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 = 3;
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 = 3;
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
109 static char mlx4_version[] =
110 DRV_NAME ": Mellanox ConnectX core driver v"
111 DRV_VERSION " (" DRV_RELDATE ")\n";
113 static struct mlx4_profile default_profile = {
116 .rdmarc_per_qp = 1 << 4,
120 .num_mtt = 1 << 20, /* It is really num mtt segements */
123 static struct mlx4_profile low_mem_profile = {
126 .rdmarc_per_qp = 1 << 4,
133 static int log_num_mac = 7;
134 module_param_named(log_num_mac, log_num_mac, int, 0444);
135 MODULE_PARM_DESC(log_num_mac, "Log2 max number of MACs per ETH port (1-7)");
137 static int log_num_vlan;
138 module_param_named(log_num_vlan, log_num_vlan, int, 0444);
139 MODULE_PARM_DESC(log_num_vlan, "Log2 max number of VLANs per ETH port (0-7)");
140 /* Log2 max number of VLANs per ETH port (0-7) */
141 #define MLX4_LOG_NUM_VLANS 7
142 #define MLX4_MIN_LOG_NUM_VLANS 0
143 #define MLX4_MIN_LOG_NUM_MAC 1
145 static bool use_prio;
146 module_param_named(use_prio, use_prio, bool, 0444);
147 MODULE_PARM_DESC(use_prio, "Enable steering by VLAN priority on ETH ports (deprecated)");
149 int log_mtts_per_seg = ilog2(MLX4_MTT_ENTRY_PER_SEG);
150 module_param_named(log_mtts_per_seg, log_mtts_per_seg, int, 0444);
151 MODULE_PARM_DESC(log_mtts_per_seg, "Log2 number of MTT entries per segment (1-7)");
153 static int port_type_array[2] = {MLX4_PORT_TYPE_NONE, MLX4_PORT_TYPE_NONE};
154 static int arr_argc = 2;
155 module_param_array(port_type_array, int, &arr_argc, 0444);
156 MODULE_PARM_DESC(port_type_array, "Array of port types: HW_DEFAULT (0) is default "
157 "1 for IB, 2 for Ethernet");
159 struct mlx4_port_config {
160 struct list_head list;
161 enum mlx4_port_type port_type[MLX4_MAX_PORTS + 1];
162 struct pci_dev *pdev;
165 static atomic_t pf_loading = ATOMIC_INIT(0);
167 int mlx4_check_port_params(struct mlx4_dev *dev,
168 enum mlx4_port_type *port_type)
172 for (i = 0; i < dev->caps.num_ports - 1; i++) {
173 if (port_type[i] != port_type[i + 1]) {
174 if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP)) {
175 mlx4_err(dev, "Only same port types supported on this HCA, aborting\n");
181 for (i = 0; i < dev->caps.num_ports; i++) {
182 if (!(port_type[i] & dev->caps.supported_type[i+1])) {
183 mlx4_err(dev, "Requested port type for port %d is not supported on this HCA\n",
191 static void mlx4_set_port_mask(struct mlx4_dev *dev)
195 for (i = 1; i <= dev->caps.num_ports; ++i)
196 dev->caps.port_mask[i] = dev->caps.port_type[i];
199 static int mlx4_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
204 err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
206 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
210 if (dev_cap->min_page_sz > PAGE_SIZE) {
211 mlx4_err(dev, "HCA minimum page size of %d bigger than kernel PAGE_SIZE of %ld, aborting\n",
212 dev_cap->min_page_sz, PAGE_SIZE);
215 if (dev_cap->num_ports > MLX4_MAX_PORTS) {
216 mlx4_err(dev, "HCA has %d ports, but we only support %d, aborting\n",
217 dev_cap->num_ports, MLX4_MAX_PORTS);
221 if (dev_cap->uar_size > pci_resource_len(dev->pdev, 2)) {
222 mlx4_err(dev, "HCA reported UAR size of 0x%x bigger than PCI resource 2 size of 0x%llx, aborting\n",
224 (unsigned long long) pci_resource_len(dev->pdev, 2));
228 dev->caps.num_ports = dev_cap->num_ports;
229 dev->phys_caps.num_phys_eqs = MLX4_MAX_EQ_NUM;
230 for (i = 1; i <= dev->caps.num_ports; ++i) {
231 dev->caps.vl_cap[i] = dev_cap->max_vl[i];
232 dev->caps.ib_mtu_cap[i] = dev_cap->ib_mtu[i];
233 dev->phys_caps.gid_phys_table_len[i] = dev_cap->max_gids[i];
234 dev->phys_caps.pkey_phys_table_len[i] = dev_cap->max_pkeys[i];
235 /* set gid and pkey table operating lengths by default
236 * to non-sriov values */
237 dev->caps.gid_table_len[i] = dev_cap->max_gids[i];
238 dev->caps.pkey_table_len[i] = dev_cap->max_pkeys[i];
239 dev->caps.port_width_cap[i] = dev_cap->max_port_width[i];
240 dev->caps.eth_mtu_cap[i] = dev_cap->eth_mtu[i];
241 dev->caps.def_mac[i] = dev_cap->def_mac[i];
242 dev->caps.supported_type[i] = dev_cap->supported_port_types[i];
243 dev->caps.suggested_type[i] = dev_cap->suggested_type[i];
244 dev->caps.default_sense[i] = dev_cap->default_sense[i];
245 dev->caps.trans_type[i] = dev_cap->trans_type[i];
246 dev->caps.vendor_oui[i] = dev_cap->vendor_oui[i];
247 dev->caps.wavelength[i] = dev_cap->wavelength[i];
248 dev->caps.trans_code[i] = dev_cap->trans_code[i];
251 dev->caps.uar_page_size = PAGE_SIZE;
252 dev->caps.num_uars = dev_cap->uar_size / PAGE_SIZE;
253 dev->caps.local_ca_ack_delay = dev_cap->local_ca_ack_delay;
254 dev->caps.bf_reg_size = dev_cap->bf_reg_size;
255 dev->caps.bf_regs_per_page = dev_cap->bf_regs_per_page;
256 dev->caps.max_sq_sg = dev_cap->max_sq_sg;
257 dev->caps.max_rq_sg = dev_cap->max_rq_sg;
258 dev->caps.max_wqes = dev_cap->max_qp_sz;
259 dev->caps.max_qp_init_rdma = dev_cap->max_requester_per_qp;
260 dev->caps.max_srq_wqes = dev_cap->max_srq_sz;
261 dev->caps.max_srq_sge = dev_cap->max_rq_sg - 1;
262 dev->caps.reserved_srqs = dev_cap->reserved_srqs;
263 dev->caps.max_sq_desc_sz = dev_cap->max_sq_desc_sz;
264 dev->caps.max_rq_desc_sz = dev_cap->max_rq_desc_sz;
266 * Subtract 1 from the limit because we need to allocate a
267 * spare CQE so the HCA HW can tell the difference between an
268 * empty CQ and a full CQ.
270 dev->caps.max_cqes = dev_cap->max_cq_sz - 1;
271 dev->caps.reserved_cqs = dev_cap->reserved_cqs;
272 dev->caps.reserved_eqs = dev_cap->reserved_eqs;
273 dev->caps.reserved_mtts = dev_cap->reserved_mtts;
274 dev->caps.reserved_mrws = dev_cap->reserved_mrws;
276 /* The first 128 UARs are used for EQ doorbells */
277 dev->caps.reserved_uars = max_t(int, 128, dev_cap->reserved_uars);
278 dev->caps.reserved_pds = dev_cap->reserved_pds;
279 dev->caps.reserved_xrcds = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
280 dev_cap->reserved_xrcds : 0;
281 dev->caps.max_xrcds = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
282 dev_cap->max_xrcds : 0;
283 dev->caps.mtt_entry_sz = dev_cap->mtt_entry_sz;
285 dev->caps.max_msg_sz = dev_cap->max_msg_sz;
286 dev->caps.page_size_cap = ~(u32) (dev_cap->min_page_sz - 1);
287 dev->caps.flags = dev_cap->flags;
288 dev->caps.flags2 = dev_cap->flags2;
289 dev->caps.bmme_flags = dev_cap->bmme_flags;
290 dev->caps.reserved_lkey = dev_cap->reserved_lkey;
291 dev->caps.stat_rate_support = dev_cap->stat_rate_support;
292 dev->caps.max_gso_sz = dev_cap->max_gso_sz;
293 dev->caps.max_rss_tbl_sz = dev_cap->max_rss_tbl_sz;
295 /* Sense port always allowed on supported devices for ConnectX-1 and -2 */
296 if (mlx4_priv(dev)->pci_dev_data & MLX4_PCI_DEV_FORCE_SENSE_PORT)
297 dev->caps.flags |= MLX4_DEV_CAP_FLAG_SENSE_SUPPORT;
298 /* Don't do sense port on multifunction devices (for now at least) */
299 if (mlx4_is_mfunc(dev))
300 dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_SENSE_SUPPORT;
302 if (mlx4_low_memory_profile()) {
303 dev->caps.log_num_macs = MLX4_MIN_LOG_NUM_MAC;
304 dev->caps.log_num_vlans = MLX4_MIN_LOG_NUM_VLANS;
306 dev->caps.log_num_macs = log_num_mac;
307 dev->caps.log_num_vlans = MLX4_LOG_NUM_VLANS;
310 for (i = 1; i <= dev->caps.num_ports; ++i) {
311 dev->caps.port_type[i] = MLX4_PORT_TYPE_NONE;
312 if (dev->caps.supported_type[i]) {
313 /* if only ETH is supported - assign ETH */
314 if (dev->caps.supported_type[i] == MLX4_PORT_TYPE_ETH)
315 dev->caps.port_type[i] = MLX4_PORT_TYPE_ETH;
316 /* if only IB is supported, assign IB */
317 else if (dev->caps.supported_type[i] ==
319 dev->caps.port_type[i] = MLX4_PORT_TYPE_IB;
321 /* if IB and ETH are supported, we set the port
322 * type according to user selection of port type;
323 * if user selected none, take the FW hint */
324 if (port_type_array[i - 1] == MLX4_PORT_TYPE_NONE)
325 dev->caps.port_type[i] = dev->caps.suggested_type[i] ?
326 MLX4_PORT_TYPE_ETH : MLX4_PORT_TYPE_IB;
328 dev->caps.port_type[i] = port_type_array[i - 1];
332 * Link sensing is allowed on the port if 3 conditions are true:
333 * 1. Both protocols are supported on the port.
334 * 2. Different types are supported on the port
335 * 3. FW declared that it supports link sensing
337 mlx4_priv(dev)->sense.sense_allowed[i] =
338 ((dev->caps.supported_type[i] == MLX4_PORT_TYPE_AUTO) &&
339 (dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
340 (dev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT));
343 * If "default_sense" bit is set, we move the port to "AUTO" mode
344 * and perform sense_port FW command to try and set the correct
345 * port type from beginning
347 if (mlx4_priv(dev)->sense.sense_allowed[i] && dev->caps.default_sense[i]) {
348 enum mlx4_port_type sensed_port = MLX4_PORT_TYPE_NONE;
349 dev->caps.possible_type[i] = MLX4_PORT_TYPE_AUTO;
350 mlx4_SENSE_PORT(dev, i, &sensed_port);
351 if (sensed_port != MLX4_PORT_TYPE_NONE)
352 dev->caps.port_type[i] = sensed_port;
354 dev->caps.possible_type[i] = dev->caps.port_type[i];
357 if (dev->caps.log_num_macs > dev_cap->log_max_macs[i]) {
358 dev->caps.log_num_macs = dev_cap->log_max_macs[i];
359 mlx4_warn(dev, "Requested number of MACs is too much for port %d, reducing to %d\n",
360 i, 1 << dev->caps.log_num_macs);
362 if (dev->caps.log_num_vlans > dev_cap->log_max_vlans[i]) {
363 dev->caps.log_num_vlans = dev_cap->log_max_vlans[i];
364 mlx4_warn(dev, "Requested number of VLANs is too much for port %d, reducing to %d\n",
365 i, 1 << dev->caps.log_num_vlans);
369 dev->caps.max_counters = 1 << ilog2(dev_cap->max_counters);
371 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] = dev_cap->reserved_qps;
372 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] =
373 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] =
374 (1 << dev->caps.log_num_macs) *
375 (1 << dev->caps.log_num_vlans) *
377 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH] = MLX4_NUM_FEXCH;
379 dev->caps.reserved_qps = dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] +
380 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] +
381 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] +
382 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH];
384 dev->caps.sqp_demux = (mlx4_is_master(dev)) ? MLX4_MAX_NUM_SLAVES : 0;
386 if (!enable_64b_cqe_eqe && !mlx4_is_slave(dev)) {
388 (MLX4_DEV_CAP_FLAG_64B_CQE | MLX4_DEV_CAP_FLAG_64B_EQE)) {
389 mlx4_warn(dev, "64B EQEs/CQEs supported by the device but not enabled\n");
390 dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_64B_CQE;
391 dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_64B_EQE;
395 if ((dev->caps.flags &
396 (MLX4_DEV_CAP_FLAG_64B_CQE | MLX4_DEV_CAP_FLAG_64B_EQE)) &&
398 dev->caps.function_caps |= MLX4_FUNC_CAP_64B_EQE_CQE;
403 static int mlx4_get_pcie_dev_link_caps(struct mlx4_dev *dev,
404 enum pci_bus_speed *speed,
405 enum pcie_link_width *width)
407 u32 lnkcap1, lnkcap2;
410 #define PCIE_MLW_CAP_SHIFT 4 /* start of MLW mask in link capabilities */
412 *speed = PCI_SPEED_UNKNOWN;
413 *width = PCIE_LNK_WIDTH_UNKNOWN;
415 err1 = pcie_capability_read_dword(dev->pdev, PCI_EXP_LNKCAP, &lnkcap1);
416 err2 = pcie_capability_read_dword(dev->pdev, PCI_EXP_LNKCAP2, &lnkcap2);
417 if (!err2 && lnkcap2) { /* PCIe r3.0-compliant */
418 if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_8_0GB)
419 *speed = PCIE_SPEED_8_0GT;
420 else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_5_0GB)
421 *speed = PCIE_SPEED_5_0GT;
422 else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_2_5GB)
423 *speed = PCIE_SPEED_2_5GT;
426 *width = (lnkcap1 & PCI_EXP_LNKCAP_MLW) >> PCIE_MLW_CAP_SHIFT;
427 if (!lnkcap2) { /* pre-r3.0 */
428 if (lnkcap1 & PCI_EXP_LNKCAP_SLS_5_0GB)
429 *speed = PCIE_SPEED_5_0GT;
430 else if (lnkcap1 & PCI_EXP_LNKCAP_SLS_2_5GB)
431 *speed = PCIE_SPEED_2_5GT;
435 if (*speed == PCI_SPEED_UNKNOWN || *width == PCIE_LNK_WIDTH_UNKNOWN) {
437 err2 ? err2 : -EINVAL;
442 static void mlx4_check_pcie_caps(struct mlx4_dev *dev)
444 enum pcie_link_width width, width_cap;
445 enum pci_bus_speed speed, speed_cap;
448 #define PCIE_SPEED_STR(speed) \
449 (speed == PCIE_SPEED_8_0GT ? "8.0GT/s" : \
450 speed == PCIE_SPEED_5_0GT ? "5.0GT/s" : \
451 speed == PCIE_SPEED_2_5GT ? "2.5GT/s" : \
454 err = mlx4_get_pcie_dev_link_caps(dev, &speed_cap, &width_cap);
457 "Unable to determine PCIe device BW capabilities\n");
461 err = pcie_get_minimum_link(dev->pdev, &speed, &width);
462 if (err || speed == PCI_SPEED_UNKNOWN ||
463 width == PCIE_LNK_WIDTH_UNKNOWN) {
465 "Unable to determine PCI device chain minimum BW\n");
469 if (width != width_cap || speed != speed_cap)
471 "PCIe BW is different than device's capability\n");
473 mlx4_info(dev, "PCIe link speed is %s, device supports %s\n",
474 PCIE_SPEED_STR(speed), PCIE_SPEED_STR(speed_cap));
475 mlx4_info(dev, "PCIe link width is x%d, device supports x%d\n",
480 /*The function checks if there are live vf, return the num of them*/
481 static int mlx4_how_many_lives_vf(struct mlx4_dev *dev)
483 struct mlx4_priv *priv = mlx4_priv(dev);
484 struct mlx4_slave_state *s_state;
488 for (i = 1/*the ppf is 0*/; i < dev->num_slaves; ++i) {
489 s_state = &priv->mfunc.master.slave_state[i];
490 if (s_state->active && s_state->last_cmd !=
491 MLX4_COMM_CMD_RESET) {
492 mlx4_warn(dev, "%s: slave: %d is still active\n",
500 int mlx4_get_parav_qkey(struct mlx4_dev *dev, u32 qpn, u32 *qkey)
502 u32 qk = MLX4_RESERVED_QKEY_BASE;
504 if (qpn >= dev->phys_caps.base_tunnel_sqpn + 8 * MLX4_MFUNC_MAX ||
505 qpn < dev->phys_caps.base_proxy_sqpn)
508 if (qpn >= dev->phys_caps.base_tunnel_sqpn)
510 qk += qpn - dev->phys_caps.base_tunnel_sqpn;
512 qk += qpn - dev->phys_caps.base_proxy_sqpn;
516 EXPORT_SYMBOL(mlx4_get_parav_qkey);
518 void mlx4_sync_pkey_table(struct mlx4_dev *dev, int slave, int port, int i, int val)
520 struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
522 if (!mlx4_is_master(dev))
525 priv->virt2phys_pkey[slave][port - 1][i] = val;
527 EXPORT_SYMBOL(mlx4_sync_pkey_table);
529 void mlx4_put_slave_node_guid(struct mlx4_dev *dev, int slave, __be64 guid)
531 struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
533 if (!mlx4_is_master(dev))
536 priv->slave_node_guids[slave] = guid;
538 EXPORT_SYMBOL(mlx4_put_slave_node_guid);
540 __be64 mlx4_get_slave_node_guid(struct mlx4_dev *dev, int slave)
542 struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
544 if (!mlx4_is_master(dev))
547 return priv->slave_node_guids[slave];
549 EXPORT_SYMBOL(mlx4_get_slave_node_guid);
551 int mlx4_is_slave_active(struct mlx4_dev *dev, int slave)
553 struct mlx4_priv *priv = mlx4_priv(dev);
554 struct mlx4_slave_state *s_slave;
556 if (!mlx4_is_master(dev))
559 s_slave = &priv->mfunc.master.slave_state[slave];
560 return !!s_slave->active;
562 EXPORT_SYMBOL(mlx4_is_slave_active);
564 static void slave_adjust_steering_mode(struct mlx4_dev *dev,
565 struct mlx4_dev_cap *dev_cap,
566 struct mlx4_init_hca_param *hca_param)
568 dev->caps.steering_mode = hca_param->steering_mode;
569 if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED) {
570 dev->caps.num_qp_per_mgm = dev_cap->fs_max_num_qp_per_entry;
571 dev->caps.fs_log_max_ucast_qp_range_size =
572 dev_cap->fs_log_max_ucast_qp_range_size;
574 dev->caps.num_qp_per_mgm =
575 4 * ((1 << hca_param->log_mc_entry_sz)/16 - 2);
577 mlx4_dbg(dev, "Steering mode is: %s\n",
578 mlx4_steering_mode_str(dev->caps.steering_mode));
581 static int mlx4_slave_cap(struct mlx4_dev *dev)
585 struct mlx4_dev_cap dev_cap;
586 struct mlx4_func_cap func_cap;
587 struct mlx4_init_hca_param hca_param;
590 memset(&hca_param, 0, sizeof(hca_param));
591 err = mlx4_QUERY_HCA(dev, &hca_param);
593 mlx4_err(dev, "QUERY_HCA command failed, aborting\n");
597 /* fail if the hca has an unknown global capability
598 * at this time global_caps should be always zeroed
600 if (hca_param.global_caps) {
601 mlx4_err(dev, "Unknown hca global capabilities\n");
605 mlx4_log_num_mgm_entry_size = hca_param.log_mc_entry_sz;
607 dev->caps.hca_core_clock = hca_param.hca_core_clock;
609 memset(&dev_cap, 0, sizeof(dev_cap));
610 dev->caps.max_qp_dest_rdma = 1 << hca_param.log_rd_per_qp;
611 err = mlx4_dev_cap(dev, &dev_cap);
613 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
617 err = mlx4_QUERY_FW(dev);
619 mlx4_err(dev, "QUERY_FW command failed: could not get FW version\n");
621 page_size = ~dev->caps.page_size_cap + 1;
622 mlx4_warn(dev, "HCA minimum page size:%d\n", page_size);
623 if (page_size > PAGE_SIZE) {
624 mlx4_err(dev, "HCA minimum page size of %d bigger than kernel PAGE_SIZE of %ld, aborting\n",
625 page_size, PAGE_SIZE);
629 /* slave gets uar page size from QUERY_HCA fw command */
630 dev->caps.uar_page_size = 1 << (hca_param.uar_page_sz + 12);
632 /* TODO: relax this assumption */
633 if (dev->caps.uar_page_size != PAGE_SIZE) {
634 mlx4_err(dev, "UAR size:%d != kernel PAGE_SIZE of %ld\n",
635 dev->caps.uar_page_size, PAGE_SIZE);
639 memset(&func_cap, 0, sizeof(func_cap));
640 err = mlx4_QUERY_FUNC_CAP(dev, 0, &func_cap);
642 mlx4_err(dev, "QUERY_FUNC_CAP general command failed, aborting (%d)\n",
647 if ((func_cap.pf_context_behaviour | PF_CONTEXT_BEHAVIOUR_MASK) !=
648 PF_CONTEXT_BEHAVIOUR_MASK) {
649 mlx4_err(dev, "Unknown pf context behaviour\n");
653 dev->caps.num_ports = func_cap.num_ports;
654 dev->quotas.qp = func_cap.qp_quota;
655 dev->quotas.srq = func_cap.srq_quota;
656 dev->quotas.cq = func_cap.cq_quota;
657 dev->quotas.mpt = func_cap.mpt_quota;
658 dev->quotas.mtt = func_cap.mtt_quota;
659 dev->caps.num_qps = 1 << hca_param.log_num_qps;
660 dev->caps.num_srqs = 1 << hca_param.log_num_srqs;
661 dev->caps.num_cqs = 1 << hca_param.log_num_cqs;
662 dev->caps.num_mpts = 1 << hca_param.log_mpt_sz;
663 dev->caps.num_eqs = func_cap.max_eq;
664 dev->caps.reserved_eqs = func_cap.reserved_eq;
665 dev->caps.num_pds = MLX4_NUM_PDS;
666 dev->caps.num_mgms = 0;
667 dev->caps.num_amgms = 0;
669 if (dev->caps.num_ports > MLX4_MAX_PORTS) {
670 mlx4_err(dev, "HCA has %d ports, but we only support %d, aborting\n",
671 dev->caps.num_ports, MLX4_MAX_PORTS);
675 dev->caps.qp0_qkey = kcalloc(dev->caps.num_ports, sizeof(u32), GFP_KERNEL);
676 dev->caps.qp0_tunnel = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
677 dev->caps.qp0_proxy = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
678 dev->caps.qp1_tunnel = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
679 dev->caps.qp1_proxy = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
681 if (!dev->caps.qp0_tunnel || !dev->caps.qp0_proxy ||
682 !dev->caps.qp1_tunnel || !dev->caps.qp1_proxy ||
683 !dev->caps.qp0_qkey) {
688 for (i = 1; i <= dev->caps.num_ports; ++i) {
689 err = mlx4_QUERY_FUNC_CAP(dev, (u32) i, &func_cap);
691 mlx4_err(dev, "QUERY_FUNC_CAP port command failed for port %d, aborting (%d)\n",
695 dev->caps.qp0_qkey[i - 1] = func_cap.qp0_qkey;
696 dev->caps.qp0_tunnel[i - 1] = func_cap.qp0_tunnel_qpn;
697 dev->caps.qp0_proxy[i - 1] = func_cap.qp0_proxy_qpn;
698 dev->caps.qp1_tunnel[i - 1] = func_cap.qp1_tunnel_qpn;
699 dev->caps.qp1_proxy[i - 1] = func_cap.qp1_proxy_qpn;
700 dev->caps.port_mask[i] = dev->caps.port_type[i];
701 dev->caps.phys_port_id[i] = func_cap.phys_port_id;
702 if (mlx4_get_slave_pkey_gid_tbl_len(dev, i,
703 &dev->caps.gid_table_len[i],
704 &dev->caps.pkey_table_len[i]))
708 if (dev->caps.uar_page_size * (dev->caps.num_uars -
709 dev->caps.reserved_uars) >
710 pci_resource_len(dev->pdev, 2)) {
711 mlx4_err(dev, "HCA reported UAR region size of 0x%x bigger than PCI resource 2 size of 0x%llx, aborting\n",
712 dev->caps.uar_page_size * dev->caps.num_uars,
713 (unsigned long long) pci_resource_len(dev->pdev, 2));
717 if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_64B_EQE_ENABLED) {
718 dev->caps.eqe_size = 64;
719 dev->caps.eqe_factor = 1;
721 dev->caps.eqe_size = 32;
722 dev->caps.eqe_factor = 0;
725 if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_64B_CQE_ENABLED) {
726 dev->caps.cqe_size = 64;
727 dev->caps.userspace_caps |= MLX4_USER_DEV_CAP_64B_CQE;
729 dev->caps.cqe_size = 32;
732 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
733 mlx4_warn(dev, "Timestamping is not supported in slave mode\n");
735 slave_adjust_steering_mode(dev, &dev_cap, &hca_param);
740 kfree(dev->caps.qp0_qkey);
741 kfree(dev->caps.qp0_tunnel);
742 kfree(dev->caps.qp0_proxy);
743 kfree(dev->caps.qp1_tunnel);
744 kfree(dev->caps.qp1_proxy);
745 dev->caps.qp0_qkey = NULL;
746 dev->caps.qp0_tunnel = NULL;
747 dev->caps.qp0_proxy = NULL;
748 dev->caps.qp1_tunnel = NULL;
749 dev->caps.qp1_proxy = NULL;
754 static void mlx4_request_modules(struct mlx4_dev *dev)
757 int has_ib_port = false;
758 int has_eth_port = false;
759 #define EN_DRV_NAME "mlx4_en"
760 #define IB_DRV_NAME "mlx4_ib"
762 for (port = 1; port <= dev->caps.num_ports; port++) {
763 if (dev->caps.port_type[port] == MLX4_PORT_TYPE_IB)
765 else if (dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH)
770 request_module_nowait(EN_DRV_NAME);
771 if (has_ib_port || (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE))
772 request_module_nowait(IB_DRV_NAME);
776 * Change the port configuration of the device.
777 * Every user of this function must hold the port mutex.
779 int mlx4_change_port_types(struct mlx4_dev *dev,
780 enum mlx4_port_type *port_types)
786 for (port = 0; port < dev->caps.num_ports; port++) {
787 /* Change the port type only if the new type is different
788 * from the current, and not set to Auto */
789 if (port_types[port] != dev->caps.port_type[port + 1])
793 mlx4_unregister_device(dev);
794 for (port = 1; port <= dev->caps.num_ports; port++) {
795 mlx4_CLOSE_PORT(dev, port);
796 dev->caps.port_type[port] = port_types[port - 1];
797 err = mlx4_SET_PORT(dev, port, -1);
799 mlx4_err(dev, "Failed to set port %d, aborting\n",
804 mlx4_set_port_mask(dev);
805 err = mlx4_register_device(dev);
807 mlx4_err(dev, "Failed to register device\n");
810 mlx4_request_modules(dev);
817 static ssize_t show_port_type(struct device *dev,
818 struct device_attribute *attr,
821 struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
823 struct mlx4_dev *mdev = info->dev;
827 (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_IB) ?
829 if (mdev->caps.possible_type[info->port] == MLX4_PORT_TYPE_AUTO)
830 sprintf(buf, "auto (%s)\n", type);
832 sprintf(buf, "%s\n", type);
837 static ssize_t set_port_type(struct device *dev,
838 struct device_attribute *attr,
839 const char *buf, size_t count)
841 struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
843 struct mlx4_dev *mdev = info->dev;
844 struct mlx4_priv *priv = mlx4_priv(mdev);
845 enum mlx4_port_type types[MLX4_MAX_PORTS];
846 enum mlx4_port_type new_types[MLX4_MAX_PORTS];
850 if (!strcmp(buf, "ib\n"))
851 info->tmp_type = MLX4_PORT_TYPE_IB;
852 else if (!strcmp(buf, "eth\n"))
853 info->tmp_type = MLX4_PORT_TYPE_ETH;
854 else if (!strcmp(buf, "auto\n"))
855 info->tmp_type = MLX4_PORT_TYPE_AUTO;
857 mlx4_err(mdev, "%s is not supported port type\n", buf);
861 mlx4_stop_sense(mdev);
862 mutex_lock(&priv->port_mutex);
863 /* Possible type is always the one that was delivered */
864 mdev->caps.possible_type[info->port] = info->tmp_type;
866 for (i = 0; i < mdev->caps.num_ports; i++) {
867 types[i] = priv->port[i+1].tmp_type ? priv->port[i+1].tmp_type :
868 mdev->caps.possible_type[i+1];
869 if (types[i] == MLX4_PORT_TYPE_AUTO)
870 types[i] = mdev->caps.port_type[i+1];
873 if (!(mdev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
874 !(mdev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT)) {
875 for (i = 1; i <= mdev->caps.num_ports; i++) {
876 if (mdev->caps.possible_type[i] == MLX4_PORT_TYPE_AUTO) {
877 mdev->caps.possible_type[i] = mdev->caps.port_type[i];
883 mlx4_err(mdev, "Auto sensing is not supported on this HCA. Set only 'eth' or 'ib' for both ports (should be the same)\n");
887 mlx4_do_sense_ports(mdev, new_types, types);
889 err = mlx4_check_port_params(mdev, new_types);
893 /* We are about to apply the changes after the configuration
894 * was verified, no need to remember the temporary types
896 for (i = 0; i < mdev->caps.num_ports; i++)
897 priv->port[i + 1].tmp_type = 0;
899 err = mlx4_change_port_types(mdev, new_types);
902 mlx4_start_sense(mdev);
903 mutex_unlock(&priv->port_mutex);
904 return err ? err : count;
915 static inline int int_to_ibta_mtu(int mtu)
918 case 256: return IB_MTU_256;
919 case 512: return IB_MTU_512;
920 case 1024: return IB_MTU_1024;
921 case 2048: return IB_MTU_2048;
922 case 4096: return IB_MTU_4096;
927 static inline int ibta_mtu_to_int(enum ibta_mtu mtu)
930 case IB_MTU_256: return 256;
931 case IB_MTU_512: return 512;
932 case IB_MTU_1024: return 1024;
933 case IB_MTU_2048: return 2048;
934 case IB_MTU_4096: return 4096;
939 static ssize_t show_port_ib_mtu(struct device *dev,
940 struct device_attribute *attr,
943 struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
945 struct mlx4_dev *mdev = info->dev;
947 if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH)
948 mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
951 ibta_mtu_to_int(mdev->caps.port_ib_mtu[info->port]));
955 static ssize_t set_port_ib_mtu(struct device *dev,
956 struct device_attribute *attr,
957 const char *buf, size_t count)
959 struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
961 struct mlx4_dev *mdev = info->dev;
962 struct mlx4_priv *priv = mlx4_priv(mdev);
963 int err, port, mtu, ibta_mtu = -1;
965 if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH) {
966 mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
970 err = kstrtoint(buf, 0, &mtu);
972 ibta_mtu = int_to_ibta_mtu(mtu);
974 if (err || ibta_mtu < 0) {
975 mlx4_err(mdev, "%s is invalid IBTA mtu\n", buf);
979 mdev->caps.port_ib_mtu[info->port] = ibta_mtu;
981 mlx4_stop_sense(mdev);
982 mutex_lock(&priv->port_mutex);
983 mlx4_unregister_device(mdev);
984 for (port = 1; port <= mdev->caps.num_ports; port++) {
985 mlx4_CLOSE_PORT(mdev, port);
986 err = mlx4_SET_PORT(mdev, port, -1);
988 mlx4_err(mdev, "Failed to set port %d, aborting\n",
993 err = mlx4_register_device(mdev);
995 mutex_unlock(&priv->port_mutex);
996 mlx4_start_sense(mdev);
997 return err ? err : count;
1000 static int mlx4_load_fw(struct mlx4_dev *dev)
1002 struct mlx4_priv *priv = mlx4_priv(dev);
1005 priv->fw.fw_icm = mlx4_alloc_icm(dev, priv->fw.fw_pages,
1006 GFP_HIGHUSER | __GFP_NOWARN, 0);
1007 if (!priv->fw.fw_icm) {
1008 mlx4_err(dev, "Couldn't allocate FW area, aborting\n");
1012 err = mlx4_MAP_FA(dev, priv->fw.fw_icm);
1014 mlx4_err(dev, "MAP_FA command failed, aborting\n");
1018 err = mlx4_RUN_FW(dev);
1020 mlx4_err(dev, "RUN_FW command failed, aborting\n");
1030 mlx4_free_icm(dev, priv->fw.fw_icm, 0);
1034 static int mlx4_init_cmpt_table(struct mlx4_dev *dev, u64 cmpt_base,
1037 struct mlx4_priv *priv = mlx4_priv(dev);
1041 err = mlx4_init_icm_table(dev, &priv->qp_table.cmpt_table,
1043 ((u64) (MLX4_CMPT_TYPE_QP *
1044 cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1045 cmpt_entry_sz, dev->caps.num_qps,
1046 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1051 err = mlx4_init_icm_table(dev, &priv->srq_table.cmpt_table,
1053 ((u64) (MLX4_CMPT_TYPE_SRQ *
1054 cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1055 cmpt_entry_sz, dev->caps.num_srqs,
1056 dev->caps.reserved_srqs, 0, 0);
1060 err = mlx4_init_icm_table(dev, &priv->cq_table.cmpt_table,
1062 ((u64) (MLX4_CMPT_TYPE_CQ *
1063 cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1064 cmpt_entry_sz, dev->caps.num_cqs,
1065 dev->caps.reserved_cqs, 0, 0);
1069 num_eqs = (mlx4_is_master(dev)) ? dev->phys_caps.num_phys_eqs :
1071 err = mlx4_init_icm_table(dev, &priv->eq_table.cmpt_table,
1073 ((u64) (MLX4_CMPT_TYPE_EQ *
1074 cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1075 cmpt_entry_sz, num_eqs, num_eqs, 0, 0);
1082 mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1085 mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1088 mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1094 static int mlx4_init_icm(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap,
1095 struct mlx4_init_hca_param *init_hca, u64 icm_size)
1097 struct mlx4_priv *priv = mlx4_priv(dev);
1102 err = mlx4_SET_ICM_SIZE(dev, icm_size, &aux_pages);
1104 mlx4_err(dev, "SET_ICM_SIZE command failed, aborting\n");
1108 mlx4_dbg(dev, "%lld KB of HCA context requires %lld KB aux memory\n",
1109 (unsigned long long) icm_size >> 10,
1110 (unsigned long long) aux_pages << 2);
1112 priv->fw.aux_icm = mlx4_alloc_icm(dev, aux_pages,
1113 GFP_HIGHUSER | __GFP_NOWARN, 0);
1114 if (!priv->fw.aux_icm) {
1115 mlx4_err(dev, "Couldn't allocate aux memory, aborting\n");
1119 err = mlx4_MAP_ICM_AUX(dev, priv->fw.aux_icm);
1121 mlx4_err(dev, "MAP_ICM_AUX command failed, aborting\n");
1125 err = mlx4_init_cmpt_table(dev, init_hca->cmpt_base, dev_cap->cmpt_entry_sz);
1127 mlx4_err(dev, "Failed to map cMPT context memory, aborting\n");
1132 num_eqs = (mlx4_is_master(dev)) ? dev->phys_caps.num_phys_eqs :
1134 err = mlx4_init_icm_table(dev, &priv->eq_table.table,
1135 init_hca->eqc_base, dev_cap->eqc_entry_sz,
1136 num_eqs, num_eqs, 0, 0);
1138 mlx4_err(dev, "Failed to map EQ context memory, aborting\n");
1139 goto err_unmap_cmpt;
1143 * Reserved MTT entries must be aligned up to a cacheline
1144 * boundary, since the FW will write to them, while the driver
1145 * writes to all other MTT entries. (The variable
1146 * dev->caps.mtt_entry_sz below is really the MTT segment
1147 * size, not the raw entry size)
1149 dev->caps.reserved_mtts =
1150 ALIGN(dev->caps.reserved_mtts * dev->caps.mtt_entry_sz,
1151 dma_get_cache_alignment()) / dev->caps.mtt_entry_sz;
1153 err = mlx4_init_icm_table(dev, &priv->mr_table.mtt_table,
1155 dev->caps.mtt_entry_sz,
1157 dev->caps.reserved_mtts, 1, 0);
1159 mlx4_err(dev, "Failed to map MTT context memory, aborting\n");
1163 err = mlx4_init_icm_table(dev, &priv->mr_table.dmpt_table,
1164 init_hca->dmpt_base,
1165 dev_cap->dmpt_entry_sz,
1167 dev->caps.reserved_mrws, 1, 1);
1169 mlx4_err(dev, "Failed to map dMPT context memory, aborting\n");
1173 err = mlx4_init_icm_table(dev, &priv->qp_table.qp_table,
1175 dev_cap->qpc_entry_sz,
1177 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1180 mlx4_err(dev, "Failed to map QP context memory, aborting\n");
1181 goto err_unmap_dmpt;
1184 err = mlx4_init_icm_table(dev, &priv->qp_table.auxc_table,
1185 init_hca->auxc_base,
1186 dev_cap->aux_entry_sz,
1188 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1191 mlx4_err(dev, "Failed to map AUXC context memory, aborting\n");
1195 err = mlx4_init_icm_table(dev, &priv->qp_table.altc_table,
1196 init_hca->altc_base,
1197 dev_cap->altc_entry_sz,
1199 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1202 mlx4_err(dev, "Failed to map ALTC context memory, aborting\n");
1203 goto err_unmap_auxc;
1206 err = mlx4_init_icm_table(dev, &priv->qp_table.rdmarc_table,
1207 init_hca->rdmarc_base,
1208 dev_cap->rdmarc_entry_sz << priv->qp_table.rdmarc_shift,
1210 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1213 mlx4_err(dev, "Failed to map RDMARC context memory, aborting\n");
1214 goto err_unmap_altc;
1217 err = mlx4_init_icm_table(dev, &priv->cq_table.table,
1219 dev_cap->cqc_entry_sz,
1221 dev->caps.reserved_cqs, 0, 0);
1223 mlx4_err(dev, "Failed to map CQ context memory, aborting\n");
1224 goto err_unmap_rdmarc;
1227 err = mlx4_init_icm_table(dev, &priv->srq_table.table,
1228 init_hca->srqc_base,
1229 dev_cap->srq_entry_sz,
1231 dev->caps.reserved_srqs, 0, 0);
1233 mlx4_err(dev, "Failed to map SRQ context memory, aborting\n");
1238 * For flow steering device managed mode it is required to use
1239 * mlx4_init_icm_table. For B0 steering mode it's not strictly
1240 * required, but for simplicity just map the whole multicast
1241 * group table now. The table isn't very big and it's a lot
1242 * easier than trying to track ref counts.
1244 err = mlx4_init_icm_table(dev, &priv->mcg_table.table,
1246 mlx4_get_mgm_entry_size(dev),
1247 dev->caps.num_mgms + dev->caps.num_amgms,
1248 dev->caps.num_mgms + dev->caps.num_amgms,
1251 mlx4_err(dev, "Failed to map MCG context memory, aborting\n");
1258 mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1261 mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1264 mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1267 mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1270 mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1273 mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1276 mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1279 mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
1282 mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
1285 mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1286 mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1287 mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1288 mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1291 mlx4_UNMAP_ICM_AUX(dev);
1294 mlx4_free_icm(dev, priv->fw.aux_icm, 0);
1299 static void mlx4_free_icms(struct mlx4_dev *dev)
1301 struct mlx4_priv *priv = mlx4_priv(dev);
1303 mlx4_cleanup_icm_table(dev, &priv->mcg_table.table);
1304 mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1305 mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1306 mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1307 mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1308 mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1309 mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1310 mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1311 mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
1312 mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
1313 mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1314 mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1315 mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1316 mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1318 mlx4_UNMAP_ICM_AUX(dev);
1319 mlx4_free_icm(dev, priv->fw.aux_icm, 0);
1322 static void mlx4_slave_exit(struct mlx4_dev *dev)
1324 struct mlx4_priv *priv = mlx4_priv(dev);
1326 mutex_lock(&priv->cmd.slave_cmd_mutex);
1327 if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, MLX4_COMM_TIME))
1328 mlx4_warn(dev, "Failed to close slave function\n");
1329 mutex_unlock(&priv->cmd.slave_cmd_mutex);
1332 static int map_bf_area(struct mlx4_dev *dev)
1334 struct mlx4_priv *priv = mlx4_priv(dev);
1335 resource_size_t bf_start;
1336 resource_size_t bf_len;
1339 if (!dev->caps.bf_reg_size)
1342 bf_start = pci_resource_start(dev->pdev, 2) +
1343 (dev->caps.num_uars << PAGE_SHIFT);
1344 bf_len = pci_resource_len(dev->pdev, 2) -
1345 (dev->caps.num_uars << PAGE_SHIFT);
1346 priv->bf_mapping = io_mapping_create_wc(bf_start, bf_len);
1347 if (!priv->bf_mapping)
1353 static void unmap_bf_area(struct mlx4_dev *dev)
1355 if (mlx4_priv(dev)->bf_mapping)
1356 io_mapping_free(mlx4_priv(dev)->bf_mapping);
1359 cycle_t mlx4_read_clock(struct mlx4_dev *dev)
1361 u32 clockhi, clocklo, clockhi1;
1364 struct mlx4_priv *priv = mlx4_priv(dev);
1366 for (i = 0; i < 10; i++) {
1367 clockhi = swab32(readl(priv->clock_mapping));
1368 clocklo = swab32(readl(priv->clock_mapping + 4));
1369 clockhi1 = swab32(readl(priv->clock_mapping));
1370 if (clockhi == clockhi1)
1374 cycles = (u64) clockhi << 32 | (u64) clocklo;
1378 EXPORT_SYMBOL_GPL(mlx4_read_clock);
1381 static int map_internal_clock(struct mlx4_dev *dev)
1383 struct mlx4_priv *priv = mlx4_priv(dev);
1385 priv->clock_mapping =
1386 ioremap(pci_resource_start(dev->pdev, priv->fw.clock_bar) +
1387 priv->fw.clock_offset, MLX4_CLOCK_SIZE);
1389 if (!priv->clock_mapping)
1395 static void unmap_internal_clock(struct mlx4_dev *dev)
1397 struct mlx4_priv *priv = mlx4_priv(dev);
1399 if (priv->clock_mapping)
1400 iounmap(priv->clock_mapping);
1403 static void mlx4_close_hca(struct mlx4_dev *dev)
1405 unmap_internal_clock(dev);
1407 if (mlx4_is_slave(dev))
1408 mlx4_slave_exit(dev);
1410 mlx4_CLOSE_HCA(dev, 0);
1411 mlx4_free_icms(dev);
1413 mlx4_free_icm(dev, mlx4_priv(dev)->fw.fw_icm, 0);
1417 static int mlx4_init_slave(struct mlx4_dev *dev)
1419 struct mlx4_priv *priv = mlx4_priv(dev);
1420 u64 dma = (u64) priv->mfunc.vhcr_dma;
1421 int ret_from_reset = 0;
1423 u32 cmd_channel_ver;
1425 if (atomic_read(&pf_loading)) {
1426 mlx4_warn(dev, "PF is not ready - Deferring probe\n");
1427 return -EPROBE_DEFER;
1430 mutex_lock(&priv->cmd.slave_cmd_mutex);
1431 priv->cmd.max_cmds = 1;
1432 mlx4_warn(dev, "Sending reset\n");
1433 ret_from_reset = mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0,
1435 /* if we are in the middle of flr the slave will try
1436 * NUM_OF_RESET_RETRIES times before leaving.*/
1437 if (ret_from_reset) {
1438 if (MLX4_DELAY_RESET_SLAVE == ret_from_reset) {
1439 mlx4_warn(dev, "slave is currently in the middle of FLR - Deferring probe\n");
1440 mutex_unlock(&priv->cmd.slave_cmd_mutex);
1441 return -EPROBE_DEFER;
1446 /* check the driver version - the slave I/F revision
1447 * must match the master's */
1448 slave_read = swab32(readl(&priv->mfunc.comm->slave_read));
1449 cmd_channel_ver = mlx4_comm_get_version();
1451 if (MLX4_COMM_GET_IF_REV(cmd_channel_ver) !=
1452 MLX4_COMM_GET_IF_REV(slave_read)) {
1453 mlx4_err(dev, "slave driver version is not supported by the master\n");
1457 mlx4_warn(dev, "Sending vhcr0\n");
1458 if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR0, dma >> 48,
1461 if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR1, dma >> 32,
1464 if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR2, dma >> 16,
1467 if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR_EN, dma, MLX4_COMM_TIME))
1470 mutex_unlock(&priv->cmd.slave_cmd_mutex);
1474 mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, 0);
1475 mutex_unlock(&priv->cmd.slave_cmd_mutex);
1479 static void mlx4_parav_master_pf_caps(struct mlx4_dev *dev)
1483 for (i = 1; i <= dev->caps.num_ports; i++) {
1484 if (dev->caps.port_type[i] == MLX4_PORT_TYPE_ETH)
1485 dev->caps.gid_table_len[i] =
1486 mlx4_get_slave_num_gids(dev, 0, i);
1488 dev->caps.gid_table_len[i] = 1;
1489 dev->caps.pkey_table_len[i] =
1490 dev->phys_caps.pkey_phys_table_len[i] - 1;
1494 static int choose_log_fs_mgm_entry_size(int qp_per_entry)
1496 int i = MLX4_MIN_MGM_LOG_ENTRY_SIZE;
1498 for (i = MLX4_MIN_MGM_LOG_ENTRY_SIZE; i <= MLX4_MAX_MGM_LOG_ENTRY_SIZE;
1500 if (qp_per_entry <= 4 * ((1 << i) / 16 - 2))
1504 return (i <= MLX4_MAX_MGM_LOG_ENTRY_SIZE) ? i : -1;
1507 static void choose_steering_mode(struct mlx4_dev *dev,
1508 struct mlx4_dev_cap *dev_cap)
1510 if (mlx4_log_num_mgm_entry_size == -1 &&
1511 dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_FS_EN &&
1512 (!mlx4_is_mfunc(dev) ||
1513 (dev_cap->fs_max_num_qp_per_entry >= (dev->num_vfs + 1))) &&
1514 choose_log_fs_mgm_entry_size(dev_cap->fs_max_num_qp_per_entry) >=
1515 MLX4_MIN_MGM_LOG_ENTRY_SIZE) {
1516 dev->oper_log_mgm_entry_size =
1517 choose_log_fs_mgm_entry_size(dev_cap->fs_max_num_qp_per_entry);
1518 dev->caps.steering_mode = MLX4_STEERING_MODE_DEVICE_MANAGED;
1519 dev->caps.num_qp_per_mgm = dev_cap->fs_max_num_qp_per_entry;
1520 dev->caps.fs_log_max_ucast_qp_range_size =
1521 dev_cap->fs_log_max_ucast_qp_range_size;
1523 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER &&
1524 dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER)
1525 dev->caps.steering_mode = MLX4_STEERING_MODE_B0;
1527 dev->caps.steering_mode = MLX4_STEERING_MODE_A0;
1529 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER ||
1530 dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER)
1531 mlx4_warn(dev, "Must have both UC_STEER and MC_STEER flags set to use B0 steering - falling back to A0 steering mode\n");
1533 dev->oper_log_mgm_entry_size =
1534 mlx4_log_num_mgm_entry_size > 0 ?
1535 mlx4_log_num_mgm_entry_size :
1536 MLX4_DEFAULT_MGM_LOG_ENTRY_SIZE;
1537 dev->caps.num_qp_per_mgm = mlx4_get_qp_per_mgm(dev);
1539 mlx4_dbg(dev, "Steering mode is: %s, oper_log_mgm_entry_size = %d, modparam log_num_mgm_entry_size = %d\n",
1540 mlx4_steering_mode_str(dev->caps.steering_mode),
1541 dev->oper_log_mgm_entry_size,
1542 mlx4_log_num_mgm_entry_size);
1545 static void choose_tunnel_offload_mode(struct mlx4_dev *dev,
1546 struct mlx4_dev_cap *dev_cap)
1548 if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED &&
1549 dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_VXLAN_OFFLOADS)
1550 dev->caps.tunnel_offload_mode = MLX4_TUNNEL_OFFLOAD_MODE_VXLAN;
1552 dev->caps.tunnel_offload_mode = MLX4_TUNNEL_OFFLOAD_MODE_NONE;
1554 mlx4_dbg(dev, "Tunneling offload mode is: %s\n", (dev->caps.tunnel_offload_mode
1555 == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) ? "vxlan" : "none");
1558 static int mlx4_init_hca(struct mlx4_dev *dev)
1560 struct mlx4_priv *priv = mlx4_priv(dev);
1561 struct mlx4_adapter adapter;
1562 struct mlx4_dev_cap dev_cap;
1563 struct mlx4_mod_stat_cfg mlx4_cfg;
1564 struct mlx4_profile profile;
1565 struct mlx4_init_hca_param init_hca;
1569 if (!mlx4_is_slave(dev)) {
1570 err = mlx4_QUERY_FW(dev);
1573 mlx4_info(dev, "non-primary physical function, skipping\n");
1575 mlx4_err(dev, "QUERY_FW command failed, aborting\n");
1579 err = mlx4_load_fw(dev);
1581 mlx4_err(dev, "Failed to start FW, aborting\n");
1585 mlx4_cfg.log_pg_sz_m = 1;
1586 mlx4_cfg.log_pg_sz = 0;
1587 err = mlx4_MOD_STAT_CFG(dev, &mlx4_cfg);
1589 mlx4_warn(dev, "Failed to override log_pg_sz parameter\n");
1591 err = mlx4_dev_cap(dev, &dev_cap);
1593 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
1597 choose_steering_mode(dev, &dev_cap);
1598 choose_tunnel_offload_mode(dev, &dev_cap);
1600 err = mlx4_get_phys_port_id(dev);
1602 mlx4_err(dev, "Fail to get physical port id\n");
1604 if (mlx4_is_master(dev))
1605 mlx4_parav_master_pf_caps(dev);
1607 if (mlx4_low_memory_profile()) {
1608 mlx4_info(dev, "Running from within kdump kernel. Using low memory profile\n");
1609 profile = low_mem_profile;
1611 profile = default_profile;
1613 if (dev->caps.steering_mode ==
1614 MLX4_STEERING_MODE_DEVICE_MANAGED)
1615 profile.num_mcg = MLX4_FS_NUM_MCG;
1617 icm_size = mlx4_make_profile(dev, &profile, &dev_cap,
1619 if ((long long) icm_size < 0) {
1624 dev->caps.max_fmr_maps = (1 << (32 - ilog2(dev->caps.num_mpts))) - 1;
1626 init_hca.log_uar_sz = ilog2(dev->caps.num_uars);
1627 init_hca.uar_page_sz = PAGE_SHIFT - 12;
1628 init_hca.mw_enabled = 0;
1629 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
1630 dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN)
1631 init_hca.mw_enabled = INIT_HCA_TPT_MW_ENABLE;
1633 err = mlx4_init_icm(dev, &dev_cap, &init_hca, icm_size);
1637 err = mlx4_INIT_HCA(dev, &init_hca);
1639 mlx4_err(dev, "INIT_HCA command failed, aborting\n");
1643 * If TS is supported by FW
1644 * read HCA frequency by QUERY_HCA command
1646 if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS) {
1647 memset(&init_hca, 0, sizeof(init_hca));
1648 err = mlx4_QUERY_HCA(dev, &init_hca);
1650 mlx4_err(dev, "QUERY_HCA command failed, disable timestamp\n");
1651 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
1653 dev->caps.hca_core_clock =
1654 init_hca.hca_core_clock;
1657 /* In case we got HCA frequency 0 - disable timestamping
1658 * to avoid dividing by zero
1660 if (!dev->caps.hca_core_clock) {
1661 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
1663 "HCA frequency is 0 - timestamping is not supported\n");
1664 } else if (map_internal_clock(dev)) {
1666 * Map internal clock,
1667 * in case of failure disable timestamping
1669 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
1670 mlx4_err(dev, "Failed to map internal clock. Timestamping is not supported\n");
1674 err = mlx4_init_slave(dev);
1676 if (err != -EPROBE_DEFER)
1677 mlx4_err(dev, "Failed to initialize slave\n");
1681 err = mlx4_slave_cap(dev);
1683 mlx4_err(dev, "Failed to obtain slave caps\n");
1688 if (map_bf_area(dev))
1689 mlx4_dbg(dev, "Failed to map blue flame area\n");
1691 /*Only the master set the ports, all the rest got it from it.*/
1692 if (!mlx4_is_slave(dev))
1693 mlx4_set_port_mask(dev);
1695 err = mlx4_QUERY_ADAPTER(dev, &adapter);
1697 mlx4_err(dev, "QUERY_ADAPTER command failed, aborting\n");
1701 priv->eq_table.inta_pin = adapter.inta_pin;
1702 memcpy(dev->board_id, adapter.board_id, sizeof dev->board_id);
1707 unmap_internal_clock(dev);
1710 if (mlx4_is_slave(dev)) {
1711 kfree(dev->caps.qp0_qkey);
1712 kfree(dev->caps.qp0_tunnel);
1713 kfree(dev->caps.qp0_proxy);
1714 kfree(dev->caps.qp1_tunnel);
1715 kfree(dev->caps.qp1_proxy);
1719 if (mlx4_is_slave(dev))
1720 mlx4_slave_exit(dev);
1722 mlx4_CLOSE_HCA(dev, 0);
1725 if (!mlx4_is_slave(dev))
1726 mlx4_free_icms(dev);
1729 if (!mlx4_is_slave(dev)) {
1731 mlx4_free_icm(dev, priv->fw.fw_icm, 0);
1736 static int mlx4_init_counters_table(struct mlx4_dev *dev)
1738 struct mlx4_priv *priv = mlx4_priv(dev);
1741 if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
1744 nent = dev->caps.max_counters;
1745 return mlx4_bitmap_init(&priv->counters_bitmap, nent, nent - 1, 0, 0);
1748 static void mlx4_cleanup_counters_table(struct mlx4_dev *dev)
1750 mlx4_bitmap_cleanup(&mlx4_priv(dev)->counters_bitmap);
1753 int __mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx)
1755 struct mlx4_priv *priv = mlx4_priv(dev);
1757 if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
1760 *idx = mlx4_bitmap_alloc(&priv->counters_bitmap);
1767 int mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx)
1772 if (mlx4_is_mfunc(dev)) {
1773 err = mlx4_cmd_imm(dev, 0, &out_param, RES_COUNTER,
1774 RES_OP_RESERVE, MLX4_CMD_ALLOC_RES,
1775 MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
1777 *idx = get_param_l(&out_param);
1781 return __mlx4_counter_alloc(dev, idx);
1783 EXPORT_SYMBOL_GPL(mlx4_counter_alloc);
1785 void __mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
1787 mlx4_bitmap_free(&mlx4_priv(dev)->counters_bitmap, idx, MLX4_USE_RR);
1791 void mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
1795 if (mlx4_is_mfunc(dev)) {
1796 set_param_l(&in_param, idx);
1797 mlx4_cmd(dev, in_param, RES_COUNTER, RES_OP_RESERVE,
1798 MLX4_CMD_FREE_RES, MLX4_CMD_TIME_CLASS_A,
1802 __mlx4_counter_free(dev, idx);
1804 EXPORT_SYMBOL_GPL(mlx4_counter_free);
1806 static int mlx4_setup_hca(struct mlx4_dev *dev)
1808 struct mlx4_priv *priv = mlx4_priv(dev);
1811 __be32 ib_port_default_caps;
1813 err = mlx4_init_uar_table(dev);
1815 mlx4_err(dev, "Failed to initialize user access region table, aborting\n");
1819 err = mlx4_uar_alloc(dev, &priv->driver_uar);
1821 mlx4_err(dev, "Failed to allocate driver access region, aborting\n");
1822 goto err_uar_table_free;
1825 priv->kar = ioremap((phys_addr_t) priv->driver_uar.pfn << PAGE_SHIFT, PAGE_SIZE);
1827 mlx4_err(dev, "Couldn't map kernel access region, aborting\n");
1832 err = mlx4_init_pd_table(dev);
1834 mlx4_err(dev, "Failed to initialize protection domain table, aborting\n");
1838 err = mlx4_init_xrcd_table(dev);
1840 mlx4_err(dev, "Failed to initialize reliable connection domain table, aborting\n");
1841 goto err_pd_table_free;
1844 err = mlx4_init_mr_table(dev);
1846 mlx4_err(dev, "Failed to initialize memory region table, aborting\n");
1847 goto err_xrcd_table_free;
1850 if (!mlx4_is_slave(dev)) {
1851 err = mlx4_init_mcg_table(dev);
1853 mlx4_err(dev, "Failed to initialize multicast group table, aborting\n");
1854 goto err_mr_table_free;
1858 err = mlx4_init_eq_table(dev);
1860 mlx4_err(dev, "Failed to initialize event queue table, aborting\n");
1861 goto err_mcg_table_free;
1864 err = mlx4_cmd_use_events(dev);
1866 mlx4_err(dev, "Failed to switch to event-driven firmware commands, aborting\n");
1867 goto err_eq_table_free;
1870 err = mlx4_NOP(dev);
1872 if (dev->flags & MLX4_FLAG_MSI_X) {
1873 mlx4_warn(dev, "NOP command failed to generate MSI-X interrupt IRQ %d)\n",
1874 priv->eq_table.eq[dev->caps.num_comp_vectors].irq);
1875 mlx4_warn(dev, "Trying again without MSI-X\n");
1877 mlx4_err(dev, "NOP command failed to generate interrupt (IRQ %d), aborting\n",
1878 priv->eq_table.eq[dev->caps.num_comp_vectors].irq);
1879 mlx4_err(dev, "BIOS or ACPI interrupt routing problem?\n");
1885 mlx4_dbg(dev, "NOP command IRQ test passed\n");
1887 err = mlx4_init_cq_table(dev);
1889 mlx4_err(dev, "Failed to initialize completion queue table, aborting\n");
1893 err = mlx4_init_srq_table(dev);
1895 mlx4_err(dev, "Failed to initialize shared receive queue table, aborting\n");
1896 goto err_cq_table_free;
1899 err = mlx4_init_qp_table(dev);
1901 mlx4_err(dev, "Failed to initialize queue pair table, aborting\n");
1902 goto err_srq_table_free;
1905 err = mlx4_init_counters_table(dev);
1906 if (err && err != -ENOENT) {
1907 mlx4_err(dev, "Failed to initialize counters table, aborting\n");
1908 goto err_qp_table_free;
1911 if (!mlx4_is_slave(dev)) {
1912 for (port = 1; port <= dev->caps.num_ports; port++) {
1913 ib_port_default_caps = 0;
1914 err = mlx4_get_port_ib_caps(dev, port,
1915 &ib_port_default_caps);
1917 mlx4_warn(dev, "failed to get port %d default ib capabilities (%d). Continuing with caps = 0\n",
1919 dev->caps.ib_port_def_cap[port] = ib_port_default_caps;
1921 /* initialize per-slave default ib port capabilities */
1922 if (mlx4_is_master(dev)) {
1924 for (i = 0; i < dev->num_slaves; i++) {
1925 if (i == mlx4_master_func_num(dev))
1927 priv->mfunc.master.slave_state[i].ib_cap_mask[port] =
1928 ib_port_default_caps;
1932 if (mlx4_is_mfunc(dev))
1933 dev->caps.port_ib_mtu[port] = IB_MTU_2048;
1935 dev->caps.port_ib_mtu[port] = IB_MTU_4096;
1937 err = mlx4_SET_PORT(dev, port, mlx4_is_master(dev) ?
1938 dev->caps.pkey_table_len[port] : -1);
1940 mlx4_err(dev, "Failed to set port %d, aborting\n",
1942 goto err_counters_table_free;
1949 err_counters_table_free:
1950 mlx4_cleanup_counters_table(dev);
1953 mlx4_cleanup_qp_table(dev);
1956 mlx4_cleanup_srq_table(dev);
1959 mlx4_cleanup_cq_table(dev);
1962 mlx4_cmd_use_polling(dev);
1965 mlx4_cleanup_eq_table(dev);
1968 if (!mlx4_is_slave(dev))
1969 mlx4_cleanup_mcg_table(dev);
1972 mlx4_cleanup_mr_table(dev);
1974 err_xrcd_table_free:
1975 mlx4_cleanup_xrcd_table(dev);
1978 mlx4_cleanup_pd_table(dev);
1984 mlx4_uar_free(dev, &priv->driver_uar);
1987 mlx4_cleanup_uar_table(dev);
1991 static void mlx4_enable_msi_x(struct mlx4_dev *dev)
1993 struct mlx4_priv *priv = mlx4_priv(dev);
1994 struct msix_entry *entries;
1995 int nreq = min_t(int, dev->caps.num_ports *
1996 min_t(int, num_online_cpus() + 1,
1997 MAX_MSIX_P_PORT) + MSIX_LEGACY_SZ, MAX_MSIX);
2001 nreq = min_t(int, dev->caps.num_eqs - dev->caps.reserved_eqs,
2004 entries = kcalloc(nreq, sizeof *entries, GFP_KERNEL);
2008 for (i = 0; i < nreq; ++i)
2009 entries[i].entry = i;
2011 nreq = pci_enable_msix_range(dev->pdev, entries, 2, nreq);
2016 } else if (nreq < MSIX_LEGACY_SZ +
2017 dev->caps.num_ports * MIN_MSIX_P_PORT) {
2018 /*Working in legacy mode , all EQ's shared*/
2019 dev->caps.comp_pool = 0;
2020 dev->caps.num_comp_vectors = nreq - 1;
2022 dev->caps.comp_pool = nreq - MSIX_LEGACY_SZ;
2023 dev->caps.num_comp_vectors = MSIX_LEGACY_SZ - 1;
2025 for (i = 0; i < nreq; ++i)
2026 priv->eq_table.eq[i].irq = entries[i].vector;
2028 dev->flags |= MLX4_FLAG_MSI_X;
2035 dev->caps.num_comp_vectors = 1;
2036 dev->caps.comp_pool = 0;
2038 for (i = 0; i < 2; ++i)
2039 priv->eq_table.eq[i].irq = dev->pdev->irq;
2042 static int mlx4_init_port_info(struct mlx4_dev *dev, int port)
2044 struct mlx4_port_info *info = &mlx4_priv(dev)->port[port];
2049 if (!mlx4_is_slave(dev)) {
2050 mlx4_init_mac_table(dev, &info->mac_table);
2051 mlx4_init_vlan_table(dev, &info->vlan_table);
2052 mlx4_init_roce_gid_table(dev, &info->gid_table);
2053 info->base_qpn = mlx4_get_base_qpn(dev, port);
2056 sprintf(info->dev_name, "mlx4_port%d", port);
2057 info->port_attr.attr.name = info->dev_name;
2058 if (mlx4_is_mfunc(dev))
2059 info->port_attr.attr.mode = S_IRUGO;
2061 info->port_attr.attr.mode = S_IRUGO | S_IWUSR;
2062 info->port_attr.store = set_port_type;
2064 info->port_attr.show = show_port_type;
2065 sysfs_attr_init(&info->port_attr.attr);
2067 err = device_create_file(&dev->pdev->dev, &info->port_attr);
2069 mlx4_err(dev, "Failed to create file for port %d\n", port);
2073 sprintf(info->dev_mtu_name, "mlx4_port%d_mtu", port);
2074 info->port_mtu_attr.attr.name = info->dev_mtu_name;
2075 if (mlx4_is_mfunc(dev))
2076 info->port_mtu_attr.attr.mode = S_IRUGO;
2078 info->port_mtu_attr.attr.mode = S_IRUGO | S_IWUSR;
2079 info->port_mtu_attr.store = set_port_ib_mtu;
2081 info->port_mtu_attr.show = show_port_ib_mtu;
2082 sysfs_attr_init(&info->port_mtu_attr.attr);
2084 err = device_create_file(&dev->pdev->dev, &info->port_mtu_attr);
2086 mlx4_err(dev, "Failed to create mtu file for port %d\n", port);
2087 device_remove_file(&info->dev->pdev->dev, &info->port_attr);
2094 static void mlx4_cleanup_port_info(struct mlx4_port_info *info)
2099 device_remove_file(&info->dev->pdev->dev, &info->port_attr);
2100 device_remove_file(&info->dev->pdev->dev, &info->port_mtu_attr);
2103 static int mlx4_init_steering(struct mlx4_dev *dev)
2105 struct mlx4_priv *priv = mlx4_priv(dev);
2106 int num_entries = dev->caps.num_ports;
2109 priv->steer = kzalloc(sizeof(struct mlx4_steer) * num_entries, GFP_KERNEL);
2113 for (i = 0; i < num_entries; i++)
2114 for (j = 0; j < MLX4_NUM_STEERS; j++) {
2115 INIT_LIST_HEAD(&priv->steer[i].promisc_qps[j]);
2116 INIT_LIST_HEAD(&priv->steer[i].steer_entries[j]);
2121 static void mlx4_clear_steering(struct mlx4_dev *dev)
2123 struct mlx4_priv *priv = mlx4_priv(dev);
2124 struct mlx4_steer_index *entry, *tmp_entry;
2125 struct mlx4_promisc_qp *pqp, *tmp_pqp;
2126 int num_entries = dev->caps.num_ports;
2129 for (i = 0; i < num_entries; i++) {
2130 for (j = 0; j < MLX4_NUM_STEERS; j++) {
2131 list_for_each_entry_safe(pqp, tmp_pqp,
2132 &priv->steer[i].promisc_qps[j],
2134 list_del(&pqp->list);
2137 list_for_each_entry_safe(entry, tmp_entry,
2138 &priv->steer[i].steer_entries[j],
2140 list_del(&entry->list);
2141 list_for_each_entry_safe(pqp, tmp_pqp,
2144 list_del(&pqp->list);
2154 static int extended_func_num(struct pci_dev *pdev)
2156 return PCI_SLOT(pdev->devfn) * 8 + PCI_FUNC(pdev->devfn);
2159 #define MLX4_OWNER_BASE 0x8069c
2160 #define MLX4_OWNER_SIZE 4
2162 static int mlx4_get_ownership(struct mlx4_dev *dev)
2164 void __iomem *owner;
2167 if (pci_channel_offline(dev->pdev))
2170 owner = ioremap(pci_resource_start(dev->pdev, 0) + MLX4_OWNER_BASE,
2173 mlx4_err(dev, "Failed to obtain ownership bit\n");
2182 static void mlx4_free_ownership(struct mlx4_dev *dev)
2184 void __iomem *owner;
2186 if (pci_channel_offline(dev->pdev))
2189 owner = ioremap(pci_resource_start(dev->pdev, 0) + MLX4_OWNER_BASE,
2192 mlx4_err(dev, "Failed to obtain ownership bit\n");
2200 static int __mlx4_init_one(struct pci_dev *pdev, int pci_dev_data)
2202 struct mlx4_priv *priv;
2203 struct mlx4_dev *dev;
2206 int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
2207 int prb_vf[MLX4_MAX_PORTS + 1] = {0, 0, 0};
2208 const int param_map[MLX4_MAX_PORTS + 1][MLX4_MAX_PORTS + 1] = {
2209 {2, 0, 0}, {0, 1, 2}, {0, 1, 2} };
2210 unsigned total_vfs = 0;
2211 int sriov_initialized = 0;
2214 pr_info(DRV_NAME ": Initializing %s\n", pci_name(pdev));
2216 err = pci_enable_device(pdev);
2218 dev_err(&pdev->dev, "Cannot enable PCI device, aborting\n");
2222 /* Due to requirement that all VFs and the PF are *guaranteed* 2 MACS
2223 * per port, we must limit the number of VFs to 63 (since their are
2226 for (i = 0; i < sizeof(nvfs)/sizeof(nvfs[0]) && i < num_vfs_argc;
2227 total_vfs += nvfs[param_map[num_vfs_argc - 1][i]], i++) {
2228 nvfs[param_map[num_vfs_argc - 1][i]] = num_vfs[i];
2230 dev_err(&pdev->dev, "num_vfs module parameter cannot be negative\n");
2234 for (i = 0; i < sizeof(prb_vf)/sizeof(prb_vf[0]) && i < probe_vfs_argc;
2236 prb_vf[param_map[probe_vfs_argc - 1][i]] = probe_vf[i];
2237 if (prb_vf[i] < 0 || prb_vf[i] > nvfs[i]) {
2238 dev_err(&pdev->dev, "probe_vf module parameter cannot be negative or greater than num_vfs\n");
2242 if (total_vfs >= MLX4_MAX_NUM_VF) {
2244 "Requested more VF's (%d) than allowed (%d)\n",
2245 total_vfs, MLX4_MAX_NUM_VF - 1);
2249 for (i = 0; i < MLX4_MAX_PORTS; i++) {
2250 if (nvfs[i] + nvfs[2] >= MLX4_MAX_NUM_VF_P_PORT) {
2252 "Requested more VF's (%d) for port (%d) than allowed (%d)\n",
2253 nvfs[i] + nvfs[2], i + 1,
2254 MLX4_MAX_NUM_VF_P_PORT - 1);
2263 if (!(pci_dev_data & MLX4_PCI_DEV_IS_VF) &&
2264 !(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) {
2265 dev_err(&pdev->dev, "Missing DCS, aborting (driver_data: 0x%x, pci_resource_flags(pdev, 0):0x%lx)\n",
2266 pci_dev_data, pci_resource_flags(pdev, 0));
2268 goto err_disable_pdev;
2270 if (!(pci_resource_flags(pdev, 2) & IORESOURCE_MEM)) {
2271 dev_err(&pdev->dev, "Missing UAR, aborting\n");
2273 goto err_disable_pdev;
2276 err = pci_request_regions(pdev, DRV_NAME);
2278 dev_err(&pdev->dev, "Couldn't get PCI resources, aborting\n");
2279 goto err_disable_pdev;
2282 pci_set_master(pdev);
2284 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
2286 dev_warn(&pdev->dev, "Warning: couldn't set 64-bit PCI DMA mask\n");
2287 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
2289 dev_err(&pdev->dev, "Can't set PCI DMA mask, aborting\n");
2290 goto err_release_regions;
2293 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
2295 dev_warn(&pdev->dev, "Warning: couldn't set 64-bit consistent PCI DMA mask\n");
2296 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
2298 dev_err(&pdev->dev, "Can't set consistent PCI DMA mask, aborting\n");
2299 goto err_release_regions;
2303 /* Allow large DMA segments, up to the firmware limit of 1 GB */
2304 dma_set_max_seg_size(&pdev->dev, 1024 * 1024 * 1024);
2306 dev = pci_get_drvdata(pdev);
2307 priv = mlx4_priv(dev);
2309 INIT_LIST_HEAD(&priv->ctx_list);
2310 spin_lock_init(&priv->ctx_lock);
2312 mutex_init(&priv->port_mutex);
2314 INIT_LIST_HEAD(&priv->pgdir_list);
2315 mutex_init(&priv->pgdir_mutex);
2317 INIT_LIST_HEAD(&priv->bf_list);
2318 mutex_init(&priv->bf_mutex);
2320 dev->rev_id = pdev->revision;
2321 dev->numa_node = dev_to_node(&pdev->dev);
2322 /* Detect if this device is a virtual function */
2323 if (pci_dev_data & MLX4_PCI_DEV_IS_VF) {
2324 /* When acting as pf, we normally skip vfs unless explicitly
2325 * requested to probe them. */
2327 unsigned vfs_offset = 0;
2328 for (i = 0; i < sizeof(nvfs)/sizeof(nvfs[0]) &&
2329 vfs_offset + nvfs[i] < extended_func_num(pdev);
2330 vfs_offset += nvfs[i], i++)
2332 if (i == sizeof(nvfs)/sizeof(nvfs[0])) {
2336 if ((extended_func_num(pdev) - vfs_offset)
2338 mlx4_warn(dev, "Skipping virtual function:%d\n",
2339 extended_func_num(pdev));
2344 mlx4_warn(dev, "Detected virtual function - running in slave mode\n");
2345 dev->flags |= MLX4_FLAG_SLAVE;
2347 /* We reset the device and enable SRIOV only for physical
2348 * devices. Try to claim ownership on the device;
2349 * if already taken, skip -- do not allow multiple PFs */
2350 err = mlx4_get_ownership(dev);
2355 mlx4_warn(dev, "Multiple PFs not yet supported - Skipping PF\n");
2362 mlx4_warn(dev, "Enabling SR-IOV with %d VFs\n",
2364 dev->dev_vfs = kzalloc(
2365 total_vfs * sizeof(*dev->dev_vfs),
2367 if (NULL == dev->dev_vfs) {
2368 mlx4_err(dev, "Failed to allocate memory for VFs\n");
2371 atomic_inc(&pf_loading);
2372 err = pci_enable_sriov(pdev, total_vfs);
2374 mlx4_err(dev, "Failed to enable SR-IOV, continuing without SR-IOV (err = %d)\n",
2376 atomic_dec(&pf_loading);
2379 mlx4_warn(dev, "Running in master mode\n");
2380 dev->flags |= MLX4_FLAG_SRIOV |
2382 dev->num_vfs = total_vfs;
2383 sriov_initialized = 1;
2388 atomic_set(&priv->opreq_count, 0);
2389 INIT_WORK(&priv->opreq_task, mlx4_opreq_action);
2392 * Now reset the HCA before we touch the PCI capabilities or
2393 * attempt a firmware command, since a boot ROM may have left
2394 * the HCA in an undefined state.
2396 err = mlx4_reset(dev);
2398 mlx4_err(dev, "Failed to reset HCA, aborting\n");
2404 err = mlx4_cmd_init(dev);
2406 mlx4_err(dev, "Failed to init command interface, aborting\n");
2410 /* In slave functions, the communication channel must be initialized
2411 * before posting commands. Also, init num_slaves before calling
2413 if (mlx4_is_mfunc(dev)) {
2414 if (mlx4_is_master(dev))
2415 dev->num_slaves = MLX4_MAX_NUM_SLAVES;
2417 dev->num_slaves = 0;
2418 err = mlx4_multi_func_init(dev);
2420 mlx4_err(dev, "Failed to init slave mfunc interface, aborting\n");
2426 err = mlx4_init_hca(dev);
2428 if (err == -EACCES) {
2429 /* Not primary Physical function
2430 * Running in slave mode */
2431 mlx4_cmd_cleanup(dev);
2432 dev->flags |= MLX4_FLAG_SLAVE;
2433 dev->flags &= ~MLX4_FLAG_MASTER;
2439 /* check if the device is functioning at its maximum possible speed.
2440 * No return code for this call, just warn the user in case of PCI
2441 * express device capabilities are under-satisfied by the bus.
2443 if (!mlx4_is_slave(dev))
2444 mlx4_check_pcie_caps(dev);
2446 /* In master functions, the communication channel must be initialized
2447 * after obtaining its address from fw */
2448 if (mlx4_is_master(dev)) {
2450 err = mlx4_multi_func_init(dev);
2452 mlx4_err(dev, "Failed to init master mfunc interface, aborting\n");
2455 if (sriov_initialized) {
2457 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2461 (num_vfs_argc > 1 || probe_vfs_argc > 1)) {
2463 "Invalid syntax of num_vfs/probe_vfs with IB port - single port VFs syntax is only supported when all ports are configured as ethernet\n");
2465 goto err_master_mfunc;
2467 for (i = 0; i < sizeof(nvfs)/sizeof(nvfs[0]); i++) {
2469 for (j = 0; j < nvfs[i]; ++sum, ++j) {
2470 dev->dev_vfs[sum].min_port =
2472 dev->dev_vfs[sum].n_ports = i < 2 ? 1 :
2473 dev->caps.num_ports;
2479 err = mlx4_alloc_eq_table(dev);
2481 goto err_master_mfunc;
2483 priv->msix_ctl.pool_bm = 0;
2484 mutex_init(&priv->msix_ctl.pool_lock);
2486 mlx4_enable_msi_x(dev);
2487 if ((mlx4_is_mfunc(dev)) &&
2488 !(dev->flags & MLX4_FLAG_MSI_X)) {
2490 mlx4_err(dev, "INTx is not supported in multi-function mode, aborting\n");
2494 if (!mlx4_is_slave(dev)) {
2495 err = mlx4_init_steering(dev);
2500 err = mlx4_setup_hca(dev);
2501 if (err == -EBUSY && (dev->flags & MLX4_FLAG_MSI_X) &&
2502 !mlx4_is_mfunc(dev)) {
2503 dev->flags &= ~MLX4_FLAG_MSI_X;
2504 dev->caps.num_comp_vectors = 1;
2505 dev->caps.comp_pool = 0;
2506 pci_disable_msix(pdev);
2507 err = mlx4_setup_hca(dev);
2513 mlx4_init_quotas(dev);
2515 for (port = 1; port <= dev->caps.num_ports; port++) {
2516 err = mlx4_init_port_info(dev, port);
2521 err = mlx4_register_device(dev);
2525 mlx4_request_modules(dev);
2527 mlx4_sense_init(dev);
2528 mlx4_start_sense(dev);
2532 if (mlx4_is_master(dev) && dev->num_vfs)
2533 atomic_dec(&pf_loading);
2538 for (--port; port >= 1; --port)
2539 mlx4_cleanup_port_info(&priv->port[port]);
2541 mlx4_cleanup_counters_table(dev);
2542 mlx4_cleanup_qp_table(dev);
2543 mlx4_cleanup_srq_table(dev);
2544 mlx4_cleanup_cq_table(dev);
2545 mlx4_cmd_use_polling(dev);
2546 mlx4_cleanup_eq_table(dev);
2547 mlx4_cleanup_mcg_table(dev);
2548 mlx4_cleanup_mr_table(dev);
2549 mlx4_cleanup_xrcd_table(dev);
2550 mlx4_cleanup_pd_table(dev);
2551 mlx4_cleanup_uar_table(dev);
2554 if (!mlx4_is_slave(dev))
2555 mlx4_clear_steering(dev);
2558 mlx4_free_eq_table(dev);
2561 if (mlx4_is_master(dev))
2562 mlx4_multi_func_cleanup(dev);
2564 if (mlx4_is_slave(dev)) {
2565 kfree(dev->caps.qp0_qkey);
2566 kfree(dev->caps.qp0_tunnel);
2567 kfree(dev->caps.qp0_proxy);
2568 kfree(dev->caps.qp1_tunnel);
2569 kfree(dev->caps.qp1_proxy);
2573 if (dev->flags & MLX4_FLAG_MSI_X)
2574 pci_disable_msix(pdev);
2576 mlx4_close_hca(dev);
2579 if (mlx4_is_slave(dev))
2580 mlx4_multi_func_cleanup(dev);
2583 mlx4_cmd_cleanup(dev);
2586 if (dev->flags & MLX4_FLAG_SRIOV)
2587 pci_disable_sriov(pdev);
2590 if (!mlx4_is_slave(dev))
2591 mlx4_free_ownership(dev);
2593 if (mlx4_is_master(dev) && dev->num_vfs)
2594 atomic_dec(&pf_loading);
2596 kfree(priv->dev.dev_vfs);
2601 err_release_regions:
2602 pci_release_regions(pdev);
2605 pci_disable_device(pdev);
2606 pci_set_drvdata(pdev, NULL);
2610 static int mlx4_init_one(struct pci_dev *pdev, const struct pci_device_id *id)
2612 struct mlx4_priv *priv;
2613 struct mlx4_dev *dev;
2615 printk_once(KERN_INFO "%s", mlx4_version);
2617 priv = kzalloc(sizeof(*priv), GFP_KERNEL);
2622 pci_set_drvdata(pdev, dev);
2623 priv->pci_dev_data = id->driver_data;
2625 return __mlx4_init_one(pdev, id->driver_data);
2628 static void __mlx4_remove_one(struct pci_dev *pdev)
2630 struct mlx4_dev *dev = pci_get_drvdata(pdev);
2631 struct mlx4_priv *priv = mlx4_priv(dev);
2638 pci_dev_data = priv->pci_dev_data;
2640 /* in SRIOV it is not allowed to unload the pf's
2641 * driver while there are alive vf's */
2642 if (mlx4_is_master(dev) && mlx4_how_many_lives_vf(dev))
2643 pr_warn("Removing PF when there are assigned VF's !!!\n");
2644 mlx4_stop_sense(dev);
2645 mlx4_unregister_device(dev);
2647 for (p = 1; p <= dev->caps.num_ports; p++) {
2648 mlx4_cleanup_port_info(&priv->port[p]);
2649 mlx4_CLOSE_PORT(dev, p);
2652 if (mlx4_is_master(dev))
2653 mlx4_free_resource_tracker(dev,
2654 RES_TR_FREE_SLAVES_ONLY);
2656 mlx4_cleanup_counters_table(dev);
2657 mlx4_cleanup_qp_table(dev);
2658 mlx4_cleanup_srq_table(dev);
2659 mlx4_cleanup_cq_table(dev);
2660 mlx4_cmd_use_polling(dev);
2661 mlx4_cleanup_eq_table(dev);
2662 mlx4_cleanup_mcg_table(dev);
2663 mlx4_cleanup_mr_table(dev);
2664 mlx4_cleanup_xrcd_table(dev);
2665 mlx4_cleanup_pd_table(dev);
2667 if (mlx4_is_master(dev))
2668 mlx4_free_resource_tracker(dev,
2669 RES_TR_FREE_STRUCTS_ONLY);
2672 mlx4_uar_free(dev, &priv->driver_uar);
2673 mlx4_cleanup_uar_table(dev);
2674 if (!mlx4_is_slave(dev))
2675 mlx4_clear_steering(dev);
2676 mlx4_free_eq_table(dev);
2677 if (mlx4_is_master(dev))
2678 mlx4_multi_func_cleanup(dev);
2679 mlx4_close_hca(dev);
2680 if (mlx4_is_slave(dev))
2681 mlx4_multi_func_cleanup(dev);
2682 mlx4_cmd_cleanup(dev);
2684 if (dev->flags & MLX4_FLAG_MSI_X)
2685 pci_disable_msix(pdev);
2686 if (dev->flags & MLX4_FLAG_SRIOV) {
2687 mlx4_warn(dev, "Disabling SR-IOV\n");
2688 pci_disable_sriov(pdev);
2692 if (!mlx4_is_slave(dev))
2693 mlx4_free_ownership(dev);
2695 kfree(dev->caps.qp0_qkey);
2696 kfree(dev->caps.qp0_tunnel);
2697 kfree(dev->caps.qp0_proxy);
2698 kfree(dev->caps.qp1_tunnel);
2699 kfree(dev->caps.qp1_proxy);
2700 kfree(dev->dev_vfs);
2702 pci_release_regions(pdev);
2703 pci_disable_device(pdev);
2704 memset(priv, 0, sizeof(*priv));
2705 priv->pci_dev_data = pci_dev_data;
2709 static void mlx4_remove_one(struct pci_dev *pdev)
2711 struct mlx4_dev *dev = pci_get_drvdata(pdev);
2712 struct mlx4_priv *priv = mlx4_priv(dev);
2714 __mlx4_remove_one(pdev);
2716 pci_set_drvdata(pdev, NULL);
2719 int mlx4_restart_one(struct pci_dev *pdev)
2721 struct mlx4_dev *dev = pci_get_drvdata(pdev);
2722 struct mlx4_priv *priv = mlx4_priv(dev);
2725 pci_dev_data = priv->pci_dev_data;
2726 __mlx4_remove_one(pdev);
2727 return __mlx4_init_one(pdev, pci_dev_data);
2730 static DEFINE_PCI_DEVICE_TABLE(mlx4_pci_table) = {
2731 /* MT25408 "Hermon" SDR */
2732 { PCI_VDEVICE(MELLANOX, 0x6340), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2733 /* MT25408 "Hermon" DDR */
2734 { PCI_VDEVICE(MELLANOX, 0x634a), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2735 /* MT25408 "Hermon" QDR */
2736 { PCI_VDEVICE(MELLANOX, 0x6354), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2737 /* MT25408 "Hermon" DDR PCIe gen2 */
2738 { PCI_VDEVICE(MELLANOX, 0x6732), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2739 /* MT25408 "Hermon" QDR PCIe gen2 */
2740 { PCI_VDEVICE(MELLANOX, 0x673c), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2741 /* MT25408 "Hermon" EN 10GigE */
2742 { PCI_VDEVICE(MELLANOX, 0x6368), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2743 /* MT25408 "Hermon" EN 10GigE PCIe gen2 */
2744 { PCI_VDEVICE(MELLANOX, 0x6750), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2745 /* MT25458 ConnectX EN 10GBASE-T 10GigE */
2746 { PCI_VDEVICE(MELLANOX, 0x6372), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2747 /* MT25458 ConnectX EN 10GBASE-T+Gen2 10GigE */
2748 { PCI_VDEVICE(MELLANOX, 0x675a), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2749 /* MT26468 ConnectX EN 10GigE PCIe gen2*/
2750 { PCI_VDEVICE(MELLANOX, 0x6764), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2751 /* MT26438 ConnectX EN 40GigE PCIe gen2 5GT/s */
2752 { PCI_VDEVICE(MELLANOX, 0x6746), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2753 /* MT26478 ConnectX2 40GigE PCIe gen2 */
2754 { PCI_VDEVICE(MELLANOX, 0x676e), MLX4_PCI_DEV_FORCE_SENSE_PORT },
2755 /* MT25400 Family [ConnectX-2 Virtual Function] */
2756 { PCI_VDEVICE(MELLANOX, 0x1002), MLX4_PCI_DEV_IS_VF },
2757 /* MT27500 Family [ConnectX-3] */
2758 { PCI_VDEVICE(MELLANOX, 0x1003), 0 },
2759 /* MT27500 Family [ConnectX-3 Virtual Function] */
2760 { PCI_VDEVICE(MELLANOX, 0x1004), MLX4_PCI_DEV_IS_VF },
2761 { PCI_VDEVICE(MELLANOX, 0x1005), 0 }, /* MT27510 Family */
2762 { PCI_VDEVICE(MELLANOX, 0x1006), 0 }, /* MT27511 Family */
2763 { PCI_VDEVICE(MELLANOX, 0x1007), 0 }, /* MT27520 Family */
2764 { PCI_VDEVICE(MELLANOX, 0x1008), 0 }, /* MT27521 Family */
2765 { PCI_VDEVICE(MELLANOX, 0x1009), 0 }, /* MT27530 Family */
2766 { PCI_VDEVICE(MELLANOX, 0x100a), 0 }, /* MT27531 Family */
2767 { PCI_VDEVICE(MELLANOX, 0x100b), 0 }, /* MT27540 Family */
2768 { PCI_VDEVICE(MELLANOX, 0x100c), 0 }, /* MT27541 Family */
2769 { PCI_VDEVICE(MELLANOX, 0x100d), 0 }, /* MT27550 Family */
2770 { PCI_VDEVICE(MELLANOX, 0x100e), 0 }, /* MT27551 Family */
2771 { PCI_VDEVICE(MELLANOX, 0x100f), 0 }, /* MT27560 Family */
2772 { PCI_VDEVICE(MELLANOX, 0x1010), 0 }, /* MT27561 Family */
2776 MODULE_DEVICE_TABLE(pci, mlx4_pci_table);
2778 static pci_ers_result_t mlx4_pci_err_detected(struct pci_dev *pdev,
2779 pci_channel_state_t state)
2781 __mlx4_remove_one(pdev);
2783 return state == pci_channel_io_perm_failure ?
2784 PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_NEED_RESET;
2787 static pci_ers_result_t mlx4_pci_slot_reset(struct pci_dev *pdev)
2789 struct mlx4_dev *dev = pci_get_drvdata(pdev);
2790 struct mlx4_priv *priv = mlx4_priv(dev);
2793 ret = __mlx4_init_one(pdev, priv->pci_dev_data);
2795 return ret ? PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_RECOVERED;
2798 static const struct pci_error_handlers mlx4_err_handler = {
2799 .error_detected = mlx4_pci_err_detected,
2800 .slot_reset = mlx4_pci_slot_reset,
2803 static struct pci_driver mlx4_driver = {
2805 .id_table = mlx4_pci_table,
2806 .probe = mlx4_init_one,
2807 .shutdown = __mlx4_remove_one,
2808 .remove = mlx4_remove_one,
2809 .err_handler = &mlx4_err_handler,
2812 static int __init mlx4_verify_params(void)
2814 if ((log_num_mac < 0) || (log_num_mac > 7)) {
2815 pr_warn("mlx4_core: bad num_mac: %d\n", log_num_mac);
2819 if (log_num_vlan != 0)
2820 pr_warn("mlx4_core: log_num_vlan - obsolete module param, using %d\n",
2821 MLX4_LOG_NUM_VLANS);
2824 pr_warn("mlx4_core: use_prio - obsolete module param, ignored\n");
2826 if ((log_mtts_per_seg < 1) || (log_mtts_per_seg > 7)) {
2827 pr_warn("mlx4_core: bad log_mtts_per_seg: %d\n",
2832 /* Check if module param for ports type has legal combination */
2833 if (port_type_array[0] == false && port_type_array[1] == true) {
2834 pr_warn("Module parameter configuration ETH/IB is not supported. Switching to default configuration IB/IB\n");
2835 port_type_array[0] = true;
2838 if (mlx4_log_num_mgm_entry_size != -1 &&
2839 (mlx4_log_num_mgm_entry_size < MLX4_MIN_MGM_LOG_ENTRY_SIZE ||
2840 mlx4_log_num_mgm_entry_size > MLX4_MAX_MGM_LOG_ENTRY_SIZE)) {
2841 pr_warn("mlx4_core: mlx4_log_num_mgm_entry_size (%d) not in legal range (-1 or %d..%d)\n",
2842 mlx4_log_num_mgm_entry_size,
2843 MLX4_MIN_MGM_LOG_ENTRY_SIZE,
2844 MLX4_MAX_MGM_LOG_ENTRY_SIZE);
2851 static int __init mlx4_init(void)
2855 if (mlx4_verify_params())
2860 mlx4_wq = create_singlethread_workqueue("mlx4");
2864 ret = pci_register_driver(&mlx4_driver);
2866 destroy_workqueue(mlx4_wq);
2867 return ret < 0 ? ret : 0;
2870 static void __exit mlx4_cleanup(void)
2872 pci_unregister_driver(&mlx4_driver);
2873 destroy_workqueue(mlx4_wq);
2876 module_init(mlx4_init);
2877 module_exit(mlx4_cleanup);