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 if (cache_line_size() != 32 && cache_line_size() != 64)
255 mlx4_dbg(dev, "Disabling CQE stride, cacheLine size unsupported\n");
256 dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
257 dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
261 static int _mlx4_dev_port(struct mlx4_dev *dev, int port,
262 struct mlx4_port_cap *port_cap)
264 dev->caps.vl_cap[port] = port_cap->max_vl;
265 dev->caps.ib_mtu_cap[port] = port_cap->ib_mtu;
266 dev->phys_caps.gid_phys_table_len[port] = port_cap->max_gids;
267 dev->phys_caps.pkey_phys_table_len[port] = port_cap->max_pkeys;
268 /* set gid and pkey table operating lengths by default
269 * to non-sriov values
271 dev->caps.gid_table_len[port] = port_cap->max_gids;
272 dev->caps.pkey_table_len[port] = port_cap->max_pkeys;
273 dev->caps.port_width_cap[port] = port_cap->max_port_width;
274 dev->caps.eth_mtu_cap[port] = port_cap->eth_mtu;
275 dev->caps.def_mac[port] = port_cap->def_mac;
276 dev->caps.supported_type[port] = port_cap->supported_port_types;
277 dev->caps.suggested_type[port] = port_cap->suggested_type;
278 dev->caps.default_sense[port] = port_cap->default_sense;
279 dev->caps.trans_type[port] = port_cap->trans_type;
280 dev->caps.vendor_oui[port] = port_cap->vendor_oui;
281 dev->caps.wavelength[port] = port_cap->wavelength;
282 dev->caps.trans_code[port] = port_cap->trans_code;
287 static int mlx4_dev_port(struct mlx4_dev *dev, int port,
288 struct mlx4_port_cap *port_cap)
292 err = mlx4_QUERY_PORT(dev, port, port_cap);
295 mlx4_err(dev, "QUERY_PORT command failed.\n");
300 static inline void mlx4_enable_ignore_fcs(struct mlx4_dev *dev)
302 if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_IGNORE_FCS))
305 if (mlx4_is_mfunc(dev)) {
306 mlx4_dbg(dev, "SRIOV mode - Disabling Ignore FCS");
307 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_IGNORE_FCS;
311 if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_FCS_KEEP)) {
313 "Keep FCS is not supported - Disabling Ignore FCS");
314 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_IGNORE_FCS;
319 #define MLX4_A0_STEERING_TABLE_SIZE 256
320 static int mlx4_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
325 err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
327 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
330 mlx4_dev_cap_dump(dev, dev_cap);
332 if (dev_cap->min_page_sz > PAGE_SIZE) {
333 mlx4_err(dev, "HCA minimum page size of %d bigger than kernel PAGE_SIZE of %ld, aborting\n",
334 dev_cap->min_page_sz, PAGE_SIZE);
337 if (dev_cap->num_ports > MLX4_MAX_PORTS) {
338 mlx4_err(dev, "HCA has %d ports, but we only support %d, aborting\n",
339 dev_cap->num_ports, MLX4_MAX_PORTS);
343 if (dev_cap->uar_size > pci_resource_len(dev->persist->pdev, 2)) {
344 mlx4_err(dev, "HCA reported UAR size of 0x%x bigger than PCI resource 2 size of 0x%llx, aborting\n",
347 pci_resource_len(dev->persist->pdev, 2));
351 dev->caps.num_ports = dev_cap->num_ports;
352 dev->caps.num_sys_eqs = dev_cap->num_sys_eqs;
353 dev->phys_caps.num_phys_eqs = dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS ?
354 dev->caps.num_sys_eqs :
356 for (i = 1; i <= dev->caps.num_ports; ++i) {
357 err = _mlx4_dev_port(dev, i, dev_cap->port_cap + i);
359 mlx4_err(dev, "QUERY_PORT command failed, aborting\n");
364 dev->caps.uar_page_size = PAGE_SIZE;
365 dev->caps.num_uars = dev_cap->uar_size / PAGE_SIZE;
366 dev->caps.local_ca_ack_delay = dev_cap->local_ca_ack_delay;
367 dev->caps.bf_reg_size = dev_cap->bf_reg_size;
368 dev->caps.bf_regs_per_page = dev_cap->bf_regs_per_page;
369 dev->caps.max_sq_sg = dev_cap->max_sq_sg;
370 dev->caps.max_rq_sg = dev_cap->max_rq_sg;
371 dev->caps.max_wqes = dev_cap->max_qp_sz;
372 dev->caps.max_qp_init_rdma = dev_cap->max_requester_per_qp;
373 dev->caps.max_srq_wqes = dev_cap->max_srq_sz;
374 dev->caps.max_srq_sge = dev_cap->max_rq_sg - 1;
375 dev->caps.reserved_srqs = dev_cap->reserved_srqs;
376 dev->caps.max_sq_desc_sz = dev_cap->max_sq_desc_sz;
377 dev->caps.max_rq_desc_sz = dev_cap->max_rq_desc_sz;
379 * Subtract 1 from the limit because we need to allocate a
380 * spare CQE so the HCA HW can tell the difference between an
381 * empty CQ and a full CQ.
383 dev->caps.max_cqes = dev_cap->max_cq_sz - 1;
384 dev->caps.reserved_cqs = dev_cap->reserved_cqs;
385 dev->caps.reserved_eqs = dev_cap->reserved_eqs;
386 dev->caps.reserved_mtts = dev_cap->reserved_mtts;
387 dev->caps.reserved_mrws = dev_cap->reserved_mrws;
389 /* The first 128 UARs are used for EQ doorbells */
390 dev->caps.reserved_uars = max_t(int, 128, dev_cap->reserved_uars);
391 dev->caps.reserved_pds = dev_cap->reserved_pds;
392 dev->caps.reserved_xrcds = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
393 dev_cap->reserved_xrcds : 0;
394 dev->caps.max_xrcds = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
395 dev_cap->max_xrcds : 0;
396 dev->caps.mtt_entry_sz = dev_cap->mtt_entry_sz;
398 dev->caps.max_msg_sz = dev_cap->max_msg_sz;
399 dev->caps.page_size_cap = ~(u32) (dev_cap->min_page_sz - 1);
400 dev->caps.flags = dev_cap->flags;
401 dev->caps.flags2 = dev_cap->flags2;
402 dev->caps.bmme_flags = dev_cap->bmme_flags;
403 dev->caps.reserved_lkey = dev_cap->reserved_lkey;
404 dev->caps.stat_rate_support = dev_cap->stat_rate_support;
405 dev->caps.max_gso_sz = dev_cap->max_gso_sz;
406 dev->caps.max_rss_tbl_sz = dev_cap->max_rss_tbl_sz;
408 if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_PHV_EN) {
409 struct mlx4_init_hca_param hca_param;
411 memset(&hca_param, 0, sizeof(hca_param));
412 err = mlx4_QUERY_HCA(dev, &hca_param);
413 /* Turn off PHV_EN flag in case phv_check_en is set.
414 * phv_check_en is a HW check that parse the packet and verify
415 * phv bit was reported correctly in the wqe. To allow QinQ
416 * PHV_EN flag should be set and phv_check_en must be cleared
417 * otherwise QinQ packets will be drop by the HW.
419 if (err || hca_param.phv_check_en)
420 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_PHV_EN;
423 /* Sense port always allowed on supported devices for ConnectX-1 and -2 */
424 if (mlx4_priv(dev)->pci_dev_data & MLX4_PCI_DEV_FORCE_SENSE_PORT)
425 dev->caps.flags |= MLX4_DEV_CAP_FLAG_SENSE_SUPPORT;
426 /* Don't do sense port on multifunction devices (for now at least) */
427 if (mlx4_is_mfunc(dev))
428 dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_SENSE_SUPPORT;
430 if (mlx4_low_memory_profile()) {
431 dev->caps.log_num_macs = MLX4_MIN_LOG_NUM_MAC;
432 dev->caps.log_num_vlans = MLX4_MIN_LOG_NUM_VLANS;
434 dev->caps.log_num_macs = log_num_mac;
435 dev->caps.log_num_vlans = MLX4_LOG_NUM_VLANS;
438 for (i = 1; i <= dev->caps.num_ports; ++i) {
439 dev->caps.port_type[i] = MLX4_PORT_TYPE_NONE;
440 if (dev->caps.supported_type[i]) {
441 /* if only ETH is supported - assign ETH */
442 if (dev->caps.supported_type[i] == MLX4_PORT_TYPE_ETH)
443 dev->caps.port_type[i] = MLX4_PORT_TYPE_ETH;
444 /* if only IB is supported, assign IB */
445 else if (dev->caps.supported_type[i] ==
447 dev->caps.port_type[i] = MLX4_PORT_TYPE_IB;
449 /* if IB and ETH are supported, we set the port
450 * type according to user selection of port type;
451 * if user selected none, take the FW hint */
452 if (port_type_array[i - 1] == MLX4_PORT_TYPE_NONE)
453 dev->caps.port_type[i] = dev->caps.suggested_type[i] ?
454 MLX4_PORT_TYPE_ETH : MLX4_PORT_TYPE_IB;
456 dev->caps.port_type[i] = port_type_array[i - 1];
460 * Link sensing is allowed on the port if 3 conditions are true:
461 * 1. Both protocols are supported on the port.
462 * 2. Different types are supported on the port
463 * 3. FW declared that it supports link sensing
465 mlx4_priv(dev)->sense.sense_allowed[i] =
466 ((dev->caps.supported_type[i] == MLX4_PORT_TYPE_AUTO) &&
467 (dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
468 (dev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT));
471 * If "default_sense" bit is set, we move the port to "AUTO" mode
472 * and perform sense_port FW command to try and set the correct
473 * port type from beginning
475 if (mlx4_priv(dev)->sense.sense_allowed[i] && dev->caps.default_sense[i]) {
476 enum mlx4_port_type sensed_port = MLX4_PORT_TYPE_NONE;
477 dev->caps.possible_type[i] = MLX4_PORT_TYPE_AUTO;
478 mlx4_SENSE_PORT(dev, i, &sensed_port);
479 if (sensed_port != MLX4_PORT_TYPE_NONE)
480 dev->caps.port_type[i] = sensed_port;
482 dev->caps.possible_type[i] = dev->caps.port_type[i];
485 if (dev->caps.log_num_macs > dev_cap->port_cap[i].log_max_macs) {
486 dev->caps.log_num_macs = dev_cap->port_cap[i].log_max_macs;
487 mlx4_warn(dev, "Requested number of MACs is too much for port %d, reducing to %d\n",
488 i, 1 << dev->caps.log_num_macs);
490 if (dev->caps.log_num_vlans > dev_cap->port_cap[i].log_max_vlans) {
491 dev->caps.log_num_vlans = dev_cap->port_cap[i].log_max_vlans;
492 mlx4_warn(dev, "Requested number of VLANs is too much for port %d, reducing to %d\n",
493 i, 1 << dev->caps.log_num_vlans);
497 if (mlx4_is_master(dev) && (dev->caps.num_ports == 2) &&
498 (port_type_array[0] == MLX4_PORT_TYPE_IB) &&
499 (port_type_array[1] == MLX4_PORT_TYPE_ETH)) {
501 "Granular QoS per VF not supported with IB/Eth configuration\n");
502 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_QOS_VPP;
505 dev->caps.max_counters = dev_cap->max_counters;
507 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] = dev_cap->reserved_qps;
508 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] =
509 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] =
510 (1 << dev->caps.log_num_macs) *
511 (1 << dev->caps.log_num_vlans) *
513 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH] = MLX4_NUM_FEXCH;
515 if (dev_cap->dmfs_high_rate_qpn_base > 0 &&
516 dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN)
517 dev->caps.dmfs_high_rate_qpn_base = dev_cap->dmfs_high_rate_qpn_base;
519 dev->caps.dmfs_high_rate_qpn_base =
520 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
522 if (dev_cap->dmfs_high_rate_qpn_range > 0 &&
523 dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN) {
524 dev->caps.dmfs_high_rate_qpn_range = dev_cap->dmfs_high_rate_qpn_range;
525 dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_DEFAULT;
526 dev->caps.flags2 |= MLX4_DEV_CAP_FLAG2_FS_A0;
528 dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_NOT_SUPPORTED;
529 dev->caps.dmfs_high_rate_qpn_base =
530 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
531 dev->caps.dmfs_high_rate_qpn_range = MLX4_A0_STEERING_TABLE_SIZE;
534 dev->caps.rl_caps = dev_cap->rl_caps;
536 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_RSS_RAW_ETH] =
537 dev->caps.dmfs_high_rate_qpn_range;
539 dev->caps.reserved_qps = dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] +
540 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] +
541 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] +
542 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH];
544 dev->caps.sqp_demux = (mlx4_is_master(dev)) ? MLX4_MAX_NUM_SLAVES : 0;
546 if (!enable_64b_cqe_eqe && !mlx4_is_slave(dev)) {
548 (MLX4_DEV_CAP_FLAG_64B_CQE | MLX4_DEV_CAP_FLAG_64B_EQE)) {
549 mlx4_warn(dev, "64B EQEs/CQEs supported by the device but not enabled\n");
550 dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_64B_CQE;
551 dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_64B_EQE;
554 if (dev_cap->flags2 &
555 (MLX4_DEV_CAP_FLAG2_CQE_STRIDE |
556 MLX4_DEV_CAP_FLAG2_EQE_STRIDE)) {
557 mlx4_warn(dev, "Disabling EQE/CQE stride per user request\n");
558 dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
559 dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
563 if ((dev->caps.flags &
564 (MLX4_DEV_CAP_FLAG_64B_CQE | MLX4_DEV_CAP_FLAG_64B_EQE)) &&
566 dev->caps.function_caps |= MLX4_FUNC_CAP_64B_EQE_CQE;
568 if (!mlx4_is_slave(dev)) {
569 mlx4_enable_cqe_eqe_stride(dev);
570 dev->caps.alloc_res_qp_mask =
571 (dev->caps.bf_reg_size ? MLX4_RESERVE_ETH_BF_QP : 0) |
574 if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ETS_CFG) &&
575 dev->caps.flags & MLX4_DEV_CAP_FLAG_SET_ETH_SCHED) {
576 mlx4_warn(dev, "Old device ETS support detected\n");
577 mlx4_warn(dev, "Consider upgrading device FW.\n");
578 dev->caps.flags2 |= MLX4_DEV_CAP_FLAG2_ETS_CFG;
582 dev->caps.alloc_res_qp_mask = 0;
585 mlx4_enable_ignore_fcs(dev);
590 static int mlx4_get_pcie_dev_link_caps(struct mlx4_dev *dev,
591 enum pci_bus_speed *speed,
592 enum pcie_link_width *width)
594 u32 lnkcap1, lnkcap2;
597 #define PCIE_MLW_CAP_SHIFT 4 /* start of MLW mask in link capabilities */
599 *speed = PCI_SPEED_UNKNOWN;
600 *width = PCIE_LNK_WIDTH_UNKNOWN;
602 err1 = pcie_capability_read_dword(dev->persist->pdev, PCI_EXP_LNKCAP,
604 err2 = pcie_capability_read_dword(dev->persist->pdev, PCI_EXP_LNKCAP2,
606 if (!err2 && lnkcap2) { /* PCIe r3.0-compliant */
607 if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_8_0GB)
608 *speed = PCIE_SPEED_8_0GT;
609 else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_5_0GB)
610 *speed = PCIE_SPEED_5_0GT;
611 else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_2_5GB)
612 *speed = PCIE_SPEED_2_5GT;
615 *width = (lnkcap1 & PCI_EXP_LNKCAP_MLW) >> PCIE_MLW_CAP_SHIFT;
616 if (!lnkcap2) { /* pre-r3.0 */
617 if (lnkcap1 & PCI_EXP_LNKCAP_SLS_5_0GB)
618 *speed = PCIE_SPEED_5_0GT;
619 else if (lnkcap1 & PCI_EXP_LNKCAP_SLS_2_5GB)
620 *speed = PCIE_SPEED_2_5GT;
624 if (*speed == PCI_SPEED_UNKNOWN || *width == PCIE_LNK_WIDTH_UNKNOWN) {
626 err2 ? err2 : -EINVAL;
631 static void mlx4_check_pcie_caps(struct mlx4_dev *dev)
633 enum pcie_link_width width, width_cap;
634 enum pci_bus_speed speed, speed_cap;
637 #define PCIE_SPEED_STR(speed) \
638 (speed == PCIE_SPEED_8_0GT ? "8.0GT/s" : \
639 speed == PCIE_SPEED_5_0GT ? "5.0GT/s" : \
640 speed == PCIE_SPEED_2_5GT ? "2.5GT/s" : \
643 err = mlx4_get_pcie_dev_link_caps(dev, &speed_cap, &width_cap);
646 "Unable to determine PCIe device BW capabilities\n");
650 err = pcie_get_minimum_link(dev->persist->pdev, &speed, &width);
651 if (err || speed == PCI_SPEED_UNKNOWN ||
652 width == PCIE_LNK_WIDTH_UNKNOWN) {
654 "Unable to determine PCI device chain minimum BW\n");
658 if (width != width_cap || speed != speed_cap)
660 "PCIe BW is different than device's capability\n");
662 mlx4_info(dev, "PCIe link speed is %s, device supports %s\n",
663 PCIE_SPEED_STR(speed), PCIE_SPEED_STR(speed_cap));
664 mlx4_info(dev, "PCIe link width is x%d, device supports x%d\n",
669 /*The function checks if there are live vf, return the num of them*/
670 static int mlx4_how_many_lives_vf(struct mlx4_dev *dev)
672 struct mlx4_priv *priv = mlx4_priv(dev);
673 struct mlx4_slave_state *s_state;
677 for (i = 1/*the ppf is 0*/; i < dev->num_slaves; ++i) {
678 s_state = &priv->mfunc.master.slave_state[i];
679 if (s_state->active && s_state->last_cmd !=
680 MLX4_COMM_CMD_RESET) {
681 mlx4_warn(dev, "%s: slave: %d is still active\n",
689 int mlx4_get_parav_qkey(struct mlx4_dev *dev, u32 qpn, u32 *qkey)
691 u32 qk = MLX4_RESERVED_QKEY_BASE;
693 if (qpn >= dev->phys_caps.base_tunnel_sqpn + 8 * MLX4_MFUNC_MAX ||
694 qpn < dev->phys_caps.base_proxy_sqpn)
697 if (qpn >= dev->phys_caps.base_tunnel_sqpn)
699 qk += qpn - dev->phys_caps.base_tunnel_sqpn;
701 qk += qpn - dev->phys_caps.base_proxy_sqpn;
705 EXPORT_SYMBOL(mlx4_get_parav_qkey);
707 void mlx4_sync_pkey_table(struct mlx4_dev *dev, int slave, int port, int i, int val)
709 struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
711 if (!mlx4_is_master(dev))
714 priv->virt2phys_pkey[slave][port - 1][i] = val;
716 EXPORT_SYMBOL(mlx4_sync_pkey_table);
718 void mlx4_put_slave_node_guid(struct mlx4_dev *dev, int slave, __be64 guid)
720 struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
722 if (!mlx4_is_master(dev))
725 priv->slave_node_guids[slave] = guid;
727 EXPORT_SYMBOL(mlx4_put_slave_node_guid);
729 __be64 mlx4_get_slave_node_guid(struct mlx4_dev *dev, int slave)
731 struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
733 if (!mlx4_is_master(dev))
736 return priv->slave_node_guids[slave];
738 EXPORT_SYMBOL(mlx4_get_slave_node_guid);
740 int mlx4_is_slave_active(struct mlx4_dev *dev, int slave)
742 struct mlx4_priv *priv = mlx4_priv(dev);
743 struct mlx4_slave_state *s_slave;
745 if (!mlx4_is_master(dev))
748 s_slave = &priv->mfunc.master.slave_state[slave];
749 return !!s_slave->active;
751 EXPORT_SYMBOL(mlx4_is_slave_active);
753 static void slave_adjust_steering_mode(struct mlx4_dev *dev,
754 struct mlx4_dev_cap *dev_cap,
755 struct mlx4_init_hca_param *hca_param)
757 dev->caps.steering_mode = hca_param->steering_mode;
758 if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED) {
759 dev->caps.num_qp_per_mgm = dev_cap->fs_max_num_qp_per_entry;
760 dev->caps.fs_log_max_ucast_qp_range_size =
761 dev_cap->fs_log_max_ucast_qp_range_size;
763 dev->caps.num_qp_per_mgm =
764 4 * ((1 << hca_param->log_mc_entry_sz)/16 - 2);
766 mlx4_dbg(dev, "Steering mode is: %s\n",
767 mlx4_steering_mode_str(dev->caps.steering_mode));
770 static int mlx4_slave_cap(struct mlx4_dev *dev)
774 struct mlx4_dev_cap dev_cap;
775 struct mlx4_func_cap func_cap;
776 struct mlx4_init_hca_param hca_param;
779 memset(&hca_param, 0, sizeof(hca_param));
780 err = mlx4_QUERY_HCA(dev, &hca_param);
782 mlx4_err(dev, "QUERY_HCA command failed, aborting\n");
786 /* fail if the hca has an unknown global capability
787 * at this time global_caps should be always zeroed
789 if (hca_param.global_caps) {
790 mlx4_err(dev, "Unknown hca global capabilities\n");
794 mlx4_log_num_mgm_entry_size = hca_param.log_mc_entry_sz;
796 dev->caps.hca_core_clock = hca_param.hca_core_clock;
798 memset(&dev_cap, 0, sizeof(dev_cap));
799 dev->caps.max_qp_dest_rdma = 1 << hca_param.log_rd_per_qp;
800 err = mlx4_dev_cap(dev, &dev_cap);
802 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
806 err = mlx4_QUERY_FW(dev);
808 mlx4_err(dev, "QUERY_FW command failed: could not get FW version\n");
810 page_size = ~dev->caps.page_size_cap + 1;
811 mlx4_warn(dev, "HCA minimum page size:%d\n", page_size);
812 if (page_size > PAGE_SIZE) {
813 mlx4_err(dev, "HCA minimum page size of %d bigger than kernel PAGE_SIZE of %ld, aborting\n",
814 page_size, PAGE_SIZE);
818 /* slave gets uar page size from QUERY_HCA fw command */
819 dev->caps.uar_page_size = 1 << (hca_param.uar_page_sz + 12);
821 /* TODO: relax this assumption */
822 if (dev->caps.uar_page_size != PAGE_SIZE) {
823 mlx4_err(dev, "UAR size:%d != kernel PAGE_SIZE of %ld\n",
824 dev->caps.uar_page_size, PAGE_SIZE);
828 memset(&func_cap, 0, sizeof(func_cap));
829 err = mlx4_QUERY_FUNC_CAP(dev, 0, &func_cap);
831 mlx4_err(dev, "QUERY_FUNC_CAP general command failed, aborting (%d)\n",
836 if ((func_cap.pf_context_behaviour | PF_CONTEXT_BEHAVIOUR_MASK) !=
837 PF_CONTEXT_BEHAVIOUR_MASK) {
838 mlx4_err(dev, "Unknown pf context behaviour %x known flags %x\n",
839 func_cap.pf_context_behaviour, PF_CONTEXT_BEHAVIOUR_MASK);
843 dev->caps.num_ports = func_cap.num_ports;
844 dev->quotas.qp = func_cap.qp_quota;
845 dev->quotas.srq = func_cap.srq_quota;
846 dev->quotas.cq = func_cap.cq_quota;
847 dev->quotas.mpt = func_cap.mpt_quota;
848 dev->quotas.mtt = func_cap.mtt_quota;
849 dev->caps.num_qps = 1 << hca_param.log_num_qps;
850 dev->caps.num_srqs = 1 << hca_param.log_num_srqs;
851 dev->caps.num_cqs = 1 << hca_param.log_num_cqs;
852 dev->caps.num_mpts = 1 << hca_param.log_mpt_sz;
853 dev->caps.num_eqs = func_cap.max_eq;
854 dev->caps.reserved_eqs = func_cap.reserved_eq;
855 dev->caps.reserved_lkey = func_cap.reserved_lkey;
856 dev->caps.num_pds = MLX4_NUM_PDS;
857 dev->caps.num_mgms = 0;
858 dev->caps.num_amgms = 0;
860 if (dev->caps.num_ports > MLX4_MAX_PORTS) {
861 mlx4_err(dev, "HCA has %d ports, but we only support %d, aborting\n",
862 dev->caps.num_ports, MLX4_MAX_PORTS);
866 mlx4_replace_zero_macs(dev);
868 dev->caps.qp0_qkey = kcalloc(dev->caps.num_ports, sizeof(u32), GFP_KERNEL);
869 dev->caps.qp0_tunnel = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
870 dev->caps.qp0_proxy = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
871 dev->caps.qp1_tunnel = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
872 dev->caps.qp1_proxy = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
874 if (!dev->caps.qp0_tunnel || !dev->caps.qp0_proxy ||
875 !dev->caps.qp1_tunnel || !dev->caps.qp1_proxy ||
876 !dev->caps.qp0_qkey) {
881 for (i = 1; i <= dev->caps.num_ports; ++i) {
882 err = mlx4_QUERY_FUNC_CAP(dev, i, &func_cap);
884 mlx4_err(dev, "QUERY_FUNC_CAP port command failed for port %d, aborting (%d)\n",
888 dev->caps.qp0_qkey[i - 1] = func_cap.qp0_qkey;
889 dev->caps.qp0_tunnel[i - 1] = func_cap.qp0_tunnel_qpn;
890 dev->caps.qp0_proxy[i - 1] = func_cap.qp0_proxy_qpn;
891 dev->caps.qp1_tunnel[i - 1] = func_cap.qp1_tunnel_qpn;
892 dev->caps.qp1_proxy[i - 1] = func_cap.qp1_proxy_qpn;
893 dev->caps.port_mask[i] = dev->caps.port_type[i];
894 dev->caps.phys_port_id[i] = func_cap.phys_port_id;
895 err = mlx4_get_slave_pkey_gid_tbl_len(dev, i,
896 &dev->caps.gid_table_len[i],
897 &dev->caps.pkey_table_len[i]);
902 if (dev->caps.uar_page_size * (dev->caps.num_uars -
903 dev->caps.reserved_uars) >
904 pci_resource_len(dev->persist->pdev,
906 mlx4_err(dev, "HCA reported UAR region size of 0x%x bigger than PCI resource 2 size of 0x%llx, aborting\n",
907 dev->caps.uar_page_size * dev->caps.num_uars,
909 pci_resource_len(dev->persist->pdev, 2));
914 if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_64B_EQE_ENABLED) {
915 dev->caps.eqe_size = 64;
916 dev->caps.eqe_factor = 1;
918 dev->caps.eqe_size = 32;
919 dev->caps.eqe_factor = 0;
922 if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_64B_CQE_ENABLED) {
923 dev->caps.cqe_size = 64;
924 dev->caps.userspace_caps |= MLX4_USER_DEV_CAP_LARGE_CQE;
926 dev->caps.cqe_size = 32;
929 if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_EQE_STRIDE_ENABLED) {
930 dev->caps.eqe_size = hca_param.eqe_size;
931 dev->caps.eqe_factor = 0;
934 if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_CQE_STRIDE_ENABLED) {
935 dev->caps.cqe_size = hca_param.cqe_size;
936 /* User still need to know when CQE > 32B */
937 dev->caps.userspace_caps |= MLX4_USER_DEV_CAP_LARGE_CQE;
940 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
941 mlx4_warn(dev, "Timestamping is not supported in slave mode\n");
943 slave_adjust_steering_mode(dev, &dev_cap, &hca_param);
944 mlx4_dbg(dev, "RSS support for IP fragments is %s\n",
945 hca_param.rss_ip_frags ? "on" : "off");
947 if (func_cap.extra_flags & MLX4_QUERY_FUNC_FLAGS_BF_RES_QP &&
948 dev->caps.bf_reg_size)
949 dev->caps.alloc_res_qp_mask |= MLX4_RESERVE_ETH_BF_QP;
951 if (func_cap.extra_flags & MLX4_QUERY_FUNC_FLAGS_A0_RES_QP)
952 dev->caps.alloc_res_qp_mask |= MLX4_RESERVE_A0_QP;
957 kfree(dev->caps.qp0_qkey);
958 kfree(dev->caps.qp0_tunnel);
959 kfree(dev->caps.qp0_proxy);
960 kfree(dev->caps.qp1_tunnel);
961 kfree(dev->caps.qp1_proxy);
962 dev->caps.qp0_qkey = NULL;
963 dev->caps.qp0_tunnel = NULL;
964 dev->caps.qp0_proxy = NULL;
965 dev->caps.qp1_tunnel = NULL;
966 dev->caps.qp1_proxy = NULL;
971 static void mlx4_request_modules(struct mlx4_dev *dev)
974 int has_ib_port = false;
975 int has_eth_port = false;
976 #define EN_DRV_NAME "mlx4_en"
977 #define IB_DRV_NAME "mlx4_ib"
979 for (port = 1; port <= dev->caps.num_ports; port++) {
980 if (dev->caps.port_type[port] == MLX4_PORT_TYPE_IB)
982 else if (dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH)
987 request_module_nowait(EN_DRV_NAME);
988 if (has_ib_port || (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE))
989 request_module_nowait(IB_DRV_NAME);
993 * Change the port configuration of the device.
994 * Every user of this function must hold the port mutex.
996 int mlx4_change_port_types(struct mlx4_dev *dev,
997 enum mlx4_port_type *port_types)
1003 for (port = 0; port < dev->caps.num_ports; port++) {
1004 /* Change the port type only if the new type is different
1005 * from the current, and not set to Auto */
1006 if (port_types[port] != dev->caps.port_type[port + 1])
1010 mlx4_unregister_device(dev);
1011 for (port = 1; port <= dev->caps.num_ports; port++) {
1012 mlx4_CLOSE_PORT(dev, port);
1013 dev->caps.port_type[port] = port_types[port - 1];
1014 err = mlx4_SET_PORT(dev, port, -1);
1016 mlx4_err(dev, "Failed to set port %d, aborting\n",
1021 mlx4_set_port_mask(dev);
1022 err = mlx4_register_device(dev);
1024 mlx4_err(dev, "Failed to register device\n");
1027 mlx4_request_modules(dev);
1034 static ssize_t show_port_type(struct device *dev,
1035 struct device_attribute *attr,
1038 struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1040 struct mlx4_dev *mdev = info->dev;
1044 (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_IB) ?
1046 if (mdev->caps.possible_type[info->port] == MLX4_PORT_TYPE_AUTO)
1047 sprintf(buf, "auto (%s)\n", type);
1049 sprintf(buf, "%s\n", type);
1054 static ssize_t set_port_type(struct device *dev,
1055 struct device_attribute *attr,
1056 const char *buf, size_t count)
1058 struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1060 struct mlx4_dev *mdev = info->dev;
1061 struct mlx4_priv *priv = mlx4_priv(mdev);
1062 enum mlx4_port_type types[MLX4_MAX_PORTS];
1063 enum mlx4_port_type new_types[MLX4_MAX_PORTS];
1064 static DEFINE_MUTEX(set_port_type_mutex);
1068 mutex_lock(&set_port_type_mutex);
1070 if (!strcmp(buf, "ib\n"))
1071 info->tmp_type = MLX4_PORT_TYPE_IB;
1072 else if (!strcmp(buf, "eth\n"))
1073 info->tmp_type = MLX4_PORT_TYPE_ETH;
1074 else if (!strcmp(buf, "auto\n"))
1075 info->tmp_type = MLX4_PORT_TYPE_AUTO;
1077 mlx4_err(mdev, "%s is not supported port type\n", buf);
1082 mlx4_stop_sense(mdev);
1083 mutex_lock(&priv->port_mutex);
1084 /* Possible type is always the one that was delivered */
1085 mdev->caps.possible_type[info->port] = info->tmp_type;
1087 for (i = 0; i < mdev->caps.num_ports; i++) {
1088 types[i] = priv->port[i+1].tmp_type ? priv->port[i+1].tmp_type :
1089 mdev->caps.possible_type[i+1];
1090 if (types[i] == MLX4_PORT_TYPE_AUTO)
1091 types[i] = mdev->caps.port_type[i+1];
1094 if (!(mdev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
1095 !(mdev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT)) {
1096 for (i = 1; i <= mdev->caps.num_ports; i++) {
1097 if (mdev->caps.possible_type[i] == MLX4_PORT_TYPE_AUTO) {
1098 mdev->caps.possible_type[i] = mdev->caps.port_type[i];
1104 mlx4_err(mdev, "Auto sensing is not supported on this HCA. Set only 'eth' or 'ib' for both ports (should be the same)\n");
1108 mlx4_do_sense_ports(mdev, new_types, types);
1110 err = mlx4_check_port_params(mdev, new_types);
1114 /* We are about to apply the changes after the configuration
1115 * was verified, no need to remember the temporary types
1117 for (i = 0; i < mdev->caps.num_ports; i++)
1118 priv->port[i + 1].tmp_type = 0;
1120 err = mlx4_change_port_types(mdev, new_types);
1123 mlx4_start_sense(mdev);
1124 mutex_unlock(&priv->port_mutex);
1126 mutex_unlock(&set_port_type_mutex);
1128 return err ? err : count;
1139 static inline int int_to_ibta_mtu(int mtu)
1142 case 256: return IB_MTU_256;
1143 case 512: return IB_MTU_512;
1144 case 1024: return IB_MTU_1024;
1145 case 2048: return IB_MTU_2048;
1146 case 4096: return IB_MTU_4096;
1151 static inline int ibta_mtu_to_int(enum ibta_mtu mtu)
1154 case IB_MTU_256: return 256;
1155 case IB_MTU_512: return 512;
1156 case IB_MTU_1024: return 1024;
1157 case IB_MTU_2048: return 2048;
1158 case IB_MTU_4096: return 4096;
1163 static ssize_t show_port_ib_mtu(struct device *dev,
1164 struct device_attribute *attr,
1167 struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1169 struct mlx4_dev *mdev = info->dev;
1171 if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH)
1172 mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
1174 sprintf(buf, "%d\n",
1175 ibta_mtu_to_int(mdev->caps.port_ib_mtu[info->port]));
1179 static ssize_t set_port_ib_mtu(struct device *dev,
1180 struct device_attribute *attr,
1181 const char *buf, size_t count)
1183 struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1185 struct mlx4_dev *mdev = info->dev;
1186 struct mlx4_priv *priv = mlx4_priv(mdev);
1187 int err, port, mtu, ibta_mtu = -1;
1189 if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH) {
1190 mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
1194 err = kstrtoint(buf, 0, &mtu);
1196 ibta_mtu = int_to_ibta_mtu(mtu);
1198 if (err || ibta_mtu < 0) {
1199 mlx4_err(mdev, "%s is invalid IBTA mtu\n", buf);
1203 mdev->caps.port_ib_mtu[info->port] = ibta_mtu;
1205 mlx4_stop_sense(mdev);
1206 mutex_lock(&priv->port_mutex);
1207 mlx4_unregister_device(mdev);
1208 for (port = 1; port <= mdev->caps.num_ports; port++) {
1209 mlx4_CLOSE_PORT(mdev, port);
1210 err = mlx4_SET_PORT(mdev, port, -1);
1212 mlx4_err(mdev, "Failed to set port %d, aborting\n",
1217 err = mlx4_register_device(mdev);
1219 mutex_unlock(&priv->port_mutex);
1220 mlx4_start_sense(mdev);
1221 return err ? err : count;
1224 int mlx4_bond(struct mlx4_dev *dev)
1227 struct mlx4_priv *priv = mlx4_priv(dev);
1229 mutex_lock(&priv->bond_mutex);
1231 if (!mlx4_is_bonded(dev))
1232 ret = mlx4_do_bond(dev, true);
1236 mutex_unlock(&priv->bond_mutex);
1238 mlx4_err(dev, "Failed to bond device: %d\n", ret);
1240 mlx4_dbg(dev, "Device is bonded\n");
1243 EXPORT_SYMBOL_GPL(mlx4_bond);
1245 int mlx4_unbond(struct mlx4_dev *dev)
1248 struct mlx4_priv *priv = mlx4_priv(dev);
1250 mutex_lock(&priv->bond_mutex);
1252 if (mlx4_is_bonded(dev))
1253 ret = mlx4_do_bond(dev, false);
1255 mutex_unlock(&priv->bond_mutex);
1257 mlx4_err(dev, "Failed to unbond device: %d\n", ret);
1259 mlx4_dbg(dev, "Device is unbonded\n");
1262 EXPORT_SYMBOL_GPL(mlx4_unbond);
1265 int mlx4_port_map_set(struct mlx4_dev *dev, struct mlx4_port_map *v2p)
1267 u8 port1 = v2p->port1;
1268 u8 port2 = v2p->port2;
1269 struct mlx4_priv *priv = mlx4_priv(dev);
1272 if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_PORT_REMAP))
1275 mutex_lock(&priv->bond_mutex);
1277 /* zero means keep current mapping for this port */
1279 port1 = priv->v2p.port1;
1281 port2 = priv->v2p.port2;
1283 if ((port1 < 1) || (port1 > MLX4_MAX_PORTS) ||
1284 (port2 < 1) || (port2 > MLX4_MAX_PORTS) ||
1285 (port1 == 2 && port2 == 1)) {
1286 /* besides boundary checks cross mapping makes
1287 * no sense and therefore not allowed */
1289 } else if ((port1 == priv->v2p.port1) &&
1290 (port2 == priv->v2p.port2)) {
1293 err = mlx4_virt2phy_port_map(dev, port1, port2);
1295 mlx4_dbg(dev, "port map changed: [%d][%d]\n",
1297 priv->v2p.port1 = port1;
1298 priv->v2p.port2 = port2;
1300 mlx4_err(dev, "Failed to change port mape: %d\n", err);
1304 mutex_unlock(&priv->bond_mutex);
1307 EXPORT_SYMBOL_GPL(mlx4_port_map_set);
1309 static int mlx4_load_fw(struct mlx4_dev *dev)
1311 struct mlx4_priv *priv = mlx4_priv(dev);
1314 priv->fw.fw_icm = mlx4_alloc_icm(dev, priv->fw.fw_pages,
1315 GFP_HIGHUSER | __GFP_NOWARN, 0);
1316 if (!priv->fw.fw_icm) {
1317 mlx4_err(dev, "Couldn't allocate FW area, aborting\n");
1321 err = mlx4_MAP_FA(dev, priv->fw.fw_icm);
1323 mlx4_err(dev, "MAP_FA command failed, aborting\n");
1327 err = mlx4_RUN_FW(dev);
1329 mlx4_err(dev, "RUN_FW command failed, aborting\n");
1339 mlx4_free_icm(dev, priv->fw.fw_icm, 0);
1343 static int mlx4_init_cmpt_table(struct mlx4_dev *dev, u64 cmpt_base,
1346 struct mlx4_priv *priv = mlx4_priv(dev);
1350 err = mlx4_init_icm_table(dev, &priv->qp_table.cmpt_table,
1352 ((u64) (MLX4_CMPT_TYPE_QP *
1353 cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1354 cmpt_entry_sz, dev->caps.num_qps,
1355 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1360 err = mlx4_init_icm_table(dev, &priv->srq_table.cmpt_table,
1362 ((u64) (MLX4_CMPT_TYPE_SRQ *
1363 cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1364 cmpt_entry_sz, dev->caps.num_srqs,
1365 dev->caps.reserved_srqs, 0, 0);
1369 err = mlx4_init_icm_table(dev, &priv->cq_table.cmpt_table,
1371 ((u64) (MLX4_CMPT_TYPE_CQ *
1372 cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1373 cmpt_entry_sz, dev->caps.num_cqs,
1374 dev->caps.reserved_cqs, 0, 0);
1378 num_eqs = dev->phys_caps.num_phys_eqs;
1379 err = mlx4_init_icm_table(dev, &priv->eq_table.cmpt_table,
1381 ((u64) (MLX4_CMPT_TYPE_EQ *
1382 cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1383 cmpt_entry_sz, num_eqs, num_eqs, 0, 0);
1390 mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1393 mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1396 mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1402 static int mlx4_init_icm(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap,
1403 struct mlx4_init_hca_param *init_hca, u64 icm_size)
1405 struct mlx4_priv *priv = mlx4_priv(dev);
1410 err = mlx4_SET_ICM_SIZE(dev, icm_size, &aux_pages);
1412 mlx4_err(dev, "SET_ICM_SIZE command failed, aborting\n");
1416 mlx4_dbg(dev, "%lld KB of HCA context requires %lld KB aux memory\n",
1417 (unsigned long long) icm_size >> 10,
1418 (unsigned long long) aux_pages << 2);
1420 priv->fw.aux_icm = mlx4_alloc_icm(dev, aux_pages,
1421 GFP_HIGHUSER | __GFP_NOWARN, 0);
1422 if (!priv->fw.aux_icm) {
1423 mlx4_err(dev, "Couldn't allocate aux memory, aborting\n");
1427 err = mlx4_MAP_ICM_AUX(dev, priv->fw.aux_icm);
1429 mlx4_err(dev, "MAP_ICM_AUX command failed, aborting\n");
1433 err = mlx4_init_cmpt_table(dev, init_hca->cmpt_base, dev_cap->cmpt_entry_sz);
1435 mlx4_err(dev, "Failed to map cMPT context memory, aborting\n");
1440 num_eqs = dev->phys_caps.num_phys_eqs;
1441 err = mlx4_init_icm_table(dev, &priv->eq_table.table,
1442 init_hca->eqc_base, dev_cap->eqc_entry_sz,
1443 num_eqs, num_eqs, 0, 0);
1445 mlx4_err(dev, "Failed to map EQ context memory, aborting\n");
1446 goto err_unmap_cmpt;
1450 * Reserved MTT entries must be aligned up to a cacheline
1451 * boundary, since the FW will write to them, while the driver
1452 * writes to all other MTT entries. (The variable
1453 * dev->caps.mtt_entry_sz below is really the MTT segment
1454 * size, not the raw entry size)
1456 dev->caps.reserved_mtts =
1457 ALIGN(dev->caps.reserved_mtts * dev->caps.mtt_entry_sz,
1458 dma_get_cache_alignment()) / dev->caps.mtt_entry_sz;
1460 err = mlx4_init_icm_table(dev, &priv->mr_table.mtt_table,
1462 dev->caps.mtt_entry_sz,
1464 dev->caps.reserved_mtts, 1, 0);
1466 mlx4_err(dev, "Failed to map MTT context memory, aborting\n");
1470 err = mlx4_init_icm_table(dev, &priv->mr_table.dmpt_table,
1471 init_hca->dmpt_base,
1472 dev_cap->dmpt_entry_sz,
1474 dev->caps.reserved_mrws, 1, 1);
1476 mlx4_err(dev, "Failed to map dMPT context memory, aborting\n");
1480 err = mlx4_init_icm_table(dev, &priv->qp_table.qp_table,
1482 dev_cap->qpc_entry_sz,
1484 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1487 mlx4_err(dev, "Failed to map QP context memory, aborting\n");
1488 goto err_unmap_dmpt;
1491 err = mlx4_init_icm_table(dev, &priv->qp_table.auxc_table,
1492 init_hca->auxc_base,
1493 dev_cap->aux_entry_sz,
1495 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1498 mlx4_err(dev, "Failed to map AUXC context memory, aborting\n");
1502 err = mlx4_init_icm_table(dev, &priv->qp_table.altc_table,
1503 init_hca->altc_base,
1504 dev_cap->altc_entry_sz,
1506 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1509 mlx4_err(dev, "Failed to map ALTC context memory, aborting\n");
1510 goto err_unmap_auxc;
1513 err = mlx4_init_icm_table(dev, &priv->qp_table.rdmarc_table,
1514 init_hca->rdmarc_base,
1515 dev_cap->rdmarc_entry_sz << priv->qp_table.rdmarc_shift,
1517 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1520 mlx4_err(dev, "Failed to map RDMARC context memory, aborting\n");
1521 goto err_unmap_altc;
1524 err = mlx4_init_icm_table(dev, &priv->cq_table.table,
1526 dev_cap->cqc_entry_sz,
1528 dev->caps.reserved_cqs, 0, 0);
1530 mlx4_err(dev, "Failed to map CQ context memory, aborting\n");
1531 goto err_unmap_rdmarc;
1534 err = mlx4_init_icm_table(dev, &priv->srq_table.table,
1535 init_hca->srqc_base,
1536 dev_cap->srq_entry_sz,
1538 dev->caps.reserved_srqs, 0, 0);
1540 mlx4_err(dev, "Failed to map SRQ context memory, aborting\n");
1545 * For flow steering device managed mode it is required to use
1546 * mlx4_init_icm_table. For B0 steering mode it's not strictly
1547 * required, but for simplicity just map the whole multicast
1548 * group table now. The table isn't very big and it's a lot
1549 * easier than trying to track ref counts.
1551 err = mlx4_init_icm_table(dev, &priv->mcg_table.table,
1553 mlx4_get_mgm_entry_size(dev),
1554 dev->caps.num_mgms + dev->caps.num_amgms,
1555 dev->caps.num_mgms + dev->caps.num_amgms,
1558 mlx4_err(dev, "Failed to map MCG context memory, aborting\n");
1565 mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1568 mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1571 mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1574 mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1577 mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1580 mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1583 mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1586 mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
1589 mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
1592 mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1593 mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1594 mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1595 mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1598 mlx4_UNMAP_ICM_AUX(dev);
1601 mlx4_free_icm(dev, priv->fw.aux_icm, 0);
1606 static void mlx4_free_icms(struct mlx4_dev *dev)
1608 struct mlx4_priv *priv = mlx4_priv(dev);
1610 mlx4_cleanup_icm_table(dev, &priv->mcg_table.table);
1611 mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1612 mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1613 mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1614 mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1615 mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1616 mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1617 mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1618 mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
1619 mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
1620 mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1621 mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1622 mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1623 mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1625 mlx4_UNMAP_ICM_AUX(dev);
1626 mlx4_free_icm(dev, priv->fw.aux_icm, 0);
1629 static void mlx4_slave_exit(struct mlx4_dev *dev)
1631 struct mlx4_priv *priv = mlx4_priv(dev);
1633 mutex_lock(&priv->cmd.slave_cmd_mutex);
1634 if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, MLX4_COMM_CMD_NA_OP,
1636 mlx4_warn(dev, "Failed to close slave function\n");
1637 mutex_unlock(&priv->cmd.slave_cmd_mutex);
1640 static int map_bf_area(struct mlx4_dev *dev)
1642 struct mlx4_priv *priv = mlx4_priv(dev);
1643 resource_size_t bf_start;
1644 resource_size_t bf_len;
1647 if (!dev->caps.bf_reg_size)
1650 bf_start = pci_resource_start(dev->persist->pdev, 2) +
1651 (dev->caps.num_uars << PAGE_SHIFT);
1652 bf_len = pci_resource_len(dev->persist->pdev, 2) -
1653 (dev->caps.num_uars << PAGE_SHIFT);
1654 priv->bf_mapping = io_mapping_create_wc(bf_start, bf_len);
1655 if (!priv->bf_mapping)
1661 static void unmap_bf_area(struct mlx4_dev *dev)
1663 if (mlx4_priv(dev)->bf_mapping)
1664 io_mapping_free(mlx4_priv(dev)->bf_mapping);
1667 cycle_t mlx4_read_clock(struct mlx4_dev *dev)
1669 u32 clockhi, clocklo, clockhi1;
1672 struct mlx4_priv *priv = mlx4_priv(dev);
1674 for (i = 0; i < 10; i++) {
1675 clockhi = swab32(readl(priv->clock_mapping));
1676 clocklo = swab32(readl(priv->clock_mapping + 4));
1677 clockhi1 = swab32(readl(priv->clock_mapping));
1678 if (clockhi == clockhi1)
1682 cycles = (u64) clockhi << 32 | (u64) clocklo;
1686 EXPORT_SYMBOL_GPL(mlx4_read_clock);
1689 static int map_internal_clock(struct mlx4_dev *dev)
1691 struct mlx4_priv *priv = mlx4_priv(dev);
1693 priv->clock_mapping =
1694 ioremap(pci_resource_start(dev->persist->pdev,
1695 priv->fw.clock_bar) +
1696 priv->fw.clock_offset, MLX4_CLOCK_SIZE);
1698 if (!priv->clock_mapping)
1704 int mlx4_get_internal_clock_params(struct mlx4_dev *dev,
1705 struct mlx4_clock_params *params)
1707 struct mlx4_priv *priv = mlx4_priv(dev);
1709 if (mlx4_is_slave(dev))
1715 params->bar = priv->fw.clock_bar;
1716 params->offset = priv->fw.clock_offset;
1717 params->size = MLX4_CLOCK_SIZE;
1721 EXPORT_SYMBOL_GPL(mlx4_get_internal_clock_params);
1723 static void unmap_internal_clock(struct mlx4_dev *dev)
1725 struct mlx4_priv *priv = mlx4_priv(dev);
1727 if (priv->clock_mapping)
1728 iounmap(priv->clock_mapping);
1731 static void mlx4_close_hca(struct mlx4_dev *dev)
1733 unmap_internal_clock(dev);
1735 if (mlx4_is_slave(dev))
1736 mlx4_slave_exit(dev);
1738 mlx4_CLOSE_HCA(dev, 0);
1739 mlx4_free_icms(dev);
1743 static void mlx4_close_fw(struct mlx4_dev *dev)
1745 if (!mlx4_is_slave(dev)) {
1747 mlx4_free_icm(dev, mlx4_priv(dev)->fw.fw_icm, 0);
1751 static int mlx4_comm_check_offline(struct mlx4_dev *dev)
1753 #define COMM_CHAN_OFFLINE_OFFSET 0x09
1758 struct mlx4_priv *priv = mlx4_priv(dev);
1760 end = msecs_to_jiffies(MLX4_COMM_OFFLINE_TIME_OUT) + jiffies;
1761 while (time_before(jiffies, end)) {
1762 comm_flags = swab32(readl((__iomem char *)priv->mfunc.comm +
1763 MLX4_COMM_CHAN_FLAGS));
1764 offline_bit = (comm_flags &
1765 (u32)(1 << COMM_CHAN_OFFLINE_OFFSET));
1768 /* There are cases as part of AER/Reset flow that PF needs
1769 * around 100 msec to load. We therefore sleep for 100 msec
1770 * to allow other tasks to make use of that CPU during this
1775 mlx4_err(dev, "Communication channel is offline.\n");
1779 static void mlx4_reset_vf_support(struct mlx4_dev *dev)
1781 #define COMM_CHAN_RST_OFFSET 0x1e
1783 struct mlx4_priv *priv = mlx4_priv(dev);
1787 comm_caps = swab32(readl((__iomem char *)priv->mfunc.comm +
1788 MLX4_COMM_CHAN_CAPS));
1789 comm_rst = (comm_caps & (u32)(1 << COMM_CHAN_RST_OFFSET));
1792 dev->caps.vf_caps |= MLX4_VF_CAP_FLAG_RESET;
1795 static int mlx4_init_slave(struct mlx4_dev *dev)
1797 struct mlx4_priv *priv = mlx4_priv(dev);
1798 u64 dma = (u64) priv->mfunc.vhcr_dma;
1799 int ret_from_reset = 0;
1801 u32 cmd_channel_ver;
1803 if (atomic_read(&pf_loading)) {
1804 mlx4_warn(dev, "PF is not ready - Deferring probe\n");
1805 return -EPROBE_DEFER;
1808 mutex_lock(&priv->cmd.slave_cmd_mutex);
1809 priv->cmd.max_cmds = 1;
1810 if (mlx4_comm_check_offline(dev)) {
1811 mlx4_err(dev, "PF is not responsive, skipping initialization\n");
1815 mlx4_reset_vf_support(dev);
1816 mlx4_warn(dev, "Sending reset\n");
1817 ret_from_reset = mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0,
1818 MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME);
1819 /* if we are in the middle of flr the slave will try
1820 * NUM_OF_RESET_RETRIES times before leaving.*/
1821 if (ret_from_reset) {
1822 if (MLX4_DELAY_RESET_SLAVE == ret_from_reset) {
1823 mlx4_warn(dev, "slave is currently in the middle of FLR - Deferring probe\n");
1824 mutex_unlock(&priv->cmd.slave_cmd_mutex);
1825 return -EPROBE_DEFER;
1830 /* check the driver version - the slave I/F revision
1831 * must match the master's */
1832 slave_read = swab32(readl(&priv->mfunc.comm->slave_read));
1833 cmd_channel_ver = mlx4_comm_get_version();
1835 if (MLX4_COMM_GET_IF_REV(cmd_channel_ver) !=
1836 MLX4_COMM_GET_IF_REV(slave_read)) {
1837 mlx4_err(dev, "slave driver version is not supported by the master\n");
1841 mlx4_warn(dev, "Sending vhcr0\n");
1842 if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR0, dma >> 48,
1843 MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
1845 if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR1, dma >> 32,
1846 MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
1848 if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR2, dma >> 16,
1849 MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
1851 if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR_EN, dma,
1852 MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
1855 mutex_unlock(&priv->cmd.slave_cmd_mutex);
1859 mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, MLX4_COMM_CMD_NA_OP, 0);
1861 mutex_unlock(&priv->cmd.slave_cmd_mutex);
1865 static void mlx4_parav_master_pf_caps(struct mlx4_dev *dev)
1869 for (i = 1; i <= dev->caps.num_ports; i++) {
1870 if (dev->caps.port_type[i] == MLX4_PORT_TYPE_ETH)
1871 dev->caps.gid_table_len[i] =
1872 mlx4_get_slave_num_gids(dev, 0, i);
1874 dev->caps.gid_table_len[i] = 1;
1875 dev->caps.pkey_table_len[i] =
1876 dev->phys_caps.pkey_phys_table_len[i] - 1;
1880 static int choose_log_fs_mgm_entry_size(int qp_per_entry)
1882 int i = MLX4_MIN_MGM_LOG_ENTRY_SIZE;
1884 for (i = MLX4_MIN_MGM_LOG_ENTRY_SIZE; i <= MLX4_MAX_MGM_LOG_ENTRY_SIZE;
1886 if (qp_per_entry <= 4 * ((1 << i) / 16 - 2))
1890 return (i <= MLX4_MAX_MGM_LOG_ENTRY_SIZE) ? i : -1;
1893 static const char *dmfs_high_rate_steering_mode_str(int dmfs_high_steer_mode)
1895 switch (dmfs_high_steer_mode) {
1896 case MLX4_STEERING_DMFS_A0_DEFAULT:
1897 return "default performance";
1899 case MLX4_STEERING_DMFS_A0_DYNAMIC:
1900 return "dynamic hybrid mode";
1902 case MLX4_STEERING_DMFS_A0_STATIC:
1903 return "performance optimized for limited rule configuration (static)";
1905 case MLX4_STEERING_DMFS_A0_DISABLE:
1906 return "disabled performance optimized steering";
1908 case MLX4_STEERING_DMFS_A0_NOT_SUPPORTED:
1909 return "performance optimized steering not supported";
1912 return "Unrecognized mode";
1916 #define MLX4_DMFS_A0_STEERING (1UL << 2)
1918 static void choose_steering_mode(struct mlx4_dev *dev,
1919 struct mlx4_dev_cap *dev_cap)
1921 if (mlx4_log_num_mgm_entry_size <= 0) {
1922 if ((-mlx4_log_num_mgm_entry_size) & MLX4_DMFS_A0_STEERING) {
1923 if (dev->caps.dmfs_high_steer_mode ==
1924 MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
1925 mlx4_err(dev, "DMFS high rate mode not supported\n");
1927 dev->caps.dmfs_high_steer_mode =
1928 MLX4_STEERING_DMFS_A0_STATIC;
1932 if (mlx4_log_num_mgm_entry_size <= 0 &&
1933 dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_FS_EN &&
1934 (!mlx4_is_mfunc(dev) ||
1935 (dev_cap->fs_max_num_qp_per_entry >=
1936 (dev->persist->num_vfs + 1))) &&
1937 choose_log_fs_mgm_entry_size(dev_cap->fs_max_num_qp_per_entry) >=
1938 MLX4_MIN_MGM_LOG_ENTRY_SIZE) {
1939 dev->oper_log_mgm_entry_size =
1940 choose_log_fs_mgm_entry_size(dev_cap->fs_max_num_qp_per_entry);
1941 dev->caps.steering_mode = MLX4_STEERING_MODE_DEVICE_MANAGED;
1942 dev->caps.num_qp_per_mgm = dev_cap->fs_max_num_qp_per_entry;
1943 dev->caps.fs_log_max_ucast_qp_range_size =
1944 dev_cap->fs_log_max_ucast_qp_range_size;
1946 if (dev->caps.dmfs_high_steer_mode !=
1947 MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
1948 dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_DISABLE;
1949 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER &&
1950 dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER)
1951 dev->caps.steering_mode = MLX4_STEERING_MODE_B0;
1953 dev->caps.steering_mode = MLX4_STEERING_MODE_A0;
1955 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER ||
1956 dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER)
1957 mlx4_warn(dev, "Must have both UC_STEER and MC_STEER flags set to use B0 steering - falling back to A0 steering mode\n");
1959 dev->oper_log_mgm_entry_size =
1960 mlx4_log_num_mgm_entry_size > 0 ?
1961 mlx4_log_num_mgm_entry_size :
1962 MLX4_DEFAULT_MGM_LOG_ENTRY_SIZE;
1963 dev->caps.num_qp_per_mgm = mlx4_get_qp_per_mgm(dev);
1965 mlx4_dbg(dev, "Steering mode is: %s, oper_log_mgm_entry_size = %d, modparam log_num_mgm_entry_size = %d\n",
1966 mlx4_steering_mode_str(dev->caps.steering_mode),
1967 dev->oper_log_mgm_entry_size,
1968 mlx4_log_num_mgm_entry_size);
1971 static void choose_tunnel_offload_mode(struct mlx4_dev *dev,
1972 struct mlx4_dev_cap *dev_cap)
1974 if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED &&
1975 dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_VXLAN_OFFLOADS)
1976 dev->caps.tunnel_offload_mode = MLX4_TUNNEL_OFFLOAD_MODE_VXLAN;
1978 dev->caps.tunnel_offload_mode = MLX4_TUNNEL_OFFLOAD_MODE_NONE;
1980 mlx4_dbg(dev, "Tunneling offload mode is: %s\n", (dev->caps.tunnel_offload_mode
1981 == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) ? "vxlan" : "none");
1984 static int mlx4_validate_optimized_steering(struct mlx4_dev *dev)
1987 struct mlx4_port_cap port_cap;
1989 if (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
1992 for (i = 1; i <= dev->caps.num_ports; i++) {
1993 if (mlx4_dev_port(dev, i, &port_cap)) {
1995 "QUERY_DEV_CAP command failed, can't veify DMFS high rate steering.\n");
1996 } else if ((dev->caps.dmfs_high_steer_mode !=
1997 MLX4_STEERING_DMFS_A0_DEFAULT) &&
1998 (port_cap.dmfs_optimized_state ==
1999 !!(dev->caps.dmfs_high_steer_mode ==
2000 MLX4_STEERING_DMFS_A0_DISABLE))) {
2002 "DMFS high rate steer mode differ, driver requested %s but %s in FW.\n",
2003 dmfs_high_rate_steering_mode_str(
2004 dev->caps.dmfs_high_steer_mode),
2005 (port_cap.dmfs_optimized_state ?
2006 "enabled" : "disabled"));
2013 static int mlx4_init_fw(struct mlx4_dev *dev)
2015 struct mlx4_mod_stat_cfg mlx4_cfg;
2018 if (!mlx4_is_slave(dev)) {
2019 err = mlx4_QUERY_FW(dev);
2022 mlx4_info(dev, "non-primary physical function, skipping\n");
2024 mlx4_err(dev, "QUERY_FW command failed, aborting\n");
2028 err = mlx4_load_fw(dev);
2030 mlx4_err(dev, "Failed to start FW, aborting\n");
2034 mlx4_cfg.log_pg_sz_m = 1;
2035 mlx4_cfg.log_pg_sz = 0;
2036 err = mlx4_MOD_STAT_CFG(dev, &mlx4_cfg);
2038 mlx4_warn(dev, "Failed to override log_pg_sz parameter\n");
2044 static int mlx4_init_hca(struct mlx4_dev *dev)
2046 struct mlx4_priv *priv = mlx4_priv(dev);
2047 struct mlx4_adapter adapter;
2048 struct mlx4_dev_cap dev_cap;
2049 struct mlx4_profile profile;
2050 struct mlx4_init_hca_param init_hca;
2052 struct mlx4_config_dev_params params;
2055 if (!mlx4_is_slave(dev)) {
2056 err = mlx4_dev_cap(dev, &dev_cap);
2058 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
2062 choose_steering_mode(dev, &dev_cap);
2063 choose_tunnel_offload_mode(dev, &dev_cap);
2065 if (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC &&
2066 mlx4_is_master(dev))
2067 dev->caps.function_caps |= MLX4_FUNC_CAP_DMFS_A0_STATIC;
2069 err = mlx4_get_phys_port_id(dev);
2071 mlx4_err(dev, "Fail to get physical port id\n");
2073 if (mlx4_is_master(dev))
2074 mlx4_parav_master_pf_caps(dev);
2076 if (mlx4_low_memory_profile()) {
2077 mlx4_info(dev, "Running from within kdump kernel. Using low memory profile\n");
2078 profile = low_mem_profile;
2080 profile = default_profile;
2082 if (dev->caps.steering_mode ==
2083 MLX4_STEERING_MODE_DEVICE_MANAGED)
2084 profile.num_mcg = MLX4_FS_NUM_MCG;
2086 icm_size = mlx4_make_profile(dev, &profile, &dev_cap,
2088 if ((long long) icm_size < 0) {
2093 dev->caps.max_fmr_maps = (1 << (32 - ilog2(dev->caps.num_mpts))) - 1;
2095 init_hca.log_uar_sz = ilog2(dev->caps.num_uars);
2096 init_hca.uar_page_sz = PAGE_SHIFT - 12;
2097 init_hca.mw_enabled = 0;
2098 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2099 dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN)
2100 init_hca.mw_enabled = INIT_HCA_TPT_MW_ENABLE;
2102 err = mlx4_init_icm(dev, &dev_cap, &init_hca, icm_size);
2106 err = mlx4_INIT_HCA(dev, &init_hca);
2108 mlx4_err(dev, "INIT_HCA command failed, aborting\n");
2112 if (dev_cap.flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS) {
2113 err = mlx4_query_func(dev, &dev_cap);
2115 mlx4_err(dev, "QUERY_FUNC command failed, aborting.\n");
2117 } else if (err & MLX4_QUERY_FUNC_NUM_SYS_EQS) {
2118 dev->caps.num_eqs = dev_cap.max_eqs;
2119 dev->caps.reserved_eqs = dev_cap.reserved_eqs;
2120 dev->caps.reserved_uars = dev_cap.reserved_uars;
2125 * If TS is supported by FW
2126 * read HCA frequency by QUERY_HCA command
2128 if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS) {
2129 memset(&init_hca, 0, sizeof(init_hca));
2130 err = mlx4_QUERY_HCA(dev, &init_hca);
2132 mlx4_err(dev, "QUERY_HCA command failed, disable timestamp\n");
2133 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
2135 dev->caps.hca_core_clock =
2136 init_hca.hca_core_clock;
2139 /* In case we got HCA frequency 0 - disable timestamping
2140 * to avoid dividing by zero
2142 if (!dev->caps.hca_core_clock) {
2143 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
2145 "HCA frequency is 0 - timestamping is not supported\n");
2146 } else if (map_internal_clock(dev)) {
2148 * Map internal clock,
2149 * in case of failure disable timestamping
2151 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
2152 mlx4_err(dev, "Failed to map internal clock. Timestamping is not supported\n");
2156 if (dev->caps.dmfs_high_steer_mode !=
2157 MLX4_STEERING_DMFS_A0_NOT_SUPPORTED) {
2158 if (mlx4_validate_optimized_steering(dev))
2159 mlx4_warn(dev, "Optimized steering validation failed\n");
2161 if (dev->caps.dmfs_high_steer_mode ==
2162 MLX4_STEERING_DMFS_A0_DISABLE) {
2163 dev->caps.dmfs_high_rate_qpn_base =
2164 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
2165 dev->caps.dmfs_high_rate_qpn_range =
2166 MLX4_A0_STEERING_TABLE_SIZE;
2169 mlx4_dbg(dev, "DMFS high rate steer mode is: %s\n",
2170 dmfs_high_rate_steering_mode_str(
2171 dev->caps.dmfs_high_steer_mode));
2174 err = mlx4_init_slave(dev);
2176 if (err != -EPROBE_DEFER)
2177 mlx4_err(dev, "Failed to initialize slave\n");
2181 err = mlx4_slave_cap(dev);
2183 mlx4_err(dev, "Failed to obtain slave caps\n");
2188 if (map_bf_area(dev))
2189 mlx4_dbg(dev, "Failed to map blue flame area\n");
2191 /*Only the master set the ports, all the rest got it from it.*/
2192 if (!mlx4_is_slave(dev))
2193 mlx4_set_port_mask(dev);
2195 err = mlx4_QUERY_ADAPTER(dev, &adapter);
2197 mlx4_err(dev, "QUERY_ADAPTER command failed, aborting\n");
2201 /* Query CONFIG_DEV parameters */
2202 err = mlx4_config_dev_retrieval(dev, ¶ms);
2203 if (err && err != -ENOTSUPP) {
2204 mlx4_err(dev, "Failed to query CONFIG_DEV parameters\n");
2206 dev->caps.rx_checksum_flags_port[1] = params.rx_csum_flags_port_1;
2207 dev->caps.rx_checksum_flags_port[2] = params.rx_csum_flags_port_2;
2209 priv->eq_table.inta_pin = adapter.inta_pin;
2210 memcpy(dev->board_id, adapter.board_id, sizeof dev->board_id);
2215 unmap_internal_clock(dev);
2218 if (mlx4_is_slave(dev)) {
2219 kfree(dev->caps.qp0_qkey);
2220 kfree(dev->caps.qp0_tunnel);
2221 kfree(dev->caps.qp0_proxy);
2222 kfree(dev->caps.qp1_tunnel);
2223 kfree(dev->caps.qp1_proxy);
2227 if (mlx4_is_slave(dev))
2228 mlx4_slave_exit(dev);
2230 mlx4_CLOSE_HCA(dev, 0);
2233 if (!mlx4_is_slave(dev))
2234 mlx4_free_icms(dev);
2239 static int mlx4_init_counters_table(struct mlx4_dev *dev)
2241 struct mlx4_priv *priv = mlx4_priv(dev);
2244 if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2247 if (!dev->caps.max_counters)
2250 nent_pow2 = roundup_pow_of_two(dev->caps.max_counters);
2251 /* reserve last counter index for sink counter */
2252 return mlx4_bitmap_init(&priv->counters_bitmap, nent_pow2,
2254 nent_pow2 - dev->caps.max_counters + 1);
2257 static void mlx4_cleanup_counters_table(struct mlx4_dev *dev)
2259 if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2262 if (!dev->caps.max_counters)
2265 mlx4_bitmap_cleanup(&mlx4_priv(dev)->counters_bitmap);
2268 static void mlx4_cleanup_default_counters(struct mlx4_dev *dev)
2270 struct mlx4_priv *priv = mlx4_priv(dev);
2273 for (port = 0; port < dev->caps.num_ports; port++)
2274 if (priv->def_counter[port] != -1)
2275 mlx4_counter_free(dev, priv->def_counter[port]);
2278 static int mlx4_allocate_default_counters(struct mlx4_dev *dev)
2280 struct mlx4_priv *priv = mlx4_priv(dev);
2284 for (port = 0; port < dev->caps.num_ports; port++)
2285 priv->def_counter[port] = -1;
2287 for (port = 0; port < dev->caps.num_ports; port++) {
2288 err = mlx4_counter_alloc(dev, &idx);
2290 if (!err || err == -ENOSPC) {
2291 priv->def_counter[port] = idx;
2292 } else if (err == -ENOENT) {
2295 } else if (mlx4_is_slave(dev) && err == -EINVAL) {
2296 priv->def_counter[port] = MLX4_SINK_COUNTER_INDEX(dev);
2297 mlx4_warn(dev, "can't allocate counter from old PF driver, using index %d\n",
2298 MLX4_SINK_COUNTER_INDEX(dev));
2301 mlx4_err(dev, "%s: failed to allocate default counter port %d err %d\n",
2302 __func__, port + 1, err);
2303 mlx4_cleanup_default_counters(dev);
2307 mlx4_dbg(dev, "%s: default counter index %d for port %d\n",
2308 __func__, priv->def_counter[port], port + 1);
2314 int __mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx)
2316 struct mlx4_priv *priv = mlx4_priv(dev);
2318 if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2321 *idx = mlx4_bitmap_alloc(&priv->counters_bitmap);
2323 *idx = MLX4_SINK_COUNTER_INDEX(dev);
2330 int mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx)
2335 if (mlx4_is_mfunc(dev)) {
2336 err = mlx4_cmd_imm(dev, 0, &out_param, RES_COUNTER,
2337 RES_OP_RESERVE, MLX4_CMD_ALLOC_RES,
2338 MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
2340 *idx = get_param_l(&out_param);
2344 return __mlx4_counter_alloc(dev, idx);
2346 EXPORT_SYMBOL_GPL(mlx4_counter_alloc);
2348 static int __mlx4_clear_if_stat(struct mlx4_dev *dev,
2351 struct mlx4_cmd_mailbox *if_stat_mailbox;
2353 u32 if_stat_in_mod = (counter_index & 0xff) | MLX4_QUERY_IF_STAT_RESET;
2355 if_stat_mailbox = mlx4_alloc_cmd_mailbox(dev);
2356 if (IS_ERR(if_stat_mailbox))
2357 return PTR_ERR(if_stat_mailbox);
2359 err = mlx4_cmd_box(dev, 0, if_stat_mailbox->dma, if_stat_in_mod, 0,
2360 MLX4_CMD_QUERY_IF_STAT, MLX4_CMD_TIME_CLASS_C,
2363 mlx4_free_cmd_mailbox(dev, if_stat_mailbox);
2367 void __mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
2369 if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2372 if (idx == MLX4_SINK_COUNTER_INDEX(dev))
2375 __mlx4_clear_if_stat(dev, idx);
2377 mlx4_bitmap_free(&mlx4_priv(dev)->counters_bitmap, idx, MLX4_USE_RR);
2381 void mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
2385 if (mlx4_is_mfunc(dev)) {
2386 set_param_l(&in_param, idx);
2387 mlx4_cmd(dev, in_param, RES_COUNTER, RES_OP_RESERVE,
2388 MLX4_CMD_FREE_RES, MLX4_CMD_TIME_CLASS_A,
2392 __mlx4_counter_free(dev, idx);
2394 EXPORT_SYMBOL_GPL(mlx4_counter_free);
2396 int mlx4_get_default_counter_index(struct mlx4_dev *dev, int port)
2398 struct mlx4_priv *priv = mlx4_priv(dev);
2400 return priv->def_counter[port - 1];
2402 EXPORT_SYMBOL_GPL(mlx4_get_default_counter_index);
2404 void mlx4_set_admin_guid(struct mlx4_dev *dev, __be64 guid, int entry, int port)
2406 struct mlx4_priv *priv = mlx4_priv(dev);
2408 priv->mfunc.master.vf_admin[entry].vport[port].guid = guid;
2410 EXPORT_SYMBOL_GPL(mlx4_set_admin_guid);
2412 __be64 mlx4_get_admin_guid(struct mlx4_dev *dev, int entry, int port)
2414 struct mlx4_priv *priv = mlx4_priv(dev);
2416 return priv->mfunc.master.vf_admin[entry].vport[port].guid;
2418 EXPORT_SYMBOL_GPL(mlx4_get_admin_guid);
2420 void mlx4_set_random_admin_guid(struct mlx4_dev *dev, int entry, int port)
2422 struct mlx4_priv *priv = mlx4_priv(dev);
2429 get_random_bytes((char *)&guid, sizeof(guid));
2430 guid &= ~(cpu_to_be64(1ULL << 56));
2431 guid |= cpu_to_be64(1ULL << 57);
2432 priv->mfunc.master.vf_admin[entry].vport[port].guid = guid;
2435 static int mlx4_setup_hca(struct mlx4_dev *dev)
2437 struct mlx4_priv *priv = mlx4_priv(dev);
2440 __be32 ib_port_default_caps;
2442 err = mlx4_init_uar_table(dev);
2444 mlx4_err(dev, "Failed to initialize user access region table, aborting\n");
2448 err = mlx4_uar_alloc(dev, &priv->driver_uar);
2450 mlx4_err(dev, "Failed to allocate driver access region, aborting\n");
2451 goto err_uar_table_free;
2454 priv->kar = ioremap((phys_addr_t) priv->driver_uar.pfn << PAGE_SHIFT, PAGE_SIZE);
2456 mlx4_err(dev, "Couldn't map kernel access region, aborting\n");
2461 err = mlx4_init_pd_table(dev);
2463 mlx4_err(dev, "Failed to initialize protection domain table, aborting\n");
2467 err = mlx4_init_xrcd_table(dev);
2469 mlx4_err(dev, "Failed to initialize reliable connection domain table, aborting\n");
2470 goto err_pd_table_free;
2473 err = mlx4_init_mr_table(dev);
2475 mlx4_err(dev, "Failed to initialize memory region table, aborting\n");
2476 goto err_xrcd_table_free;
2479 if (!mlx4_is_slave(dev)) {
2480 err = mlx4_init_mcg_table(dev);
2482 mlx4_err(dev, "Failed to initialize multicast group table, aborting\n");
2483 goto err_mr_table_free;
2485 err = mlx4_config_mad_demux(dev);
2487 mlx4_err(dev, "Failed in config_mad_demux, aborting\n");
2488 goto err_mcg_table_free;
2492 err = mlx4_init_eq_table(dev);
2494 mlx4_err(dev, "Failed to initialize event queue table, aborting\n");
2495 goto err_mcg_table_free;
2498 err = mlx4_cmd_use_events(dev);
2500 mlx4_err(dev, "Failed to switch to event-driven firmware commands, aborting\n");
2501 goto err_eq_table_free;
2504 err = mlx4_NOP(dev);
2506 if (dev->flags & MLX4_FLAG_MSI_X) {
2507 mlx4_warn(dev, "NOP command failed to generate MSI-X interrupt IRQ %d)\n",
2508 priv->eq_table.eq[MLX4_EQ_ASYNC].irq);
2509 mlx4_warn(dev, "Trying again without MSI-X\n");
2511 mlx4_err(dev, "NOP command failed to generate interrupt (IRQ %d), aborting\n",
2512 priv->eq_table.eq[MLX4_EQ_ASYNC].irq);
2513 mlx4_err(dev, "BIOS or ACPI interrupt routing problem?\n");
2519 mlx4_dbg(dev, "NOP command IRQ test passed\n");
2521 err = mlx4_init_cq_table(dev);
2523 mlx4_err(dev, "Failed to initialize completion queue table, aborting\n");
2527 err = mlx4_init_srq_table(dev);
2529 mlx4_err(dev, "Failed to initialize shared receive queue table, aborting\n");
2530 goto err_cq_table_free;
2533 err = mlx4_init_qp_table(dev);
2535 mlx4_err(dev, "Failed to initialize queue pair table, aborting\n");
2536 goto err_srq_table_free;
2539 if (!mlx4_is_slave(dev)) {
2540 err = mlx4_init_counters_table(dev);
2541 if (err && err != -ENOENT) {
2542 mlx4_err(dev, "Failed to initialize counters table, aborting\n");
2543 goto err_qp_table_free;
2547 err = mlx4_allocate_default_counters(dev);
2549 mlx4_err(dev, "Failed to allocate default counters, aborting\n");
2550 goto err_counters_table_free;
2553 if (!mlx4_is_slave(dev)) {
2554 for (port = 1; port <= dev->caps.num_ports; port++) {
2555 ib_port_default_caps = 0;
2556 err = mlx4_get_port_ib_caps(dev, port,
2557 &ib_port_default_caps);
2559 mlx4_warn(dev, "failed to get port %d default ib capabilities (%d). Continuing with caps = 0\n",
2561 dev->caps.ib_port_def_cap[port] = ib_port_default_caps;
2563 /* initialize per-slave default ib port capabilities */
2564 if (mlx4_is_master(dev)) {
2566 for (i = 0; i < dev->num_slaves; i++) {
2567 if (i == mlx4_master_func_num(dev))
2569 priv->mfunc.master.slave_state[i].ib_cap_mask[port] =
2570 ib_port_default_caps;
2574 if (mlx4_is_mfunc(dev))
2575 dev->caps.port_ib_mtu[port] = IB_MTU_2048;
2577 dev->caps.port_ib_mtu[port] = IB_MTU_4096;
2579 err = mlx4_SET_PORT(dev, port, mlx4_is_master(dev) ?
2580 dev->caps.pkey_table_len[port] : -1);
2582 mlx4_err(dev, "Failed to set port %d, aborting\n",
2584 goto err_default_countes_free;
2591 err_default_countes_free:
2592 mlx4_cleanup_default_counters(dev);
2594 err_counters_table_free:
2595 if (!mlx4_is_slave(dev))
2596 mlx4_cleanup_counters_table(dev);
2599 mlx4_cleanup_qp_table(dev);
2602 mlx4_cleanup_srq_table(dev);
2605 mlx4_cleanup_cq_table(dev);
2608 mlx4_cmd_use_polling(dev);
2611 mlx4_cleanup_eq_table(dev);
2614 if (!mlx4_is_slave(dev))
2615 mlx4_cleanup_mcg_table(dev);
2618 mlx4_cleanup_mr_table(dev);
2620 err_xrcd_table_free:
2621 mlx4_cleanup_xrcd_table(dev);
2624 mlx4_cleanup_pd_table(dev);
2630 mlx4_uar_free(dev, &priv->driver_uar);
2633 mlx4_cleanup_uar_table(dev);
2637 static int mlx4_init_affinity_hint(struct mlx4_dev *dev, int port, int eqn)
2639 int requested_cpu = 0;
2640 struct mlx4_priv *priv = mlx4_priv(dev);
2645 if (eqn > dev->caps.num_comp_vectors)
2648 for (i = 1; i < port; i++)
2649 off += mlx4_get_eqs_per_port(dev, i);
2651 requested_cpu = eqn - off - !!(eqn > MLX4_EQ_ASYNC);
2653 /* Meaning EQs are shared, and this call comes from the second port */
2654 if (requested_cpu < 0)
2657 eq = &priv->eq_table.eq[eqn];
2659 if (!zalloc_cpumask_var(&eq->affinity_mask, GFP_KERNEL))
2662 cpumask_set_cpu(requested_cpu, eq->affinity_mask);
2667 static void mlx4_enable_msi_x(struct mlx4_dev *dev)
2669 struct mlx4_priv *priv = mlx4_priv(dev);
2670 struct msix_entry *entries;
2675 int nreq = dev->caps.num_ports * num_online_cpus() + 1;
2677 nreq = min_t(int, dev->caps.num_eqs - dev->caps.reserved_eqs,
2679 if (nreq > MAX_MSIX)
2682 entries = kcalloc(nreq, sizeof *entries, GFP_KERNEL);
2686 for (i = 0; i < nreq; ++i)
2687 entries[i].entry = i;
2689 nreq = pci_enable_msix_range(dev->persist->pdev, entries, 2,
2692 if (nreq < 0 || nreq < MLX4_EQ_ASYNC) {
2696 /* 1 is reserved for events (asyncrounous EQ) */
2697 dev->caps.num_comp_vectors = nreq - 1;
2699 priv->eq_table.eq[MLX4_EQ_ASYNC].irq = entries[0].vector;
2700 bitmap_zero(priv->eq_table.eq[MLX4_EQ_ASYNC].actv_ports.ports,
2701 dev->caps.num_ports);
2703 for (i = 0; i < dev->caps.num_comp_vectors + 1; i++) {
2704 if (i == MLX4_EQ_ASYNC)
2707 priv->eq_table.eq[i].irq =
2708 entries[i + 1 - !!(i > MLX4_EQ_ASYNC)].vector;
2710 if (MLX4_IS_LEGACY_EQ_MODE(dev->caps)) {
2711 bitmap_fill(priv->eq_table.eq[i].actv_ports.ports,
2712 dev->caps.num_ports);
2713 /* We don't set affinity hint when there
2718 priv->eq_table.eq[i].actv_ports.ports);
2719 if (mlx4_init_affinity_hint(dev, port + 1, i))
2720 mlx4_warn(dev, "Couldn't init hint cpumask for EQ %d\n",
2723 /* We divide the Eqs evenly between the two ports.
2724 * (dev->caps.num_comp_vectors / dev->caps.num_ports)
2725 * refers to the number of Eqs per port
2726 * (i.e eqs_per_port). Theoretically, we would like to
2727 * write something like (i + 1) % eqs_per_port == 0.
2728 * However, since there's an asynchronous Eq, we have
2729 * to skip over it by comparing this condition to
2730 * !!((i + 1) > MLX4_EQ_ASYNC).
2732 if ((dev->caps.num_comp_vectors > dev->caps.num_ports) &&
2734 (dev->caps.num_comp_vectors / dev->caps.num_ports)) ==
2735 !!((i + 1) > MLX4_EQ_ASYNC))
2736 /* If dev->caps.num_comp_vectors < dev->caps.num_ports,
2737 * everything is shared anyway.
2742 dev->flags |= MLX4_FLAG_MSI_X;
2749 dev->caps.num_comp_vectors = 1;
2751 BUG_ON(MLX4_EQ_ASYNC >= 2);
2752 for (i = 0; i < 2; ++i) {
2753 priv->eq_table.eq[i].irq = dev->persist->pdev->irq;
2754 if (i != MLX4_EQ_ASYNC) {
2755 bitmap_fill(priv->eq_table.eq[i].actv_ports.ports,
2756 dev->caps.num_ports);
2761 static int mlx4_init_port_info(struct mlx4_dev *dev, int port)
2763 struct mlx4_port_info *info = &mlx4_priv(dev)->port[port];
2768 if (!mlx4_is_slave(dev)) {
2769 mlx4_init_mac_table(dev, &info->mac_table);
2770 mlx4_init_vlan_table(dev, &info->vlan_table);
2771 mlx4_init_roce_gid_table(dev, &info->gid_table);
2772 info->base_qpn = mlx4_get_base_qpn(dev, port);
2775 sprintf(info->dev_name, "mlx4_port%d", port);
2776 info->port_attr.attr.name = info->dev_name;
2777 if (mlx4_is_mfunc(dev))
2778 info->port_attr.attr.mode = S_IRUGO;
2780 info->port_attr.attr.mode = S_IRUGO | S_IWUSR;
2781 info->port_attr.store = set_port_type;
2783 info->port_attr.show = show_port_type;
2784 sysfs_attr_init(&info->port_attr.attr);
2786 err = device_create_file(&dev->persist->pdev->dev, &info->port_attr);
2788 mlx4_err(dev, "Failed to create file for port %d\n", port);
2792 sprintf(info->dev_mtu_name, "mlx4_port%d_mtu", port);
2793 info->port_mtu_attr.attr.name = info->dev_mtu_name;
2794 if (mlx4_is_mfunc(dev))
2795 info->port_mtu_attr.attr.mode = S_IRUGO;
2797 info->port_mtu_attr.attr.mode = S_IRUGO | S_IWUSR;
2798 info->port_mtu_attr.store = set_port_ib_mtu;
2800 info->port_mtu_attr.show = show_port_ib_mtu;
2801 sysfs_attr_init(&info->port_mtu_attr.attr);
2803 err = device_create_file(&dev->persist->pdev->dev,
2804 &info->port_mtu_attr);
2806 mlx4_err(dev, "Failed to create mtu file for port %d\n", port);
2807 device_remove_file(&info->dev->persist->pdev->dev,
2815 static void mlx4_cleanup_port_info(struct mlx4_port_info *info)
2820 device_remove_file(&info->dev->persist->pdev->dev, &info->port_attr);
2821 device_remove_file(&info->dev->persist->pdev->dev,
2822 &info->port_mtu_attr);
2823 #ifdef CONFIG_RFS_ACCEL
2824 free_irq_cpu_rmap(info->rmap);
2829 static int mlx4_init_steering(struct mlx4_dev *dev)
2831 struct mlx4_priv *priv = mlx4_priv(dev);
2832 int num_entries = dev->caps.num_ports;
2835 priv->steer = kzalloc(sizeof(struct mlx4_steer) * num_entries, GFP_KERNEL);
2839 for (i = 0; i < num_entries; i++)
2840 for (j = 0; j < MLX4_NUM_STEERS; j++) {
2841 INIT_LIST_HEAD(&priv->steer[i].promisc_qps[j]);
2842 INIT_LIST_HEAD(&priv->steer[i].steer_entries[j]);
2847 static void mlx4_clear_steering(struct mlx4_dev *dev)
2849 struct mlx4_priv *priv = mlx4_priv(dev);
2850 struct mlx4_steer_index *entry, *tmp_entry;
2851 struct mlx4_promisc_qp *pqp, *tmp_pqp;
2852 int num_entries = dev->caps.num_ports;
2855 for (i = 0; i < num_entries; i++) {
2856 for (j = 0; j < MLX4_NUM_STEERS; j++) {
2857 list_for_each_entry_safe(pqp, tmp_pqp,
2858 &priv->steer[i].promisc_qps[j],
2860 list_del(&pqp->list);
2863 list_for_each_entry_safe(entry, tmp_entry,
2864 &priv->steer[i].steer_entries[j],
2866 list_del(&entry->list);
2867 list_for_each_entry_safe(pqp, tmp_pqp,
2870 list_del(&pqp->list);
2880 static int extended_func_num(struct pci_dev *pdev)
2882 return PCI_SLOT(pdev->devfn) * 8 + PCI_FUNC(pdev->devfn);
2885 #define MLX4_OWNER_BASE 0x8069c
2886 #define MLX4_OWNER_SIZE 4
2888 static int mlx4_get_ownership(struct mlx4_dev *dev)
2890 void __iomem *owner;
2893 if (pci_channel_offline(dev->persist->pdev))
2896 owner = ioremap(pci_resource_start(dev->persist->pdev, 0) +
2900 mlx4_err(dev, "Failed to obtain ownership bit\n");
2909 static void mlx4_free_ownership(struct mlx4_dev *dev)
2911 void __iomem *owner;
2913 if (pci_channel_offline(dev->persist->pdev))
2916 owner = ioremap(pci_resource_start(dev->persist->pdev, 0) +
2920 mlx4_err(dev, "Failed to obtain ownership bit\n");
2928 #define SRIOV_VALID_STATE(flags) (!!((flags) & MLX4_FLAG_SRIOV) ==\
2929 !!((flags) & MLX4_FLAG_MASTER))
2931 static u64 mlx4_enable_sriov(struct mlx4_dev *dev, struct pci_dev *pdev,
2932 u8 total_vfs, int existing_vfs, int reset_flow)
2934 u64 dev_flags = dev->flags;
2936 int fw_enabled_sriov_vfs = min(pci_sriov_get_totalvfs(pdev),
2940 dev->dev_vfs = kcalloc(total_vfs, sizeof(*dev->dev_vfs),
2947 atomic_inc(&pf_loading);
2948 if (dev->flags & MLX4_FLAG_SRIOV) {
2949 if (existing_vfs != total_vfs) {
2950 mlx4_err(dev, "SR-IOV was already enabled, but with num_vfs (%d) different than requested (%d)\n",
2951 existing_vfs, total_vfs);
2952 total_vfs = existing_vfs;
2956 dev->dev_vfs = kzalloc(total_vfs * sizeof(*dev->dev_vfs), GFP_KERNEL);
2957 if (NULL == dev->dev_vfs) {
2958 mlx4_err(dev, "Failed to allocate memory for VFs\n");
2962 if (!(dev->flags & MLX4_FLAG_SRIOV)) {
2963 if (total_vfs > fw_enabled_sriov_vfs) {
2964 mlx4_err(dev, "requested vfs (%d) > available vfs (%d). Continuing without SR_IOV\n",
2965 total_vfs, fw_enabled_sriov_vfs);
2969 mlx4_warn(dev, "Enabling SR-IOV with %d VFs\n", total_vfs);
2970 err = pci_enable_sriov(pdev, total_vfs);
2973 mlx4_err(dev, "Failed to enable SR-IOV, continuing without SR-IOV (err = %d)\n",
2977 mlx4_warn(dev, "Running in master mode\n");
2978 dev_flags |= MLX4_FLAG_SRIOV |
2980 dev_flags &= ~MLX4_FLAG_SLAVE;
2981 dev->persist->num_vfs = total_vfs;
2986 atomic_dec(&pf_loading);
2988 dev->persist->num_vfs = 0;
2989 kfree(dev->dev_vfs);
2990 dev->dev_vfs = NULL;
2991 return dev_flags & ~MLX4_FLAG_MASTER;
2995 MLX4_DEV_CAP_CHECK_NUM_VFS_ABOVE_64 = -1,
2998 static int mlx4_check_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap,
3001 int requested_vfs = nvfs[0] + nvfs[1] + nvfs[2];
3002 /* Checking for 64 VFs as a limitation of CX2 */
3003 if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_80_VFS) &&
3004 requested_vfs >= 64) {
3005 mlx4_err(dev, "Requested %d VFs, but FW does not support more than 64\n",
3007 return MLX4_DEV_CAP_CHECK_NUM_VFS_ABOVE_64;
3012 static int mlx4_load_one(struct pci_dev *pdev, int pci_dev_data,
3013 int total_vfs, int *nvfs, struct mlx4_priv *priv,
3016 struct mlx4_dev *dev;
3021 struct mlx4_dev_cap *dev_cap = NULL;
3022 int existing_vfs = 0;
3026 INIT_LIST_HEAD(&priv->ctx_list);
3027 spin_lock_init(&priv->ctx_lock);
3029 mutex_init(&priv->port_mutex);
3030 mutex_init(&priv->bond_mutex);
3032 INIT_LIST_HEAD(&priv->pgdir_list);
3033 mutex_init(&priv->pgdir_mutex);
3035 INIT_LIST_HEAD(&priv->bf_list);
3036 mutex_init(&priv->bf_mutex);
3038 dev->rev_id = pdev->revision;
3039 dev->numa_node = dev_to_node(&pdev->dev);
3041 /* Detect if this device is a virtual function */
3042 if (pci_dev_data & MLX4_PCI_DEV_IS_VF) {
3043 mlx4_warn(dev, "Detected virtual function - running in slave mode\n");
3044 dev->flags |= MLX4_FLAG_SLAVE;
3046 /* We reset the device and enable SRIOV only for physical
3047 * devices. Try to claim ownership on the device;
3048 * if already taken, skip -- do not allow multiple PFs */
3049 err = mlx4_get_ownership(dev);
3054 mlx4_warn(dev, "Multiple PFs not yet supported - Skipping PF\n");
3059 atomic_set(&priv->opreq_count, 0);
3060 INIT_WORK(&priv->opreq_task, mlx4_opreq_action);
3063 * Now reset the HCA before we touch the PCI capabilities or
3064 * attempt a firmware command, since a boot ROM may have left
3065 * the HCA in an undefined state.
3067 err = mlx4_reset(dev);
3069 mlx4_err(dev, "Failed to reset HCA, aborting\n");
3074 dev->flags = MLX4_FLAG_MASTER;
3075 existing_vfs = pci_num_vf(pdev);
3077 dev->flags |= MLX4_FLAG_SRIOV;
3078 dev->persist->num_vfs = total_vfs;
3082 /* on load remove any previous indication of internal error,
3085 dev->persist->state = MLX4_DEVICE_STATE_UP;
3088 err = mlx4_cmd_init(dev);
3090 mlx4_err(dev, "Failed to init command interface, aborting\n");
3094 /* In slave functions, the communication channel must be initialized
3095 * before posting commands. Also, init num_slaves before calling
3097 if (mlx4_is_mfunc(dev)) {
3098 if (mlx4_is_master(dev)) {
3099 dev->num_slaves = MLX4_MAX_NUM_SLAVES;
3102 dev->num_slaves = 0;
3103 err = mlx4_multi_func_init(dev);
3105 mlx4_err(dev, "Failed to init slave mfunc interface, aborting\n");
3111 err = mlx4_init_fw(dev);
3113 mlx4_err(dev, "Failed to init fw, aborting.\n");
3117 if (mlx4_is_master(dev)) {
3118 /* when we hit the goto slave_start below, dev_cap already initialized */
3120 dev_cap = kzalloc(sizeof(*dev_cap), GFP_KERNEL);
3127 err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
3129 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
3133 if (mlx4_check_dev_cap(dev, dev_cap, nvfs))
3136 if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS)) {
3137 u64 dev_flags = mlx4_enable_sriov(dev, pdev,
3143 mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3144 dev->flags = dev_flags;
3145 if (!SRIOV_VALID_STATE(dev->flags)) {
3146 mlx4_err(dev, "Invalid SRIOV state\n");
3149 err = mlx4_reset(dev);
3151 mlx4_err(dev, "Failed to reset HCA, aborting.\n");
3157 /* Legacy mode FW requires SRIOV to be enabled before
3158 * doing QUERY_DEV_CAP, since max_eq's value is different if
3161 memset(dev_cap, 0, sizeof(*dev_cap));
3162 err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
3164 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
3168 if (mlx4_check_dev_cap(dev, dev_cap, nvfs))
3173 err = mlx4_init_hca(dev);
3175 if (err == -EACCES) {
3176 /* Not primary Physical function
3177 * Running in slave mode */
3178 mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3179 /* We're not a PF */
3180 if (dev->flags & MLX4_FLAG_SRIOV) {
3182 pci_disable_sriov(pdev);
3183 if (mlx4_is_master(dev) && !reset_flow)
3184 atomic_dec(&pf_loading);
3185 dev->flags &= ~MLX4_FLAG_SRIOV;
3187 if (!mlx4_is_slave(dev))
3188 mlx4_free_ownership(dev);
3189 dev->flags |= MLX4_FLAG_SLAVE;
3190 dev->flags &= ~MLX4_FLAG_MASTER;
3196 if (mlx4_is_master(dev) && (dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS)) {
3197 u64 dev_flags = mlx4_enable_sriov(dev, pdev, total_vfs,
3198 existing_vfs, reset_flow);
3200 if ((dev->flags ^ dev_flags) & (MLX4_FLAG_MASTER | MLX4_FLAG_SLAVE)) {
3201 mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_VHCR);
3202 dev->flags = dev_flags;
3203 err = mlx4_cmd_init(dev);
3205 /* Only VHCR is cleaned up, so could still
3208 mlx4_err(dev, "Failed to init VHCR command interface, aborting\n");
3212 dev->flags = dev_flags;
3215 if (!SRIOV_VALID_STATE(dev->flags)) {
3216 mlx4_err(dev, "Invalid SRIOV state\n");
3221 /* check if the device is functioning at its maximum possible speed.
3222 * No return code for this call, just warn the user in case of PCI
3223 * express device capabilities are under-satisfied by the bus.
3225 if (!mlx4_is_slave(dev))
3226 mlx4_check_pcie_caps(dev);
3228 /* In master functions, the communication channel must be initialized
3229 * after obtaining its address from fw */
3230 if (mlx4_is_master(dev)) {
3231 if (dev->caps.num_ports < 2 &&
3235 "Error: Trying to configure VFs on port 2, but HCA has only %d physical ports\n",
3236 dev->caps.num_ports);
3239 memcpy(dev->persist->nvfs, nvfs, sizeof(dev->persist->nvfs));
3242 i < sizeof(dev->persist->nvfs)/
3243 sizeof(dev->persist->nvfs[0]); i++) {
3246 for (j = 0; j < dev->persist->nvfs[i]; ++sum, ++j) {
3247 dev->dev_vfs[sum].min_port = i < 2 ? i + 1 : 1;
3248 dev->dev_vfs[sum].n_ports = i < 2 ? 1 :
3249 dev->caps.num_ports;
3253 /* In master functions, the communication channel
3254 * must be initialized after obtaining its address from fw
3256 err = mlx4_multi_func_init(dev);
3258 mlx4_err(dev, "Failed to init master mfunc interface, aborting.\n");
3263 err = mlx4_alloc_eq_table(dev);
3265 goto err_master_mfunc;
3267 bitmap_zero(priv->msix_ctl.pool_bm, MAX_MSIX);
3268 mutex_init(&priv->msix_ctl.pool_lock);
3270 mlx4_enable_msi_x(dev);
3271 if ((mlx4_is_mfunc(dev)) &&
3272 !(dev->flags & MLX4_FLAG_MSI_X)) {
3274 mlx4_err(dev, "INTx is not supported in multi-function mode, aborting\n");
3278 if (!mlx4_is_slave(dev)) {
3279 err = mlx4_init_steering(dev);
3281 goto err_disable_msix;
3284 err = mlx4_setup_hca(dev);
3285 if (err == -EBUSY && (dev->flags & MLX4_FLAG_MSI_X) &&
3286 !mlx4_is_mfunc(dev)) {
3287 dev->flags &= ~MLX4_FLAG_MSI_X;
3288 dev->caps.num_comp_vectors = 1;
3289 pci_disable_msix(pdev);
3290 err = mlx4_setup_hca(dev);
3296 mlx4_init_quotas(dev);
3297 /* When PF resources are ready arm its comm channel to enable
3300 if (mlx4_is_master(dev)) {
3301 err = mlx4_ARM_COMM_CHANNEL(dev);
3303 mlx4_err(dev, " Failed to arm comm channel eq: %x\n",
3309 for (port = 1; port <= dev->caps.num_ports; port++) {
3310 err = mlx4_init_port_info(dev, port);
3315 priv->v2p.port1 = 1;
3316 priv->v2p.port2 = 2;
3318 err = mlx4_register_device(dev);
3322 mlx4_request_modules(dev);
3324 mlx4_sense_init(dev);
3325 mlx4_start_sense(dev);
3329 if (mlx4_is_master(dev) && dev->persist->num_vfs && !reset_flow)
3330 atomic_dec(&pf_loading);
3336 for (--port; port >= 1; --port)
3337 mlx4_cleanup_port_info(&priv->port[port]);
3339 mlx4_cleanup_default_counters(dev);
3340 if (!mlx4_is_slave(dev))
3341 mlx4_cleanup_counters_table(dev);
3342 mlx4_cleanup_qp_table(dev);
3343 mlx4_cleanup_srq_table(dev);
3344 mlx4_cleanup_cq_table(dev);
3345 mlx4_cmd_use_polling(dev);
3346 mlx4_cleanup_eq_table(dev);
3347 mlx4_cleanup_mcg_table(dev);
3348 mlx4_cleanup_mr_table(dev);
3349 mlx4_cleanup_xrcd_table(dev);
3350 mlx4_cleanup_pd_table(dev);
3351 mlx4_cleanup_uar_table(dev);
3354 if (!mlx4_is_slave(dev))
3355 mlx4_clear_steering(dev);
3358 if (dev->flags & MLX4_FLAG_MSI_X)
3359 pci_disable_msix(pdev);
3362 mlx4_free_eq_table(dev);
3365 if (mlx4_is_master(dev)) {
3366 mlx4_free_resource_tracker(dev, RES_TR_FREE_STRUCTS_ONLY);
3367 mlx4_multi_func_cleanup(dev);
3370 if (mlx4_is_slave(dev)) {
3371 kfree(dev->caps.qp0_qkey);
3372 kfree(dev->caps.qp0_tunnel);
3373 kfree(dev->caps.qp0_proxy);
3374 kfree(dev->caps.qp1_tunnel);
3375 kfree(dev->caps.qp1_proxy);
3379 mlx4_close_hca(dev);
3385 if (mlx4_is_slave(dev))
3386 mlx4_multi_func_cleanup(dev);
3389 mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3392 if (dev->flags & MLX4_FLAG_SRIOV && !existing_vfs) {
3393 pci_disable_sriov(pdev);
3394 dev->flags &= ~MLX4_FLAG_SRIOV;
3397 if (mlx4_is_master(dev) && dev->persist->num_vfs && !reset_flow)
3398 atomic_dec(&pf_loading);
3400 kfree(priv->dev.dev_vfs);
3402 if (!mlx4_is_slave(dev))
3403 mlx4_free_ownership(dev);
3409 static int __mlx4_init_one(struct pci_dev *pdev, int pci_dev_data,
3410 struct mlx4_priv *priv)
3413 int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
3414 int prb_vf[MLX4_MAX_PORTS + 1] = {0, 0, 0};
3415 const int param_map[MLX4_MAX_PORTS + 1][MLX4_MAX_PORTS + 1] = {
3416 {2, 0, 0}, {0, 1, 2}, {0, 1, 2} };
3417 unsigned total_vfs = 0;
3420 pr_info(DRV_NAME ": Initializing %s\n", pci_name(pdev));
3422 err = pci_enable_device(pdev);
3424 dev_err(&pdev->dev, "Cannot enable PCI device, aborting\n");
3428 /* Due to requirement that all VFs and the PF are *guaranteed* 2 MACS
3429 * per port, we must limit the number of VFs to 63 (since their are
3432 for (i = 0; i < sizeof(nvfs)/sizeof(nvfs[0]) && i < num_vfs_argc;
3433 total_vfs += nvfs[param_map[num_vfs_argc - 1][i]], i++) {
3434 nvfs[param_map[num_vfs_argc - 1][i]] = num_vfs[i];
3436 dev_err(&pdev->dev, "num_vfs module parameter cannot be negative\n");
3438 goto err_disable_pdev;
3441 for (i = 0; i < sizeof(prb_vf)/sizeof(prb_vf[0]) && i < probe_vfs_argc;
3443 prb_vf[param_map[probe_vfs_argc - 1][i]] = probe_vf[i];
3444 if (prb_vf[i] < 0 || prb_vf[i] > nvfs[i]) {
3445 dev_err(&pdev->dev, "probe_vf module parameter cannot be negative or greater than num_vfs\n");
3447 goto err_disable_pdev;
3450 if (total_vfs > MLX4_MAX_NUM_VF) {
3452 "Requested more VF's (%d) than allowed by hw (%d)\n",
3453 total_vfs, MLX4_MAX_NUM_VF);
3455 goto err_disable_pdev;
3458 for (i = 0; i < MLX4_MAX_PORTS; i++) {
3459 if (nvfs[i] + nvfs[2] > MLX4_MAX_NUM_VF_P_PORT) {
3461 "Requested more VF's (%d) for port (%d) than allowed by driver (%d)\n",
3462 nvfs[i] + nvfs[2], i + 1,
3463 MLX4_MAX_NUM_VF_P_PORT);
3465 goto err_disable_pdev;
3469 /* Check for BARs. */
3470 if (!(pci_dev_data & MLX4_PCI_DEV_IS_VF) &&
3471 !(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) {
3472 dev_err(&pdev->dev, "Missing DCS, aborting (driver_data: 0x%x, pci_resource_flags(pdev, 0):0x%lx)\n",
3473 pci_dev_data, pci_resource_flags(pdev, 0));
3475 goto err_disable_pdev;
3477 if (!(pci_resource_flags(pdev, 2) & IORESOURCE_MEM)) {
3478 dev_err(&pdev->dev, "Missing UAR, aborting\n");
3480 goto err_disable_pdev;
3483 err = pci_request_regions(pdev, DRV_NAME);
3485 dev_err(&pdev->dev, "Couldn't get PCI resources, aborting\n");
3486 goto err_disable_pdev;
3489 pci_set_master(pdev);
3491 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
3493 dev_warn(&pdev->dev, "Warning: couldn't set 64-bit PCI DMA mask\n");
3494 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
3496 dev_err(&pdev->dev, "Can't set PCI DMA mask, aborting\n");
3497 goto err_release_regions;
3500 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
3502 dev_warn(&pdev->dev, "Warning: couldn't set 64-bit consistent PCI DMA mask\n");
3503 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
3505 dev_err(&pdev->dev, "Can't set consistent PCI DMA mask, aborting\n");
3506 goto err_release_regions;
3510 /* Allow large DMA segments, up to the firmware limit of 1 GB */
3511 dma_set_max_seg_size(&pdev->dev, 1024 * 1024 * 1024);
3512 /* Detect if this device is a virtual function */
3513 if (pci_dev_data & MLX4_PCI_DEV_IS_VF) {
3514 /* When acting as pf, we normally skip vfs unless explicitly
3515 * requested to probe them.
3518 unsigned vfs_offset = 0;
3520 for (i = 0; i < sizeof(nvfs)/sizeof(nvfs[0]) &&
3521 vfs_offset + nvfs[i] < extended_func_num(pdev);
3522 vfs_offset += nvfs[i], i++)
3524 if (i == sizeof(nvfs)/sizeof(nvfs[0])) {
3526 goto err_release_regions;
3528 if ((extended_func_num(pdev) - vfs_offset)
3530 dev_warn(&pdev->dev, "Skipping virtual function:%d\n",
3531 extended_func_num(pdev));
3533 goto err_release_regions;
3538 err = mlx4_catas_init(&priv->dev);
3540 goto err_release_regions;
3542 err = mlx4_load_one(pdev, pci_dev_data, total_vfs, nvfs, priv, 0);
3549 mlx4_catas_end(&priv->dev);
3551 err_release_regions:
3552 pci_release_regions(pdev);
3555 pci_disable_device(pdev);
3556 pci_set_drvdata(pdev, NULL);
3560 static int mlx4_init_one(struct pci_dev *pdev, const struct pci_device_id *id)
3562 struct mlx4_priv *priv;
3563 struct mlx4_dev *dev;
3566 printk_once(KERN_INFO "%s", mlx4_version);
3568 priv = kzalloc(sizeof(*priv), GFP_KERNEL);
3573 dev->persist = kzalloc(sizeof(*dev->persist), GFP_KERNEL);
3574 if (!dev->persist) {
3578 dev->persist->pdev = pdev;
3579 dev->persist->dev = dev;
3580 pci_set_drvdata(pdev, dev->persist);
3581 priv->pci_dev_data = id->driver_data;
3582 mutex_init(&dev->persist->device_state_mutex);
3583 mutex_init(&dev->persist->interface_state_mutex);
3585 ret = __mlx4_init_one(pdev, id->driver_data, priv);
3587 kfree(dev->persist);
3590 pci_save_state(pdev);
3596 static void mlx4_clean_dev(struct mlx4_dev *dev)
3598 struct mlx4_dev_persistent *persist = dev->persist;
3599 struct mlx4_priv *priv = mlx4_priv(dev);
3600 unsigned long flags = (dev->flags & RESET_PERSIST_MASK_FLAGS);
3602 memset(priv, 0, sizeof(*priv));
3603 priv->dev.persist = persist;
3604 priv->dev.flags = flags;
3607 static void mlx4_unload_one(struct pci_dev *pdev)
3609 struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
3610 struct mlx4_dev *dev = persist->dev;
3611 struct mlx4_priv *priv = mlx4_priv(dev);
3618 /* saving current ports type for further use */
3619 for (i = 0; i < dev->caps.num_ports; i++) {
3620 dev->persist->curr_port_type[i] = dev->caps.port_type[i + 1];
3621 dev->persist->curr_port_poss_type[i] = dev->caps.
3622 possible_type[i + 1];
3625 pci_dev_data = priv->pci_dev_data;
3627 mlx4_stop_sense(dev);
3628 mlx4_unregister_device(dev);
3630 for (p = 1; p <= dev->caps.num_ports; p++) {
3631 mlx4_cleanup_port_info(&priv->port[p]);
3632 mlx4_CLOSE_PORT(dev, p);
3635 if (mlx4_is_master(dev))
3636 mlx4_free_resource_tracker(dev,
3637 RES_TR_FREE_SLAVES_ONLY);
3639 mlx4_cleanup_default_counters(dev);
3640 if (!mlx4_is_slave(dev))
3641 mlx4_cleanup_counters_table(dev);
3642 mlx4_cleanup_qp_table(dev);
3643 mlx4_cleanup_srq_table(dev);
3644 mlx4_cleanup_cq_table(dev);
3645 mlx4_cmd_use_polling(dev);
3646 mlx4_cleanup_eq_table(dev);
3647 mlx4_cleanup_mcg_table(dev);
3648 mlx4_cleanup_mr_table(dev);
3649 mlx4_cleanup_xrcd_table(dev);
3650 mlx4_cleanup_pd_table(dev);
3652 if (mlx4_is_master(dev))
3653 mlx4_free_resource_tracker(dev,
3654 RES_TR_FREE_STRUCTS_ONLY);
3657 mlx4_uar_free(dev, &priv->driver_uar);
3658 mlx4_cleanup_uar_table(dev);
3659 if (!mlx4_is_slave(dev))
3660 mlx4_clear_steering(dev);
3661 mlx4_free_eq_table(dev);
3662 if (mlx4_is_master(dev))
3663 mlx4_multi_func_cleanup(dev);
3664 mlx4_close_hca(dev);
3666 if (mlx4_is_slave(dev))
3667 mlx4_multi_func_cleanup(dev);
3668 mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3670 if (dev->flags & MLX4_FLAG_MSI_X)
3671 pci_disable_msix(pdev);
3673 if (!mlx4_is_slave(dev))
3674 mlx4_free_ownership(dev);
3676 kfree(dev->caps.qp0_qkey);
3677 kfree(dev->caps.qp0_tunnel);
3678 kfree(dev->caps.qp0_proxy);
3679 kfree(dev->caps.qp1_tunnel);
3680 kfree(dev->caps.qp1_proxy);
3681 kfree(dev->dev_vfs);
3683 mlx4_clean_dev(dev);
3684 priv->pci_dev_data = pci_dev_data;
3688 static void mlx4_remove_one(struct pci_dev *pdev)
3690 struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
3691 struct mlx4_dev *dev = persist->dev;
3692 struct mlx4_priv *priv = mlx4_priv(dev);
3695 mutex_lock(&persist->interface_state_mutex);
3696 persist->interface_state |= MLX4_INTERFACE_STATE_DELETION;
3697 mutex_unlock(&persist->interface_state_mutex);
3699 /* Disabling SR-IOV is not allowed while there are active vf's */
3700 if (mlx4_is_master(dev) && dev->flags & MLX4_FLAG_SRIOV) {
3701 active_vfs = mlx4_how_many_lives_vf(dev);
3703 pr_warn("Removing PF when there are active VF's !!\n");
3704 pr_warn("Will not disable SR-IOV.\n");
3708 /* device marked to be under deletion running now without the lock
3709 * letting other tasks to be terminated
3711 if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
3712 mlx4_unload_one(pdev);
3714 mlx4_info(dev, "%s: interface is down\n", __func__);
3715 mlx4_catas_end(dev);
3716 if (dev->flags & MLX4_FLAG_SRIOV && !active_vfs) {
3717 mlx4_warn(dev, "Disabling SR-IOV\n");
3718 pci_disable_sriov(pdev);
3721 pci_release_regions(pdev);
3722 pci_disable_device(pdev);
3723 kfree(dev->persist);
3725 pci_set_drvdata(pdev, NULL);
3728 static int restore_current_port_types(struct mlx4_dev *dev,
3729 enum mlx4_port_type *types,
3730 enum mlx4_port_type *poss_types)
3732 struct mlx4_priv *priv = mlx4_priv(dev);
3735 mlx4_stop_sense(dev);
3737 mutex_lock(&priv->port_mutex);
3738 for (i = 0; i < dev->caps.num_ports; i++)
3739 dev->caps.possible_type[i + 1] = poss_types[i];
3740 err = mlx4_change_port_types(dev, types);
3741 mlx4_start_sense(dev);
3742 mutex_unlock(&priv->port_mutex);
3747 int mlx4_restart_one(struct pci_dev *pdev)
3749 struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
3750 struct mlx4_dev *dev = persist->dev;
3751 struct mlx4_priv *priv = mlx4_priv(dev);
3752 int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
3753 int pci_dev_data, err, total_vfs;
3755 pci_dev_data = priv->pci_dev_data;
3756 total_vfs = dev->persist->num_vfs;
3757 memcpy(nvfs, dev->persist->nvfs, sizeof(dev->persist->nvfs));
3759 mlx4_unload_one(pdev);
3760 err = mlx4_load_one(pdev, pci_dev_data, total_vfs, nvfs, priv, 1);
3762 mlx4_err(dev, "%s: ERROR: mlx4_load_one failed, pci_name=%s, err=%d\n",
3763 __func__, pci_name(pdev), err);
3767 err = restore_current_port_types(dev, dev->persist->curr_port_type,
3768 dev->persist->curr_port_poss_type);
3770 mlx4_err(dev, "could not restore original port types (%d)\n",
3776 static const struct pci_device_id mlx4_pci_table[] = {
3777 /* MT25408 "Hermon" SDR */
3778 { PCI_VDEVICE(MELLANOX, 0x6340), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3779 /* MT25408 "Hermon" DDR */
3780 { PCI_VDEVICE(MELLANOX, 0x634a), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3781 /* MT25408 "Hermon" QDR */
3782 { PCI_VDEVICE(MELLANOX, 0x6354), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3783 /* MT25408 "Hermon" DDR PCIe gen2 */
3784 { PCI_VDEVICE(MELLANOX, 0x6732), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3785 /* MT25408 "Hermon" QDR PCIe gen2 */
3786 { PCI_VDEVICE(MELLANOX, 0x673c), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3787 /* MT25408 "Hermon" EN 10GigE */
3788 { PCI_VDEVICE(MELLANOX, 0x6368), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3789 /* MT25408 "Hermon" EN 10GigE PCIe gen2 */
3790 { PCI_VDEVICE(MELLANOX, 0x6750), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3791 /* MT25458 ConnectX EN 10GBASE-T 10GigE */
3792 { PCI_VDEVICE(MELLANOX, 0x6372), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3793 /* MT25458 ConnectX EN 10GBASE-T+Gen2 10GigE */
3794 { PCI_VDEVICE(MELLANOX, 0x675a), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3795 /* MT26468 ConnectX EN 10GigE PCIe gen2*/
3796 { PCI_VDEVICE(MELLANOX, 0x6764), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3797 /* MT26438 ConnectX EN 40GigE PCIe gen2 5GT/s */
3798 { PCI_VDEVICE(MELLANOX, 0x6746), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3799 /* MT26478 ConnectX2 40GigE PCIe gen2 */
3800 { PCI_VDEVICE(MELLANOX, 0x676e), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3801 /* MT25400 Family [ConnectX-2 Virtual Function] */
3802 { PCI_VDEVICE(MELLANOX, 0x1002), MLX4_PCI_DEV_IS_VF },
3803 /* MT27500 Family [ConnectX-3] */
3804 { PCI_VDEVICE(MELLANOX, 0x1003), 0 },
3805 /* MT27500 Family [ConnectX-3 Virtual Function] */
3806 { PCI_VDEVICE(MELLANOX, 0x1004), MLX4_PCI_DEV_IS_VF },
3807 { PCI_VDEVICE(MELLANOX, 0x1005), 0 }, /* MT27510 Family */
3808 { PCI_VDEVICE(MELLANOX, 0x1006), 0 }, /* MT27511 Family */
3809 { PCI_VDEVICE(MELLANOX, 0x1007), 0 }, /* MT27520 Family */
3810 { PCI_VDEVICE(MELLANOX, 0x1008), 0 }, /* MT27521 Family */
3811 { PCI_VDEVICE(MELLANOX, 0x1009), 0 }, /* MT27530 Family */
3812 { PCI_VDEVICE(MELLANOX, 0x100a), 0 }, /* MT27531 Family */
3813 { PCI_VDEVICE(MELLANOX, 0x100b), 0 }, /* MT27540 Family */
3814 { PCI_VDEVICE(MELLANOX, 0x100c), 0 }, /* MT27541 Family */
3815 { PCI_VDEVICE(MELLANOX, 0x100d), 0 }, /* MT27550 Family */
3816 { PCI_VDEVICE(MELLANOX, 0x100e), 0 }, /* MT27551 Family */
3817 { PCI_VDEVICE(MELLANOX, 0x100f), 0 }, /* MT27560 Family */
3818 { PCI_VDEVICE(MELLANOX, 0x1010), 0 }, /* MT27561 Family */
3822 MODULE_DEVICE_TABLE(pci, mlx4_pci_table);
3824 static pci_ers_result_t mlx4_pci_err_detected(struct pci_dev *pdev,
3825 pci_channel_state_t state)
3827 struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
3829 mlx4_err(persist->dev, "mlx4_pci_err_detected was called\n");
3830 mlx4_enter_error_state(persist);
3832 mutex_lock(&persist->interface_state_mutex);
3833 if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
3834 mlx4_unload_one(pdev);
3836 mutex_unlock(&persist->interface_state_mutex);
3837 if (state == pci_channel_io_perm_failure)
3838 return PCI_ERS_RESULT_DISCONNECT;
3840 pci_disable_device(pdev);
3841 return PCI_ERS_RESULT_NEED_RESET;
3844 static pci_ers_result_t mlx4_pci_slot_reset(struct pci_dev *pdev)
3846 struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
3847 struct mlx4_dev *dev = persist->dev;
3848 struct mlx4_priv *priv = mlx4_priv(dev);
3850 int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
3853 mlx4_err(dev, "mlx4_pci_slot_reset was called\n");
3854 ret = pci_enable_device(pdev);
3856 mlx4_err(dev, "Can not re-enable device, ret=%d\n", ret);
3857 return PCI_ERS_RESULT_DISCONNECT;
3860 pci_set_master(pdev);
3861 pci_restore_state(pdev);
3862 pci_save_state(pdev);
3864 total_vfs = dev->persist->num_vfs;
3865 memcpy(nvfs, dev->persist->nvfs, sizeof(dev->persist->nvfs));
3867 mutex_lock(&persist->interface_state_mutex);
3868 if (!(persist->interface_state & MLX4_INTERFACE_STATE_UP)) {
3869 ret = mlx4_load_one(pdev, priv->pci_dev_data, total_vfs, nvfs,
3872 mlx4_err(dev, "%s: mlx4_load_one failed, ret=%d\n",
3877 ret = restore_current_port_types(dev, dev->persist->
3878 curr_port_type, dev->persist->
3879 curr_port_poss_type);
3881 mlx4_err(dev, "could not restore original port types (%d)\n", ret);
3884 mutex_unlock(&persist->interface_state_mutex);
3886 return ret ? PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_RECOVERED;
3889 static void mlx4_shutdown(struct pci_dev *pdev)
3891 struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
3893 mlx4_info(persist->dev, "mlx4_shutdown was called\n");
3894 mutex_lock(&persist->interface_state_mutex);
3895 if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
3896 mlx4_unload_one(pdev);
3897 mutex_unlock(&persist->interface_state_mutex);
3900 static const struct pci_error_handlers mlx4_err_handler = {
3901 .error_detected = mlx4_pci_err_detected,
3902 .slot_reset = mlx4_pci_slot_reset,
3905 static struct pci_driver mlx4_driver = {
3907 .id_table = mlx4_pci_table,
3908 .probe = mlx4_init_one,
3909 .shutdown = mlx4_shutdown,
3910 .remove = mlx4_remove_one,
3911 .err_handler = &mlx4_err_handler,
3914 static int __init mlx4_verify_params(void)
3916 if ((log_num_mac < 0) || (log_num_mac > 7)) {
3917 pr_warn("mlx4_core: bad num_mac: %d\n", log_num_mac);
3921 if (log_num_vlan != 0)
3922 pr_warn("mlx4_core: log_num_vlan - obsolete module param, using %d\n",
3923 MLX4_LOG_NUM_VLANS);
3926 pr_warn("mlx4_core: use_prio - obsolete module param, ignored\n");
3928 if ((log_mtts_per_seg < 1) || (log_mtts_per_seg > 7)) {
3929 pr_warn("mlx4_core: bad log_mtts_per_seg: %d\n",
3934 /* Check if module param for ports type has legal combination */
3935 if (port_type_array[0] == false && port_type_array[1] == true) {
3936 pr_warn("Module parameter configuration ETH/IB is not supported. Switching to default configuration IB/IB\n");
3937 port_type_array[0] = true;
3940 if (mlx4_log_num_mgm_entry_size < -7 ||
3941 (mlx4_log_num_mgm_entry_size > 0 &&
3942 (mlx4_log_num_mgm_entry_size < MLX4_MIN_MGM_LOG_ENTRY_SIZE ||
3943 mlx4_log_num_mgm_entry_size > MLX4_MAX_MGM_LOG_ENTRY_SIZE))) {
3944 pr_warn("mlx4_core: mlx4_log_num_mgm_entry_size (%d) not in legal range (-7..0 or %d..%d)\n",
3945 mlx4_log_num_mgm_entry_size,
3946 MLX4_MIN_MGM_LOG_ENTRY_SIZE,
3947 MLX4_MAX_MGM_LOG_ENTRY_SIZE);
3954 static int __init mlx4_init(void)
3958 if (mlx4_verify_params())
3962 mlx4_wq = create_singlethread_workqueue("mlx4");
3966 ret = pci_register_driver(&mlx4_driver);
3968 destroy_workqueue(mlx4_wq);
3969 return ret < 0 ? ret : 0;
3972 static void __exit mlx4_cleanup(void)
3974 pci_unregister_driver(&mlx4_driver);
3975 destroy_workqueue(mlx4_wq);
3978 module_init(mlx4_init);
3979 module_exit(mlx4_cleanup);