projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
IB/mlx4: Miscellaneous adjustments for SR-IOV IB support
[linux-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
a0c64a17
JM		 62int mlx4_ib_sm_guid_assign = 1;
63module_param_named(sm_guid_assign, mlx4_ib_sm_guid_assign, int, 0444);
64MODULE_PARM_DESC(sm_guid_assign, "Enable SM alias_GUID assignment if sm_guid_assign > 0 (Default: 1)");
65
68f3948d 66static const char mlx4_ib_version[] =
225c7b1f
RD		 67 DRV_NAME ": Mellanox ConnectX InfiniBand driver v"
68 DRV_VERSION " (" DRV_RELDATE ")\n";
69
fa417f7b
EC		 70struct update_gid_work {
71 struct work_struct work;
72 union ib_gid gids[128];
73 struct mlx4_ib_dev *dev;
74 int port;
75};
76
77static struct workqueue_struct *wq;
78
225c7b1f
RD		 79static void init_query_mad(struct ib_smp *mad)
80{
81 mad->base_version = 1;
82 mad->mgmt_class = IB_MGMT_CLASS_SUBN_LID_ROUTED;
83 mad->class_version = 1;
84 mad->method = IB_MGMT_METHOD_GET;
85}
86
4c3eb3ca
EC		 87static union ib_gid zgid;
88
225c7b1f
RD		 89static int mlx4_ib_query_device(struct ib_device *ibdev,
90 struct ib_device_attr *props)
91{
92 struct mlx4_ib_dev *dev = to_mdev(ibdev);
93 struct ib_smp *in_mad = NULL;
94 struct ib_smp *out_mad = NULL;
95 int err = -ENOMEM;
96
97 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
98 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
99 if (!in_mad || !out_mad)
100 goto out;
101
102 init_query_mad(in_mad);
103 in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
104
0a9a0188
JM		 105 err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
106 1, NULL, NULL, in_mad, out_mad);
225c7b1f
RD		 107 if (err)
108 goto out;
109
110 memset(props, 0, sizeof *props);
111
112 props->fw_ver = dev->dev->caps.fw_ver;
113 props->device_cap_flags = IB_DEVICE_CHANGE_PHY_PORT |
114 IB_DEVICE_PORT_ACTIVE_EVENT |
115 IB_DEVICE_SYS_IMAGE_GUID |
521e575b
RL		 116 IB_DEVICE_RC_RNR_NAK_GEN |
117 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
225c7b1f
RD		 118 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
119 props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
120 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
121 props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
122 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM)
123 props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
124 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
125 props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
8ff095ec
EC		 126 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
127 props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
417608c2 128 if (dev->dev->caps.max_gso_sz && dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH)
b832be1e 129 props->device_cap_flags |= IB_DEVICE_UD_TSO;
95d04f07
RD		 130 if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
131 props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
132 if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
133 (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
134 (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
135 props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
0a1405da
SH		 136 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
137 props->device_cap_flags |= IB_DEVICE_XRC;
225c7b1f
RD		 138
139 props->vendor_id = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
140 0xffffff;
992e8e6e 141 props->vendor_part_id = dev->dev->pdev->device;
225c7b1f
RD		 142 props->hw_ver = be32_to_cpup((__be32 *) (out_mad->data + 32));
143 memcpy(&props->sys_image_guid, out_mad->data + 4, 8);
144
145 props->max_mr_size = ~0ull;
146 props->page_size_cap = dev->dev->caps.page_size_cap;
147 props->max_qp = dev->dev->caps.num_qps - dev->dev->caps.reserved_qps;
fc2d0044 148 props->max_qp_wr = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
225c7b1f
RD		 149 props->max_sge = min(dev->dev->caps.max_sq_sg,
150 dev->dev->caps.max_rq_sg);
151 props->max_cq = dev->dev->caps.num_cqs - dev->dev->caps.reserved_cqs;
152 props->max_cqe = dev->dev->caps.max_cqes;
153 props->max_mr = dev->dev->caps.num_mpts - dev->dev->caps.reserved_mrws;
154 props->max_pd = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
155 props->max_qp_rd_atom = dev->dev->caps.max_qp_dest_rdma;
156 props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
157 props->max_res_rd_atom = props->max_qp_rd_atom * props->max_qp;
158 props->max_srq = dev->dev->caps.num_srqs - dev->dev->caps.reserved_srqs;
c8681f14 159 props->max_srq_wr = dev->dev->caps.max_srq_wqes - 1;
225c7b1f 160 props->max_srq_sge = dev->dev->caps.max_srq_sge;
5a0fd094 161 props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
225c7b1f
RD		 162 props->local_ca_ack_delay = dev->dev->caps.local_ca_ack_delay;
163 props->atomic_cap = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
164 IB_ATOMIC_HCA : IB_ATOMIC_NONE;
47e956b2 165 props->masked_atomic_cap = props->atomic_cap;
5ae2a7a8 166 props->max_pkeys = dev->dev->caps.pkey_table_len[1];
225c7b1f
RD		 167 props->max_mcast_grp = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
168 props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
169 props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
170 props->max_mcast_grp;
a5bbe892 171 props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
225c7b1f
RD		 172
173out:
174 kfree(in_mad);
175 kfree(out_mad);
176
177 return err;
178}
179
fa417f7b
EC		 180static enum rdma_link_layer
181mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
225c7b1f 182{
fa417f7b 183 struct mlx4_dev *dev = to_mdev(device)->dev;
225c7b1f 184
65dab25d 185 return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
fa417f7b
EC		 186 IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
187}
225c7b1f 188
fa417f7b 189static int ib_link_query_port(struct ib_device *ibdev, u8 port,
0a9a0188 190 struct ib_port_attr *props, int netw_view)
fa417f7b 191{
a9c766bb
OG		 192 struct ib_smp *in_mad = NULL;
193 struct ib_smp *out_mad = NULL;
a5e12dff 194 int ext_active_speed;
0a9a0188 195 int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
a9c766bb
OG		 196 int err = -ENOMEM;
197
198 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
199 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
200 if (!in_mad || !out_mad)
201 goto out;
202
203 init_query_mad(in_mad);
204 in_mad->attr_id = IB_SMP_ATTR_PORT_INFO;
205 in_mad->attr_mod = cpu_to_be32(port);
206
0a9a0188
JM		 207 if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
208 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
209
210 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
a9c766bb
OG		 211 in_mad, out_mad);
212 if (err)
213 goto out;
214
a5e12dff 215
225c7b1f
RD		 216 props->lid = be16_to_cpup((__be16 *) (out_mad->data + 16));
217 props->lmc = out_mad->data[34] & 0x7;
218 props->sm_lid = be16_to_cpup((__be16 *) (out_mad->data + 18));
219 props->sm_sl = out_mad->data[36] & 0xf;
220 props->state = out_mad->data[32] & 0xf;
221 props->phys_state = out_mad->data[33] >> 4;
222 props->port_cap_flags = be32_to_cpup((__be32 *) (out_mad->data + 20));
0a9a0188
JM		 223 if (netw_view)
224 props->gid_tbl_len = out_mad->data[50];
225 else
226 props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
149983af 227 props->max_msg_sz = to_mdev(ibdev)->dev->caps.max_msg_sz;
5ae2a7a8 228 props->pkey_tbl_len = to_mdev(ibdev)->dev->caps.pkey_table_len[port];
225c7b1f
RD		 229 props->bad_pkey_cntr = be16_to_cpup((__be16 *) (out_mad->data + 46));
230 props->qkey_viol_cntr = be16_to_cpup((__be16 *) (out_mad->data + 48));
231 props->active_width = out_mad->data[31] & 0xf;
232 props->active_speed = out_mad->data[35] >> 4;
233 props->max_mtu = out_mad->data[41] & 0xf;
234 props->active_mtu = out_mad->data[36] >> 4;
235 props->subnet_timeout = out_mad->data[51] & 0x1f;
236 props->max_vl_num = out_mad->data[37] >> 4;
237 props->init_type_reply = out_mad->data[41] >> 4;
238
a5e12dff
MA		 239 /* Check if extended speeds (EDR/FDR/...) are supported */
240 if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
241 ext_active_speed = out_mad->data[62] >> 4;
242
243 switch (ext_active_speed) {
244 case 1:
2e96691c 245 props->active_speed = IB_SPEED_FDR;
a5e12dff
MA		 246 break;
247 case 2:
2e96691c 248 props->active_speed = IB_SPEED_EDR;
a5e12dff
MA		 249 break;
250 }
251 }
252
253 /* If reported active speed is QDR, check if is FDR-10 */
2e96691c 254 if (props->active_speed == IB_SPEED_QDR) {
8154c07f
OG		 255 init_query_mad(in_mad);
256 in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
257 in_mad->attr_mod = cpu_to_be32(port);
258
0a9a0188 259 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
8154c07f
OG		 260 NULL, NULL, in_mad, out_mad);
261 if (err)
bf6b47de 262 goto out;
8154c07f
OG		 263
264 /* Checking LinkSpeedActive for FDR-10 */
265 if (out_mad->data[15] & 0x1)
266 props->active_speed = IB_SPEED_FDR10;
a5e12dff 267 }
d2ef4068
OG		 268
269 /* Avoid wrong speed value returned by FW if the IB link is down. */
270 if (props->state == IB_PORT_DOWN)
271 props->active_speed = IB_SPEED_SDR;
272
a9c766bb
OG		 273out:
274 kfree(in_mad);
275 kfree(out_mad);
276 return err;
fa417f7b
EC		 277}
278
279static u8 state_to_phys_state(enum ib_port_state state)
280{
281 return state == IB_PORT_ACTIVE ? 5 : 3;
282}
283
284static int eth_link_query_port(struct ib_device *ibdev, u8 port,
0a9a0188 285 struct ib_port_attr *props, int netw_view)
fa417f7b 286{
a9c766bb
OG		 287
288 struct mlx4_ib_dev *mdev = to_mdev(ibdev);
289 struct mlx4_ib_iboe *iboe = &mdev->iboe;
fa417f7b
EC		 290 struct net_device *ndev;
291 enum ib_mtu tmp;
a9c766bb
OG		 292 struct mlx4_cmd_mailbox *mailbox;
293 int err = 0;
294
295 mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
296 if (IS_ERR(mailbox))
297 return PTR_ERR(mailbox);
fa417f7b 298
a9c766bb
OG		 299 err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
300 MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
301 MLX4_CMD_WRAPPED);
302 if (err)
303 goto out;
304
305 props->active_width = (((u8 *)mailbox->buf)[5] == 0x40) ?
306 IB_WIDTH_4X : IB_WIDTH_1X;
2e96691c 307 props->active_speed = IB_SPEED_QDR;
fa417f7b 308 props->port_cap_flags = IB_PORT_CM_SUP;
a9c766bb
OG		 309 props->gid_tbl_len = mdev->dev->caps.gid_table_len[port];
310 props->max_msg_sz = mdev->dev->caps.max_msg_sz;
fa417f7b 311 props->pkey_tbl_len = 1;
bcacb897 312 props->max_mtu = IB_MTU_4096;
a9c766bb 313 props->max_vl_num = 2;
fa417f7b
EC		 314 props->state = IB_PORT_DOWN;
315 props->phys_state = state_to_phys_state(props->state);
316 props->active_mtu = IB_MTU_256;
317 spin_lock(&iboe->lock);
318 ndev = iboe->netdevs[port - 1];
319 if (!ndev)
a9c766bb 320 goto out_unlock;
fa417f7b
EC		 321
322 tmp = iboe_get_mtu(ndev->mtu);
323 props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
324
21d60609 325 props->state = (netif_running(ndev) && netif_carrier_ok(ndev)) ?
fa417f7b
EC		 326 IB_PORT_ACTIVE : IB_PORT_DOWN;
327 props->phys_state = state_to_phys_state(props->state);
a9c766bb 328out_unlock:
fa417f7b 329 spin_unlock(&iboe->lock);
a9c766bb
OG		 330out:
331 mlx4_free_cmd_mailbox(mdev->dev, mailbox);
332 return err;
fa417f7b
EC		 333}
334
0a9a0188
JM		 335int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
336 struct ib_port_attr *props, int netw_view)
fa417f7b 337{
a9c766bb 338 int err;
fa417f7b
EC		 339
340 memset(props, 0, sizeof *props);
341
fa417f7b 342 err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
0a9a0188
JM		 343 ib_link_query_port(ibdev, port, props, netw_view) :
344 eth_link_query_port(ibdev, port, props, netw_view);
225c7b1f
RD		 345
346 return err;
347}
348
0a9a0188
JM		 349static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
350 struct ib_port_attr *props)
351{
352 /* returns host view */
353 return __mlx4_ib_query_port(ibdev, port, props, 0);
354}
355
a0c64a17
JM		 356int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
357 union ib_gid *gid, int netw_view)
225c7b1f
RD		 358{
359 struct ib_smp *in_mad = NULL;
360 struct ib_smp *out_mad = NULL;
361 int err = -ENOMEM;
a0c64a17
JM		 362 struct mlx4_ib_dev *dev = to_mdev(ibdev);
363 int clear = 0;
364 int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
225c7b1f
RD		 365
366 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
367 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
368 if (!in_mad || !out_mad)
369 goto out;
370
371 init_query_mad(in_mad);
372 in_mad->attr_id = IB_SMP_ATTR_PORT_INFO;
373 in_mad->attr_mod = cpu_to_be32(port);
374
a0c64a17
JM		 375 if (mlx4_is_mfunc(dev->dev) && netw_view)
376 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
377
378 err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
225c7b1f
RD		 379 if (err)
380 goto out;
381
382 memcpy(gid->raw, out_mad->data + 8, 8);
383
a0c64a17
JM		 384 if (mlx4_is_mfunc(dev->dev) && !netw_view) {
385 if (index) {
386 /* For any index > 0, return the null guid */
387 err = 0;
388 clear = 1;
389 goto out;
390 }
391 }
392
225c7b1f
RD		 393 init_query_mad(in_mad);
394 in_mad->attr_id = IB_SMP_ATTR_GUID_INFO;
395 in_mad->attr_mod = cpu_to_be32(index / 8);
396
a0c64a17 397 err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
0a9a0188 398 NULL, NULL, in_mad, out_mad);
225c7b1f
RD		 399 if (err)
400 goto out;
401
402 memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
403
404out:
a0c64a17
JM		 405 if (clear)
406 memset(gid->raw + 8, 0, 8);
225c7b1f
RD		 407 kfree(in_mad);
408 kfree(out_mad);
409 return err;
410}
411
fa417f7b
EC		 412static int iboe_query_gid(struct ib_device *ibdev, u8 port, int index,
413 union ib_gid *gid)
414{
415 struct mlx4_ib_dev *dev = to_mdev(ibdev);
416
417 *gid = dev->iboe.gid_table[port - 1][index];
418
419 return 0;
420}
421
422static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
423 union ib_gid *gid)
424{
425 if (rdma_port_get_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND)
a0c64a17 426 return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
fa417f7b
EC		 427 else
428 return iboe_query_gid(ibdev, port, index, gid);
429}
430
0a9a0188
JM		 431int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
432 u16 *pkey, int netw_view)
225c7b1f
RD		 433{
434 struct ib_smp *in_mad = NULL;
435 struct ib_smp *out_mad = NULL;
0a9a0188 436 int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
225c7b1f
RD		 437 int err = -ENOMEM;
438
439 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
440 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
441 if (!in_mad || !out_mad)
442 goto out;
443
444 init_query_mad(in_mad);
445 in_mad->attr_id = IB_SMP_ATTR_PKEY_TABLE;
446 in_mad->attr_mod = cpu_to_be32(index / 32);
447
0a9a0188
JM		 448 if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
449 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
450
451 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
452 in_mad, out_mad);
225c7b1f
RD		 453 if (err)
454 goto out;
455
456 *pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
457
458out:
459 kfree(in_mad);
460 kfree(out_mad);
461 return err;
462}
463
0a9a0188
JM		 464static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
465{
466 return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
467}
468
225c7b1f
RD		 469static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
470 struct ib_device_modify *props)
471{
d0d68b86 472 struct mlx4_cmd_mailbox *mailbox;
df7fba66 473 unsigned long flags;
d0d68b86 474
225c7b1f
RD		 475 if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
476 return -EOPNOTSUPP;
477
d0d68b86
JM		 478 if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
479 return 0;
480
992e8e6e
JM		 481 if (mlx4_is_slave(to_mdev(ibdev)->dev))
482 return -EOPNOTSUPP;
483
df7fba66 484 spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
d0d68b86 485 memcpy(ibdev->node_desc, props->node_desc, 64);
df7fba66 486 spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
d0d68b86
JM		 487
488 /*
489 * If possible, pass node desc to FW, so it can generate
490 * a 144 trap. If cmd fails, just ignore.
491 */
492 mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
493 if (IS_ERR(mailbox))
494 return 0;
495
496 memset(mailbox->buf, 0, 256);
497 memcpy(mailbox->buf, props->node_desc, 64);
498 mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
992e8e6e 499 MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
d0d68b86
JM		 500
501 mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
225c7b1f
RD		 502
503 return 0;
504}
505
506static int mlx4_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
507 u32 cap_mask)
508{
509 struct mlx4_cmd_mailbox *mailbox;
510 int err;
fa417f7b 511 u8 is_eth = dev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
225c7b1f
RD		 512
513 mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
514 if (IS_ERR(mailbox))
515 return PTR_ERR(mailbox);
516
517 memset(mailbox->buf, 0, 256);
5ae2a7a8
RD		 518
519 if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
520 *(u8 *) mailbox->buf = !!reset_qkey_viols << 6;
521 ((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
522 } else {
523 ((u8 *) mailbox->buf)[3] = !!reset_qkey_viols;
524 ((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
525 }
225c7b1f 526
fa417f7b 527 err = mlx4_cmd(dev->dev, mailbox->dma, port, is_eth, MLX4_CMD_SET_PORT,
f9baff50 528 MLX4_CMD_TIME_CLASS_B, MLX4_CMD_NATIVE);
225c7b1f
RD		 529
530 mlx4_free_cmd_mailbox(dev->dev, mailbox);
531 return err;
532}
533
534static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
535 struct ib_port_modify *props)
536{
537 struct ib_port_attr attr;
538 u32 cap_mask;
539 int err;
540
541 mutex_lock(&to_mdev(ibdev)->cap_mask_mutex);
542
543 err = mlx4_ib_query_port(ibdev, port, &attr);
544 if (err)
545 goto out;
546
547 cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
548 ~props->clr_port_cap_mask;
549
550 err = mlx4_SET_PORT(to_mdev(ibdev), port,
551 !!(mask & IB_PORT_RESET_QKEY_CNTR),
552 cap_mask);
553
554out:
555 mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
556 return err;
557}
558
559static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
560 struct ib_udata *udata)
561{
562 struct mlx4_ib_dev *dev = to_mdev(ibdev);
563 struct mlx4_ib_ucontext *context;
564 struct mlx4_ib_alloc_ucontext_resp resp;
565 int err;
566
3b4a8cd5
JM		 567 if (!dev->ib_active)
568 return ERR_PTR(-EAGAIN);
569
225c7b1f
RD		 570 resp.qp_tab_size = dev->dev->caps.num_qps;
571 resp.bf_reg_size = dev->dev->caps.bf_reg_size;
572 resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
573
574 context = kmalloc(sizeof *context, GFP_KERNEL);
575 if (!context)
576 return ERR_PTR(-ENOMEM);
577
578 err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
579 if (err) {
580 kfree(context);
581 return ERR_PTR(err);
582 }
583
584 INIT_LIST_HEAD(&context->db_page_list);
585 mutex_init(&context->db_page_mutex);
586
587 err = ib_copy_to_udata(udata, &resp, sizeof resp);
588 if (err) {
589 mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
590 kfree(context);
591 return ERR_PTR(-EFAULT);
592 }
593
594 return &context->ibucontext;
595}
596
597static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
598{
599 struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
600
601 mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
602 kfree(context);
603
604 return 0;
605}
606
607static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
608{
609 struct mlx4_ib_dev *dev = to_mdev(context->device);
610
611 if (vma->vm_end - vma->vm_start != PAGE_SIZE)
612 return -EINVAL;
613
614 if (vma->vm_pgoff == 0) {
615 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
616
617 if (io_remap_pfn_range(vma, vma->vm_start,
618 to_mucontext(context)->uar.pfn,
619 PAGE_SIZE, vma->vm_page_prot))
620 return -EAGAIN;
621 } else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) {
e1d60ec6 622 vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
225c7b1f
RD		 623
624 if (io_remap_pfn_range(vma, vma->vm_start,
625 to_mucontext(context)->uar.pfn +
626 dev->dev->caps.num_uars,
627 PAGE_SIZE, vma->vm_page_prot))
628 return -EAGAIN;
629 } else
630 return -EINVAL;
631
632 return 0;
633}
634
635static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
636 struct ib_ucontext *context,
637 struct ib_udata *udata)
638{
639 struct mlx4_ib_pd *pd;
640 int err;
641
642 pd = kmalloc(sizeof *pd, GFP_KERNEL);
643 if (!pd)
644 return ERR_PTR(-ENOMEM);
645
646 err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
647 if (err) {
648 kfree(pd);
649 return ERR_PTR(err);
650 }
651
652 if (context)
653 if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
654 mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
655 kfree(pd);
656 return ERR_PTR(-EFAULT);
657 }
658
659 return &pd->ibpd;
660}
661
662static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
663{
664 mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
665 kfree(pd);
666
667 return 0;
668}
669
012a8ff5
SH		 670static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
671 struct ib_ucontext *context,
672 struct ib_udata *udata)
673{
674 struct mlx4_ib_xrcd *xrcd;
675 int err;
676
677 if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
678 return ERR_PTR(-ENOSYS);
679
680 xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
681 if (!xrcd)
682 return ERR_PTR(-ENOMEM);
683
684 err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
685 if (err)
686 goto err1;
687
688 xrcd->pd = ib_alloc_pd(ibdev);
689 if (IS_ERR(xrcd->pd)) {
690 err = PTR_ERR(xrcd->pd);
691 goto err2;
692 }
693
694 xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, 1, 0);
695 if (IS_ERR(xrcd->cq)) {
696 err = PTR_ERR(xrcd->cq);
697 goto err3;
698 }
699
700 return &xrcd->ibxrcd;
701
702err3:
703 ib_dealloc_pd(xrcd->pd);
704err2:
705 mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
706err1:
707 kfree(xrcd);
708 return ERR_PTR(err);
709}
710
711static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
712{
713 ib_destroy_cq(to_mxrcd(xrcd)->cq);
714 ib_dealloc_pd(to_mxrcd(xrcd)->pd);
715 mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
716 kfree(xrcd);
717
718 return 0;
719}
720
fa417f7b
EC		 721static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
722{
723 struct mlx4_ib_qp *mqp = to_mqp(ibqp);
724 struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
725 struct mlx4_ib_gid_entry *ge;
726
727 ge = kzalloc(sizeof *ge, GFP_KERNEL);
728 if (!ge)
729 return -ENOMEM;
730
731 ge->gid = *gid;
732 if (mlx4_ib_add_mc(mdev, mqp, gid)) {
733 ge->port = mqp->port;
734 ge->added = 1;
735 }
736
737 mutex_lock(&mqp->mutex);
738 list_add_tail(&ge->list, &mqp->gid_list);
739 mutex_unlock(&mqp->mutex);
740
741 return 0;
742}
743
744int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
745 union ib_gid *gid)
746{
747 u8 mac[6];
748 struct net_device *ndev;
749 int ret = 0;
750
751 if (!mqp->port)
752 return 0;
753
754 spin_lock(&mdev->iboe.lock);
755 ndev = mdev->iboe.netdevs[mqp->port - 1];
756 if (ndev)
757 dev_hold(ndev);
758 spin_unlock(&mdev->iboe.lock);
759
760 if (ndev) {
761 rdma_get_mcast_mac((struct in6_addr *)gid, mac);
762 rtnl_lock();
763 dev_mc_add(mdev->iboe.netdevs[mqp->port - 1], mac);
764 ret = 1;
765 rtnl_unlock();
766 dev_put(ndev);
767 }
768
769 return ret;
770}
771
0ff1fb65
HHZ		 772struct mlx4_ib_steering {
773 struct list_head list;
774 u64 reg_id;
775 union ib_gid gid;
776};
777
225c7b1f
RD		 778static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
779{
fa417f7b
EC		 780 int err;
781 struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
782 struct mlx4_ib_qp *mqp = to_mqp(ibqp);
0ff1fb65
HHZ		 783 u64 reg_id;
784 struct mlx4_ib_steering *ib_steering = NULL;
785
786 if (mdev->dev->caps.steering_mode ==
787 MLX4_STEERING_MODE_DEVICE_MANAGED) {
788 ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
789 if (!ib_steering)
790 return -ENOMEM;
791 }
fa417f7b 792
0ff1fb65
HHZ		 793 err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
794 !!(mqp->flags &
795 MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
796 MLX4_PROT_IB_IPV6, &reg_id);
fa417f7b 797 if (err)
0ff1fb65 798 goto err_malloc;
fa417f7b
EC		 799
800 err = add_gid_entry(ibqp, gid);
801 if (err)
802 goto err_add;
803
0ff1fb65
HHZ		 804 if (ib_steering) {
805 memcpy(ib_steering->gid.raw, gid->raw, 16);
806 ib_steering->reg_id = reg_id;
807 mutex_lock(&mqp->mutex);
808 list_add(&ib_steering->list, &mqp->steering_rules);
809 mutex_unlock(&mqp->mutex);
810 }
fa417f7b
EC		 811 return 0;
812
813err_add:
0ff1fb65
HHZ		 814 mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
815 MLX4_PROT_IB_IPV6, reg_id);
816err_malloc:
817 kfree(ib_steering);
818
fa417f7b
EC		 819 return err;
820}
821
822static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
823{
824 struct mlx4_ib_gid_entry *ge;
825 struct mlx4_ib_gid_entry *tmp;
826 struct mlx4_ib_gid_entry *ret = NULL;
827
828 list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
829 if (!memcmp(raw, ge->gid.raw, 16)) {
830 ret = ge;
831 break;
832 }
833 }
834
835 return ret;
225c7b1f
RD		 836}
837
838static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
839{
fa417f7b
EC		 840 int err;
841 struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
842 struct mlx4_ib_qp *mqp = to_mqp(ibqp);
843 u8 mac[6];
844 struct net_device *ndev;
845 struct mlx4_ib_gid_entry *ge;
0ff1fb65
HHZ		 846 u64 reg_id = 0;
847
848 if (mdev->dev->caps.steering_mode ==
849 MLX4_STEERING_MODE_DEVICE_MANAGED) {
850 struct mlx4_ib_steering *ib_steering;
851
852 mutex_lock(&mqp->mutex);
853 list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
854 if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
855 list_del(&ib_steering->list);
856 break;
857 }
858 }
859 mutex_unlock(&mqp->mutex);
860 if (&ib_steering->list == &mqp->steering_rules) {
861 pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
862 return -EINVAL;
863 }
864 reg_id = ib_steering->reg_id;
865 kfree(ib_steering);
866 }
fa417f7b 867
0ff1fb65
HHZ		 868 err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
869 MLX4_PROT_IB_IPV6, reg_id);
fa417f7b
EC		 870 if (err)
871 return err;
872
873 mutex_lock(&mqp->mutex);
874 ge = find_gid_entry(mqp, gid->raw);
875 if (ge) {
876 spin_lock(&mdev->iboe.lock);
877 ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
878 if (ndev)
879 dev_hold(ndev);
880 spin_unlock(&mdev->iboe.lock);
881 rdma_get_mcast_mac((struct in6_addr *)gid, mac);
882 if (ndev) {
883 rtnl_lock();
884 dev_mc_del(mdev->iboe.netdevs[ge->port - 1], mac);
885 rtnl_unlock();
886 dev_put(ndev);
887 }
888 list_del(&ge->list);
889 kfree(ge);
890 } else
987c8f8f 891 pr_warn("could not find mgid entry\n");
fa417f7b
EC		 892
893 mutex_unlock(&mqp->mutex);
894
895 return 0;
225c7b1f
RD		 896}
897
898static int init_node_data(struct mlx4_ib_dev *dev)
899{
900 struct ib_smp *in_mad = NULL;
901 struct ib_smp *out_mad = NULL;
0a9a0188 902 int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
225c7b1f
RD		 903 int err = -ENOMEM;
904
905 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
906 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
907 if (!in_mad || !out_mad)
908 goto out;
909
910 init_query_mad(in_mad);
911 in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
0a9a0188
JM		 912 if (mlx4_is_master(dev->dev))
913 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
225c7b1f 914
0a9a0188 915 err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
225c7b1f
RD		 916 if (err)
917 goto out;
918
919 memcpy(dev->ib_dev.node_desc, out_mad->data, 64);
920
921 in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
922
0a9a0188 923 err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
225c7b1f
RD		 924 if (err)
925 goto out;
926
992e8e6e 927 dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
225c7b1f
RD		 928 memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
929
930out:
931 kfree(in_mad);
932 kfree(out_mad);
933 return err;
934}
935
f4e91eb4
TJ		 936static ssize_t show_hca(struct device *device, struct device_attribute *attr,
937 char *buf)
cd9281d8 938{
f4e91eb4
TJ		 939 struct mlx4_ib_dev *dev =
940 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
cd9281d8
JM		 941 return sprintf(buf, "MT%d\n", dev->dev->pdev->device);
942}
943
f4e91eb4
TJ		 944static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
945 char *buf)
cd9281d8 946{
f4e91eb4
TJ		 947 struct mlx4_ib_dev *dev =
948 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
cd9281d8
JM		 949 return sprintf(buf, "%d.%d.%d\n", (int) (dev->dev->caps.fw_ver >> 32),
950 (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
951 (int) dev->dev->caps.fw_ver & 0xffff);
952}
953
f4e91eb4
TJ		 954static ssize_t show_rev(struct device *device, struct device_attribute *attr,
955 char *buf)
cd9281d8 956{
f4e91eb4
TJ		 957 struct mlx4_ib_dev *dev =
958 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
cd9281d8
JM		 959 return sprintf(buf, "%x\n", dev->dev->rev_id);
960}
961
f4e91eb4
TJ		 962static ssize_t show_board(struct device *device, struct device_attribute *attr,
963 char *buf)
cd9281d8 964{
f4e91eb4
TJ		 965 struct mlx4_ib_dev *dev =
966 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
967 return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
968 dev->dev->board_id);
cd9281d8
JM		 969}
970
f4e91eb4
TJ		 971static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL);
972static DEVICE_ATTR(fw_ver, S_IRUGO, show_fw_ver, NULL);
973static DEVICE_ATTR(hca_type, S_IRUGO, show_hca, NULL);
974static DEVICE_ATTR(board_id, S_IRUGO, show_board, NULL);
cd9281d8 975
f4e91eb4
TJ		 976static struct device_attribute *mlx4_class_attributes[] = {
977 &dev_attr_hw_rev,
978 &dev_attr_fw_ver,
979 &dev_attr_hca_type,
980 &dev_attr_board_id
cd9281d8
JM		 981};
982
4c3eb3ca 983static void mlx4_addrconf_ifid_eui48(u8 *eui, u16 vlan_id, struct net_device *dev)
fa417f7b
EC		 984{
985 memcpy(eui, dev->dev_addr, 3);
986 memcpy(eui + 5, dev->dev_addr + 3, 3);
4c3eb3ca
EC		 987 if (vlan_id < 0x1000) {
988 eui[3] = vlan_id >> 8;
989 eui[4] = vlan_id & 0xff;
990 } else {
991 eui[3] = 0xff;
992 eui[4] = 0xfe;
993 }
fa417f7b
EC		 994 eui[0] ^= 2;
995}
996
997static void update_gids_task(struct work_struct *work)
998{
999 struct update_gid_work *gw = container_of(work, struct update_gid_work, work);
1000 struct mlx4_cmd_mailbox *mailbox;
1001 union ib_gid *gids;
1002 int err;
1003 struct mlx4_dev *dev = gw->dev->dev;
fa417f7b
EC		 1004
1005 mailbox = mlx4_alloc_cmd_mailbox(dev);
1006 if (IS_ERR(mailbox)) {
987c8f8f 1007 pr_warn("update gid table failed %ld\n", PTR_ERR(mailbox));
fa417f7b
EC		 1008 return;
1009 }
1010
1011 gids = mailbox->buf;
1012 memcpy(gids, gw->gids, sizeof gw->gids);
1013
1014 err = mlx4_cmd(dev, mailbox->dma, MLX4_SET_PORT_GID_TABLE << 8 | gw->port,
f9baff50 1015 1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
992e8e6e 1016 MLX4_CMD_WRAPPED);
fa417f7b 1017 if (err)
987c8f8f 1018 pr_warn("set port command failed\n");
fa417f7b
EC		 1019 else {
1020 memcpy(gw->dev->iboe.gid_table[gw->port - 1], gw->gids, sizeof gw->gids);
00f5ce99 1021 mlx4_ib_dispatch_event(gw->dev, gw->port, IB_EVENT_GID_CHANGE);
fa417f7b
EC		 1022 }
1023
1024 mlx4_free_cmd_mailbox(dev, mailbox);
1025 kfree(gw);
1026}
1027
1028static int update_ipv6_gids(struct mlx4_ib_dev *dev, int port, int clear)
1029{
1030 struct net_device *ndev = dev->iboe.netdevs[port - 1];
1031 struct update_gid_work *work;
4c3eb3ca
EC		 1032 struct net_device *tmp;
1033 int i;
1034 u8 *hits;
1035 int ret;
1036 union ib_gid gid;
1037 int free;
1038 int found;
1039 int need_update = 0;
1040 u16 vid;
fa417f7b
EC		 1041
1042 work = kzalloc(sizeof *work, GFP_ATOMIC);
1043 if (!work)
1044 return -ENOMEM;
1045
4c3eb3ca
EC		 1046 hits = kzalloc(128, GFP_ATOMIC);
1047 if (!hits) {
1048 ret = -ENOMEM;
1049 goto out;
1050 }
1051
22f4fbd9
ED		 1052 rcu_read_lock();
1053 for_each_netdev_rcu(&init_net, tmp) {
4c3eb3ca
EC		 1054 if (ndev && (tmp == ndev || rdma_vlan_dev_real_dev(tmp) == ndev)) {
1055 gid.global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL);
1056 vid = rdma_vlan_dev_vlan_id(tmp);
1057 mlx4_addrconf_ifid_eui48(&gid.raw[8], vid, ndev);
1058 found = 0;
1059 free = -1;
1060 for (i = 0; i < 128; ++i) {
1061 if (free < 0 &&
1062 !memcmp(&dev->iboe.gid_table[port - 1][i], &zgid, sizeof zgid))
1063 free = i;
1064 if (!memcmp(&dev->iboe.gid_table[port - 1][i], &gid, sizeof gid)) {
1065 hits[i] = 1;
1066 found = 1;
1067 break;
1068 }
1069 }
1070
1071 if (!found) {
1072 if (tmp == ndev &&
1073 (memcmp(&dev->iboe.gid_table[port - 1][0],
1074 &gid, sizeof gid) ||
1075 !memcmp(&dev->iboe.gid_table[port - 1][0],
1076 &zgid, sizeof gid))) {
1077 dev->iboe.gid_table[port - 1][0] = gid;
1078 ++need_update;
1079 hits[0] = 1;
1080 } else if (free >= 0) {
1081 dev->iboe.gid_table[port - 1][free] = gid;
1082 hits[free] = 1;
1083 ++need_update;
1084 }
1085 }
1086 }
fa417f7b 1087 }
22f4fbd9 1088 rcu_read_unlock();
4c3eb3ca
EC		 1089
1090 for (i = 0; i < 128; ++i)
1091 if (!hits[i]) {
1092 if (memcmp(&dev->iboe.gid_table[port - 1][i], &zgid, sizeof zgid))
1093 ++need_update;
1094 dev->iboe.gid_table[port - 1][i] = zgid;
1095 }
fa417f7b 1096
4c3eb3ca
EC		 1097 if (need_update) {
1098 memcpy(work->gids, dev->iboe.gid_table[port - 1], sizeof work->gids);
1099 INIT_WORK(&work->work, update_gids_task);
1100 work->port = port;
1101 work->dev = dev;
1102 queue_work(wq, &work->work);
1103 } else
1104 kfree(work);
fa417f7b 1105
4c3eb3ca 1106 kfree(hits);
fa417f7b 1107 return 0;
4c3eb3ca
EC		 1108
1109out:
1110 kfree(work);
1111 return ret;
fa417f7b
EC		 1112}
1113
1114static void handle_en_event(struct mlx4_ib_dev *dev, int port, unsigned long event)
1115{
1116 switch (event) {
1117 case NETDEV_UP:
4c3eb3ca 1118 case NETDEV_CHANGEADDR:
fa417f7b
EC		 1119 update_ipv6_gids(dev, port, 0);
1120 break;
1121
1122 case NETDEV_DOWN:
1123 update_ipv6_gids(dev, port, 1);
1124 dev->iboe.netdevs[port - 1] = NULL;
1125 }
1126}
1127
1128static void netdev_added(struct mlx4_ib_dev *dev, int port)
1129{
1130 update_ipv6_gids(dev, port, 0);
1131}
1132
1133static void netdev_removed(struct mlx4_ib_dev *dev, int port)
1134{
1135 update_ipv6_gids(dev, port, 1);
1136}
1137
1138static int mlx4_ib_netdev_event(struct notifier_block *this, unsigned long event,
1139 void *ptr)
1140{
1141 struct net_device *dev = ptr;
1142 struct mlx4_ib_dev *ibdev;
1143 struct net_device *oldnd;
1144 struct mlx4_ib_iboe *iboe;
1145 int port;
1146
1147 if (!net_eq(dev_net(dev), &init_net))
1148 return NOTIFY_DONE;
1149
1150 ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
1151 iboe = &ibdev->iboe;
1152
1153 spin_lock(&iboe->lock);
1154 mlx4_foreach_ib_transport_port(port, ibdev->dev) {
1155 oldnd = iboe->netdevs[port - 1];
1156 iboe->netdevs[port - 1] =
0345584e 1157 mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
fa417f7b
EC		 1158 if (oldnd != iboe->netdevs[port - 1]) {
1159 if (iboe->netdevs[port - 1])
1160 netdev_added(ibdev, port);
1161 else
1162 netdev_removed(ibdev, port);
1163 }
1164 }
1165
4c3eb3ca
EC		 1166 if (dev == iboe->netdevs[0] ||
1167 (iboe->netdevs[0] && rdma_vlan_dev_real_dev(dev) == iboe->netdevs[0]))
fa417f7b 1168 handle_en_event(ibdev, 1, event);
4c3eb3ca
EC		 1169 else if (dev == iboe->netdevs[1]
1170 || (iboe->netdevs[1] && rdma_vlan_dev_real_dev(dev) == iboe->netdevs[1]))
fa417f7b
EC		 1171 handle_en_event(ibdev, 2, event);
1172
1173 spin_unlock(&iboe->lock);
1174
1175 return NOTIFY_DONE;
1176}
1177
54679e14
JM		 1178static void init_pkeys(struct mlx4_ib_dev *ibdev)
1179{
1180 int port;
1181 int slave;
1182 int i;
1183
1184 if (mlx4_is_master(ibdev->dev)) {
1185 for (slave = 0; slave <= ibdev->dev->num_vfs; ++slave) {
1186 for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
1187 for (i = 0;
1188 i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
1189 ++i) {
1190 ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
1191 /* master has the identity virt2phys pkey mapping */
1192 (slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
1193 ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
1194 mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
1195 ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
1196 }
1197 }
1198 }
1199 /* initialize pkey cache */
1200 for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
1201 for (i = 0;
1202 i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
1203 ++i)
1204 ibdev->pkeys.phys_pkey_cache[port-1][i] =
1205 (i) ? 0 : 0xFFFF;
1206 }
1207 }
1208}
1209
e605b743
SP		 1210static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
1211{
1212 char name[32];
1213 int eq_per_port = 0;
1214 int added_eqs = 0;
1215 int total_eqs = 0;
1216 int i, j, eq;
1217
3aac6ff1
SP		 1218 /* Legacy mode or comp_pool is not large enough */
1219 if (dev->caps.comp_pool == 0 ||
1220 dev->caps.num_ports > dev->caps.comp_pool)
e605b743
SP		 1221 return;
1222
1223 eq_per_port = rounddown_pow_of_two(dev->caps.comp_pool/
1224 dev->caps.num_ports);
1225
1226 /* Init eq table */
1227 added_eqs = 0;
1228 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
1229 added_eqs += eq_per_port;
1230
1231 total_eqs = dev->caps.num_comp_vectors + added_eqs;
1232
1233 ibdev->eq_table = kzalloc(total_eqs * sizeof(int), GFP_KERNEL);
1234 if (!ibdev->eq_table)
1235 return;
1236
1237 ibdev->eq_added = added_eqs;
1238
1239 eq = 0;
1240 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB) {
1241 for (j = 0; j < eq_per_port; j++) {
1242 sprintf(name, "mlx4-ib-%d-%d@%s",
1243 i, j, dev->pdev->bus->name);
1244 /* Set IRQ for specific name (per ring) */
d9236c3f
AV		 1245 if (mlx4_assign_eq(dev, name, NULL,
1246 &ibdev->eq_table[eq])) {
e605b743
SP		 1247 /* Use legacy (same as mlx4_en driver) */
1248 pr_warn("Can't allocate EQ %d; reverting to legacy\n", eq);
1249 ibdev->eq_table[eq] =
1250 (eq % dev->caps.num_comp_vectors);
1251 }
1252 eq++;
1253 }
1254 }
1255
1256 /* Fill the reset of the vector with legacy EQ */
1257 for (i = 0, eq = added_eqs; i < dev->caps.num_comp_vectors; i++)
1258 ibdev->eq_table[eq++] = i;
1259
1260 /* Advertise the new number of EQs to clients */
1261 ibdev->ib_dev.num_comp_vectors = total_eqs;
1262}
1263
1264static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
1265{
1266 int i;
3aac6ff1
SP		 1267
1268 /* no additional eqs were added */
1269 if (!ibdev->eq_table)
1270 return;
e605b743
SP		 1271
1272 /* Reset the advertised EQ number */
1273 ibdev->ib_dev.num_comp_vectors = dev->caps.num_comp_vectors;
1274
1275 /* Free only the added eqs */
1276 for (i = 0; i < ibdev->eq_added; i++) {
1277 /* Don't free legacy eqs if used */
1278 if (ibdev->eq_table[i] <= dev->caps.num_comp_vectors)
1279 continue;
1280 mlx4_release_eq(dev, ibdev->eq_table[i]);
1281 }
1282
e605b743 1283 kfree(ibdev->eq_table);
e605b743
SP		 1284}
1285
225c7b1f
RD		 1286static void *mlx4_ib_add(struct mlx4_dev *dev)
1287{
1288 struct mlx4_ib_dev *ibdev;
22e7ef9c 1289 int num_ports = 0;
035b1032 1290 int i, j;
fa417f7b
EC		 1291 int err;
1292 struct mlx4_ib_iboe *iboe;
225c7b1f 1293
987c8f8f 1294 pr_info_once("%s", mlx4_ib_version);
68f3948d 1295
8e59d254 1296 if (mlx4_is_mfunc(dev)) {
987c8f8f 1297 pr_warn("IB not yet supported in SRIOV\n");
8e59d254
JM		 1298 return NULL;
1299 }
1300
fa417f7b 1301 mlx4_foreach_ib_transport_port(i, dev)
22e7ef9c
RD		 1302 num_ports++;
1303
1304 /* No point in registering a device with no ports... */
1305 if (num_ports == 0)
1306 return NULL;
1307
225c7b1f
RD		 1308 ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
1309 if (!ibdev) {
1310 dev_err(&dev->pdev->dev, "Device struct alloc failed\n");
1311 return NULL;
1312 }
1313
fa417f7b
EC		 1314 iboe = &ibdev->iboe;
1315
225c7b1f
RD		 1316 if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
1317 goto err_dealloc;
1318
1319 if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
1320 goto err_pd;
1321
4979d18f
RD		 1322 ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
1323 PAGE_SIZE);
225c7b1f
RD		 1324 if (!ibdev->uar_map)
1325 goto err_uar;
26c6bc7b 1326 MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
225c7b1f 1327
225c7b1f
RD		 1328 ibdev->dev = dev;
1329
1330 strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
1331 ibdev->ib_dev.owner = THIS_MODULE;
1332 ibdev->ib_dev.node_type = RDMA_NODE_IB_CA;
95d04f07 1333 ibdev->ib_dev.local_dma_lkey = dev->caps.reserved_lkey;
22e7ef9c 1334 ibdev->num_ports = num_ports;
7ff93f8b 1335 ibdev->ib_dev.phys_port_cnt = ibdev->num_ports;
b8dd786f 1336 ibdev->ib_dev.num_comp_vectors = dev->caps.num_comp_vectors;
225c7b1f
RD		 1337 ibdev->ib_dev.dma_device = &dev->pdev->dev;
1338
1339 ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
1340 ibdev->ib_dev.uverbs_cmd_mask =
1341 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT) |
1342 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE) |
1343 (1ull << IB_USER_VERBS_CMD_QUERY_PORT) |
1344 (1ull << IB_USER_VERBS_CMD_ALLOC_PD) |
1345 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD) |
1346 (1ull << IB_USER_VERBS_CMD_REG_MR) |
1347 (1ull << IB_USER_VERBS_CMD_DEREG_MR) |
1348 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
1349 (1ull << IB_USER_VERBS_CMD_CREATE_CQ) |
bbf8eed1 1350 (1ull << IB_USER_VERBS_CMD_RESIZE_CQ) |
225c7b1f
RD		 1351 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ) |
1352 (1ull << IB_USER_VERBS_CMD_CREATE_QP) |
1353 (1ull << IB_USER_VERBS_CMD_MODIFY_QP) |
6a775e2b 1354 (1ull << IB_USER_VERBS_CMD_QUERY_QP) |
225c7b1f
RD		 1355 (1ull << IB_USER_VERBS_CMD_DESTROY_QP) |
1356 (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST) |
1357 (1ull << IB_USER_VERBS_CMD_DETACH_MCAST) |
1358 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ) |
1359 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ) |
65541cb7 1360 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ) |
18abd5ea 1361 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ) |
42849b26
SH		 1362 (1ull << IB_USER_VERBS_CMD_CREATE_XSRQ) |
1363 (1ull << IB_USER_VERBS_CMD_OPEN_QP);
225c7b1f
RD		 1364
1365 ibdev->ib_dev.query_device = mlx4_ib_query_device;
1366 ibdev->ib_dev.query_port = mlx4_ib_query_port;
fa417f7b 1367 ibdev->ib_dev.get_link_layer = mlx4_ib_port_link_layer;
225c7b1f
RD		 1368 ibdev->ib_dev.query_gid = mlx4_ib_query_gid;
1369 ibdev->ib_dev.query_pkey = mlx4_ib_query_pkey;
1370 ibdev->ib_dev.modify_device = mlx4_ib_modify_device;
1371 ibdev->ib_dev.modify_port = mlx4_ib_modify_port;
1372 ibdev->ib_dev.alloc_ucontext = mlx4_ib_alloc_ucontext;
1373 ibdev->ib_dev.dealloc_ucontext = mlx4_ib_dealloc_ucontext;
1374 ibdev->ib_dev.mmap = mlx4_ib_mmap;
1375 ibdev->ib_dev.alloc_pd = mlx4_ib_alloc_pd;
1376 ibdev->ib_dev.dealloc_pd = mlx4_ib_dealloc_pd;
1377 ibdev->ib_dev.create_ah = mlx4_ib_create_ah;
1378 ibdev->ib_dev.query_ah = mlx4_ib_query_ah;
1379 ibdev->ib_dev.destroy_ah = mlx4_ib_destroy_ah;
1380 ibdev->ib_dev.create_srq = mlx4_ib_create_srq;
1381 ibdev->ib_dev.modify_srq = mlx4_ib_modify_srq;
65541cb7 1382 ibdev->ib_dev.query_srq = mlx4_ib_query_srq;
225c7b1f
RD		 1383 ibdev->ib_dev.destroy_srq = mlx4_ib_destroy_srq;
1384 ibdev->ib_dev.post_srq_recv = mlx4_ib_post_srq_recv;
1385 ibdev->ib_dev.create_qp = mlx4_ib_create_qp;
1386 ibdev->ib_dev.modify_qp = mlx4_ib_modify_qp;
6a775e2b 1387 ibdev->ib_dev.query_qp = mlx4_ib_query_qp;
225c7b1f
RD		 1388 ibdev->ib_dev.destroy_qp = mlx4_ib_destroy_qp;
1389 ibdev->ib_dev.post_send = mlx4_ib_post_send;
1390 ibdev->ib_dev.post_recv = mlx4_ib_post_recv;
1391 ibdev->ib_dev.create_cq = mlx4_ib_create_cq;
3fdcb97f 1392 ibdev->ib_dev.modify_cq = mlx4_ib_modify_cq;
bbf8eed1 1393 ibdev->ib_dev.resize_cq = mlx4_ib_resize_cq;
225c7b1f
RD		 1394 ibdev->ib_dev.destroy_cq = mlx4_ib_destroy_cq;
1395 ibdev->ib_dev.poll_cq = mlx4_ib_poll_cq;
1396 ibdev->ib_dev.req_notify_cq = mlx4_ib_arm_cq;
1397 ibdev->ib_dev.get_dma_mr = mlx4_ib_get_dma_mr;
1398 ibdev->ib_dev.reg_user_mr = mlx4_ib_reg_user_mr;
1399 ibdev->ib_dev.dereg_mr = mlx4_ib_dereg_mr;
95d04f07
RD		 1400 ibdev->ib_dev.alloc_fast_reg_mr = mlx4_ib_alloc_fast_reg_mr;
1401 ibdev->ib_dev.alloc_fast_reg_page_list = mlx4_ib_alloc_fast_reg_page_list;
1402 ibdev->ib_dev.free_fast_reg_page_list = mlx4_ib_free_fast_reg_page_list;
225c7b1f
RD		 1403 ibdev->ib_dev.attach_mcast = mlx4_ib_mcg_attach;
1404 ibdev->ib_dev.detach_mcast = mlx4_ib_mcg_detach;
1405 ibdev->ib_dev.process_mad = mlx4_ib_process_mad;
1406
992e8e6e
JM		 1407 if (!mlx4_is_slave(ibdev->dev)) {
1408 ibdev->ib_dev.alloc_fmr = mlx4_ib_fmr_alloc;
1409 ibdev->ib_dev.map_phys_fmr = mlx4_ib_map_phys_fmr;
1410 ibdev->ib_dev.unmap_fmr = mlx4_ib_unmap_fmr;
1411 ibdev->ib_dev.dealloc_fmr = mlx4_ib_fmr_dealloc;
1412 }
8ad11fb6 1413
012a8ff5
SH		 1414 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
1415 ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
1416 ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
1417 ibdev->ib_dev.uverbs_cmd_mask |=
1418 (1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
1419 (1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
1420 }
1421
e605b743
SP		 1422 mlx4_ib_alloc_eqs(dev, ibdev);
1423
fa417f7b
EC		 1424 spin_lock_init(&iboe->lock);
1425
225c7b1f
RD		 1426 if (init_node_data(ibdev))
1427 goto err_map;
1428
cfcde11c
OG		 1429 for (i = 0; i < ibdev->num_ports; ++i) {
1430 if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
1431 IB_LINK_LAYER_ETHERNET) {
1432 err = mlx4_counter_alloc(ibdev->dev, &ibdev->counters[i]);
1433 if (err)
1434 ibdev->counters[i] = -1;
1435 } else
1436 ibdev->counters[i] = -1;
1437 }
1438
225c7b1f
RD		 1439 spin_lock_init(&ibdev->sm_lock);
1440 mutex_init(&ibdev->cap_mask_mutex);
1441
9a6edb60 1442 if (ib_register_device(&ibdev->ib_dev, NULL))
cfcde11c 1443 goto err_counter;
225c7b1f
RD		 1444
1445 if (mlx4_ib_mad_init(ibdev))
1446 goto err_reg;
1447
fc06573d
JM		 1448 if (mlx4_ib_init_sriov(ibdev))
1449 goto err_mad;
1450
fa417f7b
EC		 1451 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE && !iboe->nb.notifier_call) {
1452 iboe->nb.notifier_call = mlx4_ib_netdev_event;
1453 err = register_netdevice_notifier(&iboe->nb);
1454 if (err)
fc06573d 1455 goto err_sriov;
fa417f7b
EC		 1456 }
1457
035b1032 1458 for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
f4e91eb4 1459 if (device_create_file(&ibdev->ib_dev.dev,
035b1032 1460 mlx4_class_attributes[j]))
fa417f7b 1461 goto err_notif;
cd9281d8
JM		 1462 }
1463
3b4a8cd5
JM		 1464 ibdev->ib_active = true;
1465
54679e14
JM		 1466 if (mlx4_is_mfunc(ibdev->dev))
1467 init_pkeys(ibdev);
1468
225c7b1f
RD		 1469 return ibdev;
1470
fa417f7b
EC		 1471err_notif:
1472 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
987c8f8f 1473 pr_warn("failure unregistering notifier\n");
fa417f7b
EC		 1474 flush_workqueue(wq);
1475
fc06573d
JM		 1476err_sriov:
1477 mlx4_ib_close_sriov(ibdev);
1478
1479err_mad:
1480 mlx4_ib_mad_cleanup(ibdev);
1481
225c7b1f
RD		 1482err_reg:
1483 ib_unregister_device(&ibdev->ib_dev);
1484
cfcde11c
OG		 1485err_counter:
1486 for (; i; --i)
4af3ce0d
RD		 1487 if (ibdev->counters[i - 1] != -1)
1488 mlx4_counter_free(ibdev->dev, ibdev->counters[i - 1]);
cfcde11c 1489
225c7b1f
RD		 1490err_map:
1491 iounmap(ibdev->uar_map);
1492
1493err_uar:
1494 mlx4_uar_free(dev, &ibdev->priv_uar);
1495
1496err_pd:
1497 mlx4_pd_free(dev, ibdev->priv_pdn);
1498
1499err_dealloc:
1500 ib_dealloc_device(&ibdev->ib_dev);
1501
1502 return NULL;
1503}
1504
1505static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
1506{
1507 struct mlx4_ib_dev *ibdev = ibdev_ptr;
1508 int p;
1509
fc06573d 1510 mlx4_ib_close_sriov(ibdev);
a6a47771
YP		 1511 mlx4_ib_mad_cleanup(ibdev);
1512 ib_unregister_device(&ibdev->ib_dev);
fa417f7b
EC		 1513 if (ibdev->iboe.nb.notifier_call) {
1514 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
987c8f8f 1515 pr_warn("failure unregistering notifier\n");
fa417f7b
EC		 1516 ibdev->iboe.nb.notifier_call = NULL;
1517 }
1518 iounmap(ibdev->uar_map);
cfcde11c 1519 for (p = 0; p < ibdev->num_ports; ++p)
4af3ce0d
RD		 1520 if (ibdev->counters[p] != -1)
1521 mlx4_counter_free(ibdev->dev, ibdev->counters[p]);
fa417f7b 1522 mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
225c7b1f
RD		 1523 mlx4_CLOSE_PORT(dev, p);
1524
e605b743
SP		 1525 mlx4_ib_free_eqs(dev, ibdev);
1526
225c7b1f
RD		 1527 mlx4_uar_free(dev, &ibdev->priv_uar);
1528 mlx4_pd_free(dev, ibdev->priv_pdn);
1529 ib_dealloc_device(&ibdev->ib_dev);
1530}
1531
fc06573d
JM		 1532static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
1533{
1534 struct mlx4_ib_demux_work **dm = NULL;
1535 struct mlx4_dev *dev = ibdev->dev;
1536 int i;
1537 unsigned long flags;
1538
1539 if (!mlx4_is_master(dev))
1540 return;
1541
1542 dm = kcalloc(dev->caps.num_ports, sizeof *dm, GFP_ATOMIC);
1543 if (!dm) {
1544 pr_err("failed to allocate memory for tunneling qp update\n");
1545 goto out;
1546 }
1547
1548 for (i = 0; i < dev->caps.num_ports; i++) {
1549 dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
1550 if (!dm[i]) {
1551 pr_err("failed to allocate memory for tunneling qp update work struct\n");
1552 for (i = 0; i < dev->caps.num_ports; i++) {
1553 if (dm[i])
1554 kfree(dm[i]);
1555 }
1556 goto out;
1557 }
1558 }
1559 /* initialize or tear down tunnel QPs for the slave */
1560 for (i = 0; i < dev->caps.num_ports; i++) {
1561 INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
1562 dm[i]->port = i + 1;
1563 dm[i]->slave = slave;
1564 dm[i]->do_init = do_init;
1565 dm[i]->dev = ibdev;
1566 spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
1567 if (!ibdev->sriov.is_going_down)
1568 queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
1569 spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
1570 }
1571out:
1572 if (dm)
1573 kfree(dm);
1574 return;
1575}
1576
225c7b1f 1577static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
00f5ce99 1578 enum mlx4_dev_event event, unsigned long param)
225c7b1f
RD		 1579{
1580 struct ib_event ibev;
7ff93f8b 1581 struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
00f5ce99
JM		 1582 struct mlx4_eqe *eqe = NULL;
1583 struct ib_event_work *ew;
fc06573d 1584 int p = 0;
00f5ce99
JM		 1585
1586 if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
1587 eqe = (struct mlx4_eqe *)param;
1588 else
fc06573d 1589 p = (int) param;
225c7b1f
RD		 1590
1591 switch (event) {
37608eea 1592 case MLX4_DEV_EVENT_PORT_UP:
fc06573d
JM		 1593 if (p > ibdev->num_ports)
1594 return;
a0c64a17
JM		 1595 if (mlx4_is_master(dev) &&
1596 rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
1597 IB_LINK_LAYER_INFINIBAND) {
1598 mlx4_ib_invalidate_all_guid_record(ibdev, p);
1599 }
37608eea 1600 ibev.event = IB_EVENT_PORT_ACTIVE;
225c7b1f
RD		 1601 break;
1602
37608eea 1603 case MLX4_DEV_EVENT_PORT_DOWN:
fc06573d
JM		 1604 if (p > ibdev->num_ports)
1605 return;
37608eea
RD		 1606 ibev.event = IB_EVENT_PORT_ERR;
1607 break;
1608
1609 case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
3b4a8cd5 1610 ibdev->ib_active = false;
225c7b1f
RD		 1611 ibev.event = IB_EVENT_DEVICE_FATAL;
1612 break;
1613
00f5ce99
JM		 1614 case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
1615 ew = kmalloc(sizeof *ew, GFP_ATOMIC);
1616 if (!ew) {
1617 pr_err("failed to allocate memory for events work\n");
1618 break;
1619 }
1620
1621 INIT_WORK(&ew->work, handle_port_mgmt_change_event);
1622 memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
1623 ew->ib_dev = ibdev;
992e8e6e
JM		 1624 /* need to queue only for port owner, which uses GEN_EQE */
1625 if (mlx4_is_master(dev))
1626 queue_work(wq, &ew->work);
1627 else
1628 handle_port_mgmt_change_event(&ew->work);
00f5ce99
JM		 1629 return;
1630
fc06573d
JM		 1631 case MLX4_DEV_EVENT_SLAVE_INIT:
1632 /* here, p is the slave id */
1633 do_slave_init(ibdev, p, 1);
1634 return;
1635
1636 case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
1637 /* here, p is the slave id */
1638 do_slave_init(ibdev, p, 0);
1639 return;
1640
225c7b1f
RD		 1641 default:
1642 return;
1643 }
1644
1645 ibev.device = ibdev_ptr;
fc06573d 1646 ibev.element.port_num = (u8) p;
225c7b1f
RD		 1647
1648 ib_dispatch_event(&ibev);
1649}
1650
1651static struct mlx4_interface mlx4_ib_interface = {
fa417f7b
EC		 1652 .add = mlx4_ib_add,
1653 .remove = mlx4_ib_remove,
1654 .event = mlx4_ib_event,
0345584e 1655 .protocol = MLX4_PROT_IB_IPV6
225c7b1f
RD		 1656};
1657
1658static int __init mlx4_ib_init(void)
1659{
fa417f7b
EC		 1660 int err;
1661
1662 wq = create_singlethread_workqueue("mlx4_ib");
1663 if (!wq)
1664 return -ENOMEM;
1665
b9c5d6a6
OD		 1666 err = mlx4_ib_mcg_init();
1667 if (err)
1668 goto clean_wq;
1669
fa417f7b 1670 err = mlx4_register_interface(&mlx4_ib_interface);
b9c5d6a6
OD		 1671 if (err)
1672 goto clean_mcg;
fa417f7b
EC		 1673
1674 return 0;
b9c5d6a6
OD		 1675
1676clean_mcg:
1677 mlx4_ib_mcg_destroy();
1678
1679clean_wq:
1680 destroy_workqueue(wq);
1681 return err;
225c7b1f
RD		 1682}
1683
1684static void __exit mlx4_ib_cleanup(void)
1685{
1686 mlx4_unregister_interface(&mlx4_ib_interface);
b9c5d6a6 1687 mlx4_ib_mcg_destroy();
fa417f7b 1688 destroy_workqueue(wq);
225c7b1f
RD		 1689}
1690
1691module_init(mlx4_ib_init);
1692module_exit(mlx4_ib_cleanup);
Linux 6.x block layer and io_uring tree(s)