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/init.h>
38 #include <linux/errno.h>
39 #include <linux/pci.h>
40 #include <linux/dma-mapping.h>
41 #include <linux/slab.h>
42 #include <linux/io-mapping.h>
43 #include <linux/delay.h>
44 #include <linux/kmod.h>
46 #include <linux/mlx4/device.h>
47 #include <linux/mlx4/doorbell.h>
53 MODULE_AUTHOR("Roland Dreier");
54 MODULE_DESCRIPTION("Mellanox ConnectX HCA low-level driver");
55 MODULE_LICENSE("Dual BSD/GPL");
56 MODULE_VERSION(DRV_VERSION);
58 struct workqueue_struct *mlx4_wq;
60 #ifdef CONFIG_MLX4_DEBUG
62 int mlx4_debug_level = 0;
63 module_param_named(debug_level, mlx4_debug_level, int, 0644);
64 MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0");
66 #endif /* CONFIG_MLX4_DEBUG */
71 module_param(msi_x, int, 0444);
72 MODULE_PARM_DESC(msi_x, "attempt to use MSI-X if nonzero");
74 #else /* CONFIG_PCI_MSI */
78 #endif /* CONFIG_PCI_MSI */
80 static uint8_t num_vfs[3] = {0, 0, 0};
81 static int num_vfs_argc;
82 module_param_array(num_vfs, byte , &num_vfs_argc, 0444);
83 MODULE_PARM_DESC(num_vfs, "enable #num_vfs functions if num_vfs > 0\n"
84 "num_vfs=port1,port2,port1+2");
86 static uint8_t probe_vf[3] = {0, 0, 0};
87 static int probe_vfs_argc;
88 module_param_array(probe_vf, byte, &probe_vfs_argc, 0444);
89 MODULE_PARM_DESC(probe_vf, "number of vfs to probe by pf driver (num_vfs > 0)\n"
90 "probe_vf=port1,port2,port1+2");
92 int mlx4_log_num_mgm_entry_size = MLX4_DEFAULT_MGM_LOG_ENTRY_SIZE;
93 module_param_named(log_num_mgm_entry_size,
94 mlx4_log_num_mgm_entry_size, int, 0444);
95 MODULE_PARM_DESC(log_num_mgm_entry_size, "log mgm size, that defines the num"
96 " of qp per mcg, for example:"
97 " 10 gives 248.range: 7 <="
98 " log_num_mgm_entry_size <= 12."
99 " To activate device managed"
100 " flow steering when available, set to -1");
102 static bool enable_64b_cqe_eqe = true;
103 module_param(enable_64b_cqe_eqe, bool, 0444);
104 MODULE_PARM_DESC(enable_64b_cqe_eqe,
105 "Enable 64 byte CQEs/EQEs when the FW supports this (default: True)");
107 #define PF_CONTEXT_BEHAVIOUR_MASK (MLX4_FUNC_CAP_64B_EQE_CQE | \
108 MLX4_FUNC_CAP_EQE_CQE_STRIDE | \
109 MLX4_FUNC_CAP_DMFS_A0_STATIC)
111 #define RESET_PERSIST_MASK_FLAGS (MLX4_FLAG_SRIOV)
113 static char mlx4_version[] =
114 DRV_NAME ": Mellanox ConnectX core driver v"
115 DRV_VERSION " (" DRV_RELDATE ")\n";
117 static struct mlx4_profile default_profile = {
120 .rdmarc_per_qp = 1 << 4,
124 .num_mtt = 1 << 20, /* It is really num mtt segements */
127 static struct mlx4_profile low_mem_profile = {
130 .rdmarc_per_qp = 1 << 4,
137 static int log_num_mac = 7;
138 module_param_named(log_num_mac, log_num_mac, int, 0444);
139 MODULE_PARM_DESC(log_num_mac, "Log2 max number of MACs per ETH port (1-7)");
141 static int log_num_vlan;
142 module_param_named(log_num_vlan, log_num_vlan, int, 0444);
143 MODULE_PARM_DESC(log_num_vlan, "Log2 max number of VLANs per ETH port (0-7)");
144 /* Log2 max number of VLANs per ETH port (0-7) */
145 #define MLX4_LOG_NUM_VLANS 7
146 #define MLX4_MIN_LOG_NUM_VLANS 0
147 #define MLX4_MIN_LOG_NUM_MAC 1
149 static bool use_prio;
150 module_param_named(use_prio, use_prio, bool, 0444);
151 MODULE_PARM_DESC(use_prio, "Enable steering by VLAN priority on ETH ports (deprecated)");
153 int log_mtts_per_seg = ilog2(MLX4_MTT_ENTRY_PER_SEG);
154 module_param_named(log_mtts_per_seg, log_mtts_per_seg, int, 0444);
155 MODULE_PARM_DESC(log_mtts_per_seg, "Log2 number of MTT entries per segment (1-7)");
157 static int port_type_array[2] = {MLX4_PORT_TYPE_NONE, MLX4_PORT_TYPE_NONE};
158 static int arr_argc = 2;
159 module_param_array(port_type_array, int, &arr_argc, 0444);
160 MODULE_PARM_DESC(port_type_array, "Array of port types: HW_DEFAULT (0) is default "
161 "1 for IB, 2 for Ethernet");
163 struct mlx4_port_config {
164 struct list_head list;
165 enum mlx4_port_type port_type[MLX4_MAX_PORTS + 1];
166 struct pci_dev *pdev;
169 static atomic_t pf_loading = ATOMIC_INIT(0);
171 int mlx4_check_port_params(struct mlx4_dev *dev,
172 enum mlx4_port_type *port_type)
176 if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP)) {
177 for (i = 0; i < dev->caps.num_ports - 1; i++) {
178 if (port_type[i] != port_type[i + 1]) {
179 mlx4_err(dev, "Only same port types supported on this HCA, aborting\n");
185 for (i = 0; i < dev->caps.num_ports; i++) {
186 if (!(port_type[i] & dev->caps.supported_type[i+1])) {
187 mlx4_err(dev, "Requested port type for port %d is not supported on this HCA\n",
195 static void mlx4_set_port_mask(struct mlx4_dev *dev)
199 for (i = 1; i <= dev->caps.num_ports; ++i)
200 dev->caps.port_mask[i] = dev->caps.port_type[i];
204 MLX4_QUERY_FUNC_NUM_SYS_EQS = 1 << 0,
207 static int mlx4_query_func(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
210 struct mlx4_func func;
212 if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS) {
213 err = mlx4_QUERY_FUNC(dev, &func, 0);
215 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
218 dev_cap->max_eqs = func.max_eq;
219 dev_cap->reserved_eqs = func.rsvd_eqs;
220 dev_cap->reserved_uars = func.rsvd_uars;
221 err |= MLX4_QUERY_FUNC_NUM_SYS_EQS;
226 static void mlx4_enable_cqe_eqe_stride(struct mlx4_dev *dev)
228 struct mlx4_caps *dev_cap = &dev->caps;
230 /* FW not supporting or cancelled by user */
231 if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_EQE_STRIDE) ||
232 !(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_CQE_STRIDE))
235 /* Must have 64B CQE_EQE enabled by FW to use bigger stride
236 * When FW has NCSI it may decide not to report 64B CQE/EQEs
238 if (!(dev_cap->flags & MLX4_DEV_CAP_FLAG_64B_EQE) ||
239 !(dev_cap->flags & MLX4_DEV_CAP_FLAG_64B_CQE)) {
240 dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
241 dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
245 if (cache_line_size() == 128 || cache_line_size() == 256) {
246 mlx4_dbg(dev, "Enabling CQE stride cacheLine supported\n");
247 /* Changing the real data inside CQE size to 32B */
248 dev_cap->flags &= ~MLX4_DEV_CAP_FLAG_64B_CQE;
249 dev_cap->flags &= ~MLX4_DEV_CAP_FLAG_64B_EQE;
251 if (mlx4_is_master(dev))
252 dev_cap->function_caps |= MLX4_FUNC_CAP_EQE_CQE_STRIDE;
254 mlx4_dbg(dev, "Disabling CQE stride cacheLine unsupported\n");
255 dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
256 dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
260 static int _mlx4_dev_port(struct mlx4_dev *dev, int port,
261 struct mlx4_port_cap *port_cap)
263 dev->caps.vl_cap[port] = port_cap->max_vl;
264 dev->caps.ib_mtu_cap[port] = port_cap->ib_mtu;
265 dev->phys_caps.gid_phys_table_len[port] = port_cap->max_gids;
266 dev->phys_caps.pkey_phys_table_len[port] = port_cap->max_pkeys;
267 /* set gid and pkey table operating lengths by default
268 * to non-sriov values
270 dev->caps.gid_table_len[port] = port_cap->max_gids;
271 dev->caps.pkey_table_len[port] = port_cap->max_pkeys;
272 dev->caps.port_width_cap[port] = port_cap->max_port_width;
273 dev->caps.eth_mtu_cap[port] = port_cap->eth_mtu;
274 dev->caps.def_mac[port] = port_cap->def_mac;
275 dev->caps.supported_type[port] = port_cap->supported_port_types;
276 dev->caps.suggested_type[port] = port_cap->suggested_type;
277 dev->caps.default_sense[port] = port_cap->default_sense;
278 dev->caps.trans_type[port] = port_cap->trans_type;
279 dev->caps.vendor_oui[port] = port_cap->vendor_oui;
280 dev->caps.wavelength[port] = port_cap->wavelength;
281 dev->caps.trans_code[port] = port_cap->trans_code;
286 static int mlx4_dev_port(struct mlx4_dev *dev, int port,
287 struct mlx4_port_cap *port_cap)
291 err = mlx4_QUERY_PORT(dev, port, port_cap);
294 mlx4_err(dev, "QUERY_PORT command failed.\n");
299 #define MLX4_A0_STEERING_TABLE_SIZE 256
300 static int mlx4_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
305 err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
307 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
310 mlx4_dev_cap_dump(dev, dev_cap);
312 if (dev_cap->min_page_sz > PAGE_SIZE) {
313 mlx4_err(dev, "HCA minimum page size of %d bigger than kernel PAGE_SIZE of %ld, aborting\n",
314 dev_cap->min_page_sz, PAGE_SIZE);
317 if (dev_cap->num_ports > MLX4_MAX_PORTS) {
318 mlx4_err(dev, "HCA has %d ports, but we only support %d, aborting\n",
319 dev_cap->num_ports, MLX4_MAX_PORTS);
323 if (dev_cap->uar_size > pci_resource_len(dev->persist->pdev, 2)) {
324 mlx4_err(dev, "HCA reported UAR size of 0x%x bigger than PCI resource 2 size of 0x%llx, aborting\n",
327 pci_resource_len(dev->persist->pdev, 2));
331 dev->caps.num_ports = dev_cap->num_ports;
332 dev->caps.num_sys_eqs = dev_cap->num_sys_eqs;
333 dev->phys_caps.num_phys_eqs = dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS ?
334 dev->caps.num_sys_eqs :
336 for (i = 1; i <= dev->caps.num_ports; ++i) {
337 err = _mlx4_dev_port(dev, i, dev_cap->port_cap + i);
339 mlx4_err(dev, "QUERY_PORT command failed, aborting\n");
344 dev->caps.uar_page_size = PAGE_SIZE;
345 dev->caps.num_uars = dev_cap->uar_size / PAGE_SIZE;
346 dev->caps.local_ca_ack_delay = dev_cap->local_ca_ack_delay;
347 dev->caps.bf_reg_size = dev_cap->bf_reg_size;
348 dev->caps.bf_regs_per_page = dev_cap->bf_regs_per_page;
349 dev->caps.max_sq_sg = dev_cap->max_sq_sg;
350 dev->caps.max_rq_sg = dev_cap->max_rq_sg;
351 dev->caps.max_wqes = dev_cap->max_qp_sz;
352 dev->caps.max_qp_init_rdma = dev_cap->max_requester_per_qp;
353 dev->caps.max_srq_wqes = dev_cap->max_srq_sz;
354 dev->caps.max_srq_sge = dev_cap->max_rq_sg - 1;
355 dev->caps.reserved_srqs = dev_cap->reserved_srqs;
356 dev->caps.max_sq_desc_sz = dev_cap->max_sq_desc_sz;
357 dev->caps.max_rq_desc_sz = dev_cap->max_rq_desc_sz;
359 * Subtract 1 from the limit because we need to allocate a
360 * spare CQE so the HCA HW can tell the difference between an
361 * empty CQ and a full CQ.
363 dev->caps.max_cqes = dev_cap->max_cq_sz - 1;
364 dev->caps.reserved_cqs = dev_cap->reserved_cqs;
365 dev->caps.reserved_eqs = dev_cap->reserved_eqs;
366 dev->caps.reserved_mtts = dev_cap->reserved_mtts;
367 dev->caps.reserved_mrws = dev_cap->reserved_mrws;
369 /* The first 128 UARs are used for EQ doorbells */
370 dev->caps.reserved_uars = max_t(int, 128, dev_cap->reserved_uars);
371 dev->caps.reserved_pds = dev_cap->reserved_pds;
372 dev->caps.reserved_xrcds = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
373 dev_cap->reserved_xrcds : 0;
374 dev->caps.max_xrcds = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
375 dev_cap->max_xrcds : 0;
376 dev->caps.mtt_entry_sz = dev_cap->mtt_entry_sz;
378 dev->caps.max_msg_sz = dev_cap->max_msg_sz;
379 dev->caps.page_size_cap = ~(u32) (dev_cap->min_page_sz - 1);
380 dev->caps.flags = dev_cap->flags;
381 dev->caps.flags2 = dev_cap->flags2;
382 dev->caps.bmme_flags = dev_cap->bmme_flags;
383 dev->caps.reserved_lkey = dev_cap->reserved_lkey;
384 dev->caps.stat_rate_support = dev_cap->stat_rate_support;
385 dev->caps.max_gso_sz = dev_cap->max_gso_sz;
386 dev->caps.max_rss_tbl_sz = dev_cap->max_rss_tbl_sz;
388 /* Sense port always allowed on supported devices for ConnectX-1 and -2 */
389 if (mlx4_priv(dev)->pci_dev_data & MLX4_PCI_DEV_FORCE_SENSE_PORT)
390 dev->caps.flags |= MLX4_DEV_CAP_FLAG_SENSE_SUPPORT;
391 /* Don't do sense port on multifunction devices (for now at least) */
392 if (mlx4_is_mfunc(dev))
393 dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_SENSE_SUPPORT;
395 if (mlx4_low_memory_profile()) {
396 dev->caps.log_num_macs = MLX4_MIN_LOG_NUM_MAC;
397 dev->caps.log_num_vlans = MLX4_MIN_LOG_NUM_VLANS;
399 dev->caps.log_num_macs = log_num_mac;
400 dev->caps.log_num_vlans = MLX4_LOG_NUM_VLANS;
403 for (i = 1; i <= dev->caps.num_ports; ++i) {
404 dev->caps.port_type[i] = MLX4_PORT_TYPE_NONE;
405 if (dev->caps.supported_type[i]) {
406 /* if only ETH is supported - assign ETH */
407 if (dev->caps.supported_type[i] == MLX4_PORT_TYPE_ETH)
408 dev->caps.port_type[i] = MLX4_PORT_TYPE_ETH;
409 /* if only IB is supported, assign IB */
410 else if (dev->caps.supported_type[i] ==
412 dev->caps.port_type[i] = MLX4_PORT_TYPE_IB;
414 /* if IB and ETH are supported, we set the port
415 * type according to user selection of port type;
416 * if user selected none, take the FW hint */
417 if (port_type_array[i - 1] == MLX4_PORT_TYPE_NONE)
418 dev->caps.port_type[i] = dev->caps.suggested_type[i] ?
419 MLX4_PORT_TYPE_ETH : MLX4_PORT_TYPE_IB;
421 dev->caps.port_type[i] = port_type_array[i - 1];
425 * Link sensing is allowed on the port if 3 conditions are true:
426 * 1. Both protocols are supported on the port.
427 * 2. Different types are supported on the port
428 * 3. FW declared that it supports link sensing
430 mlx4_priv(dev)->sense.sense_allowed[i] =
431 ((dev->caps.supported_type[i] == MLX4_PORT_TYPE_AUTO) &&
432 (dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
433 (dev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT));
436 * If "default_sense" bit is set, we move the port to "AUTO" mode
437 * and perform sense_port FW command to try and set the correct
438 * port type from beginning
440 if (mlx4_priv(dev)->sense.sense_allowed[i] && dev->caps.default_sense[i]) {
441 enum mlx4_port_type sensed_port = MLX4_PORT_TYPE_NONE;
442 dev->caps.possible_type[i] = MLX4_PORT_TYPE_AUTO;
443 mlx4_SENSE_PORT(dev, i, &sensed_port);
444 if (sensed_port != MLX4_PORT_TYPE_NONE)
445 dev->caps.port_type[i] = sensed_port;
447 dev->caps.possible_type[i] = dev->caps.port_type[i];
450 if (dev->caps.log_num_macs > dev_cap->port_cap[i].log_max_macs) {
451 dev->caps.log_num_macs = dev_cap->port_cap[i].log_max_macs;
452 mlx4_warn(dev, "Requested number of MACs is too much for port %d, reducing to %d\n",
453 i, 1 << dev->caps.log_num_macs);
455 if (dev->caps.log_num_vlans > dev_cap->port_cap[i].log_max_vlans) {
456 dev->caps.log_num_vlans = dev_cap->port_cap[i].log_max_vlans;
457 mlx4_warn(dev, "Requested number of VLANs is too much for port %d, reducing to %d\n",
458 i, 1 << dev->caps.log_num_vlans);
462 dev->caps.max_counters = 1 << ilog2(dev_cap->max_counters);
464 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] = dev_cap->reserved_qps;
465 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] =
466 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] =
467 (1 << dev->caps.log_num_macs) *
468 (1 << dev->caps.log_num_vlans) *
470 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH] = MLX4_NUM_FEXCH;
472 if (dev_cap->dmfs_high_rate_qpn_base > 0 &&
473 dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN)
474 dev->caps.dmfs_high_rate_qpn_base = dev_cap->dmfs_high_rate_qpn_base;
476 dev->caps.dmfs_high_rate_qpn_base =
477 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
479 if (dev_cap->dmfs_high_rate_qpn_range > 0 &&
480 dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN) {
481 dev->caps.dmfs_high_rate_qpn_range = dev_cap->dmfs_high_rate_qpn_range;
482 dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_DEFAULT;
483 dev->caps.flags2 |= MLX4_DEV_CAP_FLAG2_FS_A0;
485 dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_NOT_SUPPORTED;
486 dev->caps.dmfs_high_rate_qpn_base =
487 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
488 dev->caps.dmfs_high_rate_qpn_range = MLX4_A0_STEERING_TABLE_SIZE;
491 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_RSS_RAW_ETH] =
492 dev->caps.dmfs_high_rate_qpn_range;
494 dev->caps.reserved_qps = dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] +
495 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] +
496 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] +
497 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH];
499 dev->caps.sqp_demux = (mlx4_is_master(dev)) ? MLX4_MAX_NUM_SLAVES : 0;
501 if (!enable_64b_cqe_eqe && !mlx4_is_slave(dev)) {
503 (MLX4_DEV_CAP_FLAG_64B_CQE | MLX4_DEV_CAP_FLAG_64B_EQE)) {
504 mlx4_warn(dev, "64B EQEs/CQEs supported by the device but not enabled\n");
505 dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_64B_CQE;
506 dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_64B_EQE;
509 if (dev_cap->flags2 &
510 (MLX4_DEV_CAP_FLAG2_CQE_STRIDE |
511 MLX4_DEV_CAP_FLAG2_EQE_STRIDE)) {
512 mlx4_warn(dev, "Disabling EQE/CQE stride per user request\n");
513 dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
514 dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
518 if ((dev->caps.flags &
519 (MLX4_DEV_CAP_FLAG_64B_CQE | MLX4_DEV_CAP_FLAG_64B_EQE)) &&
521 dev->caps.function_caps |= MLX4_FUNC_CAP_64B_EQE_CQE;
523 if (!mlx4_is_slave(dev)) {
524 mlx4_enable_cqe_eqe_stride(dev);
525 dev->caps.alloc_res_qp_mask =
526 (dev->caps.bf_reg_size ? MLX4_RESERVE_ETH_BF_QP : 0) |
529 dev->caps.alloc_res_qp_mask = 0;
535 static int mlx4_get_pcie_dev_link_caps(struct mlx4_dev *dev,
536 enum pci_bus_speed *speed,
537 enum pcie_link_width *width)
539 u32 lnkcap1, lnkcap2;
542 #define PCIE_MLW_CAP_SHIFT 4 /* start of MLW mask in link capabilities */
544 *speed = PCI_SPEED_UNKNOWN;
545 *width = PCIE_LNK_WIDTH_UNKNOWN;
547 err1 = pcie_capability_read_dword(dev->persist->pdev, PCI_EXP_LNKCAP,
549 err2 = pcie_capability_read_dword(dev->persist->pdev, PCI_EXP_LNKCAP2,
551 if (!err2 && lnkcap2) { /* PCIe r3.0-compliant */
552 if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_8_0GB)
553 *speed = PCIE_SPEED_8_0GT;
554 else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_5_0GB)
555 *speed = PCIE_SPEED_5_0GT;
556 else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_2_5GB)
557 *speed = PCIE_SPEED_2_5GT;
560 *width = (lnkcap1 & PCI_EXP_LNKCAP_MLW) >> PCIE_MLW_CAP_SHIFT;
561 if (!lnkcap2) { /* pre-r3.0 */
562 if (lnkcap1 & PCI_EXP_LNKCAP_SLS_5_0GB)
563 *speed = PCIE_SPEED_5_0GT;
564 else if (lnkcap1 & PCI_EXP_LNKCAP_SLS_2_5GB)
565 *speed = PCIE_SPEED_2_5GT;
569 if (*speed == PCI_SPEED_UNKNOWN || *width == PCIE_LNK_WIDTH_UNKNOWN) {
571 err2 ? err2 : -EINVAL;
576 static void mlx4_check_pcie_caps(struct mlx4_dev *dev)
578 enum pcie_link_width width, width_cap;
579 enum pci_bus_speed speed, speed_cap;
582 #define PCIE_SPEED_STR(speed) \
583 (speed == PCIE_SPEED_8_0GT ? "8.0GT/s" : \
584 speed == PCIE_SPEED_5_0GT ? "5.0GT/s" : \
585 speed == PCIE_SPEED_2_5GT ? "2.5GT/s" : \
588 err = mlx4_get_pcie_dev_link_caps(dev, &speed_cap, &width_cap);
591 "Unable to determine PCIe device BW capabilities\n");
595 err = pcie_get_minimum_link(dev->persist->pdev, &speed, &width);
596 if (err || speed == PCI_SPEED_UNKNOWN ||
597 width == PCIE_LNK_WIDTH_UNKNOWN) {
599 "Unable to determine PCI device chain minimum BW\n");
603 if (width != width_cap || speed != speed_cap)
605 "PCIe BW is different than device's capability\n");
607 mlx4_info(dev, "PCIe link speed is %s, device supports %s\n",
608 PCIE_SPEED_STR(speed), PCIE_SPEED_STR(speed_cap));
609 mlx4_info(dev, "PCIe link width is x%d, device supports x%d\n",
614 /*The function checks if there are live vf, return the num of them*/
615 static int mlx4_how_many_lives_vf(struct mlx4_dev *dev)
617 struct mlx4_priv *priv = mlx4_priv(dev);
618 struct mlx4_slave_state *s_state;
622 for (i = 1/*the ppf is 0*/; i < dev->num_slaves; ++i) {
623 s_state = &priv->mfunc.master.slave_state[i];
624 if (s_state->active && s_state->last_cmd !=
625 MLX4_COMM_CMD_RESET) {
626 mlx4_warn(dev, "%s: slave: %d is still active\n",
634 int mlx4_get_parav_qkey(struct mlx4_dev *dev, u32 qpn, u32 *qkey)
636 u32 qk = MLX4_RESERVED_QKEY_BASE;
638 if (qpn >= dev->phys_caps.base_tunnel_sqpn + 8 * MLX4_MFUNC_MAX ||
639 qpn < dev->phys_caps.base_proxy_sqpn)
642 if (qpn >= dev->phys_caps.base_tunnel_sqpn)
644 qk += qpn - dev->phys_caps.base_tunnel_sqpn;
646 qk += qpn - dev->phys_caps.base_proxy_sqpn;
650 EXPORT_SYMBOL(mlx4_get_parav_qkey);
652 void mlx4_sync_pkey_table(struct mlx4_dev *dev, int slave, int port, int i, int val)
654 struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
656 if (!mlx4_is_master(dev))
659 priv->virt2phys_pkey[slave][port - 1][i] = val;
661 EXPORT_SYMBOL(mlx4_sync_pkey_table);
663 void mlx4_put_slave_node_guid(struct mlx4_dev *dev, int slave, __be64 guid)
665 struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
667 if (!mlx4_is_master(dev))
670 priv->slave_node_guids[slave] = guid;
672 EXPORT_SYMBOL(mlx4_put_slave_node_guid);
674 __be64 mlx4_get_slave_node_guid(struct mlx4_dev *dev, int slave)
676 struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
678 if (!mlx4_is_master(dev))
681 return priv->slave_node_guids[slave];
683 EXPORT_SYMBOL(mlx4_get_slave_node_guid);
685 int mlx4_is_slave_active(struct mlx4_dev *dev, int slave)
687 struct mlx4_priv *priv = mlx4_priv(dev);
688 struct mlx4_slave_state *s_slave;
690 if (!mlx4_is_master(dev))
693 s_slave = &priv->mfunc.master.slave_state[slave];
694 return !!s_slave->active;
696 EXPORT_SYMBOL(mlx4_is_slave_active);
698 static void slave_adjust_steering_mode(struct mlx4_dev *dev,
699 struct mlx4_dev_cap *dev_cap,
700 struct mlx4_init_hca_param *hca_param)
702 dev->caps.steering_mode = hca_param->steering_mode;
703 if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED) {
704 dev->caps.num_qp_per_mgm = dev_cap->fs_max_num_qp_per_entry;
705 dev->caps.fs_log_max_ucast_qp_range_size =
706 dev_cap->fs_log_max_ucast_qp_range_size;
708 dev->caps.num_qp_per_mgm =
709 4 * ((1 << hca_param->log_mc_entry_sz)/16 - 2);
711 mlx4_dbg(dev, "Steering mode is: %s\n",
712 mlx4_steering_mode_str(dev->caps.steering_mode));
715 static int mlx4_slave_cap(struct mlx4_dev *dev)
719 struct mlx4_dev_cap dev_cap;
720 struct mlx4_func_cap func_cap;
721 struct mlx4_init_hca_param hca_param;
724 memset(&hca_param, 0, sizeof(hca_param));
725 err = mlx4_QUERY_HCA(dev, &hca_param);
727 mlx4_err(dev, "QUERY_HCA command failed, aborting\n");
731 /* fail if the hca has an unknown global capability
732 * at this time global_caps should be always zeroed
734 if (hca_param.global_caps) {
735 mlx4_err(dev, "Unknown hca global capabilities\n");
739 mlx4_log_num_mgm_entry_size = hca_param.log_mc_entry_sz;
741 dev->caps.hca_core_clock = hca_param.hca_core_clock;
743 memset(&dev_cap, 0, sizeof(dev_cap));
744 dev->caps.max_qp_dest_rdma = 1 << hca_param.log_rd_per_qp;
745 err = mlx4_dev_cap(dev, &dev_cap);
747 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
751 err = mlx4_QUERY_FW(dev);
753 mlx4_err(dev, "QUERY_FW command failed: could not get FW version\n");
755 page_size = ~dev->caps.page_size_cap + 1;
756 mlx4_warn(dev, "HCA minimum page size:%d\n", page_size);
757 if (page_size > PAGE_SIZE) {
758 mlx4_err(dev, "HCA minimum page size of %d bigger than kernel PAGE_SIZE of %ld, aborting\n",
759 page_size, PAGE_SIZE);
763 /* slave gets uar page size from QUERY_HCA fw command */
764 dev->caps.uar_page_size = 1 << (hca_param.uar_page_sz + 12);
766 /* TODO: relax this assumption */
767 if (dev->caps.uar_page_size != PAGE_SIZE) {
768 mlx4_err(dev, "UAR size:%d != kernel PAGE_SIZE of %ld\n",
769 dev->caps.uar_page_size, PAGE_SIZE);
773 memset(&func_cap, 0, sizeof(func_cap));
774 err = mlx4_QUERY_FUNC_CAP(dev, 0, &func_cap);
776 mlx4_err(dev, "QUERY_FUNC_CAP general command failed, aborting (%d)\n",
781 if ((func_cap.pf_context_behaviour | PF_CONTEXT_BEHAVIOUR_MASK) !=
782 PF_CONTEXT_BEHAVIOUR_MASK) {
783 mlx4_err(dev, "Unknown pf context behaviour %x known flags %x\n",
784 func_cap.pf_context_behaviour, PF_CONTEXT_BEHAVIOUR_MASK);
788 dev->caps.num_ports = func_cap.num_ports;
789 dev->quotas.qp = func_cap.qp_quota;
790 dev->quotas.srq = func_cap.srq_quota;
791 dev->quotas.cq = func_cap.cq_quota;
792 dev->quotas.mpt = func_cap.mpt_quota;
793 dev->quotas.mtt = func_cap.mtt_quota;
794 dev->caps.num_qps = 1 << hca_param.log_num_qps;
795 dev->caps.num_srqs = 1 << hca_param.log_num_srqs;
796 dev->caps.num_cqs = 1 << hca_param.log_num_cqs;
797 dev->caps.num_mpts = 1 << hca_param.log_mpt_sz;
798 dev->caps.num_eqs = func_cap.max_eq;
799 dev->caps.reserved_eqs = func_cap.reserved_eq;
800 dev->caps.reserved_lkey = func_cap.reserved_lkey;
801 dev->caps.num_pds = MLX4_NUM_PDS;
802 dev->caps.num_mgms = 0;
803 dev->caps.num_amgms = 0;
805 if (dev->caps.num_ports > MLX4_MAX_PORTS) {
806 mlx4_err(dev, "HCA has %d ports, but we only support %d, aborting\n",
807 dev->caps.num_ports, MLX4_MAX_PORTS);
811 dev->caps.qp0_qkey = kcalloc(dev->caps.num_ports, sizeof(u32), GFP_KERNEL);
812 dev->caps.qp0_tunnel = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
813 dev->caps.qp0_proxy = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
814 dev->caps.qp1_tunnel = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
815 dev->caps.qp1_proxy = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
817 if (!dev->caps.qp0_tunnel || !dev->caps.qp0_proxy ||
818 !dev->caps.qp1_tunnel || !dev->caps.qp1_proxy ||
819 !dev->caps.qp0_qkey) {
824 for (i = 1; i <= dev->caps.num_ports; ++i) {
825 err = mlx4_QUERY_FUNC_CAP(dev, i, &func_cap);
827 mlx4_err(dev, "QUERY_FUNC_CAP port command failed for port %d, aborting (%d)\n",
831 dev->caps.qp0_qkey[i - 1] = func_cap.qp0_qkey;
832 dev->caps.qp0_tunnel[i - 1] = func_cap.qp0_tunnel_qpn;
833 dev->caps.qp0_proxy[i - 1] = func_cap.qp0_proxy_qpn;
834 dev->caps.qp1_tunnel[i - 1] = func_cap.qp1_tunnel_qpn;
835 dev->caps.qp1_proxy[i - 1] = func_cap.qp1_proxy_qpn;
836 dev->caps.port_mask[i] = dev->caps.port_type[i];
837 dev->caps.phys_port_id[i] = func_cap.phys_port_id;
838 if (mlx4_get_slave_pkey_gid_tbl_len(dev, i,
839 &dev->caps.gid_table_len[i],
840 &dev->caps.pkey_table_len[i]))
844 if (dev->caps.uar_page_size * (dev->caps.num_uars -
845 dev->caps.reserved_uars) >
846 pci_resource_len(dev->persist->pdev,
848 mlx4_err(dev, "HCA reported UAR region size of 0x%x bigger than PCI resource 2 size of 0x%llx, aborting\n",
849 dev->caps.uar_page_size * dev->caps.num_uars,
851 pci_resource_len(dev->persist->pdev, 2));
855 if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_64B_EQE_ENABLED) {
856 dev->caps.eqe_size = 64;
857 dev->caps.eqe_factor = 1;
859 dev->caps.eqe_size = 32;
860 dev->caps.eqe_factor = 0;
863 if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_64B_CQE_ENABLED) {
864 dev->caps.cqe_size = 64;
865 dev->caps.userspace_caps |= MLX4_USER_DEV_CAP_LARGE_CQE;
867 dev->caps.cqe_size = 32;
870 if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_EQE_STRIDE_ENABLED) {
871 dev->caps.eqe_size = hca_param.eqe_size;
872 dev->caps.eqe_factor = 0;
875 if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_CQE_STRIDE_ENABLED) {
876 dev->caps.cqe_size = hca_param.cqe_size;
877 /* User still need to know when CQE > 32B */
878 dev->caps.userspace_caps |= MLX4_USER_DEV_CAP_LARGE_CQE;
881 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
882 mlx4_warn(dev, "Timestamping is not supported in slave mode\n");
884 slave_adjust_steering_mode(dev, &dev_cap, &hca_param);
886 if (func_cap.extra_flags & MLX4_QUERY_FUNC_FLAGS_BF_RES_QP &&
887 dev->caps.bf_reg_size)
888 dev->caps.alloc_res_qp_mask |= MLX4_RESERVE_ETH_BF_QP;
890 if (func_cap.extra_flags & MLX4_QUERY_FUNC_FLAGS_A0_RES_QP)
891 dev->caps.alloc_res_qp_mask |= MLX4_RESERVE_A0_QP;
896 kfree(dev->caps.qp0_qkey);
897 kfree(dev->caps.qp0_tunnel);
898 kfree(dev->caps.qp0_proxy);
899 kfree(dev->caps.qp1_tunnel);
900 kfree(dev->caps.qp1_proxy);
901 dev->caps.qp0_qkey = NULL;
902 dev->caps.qp0_tunnel = NULL;
903 dev->caps.qp0_proxy = NULL;
904 dev->caps.qp1_tunnel = NULL;
905 dev->caps.qp1_proxy = NULL;
910 static void mlx4_request_modules(struct mlx4_dev *dev)
913 int has_ib_port = false;
914 int has_eth_port = false;
915 #define EN_DRV_NAME "mlx4_en"
916 #define IB_DRV_NAME "mlx4_ib"
918 for (port = 1; port <= dev->caps.num_ports; port++) {
919 if (dev->caps.port_type[port] == MLX4_PORT_TYPE_IB)
921 else if (dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH)
926 request_module_nowait(EN_DRV_NAME);
927 if (has_ib_port || (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE))
928 request_module_nowait(IB_DRV_NAME);
932 * Change the port configuration of the device.
933 * Every user of this function must hold the port mutex.
935 int mlx4_change_port_types(struct mlx4_dev *dev,
936 enum mlx4_port_type *port_types)
942 for (port = 0; port < dev->caps.num_ports; port++) {
943 /* Change the port type only if the new type is different
944 * from the current, and not set to Auto */
945 if (port_types[port] != dev->caps.port_type[port + 1])
949 mlx4_unregister_device(dev);
950 for (port = 1; port <= dev->caps.num_ports; port++) {
951 mlx4_CLOSE_PORT(dev, port);
952 dev->caps.port_type[port] = port_types[port - 1];
953 err = mlx4_SET_PORT(dev, port, -1);
955 mlx4_err(dev, "Failed to set port %d, aborting\n",
960 mlx4_set_port_mask(dev);
961 err = mlx4_register_device(dev);
963 mlx4_err(dev, "Failed to register device\n");
966 mlx4_request_modules(dev);
973 static ssize_t show_port_type(struct device *dev,
974 struct device_attribute *attr,
977 struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
979 struct mlx4_dev *mdev = info->dev;
983 (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_IB) ?
985 if (mdev->caps.possible_type[info->port] == MLX4_PORT_TYPE_AUTO)
986 sprintf(buf, "auto (%s)\n", type);
988 sprintf(buf, "%s\n", type);
993 static ssize_t set_port_type(struct device *dev,
994 struct device_attribute *attr,
995 const char *buf, size_t count)
997 struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
999 struct mlx4_dev *mdev = info->dev;
1000 struct mlx4_priv *priv = mlx4_priv(mdev);
1001 enum mlx4_port_type types[MLX4_MAX_PORTS];
1002 enum mlx4_port_type new_types[MLX4_MAX_PORTS];
1003 static DEFINE_MUTEX(set_port_type_mutex);
1007 mutex_lock(&set_port_type_mutex);
1009 if (!strcmp(buf, "ib\n"))
1010 info->tmp_type = MLX4_PORT_TYPE_IB;
1011 else if (!strcmp(buf, "eth\n"))
1012 info->tmp_type = MLX4_PORT_TYPE_ETH;
1013 else if (!strcmp(buf, "auto\n"))
1014 info->tmp_type = MLX4_PORT_TYPE_AUTO;
1016 mlx4_err(mdev, "%s is not supported port type\n", buf);
1021 mlx4_stop_sense(mdev);
1022 mutex_lock(&priv->port_mutex);
1023 /* Possible type is always the one that was delivered */
1024 mdev->caps.possible_type[info->port] = info->tmp_type;
1026 for (i = 0; i < mdev->caps.num_ports; i++) {
1027 types[i] = priv->port[i+1].tmp_type ? priv->port[i+1].tmp_type :
1028 mdev->caps.possible_type[i+1];
1029 if (types[i] == MLX4_PORT_TYPE_AUTO)
1030 types[i] = mdev->caps.port_type[i+1];
1033 if (!(mdev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
1034 !(mdev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT)) {
1035 for (i = 1; i <= mdev->caps.num_ports; i++) {
1036 if (mdev->caps.possible_type[i] == MLX4_PORT_TYPE_AUTO) {
1037 mdev->caps.possible_type[i] = mdev->caps.port_type[i];
1043 mlx4_err(mdev, "Auto sensing is not supported on this HCA. Set only 'eth' or 'ib' for both ports (should be the same)\n");
1047 mlx4_do_sense_ports(mdev, new_types, types);
1049 err = mlx4_check_port_params(mdev, new_types);
1053 /* We are about to apply the changes after the configuration
1054 * was verified, no need to remember the temporary types
1056 for (i = 0; i < mdev->caps.num_ports; i++)
1057 priv->port[i + 1].tmp_type = 0;
1059 err = mlx4_change_port_types(mdev, new_types);
1062 mlx4_start_sense(mdev);
1063 mutex_unlock(&priv->port_mutex);
1065 mutex_unlock(&set_port_type_mutex);
1067 return err ? err : count;
1078 static inline int int_to_ibta_mtu(int mtu)
1081 case 256: return IB_MTU_256;
1082 case 512: return IB_MTU_512;
1083 case 1024: return IB_MTU_1024;
1084 case 2048: return IB_MTU_2048;
1085 case 4096: return IB_MTU_4096;
1090 static inline int ibta_mtu_to_int(enum ibta_mtu mtu)
1093 case IB_MTU_256: return 256;
1094 case IB_MTU_512: return 512;
1095 case IB_MTU_1024: return 1024;
1096 case IB_MTU_2048: return 2048;
1097 case IB_MTU_4096: return 4096;
1102 static ssize_t show_port_ib_mtu(struct device *dev,
1103 struct device_attribute *attr,
1106 struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1108 struct mlx4_dev *mdev = info->dev;
1110 if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH)
1111 mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
1113 sprintf(buf, "%d\n",
1114 ibta_mtu_to_int(mdev->caps.port_ib_mtu[info->port]));
1118 static ssize_t set_port_ib_mtu(struct device *dev,
1119 struct device_attribute *attr,
1120 const char *buf, size_t count)
1122 struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1124 struct mlx4_dev *mdev = info->dev;
1125 struct mlx4_priv *priv = mlx4_priv(mdev);
1126 int err, port, mtu, ibta_mtu = -1;
1128 if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH) {
1129 mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
1133 err = kstrtoint(buf, 0, &mtu);
1135 ibta_mtu = int_to_ibta_mtu(mtu);
1137 if (err || ibta_mtu < 0) {
1138 mlx4_err(mdev, "%s is invalid IBTA mtu\n", buf);
1142 mdev->caps.port_ib_mtu[info->port] = ibta_mtu;
1144 mlx4_stop_sense(mdev);
1145 mutex_lock(&priv->port_mutex);
1146 mlx4_unregister_device(mdev);
1147 for (port = 1; port <= mdev->caps.num_ports; port++) {
1148 mlx4_CLOSE_PORT(mdev, port);
1149 err = mlx4_SET_PORT(mdev, port, -1);
1151 mlx4_err(mdev, "Failed to set port %d, aborting\n",
1156 err = mlx4_register_device(mdev);
1158 mutex_unlock(&priv->port_mutex);
1159 mlx4_start_sense(mdev);
1160 return err ? err : count;
1163 static int mlx4_load_fw(struct mlx4_dev *dev)
1165 struct mlx4_priv *priv = mlx4_priv(dev);
1168 priv->fw.fw_icm = mlx4_alloc_icm(dev, priv->fw.fw_pages,
1169 GFP_HIGHUSER | __GFP_NOWARN, 0);
1170 if (!priv->fw.fw_icm) {
1171 mlx4_err(dev, "Couldn't allocate FW area, aborting\n");
1175 err = mlx4_MAP_FA(dev, priv->fw.fw_icm);
1177 mlx4_err(dev, "MAP_FA command failed, aborting\n");
1181 err = mlx4_RUN_FW(dev);
1183 mlx4_err(dev, "RUN_FW command failed, aborting\n");
1193 mlx4_free_icm(dev, priv->fw.fw_icm, 0);
1197 static int mlx4_init_cmpt_table(struct mlx4_dev *dev, u64 cmpt_base,
1200 struct mlx4_priv *priv = mlx4_priv(dev);
1204 err = mlx4_init_icm_table(dev, &priv->qp_table.cmpt_table,
1206 ((u64) (MLX4_CMPT_TYPE_QP *
1207 cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1208 cmpt_entry_sz, dev->caps.num_qps,
1209 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1214 err = mlx4_init_icm_table(dev, &priv->srq_table.cmpt_table,
1216 ((u64) (MLX4_CMPT_TYPE_SRQ *
1217 cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1218 cmpt_entry_sz, dev->caps.num_srqs,
1219 dev->caps.reserved_srqs, 0, 0);
1223 err = mlx4_init_icm_table(dev, &priv->cq_table.cmpt_table,
1225 ((u64) (MLX4_CMPT_TYPE_CQ *
1226 cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1227 cmpt_entry_sz, dev->caps.num_cqs,
1228 dev->caps.reserved_cqs, 0, 0);
1232 num_eqs = dev->phys_caps.num_phys_eqs;
1233 err = mlx4_init_icm_table(dev, &priv->eq_table.cmpt_table,
1235 ((u64) (MLX4_CMPT_TYPE_EQ *
1236 cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1237 cmpt_entry_sz, num_eqs, num_eqs, 0, 0);
1244 mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1247 mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1250 mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1256 static int mlx4_init_icm(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap,
1257 struct mlx4_init_hca_param *init_hca, u64 icm_size)
1259 struct mlx4_priv *priv = mlx4_priv(dev);
1264 err = mlx4_SET_ICM_SIZE(dev, icm_size, &aux_pages);
1266 mlx4_err(dev, "SET_ICM_SIZE command failed, aborting\n");
1270 mlx4_dbg(dev, "%lld KB of HCA context requires %lld KB aux memory\n",
1271 (unsigned long long) icm_size >> 10,
1272 (unsigned long long) aux_pages << 2);
1274 priv->fw.aux_icm = mlx4_alloc_icm(dev, aux_pages,
1275 GFP_HIGHUSER | __GFP_NOWARN, 0);
1276 if (!priv->fw.aux_icm) {
1277 mlx4_err(dev, "Couldn't allocate aux memory, aborting\n");
1281 err = mlx4_MAP_ICM_AUX(dev, priv->fw.aux_icm);
1283 mlx4_err(dev, "MAP_ICM_AUX command failed, aborting\n");
1287 err = mlx4_init_cmpt_table(dev, init_hca->cmpt_base, dev_cap->cmpt_entry_sz);
1289 mlx4_err(dev, "Failed to map cMPT context memory, aborting\n");
1294 num_eqs = dev->phys_caps.num_phys_eqs;
1295 err = mlx4_init_icm_table(dev, &priv->eq_table.table,
1296 init_hca->eqc_base, dev_cap->eqc_entry_sz,
1297 num_eqs, num_eqs, 0, 0);
1299 mlx4_err(dev, "Failed to map EQ context memory, aborting\n");
1300 goto err_unmap_cmpt;
1304 * Reserved MTT entries must be aligned up to a cacheline
1305 * boundary, since the FW will write to them, while the driver
1306 * writes to all other MTT entries. (The variable
1307 * dev->caps.mtt_entry_sz below is really the MTT segment
1308 * size, not the raw entry size)
1310 dev->caps.reserved_mtts =
1311 ALIGN(dev->caps.reserved_mtts * dev->caps.mtt_entry_sz,
1312 dma_get_cache_alignment()) / dev->caps.mtt_entry_sz;
1314 err = mlx4_init_icm_table(dev, &priv->mr_table.mtt_table,
1316 dev->caps.mtt_entry_sz,
1318 dev->caps.reserved_mtts, 1, 0);
1320 mlx4_err(dev, "Failed to map MTT context memory, aborting\n");
1324 err = mlx4_init_icm_table(dev, &priv->mr_table.dmpt_table,
1325 init_hca->dmpt_base,
1326 dev_cap->dmpt_entry_sz,
1328 dev->caps.reserved_mrws, 1, 1);
1330 mlx4_err(dev, "Failed to map dMPT context memory, aborting\n");
1334 err = mlx4_init_icm_table(dev, &priv->qp_table.qp_table,
1336 dev_cap->qpc_entry_sz,
1338 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1341 mlx4_err(dev, "Failed to map QP context memory, aborting\n");
1342 goto err_unmap_dmpt;
1345 err = mlx4_init_icm_table(dev, &priv->qp_table.auxc_table,
1346 init_hca->auxc_base,
1347 dev_cap->aux_entry_sz,
1349 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1352 mlx4_err(dev, "Failed to map AUXC context memory, aborting\n");
1356 err = mlx4_init_icm_table(dev, &priv->qp_table.altc_table,
1357 init_hca->altc_base,
1358 dev_cap->altc_entry_sz,
1360 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1363 mlx4_err(dev, "Failed to map ALTC context memory, aborting\n");
1364 goto err_unmap_auxc;
1367 err = mlx4_init_icm_table(dev, &priv->qp_table.rdmarc_table,
1368 init_hca->rdmarc_base,
1369 dev_cap->rdmarc_entry_sz << priv->qp_table.rdmarc_shift,
1371 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1374 mlx4_err(dev, "Failed to map RDMARC context memory, aborting\n");
1375 goto err_unmap_altc;
1378 err = mlx4_init_icm_table(dev, &priv->cq_table.table,
1380 dev_cap->cqc_entry_sz,
1382 dev->caps.reserved_cqs, 0, 0);
1384 mlx4_err(dev, "Failed to map CQ context memory, aborting\n");
1385 goto err_unmap_rdmarc;
1388 err = mlx4_init_icm_table(dev, &priv->srq_table.table,
1389 init_hca->srqc_base,
1390 dev_cap->srq_entry_sz,
1392 dev->caps.reserved_srqs, 0, 0);
1394 mlx4_err(dev, "Failed to map SRQ context memory, aborting\n");
1399 * For flow steering device managed mode it is required to use
1400 * mlx4_init_icm_table. For B0 steering mode it's not strictly
1401 * required, but for simplicity just map the whole multicast
1402 * group table now. The table isn't very big and it's a lot
1403 * easier than trying to track ref counts.
1405 err = mlx4_init_icm_table(dev, &priv->mcg_table.table,
1407 mlx4_get_mgm_entry_size(dev),
1408 dev->caps.num_mgms + dev->caps.num_amgms,
1409 dev->caps.num_mgms + dev->caps.num_amgms,
1412 mlx4_err(dev, "Failed to map MCG context memory, aborting\n");
1419 mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1422 mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1425 mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1428 mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1431 mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1434 mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1437 mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1440 mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
1443 mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
1446 mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1447 mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1448 mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1449 mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1452 mlx4_UNMAP_ICM_AUX(dev);
1455 mlx4_free_icm(dev, priv->fw.aux_icm, 0);
1460 static void mlx4_free_icms(struct mlx4_dev *dev)
1462 struct mlx4_priv *priv = mlx4_priv(dev);
1464 mlx4_cleanup_icm_table(dev, &priv->mcg_table.table);
1465 mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1466 mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1467 mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1468 mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1469 mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1470 mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1471 mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1472 mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
1473 mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
1474 mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1475 mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1476 mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1477 mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1479 mlx4_UNMAP_ICM_AUX(dev);
1480 mlx4_free_icm(dev, priv->fw.aux_icm, 0);
1483 static void mlx4_slave_exit(struct mlx4_dev *dev)
1485 struct mlx4_priv *priv = mlx4_priv(dev);
1487 mutex_lock(&priv->cmd.slave_cmd_mutex);
1488 if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, MLX4_COMM_CMD_NA_OP,
1490 mlx4_warn(dev, "Failed to close slave function\n");
1491 mutex_unlock(&priv->cmd.slave_cmd_mutex);
1494 static int map_bf_area(struct mlx4_dev *dev)
1496 struct mlx4_priv *priv = mlx4_priv(dev);
1497 resource_size_t bf_start;
1498 resource_size_t bf_len;
1501 if (!dev->caps.bf_reg_size)
1504 bf_start = pci_resource_start(dev->persist->pdev, 2) +
1505 (dev->caps.num_uars << PAGE_SHIFT);
1506 bf_len = pci_resource_len(dev->persist->pdev, 2) -
1507 (dev->caps.num_uars << PAGE_SHIFT);
1508 priv->bf_mapping = io_mapping_create_wc(bf_start, bf_len);
1509 if (!priv->bf_mapping)
1515 static void unmap_bf_area(struct mlx4_dev *dev)
1517 if (mlx4_priv(dev)->bf_mapping)
1518 io_mapping_free(mlx4_priv(dev)->bf_mapping);
1521 cycle_t mlx4_read_clock(struct mlx4_dev *dev)
1523 u32 clockhi, clocklo, clockhi1;
1526 struct mlx4_priv *priv = mlx4_priv(dev);
1528 for (i = 0; i < 10; i++) {
1529 clockhi = swab32(readl(priv->clock_mapping));
1530 clocklo = swab32(readl(priv->clock_mapping + 4));
1531 clockhi1 = swab32(readl(priv->clock_mapping));
1532 if (clockhi == clockhi1)
1536 cycles = (u64) clockhi << 32 | (u64) clocklo;
1540 EXPORT_SYMBOL_GPL(mlx4_read_clock);
1543 static int map_internal_clock(struct mlx4_dev *dev)
1545 struct mlx4_priv *priv = mlx4_priv(dev);
1547 priv->clock_mapping =
1548 ioremap(pci_resource_start(dev->persist->pdev,
1549 priv->fw.clock_bar) +
1550 priv->fw.clock_offset, MLX4_CLOCK_SIZE);
1552 if (!priv->clock_mapping)
1558 static void unmap_internal_clock(struct mlx4_dev *dev)
1560 struct mlx4_priv *priv = mlx4_priv(dev);
1562 if (priv->clock_mapping)
1563 iounmap(priv->clock_mapping);
1566 static void mlx4_close_hca(struct mlx4_dev *dev)
1568 unmap_internal_clock(dev);
1570 if (mlx4_is_slave(dev))
1571 mlx4_slave_exit(dev);
1573 mlx4_CLOSE_HCA(dev, 0);
1574 mlx4_free_icms(dev);
1578 static void mlx4_close_fw(struct mlx4_dev *dev)
1580 if (!mlx4_is_slave(dev)) {
1582 mlx4_free_icm(dev, mlx4_priv(dev)->fw.fw_icm, 0);
1586 static int mlx4_comm_check_offline(struct mlx4_dev *dev)
1588 #define COMM_CHAN_OFFLINE_OFFSET 0x09
1593 struct mlx4_priv *priv = mlx4_priv(dev);
1595 end = msecs_to_jiffies(MLX4_COMM_OFFLINE_TIME_OUT) + jiffies;
1596 while (time_before(jiffies, end)) {
1597 comm_flags = swab32(readl((__iomem char *)priv->mfunc.comm +
1598 MLX4_COMM_CHAN_FLAGS));
1599 offline_bit = (comm_flags &
1600 (u32)(1 << COMM_CHAN_OFFLINE_OFFSET));
1603 /* There are cases as part of AER/Reset flow that PF needs
1604 * around 100 msec to load. We therefore sleep for 100 msec
1605 * to allow other tasks to make use of that CPU during this
1610 mlx4_err(dev, "Communication channel is offline.\n");
1614 static void mlx4_reset_vf_support(struct mlx4_dev *dev)
1616 #define COMM_CHAN_RST_OFFSET 0x1e
1618 struct mlx4_priv *priv = mlx4_priv(dev);
1622 comm_caps = swab32(readl((__iomem char *)priv->mfunc.comm +
1623 MLX4_COMM_CHAN_CAPS));
1624 comm_rst = (comm_caps & (u32)(1 << COMM_CHAN_RST_OFFSET));
1627 dev->caps.vf_caps |= MLX4_VF_CAP_FLAG_RESET;
1630 static int mlx4_init_slave(struct mlx4_dev *dev)
1632 struct mlx4_priv *priv = mlx4_priv(dev);
1633 u64 dma = (u64) priv->mfunc.vhcr_dma;
1634 int ret_from_reset = 0;
1636 u32 cmd_channel_ver;
1638 if (atomic_read(&pf_loading)) {
1639 mlx4_warn(dev, "PF is not ready - Deferring probe\n");
1640 return -EPROBE_DEFER;
1643 mutex_lock(&priv->cmd.slave_cmd_mutex);
1644 priv->cmd.max_cmds = 1;
1645 if (mlx4_comm_check_offline(dev)) {
1646 mlx4_err(dev, "PF is not responsive, skipping initialization\n");
1650 mlx4_reset_vf_support(dev);
1651 mlx4_warn(dev, "Sending reset\n");
1652 ret_from_reset = mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0,
1653 MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME);
1654 /* if we are in the middle of flr the slave will try
1655 * NUM_OF_RESET_RETRIES times before leaving.*/
1656 if (ret_from_reset) {
1657 if (MLX4_DELAY_RESET_SLAVE == ret_from_reset) {
1658 mlx4_warn(dev, "slave is currently in the middle of FLR - Deferring probe\n");
1659 mutex_unlock(&priv->cmd.slave_cmd_mutex);
1660 return -EPROBE_DEFER;
1665 /* check the driver version - the slave I/F revision
1666 * must match the master's */
1667 slave_read = swab32(readl(&priv->mfunc.comm->slave_read));
1668 cmd_channel_ver = mlx4_comm_get_version();
1670 if (MLX4_COMM_GET_IF_REV(cmd_channel_ver) !=
1671 MLX4_COMM_GET_IF_REV(slave_read)) {
1672 mlx4_err(dev, "slave driver version is not supported by the master\n");
1676 mlx4_warn(dev, "Sending vhcr0\n");
1677 if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR0, dma >> 48,
1678 MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
1680 if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR1, dma >> 32,
1681 MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
1683 if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR2, dma >> 16,
1684 MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
1686 if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR_EN, dma,
1687 MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
1690 mutex_unlock(&priv->cmd.slave_cmd_mutex);
1694 mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, MLX4_COMM_CMD_NA_OP, 0);
1696 mutex_unlock(&priv->cmd.slave_cmd_mutex);
1700 static void mlx4_parav_master_pf_caps(struct mlx4_dev *dev)
1704 for (i = 1; i <= dev->caps.num_ports; i++) {
1705 if (dev->caps.port_type[i] == MLX4_PORT_TYPE_ETH)
1706 dev->caps.gid_table_len[i] =
1707 mlx4_get_slave_num_gids(dev, 0, i);
1709 dev->caps.gid_table_len[i] = 1;
1710 dev->caps.pkey_table_len[i] =
1711 dev->phys_caps.pkey_phys_table_len[i] - 1;
1715 static int choose_log_fs_mgm_entry_size(int qp_per_entry)
1717 int i = MLX4_MIN_MGM_LOG_ENTRY_SIZE;
1719 for (i = MLX4_MIN_MGM_LOG_ENTRY_SIZE; i <= MLX4_MAX_MGM_LOG_ENTRY_SIZE;
1721 if (qp_per_entry <= 4 * ((1 << i) / 16 - 2))
1725 return (i <= MLX4_MAX_MGM_LOG_ENTRY_SIZE) ? i : -1;
1728 static const char *dmfs_high_rate_steering_mode_str(int dmfs_high_steer_mode)
1730 switch (dmfs_high_steer_mode) {
1731 case MLX4_STEERING_DMFS_A0_DEFAULT:
1732 return "default performance";
1734 case MLX4_STEERING_DMFS_A0_DYNAMIC:
1735 return "dynamic hybrid mode";
1737 case MLX4_STEERING_DMFS_A0_STATIC:
1738 return "performance optimized for limited rule configuration (static)";
1740 case MLX4_STEERING_DMFS_A0_DISABLE:
1741 return "disabled performance optimized steering";
1743 case MLX4_STEERING_DMFS_A0_NOT_SUPPORTED:
1744 return "performance optimized steering not supported";
1747 return "Unrecognized mode";
1751 #define MLX4_DMFS_A0_STEERING (1UL << 2)
1753 static void choose_steering_mode(struct mlx4_dev *dev,
1754 struct mlx4_dev_cap *dev_cap)
1756 if (mlx4_log_num_mgm_entry_size <= 0) {
1757 if ((-mlx4_log_num_mgm_entry_size) & MLX4_DMFS_A0_STEERING) {
1758 if (dev->caps.dmfs_high_steer_mode ==
1759 MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
1760 mlx4_err(dev, "DMFS high rate mode not supported\n");
1762 dev->caps.dmfs_high_steer_mode =
1763 MLX4_STEERING_DMFS_A0_STATIC;
1767 if (mlx4_log_num_mgm_entry_size <= 0 &&
1768 dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_FS_EN &&
1769 (!mlx4_is_mfunc(dev) ||
1770 (dev_cap->fs_max_num_qp_per_entry >=
1771 (dev->persist->num_vfs + 1))) &&
1772 choose_log_fs_mgm_entry_size(dev_cap->fs_max_num_qp_per_entry) >=
1773 MLX4_MIN_MGM_LOG_ENTRY_SIZE) {
1774 dev->oper_log_mgm_entry_size =
1775 choose_log_fs_mgm_entry_size(dev_cap->fs_max_num_qp_per_entry);
1776 dev->caps.steering_mode = MLX4_STEERING_MODE_DEVICE_MANAGED;
1777 dev->caps.num_qp_per_mgm = dev_cap->fs_max_num_qp_per_entry;
1778 dev->caps.fs_log_max_ucast_qp_range_size =
1779 dev_cap->fs_log_max_ucast_qp_range_size;
1781 if (dev->caps.dmfs_high_steer_mode !=
1782 MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
1783 dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_DISABLE;
1784 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER &&
1785 dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER)
1786 dev->caps.steering_mode = MLX4_STEERING_MODE_B0;
1788 dev->caps.steering_mode = MLX4_STEERING_MODE_A0;
1790 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER ||
1791 dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER)
1792 mlx4_warn(dev, "Must have both UC_STEER and MC_STEER flags set to use B0 steering - falling back to A0 steering mode\n");
1794 dev->oper_log_mgm_entry_size =
1795 mlx4_log_num_mgm_entry_size > 0 ?
1796 mlx4_log_num_mgm_entry_size :
1797 MLX4_DEFAULT_MGM_LOG_ENTRY_SIZE;
1798 dev->caps.num_qp_per_mgm = mlx4_get_qp_per_mgm(dev);
1800 mlx4_dbg(dev, "Steering mode is: %s, oper_log_mgm_entry_size = %d, modparam log_num_mgm_entry_size = %d\n",
1801 mlx4_steering_mode_str(dev->caps.steering_mode),
1802 dev->oper_log_mgm_entry_size,
1803 mlx4_log_num_mgm_entry_size);
1806 static void choose_tunnel_offload_mode(struct mlx4_dev *dev,
1807 struct mlx4_dev_cap *dev_cap)
1809 if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED &&
1810 dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_VXLAN_OFFLOADS)
1811 dev->caps.tunnel_offload_mode = MLX4_TUNNEL_OFFLOAD_MODE_VXLAN;
1813 dev->caps.tunnel_offload_mode = MLX4_TUNNEL_OFFLOAD_MODE_NONE;
1815 mlx4_dbg(dev, "Tunneling offload mode is: %s\n", (dev->caps.tunnel_offload_mode
1816 == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) ? "vxlan" : "none");
1819 static int mlx4_validate_optimized_steering(struct mlx4_dev *dev)
1822 struct mlx4_port_cap port_cap;
1824 if (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
1827 for (i = 1; i <= dev->caps.num_ports; i++) {
1828 if (mlx4_dev_port(dev, i, &port_cap)) {
1830 "QUERY_DEV_CAP command failed, can't veify DMFS high rate steering.\n");
1831 } else if ((dev->caps.dmfs_high_steer_mode !=
1832 MLX4_STEERING_DMFS_A0_DEFAULT) &&
1833 (port_cap.dmfs_optimized_state ==
1834 !!(dev->caps.dmfs_high_steer_mode ==
1835 MLX4_STEERING_DMFS_A0_DISABLE))) {
1837 "DMFS high rate steer mode differ, driver requested %s but %s in FW.\n",
1838 dmfs_high_rate_steering_mode_str(
1839 dev->caps.dmfs_high_steer_mode),
1840 (port_cap.dmfs_optimized_state ?
1841 "enabled" : "disabled"));
1848 static int mlx4_init_fw(struct mlx4_dev *dev)
1850 struct mlx4_mod_stat_cfg mlx4_cfg;
1853 if (!mlx4_is_slave(dev)) {
1854 err = mlx4_QUERY_FW(dev);
1857 mlx4_info(dev, "non-primary physical function, skipping\n");
1859 mlx4_err(dev, "QUERY_FW command failed, aborting\n");
1863 err = mlx4_load_fw(dev);
1865 mlx4_err(dev, "Failed to start FW, aborting\n");
1869 mlx4_cfg.log_pg_sz_m = 1;
1870 mlx4_cfg.log_pg_sz = 0;
1871 err = mlx4_MOD_STAT_CFG(dev, &mlx4_cfg);
1873 mlx4_warn(dev, "Failed to override log_pg_sz parameter\n");
1879 static int mlx4_init_hca(struct mlx4_dev *dev)
1881 struct mlx4_priv *priv = mlx4_priv(dev);
1882 struct mlx4_adapter adapter;
1883 struct mlx4_dev_cap dev_cap;
1884 struct mlx4_profile profile;
1885 struct mlx4_init_hca_param init_hca;
1887 struct mlx4_config_dev_params params;
1890 if (!mlx4_is_slave(dev)) {
1891 err = mlx4_dev_cap(dev, &dev_cap);
1893 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
1897 choose_steering_mode(dev, &dev_cap);
1898 choose_tunnel_offload_mode(dev, &dev_cap);
1900 if (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC &&
1901 mlx4_is_master(dev))
1902 dev->caps.function_caps |= MLX4_FUNC_CAP_DMFS_A0_STATIC;
1904 err = mlx4_get_phys_port_id(dev);
1906 mlx4_err(dev, "Fail to get physical port id\n");
1908 if (mlx4_is_master(dev))
1909 mlx4_parav_master_pf_caps(dev);
1911 if (mlx4_low_memory_profile()) {
1912 mlx4_info(dev, "Running from within kdump kernel. Using low memory profile\n");
1913 profile = low_mem_profile;
1915 profile = default_profile;
1917 if (dev->caps.steering_mode ==
1918 MLX4_STEERING_MODE_DEVICE_MANAGED)
1919 profile.num_mcg = MLX4_FS_NUM_MCG;
1921 icm_size = mlx4_make_profile(dev, &profile, &dev_cap,
1923 if ((long long) icm_size < 0) {
1928 dev->caps.max_fmr_maps = (1 << (32 - ilog2(dev->caps.num_mpts))) - 1;
1930 init_hca.log_uar_sz = ilog2(dev->caps.num_uars);
1931 init_hca.uar_page_sz = PAGE_SHIFT - 12;
1932 init_hca.mw_enabled = 0;
1933 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
1934 dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN)
1935 init_hca.mw_enabled = INIT_HCA_TPT_MW_ENABLE;
1937 err = mlx4_init_icm(dev, &dev_cap, &init_hca, icm_size);
1941 err = mlx4_INIT_HCA(dev, &init_hca);
1943 mlx4_err(dev, "INIT_HCA command failed, aborting\n");
1947 if (dev_cap.flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS) {
1948 err = mlx4_query_func(dev, &dev_cap);
1950 mlx4_err(dev, "QUERY_FUNC command failed, aborting.\n");
1952 } else if (err & MLX4_QUERY_FUNC_NUM_SYS_EQS) {
1953 dev->caps.num_eqs = dev_cap.max_eqs;
1954 dev->caps.reserved_eqs = dev_cap.reserved_eqs;
1955 dev->caps.reserved_uars = dev_cap.reserved_uars;
1960 * If TS is supported by FW
1961 * read HCA frequency by QUERY_HCA command
1963 if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS) {
1964 memset(&init_hca, 0, sizeof(init_hca));
1965 err = mlx4_QUERY_HCA(dev, &init_hca);
1967 mlx4_err(dev, "QUERY_HCA command failed, disable timestamp\n");
1968 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
1970 dev->caps.hca_core_clock =
1971 init_hca.hca_core_clock;
1974 /* In case we got HCA frequency 0 - disable timestamping
1975 * to avoid dividing by zero
1977 if (!dev->caps.hca_core_clock) {
1978 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
1980 "HCA frequency is 0 - timestamping is not supported\n");
1981 } else if (map_internal_clock(dev)) {
1983 * Map internal clock,
1984 * in case of failure disable timestamping
1986 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
1987 mlx4_err(dev, "Failed to map internal clock. Timestamping is not supported\n");
1991 if (dev->caps.dmfs_high_steer_mode !=
1992 MLX4_STEERING_DMFS_A0_NOT_SUPPORTED) {
1993 if (mlx4_validate_optimized_steering(dev))
1994 mlx4_warn(dev, "Optimized steering validation failed\n");
1996 if (dev->caps.dmfs_high_steer_mode ==
1997 MLX4_STEERING_DMFS_A0_DISABLE) {
1998 dev->caps.dmfs_high_rate_qpn_base =
1999 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
2000 dev->caps.dmfs_high_rate_qpn_range =
2001 MLX4_A0_STEERING_TABLE_SIZE;
2004 mlx4_dbg(dev, "DMFS high rate steer mode is: %s\n",
2005 dmfs_high_rate_steering_mode_str(
2006 dev->caps.dmfs_high_steer_mode));
2009 err = mlx4_init_slave(dev);
2011 if (err != -EPROBE_DEFER)
2012 mlx4_err(dev, "Failed to initialize slave\n");
2016 err = mlx4_slave_cap(dev);
2018 mlx4_err(dev, "Failed to obtain slave caps\n");
2023 if (map_bf_area(dev))
2024 mlx4_dbg(dev, "Failed to map blue flame area\n");
2026 /*Only the master set the ports, all the rest got it from it.*/
2027 if (!mlx4_is_slave(dev))
2028 mlx4_set_port_mask(dev);
2030 err = mlx4_QUERY_ADAPTER(dev, &adapter);
2032 mlx4_err(dev, "QUERY_ADAPTER command failed, aborting\n");
2036 /* Query CONFIG_DEV parameters */
2037 err = mlx4_config_dev_retrieval(dev, ¶ms);
2038 if (err && err != -ENOTSUPP) {
2039 mlx4_err(dev, "Failed to query CONFIG_DEV parameters\n");
2041 dev->caps.rx_checksum_flags_port[1] = params.rx_csum_flags_port_1;
2042 dev->caps.rx_checksum_flags_port[2] = params.rx_csum_flags_port_2;
2044 priv->eq_table.inta_pin = adapter.inta_pin;
2045 memcpy(dev->board_id, adapter.board_id, sizeof dev->board_id);
2050 unmap_internal_clock(dev);
2053 if (mlx4_is_slave(dev)) {
2054 kfree(dev->caps.qp0_qkey);
2055 kfree(dev->caps.qp0_tunnel);
2056 kfree(dev->caps.qp0_proxy);
2057 kfree(dev->caps.qp1_tunnel);
2058 kfree(dev->caps.qp1_proxy);
2062 if (mlx4_is_slave(dev))
2063 mlx4_slave_exit(dev);
2065 mlx4_CLOSE_HCA(dev, 0);
2068 if (!mlx4_is_slave(dev))
2069 mlx4_free_icms(dev);
2074 static int mlx4_init_counters_table(struct mlx4_dev *dev)
2076 struct mlx4_priv *priv = mlx4_priv(dev);
2079 if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2082 nent = dev->caps.max_counters;
2083 return mlx4_bitmap_init(&priv->counters_bitmap, nent, nent - 1, 0, 0);
2086 static void mlx4_cleanup_counters_table(struct mlx4_dev *dev)
2088 mlx4_bitmap_cleanup(&mlx4_priv(dev)->counters_bitmap);
2091 int __mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx)
2093 struct mlx4_priv *priv = mlx4_priv(dev);
2095 if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2098 *idx = mlx4_bitmap_alloc(&priv->counters_bitmap);
2105 int mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx)
2110 if (mlx4_is_mfunc(dev)) {
2111 err = mlx4_cmd_imm(dev, 0, &out_param, RES_COUNTER,
2112 RES_OP_RESERVE, MLX4_CMD_ALLOC_RES,
2113 MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
2115 *idx = get_param_l(&out_param);
2119 return __mlx4_counter_alloc(dev, idx);
2121 EXPORT_SYMBOL_GPL(mlx4_counter_alloc);
2123 void __mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
2125 mlx4_bitmap_free(&mlx4_priv(dev)->counters_bitmap, idx, MLX4_USE_RR);
2129 void mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
2133 if (mlx4_is_mfunc(dev)) {
2134 set_param_l(&in_param, idx);
2135 mlx4_cmd(dev, in_param, RES_COUNTER, RES_OP_RESERVE,
2136 MLX4_CMD_FREE_RES, MLX4_CMD_TIME_CLASS_A,
2140 __mlx4_counter_free(dev, idx);
2142 EXPORT_SYMBOL_GPL(mlx4_counter_free);
2144 static int mlx4_setup_hca(struct mlx4_dev *dev)
2146 struct mlx4_priv *priv = mlx4_priv(dev);
2149 __be32 ib_port_default_caps;
2151 err = mlx4_init_uar_table(dev);
2153 mlx4_err(dev, "Failed to initialize user access region table, aborting\n");
2157 err = mlx4_uar_alloc(dev, &priv->driver_uar);
2159 mlx4_err(dev, "Failed to allocate driver access region, aborting\n");
2160 goto err_uar_table_free;
2163 priv->kar = ioremap((phys_addr_t) priv->driver_uar.pfn << PAGE_SHIFT, PAGE_SIZE);
2165 mlx4_err(dev, "Couldn't map kernel access region, aborting\n");
2170 err = mlx4_init_pd_table(dev);
2172 mlx4_err(dev, "Failed to initialize protection domain table, aborting\n");
2176 err = mlx4_init_xrcd_table(dev);
2178 mlx4_err(dev, "Failed to initialize reliable connection domain table, aborting\n");
2179 goto err_pd_table_free;
2182 err = mlx4_init_mr_table(dev);
2184 mlx4_err(dev, "Failed to initialize memory region table, aborting\n");
2185 goto err_xrcd_table_free;
2188 if (!mlx4_is_slave(dev)) {
2189 err = mlx4_init_mcg_table(dev);
2191 mlx4_err(dev, "Failed to initialize multicast group table, aborting\n");
2192 goto err_mr_table_free;
2194 err = mlx4_config_mad_demux(dev);
2196 mlx4_err(dev, "Failed in config_mad_demux, aborting\n");
2197 goto err_mcg_table_free;
2201 err = mlx4_init_eq_table(dev);
2203 mlx4_err(dev, "Failed to initialize event queue table, aborting\n");
2204 goto err_mcg_table_free;
2207 err = mlx4_cmd_use_events(dev);
2209 mlx4_err(dev, "Failed to switch to event-driven firmware commands, aborting\n");
2210 goto err_eq_table_free;
2213 err = mlx4_NOP(dev);
2215 if (dev->flags & MLX4_FLAG_MSI_X) {
2216 mlx4_warn(dev, "NOP command failed to generate MSI-X interrupt IRQ %d)\n",
2217 priv->eq_table.eq[dev->caps.num_comp_vectors].irq);
2218 mlx4_warn(dev, "Trying again without MSI-X\n");
2220 mlx4_err(dev, "NOP command failed to generate interrupt (IRQ %d), aborting\n",
2221 priv->eq_table.eq[dev->caps.num_comp_vectors].irq);
2222 mlx4_err(dev, "BIOS or ACPI interrupt routing problem?\n");
2228 mlx4_dbg(dev, "NOP command IRQ test passed\n");
2230 err = mlx4_init_cq_table(dev);
2232 mlx4_err(dev, "Failed to initialize completion queue table, aborting\n");
2236 err = mlx4_init_srq_table(dev);
2238 mlx4_err(dev, "Failed to initialize shared receive queue table, aborting\n");
2239 goto err_cq_table_free;
2242 err = mlx4_init_qp_table(dev);
2244 mlx4_err(dev, "Failed to initialize queue pair table, aborting\n");
2245 goto err_srq_table_free;
2248 err = mlx4_init_counters_table(dev);
2249 if (err && err != -ENOENT) {
2250 mlx4_err(dev, "Failed to initialize counters table, aborting\n");
2251 goto err_qp_table_free;
2254 if (!mlx4_is_slave(dev)) {
2255 for (port = 1; port <= dev->caps.num_ports; port++) {
2256 ib_port_default_caps = 0;
2257 err = mlx4_get_port_ib_caps(dev, port,
2258 &ib_port_default_caps);
2260 mlx4_warn(dev, "failed to get port %d default ib capabilities (%d). Continuing with caps = 0\n",
2262 dev->caps.ib_port_def_cap[port] = ib_port_default_caps;
2264 /* initialize per-slave default ib port capabilities */
2265 if (mlx4_is_master(dev)) {
2267 for (i = 0; i < dev->num_slaves; i++) {
2268 if (i == mlx4_master_func_num(dev))
2270 priv->mfunc.master.slave_state[i].ib_cap_mask[port] =
2271 ib_port_default_caps;
2275 if (mlx4_is_mfunc(dev))
2276 dev->caps.port_ib_mtu[port] = IB_MTU_2048;
2278 dev->caps.port_ib_mtu[port] = IB_MTU_4096;
2280 err = mlx4_SET_PORT(dev, port, mlx4_is_master(dev) ?
2281 dev->caps.pkey_table_len[port] : -1);
2283 mlx4_err(dev, "Failed to set port %d, aborting\n",
2285 goto err_counters_table_free;
2292 err_counters_table_free:
2293 mlx4_cleanup_counters_table(dev);
2296 mlx4_cleanup_qp_table(dev);
2299 mlx4_cleanup_srq_table(dev);
2302 mlx4_cleanup_cq_table(dev);
2305 mlx4_cmd_use_polling(dev);
2308 mlx4_cleanup_eq_table(dev);
2311 if (!mlx4_is_slave(dev))
2312 mlx4_cleanup_mcg_table(dev);
2315 mlx4_cleanup_mr_table(dev);
2317 err_xrcd_table_free:
2318 mlx4_cleanup_xrcd_table(dev);
2321 mlx4_cleanup_pd_table(dev);
2327 mlx4_uar_free(dev, &priv->driver_uar);
2330 mlx4_cleanup_uar_table(dev);
2334 static void mlx4_enable_msi_x(struct mlx4_dev *dev)
2336 struct mlx4_priv *priv = mlx4_priv(dev);
2337 struct msix_entry *entries;
2341 int nreq = dev->caps.num_ports * num_online_cpus() + MSIX_LEGACY_SZ;
2343 nreq = min_t(int, dev->caps.num_eqs - dev->caps.reserved_eqs,
2346 entries = kcalloc(nreq, sizeof *entries, GFP_KERNEL);
2350 for (i = 0; i < nreq; ++i)
2351 entries[i].entry = i;
2353 nreq = pci_enable_msix_range(dev->persist->pdev, entries, 2,
2359 } else if (nreq < MSIX_LEGACY_SZ +
2360 dev->caps.num_ports * MIN_MSIX_P_PORT) {
2361 /*Working in legacy mode , all EQ's shared*/
2362 dev->caps.comp_pool = 0;
2363 dev->caps.num_comp_vectors = nreq - 1;
2365 dev->caps.comp_pool = nreq - MSIX_LEGACY_SZ;
2366 dev->caps.num_comp_vectors = MSIX_LEGACY_SZ - 1;
2368 for (i = 0; i < nreq; ++i)
2369 priv->eq_table.eq[i].irq = entries[i].vector;
2371 dev->flags |= MLX4_FLAG_MSI_X;
2378 dev->caps.num_comp_vectors = 1;
2379 dev->caps.comp_pool = 0;
2381 for (i = 0; i < 2; ++i)
2382 priv->eq_table.eq[i].irq = dev->persist->pdev->irq;
2385 static int mlx4_init_port_info(struct mlx4_dev *dev, int port)
2387 struct mlx4_port_info *info = &mlx4_priv(dev)->port[port];
2392 if (!mlx4_is_slave(dev)) {
2393 mlx4_init_mac_table(dev, &info->mac_table);
2394 mlx4_init_vlan_table(dev, &info->vlan_table);
2395 mlx4_init_roce_gid_table(dev, &info->gid_table);
2396 info->base_qpn = mlx4_get_base_qpn(dev, port);
2399 sprintf(info->dev_name, "mlx4_port%d", port);
2400 info->port_attr.attr.name = info->dev_name;
2401 if (mlx4_is_mfunc(dev))
2402 info->port_attr.attr.mode = S_IRUGO;
2404 info->port_attr.attr.mode = S_IRUGO | S_IWUSR;
2405 info->port_attr.store = set_port_type;
2407 info->port_attr.show = show_port_type;
2408 sysfs_attr_init(&info->port_attr.attr);
2410 err = device_create_file(&dev->persist->pdev->dev, &info->port_attr);
2412 mlx4_err(dev, "Failed to create file for port %d\n", port);
2416 sprintf(info->dev_mtu_name, "mlx4_port%d_mtu", port);
2417 info->port_mtu_attr.attr.name = info->dev_mtu_name;
2418 if (mlx4_is_mfunc(dev))
2419 info->port_mtu_attr.attr.mode = S_IRUGO;
2421 info->port_mtu_attr.attr.mode = S_IRUGO | S_IWUSR;
2422 info->port_mtu_attr.store = set_port_ib_mtu;
2424 info->port_mtu_attr.show = show_port_ib_mtu;
2425 sysfs_attr_init(&info->port_mtu_attr.attr);
2427 err = device_create_file(&dev->persist->pdev->dev,
2428 &info->port_mtu_attr);
2430 mlx4_err(dev, "Failed to create mtu file for port %d\n", port);
2431 device_remove_file(&info->dev->persist->pdev->dev,
2439 static void mlx4_cleanup_port_info(struct mlx4_port_info *info)
2444 device_remove_file(&info->dev->persist->pdev->dev, &info->port_attr);
2445 device_remove_file(&info->dev->persist->pdev->dev,
2446 &info->port_mtu_attr);
2449 static int mlx4_init_steering(struct mlx4_dev *dev)
2451 struct mlx4_priv *priv = mlx4_priv(dev);
2452 int num_entries = dev->caps.num_ports;
2455 priv->steer = kzalloc(sizeof(struct mlx4_steer) * num_entries, GFP_KERNEL);
2459 for (i = 0; i < num_entries; i++)
2460 for (j = 0; j < MLX4_NUM_STEERS; j++) {
2461 INIT_LIST_HEAD(&priv->steer[i].promisc_qps[j]);
2462 INIT_LIST_HEAD(&priv->steer[i].steer_entries[j]);
2467 static void mlx4_clear_steering(struct mlx4_dev *dev)
2469 struct mlx4_priv *priv = mlx4_priv(dev);
2470 struct mlx4_steer_index *entry, *tmp_entry;
2471 struct mlx4_promisc_qp *pqp, *tmp_pqp;
2472 int num_entries = dev->caps.num_ports;
2475 for (i = 0; i < num_entries; i++) {
2476 for (j = 0; j < MLX4_NUM_STEERS; j++) {
2477 list_for_each_entry_safe(pqp, tmp_pqp,
2478 &priv->steer[i].promisc_qps[j],
2480 list_del(&pqp->list);
2483 list_for_each_entry_safe(entry, tmp_entry,
2484 &priv->steer[i].steer_entries[j],
2486 list_del(&entry->list);
2487 list_for_each_entry_safe(pqp, tmp_pqp,
2490 list_del(&pqp->list);
2500 static int extended_func_num(struct pci_dev *pdev)
2502 return PCI_SLOT(pdev->devfn) * 8 + PCI_FUNC(pdev->devfn);
2505 #define MLX4_OWNER_BASE 0x8069c
2506 #define MLX4_OWNER_SIZE 4
2508 static int mlx4_get_ownership(struct mlx4_dev *dev)
2510 void __iomem *owner;
2513 if (pci_channel_offline(dev->persist->pdev))
2516 owner = ioremap(pci_resource_start(dev->persist->pdev, 0) +
2520 mlx4_err(dev, "Failed to obtain ownership bit\n");
2529 static void mlx4_free_ownership(struct mlx4_dev *dev)
2531 void __iomem *owner;
2533 if (pci_channel_offline(dev->persist->pdev))
2536 owner = ioremap(pci_resource_start(dev->persist->pdev, 0) +
2540 mlx4_err(dev, "Failed to obtain ownership bit\n");
2548 #define SRIOV_VALID_STATE(flags) (!!((flags) & MLX4_FLAG_SRIOV) ==\
2549 !!((flags) & MLX4_FLAG_MASTER))
2551 static u64 mlx4_enable_sriov(struct mlx4_dev *dev, struct pci_dev *pdev,
2552 u8 total_vfs, int existing_vfs, int reset_flow)
2554 u64 dev_flags = dev->flags;
2558 dev->dev_vfs = kcalloc(total_vfs, sizeof(*dev->dev_vfs),
2565 atomic_inc(&pf_loading);
2566 if (dev->flags & MLX4_FLAG_SRIOV) {
2567 if (existing_vfs != total_vfs) {
2568 mlx4_err(dev, "SR-IOV was already enabled, but with num_vfs (%d) different than requested (%d)\n",
2569 existing_vfs, total_vfs);
2570 total_vfs = existing_vfs;
2574 dev->dev_vfs = kzalloc(total_vfs * sizeof(*dev->dev_vfs), GFP_KERNEL);
2575 if (NULL == dev->dev_vfs) {
2576 mlx4_err(dev, "Failed to allocate memory for VFs\n");
2580 if (!(dev->flags & MLX4_FLAG_SRIOV)) {
2581 mlx4_warn(dev, "Enabling SR-IOV with %d VFs\n", total_vfs);
2582 err = pci_enable_sriov(pdev, total_vfs);
2585 mlx4_err(dev, "Failed to enable SR-IOV, continuing without SR-IOV (err = %d)\n",
2589 mlx4_warn(dev, "Running in master mode\n");
2590 dev_flags |= MLX4_FLAG_SRIOV |
2592 dev_flags &= ~MLX4_FLAG_SLAVE;
2593 dev->persist->num_vfs = total_vfs;
2598 atomic_dec(&pf_loading);
2600 dev->persist->num_vfs = 0;
2601 kfree(dev->dev_vfs);
2602 return dev_flags & ~MLX4_FLAG_MASTER;
2606 MLX4_DEV_CAP_CHECK_NUM_VFS_ABOVE_64 = -1,
2609 static int mlx4_check_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap,
2612 int requested_vfs = nvfs[0] + nvfs[1] + nvfs[2];
2613 /* Checking for 64 VFs as a limitation of CX2 */
2614 if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_80_VFS) &&
2615 requested_vfs >= 64) {
2616 mlx4_err(dev, "Requested %d VFs, but FW does not support more than 64\n",
2618 return MLX4_DEV_CAP_CHECK_NUM_VFS_ABOVE_64;
2623 static int mlx4_load_one(struct pci_dev *pdev, int pci_dev_data,
2624 int total_vfs, int *nvfs, struct mlx4_priv *priv,
2627 struct mlx4_dev *dev;
2632 struct mlx4_dev_cap *dev_cap = NULL;
2633 int existing_vfs = 0;
2637 INIT_LIST_HEAD(&priv->ctx_list);
2638 spin_lock_init(&priv->ctx_lock);
2640 mutex_init(&priv->port_mutex);
2642 INIT_LIST_HEAD(&priv->pgdir_list);
2643 mutex_init(&priv->pgdir_mutex);
2645 INIT_LIST_HEAD(&priv->bf_list);
2646 mutex_init(&priv->bf_mutex);
2648 dev->rev_id = pdev->revision;
2649 dev->numa_node = dev_to_node(&pdev->dev);
2651 /* Detect if this device is a virtual function */
2652 if (pci_dev_data & MLX4_PCI_DEV_IS_VF) {
2653 mlx4_warn(dev, "Detected virtual function - running in slave mode\n");
2654 dev->flags |= MLX4_FLAG_SLAVE;
2656 /* We reset the device and enable SRIOV only for physical
2657 * devices. Try to claim ownership on the device;
2658 * if already taken, skip -- do not allow multiple PFs */
2659 err = mlx4_get_ownership(dev);
2664 mlx4_warn(dev, "Multiple PFs not yet supported - Skipping PF\n");
2669 atomic_set(&priv->opreq_count, 0);
2670 INIT_WORK(&priv->opreq_task, mlx4_opreq_action);
2673 * Now reset the HCA before we touch the PCI capabilities or
2674 * attempt a firmware command, since a boot ROM may have left
2675 * the HCA in an undefined state.
2677 err = mlx4_reset(dev);
2679 mlx4_err(dev, "Failed to reset HCA, aborting\n");
2684 dev->flags = MLX4_FLAG_MASTER;
2685 existing_vfs = pci_num_vf(pdev);
2687 dev->flags |= MLX4_FLAG_SRIOV;
2688 dev->persist->num_vfs = total_vfs;
2692 /* on load remove any previous indication of internal error,
2695 dev->persist->state = MLX4_DEVICE_STATE_UP;
2698 err = mlx4_cmd_init(dev);
2700 mlx4_err(dev, "Failed to init command interface, aborting\n");
2704 /* In slave functions, the communication channel must be initialized
2705 * before posting commands. Also, init num_slaves before calling
2707 if (mlx4_is_mfunc(dev)) {
2708 if (mlx4_is_master(dev)) {
2709 dev->num_slaves = MLX4_MAX_NUM_SLAVES;
2712 dev->num_slaves = 0;
2713 err = mlx4_multi_func_init(dev);
2715 mlx4_err(dev, "Failed to init slave mfunc interface, aborting\n");
2721 err = mlx4_init_fw(dev);
2723 mlx4_err(dev, "Failed to init fw, aborting.\n");
2727 if (mlx4_is_master(dev)) {
2728 /* when we hit the goto slave_start below, dev_cap already initialized */
2730 dev_cap = kzalloc(sizeof(*dev_cap), GFP_KERNEL);
2737 err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
2739 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
2743 if (mlx4_check_dev_cap(dev, dev_cap, nvfs))
2746 if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS)) {
2747 u64 dev_flags = mlx4_enable_sriov(dev, pdev,
2752 mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
2753 dev->flags = dev_flags;
2754 if (!SRIOV_VALID_STATE(dev->flags)) {
2755 mlx4_err(dev, "Invalid SRIOV state\n");
2758 err = mlx4_reset(dev);
2760 mlx4_err(dev, "Failed to reset HCA, aborting.\n");
2766 /* Legacy mode FW requires SRIOV to be enabled before
2767 * doing QUERY_DEV_CAP, since max_eq's value is different if
2770 memset(dev_cap, 0, sizeof(*dev_cap));
2771 err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
2773 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
2777 if (mlx4_check_dev_cap(dev, dev_cap, nvfs))
2782 err = mlx4_init_hca(dev);
2784 if (err == -EACCES) {
2785 /* Not primary Physical function
2786 * Running in slave mode */
2787 mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
2788 /* We're not a PF */
2789 if (dev->flags & MLX4_FLAG_SRIOV) {
2791 pci_disable_sriov(pdev);
2792 if (mlx4_is_master(dev) && !reset_flow)
2793 atomic_dec(&pf_loading);
2794 dev->flags &= ~MLX4_FLAG_SRIOV;
2796 if (!mlx4_is_slave(dev))
2797 mlx4_free_ownership(dev);
2798 dev->flags |= MLX4_FLAG_SLAVE;
2799 dev->flags &= ~MLX4_FLAG_MASTER;
2805 if (mlx4_is_master(dev) && (dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS)) {
2806 u64 dev_flags = mlx4_enable_sriov(dev, pdev, total_vfs,
2807 existing_vfs, reset_flow);
2809 if ((dev->flags ^ dev_flags) & (MLX4_FLAG_MASTER | MLX4_FLAG_SLAVE)) {
2810 mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_VHCR);
2811 dev->flags = dev_flags;
2812 err = mlx4_cmd_init(dev);
2814 /* Only VHCR is cleaned up, so could still
2817 mlx4_err(dev, "Failed to init VHCR command interface, aborting\n");
2821 dev->flags = dev_flags;
2824 if (!SRIOV_VALID_STATE(dev->flags)) {
2825 mlx4_err(dev, "Invalid SRIOV state\n");
2830 /* check if the device is functioning at its maximum possible speed.
2831 * No return code for this call, just warn the user in case of PCI
2832 * express device capabilities are under-satisfied by the bus.
2834 if (!mlx4_is_slave(dev))
2835 mlx4_check_pcie_caps(dev);
2837 /* In master functions, the communication channel must be initialized
2838 * after obtaining its address from fw */
2839 if (mlx4_is_master(dev)) {
2842 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2846 (num_vfs_argc > 1 || probe_vfs_argc > 1)) {
2848 "Invalid syntax of num_vfs/probe_vfs with IB port - single port VFs syntax is only supported when all ports are configured as ethernet\n");
2852 if (dev->caps.num_ports < 2 &&
2856 "Error: Trying to configure VFs on port 2, but HCA has only %d physical ports\n",
2857 dev->caps.num_ports);
2860 memcpy(dev->persist->nvfs, nvfs, sizeof(dev->persist->nvfs));
2863 i < sizeof(dev->persist->nvfs)/
2864 sizeof(dev->persist->nvfs[0]); i++) {
2867 for (j = 0; j < dev->persist->nvfs[i]; ++sum, ++j) {
2868 dev->dev_vfs[sum].min_port = i < 2 ? i + 1 : 1;
2869 dev->dev_vfs[sum].n_ports = i < 2 ? 1 :
2870 dev->caps.num_ports;
2874 /* In master functions, the communication channel
2875 * must be initialized after obtaining its address from fw
2877 err = mlx4_multi_func_init(dev);
2879 mlx4_err(dev, "Failed to init master mfunc interface, aborting.\n");
2884 err = mlx4_alloc_eq_table(dev);
2886 goto err_master_mfunc;
2888 priv->msix_ctl.pool_bm = 0;
2889 mutex_init(&priv->msix_ctl.pool_lock);
2891 mlx4_enable_msi_x(dev);
2892 if ((mlx4_is_mfunc(dev)) &&
2893 !(dev->flags & MLX4_FLAG_MSI_X)) {
2895 mlx4_err(dev, "INTx is not supported in multi-function mode, aborting\n");
2899 if (!mlx4_is_slave(dev)) {
2900 err = mlx4_init_steering(dev);
2902 goto err_disable_msix;
2905 err = mlx4_setup_hca(dev);
2906 if (err == -EBUSY && (dev->flags & MLX4_FLAG_MSI_X) &&
2907 !mlx4_is_mfunc(dev)) {
2908 dev->flags &= ~MLX4_FLAG_MSI_X;
2909 dev->caps.num_comp_vectors = 1;
2910 dev->caps.comp_pool = 0;
2911 pci_disable_msix(pdev);
2912 err = mlx4_setup_hca(dev);
2918 mlx4_init_quotas(dev);
2919 /* When PF resources are ready arm its comm channel to enable
2922 if (mlx4_is_master(dev)) {
2923 err = mlx4_ARM_COMM_CHANNEL(dev);
2925 mlx4_err(dev, " Failed to arm comm channel eq: %x\n",
2931 for (port = 1; port <= dev->caps.num_ports; port++) {
2932 err = mlx4_init_port_info(dev, port);
2937 err = mlx4_register_device(dev);
2941 mlx4_request_modules(dev);
2943 mlx4_sense_init(dev);
2944 mlx4_start_sense(dev);
2948 if (mlx4_is_master(dev) && dev->persist->num_vfs && !reset_flow)
2949 atomic_dec(&pf_loading);
2955 for (--port; port >= 1; --port)
2956 mlx4_cleanup_port_info(&priv->port[port]);
2958 mlx4_cleanup_counters_table(dev);
2959 mlx4_cleanup_qp_table(dev);
2960 mlx4_cleanup_srq_table(dev);
2961 mlx4_cleanup_cq_table(dev);
2962 mlx4_cmd_use_polling(dev);
2963 mlx4_cleanup_eq_table(dev);
2964 mlx4_cleanup_mcg_table(dev);
2965 mlx4_cleanup_mr_table(dev);
2966 mlx4_cleanup_xrcd_table(dev);
2967 mlx4_cleanup_pd_table(dev);
2968 mlx4_cleanup_uar_table(dev);
2971 if (!mlx4_is_slave(dev))
2972 mlx4_clear_steering(dev);
2975 if (dev->flags & MLX4_FLAG_MSI_X)
2976 pci_disable_msix(pdev);
2979 mlx4_free_eq_table(dev);
2982 if (mlx4_is_master(dev)) {
2983 mlx4_free_resource_tracker(dev, RES_TR_FREE_STRUCTS_ONLY);
2984 mlx4_multi_func_cleanup(dev);
2987 if (mlx4_is_slave(dev)) {
2988 kfree(dev->caps.qp0_qkey);
2989 kfree(dev->caps.qp0_tunnel);
2990 kfree(dev->caps.qp0_proxy);
2991 kfree(dev->caps.qp1_tunnel);
2992 kfree(dev->caps.qp1_proxy);
2996 mlx4_close_hca(dev);
3002 if (mlx4_is_slave(dev))
3003 mlx4_multi_func_cleanup(dev);
3006 mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3009 if (dev->flags & MLX4_FLAG_SRIOV && !existing_vfs) {
3010 pci_disable_sriov(pdev);
3011 dev->flags &= ~MLX4_FLAG_SRIOV;
3014 if (mlx4_is_master(dev) && dev->persist->num_vfs && !reset_flow)
3015 atomic_dec(&pf_loading);
3017 kfree(priv->dev.dev_vfs);
3019 if (!mlx4_is_slave(dev))
3020 mlx4_free_ownership(dev);
3026 static int __mlx4_init_one(struct pci_dev *pdev, int pci_dev_data,
3027 struct mlx4_priv *priv)
3030 int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
3031 int prb_vf[MLX4_MAX_PORTS + 1] = {0, 0, 0};
3032 const int param_map[MLX4_MAX_PORTS + 1][MLX4_MAX_PORTS + 1] = {
3033 {2, 0, 0}, {0, 1, 2}, {0, 1, 2} };
3034 unsigned total_vfs = 0;
3037 pr_info(DRV_NAME ": Initializing %s\n", pci_name(pdev));
3039 err = pci_enable_device(pdev);
3041 dev_err(&pdev->dev, "Cannot enable PCI device, aborting\n");
3045 /* Due to requirement that all VFs and the PF are *guaranteed* 2 MACS
3046 * per port, we must limit the number of VFs to 63 (since their are
3049 for (i = 0; i < sizeof(nvfs)/sizeof(nvfs[0]) && i < num_vfs_argc;
3050 total_vfs += nvfs[param_map[num_vfs_argc - 1][i]], i++) {
3051 nvfs[param_map[num_vfs_argc - 1][i]] = num_vfs[i];
3053 dev_err(&pdev->dev, "num_vfs module parameter cannot be negative\n");
3055 goto err_disable_pdev;
3058 for (i = 0; i < sizeof(prb_vf)/sizeof(prb_vf[0]) && i < probe_vfs_argc;
3060 prb_vf[param_map[probe_vfs_argc - 1][i]] = probe_vf[i];
3061 if (prb_vf[i] < 0 || prb_vf[i] > nvfs[i]) {
3062 dev_err(&pdev->dev, "probe_vf module parameter cannot be negative or greater than num_vfs\n");
3064 goto err_disable_pdev;
3067 if (total_vfs >= MLX4_MAX_NUM_VF) {
3069 "Requested more VF's (%d) than allowed (%d)\n",
3070 total_vfs, MLX4_MAX_NUM_VF - 1);
3072 goto err_disable_pdev;
3075 for (i = 0; i < MLX4_MAX_PORTS; i++) {
3076 if (nvfs[i] + nvfs[2] >= MLX4_MAX_NUM_VF_P_PORT) {
3078 "Requested more VF's (%d) for port (%d) than allowed (%d)\n",
3079 nvfs[i] + nvfs[2], i + 1,
3080 MLX4_MAX_NUM_VF_P_PORT - 1);
3082 goto err_disable_pdev;
3086 /* Check for BARs. */
3087 if (!(pci_dev_data & MLX4_PCI_DEV_IS_VF) &&
3088 !(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) {
3089 dev_err(&pdev->dev, "Missing DCS, aborting (driver_data: 0x%x, pci_resource_flags(pdev, 0):0x%lx)\n",
3090 pci_dev_data, pci_resource_flags(pdev, 0));
3092 goto err_disable_pdev;
3094 if (!(pci_resource_flags(pdev, 2) & IORESOURCE_MEM)) {
3095 dev_err(&pdev->dev, "Missing UAR, aborting\n");
3097 goto err_disable_pdev;
3100 err = pci_request_regions(pdev, DRV_NAME);
3102 dev_err(&pdev->dev, "Couldn't get PCI resources, aborting\n");
3103 goto err_disable_pdev;
3106 pci_set_master(pdev);
3108 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
3110 dev_warn(&pdev->dev, "Warning: couldn't set 64-bit PCI DMA mask\n");
3111 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
3113 dev_err(&pdev->dev, "Can't set PCI DMA mask, aborting\n");
3114 goto err_release_regions;
3117 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
3119 dev_warn(&pdev->dev, "Warning: couldn't set 64-bit consistent PCI DMA mask\n");
3120 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
3122 dev_err(&pdev->dev, "Can't set consistent PCI DMA mask, aborting\n");
3123 goto err_release_regions;
3127 /* Allow large DMA segments, up to the firmware limit of 1 GB */
3128 dma_set_max_seg_size(&pdev->dev, 1024 * 1024 * 1024);
3129 /* Detect if this device is a virtual function */
3130 if (pci_dev_data & MLX4_PCI_DEV_IS_VF) {
3131 /* When acting as pf, we normally skip vfs unless explicitly
3132 * requested to probe them.
3135 unsigned vfs_offset = 0;
3137 for (i = 0; i < sizeof(nvfs)/sizeof(nvfs[0]) &&
3138 vfs_offset + nvfs[i] < extended_func_num(pdev);
3139 vfs_offset += nvfs[i], i++)
3141 if (i == sizeof(nvfs)/sizeof(nvfs[0])) {
3143 goto err_release_regions;
3145 if ((extended_func_num(pdev) - vfs_offset)
3147 dev_warn(&pdev->dev, "Skipping virtual function:%d\n",
3148 extended_func_num(pdev));
3150 goto err_release_regions;
3155 err = mlx4_catas_init(&priv->dev);
3157 goto err_release_regions;
3159 err = mlx4_load_one(pdev, pci_dev_data, total_vfs, nvfs, priv, 0);
3166 mlx4_catas_end(&priv->dev);
3168 err_release_regions:
3169 pci_release_regions(pdev);
3172 pci_disable_device(pdev);
3173 pci_set_drvdata(pdev, NULL);
3177 static int mlx4_init_one(struct pci_dev *pdev, const struct pci_device_id *id)
3179 struct mlx4_priv *priv;
3180 struct mlx4_dev *dev;
3183 printk_once(KERN_INFO "%s", mlx4_version);
3185 priv = kzalloc(sizeof(*priv), GFP_KERNEL);
3190 dev->persist = kzalloc(sizeof(*dev->persist), GFP_KERNEL);
3191 if (!dev->persist) {
3195 dev->persist->pdev = pdev;
3196 dev->persist->dev = dev;
3197 pci_set_drvdata(pdev, dev->persist);
3198 priv->pci_dev_data = id->driver_data;
3199 mutex_init(&dev->persist->device_state_mutex);
3200 mutex_init(&dev->persist->interface_state_mutex);
3202 ret = __mlx4_init_one(pdev, id->driver_data, priv);
3204 kfree(dev->persist);
3207 pci_save_state(pdev);
3213 static void mlx4_clean_dev(struct mlx4_dev *dev)
3215 struct mlx4_dev_persistent *persist = dev->persist;
3216 struct mlx4_priv *priv = mlx4_priv(dev);
3217 unsigned long flags = (dev->flags & RESET_PERSIST_MASK_FLAGS);
3219 memset(priv, 0, sizeof(*priv));
3220 priv->dev.persist = persist;
3221 priv->dev.flags = flags;
3224 static void mlx4_unload_one(struct pci_dev *pdev)
3226 struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
3227 struct mlx4_dev *dev = persist->dev;
3228 struct mlx4_priv *priv = mlx4_priv(dev);
3235 /* saving current ports type for further use */
3236 for (i = 0; i < dev->caps.num_ports; i++) {
3237 dev->persist->curr_port_type[i] = dev->caps.port_type[i + 1];
3238 dev->persist->curr_port_poss_type[i] = dev->caps.
3239 possible_type[i + 1];
3242 pci_dev_data = priv->pci_dev_data;
3244 mlx4_stop_sense(dev);
3245 mlx4_unregister_device(dev);
3247 for (p = 1; p <= dev->caps.num_ports; p++) {
3248 mlx4_cleanup_port_info(&priv->port[p]);
3249 mlx4_CLOSE_PORT(dev, p);
3252 if (mlx4_is_master(dev))
3253 mlx4_free_resource_tracker(dev,
3254 RES_TR_FREE_SLAVES_ONLY);
3256 mlx4_cleanup_counters_table(dev);
3257 mlx4_cleanup_qp_table(dev);
3258 mlx4_cleanup_srq_table(dev);
3259 mlx4_cleanup_cq_table(dev);
3260 mlx4_cmd_use_polling(dev);
3261 mlx4_cleanup_eq_table(dev);
3262 mlx4_cleanup_mcg_table(dev);
3263 mlx4_cleanup_mr_table(dev);
3264 mlx4_cleanup_xrcd_table(dev);
3265 mlx4_cleanup_pd_table(dev);
3267 if (mlx4_is_master(dev))
3268 mlx4_free_resource_tracker(dev,
3269 RES_TR_FREE_STRUCTS_ONLY);
3272 mlx4_uar_free(dev, &priv->driver_uar);
3273 mlx4_cleanup_uar_table(dev);
3274 if (!mlx4_is_slave(dev))
3275 mlx4_clear_steering(dev);
3276 mlx4_free_eq_table(dev);
3277 if (mlx4_is_master(dev))
3278 mlx4_multi_func_cleanup(dev);
3279 mlx4_close_hca(dev);
3281 if (mlx4_is_slave(dev))
3282 mlx4_multi_func_cleanup(dev);
3283 mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3285 if (dev->flags & MLX4_FLAG_MSI_X)
3286 pci_disable_msix(pdev);
3288 if (!mlx4_is_slave(dev))
3289 mlx4_free_ownership(dev);
3291 kfree(dev->caps.qp0_qkey);
3292 kfree(dev->caps.qp0_tunnel);
3293 kfree(dev->caps.qp0_proxy);
3294 kfree(dev->caps.qp1_tunnel);
3295 kfree(dev->caps.qp1_proxy);
3296 kfree(dev->dev_vfs);
3298 mlx4_clean_dev(dev);
3299 priv->pci_dev_data = pci_dev_data;
3303 static void mlx4_remove_one(struct pci_dev *pdev)
3305 struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
3306 struct mlx4_dev *dev = persist->dev;
3307 struct mlx4_priv *priv = mlx4_priv(dev);
3310 mutex_lock(&persist->interface_state_mutex);
3311 persist->interface_state |= MLX4_INTERFACE_STATE_DELETION;
3312 mutex_unlock(&persist->interface_state_mutex);
3314 /* Disabling SR-IOV is not allowed while there are active vf's */
3315 if (mlx4_is_master(dev) && dev->flags & MLX4_FLAG_SRIOV) {
3316 active_vfs = mlx4_how_many_lives_vf(dev);
3318 pr_warn("Removing PF when there are active VF's !!\n");
3319 pr_warn("Will not disable SR-IOV.\n");
3323 /* device marked to be under deletion running now without the lock
3324 * letting other tasks to be terminated
3326 if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
3327 mlx4_unload_one(pdev);
3329 mlx4_info(dev, "%s: interface is down\n", __func__);
3330 mlx4_catas_end(dev);
3331 if (dev->flags & MLX4_FLAG_SRIOV && !active_vfs) {
3332 mlx4_warn(dev, "Disabling SR-IOV\n");
3333 pci_disable_sriov(pdev);
3336 pci_release_regions(pdev);
3337 pci_disable_device(pdev);
3338 kfree(dev->persist);
3340 pci_set_drvdata(pdev, NULL);
3343 static int restore_current_port_types(struct mlx4_dev *dev,
3344 enum mlx4_port_type *types,
3345 enum mlx4_port_type *poss_types)
3347 struct mlx4_priv *priv = mlx4_priv(dev);
3350 mlx4_stop_sense(dev);
3352 mutex_lock(&priv->port_mutex);
3353 for (i = 0; i < dev->caps.num_ports; i++)
3354 dev->caps.possible_type[i + 1] = poss_types[i];
3355 err = mlx4_change_port_types(dev, types);
3356 mlx4_start_sense(dev);
3357 mutex_unlock(&priv->port_mutex);
3362 int mlx4_restart_one(struct pci_dev *pdev)
3364 struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
3365 struct mlx4_dev *dev = persist->dev;
3366 struct mlx4_priv *priv = mlx4_priv(dev);
3367 int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
3368 int pci_dev_data, err, total_vfs;
3370 pci_dev_data = priv->pci_dev_data;
3371 total_vfs = dev->persist->num_vfs;
3372 memcpy(nvfs, dev->persist->nvfs, sizeof(dev->persist->nvfs));
3374 mlx4_unload_one(pdev);
3375 err = mlx4_load_one(pdev, pci_dev_data, total_vfs, nvfs, priv, 1);
3377 mlx4_err(dev, "%s: ERROR: mlx4_load_one failed, pci_name=%s, err=%d\n",
3378 __func__, pci_name(pdev), err);
3382 err = restore_current_port_types(dev, dev->persist->curr_port_type,
3383 dev->persist->curr_port_poss_type);
3385 mlx4_err(dev, "could not restore original port types (%d)\n",
3391 static const struct pci_device_id mlx4_pci_table[] = {
3392 /* MT25408 "Hermon" SDR */
3393 { PCI_VDEVICE(MELLANOX, 0x6340), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3394 /* MT25408 "Hermon" DDR */
3395 { PCI_VDEVICE(MELLANOX, 0x634a), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3396 /* MT25408 "Hermon" QDR */
3397 { PCI_VDEVICE(MELLANOX, 0x6354), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3398 /* MT25408 "Hermon" DDR PCIe gen2 */
3399 { PCI_VDEVICE(MELLANOX, 0x6732), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3400 /* MT25408 "Hermon" QDR PCIe gen2 */
3401 { PCI_VDEVICE(MELLANOX, 0x673c), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3402 /* MT25408 "Hermon" EN 10GigE */
3403 { PCI_VDEVICE(MELLANOX, 0x6368), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3404 /* MT25408 "Hermon" EN 10GigE PCIe gen2 */
3405 { PCI_VDEVICE(MELLANOX, 0x6750), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3406 /* MT25458 ConnectX EN 10GBASE-T 10GigE */
3407 { PCI_VDEVICE(MELLANOX, 0x6372), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3408 /* MT25458 ConnectX EN 10GBASE-T+Gen2 10GigE */
3409 { PCI_VDEVICE(MELLANOX, 0x675a), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3410 /* MT26468 ConnectX EN 10GigE PCIe gen2*/
3411 { PCI_VDEVICE(MELLANOX, 0x6764), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3412 /* MT26438 ConnectX EN 40GigE PCIe gen2 5GT/s */
3413 { PCI_VDEVICE(MELLANOX, 0x6746), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3414 /* MT26478 ConnectX2 40GigE PCIe gen2 */
3415 { PCI_VDEVICE(MELLANOX, 0x676e), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3416 /* MT25400 Family [ConnectX-2 Virtual Function] */
3417 { PCI_VDEVICE(MELLANOX, 0x1002), MLX4_PCI_DEV_IS_VF },
3418 /* MT27500 Family [ConnectX-3] */
3419 { PCI_VDEVICE(MELLANOX, 0x1003), 0 },
3420 /* MT27500 Family [ConnectX-3 Virtual Function] */
3421 { PCI_VDEVICE(MELLANOX, 0x1004), MLX4_PCI_DEV_IS_VF },
3422 { PCI_VDEVICE(MELLANOX, 0x1005), 0 }, /* MT27510 Family */
3423 { PCI_VDEVICE(MELLANOX, 0x1006), 0 }, /* MT27511 Family */
3424 { PCI_VDEVICE(MELLANOX, 0x1007), 0 }, /* MT27520 Family */
3425 { PCI_VDEVICE(MELLANOX, 0x1008), 0 }, /* MT27521 Family */
3426 { PCI_VDEVICE(MELLANOX, 0x1009), 0 }, /* MT27530 Family */
3427 { PCI_VDEVICE(MELLANOX, 0x100a), 0 }, /* MT27531 Family */
3428 { PCI_VDEVICE(MELLANOX, 0x100b), 0 }, /* MT27540 Family */
3429 { PCI_VDEVICE(MELLANOX, 0x100c), 0 }, /* MT27541 Family */
3430 { PCI_VDEVICE(MELLANOX, 0x100d), 0 }, /* MT27550 Family */
3431 { PCI_VDEVICE(MELLANOX, 0x100e), 0 }, /* MT27551 Family */
3432 { PCI_VDEVICE(MELLANOX, 0x100f), 0 }, /* MT27560 Family */
3433 { PCI_VDEVICE(MELLANOX, 0x1010), 0 }, /* MT27561 Family */
3437 MODULE_DEVICE_TABLE(pci, mlx4_pci_table);
3439 static pci_ers_result_t mlx4_pci_err_detected(struct pci_dev *pdev,
3440 pci_channel_state_t state)
3442 struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
3444 mlx4_err(persist->dev, "mlx4_pci_err_detected was called\n");
3445 mlx4_enter_error_state(persist);
3447 mutex_lock(&persist->interface_state_mutex);
3448 if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
3449 mlx4_unload_one(pdev);
3451 mutex_unlock(&persist->interface_state_mutex);
3452 if (state == pci_channel_io_perm_failure)
3453 return PCI_ERS_RESULT_DISCONNECT;
3455 pci_disable_device(pdev);
3456 return PCI_ERS_RESULT_NEED_RESET;
3459 static pci_ers_result_t mlx4_pci_slot_reset(struct pci_dev *pdev)
3461 struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
3462 struct mlx4_dev *dev = persist->dev;
3463 struct mlx4_priv *priv = mlx4_priv(dev);
3465 int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
3468 mlx4_err(dev, "mlx4_pci_slot_reset was called\n");
3469 ret = pci_enable_device(pdev);
3471 mlx4_err(dev, "Can not re-enable device, ret=%d\n", ret);
3472 return PCI_ERS_RESULT_DISCONNECT;
3475 pci_set_master(pdev);
3476 pci_restore_state(pdev);
3477 pci_save_state(pdev);
3479 total_vfs = dev->persist->num_vfs;
3480 memcpy(nvfs, dev->persist->nvfs, sizeof(dev->persist->nvfs));
3482 mutex_lock(&persist->interface_state_mutex);
3483 if (!(persist->interface_state & MLX4_INTERFACE_STATE_UP)) {
3484 ret = mlx4_load_one(pdev, priv->pci_dev_data, total_vfs, nvfs,
3487 mlx4_err(dev, "%s: mlx4_load_one failed, ret=%d\n",
3492 ret = restore_current_port_types(dev, dev->persist->
3493 curr_port_type, dev->persist->
3494 curr_port_poss_type);
3496 mlx4_err(dev, "could not restore original port types (%d)\n", ret);
3499 mutex_unlock(&persist->interface_state_mutex);
3501 return ret ? PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_RECOVERED;
3504 static void mlx4_shutdown(struct pci_dev *pdev)
3506 struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
3508 mlx4_info(persist->dev, "mlx4_shutdown was called\n");
3509 mutex_lock(&persist->interface_state_mutex);
3510 if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
3511 mlx4_unload_one(pdev);
3512 mutex_unlock(&persist->interface_state_mutex);
3515 static const struct pci_error_handlers mlx4_err_handler = {
3516 .error_detected = mlx4_pci_err_detected,
3517 .slot_reset = mlx4_pci_slot_reset,
3520 static struct pci_driver mlx4_driver = {
3522 .id_table = mlx4_pci_table,
3523 .probe = mlx4_init_one,
3524 .shutdown = mlx4_shutdown,
3525 .remove = mlx4_remove_one,
3526 .err_handler = &mlx4_err_handler,
3529 static int __init mlx4_verify_params(void)
3531 if ((log_num_mac < 0) || (log_num_mac > 7)) {
3532 pr_warn("mlx4_core: bad num_mac: %d\n", log_num_mac);
3536 if (log_num_vlan != 0)
3537 pr_warn("mlx4_core: log_num_vlan - obsolete module param, using %d\n",
3538 MLX4_LOG_NUM_VLANS);
3541 pr_warn("mlx4_core: use_prio - obsolete module param, ignored\n");
3543 if ((log_mtts_per_seg < 1) || (log_mtts_per_seg > 7)) {
3544 pr_warn("mlx4_core: bad log_mtts_per_seg: %d\n",
3549 /* Check if module param for ports type has legal combination */
3550 if (port_type_array[0] == false && port_type_array[1] == true) {
3551 pr_warn("Module parameter configuration ETH/IB is not supported. Switching to default configuration IB/IB\n");
3552 port_type_array[0] = true;
3555 if (mlx4_log_num_mgm_entry_size < -7 ||
3556 (mlx4_log_num_mgm_entry_size > 0 &&
3557 (mlx4_log_num_mgm_entry_size < MLX4_MIN_MGM_LOG_ENTRY_SIZE ||
3558 mlx4_log_num_mgm_entry_size > MLX4_MAX_MGM_LOG_ENTRY_SIZE))) {
3559 pr_warn("mlx4_core: mlx4_log_num_mgm_entry_size (%d) not in legal range (-7..0 or %d..%d)\n",
3560 mlx4_log_num_mgm_entry_size,
3561 MLX4_MIN_MGM_LOG_ENTRY_SIZE,
3562 MLX4_MAX_MGM_LOG_ENTRY_SIZE);
3569 static int __init mlx4_init(void)
3573 if (mlx4_verify_params())
3577 mlx4_wq = create_singlethread_workqueue("mlx4");
3581 ret = pci_register_driver(&mlx4_driver);
3583 destroy_workqueue(mlx4_wq);
3584 return ret < 0 ? ret : 0;
3587 static void __exit mlx4_cleanup(void)
3589 pci_unregister_driver(&mlx4_driver);
3590 destroy_workqueue(mlx4_wq);
3593 module_init(mlx4_init);
3594 module_exit(mlx4_cleanup);