projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
mlx4_core: Adjustments to Flow Steering activation logic for SR-IOV
[linux-2.6-block.git] / drivers / net / ethernet / mellanox / mlx4 / main.c
CommitLineData
225c7b1f
RD		 1/*
2 * Copyright (c) 2004, 2005 Topspin Communications. All rights reserved.
3 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
51a379d0 4 * Copyright (c) 2005, 2006, 2007, 2008 Mellanox Technologies. All rights reserved.
225c7b1f
RD		 5 * Copyright (c) 2006, 2007 Cisco Systems, Inc. All rights reserved.
6 *
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:
12 *
13 * Redistribution and use in source and binary forms, with or
14 * without modification, are permitted provided that the following
15 * conditions are met:
16 *
17 * - Redistributions of source code must retain the above
18 * copyright notice, this list of conditions and the following
19 * disclaimer.
20 *
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.
25 *
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
33 * SOFTWARE.
34 */
35
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>
5a0e3ad6 41#include <linux/slab.h>
c1b43dca 42#include <linux/io-mapping.h>
ab9c17a0 43#include <linux/delay.h>
90b1ebe7 44#include <linux/netdevice.h>
225c7b1f
RD		 45
46#include <linux/mlx4/device.h>
47#include <linux/mlx4/doorbell.h>
48
49#include "mlx4.h"
50#include "fw.h"
51#include "icm.h"
52
53MODULE_AUTHOR("Roland Dreier");
54MODULE_DESCRIPTION("Mellanox ConnectX HCA low-level driver");
55MODULE_LICENSE("Dual BSD/GPL");
56MODULE_VERSION(DRV_VERSION);
57
27bf91d6
YP		 58struct workqueue_struct *mlx4_wq;
59
225c7b1f
RD		 60#ifdef CONFIG_MLX4_DEBUG
61
62int mlx4_debug_level = 0;
63module_param_named(debug_level, mlx4_debug_level, int, 0644);
64MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0");
65
66#endif /* CONFIG_MLX4_DEBUG */
67
68#ifdef CONFIG_PCI_MSI
69
08fb1055 70static int msi_x = 1;
225c7b1f
RD		 71module_param(msi_x, int, 0444);
72MODULE_PARM_DESC(msi_x, "attempt to use MSI-X if nonzero");
73
74#else /* CONFIG_PCI_MSI */
75
76#define msi_x (0)
77
78#endif /* CONFIG_PCI_MSI */
79
ab9c17a0
JM		 80static int num_vfs;
81module_param(num_vfs, int, 0444);
82MODULE_PARM_DESC(num_vfs, "enable #num_vfs functions if num_vfs > 0");
83
84static int probe_vf;
85module_param(probe_vf, int, 0644);
86MODULE_PARM_DESC(probe_vf, "number of vfs to probe by pf driver (num_vfs > 0)");
87
0ec2c0f8
EE		 88int mlx4_log_num_mgm_entry_size = 10;
89module_param_named(log_num_mgm_entry_size,
90 mlx4_log_num_mgm_entry_size, int, 0444);
91MODULE_PARM_DESC(log_num_mgm_entry_size, "log mgm size, that defines the num"
92 " of qp per mcg, for example:"
93 " 10 gives 248.range: 9<="
0ff1fb65
HHZ		 94 " log_num_mgm_entry_size <= 12."
95 " Not in use with device managed"
96 " flow steering");
0ec2c0f8 97
08ff3235
OG		 98static bool enable_64b_cqe_eqe;
99module_param(enable_64b_cqe_eqe, bool, 0444);
100MODULE_PARM_DESC(enable_64b_cqe_eqe,
101 "Enable 64 byte CQEs/EQEs when the the FW supports this");
102
ab9c17a0 103#define HCA_GLOBAL_CAP_MASK 0
08ff3235
OG		 104
105#define PF_CONTEXT_BEHAVIOUR_MASK MLX4_FUNC_CAP_64B_EQE_CQE
ab9c17a0 106
f57e6848 107static char mlx4_version[] =
225c7b1f
RD		 108 DRV_NAME ": Mellanox ConnectX core driver v"
109 DRV_VERSION " (" DRV_RELDATE ")\n";
110
111static struct mlx4_profile default_profile = {
ab9c17a0 112 .num_qp = 1 << 18,
225c7b1f 113 .num_srq = 1 << 16,
c9f2ba5e 114 .rdmarc_per_qp = 1 << 4,
225c7b1f
RD		 115 .num_cq = 1 << 16,
116 .num_mcg = 1 << 13,
ab9c17a0 117 .num_mpt = 1 << 19,
9fd7a1e1 118 .num_mtt = 1 << 20, /* It is really num mtt segements */
225c7b1f
RD		 119};
120
ab9c17a0 121static int log_num_mac = 7;
93fc9e1b
YP		 122module_param_named(log_num_mac, log_num_mac, int, 0444);
123MODULE_PARM_DESC(log_num_mac, "Log2 max number of MACs per ETH port (1-7)");
124
125static int log_num_vlan;
126module_param_named(log_num_vlan, log_num_vlan, int, 0444);
127MODULE_PARM_DESC(log_num_vlan, "Log2 max number of VLANs per ETH port (0-7)");
cb29688a
OG		 128/* Log2 max number of VLANs per ETH port (0-7) */
129#define MLX4_LOG_NUM_VLANS 7
93fc9e1b 130
eb939922 131static bool use_prio;
93fc9e1b
YP		 132module_param_named(use_prio, use_prio, bool, 0444);
133MODULE_PARM_DESC(use_prio, "Enable steering by VLAN priority on ETH ports "
134 "(0/1, default 0)");
135
2b8fb286 136int log_mtts_per_seg = ilog2(MLX4_MTT_ENTRY_PER_SEG);
ab6bf42e 137module_param_named(log_mtts_per_seg, log_mtts_per_seg, int, 0444);
0498628f 138MODULE_PARM_DESC(log_mtts_per_seg, "Log2 number of MTT entries per segment (1-7)");
ab6bf42e 139
8d0fc7b6 140static int port_type_array[2] = {MLX4_PORT_TYPE_NONE, MLX4_PORT_TYPE_NONE};
ab9c17a0
JM		 141static int arr_argc = 2;
142module_param_array(port_type_array, int, &arr_argc, 0444);
8d0fc7b6
YP		 143MODULE_PARM_DESC(port_type_array, "Array of port types: HW_DEFAULT (0) is default "
144 "1 for IB, 2 for Ethernet");
ab9c17a0
JM		 145
146struct mlx4_port_config {
147 struct list_head list;
148 enum mlx4_port_type port_type[MLX4_MAX_PORTS + 1];
149 struct pci_dev *pdev;
150};
151
27bf91d6
YP		 152int mlx4_check_port_params(struct mlx4_dev *dev,
153 enum mlx4_port_type *port_type)
7ff93f8b
YP		 154{
155 int i;
156
157 for (i = 0; i < dev->caps.num_ports - 1; i++) {
27bf91d6
YP		 158 if (port_type[i] != port_type[i + 1]) {
159 if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP)) {
160 mlx4_err(dev, "Only same port types supported "
161 "on this HCA, aborting.\n");
162 return -EINVAL;
163 }
7ff93f8b
YP		 164 }
165 }
7ff93f8b
YP		 166
167 for (i = 0; i < dev->caps.num_ports; i++) {
168 if (!(port_type[i] & dev->caps.supported_type[i+1])) {
169 mlx4_err(dev, "Requested port type for port %d is not "
170 "supported on this HCA\n", i + 1);
171 return -EINVAL;
172 }
173 }
174 return 0;
175}
176
177static void mlx4_set_port_mask(struct mlx4_dev *dev)
178{
179 int i;
180
7ff93f8b 181 for (i = 1; i <= dev->caps.num_ports; ++i)
65dab25d 182 dev->caps.port_mask[i] = dev->caps.port_type[i];
7ff93f8b 183}
f2a3f6a3 184
3d73c288 185static int mlx4_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
225c7b1f
RD		 186{
187 int err;
5ae2a7a8 188 int i;
225c7b1f
RD		 189
190 err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
191 if (err) {
192 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
193 return err;
194 }
195
196 if (dev_cap->min_page_sz > PAGE_SIZE) {
197 mlx4_err(dev, "HCA minimum page size of %d bigger than "
198 "kernel PAGE_SIZE of %ld, aborting.\n",
199 dev_cap->min_page_sz, PAGE_SIZE);
200 return -ENODEV;
201 }
202 if (dev_cap->num_ports > MLX4_MAX_PORTS) {
203 mlx4_err(dev, "HCA has %d ports, but we only support %d, "
204 "aborting.\n",
205 dev_cap->num_ports, MLX4_MAX_PORTS);
206 return -ENODEV;
207 }
208
209 if (dev_cap->uar_size > pci_resource_len(dev->pdev, 2)) {
210 mlx4_err(dev, "HCA reported UAR size of 0x%x bigger than "
211 "PCI resource 2 size of 0x%llx, aborting.\n",
212 dev_cap->uar_size,
213 (unsigned long long) pci_resource_len(dev->pdev, 2));
214 return -ENODEV;
215 }
216
217 dev->caps.num_ports = dev_cap->num_ports;
3fc929e2 218 dev->phys_caps.num_phys_eqs = MLX4_MAX_EQ_NUM;
5ae2a7a8
RD		 219 for (i = 1; i <= dev->caps.num_ports; ++i) {
220 dev->caps.vl_cap[i] = dev_cap->max_vl[i];
b79acb49 221 dev->caps.ib_mtu_cap[i] = dev_cap->ib_mtu[i];
6634961c
JM		 222 dev->phys_caps.gid_phys_table_len[i] = dev_cap->max_gids[i];
223 dev->phys_caps.pkey_phys_table_len[i] = dev_cap->max_pkeys[i];
224 /* set gid and pkey table operating lengths by default
225 * to non-sriov values */
5ae2a7a8
RD		 226 dev->caps.gid_table_len[i] = dev_cap->max_gids[i];
227 dev->caps.pkey_table_len[i] = dev_cap->max_pkeys[i];
228 dev->caps.port_width_cap[i] = dev_cap->max_port_width[i];
b79acb49
YP		 229 dev->caps.eth_mtu_cap[i] = dev_cap->eth_mtu[i];
230 dev->caps.def_mac[i] = dev_cap->def_mac[i];
7ff93f8b 231 dev->caps.supported_type[i] = dev_cap->supported_port_types[i];
8d0fc7b6
YP		 232 dev->caps.suggested_type[i] = dev_cap->suggested_type[i];
233 dev->caps.default_sense[i] = dev_cap->default_sense[i];
7699517d
YP		 234 dev->caps.trans_type[i] = dev_cap->trans_type[i];
235 dev->caps.vendor_oui[i] = dev_cap->vendor_oui[i];
236 dev->caps.wavelength[i] = dev_cap->wavelength[i];
237 dev->caps.trans_code[i] = dev_cap->trans_code[i];
5ae2a7a8
RD		 238 }
239
ab9c17a0 240 dev->caps.uar_page_size = PAGE_SIZE;
225c7b1f 241 dev->caps.num_uars = dev_cap->uar_size / PAGE_SIZE;
225c7b1f
RD		 242 dev->caps.local_ca_ack_delay = dev_cap->local_ca_ack_delay;
243 dev->caps.bf_reg_size = dev_cap->bf_reg_size;
244 dev->caps.bf_regs_per_page = dev_cap->bf_regs_per_page;
245 dev->caps.max_sq_sg = dev_cap->max_sq_sg;
246 dev->caps.max_rq_sg = dev_cap->max_rq_sg;
247 dev->caps.max_wqes = dev_cap->max_qp_sz;
248 dev->caps.max_qp_init_rdma = dev_cap->max_requester_per_qp;
225c7b1f
RD		 249 dev->caps.max_srq_wqes = dev_cap->max_srq_sz;
250 dev->caps.max_srq_sge = dev_cap->max_rq_sg - 1;
251 dev->caps.reserved_srqs = dev_cap->reserved_srqs;
252 dev->caps.max_sq_desc_sz = dev_cap->max_sq_desc_sz;
253 dev->caps.max_rq_desc_sz = dev_cap->max_rq_desc_sz;
225c7b1f
RD		 254 /*
255 * Subtract 1 from the limit because we need to allocate a
256 * spare CQE so the HCA HW can tell the difference between an
257 * empty CQ and a full CQ.
258 */
259 dev->caps.max_cqes = dev_cap->max_cq_sz - 1;
260 dev->caps.reserved_cqs = dev_cap->reserved_cqs;
261 dev->caps.reserved_eqs = dev_cap->reserved_eqs;
2b8fb286 262 dev->caps.reserved_mtts = dev_cap->reserved_mtts;
225c7b1f 263 dev->caps.reserved_mrws = dev_cap->reserved_mrws;
ab9c17a0
JM		 264
265 /* The first 128 UARs are used for EQ doorbells */
266 dev->caps.reserved_uars = max_t(int, 128, dev_cap->reserved_uars);
225c7b1f 267 dev->caps.reserved_pds = dev_cap->reserved_pds;
012a8ff5
SH		 268 dev->caps.reserved_xrcds = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
269 dev_cap->reserved_xrcds : 0;
270 dev->caps.max_xrcds = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
271 dev_cap->max_xrcds : 0;
2b8fb286
MA		 272 dev->caps.mtt_entry_sz = dev_cap->mtt_entry_sz;
273
149983af 274 dev->caps.max_msg_sz = dev_cap->max_msg_sz;
225c7b1f
RD		 275 dev->caps.page_size_cap = ~(u32) (dev_cap->min_page_sz - 1);
276 dev->caps.flags = dev_cap->flags;
b3416f44 277 dev->caps.flags2 = dev_cap->flags2;
95d04f07
RD		 278 dev->caps.bmme_flags = dev_cap->bmme_flags;
279 dev->caps.reserved_lkey = dev_cap->reserved_lkey;
225c7b1f 280 dev->caps.stat_rate_support = dev_cap->stat_rate_support;
b832be1e 281 dev->caps.max_gso_sz = dev_cap->max_gso_sz;
b3416f44 282 dev->caps.max_rss_tbl_sz = dev_cap->max_rss_tbl_sz;
225c7b1f 283
ca3e57a5
RD		 284 /* Sense port always allowed on supported devices for ConnectX-1 and -2 */
285 if (mlx4_priv(dev)->pci_dev_data & MLX4_PCI_DEV_FORCE_SENSE_PORT)
58a60168 286 dev->caps.flags |= MLX4_DEV_CAP_FLAG_SENSE_SUPPORT;
aadf4f3f
RD		 287 /* Don't do sense port on multifunction devices (for now at least) */
288 if (mlx4_is_mfunc(dev))
289 dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_SENSE_SUPPORT;
58a60168 290
93fc9e1b 291 dev->caps.log_num_macs = log_num_mac;
cb29688a 292 dev->caps.log_num_vlans = MLX4_LOG_NUM_VLANS;
93fc9e1b
YP		 293 dev->caps.log_num_prios = use_prio ? 3 : 0;
294
295 for (i = 1; i <= dev->caps.num_ports; ++i) {
ab9c17a0
JM		 296 dev->caps.port_type[i] = MLX4_PORT_TYPE_NONE;
297 if (dev->caps.supported_type[i]) {
298 /* if only ETH is supported - assign ETH */
299 if (dev->caps.supported_type[i] == MLX4_PORT_TYPE_ETH)
300 dev->caps.port_type[i] = MLX4_PORT_TYPE_ETH;
105c320f 301 /* if only IB is supported, assign IB */
ab9c17a0 302 else if (dev->caps.supported_type[i] ==
105c320f
JM		 303 MLX4_PORT_TYPE_IB)
304 dev->caps.port_type[i] = MLX4_PORT_TYPE_IB;
ab9c17a0 305 else {
105c320f
JM		 306 /* if IB and ETH are supported, we set the port
307 * type according to user selection of port type;
308 * if user selected none, take the FW hint */
309 if (port_type_array[i - 1] == MLX4_PORT_TYPE_NONE)
8d0fc7b6
YP		 310 dev->caps.port_type[i] = dev->caps.suggested_type[i] ?
311 MLX4_PORT_TYPE_ETH : MLX4_PORT_TYPE_IB;
ab9c17a0 312 else
105c320f 313 dev->caps.port_type[i] = port_type_array[i - 1];
ab9c17a0
JM		 314 }
315 }
8d0fc7b6
YP		 316 /*
317 * Link sensing is allowed on the port if 3 conditions are true:
318 * 1. Both protocols are supported on the port.
319 * 2. Different types are supported on the port
320 * 3. FW declared that it supports link sensing
321 */
27bf91d6 322 mlx4_priv(dev)->sense.sense_allowed[i] =
58a60168 323 ((dev->caps.supported_type[i] == MLX4_PORT_TYPE_AUTO) &&
8d0fc7b6 324 (dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
58a60168 325 (dev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT));
7ff93f8b 326
8d0fc7b6
YP		 327 /*
328 * If "default_sense" bit is set, we move the port to "AUTO" mode
329 * and perform sense_port FW command to try and set the correct
330 * port type from beginning
331 */
46c46747 332 if (mlx4_priv(dev)->sense.sense_allowed[i] && dev->caps.default_sense[i]) {
8d0fc7b6
YP		 333 enum mlx4_port_type sensed_port = MLX4_PORT_TYPE_NONE;
334 dev->caps.possible_type[i] = MLX4_PORT_TYPE_AUTO;
335 mlx4_SENSE_PORT(dev, i, &sensed_port);
336 if (sensed_port != MLX4_PORT_TYPE_NONE)
337 dev->caps.port_type[i] = sensed_port;
338 } else {
339 dev->caps.possible_type[i] = dev->caps.port_type[i];
340 }
341
93fc9e1b
YP		 342 if (dev->caps.log_num_macs > dev_cap->log_max_macs[i]) {
343 dev->caps.log_num_macs = dev_cap->log_max_macs[i];
344 mlx4_warn(dev, "Requested number of MACs is too much "
345 "for port %d, reducing to %d.\n",
346 i, 1 << dev->caps.log_num_macs);
347 }
348 if (dev->caps.log_num_vlans > dev_cap->log_max_vlans[i]) {
349 dev->caps.log_num_vlans = dev_cap->log_max_vlans[i];
350 mlx4_warn(dev, "Requested number of VLANs is too much "
351 "for port %d, reducing to %d.\n",
352 i, 1 << dev->caps.log_num_vlans);
353 }
354 }
355
f2a3f6a3
OG		 356 dev->caps.max_counters = 1 << ilog2(dev_cap->max_counters);
357
93fc9e1b
YP		 358 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] = dev_cap->reserved_qps;
359 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] =
360 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] =
361 (1 << dev->caps.log_num_macs) *
362 (1 << dev->caps.log_num_vlans) *
363 (1 << dev->caps.log_num_prios) *
364 dev->caps.num_ports;
365 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH] = MLX4_NUM_FEXCH;
366
367 dev->caps.reserved_qps = dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] +
368 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] +
369 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] +
370 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH];
371
e2c76824 372 dev->caps.sqp_demux = (mlx4_is_master(dev)) ? MLX4_MAX_NUM_SLAVES : 0;
08ff3235
OG		 373
374 if (!enable_64b_cqe_eqe) {
375 if (dev_cap->flags &
376 (MLX4_DEV_CAP_FLAG_64B_CQE | MLX4_DEV_CAP_FLAG_64B_EQE)) {
377 mlx4_warn(dev, "64B EQEs/CQEs supported by the device but not enabled\n");
378 dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_64B_CQE;
379 dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_64B_EQE;
380 }
381 }
382
383 if ((dev_cap->flags &
384 (MLX4_DEV_CAP_FLAG_64B_CQE | MLX4_DEV_CAP_FLAG_64B_EQE)) &&
385 mlx4_is_master(dev))
386 dev->caps.function_caps |= MLX4_FUNC_CAP_64B_EQE_CQE;
387
225c7b1f
RD		 388 return 0;
389}
ab9c17a0
JM		 390/*The function checks if there are live vf, return the num of them*/
391static int mlx4_how_many_lives_vf(struct mlx4_dev *dev)
392{
393 struct mlx4_priv *priv = mlx4_priv(dev);
394 struct mlx4_slave_state *s_state;
395 int i;
396 int ret = 0;
397
398 for (i = 1/*the ppf is 0*/; i < dev->num_slaves; ++i) {
399 s_state = &priv->mfunc.master.slave_state[i];
400 if (s_state->active && s_state->last_cmd !=
401 MLX4_COMM_CMD_RESET) {
402 mlx4_warn(dev, "%s: slave: %d is still active\n",
403 __func__, i);
404 ret++;
405 }
406 }
407 return ret;
408}
409
396f2feb
JM		 410int mlx4_get_parav_qkey(struct mlx4_dev *dev, u32 qpn, u32 *qkey)
411{
412 u32 qk = MLX4_RESERVED_QKEY_BASE;
47605df9
JM		 413
414 if (qpn >= dev->phys_caps.base_tunnel_sqpn + 8 * MLX4_MFUNC_MAX ||
415 qpn < dev->phys_caps.base_proxy_sqpn)
396f2feb
JM		 416 return -EINVAL;
417
47605df9 418 if (qpn >= dev->phys_caps.base_tunnel_sqpn)
396f2feb 419 /* tunnel qp */
47605df9 420 qk += qpn - dev->phys_caps.base_tunnel_sqpn;
396f2feb 421 else
47605df9 422 qk += qpn - dev->phys_caps.base_proxy_sqpn;
396f2feb
JM		 423 *qkey = qk;
424 return 0;
425}
426EXPORT_SYMBOL(mlx4_get_parav_qkey);
427
54679e14
JM		 428void mlx4_sync_pkey_table(struct mlx4_dev *dev, int slave, int port, int i, int val)
429{
430 struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
431
432 if (!mlx4_is_master(dev))
433 return;
434
435 priv->virt2phys_pkey[slave][port - 1][i] = val;
436}
437EXPORT_SYMBOL(mlx4_sync_pkey_table);
438
afa8fd1d
JM		 439void mlx4_put_slave_node_guid(struct mlx4_dev *dev, int slave, __be64 guid)
440{
441 struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
442
443 if (!mlx4_is_master(dev))
444 return;
445
446 priv->slave_node_guids[slave] = guid;
447}
448EXPORT_SYMBOL(mlx4_put_slave_node_guid);
449
450__be64 mlx4_get_slave_node_guid(struct mlx4_dev *dev, int slave)
451{
452 struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
453
454 if (!mlx4_is_master(dev))
455 return 0;
456
457 return priv->slave_node_guids[slave];
458}
459EXPORT_SYMBOL(mlx4_get_slave_node_guid);
460
e10903b0 461int mlx4_is_slave_active(struct mlx4_dev *dev, int slave)
ab9c17a0
JM		 462{
463 struct mlx4_priv *priv = mlx4_priv(dev);
464 struct mlx4_slave_state *s_slave;
465
466 if (!mlx4_is_master(dev))
467 return 0;
468
469 s_slave = &priv->mfunc.master.slave_state[slave];
470 return !!s_slave->active;
471}
472EXPORT_SYMBOL(mlx4_is_slave_active);
473
7b8157be
JM		 474static void slave_adjust_steering_mode(struct mlx4_dev *dev,
475 struct mlx4_dev_cap *dev_cap,
476 struct mlx4_init_hca_param *hca_param)
477{
478 dev->caps.steering_mode = hca_param->steering_mode;
479 if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED) {
480 dev->caps.num_qp_per_mgm = dev_cap->fs_max_num_qp_per_entry;
481 dev->caps.fs_log_max_ucast_qp_range_size =
482 dev_cap->fs_log_max_ucast_qp_range_size;
483 } else
484 dev->caps.num_qp_per_mgm =
485 4 * ((1 << hca_param->log_mc_entry_sz)/16 - 2);
486
487 mlx4_dbg(dev, "Steering mode is: %s\n",
488 mlx4_steering_mode_str(dev->caps.steering_mode));
489}
490
ab9c17a0
JM		 491static int mlx4_slave_cap(struct mlx4_dev *dev)
492{
493 int err;
494 u32 page_size;
495 struct mlx4_dev_cap dev_cap;
496 struct mlx4_func_cap func_cap;
497 struct mlx4_init_hca_param hca_param;
498 int i;
499
500 memset(&hca_param, 0, sizeof(hca_param));
501 err = mlx4_QUERY_HCA(dev, &hca_param);
502 if (err) {
503 mlx4_err(dev, "QUERY_HCA command failed, aborting.\n");
504 return err;
505 }
506
507 /*fail if the hca has an unknown capability */
508 if ((hca_param.global_caps | HCA_GLOBAL_CAP_MASK) !=
509 HCA_GLOBAL_CAP_MASK) {
510 mlx4_err(dev, "Unknown hca global capabilities\n");
511 return -ENOSYS;
512 }
513
514 mlx4_log_num_mgm_entry_size = hca_param.log_mc_entry_sz;
515
516 memset(&dev_cap, 0, sizeof(dev_cap));
b91cb3eb 517 dev->caps.max_qp_dest_rdma = 1 << hca_param.log_rd_per_qp;
ab9c17a0
JM		 518 err = mlx4_dev_cap(dev, &dev_cap);
519 if (err) {
520 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
521 return err;
522 }
523
b91cb3eb
JM		 524 err = mlx4_QUERY_FW(dev);
525 if (err)
526 mlx4_err(dev, "QUERY_FW command failed: could not get FW version.\n");
527
ab9c17a0
JM		 528 page_size = ~dev->caps.page_size_cap + 1;
529 mlx4_warn(dev, "HCA minimum page size:%d\n", page_size);
530 if (page_size > PAGE_SIZE) {
531 mlx4_err(dev, "HCA minimum page size of %d bigger than "
532 "kernel PAGE_SIZE of %ld, aborting.\n",
533 page_size, PAGE_SIZE);
534 return -ENODEV;
535 }
536
537 /* slave gets uar page size from QUERY_HCA fw command */
538 dev->caps.uar_page_size = 1 << (hca_param.uar_page_sz + 12);
539
540 /* TODO: relax this assumption */
541 if (dev->caps.uar_page_size != PAGE_SIZE) {
542 mlx4_err(dev, "UAR size:%d != kernel PAGE_SIZE of %ld\n",
543 dev->caps.uar_page_size, PAGE_SIZE);
544 return -ENODEV;
545 }
546
547 memset(&func_cap, 0, sizeof(func_cap));
47605df9 548 err = mlx4_QUERY_FUNC_CAP(dev, 0, &func_cap);
ab9c17a0 549 if (err) {
47605df9
JM		 550 mlx4_err(dev, "QUERY_FUNC_CAP general command failed, aborting (%d).\n",
551 err);
ab9c17a0
JM		 552 return err;
553 }
554
555 if ((func_cap.pf_context_behaviour | PF_CONTEXT_BEHAVIOUR_MASK) !=
556 PF_CONTEXT_BEHAVIOUR_MASK) {
557 mlx4_err(dev, "Unknown pf context behaviour\n");
558 return -ENOSYS;
559 }
560
ab9c17a0
JM		 561 dev->caps.num_ports = func_cap.num_ports;
562 dev->caps.num_qps = func_cap.qp_quota;
563 dev->caps.num_srqs = func_cap.srq_quota;
564 dev->caps.num_cqs = func_cap.cq_quota;
565 dev->caps.num_eqs = func_cap.max_eq;
566 dev->caps.reserved_eqs = func_cap.reserved_eq;
567 dev->caps.num_mpts = func_cap.mpt_quota;
568 dev->caps.num_mtts = func_cap.mtt_quota;
569 dev->caps.num_pds = MLX4_NUM_PDS;
570 dev->caps.num_mgms = 0;
571 dev->caps.num_amgms = 0;
572
ab9c17a0
JM		 573 if (dev->caps.num_ports > MLX4_MAX_PORTS) {
574 mlx4_err(dev, "HCA has %d ports, but we only support %d, "
575 "aborting.\n", dev->caps.num_ports, MLX4_MAX_PORTS);
576 return -ENODEV;
577 }
578
47605df9
JM		 579 dev->caps.qp0_tunnel = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
580 dev->caps.qp0_proxy = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
581 dev->caps.qp1_tunnel = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
582 dev->caps.qp1_proxy = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
583
584 if (!dev->caps.qp0_tunnel || !dev->caps.qp0_proxy ||
585 !dev->caps.qp1_tunnel || !dev->caps.qp1_proxy) {
586 err = -ENOMEM;
587 goto err_mem;
588 }
589
6634961c 590 for (i = 1; i <= dev->caps.num_ports; ++i) {
47605df9
JM		 591 err = mlx4_QUERY_FUNC_CAP(dev, (u32) i, &func_cap);
592 if (err) {
593 mlx4_err(dev, "QUERY_FUNC_CAP port command failed for"
594 " port %d, aborting (%d).\n", i, err);
595 goto err_mem;
596 }
597 dev->caps.qp0_tunnel[i - 1] = func_cap.qp0_tunnel_qpn;
598 dev->caps.qp0_proxy[i - 1] = func_cap.qp0_proxy_qpn;
599 dev->caps.qp1_tunnel[i - 1] = func_cap.qp1_tunnel_qpn;
600 dev->caps.qp1_proxy[i - 1] = func_cap.qp1_proxy_qpn;
6230bb23 601 dev->caps.port_mask[i] = dev->caps.port_type[i];
6634961c
JM		 602 if (mlx4_get_slave_pkey_gid_tbl_len(dev, i,
603 &dev->caps.gid_table_len[i],
604 &dev->caps.pkey_table_len[i]))
47605df9 605 goto err_mem;
6634961c 606 }
6230bb23 607
ab9c17a0
JM		 608 if (dev->caps.uar_page_size * (dev->caps.num_uars -
609 dev->caps.reserved_uars) >
610 pci_resource_len(dev->pdev, 2)) {
611 mlx4_err(dev, "HCA reported UAR region size of 0x%x bigger than "
612 "PCI resource 2 size of 0x%llx, aborting.\n",
613 dev->caps.uar_page_size * dev->caps.num_uars,
614 (unsigned long long) pci_resource_len(dev->pdev, 2));
47605df9 615 goto err_mem;
ab9c17a0
JM		 616 }
617
08ff3235
OG		 618 if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_64B_EQE_ENABLED) {
619 dev->caps.eqe_size = 64;
620 dev->caps.eqe_factor = 1;
621 } else {
622 dev->caps.eqe_size = 32;
623 dev->caps.eqe_factor = 0;
624 }
625
626 if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_64B_CQE_ENABLED) {
627 dev->caps.cqe_size = 64;
628 dev->caps.userspace_caps |= MLX4_USER_DEV_CAP_64B_CQE;
629 } else {
630 dev->caps.cqe_size = 32;
631 }
632
7b8157be
JM		 633 slave_adjust_steering_mode(dev, &dev_cap, &hca_param);
634
ab9c17a0 635 return 0;
47605df9
JM		 636
637err_mem:
638 kfree(dev->caps.qp0_tunnel);
639 kfree(dev->caps.qp0_proxy);
640 kfree(dev->caps.qp1_tunnel);
641 kfree(dev->caps.qp1_proxy);
642 dev->caps.qp0_tunnel = dev->caps.qp0_proxy =
643 dev->caps.qp1_tunnel = dev->caps.qp1_proxy = NULL;
644
645 return err;
ab9c17a0 646}
225c7b1f 647
7ff93f8b
YP		 648/*
649 * Change the port configuration of the device.
650 * Every user of this function must hold the port mutex.
651 */
27bf91d6
YP		 652int mlx4_change_port_types(struct mlx4_dev *dev,
653 enum mlx4_port_type *port_types)
7ff93f8b
YP		 654{
655 int err = 0;
656 int change = 0;
657 int port;
658
659 for (port = 0; port < dev->caps.num_ports; port++) {
27bf91d6
YP		 660 /* Change the port type only if the new type is different
661 * from the current, and not set to Auto */
3d8f9308 662 if (port_types[port] != dev->caps.port_type[port + 1])
7ff93f8b 663 change = 1;
7ff93f8b
YP		 664 }
665 if (change) {
666 mlx4_unregister_device(dev);
667 for (port = 1; port <= dev->caps.num_ports; port++) {
668 mlx4_CLOSE_PORT(dev, port);
1e0f03d5 669 dev->caps.port_type[port] = port_types[port - 1];
6634961c 670 err = mlx4_SET_PORT(dev, port, -1);
7ff93f8b
YP		 671 if (err) {
672 mlx4_err(dev, "Failed to set port %d, "
673 "aborting\n", port);
674 goto out;
675 }
676 }
677 mlx4_set_port_mask(dev);
678 err = mlx4_register_device(dev);
679 }
680
681out:
682 return err;
683}
684
685static ssize_t show_port_type(struct device *dev,
686 struct device_attribute *attr,
687 char *buf)
688{
689 struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
690 port_attr);
691 struct mlx4_dev *mdev = info->dev;
27bf91d6
YP		 692 char type[8];
693
694 sprintf(type, "%s",
695 (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_IB) ?
696 "ib" : "eth");
697 if (mdev->caps.possible_type[info->port] == MLX4_PORT_TYPE_AUTO)
698 sprintf(buf, "auto (%s)\n", type);
699 else
700 sprintf(buf, "%s\n", type);
7ff93f8b 701
27bf91d6 702 return strlen(buf);
7ff93f8b
YP		 703}
704
705static ssize_t set_port_type(struct device *dev,
706 struct device_attribute *attr,
707 const char *buf, size_t count)
708{
709 struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
710 port_attr);
711 struct mlx4_dev *mdev = info->dev;
712 struct mlx4_priv *priv = mlx4_priv(mdev);
713 enum mlx4_port_type types[MLX4_MAX_PORTS];
27bf91d6 714 enum mlx4_port_type new_types[MLX4_MAX_PORTS];
7ff93f8b
YP		 715 int i;
716 int err = 0;
717
718 if (!strcmp(buf, "ib\n"))
719 info->tmp_type = MLX4_PORT_TYPE_IB;
720 else if (!strcmp(buf, "eth\n"))
721 info->tmp_type = MLX4_PORT_TYPE_ETH;
27bf91d6
YP		 722 else if (!strcmp(buf, "auto\n"))
723 info->tmp_type = MLX4_PORT_TYPE_AUTO;
7ff93f8b
YP		 724 else {
725 mlx4_err(mdev, "%s is not supported port type\n", buf);
726 return -EINVAL;
727 }
728
27bf91d6 729 mlx4_stop_sense(mdev);
7ff93f8b 730 mutex_lock(&priv->port_mutex);
27bf91d6
YP		 731 /* Possible type is always the one that was delivered */
732 mdev->caps.possible_type[info->port] = info->tmp_type;
733
734 for (i = 0; i < mdev->caps.num_ports; i++) {
7ff93f8b 735 types[i] = priv->port[i+1].tmp_type ? priv->port[i+1].tmp_type :
27bf91d6
YP		 736 mdev->caps.possible_type[i+1];
737 if (types[i] == MLX4_PORT_TYPE_AUTO)
738 types[i] = mdev->caps.port_type[i+1];
739 }
7ff93f8b 740
58a60168
YP		 741 if (!(mdev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
742 !(mdev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT)) {
27bf91d6
YP		 743 for (i = 1; i <= mdev->caps.num_ports; i++) {
744 if (mdev->caps.possible_type[i] == MLX4_PORT_TYPE_AUTO) {
745 mdev->caps.possible_type[i] = mdev->caps.port_type[i];
746 err = -EINVAL;
747 }
748 }
749 }
750 if (err) {
751 mlx4_err(mdev, "Auto sensing is not supported on this HCA. "
752 "Set only 'eth' or 'ib' for both ports "
753 "(should be the same)\n");
754 goto out;
755 }
756
757 mlx4_do_sense_ports(mdev, new_types, types);
758
759 err = mlx4_check_port_params(mdev, new_types);
7ff93f8b
YP		 760 if (err)
761 goto out;
762
27bf91d6
YP		 763 /* We are about to apply the changes after the configuration
764 * was verified, no need to remember the temporary types
765 * any more */
766 for (i = 0; i < mdev->caps.num_ports; i++)
767 priv->port[i + 1].tmp_type = 0;
7ff93f8b 768
27bf91d6 769 err = mlx4_change_port_types(mdev, new_types);
7ff93f8b
YP		 770
771out:
27bf91d6 772 mlx4_start_sense(mdev);
7ff93f8b
YP		 773 mutex_unlock(&priv->port_mutex);
774 return err ? err : count;
775}
776
096335b3
OG		 777enum ibta_mtu {
778 IB_MTU_256 = 1,
779 IB_MTU_512 = 2,
780 IB_MTU_1024 = 3,
781 IB_MTU_2048 = 4,
782 IB_MTU_4096 = 5
783};
784
785static inline int int_to_ibta_mtu(int mtu)
786{
787 switch (mtu) {
788 case 256: return IB_MTU_256;
789 case 512: return IB_MTU_512;
790 case 1024: return IB_MTU_1024;
791 case 2048: return IB_MTU_2048;
792 case 4096: return IB_MTU_4096;
793 default: return -1;
794 }
795}
796
797static inline int ibta_mtu_to_int(enum ibta_mtu mtu)
798{
799 switch (mtu) {
800 case IB_MTU_256: return 256;
801 case IB_MTU_512: return 512;
802 case IB_MTU_1024: return 1024;
803 case IB_MTU_2048: return 2048;
804 case IB_MTU_4096: return 4096;
805 default: return -1;
806 }
807}
808
809static ssize_t show_port_ib_mtu(struct device *dev,
810 struct device_attribute *attr,
811 char *buf)
812{
813 struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
814 port_mtu_attr);
815 struct mlx4_dev *mdev = info->dev;
816
817 if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH)
818 mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
819
820 sprintf(buf, "%d\n",
821 ibta_mtu_to_int(mdev->caps.port_ib_mtu[info->port]));
822 return strlen(buf);
823}
824
825static ssize_t set_port_ib_mtu(struct device *dev,
826 struct device_attribute *attr,
827 const char *buf, size_t count)
828{
829 struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
830 port_mtu_attr);
831 struct mlx4_dev *mdev = info->dev;
832 struct mlx4_priv *priv = mlx4_priv(mdev);
833 int err, port, mtu, ibta_mtu = -1;
834
835 if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH) {
836 mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
837 return -EINVAL;
838 }
839
840 err = sscanf(buf, "%d", &mtu);
841 if (err > 0)
842 ibta_mtu = int_to_ibta_mtu(mtu);
843
844 if (err <= 0 || ibta_mtu < 0) {
845 mlx4_err(mdev, "%s is invalid IBTA mtu\n", buf);
846 return -EINVAL;
847 }
848
849 mdev->caps.port_ib_mtu[info->port] = ibta_mtu;
850
851 mlx4_stop_sense(mdev);
852 mutex_lock(&priv->port_mutex);
853 mlx4_unregister_device(mdev);
854 for (port = 1; port <= mdev->caps.num_ports; port++) {
855 mlx4_CLOSE_PORT(mdev, port);
6634961c 856 err = mlx4_SET_PORT(mdev, port, -1);
096335b3
OG		 857 if (err) {
858 mlx4_err(mdev, "Failed to set port %d, "
859 "aborting\n", port);
860 goto err_set_port;
861 }
862 }
863 err = mlx4_register_device(mdev);
864err_set_port:
865 mutex_unlock(&priv->port_mutex);
866 mlx4_start_sense(mdev);
867 return err ? err : count;
868}
869
e8f9b2ed 870static int mlx4_load_fw(struct mlx4_dev *dev)
225c7b1f
RD		 871{
872 struct mlx4_priv *priv = mlx4_priv(dev);
873 int err;
874
875 priv->fw.fw_icm = mlx4_alloc_icm(dev, priv->fw.fw_pages,
5b0bf5e2 876 GFP_HIGHUSER | __GFP_NOWARN, 0);
225c7b1f
RD		 877 if (!priv->fw.fw_icm) {
878 mlx4_err(dev, "Couldn't allocate FW area, aborting.\n");
879 return -ENOMEM;
880 }
881
882 err = mlx4_MAP_FA(dev, priv->fw.fw_icm);
883 if (err) {
884 mlx4_err(dev, "MAP_FA command failed, aborting.\n");
885 goto err_free;
886 }
887
888 err = mlx4_RUN_FW(dev);
889 if (err) {
890 mlx4_err(dev, "RUN_FW command failed, aborting.\n");
891 goto err_unmap_fa;
892 }
893
894 return 0;
895
896err_unmap_fa:
897 mlx4_UNMAP_FA(dev);
898
899err_free:
5b0bf5e2 900 mlx4_free_icm(dev, priv->fw.fw_icm, 0);
225c7b1f
RD		 901 return err;
902}
903
e8f9b2ed
RD		 904static int mlx4_init_cmpt_table(struct mlx4_dev *dev, u64 cmpt_base,
905 int cmpt_entry_sz)
225c7b1f
RD		 906{
907 struct mlx4_priv *priv = mlx4_priv(dev);
908 int err;
ab9c17a0 909 int num_eqs;
225c7b1f
RD		 910
911 err = mlx4_init_icm_table(dev, &priv->qp_table.cmpt_table,
912 cmpt_base +
913 ((u64) (MLX4_CMPT_TYPE_QP *
914 cmpt_entry_sz) << MLX4_CMPT_SHIFT),
915 cmpt_entry_sz, dev->caps.num_qps,
93fc9e1b
YP		 916 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
917 0, 0);
225c7b1f
RD		 918 if (err)
919 goto err;
920
921 err = mlx4_init_icm_table(dev, &priv->srq_table.cmpt_table,
922 cmpt_base +
923 ((u64) (MLX4_CMPT_TYPE_SRQ *
924 cmpt_entry_sz) << MLX4_CMPT_SHIFT),
925 cmpt_entry_sz, dev->caps.num_srqs,
5b0bf5e2 926 dev->caps.reserved_srqs, 0, 0);
225c7b1f
RD		 927 if (err)
928 goto err_qp;
929
930 err = mlx4_init_icm_table(dev, &priv->cq_table.cmpt_table,
931 cmpt_base +
932 ((u64) (MLX4_CMPT_TYPE_CQ *
933 cmpt_entry_sz) << MLX4_CMPT_SHIFT),
934 cmpt_entry_sz, dev->caps.num_cqs,
5b0bf5e2 935 dev->caps.reserved_cqs, 0, 0);
225c7b1f
RD		 936 if (err)
937 goto err_srq;
938
3fc929e2
MA		 939 num_eqs = (mlx4_is_master(dev)) ? dev->phys_caps.num_phys_eqs :
940 dev->caps.num_eqs;
225c7b1f
RD		 941 err = mlx4_init_icm_table(dev, &priv->eq_table.cmpt_table,
942 cmpt_base +
943 ((u64) (MLX4_CMPT_TYPE_EQ *
944 cmpt_entry_sz) << MLX4_CMPT_SHIFT),
ab9c17a0 945 cmpt_entry_sz, num_eqs, num_eqs, 0, 0);
225c7b1f
RD		 946 if (err)
947 goto err_cq;
948
949 return 0;
950
951err_cq:
952 mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
953
954err_srq:
955 mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
956
957err_qp:
958 mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
959
960err:
961 return err;
962}
963
3d73c288
RD		 964static int mlx4_init_icm(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap,
965 struct mlx4_init_hca_param *init_hca, u64 icm_size)
225c7b1f
RD		 966{
967 struct mlx4_priv *priv = mlx4_priv(dev);
968 u64 aux_pages;
ab9c17a0 969 int num_eqs;
225c7b1f
RD		 970 int err;
971
972 err = mlx4_SET_ICM_SIZE(dev, icm_size, &aux_pages);
973 if (err) {
974 mlx4_err(dev, "SET_ICM_SIZE command failed, aborting.\n");
975 return err;
976 }
977
978 mlx4_dbg(dev, "%lld KB of HCA context requires %lld KB aux memory.\n",
979 (unsigned long long) icm_size >> 10,
980 (unsigned long long) aux_pages << 2);
981
982 priv->fw.aux_icm = mlx4_alloc_icm(dev, aux_pages,
5b0bf5e2 983 GFP_HIGHUSER | __GFP_NOWARN, 0);
225c7b1f
RD		 984 if (!priv->fw.aux_icm) {
985 mlx4_err(dev, "Couldn't allocate aux memory, aborting.\n");
986 return -ENOMEM;
987 }
988
989 err = mlx4_MAP_ICM_AUX(dev, priv->fw.aux_icm);
990 if (err) {
991 mlx4_err(dev, "MAP_ICM_AUX command failed, aborting.\n");
992 goto err_free_aux;
993 }
994
995 err = mlx4_init_cmpt_table(dev, init_hca->cmpt_base, dev_cap->cmpt_entry_sz);
996 if (err) {
997 mlx4_err(dev, "Failed to map cMPT context memory, aborting.\n");
998 goto err_unmap_aux;
999 }
1000
ab9c17a0 1001
3fc929e2
MA		 1002 num_eqs = (mlx4_is_master(dev)) ? dev->phys_caps.num_phys_eqs :
1003 dev->caps.num_eqs;
fa0681d2
RD		 1004 err = mlx4_init_icm_table(dev, &priv->eq_table.table,
1005 init_hca->eqc_base, dev_cap->eqc_entry_sz,
ab9c17a0 1006 num_eqs, num_eqs, 0, 0);
225c7b1f
RD		 1007 if (err) {
1008 mlx4_err(dev, "Failed to map EQ context memory, aborting.\n");
1009 goto err_unmap_cmpt;
1010 }
1011
d7bb58fb
JM		 1012 /*
1013 * Reserved MTT entries must be aligned up to a cacheline
1014 * boundary, since the FW will write to them, while the driver
1015 * writes to all other MTT entries. (The variable
1016 * dev->caps.mtt_entry_sz below is really the MTT segment
1017 * size, not the raw entry size)
1018 */
1019 dev->caps.reserved_mtts =
1020 ALIGN(dev->caps.reserved_mtts * dev->caps.mtt_entry_sz,
1021 dma_get_cache_alignment()) / dev->caps.mtt_entry_sz;
1022
225c7b1f
RD		 1023 err = mlx4_init_icm_table(dev, &priv->mr_table.mtt_table,
1024 init_hca->mtt_base,
1025 dev->caps.mtt_entry_sz,
2b8fb286 1026 dev->caps.num_mtts,
5b0bf5e2 1027 dev->caps.reserved_mtts, 1, 0);
225c7b1f
RD		 1028 if (err) {
1029 mlx4_err(dev, "Failed to map MTT context memory, aborting.\n");
1030 goto err_unmap_eq;
1031 }
1032
1033 err = mlx4_init_icm_table(dev, &priv->mr_table.dmpt_table,
1034 init_hca->dmpt_base,
1035 dev_cap->dmpt_entry_sz,
1036 dev->caps.num_mpts,
5b0bf5e2 1037 dev->caps.reserved_mrws, 1, 1);
225c7b1f
RD		 1038 if (err) {
1039 mlx4_err(dev, "Failed to map dMPT context memory, aborting.\n");
1040 goto err_unmap_mtt;
1041 }
1042
1043 err = mlx4_init_icm_table(dev, &priv->qp_table.qp_table,
1044 init_hca->qpc_base,
1045 dev_cap->qpc_entry_sz,
1046 dev->caps.num_qps,
93fc9e1b
YP		 1047 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1048 0, 0);
225c7b1f
RD		 1049 if (err) {
1050 mlx4_err(dev, "Failed to map QP context memory, aborting.\n");
1051 goto err_unmap_dmpt;
1052 }
1053
1054 err = mlx4_init_icm_table(dev, &priv->qp_table.auxc_table,
1055 init_hca->auxc_base,
1056 dev_cap->aux_entry_sz,
1057 dev->caps.num_qps,
93fc9e1b
YP		 1058 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1059 0, 0);
225c7b1f
RD		 1060 if (err) {
1061 mlx4_err(dev, "Failed to map AUXC context memory, aborting.\n");
1062 goto err_unmap_qp;
1063 }
1064
1065 err = mlx4_init_icm_table(dev, &priv->qp_table.altc_table,
1066 init_hca->altc_base,
1067 dev_cap->altc_entry_sz,
1068 dev->caps.num_qps,
93fc9e1b
YP		 1069 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1070 0, 0);
225c7b1f
RD		 1071 if (err) {
1072 mlx4_err(dev, "Failed to map ALTC context memory, aborting.\n");
1073 goto err_unmap_auxc;
1074 }
1075
1076 err = mlx4_init_icm_table(dev, &priv->qp_table.rdmarc_table,
1077 init_hca->rdmarc_base,
1078 dev_cap->rdmarc_entry_sz << priv->qp_table.rdmarc_shift,
1079 dev->caps.num_qps,
93fc9e1b
YP		 1080 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1081 0, 0);
225c7b1f
RD		 1082 if (err) {
1083 mlx4_err(dev, "Failed to map RDMARC context memory, aborting\n");
1084 goto err_unmap_altc;
1085 }
1086
1087 err = mlx4_init_icm_table(dev, &priv->cq_table.table,
1088 init_hca->cqc_base,
1089 dev_cap->cqc_entry_sz,
1090 dev->caps.num_cqs,
5b0bf5e2 1091 dev->caps.reserved_cqs, 0, 0);
225c7b1f
RD		 1092 if (err) {
1093 mlx4_err(dev, "Failed to map CQ context memory, aborting.\n");
1094 goto err_unmap_rdmarc;
1095 }
1096
1097 err = mlx4_init_icm_table(dev, &priv->srq_table.table,
1098 init_hca->srqc_base,
1099 dev_cap->srq_entry_sz,
1100 dev->caps.num_srqs,
5b0bf5e2 1101 dev->caps.reserved_srqs, 0, 0);
225c7b1f
RD		 1102 if (err) {
1103 mlx4_err(dev, "Failed to map SRQ context memory, aborting.\n");
1104 goto err_unmap_cq;
1105 }
1106
1107 /*
0ff1fb65
HHZ		 1108 * For flow steering device managed mode it is required to use
1109 * mlx4_init_icm_table. For B0 steering mode it's not strictly
1110 * required, but for simplicity just map the whole multicast
1111 * group table now. The table isn't very big and it's a lot
1112 * easier than trying to track ref counts.
225c7b1f
RD		 1113 */
1114 err = mlx4_init_icm_table(dev, &priv->mcg_table.table,
0ec2c0f8
EE		 1115 init_hca->mc_base,
1116 mlx4_get_mgm_entry_size(dev),
225c7b1f
RD		 1117 dev->caps.num_mgms + dev->caps.num_amgms,
1118 dev->caps.num_mgms + dev->caps.num_amgms,
5b0bf5e2 1119 0, 0);
225c7b1f
RD		 1120 if (err) {
1121 mlx4_err(dev, "Failed to map MCG context memory, aborting.\n");
1122 goto err_unmap_srq;
1123 }
1124
1125 return 0;
1126
1127err_unmap_srq:
1128 mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1129
1130err_unmap_cq:
1131 mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1132
1133err_unmap_rdmarc:
1134 mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1135
1136err_unmap_altc:
1137 mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1138
1139err_unmap_auxc:
1140 mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1141
1142err_unmap_qp:
1143 mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1144
1145err_unmap_dmpt:
1146 mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1147
1148err_unmap_mtt:
1149 mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
1150
1151err_unmap_eq:
fa0681d2 1152 mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
225c7b1f
RD		 1153
1154err_unmap_cmpt:
1155 mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1156 mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1157 mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1158 mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1159
1160err_unmap_aux:
1161 mlx4_UNMAP_ICM_AUX(dev);
1162
1163err_free_aux:
5b0bf5e2 1164 mlx4_free_icm(dev, priv->fw.aux_icm, 0);
225c7b1f
RD		 1165
1166 return err;
1167}
1168
1169static void mlx4_free_icms(struct mlx4_dev *dev)
1170{
1171 struct mlx4_priv *priv = mlx4_priv(dev);
1172
1173 mlx4_cleanup_icm_table(dev, &priv->mcg_table.table);
1174 mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1175 mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1176 mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1177 mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1178 mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1179 mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1180 mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1181 mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
fa0681d2 1182 mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
225c7b1f
RD		 1183 mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1184 mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1185 mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1186 mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
225c7b1f
RD		 1187
1188 mlx4_UNMAP_ICM_AUX(dev);
5b0bf5e2 1189 mlx4_free_icm(dev, priv->fw.aux_icm, 0);
225c7b1f
RD		 1190}
1191
ab9c17a0
JM		 1192static void mlx4_slave_exit(struct mlx4_dev *dev)
1193{
1194 struct mlx4_priv *priv = mlx4_priv(dev);
1195
f3d4c89e 1196 mutex_lock(&priv->cmd.slave_cmd_mutex);
ab9c17a0
JM		 1197 if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, MLX4_COMM_TIME))
1198 mlx4_warn(dev, "Failed to close slave function.\n");
f3d4c89e 1199 mutex_unlock(&priv->cmd.slave_cmd_mutex);
ab9c17a0
JM		 1200}
1201
c1b43dca
EC		 1202static int map_bf_area(struct mlx4_dev *dev)
1203{
1204 struct mlx4_priv *priv = mlx4_priv(dev);
1205 resource_size_t bf_start;
1206 resource_size_t bf_len;
1207 int err = 0;
1208
3d747473
JM		 1209 if (!dev->caps.bf_reg_size)
1210 return -ENXIO;
1211
ab9c17a0
JM		 1212 bf_start = pci_resource_start(dev->pdev, 2) +
1213 (dev->caps.num_uars << PAGE_SHIFT);
1214 bf_len = pci_resource_len(dev->pdev, 2) -
1215 (dev->caps.num_uars << PAGE_SHIFT);
c1b43dca
EC		 1216 priv->bf_mapping = io_mapping_create_wc(bf_start, bf_len);
1217 if (!priv->bf_mapping)
1218 err = -ENOMEM;
1219
1220 return err;
1221}
1222
1223static void unmap_bf_area(struct mlx4_dev *dev)
1224{
1225 if (mlx4_priv(dev)->bf_mapping)
1226 io_mapping_free(mlx4_priv(dev)->bf_mapping);
1227}
1228
225c7b1f
RD		 1229static void mlx4_close_hca(struct mlx4_dev *dev)
1230{
c1b43dca 1231 unmap_bf_area(dev);
ab9c17a0
JM		 1232 if (mlx4_is_slave(dev))
1233 mlx4_slave_exit(dev);
1234 else {
1235 mlx4_CLOSE_HCA(dev, 0);
1236 mlx4_free_icms(dev);
1237 mlx4_UNMAP_FA(dev);
1238 mlx4_free_icm(dev, mlx4_priv(dev)->fw.fw_icm, 0);
1239 }
1240}
1241
1242static int mlx4_init_slave(struct mlx4_dev *dev)
1243{
1244 struct mlx4_priv *priv = mlx4_priv(dev);
1245 u64 dma = (u64) priv->mfunc.vhcr_dma;
1246 int num_of_reset_retries = NUM_OF_RESET_RETRIES;
1247 int ret_from_reset = 0;
1248 u32 slave_read;
1249 u32 cmd_channel_ver;
1250
f3d4c89e 1251 mutex_lock(&priv->cmd.slave_cmd_mutex);
ab9c17a0
JM		 1252 priv->cmd.max_cmds = 1;
1253 mlx4_warn(dev, "Sending reset\n");
1254 ret_from_reset = mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0,
1255 MLX4_COMM_TIME);
1256 /* if we are in the middle of flr the slave will try
1257 * NUM_OF_RESET_RETRIES times before leaving.*/
1258 if (ret_from_reset) {
1259 if (MLX4_DELAY_RESET_SLAVE == ret_from_reset) {
1260 msleep(SLEEP_TIME_IN_RESET);
1261 while (ret_from_reset && num_of_reset_retries) {
1262 mlx4_warn(dev, "slave is currently in the"
1263 "middle of FLR. retrying..."
1264 "(try num:%d)\n",
1265 (NUM_OF_RESET_RETRIES -
1266 num_of_reset_retries + 1));
1267 ret_from_reset =
1268 mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET,
1269 0, MLX4_COMM_TIME);
1270 num_of_reset_retries = num_of_reset_retries - 1;
1271 }
1272 } else
1273 goto err;
1274 }
1275
1276 /* check the driver version - the slave I/F revision
1277 * must match the master's */
1278 slave_read = swab32(readl(&priv->mfunc.comm->slave_read));
1279 cmd_channel_ver = mlx4_comm_get_version();
1280
1281 if (MLX4_COMM_GET_IF_REV(cmd_channel_ver) !=
1282 MLX4_COMM_GET_IF_REV(slave_read)) {
1283 mlx4_err(dev, "slave driver version is not supported"
1284 " by the master\n");
1285 goto err;
1286 }
1287
1288 mlx4_warn(dev, "Sending vhcr0\n");
1289 if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR0, dma >> 48,
1290 MLX4_COMM_TIME))
1291 goto err;
1292 if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR1, dma >> 32,
1293 MLX4_COMM_TIME))
1294 goto err;
1295 if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR2, dma >> 16,
1296 MLX4_COMM_TIME))
1297 goto err;
1298 if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR_EN, dma, MLX4_COMM_TIME))
1299 goto err;
f3d4c89e
RD		 1300
1301 mutex_unlock(&priv->cmd.slave_cmd_mutex);
ab9c17a0
JM		 1302 return 0;
1303
1304err:
1305 mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, 0);
f3d4c89e 1306 mutex_unlock(&priv->cmd.slave_cmd_mutex);
ab9c17a0 1307 return -EIO;
225c7b1f
RD		 1308}
1309
6634961c
JM		 1310static void mlx4_parav_master_pf_caps(struct mlx4_dev *dev)
1311{
1312 int i;
1313
1314 for (i = 1; i <= dev->caps.num_ports; i++) {
1315 dev->caps.gid_table_len[i] = 1;
1316 dev->caps.pkey_table_len[i] =
1317 dev->phys_caps.pkey_phys_table_len[i] - 1;
1318 }
1319}
1320
7b8157be
JM		 1321static void choose_steering_mode(struct mlx4_dev *dev,
1322 struct mlx4_dev_cap *dev_cap)
1323{
1324 if (dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_FS_EN &&
1325 (!mlx4_is_mfunc(dev) ||
1326 (dev_cap->fs_max_num_qp_per_entry >= (num_vfs + 1)))) {
1327 dev->caps.steering_mode = MLX4_STEERING_MODE_DEVICE_MANAGED;
1328 dev->caps.num_qp_per_mgm = dev_cap->fs_max_num_qp_per_entry;
1329 dev->caps.fs_log_max_ucast_qp_range_size =
1330 dev_cap->fs_log_max_ucast_qp_range_size;
1331 } else {
1332 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER &&
1333 dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER)
1334 dev->caps.steering_mode = MLX4_STEERING_MODE_B0;
1335 else {
1336 dev->caps.steering_mode = MLX4_STEERING_MODE_A0;
1337
1338 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER ||
1339 dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER)
1340 mlx4_warn(dev, "Must have both UC_STEER and MC_STEER flags "
1341 "set to use B0 steering. Falling back to A0 steering mode.\n");
1342 }
1343 dev->caps.num_qp_per_mgm = mlx4_get_qp_per_mgm(dev);
1344 }
1345 mlx4_dbg(dev, "Steering mode is: %s\n",
1346 mlx4_steering_mode_str(dev->caps.steering_mode));
1347}
1348
3d73c288 1349static int mlx4_init_hca(struct mlx4_dev *dev)
225c7b1f
RD		 1350{
1351 struct mlx4_priv *priv = mlx4_priv(dev);
1352 struct mlx4_adapter adapter;
1353 struct mlx4_dev_cap dev_cap;
2d928651 1354 struct mlx4_mod_stat_cfg mlx4_cfg;
225c7b1f
RD		 1355 struct mlx4_profile profile;
1356 struct mlx4_init_hca_param init_hca;
1357 u64 icm_size;
1358 int err;
1359
ab9c17a0
JM		 1360 if (!mlx4_is_slave(dev)) {
1361 err = mlx4_QUERY_FW(dev);
1362 if (err) {
1363 if (err == -EACCES)
1364 mlx4_info(dev, "non-primary physical function, skipping.\n");
1365 else
1366 mlx4_err(dev, "QUERY_FW command failed, aborting.\n");
bef772eb 1367 return err;
ab9c17a0 1368 }
225c7b1f 1369
ab9c17a0
JM		 1370 err = mlx4_load_fw(dev);
1371 if (err) {
1372 mlx4_err(dev, "Failed to start FW, aborting.\n");
bef772eb 1373 return err;
ab9c17a0 1374 }
225c7b1f 1375
ab9c17a0
JM		 1376 mlx4_cfg.log_pg_sz_m = 1;
1377 mlx4_cfg.log_pg_sz = 0;
1378 err = mlx4_MOD_STAT_CFG(dev, &mlx4_cfg);
1379 if (err)
1380 mlx4_warn(dev, "Failed to override log_pg_sz parameter\n");
2d928651 1381
ab9c17a0
JM		 1382 err = mlx4_dev_cap(dev, &dev_cap);
1383 if (err) {
1384 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
1385 goto err_stop_fw;
1386 }
225c7b1f 1387
7b8157be
JM		 1388 choose_steering_mode(dev, &dev_cap);
1389
6634961c
JM		 1390 if (mlx4_is_master(dev))
1391 mlx4_parav_master_pf_caps(dev);
1392
0ff1fb65
HHZ		 1393 priv->fs_hash_mode = MLX4_FS_L2_HASH;
1394
1395 switch (priv->fs_hash_mode) {
1396 case MLX4_FS_L2_HASH:
1397 init_hca.fs_hash_enable_bits = 0;
1398 break;
1399
1400 case MLX4_FS_L2_L3_L4_HASH:
1401 /* Enable flow steering with
1402 * udp unicast and tcp unicast
1403 */
1404 init_hca.fs_hash_enable_bits =
1405 MLX4_FS_UDP_UC_EN | MLX4_FS_TCP_UC_EN;
1406 break;
1407 }
1408
ab9c17a0 1409 profile = default_profile;
0ff1fb65
HHZ		 1410 if (dev->caps.steering_mode ==
1411 MLX4_STEERING_MODE_DEVICE_MANAGED)
1412 profile.num_mcg = MLX4_FS_NUM_MCG;
225c7b1f 1413
ab9c17a0
JM		 1414 icm_size = mlx4_make_profile(dev, &profile, &dev_cap,
1415 &init_hca);
1416 if ((long long) icm_size < 0) {
1417 err = icm_size;
1418 goto err_stop_fw;
1419 }
225c7b1f 1420
a5bbe892
EC		 1421 dev->caps.max_fmr_maps = (1 << (32 - ilog2(dev->caps.num_mpts))) - 1;
1422
ab9c17a0
JM		 1423 init_hca.log_uar_sz = ilog2(dev->caps.num_uars);
1424 init_hca.uar_page_sz = PAGE_SHIFT - 12;
c1b43dca 1425
ab9c17a0
JM		 1426 err = mlx4_init_icm(dev, &dev_cap, &init_hca, icm_size);
1427 if (err)
1428 goto err_stop_fw;
225c7b1f 1429
ab9c17a0
JM		 1430 err = mlx4_INIT_HCA(dev, &init_hca);
1431 if (err) {
1432 mlx4_err(dev, "INIT_HCA command failed, aborting.\n");
1433 goto err_free_icm;
1434 }
1435 } else {
1436 err = mlx4_init_slave(dev);
1437 if (err) {
1438 mlx4_err(dev, "Failed to initialize slave\n");
bef772eb 1439 return err;
ab9c17a0 1440 }
225c7b1f 1441
ab9c17a0
JM		 1442 err = mlx4_slave_cap(dev);
1443 if (err) {
1444 mlx4_err(dev, "Failed to obtain slave caps\n");
1445 goto err_close;
1446 }
225c7b1f
RD		 1447 }
1448
ab9c17a0
JM		 1449 if (map_bf_area(dev))
1450 mlx4_dbg(dev, "Failed to map blue flame area\n");
1451
1452 /*Only the master set the ports, all the rest got it from it.*/
1453 if (!mlx4_is_slave(dev))
1454 mlx4_set_port_mask(dev);
1455
225c7b1f
RD		 1456 err = mlx4_QUERY_ADAPTER(dev, &adapter);
1457 if (err) {
1458 mlx4_err(dev, "QUERY_ADAPTER command failed, aborting.\n");
bef772eb 1459 goto unmap_bf;
225c7b1f
RD		 1460 }
1461
1462 priv->eq_table.inta_pin = adapter.inta_pin;
cd9281d8 1463 memcpy(dev->board_id, adapter.board_id, sizeof dev->board_id);
225c7b1f
RD		 1464
1465 return 0;
1466
bef772eb
AY		 1467unmap_bf:
1468 unmap_bf_area(dev);
1469
225c7b1f 1470err_close:
41929ed2
DB		 1471 if (mlx4_is_slave(dev))
1472 mlx4_slave_exit(dev);
1473 else
1474 mlx4_CLOSE_HCA(dev, 0);
225c7b1f
RD		 1475
1476err_free_icm:
ab9c17a0
JM		 1477 if (!mlx4_is_slave(dev))
1478 mlx4_free_icms(dev);
225c7b1f
RD		 1479
1480err_stop_fw:
ab9c17a0
JM		 1481 if (!mlx4_is_slave(dev)) {
1482 mlx4_UNMAP_FA(dev);
1483 mlx4_free_icm(dev, priv->fw.fw_icm, 0);
1484 }
225c7b1f
RD		 1485 return err;
1486}
1487
f2a3f6a3
OG		 1488static int mlx4_init_counters_table(struct mlx4_dev *dev)
1489{
1490 struct mlx4_priv *priv = mlx4_priv(dev);
1491 int nent;
1492
1493 if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
1494 return -ENOENT;
1495
1496 nent = dev->caps.max_counters;
1497 return mlx4_bitmap_init(&priv->counters_bitmap, nent, nent - 1, 0, 0);
1498}
1499
1500static void mlx4_cleanup_counters_table(struct mlx4_dev *dev)
1501{
1502 mlx4_bitmap_cleanup(&mlx4_priv(dev)->counters_bitmap);
1503}
1504
ba062d52 1505int __mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx)
f2a3f6a3
OG		 1506{
1507 struct mlx4_priv *priv = mlx4_priv(dev);
1508
1509 if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
1510 return -ENOENT;
1511
1512 *idx = mlx4_bitmap_alloc(&priv->counters_bitmap);
1513 if (*idx == -1)
1514 return -ENOMEM;
1515
1516 return 0;
1517}
ba062d52
JM		 1518
1519int mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx)
1520{
1521 u64 out_param;
1522 int err;
1523
1524 if (mlx4_is_mfunc(dev)) {
1525 err = mlx4_cmd_imm(dev, 0, &out_param, RES_COUNTER,
1526 RES_OP_RESERVE, MLX4_CMD_ALLOC_RES,
1527 MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
1528 if (!err)
1529 *idx = get_param_l(&out_param);
1530
1531 return err;
1532 }
1533 return __mlx4_counter_alloc(dev, idx);
1534}
f2a3f6a3
OG		 1535EXPORT_SYMBOL_GPL(mlx4_counter_alloc);
1536
ba062d52 1537void __mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
f2a3f6a3
OG		 1538{
1539 mlx4_bitmap_free(&mlx4_priv(dev)->counters_bitmap, idx);
1540 return;
1541}
ba062d52
JM		 1542
1543void mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
1544{
1545 u64 in_param;
1546
1547 if (mlx4_is_mfunc(dev)) {
1548 set_param_l(&in_param, idx);
1549 mlx4_cmd(dev, in_param, RES_COUNTER, RES_OP_RESERVE,
1550 MLX4_CMD_FREE_RES, MLX4_CMD_TIME_CLASS_A,
1551 MLX4_CMD_WRAPPED);
1552 return;
1553 }
1554 __mlx4_counter_free(dev, idx);
1555}
f2a3f6a3
OG		 1556EXPORT_SYMBOL_GPL(mlx4_counter_free);
1557
3d73c288 1558static int mlx4_setup_hca(struct mlx4_dev *dev)
225c7b1f
RD		 1559{
1560 struct mlx4_priv *priv = mlx4_priv(dev);
1561 int err;
7ff93f8b 1562 int port;
9a5aa622 1563 __be32 ib_port_default_caps;
225c7b1f 1564
225c7b1f
RD		 1565 err = mlx4_init_uar_table(dev);
1566 if (err) {
1567 mlx4_err(dev, "Failed to initialize "
1568 "user access region table, aborting.\n");
1569 return err;
1570 }
1571
1572 err = mlx4_uar_alloc(dev, &priv->driver_uar);
1573 if (err) {
1574 mlx4_err(dev, "Failed to allocate driver access region, "
1575 "aborting.\n");
1576 goto err_uar_table_free;
1577 }
1578
4979d18f 1579 priv->kar = ioremap((phys_addr_t) priv->driver_uar.pfn << PAGE_SHIFT, PAGE_SIZE);
225c7b1f
RD		 1580 if (!priv->kar) {
1581 mlx4_err(dev, "Couldn't map kernel access region, "
1582 "aborting.\n");
1583 err = -ENOMEM;
1584 goto err_uar_free;
1585 }
1586
1587 err = mlx4_init_pd_table(dev);
1588 if (err) {
1589 mlx4_err(dev, "Failed to initialize "
1590 "protection domain table, aborting.\n");
1591 goto err_kar_unmap;
1592 }
1593
012a8ff5
SH		 1594 err = mlx4_init_xrcd_table(dev);
1595 if (err) {
1596 mlx4_err(dev, "Failed to initialize "
1597 "reliable connection domain table, aborting.\n");
1598 goto err_pd_table_free;
1599 }
1600
225c7b1f
RD		 1601 err = mlx4_init_mr_table(dev);
1602 if (err) {
1603 mlx4_err(dev, "Failed to initialize "
1604 "memory region table, aborting.\n");
012a8ff5 1605 goto err_xrcd_table_free;
225c7b1f
RD		 1606 }
1607
225c7b1f
RD		 1608 err = mlx4_init_eq_table(dev);
1609 if (err) {
1610 mlx4_err(dev, "Failed to initialize "
1611 "event queue table, aborting.\n");
ee49bd93 1612 goto err_mr_table_free;
225c7b1f
RD		 1613 }
1614
1615 err = mlx4_cmd_use_events(dev);
1616 if (err) {
1617 mlx4_err(dev, "Failed to switch to event-driven "
1618 "firmware commands, aborting.\n");
1619 goto err_eq_table_free;
1620 }
1621
1622 err = mlx4_NOP(dev);
1623 if (err) {
08fb1055
MT		 1624 if (dev->flags & MLX4_FLAG_MSI_X) {
1625 mlx4_warn(dev, "NOP command failed to generate MSI-X "
1626 "interrupt IRQ %d).\n",
b8dd786f 1627 priv->eq_table.eq[dev->caps.num_comp_vectors].irq);
08fb1055
MT		 1628 mlx4_warn(dev, "Trying again without MSI-X.\n");
1629 } else {
1630 mlx4_err(dev, "NOP command failed to generate interrupt "
1631 "(IRQ %d), aborting.\n",
b8dd786f 1632 priv->eq_table.eq[dev->caps.num_comp_vectors].irq);
225c7b1f 1633 mlx4_err(dev, "BIOS or ACPI interrupt routing problem?\n");
08fb1055 1634 }
225c7b1f
RD		 1635
1636 goto err_cmd_poll;
1637 }
1638
1639 mlx4_dbg(dev, "NOP command IRQ test passed\n");
1640
1641 err = mlx4_init_cq_table(dev);
1642 if (err) {
1643 mlx4_err(dev, "Failed to initialize "
1644 "completion queue table, aborting.\n");
1645 goto err_cmd_poll;
1646 }
1647
1648 err = mlx4_init_srq_table(dev);
1649 if (err) {
1650 mlx4_err(dev, "Failed to initialize "
1651 "shared receive queue table, aborting.\n");
1652 goto err_cq_table_free;
1653 }
1654
1655 err = mlx4_init_qp_table(dev);
1656 if (err) {
1657 mlx4_err(dev, "Failed to initialize "
1658 "queue pair table, aborting.\n");
1659 goto err_srq_table_free;
1660 }
1661
ab9c17a0
JM		 1662 if (!mlx4_is_slave(dev)) {
1663 err = mlx4_init_mcg_table(dev);
1664 if (err) {
1665 mlx4_err(dev, "Failed to initialize "
1666 "multicast group table, aborting.\n");
1667 goto err_qp_table_free;
1668 }
225c7b1f
RD		 1669 }
1670
f2a3f6a3
OG		 1671 err = mlx4_init_counters_table(dev);
1672 if (err && err != -ENOENT) {
1673 mlx4_err(dev, "Failed to initialize counters table, aborting.\n");
ab9c17a0 1674 goto err_mcg_table_free;
f2a3f6a3
OG		 1675 }
1676
ab9c17a0
JM		 1677 if (!mlx4_is_slave(dev)) {
1678 for (port = 1; port <= dev->caps.num_ports; port++) {
ab9c17a0
JM		 1679 ib_port_default_caps = 0;
1680 err = mlx4_get_port_ib_caps(dev, port,
1681 &ib_port_default_caps);
1682 if (err)
1683 mlx4_warn(dev, "failed to get port %d default "
1684 "ib capabilities (%d). Continuing "
1685 "with caps = 0\n", port, err);
1686 dev->caps.ib_port_def_cap[port] = ib_port_default_caps;
1687
2aca1172
JM		 1688 /* initialize per-slave default ib port capabilities */
1689 if (mlx4_is_master(dev)) {
1690 int i;
1691 for (i = 0; i < dev->num_slaves; i++) {
1692 if (i == mlx4_master_func_num(dev))
1693 continue;
1694 priv->mfunc.master.slave_state[i].ib_cap_mask[port] =
1695 ib_port_default_caps;
1696 }
1697 }
1698
096335b3
OG		 1699 if (mlx4_is_mfunc(dev))
1700 dev->caps.port_ib_mtu[port] = IB_MTU_2048;
1701 else
1702 dev->caps.port_ib_mtu[port] = IB_MTU_4096;
97285b78 1703
6634961c
JM		 1704 err = mlx4_SET_PORT(dev, port, mlx4_is_master(dev) ?
1705 dev->caps.pkey_table_len[port] : -1);
ab9c17a0
JM		 1706 if (err) {
1707 mlx4_err(dev, "Failed to set port %d, aborting\n",
1708 port);
1709 goto err_counters_table_free;
1710 }
7ff93f8b
YP		 1711 }
1712 }
1713
225c7b1f
RD		 1714 return 0;
1715
f2a3f6a3
OG		 1716err_counters_table_free:
1717 mlx4_cleanup_counters_table(dev);
1718
ab9c17a0
JM		 1719err_mcg_table_free:
1720 mlx4_cleanup_mcg_table(dev);
1721
225c7b1f
RD		 1722err_qp_table_free:
1723 mlx4_cleanup_qp_table(dev);
1724
1725err_srq_table_free:
1726 mlx4_cleanup_srq_table(dev);
1727
1728err_cq_table_free:
1729 mlx4_cleanup_cq_table(dev);
1730
1731err_cmd_poll:
1732 mlx4_cmd_use_polling(dev);
1733
1734err_eq_table_free:
1735 mlx4_cleanup_eq_table(dev);
1736
ee49bd93 1737err_mr_table_free:
225c7b1f
RD		 1738 mlx4_cleanup_mr_table(dev);
1739
012a8ff5
SH		 1740err_xrcd_table_free:
1741 mlx4_cleanup_xrcd_table(dev);
1742
225c7b1f
RD		 1743err_pd_table_free:
1744 mlx4_cleanup_pd_table(dev);
1745
1746err_kar_unmap:
1747 iounmap(priv->kar);
1748
1749err_uar_free:
1750 mlx4_uar_free(dev, &priv->driver_uar);
1751
1752err_uar_table_free:
1753 mlx4_cleanup_uar_table(dev);
1754 return err;
1755}
1756
e8f9b2ed 1757static void mlx4_enable_msi_x(struct mlx4_dev *dev)
225c7b1f
RD		 1758{
1759 struct mlx4_priv *priv = mlx4_priv(dev);
b8dd786f 1760 struct msix_entry *entries;
0b7ca5a9 1761 int nreq = min_t(int, dev->caps.num_ports *
90b1ebe7
YM		 1762 min_t(int, netif_get_num_default_rss_queues() + 1,
1763 MAX_MSIX_P_PORT) + MSIX_LEGACY_SZ, MAX_MSIX);
225c7b1f
RD		 1764 int err;
1765 int i;
1766
1767 if (msi_x) {
ab9c17a0
JM		 1768 /* In multifunction mode each function gets 2 msi-X vectors
1769 * one for data path completions anf the other for asynch events
1770 * or command completions */
1771 if (mlx4_is_mfunc(dev)) {
1772 nreq = 2;
1773 } else {
1774 nreq = min_t(int, dev->caps.num_eqs -
1775 dev->caps.reserved_eqs, nreq);
1776 }
1777
b8dd786f
YP		 1778 entries = kcalloc(nreq, sizeof *entries, GFP_KERNEL);
1779 if (!entries)
1780 goto no_msi;
1781
1782 for (i = 0; i < nreq; ++i)
225c7b1f
RD		 1783 entries[i].entry = i;
1784
b8dd786f
YP		 1785 retry:
1786 err = pci_enable_msix(dev->pdev, entries, nreq);
225c7b1f 1787 if (err) {
b8dd786f
YP		 1788 /* Try again if at least 2 vectors are available */
1789 if (err > 1) {
1790 mlx4_info(dev, "Requested %d vectors, "
1791 "but only %d MSI-X vectors available, "
1792 "trying again\n", nreq, err);
1793 nreq = err;
1794 goto retry;
1795 }
5bf0da7d 1796 kfree(entries);
225c7b1f
RD		 1797 goto no_msi;
1798 }
1799
0b7ca5a9
YP		 1800 if (nreq <
1801 MSIX_LEGACY_SZ + dev->caps.num_ports * MIN_MSIX_P_PORT) {
1802 /*Working in legacy mode , all EQ's shared*/
1803 dev->caps.comp_pool = 0;
1804 dev->caps.num_comp_vectors = nreq - 1;
1805 } else {
1806 dev->caps.comp_pool = nreq - MSIX_LEGACY_SZ;
1807 dev->caps.num_comp_vectors = MSIX_LEGACY_SZ - 1;
1808 }
b8dd786f 1809 for (i = 0; i < nreq; ++i)
225c7b1f
RD		 1810 priv->eq_table.eq[i].irq = entries[i].vector;
1811
1812 dev->flags |= MLX4_FLAG_MSI_X;
b8dd786f
YP		 1813
1814 kfree(entries);
225c7b1f
RD		 1815 return;
1816 }
1817
1818no_msi:
b8dd786f 1819 dev->caps.num_comp_vectors = 1;
0b7ca5a9 1820 dev->caps.comp_pool = 0;
b8dd786f
YP		 1821
1822 for (i = 0; i < 2; ++i)
225c7b1f
RD		 1823 priv->eq_table.eq[i].irq = dev->pdev->irq;
1824}
1825
7ff93f8b 1826static int mlx4_init_port_info(struct mlx4_dev *dev, int port)
2a2336f8
YP		 1827{
1828 struct mlx4_port_info *info = &mlx4_priv(dev)->port[port];
7ff93f8b 1829 int err = 0;
2a2336f8
YP		 1830
1831 info->dev = dev;
1832 info->port = port;
ab9c17a0
JM		 1833 if (!mlx4_is_slave(dev)) {
1834 INIT_RADIX_TREE(&info->mac_tree, GFP_KERNEL);
1835 mlx4_init_mac_table(dev, &info->mac_table);
1836 mlx4_init_vlan_table(dev, &info->vlan_table);
1837 info->base_qpn =
1838 dev->caps.reserved_qps_base[MLX4_QP_REGION_ETH_ADDR] +
06fa0a88 1839 (port - 1) * (1 << log_num_mac);
ab9c17a0 1840 }
7ff93f8b
YP		 1841
1842 sprintf(info->dev_name, "mlx4_port%d", port);
1843 info->port_attr.attr.name = info->dev_name;
ab9c17a0
JM		 1844 if (mlx4_is_mfunc(dev))
1845 info->port_attr.attr.mode = S_IRUGO;
1846 else {
1847 info->port_attr.attr.mode = S_IRUGO | S_IWUSR;
1848 info->port_attr.store = set_port_type;
1849 }
7ff93f8b 1850 info->port_attr.show = show_port_type;
3691c964 1851 sysfs_attr_init(&info->port_attr.attr);
7ff93f8b
YP		 1852
1853 err = device_create_file(&dev->pdev->dev, &info->port_attr);
1854 if (err) {
1855 mlx4_err(dev, "Failed to create file for port %d\n", port);
1856 info->port = -1;
1857 }
1858
096335b3
OG		 1859 sprintf(info->dev_mtu_name, "mlx4_port%d_mtu", port);
1860 info->port_mtu_attr.attr.name = info->dev_mtu_name;
1861 if (mlx4_is_mfunc(dev))
1862 info->port_mtu_attr.attr.mode = S_IRUGO;
1863 else {
1864 info->port_mtu_attr.attr.mode = S_IRUGO | S_IWUSR;
1865 info->port_mtu_attr.store = set_port_ib_mtu;
1866 }
1867 info->port_mtu_attr.show = show_port_ib_mtu;
1868 sysfs_attr_init(&info->port_mtu_attr.attr);
1869
1870 err = device_create_file(&dev->pdev->dev, &info->port_mtu_attr);
1871 if (err) {
1872 mlx4_err(dev, "Failed to create mtu file for port %d\n", port);
1873 device_remove_file(&info->dev->pdev->dev, &info->port_attr);
1874 info->port = -1;
1875 }
1876
7ff93f8b
YP		 1877 return err;
1878}
1879
1880static void mlx4_cleanup_port_info(struct mlx4_port_info *info)
1881{
1882 if (info->port < 0)
1883 return;
1884
1885 device_remove_file(&info->dev->pdev->dev, &info->port_attr);
096335b3 1886 device_remove_file(&info->dev->pdev->dev, &info->port_mtu_attr);
2a2336f8
YP		 1887}
1888
b12d93d6
YP		 1889static int mlx4_init_steering(struct mlx4_dev *dev)
1890{
1891 struct mlx4_priv *priv = mlx4_priv(dev);
1892 int num_entries = dev->caps.num_ports;
1893 int i, j;
1894
1895 priv->steer = kzalloc(sizeof(struct mlx4_steer) * num_entries, GFP_KERNEL);
1896 if (!priv->steer)
1897 return -ENOMEM;
1898
45b51365 1899 for (i = 0; i < num_entries; i++)
b12d93d6
YP		 1900 for (j = 0; j < MLX4_NUM_STEERS; j++) {
1901 INIT_LIST_HEAD(&priv->steer[i].promisc_qps[j]);
1902 INIT_LIST_HEAD(&priv->steer[i].steer_entries[j]);
1903 }
b12d93d6
YP		 1904 return 0;
1905}
1906
1907static void mlx4_clear_steering(struct mlx4_dev *dev)
1908{
1909 struct mlx4_priv *priv = mlx4_priv(dev);
1910 struct mlx4_steer_index *entry, *tmp_entry;
1911 struct mlx4_promisc_qp *pqp, *tmp_pqp;
1912 int num_entries = dev->caps.num_ports;
1913 int i, j;
1914
1915 for (i = 0; i < num_entries; i++) {
1916 for (j = 0; j < MLX4_NUM_STEERS; j++) {
1917 list_for_each_entry_safe(pqp, tmp_pqp,
1918 &priv->steer[i].promisc_qps[j],
1919 list) {
1920 list_del(&pqp->list);
1921 kfree(pqp);
1922 }
1923 list_for_each_entry_safe(entry, tmp_entry,
1924 &priv->steer[i].steer_entries[j],
1925 list) {
1926 list_del(&entry->list);
1927 list_for_each_entry_safe(pqp, tmp_pqp,
1928 &entry->duplicates,
1929 list) {
1930 list_del(&pqp->list);
1931 kfree(pqp);
1932 }
1933 kfree(entry);
1934 }
1935 }
1936 }
1937 kfree(priv->steer);
1938}
1939
ab9c17a0
JM		 1940static int extended_func_num(struct pci_dev *pdev)
1941{
1942 return PCI_SLOT(pdev->devfn) * 8 + PCI_FUNC(pdev->devfn);
1943}
1944
1945#define MLX4_OWNER_BASE 0x8069c
1946#define MLX4_OWNER_SIZE 4
1947
1948static int mlx4_get_ownership(struct mlx4_dev *dev)
1949{
1950 void __iomem *owner;
1951 u32 ret;
1952
57dbf29a
KSS		 1953 if (pci_channel_offline(dev->pdev))
1954 return -EIO;
1955
ab9c17a0
JM		 1956 owner = ioremap(pci_resource_start(dev->pdev, 0) + MLX4_OWNER_BASE,
1957 MLX4_OWNER_SIZE);
1958 if (!owner) {
1959 mlx4_err(dev, "Failed to obtain ownership bit\n");
1960 return -ENOMEM;
1961 }
1962
1963 ret = readl(owner);
1964 iounmap(owner);
1965 return (int) !!ret;
1966}
1967
1968static void mlx4_free_ownership(struct mlx4_dev *dev)
1969{
1970 void __iomem *owner;
1971
57dbf29a
KSS		 1972 if (pci_channel_offline(dev->pdev))
1973 return;
1974
ab9c17a0
JM		 1975 owner = ioremap(pci_resource_start(dev->pdev, 0) + MLX4_OWNER_BASE,
1976 MLX4_OWNER_SIZE);
1977 if (!owner) {
1978 mlx4_err(dev, "Failed to obtain ownership bit\n");
1979 return;
1980 }
1981 writel(0, owner);
1982 msleep(1000);
1983 iounmap(owner);
1984}
1985
839f1243 1986static int __mlx4_init_one(struct pci_dev *pdev, int pci_dev_data)
225c7b1f 1987{
225c7b1f
RD		 1988 struct mlx4_priv *priv;
1989 struct mlx4_dev *dev;
1990 int err;
2a2336f8 1991 int port;
225c7b1f 1992
0a645e80 1993 pr_info(DRV_NAME ": Initializing %s\n", pci_name(pdev));
225c7b1f
RD		 1994
1995 err = pci_enable_device(pdev);
1996 if (err) {
1997 dev_err(&pdev->dev, "Cannot enable PCI device, "
1998 "aborting.\n");
1999 return err;
2000 }
ab9c17a0
JM		 2001 if (num_vfs > MLX4_MAX_NUM_VF) {
2002 printk(KERN_ERR "There are more VF's (%d) than allowed(%d)\n",
2003 num_vfs, MLX4_MAX_NUM_VF);
2004 return -EINVAL;
2005 }
225c7b1f 2006 /*
ab9c17a0 2007 * Check for BARs.
225c7b1f 2008 */
839f1243 2009 if (!(pci_dev_data & MLX4_PCI_DEV_IS_VF) &&
ab9c17a0
JM		 2010 !(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) {
2011 dev_err(&pdev->dev, "Missing DCS, aborting."
839f1243
RD		 2012 "(driver_data: 0x%x, pci_resource_flags(pdev, 0):0x%lx)\n",
2013 pci_dev_data, pci_resource_flags(pdev, 0));
225c7b1f
RD		 2014 err = -ENODEV;
2015 goto err_disable_pdev;
2016 }
2017 if (!(pci_resource_flags(pdev, 2) & IORESOURCE_MEM)) {
2018 dev_err(&pdev->dev, "Missing UAR, aborting.\n");
2019 err = -ENODEV;
2020 goto err_disable_pdev;
2021 }
2022
a01df0fe 2023 err = pci_request_regions(pdev, DRV_NAME);
225c7b1f 2024 if (err) {
a01df0fe 2025 dev_err(&pdev->dev, "Couldn't get PCI resources, aborting\n");
225c7b1f
RD		 2026 goto err_disable_pdev;
2027 }
2028
225c7b1f
RD		 2029 pci_set_master(pdev);
2030
6a35528a 2031 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
225c7b1f
RD		 2032 if (err) {
2033 dev_warn(&pdev->dev, "Warning: couldn't set 64-bit PCI DMA mask.\n");
284901a9 2034 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
225c7b1f
RD		 2035 if (err) {
2036 dev_err(&pdev->dev, "Can't set PCI DMA mask, aborting.\n");
a01df0fe 2037 goto err_release_regions;
225c7b1f
RD		 2038 }
2039 }
6a35528a 2040 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
225c7b1f
RD		 2041 if (err) {
2042 dev_warn(&pdev->dev, "Warning: couldn't set 64-bit "
2043 "consistent PCI DMA mask.\n");
284901a9 2044 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
225c7b1f
RD		 2045 if (err) {
2046 dev_err(&pdev->dev, "Can't set consistent PCI DMA mask, "
2047 "aborting.\n");
a01df0fe 2048 goto err_release_regions;
225c7b1f
RD		 2049 }
2050 }
2051
7f9e5c48
DD		 2052 /* Allow large DMA segments, up to the firmware limit of 1 GB */
2053 dma_set_max_seg_size(&pdev->dev, 1024 * 1024 * 1024);
2054
225c7b1f
RD		 2055 priv = kzalloc(sizeof *priv, GFP_KERNEL);
2056 if (!priv) {
2057 dev_err(&pdev->dev, "Device struct alloc failed, "
2058 "aborting.\n");
2059 err = -ENOMEM;
a01df0fe 2060 goto err_release_regions;
225c7b1f
RD		 2061 }
2062
2063 dev = &priv->dev;
2064 dev->pdev = pdev;
b581401e
RD		 2065 INIT_LIST_HEAD(&priv->ctx_list);
2066 spin_lock_init(&priv->ctx_lock);
225c7b1f 2067
7ff93f8b
YP		 2068 mutex_init(&priv->port_mutex);
2069
6296883c
YP		 2070 INIT_LIST_HEAD(&priv->pgdir_list);
2071 mutex_init(&priv->pgdir_mutex);
2072
c1b43dca
EC		 2073 INIT_LIST_HEAD(&priv->bf_list);
2074 mutex_init(&priv->bf_mutex);
2075
aca7a3ac 2076 dev->rev_id = pdev->revision;
ab9c17a0 2077 /* Detect if this device is a virtual function */
839f1243 2078 if (pci_dev_data & MLX4_PCI_DEV_IS_VF) {
ab9c17a0
JM		 2079 /* When acting as pf, we normally skip vfs unless explicitly
2080 * requested to probe them. */
2081 if (num_vfs && extended_func_num(pdev) > probe_vf) {
2082 mlx4_warn(dev, "Skipping virtual function:%d\n",
2083 extended_func_num(pdev));
2084 err = -ENODEV;
2085 goto err_free_dev;
2086 }
2087 mlx4_warn(dev, "Detected virtual function - running in slave mode\n");
2088 dev->flags |= MLX4_FLAG_SLAVE;
2089 } else {
2090 /* We reset the device and enable SRIOV only for physical
2091 * devices. Try to claim ownership on the device;
2092 * if already taken, skip -- do not allow multiple PFs */
2093 err = mlx4_get_ownership(dev);
2094 if (err) {
2095 if (err < 0)
2096 goto err_free_dev;
2097 else {
2098 mlx4_warn(dev, "Multiple PFs not yet supported."
2099 " Skipping PF.\n");
2100 err = -EINVAL;
2101 goto err_free_dev;
2102 }
2103 }
aca7a3ac 2104
ab9c17a0 2105 if (num_vfs) {
84b1f153 2106 mlx4_warn(dev, "Enabling SR-IOV with %d VFs\n", num_vfs);
ab9c17a0
JM		 2107 err = pci_enable_sriov(pdev, num_vfs);
2108 if (err) {
84b1f153
RD		 2109 mlx4_err(dev, "Failed to enable SR-IOV, continuing without SR-IOV (err = %d).\n",
2110 err);
ab9c17a0
JM		 2111 err = 0;
2112 } else {
2113 mlx4_warn(dev, "Running in master mode\n");
2114 dev->flags |= MLX4_FLAG_SRIOV |
2115 MLX4_FLAG_MASTER;
2116 dev->num_vfs = num_vfs;
2117 }
2118 }
2119
2120 /*
2121 * Now reset the HCA before we touch the PCI capabilities or
2122 * attempt a firmware command, since a boot ROM may have left
2123 * the HCA in an undefined state.
2124 */
2125 err = mlx4_reset(dev);
2126 if (err) {
2127 mlx4_err(dev, "Failed to reset HCA, aborting.\n");
2128 goto err_rel_own;
2129 }
225c7b1f
RD		 2130 }
2131
ab9c17a0 2132slave_start:
521130d1
EE		 2133 err = mlx4_cmd_init(dev);
2134 if (err) {
225c7b1f 2135 mlx4_err(dev, "Failed to init command interface, aborting.\n");
ab9c17a0
JM		 2136 goto err_sriov;
2137 }
2138
2139 /* In slave functions, the communication channel must be initialized
2140 * before posting commands. Also, init num_slaves before calling
2141 * mlx4_init_hca */
2142 if (mlx4_is_mfunc(dev)) {
2143 if (mlx4_is_master(dev))
2144 dev->num_slaves = MLX4_MAX_NUM_SLAVES;
2145 else {
2146 dev->num_slaves = 0;
2147 if (mlx4_multi_func_init(dev)) {
2148 mlx4_err(dev, "Failed to init slave mfunc"
2149 " interface, aborting.\n");
2150 goto err_cmd;
2151 }
2152 }
225c7b1f
RD		 2153 }
2154
2155 err = mlx4_init_hca(dev);
ab9c17a0
JM		 2156 if (err) {
2157 if (err == -EACCES) {
2158 /* Not primary Physical function
2159 * Running in slave mode */
2160 mlx4_cmd_cleanup(dev);
2161 dev->flags |= MLX4_FLAG_SLAVE;
2162 dev->flags &= ~MLX4_FLAG_MASTER;
2163 goto slave_start;
2164 } else
2165 goto err_mfunc;
2166 }
2167
2168 /* In master functions, the communication channel must be initialized
2169 * after obtaining its address from fw */
2170 if (mlx4_is_master(dev)) {
2171 if (mlx4_multi_func_init(dev)) {
2172 mlx4_err(dev, "Failed to init master mfunc"
2173 "interface, aborting.\n");
2174 goto err_close;
2175 }
2176 }
225c7b1f 2177
b8dd786f
YP		 2178 err = mlx4_alloc_eq_table(dev);
2179 if (err)
ab9c17a0 2180 goto err_master_mfunc;
b8dd786f 2181
0b7ca5a9 2182 priv->msix_ctl.pool_bm = 0;
730c41d5 2183 mutex_init(&priv->msix_ctl.pool_lock);
0b7ca5a9 2184
08fb1055 2185 mlx4_enable_msi_x(dev);
ab9c17a0
JM		 2186 if ((mlx4_is_mfunc(dev)) &&
2187 !(dev->flags & MLX4_FLAG_MSI_X)) {
2188 mlx4_err(dev, "INTx is not supported in multi-function mode."
2189 " aborting.\n");
b12d93d6 2190 goto err_free_eq;
ab9c17a0
JM		 2191 }
2192
2193 if (!mlx4_is_slave(dev)) {
2194 err = mlx4_init_steering(dev);
2195 if (err)
2196 goto err_free_eq;
2197 }
b12d93d6 2198
225c7b1f 2199 err = mlx4_setup_hca(dev);
ab9c17a0
JM		 2200 if (err == -EBUSY && (dev->flags & MLX4_FLAG_MSI_X) &&
2201 !mlx4_is_mfunc(dev)) {
08fb1055 2202 dev->flags &= ~MLX4_FLAG_MSI_X;
9858d2d1
YP		 2203 dev->caps.num_comp_vectors = 1;
2204 dev->caps.comp_pool = 0;
08fb1055
MT		 2205 pci_disable_msix(pdev);
2206 err = mlx4_setup_hca(dev);
2207 }
2208
225c7b1f 2209 if (err)
b12d93d6 2210 goto err_steer;
225c7b1f 2211
7ff93f8b
YP		 2212 for (port = 1; port <= dev->caps.num_ports; port++) {
2213 err = mlx4_init_port_info(dev, port);
2214 if (err)
2215 goto err_port;
2216 }
2a2336f8 2217
225c7b1f
RD		 2218 err = mlx4_register_device(dev);
2219 if (err)
7ff93f8b 2220 goto err_port;
225c7b1f 2221
27bf91d6
YP		 2222 mlx4_sense_init(dev);
2223 mlx4_start_sense(dev);
2224
839f1243 2225 priv->pci_dev_data = pci_dev_data;
225c7b1f
RD		 2226 pci_set_drvdata(pdev, dev);
2227
2228 return 0;
2229
7ff93f8b 2230err_port:
b4f77264 2231 for (--port; port >= 1; --port)
7ff93f8b
YP		 2232 mlx4_cleanup_port_info(&priv->port[port]);
2233
f2a3f6a3 2234 mlx4_cleanup_counters_table(dev);
225c7b1f
RD		 2235 mlx4_cleanup_mcg_table(dev);
2236 mlx4_cleanup_qp_table(dev);
2237 mlx4_cleanup_srq_table(dev);
2238 mlx4_cleanup_cq_table(dev);
2239 mlx4_cmd_use_polling(dev);
2240 mlx4_cleanup_eq_table(dev);
225c7b1f 2241 mlx4_cleanup_mr_table(dev);
012a8ff5 2242 mlx4_cleanup_xrcd_table(dev);
225c7b1f
RD		 2243 mlx4_cleanup_pd_table(dev);
2244 mlx4_cleanup_uar_table(dev);
2245
b12d93d6 2246err_steer:
ab9c17a0
JM		 2247 if (!mlx4_is_slave(dev))
2248 mlx4_clear_steering(dev);
b12d93d6 2249
b8dd786f
YP		 2250err_free_eq:
2251 mlx4_free_eq_table(dev);
2252
ab9c17a0
JM		 2253err_master_mfunc:
2254 if (mlx4_is_master(dev))
2255 mlx4_multi_func_cleanup(dev);
2256
225c7b1f 2257err_close:
08fb1055
MT		 2258 if (dev->flags & MLX4_FLAG_MSI_X)
2259 pci_disable_msix(pdev);
2260
225c7b1f
RD		 2261 mlx4_close_hca(dev);
2262
ab9c17a0
JM		 2263err_mfunc:
2264 if (mlx4_is_slave(dev))
2265 mlx4_multi_func_cleanup(dev);
2266
225c7b1f
RD		 2267err_cmd:
2268 mlx4_cmd_cleanup(dev);
2269
ab9c17a0 2270err_sriov:
681372a7 2271 if (dev->flags & MLX4_FLAG_SRIOV)
ab9c17a0
JM		 2272 pci_disable_sriov(pdev);
2273
2274err_rel_own:
2275 if (!mlx4_is_slave(dev))
2276 mlx4_free_ownership(dev);
2277
225c7b1f 2278err_free_dev:
225c7b1f
RD		 2279 kfree(priv);
2280
a01df0fe
RD		 2281err_release_regions:
2282 pci_release_regions(pdev);
225c7b1f
RD		 2283
2284err_disable_pdev:
2285 pci_disable_device(pdev);
2286 pci_set_drvdata(pdev, NULL);
2287 return err;
2288}
2289
1dd06ae8 2290static int mlx4_init_one(struct pci_dev *pdev, const struct pci_device_id *id)
3d73c288 2291{
0a645e80 2292 printk_once(KERN_INFO "%s", mlx4_version);
3d73c288 2293
839f1243 2294 return __mlx4_init_one(pdev, id->driver_data);
3d73c288
RD		 2295}
2296
2297static void mlx4_remove_one(struct pci_dev *pdev)
225c7b1f
RD		 2298{
2299 struct mlx4_dev *dev = pci_get_drvdata(pdev);
2300 struct mlx4_priv *priv = mlx4_priv(dev);
2301 int p;
2302
2303 if (dev) {
ab9c17a0
JM		 2304 /* in SRIOV it is not allowed to unload the pf's
2305 * driver while there are alive vf's */
2306 if (mlx4_is_master(dev)) {
2307 if (mlx4_how_many_lives_vf(dev))
2308 printk(KERN_ERR "Removing PF when there are assigned VF's !!!\n");
2309 }
27bf91d6 2310 mlx4_stop_sense(dev);
225c7b1f
RD		 2311 mlx4_unregister_device(dev);
2312
7ff93f8b
YP		 2313 for (p = 1; p <= dev->caps.num_ports; p++) {
2314 mlx4_cleanup_port_info(&priv->port[p]);
225c7b1f 2315 mlx4_CLOSE_PORT(dev, p);
7ff93f8b 2316 }
225c7b1f 2317
b8924951
JM		 2318 if (mlx4_is_master(dev))
2319 mlx4_free_resource_tracker(dev,
2320 RES_TR_FREE_SLAVES_ONLY);
2321
f2a3f6a3 2322 mlx4_cleanup_counters_table(dev);
225c7b1f
RD		 2323 mlx4_cleanup_mcg_table(dev);
2324 mlx4_cleanup_qp_table(dev);
2325 mlx4_cleanup_srq_table(dev);
2326 mlx4_cleanup_cq_table(dev);
2327 mlx4_cmd_use_polling(dev);
2328 mlx4_cleanup_eq_table(dev);
225c7b1f 2329 mlx4_cleanup_mr_table(dev);
012a8ff5 2330 mlx4_cleanup_xrcd_table(dev);
225c7b1f
RD		 2331 mlx4_cleanup_pd_table(dev);
2332
ab9c17a0 2333 if (mlx4_is_master(dev))
b8924951
JM		 2334 mlx4_free_resource_tracker(dev,
2335 RES_TR_FREE_STRUCTS_ONLY);
ab9c17a0 2336
225c7b1f
RD		 2337 iounmap(priv->kar);
2338 mlx4_uar_free(dev, &priv->driver_uar);
2339 mlx4_cleanup_uar_table(dev);
ab9c17a0
JM		 2340 if (!mlx4_is_slave(dev))
2341 mlx4_clear_steering(dev);
b8dd786f 2342 mlx4_free_eq_table(dev);
ab9c17a0
JM		 2343 if (mlx4_is_master(dev))
2344 mlx4_multi_func_cleanup(dev);
225c7b1f 2345 mlx4_close_hca(dev);
ab9c17a0
JM		 2346 if (mlx4_is_slave(dev))
2347 mlx4_multi_func_cleanup(dev);
225c7b1f
RD		 2348 mlx4_cmd_cleanup(dev);
2349
2350 if (dev->flags & MLX4_FLAG_MSI_X)
2351 pci_disable_msix(pdev);
681372a7 2352 if (dev->flags & MLX4_FLAG_SRIOV) {
84b1f153 2353 mlx4_warn(dev, "Disabling SR-IOV\n");
ab9c17a0
JM		 2354 pci_disable_sriov(pdev);
2355 }
225c7b1f 2356
ab9c17a0
JM		 2357 if (!mlx4_is_slave(dev))
2358 mlx4_free_ownership(dev);
47605df9
JM		 2359
2360 kfree(dev->caps.qp0_tunnel);
2361 kfree(dev->caps.qp0_proxy);
2362 kfree(dev->caps.qp1_tunnel);
2363 kfree(dev->caps.qp1_proxy);
2364
225c7b1f 2365 kfree(priv);
a01df0fe 2366 pci_release_regions(pdev);
225c7b1f
RD		 2367 pci_disable_device(pdev);
2368 pci_set_drvdata(pdev, NULL);
2369 }
2370}
2371
ee49bd93
JM		 2372int mlx4_restart_one(struct pci_dev *pdev)
2373{
839f1243
RD		 2374 struct mlx4_dev *dev = pci_get_drvdata(pdev);
2375 struct mlx4_priv *priv = mlx4_priv(dev);
2376 int pci_dev_data;
2377
2378 pci_dev_data = priv->pci_dev_data;
ee49bd93 2379 mlx4_remove_one(pdev);
839f1243 2380 return __mlx4_init_one(pdev, pci_dev_data);
ee49bd93
JM		 2381}
2382
a3aa1884 2383static DEFINE_PCI_DEVICE_TABLE(mlx4_pci_table) = {
ab9c17a0 2384 /* MT25408 "Hermon" SDR */
ca3e57a5 2385 { PCI_VDEVICE(MELLANOX, 0x6340), MLX4_PCI_DEV_FORCE_SENSE_PORT },
ab9c17a0 2386 /* MT25408 "Hermon" DDR */
ca3e57a5 2387 { PCI_VDEVICE(MELLANOX, 0x634a), MLX4_PCI_DEV_FORCE_SENSE_PORT },
ab9c17a0 2388 /* MT25408 "Hermon" QDR */
ca3e57a5 2389 { PCI_VDEVICE(MELLANOX, 0x6354), MLX4_PCI_DEV_FORCE_SENSE_PORT },
ab9c17a0 2390 /* MT25408 "Hermon" DDR PCIe gen2 */
ca3e57a5 2391 { PCI_VDEVICE(MELLANOX, 0x6732), MLX4_PCI_DEV_FORCE_SENSE_PORT },
ab9c17a0 2392 /* MT25408 "Hermon" QDR PCIe gen2 */
ca3e57a5 2393 { PCI_VDEVICE(MELLANOX, 0x673c), MLX4_PCI_DEV_FORCE_SENSE_PORT },
ab9c17a0 2394 /* MT25408 "Hermon" EN 10GigE */
ca3e57a5 2395 { PCI_VDEVICE(MELLANOX, 0x6368), MLX4_PCI_DEV_FORCE_SENSE_PORT },
ab9c17a0 2396 /* MT25408 "Hermon" EN 10GigE PCIe gen2 */
ca3e57a5 2397 { PCI_VDEVICE(MELLANOX, 0x6750), MLX4_PCI_DEV_FORCE_SENSE_PORT },
ab9c17a0 2398 /* MT25458 ConnectX EN 10GBASE-T 10GigE */
ca3e57a5 2399 { PCI_VDEVICE(MELLANOX, 0x6372), MLX4_PCI_DEV_FORCE_SENSE_PORT },
ab9c17a0 2400 /* MT25458 ConnectX EN 10GBASE-T+Gen2 10GigE */
ca3e57a5 2401 { PCI_VDEVICE(MELLANOX, 0x675a), MLX4_PCI_DEV_FORCE_SENSE_PORT },
ab9c17a0 2402 /* MT26468 ConnectX EN 10GigE PCIe gen2*/
ca3e57a5 2403 { PCI_VDEVICE(MELLANOX, 0x6764), MLX4_PCI_DEV_FORCE_SENSE_PORT },
ab9c17a0 2404 /* MT26438 ConnectX EN 40GigE PCIe gen2 5GT/s */
ca3e57a5 2405 { PCI_VDEVICE(MELLANOX, 0x6746), MLX4_PCI_DEV_FORCE_SENSE_PORT },
ab9c17a0 2406 /* MT26478 ConnectX2 40GigE PCIe gen2 */
ca3e57a5 2407 { PCI_VDEVICE(MELLANOX, 0x676e), MLX4_PCI_DEV_FORCE_SENSE_PORT },
ab9c17a0 2408 /* MT25400 Family [ConnectX-2 Virtual Function] */
839f1243 2409 { PCI_VDEVICE(MELLANOX, 0x1002), MLX4_PCI_DEV_IS_VF },
ab9c17a0
JM		 2410 /* MT27500 Family [ConnectX-3] */
2411 { PCI_VDEVICE(MELLANOX, 0x1003), 0 },
2412 /* MT27500 Family [ConnectX-3 Virtual Function] */
839f1243 2413 { PCI_VDEVICE(MELLANOX, 0x1004), MLX4_PCI_DEV_IS_VF },
ab9c17a0
JM		 2414 { PCI_VDEVICE(MELLANOX, 0x1005), 0 }, /* MT27510 Family */
2415 { PCI_VDEVICE(MELLANOX, 0x1006), 0 }, /* MT27511 Family */
2416 { PCI_VDEVICE(MELLANOX, 0x1007), 0 }, /* MT27520 Family */
2417 { PCI_VDEVICE(MELLANOX, 0x1008), 0 }, /* MT27521 Family */
2418 { PCI_VDEVICE(MELLANOX, 0x1009), 0 }, /* MT27530 Family */
2419 { PCI_VDEVICE(MELLANOX, 0x100a), 0 }, /* MT27531 Family */
2420 { PCI_VDEVICE(MELLANOX, 0x100b), 0 }, /* MT27540 Family */
2421 { PCI_VDEVICE(MELLANOX, 0x100c), 0 }, /* MT27541 Family */
2422 { PCI_VDEVICE(MELLANOX, 0x100d), 0 }, /* MT27550 Family */
2423 { PCI_VDEVICE(MELLANOX, 0x100e), 0 }, /* MT27551 Family */
2424 { PCI_VDEVICE(MELLANOX, 0x100f), 0 }, /* MT27560 Family */
2425 { PCI_VDEVICE(MELLANOX, 0x1010), 0 }, /* MT27561 Family */
225c7b1f
RD		 2426 { 0, }
2427};
2428
2429MODULE_DEVICE_TABLE(pci, mlx4_pci_table);
2430
57dbf29a
KSS		 2431static pci_ers_result_t mlx4_pci_err_detected(struct pci_dev *pdev,
2432 pci_channel_state_t state)
2433{
2434 mlx4_remove_one(pdev);
2435
2436 return state == pci_channel_io_perm_failure ?
2437 PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_NEED_RESET;
2438}
2439
2440static pci_ers_result_t mlx4_pci_slot_reset(struct pci_dev *pdev)
2441{
839f1243 2442 int ret = __mlx4_init_one(pdev, 0);
57dbf29a
KSS		 2443
2444 return ret ? PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_RECOVERED;
2445}
2446
3646f0e5 2447static const struct pci_error_handlers mlx4_err_handler = {
57dbf29a
KSS		 2448 .error_detected = mlx4_pci_err_detected,
2449 .slot_reset = mlx4_pci_slot_reset,
2450};
2451
225c7b1f
RD		 2452static struct pci_driver mlx4_driver = {
2453 .name = DRV_NAME,
2454 .id_table = mlx4_pci_table,
2455 .probe = mlx4_init_one,
f57e6848 2456 .remove = mlx4_remove_one,
57dbf29a 2457 .err_handler = &mlx4_err_handler,
225c7b1f
RD		 2458};
2459
7ff93f8b
YP		 2460static int __init mlx4_verify_params(void)
2461{
2462 if ((log_num_mac < 0) || (log_num_mac > 7)) {
0a645e80 2463 pr_warning("mlx4_core: bad num_mac: %d\n", log_num_mac);
7ff93f8b
YP		 2464 return -1;
2465 }
2466
cb29688a
OG		 2467 if (log_num_vlan != 0)
2468 pr_warning("mlx4_core: log_num_vlan - obsolete module param, using %d\n",
2469 MLX4_LOG_NUM_VLANS);
7ff93f8b 2470
0498628f 2471 if ((log_mtts_per_seg < 1) || (log_mtts_per_seg > 7)) {
0a645e80 2472 pr_warning("mlx4_core: bad log_mtts_per_seg: %d\n", log_mtts_per_seg);
ab6bf42e
EC		 2473 return -1;
2474 }
2475
ab9c17a0
JM		 2476 /* Check if module param for ports type has legal combination */
2477 if (port_type_array[0] == false && port_type_array[1] == true) {
2478 printk(KERN_WARNING "Module parameter configuration ETH/IB is not supported. Switching to default configuration IB/IB\n");
2479 port_type_array[0] = true;
2480 }
2481
7ff93f8b
YP		 2482 return 0;
2483}
2484
225c7b1f
RD		 2485static int __init mlx4_init(void)
2486{
2487 int ret;
2488
7ff93f8b
YP		 2489 if (mlx4_verify_params())
2490 return -EINVAL;
2491
27bf91d6
YP		 2492 mlx4_catas_init();
2493
2494 mlx4_wq = create_singlethread_workqueue("mlx4");
2495 if (!mlx4_wq)
2496 return -ENOMEM;
ee49bd93 2497
225c7b1f
RD		 2498 ret = pci_register_driver(&mlx4_driver);
2499 return ret < 0 ? ret : 0;
2500}
2501
2502static void __exit mlx4_cleanup(void)
2503{
2504 pci_unregister_driver(&mlx4_driver);
27bf91d6 2505 destroy_workqueue(mlx4_wq);
225c7b1f
RD		 2506}
2507
2508module_init(mlx4_init);
2509module_exit(mlx4_cleanup);
Linux 6.x block layer and io_uring tree(s)