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 struct workqueue_struct *mlx4_wq;
56 #ifdef CONFIG_MLX4_DEBUG
58 int mlx4_debug_level = 0;
59 module_param_named(debug_level, mlx4_debug_level, int, 0644);
60 MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0");
62 #endif /* CONFIG_MLX4_DEBUG */
67 module_param(msi_x, int, 0444);
68 MODULE_PARM_DESC(msi_x, "attempt to use MSI-X if nonzero");
70 #else /* CONFIG_PCI_MSI */
74 #endif /* CONFIG_PCI_MSI */
76 static char mlx4_version[] __devinitdata =
77 DRV_NAME ": Mellanox ConnectX core driver v"
78 DRV_VERSION " (" DRV_RELDATE ")\n";
80 static struct mlx4_profile default_profile = {
83 .rdmarc_per_qp = 1 << 4,
90 static int log_num_mac = 2;
91 module_param_named(log_num_mac, log_num_mac, int, 0444);
92 MODULE_PARM_DESC(log_num_mac, "Log2 max number of MACs per ETH port (1-7)");
94 static int log_num_vlan;
95 module_param_named(log_num_vlan, log_num_vlan, int, 0444);
96 MODULE_PARM_DESC(log_num_vlan, "Log2 max number of VLANs per ETH port (0-7)");
99 module_param_named(use_prio, use_prio, bool, 0444);
100 MODULE_PARM_DESC(use_prio, "Enable steering by VLAN priority on ETH ports "
103 int mlx4_check_port_params(struct mlx4_dev *dev,
104 enum mlx4_port_type *port_type)
108 for (i = 0; i < dev->caps.num_ports - 1; i++) {
109 if (port_type[i] != port_type[i + 1]) {
110 if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP)) {
111 mlx4_err(dev, "Only same port types supported "
112 "on this HCA, aborting.\n");
115 if (port_type[i] == MLX4_PORT_TYPE_ETH &&
116 port_type[i + 1] == MLX4_PORT_TYPE_IB)
121 for (i = 0; i < dev->caps.num_ports; i++) {
122 if (!(port_type[i] & dev->caps.supported_type[i+1])) {
123 mlx4_err(dev, "Requested port type for port %d is not "
124 "supported on this HCA\n", i + 1);
131 static void mlx4_set_port_mask(struct mlx4_dev *dev)
135 dev->caps.port_mask = 0;
136 for (i = 1; i <= dev->caps.num_ports; ++i)
137 if (dev->caps.port_type[i] == MLX4_PORT_TYPE_IB)
138 dev->caps.port_mask |= 1 << (i - 1);
140 static int mlx4_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
145 err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
147 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
151 if (dev_cap->min_page_sz > PAGE_SIZE) {
152 mlx4_err(dev, "HCA minimum page size of %d bigger than "
153 "kernel PAGE_SIZE of %ld, aborting.\n",
154 dev_cap->min_page_sz, PAGE_SIZE);
157 if (dev_cap->num_ports > MLX4_MAX_PORTS) {
158 mlx4_err(dev, "HCA has %d ports, but we only support %d, "
160 dev_cap->num_ports, MLX4_MAX_PORTS);
164 if (dev_cap->uar_size > pci_resource_len(dev->pdev, 2)) {
165 mlx4_err(dev, "HCA reported UAR size of 0x%x bigger than "
166 "PCI resource 2 size of 0x%llx, aborting.\n",
168 (unsigned long long) pci_resource_len(dev->pdev, 2));
172 dev->caps.num_ports = dev_cap->num_ports;
173 for (i = 1; i <= dev->caps.num_ports; ++i) {
174 dev->caps.vl_cap[i] = dev_cap->max_vl[i];
175 dev->caps.ib_mtu_cap[i] = dev_cap->ib_mtu[i];
176 dev->caps.gid_table_len[i] = dev_cap->max_gids[i];
177 dev->caps.pkey_table_len[i] = dev_cap->max_pkeys[i];
178 dev->caps.port_width_cap[i] = dev_cap->max_port_width[i];
179 dev->caps.eth_mtu_cap[i] = dev_cap->eth_mtu[i];
180 dev->caps.def_mac[i] = dev_cap->def_mac[i];
181 dev->caps.supported_type[i] = dev_cap->supported_port_types[i];
184 dev->caps.num_uars = dev_cap->uar_size / PAGE_SIZE;
185 dev->caps.local_ca_ack_delay = dev_cap->local_ca_ack_delay;
186 dev->caps.bf_reg_size = dev_cap->bf_reg_size;
187 dev->caps.bf_regs_per_page = dev_cap->bf_regs_per_page;
188 dev->caps.max_sq_sg = dev_cap->max_sq_sg;
189 dev->caps.max_rq_sg = dev_cap->max_rq_sg;
190 dev->caps.max_wqes = dev_cap->max_qp_sz;
191 dev->caps.max_qp_init_rdma = dev_cap->max_requester_per_qp;
192 dev->caps.max_srq_wqes = dev_cap->max_srq_sz;
193 dev->caps.max_srq_sge = dev_cap->max_rq_sg - 1;
194 dev->caps.reserved_srqs = dev_cap->reserved_srqs;
195 dev->caps.max_sq_desc_sz = dev_cap->max_sq_desc_sz;
196 dev->caps.max_rq_desc_sz = dev_cap->max_rq_desc_sz;
197 dev->caps.num_qp_per_mgm = MLX4_QP_PER_MGM;
199 * Subtract 1 from the limit because we need to allocate a
200 * spare CQE so the HCA HW can tell the difference between an
201 * empty CQ and a full CQ.
203 dev->caps.max_cqes = dev_cap->max_cq_sz - 1;
204 dev->caps.reserved_cqs = dev_cap->reserved_cqs;
205 dev->caps.reserved_eqs = dev_cap->reserved_eqs;
206 dev->caps.reserved_mtts = DIV_ROUND_UP(dev_cap->reserved_mtts,
207 MLX4_MTT_ENTRY_PER_SEG);
208 dev->caps.reserved_mrws = dev_cap->reserved_mrws;
209 dev->caps.reserved_uars = dev_cap->reserved_uars;
210 dev->caps.reserved_pds = dev_cap->reserved_pds;
211 dev->caps.mtt_entry_sz = MLX4_MTT_ENTRY_PER_SEG * dev_cap->mtt_entry_sz;
212 dev->caps.max_msg_sz = dev_cap->max_msg_sz;
213 dev->caps.page_size_cap = ~(u32) (dev_cap->min_page_sz - 1);
214 dev->caps.flags = dev_cap->flags;
215 dev->caps.bmme_flags = dev_cap->bmme_flags;
216 dev->caps.reserved_lkey = dev_cap->reserved_lkey;
217 dev->caps.stat_rate_support = dev_cap->stat_rate_support;
218 dev->caps.max_gso_sz = dev_cap->max_gso_sz;
220 dev->caps.log_num_macs = log_num_mac;
221 dev->caps.log_num_vlans = log_num_vlan;
222 dev->caps.log_num_prios = use_prio ? 3 : 0;
224 for (i = 1; i <= dev->caps.num_ports; ++i) {
225 if (dev->caps.supported_type[i] != MLX4_PORT_TYPE_ETH)
226 dev->caps.port_type[i] = MLX4_PORT_TYPE_IB;
228 dev->caps.port_type[i] = MLX4_PORT_TYPE_ETH;
229 dev->caps.possible_type[i] = dev->caps.port_type[i];
230 mlx4_priv(dev)->sense.sense_allowed[i] =
231 dev->caps.supported_type[i] == MLX4_PORT_TYPE_AUTO;
233 if (dev->caps.log_num_macs > dev_cap->log_max_macs[i]) {
234 dev->caps.log_num_macs = dev_cap->log_max_macs[i];
235 mlx4_warn(dev, "Requested number of MACs is too much "
236 "for port %d, reducing to %d.\n",
237 i, 1 << dev->caps.log_num_macs);
239 if (dev->caps.log_num_vlans > dev_cap->log_max_vlans[i]) {
240 dev->caps.log_num_vlans = dev_cap->log_max_vlans[i];
241 mlx4_warn(dev, "Requested number of VLANs is too much "
242 "for port %d, reducing to %d.\n",
243 i, 1 << dev->caps.log_num_vlans);
247 mlx4_set_port_mask(dev);
249 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] = dev_cap->reserved_qps;
250 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] =
251 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] =
252 (1 << dev->caps.log_num_macs) *
253 (1 << dev->caps.log_num_vlans) *
254 (1 << dev->caps.log_num_prios) *
256 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH] = MLX4_NUM_FEXCH;
258 dev->caps.reserved_qps = dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] +
259 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] +
260 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] +
261 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH];
267 * Change the port configuration of the device.
268 * Every user of this function must hold the port mutex.
270 int mlx4_change_port_types(struct mlx4_dev *dev,
271 enum mlx4_port_type *port_types)
277 for (port = 0; port < dev->caps.num_ports; port++) {
278 /* Change the port type only if the new type is different
279 * from the current, and not set to Auto */
280 if (port_types[port] != dev->caps.port_type[port + 1]) {
282 dev->caps.port_type[port + 1] = port_types[port];
286 mlx4_unregister_device(dev);
287 for (port = 1; port <= dev->caps.num_ports; port++) {
288 mlx4_CLOSE_PORT(dev, port);
289 err = mlx4_SET_PORT(dev, port);
291 mlx4_err(dev, "Failed to set port %d, "
296 mlx4_set_port_mask(dev);
297 err = mlx4_register_device(dev);
304 static ssize_t show_port_type(struct device *dev,
305 struct device_attribute *attr,
308 struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
310 struct mlx4_dev *mdev = info->dev;
314 (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_IB) ?
316 if (mdev->caps.possible_type[info->port] == MLX4_PORT_TYPE_AUTO)
317 sprintf(buf, "auto (%s)\n", type);
319 sprintf(buf, "%s\n", type);
324 static ssize_t set_port_type(struct device *dev,
325 struct device_attribute *attr,
326 const char *buf, size_t count)
328 struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
330 struct mlx4_dev *mdev = info->dev;
331 struct mlx4_priv *priv = mlx4_priv(mdev);
332 enum mlx4_port_type types[MLX4_MAX_PORTS];
333 enum mlx4_port_type new_types[MLX4_MAX_PORTS];
337 if (!strcmp(buf, "ib\n"))
338 info->tmp_type = MLX4_PORT_TYPE_IB;
339 else if (!strcmp(buf, "eth\n"))
340 info->tmp_type = MLX4_PORT_TYPE_ETH;
341 else if (!strcmp(buf, "auto\n"))
342 info->tmp_type = MLX4_PORT_TYPE_AUTO;
344 mlx4_err(mdev, "%s is not supported port type\n", buf);
348 mlx4_stop_sense(mdev);
349 mutex_lock(&priv->port_mutex);
350 /* Possible type is always the one that was delivered */
351 mdev->caps.possible_type[info->port] = info->tmp_type;
353 for (i = 0; i < mdev->caps.num_ports; i++) {
354 types[i] = priv->port[i+1].tmp_type ? priv->port[i+1].tmp_type :
355 mdev->caps.possible_type[i+1];
356 if (types[i] == MLX4_PORT_TYPE_AUTO)
357 types[i] = mdev->caps.port_type[i+1];
360 if (!(mdev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP)) {
361 for (i = 1; i <= mdev->caps.num_ports; i++) {
362 if (mdev->caps.possible_type[i] == MLX4_PORT_TYPE_AUTO) {
363 mdev->caps.possible_type[i] = mdev->caps.port_type[i];
369 mlx4_err(mdev, "Auto sensing is not supported on this HCA. "
370 "Set only 'eth' or 'ib' for both ports "
371 "(should be the same)\n");
375 mlx4_do_sense_ports(mdev, new_types, types);
377 err = mlx4_check_port_params(mdev, new_types);
381 /* We are about to apply the changes after the configuration
382 * was verified, no need to remember the temporary types
384 for (i = 0; i < mdev->caps.num_ports; i++)
385 priv->port[i + 1].tmp_type = 0;
387 err = mlx4_change_port_types(mdev, new_types);
390 mlx4_start_sense(mdev);
391 mutex_unlock(&priv->port_mutex);
392 return err ? err : count;
395 static int mlx4_load_fw(struct mlx4_dev *dev)
397 struct mlx4_priv *priv = mlx4_priv(dev);
400 priv->fw.fw_icm = mlx4_alloc_icm(dev, priv->fw.fw_pages,
401 GFP_HIGHUSER | __GFP_NOWARN, 0);
402 if (!priv->fw.fw_icm) {
403 mlx4_err(dev, "Couldn't allocate FW area, aborting.\n");
407 err = mlx4_MAP_FA(dev, priv->fw.fw_icm);
409 mlx4_err(dev, "MAP_FA command failed, aborting.\n");
413 err = mlx4_RUN_FW(dev);
415 mlx4_err(dev, "RUN_FW command failed, aborting.\n");
425 mlx4_free_icm(dev, priv->fw.fw_icm, 0);
429 static int mlx4_init_cmpt_table(struct mlx4_dev *dev, u64 cmpt_base,
432 struct mlx4_priv *priv = mlx4_priv(dev);
435 err = mlx4_init_icm_table(dev, &priv->qp_table.cmpt_table,
437 ((u64) (MLX4_CMPT_TYPE_QP *
438 cmpt_entry_sz) << MLX4_CMPT_SHIFT),
439 cmpt_entry_sz, dev->caps.num_qps,
440 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
445 err = mlx4_init_icm_table(dev, &priv->srq_table.cmpt_table,
447 ((u64) (MLX4_CMPT_TYPE_SRQ *
448 cmpt_entry_sz) << MLX4_CMPT_SHIFT),
449 cmpt_entry_sz, dev->caps.num_srqs,
450 dev->caps.reserved_srqs, 0, 0);
454 err = mlx4_init_icm_table(dev, &priv->cq_table.cmpt_table,
456 ((u64) (MLX4_CMPT_TYPE_CQ *
457 cmpt_entry_sz) << MLX4_CMPT_SHIFT),
458 cmpt_entry_sz, dev->caps.num_cqs,
459 dev->caps.reserved_cqs, 0, 0);
463 err = mlx4_init_icm_table(dev, &priv->eq_table.cmpt_table,
465 ((u64) (MLX4_CMPT_TYPE_EQ *
466 cmpt_entry_sz) << MLX4_CMPT_SHIFT),
468 dev->caps.num_eqs, dev->caps.num_eqs, 0, 0);
475 mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
478 mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
481 mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
487 static int mlx4_init_icm(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap,
488 struct mlx4_init_hca_param *init_hca, u64 icm_size)
490 struct mlx4_priv *priv = mlx4_priv(dev);
494 err = mlx4_SET_ICM_SIZE(dev, icm_size, &aux_pages);
496 mlx4_err(dev, "SET_ICM_SIZE command failed, aborting.\n");
500 mlx4_dbg(dev, "%lld KB of HCA context requires %lld KB aux memory.\n",
501 (unsigned long long) icm_size >> 10,
502 (unsigned long long) aux_pages << 2);
504 priv->fw.aux_icm = mlx4_alloc_icm(dev, aux_pages,
505 GFP_HIGHUSER | __GFP_NOWARN, 0);
506 if (!priv->fw.aux_icm) {
507 mlx4_err(dev, "Couldn't allocate aux memory, aborting.\n");
511 err = mlx4_MAP_ICM_AUX(dev, priv->fw.aux_icm);
513 mlx4_err(dev, "MAP_ICM_AUX command failed, aborting.\n");
517 err = mlx4_init_cmpt_table(dev, init_hca->cmpt_base, dev_cap->cmpt_entry_sz);
519 mlx4_err(dev, "Failed to map cMPT context memory, aborting.\n");
523 err = mlx4_map_eq_icm(dev, init_hca->eqc_base);
525 mlx4_err(dev, "Failed to map EQ context memory, aborting.\n");
530 * Reserved MTT entries must be aligned up to a cacheline
531 * boundary, since the FW will write to them, while the driver
532 * writes to all other MTT entries. (The variable
533 * dev->caps.mtt_entry_sz below is really the MTT segment
534 * size, not the raw entry size)
536 dev->caps.reserved_mtts =
537 ALIGN(dev->caps.reserved_mtts * dev->caps.mtt_entry_sz,
538 dma_get_cache_alignment()) / dev->caps.mtt_entry_sz;
540 err = mlx4_init_icm_table(dev, &priv->mr_table.mtt_table,
542 dev->caps.mtt_entry_sz,
543 dev->caps.num_mtt_segs,
544 dev->caps.reserved_mtts, 1, 0);
546 mlx4_err(dev, "Failed to map MTT context memory, aborting.\n");
550 err = mlx4_init_icm_table(dev, &priv->mr_table.dmpt_table,
552 dev_cap->dmpt_entry_sz,
554 dev->caps.reserved_mrws, 1, 1);
556 mlx4_err(dev, "Failed to map dMPT context memory, aborting.\n");
560 err = mlx4_init_icm_table(dev, &priv->qp_table.qp_table,
562 dev_cap->qpc_entry_sz,
564 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
567 mlx4_err(dev, "Failed to map QP context memory, aborting.\n");
571 err = mlx4_init_icm_table(dev, &priv->qp_table.auxc_table,
573 dev_cap->aux_entry_sz,
575 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
578 mlx4_err(dev, "Failed to map AUXC context memory, aborting.\n");
582 err = mlx4_init_icm_table(dev, &priv->qp_table.altc_table,
584 dev_cap->altc_entry_sz,
586 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
589 mlx4_err(dev, "Failed to map ALTC context memory, aborting.\n");
593 err = mlx4_init_icm_table(dev, &priv->qp_table.rdmarc_table,
594 init_hca->rdmarc_base,
595 dev_cap->rdmarc_entry_sz << priv->qp_table.rdmarc_shift,
597 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
600 mlx4_err(dev, "Failed to map RDMARC context memory, aborting\n");
604 err = mlx4_init_icm_table(dev, &priv->cq_table.table,
606 dev_cap->cqc_entry_sz,
608 dev->caps.reserved_cqs, 0, 0);
610 mlx4_err(dev, "Failed to map CQ context memory, aborting.\n");
611 goto err_unmap_rdmarc;
614 err = mlx4_init_icm_table(dev, &priv->srq_table.table,
616 dev_cap->srq_entry_sz,
618 dev->caps.reserved_srqs, 0, 0);
620 mlx4_err(dev, "Failed to map SRQ context memory, aborting.\n");
625 * It's not strictly required, but for simplicity just map the
626 * whole multicast group table now. The table isn't very big
627 * and it's a lot easier than trying to track ref counts.
629 err = mlx4_init_icm_table(dev, &priv->mcg_table.table,
630 init_hca->mc_base, MLX4_MGM_ENTRY_SIZE,
631 dev->caps.num_mgms + dev->caps.num_amgms,
632 dev->caps.num_mgms + dev->caps.num_amgms,
635 mlx4_err(dev, "Failed to map MCG context memory, aborting.\n");
642 mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
645 mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
648 mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
651 mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
654 mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
657 mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
660 mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
663 mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
666 mlx4_unmap_eq_icm(dev);
669 mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
670 mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
671 mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
672 mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
675 mlx4_UNMAP_ICM_AUX(dev);
678 mlx4_free_icm(dev, priv->fw.aux_icm, 0);
683 static void mlx4_free_icms(struct mlx4_dev *dev)
685 struct mlx4_priv *priv = mlx4_priv(dev);
687 mlx4_cleanup_icm_table(dev, &priv->mcg_table.table);
688 mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
689 mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
690 mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
691 mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
692 mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
693 mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
694 mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
695 mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
696 mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
697 mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
698 mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
699 mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
700 mlx4_unmap_eq_icm(dev);
702 mlx4_UNMAP_ICM_AUX(dev);
703 mlx4_free_icm(dev, priv->fw.aux_icm, 0);
706 static void mlx4_close_hca(struct mlx4_dev *dev)
708 mlx4_CLOSE_HCA(dev, 0);
711 mlx4_free_icm(dev, mlx4_priv(dev)->fw.fw_icm, 0);
714 static int mlx4_init_hca(struct mlx4_dev *dev)
716 struct mlx4_priv *priv = mlx4_priv(dev);
717 struct mlx4_adapter adapter;
718 struct mlx4_dev_cap dev_cap;
719 struct mlx4_mod_stat_cfg mlx4_cfg;
720 struct mlx4_profile profile;
721 struct mlx4_init_hca_param init_hca;
725 err = mlx4_QUERY_FW(dev);
727 mlx4_err(dev, "QUERY_FW command failed, aborting.\n");
731 err = mlx4_load_fw(dev);
733 mlx4_err(dev, "Failed to start FW, aborting.\n");
737 mlx4_cfg.log_pg_sz_m = 1;
738 mlx4_cfg.log_pg_sz = 0;
739 err = mlx4_MOD_STAT_CFG(dev, &mlx4_cfg);
741 mlx4_warn(dev, "Failed to override log_pg_sz parameter\n");
743 err = mlx4_dev_cap(dev, &dev_cap);
745 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
749 profile = default_profile;
751 icm_size = mlx4_make_profile(dev, &profile, &dev_cap, &init_hca);
752 if ((long long) icm_size < 0) {
757 init_hca.log_uar_sz = ilog2(dev->caps.num_uars);
759 err = mlx4_init_icm(dev, &dev_cap, &init_hca, icm_size);
763 err = mlx4_INIT_HCA(dev, &init_hca);
765 mlx4_err(dev, "INIT_HCA command failed, aborting.\n");
769 err = mlx4_QUERY_ADAPTER(dev, &adapter);
771 mlx4_err(dev, "QUERY_ADAPTER command failed, aborting.\n");
775 priv->eq_table.inta_pin = adapter.inta_pin;
776 memcpy(dev->board_id, adapter.board_id, sizeof dev->board_id);
788 mlx4_free_icm(dev, priv->fw.fw_icm, 0);
793 static int mlx4_setup_hca(struct mlx4_dev *dev)
795 struct mlx4_priv *priv = mlx4_priv(dev);
798 __be32 ib_port_default_caps;
800 err = mlx4_init_uar_table(dev);
802 mlx4_err(dev, "Failed to initialize "
803 "user access region table, aborting.\n");
807 err = mlx4_uar_alloc(dev, &priv->driver_uar);
809 mlx4_err(dev, "Failed to allocate driver access region, "
811 goto err_uar_table_free;
814 priv->kar = ioremap(priv->driver_uar.pfn << PAGE_SHIFT, PAGE_SIZE);
816 mlx4_err(dev, "Couldn't map kernel access region, "
822 err = mlx4_init_pd_table(dev);
824 mlx4_err(dev, "Failed to initialize "
825 "protection domain table, aborting.\n");
829 err = mlx4_init_mr_table(dev);
831 mlx4_err(dev, "Failed to initialize "
832 "memory region table, aborting.\n");
833 goto err_pd_table_free;
836 err = mlx4_init_eq_table(dev);
838 mlx4_err(dev, "Failed to initialize "
839 "event queue table, aborting.\n");
840 goto err_mr_table_free;
843 err = mlx4_cmd_use_events(dev);
845 mlx4_err(dev, "Failed to switch to event-driven "
846 "firmware commands, aborting.\n");
847 goto err_eq_table_free;
852 if (dev->flags & MLX4_FLAG_MSI_X) {
853 mlx4_warn(dev, "NOP command failed to generate MSI-X "
854 "interrupt IRQ %d).\n",
855 priv->eq_table.eq[dev->caps.num_comp_vectors].irq);
856 mlx4_warn(dev, "Trying again without MSI-X.\n");
858 mlx4_err(dev, "NOP command failed to generate interrupt "
859 "(IRQ %d), aborting.\n",
860 priv->eq_table.eq[dev->caps.num_comp_vectors].irq);
861 mlx4_err(dev, "BIOS or ACPI interrupt routing problem?\n");
867 mlx4_dbg(dev, "NOP command IRQ test passed\n");
869 err = mlx4_init_cq_table(dev);
871 mlx4_err(dev, "Failed to initialize "
872 "completion queue table, aborting.\n");
876 err = mlx4_init_srq_table(dev);
878 mlx4_err(dev, "Failed to initialize "
879 "shared receive queue table, aborting.\n");
880 goto err_cq_table_free;
883 err = mlx4_init_qp_table(dev);
885 mlx4_err(dev, "Failed to initialize "
886 "queue pair table, aborting.\n");
887 goto err_srq_table_free;
890 err = mlx4_init_mcg_table(dev);
892 mlx4_err(dev, "Failed to initialize "
893 "multicast group table, aborting.\n");
894 goto err_qp_table_free;
897 for (port = 1; port <= dev->caps.num_ports; port++) {
898 ib_port_default_caps = 0;
899 err = mlx4_get_port_ib_caps(dev, port, &ib_port_default_caps);
901 mlx4_warn(dev, "failed to get port %d default "
902 "ib capabilities (%d). Continuing with "
903 "caps = 0\n", port, err);
904 dev->caps.ib_port_def_cap[port] = ib_port_default_caps;
905 err = mlx4_SET_PORT(dev, port);
907 mlx4_err(dev, "Failed to set port %d, aborting\n",
909 goto err_mcg_table_free;
916 mlx4_cleanup_mcg_table(dev);
919 mlx4_cleanup_qp_table(dev);
922 mlx4_cleanup_srq_table(dev);
925 mlx4_cleanup_cq_table(dev);
928 mlx4_cmd_use_polling(dev);
931 mlx4_cleanup_eq_table(dev);
934 mlx4_cleanup_mr_table(dev);
937 mlx4_cleanup_pd_table(dev);
943 mlx4_uar_free(dev, &priv->driver_uar);
946 mlx4_cleanup_uar_table(dev);
950 static void mlx4_enable_msi_x(struct mlx4_dev *dev)
952 struct mlx4_priv *priv = mlx4_priv(dev);
953 struct msix_entry *entries;
959 nreq = min_t(int, dev->caps.num_eqs - dev->caps.reserved_eqs,
960 num_possible_cpus() + 1);
961 entries = kcalloc(nreq, sizeof *entries, GFP_KERNEL);
965 for (i = 0; i < nreq; ++i)
966 entries[i].entry = i;
969 err = pci_enable_msix(dev->pdev, entries, nreq);
971 /* Try again if at least 2 vectors are available */
973 mlx4_info(dev, "Requested %d vectors, "
974 "but only %d MSI-X vectors available, "
975 "trying again\n", nreq, err);
983 dev->caps.num_comp_vectors = nreq - 1;
984 for (i = 0; i < nreq; ++i)
985 priv->eq_table.eq[i].irq = entries[i].vector;
987 dev->flags |= MLX4_FLAG_MSI_X;
994 dev->caps.num_comp_vectors = 1;
996 for (i = 0; i < 2; ++i)
997 priv->eq_table.eq[i].irq = dev->pdev->irq;
1000 static int mlx4_init_port_info(struct mlx4_dev *dev, int port)
1002 struct mlx4_port_info *info = &mlx4_priv(dev)->port[port];
1007 mlx4_init_mac_table(dev, &info->mac_table);
1008 mlx4_init_vlan_table(dev, &info->vlan_table);
1010 sprintf(info->dev_name, "mlx4_port%d", port);
1011 info->port_attr.attr.name = info->dev_name;
1012 info->port_attr.attr.mode = S_IRUGO | S_IWUSR;
1013 info->port_attr.show = show_port_type;
1014 info->port_attr.store = set_port_type;
1016 err = device_create_file(&dev->pdev->dev, &info->port_attr);
1018 mlx4_err(dev, "Failed to create file for port %d\n", port);
1025 static void mlx4_cleanup_port_info(struct mlx4_port_info *info)
1030 device_remove_file(&info->dev->pdev->dev, &info->port_attr);
1033 static int __mlx4_init_one(struct pci_dev *pdev, const struct pci_device_id *id)
1035 struct mlx4_priv *priv;
1036 struct mlx4_dev *dev;
1040 printk(KERN_INFO PFX "Initializing %s\n",
1043 err = pci_enable_device(pdev);
1045 dev_err(&pdev->dev, "Cannot enable PCI device, "
1051 * Check for BARs. We expect 0: 1MB
1053 if (!(pci_resource_flags(pdev, 0) & IORESOURCE_MEM) ||
1054 pci_resource_len(pdev, 0) != 1 << 20) {
1055 dev_err(&pdev->dev, "Missing DCS, aborting.\n");
1057 goto err_disable_pdev;
1059 if (!(pci_resource_flags(pdev, 2) & IORESOURCE_MEM)) {
1060 dev_err(&pdev->dev, "Missing UAR, aborting.\n");
1062 goto err_disable_pdev;
1065 err = pci_request_region(pdev, 0, DRV_NAME);
1067 dev_err(&pdev->dev, "Cannot request control region, aborting.\n");
1068 goto err_disable_pdev;
1071 err = pci_request_region(pdev, 2, DRV_NAME);
1073 dev_err(&pdev->dev, "Cannot request UAR region, aborting.\n");
1074 goto err_release_bar0;
1077 pci_set_master(pdev);
1079 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
1081 dev_warn(&pdev->dev, "Warning: couldn't set 64-bit PCI DMA mask.\n");
1082 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
1084 dev_err(&pdev->dev, "Can't set PCI DMA mask, aborting.\n");
1085 goto err_release_bar2;
1088 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
1090 dev_warn(&pdev->dev, "Warning: couldn't set 64-bit "
1091 "consistent PCI DMA mask.\n");
1092 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
1094 dev_err(&pdev->dev, "Can't set consistent PCI DMA mask, "
1096 goto err_release_bar2;
1100 priv = kzalloc(sizeof *priv, GFP_KERNEL);
1102 dev_err(&pdev->dev, "Device struct alloc failed, "
1105 goto err_release_bar2;
1110 INIT_LIST_HEAD(&priv->ctx_list);
1111 spin_lock_init(&priv->ctx_lock);
1113 mutex_init(&priv->port_mutex);
1115 INIT_LIST_HEAD(&priv->pgdir_list);
1116 mutex_init(&priv->pgdir_mutex);
1119 * Now reset the HCA before we touch the PCI capabilities or
1120 * attempt a firmware command, since a boot ROM may have left
1121 * the HCA in an undefined state.
1123 err = mlx4_reset(dev);
1125 mlx4_err(dev, "Failed to reset HCA, aborting.\n");
1129 if (mlx4_cmd_init(dev)) {
1130 mlx4_err(dev, "Failed to init command interface, aborting.\n");
1134 err = mlx4_init_hca(dev);
1138 err = mlx4_alloc_eq_table(dev);
1142 mlx4_enable_msi_x(dev);
1144 err = mlx4_setup_hca(dev);
1145 if (err == -EBUSY && (dev->flags & MLX4_FLAG_MSI_X)) {
1146 dev->flags &= ~MLX4_FLAG_MSI_X;
1147 pci_disable_msix(pdev);
1148 err = mlx4_setup_hca(dev);
1154 for (port = 1; port <= dev->caps.num_ports; port++) {
1155 err = mlx4_init_port_info(dev, port);
1160 err = mlx4_register_device(dev);
1164 mlx4_sense_init(dev);
1165 mlx4_start_sense(dev);
1167 pci_set_drvdata(pdev, dev);
1172 for (port = 1; port <= dev->caps.num_ports; port++)
1173 mlx4_cleanup_port_info(&priv->port[port]);
1175 mlx4_cleanup_mcg_table(dev);
1176 mlx4_cleanup_qp_table(dev);
1177 mlx4_cleanup_srq_table(dev);
1178 mlx4_cleanup_cq_table(dev);
1179 mlx4_cmd_use_polling(dev);
1180 mlx4_cleanup_eq_table(dev);
1181 mlx4_cleanup_mr_table(dev);
1182 mlx4_cleanup_pd_table(dev);
1183 mlx4_cleanup_uar_table(dev);
1186 mlx4_free_eq_table(dev);
1189 if (dev->flags & MLX4_FLAG_MSI_X)
1190 pci_disable_msix(pdev);
1192 mlx4_close_hca(dev);
1195 mlx4_cmd_cleanup(dev);
1201 pci_release_region(pdev, 2);
1204 pci_release_region(pdev, 0);
1207 pci_disable_device(pdev);
1208 pci_set_drvdata(pdev, NULL);
1212 static int __devinit mlx4_init_one(struct pci_dev *pdev,
1213 const struct pci_device_id *id)
1215 static int mlx4_version_printed;
1217 if (!mlx4_version_printed) {
1218 printk(KERN_INFO "%s", mlx4_version);
1219 ++mlx4_version_printed;
1222 return __mlx4_init_one(pdev, id);
1225 static void mlx4_remove_one(struct pci_dev *pdev)
1227 struct mlx4_dev *dev = pci_get_drvdata(pdev);
1228 struct mlx4_priv *priv = mlx4_priv(dev);
1232 mlx4_stop_sense(dev);
1233 mlx4_unregister_device(dev);
1235 for (p = 1; p <= dev->caps.num_ports; p++) {
1236 mlx4_cleanup_port_info(&priv->port[p]);
1237 mlx4_CLOSE_PORT(dev, p);
1240 mlx4_cleanup_mcg_table(dev);
1241 mlx4_cleanup_qp_table(dev);
1242 mlx4_cleanup_srq_table(dev);
1243 mlx4_cleanup_cq_table(dev);
1244 mlx4_cmd_use_polling(dev);
1245 mlx4_cleanup_eq_table(dev);
1246 mlx4_cleanup_mr_table(dev);
1247 mlx4_cleanup_pd_table(dev);
1250 mlx4_uar_free(dev, &priv->driver_uar);
1251 mlx4_cleanup_uar_table(dev);
1252 mlx4_free_eq_table(dev);
1253 mlx4_close_hca(dev);
1254 mlx4_cmd_cleanup(dev);
1256 if (dev->flags & MLX4_FLAG_MSI_X)
1257 pci_disable_msix(pdev);
1260 pci_release_region(pdev, 2);
1261 pci_release_region(pdev, 0);
1262 pci_disable_device(pdev);
1263 pci_set_drvdata(pdev, NULL);
1267 int mlx4_restart_one(struct pci_dev *pdev)
1269 mlx4_remove_one(pdev);
1270 return __mlx4_init_one(pdev, NULL);
1273 static struct pci_device_id mlx4_pci_table[] = {
1274 { PCI_VDEVICE(MELLANOX, 0x6340) }, /* MT25408 "Hermon" SDR */
1275 { PCI_VDEVICE(MELLANOX, 0x634a) }, /* MT25408 "Hermon" DDR */
1276 { PCI_VDEVICE(MELLANOX, 0x6354) }, /* MT25408 "Hermon" QDR */
1277 { PCI_VDEVICE(MELLANOX, 0x6732) }, /* MT25408 "Hermon" DDR PCIe gen2 */
1278 { PCI_VDEVICE(MELLANOX, 0x673c) }, /* MT25408 "Hermon" QDR PCIe gen2 */
1279 { PCI_VDEVICE(MELLANOX, 0x6368) }, /* MT25408 "Hermon" EN 10GigE */
1280 { PCI_VDEVICE(MELLANOX, 0x6750) }, /* MT25408 "Hermon" EN 10GigE PCIe gen2 */
1281 { PCI_VDEVICE(MELLANOX, 0x6372) }, /* MT25458 ConnectX EN 10GBASE-T 10GigE */
1282 { PCI_VDEVICE(MELLANOX, 0x675a) }, /* MT25458 ConnectX EN 10GBASE-T+Gen2 10GigE */
1286 MODULE_DEVICE_TABLE(pci, mlx4_pci_table);
1288 static struct pci_driver mlx4_driver = {
1290 .id_table = mlx4_pci_table,
1291 .probe = mlx4_init_one,
1292 .remove = __devexit_p(mlx4_remove_one)
1295 static int __init mlx4_verify_params(void)
1297 if ((log_num_mac < 0) || (log_num_mac > 7)) {
1298 printk(KERN_WARNING "mlx4_core: bad num_mac: %d\n", log_num_mac);
1302 if ((log_num_vlan < 0) || (log_num_vlan > 7)) {
1303 printk(KERN_WARNING "mlx4_core: bad num_vlan: %d\n", log_num_vlan);
1310 static int __init mlx4_init(void)
1314 if (mlx4_verify_params())
1319 mlx4_wq = create_singlethread_workqueue("mlx4");
1323 ret = pci_register_driver(&mlx4_driver);
1324 return ret < 0 ? ret : 0;
1327 static void __exit mlx4_cleanup(void)
1329 pci_unregister_driver(&mlx4_driver);
1330 destroy_workqueue(mlx4_wq);
1333 module_init(mlx4_init);
1334 module_exit(mlx4_cleanup);