1 /*
2 * Copyright (c) 2006, 2007 Cisco Systems, Inc. All rights reserved.
3 * Copyright (c) 2007, 2008 Mellanox Technologies. All rights reserved.
4 *
5 * This software is available to you under a choice of one of two
6 * licenses. You may choose to be licensed under the terms of the GNU
7 * General Public License (GPL) Version 2, available from the file
8 * COPYING in the main directory of this source tree, or the
9 * OpenIB.org BSD license below:
10 *
11 * Redistribution and use in source and binary forms, with or
12 * without modification, are permitted provided that the following
13 * conditions are met:
14 *
15 * - Redistributions of source code must retain the above
16 * copyright notice, this list of conditions and the following
17 * disclaimer.
18 *
19 * - Redistributions in binary form must reproduce the above
20 * copyright notice, this list of conditions and the following
21 * disclaimer in the documentation and/or other materials
22 * provided with the distribution.
23 *
24 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
25 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
26 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
27 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
28 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
29 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
30 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
31 * SOFTWARE.
32 */
33
34 #include <linux/module.h>
35 #include <linux/slab.h>
36 #include <linux/errno.h>
37 #include <linux/netdevice.h>
38 #include <linux/inetdevice.h>
39 #include <linux/if_vlan.h>
40 #include <linux/fs.h>
41 #include <net/ipv6.h>
42
43 #include <rdma/ib_smi.h>
44 #include <rdma/ib_user_verbs.h>
45 #include <rdma/ib_user_verbs_exp.h>
46 #include <rdma/ib_addr.h>
47
48 #include <linux/mlx4/driver.h>
49 #include <linux/mlx4/cmd.h>
50 #include <linux/sched.h>
51 #include <linux/page.h>
52 #include <linux/printk.h>
53 #include "mlx4_ib.h"
54 #include "mlx4_exp.h"
55 #include "user.h"
56 #include "wc.h"
57
58 #define DRV_NAME MLX4_IB_DRV_NAME
59 #define DRV_VERSION "1.0"
60 #define DRV_RELDATE __DATE__
61
62 #define MLX4_IB_DRIVER_PROC_DIR_NAME "driver/mlx4_ib"
63 #define MLX4_IB_MRS_PROC_DIR_NAME "mrs"
64 #define MLX4_IB_FLOW_MAX_PRIO 0xFFF
65 #define MLX4_IB_FLOW_QPN_MASK 0xFFFFFF
66
67 MODULE_AUTHOR("Roland Dreier");
68 MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
69 MODULE_LICENSE("Dual BSD/GPL");
70 #ifdef __linux__
71 MODULE_VERSION(DRV_VERSION);
72 #endif
73
74 int mlx4_ib_sm_guid_assign = 1;
75
76 module_param_named(sm_guid_assign, mlx4_ib_sm_guid_assign, int, 0444);
77 MODULE_PARM_DESC(sm_guid_assign, "Enable SM alias_GUID assignment if sm_guid_assign > 0 (Default: 1)");
78
79 enum {
80 MAX_NUM_STR_BITMAP = 1 << 15,
81 DEFAULT_TBL_VAL = -1
82 };
83
84 static struct mlx4_dbdf2val_lst dev_assign_str = {
85 .name = "dev_assign_str param",
86 .num_vals = 1,
87 .def_val = {DEFAULT_TBL_VAL},
88 .range = {0, MAX_NUM_STR_BITMAP - 1}
89 };
90 module_param_string(dev_assign_str, dev_assign_str.str,
91 sizeof(dev_assign_str.str), 0444);
92 MODULE_PARM_DESC(dev_assign_str,
93 "Map device function numbers to IB device numbers (e.g. '0000:04:00.0-0,002b:1c:0b.a-1,...').\n"
94 "\t\tHexadecimal digits for the device function (e.g. 002b:1c:0b.a) and decimal for IB device numbers (e.g. 1).\n"
95 "\t\tMax supported devices - 32");
96
97
98 static unsigned long *dev_num_str_bitmap;
99 static spinlock_t dev_num_str_lock;
100
101 static const char mlx4_ib_version[] =
102 DRV_NAME ": Mellanox ConnectX InfiniBand driver v"
103 DRV_VERSION " (" DRV_RELDATE ")\n";
104
105 struct update_gid_work {
106 struct work_struct work;
107 union ib_gid gids[128];
108 struct mlx4_ib_dev *dev;
109 int port;
110 };
111
112 struct dev_rec {
113 int bus;
114 int dev;
115 int func;
116 int nr;
117 };
118
119 static int dr_active;
120
121 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init);
122
123 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev, struct net_device*,
124 unsigned long);
125
126 static u8 mlx4_ib_get_dev_port(struct net_device *dev,
127 struct mlx4_ib_dev *ibdev);
128
129 static struct workqueue_struct *wq;
130
init_query_mad(struct ib_smp * mad)131 static void init_query_mad(struct ib_smp *mad)
132 {
133 mad->base_version = 1;
134 mad->mgmt_class = IB_MGMT_CLASS_SUBN_LID_ROUTED;
135 mad->class_version = 1;
136 mad->method = IB_MGMT_METHOD_GET;
137 }
138
139 static union ib_gid zgid;
140
check_flow_steering_support(struct mlx4_dev * dev)141 static int check_flow_steering_support(struct mlx4_dev *dev)
142 {
143 int eth_num_ports = 0;
144 int ib_num_ports = 0;
145 int dmfs = dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED;
146
147 if (dmfs) {
148 int i;
149 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH)
150 eth_num_ports++;
151 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
152 ib_num_ports++;
153 dmfs &= (!ib_num_ports ||
154 (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB)) &&
155 (!eth_num_ports ||
156 (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN));
157 if (ib_num_ports && mlx4_is_mfunc(dev)) {
158 dmfs = 0;
159 }
160 }
161 return dmfs;
162 }
163
mlx4_ib_query_device(struct ib_device * ibdev,struct ib_device_attr * props)164 int mlx4_ib_query_device(struct ib_device *ibdev,
165 struct ib_device_attr *props)
166 {
167 struct mlx4_ib_dev *dev = to_mdev(ibdev);
168 struct ib_smp *in_mad = NULL;
169 struct ib_smp *out_mad = NULL;
170 int err = -ENOMEM;
171
172 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
173 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
174 if (!in_mad || !out_mad)
175 goto out;
176
177 init_query_mad(in_mad);
178 in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
179
180 err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
181 1, NULL, NULL, in_mad, out_mad);
182 if (err)
183 goto out;
184
185 memset(props, 0, sizeof *props);
186
187 props->fw_ver = dev->dev->caps.fw_ver;
188 props->device_cap_flags = IB_DEVICE_CHANGE_PHY_PORT |
189 IB_DEVICE_PORT_ACTIVE_EVENT |
190 IB_DEVICE_SYS_IMAGE_GUID |
191 IB_DEVICE_RC_RNR_NAK_GEN |
192 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK |
193 IB_DEVICE_SHARED_MR;
194
195 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
196 props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
197 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
198 props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
199 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM)
200 props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
201 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
202 props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
203 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
204 props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
205 if (dev->dev->caps.max_gso_sz && dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH)
206 props->device_cap_flags |= IB_DEVICE_UD_TSO;
207 if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
208 props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
209 if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
210 (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
211 (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
212 props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
213 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
214 props->device_cap_flags |= IB_DEVICE_XRC;
215 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_CROSS_CHANNEL)
216 props->device_cap_flags |= IB_DEVICE_CROSS_CHANNEL;
217
218 if (check_flow_steering_support(dev->dev))
219 props->device_cap_flags |= IB_DEVICE_MANAGED_FLOW_STEERING;
220
221
222 props->device_cap_flags |= IB_DEVICE_QPG;
223 if (dev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS) {
224 props->device_cap_flags |= IB_DEVICE_UD_RSS;
225 props->max_rss_tbl_sz = dev->dev->caps.max_rss_tbl_sz;
226 }
227 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW)
228 props->device_cap_flags |= IB_DEVICE_MEM_WINDOW;
229 if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
230 if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_WIN_TYPE_2B)
231 props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B;
232 else
233 props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2A;
234 }
235 props->vendor_id = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
236 0xffffff;
237 props->vendor_part_id = dev->dev->pdev->device;
238 props->hw_ver = be32_to_cpup((__be32 *) (out_mad->data + 32));
239 memcpy(&props->sys_image_guid, out_mad->data + 4, 8);
240
241 props->max_mr_size = ~0ull;
242 props->page_size_cap = dev->dev->caps.page_size_cap;
243 props->max_qp = dev->dev->quotas.qp;
244 props->max_qp_wr = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
245 props->max_sge = min(dev->dev->caps.max_sq_sg,
246 dev->dev->caps.max_rq_sg);
247 props->max_cq = dev->dev->quotas.cq;
248 props->max_cqe = dev->dev->caps.max_cqes;
249 props->max_mr = dev->dev->quotas.mpt;
250 props->max_pd = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
251 props->max_qp_rd_atom = dev->dev->caps.max_qp_dest_rdma;
252 props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
253 props->max_res_rd_atom = props->max_qp_rd_atom * props->max_qp;
254 props->max_srq = dev->dev->quotas.srq;
255 props->max_srq_wr = dev->dev->caps.max_srq_wqes - 1;
256 props->max_srq_sge = dev->dev->caps.max_srq_sge;
257 props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
258 props->local_ca_ack_delay = dev->dev->caps.local_ca_ack_delay;
259 props->atomic_cap = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
260 IB_ATOMIC_HCA : IB_ATOMIC_NONE;
261 props->masked_atomic_cap = props->atomic_cap;
262 props->max_pkeys = dev->dev->caps.pkey_table_len[1];
263 props->max_mcast_grp = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
264 props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
265 props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
266 props->max_mcast_grp;
267 props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
268 props->hca_core_clock = dev->dev->caps.hca_core_clock;
269 if (dev->dev->caps.hca_core_clock > 0)
270 props->comp_mask |= IB_DEVICE_ATTR_WITH_HCA_CORE_CLOCK;
271 if (dev->dev->caps.cq_timestamp) {
272 props->timestamp_mask = 0xFFFFFFFFFFFF;
273 props->comp_mask |= IB_DEVICE_ATTR_WITH_TIMESTAMP_MASK;
274 }
275
276 out:
277 kfree(in_mad);
278 kfree(out_mad);
279
280 return err;
281 }
282
283 static enum rdma_link_layer
mlx4_ib_port_link_layer(struct ib_device * device,u8 port_num)284 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
285 {
286 struct mlx4_dev *dev = to_mdev(device)->dev;
287
288 return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
289 IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
290 }
291
ib_link_query_port(struct ib_device * ibdev,u8 port,struct ib_port_attr * props,int netw_view)292 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
293 struct ib_port_attr *props, int netw_view)
294 {
295 struct ib_smp *in_mad = NULL;
296 struct ib_smp *out_mad = NULL;
297 int ext_active_speed;
298 int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
299 int err = -ENOMEM;
300
301 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
302 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
303 if (!in_mad || !out_mad)
304 goto out;
305
306 init_query_mad(in_mad);
307 in_mad->attr_id = IB_SMP_ATTR_PORT_INFO;
308 in_mad->attr_mod = cpu_to_be32(port);
309
310 if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
311 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
312
313 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
314 in_mad, out_mad);
315 if (err)
316 goto out;
317
318
319 props->lid = be16_to_cpup((__be16 *) (out_mad->data + 16));
320 props->lmc = out_mad->data[34] & 0x7;
321 props->sm_lid = be16_to_cpup((__be16 *) (out_mad->data + 18));
322 props->sm_sl = out_mad->data[36] & 0xf;
323 props->state = out_mad->data[32] & 0xf;
324 props->phys_state = out_mad->data[33] >> 4;
325 props->port_cap_flags = be32_to_cpup((__be32 *) (out_mad->data + 20));
326 if (netw_view)
327 props->gid_tbl_len = out_mad->data[50];
328 else
329 props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
330 props->max_msg_sz = to_mdev(ibdev)->dev->caps.max_msg_sz;
331 props->pkey_tbl_len = to_mdev(ibdev)->dev->caps.pkey_table_len[port];
332 props->bad_pkey_cntr = be16_to_cpup((__be16 *) (out_mad->data + 46));
333 props->qkey_viol_cntr = be16_to_cpup((__be16 *) (out_mad->data + 48));
334 props->active_width = out_mad->data[31] & 0xf;
335 props->active_speed = out_mad->data[35] >> 4;
336 props->max_mtu = out_mad->data[41] & 0xf;
337 props->active_mtu = out_mad->data[36] >> 4;
338 props->subnet_timeout = out_mad->data[51] & 0x1f;
339 props->max_vl_num = out_mad->data[37] >> 4;
340 props->init_type_reply = out_mad->data[41] >> 4;
341
342 /* Check if extended speeds (EDR/FDR/...) are supported */
343 if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
344 ext_active_speed = out_mad->data[62] >> 4;
345
346 switch (ext_active_speed) {
347 case 1:
348 props->active_speed = IB_SPEED_FDR;
349 break;
350 case 2:
351 props->active_speed = IB_SPEED_EDR;
352 break;
353 }
354 }
355
356 /* If reported active speed is QDR, check if is FDR-10 */
357 if (props->active_speed == IB_SPEED_QDR) {
358 init_query_mad(in_mad);
359 in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
360 in_mad->attr_mod = cpu_to_be32(port);
361
362 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
363 NULL, NULL, in_mad, out_mad);
364 if (err)
365 goto out;
366
367 /* Checking LinkSpeedActive for FDR-10 */
368 if (out_mad->data[15] & 0x1)
369 props->active_speed = IB_SPEED_FDR10;
370 }
371
372 /* Avoid wrong speed value returned by FW if the IB link is down. */
373 if (props->state == IB_PORT_DOWN)
374 props->active_speed = IB_SPEED_SDR;
375
376 out:
377 kfree(in_mad);
378 kfree(out_mad);
379 return err;
380 }
381
state_to_phys_state(enum ib_port_state state)382 static u8 state_to_phys_state(enum ib_port_state state)
383 {
384 return state == IB_PORT_ACTIVE ? 5 : 3;
385 }
386
eth_link_query_port(struct ib_device * ibdev,u8 port,struct ib_port_attr * props,int netw_view)387 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
388 struct ib_port_attr *props, int netw_view)
389 {
390
391 struct mlx4_ib_dev *mdev = to_mdev(ibdev);
392 struct mlx4_ib_iboe *iboe = &mdev->iboe;
393 struct net_device *ndev;
394 enum ib_mtu tmp;
395 struct mlx4_cmd_mailbox *mailbox;
396 unsigned long flags;
397 int err = 0;
398
399 mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
400 if (IS_ERR(mailbox))
401 return PTR_ERR(mailbox);
402
403 err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
404 MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
405 MLX4_CMD_WRAPPED);
406 if (err)
407 goto out;
408
409 props->active_width = (((u8 *)mailbox->buf)[5] == 0x40) ?
410 IB_WIDTH_4X : IB_WIDTH_1X;
411 props->active_speed = IB_SPEED_QDR;
412 props->port_cap_flags = IB_PORT_CM_SUP;
413 if (netw_view)
414 props->gid_tbl_len = MLX4_ROCE_MAX_GIDS;
415 else
416 props->gid_tbl_len = mdev->dev->caps.gid_table_len[port];
417
418 props->max_msg_sz = mdev->dev->caps.max_msg_sz;
419 props->pkey_tbl_len = 1;
420 props->max_mtu = IB_MTU_4096;
421 props->max_vl_num = 2;
422 props->state = IB_PORT_DOWN;
423 props->phys_state = state_to_phys_state(props->state);
424 props->active_mtu = IB_MTU_256;
425 spin_lock_irqsave(&iboe->lock, flags);
426 ndev = iboe->netdevs[port - 1];
427 if (!ndev)
428 goto out_unlock;
429
430 tmp = iboe_get_mtu(ndev->if_mtu);
431 props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
432
433 props->state = (netif_running(ndev) && netif_carrier_ok(ndev)) ?
434 IB_PORT_ACTIVE : IB_PORT_DOWN;
435 props->phys_state = state_to_phys_state(props->state);
436 out_unlock:
437 spin_unlock_irqrestore(&iboe->lock, flags);
438 out:
439 mlx4_free_cmd_mailbox(mdev->dev, mailbox);
440 return err;
441 }
442
__mlx4_ib_query_port(struct ib_device * ibdev,u8 port,struct ib_port_attr * props,int netw_view)443 int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
444 struct ib_port_attr *props, int netw_view)
445 {
446 int err;
447
448 memset(props, 0, sizeof *props);
449
450 err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
451 ib_link_query_port(ibdev, port, props, netw_view) :
452 eth_link_query_port(ibdev, port, props, netw_view);
453
454 return err;
455 }
456
mlx4_ib_query_port(struct ib_device * ibdev,u8 port,struct ib_port_attr * props)457 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
458 struct ib_port_attr *props)
459 {
460 /* returns host view */
461 return __mlx4_ib_query_port(ibdev, port, props, 0);
462 }
463
__mlx4_ib_query_gid(struct ib_device * ibdev,u8 port,int index,union ib_gid * gid,int netw_view)464 int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
465 union ib_gid *gid, int netw_view)
466 {
467 struct ib_smp *in_mad = NULL;
468 struct ib_smp *out_mad = NULL;
469 int err = -ENOMEM;
470 struct mlx4_ib_dev *dev = to_mdev(ibdev);
471 int clear = 0;
472 int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
473
474 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
475 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
476 if (!in_mad || !out_mad)
477 goto out;
478
479 init_query_mad(in_mad);
480 in_mad->attr_id = IB_SMP_ATTR_PORT_INFO;
481 in_mad->attr_mod = cpu_to_be32(port);
482
483 if (mlx4_is_mfunc(dev->dev) && netw_view)
484 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
485
486 err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
487 if (err)
488 goto out;
489
490 memcpy(gid->raw, out_mad->data + 8, 8);
491
492 if (mlx4_is_mfunc(dev->dev) && !netw_view) {
493 if (index) {
494 /* For any index > 0, return the null guid */
495 err = 0;
496 clear = 1;
497 goto out;
498 }
499 }
500
501 init_query_mad(in_mad);
502 in_mad->attr_id = IB_SMP_ATTR_GUID_INFO;
503 in_mad->attr_mod = cpu_to_be32(index / 8);
504
505 err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
506 NULL, NULL, in_mad, out_mad);
507 if (err)
508 goto out;
509
510 memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
511
512 out:
513 if (clear)
514 memset(gid->raw + 8, 0, 8);
515 kfree(in_mad);
516 kfree(out_mad);
517 return err;
518 }
519
iboe_query_gid(struct ib_device * ibdev,u8 port,int index,union ib_gid * gid)520 static int iboe_query_gid(struct ib_device *ibdev, u8 port, int index,
521 union ib_gid *gid)
522 {
523 struct mlx4_ib_dev *dev = to_mdev(ibdev);
524
525 *gid = dev->iboe.gid_table[port - 1][index];
526
527 return 0;
528 }
529
mlx4_ib_query_gid(struct ib_device * ibdev,u8 port,int index,union ib_gid * gid)530 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
531 union ib_gid *gid)
532 {
533 if (rdma_port_get_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND)
534 return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
535 else
536 return iboe_query_gid(ibdev, port, index, gid);
537 }
538
__mlx4_ib_query_pkey(struct ib_device * ibdev,u8 port,u16 index,u16 * pkey,int netw_view)539 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
540 u16 *pkey, int netw_view)
541 {
542 struct ib_smp *in_mad = NULL;
543 struct ib_smp *out_mad = NULL;
544 int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
545 int err = -ENOMEM;
546
547 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
548 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
549 if (!in_mad || !out_mad)
550 goto out;
551
552 init_query_mad(in_mad);
553 in_mad->attr_id = IB_SMP_ATTR_PKEY_TABLE;
554 in_mad->attr_mod = cpu_to_be32(index / 32);
555
556 if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
557 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
558
559 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
560 in_mad, out_mad);
561 if (err)
562 goto out;
563
564 *pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
565
566 out:
567 kfree(in_mad);
568 kfree(out_mad);
569 return err;
570 }
571
mlx4_ib_query_pkey(struct ib_device * ibdev,u8 port,u16 index,u16 * pkey)572 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
573 {
574 return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
575 }
576
mlx4_ib_modify_device(struct ib_device * ibdev,int mask,struct ib_device_modify * props)577 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
578 struct ib_device_modify *props)
579 {
580 struct mlx4_cmd_mailbox *mailbox;
581 unsigned long flags;
582
583 if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
584 return -EOPNOTSUPP;
585
586 if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
587 return 0;
588
589 if (mlx4_is_slave(to_mdev(ibdev)->dev))
590 return -EOPNOTSUPP;
591
592 spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
593 memcpy(ibdev->node_desc, props->node_desc, 64);
594 spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
595
596 /*
597 * If possible, pass node desc to FW, so it can generate
598 * a 144 trap. If cmd fails, just ignore.
599 */
600 mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
601 if (IS_ERR(mailbox))
602 return 0;
603
604 memset(mailbox->buf, 0, 256);
605 memcpy(mailbox->buf, props->node_desc, 64);
606 mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
607 MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
608
609 mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
610
611 return 0;
612 }
613
mlx4_SET_PORT(struct mlx4_ib_dev * dev,u8 port,int reset_qkey_viols,u32 cap_mask)614 static int mlx4_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
615 u32 cap_mask)
616 {
617 struct mlx4_cmd_mailbox *mailbox;
618 int err;
619 u8 is_eth = dev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
620
621 mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
622 if (IS_ERR(mailbox))
623 return PTR_ERR(mailbox);
624
625 memset(mailbox->buf, 0, 256);
626
627 if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
628 *(u8 *) mailbox->buf = !!reset_qkey_viols << 6;
629 ((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
630 } else {
631 ((u8 *) mailbox->buf)[3] = !!reset_qkey_viols;
632 ((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
633 }
634
635 err = mlx4_cmd(dev->dev, mailbox->dma, port, is_eth, MLX4_CMD_SET_PORT,
636 MLX4_CMD_TIME_CLASS_B, MLX4_CMD_NATIVE);
637
638 mlx4_free_cmd_mailbox(dev->dev, mailbox);
639 return err;
640 }
641
mlx4_ib_modify_port(struct ib_device * ibdev,u8 port,int mask,struct ib_port_modify * props)642 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
643 struct ib_port_modify *props)
644 {
645 struct ib_port_attr attr;
646 u32 cap_mask;
647 int err;
648
649 mutex_lock(&to_mdev(ibdev)->cap_mask_mutex);
650
651 err = mlx4_ib_query_port(ibdev, port, &attr);
652 if (err)
653 goto out;
654
655 cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
656 ~props->clr_port_cap_mask;
657
658 err = mlx4_SET_PORT(to_mdev(ibdev), port,
659 !!(mask & IB_PORT_RESET_QKEY_CNTR),
660 cap_mask);
661
662 out:
663 mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
664 return err;
665 }
666
mlx4_ib_alloc_ucontext(struct ib_device * ibdev,struct ib_udata * udata)667 static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
668 struct ib_udata *udata)
669 {
670 struct mlx4_ib_dev *dev = to_mdev(ibdev);
671 struct mlx4_ib_ucontext *context;
672 struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
673 struct mlx4_ib_alloc_ucontext_resp resp;
674 int err;
675
676 if (!dev->ib_active)
677 return ERR_PTR(-EAGAIN);
678
679 if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
680 resp_v3.qp_tab_size = dev->dev->caps.num_qps;
681 if (mlx4_wc_enabled()) {
682 resp_v3.bf_reg_size = dev->dev->caps.bf_reg_size;
683 resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
684 } else {
685 resp_v3.bf_reg_size = 0;
686 resp_v3.bf_regs_per_page = 0;
687 }
688 } else {
689 resp.dev_caps = dev->dev->caps.userspace_caps;
690 resp.qp_tab_size = dev->dev->caps.num_qps;
691 if (mlx4_wc_enabled()) {
692 resp.bf_reg_size = dev->dev->caps.bf_reg_size;
693 resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
694 } else {
695 resp.bf_reg_size = 0;
696 resp.bf_regs_per_page = 0;
697 }
698 resp.cqe_size = dev->dev->caps.cqe_size;
699 }
700
701 context = kmalloc(sizeof *context, GFP_KERNEL);
702 if (!context)
703 return ERR_PTR(-ENOMEM);
704
705 err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
706 if (err) {
707 kfree(context);
708 return ERR_PTR(err);
709 }
710
711 INIT_LIST_HEAD(&context->db_page_list);
712 mutex_init(&context->db_page_mutex);
713
714 if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
715 err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
716 else
717 err = ib_copy_to_udata(udata, &resp, sizeof(resp));
718
719 if (err) {
720 mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
721 kfree(context);
722 return ERR_PTR(-EFAULT);
723 }
724
725 return &context->ibucontext;
726 }
727
mlx4_ib_dealloc_ucontext(struct ib_ucontext * ibcontext)728 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
729 {
730 struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
731
732 mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
733 kfree(context);
734
735 return 0;
736 }
737
738 /* XXX FBSD has no support for get_unmapped_area function */
739 #if 0
740 static unsigned long mlx4_ib_get_unmapped_area(struct file *file,
741 unsigned long addr,
742 unsigned long len, unsigned long pgoff,
743 unsigned long flags)
744 {
745 struct mm_struct *mm;
746 struct vm_area_struct *vma;
747 unsigned long start_addr;
748 unsigned long page_size_order;
749 unsigned long command;
750
751 mm = current->mm;
752 if (addr)
753 return current->mm->get_unmapped_area(file, addr, len,
754 pgoff, flags);
755
756 /* Last 8 bits hold the command others are data per that command */
757 command = pgoff & MLX4_IB_MMAP_CMD_MASK;
758 if (command != MLX4_IB_MMAP_GET_CONTIGUOUS_PAGES)
759 return current->mm->get_unmapped_area(file, addr, len,
760 pgoff, flags);
761
762 page_size_order = pgoff >> MLX4_IB_MMAP_CMD_BITS;
763 /* code is based on the huge-pages get_unmapped_area code */
764 start_addr = mm->free_area_cache;
765
766 if (len <= mm->cached_hole_size)
767 start_addr = TASK_UNMAPPED_BASE;
768
769
770 full_search:
771 addr = ALIGN(start_addr, 1 << page_size_order);
772
773 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
774 /* At this point: (!vma || addr < vma->vm_end). */
775 if (TASK_SIZE - len < addr) {
776 /*
777 * Start a new search - just in case we missed
778 * some holes.
779 */
780 if (start_addr != TASK_UNMAPPED_BASE) {
781 start_addr = TASK_UNMAPPED_BASE;
782 goto full_search;
783 }
784 return -ENOMEM;
785 }
786
787 if (!vma || addr + len <= vma->vm_start)
788 return addr;
789 addr = ALIGN(vma->vm_end, 1 << page_size_order);
790 }
791 }
792 #endif
793
mlx4_ib_mmap(struct ib_ucontext * context,struct vm_area_struct * vma)794 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
795 {
796 struct mlx4_ib_dev *dev = to_mdev(context->device);
797
798 /* Last 8 bits hold the command others are data per that command */
799 unsigned long command = vma->vm_pgoff & MLX4_IB_MMAP_CMD_MASK;
800
801 if (command < MLX4_IB_MMAP_GET_CONTIGUOUS_PAGES) {
802 /* compatability handling for commands 0 & 1*/
803 if (vma->vm_end - vma->vm_start != PAGE_SIZE)
804 return -EINVAL;
805 }
806 if (command == MLX4_IB_MMAP_UAR_PAGE) {
807 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
808
809 if (io_remap_pfn_range(vma, vma->vm_start,
810 to_mucontext(context)->uar.pfn,
811 PAGE_SIZE, vma->vm_page_prot))
812 return -EAGAIN;
813 } else if (command == MLX4_IB_MMAP_BLUE_FLAME_PAGE &&
814 dev->dev->caps.bf_reg_size != 0) {
815 vma->vm_page_prot = pgprot_wc(vma->vm_page_prot);
816
817 if (io_remap_pfn_range(vma, vma->vm_start,
818 to_mucontext(context)->uar.pfn +
819 dev->dev->caps.num_uars,
820 PAGE_SIZE, vma->vm_page_prot))
821 return -EAGAIN;
822 } else if (command == MLX4_IB_MMAP_GET_HW_CLOCK) {
823 struct mlx4_clock_params params;
824 int ret;
825
826 ret = mlx4_get_internal_clock_params(dev->dev, ¶ms);
827 if (ret)
828 return ret;
829
830 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
831
832 if (io_remap_pfn_range(vma, vma->vm_start,
833 (pci_resource_start(dev->dev->pdev,
834 params.bar) + params.offset)
835 >> PAGE_SHIFT,
836 PAGE_SIZE, vma->vm_page_prot))
837 return -EAGAIN;
838 } else
839 return -EINVAL;
840
841 return 0;
842 }
843
mlx4_ib_ioctl(struct ib_ucontext * context,unsigned int cmd,unsigned long arg)844 static int mlx4_ib_ioctl(struct ib_ucontext *context, unsigned int cmd,
845 unsigned long arg)
846 {
847 struct mlx4_ib_dev *dev = to_mdev(context->device);
848 int ret;
849 int offset;
850
851 switch (cmd) {
852 case MLX4_IOCHWCLOCKOFFSET: {
853 struct mlx4_clock_params params;
854 int ret;
855 ret = mlx4_get_internal_clock_params(dev->dev, ¶ms);
856 if (!ret) {
857 offset = params.offset % PAGE_SIZE;
858 ret = put_user(offset,
859 (int *)arg);
860 return sizeof(int);
861 } else {
862 return ret;
863 }
864 }
865 default: {
866 pr_err("mlx4_ib: invalid ioctl %u command with arg %lX\n",
867 cmd, arg);
868 return -ENOTTY;
869 }
870 }
871
872 return ret;
873 }
874
mlx4_ib_query_values(struct ib_device * device,int q_values,struct ib_device_values * values)875 static int mlx4_ib_query_values(struct ib_device *device, int q_values,
876 struct ib_device_values *values)
877 {
878 struct mlx4_ib_dev *dev = to_mdev(device);
879 cycle_t cycles;
880
881 values->values_mask = 0;
882 if (q_values & IBV_VALUES_HW_CLOCK) {
883 cycles = mlx4_read_clock(dev->dev);
884 if (cycles < 0) {
885 values->hwclock = cycles & CORE_CLOCK_MASK;
886 values->values_mask |= IBV_VALUES_HW_CLOCK;
887 }
888 q_values &= ~IBV_VALUES_HW_CLOCK;
889 }
890
891 if (q_values)
892 return -ENOTTY;
893
894 return 0;
895 }
896
mlx4_ib_alloc_pd(struct ib_device * ibdev,struct ib_ucontext * context,struct ib_udata * udata)897 static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
898 struct ib_ucontext *context,
899 struct ib_udata *udata)
900 {
901 struct mlx4_ib_pd *pd;
902 int err;
903
904 pd = kmalloc(sizeof *pd, GFP_KERNEL);
905 if (!pd)
906 return ERR_PTR(-ENOMEM);
907
908 err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
909 if (err) {
910 kfree(pd);
911 return ERR_PTR(err);
912 }
913
914 if (context)
915 if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
916 mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
917 kfree(pd);
918 return ERR_PTR(-EFAULT);
919 }
920
921 return &pd->ibpd;
922 }
923
mlx4_ib_dealloc_pd(struct ib_pd * pd)924 static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
925 {
926 mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
927 kfree(pd);
928
929 return 0;
930 }
931
mlx4_ib_alloc_xrcd(struct ib_device * ibdev,struct ib_ucontext * context,struct ib_udata * udata)932 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
933 struct ib_ucontext *context,
934 struct ib_udata *udata)
935 {
936 struct mlx4_ib_xrcd *xrcd;
937 int err;
938
939 if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
940 return ERR_PTR(-ENOSYS);
941
942 xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
943 if (!xrcd)
944 return ERR_PTR(-ENOMEM);
945
946 err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
947 if (err)
948 goto err1;
949
950 xrcd->pd = ib_alloc_pd(ibdev);
951 if (IS_ERR(xrcd->pd)) {
952 err = PTR_ERR(xrcd->pd);
953 goto err2;
954 }
955
956 xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, 1, 0);
957 if (IS_ERR(xrcd->cq)) {
958 err = PTR_ERR(xrcd->cq);
959 goto err3;
960 }
961
962 return &xrcd->ibxrcd;
963
964 err3:
965 ib_dealloc_pd(xrcd->pd);
966 err2:
967 mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
968 err1:
969 kfree(xrcd);
970 return ERR_PTR(err);
971 }
972
mlx4_ib_dealloc_xrcd(struct ib_xrcd * xrcd)973 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
974 {
975 ib_destroy_cq(to_mxrcd(xrcd)->cq);
976 ib_dealloc_pd(to_mxrcd(xrcd)->pd);
977 mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
978 kfree(xrcd);
979
980 return 0;
981 }
982
add_gid_entry(struct ib_qp * ibqp,union ib_gid * gid)983 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
984 {
985 struct mlx4_ib_qp *mqp = to_mqp(ibqp);
986 struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
987 struct mlx4_ib_gid_entry *ge;
988
989 ge = kzalloc(sizeof *ge, GFP_KERNEL);
990 if (!ge)
991 return -ENOMEM;
992
993 ge->gid = *gid;
994 if (mlx4_ib_add_mc(mdev, mqp, gid)) {
995 ge->port = mqp->port;
996 ge->added = 1;
997 }
998
999 mutex_lock(&mqp->mutex);
1000 list_add_tail(&ge->list, &mqp->gid_list);
1001 mutex_unlock(&mqp->mutex);
1002
1003 return 0;
1004 }
1005
mlx4_ib_add_mc(struct mlx4_ib_dev * mdev,struct mlx4_ib_qp * mqp,union ib_gid * gid)1006 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
1007 union ib_gid *gid)
1008 {
1009 u8 mac[6];
1010 struct net_device *ndev;
1011 int ret = 0;
1012
1013 if (!mqp->port)
1014 return 0;
1015
1016 spin_lock(&mdev->iboe.lock);
1017 ndev = mdev->iboe.netdevs[mqp->port - 1];
1018 if (ndev)
1019 dev_hold(ndev);
1020 spin_unlock(&mdev->iboe.lock);
1021
1022 if (ndev) {
1023 rdma_get_mcast_mac((struct in6_addr *)gid, mac);
1024 rtnl_lock();
1025 dev_mc_add(mdev->iboe.netdevs[mqp->port - 1], mac, 6, 0);
1026 ret = 1;
1027 rtnl_unlock();
1028 dev_put(ndev);
1029 }
1030
1031 return ret;
1032 }
1033
1034 struct mlx4_ib_steering {
1035 struct list_head list;
1036 u64 reg_id;
1037 union ib_gid gid;
1038 };
1039
parse_flow_attr(struct mlx4_dev * dev,union ib_flow_spec * ib_spec,struct _rule_hw * mlx4_spec)1040 static int parse_flow_attr(struct mlx4_dev *dev,
1041 union ib_flow_spec *ib_spec,
1042 struct _rule_hw *mlx4_spec)
1043 {
1044 enum mlx4_net_trans_rule_id type;
1045
1046 switch (ib_spec->type) {
1047 case IB_FLOW_SPEC_ETH:
1048 type = MLX4_NET_TRANS_RULE_ID_ETH;
1049 memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac,
1050 ETH_ALEN);
1051 memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac,
1052 ETH_ALEN);
1053 mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
1054 mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
1055 break;
1056
1057 case IB_FLOW_SPEC_IB:
1058 type = MLX4_NET_TRANS_RULE_ID_IB;
1059 mlx4_spec->ib.l3_qpn = ib_spec->ib.val.l3_type_qpn;
1060 mlx4_spec->ib.qpn_mask = ib_spec->ib.mask.l3_type_qpn;
1061 memcpy(&mlx4_spec->ib.dst_gid, ib_spec->ib.val.dst_gid, 16);
1062 memcpy(&mlx4_spec->ib.dst_gid_msk,
1063 ib_spec->ib.mask.dst_gid, 16);
1064 break;
1065
1066 case IB_FLOW_SPEC_IPV4:
1067 type = MLX4_NET_TRANS_RULE_ID_IPV4;
1068 mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip;
1069 mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip;
1070 mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip;
1071 mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip;
1072 break;
1073
1074 case IB_FLOW_SPEC_TCP:
1075 case IB_FLOW_SPEC_UDP:
1076 type = ib_spec->type == IB_FLOW_SPEC_TCP ?
1077 MLX4_NET_TRANS_RULE_ID_TCP :
1078 MLX4_NET_TRANS_RULE_ID_UDP;
1079 mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port;
1080 mlx4_spec->tcp_udp.dst_port_msk =
1081 ib_spec->tcp_udp.mask.dst_port;
1082 mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port;
1083 mlx4_spec->tcp_udp.src_port_msk =
1084 ib_spec->tcp_udp.mask.src_port;
1085 break;
1086
1087 default:
1088 return -EINVAL;
1089 }
1090 if (map_sw_to_hw_steering_id(dev, type) < 0 ||
1091 hw_rule_sz(dev, type) < 0)
1092 return -EINVAL;
1093 mlx4_spec->id = cpu_to_be16(map_sw_to_hw_steering_id(dev, type));
1094 mlx4_spec->size = hw_rule_sz(dev, type) >> 2;
1095 return hw_rule_sz(dev, type);
1096 }
1097
__mlx4_ib_create_flow(struct ib_qp * qp,struct ib_flow_attr * flow_attr,int domain,enum mlx4_net_trans_promisc_mode flow_type,u64 * reg_id)1098 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1099 int domain,
1100 enum mlx4_net_trans_promisc_mode flow_type,
1101 u64 *reg_id)
1102 {
1103 int ret, i;
1104 int size = 0;
1105 void *ib_flow;
1106 struct mlx4_ib_dev *mdev = to_mdev(qp->device);
1107 struct mlx4_cmd_mailbox *mailbox;
1108 struct mlx4_net_trans_rule_hw_ctrl *ctrl;
1109 size_t rule_size = sizeof(struct mlx4_net_trans_rule_hw_ctrl) +
1110 (sizeof(struct _rule_hw) * flow_attr->num_of_specs);
1111
1112 static const u16 __mlx4_domain[] = {
1113 [IB_FLOW_DOMAIN_USER] = MLX4_DOMAIN_UVERBS,
1114 [IB_FLOW_DOMAIN_ETHTOOL] = MLX4_DOMAIN_ETHTOOL,
1115 [IB_FLOW_DOMAIN_RFS] = MLX4_DOMAIN_RFS,
1116 [IB_FLOW_DOMAIN_NIC] = MLX4_DOMAIN_NIC,
1117 };
1118
1119 if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
1120 pr_err("Invalid priority value.\n");
1121 return -EINVAL;
1122 }
1123 if (domain >= IB_FLOW_DOMAIN_NUM) {
1124 pr_err("Invalid domain value.\n");
1125 return -EINVAL;
1126 }
1127 if (map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
1128 return -EINVAL;
1129
1130 mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
1131 if (IS_ERR(mailbox))
1132 return PTR_ERR(mailbox);
1133 memset(mailbox->buf, 0, rule_size);
1134 ctrl = mailbox->buf;
1135
1136 ctrl->prio = cpu_to_be16(__mlx4_domain[domain] |
1137 flow_attr->priority);
1138 ctrl->type = map_sw_to_hw_steering_mode(mdev->dev, flow_type);
1139 ctrl->port = flow_attr->port;
1140 ctrl->qpn = cpu_to_be32(qp->qp_num);
1141
1142 if (flow_attr->flags & IB_FLOW_ATTR_FLAGS_ALLOW_LOOP_BACK)
1143 ctrl->flags = (1 << 3);
1144
1145 ib_flow = flow_attr + 1;
1146 size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
1147 for (i = 0; i < flow_attr->num_of_specs; i++) {
1148 ret = parse_flow_attr(mdev->dev, ib_flow, mailbox->buf + size);
1149 if (ret < 0) {
1150 mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1151 return -EINVAL;
1152 }
1153 ib_flow += ((union ib_flow_spec *)ib_flow)->size;
1154 size += ret;
1155 }
1156
1157 ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
1158 MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
1159 MLX4_CMD_NATIVE);
1160 if (ret == -ENOMEM)
1161 pr_err("mcg table is full. Fail to register network rule.\n");
1162 else if (ret == -ENXIO)
1163 pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
1164 else if (ret)
1165 pr_err("Invalid argumant. Fail to register network rule.\n");
1166 mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1167 return ret;
1168 }
1169
__mlx4_ib_destroy_flow(struct mlx4_dev * dev,u64 reg_id)1170 static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
1171 {
1172 int err;
1173 err = mlx4_cmd(dev, reg_id, 0, 0,
1174 MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
1175 MLX4_CMD_NATIVE);
1176 if (err)
1177 pr_err("Fail to detach network rule. registration id = 0x%llx\n",
1178 (unsigned long long)reg_id);
1179 return err;
1180 }
1181
mlx4_ib_create_flow(struct ib_qp * qp,struct ib_flow_attr * flow_attr,int domain)1182 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
1183 struct ib_flow_attr *flow_attr,
1184 int domain)
1185 {
1186 int err = 0, i = 0;
1187 struct mlx4_ib_flow *mflow;
1188 enum mlx4_net_trans_promisc_mode type[2];
1189
1190 memset(type, 0, sizeof(type));
1191
1192 mflow = kzalloc(sizeof(struct mlx4_ib_flow), GFP_KERNEL);
1193 if (!mflow) {
1194 err = -ENOMEM;
1195 goto err_free;
1196 }
1197
1198 switch (flow_attr->type) {
1199 case IB_FLOW_ATTR_NORMAL:
1200 type[0] = MLX4_FS_REGULAR;
1201 break;
1202
1203 case IB_FLOW_ATTR_ALL_DEFAULT:
1204 type[0] = MLX4_FS_ALL_DEFAULT;
1205 break;
1206
1207 case IB_FLOW_ATTR_MC_DEFAULT:
1208 type[0] = MLX4_FS_MC_DEFAULT;
1209 break;
1210
1211 case IB_FLOW_ATTR_SNIFFER:
1212 type[0] = MLX4_FS_UC_SNIFFER;
1213 type[1] = MLX4_FS_MC_SNIFFER;
1214 break;
1215
1216 default:
1217 err = -EINVAL;
1218 goto err_free;
1219 }
1220
1221 while (i < ARRAY_SIZE(type) && type[i]) {
1222 err = __mlx4_ib_create_flow(qp, flow_attr, domain, type[i],
1223 &mflow->reg_id[i]);
1224 if (err)
1225 goto err_free;
1226 i++;
1227 }
1228
1229 return &mflow->ibflow;
1230
1231 err_free:
1232 kfree(mflow);
1233 return ERR_PTR(err);
1234 }
1235
mlx4_ib_destroy_flow(struct ib_flow * flow_id)1236 static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
1237 {
1238 int err, ret = 0;
1239 int i = 0;
1240 struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
1241 struct mlx4_ib_flow *mflow = to_mflow(flow_id);
1242
1243 while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i]) {
1244 err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i]);
1245 if (err)
1246 ret = err;
1247 i++;
1248 }
1249
1250 kfree(mflow);
1251 return ret;
1252 }
1253
find_gid_entry(struct mlx4_ib_qp * qp,u8 * raw)1254 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
1255 {
1256 struct mlx4_ib_gid_entry *ge;
1257 struct mlx4_ib_gid_entry *tmp;
1258 struct mlx4_ib_gid_entry *ret = NULL;
1259
1260 list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
1261 if (!memcmp(raw, ge->gid.raw, 16)) {
1262 ret = ge;
1263 break;
1264 }
1265 }
1266
1267 return ret;
1268 }
1269
1270
del_gid_entry(struct ib_qp * ibqp,union ib_gid * gid)1271 static int del_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
1272 {
1273 struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1274 struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1275 struct mlx4_ib_gid_entry *ge;
1276 struct net_device *ndev;
1277 u8 mac[6];
1278
1279 mutex_lock(&mqp->mutex);
1280 ge = find_gid_entry(mqp, gid->raw);
1281 if (ge) {
1282 spin_lock(&mdev->iboe.lock);
1283 ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
1284 if (ndev)
1285 dev_hold(ndev);
1286 spin_unlock(&mdev->iboe.lock);
1287 rdma_get_mcast_mac((struct in6_addr *)gid, mac);
1288 if (ndev) {
1289 rtnl_lock();
1290 dev_mc_delete(mdev->iboe.netdevs[ge->port - 1], mac, 6, 0);
1291 rtnl_unlock();
1292 dev_put(ndev);
1293 }
1294 list_del(&ge->list);
1295 kfree(ge);
1296 } else
1297 pr_warn("could not find mgid entry\n");
1298
1299 mutex_unlock(&mqp->mutex);
1300 return ge != 0 ? 0 : -EINVAL;
1301 }
1302
_mlx4_ib_mcg_detach(struct ib_qp * ibqp,union ib_gid * gid,u16 lid,int count)1303 static int _mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid,
1304 int count)
1305 {
1306 int err;
1307 struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1308 struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1309 u64 reg_id = 0;
1310 int record_err = 0;
1311
1312 if (mdev->dev->caps.steering_mode ==
1313 MLX4_STEERING_MODE_DEVICE_MANAGED) {
1314 struct mlx4_ib_steering *ib_steering;
1315 struct mlx4_ib_steering *tmp;
1316 LIST_HEAD(temp);
1317
1318 mutex_lock(&mqp->mutex);
1319 list_for_each_entry_safe(ib_steering, tmp, &mqp->steering_rules,
1320 list) {
1321 if (memcmp(ib_steering->gid.raw, gid->raw, 16))
1322 continue;
1323
1324 if (--count < 0)
1325 break;
1326
1327 list_del(&ib_steering->list);
1328 list_add(&ib_steering->list, &temp);
1329 }
1330 mutex_unlock(&mqp->mutex);
1331 list_for_each_entry_safe(ib_steering, tmp, &temp,
1332 list) {
1333 reg_id = ib_steering->reg_id;
1334
1335 err = mlx4_multicast_detach(mdev->dev, &mqp->mqp,
1336 gid->raw,
1337 (ibqp->qp_type == IB_QPT_RAW_PACKET) ?
1338 MLX4_PROT_ETH : MLX4_PROT_IB_IPV6,
1339 reg_id);
1340 if (err) {
1341 record_err = record_err ?: err;
1342 continue;
1343 }
1344
1345 err = del_gid_entry(ibqp, gid);
1346 if (err) {
1347 record_err = record_err ?: err;
1348 continue;
1349 }
1350
1351 list_del(&ib_steering->list);
1352 kfree(ib_steering);
1353 }
1354 mutex_lock(&mqp->mutex);
1355 list_for_each_entry(ib_steering, &temp, list) {
1356 list_add(&ib_steering->list, &mqp->steering_rules);
1357 }
1358 mutex_unlock(&mqp->mutex);
1359 if (count) {
1360 pr_warn("Couldn't release all reg_ids for mgid. Steering rule is left attached\n");
1361 return -EINVAL;
1362 }
1363
1364 } else {
1365 if (mdev->dev->caps.steering_mode == MLX4_STEERING_MODE_B0 &&
1366 ibqp->qp_type == IB_QPT_RAW_PACKET)
1367 gid->raw[5] = mqp->port;
1368
1369 err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1370 (ibqp->qp_type == IB_QPT_RAW_PACKET) ?
1371 MLX4_PROT_ETH : MLX4_PROT_IB_IPV6,
1372 reg_id);
1373 if (err)
1374 return err;
1375
1376 err = del_gid_entry(ibqp, gid);
1377
1378 if (err)
1379 return err;
1380 }
1381
1382 return record_err;
1383 }
1384
mlx4_ib_mcg_detach(struct ib_qp * ibqp,union ib_gid * gid,u16 lid)1385 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1386 {
1387 struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1388 int count = (mdev->dev->caps.steering_mode ==
1389 MLX4_STEERING_MODE_DEVICE_MANAGED) ?
1390 mdev->dev->caps.num_ports : 1;
1391
1392 return _mlx4_ib_mcg_detach(ibqp, gid, lid, count);
1393 }
1394
mlx4_ib_mcg_attach(struct ib_qp * ibqp,union ib_gid * gid,u16 lid)1395 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1396 {
1397 int err = -ENODEV;
1398 struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1399 struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1400 DECLARE_BITMAP(ports, MLX4_MAX_PORTS);
1401 int i = 0;
1402
1403 if (mdev->dev->caps.steering_mode == MLX4_STEERING_MODE_B0 &&
1404 ibqp->qp_type == IB_QPT_RAW_PACKET)
1405 gid->raw[5] = mqp->port;
1406
1407 if (mdev->dev->caps.steering_mode ==
1408 MLX4_STEERING_MODE_DEVICE_MANAGED) {
1409 bitmap_fill(ports, mdev->dev->caps.num_ports);
1410 } else {
1411 if (mqp->port <= mdev->dev->caps.num_ports) {
1412 bitmap_zero(ports, mdev->dev->caps.num_ports);
1413 set_bit(0, ports);
1414 } else {
1415 return -EINVAL;
1416 }
1417 }
1418
1419 for (; i < mdev->dev->caps.num_ports; i++) {
1420 u64 reg_id;
1421 struct mlx4_ib_steering *ib_steering = NULL;
1422 if (!test_bit(i, ports))
1423 continue;
1424 if (mdev->dev->caps.steering_mode ==
1425 MLX4_STEERING_MODE_DEVICE_MANAGED) {
1426 ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
1427 if (!ib_steering)
1428 goto err_add;
1429 }
1430
1431 err = mlx4_multicast_attach(mdev->dev, &mqp->mqp,
1432 gid->raw, i + 1,
1433 !!(mqp->flags &
1434 MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1435 (ibqp->qp_type == IB_QPT_RAW_PACKET) ?
1436 MLX4_PROT_ETH : MLX4_PROT_IB_IPV6,
1437 ®_id);
1438 if (err) {
1439 kfree(ib_steering);
1440 goto err_add;
1441 }
1442
1443 err = add_gid_entry(ibqp, gid);
1444 if (err) {
1445 mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1446 MLX4_PROT_IB_IPV6, reg_id);
1447 kfree(ib_steering);
1448 goto err_add;
1449 }
1450
1451 if (ib_steering) {
1452 memcpy(ib_steering->gid.raw, gid->raw, 16);
1453 mutex_lock(&mqp->mutex);
1454 list_add(&ib_steering->list, &mqp->steering_rules);
1455 mutex_unlock(&mqp->mutex);
1456 ib_steering->reg_id = reg_id;
1457 }
1458 }
1459
1460
1461 return 0;
1462
1463 err_add:
1464 if (i > 0)
1465 _mlx4_ib_mcg_detach(ibqp, gid, lid, i);
1466
1467 return err;
1468 }
1469
init_node_data(struct mlx4_ib_dev * dev)1470 static int init_node_data(struct mlx4_ib_dev *dev)
1471 {
1472 struct ib_smp *in_mad = NULL;
1473 struct ib_smp *out_mad = NULL;
1474 int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
1475 int err = -ENOMEM;
1476
1477 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
1478 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
1479 if (!in_mad || !out_mad)
1480 goto out;
1481
1482 init_query_mad(in_mad);
1483 in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
1484 if (mlx4_is_master(dev->dev))
1485 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
1486
1487 err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1488 if (err)
1489 goto out;
1490
1491 memcpy(dev->ib_dev.node_desc, out_mad->data, 64);
1492
1493 in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
1494
1495 err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1496 if (err)
1497 goto out;
1498
1499 dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
1500 memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
1501
1502 out:
1503 kfree(in_mad);
1504 kfree(out_mad);
1505 return err;
1506 }
1507
show_hca(struct device * device,struct device_attribute * attr,char * buf)1508 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
1509 char *buf)
1510 {
1511 struct mlx4_ib_dev *dev =
1512 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1513 return sprintf(buf, "MT%d\n", dev->dev->pdev->device);
1514 }
1515
show_fw_ver(struct device * device,struct device_attribute * attr,char * buf)1516 static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
1517 char *buf)
1518 {
1519 struct mlx4_ib_dev *dev =
1520 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1521 return sprintf(buf, "%d.%d.%d\n", (int) (dev->dev->caps.fw_ver >> 32),
1522 (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
1523 (int) dev->dev->caps.fw_ver & 0xffff);
1524 }
1525
show_rev(struct device * device,struct device_attribute * attr,char * buf)1526 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
1527 char *buf)
1528 {
1529 struct mlx4_ib_dev *dev =
1530 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1531 return sprintf(buf, "%x\n", dev->dev->rev_id);
1532 }
1533
show_board(struct device * device,struct device_attribute * attr,char * buf)1534 static ssize_t show_board(struct device *device, struct device_attribute *attr,
1535 char *buf)
1536 {
1537 struct mlx4_ib_dev *dev =
1538 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1539 return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
1540 dev->dev->board_id);
1541 }
1542
show_vsd(struct device * device,struct device_attribute * attr,char * buf)1543 static ssize_t show_vsd(struct device *device, struct device_attribute *attr,
1544 char *buf)
1545 {
1546 struct mlx4_ib_dev *dev =
1547 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1548 ssize_t len = MLX4_VSD_LEN;
1549
1550 if (dev->dev->vsd_vendor_id == PCI_VENDOR_ID_MELLANOX)
1551 len = sprintf(buf, "%.*s\n", MLX4_VSD_LEN, dev->dev->vsd);
1552 else
1553 memcpy(buf, dev->dev->vsd, MLX4_VSD_LEN);
1554
1555 return len;
1556 }
1557
1558 static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL);
1559 static DEVICE_ATTR(fw_ver, S_IRUGO, show_fw_ver, NULL);
1560 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca, NULL);
1561 static DEVICE_ATTR(board_id, S_IRUGO, show_board, NULL);
1562 static DEVICE_ATTR(vsd, S_IRUGO, show_vsd, NULL);
1563
1564 static struct device_attribute *mlx4_class_attributes[] = {
1565 &dev_attr_hw_rev,
1566 &dev_attr_fw_ver,
1567 &dev_attr_hca_type,
1568 &dev_attr_board_id,
1569 &dev_attr_vsd
1570 };
1571
mlx4_addrconf_ifid_eui48(u8 * eui,u16 vlan_id,struct net_device * dev,u8 port)1572 static void mlx4_addrconf_ifid_eui48(u8 *eui, u16 vlan_id, struct net_device *dev, u8 port)
1573 {
1574 memcpy(eui, IF_LLADDR(dev), 3);
1575 memcpy(eui + 5, IF_LLADDR(dev) + 3, 3);
1576 if (vlan_id < 0x1000) {
1577 eui[3] = vlan_id >> 8;
1578 eui[4] = vlan_id & 0xff;
1579 } else {
1580 eui[3] = 0xff;
1581 eui[4] = 0xfe;
1582 }
1583 eui[0] ^= 2;
1584 }
1585
update_gids_task(struct work_struct * work)1586 static void update_gids_task(struct work_struct *work)
1587 {
1588 struct update_gid_work *gw = container_of(work, struct update_gid_work, work);
1589 struct mlx4_cmd_mailbox *mailbox;
1590 union ib_gid *gids;
1591 int err;
1592 struct mlx4_dev *dev = gw->dev->dev;
1593
1594
1595 mailbox = mlx4_alloc_cmd_mailbox(dev);
1596 if (IS_ERR(mailbox)) {
1597 pr_warn("update gid table failed %ld\n", PTR_ERR(mailbox));
1598 goto free;
1599 }
1600
1601 gids = mailbox->buf;
1602 memcpy(gids, gw->gids, sizeof gw->gids);
1603
1604 if (mlx4_ib_port_link_layer(&gw->dev->ib_dev, gw->port) ==
1605 IB_LINK_LAYER_ETHERNET) {
1606 err = mlx4_cmd(dev, mailbox->dma,
1607 MLX4_SET_PORT_GID_TABLE << 8 | gw->port,
1608 1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
1609 MLX4_CMD_WRAPPED);
1610
1611 if (err)
1612 pr_warn("set port command failed\n");
1613 else
1614 mlx4_ib_dispatch_event(gw->dev, gw->port,
1615 IB_EVENT_GID_CHANGE);
1616 }
1617
1618 mlx4_free_cmd_mailbox(dev, mailbox);
1619 free:
1620 kfree(gw);
1621 }
1622
reset_gids_task(struct work_struct * work)1623 static void reset_gids_task(struct work_struct *work)
1624 {
1625 struct update_gid_work *gw =
1626 container_of(work, struct update_gid_work, work);
1627 struct mlx4_cmd_mailbox *mailbox;
1628 union ib_gid *gids;
1629 int err;
1630 struct mlx4_dev *dev = gw->dev->dev;
1631
1632 mailbox = mlx4_alloc_cmd_mailbox(dev);
1633 if (IS_ERR(mailbox)) {
1634 pr_warn("reset gid table failed\n");
1635 goto free;
1636 }
1637
1638 gids = mailbox->buf;
1639 memcpy(gids, gw->gids, sizeof(gw->gids));
1640
1641 if (mlx4_ib_port_link_layer(&gw->dev->ib_dev, 1) ==
1642 IB_LINK_LAYER_ETHERNET &&
1643 dev->caps.num_ports > 0) {
1644 err = mlx4_cmd(dev, mailbox->dma,
1645 MLX4_SET_PORT_GID_TABLE << 8 | 1,
1646 1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
1647 MLX4_CMD_WRAPPED);
1648 if (err)
1649 pr_warn("set port 1 command failed\n");
1650 }
1651
1652 if (mlx4_ib_port_link_layer(&gw->dev->ib_dev, 2) ==
1653 IB_LINK_LAYER_ETHERNET &&
1654 dev->caps.num_ports > 1) {
1655 err = mlx4_cmd(dev, mailbox->dma,
1656 MLX4_SET_PORT_GID_TABLE << 8 | 2,
1657 1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
1658 MLX4_CMD_WRAPPED);
1659 if (err)
1660 pr_warn("set port 2 command failed\n");
1661 }
1662
1663 mlx4_free_cmd_mailbox(dev, mailbox);
1664 free:
1665 kfree(gw);
1666 }
1667
update_gid_table(struct mlx4_ib_dev * dev,int port,union ib_gid * gid,int clear,int default_gid)1668 static int update_gid_table(struct mlx4_ib_dev *dev, int port,
1669 union ib_gid *gid, int clear, int default_gid)
1670 {
1671 struct update_gid_work *work;
1672 int i;
1673 int need_update = 0;
1674 int free = -1;
1675 int found = -1;
1676 int max_gids;
1677 int start_index = !default_gid;
1678
1679 max_gids = dev->dev->caps.gid_table_len[port];
1680 for (i = start_index; i < max_gids; ++i) {
1681 if (!memcmp(&dev->iboe.gid_table[port - 1][i], gid,
1682 sizeof(*gid)))
1683 found = i;
1684
1685 if (clear) {
1686 if (found >= 0) {
1687 need_update = 1;
1688 dev->iboe.gid_table[port - 1][found] = zgid;
1689 break;
1690 }
1691 } else {
1692 if (found >= 0)
1693 break;
1694
1695 if (free < 0 &&
1696 !memcmp(&dev->iboe.gid_table[port - 1][i],
1697 &zgid, sizeof(*gid)))
1698 free = i;
1699 }
1700 }
1701
1702 if (found == -1 && !clear && free < 0) {
1703 pr_err("GID table of port %d is full. Can't add "GID_PRINT_FMT"\n",
1704 port, GID_PRINT_ARGS(gid));
1705 return -ENOMEM;
1706 }
1707 if (found == -1 && clear) {
1708 pr_err(GID_PRINT_FMT" is not in GID table of port %d\n", GID_PRINT_ARGS(gid), port);
1709 return -EINVAL;
1710 }
1711 if (found == -1 && !clear && free >= 0) {
1712 dev->iboe.gid_table[port - 1][free] = *gid;
1713 need_update = 1;
1714 }
1715
1716 if (!need_update)
1717 return 0;
1718
1719 work = kzalloc(sizeof *work, GFP_ATOMIC);
1720 if (!work)
1721 return -ENOMEM;
1722
1723 memcpy(work->gids, dev->iboe.gid_table[port - 1], sizeof(work->gids));
1724 INIT_WORK(&work->work, update_gids_task);
1725 work->port = port;
1726 work->dev = dev;
1727 queue_work(wq, &work->work);
1728
1729 return 0;
1730 }
1731
reset_gid_table(struct mlx4_ib_dev * dev)1732 static int reset_gid_table(struct mlx4_ib_dev *dev)
1733 {
1734 struct update_gid_work *work;
1735
1736
1737 work = kzalloc(sizeof(*work), GFP_ATOMIC);
1738 if (!work)
1739 return -ENOMEM;
1740
1741 memset(dev->iboe.gid_table, 0, sizeof(dev->iboe.gid_table));
1742 memset(work->gids, 0, sizeof(work->gids));
1743 INIT_WORK(&work->work, reset_gids_task);
1744 work->dev = dev;
1745 queue_work(wq, &work->work);
1746 return 0;
1747 }
1748
1749 /* XXX BOND Related - stub (no support for these flags in FBSD)*/
netif_is_bond_master(struct net_device * dev)1750 static inline int netif_is_bond_master(struct net_device *dev)
1751 {
1752 #if 0
1753 return (dev->flags & IFF_MASTER) && (dev->priv_flags & IFF_BONDING);
1754 #endif
1755 return 0;
1756 }
1757
mlx4_make_default_gid(struct net_device * dev,union ib_gid * gid,u8 port)1758 static void mlx4_make_default_gid(struct net_device *dev, union ib_gid *gid, u8 port)
1759 {
1760 gid->global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL);
1761 mlx4_addrconf_ifid_eui48(&gid->raw[8], 0xffff, dev, port);
1762 }
1763
mlx4_ib_get_dev_port(struct net_device * dev,struct mlx4_ib_dev * ibdev)1764 static u8 mlx4_ib_get_dev_port(struct net_device *dev, struct mlx4_ib_dev *ibdev)
1765 {
1766 u8 port = 0;
1767 struct mlx4_ib_iboe *iboe;
1768 struct net_device *real_dev = rdma_vlan_dev_real_dev(dev) ?
1769 rdma_vlan_dev_real_dev(dev) : dev;
1770
1771 iboe = &ibdev->iboe;
1772
1773 for (port = 1; port <= MLX4_MAX_PORTS; ++port)
1774 if ((netif_is_bond_master(real_dev) && (real_dev == iboe->masters[port - 1])) ||
1775 (!netif_is_bond_master(real_dev) && (real_dev == iboe->netdevs[port - 1])))
1776 break;
1777
1778 return port > MLX4_MAX_PORTS ? 0 : port;
1779 }
1780
mlx4_ib_get_dev_addr(struct net_device * dev,struct mlx4_ib_dev * ibdev,u8 port)1781 static void mlx4_ib_get_dev_addr(struct net_device *dev, struct mlx4_ib_dev *ibdev, u8 port)
1782 {
1783 struct ifaddr *ifa;
1784 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1785 struct inet6_dev *in6_dev;
1786 union ib_gid *pgid;
1787 struct inet6_ifaddr *ifp;
1788 #endif
1789 union ib_gid gid;
1790
1791
1792 if ((port == 0) || (port > MLX4_MAX_PORTS))
1793 return;
1794
1795 /* IPv4 gids */
1796 TAILQ_FOREACH(ifa, &dev->if_addrhead, ifa_link) {
1797 if (ifa->ifa_addr && ifa->ifa_addr->sa_family == AF_INET){
1798 ipv6_addr_set_v4mapped(
1799 ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr.s_addr,
1800 (struct in6_addr *)&gid);
1801 update_gid_table(ibdev, port, &gid, 0, 0);
1802 }
1803
1804 }
1805 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1806 /* IPv6 gids */
1807 in6_dev = in6_dev_get(dev);
1808 if (in6_dev) {
1809 read_lock_bh(&in6_dev->lock);
1810 list_for_each_entry(ifp, &in6_dev->addr_list, if_list) {
1811 pgid = (union ib_gid *)&ifp->addr;
1812 update_gid_table(ibdev, port, pgid, 0, 0);
1813 }
1814 read_unlock_bh(&in6_dev->lock);
1815 in6_dev_put(in6_dev);
1816 }
1817 #endif
1818 }
1819
mlx4_set_default_gid(struct mlx4_ib_dev * ibdev,struct net_device * dev,u8 port)1820 static void mlx4_set_default_gid(struct mlx4_ib_dev *ibdev,
1821 struct net_device *dev, u8 port)
1822 {
1823 union ib_gid gid;
1824 mlx4_make_default_gid(dev, &gid, port);
1825 update_gid_table(ibdev, port, &gid, 0, 1);
1826 }
1827
mlx4_ib_init_gid_table(struct mlx4_ib_dev * ibdev)1828 static int mlx4_ib_init_gid_table(struct mlx4_ib_dev *ibdev)
1829 {
1830 struct net_device *dev;
1831
1832 if (reset_gid_table(ibdev))
1833 return -1;
1834
1835 IFNET_RLOCK_NOSLEEP();
1836 TAILQ_FOREACH(dev, &V_ifnet, if_link) {
1837 u8 port = mlx4_ib_get_dev_port(dev, ibdev);
1838 if (port) {
1839 if (!rdma_vlan_dev_real_dev(dev) &&
1840 !netif_is_bond_master(dev))
1841 mlx4_set_default_gid(ibdev, dev, port);
1842 mlx4_ib_get_dev_addr(dev, ibdev, port);
1843 }
1844 }
1845
1846 IFNET_RUNLOCK_NOSLEEP();
1847
1848 return 0;
1849 }
1850
mlx4_ib_scan_netdevs(struct mlx4_ib_dev * ibdev,struct net_device * dev,unsigned long event)1851 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev,
1852 struct net_device *dev, unsigned long event)
1853 {
1854 struct mlx4_ib_iboe *iboe;
1855 int port;
1856 int init = 0;
1857 unsigned long flags;
1858
1859 iboe = &ibdev->iboe;
1860
1861 spin_lock_irqsave(&iboe->lock, flags);
1862 mlx4_foreach_ib_transport_port(port, ibdev->dev) {
1863 struct net_device *old_netdev = iboe->netdevs[port - 1];
1864 /* XXX BOND related */
1865 #if 0
1866 struct net_device *old_master = iboe->masters[port - 1];
1867 #endif
1868 iboe->masters[port - 1] = NULL;
1869 iboe->netdevs[port - 1] =
1870 mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
1871
1872
1873 if (old_netdev != iboe->netdevs[port - 1])
1874 init = 1;
1875 if (dev == iboe->netdevs[port - 1] &&
1876 event == NETDEV_CHANGEADDR)
1877 init = 1;
1878 /* XXX BOND related */
1879 #if 0
1880 if (iboe->netdevs[port - 1] && netif_is_bond_slave(iboe->netdevs[port - 1]))
1881 iboe->masters[port - 1] = iboe->netdevs[port - 1]->master;
1882
1883 /* if bonding is used it is possible that we add it to masters only after
1884 IP address is assigned to the net bonding interface */
1885 if (old_master != iboe->masters[port - 1])
1886 init = 1;
1887 #endif
1888 }
1889
1890 spin_unlock_irqrestore(&iboe->lock, flags);
1891
1892 if (init)
1893 if (mlx4_ib_init_gid_table(ibdev))
1894 pr_warn("Fail to reset gid table\n");
1895 }
1896
mlx4_ib_netdev_event(struct notifier_block * this,unsigned long event,void * ptr)1897 static int mlx4_ib_netdev_event(struct notifier_block *this, unsigned long event,
1898 void *ptr)
1899 {
1900 struct net_device *dev = ptr;
1901 struct mlx4_ib_dev *ibdev;
1902
1903 ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
1904
1905 mlx4_ib_scan_netdevs(ibdev, dev, event);
1906
1907 return NOTIFY_DONE;
1908 }
1909
1910 /* This function initializes the gid table only if the event_netdev real device is an iboe
1911 * device, will be invoked by the inet/inet6 events */
mlx4_ib_inet_event(struct notifier_block * this,unsigned long event,void * ptr)1912 static int mlx4_ib_inet_event(struct notifier_block *this, unsigned long event,
1913 void *ptr)
1914 {
1915 struct net_device *event_netdev = ptr;
1916 struct mlx4_ib_dev *ibdev;
1917 struct mlx4_ib_iboe *ibdev_iboe;
1918 int port = 0;
1919
1920 ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb_inet);
1921
1922 struct net_device *real_dev = rdma_vlan_dev_real_dev(event_netdev) ?
1923 rdma_vlan_dev_real_dev(event_netdev) :
1924 event_netdev;
1925
1926 ibdev_iboe = &ibdev->iboe;
1927
1928 port = mlx4_ib_get_dev_port(real_dev, ibdev);
1929
1930 /* Perform init_gid_table if the event real_dev is the net_device which represents this port,
1931 * otherwise this event is not related and would be ignored.*/
1932 if(port && (real_dev == ibdev_iboe->netdevs[port - 1]))
1933 if (mlx4_ib_init_gid_table(ibdev))
1934 pr_warn("Fail to reset gid table\n");
1935
1936 return NOTIFY_DONE;
1937 }
1938
1939
init_pkeys(struct mlx4_ib_dev * ibdev)1940 static void init_pkeys(struct mlx4_ib_dev *ibdev)
1941 {
1942 int port;
1943 int slave;
1944 int i;
1945
1946 if (mlx4_is_master(ibdev->dev)) {
1947 for (slave = 0; slave <= ibdev->dev->num_vfs; ++slave) {
1948 for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
1949 for (i = 0;
1950 i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
1951 ++i) {
1952 ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
1953 /* master has the identity virt2phys pkey mapping */
1954 (slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
1955 ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
1956 mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
1957 ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
1958 }
1959 }
1960 }
1961 /* initialize pkey cache */
1962 for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
1963 for (i = 0;
1964 i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
1965 ++i)
1966 ibdev->pkeys.phys_pkey_cache[port-1][i] =
1967 (i) ? 0 : 0xFFFF;
1968 }
1969 }
1970 }
1971
mlx4_ib_alloc_eqs(struct mlx4_dev * dev,struct mlx4_ib_dev * ibdev)1972 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
1973 {
1974 char name[32];
1975 int eq_per_port = 0;
1976 int added_eqs = 0;
1977 int total_eqs = 0;
1978 int i, j, eq;
1979
1980 /* Legacy mode or comp_pool is not large enough */
1981 if (dev->caps.comp_pool == 0 ||
1982 dev->caps.num_ports > dev->caps.comp_pool)
1983 return;
1984
1985 eq_per_port = rounddown_pow_of_two(dev->caps.comp_pool/
1986 dev->caps.num_ports);
1987
1988 /* Init eq table */
1989 added_eqs = 0;
1990 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
1991 added_eqs += eq_per_port;
1992
1993 total_eqs = dev->caps.num_comp_vectors + added_eqs;
1994
1995 ibdev->eq_table = kzalloc(total_eqs * sizeof(int), GFP_KERNEL);
1996 if (!ibdev->eq_table)
1997 return;
1998
1999 ibdev->eq_added = added_eqs;
2000
2001 eq = 0;
2002 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB) {
2003 for (j = 0; j < eq_per_port; j++) {
2004 sprintf(name, "mlx4-ib-%d-%d@%d:%d:%d:%d", i, j,
2005 pci_get_domain(dev->pdev->dev.bsddev),
2006 pci_get_bus(dev->pdev->dev.bsddev),
2007 PCI_SLOT(dev->pdev->devfn),
2008 PCI_FUNC(dev->pdev->devfn));
2009
2010 /* Set IRQ for specific name (per ring) */
2011 if (mlx4_assign_eq(dev, name,
2012 &ibdev->eq_table[eq])) {
2013 /* Use legacy (same as mlx4_en driver) */
2014 pr_warn("Can't allocate EQ %d; reverting to legacy\n", eq);
2015 ibdev->eq_table[eq] =
2016 (eq % dev->caps.num_comp_vectors);
2017 }
2018 eq++;
2019 }
2020 }
2021
2022 /* Fill the reset of the vector with legacy EQ */
2023 for (i = 0, eq = added_eqs; i < dev->caps.num_comp_vectors; i++)
2024 ibdev->eq_table[eq++] = i;
2025
2026 /* Advertise the new number of EQs to clients */
2027 ibdev->ib_dev.num_comp_vectors = total_eqs;
2028 }
2029
mlx4_ib_free_eqs(struct mlx4_dev * dev,struct mlx4_ib_dev * ibdev)2030 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2031 {
2032 int i;
2033
2034 /* no additional eqs were added */
2035 if (!ibdev->eq_table)
2036 return;
2037
2038 /* Reset the advertised EQ number */
2039 ibdev->ib_dev.num_comp_vectors = dev->caps.num_comp_vectors;
2040
2041 /* Free only the added eqs */
2042 for (i = 0; i < ibdev->eq_added; i++) {
2043 /* Don't free legacy eqs if used */
2044 if (ibdev->eq_table[i] <= dev->caps.num_comp_vectors)
2045 continue;
2046 mlx4_release_eq(dev, ibdev->eq_table[i]);
2047 }
2048
2049 kfree(ibdev->eq_table);
2050 }
2051
2052 /*
2053 * create show function and a device_attribute struct pointing to
2054 * the function for _name
2055 */
2056 #define DEVICE_DIAG_RPRT_ATTR(_name, _offset, _op_mod) \
2057 static ssize_t show_rprt_##_name(struct device *dev, \
2058 struct device_attribute *attr, \
2059 char *buf){ \
2060 return show_diag_rprt(dev, buf, _offset, _op_mod); \
2061 } \
2062 static DEVICE_ATTR(_name, S_IRUGO, show_rprt_##_name, NULL);
2063
2064 #define MLX4_DIAG_RPRT_CLEAR_DIAGS 3
2065
show_diag_rprt(struct device * device,char * buf,u32 offset,u8 op_modifier)2066 static size_t show_diag_rprt(struct device *device, char *buf,
2067 u32 offset, u8 op_modifier)
2068 {
2069 size_t ret;
2070 u32 counter_offset = offset;
2071 u32 diag_counter = 0;
2072 struct mlx4_ib_dev *dev = container_of(device, struct mlx4_ib_dev,
2073 ib_dev.dev);
2074
2075 ret = mlx4_query_diag_counters(dev->dev, 1, op_modifier,
2076 &counter_offset, &diag_counter);
2077 if (ret)
2078 return ret;
2079
2080 return sprintf(buf, "%d\n", diag_counter);
2081 }
2082
clear_diag_counters(struct device * device,struct device_attribute * attr,const char * buf,size_t length)2083 static ssize_t clear_diag_counters(struct device *device,
2084 struct device_attribute *attr,
2085 const char *buf, size_t length)
2086 {
2087 size_t ret;
2088 struct mlx4_ib_dev *dev = container_of(device, struct mlx4_ib_dev,
2089 ib_dev.dev);
2090
2091 ret = mlx4_query_diag_counters(dev->dev, 0, MLX4_DIAG_RPRT_CLEAR_DIAGS,
2092 NULL, NULL);
2093 if (ret)
2094 return ret;
2095
2096 return length;
2097 }
2098
2099 DEVICE_DIAG_RPRT_ATTR(rq_num_lle , 0x00, 2);
2100 DEVICE_DIAG_RPRT_ATTR(sq_num_lle , 0x04, 2);
2101 DEVICE_DIAG_RPRT_ATTR(rq_num_lqpoe , 0x08, 2);
2102 DEVICE_DIAG_RPRT_ATTR(sq_num_lqpoe , 0x0C, 2);
2103 DEVICE_DIAG_RPRT_ATTR(rq_num_lpe , 0x18, 2);
2104 DEVICE_DIAG_RPRT_ATTR(sq_num_lpe , 0x1C, 2);
2105 DEVICE_DIAG_RPRT_ATTR(rq_num_wrfe , 0x20, 2);
2106 DEVICE_DIAG_RPRT_ATTR(sq_num_wrfe , 0x24, 2);
2107 DEVICE_DIAG_RPRT_ATTR(sq_num_mwbe , 0x2C, 2);
2108 DEVICE_DIAG_RPRT_ATTR(sq_num_bre , 0x34, 2);
2109 DEVICE_DIAG_RPRT_ATTR(rq_num_lae , 0x38, 2);
2110 DEVICE_DIAG_RPRT_ATTR(sq_num_rire , 0x44, 2);
2111 DEVICE_DIAG_RPRT_ATTR(rq_num_rire , 0x48, 2);
2112 DEVICE_DIAG_RPRT_ATTR(sq_num_rae , 0x4C, 2);
2113 DEVICE_DIAG_RPRT_ATTR(rq_num_rae , 0x50, 2);
2114 DEVICE_DIAG_RPRT_ATTR(sq_num_roe , 0x54, 2);
2115 DEVICE_DIAG_RPRT_ATTR(sq_num_tree , 0x5C, 2);
2116 DEVICE_DIAG_RPRT_ATTR(sq_num_rree , 0x64, 2);
2117 DEVICE_DIAG_RPRT_ATTR(rq_num_rnr , 0x68, 2);
2118 DEVICE_DIAG_RPRT_ATTR(sq_num_rnr , 0x6C, 2);
2119 DEVICE_DIAG_RPRT_ATTR(rq_num_oos , 0x100, 2);
2120 DEVICE_DIAG_RPRT_ATTR(sq_num_oos , 0x104, 2);
2121 DEVICE_DIAG_RPRT_ATTR(rq_num_mce , 0x108, 2);
2122 DEVICE_DIAG_RPRT_ATTR(rq_num_udsdprd , 0x118, 2);
2123 DEVICE_DIAG_RPRT_ATTR(rq_num_ucsdprd , 0x120, 2);
2124 DEVICE_DIAG_RPRT_ATTR(num_cqovf , 0x1A0, 2);
2125 DEVICE_DIAG_RPRT_ATTR(num_eqovf , 0x1A4, 2);
2126 DEVICE_DIAG_RPRT_ATTR(num_baddb , 0x1A8, 2);
2127
2128 static DEVICE_ATTR(clear_diag, S_IWUSR, NULL, clear_diag_counters);
2129
2130 static struct attribute *diag_rprt_attrs[] = {
2131 &dev_attr_rq_num_lle.attr,
2132 &dev_attr_sq_num_lle.attr,
2133 &dev_attr_rq_num_lqpoe.attr,
2134 &dev_attr_sq_num_lqpoe.attr,
2135 &dev_attr_rq_num_lpe.attr,
2136 &dev_attr_sq_num_lpe.attr,
2137 &dev_attr_rq_num_wrfe.attr,
2138 &dev_attr_sq_num_wrfe.attr,
2139 &dev_attr_sq_num_mwbe.attr,
2140 &dev_attr_sq_num_bre.attr,
2141 &dev_attr_rq_num_lae.attr,
2142 &dev_attr_sq_num_rire.attr,
2143 &dev_attr_rq_num_rire.attr,
2144 &dev_attr_sq_num_rae.attr,
2145 &dev_attr_rq_num_rae.attr,
2146 &dev_attr_sq_num_roe.attr,
2147 &dev_attr_sq_num_tree.attr,
2148 &dev_attr_sq_num_rree.attr,
2149 &dev_attr_rq_num_rnr.attr,
2150 &dev_attr_sq_num_rnr.attr,
2151 &dev_attr_rq_num_oos.attr,
2152 &dev_attr_sq_num_oos.attr,
2153 &dev_attr_rq_num_mce.attr,
2154 &dev_attr_rq_num_udsdprd.attr,
2155 &dev_attr_rq_num_ucsdprd.attr,
2156 &dev_attr_num_cqovf.attr,
2157 &dev_attr_num_eqovf.attr,
2158 &dev_attr_num_baddb.attr,
2159 &dev_attr_clear_diag.attr,
2160 NULL
2161 };
2162
2163 static struct attribute_group diag_counters_group = {
2164 .name = "diag_counters",
2165 .attrs = diag_rprt_attrs
2166 };
2167
init_dev_assign(void)2168 static void init_dev_assign(void)
2169 {
2170 int i = 1;
2171
2172 spin_lock_init(&dev_num_str_lock);
2173 if (mlx4_fill_dbdf2val_tbl(&dev_assign_str))
2174 return;
2175 dev_num_str_bitmap =
2176 kmalloc(BITS_TO_LONGS(MAX_NUM_STR_BITMAP) * sizeof(long),
2177 GFP_KERNEL);
2178 if (!dev_num_str_bitmap) {
2179 pr_warn("bitmap alloc failed -- cannot apply dev_assign_str parameter\n");
2180 return;
2181 }
2182 bitmap_zero(dev_num_str_bitmap, MAX_NUM_STR_BITMAP);
2183 while ((i < MLX4_DEVS_TBL_SIZE) && (dev_assign_str.tbl[i].dbdf !=
2184 MLX4_ENDOF_TBL)) {
2185 if (bitmap_allocate_region(dev_num_str_bitmap,
2186 dev_assign_str.tbl[i].val[0], 0))
2187 goto err;
2188 i++;
2189 }
2190 dr_active = 1;
2191 return;
2192
2193 err:
2194 kfree(dev_num_str_bitmap);
2195 dev_num_str_bitmap = NULL;
2196 pr_warn("mlx4_ib: The value of 'dev_assign_str' parameter "
2197 "is incorrect. The parameter value is discarded!");
2198 }
2199
mlx4_ib_dev_idx(struct mlx4_dev * dev)2200 static int mlx4_ib_dev_idx(struct mlx4_dev *dev)
2201 {
2202 int i, val;
2203
2204 if (!dr_active)
2205 return -1;
2206 if (!dev)
2207 return -1;
2208 if (mlx4_get_val(dev_assign_str.tbl, dev->pdev, 0, &val))
2209 return -1;
2210
2211 if (val != DEFAULT_TBL_VAL) {
2212 dev->flags |= MLX4_FLAG_DEV_NUM_STR;
2213 return val;
2214 }
2215
2216 spin_lock(&dev_num_str_lock);
2217 i = bitmap_find_free_region(dev_num_str_bitmap, MAX_NUM_STR_BITMAP, 0);
2218 spin_unlock(&dev_num_str_lock);
2219 if (i >= 0)
2220 return i;
2221
2222 return -1;
2223 }
2224
mlx4_ib_add(struct mlx4_dev * dev)2225 static void *mlx4_ib_add(struct mlx4_dev *dev)
2226 {
2227 struct mlx4_ib_dev *ibdev;
2228 int num_ports = 0;
2229 int i, j;
2230 int err;
2231 struct mlx4_ib_iboe *iboe;
2232 int dev_idx;
2233
2234 pr_info_once("%s", mlx4_ib_version);
2235
2236 mlx4_foreach_ib_transport_port(i, dev)
2237 num_ports++;
2238
2239 /* No point in registering a device with no ports... */
2240 if (num_ports == 0)
2241 return NULL;
2242
2243 ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
2244 if (!ibdev) {
2245 dev_err(&dev->pdev->dev, "Device struct alloc failed\n");
2246 return NULL;
2247 }
2248
2249 iboe = &ibdev->iboe;
2250
2251 if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
2252 goto err_dealloc;
2253
2254 if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
2255 goto err_pd;
2256
2257 ibdev->priv_uar.map = ioremap(ibdev->priv_uar.pfn << PAGE_SHIFT,
2258 PAGE_SIZE);
2259
2260 if (!ibdev->priv_uar.map)
2261 goto err_uar;
2262
2263 MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
2264
2265 ibdev->dev = dev;
2266
2267 dev_idx = mlx4_ib_dev_idx(dev);
2268 if (dev_idx >= 0)
2269 sprintf(ibdev->ib_dev.name, "mlx4_%d", dev_idx);
2270 else
2271 strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
2272
2273 ibdev->ib_dev.owner = THIS_MODULE;
2274 ibdev->ib_dev.node_type = RDMA_NODE_IB_CA;
2275 ibdev->ib_dev.local_dma_lkey = dev->caps.reserved_lkey;
2276 ibdev->num_ports = num_ports;
2277 ibdev->ib_dev.phys_port_cnt = ibdev->num_ports;
2278 ibdev->ib_dev.num_comp_vectors = dev->caps.num_comp_vectors;
2279 ibdev->ib_dev.dma_device = &dev->pdev->dev;
2280
2281 if (dev->caps.userspace_caps)
2282 ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
2283 else
2284 ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
2285
2286 ibdev->ib_dev.uverbs_cmd_mask =
2287 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT) |
2288 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE) |
2289 (1ull << IB_USER_VERBS_CMD_QUERY_PORT) |
2290 (1ull << IB_USER_VERBS_CMD_ALLOC_PD) |
2291 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD) |
2292 (1ull << IB_USER_VERBS_CMD_REG_MR) |
2293 (1ull << IB_USER_VERBS_CMD_DEREG_MR) |
2294 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
2295 (1ull << IB_USER_VERBS_CMD_CREATE_CQ) |
2296 (1ull << IB_USER_VERBS_CMD_RESIZE_CQ) |
2297 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ) |
2298 (1ull << IB_USER_VERBS_CMD_CREATE_QP) |
2299 (1ull << IB_USER_VERBS_CMD_MODIFY_QP) |
2300 (1ull << IB_USER_VERBS_CMD_QUERY_QP) |
2301 (1ull << IB_USER_VERBS_CMD_DESTROY_QP) |
2302 (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST) |
2303 (1ull << IB_USER_VERBS_CMD_DETACH_MCAST) |
2304 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ) |
2305 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ) |
2306 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ) |
2307 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ) |
2308 (1ull << IB_USER_VERBS_CMD_CREATE_XSRQ) |
2309 (1ull << IB_USER_VERBS_CMD_OPEN_QP);
2310
2311 ibdev->ib_dev.query_device = mlx4_ib_query_device;
2312 ibdev->ib_dev.query_port = mlx4_ib_query_port;
2313 ibdev->ib_dev.get_link_layer = mlx4_ib_port_link_layer;
2314 ibdev->ib_dev.query_gid = mlx4_ib_query_gid;
2315 ibdev->ib_dev.query_pkey = mlx4_ib_query_pkey;
2316 ibdev->ib_dev.modify_device = mlx4_ib_modify_device;
2317 ibdev->ib_dev.modify_port = mlx4_ib_modify_port;
2318 ibdev->ib_dev.alloc_ucontext = mlx4_ib_alloc_ucontext;
2319 ibdev->ib_dev.dealloc_ucontext = mlx4_ib_dealloc_ucontext;
2320 ibdev->ib_dev.mmap = mlx4_ib_mmap;
2321 /* XXX FBSD has no support for get_unmapped_area function */
2322 #if 0
2323 ibdev->ib_dev.get_unmapped_area = mlx4_ib_get_unmapped_area;
2324 #endif
2325 ibdev->ib_dev.alloc_pd = mlx4_ib_alloc_pd;
2326 ibdev->ib_dev.dealloc_pd = mlx4_ib_dealloc_pd;
2327 ibdev->ib_dev.create_ah = mlx4_ib_create_ah;
2328 ibdev->ib_dev.query_ah = mlx4_ib_query_ah;
2329 ibdev->ib_dev.destroy_ah = mlx4_ib_destroy_ah;
2330 ibdev->ib_dev.create_srq = mlx4_ib_create_srq;
2331 ibdev->ib_dev.modify_srq = mlx4_ib_modify_srq;
2332 ibdev->ib_dev.query_srq = mlx4_ib_query_srq;
2333 ibdev->ib_dev.destroy_srq = mlx4_ib_destroy_srq;
2334 ibdev->ib_dev.post_srq_recv = mlx4_ib_post_srq_recv;
2335 ibdev->ib_dev.create_qp = mlx4_ib_create_qp;
2336 ibdev->ib_dev.modify_qp = mlx4_ib_modify_qp;
2337 ibdev->ib_dev.query_qp = mlx4_ib_query_qp;
2338 ibdev->ib_dev.destroy_qp = mlx4_ib_destroy_qp;
2339 ibdev->ib_dev.post_send = mlx4_ib_post_send;
2340 ibdev->ib_dev.post_recv = mlx4_ib_post_recv;
2341 ibdev->ib_dev.create_cq = mlx4_ib_create_cq;
2342 ibdev->ib_dev.modify_cq = mlx4_ib_modify_cq;
2343 ibdev->ib_dev.resize_cq = mlx4_ib_resize_cq;
2344 ibdev->ib_dev.destroy_cq = mlx4_ib_destroy_cq;
2345 ibdev->ib_dev.poll_cq = mlx4_ib_poll_cq;
2346 ibdev->ib_dev.req_notify_cq = mlx4_ib_arm_cq;
2347 ibdev->ib_dev.get_dma_mr = mlx4_ib_get_dma_mr;
2348 ibdev->ib_dev.reg_user_mr = mlx4_ib_reg_user_mr;
2349 ibdev->ib_dev.dereg_mr = mlx4_ib_dereg_mr;
2350 ibdev->ib_dev.alloc_fast_reg_mr = mlx4_ib_alloc_fast_reg_mr;
2351 ibdev->ib_dev.alloc_fast_reg_page_list = mlx4_ib_alloc_fast_reg_page_list;
2352 ibdev->ib_dev.free_fast_reg_page_list = mlx4_ib_free_fast_reg_page_list;
2353 ibdev->ib_dev.attach_mcast = mlx4_ib_mcg_attach;
2354 ibdev->ib_dev.detach_mcast = mlx4_ib_mcg_detach;
2355 ibdev->ib_dev.process_mad = mlx4_ib_process_mad;
2356 ibdev->ib_dev.ioctl = mlx4_ib_ioctl;
2357 ibdev->ib_dev.query_values = mlx4_ib_query_values;
2358
2359 if (!mlx4_is_slave(ibdev->dev)) {
2360 ibdev->ib_dev.alloc_fmr = mlx4_ib_fmr_alloc;
2361 ibdev->ib_dev.map_phys_fmr = mlx4_ib_map_phys_fmr;
2362 ibdev->ib_dev.unmap_fmr = mlx4_ib_unmap_fmr;
2363 ibdev->ib_dev.dealloc_fmr = mlx4_ib_fmr_dealloc;
2364 }
2365
2366 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW) {
2367 ibdev->ib_dev.alloc_mw = mlx4_ib_alloc_mw;
2368 ibdev->ib_dev.bind_mw = mlx4_ib_bind_mw;
2369 ibdev->ib_dev.dealloc_mw = mlx4_ib_dealloc_mw;
2370
2371 ibdev->ib_dev.uverbs_cmd_mask |=
2372 (1ull << IB_USER_VERBS_CMD_ALLOC_MW) |
2373 (1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
2374 }
2375
2376 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
2377 ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
2378 ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
2379 ibdev->ib_dev.uverbs_cmd_mask |=
2380 (1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
2381 (1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
2382 }
2383
2384 /*
2385 * Set experimental data
2386 */
2387 ibdev->ib_dev.uverbs_exp_cmd_mask =
2388 (1ull << IB_USER_VERBS_EXP_CMD_CREATE_QP) |
2389 (1ull << IB_USER_VERBS_EXP_CMD_MODIFY_CQ) |
2390 (1ull << IB_USER_VERBS_EXP_CMD_QUERY_DEVICE) |
2391 (1ull << IB_USER_VERBS_EXP_CMD_CREATE_CQ);
2392 ibdev->ib_dev.exp_create_qp = mlx4_ib_exp_create_qp;
2393 ibdev->ib_dev.exp_query_device = mlx4_ib_exp_query_device;
2394 if (check_flow_steering_support(dev)) {
2395 ibdev->ib_dev.uverbs_ex_cmd_mask |=
2396 (1ull << IB_USER_VERBS_EX_CMD_CREATE_FLOW) |
2397 (1ull << IB_USER_VERBS_EX_CMD_DESTROY_FLOW);
2398 ibdev->ib_dev.create_flow = mlx4_ib_create_flow;
2399 ibdev->ib_dev.destroy_flow = mlx4_ib_destroy_flow;
2400 } else {
2401 pr_debug("Device managed flow steering is unavailable for this configuration.\n");
2402 }
2403 /*
2404 * End of experimental data
2405 */
2406
2407 mlx4_ib_alloc_eqs(dev, ibdev);
2408
2409 spin_lock_init(&iboe->lock);
2410
2411 if (init_node_data(ibdev))
2412 goto err_map;
2413
2414 for (i = 0; i < ibdev->num_ports; ++i) {
2415 if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
2416 IB_LINK_LAYER_ETHERNET) {
2417 if (mlx4_is_slave(dev)) {
2418 ibdev->counters[i].status = mlx4_counter_alloc(ibdev->dev,
2419 i + 1,
2420 &ibdev->counters[i].counter_index);
2421 } else {/* allocating the PF IB default counter indices reserved in mlx4_init_counters_table */
2422 ibdev->counters[i].counter_index = ((i + 1) << 1) - 1;
2423 ibdev->counters[i].status = 0;
2424 }
2425
2426 dev_info(&dev->pdev->dev,
2427 "%s: allocated counter index %d for port %d\n",
2428 __func__, ibdev->counters[i].counter_index, i+1);
2429 } else {
2430 ibdev->counters[i].counter_index = MLX4_SINK_COUNTER_INDEX;
2431 ibdev->counters[i].status = -ENOSPC;
2432 }
2433 }
2434
2435 spin_lock_init(&ibdev->sm_lock);
2436 mutex_init(&ibdev->cap_mask_mutex);
2437
2438 if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2439 !mlx4_is_mfunc(dev)) {
2440 ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
2441 err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
2442 MLX4_IB_UC_STEER_QPN_ALIGN, &ibdev->steer_qpn_base, 0);
2443 if (err)
2444 goto err_counter;
2445
2446 ibdev->ib_uc_qpns_bitmap =
2447 kmalloc(BITS_TO_LONGS(ibdev->steer_qpn_count) *
2448 sizeof(long),
2449 GFP_KERNEL);
2450 if (!ibdev->ib_uc_qpns_bitmap) {
2451 dev_err(&dev->pdev->dev, "bit map alloc failed\n");
2452 goto err_steer_qp_release;
2453 }
2454
2455 bitmap_zero(ibdev->ib_uc_qpns_bitmap, ibdev->steer_qpn_count);
2456
2457 err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(dev, ibdev->steer_qpn_base,
2458 ibdev->steer_qpn_base + ibdev->steer_qpn_count - 1);
2459 if (err)
2460 goto err_steer_free_bitmap;
2461 }
2462
2463 if (ib_register_device(&ibdev->ib_dev, NULL))
2464 goto err_steer_free_bitmap;
2465
2466 if (mlx4_ib_mad_init(ibdev))
2467 goto err_reg;
2468
2469 if (mlx4_ib_init_sriov(ibdev))
2470 goto err_mad;
2471
2472 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE) {
2473 if (!iboe->nb.notifier_call) {
2474 iboe->nb.notifier_call = mlx4_ib_netdev_event;
2475 err = register_netdevice_notifier(&iboe->nb);
2476 if (err) {
2477 iboe->nb.notifier_call = NULL;
2478 goto err_notify;
2479 }
2480 }
2481 if (!iboe->nb_inet.notifier_call) {
2482 iboe->nb_inet.notifier_call = mlx4_ib_inet_event;
2483 err = register_inetaddr_notifier(&iboe->nb_inet);
2484 if (err) {
2485 iboe->nb_inet.notifier_call = NULL;
2486 goto err_notify;
2487 }
2488 }
2489 mlx4_ib_scan_netdevs(ibdev, NULL, 0);
2490 }
2491 for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
2492 if (device_create_file(&ibdev->ib_dev.dev,
2493 mlx4_class_attributes[j]))
2494 goto err_notify;
2495 }
2496 if (sysfs_create_group(&ibdev->ib_dev.dev.kobj, &diag_counters_group))
2497 goto err_notify;
2498
2499 ibdev->ib_active = true;
2500
2501 if (mlx4_is_mfunc(ibdev->dev))
2502 init_pkeys(ibdev);
2503
2504 /* create paravirt contexts for any VFs which are active */
2505 if (mlx4_is_master(ibdev->dev)) {
2506 for (j = 0; j < MLX4_MFUNC_MAX; j++) {
2507 if (j == mlx4_master_func_num(ibdev->dev))
2508 continue;
2509 if (mlx4_is_slave_active(ibdev->dev, j))
2510 do_slave_init(ibdev, j, 1);
2511 }
2512 }
2513 return ibdev;
2514
2515 err_notify:
2516 for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
2517 device_remove_file(&ibdev->ib_dev.dev,
2518 mlx4_class_attributes[j]);
2519 }
2520
2521 if (ibdev->iboe.nb.notifier_call) {
2522 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2523 pr_warn("failure unregistering notifier\n");
2524 ibdev->iboe.nb.notifier_call = NULL;
2525 }
2526 if (ibdev->iboe.nb_inet.notifier_call) {
2527 if (unregister_inetaddr_notifier(&ibdev->iboe.nb_inet))
2528 pr_warn("failure unregistering notifier\n");
2529 ibdev->iboe.nb_inet.notifier_call = NULL;
2530 }
2531 flush_workqueue(wq);
2532
2533 mlx4_ib_close_sriov(ibdev);
2534
2535 err_mad:
2536 mlx4_ib_mad_cleanup(ibdev);
2537
2538 err_reg:
2539 ib_unregister_device(&ibdev->ib_dev);
2540
2541 err_steer_free_bitmap:
2542 kfree(ibdev->ib_uc_qpns_bitmap);
2543
2544 err_steer_qp_release:
2545 if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED)
2546 mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2547 ibdev->steer_qpn_count);
2548 err_counter:
2549 for (; i; --i) {
2550 if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i) ==
2551 IB_LINK_LAYER_ETHERNET) {
2552 mlx4_counter_free(ibdev->dev,
2553 i,
2554 ibdev->counters[i - 1].counter_index);
2555 }
2556 }
2557
2558 err_map:
2559 iounmap(ibdev->priv_uar.map);
2560 mlx4_ib_free_eqs(dev, ibdev);
2561
2562 err_uar:
2563 mlx4_uar_free(dev, &ibdev->priv_uar);
2564
2565 err_pd:
2566 mlx4_pd_free(dev, ibdev->priv_pdn);
2567
2568 err_dealloc:
2569 ib_dealloc_device(&ibdev->ib_dev);
2570
2571 return NULL;
2572 }
2573
mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev * dev,int count,int * qpn)2574 int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn)
2575 {
2576 int offset;
2577
2578 WARN_ON(!dev->ib_uc_qpns_bitmap);
2579
2580 offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap,
2581 dev->steer_qpn_count,
2582 get_count_order(count));
2583 if (offset < 0)
2584 return offset;
2585
2586 *qpn = dev->steer_qpn_base + offset;
2587 return 0;
2588 }
2589
mlx4_ib_steer_qp_free(struct mlx4_ib_dev * dev,u32 qpn,int count)2590 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count)
2591 {
2592 if (!qpn ||
2593 dev->dev->caps.steering_mode != MLX4_STEERING_MODE_DEVICE_MANAGED)
2594 return;
2595
2596 BUG_ON(qpn < dev->steer_qpn_base);
2597
2598 bitmap_release_region(dev->ib_uc_qpns_bitmap,
2599 qpn - dev->steer_qpn_base, get_count_order(count));
2600 }
2601
mlx4_ib_steer_qp_reg(struct mlx4_ib_dev * mdev,struct mlx4_ib_qp * mqp,int is_attach)2602 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
2603 int is_attach)
2604 {
2605 int err;
2606 size_t flow_size;
2607 struct ib_flow_attr *flow = NULL;
2608 struct ib_flow_spec_ib *ib_spec;
2609
2610 if (is_attach) {
2611 flow_size = sizeof(struct ib_flow_attr) +
2612 sizeof(struct ib_flow_spec_ib);
2613 flow = kzalloc(flow_size, GFP_KERNEL);
2614 if (!flow)
2615 return -ENOMEM;
2616 flow->port = mqp->port;
2617 flow->num_of_specs = 1;
2618 flow->size = flow_size;
2619 ib_spec = (struct ib_flow_spec_ib *)(flow + 1);
2620 ib_spec->type = IB_FLOW_SPEC_IB;
2621 ib_spec->size = sizeof(struct ib_flow_spec_ib);
2622 ib_spec->val.l3_type_qpn = mqp->ibqp.qp_num;
2623 ib_spec->mask.l3_type_qpn = MLX4_IB_FLOW_QPN_MASK;
2624
2625 err = __mlx4_ib_create_flow(&mqp->ibqp, flow,
2626 IB_FLOW_DOMAIN_NIC,
2627 MLX4_FS_REGULAR,
2628 &mqp->reg_id);
2629 } else {
2630 err = __mlx4_ib_destroy_flow(mdev->dev, mqp->reg_id);
2631 }
2632 kfree(flow);
2633 return err;
2634 }
2635
mlx4_ib_remove(struct mlx4_dev * dev,void * ibdev_ptr)2636 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
2637 {
2638 struct mlx4_ib_dev *ibdev = ibdev_ptr;
2639 int p, j;
2640 int dev_idx, ret;
2641
2642 if (ibdev->iboe.nb_inet.notifier_call) {
2643 if (unregister_inetaddr_notifier(&ibdev->iboe.nb_inet))
2644 pr_warn("failure unregistering notifier\n");
2645 ibdev->iboe.nb_inet.notifier_call = NULL;
2646 }
2647
2648 mlx4_ib_close_sriov(ibdev);
2649 sysfs_remove_group(&ibdev->ib_dev.dev.kobj, &diag_counters_group);
2650 mlx4_ib_mad_cleanup(ibdev);
2651
2652 for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
2653 device_remove_file(&ibdev->ib_dev.dev,
2654 mlx4_class_attributes[j]);
2655 }
2656
2657
2658 dev_idx = -1;
2659 if (dr_active && !(ibdev->dev->flags & MLX4_FLAG_DEV_NUM_STR)) {
2660 ret = sscanf(ibdev->ib_dev.name, "mlx4_%d", &dev_idx);
2661 if (ret != 1)
2662 dev_idx = -1;
2663 }
2664 ib_unregister_device(&ibdev->ib_dev);
2665 if (dev_idx >= 0) {
2666 spin_lock(&dev_num_str_lock);
2667 bitmap_release_region(dev_num_str_bitmap, dev_idx, 0);
2668 spin_unlock(&dev_num_str_lock);
2669 }
2670
2671 if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED) {
2672 mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2673 ibdev->steer_qpn_count);
2674 kfree(ibdev->ib_uc_qpns_bitmap);
2675 }
2676
2677 if (ibdev->iboe.nb.notifier_call) {
2678 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2679 pr_warn("failure unregistering notifier\n");
2680 ibdev->iboe.nb.notifier_call = NULL;
2681 }
2682 iounmap(ibdev->priv_uar.map);
2683
2684 for (p = 0; p < ibdev->num_ports; ++p) {
2685 if (mlx4_ib_port_link_layer(&ibdev->ib_dev, p + 1) ==
2686 IB_LINK_LAYER_ETHERNET) {
2687 mlx4_counter_free(ibdev->dev,
2688 p + 1,
2689 ibdev->counters[p].counter_index);
2690 }
2691 }
2692
2693 mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
2694 mlx4_CLOSE_PORT(dev, p);
2695
2696 mlx4_ib_free_eqs(dev, ibdev);
2697
2698 mlx4_uar_free(dev, &ibdev->priv_uar);
2699 mlx4_pd_free(dev, ibdev->priv_pdn);
2700 ib_dealloc_device(&ibdev->ib_dev);
2701 }
2702
do_slave_init(struct mlx4_ib_dev * ibdev,int slave,int do_init)2703 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
2704 {
2705 struct mlx4_ib_demux_work **dm = NULL;
2706 struct mlx4_dev *dev = ibdev->dev;
2707 int i;
2708 unsigned long flags;
2709
2710 if (!mlx4_is_master(dev))
2711 return;
2712
2713 dm = kcalloc(dev->caps.num_ports, sizeof *dm, GFP_ATOMIC);
2714 if (!dm) {
2715 pr_err("failed to allocate memory for tunneling qp update\n");
2716 goto out;
2717 }
2718
2719 for (i = 0; i < dev->caps.num_ports; i++) {
2720 dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
2721 if (!dm[i]) {
2722 pr_err("failed to allocate memory for tunneling qp update work struct\n");
2723 for (i = 0; i < dev->caps.num_ports; i++) {
2724 if (dm[i])
2725 kfree(dm[i]);
2726 }
2727 goto out;
2728 }
2729 }
2730 /* initialize or tear down tunnel QPs for the slave */
2731 for (i = 0; i < dev->caps.num_ports; i++) {
2732 INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
2733 dm[i]->port = i + 1;
2734 dm[i]->slave = slave;
2735 dm[i]->do_init = do_init;
2736 dm[i]->dev = ibdev;
2737 spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
2738 if (!ibdev->sriov.is_going_down)
2739 queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
2740 spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
2741 }
2742 out:
2743 if (dm)
2744 kfree(dm);
2745 return;
2746 }
2747
mlx4_ib_event(struct mlx4_dev * dev,void * ibdev_ptr,enum mlx4_dev_event event,unsigned long param)2748 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
2749 enum mlx4_dev_event event, unsigned long param)
2750 {
2751 struct ib_event ibev;
2752 struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
2753 struct mlx4_eqe *eqe = NULL;
2754 struct ib_event_work *ew;
2755 int p = 0;
2756
2757 if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
2758 eqe = (struct mlx4_eqe *)param;
2759 else
2760 p = (int) param;
2761
2762 switch (event) {
2763 case MLX4_DEV_EVENT_PORT_UP:
2764 if (p > ibdev->num_ports)
2765 return;
2766 if (mlx4_is_master(dev) &&
2767 rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
2768 IB_LINK_LAYER_INFINIBAND) {
2769 mlx4_ib_invalidate_all_guid_record(ibdev, p);
2770 }
2771 mlx4_ib_info((struct ib_device *) ibdev_ptr,
2772 "Port %d logical link is up\n", p);
2773 ibev.event = IB_EVENT_PORT_ACTIVE;
2774 break;
2775
2776 case MLX4_DEV_EVENT_PORT_DOWN:
2777 if (p > ibdev->num_ports)
2778 return;
2779 mlx4_ib_info((struct ib_device *) ibdev_ptr,
2780 "Port %d logical link is down\n", p);
2781 ibev.event = IB_EVENT_PORT_ERR;
2782 break;
2783
2784 case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
2785 ibdev->ib_active = false;
2786 ibev.event = IB_EVENT_DEVICE_FATAL;
2787 break;
2788
2789 case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
2790 ew = kmalloc(sizeof *ew, GFP_ATOMIC);
2791 if (!ew) {
2792 pr_err("failed to allocate memory for events work\n");
2793 break;
2794 }
2795
2796 INIT_WORK(&ew->work, handle_port_mgmt_change_event);
2797 memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
2798 ew->ib_dev = ibdev;
2799 /* need to queue only for port owner, which uses GEN_EQE */
2800 if (mlx4_is_master(dev))
2801 queue_work(wq, &ew->work);
2802 else
2803 handle_port_mgmt_change_event(&ew->work);
2804 return;
2805
2806 case MLX4_DEV_EVENT_SLAVE_INIT:
2807 /* here, p is the slave id */
2808 do_slave_init(ibdev, p, 1);
2809 return;
2810
2811 case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
2812 /* here, p is the slave id */
2813 do_slave_init(ibdev, p, 0);
2814 return;
2815
2816 default:
2817 return;
2818 }
2819
2820 ibev.device = ibdev_ptr;
2821 ibev.element.port_num = (u8) p;
2822
2823 ib_dispatch_event(&ibev);
2824 }
2825
2826 static struct mlx4_interface mlx4_ib_interface = {
2827 .add = mlx4_ib_add,
2828 .remove = mlx4_ib_remove,
2829 .event = mlx4_ib_event,
2830 .protocol = MLX4_PROT_IB_IPV6
2831 };
2832
mlx4_ib_init(void)2833 static int __init mlx4_ib_init(void)
2834 {
2835 int err;
2836
2837 wq = create_singlethread_workqueue("mlx4_ib");
2838 if (!wq)
2839 return -ENOMEM;
2840
2841 err = mlx4_ib_mcg_init();
2842 if (err)
2843 goto clean_proc;
2844
2845 init_dev_assign();
2846
2847 err = mlx4_register_interface(&mlx4_ib_interface);
2848 if (err)
2849 goto clean_mcg;
2850
2851 return 0;
2852
2853 clean_mcg:
2854 mlx4_ib_mcg_destroy();
2855
2856 clean_proc:
2857 destroy_workqueue(wq);
2858 return err;
2859 }
2860
mlx4_ib_cleanup(void)2861 static void __exit mlx4_ib_cleanup(void)
2862 {
2863 mlx4_unregister_interface(&mlx4_ib_interface);
2864 mlx4_ib_mcg_destroy();
2865 destroy_workqueue(wq);
2866
2867 kfree(dev_num_str_bitmap);
2868 }
2869
2870 module_init_order(mlx4_ib_init, SI_ORDER_MIDDLE);
2871 module_exit(mlx4_ib_cleanup);
2872
2873 static int
mlx4ib_evhand(module_t mod,int event,void * arg)2874 mlx4ib_evhand(module_t mod, int event, void *arg)
2875 {
2876 return (0);
2877 }
2878
2879 static moduledata_t mlx4ib_mod = {
2880 .name = "mlx4ib",
2881 .evhand = mlx4ib_evhand,
2882 };
2883
2884 DECLARE_MODULE(mlx4ib, mlx4ib_mod, SI_SUB_SMP, SI_ORDER_ANY);
2885 MODULE_DEPEND(mlx4ib, mlx4, 1, 1, 1);
2886 MODULE_DEPEND(mlx4ib, ibcore, 1, 1, 1);
2887 MODULE_DEPEND(mlx4ib, linuxkpi, 1, 1, 1);
2888