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>
42 #include <linux/mlx4/device.h>
43 #include <linux/mlx4/doorbell.h>
49 MODULE_AUTHOR("Roland Dreier");
50 MODULE_DESCRIPTION("Mellanox ConnectX HCA low-level driver");
51 MODULE_LICENSE("Dual BSD/GPL");
52 MODULE_VERSION(DRV_VERSION);
54 #ifdef CONFIG_MLX4_DEBUG
56 int mlx4_debug_level = 0;
57 module_param_named(debug_level, mlx4_debug_level, int, 0644);
58 MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0");
60 #endif /* CONFIG_MLX4_DEBUG */
65 module_param(msi_x, int, 0444);
66 MODULE_PARM_DESC(msi_x, "attempt to use MSI-X if nonzero");
68 #else /* CONFIG_PCI_MSI */
72 #endif /* CONFIG_PCI_MSI */
74 static char mlx4_version[] __devinitdata =
75 DRV_NAME ": Mellanox ConnectX core driver v"
76 DRV_VERSION " (" DRV_RELDATE ")\n";
78 static struct mlx4_profile default_profile = {
81 .rdmarc_per_qp = 1 << 4,
88 static int log_num_mac = 2;
89 module_param_named(log_num_mac, log_num_mac, int, 0444);
90 MODULE_PARM_DESC(log_num_mac, "Log2 max number of MACs per ETH port (1-7)");
92 static int log_num_vlan;
93 module_param_named(log_num_vlan, log_num_vlan, int, 0444);
94 MODULE_PARM_DESC(log_num_vlan, "Log2 max number of VLANs per ETH port (0-7)");
97 module_param_named(use_prio, use_prio, bool, 0444);
98 MODULE_PARM_DESC(use_prio, "Enable steering by VLAN priority on ETH ports "
101 static int mlx4_check_port_params(struct mlx4_dev *dev,
102 enum mlx4_port_type *port_type)
106 for (i = 0; i < dev->caps.num_ports - 1; i++) {
107 if (port_type[i] != port_type[i+1] &&
108 !(dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP)) {
109 mlx4_err(dev, "Only same port types supported "
110 "on this HCA, aborting.\n");
114 if ((port_type[0] == MLX4_PORT_TYPE_ETH) &&
115 (port_type[1] == MLX4_PORT_TYPE_IB)) {
116 mlx4_err(dev, "eth-ib configuration is not supported.\n");
120 for (i = 0; i < dev->caps.num_ports; i++) {
121 if (!(port_type[i] & dev->caps.supported_type[i+1])) {
122 mlx4_err(dev, "Requested port type for port %d is not "
123 "supported on this HCA\n", i + 1);
130 static void mlx4_set_port_mask(struct mlx4_dev *dev)
134 dev->caps.port_mask = 0;
135 for (i = 1; i <= dev->caps.num_ports; ++i)
136 if (dev->caps.port_type[i] == MLX4_PORT_TYPE_IB)
137 dev->caps.port_mask |= 1 << (i - 1);
139 static int mlx4_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
144 err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
146 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
150 if (dev_cap->min_page_sz > PAGE_SIZE) {
151 mlx4_err(dev, "HCA minimum page size of %d bigger than "
152 "kernel PAGE_SIZE of %ld, aborting.\n",
153 dev_cap->min_page_sz, PAGE_SIZE);
156 if (dev_cap->num_ports > MLX4_MAX_PORTS) {
157 mlx4_err(dev, "HCA has %d ports, but we only support %d, "
159 dev_cap->num_ports, MLX4_MAX_PORTS);
163 if (dev_cap->uar_size > pci_resource_len(dev->pdev, 2)) {
164 mlx4_err(dev, "HCA reported UAR size of 0x%x bigger than "
165 "PCI resource 2 size of 0x%llx, aborting.\n",
167 (unsigned long long) pci_resource_len(dev->pdev, 2));
171 dev->caps.num_ports = dev_cap->num_ports;
172 for (i = 1; i <= dev->caps.num_ports; ++i) {
173 dev->caps.vl_cap[i] = dev_cap->max_vl[i];
174 dev->caps.ib_mtu_cap[i] = dev_cap->ib_mtu[i];
175 dev->caps.gid_table_len[i] = dev_cap->max_gids[i];
176 dev->caps.pkey_table_len[i] = dev_cap->max_pkeys[i];
177 dev->caps.port_width_cap[i] = dev_cap->max_port_width[i];
178 dev->caps.eth_mtu_cap[i] = dev_cap->eth_mtu[i];
179 dev->caps.def_mac[i] = dev_cap->def_mac[i];
180 dev->caps.supported_type[i] = dev_cap->supported_port_types[i];
183 dev->caps.num_uars = dev_cap->uar_size / PAGE_SIZE;
184 dev->caps.local_ca_ack_delay = dev_cap->local_ca_ack_delay;
185 dev->caps.bf_reg_size = dev_cap->bf_reg_size;
186 dev->caps.bf_regs_per_page = dev_cap->bf_regs_per_page;
187 dev->caps.max_sq_sg = dev_cap->max_sq_sg;
188 dev->caps.max_rq_sg = dev_cap->max_rq_sg;
189 dev->caps.max_wqes = dev_cap->max_qp_sz;
190 dev->caps.max_qp_init_rdma = dev_cap->max_requester_per_qp;
191 dev->caps.max_srq_wqes = dev_cap->max_srq_sz;
192 dev->caps.max_srq_sge = dev_cap->max_rq_sg - 1;
193 dev->caps.reserved_srqs = dev_cap->reserved_srqs;
194 dev->caps.max_sq_desc_sz = dev_cap->max_sq_desc_sz;
195 dev->caps.max_rq_desc_sz = dev_cap->max_rq_desc_sz;
196 dev->caps.num_qp_per_mgm = MLX4_QP_PER_MGM;
198 * Subtract 1 from the limit because we need to allocate a
199 * spare CQE so the HCA HW can tell the difference between an
200 * empty CQ and a full CQ.
202 dev->caps.max_cqes = dev_cap->max_cq_sz - 1;
203 dev->caps.reserved_cqs = dev_cap->reserved_cqs;
204 dev->caps.reserved_eqs = dev_cap->reserved_eqs;
205 dev->caps.reserved_mtts = DIV_ROUND_UP(dev_cap->reserved_mtts,
206 MLX4_MTT_ENTRY_PER_SEG);
207 dev->caps.reserved_mrws = dev_cap->reserved_mrws;
208 dev->caps.reserved_uars = dev_cap->reserved_uars;
209 dev->caps.reserved_pds = dev_cap->reserved_pds;
210 dev->caps.mtt_entry_sz = MLX4_MTT_ENTRY_PER_SEG * dev_cap->mtt_entry_sz;
211 dev->caps.max_msg_sz = dev_cap->max_msg_sz;
212 dev->caps.page_size_cap = ~(u32) (dev_cap->min_page_sz - 1);
213 dev->caps.flags = dev_cap->flags;
214 dev->caps.bmme_flags = dev_cap->bmme_flags;
215 dev->caps.reserved_lkey = dev_cap->reserved_lkey;
216 dev->caps.stat_rate_support = dev_cap->stat_rate_support;
217 dev->caps.max_gso_sz = dev_cap->max_gso_sz;
219 dev->caps.log_num_macs = log_num_mac;
220 dev->caps.log_num_vlans = log_num_vlan;
221 dev->caps.log_num_prios = use_prio ? 3 : 0;
223 for (i = 1; i <= dev->caps.num_ports; ++i) {
224 if (dev->caps.supported_type[i] != MLX4_PORT_TYPE_ETH)
225 dev->caps.port_type[i] = MLX4_PORT_TYPE_IB;
227 dev->caps.port_type[i] = MLX4_PORT_TYPE_ETH;
229 if (dev->caps.log_num_macs > dev_cap->log_max_macs[i]) {
230 dev->caps.log_num_macs = dev_cap->log_max_macs[i];
231 mlx4_warn(dev, "Requested number of MACs is too much "
232 "for port %d, reducing to %d.\n",
233 i, 1 << dev->caps.log_num_macs);
235 if (dev->caps.log_num_vlans > dev_cap->log_max_vlans[i]) {
236 dev->caps.log_num_vlans = dev_cap->log_max_vlans[i];
237 mlx4_warn(dev, "Requested number of VLANs is too much "
238 "for port %d, reducing to %d.\n",
239 i, 1 << dev->caps.log_num_vlans);
243 mlx4_set_port_mask(dev);
245 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] = dev_cap->reserved_qps;
246 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] =
247 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] =
248 (1 << dev->caps.log_num_macs) *
249 (1 << dev->caps.log_num_vlans) *
250 (1 << dev->caps.log_num_prios) *
252 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH] = MLX4_NUM_FEXCH;
254 dev->caps.reserved_qps = dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] +
255 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] +
256 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] +
257 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH];
263 * Change the port configuration of the device.
264 * Every user of this function must hold the port mutex.
266 static int mlx4_change_port_types(struct mlx4_dev *dev,
267 enum mlx4_port_type *port_types)
273 for (port = 0; port < dev->caps.num_ports; port++) {
274 if (port_types[port] != dev->caps.port_type[port + 1]) {
276 dev->caps.port_type[port + 1] = port_types[port];
280 mlx4_unregister_device(dev);
281 for (port = 1; port <= dev->caps.num_ports; port++) {
282 mlx4_CLOSE_PORT(dev, port);
283 err = mlx4_SET_PORT(dev, port);
285 mlx4_err(dev, "Failed to set port %d, "
290 mlx4_set_port_mask(dev);
291 err = mlx4_register_device(dev);
298 static ssize_t show_port_type(struct device *dev,
299 struct device_attribute *attr,
302 struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
304 struct mlx4_dev *mdev = info->dev;
306 return sprintf(buf, "%s\n",
307 mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_IB ?
311 static ssize_t set_port_type(struct device *dev,
312 struct device_attribute *attr,
313 const char *buf, size_t count)
315 struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
317 struct mlx4_dev *mdev = info->dev;
318 struct mlx4_priv *priv = mlx4_priv(mdev);
319 enum mlx4_port_type types[MLX4_MAX_PORTS];
323 if (!strcmp(buf, "ib\n"))
324 info->tmp_type = MLX4_PORT_TYPE_IB;
325 else if (!strcmp(buf, "eth\n"))
326 info->tmp_type = MLX4_PORT_TYPE_ETH;
328 mlx4_err(mdev, "%s is not supported port type\n", buf);
332 mutex_lock(&priv->port_mutex);
333 for (i = 0; i < mdev->caps.num_ports; i++)
334 types[i] = priv->port[i+1].tmp_type ? priv->port[i+1].tmp_type :
335 mdev->caps.port_type[i+1];
337 err = mlx4_check_port_params(mdev, types);
341 for (i = 1; i <= mdev->caps.num_ports; i++)
342 priv->port[i].tmp_type = 0;
344 err = mlx4_change_port_types(mdev, types);
347 mutex_unlock(&priv->port_mutex);
348 return err ? err : count;
351 static int mlx4_load_fw(struct mlx4_dev *dev)
353 struct mlx4_priv *priv = mlx4_priv(dev);
356 priv->fw.fw_icm = mlx4_alloc_icm(dev, priv->fw.fw_pages,
357 GFP_HIGHUSER | __GFP_NOWARN, 0);
358 if (!priv->fw.fw_icm) {
359 mlx4_err(dev, "Couldn't allocate FW area, aborting.\n");
363 err = mlx4_MAP_FA(dev, priv->fw.fw_icm);
365 mlx4_err(dev, "MAP_FA command failed, aborting.\n");
369 err = mlx4_RUN_FW(dev);
371 mlx4_err(dev, "RUN_FW command failed, aborting.\n");
381 mlx4_free_icm(dev, priv->fw.fw_icm, 0);
385 static int mlx4_init_cmpt_table(struct mlx4_dev *dev, u64 cmpt_base,
388 struct mlx4_priv *priv = mlx4_priv(dev);
391 err = mlx4_init_icm_table(dev, &priv->qp_table.cmpt_table,
393 ((u64) (MLX4_CMPT_TYPE_QP *
394 cmpt_entry_sz) << MLX4_CMPT_SHIFT),
395 cmpt_entry_sz, dev->caps.num_qps,
396 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
401 err = mlx4_init_icm_table(dev, &priv->srq_table.cmpt_table,
403 ((u64) (MLX4_CMPT_TYPE_SRQ *
404 cmpt_entry_sz) << MLX4_CMPT_SHIFT),
405 cmpt_entry_sz, dev->caps.num_srqs,
406 dev->caps.reserved_srqs, 0, 0);
410 err = mlx4_init_icm_table(dev, &priv->cq_table.cmpt_table,
412 ((u64) (MLX4_CMPT_TYPE_CQ *
413 cmpt_entry_sz) << MLX4_CMPT_SHIFT),
414 cmpt_entry_sz, dev->caps.num_cqs,
415 dev->caps.reserved_cqs, 0, 0);
419 err = mlx4_init_icm_table(dev, &priv->eq_table.cmpt_table,
421 ((u64) (MLX4_CMPT_TYPE_EQ *
422 cmpt_entry_sz) << MLX4_CMPT_SHIFT),
424 roundup_pow_of_two(MLX4_NUM_EQ +
425 dev->caps.reserved_eqs),
426 MLX4_NUM_EQ + dev->caps.reserved_eqs, 0, 0);
433 mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
436 mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
439 mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
445 static int mlx4_init_icm(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap,
446 struct mlx4_init_hca_param *init_hca, u64 icm_size)
448 struct mlx4_priv *priv = mlx4_priv(dev);
452 err = mlx4_SET_ICM_SIZE(dev, icm_size, &aux_pages);
454 mlx4_err(dev, "SET_ICM_SIZE command failed, aborting.\n");
458 mlx4_dbg(dev, "%lld KB of HCA context requires %lld KB aux memory.\n",
459 (unsigned long long) icm_size >> 10,
460 (unsigned long long) aux_pages << 2);
462 priv->fw.aux_icm = mlx4_alloc_icm(dev, aux_pages,
463 GFP_HIGHUSER | __GFP_NOWARN, 0);
464 if (!priv->fw.aux_icm) {
465 mlx4_err(dev, "Couldn't allocate aux memory, aborting.\n");
469 err = mlx4_MAP_ICM_AUX(dev, priv->fw.aux_icm);
471 mlx4_err(dev, "MAP_ICM_AUX command failed, aborting.\n");
475 err = mlx4_init_cmpt_table(dev, init_hca->cmpt_base, dev_cap->cmpt_entry_sz);
477 mlx4_err(dev, "Failed to map cMPT context memory, aborting.\n");
481 err = mlx4_map_eq_icm(dev, init_hca->eqc_base);
483 mlx4_err(dev, "Failed to map EQ context memory, aborting.\n");
488 * Reserved MTT entries must be aligned up to a cacheline
489 * boundary, since the FW will write to them, while the driver
490 * writes to all other MTT entries. (The variable
491 * dev->caps.mtt_entry_sz below is really the MTT segment
492 * size, not the raw entry size)
494 dev->caps.reserved_mtts =
495 ALIGN(dev->caps.reserved_mtts * dev->caps.mtt_entry_sz,
496 dma_get_cache_alignment()) / dev->caps.mtt_entry_sz;
498 err = mlx4_init_icm_table(dev, &priv->mr_table.mtt_table,
500 dev->caps.mtt_entry_sz,
501 dev->caps.num_mtt_segs,
502 dev->caps.reserved_mtts, 1, 0);
504 mlx4_err(dev, "Failed to map MTT context memory, aborting.\n");
508 err = mlx4_init_icm_table(dev, &priv->mr_table.dmpt_table,
510 dev_cap->dmpt_entry_sz,
512 dev->caps.reserved_mrws, 1, 1);
514 mlx4_err(dev, "Failed to map dMPT context memory, aborting.\n");
518 err = mlx4_init_icm_table(dev, &priv->qp_table.qp_table,
520 dev_cap->qpc_entry_sz,
522 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
525 mlx4_err(dev, "Failed to map QP context memory, aborting.\n");
529 err = mlx4_init_icm_table(dev, &priv->qp_table.auxc_table,
531 dev_cap->aux_entry_sz,
533 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
536 mlx4_err(dev, "Failed to map AUXC context memory, aborting.\n");
540 err = mlx4_init_icm_table(dev, &priv->qp_table.altc_table,
542 dev_cap->altc_entry_sz,
544 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
547 mlx4_err(dev, "Failed to map ALTC context memory, aborting.\n");
551 err = mlx4_init_icm_table(dev, &priv->qp_table.rdmarc_table,
552 init_hca->rdmarc_base,
553 dev_cap->rdmarc_entry_sz << priv->qp_table.rdmarc_shift,
555 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
558 mlx4_err(dev, "Failed to map RDMARC context memory, aborting\n");
562 err = mlx4_init_icm_table(dev, &priv->cq_table.table,
564 dev_cap->cqc_entry_sz,
566 dev->caps.reserved_cqs, 0, 0);
568 mlx4_err(dev, "Failed to map CQ context memory, aborting.\n");
569 goto err_unmap_rdmarc;
572 err = mlx4_init_icm_table(dev, &priv->srq_table.table,
574 dev_cap->srq_entry_sz,
576 dev->caps.reserved_srqs, 0, 0);
578 mlx4_err(dev, "Failed to map SRQ context memory, aborting.\n");
583 * It's not strictly required, but for simplicity just map the
584 * whole multicast group table now. The table isn't very big
585 * and it's a lot easier than trying to track ref counts.
587 err = mlx4_init_icm_table(dev, &priv->mcg_table.table,
588 init_hca->mc_base, MLX4_MGM_ENTRY_SIZE,
589 dev->caps.num_mgms + dev->caps.num_amgms,
590 dev->caps.num_mgms + dev->caps.num_amgms,
593 mlx4_err(dev, "Failed to map MCG context memory, aborting.\n");
600 mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
603 mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
606 mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
609 mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
612 mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
615 mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
618 mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
621 mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
624 mlx4_unmap_eq_icm(dev);
627 mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
628 mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
629 mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
630 mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
633 mlx4_UNMAP_ICM_AUX(dev);
636 mlx4_free_icm(dev, priv->fw.aux_icm, 0);
641 static void mlx4_free_icms(struct mlx4_dev *dev)
643 struct mlx4_priv *priv = mlx4_priv(dev);
645 mlx4_cleanup_icm_table(dev, &priv->mcg_table.table);
646 mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
647 mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
648 mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
649 mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
650 mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
651 mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
652 mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
653 mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
654 mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
655 mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
656 mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
657 mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
658 mlx4_unmap_eq_icm(dev);
660 mlx4_UNMAP_ICM_AUX(dev);
661 mlx4_free_icm(dev, priv->fw.aux_icm, 0);
664 static void mlx4_close_hca(struct mlx4_dev *dev)
666 mlx4_CLOSE_HCA(dev, 0);
669 mlx4_free_icm(dev, mlx4_priv(dev)->fw.fw_icm, 0);
672 static int mlx4_init_hca(struct mlx4_dev *dev)
674 struct mlx4_priv *priv = mlx4_priv(dev);
675 struct mlx4_adapter adapter;
676 struct mlx4_dev_cap dev_cap;
677 struct mlx4_mod_stat_cfg mlx4_cfg;
678 struct mlx4_profile profile;
679 struct mlx4_init_hca_param init_hca;
683 err = mlx4_QUERY_FW(dev);
685 mlx4_err(dev, "QUERY_FW command failed, aborting.\n");
689 err = mlx4_load_fw(dev);
691 mlx4_err(dev, "Failed to start FW, aborting.\n");
695 mlx4_cfg.log_pg_sz_m = 1;
696 mlx4_cfg.log_pg_sz = 0;
697 err = mlx4_MOD_STAT_CFG(dev, &mlx4_cfg);
699 mlx4_warn(dev, "Failed to override log_pg_sz parameter\n");
701 err = mlx4_dev_cap(dev, &dev_cap);
703 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
707 profile = default_profile;
709 icm_size = mlx4_make_profile(dev, &profile, &dev_cap, &init_hca);
710 if ((long long) icm_size < 0) {
715 init_hca.log_uar_sz = ilog2(dev->caps.num_uars);
717 err = mlx4_init_icm(dev, &dev_cap, &init_hca, icm_size);
721 err = mlx4_INIT_HCA(dev, &init_hca);
723 mlx4_err(dev, "INIT_HCA command failed, aborting.\n");
727 err = mlx4_QUERY_ADAPTER(dev, &adapter);
729 mlx4_err(dev, "QUERY_ADAPTER command failed, aborting.\n");
733 priv->eq_table.inta_pin = adapter.inta_pin;
734 memcpy(dev->board_id, adapter.board_id, sizeof dev->board_id);
746 mlx4_free_icm(dev, priv->fw.fw_icm, 0);
751 static int mlx4_setup_hca(struct mlx4_dev *dev)
753 struct mlx4_priv *priv = mlx4_priv(dev);
756 __be32 ib_port_default_caps;
758 err = mlx4_init_uar_table(dev);
760 mlx4_err(dev, "Failed to initialize "
761 "user access region table, aborting.\n");
765 err = mlx4_uar_alloc(dev, &priv->driver_uar);
767 mlx4_err(dev, "Failed to allocate driver access region, "
769 goto err_uar_table_free;
772 priv->kar = ioremap(priv->driver_uar.pfn << PAGE_SHIFT, PAGE_SIZE);
774 mlx4_err(dev, "Couldn't map kernel access region, "
780 err = mlx4_init_pd_table(dev);
782 mlx4_err(dev, "Failed to initialize "
783 "protection domain table, aborting.\n");
787 err = mlx4_init_mr_table(dev);
789 mlx4_err(dev, "Failed to initialize "
790 "memory region table, aborting.\n");
791 goto err_pd_table_free;
794 err = mlx4_init_eq_table(dev);
796 mlx4_err(dev, "Failed to initialize "
797 "event queue table, aborting.\n");
798 goto err_mr_table_free;
801 err = mlx4_cmd_use_events(dev);
803 mlx4_err(dev, "Failed to switch to event-driven "
804 "firmware commands, aborting.\n");
805 goto err_eq_table_free;
810 if (dev->flags & MLX4_FLAG_MSI_X) {
811 mlx4_warn(dev, "NOP command failed to generate MSI-X "
812 "interrupt IRQ %d).\n",
813 priv->eq_table.eq[MLX4_EQ_ASYNC].irq);
814 mlx4_warn(dev, "Trying again without MSI-X.\n");
816 mlx4_err(dev, "NOP command failed to generate interrupt "
817 "(IRQ %d), aborting.\n",
818 priv->eq_table.eq[MLX4_EQ_ASYNC].irq);
819 mlx4_err(dev, "BIOS or ACPI interrupt routing problem?\n");
825 mlx4_dbg(dev, "NOP command IRQ test passed\n");
827 err = mlx4_init_cq_table(dev);
829 mlx4_err(dev, "Failed to initialize "
830 "completion queue table, aborting.\n");
834 err = mlx4_init_srq_table(dev);
836 mlx4_err(dev, "Failed to initialize "
837 "shared receive queue table, aborting.\n");
838 goto err_cq_table_free;
841 err = mlx4_init_qp_table(dev);
843 mlx4_err(dev, "Failed to initialize "
844 "queue pair table, aborting.\n");
845 goto err_srq_table_free;
848 err = mlx4_init_mcg_table(dev);
850 mlx4_err(dev, "Failed to initialize "
851 "multicast group table, aborting.\n");
852 goto err_qp_table_free;
855 for (port = 1; port <= dev->caps.num_ports; port++) {
856 ib_port_default_caps = 0;
857 err = mlx4_get_port_ib_caps(dev, port, &ib_port_default_caps);
859 mlx4_warn(dev, "failed to get port %d default "
860 "ib capabilities (%d). Continuing with "
861 "caps = 0\n", port, err);
862 dev->caps.ib_port_def_cap[port] = ib_port_default_caps;
863 err = mlx4_SET_PORT(dev, port);
865 mlx4_err(dev, "Failed to set port %d, aborting\n",
867 goto err_mcg_table_free;
874 mlx4_cleanup_mcg_table(dev);
877 mlx4_cleanup_qp_table(dev);
880 mlx4_cleanup_srq_table(dev);
883 mlx4_cleanup_cq_table(dev);
886 mlx4_cmd_use_polling(dev);
889 mlx4_cleanup_eq_table(dev);
892 mlx4_cleanup_mr_table(dev);
895 mlx4_cleanup_pd_table(dev);
901 mlx4_uar_free(dev, &priv->driver_uar);
904 mlx4_cleanup_uar_table(dev);
908 static void mlx4_enable_msi_x(struct mlx4_dev *dev)
910 struct mlx4_priv *priv = mlx4_priv(dev);
911 struct msix_entry entries[MLX4_NUM_EQ];
916 for (i = 0; i < MLX4_NUM_EQ; ++i)
917 entries[i].entry = i;
919 err = pci_enable_msix(dev->pdev, entries, ARRAY_SIZE(entries));
922 mlx4_info(dev, "Only %d MSI-X vectors available, "
923 "not using MSI-X\n", err);
927 for (i = 0; i < MLX4_NUM_EQ; ++i)
928 priv->eq_table.eq[i].irq = entries[i].vector;
930 dev->flags |= MLX4_FLAG_MSI_X;
935 for (i = 0; i < MLX4_NUM_EQ; ++i)
936 priv->eq_table.eq[i].irq = dev->pdev->irq;
939 static int mlx4_init_port_info(struct mlx4_dev *dev, int port)
941 struct mlx4_port_info *info = &mlx4_priv(dev)->port[port];
946 mlx4_init_mac_table(dev, &info->mac_table);
947 mlx4_init_vlan_table(dev, &info->vlan_table);
949 sprintf(info->dev_name, "mlx4_port%d", port);
950 info->port_attr.attr.name = info->dev_name;
951 info->port_attr.attr.mode = S_IRUGO | S_IWUSR;
952 info->port_attr.show = show_port_type;
953 info->port_attr.store = set_port_type;
955 err = device_create_file(&dev->pdev->dev, &info->port_attr);
957 mlx4_err(dev, "Failed to create file for port %d\n", port);
964 static void mlx4_cleanup_port_info(struct mlx4_port_info *info)
969 device_remove_file(&info->dev->pdev->dev, &info->port_attr);
972 static int __mlx4_init_one(struct pci_dev *pdev, const struct pci_device_id *id)
974 struct mlx4_priv *priv;
975 struct mlx4_dev *dev;
979 printk(KERN_INFO PFX "Initializing %s\n",
982 err = pci_enable_device(pdev);
984 dev_err(&pdev->dev, "Cannot enable PCI device, "
990 * Check for BARs. We expect 0: 1MB
992 if (!(pci_resource_flags(pdev, 0) & IORESOURCE_MEM) ||
993 pci_resource_len(pdev, 0) != 1 << 20) {
994 dev_err(&pdev->dev, "Missing DCS, aborting.\n");
996 goto err_disable_pdev;
998 if (!(pci_resource_flags(pdev, 2) & IORESOURCE_MEM)) {
999 dev_err(&pdev->dev, "Missing UAR, aborting.\n");
1001 goto err_disable_pdev;
1004 err = pci_request_region(pdev, 0, DRV_NAME);
1006 dev_err(&pdev->dev, "Cannot request control region, aborting.\n");
1007 goto err_disable_pdev;
1010 err = pci_request_region(pdev, 2, DRV_NAME);
1012 dev_err(&pdev->dev, "Cannot request UAR region, aborting.\n");
1013 goto err_release_bar0;
1016 pci_set_master(pdev);
1018 err = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
1020 dev_warn(&pdev->dev, "Warning: couldn't set 64-bit PCI DMA mask.\n");
1021 err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
1023 dev_err(&pdev->dev, "Can't set PCI DMA mask, aborting.\n");
1024 goto err_release_bar2;
1027 err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK);
1029 dev_warn(&pdev->dev, "Warning: couldn't set 64-bit "
1030 "consistent PCI DMA mask.\n");
1031 err = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK);
1033 dev_err(&pdev->dev, "Can't set consistent PCI DMA mask, "
1035 goto err_release_bar2;
1039 priv = kzalloc(sizeof *priv, GFP_KERNEL);
1041 dev_err(&pdev->dev, "Device struct alloc failed, "
1044 goto err_release_bar2;
1049 INIT_LIST_HEAD(&priv->ctx_list);
1050 spin_lock_init(&priv->ctx_lock);
1052 mutex_init(&priv->port_mutex);
1054 INIT_LIST_HEAD(&priv->pgdir_list);
1055 mutex_init(&priv->pgdir_mutex);
1058 * Now reset the HCA before we touch the PCI capabilities or
1059 * attempt a firmware command, since a boot ROM may have left
1060 * the HCA in an undefined state.
1062 err = mlx4_reset(dev);
1064 mlx4_err(dev, "Failed to reset HCA, aborting.\n");
1068 if (mlx4_cmd_init(dev)) {
1069 mlx4_err(dev, "Failed to init command interface, aborting.\n");
1073 err = mlx4_init_hca(dev);
1077 mlx4_enable_msi_x(dev);
1079 err = mlx4_setup_hca(dev);
1080 if (err == -EBUSY && (dev->flags & MLX4_FLAG_MSI_X)) {
1081 dev->flags &= ~MLX4_FLAG_MSI_X;
1082 pci_disable_msix(pdev);
1083 err = mlx4_setup_hca(dev);
1089 for (port = 1; port <= dev->caps.num_ports; port++) {
1090 err = mlx4_init_port_info(dev, port);
1095 err = mlx4_register_device(dev);
1099 pci_set_drvdata(pdev, dev);
1104 for (port = 1; port <= dev->caps.num_ports; port++)
1105 mlx4_cleanup_port_info(&priv->port[port]);
1107 mlx4_cleanup_mcg_table(dev);
1108 mlx4_cleanup_qp_table(dev);
1109 mlx4_cleanup_srq_table(dev);
1110 mlx4_cleanup_cq_table(dev);
1111 mlx4_cmd_use_polling(dev);
1112 mlx4_cleanup_eq_table(dev);
1113 mlx4_cleanup_mr_table(dev);
1114 mlx4_cleanup_pd_table(dev);
1115 mlx4_cleanup_uar_table(dev);
1118 if (dev->flags & MLX4_FLAG_MSI_X)
1119 pci_disable_msix(pdev);
1121 mlx4_close_hca(dev);
1124 mlx4_cmd_cleanup(dev);
1130 pci_release_region(pdev, 2);
1133 pci_release_region(pdev, 0);
1136 pci_disable_device(pdev);
1137 pci_set_drvdata(pdev, NULL);
1141 static int __devinit mlx4_init_one(struct pci_dev *pdev,
1142 const struct pci_device_id *id)
1144 static int mlx4_version_printed;
1146 if (!mlx4_version_printed) {
1147 printk(KERN_INFO "%s", mlx4_version);
1148 ++mlx4_version_printed;
1151 return __mlx4_init_one(pdev, id);
1154 static void mlx4_remove_one(struct pci_dev *pdev)
1156 struct mlx4_dev *dev = pci_get_drvdata(pdev);
1157 struct mlx4_priv *priv = mlx4_priv(dev);
1161 mlx4_unregister_device(dev);
1163 for (p = 1; p <= dev->caps.num_ports; p++) {
1164 mlx4_cleanup_port_info(&priv->port[p]);
1165 mlx4_CLOSE_PORT(dev, p);
1168 mlx4_cleanup_mcg_table(dev);
1169 mlx4_cleanup_qp_table(dev);
1170 mlx4_cleanup_srq_table(dev);
1171 mlx4_cleanup_cq_table(dev);
1172 mlx4_cmd_use_polling(dev);
1173 mlx4_cleanup_eq_table(dev);
1174 mlx4_cleanup_mr_table(dev);
1175 mlx4_cleanup_pd_table(dev);
1178 mlx4_uar_free(dev, &priv->driver_uar);
1179 mlx4_cleanup_uar_table(dev);
1180 mlx4_close_hca(dev);
1181 mlx4_cmd_cleanup(dev);
1183 if (dev->flags & MLX4_FLAG_MSI_X)
1184 pci_disable_msix(pdev);
1187 pci_release_region(pdev, 2);
1188 pci_release_region(pdev, 0);
1189 pci_disable_device(pdev);
1190 pci_set_drvdata(pdev, NULL);
1194 int mlx4_restart_one(struct pci_dev *pdev)
1196 mlx4_remove_one(pdev);
1197 return __mlx4_init_one(pdev, NULL);
1200 static struct pci_device_id mlx4_pci_table[] = {
1201 { PCI_VDEVICE(MELLANOX, 0x6340) }, /* MT25408 "Hermon" SDR */
1202 { PCI_VDEVICE(MELLANOX, 0x634a) }, /* MT25408 "Hermon" DDR */
1203 { PCI_VDEVICE(MELLANOX, 0x6354) }, /* MT25408 "Hermon" QDR */
1204 { PCI_VDEVICE(MELLANOX, 0x6732) }, /* MT25408 "Hermon" DDR PCIe gen2 */
1205 { PCI_VDEVICE(MELLANOX, 0x673c) }, /* MT25408 "Hermon" QDR PCIe gen2 */
1206 { PCI_VDEVICE(MELLANOX, 0x6368) }, /* MT25408 "Hermon" EN 10GigE */
1207 { PCI_VDEVICE(MELLANOX, 0x6750) }, /* MT25408 "Hermon" EN 10GigE PCIe gen2 */
1211 MODULE_DEVICE_TABLE(pci, mlx4_pci_table);
1213 static struct pci_driver mlx4_driver = {
1215 .id_table = mlx4_pci_table,
1216 .probe = mlx4_init_one,
1217 .remove = __devexit_p(mlx4_remove_one)
1220 static int __init mlx4_verify_params(void)
1222 if ((log_num_mac < 0) || (log_num_mac > 7)) {
1223 printk(KERN_WARNING "mlx4_core: bad num_mac: %d\n", log_num_mac);
1227 if ((log_num_vlan < 0) || (log_num_vlan > 7)) {
1228 printk(KERN_WARNING "mlx4_core: bad num_vlan: %d\n", log_num_vlan);
1235 static int __init mlx4_init(void)
1239 if (mlx4_verify_params())
1242 ret = mlx4_catas_init();
1246 ret = pci_register_driver(&mlx4_driver);
1247 return ret < 0 ? ret : 0;
1250 static void __exit mlx4_cleanup(void)
1252 pci_unregister_driver(&mlx4_driver);
1253 mlx4_catas_cleanup();
1256 module_init(mlx4_init);
1257 module_exit(mlx4_cleanup);