projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
net/mlx4_core: Add sink counter
[linux-2.6-block.git] / drivers / net / ethernet / mellanox / mlx4 / main.c
CommitLineData
225c7b1f
RD		 1/*
2 * Copyright (c) 2004, 2005 Topspin Communications. All rights reserved.
3 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
51a379d0 4 * Copyright (c) 2005, 2006, 2007, 2008 Mellanox Technologies. All rights reserved.
225c7b1f
RD		 5 * Copyright (c) 2006, 2007 Cisco Systems, Inc. All rights reserved.
6 *
7 * This software is available to you under a choice of one of two
8 * licenses. You may choose to be licensed under the terms of the GNU
9 * General Public License (GPL) Version 2, available from the file
10 * COPYING in the main directory of this source tree, or the
11 * OpenIB.org BSD license below:
12 *
13 * Redistribution and use in source and binary forms, with or
14 * without modification, are permitted provided that the following
15 * conditions are met:
16 *
17 * - Redistributions of source code must retain the above
18 * copyright notice, this list of conditions and the following
19 * disclaimer.
20 *
21 * - Redistributions in binary form must reproduce the above
22 * copyright notice, this list of conditions and the following
23 * disclaimer in the documentation and/or other materials
24 * provided with the distribution.
25 *
26 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
27 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
28 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
29 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
30 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
31 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
32 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
33 * SOFTWARE.
34 */
35
36#include <linux/module.h>
37#include <linux/init.h>
38#include <linux/errno.h>
39#include <linux/pci.h>
40#include <linux/dma-mapping.h>
5a0e3ad6 41#include <linux/slab.h>
c1b43dca 42#include <linux/io-mapping.h>
ab9c17a0 43#include <linux/delay.h>
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
47d8417f 482 dev->caps.max_counters = dev_cap->max_counters;
f2a3f6a3 483
93fc9e1b
YP		 484 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] = dev_cap->reserved_qps;
485 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] =
486 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] =
487 (1 << dev->caps.log_num_macs) *
488 (1 << dev->caps.log_num_vlans) *
93fc9e1b
YP		 489 dev->caps.num_ports;
490 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH] = MLX4_NUM_FEXCH;
7d077cd3
MB		 491
492 if (dev_cap->dmfs_high_rate_qpn_base > 0 &&
493 dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN)
494 dev->caps.dmfs_high_rate_qpn_base = dev_cap->dmfs_high_rate_qpn_base;
495 else
496 dev->caps.dmfs_high_rate_qpn_base =
497 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
498
499 if (dev_cap->dmfs_high_rate_qpn_range > 0 &&
500 dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN) {
501 dev->caps.dmfs_high_rate_qpn_range = dev_cap->dmfs_high_rate_qpn_range;
502 dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_DEFAULT;
503 dev->caps.flags2 |= MLX4_DEV_CAP_FLAG2_FS_A0;
504 } else {
505 dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_NOT_SUPPORTED;
506 dev->caps.dmfs_high_rate_qpn_base =
507 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
508 dev->caps.dmfs_high_rate_qpn_range = MLX4_A0_STEERING_TABLE_SIZE;
509 }
510
fc31e256
OG		 511 dev->caps.rl_caps = dev_cap->rl_caps;
512
d57febe1 513 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_RSS_RAW_ETH] =
7d077cd3 514 dev->caps.dmfs_high_rate_qpn_range;
93fc9e1b
YP		 515
516 dev->caps.reserved_qps = dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] +
517 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] +
518 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] +
519 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH];
520
e2c76824 521 dev->caps.sqp_demux = (mlx4_is_master(dev)) ? MLX4_MAX_NUM_SLAVES : 0;
08ff3235 522
b3051320 523 if (!enable_64b_cqe_eqe && !mlx4_is_slave(dev)) {
08ff3235
OG		 524 if (dev_cap->flags &
525 (MLX4_DEV_CAP_FLAG_64B_CQE | MLX4_DEV_CAP_FLAG_64B_EQE)) {
526 mlx4_warn(dev, "64B EQEs/CQEs supported by the device but not enabled\n");
527 dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_64B_CQE;
528 dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_64B_EQE;
529 }
77507aa2
IS		 530
531 if (dev_cap->flags2 &
532 (MLX4_DEV_CAP_FLAG2_CQE_STRIDE |
533 MLX4_DEV_CAP_FLAG2_EQE_STRIDE)) {
534 mlx4_warn(dev, "Disabling EQE/CQE stride per user request\n");
535 dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
536 dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
537 }
08ff3235
OG		 538 }
539
f97b4b5d 540 if ((dev->caps.flags &
08ff3235
OG		 541 (MLX4_DEV_CAP_FLAG_64B_CQE | MLX4_DEV_CAP_FLAG_64B_EQE)) &&
542 mlx4_is_master(dev))
543 dev->caps.function_caps |= MLX4_FUNC_CAP_64B_EQE_CQE;
544
ddae0349 545 if (!mlx4_is_slave(dev)) {
77507aa2 546 mlx4_enable_cqe_eqe_stride(dev);
ddae0349 547 dev->caps.alloc_res_qp_mask =
d57febe1
MB		 548 (dev->caps.bf_reg_size ? MLX4_RESERVE_ETH_BF_QP : 0) |
549 MLX4_RESERVE_A0_QP;
3742cc65
IS		 550
551 if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ETS_CFG) &&
552 dev->caps.flags & MLX4_DEV_CAP_FLAG_SET_ETH_SCHED) {
553 mlx4_warn(dev, "Old device ETS support detected\n");
554 mlx4_warn(dev, "Consider upgrading device FW.\n");
555 dev->caps.flags2 |= MLX4_DEV_CAP_FLAG2_ETS_CFG;
556 }
557
ddae0349
EE		 558 } else {
559 dev->caps.alloc_res_qp_mask = 0;
560 }
77507aa2 561
78500b8c
MM		 562 mlx4_enable_ignore_fcs(dev);
563
225c7b1f
RD		 564 return 0;
565}
b912b2f8
EP		 566
567static int mlx4_get_pcie_dev_link_caps(struct mlx4_dev *dev,
568 enum pci_bus_speed *speed,
569 enum pcie_link_width *width)
570{
571 u32 lnkcap1, lnkcap2;
572 int err1, err2;
573
574#define PCIE_MLW_CAP_SHIFT 4 /* start of MLW mask in link capabilities */
575
576 *speed = PCI_SPEED_UNKNOWN;
577 *width = PCIE_LNK_WIDTH_UNKNOWN;
578
872bf2fb
YH		 579 err1 = pcie_capability_read_dword(dev->persist->pdev, PCI_EXP_LNKCAP,
580 &lnkcap1);
581 err2 = pcie_capability_read_dword(dev->persist->pdev, PCI_EXP_LNKCAP2,
582 &lnkcap2);
b912b2f8
EP		 583 if (!err2 && lnkcap2) { /* PCIe r3.0-compliant */
584 if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_8_0GB)
585 *speed = PCIE_SPEED_8_0GT;
586 else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_5_0GB)
587 *speed = PCIE_SPEED_5_0GT;
588 else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_2_5GB)
589 *speed = PCIE_SPEED_2_5GT;
590 }
591 if (!err1) {
592 *width = (lnkcap1 & PCI_EXP_LNKCAP_MLW) >> PCIE_MLW_CAP_SHIFT;
593 if (!lnkcap2) { /* pre-r3.0 */
594 if (lnkcap1 & PCI_EXP_LNKCAP_SLS_5_0GB)
595 *speed = PCIE_SPEED_5_0GT;
596 else if (lnkcap1 & PCI_EXP_LNKCAP_SLS_2_5GB)
597 *speed = PCIE_SPEED_2_5GT;
598 }
599 }
600
601 if (*speed == PCI_SPEED_UNKNOWN || *width == PCIE_LNK_WIDTH_UNKNOWN) {
602 return err1 ? err1 :
603 err2 ? err2 : -EINVAL;
604 }
605 return 0;
606}
607
608static void mlx4_check_pcie_caps(struct mlx4_dev *dev)
609{
610 enum pcie_link_width width, width_cap;
611 enum pci_bus_speed speed, speed_cap;
612 int err;
613
614#define PCIE_SPEED_STR(speed) \
615 (speed == PCIE_SPEED_8_0GT ? "8.0GT/s" : \
616 speed == PCIE_SPEED_5_0GT ? "5.0GT/s" : \
617 speed == PCIE_SPEED_2_5GT ? "2.5GT/s" : \
618 "Unknown")
619
620 err = mlx4_get_pcie_dev_link_caps(dev, &speed_cap, &width_cap);
621 if (err) {
622 mlx4_warn(dev,
623 "Unable to determine PCIe device BW capabilities\n");
624 return;
625 }
626
872bf2fb 627 err = pcie_get_minimum_link(dev->persist->pdev, &speed, &width);
b912b2f8
EP		 628 if (err || speed == PCI_SPEED_UNKNOWN ||
629 width == PCIE_LNK_WIDTH_UNKNOWN) {
630 mlx4_warn(dev,
631 "Unable to determine PCI device chain minimum BW\n");
632 return;
633 }
634
635 if (width != width_cap || speed != speed_cap)
636 mlx4_warn(dev,
637 "PCIe BW is different than device's capability\n");
638
639 mlx4_info(dev, "PCIe link speed is %s, device supports %s\n",
640 PCIE_SPEED_STR(speed), PCIE_SPEED_STR(speed_cap));
641 mlx4_info(dev, "PCIe link width is x%d, device supports x%d\n",
642 width, width_cap);
643 return;
644}
645
ab9c17a0
JM		 646/*The function checks if there are live vf, return the num of them*/
647static int mlx4_how_many_lives_vf(struct mlx4_dev *dev)
648{
649 struct mlx4_priv *priv = mlx4_priv(dev);
650 struct mlx4_slave_state *s_state;
651 int i;
652 int ret = 0;
653
654 for (i = 1/*the ppf is 0*/; i < dev->num_slaves; ++i) {
655 s_state = &priv->mfunc.master.slave_state[i];
656 if (s_state->active && s_state->last_cmd !=
657 MLX4_COMM_CMD_RESET) {
658 mlx4_warn(dev, "%s: slave: %d is still active\n",
659 __func__, i);
660 ret++;
661 }
662 }
663 return ret;
664}
665
396f2feb
JM		 666int mlx4_get_parav_qkey(struct mlx4_dev *dev, u32 qpn, u32 *qkey)
667{
668 u32 qk = MLX4_RESERVED_QKEY_BASE;
47605df9
JM		 669
670 if (qpn >= dev->phys_caps.base_tunnel_sqpn + 8 * MLX4_MFUNC_MAX ||
671 qpn < dev->phys_caps.base_proxy_sqpn)
396f2feb
JM		 672 return -EINVAL;
673
47605df9 674 if (qpn >= dev->phys_caps.base_tunnel_sqpn)
396f2feb 675 /* tunnel qp */
47605df9 676 qk += qpn - dev->phys_caps.base_tunnel_sqpn;
396f2feb 677 else
47605df9 678 qk += qpn - dev->phys_caps.base_proxy_sqpn;
396f2feb
JM		 679 *qkey = qk;
680 return 0;
681}
682EXPORT_SYMBOL(mlx4_get_parav_qkey);
683
54679e14
JM		 684void mlx4_sync_pkey_table(struct mlx4_dev *dev, int slave, int port, int i, int val)
685{
686 struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
687
688 if (!mlx4_is_master(dev))
689 return;
690
691 priv->virt2phys_pkey[slave][port - 1][i] = val;
692}
693EXPORT_SYMBOL(mlx4_sync_pkey_table);
694
afa8fd1d
JM		 695void mlx4_put_slave_node_guid(struct mlx4_dev *dev, int slave, __be64 guid)
696{
697 struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
698
699 if (!mlx4_is_master(dev))
700 return;
701
702 priv->slave_node_guids[slave] = guid;
703}
704EXPORT_SYMBOL(mlx4_put_slave_node_guid);
705
706__be64 mlx4_get_slave_node_guid(struct mlx4_dev *dev, int slave)
707{
708 struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
709
710 if (!mlx4_is_master(dev))
711 return 0;
712
713 return priv->slave_node_guids[slave];
714}
715EXPORT_SYMBOL(mlx4_get_slave_node_guid);
716
e10903b0 717int mlx4_is_slave_active(struct mlx4_dev *dev, int slave)
ab9c17a0
JM		 718{
719 struct mlx4_priv *priv = mlx4_priv(dev);
720 struct mlx4_slave_state *s_slave;
721
722 if (!mlx4_is_master(dev))
723 return 0;
724
725 s_slave = &priv->mfunc.master.slave_state[slave];
726 return !!s_slave->active;
727}
728EXPORT_SYMBOL(mlx4_is_slave_active);
729
7b8157be
JM		 730static void slave_adjust_steering_mode(struct mlx4_dev *dev,
731 struct mlx4_dev_cap *dev_cap,
732 struct mlx4_init_hca_param *hca_param)
733{
734 dev->caps.steering_mode = hca_param->steering_mode;
735 if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED) {
736 dev->caps.num_qp_per_mgm = dev_cap->fs_max_num_qp_per_entry;
737 dev->caps.fs_log_max_ucast_qp_range_size =
738 dev_cap->fs_log_max_ucast_qp_range_size;
739 } else
740 dev->caps.num_qp_per_mgm =
741 4 * ((1 << hca_param->log_mc_entry_sz)/16 - 2);
742
743 mlx4_dbg(dev, "Steering mode is: %s\n",
744 mlx4_steering_mode_str(dev->caps.steering_mode));
745}
746
ab9c17a0
JM		 747static int mlx4_slave_cap(struct mlx4_dev *dev)
748{
749 int err;
750 u32 page_size;
751 struct mlx4_dev_cap dev_cap;
752 struct mlx4_func_cap func_cap;
753 struct mlx4_init_hca_param hca_param;
225c6c8c 754 u8 i;
ab9c17a0
JM		 755
756 memset(&hca_param, 0, sizeof(hca_param));
757 err = mlx4_QUERY_HCA(dev, &hca_param);
758 if (err) {
1a91de28 759 mlx4_err(dev, "QUERY_HCA command failed, aborting\n");
ab9c17a0
JM		 760 return err;
761 }
762
483e0132
EP		 763 /* fail if the hca has an unknown global capability
764 * at this time global_caps should be always zeroed
765 */
766 if (hca_param.global_caps) {
ab9c17a0
JM		 767 mlx4_err(dev, "Unknown hca global capabilities\n");
768 return -ENOSYS;
769 }
770
771 mlx4_log_num_mgm_entry_size = hca_param.log_mc_entry_sz;
772
ddd8a6c1
EE		 773 dev->caps.hca_core_clock = hca_param.hca_core_clock;
774
ab9c17a0 775 memset(&dev_cap, 0, sizeof(dev_cap));
b91cb3eb 776 dev->caps.max_qp_dest_rdma = 1 << hca_param.log_rd_per_qp;
ab9c17a0
JM		 777 err = mlx4_dev_cap(dev, &dev_cap);
778 if (err) {
1a91de28 779 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
ab9c17a0
JM		 780 return err;
781 }
782
b91cb3eb
JM		 783 err = mlx4_QUERY_FW(dev);
784 if (err)
1a91de28 785 mlx4_err(dev, "QUERY_FW command failed: could not get FW version\n");
b91cb3eb 786
ab9c17a0
JM		 787 page_size = ~dev->caps.page_size_cap + 1;
788 mlx4_warn(dev, "HCA minimum page size:%d\n", page_size);
789 if (page_size > PAGE_SIZE) {
1a91de28 790 mlx4_err(dev, "HCA minimum page size of %d bigger than kernel PAGE_SIZE of %ld, aborting\n",
ab9c17a0
JM		 791 page_size, PAGE_SIZE);
792 return -ENODEV;
793 }
794
795 /* slave gets uar page size from QUERY_HCA fw command */
796 dev->caps.uar_page_size = 1 << (hca_param.uar_page_sz + 12);
797
798 /* TODO: relax this assumption */
799 if (dev->caps.uar_page_size != PAGE_SIZE) {
800 mlx4_err(dev, "UAR size:%d != kernel PAGE_SIZE of %ld\n",
801 dev->caps.uar_page_size, PAGE_SIZE);
802 return -ENODEV;
803 }
804
805 memset(&func_cap, 0, sizeof(func_cap));
47605df9 806 err = mlx4_QUERY_FUNC_CAP(dev, 0, &func_cap);
ab9c17a0 807 if (err) {
1a91de28
JP		 808 mlx4_err(dev, "QUERY_FUNC_CAP general command failed, aborting (%d)\n",
809 err);
ab9c17a0
JM		 810 return err;
811 }
812
813 if ((func_cap.pf_context_behaviour | PF_CONTEXT_BEHAVIOUR_MASK) !=
814 PF_CONTEXT_BEHAVIOUR_MASK) {
7d077cd3
MB		 815 mlx4_err(dev, "Unknown pf context behaviour %x known flags %x\n",
816 func_cap.pf_context_behaviour, PF_CONTEXT_BEHAVIOUR_MASK);
ab9c17a0
JM		 817 return -ENOSYS;
818 }
819
ab9c17a0 820 dev->caps.num_ports = func_cap.num_ports;
5a0d0a61
JM		 821 dev->quotas.qp = func_cap.qp_quota;
822 dev->quotas.srq = func_cap.srq_quota;
823 dev->quotas.cq = func_cap.cq_quota;
824 dev->quotas.mpt = func_cap.mpt_quota;
825 dev->quotas.mtt = func_cap.mtt_quota;
826 dev->caps.num_qps = 1 << hca_param.log_num_qps;
827 dev->caps.num_srqs = 1 << hca_param.log_num_srqs;
828 dev->caps.num_cqs = 1 << hca_param.log_num_cqs;
829 dev->caps.num_mpts = 1 << hca_param.log_mpt_sz;
830 dev->caps.num_eqs = func_cap.max_eq;
831 dev->caps.reserved_eqs = func_cap.reserved_eq;
f0ce0615 832 dev->caps.reserved_lkey = func_cap.reserved_lkey;
ab9c17a0
JM		 833 dev->caps.num_pds = MLX4_NUM_PDS;
834 dev->caps.num_mgms = 0;
835 dev->caps.num_amgms = 0;
836
ab9c17a0 837 if (dev->caps.num_ports > MLX4_MAX_PORTS) {
1a91de28
JP		 838 mlx4_err(dev, "HCA has %d ports, but we only support %d, aborting\n",
839 dev->caps.num_ports, MLX4_MAX_PORTS);
ab9c17a0
JM		 840 return -ENODEV;
841 }
842
99ec41d0 843 dev->caps.qp0_qkey = kcalloc(dev->caps.num_ports, sizeof(u32), GFP_KERNEL);
47605df9
JM		 844 dev->caps.qp0_tunnel = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
845 dev->caps.qp0_proxy = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
846 dev->caps.qp1_tunnel = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
847 dev->caps.qp1_proxy = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
848
849 if (!dev->caps.qp0_tunnel || !dev->caps.qp0_proxy ||
99ec41d0
JM		 850 !dev->caps.qp1_tunnel || !dev->caps.qp1_proxy ||
851 !dev->caps.qp0_qkey) {
47605df9
JM		 852 err = -ENOMEM;
853 goto err_mem;
854 }
855
6634961c 856 for (i = 1; i <= dev->caps.num_ports; ++i) {
225c6c8c 857 err = mlx4_QUERY_FUNC_CAP(dev, i, &func_cap);
47605df9 858 if (err) {
1a91de28
JP		 859 mlx4_err(dev, "QUERY_FUNC_CAP port command failed for port %d, aborting (%d)\n",
860 i, err);
47605df9
JM		 861 goto err_mem;
862 }
99ec41d0 863 dev->caps.qp0_qkey[i - 1] = func_cap.qp0_qkey;
47605df9
JM		 864 dev->caps.qp0_tunnel[i - 1] = func_cap.qp0_tunnel_qpn;
865 dev->caps.qp0_proxy[i - 1] = func_cap.qp0_proxy_qpn;
866 dev->caps.qp1_tunnel[i - 1] = func_cap.qp1_tunnel_qpn;
867 dev->caps.qp1_proxy[i - 1] = func_cap.qp1_proxy_qpn;
6230bb23 868 dev->caps.port_mask[i] = dev->caps.port_type[i];
8e1a28e8 869 dev->caps.phys_port_id[i] = func_cap.phys_port_id;
6634961c
JM		 870 if (mlx4_get_slave_pkey_gid_tbl_len(dev, i,
871 &dev->caps.gid_table_len[i],
872 &dev->caps.pkey_table_len[i]))
47605df9 873 goto err_mem;
6634961c 874 }
6230bb23 875
ab9c17a0
JM		 876 if (dev->caps.uar_page_size * (dev->caps.num_uars -
877 dev->caps.reserved_uars) >
872bf2fb
YH		 878 pci_resource_len(dev->persist->pdev,
879 2)) {
1a91de28 880 mlx4_err(dev, "HCA reported UAR region size of 0x%x bigger than PCI resource 2 size of 0x%llx, aborting\n",
ab9c17a0 881 dev->caps.uar_page_size * dev->caps.num_uars,
872bf2fb
YH		 882 (unsigned long long)
883 pci_resource_len(dev->persist->pdev, 2));
47605df9 884 goto err_mem;
ab9c17a0
JM		 885 }
886
08ff3235
OG		 887 if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_64B_EQE_ENABLED) {
888 dev->caps.eqe_size = 64;
889 dev->caps.eqe_factor = 1;
890 } else {
891 dev->caps.eqe_size = 32;
892 dev->caps.eqe_factor = 0;
893 }
894
895 if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_64B_CQE_ENABLED) {
896 dev->caps.cqe_size = 64;
77507aa2 897 dev->caps.userspace_caps |= MLX4_USER_DEV_CAP_LARGE_CQE;
08ff3235
OG		 898 } else {
899 dev->caps.cqe_size = 32;
900 }
901
77507aa2
IS		 902 if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_EQE_STRIDE_ENABLED) {
903 dev->caps.eqe_size = hca_param.eqe_size;
904 dev->caps.eqe_factor = 0;
905 }
906
907 if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_CQE_STRIDE_ENABLED) {
908 dev->caps.cqe_size = hca_param.cqe_size;
909 /* User still need to know when CQE > 32B */
910 dev->caps.userspace_caps |= MLX4_USER_DEV_CAP_LARGE_CQE;
911 }
912
f9bd2d7f 913 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
1a91de28 914 mlx4_warn(dev, "Timestamping is not supported in slave mode\n");
f9bd2d7f 915
7b8157be 916 slave_adjust_steering_mode(dev, &dev_cap, &hca_param);
802f42a8
IS		 917 mlx4_dbg(dev, "RSS support for IP fragments is %s\n",
918 hca_param.rss_ip_frags ? "on" : "off");
7b8157be 919
ddae0349
EE		 920 if (func_cap.extra_flags & MLX4_QUERY_FUNC_FLAGS_BF_RES_QP &&
921 dev->caps.bf_reg_size)
922 dev->caps.alloc_res_qp_mask |= MLX4_RESERVE_ETH_BF_QP;
923
d57febe1
MB		 924 if (func_cap.extra_flags & MLX4_QUERY_FUNC_FLAGS_A0_RES_QP)
925 dev->caps.alloc_res_qp_mask |= MLX4_RESERVE_A0_QP;
926
ab9c17a0 927 return 0;
47605df9
JM		 928
929err_mem:
99ec41d0 930 kfree(dev->caps.qp0_qkey);
47605df9
JM		 931 kfree(dev->caps.qp0_tunnel);
932 kfree(dev->caps.qp0_proxy);
933 kfree(dev->caps.qp1_tunnel);
934 kfree(dev->caps.qp1_proxy);
99ec41d0
JM		 935 dev->caps.qp0_qkey = NULL;
936 dev->caps.qp0_tunnel = NULL;
937 dev->caps.qp0_proxy = NULL;
938 dev->caps.qp1_tunnel = NULL;
939 dev->caps.qp1_proxy = NULL;
47605df9
JM		 940
941 return err;
ab9c17a0 942}
225c7b1f 943
b046ffe5
EP		 944static void mlx4_request_modules(struct mlx4_dev *dev)
945{
946 int port;
947 int has_ib_port = false;
948 int has_eth_port = false;
949#define EN_DRV_NAME "mlx4_en"
950#define IB_DRV_NAME "mlx4_ib"
951
952 for (port = 1; port <= dev->caps.num_ports; port++) {
953 if (dev->caps.port_type[port] == MLX4_PORT_TYPE_IB)
954 has_ib_port = true;
955 else if (dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH)
956 has_eth_port = true;
957 }
958
b046ffe5
EP		 959 if (has_eth_port)
960 request_module_nowait(EN_DRV_NAME);
f24f790f
OG		 961 if (has_ib_port || (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE))
962 request_module_nowait(IB_DRV_NAME);
b046ffe5
EP		 963}
964
7ff93f8b
YP		 965/*
966 * Change the port configuration of the device.
967 * Every user of this function must hold the port mutex.
968 */
27bf91d6
YP		 969int mlx4_change_port_types(struct mlx4_dev *dev,
970 enum mlx4_port_type *port_types)
7ff93f8b
YP		 971{
972 int err = 0;
973 int change = 0;
974 int port;
975
976 for (port = 0; port < dev->caps.num_ports; port++) {
27bf91d6
YP		 977 /* Change the port type only if the new type is different
978 * from the current, and not set to Auto */
3d8f9308 979 if (port_types[port] != dev->caps.port_type[port + 1])
7ff93f8b 980 change = 1;
7ff93f8b
YP		 981 }
982 if (change) {
983 mlx4_unregister_device(dev);
984 for (port = 1; port <= dev->caps.num_ports; port++) {
985 mlx4_CLOSE_PORT(dev, port);
1e0f03d5 986 dev->caps.port_type[port] = port_types[port - 1];
6634961c 987 err = mlx4_SET_PORT(dev, port, -1);
7ff93f8b 988 if (err) {
1a91de28
JP		 989 mlx4_err(dev, "Failed to set port %d, aborting\n",
990 port);
7ff93f8b
YP		 991 goto out;
992 }
993 }
994 mlx4_set_port_mask(dev);
995 err = mlx4_register_device(dev);
b046ffe5
EP		 996 if (err) {
997 mlx4_err(dev, "Failed to register device\n");
998 goto out;
999 }
1000 mlx4_request_modules(dev);
7ff93f8b
YP		 1001 }
1002
1003out:
1004 return err;
1005}
1006
1007static ssize_t show_port_type(struct device *dev,
1008 struct device_attribute *attr,
1009 char *buf)
1010{
1011 struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1012 port_attr);
1013 struct mlx4_dev *mdev = info->dev;
27bf91d6
YP		 1014 char type[8];
1015
1016 sprintf(type, "%s",
1017 (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_IB) ?
1018 "ib" : "eth");
1019 if (mdev->caps.possible_type[info->port] == MLX4_PORT_TYPE_AUTO)
1020 sprintf(buf, "auto (%s)\n", type);
1021 else
1022 sprintf(buf, "%s\n", type);
7ff93f8b 1023
27bf91d6 1024 return strlen(buf);
7ff93f8b
YP		 1025}
1026
1027static ssize_t set_port_type(struct device *dev,
1028 struct device_attribute *attr,
1029 const char *buf, size_t count)
1030{
1031 struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1032 port_attr);
1033 struct mlx4_dev *mdev = info->dev;
1034 struct mlx4_priv *priv = mlx4_priv(mdev);
1035 enum mlx4_port_type types[MLX4_MAX_PORTS];
27bf91d6 1036 enum mlx4_port_type new_types[MLX4_MAX_PORTS];
0a984556 1037 static DEFINE_MUTEX(set_port_type_mutex);
7ff93f8b
YP		 1038 int i;
1039 int err = 0;
1040
0a984556
AV		 1041 mutex_lock(&set_port_type_mutex);
1042
7ff93f8b
YP		 1043 if (!strcmp(buf, "ib\n"))
1044 info->tmp_type = MLX4_PORT_TYPE_IB;
1045 else if (!strcmp(buf, "eth\n"))
1046 info->tmp_type = MLX4_PORT_TYPE_ETH;
27bf91d6
YP		 1047 else if (!strcmp(buf, "auto\n"))
1048 info->tmp_type = MLX4_PORT_TYPE_AUTO;
7ff93f8b
YP		 1049 else {
1050 mlx4_err(mdev, "%s is not supported port type\n", buf);
0a984556
AV		 1051 err = -EINVAL;
1052 goto err_out;
7ff93f8b
YP		 1053 }
1054
27bf91d6 1055 mlx4_stop_sense(mdev);
7ff93f8b 1056 mutex_lock(&priv->port_mutex);
27bf91d6
YP		 1057 /* Possible type is always the one that was delivered */
1058 mdev->caps.possible_type[info->port] = info->tmp_type;
1059
1060 for (i = 0; i < mdev->caps.num_ports; i++) {
7ff93f8b 1061 types[i] = priv->port[i+1].tmp_type ? priv->port[i+1].tmp_type :
27bf91d6
YP		 1062 mdev->caps.possible_type[i+1];
1063 if (types[i] == MLX4_PORT_TYPE_AUTO)
1064 types[i] = mdev->caps.port_type[i+1];
1065 }
7ff93f8b 1066
58a60168
YP		 1067 if (!(mdev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
1068 !(mdev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT)) {
27bf91d6
YP		 1069 for (i = 1; i <= mdev->caps.num_ports; i++) {
1070 if (mdev->caps.possible_type[i] == MLX4_PORT_TYPE_AUTO) {
1071 mdev->caps.possible_type[i] = mdev->caps.port_type[i];
1072 err = -EINVAL;
1073 }
1074 }
1075 }
1076 if (err) {
1a91de28 1077 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		 1078 goto out;
1079 }
1080
1081 mlx4_do_sense_ports(mdev, new_types, types);
1082
1083 err = mlx4_check_port_params(mdev, new_types);
7ff93f8b
YP		 1084 if (err)
1085 goto out;
1086
27bf91d6
YP		 1087 /* We are about to apply the changes after the configuration
1088 * was verified, no need to remember the temporary types
1089 * any more */
1090 for (i = 0; i < mdev->caps.num_ports; i++)
1091 priv->port[i + 1].tmp_type = 0;
7ff93f8b 1092
27bf91d6 1093 err = mlx4_change_port_types(mdev, new_types);
7ff93f8b
YP		 1094
1095out:
27bf91d6 1096 mlx4_start_sense(mdev);
7ff93f8b 1097 mutex_unlock(&priv->port_mutex);
0a984556
AV		 1098err_out:
1099 mutex_unlock(&set_port_type_mutex);
1100
7ff93f8b
YP		 1101 return err ? err : count;
1102}
1103
096335b3
OG		 1104enum ibta_mtu {
1105 IB_MTU_256 = 1,
1106 IB_MTU_512 = 2,
1107 IB_MTU_1024 = 3,
1108 IB_MTU_2048 = 4,
1109 IB_MTU_4096 = 5
1110};
1111
1112static inline int int_to_ibta_mtu(int mtu)
1113{
1114 switch (mtu) {
1115 case 256: return IB_MTU_256;
1116 case 512: return IB_MTU_512;
1117 case 1024: return IB_MTU_1024;
1118 case 2048: return IB_MTU_2048;
1119 case 4096: return IB_MTU_4096;
1120 default: return -1;
1121 }
1122}
1123
1124static inline int ibta_mtu_to_int(enum ibta_mtu mtu)
1125{
1126 switch (mtu) {
1127 case IB_MTU_256: return 256;
1128 case IB_MTU_512: return 512;
1129 case IB_MTU_1024: return 1024;
1130 case IB_MTU_2048: return 2048;
1131 case IB_MTU_4096: return 4096;
1132 default: return -1;
1133 }
1134}
1135
1136static ssize_t show_port_ib_mtu(struct device *dev,
1137 struct device_attribute *attr,
1138 char *buf)
1139{
1140 struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1141 port_mtu_attr);
1142 struct mlx4_dev *mdev = info->dev;
1143
1144 if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH)
1145 mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
1146
1147 sprintf(buf, "%d\n",
1148 ibta_mtu_to_int(mdev->caps.port_ib_mtu[info->port]));
1149 return strlen(buf);
1150}
1151
1152static ssize_t set_port_ib_mtu(struct device *dev,
1153 struct device_attribute *attr,
1154 const char *buf, size_t count)
1155{
1156 struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1157 port_mtu_attr);
1158 struct mlx4_dev *mdev = info->dev;
1159 struct mlx4_priv *priv = mlx4_priv(mdev);
1160 int err, port, mtu, ibta_mtu = -1;
1161
1162 if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH) {
1163 mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
1164 return -EINVAL;
1165 }
1166
618fad95
DB		 1167 err = kstrtoint(buf, 0, &mtu);
1168 if (!err)
096335b3
OG		 1169 ibta_mtu = int_to_ibta_mtu(mtu);
1170
618fad95 1171 if (err || ibta_mtu < 0) {
096335b3
OG		 1172 mlx4_err(mdev, "%s is invalid IBTA mtu\n", buf);
1173 return -EINVAL;
1174 }
1175
1176 mdev->caps.port_ib_mtu[info->port] = ibta_mtu;
1177
1178 mlx4_stop_sense(mdev);
1179 mutex_lock(&priv->port_mutex);
1180 mlx4_unregister_device(mdev);
1181 for (port = 1; port <= mdev->caps.num_ports; port++) {
1182 mlx4_CLOSE_PORT(mdev, port);
6634961c 1183 err = mlx4_SET_PORT(mdev, port, -1);
096335b3 1184 if (err) {
1a91de28
JP		 1185 mlx4_err(mdev, "Failed to set port %d, aborting\n",
1186 port);
096335b3
OG		 1187 goto err_set_port;
1188 }
1189 }
1190 err = mlx4_register_device(mdev);
1191err_set_port:
1192 mutex_unlock(&priv->port_mutex);
1193 mlx4_start_sense(mdev);
1194 return err ? err : count;
1195}
1196
53f33ae2
MS		 1197int mlx4_bond(struct mlx4_dev *dev)
1198{
1199 int ret = 0;
1200 struct mlx4_priv *priv = mlx4_priv(dev);
1201
1202 mutex_lock(&priv->bond_mutex);
1203
1204 if (!mlx4_is_bonded(dev))
1205 ret = mlx4_do_bond(dev, true);
1206 else
1207 ret = 0;
1208
1209 mutex_unlock(&priv->bond_mutex);
1210 if (ret)
1211 mlx4_err(dev, "Failed to bond device: %d\n", ret);
1212 else
1213 mlx4_dbg(dev, "Device is bonded\n");
1214 return ret;
1215}
1216EXPORT_SYMBOL_GPL(mlx4_bond);
1217
1218int mlx4_unbond(struct mlx4_dev *dev)
1219{
1220 int ret = 0;
1221 struct mlx4_priv *priv = mlx4_priv(dev);
1222
1223 mutex_lock(&priv->bond_mutex);
1224
1225 if (mlx4_is_bonded(dev))
1226 ret = mlx4_do_bond(dev, false);
1227
1228 mutex_unlock(&priv->bond_mutex);
1229 if (ret)
1230 mlx4_err(dev, "Failed to unbond device: %d\n", ret);
1231 else
1232 mlx4_dbg(dev, "Device is unbonded\n");
1233 return ret;
1234}
1235EXPORT_SYMBOL_GPL(mlx4_unbond);
1236
1237
1238int mlx4_port_map_set(struct mlx4_dev *dev, struct mlx4_port_map *v2p)
1239{
1240 u8 port1 = v2p->port1;
1241 u8 port2 = v2p->port2;
1242 struct mlx4_priv *priv = mlx4_priv(dev);
1243 int err;
1244
1245 if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_PORT_REMAP))
1246 return -ENOTSUPP;
1247
1248 mutex_lock(&priv->bond_mutex);
1249
1250 /* zero means keep current mapping for this port */
1251 if (port1 == 0)
1252 port1 = priv->v2p.port1;
1253 if (port2 == 0)
1254 port2 = priv->v2p.port2;
1255
1256 if ((port1 < 1) || (port1 > MLX4_MAX_PORTS) ||
1257 (port2 < 1) || (port2 > MLX4_MAX_PORTS) ||
1258 (port1 == 2 && port2 == 1)) {
1259 /* besides boundary checks cross mapping makes
1260 * no sense and therefore not allowed */
1261 err = -EINVAL;
1262 } else if ((port1 == priv->v2p.port1) &&
1263 (port2 == priv->v2p.port2)) {
1264 err = 0;
1265 } else {
1266 err = mlx4_virt2phy_port_map(dev, port1, port2);
1267 if (!err) {
1268 mlx4_dbg(dev, "port map changed: [%d][%d]\n",
1269 port1, port2);
1270 priv->v2p.port1 = port1;
1271 priv->v2p.port2 = port2;
1272 } else {
1273 mlx4_err(dev, "Failed to change port mape: %d\n", err);
1274 }
1275 }
1276
1277 mutex_unlock(&priv->bond_mutex);
1278 return err;
1279}
1280EXPORT_SYMBOL_GPL(mlx4_port_map_set);
1281
e8f9b2ed 1282static int mlx4_load_fw(struct mlx4_dev *dev)
225c7b1f
RD		 1283{
1284 struct mlx4_priv *priv = mlx4_priv(dev);
1285 int err;
1286
1287 priv->fw.fw_icm = mlx4_alloc_icm(dev, priv->fw.fw_pages,
5b0bf5e2 1288 GFP_HIGHUSER | __GFP_NOWARN, 0);
225c7b1f 1289 if (!priv->fw.fw_icm) {
1a91de28 1290 mlx4_err(dev, "Couldn't allocate FW area, aborting\n");
225c7b1f
RD		 1291 return -ENOMEM;
1292 }
1293
1294 err = mlx4_MAP_FA(dev, priv->fw.fw_icm);
1295 if (err) {
1a91de28 1296 mlx4_err(dev, "MAP_FA command failed, aborting\n");
225c7b1f
RD		 1297 goto err_free;
1298 }
1299
1300 err = mlx4_RUN_FW(dev);
1301 if (err) {
1a91de28 1302 mlx4_err(dev, "RUN_FW command failed, aborting\n");
225c7b1f
RD		 1303 goto err_unmap_fa;
1304 }
1305
1306 return 0;
1307
1308err_unmap_fa:
1309 mlx4_UNMAP_FA(dev);
1310
1311err_free:
5b0bf5e2 1312 mlx4_free_icm(dev, priv->fw.fw_icm, 0);
225c7b1f
RD		 1313 return err;
1314}
1315
e8f9b2ed
RD		 1316static int mlx4_init_cmpt_table(struct mlx4_dev *dev, u64 cmpt_base,
1317 int cmpt_entry_sz)
225c7b1f
RD		 1318{
1319 struct mlx4_priv *priv = mlx4_priv(dev);
1320 int err;
ab9c17a0 1321 int num_eqs;
225c7b1f
RD		 1322
1323 err = mlx4_init_icm_table(dev, &priv->qp_table.cmpt_table,
1324 cmpt_base +
1325 ((u64) (MLX4_CMPT_TYPE_QP *
1326 cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1327 cmpt_entry_sz, dev->caps.num_qps,
93fc9e1b
YP		 1328 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1329 0, 0);
225c7b1f
RD		 1330 if (err)
1331 goto err;
1332
1333 err = mlx4_init_icm_table(dev, &priv->srq_table.cmpt_table,
1334 cmpt_base +
1335 ((u64) (MLX4_CMPT_TYPE_SRQ *
1336 cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1337 cmpt_entry_sz, dev->caps.num_srqs,
5b0bf5e2 1338 dev->caps.reserved_srqs, 0, 0);
225c7b1f
RD		 1339 if (err)
1340 goto err_qp;
1341
1342 err = mlx4_init_icm_table(dev, &priv->cq_table.cmpt_table,
1343 cmpt_base +
1344 ((u64) (MLX4_CMPT_TYPE_CQ *
1345 cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1346 cmpt_entry_sz, dev->caps.num_cqs,
5b0bf5e2 1347 dev->caps.reserved_cqs, 0, 0);
225c7b1f
RD		 1348 if (err)
1349 goto err_srq;
1350
7ae0e400 1351 num_eqs = dev->phys_caps.num_phys_eqs;
225c7b1f
RD		 1352 err = mlx4_init_icm_table(dev, &priv->eq_table.cmpt_table,
1353 cmpt_base +
1354 ((u64) (MLX4_CMPT_TYPE_EQ *
1355 cmpt_entry_sz) << MLX4_CMPT_SHIFT),
ab9c17a0 1356 cmpt_entry_sz, num_eqs, num_eqs, 0, 0);
225c7b1f
RD		 1357 if (err)
1358 goto err_cq;
1359
1360 return 0;
1361
1362err_cq:
1363 mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1364
1365err_srq:
1366 mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1367
1368err_qp:
1369 mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1370
1371err:
1372 return err;
1373}
1374
3d73c288
RD		 1375static int mlx4_init_icm(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap,
1376 struct mlx4_init_hca_param *init_hca, u64 icm_size)
225c7b1f
RD		 1377{
1378 struct mlx4_priv *priv = mlx4_priv(dev);
1379 u64 aux_pages;
ab9c17a0 1380 int num_eqs;
225c7b1f
RD		 1381 int err;
1382
1383 err = mlx4_SET_ICM_SIZE(dev, icm_size, &aux_pages);
1384 if (err) {
1a91de28 1385 mlx4_err(dev, "SET_ICM_SIZE command failed, aborting\n");
225c7b1f
RD		 1386 return err;
1387 }
1388
1a91de28 1389 mlx4_dbg(dev, "%lld KB of HCA context requires %lld KB aux memory\n",
225c7b1f
RD		 1390 (unsigned long long) icm_size >> 10,
1391 (unsigned long long) aux_pages << 2);
1392
1393 priv->fw.aux_icm = mlx4_alloc_icm(dev, aux_pages,
5b0bf5e2 1394 GFP_HIGHUSER | __GFP_NOWARN, 0);
225c7b1f 1395 if (!priv->fw.aux_icm) {
1a91de28 1396 mlx4_err(dev, "Couldn't allocate aux memory, aborting\n");
225c7b1f
RD		 1397 return -ENOMEM;
1398 }
1399
1400 err = mlx4_MAP_ICM_AUX(dev, priv->fw.aux_icm);
1401 if (err) {
1a91de28 1402 mlx4_err(dev, "MAP_ICM_AUX command failed, aborting\n");
225c7b1f
RD		 1403 goto err_free_aux;
1404 }
1405
1406 err = mlx4_init_cmpt_table(dev, init_hca->cmpt_base, dev_cap->cmpt_entry_sz);
1407 if (err) {
1a91de28 1408 mlx4_err(dev, "Failed to map cMPT context memory, aborting\n");
225c7b1f
RD		 1409 goto err_unmap_aux;
1410 }
1411
ab9c17a0 1412
7ae0e400 1413 num_eqs = dev->phys_caps.num_phys_eqs;
fa0681d2
RD		 1414 err = mlx4_init_icm_table(dev, &priv->eq_table.table,
1415 init_hca->eqc_base, dev_cap->eqc_entry_sz,
ab9c17a0 1416 num_eqs, num_eqs, 0, 0);
225c7b1f 1417 if (err) {
1a91de28 1418 mlx4_err(dev, "Failed to map EQ context memory, aborting\n");
225c7b1f
RD		 1419 goto err_unmap_cmpt;
1420 }
1421
d7bb58fb
JM		 1422 /*
1423 * Reserved MTT entries must be aligned up to a cacheline
1424 * boundary, since the FW will write to them, while the driver
1425 * writes to all other MTT entries. (The variable
1426 * dev->caps.mtt_entry_sz below is really the MTT segment
1427 * size, not the raw entry size)
1428 */
1429 dev->caps.reserved_mtts =
1430 ALIGN(dev->caps.reserved_mtts * dev->caps.mtt_entry_sz,
1431 dma_get_cache_alignment()) / dev->caps.mtt_entry_sz;
1432
225c7b1f
RD		 1433 err = mlx4_init_icm_table(dev, &priv->mr_table.mtt_table,
1434 init_hca->mtt_base,
1435 dev->caps.mtt_entry_sz,
2b8fb286 1436 dev->caps.num_mtts,
5b0bf5e2 1437 dev->caps.reserved_mtts, 1, 0);
225c7b1f 1438 if (err) {
1a91de28 1439 mlx4_err(dev, "Failed to map MTT context memory, aborting\n");
225c7b1f
RD		 1440 goto err_unmap_eq;
1441 }
1442
1443 err = mlx4_init_icm_table(dev, &priv->mr_table.dmpt_table,
1444 init_hca->dmpt_base,
1445 dev_cap->dmpt_entry_sz,
1446 dev->caps.num_mpts,
5b0bf5e2 1447 dev->caps.reserved_mrws, 1, 1);
225c7b1f 1448 if (err) {
1a91de28 1449 mlx4_err(dev, "Failed to map dMPT context memory, aborting\n");
225c7b1f
RD		 1450 goto err_unmap_mtt;
1451 }
1452
1453 err = mlx4_init_icm_table(dev, &priv->qp_table.qp_table,
1454 init_hca->qpc_base,
1455 dev_cap->qpc_entry_sz,
1456 dev->caps.num_qps,
93fc9e1b
YP		 1457 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1458 0, 0);
225c7b1f 1459 if (err) {
1a91de28 1460 mlx4_err(dev, "Failed to map QP context memory, aborting\n");
225c7b1f
RD		 1461 goto err_unmap_dmpt;
1462 }
1463
1464 err = mlx4_init_icm_table(dev, &priv->qp_table.auxc_table,
1465 init_hca->auxc_base,
1466 dev_cap->aux_entry_sz,
1467 dev->caps.num_qps,
93fc9e1b
YP		 1468 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1469 0, 0);
225c7b1f 1470 if (err) {
1a91de28 1471 mlx4_err(dev, "Failed to map AUXC context memory, aborting\n");
225c7b1f
RD		 1472 goto err_unmap_qp;
1473 }
1474
1475 err = mlx4_init_icm_table(dev, &priv->qp_table.altc_table,
1476 init_hca->altc_base,
1477 dev_cap->altc_entry_sz,
1478 dev->caps.num_qps,
93fc9e1b
YP		 1479 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1480 0, 0);
225c7b1f 1481 if (err) {
1a91de28 1482 mlx4_err(dev, "Failed to map ALTC context memory, aborting\n");
225c7b1f
RD		 1483 goto err_unmap_auxc;
1484 }
1485
1486 err = mlx4_init_icm_table(dev, &priv->qp_table.rdmarc_table,
1487 init_hca->rdmarc_base,
1488 dev_cap->rdmarc_entry_sz << priv->qp_table.rdmarc_shift,
1489 dev->caps.num_qps,
93fc9e1b
YP		 1490 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1491 0, 0);
225c7b1f
RD		 1492 if (err) {
1493 mlx4_err(dev, "Failed to map RDMARC context memory, aborting\n");
1494 goto err_unmap_altc;
1495 }
1496
1497 err = mlx4_init_icm_table(dev, &priv->cq_table.table,
1498 init_hca->cqc_base,
1499 dev_cap->cqc_entry_sz,
1500 dev->caps.num_cqs,
5b0bf5e2 1501 dev->caps.reserved_cqs, 0, 0);
225c7b1f 1502 if (err) {
1a91de28 1503 mlx4_err(dev, "Failed to map CQ context memory, aborting\n");
225c7b1f
RD		 1504 goto err_unmap_rdmarc;
1505 }
1506
1507 err = mlx4_init_icm_table(dev, &priv->srq_table.table,
1508 init_hca->srqc_base,
1509 dev_cap->srq_entry_sz,
1510 dev->caps.num_srqs,
5b0bf5e2 1511 dev->caps.reserved_srqs, 0, 0);
225c7b1f 1512 if (err) {
1a91de28 1513 mlx4_err(dev, "Failed to map SRQ context memory, aborting\n");
225c7b1f
RD		 1514 goto err_unmap_cq;
1515 }
1516
1517 /*
0ff1fb65
HHZ		 1518 * For flow steering device managed mode it is required to use
1519 * mlx4_init_icm_table. For B0 steering mode it's not strictly
1520 * required, but for simplicity just map the whole multicast
1521 * group table now. The table isn't very big and it's a lot
1522 * easier than trying to track ref counts.
225c7b1f
RD		 1523 */
1524 err = mlx4_init_icm_table(dev, &priv->mcg_table.table,
0ec2c0f8
EE		 1525 init_hca->mc_base,
1526 mlx4_get_mgm_entry_size(dev),
225c7b1f
RD		 1527 dev->caps.num_mgms + dev->caps.num_amgms,
1528 dev->caps.num_mgms + dev->caps.num_amgms,
5b0bf5e2 1529 0, 0);
225c7b1f 1530 if (err) {
1a91de28 1531 mlx4_err(dev, "Failed to map MCG context memory, aborting\n");
225c7b1f
RD		 1532 goto err_unmap_srq;
1533 }
1534
1535 return 0;
1536
1537err_unmap_srq:
1538 mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1539
1540err_unmap_cq:
1541 mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1542
1543err_unmap_rdmarc:
1544 mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1545
1546err_unmap_altc:
1547 mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1548
1549err_unmap_auxc:
1550 mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1551
1552err_unmap_qp:
1553 mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1554
1555err_unmap_dmpt:
1556 mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1557
1558err_unmap_mtt:
1559 mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
1560
1561err_unmap_eq:
fa0681d2 1562 mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
225c7b1f
RD		 1563
1564err_unmap_cmpt:
1565 mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1566 mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1567 mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1568 mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1569
1570err_unmap_aux:
1571 mlx4_UNMAP_ICM_AUX(dev);
1572
1573err_free_aux:
5b0bf5e2 1574 mlx4_free_icm(dev, priv->fw.aux_icm, 0);
225c7b1f
RD		 1575
1576 return err;
1577}
1578
1579static void mlx4_free_icms(struct mlx4_dev *dev)
1580{
1581 struct mlx4_priv *priv = mlx4_priv(dev);
1582
1583 mlx4_cleanup_icm_table(dev, &priv->mcg_table.table);
1584 mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1585 mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1586 mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1587 mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1588 mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1589 mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1590 mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1591 mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
fa0681d2 1592 mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
225c7b1f
RD		 1593 mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1594 mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1595 mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1596 mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
225c7b1f
RD		 1597
1598 mlx4_UNMAP_ICM_AUX(dev);
5b0bf5e2 1599 mlx4_free_icm(dev, priv->fw.aux_icm, 0);
225c7b1f
RD		 1600}
1601
ab9c17a0
JM		 1602static void mlx4_slave_exit(struct mlx4_dev *dev)
1603{
1604 struct mlx4_priv *priv = mlx4_priv(dev);
1605
f3d4c89e 1606 mutex_lock(&priv->cmd.slave_cmd_mutex);
0cd93027
YH		 1607 if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, MLX4_COMM_CMD_NA_OP,
1608 MLX4_COMM_TIME))
1a91de28 1609 mlx4_warn(dev, "Failed to close slave function\n");
f3d4c89e 1610 mutex_unlock(&priv->cmd.slave_cmd_mutex);
ab9c17a0
JM		 1611}
1612
c1b43dca
EC		 1613static int map_bf_area(struct mlx4_dev *dev)
1614{
1615 struct mlx4_priv *priv = mlx4_priv(dev);
1616 resource_size_t bf_start;
1617 resource_size_t bf_len;
1618 int err = 0;
1619
3d747473
JM		 1620 if (!dev->caps.bf_reg_size)
1621 return -ENXIO;
1622
872bf2fb 1623 bf_start = pci_resource_start(dev->persist->pdev, 2) +
ab9c17a0 1624 (dev->caps.num_uars << PAGE_SHIFT);
872bf2fb 1625 bf_len = pci_resource_len(dev->persist->pdev, 2) -
ab9c17a0 1626 (dev->caps.num_uars << PAGE_SHIFT);
c1b43dca
EC		 1627 priv->bf_mapping = io_mapping_create_wc(bf_start, bf_len);
1628 if (!priv->bf_mapping)
1629 err = -ENOMEM;
1630
1631 return err;
1632}
1633
1634static void unmap_bf_area(struct mlx4_dev *dev)
1635{
1636 if (mlx4_priv(dev)->bf_mapping)
1637 io_mapping_free(mlx4_priv(dev)->bf_mapping);
1638}
1639
ec693d47
AV		 1640cycle_t mlx4_read_clock(struct mlx4_dev *dev)
1641{
1642 u32 clockhi, clocklo, clockhi1;
1643 cycle_t cycles;
1644 int i;
1645 struct mlx4_priv *priv = mlx4_priv(dev);
1646
1647 for (i = 0; i < 10; i++) {
1648 clockhi = swab32(readl(priv->clock_mapping));
1649 clocklo = swab32(readl(priv->clock_mapping + 4));
1650 clockhi1 = swab32(readl(priv->clock_mapping));
1651 if (clockhi == clockhi1)
1652 break;
1653 }
1654
1655 cycles = (u64) clockhi << 32 | (u64) clocklo;
1656
1657 return cycles;
1658}
1659EXPORT_SYMBOL_GPL(mlx4_read_clock);
1660
1661
ddd8a6c1
EE		 1662static int map_internal_clock(struct mlx4_dev *dev)
1663{
1664 struct mlx4_priv *priv = mlx4_priv(dev);
1665
1666 priv->clock_mapping =
872bf2fb
YH		 1667 ioremap(pci_resource_start(dev->persist->pdev,
1668 priv->fw.clock_bar) +
ddd8a6c1
EE		 1669 priv->fw.clock_offset, MLX4_CLOCK_SIZE);
1670
1671 if (!priv->clock_mapping)
1672 return -ENOMEM;
1673
1674 return 0;
1675}
1676
1677static void unmap_internal_clock(struct mlx4_dev *dev)
1678{
1679 struct mlx4_priv *priv = mlx4_priv(dev);
1680
1681 if (priv->clock_mapping)
1682 iounmap(priv->clock_mapping);
1683}
1684
225c7b1f
RD		 1685static void mlx4_close_hca(struct mlx4_dev *dev)
1686{
ddd8a6c1 1687 unmap_internal_clock(dev);
c1b43dca 1688 unmap_bf_area(dev);
ab9c17a0
JM		 1689 if (mlx4_is_slave(dev))
1690 mlx4_slave_exit(dev);
1691 else {
1692 mlx4_CLOSE_HCA(dev, 0);
1693 mlx4_free_icms(dev);
a0eacca9
MB		 1694 }
1695}
1696
1697static void mlx4_close_fw(struct mlx4_dev *dev)
1698{
1699 if (!mlx4_is_slave(dev)) {
ab9c17a0
JM		 1700 mlx4_UNMAP_FA(dev);
1701 mlx4_free_icm(dev, mlx4_priv(dev)->fw.fw_icm, 0);
1702 }
1703}
1704
55ad3592
YH		 1705static int mlx4_comm_check_offline(struct mlx4_dev *dev)
1706{
1707#define COMM_CHAN_OFFLINE_OFFSET 0x09
1708
1709 u32 comm_flags;
1710 u32 offline_bit;
1711 unsigned long end;
1712 struct mlx4_priv *priv = mlx4_priv(dev);
1713
1714 end = msecs_to_jiffies(MLX4_COMM_OFFLINE_TIME_OUT) + jiffies;
1715 while (time_before(jiffies, end)) {
1716 comm_flags = swab32(readl((__iomem char *)priv->mfunc.comm +
1717 MLX4_COMM_CHAN_FLAGS));
1718 offline_bit = (comm_flags &
1719 (u32)(1 << COMM_CHAN_OFFLINE_OFFSET));
1720 if (!offline_bit)
1721 return 0;
1722 /* There are cases as part of AER/Reset flow that PF needs
1723 * around 100 msec to load. We therefore sleep for 100 msec
1724 * to allow other tasks to make use of that CPU during this
1725 * time interval.
1726 */
1727 msleep(100);
1728 }
1729 mlx4_err(dev, "Communication channel is offline.\n");
1730 return -EIO;
1731}
1732
1733static void mlx4_reset_vf_support(struct mlx4_dev *dev)
1734{
1735#define COMM_CHAN_RST_OFFSET 0x1e
1736
1737 struct mlx4_priv *priv = mlx4_priv(dev);
1738 u32 comm_rst;
1739 u32 comm_caps;
1740
1741 comm_caps = swab32(readl((__iomem char *)priv->mfunc.comm +
1742 MLX4_COMM_CHAN_CAPS));
1743 comm_rst = (comm_caps & (u32)(1 << COMM_CHAN_RST_OFFSET));
1744
1745 if (comm_rst)
1746 dev->caps.vf_caps |= MLX4_VF_CAP_FLAG_RESET;
1747}
1748
ab9c17a0
JM		 1749static int mlx4_init_slave(struct mlx4_dev *dev)
1750{
1751 struct mlx4_priv *priv = mlx4_priv(dev);
1752 u64 dma = (u64) priv->mfunc.vhcr_dma;
ab9c17a0
JM		 1753 int ret_from_reset = 0;
1754 u32 slave_read;
1755 u32 cmd_channel_ver;
1756
97989356 1757 if (atomic_read(&pf_loading)) {
1a91de28 1758 mlx4_warn(dev, "PF is not ready - Deferring probe\n");
97989356
AV		 1759 return -EPROBE_DEFER;
1760 }
1761
f3d4c89e 1762 mutex_lock(&priv->cmd.slave_cmd_mutex);
ab9c17a0 1763 priv->cmd.max_cmds = 1;
55ad3592
YH		 1764 if (mlx4_comm_check_offline(dev)) {
1765 mlx4_err(dev, "PF is not responsive, skipping initialization\n");
1766 goto err_offline;
1767 }
1768
1769 mlx4_reset_vf_support(dev);
ab9c17a0
JM		 1770 mlx4_warn(dev, "Sending reset\n");
1771 ret_from_reset = mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0,
0cd93027 1772 MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME);
ab9c17a0
JM		 1773 /* if we are in the middle of flr the slave will try
1774 * NUM_OF_RESET_RETRIES times before leaving.*/
1775 if (ret_from_reset) {
1776 if (MLX4_DELAY_RESET_SLAVE == ret_from_reset) {
1a91de28 1777 mlx4_warn(dev, "slave is currently in the middle of FLR - Deferring probe\n");
5efe5355
JM		 1778 mutex_unlock(&priv->cmd.slave_cmd_mutex);
1779 return -EPROBE_DEFER;
ab9c17a0
JM		 1780 } else
1781 goto err;
1782 }
1783
1784 /* check the driver version - the slave I/F revision
1785 * must match the master's */
1786 slave_read = swab32(readl(&priv->mfunc.comm->slave_read));
1787 cmd_channel_ver = mlx4_comm_get_version();
1788
1789 if (MLX4_COMM_GET_IF_REV(cmd_channel_ver) !=
1790 MLX4_COMM_GET_IF_REV(slave_read)) {
1a91de28 1791 mlx4_err(dev, "slave driver version is not supported by the master\n");
ab9c17a0
JM		 1792 goto err;
1793 }
1794
1795 mlx4_warn(dev, "Sending vhcr0\n");
1796 if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR0, dma >> 48,
0cd93027 1797 MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
ab9c17a0
JM		 1798 goto err;
1799 if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR1, dma >> 32,
0cd93027 1800 MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
ab9c17a0
JM		 1801 goto err;
1802 if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR2, dma >> 16,
0cd93027 1803 MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
ab9c17a0 1804 goto err;
0cd93027
YH		 1805 if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR_EN, dma,
1806 MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
ab9c17a0 1807 goto err;
f3d4c89e
RD		 1808
1809 mutex_unlock(&priv->cmd.slave_cmd_mutex);
ab9c17a0
JM		 1810 return 0;
1811
1812err:
0cd93027 1813 mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, MLX4_COMM_CMD_NA_OP, 0);
55ad3592 1814err_offline:
f3d4c89e 1815 mutex_unlock(&priv->cmd.slave_cmd_mutex);
ab9c17a0 1816 return -EIO;
225c7b1f
RD		 1817}
1818
6634961c
JM		 1819static void mlx4_parav_master_pf_caps(struct mlx4_dev *dev)
1820{
1821 int i;
1822
1823 for (i = 1; i <= dev->caps.num_ports; i++) {
b6ffaeff
JM		 1824 if (dev->caps.port_type[i] == MLX4_PORT_TYPE_ETH)
1825 dev->caps.gid_table_len[i] =
449fc488 1826 mlx4_get_slave_num_gids(dev, 0, i);
b6ffaeff
JM		 1827 else
1828 dev->caps.gid_table_len[i] = 1;
6634961c
JM		 1829 dev->caps.pkey_table_len[i] =
1830 dev->phys_caps.pkey_phys_table_len[i] - 1;
1831 }
1832}
1833
3c439b55
JM		 1834static int choose_log_fs_mgm_entry_size(int qp_per_entry)
1835{
1836 int i = MLX4_MIN_MGM_LOG_ENTRY_SIZE;
1837
1838 for (i = MLX4_MIN_MGM_LOG_ENTRY_SIZE; i <= MLX4_MAX_MGM_LOG_ENTRY_SIZE;
1839 i++) {
1840 if (qp_per_entry <= 4 * ((1 << i) / 16 - 2))
1841 break;
1842 }
1843
1844 return (i <= MLX4_MAX_MGM_LOG_ENTRY_SIZE) ? i : -1;
1845}
1846
7d077cd3
MB		 1847static const char *dmfs_high_rate_steering_mode_str(int dmfs_high_steer_mode)
1848{
1849 switch (dmfs_high_steer_mode) {
1850 case MLX4_STEERING_DMFS_A0_DEFAULT:
1851 return "default performance";
1852
1853 case MLX4_STEERING_DMFS_A0_DYNAMIC:
1854 return "dynamic hybrid mode";
1855
1856 case MLX4_STEERING_DMFS_A0_STATIC:
1857 return "performance optimized for limited rule configuration (static)";
1858
1859 case MLX4_STEERING_DMFS_A0_DISABLE:
1860 return "disabled performance optimized steering";
1861
1862 case MLX4_STEERING_DMFS_A0_NOT_SUPPORTED:
1863 return "performance optimized steering not supported";
1864
1865 default:
1866 return "Unrecognized mode";
1867 }
1868}
1869
1870#define MLX4_DMFS_A0_STEERING (1UL << 2)
1871
7b8157be
JM		 1872static void choose_steering_mode(struct mlx4_dev *dev,
1873 struct mlx4_dev_cap *dev_cap)
1874{
7d077cd3
MB		 1875 if (mlx4_log_num_mgm_entry_size <= 0) {
1876 if ((-mlx4_log_num_mgm_entry_size) & MLX4_DMFS_A0_STEERING) {
1877 if (dev->caps.dmfs_high_steer_mode ==
1878 MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
1879 mlx4_err(dev, "DMFS high rate mode not supported\n");
1880 else
1881 dev->caps.dmfs_high_steer_mode =
1882 MLX4_STEERING_DMFS_A0_STATIC;
1883 }
1884 }
1885
1886 if (mlx4_log_num_mgm_entry_size <= 0 &&
3c439b55 1887 dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_FS_EN &&
7b8157be 1888 (!mlx4_is_mfunc(dev) ||
872bf2fb
YH		 1889 (dev_cap->fs_max_num_qp_per_entry >=
1890 (dev->persist->num_vfs + 1))) &&
3c439b55
JM		 1891 choose_log_fs_mgm_entry_size(dev_cap->fs_max_num_qp_per_entry) >=
1892 MLX4_MIN_MGM_LOG_ENTRY_SIZE) {
1893 dev->oper_log_mgm_entry_size =
1894 choose_log_fs_mgm_entry_size(dev_cap->fs_max_num_qp_per_entry);
7b8157be
JM		 1895 dev->caps.steering_mode = MLX4_STEERING_MODE_DEVICE_MANAGED;
1896 dev->caps.num_qp_per_mgm = dev_cap->fs_max_num_qp_per_entry;
1897 dev->caps.fs_log_max_ucast_qp_range_size =
1898 dev_cap->fs_log_max_ucast_qp_range_size;
1899 } else {
7d077cd3
MB		 1900 if (dev->caps.dmfs_high_steer_mode !=
1901 MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
1902 dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_DISABLE;
7b8157be
JM		 1903 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER &&
1904 dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER)
1905 dev->caps.steering_mode = MLX4_STEERING_MODE_B0;
1906 else {
1907 dev->caps.steering_mode = MLX4_STEERING_MODE_A0;
1908
1909 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER ||
1910 dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER)
1a91de28 1911 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 1912 }
3c439b55
JM		 1913 dev->oper_log_mgm_entry_size =
1914 mlx4_log_num_mgm_entry_size > 0 ?
1915 mlx4_log_num_mgm_entry_size :
1916 MLX4_DEFAULT_MGM_LOG_ENTRY_SIZE;
7b8157be
JM		 1917 dev->caps.num_qp_per_mgm = mlx4_get_qp_per_mgm(dev);
1918 }
1a91de28 1919 mlx4_dbg(dev, "Steering mode is: %s, oper_log_mgm_entry_size = %d, modparam log_num_mgm_entry_size = %d\n",
3c439b55
JM		 1920 mlx4_steering_mode_str(dev->caps.steering_mode),
1921 dev->oper_log_mgm_entry_size,
1922 mlx4_log_num_mgm_entry_size);
7b8157be
JM		 1923}
1924
7ffdf726
OG		 1925static void choose_tunnel_offload_mode(struct mlx4_dev *dev,
1926 struct mlx4_dev_cap *dev_cap)
1927{
1928 if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED &&
5eff6dad 1929 dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_VXLAN_OFFLOADS)
7ffdf726
OG		 1930 dev->caps.tunnel_offload_mode = MLX4_TUNNEL_OFFLOAD_MODE_VXLAN;
1931 else
1932 dev->caps.tunnel_offload_mode = MLX4_TUNNEL_OFFLOAD_MODE_NONE;
1933
1934 mlx4_dbg(dev, "Tunneling offload mode is: %s\n", (dev->caps.tunnel_offload_mode
1935 == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) ? "vxlan" : "none");
1936}
1937
7d077cd3
MB		 1938static int mlx4_validate_optimized_steering(struct mlx4_dev *dev)
1939{
1940 int i;
1941 struct mlx4_port_cap port_cap;
1942
1943 if (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
1944 return -EINVAL;
1945
1946 for (i = 1; i <= dev->caps.num_ports; i++) {
1947 if (mlx4_dev_port(dev, i, &port_cap)) {
1948 mlx4_err(dev,
1949 "QUERY_DEV_CAP command failed, can't veify DMFS high rate steering.\n");
1950 } else if ((dev->caps.dmfs_high_steer_mode !=
1951 MLX4_STEERING_DMFS_A0_DEFAULT) &&
1952 (port_cap.dmfs_optimized_state ==
1953 !!(dev->caps.dmfs_high_steer_mode ==
1954 MLX4_STEERING_DMFS_A0_DISABLE))) {
1955 mlx4_err(dev,
1956 "DMFS high rate steer mode differ, driver requested %s but %s in FW.\n",
1957 dmfs_high_rate_steering_mode_str(
1958 dev->caps.dmfs_high_steer_mode),
1959 (port_cap.dmfs_optimized_state ?
1960 "enabled" : "disabled"));
1961 }
1962 }
1963
1964 return 0;
1965}
1966
a0eacca9 1967static int mlx4_init_fw(struct mlx4_dev *dev)
225c7b1f 1968{
2d928651 1969 struct mlx4_mod_stat_cfg mlx4_cfg;
a0eacca9 1970 int err = 0;
225c7b1f 1971
ab9c17a0
JM		 1972 if (!mlx4_is_slave(dev)) {
1973 err = mlx4_QUERY_FW(dev);
1974 if (err) {
1975 if (err == -EACCES)
1a91de28 1976 mlx4_info(dev, "non-primary physical function, skipping\n");
ab9c17a0 1977 else
1a91de28 1978 mlx4_err(dev, "QUERY_FW command failed, aborting\n");
bef772eb 1979 return err;
ab9c17a0 1980 }
225c7b1f 1981
ab9c17a0
JM		 1982 err = mlx4_load_fw(dev);
1983 if (err) {
1a91de28 1984 mlx4_err(dev, "Failed to start FW, aborting\n");
bef772eb 1985 return err;
ab9c17a0 1986 }
225c7b1f 1987
ab9c17a0
JM		 1988 mlx4_cfg.log_pg_sz_m = 1;
1989 mlx4_cfg.log_pg_sz = 0;
1990 err = mlx4_MOD_STAT_CFG(dev, &mlx4_cfg);
1991 if (err)
1992 mlx4_warn(dev, "Failed to override log_pg_sz parameter\n");
a0eacca9 1993 }
2d928651 1994
a0eacca9
MB		 1995 return err;
1996}
1997
1998static int mlx4_init_hca(struct mlx4_dev *dev)
1999{
2000 struct mlx4_priv *priv = mlx4_priv(dev);
2001 struct mlx4_adapter adapter;
2002 struct mlx4_dev_cap dev_cap;
2003 struct mlx4_profile profile;
2004 struct mlx4_init_hca_param init_hca;
2005 u64 icm_size;
2006 struct mlx4_config_dev_params params;
2007 int err;
2008
2009 if (!mlx4_is_slave(dev)) {
ab9c17a0
JM		 2010 err = mlx4_dev_cap(dev, &dev_cap);
2011 if (err) {
1a91de28 2012 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
d0d01250 2013 return err;
ab9c17a0 2014 }
225c7b1f 2015
7b8157be 2016 choose_steering_mode(dev, &dev_cap);
7ffdf726 2017 choose_tunnel_offload_mode(dev, &dev_cap);
7b8157be 2018
7d077cd3
MB		 2019 if (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC &&
2020 mlx4_is_master(dev))
2021 dev->caps.function_caps |= MLX4_FUNC_CAP_DMFS_A0_STATIC;
2022
8e1a28e8
HHZ		 2023 err = mlx4_get_phys_port_id(dev);
2024 if (err)
2025 mlx4_err(dev, "Fail to get physical port id\n");
2026
6634961c
JM		 2027 if (mlx4_is_master(dev))
2028 mlx4_parav_master_pf_caps(dev);
2029
2599d858
AV		 2030 if (mlx4_low_memory_profile()) {
2031 mlx4_info(dev, "Running from within kdump kernel. Using low memory profile\n");
2032 profile = low_mem_profile;
2033 } else {
2034 profile = default_profile;
2035 }
0ff1fb65
HHZ		 2036 if (dev->caps.steering_mode ==
2037 MLX4_STEERING_MODE_DEVICE_MANAGED)
2038 profile.num_mcg = MLX4_FS_NUM_MCG;
225c7b1f 2039
ab9c17a0
JM		 2040 icm_size = mlx4_make_profile(dev, &profile, &dev_cap,
2041 &init_hca);
2042 if ((long long) icm_size < 0) {
2043 err = icm_size;
d0d01250 2044 return err;
ab9c17a0 2045 }
225c7b1f 2046
a5bbe892
EC		 2047 dev->caps.max_fmr_maps = (1 << (32 - ilog2(dev->caps.num_mpts))) - 1;
2048
ab9c17a0
JM		 2049 init_hca.log_uar_sz = ilog2(dev->caps.num_uars);
2050 init_hca.uar_page_sz = PAGE_SHIFT - 12;
e448834e
SM		 2051 init_hca.mw_enabled = 0;
2052 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2053 dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN)
2054 init_hca.mw_enabled = INIT_HCA_TPT_MW_ENABLE;
c1b43dca 2055
ab9c17a0
JM		 2056 err = mlx4_init_icm(dev, &dev_cap, &init_hca, icm_size);
2057 if (err)
d0d01250 2058 return err;
225c7b1f 2059
ab9c17a0
JM		 2060 err = mlx4_INIT_HCA(dev, &init_hca);
2061 if (err) {
1a91de28 2062 mlx4_err(dev, "INIT_HCA command failed, aborting\n");
ab9c17a0
JM		 2063 goto err_free_icm;
2064 }
7ae0e400
MB		 2065
2066 if (dev_cap.flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS) {
2067 err = mlx4_query_func(dev, &dev_cap);
2068 if (err < 0) {
2069 mlx4_err(dev, "QUERY_FUNC command failed, aborting.\n");
d0d01250 2070 goto err_close;
7ae0e400
MB		 2071 } else if (err & MLX4_QUERY_FUNC_NUM_SYS_EQS) {
2072 dev->caps.num_eqs = dev_cap.max_eqs;
2073 dev->caps.reserved_eqs = dev_cap.reserved_eqs;
2074 dev->caps.reserved_uars = dev_cap.reserved_uars;
2075 }
2076 }
2077
ddd8a6c1
EE		 2078 /*
2079 * If TS is supported by FW
2080 * read HCA frequency by QUERY_HCA command
2081 */
2082 if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS) {
2083 memset(&init_hca, 0, sizeof(init_hca));
2084 err = mlx4_QUERY_HCA(dev, &init_hca);
2085 if (err) {
1a91de28 2086 mlx4_err(dev, "QUERY_HCA command failed, disable timestamp\n");
ddd8a6c1
EE		 2087 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
2088 } else {
2089 dev->caps.hca_core_clock =
2090 init_hca.hca_core_clock;
2091 }
2092
2093 /* In case we got HCA frequency 0 - disable timestamping
2094 * to avoid dividing by zero
2095 */
2096 if (!dev->caps.hca_core_clock) {
2097 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
2098 mlx4_err(dev,
1a91de28 2099 "HCA frequency is 0 - timestamping is not supported\n");
ddd8a6c1
EE		 2100 } else if (map_internal_clock(dev)) {
2101 /*
2102 * Map internal clock,
2103 * in case of failure disable timestamping
2104 */
2105 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
1a91de28 2106 mlx4_err(dev, "Failed to map internal clock. Timestamping is not supported\n");
ddd8a6c1
EE		 2107 }
2108 }
7d077cd3
MB		 2109
2110 if (dev->caps.dmfs_high_steer_mode !=
2111 MLX4_STEERING_DMFS_A0_NOT_SUPPORTED) {
2112 if (mlx4_validate_optimized_steering(dev))
2113 mlx4_warn(dev, "Optimized steering validation failed\n");
2114
2115 if (dev->caps.dmfs_high_steer_mode ==
2116 MLX4_STEERING_DMFS_A0_DISABLE) {
2117 dev->caps.dmfs_high_rate_qpn_base =
2118 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
2119 dev->caps.dmfs_high_rate_qpn_range =
2120 MLX4_A0_STEERING_TABLE_SIZE;
2121 }
2122
2123 mlx4_dbg(dev, "DMFS high rate steer mode is: %s\n",
2124 dmfs_high_rate_steering_mode_str(
2125 dev->caps.dmfs_high_steer_mode));
2126 }
ab9c17a0
JM		 2127 } else {
2128 err = mlx4_init_slave(dev);
2129 if (err) {
5efe5355
JM		 2130 if (err != -EPROBE_DEFER)
2131 mlx4_err(dev, "Failed to initialize slave\n");
bef772eb 2132 return err;
ab9c17a0 2133 }
225c7b1f 2134
ab9c17a0
JM		 2135 err = mlx4_slave_cap(dev);
2136 if (err) {
2137 mlx4_err(dev, "Failed to obtain slave caps\n");
2138 goto err_close;
2139 }
225c7b1f
RD		 2140 }
2141
ab9c17a0
JM		 2142 if (map_bf_area(dev))
2143 mlx4_dbg(dev, "Failed to map blue flame area\n");
2144
2145 /*Only the master set the ports, all the rest got it from it.*/
2146 if (!mlx4_is_slave(dev))
2147 mlx4_set_port_mask(dev);
2148
225c7b1f
RD		 2149 err = mlx4_QUERY_ADAPTER(dev, &adapter);
2150 if (err) {
1a91de28 2151 mlx4_err(dev, "QUERY_ADAPTER command failed, aborting\n");
bef772eb 2152 goto unmap_bf;
225c7b1f
RD		 2153 }
2154
f8c6455b
SM		 2155 /* Query CONFIG_DEV parameters */
2156 err = mlx4_config_dev_retrieval(dev, &params);
2157 if (err && err != -ENOTSUPP) {
2158 mlx4_err(dev, "Failed to query CONFIG_DEV parameters\n");
2159 } else if (!err) {
2160 dev->caps.rx_checksum_flags_port[1] = params.rx_csum_flags_port_1;
2161 dev->caps.rx_checksum_flags_port[2] = params.rx_csum_flags_port_2;
2162 }
225c7b1f 2163 priv->eq_table.inta_pin = adapter.inta_pin;
cd9281d8 2164 memcpy(dev->board_id, adapter.board_id, sizeof dev->board_id);
225c7b1f
RD		 2165
2166 return 0;
2167
bef772eb 2168unmap_bf:
ddd8a6c1 2169 unmap_internal_clock(dev);
bef772eb
AY		 2170 unmap_bf_area(dev);
2171
b38f2879 2172 if (mlx4_is_slave(dev)) {
99ec41d0 2173 kfree(dev->caps.qp0_qkey);
b38f2879
DB		 2174 kfree(dev->caps.qp0_tunnel);
2175 kfree(dev->caps.qp0_proxy);
2176 kfree(dev->caps.qp1_tunnel);
2177 kfree(dev->caps.qp1_proxy);
2178 }
2179
225c7b1f 2180err_close:
41929ed2
DB		 2181 if (mlx4_is_slave(dev))
2182 mlx4_slave_exit(dev);
2183 else
2184 mlx4_CLOSE_HCA(dev, 0);
225c7b1f
RD		 2185
2186err_free_icm:
ab9c17a0
JM		 2187 if (!mlx4_is_slave(dev))
2188 mlx4_free_icms(dev);
225c7b1f 2189
225c7b1f
RD		 2190 return err;
2191}
2192
f2a3f6a3
OG		 2193static int mlx4_init_counters_table(struct mlx4_dev *dev)
2194{
2195 struct mlx4_priv *priv = mlx4_priv(dev);
47d8417f 2196 int nent_pow2;
f2a3f6a3
OG		 2197
2198 if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2199 return -ENOENT;
2200
47d8417f
EBE		 2201 nent_pow2 = roundup_pow_of_two(dev->caps.max_counters);
2202 /* reserve last counter index for sink counter */
2203 return mlx4_bitmap_init(&priv->counters_bitmap, nent_pow2,
2204 nent_pow2 - 1, 0,
2205 nent_pow2 - dev->caps.max_counters + 1);
f2a3f6a3
OG		 2206}
2207
2208static void mlx4_cleanup_counters_table(struct mlx4_dev *dev)
2209{
efa6bc91
EBE		 2210 if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2211 return;
2212
f2a3f6a3
OG		 2213 mlx4_bitmap_cleanup(&mlx4_priv(dev)->counters_bitmap);
2214}
2215
ba062d52 2216int __mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx)
f2a3f6a3
OG		 2217{
2218 struct mlx4_priv *priv = mlx4_priv(dev);
2219
2220 if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2221 return -ENOENT;
2222
2223 *idx = mlx4_bitmap_alloc(&priv->counters_bitmap);
2224 if (*idx == -1)
2225 return -ENOMEM;
2226
2227 return 0;
2228}
ba062d52
JM		 2229
2230int mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx)
2231{
2232 u64 out_param;
2233 int err;
2234
2235 if (mlx4_is_mfunc(dev)) {
2236 err = mlx4_cmd_imm(dev, 0, &out_param, RES_COUNTER,
2237 RES_OP_RESERVE, MLX4_CMD_ALLOC_RES,
2238 MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
2239 if (!err)
2240 *idx = get_param_l(&out_param);
2241
2242 return err;
2243 }
2244 return __mlx4_counter_alloc(dev, idx);
2245}
f2a3f6a3
OG		 2246EXPORT_SYMBOL_GPL(mlx4_counter_alloc);
2247
b72ca7e9
EBE		 2248static int __mlx4_clear_if_stat(struct mlx4_dev *dev,
2249 u8 counter_index)
2250{
2251 struct mlx4_cmd_mailbox *if_stat_mailbox;
2252 int err;
2253 u32 if_stat_in_mod = (counter_index & 0xff) | MLX4_QUERY_IF_STAT_RESET;
2254
2255 if_stat_mailbox = mlx4_alloc_cmd_mailbox(dev);
2256 if (IS_ERR(if_stat_mailbox))
2257 return PTR_ERR(if_stat_mailbox);
2258
2259 err = mlx4_cmd_box(dev, 0, if_stat_mailbox->dma, if_stat_in_mod, 0,
2260 MLX4_CMD_QUERY_IF_STAT, MLX4_CMD_TIME_CLASS_C,
2261 MLX4_CMD_NATIVE);
2262
2263 mlx4_free_cmd_mailbox(dev, if_stat_mailbox);
2264 return err;
2265}
2266
ba062d52 2267void __mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
f2a3f6a3 2268{
efa6bc91
EBE		 2269 if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2270 return;
2271
b72ca7e9
EBE		 2272 __mlx4_clear_if_stat(dev, idx);
2273
7c6d74d2 2274 mlx4_bitmap_free(&mlx4_priv(dev)->counters_bitmap, idx, MLX4_USE_RR);
f2a3f6a3
OG		 2275 return;
2276}
ba062d52
JM		 2277
2278void mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
2279{
e7dbeba8 2280 u64 in_param = 0;
ba062d52
JM		 2281
2282 if (mlx4_is_mfunc(dev)) {
2283 set_param_l(&in_param, idx);
2284 mlx4_cmd(dev, in_param, RES_COUNTER, RES_OP_RESERVE,
2285 MLX4_CMD_FREE_RES, MLX4_CMD_TIME_CLASS_A,
2286 MLX4_CMD_WRAPPED);
2287 return;
2288 }
2289 __mlx4_counter_free(dev, idx);
2290}
f2a3f6a3
OG		 2291EXPORT_SYMBOL_GPL(mlx4_counter_free);
2292
773af94e
YH		 2293void mlx4_set_admin_guid(struct mlx4_dev *dev, __be64 guid, int entry, int port)
2294{
2295 struct mlx4_priv *priv = mlx4_priv(dev);
2296
2297 priv->mfunc.master.vf_admin[entry].vport[port].guid = guid;
2298}
2299EXPORT_SYMBOL_GPL(mlx4_set_admin_guid);
2300
2301__be64 mlx4_get_admin_guid(struct mlx4_dev *dev, int entry, int port)
2302{
2303 struct mlx4_priv *priv = mlx4_priv(dev);
2304
2305 return priv->mfunc.master.vf_admin[entry].vport[port].guid;
2306}
2307EXPORT_SYMBOL_GPL(mlx4_get_admin_guid);
2308
fb517a4f
YH		 2309void mlx4_set_random_admin_guid(struct mlx4_dev *dev, int entry, int port)
2310{
2311 struct mlx4_priv *priv = mlx4_priv(dev);
2312 __be64 guid;
2313
2314 /* hw GUID */
2315 if (entry == 0)
2316 return;
2317
2318 get_random_bytes((char *)&guid, sizeof(guid));
2319 guid &= ~(cpu_to_be64(1ULL << 56));
2320 guid |= cpu_to_be64(1ULL << 57);
2321 priv->mfunc.master.vf_admin[entry].vport[port].guid = guid;
2322}
2323
3d73c288 2324static int mlx4_setup_hca(struct mlx4_dev *dev)
225c7b1f
RD		 2325{
2326 struct mlx4_priv *priv = mlx4_priv(dev);
2327 int err;
7ff93f8b 2328 int port;
9a5aa622 2329 __be32 ib_port_default_caps;
225c7b1f 2330
225c7b1f
RD		 2331 err = mlx4_init_uar_table(dev);
2332 if (err) {
1a91de28
JP		 2333 mlx4_err(dev, "Failed to initialize user access region table, aborting\n");
2334 return err;
225c7b1f
RD		 2335 }
2336
2337 err = mlx4_uar_alloc(dev, &priv->driver_uar);
2338 if (err) {
1a91de28 2339 mlx4_err(dev, "Failed to allocate driver access region, aborting\n");
225c7b1f
RD		 2340 goto err_uar_table_free;
2341 }
2342
4979d18f 2343 priv->kar = ioremap((phys_addr_t) priv->driver_uar.pfn << PAGE_SHIFT, PAGE_SIZE);
225c7b1f 2344 if (!priv->kar) {
1a91de28 2345 mlx4_err(dev, "Couldn't map kernel access region, aborting\n");
225c7b1f
RD		 2346 err = -ENOMEM;
2347 goto err_uar_free;
2348 }
2349
2350 err = mlx4_init_pd_table(dev);
2351 if (err) {
1a91de28 2352 mlx4_err(dev, "Failed to initialize protection domain table, aborting\n");
225c7b1f
RD		 2353 goto err_kar_unmap;
2354 }
2355
012a8ff5
SH		 2356 err = mlx4_init_xrcd_table(dev);
2357 if (err) {
1a91de28 2358 mlx4_err(dev, "Failed to initialize reliable connection domain table, aborting\n");
012a8ff5
SH		 2359 goto err_pd_table_free;
2360 }
2361
225c7b1f
RD		 2362 err = mlx4_init_mr_table(dev);
2363 if (err) {
1a91de28 2364 mlx4_err(dev, "Failed to initialize memory region table, aborting\n");
012a8ff5 2365 goto err_xrcd_table_free;
225c7b1f
RD		 2366 }
2367
fe6f700d
YP		 2368 if (!mlx4_is_slave(dev)) {
2369 err = mlx4_init_mcg_table(dev);
2370 if (err) {
1a91de28 2371 mlx4_err(dev, "Failed to initialize multicast group table, aborting\n");
fe6f700d
YP		 2372 goto err_mr_table_free;
2373 }
114840c3
JM		 2374 err = mlx4_config_mad_demux(dev);
2375 if (err) {
2376 mlx4_err(dev, "Failed in config_mad_demux, aborting\n");
2377 goto err_mcg_table_free;
2378 }
fe6f700d
YP		 2379 }
2380
225c7b1f
RD		 2381 err = mlx4_init_eq_table(dev);
2382 if (err) {
1a91de28 2383 mlx4_err(dev, "Failed to initialize event queue table, aborting\n");
fe6f700d 2384 goto err_mcg_table_free;
225c7b1f
RD		 2385 }
2386
2387 err = mlx4_cmd_use_events(dev);
2388 if (err) {
1a91de28 2389 mlx4_err(dev, "Failed to switch to event-driven firmware commands, aborting\n");
225c7b1f
RD		 2390 goto err_eq_table_free;
2391 }
2392
2393 err = mlx4_NOP(dev);
2394 if (err) {
08fb1055 2395 if (dev->flags & MLX4_FLAG_MSI_X) {
1a91de28 2396 mlx4_warn(dev, "NOP command failed to generate MSI-X interrupt IRQ %d)\n",
c66fa19c 2397 priv->eq_table.eq[MLX4_EQ_ASYNC].irq);
1a91de28 2398 mlx4_warn(dev, "Trying again without MSI-X\n");
08fb1055 2399 } else {
1a91de28 2400 mlx4_err(dev, "NOP command failed to generate interrupt (IRQ %d), aborting\n",
c66fa19c 2401 priv->eq_table.eq[MLX4_EQ_ASYNC].irq);
225c7b1f 2402 mlx4_err(dev, "BIOS or ACPI interrupt routing problem?\n");
08fb1055 2403 }
225c7b1f
RD		 2404
2405 goto err_cmd_poll;
2406 }
2407
2408 mlx4_dbg(dev, "NOP command IRQ test passed\n");
2409
2410 err = mlx4_init_cq_table(dev);
2411 if (err) {
1a91de28 2412 mlx4_err(dev, "Failed to initialize completion queue table, aborting\n");
225c7b1f
RD		 2413 goto err_cmd_poll;
2414 }
2415
2416 err = mlx4_init_srq_table(dev);
2417 if (err) {
1a91de28 2418 mlx4_err(dev, "Failed to initialize shared receive queue table, aborting\n");
225c7b1f
RD		 2419 goto err_cq_table_free;
2420 }
2421
2422 err = mlx4_init_qp_table(dev);
2423 if (err) {
1a91de28 2424 mlx4_err(dev, "Failed to initialize queue pair table, aborting\n");
225c7b1f
RD		 2425 goto err_srq_table_free;
2426 }
2427
f2a3f6a3
OG		 2428 err = mlx4_init_counters_table(dev);
2429 if (err && err != -ENOENT) {
1a91de28 2430 mlx4_err(dev, "Failed to initialize counters table, aborting\n");
fe6f700d 2431 goto err_qp_table_free;
f2a3f6a3
OG		 2432 }
2433
ab9c17a0
JM		 2434 if (!mlx4_is_slave(dev)) {
2435 for (port = 1; port <= dev->caps.num_ports; port++) {
ab9c17a0
JM		 2436 ib_port_default_caps = 0;
2437 err = mlx4_get_port_ib_caps(dev, port,
2438 &ib_port_default_caps);
2439 if (err)
1a91de28
JP		 2440 mlx4_warn(dev, "failed to get port %d default ib capabilities (%d). Continuing with caps = 0\n",
2441 port, err);
ab9c17a0
JM		 2442 dev->caps.ib_port_def_cap[port] = ib_port_default_caps;
2443
2aca1172
JM		 2444 /* initialize per-slave default ib port capabilities */
2445 if (mlx4_is_master(dev)) {
2446 int i;
2447 for (i = 0; i < dev->num_slaves; i++) {
2448 if (i == mlx4_master_func_num(dev))
2449 continue;
2450 priv->mfunc.master.slave_state[i].ib_cap_mask[port] =
1a91de28 2451 ib_port_default_caps;
2aca1172
JM		 2452 }
2453 }
2454
096335b3
OG		 2455 if (mlx4_is_mfunc(dev))
2456 dev->caps.port_ib_mtu[port] = IB_MTU_2048;
2457 else
2458 dev->caps.port_ib_mtu[port] = IB_MTU_4096;
97285b78 2459
6634961c
JM		 2460 err = mlx4_SET_PORT(dev, port, mlx4_is_master(dev) ?
2461 dev->caps.pkey_table_len[port] : -1);
ab9c17a0
JM		 2462 if (err) {
2463 mlx4_err(dev, "Failed to set port %d, aborting\n",
1a91de28 2464 port);
ab9c17a0
JM		 2465 goto err_counters_table_free;
2466 }
7ff93f8b
YP		 2467 }
2468 }
2469
225c7b1f
RD		 2470 return 0;
2471
f2a3f6a3
OG		 2472err_counters_table_free:
2473 mlx4_cleanup_counters_table(dev);
2474
225c7b1f
RD		 2475err_qp_table_free:
2476 mlx4_cleanup_qp_table(dev);
2477
2478err_srq_table_free:
2479 mlx4_cleanup_srq_table(dev);
2480
2481err_cq_table_free:
2482 mlx4_cleanup_cq_table(dev);
2483
2484err_cmd_poll:
2485 mlx4_cmd_use_polling(dev);
2486
2487err_eq_table_free:
2488 mlx4_cleanup_eq_table(dev);
2489
fe6f700d
YP		 2490err_mcg_table_free:
2491 if (!mlx4_is_slave(dev))
2492 mlx4_cleanup_mcg_table(dev);
2493
ee49bd93 2494err_mr_table_free:
225c7b1f
RD		 2495 mlx4_cleanup_mr_table(dev);
2496
012a8ff5
SH		 2497err_xrcd_table_free:
2498 mlx4_cleanup_xrcd_table(dev);
2499
225c7b1f
RD		 2500err_pd_table_free:
2501 mlx4_cleanup_pd_table(dev);
2502
2503err_kar_unmap:
2504 iounmap(priv->kar);
2505
2506err_uar_free:
2507 mlx4_uar_free(dev, &priv->driver_uar);
2508
2509err_uar_table_free:
2510 mlx4_cleanup_uar_table(dev);
2511 return err;
2512}
2513
de161803
IS		 2514static int mlx4_init_affinity_hint(struct mlx4_dev *dev, int port, int eqn)
2515{
2516 int requested_cpu = 0;
2517 struct mlx4_priv *priv = mlx4_priv(dev);
2518 struct mlx4_eq *eq;
2519 int off = 0;
2520 int i;
2521
2522 if (eqn > dev->caps.num_comp_vectors)
2523 return -EINVAL;
2524
2525 for (i = 1; i < port; i++)
2526 off += mlx4_get_eqs_per_port(dev, i);
2527
2528 requested_cpu = eqn - off - !!(eqn > MLX4_EQ_ASYNC);
2529
2530 /* Meaning EQs are shared, and this call comes from the second port */
2531 if (requested_cpu < 0)
2532 return 0;
2533
2534 eq = &priv->eq_table.eq[eqn];
2535
2536 if (!zalloc_cpumask_var(&eq->affinity_mask, GFP_KERNEL))
2537 return -ENOMEM;
2538
2539 cpumask_set_cpu(requested_cpu, eq->affinity_mask);
2540
2541 return 0;
2542}
2543
e8f9b2ed 2544static void mlx4_enable_msi_x(struct mlx4_dev *dev)
225c7b1f
RD		 2545{
2546 struct mlx4_priv *priv = mlx4_priv(dev);
b8dd786f 2547 struct msix_entry *entries;
225c7b1f 2548 int i;
c66fa19c 2549 int port = 0;
225c7b1f
RD		 2550
2551 if (msi_x) {
c66fa19c 2552 int nreq = dev->caps.num_ports * num_online_cpus() + 1;
7ae0e400 2553
ca4c7b35
OG		 2554 nreq = min_t(int, dev->caps.num_eqs - dev->caps.reserved_eqs,
2555 nreq);
ab9c17a0 2556
b8dd786f
YP		 2557 entries = kcalloc(nreq, sizeof *entries, GFP_KERNEL);
2558 if (!entries)
2559 goto no_msi;
2560
2561 for (i = 0; i < nreq; ++i)
225c7b1f
RD		 2562 entries[i].entry = i;
2563
872bf2fb
YH		 2564 nreq = pci_enable_msix_range(dev->persist->pdev, entries, 2,
2565 nreq);
66e2f9c1 2566
c66fa19c 2567 if (nreq < 0 || nreq < MLX4_EQ_ASYNC) {
5bf0da7d 2568 kfree(entries);
225c7b1f 2569 goto no_msi;
0b7ca5a9 2570 }
c66fa19c
MB		 2571 /* 1 is reserved for events (asyncrounous EQ) */
2572 dev->caps.num_comp_vectors = nreq - 1;
2573
2574 priv->eq_table.eq[MLX4_EQ_ASYNC].irq = entries[0].vector;
2575 bitmap_zero(priv->eq_table.eq[MLX4_EQ_ASYNC].actv_ports.ports,
2576 dev->caps.num_ports);
2577
2578 for (i = 0; i < dev->caps.num_comp_vectors + 1; i++) {
2579 if (i == MLX4_EQ_ASYNC)
2580 continue;
2581
2582 priv->eq_table.eq[i].irq =
2583 entries[i + 1 - !!(i > MLX4_EQ_ASYNC)].vector;
2584
2585 if (MLX4_IS_LEGACY_EQ_MODE(dev->caps)) {
2586 bitmap_fill(priv->eq_table.eq[i].actv_ports.ports,
2587 dev->caps.num_ports);
de161803
IS		 2588 /* We don't set affinity hint when there
2589 * aren't enough EQs
2590 */
c66fa19c
MB		 2591 } else {
2592 set_bit(port,
2593 priv->eq_table.eq[i].actv_ports.ports);
de161803
IS		 2594 if (mlx4_init_affinity_hint(dev, port + 1, i))
2595 mlx4_warn(dev, "Couldn't init hint cpumask for EQ %d\n",
2596 i);
c66fa19c
MB		 2597 }
2598 /* We divide the Eqs evenly between the two ports.
2599 * (dev->caps.num_comp_vectors / dev->caps.num_ports)
2600 * refers to the number of Eqs per port
2601 * (i.e eqs_per_port). Theoretically, we would like to
2602 * write something like (i + 1) % eqs_per_port == 0.
2603 * However, since there's an asynchronous Eq, we have
2604 * to skip over it by comparing this condition to
2605 * !!((i + 1) > MLX4_EQ_ASYNC).
2606 */
2607 if ((dev->caps.num_comp_vectors > dev->caps.num_ports) &&
2608 ((i + 1) %
2609 (dev->caps.num_comp_vectors / dev->caps.num_ports)) ==
2610 !!((i + 1) > MLX4_EQ_ASYNC))
2611 /* If dev->caps.num_comp_vectors < dev->caps.num_ports,
2612 * everything is shared anyway.
2613 */
2614 port++;
2615 }
225c7b1f
RD		 2616
2617 dev->flags |= MLX4_FLAG_MSI_X;
b8dd786f
YP		 2618
2619 kfree(entries);
225c7b1f
RD		 2620 return;
2621 }
2622
2623no_msi:
b8dd786f
YP		 2624 dev->caps.num_comp_vectors = 1;
2625
c66fa19c
MB		 2626 BUG_ON(MLX4_EQ_ASYNC >= 2);
2627 for (i = 0; i < 2; ++i) {
872bf2fb 2628 priv->eq_table.eq[i].irq = dev->persist->pdev->irq;
c66fa19c
MB		 2629 if (i != MLX4_EQ_ASYNC) {
2630 bitmap_fill(priv->eq_table.eq[i].actv_ports.ports,
2631 dev->caps.num_ports);
2632 }
2633 }
225c7b1f
RD		 2634}
2635
7ff93f8b 2636static int mlx4_init_port_info(struct mlx4_dev *dev, int port)
2a2336f8
YP		 2637{
2638 struct mlx4_port_info *info = &mlx4_priv(dev)->port[port];
7ff93f8b 2639 int err = 0;
2a2336f8
YP		 2640
2641 info->dev = dev;
2642 info->port = port;
ab9c17a0 2643 if (!mlx4_is_slave(dev)) {
ab9c17a0
JM		 2644 mlx4_init_mac_table(dev, &info->mac_table);
2645 mlx4_init_vlan_table(dev, &info->vlan_table);
111c6094 2646 mlx4_init_roce_gid_table(dev, &info->gid_table);
16a10ffd 2647 info->base_qpn = mlx4_get_base_qpn(dev, port);
ab9c17a0 2648 }
7ff93f8b
YP		 2649
2650 sprintf(info->dev_name, "mlx4_port%d", port);
2651 info->port_attr.attr.name = info->dev_name;
ab9c17a0
JM		 2652 if (mlx4_is_mfunc(dev))
2653 info->port_attr.attr.mode = S_IRUGO;
2654 else {
2655 info->port_attr.attr.mode = S_IRUGO | S_IWUSR;
2656 info->port_attr.store = set_port_type;
2657 }
7ff93f8b 2658 info->port_attr.show = show_port_type;
3691c964 2659 sysfs_attr_init(&info->port_attr.attr);
7ff93f8b 2660
872bf2fb 2661 err = device_create_file(&dev->persist->pdev->dev, &info->port_attr);
7ff93f8b
YP		 2662 if (err) {
2663 mlx4_err(dev, "Failed to create file for port %d\n", port);
2664 info->port = -1;
2665 }
2666
096335b3
OG		 2667 sprintf(info->dev_mtu_name, "mlx4_port%d_mtu", port);
2668 info->port_mtu_attr.attr.name = info->dev_mtu_name;
2669 if (mlx4_is_mfunc(dev))
2670 info->port_mtu_attr.attr.mode = S_IRUGO;
2671 else {
2672 info->port_mtu_attr.attr.mode = S_IRUGO | S_IWUSR;
2673 info->port_mtu_attr.store = set_port_ib_mtu;
2674 }
2675 info->port_mtu_attr.show = show_port_ib_mtu;
2676 sysfs_attr_init(&info->port_mtu_attr.attr);
2677
872bf2fb
YH		 2678 err = device_create_file(&dev->persist->pdev->dev,
2679 &info->port_mtu_attr);
096335b3
OG		 2680 if (err) {
2681 mlx4_err(dev, "Failed to create mtu file for port %d\n", port);
872bf2fb
YH		 2682 device_remove_file(&info->dev->persist->pdev->dev,
2683 &info->port_attr);
096335b3
OG		 2684 info->port = -1;
2685 }
2686
7ff93f8b
YP		 2687 return err;
2688}
2689
2690static void mlx4_cleanup_port_info(struct mlx4_port_info *info)
2691{
2692 if (info->port < 0)
2693 return;
2694
872bf2fb
YH		 2695 device_remove_file(&info->dev->persist->pdev->dev, &info->port_attr);
2696 device_remove_file(&info->dev->persist->pdev->dev,
2697 &info->port_mtu_attr);
c66fa19c
MB		 2698#ifdef CONFIG_RFS_ACCEL
2699 free_irq_cpu_rmap(info->rmap);
2700 info->rmap = NULL;
2701#endif
2a2336f8
YP		 2702}
2703
b12d93d6
YP		 2704static int mlx4_init_steering(struct mlx4_dev *dev)
2705{
2706 struct mlx4_priv *priv = mlx4_priv(dev);
2707 int num_entries = dev->caps.num_ports;
2708 int i, j;
2709
2710 priv->steer = kzalloc(sizeof(struct mlx4_steer) * num_entries, GFP_KERNEL);
2711 if (!priv->steer)
2712 return -ENOMEM;
2713
45b51365 2714 for (i = 0; i < num_entries; i++)
b12d93d6
YP		 2715 for (j = 0; j < MLX4_NUM_STEERS; j++) {
2716 INIT_LIST_HEAD(&priv->steer[i].promisc_qps[j]);
2717 INIT_LIST_HEAD(&priv->steer[i].steer_entries[j]);
2718 }
b12d93d6
YP		 2719 return 0;
2720}
2721
2722static void mlx4_clear_steering(struct mlx4_dev *dev)
2723{
2724 struct mlx4_priv *priv = mlx4_priv(dev);
2725 struct mlx4_steer_index *entry, *tmp_entry;
2726 struct mlx4_promisc_qp *pqp, *tmp_pqp;
2727 int num_entries = dev->caps.num_ports;
2728 int i, j;
2729
2730 for (i = 0; i < num_entries; i++) {
2731 for (j = 0; j < MLX4_NUM_STEERS; j++) {
2732 list_for_each_entry_safe(pqp, tmp_pqp,
2733 &priv->steer[i].promisc_qps[j],
2734 list) {
2735 list_del(&pqp->list);
2736 kfree(pqp);
2737 }
2738 list_for_each_entry_safe(entry, tmp_entry,
2739 &priv->steer[i].steer_entries[j],
2740 list) {
2741 list_del(&entry->list);
2742 list_for_each_entry_safe(pqp, tmp_pqp,
2743 &entry->duplicates,
2744 list) {
2745 list_del(&pqp->list);
2746 kfree(pqp);
2747 }
2748 kfree(entry);
2749 }
2750 }
2751 }
2752 kfree(priv->steer);
2753}
2754
ab9c17a0
JM		 2755static int extended_func_num(struct pci_dev *pdev)
2756{
2757 return PCI_SLOT(pdev->devfn) * 8 + PCI_FUNC(pdev->devfn);
2758}
2759
2760#define MLX4_OWNER_BASE 0x8069c
2761#define MLX4_OWNER_SIZE 4
2762
2763static int mlx4_get_ownership(struct mlx4_dev *dev)
2764{
2765 void __iomem *owner;
2766 u32 ret;
2767
872bf2fb 2768 if (pci_channel_offline(dev->persist->pdev))
57dbf29a
KSS		 2769 return -EIO;
2770
872bf2fb
YH		 2771 owner = ioremap(pci_resource_start(dev->persist->pdev, 0) +
2772 MLX4_OWNER_BASE,
ab9c17a0
JM		 2773 MLX4_OWNER_SIZE);
2774 if (!owner) {
2775 mlx4_err(dev, "Failed to obtain ownership bit\n");
2776 return -ENOMEM;
2777 }
2778
2779 ret = readl(owner);
2780 iounmap(owner);
2781 return (int) !!ret;
2782}
2783
2784static void mlx4_free_ownership(struct mlx4_dev *dev)
2785{
2786 void __iomem *owner;
2787
872bf2fb 2788 if (pci_channel_offline(dev->persist->pdev))
57dbf29a
KSS		 2789 return;
2790
872bf2fb
YH		 2791 owner = ioremap(pci_resource_start(dev->persist->pdev, 0) +
2792 MLX4_OWNER_BASE,
ab9c17a0
JM		 2793 MLX4_OWNER_SIZE);
2794 if (!owner) {
2795 mlx4_err(dev, "Failed to obtain ownership bit\n");
2796 return;
2797 }
2798 writel(0, owner);
2799 msleep(1000);
2800 iounmap(owner);
2801}
2802
a0eacca9
MB		 2803#define SRIOV_VALID_STATE(flags) (!!((flags) & MLX4_FLAG_SRIOV) ==\
2804 !!((flags) & MLX4_FLAG_MASTER))
2805
2806static u64 mlx4_enable_sriov(struct mlx4_dev *dev, struct pci_dev *pdev,
55ad3592 2807 u8 total_vfs, int existing_vfs, int reset_flow)
a0eacca9
MB		 2808{
2809 u64 dev_flags = dev->flags;
da315679 2810 int err = 0;
a0eacca9 2811
55ad3592
YH		 2812 if (reset_flow) {
2813 dev->dev_vfs = kcalloc(total_vfs, sizeof(*dev->dev_vfs),
2814 GFP_KERNEL);
2815 if (!dev->dev_vfs)
2816 goto free_mem;
2817 return dev_flags;
2818 }
2819
da315679
MB		 2820 atomic_inc(&pf_loading);
2821 if (dev->flags & MLX4_FLAG_SRIOV) {
2822 if (existing_vfs != total_vfs) {
2823 mlx4_err(dev, "SR-IOV was already enabled, but with num_vfs (%d) different than requested (%d)\n",
2824 existing_vfs, total_vfs);
2825 total_vfs = existing_vfs;
2826 }
2827 }
2828
2829 dev->dev_vfs = kzalloc(total_vfs * sizeof(*dev->dev_vfs), GFP_KERNEL);
a0eacca9
MB		 2830 if (NULL == dev->dev_vfs) {
2831 mlx4_err(dev, "Failed to allocate memory for VFs\n");
2832 goto disable_sriov;
da315679
MB		 2833 }
2834
2835 if (!(dev->flags & MLX4_FLAG_SRIOV)) {
2836 mlx4_warn(dev, "Enabling SR-IOV with %d VFs\n", total_vfs);
2837 err = pci_enable_sriov(pdev, total_vfs);
2838 }
2839 if (err) {
2840 mlx4_err(dev, "Failed to enable SR-IOV, continuing without SR-IOV (err = %d)\n",
2841 err);
2842 goto disable_sriov;
2843 } else {
2844 mlx4_warn(dev, "Running in master mode\n");
2845 dev_flags |= MLX4_FLAG_SRIOV |
2846 MLX4_FLAG_MASTER;
2847 dev_flags &= ~MLX4_FLAG_SLAVE;
872bf2fb 2848 dev->persist->num_vfs = total_vfs;
a0eacca9
MB		 2849 }
2850 return dev_flags;
2851
2852disable_sriov:
da315679 2853 atomic_dec(&pf_loading);
55ad3592 2854free_mem:
872bf2fb 2855 dev->persist->num_vfs = 0;
a0eacca9 2856 kfree(dev->dev_vfs);
5114a04e 2857 dev->dev_vfs = NULL;
a0eacca9
MB		 2858 return dev_flags & ~MLX4_FLAG_MASTER;
2859}
2860
de966c59
MB		 2861enum {
2862 MLX4_DEV_CAP_CHECK_NUM_VFS_ABOVE_64 = -1,
2863};
2864
2865static int mlx4_check_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap,
2866 int *nvfs)
2867{
2868 int requested_vfs = nvfs[0] + nvfs[1] + nvfs[2];
2869 /* Checking for 64 VFs as a limitation of CX2 */
2870 if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_80_VFS) &&
2871 requested_vfs >= 64) {
2872 mlx4_err(dev, "Requested %d VFs, but FW does not support more than 64\n",
2873 requested_vfs);
2874 return MLX4_DEV_CAP_CHECK_NUM_VFS_ABOVE_64;
2875 }
2876 return 0;
2877}
2878
e1c00e10 2879static int mlx4_load_one(struct pci_dev *pdev, int pci_dev_data,
55ad3592
YH		 2880 int total_vfs, int *nvfs, struct mlx4_priv *priv,
2881 int reset_flow)
225c7b1f 2882{
225c7b1f 2883 struct mlx4_dev *dev;
e1c00e10 2884 unsigned sum = 0;
225c7b1f 2885 int err;
2a2336f8 2886 int port;
e1c00e10 2887 int i;
7ae0e400 2888 struct mlx4_dev_cap *dev_cap = NULL;
bbb07af4 2889 int existing_vfs = 0;
225c7b1f 2890
e1c00e10 2891 dev = &priv->dev;
225c7b1f 2892
b581401e
RD		 2893 INIT_LIST_HEAD(&priv->ctx_list);
2894 spin_lock_init(&priv->ctx_lock);
225c7b1f 2895
7ff93f8b 2896 mutex_init(&priv->port_mutex);
53f33ae2 2897 mutex_init(&priv->bond_mutex);
7ff93f8b 2898
6296883c
YP		 2899 INIT_LIST_HEAD(&priv->pgdir_list);
2900 mutex_init(&priv->pgdir_mutex);
2901
c1b43dca
EC		 2902 INIT_LIST_HEAD(&priv->bf_list);
2903 mutex_init(&priv->bf_mutex);
2904
aca7a3ac 2905 dev->rev_id = pdev->revision;
6e7136ed 2906 dev->numa_node = dev_to_node(&pdev->dev);
e1c00e10 2907
ab9c17a0 2908 /* Detect if this device is a virtual function */
839f1243 2909 if (pci_dev_data & MLX4_PCI_DEV_IS_VF) {
ab9c17a0
JM		 2910 mlx4_warn(dev, "Detected virtual function - running in slave mode\n");
2911 dev->flags |= MLX4_FLAG_SLAVE;
2912 } else {
2913 /* We reset the device and enable SRIOV only for physical
2914 * devices. Try to claim ownership on the device;
2915 * if already taken, skip -- do not allow multiple PFs */
2916 err = mlx4_get_ownership(dev);
2917 if (err) {
2918 if (err < 0)
e1c00e10 2919 return err;
ab9c17a0 2920 else {
1a91de28 2921 mlx4_warn(dev, "Multiple PFs not yet supported - Skipping PF\n");
e1c00e10 2922 return -EINVAL;
ab9c17a0
JM		 2923 }
2924 }
aca7a3ac 2925
fe6f700d
YP		 2926 atomic_set(&priv->opreq_count, 0);
2927 INIT_WORK(&priv->opreq_task, mlx4_opreq_action);
2928
ab9c17a0
JM		 2929 /*
2930 * Now reset the HCA before we touch the PCI capabilities or
2931 * attempt a firmware command, since a boot ROM may have left
2932 * the HCA in an undefined state.
2933 */
2934 err = mlx4_reset(dev);
2935 if (err) {
1a91de28 2936 mlx4_err(dev, "Failed to reset HCA, aborting\n");
e1c00e10 2937 goto err_sriov;
ab9c17a0 2938 }
7ae0e400
MB		 2939
2940 if (total_vfs) {
7ae0e400 2941 dev->flags = MLX4_FLAG_MASTER;
da315679
MB		 2942 existing_vfs = pci_num_vf(pdev);
2943 if (existing_vfs)
2944 dev->flags |= MLX4_FLAG_SRIOV;
872bf2fb 2945 dev->persist->num_vfs = total_vfs;
7ae0e400 2946 }
225c7b1f
RD		 2947 }
2948
f6bc11e4
YH		 2949 /* on load remove any previous indication of internal error,
2950 * device is up.
2951 */
2952 dev->persist->state = MLX4_DEVICE_STATE_UP;
2953
ab9c17a0 2954slave_start:
521130d1
EE		 2955 err = mlx4_cmd_init(dev);
2956 if (err) {
1a91de28 2957 mlx4_err(dev, "Failed to init command interface, aborting\n");
ab9c17a0
JM		 2958 goto err_sriov;
2959 }
2960
2961 /* In slave functions, the communication channel must be initialized
2962 * before posting commands. Also, init num_slaves before calling
2963 * mlx4_init_hca */
2964 if (mlx4_is_mfunc(dev)) {
7ae0e400 2965 if (mlx4_is_master(dev)) {
ab9c17a0 2966 dev->num_slaves = MLX4_MAX_NUM_SLAVES;
7ae0e400
MB		 2967
2968 } else {
ab9c17a0 2969 dev->num_slaves = 0;
f356fcbe
JM		 2970 err = mlx4_multi_func_init(dev);
2971 if (err) {
1a91de28 2972 mlx4_err(dev, "Failed to init slave mfunc interface, aborting\n");
ab9c17a0
JM		 2973 goto err_cmd;
2974 }
2975 }
225c7b1f
RD		 2976 }
2977
a0eacca9
MB		 2978 err = mlx4_init_fw(dev);
2979 if (err) {
2980 mlx4_err(dev, "Failed to init fw, aborting.\n");
2981 goto err_mfunc;
2982 }
2983
7ae0e400 2984 if (mlx4_is_master(dev)) {
da315679 2985 /* when we hit the goto slave_start below, dev_cap already initialized */
7ae0e400
MB		 2986 if (!dev_cap) {
2987 dev_cap = kzalloc(sizeof(*dev_cap), GFP_KERNEL);
2988
2989 if (!dev_cap) {
2990 err = -ENOMEM;
2991 goto err_fw;
2992 }
2993
2994 err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
2995 if (err) {
2996 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
2997 goto err_fw;
2998 }
2999
de966c59
MB		 3000 if (mlx4_check_dev_cap(dev, dev_cap, nvfs))
3001 goto err_fw;
3002
7ae0e400 3003 if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS)) {
55ad3592
YH		 3004 u64 dev_flags = mlx4_enable_sriov(dev, pdev,
3005 total_vfs,
3006 existing_vfs,
3007 reset_flow);
7ae0e400 3008
ed3d2276 3009 mlx4_close_fw(dev);
7ae0e400
MB		 3010 mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3011 dev->flags = dev_flags;
3012 if (!SRIOV_VALID_STATE(dev->flags)) {
3013 mlx4_err(dev, "Invalid SRIOV state\n");
3014 goto err_sriov;
3015 }
3016 err = mlx4_reset(dev);
3017 if (err) {
3018 mlx4_err(dev, "Failed to reset HCA, aborting.\n");
3019 goto err_sriov;
3020 }
3021 goto slave_start;
3022 }
3023 } else {
3024 /* Legacy mode FW requires SRIOV to be enabled before
3025 * doing QUERY_DEV_CAP, since max_eq's value is different if
3026 * SRIOV is enabled.
3027 */
3028 memset(dev_cap, 0, sizeof(*dev_cap));
3029 err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
3030 if (err) {
3031 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
3032 goto err_fw;
3033 }
de966c59
MB		 3034
3035 if (mlx4_check_dev_cap(dev, dev_cap, nvfs))
3036 goto err_fw;
7ae0e400
MB		 3037 }
3038 }
3039
225c7b1f 3040 err = mlx4_init_hca(dev);
ab9c17a0
JM		 3041 if (err) {
3042 if (err == -EACCES) {
3043 /* Not primary Physical function
3044 * Running in slave mode */
ffc39f6d 3045 mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
a0eacca9
MB		 3046 /* We're not a PF */
3047 if (dev->flags & MLX4_FLAG_SRIOV) {
3048 if (!existing_vfs)
3049 pci_disable_sriov(pdev);
55ad3592 3050 if (mlx4_is_master(dev) && !reset_flow)
a0eacca9
MB		 3051 atomic_dec(&pf_loading);
3052 dev->flags &= ~MLX4_FLAG_SRIOV;
3053 }
3054 if (!mlx4_is_slave(dev))
3055 mlx4_free_ownership(dev);
ab9c17a0
JM		 3056 dev->flags |= MLX4_FLAG_SLAVE;
3057 dev->flags &= ~MLX4_FLAG_MASTER;
3058 goto slave_start;
3059 } else
a0eacca9 3060 goto err_fw;
ab9c17a0
JM		 3061 }
3062
7ae0e400 3063 if (mlx4_is_master(dev) && (dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS)) {
55ad3592
YH		 3064 u64 dev_flags = mlx4_enable_sriov(dev, pdev, total_vfs,
3065 existing_vfs, reset_flow);
7ae0e400
MB		 3066
3067 if ((dev->flags ^ dev_flags) & (MLX4_FLAG_MASTER | MLX4_FLAG_SLAVE)) {
3068 mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_VHCR);
3069 dev->flags = dev_flags;
3070 err = mlx4_cmd_init(dev);
3071 if (err) {
3072 /* Only VHCR is cleaned up, so could still
3073 * send FW commands
3074 */
3075 mlx4_err(dev, "Failed to init VHCR command interface, aborting\n");
3076 goto err_close;
3077 }
3078 } else {
3079 dev->flags = dev_flags;
3080 }
3081
3082 if (!SRIOV_VALID_STATE(dev->flags)) {
3083 mlx4_err(dev, "Invalid SRIOV state\n");
3084 goto err_close;
3085 }
3086 }
3087
b912b2f8
EP		 3088 /* check if the device is functioning at its maximum possible speed.
3089 * No return code for this call, just warn the user in case of PCI
3090 * express device capabilities are under-satisfied by the bus.
3091 */
83d3459a
EP		 3092 if (!mlx4_is_slave(dev))
3093 mlx4_check_pcie_caps(dev);
b912b2f8 3094
ab9c17a0
JM		 3095 /* In master functions, the communication channel must be initialized
3096 * after obtaining its address from fw */
3097 if (mlx4_is_master(dev)) {
e1c00e10
MD		 3098 if (dev->caps.num_ports < 2 &&
3099 num_vfs_argc > 1) {
3100 err = -EINVAL;
3101 mlx4_err(dev,
3102 "Error: Trying to configure VFs on port 2, but HCA has only %d physical ports\n",
3103 dev->caps.num_ports);
ab9c17a0
JM		 3104 goto err_close;
3105 }
872bf2fb 3106 memcpy(dev->persist->nvfs, nvfs, sizeof(dev->persist->nvfs));
dd41cc3b 3107
872bf2fb
YH		 3108 for (i = 0;
3109 i < sizeof(dev->persist->nvfs)/
3110 sizeof(dev->persist->nvfs[0]); i++) {
e1c00e10
MD		 3111 unsigned j;
3112
872bf2fb 3113 for (j = 0; j < dev->persist->nvfs[i]; ++sum, ++j) {
e1c00e10
MD		 3114 dev->dev_vfs[sum].min_port = i < 2 ? i + 1 : 1;
3115 dev->dev_vfs[sum].n_ports = i < 2 ? 1 :
3116 dev->caps.num_ports;
1ab95d37
MB		 3117 }
3118 }
e1c00e10
MD		 3119
3120 /* In master functions, the communication channel
3121 * must be initialized after obtaining its address from fw
3122 */
3123 err = mlx4_multi_func_init(dev);
3124 if (err) {
3125 mlx4_err(dev, "Failed to init master mfunc interface, aborting.\n");
3126 goto err_close;
3127 }
ab9c17a0 3128 }
225c7b1f 3129
b8dd786f
YP		 3130 err = mlx4_alloc_eq_table(dev);
3131 if (err)
ab9c17a0 3132 goto err_master_mfunc;
b8dd786f 3133
c66fa19c 3134 bitmap_zero(priv->msix_ctl.pool_bm, MAX_MSIX);
730c41d5 3135 mutex_init(&priv->msix_ctl.pool_lock);
0b7ca5a9 3136
08fb1055 3137 mlx4_enable_msi_x(dev);
ab9c17a0
JM		 3138 if ((mlx4_is_mfunc(dev)) &&
3139 !(dev->flags & MLX4_FLAG_MSI_X)) {
f356fcbe 3140 err = -ENOSYS;
1a91de28 3141 mlx4_err(dev, "INTx is not supported in multi-function mode, aborting\n");
b12d93d6 3142 goto err_free_eq;
ab9c17a0
JM		 3143 }
3144
3145 if (!mlx4_is_slave(dev)) {
3146 err = mlx4_init_steering(dev);
3147 if (err)
e1c00e10 3148 goto err_disable_msix;
ab9c17a0 3149 }
b12d93d6 3150
225c7b1f 3151 err = mlx4_setup_hca(dev);
ab9c17a0
JM		 3152 if (err == -EBUSY && (dev->flags & MLX4_FLAG_MSI_X) &&
3153 !mlx4_is_mfunc(dev)) {
08fb1055 3154 dev->flags &= ~MLX4_FLAG_MSI_X;
9858d2d1 3155 dev->caps.num_comp_vectors = 1;
08fb1055
MT		 3156 pci_disable_msix(pdev);
3157 err = mlx4_setup_hca(dev);
3158 }
3159
225c7b1f 3160 if (err)
b12d93d6 3161 goto err_steer;
225c7b1f 3162
5a0d0a61 3163 mlx4_init_quotas(dev);
55ad3592
YH		 3164 /* When PF resources are ready arm its comm channel to enable
3165 * getting commands
3166 */
3167 if (mlx4_is_master(dev)) {
3168 err = mlx4_ARM_COMM_CHANNEL(dev);
3169 if (err) {
3170 mlx4_err(dev, " Failed to arm comm channel eq: %x\n",
3171 err);
3172 goto err_steer;
3173 }
3174 }
5a0d0a61 3175
7ff93f8b
YP		 3176 for (port = 1; port <= dev->caps.num_ports; port++) {
3177 err = mlx4_init_port_info(dev, port);
3178 if (err)
3179 goto err_port;
3180 }
2a2336f8 3181
53f33ae2
MS		 3182 priv->v2p.port1 = 1;
3183 priv->v2p.port2 = 2;
3184
225c7b1f
RD		 3185 err = mlx4_register_device(dev);
3186 if (err)
7ff93f8b 3187 goto err_port;
225c7b1f 3188
b046ffe5
EP		 3189 mlx4_request_modules(dev);
3190
27bf91d6
YP		 3191 mlx4_sense_init(dev);
3192 mlx4_start_sense(dev);
3193
befdf897 3194 priv->removed = 0;
225c7b1f 3195
55ad3592 3196 if (mlx4_is_master(dev) && dev->persist->num_vfs && !reset_flow)
e1a5ddc5
AV		 3197 atomic_dec(&pf_loading);
3198
da315679 3199 kfree(dev_cap);
225c7b1f
RD		 3200 return 0;
3201
7ff93f8b 3202err_port:
b4f77264 3203 for (--port; port >= 1; --port)
7ff93f8b
YP		 3204 mlx4_cleanup_port_info(&priv->port[port]);
3205
f2a3f6a3 3206 mlx4_cleanup_counters_table(dev);
225c7b1f
RD		 3207 mlx4_cleanup_qp_table(dev);
3208 mlx4_cleanup_srq_table(dev);
3209 mlx4_cleanup_cq_table(dev);
3210 mlx4_cmd_use_polling(dev);
3211 mlx4_cleanup_eq_table(dev);
fe6f700d 3212 mlx4_cleanup_mcg_table(dev);
225c7b1f 3213 mlx4_cleanup_mr_table(dev);
012a8ff5 3214 mlx4_cleanup_xrcd_table(dev);
225c7b1f
RD		 3215 mlx4_cleanup_pd_table(dev);
3216 mlx4_cleanup_uar_table(dev);
3217
b12d93d6 3218err_steer:
ab9c17a0
JM		 3219 if (!mlx4_is_slave(dev))
3220 mlx4_clear_steering(dev);
b12d93d6 3221
e1c00e10
MD		 3222err_disable_msix:
3223 if (dev->flags & MLX4_FLAG_MSI_X)
3224 pci_disable_msix(pdev);
3225
b8dd786f
YP		 3226err_free_eq:
3227 mlx4_free_eq_table(dev);
3228
ab9c17a0 3229err_master_mfunc:
772103e6
JM		 3230 if (mlx4_is_master(dev)) {
3231 mlx4_free_resource_tracker(dev, RES_TR_FREE_STRUCTS_ONLY);
ab9c17a0 3232 mlx4_multi_func_cleanup(dev);
772103e6 3233 }
ab9c17a0 3234
b38f2879 3235 if (mlx4_is_slave(dev)) {
99ec41d0 3236 kfree(dev->caps.qp0_qkey);
b38f2879
DB		 3237 kfree(dev->caps.qp0_tunnel);
3238 kfree(dev->caps.qp0_proxy);
3239 kfree(dev->caps.qp1_tunnel);
3240 kfree(dev->caps.qp1_proxy);
3241 }
3242
225c7b1f
RD		 3243err_close:
3244 mlx4_close_hca(dev);
3245
a0eacca9
MB		 3246err_fw:
3247 mlx4_close_fw(dev);
3248
ab9c17a0
JM		 3249err_mfunc:
3250 if (mlx4_is_slave(dev))
3251 mlx4_multi_func_cleanup(dev);
3252
225c7b1f 3253err_cmd:
ffc39f6d 3254 mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
225c7b1f 3255
ab9c17a0 3256err_sriov:
55ad3592 3257 if (dev->flags & MLX4_FLAG_SRIOV && !existing_vfs) {
ab9c17a0 3258 pci_disable_sriov(pdev);
55ad3592
YH		 3259 dev->flags &= ~MLX4_FLAG_SRIOV;
3260 }
ab9c17a0 3261
55ad3592 3262 if (mlx4_is_master(dev) && dev->persist->num_vfs && !reset_flow)
e1a5ddc5
AV		 3263 atomic_dec(&pf_loading);
3264
1ab95d37
MB		 3265 kfree(priv->dev.dev_vfs);
3266
e1c00e10
MD		 3267 if (!mlx4_is_slave(dev))
3268 mlx4_free_ownership(dev);
3269
7ae0e400 3270 kfree(dev_cap);
e1c00e10
MD		 3271 return err;
3272}
3273
3274static int __mlx4_init_one(struct pci_dev *pdev, int pci_dev_data,
3275 struct mlx4_priv *priv)
3276{
3277 int err;
3278 int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
3279 int prb_vf[MLX4_MAX_PORTS + 1] = {0, 0, 0};
3280 const int param_map[MLX4_MAX_PORTS + 1][MLX4_MAX_PORTS + 1] = {
3281 {2, 0, 0}, {0, 1, 2}, {0, 1, 2} };
3282 unsigned total_vfs = 0;
3283 unsigned int i;
3284
3285 pr_info(DRV_NAME ": Initializing %s\n", pci_name(pdev));
3286
3287 err = pci_enable_device(pdev);
3288 if (err) {
3289 dev_err(&pdev->dev, "Cannot enable PCI device, aborting\n");
3290 return err;
3291 }
3292
3293 /* Due to requirement that all VFs and the PF are *guaranteed* 2 MACS
3294 * per port, we must limit the number of VFs to 63 (since their are
3295 * 128 MACs)
3296 */
3297 for (i = 0; i < sizeof(nvfs)/sizeof(nvfs[0]) && i < num_vfs_argc;
3298 total_vfs += nvfs[param_map[num_vfs_argc - 1][i]], i++) {
3299 nvfs[param_map[num_vfs_argc - 1][i]] = num_vfs[i];
3300 if (nvfs[i] < 0) {
3301 dev_err(&pdev->dev, "num_vfs module parameter cannot be negative\n");
3302 err = -EINVAL;
3303 goto err_disable_pdev;
3304 }
3305 }
3306 for (i = 0; i < sizeof(prb_vf)/sizeof(prb_vf[0]) && i < probe_vfs_argc;
3307 i++) {
3308 prb_vf[param_map[probe_vfs_argc - 1][i]] = probe_vf[i];
3309 if (prb_vf[i] < 0 || prb_vf[i] > nvfs[i]) {
3310 dev_err(&pdev->dev, "probe_vf module parameter cannot be negative or greater than num_vfs\n");
3311 err = -EINVAL;
3312 goto err_disable_pdev;
3313 }
3314 }
3315 if (total_vfs >= MLX4_MAX_NUM_VF) {
3316 dev_err(&pdev->dev,
3317 "Requested more VF's (%d) than allowed (%d)\n",
3318 total_vfs, MLX4_MAX_NUM_VF - 1);
3319 err = -EINVAL;
3320 goto err_disable_pdev;
3321 }
3322
3323 for (i = 0; i < MLX4_MAX_PORTS; i++) {
3324 if (nvfs[i] + nvfs[2] >= MLX4_MAX_NUM_VF_P_PORT) {
3325 dev_err(&pdev->dev,
3326 "Requested more VF's (%d) for port (%d) than allowed (%d)\n",
3327 nvfs[i] + nvfs[2], i + 1,
3328 MLX4_MAX_NUM_VF_P_PORT - 1);
3329 err = -EINVAL;
3330 goto err_disable_pdev;
3331 }
3332 }
3333
3334 /* Check for BARs. */
3335 if (!(pci_dev_data & MLX4_PCI_DEV_IS_VF) &&
3336 !(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) {
3337 dev_err(&pdev->dev, "Missing DCS, aborting (driver_data: 0x%x, pci_resource_flags(pdev, 0):0x%lx)\n",
3338 pci_dev_data, pci_resource_flags(pdev, 0));
3339 err = -ENODEV;
3340 goto err_disable_pdev;
3341 }
3342 if (!(pci_resource_flags(pdev, 2) & IORESOURCE_MEM)) {
3343 dev_err(&pdev->dev, "Missing UAR, aborting\n");
3344 err = -ENODEV;
3345 goto err_disable_pdev;
3346 }
3347
3348 err = pci_request_regions(pdev, DRV_NAME);
3349 if (err) {
3350 dev_err(&pdev->dev, "Couldn't get PCI resources, aborting\n");
3351 goto err_disable_pdev;
3352 }
3353
3354 pci_set_master(pdev);
3355
3356 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
3357 if (err) {
3358 dev_warn(&pdev->dev, "Warning: couldn't set 64-bit PCI DMA mask\n");
3359 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
3360 if (err) {
3361 dev_err(&pdev->dev, "Can't set PCI DMA mask, aborting\n");
3362 goto err_release_regions;
3363 }
3364 }
3365 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
3366 if (err) {
3367 dev_warn(&pdev->dev, "Warning: couldn't set 64-bit consistent PCI DMA mask\n");
3368 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
3369 if (err) {
3370 dev_err(&pdev->dev, "Can't set consistent PCI DMA mask, aborting\n");
3371 goto err_release_regions;
3372 }
3373 }
3374
3375 /* Allow large DMA segments, up to the firmware limit of 1 GB */
3376 dma_set_max_seg_size(&pdev->dev, 1024 * 1024 * 1024);
3377 /* Detect if this device is a virtual function */
3378 if (pci_dev_data & MLX4_PCI_DEV_IS_VF) {
3379 /* When acting as pf, we normally skip vfs unless explicitly
3380 * requested to probe them.
3381 */
3382 if (total_vfs) {
3383 unsigned vfs_offset = 0;
3384
3385 for (i = 0; i < sizeof(nvfs)/sizeof(nvfs[0]) &&
3386 vfs_offset + nvfs[i] < extended_func_num(pdev);
3387 vfs_offset += nvfs[i], i++)
3388 ;
3389 if (i == sizeof(nvfs)/sizeof(nvfs[0])) {
3390 err = -ENODEV;
3391 goto err_release_regions;
3392 }
3393 if ((extended_func_num(pdev) - vfs_offset)
3394 > prb_vf[i]) {
3395 dev_warn(&pdev->dev, "Skipping virtual function:%d\n",
3396 extended_func_num(pdev));
3397 err = -ENODEV;
3398 goto err_release_regions;
3399 }
3400 }
3401 }
3402
ad9a0bf0 3403 err = mlx4_catas_init(&priv->dev);
e1c00e10
MD		 3404 if (err)
3405 goto err_release_regions;
ad9a0bf0 3406
55ad3592 3407 err = mlx4_load_one(pdev, pci_dev_data, total_vfs, nvfs, priv, 0);
ad9a0bf0
YH		 3408 if (err)
3409 goto err_catas;
3410
e1c00e10 3411 return 0;
225c7b1f 3412
ad9a0bf0
YH		 3413err_catas:
3414 mlx4_catas_end(&priv->dev);
3415
a01df0fe
RD		 3416err_release_regions:
3417 pci_release_regions(pdev);
225c7b1f
RD		 3418
3419err_disable_pdev:
3420 pci_disable_device(pdev);
3421 pci_set_drvdata(pdev, NULL);
3422 return err;
3423}
3424
1dd06ae8 3425static int mlx4_init_one(struct pci_dev *pdev, const struct pci_device_id *id)
3d73c288 3426{
befdf897
WY		 3427 struct mlx4_priv *priv;
3428 struct mlx4_dev *dev;
e1c00e10 3429 int ret;
befdf897 3430
0a645e80 3431 printk_once(KERN_INFO "%s", mlx4_version);
3d73c288 3432
befdf897
WY		 3433 priv = kzalloc(sizeof(*priv), GFP_KERNEL);
3434 if (!priv)
3435 return -ENOMEM;
3436
3437 dev = &priv->dev;
872bf2fb
YH		 3438 dev->persist = kzalloc(sizeof(*dev->persist), GFP_KERNEL);
3439 if (!dev->persist) {
3440 kfree(priv);
3441 return -ENOMEM;
3442 }
3443 dev->persist->pdev = pdev;
3444 dev->persist->dev = dev;
3445 pci_set_drvdata(pdev, dev->persist);
befdf897 3446 priv->pci_dev_data = id->driver_data;
f6bc11e4 3447 mutex_init(&dev->persist->device_state_mutex);
c69453e2 3448 mutex_init(&dev->persist->interface_state_mutex);
befdf897 3449
e1c00e10 3450 ret = __mlx4_init_one(pdev, id->driver_data, priv);
872bf2fb
YH		 3451 if (ret) {
3452 kfree(dev->persist);
e1c00e10 3453 kfree(priv);
2ba5fbd6
YH		 3454 } else {
3455 pci_save_state(pdev);
872bf2fb 3456 }
2ba5fbd6 3457
e1c00e10 3458 return ret;
3d73c288
RD		 3459}
3460
dd0eefe3
YH		 3461static void mlx4_clean_dev(struct mlx4_dev *dev)
3462{
3463 struct mlx4_dev_persistent *persist = dev->persist;
3464 struct mlx4_priv *priv = mlx4_priv(dev);
55ad3592 3465 unsigned long flags = (dev->flags & RESET_PERSIST_MASK_FLAGS);
dd0eefe3
YH		 3466
3467 memset(priv, 0, sizeof(*priv));
3468 priv->dev.persist = persist;
55ad3592 3469 priv->dev.flags = flags;
dd0eefe3
YH		 3470}
3471
e1c00e10 3472static void mlx4_unload_one(struct pci_dev *pdev)
225c7b1f 3473{
872bf2fb
YH		 3474 struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
3475 struct mlx4_dev *dev = persist->dev;
225c7b1f 3476 struct mlx4_priv *priv = mlx4_priv(dev);
befdf897 3477 int pci_dev_data;
dd0eefe3 3478 int p, i;
225c7b1f 3479
befdf897
WY		 3480 if (priv->removed)
3481 return;
225c7b1f 3482
dd0eefe3
YH		 3483 /* saving current ports type for further use */
3484 for (i = 0; i < dev->caps.num_ports; i++) {
3485 dev->persist->curr_port_type[i] = dev->caps.port_type[i + 1];
3486 dev->persist->curr_port_poss_type[i] = dev->caps.
3487 possible_type[i + 1];
3488 }
3489
befdf897 3490 pci_dev_data = priv->pci_dev_data;
225c7b1f 3491
befdf897
WY		 3492 mlx4_stop_sense(dev);
3493 mlx4_unregister_device(dev);
225c7b1f 3494
befdf897
WY		 3495 for (p = 1; p <= dev->caps.num_ports; p++) {
3496 mlx4_cleanup_port_info(&priv->port[p]);
3497 mlx4_CLOSE_PORT(dev, p);
3498 }
3499
3500 if (mlx4_is_master(dev))
3501 mlx4_free_resource_tracker(dev,
3502 RES_TR_FREE_SLAVES_ONLY);
3503
3504 mlx4_cleanup_counters_table(dev);
3505 mlx4_cleanup_qp_table(dev);
3506 mlx4_cleanup_srq_table(dev);
3507 mlx4_cleanup_cq_table(dev);
3508 mlx4_cmd_use_polling(dev);
3509 mlx4_cleanup_eq_table(dev);
3510 mlx4_cleanup_mcg_table(dev);
3511 mlx4_cleanup_mr_table(dev);
3512 mlx4_cleanup_xrcd_table(dev);
3513 mlx4_cleanup_pd_table(dev);
225c7b1f 3514
befdf897
WY		 3515 if (mlx4_is_master(dev))
3516 mlx4_free_resource_tracker(dev,
3517 RES_TR_FREE_STRUCTS_ONLY);
47605df9 3518
befdf897
WY		 3519 iounmap(priv->kar);
3520 mlx4_uar_free(dev, &priv->driver_uar);
3521 mlx4_cleanup_uar_table(dev);
3522 if (!mlx4_is_slave(dev))
3523 mlx4_clear_steering(dev);
3524 mlx4_free_eq_table(dev);
3525 if (mlx4_is_master(dev))
3526 mlx4_multi_func_cleanup(dev);
3527 mlx4_close_hca(dev);
a0eacca9 3528 mlx4_close_fw(dev);
befdf897
WY		 3529 if (mlx4_is_slave(dev))
3530 mlx4_multi_func_cleanup(dev);
ffc39f6d 3531 mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
47605df9 3532
befdf897
WY		 3533 if (dev->flags & MLX4_FLAG_MSI_X)
3534 pci_disable_msix(pdev);
befdf897
WY		 3535
3536 if (!mlx4_is_slave(dev))
3537 mlx4_free_ownership(dev);
3538
99ec41d0 3539 kfree(dev->caps.qp0_qkey);
befdf897
WY		 3540 kfree(dev->caps.qp0_tunnel);
3541 kfree(dev->caps.qp0_proxy);
3542 kfree(dev->caps.qp1_tunnel);
3543 kfree(dev->caps.qp1_proxy);
3544 kfree(dev->dev_vfs);
3545
dd0eefe3 3546 mlx4_clean_dev(dev);
befdf897
WY		 3547 priv->pci_dev_data = pci_dev_data;
3548 priv->removed = 1;
3549}
3550
3551static void mlx4_remove_one(struct pci_dev *pdev)
3552{
872bf2fb
YH		 3553 struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
3554 struct mlx4_dev *dev = persist->dev;
befdf897 3555 struct mlx4_priv *priv = mlx4_priv(dev);
55ad3592 3556 int active_vfs = 0;
befdf897 3557
c69453e2
YH		 3558 mutex_lock(&persist->interface_state_mutex);
3559 persist->interface_state |= MLX4_INTERFACE_STATE_DELETION;
3560 mutex_unlock(&persist->interface_state_mutex);
3561
55ad3592
YH		 3562 /* Disabling SR-IOV is not allowed while there are active vf's */
3563 if (mlx4_is_master(dev) && dev->flags & MLX4_FLAG_SRIOV) {
3564 active_vfs = mlx4_how_many_lives_vf(dev);
3565 if (active_vfs) {
3566 pr_warn("Removing PF when there are active VF's !!\n");
3567 pr_warn("Will not disable SR-IOV.\n");
3568 }
3569 }
3570
c69453e2
YH		 3571 /* device marked to be under deletion running now without the lock
3572 * letting other tasks to be terminated
3573 */
3574 if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
3575 mlx4_unload_one(pdev);
3576 else
3577 mlx4_info(dev, "%s: interface is down\n", __func__);
ad9a0bf0 3578 mlx4_catas_end(dev);
55ad3592
YH		 3579 if (dev->flags & MLX4_FLAG_SRIOV && !active_vfs) {
3580 mlx4_warn(dev, "Disabling SR-IOV\n");
3581 pci_disable_sriov(pdev);
3582 }
3583
e1c00e10
MD		 3584 pci_release_regions(pdev);
3585 pci_disable_device(pdev);
872bf2fb 3586 kfree(dev->persist);
befdf897
WY		 3587 kfree(priv);
3588 pci_set_drvdata(pdev, NULL);
225c7b1f
RD		 3589}
3590
dd0eefe3
YH		 3591static int restore_current_port_types(struct mlx4_dev *dev,
3592 enum mlx4_port_type *types,
3593 enum mlx4_port_type *poss_types)
3594{
3595 struct mlx4_priv *priv = mlx4_priv(dev);
3596 int err, i;
3597
3598 mlx4_stop_sense(dev);
3599
3600 mutex_lock(&priv->port_mutex);
3601 for (i = 0; i < dev->caps.num_ports; i++)
3602 dev->caps.possible_type[i + 1] = poss_types[i];
3603 err = mlx4_change_port_types(dev, types);
3604 mlx4_start_sense(dev);
3605 mutex_unlock(&priv->port_mutex);
3606
3607 return err;
3608}
3609
ee49bd93
JM		 3610int mlx4_restart_one(struct pci_dev *pdev)
3611{
872bf2fb
YH		 3612 struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
3613 struct mlx4_dev *dev = persist->dev;
839f1243 3614 struct mlx4_priv *priv = mlx4_priv(dev);
e1c00e10
MD		 3615 int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
3616 int pci_dev_data, err, total_vfs;
839f1243
RD		 3617
3618 pci_dev_data = priv->pci_dev_data;
872bf2fb
YH		 3619 total_vfs = dev->persist->num_vfs;
3620 memcpy(nvfs, dev->persist->nvfs, sizeof(dev->persist->nvfs));
e1c00e10
MD		 3621
3622 mlx4_unload_one(pdev);
55ad3592 3623 err = mlx4_load_one(pdev, pci_dev_data, total_vfs, nvfs, priv, 1);
e1c00e10
MD		 3624 if (err) {
3625 mlx4_err(dev, "%s: ERROR: mlx4_load_one failed, pci_name=%s, err=%d\n",
3626 __func__, pci_name(pdev), err);
3627 return err;
3628 }
3629
dd0eefe3
YH		 3630 err = restore_current_port_types(dev, dev->persist->curr_port_type,
3631 dev->persist->curr_port_poss_type);
3632 if (err)
3633 mlx4_err(dev, "could not restore original port types (%d)\n",
3634 err);
3635
e1c00e10 3636 return err;
ee49bd93
JM		 3637}
3638
9baa3c34 3639static const struct pci_device_id mlx4_pci_table[] = {
ab9c17a0 3640 /* MT25408 "Hermon" SDR */
ca3e57a5 3641 { PCI_VDEVICE(MELLANOX, 0x6340), MLX4_PCI_DEV_FORCE_SENSE_PORT },
ab9c17a0 3642 /* MT25408 "Hermon" DDR */
ca3e57a5 3643 { PCI_VDEVICE(MELLANOX, 0x634a), MLX4_PCI_DEV_FORCE_SENSE_PORT },
ab9c17a0 3644 /* MT25408 "Hermon" QDR */
ca3e57a5 3645 { PCI_VDEVICE(MELLANOX, 0x6354), MLX4_PCI_DEV_FORCE_SENSE_PORT },
ab9c17a0 3646 /* MT25408 "Hermon" DDR PCIe gen2 */
ca3e57a5 3647 { PCI_VDEVICE(MELLANOX, 0x6732), MLX4_PCI_DEV_FORCE_SENSE_PORT },
ab9c17a0 3648 /* MT25408 "Hermon" QDR PCIe gen2 */
ca3e57a5 3649 { PCI_VDEVICE(MELLANOX, 0x673c), MLX4_PCI_DEV_FORCE_SENSE_PORT },
ab9c17a0 3650 /* MT25408 "Hermon" EN 10GigE */
ca3e57a5 3651 { PCI_VDEVICE(MELLANOX, 0x6368), MLX4_PCI_DEV_FORCE_SENSE_PORT },
ab9c17a0 3652 /* MT25408 "Hermon" EN 10GigE PCIe gen2 */
ca3e57a5 3653 { PCI_VDEVICE(MELLANOX, 0x6750), MLX4_PCI_DEV_FORCE_SENSE_PORT },
ab9c17a0 3654 /* MT25458 ConnectX EN 10GBASE-T 10GigE */
ca3e57a5 3655 { PCI_VDEVICE(MELLANOX, 0x6372), MLX4_PCI_DEV_FORCE_SENSE_PORT },
ab9c17a0 3656 /* MT25458 ConnectX EN 10GBASE-T+Gen2 10GigE */
ca3e57a5 3657 { PCI_VDEVICE(MELLANOX, 0x675a), MLX4_PCI_DEV_FORCE_SENSE_PORT },
ab9c17a0 3658 /* MT26468 ConnectX EN 10GigE PCIe gen2*/
ca3e57a5 3659 { PCI_VDEVICE(MELLANOX, 0x6764), MLX4_PCI_DEV_FORCE_SENSE_PORT },
ab9c17a0 3660 /* MT26438 ConnectX EN 40GigE PCIe gen2 5GT/s */
ca3e57a5 3661 { PCI_VDEVICE(MELLANOX, 0x6746), MLX4_PCI_DEV_FORCE_SENSE_PORT },
ab9c17a0 3662 /* MT26478 ConnectX2 40GigE PCIe gen2 */
ca3e57a5 3663 { PCI_VDEVICE(MELLANOX, 0x676e), MLX4_PCI_DEV_FORCE_SENSE_PORT },
ab9c17a0 3664 /* MT25400 Family [ConnectX-2 Virtual Function] */
839f1243 3665 { PCI_VDEVICE(MELLANOX, 0x1002), MLX4_PCI_DEV_IS_VF },
ab9c17a0
JM		 3666 /* MT27500 Family [ConnectX-3] */
3667 { PCI_VDEVICE(MELLANOX, 0x1003), 0 },
3668 /* MT27500 Family [ConnectX-3 Virtual Function] */
839f1243 3669 { PCI_VDEVICE(MELLANOX, 0x1004), MLX4_PCI_DEV_IS_VF },
ab9c17a0
JM		 3670 { PCI_VDEVICE(MELLANOX, 0x1005), 0 }, /* MT27510 Family */
3671 { PCI_VDEVICE(MELLANOX, 0x1006), 0 }, /* MT27511 Family */
3672 { PCI_VDEVICE(MELLANOX, 0x1007), 0 }, /* MT27520 Family */
3673 { PCI_VDEVICE(MELLANOX, 0x1008), 0 }, /* MT27521 Family */
3674 { PCI_VDEVICE(MELLANOX, 0x1009), 0 }, /* MT27530 Family */
3675 { PCI_VDEVICE(MELLANOX, 0x100a), 0 }, /* MT27531 Family */
3676 { PCI_VDEVICE(MELLANOX, 0x100b), 0 }, /* MT27540 Family */
3677 { PCI_VDEVICE(MELLANOX, 0x100c), 0 }, /* MT27541 Family */
3678 { PCI_VDEVICE(MELLANOX, 0x100d), 0 }, /* MT27550 Family */
3679 { PCI_VDEVICE(MELLANOX, 0x100e), 0 }, /* MT27551 Family */
3680 { PCI_VDEVICE(MELLANOX, 0x100f), 0 }, /* MT27560 Family */
3681 { PCI_VDEVICE(MELLANOX, 0x1010), 0 }, /* MT27561 Family */
225c7b1f
RD		 3682 { 0, }
3683};
3684
3685MODULE_DEVICE_TABLE(pci, mlx4_pci_table);
3686
57dbf29a
KSS		 3687static pci_ers_result_t mlx4_pci_err_detected(struct pci_dev *pdev,
3688 pci_channel_state_t state)
3689{
2ba5fbd6
YH		 3690 struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
3691
3692 mlx4_err(persist->dev, "mlx4_pci_err_detected was called\n");
3693 mlx4_enter_error_state(persist);
57dbf29a 3694
2ba5fbd6
YH		 3695 mutex_lock(&persist->interface_state_mutex);
3696 if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
3697 mlx4_unload_one(pdev);
3698
3699 mutex_unlock(&persist->interface_state_mutex);
3700 if (state == pci_channel_io_perm_failure)
3701 return PCI_ERS_RESULT_DISCONNECT;
3702
3703 pci_disable_device(pdev);
3704 return PCI_ERS_RESULT_NEED_RESET;
57dbf29a
KSS		 3705}
3706
3707static pci_ers_result_t mlx4_pci_slot_reset(struct pci_dev *pdev)
3708{
2ba5fbd6
YH		 3709 struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
3710 struct mlx4_dev *dev = persist->dev;
befdf897
WY		 3711 struct mlx4_priv *priv = mlx4_priv(dev);
3712 int ret;
2ba5fbd6
YH		 3713 int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
3714 int total_vfs;
97a5221f 3715
2ba5fbd6
YH		 3716 mlx4_err(dev, "mlx4_pci_slot_reset was called\n");
3717 ret = pci_enable_device(pdev);
3718 if (ret) {
3719 mlx4_err(dev, "Can not re-enable device, ret=%d\n", ret);
3720 return PCI_ERS_RESULT_DISCONNECT;
3721 }
3722
3723 pci_set_master(pdev);
3724 pci_restore_state(pdev);
3725 pci_save_state(pdev);
3726
3727 total_vfs = dev->persist->num_vfs;
3728 memcpy(nvfs, dev->persist->nvfs, sizeof(dev->persist->nvfs));
3729
3730 mutex_lock(&persist->interface_state_mutex);
3731 if (!(persist->interface_state & MLX4_INTERFACE_STATE_UP)) {
3732 ret = mlx4_load_one(pdev, priv->pci_dev_data, total_vfs, nvfs,
55ad3592 3733 priv, 1);
2ba5fbd6
YH		 3734 if (ret) {
3735 mlx4_err(dev, "%s: mlx4_load_one failed, ret=%d\n",
3736 __func__, ret);
3737 goto end;
3738 }
3739
3740 ret = restore_current_port_types(dev, dev->persist->
3741 curr_port_type, dev->persist->
3742 curr_port_poss_type);
3743 if (ret)
3744 mlx4_err(dev, "could not restore original port types (%d)\n", ret);
3745 }
3746end:
3747 mutex_unlock(&persist->interface_state_mutex);
57dbf29a
KSS		 3748
3749 return ret ? PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_RECOVERED;
3750}
3751
2ba5fbd6
YH		 3752static void mlx4_shutdown(struct pci_dev *pdev)
3753{
3754 struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
3755
3756 mlx4_info(persist->dev, "mlx4_shutdown was called\n");
3757 mutex_lock(&persist->interface_state_mutex);
3758 if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
3759 mlx4_unload_one(pdev);
3760 mutex_unlock(&persist->interface_state_mutex);
3761}
3762
3646f0e5 3763static const struct pci_error_handlers mlx4_err_handler = {
57dbf29a
KSS		 3764 .error_detected = mlx4_pci_err_detected,
3765 .slot_reset = mlx4_pci_slot_reset,
3766};
3767
225c7b1f
RD		 3768static struct pci_driver mlx4_driver = {
3769 .name = DRV_NAME,
3770 .id_table = mlx4_pci_table,
3771 .probe = mlx4_init_one,
2ba5fbd6 3772 .shutdown = mlx4_shutdown,
f57e6848 3773 .remove = mlx4_remove_one,
57dbf29a 3774 .err_handler = &mlx4_err_handler,
225c7b1f
RD		 3775};
3776
7ff93f8b
YP		 3777static int __init mlx4_verify_params(void)
3778{
3779 if ((log_num_mac < 0) || (log_num_mac > 7)) {
c20862c8 3780 pr_warn("mlx4_core: bad num_mac: %d\n", log_num_mac);
7ff93f8b
YP		 3781 return -1;
3782 }
3783
cb29688a 3784 if (log_num_vlan != 0)
c20862c8
AV		 3785 pr_warn("mlx4_core: log_num_vlan - obsolete module param, using %d\n",
3786 MLX4_LOG_NUM_VLANS);
7ff93f8b 3787
ecc8fb11
AV		 3788 if (use_prio != 0)
3789 pr_warn("mlx4_core: use_prio - obsolete module param, ignored\n");
7ff93f8b 3790
0498628f 3791 if ((log_mtts_per_seg < 1) || (log_mtts_per_seg > 7)) {
c20862c8
AV		 3792 pr_warn("mlx4_core: bad log_mtts_per_seg: %d\n",
3793 log_mtts_per_seg);
ab6bf42e
EC		 3794 return -1;
3795 }
3796
ab9c17a0
JM		 3797 /* Check if module param for ports type has legal combination */
3798 if (port_type_array[0] == false && port_type_array[1] == true) {
c20862c8 3799 pr_warn("Module parameter configuration ETH/IB is not supported. Switching to default configuration IB/IB\n");
ab9c17a0
JM		 3800 port_type_array[0] = true;
3801 }
3802
7d077cd3
MB		 3803 if (mlx4_log_num_mgm_entry_size < -7 ||
3804 (mlx4_log_num_mgm_entry_size > 0 &&
3805 (mlx4_log_num_mgm_entry_size < MLX4_MIN_MGM_LOG_ENTRY_SIZE ||
3806 mlx4_log_num_mgm_entry_size > MLX4_MAX_MGM_LOG_ENTRY_SIZE))) {
3807 pr_warn("mlx4_core: mlx4_log_num_mgm_entry_size (%d) not in legal range (-7..0 or %d..%d)\n",
1a91de28
JP		 3808 mlx4_log_num_mgm_entry_size,
3809 MLX4_MIN_MGM_LOG_ENTRY_SIZE,
3810 MLX4_MAX_MGM_LOG_ENTRY_SIZE);
3c439b55
JM		 3811 return -1;
3812 }
3813
7ff93f8b
YP		 3814 return 0;
3815}
3816
225c7b1f
RD		 3817static int __init mlx4_init(void)
3818{
3819 int ret;
3820
7ff93f8b
YP		 3821 if (mlx4_verify_params())
3822 return -EINVAL;
3823
27bf91d6
YP		 3824
3825 mlx4_wq = create_singlethread_workqueue("mlx4");
3826 if (!mlx4_wq)
3827 return -ENOMEM;
ee49bd93 3828
225c7b1f 3829 ret = pci_register_driver(&mlx4_driver);
1b85ee09
WY		 3830 if (ret < 0)
3831 destroy_workqueue(mlx4_wq);
225c7b1f
RD		 3832 return ret < 0 ? ret : 0;
3833}
3834
3835static void __exit mlx4_cleanup(void)
3836{
3837 pci_unregister_driver(&mlx4_driver);
27bf91d6 3838 destroy_workqueue(mlx4_wq);
225c7b1f
RD		 3839}
3840
3841module_init(mlx4_init);
3842module_exit(mlx4_cleanup);
Linux 6.x block layer and io_uring tree(s)