projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
bridge: netlink: fix slave_changelink/br_setport race conditions
[deliverable/linux.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>
b046ffe5 44#include <linux/kmod.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
dd41cc3b 80static uint8_t num_vfs[3] = {0, 0, 0};
effa4bc4 81static int num_vfs_argc;
dd41cc3b
MB		 82module_param_array(num_vfs, byte , &num_vfs_argc, 0444);
83MODULE_PARM_DESC(num_vfs, "enable #num_vfs functions if num_vfs > 0\n"
84 "num_vfs=port1,port2,port1+2");
85
86static uint8_t probe_vf[3] = {0, 0, 0};
effa4bc4 87static int probe_vfs_argc;
dd41cc3b
MB		 88module_param_array(probe_vf, byte, &probe_vfs_argc, 0444);
89MODULE_PARM_DESC(probe_vf, "number of vfs to probe by pf driver (num_vfs > 0)\n"
90 "probe_vf=port1,port2,port1+2");
ab9c17a0 91
3c439b55 92int mlx4_log_num_mgm_entry_size = MLX4_DEFAULT_MGM_LOG_ENTRY_SIZE;
0ec2c0f8
EE		 93module_param_named(log_num_mgm_entry_size,
94 mlx4_log_num_mgm_entry_size, int, 0444);
95MODULE_PARM_DESC(log_num_mgm_entry_size, "log mgm size, that defines the num"
96 " of qp per mcg, for example:"
3c439b55 97 " 10 gives 248.range: 7 <="
0ff1fb65 98 " log_num_mgm_entry_size <= 12."
3c439b55
JM		 99 " To activate device managed"
100 " flow steering when available, set to -1");
0ec2c0f8 101
be902ab1 102static bool enable_64b_cqe_eqe = true;
08ff3235
OG		 103module_param(enable_64b_cqe_eqe, bool, 0444);
104MODULE_PARM_DESC(enable_64b_cqe_eqe,
be902ab1 105 "Enable 64 byte CQEs/EQEs when the FW supports this (default: True)");
08ff3235 106
77507aa2 107#define PF_CONTEXT_BEHAVIOUR_MASK (MLX4_FUNC_CAP_64B_EQE_CQE | \
7d077cd3
MB		 108 MLX4_FUNC_CAP_EQE_CQE_STRIDE | \
109 MLX4_FUNC_CAP_DMFS_A0_STATIC)
ab9c17a0 110
55ad3592
YH		 111#define RESET_PERSIST_MASK_FLAGS (MLX4_FLAG_SRIOV)
112
f57e6848 113static char mlx4_version[] =
225c7b1f
RD		 114 DRV_NAME ": Mellanox ConnectX core driver v"
115 DRV_VERSION " (" DRV_RELDATE ")\n";
116
117static struct mlx4_profile default_profile = {
ab9c17a0 118 .num_qp = 1 << 18,
225c7b1f 119 .num_srq = 1 << 16,
c9f2ba5e 120 .rdmarc_per_qp = 1 << 4,
225c7b1f
RD		 121 .num_cq = 1 << 16,
122 .num_mcg = 1 << 13,
ab9c17a0 123 .num_mpt = 1 << 19,
9fd7a1e1 124 .num_mtt = 1 << 20, /* It is really num mtt segements */
225c7b1f
RD		 125};
126
2599d858
AV		 127static struct mlx4_profile low_mem_profile = {
128 .num_qp = 1 << 17,
129 .num_srq = 1 << 6,
130 .rdmarc_per_qp = 1 << 4,
131 .num_cq = 1 << 8,
132 .num_mcg = 1 << 8,
133 .num_mpt = 1 << 9,
134 .num_mtt = 1 << 7,
135};
136
ab9c17a0 137static int log_num_mac = 7;
93fc9e1b
YP		 138module_param_named(log_num_mac, log_num_mac, int, 0444);
139MODULE_PARM_DESC(log_num_mac, "Log2 max number of MACs per ETH port (1-7)");
140
141static int log_num_vlan;
142module_param_named(log_num_vlan, log_num_vlan, int, 0444);
143MODULE_PARM_DESC(log_num_vlan, "Log2 max number of VLANs per ETH port (0-7)");
cb29688a
OG		 144/* Log2 max number of VLANs per ETH port (0-7) */
145#define MLX4_LOG_NUM_VLANS 7
2599d858
AV		 146#define MLX4_MIN_LOG_NUM_VLANS 0
147#define MLX4_MIN_LOG_NUM_MAC 1
93fc9e1b 148
eb939922 149static bool use_prio;
93fc9e1b 150module_param_named(use_prio, use_prio, bool, 0444);
ecc8fb11 151MODULE_PARM_DESC(use_prio, "Enable steering by VLAN priority on ETH ports (deprecated)");
93fc9e1b 152
2b8fb286 153int log_mtts_per_seg = ilog2(MLX4_MTT_ENTRY_PER_SEG);
ab6bf42e 154module_param_named(log_mtts_per_seg, log_mtts_per_seg, int, 0444);
0498628f 155MODULE_PARM_DESC(log_mtts_per_seg, "Log2 number of MTT entries per segment (1-7)");
ab6bf42e 156
8d0fc7b6 157static int port_type_array[2] = {MLX4_PORT_TYPE_NONE, MLX4_PORT_TYPE_NONE};
ab9c17a0
JM		 158static int arr_argc = 2;
159module_param_array(port_type_array, int, &arr_argc, 0444);
8d0fc7b6
YP		 160MODULE_PARM_DESC(port_type_array, "Array of port types: HW_DEFAULT (0) is default "
161 "1 for IB, 2 for Ethernet");
ab9c17a0
JM		 162
163struct mlx4_port_config {
164 struct list_head list;
165 enum mlx4_port_type port_type[MLX4_MAX_PORTS + 1];
166 struct pci_dev *pdev;
167};
168
97989356
AV		 169static atomic_t pf_loading = ATOMIC_INIT(0);
170
27bf91d6
YP		 171int mlx4_check_port_params(struct mlx4_dev *dev,
172 enum mlx4_port_type *port_type)
7ff93f8b
YP		 173{
174 int i;
175
0b997657
YS		 176 if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP)) {
177 for (i = 0; i < dev->caps.num_ports - 1; i++) {
178 if (port_type[i] != port_type[i + 1]) {
1a91de28 179 mlx4_err(dev, "Only same port types supported on this HCA, aborting\n");
27bf91d6
YP		 180 return -EINVAL;
181 }
7ff93f8b
YP		 182 }
183 }
7ff93f8b
YP		 184
185 for (i = 0; i < dev->caps.num_ports; i++) {
186 if (!(port_type[i] & dev->caps.supported_type[i+1])) {
1a91de28
JP		 187 mlx4_err(dev, "Requested port type for port %d is not supported on this HCA\n",
188 i + 1);
7ff93f8b
YP		 189 return -EINVAL;
190 }
191 }
192 return 0;
193}
194
195static void mlx4_set_port_mask(struct mlx4_dev *dev)
196{
197 int i;
198
7ff93f8b 199 for (i = 1; i <= dev->caps.num_ports; ++i)
65dab25d 200 dev->caps.port_mask[i] = dev->caps.port_type[i];
7ff93f8b 201}
f2a3f6a3 202
7ae0e400
MB		 203enum {
204 MLX4_QUERY_FUNC_NUM_SYS_EQS = 1 << 0,
205};
206
207static int mlx4_query_func(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
208{
209 int err = 0;
210 struct mlx4_func func;
211
212 if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS) {
213 err = mlx4_QUERY_FUNC(dev, &func, 0);
214 if (err) {
215 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
216 return err;
217 }
218 dev_cap->max_eqs = func.max_eq;
219 dev_cap->reserved_eqs = func.rsvd_eqs;
220 dev_cap->reserved_uars = func.rsvd_uars;
221 err |= MLX4_QUERY_FUNC_NUM_SYS_EQS;
222 }
223 return err;
224}
225
77507aa2
IS		 226static void mlx4_enable_cqe_eqe_stride(struct mlx4_dev *dev)
227{
228 struct mlx4_caps *dev_cap = &dev->caps;
229
230 /* FW not supporting or cancelled by user */
231 if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_EQE_STRIDE) ||
232 !(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_CQE_STRIDE))
233 return;
234
235 /* Must have 64B CQE_EQE enabled by FW to use bigger stride
236 * When FW has NCSI it may decide not to report 64B CQE/EQEs
237 */
238 if (!(dev_cap->flags & MLX4_DEV_CAP_FLAG_64B_EQE) ||
239 !(dev_cap->flags & MLX4_DEV_CAP_FLAG_64B_CQE)) {
240 dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
241 dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
242 return;
243 }
244
245 if (cache_line_size() == 128 || cache_line_size() == 256) {
246 mlx4_dbg(dev, "Enabling CQE stride cacheLine supported\n");
247 /* Changing the real data inside CQE size to 32B */
248 dev_cap->flags &= ~MLX4_DEV_CAP_FLAG_64B_CQE;
249 dev_cap->flags &= ~MLX4_DEV_CAP_FLAG_64B_EQE;
250
251 if (mlx4_is_master(dev))
252 dev_cap->function_caps |= MLX4_FUNC_CAP_EQE_CQE_STRIDE;
253 } else {
0fab541a
OG		 254 if (cache_line_size() != 32 && cache_line_size() != 64)
255 mlx4_dbg(dev, "Disabling CQE stride, cacheLine size unsupported\n");
77507aa2
IS		 256 dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
257 dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
258 }
259}
260
431df8c7
MB		 261static int _mlx4_dev_port(struct mlx4_dev *dev, int port,
262 struct mlx4_port_cap *port_cap)
263{
264 dev->caps.vl_cap[port] = port_cap->max_vl;
265 dev->caps.ib_mtu_cap[port] = port_cap->ib_mtu;
266 dev->phys_caps.gid_phys_table_len[port] = port_cap->max_gids;
267 dev->phys_caps.pkey_phys_table_len[port] = port_cap->max_pkeys;
268 /* set gid and pkey table operating lengths by default
269 * to non-sriov values
270 */
271 dev->caps.gid_table_len[port] = port_cap->max_gids;
272 dev->caps.pkey_table_len[port] = port_cap->max_pkeys;
273 dev->caps.port_width_cap[port] = port_cap->max_port_width;
274 dev->caps.eth_mtu_cap[port] = port_cap->eth_mtu;
275 dev->caps.def_mac[port] = port_cap->def_mac;
276 dev->caps.supported_type[port] = port_cap->supported_port_types;
277 dev->caps.suggested_type[port] = port_cap->suggested_type;
278 dev->caps.default_sense[port] = port_cap->default_sense;
279 dev->caps.trans_type[port] = port_cap->trans_type;
280 dev->caps.vendor_oui[port] = port_cap->vendor_oui;
281 dev->caps.wavelength[port] = port_cap->wavelength;
282 dev->caps.trans_code[port] = port_cap->trans_code;
283
284 return 0;
285}
286
287static int mlx4_dev_port(struct mlx4_dev *dev, int port,
288 struct mlx4_port_cap *port_cap)
289{
290 int err = 0;
291
292 err = mlx4_QUERY_PORT(dev, port, port_cap);
293
294 if (err)
295 mlx4_err(dev, "QUERY_PORT command failed.\n");
296
297 return err;
298}
299
78500b8c
MM		 300static inline void mlx4_enable_ignore_fcs(struct mlx4_dev *dev)
301{
302 if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_IGNORE_FCS))
303 return;
304
305 if (mlx4_is_mfunc(dev)) {
306 mlx4_dbg(dev, "SRIOV mode - Disabling Ignore FCS");
307 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_IGNORE_FCS;
308 return;
309 }
310
311 if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_FCS_KEEP)) {
312 mlx4_dbg(dev,
313 "Keep FCS is not supported - Disabling Ignore FCS");
314 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_IGNORE_FCS;
315 return;
316 }
317}
318
431df8c7 319#define MLX4_A0_STEERING_TABLE_SIZE 256
3d73c288 320static int mlx4_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
225c7b1f
RD		 321{
322 int err;
5ae2a7a8 323 int i;
225c7b1f
RD		 324
325 err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
326 if (err) {
1a91de28 327 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
225c7b1f
RD		 328 return err;
329 }
c78e25ed 330 mlx4_dev_cap_dump(dev, dev_cap);
225c7b1f
RD		 331
332 if (dev_cap->min_page_sz > PAGE_SIZE) {
1a91de28 333 mlx4_err(dev, "HCA minimum page size of %d bigger than kernel PAGE_SIZE of %ld, aborting\n",
225c7b1f
RD		 334 dev_cap->min_page_sz, PAGE_SIZE);
335 return -ENODEV;
336 }
337 if (dev_cap->num_ports > MLX4_MAX_PORTS) {
1a91de28 338 mlx4_err(dev, "HCA has %d ports, but we only support %d, aborting\n",
225c7b1f
RD		 339 dev_cap->num_ports, MLX4_MAX_PORTS);
340 return -ENODEV;
341 }
342
872bf2fb 343 if (dev_cap->uar_size > pci_resource_len(dev->persist->pdev, 2)) {
1a91de28 344 mlx4_err(dev, "HCA reported UAR size of 0x%x bigger than PCI resource 2 size of 0x%llx, aborting\n",
225c7b1f 345 dev_cap->uar_size,
872bf2fb
YH		 346 (unsigned long long)
347 pci_resource_len(dev->persist->pdev, 2));
225c7b1f
RD		 348 return -ENODEV;
349 }
350
351 dev->caps.num_ports = dev_cap->num_ports;
7ae0e400
MB		 352 dev->caps.num_sys_eqs = dev_cap->num_sys_eqs;
353 dev->phys_caps.num_phys_eqs = dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS ?
354 dev->caps.num_sys_eqs :
355 MLX4_MAX_EQ_NUM;
5ae2a7a8 356 for (i = 1; i <= dev->caps.num_ports; ++i) {
431df8c7
MB		 357 err = _mlx4_dev_port(dev, i, dev_cap->port_cap + i);
358 if (err) {
359 mlx4_err(dev, "QUERY_PORT command failed, aborting\n");
360 return err;
361 }
5ae2a7a8
RD		 362 }
363
ab9c17a0 364 dev->caps.uar_page_size = PAGE_SIZE;
225c7b1f 365 dev->caps.num_uars = dev_cap->uar_size / PAGE_SIZE;
225c7b1f
RD		 366 dev->caps.local_ca_ack_delay = dev_cap->local_ca_ack_delay;
367 dev->caps.bf_reg_size = dev_cap->bf_reg_size;
368 dev->caps.bf_regs_per_page = dev_cap->bf_regs_per_page;
369 dev->caps.max_sq_sg = dev_cap->max_sq_sg;
370 dev->caps.max_rq_sg = dev_cap->max_rq_sg;
371 dev->caps.max_wqes = dev_cap->max_qp_sz;
372 dev->caps.max_qp_init_rdma = dev_cap->max_requester_per_qp;
225c7b1f
RD		 373 dev->caps.max_srq_wqes = dev_cap->max_srq_sz;
374 dev->caps.max_srq_sge = dev_cap->max_rq_sg - 1;
375 dev->caps.reserved_srqs = dev_cap->reserved_srqs;
376 dev->caps.max_sq_desc_sz = dev_cap->max_sq_desc_sz;
377 dev->caps.max_rq_desc_sz = dev_cap->max_rq_desc_sz;
225c7b1f
RD		 378 /*
379 * Subtract 1 from the limit because we need to allocate a
380 * spare CQE so the HCA HW can tell the difference between an
381 * empty CQ and a full CQ.
382 */
383 dev->caps.max_cqes = dev_cap->max_cq_sz - 1;
384 dev->caps.reserved_cqs = dev_cap->reserved_cqs;
385 dev->caps.reserved_eqs = dev_cap->reserved_eqs;
2b8fb286 386 dev->caps.reserved_mtts = dev_cap->reserved_mtts;
225c7b1f 387 dev->caps.reserved_mrws = dev_cap->reserved_mrws;
ab9c17a0
JM		 388
389 /* The first 128 UARs are used for EQ doorbells */
390 dev->caps.reserved_uars = max_t(int, 128, dev_cap->reserved_uars);
225c7b1f 391 dev->caps.reserved_pds = dev_cap->reserved_pds;
012a8ff5
SH		 392 dev->caps.reserved_xrcds = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
393 dev_cap->reserved_xrcds : 0;
394 dev->caps.max_xrcds = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
395 dev_cap->max_xrcds : 0;
2b8fb286
MA		 396 dev->caps.mtt_entry_sz = dev_cap->mtt_entry_sz;
397
149983af 398 dev->caps.max_msg_sz = dev_cap->max_msg_sz;
225c7b1f
RD		 399 dev->caps.page_size_cap = ~(u32) (dev_cap->min_page_sz - 1);
400 dev->caps.flags = dev_cap->flags;
b3416f44 401 dev->caps.flags2 = dev_cap->flags2;
95d04f07
RD		 402 dev->caps.bmme_flags = dev_cap->bmme_flags;
403 dev->caps.reserved_lkey = dev_cap->reserved_lkey;
225c7b1f 404 dev->caps.stat_rate_support = dev_cap->stat_rate_support;
b832be1e 405 dev->caps.max_gso_sz = dev_cap->max_gso_sz;
b3416f44 406 dev->caps.max_rss_tbl_sz = dev_cap->max_rss_tbl_sz;
225c7b1f 407
ca3e57a5
RD		 408 /* Sense port always allowed on supported devices for ConnectX-1 and -2 */
409 if (mlx4_priv(dev)->pci_dev_data & MLX4_PCI_DEV_FORCE_SENSE_PORT)
58a60168 410 dev->caps.flags |= MLX4_DEV_CAP_FLAG_SENSE_SUPPORT;
aadf4f3f
RD		 411 /* Don't do sense port on multifunction devices (for now at least) */
412 if (mlx4_is_mfunc(dev))
413 dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_SENSE_SUPPORT;
58a60168 414
2599d858
AV		 415 if (mlx4_low_memory_profile()) {
416 dev->caps.log_num_macs = MLX4_MIN_LOG_NUM_MAC;
417 dev->caps.log_num_vlans = MLX4_MIN_LOG_NUM_VLANS;
418 } else {
419 dev->caps.log_num_macs = log_num_mac;
420 dev->caps.log_num_vlans = MLX4_LOG_NUM_VLANS;
421 }
93fc9e1b
YP		 422
423 for (i = 1; i <= dev->caps.num_ports; ++i) {
ab9c17a0
JM		 424 dev->caps.port_type[i] = MLX4_PORT_TYPE_NONE;
425 if (dev->caps.supported_type[i]) {
426 /* if only ETH is supported - assign ETH */
427 if (dev->caps.supported_type[i] == MLX4_PORT_TYPE_ETH)
428 dev->caps.port_type[i] = MLX4_PORT_TYPE_ETH;
105c320f 429 /* if only IB is supported, assign IB */
ab9c17a0 430 else if (dev->caps.supported_type[i] ==
105c320f
JM		 431 MLX4_PORT_TYPE_IB)
432 dev->caps.port_type[i] = MLX4_PORT_TYPE_IB;
ab9c17a0 433 else {
105c320f
JM		 434 /* if IB and ETH are supported, we set the port
435 * type according to user selection of port type;
436 * if user selected none, take the FW hint */
437 if (port_type_array[i - 1] == MLX4_PORT_TYPE_NONE)
8d0fc7b6
YP		 438 dev->caps.port_type[i] = dev->caps.suggested_type[i] ?
439 MLX4_PORT_TYPE_ETH : MLX4_PORT_TYPE_IB;
ab9c17a0 440 else
105c320f 441 dev->caps.port_type[i] = port_type_array[i - 1];
ab9c17a0
JM		 442 }
443 }
8d0fc7b6
YP		 444 /*
445 * Link sensing is allowed on the port if 3 conditions are true:
446 * 1. Both protocols are supported on the port.
447 * 2. Different types are supported on the port
448 * 3. FW declared that it supports link sensing
449 */
27bf91d6 450 mlx4_priv(dev)->sense.sense_allowed[i] =
58a60168 451 ((dev->caps.supported_type[i] == MLX4_PORT_TYPE_AUTO) &&
8d0fc7b6 452 (dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
58a60168 453 (dev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT));
7ff93f8b 454
8d0fc7b6
YP		 455 /*
456 * If "default_sense" bit is set, we move the port to "AUTO" mode
457 * and perform sense_port FW command to try and set the correct
458 * port type from beginning
459 */
46c46747 460 if (mlx4_priv(dev)->sense.sense_allowed[i] && dev->caps.default_sense[i]) {
8d0fc7b6
YP		 461 enum mlx4_port_type sensed_port = MLX4_PORT_TYPE_NONE;
462 dev->caps.possible_type[i] = MLX4_PORT_TYPE_AUTO;
463 mlx4_SENSE_PORT(dev, i, &sensed_port);
464 if (sensed_port != MLX4_PORT_TYPE_NONE)
465 dev->caps.port_type[i] = sensed_port;
466 } else {
467 dev->caps.possible_type[i] = dev->caps.port_type[i];
468 }
469
431df8c7
MB		 470 if (dev->caps.log_num_macs > dev_cap->port_cap[i].log_max_macs) {
471 dev->caps.log_num_macs = dev_cap->port_cap[i].log_max_macs;
1a91de28 472 mlx4_warn(dev, "Requested number of MACs is too much for port %d, reducing to %d\n",
93fc9e1b
YP		 473 i, 1 << dev->caps.log_num_macs);
474 }
431df8c7
MB		 475 if (dev->caps.log_num_vlans > dev_cap->port_cap[i].log_max_vlans) {
476 dev->caps.log_num_vlans = dev_cap->port_cap[i].log_max_vlans;
1a91de28 477 mlx4_warn(dev, "Requested number of VLANs is too much for port %d, reducing to %d\n",
93fc9e1b
YP		 478 i, 1 << dev->caps.log_num_vlans);
479 }
480 }
481
ac0a72a3
OG		 482 if (mlx4_is_master(dev) && (dev->caps.num_ports == 2) &&
483 (port_type_array[0] == MLX4_PORT_TYPE_IB) &&
484 (port_type_array[1] == MLX4_PORT_TYPE_ETH)) {
485 mlx4_warn(dev,
486 "Granular QoS per VF not supported with IB/Eth configuration\n");
487 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_QOS_VPP;
488 }
489
47d8417f 490 dev->caps.max_counters = dev_cap->max_counters;
f2a3f6a3 491
93fc9e1b
YP		 492 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] = dev_cap->reserved_qps;
493 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] =
494 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] =
495 (1 << dev->caps.log_num_macs) *
496 (1 << dev->caps.log_num_vlans) *
93fc9e1b
YP		 497 dev->caps.num_ports;
498 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH] = MLX4_NUM_FEXCH;
7d077cd3
MB		 499
500 if (dev_cap->dmfs_high_rate_qpn_base > 0 &&
501 dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN)
502 dev->caps.dmfs_high_rate_qpn_base = dev_cap->dmfs_high_rate_qpn_base;
503 else
504 dev->caps.dmfs_high_rate_qpn_base =
505 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
506
507 if (dev_cap->dmfs_high_rate_qpn_range > 0 &&
508 dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN) {
509 dev->caps.dmfs_high_rate_qpn_range = dev_cap->dmfs_high_rate_qpn_range;
510 dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_DEFAULT;
511 dev->caps.flags2 |= MLX4_DEV_CAP_FLAG2_FS_A0;
512 } else {
513 dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_NOT_SUPPORTED;
514 dev->caps.dmfs_high_rate_qpn_base =
515 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
516 dev->caps.dmfs_high_rate_qpn_range = MLX4_A0_STEERING_TABLE_SIZE;
517 }
518
fc31e256
OG		 519 dev->caps.rl_caps = dev_cap->rl_caps;
520
d57febe1 521 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_RSS_RAW_ETH] =
7d077cd3 522 dev->caps.dmfs_high_rate_qpn_range;
93fc9e1b
YP		 523
524 dev->caps.reserved_qps = dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] +
525 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] +
526 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] +
527 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH];
528
e2c76824 529 dev->caps.sqp_demux = (mlx4_is_master(dev)) ? MLX4_MAX_NUM_SLAVES : 0;
08ff3235 530
b3051320 531 if (!enable_64b_cqe_eqe && !mlx4_is_slave(dev)) {
08ff3235
OG		 532 if (dev_cap->flags &
533 (MLX4_DEV_CAP_FLAG_64B_CQE | MLX4_DEV_CAP_FLAG_64B_EQE)) {
534 mlx4_warn(dev, "64B EQEs/CQEs supported by the device but not enabled\n");
535 dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_64B_CQE;
536 dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_64B_EQE;
537 }
77507aa2
IS		 538
539 if (dev_cap->flags2 &
540 (MLX4_DEV_CAP_FLAG2_CQE_STRIDE |
541 MLX4_DEV_CAP_FLAG2_EQE_STRIDE)) {
542 mlx4_warn(dev, "Disabling EQE/CQE stride per user request\n");
543 dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
544 dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
545 }
08ff3235
OG		 546 }
547
f97b4b5d 548 if ((dev->caps.flags &
08ff3235
OG		 549 (MLX4_DEV_CAP_FLAG_64B_CQE | MLX4_DEV_CAP_FLAG_64B_EQE)) &&
550 mlx4_is_master(dev))
551 dev->caps.function_caps |= MLX4_FUNC_CAP_64B_EQE_CQE;
552
ddae0349 553 if (!mlx4_is_slave(dev)) {
77507aa2 554 mlx4_enable_cqe_eqe_stride(dev);
ddae0349 555 dev->caps.alloc_res_qp_mask =
d57febe1
MB		 556 (dev->caps.bf_reg_size ? MLX4_RESERVE_ETH_BF_QP : 0) |
557 MLX4_RESERVE_A0_QP;
3742cc65
IS		 558
559 if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ETS_CFG) &&
560 dev->caps.flags & MLX4_DEV_CAP_FLAG_SET_ETH_SCHED) {
561 mlx4_warn(dev, "Old device ETS support detected\n");
562 mlx4_warn(dev, "Consider upgrading device FW.\n");
563 dev->caps.flags2 |= MLX4_DEV_CAP_FLAG2_ETS_CFG;
564 }
565
ddae0349
EE		 566 } else {
567 dev->caps.alloc_res_qp_mask = 0;
568 }
77507aa2 569
78500b8c
MM		 570 mlx4_enable_ignore_fcs(dev);
571
225c7b1f
RD		 572 return 0;
573}
b912b2f8
EP		 574
575static int mlx4_get_pcie_dev_link_caps(struct mlx4_dev *dev,
576 enum pci_bus_speed *speed,
577 enum pcie_link_width *width)
578{
579 u32 lnkcap1, lnkcap2;
580 int err1, err2;
581
582#define PCIE_MLW_CAP_SHIFT 4 /* start of MLW mask in link capabilities */
583
584 *speed = PCI_SPEED_UNKNOWN;
585 *width = PCIE_LNK_WIDTH_UNKNOWN;
586
872bf2fb
YH		 587 err1 = pcie_capability_read_dword(dev->persist->pdev, PCI_EXP_LNKCAP,
588 &lnkcap1);
589 err2 = pcie_capability_read_dword(dev->persist->pdev, PCI_EXP_LNKCAP2,
590 &lnkcap2);
b912b2f8
EP		 591 if (!err2 && lnkcap2) { /* PCIe r3.0-compliant */
592 if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_8_0GB)
593 *speed = PCIE_SPEED_8_0GT;
594 else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_5_0GB)
595 *speed = PCIE_SPEED_5_0GT;
596 else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_2_5GB)
597 *speed = PCIE_SPEED_2_5GT;
598 }
599 if (!err1) {
600 *width = (lnkcap1 & PCI_EXP_LNKCAP_MLW) >> PCIE_MLW_CAP_SHIFT;
601 if (!lnkcap2) { /* pre-r3.0 */
602 if (lnkcap1 & PCI_EXP_LNKCAP_SLS_5_0GB)
603 *speed = PCIE_SPEED_5_0GT;
604 else if (lnkcap1 & PCI_EXP_LNKCAP_SLS_2_5GB)
605 *speed = PCIE_SPEED_2_5GT;
606 }
607 }
608
609 if (*speed == PCI_SPEED_UNKNOWN || *width == PCIE_LNK_WIDTH_UNKNOWN) {
610 return err1 ? err1 :
611 err2 ? err2 : -EINVAL;
612 }
613 return 0;
614}
615
616static void mlx4_check_pcie_caps(struct mlx4_dev *dev)
617{
618 enum pcie_link_width width, width_cap;
619 enum pci_bus_speed speed, speed_cap;
620 int err;
621
622#define PCIE_SPEED_STR(speed) \
623 (speed == PCIE_SPEED_8_0GT ? "8.0GT/s" : \
624 speed == PCIE_SPEED_5_0GT ? "5.0GT/s" : \
625 speed == PCIE_SPEED_2_5GT ? "2.5GT/s" : \
626 "Unknown")
627
628 err = mlx4_get_pcie_dev_link_caps(dev, &speed_cap, &width_cap);
629 if (err) {
630 mlx4_warn(dev,
631 "Unable to determine PCIe device BW capabilities\n");
632 return;
633 }
634
872bf2fb 635 err = pcie_get_minimum_link(dev->persist->pdev, &speed, &width);
b912b2f8
EP		 636 if (err || speed == PCI_SPEED_UNKNOWN ||
637 width == PCIE_LNK_WIDTH_UNKNOWN) {
638 mlx4_warn(dev,
639 "Unable to determine PCI device chain minimum BW\n");
640 return;
641 }
642
643 if (width != width_cap || speed != speed_cap)
644 mlx4_warn(dev,
645 "PCIe BW is different than device's capability\n");
646
647 mlx4_info(dev, "PCIe link speed is %s, device supports %s\n",
648 PCIE_SPEED_STR(speed), PCIE_SPEED_STR(speed_cap));
649 mlx4_info(dev, "PCIe link width is x%d, device supports x%d\n",
650 width, width_cap);
651 return;
652}
653
ab9c17a0
JM		 654/*The function checks if there are live vf, return the num of them*/
655static int mlx4_how_many_lives_vf(struct mlx4_dev *dev)
656{
657 struct mlx4_priv *priv = mlx4_priv(dev);
658 struct mlx4_slave_state *s_state;
659 int i;
660 int ret = 0;
661
662 for (i = 1/*the ppf is 0*/; i < dev->num_slaves; ++i) {
663 s_state = &priv->mfunc.master.slave_state[i];
664 if (s_state->active && s_state->last_cmd !=
665 MLX4_COMM_CMD_RESET) {
666 mlx4_warn(dev, "%s: slave: %d is still active\n",
667 __func__, i);
668 ret++;
669 }
670 }
671 return ret;
672}
673
396f2feb
JM		 674int mlx4_get_parav_qkey(struct mlx4_dev *dev, u32 qpn, u32 *qkey)
675{
676 u32 qk = MLX4_RESERVED_QKEY_BASE;
47605df9
JM		 677
678 if (qpn >= dev->phys_caps.base_tunnel_sqpn + 8 * MLX4_MFUNC_MAX ||
679 qpn < dev->phys_caps.base_proxy_sqpn)
396f2feb
JM		 680 return -EINVAL;
681
47605df9 682 if (qpn >= dev->phys_caps.base_tunnel_sqpn)
396f2feb 683 /* tunnel qp */
47605df9 684 qk += qpn - dev->phys_caps.base_tunnel_sqpn;
396f2feb 685 else
47605df9 686 qk += qpn - dev->phys_caps.base_proxy_sqpn;
396f2feb
JM		 687 *qkey = qk;
688 return 0;
689}
690EXPORT_SYMBOL(mlx4_get_parav_qkey);
691
54679e14
JM		 692void mlx4_sync_pkey_table(struct mlx4_dev *dev, int slave, int port, int i, int val)
693{
694 struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
695
696 if (!mlx4_is_master(dev))
697 return;
698
699 priv->virt2phys_pkey[slave][port - 1][i] = val;
700}
701EXPORT_SYMBOL(mlx4_sync_pkey_table);
702
afa8fd1d
JM		 703void mlx4_put_slave_node_guid(struct mlx4_dev *dev, int slave, __be64 guid)
704{
705 struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
706
707 if (!mlx4_is_master(dev))
708 return;
709
710 priv->slave_node_guids[slave] = guid;
711}
712EXPORT_SYMBOL(mlx4_put_slave_node_guid);
713
714__be64 mlx4_get_slave_node_guid(struct mlx4_dev *dev, int slave)
715{
716 struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
717
718 if (!mlx4_is_master(dev))
719 return 0;
720
721 return priv->slave_node_guids[slave];
722}
723EXPORT_SYMBOL(mlx4_get_slave_node_guid);
724
e10903b0 725int mlx4_is_slave_active(struct mlx4_dev *dev, int slave)
ab9c17a0
JM		 726{
727 struct mlx4_priv *priv = mlx4_priv(dev);
728 struct mlx4_slave_state *s_slave;
729
730 if (!mlx4_is_master(dev))
731 return 0;
732
733 s_slave = &priv->mfunc.master.slave_state[slave];
734 return !!s_slave->active;
735}
736EXPORT_SYMBOL(mlx4_is_slave_active);
737
7b8157be
JM		 738static void slave_adjust_steering_mode(struct mlx4_dev *dev,
739 struct mlx4_dev_cap *dev_cap,
740 struct mlx4_init_hca_param *hca_param)
741{
742 dev->caps.steering_mode = hca_param->steering_mode;
743 if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED) {
744 dev->caps.num_qp_per_mgm = dev_cap->fs_max_num_qp_per_entry;
745 dev->caps.fs_log_max_ucast_qp_range_size =
746 dev_cap->fs_log_max_ucast_qp_range_size;
747 } else
748 dev->caps.num_qp_per_mgm =
749 4 * ((1 << hca_param->log_mc_entry_sz)/16 - 2);
750
751 mlx4_dbg(dev, "Steering mode is: %s\n",
752 mlx4_steering_mode_str(dev->caps.steering_mode));
753}
754
ab9c17a0
JM		 755static int mlx4_slave_cap(struct mlx4_dev *dev)
756{
757 int err;
758 u32 page_size;
759 struct mlx4_dev_cap dev_cap;
760 struct mlx4_func_cap func_cap;
761 struct mlx4_init_hca_param hca_param;
225c6c8c 762 u8 i;
ab9c17a0
JM		 763
764 memset(&hca_param, 0, sizeof(hca_param));
765 err = mlx4_QUERY_HCA(dev, &hca_param);
766 if (err) {
1a91de28 767 mlx4_err(dev, "QUERY_HCA command failed, aborting\n");
ab9c17a0
JM		 768 return err;
769 }
770
483e0132
EP		 771 /* fail if the hca has an unknown global capability
772 * at this time global_caps should be always zeroed
773 */
774 if (hca_param.global_caps) {
ab9c17a0
JM		 775 mlx4_err(dev, "Unknown hca global capabilities\n");
776 return -ENOSYS;
777 }
778
779 mlx4_log_num_mgm_entry_size = hca_param.log_mc_entry_sz;
780
ddd8a6c1
EE		 781 dev->caps.hca_core_clock = hca_param.hca_core_clock;
782
ab9c17a0 783 memset(&dev_cap, 0, sizeof(dev_cap));
b91cb3eb 784 dev->caps.max_qp_dest_rdma = 1 << hca_param.log_rd_per_qp;
ab9c17a0
JM		 785 err = mlx4_dev_cap(dev, &dev_cap);
786 if (err) {
1a91de28 787 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
ab9c17a0
JM		 788 return err;
789 }
790
b91cb3eb
JM		 791 err = mlx4_QUERY_FW(dev);
792 if (err)
1a91de28 793 mlx4_err(dev, "QUERY_FW command failed: could not get FW version\n");
b91cb3eb 794
ab9c17a0
JM		 795 page_size = ~dev->caps.page_size_cap + 1;
796 mlx4_warn(dev, "HCA minimum page size:%d\n", page_size);
797 if (page_size > PAGE_SIZE) {
1a91de28 798 mlx4_err(dev, "HCA minimum page size of %d bigger than kernel PAGE_SIZE of %ld, aborting\n",
ab9c17a0
JM		 799 page_size, PAGE_SIZE);
800 return -ENODEV;
801 }
802
803 /* slave gets uar page size from QUERY_HCA fw command */
804 dev->caps.uar_page_size = 1 << (hca_param.uar_page_sz + 12);
805
806 /* TODO: relax this assumption */
807 if (dev->caps.uar_page_size != PAGE_SIZE) {
808 mlx4_err(dev, "UAR size:%d != kernel PAGE_SIZE of %ld\n",
809 dev->caps.uar_page_size, PAGE_SIZE);
810 return -ENODEV;
811 }
812
813 memset(&func_cap, 0, sizeof(func_cap));
47605df9 814 err = mlx4_QUERY_FUNC_CAP(dev, 0, &func_cap);
ab9c17a0 815 if (err) {
1a91de28
JP		 816 mlx4_err(dev, "QUERY_FUNC_CAP general command failed, aborting (%d)\n",
817 err);
ab9c17a0
JM		 818 return err;
819 }
820
821 if ((func_cap.pf_context_behaviour | PF_CONTEXT_BEHAVIOUR_MASK) !=
822 PF_CONTEXT_BEHAVIOUR_MASK) {
7d077cd3
MB		 823 mlx4_err(dev, "Unknown pf context behaviour %x known flags %x\n",
824 func_cap.pf_context_behaviour, PF_CONTEXT_BEHAVIOUR_MASK);
ab9c17a0
JM		 825 return -ENOSYS;
826 }
827
ab9c17a0 828 dev->caps.num_ports = func_cap.num_ports;
5a0d0a61
JM		 829 dev->quotas.qp = func_cap.qp_quota;
830 dev->quotas.srq = func_cap.srq_quota;
831 dev->quotas.cq = func_cap.cq_quota;
832 dev->quotas.mpt = func_cap.mpt_quota;
833 dev->quotas.mtt = func_cap.mtt_quota;
834 dev->caps.num_qps = 1 << hca_param.log_num_qps;
835 dev->caps.num_srqs = 1 << hca_param.log_num_srqs;
836 dev->caps.num_cqs = 1 << hca_param.log_num_cqs;
837 dev->caps.num_mpts = 1 << hca_param.log_mpt_sz;
838 dev->caps.num_eqs = func_cap.max_eq;
839 dev->caps.reserved_eqs = func_cap.reserved_eq;
f0ce0615 840 dev->caps.reserved_lkey = func_cap.reserved_lkey;
ab9c17a0
JM		 841 dev->caps.num_pds = MLX4_NUM_PDS;
842 dev->caps.num_mgms = 0;
843 dev->caps.num_amgms = 0;
844
ab9c17a0 845 if (dev->caps.num_ports > MLX4_MAX_PORTS) {
1a91de28
JP		 846 mlx4_err(dev, "HCA has %d ports, but we only support %d, aborting\n",
847 dev->caps.num_ports, MLX4_MAX_PORTS);
ab9c17a0
JM		 848 return -ENODEV;
849 }
850
99ec41d0 851 dev->caps.qp0_qkey = kcalloc(dev->caps.num_ports, sizeof(u32), GFP_KERNEL);
47605df9
JM		 852 dev->caps.qp0_tunnel = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
853 dev->caps.qp0_proxy = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
854 dev->caps.qp1_tunnel = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
855 dev->caps.qp1_proxy = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
856
857 if (!dev->caps.qp0_tunnel || !dev->caps.qp0_proxy ||
99ec41d0
JM		 858 !dev->caps.qp1_tunnel || !dev->caps.qp1_proxy ||
859 !dev->caps.qp0_qkey) {
47605df9
JM		 860 err = -ENOMEM;
861 goto err_mem;
862 }
863
6634961c 864 for (i = 1; i <= dev->caps.num_ports; ++i) {
225c6c8c 865 err = mlx4_QUERY_FUNC_CAP(dev, i, &func_cap);
47605df9 866 if (err) {
1a91de28
JP		 867 mlx4_err(dev, "QUERY_FUNC_CAP port command failed for port %d, aborting (%d)\n",
868 i, err);
47605df9
JM		 869 goto err_mem;
870 }
99ec41d0 871 dev->caps.qp0_qkey[i - 1] = func_cap.qp0_qkey;
47605df9
JM		 872 dev->caps.qp0_tunnel[i - 1] = func_cap.qp0_tunnel_qpn;
873 dev->caps.qp0_proxy[i - 1] = func_cap.qp0_proxy_qpn;
874 dev->caps.qp1_tunnel[i - 1] = func_cap.qp1_tunnel_qpn;
875 dev->caps.qp1_proxy[i - 1] = func_cap.qp1_proxy_qpn;
6230bb23 876 dev->caps.port_mask[i] = dev->caps.port_type[i];
8e1a28e8 877 dev->caps.phys_port_id[i] = func_cap.phys_port_id;
6634961c
JM		 878 if (mlx4_get_slave_pkey_gid_tbl_len(dev, i,
879 &dev->caps.gid_table_len[i],
880 &dev->caps.pkey_table_len[i]))
47605df9 881 goto err_mem;
6634961c 882 }
6230bb23 883
ab9c17a0
JM		 884 if (dev->caps.uar_page_size * (dev->caps.num_uars -
885 dev->caps.reserved_uars) >
872bf2fb
YH		 886 pci_resource_len(dev->persist->pdev,
887 2)) {
1a91de28 888 mlx4_err(dev, "HCA reported UAR region size of 0x%x bigger than PCI resource 2 size of 0x%llx, aborting\n",
ab9c17a0 889 dev->caps.uar_page_size * dev->caps.num_uars,
872bf2fb
YH		 890 (unsigned long long)
891 pci_resource_len(dev->persist->pdev, 2));
47605df9 892 goto err_mem;
ab9c17a0
JM		 893 }
894
08ff3235
OG		 895 if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_64B_EQE_ENABLED) {
896 dev->caps.eqe_size = 64;
897 dev->caps.eqe_factor = 1;
898 } else {
899 dev->caps.eqe_size = 32;
900 dev->caps.eqe_factor = 0;
901 }
902
903 if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_64B_CQE_ENABLED) {
904 dev->caps.cqe_size = 64;
77507aa2 905 dev->caps.userspace_caps |= MLX4_USER_DEV_CAP_LARGE_CQE;
08ff3235
OG		 906 } else {
907 dev->caps.cqe_size = 32;
908 }
909
77507aa2
IS		 910 if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_EQE_STRIDE_ENABLED) {
911 dev->caps.eqe_size = hca_param.eqe_size;
912 dev->caps.eqe_factor = 0;
913 }
914
915 if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_CQE_STRIDE_ENABLED) {
916 dev->caps.cqe_size = hca_param.cqe_size;
917 /* User still need to know when CQE > 32B */
918 dev->caps.userspace_caps |= MLX4_USER_DEV_CAP_LARGE_CQE;
919 }
920
f9bd2d7f 921 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
1a91de28 922 mlx4_warn(dev, "Timestamping is not supported in slave mode\n");
f9bd2d7f 923
7b8157be 924 slave_adjust_steering_mode(dev, &dev_cap, &hca_param);
802f42a8
IS		 925 mlx4_dbg(dev, "RSS support for IP fragments is %s\n",
926 hca_param.rss_ip_frags ? "on" : "off");
7b8157be 927
ddae0349
EE		 928 if (func_cap.extra_flags & MLX4_QUERY_FUNC_FLAGS_BF_RES_QP &&
929 dev->caps.bf_reg_size)
930 dev->caps.alloc_res_qp_mask |= MLX4_RESERVE_ETH_BF_QP;
931
d57febe1
MB		 932 if (func_cap.extra_flags & MLX4_QUERY_FUNC_FLAGS_A0_RES_QP)
933 dev->caps.alloc_res_qp_mask |= MLX4_RESERVE_A0_QP;
934
ab9c17a0 935 return 0;
47605df9
JM		 936
937err_mem:
99ec41d0 938 kfree(dev->caps.qp0_qkey);
47605df9
JM		 939 kfree(dev->caps.qp0_tunnel);
940 kfree(dev->caps.qp0_proxy);
941 kfree(dev->caps.qp1_tunnel);
942 kfree(dev->caps.qp1_proxy);
99ec41d0
JM		 943 dev->caps.qp0_qkey = NULL;
944 dev->caps.qp0_tunnel = NULL;
945 dev->caps.qp0_proxy = NULL;
946 dev->caps.qp1_tunnel = NULL;
947 dev->caps.qp1_proxy = NULL;
47605df9
JM		 948
949 return err;
ab9c17a0 950}
225c7b1f 951
b046ffe5
EP		 952static void mlx4_request_modules(struct mlx4_dev *dev)
953{
954 int port;
955 int has_ib_port = false;
956 int has_eth_port = false;
957#define EN_DRV_NAME "mlx4_en"
958#define IB_DRV_NAME "mlx4_ib"
959
960 for (port = 1; port <= dev->caps.num_ports; port++) {
961 if (dev->caps.port_type[port] == MLX4_PORT_TYPE_IB)
962 has_ib_port = true;
963 else if (dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH)
964 has_eth_port = true;
965 }
966
b046ffe5
EP		 967 if (has_eth_port)
968 request_module_nowait(EN_DRV_NAME);
f24f790f
OG		 969 if (has_ib_port || (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE))
970 request_module_nowait(IB_DRV_NAME);
b046ffe5
EP		 971}
972
7ff93f8b
YP		 973/*
974 * Change the port configuration of the device.
975 * Every user of this function must hold the port mutex.
976 */
27bf91d6
YP		 977int mlx4_change_port_types(struct mlx4_dev *dev,
978 enum mlx4_port_type *port_types)
7ff93f8b
YP		 979{
980 int err = 0;
981 int change = 0;
982 int port;
983
984 for (port = 0; port < dev->caps.num_ports; port++) {
27bf91d6
YP		 985 /* Change the port type only if the new type is different
986 * from the current, and not set to Auto */
3d8f9308 987 if (port_types[port] != dev->caps.port_type[port + 1])
7ff93f8b 988 change = 1;
7ff93f8b
YP		 989 }
990 if (change) {
991 mlx4_unregister_device(dev);
992 for (port = 1; port <= dev->caps.num_ports; port++) {
993 mlx4_CLOSE_PORT(dev, port);
1e0f03d5 994 dev->caps.port_type[port] = port_types[port - 1];
6634961c 995 err = mlx4_SET_PORT(dev, port, -1);
7ff93f8b 996 if (err) {
1a91de28
JP		 997 mlx4_err(dev, "Failed to set port %d, aborting\n",
998 port);
7ff93f8b
YP		 999 goto out;
1000 }
1001 }
1002 mlx4_set_port_mask(dev);
1003 err = mlx4_register_device(dev);
b046ffe5
EP		 1004 if (err) {
1005 mlx4_err(dev, "Failed to register device\n");
1006 goto out;
1007 }
1008 mlx4_request_modules(dev);
7ff93f8b
YP		 1009 }
1010
1011out:
1012 return err;
1013}
1014
1015static ssize_t show_port_type(struct device *dev,
1016 struct device_attribute *attr,
1017 char *buf)
1018{
1019 struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1020 port_attr);
1021 struct mlx4_dev *mdev = info->dev;
27bf91d6
YP		 1022 char type[8];
1023
1024 sprintf(type, "%s",
1025 (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_IB) ?
1026 "ib" : "eth");
1027 if (mdev->caps.possible_type[info->port] == MLX4_PORT_TYPE_AUTO)
1028 sprintf(buf, "auto (%s)\n", type);
1029 else
1030 sprintf(buf, "%s\n", type);
7ff93f8b 1031
27bf91d6 1032 return strlen(buf);
7ff93f8b
YP		 1033}
1034
1035static ssize_t set_port_type(struct device *dev,
1036 struct device_attribute *attr,
1037 const char *buf, size_t count)
1038{
1039 struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1040 port_attr);
1041 struct mlx4_dev *mdev = info->dev;
1042 struct mlx4_priv *priv = mlx4_priv(mdev);
1043 enum mlx4_port_type types[MLX4_MAX_PORTS];
27bf91d6 1044 enum mlx4_port_type new_types[MLX4_MAX_PORTS];
0a984556 1045 static DEFINE_MUTEX(set_port_type_mutex);
7ff93f8b
YP		 1046 int i;
1047 int err = 0;
1048
0a984556
AV		 1049 mutex_lock(&set_port_type_mutex);
1050
7ff93f8b
YP		 1051 if (!strcmp(buf, "ib\n"))
1052 info->tmp_type = MLX4_PORT_TYPE_IB;
1053 else if (!strcmp(buf, "eth\n"))
1054 info->tmp_type = MLX4_PORT_TYPE_ETH;
27bf91d6
YP		 1055 else if (!strcmp(buf, "auto\n"))
1056 info->tmp_type = MLX4_PORT_TYPE_AUTO;
7ff93f8b
YP		 1057 else {
1058 mlx4_err(mdev, "%s is not supported port type\n", buf);
0a984556
AV		 1059 err = -EINVAL;
1060 goto err_out;
7ff93f8b
YP		 1061 }
1062
27bf91d6 1063 mlx4_stop_sense(mdev);
7ff93f8b 1064 mutex_lock(&priv->port_mutex);
27bf91d6
YP		 1065 /* Possible type is always the one that was delivered */
1066 mdev->caps.possible_type[info->port] = info->tmp_type;
1067
1068 for (i = 0; i < mdev->caps.num_ports; i++) {
7ff93f8b 1069 types[i] = priv->port[i+1].tmp_type ? priv->port[i+1].tmp_type :
27bf91d6
YP		 1070 mdev->caps.possible_type[i+1];
1071 if (types[i] == MLX4_PORT_TYPE_AUTO)
1072 types[i] = mdev->caps.port_type[i+1];
1073 }
7ff93f8b 1074
58a60168
YP		 1075 if (!(mdev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
1076 !(mdev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT)) {
27bf91d6
YP		 1077 for (i = 1; i <= mdev->caps.num_ports; i++) {
1078 if (mdev->caps.possible_type[i] == MLX4_PORT_TYPE_AUTO) {
1079 mdev->caps.possible_type[i] = mdev->caps.port_type[i];
1080 err = -EINVAL;
1081 }
1082 }
1083 }
1084 if (err) {
1a91de28 1085 mlx4_err(mdev, "Auto sensing is not supported on this HCA. Set only 'eth' or 'ib' for both ports (should be the same)\n");
27bf91d6
YP		 1086 goto out;
1087 }
1088
1089 mlx4_do_sense_ports(mdev, new_types, types);
1090
1091 err = mlx4_check_port_params(mdev, new_types);
7ff93f8b
YP		 1092 if (err)
1093 goto out;
1094
27bf91d6
YP		 1095 /* We are about to apply the changes after the configuration
1096 * was verified, no need to remember the temporary types
1097 * any more */
1098 for (i = 0; i < mdev->caps.num_ports; i++)
1099 priv->port[i + 1].tmp_type = 0;
7ff93f8b 1100
27bf91d6 1101 err = mlx4_change_port_types(mdev, new_types);
7ff93f8b
YP		 1102
1103out:
27bf91d6 1104 mlx4_start_sense(mdev);
7ff93f8b 1105 mutex_unlock(&priv->port_mutex);
0a984556
AV		 1106err_out:
1107 mutex_unlock(&set_port_type_mutex);
1108
7ff93f8b
YP		 1109 return err ? err : count;
1110}
1111
096335b3
OG		 1112enum ibta_mtu {
1113 IB_MTU_256 = 1,
1114 IB_MTU_512 = 2,
1115 IB_MTU_1024 = 3,
1116 IB_MTU_2048 = 4,
1117 IB_MTU_4096 = 5
1118};
1119
1120static inline int int_to_ibta_mtu(int mtu)
1121{
1122 switch (mtu) {
1123 case 256: return IB_MTU_256;
1124 case 512: return IB_MTU_512;
1125 case 1024: return IB_MTU_1024;
1126 case 2048: return IB_MTU_2048;
1127 case 4096: return IB_MTU_4096;
1128 default: return -1;
1129 }
1130}
1131
1132static inline int ibta_mtu_to_int(enum ibta_mtu mtu)
1133{
1134 switch (mtu) {
1135 case IB_MTU_256: return 256;
1136 case IB_MTU_512: return 512;
1137 case IB_MTU_1024: return 1024;
1138 case IB_MTU_2048: return 2048;
1139 case IB_MTU_4096: return 4096;
1140 default: return -1;
1141 }
1142}
1143
1144static ssize_t show_port_ib_mtu(struct device *dev,
1145 struct device_attribute *attr,
1146 char *buf)
1147{
1148 struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1149 port_mtu_attr);
1150 struct mlx4_dev *mdev = info->dev;
1151
1152 if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH)
1153 mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
1154
1155 sprintf(buf, "%d\n",
1156 ibta_mtu_to_int(mdev->caps.port_ib_mtu[info->port]));
1157 return strlen(buf);
1158}
1159
1160static ssize_t set_port_ib_mtu(struct device *dev,
1161 struct device_attribute *attr,
1162 const char *buf, size_t count)
1163{
1164 struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1165 port_mtu_attr);
1166 struct mlx4_dev *mdev = info->dev;
1167 struct mlx4_priv *priv = mlx4_priv(mdev);
1168 int err, port, mtu, ibta_mtu = -1;
1169
1170 if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH) {
1171 mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
1172 return -EINVAL;
1173 }
1174
618fad95
DB		 1175 err = kstrtoint(buf, 0, &mtu);
1176 if (!err)
096335b3
OG		 1177 ibta_mtu = int_to_ibta_mtu(mtu);
1178
618fad95 1179 if (err || ibta_mtu < 0) {
096335b3
OG		 1180 mlx4_err(mdev, "%s is invalid IBTA mtu\n", buf);
1181 return -EINVAL;
1182 }
1183
1184 mdev->caps.port_ib_mtu[info->port] = ibta_mtu;
1185
1186 mlx4_stop_sense(mdev);
1187 mutex_lock(&priv->port_mutex);
1188 mlx4_unregister_device(mdev);
1189 for (port = 1; port <= mdev->caps.num_ports; port++) {
1190 mlx4_CLOSE_PORT(mdev, port);
6634961c 1191 err = mlx4_SET_PORT(mdev, port, -1);
096335b3 1192 if (err) {
1a91de28
JP		 1193 mlx4_err(mdev, "Failed to set port %d, aborting\n",
1194 port);
096335b3
OG		 1195 goto err_set_port;
1196 }
1197 }
1198 err = mlx4_register_device(mdev);
1199err_set_port:
1200 mutex_unlock(&priv->port_mutex);
1201 mlx4_start_sense(mdev);
1202 return err ? err : count;
1203}
1204
53f33ae2
MS		 1205int mlx4_bond(struct mlx4_dev *dev)
1206{
1207 int ret = 0;
1208 struct mlx4_priv *priv = mlx4_priv(dev);
1209
1210 mutex_lock(&priv->bond_mutex);
1211
1212 if (!mlx4_is_bonded(dev))
1213 ret = mlx4_do_bond(dev, true);
1214 else
1215 ret = 0;
1216
1217 mutex_unlock(&priv->bond_mutex);
1218 if (ret)
1219 mlx4_err(dev, "Failed to bond device: %d\n", ret);
1220 else
1221 mlx4_dbg(dev, "Device is bonded\n");
1222 return ret;
1223}
1224EXPORT_SYMBOL_GPL(mlx4_bond);
1225
1226int mlx4_unbond(struct mlx4_dev *dev)
1227{
1228 int ret = 0;
1229 struct mlx4_priv *priv = mlx4_priv(dev);
1230
1231 mutex_lock(&priv->bond_mutex);
1232
1233 if (mlx4_is_bonded(dev))
1234 ret = mlx4_do_bond(dev, false);
1235
1236 mutex_unlock(&priv->bond_mutex);
1237 if (ret)
1238 mlx4_err(dev, "Failed to unbond device: %d\n", ret);
1239 else
1240 mlx4_dbg(dev, "Device is unbonded\n");
1241 return ret;
1242}
1243EXPORT_SYMBOL_GPL(mlx4_unbond);
1244
1245
1246int mlx4_port_map_set(struct mlx4_dev *dev, struct mlx4_port_map *v2p)
1247{
1248 u8 port1 = v2p->port1;
1249 u8 port2 = v2p->port2;
1250 struct mlx4_priv *priv = mlx4_priv(dev);
1251 int err;
1252
1253 if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_PORT_REMAP))
1254 return -ENOTSUPP;
1255
1256 mutex_lock(&priv->bond_mutex);
1257
1258 /* zero means keep current mapping for this port */
1259 if (port1 == 0)
1260 port1 = priv->v2p.port1;
1261 if (port2 == 0)
1262 port2 = priv->v2p.port2;
1263
1264 if ((port1 < 1) || (port1 > MLX4_MAX_PORTS) ||
1265 (port2 < 1) || (port2 > MLX4_MAX_PORTS) ||
1266 (port1 == 2 && port2 == 1)) {
1267 /* besides boundary checks cross mapping makes
1268 * no sense and therefore not allowed */
1269 err = -EINVAL;
1270 } else if ((port1 == priv->v2p.port1) &&
1271 (port2 == priv->v2p.port2)) {
1272 err = 0;
1273 } else {
1274 err = mlx4_virt2phy_port_map(dev, port1, port2);
1275 if (!err) {
1276 mlx4_dbg(dev, "port map changed: [%d][%d]\n",
1277 port1, port2);
1278 priv->v2p.port1 = port1;
1279 priv->v2p.port2 = port2;
1280 } else {
1281 mlx4_err(dev, "Failed to change port mape: %d\n", err);
1282 }
1283 }
1284
1285 mutex_unlock(&priv->bond_mutex);
1286 return err;
1287}
1288EXPORT_SYMBOL_GPL(mlx4_port_map_set);
1289
e8f9b2ed 1290static int mlx4_load_fw(struct mlx4_dev *dev)
225c7b1f
RD		 1291{
1292 struct mlx4_priv *priv = mlx4_priv(dev);
1293 int err;
1294
1295 priv->fw.fw_icm = mlx4_alloc_icm(dev, priv->fw.fw_pages,
5b0bf5e2 1296 GFP_HIGHUSER | __GFP_NOWARN, 0);
225c7b1f 1297 if (!priv->fw.fw_icm) {
1a91de28 1298 mlx4_err(dev, "Couldn't allocate FW area, aborting\n");
225c7b1f
RD		 1299 return -ENOMEM;
1300 }
1301
1302 err = mlx4_MAP_FA(dev, priv->fw.fw_icm);
1303 if (err) {
1a91de28 1304 mlx4_err(dev, "MAP_FA command failed, aborting\n");
225c7b1f
RD		 1305 goto err_free;
1306 }
1307
1308 err = mlx4_RUN_FW(dev);
1309 if (err) {
1a91de28 1310 mlx4_err(dev, "RUN_FW command failed, aborting\n");
225c7b1f
RD		 1311 goto err_unmap_fa;
1312 }
1313
1314 return 0;
1315
1316err_unmap_fa:
1317 mlx4_UNMAP_FA(dev);
1318
1319err_free:
5b0bf5e2 1320 mlx4_free_icm(dev, priv->fw.fw_icm, 0);
225c7b1f
RD		 1321 return err;
1322}
1323
e8f9b2ed
RD		 1324static int mlx4_init_cmpt_table(struct mlx4_dev *dev, u64 cmpt_base,
1325 int cmpt_entry_sz)
225c7b1f
RD		 1326{
1327 struct mlx4_priv *priv = mlx4_priv(dev);
1328 int err;
ab9c17a0 1329 int num_eqs;
225c7b1f
RD		 1330
1331 err = mlx4_init_icm_table(dev, &priv->qp_table.cmpt_table,
1332 cmpt_base +
1333 ((u64) (MLX4_CMPT_TYPE_QP *
1334 cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1335 cmpt_entry_sz, dev->caps.num_qps,
93fc9e1b
YP		 1336 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1337 0, 0);
225c7b1f
RD		 1338 if (err)
1339 goto err;
1340
1341 err = mlx4_init_icm_table(dev, &priv->srq_table.cmpt_table,
1342 cmpt_base +
1343 ((u64) (MLX4_CMPT_TYPE_SRQ *
1344 cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1345 cmpt_entry_sz, dev->caps.num_srqs,
5b0bf5e2 1346 dev->caps.reserved_srqs, 0, 0);
225c7b1f
RD		 1347 if (err)
1348 goto err_qp;
1349
1350 err = mlx4_init_icm_table(dev, &priv->cq_table.cmpt_table,
1351 cmpt_base +
1352 ((u64) (MLX4_CMPT_TYPE_CQ *
1353 cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1354 cmpt_entry_sz, dev->caps.num_cqs,
5b0bf5e2 1355 dev->caps.reserved_cqs, 0, 0);
225c7b1f
RD		 1356 if (err)
1357 goto err_srq;
1358
7ae0e400 1359 num_eqs = dev->phys_caps.num_phys_eqs;
225c7b1f
RD		 1360 err = mlx4_init_icm_table(dev, &priv->eq_table.cmpt_table,
1361 cmpt_base +
1362 ((u64) (MLX4_CMPT_TYPE_EQ *
1363 cmpt_entry_sz) << MLX4_CMPT_SHIFT),
ab9c17a0 1364 cmpt_entry_sz, num_eqs, num_eqs, 0, 0);
225c7b1f
RD		 1365 if (err)
1366 goto err_cq;
1367
1368 return 0;
1369
1370err_cq:
1371 mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1372
1373err_srq:
1374 mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1375
1376err_qp:
1377 mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1378
1379err:
1380 return err;
1381}
1382
3d73c288
RD		 1383static int mlx4_init_icm(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap,
1384 struct mlx4_init_hca_param *init_hca, u64 icm_size)
225c7b1f
RD		 1385{
1386 struct mlx4_priv *priv = mlx4_priv(dev);
1387 u64 aux_pages;
ab9c17a0 1388 int num_eqs;
225c7b1f
RD		 1389 int err;
1390
1391 err = mlx4_SET_ICM_SIZE(dev, icm_size, &aux_pages);
1392 if (err) {
1a91de28 1393 mlx4_err(dev, "SET_ICM_SIZE command failed, aborting\n");
225c7b1f
RD		 1394 return err;
1395 }
1396
1a91de28 1397 mlx4_dbg(dev, "%lld KB of HCA context requires %lld KB aux memory\n",
225c7b1f
RD		 1398 (unsigned long long) icm_size >> 10,
1399 (unsigned long long) aux_pages << 2);
1400
1401 priv->fw.aux_icm = mlx4_alloc_icm(dev, aux_pages,
5b0bf5e2 1402 GFP_HIGHUSER | __GFP_NOWARN, 0);
225c7b1f 1403 if (!priv->fw.aux_icm) {
1a91de28 1404 mlx4_err(dev, "Couldn't allocate aux memory, aborting\n");
225c7b1f
RD		 1405 return -ENOMEM;
1406 }
1407
1408 err = mlx4_MAP_ICM_AUX(dev, priv->fw.aux_icm);
1409 if (err) {
1a91de28 1410 mlx4_err(dev, "MAP_ICM_AUX command failed, aborting\n");
225c7b1f
RD		 1411 goto err_free_aux;
1412 }
1413
1414 err = mlx4_init_cmpt_table(dev, init_hca->cmpt_base, dev_cap->cmpt_entry_sz);
1415 if (err) {
1a91de28 1416 mlx4_err(dev, "Failed to map cMPT context memory, aborting\n");
225c7b1f
RD		 1417 goto err_unmap_aux;
1418 }
1419
ab9c17a0 1420
7ae0e400 1421 num_eqs = dev->phys_caps.num_phys_eqs;
fa0681d2
RD		 1422 err = mlx4_init_icm_table(dev, &priv->eq_table.table,
1423 init_hca->eqc_base, dev_cap->eqc_entry_sz,
ab9c17a0 1424 num_eqs, num_eqs, 0, 0);
225c7b1f 1425 if (err) {
1a91de28 1426 mlx4_err(dev, "Failed to map EQ context memory, aborting\n");
225c7b1f
RD		 1427 goto err_unmap_cmpt;
1428 }
1429
d7bb58fb
JM		 1430 /*
1431 * Reserved MTT entries must be aligned up to a cacheline
1432 * boundary, since the FW will write to them, while the driver
1433 * writes to all other MTT entries. (The variable
1434 * dev->caps.mtt_entry_sz below is really the MTT segment
1435 * size, not the raw entry size)
1436 */
1437 dev->caps.reserved_mtts =
1438 ALIGN(dev->caps.reserved_mtts * dev->caps.mtt_entry_sz,
1439 dma_get_cache_alignment()) / dev->caps.mtt_entry_sz;
1440
225c7b1f
RD		 1441 err = mlx4_init_icm_table(dev, &priv->mr_table.mtt_table,
1442 init_hca->mtt_base,
1443 dev->caps.mtt_entry_sz,
2b8fb286 1444 dev->caps.num_mtts,
5b0bf5e2 1445 dev->caps.reserved_mtts, 1, 0);
225c7b1f 1446 if (err) {
1a91de28 1447 mlx4_err(dev, "Failed to map MTT context memory, aborting\n");
225c7b1f
RD		 1448 goto err_unmap_eq;
1449 }
1450
1451 err = mlx4_init_icm_table(dev, &priv->mr_table.dmpt_table,
1452 init_hca->dmpt_base,
1453 dev_cap->dmpt_entry_sz,
1454 dev->caps.num_mpts,
5b0bf5e2 1455 dev->caps.reserved_mrws, 1, 1);
225c7b1f 1456 if (err) {
1a91de28 1457 mlx4_err(dev, "Failed to map dMPT context memory, aborting\n");
225c7b1f
RD		 1458 goto err_unmap_mtt;
1459 }
1460
1461 err = mlx4_init_icm_table(dev, &priv->qp_table.qp_table,
1462 init_hca->qpc_base,
1463 dev_cap->qpc_entry_sz,
1464 dev->caps.num_qps,
93fc9e1b
YP		 1465 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1466 0, 0);
225c7b1f 1467 if (err) {
1a91de28 1468 mlx4_err(dev, "Failed to map QP context memory, aborting\n");
225c7b1f
RD		 1469 goto err_unmap_dmpt;
1470 }
1471
1472 err = mlx4_init_icm_table(dev, &priv->qp_table.auxc_table,
1473 init_hca->auxc_base,
1474 dev_cap->aux_entry_sz,
1475 dev->caps.num_qps,
93fc9e1b
YP		 1476 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1477 0, 0);
225c7b1f 1478 if (err) {
1a91de28 1479 mlx4_err(dev, "Failed to map AUXC context memory, aborting\n");
225c7b1f
RD		 1480 goto err_unmap_qp;
1481 }
1482
1483 err = mlx4_init_icm_table(dev, &priv->qp_table.altc_table,
1484 init_hca->altc_base,
1485 dev_cap->altc_entry_sz,
1486 dev->caps.num_qps,
93fc9e1b
YP		 1487 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1488 0, 0);
225c7b1f 1489 if (err) {
1a91de28 1490 mlx4_err(dev, "Failed to map ALTC context memory, aborting\n");
225c7b1f
RD		 1491 goto err_unmap_auxc;
1492 }
1493
1494 err = mlx4_init_icm_table(dev, &priv->qp_table.rdmarc_table,
1495 init_hca->rdmarc_base,
1496 dev_cap->rdmarc_entry_sz << priv->qp_table.rdmarc_shift,
1497 dev->caps.num_qps,
93fc9e1b
YP		 1498 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1499 0, 0);
225c7b1f
RD		 1500 if (err) {
1501 mlx4_err(dev, "Failed to map RDMARC context memory, aborting\n");
1502 goto err_unmap_altc;
1503 }
1504
1505 err = mlx4_init_icm_table(dev, &priv->cq_table.table,
1506 init_hca->cqc_base,
1507 dev_cap->cqc_entry_sz,
1508 dev->caps.num_cqs,
5b0bf5e2 1509 dev->caps.reserved_cqs, 0, 0);
225c7b1f 1510 if (err) {
1a91de28 1511 mlx4_err(dev, "Failed to map CQ context memory, aborting\n");
225c7b1f
RD		 1512 goto err_unmap_rdmarc;
1513 }
1514
1515 err = mlx4_init_icm_table(dev, &priv->srq_table.table,
1516 init_hca->srqc_base,
1517 dev_cap->srq_entry_sz,
1518 dev->caps.num_srqs,
5b0bf5e2 1519 dev->caps.reserved_srqs, 0, 0);
225c7b1f 1520 if (err) {
1a91de28 1521 mlx4_err(dev, "Failed to map SRQ context memory, aborting\n");
225c7b1f
RD		 1522 goto err_unmap_cq;
1523 }
1524
1525 /*
0ff1fb65
HHZ		 1526 * For flow steering device managed mode it is required to use
1527 * mlx4_init_icm_table. For B0 steering mode it's not strictly
1528 * required, but for simplicity just map the whole multicast
1529 * group table now. The table isn't very big and it's a lot
1530 * easier than trying to track ref counts.
225c7b1f
RD		 1531 */
1532 err = mlx4_init_icm_table(dev, &priv->mcg_table.table,
0ec2c0f8
EE		 1533 init_hca->mc_base,
1534 mlx4_get_mgm_entry_size(dev),
225c7b1f
RD		 1535 dev->caps.num_mgms + dev->caps.num_amgms,
1536 dev->caps.num_mgms + dev->caps.num_amgms,
5b0bf5e2 1537 0, 0);
225c7b1f 1538 if (err) {
1a91de28 1539 mlx4_err(dev, "Failed to map MCG context memory, aborting\n");
225c7b1f
RD		 1540 goto err_unmap_srq;
1541 }
1542
1543 return 0;
1544
1545err_unmap_srq:
1546 mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1547
1548err_unmap_cq:
1549 mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1550
1551err_unmap_rdmarc:
1552 mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1553
1554err_unmap_altc:
1555 mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1556
1557err_unmap_auxc:
1558 mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1559
1560err_unmap_qp:
1561 mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1562
1563err_unmap_dmpt:
1564 mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1565
1566err_unmap_mtt:
1567 mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
1568
1569err_unmap_eq:
fa0681d2 1570 mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
225c7b1f
RD		 1571
1572err_unmap_cmpt:
1573 mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1574 mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1575 mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1576 mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1577
1578err_unmap_aux:
1579 mlx4_UNMAP_ICM_AUX(dev);
1580
1581err_free_aux:
5b0bf5e2 1582 mlx4_free_icm(dev, priv->fw.aux_icm, 0);
225c7b1f
RD		 1583
1584 return err;
1585}
1586
1587static void mlx4_free_icms(struct mlx4_dev *dev)
1588{
1589 struct mlx4_priv *priv = mlx4_priv(dev);
1590
1591 mlx4_cleanup_icm_table(dev, &priv->mcg_table.table);
1592 mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1593 mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1594 mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1595 mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1596 mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1597 mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1598 mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1599 mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
fa0681d2 1600 mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
225c7b1f
RD		 1601 mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1602 mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1603 mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1604 mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
225c7b1f
RD		 1605
1606 mlx4_UNMAP_ICM_AUX(dev);
5b0bf5e2 1607 mlx4_free_icm(dev, priv->fw.aux_icm, 0);
225c7b1f
RD		 1608}
1609
ab9c17a0
JM		 1610static void mlx4_slave_exit(struct mlx4_dev *dev)
1611{
1612 struct mlx4_priv *priv = mlx4_priv(dev);
1613
f3d4c89e 1614 mutex_lock(&priv->cmd.slave_cmd_mutex);
0cd93027
YH		 1615 if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, MLX4_COMM_CMD_NA_OP,
1616 MLX4_COMM_TIME))
1a91de28 1617 mlx4_warn(dev, "Failed to close slave function\n");
f3d4c89e 1618 mutex_unlock(&priv->cmd.slave_cmd_mutex);
ab9c17a0
JM		 1619}
1620
c1b43dca
EC		 1621static int map_bf_area(struct mlx4_dev *dev)
1622{
1623 struct mlx4_priv *priv = mlx4_priv(dev);
1624 resource_size_t bf_start;
1625 resource_size_t bf_len;
1626 int err = 0;
1627
3d747473
JM		 1628 if (!dev->caps.bf_reg_size)
1629 return -ENXIO;
1630
872bf2fb 1631 bf_start = pci_resource_start(dev->persist->pdev, 2) +
ab9c17a0 1632 (dev->caps.num_uars << PAGE_SHIFT);
872bf2fb 1633 bf_len = pci_resource_len(dev->persist->pdev, 2) -
ab9c17a0 1634 (dev->caps.num_uars << PAGE_SHIFT);
c1b43dca
EC		 1635 priv->bf_mapping = io_mapping_create_wc(bf_start, bf_len);
1636 if (!priv->bf_mapping)
1637 err = -ENOMEM;
1638
1639 return err;
1640}
1641
1642static void unmap_bf_area(struct mlx4_dev *dev)
1643{
1644 if (mlx4_priv(dev)->bf_mapping)
1645 io_mapping_free(mlx4_priv(dev)->bf_mapping);
1646}
1647
ec693d47
AV		 1648cycle_t mlx4_read_clock(struct mlx4_dev *dev)
1649{
1650 u32 clockhi, clocklo, clockhi1;
1651 cycle_t cycles;
1652 int i;
1653 struct mlx4_priv *priv = mlx4_priv(dev);
1654
1655 for (i = 0; i < 10; i++) {
1656 clockhi = swab32(readl(priv->clock_mapping));
1657 clocklo = swab32(readl(priv->clock_mapping + 4));
1658 clockhi1 = swab32(readl(priv->clock_mapping));
1659 if (clockhi == clockhi1)
1660 break;
1661 }
1662
1663 cycles = (u64) clockhi << 32 | (u64) clocklo;
1664
1665 return cycles;
1666}
1667EXPORT_SYMBOL_GPL(mlx4_read_clock);
1668
1669
ddd8a6c1
EE		 1670static int map_internal_clock(struct mlx4_dev *dev)
1671{
1672 struct mlx4_priv *priv = mlx4_priv(dev);
1673
1674 priv->clock_mapping =
872bf2fb
YH		 1675 ioremap(pci_resource_start(dev->persist->pdev,
1676 priv->fw.clock_bar) +
ddd8a6c1
EE		 1677 priv->fw.clock_offset, MLX4_CLOCK_SIZE);
1678
1679 if (!priv->clock_mapping)
1680 return -ENOMEM;
1681
1682 return 0;
1683}
1684
52033cfb
MB		 1685int mlx4_get_internal_clock_params(struct mlx4_dev *dev,
1686 struct mlx4_clock_params *params)
1687{
1688 struct mlx4_priv *priv = mlx4_priv(dev);
1689
1690 if (mlx4_is_slave(dev))
1691 return -ENOTSUPP;
1692
1693 if (!params)
1694 return -EINVAL;
1695
1696 params->bar = priv->fw.clock_bar;
1697 params->offset = priv->fw.clock_offset;
1698 params->size = MLX4_CLOCK_SIZE;
1699
1700 return 0;
1701}
1702EXPORT_SYMBOL_GPL(mlx4_get_internal_clock_params);
1703
ddd8a6c1
EE		 1704static void unmap_internal_clock(struct mlx4_dev *dev)
1705{
1706 struct mlx4_priv *priv = mlx4_priv(dev);
1707
1708 if (priv->clock_mapping)
1709 iounmap(priv->clock_mapping);
1710}
1711
225c7b1f
RD		 1712static void mlx4_close_hca(struct mlx4_dev *dev)
1713{
ddd8a6c1 1714 unmap_internal_clock(dev);
c1b43dca 1715 unmap_bf_area(dev);
ab9c17a0
JM		 1716 if (mlx4_is_slave(dev))
1717 mlx4_slave_exit(dev);
1718 else {
1719 mlx4_CLOSE_HCA(dev, 0);
1720 mlx4_free_icms(dev);
a0eacca9
MB		 1721 }
1722}
1723
1724static void mlx4_close_fw(struct mlx4_dev *dev)
1725{
1726 if (!mlx4_is_slave(dev)) {
ab9c17a0
JM		 1727 mlx4_UNMAP_FA(dev);
1728 mlx4_free_icm(dev, mlx4_priv(dev)->fw.fw_icm, 0);
1729 }
1730}
1731
55ad3592
YH		 1732static int mlx4_comm_check_offline(struct mlx4_dev *dev)
1733{
1734#define COMM_CHAN_OFFLINE_OFFSET 0x09
1735
1736 u32 comm_flags;
1737 u32 offline_bit;
1738 unsigned long end;
1739 struct mlx4_priv *priv = mlx4_priv(dev);
1740
1741 end = msecs_to_jiffies(MLX4_COMM_OFFLINE_TIME_OUT) + jiffies;
1742 while (time_before(jiffies, end)) {
1743 comm_flags = swab32(readl((__iomem char *)priv->mfunc.comm +
1744 MLX4_COMM_CHAN_FLAGS));
1745 offline_bit = (comm_flags &
1746 (u32)(1 << COMM_CHAN_OFFLINE_OFFSET));
1747 if (!offline_bit)
1748 return 0;
1749 /* There are cases as part of AER/Reset flow that PF needs
1750 * around 100 msec to load. We therefore sleep for 100 msec
1751 * to allow other tasks to make use of that CPU during this
1752 * time interval.
1753 */
1754 msleep(100);
1755 }
1756 mlx4_err(dev, "Communication channel is offline.\n");
1757 return -EIO;
1758}
1759
1760static void mlx4_reset_vf_support(struct mlx4_dev *dev)
1761{
1762#define COMM_CHAN_RST_OFFSET 0x1e
1763
1764 struct mlx4_priv *priv = mlx4_priv(dev);
1765 u32 comm_rst;
1766 u32 comm_caps;
1767
1768 comm_caps = swab32(readl((__iomem char *)priv->mfunc.comm +
1769 MLX4_COMM_CHAN_CAPS));
1770 comm_rst = (comm_caps & (u32)(1 << COMM_CHAN_RST_OFFSET));
1771
1772 if (comm_rst)
1773 dev->caps.vf_caps |= MLX4_VF_CAP_FLAG_RESET;
1774}
1775
ab9c17a0
JM		 1776static int mlx4_init_slave(struct mlx4_dev *dev)
1777{
1778 struct mlx4_priv *priv = mlx4_priv(dev);
1779 u64 dma = (u64) priv->mfunc.vhcr_dma;
ab9c17a0
JM		 1780 int ret_from_reset = 0;
1781 u32 slave_read;
1782 u32 cmd_channel_ver;
1783
97989356 1784 if (atomic_read(&pf_loading)) {
1a91de28 1785 mlx4_warn(dev, "PF is not ready - Deferring probe\n");
97989356
AV		 1786 return -EPROBE_DEFER;
1787 }
1788
f3d4c89e 1789 mutex_lock(&priv->cmd.slave_cmd_mutex);
ab9c17a0 1790 priv->cmd.max_cmds = 1;
55ad3592
YH		 1791 if (mlx4_comm_check_offline(dev)) {
1792 mlx4_err(dev, "PF is not responsive, skipping initialization\n");
1793 goto err_offline;
1794 }
1795
1796 mlx4_reset_vf_support(dev);
ab9c17a0
JM		 1797 mlx4_warn(dev, "Sending reset\n");
1798 ret_from_reset = mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0,
0cd93027 1799 MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME);
ab9c17a0
JM		 1800 /* if we are in the middle of flr the slave will try
1801 * NUM_OF_RESET_RETRIES times before leaving.*/
1802 if (ret_from_reset) {
1803 if (MLX4_DELAY_RESET_SLAVE == ret_from_reset) {
1a91de28 1804 mlx4_warn(dev, "slave is currently in the middle of FLR - Deferring probe\n");
5efe5355
JM		 1805 mutex_unlock(&priv->cmd.slave_cmd_mutex);
1806 return -EPROBE_DEFER;
ab9c17a0
JM		 1807 } else
1808 goto err;
1809 }
1810
1811 /* check the driver version - the slave I/F revision
1812 * must match the master's */
1813 slave_read = swab32(readl(&priv->mfunc.comm->slave_read));
1814 cmd_channel_ver = mlx4_comm_get_version();
1815
1816 if (MLX4_COMM_GET_IF_REV(cmd_channel_ver) !=
1817 MLX4_COMM_GET_IF_REV(slave_read)) {
1a91de28 1818 mlx4_err(dev, "slave driver version is not supported by the master\n");
ab9c17a0
JM		 1819 goto err;
1820 }
1821
1822 mlx4_warn(dev, "Sending vhcr0\n");
1823 if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR0, dma >> 48,
0cd93027 1824 MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
ab9c17a0
JM		 1825 goto err;
1826 if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR1, dma >> 32,
0cd93027 1827 MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
ab9c17a0
JM		 1828 goto err;
1829 if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR2, dma >> 16,
0cd93027 1830 MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
ab9c17a0 1831 goto err;
0cd93027
YH		 1832 if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR_EN, dma,
1833 MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
ab9c17a0 1834 goto err;
f3d4c89e
RD		 1835
1836 mutex_unlock(&priv->cmd.slave_cmd_mutex);
ab9c17a0
JM		 1837 return 0;
1838
1839err:
0cd93027 1840 mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, MLX4_COMM_CMD_NA_OP, 0);
55ad3592 1841err_offline:
f3d4c89e 1842 mutex_unlock(&priv->cmd.slave_cmd_mutex);
ab9c17a0 1843 return -EIO;
225c7b1f
RD		 1844}
1845
6634961c
JM		 1846static void mlx4_parav_master_pf_caps(struct mlx4_dev *dev)
1847{
1848 int i;
1849
1850 for (i = 1; i <= dev->caps.num_ports; i++) {
b6ffaeff
JM		 1851 if (dev->caps.port_type[i] == MLX4_PORT_TYPE_ETH)
1852 dev->caps.gid_table_len[i] =
449fc488 1853 mlx4_get_slave_num_gids(dev, 0, i);
b6ffaeff
JM		 1854 else
1855 dev->caps.gid_table_len[i] = 1;
6634961c
JM		 1856 dev->caps.pkey_table_len[i] =
1857 dev->phys_caps.pkey_phys_table_len[i] - 1;
1858 }
1859}
1860
3c439b55
JM		 1861static int choose_log_fs_mgm_entry_size(int qp_per_entry)
1862{
1863 int i = MLX4_MIN_MGM_LOG_ENTRY_SIZE;
1864
1865 for (i = MLX4_MIN_MGM_LOG_ENTRY_SIZE; i <= MLX4_MAX_MGM_LOG_ENTRY_SIZE;
1866 i++) {
1867 if (qp_per_entry <= 4 * ((1 << i) / 16 - 2))
1868 break;
1869 }
1870
1871 return (i <= MLX4_MAX_MGM_LOG_ENTRY_SIZE) ? i : -1;
1872}
1873
7d077cd3
MB		 1874static const char *dmfs_high_rate_steering_mode_str(int dmfs_high_steer_mode)
1875{
1876 switch (dmfs_high_steer_mode) {
1877 case MLX4_STEERING_DMFS_A0_DEFAULT:
1878 return "default performance";
1879
1880 case MLX4_STEERING_DMFS_A0_DYNAMIC:
1881 return "dynamic hybrid mode";
1882
1883 case MLX4_STEERING_DMFS_A0_STATIC:
1884 return "performance optimized for limited rule configuration (static)";
1885
1886 case MLX4_STEERING_DMFS_A0_DISABLE:
1887 return "disabled performance optimized steering";
1888
1889 case MLX4_STEERING_DMFS_A0_NOT_SUPPORTED:
1890 return "performance optimized steering not supported";
1891
1892 default:
1893 return "Unrecognized mode";
1894 }
1895}
1896
1897#define MLX4_DMFS_A0_STEERING (1UL << 2)
1898
7b8157be
JM		 1899static void choose_steering_mode(struct mlx4_dev *dev,
1900 struct mlx4_dev_cap *dev_cap)
1901{
7d077cd3
MB		 1902 if (mlx4_log_num_mgm_entry_size <= 0) {
1903 if ((-mlx4_log_num_mgm_entry_size) & MLX4_DMFS_A0_STEERING) {
1904 if (dev->caps.dmfs_high_steer_mode ==
1905 MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
1906 mlx4_err(dev, "DMFS high rate mode not supported\n");
1907 else
1908 dev->caps.dmfs_high_steer_mode =
1909 MLX4_STEERING_DMFS_A0_STATIC;
1910 }
1911 }
1912
1913 if (mlx4_log_num_mgm_entry_size <= 0 &&
3c439b55 1914 dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_FS_EN &&
7b8157be 1915 (!mlx4_is_mfunc(dev) ||
872bf2fb
YH		 1916 (dev_cap->fs_max_num_qp_per_entry >=
1917 (dev->persist->num_vfs + 1))) &&
3c439b55
JM		 1918 choose_log_fs_mgm_entry_size(dev_cap->fs_max_num_qp_per_entry) >=
1919 MLX4_MIN_MGM_LOG_ENTRY_SIZE) {
1920 dev->oper_log_mgm_entry_size =
1921 choose_log_fs_mgm_entry_size(dev_cap->fs_max_num_qp_per_entry);
7b8157be
JM		 1922 dev->caps.steering_mode = MLX4_STEERING_MODE_DEVICE_MANAGED;
1923 dev->caps.num_qp_per_mgm = dev_cap->fs_max_num_qp_per_entry;
1924 dev->caps.fs_log_max_ucast_qp_range_size =
1925 dev_cap->fs_log_max_ucast_qp_range_size;
1926 } else {
7d077cd3
MB		 1927 if (dev->caps.dmfs_high_steer_mode !=
1928 MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
1929 dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_DISABLE;
7b8157be
JM		 1930 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER &&
1931 dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER)
1932 dev->caps.steering_mode = MLX4_STEERING_MODE_B0;
1933 else {
1934 dev->caps.steering_mode = MLX4_STEERING_MODE_A0;
1935
1936 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER ||
1937 dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER)
1a91de28 1938 mlx4_warn(dev, "Must have both UC_STEER and MC_STEER flags set to use B0 steering - falling back to A0 steering mode\n");
7b8157be 1939 }
3c439b55
JM		 1940 dev->oper_log_mgm_entry_size =
1941 mlx4_log_num_mgm_entry_size > 0 ?
1942 mlx4_log_num_mgm_entry_size :
1943 MLX4_DEFAULT_MGM_LOG_ENTRY_SIZE;
7b8157be
JM		 1944 dev->caps.num_qp_per_mgm = mlx4_get_qp_per_mgm(dev);
1945 }
1a91de28 1946 mlx4_dbg(dev, "Steering mode is: %s, oper_log_mgm_entry_size = %d, modparam log_num_mgm_entry_size = %d\n",
3c439b55
JM		 1947 mlx4_steering_mode_str(dev->caps.steering_mode),
1948 dev->oper_log_mgm_entry_size,
1949 mlx4_log_num_mgm_entry_size);
7b8157be
JM		 1950}
1951
7ffdf726
OG		 1952static void choose_tunnel_offload_mode(struct mlx4_dev *dev,
1953 struct mlx4_dev_cap *dev_cap)
1954{
1955 if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED &&
5eff6dad 1956 dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_VXLAN_OFFLOADS)
7ffdf726
OG		 1957 dev->caps.tunnel_offload_mode = MLX4_TUNNEL_OFFLOAD_MODE_VXLAN;
1958 else
1959 dev->caps.tunnel_offload_mode = MLX4_TUNNEL_OFFLOAD_MODE_NONE;
1960
1961 mlx4_dbg(dev, "Tunneling offload mode is: %s\n", (dev->caps.tunnel_offload_mode
1962 == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) ? "vxlan" : "none");
1963}
1964
7d077cd3
MB		 1965static int mlx4_validate_optimized_steering(struct mlx4_dev *dev)
1966{
1967 int i;
1968 struct mlx4_port_cap port_cap;
1969
1970 if (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
1971 return -EINVAL;
1972
1973 for (i = 1; i <= dev->caps.num_ports; i++) {
1974 if (mlx4_dev_port(dev, i, &port_cap)) {
1975 mlx4_err(dev,
1976 "QUERY_DEV_CAP command failed, can't veify DMFS high rate steering.\n");
1977 } else if ((dev->caps.dmfs_high_steer_mode !=
1978 MLX4_STEERING_DMFS_A0_DEFAULT) &&
1979 (port_cap.dmfs_optimized_state ==
1980 !!(dev->caps.dmfs_high_steer_mode ==
1981 MLX4_STEERING_DMFS_A0_DISABLE))) {
1982 mlx4_err(dev,
1983 "DMFS high rate steer mode differ, driver requested %s but %s in FW.\n",
1984 dmfs_high_rate_steering_mode_str(
1985 dev->caps.dmfs_high_steer_mode),
1986 (port_cap.dmfs_optimized_state ?
1987 "enabled" : "disabled"));
1988 }
1989 }
1990
1991 return 0;
1992}
1993
a0eacca9 1994static int mlx4_init_fw(struct mlx4_dev *dev)
225c7b1f 1995{
2d928651 1996 struct mlx4_mod_stat_cfg mlx4_cfg;
a0eacca9 1997 int err = 0;
225c7b1f 1998
ab9c17a0
JM		 1999 if (!mlx4_is_slave(dev)) {
2000 err = mlx4_QUERY_FW(dev);
2001 if (err) {
2002 if (err == -EACCES)
1a91de28 2003 mlx4_info(dev, "non-primary physical function, skipping\n");
ab9c17a0 2004 else
1a91de28 2005 mlx4_err(dev, "QUERY_FW command failed, aborting\n");
bef772eb 2006 return err;
ab9c17a0 2007 }
225c7b1f 2008
ab9c17a0
JM		 2009 err = mlx4_load_fw(dev);
2010 if (err) {
1a91de28 2011 mlx4_err(dev, "Failed to start FW, aborting\n");
bef772eb 2012 return err;
ab9c17a0 2013 }
225c7b1f 2014
ab9c17a0
JM		 2015 mlx4_cfg.log_pg_sz_m = 1;
2016 mlx4_cfg.log_pg_sz = 0;
2017 err = mlx4_MOD_STAT_CFG(dev, &mlx4_cfg);
2018 if (err)
2019 mlx4_warn(dev, "Failed to override log_pg_sz parameter\n");
a0eacca9 2020 }
2d928651 2021
a0eacca9
MB		 2022 return err;
2023}
2024
2025static int mlx4_init_hca(struct mlx4_dev *dev)
2026{
2027 struct mlx4_priv *priv = mlx4_priv(dev);
2028 struct mlx4_adapter adapter;
2029 struct mlx4_dev_cap dev_cap;
2030 struct mlx4_profile profile;
2031 struct mlx4_init_hca_param init_hca;
2032 u64 icm_size;
2033 struct mlx4_config_dev_params params;
2034 int err;
2035
2036 if (!mlx4_is_slave(dev)) {
ab9c17a0
JM		 2037 err = mlx4_dev_cap(dev, &dev_cap);
2038 if (err) {
1a91de28 2039 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
d0d01250 2040 return err;
ab9c17a0 2041 }
225c7b1f 2042
7b8157be 2043 choose_steering_mode(dev, &dev_cap);
7ffdf726 2044 choose_tunnel_offload_mode(dev, &dev_cap);
7b8157be 2045
7d077cd3
MB		 2046 if (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC &&
2047 mlx4_is_master(dev))
2048 dev->caps.function_caps |= MLX4_FUNC_CAP_DMFS_A0_STATIC;
2049
8e1a28e8
HHZ		 2050 err = mlx4_get_phys_port_id(dev);
2051 if (err)
2052 mlx4_err(dev, "Fail to get physical port id\n");
2053
6634961c
JM		 2054 if (mlx4_is_master(dev))
2055 mlx4_parav_master_pf_caps(dev);
2056
2599d858
AV		 2057 if (mlx4_low_memory_profile()) {
2058 mlx4_info(dev, "Running from within kdump kernel. Using low memory profile\n");
2059 profile = low_mem_profile;
2060 } else {
2061 profile = default_profile;
2062 }
0ff1fb65
HHZ		 2063 if (dev->caps.steering_mode ==
2064 MLX4_STEERING_MODE_DEVICE_MANAGED)
2065 profile.num_mcg = MLX4_FS_NUM_MCG;
225c7b1f 2066
ab9c17a0
JM		 2067 icm_size = mlx4_make_profile(dev, &profile, &dev_cap,
2068 &init_hca);
2069 if ((long long) icm_size < 0) {
2070 err = icm_size;
d0d01250 2071 return err;
ab9c17a0 2072 }
225c7b1f 2073
a5bbe892
EC		 2074 dev->caps.max_fmr_maps = (1 << (32 - ilog2(dev->caps.num_mpts))) - 1;
2075
ab9c17a0
JM		 2076 init_hca.log_uar_sz = ilog2(dev->caps.num_uars);
2077 init_hca.uar_page_sz = PAGE_SHIFT - 12;
e448834e
SM		 2078 init_hca.mw_enabled = 0;
2079 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2080 dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN)
2081 init_hca.mw_enabled = INIT_HCA_TPT_MW_ENABLE;
c1b43dca 2082
ab9c17a0
JM		 2083 err = mlx4_init_icm(dev, &dev_cap, &init_hca, icm_size);
2084 if (err)
d0d01250 2085 return err;
225c7b1f 2086
ab9c17a0
JM		 2087 err = mlx4_INIT_HCA(dev, &init_hca);
2088 if (err) {
1a91de28 2089 mlx4_err(dev, "INIT_HCA command failed, aborting\n");
ab9c17a0
JM		 2090 goto err_free_icm;
2091 }
7ae0e400
MB		 2092
2093 if (dev_cap.flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS) {
2094 err = mlx4_query_func(dev, &dev_cap);
2095 if (err < 0) {
2096 mlx4_err(dev, "QUERY_FUNC command failed, aborting.\n");
d0d01250 2097 goto err_close;
7ae0e400
MB		 2098 } else if (err & MLX4_QUERY_FUNC_NUM_SYS_EQS) {
2099 dev->caps.num_eqs = dev_cap.max_eqs;
2100 dev->caps.reserved_eqs = dev_cap.reserved_eqs;
2101 dev->caps.reserved_uars = dev_cap.reserved_uars;
2102 }
2103 }
2104
ddd8a6c1
EE		 2105 /*
2106 * If TS is supported by FW
2107 * read HCA frequency by QUERY_HCA command
2108 */
2109 if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS) {
2110 memset(&init_hca, 0, sizeof(init_hca));
2111 err = mlx4_QUERY_HCA(dev, &init_hca);
2112 if (err) {
1a91de28 2113 mlx4_err(dev, "QUERY_HCA command failed, disable timestamp\n");
ddd8a6c1
EE		 2114 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
2115 } else {
2116 dev->caps.hca_core_clock =
2117 init_hca.hca_core_clock;
2118 }
2119
2120 /* In case we got HCA frequency 0 - disable timestamping
2121 * to avoid dividing by zero
2122 */
2123 if (!dev->caps.hca_core_clock) {
2124 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
2125 mlx4_err(dev,
1a91de28 2126 "HCA frequency is 0 - timestamping is not supported\n");
ddd8a6c1
EE		 2127 } else if (map_internal_clock(dev)) {
2128 /*
2129 * Map internal clock,
2130 * in case of failure disable timestamping
2131 */
2132 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
1a91de28 2133 mlx4_err(dev, "Failed to map internal clock. Timestamping is not supported\n");
ddd8a6c1
EE		 2134 }
2135 }
7d077cd3
MB		 2136
2137 if (dev->caps.dmfs_high_steer_mode !=
2138 MLX4_STEERING_DMFS_A0_NOT_SUPPORTED) {
2139 if (mlx4_validate_optimized_steering(dev))
2140 mlx4_warn(dev, "Optimized steering validation failed\n");
2141
2142 if (dev->caps.dmfs_high_steer_mode ==
2143 MLX4_STEERING_DMFS_A0_DISABLE) {
2144 dev->caps.dmfs_high_rate_qpn_base =
2145 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
2146 dev->caps.dmfs_high_rate_qpn_range =
2147 MLX4_A0_STEERING_TABLE_SIZE;
2148 }
2149
2150 mlx4_dbg(dev, "DMFS high rate steer mode is: %s\n",
2151 dmfs_high_rate_steering_mode_str(
2152 dev->caps.dmfs_high_steer_mode));
2153 }
ab9c17a0
JM		 2154 } else {
2155 err = mlx4_init_slave(dev);
2156 if (err) {
5efe5355
JM		 2157 if (err != -EPROBE_DEFER)
2158 mlx4_err(dev, "Failed to initialize slave\n");
bef772eb 2159 return err;
ab9c17a0 2160 }
225c7b1f 2161
ab9c17a0
JM		 2162 err = mlx4_slave_cap(dev);
2163 if (err) {
2164 mlx4_err(dev, "Failed to obtain slave caps\n");
2165 goto err_close;
2166 }
225c7b1f
RD		 2167 }
2168
ab9c17a0
JM		 2169 if (map_bf_area(dev))
2170 mlx4_dbg(dev, "Failed to map blue flame area\n");
2171
2172 /*Only the master set the ports, all the rest got it from it.*/
2173 if (!mlx4_is_slave(dev))
2174 mlx4_set_port_mask(dev);
2175
225c7b1f
RD		 2176 err = mlx4_QUERY_ADAPTER(dev, &adapter);
2177 if (err) {
1a91de28 2178 mlx4_err(dev, "QUERY_ADAPTER command failed, aborting\n");
bef772eb 2179 goto unmap_bf;
225c7b1f
RD		 2180 }
2181
f8c6455b
SM		 2182 /* Query CONFIG_DEV parameters */
2183 err = mlx4_config_dev_retrieval(dev, &params);
2184 if (err && err != -ENOTSUPP) {
2185 mlx4_err(dev, "Failed to query CONFIG_DEV parameters\n");
2186 } else if (!err) {
2187 dev->caps.rx_checksum_flags_port[1] = params.rx_csum_flags_port_1;
2188 dev->caps.rx_checksum_flags_port[2] = params.rx_csum_flags_port_2;
2189 }
225c7b1f 2190 priv->eq_table.inta_pin = adapter.inta_pin;
cd9281d8 2191 memcpy(dev->board_id, adapter.board_id, sizeof dev->board_id);
225c7b1f
RD		 2192
2193 return 0;
2194
bef772eb 2195unmap_bf:
ddd8a6c1 2196 unmap_internal_clock(dev);
bef772eb
AY		 2197 unmap_bf_area(dev);
2198
b38f2879 2199 if (mlx4_is_slave(dev)) {
99ec41d0 2200 kfree(dev->caps.qp0_qkey);
b38f2879
DB		 2201 kfree(dev->caps.qp0_tunnel);
2202 kfree(dev->caps.qp0_proxy);
2203 kfree(dev->caps.qp1_tunnel);
2204 kfree(dev->caps.qp1_proxy);
2205 }
2206
225c7b1f 2207err_close:
41929ed2
DB		 2208 if (mlx4_is_slave(dev))
2209 mlx4_slave_exit(dev);
2210 else
2211 mlx4_CLOSE_HCA(dev, 0);
225c7b1f
RD		 2212
2213err_free_icm:
ab9c17a0
JM		 2214 if (!mlx4_is_slave(dev))
2215 mlx4_free_icms(dev);
225c7b1f 2216
225c7b1f
RD		 2217 return err;
2218}
2219
f2a3f6a3
OG		 2220static int mlx4_init_counters_table(struct mlx4_dev *dev)
2221{
2222 struct mlx4_priv *priv = mlx4_priv(dev);
47d8417f 2223 int nent_pow2;
f2a3f6a3
OG		 2224
2225 if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2226 return -ENOENT;
2227
2632d18d
EBE		 2228 if (!dev->caps.max_counters)
2229 return -ENOSPC;
2230
47d8417f
EBE		 2231 nent_pow2 = roundup_pow_of_two(dev->caps.max_counters);
2232 /* reserve last counter index for sink counter */
2233 return mlx4_bitmap_init(&priv->counters_bitmap, nent_pow2,
2234 nent_pow2 - 1, 0,
2235 nent_pow2 - dev->caps.max_counters + 1);
f2a3f6a3
OG		 2236}
2237
2238static void mlx4_cleanup_counters_table(struct mlx4_dev *dev)
2239{
efa6bc91
EBE		 2240 if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2241 return;
2242
2632d18d
EBE		 2243 if (!dev->caps.max_counters)
2244 return;
2245
f2a3f6a3
OG		 2246 mlx4_bitmap_cleanup(&mlx4_priv(dev)->counters_bitmap);
2247}
2248
6de5f7f6
EBE		 2249static void mlx4_cleanup_default_counters(struct mlx4_dev *dev)
2250{
2251 struct mlx4_priv *priv = mlx4_priv(dev);
2252 int port;
2253
2254 for (port = 0; port < dev->caps.num_ports; port++)
2255 if (priv->def_counter[port] != -1)
2256 mlx4_counter_free(dev, priv->def_counter[port]);
2257}
2258
2259static int mlx4_allocate_default_counters(struct mlx4_dev *dev)
2260{
2261 struct mlx4_priv *priv = mlx4_priv(dev);
2262 int port, err = 0;
2263 u32 idx;
2264
2265 for (port = 0; port < dev->caps.num_ports; port++)
2266 priv->def_counter[port] = -1;
2267
2268 for (port = 0; port < dev->caps.num_ports; port++) {
2269 err = mlx4_counter_alloc(dev, &idx);
2270
2271 if (!err || err == -ENOSPC) {
2272 priv->def_counter[port] = idx;
2273 } else if (err == -ENOENT) {
2274 err = 0;
2275 continue;
2276 } else {
2277 mlx4_err(dev, "%s: failed to allocate default counter port %d err %d\n",
2278 __func__, port + 1, err);
2279 mlx4_cleanup_default_counters(dev);
2280 return err;
2281 }
2282
2283 mlx4_dbg(dev, "%s: default counter index %d for port %d\n",
2284 __func__, priv->def_counter[port], port + 1);
2285 }
2286
2287 return err;
2288}
2289
ba062d52 2290int __mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx)
f2a3f6a3
OG		 2291{
2292 struct mlx4_priv *priv = mlx4_priv(dev);
2293
2294 if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2295 return -ENOENT;
2296
2297 *idx = mlx4_bitmap_alloc(&priv->counters_bitmap);
6de5f7f6
EBE		 2298 if (*idx == -1) {
2299 *idx = MLX4_SINK_COUNTER_INDEX(dev);
2300 return -ENOSPC;
2301 }
f2a3f6a3
OG		 2302
2303 return 0;
2304}
ba062d52
JM		 2305
2306int mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx)
2307{
2308 u64 out_param;
2309 int err;
2310
2311 if (mlx4_is_mfunc(dev)) {
2312 err = mlx4_cmd_imm(dev, 0, &out_param, RES_COUNTER,
2313 RES_OP_RESERVE, MLX4_CMD_ALLOC_RES,
2314 MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
2315 if (!err)
2316 *idx = get_param_l(&out_param);
2317
2318 return err;
2319 }
2320 return __mlx4_counter_alloc(dev, idx);
2321}
f2a3f6a3
OG		 2322EXPORT_SYMBOL_GPL(mlx4_counter_alloc);
2323
b72ca7e9
EBE		 2324static int __mlx4_clear_if_stat(struct mlx4_dev *dev,
2325 u8 counter_index)
2326{
2327 struct mlx4_cmd_mailbox *if_stat_mailbox;
2328 int err;
2329 u32 if_stat_in_mod = (counter_index & 0xff) | MLX4_QUERY_IF_STAT_RESET;
2330
2331 if_stat_mailbox = mlx4_alloc_cmd_mailbox(dev);
2332 if (IS_ERR(if_stat_mailbox))
2333 return PTR_ERR(if_stat_mailbox);
2334
2335 err = mlx4_cmd_box(dev, 0, if_stat_mailbox->dma, if_stat_in_mod, 0,
2336 MLX4_CMD_QUERY_IF_STAT, MLX4_CMD_TIME_CLASS_C,
2337 MLX4_CMD_NATIVE);
2338
2339 mlx4_free_cmd_mailbox(dev, if_stat_mailbox);
2340 return err;
2341}
2342
ba062d52 2343void __mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
f2a3f6a3 2344{
efa6bc91
EBE		 2345 if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2346 return;
2347
6de5f7f6
EBE		 2348 if (idx == MLX4_SINK_COUNTER_INDEX(dev))
2349 return;
2350
b72ca7e9
EBE		 2351 __mlx4_clear_if_stat(dev, idx);
2352
7c6d74d2 2353 mlx4_bitmap_free(&mlx4_priv(dev)->counters_bitmap, idx, MLX4_USE_RR);
f2a3f6a3
OG		 2354 return;
2355}
ba062d52
JM		 2356
2357void mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
2358{
e7dbeba8 2359 u64 in_param = 0;
ba062d52
JM		 2360
2361 if (mlx4_is_mfunc(dev)) {
2362 set_param_l(&in_param, idx);
2363 mlx4_cmd(dev, in_param, RES_COUNTER, RES_OP_RESERVE,
2364 MLX4_CMD_FREE_RES, MLX4_CMD_TIME_CLASS_A,
2365 MLX4_CMD_WRAPPED);
2366 return;
2367 }
2368 __mlx4_counter_free(dev, idx);
2369}
f2a3f6a3
OG		 2370EXPORT_SYMBOL_GPL(mlx4_counter_free);
2371
6de5f7f6
EBE		 2372int mlx4_get_default_counter_index(struct mlx4_dev *dev, int port)
2373{
2374 struct mlx4_priv *priv = mlx4_priv(dev);
2375
2376 return priv->def_counter[port - 1];
2377}
2378EXPORT_SYMBOL_GPL(mlx4_get_default_counter_index);
2379
773af94e
YH		 2380void mlx4_set_admin_guid(struct mlx4_dev *dev, __be64 guid, int entry, int port)
2381{
2382 struct mlx4_priv *priv = mlx4_priv(dev);
2383
2384 priv->mfunc.master.vf_admin[entry].vport[port].guid = guid;
2385}
2386EXPORT_SYMBOL_GPL(mlx4_set_admin_guid);
2387
2388__be64 mlx4_get_admin_guid(struct mlx4_dev *dev, int entry, int port)
2389{
2390 struct mlx4_priv *priv = mlx4_priv(dev);
2391
2392 return priv->mfunc.master.vf_admin[entry].vport[port].guid;
2393}
2394EXPORT_SYMBOL_GPL(mlx4_get_admin_guid);
2395
fb517a4f
YH		 2396void mlx4_set_random_admin_guid(struct mlx4_dev *dev, int entry, int port)
2397{
2398 struct mlx4_priv *priv = mlx4_priv(dev);
2399 __be64 guid;
2400
2401 /* hw GUID */
2402 if (entry == 0)
2403 return;
2404
2405 get_random_bytes((char *)&guid, sizeof(guid));
2406 guid &= ~(cpu_to_be64(1ULL << 56));
2407 guid |= cpu_to_be64(1ULL << 57);
2408 priv->mfunc.master.vf_admin[entry].vport[port].guid = guid;
2409}
2410
3d73c288 2411static int mlx4_setup_hca(struct mlx4_dev *dev)
225c7b1f
RD		 2412{
2413 struct mlx4_priv *priv = mlx4_priv(dev);
2414 int err;
7ff93f8b 2415 int port;
9a5aa622 2416 __be32 ib_port_default_caps;
225c7b1f 2417
225c7b1f
RD		 2418 err = mlx4_init_uar_table(dev);
2419 if (err) {
1a91de28
JP		 2420 mlx4_err(dev, "Failed to initialize user access region table, aborting\n");
2421 return err;
225c7b1f
RD		 2422 }
2423
2424 err = mlx4_uar_alloc(dev, &priv->driver_uar);
2425 if (err) {
1a91de28 2426 mlx4_err(dev, "Failed to allocate driver access region, aborting\n");
225c7b1f
RD		 2427 goto err_uar_table_free;
2428 }
2429
4979d18f 2430 priv->kar = ioremap((phys_addr_t) priv->driver_uar.pfn << PAGE_SHIFT, PAGE_SIZE);
225c7b1f 2431 if (!priv->kar) {
1a91de28 2432 mlx4_err(dev, "Couldn't map kernel access region, aborting\n");
225c7b1f
RD		 2433 err = -ENOMEM;
2434 goto err_uar_free;
2435 }
2436
2437 err = mlx4_init_pd_table(dev);
2438 if (err) {
1a91de28 2439 mlx4_err(dev, "Failed to initialize protection domain table, aborting\n");
225c7b1f
RD		 2440 goto err_kar_unmap;
2441 }
2442
012a8ff5
SH		 2443 err = mlx4_init_xrcd_table(dev);
2444 if (err) {
1a91de28 2445 mlx4_err(dev, "Failed to initialize reliable connection domain table, aborting\n");
012a8ff5
SH		 2446 goto err_pd_table_free;
2447 }
2448
225c7b1f
RD		 2449 err = mlx4_init_mr_table(dev);
2450 if (err) {
1a91de28 2451 mlx4_err(dev, "Failed to initialize memory region table, aborting\n");
012a8ff5 2452 goto err_xrcd_table_free;
225c7b1f
RD		 2453 }
2454
fe6f700d
YP		 2455 if (!mlx4_is_slave(dev)) {
2456 err = mlx4_init_mcg_table(dev);
2457 if (err) {
1a91de28 2458 mlx4_err(dev, "Failed to initialize multicast group table, aborting\n");
fe6f700d
YP		 2459 goto err_mr_table_free;
2460 }
114840c3
JM		 2461 err = mlx4_config_mad_demux(dev);
2462 if (err) {
2463 mlx4_err(dev, "Failed in config_mad_demux, aborting\n");
2464 goto err_mcg_table_free;
2465 }
fe6f700d
YP		 2466 }
2467
225c7b1f
RD		 2468 err = mlx4_init_eq_table(dev);
2469 if (err) {
1a91de28 2470 mlx4_err(dev, "Failed to initialize event queue table, aborting\n");
fe6f700d 2471 goto err_mcg_table_free;
225c7b1f
RD		 2472 }
2473
2474 err = mlx4_cmd_use_events(dev);
2475 if (err) {
1a91de28 2476 mlx4_err(dev, "Failed to switch to event-driven firmware commands, aborting\n");
225c7b1f
RD		 2477 goto err_eq_table_free;
2478 }
2479
2480 err = mlx4_NOP(dev);
2481 if (err) {
08fb1055 2482 if (dev->flags & MLX4_FLAG_MSI_X) {
1a91de28 2483 mlx4_warn(dev, "NOP command failed to generate MSI-X interrupt IRQ %d)\n",
c66fa19c 2484 priv->eq_table.eq[MLX4_EQ_ASYNC].irq);
1a91de28 2485 mlx4_warn(dev, "Trying again without MSI-X\n");
08fb1055 2486 } else {
1a91de28 2487 mlx4_err(dev, "NOP command failed to generate interrupt (IRQ %d), aborting\n",
c66fa19c 2488 priv->eq_table.eq[MLX4_EQ_ASYNC].irq);
225c7b1f 2489 mlx4_err(dev, "BIOS or ACPI interrupt routing problem?\n");
08fb1055 2490 }
225c7b1f
RD		 2491
2492 goto err_cmd_poll;
2493 }
2494
2495 mlx4_dbg(dev, "NOP command IRQ test passed\n");
2496
2497 err = mlx4_init_cq_table(dev);
2498 if (err) {
1a91de28 2499 mlx4_err(dev, "Failed to initialize completion queue table, aborting\n");
225c7b1f
RD		 2500 goto err_cmd_poll;
2501 }
2502
2503 err = mlx4_init_srq_table(dev);
2504 if (err) {
1a91de28 2505 mlx4_err(dev, "Failed to initialize shared receive queue table, aborting\n");
225c7b1f
RD		 2506 goto err_cq_table_free;
2507 }
2508
2509 err = mlx4_init_qp_table(dev);
2510 if (err) {
1a91de28 2511 mlx4_err(dev, "Failed to initialize queue pair table, aborting\n");
225c7b1f
RD		 2512 goto err_srq_table_free;
2513 }
2514
2632d18d
EBE		 2515 if (!mlx4_is_slave(dev)) {
2516 err = mlx4_init_counters_table(dev);
2517 if (err && err != -ENOENT) {
2518 mlx4_err(dev, "Failed to initialize counters table, aborting\n");
2519 goto err_qp_table_free;
2520 }
f2a3f6a3
OG		 2521 }
2522
6de5f7f6
EBE		 2523 err = mlx4_allocate_default_counters(dev);
2524 if (err) {
2525 mlx4_err(dev, "Failed to allocate default counters, aborting\n");
2526 goto err_counters_table_free;
f2a3f6a3
OG		 2527 }
2528
ab9c17a0
JM		 2529 if (!mlx4_is_slave(dev)) {
2530 for (port = 1; port <= dev->caps.num_ports; port++) {
ab9c17a0
JM		 2531 ib_port_default_caps = 0;
2532 err = mlx4_get_port_ib_caps(dev, port,
2533 &ib_port_default_caps);
2534 if (err)
1a91de28
JP		 2535 mlx4_warn(dev, "failed to get port %d default ib capabilities (%d). Continuing with caps = 0\n",
2536 port, err);
ab9c17a0
JM		 2537 dev->caps.ib_port_def_cap[port] = ib_port_default_caps;
2538
2aca1172
JM		 2539 /* initialize per-slave default ib port capabilities */
2540 if (mlx4_is_master(dev)) {
2541 int i;
2542 for (i = 0; i < dev->num_slaves; i++) {
2543 if (i == mlx4_master_func_num(dev))
2544 continue;
2545 priv->mfunc.master.slave_state[i].ib_cap_mask[port] =
1a91de28 2546 ib_port_default_caps;
2aca1172
JM		 2547 }
2548 }
2549
096335b3
OG		 2550 if (mlx4_is_mfunc(dev))
2551 dev->caps.port_ib_mtu[port] = IB_MTU_2048;
2552 else
2553 dev->caps.port_ib_mtu[port] = IB_MTU_4096;
97285b78 2554
6634961c
JM		 2555 err = mlx4_SET_PORT(dev, port, mlx4_is_master(dev) ?
2556 dev->caps.pkey_table_len[port] : -1);
ab9c17a0
JM		 2557 if (err) {
2558 mlx4_err(dev, "Failed to set port %d, aborting\n",
1a91de28 2559 port);
6de5f7f6 2560 goto err_default_countes_free;
ab9c17a0 2561 }
7ff93f8b
YP		 2562 }
2563 }
2564
225c7b1f
RD		 2565 return 0;
2566
6de5f7f6
EBE		 2567err_default_countes_free:
2568 mlx4_cleanup_default_counters(dev);
2569
f2a3f6a3 2570err_counters_table_free:
2632d18d
EBE		 2571 if (!mlx4_is_slave(dev))
2572 mlx4_cleanup_counters_table(dev);
f2a3f6a3 2573
225c7b1f
RD		 2574err_qp_table_free:
2575 mlx4_cleanup_qp_table(dev);
2576
2577err_srq_table_free:
2578 mlx4_cleanup_srq_table(dev);
2579
2580err_cq_table_free:
2581 mlx4_cleanup_cq_table(dev);
2582
2583err_cmd_poll:
2584 mlx4_cmd_use_polling(dev);
2585
2586err_eq_table_free:
2587 mlx4_cleanup_eq_table(dev);
2588
fe6f700d
YP		 2589err_mcg_table_free:
2590 if (!mlx4_is_slave(dev))
2591 mlx4_cleanup_mcg_table(dev);
2592
ee49bd93 2593err_mr_table_free:
225c7b1f
RD		 2594 mlx4_cleanup_mr_table(dev);
2595
012a8ff5
SH		 2596err_xrcd_table_free:
2597 mlx4_cleanup_xrcd_table(dev);
2598
225c7b1f
RD		 2599err_pd_table_free:
2600 mlx4_cleanup_pd_table(dev);
2601
2602err_kar_unmap:
2603 iounmap(priv->kar);
2604
2605err_uar_free:
2606 mlx4_uar_free(dev, &priv->driver_uar);
2607
2608err_uar_table_free:
2609 mlx4_cleanup_uar_table(dev);
2610 return err;
2611}
2612
de161803
IS		 2613static int mlx4_init_affinity_hint(struct mlx4_dev *dev, int port, int eqn)
2614{
2615 int requested_cpu = 0;
2616 struct mlx4_priv *priv = mlx4_priv(dev);
2617 struct mlx4_eq *eq;
2618 int off = 0;
2619 int i;
2620
2621 if (eqn > dev->caps.num_comp_vectors)
2622 return -EINVAL;
2623
2624 for (i = 1; i < port; i++)
2625 off += mlx4_get_eqs_per_port(dev, i);
2626
2627 requested_cpu = eqn - off - !!(eqn > MLX4_EQ_ASYNC);
2628
2629 /* Meaning EQs are shared, and this call comes from the second port */
2630 if (requested_cpu < 0)
2631 return 0;
2632
2633 eq = &priv->eq_table.eq[eqn];
2634
2635 if (!zalloc_cpumask_var(&eq->affinity_mask, GFP_KERNEL))
2636 return -ENOMEM;
2637
2638 cpumask_set_cpu(requested_cpu, eq->affinity_mask);
2639
2640 return 0;
2641}
2642
e8f9b2ed 2643static void mlx4_enable_msi_x(struct mlx4_dev *dev)
225c7b1f
RD		 2644{
2645 struct mlx4_priv *priv = mlx4_priv(dev);
b8dd786f 2646 struct msix_entry *entries;
225c7b1f 2647 int i;
c66fa19c 2648 int port = 0;
225c7b1f
RD		 2649
2650 if (msi_x) {
c66fa19c 2651 int nreq = dev->caps.num_ports * num_online_cpus() + 1;
7ae0e400 2652
ca4c7b35
OG		 2653 nreq = min_t(int, dev->caps.num_eqs - dev->caps.reserved_eqs,
2654 nreq);
ab9c17a0 2655
b8dd786f
YP		 2656 entries = kcalloc(nreq, sizeof *entries, GFP_KERNEL);
2657 if (!entries)
2658 goto no_msi;
2659
2660 for (i = 0; i < nreq; ++i)
225c7b1f
RD		 2661 entries[i].entry = i;
2662
872bf2fb
YH		 2663 nreq = pci_enable_msix_range(dev->persist->pdev, entries, 2,
2664 nreq);
66e2f9c1 2665
c66fa19c 2666 if (nreq < 0 || nreq < MLX4_EQ_ASYNC) {
5bf0da7d 2667 kfree(entries);
225c7b1f 2668 goto no_msi;
0b7ca5a9 2669 }
c66fa19c
MB		 2670 /* 1 is reserved for events (asyncrounous EQ) */
2671 dev->caps.num_comp_vectors = nreq - 1;
2672
2673 priv->eq_table.eq[MLX4_EQ_ASYNC].irq = entries[0].vector;
2674 bitmap_zero(priv->eq_table.eq[MLX4_EQ_ASYNC].actv_ports.ports,
2675 dev->caps.num_ports);
2676
2677 for (i = 0; i < dev->caps.num_comp_vectors + 1; i++) {
2678 if (i == MLX4_EQ_ASYNC)
2679 continue;
2680
2681 priv->eq_table.eq[i].irq =
2682 entries[i + 1 - !!(i > MLX4_EQ_ASYNC)].vector;
2683
2684 if (MLX4_IS_LEGACY_EQ_MODE(dev->caps)) {
2685 bitmap_fill(priv->eq_table.eq[i].actv_ports.ports,
2686 dev->caps.num_ports);
de161803
IS		 2687 /* We don't set affinity hint when there
2688 * aren't enough EQs
2689 */
c66fa19c
MB		 2690 } else {
2691 set_bit(port,
2692 priv->eq_table.eq[i].actv_ports.ports);
de161803
IS		 2693 if (mlx4_init_affinity_hint(dev, port + 1, i))
2694 mlx4_warn(dev, "Couldn't init hint cpumask for EQ %d\n",
2695 i);
c66fa19c
MB		 2696 }
2697 /* We divide the Eqs evenly between the two ports.
2698 * (dev->caps.num_comp_vectors / dev->caps.num_ports)
2699 * refers to the number of Eqs per port
2700 * (i.e eqs_per_port). Theoretically, we would like to
2701 * write something like (i + 1) % eqs_per_port == 0.
2702 * However, since there's an asynchronous Eq, we have
2703 * to skip over it by comparing this condition to
2704 * !!((i + 1) > MLX4_EQ_ASYNC).
2705 */
2706 if ((dev->caps.num_comp_vectors > dev->caps.num_ports) &&
2707 ((i + 1) %
2708 (dev->caps.num_comp_vectors / dev->caps.num_ports)) ==
2709 !!((i + 1) > MLX4_EQ_ASYNC))
2710 /* If dev->caps.num_comp_vectors < dev->caps.num_ports,
2711 * everything is shared anyway.
2712 */
2713 port++;
2714 }
225c7b1f
RD		 2715
2716 dev->flags |= MLX4_FLAG_MSI_X;
b8dd786f
YP		 2717
2718 kfree(entries);
225c7b1f
RD		 2719 return;
2720 }
2721
2722no_msi:
b8dd786f
YP		 2723 dev->caps.num_comp_vectors = 1;
2724
c66fa19c
MB		 2725 BUG_ON(MLX4_EQ_ASYNC >= 2);
2726 for (i = 0; i < 2; ++i) {
872bf2fb 2727 priv->eq_table.eq[i].irq = dev->persist->pdev->irq;
c66fa19c
MB		 2728 if (i != MLX4_EQ_ASYNC) {
2729 bitmap_fill(priv->eq_table.eq[i].actv_ports.ports,
2730 dev->caps.num_ports);
2731 }
2732 }
225c7b1f
RD		 2733}
2734
7ff93f8b 2735static int mlx4_init_port_info(struct mlx4_dev *dev, int port)
2a2336f8
YP		 2736{
2737 struct mlx4_port_info *info = &mlx4_priv(dev)->port[port];
7ff93f8b 2738 int err = 0;
2a2336f8
YP		 2739
2740 info->dev = dev;
2741 info->port = port;
ab9c17a0 2742 if (!mlx4_is_slave(dev)) {
ab9c17a0
JM		 2743 mlx4_init_mac_table(dev, &info->mac_table);
2744 mlx4_init_vlan_table(dev, &info->vlan_table);
111c6094 2745 mlx4_init_roce_gid_table(dev, &info->gid_table);
16a10ffd 2746 info->base_qpn = mlx4_get_base_qpn(dev, port);
ab9c17a0 2747 }
7ff93f8b
YP		 2748
2749 sprintf(info->dev_name, "mlx4_port%d", port);
2750 info->port_attr.attr.name = info->dev_name;
ab9c17a0
JM		 2751 if (mlx4_is_mfunc(dev))
2752 info->port_attr.attr.mode = S_IRUGO;
2753 else {
2754 info->port_attr.attr.mode = S_IRUGO | S_IWUSR;
2755 info->port_attr.store = set_port_type;
2756 }
7ff93f8b 2757 info->port_attr.show = show_port_type;
3691c964 2758 sysfs_attr_init(&info->port_attr.attr);
7ff93f8b 2759
872bf2fb 2760 err = device_create_file(&dev->persist->pdev->dev, &info->port_attr);
7ff93f8b
YP		 2761 if (err) {
2762 mlx4_err(dev, "Failed to create file for port %d\n", port);
2763 info->port = -1;
2764 }
2765
096335b3
OG		 2766 sprintf(info->dev_mtu_name, "mlx4_port%d_mtu", port);
2767 info->port_mtu_attr.attr.name = info->dev_mtu_name;
2768 if (mlx4_is_mfunc(dev))
2769 info->port_mtu_attr.attr.mode = S_IRUGO;
2770 else {
2771 info->port_mtu_attr.attr.mode = S_IRUGO | S_IWUSR;
2772 info->port_mtu_attr.store = set_port_ib_mtu;
2773 }
2774 info->port_mtu_attr.show = show_port_ib_mtu;
2775 sysfs_attr_init(&info->port_mtu_attr.attr);
2776
872bf2fb
YH		 2777 err = device_create_file(&dev->persist->pdev->dev,
2778 &info->port_mtu_attr);
096335b3
OG		 2779 if (err) {
2780 mlx4_err(dev, "Failed to create mtu file for port %d\n", port);
872bf2fb
YH		 2781 device_remove_file(&info->dev->persist->pdev->dev,
2782 &info->port_attr);
096335b3
OG		 2783 info->port = -1;
2784 }
2785
7ff93f8b
YP		 2786 return err;
2787}
2788
2789static void mlx4_cleanup_port_info(struct mlx4_port_info *info)
2790{
2791 if (info->port < 0)
2792 return;
2793
872bf2fb
YH		 2794 device_remove_file(&info->dev->persist->pdev->dev, &info->port_attr);
2795 device_remove_file(&info->dev->persist->pdev->dev,
2796 &info->port_mtu_attr);
c66fa19c
MB		 2797#ifdef CONFIG_RFS_ACCEL
2798 free_irq_cpu_rmap(info->rmap);
2799 info->rmap = NULL;
2800#endif
2a2336f8
YP		 2801}
2802
b12d93d6
YP		 2803static int mlx4_init_steering(struct mlx4_dev *dev)
2804{
2805 struct mlx4_priv *priv = mlx4_priv(dev);
2806 int num_entries = dev->caps.num_ports;
2807 int i, j;
2808
2809 priv->steer = kzalloc(sizeof(struct mlx4_steer) * num_entries, GFP_KERNEL);
2810 if (!priv->steer)
2811 return -ENOMEM;
2812
45b51365 2813 for (i = 0; i < num_entries; i++)
b12d93d6
YP		 2814 for (j = 0; j < MLX4_NUM_STEERS; j++) {
2815 INIT_LIST_HEAD(&priv->steer[i].promisc_qps[j]);
2816 INIT_LIST_HEAD(&priv->steer[i].steer_entries[j]);
2817 }
b12d93d6
YP		 2818 return 0;
2819}
2820
2821static void mlx4_clear_steering(struct mlx4_dev *dev)
2822{
2823 struct mlx4_priv *priv = mlx4_priv(dev);
2824 struct mlx4_steer_index *entry, *tmp_entry;
2825 struct mlx4_promisc_qp *pqp, *tmp_pqp;
2826 int num_entries = dev->caps.num_ports;
2827 int i, j;
2828
2829 for (i = 0; i < num_entries; i++) {
2830 for (j = 0; j < MLX4_NUM_STEERS; j++) {
2831 list_for_each_entry_safe(pqp, tmp_pqp,
2832 &priv->steer[i].promisc_qps[j],
2833 list) {
2834 list_del(&pqp->list);
2835 kfree(pqp);
2836 }
2837 list_for_each_entry_safe(entry, tmp_entry,
2838 &priv->steer[i].steer_entries[j],
2839 list) {
2840 list_del(&entry->list);
2841 list_for_each_entry_safe(pqp, tmp_pqp,
2842 &entry->duplicates,
2843 list) {
2844 list_del(&pqp->list);
2845 kfree(pqp);
2846 }
2847 kfree(entry);
2848 }
2849 }
2850 }
2851 kfree(priv->steer);
2852}
2853
ab9c17a0
JM		 2854static int extended_func_num(struct pci_dev *pdev)
2855{
2856 return PCI_SLOT(pdev->devfn) * 8 + PCI_FUNC(pdev->devfn);
2857}
2858
2859#define MLX4_OWNER_BASE 0x8069c
2860#define MLX4_OWNER_SIZE 4
2861
2862static int mlx4_get_ownership(struct mlx4_dev *dev)
2863{
2864 void __iomem *owner;
2865 u32 ret;
2866
872bf2fb 2867 if (pci_channel_offline(dev->persist->pdev))
57dbf29a
KSS		 2868 return -EIO;
2869
872bf2fb
YH		 2870 owner = ioremap(pci_resource_start(dev->persist->pdev, 0) +
2871 MLX4_OWNER_BASE,
ab9c17a0
JM		 2872 MLX4_OWNER_SIZE);
2873 if (!owner) {
2874 mlx4_err(dev, "Failed to obtain ownership bit\n");
2875 return -ENOMEM;
2876 }
2877
2878 ret = readl(owner);
2879 iounmap(owner);
2880 return (int) !!ret;
2881}
2882
2883static void mlx4_free_ownership(struct mlx4_dev *dev)
2884{
2885 void __iomem *owner;
2886
872bf2fb 2887 if (pci_channel_offline(dev->persist->pdev))
57dbf29a
KSS		 2888 return;
2889
872bf2fb
YH		 2890 owner = ioremap(pci_resource_start(dev->persist->pdev, 0) +
2891 MLX4_OWNER_BASE,
ab9c17a0
JM		 2892 MLX4_OWNER_SIZE);
2893 if (!owner) {
2894 mlx4_err(dev, "Failed to obtain ownership bit\n");
2895 return;
2896 }
2897 writel(0, owner);
2898 msleep(1000);
2899 iounmap(owner);
2900}
2901
a0eacca9
MB		 2902#define SRIOV_VALID_STATE(flags) (!!((flags) & MLX4_FLAG_SRIOV) ==\
2903 !!((flags) & MLX4_FLAG_MASTER))
2904
2905static u64 mlx4_enable_sriov(struct mlx4_dev *dev, struct pci_dev *pdev,
55ad3592 2906 u8 total_vfs, int existing_vfs, int reset_flow)
a0eacca9
MB		 2907{
2908 u64 dev_flags = dev->flags;
da315679 2909 int err = 0;
a0eacca9 2910
55ad3592
YH		 2911 if (reset_flow) {
2912 dev->dev_vfs = kcalloc(total_vfs, sizeof(*dev->dev_vfs),
2913 GFP_KERNEL);
2914 if (!dev->dev_vfs)
2915 goto free_mem;
2916 return dev_flags;
2917 }
2918
da315679
MB		 2919 atomic_inc(&pf_loading);
2920 if (dev->flags & MLX4_FLAG_SRIOV) {
2921 if (existing_vfs != total_vfs) {
2922 mlx4_err(dev, "SR-IOV was already enabled, but with num_vfs (%d) different than requested (%d)\n",
2923 existing_vfs, total_vfs);
2924 total_vfs = existing_vfs;
2925 }
2926 }
2927
2928 dev->dev_vfs = kzalloc(total_vfs * sizeof(*dev->dev_vfs), GFP_KERNEL);
a0eacca9
MB		 2929 if (NULL == dev->dev_vfs) {
2930 mlx4_err(dev, "Failed to allocate memory for VFs\n");
2931 goto disable_sriov;
da315679
MB		 2932 }
2933
2934 if (!(dev->flags & MLX4_FLAG_SRIOV)) {
2935 mlx4_warn(dev, "Enabling SR-IOV with %d VFs\n", total_vfs);
2936 err = pci_enable_sriov(pdev, total_vfs);
2937 }
2938 if (err) {
2939 mlx4_err(dev, "Failed to enable SR-IOV, continuing without SR-IOV (err = %d)\n",
2940 err);
2941 goto disable_sriov;
2942 } else {
2943 mlx4_warn(dev, "Running in master mode\n");
2944 dev_flags |= MLX4_FLAG_SRIOV |
2945 MLX4_FLAG_MASTER;
2946 dev_flags &= ~MLX4_FLAG_SLAVE;
872bf2fb 2947 dev->persist->num_vfs = total_vfs;
a0eacca9
MB		 2948 }
2949 return dev_flags;
2950
2951disable_sriov:
da315679 2952 atomic_dec(&pf_loading);
55ad3592 2953free_mem:
872bf2fb 2954 dev->persist->num_vfs = 0;
a0eacca9 2955 kfree(dev->dev_vfs);
5114a04e 2956 dev->dev_vfs = NULL;
a0eacca9
MB		 2957 return dev_flags & ~MLX4_FLAG_MASTER;
2958}
2959
de966c59
MB		 2960enum {
2961 MLX4_DEV_CAP_CHECK_NUM_VFS_ABOVE_64 = -1,
2962};
2963
2964static int mlx4_check_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap,
2965 int *nvfs)
2966{
2967 int requested_vfs = nvfs[0] + nvfs[1] + nvfs[2];
2968 /* Checking for 64 VFs as a limitation of CX2 */
2969 if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_80_VFS) &&
2970 requested_vfs >= 64) {
2971 mlx4_err(dev, "Requested %d VFs, but FW does not support more than 64\n",
2972 requested_vfs);
2973 return MLX4_DEV_CAP_CHECK_NUM_VFS_ABOVE_64;
2974 }
2975 return 0;
2976}
2977
e1c00e10 2978static int mlx4_load_one(struct pci_dev *pdev, int pci_dev_data,
55ad3592
YH		 2979 int total_vfs, int *nvfs, struct mlx4_priv *priv,
2980 int reset_flow)
225c7b1f 2981{
225c7b1f 2982 struct mlx4_dev *dev;
e1c00e10 2983 unsigned sum = 0;
225c7b1f 2984 int err;
2a2336f8 2985 int port;
e1c00e10 2986 int i;
7ae0e400 2987 struct mlx4_dev_cap *dev_cap = NULL;
bbb07af4 2988 int existing_vfs = 0;
225c7b1f 2989
e1c00e10 2990 dev = &priv->dev;
225c7b1f 2991
b581401e
RD		 2992 INIT_LIST_HEAD(&priv->ctx_list);
2993 spin_lock_init(&priv->ctx_lock);
225c7b1f 2994
7ff93f8b 2995 mutex_init(&priv->port_mutex);
53f33ae2 2996 mutex_init(&priv->bond_mutex);
7ff93f8b 2997
6296883c
YP		 2998 INIT_LIST_HEAD(&priv->pgdir_list);
2999 mutex_init(&priv->pgdir_mutex);
3000
c1b43dca
EC		 3001 INIT_LIST_HEAD(&priv->bf_list);
3002 mutex_init(&priv->bf_mutex);
3003
aca7a3ac 3004 dev->rev_id = pdev->revision;
6e7136ed 3005 dev->numa_node = dev_to_node(&pdev->dev);
e1c00e10 3006
ab9c17a0 3007 /* Detect if this device is a virtual function */
839f1243 3008 if (pci_dev_data & MLX4_PCI_DEV_IS_VF) {
ab9c17a0
JM		 3009 mlx4_warn(dev, "Detected virtual function - running in slave mode\n");
3010 dev->flags |= MLX4_FLAG_SLAVE;
3011 } else {
3012 /* We reset the device and enable SRIOV only for physical
3013 * devices. Try to claim ownership on the device;
3014 * if already taken, skip -- do not allow multiple PFs */
3015 err = mlx4_get_ownership(dev);
3016 if (err) {
3017 if (err < 0)
e1c00e10 3018 return err;
ab9c17a0 3019 else {
1a91de28 3020 mlx4_warn(dev, "Multiple PFs not yet supported - Skipping PF\n");
e1c00e10 3021 return -EINVAL;
ab9c17a0
JM		 3022 }
3023 }
aca7a3ac 3024
fe6f700d
YP		 3025 atomic_set(&priv->opreq_count, 0);
3026 INIT_WORK(&priv->opreq_task, mlx4_opreq_action);
3027
ab9c17a0
JM		 3028 /*
3029 * Now reset the HCA before we touch the PCI capabilities or
3030 * attempt a firmware command, since a boot ROM may have left
3031 * the HCA in an undefined state.
3032 */
3033 err = mlx4_reset(dev);
3034 if (err) {
1a91de28 3035 mlx4_err(dev, "Failed to reset HCA, aborting\n");
e1c00e10 3036 goto err_sriov;
ab9c17a0 3037 }
7ae0e400
MB		 3038
3039 if (total_vfs) {
7ae0e400 3040 dev->flags = MLX4_FLAG_MASTER;
da315679
MB		 3041 existing_vfs = pci_num_vf(pdev);
3042 if (existing_vfs)
3043 dev->flags |= MLX4_FLAG_SRIOV;
872bf2fb 3044 dev->persist->num_vfs = total_vfs;
7ae0e400 3045 }
225c7b1f
RD		 3046 }
3047
f6bc11e4
YH		 3048 /* on load remove any previous indication of internal error,
3049 * device is up.
3050 */
3051 dev->persist->state = MLX4_DEVICE_STATE_UP;
3052
ab9c17a0 3053slave_start:
521130d1
EE		 3054 err = mlx4_cmd_init(dev);
3055 if (err) {
1a91de28 3056 mlx4_err(dev, "Failed to init command interface, aborting\n");
ab9c17a0
JM		 3057 goto err_sriov;
3058 }
3059
3060 /* In slave functions, the communication channel must be initialized
3061 * before posting commands. Also, init num_slaves before calling
3062 * mlx4_init_hca */
3063 if (mlx4_is_mfunc(dev)) {
7ae0e400 3064 if (mlx4_is_master(dev)) {
ab9c17a0 3065 dev->num_slaves = MLX4_MAX_NUM_SLAVES;
7ae0e400
MB		 3066
3067 } else {
ab9c17a0 3068 dev->num_slaves = 0;
f356fcbe
JM		 3069 err = mlx4_multi_func_init(dev);
3070 if (err) {
1a91de28 3071 mlx4_err(dev, "Failed to init slave mfunc interface, aborting\n");
ab9c17a0
JM		 3072 goto err_cmd;
3073 }
3074 }
225c7b1f
RD		 3075 }
3076
a0eacca9
MB		 3077 err = mlx4_init_fw(dev);
3078 if (err) {
3079 mlx4_err(dev, "Failed to init fw, aborting.\n");
3080 goto err_mfunc;
3081 }
3082
7ae0e400 3083 if (mlx4_is_master(dev)) {
da315679 3084 /* when we hit the goto slave_start below, dev_cap already initialized */
7ae0e400
MB		 3085 if (!dev_cap) {
3086 dev_cap = kzalloc(sizeof(*dev_cap), GFP_KERNEL);
3087
3088 if (!dev_cap) {
3089 err = -ENOMEM;
3090 goto err_fw;
3091 }
3092
3093 err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
3094 if (err) {
3095 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
3096 goto err_fw;
3097 }
3098
de966c59
MB		 3099 if (mlx4_check_dev_cap(dev, dev_cap, nvfs))
3100 goto err_fw;
3101
7ae0e400 3102 if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS)) {
55ad3592
YH		 3103 u64 dev_flags = mlx4_enable_sriov(dev, pdev,
3104 total_vfs,
3105 existing_vfs,
3106 reset_flow);
7ae0e400 3107
ed3d2276 3108 mlx4_close_fw(dev);
7ae0e400
MB		 3109 mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3110 dev->flags = dev_flags;
3111 if (!SRIOV_VALID_STATE(dev->flags)) {
3112 mlx4_err(dev, "Invalid SRIOV state\n");
3113 goto err_sriov;
3114 }
3115 err = mlx4_reset(dev);
3116 if (err) {
3117 mlx4_err(dev, "Failed to reset HCA, aborting.\n");
3118 goto err_sriov;
3119 }
3120 goto slave_start;
3121 }
3122 } else {
3123 /* Legacy mode FW requires SRIOV to be enabled before
3124 * doing QUERY_DEV_CAP, since max_eq's value is different if
3125 * SRIOV is enabled.
3126 */
3127 memset(dev_cap, 0, sizeof(*dev_cap));
3128 err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
3129 if (err) {
3130 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
3131 goto err_fw;
3132 }
de966c59
MB		 3133
3134 if (mlx4_check_dev_cap(dev, dev_cap, nvfs))
3135 goto err_fw;
7ae0e400
MB		 3136 }
3137 }
3138
225c7b1f 3139 err = mlx4_init_hca(dev);
ab9c17a0
JM		 3140 if (err) {
3141 if (err == -EACCES) {
3142 /* Not primary Physical function
3143 * Running in slave mode */
ffc39f6d 3144 mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
a0eacca9
MB		 3145 /* We're not a PF */
3146 if (dev->flags & MLX4_FLAG_SRIOV) {
3147 if (!existing_vfs)
3148 pci_disable_sriov(pdev);
55ad3592 3149 if (mlx4_is_master(dev) && !reset_flow)
a0eacca9
MB		 3150 atomic_dec(&pf_loading);
3151 dev->flags &= ~MLX4_FLAG_SRIOV;
3152 }
3153 if (!mlx4_is_slave(dev))
3154 mlx4_free_ownership(dev);
ab9c17a0
JM		 3155 dev->flags |= MLX4_FLAG_SLAVE;
3156 dev->flags &= ~MLX4_FLAG_MASTER;
3157 goto slave_start;
3158 } else
a0eacca9 3159 goto err_fw;
ab9c17a0
JM		 3160 }
3161
7ae0e400 3162 if (mlx4_is_master(dev) && (dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS)) {
55ad3592
YH		 3163 u64 dev_flags = mlx4_enable_sriov(dev, pdev, total_vfs,
3164 existing_vfs, reset_flow);
7ae0e400
MB		 3165
3166 if ((dev->flags ^ dev_flags) & (MLX4_FLAG_MASTER | MLX4_FLAG_SLAVE)) {
3167 mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_VHCR);
3168 dev->flags = dev_flags;
3169 err = mlx4_cmd_init(dev);
3170 if (err) {
3171 /* Only VHCR is cleaned up, so could still
3172 * send FW commands
3173 */
3174 mlx4_err(dev, "Failed to init VHCR command interface, aborting\n");
3175 goto err_close;
3176 }
3177 } else {
3178 dev->flags = dev_flags;
3179 }
3180
3181 if (!SRIOV_VALID_STATE(dev->flags)) {
3182 mlx4_err(dev, "Invalid SRIOV state\n");
3183 goto err_close;
3184 }
3185 }
3186
b912b2f8
EP		 3187 /* check if the device is functioning at its maximum possible speed.
3188 * No return code for this call, just warn the user in case of PCI
3189 * express device capabilities are under-satisfied by the bus.
3190 */
83d3459a
EP		 3191 if (!mlx4_is_slave(dev))
3192 mlx4_check_pcie_caps(dev);
b912b2f8 3193
ab9c17a0
JM		 3194 /* In master functions, the communication channel must be initialized
3195 * after obtaining its address from fw */
3196 if (mlx4_is_master(dev)) {
e1c00e10
MD		 3197 if (dev->caps.num_ports < 2 &&
3198 num_vfs_argc > 1) {
3199 err = -EINVAL;
3200 mlx4_err(dev,
3201 "Error: Trying to configure VFs on port 2, but HCA has only %d physical ports\n",
3202 dev->caps.num_ports);
ab9c17a0
JM		 3203 goto err_close;
3204 }
872bf2fb 3205 memcpy(dev->persist->nvfs, nvfs, sizeof(dev->persist->nvfs));
dd41cc3b 3206
872bf2fb
YH		 3207 for (i = 0;
3208 i < sizeof(dev->persist->nvfs)/
3209 sizeof(dev->persist->nvfs[0]); i++) {
e1c00e10
MD		 3210 unsigned j;
3211
872bf2fb 3212 for (j = 0; j < dev->persist->nvfs[i]; ++sum, ++j) {
e1c00e10
MD		 3213 dev->dev_vfs[sum].min_port = i < 2 ? i + 1 : 1;
3214 dev->dev_vfs[sum].n_ports = i < 2 ? 1 :
3215 dev->caps.num_ports;
1ab95d37
MB		 3216 }
3217 }
e1c00e10
MD		 3218
3219 /* In master functions, the communication channel
3220 * must be initialized after obtaining its address from fw
3221 */
3222 err = mlx4_multi_func_init(dev);
3223 if (err) {
3224 mlx4_err(dev, "Failed to init master mfunc interface, aborting.\n");
3225 goto err_close;
3226 }
ab9c17a0 3227 }
225c7b1f 3228
b8dd786f
YP		 3229 err = mlx4_alloc_eq_table(dev);
3230 if (err)
ab9c17a0 3231 goto err_master_mfunc;
b8dd786f 3232
c66fa19c 3233 bitmap_zero(priv->msix_ctl.pool_bm, MAX_MSIX);
730c41d5 3234 mutex_init(&priv->msix_ctl.pool_lock);
0b7ca5a9 3235
08fb1055 3236 mlx4_enable_msi_x(dev);
ab9c17a0
JM		 3237 if ((mlx4_is_mfunc(dev)) &&
3238 !(dev->flags & MLX4_FLAG_MSI_X)) {
f356fcbe 3239 err = -ENOSYS;
1a91de28 3240 mlx4_err(dev, "INTx is not supported in multi-function mode, aborting\n");
b12d93d6 3241 goto err_free_eq;
ab9c17a0
JM		 3242 }
3243
3244 if (!mlx4_is_slave(dev)) {
3245 err = mlx4_init_steering(dev);
3246 if (err)
e1c00e10 3247 goto err_disable_msix;
ab9c17a0 3248 }
b12d93d6 3249
225c7b1f 3250 err = mlx4_setup_hca(dev);
ab9c17a0
JM		 3251 if (err == -EBUSY && (dev->flags & MLX4_FLAG_MSI_X) &&
3252 !mlx4_is_mfunc(dev)) {
08fb1055 3253 dev->flags &= ~MLX4_FLAG_MSI_X;
9858d2d1 3254 dev->caps.num_comp_vectors = 1;
08fb1055
MT		 3255 pci_disable_msix(pdev);
3256 err = mlx4_setup_hca(dev);
3257 }
3258
225c7b1f 3259 if (err)
b12d93d6 3260 goto err_steer;
225c7b1f 3261
5a0d0a61 3262 mlx4_init_quotas(dev);
55ad3592
YH		 3263 /* When PF resources are ready arm its comm channel to enable
3264 * getting commands
3265 */
3266 if (mlx4_is_master(dev)) {
3267 err = mlx4_ARM_COMM_CHANNEL(dev);
3268 if (err) {
3269 mlx4_err(dev, " Failed to arm comm channel eq: %x\n",
3270 err);
3271 goto err_steer;
3272 }
3273 }
5a0d0a61 3274
7ff93f8b
YP		 3275 for (port = 1; port <= dev->caps.num_ports; port++) {
3276 err = mlx4_init_port_info(dev, port);
3277 if (err)
3278 goto err_port;
3279 }
2a2336f8 3280
53f33ae2
MS		 3281 priv->v2p.port1 = 1;
3282 priv->v2p.port2 = 2;
3283
225c7b1f
RD		 3284 err = mlx4_register_device(dev);
3285 if (err)
7ff93f8b 3286 goto err_port;
225c7b1f 3287
b046ffe5
EP		 3288 mlx4_request_modules(dev);
3289
27bf91d6
YP		 3290 mlx4_sense_init(dev);
3291 mlx4_start_sense(dev);
3292
befdf897 3293 priv->removed = 0;
225c7b1f 3294
55ad3592 3295 if (mlx4_is_master(dev) && dev->persist->num_vfs && !reset_flow)
e1a5ddc5
AV		 3296 atomic_dec(&pf_loading);
3297
da315679 3298 kfree(dev_cap);
225c7b1f
RD		 3299 return 0;
3300
7ff93f8b 3301err_port:
b4f77264 3302 for (--port; port >= 1; --port)
7ff93f8b
YP		 3303 mlx4_cleanup_port_info(&priv->port[port]);
3304
6de5f7f6 3305 mlx4_cleanup_default_counters(dev);
2632d18d
EBE		 3306 if (!mlx4_is_slave(dev))
3307 mlx4_cleanup_counters_table(dev);
225c7b1f
RD		 3308 mlx4_cleanup_qp_table(dev);
3309 mlx4_cleanup_srq_table(dev);
3310 mlx4_cleanup_cq_table(dev);
3311 mlx4_cmd_use_polling(dev);
3312 mlx4_cleanup_eq_table(dev);
fe6f700d 3313 mlx4_cleanup_mcg_table(dev);
225c7b1f 3314 mlx4_cleanup_mr_table(dev);
012a8ff5 3315 mlx4_cleanup_xrcd_table(dev);
225c7b1f
RD		 3316 mlx4_cleanup_pd_table(dev);
3317 mlx4_cleanup_uar_table(dev);
3318
b12d93d6 3319err_steer:
ab9c17a0
JM		 3320 if (!mlx4_is_slave(dev))
3321 mlx4_clear_steering(dev);
b12d93d6 3322
e1c00e10
MD		 3323err_disable_msix:
3324 if (dev->flags & MLX4_FLAG_MSI_X)
3325 pci_disable_msix(pdev);
3326
b8dd786f
YP		 3327err_free_eq:
3328 mlx4_free_eq_table(dev);
3329
ab9c17a0 3330err_master_mfunc:
772103e6
JM		 3331 if (mlx4_is_master(dev)) {
3332 mlx4_free_resource_tracker(dev, RES_TR_FREE_STRUCTS_ONLY);
ab9c17a0 3333 mlx4_multi_func_cleanup(dev);
772103e6 3334 }
ab9c17a0 3335
b38f2879 3336 if (mlx4_is_slave(dev)) {
99ec41d0 3337 kfree(dev->caps.qp0_qkey);
b38f2879
DB		 3338 kfree(dev->caps.qp0_tunnel);
3339 kfree(dev->caps.qp0_proxy);
3340 kfree(dev->caps.qp1_tunnel);
3341 kfree(dev->caps.qp1_proxy);
3342 }
3343
225c7b1f
RD		 3344err_close:
3345 mlx4_close_hca(dev);
3346
a0eacca9
MB		 3347err_fw:
3348 mlx4_close_fw(dev);
3349
ab9c17a0
JM		 3350err_mfunc:
3351 if (mlx4_is_slave(dev))
3352 mlx4_multi_func_cleanup(dev);
3353
225c7b1f 3354err_cmd:
ffc39f6d 3355 mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
225c7b1f 3356
ab9c17a0 3357err_sriov:
55ad3592 3358 if (dev->flags & MLX4_FLAG_SRIOV && !existing_vfs) {
ab9c17a0 3359 pci_disable_sriov(pdev);
55ad3592
YH		 3360 dev->flags &= ~MLX4_FLAG_SRIOV;
3361 }
ab9c17a0 3362
55ad3592 3363 if (mlx4_is_master(dev) && dev->persist->num_vfs && !reset_flow)
e1a5ddc5
AV		 3364 atomic_dec(&pf_loading);
3365
1ab95d37
MB		 3366 kfree(priv->dev.dev_vfs);
3367
e1c00e10
MD		 3368 if (!mlx4_is_slave(dev))
3369 mlx4_free_ownership(dev);
3370
7ae0e400 3371 kfree(dev_cap);
e1c00e10
MD		 3372 return err;
3373}
3374
3375static int __mlx4_init_one(struct pci_dev *pdev, int pci_dev_data,
3376 struct mlx4_priv *priv)
3377{
3378 int err;
3379 int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
3380 int prb_vf[MLX4_MAX_PORTS + 1] = {0, 0, 0};
3381 const int param_map[MLX4_MAX_PORTS + 1][MLX4_MAX_PORTS + 1] = {
3382 {2, 0, 0}, {0, 1, 2}, {0, 1, 2} };
3383 unsigned total_vfs = 0;
3384 unsigned int i;
3385
3386 pr_info(DRV_NAME ": Initializing %s\n", pci_name(pdev));
3387
3388 err = pci_enable_device(pdev);
3389 if (err) {
3390 dev_err(&pdev->dev, "Cannot enable PCI device, aborting\n");
3391 return err;
3392 }
3393
3394 /* Due to requirement that all VFs and the PF are *guaranteed* 2 MACS
3395 * per port, we must limit the number of VFs to 63 (since their are
3396 * 128 MACs)
3397 */
3398 for (i = 0; i < sizeof(nvfs)/sizeof(nvfs[0]) && i < num_vfs_argc;
3399 total_vfs += nvfs[param_map[num_vfs_argc - 1][i]], i++) {
3400 nvfs[param_map[num_vfs_argc - 1][i]] = num_vfs[i];
3401 if (nvfs[i] < 0) {
3402 dev_err(&pdev->dev, "num_vfs module parameter cannot be negative\n");
3403 err = -EINVAL;
3404 goto err_disable_pdev;
3405 }
3406 }
3407 for (i = 0; i < sizeof(prb_vf)/sizeof(prb_vf[0]) && i < probe_vfs_argc;
3408 i++) {
3409 prb_vf[param_map[probe_vfs_argc - 1][i]] = probe_vf[i];
3410 if (prb_vf[i] < 0 || prb_vf[i] > nvfs[i]) {
3411 dev_err(&pdev->dev, "probe_vf module parameter cannot be negative or greater than num_vfs\n");
3412 err = -EINVAL;
3413 goto err_disable_pdev;
3414 }
3415 }
3416 if (total_vfs >= MLX4_MAX_NUM_VF) {
3417 dev_err(&pdev->dev,
3418 "Requested more VF's (%d) than allowed (%d)\n",
3419 total_vfs, MLX4_MAX_NUM_VF - 1);
3420 err = -EINVAL;
3421 goto err_disable_pdev;
3422 }
3423
3424 for (i = 0; i < MLX4_MAX_PORTS; i++) {
3425 if (nvfs[i] + nvfs[2] >= MLX4_MAX_NUM_VF_P_PORT) {
3426 dev_err(&pdev->dev,
3427 "Requested more VF's (%d) for port (%d) than allowed (%d)\n",
3428 nvfs[i] + nvfs[2], i + 1,
3429 MLX4_MAX_NUM_VF_P_PORT - 1);
3430 err = -EINVAL;
3431 goto err_disable_pdev;
3432 }
3433 }
3434
3435 /* Check for BARs. */
3436 if (!(pci_dev_data & MLX4_PCI_DEV_IS_VF) &&
3437 !(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) {
3438 dev_err(&pdev->dev, "Missing DCS, aborting (driver_data: 0x%x, pci_resource_flags(pdev, 0):0x%lx)\n",
3439 pci_dev_data, pci_resource_flags(pdev, 0));
3440 err = -ENODEV;
3441 goto err_disable_pdev;
3442 }
3443 if (!(pci_resource_flags(pdev, 2) & IORESOURCE_MEM)) {
3444 dev_err(&pdev->dev, "Missing UAR, aborting\n");
3445 err = -ENODEV;
3446 goto err_disable_pdev;
3447 }
3448
3449 err = pci_request_regions(pdev, DRV_NAME);
3450 if (err) {
3451 dev_err(&pdev->dev, "Couldn't get PCI resources, aborting\n");
3452 goto err_disable_pdev;
3453 }
3454
3455 pci_set_master(pdev);
3456
3457 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
3458 if (err) {
3459 dev_warn(&pdev->dev, "Warning: couldn't set 64-bit PCI DMA mask\n");
3460 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
3461 if (err) {
3462 dev_err(&pdev->dev, "Can't set PCI DMA mask, aborting\n");
3463 goto err_release_regions;
3464 }
3465 }
3466 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
3467 if (err) {
3468 dev_warn(&pdev->dev, "Warning: couldn't set 64-bit consistent PCI DMA mask\n");
3469 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
3470 if (err) {
3471 dev_err(&pdev->dev, "Can't set consistent PCI DMA mask, aborting\n");
3472 goto err_release_regions;
3473 }
3474 }
3475
3476 /* Allow large DMA segments, up to the firmware limit of 1 GB */
3477 dma_set_max_seg_size(&pdev->dev, 1024 * 1024 * 1024);
3478 /* Detect if this device is a virtual function */
3479 if (pci_dev_data & MLX4_PCI_DEV_IS_VF) {
3480 /* When acting as pf, we normally skip vfs unless explicitly
3481 * requested to probe them.
3482 */
3483 if (total_vfs) {
3484 unsigned vfs_offset = 0;
3485
3486 for (i = 0; i < sizeof(nvfs)/sizeof(nvfs[0]) &&
3487 vfs_offset + nvfs[i] < extended_func_num(pdev);
3488 vfs_offset += nvfs[i], i++)
3489 ;
3490 if (i == sizeof(nvfs)/sizeof(nvfs[0])) {
3491 err = -ENODEV;
3492 goto err_release_regions;
3493 }
3494 if ((extended_func_num(pdev) - vfs_offset)
3495 > prb_vf[i]) {
3496 dev_warn(&pdev->dev, "Skipping virtual function:%d\n",
3497 extended_func_num(pdev));
3498 err = -ENODEV;
3499 goto err_release_regions;
3500 }
3501 }
3502 }
3503
ad9a0bf0 3504 err = mlx4_catas_init(&priv->dev);
e1c00e10
MD		 3505 if (err)
3506 goto err_release_regions;
ad9a0bf0 3507
55ad3592 3508 err = mlx4_load_one(pdev, pci_dev_data, total_vfs, nvfs, priv, 0);
ad9a0bf0
YH		 3509 if (err)
3510 goto err_catas;
3511
e1c00e10 3512 return 0;
225c7b1f 3513
ad9a0bf0
YH		 3514err_catas:
3515 mlx4_catas_end(&priv->dev);
3516
a01df0fe
RD		 3517err_release_regions:
3518 pci_release_regions(pdev);
225c7b1f
RD		 3519
3520err_disable_pdev:
3521 pci_disable_device(pdev);
3522 pci_set_drvdata(pdev, NULL);
3523 return err;
3524}
3525
1dd06ae8 3526static int mlx4_init_one(struct pci_dev *pdev, const struct pci_device_id *id)
3d73c288 3527{
befdf897
WY		 3528 struct mlx4_priv *priv;
3529 struct mlx4_dev *dev;
e1c00e10 3530 int ret;
befdf897 3531
0a645e80 3532 printk_once(KERN_INFO "%s", mlx4_version);
3d73c288 3533
befdf897
WY		 3534 priv = kzalloc(sizeof(*priv), GFP_KERNEL);
3535 if (!priv)
3536 return -ENOMEM;
3537
3538 dev = &priv->dev;
872bf2fb
YH		 3539 dev->persist = kzalloc(sizeof(*dev->persist), GFP_KERNEL);
3540 if (!dev->persist) {
3541 kfree(priv);
3542 return -ENOMEM;
3543 }
3544 dev->persist->pdev = pdev;
3545 dev->persist->dev = dev;
3546 pci_set_drvdata(pdev, dev->persist);
befdf897 3547 priv->pci_dev_data = id->driver_data;
f6bc11e4 3548 mutex_init(&dev->persist->device_state_mutex);
c69453e2 3549 mutex_init(&dev->persist->interface_state_mutex);
befdf897 3550
e1c00e10 3551 ret = __mlx4_init_one(pdev, id->driver_data, priv);
872bf2fb
YH		 3552 if (ret) {
3553 kfree(dev->persist);
e1c00e10 3554 kfree(priv);
2ba5fbd6
YH		 3555 } else {
3556 pci_save_state(pdev);
872bf2fb 3557 }
2ba5fbd6 3558
e1c00e10 3559 return ret;
3d73c288
RD		 3560}
3561
dd0eefe3
YH		 3562static void mlx4_clean_dev(struct mlx4_dev *dev)
3563{
3564 struct mlx4_dev_persistent *persist = dev->persist;
3565 struct mlx4_priv *priv = mlx4_priv(dev);
55ad3592 3566 unsigned long flags = (dev->flags & RESET_PERSIST_MASK_FLAGS);
dd0eefe3
YH		 3567
3568 memset(priv, 0, sizeof(*priv));
3569 priv->dev.persist = persist;
55ad3592 3570 priv->dev.flags = flags;
dd0eefe3
YH		 3571}
3572
e1c00e10 3573static void mlx4_unload_one(struct pci_dev *pdev)
225c7b1f 3574{
872bf2fb
YH		 3575 struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
3576 struct mlx4_dev *dev = persist->dev;
225c7b1f 3577 struct mlx4_priv *priv = mlx4_priv(dev);
befdf897 3578 int pci_dev_data;
dd0eefe3 3579 int p, i;
225c7b1f 3580
befdf897
WY		 3581 if (priv->removed)
3582 return;
225c7b1f 3583
dd0eefe3
YH		 3584 /* saving current ports type for further use */
3585 for (i = 0; i < dev->caps.num_ports; i++) {
3586 dev->persist->curr_port_type[i] = dev->caps.port_type[i + 1];
3587 dev->persist->curr_port_poss_type[i] = dev->caps.
3588 possible_type[i + 1];
3589 }
3590
befdf897 3591 pci_dev_data = priv->pci_dev_data;
225c7b1f 3592
befdf897
WY		 3593 mlx4_stop_sense(dev);
3594 mlx4_unregister_device(dev);
225c7b1f 3595
befdf897
WY		 3596 for (p = 1; p <= dev->caps.num_ports; p++) {
3597 mlx4_cleanup_port_info(&priv->port[p]);
3598 mlx4_CLOSE_PORT(dev, p);
3599 }
3600
3601 if (mlx4_is_master(dev))
3602 mlx4_free_resource_tracker(dev,
3603 RES_TR_FREE_SLAVES_ONLY);
3604
6de5f7f6 3605 mlx4_cleanup_default_counters(dev);
2632d18d
EBE		 3606 if (!mlx4_is_slave(dev))
3607 mlx4_cleanup_counters_table(dev);
befdf897
WY		 3608 mlx4_cleanup_qp_table(dev);
3609 mlx4_cleanup_srq_table(dev);
3610 mlx4_cleanup_cq_table(dev);
3611 mlx4_cmd_use_polling(dev);
3612 mlx4_cleanup_eq_table(dev);
3613 mlx4_cleanup_mcg_table(dev);
3614 mlx4_cleanup_mr_table(dev);
3615 mlx4_cleanup_xrcd_table(dev);
3616 mlx4_cleanup_pd_table(dev);
225c7b1f 3617
befdf897
WY		 3618 if (mlx4_is_master(dev))
3619 mlx4_free_resource_tracker(dev,
3620 RES_TR_FREE_STRUCTS_ONLY);
47605df9 3621
befdf897
WY		 3622 iounmap(priv->kar);
3623 mlx4_uar_free(dev, &priv->driver_uar);
3624 mlx4_cleanup_uar_table(dev);
3625 if (!mlx4_is_slave(dev))
3626 mlx4_clear_steering(dev);
3627 mlx4_free_eq_table(dev);
3628 if (mlx4_is_master(dev))
3629 mlx4_multi_func_cleanup(dev);
3630 mlx4_close_hca(dev);
a0eacca9 3631 mlx4_close_fw(dev);
befdf897
WY		 3632 if (mlx4_is_slave(dev))
3633 mlx4_multi_func_cleanup(dev);
ffc39f6d 3634 mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
47605df9 3635
befdf897
WY		 3636 if (dev->flags & MLX4_FLAG_MSI_X)
3637 pci_disable_msix(pdev);
befdf897
WY		 3638
3639 if (!mlx4_is_slave(dev))
3640 mlx4_free_ownership(dev);
3641
99ec41d0 3642 kfree(dev->caps.qp0_qkey);
befdf897
WY		 3643 kfree(dev->caps.qp0_tunnel);
3644 kfree(dev->caps.qp0_proxy);
3645 kfree(dev->caps.qp1_tunnel);
3646 kfree(dev->caps.qp1_proxy);
3647 kfree(dev->dev_vfs);
3648
dd0eefe3 3649 mlx4_clean_dev(dev);
befdf897
WY		 3650 priv->pci_dev_data = pci_dev_data;
3651 priv->removed = 1;
3652}
3653
3654static void mlx4_remove_one(struct pci_dev *pdev)
3655{
872bf2fb
YH		 3656 struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
3657 struct mlx4_dev *dev = persist->dev;
befdf897 3658 struct mlx4_priv *priv = mlx4_priv(dev);
55ad3592 3659 int active_vfs = 0;
befdf897 3660
c69453e2
YH		 3661 mutex_lock(&persist->interface_state_mutex);
3662 persist->interface_state |= MLX4_INTERFACE_STATE_DELETION;
3663 mutex_unlock(&persist->interface_state_mutex);
3664
55ad3592
YH		 3665 /* Disabling SR-IOV is not allowed while there are active vf's */
3666 if (mlx4_is_master(dev) && dev->flags & MLX4_FLAG_SRIOV) {
3667 active_vfs = mlx4_how_many_lives_vf(dev);
3668 if (active_vfs) {
3669 pr_warn("Removing PF when there are active VF's !!\n");
3670 pr_warn("Will not disable SR-IOV.\n");
3671 }
3672 }
3673
c69453e2
YH		 3674 /* device marked to be under deletion running now without the lock
3675 * letting other tasks to be terminated
3676 */
3677 if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
3678 mlx4_unload_one(pdev);
3679 else
3680 mlx4_info(dev, "%s: interface is down\n", __func__);
ad9a0bf0 3681 mlx4_catas_end(dev);
55ad3592
YH		 3682 if (dev->flags & MLX4_FLAG_SRIOV && !active_vfs) {
3683 mlx4_warn(dev, "Disabling SR-IOV\n");
3684 pci_disable_sriov(pdev);
3685 }
3686
e1c00e10
MD		 3687 pci_release_regions(pdev);
3688 pci_disable_device(pdev);
872bf2fb 3689 kfree(dev->persist);
befdf897
WY		 3690 kfree(priv);
3691 pci_set_drvdata(pdev, NULL);
225c7b1f
RD		 3692}
3693
dd0eefe3
YH		 3694static int restore_current_port_types(struct mlx4_dev *dev,
3695 enum mlx4_port_type *types,
3696 enum mlx4_port_type *poss_types)
3697{
3698 struct mlx4_priv *priv = mlx4_priv(dev);
3699 int err, i;
3700
3701 mlx4_stop_sense(dev);
3702
3703 mutex_lock(&priv->port_mutex);
3704 for (i = 0; i < dev->caps.num_ports; i++)
3705 dev->caps.possible_type[i + 1] = poss_types[i];
3706 err = mlx4_change_port_types(dev, types);
3707 mlx4_start_sense(dev);
3708 mutex_unlock(&priv->port_mutex);
3709
3710 return err;
3711}
3712
ee49bd93
JM		 3713int mlx4_restart_one(struct pci_dev *pdev)
3714{
872bf2fb
YH		 3715 struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
3716 struct mlx4_dev *dev = persist->dev;
839f1243 3717 struct mlx4_priv *priv = mlx4_priv(dev);
e1c00e10
MD		 3718 int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
3719 int pci_dev_data, err, total_vfs;
839f1243
RD		 3720
3721 pci_dev_data = priv->pci_dev_data;
872bf2fb
YH		 3722 total_vfs = dev->persist->num_vfs;
3723 memcpy(nvfs, dev->persist->nvfs, sizeof(dev->persist->nvfs));
e1c00e10
MD		 3724
3725 mlx4_unload_one(pdev);
55ad3592 3726 err = mlx4_load_one(pdev, pci_dev_data, total_vfs, nvfs, priv, 1);
e1c00e10
MD		 3727 if (err) {
3728 mlx4_err(dev, "%s: ERROR: mlx4_load_one failed, pci_name=%s, err=%d\n",
3729 __func__, pci_name(pdev), err);
3730 return err;
3731 }
3732
dd0eefe3
YH		 3733 err = restore_current_port_types(dev, dev->persist->curr_port_type,
3734 dev->persist->curr_port_poss_type);
3735 if (err)
3736 mlx4_err(dev, "could not restore original port types (%d)\n",
3737 err);
3738
e1c00e10 3739 return err;
ee49bd93
JM		 3740}
3741
9baa3c34 3742static const struct pci_device_id mlx4_pci_table[] = {
ab9c17a0 3743 /* MT25408 "Hermon" SDR */
ca3e57a5 3744 { PCI_VDEVICE(MELLANOX, 0x6340), MLX4_PCI_DEV_FORCE_SENSE_PORT },
ab9c17a0 3745 /* MT25408 "Hermon" DDR */
ca3e57a5 3746 { PCI_VDEVICE(MELLANOX, 0x634a), MLX4_PCI_DEV_FORCE_SENSE_PORT },
ab9c17a0 3747 /* MT25408 "Hermon" QDR */
ca3e57a5 3748 { PCI_VDEVICE(MELLANOX, 0x6354), MLX4_PCI_DEV_FORCE_SENSE_PORT },
ab9c17a0 3749 /* MT25408 "Hermon" DDR PCIe gen2 */
ca3e57a5 3750 { PCI_VDEVICE(MELLANOX, 0x6732), MLX4_PCI_DEV_FORCE_SENSE_PORT },
ab9c17a0 3751 /* MT25408 "Hermon" QDR PCIe gen2 */
ca3e57a5 3752 { PCI_VDEVICE(MELLANOX, 0x673c), MLX4_PCI_DEV_FORCE_SENSE_PORT },
ab9c17a0 3753 /* MT25408 "Hermon" EN 10GigE */
ca3e57a5 3754 { PCI_VDEVICE(MELLANOX, 0x6368), MLX4_PCI_DEV_FORCE_SENSE_PORT },
ab9c17a0 3755 /* MT25408 "Hermon" EN 10GigE PCIe gen2 */
ca3e57a5 3756 { PCI_VDEVICE(MELLANOX, 0x6750), MLX4_PCI_DEV_FORCE_SENSE_PORT },
ab9c17a0 3757 /* MT25458 ConnectX EN 10GBASE-T 10GigE */
ca3e57a5 3758 { PCI_VDEVICE(MELLANOX, 0x6372), MLX4_PCI_DEV_FORCE_SENSE_PORT },
ab9c17a0 3759 /* MT25458 ConnectX EN 10GBASE-T+Gen2 10GigE */
ca3e57a5 3760 { PCI_VDEVICE(MELLANOX, 0x675a), MLX4_PCI_DEV_FORCE_SENSE_PORT },
ab9c17a0 3761 /* MT26468 ConnectX EN 10GigE PCIe gen2*/
ca3e57a5 3762 { PCI_VDEVICE(MELLANOX, 0x6764), MLX4_PCI_DEV_FORCE_SENSE_PORT },
ab9c17a0 3763 /* MT26438 ConnectX EN 40GigE PCIe gen2 5GT/s */
ca3e57a5 3764 { PCI_VDEVICE(MELLANOX, 0x6746), MLX4_PCI_DEV_FORCE_SENSE_PORT },
ab9c17a0 3765 /* MT26478 ConnectX2 40GigE PCIe gen2 */
ca3e57a5 3766 { PCI_VDEVICE(MELLANOX, 0x676e), MLX4_PCI_DEV_FORCE_SENSE_PORT },
ab9c17a0 3767 /* MT25400 Family [ConnectX-2 Virtual Function] */
839f1243 3768 { PCI_VDEVICE(MELLANOX, 0x1002), MLX4_PCI_DEV_IS_VF },
ab9c17a0
JM		 3769 /* MT27500 Family [ConnectX-3] */
3770 { PCI_VDEVICE(MELLANOX, 0x1003), 0 },
3771 /* MT27500 Family [ConnectX-3 Virtual Function] */
839f1243 3772 { PCI_VDEVICE(MELLANOX, 0x1004), MLX4_PCI_DEV_IS_VF },
ab9c17a0
JM		 3773 { PCI_VDEVICE(MELLANOX, 0x1005), 0 }, /* MT27510 Family */
3774 { PCI_VDEVICE(MELLANOX, 0x1006), 0 }, /* MT27511 Family */
3775 { PCI_VDEVICE(MELLANOX, 0x1007), 0 }, /* MT27520 Family */
3776 { PCI_VDEVICE(MELLANOX, 0x1008), 0 }, /* MT27521 Family */
3777 { PCI_VDEVICE(MELLANOX, 0x1009), 0 }, /* MT27530 Family */
3778 { PCI_VDEVICE(MELLANOX, 0x100a), 0 }, /* MT27531 Family */
3779 { PCI_VDEVICE(MELLANOX, 0x100b), 0 }, /* MT27540 Family */
3780 { PCI_VDEVICE(MELLANOX, 0x100c), 0 }, /* MT27541 Family */
3781 { PCI_VDEVICE(MELLANOX, 0x100d), 0 }, /* MT27550 Family */
3782 { PCI_VDEVICE(MELLANOX, 0x100e), 0 }, /* MT27551 Family */
3783 { PCI_VDEVICE(MELLANOX, 0x100f), 0 }, /* MT27560 Family */
3784 { PCI_VDEVICE(MELLANOX, 0x1010), 0 }, /* MT27561 Family */
225c7b1f
RD		 3785 { 0, }
3786};
3787
3788MODULE_DEVICE_TABLE(pci, mlx4_pci_table);
3789
57dbf29a
KSS		 3790static pci_ers_result_t mlx4_pci_err_detected(struct pci_dev *pdev,
3791 pci_channel_state_t state)
3792{
2ba5fbd6
YH		 3793 struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
3794
3795 mlx4_err(persist->dev, "mlx4_pci_err_detected was called\n");
3796 mlx4_enter_error_state(persist);
57dbf29a 3797
2ba5fbd6
YH		 3798 mutex_lock(&persist->interface_state_mutex);
3799 if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
3800 mlx4_unload_one(pdev);
3801
3802 mutex_unlock(&persist->interface_state_mutex);
3803 if (state == pci_channel_io_perm_failure)
3804 return PCI_ERS_RESULT_DISCONNECT;
3805
3806 pci_disable_device(pdev);
3807 return PCI_ERS_RESULT_NEED_RESET;
57dbf29a
KSS		 3808}
3809
3810static pci_ers_result_t mlx4_pci_slot_reset(struct pci_dev *pdev)
3811{
2ba5fbd6
YH		 3812 struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
3813 struct mlx4_dev *dev = persist->dev;
befdf897
WY		 3814 struct mlx4_priv *priv = mlx4_priv(dev);
3815 int ret;
2ba5fbd6
YH		 3816 int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
3817 int total_vfs;
97a5221f 3818
2ba5fbd6
YH		 3819 mlx4_err(dev, "mlx4_pci_slot_reset was called\n");
3820 ret = pci_enable_device(pdev);
3821 if (ret) {
3822 mlx4_err(dev, "Can not re-enable device, ret=%d\n", ret);
3823 return PCI_ERS_RESULT_DISCONNECT;
3824 }
3825
3826 pci_set_master(pdev);
3827 pci_restore_state(pdev);
3828 pci_save_state(pdev);
3829
3830 total_vfs = dev->persist->num_vfs;
3831 memcpy(nvfs, dev->persist->nvfs, sizeof(dev->persist->nvfs));
3832
3833 mutex_lock(&persist->interface_state_mutex);
3834 if (!(persist->interface_state & MLX4_INTERFACE_STATE_UP)) {
3835 ret = mlx4_load_one(pdev, priv->pci_dev_data, total_vfs, nvfs,
55ad3592 3836 priv, 1);
2ba5fbd6
YH		 3837 if (ret) {
3838 mlx4_err(dev, "%s: mlx4_load_one failed, ret=%d\n",
3839 __func__, ret);
3840 goto end;
3841 }
3842
3843 ret = restore_current_port_types(dev, dev->persist->
3844 curr_port_type, dev->persist->
3845 curr_port_poss_type);
3846 if (ret)
3847 mlx4_err(dev, "could not restore original port types (%d)\n", ret);
3848 }
3849end:
3850 mutex_unlock(&persist->interface_state_mutex);
57dbf29a
KSS		 3851
3852 return ret ? PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_RECOVERED;
3853}
3854
2ba5fbd6
YH		 3855static void mlx4_shutdown(struct pci_dev *pdev)
3856{
3857 struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
3858
3859 mlx4_info(persist->dev, "mlx4_shutdown was called\n");
3860 mutex_lock(&persist->interface_state_mutex);
3861 if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
3862 mlx4_unload_one(pdev);
3863 mutex_unlock(&persist->interface_state_mutex);
3864}
3865
3646f0e5 3866static const struct pci_error_handlers mlx4_err_handler = {
57dbf29a
KSS		 3867 .error_detected = mlx4_pci_err_detected,
3868 .slot_reset = mlx4_pci_slot_reset,
3869};
3870
225c7b1f
RD		 3871static struct pci_driver mlx4_driver = {
3872 .name = DRV_NAME,
3873 .id_table = mlx4_pci_table,
3874 .probe = mlx4_init_one,
2ba5fbd6 3875 .shutdown = mlx4_shutdown,
f57e6848 3876 .remove = mlx4_remove_one,
57dbf29a 3877 .err_handler = &mlx4_err_handler,
225c7b1f
RD		 3878};
3879
7ff93f8b
YP		 3880static int __init mlx4_verify_params(void)
3881{
3882 if ((log_num_mac < 0) || (log_num_mac > 7)) {
c20862c8 3883 pr_warn("mlx4_core: bad num_mac: %d\n", log_num_mac);
7ff93f8b
YP		 3884 return -1;
3885 }
3886
cb29688a 3887 if (log_num_vlan != 0)
c20862c8
AV		 3888 pr_warn("mlx4_core: log_num_vlan - obsolete module param, using %d\n",
3889 MLX4_LOG_NUM_VLANS);
7ff93f8b 3890
ecc8fb11
AV		 3891 if (use_prio != 0)
3892 pr_warn("mlx4_core: use_prio - obsolete module param, ignored\n");
7ff93f8b 3893
0498628f 3894 if ((log_mtts_per_seg < 1) || (log_mtts_per_seg > 7)) {
c20862c8
AV		 3895 pr_warn("mlx4_core: bad log_mtts_per_seg: %d\n",
3896 log_mtts_per_seg);
ab6bf42e
EC		 3897 return -1;
3898 }
3899
ab9c17a0
JM		 3900 /* Check if module param for ports type has legal combination */
3901 if (port_type_array[0] == false && port_type_array[1] == true) {
c20862c8 3902 pr_warn("Module parameter configuration ETH/IB is not supported. Switching to default configuration IB/IB\n");
ab9c17a0
JM		 3903 port_type_array[0] = true;
3904 }
3905
7d077cd3
MB		 3906 if (mlx4_log_num_mgm_entry_size < -7 ||
3907 (mlx4_log_num_mgm_entry_size > 0 &&
3908 (mlx4_log_num_mgm_entry_size < MLX4_MIN_MGM_LOG_ENTRY_SIZE ||
3909 mlx4_log_num_mgm_entry_size > MLX4_MAX_MGM_LOG_ENTRY_SIZE))) {
3910 pr_warn("mlx4_core: mlx4_log_num_mgm_entry_size (%d) not in legal range (-7..0 or %d..%d)\n",
1a91de28
JP		 3911 mlx4_log_num_mgm_entry_size,
3912 MLX4_MIN_MGM_LOG_ENTRY_SIZE,
3913 MLX4_MAX_MGM_LOG_ENTRY_SIZE);
3c439b55
JM		 3914 return -1;
3915 }
3916
7ff93f8b
YP		 3917 return 0;
3918}
3919
225c7b1f
RD		 3920static int __init mlx4_init(void)
3921{
3922 int ret;
3923
7ff93f8b
YP		 3924 if (mlx4_verify_params())
3925 return -EINVAL;
3926
27bf91d6
YP		 3927
3928 mlx4_wq = create_singlethread_workqueue("mlx4");
3929 if (!mlx4_wq)
3930 return -ENOMEM;
ee49bd93 3931
225c7b1f 3932 ret = pci_register_driver(&mlx4_driver);
1b85ee09
WY		 3933 if (ret < 0)
3934 destroy_workqueue(mlx4_wq);
225c7b1f
RD		 3935 return ret < 0 ? ret : 0;
3936}
3937
3938static void __exit mlx4_cleanup(void)
3939{
3940 pci_unregister_driver(&mlx4_driver);
27bf91d6 3941 destroy_workqueue(mlx4_wq);
225c7b1f
RD		 3942}
3943
3944module_init(mlx4_init);
3945module_exit(mlx4_cleanup);
Linux kernel - Deliverables
This page took 0.984876 seconds and 5 git commands to generate.