This source file includes following definitions.
- mlx4_devlink_ierr_reset_get
- mlx4_devlink_ierr_reset_set
- mlx4_devlink_crdump_snapshot_get
- mlx4_devlink_crdump_snapshot_set
- mlx4_devlink_max_macs_validate
- mlx4_devlink_set_params_init_values
- mlx4_set_num_reserved_uars
- mlx4_check_port_params
- mlx4_set_port_mask
- mlx4_query_func
- mlx4_enable_cqe_eqe_stride
- _mlx4_dev_port
- mlx4_dev_port
- mlx4_enable_ignore_fcs
- mlx4_dev_cap
- mlx4_how_many_lives_vf
- mlx4_get_parav_qkey
- mlx4_sync_pkey_table
- mlx4_put_slave_node_guid
- mlx4_get_slave_node_guid
- mlx4_is_slave_active
- mlx4_handle_eth_header_mcast_prio
- slave_adjust_steering_mode
- mlx4_slave_destroy_special_qp_cap
- mlx4_slave_special_qp_cap
- mlx4_slave_cap
- mlx4_request_modules
- mlx4_change_port_types
- show_port_type
- __set_port_type
- set_port_type
- int_to_ibta_mtu
- ibta_mtu_to_int
- show_port_ib_mtu
- set_port_ib_mtu
- mlx4_mf_bond
- mlx4_mf_unbond
- mlx4_bond
- mlx4_unbond
- mlx4_port_map_set
- mlx4_load_fw
- mlx4_init_cmpt_table
- mlx4_init_icm
- mlx4_free_icms
- mlx4_slave_exit
- map_bf_area
- unmap_bf_area
- mlx4_read_clock
- map_internal_clock
- mlx4_get_internal_clock_params
- unmap_internal_clock
- mlx4_close_hca
- mlx4_close_fw
- mlx4_comm_check_offline
- mlx4_reset_vf_support
- mlx4_init_slave
- mlx4_parav_master_pf_caps
- choose_log_fs_mgm_entry_size
- dmfs_high_rate_steering_mode_str
- choose_steering_mode
- choose_tunnel_offload_mode
- mlx4_validate_optimized_steering
- mlx4_init_fw
- mlx4_init_hca
- mlx4_init_counters_table
- mlx4_cleanup_counters_table
- mlx4_cleanup_default_counters
- mlx4_allocate_default_counters
- __mlx4_counter_alloc
- mlx4_counter_alloc
- __mlx4_clear_if_stat
- __mlx4_counter_free
- mlx4_counter_free
- mlx4_get_default_counter_index
- mlx4_set_admin_guid
- mlx4_get_admin_guid
- mlx4_set_random_admin_guid
- mlx4_setup_hca
- mlx4_init_affinity_hint
- mlx4_enable_msi_x
- mlx4_init_port_info
- mlx4_cleanup_port_info
- mlx4_init_steering
- mlx4_clear_steering
- extended_func_num
- mlx4_get_ownership
- mlx4_free_ownership
- mlx4_enable_sriov
- mlx4_check_dev_cap
- mlx4_pci_enable_device
- mlx4_pci_disable_device
- mlx4_load_one
- __mlx4_init_one
- mlx4_devlink_port_type_set
- mlx4_devlink_param_load_driverinit_values
- mlx4_devlink_reload_down
- mlx4_devlink_reload_up
- mlx4_init_one
- mlx4_clean_dev
- mlx4_unload_one
- mlx4_remove_one
- restore_current_port_types
- mlx4_restart_one_down
- mlx4_restart_one_up
- mlx4_restart_one
- mlx4_pci_err_detected
- mlx4_pci_slot_reset
- mlx4_pci_resume
- mlx4_shutdown
- mlx4_suspend
- mlx4_resume
- mlx4_verify_params
- mlx4_init
- mlx4_cleanup
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36 #include <linux/module.h>
37 #include <linux/kernel.h>
38 #include <linux/init.h>
39 #include <linux/errno.h>
40 #include <linux/pci.h>
41 #include <linux/dma-mapping.h>
42 #include <linux/slab.h>
43 #include <linux/io-mapping.h>
44 #include <linux/delay.h>
45 #include <linux/kmod.h>
46 #include <linux/etherdevice.h>
47 #include <net/devlink.h>
48
49 #include <uapi/rdma/mlx4-abi.h>
50 #include <linux/mlx4/device.h>
51 #include <linux/mlx4/doorbell.h>
52
53 #include "mlx4.h"
54 #include "fw.h"
55 #include "icm.h"
56
57 MODULE_AUTHOR("Roland Dreier");
58 MODULE_DESCRIPTION("Mellanox ConnectX HCA low-level driver");
59 MODULE_LICENSE("Dual BSD/GPL");
60 MODULE_VERSION(DRV_VERSION);
61
62 struct workqueue_struct *mlx4_wq;
63
64 #ifdef CONFIG_MLX4_DEBUG
65
66 int mlx4_debug_level;
67 module_param_named(debug_level, mlx4_debug_level, int, 0644);
68 MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0");
69
70 #endif
71
72 #ifdef CONFIG_PCI_MSI
73
74 static int msi_x = 1;
75 module_param(msi_x, int, 0444);
76 MODULE_PARM_DESC(msi_x, "0 - don't use MSI-X, 1 - use MSI-X, >1 - limit number of MSI-X irqs to msi_x");
77
78 #else
79
80 #define msi_x (0)
81
82 #endif
83
84 static uint8_t num_vfs[3] = {0, 0, 0};
85 static int num_vfs_argc;
86 module_param_array(num_vfs, byte, &num_vfs_argc, 0444);
87 MODULE_PARM_DESC(num_vfs, "enable #num_vfs functions if num_vfs > 0\n"
88 "num_vfs=port1,port2,port1+2");
89
90 static uint8_t probe_vf[3] = {0, 0, 0};
91 static int probe_vfs_argc;
92 module_param_array(probe_vf, byte, &probe_vfs_argc, 0444);
93 MODULE_PARM_DESC(probe_vf, "number of vfs to probe by pf driver (num_vfs > 0)\n"
94 "probe_vf=port1,port2,port1+2");
95
96 static int mlx4_log_num_mgm_entry_size = MLX4_DEFAULT_MGM_LOG_ENTRY_SIZE;
97 module_param_named(log_num_mgm_entry_size,
98 mlx4_log_num_mgm_entry_size, int, 0444);
99 MODULE_PARM_DESC(log_num_mgm_entry_size, "log mgm size, that defines the num"
100 " of qp per mcg, for example:"
101 " 10 gives 248.range: 7 <="
102 " log_num_mgm_entry_size <= 12."
103 " To activate device managed"
104 " flow steering when available, set to -1");
105
106 static bool enable_64b_cqe_eqe = true;
107 module_param(enable_64b_cqe_eqe, bool, 0444);
108 MODULE_PARM_DESC(enable_64b_cqe_eqe,
109 "Enable 64 byte CQEs/EQEs when the FW supports this (default: True)");
110
111 static bool enable_4k_uar;
112 module_param(enable_4k_uar, bool, 0444);
113 MODULE_PARM_DESC(enable_4k_uar,
114 "Enable using 4K UAR. Should not be enabled if have VFs which do not support 4K UARs (default: false)");
115
116 #define PF_CONTEXT_BEHAVIOUR_MASK (MLX4_FUNC_CAP_64B_EQE_CQE | \
117 MLX4_FUNC_CAP_EQE_CQE_STRIDE | \
118 MLX4_FUNC_CAP_DMFS_A0_STATIC)
119
120 #define RESET_PERSIST_MASK_FLAGS (MLX4_FLAG_SRIOV)
121
122 static char mlx4_version[] =
123 DRV_NAME ": Mellanox ConnectX core driver v"
124 DRV_VERSION "\n";
125
126 static const struct mlx4_profile default_profile = {
127 .num_qp = 1 << 18,
128 .num_srq = 1 << 16,
129 .rdmarc_per_qp = 1 << 4,
130 .num_cq = 1 << 16,
131 .num_mcg = 1 << 13,
132 .num_mpt = 1 << 19,
133 .num_mtt = 1 << 20,
134 };
135
136 static const struct mlx4_profile low_mem_profile = {
137 .num_qp = 1 << 17,
138 .num_srq = 1 << 6,
139 .rdmarc_per_qp = 1 << 4,
140 .num_cq = 1 << 8,
141 .num_mcg = 1 << 8,
142 .num_mpt = 1 << 9,
143 .num_mtt = 1 << 7,
144 };
145
146 static int log_num_mac = 7;
147 module_param_named(log_num_mac, log_num_mac, int, 0444);
148 MODULE_PARM_DESC(log_num_mac, "Log2 max number of MACs per ETH port (1-7)");
149
150 static int log_num_vlan;
151 module_param_named(log_num_vlan, log_num_vlan, int, 0444);
152 MODULE_PARM_DESC(log_num_vlan, "Log2 max number of VLANs per ETH port (0-7)");
153
154 #define MLX4_LOG_NUM_VLANS 7
155 #define MLX4_MIN_LOG_NUM_VLANS 0
156 #define MLX4_MIN_LOG_NUM_MAC 1
157
158 static bool use_prio;
159 module_param_named(use_prio, use_prio, bool, 0444);
160 MODULE_PARM_DESC(use_prio, "Enable steering by VLAN priority on ETH ports (deprecated)");
161
162 int log_mtts_per_seg = ilog2(1);
163 module_param_named(log_mtts_per_seg, log_mtts_per_seg, int, 0444);
164 MODULE_PARM_DESC(log_mtts_per_seg, "Log2 number of MTT entries per segment "
165 "(0-7) (default: 0)");
166
167 static int port_type_array[2] = {MLX4_PORT_TYPE_NONE, MLX4_PORT_TYPE_NONE};
168 static int arr_argc = 2;
169 module_param_array(port_type_array, int, &arr_argc, 0444);
170 MODULE_PARM_DESC(port_type_array, "Array of port types: HW_DEFAULT (0) is default "
171 "1 for IB, 2 for Ethernet");
172
173 struct mlx4_port_config {
174 struct list_head list;
175 enum mlx4_port_type port_type[MLX4_MAX_PORTS + 1];
176 struct pci_dev *pdev;
177 };
178
179 static atomic_t pf_loading = ATOMIC_INIT(0);
180
181 static int mlx4_devlink_ierr_reset_get(struct devlink *devlink, u32 id,
182 struct devlink_param_gset_ctx *ctx)
183 {
184 ctx->val.vbool = !!mlx4_internal_err_reset;
185 return 0;
186 }
187
188 static int mlx4_devlink_ierr_reset_set(struct devlink *devlink, u32 id,
189 struct devlink_param_gset_ctx *ctx)
190 {
191 mlx4_internal_err_reset = ctx->val.vbool;
192 return 0;
193 }
194
195 static int mlx4_devlink_crdump_snapshot_get(struct devlink *devlink, u32 id,
196 struct devlink_param_gset_ctx *ctx)
197 {
198 struct mlx4_priv *priv = devlink_priv(devlink);
199 struct mlx4_dev *dev = &priv->dev;
200
201 ctx->val.vbool = dev->persist->crdump.snapshot_enable;
202 return 0;
203 }
204
205 static int mlx4_devlink_crdump_snapshot_set(struct devlink *devlink, u32 id,
206 struct devlink_param_gset_ctx *ctx)
207 {
208 struct mlx4_priv *priv = devlink_priv(devlink);
209 struct mlx4_dev *dev = &priv->dev;
210
211 dev->persist->crdump.snapshot_enable = ctx->val.vbool;
212 return 0;
213 }
214
215 static int
216 mlx4_devlink_max_macs_validate(struct devlink *devlink, u32 id,
217 union devlink_param_value val,
218 struct netlink_ext_ack *extack)
219 {
220 u32 value = val.vu32;
221
222 if (value < 1 || value > 128)
223 return -ERANGE;
224
225 if (!is_power_of_2(value)) {
226 NL_SET_ERR_MSG_MOD(extack, "max_macs supported must be power of 2");
227 return -EINVAL;
228 }
229
230 return 0;
231 }
232
233 enum mlx4_devlink_param_id {
234 MLX4_DEVLINK_PARAM_ID_BASE = DEVLINK_PARAM_GENERIC_ID_MAX,
235 MLX4_DEVLINK_PARAM_ID_ENABLE_64B_CQE_EQE,
236 MLX4_DEVLINK_PARAM_ID_ENABLE_4K_UAR,
237 };
238
239 static const struct devlink_param mlx4_devlink_params[] = {
240 DEVLINK_PARAM_GENERIC(INT_ERR_RESET,
241 BIT(DEVLINK_PARAM_CMODE_RUNTIME) |
242 BIT(DEVLINK_PARAM_CMODE_DRIVERINIT),
243 mlx4_devlink_ierr_reset_get,
244 mlx4_devlink_ierr_reset_set, NULL),
245 DEVLINK_PARAM_GENERIC(MAX_MACS,
246 BIT(DEVLINK_PARAM_CMODE_DRIVERINIT),
247 NULL, NULL, mlx4_devlink_max_macs_validate),
248 DEVLINK_PARAM_GENERIC(REGION_SNAPSHOT,
249 BIT(DEVLINK_PARAM_CMODE_RUNTIME) |
250 BIT(DEVLINK_PARAM_CMODE_DRIVERINIT),
251 mlx4_devlink_crdump_snapshot_get,
252 mlx4_devlink_crdump_snapshot_set, NULL),
253 DEVLINK_PARAM_DRIVER(MLX4_DEVLINK_PARAM_ID_ENABLE_64B_CQE_EQE,
254 "enable_64b_cqe_eqe", DEVLINK_PARAM_TYPE_BOOL,
255 BIT(DEVLINK_PARAM_CMODE_DRIVERINIT),
256 NULL, NULL, NULL),
257 DEVLINK_PARAM_DRIVER(MLX4_DEVLINK_PARAM_ID_ENABLE_4K_UAR,
258 "enable_4k_uar", DEVLINK_PARAM_TYPE_BOOL,
259 BIT(DEVLINK_PARAM_CMODE_DRIVERINIT),
260 NULL, NULL, NULL),
261 };
262
263 static void mlx4_devlink_set_params_init_values(struct devlink *devlink)
264 {
265 union devlink_param_value value;
266
267 value.vbool = !!mlx4_internal_err_reset;
268 devlink_param_driverinit_value_set(devlink,
269 DEVLINK_PARAM_GENERIC_ID_INT_ERR_RESET,
270 value);
271
272 value.vu32 = 1UL << log_num_mac;
273 devlink_param_driverinit_value_set(devlink,
274 DEVLINK_PARAM_GENERIC_ID_MAX_MACS,
275 value);
276
277 value.vbool = enable_64b_cqe_eqe;
278 devlink_param_driverinit_value_set(devlink,
279 MLX4_DEVLINK_PARAM_ID_ENABLE_64B_CQE_EQE,
280 value);
281
282 value.vbool = enable_4k_uar;
283 devlink_param_driverinit_value_set(devlink,
284 MLX4_DEVLINK_PARAM_ID_ENABLE_4K_UAR,
285 value);
286
287 value.vbool = false;
288 devlink_param_driverinit_value_set(devlink,
289 DEVLINK_PARAM_GENERIC_ID_REGION_SNAPSHOT,
290 value);
291 }
292
293 static inline void mlx4_set_num_reserved_uars(struct mlx4_dev *dev,
294 struct mlx4_dev_cap *dev_cap)
295 {
296
297
298
299
300 dev->caps.reserved_uars =
301 max_t(int,
302 mlx4_get_num_reserved_uar(dev),
303 dev_cap->reserved_uars /
304 (1 << (PAGE_SHIFT - dev->uar_page_shift)));
305 }
306
307 int mlx4_check_port_params(struct mlx4_dev *dev,
308 enum mlx4_port_type *port_type)
309 {
310 int i;
311
312 if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP)) {
313 for (i = 0; i < dev->caps.num_ports - 1; i++) {
314 if (port_type[i] != port_type[i + 1]) {
315 mlx4_err(dev, "Only same port types supported on this HCA, aborting\n");
316 return -EOPNOTSUPP;
317 }
318 }
319 }
320
321 for (i = 0; i < dev->caps.num_ports; i++) {
322 if (!(port_type[i] & dev->caps.supported_type[i+1])) {
323 mlx4_err(dev, "Requested port type for port %d is not supported on this HCA\n",
324 i + 1);
325 return -EOPNOTSUPP;
326 }
327 }
328 return 0;
329 }
330
331 static void mlx4_set_port_mask(struct mlx4_dev *dev)
332 {
333 int i;
334
335 for (i = 1; i <= dev->caps.num_ports; ++i)
336 dev->caps.port_mask[i] = dev->caps.port_type[i];
337 }
338
339 enum {
340 MLX4_QUERY_FUNC_NUM_SYS_EQS = 1 << 0,
341 };
342
343 static int mlx4_query_func(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
344 {
345 int err = 0;
346 struct mlx4_func func;
347
348 if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS) {
349 err = mlx4_QUERY_FUNC(dev, &func, 0);
350 if (err) {
351 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
352 return err;
353 }
354 dev_cap->max_eqs = func.max_eq;
355 dev_cap->reserved_eqs = func.rsvd_eqs;
356 dev_cap->reserved_uars = func.rsvd_uars;
357 err |= MLX4_QUERY_FUNC_NUM_SYS_EQS;
358 }
359 return err;
360 }
361
362 static void mlx4_enable_cqe_eqe_stride(struct mlx4_dev *dev)
363 {
364 struct mlx4_caps *dev_cap = &dev->caps;
365
366
367 if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_EQE_STRIDE) ||
368 !(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_CQE_STRIDE))
369 return;
370
371
372
373
374 if (!(dev_cap->flags & MLX4_DEV_CAP_FLAG_64B_EQE) ||
375 !(dev_cap->flags & MLX4_DEV_CAP_FLAG_64B_CQE)) {
376 dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
377 dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
378 return;
379 }
380
381 if (cache_line_size() == 128 || cache_line_size() == 256) {
382 mlx4_dbg(dev, "Enabling CQE stride cacheLine supported\n");
383
384 dev_cap->flags &= ~MLX4_DEV_CAP_FLAG_64B_CQE;
385 dev_cap->flags &= ~MLX4_DEV_CAP_FLAG_64B_EQE;
386
387 if (mlx4_is_master(dev))
388 dev_cap->function_caps |= MLX4_FUNC_CAP_EQE_CQE_STRIDE;
389 } else {
390 if (cache_line_size() != 32 && cache_line_size() != 64)
391 mlx4_dbg(dev, "Disabling CQE stride, cacheLine size unsupported\n");
392 dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
393 dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
394 }
395 }
396
397 static int _mlx4_dev_port(struct mlx4_dev *dev, int port,
398 struct mlx4_port_cap *port_cap)
399 {
400 dev->caps.vl_cap[port] = port_cap->max_vl;
401 dev->caps.ib_mtu_cap[port] = port_cap->ib_mtu;
402 dev->phys_caps.gid_phys_table_len[port] = port_cap->max_gids;
403 dev->phys_caps.pkey_phys_table_len[port] = port_cap->max_pkeys;
404
405
406
407 dev->caps.gid_table_len[port] = port_cap->max_gids;
408 dev->caps.pkey_table_len[port] = port_cap->max_pkeys;
409 dev->caps.port_width_cap[port] = port_cap->max_port_width;
410 dev->caps.eth_mtu_cap[port] = port_cap->eth_mtu;
411 dev->caps.max_tc_eth = port_cap->max_tc_eth;
412 dev->caps.def_mac[port] = port_cap->def_mac;
413 dev->caps.supported_type[port] = port_cap->supported_port_types;
414 dev->caps.suggested_type[port] = port_cap->suggested_type;
415 dev->caps.default_sense[port] = port_cap->default_sense;
416 dev->caps.trans_type[port] = port_cap->trans_type;
417 dev->caps.vendor_oui[port] = port_cap->vendor_oui;
418 dev->caps.wavelength[port] = port_cap->wavelength;
419 dev->caps.trans_code[port] = port_cap->trans_code;
420
421 return 0;
422 }
423
424 static int mlx4_dev_port(struct mlx4_dev *dev, int port,
425 struct mlx4_port_cap *port_cap)
426 {
427 int err = 0;
428
429 err = mlx4_QUERY_PORT(dev, port, port_cap);
430
431 if (err)
432 mlx4_err(dev, "QUERY_PORT command failed.\n");
433
434 return err;
435 }
436
437 static inline void mlx4_enable_ignore_fcs(struct mlx4_dev *dev)
438 {
439 if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_IGNORE_FCS))
440 return;
441
442 if (mlx4_is_mfunc(dev)) {
443 mlx4_dbg(dev, "SRIOV mode - Disabling Ignore FCS");
444 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_IGNORE_FCS;
445 return;
446 }
447
448 if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_FCS_KEEP)) {
449 mlx4_dbg(dev,
450 "Keep FCS is not supported - Disabling Ignore FCS");
451 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_IGNORE_FCS;
452 return;
453 }
454 }
455
456 #define MLX4_A0_STEERING_TABLE_SIZE 256
457 static int mlx4_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
458 {
459 int err;
460 int i;
461
462 err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
463 if (err) {
464 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
465 return err;
466 }
467 mlx4_dev_cap_dump(dev, dev_cap);
468
469 if (dev_cap->min_page_sz > PAGE_SIZE) {
470 mlx4_err(dev, "HCA minimum page size of %d bigger than kernel PAGE_SIZE of %ld, aborting\n",
471 dev_cap->min_page_sz, PAGE_SIZE);
472 return -ENODEV;
473 }
474 if (dev_cap->num_ports > MLX4_MAX_PORTS) {
475 mlx4_err(dev, "HCA has %d ports, but we only support %d, aborting\n",
476 dev_cap->num_ports, MLX4_MAX_PORTS);
477 return -ENODEV;
478 }
479
480 if (dev_cap->uar_size > pci_resource_len(dev->persist->pdev, 2)) {
481 mlx4_err(dev, "HCA reported UAR size of 0x%x bigger than PCI resource 2 size of 0x%llx, aborting\n",
482 dev_cap->uar_size,
483 (unsigned long long)
484 pci_resource_len(dev->persist->pdev, 2));
485 return -ENODEV;
486 }
487
488 dev->caps.num_ports = dev_cap->num_ports;
489 dev->caps.num_sys_eqs = dev_cap->num_sys_eqs;
490 dev->phys_caps.num_phys_eqs = dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS ?
491 dev->caps.num_sys_eqs :
492 MLX4_MAX_EQ_NUM;
493 for (i = 1; i <= dev->caps.num_ports; ++i) {
494 err = _mlx4_dev_port(dev, i, dev_cap->port_cap + i);
495 if (err) {
496 mlx4_err(dev, "QUERY_PORT command failed, aborting\n");
497 return err;
498 }
499 }
500
501 dev->caps.uar_page_size = PAGE_SIZE;
502 dev->caps.num_uars = dev_cap->uar_size / PAGE_SIZE;
503 dev->caps.local_ca_ack_delay = dev_cap->local_ca_ack_delay;
504 dev->caps.bf_reg_size = dev_cap->bf_reg_size;
505 dev->caps.bf_regs_per_page = dev_cap->bf_regs_per_page;
506 dev->caps.max_sq_sg = dev_cap->max_sq_sg;
507 dev->caps.max_rq_sg = dev_cap->max_rq_sg;
508 dev->caps.max_wqes = dev_cap->max_qp_sz;
509 dev->caps.max_qp_init_rdma = dev_cap->max_requester_per_qp;
510 dev->caps.max_srq_wqes = dev_cap->max_srq_sz;
511 dev->caps.max_srq_sge = dev_cap->max_rq_sg - 1;
512 dev->caps.reserved_srqs = dev_cap->reserved_srqs;
513 dev->caps.max_sq_desc_sz = dev_cap->max_sq_desc_sz;
514 dev->caps.max_rq_desc_sz = dev_cap->max_rq_desc_sz;
515
516
517
518
519 dev->caps.max_cqes = dev_cap->max_cq_sz - 1;
520 dev->caps.reserved_cqs = dev_cap->reserved_cqs;
521 dev->caps.reserved_eqs = dev_cap->reserved_eqs;
522 dev->caps.reserved_mtts = dev_cap->reserved_mtts;
523 dev->caps.reserved_mrws = dev_cap->reserved_mrws;
524
525 dev->caps.reserved_pds = dev_cap->reserved_pds;
526 dev->caps.reserved_xrcds = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
527 dev_cap->reserved_xrcds : 0;
528 dev->caps.max_xrcds = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
529 dev_cap->max_xrcds : 0;
530 dev->caps.mtt_entry_sz = dev_cap->mtt_entry_sz;
531
532 dev->caps.max_msg_sz = dev_cap->max_msg_sz;
533 dev->caps.page_size_cap = ~(u32) (dev_cap->min_page_sz - 1);
534 dev->caps.flags = dev_cap->flags;
535 dev->caps.flags2 = dev_cap->flags2;
536 dev->caps.bmme_flags = dev_cap->bmme_flags;
537 dev->caps.reserved_lkey = dev_cap->reserved_lkey;
538 dev->caps.stat_rate_support = dev_cap->stat_rate_support;
539 dev->caps.max_gso_sz = dev_cap->max_gso_sz;
540 dev->caps.max_rss_tbl_sz = dev_cap->max_rss_tbl_sz;
541 dev->caps.wol_port[1] = dev_cap->wol_port[1];
542 dev->caps.wol_port[2] = dev_cap->wol_port[2];
543 dev->caps.health_buffer_addrs = dev_cap->health_buffer_addrs;
544
545
546 if (!mlx4_is_slave(dev)) {
547
548
549
550 if (enable_4k_uar || !dev->persist->num_vfs)
551 dev->uar_page_shift = DEFAULT_UAR_PAGE_SHIFT;
552 else
553 dev->uar_page_shift = PAGE_SHIFT;
554
555 mlx4_set_num_reserved_uars(dev, dev_cap);
556 }
557
558 if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_PHV_EN) {
559 struct mlx4_init_hca_param hca_param;
560
561 memset(&hca_param, 0, sizeof(hca_param));
562 err = mlx4_QUERY_HCA(dev, &hca_param);
563
564
565
566
567
568
569 if (err || hca_param.phv_check_en)
570 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_PHV_EN;
571 }
572
573
574 if (mlx4_priv(dev)->pci_dev_data & MLX4_PCI_DEV_FORCE_SENSE_PORT)
575 dev->caps.flags |= MLX4_DEV_CAP_FLAG_SENSE_SUPPORT;
576
577 if (mlx4_is_mfunc(dev))
578 dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_SENSE_SUPPORT;
579
580 if (mlx4_low_memory_profile()) {
581 dev->caps.log_num_macs = MLX4_MIN_LOG_NUM_MAC;
582 dev->caps.log_num_vlans = MLX4_MIN_LOG_NUM_VLANS;
583 } else {
584 dev->caps.log_num_macs = log_num_mac;
585 dev->caps.log_num_vlans = MLX4_LOG_NUM_VLANS;
586 }
587
588 for (i = 1; i <= dev->caps.num_ports; ++i) {
589 dev->caps.port_type[i] = MLX4_PORT_TYPE_NONE;
590 if (dev->caps.supported_type[i]) {
591
592 if (dev->caps.supported_type[i] == MLX4_PORT_TYPE_ETH)
593 dev->caps.port_type[i] = MLX4_PORT_TYPE_ETH;
594
595 else if (dev->caps.supported_type[i] ==
596 MLX4_PORT_TYPE_IB)
597 dev->caps.port_type[i] = MLX4_PORT_TYPE_IB;
598 else {
599
600
601
602 if (port_type_array[i - 1] == MLX4_PORT_TYPE_NONE)
603 dev->caps.port_type[i] = dev->caps.suggested_type[i] ?
604 MLX4_PORT_TYPE_ETH : MLX4_PORT_TYPE_IB;
605 else
606 dev->caps.port_type[i] = port_type_array[i - 1];
607 }
608 }
609
610
611
612
613
614
615 mlx4_priv(dev)->sense.sense_allowed[i] =
616 ((dev->caps.supported_type[i] == MLX4_PORT_TYPE_AUTO) &&
617 (dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
618 (dev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT));
619
620
621
622
623
624
625 if (mlx4_priv(dev)->sense.sense_allowed[i] && dev->caps.default_sense[i]) {
626 enum mlx4_port_type sensed_port = MLX4_PORT_TYPE_NONE;
627 dev->caps.possible_type[i] = MLX4_PORT_TYPE_AUTO;
628 mlx4_SENSE_PORT(dev, i, &sensed_port);
629 if (sensed_port != MLX4_PORT_TYPE_NONE)
630 dev->caps.port_type[i] = sensed_port;
631 } else {
632 dev->caps.possible_type[i] = dev->caps.port_type[i];
633 }
634
635 if (dev->caps.log_num_macs > dev_cap->port_cap[i].log_max_macs) {
636 dev->caps.log_num_macs = dev_cap->port_cap[i].log_max_macs;
637 mlx4_warn(dev, "Requested number of MACs is too much for port %d, reducing to %d\n",
638 i, 1 << dev->caps.log_num_macs);
639 }
640 if (dev->caps.log_num_vlans > dev_cap->port_cap[i].log_max_vlans) {
641 dev->caps.log_num_vlans = dev_cap->port_cap[i].log_max_vlans;
642 mlx4_warn(dev, "Requested number of VLANs is too much for port %d, reducing to %d\n",
643 i, 1 << dev->caps.log_num_vlans);
644 }
645 }
646
647 if (mlx4_is_master(dev) && (dev->caps.num_ports == 2) &&
648 (port_type_array[0] == MLX4_PORT_TYPE_IB) &&
649 (port_type_array[1] == MLX4_PORT_TYPE_ETH)) {
650 mlx4_warn(dev,
651 "Granular QoS per VF not supported with IB/Eth configuration\n");
652 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_QOS_VPP;
653 }
654
655 dev->caps.max_counters = dev_cap->max_counters;
656
657 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] = dev_cap->reserved_qps;
658 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] =
659 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] =
660 (1 << dev->caps.log_num_macs) *
661 (1 << dev->caps.log_num_vlans) *
662 dev->caps.num_ports;
663 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH] = MLX4_NUM_FEXCH;
664
665 if (dev_cap->dmfs_high_rate_qpn_base > 0 &&
666 dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN)
667 dev->caps.dmfs_high_rate_qpn_base = dev_cap->dmfs_high_rate_qpn_base;
668 else
669 dev->caps.dmfs_high_rate_qpn_base =
670 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
671
672 if (dev_cap->dmfs_high_rate_qpn_range > 0 &&
673 dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN) {
674 dev->caps.dmfs_high_rate_qpn_range = dev_cap->dmfs_high_rate_qpn_range;
675 dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_DEFAULT;
676 dev->caps.flags2 |= MLX4_DEV_CAP_FLAG2_FS_A0;
677 } else {
678 dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_NOT_SUPPORTED;
679 dev->caps.dmfs_high_rate_qpn_base =
680 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
681 dev->caps.dmfs_high_rate_qpn_range = MLX4_A0_STEERING_TABLE_SIZE;
682 }
683
684 dev->caps.rl_caps = dev_cap->rl_caps;
685
686 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_RSS_RAW_ETH] =
687 dev->caps.dmfs_high_rate_qpn_range;
688
689 dev->caps.reserved_qps = dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] +
690 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] +
691 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] +
692 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH];
693
694 dev->caps.sqp_demux = (mlx4_is_master(dev)) ? MLX4_MAX_NUM_SLAVES : 0;
695
696 if (!enable_64b_cqe_eqe && !mlx4_is_slave(dev)) {
697 if (dev_cap->flags &
698 (MLX4_DEV_CAP_FLAG_64B_CQE | MLX4_DEV_CAP_FLAG_64B_EQE)) {
699 mlx4_warn(dev, "64B EQEs/CQEs supported by the device but not enabled\n");
700 dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_64B_CQE;
701 dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_64B_EQE;
702 }
703
704 if (dev_cap->flags2 &
705 (MLX4_DEV_CAP_FLAG2_CQE_STRIDE |
706 MLX4_DEV_CAP_FLAG2_EQE_STRIDE)) {
707 mlx4_warn(dev, "Disabling EQE/CQE stride per user request\n");
708 dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
709 dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
710 }
711 }
712
713 if ((dev->caps.flags &
714 (MLX4_DEV_CAP_FLAG_64B_CQE | MLX4_DEV_CAP_FLAG_64B_EQE)) &&
715 mlx4_is_master(dev))
716 dev->caps.function_caps |= MLX4_FUNC_CAP_64B_EQE_CQE;
717
718 if (!mlx4_is_slave(dev)) {
719 mlx4_enable_cqe_eqe_stride(dev);
720 dev->caps.alloc_res_qp_mask =
721 (dev->caps.bf_reg_size ? MLX4_RESERVE_ETH_BF_QP : 0) |
722 MLX4_RESERVE_A0_QP;
723
724 if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ETS_CFG) &&
725 dev->caps.flags & MLX4_DEV_CAP_FLAG_SET_ETH_SCHED) {
726 mlx4_warn(dev, "Old device ETS support detected\n");
727 mlx4_warn(dev, "Consider upgrading device FW.\n");
728 dev->caps.flags2 |= MLX4_DEV_CAP_FLAG2_ETS_CFG;
729 }
730
731 } else {
732 dev->caps.alloc_res_qp_mask = 0;
733 }
734
735 mlx4_enable_ignore_fcs(dev);
736
737 return 0;
738 }
739
740
741 static int mlx4_how_many_lives_vf(struct mlx4_dev *dev)
742 {
743 struct mlx4_priv *priv = mlx4_priv(dev);
744 struct mlx4_slave_state *s_state;
745 int i;
746 int ret = 0;
747
748 for (i = 1; i < dev->num_slaves; ++i) {
749 s_state = &priv->mfunc.master.slave_state[i];
750 if (s_state->active && s_state->last_cmd !=
751 MLX4_COMM_CMD_RESET) {
752 mlx4_warn(dev, "%s: slave: %d is still active\n",
753 __func__, i);
754 ret++;
755 }
756 }
757 return ret;
758 }
759
760 int mlx4_get_parav_qkey(struct mlx4_dev *dev, u32 qpn, u32 *qkey)
761 {
762 u32 qk = MLX4_RESERVED_QKEY_BASE;
763
764 if (qpn >= dev->phys_caps.base_tunnel_sqpn + 8 * MLX4_MFUNC_MAX ||
765 qpn < dev->phys_caps.base_proxy_sqpn)
766 return -EINVAL;
767
768 if (qpn >= dev->phys_caps.base_tunnel_sqpn)
769
770 qk += qpn - dev->phys_caps.base_tunnel_sqpn;
771 else
772 qk += qpn - dev->phys_caps.base_proxy_sqpn;
773 *qkey = qk;
774 return 0;
775 }
776 EXPORT_SYMBOL(mlx4_get_parav_qkey);
777
778 void mlx4_sync_pkey_table(struct mlx4_dev *dev, int slave, int port, int i, int val)
779 {
780 struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
781
782 if (!mlx4_is_master(dev))
783 return;
784
785 priv->virt2phys_pkey[slave][port - 1][i] = val;
786 }
787 EXPORT_SYMBOL(mlx4_sync_pkey_table);
788
789 void mlx4_put_slave_node_guid(struct mlx4_dev *dev, int slave, __be64 guid)
790 {
791 struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
792
793 if (!mlx4_is_master(dev))
794 return;
795
796 priv->slave_node_guids[slave] = guid;
797 }
798 EXPORT_SYMBOL(mlx4_put_slave_node_guid);
799
800 __be64 mlx4_get_slave_node_guid(struct mlx4_dev *dev, int slave)
801 {
802 struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
803
804 if (!mlx4_is_master(dev))
805 return 0;
806
807 return priv->slave_node_guids[slave];
808 }
809 EXPORT_SYMBOL(mlx4_get_slave_node_guid);
810
811 int mlx4_is_slave_active(struct mlx4_dev *dev, int slave)
812 {
813 struct mlx4_priv *priv = mlx4_priv(dev);
814 struct mlx4_slave_state *s_slave;
815
816 if (!mlx4_is_master(dev))
817 return 0;
818
819 s_slave = &priv->mfunc.master.slave_state[slave];
820 return !!s_slave->active;
821 }
822 EXPORT_SYMBOL(mlx4_is_slave_active);
823
824 void mlx4_handle_eth_header_mcast_prio(struct mlx4_net_trans_rule_hw_ctrl *ctrl,
825 struct _rule_hw *eth_header)
826 {
827 if (is_multicast_ether_addr(eth_header->eth.dst_mac) ||
828 is_broadcast_ether_addr(eth_header->eth.dst_mac)) {
829 struct mlx4_net_trans_rule_hw_eth *eth =
830 (struct mlx4_net_trans_rule_hw_eth *)eth_header;
831 struct _rule_hw *next_rule = (struct _rule_hw *)(eth + 1);
832 bool last_rule = next_rule->size == 0 && next_rule->id == 0 &&
833 next_rule->rsvd == 0;
834
835 if (last_rule)
836 ctrl->prio = cpu_to_be16(MLX4_DOMAIN_NIC);
837 }
838 }
839 EXPORT_SYMBOL(mlx4_handle_eth_header_mcast_prio);
840
841 static void slave_adjust_steering_mode(struct mlx4_dev *dev,
842 struct mlx4_dev_cap *dev_cap,
843 struct mlx4_init_hca_param *hca_param)
844 {
845 dev->caps.steering_mode = hca_param->steering_mode;
846 if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED) {
847 dev->caps.num_qp_per_mgm = dev_cap->fs_max_num_qp_per_entry;
848 dev->caps.fs_log_max_ucast_qp_range_size =
849 dev_cap->fs_log_max_ucast_qp_range_size;
850 } else
851 dev->caps.num_qp_per_mgm =
852 4 * ((1 << hca_param->log_mc_entry_sz)/16 - 2);
853
854 mlx4_dbg(dev, "Steering mode is: %s\n",
855 mlx4_steering_mode_str(dev->caps.steering_mode));
856 }
857
858 static void mlx4_slave_destroy_special_qp_cap(struct mlx4_dev *dev)
859 {
860 kfree(dev->caps.spec_qps);
861 dev->caps.spec_qps = NULL;
862 }
863
864 static int mlx4_slave_special_qp_cap(struct mlx4_dev *dev)
865 {
866 struct mlx4_func_cap *func_cap = NULL;
867 struct mlx4_caps *caps = &dev->caps;
868 int i, err = 0;
869
870 func_cap = kzalloc(sizeof(*func_cap), GFP_KERNEL);
871 caps->spec_qps = kcalloc(caps->num_ports, sizeof(*caps->spec_qps), GFP_KERNEL);
872
873 if (!func_cap || !caps->spec_qps) {
874 mlx4_err(dev, "Failed to allocate memory for special qps cap\n");
875 err = -ENOMEM;
876 goto err_mem;
877 }
878
879 for (i = 1; i <= caps->num_ports; ++i) {
880 err = mlx4_QUERY_FUNC_CAP(dev, i, func_cap);
881 if (err) {
882 mlx4_err(dev, "QUERY_FUNC_CAP port command failed for port %d, aborting (%d)\n",
883 i, err);
884 goto err_mem;
885 }
886 caps->spec_qps[i - 1] = func_cap->spec_qps;
887 caps->port_mask[i] = caps->port_type[i];
888 caps->phys_port_id[i] = func_cap->phys_port_id;
889 err = mlx4_get_slave_pkey_gid_tbl_len(dev, i,
890 &caps->gid_table_len[i],
891 &caps->pkey_table_len[i]);
892 if (err) {
893 mlx4_err(dev, "QUERY_PORT command failed for port %d, aborting (%d)\n",
894 i, err);
895 goto err_mem;
896 }
897 }
898
899 err_mem:
900 if (err)
901 mlx4_slave_destroy_special_qp_cap(dev);
902 kfree(func_cap);
903 return err;
904 }
905
906 static int mlx4_slave_cap(struct mlx4_dev *dev)
907 {
908 int err;
909 u32 page_size;
910 struct mlx4_dev_cap *dev_cap = NULL;
911 struct mlx4_func_cap *func_cap = NULL;
912 struct mlx4_init_hca_param *hca_param = NULL;
913
914 hca_param = kzalloc(sizeof(*hca_param), GFP_KERNEL);
915 func_cap = kzalloc(sizeof(*func_cap), GFP_KERNEL);
916 dev_cap = kzalloc(sizeof(*dev_cap), GFP_KERNEL);
917 if (!hca_param || !func_cap || !dev_cap) {
918 mlx4_err(dev, "Failed to allocate memory for slave_cap\n");
919 err = -ENOMEM;
920 goto free_mem;
921 }
922
923 err = mlx4_QUERY_HCA(dev, hca_param);
924 if (err) {
925 mlx4_err(dev, "QUERY_HCA command failed, aborting\n");
926 goto free_mem;
927 }
928
929
930
931
932 if (hca_param->global_caps) {
933 mlx4_err(dev, "Unknown hca global capabilities\n");
934 err = -EINVAL;
935 goto free_mem;
936 }
937
938 dev->caps.hca_core_clock = hca_param->hca_core_clock;
939
940 dev->caps.max_qp_dest_rdma = 1 << hca_param->log_rd_per_qp;
941 err = mlx4_dev_cap(dev, dev_cap);
942 if (err) {
943 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
944 goto free_mem;
945 }
946
947 err = mlx4_QUERY_FW(dev);
948 if (err)
949 mlx4_err(dev, "QUERY_FW command failed: could not get FW version\n");
950
951 page_size = ~dev->caps.page_size_cap + 1;
952 mlx4_warn(dev, "HCA minimum page size:%d\n", page_size);
953 if (page_size > PAGE_SIZE) {
954 mlx4_err(dev, "HCA minimum page size of %d bigger than kernel PAGE_SIZE of %ld, aborting\n",
955 page_size, PAGE_SIZE);
956 err = -ENODEV;
957 goto free_mem;
958 }
959
960
961 dev->uar_page_shift = hca_param->uar_page_sz + 12;
962
963
964 if (dev->uar_page_shift > PAGE_SHIFT) {
965 mlx4_err(dev,
966 "Invalid configuration: uar page size is larger than system page size\n");
967 err = -ENODEV;
968 goto free_mem;
969 }
970
971
972 mlx4_set_num_reserved_uars(dev, dev_cap);
973
974
975
976
977
978 dev->caps.uar_page_size = PAGE_SIZE;
979
980 err = mlx4_QUERY_FUNC_CAP(dev, 0, func_cap);
981 if (err) {
982 mlx4_err(dev, "QUERY_FUNC_CAP general command failed, aborting (%d)\n",
983 err);
984 goto free_mem;
985 }
986
987 if ((func_cap->pf_context_behaviour | PF_CONTEXT_BEHAVIOUR_MASK) !=
988 PF_CONTEXT_BEHAVIOUR_MASK) {
989 mlx4_err(dev, "Unknown pf context behaviour %x known flags %x\n",
990 func_cap->pf_context_behaviour,
991 PF_CONTEXT_BEHAVIOUR_MASK);
992 err = -EINVAL;
993 goto free_mem;
994 }
995
996 dev->caps.num_ports = func_cap->num_ports;
997 dev->quotas.qp = func_cap->qp_quota;
998 dev->quotas.srq = func_cap->srq_quota;
999 dev->quotas.cq = func_cap->cq_quota;
1000 dev->quotas.mpt = func_cap->mpt_quota;
1001 dev->quotas.mtt = func_cap->mtt_quota;
1002 dev->caps.num_qps = 1 << hca_param->log_num_qps;
1003 dev->caps.num_srqs = 1 << hca_param->log_num_srqs;
1004 dev->caps.num_cqs = 1 << hca_param->log_num_cqs;
1005 dev->caps.num_mpts = 1 << hca_param->log_mpt_sz;
1006 dev->caps.num_eqs = func_cap->max_eq;
1007 dev->caps.reserved_eqs = func_cap->reserved_eq;
1008 dev->caps.reserved_lkey = func_cap->reserved_lkey;
1009 dev->caps.num_pds = MLX4_NUM_PDS;
1010 dev->caps.num_mgms = 0;
1011 dev->caps.num_amgms = 0;
1012
1013 if (dev->caps.num_ports > MLX4_MAX_PORTS) {
1014 mlx4_err(dev, "HCA has %d ports, but we only support %d, aborting\n",
1015 dev->caps.num_ports, MLX4_MAX_PORTS);
1016 err = -ENODEV;
1017 goto free_mem;
1018 }
1019
1020 mlx4_replace_zero_macs(dev);
1021
1022 err = mlx4_slave_special_qp_cap(dev);
1023 if (err) {
1024 mlx4_err(dev, "Set special QP caps failed. aborting\n");
1025 goto free_mem;
1026 }
1027
1028 if (dev->caps.uar_page_size * (dev->caps.num_uars -
1029 dev->caps.reserved_uars) >
1030 pci_resource_len(dev->persist->pdev,
1031 2)) {
1032 mlx4_err(dev, "HCA reported UAR region size of 0x%x bigger than PCI resource 2 size of 0x%llx, aborting\n",
1033 dev->caps.uar_page_size * dev->caps.num_uars,
1034 (unsigned long long)
1035 pci_resource_len(dev->persist->pdev, 2));
1036 err = -ENOMEM;
1037 goto err_mem;
1038 }
1039
1040 if (hca_param->dev_cap_enabled & MLX4_DEV_CAP_64B_EQE_ENABLED) {
1041 dev->caps.eqe_size = 64;
1042 dev->caps.eqe_factor = 1;
1043 } else {
1044 dev->caps.eqe_size = 32;
1045 dev->caps.eqe_factor = 0;
1046 }
1047
1048 if (hca_param->dev_cap_enabled & MLX4_DEV_CAP_64B_CQE_ENABLED) {
1049 dev->caps.cqe_size = 64;
1050 dev->caps.userspace_caps |= MLX4_USER_DEV_CAP_LARGE_CQE;
1051 } else {
1052 dev->caps.cqe_size = 32;
1053 }
1054
1055 if (hca_param->dev_cap_enabled & MLX4_DEV_CAP_EQE_STRIDE_ENABLED) {
1056 dev->caps.eqe_size = hca_param->eqe_size;
1057 dev->caps.eqe_factor = 0;
1058 }
1059
1060 if (hca_param->dev_cap_enabled & MLX4_DEV_CAP_CQE_STRIDE_ENABLED) {
1061 dev->caps.cqe_size = hca_param->cqe_size;
1062
1063 dev->caps.userspace_caps |= MLX4_USER_DEV_CAP_LARGE_CQE;
1064 }
1065
1066 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
1067 mlx4_warn(dev, "Timestamping is not supported in slave mode\n");
1068
1069 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_USER_MAC_EN;
1070 mlx4_dbg(dev, "User MAC FW update is not supported in slave mode\n");
1071
1072 slave_adjust_steering_mode(dev, dev_cap, hca_param);
1073 mlx4_dbg(dev, "RSS support for IP fragments is %s\n",
1074 hca_param->rss_ip_frags ? "on" : "off");
1075
1076 if (func_cap->extra_flags & MLX4_QUERY_FUNC_FLAGS_BF_RES_QP &&
1077 dev->caps.bf_reg_size)
1078 dev->caps.alloc_res_qp_mask |= MLX4_RESERVE_ETH_BF_QP;
1079
1080 if (func_cap->extra_flags & MLX4_QUERY_FUNC_FLAGS_A0_RES_QP)
1081 dev->caps.alloc_res_qp_mask |= MLX4_RESERVE_A0_QP;
1082
1083 err_mem:
1084 if (err)
1085 mlx4_slave_destroy_special_qp_cap(dev);
1086 free_mem:
1087 kfree(hca_param);
1088 kfree(func_cap);
1089 kfree(dev_cap);
1090 return err;
1091 }
1092
1093 static void mlx4_request_modules(struct mlx4_dev *dev)
1094 {
1095 int port;
1096 int has_ib_port = false;
1097 int has_eth_port = false;
1098 #define EN_DRV_NAME "mlx4_en"
1099 #define IB_DRV_NAME "mlx4_ib"
1100
1101 for (port = 1; port <= dev->caps.num_ports; port++) {
1102 if (dev->caps.port_type[port] == MLX4_PORT_TYPE_IB)
1103 has_ib_port = true;
1104 else if (dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH)
1105 has_eth_port = true;
1106 }
1107
1108 if (has_eth_port)
1109 request_module_nowait(EN_DRV_NAME);
1110 if (has_ib_port || (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE))
1111 request_module_nowait(IB_DRV_NAME);
1112 }
1113
1114
1115
1116
1117
1118 int mlx4_change_port_types(struct mlx4_dev *dev,
1119 enum mlx4_port_type *port_types)
1120 {
1121 int err = 0;
1122 int change = 0;
1123 int port;
1124
1125 for (port = 0; port < dev->caps.num_ports; port++) {
1126
1127
1128 if (port_types[port] != dev->caps.port_type[port + 1])
1129 change = 1;
1130 }
1131 if (change) {
1132 mlx4_unregister_device(dev);
1133 for (port = 1; port <= dev->caps.num_ports; port++) {
1134 mlx4_CLOSE_PORT(dev, port);
1135 dev->caps.port_type[port] = port_types[port - 1];
1136 err = mlx4_SET_PORT(dev, port, -1);
1137 if (err) {
1138 mlx4_err(dev, "Failed to set port %d, aborting\n",
1139 port);
1140 goto out;
1141 }
1142 }
1143 mlx4_set_port_mask(dev);
1144 err = mlx4_register_device(dev);
1145 if (err) {
1146 mlx4_err(dev, "Failed to register device\n");
1147 goto out;
1148 }
1149 mlx4_request_modules(dev);
1150 }
1151
1152 out:
1153 return err;
1154 }
1155
1156 static ssize_t show_port_type(struct device *dev,
1157 struct device_attribute *attr,
1158 char *buf)
1159 {
1160 struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1161 port_attr);
1162 struct mlx4_dev *mdev = info->dev;
1163 char type[8];
1164
1165 sprintf(type, "%s",
1166 (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_IB) ?
1167 "ib" : "eth");
1168 if (mdev->caps.possible_type[info->port] == MLX4_PORT_TYPE_AUTO)
1169 sprintf(buf, "auto (%s)\n", type);
1170 else
1171 sprintf(buf, "%s\n", type);
1172
1173 return strlen(buf);
1174 }
1175
1176 static int __set_port_type(struct mlx4_port_info *info,
1177 enum mlx4_port_type port_type)
1178 {
1179 struct mlx4_dev *mdev = info->dev;
1180 struct mlx4_priv *priv = mlx4_priv(mdev);
1181 enum mlx4_port_type types[MLX4_MAX_PORTS];
1182 enum mlx4_port_type new_types[MLX4_MAX_PORTS];
1183 int i;
1184 int err = 0;
1185
1186 if ((port_type & mdev->caps.supported_type[info->port]) != port_type) {
1187 mlx4_err(mdev,
1188 "Requested port type for port %d is not supported on this HCA\n",
1189 info->port);
1190 return -EOPNOTSUPP;
1191 }
1192
1193 mlx4_stop_sense(mdev);
1194 mutex_lock(&priv->port_mutex);
1195 info->tmp_type = port_type;
1196
1197
1198 mdev->caps.possible_type[info->port] = info->tmp_type;
1199
1200 for (i = 0; i < mdev->caps.num_ports; i++) {
1201 types[i] = priv->port[i+1].tmp_type ? priv->port[i+1].tmp_type :
1202 mdev->caps.possible_type[i+1];
1203 if (types[i] == MLX4_PORT_TYPE_AUTO)
1204 types[i] = mdev->caps.port_type[i+1];
1205 }
1206
1207 if (!(mdev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
1208 !(mdev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT)) {
1209 for (i = 1; i <= mdev->caps.num_ports; i++) {
1210 if (mdev->caps.possible_type[i] == MLX4_PORT_TYPE_AUTO) {
1211 mdev->caps.possible_type[i] = mdev->caps.port_type[i];
1212 err = -EOPNOTSUPP;
1213 }
1214 }
1215 }
1216 if (err) {
1217 mlx4_err(mdev, "Auto sensing is not supported on this HCA. Set only 'eth' or 'ib' for both ports (should be the same)\n");
1218 goto out;
1219 }
1220
1221 mlx4_do_sense_ports(mdev, new_types, types);
1222
1223 err = mlx4_check_port_params(mdev, new_types);
1224 if (err)
1225 goto out;
1226
1227
1228
1229
1230 for (i = 0; i < mdev->caps.num_ports; i++)
1231 priv->port[i + 1].tmp_type = 0;
1232
1233 err = mlx4_change_port_types(mdev, new_types);
1234
1235 out:
1236 mlx4_start_sense(mdev);
1237 mutex_unlock(&priv->port_mutex);
1238
1239 return err;
1240 }
1241
1242 static ssize_t set_port_type(struct device *dev,
1243 struct device_attribute *attr,
1244 const char *buf, size_t count)
1245 {
1246 struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1247 port_attr);
1248 struct mlx4_dev *mdev = info->dev;
1249 enum mlx4_port_type port_type;
1250 static DEFINE_MUTEX(set_port_type_mutex);
1251 int err;
1252
1253 mutex_lock(&set_port_type_mutex);
1254
1255 if (!strcmp(buf, "ib\n")) {
1256 port_type = MLX4_PORT_TYPE_IB;
1257 } else if (!strcmp(buf, "eth\n")) {
1258 port_type = MLX4_PORT_TYPE_ETH;
1259 } else if (!strcmp(buf, "auto\n")) {
1260 port_type = MLX4_PORT_TYPE_AUTO;
1261 } else {
1262 mlx4_err(mdev, "%s is not supported port type\n", buf);
1263 err = -EINVAL;
1264 goto err_out;
1265 }
1266
1267 err = __set_port_type(info, port_type);
1268
1269 err_out:
1270 mutex_unlock(&set_port_type_mutex);
1271
1272 return err ? err : count;
1273 }
1274
1275 enum ibta_mtu {
1276 IB_MTU_256 = 1,
1277 IB_MTU_512 = 2,
1278 IB_MTU_1024 = 3,
1279 IB_MTU_2048 = 4,
1280 IB_MTU_4096 = 5
1281 };
1282
1283 static inline int int_to_ibta_mtu(int mtu)
1284 {
1285 switch (mtu) {
1286 case 256: return IB_MTU_256;
1287 case 512: return IB_MTU_512;
1288 case 1024: return IB_MTU_1024;
1289 case 2048: return IB_MTU_2048;
1290 case 4096: return IB_MTU_4096;
1291 default: return -1;
1292 }
1293 }
1294
1295 static inline int ibta_mtu_to_int(enum ibta_mtu mtu)
1296 {
1297 switch (mtu) {
1298 case IB_MTU_256: return 256;
1299 case IB_MTU_512: return 512;
1300 case IB_MTU_1024: return 1024;
1301 case IB_MTU_2048: return 2048;
1302 case IB_MTU_4096: return 4096;
1303 default: return -1;
1304 }
1305 }
1306
1307 static ssize_t show_port_ib_mtu(struct device *dev,
1308 struct device_attribute *attr,
1309 char *buf)
1310 {
1311 struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1312 port_mtu_attr);
1313 struct mlx4_dev *mdev = info->dev;
1314
1315 if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH)
1316 mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
1317
1318 sprintf(buf, "%d\n",
1319 ibta_mtu_to_int(mdev->caps.port_ib_mtu[info->port]));
1320 return strlen(buf);
1321 }
1322
1323 static ssize_t set_port_ib_mtu(struct device *dev,
1324 struct device_attribute *attr,
1325 const char *buf, size_t count)
1326 {
1327 struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1328 port_mtu_attr);
1329 struct mlx4_dev *mdev = info->dev;
1330 struct mlx4_priv *priv = mlx4_priv(mdev);
1331 int err, port, mtu, ibta_mtu = -1;
1332
1333 if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH) {
1334 mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
1335 return -EINVAL;
1336 }
1337
1338 err = kstrtoint(buf, 0, &mtu);
1339 if (!err)
1340 ibta_mtu = int_to_ibta_mtu(mtu);
1341
1342 if (err || ibta_mtu < 0) {
1343 mlx4_err(mdev, "%s is invalid IBTA mtu\n", buf);
1344 return -EINVAL;
1345 }
1346
1347 mdev->caps.port_ib_mtu[info->port] = ibta_mtu;
1348
1349 mlx4_stop_sense(mdev);
1350 mutex_lock(&priv->port_mutex);
1351 mlx4_unregister_device(mdev);
1352 for (port = 1; port <= mdev->caps.num_ports; port++) {
1353 mlx4_CLOSE_PORT(mdev, port);
1354 err = mlx4_SET_PORT(mdev, port, -1);
1355 if (err) {
1356 mlx4_err(mdev, "Failed to set port %d, aborting\n",
1357 port);
1358 goto err_set_port;
1359 }
1360 }
1361 err = mlx4_register_device(mdev);
1362 err_set_port:
1363 mutex_unlock(&priv->port_mutex);
1364 mlx4_start_sense(mdev);
1365 return err ? err : count;
1366 }
1367
1368
1369 #define MAX_MF_BOND_ALLOWED_SLAVES 63
1370 static int mlx4_mf_bond(struct mlx4_dev *dev)
1371 {
1372 int err = 0;
1373 int nvfs;
1374 struct mlx4_slaves_pport slaves_port1;
1375 struct mlx4_slaves_pport slaves_port2;
1376 DECLARE_BITMAP(slaves_port_1_2, MLX4_MFUNC_MAX);
1377
1378 slaves_port1 = mlx4_phys_to_slaves_pport(dev, 1);
1379 slaves_port2 = mlx4_phys_to_slaves_pport(dev, 2);
1380 bitmap_and(slaves_port_1_2,
1381 slaves_port1.slaves, slaves_port2.slaves,
1382 dev->persist->num_vfs + 1);
1383
1384
1385 if (bitmap_weight(slaves_port_1_2, dev->persist->num_vfs + 1) > 1) {
1386 mlx4_warn(dev, "HA mode unsupported for dual ported VFs\n");
1387 return -EINVAL;
1388 }
1389
1390
1391
1392
1393 nvfs = bitmap_weight(slaves_port1.slaves, dev->persist->num_vfs + 1) +
1394 bitmap_weight(slaves_port2.slaves, dev->persist->num_vfs + 1) - 2;
1395
1396
1397 if (nvfs > MAX_MF_BOND_ALLOWED_SLAVES) {
1398 mlx4_warn(dev, "HA mode is not supported for %d VFs (max %d are allowed)\n",
1399 nvfs, MAX_MF_BOND_ALLOWED_SLAVES);
1400 return -EINVAL;
1401 }
1402
1403 if (dev->caps.steering_mode != MLX4_STEERING_MODE_DEVICE_MANAGED) {
1404 mlx4_warn(dev, "HA mode unsupported for NON DMFS steering\n");
1405 return -EINVAL;
1406 }
1407
1408 err = mlx4_bond_mac_table(dev);
1409 if (err)
1410 return err;
1411 err = mlx4_bond_vlan_table(dev);
1412 if (err)
1413 goto err1;
1414 err = mlx4_bond_fs_rules(dev);
1415 if (err)
1416 goto err2;
1417
1418 return 0;
1419 err2:
1420 (void)mlx4_unbond_vlan_table(dev);
1421 err1:
1422 (void)mlx4_unbond_mac_table(dev);
1423 return err;
1424 }
1425
1426 static int mlx4_mf_unbond(struct mlx4_dev *dev)
1427 {
1428 int ret, ret1;
1429
1430 ret = mlx4_unbond_fs_rules(dev);
1431 if (ret)
1432 mlx4_warn(dev, "multifunction unbond for flow rules failed (%d)\n", ret);
1433 ret1 = mlx4_unbond_mac_table(dev);
1434 if (ret1) {
1435 mlx4_warn(dev, "multifunction unbond for MAC table failed (%d)\n", ret1);
1436 ret = ret1;
1437 }
1438 ret1 = mlx4_unbond_vlan_table(dev);
1439 if (ret1) {
1440 mlx4_warn(dev, "multifunction unbond for VLAN table failed (%d)\n", ret1);
1441 ret = ret1;
1442 }
1443 return ret;
1444 }
1445
1446 int mlx4_bond(struct mlx4_dev *dev)
1447 {
1448 int ret = 0;
1449 struct mlx4_priv *priv = mlx4_priv(dev);
1450
1451 mutex_lock(&priv->bond_mutex);
1452
1453 if (!mlx4_is_bonded(dev)) {
1454 ret = mlx4_do_bond(dev, true);
1455 if (ret)
1456 mlx4_err(dev, "Failed to bond device: %d\n", ret);
1457 if (!ret && mlx4_is_master(dev)) {
1458 ret = mlx4_mf_bond(dev);
1459 if (ret) {
1460 mlx4_err(dev, "bond for multifunction failed\n");
1461 mlx4_do_bond(dev, false);
1462 }
1463 }
1464 }
1465
1466 mutex_unlock(&priv->bond_mutex);
1467 if (!ret)
1468 mlx4_dbg(dev, "Device is bonded\n");
1469
1470 return ret;
1471 }
1472 EXPORT_SYMBOL_GPL(mlx4_bond);
1473
1474 int mlx4_unbond(struct mlx4_dev *dev)
1475 {
1476 int ret = 0;
1477 struct mlx4_priv *priv = mlx4_priv(dev);
1478
1479 mutex_lock(&priv->bond_mutex);
1480
1481 if (mlx4_is_bonded(dev)) {
1482 int ret2 = 0;
1483
1484 ret = mlx4_do_bond(dev, false);
1485 if (ret)
1486 mlx4_err(dev, "Failed to unbond device: %d\n", ret);
1487 if (mlx4_is_master(dev))
1488 ret2 = mlx4_mf_unbond(dev);
1489 if (ret2) {
1490 mlx4_warn(dev, "Failed to unbond device for multifunction (%d)\n", ret2);
1491 ret = ret2;
1492 }
1493 }
1494
1495 mutex_unlock(&priv->bond_mutex);
1496 if (!ret)
1497 mlx4_dbg(dev, "Device is unbonded\n");
1498
1499 return ret;
1500 }
1501 EXPORT_SYMBOL_GPL(mlx4_unbond);
1502
1503
1504 int mlx4_port_map_set(struct mlx4_dev *dev, struct mlx4_port_map *v2p)
1505 {
1506 u8 port1 = v2p->port1;
1507 u8 port2 = v2p->port2;
1508 struct mlx4_priv *priv = mlx4_priv(dev);
1509 int err;
1510
1511 if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_PORT_REMAP))
1512 return -EOPNOTSUPP;
1513
1514 mutex_lock(&priv->bond_mutex);
1515
1516
1517 if (port1 == 0)
1518 port1 = priv->v2p.port1;
1519 if (port2 == 0)
1520 port2 = priv->v2p.port2;
1521
1522 if ((port1 < 1) || (port1 > MLX4_MAX_PORTS) ||
1523 (port2 < 1) || (port2 > MLX4_MAX_PORTS) ||
1524 (port1 == 2 && port2 == 1)) {
1525
1526
1527 err = -EINVAL;
1528 } else if ((port1 == priv->v2p.port1) &&
1529 (port2 == priv->v2p.port2)) {
1530 err = 0;
1531 } else {
1532 err = mlx4_virt2phy_port_map(dev, port1, port2);
1533 if (!err) {
1534 mlx4_dbg(dev, "port map changed: [%d][%d]\n",
1535 port1, port2);
1536 priv->v2p.port1 = port1;
1537 priv->v2p.port2 = port2;
1538 } else {
1539 mlx4_err(dev, "Failed to change port mape: %d\n", err);
1540 }
1541 }
1542
1543 mutex_unlock(&priv->bond_mutex);
1544 return err;
1545 }
1546 EXPORT_SYMBOL_GPL(mlx4_port_map_set);
1547
1548 static int mlx4_load_fw(struct mlx4_dev *dev)
1549 {
1550 struct mlx4_priv *priv = mlx4_priv(dev);
1551 int err;
1552
1553 priv->fw.fw_icm = mlx4_alloc_icm(dev, priv->fw.fw_pages,
1554 GFP_HIGHUSER | __GFP_NOWARN, 0);
1555 if (!priv->fw.fw_icm) {
1556 mlx4_err(dev, "Couldn't allocate FW area, aborting\n");
1557 return -ENOMEM;
1558 }
1559
1560 err = mlx4_MAP_FA(dev, priv->fw.fw_icm);
1561 if (err) {
1562 mlx4_err(dev, "MAP_FA command failed, aborting\n");
1563 goto err_free;
1564 }
1565
1566 err = mlx4_RUN_FW(dev);
1567 if (err) {
1568 mlx4_err(dev, "RUN_FW command failed, aborting\n");
1569 goto err_unmap_fa;
1570 }
1571
1572 return 0;
1573
1574 err_unmap_fa:
1575 mlx4_UNMAP_FA(dev);
1576
1577 err_free:
1578 mlx4_free_icm(dev, priv->fw.fw_icm, 0);
1579 return err;
1580 }
1581
1582 static int mlx4_init_cmpt_table(struct mlx4_dev *dev, u64 cmpt_base,
1583 int cmpt_entry_sz)
1584 {
1585 struct mlx4_priv *priv = mlx4_priv(dev);
1586 int err;
1587 int num_eqs;
1588
1589 err = mlx4_init_icm_table(dev, &priv->qp_table.cmpt_table,
1590 cmpt_base +
1591 ((u64) (MLX4_CMPT_TYPE_QP *
1592 cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1593 cmpt_entry_sz, dev->caps.num_qps,
1594 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1595 0, 0);
1596 if (err)
1597 goto err;
1598
1599 err = mlx4_init_icm_table(dev, &priv->srq_table.cmpt_table,
1600 cmpt_base +
1601 ((u64) (MLX4_CMPT_TYPE_SRQ *
1602 cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1603 cmpt_entry_sz, dev->caps.num_srqs,
1604 dev->caps.reserved_srqs, 0, 0);
1605 if (err)
1606 goto err_qp;
1607
1608 err = mlx4_init_icm_table(dev, &priv->cq_table.cmpt_table,
1609 cmpt_base +
1610 ((u64) (MLX4_CMPT_TYPE_CQ *
1611 cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1612 cmpt_entry_sz, dev->caps.num_cqs,
1613 dev->caps.reserved_cqs, 0, 0);
1614 if (err)
1615 goto err_srq;
1616
1617 num_eqs = dev->phys_caps.num_phys_eqs;
1618 err = mlx4_init_icm_table(dev, &priv->eq_table.cmpt_table,
1619 cmpt_base +
1620 ((u64) (MLX4_CMPT_TYPE_EQ *
1621 cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1622 cmpt_entry_sz, num_eqs, num_eqs, 0, 0);
1623 if (err)
1624 goto err_cq;
1625
1626 return 0;
1627
1628 err_cq:
1629 mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1630
1631 err_srq:
1632 mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1633
1634 err_qp:
1635 mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1636
1637 err:
1638 return err;
1639 }
1640
1641 static int mlx4_init_icm(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap,
1642 struct mlx4_init_hca_param *init_hca, u64 icm_size)
1643 {
1644 struct mlx4_priv *priv = mlx4_priv(dev);
1645 u64 aux_pages;
1646 int num_eqs;
1647 int err;
1648
1649 err = mlx4_SET_ICM_SIZE(dev, icm_size, &aux_pages);
1650 if (err) {
1651 mlx4_err(dev, "SET_ICM_SIZE command failed, aborting\n");
1652 return err;
1653 }
1654
1655 mlx4_dbg(dev, "%lld KB of HCA context requires %lld KB aux memory\n",
1656 (unsigned long long) icm_size >> 10,
1657 (unsigned long long) aux_pages << 2);
1658
1659 priv->fw.aux_icm = mlx4_alloc_icm(dev, aux_pages,
1660 GFP_HIGHUSER | __GFP_NOWARN, 0);
1661 if (!priv->fw.aux_icm) {
1662 mlx4_err(dev, "Couldn't allocate aux memory, aborting\n");
1663 return -ENOMEM;
1664 }
1665
1666 err = mlx4_MAP_ICM_AUX(dev, priv->fw.aux_icm);
1667 if (err) {
1668 mlx4_err(dev, "MAP_ICM_AUX command failed, aborting\n");
1669 goto err_free_aux;
1670 }
1671
1672 err = mlx4_init_cmpt_table(dev, init_hca->cmpt_base, dev_cap->cmpt_entry_sz);
1673 if (err) {
1674 mlx4_err(dev, "Failed to map cMPT context memory, aborting\n");
1675 goto err_unmap_aux;
1676 }
1677
1678
1679 num_eqs = dev->phys_caps.num_phys_eqs;
1680 err = mlx4_init_icm_table(dev, &priv->eq_table.table,
1681 init_hca->eqc_base, dev_cap->eqc_entry_sz,
1682 num_eqs, num_eqs, 0, 0);
1683 if (err) {
1684 mlx4_err(dev, "Failed to map EQ context memory, aborting\n");
1685 goto err_unmap_cmpt;
1686 }
1687
1688
1689
1690
1691
1692
1693
1694
1695 dev->caps.reserved_mtts =
1696 ALIGN(dev->caps.reserved_mtts * dev->caps.mtt_entry_sz,
1697 dma_get_cache_alignment()) / dev->caps.mtt_entry_sz;
1698
1699 err = mlx4_init_icm_table(dev, &priv->mr_table.mtt_table,
1700 init_hca->mtt_base,
1701 dev->caps.mtt_entry_sz,
1702 dev->caps.num_mtts,
1703 dev->caps.reserved_mtts, 1, 0);
1704 if (err) {
1705 mlx4_err(dev, "Failed to map MTT context memory, aborting\n");
1706 goto err_unmap_eq;
1707 }
1708
1709 err = mlx4_init_icm_table(dev, &priv->mr_table.dmpt_table,
1710 init_hca->dmpt_base,
1711 dev_cap->dmpt_entry_sz,
1712 dev->caps.num_mpts,
1713 dev->caps.reserved_mrws, 1, 1);
1714 if (err) {
1715 mlx4_err(dev, "Failed to map dMPT context memory, aborting\n");
1716 goto err_unmap_mtt;
1717 }
1718
1719 err = mlx4_init_icm_table(dev, &priv->qp_table.qp_table,
1720 init_hca->qpc_base,
1721 dev_cap->qpc_entry_sz,
1722 dev->caps.num_qps,
1723 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1724 0, 0);
1725 if (err) {
1726 mlx4_err(dev, "Failed to map QP context memory, aborting\n");
1727 goto err_unmap_dmpt;
1728 }
1729
1730 err = mlx4_init_icm_table(dev, &priv->qp_table.auxc_table,
1731 init_hca->auxc_base,
1732 dev_cap->aux_entry_sz,
1733 dev->caps.num_qps,
1734 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1735 0, 0);
1736 if (err) {
1737 mlx4_err(dev, "Failed to map AUXC context memory, aborting\n");
1738 goto err_unmap_qp;
1739 }
1740
1741 err = mlx4_init_icm_table(dev, &priv->qp_table.altc_table,
1742 init_hca->altc_base,
1743 dev_cap->altc_entry_sz,
1744 dev->caps.num_qps,
1745 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1746 0, 0);
1747 if (err) {
1748 mlx4_err(dev, "Failed to map ALTC context memory, aborting\n");
1749 goto err_unmap_auxc;
1750 }
1751
1752 err = mlx4_init_icm_table(dev, &priv->qp_table.rdmarc_table,
1753 init_hca->rdmarc_base,
1754 dev_cap->rdmarc_entry_sz << priv->qp_table.rdmarc_shift,
1755 dev->caps.num_qps,
1756 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1757 0, 0);
1758 if (err) {
1759 mlx4_err(dev, "Failed to map RDMARC context memory, aborting\n");
1760 goto err_unmap_altc;
1761 }
1762
1763 err = mlx4_init_icm_table(dev, &priv->cq_table.table,
1764 init_hca->cqc_base,
1765 dev_cap->cqc_entry_sz,
1766 dev->caps.num_cqs,
1767 dev->caps.reserved_cqs, 0, 0);
1768 if (err) {
1769 mlx4_err(dev, "Failed to map CQ context memory, aborting\n");
1770 goto err_unmap_rdmarc;
1771 }
1772
1773 err = mlx4_init_icm_table(dev, &priv->srq_table.table,
1774 init_hca->srqc_base,
1775 dev_cap->srq_entry_sz,
1776 dev->caps.num_srqs,
1777 dev->caps.reserved_srqs, 0, 0);
1778 if (err) {
1779 mlx4_err(dev, "Failed to map SRQ context memory, aborting\n");
1780 goto err_unmap_cq;
1781 }
1782
1783
1784
1785
1786
1787
1788
1789
1790 err = mlx4_init_icm_table(dev, &priv->mcg_table.table,
1791 init_hca->mc_base,
1792 mlx4_get_mgm_entry_size(dev),
1793 dev->caps.num_mgms + dev->caps.num_amgms,
1794 dev->caps.num_mgms + dev->caps.num_amgms,
1795 0, 0);
1796 if (err) {
1797 mlx4_err(dev, "Failed to map MCG context memory, aborting\n");
1798 goto err_unmap_srq;
1799 }
1800
1801 return 0;
1802
1803 err_unmap_srq:
1804 mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1805
1806 err_unmap_cq:
1807 mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1808
1809 err_unmap_rdmarc:
1810 mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1811
1812 err_unmap_altc:
1813 mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1814
1815 err_unmap_auxc:
1816 mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1817
1818 err_unmap_qp:
1819 mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1820
1821 err_unmap_dmpt:
1822 mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1823
1824 err_unmap_mtt:
1825 mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
1826
1827 err_unmap_eq:
1828 mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
1829
1830 err_unmap_cmpt:
1831 mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1832 mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1833 mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1834 mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1835
1836 err_unmap_aux:
1837 mlx4_UNMAP_ICM_AUX(dev);
1838
1839 err_free_aux:
1840 mlx4_free_icm(dev, priv->fw.aux_icm, 0);
1841
1842 return err;
1843 }
1844
1845 static void mlx4_free_icms(struct mlx4_dev *dev)
1846 {
1847 struct mlx4_priv *priv = mlx4_priv(dev);
1848
1849 mlx4_cleanup_icm_table(dev, &priv->mcg_table.table);
1850 mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1851 mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1852 mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1853 mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1854 mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1855 mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1856 mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1857 mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
1858 mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
1859 mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1860 mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1861 mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1862 mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1863
1864 mlx4_UNMAP_ICM_AUX(dev);
1865 mlx4_free_icm(dev, priv->fw.aux_icm, 0);
1866 }
1867
1868 static void mlx4_slave_exit(struct mlx4_dev *dev)
1869 {
1870 struct mlx4_priv *priv = mlx4_priv(dev);
1871
1872 mutex_lock(&priv->cmd.slave_cmd_mutex);
1873 if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, MLX4_COMM_CMD_NA_OP,
1874 MLX4_COMM_TIME))
1875 mlx4_warn(dev, "Failed to close slave function\n");
1876 mutex_unlock(&priv->cmd.slave_cmd_mutex);
1877 }
1878
1879 static int map_bf_area(struct mlx4_dev *dev)
1880 {
1881 struct mlx4_priv *priv = mlx4_priv(dev);
1882 resource_size_t bf_start;
1883 resource_size_t bf_len;
1884 int err = 0;
1885
1886 if (!dev->caps.bf_reg_size)
1887 return -ENXIO;
1888
1889 bf_start = pci_resource_start(dev->persist->pdev, 2) +
1890 (dev->caps.num_uars << PAGE_SHIFT);
1891 bf_len = pci_resource_len(dev->persist->pdev, 2) -
1892 (dev->caps.num_uars << PAGE_SHIFT);
1893 priv->bf_mapping = io_mapping_create_wc(bf_start, bf_len);
1894 if (!priv->bf_mapping)
1895 err = -ENOMEM;
1896
1897 return err;
1898 }
1899
1900 static void unmap_bf_area(struct mlx4_dev *dev)
1901 {
1902 if (mlx4_priv(dev)->bf_mapping)
1903 io_mapping_free(mlx4_priv(dev)->bf_mapping);
1904 }
1905
1906 u64 mlx4_read_clock(struct mlx4_dev *dev)
1907 {
1908 u32 clockhi, clocklo, clockhi1;
1909 u64 cycles;
1910 int i;
1911 struct mlx4_priv *priv = mlx4_priv(dev);
1912
1913 for (i = 0; i < 10; i++) {
1914 clockhi = swab32(readl(priv->clock_mapping));
1915 clocklo = swab32(readl(priv->clock_mapping + 4));
1916 clockhi1 = swab32(readl(priv->clock_mapping));
1917 if (clockhi == clockhi1)
1918 break;
1919 }
1920
1921 cycles = (u64) clockhi << 32 | (u64) clocklo;
1922
1923 return cycles;
1924 }
1925 EXPORT_SYMBOL_GPL(mlx4_read_clock);
1926
1927
1928 static int map_internal_clock(struct mlx4_dev *dev)
1929 {
1930 struct mlx4_priv *priv = mlx4_priv(dev);
1931
1932 priv->clock_mapping =
1933 ioremap(pci_resource_start(dev->persist->pdev,
1934 priv->fw.clock_bar) +
1935 priv->fw.clock_offset, MLX4_CLOCK_SIZE);
1936
1937 if (!priv->clock_mapping)
1938 return -ENOMEM;
1939
1940 return 0;
1941 }
1942
1943 int mlx4_get_internal_clock_params(struct mlx4_dev *dev,
1944 struct mlx4_clock_params *params)
1945 {
1946 struct mlx4_priv *priv = mlx4_priv(dev);
1947
1948 if (mlx4_is_slave(dev))
1949 return -EOPNOTSUPP;
1950
1951 if (!params)
1952 return -EINVAL;
1953
1954 params->bar = priv->fw.clock_bar;
1955 params->offset = priv->fw.clock_offset;
1956 params->size = MLX4_CLOCK_SIZE;
1957
1958 return 0;
1959 }
1960 EXPORT_SYMBOL_GPL(mlx4_get_internal_clock_params);
1961
1962 static void unmap_internal_clock(struct mlx4_dev *dev)
1963 {
1964 struct mlx4_priv *priv = mlx4_priv(dev);
1965
1966 if (priv->clock_mapping)
1967 iounmap(priv->clock_mapping);
1968 }
1969
1970 static void mlx4_close_hca(struct mlx4_dev *dev)
1971 {
1972 unmap_internal_clock(dev);
1973 unmap_bf_area(dev);
1974 if (mlx4_is_slave(dev))
1975 mlx4_slave_exit(dev);
1976 else {
1977 mlx4_CLOSE_HCA(dev, 0);
1978 mlx4_free_icms(dev);
1979 }
1980 }
1981
1982 static void mlx4_close_fw(struct mlx4_dev *dev)
1983 {
1984 if (!mlx4_is_slave(dev)) {
1985 mlx4_UNMAP_FA(dev);
1986 mlx4_free_icm(dev, mlx4_priv(dev)->fw.fw_icm, 0);
1987 }
1988 }
1989
1990 static int mlx4_comm_check_offline(struct mlx4_dev *dev)
1991 {
1992 #define COMM_CHAN_OFFLINE_OFFSET 0x09
1993
1994 u32 comm_flags;
1995 u32 offline_bit;
1996 unsigned long end;
1997 struct mlx4_priv *priv = mlx4_priv(dev);
1998
1999 end = msecs_to_jiffies(MLX4_COMM_OFFLINE_TIME_OUT) + jiffies;
2000 while (time_before(jiffies, end)) {
2001 comm_flags = swab32(readl((__iomem char *)priv->mfunc.comm +
2002 MLX4_COMM_CHAN_FLAGS));
2003 offline_bit = (comm_flags &
2004 (u32)(1 << COMM_CHAN_OFFLINE_OFFSET));
2005 if (!offline_bit)
2006 return 0;
2007
2008
2009
2010
2011 if (dev->persist->interface_state &
2012 MLX4_INTERFACE_STATE_NOWAIT)
2013 break;
2014
2015
2016
2017
2018
2019
2020 msleep(100);
2021 }
2022 mlx4_err(dev, "Communication channel is offline.\n");
2023 return -EIO;
2024 }
2025
2026 static void mlx4_reset_vf_support(struct mlx4_dev *dev)
2027 {
2028 #define COMM_CHAN_RST_OFFSET 0x1e
2029
2030 struct mlx4_priv *priv = mlx4_priv(dev);
2031 u32 comm_rst;
2032 u32 comm_caps;
2033
2034 comm_caps = swab32(readl((__iomem char *)priv->mfunc.comm +
2035 MLX4_COMM_CHAN_CAPS));
2036 comm_rst = (comm_caps & (u32)(1 << COMM_CHAN_RST_OFFSET));
2037
2038 if (comm_rst)
2039 dev->caps.vf_caps |= MLX4_VF_CAP_FLAG_RESET;
2040 }
2041
2042 static int mlx4_init_slave(struct mlx4_dev *dev)
2043 {
2044 struct mlx4_priv *priv = mlx4_priv(dev);
2045 u64 dma = (u64) priv->mfunc.vhcr_dma;
2046 int ret_from_reset = 0;
2047 u32 slave_read;
2048 u32 cmd_channel_ver;
2049
2050 if (atomic_read(&pf_loading)) {
2051 mlx4_warn(dev, "PF is not ready - Deferring probe\n");
2052 return -EPROBE_DEFER;
2053 }
2054
2055 mutex_lock(&priv->cmd.slave_cmd_mutex);
2056 priv->cmd.max_cmds = 1;
2057 if (mlx4_comm_check_offline(dev)) {
2058 mlx4_err(dev, "PF is not responsive, skipping initialization\n");
2059 goto err_offline;
2060 }
2061
2062 mlx4_reset_vf_support(dev);
2063 mlx4_warn(dev, "Sending reset\n");
2064 ret_from_reset = mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0,
2065 MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME);
2066
2067
2068 if (ret_from_reset) {
2069 if (MLX4_DELAY_RESET_SLAVE == ret_from_reset) {
2070 mlx4_warn(dev, "slave is currently in the middle of FLR - Deferring probe\n");
2071 mutex_unlock(&priv->cmd.slave_cmd_mutex);
2072 return -EPROBE_DEFER;
2073 } else
2074 goto err;
2075 }
2076
2077
2078
2079 slave_read = swab32(readl(&priv->mfunc.comm->slave_read));
2080 cmd_channel_ver = mlx4_comm_get_version();
2081
2082 if (MLX4_COMM_GET_IF_REV(cmd_channel_ver) !=
2083 MLX4_COMM_GET_IF_REV(slave_read)) {
2084 mlx4_err(dev, "slave driver version is not supported by the master\n");
2085 goto err;
2086 }
2087
2088 mlx4_warn(dev, "Sending vhcr0\n");
2089 if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR0, dma >> 48,
2090 MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
2091 goto err;
2092 if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR1, dma >> 32,
2093 MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
2094 goto err;
2095 if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR2, dma >> 16,
2096 MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
2097 goto err;
2098 if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR_EN, dma,
2099 MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
2100 goto err;
2101
2102 mutex_unlock(&priv->cmd.slave_cmd_mutex);
2103 return 0;
2104
2105 err:
2106 mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, MLX4_COMM_CMD_NA_OP, 0);
2107 err_offline:
2108 mutex_unlock(&priv->cmd.slave_cmd_mutex);
2109 return -EIO;
2110 }
2111
2112 static void mlx4_parav_master_pf_caps(struct mlx4_dev *dev)
2113 {
2114 int i;
2115
2116 for (i = 1; i <= dev->caps.num_ports; i++) {
2117 if (dev->caps.port_type[i] == MLX4_PORT_TYPE_ETH)
2118 dev->caps.gid_table_len[i] =
2119 mlx4_get_slave_num_gids(dev, 0, i);
2120 else
2121 dev->caps.gid_table_len[i] = 1;
2122 dev->caps.pkey_table_len[i] =
2123 dev->phys_caps.pkey_phys_table_len[i] - 1;
2124 }
2125 }
2126
2127 static int choose_log_fs_mgm_entry_size(int qp_per_entry)
2128 {
2129 int i = MLX4_MIN_MGM_LOG_ENTRY_SIZE;
2130
2131 for (i = MLX4_MIN_MGM_LOG_ENTRY_SIZE; i <= MLX4_MAX_MGM_LOG_ENTRY_SIZE;
2132 i++) {
2133 if (qp_per_entry <= 4 * ((1 << i) / 16 - 2))
2134 break;
2135 }
2136
2137 return (i <= MLX4_MAX_MGM_LOG_ENTRY_SIZE) ? i : -1;
2138 }
2139
2140 static const char *dmfs_high_rate_steering_mode_str(int dmfs_high_steer_mode)
2141 {
2142 switch (dmfs_high_steer_mode) {
2143 case MLX4_STEERING_DMFS_A0_DEFAULT:
2144 return "default performance";
2145
2146 case MLX4_STEERING_DMFS_A0_DYNAMIC:
2147 return "dynamic hybrid mode";
2148
2149 case MLX4_STEERING_DMFS_A0_STATIC:
2150 return "performance optimized for limited rule configuration (static)";
2151
2152 case MLX4_STEERING_DMFS_A0_DISABLE:
2153 return "disabled performance optimized steering";
2154
2155 case MLX4_STEERING_DMFS_A0_NOT_SUPPORTED:
2156 return "performance optimized steering not supported";
2157
2158 default:
2159 return "Unrecognized mode";
2160 }
2161 }
2162
2163 #define MLX4_DMFS_A0_STEERING (1UL << 2)
2164
2165 static void choose_steering_mode(struct mlx4_dev *dev,
2166 struct mlx4_dev_cap *dev_cap)
2167 {
2168 if (mlx4_log_num_mgm_entry_size <= 0) {
2169 if ((-mlx4_log_num_mgm_entry_size) & MLX4_DMFS_A0_STEERING) {
2170 if (dev->caps.dmfs_high_steer_mode ==
2171 MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
2172 mlx4_err(dev, "DMFS high rate mode not supported\n");
2173 else
2174 dev->caps.dmfs_high_steer_mode =
2175 MLX4_STEERING_DMFS_A0_STATIC;
2176 }
2177 }
2178
2179 if (mlx4_log_num_mgm_entry_size <= 0 &&
2180 dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_FS_EN &&
2181 (!mlx4_is_mfunc(dev) ||
2182 (dev_cap->fs_max_num_qp_per_entry >=
2183 (dev->persist->num_vfs + 1))) &&
2184 choose_log_fs_mgm_entry_size(dev_cap->fs_max_num_qp_per_entry) >=
2185 MLX4_MIN_MGM_LOG_ENTRY_SIZE) {
2186 dev->oper_log_mgm_entry_size =
2187 choose_log_fs_mgm_entry_size(dev_cap->fs_max_num_qp_per_entry);
2188 dev->caps.steering_mode = MLX4_STEERING_MODE_DEVICE_MANAGED;
2189 dev->caps.num_qp_per_mgm = dev_cap->fs_max_num_qp_per_entry;
2190 dev->caps.fs_log_max_ucast_qp_range_size =
2191 dev_cap->fs_log_max_ucast_qp_range_size;
2192 } else {
2193 if (dev->caps.dmfs_high_steer_mode !=
2194 MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
2195 dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_DISABLE;
2196 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER &&
2197 dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER)
2198 dev->caps.steering_mode = MLX4_STEERING_MODE_B0;
2199 else {
2200 dev->caps.steering_mode = MLX4_STEERING_MODE_A0;
2201
2202 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER ||
2203 dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER)
2204 mlx4_warn(dev, "Must have both UC_STEER and MC_STEER flags set to use B0 steering - falling back to A0 steering mode\n");
2205 }
2206 dev->oper_log_mgm_entry_size =
2207 mlx4_log_num_mgm_entry_size > 0 ?
2208 mlx4_log_num_mgm_entry_size :
2209 MLX4_DEFAULT_MGM_LOG_ENTRY_SIZE;
2210 dev->caps.num_qp_per_mgm = mlx4_get_qp_per_mgm(dev);
2211 }
2212 mlx4_dbg(dev, "Steering mode is: %s, oper_log_mgm_entry_size = %d, modparam log_num_mgm_entry_size = %d\n",
2213 mlx4_steering_mode_str(dev->caps.steering_mode),
2214 dev->oper_log_mgm_entry_size,
2215 mlx4_log_num_mgm_entry_size);
2216 }
2217
2218 static void choose_tunnel_offload_mode(struct mlx4_dev *dev,
2219 struct mlx4_dev_cap *dev_cap)
2220 {
2221 if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2222 dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_VXLAN_OFFLOADS)
2223 dev->caps.tunnel_offload_mode = MLX4_TUNNEL_OFFLOAD_MODE_VXLAN;
2224 else
2225 dev->caps.tunnel_offload_mode = MLX4_TUNNEL_OFFLOAD_MODE_NONE;
2226
2227 mlx4_dbg(dev, "Tunneling offload mode is: %s\n", (dev->caps.tunnel_offload_mode
2228 == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) ? "vxlan" : "none");
2229 }
2230
2231 static int mlx4_validate_optimized_steering(struct mlx4_dev *dev)
2232 {
2233 int i;
2234 struct mlx4_port_cap port_cap;
2235
2236 if (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
2237 return -EINVAL;
2238
2239 for (i = 1; i <= dev->caps.num_ports; i++) {
2240 if (mlx4_dev_port(dev, i, &port_cap)) {
2241 mlx4_err(dev,
2242 "QUERY_DEV_CAP command failed, can't verify DMFS high rate steering.\n");
2243 } else if ((dev->caps.dmfs_high_steer_mode !=
2244 MLX4_STEERING_DMFS_A0_DEFAULT) &&
2245 (port_cap.dmfs_optimized_state ==
2246 !!(dev->caps.dmfs_high_steer_mode ==
2247 MLX4_STEERING_DMFS_A0_DISABLE))) {
2248 mlx4_err(dev,
2249 "DMFS high rate steer mode differ, driver requested %s but %s in FW.\n",
2250 dmfs_high_rate_steering_mode_str(
2251 dev->caps.dmfs_high_steer_mode),
2252 (port_cap.dmfs_optimized_state ?
2253 "enabled" : "disabled"));
2254 }
2255 }
2256
2257 return 0;
2258 }
2259
2260 static int mlx4_init_fw(struct mlx4_dev *dev)
2261 {
2262 struct mlx4_mod_stat_cfg mlx4_cfg;
2263 int err = 0;
2264
2265 if (!mlx4_is_slave(dev)) {
2266 err = mlx4_QUERY_FW(dev);
2267 if (err) {
2268 if (err == -EACCES)
2269 mlx4_info(dev, "non-primary physical function, skipping\n");
2270 else
2271 mlx4_err(dev, "QUERY_FW command failed, aborting\n");
2272 return err;
2273 }
2274
2275 err = mlx4_load_fw(dev);
2276 if (err) {
2277 mlx4_err(dev, "Failed to start FW, aborting\n");
2278 return err;
2279 }
2280
2281 mlx4_cfg.log_pg_sz_m = 1;
2282 mlx4_cfg.log_pg_sz = 0;
2283 err = mlx4_MOD_STAT_CFG(dev, &mlx4_cfg);
2284 if (err)
2285 mlx4_warn(dev, "Failed to override log_pg_sz parameter\n");
2286 }
2287
2288 return err;
2289 }
2290
2291 static int mlx4_init_hca(struct mlx4_dev *dev)
2292 {
2293 struct mlx4_priv *priv = mlx4_priv(dev);
2294 struct mlx4_init_hca_param *init_hca = NULL;
2295 struct mlx4_dev_cap *dev_cap = NULL;
2296 struct mlx4_adapter adapter;
2297 struct mlx4_profile profile;
2298 u64 icm_size;
2299 struct mlx4_config_dev_params params;
2300 int err;
2301
2302 if (!mlx4_is_slave(dev)) {
2303 dev_cap = kzalloc(sizeof(*dev_cap), GFP_KERNEL);
2304 init_hca = kzalloc(sizeof(*init_hca), GFP_KERNEL);
2305
2306 if (!dev_cap || !init_hca) {
2307 err = -ENOMEM;
2308 goto out_free;
2309 }
2310
2311 err = mlx4_dev_cap(dev, dev_cap);
2312 if (err) {
2313 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
2314 goto out_free;
2315 }
2316
2317 choose_steering_mode(dev, dev_cap);
2318 choose_tunnel_offload_mode(dev, dev_cap);
2319
2320 if (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC &&
2321 mlx4_is_master(dev))
2322 dev->caps.function_caps |= MLX4_FUNC_CAP_DMFS_A0_STATIC;
2323
2324 err = mlx4_get_phys_port_id(dev);
2325 if (err)
2326 mlx4_err(dev, "Fail to get physical port id\n");
2327
2328 if (mlx4_is_master(dev))
2329 mlx4_parav_master_pf_caps(dev);
2330
2331 if (mlx4_low_memory_profile()) {
2332 mlx4_info(dev, "Running from within kdump kernel. Using low memory profile\n");
2333 profile = low_mem_profile;
2334 } else {
2335 profile = default_profile;
2336 }
2337 if (dev->caps.steering_mode ==
2338 MLX4_STEERING_MODE_DEVICE_MANAGED)
2339 profile.num_mcg = MLX4_FS_NUM_MCG;
2340
2341 icm_size = mlx4_make_profile(dev, &profile, dev_cap,
2342 init_hca);
2343 if ((long long) icm_size < 0) {
2344 err = icm_size;
2345 goto out_free;
2346 }
2347
2348 dev->caps.max_fmr_maps = (1 << (32 - ilog2(dev->caps.num_mpts))) - 1;
2349
2350 if (enable_4k_uar || !dev->persist->num_vfs) {
2351 init_hca->log_uar_sz = ilog2(dev->caps.num_uars) +
2352 PAGE_SHIFT - DEFAULT_UAR_PAGE_SHIFT;
2353 init_hca->uar_page_sz = DEFAULT_UAR_PAGE_SHIFT - 12;
2354 } else {
2355 init_hca->log_uar_sz = ilog2(dev->caps.num_uars);
2356 init_hca->uar_page_sz = PAGE_SHIFT - 12;
2357 }
2358
2359 init_hca->mw_enabled = 0;
2360 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2361 dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN)
2362 init_hca->mw_enabled = INIT_HCA_TPT_MW_ENABLE;
2363
2364 err = mlx4_init_icm(dev, dev_cap, init_hca, icm_size);
2365 if (err)
2366 goto out_free;
2367
2368 err = mlx4_INIT_HCA(dev, init_hca);
2369 if (err) {
2370 mlx4_err(dev, "INIT_HCA command failed, aborting\n");
2371 goto err_free_icm;
2372 }
2373
2374 if (dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS) {
2375 err = mlx4_query_func(dev, dev_cap);
2376 if (err < 0) {
2377 mlx4_err(dev, "QUERY_FUNC command failed, aborting.\n");
2378 goto err_close;
2379 } else if (err & MLX4_QUERY_FUNC_NUM_SYS_EQS) {
2380 dev->caps.num_eqs = dev_cap->max_eqs;
2381 dev->caps.reserved_eqs = dev_cap->reserved_eqs;
2382 dev->caps.reserved_uars = dev_cap->reserved_uars;
2383 }
2384 }
2385
2386
2387
2388
2389
2390 if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS) {
2391 err = mlx4_QUERY_HCA(dev, init_hca);
2392 if (err) {
2393 mlx4_err(dev, "QUERY_HCA command failed, disable timestamp\n");
2394 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
2395 } else {
2396 dev->caps.hca_core_clock =
2397 init_hca->hca_core_clock;
2398 }
2399
2400
2401
2402
2403 if (!dev->caps.hca_core_clock) {
2404 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
2405 mlx4_err(dev,
2406 "HCA frequency is 0 - timestamping is not supported\n");
2407 } else if (map_internal_clock(dev)) {
2408
2409
2410
2411
2412 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
2413 mlx4_err(dev, "Failed to map internal clock. Timestamping is not supported\n");
2414 }
2415 }
2416
2417 if (dev->caps.dmfs_high_steer_mode !=
2418 MLX4_STEERING_DMFS_A0_NOT_SUPPORTED) {
2419 if (mlx4_validate_optimized_steering(dev))
2420 mlx4_warn(dev, "Optimized steering validation failed\n");
2421
2422 if (dev->caps.dmfs_high_steer_mode ==
2423 MLX4_STEERING_DMFS_A0_DISABLE) {
2424 dev->caps.dmfs_high_rate_qpn_base =
2425 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
2426 dev->caps.dmfs_high_rate_qpn_range =
2427 MLX4_A0_STEERING_TABLE_SIZE;
2428 }
2429
2430 mlx4_info(dev, "DMFS high rate steer mode is: %s\n",
2431 dmfs_high_rate_steering_mode_str(
2432 dev->caps.dmfs_high_steer_mode));
2433 }
2434 } else {
2435 err = mlx4_init_slave(dev);
2436 if (err) {
2437 if (err != -EPROBE_DEFER)
2438 mlx4_err(dev, "Failed to initialize slave\n");
2439 return err;
2440 }
2441
2442 err = mlx4_slave_cap(dev);
2443 if (err) {
2444 mlx4_err(dev, "Failed to obtain slave caps\n");
2445 goto err_close;
2446 }
2447 }
2448
2449 if (map_bf_area(dev))
2450 mlx4_dbg(dev, "Failed to map blue flame area\n");
2451
2452
2453 if (!mlx4_is_slave(dev))
2454 mlx4_set_port_mask(dev);
2455
2456 err = mlx4_QUERY_ADAPTER(dev, &adapter);
2457 if (err) {
2458 mlx4_err(dev, "QUERY_ADAPTER command failed, aborting\n");
2459 goto unmap_bf;
2460 }
2461
2462
2463 err = mlx4_config_dev_retrieval(dev, ¶ms);
2464 if (err && err != -EOPNOTSUPP) {
2465 mlx4_err(dev, "Failed to query CONFIG_DEV parameters\n");
2466 } else if (!err) {
2467 dev->caps.rx_checksum_flags_port[1] = params.rx_csum_flags_port_1;
2468 dev->caps.rx_checksum_flags_port[2] = params.rx_csum_flags_port_2;
2469 }
2470 priv->eq_table.inta_pin = adapter.inta_pin;
2471 memcpy(dev->board_id, adapter.board_id, sizeof(dev->board_id));
2472
2473 err = 0;
2474 goto out_free;
2475
2476 unmap_bf:
2477 unmap_internal_clock(dev);
2478 unmap_bf_area(dev);
2479
2480 if (mlx4_is_slave(dev))
2481 mlx4_slave_destroy_special_qp_cap(dev);
2482
2483 err_close:
2484 if (mlx4_is_slave(dev))
2485 mlx4_slave_exit(dev);
2486 else
2487 mlx4_CLOSE_HCA(dev, 0);
2488
2489 err_free_icm:
2490 if (!mlx4_is_slave(dev))
2491 mlx4_free_icms(dev);
2492
2493 out_free:
2494 kfree(dev_cap);
2495 kfree(init_hca);
2496
2497 return err;
2498 }
2499
2500 static int mlx4_init_counters_table(struct mlx4_dev *dev)
2501 {
2502 struct mlx4_priv *priv = mlx4_priv(dev);
2503 int nent_pow2;
2504
2505 if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2506 return -ENOENT;
2507
2508 if (!dev->caps.max_counters)
2509 return -ENOSPC;
2510
2511 nent_pow2 = roundup_pow_of_two(dev->caps.max_counters);
2512
2513 return mlx4_bitmap_init(&priv->counters_bitmap, nent_pow2,
2514 nent_pow2 - 1, 0,
2515 nent_pow2 - dev->caps.max_counters + 1);
2516 }
2517
2518 static void mlx4_cleanup_counters_table(struct mlx4_dev *dev)
2519 {
2520 if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2521 return;
2522
2523 if (!dev->caps.max_counters)
2524 return;
2525
2526 mlx4_bitmap_cleanup(&mlx4_priv(dev)->counters_bitmap);
2527 }
2528
2529 static void mlx4_cleanup_default_counters(struct mlx4_dev *dev)
2530 {
2531 struct mlx4_priv *priv = mlx4_priv(dev);
2532 int port;
2533
2534 for (port = 0; port < dev->caps.num_ports; port++)
2535 if (priv->def_counter[port] != -1)
2536 mlx4_counter_free(dev, priv->def_counter[port]);
2537 }
2538
2539 static int mlx4_allocate_default_counters(struct mlx4_dev *dev)
2540 {
2541 struct mlx4_priv *priv = mlx4_priv(dev);
2542 int port, err = 0;
2543 u32 idx;
2544
2545 for (port = 0; port < dev->caps.num_ports; port++)
2546 priv->def_counter[port] = -1;
2547
2548 for (port = 0; port < dev->caps.num_ports; port++) {
2549 err = mlx4_counter_alloc(dev, &idx, MLX4_RES_USAGE_DRIVER);
2550
2551 if (!err || err == -ENOSPC) {
2552 priv->def_counter[port] = idx;
2553 err = 0;
2554 } else if (err == -ENOENT) {
2555 err = 0;
2556 continue;
2557 } else if (mlx4_is_slave(dev) && err == -EINVAL) {
2558 priv->def_counter[port] = MLX4_SINK_COUNTER_INDEX(dev);
2559 mlx4_warn(dev, "can't allocate counter from old PF driver, using index %d\n",
2560 MLX4_SINK_COUNTER_INDEX(dev));
2561 err = 0;
2562 } else {
2563 mlx4_err(dev, "%s: failed to allocate default counter port %d err %d\n",
2564 __func__, port + 1, err);
2565 mlx4_cleanup_default_counters(dev);
2566 return err;
2567 }
2568
2569 mlx4_dbg(dev, "%s: default counter index %d for port %d\n",
2570 __func__, priv->def_counter[port], port + 1);
2571 }
2572
2573 return err;
2574 }
2575
2576 int __mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx)
2577 {
2578 struct mlx4_priv *priv = mlx4_priv(dev);
2579
2580 if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2581 return -ENOENT;
2582
2583 *idx = mlx4_bitmap_alloc(&priv->counters_bitmap);
2584 if (*idx == -1) {
2585 *idx = MLX4_SINK_COUNTER_INDEX(dev);
2586 return -ENOSPC;
2587 }
2588
2589 return 0;
2590 }
2591
2592 int mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx, u8 usage)
2593 {
2594 u32 in_modifier = RES_COUNTER | (((u32)usage & 3) << 30);
2595 u64 out_param;
2596 int err;
2597
2598 if (mlx4_is_mfunc(dev)) {
2599 err = mlx4_cmd_imm(dev, 0, &out_param, in_modifier,
2600 RES_OP_RESERVE, MLX4_CMD_ALLOC_RES,
2601 MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
2602 if (!err)
2603 *idx = get_param_l(&out_param);
2604 if (WARN_ON(err == -ENOSPC))
2605 err = -EINVAL;
2606 return err;
2607 }
2608 return __mlx4_counter_alloc(dev, idx);
2609 }
2610 EXPORT_SYMBOL_GPL(mlx4_counter_alloc);
2611
2612 static int __mlx4_clear_if_stat(struct mlx4_dev *dev,
2613 u8 counter_index)
2614 {
2615 struct mlx4_cmd_mailbox *if_stat_mailbox;
2616 int err;
2617 u32 if_stat_in_mod = (counter_index & 0xff) | MLX4_QUERY_IF_STAT_RESET;
2618
2619 if_stat_mailbox = mlx4_alloc_cmd_mailbox(dev);
2620 if (IS_ERR(if_stat_mailbox))
2621 return PTR_ERR(if_stat_mailbox);
2622
2623 err = mlx4_cmd_box(dev, 0, if_stat_mailbox->dma, if_stat_in_mod, 0,
2624 MLX4_CMD_QUERY_IF_STAT, MLX4_CMD_TIME_CLASS_C,
2625 MLX4_CMD_NATIVE);
2626
2627 mlx4_free_cmd_mailbox(dev, if_stat_mailbox);
2628 return err;
2629 }
2630
2631 void __mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
2632 {
2633 if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2634 return;
2635
2636 if (idx == MLX4_SINK_COUNTER_INDEX(dev))
2637 return;
2638
2639 __mlx4_clear_if_stat(dev, idx);
2640
2641 mlx4_bitmap_free(&mlx4_priv(dev)->counters_bitmap, idx, MLX4_USE_RR);
2642 return;
2643 }
2644
2645 void mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
2646 {
2647 u64 in_param = 0;
2648
2649 if (mlx4_is_mfunc(dev)) {
2650 set_param_l(&in_param, idx);
2651 mlx4_cmd(dev, in_param, RES_COUNTER, RES_OP_RESERVE,
2652 MLX4_CMD_FREE_RES, MLX4_CMD_TIME_CLASS_A,
2653 MLX4_CMD_WRAPPED);
2654 return;
2655 }
2656 __mlx4_counter_free(dev, idx);
2657 }
2658 EXPORT_SYMBOL_GPL(mlx4_counter_free);
2659
2660 int mlx4_get_default_counter_index(struct mlx4_dev *dev, int port)
2661 {
2662 struct mlx4_priv *priv = mlx4_priv(dev);
2663
2664 return priv->def_counter[port - 1];
2665 }
2666 EXPORT_SYMBOL_GPL(mlx4_get_default_counter_index);
2667
2668 void mlx4_set_admin_guid(struct mlx4_dev *dev, __be64 guid, int entry, int port)
2669 {
2670 struct mlx4_priv *priv = mlx4_priv(dev);
2671
2672 priv->mfunc.master.vf_admin[entry].vport[port].guid = guid;
2673 }
2674 EXPORT_SYMBOL_GPL(mlx4_set_admin_guid);
2675
2676 __be64 mlx4_get_admin_guid(struct mlx4_dev *dev, int entry, int port)
2677 {
2678 struct mlx4_priv *priv = mlx4_priv(dev);
2679
2680 return priv->mfunc.master.vf_admin[entry].vport[port].guid;
2681 }
2682 EXPORT_SYMBOL_GPL(mlx4_get_admin_guid);
2683
2684 void mlx4_set_random_admin_guid(struct mlx4_dev *dev, int entry, int port)
2685 {
2686 struct mlx4_priv *priv = mlx4_priv(dev);
2687 __be64 guid;
2688
2689
2690 if (entry == 0)
2691 return;
2692
2693 get_random_bytes((char *)&guid, sizeof(guid));
2694 guid &= ~(cpu_to_be64(1ULL << 56));
2695 guid |= cpu_to_be64(1ULL << 57);
2696 priv->mfunc.master.vf_admin[entry].vport[port].guid = guid;
2697 }
2698
2699 static int mlx4_setup_hca(struct mlx4_dev *dev)
2700 {
2701 struct mlx4_priv *priv = mlx4_priv(dev);
2702 int err;
2703 int port;
2704 __be32 ib_port_default_caps;
2705
2706 err = mlx4_init_uar_table(dev);
2707 if (err) {
2708 mlx4_err(dev, "Failed to initialize user access region table, aborting\n");
2709 return err;
2710 }
2711
2712 err = mlx4_uar_alloc(dev, &priv->driver_uar);
2713 if (err) {
2714 mlx4_err(dev, "Failed to allocate driver access region, aborting\n");
2715 goto err_uar_table_free;
2716 }
2717
2718 priv->kar = ioremap((phys_addr_t) priv->driver_uar.pfn << PAGE_SHIFT, PAGE_SIZE);
2719 if (!priv->kar) {
2720 mlx4_err(dev, "Couldn't map kernel access region, aborting\n");
2721 err = -ENOMEM;
2722 goto err_uar_free;
2723 }
2724
2725 err = mlx4_init_pd_table(dev);
2726 if (err) {
2727 mlx4_err(dev, "Failed to initialize protection domain table, aborting\n");
2728 goto err_kar_unmap;
2729 }
2730
2731 err = mlx4_init_xrcd_table(dev);
2732 if (err) {
2733 mlx4_err(dev, "Failed to initialize reliable connection domain table, aborting\n");
2734 goto err_pd_table_free;
2735 }
2736
2737 err = mlx4_init_mr_table(dev);
2738 if (err) {
2739 mlx4_err(dev, "Failed to initialize memory region table, aborting\n");
2740 goto err_xrcd_table_free;
2741 }
2742
2743 if (!mlx4_is_slave(dev)) {
2744 err = mlx4_init_mcg_table(dev);
2745 if (err) {
2746 mlx4_err(dev, "Failed to initialize multicast group table, aborting\n");
2747 goto err_mr_table_free;
2748 }
2749 err = mlx4_config_mad_demux(dev);
2750 if (err) {
2751 mlx4_err(dev, "Failed in config_mad_demux, aborting\n");
2752 goto err_mcg_table_free;
2753 }
2754 }
2755
2756 err = mlx4_init_eq_table(dev);
2757 if (err) {
2758 mlx4_err(dev, "Failed to initialize event queue table, aborting\n");
2759 goto err_mcg_table_free;
2760 }
2761
2762 err = mlx4_cmd_use_events(dev);
2763 if (err) {
2764 mlx4_err(dev, "Failed to switch to event-driven firmware commands, aborting\n");
2765 goto err_eq_table_free;
2766 }
2767
2768 err = mlx4_NOP(dev);
2769 if (err) {
2770 if (dev->flags & MLX4_FLAG_MSI_X) {
2771 mlx4_warn(dev, "NOP command failed to generate MSI-X interrupt IRQ %d)\n",
2772 priv->eq_table.eq[MLX4_EQ_ASYNC].irq);
2773 mlx4_warn(dev, "Trying again without MSI-X\n");
2774 } else {
2775 mlx4_err(dev, "NOP command failed to generate interrupt (IRQ %d), aborting\n",
2776 priv->eq_table.eq[MLX4_EQ_ASYNC].irq);
2777 mlx4_err(dev, "BIOS or ACPI interrupt routing problem?\n");
2778 }
2779
2780 goto err_cmd_poll;
2781 }
2782
2783 mlx4_dbg(dev, "NOP command IRQ test passed\n");
2784
2785 err = mlx4_init_cq_table(dev);
2786 if (err) {
2787 mlx4_err(dev, "Failed to initialize completion queue table, aborting\n");
2788 goto err_cmd_poll;
2789 }
2790
2791 err = mlx4_init_srq_table(dev);
2792 if (err) {
2793 mlx4_err(dev, "Failed to initialize shared receive queue table, aborting\n");
2794 goto err_cq_table_free;
2795 }
2796
2797 err = mlx4_init_qp_table(dev);
2798 if (err) {
2799 mlx4_err(dev, "Failed to initialize queue pair table, aborting\n");
2800 goto err_srq_table_free;
2801 }
2802
2803 if (!mlx4_is_slave(dev)) {
2804 err = mlx4_init_counters_table(dev);
2805 if (err && err != -ENOENT) {
2806 mlx4_err(dev, "Failed to initialize counters table, aborting\n");
2807 goto err_qp_table_free;
2808 }
2809 }
2810
2811 err = mlx4_allocate_default_counters(dev);
2812 if (err) {
2813 mlx4_err(dev, "Failed to allocate default counters, aborting\n");
2814 goto err_counters_table_free;
2815 }
2816
2817 if (!mlx4_is_slave(dev)) {
2818 for (port = 1; port <= dev->caps.num_ports; port++) {
2819 ib_port_default_caps = 0;
2820 err = mlx4_get_port_ib_caps(dev, port,
2821 &ib_port_default_caps);
2822 if (err)
2823 mlx4_warn(dev, "failed to get port %d default ib capabilities (%d). Continuing with caps = 0\n",
2824 port, err);
2825 dev->caps.ib_port_def_cap[port] = ib_port_default_caps;
2826
2827
2828 if (mlx4_is_master(dev)) {
2829 int i;
2830 for (i = 0; i < dev->num_slaves; i++) {
2831 if (i == mlx4_master_func_num(dev))
2832 continue;
2833 priv->mfunc.master.slave_state[i].ib_cap_mask[port] =
2834 ib_port_default_caps;
2835 }
2836 }
2837
2838 if (mlx4_is_mfunc(dev))
2839 dev->caps.port_ib_mtu[port] = IB_MTU_2048;
2840 else
2841 dev->caps.port_ib_mtu[port] = IB_MTU_4096;
2842
2843 err = mlx4_SET_PORT(dev, port, mlx4_is_master(dev) ?
2844 dev->caps.pkey_table_len[port] : -1);
2845 if (err) {
2846 mlx4_err(dev, "Failed to set port %d, aborting\n",
2847 port);
2848 goto err_default_countes_free;
2849 }
2850 }
2851 }
2852
2853 return 0;
2854
2855 err_default_countes_free:
2856 mlx4_cleanup_default_counters(dev);
2857
2858 err_counters_table_free:
2859 if (!mlx4_is_slave(dev))
2860 mlx4_cleanup_counters_table(dev);
2861
2862 err_qp_table_free:
2863 mlx4_cleanup_qp_table(dev);
2864
2865 err_srq_table_free:
2866 mlx4_cleanup_srq_table(dev);
2867
2868 err_cq_table_free:
2869 mlx4_cleanup_cq_table(dev);
2870
2871 err_cmd_poll:
2872 mlx4_cmd_use_polling(dev);
2873
2874 err_eq_table_free:
2875 mlx4_cleanup_eq_table(dev);
2876
2877 err_mcg_table_free:
2878 if (!mlx4_is_slave(dev))
2879 mlx4_cleanup_mcg_table(dev);
2880
2881 err_mr_table_free:
2882 mlx4_cleanup_mr_table(dev);
2883
2884 err_xrcd_table_free:
2885 mlx4_cleanup_xrcd_table(dev);
2886
2887 err_pd_table_free:
2888 mlx4_cleanup_pd_table(dev);
2889
2890 err_kar_unmap:
2891 iounmap(priv->kar);
2892
2893 err_uar_free:
2894 mlx4_uar_free(dev, &priv->driver_uar);
2895
2896 err_uar_table_free:
2897 mlx4_cleanup_uar_table(dev);
2898 return err;
2899 }
2900
2901 static int mlx4_init_affinity_hint(struct mlx4_dev *dev, int port, int eqn)
2902 {
2903 int requested_cpu = 0;
2904 struct mlx4_priv *priv = mlx4_priv(dev);
2905 struct mlx4_eq *eq;
2906 int off = 0;
2907 int i;
2908
2909 if (eqn > dev->caps.num_comp_vectors)
2910 return -EINVAL;
2911
2912 for (i = 1; i < port; i++)
2913 off += mlx4_get_eqs_per_port(dev, i);
2914
2915 requested_cpu = eqn - off - !!(eqn > MLX4_EQ_ASYNC);
2916
2917
2918 if (requested_cpu < 0)
2919 return 0;
2920
2921 eq = &priv->eq_table.eq[eqn];
2922
2923 if (!zalloc_cpumask_var(&eq->affinity_mask, GFP_KERNEL))
2924 return -ENOMEM;
2925
2926 cpumask_set_cpu(requested_cpu, eq->affinity_mask);
2927
2928 return 0;
2929 }
2930
2931 static void mlx4_enable_msi_x(struct mlx4_dev *dev)
2932 {
2933 struct mlx4_priv *priv = mlx4_priv(dev);
2934 struct msix_entry *entries;
2935 int i;
2936 int port = 0;
2937
2938 if (msi_x) {
2939 int nreq = min3(dev->caps.num_ports *
2940 (int)num_online_cpus() + 1,
2941 dev->caps.num_eqs - dev->caps.reserved_eqs,
2942 MAX_MSIX);
2943
2944 if (msi_x > 1)
2945 nreq = min_t(int, nreq, msi_x);
2946
2947 entries = kcalloc(nreq, sizeof(*entries), GFP_KERNEL);
2948 if (!entries)
2949 goto no_msi;
2950
2951 for (i = 0; i < nreq; ++i)
2952 entries[i].entry = i;
2953
2954 nreq = pci_enable_msix_range(dev->persist->pdev, entries, 2,
2955 nreq);
2956
2957 if (nreq < 0 || nreq < MLX4_EQ_ASYNC) {
2958 kfree(entries);
2959 goto no_msi;
2960 }
2961
2962 dev->caps.num_comp_vectors = nreq - 1;
2963
2964 priv->eq_table.eq[MLX4_EQ_ASYNC].irq = entries[0].vector;
2965 bitmap_zero(priv->eq_table.eq[MLX4_EQ_ASYNC].actv_ports.ports,
2966 dev->caps.num_ports);
2967
2968 for (i = 0; i < dev->caps.num_comp_vectors + 1; i++) {
2969 if (i == MLX4_EQ_ASYNC)
2970 continue;
2971
2972 priv->eq_table.eq[i].irq =
2973 entries[i + 1 - !!(i > MLX4_EQ_ASYNC)].vector;
2974
2975 if (MLX4_IS_LEGACY_EQ_MODE(dev->caps)) {
2976 bitmap_fill(priv->eq_table.eq[i].actv_ports.ports,
2977 dev->caps.num_ports);
2978
2979
2980
2981 } else {
2982 set_bit(port,
2983 priv->eq_table.eq[i].actv_ports.ports);
2984 if (mlx4_init_affinity_hint(dev, port + 1, i))
2985 mlx4_warn(dev, "Couldn't init hint cpumask for EQ %d\n",
2986 i);
2987 }
2988
2989
2990
2991
2992
2993
2994
2995
2996
2997 if ((dev->caps.num_comp_vectors > dev->caps.num_ports) &&
2998 ((i + 1) %
2999 (dev->caps.num_comp_vectors / dev->caps.num_ports)) ==
3000 !!((i + 1) > MLX4_EQ_ASYNC))
3001
3002
3003
3004 port++;
3005 }
3006
3007 dev->flags |= MLX4_FLAG_MSI_X;
3008
3009 kfree(entries);
3010 return;
3011 }
3012
3013 no_msi:
3014 dev->caps.num_comp_vectors = 1;
3015
3016 BUG_ON(MLX4_EQ_ASYNC >= 2);
3017 for (i = 0; i < 2; ++i) {
3018 priv->eq_table.eq[i].irq = dev->persist->pdev->irq;
3019 if (i != MLX4_EQ_ASYNC) {
3020 bitmap_fill(priv->eq_table.eq[i].actv_ports.ports,
3021 dev->caps.num_ports);
3022 }
3023 }
3024 }
3025
3026 static int mlx4_init_port_info(struct mlx4_dev *dev, int port)
3027 {
3028 struct devlink *devlink = priv_to_devlink(mlx4_priv(dev));
3029 struct mlx4_port_info *info = &mlx4_priv(dev)->port[port];
3030 int err;
3031
3032 err = devlink_port_register(devlink, &info->devlink_port, port);
3033 if (err)
3034 return err;
3035
3036 info->dev = dev;
3037 info->port = port;
3038 if (!mlx4_is_slave(dev)) {
3039 mlx4_init_mac_table(dev, &info->mac_table);
3040 mlx4_init_vlan_table(dev, &info->vlan_table);
3041 mlx4_init_roce_gid_table(dev, &info->gid_table);
3042 info->base_qpn = mlx4_get_base_qpn(dev, port);
3043 }
3044
3045 sprintf(info->dev_name, "mlx4_port%d", port);
3046 info->port_attr.attr.name = info->dev_name;
3047 if (mlx4_is_mfunc(dev)) {
3048 info->port_attr.attr.mode = 0444;
3049 } else {
3050 info->port_attr.attr.mode = 0644;
3051 info->port_attr.store = set_port_type;
3052 }
3053 info->port_attr.show = show_port_type;
3054 sysfs_attr_init(&info->port_attr.attr);
3055
3056 err = device_create_file(&dev->persist->pdev->dev, &info->port_attr);
3057 if (err) {
3058 mlx4_err(dev, "Failed to create file for port %d\n", port);
3059 devlink_port_unregister(&info->devlink_port);
3060 info->port = -1;
3061 return err;
3062 }
3063
3064 sprintf(info->dev_mtu_name, "mlx4_port%d_mtu", port);
3065 info->port_mtu_attr.attr.name = info->dev_mtu_name;
3066 if (mlx4_is_mfunc(dev)) {
3067 info->port_mtu_attr.attr.mode = 0444;
3068 } else {
3069 info->port_mtu_attr.attr.mode = 0644;
3070 info->port_mtu_attr.store = set_port_ib_mtu;
3071 }
3072 info->port_mtu_attr.show = show_port_ib_mtu;
3073 sysfs_attr_init(&info->port_mtu_attr.attr);
3074
3075 err = device_create_file(&dev->persist->pdev->dev,
3076 &info->port_mtu_attr);
3077 if (err) {
3078 mlx4_err(dev, "Failed to create mtu file for port %d\n", port);
3079 device_remove_file(&info->dev->persist->pdev->dev,
3080 &info->port_attr);
3081 devlink_port_unregister(&info->devlink_port);
3082 info->port = -1;
3083 return err;
3084 }
3085
3086 return 0;
3087 }
3088
3089 static void mlx4_cleanup_port_info(struct mlx4_port_info *info)
3090 {
3091 if (info->port < 0)
3092 return;
3093
3094 device_remove_file(&info->dev->persist->pdev->dev, &info->port_attr);
3095 device_remove_file(&info->dev->persist->pdev->dev,
3096 &info->port_mtu_attr);
3097 devlink_port_unregister(&info->devlink_port);
3098
3099 #ifdef CONFIG_RFS_ACCEL
3100 free_irq_cpu_rmap(info->rmap);
3101 info->rmap = NULL;
3102 #endif
3103 }
3104
3105 static int mlx4_init_steering(struct mlx4_dev *dev)
3106 {
3107 struct mlx4_priv *priv = mlx4_priv(dev);
3108 int num_entries = dev->caps.num_ports;
3109 int i, j;
3110
3111 priv->steer = kcalloc(num_entries, sizeof(struct mlx4_steer),
3112 GFP_KERNEL);
3113 if (!priv->steer)
3114 return -ENOMEM;
3115
3116 for (i = 0; i < num_entries; i++)
3117 for (j = 0; j < MLX4_NUM_STEERS; j++) {
3118 INIT_LIST_HEAD(&priv->steer[i].promisc_qps[j]);
3119 INIT_LIST_HEAD(&priv->steer[i].steer_entries[j]);
3120 }
3121 return 0;
3122 }
3123
3124 static void mlx4_clear_steering(struct mlx4_dev *dev)
3125 {
3126 struct mlx4_priv *priv = mlx4_priv(dev);
3127 struct mlx4_steer_index *entry, *tmp_entry;
3128 struct mlx4_promisc_qp *pqp, *tmp_pqp;
3129 int num_entries = dev->caps.num_ports;
3130 int i, j;
3131
3132 for (i = 0; i < num_entries; i++) {
3133 for (j = 0; j < MLX4_NUM_STEERS; j++) {
3134 list_for_each_entry_safe(pqp, tmp_pqp,
3135 &priv->steer[i].promisc_qps[j],
3136 list) {
3137 list_del(&pqp->list);
3138 kfree(pqp);
3139 }
3140 list_for_each_entry_safe(entry, tmp_entry,
3141 &priv->steer[i].steer_entries[j],
3142 list) {
3143 list_del(&entry->list);
3144 list_for_each_entry_safe(pqp, tmp_pqp,
3145 &entry->duplicates,
3146 list) {
3147 list_del(&pqp->list);
3148 kfree(pqp);
3149 }
3150 kfree(entry);
3151 }
3152 }
3153 }
3154 kfree(priv->steer);
3155 }
3156
3157 static int extended_func_num(struct pci_dev *pdev)
3158 {
3159 return PCI_SLOT(pdev->devfn) * 8 + PCI_FUNC(pdev->devfn);
3160 }
3161
3162 #define MLX4_OWNER_BASE 0x8069c
3163 #define MLX4_OWNER_SIZE 4
3164
3165 static int mlx4_get_ownership(struct mlx4_dev *dev)
3166 {
3167 void __iomem *owner;
3168 u32 ret;
3169
3170 if (pci_channel_offline(dev->persist->pdev))
3171 return -EIO;
3172
3173 owner = ioremap(pci_resource_start(dev->persist->pdev, 0) +
3174 MLX4_OWNER_BASE,
3175 MLX4_OWNER_SIZE);
3176 if (!owner) {
3177 mlx4_err(dev, "Failed to obtain ownership bit\n");
3178 return -ENOMEM;
3179 }
3180
3181 ret = readl(owner);
3182 iounmap(owner);
3183 return (int) !!ret;
3184 }
3185
3186 static void mlx4_free_ownership(struct mlx4_dev *dev)
3187 {
3188 void __iomem *owner;
3189
3190 if (pci_channel_offline(dev->persist->pdev))
3191 return;
3192
3193 owner = ioremap(pci_resource_start(dev->persist->pdev, 0) +
3194 MLX4_OWNER_BASE,
3195 MLX4_OWNER_SIZE);
3196 if (!owner) {
3197 mlx4_err(dev, "Failed to obtain ownership bit\n");
3198 return;
3199 }
3200 writel(0, owner);
3201 msleep(1000);
3202 iounmap(owner);
3203 }
3204
3205 #define SRIOV_VALID_STATE(flags) (!!((flags) & MLX4_FLAG_SRIOV) ==\
3206 !!((flags) & MLX4_FLAG_MASTER))
3207
3208 static u64 mlx4_enable_sriov(struct mlx4_dev *dev, struct pci_dev *pdev,
3209 u8 total_vfs, int existing_vfs, int reset_flow)
3210 {
3211 u64 dev_flags = dev->flags;
3212 int err = 0;
3213 int fw_enabled_sriov_vfs = min(pci_sriov_get_totalvfs(pdev),
3214 MLX4_MAX_NUM_VF);
3215
3216 if (reset_flow) {
3217 dev->dev_vfs = kcalloc(total_vfs, sizeof(*dev->dev_vfs),
3218 GFP_KERNEL);
3219 if (!dev->dev_vfs)
3220 goto free_mem;
3221 return dev_flags;
3222 }
3223
3224 atomic_inc(&pf_loading);
3225 if (dev->flags & MLX4_FLAG_SRIOV) {
3226 if (existing_vfs != total_vfs) {
3227 mlx4_err(dev, "SR-IOV was already enabled, but with num_vfs (%d) different than requested (%d)\n",
3228 existing_vfs, total_vfs);
3229 total_vfs = existing_vfs;
3230 }
3231 }
3232
3233 dev->dev_vfs = kcalloc(total_vfs, sizeof(*dev->dev_vfs), GFP_KERNEL);
3234 if (NULL == dev->dev_vfs) {
3235 mlx4_err(dev, "Failed to allocate memory for VFs\n");
3236 goto disable_sriov;
3237 }
3238
3239 if (!(dev->flags & MLX4_FLAG_SRIOV)) {
3240 if (total_vfs > fw_enabled_sriov_vfs) {
3241 mlx4_err(dev, "requested vfs (%d) > available vfs (%d). Continuing without SR_IOV\n",
3242 total_vfs, fw_enabled_sriov_vfs);
3243 err = -ENOMEM;
3244 goto disable_sriov;
3245 }
3246 mlx4_warn(dev, "Enabling SR-IOV with %d VFs\n", total_vfs);
3247 err = pci_enable_sriov(pdev, total_vfs);
3248 }
3249 if (err) {
3250 mlx4_err(dev, "Failed to enable SR-IOV, continuing without SR-IOV (err = %d)\n",
3251 err);
3252 goto disable_sriov;
3253 } else {
3254 mlx4_warn(dev, "Running in master mode\n");
3255 dev_flags |= MLX4_FLAG_SRIOV |
3256 MLX4_FLAG_MASTER;
3257 dev_flags &= ~MLX4_FLAG_SLAVE;
3258 dev->persist->num_vfs = total_vfs;
3259 }
3260 return dev_flags;
3261
3262 disable_sriov:
3263 atomic_dec(&pf_loading);
3264 free_mem:
3265 dev->persist->num_vfs = 0;
3266 kfree(dev->dev_vfs);
3267 dev->dev_vfs = NULL;
3268 return dev_flags & ~MLX4_FLAG_MASTER;
3269 }
3270
3271 enum {
3272 MLX4_DEV_CAP_CHECK_NUM_VFS_ABOVE_64 = -1,
3273 };
3274
3275 static int mlx4_check_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap,
3276 int *nvfs)
3277 {
3278 int requested_vfs = nvfs[0] + nvfs[1] + nvfs[2];
3279
3280 if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_80_VFS) &&
3281 requested_vfs >= 64) {
3282 mlx4_err(dev, "Requested %d VFs, but FW does not support more than 64\n",
3283 requested_vfs);
3284 return MLX4_DEV_CAP_CHECK_NUM_VFS_ABOVE_64;
3285 }
3286 return 0;
3287 }
3288
3289 static int mlx4_pci_enable_device(struct mlx4_dev *dev)
3290 {
3291 struct pci_dev *pdev = dev->persist->pdev;
3292 int err = 0;
3293
3294 mutex_lock(&dev->persist->pci_status_mutex);
3295 if (dev->persist->pci_status == MLX4_PCI_STATUS_DISABLED) {
3296 err = pci_enable_device(pdev);
3297 if (!err)
3298 dev->persist->pci_status = MLX4_PCI_STATUS_ENABLED;
3299 }
3300 mutex_unlock(&dev->persist->pci_status_mutex);
3301
3302 return err;
3303 }
3304
3305 static void mlx4_pci_disable_device(struct mlx4_dev *dev)
3306 {
3307 struct pci_dev *pdev = dev->persist->pdev;
3308
3309 mutex_lock(&dev->persist->pci_status_mutex);
3310 if (dev->persist->pci_status == MLX4_PCI_STATUS_ENABLED) {
3311 pci_disable_device(pdev);
3312 dev->persist->pci_status = MLX4_PCI_STATUS_DISABLED;
3313 }
3314 mutex_unlock(&dev->persist->pci_status_mutex);
3315 }
3316
3317 static int mlx4_load_one(struct pci_dev *pdev, int pci_dev_data,
3318 int total_vfs, int *nvfs, struct mlx4_priv *priv,
3319 int reset_flow)
3320 {
3321 struct mlx4_dev *dev;
3322 unsigned sum = 0;
3323 int err;
3324 int port;
3325 int i;
3326 struct mlx4_dev_cap *dev_cap = NULL;
3327 int existing_vfs = 0;
3328
3329 dev = &priv->dev;
3330
3331 INIT_LIST_HEAD(&priv->ctx_list);
3332 spin_lock_init(&priv->ctx_lock);
3333
3334 mutex_init(&priv->port_mutex);
3335 mutex_init(&priv->bond_mutex);
3336
3337 INIT_LIST_HEAD(&priv->pgdir_list);
3338 mutex_init(&priv->pgdir_mutex);
3339 spin_lock_init(&priv->cmd.context_lock);
3340
3341 INIT_LIST_HEAD(&priv->bf_list);
3342 mutex_init(&priv->bf_mutex);
3343
3344 dev->rev_id = pdev->revision;
3345 dev->numa_node = dev_to_node(&pdev->dev);
3346
3347
3348 if (pci_dev_data & MLX4_PCI_DEV_IS_VF) {
3349 mlx4_warn(dev, "Detected virtual function - running in slave mode\n");
3350 dev->flags |= MLX4_FLAG_SLAVE;
3351 } else {
3352
3353
3354
3355 err = mlx4_get_ownership(dev);
3356 if (err) {
3357 if (err < 0)
3358 return err;
3359 else {
3360 mlx4_warn(dev, "Multiple PFs not yet supported - Skipping PF\n");
3361 return -EINVAL;
3362 }
3363 }
3364
3365 atomic_set(&priv->opreq_count, 0);
3366 INIT_WORK(&priv->opreq_task, mlx4_opreq_action);
3367
3368
3369
3370
3371
3372
3373 err = mlx4_reset(dev);
3374 if (err) {
3375 mlx4_err(dev, "Failed to reset HCA, aborting\n");
3376 goto err_sriov;
3377 }
3378
3379 if (total_vfs) {
3380 dev->flags = MLX4_FLAG_MASTER;
3381 existing_vfs = pci_num_vf(pdev);
3382 if (existing_vfs)
3383 dev->flags |= MLX4_FLAG_SRIOV;
3384 dev->persist->num_vfs = total_vfs;
3385 }
3386 }
3387
3388
3389
3390
3391 dev->persist->state = MLX4_DEVICE_STATE_UP;
3392
3393 slave_start:
3394 err = mlx4_cmd_init(dev);
3395 if (err) {
3396 mlx4_err(dev, "Failed to init command interface, aborting\n");
3397 goto err_sriov;
3398 }
3399
3400
3401
3402
3403 if (mlx4_is_mfunc(dev)) {
3404 if (mlx4_is_master(dev)) {
3405 dev->num_slaves = MLX4_MAX_NUM_SLAVES;
3406
3407 } else {
3408 dev->num_slaves = 0;
3409 err = mlx4_multi_func_init(dev);
3410 if (err) {
3411 mlx4_err(dev, "Failed to init slave mfunc interface, aborting\n");
3412 goto err_cmd;
3413 }
3414 }
3415 }
3416
3417 err = mlx4_init_fw(dev);
3418 if (err) {
3419 mlx4_err(dev, "Failed to init fw, aborting.\n");
3420 goto err_mfunc;
3421 }
3422
3423 if (mlx4_is_master(dev)) {
3424
3425 if (!dev_cap) {
3426 dev_cap = kzalloc(sizeof(*dev_cap), GFP_KERNEL);
3427
3428 if (!dev_cap) {
3429 err = -ENOMEM;
3430 goto err_fw;
3431 }
3432
3433 err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
3434 if (err) {
3435 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
3436 goto err_fw;
3437 }
3438
3439 if (mlx4_check_dev_cap(dev, dev_cap, nvfs))
3440 goto err_fw;
3441
3442 if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS)) {
3443 u64 dev_flags = mlx4_enable_sriov(dev, pdev,
3444 total_vfs,
3445 existing_vfs,
3446 reset_flow);
3447
3448 mlx4_close_fw(dev);
3449 mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3450 dev->flags = dev_flags;
3451 if (!SRIOV_VALID_STATE(dev->flags)) {
3452 mlx4_err(dev, "Invalid SRIOV state\n");
3453 goto err_sriov;
3454 }
3455 err = mlx4_reset(dev);
3456 if (err) {
3457 mlx4_err(dev, "Failed to reset HCA, aborting.\n");
3458 goto err_sriov;
3459 }
3460 goto slave_start;
3461 }
3462 } else {
3463
3464
3465
3466
3467 memset(dev_cap, 0, sizeof(*dev_cap));
3468 err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
3469 if (err) {
3470 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
3471 goto err_fw;
3472 }
3473
3474 if (mlx4_check_dev_cap(dev, dev_cap, nvfs))
3475 goto err_fw;
3476 }
3477 }
3478
3479 err = mlx4_init_hca(dev);
3480 if (err) {
3481 if (err == -EACCES) {
3482
3483
3484 mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3485
3486 if (dev->flags & MLX4_FLAG_SRIOV) {
3487 if (!existing_vfs)
3488 pci_disable_sriov(pdev);
3489 if (mlx4_is_master(dev) && !reset_flow)
3490 atomic_dec(&pf_loading);
3491 dev->flags &= ~MLX4_FLAG_SRIOV;
3492 }
3493 if (!mlx4_is_slave(dev))
3494 mlx4_free_ownership(dev);
3495 dev->flags |= MLX4_FLAG_SLAVE;
3496 dev->flags &= ~MLX4_FLAG_MASTER;
3497 goto slave_start;
3498 } else
3499 goto err_fw;
3500 }
3501
3502 if (mlx4_is_master(dev) && (dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS)) {
3503 u64 dev_flags = mlx4_enable_sriov(dev, pdev, total_vfs,
3504 existing_vfs, reset_flow);
3505
3506 if ((dev->flags ^ dev_flags) & (MLX4_FLAG_MASTER | MLX4_FLAG_SLAVE)) {
3507 mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_VHCR);
3508 dev->flags = dev_flags;
3509 err = mlx4_cmd_init(dev);
3510 if (err) {
3511
3512
3513
3514 mlx4_err(dev, "Failed to init VHCR command interface, aborting\n");
3515 goto err_close;
3516 }
3517 } else {
3518 dev->flags = dev_flags;
3519 }
3520
3521 if (!SRIOV_VALID_STATE(dev->flags)) {
3522 mlx4_err(dev, "Invalid SRIOV state\n");
3523 goto err_close;
3524 }
3525 }
3526
3527
3528
3529
3530
3531 if (!mlx4_is_slave(dev))
3532 pcie_print_link_status(dev->persist->pdev);
3533
3534
3535
3536 if (mlx4_is_master(dev)) {
3537 if (dev->caps.num_ports < 2 &&
3538 num_vfs_argc > 1) {
3539 err = -EINVAL;
3540 mlx4_err(dev,
3541 "Error: Trying to configure VFs on port 2, but HCA has only %d physical ports\n",
3542 dev->caps.num_ports);
3543 goto err_close;
3544 }
3545 memcpy(dev->persist->nvfs, nvfs, sizeof(dev->persist->nvfs));
3546
3547 for (i = 0;
3548 i < sizeof(dev->persist->nvfs)/
3549 sizeof(dev->persist->nvfs[0]); i++) {
3550 unsigned j;
3551
3552 for (j = 0; j < dev->persist->nvfs[i]; ++sum, ++j) {
3553 dev->dev_vfs[sum].min_port = i < 2 ? i + 1 : 1;
3554 dev->dev_vfs[sum].n_ports = i < 2 ? 1 :
3555 dev->caps.num_ports;
3556 }
3557 }
3558
3559
3560
3561
3562 err = mlx4_multi_func_init(dev);
3563 if (err) {
3564 mlx4_err(dev, "Failed to init master mfunc interface, aborting.\n");
3565 goto err_close;
3566 }
3567 }
3568
3569 err = mlx4_alloc_eq_table(dev);
3570 if (err)
3571 goto err_master_mfunc;
3572
3573 bitmap_zero(priv->msix_ctl.pool_bm, MAX_MSIX);
3574 mutex_init(&priv->msix_ctl.pool_lock);
3575
3576 mlx4_enable_msi_x(dev);
3577 if ((mlx4_is_mfunc(dev)) &&
3578 !(dev->flags & MLX4_FLAG_MSI_X)) {
3579 err = -EOPNOTSUPP;
3580 mlx4_err(dev, "INTx is not supported in multi-function mode, aborting\n");
3581 goto err_free_eq;
3582 }
3583
3584 if (!mlx4_is_slave(dev)) {
3585 err = mlx4_init_steering(dev);
3586 if (err)
3587 goto err_disable_msix;
3588 }
3589
3590 mlx4_init_quotas(dev);
3591
3592 err = mlx4_setup_hca(dev);
3593 if (err == -EBUSY && (dev->flags & MLX4_FLAG_MSI_X) &&
3594 !mlx4_is_mfunc(dev)) {
3595 dev->flags &= ~MLX4_FLAG_MSI_X;
3596 dev->caps.num_comp_vectors = 1;
3597 pci_disable_msix(pdev);
3598 err = mlx4_setup_hca(dev);
3599 }
3600
3601 if (err)
3602 goto err_steer;
3603
3604
3605
3606
3607 if (mlx4_is_master(dev)) {
3608 err = mlx4_ARM_COMM_CHANNEL(dev);
3609 if (err) {
3610 mlx4_err(dev, " Failed to arm comm channel eq: %x\n",
3611 err);
3612 goto err_steer;
3613 }
3614 }
3615
3616 for (port = 1; port <= dev->caps.num_ports; port++) {
3617 err = mlx4_init_port_info(dev, port);
3618 if (err)
3619 goto err_port;
3620 }
3621
3622 priv->v2p.port1 = 1;
3623 priv->v2p.port2 = 2;
3624
3625 err = mlx4_register_device(dev);
3626 if (err)
3627 goto err_port;
3628
3629 mlx4_request_modules(dev);
3630
3631 mlx4_sense_init(dev);
3632 mlx4_start_sense(dev);
3633
3634 priv->removed = 0;
3635
3636 if (mlx4_is_master(dev) && dev->persist->num_vfs && !reset_flow)
3637 atomic_dec(&pf_loading);
3638
3639 kfree(dev_cap);
3640 return 0;
3641
3642 err_port:
3643 for (--port; port >= 1; --port)
3644 mlx4_cleanup_port_info(&priv->port[port]);
3645
3646 mlx4_cleanup_default_counters(dev);
3647 if (!mlx4_is_slave(dev))
3648 mlx4_cleanup_counters_table(dev);
3649 mlx4_cleanup_qp_table(dev);
3650 mlx4_cleanup_srq_table(dev);
3651 mlx4_cleanup_cq_table(dev);
3652 mlx4_cmd_use_polling(dev);
3653 mlx4_cleanup_eq_table(dev);
3654 mlx4_cleanup_mcg_table(dev);
3655 mlx4_cleanup_mr_table(dev);
3656 mlx4_cleanup_xrcd_table(dev);
3657 mlx4_cleanup_pd_table(dev);
3658 mlx4_cleanup_uar_table(dev);
3659
3660 err_steer:
3661 if (!mlx4_is_slave(dev))
3662 mlx4_clear_steering(dev);
3663
3664 err_disable_msix:
3665 if (dev->flags & MLX4_FLAG_MSI_X)
3666 pci_disable_msix(pdev);
3667
3668 err_free_eq:
3669 mlx4_free_eq_table(dev);
3670
3671 err_master_mfunc:
3672 if (mlx4_is_master(dev)) {
3673 mlx4_free_resource_tracker(dev, RES_TR_FREE_STRUCTS_ONLY);
3674 mlx4_multi_func_cleanup(dev);
3675 }
3676
3677 if (mlx4_is_slave(dev))
3678 mlx4_slave_destroy_special_qp_cap(dev);
3679
3680 err_close:
3681 mlx4_close_hca(dev);
3682
3683 err_fw:
3684 mlx4_close_fw(dev);
3685
3686 err_mfunc:
3687 if (mlx4_is_slave(dev))
3688 mlx4_multi_func_cleanup(dev);
3689
3690 err_cmd:
3691 mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3692
3693 err_sriov:
3694 if (dev->flags & MLX4_FLAG_SRIOV && !existing_vfs) {
3695 pci_disable_sriov(pdev);
3696 dev->flags &= ~MLX4_FLAG_SRIOV;
3697 }
3698
3699 if (mlx4_is_master(dev) && dev->persist->num_vfs && !reset_flow)
3700 atomic_dec(&pf_loading);
3701
3702 kfree(priv->dev.dev_vfs);
3703
3704 if (!mlx4_is_slave(dev))
3705 mlx4_free_ownership(dev);
3706
3707 kfree(dev_cap);
3708 return err;
3709 }
3710
3711 static int __mlx4_init_one(struct pci_dev *pdev, int pci_dev_data,
3712 struct mlx4_priv *priv)
3713 {
3714 int err;
3715 int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
3716 int prb_vf[MLX4_MAX_PORTS + 1] = {0, 0, 0};
3717 const int param_map[MLX4_MAX_PORTS + 1][MLX4_MAX_PORTS + 1] = {
3718 {2, 0, 0}, {0, 1, 2}, {0, 1, 2} };
3719 unsigned total_vfs = 0;
3720 unsigned int i;
3721
3722 pr_info(DRV_NAME ": Initializing %s\n", pci_name(pdev));
3723
3724 err = mlx4_pci_enable_device(&priv->dev);
3725 if (err) {
3726 dev_err(&pdev->dev, "Cannot enable PCI device, aborting\n");
3727 return err;
3728 }
3729
3730
3731
3732
3733
3734 for (i = 0; i < ARRAY_SIZE(nvfs) && i < num_vfs_argc;
3735 total_vfs += nvfs[param_map[num_vfs_argc - 1][i]], i++) {
3736 nvfs[param_map[num_vfs_argc - 1][i]] = num_vfs[i];
3737 if (nvfs[i] < 0) {
3738 dev_err(&pdev->dev, "num_vfs module parameter cannot be negative\n");
3739 err = -EINVAL;
3740 goto err_disable_pdev;
3741 }
3742 }
3743 for (i = 0; i < ARRAY_SIZE(prb_vf) && i < probe_vfs_argc;
3744 i++) {
3745 prb_vf[param_map[probe_vfs_argc - 1][i]] = probe_vf[i];
3746 if (prb_vf[i] < 0 || prb_vf[i] > nvfs[i]) {
3747 dev_err(&pdev->dev, "probe_vf module parameter cannot be negative or greater than num_vfs\n");
3748 err = -EINVAL;
3749 goto err_disable_pdev;
3750 }
3751 }
3752 if (total_vfs > MLX4_MAX_NUM_VF) {
3753 dev_err(&pdev->dev,
3754 "Requested more VF's (%d) than allowed by hw (%d)\n",
3755 total_vfs, MLX4_MAX_NUM_VF);
3756 err = -EINVAL;
3757 goto err_disable_pdev;
3758 }
3759
3760 for (i = 0; i < MLX4_MAX_PORTS; i++) {
3761 if (nvfs[i] + nvfs[2] > MLX4_MAX_NUM_VF_P_PORT) {
3762 dev_err(&pdev->dev,
3763 "Requested more VF's (%d) for port (%d) than allowed by driver (%d)\n",
3764 nvfs[i] + nvfs[2], i + 1,
3765 MLX4_MAX_NUM_VF_P_PORT);
3766 err = -EINVAL;
3767 goto err_disable_pdev;
3768 }
3769 }
3770
3771
3772 if (!(pci_dev_data & MLX4_PCI_DEV_IS_VF) &&
3773 !(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) {
3774 dev_err(&pdev->dev, "Missing DCS, aborting (driver_data: 0x%x, pci_resource_flags(pdev, 0):0x%lx)\n",
3775 pci_dev_data, pci_resource_flags(pdev, 0));
3776 err = -ENODEV;
3777 goto err_disable_pdev;
3778 }
3779 if (!(pci_resource_flags(pdev, 2) & IORESOURCE_MEM)) {
3780 dev_err(&pdev->dev, "Missing UAR, aborting\n");
3781 err = -ENODEV;
3782 goto err_disable_pdev;
3783 }
3784
3785 err = pci_request_regions(pdev, DRV_NAME);
3786 if (err) {
3787 dev_err(&pdev->dev, "Couldn't get PCI resources, aborting\n");
3788 goto err_disable_pdev;
3789 }
3790
3791 pci_set_master(pdev);
3792
3793 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
3794 if (err) {
3795 dev_warn(&pdev->dev, "Warning: couldn't set 64-bit PCI DMA mask\n");
3796 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
3797 if (err) {
3798 dev_err(&pdev->dev, "Can't set PCI DMA mask, aborting\n");
3799 goto err_release_regions;
3800 }
3801 }
3802 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
3803 if (err) {
3804 dev_warn(&pdev->dev, "Warning: couldn't set 64-bit consistent PCI DMA mask\n");
3805 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
3806 if (err) {
3807 dev_err(&pdev->dev, "Can't set consistent PCI DMA mask, aborting\n");
3808 goto err_release_regions;
3809 }
3810 }
3811
3812
3813 dma_set_max_seg_size(&pdev->dev, 1024 * 1024 * 1024);
3814
3815 if (pci_dev_data & MLX4_PCI_DEV_IS_VF) {
3816
3817
3818
3819 if (total_vfs) {
3820 unsigned vfs_offset = 0;
3821
3822 for (i = 0; i < ARRAY_SIZE(nvfs) &&
3823 vfs_offset + nvfs[i] < extended_func_num(pdev);
3824 vfs_offset += nvfs[i], i++)
3825 ;
3826 if (i == ARRAY_SIZE(nvfs)) {
3827 err = -ENODEV;
3828 goto err_release_regions;
3829 }
3830 if ((extended_func_num(pdev) - vfs_offset)
3831 > prb_vf[i]) {
3832 dev_warn(&pdev->dev, "Skipping virtual function:%d\n",
3833 extended_func_num(pdev));
3834 err = -ENODEV;
3835 goto err_release_regions;
3836 }
3837 }
3838 }
3839
3840 err = mlx4_crdump_init(&priv->dev);
3841 if (err)
3842 goto err_release_regions;
3843
3844 err = mlx4_catas_init(&priv->dev);
3845 if (err)
3846 goto err_crdump;
3847
3848 err = mlx4_load_one(pdev, pci_dev_data, total_vfs, nvfs, priv, 0);
3849 if (err)
3850 goto err_catas;
3851
3852 return 0;
3853
3854 err_catas:
3855 mlx4_catas_end(&priv->dev);
3856
3857 err_crdump:
3858 mlx4_crdump_end(&priv->dev);
3859
3860 err_release_regions:
3861 pci_release_regions(pdev);
3862
3863 err_disable_pdev:
3864 mlx4_pci_disable_device(&priv->dev);
3865 return err;
3866 }
3867
3868 static int mlx4_devlink_port_type_set(struct devlink_port *devlink_port,
3869 enum devlink_port_type port_type)
3870 {
3871 struct mlx4_port_info *info = container_of(devlink_port,
3872 struct mlx4_port_info,
3873 devlink_port);
3874 enum mlx4_port_type mlx4_port_type;
3875
3876 switch (port_type) {
3877 case DEVLINK_PORT_TYPE_AUTO:
3878 mlx4_port_type = MLX4_PORT_TYPE_AUTO;
3879 break;
3880 case DEVLINK_PORT_TYPE_ETH:
3881 mlx4_port_type = MLX4_PORT_TYPE_ETH;
3882 break;
3883 case DEVLINK_PORT_TYPE_IB:
3884 mlx4_port_type = MLX4_PORT_TYPE_IB;
3885 break;
3886 default:
3887 return -EOPNOTSUPP;
3888 }
3889
3890 return __set_port_type(info, mlx4_port_type);
3891 }
3892
3893 static void mlx4_devlink_param_load_driverinit_values(struct devlink *devlink)
3894 {
3895 struct mlx4_priv *priv = devlink_priv(devlink);
3896 struct mlx4_dev *dev = &priv->dev;
3897 struct mlx4_fw_crdump *crdump = &dev->persist->crdump;
3898 union devlink_param_value saved_value;
3899 int err;
3900
3901 err = devlink_param_driverinit_value_get(devlink,
3902 DEVLINK_PARAM_GENERIC_ID_INT_ERR_RESET,
3903 &saved_value);
3904 if (!err && mlx4_internal_err_reset != saved_value.vbool) {
3905 mlx4_internal_err_reset = saved_value.vbool;
3906
3907 devlink_param_value_changed(devlink,
3908 DEVLINK_PARAM_GENERIC_ID_INT_ERR_RESET);
3909 }
3910 err = devlink_param_driverinit_value_get(devlink,
3911 DEVLINK_PARAM_GENERIC_ID_MAX_MACS,
3912 &saved_value);
3913 if (!err)
3914 log_num_mac = order_base_2(saved_value.vu32);
3915 err = devlink_param_driverinit_value_get(devlink,
3916 MLX4_DEVLINK_PARAM_ID_ENABLE_64B_CQE_EQE,
3917 &saved_value);
3918 if (!err)
3919 enable_64b_cqe_eqe = saved_value.vbool;
3920 err = devlink_param_driverinit_value_get(devlink,
3921 MLX4_DEVLINK_PARAM_ID_ENABLE_4K_UAR,
3922 &saved_value);
3923 if (!err)
3924 enable_4k_uar = saved_value.vbool;
3925 err = devlink_param_driverinit_value_get(devlink,
3926 DEVLINK_PARAM_GENERIC_ID_REGION_SNAPSHOT,
3927 &saved_value);
3928 if (!err && crdump->snapshot_enable != saved_value.vbool) {
3929 crdump->snapshot_enable = saved_value.vbool;
3930 devlink_param_value_changed(devlink,
3931 DEVLINK_PARAM_GENERIC_ID_REGION_SNAPSHOT);
3932 }
3933 }
3934
3935 static void mlx4_restart_one_down(struct pci_dev *pdev);
3936 static int mlx4_restart_one_up(struct pci_dev *pdev, bool reload,
3937 struct devlink *devlink);
3938
3939 static int mlx4_devlink_reload_down(struct devlink *devlink,
3940 struct netlink_ext_ack *extack)
3941 {
3942 struct mlx4_priv *priv = devlink_priv(devlink);
3943 struct mlx4_dev *dev = &priv->dev;
3944 struct mlx4_dev_persistent *persist = dev->persist;
3945
3946 if (persist->num_vfs)
3947 mlx4_warn(persist->dev, "Reload performed on PF, will cause reset on operating Virtual Functions\n");
3948 mlx4_restart_one_down(persist->pdev);
3949 return 0;
3950 }
3951
3952 static int mlx4_devlink_reload_up(struct devlink *devlink,
3953 struct netlink_ext_ack *extack)
3954 {
3955 struct mlx4_priv *priv = devlink_priv(devlink);
3956 struct mlx4_dev *dev = &priv->dev;
3957 struct mlx4_dev_persistent *persist = dev->persist;
3958 int err;
3959
3960 err = mlx4_restart_one_up(persist->pdev, true, devlink);
3961 if (err)
3962 mlx4_err(persist->dev, "mlx4_restart_one_up failed, ret=%d\n",
3963 err);
3964
3965 return err;
3966 }
3967
3968 static const struct devlink_ops mlx4_devlink_ops = {
3969 .port_type_set = mlx4_devlink_port_type_set,
3970 .reload_down = mlx4_devlink_reload_down,
3971 .reload_up = mlx4_devlink_reload_up,
3972 };
3973
3974 static int mlx4_init_one(struct pci_dev *pdev, const struct pci_device_id *id)
3975 {
3976 struct devlink *devlink;
3977 struct mlx4_priv *priv;
3978 struct mlx4_dev *dev;
3979 int ret;
3980
3981 printk_once(KERN_INFO "%s", mlx4_version);
3982
3983 devlink = devlink_alloc(&mlx4_devlink_ops, sizeof(*priv));
3984 if (!devlink)
3985 return -ENOMEM;
3986 priv = devlink_priv(devlink);
3987
3988 dev = &priv->dev;
3989 dev->persist = kzalloc(sizeof(*dev->persist), GFP_KERNEL);
3990 if (!dev->persist) {
3991 ret = -ENOMEM;
3992 goto err_devlink_free;
3993 }
3994 dev->persist->pdev = pdev;
3995 dev->persist->dev = dev;
3996 pci_set_drvdata(pdev, dev->persist);
3997 priv->pci_dev_data = id->driver_data;
3998 mutex_init(&dev->persist->device_state_mutex);
3999 mutex_init(&dev->persist->interface_state_mutex);
4000 mutex_init(&dev->persist->pci_status_mutex);
4001
4002 ret = devlink_register(devlink, &pdev->dev);
4003 if (ret)
4004 goto err_persist_free;
4005 ret = devlink_params_register(devlink, mlx4_devlink_params,
4006 ARRAY_SIZE(mlx4_devlink_params));
4007 if (ret)
4008 goto err_devlink_unregister;
4009 mlx4_devlink_set_params_init_values(devlink);
4010 ret = __mlx4_init_one(pdev, id->driver_data, priv);
4011 if (ret)
4012 goto err_params_unregister;
4013
4014 devlink_params_publish(devlink);
4015 devlink_reload_enable(devlink);
4016 pci_save_state(pdev);
4017 return 0;
4018
4019 err_params_unregister:
4020 devlink_params_unregister(devlink, mlx4_devlink_params,
4021 ARRAY_SIZE(mlx4_devlink_params));
4022 err_devlink_unregister:
4023 devlink_unregister(devlink);
4024 err_persist_free:
4025 kfree(dev->persist);
4026 err_devlink_free:
4027 devlink_free(devlink);
4028 return ret;
4029 }
4030
4031 static void mlx4_clean_dev(struct mlx4_dev *dev)
4032 {
4033 struct mlx4_dev_persistent *persist = dev->persist;
4034 struct mlx4_priv *priv = mlx4_priv(dev);
4035 unsigned long flags = (dev->flags & RESET_PERSIST_MASK_FLAGS);
4036
4037 memset(priv, 0, sizeof(*priv));
4038 priv->dev.persist = persist;
4039 priv->dev.flags = flags;
4040 }
4041
4042 static void mlx4_unload_one(struct pci_dev *pdev)
4043 {
4044 struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4045 struct mlx4_dev *dev = persist->dev;
4046 struct mlx4_priv *priv = mlx4_priv(dev);
4047 int pci_dev_data;
4048 int p, i;
4049
4050 if (priv->removed)
4051 return;
4052
4053
4054 for (i = 0; i < dev->caps.num_ports; i++) {
4055 dev->persist->curr_port_type[i] = dev->caps.port_type[i + 1];
4056 dev->persist->curr_port_poss_type[i] = dev->caps.
4057 possible_type[i + 1];
4058 }
4059
4060 pci_dev_data = priv->pci_dev_data;
4061
4062 mlx4_stop_sense(dev);
4063 mlx4_unregister_device(dev);
4064
4065 for (p = 1; p <= dev->caps.num_ports; p++) {
4066 mlx4_cleanup_port_info(&priv->port[p]);
4067 mlx4_CLOSE_PORT(dev, p);
4068 }
4069
4070 if (mlx4_is_master(dev))
4071 mlx4_free_resource_tracker(dev,
4072 RES_TR_FREE_SLAVES_ONLY);
4073
4074 mlx4_cleanup_default_counters(dev);
4075 if (!mlx4_is_slave(dev))
4076 mlx4_cleanup_counters_table(dev);
4077 mlx4_cleanup_qp_table(dev);
4078 mlx4_cleanup_srq_table(dev);
4079 mlx4_cleanup_cq_table(dev);
4080 mlx4_cmd_use_polling(dev);
4081 mlx4_cleanup_eq_table(dev);
4082 mlx4_cleanup_mcg_table(dev);
4083 mlx4_cleanup_mr_table(dev);
4084 mlx4_cleanup_xrcd_table(dev);
4085 mlx4_cleanup_pd_table(dev);
4086
4087 if (mlx4_is_master(dev))
4088 mlx4_free_resource_tracker(dev,
4089 RES_TR_FREE_STRUCTS_ONLY);
4090
4091 iounmap(priv->kar);
4092 mlx4_uar_free(dev, &priv->driver_uar);
4093 mlx4_cleanup_uar_table(dev);
4094 if (!mlx4_is_slave(dev))
4095 mlx4_clear_steering(dev);
4096 mlx4_free_eq_table(dev);
4097 if (mlx4_is_master(dev))
4098 mlx4_multi_func_cleanup(dev);
4099 mlx4_close_hca(dev);
4100 mlx4_close_fw(dev);
4101 if (mlx4_is_slave(dev))
4102 mlx4_multi_func_cleanup(dev);
4103 mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
4104
4105 if (dev->flags & MLX4_FLAG_MSI_X)
4106 pci_disable_msix(pdev);
4107
4108 if (!mlx4_is_slave(dev))
4109 mlx4_free_ownership(dev);
4110
4111 mlx4_slave_destroy_special_qp_cap(dev);
4112 kfree(dev->dev_vfs);
4113
4114 mlx4_clean_dev(dev);
4115 priv->pci_dev_data = pci_dev_data;
4116 priv->removed = 1;
4117 }
4118
4119 static void mlx4_remove_one(struct pci_dev *pdev)
4120 {
4121 struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4122 struct mlx4_dev *dev = persist->dev;
4123 struct mlx4_priv *priv = mlx4_priv(dev);
4124 struct devlink *devlink = priv_to_devlink(priv);
4125 int active_vfs = 0;
4126
4127 devlink_reload_disable(devlink);
4128
4129 if (mlx4_is_slave(dev))
4130 persist->interface_state |= MLX4_INTERFACE_STATE_NOWAIT;
4131
4132 mutex_lock(&persist->interface_state_mutex);
4133 persist->interface_state |= MLX4_INTERFACE_STATE_DELETION;
4134 mutex_unlock(&persist->interface_state_mutex);
4135
4136
4137 if (mlx4_is_master(dev) && dev->flags & MLX4_FLAG_SRIOV) {
4138 active_vfs = mlx4_how_many_lives_vf(dev);
4139 if (active_vfs) {
4140 pr_warn("Removing PF when there are active VF's !!\n");
4141 pr_warn("Will not disable SR-IOV.\n");
4142 }
4143 }
4144
4145
4146
4147
4148 if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
4149 mlx4_unload_one(pdev);
4150 else
4151 mlx4_info(dev, "%s: interface is down\n", __func__);
4152 mlx4_catas_end(dev);
4153 mlx4_crdump_end(dev);
4154 if (dev->flags & MLX4_FLAG_SRIOV && !active_vfs) {
4155 mlx4_warn(dev, "Disabling SR-IOV\n");
4156 pci_disable_sriov(pdev);
4157 }
4158
4159 pci_release_regions(pdev);
4160 mlx4_pci_disable_device(dev);
4161 devlink_params_unregister(devlink, mlx4_devlink_params,
4162 ARRAY_SIZE(mlx4_devlink_params));
4163 devlink_unregister(devlink);
4164 kfree(dev->persist);
4165 devlink_free(devlink);
4166 }
4167
4168 static int restore_current_port_types(struct mlx4_dev *dev,
4169 enum mlx4_port_type *types,
4170 enum mlx4_port_type *poss_types)
4171 {
4172 struct mlx4_priv *priv = mlx4_priv(dev);
4173 int err, i;
4174
4175 mlx4_stop_sense(dev);
4176
4177 mutex_lock(&priv->port_mutex);
4178 for (i = 0; i < dev->caps.num_ports; i++)
4179 dev->caps.possible_type[i + 1] = poss_types[i];
4180 err = mlx4_change_port_types(dev, types);
4181 mlx4_start_sense(dev);
4182 mutex_unlock(&priv->port_mutex);
4183
4184 return err;
4185 }
4186
4187 static void mlx4_restart_one_down(struct pci_dev *pdev)
4188 {
4189 mlx4_unload_one(pdev);
4190 }
4191
4192 static int mlx4_restart_one_up(struct pci_dev *pdev, bool reload,
4193 struct devlink *devlink)
4194 {
4195 struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4196 struct mlx4_dev *dev = persist->dev;
4197 struct mlx4_priv *priv = mlx4_priv(dev);
4198 int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
4199 int pci_dev_data, err, total_vfs;
4200
4201 pci_dev_data = priv->pci_dev_data;
4202 total_vfs = dev->persist->num_vfs;
4203 memcpy(nvfs, dev->persist->nvfs, sizeof(dev->persist->nvfs));
4204
4205 if (reload)
4206 mlx4_devlink_param_load_driverinit_values(devlink);
4207 err = mlx4_load_one(pdev, pci_dev_data, total_vfs, nvfs, priv, 1);
4208 if (err) {
4209 mlx4_err(dev, "%s: ERROR: mlx4_load_one failed, pci_name=%s, err=%d\n",
4210 __func__, pci_name(pdev), err);
4211 return err;
4212 }
4213
4214 err = restore_current_port_types(dev, dev->persist->curr_port_type,
4215 dev->persist->curr_port_poss_type);
4216 if (err)
4217 mlx4_err(dev, "could not restore original port types (%d)\n",
4218 err);
4219
4220 return err;
4221 }
4222
4223 int mlx4_restart_one(struct pci_dev *pdev)
4224 {
4225 mlx4_restart_one_down(pdev);
4226 return mlx4_restart_one_up(pdev, false, NULL);
4227 }
4228
4229 #define MLX_SP(id) { PCI_VDEVICE(MELLANOX, id), MLX4_PCI_DEV_FORCE_SENSE_PORT }
4230 #define MLX_VF(id) { PCI_VDEVICE(MELLANOX, id), MLX4_PCI_DEV_IS_VF }
4231 #define MLX_GN(id) { PCI_VDEVICE(MELLANOX, id), 0 }
4232
4233 static const struct pci_device_id mlx4_pci_table[] = {
4234 #ifdef CONFIG_MLX4_CORE_GEN2
4235
4236 MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_SDR),
4237 MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_DDR),
4238 MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_QDR),
4239 MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_DDR_GEN2),
4240 MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_QDR_GEN2),
4241 MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_EN),
4242 MLX_SP(PCI_DEVICE_ID_MELLANOX_HERMON_EN_GEN2),
4243
4244 MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX_EN),
4245 MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX_EN_T_GEN2),
4246
4247 MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX_EN_GEN2),
4248
4249 MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX_EN_5_GEN2),
4250
4251 MLX_SP(PCI_DEVICE_ID_MELLANOX_CONNECTX2),
4252
4253 MLX_VF(0x1002),
4254 #endif
4255
4256 MLX_GN(PCI_DEVICE_ID_MELLANOX_CONNECTX3),
4257 MLX_VF(0x1004),
4258 MLX_GN(0x1005),
4259 MLX_GN(0x1006),
4260 MLX_GN(PCI_DEVICE_ID_MELLANOX_CONNECTX3_PRO),
4261 MLX_GN(0x1008),
4262 MLX_GN(0x1009),
4263 MLX_GN(0x100a),
4264 MLX_GN(0x100b),
4265 MLX_GN(0x100c),
4266 MLX_GN(0x100d),
4267 MLX_GN(0x100e),
4268 MLX_GN(0x100f),
4269 MLX_GN(0x1010),
4270
4271
4272
4273
4274
4275
4276 { 0, }
4277 };
4278
4279 MODULE_DEVICE_TABLE(pci, mlx4_pci_table);
4280
4281 static pci_ers_result_t mlx4_pci_err_detected(struct pci_dev *pdev,
4282 pci_channel_state_t state)
4283 {
4284 struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4285
4286 mlx4_err(persist->dev, "mlx4_pci_err_detected was called\n");
4287 mlx4_enter_error_state(persist);
4288
4289 mutex_lock(&persist->interface_state_mutex);
4290 if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
4291 mlx4_unload_one(pdev);
4292
4293 mutex_unlock(&persist->interface_state_mutex);
4294 if (state == pci_channel_io_perm_failure)
4295 return PCI_ERS_RESULT_DISCONNECT;
4296
4297 mlx4_pci_disable_device(persist->dev);
4298 return PCI_ERS_RESULT_NEED_RESET;
4299 }
4300
4301 static pci_ers_result_t mlx4_pci_slot_reset(struct pci_dev *pdev)
4302 {
4303 struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4304 struct mlx4_dev *dev = persist->dev;
4305 int err;
4306
4307 mlx4_err(dev, "mlx4_pci_slot_reset was called\n");
4308 err = mlx4_pci_enable_device(dev);
4309 if (err) {
4310 mlx4_err(dev, "Can not re-enable device, err=%d\n", err);
4311 return PCI_ERS_RESULT_DISCONNECT;
4312 }
4313
4314 pci_set_master(pdev);
4315 pci_restore_state(pdev);
4316 pci_save_state(pdev);
4317 return PCI_ERS_RESULT_RECOVERED;
4318 }
4319
4320 static void mlx4_pci_resume(struct pci_dev *pdev)
4321 {
4322 struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4323 struct mlx4_dev *dev = persist->dev;
4324 struct mlx4_priv *priv = mlx4_priv(dev);
4325 int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
4326 int total_vfs;
4327 int err;
4328
4329 mlx4_err(dev, "%s was called\n", __func__);
4330 total_vfs = dev->persist->num_vfs;
4331 memcpy(nvfs, dev->persist->nvfs, sizeof(dev->persist->nvfs));
4332
4333 mutex_lock(&persist->interface_state_mutex);
4334 if (!(persist->interface_state & MLX4_INTERFACE_STATE_UP)) {
4335 err = mlx4_load_one(pdev, priv->pci_dev_data, total_vfs, nvfs,
4336 priv, 1);
4337 if (err) {
4338 mlx4_err(dev, "%s: mlx4_load_one failed, err=%d\n",
4339 __func__, err);
4340 goto end;
4341 }
4342
4343 err = restore_current_port_types(dev, dev->persist->
4344 curr_port_type, dev->persist->
4345 curr_port_poss_type);
4346 if (err)
4347 mlx4_err(dev, "could not restore original port types (%d)\n", err);
4348 }
4349 end:
4350 mutex_unlock(&persist->interface_state_mutex);
4351
4352 }
4353
4354 static void mlx4_shutdown(struct pci_dev *pdev)
4355 {
4356 struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4357
4358 mlx4_info(persist->dev, "mlx4_shutdown was called\n");
4359 mutex_lock(&persist->interface_state_mutex);
4360 if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
4361 mlx4_unload_one(pdev);
4362 mutex_unlock(&persist->interface_state_mutex);
4363 }
4364
4365 static const struct pci_error_handlers mlx4_err_handler = {
4366 .error_detected = mlx4_pci_err_detected,
4367 .slot_reset = mlx4_pci_slot_reset,
4368 .resume = mlx4_pci_resume,
4369 };
4370
4371 static int mlx4_suspend(struct pci_dev *pdev, pm_message_t state)
4372 {
4373 struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4374 struct mlx4_dev *dev = persist->dev;
4375
4376 mlx4_err(dev, "suspend was called\n");
4377 mutex_lock(&persist->interface_state_mutex);
4378 if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
4379 mlx4_unload_one(pdev);
4380 mutex_unlock(&persist->interface_state_mutex);
4381
4382 return 0;
4383 }
4384
4385 static int mlx4_resume(struct pci_dev *pdev)
4386 {
4387 struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
4388 struct mlx4_dev *dev = persist->dev;
4389 struct mlx4_priv *priv = mlx4_priv(dev);
4390 int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
4391 int total_vfs;
4392 int ret = 0;
4393
4394 mlx4_err(dev, "resume was called\n");
4395 total_vfs = dev->persist->num_vfs;
4396 memcpy(nvfs, dev->persist->nvfs, sizeof(dev->persist->nvfs));
4397
4398 mutex_lock(&persist->interface_state_mutex);
4399 if (!(persist->interface_state & MLX4_INTERFACE_STATE_UP)) {
4400 ret = mlx4_load_one(pdev, priv->pci_dev_data, total_vfs,
4401 nvfs, priv, 1);
4402 if (!ret) {
4403 ret = restore_current_port_types(dev,
4404 dev->persist->curr_port_type,
4405 dev->persist->curr_port_poss_type);
4406 if (ret)
4407 mlx4_err(dev, "resume: could not restore original port types (%d)\n", ret);
4408 }
4409 }
4410 mutex_unlock(&persist->interface_state_mutex);
4411
4412 return ret;
4413 }
4414
4415 static struct pci_driver mlx4_driver = {
4416 .name = DRV_NAME,
4417 .id_table = mlx4_pci_table,
4418 .probe = mlx4_init_one,
4419 .shutdown = mlx4_shutdown,
4420 .remove = mlx4_remove_one,
4421 .suspend = mlx4_suspend,
4422 .resume = mlx4_resume,
4423 .err_handler = &mlx4_err_handler,
4424 };
4425
4426 static int __init mlx4_verify_params(void)
4427 {
4428 if (msi_x < 0) {
4429 pr_warn("mlx4_core: bad msi_x: %d\n", msi_x);
4430 return -1;
4431 }
4432
4433 if ((log_num_mac < 0) || (log_num_mac > 7)) {
4434 pr_warn("mlx4_core: bad num_mac: %d\n", log_num_mac);
4435 return -1;
4436 }
4437
4438 if (log_num_vlan != 0)
4439 pr_warn("mlx4_core: log_num_vlan - obsolete module param, using %d\n",
4440 MLX4_LOG_NUM_VLANS);
4441
4442 if (use_prio != 0)
4443 pr_warn("mlx4_core: use_prio - obsolete module param, ignored\n");
4444
4445 if ((log_mtts_per_seg < 0) || (log_mtts_per_seg > 7)) {
4446 pr_warn("mlx4_core: bad log_mtts_per_seg: %d\n",
4447 log_mtts_per_seg);
4448 return -1;
4449 }
4450
4451
4452 if (port_type_array[0] == false && port_type_array[1] == true) {
4453 pr_warn("Module parameter configuration ETH/IB is not supported. Switching to default configuration IB/IB\n");
4454 port_type_array[0] = true;
4455 }
4456
4457 if (mlx4_log_num_mgm_entry_size < -7 ||
4458 (mlx4_log_num_mgm_entry_size > 0 &&
4459 (mlx4_log_num_mgm_entry_size < MLX4_MIN_MGM_LOG_ENTRY_SIZE ||
4460 mlx4_log_num_mgm_entry_size > MLX4_MAX_MGM_LOG_ENTRY_SIZE))) {
4461 pr_warn("mlx4_core: mlx4_log_num_mgm_entry_size (%d) not in legal range (-7..0 or %d..%d)\n",
4462 mlx4_log_num_mgm_entry_size,
4463 MLX4_MIN_MGM_LOG_ENTRY_SIZE,
4464 MLX4_MAX_MGM_LOG_ENTRY_SIZE);
4465 return -1;
4466 }
4467
4468 return 0;
4469 }
4470
4471 static int __init mlx4_init(void)
4472 {
4473 int ret;
4474
4475 if (mlx4_verify_params())
4476 return -EINVAL;
4477
4478
4479 mlx4_wq = create_singlethread_workqueue("mlx4");
4480 if (!mlx4_wq)
4481 return -ENOMEM;
4482
4483 ret = pci_register_driver(&mlx4_driver);
4484 if (ret < 0)
4485 destroy_workqueue(mlx4_wq);
4486 return ret < 0 ? ret : 0;
4487 }
4488
4489 static void __exit mlx4_cleanup(void)
4490 {
4491 pci_unregister_driver(&mlx4_driver);
4492 destroy_workqueue(mlx4_wq);
4493 }
4494
4495 module_init(mlx4_init);
4496 module_exit(mlx4_cleanup);