Merge tag 'md-3.6' of git://neil.brown.name/md
[linux-2.6-block.git] / drivers / infiniband / hw / mlx4 / main.c
CommitLineData
225c7b1f
RD
1/*
2 * Copyright (c) 2006, 2007 Cisco Systems, Inc. All rights reserved.
51a379d0 3 * Copyright (c) 2007, 2008 Mellanox Technologies. All rights reserved.
225c7b1f
RD
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/init.h>
5a0e3ad6 36#include <linux/slab.h>
225c7b1f 37#include <linux/errno.h>
fa417f7b
EC
38#include <linux/netdevice.h>
39#include <linux/inetdevice.h>
40#include <linux/rtnetlink.h>
4c3eb3ca 41#include <linux/if_vlan.h>
225c7b1f
RD
42
43#include <rdma/ib_smi.h>
44#include <rdma/ib_user_verbs.h>
fa417f7b 45#include <rdma/ib_addr.h>
225c7b1f
RD
46
47#include <linux/mlx4/driver.h>
48#include <linux/mlx4/cmd.h>
49
50#include "mlx4_ib.h"
51#include "user.h"
52
b1d8eb5a 53#define DRV_NAME MLX4_IB_DRV_NAME
068c4ea1
JM
54#define DRV_VERSION "1.0"
55#define DRV_RELDATE "April 4, 2008"
225c7b1f
RD
56
57MODULE_AUTHOR("Roland Dreier");
58MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
59MODULE_LICENSE("Dual BSD/GPL");
60MODULE_VERSION(DRV_VERSION);
61
68f3948d 62static const char mlx4_ib_version[] =
225c7b1f
RD
63 DRV_NAME ": Mellanox ConnectX InfiniBand driver v"
64 DRV_VERSION " (" DRV_RELDATE ")\n";
65
fa417f7b
EC
66struct update_gid_work {
67 struct work_struct work;
68 union ib_gid gids[128];
69 struct mlx4_ib_dev *dev;
70 int port;
71};
72
73static struct workqueue_struct *wq;
74
225c7b1f
RD
75static void init_query_mad(struct ib_smp *mad)
76{
77 mad->base_version = 1;
78 mad->mgmt_class = IB_MGMT_CLASS_SUBN_LID_ROUTED;
79 mad->class_version = 1;
80 mad->method = IB_MGMT_METHOD_GET;
81}
82
4c3eb3ca
EC
83static union ib_gid zgid;
84
225c7b1f
RD
85static int mlx4_ib_query_device(struct ib_device *ibdev,
86 struct ib_device_attr *props)
87{
88 struct mlx4_ib_dev *dev = to_mdev(ibdev);
89 struct ib_smp *in_mad = NULL;
90 struct ib_smp *out_mad = NULL;
91 int err = -ENOMEM;
92
93 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
94 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
95 if (!in_mad || !out_mad)
96 goto out;
97
98 init_query_mad(in_mad);
99 in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
100
101 err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, 1, NULL, NULL, in_mad, out_mad);
102 if (err)
103 goto out;
104
105 memset(props, 0, sizeof *props);
106
107 props->fw_ver = dev->dev->caps.fw_ver;
108 props->device_cap_flags = IB_DEVICE_CHANGE_PHY_PORT |
109 IB_DEVICE_PORT_ACTIVE_EVENT |
110 IB_DEVICE_SYS_IMAGE_GUID |
521e575b
RL
111 IB_DEVICE_RC_RNR_NAK_GEN |
112 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
225c7b1f
RD
113 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
114 props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
115 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
116 props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
117 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM)
118 props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
119 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
120 props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
8ff095ec
EC
121 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
122 props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
417608c2 123 if (dev->dev->caps.max_gso_sz && dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH)
b832be1e 124 props->device_cap_flags |= IB_DEVICE_UD_TSO;
95d04f07
RD
125 if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
126 props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
127 if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
128 (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
129 (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
130 props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
0a1405da
SH
131 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
132 props->device_cap_flags |= IB_DEVICE_XRC;
225c7b1f
RD
133
134 props->vendor_id = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
135 0xffffff;
136 props->vendor_part_id = be16_to_cpup((__be16 *) (out_mad->data + 30));
137 props->hw_ver = be32_to_cpup((__be32 *) (out_mad->data + 32));
138 memcpy(&props->sys_image_guid, out_mad->data + 4, 8);
139
140 props->max_mr_size = ~0ull;
141 props->page_size_cap = dev->dev->caps.page_size_cap;
142 props->max_qp = dev->dev->caps.num_qps - dev->dev->caps.reserved_qps;
fc2d0044 143 props->max_qp_wr = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
225c7b1f
RD
144 props->max_sge = min(dev->dev->caps.max_sq_sg,
145 dev->dev->caps.max_rq_sg);
146 props->max_cq = dev->dev->caps.num_cqs - dev->dev->caps.reserved_cqs;
147 props->max_cqe = dev->dev->caps.max_cqes;
148 props->max_mr = dev->dev->caps.num_mpts - dev->dev->caps.reserved_mrws;
149 props->max_pd = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
150 props->max_qp_rd_atom = dev->dev->caps.max_qp_dest_rdma;
151 props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
152 props->max_res_rd_atom = props->max_qp_rd_atom * props->max_qp;
153 props->max_srq = dev->dev->caps.num_srqs - dev->dev->caps.reserved_srqs;
c8681f14 154 props->max_srq_wr = dev->dev->caps.max_srq_wqes - 1;
225c7b1f 155 props->max_srq_sge = dev->dev->caps.max_srq_sge;
5a0fd094 156 props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
225c7b1f
RD
157 props->local_ca_ack_delay = dev->dev->caps.local_ca_ack_delay;
158 props->atomic_cap = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
159 IB_ATOMIC_HCA : IB_ATOMIC_NONE;
47e956b2 160 props->masked_atomic_cap = props->atomic_cap;
5ae2a7a8 161 props->max_pkeys = dev->dev->caps.pkey_table_len[1];
225c7b1f
RD
162 props->max_mcast_grp = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
163 props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
164 props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
165 props->max_mcast_grp;
a5bbe892 166 props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
225c7b1f
RD
167
168out:
169 kfree(in_mad);
170 kfree(out_mad);
171
172 return err;
173}
174
fa417f7b
EC
175static enum rdma_link_layer
176mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
225c7b1f 177{
fa417f7b 178 struct mlx4_dev *dev = to_mdev(device)->dev;
225c7b1f 179
65dab25d 180 return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
fa417f7b
EC
181 IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
182}
225c7b1f 183
fa417f7b 184static int ib_link_query_port(struct ib_device *ibdev, u8 port,
a9c766bb 185 struct ib_port_attr *props)
fa417f7b 186{
a9c766bb
OG
187 struct ib_smp *in_mad = NULL;
188 struct ib_smp *out_mad = NULL;
a5e12dff 189 int ext_active_speed;
a9c766bb
OG
190 int err = -ENOMEM;
191
192 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
193 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
194 if (!in_mad || !out_mad)
195 goto out;
196
197 init_query_mad(in_mad);
198 in_mad->attr_id = IB_SMP_ATTR_PORT_INFO;
199 in_mad->attr_mod = cpu_to_be32(port);
200
201 err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL,
202 in_mad, out_mad);
203 if (err)
204 goto out;
205
a5e12dff 206
225c7b1f
RD
207 props->lid = be16_to_cpup((__be16 *) (out_mad->data + 16));
208 props->lmc = out_mad->data[34] & 0x7;
209 props->sm_lid = be16_to_cpup((__be16 *) (out_mad->data + 18));
210 props->sm_sl = out_mad->data[36] & 0xf;
211 props->state = out_mad->data[32] & 0xf;
212 props->phys_state = out_mad->data[33] >> 4;
213 props->port_cap_flags = be32_to_cpup((__be32 *) (out_mad->data + 20));
5ae2a7a8 214 props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
149983af 215 props->max_msg_sz = to_mdev(ibdev)->dev->caps.max_msg_sz;
5ae2a7a8 216 props->pkey_tbl_len = to_mdev(ibdev)->dev->caps.pkey_table_len[port];
225c7b1f
RD
217 props->bad_pkey_cntr = be16_to_cpup((__be16 *) (out_mad->data + 46));
218 props->qkey_viol_cntr = be16_to_cpup((__be16 *) (out_mad->data + 48));
219 props->active_width = out_mad->data[31] & 0xf;
220 props->active_speed = out_mad->data[35] >> 4;
221 props->max_mtu = out_mad->data[41] & 0xf;
222 props->active_mtu = out_mad->data[36] >> 4;
223 props->subnet_timeout = out_mad->data[51] & 0x1f;
224 props->max_vl_num = out_mad->data[37] >> 4;
225 props->init_type_reply = out_mad->data[41] >> 4;
226
a5e12dff
MA
227 /* Check if extended speeds (EDR/FDR/...) are supported */
228 if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
229 ext_active_speed = out_mad->data[62] >> 4;
230
231 switch (ext_active_speed) {
232 case 1:
2e96691c 233 props->active_speed = IB_SPEED_FDR;
a5e12dff
MA
234 break;
235 case 2:
2e96691c 236 props->active_speed = IB_SPEED_EDR;
a5e12dff
MA
237 break;
238 }
239 }
240
241 /* If reported active speed is QDR, check if is FDR-10 */
2e96691c 242 if (props->active_speed == IB_SPEED_QDR) {
8154c07f
OG
243 init_query_mad(in_mad);
244 in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
245 in_mad->attr_mod = cpu_to_be32(port);
246
247 err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port,
248 NULL, NULL, in_mad, out_mad);
249 if (err)
bf6b47de 250 goto out;
8154c07f
OG
251
252 /* Checking LinkSpeedActive for FDR-10 */
253 if (out_mad->data[15] & 0x1)
254 props->active_speed = IB_SPEED_FDR10;
a5e12dff 255 }
d2ef4068
OG
256
257 /* Avoid wrong speed value returned by FW if the IB link is down. */
258 if (props->state == IB_PORT_DOWN)
259 props->active_speed = IB_SPEED_SDR;
260
a9c766bb
OG
261out:
262 kfree(in_mad);
263 kfree(out_mad);
264 return err;
fa417f7b
EC
265}
266
267static u8 state_to_phys_state(enum ib_port_state state)
268{
269 return state == IB_PORT_ACTIVE ? 5 : 3;
270}
271
272static int eth_link_query_port(struct ib_device *ibdev, u8 port,
a9c766bb 273 struct ib_port_attr *props)
fa417f7b 274{
a9c766bb
OG
275
276 struct mlx4_ib_dev *mdev = to_mdev(ibdev);
277 struct mlx4_ib_iboe *iboe = &mdev->iboe;
fa417f7b
EC
278 struct net_device *ndev;
279 enum ib_mtu tmp;
a9c766bb
OG
280 struct mlx4_cmd_mailbox *mailbox;
281 int err = 0;
282
283 mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
284 if (IS_ERR(mailbox))
285 return PTR_ERR(mailbox);
fa417f7b 286
a9c766bb
OG
287 err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
288 MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
289 MLX4_CMD_WRAPPED);
290 if (err)
291 goto out;
292
293 props->active_width = (((u8 *)mailbox->buf)[5] == 0x40) ?
294 IB_WIDTH_4X : IB_WIDTH_1X;
2e96691c 295 props->active_speed = IB_SPEED_QDR;
fa417f7b 296 props->port_cap_flags = IB_PORT_CM_SUP;
a9c766bb
OG
297 props->gid_tbl_len = mdev->dev->caps.gid_table_len[port];
298 props->max_msg_sz = mdev->dev->caps.max_msg_sz;
fa417f7b 299 props->pkey_tbl_len = 1;
bcacb897 300 props->max_mtu = IB_MTU_4096;
a9c766bb 301 props->max_vl_num = 2;
fa417f7b
EC
302 props->state = IB_PORT_DOWN;
303 props->phys_state = state_to_phys_state(props->state);
304 props->active_mtu = IB_MTU_256;
305 spin_lock(&iboe->lock);
306 ndev = iboe->netdevs[port - 1];
307 if (!ndev)
a9c766bb 308 goto out_unlock;
fa417f7b
EC
309
310 tmp = iboe_get_mtu(ndev->mtu);
311 props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
312
21d60609 313 props->state = (netif_running(ndev) && netif_carrier_ok(ndev)) ?
fa417f7b
EC
314 IB_PORT_ACTIVE : IB_PORT_DOWN;
315 props->phys_state = state_to_phys_state(props->state);
a9c766bb 316out_unlock:
fa417f7b 317 spin_unlock(&iboe->lock);
a9c766bb
OG
318out:
319 mlx4_free_cmd_mailbox(mdev->dev, mailbox);
320 return err;
fa417f7b
EC
321}
322
323static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
324 struct ib_port_attr *props)
325{
a9c766bb 326 int err;
fa417f7b
EC
327
328 memset(props, 0, sizeof *props);
329
fa417f7b 330 err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
a9c766bb
OG
331 ib_link_query_port(ibdev, port, props) :
332 eth_link_query_port(ibdev, port, props);
225c7b1f
RD
333
334 return err;
335}
336
fa417f7b
EC
337static int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
338 union ib_gid *gid)
225c7b1f
RD
339{
340 struct ib_smp *in_mad = NULL;
341 struct ib_smp *out_mad = NULL;
342 int err = -ENOMEM;
343
344 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
345 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
346 if (!in_mad || !out_mad)
347 goto out;
348
349 init_query_mad(in_mad);
350 in_mad->attr_id = IB_SMP_ATTR_PORT_INFO;
351 in_mad->attr_mod = cpu_to_be32(port);
352
353 err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL, in_mad, out_mad);
354 if (err)
355 goto out;
356
357 memcpy(gid->raw, out_mad->data + 8, 8);
358
359 init_query_mad(in_mad);
360 in_mad->attr_id = IB_SMP_ATTR_GUID_INFO;
361 in_mad->attr_mod = cpu_to_be32(index / 8);
362
363 err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL, in_mad, out_mad);
364 if (err)
365 goto out;
366
367 memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
368
369out:
370 kfree(in_mad);
371 kfree(out_mad);
372 return err;
373}
374
fa417f7b
EC
375static int iboe_query_gid(struct ib_device *ibdev, u8 port, int index,
376 union ib_gid *gid)
377{
378 struct mlx4_ib_dev *dev = to_mdev(ibdev);
379
380 *gid = dev->iboe.gid_table[port - 1][index];
381
382 return 0;
383}
384
385static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
386 union ib_gid *gid)
387{
388 if (rdma_port_get_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND)
389 return __mlx4_ib_query_gid(ibdev, port, index, gid);
390 else
391 return iboe_query_gid(ibdev, port, index, gid);
392}
393
225c7b1f
RD
394static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
395 u16 *pkey)
396{
397 struct ib_smp *in_mad = NULL;
398 struct ib_smp *out_mad = NULL;
399 int err = -ENOMEM;
400
401 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
402 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
403 if (!in_mad || !out_mad)
404 goto out;
405
406 init_query_mad(in_mad);
407 in_mad->attr_id = IB_SMP_ATTR_PKEY_TABLE;
408 in_mad->attr_mod = cpu_to_be32(index / 32);
409
410 err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL, in_mad, out_mad);
411 if (err)
412 goto out;
413
414 *pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
415
416out:
417 kfree(in_mad);
418 kfree(out_mad);
419 return err;
420}
421
422static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
423 struct ib_device_modify *props)
424{
d0d68b86
JM
425 struct mlx4_cmd_mailbox *mailbox;
426
225c7b1f
RD
427 if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
428 return -EOPNOTSUPP;
429
d0d68b86
JM
430 if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
431 return 0;
432
433 spin_lock(&to_mdev(ibdev)->sm_lock);
434 memcpy(ibdev->node_desc, props->node_desc, 64);
435 spin_unlock(&to_mdev(ibdev)->sm_lock);
436
437 /*
438 * If possible, pass node desc to FW, so it can generate
439 * a 144 trap. If cmd fails, just ignore.
440 */
441 mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
442 if (IS_ERR(mailbox))
443 return 0;
444
445 memset(mailbox->buf, 0, 256);
446 memcpy(mailbox->buf, props->node_desc, 64);
447 mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
f9baff50 448 MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
d0d68b86
JM
449
450 mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
225c7b1f
RD
451
452 return 0;
453}
454
455static int mlx4_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
456 u32 cap_mask)
457{
458 struct mlx4_cmd_mailbox *mailbox;
459 int err;
fa417f7b 460 u8 is_eth = dev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
225c7b1f
RD
461
462 mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
463 if (IS_ERR(mailbox))
464 return PTR_ERR(mailbox);
465
466 memset(mailbox->buf, 0, 256);
5ae2a7a8
RD
467
468 if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
469 *(u8 *) mailbox->buf = !!reset_qkey_viols << 6;
470 ((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
471 } else {
472 ((u8 *) mailbox->buf)[3] = !!reset_qkey_viols;
473 ((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
474 }
225c7b1f 475
fa417f7b 476 err = mlx4_cmd(dev->dev, mailbox->dma, port, is_eth, MLX4_CMD_SET_PORT,
f9baff50 477 MLX4_CMD_TIME_CLASS_B, MLX4_CMD_NATIVE);
225c7b1f
RD
478
479 mlx4_free_cmd_mailbox(dev->dev, mailbox);
480 return err;
481}
482
483static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
484 struct ib_port_modify *props)
485{
486 struct ib_port_attr attr;
487 u32 cap_mask;
488 int err;
489
490 mutex_lock(&to_mdev(ibdev)->cap_mask_mutex);
491
492 err = mlx4_ib_query_port(ibdev, port, &attr);
493 if (err)
494 goto out;
495
496 cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
497 ~props->clr_port_cap_mask;
498
499 err = mlx4_SET_PORT(to_mdev(ibdev), port,
500 !!(mask & IB_PORT_RESET_QKEY_CNTR),
501 cap_mask);
502
503out:
504 mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
505 return err;
506}
507
508static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
509 struct ib_udata *udata)
510{
511 struct mlx4_ib_dev *dev = to_mdev(ibdev);
512 struct mlx4_ib_ucontext *context;
513 struct mlx4_ib_alloc_ucontext_resp resp;
514 int err;
515
3b4a8cd5
JM
516 if (!dev->ib_active)
517 return ERR_PTR(-EAGAIN);
518
225c7b1f
RD
519 resp.qp_tab_size = dev->dev->caps.num_qps;
520 resp.bf_reg_size = dev->dev->caps.bf_reg_size;
521 resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
522
523 context = kmalloc(sizeof *context, GFP_KERNEL);
524 if (!context)
525 return ERR_PTR(-ENOMEM);
526
527 err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
528 if (err) {
529 kfree(context);
530 return ERR_PTR(err);
531 }
532
533 INIT_LIST_HEAD(&context->db_page_list);
534 mutex_init(&context->db_page_mutex);
535
536 err = ib_copy_to_udata(udata, &resp, sizeof resp);
537 if (err) {
538 mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
539 kfree(context);
540 return ERR_PTR(-EFAULT);
541 }
542
543 return &context->ibucontext;
544}
545
546static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
547{
548 struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
549
550 mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
551 kfree(context);
552
553 return 0;
554}
555
556static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
557{
558 struct mlx4_ib_dev *dev = to_mdev(context->device);
559
560 if (vma->vm_end - vma->vm_start != PAGE_SIZE)
561 return -EINVAL;
562
563 if (vma->vm_pgoff == 0) {
564 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
565
566 if (io_remap_pfn_range(vma, vma->vm_start,
567 to_mucontext(context)->uar.pfn,
568 PAGE_SIZE, vma->vm_page_prot))
569 return -EAGAIN;
570 } else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) {
e1d60ec6 571 vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
225c7b1f
RD
572
573 if (io_remap_pfn_range(vma, vma->vm_start,
574 to_mucontext(context)->uar.pfn +
575 dev->dev->caps.num_uars,
576 PAGE_SIZE, vma->vm_page_prot))
577 return -EAGAIN;
578 } else
579 return -EINVAL;
580
581 return 0;
582}
583
584static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
585 struct ib_ucontext *context,
586 struct ib_udata *udata)
587{
588 struct mlx4_ib_pd *pd;
589 int err;
590
591 pd = kmalloc(sizeof *pd, GFP_KERNEL);
592 if (!pd)
593 return ERR_PTR(-ENOMEM);
594
595 err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
596 if (err) {
597 kfree(pd);
598 return ERR_PTR(err);
599 }
600
601 if (context)
602 if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
603 mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
604 kfree(pd);
605 return ERR_PTR(-EFAULT);
606 }
607
608 return &pd->ibpd;
609}
610
611static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
612{
613 mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
614 kfree(pd);
615
616 return 0;
617}
618
012a8ff5
SH
619static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
620 struct ib_ucontext *context,
621 struct ib_udata *udata)
622{
623 struct mlx4_ib_xrcd *xrcd;
624 int err;
625
626 if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
627 return ERR_PTR(-ENOSYS);
628
629 xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
630 if (!xrcd)
631 return ERR_PTR(-ENOMEM);
632
633 err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
634 if (err)
635 goto err1;
636
637 xrcd->pd = ib_alloc_pd(ibdev);
638 if (IS_ERR(xrcd->pd)) {
639 err = PTR_ERR(xrcd->pd);
640 goto err2;
641 }
642
643 xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, 1, 0);
644 if (IS_ERR(xrcd->cq)) {
645 err = PTR_ERR(xrcd->cq);
646 goto err3;
647 }
648
649 return &xrcd->ibxrcd;
650
651err3:
652 ib_dealloc_pd(xrcd->pd);
653err2:
654 mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
655err1:
656 kfree(xrcd);
657 return ERR_PTR(err);
658}
659
660static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
661{
662 ib_destroy_cq(to_mxrcd(xrcd)->cq);
663 ib_dealloc_pd(to_mxrcd(xrcd)->pd);
664 mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
665 kfree(xrcd);
666
667 return 0;
668}
669
fa417f7b
EC
670static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
671{
672 struct mlx4_ib_qp *mqp = to_mqp(ibqp);
673 struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
674 struct mlx4_ib_gid_entry *ge;
675
676 ge = kzalloc(sizeof *ge, GFP_KERNEL);
677 if (!ge)
678 return -ENOMEM;
679
680 ge->gid = *gid;
681 if (mlx4_ib_add_mc(mdev, mqp, gid)) {
682 ge->port = mqp->port;
683 ge->added = 1;
684 }
685
686 mutex_lock(&mqp->mutex);
687 list_add_tail(&ge->list, &mqp->gid_list);
688 mutex_unlock(&mqp->mutex);
689
690 return 0;
691}
692
693int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
694 union ib_gid *gid)
695{
696 u8 mac[6];
697 struct net_device *ndev;
698 int ret = 0;
699
700 if (!mqp->port)
701 return 0;
702
703 spin_lock(&mdev->iboe.lock);
704 ndev = mdev->iboe.netdevs[mqp->port - 1];
705 if (ndev)
706 dev_hold(ndev);
707 spin_unlock(&mdev->iboe.lock);
708
709 if (ndev) {
710 rdma_get_mcast_mac((struct in6_addr *)gid, mac);
711 rtnl_lock();
712 dev_mc_add(mdev->iboe.netdevs[mqp->port - 1], mac);
713 ret = 1;
714 rtnl_unlock();
715 dev_put(ndev);
716 }
717
718 return ret;
719}
720
0ff1fb65
HHZ
721struct mlx4_ib_steering {
722 struct list_head list;
723 u64 reg_id;
724 union ib_gid gid;
725};
726
225c7b1f
RD
727static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
728{
fa417f7b
EC
729 int err;
730 struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
731 struct mlx4_ib_qp *mqp = to_mqp(ibqp);
0ff1fb65
HHZ
732 u64 reg_id;
733 struct mlx4_ib_steering *ib_steering = NULL;
734
735 if (mdev->dev->caps.steering_mode ==
736 MLX4_STEERING_MODE_DEVICE_MANAGED) {
737 ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
738 if (!ib_steering)
739 return -ENOMEM;
740 }
fa417f7b 741
0ff1fb65
HHZ
742 err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
743 !!(mqp->flags &
744 MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
745 MLX4_PROT_IB_IPV6, &reg_id);
fa417f7b 746 if (err)
0ff1fb65 747 goto err_malloc;
fa417f7b
EC
748
749 err = add_gid_entry(ibqp, gid);
750 if (err)
751 goto err_add;
752
0ff1fb65
HHZ
753 if (ib_steering) {
754 memcpy(ib_steering->gid.raw, gid->raw, 16);
755 ib_steering->reg_id = reg_id;
756 mutex_lock(&mqp->mutex);
757 list_add(&ib_steering->list, &mqp->steering_rules);
758 mutex_unlock(&mqp->mutex);
759 }
fa417f7b
EC
760 return 0;
761
762err_add:
0ff1fb65
HHZ
763 mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
764 MLX4_PROT_IB_IPV6, reg_id);
765err_malloc:
766 kfree(ib_steering);
767
fa417f7b
EC
768 return err;
769}
770
771static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
772{
773 struct mlx4_ib_gid_entry *ge;
774 struct mlx4_ib_gid_entry *tmp;
775 struct mlx4_ib_gid_entry *ret = NULL;
776
777 list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
778 if (!memcmp(raw, ge->gid.raw, 16)) {
779 ret = ge;
780 break;
781 }
782 }
783
784 return ret;
225c7b1f
RD
785}
786
787static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
788{
fa417f7b
EC
789 int err;
790 struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
791 struct mlx4_ib_qp *mqp = to_mqp(ibqp);
792 u8 mac[6];
793 struct net_device *ndev;
794 struct mlx4_ib_gid_entry *ge;
0ff1fb65
HHZ
795 u64 reg_id = 0;
796
797 if (mdev->dev->caps.steering_mode ==
798 MLX4_STEERING_MODE_DEVICE_MANAGED) {
799 struct mlx4_ib_steering *ib_steering;
800
801 mutex_lock(&mqp->mutex);
802 list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
803 if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
804 list_del(&ib_steering->list);
805 break;
806 }
807 }
808 mutex_unlock(&mqp->mutex);
809 if (&ib_steering->list == &mqp->steering_rules) {
810 pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
811 return -EINVAL;
812 }
813 reg_id = ib_steering->reg_id;
814 kfree(ib_steering);
815 }
fa417f7b 816
0ff1fb65
HHZ
817 err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
818 MLX4_PROT_IB_IPV6, reg_id);
fa417f7b
EC
819 if (err)
820 return err;
821
822 mutex_lock(&mqp->mutex);
823 ge = find_gid_entry(mqp, gid->raw);
824 if (ge) {
825 spin_lock(&mdev->iboe.lock);
826 ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
827 if (ndev)
828 dev_hold(ndev);
829 spin_unlock(&mdev->iboe.lock);
830 rdma_get_mcast_mac((struct in6_addr *)gid, mac);
831 if (ndev) {
832 rtnl_lock();
833 dev_mc_del(mdev->iboe.netdevs[ge->port - 1], mac);
834 rtnl_unlock();
835 dev_put(ndev);
836 }
837 list_del(&ge->list);
838 kfree(ge);
839 } else
987c8f8f 840 pr_warn("could not find mgid entry\n");
fa417f7b
EC
841
842 mutex_unlock(&mqp->mutex);
843
844 return 0;
225c7b1f
RD
845}
846
847static int init_node_data(struct mlx4_ib_dev *dev)
848{
849 struct ib_smp *in_mad = NULL;
850 struct ib_smp *out_mad = NULL;
851 int err = -ENOMEM;
852
853 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
854 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
855 if (!in_mad || !out_mad)
856 goto out;
857
858 init_query_mad(in_mad);
859 in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
860
861 err = mlx4_MAD_IFC(dev, 1, 1, 1, NULL, NULL, in_mad, out_mad);
862 if (err)
863 goto out;
864
865 memcpy(dev->ib_dev.node_desc, out_mad->data, 64);
866
867 in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
868
869 err = mlx4_MAD_IFC(dev, 1, 1, 1, NULL, NULL, in_mad, out_mad);
870 if (err)
871 goto out;
872
873 memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
874
875out:
876 kfree(in_mad);
877 kfree(out_mad);
878 return err;
879}
880
f4e91eb4
TJ
881static ssize_t show_hca(struct device *device, struct device_attribute *attr,
882 char *buf)
cd9281d8 883{
f4e91eb4
TJ
884 struct mlx4_ib_dev *dev =
885 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
cd9281d8
JM
886 return sprintf(buf, "MT%d\n", dev->dev->pdev->device);
887}
888
f4e91eb4
TJ
889static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
890 char *buf)
cd9281d8 891{
f4e91eb4
TJ
892 struct mlx4_ib_dev *dev =
893 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
cd9281d8
JM
894 return sprintf(buf, "%d.%d.%d\n", (int) (dev->dev->caps.fw_ver >> 32),
895 (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
896 (int) dev->dev->caps.fw_ver & 0xffff);
897}
898
f4e91eb4
TJ
899static ssize_t show_rev(struct device *device, struct device_attribute *attr,
900 char *buf)
cd9281d8 901{
f4e91eb4
TJ
902 struct mlx4_ib_dev *dev =
903 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
cd9281d8
JM
904 return sprintf(buf, "%x\n", dev->dev->rev_id);
905}
906
f4e91eb4
TJ
907static ssize_t show_board(struct device *device, struct device_attribute *attr,
908 char *buf)
cd9281d8 909{
f4e91eb4
TJ
910 struct mlx4_ib_dev *dev =
911 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
912 return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
913 dev->dev->board_id);
cd9281d8
JM
914}
915
f4e91eb4
TJ
916static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL);
917static DEVICE_ATTR(fw_ver, S_IRUGO, show_fw_ver, NULL);
918static DEVICE_ATTR(hca_type, S_IRUGO, show_hca, NULL);
919static DEVICE_ATTR(board_id, S_IRUGO, show_board, NULL);
cd9281d8 920
f4e91eb4
TJ
921static struct device_attribute *mlx4_class_attributes[] = {
922 &dev_attr_hw_rev,
923 &dev_attr_fw_ver,
924 &dev_attr_hca_type,
925 &dev_attr_board_id
cd9281d8
JM
926};
927
4c3eb3ca 928static void mlx4_addrconf_ifid_eui48(u8 *eui, u16 vlan_id, struct net_device *dev)
fa417f7b
EC
929{
930 memcpy(eui, dev->dev_addr, 3);
931 memcpy(eui + 5, dev->dev_addr + 3, 3);
4c3eb3ca
EC
932 if (vlan_id < 0x1000) {
933 eui[3] = vlan_id >> 8;
934 eui[4] = vlan_id & 0xff;
935 } else {
936 eui[3] = 0xff;
937 eui[4] = 0xfe;
938 }
fa417f7b
EC
939 eui[0] ^= 2;
940}
941
942static void update_gids_task(struct work_struct *work)
943{
944 struct update_gid_work *gw = container_of(work, struct update_gid_work, work);
945 struct mlx4_cmd_mailbox *mailbox;
946 union ib_gid *gids;
947 int err;
948 struct mlx4_dev *dev = gw->dev->dev;
fa417f7b
EC
949
950 mailbox = mlx4_alloc_cmd_mailbox(dev);
951 if (IS_ERR(mailbox)) {
987c8f8f 952 pr_warn("update gid table failed %ld\n", PTR_ERR(mailbox));
fa417f7b
EC
953 return;
954 }
955
956 gids = mailbox->buf;
957 memcpy(gids, gw->gids, sizeof gw->gids);
958
959 err = mlx4_cmd(dev, mailbox->dma, MLX4_SET_PORT_GID_TABLE << 8 | gw->port,
f9baff50
JM
960 1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
961 MLX4_CMD_NATIVE);
fa417f7b 962 if (err)
987c8f8f 963 pr_warn("set port command failed\n");
fa417f7b
EC
964 else {
965 memcpy(gw->dev->iboe.gid_table[gw->port - 1], gw->gids, sizeof gw->gids);
00f5ce99 966 mlx4_ib_dispatch_event(gw->dev, gw->port, IB_EVENT_GID_CHANGE);
fa417f7b
EC
967 }
968
969 mlx4_free_cmd_mailbox(dev, mailbox);
970 kfree(gw);
971}
972
973static int update_ipv6_gids(struct mlx4_ib_dev *dev, int port, int clear)
974{
975 struct net_device *ndev = dev->iboe.netdevs[port - 1];
976 struct update_gid_work *work;
4c3eb3ca
EC
977 struct net_device *tmp;
978 int i;
979 u8 *hits;
980 int ret;
981 union ib_gid gid;
982 int free;
983 int found;
984 int need_update = 0;
985 u16 vid;
fa417f7b
EC
986
987 work = kzalloc(sizeof *work, GFP_ATOMIC);
988 if (!work)
989 return -ENOMEM;
990
4c3eb3ca
EC
991 hits = kzalloc(128, GFP_ATOMIC);
992 if (!hits) {
993 ret = -ENOMEM;
994 goto out;
995 }
996
22f4fbd9
ED
997 rcu_read_lock();
998 for_each_netdev_rcu(&init_net, tmp) {
4c3eb3ca
EC
999 if (ndev && (tmp == ndev || rdma_vlan_dev_real_dev(tmp) == ndev)) {
1000 gid.global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL);
1001 vid = rdma_vlan_dev_vlan_id(tmp);
1002 mlx4_addrconf_ifid_eui48(&gid.raw[8], vid, ndev);
1003 found = 0;
1004 free = -1;
1005 for (i = 0; i < 128; ++i) {
1006 if (free < 0 &&
1007 !memcmp(&dev->iboe.gid_table[port - 1][i], &zgid, sizeof zgid))
1008 free = i;
1009 if (!memcmp(&dev->iboe.gid_table[port - 1][i], &gid, sizeof gid)) {
1010 hits[i] = 1;
1011 found = 1;
1012 break;
1013 }
1014 }
1015
1016 if (!found) {
1017 if (tmp == ndev &&
1018 (memcmp(&dev->iboe.gid_table[port - 1][0],
1019 &gid, sizeof gid) ||
1020 !memcmp(&dev->iboe.gid_table[port - 1][0],
1021 &zgid, sizeof gid))) {
1022 dev->iboe.gid_table[port - 1][0] = gid;
1023 ++need_update;
1024 hits[0] = 1;
1025 } else if (free >= 0) {
1026 dev->iboe.gid_table[port - 1][free] = gid;
1027 hits[free] = 1;
1028 ++need_update;
1029 }
1030 }
1031 }
fa417f7b 1032 }
22f4fbd9 1033 rcu_read_unlock();
4c3eb3ca
EC
1034
1035 for (i = 0; i < 128; ++i)
1036 if (!hits[i]) {
1037 if (memcmp(&dev->iboe.gid_table[port - 1][i], &zgid, sizeof zgid))
1038 ++need_update;
1039 dev->iboe.gid_table[port - 1][i] = zgid;
1040 }
fa417f7b 1041
4c3eb3ca
EC
1042 if (need_update) {
1043 memcpy(work->gids, dev->iboe.gid_table[port - 1], sizeof work->gids);
1044 INIT_WORK(&work->work, update_gids_task);
1045 work->port = port;
1046 work->dev = dev;
1047 queue_work(wq, &work->work);
1048 } else
1049 kfree(work);
fa417f7b 1050
4c3eb3ca 1051 kfree(hits);
fa417f7b 1052 return 0;
4c3eb3ca
EC
1053
1054out:
1055 kfree(work);
1056 return ret;
fa417f7b
EC
1057}
1058
1059static void handle_en_event(struct mlx4_ib_dev *dev, int port, unsigned long event)
1060{
1061 switch (event) {
1062 case NETDEV_UP:
4c3eb3ca 1063 case NETDEV_CHANGEADDR:
fa417f7b
EC
1064 update_ipv6_gids(dev, port, 0);
1065 break;
1066
1067 case NETDEV_DOWN:
1068 update_ipv6_gids(dev, port, 1);
1069 dev->iboe.netdevs[port - 1] = NULL;
1070 }
1071}
1072
1073static void netdev_added(struct mlx4_ib_dev *dev, int port)
1074{
1075 update_ipv6_gids(dev, port, 0);
1076}
1077
1078static void netdev_removed(struct mlx4_ib_dev *dev, int port)
1079{
1080 update_ipv6_gids(dev, port, 1);
1081}
1082
1083static int mlx4_ib_netdev_event(struct notifier_block *this, unsigned long event,
1084 void *ptr)
1085{
1086 struct net_device *dev = ptr;
1087 struct mlx4_ib_dev *ibdev;
1088 struct net_device *oldnd;
1089 struct mlx4_ib_iboe *iboe;
1090 int port;
1091
1092 if (!net_eq(dev_net(dev), &init_net))
1093 return NOTIFY_DONE;
1094
1095 ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
1096 iboe = &ibdev->iboe;
1097
1098 spin_lock(&iboe->lock);
1099 mlx4_foreach_ib_transport_port(port, ibdev->dev) {
1100 oldnd = iboe->netdevs[port - 1];
1101 iboe->netdevs[port - 1] =
0345584e 1102 mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
fa417f7b
EC
1103 if (oldnd != iboe->netdevs[port - 1]) {
1104 if (iboe->netdevs[port - 1])
1105 netdev_added(ibdev, port);
1106 else
1107 netdev_removed(ibdev, port);
1108 }
1109 }
1110
4c3eb3ca
EC
1111 if (dev == iboe->netdevs[0] ||
1112 (iboe->netdevs[0] && rdma_vlan_dev_real_dev(dev) == iboe->netdevs[0]))
fa417f7b 1113 handle_en_event(ibdev, 1, event);
4c3eb3ca
EC
1114 else if (dev == iboe->netdevs[1]
1115 || (iboe->netdevs[1] && rdma_vlan_dev_real_dev(dev) == iboe->netdevs[1]))
fa417f7b
EC
1116 handle_en_event(ibdev, 2, event);
1117
1118 spin_unlock(&iboe->lock);
1119
1120 return NOTIFY_DONE;
1121}
1122
e605b743
SP
1123static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
1124{
1125 char name[32];
1126 int eq_per_port = 0;
1127 int added_eqs = 0;
1128 int total_eqs = 0;
1129 int i, j, eq;
1130
3aac6ff1
SP
1131 /* Legacy mode or comp_pool is not large enough */
1132 if (dev->caps.comp_pool == 0 ||
1133 dev->caps.num_ports > dev->caps.comp_pool)
e605b743
SP
1134 return;
1135
1136 eq_per_port = rounddown_pow_of_two(dev->caps.comp_pool/
1137 dev->caps.num_ports);
1138
1139 /* Init eq table */
1140 added_eqs = 0;
1141 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
1142 added_eqs += eq_per_port;
1143
1144 total_eqs = dev->caps.num_comp_vectors + added_eqs;
1145
1146 ibdev->eq_table = kzalloc(total_eqs * sizeof(int), GFP_KERNEL);
1147 if (!ibdev->eq_table)
1148 return;
1149
1150 ibdev->eq_added = added_eqs;
1151
1152 eq = 0;
1153 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB) {
1154 for (j = 0; j < eq_per_port; j++) {
1155 sprintf(name, "mlx4-ib-%d-%d@%s",
1156 i, j, dev->pdev->bus->name);
1157 /* Set IRQ for specific name (per ring) */
d9236c3f
AV
1158 if (mlx4_assign_eq(dev, name, NULL,
1159 &ibdev->eq_table[eq])) {
e605b743
SP
1160 /* Use legacy (same as mlx4_en driver) */
1161 pr_warn("Can't allocate EQ %d; reverting to legacy\n", eq);
1162 ibdev->eq_table[eq] =
1163 (eq % dev->caps.num_comp_vectors);
1164 }
1165 eq++;
1166 }
1167 }
1168
1169 /* Fill the reset of the vector with legacy EQ */
1170 for (i = 0, eq = added_eqs; i < dev->caps.num_comp_vectors; i++)
1171 ibdev->eq_table[eq++] = i;
1172
1173 /* Advertise the new number of EQs to clients */
1174 ibdev->ib_dev.num_comp_vectors = total_eqs;
1175}
1176
1177static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
1178{
1179 int i;
3aac6ff1
SP
1180
1181 /* no additional eqs were added */
1182 if (!ibdev->eq_table)
1183 return;
e605b743
SP
1184
1185 /* Reset the advertised EQ number */
1186 ibdev->ib_dev.num_comp_vectors = dev->caps.num_comp_vectors;
1187
1188 /* Free only the added eqs */
1189 for (i = 0; i < ibdev->eq_added; i++) {
1190 /* Don't free legacy eqs if used */
1191 if (ibdev->eq_table[i] <= dev->caps.num_comp_vectors)
1192 continue;
1193 mlx4_release_eq(dev, ibdev->eq_table[i]);
1194 }
1195
e605b743 1196 kfree(ibdev->eq_table);
e605b743
SP
1197}
1198
225c7b1f
RD
1199static void *mlx4_ib_add(struct mlx4_dev *dev)
1200{
1201 struct mlx4_ib_dev *ibdev;
22e7ef9c 1202 int num_ports = 0;
035b1032 1203 int i, j;
fa417f7b
EC
1204 int err;
1205 struct mlx4_ib_iboe *iboe;
225c7b1f 1206
987c8f8f 1207 pr_info_once("%s", mlx4_ib_version);
68f3948d 1208
8e59d254 1209 if (mlx4_is_mfunc(dev)) {
987c8f8f 1210 pr_warn("IB not yet supported in SRIOV\n");
8e59d254
JM
1211 return NULL;
1212 }
1213
fa417f7b 1214 mlx4_foreach_ib_transport_port(i, dev)
22e7ef9c
RD
1215 num_ports++;
1216
1217 /* No point in registering a device with no ports... */
1218 if (num_ports == 0)
1219 return NULL;
1220
225c7b1f
RD
1221 ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
1222 if (!ibdev) {
1223 dev_err(&dev->pdev->dev, "Device struct alloc failed\n");
1224 return NULL;
1225 }
1226
fa417f7b
EC
1227 iboe = &ibdev->iboe;
1228
225c7b1f
RD
1229 if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
1230 goto err_dealloc;
1231
1232 if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
1233 goto err_pd;
1234
4979d18f
RD
1235 ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
1236 PAGE_SIZE);
225c7b1f
RD
1237 if (!ibdev->uar_map)
1238 goto err_uar;
26c6bc7b 1239 MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
225c7b1f 1240
225c7b1f
RD
1241 ibdev->dev = dev;
1242
1243 strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
1244 ibdev->ib_dev.owner = THIS_MODULE;
1245 ibdev->ib_dev.node_type = RDMA_NODE_IB_CA;
95d04f07 1246 ibdev->ib_dev.local_dma_lkey = dev->caps.reserved_lkey;
22e7ef9c 1247 ibdev->num_ports = num_ports;
7ff93f8b 1248 ibdev->ib_dev.phys_port_cnt = ibdev->num_ports;
b8dd786f 1249 ibdev->ib_dev.num_comp_vectors = dev->caps.num_comp_vectors;
225c7b1f
RD
1250 ibdev->ib_dev.dma_device = &dev->pdev->dev;
1251
1252 ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
1253 ibdev->ib_dev.uverbs_cmd_mask =
1254 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT) |
1255 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE) |
1256 (1ull << IB_USER_VERBS_CMD_QUERY_PORT) |
1257 (1ull << IB_USER_VERBS_CMD_ALLOC_PD) |
1258 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD) |
1259 (1ull << IB_USER_VERBS_CMD_REG_MR) |
1260 (1ull << IB_USER_VERBS_CMD_DEREG_MR) |
1261 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
1262 (1ull << IB_USER_VERBS_CMD_CREATE_CQ) |
bbf8eed1 1263 (1ull << IB_USER_VERBS_CMD_RESIZE_CQ) |
225c7b1f
RD
1264 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ) |
1265 (1ull << IB_USER_VERBS_CMD_CREATE_QP) |
1266 (1ull << IB_USER_VERBS_CMD_MODIFY_QP) |
6a775e2b 1267 (1ull << IB_USER_VERBS_CMD_QUERY_QP) |
225c7b1f
RD
1268 (1ull << IB_USER_VERBS_CMD_DESTROY_QP) |
1269 (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST) |
1270 (1ull << IB_USER_VERBS_CMD_DETACH_MCAST) |
1271 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ) |
1272 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ) |
65541cb7 1273 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ) |
18abd5ea 1274 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ) |
42849b26
SH
1275 (1ull << IB_USER_VERBS_CMD_CREATE_XSRQ) |
1276 (1ull << IB_USER_VERBS_CMD_OPEN_QP);
225c7b1f
RD
1277
1278 ibdev->ib_dev.query_device = mlx4_ib_query_device;
1279 ibdev->ib_dev.query_port = mlx4_ib_query_port;
fa417f7b 1280 ibdev->ib_dev.get_link_layer = mlx4_ib_port_link_layer;
225c7b1f
RD
1281 ibdev->ib_dev.query_gid = mlx4_ib_query_gid;
1282 ibdev->ib_dev.query_pkey = mlx4_ib_query_pkey;
1283 ibdev->ib_dev.modify_device = mlx4_ib_modify_device;
1284 ibdev->ib_dev.modify_port = mlx4_ib_modify_port;
1285 ibdev->ib_dev.alloc_ucontext = mlx4_ib_alloc_ucontext;
1286 ibdev->ib_dev.dealloc_ucontext = mlx4_ib_dealloc_ucontext;
1287 ibdev->ib_dev.mmap = mlx4_ib_mmap;
1288 ibdev->ib_dev.alloc_pd = mlx4_ib_alloc_pd;
1289 ibdev->ib_dev.dealloc_pd = mlx4_ib_dealloc_pd;
1290 ibdev->ib_dev.create_ah = mlx4_ib_create_ah;
1291 ibdev->ib_dev.query_ah = mlx4_ib_query_ah;
1292 ibdev->ib_dev.destroy_ah = mlx4_ib_destroy_ah;
1293 ibdev->ib_dev.create_srq = mlx4_ib_create_srq;
1294 ibdev->ib_dev.modify_srq = mlx4_ib_modify_srq;
65541cb7 1295 ibdev->ib_dev.query_srq = mlx4_ib_query_srq;
225c7b1f
RD
1296 ibdev->ib_dev.destroy_srq = mlx4_ib_destroy_srq;
1297 ibdev->ib_dev.post_srq_recv = mlx4_ib_post_srq_recv;
1298 ibdev->ib_dev.create_qp = mlx4_ib_create_qp;
1299 ibdev->ib_dev.modify_qp = mlx4_ib_modify_qp;
6a775e2b 1300 ibdev->ib_dev.query_qp = mlx4_ib_query_qp;
225c7b1f
RD
1301 ibdev->ib_dev.destroy_qp = mlx4_ib_destroy_qp;
1302 ibdev->ib_dev.post_send = mlx4_ib_post_send;
1303 ibdev->ib_dev.post_recv = mlx4_ib_post_recv;
1304 ibdev->ib_dev.create_cq = mlx4_ib_create_cq;
3fdcb97f 1305 ibdev->ib_dev.modify_cq = mlx4_ib_modify_cq;
bbf8eed1 1306 ibdev->ib_dev.resize_cq = mlx4_ib_resize_cq;
225c7b1f
RD
1307 ibdev->ib_dev.destroy_cq = mlx4_ib_destroy_cq;
1308 ibdev->ib_dev.poll_cq = mlx4_ib_poll_cq;
1309 ibdev->ib_dev.req_notify_cq = mlx4_ib_arm_cq;
1310 ibdev->ib_dev.get_dma_mr = mlx4_ib_get_dma_mr;
1311 ibdev->ib_dev.reg_user_mr = mlx4_ib_reg_user_mr;
1312 ibdev->ib_dev.dereg_mr = mlx4_ib_dereg_mr;
95d04f07
RD
1313 ibdev->ib_dev.alloc_fast_reg_mr = mlx4_ib_alloc_fast_reg_mr;
1314 ibdev->ib_dev.alloc_fast_reg_page_list = mlx4_ib_alloc_fast_reg_page_list;
1315 ibdev->ib_dev.free_fast_reg_page_list = mlx4_ib_free_fast_reg_page_list;
225c7b1f
RD
1316 ibdev->ib_dev.attach_mcast = mlx4_ib_mcg_attach;
1317 ibdev->ib_dev.detach_mcast = mlx4_ib_mcg_detach;
1318 ibdev->ib_dev.process_mad = mlx4_ib_process_mad;
1319
8ad11fb6
JM
1320 ibdev->ib_dev.alloc_fmr = mlx4_ib_fmr_alloc;
1321 ibdev->ib_dev.map_phys_fmr = mlx4_ib_map_phys_fmr;
1322 ibdev->ib_dev.unmap_fmr = mlx4_ib_unmap_fmr;
1323 ibdev->ib_dev.dealloc_fmr = mlx4_ib_fmr_dealloc;
1324
012a8ff5
SH
1325 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
1326 ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
1327 ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
1328 ibdev->ib_dev.uverbs_cmd_mask |=
1329 (1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
1330 (1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
1331 }
1332
e605b743
SP
1333 mlx4_ib_alloc_eqs(dev, ibdev);
1334
fa417f7b
EC
1335 spin_lock_init(&iboe->lock);
1336
225c7b1f
RD
1337 if (init_node_data(ibdev))
1338 goto err_map;
1339
cfcde11c
OG
1340 for (i = 0; i < ibdev->num_ports; ++i) {
1341 if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
1342 IB_LINK_LAYER_ETHERNET) {
1343 err = mlx4_counter_alloc(ibdev->dev, &ibdev->counters[i]);
1344 if (err)
1345 ibdev->counters[i] = -1;
1346 } else
1347 ibdev->counters[i] = -1;
1348 }
1349
225c7b1f
RD
1350 spin_lock_init(&ibdev->sm_lock);
1351 mutex_init(&ibdev->cap_mask_mutex);
1352
9a6edb60 1353 if (ib_register_device(&ibdev->ib_dev, NULL))
cfcde11c 1354 goto err_counter;
225c7b1f
RD
1355
1356 if (mlx4_ib_mad_init(ibdev))
1357 goto err_reg;
1358
fa417f7b
EC
1359 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE && !iboe->nb.notifier_call) {
1360 iboe->nb.notifier_call = mlx4_ib_netdev_event;
1361 err = register_netdevice_notifier(&iboe->nb);
1362 if (err)
1363 goto err_reg;
1364 }
1365
035b1032 1366 for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
f4e91eb4 1367 if (device_create_file(&ibdev->ib_dev.dev,
035b1032 1368 mlx4_class_attributes[j]))
fa417f7b 1369 goto err_notif;
cd9281d8
JM
1370 }
1371
3b4a8cd5
JM
1372 ibdev->ib_active = true;
1373
225c7b1f
RD
1374 return ibdev;
1375
fa417f7b
EC
1376err_notif:
1377 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
987c8f8f 1378 pr_warn("failure unregistering notifier\n");
fa417f7b
EC
1379 flush_workqueue(wq);
1380
225c7b1f
RD
1381err_reg:
1382 ib_unregister_device(&ibdev->ib_dev);
1383
cfcde11c
OG
1384err_counter:
1385 for (; i; --i)
4af3ce0d
RD
1386 if (ibdev->counters[i - 1] != -1)
1387 mlx4_counter_free(ibdev->dev, ibdev->counters[i - 1]);
cfcde11c 1388
225c7b1f
RD
1389err_map:
1390 iounmap(ibdev->uar_map);
1391
1392err_uar:
1393 mlx4_uar_free(dev, &ibdev->priv_uar);
1394
1395err_pd:
1396 mlx4_pd_free(dev, ibdev->priv_pdn);
1397
1398err_dealloc:
1399 ib_dealloc_device(&ibdev->ib_dev);
1400
1401 return NULL;
1402}
1403
1404static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
1405{
1406 struct mlx4_ib_dev *ibdev = ibdev_ptr;
1407 int p;
1408
a6a47771
YP
1409 mlx4_ib_mad_cleanup(ibdev);
1410 ib_unregister_device(&ibdev->ib_dev);
fa417f7b
EC
1411 if (ibdev->iboe.nb.notifier_call) {
1412 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
987c8f8f 1413 pr_warn("failure unregistering notifier\n");
fa417f7b
EC
1414 ibdev->iboe.nb.notifier_call = NULL;
1415 }
1416 iounmap(ibdev->uar_map);
cfcde11c 1417 for (p = 0; p < ibdev->num_ports; ++p)
4af3ce0d
RD
1418 if (ibdev->counters[p] != -1)
1419 mlx4_counter_free(ibdev->dev, ibdev->counters[p]);
fa417f7b 1420 mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
225c7b1f
RD
1421 mlx4_CLOSE_PORT(dev, p);
1422
e605b743
SP
1423 mlx4_ib_free_eqs(dev, ibdev);
1424
225c7b1f
RD
1425 mlx4_uar_free(dev, &ibdev->priv_uar);
1426 mlx4_pd_free(dev, ibdev->priv_pdn);
1427 ib_dealloc_device(&ibdev->ib_dev);
1428}
1429
1430static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
00f5ce99 1431 enum mlx4_dev_event event, unsigned long param)
225c7b1f
RD
1432{
1433 struct ib_event ibev;
7ff93f8b 1434 struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
00f5ce99
JM
1435 struct mlx4_eqe *eqe = NULL;
1436 struct ib_event_work *ew;
1437 int port = 0;
1438
1439 if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
1440 eqe = (struct mlx4_eqe *)param;
1441 else
1442 port = (u8)param;
7ff93f8b
YP
1443
1444 if (port > ibdev->num_ports)
1445 return;
225c7b1f
RD
1446
1447 switch (event) {
37608eea
RD
1448 case MLX4_DEV_EVENT_PORT_UP:
1449 ibev.event = IB_EVENT_PORT_ACTIVE;
225c7b1f
RD
1450 break;
1451
37608eea
RD
1452 case MLX4_DEV_EVENT_PORT_DOWN:
1453 ibev.event = IB_EVENT_PORT_ERR;
1454 break;
1455
1456 case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
3b4a8cd5 1457 ibdev->ib_active = false;
225c7b1f
RD
1458 ibev.event = IB_EVENT_DEVICE_FATAL;
1459 break;
1460
00f5ce99
JM
1461 case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
1462 ew = kmalloc(sizeof *ew, GFP_ATOMIC);
1463 if (!ew) {
1464 pr_err("failed to allocate memory for events work\n");
1465 break;
1466 }
1467
1468 INIT_WORK(&ew->work, handle_port_mgmt_change_event);
1469 memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
1470 ew->ib_dev = ibdev;
1471 handle_port_mgmt_change_event(&ew->work);
1472 return;
1473
225c7b1f
RD
1474 default:
1475 return;
1476 }
1477
1478 ibev.device = ibdev_ptr;
1479 ibev.element.port_num = port;
1480
1481 ib_dispatch_event(&ibev);
1482}
1483
1484static struct mlx4_interface mlx4_ib_interface = {
fa417f7b
EC
1485 .add = mlx4_ib_add,
1486 .remove = mlx4_ib_remove,
1487 .event = mlx4_ib_event,
0345584e 1488 .protocol = MLX4_PROT_IB_IPV6
225c7b1f
RD
1489};
1490
1491static int __init mlx4_ib_init(void)
1492{
fa417f7b
EC
1493 int err;
1494
1495 wq = create_singlethread_workqueue("mlx4_ib");
1496 if (!wq)
1497 return -ENOMEM;
1498
1499 err = mlx4_register_interface(&mlx4_ib_interface);
1500 if (err) {
1501 destroy_workqueue(wq);
1502 return err;
1503 }
1504
1505 return 0;
225c7b1f
RD
1506}
1507
1508static void __exit mlx4_ib_cleanup(void)
1509{
1510 mlx4_unregister_interface(&mlx4_ib_interface);
fa417f7b 1511 destroy_workqueue(wq);
225c7b1f
RD
1512}
1513
1514module_init(mlx4_ib_init);
1515module_exit(mlx4_ib_cleanup);