock(&priv->port_mutex);
1220 static ssize_t set_port_type(struct device *dev,
1221 struct device_attribute *attr,
1222 const char *buf, size_t count)
1224 struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1226 struct mlx4_dev *mdev = info->dev;
1227 enum mlx4_port_type port_type;
1228 static DEFINE_MUTEX(set_port_type_mutex);
1231 mutex_lock(&set_port_type_mutex);
1233 if (!strcmp(buf, "ib\n")) {
1234 port_type = MLX4_PORT_TYPE_IB;
1235 } else if (!strcmp(buf, "eth\n")) {
1236 port_type = MLX4_PORT_TYPE_ETH;
1237 } else if (!strcmp(buf, "auto\n")) {
1238 port_type = MLX4_PORT_TYPE_AUTO;
1240 mlx4_err(mdev, "%s is not supported port type\n", buf);
1245 err = __set_port_type(info, port_type);
1248 mutex_unlock(&set_port_type_mutex);
1250 return err ? err : count;
1261 static inline int int_to_ibta_mtu(int mtu)
1264 case 256: return IB_MTU_256;
1265 case 512: return IB_MTU_512;
1266 case 1024: return IB_MTU_1024;
1267 case 2048: return IB_MTU_2048;
1268 case 4096: return IB_MTU_4096;
1273 static inline int ibta_mtu_to_int(enum ibta_mtu mtu)
1276 case IB_MTU_256: return 256;
1277 case IB_MTU_512: return 512;
1278 case IB_MTU_1024: return 1024;
1279 case IB_MTU_2048: return 2048;
1280 case IB_MTU_4096: return 4096;
1285 static ssize_t show_port_ib_mtu(struct device *dev,
1286 struct device_attribute *attr,
1289 struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1291 struct mlx4_dev *mdev = info->dev;
1293 if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH)
1294 mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
1296 sprintf(buf, "%d\n",
1297 ibta_mtu_to_int(mdev->caps.port_ib_mtu[info->port]));
1301 static ssize_t set_port_ib_mtu(struct device *dev,
1302 struct device_attribute *attr,
1303 const char *buf, size_t count)
1305 struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1307 struct mlx4_dev *mdev = info->dev;
1308 struct mlx4_priv *priv = mlx4_priv(mdev);
1309 int err, port, mtu, ibta_mtu = -1;
1311 if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH) {
1312 mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
1316 err = kstrtoint(buf, 0, &mtu);
1318 ibta_mtu = int_to_ibta_mtu(mtu);
1320 if (err || ibta_mtu < 0) {
1321 mlx4_err(mdev, "%s is invalid IBTA mtu\n", buf);
1325 mdev->caps.port_ib_mtu[info->port] = ibta_mtu;
1327 mlx4_stop_sense(mdev);
1328 mutex_lock(&priv->port_mutex);
1329 mlx4_unregister_device(mdev);
1330 for (port = 1; port <= mdev->caps.num_ports; port++) {
1331 mlx4_CLOSE_PORT(mdev, port);
1332 err = mlx4_SET_PORT(mdev, port, -1);
1334 mlx4_err(mdev, "Failed to set port %d, aborting\n",
1339 err = mlx4_register_device(mdev);
1341 mutex_unlock(&priv->port_mutex);
1342 mlx4_start_sense(mdev);
1343 return err ? err : count;
1346 /* bond for multi-function device */
1347 #define MAX_MF_BOND_ALLOWED_SLAVES 63
1348 static int mlx4_mf_bond(struct mlx4_dev *dev)
1352 struct mlx4_slaves_pport slaves_port1;
1353 struct mlx4_slaves_pport slaves_port2;
1355 slaves_port1 = mlx4_phys_to_slaves_pport(dev, 1);
1356 slaves_port2 = mlx4_phys_to_slaves_pport(dev, 2);
1358 /* only single port vfs are allowed */
1359 if (bitmap_weight_and(slaves_port1.slaves, slaves_port2.slaves,
1360 dev->persist->num_vfs + 1) > 1) {
1361 mlx4_warn(dev, "HA mode unsupported for dual ported VFs\n");
1365 /* number of virtual functions is number of total functions minus one
1366 * physical function for each port.
1368 nvfs = bitmap_weight(slaves_port1.slaves, dev->persist->num_vfs + 1) +
1369 bitmap_weight(slaves_port2.slaves, dev->persist->num_vfs + 1) - 2;
1371 /* limit on maximum allowed VFs */
1372 if (nvfs > MAX_MF_BOND_ALLOWED_SLAVES) {
1373 mlx4_warn(dev, "HA mode is not supported for %d VFs (max %d are allowed)\n",
1374 nvfs, MAX_MF_BOND_ALLOWED_SLAVES);
1378 if (dev->caps.steering_mode != MLX4_STEERING_MODE_DEVICE_MANAGED) {
1379 mlx4_warn(dev, "HA mode unsupported for NON DMFS steering\n");
1383 err = mlx4_bond_mac_table(dev);
1386 err = mlx4_bond_vlan_table(dev);
1389 err = mlx4_bond_fs_rules(dev);
1395 (void)mlx4_unbond_vlan_table(dev);
1397 (void)mlx4_unbond_mac_table(dev);
1401 static int mlx4_mf_unbond(struct mlx4_dev *dev)
1405 ret = mlx4_unbond_fs_rules(dev);
1407 mlx4_warn(dev, "multifunction unbond for flow rules failed (%d)\n", ret);
1408 ret1 = mlx4_unbond_mac_table(dev);
1410 mlx4_warn(dev, "multifunction unbond for MAC table failed (%d)\n", ret1);
1413 ret1 = mlx4_unbond_vlan_table(dev);
1415 mlx4_warn(dev, "multifunction unbond for VLAN table failed (%d)\n", ret1);
1421 static int mlx4_bond(struct mlx4_dev *dev)
1424 struct mlx4_priv *priv = mlx4_priv(dev);
1426 mutex_lock(&priv->bond_mutex);
1428 if (!mlx4_is_bonded(dev)) {
1429 ret = mlx4_do_bond(dev, true);
1431 mlx4_err(dev, "Failed to bond device: %d\n", ret);
1432 if (!ret && mlx4_is_master(dev)) {
1433 ret = mlx4_mf_bond(dev);
1435 mlx4_err(dev, "bond for multifunction failed\n");
1436 mlx4_do_bond(dev, false);
1441 mutex_unlock(&priv->bond_mutex);
1443 mlx4_dbg(dev, "Device is bonded\n");
1448 static int mlx4_unbond(struct mlx4_dev *dev)
1451 struct mlx4_priv *priv = mlx4_priv(dev);
1453 mutex_lock(&priv->bond_mutex);
1455 if (mlx4_is_bonded(dev)) {
1458 ret = mlx4_do_bond(dev, false);
1460 mlx4_err(dev, "Failed to unbond device: %d\n", ret);
1461 if (mlx4_is_master(dev))
1462 ret2 = mlx4_mf_unbond(dev);
1464 mlx4_warn(dev, "Failed to unbond device for multifunction (%d)\n", ret2);
1469 mutex_unlock(&priv->bond_mutex);
1471 mlx4_dbg(dev, "Device is unbonded\n");
1476 static int mlx4_port_map_set(struct mlx4_dev *dev, struct mlx4_port_map *v2p)
1478 u8 port1 = v2p->port1;
1479 u8 port2 = v2p->port2;
1480 struct mlx4_priv *priv = mlx4_priv(dev);
1483 if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_PORT_REMAP))
1486 mutex_lock(&priv->bond_mutex);
1488 /* zero means keep current mapping for this port */
1490 port1 = priv->v2p.port1;
1492 port2 = priv->v2p.port2;
1494 if ((port1 < 1) || (port1 > MLX4_MAX_PORTS) ||
1495 (port2 < 1) || (port2 > MLX4_MAX_PORTS) ||
1496 (port1 == 2 && port2 == 1)) {
1497 /* besides boundary checks cross mapping makes
1498 * no sense and therefore not allowed */
1500 } else if ((port1 == priv->v2p.port1) &&
1501 (port2 == priv->v2p.port2)) {
1504 err = mlx4_virt2phy_port_map(dev, port1, port2);
1506 mlx4_dbg(dev, "port map changed: [%d][%d]\n",
1508 priv->v2p.port1 = port1;
1509 priv->v2p.port2 = port2;
1511 mlx4_err(dev, "Failed to change port mape: %d\n", err);
1515 mutex_unlock(&priv->bond_mutex);
1520 struct work_struct work;
1521 struct mlx4_dev *dev;
1523 struct mlx4_port_map port_map;
1526 static void mlx4_bond_work(struct work_struct *work)
1528 struct mlx4_bond *bond = container_of(work, struct mlx4_bond, work);
1531 if (bond->is_bonded) {
1532 if (!mlx4_is_bonded(bond->dev)) {
1533 err = mlx4_bond(bond->dev);
1535 mlx4_err(bond->dev, "Fail to bond device\n");
1538 err = mlx4_port_map_set(bond->dev, &bond->port_map);
1541 "Fail to set port map [%d][%d]: %d\n",
1542 bond->port_map.port1,
1543 bond->port_map.port2, err);
1545 } else if (mlx4_is_bonded(bond->dev)) {
1546 err = mlx4_unbond(bond->dev);
1548 mlx4_err(bond->dev, "Fail to unbond device\n");
1550 put_device(&bond->dev->persist->pdev->dev);
1554 int mlx4_queue_bond_work(struct mlx4_dev *dev, int is_bonded, u8 v2p_p1,
1557 struct mlx4_bond *bond;
1559 bond = kzalloc(sizeof(*bond), GFP_ATOMIC);
1563 INIT_WORK(&bond->work, mlx4_bond_work);
1564 get_device(&dev->persist->pdev->dev);
1566 bond->is_bonded = is_bonded;
1567 bond->port_map.port1 = v2p_p1;
1568 bond->port_map.port2 = v2p_p2;
1569 queue_work(mlx4_wq, &bond->work);
1572 EXPORT_SYMBOL(mlx4_queue_bond_work);
1574 static int mlx4_load_fw(struct mlx4_dev *dev)
1576 struct mlx4_priv *priv = mlx4_priv(dev);
1579 priv->fw.fw_icm = mlx4_alloc_icm(dev, priv->fw.fw_pages,
1580 GFP_HIGHUSER | __GFP_NOWARN, 0);
1581 if (!priv->fw.fw_icm) {
1582 mlx4_err(dev, "Couldn't allocate FW area, aborting\n");
1586 err = mlx4_MAP_FA(dev, priv->fw.fw_icm);
1588 mlx4_err(dev, "MAP_FA command failed, aborting\n");
1592 err = mlx4_RUN_FW(dev);
1594 mlx4_err(dev, "RUN_FW command failed, aborting\n");
1604 mlx4_free_icm(dev, priv->fw.fw_icm, 0);
1608 static int mlx4_init_cmpt_table(struct mlx4_dev *dev, u64 cmpt_base,
1611 struct mlx4_priv *priv = mlx4_priv(dev);
1615 err = mlx4_init_icm_table(dev, &priv->qp_table.cmpt_table,
1617 ((u64) (MLX4_CMPT_TYPE_QP *
1618 cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1619 cmpt_entry_sz, dev->caps.num_qps,
1620 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1625 err = mlx4_init_icm_table(dev, &priv->srq_table.cmpt_table,
1627 ((u64) (MLX4_CMPT_TYPE_SRQ *
1628 cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1629 cmpt_entry_sz, dev->caps.num_srqs,
1630 dev->caps.reserved_srqs, 0, 0);
1634 err = mlx4_init_icm_table(dev, &priv->cq_table.cmpt_table,
1636 ((u64) (MLX4_CMPT_TYPE_CQ *
1637 cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1638 cmpt_entry_sz, dev->caps.num_cqs,
1639 dev->caps.reserved_cqs, 0, 0);
1643 num_eqs = dev->phys_caps.num_phys_eqs;
1644 err = mlx4_init_icm_table(dev, &priv->eq_table.cmpt_table,
1646 ((u64) (MLX4_CMPT_TYPE_EQ *
1647 cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1648 cmpt_entry_sz, num_eqs, num_eqs, 0, 0);
1655 mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1658 mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1661 mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1667 static int mlx4_init_icm(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap,
1668 struct mlx4_init_hca_param *init_hca, u64 icm_size)
1670 struct mlx4_priv *priv = mlx4_priv(dev);
1675 err = mlx4_SET_ICM_SIZE(dev, icm_size, &aux_pages);
1677 mlx4_err(dev, "SET_ICM_SIZE command failed, aborting\n");
1681 mlx4_dbg(dev, "%lld KB of HCA context requires %lld KB aux memory\n",
1682 (unsigned long long) icm_size >> 10,
1683 (unsigned long long) aux_pages << 2);
1685 priv->fw.aux_icm = mlx4_alloc_icm(dev, aux_pages,
1686 GFP_HIGHUSER | __GFP_NOWARN, 0);
1687 if (!priv->fw.aux_icm) {
1688 mlx4_err(dev, "Couldn't allocate aux memory, aborting\n");
1692 err = mlx4_MAP_ICM_AUX(dev, priv->fw.aux_icm);
1694 mlx4_err(dev, "MAP_ICM_AUX command failed, aborting\n");
1698 err = mlx4_init_cmpt_table(dev, init_hca->cmpt_base, dev_cap->cmpt_entry_sz);
1700 mlx4_err(dev, "Failed to map cMPT context memory, aborting\n");
1705 num_eqs = dev->phys_caps.num_phys_eqs;
1706 err = mlx4_init_icm_table(dev, &priv->eq_table.table,
1707 init_hca->eqc_base, dev_cap->eqc_entry_sz,
1708 num_eqs, num_eqs, 0, 0);
1710 mlx4_err(dev, "Failed to map EQ context memory, aborting\n");
1711 goto err_unmap_cmpt;
1715 * Reserved MTT entries must be aligned up to a cacheline
1716 * boundary, since the FW will write to them, while the driver
1717 * writes to all other MTT entries. (The variable
1718 * dev->caps.mtt_entry_sz below is really the MTT segment
1719 * size, not the raw entry size)
1721 dev->caps.reserved_mtts =
1722 ALIGN(dev->caps.reserved_mtts * dev->caps.mtt_entry_sz,
1723 dma_get_cache_alignment()) / dev->caps.mtt_entry_sz;
1725 err = mlx4_init_icm_table(dev, &priv->mr_table.mtt_table,
1727 dev->caps.mtt_entry_sz,
1729 dev->caps.reserved_mtts, 1, 0);
1731 mlx4_err(dev, "Failed to map MTT context memory, aborting\n");
1735 err = mlx4_init_icm_table(dev, &priv->mr_table.dmpt_table,
1736 init_hca->dmpt_base,
1737 dev_cap->dmpt_entry_sz,
1739 dev->caps.reserved_mrws, 1, 1);
1741 mlx4_err(dev, "Failed to map dMPT context memory, aborting\n");
1745 err = mlx4_init_icm_table(dev, &priv->qp_table.qp_table,
1747 dev_cap->qpc_entry_sz,
1749 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1752 mlx4_err(dev, "Failed to map QP context memory, aborting\n");
1753 goto err_unmap_dmpt;
1756 err = mlx4_init_icm_table(dev, &priv->qp_table.auxc_table,
1757 init_hca->auxc_base,
1758 dev_cap->aux_entry_sz,
1760 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1763 mlx4_err(dev, "Failed to map AUXC context memory, aborting\n");
1767 err = mlx4_init_icm_table(dev, &priv->qp_table.altc_table,
1768 init_hca->altc_base,
1769 dev_cap->altc_entry_sz,
1771 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1774 mlx4_err(dev, "Failed to map ALTC context memory, aborting\n");
1775 goto err_unmap_auxc;
1778 err = mlx4_init_icm_table(dev, &priv->qp_table.rdmarc_table,
1779 init_hca->rdmarc_base,
1780 dev_cap->rdmarc_entry_sz << priv->qp_table.rdmarc_shift,
1782 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1785 mlx4_err(dev, "Failed to map RDMARC context memory, aborting\n");
1786 goto err_unmap_altc;
1789 err = mlx4_init_icm_table(dev, &priv->cq_table.table,
1791 dev_cap->cqc_entry_sz,
1793 dev->caps.reserved_cqs, 0, 0);
1795 mlx4_err(dev, "Failed to map CQ context memory, aborting\n");
1796 goto err_unmap_rdmarc;
1799 err = mlx4_init_icm_table(dev, &priv->srq_table.table,
1800 init_hca->srqc_base,
1801 dev_cap->srq_entry_sz,
1803 dev->caps.reserved_srqs, 0, 0);
1805 mlx4_err(dev, "Failed to map SRQ context memory, aborting\n");
1810 * For flow steering device managed mode it is required to use
1811 * mlx4_init_icm_table. For B0 steering mode it's not strictly
1812 * required, but for simplicity just map the whole multicast
1813 * group table now. The table isn't very big and it's a lot
1814 * easier than trying to track ref counts.
1816 err = mlx4_init_icm_table(dev, &priv->mcg_table.table,
1818 mlx4_get_mgm_entry_size(dev),
1819 dev->caps.num_mgms + dev->caps.num_amgms,
1820 dev->caps.num_mgms + dev->caps.num_amgms,
1823 mlx4_err(dev, "Failed to map MCG context memory, aborting\n");
1830 mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1833 mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1836 mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1839 mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1842 mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1845 mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1848 mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1851 mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
1854 mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
1857 mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1858 mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1859 mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1860 mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1863 mlx4_UNMAP_ICM_AUX(dev);
1866 mlx4_free_icm(dev, priv->fw.aux_icm, 0);
1871 static void mlx4_free_icms(struct mlx4_dev *dev)
1873 struct mlx4_priv *priv = mlx4_priv(dev);
1875 mlx4_cleanup_icm_table(dev, &priv->mcg_table.table);
1876 mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1877 mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1878 mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1879 mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1880 mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1881 mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1882 mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1883 mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
1884 mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
1885 mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1886 mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1887 mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1888 mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1890 mlx4_UNMAP_ICM_AUX(dev);
1891 mlx4_free_icm(dev, priv->fw.aux_icm, 0);
1894 static void mlx4_slave_exit(struct mlx4_dev *dev)
1896 struct mlx4_priv *priv = mlx4_priv(dev);
1898 mutex_lock(&priv->cmd.slave_cmd_mutex);
1899 if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, MLX4_COMM_CMD_NA_OP,
1901 mlx4_warn(dev, "Failed to close slave function\n");
1902 mutex_unlock(&priv->cmd.slave_cmd_mutex);
1905 static int map_bf_area(struct mlx4_dev *dev)
1907 struct mlx4_priv *priv = mlx4_priv(dev);
1908 resource_size_t bf_start;
1909 resource_size_t bf_len;
1912 if (!dev->caps.bf_reg_size)
1915 bf_start = pci_resource_start(dev->persist->pdev, 2) +
1916 (dev->caps.num_uars << PAGE_SHIFT);
1917 bf_len = pci_resource_len(dev->persist->pdev, 2) -
1918 (dev->caps.num_uars << PAGE_SHIFT);
1919 priv->bf_mapping = io_mapping_create_wc(bf_start, bf_len);
1920 if (!priv->bf_mapping)
1926 static void unmap_bf_area(struct mlx4_dev *dev)
1928 if (mlx4_priv(dev)->bf_mapping)
1929 io_mapping_free(mlx4_priv(dev)->bf_mapping);
1932 u64 mlx4_read_clock(struct mlx4_dev *dev)
1934 u32 clockhi, clocklo, clockhi1;
1937 struct mlx4_priv *priv = mlx4_priv(dev);
1939 for (i = 0; i < 10; i++) {
1940 clockhi = swab32(readl(priv->clock_mapping));
1941 clocklo = swab32(readl(priv->clock_mapping + 4));
1942 clockhi1 = swab32(readl(priv->clock_mapping));
1943 if (clockhi == clockhi1)
1947 cycles = (u64) clockhi << 32 | (u64) clocklo;
1951 EXPORT_SYMBOL_GPL(mlx4_read_clock);
1954 static int map_internal_clock(struct mlx4_dev *dev)
1956 struct mlx4_priv *priv = mlx4_priv(dev);
1958 priv->clock_mapping =
1959 ioremap(pci_resource_start(dev->persist->pdev,
1960 priv->fw.clock_bar) +
1961 priv->fw.clock_offset, MLX4_CLOCK_SIZE);
1963 if (!priv->clock_mapping)
1969 int mlx4_get_internal_clock_params(struct mlx4_dev *dev,
1970 struct mlx4_clock_params *params)
1972 struct mlx4_priv *priv = mlx4_priv(dev);
1974 if (mlx4_is_slave(dev))
1977 if (!dev->caps.map_clock_to_user) {
1978 mlx4_dbg(dev, "Map clock to user is not supported.\n");
1985 params->bar = priv->fw.clock_bar;
1986 params->offset = priv->fw.clock_offset;
1987 params->size = MLX4_CLOCK_SIZE;
1991 EXPORT_SYMBOL_GPL(mlx4_get_internal_clock_params);
1993 static void unmap_internal_clock(struct mlx4_dev *dev)
1995 struct mlx4_priv *priv = mlx4_priv(dev);
1997 if (priv->clock_mapping)
1998 iounmap(priv->clock_mapping);
2001 static void mlx4_close_hca(struct mlx4_dev *dev)
2003 unmap_internal_clock(dev);
2005 if (mlx4_is_slave(dev))
2006 mlx4_slave_exit(dev);
2008 mlx4_CLOSE_HCA(dev, 0);
2009 mlx4_free_icms(dev);
2013 static void mlx4_close_fw(struct mlx4_dev *dev)
2015 if (!mlx4_is_slave(dev)) {
2017 mlx4_free_icm(dev, mlx4_priv(dev)->fw.fw_icm, 0);
2021 static int mlx4_comm_check_offline(struct mlx4_dev *dev)
2023 #define COMM_CHAN_OFFLINE_OFFSET 0x09
2028 struct mlx4_priv *priv = mlx4_priv(dev);
2030 end = msecs_to_jiffies(MLX4_COMM_OFFLINE_TIME_OUT) + jiffies;
2031 while (time_before(jiffies, end)) {
2032 comm_flags = swab32(readl((__iomem char *)priv->mfunc.comm +
2033 MLX4_COMM_CHAN_FLAGS));
2034 offline_bit = (comm_flags &
2035 (u32)(1 << COMM_CHAN_OFFLINE_OFFSET));
2039 /* If device removal has been requested,
2040 * do not continue retrying.
2042 if (dev->persist->interface_state &
2043 MLX4_INTERFACE_STATE_NOWAIT)
2046 /* There are cases as part of AER/Reset flow that PF needs
2047 * around 100 msec to load. We therefore sleep for 100 msec
2048 * to allow other tasks to make use of that CPU during this
2053 mlx4_err(dev, "Communication channel is offline.\n");
2057 static void mlx4_reset_vf_support(struct mlx4_dev *dev)
2059 #define COMM_CHAN_RST_OFFSET 0x1e
2061 struct mlx4_priv *priv = mlx4_priv(dev);
2065 comm_caps = swab32(readl((__iomem char *)priv->mfunc.comm +
2066 MLX4_COMM_CHAN_CAPS));
2067 comm_rst = (comm_caps & (u32)(1 << COMM_CHAN_RST_OFFSET));
2070 dev->caps.vf_caps |= MLX4_VF_CAP_FLAG_RESET;
2073 static int mlx4_init_slave(struct mlx4_dev *dev)
2075 struct mlx4_priv *priv = mlx4_priv(dev);
2076 u64 dma = (u64) priv->mfunc.vhcr_dma;
2077 int ret_from_reset = 0;
2079 u32 cmd_channel_ver;
2081 if (atomic_read(&pf_loading)) {
2082 mlx4_warn(dev, "PF is not ready - Deferring probe\n");
2083 return -EPROBE_DEFER;
2086 mutex_lock(&priv->cmd.slave_cmd_mutex);
2087 priv->cmd.max_cmds = 1;
2088 if (mlx4_comm_check_offline(dev)) {
2089 mlx4_err(dev, "PF is not responsive, skipping initialization\n");
2093 mlx4_reset_vf_support(dev);
2094 mlx4_warn(dev, "Sending reset\n");
2095 ret_from_reset = mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0,
2096 MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME);
2097 /* if we are in the middle of flr the slave will try
2098 * NUM_OF_RESET_RETRIES times before leaving.*/
2099 if (ret_from_reset) {
2100 if (MLX4_DELAY_RESET_SLAVE == ret_from_reset) {
2101 mlx4_warn(dev, "slave is currently in the middle of FLR - Deferring probe\n");
2102 mutex_unlock(&priv->cmd.slave_cmd_mutex);
2103 return -EPROBE_DEFER;
2108 /* check the driver version - the slave I/F revision
2109 * must match the master's */
2110 slave_read = swab32(readl(&priv->mfunc.comm->slave_read));
2111 cmd_channel_ver = mlx4_comm_get_version();
2113 if (MLX4_COMM_GET_IF_REV(cmd_channel_ver) !=
2114 MLX4_COMM_GET_IF_REV(slave_read)) {
2115 mlx4_err(dev, "slave driver version is not supported by the master\n");
2119 mlx4_warn(dev, "Sending vhcr0\n");
2120 if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR0, dma >> 48,
2121 MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
2123 if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR1, dma >> 32,
2124 MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
2126 if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR2, dma >> 16,
2127 MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
2129 if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR_EN, dma,
2130 MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
2133 mutex_unlock(&priv->cmd.slave_cmd_mutex);
2137 mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, MLX4_COMM_CMD_NA_OP, 0);
2139 mutex_unlock(&priv->cmd.slave_cmd_mutex);
2143 static void mlx4_parav_master_pf_caps(struct mlx4_dev *dev)
2147 for (i = 1; i <= dev->caps.num_ports; i++) {
2148 if (dev->caps.port_type[i] == MLX4_PORT_TYPE_ETH)
2149 dev->caps.gid_table_len[i] =
2150 mlx4_get_slave_num_gids(dev, 0, i);
2152 dev->caps.gid_table_len[i] = 1;
2153 dev->caps.pkey_table_len[i] =
2154 dev->phys_caps.pkey_phys_table_len[i] - 1;
2158 static int choose_log_fs_mgm_entry_size(int qp_per_entry)
2160 int i = MLX4_MIN_MGM_LOG_ENTRY_SIZE;
2162 for (i = MLX4_MIN_MGM_LOG_ENTRY_SIZE; i <= MLX4_MAX_MGM_LOG_ENTRY_SIZE;
2164 if (qp_per_entry <= 4 * ((1 << i) / 16 - 2))
2168 return (i <= MLX4_MAX_MGM_LOG_ENTRY_SIZE) ? i : -1;
2171 static const char *dmfs_high_rate_steering_mode_str(int dmfs_high_steer_mode)
2173 switch (dmfs_high_steer_mode) {
2174 case MLX4_STEERING_DMFS_A0_DEFAULT:
2175 return "default performance";
2177 case MLX4_STEERING_DMFS_A0_DYNAMIC:
2178 return "dynamic hybrid mode";
2180 case MLX4_STEERING_DMFS_A0_STATIC:
2181 return "performance optimized for limited rule configuration (static)";
2183 case MLX4_STEERING_DMFS_A0_DISABLE:
2184 return "disabled performance optimized steering";
2186 case MLX4_STEERING_DMFS_A0_NOT_SUPPORTED:
2187 return "performance optimized steering not supported";
2190 return "Unrecognized mode";
2194 #define MLX4_DMFS_A0_STEERING (1UL << 2)
2196 static void choose_steering_mode(struct mlx4_dev *dev,
2197 struct mlx4_dev_cap *dev_cap)
2199 if (mlx4_log_num_mgm_entry_size <= 0) {
2200 if ((-mlx4_log_num_mgm_entry_size) & MLX4_DMFS_A0_STEERING) {
2201 if (dev->caps.dmfs_high_steer_mode ==
2202 MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
2203 mlx4_err(dev, "DMFS high rate mode not supported\n");
2205 dev->caps.dmfs_high_steer_mode =
2206 MLX4_STEERING_DMFS_A0_STATIC;
2210 if (mlx4_log_num_mgm_entry_size <= 0 &&
2211 dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_FS_EN &&
2212 (!mlx4_is_mfunc(dev) ||
2213 (dev_cap->fs_max_num_qp_per_entry >=
2214 (dev->persist->num_vfs + 1))) &&
2215 choose_log_fs_mgm_entry_size(dev_cap->fs_max_num_qp_per_entry) >=
2216 MLX4_MIN_MGM_LOG_ENTRY_SIZE) {
2217 dev->oper_log_mgm_entry_size =
2218 choose_log_fs_mgm_entry_size(dev_cap->fs_max_num_qp_per_entry);
2219 dev->caps.steering_mode = MLX4_STEERING_MODE_DEVICE_MANAGED;
2220 dev->caps.num_qp_per_mgm = dev_cap->fs_max_num_qp_per_entry;
2221 dev->caps.fs_log_max_ucast_qp_range_size =
2222 dev_cap->fs_log_max_ucast_qp_range_size;
2224 if (dev->caps.dmfs_high_steer_mode !=
2225 MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
2226 dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_DISABLE;
2227 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER &&
2228 dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER)
2229 dev->caps.steering_mode = MLX4_STEERING_MODE_B0;
2231 dev->caps.steering_mode = MLX4_STEERING_MODE_A0;
2233 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER ||
2234 dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER)
2235 mlx4_warn(dev, "Must have both UC_STEER and MC_STEER flags set to use B0 steering - falling back to A0 steering mode\n");
2237 dev->oper_log_mgm_entry_size =
2238 mlx4_log_num_mgm_entry_size > 0 ?
2239 mlx4_log_num_mgm_entry_size :
2240 MLX4_DEFAULT_MGM_LOG_ENTRY_SIZE;
2241 dev->caps.num_qp_per_mgm = mlx4_get_qp_per_mgm(dev);
2243 mlx4_dbg(dev, "Steering mode is: %s, oper_log_mgm_entry_size = %d, modparam log_num_mgm_entry_size = %d\n",
2244 mlx4_steering_mode_str(dev->caps.steering_mode),
2245 dev->oper_log_mgm_entry_size,
2246 mlx4_log_num_mgm_entry_size);
2249 static void choose_tunnel_offload_mode(struct mlx4_dev *dev,
2250 struct mlx4_dev_cap *dev_cap)
2252 if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2253 dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_VXLAN_OFFLOADS)
2254 dev->caps.tunnel_offload_mode = MLX4_TUNNEL_OFFLOAD_MODE_VXLAN;
2256 dev->caps.tunnel_offload_mode = MLX4_TUNNEL_OFFLOAD_MODE_NONE;
2258 mlx4_dbg(dev, "Tunneling offload mode is: %s\n", (dev->caps.tunnel_offload_mode
2259 == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) ? "vxlan" : "none");
2262 static int mlx4_validate_optimized_steering(struct mlx4_dev *dev)
2265 struct mlx4_port_cap port_cap;
2267 if (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
2270 for (i = 1; i <= dev->caps.num_ports; i++) {
2271 if (mlx4_dev_port(dev, i, &port_cap)) {
2273 "QUERY_DEV_CAP command failed, can't verify DMFS high rate steering.\n");
2274 } else if ((dev->caps.dmfs_high_steer_mode !=
2275 MLX4_STEERING_DMFS_A0_DEFAULT) &&
2276 (port_cap.dmfs_optimized_state ==
2277 !!(dev->caps.dmfs_high_steer_mode ==
2278 MLX4_STEERING_DMFS_A0_DISABLE))) {
2280 "DMFS high rate steer mode differ, driver requested %s but %s in FW.\n",
2281 dmfs_high_rate_steering_mode_str(
2282 dev->caps.dmfs_high_steer_mode),
2283 (port_cap.dmfs_optimized_state ?
2284 "enabled" : "disabled"));
2291 static int mlx4_init_fw(struct mlx4_dev *dev)
2293 struct mlx4_mod_stat_cfg mlx4_cfg;
2296 if (!mlx4_is_slave(dev)) {
2297 err = mlx4_QUERY_FW(dev);
2300 mlx4_info(dev, "non-primary physical function, skipping\n");
2302 mlx4_err(dev, "QUERY_FW command failed, aborting\n");
2306 err = mlx4_load_fw(dev);
2308 mlx4_err(dev, "Failed to start FW, aborting\n");
2312 mlx4_cfg.log_pg_sz_m = 1;
2313 mlx4_cfg.log_pg_sz = 0;
2314 err = mlx4_MOD_STAT_CFG(dev, &mlx4_cfg);
2316 mlx4_warn(dev, "Failed to override log_pg_sz parameter\n");
2322 static int mlx4_init_hca(struct mlx4_dev *dev)
2324 struct mlx4_priv *priv = mlx4_priv(dev);
2325 struct mlx4_init_hca_param *init_hca = NULL;
2326 struct mlx4_dev_cap *dev_cap = NULL;
2327 struct mlx4_adapter adapter;
2328 struct mlx4_profile profile;
2330 struct mlx4_config_dev_params params;
2333 if (!mlx4_is_slave(dev)) {
2334 dev_cap = kzalloc(sizeof(*dev_cap), GFP_KERNEL);
2335 init_hca = kzalloc(sizeof(*init_hca), GFP_KERNEL);
2337 if (!dev_cap || !init_hca) {
2342 err = mlx4_dev_cap(dev, dev_cap);
2344 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
2348 choose_steering_mode(dev, dev_cap);
2349 choose_tunnel_offload_mode(dev, dev_cap);
2351 if (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC &&
2352 mlx4_is_master(dev))
2353 dev->caps.function_caps |= MLX4_FUNC_CAP_DMFS_A0_STATIC;
2355 err = mlx4_get_phys_port_id(dev);
2357 mlx4_err(dev, "Fail to get physical port id\n");
2359 if (mlx4_is_master(dev))
2360 mlx4_parav_master_pf_caps(dev);
2362 if (mlx4_low_memory_profile()) {
2363 mlx4_info(dev, "Running from within kdump kernel. Using low memory profile\n");
2364 profile = low_mem_profile;
2366 profile = default_profile;
2368 if (dev->caps.steering_mode ==
2369 MLX4_STEERING_MODE_DEVICE_MANAGED)
2370 profile.num_mcg = MLX4_FS_NUM_MCG;
2372 icm_size = mlx4_make_profile(dev, &profile, dev_cap,
2374 if ((long long) icm_size < 0) {
2379 if (enable_4k_uar || !dev->persist->num_vfs) {
2380 init_hca->log_uar_sz = ilog2(dev->caps.num_uars) +
2381 PAGE_SHIFT - DEFAULT_UAR_PAGE_SHIFT;
2382 init_hca->uar_page_sz = DEFAULT_UAR_PAGE_SHIFT - 12;
2384 init_hca->log_uar_sz = ilog2(dev->caps.num_uars);
2385 init_hca->uar_page_sz = PAGE_SHIFT - 12;
2388 init_hca->mw_enabled = 0;
2389 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2390 dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN)
2391 init_hca->mw_enabled = INIT_HCA_TPT_MW_ENABLE;
2393 err = mlx4_init_icm(dev, dev_cap, init_hca, icm_size);
2397 err = mlx4_INIT_HCA(dev, init_hca);
2399 mlx4_err(dev, "INIT_HCA command failed, aborting\n");
2403 if (dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS) {
2404 err = mlx4_query_func(dev, dev_cap);
2406 mlx4_err(dev, "QUERY_FUNC command failed, aborting.\n");
2408 } else if (err & MLX4_QUERY_FUNC_NUM_SYS_EQS) {
2409 dev->caps.num_eqs = dev_cap->max_eqs;
2410 dev->caps.reserved_eqs = dev_cap->reserved_eqs;
2411 dev->caps.reserved_uars = dev_cap->reserved_uars;
2416 * If TS is supported by FW
2417 * read HCA frequency by QUERY_HCA command
2419 if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS) {
2420 err = mlx4_QUERY_HCA(dev, init_hca);
2422 mlx4_err(dev, "QUERY_HCA command failed, disable timestamp\n");
2423 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
2425 dev->caps.hca_core_clock =
2426 init_hca->hca_core_clock;
2429 /* In case we got HCA frequency 0 - disable timestamping
2430 * to avoid dividing by zero
2432 if (!dev->caps.hca_core_clock) {
2433 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
2435 "HCA frequency is 0 - timestamping is not supported\n");
2436 } else if (map_internal_clock(dev)) {
2438 * Map internal clock,
2439 * in case of failure disable timestamping
2441 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
2442 mlx4_err(dev, "Failed to map internal clock. Timestamping is not supported\n");
2446 if (dev->caps.dmfs_high_steer_mode !=
2447 MLX4_STEERING_DMFS_A0_NOT_SUPPORTED) {
2448 if (mlx4_validate_optimized_steering(dev))
2449 mlx4_warn(dev, "Optimized steering validation failed\n");
2451 if (dev->caps.dmfs_high_steer_mode ==
2452 MLX4_STEERING_DMFS_A0_DISABLE) {
2453 dev->caps.dmfs_high_rate_qpn_base =
2454 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
2455 dev->caps.dmfs_high_rate_qpn_range =
2456 MLX4_A0_STEERING_TABLE_SIZE;
2459 mlx4_info(dev, "DMFS high rate steer mode is: %s\n",
2460 dmfs_high_rate_steering_mode_str(
2461 dev->caps.dmfs_high_steer_mode));
2464 err = mlx4_init_slave(dev);
2466 if (err != -EPROBE_DEFER)
2467 mlx4_err(dev, "Failed to initialize slave\n");
2471 err = mlx4_slave_cap(dev);
2473 mlx4_err(dev, "Failed to obtain slave caps\n");
2478 if (map_bf_area(dev))
2479 mlx4_dbg(dev, "Failed to map blue flame area\n");
2481 /*Only the master set the ports, all the rest got it from it.*/
2482 if (!mlx4_is_slave(dev))
2483 mlx4_set_port_mask(dev);
2485 err = mlx4_QUERY_ADAPTER(dev, &adapter);
2487 mlx4_err(dev, "QUERY_ADAPTER command failed, aborting\n");
2491 /* Query CONFIG_DEV parameters */
2492 err = mlx4_config_dev_retrieval(dev, ¶ms);
2493 if (err && err != -EOPNOTSUPP) {
2494 mlx4_err(dev, "Failed to query CONFIG_DEV parameters\n");
2496 dev->caps.rx_checksum_flags_port[1] = params.rx_csum_flags_port_1;
2497 dev->caps.rx_checksum_flags_port[2] = params.rx_csum_flags_port_2;
2499 priv->eq_table.inta_pin = adapter.inta_pin;
2500 memcpy(dev->board_id, adapter.board_id, sizeof(dev->board_id));
2506 unmap_internal_clock(dev);
2509 if (mlx4_is_slave(dev))
2510 mlx4_slave_destroy_special_qp_cap(dev);
2513 if (mlx4_is_slave(dev))
2514 mlx4_slave_exit(dev);
2516 mlx4_CLOSE_HCA(dev, 0);
2519 if (!mlx4_is_slave(dev))
2520 mlx4_free_icms(dev);
2529 static int mlx4_init_counters_table(struct mlx4_dev *dev)
2531 struct mlx4_priv *priv = mlx4_priv(dev);
2534 if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2537 if (!dev->caps.max_counters)
2540 nent_pow2 = roundup_pow_of_two(dev->caps.max_counters);
2541 /* reserve last counter index for sink counter */
2542 return mlx4_bitmap_init(&priv->counters_bitmap, nent_pow2,
2544 nent_pow2 - dev->caps.max_counters + 1);
2547 static void mlx4_cleanup_counters_table(struct mlx4_dev *dev)
2549 if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2552 if (!dev->caps.max_counters)
2555 mlx4_bitmap_cleanup(&mlx4_priv(dev)->counters_bitmap);
2558 static void mlx4_cleanup_default_counters(struct mlx4_dev *dev)
2560 struct mlx4_priv *priv = mlx4_priv(dev);
2563 for (port = 0; port < dev->caps.num_ports; port++)
2564 if (priv->def_counter[port] != -1)
2565 mlx4_counter_free(dev, priv->def_counter[port]);
2568 static int mlx4_allocate_default_counters(struct mlx4_dev *dev)
2570 struct mlx4_priv *priv = mlx4_priv(dev);
2574 for (port = 0; port < dev->caps.num_ports; port++)
2575 priv->def_counter[port] = -1;
2577 for (port = 0; port < dev->caps.num_ports; port++) {
2578 err = mlx4_counter_alloc(dev, &idx, MLX4_RES_USAGE_DRIVER);
2580 if (!err || err == -ENOSPC) {
2581 priv->def_counter[port] = idx;
2583 } else if (err == -ENOENT) {
2586 } else if (mlx4_is_slave(dev) && err == -EINVAL) {
2587 priv->def_counter[port] = MLX4_SINK_COUNTER_INDEX(dev);
2588 mlx4_warn(dev, "can't allocate counter from old PF driver, using index %d\n",
2589 MLX4_SINK_COUNTER_INDEX(dev));
2592 mlx4_err(dev, "%s: failed to allocate default counter port %d err %d\n",
2593 __func__, port + 1, err);
2594 mlx4_cleanup_default_counters(dev);
2598 mlx4_dbg(dev, "%s: default counter index %d for port %d\n",
2599 __func__, priv->def_counter[port], port + 1);
2605 int __mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx)
2607 struct mlx4_priv *priv = mlx4_priv(dev);
2609 if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2612 *idx = mlx4_bitmap_alloc(&priv->counters_bitmap);
2614 *idx = MLX4_SINK_COUNTER_INDEX(dev);
2621 int mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx, u8 usage)
2623 u32 in_modifier = RES_COUNTER | (((u32)usage & 3) << 30);
2627 if (mlx4_is_mfunc(dev)) {
2628 err = mlx4_cmd_imm(dev, 0, &out_param, in_modifier,
2629 RES_OP_RESERVE, MLX4_CMD_ALLOC_RES,
2630 MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
2632 *idx = get_param_l(&out_param);
2633 if (WARN_ON(err == -ENOSPC))
2637 return __mlx4_counter_alloc(dev, idx);
2639 EXPORT_SYMBOL_GPL(mlx4_counter_alloc);
2641 static int __mlx4_clear_if_stat(struct mlx4_dev *dev,
2644 struct mlx4_cmd_mailbox *if_stat_mailbox;
2646 u32 if_stat_in_mod = (counter_index & 0xff) | MLX4_QUERY_IF_STAT_RESET;
2648 if_stat_mailbox = mlx4_alloc_cmd_mailbox(dev);
2649 if (IS_ERR(if_stat_mailbox))
2650 return PTR_ERR(if_stat_mailbox);
2652 err = mlx4_cmd_box(dev, 0, if_stat_mailbox->dma, if_stat_in_mod, 0,
2653 MLX4_CMD_QUERY_IF_STAT, MLX4_CMD_TIME_CLASS_C,
2656 mlx4_free_cmd_mailbox(dev, if_stat_mailbox);
2660 void __mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
2662 if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2665 if (idx == MLX4_SINK_COUNTER_INDEX(dev))
2668 __mlx4_clear_if_stat(dev, idx);
2670 mlx4_bitmap_free(&mlx4_priv(dev)->counters_bitmap, idx, MLX4_USE_RR);
2674 void mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
2678 if (mlx4_is_mfunc(dev)) {
2679 set_param_l(&in_param, idx);
2680 mlx4_cmd(dev, in_param, RES_COUNTER, RES_OP_RESERVE,
2681 MLX4_CMD_FREE_RES, MLX4_CMD_TIME_CLASS_A,
2685 __mlx4_counter_free(dev, idx);
2687 EXPORT_SYMBOL_GPL(mlx4_counter_free);
2689 int mlx4_get_default_counter_index(struct mlx4_dev *dev, int port)
2691 struct mlx4_priv *priv = mlx4_priv(dev);
2693 return priv->def_counter[port - 1];
2695 EXPORT_SYMBOL_GPL(mlx4_get_default_counter_index);
2697 void mlx4_set_admin_guid(struct mlx4_dev *dev, __be64 guid, int entry, int port)
2699 struct mlx4_priv *priv = mlx4_priv(dev);
2701 priv->mfunc.master.vf_admin[entry].vport[port].guid = guid;
2703 EXPORT_SYMBOL_GPL(mlx4_set_admin_guid);
2705 __be64 mlx4_get_admin_guid(struct mlx4_dev *dev, int entry, int port)
2707 struct mlx4_priv *priv = mlx4_priv(dev);
2709 return priv->mfunc.master.vf_admin[entry].vport[port].guid;
2711 EXPORT_SYMBOL_GPL(mlx4_get_admin_guid);
2713 void mlx4_set_random_admin_guid(struct mlx4_dev *dev, int entry, int port)
2715 struct mlx4_priv *priv = mlx4_priv(dev);
2722 get_random_bytes((char *)&guid, sizeof(guid));
2723 guid &= ~(cpu_to_be64(1ULL << 56));
2724 guid |= cpu_to_be64(1ULL << 57);
2725 priv->mfunc.master.vf_admin[entry].vport[port].guid = guid;
2728 static int mlx4_setup_hca(struct mlx4_dev *dev)
2730 struct mlx4_priv *priv = mlx4_priv(dev);
2733 __be32 ib_port_default_caps;
2735 err = mlx4_init_uar_table(dev);
2737 mlx4_err(dev, "Failed to initialize user access region table, aborting\n");
2741 err = mlx4_uar_alloc(dev, &priv->driver_uar);
2743 mlx4_err(dev, "Failed to allocate driver access region, aborting\n");
2744 goto err_uar_table_free;
2747 priv->kar = ioremap((phys_addr_t) priv->driver_uar.pfn << PAGE_SHIFT, PAGE_SIZE);
2749 mlx4_err(dev, "Couldn't map kernel access region, aborting\n");
2754 err = mlx4_init_pd_table(dev);
2756 mlx4_err(dev, "Failed to initialize protection domain table, aborting\n");
2760 err = mlx4_init_xrcd_table(dev);
2762 mlx4_err(dev, "Failed to initialize reliable connection domain table, aborting\n");
2763 goto err_pd_table_free;
2766 err = mlx4_init_mr_table(dev);
2768 mlx4_err(dev, "Failed to initialize memory region table, aborting\n");
2769 goto err_xrcd_table_free;
2772 if (!mlx4_is_slave(dev)) {
2773 err = mlx4_init_mcg_table(dev);
2775 mlx4_err(dev, "Failed to initialize multicast group table, aborting\n");
2776 goto err_mr_table_free;
2778 err = mlx4_config_mad_demux(dev);
2780 mlx4_err(dev, "Failed in config_mad_demux, aborting\n");
2781 goto err_mcg_table_free;
2785 err = mlx4_init_eq_table(dev);
2787 mlx4_err(dev, "Failed to initialize event queue table, aborting\n");
2788 goto err_mcg_table_free;
2791 err = mlx4_cmd_use_events(dev);
2793 mlx4_err(dev, "Failed to switch to event-driven firmware commands, aborting\n");
2794 goto err_eq_table_free;
2797 err = mlx4_NOP(dev);
2799 if (dev->flags & MLX4_FLAG_MSI_X) {
2800 mlx4_warn(dev, "NOP command failed to generate MSI-X interrupt IRQ %d)\n",
2801 priv->eq_table.eq[MLX4_EQ_ASYNC].irq);
2802 mlx4_warn(dev, "Trying again without MSI-X\n");
2804 mlx4_err(dev, "NOP command failed to generate interrupt (IRQ %d), aborting\n",
2805 priv->eq_table.eq[MLX4_EQ_ASYNC].irq);
2806 mlx4_err(dev, "BIOS or ACPI interrupt routing problem?\n");
2812 mlx4_dbg(dev, "NOP command IRQ test passed\n");
2814 err = mlx4_init_cq_table(dev);
2816 mlx4_err(dev, "Failed to initialize completion queue table, aborting\n");
2820 err = mlx4_init_srq_table(dev);
2822 mlx4_err(dev, "Failed to initialize shared receive queue table, aborting\n");
2823 goto err_cq_table_free;
2826 err = mlx4_init_qp_table(dev);
2828 mlx4_err(dev, "Failed to initialize queue pair table, aborting\n");
2829 goto err_srq_table_free;
2832 if (!mlx4_is_slave(dev)) {
2833 err = mlx4_init_counters_table(dev);
2834 if (err && err != -ENOENT) {
2835 mlx4_err(dev, "Failed to initialize counters table, aborting\n");
2836 goto err_qp_table_free;
2840 err = mlx4_allocate_default_counters(dev);
2842 mlx4_err(dev, "Failed to allocate default counters, aborting\n");
2843 goto err_counters_table_free;
2846 if (!mlx4_is_slave(dev)) {
2847 for (port = 1; port <= dev->caps.num_ports; port++) {
2848 ib_port_default_caps = 0;
2849 err = mlx4_get_port_ib_caps(dev, port,
2850 &ib_port_default_caps);
2852 mlx4_warn(dev, "failed to get port %d default ib capabilities (%d). Continuing with caps = 0\n",
2854 dev->caps.ib_port_def_cap[port] = ib_port_default_caps;
2856 /* initialize per-slave default ib port capabilities */
2857 if (mlx4_is_master(dev)) {
2859 for (i = 0; i < dev->num_slaves; i++) {
2860 if (i == mlx4_master_func_num(dev))
2862 priv->mfunc.master.slave_state[i].ib_cap_mask[port] =
2863 ib_port_default_caps;
2867 if (mlx4_is_mfunc(dev))
2868 dev->caps.port_ib_mtu[port] = IB_MTU_2048;
2870 dev->caps.port_ib_mtu[port] = IB_MTU_4096;
2872 err = mlx4_SET_PORT(dev, port, mlx4_is_master(dev) ?
2873 dev->caps.pkey_table_len[port] : -1);
2875 mlx4_err(dev, "Failed to set port %d, aborting\n",
2877 goto err_default_countes_free;
2884 err_default_countes_free:
2885 mlx4_cleanup_default_counters(dev);
2887 err_counters_table_free:
2888 if (!mlx4_is_slave(dev))
2889 mlx4_cleanup_counters_table(dev);
2892 mlx4_cleanup_qp_table(dev);
2895 mlx4_cleanup_srq_table(dev);
2898 mlx4_cleanup_cq_table(dev);
2901 mlx4_cmd_use_polling(dev);
2904 mlx4_cleanup_eq_table(dev);
2907 if (!mlx4_is_slave(dev))
2908 mlx4_cleanup_mcg_table(dev);
2911 mlx4_cleanup_mr_table(dev);
2913 err_xrcd_table_free:
2914 mlx4_cleanup_xrcd_table(dev);
2917 mlx4_cleanup_pd_table(dev);
2923 mlx4_uar_free(dev, &priv->driver_uar);
2926 mlx4_cleanup_uar_table(dev);
2930 static int mlx4_init_affinity_hint(struct mlx4_dev *dev, int port, int eqn)
2932 int requested_cpu = 0;
2933 struct mlx4_priv *priv = mlx4_priv(dev);
2938 if (eqn > dev->caps.num_comp_vectors)
2941 for (i = 1; i < port; i++)
2942 off += mlx4_get_eqs_per_port(dev, i);
2944 requested_cpu = eqn - off - !!(eqn > MLX4_EQ_ASYNC);
2946 /* Meaning EQs are shared, and this call comes from the second port */
2947 if (requested_cpu < 0)
2950 eq = &priv->eq_table.eq[eqn];
2952 if (!zalloc_cpumask_var(&eq->affinity_mask, GFP_KERNEL))
2955 cpumask_set_cpu(requested_cpu, eq->affinity_mask);
2960 static void mlx4_enable_msi_x(struct mlx4_dev *dev)
2962 struct mlx4_priv *priv = mlx4_priv(dev);
2963 struct msix_entry *entries;
2968 int nreq = min3(dev->caps.num_ports *
2969 (int)num_online_cpus() + 1,
2970 dev->caps.num_eqs - dev->caps.reserved_eqs,
2974 nreq = min_t(int, nreq, msi_x);
2976 entries = kcalloc(nreq, sizeof(*entries), GFP_KERNEL);
2980 for (i = 0; i < nreq; ++i)
2981 entries[i].entry = i;
2983 nreq = pci_enable_msix_range(dev->persist->pdev, entries, 2,
2986 if (nreq < 0 || nreq < MLX4_EQ_ASYNC) {
2990 /* 1 is reserved for events (asyncrounous EQ) */
2991 dev->caps.num_comp_vectors = nreq - 1;
2993 priv->eq_table.eq[MLX4_EQ_ASYNC].irq = entries[0].vector;
2994 bitmap_zero(priv->eq_table.eq[MLX4_EQ_ASYNC].actv_ports.ports,
2995 dev->caps.num_ports);
2997 for (i = 0; i < dev->caps.num_comp_vectors + 1; i++) {
2998 if (i == MLX4_EQ_ASYNC)
3001 priv->eq_table.eq[i].irq =
3002 entries[i + 1 - !!(i > MLX4_EQ_ASYNC)].vector;
3004 if (MLX4_IS_LEGACY_EQ_MODE(dev->caps)) {
3005 bitmap_fill(priv->eq_table.eq[i].actv_ports.ports,
3006 dev->caps.num_ports);
3007 /* We don't set affinity hint when there
3012 priv->eq_table.eq[i].actv_ports.ports);
3013 if (mlx4_init_affinity_hint(dev, port + 1, i))
3014 mlx4_warn(dev, "Couldn't init hint cpumask for EQ %d\n",
3017 /* We divide the Eqs evenly between the two ports.
3018 * (dev->caps.num_comp_vectors / dev->caps.num_ports)
3019 * refers to the number of Eqs per port
3020 * (i.e eqs_per_port). Theoretically, we would like to
3021 * write something like (i + 1) % eqs_per_port == 0.
3022 * However, since there's an asynchronous Eq, we have
3023 * to skip over it by comparing this condition to
3024 * !!((i + 1) > MLX4_EQ_ASYNC).
3026 if ((dev->caps.num_comp_vectors > dev->caps.num_ports) &&
3028 (dev->caps.num_comp_vectors / dev->caps.num_ports)) ==
3029 !!((i + 1) > MLX4_EQ_ASYNC))
3030 /* If dev->caps.num_comp_vectors < dev->caps.num_ports,
3031 * everything is shared anyway.
3036 dev->flags |= MLX4_FLAG_MSI_X;
3043 dev->caps.num_comp_vectors = 1;
3045 BUG_ON(MLX4_EQ_ASYNC >= 2);
3046 for (i = 0; i < 2; ++i) {
3047 priv->eq_table.eq[i].irq = dev->persist->pdev->irq;
3048 if (i != MLX4_EQ_ASYNC) {
3049 bitmap_fill(priv->eq_table.eq[i].actv_ports.ports,
3050 dev->caps.num_ports);
3055 static int mlx4_devlink_port_type_set(struct devlink_port *devlink_port,
3056 enum devlink_port_type port_type)
3058 struct mlx4_port_info *info = container_of(devlink_port,
3059 struct mlx4_port_info,
3061 enum mlx4_port_type mlx4_port_type;
3063 switch (port_type) {
3064 case DEVLINK_PORT_TYPE_AUTO:
3065 mlx4_port_type = MLX4_PORT_TYPE_AUTO;
3067 case DEVLINK_PORT_TYPE_ETH:
3068 mlx4_port_type = MLX4_PORT_TYPE_ETH;
3070 case DEVLINK_PORT_TYPE_IB:
3071 mlx4_port_type = MLX4_PORT_TYPE_IB;
3077 return __set_port_type(info, mlx4_port_type);
3080 static const struct devlink_port_ops mlx4_devlink_port_ops = {
3081 .port_type_set = mlx4_devlink_port_type_set,
3084 static int mlx4_init_port_info(struct mlx4_dev *dev, int port)
3086 struct devlink *devlink = priv_to_devlink(mlx4_priv(dev));
3087 struct mlx4_port_info *info = &mlx4_priv(dev)->port[port];
3090 err = devl_port_register_with_ops(devlink, &info->devlink_port, port,
3091 &mlx4_devlink_port_ops);
3095 /* Ethernet and IB drivers will normally set the port type,
3096 * but if they are not built set the type now to prevent
3097 * devlink_port_type_warn() from firing.
3099 if (!IS_ENABLED(CONFIG_MLX4_EN) &&
3100 dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH)
3101 devlink_port_type_eth_set(&info->devlink_port);
3102 else if (!IS_ENABLED(CONFIG_MLX4_INFINIBAND) &&
3103 dev->caps.port_type[port] == MLX4_PORT_TYPE_IB)
3104 devlink_port_type_ib_set(&info->devlink_port, NULL);
3108 if (!mlx4_is_slave(dev)) {
3109 mlx4_init_mac_table(dev, &info->mac_table);
3110 mlx4_init_vlan_table(dev, &info->vlan_table);
3111 mlx4_init_roce_gid_table(dev, &info->gid_table);
3112 info->base_qpn = mlx4_get_base_qpn(dev, port);
3115 sprintf(info->dev_name, "mlx4_port%d", port);
3116 info->port_attr.attr.name = info->dev_name;
3117 if (mlx4_is_mfunc(dev)) {
3118 info->port_attr.attr.mode = 0444;
3120 info->port_attr.attr.mode = 0644;
3121 info->port_attr.store = set_port_type;
3123 info->port_attr.show = show_port_type;
3124 sysfs_attr_init(&info->port_attr.attr);
3126 err = device_create_file(&dev->persist->pdev->dev, &info->port_attr);
3128 mlx4_err(dev, "Failed to create file for port %d\n", port);
3129 devlink_port_type_clear(&info->devlink_port);
3130 devl_port_unregister(&info->devlink_port);
3135 sprintf(info->dev_mtu_name, "mlx4_port%d_mtu", port);
3136 info->port_mtu_attr.attr.name = info->dev_mtu_name;
3137 if (mlx4_is_mfunc(dev)) {
3138 info->port_mtu_attr.attr.mode = 0444;
3140 info->port_mtu_attr.attr.mode = 0644;
3141 info->port_mtu_attr.store = set_port_ib_mtu;
3143 info->port_mtu_attr.show = show_port_ib_mtu;
3144 sysfs_attr_init(&info->port_mtu_attr.attr);
3146 err = device_create_file(&dev->persist->pdev->dev,
3147 &info->port_mtu_attr);
3149 mlx4_err(dev, "Failed to create mtu file for port %d\n", port);
3150 device_remove_file(&info->dev->persist->pdev->dev,
3152 devlink_port_type_clear(&info->devlink_port);
3153 devl_port_unregister(&info->devlink_port);
3161 static void mlx4_cleanup_port_info(struct mlx4_port_info *info)
3166 device_remove_file(&info->dev->persist->pdev->dev, &info->port_attr);
3167 device_remove_file(&info->dev->persist->pdev->dev,
3168 &info->port_mtu_attr);
3169 devlink_port_type_clear(&info->devlink_port);
3170 devl_port_unregister(&info->devlink_port);
3172 #ifdef CONFIG_RFS_ACCEL
3173 free_irq_cpu_rmap(info->rmap);
3178 static int mlx4_init_steering(struct mlx4_dev *dev)
3180 struct mlx4_priv *priv = mlx4_priv(dev);
3181 int num_entries = dev->caps.num_ports;
3184 priv->steer = kcalloc(num_entries, sizeof(struct mlx4_steer),
3189 for (i = 0; i < num_entries; i++)
3190 for (j = 0; j < MLX4_NUM_STEERS; j++) {
3191 INIT_LIST_HEAD(&priv->steer[i].promisc_qps[j]);
3192 INIT_LIST_HEAD(&priv->steer[i].steer_entries[j]);
3197 static void mlx4_clear_steering(struct mlx4_dev *dev)
3199 struct mlx4_priv *priv = mlx4_priv(dev);
3200 struct mlx4_steer_index *entry, *tmp_entry;
3201 struct mlx4_promisc_qp *pqp, *tmp_pqp;
3202 int num_entries = dev->caps.num_ports;
3205 for (i = 0; i < num_entries; i++) {
3206 for (j = 0; j < MLX4_NUM_STEERS; j++) {
3207 list_for_each_entry_safe(pqp, tmp_pqp,
3208 &priv->steer[i].promisc_qps[j],
3210 list_del(&pqp->list);
3213 list_for_each_entry_safe(entry, tmp_entry,
3214 &priv->steer[i].steer_entries[j],
3216 list_del(&entry->list);
3217 list_for_each_entry_safe(pqp, tmp_pqp,
3220 list_del(&pqp->list);
3230 static int extended_func_num(struct pci_dev *pdev)
3232 return PCI_SLOT(pdev->devfn) * 8 + PCI_FUNC(pdev->devfn);
3235 #define MLX4_OWNER_BASE 0x8069c
3236 #define MLX4_OWNER_SIZE 4
3238 static int mlx4_get_ownership(struct mlx4_dev *dev)
3240 void __iomem *owner;
3243 if (pci_channel_offline(dev->persist->pdev))
3246 owner = ioremap(pci_resource_start(dev->persist->pdev, 0) +
3250 mlx4_err(dev, "Failed to obtain ownership bit\n");
3259 static void mlx4_free_ownership(struct mlx4_dev *dev)
3261 void __iomem *owner;
3263 if (pci_channel_offline(dev->persist->pdev))
3266 owner = ioremap(pci_resource_start(dev->persist->pdev, 0) +
3270 mlx4_err(dev, "Failed to obtain ownership bit\n");
3278 #define SRIOV_VALID_STATE(flags) (!!((flags) & MLX4_FLAG_SRIOV) ==\
3279 !!((flags) & MLX4_FLAG_MASTER))
3281 static u64 mlx4_enable_sriov(struct mlx4_dev *dev, struct pci_dev *pdev,
3282 u8 total_vfs, int existing_vfs, int reset_flow)
3284 u64 dev_flags = dev->flags;
3286 int fw_enabled_sriov_vfs = min(pci_sriov_get_totalvfs(pdev),
3290 dev->dev_vfs = kcalloc(total_vfs, sizeof(*dev->dev_vfs),
3297 atomic_inc(&pf_loading);
3298 if (dev->flags & MLX4_FLAG_SRIOV) {
3299 if (existing_vfs != total_vfs) {
3300 mlx4_err(dev, "SR-IOV was already enabled, but with num_vfs (%d) different than requested (%d)\n",
3301 existing_vfs, total_vfs);
3302 total_vfs = existing_vfs;
3306 dev->dev_vfs = kcalloc(total_vfs, sizeof(*dev->dev_vfs), GFP_KERNEL);
3307 if (NULL == dev->dev_vfs) {
3308 mlx4_err(dev, "Failed to allocate memory for VFs\n");
3312 if (!(dev->flags & MLX4_FLAG_SRIOV)) {
3313 if (total_vfs > fw_enabled_sriov_vfs) {
3314 mlx4_err(dev, "requested vfs (%d) > available vfs (%d). Continuing without SR_IOV\n",
3315 total_vfs, fw_enabled_sriov_vfs);
3319 mlx4_warn(dev, "Enabling SR-IOV with %d VFs\n", total_vfs);
3320 err = pci_enable_sriov(pdev, total_vfs);
3323 mlx4_err(dev, "Failed to enable SR-IOV, continuing without SR-IOV (err = %d)\n",
3327 mlx4_warn(dev, "Running in master mode\n");
3328 dev_flags |= MLX4_FLAG_SRIOV |
3330 dev_flags &= ~MLX4_FLAG_SLAVE;
3331 dev->persist->num_vfs = total_vfs;
3336 atomic_dec(&pf_loading);
3338 dev->persist->num_vfs = 0;
3339 kfree(dev->dev_vfs);
3340 dev->dev_vfs = NULL;
3341 return dev_flags & ~MLX4_FLAG_MASTER;
3345 MLX4_DEV_CAP_CHECK_NUM_VFS_ABOVE_64 = -1,
3348 static int mlx4_check_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap,
3351 int requested_vfs = nvfs[0] + nvfs[1] + nvfs[2];
3352 /* Checking for 64 VFs as a limitation of CX2 */
3353 if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_80_VFS) &&
3354 requested_vfs >= 64) {
3355 mlx4_err(dev, "Requested %d VFs, but FW does not support more than 64\n",
3357 return MLX4_DEV_CAP_CHECK_NUM_VFS_ABOVE_64;
3362 static int mlx4_pci_enable_device(struct mlx4_dev *dev)
3364 struct pci_dev *pdev = dev->persist->pdev;
3367 mutex_lock(&dev->persist->pci_status_mutex);
3368 if (dev->persist->pci_status == MLX4_PCI_STATUS_DISABLED) {
3369 err = pci_enable_device(pdev);
3371 dev->persist->pci_status = MLX4_PCI_STATUS_ENABLED;
3373 mutex_unlock(&dev->persist->pci_status_mutex);
3378 static void mlx4_pci_disable_device(struct mlx4_dev *dev)
3380 struct pci_dev *pdev = dev->persist->pdev;
3382 mutex_lock(&dev->persist->pci_status_mutex);
3383 if (dev->persist->pci_status == MLX4_PCI_STATUS_ENABLED) {
3384 pci_disable_device(pdev);
3385 dev->persist->pci_status = MLX4_PCI_STATUS_DISABLED;
3387 mutex_unlock(&dev->persist->pci_status_mutex);
3390 static int mlx4_load_one(struct pci_dev *pdev, int pci_dev_data,
3391 int total_vfs, int *nvfs, struct mlx4_priv *priv,
3394 struct devlink *devlink = priv_to_devlink(priv);
3395 struct mlx4_dev *dev;
3400 struct mlx4_dev_cap *dev_cap = NULL;
3401 int existing_vfs = 0;
3403 devl_assert_locked(devlink);
3406 err = mlx4_adev_init(dev);
3410 ATOMIC_INIT_NOTIFIER_HEAD(&priv->event_nh);
3412 mutex_init(&priv->port_mutex);
3413 mutex_init(&priv->bond_mutex);
3415 INIT_LIST_HEAD(&priv->pgdir_list);
3416 mutex_init(&priv->pgdir_mutex);
3417 spin_lock_init(&priv->cmd.context_lock);
3419 INIT_LIST_HEAD(&priv->bf_list);
3420 mutex_init(&priv->bf_mutex);
3422 dev->rev_id = pdev->revision;
3423 dev->numa_node = dev_to_node(&pdev->dev);
3425 /* Detect if this device is a virtual function */
3426 if (pci_dev_data & MLX4_PCI_DEV_IS_VF) {
3427 mlx4_warn(dev, "Detected virtual function - running in slave mode\n");
3428 dev->flags |= MLX4_FLAG_SLAVE;
3430 /* We reset the device and enable SRIOV only for physical
3431 * devices. Try to claim ownership on the device;
3432 * if already taken, skip -- do not allow multiple PFs */
3433 err = mlx4_get_ownership(dev);
3438 mlx4_warn(dev, "Multiple PFs not yet supported - Skipping PF\n");
3444 atomic_set(&priv->opreq_count, 0);
3445 INIT_WORK(&priv->opreq_task, mlx4_opreq_action);
3448 * Now reset the HCA before we touch the PCI capabilities or
3449 * attempt a firmware command, since a boot ROM may have left
3450 * the HCA in an undefined state.
3452 err = mlx4_reset(dev);
3454 mlx4_err(dev, "Failed to reset HCA, aborting\n");
3459 dev->flags = MLX4_FLAG_MASTER;
3460 existing_vfs = pci_num_vf(pdev);
3462 dev->flags |= MLX4_FLAG_SRIOV;
3463 dev->persist->num_vfs = total_vfs;
3467 /* on load remove any previous indication of internal error,
3470 dev->persist->state = MLX4_DEVICE_STATE_UP;
3473 err = mlx4_cmd_init(dev);
3475 mlx4_err(dev, "Failed to init command interface, aborting\n");
3479 /* In slave functions, the communication channel must be initialized
3480 * before posting commands. Also, init num_slaves before calling
3482 if (mlx4_is_mfunc(dev)) {
3483 if (mlx4_is_master(dev)) {
3484 dev->num_slaves = MLX4_MAX_NUM_SLAVES;
3487 dev->num_slaves = 0;
3488 err = mlx4_multi_func_init(dev);
3490 mlx4_err(dev, "Failed to init slave mfunc interface, aborting\n");
3496 err = mlx4_init_fw(dev);
3498 mlx4_err(dev, "Failed to init fw, aborting.\n");
3502 if (mlx4_is_master(dev)) {
3503 /* when we hit the goto slave_start below, dev_cap already initialized */
3505 dev_cap = kzalloc(sizeof(*dev_cap), GFP_KERNEL);
3512 err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
3514 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
3518 if (mlx4_check_dev_cap(dev, dev_cap, nvfs))
3521 if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS)) {
3522 u64 dev_flags = mlx4_enable_sriov(dev, pdev,
3528 mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3529 dev->flags = dev_flags;
3530 if (!SRIOV_VALID_STATE(dev->flags)) {
3531 mlx4_err(dev, "Invalid SRIOV state\n");
3534 err = mlx4_reset(dev);
3536 mlx4_err(dev, "Failed to reset HCA, aborting.\n");
3542 /* Legacy mode FW requires SRIOV to be enabled before
3543 * doing QUERY_DEV_CAP, since max_eq's value is different if
3546 memset(dev_cap, 0, sizeof(*dev_cap));
3547 err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
3549 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
3553 if (mlx4_check_dev_cap(dev, dev_cap, nvfs))
3558 err = mlx4_init_hca(dev);
3560 if (err == -EACCES) {
3561 /* Not primary Physical function
3562 * Running in slave mode */
3563 mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3564 /* We're not a PF */
3565 if (dev->flags & MLX4_FLAG_SRIOV) {
3567 pci_disable_sriov(pdev);
3568 if (mlx4_is_master(dev) && !reset_flow)
3569 atomic_dec(&pf_loading);
3570 dev->flags &= ~MLX4_FLAG_SRIOV;
3572 if (!mlx4_is_slave(dev))
3573 mlx4_free_ownership(dev);
3574 dev->flags |= MLX4_FLAG_SLAVE;
3575 dev->flags &= ~MLX4_FLAG_MASTER;
3581 if (mlx4_is_master(dev) && (dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS)) {
3582 u64 dev_flags = mlx4_enable_sriov(dev, pdev, total_vfs,
3583 existing_vfs, reset_flow);
3585 if ((dev->flags ^ dev_flags) & (MLX4_FLAG_MASTER | MLX4_FLAG_SLAVE)) {
3586 mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_VHCR);
3587 dev->flags = dev_flags;
3588 err = mlx4_cmd_init(dev);
3590 /* Only VHCR is cleaned up, so could still
3593 mlx4_err(dev, "Failed to init VHCR command interface, aborting\n");
3597 dev->flags = dev_flags;
3600 if (!SRIOV_VALID_STATE(dev->flags)) {
3601 mlx4_err(dev, "Invalid SRIOV state\n");
3607 /* check if the device is functioning at its maximum possible speed.
3608 * No return code for this call, just warn the user in case of PCI
3609 * express device capabilities are under-satisfied by the bus.
3611 if (!mlx4_is_slave(dev))
3612 pcie_print_link_status(dev->persist->pdev);
3614 /* In master functions, the communication channel must be initialized
3615 * after obtaining its address from fw */
3616 if (mlx4_is_master(dev)) {
3617 if (dev->caps.num_ports < 2 &&
3621 "Error: Trying to configure VFs on port 2, but HCA has only %d physical ports\n",
3622 dev->caps.num_ports);
3625 memcpy(dev->persist->nvfs, nvfs, sizeof(dev->persist->nvfs));
3628 i < sizeof(dev->persist->nvfs)/
3629 sizeof(dev->persist->nvfs[0]); i++) {
3632 for (j = 0; j < dev->persist->nvfs[i]; ++sum, ++j) {
3633 dev->dev_vfs[sum].min_port = i < 2 ? i + 1 : 1;
3634 dev->dev_vfs[sum].n_ports = i < 2 ? 1 :
3635 dev->caps.num_ports;
3639 /* In master functions, the communication channel
3640 * must be initialized after obtaining its address from fw
3642 err = mlx4_multi_func_init(dev);
3644 mlx4_err(dev, "Failed to init master mfunc interface, aborting.\n");
3649 err = mlx4_alloc_eq_table(dev);
3651 goto err_master_mfunc;
3653 bitmap_zero(priv->msix_ctl.pool_bm, MAX_MSIX);
3654 mutex_init(&priv->msix_ctl.pool_lock);
3656 mlx4_enable_msi_x(dev);
3657 if ((mlx4_is_mfunc(dev)) &&
3658 !(dev->flags & MLX4_FLAG_MSI_X)) {
3660 mlx4_err(dev, "INTx is not supported in multi-function mode, aborting\n");
3664 if (!mlx4_is_slave(dev)) {
3665 err = mlx4_init_steering(dev);
3667 goto err_disable_msix;
3670 mlx4_init_quotas(dev);
3672 err = mlx4_setup_hca(dev);
3673 if (err == -EBUSY && (dev->flags & MLX4_FLAG_MSI_X) &&
3674 !mlx4_is_mfunc(dev)) {
3675 dev->flags &= ~MLX4_FLAG_MSI_X;
3676 dev->caps.num_comp_vectors = 1;
3677 pci_disable_msix(pdev);
3678 err = mlx4_setup_hca(dev);
3684 /* When PF resources are ready arm its comm channel to enable
3687 if (mlx4_is_master(dev)) {
3688 err = mlx4_ARM_COMM_CHANNEL(dev);
3690 mlx4_err(dev, " Failed to arm comm channel eq: %x\n",
3696 for (port = 1; port <= dev->caps.num_ports; port++) {
3697 err = mlx4_init_port_info(dev, port);
3702 priv->v2p.port1 = 1;
3703 priv->v2p.port2 = 2;
3705 err = mlx4_register_device(dev);
3709 mlx4_sense_init(dev);
3710 mlx4_start_sense(dev);
3714 if (mlx4_is_master(dev) && dev->persist->num_vfs && !reset_flow)
3715 atomic_dec(&pf_loading);
3721 for (--port; port >= 1; --port)
3722 mlx4_cleanup_port_info(&priv->port[port]);
3724 mlx4_cleanup_default_counters(dev);
3725 if (!mlx4_is_slave(dev))
3726 mlx4_cleanup_counters_table(dev);
3727 mlx4_cleanup_qp_table(dev);
3728 mlx4_cleanup_srq_table(dev);
3729 mlx4_cleanup_cq_table(dev);
3730 mlx4_cmd_use_polling(dev);
3731 mlx4_cleanup_eq_table(dev);
3732 mlx4_cleanup_mcg_table(dev);
3733 mlx4_cleanup_mr_table(dev);
3734 mlx4_cleanup_xrcd_table(dev);
3735 mlx4_cleanup_pd_table(dev);
3736 mlx4_cleanup_uar_table(dev);
3739 if (!mlx4_is_slave(dev))
3740 mlx4_clear_steering(dev);
3743 if (dev->flags & MLX4_FLAG_MSI_X)
3744 pci_disable_msix(pdev);
3747 mlx4_free_eq_table(dev);
3750 if (mlx4_is_master(dev)) {
3751 mlx4_free_resource_tracker(dev, RES_TR_FREE_STRUCTS_ONLY);
3752 mlx4_multi_func_cleanup(dev);
3755 if (mlx4_is_slave(dev))
3756 mlx4_slave_destroy_special_qp_cap(dev);
3759 mlx4_close_hca(dev);
3765 if (mlx4_is_slave(dev))
3766 mlx4_multi_func_cleanup(dev);
3769 mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3772 if (dev->flags & MLX4_FLAG_SRIOV && !existing_vfs) {
3773 pci_disable_sriov(pdev);
3774 dev->flags &= ~MLX4_FLAG_SRIOV;
3777 if (mlx4_is_master(dev) && dev->persist->num_vfs && !reset_flow)
3778 atomic_dec(&pf_loading);
3780 kfree(priv->dev.dev_vfs);
3782 if (!mlx4_is_slave(dev))
3783 mlx4_free_ownership(dev);
3788 mlx4_adev_cleanup(dev);
3792 static int __mlx4_init_one(struct pci_dev *pdev, int pci_dev_data,
3793 struct mlx4_priv *priv)
3796 int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
3797 int prb_vf[MLX4_MAX_PORTS + 1] = {0, 0, 0};
3798 const int param_map[MLX4_MAX_PORTS + 1][MLX4_MAX_PORTS + 1] = {
3799 {2, 0, 0}, {0, 1, 2}, {0, 1, 2} };
3800 unsigned total_vfs = 0;
3803 pr_info(DRV_NAME ": Initializing %s\n", pci_name(pdev));
3805 err = mlx4_pci_enable_device(&priv->dev);
3807 dev_err(&pdev->dev, "Cannot enable PCI device, aborting\n");
3811 /* Due to requirement that all VFs and the PF are *guaranteed* 2 MACS
3812 * per port, we must limit the number of VFs to 63 (since their are
3815 for (i = 0; i < ARRAY_SIZE(nvfs) && i < num_vfs_argc;
3816 total_vfs += nvfs[param_map[num_vfs_argc - 1][i]], i++) {
3817 nvfs[param_map[num_vfs_argc - 1][i]] = num_vfs[i];
3819 dev_err(&pdev->dev, "num_vfs module parameter cannot be negative\n");
3821 goto err_disable_pdev;
3824 for (i = 0; i < ARRAY_SIZE(prb_vf) && i < probe_vfs_argc;
3826 prb_vf[param_map[probe_vfs_argc - 1][i]] = probe_vf[i];
3827 if (prb_vf[i] < 0 || prb_vf[i] > nvfs[i]) {
3828 dev_err(&pdev->dev, "probe_vf module parameter cannot be negative or greater than num_vfs\n");
3830 goto err_disable_pdev;
3833 if (total_vfs > MLX4_MAX_NUM_VF) {
3835 "Requested more VF's (%d) than allowed by hw (%d)\n",
3836 total_vfs, MLX4_MAX_NUM_VF);
3838 goto err_disable_pdev;
3841 for (i = 0; i < MLX4_MAX_PORTS; i++) {
3842 if (nvfs[i] + nvfs[2] > MLX4_MAX_NUM_VF_P_PORT) {
3844 "Requested more VF's (%d) for port (%d) than allowed by driver (%d)\n",
3845 nvfs[i] + nvfs[2], i + 1,
3846 MLX4_MAX_NUM_VF_P_PORT);
3848 goto err_disable_pdev;
3852 /* Check for BARs. */
3853 if (!(pci_dev_data & MLX4_PCI_DEV_IS_VF) &&
3854 !(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) {
3855 dev_err(&pdev->dev, "Missing DCS, aborting (driver_data: 0x%x, pci_resource_flags(pdev, 0):0x%lx)\n",
3856 pci_dev_data, pci_resource_flags(pdev, 0));
3858 goto err_disable_pdev;
3860 if (!(pci_resource_flags(pdev, 2) & IORESOURCE_MEM)) {
3861 dev_err(&pdev->dev, "Missing UAR, aborting\n");
3863 goto err_disable_pdev;
3866 err = pci_request_regions(pdev, DRV_NAME);
3868 dev_err(&pdev->dev, "Couldn't get PCI resources, aborting\n");
3869 goto err_disable_pdev;
3872 pci_set_master(pdev);
3874 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
3876 dev_warn(&pdev->dev, "Warning: couldn't set 64-bit PCI DMA mask\n");
3877 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
3879 dev_err(&pdev->dev, "Can't set PCI DMA mask, aborting\n");
3880 goto err_release_regions;
3884 /* Allow large DMA segments, up to the firmware limit of 1 GB */
3885 dma_set_max_seg_size(&pdev->dev, 1024 * 1024 * 1024);
3886 /* Detect if this device is a virtual function */
3887 if (pci_dev_data & MLX4_PCI_DEV_IS_VF) {
3888 /* When acting as pf, we normally skip vfs unless explicitly
3889 * requested to probe them.
3892 unsigned vfs_offset = 0;
3894 for (i = 0; i < ARRAY_SIZE(nvfs) &&
3895 vfs_offset + nvfs[i] < extended_func_num(pdev);
3896 vfs_offset += nvfs[i], i++)
3898 if (i == ARRAY_SIZE(nvfs)) {
3900 goto err_release_regions;
3902 if ((extended_func_num(pdev) - vfs_offset)
3904 dev_warn(&pdev->dev, "Skipping virtual function:%d\n",
3905 extended_func_num(pdev));
3907 goto err_release_regions;
3912 err = mlx4_crdump_init(&priv->dev);
3914 goto err_release_regions;
3916 err = mlx4_catas_init(&priv->dev);
3920 err = mlx4_load_one(pdev, pci_dev_data, total_vfs, nvfs, priv, 0);
3927 mlx4_catas_end(&priv->dev);
3930 mlx4_crdump_end(&priv->dev);
3932 err_release_regions:
3933 pci_release_regions(pdev);
3936 mlx4_pci_disable_device(&priv->dev);
3940 static void mlx4_devlink_param_load_driverinit_values(struct devlink *devlink)
3942 struct mlx4_priv *priv = devlink_priv(devlink);
3943 struct mlx4_dev *dev = &priv->dev;
3944 struct mlx4_fw_crdump *crdump = &dev->persist->crdump;
3945 union devlink_param_value saved_value;
3948 err = devl_param_driverinit_value_get(devlink,
3949 DEVLINK_PARAM_GENERIC_ID_INT_ERR_RESET,
3951 if (!err && mlx4_internal_err_reset != saved_value.vbool) {
3952 mlx4_internal_err_reset = saved_value.vbool;
3953 /* Notify on value changed on runtime configuration mode */
3954 devl_param_value_changed(devlink,
3955 DEVLINK_PARAM_GENERIC_ID_INT_ERR_RESET);
3957 err = devl_param_driverinit_value_get(devlink,
3958 DEVLINK_PARAM_GENERIC_ID_MAX_MACS,
3961 log_num_mac = order_base_2(saved_value.vu32);
3962 err = devl_param_driverinit_value_get(devlink,
3963 MLX4_DEVLINK_PARAM_ID_ENABLE_64B_CQE_EQE,
3966 enable_64b_cqe_eqe = saved_value.vbool;
3967 err = devl_param_driverinit_value_get(devlink,
3968 MLX4_DEVLINK_PARAM_ID_ENABLE_4K_UAR,
3971 enable_4k_uar = saved_value.vbool;
3972 err = devl_param_driverinit_value_get(devlink,
3973 DEVLINK_PARAM_GENERIC_ID_REGION_SNAPSHOT,
3975 if (!err && crdump->snapshot_enable != saved_value.vbool) {
3976 crdump->snapshot_enable = saved_value.vbool;
3977 devl_param_value_changed(devlink,
3978 DEVLINK_PARAM_GENERIC_ID_REGION_SNAPSHOT);
3982 static void mlx4_restart_one_down(struct pci_dev *pdev);
3983 static int mlx4_restart_one_up(struct pci_dev *pdev, bool reload,
3984 struct devlink *devlink);
3986 static int mlx4_devlink_reload_down(struct devlink *devlink, bool netns_change,
3987 enum devlink_reload_action action,
3988 enum devlink_reload_limit limit,
3989 struct netlink_ext_ack *extack)
3991 struct mlx4_priv *priv = devlink_priv(devlink);
3992 struct mlx4_dev *dev = &priv->dev;
3993 struct mlx4_dev_persistent *persist = dev->persist;
3996 NL_SET_ERR_MSG_MOD(extack, "Namespace change is not supported");
3999 if (persist->num_vfs)
4000 mlx4_warn(persist->dev, "Reload performed on PF, will cause reset on operating Virtual Functions\n");
4001 mlx4_restart_one_down(persist->pdev);
4005 static int mlx4_devlink_reload_up(struct devlink *devlink, enum devlink_reload_action action,
4006 enum devlink_reload_limit limit, u32 *actions_performed,
4007 struct netlink_ext_ack *extack)
4009 struct mlx4_priv *priv = devlink_priv(devlink);
4010 struct mlx4_dev *dev = &priv->dev;
4011 struct mlx4_dev_persistent *persist = dev->persist;
4014 *actions_performed = BIT(DEVLINK_RELOAD_ACTION_DRIVER_REINIT);
4015 err = mlx4_restart_one_up(persist->pdev, true, devlink);
4017 mlx4_err(persist->dev, "mlx4_restart_one_up failed, ret=%d\n",
4023 static const struct devlink_ops mlx4_devlink_ops = {
4024 .reload_actions = BIT(DEVLINK_RELOAD_ACTION_DRIVER_REINIT),
4025 .reload_down = mlx4_devlink_reload_down,
4026 .reload_up = mlx4_devlink_reload_up,
4029 static int mlx4_init_one(struct pci_dev *pdev, const struct pci_device_id *id)
4031 struct devlink *devlink;
4032 struct mlx4_priv *priv;
4033 struct mlx4_dev *dev;
4036 printk_once(KERN_INFO "%s", mlx4_version);
4038 devlink = devlink_alloc(&mlx4_devlink_ops, sizeof(*priv), &pdev->dev);
4042 priv = devlink_priv(devlink);
4045 dev->persist = kzalloc(sizeof(*dev->persist), GFP_KERNEL);
4046 if (!dev->persist) {
4048 goto err_devlink_free;
4050 dev->persist->pdev = pdev;
4051 dev->persist->dev = dev;
4052 pci_set_drvdata(pdev, dev->persist);
4053 priv->pci_dev_data = id->driver_data;
4054 mutex_init(&dev->persist->device_state_mutex);
4055 mutex_init(&dev->persist->interface_state_mutex);
4056 mutex_init(&dev->persist->pci_status_mutex);
4058 ret = devl_params_register(devlink, mlx4_devlink_params,
4059 ARRAY_SIZE(mlx4_devlink_params));
4061 goto err_devlink_unregister;
4062 mlx4_devlink_set_params_init_values(devlink);
4063 ret = __mlx4_init_one(pdev, id->driver_data, priv);
4065 goto err_params_unregister;
4067 pci_save_state(pdev);
4068 devl_unlock(devlink);
4069 devlink_register(devlink);
4072 err_params_unregister:
4073 devl_params_unregister(devlink, mlx4_devlink_params,
4074 ARRAY_SIZE(mlx4_devlink_params));
4075 err_devlink_unregister:
4076 kfree(dev->persist);
4078 devl_unlock(devlink);
4079 devlink_free(devlink);
4083 static void mlx4_clean_dev(struct mlx4_dev *dev)
4085 struct mlx4_dev_persistent *persist = dev->persist;
4086 struct mlx4_priv *priv = mlx4_priv(dev);
4087 unsigned long flags = (dev->flags & RESET_PERSIST_MASK_FLAGS);
4089 memset(priv, 0, sizeof(*priv));
4090 priv->dev.persist = persist;
4091 priv->dev.flags = flags;
4094 static void mlx4_unload_one(struct pci_dev *pdev)
4096 struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4097 struct mlx4_dev *dev = persist->dev;
4098 struct mlx4_priv *priv = mlx4_priv(dev);
4100 struct devlink *devlink;
4103 devlink = priv_to_devlink(priv);
4104 devl_assert_locked(devlink);
4108 /* saving current ports type for further use */
4109 for (i = 0; i < dev->caps.num_ports; i++) {
4110 dev->persist->curr_port_type[i] = dev->caps.port_type[i + 1];
4111 dev->persist->curr_port_poss_type[i] = dev->caps.
4112 possible_type[i + 1];
4115 pci_dev_data = priv->pci_dev_data;
4117 mlx4_stop_sense(dev);
4118 mlx4_unregister_device(dev);
4120 for (p = 1; p <= dev->caps.num_ports; p++) {
4121 mlx4_cleanup_port_info(&priv->port[p]);
4122 mlx4_CLOSE_PORT(dev, p);
4125 if (mlx4_is_master(dev))
4126 mlx4_free_resource_tracker(dev,
4127 RES_TR_FREE_SLAVES_ONLY);
4129 mlx4_cleanup_default_counters(dev);
4130 if (!mlx4_is_slave(dev))
4131 mlx4_cleanup_counters_table(dev);
4132 mlx4_cleanup_qp_table(dev);
4133 mlx4_cleanup_srq_table(dev);
4134 mlx4_cleanup_cq_table(dev);
4135 mlx4_cmd_use_polling(dev);
4136 mlx4_cleanup_eq_table(dev);
4137 mlx4_cleanup_mcg_table(dev);
4138 mlx4_cleanup_mr_table(dev);
4139 mlx4_cleanup_xrcd_table(dev);
4140 mlx4_cleanup_pd_table(dev);
4142 if (mlx4_is_master(dev))
4143 mlx4_free_resource_tracker(dev,
4144 RES_TR_FREE_STRUCTS_ONLY);
4147 mlx4_uar_free(dev, &priv->driver_uar);
4148 mlx4_cleanup_uar_table(dev);
4149 if (!mlx4_is_slave(dev))
4150 mlx4_clear_steering(dev);
4151 mlx4_free_eq_table(dev);
4152 if (mlx4_is_master(dev))
4153 mlx4_multi_func_cleanup(dev);
4154 mlx4_close_hca(dev);
4156 if (mlx4_is_slave(dev))
4157 mlx4_multi_func_cleanup(dev);
4158 mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
4160 if (dev->flags & MLX4_FLAG_MSI_X)
4161 pci_disable_msix(pdev);
4163 if (!mlx4_is_slave(dev))
4164 mlx4_free_ownership(dev);
4166 mlx4_slave_destroy_special_qp_cap(dev);
4167 kfree(dev->dev_vfs);
4169 mlx4_adev_cleanup(dev);
4171 mlx4_clean_dev(dev);
4172 priv->pci_dev_data = pci_dev_data;
4176 static void mlx4_remove_one(struct pci_dev *pdev)
4178 struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4179 struct mlx4_dev *dev = persist->dev;
4180 struct mlx4_priv *priv = mlx4_priv(dev);
4181 struct devlink *devlink = priv_to_devlink(priv);
4184 devlink_unregister(devlink);
4187 if (mlx4_is_slave(dev))
4188 persist->interface_state |= MLX4_INTERFACE_STATE_NOWAIT;
4190 mutex_lock(&persist->interface_state_mutex);
4191 persist->interface_state |= MLX4_INTERFACE_STATE_DELETION;
4192 mutex_unlock(&persist->interface_state_mutex);
4194 /* Disabling SR-IOV is not allowed while there are active vf's */
4195 if (mlx4_is_master(dev) && dev->flags & MLX4_FLAG_SRIOV) {
4196 active_vfs = mlx4_how_many_lives_vf(dev);
4198 pr_warn("Removing PF when there are active VF's !!\n");
4199 pr_warn("Will not disable SR-IOV.\n");
4203 /* device marked to be under deletion running now without the lock
4204 * letting other tasks to be terminated
4206 if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
4207 mlx4_unload_one(pdev);
4209 mlx4_info(dev, "%s: interface is down\n", __func__);
4210 mlx4_catas_end(dev);
4211 mlx4_crdump_end(dev);
4212 if (dev->flags & MLX4_FLAG_SRIOV && !active_vfs) {
4213 mlx4_warn(dev, "Disabling SR-IOV\n");
4214 pci_disable_sriov(pdev);
4217 pci_release_regions(pdev);
4218 mlx4_pci_disable_device(dev);
4219 devl_params_unregister(devlink, mlx4_devlink_params,
4220 ARRAY_SIZE(mlx4_devlink_params));
4221 kfree(dev->persist);
4222 devl_unlock(devlink);
4223 devlink_free(devlink);
4226 static int restore_current_port_types(struct mlx4_dev *dev,
4227 enum mlx4_port_type *types,
4228 enum mlx4_port_type *poss_types)
4230 struct mlx4_priv *priv = mlx4_priv(dev);
4233 mlx4_stop_sense(dev);
4235 mutex_lock(&priv->port_mutex);
4236 for (i = 0; i < dev->caps.num_ports; i++)
4237 dev->caps.possible_type[i + 1] = poss_types[i];
4238 err = mlx4_change_port_types(dev, types);
4239 mlx4_start_sense(dev);
4240 mutex_unlock(&priv->port_mutex);
4245 static void mlx4_restart_one_down(struct pci_dev *pdev)
4247 mlx4_unload_one(pdev);
4250 static int mlx4_restart_one_up(struct pci_dev *pdev, bool reload,
4251 struct devlink *devlink)
4253 struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4254 struct mlx4_dev *dev = persist->dev;
4255 struct mlx4_priv *priv = mlx4_priv(dev);
4256 int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
4257 int pci_dev_data, err, total_vfs;
4259 pci_dev_data = priv->pci_dev_data;
4260 total_vfs = dev->persist->num_vfs;
4261 memcpy(nvfs, dev->persist->nvfs, sizeof(dev->persist->nvfs));
4264 mlx4_devlink_param_load_driverinit_values(devlink);
4265 err = mlx4_load_one(pdev, pci_dev_data, total_vfs, nvfs, priv, 1);
4267 mlx4_err(dev, "%s: ERROR: mlx4_load_one failed, pci_name=%s, err=%d\n",
4268 __func__, pci_name(pdev), err);
4272 err = restore_current_port_types(dev, dev->persist->curr_port_type,
4273 dev->persist->curr_port_poss_type);
4275 mlx4_err(dev, "could not restore original port types (%d)\n",
4281 int mlx4_restart_one(struct pci_dev *pdev)
4283 mlx4_restart_one_down(pdev);
4284 return mlx4_restart_one_up(pdev, false, NULL);
4287 #define MLX_SP(id) { PCI_VDEVICE(MELLANOX, id), MLX4_PCI_DEV_FORCE_SENSE_PORT }
4288 #define MLX_VF(id) { PCI_VDEVICE(MELLANOX, id), MLX4_PCI_DEV_IS_VF }
4289 #define MLX_GN(id) { PCI_VDEVICE(MELLANOX, id), 0 }
4291 static const struct pci_device_id mlx4_pci_table[] = {
4292 #ifdef CONFIG_MLX4_CORE_GEN2
4293 /* MT25408 "Hermon" */
4294 MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_SDR), /* SDR */
4295 MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_DDR), /* DDR */
4296 MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_QDR), /* QDR */
4297 MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_DDR_GEN2), /* DDR Gen2 */
4298 MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_QDR_GEN2), /* QDR Gen2 */
4299 MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_EN), /* EN 10GigE */
4300 MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_EN_GEN2), /* EN 10GigE Gen2 */
4301 /* MT25458 ConnectX EN 10GBASE-T */
4302 MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX_EN),
4303 MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX_EN_T_GEN2), /* Gen2 */
4304 /* MT26468 ConnectX EN 10GigE PCIe Gen2*/
4305 MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX_EN_GEN2),
4306 /* MT26438 ConnectX EN 40GigE PCIe Gen2 5GT/s */
4307 MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX_EN_5_GEN2),
4308 /* MT26478 ConnectX2 40GigE PCIe Gen2 */
4309 MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX2),
4310 /* MT25400 Family [ConnectX-2] */
4311 MLX_VF(0x1002), /* Virtual Function */
4312 #endif /* CONFIG_MLX4_CORE_GEN2 */
4313 /* MT27500 Family [ConnectX-3] */
4314 MLX_GN(PCI_DEVICE_ID_MELLANOX_CONNECTX3),
4315 MLX_VF(0x1004), /* Virtual Function */
4316 MLX_GN(0x1005), /* MT27510 Family */
4317 MLX_GN(0x1006), /* MT27511 Family */
4318 MLX_GN(PCI_DEVICE_ID_MELLANOX_CONNECTX3_PRO), /* MT27520 Family */
4319 MLX_GN(0x1008), /* MT27521 Family */
4320 MLX_GN(0x1009), /* MT27530 Family */
4321 MLX_GN(0x100a), /* MT27531 Family */
4322 MLX_GN(0x100b), /* MT27540 Family */
4323 MLX_GN(0x100c), /* MT27541 Family */
4324 MLX_GN(0x100d), /* MT27550 Family */
4325 MLX_GN(0x100e), /* MT27551 Family */
4326 MLX_GN(0x100f), /* MT27560 Family */
4327 MLX_GN(0x1010), /* MT27561 Family */
4330 * See the mellanox_check_broken_intx_masking() quirk when
4337 MODULE_DEVICE_TABLE(pci, mlx4_pci_table);
4339 static pci_ers_result_t mlx4_pci_err_detected(struct pci_dev *pdev,
4340 pci_channel_state_t state)
4342 struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4343 struct mlx4_dev *dev = persist->dev;
4344 struct devlink *devlink;
4346 mlx4_err(persist->dev, "mlx4_pci_err_detected was called\n");
4347 mlx4_enter_error_state(persist);
4349 devlink = priv_to_devlink(mlx4_priv(dev));
4351 mutex_lock(&persist->interface_state_mutex);
4352 if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
4353 mlx4_unload_one(pdev);
4355 mutex_unlock(&persist->interface_state_mutex);
4356 devl_unlock(devlink);
4357 if (state == pci_channel_io_perm_failure)
4358 return PCI_ERS_RESULT_DISCONNECT;
4360 mlx4_pci_disable_device(persist->dev);
4361 return PCI_ERS_RESULT_NEED_RESET;
4364 static pci_ers_result_t mlx4_pci_slot_reset(struct pci_dev *pdev)
4366 struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4367 struct mlx4_dev *dev = persist->dev;
4370 mlx4_err(dev, "mlx4_pci_slot_reset was called\n");
4371 err = mlx4_pci_enable_device(dev);
4373 mlx4_err(dev, "Can not re-enable device, err=%d\n", err);
4374 return PCI_ERS_RESULT_DISCONNECT;
4377 pci_set_master(pdev);
4378 pci_restore_state(pdev);
4379 pci_save_state(pdev);
4380 return PCI_ERS_RESULT_RECOVERED;
4383 static void mlx4_pci_resume(struct pci_dev *pdev)
4385 struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4386 struct mlx4_dev *dev = persist->dev;
4387 struct mlx4_priv *priv = mlx4_priv(dev);
4388 int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
4389 struct devlink *devlink;
4393 mlx4_err(dev, "%s was called\n", __func__);
4394 total_vfs = dev->persist->num_vfs;
4395 memcpy(nvfs, dev->persist->nvfs, sizeof(dev->persist->nvfs));
4397 devlink = priv_to_devlink(priv);
4399 mutex_lock(&persist->interface_state_mutex);
4400 if (!(persist->interface_state & MLX4_INTERFACE_STATE_UP)) {
4401 err = mlx4_load_one(pdev, priv->pci_dev_data, total_vfs, nvfs,
4404 mlx4_err(dev, "%s: mlx4_load_one failed, err=%d\n",
4409 err = restore_current_port_types(dev, dev->persist->
4410 curr_port_type, dev->persist->
4411 curr_port_poss_type);
4413 mlx4_err(dev, "could not restore original port types (%d)\n", err);
4416 mutex_unlock(&persist->interface_state_mutex);
4417 devl_unlock(devlink);
4420 static void mlx4_shutdown(struct pci_dev *pdev)
4422 struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4423 struct mlx4_dev *dev = persist->dev;
4424 struct devlink *devlink;
4426 mlx4_info(persist->dev, "mlx4_shutdown was called\n");
4427 devlink = priv_to_devlink(mlx4_priv(dev));
4429 mutex_lock(&persist->interface_state_mutex);
4430 if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
4431 mlx4_unload_one(pdev);
4432 mutex_unlock(&persist->interface_state_mutex);
4433 devl_unlock(devlink);
4434 mlx4_pci_disable_device(dev);
4437 static const struct pci_error_handlers mlx4_err_handler = {
4438 .error_detected = mlx4_pci_err_detected,
4439 .slot_reset = mlx4_pci_slot_reset,
4440 .resume = mlx4_pci_resume,
4443 static int __maybe_unused mlx4_suspend(struct device *dev_d)
4445 struct pci_dev *pdev = to_pci_dev(dev_d);
4446 struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4447 struct mlx4_dev *dev = persist->dev;
4448 struct devlink *devlink;
4450 mlx4_err(dev, "suspend was called\n");
4451 devlink = priv_to_devlink(mlx4_priv(dev));
4453 mutex_lock(&persist->interface_state_mutex);
4454 if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
4455 mlx4_unload_one(pdev);
4456 mutex_unlock(&persist->interface_state_mutex);
4457 devl_unlock(devlink);
4462 static int __maybe_unused mlx4_resume(struct device *dev_d)
4464 struct pci_dev *pdev = to_pci_dev(dev_d);
4465 struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4466 struct mlx4_dev *dev = persist->dev;
4467 struct mlx4_priv *priv = mlx4_priv(dev);
4468 int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
4469 struct devlink *devlink;
4473 mlx4_err(dev, "resume was called\n");
4474 total_vfs = dev->persist->num_vfs;
4475 memcpy(nvfs, dev->persist->nvfs, sizeof(dev->persist->nvfs));
4477 devlink = priv_to_devlink(priv);
4479 mutex_lock(&persist->interface_state_mutex);
4480 if (!(persist->interface_state & MLX4_INTERFACE_STATE_UP)) {
4481 ret = mlx4_load_one(pdev, priv->pci_dev_data, total_vfs,
4484 ret = restore_current_port_types(dev,
4485 dev->persist->curr_port_type,
4486 dev->persist->curr_port_poss_type);
4488 mlx4_err(dev, "resume: could not restore original port types (%d)\n", ret);
4491 mutex_unlock(&persist->interface_state_mutex);
4492 devl_unlock(devlink);
4497 static SIMPLE_DEV_PM_OPS(mlx4_pm_ops, mlx4_suspend, mlx4_resume);
4499 static struct pci_driver mlx4_driver = {
4501 .id_table = mlx4_pci_table,
4502 .probe = mlx4_init_one,
4503 .shutdown = mlx4_shutdown,
4504 .remove = mlx4_remove_one,
4505 .driver.pm = &mlx4_pm_ops,
4506 .err_handler = &mlx4_err_handler,
4509 static int __init mlx4_verify_params(void)
4512 pr_warn("mlx4_core: bad msi_x: %d\n", msi_x);
4516 if ((log_num_mac < 0) || (log_num_mac > 7)) {
4517 pr_warn("mlx4_core: bad num_mac: %d\n", log_num_mac);
4521 if (log_num_vlan != 0)
4522 pr_warn("mlx4_core: log_num_vlan - obsolete module param, using %d\n",
4523 MLX4_LOG_NUM_VLANS);
4526 pr_warn("mlx4_core: use_prio - obsolete module param, ignored\n");
4528 if ((log_mtts_per_seg < 0) || (log_mtts_per_seg > 7)) {
4529 pr_warn("mlx4_core: bad log_mtts_per_seg: %d\n",
4534 /* Check if module param for ports type has legal combination */
4535 if (port_type_array[0] == false && port_type_array[1] == true) {
4536 pr_warn("Module parameter configuration ETH/IB is not supported. Switching to default configuration IB/IB\n");
4537 port_type_array[0] = true;
4540 if (mlx4_log_num_mgm_entry_size < -7 ||
4541 (mlx4_log_num_mgm_entry_size > 0 &&
4542 (mlx4_log_num_mgm_entry_size < MLX4_MIN_MGM_LOG_ENTRY_SIZE ||
4543 mlx4_log_num_mgm_entry_size > MLX4_MAX_MGM_LOG_ENTRY_SIZE))) {
4544 pr_warn("mlx4_core: mlx4_log_num_mgm_entry_size (%d) not in legal range (-7..0 or %d..%d)\n",
4545 mlx4_log_num_mgm_entry_size,
4546 MLX4_MIN_MGM_LOG_ENTRY_SIZE,
4547 MLX4_MAX_MGM_LOG_ENTRY_SIZE);
4554 static int __init mlx4_init(void)
4558 WARN_ONCE(strcmp(MLX4_ADEV_NAME, KBUILD_MODNAME),
4559 "mlx4_core name not in sync with kernel module name");
4561 if (mlx4_verify_params())
4565 mlx4_wq = create_singlethread_workqueue("mlx4");
4569 ret = pci_register_driver(&mlx4_driver);
4571 destroy_workqueue(mlx4_wq);
4572 return ret < 0 ? ret : 0;
4575 static void __exit mlx4_cleanup(void)
4577 pci_unregister_driver(&mlx4_driver);
4578 destroy_workqueue(mlx4_wq);
4581 module_init(mlx4_init);
4582 module_exit(mlx4_cleanup);